Adding upstream version 110.0.1.upstream/110.0.1upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

159 files changed, 42085 insertions, 0 deletions

diff --git a/third_party/aom/test/accounting_test.cc b/third_party/aom/test/accounting_test.cc
new file mode 100644
index 0000000000..8b5c8af135
--- /dev/null
+++ b/third_party/aom/test/accounting_test.cc
@@ -0,0 +1,75 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/acm_random.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/bitreader.h"
+#include "aom_dsp/bitwriter.h"
+
+using libaom_test::ACMRandom;
+
+TEST(AV1, TestAccounting) {
+  const int kBufferSize = 10000;
+  const int kSymbols = 1024;
+  aom_writer bw;
+  uint8_t bw_buffer[kBufferSize];
+  aom_start_encode(&bw, bw_buffer);
+  for (int i = 0; i < kSymbols; i++) {
+    aom_write(&bw, 0, 32);
+    aom_write(&bw, 0, 32);
+    aom_write(&bw, 0, 32);
+  }
+  aom_stop_encode(&bw);
+  aom_reader br;
+  aom_reader_init(&br, bw_buffer, bw.pos);
+
+  Accounting accounting;
+  aom_accounting_init(&accounting);
+  br.accounting = &accounting;
+  for (int i = 0; i < kSymbols; i++) {
+    aom_read(&br, 32, "A");
+  }
+  // Consecutive symbols that are the same are coalesced.
+  GTEST_ASSERT_EQ(accounting.syms.num_syms, 1);
+  GTEST_ASSERT_EQ(accounting.syms.syms[0].samples, (unsigned int)kSymbols);
+
+  aom_accounting_reset(&accounting);
+  GTEST_ASSERT_EQ(accounting.syms.num_syms, 0);
+
+  // Should record 2 * kSymbols accounting symbols.
+  aom_reader_init(&br, bw_buffer, bw.pos);
+  br.accounting = &accounting;
+  for (int i = 0; i < kSymbols; i++) {
+    aom_read(&br, 32, "A");
+    aom_read(&br, 32, "B");
+    aom_read(&br, 32, "B");
+  }
+  GTEST_ASSERT_EQ(accounting.syms.num_syms, kSymbols * 2);
+  uint32_t tell_frac = aom_reader_tell_frac(&br);
+  for (int i = 0; i < accounting.syms.num_syms; i++) {
+    tell_frac -= accounting.syms.syms[i].bits;
+  }
+  GTEST_ASSERT_EQ(tell_frac, 0U);
+
+  GTEST_ASSERT_EQ(aom_accounting_dictionary_lookup(&accounting, "A"),
+                  aom_accounting_dictionary_lookup(&accounting, "A"));
+
+  // Check for collisions. The current aom_accounting_hash function returns
+  // the same hash code for AB and BA.
+  GTEST_ASSERT_NE(aom_accounting_dictionary_lookup(&accounting, "AB"),
+                  aom_accounting_dictionary_lookup(&accounting, "BA"));
+}
diff --git a/third_party/aom/test/acm_random.h b/third_party/aom/test/acm_random.h
new file mode 100644
index 0000000000..0a8317fd57
--- /dev/null
+++ b/third_party/aom/test/acm_random.h
@@ -0,0 +1,84 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_ACM_RANDOM_H_
+#define AOM_TEST_ACM_RANDOM_H_
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "aom/aom_integer.h"
+
+namespace libaom_test {
+
+class ACMRandom {
+ public:
+  ACMRandom() : random_(DeterministicSeed()) {}
+
+  explicit ACMRandom(int seed) : random_(seed) {}
+
+  void Reset(int seed) { random_.Reseed(seed); }
+
+  uint32_t Rand31(void) {
+    return random_.Generate(testing::internal::Random::kMaxRange);
+  }
+
+  uint16_t Rand16(void) {
+    const uint32_t value =
+        random_.Generate(testing::internal::Random::kMaxRange);
+    return (value >> 15) & 0xffff;
+  }
+
+  int16_t Rand15Signed(void) {
+    const uint32_t value =
+        random_.Generate(testing::internal::Random::kMaxRange);
+    return (value >> 17) & 0xffff;
+  }
+
+  uint16_t Rand12(void) {
+    const uint32_t value =
+        random_.Generate(testing::internal::Random::kMaxRange);
+    // There's a bit more entropy in the upper bits of this implementation.
+    return (value >> 19) & 0xfff;
+  }
+
+  int16_t Rand9Signed(void) {
+    // Use 9 bits: values between 255 (0x0FF) and -256 (0x100).
+    const uint32_t value = random_.Generate(512);
+    return static_cast<int16_t>(value) - 256;
+  }
+
+  uint8_t Rand8(void) {
+    const uint32_t value =
+        random_.Generate(testing::internal::Random::kMaxRange);
+    // There's a bit more entropy in the upper bits of this implementation.
+    return (value >> 23) & 0xff;
+  }
+
+  uint8_t Rand8Extremes(void) {
+    // Returns a random value near 0 or near 255, to better exercise
+    // saturation behavior.
+    const uint8_t r = Rand8();
+    return r < 128 ? r << 4 : r >> 4;
+  }
+
+  int PseudoUniform(int range) { return random_.Generate(range); }
+
+  int operator()(int n) { return PseudoUniform(n); }
+
+  static int DeterministicSeed(void) { return 0xbaba; }
+
+ private:
+  testing::internal::Random random_;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_ACM_RANDOM_H_
diff --git a/third_party/aom/test/active_map_test.cc b/third_party/aom/test/active_map_test.cc
new file mode 100644
index 0000000000..a2b0546edb
--- /dev/null
+++ b/third_party/aom/test/active_map_test.cc
@@ -0,0 +1,103 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <climits>
+#include <vector>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+class ActiveMapTest
+    : public ::libaom_test::CodecTestWith2Params<libaom_test::TestMode, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  static const int kWidth = 208;
+  static const int kHeight = 144;
+
+  ActiveMapTest() : EncoderTest(GET_PARAM(0)) {}
+  virtual ~ActiveMapTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+    cpu_used_ = GET_PARAM(2);
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_CPUUSED, cpu_used_);
+    } else if (video->frame() == 3) {
+      aom_active_map_t map = aom_active_map_t();
+      /* clang-format off */
+      uint8_t active_map[9 * 13] = {
+        1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0,
+        1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0,
+        1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0,
+        1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0,
+        0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 1, 1,
+        0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 1, 0, 1,
+        0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 1, 0, 1,
+        0, 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 1, 1,
+        1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0,
+      };
+      /* clang-format on */
+      map.cols = (kWidth + 15) / 16;
+      map.rows = (kHeight + 15) / 16;
+      ASSERT_EQ(map.cols, 13u);
+      ASSERT_EQ(map.rows, 9u);
+      map.active_map = active_map;
+      encoder->Control(AOME_SET_ACTIVEMAP, &map);
+    } else if (video->frame() == 15) {
+      aom_active_map_t map = aom_active_map_t();
+      map.cols = (kWidth + 15) / 16;
+      map.rows = (kHeight + 15) / 16;
+      map.active_map = NULL;
+      encoder->Control(AOME_SET_ACTIVEMAP, &map);
+    }
+  }
+
+  void DoTest() {
+    // Validate that this non multiple of 64 wide clip encodes
+    cfg_.g_lag_in_frames = 0;
+    cfg_.rc_target_bitrate = 400;
+    cfg_.rc_resize_mode = 0;
+    cfg_.g_pass = AOM_RC_ONE_PASS;
+    cfg_.rc_end_usage = AOM_CBR;
+    cfg_.kf_max_dist = 90000;
+    ::libaom_test::I420VideoSource video("hantro_odd.yuv", kWidth, kHeight, 30,
+                                         1, 0, 20);
+
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  }
+
+  int cpu_used_;
+};
+
+TEST_P(ActiveMapTest, Test) { DoTest(); }
+
+class ActiveMapTestLarge : public ActiveMapTest {};
+
+TEST_P(ActiveMapTestLarge, Test) { DoTest(); }
+
+AV1_INSTANTIATE_TEST_CASE(ActiveMapTestLarge,
+                          ::testing::Values(::libaom_test::kRealTime),
+                          ::testing::Range(0, 5));
+
+AV1_INSTANTIATE_TEST_CASE(ActiveMapTest,
+                          ::testing::Values(::libaom_test::kRealTime),
+                          ::testing::Range(5, 9));
+
+}  // namespace
diff --git a/third_party/aom/test/altref_test.cc b/third_party/aom/test/altref_test.cc
new file mode 100644
index 0000000000..dabb1475a8
--- /dev/null
+++ b/third_party/aom/test/altref_test.cc
@@ -0,0 +1,97 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+namespace {
+
+class AltRefForcedKeyTestLarge
+    : public ::libaom_test::CodecTestWith2Params<libaom_test::TestMode, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  AltRefForcedKeyTestLarge()
+      : EncoderTest(GET_PARAM(0)), encoding_mode_(GET_PARAM(1)),
+        cpu_used_(GET_PARAM(2)), forced_kf_frame_num_(1), frame_num_(0) {}
+  virtual ~AltRefForcedKeyTestLarge() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+    cfg_.rc_end_usage = AOM_VBR;
+    cfg_.g_threads = 0;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 0) {
+      encoder->Control(AOME_SET_CPUUSED, cpu_used_);
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+#if CONFIG_AV1_ENCODER
+      // override test default for tile columns if necessary.
+      if (GET_PARAM(0) == &libaom_test::kAV1) {
+        encoder->Control(AV1E_SET_TILE_COLUMNS, 6);
+      }
+#endif
+    }
+    frame_flags_ =
+        (video->frame() == forced_kf_frame_num_) ? AOM_EFLAG_FORCE_KF : 0;
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (frame_num_ == forced_kf_frame_num_) {
+      ASSERT_TRUE(!!(pkt->data.frame.flags & AOM_FRAME_IS_KEY))
+          << "Frame #" << frame_num_ << " isn't a keyframe!";
+    }
+    ++frame_num_;
+  }
+
+  ::libaom_test::TestMode encoding_mode_;
+  int cpu_used_;
+  unsigned int forced_kf_frame_num_;
+  unsigned int frame_num_;
+};
+
+TEST_P(AltRefForcedKeyTestLarge, Frame1IsKey) {
+  const aom_rational timebase = { 1, 30 };
+  const int lag_values[] = { 3, 15, 25, -1 };
+
+  forced_kf_frame_num_ = 1;
+  for (int i = 0; lag_values[i] != -1; ++i) {
+    frame_num_ = 0;
+    cfg_.g_lag_in_frames = lag_values[i];
+    libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       timebase.den, timebase.num, 0, 30);
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  }
+}
+
+TEST_P(AltRefForcedKeyTestLarge, ForcedFrameIsKey) {
+  const aom_rational timebase = { 1, 30 };
+  const int lag_values[] = { 3, 15, 25, -1 };
+
+  for (int i = 0; lag_values[i] != -1; ++i) {
+    frame_num_ = 0;
+    forced_kf_frame_num_ = lag_values[i] - 1;
+    cfg_.g_lag_in_frames = lag_values[i];
+    libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       timebase.den, timebase.num, 0, 30);
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  }
+}
+
+AV1_INSTANTIATE_TEST_CASE(AltRefForcedKeyTestLarge,
+                          ::testing::Values(::libaom_test::kOnePassGood),
+                          ::testing::Values(2, 5));
+
+}  // namespace
diff --git a/third_party/aom/test/aom_integer_test.cc b/third_party/aom/test/aom_integer_test.cc
new file mode 100644
index 0000000000..fe88a54e9e
--- /dev/null
+++ b/third_party/aom/test/aom_integer_test.cc
@@ -0,0 +1,177 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "aom/aom_integer.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+const uint64_t kMaximumLeb128CodedSize = 8;
+const uint8_t kLeb128PadByte = 0x80;  // Binary: 10000000
+const uint64_t kMaximumLeb128Value = UINT32_MAX;
+const uint32_t kSizeTestNumValues = 6;
+const uint32_t kSizeTestExpectedSizes[kSizeTestNumValues] = {
+  1, 1, 2, 3, 4, 5
+};
+const uint64_t kSizeTestInputs[kSizeTestNumValues] = {
+  0, 0x7f, 0x3fff, 0x1fffff, 0xffffff, 0x10000000
+};
+
+const uint8_t kOutOfRangeLeb128Value[5] = { 0x80, 0x80, 0x80, 0x80,
+                                            0x10 };  // UINT32_MAX + 1
+}  // namespace
+
+TEST(AomLeb128, DecodeTest) {
+  const size_t num_leb128_bytes = 3;
+  const uint8_t leb128_bytes[num_leb128_bytes] = { 0xE5, 0x8E, 0x26 };
+  const uint64_t expected_value = 0x98765;  // 624485
+  const size_t expected_length = 3;
+  uint64_t value = ~0ULL;  // make sure value is cleared by the function
+  size_t length;
+  ASSERT_EQ(
+      aom_uleb_decode(&leb128_bytes[0], num_leb128_bytes, &value, &length), 0);
+  ASSERT_EQ(expected_value, value);
+  ASSERT_EQ(expected_length, length);
+
+  // Make sure the decoder stops on the last marked LEB128 byte.
+  aom_uleb_decode(&leb128_bytes[0], num_leb128_bytes + 1, &value, &length);
+  ASSERT_EQ(expected_value, value);
+  ASSERT_EQ(expected_length, length);
+}
+
+TEST(AomLeb128, EncodeTest) {
+  const uint32_t test_value = 0x98765;  // 624485
+  const uint8_t expected_bytes[3] = { 0xE5, 0x8E, 0x26 };
+  const size_t kWriteBufferSize = 4;
+  uint8_t write_buffer[kWriteBufferSize] = { 0 };
+  size_t bytes_written = 0;
+  ASSERT_EQ(aom_uleb_encode(test_value, kWriteBufferSize, &write_buffer[0],
+                            &bytes_written),
+            0);
+  ASSERT_EQ(bytes_written, 3u);
+  for (size_t i = 0; i < bytes_written; ++i) {
+    ASSERT_EQ(write_buffer[i], expected_bytes[i]);
+  }
+}
+
+TEST(AomLeb128, EncodeDecodeTest) {
+  const uint32_t value = 0x98765;  // 624485
+  const size_t kWriteBufferSize = 4;
+  uint8_t write_buffer[kWriteBufferSize] = { 0 };
+  size_t bytes_written = 0;
+  ASSERT_EQ(aom_uleb_encode(value, kWriteBufferSize, &write_buffer[0],
+                            &bytes_written),
+            0);
+  ASSERT_EQ(bytes_written, 3u);
+  uint64_t decoded_value;
+  size_t decoded_length;
+  aom_uleb_decode(&write_buffer[0], bytes_written, &decoded_value,
+                  &decoded_length);
+  ASSERT_EQ(value, decoded_value);
+  ASSERT_EQ(bytes_written, decoded_length);
+}
+
+TEST(AomLeb128, FixedSizeEncodeTest) {
+  const uint32_t test_value = 0x123;
+  const uint8_t expected_bytes[4] = { 0xa3, 0x82, 0x80, 0x00 };
+  const size_t kWriteBufferSize = 4;
+  uint8_t write_buffer[kWriteBufferSize] = { 0 };
+  size_t bytes_written = 0;
+  ASSERT_EQ(0, aom_uleb_encode_fixed_size(test_value, kWriteBufferSize,
+                                          kWriteBufferSize, &write_buffer[0],
+                                          &bytes_written));
+  ASSERT_EQ(kWriteBufferSize, bytes_written);
+  for (size_t i = 0; i < bytes_written; ++i) {
+    ASSERT_EQ(write_buffer[i], expected_bytes[i]);
+  }
+}
+
+TEST(AomLeb128, FixedSizeEncodeDecodeTest) {
+  const uint32_t value = 0x1;
+  const size_t kWriteBufferSize = 4;
+  uint8_t write_buffer[kWriteBufferSize] = { 0 };
+  size_t bytes_written = 0;
+  ASSERT_EQ(
+      aom_uleb_encode_fixed_size(value, kWriteBufferSize, kWriteBufferSize,
+                                 &write_buffer[0], &bytes_written),
+      0);
+  ASSERT_EQ(bytes_written, 4u);
+  uint64_t decoded_value;
+  size_t decoded_length;
+  aom_uleb_decode(&write_buffer[0], bytes_written, &decoded_value,
+                  &decoded_length);
+  ASSERT_EQ(value, decoded_value);
+  ASSERT_EQ(bytes_written, decoded_length);
+}
+
+TEST(AomLeb128, SizeTest) {
+  for (size_t i = 0; i < kSizeTestNumValues; ++i) {
+    ASSERT_EQ(kSizeTestExpectedSizes[i],
+              aom_uleb_size_in_bytes(kSizeTestInputs[i]));
+  }
+}
+
+TEST(AomLeb128, DecodeFailTest) {
+  // Input buffer containing what would be a valid 9 byte LEB128 encoded
+  // unsigned integer.
+  const uint8_t kAllPadBytesBuffer[kMaximumLeb128CodedSize + 1] = {
+    kLeb128PadByte, kLeb128PadByte, kLeb128PadByte,
+    kLeb128PadByte, kLeb128PadByte, kLeb128PadByte,
+    kLeb128PadByte, kLeb128PadByte, 0
+  };
+  uint64_t decoded_value;
+
+  // Test that decode fails when result would be valid 9 byte integer.
+  ASSERT_EQ(aom_uleb_decode(&kAllPadBytesBuffer[0], kMaximumLeb128CodedSize + 1,
+                            &decoded_value, NULL),
+            -1);
+
+  // Test that encoded value missing terminator byte within available buffer
+  // range causes decode error.
+  ASSERT_EQ(aom_uleb_decode(&kAllPadBytesBuffer[0], kMaximumLeb128CodedSize,
+                            &decoded_value, NULL),
+            -1);
+
+  // Test that LEB128 input that decodes to a value larger than 32-bits fails.
+  size_t value_size = 0;
+  ASSERT_EQ(aom_uleb_decode(&kOutOfRangeLeb128Value[0],
+                            sizeof(kOutOfRangeLeb128Value), &decoded_value,
+                            &value_size),
+            -1);
+}
+
+TEST(AomLeb128, EncodeFailTest) {
+  const size_t kWriteBufferSize = 4;
+  const uint32_t kValidTestValue = 1;
+  uint8_t write_buffer[kWriteBufferSize] = { 0 };
+  size_t coded_size = 0;
+  ASSERT_EQ(
+      aom_uleb_encode(kValidTestValue, kWriteBufferSize, NULL, &coded_size),
+      -1);
+  ASSERT_EQ(aom_uleb_encode(kValidTestValue, kWriteBufferSize, &write_buffer[0],
+                            NULL),
+            -1);
+
+  const uint32_t kValueOutOfRangeForBuffer = 0xFFFFFFFF;
+  ASSERT_EQ(aom_uleb_encode(kValueOutOfRangeForBuffer, kWriteBufferSize,
+                            &write_buffer[0], &coded_size),
+            -1);
+
+  const uint64_t kValueOutOfRange = kMaximumLeb128Value + 1;
+  ASSERT_EQ(aom_uleb_encode(kValueOutOfRange, kWriteBufferSize,
+                            &write_buffer[0], &coded_size),
+            -1);
+
+  const size_t kPadSizeOutOfRange = 5;
+  ASSERT_EQ(aom_uleb_encode_fixed_size(kValidTestValue, kWriteBufferSize,
+                                       kPadSizeOutOfRange, &write_buffer[0],
+                                       &coded_size),
+            -1);
+}
diff --git a/third_party/aom/test/aomcx_set_ref.sh b/third_party/aom/test/aomcx_set_ref.sh
new file mode 100755
index 0000000000..f51b73c58e
--- /dev/null
+++ b/third_party/aom/test/aomcx_set_ref.sh
@@ -0,0 +1,58 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the libaom aom_cx_set_ref example. To add new tests to this
+## file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to aom_cx_set_ref_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: $YUV_RAW_INPUT is required.
+aom_cx_set_ref_verify_environment() {
+  if [ ! -e "${YUV_RAW_INPUT}" ]; then
+    echo "Libaom test data must exist in LIBAOM_TEST_DATA_PATH."
+    return 1
+  fi
+}
+
+# Runs aom_cx_set_ref and updates the reference frame before encoding frame 90.
+# $1 is the codec name, which aom_cx_set_ref does not support at present: It's
+# currently used only to name the output file.
+# TODO(tomfinegan): Pass the codec param once the example is updated to support
+# AV1.
+aom_set_ref() {
+  local encoder="${LIBAOM_BIN_PATH}/aom_cx_set_ref${AOM_TEST_EXE_SUFFIX}"
+  local codec="$1"
+  local output_file="${AOM_TEST_OUTPUT_DIR}/aom_cx_set_ref_${codec}.ivf"
+  local ref_frame_num=4
+  local limit=10
+  if [ ! -x "${encoder}" ]; then
+    elog "${encoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${encoder}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
+      "${YUV_RAW_INPUT_HEIGHT}" "${YUV_RAW_INPUT}" "${output_file}" \
+      "${ref_frame_num}" "${limit}" ${devnull}
+
+  [ -e "${output_file}" ] || return 1
+}
+
+aom_cx_set_ref_av1() {
+  if [ "$(av1_encode_available)" = "yes" ]; then
+    aom_set_ref av1 || return 1
+  fi
+}
+
+aom_cx_set_ref_tests="aom_cx_set_ref_av1"
+
+run_tests aom_cx_set_ref_verify_environment "${aom_cx_set_ref_tests}"
+
diff --git a/third_party/aom/test/aomdec.sh b/third_party/aom/test/aomdec.sh
new file mode 100755
index 0000000000..927142287c
--- /dev/null
+++ b/third_party/aom/test/aomdec.sh
@@ -0,0 +1,147 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests aomdec. To add new tests to this file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to aomdec_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: Make sure input is available.
+aomdec_verify_environment() {
+  if [ "$(av1_encode_available)" != "yes" ] ; then
+    if [ ! -e "${AV1_IVF_FILE}" ] || \
+       [ ! -e "${AV1_OBU_ANNEXB_FILE}" ] || \
+       [ ! -e "${AV1_OBU_SEC5_FILE}" ] || \
+       [ ! -e "${AV1_WEBM_FILE}" ]; then
+      elog "Libaom test data must exist before running this test script when " \
+           " encoding is disabled. "
+      return 1
+    fi
+  fi
+  if [ -z "$(aom_tool_path aomdec)" ]; then
+    elog "aomdec not found. It must exist in LIBAOM_BIN_PATH or its parent."
+    return 1
+  fi
+}
+
+# Wrapper function for running aomdec with pipe input. Requires that
+# LIBAOM_BIN_PATH points to the directory containing aomdec. $1 is used as the
+# input file path and shifted away. All remaining parameters are passed through
+# to aomdec.
+aomdec_pipe() {
+  local input="$1"
+  shift
+  if [ ! -e "${input}" ]; then
+    elog "Input file ($input) missing in aomdec_pipe()"
+    return 1
+  fi
+  cat "${file}" | aomdec - "$@" ${devnull}
+}
+
+
+# Wrapper function for running aomdec. Requires that LIBAOM_BIN_PATH points to
+# the directory containing aomdec. $1 one is used as the input file path and
+# shifted away. All remaining parameters are passed through to aomdec.
+aomdec() {
+  local decoder="$(aom_tool_path aomdec)"
+  local input="$1"
+  shift
+  eval "${AOM_TEST_PREFIX}" "${decoder}" "$input" "$@" ${devnull}
+}
+
+aomdec_can_decode_av1() {
+  if [ "$(av1_decode_available)" = "yes" ]; then
+    echo yes
+  fi
+}
+
+aomdec_av1_ivf() {
+  if [ "$(aomdec_can_decode_av1)" = "yes" ]; then
+    local file="${AV1_IVF_FILE}"
+    if [ ! -e "${file}" ]; then
+      encode_yuv_raw_input_av1 "${file}" --ivf
+    fi
+    aomdec "${AV1_IVF_FILE}" --summary --noblit
+  fi
+}
+
+aomdec_av1_ivf_error_resilient() {
+  if [ "$(aomdec_can_decode_av1)" = "yes" ]; then
+    local file="av1.error-resilient.ivf"
+    if [ ! -e "${file}" ]; then
+      encode_yuv_raw_input_av1 "${file}" --ivf --error-resilient=1
+    fi
+    aomdec "${file}" --summary --noblit
+  fi
+}
+
+aomdec_av1_ivf_multithread() {
+  if [ "$(aomdec_can_decode_av1)" = "yes" ]; then
+    local file="${AV1_IVF_FILE}"
+    if [ ! -e "${file}" ]; then
+      encode_yuv_raw_input_av1 "${file}" --ivf
+    fi
+    for threads in 2 3 4 5 6 7 8; do
+      aomdec "${file}" --summary --noblit --threads=$threads
+    done
+  fi
+}
+
+aomdec_aom_ivf_pipe_input() {
+  if [ "$(aomdec_can_decode_av1)" = "yes" ]; then
+    local file="${AV1_IVF_FILE}"
+    if [ ! -e "${file}" ]; then
+      encode_yuv_raw_input_av1 "${file}" --ivf
+    fi
+    aomdec_pipe "${AV1_IVF_FILE}" --summary --noblit
+  fi
+}
+
+aomdec_av1_obu_annexb() {
+  if [ "$(aomdec_can_decode_av1)" = "yes" ]; then
+    local file="${AV1_OBU_ANNEXB_FILE}"
+    if [ ! -e "${file}" ]; then
+      encode_yuv_raw_input_av1 "${file}" --obu --annexb=1
+    fi
+    aomdec "${file}" --summary --noblit --annexb
+  fi
+}
+
+aomdec_av1_obu_section5() {
+  if [ "$(aomdec_can_decode_av1)" = "yes" ]; then
+    local file="${AV1_OBU_SEC5_FILE}"
+    if [ ! -e "${file}" ]; then
+      encode_yuv_raw_input_av1 "${file}" --obu
+    fi
+    aomdec "${file}" --summary --noblit
+  fi
+}
+
+aomdec_av1_webm() {
+  if [ "$(aomdec_can_decode_av1)" = "yes" ] && \
+     [ "$(webm_io_available)" = "yes" ]; then
+    local file="${AV1_WEBM_FILE}"
+    if [ ! -e "${file}" ]; then
+      encode_yuv_raw_input_av1 "${file}"
+    fi
+    aomdec "${AV1_WEBM_FILE}" --summary --noblit
+  fi
+}
+
+aomdec_tests="aomdec_av1_ivf
+              aomdec_av1_ivf_error_resilient
+              aomdec_av1_ivf_multithread
+              aomdec_aom_ivf_pipe_input
+              aomdec_av1_obu_annexb
+              aomdec_av1_obu_section5
+              aomdec_av1_webm"
+
+run_tests aomdec_verify_environment "${aomdec_tests}"
diff --git a/third_party/aom/test/aomenc.sh b/third_party/aom/test/aomenc.sh
new file mode 100755
index 0000000000..b030397a30
--- /dev/null
+++ b/third_party/aom/test/aomenc.sh
@@ -0,0 +1,269 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests aomenc using hantro_collage_w352h288.yuv as input. To add
+## new tests to this file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to aomenc_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: Make sure input is available.
+aomenc_verify_environment() {
+  if [ ! -e "${YUV_RAW_INPUT}" ]; then
+    elog "The file ${YUV_RAW_INPUT##*/} must exist in LIBAOM_TEST_DATA_PATH."
+    return 1
+  fi
+  if [ "$(aomenc_can_encode_av1)" = "yes" ]; then
+    if [ ! -e "${Y4M_NOSQ_PAR_INPUT}" ]; then
+      elog "The file ${Y4M_NOSQ_PAR_INPUT##*/} must exist in"
+      elog "LIBAOM_TEST_DATA_PATH."
+      return 1
+    fi
+  fi
+  if [ -z "$(aom_tool_path aomenc)" ]; then
+    elog "aomenc not found. It must exist in LIBAOM_BIN_PATH or its parent."
+    return 1
+  fi
+}
+
+aomenc_can_encode_av1() {
+  if [ "$(av1_encode_available)" = "yes" ]; then
+    echo yes
+  fi
+}
+
+aomenc_can_encode_av1() {
+  if [ "$(av1_encode_available)" = "yes" ]; then
+    echo yes
+  fi
+}
+
+# Utilities that echo aomenc input file parameters.
+y4m_input_non_square_par() {
+  echo ""${Y4M_NOSQ_PAR_INPUT}""
+}
+
+y4m_input_720p() {
+  echo ""${Y4M_720P_INPUT}""
+}
+
+# Wrapper function for running aomenc with pipe input. Requires that
+# LIBAOM_BIN_PATH points to the directory containing aomenc. $1 is used as the
+# input file path and shifted away. All remaining parameters are passed through
+# to aomenc.
+aomenc_pipe() {
+  local encoder="$(aom_tool_path aomenc)"
+  local input="$1"
+  shift
+  cat "${input}" | eval "${AOM_TEST_PREFIX}" "${encoder}" - \
+    --test-decode=fatal \
+    "$@" ${devnull}
+}
+
+# Wrapper function for running aomenc. Requires that LIBAOM_BIN_PATH points to
+# the directory containing aomenc. $1 one is used as the input file path and
+# shifted away. All remaining parameters are passed through to aomenc.
+aomenc() {
+  local encoder="$(aom_tool_path aomenc)"
+  local input="$1"
+  shift
+  eval "${AOM_TEST_PREFIX}" "${encoder}" "${input}" \
+    --test-decode=fatal \
+    "$@" ${devnull}
+}
+
+aomenc_av1_ivf() {
+  if [ "$(aomenc_can_encode_av1)" = "yes" ]; then
+    local output="${AV1_IVF_FILE}"
+    if [ -e "${AV1_IVF_FILE}" ]; then
+      output="${AOM_TEST_OUTPUT_DIR}/av1_test.ivf"
+    fi
+    aomenc $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --ivf \
+      --output="${output}"
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+aomenc_av1_obu_annexb() {
+   if [ "$(aomenc_can_encode_av1)" = "yes" ]; then
+    local output="${AV1_OBU_ANNEXB_FILE}"
+    if [ -e "${AV1_OBU_ANNEXB_FILE}" ]; then
+      output="${AOM_TEST_OUTPUT_DIR}/av1_test.annexb.obu"
+    fi
+    aomenc $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --obu \
+      --annexb=1 \
+      --output="${output}"
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+aomenc_av1_obu_section5() {
+   if [ "$(aomenc_can_encode_av1)" = "yes" ]; then
+    local output="${AV1_OBU_SEC5_FILE}"
+    if [ -e "${AV1_OBU_SEC5_FILE}" ]; then
+      output="${AOM_TEST_OUTPUT_DIR}/av1_test.section5.obu"
+    fi
+    aomenc $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --obu \
+      --output="${output}"
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+aomenc_av1_webm() {
+  if [ "$(aomenc_can_encode_av1)" = "yes" ] && \
+     [ "$(webm_io_available)" = "yes" ]; then
+    local output="${AV1_WEBM_FILE}"
+    if [ -e "${AV1_WEBM_FILE}" ]; then
+      output="${AOM_TEST_OUTPUT_DIR}/av1_test.webm"
+    fi
+    aomenc $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --output="${output}"
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+aomenc_av1_webm_1pass() {
+  if [ "$(aomenc_can_encode_av1)" = "yes" ] && \
+     [ "$(webm_io_available)" = "yes" ]; then
+    local output="${AOM_TEST_OUTPUT_DIR}/av1_test.webm"
+    aomenc $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --passes=1 \
+      --output="${output}"
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+aomenc_av1_ivf_lossless() {
+  if [ "$(aomenc_can_encode_av1)" = "yes" ]; then
+    local output="${AOM_TEST_OUTPUT_DIR}/av1_lossless.ivf"
+    aomenc $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --ivf \
+      --output="${output}" \
+      --lossless=1
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+aomenc_av1_ivf_minq0_maxq0() {
+  if [ "$(aomenc_can_encode_av1)" = "yes" ]; then
+    local output="${AOM_TEST_OUTPUT_DIR}/av1_lossless_minq0_maxq0.ivf"
+    aomenc $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --ivf \
+      --output="${output}" \
+      --min-q=0 \
+      --max-q=0
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+aomenc_av1_webm_lag5_frames10() {
+  if [ "$(aomenc_can_encode_av1)" = "yes" ] && \
+     [ "$(webm_io_available)" = "yes" ]; then
+    local lag_total_frames=10
+    local lag_frames=5
+    local output="${AOM_TEST_OUTPUT_DIR}/av1_lag5_frames10.webm"
+    aomenc $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --limit=${lag_total_frames} \
+      --lag-in-frames=${lag_frames} \
+      --output="${output}"
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+# TODO(fgalligan): Test that DisplayWidth is different than video width.
+aomenc_av1_webm_non_square_par() {
+  if [ "$(aomenc_can_encode_av1)" = "yes" ] && \
+     [ "$(webm_io_available)" = "yes" ]; then
+    local output="${AOM_TEST_OUTPUT_DIR}/av1_non_square_par.webm"
+    aomenc $(y4m_input_non_square_par) \
+      $(aomenc_encode_test_fast_params) \
+      --output="${output}"
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+aomenc_av1_webm_cdf_update_mode() {
+  if [ "$(aomenc_can_encode_av1)" = "yes" ] && \
+     [ "$(webm_io_available)" = "yes" ]; then
+    for mode in 0 1 2; do
+      local output="${AOM_TEST_OUTPUT_DIR}/cdf_mode_${mode}.webm"
+      aomenc $(yuv_raw_input) \
+        $(aomenc_encode_test_fast_params) \
+        --cdf-update-mode=${mode} \
+        --output="${output}"
+
+      if [ ! -e "${output}" ]; then
+        elog "Output file does not exist."
+        return 1
+      fi
+    done
+  fi
+}
+
+aomenc_tests="aomenc_av1_ivf
+              aomenc_av1_obu_annexb
+              aomenc_av1_obu_section5
+              aomenc_av1_webm
+              aomenc_av1_webm_1pass
+              aomenc_av1_ivf_lossless
+              aomenc_av1_ivf_minq0_maxq0
+              aomenc_av1_webm_lag5_frames10
+              aomenc_av1_webm_non_square_par
+              aomenc_av1_webm_cdf_update_mode"
+
+run_tests aomenc_verify_environment "${aomenc_tests}"
diff --git a/third_party/aom/test/aq_segment_test.cc b/third_party/aom/test/aq_segment_test.cc
new file mode 100644
index 0000000000..bbb5027d4d
--- /dev/null
+++ b/third_party/aom/test/aq_segment_test.cc
@@ -0,0 +1,108 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "config/aom_config.h"
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+class AqSegmentTest
+    : public ::libaom_test::CodecTestWith2Params<libaom_test::TestMode, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  AqSegmentTest() : EncoderTest(GET_PARAM(0)) {}
+  virtual ~AqSegmentTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+    set_cpu_used_ = GET_PARAM(2);
+    aq_mode_ = 0;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
+      encoder->Control(AV1E_SET_AQ_MODE, aq_mode_);
+      encoder->Control(AV1E_SET_DELTAQ_MODE, deltaq_mode_);
+      encoder->Control(AOME_SET_MAX_INTRA_BITRATE_PCT, 100);
+    }
+  }
+
+  void DoTest(int aq_mode) {
+    aq_mode_ = aq_mode;
+    deltaq_mode_ = 0;
+    cfg_.kf_max_dist = 12;
+    cfg_.rc_min_quantizer = 8;
+    cfg_.rc_max_quantizer = 56;
+    cfg_.rc_end_usage = AOM_CBR;
+    cfg_.g_lag_in_frames = 6;
+    cfg_.rc_buf_initial_sz = 500;
+    cfg_.rc_buf_optimal_sz = 500;
+    cfg_.rc_buf_sz = 1000;
+    cfg_.rc_target_bitrate = 300;
+    ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352,
+                                         288, 30, 1, 0, 15);
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  }
+
+  int set_cpu_used_;
+  int aq_mode_;
+  int deltaq_mode_;
+};
+
+// Validate that this AQ segmentation mode (AQ=1, variance_ap)
+// encodes and decodes without a mismatch.
+TEST_P(AqSegmentTest, TestNoMisMatchAQ1) { DoTest(1); }
+
+// Validate that this AQ segmentation mode (AQ=2, complexity_aq)
+// encodes and decodes without a mismatch.
+TEST_P(AqSegmentTest, TestNoMisMatchAQ2) { DoTest(2); }
+
+// Validate that this AQ segmentation mode (AQ=3, cyclic_refresh_aq)
+// encodes and decodes without a mismatch.
+TEST_P(AqSegmentTest, TestNoMisMatchAQ3) { DoTest(3); }
+
+class AqSegmentTestLarge : public AqSegmentTest {};
+
+TEST_P(AqSegmentTestLarge, TestNoMisMatchAQ1) { DoTest(1); }
+
+TEST_P(AqSegmentTestLarge, TestNoMisMatchAQ2) { DoTest(2); }
+
+TEST_P(AqSegmentTestLarge, TestNoMisMatchAQ3) { DoTest(3); }
+
+// Validate that this delta q mode
+// encodes and decodes without a mismatch.
+TEST_P(AqSegmentTest, TestNoMisMatchExtDeltaQ) {
+  cfg_.rc_end_usage = AOM_CQ;
+  aq_mode_ = 0;
+  deltaq_mode_ = 2;
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 15);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
+AV1_INSTANTIATE_TEST_CASE(AqSegmentTest,
+                          ::testing::Values(::libaom_test::kRealTime,
+                                            ::libaom_test::kOnePassGood),
+                          ::testing::Range(5, 9));
+AV1_INSTANTIATE_TEST_CASE(AqSegmentTestLarge,
+                          ::testing::Values(::libaom_test::kRealTime,
+                                            ::libaom_test::kOnePassGood),
+                          ::testing::Range(3, 5));
+}  // namespace
diff --git a/third_party/aom/test/arf_freq_test.cc b/third_party/aom/test/arf_freq_test.cc
new file mode 100644
index 0000000000..083f4022f4
--- /dev/null
+++ b/third_party/aom/test/arf_freq_test.cc
@@ -0,0 +1,223 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+#include "test/yuv_video_source.h"
+#include "av1/encoder/ratectrl.h"
+
+namespace {
+
+const unsigned int kFrames = 100;
+const int kBitrate = 500;
+
+#define ARF_NOT_SEEN 1000001
+#define ARF_SEEN_ONCE 1000000
+
+typedef struct {
+  const char *filename;
+  unsigned int width;
+  unsigned int height;
+  unsigned int framerate_num;
+  unsigned int framerate_den;
+  unsigned int input_bit_depth;
+  aom_img_fmt fmt;
+  aom_bit_depth_t bit_depth;
+  unsigned int profile;
+} TestVideoParam;
+
+typedef struct {
+  libaom_test::TestMode mode;
+  int cpu_used;
+} TestEncodeParam;
+
+const TestVideoParam kTestVectors[] = {
+  // artificially increase framerate to trigger default check
+  { "hantro_collage_w352h288.yuv", 352, 288, 5000, 1, 8, AOM_IMG_FMT_I420,
+    AOM_BITS_8, 0 },
+  { "hantro_collage_w352h288.yuv", 352, 288, 30, 1, 8, AOM_IMG_FMT_I420,
+    AOM_BITS_8, 0 },
+  { "rush_hour_444.y4m", 352, 288, 30, 1, 8, AOM_IMG_FMT_I444, AOM_BITS_8, 1 },
+  // Add list of profile 2/3 test videos here ...
+};
+
+const TestEncodeParam kEncodeVectors[] = {
+  { ::libaom_test::kOnePassGood, 2 }, { ::libaom_test::kOnePassGood, 5 },
+  { ::libaom_test::kTwoPassGood, 1 }, { ::libaom_test::kTwoPassGood, 2 },
+  { ::libaom_test::kTwoPassGood, 5 }, { ::libaom_test::kRealTime, 5 },
+};
+
+const int kMinArfVectors[] = {
+  // NOTE: 0 refers to the default built-in logic in:
+  //       av1_rc_get_default_min_gf_interval(...)
+  0, 4, 8, 12, 15
+};
+
+int is_extension_y4m(const char *filename) {
+  const char *dot = strrchr(filename, '.');
+  if (!dot || dot == filename)
+    return 0;
+  else
+    return !strcmp(dot, ".y4m");
+}
+
+class ArfFreqTestLarge
+    : public ::libaom_test::CodecTestWith3Params<TestVideoParam,
+                                                 TestEncodeParam, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  ArfFreqTestLarge()
+      : EncoderTest(GET_PARAM(0)), test_video_param_(GET_PARAM(1)),
+        test_encode_param_(GET_PARAM(2)), min_arf_requested_(GET_PARAM(3)) {}
+
+  virtual ~ArfFreqTestLarge() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(test_encode_param_.mode);
+    if (test_encode_param_.mode != ::libaom_test::kRealTime) {
+      cfg_.g_lag_in_frames = 25;
+      cfg_.rc_end_usage = AOM_VBR;
+    } else {
+      cfg_.g_lag_in_frames = 0;
+      cfg_.rc_end_usage = AOM_CBR;
+      cfg_.rc_buf_sz = 1000;
+      cfg_.rc_buf_initial_sz = 500;
+      cfg_.rc_buf_optimal_sz = 600;
+    }
+  }
+
+  virtual void BeginPassHook(unsigned int) {
+    min_run_ = ARF_NOT_SEEN;
+    run_of_visible_frames_ = 0;
+  }
+
+  int GetNumFramesInPkt(const aom_codec_cx_pkt_t *pkt) {
+    const uint8_t *buffer = reinterpret_cast<uint8_t *>(pkt->data.frame.buf);
+    const uint8_t marker = buffer[pkt->data.frame.sz - 1];
+    const int mag = ((marker >> 3) & 3) + 1;
+    int frames = (marker & 0x7) + 1;
+    const unsigned int index_sz = 2 + mag * frames;
+    // Check for superframe or not.
+    // Assume superframe has only one visible frame, the rest being
+    // invisible. If superframe index is not found, then there is only
+    // one frame.
+    if (!((marker & 0xe0) == 0xc0 && pkt->data.frame.sz >= index_sz &&
+          buffer[pkt->data.frame.sz - index_sz] == marker)) {
+      frames = 1;
+    }
+    return frames;
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (pkt->kind != AOM_CODEC_CX_FRAME_PKT) return;
+    const int frames = GetNumFramesInPkt(pkt);
+    if (frames == 1) {
+      run_of_visible_frames_++;
+    } else if (frames == 2) {
+      if (min_run_ == ARF_NOT_SEEN) {
+        min_run_ = ARF_SEEN_ONCE;
+      } else if (min_run_ == ARF_SEEN_ONCE ||
+                 run_of_visible_frames_ < min_run_) {
+        min_run_ = run_of_visible_frames_;
+      }
+      run_of_visible_frames_ = 1;
+    } else {
+      min_run_ = 0;
+      run_of_visible_frames_ = 1;
+    }
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 0) {
+      encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
+      encoder->Control(AV1E_SET_TILE_COLUMNS, 4);
+      encoder->Control(AOME_SET_CPUUSED, test_encode_param_.cpu_used);
+      encoder->Control(AV1E_SET_MIN_GF_INTERVAL, min_arf_requested_);
+      if (test_encode_param_.mode != ::libaom_test::kRealTime) {
+        encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+        encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+        encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+      }
+    }
+  }
+
+  int GetMinVisibleRun() const { return min_run_; }
+
+  int GetMinArfDistanceRequested() const {
+    if (min_arf_requested_)
+      return min_arf_requested_;
+    else
+      return av1_rc_get_default_min_gf_interval(
+          test_video_param_.width, test_video_param_.height,
+          (double)test_video_param_.framerate_num /
+              test_video_param_.framerate_den);
+  }
+
+  TestVideoParam test_video_param_;
+  TestEncodeParam test_encode_param_;
+
+ private:
+  int min_arf_requested_;
+  int min_run_;
+  int run_of_visible_frames_;
+};
+
+TEST_P(ArfFreqTestLarge, MinArfFreqTest) {
+  cfg_.rc_target_bitrate = kBitrate;
+  cfg_.g_error_resilient = 0;
+  cfg_.g_profile = test_video_param_.profile;
+  cfg_.g_input_bit_depth = test_video_param_.input_bit_depth;
+  cfg_.g_bit_depth = test_video_param_.bit_depth;
+  init_flags_ = AOM_CODEC_USE_PSNR;
+  if (cfg_.g_bit_depth > 8) init_flags_ |= AOM_CODEC_USE_HIGHBITDEPTH;
+
+  testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+  if (is_extension_y4m(test_video_param_.filename)) {
+    video.reset(new libaom_test::Y4mVideoSource(test_video_param_.filename, 0,
+                                                kFrames));
+  } else {
+    video.reset(new libaom_test::YUVVideoSource(
+        test_video_param_.filename, test_video_param_.fmt,
+        test_video_param_.width, test_video_param_.height,
+        test_video_param_.framerate_num, test_video_param_.framerate_den, 0,
+        kFrames));
+  }
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+  const int min_run = GetMinVisibleRun();
+  const int min_arf_dist_requested = GetMinArfDistanceRequested();
+  if (min_run != ARF_NOT_SEEN && min_run != ARF_SEEN_ONCE) {
+    const int min_arf_dist = min_run + 1;
+    EXPECT_GE(min_arf_dist, min_arf_dist_requested);
+  }
+}
+
+#if CONFIG_AV1_ENCODER
+// TODO(angiebird): 25-29 fail in high bitdepth mode.
+// TODO(zoeliu): This ArfFreqTest does not work with BWDREF_FRAME, as
+// BWDREF_FRAME is also a non-show frame, and the minimum run between two
+// consecutive BWDREF_FRAME's may vary between 1 and any arbitrary positive
+// number as long as it does not exceed the gf_group interval.
+INSTANTIATE_TEST_CASE_P(
+    DISABLED_AV1, ArfFreqTestLarge,
+    ::testing::Combine(
+        ::testing::Values(
+            static_cast<const libaom_test::CodecFactory *>(&libaom_test::kAV1)),
+        ::testing::ValuesIn(kTestVectors), ::testing::ValuesIn(kEncodeVectors),
+        ::testing::ValuesIn(kMinArfVectors)));
+#endif  // CONFIG_AV1_ENCODER
+}  // namespace
diff --git a/third_party/aom/test/av1_config_test.cc b/third_party/aom/test/av1_config_test.cc
new file mode 100644
index 0000000000..e2f2c53906
--- /dev/null
+++ b/third_party/aom/test/av1_config_test.cc
@@ -0,0 +1,164 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include <string.h>
+
+#include "common/av1_config.h"
+#include "test/util.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+
+//
+// Input buffers containing exactly one Sequence Header OBU.
+//
+// Each buffer is named according to the OBU storage format (Annex-B vs Low
+// Overhead Bitstream Format) and the type of Sequence Header OBU ("Full"
+// Sequence Header OBUs vs Sequence Header OBUs with the
+// reduced_still_image_flag set).
+//
+const uint8_t kAnnexBFullSequenceHeaderObu[] = {
+  0x0c, 0x08, 0x00, 0x00, 0x00, 0x04, 0x45, 0x7e, 0x3e, 0xff, 0xfc, 0xc0, 0x20
+};
+const uint8_t kAnnexBReducedStillImageSequenceHeaderObu[] = {
+  0x08, 0x08, 0x18, 0x22, 0x2b, 0xf1, 0xfe, 0xc0, 0x20
+};
+
+const uint8_t kLobfFullSequenceHeaderObu[] = {
+  0x0a, 0x0b, 0x00, 0x00, 0x00, 0x04, 0x45, 0x7e, 0x3e, 0xff, 0xfc, 0xc0, 0x20
+};
+
+const uint8_t kLobfReducedStillImageSequenceHeaderObu[] = {
+  0x0a, 0x07, 0x18, 0x22, 0x2b, 0xf1, 0xfe, 0xc0, 0x20
+};
+
+const uint8_t kAv1cAllZero[] = { 0, 0, 0, 0 };
+
+// The size of AV1 config when no configOBUs are present at the end of the
+// configuration structure.
+const size_t kAv1cNoConfigObusSize = 4;
+
+bool VerifyAv1c(const uint8_t *const obu_buffer, size_t obu_buffer_length,
+                bool is_annexb) {
+  Av1Config av1_config;
+  memset(&av1_config, 0, sizeof(av1_config));
+  bool parse_ok = get_av1config_from_obu(obu_buffer, obu_buffer_length,
+                                         is_annexb, &av1_config) == 0;
+  if (parse_ok) {
+    EXPECT_EQ(1, av1_config.marker);
+    EXPECT_EQ(1, av1_config.version);
+    EXPECT_EQ(0, av1_config.seq_profile);
+    EXPECT_EQ(0, av1_config.seq_level_idx_0);
+    EXPECT_EQ(0, av1_config.seq_tier_0);
+    EXPECT_EQ(0, av1_config.high_bitdepth);
+    EXPECT_EQ(0, av1_config.twelve_bit);
+    EXPECT_EQ(0, av1_config.monochrome);
+    EXPECT_EQ(1, av1_config.chroma_subsampling_x);
+    EXPECT_EQ(1, av1_config.chroma_subsampling_y);
+    EXPECT_EQ(0, av1_config.chroma_sample_position);
+    EXPECT_EQ(0, av1_config.initial_presentation_delay_present);
+    EXPECT_EQ(0, av1_config.initial_presentation_delay_minus_one);
+  }
+  return parse_ok && ::testing::Test::HasFailure() == false;
+}
+
+TEST(Av1Config, ObuInvalidInputs) {
+  Av1Config av1_config;
+  memset(&av1_config, 0, sizeof(av1_config));
+  ASSERT_EQ(-1, get_av1config_from_obu(NULL, 0, 0, NULL));
+  ASSERT_EQ(-1,
+            get_av1config_from_obu(&kLobfFullSequenceHeaderObu[0], 0, 0, NULL));
+  ASSERT_EQ(
+      -1, get_av1config_from_obu(&kLobfFullSequenceHeaderObu[0],
+                                 sizeof(kLobfFullSequenceHeaderObu), 0, NULL));
+  ASSERT_EQ(-1, get_av1config_from_obu(NULL, sizeof(kLobfFullSequenceHeaderObu),
+                                       0, NULL));
+  ASSERT_EQ(-1, get_av1config_from_obu(&kLobfFullSequenceHeaderObu[0], 0, 0,
+                                       &av1_config));
+}
+
+TEST(Av1Config, ReadInvalidInputs) {
+  Av1Config av1_config;
+  memset(&av1_config, 0, sizeof(av1_config));
+  size_t bytes_read = 0;
+  ASSERT_EQ(-1, read_av1config(NULL, 0, NULL, NULL));
+  ASSERT_EQ(-1, read_av1config(NULL, 4, NULL, NULL));
+  ASSERT_EQ(-1, read_av1config(&kAv1cAllZero[0], 0, NULL, NULL));
+  ASSERT_EQ(-1, read_av1config(&kAv1cAllZero[0], 4, &bytes_read, NULL));
+  ASSERT_EQ(-1, read_av1config(NULL, 4, &bytes_read, &av1_config));
+}
+
+TEST(Av1Config, WriteInvalidInputs) {
+  Av1Config av1_config;
+  memset(&av1_config, 0, sizeof(av1_config));
+  size_t bytes_written = 0;
+  uint8_t av1c_buffer[4] = { 0 };
+  ASSERT_EQ(-1, write_av1config(NULL, 0, NULL, NULL));
+  ASSERT_EQ(-1, write_av1config(&av1_config, 0, NULL, NULL));
+  ASSERT_EQ(-1, write_av1config(&av1_config, 0, &bytes_written, NULL));
+
+  ASSERT_EQ(-1,
+            write_av1config(&av1_config, 0, &bytes_written, &av1c_buffer[0]));
+  ASSERT_EQ(-1, write_av1config(&av1_config, 4, &bytes_written, NULL));
+}
+
+TEST(Av1Config, GetAv1ConfigFromLobfObu) {
+  // Test parsing of a Sequence Header OBU with the reduced_still_picture_header
+  // unset-- aka a full Sequence Header OBU.
+  ASSERT_TRUE(VerifyAv1c(kLobfFullSequenceHeaderObu,
+                         sizeof(kLobfFullSequenceHeaderObu), false));
+
+  // Test parsing of a reduced still image Sequence Header OBU.
+  ASSERT_TRUE(VerifyAv1c(kLobfReducedStillImageSequenceHeaderObu,
+                         sizeof(kLobfReducedStillImageSequenceHeaderObu),
+                         false));
+}
+
+TEST(Av1Config, GetAv1ConfigFromAnnexBObu) {
+  // Test parsing of a Sequence Header OBU with the reduced_still_picture_header
+  // unset-- aka a full Sequence Header OBU.
+  ASSERT_TRUE(VerifyAv1c(kAnnexBFullSequenceHeaderObu,
+                         sizeof(kAnnexBFullSequenceHeaderObu), true));
+
+  // Test parsing of a reduced still image Sequence Header OBU.
+  ASSERT_TRUE(VerifyAv1c(kAnnexBReducedStillImageSequenceHeaderObu,
+                         sizeof(kAnnexBReducedStillImageSequenceHeaderObu),
+                         true));
+}
+
+TEST(Av1Config, ReadWriteConfig) {
+  Av1Config av1_config;
+  memset(&av1_config, 0, sizeof(av1_config));
+
+  // Test writing out the AV1 config.
+  size_t bytes_written = 0;
+  uint8_t av1c_buffer[4] = { 0 };
+  ASSERT_EQ(0, write_av1config(&av1_config, sizeof(av1c_buffer), &bytes_written,
+                               &av1c_buffer[0]));
+  ASSERT_EQ(kAv1cNoConfigObusSize, bytes_written);
+  for (size_t i = 0; i < kAv1cNoConfigObusSize; ++i) {
+    ASSERT_EQ(kAv1cAllZero[i], av1c_buffer[i])
+        << "Mismatch in output Av1Config at offset=" << i;
+  }
+
+  // Test reading the AV1 config.
+  size_t bytes_read = 0;
+  ASSERT_EQ(0, read_av1config(&kAv1cAllZero[0], sizeof(kAv1cAllZero),
+                              &bytes_read, &av1_config));
+  ASSERT_EQ(kAv1cNoConfigObusSize, bytes_read);
+  ASSERT_EQ(0, write_av1config(&av1_config, sizeof(av1c_buffer), &bytes_written,
+                               &av1c_buffer[0]));
+  for (size_t i = 0; i < kAv1cNoConfigObusSize; ++i) {
+    ASSERT_EQ(kAv1cAllZero[i], av1c_buffer[i])
+        << "Mismatch in output Av1Config at offset=" << i;
+  }
+}
+
+}  // namespace
diff --git a/third_party/aom/test/av1_convolve_2d_test.cc b/third_party/aom/test/av1_convolve_2d_test.cc
new file mode 100644
index 0000000000..03286260e8
--- /dev/null
+++ b/third_party/aom/test/av1_convolve_2d_test.cc
@@ -0,0 +1,249 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/av1_convolve_2d_test_util.h"
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+using libaom_test::AV1Convolve2D::AV1Convolve2DSrTest;
+using libaom_test::AV1Convolve2D::AV1JntConvolve2DTest;
+using libaom_test::AV1HighbdConvolve2D::AV1HighbdConvolve2DSrTest;
+using libaom_test::AV1HighbdConvolve2D::AV1HighbdJntConvolve2DTest;
+namespace {
+
+TEST_P(AV1Convolve2DSrTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(0)); }
+
+TEST_P(AV1Convolve2DSrTest, CheckOutput) { RunCheckOutput(GET_PARAM(0)); }
+
+INSTANTIATE_TEST_CASE_P(
+    C_COPY, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_2d_copy_sr_c, 0, 0));
+INSTANTIATE_TEST_CASE_P(
+    C_X, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_x_sr_c, 1, 0));
+INSTANTIATE_TEST_CASE_P(
+    C_Y, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_y_sr_c, 0, 1));
+INSTANTIATE_TEST_CASE_P(
+    C, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_2d_sr_c, 1, 1));
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2_COPY, AV1Convolve2DSrTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_convolve_2d_copy_sr_sse2, 0, 0));
+INSTANTIATE_TEST_CASE_P(
+    SSE2_X, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_x_sr_sse2, 1, 0));
+INSTANTIATE_TEST_CASE_P(
+    SSE2_Y, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_y_sr_sse2, 0, 1));
+INSTANTIATE_TEST_CASE_P(
+    SSE2, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_2d_sr_sse2, 1, 1));
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2_COPY, AV1Convolve2DSrTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_convolve_2d_copy_sr_avx2, 0, 0));
+INSTANTIATE_TEST_CASE_P(
+    AVX2_X, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_x_sr_avx2, 1, 0));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2_Y, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_y_sr_avx2, 0, 1));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_2d_sr_avx2, 1, 1));
+#endif  // HAVE_AVX2
+#endif  // HAVE_SSE2
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(
+    NEON_X, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_x_sr_neon, 1, 0));
+
+INSTANTIATE_TEST_CASE_P(
+    NEON_Y, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_y_sr_neon, 0, 1));
+
+INSTANTIATE_TEST_CASE_P(
+    NEON, AV1Convolve2DSrTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_convolve_2d_sr_neon, 1, 1));
+
+INSTANTIATE_TEST_CASE_P(NEON_COPY, AV1Convolve2DSrTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_convolve_2d_copy_sr_neon, 0, 0));
+#endif  // HAVE_NEON
+
+TEST_P(AV1JntConvolve2DTest, CheckOutput) { RunCheckOutput(GET_PARAM(0)); }
+TEST_P(AV1JntConvolve2DTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(0)); }
+
+INSTANTIATE_TEST_CASE_P(
+    C_COPY, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_2d_copy_c, 0, 0));
+
+INSTANTIATE_TEST_CASE_P(
+    C_X, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_x_c, 1, 0));
+
+INSTANTIATE_TEST_CASE_P(
+    C_Y, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_y_c, 0, 1));
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2_COPY, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_jnt_convolve_2d_copy_sse2, 0, 0));
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE2_X, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_x_sse2, 1, 0));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2_Y, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_y_sse2, 0, 1));
+
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_2d_ssse3, 1, 1));
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2_COPY, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_jnt_convolve_2d_copy_avx2, 0, 0));
+INSTANTIATE_TEST_CASE_P(
+    AVX2_X, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_x_avx2, 1, 0));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2_Y, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_y_avx2, 0, 1));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_2d_avx2, 1, 1));
+#endif  // HAVE_AVX2
+#endif  // HAVE_SSE4_1
+#endif  // HAVE_SSE2
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON_COPY, AV1JntConvolve2DTest,
+                        libaom_test::AV1Convolve2D::BuildParams(
+                            av1_jnt_convolve_2d_copy_neon, 0, 0));
+
+INSTANTIATE_TEST_CASE_P(
+    NEON, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_2d_neon, 1, 1));
+INSTANTIATE_TEST_CASE_P(
+    NEON_X, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_x_neon, 1, 0));
+
+INSTANTIATE_TEST_CASE_P(
+    NEON_Y, AV1JntConvolve2DTest,
+    libaom_test::AV1Convolve2D::BuildParams(av1_jnt_convolve_y_neon, 0, 1));
+#endif  // HAVE_NEON
+
+TEST_P(AV1HighbdConvolve2DSrTest, CheckOutput) { RunCheckOutput(GET_PARAM(1)); }
+TEST_P(AV1HighbdConvolve2DSrTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(1));
+}
+
+INSTANTIATE_TEST_CASE_P(C_X, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_x_sr_c, 1, 0));
+
+INSTANTIATE_TEST_CASE_P(C_Y, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_y_sr_c, 0, 1));
+
+INSTANTIATE_TEST_CASE_P(C_COPY, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_2d_copy_sr_c, 0, 0));
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2_COPY, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_2d_copy_sr_sse2, 0, 0));
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(SSSE3, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_2d_sr_ssse3, 1, 1));
+INSTANTIATE_TEST_CASE_P(SSSE3_X, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_x_sr_ssse3, 1, 0));
+INSTANTIATE_TEST_CASE_P(SSSE3_Y, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_y_sr_ssse3, 0, 1));
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_2d_sr_avx2, 1, 1));
+INSTANTIATE_TEST_CASE_P(AVX2_X, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_x_sr_avx2, 1, 0));
+INSTANTIATE_TEST_CASE_P(AVX2_Y, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_y_sr_avx2, 0, 1));
+INSTANTIATE_TEST_CASE_P(AVX2_COPY, AV1HighbdConvolve2DSrTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_convolve_2d_copy_sr_avx2, 0, 0));
+#endif  // HAVE_AVX2
+#endif  // HAVE_SSSE3
+#endif  // HAVE_SSE2
+TEST_P(AV1HighbdJntConvolve2DTest, CheckOutput) {
+  RunCheckOutput(GET_PARAM(1));
+}
+
+TEST_P(AV1HighbdJntConvolve2DTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(1));
+}
+
+INSTANTIATE_TEST_CASE_P(C_X, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_x_c, 1, 0));
+
+INSTANTIATE_TEST_CASE_P(C_Y, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_y_c, 0, 1));
+
+INSTANTIATE_TEST_CASE_P(C_COPY, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_2d_copy_c, 0, 0));
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1_COPY, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_2d_copy_sse4_1, 0, 0));
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_2d_sse4_1, 1, 1));
+INSTANTIATE_TEST_CASE_P(SSE4_1_X, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_x_sse4_1, 1, 0));
+INSTANTIATE_TEST_CASE_P(SSE4_1_Y, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_y_sse4_1, 0, 1));
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2_COPY, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_2d_copy_avx2, 0, 0));
+INSTANTIATE_TEST_CASE_P(AVX2, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_2d_avx2, 1, 1));
+INSTANTIATE_TEST_CASE_P(AVX2_X, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_x_avx2, 1, 0));
+INSTANTIATE_TEST_CASE_P(AVX2_Y, AV1HighbdJntConvolve2DTest,
+                        libaom_test::AV1HighbdConvolve2D::BuildParams(
+                            av1_highbd_jnt_convolve_y_avx2, 0, 1));
+#endif  // HAVE_AVX2
+#endif  // HAVE_SSE4_1
+}  // namespace
diff --git a/third_party/aom/test/av1_convolve_2d_test_util.cc b/third_party/aom/test/av1_convolve_2d_test_util.cc
new file mode 100644
index 0000000000..409fd23e10
--- /dev/null
+++ b/third_party/aom/test/av1_convolve_2d_test_util.cc
@@ -0,0 +1,705 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "test/av1_convolve_2d_test_util.h"
+
+#include "aom_ports/aom_timer.h"
+#include "av1/common/common_data.h"
+#include "av1/common/convolve.h"
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+
+namespace libaom_test {
+
+const int kMaxSize = 128 + 32;  // padding
+namespace AV1Convolve2D {
+
+::testing::internal::ParamGenerator<Convolve2DParam> BuildParams(
+    convolve_2d_func filter, int has_subx, int has_suby) {
+  return ::testing::Combine(::testing::Values(filter),
+                            ::testing::Values(has_subx),
+                            ::testing::Values(has_suby),
+                            ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
+}
+
+AV1Convolve2DSrTest::~AV1Convolve2DSrTest() {}
+void AV1Convolve2DSrTest::SetUp() {
+  rnd_.Reset(ACMRandom::DeterministicSeed());
+}
+
+void AV1Convolve2DSrTest::TearDown() { libaom_test::ClearSystemState(); }
+
+void AV1Convolve2DSrTest::RunCheckOutput(convolve_2d_func test_impl) {
+  const int w = kMaxSize, h = kMaxSize;
+  const int has_subx = GET_PARAM(1);
+  const int has_suby = GET_PARAM(2);
+  const int block_idx = GET_PARAM(3);
+  int hfilter, vfilter, subx, suby;
+  uint8_t input[kMaxSize * kMaxSize];
+  DECLARE_ALIGNED(32, uint8_t, output[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, output2[MAX_SB_SQUARE]);
+
+  for (int i = 0; i < h; ++i)
+    for (int j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand8();
+  for (int i = 0; i < MAX_SB_SQUARE; ++i)
+    output[i] = output2[i] = rnd_.Rand31();
+
+  // Make sure that sizes 2xN and Nx2 are also tested for chroma.
+  const int num_sizes =
+      (block_size_wide[block_idx] == 4 || block_size_high[block_idx] == 4) ? 2
+                                                                           : 1;
+  for (int shift = 0; shift < num_sizes; ++shift) {  // luma and chroma
+    const int out_w = block_size_wide[block_idx] >> shift;
+    const int out_h = block_size_high[block_idx] >> shift;
+    for (hfilter = EIGHTTAP_REGULAR; hfilter < INTERP_FILTERS_ALL; ++hfilter) {
+      for (vfilter = EIGHTTAP_REGULAR; vfilter < INTERP_FILTERS_ALL;
+           ++vfilter) {
+        const InterpFilterParams *filter_params_x =
+            av1_get_interp_filter_params_with_block_size((InterpFilter)hfilter,
+                                                         out_w);
+        const InterpFilterParams *filter_params_y =
+            av1_get_interp_filter_params_with_block_size((InterpFilter)vfilter,
+                                                         out_h);
+        for (int do_average = 0; do_average < 1; ++do_average) {
+          ConvolveParams conv_params1 =
+              get_conv_params_no_round(do_average, 0, NULL, 0, 0, 8);
+          ConvolveParams conv_params2 =
+              get_conv_params_no_round(do_average, 0, NULL, 0, 0, 8);
+
+          const int subx_range = has_subx ? 16 : 1;
+          const int suby_range = has_suby ? 16 : 1;
+          for (subx = 0; subx < subx_range; ++subx) {
+            for (suby = 0; suby < suby_range; ++suby) {
+              // Choose random locations within the source block
+              const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+              const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+              av1_convolve_2d_sr_c(input + offset_r * w + offset_c, w, output,
+                                   MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                                   filter_params_y, subx, suby, &conv_params1);
+              test_impl(input + offset_r * w + offset_c, w, output2,
+                        MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                        filter_params_y, subx, suby, &conv_params2);
+
+              if (memcmp(output, output2, sizeof(output))) {
+                for (int i = 0; i < MAX_SB_SIZE; ++i) {
+                  for (int j = 0; j < MAX_SB_SIZE; ++j) {
+                    int idx = i * MAX_SB_SIZE + j;
+                    ASSERT_EQ(output[idx], output2[idx])
+                        << out_w << "x" << out_h << " Pixel mismatch at index "
+                        << idx << " = (" << i << ", " << j
+                        << "), sub pixel offset = (" << suby << ", " << subx
+                        << ")";
+                  }
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+void AV1Convolve2DSrTest::RunSpeedTest(convolve_2d_func test_impl) {
+  const int w = kMaxSize, h = kMaxSize;
+  const int has_subx = GET_PARAM(1);
+  const int has_suby = GET_PARAM(2);
+  const int block_idx = GET_PARAM(3);
+
+  uint8_t input[kMaxSize * kMaxSize];
+  DECLARE_ALIGNED(32, uint8_t, output[MAX_SB_SQUARE]);
+
+  for (int i = 0; i < h; ++i)
+    for (int j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand8();
+
+  int hfilter = EIGHTTAP_REGULAR, vfilter = EIGHTTAP_REGULAR;
+  int subx = 0, suby = 0;
+
+  const int do_average = 0;
+  ConvolveParams conv_params2 =
+      get_conv_params_no_round(do_average, 0, NULL, 0, 0, 8);
+
+  // Make sure that sizes 2xN and Nx2 are also tested for chroma.
+  const int num_sizes =
+      (block_size_wide[block_idx] == 4 || block_size_high[block_idx] == 4) ? 2
+                                                                           : 1;
+  for (int shift = 0; shift < num_sizes; ++shift) {  // luma and chroma
+    const int out_w = block_size_wide[block_idx] >> shift;
+    const int out_h = block_size_high[block_idx] >> shift;
+    const int num_loops = 1000000000 / (out_w + out_h);
+
+    const InterpFilterParams *filter_params_x =
+        av1_get_interp_filter_params_with_block_size((InterpFilter)hfilter,
+                                                     out_w);
+    const InterpFilterParams *filter_params_y =
+        av1_get_interp_filter_params_with_block_size((InterpFilter)vfilter,
+                                                     out_h);
+
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+
+    for (int i = 0; i < num_loops; ++i)
+      test_impl(input, w, output, MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                filter_params_y, subx, suby, &conv_params2);
+
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+    printf("%d,%d convolve %3dx%-3d: %7.2f us\n", has_subx, has_suby, out_w,
+           out_h, 1000.0 * elapsed_time / num_loops);
+  }
+}
+
+AV1JntConvolve2DTest::~AV1JntConvolve2DTest() {}
+void AV1JntConvolve2DTest::SetUp() {
+  rnd_.Reset(ACMRandom::DeterministicSeed());
+}
+
+void AV1JntConvolve2DTest::TearDown() { libaom_test::ClearSystemState(); }
+
+void AV1JntConvolve2DTest::RunCheckOutput(convolve_2d_func test_impl) {
+  const int w = kMaxSize, h = kMaxSize;
+  const int has_subx = GET_PARAM(1);
+  const int has_suby = GET_PARAM(2);
+  const int block_idx = GET_PARAM(3);
+  int hfilter, vfilter, subx, suby;
+  uint8_t input[kMaxSize * kMaxSize];
+  DECLARE_ALIGNED(32, CONV_BUF_TYPE, output1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, CONV_BUF_TYPE, output2[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(16, uint8_t, output8_1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(16, uint8_t, output8_2[MAX_SB_SQUARE]);
+
+  for (int i = 0; i < h; ++i)
+    for (int j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand8();
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    output1[i] = output2[i] = rnd_.Rand16();
+    output8_1[i] = output8_2[i] = rnd_.Rand8();
+  }
+
+  const int out_w = block_size_wide[block_idx];
+  const int out_h = block_size_high[block_idx];
+  for (hfilter = EIGHTTAP_REGULAR; hfilter < INTERP_FILTERS_ALL; ++hfilter) {
+    for (vfilter = EIGHTTAP_REGULAR; vfilter < INTERP_FILTERS_ALL; ++vfilter) {
+      const InterpFilterParams *filter_params_x =
+          av1_get_interp_filter_params_with_block_size((InterpFilter)hfilter,
+                                                       out_w);
+      const InterpFilterParams *filter_params_y =
+          av1_get_interp_filter_params_with_block_size((InterpFilter)vfilter,
+                                                       out_h);
+      for (int do_average = 0; do_average <= 1; ++do_average) {
+        ConvolveParams conv_params1 =
+            get_conv_params_no_round(do_average, 0, output1, MAX_SB_SIZE, 1, 8);
+        ConvolveParams conv_params2 =
+            get_conv_params_no_round(do_average, 0, output2, MAX_SB_SIZE, 1, 8);
+
+        // Test special case where jnt_comp_avg is not used
+        conv_params1.use_jnt_comp_avg = 0;
+        conv_params2.use_jnt_comp_avg = 0;
+
+        const int subx_range = has_subx ? 16 : 1;
+        const int suby_range = has_suby ? 16 : 1;
+        for (subx = 0; subx < subx_range; ++subx) {
+          for (suby = 0; suby < suby_range; ++suby) {
+            // Choose random locations within the source block
+            const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+            const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+            av1_jnt_convolve_2d_c(input + offset_r * w + offset_c, w, output8_1,
+                                  MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                                  filter_params_y, subx, suby, &conv_params1);
+            test_impl(input + offset_r * w + offset_c, w, output8_2,
+                      MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                      filter_params_y, subx, suby, &conv_params2);
+
+            for (int i = 0; i < out_h; ++i) {
+              for (int j = 0; j < out_w; ++j) {
+                int idx = i * MAX_SB_SIZE + j;
+                ASSERT_EQ(output1[idx], output2[idx])
+                    << "Mismatch at unit tests for av1_jnt_convolve_2d\n"
+                    << out_w << "x" << out_h << " Pixel mismatch at index "
+                    << idx << " = (" << i << ", " << j
+                    << "), sub pixel offset = (" << suby << ", " << subx << ")";
+              }
+            }
+
+            if (memcmp(output8_1, output8_2, sizeof(output8_1))) {
+              for (int i = 0; i < MAX_SB_SIZE; ++i) {
+                for (int j = 0; j < MAX_SB_SIZE; ++j) {
+                  int idx = i * MAX_SB_SIZE + j;
+                  ASSERT_EQ(output8_1[idx], output8_2[idx])
+                      << out_w << "x" << out_h << " Pixel mismatch at index "
+                      << idx << " = (" << i << ", " << j
+                      << "), sub pixel offset = (" << suby << ", " << subx
+                      << ")";
+                }
+              }
+            }
+          }
+        }
+
+        // Test different combination of fwd and bck offset weights
+        for (int k = 0; k < 2; ++k) {
+          for (int l = 0; l < 4; ++l) {
+            conv_params1.use_jnt_comp_avg = 1;
+            conv_params2.use_jnt_comp_avg = 1;
+            conv_params1.fwd_offset = quant_dist_lookup_table[k][l][0];
+            conv_params1.bck_offset = quant_dist_lookup_table[k][l][1];
+            conv_params2.fwd_offset = quant_dist_lookup_table[k][l][0];
+            conv_params2.bck_offset = quant_dist_lookup_table[k][l][1];
+
+            for (subx = 0; subx < subx_range; ++subx) {
+              for (suby = 0; suby < suby_range; ++suby) {
+                // Choose random locations within the source block
+                const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+                const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+                av1_jnt_convolve_2d_c(input + offset_r * w + offset_c, w,
+                                      output8_1, MAX_SB_SIZE, out_w, out_h,
+                                      filter_params_x, filter_params_y, subx,
+                                      suby, &conv_params1);
+                test_impl(input + offset_r * w + offset_c, w, output8_2,
+                          MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                          filter_params_y, subx, suby, &conv_params2);
+
+                for (int i = 0; i < out_h; ++i) {
+                  for (int j = 0; j < out_w; ++j) {
+                    int idx = i * MAX_SB_SIZE + j;
+                    ASSERT_EQ(output1[idx], output2[idx])
+                        << "Mismatch at unit tests for "
+                           "av1_jnt_convolve_2d\n"
+                        << out_w << "x" << out_h << " Pixel mismatch at index "
+                        << idx << " = (" << i << ", " << j
+                        << "), sub pixel offset = (" << suby << ", " << subx
+                        << ")";
+                  }
+                }
+                if (memcmp(output8_1, output8_2, sizeof(output8_1))) {
+                  for (int i = 0; i < MAX_SB_SIZE; ++i) {
+                    for (int j = 0; j < MAX_SB_SIZE; ++j) {
+                      int idx = i * MAX_SB_SIZE + j;
+                      ASSERT_EQ(output8_1[idx], output8_2[idx])
+                          << out_w << "x" << out_h
+                          << " Pixel mismatch at index " << idx << " = (" << i
+                          << ", " << j << "), sub pixel offset = (" << suby
+                          << ", " << subx << ")";
+                    }
+                  }
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+void AV1JntConvolve2DTest::RunSpeedTest(convolve_2d_func test_impl) {
+  const int w = kMaxSize, h = kMaxSize;
+  const int has_subx = GET_PARAM(1);
+  const int has_suby = GET_PARAM(2);
+  const int block_idx = GET_PARAM(3);
+
+  int subx = 0, suby = 0;
+  uint8_t input[kMaxSize * kMaxSize];
+  DECLARE_ALIGNED(32, CONV_BUF_TYPE, output[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(16, uint8_t, output8[MAX_SB_SQUARE]);
+  int hfilter = EIGHTTAP_REGULAR, vfilter = EIGHTTAP_REGULAR;
+  for (int i = 0; i < h; ++i)
+    for (int j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand8();
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    output[i] = rnd_.Rand16();
+    output8[i] = rnd_.Rand8();
+  }
+
+  const int out_w = block_size_wide[block_idx];
+  const int out_h = block_size_high[block_idx];
+  const int num_loops = 1000000000 / (out_w + out_h);
+  const int do_average = 0;
+
+  const InterpFilterParams *filter_params_x =
+      av1_get_interp_filter_params_with_block_size((InterpFilter)hfilter,
+                                                   out_w);
+  const InterpFilterParams *filter_params_y =
+      av1_get_interp_filter_params_with_block_size((InterpFilter)vfilter,
+                                                   out_h);
+
+  ConvolveParams conv_params =
+      get_conv_params_no_round(do_average, 0, output, MAX_SB_SIZE, 1, 8);
+
+  conv_params.use_jnt_comp_avg = 0;
+
+  // Choose random locations within the source block
+  const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+  const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+
+  for (int i = 0; i < num_loops; ++i)
+    test_impl(input + offset_r * w + offset_c, w, output8, MAX_SB_SIZE, out_w,
+              out_h, filter_params_x, filter_params_y, subx, suby,
+              &conv_params);
+
+  aom_usec_timer_mark(&timer);
+  const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+  printf("%d,%d convolve %3dx%-3d: %7.2f us\n", has_subx, has_suby, out_w,
+         out_h, 1000.0 * elapsed_time / num_loops);
+}
+}  // namespace AV1Convolve2D
+
+namespace AV1HighbdConvolve2D {
+::testing::internal::ParamGenerator<HighbdConvolve2DParam> BuildParams(
+    highbd_convolve_2d_func filter, int has_subx, int has_suby) {
+  return ::testing::Combine(
+      ::testing::Range(8, 13, 2), ::testing::Values(filter),
+      ::testing::Values(has_subx), ::testing::Values(has_suby),
+      ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
+}
+
+AV1HighbdConvolve2DSrTest::~AV1HighbdConvolve2DSrTest() {}
+void AV1HighbdConvolve2DSrTest::SetUp() {
+  rnd_.Reset(ACMRandom::DeterministicSeed());
+}
+
+void AV1HighbdConvolve2DSrTest::TearDown() { libaom_test::ClearSystemState(); }
+
+void AV1HighbdConvolve2DSrTest::RunSpeedTest(
+    highbd_convolve_2d_func test_impl) {
+  const int w = kMaxSize, h = kMaxSize;
+  const int bd = GET_PARAM(0);
+  const int has_subx = GET_PARAM(2);
+  const int has_suby = GET_PARAM(3);
+  const int block_idx = GET_PARAM(4);
+  int hfilter, vfilter, subx, suby;
+  uint16_t input[kMaxSize * kMaxSize];
+  DECLARE_ALIGNED(32, uint16_t, output[MAX_SB_SQUARE]);
+
+  for (int i = 0; i < h; ++i)
+    for (int j = 0; j < w; ++j)
+      input[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+
+  hfilter = EIGHTTAP_REGULAR;
+  vfilter = EIGHTTAP_REGULAR;
+  int do_average = 0;
+
+  const int offset_r = 3;
+  const int offset_c = 3;
+  subx = 0;
+  suby = 0;
+
+  ConvolveParams conv_params =
+      get_conv_params_no_round(do_average, 0, NULL, 0, 0, bd);
+
+  // Make sure that sizes 2xN and Nx2 are also tested for chroma.
+  const int num_sizes =
+      (block_size_wide[block_idx] == 4 || block_size_high[block_idx] == 4) ? 2
+                                                                           : 1;
+
+  for (int shift = 0; shift < num_sizes; ++shift) {  // luma and chroma
+    const int out_w = block_size_wide[block_idx] >> shift;
+    const int out_h = block_size_high[block_idx] >> shift;
+    const int num_loops = 1000000000 / (out_w + out_h);
+
+    const InterpFilterParams *filter_params_x =
+        av1_get_interp_filter_params_with_block_size((InterpFilter)hfilter,
+                                                     out_w);
+    const InterpFilterParams *filter_params_y =
+        av1_get_interp_filter_params_with_block_size((InterpFilter)vfilter,
+                                                     out_h);
+
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < num_loops; ++i)
+      test_impl(input + offset_r * w + offset_c, w, output, MAX_SB_SIZE, out_w,
+                out_h, filter_params_x, filter_params_y, subx, suby,
+                &conv_params, bd);
+
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+    printf("%d,%d convolve %3dx%-3d: %7.2f us\n", has_subx, has_suby, out_w,
+           out_h, 1000.0 * elapsed_time / num_loops);
+  }
+}
+
+void AV1HighbdConvolve2DSrTest::RunCheckOutput(
+    highbd_convolve_2d_func test_impl) {
+  const int w = kMaxSize, h = kMaxSize;
+  const int bd = GET_PARAM(0);
+  const int has_subx = GET_PARAM(2);
+  const int has_suby = GET_PARAM(3);
+  const int block_idx = GET_PARAM(4);
+  int hfilter, vfilter, subx, suby;
+  uint16_t input[kMaxSize * kMaxSize];
+  DECLARE_ALIGNED(32, uint16_t, output[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint16_t, output2[MAX_SB_SQUARE]);
+
+  for (int i = 0; i < h; ++i)
+    for (int j = 0; j < w; ++j)
+      input[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+  for (int i = 0; i < MAX_SB_SQUARE; ++i)
+    output[i] = output2[i] = rnd_.Rand31();
+
+  // Make sure that sizes 2xN and Nx2 are also tested for chroma.
+  const int num_sizes =
+      (block_size_wide[block_idx] == 4 || block_size_high[block_idx] == 4) ? 2
+                                                                           : 1;
+  for (int shift = 0; shift < num_sizes; ++shift) {  // luma and chroma
+    const int out_w = block_size_wide[block_idx] >> shift;
+    const int out_h = block_size_high[block_idx] >> shift;
+    for (hfilter = EIGHTTAP_REGULAR; hfilter < INTERP_FILTERS_ALL; ++hfilter) {
+      for (vfilter = EIGHTTAP_REGULAR; vfilter < INTERP_FILTERS_ALL;
+           ++vfilter) {
+        const InterpFilterParams *filter_params_x =
+            av1_get_interp_filter_params_with_block_size((InterpFilter)hfilter,
+                                                         out_w);
+        const InterpFilterParams *filter_params_y =
+            av1_get_interp_filter_params_with_block_size((InterpFilter)vfilter,
+                                                         out_h);
+        for (int do_average = 0; do_average < 1; ++do_average) {
+          ConvolveParams conv_params1 =
+              get_conv_params_no_round(do_average, 0, NULL, 0, 0, bd);
+          ConvolveParams conv_params2 =
+              get_conv_params_no_round(do_average, 0, NULL, 0, 0, bd);
+
+          const int subx_range = has_subx ? 16 : 1;
+          const int suby_range = has_suby ? 16 : 1;
+          for (subx = 0; subx < subx_range; ++subx) {
+            for (suby = 0; suby < suby_range; ++suby) {
+              // Choose random locations within the source block
+              const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+              const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+              av1_highbd_convolve_2d_sr_c(input + offset_r * w + offset_c, w,
+                                          output, MAX_SB_SIZE, out_w, out_h,
+                                          filter_params_x, filter_params_y,
+                                          subx, suby, &conv_params1, bd);
+              test_impl(input + offset_r * w + offset_c, w, output2,
+                        MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                        filter_params_y, subx, suby, &conv_params2, bd);
+
+              if (memcmp(output, output2, sizeof(output))) {
+                for (int i = 0; i < MAX_SB_SIZE; ++i) {
+                  for (int j = 0; j < MAX_SB_SIZE; ++j) {
+                    int idx = i * MAX_SB_SIZE + j;
+                    ASSERT_EQ(output[idx], output2[idx])
+                        << out_w << "x" << out_h << " Pixel mismatch at index "
+                        << idx << " = (" << i << ", " << j
+                        << "), sub pixel offset = (" << suby << ", " << subx
+                        << ")";
+                  }
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+AV1HighbdJntConvolve2DTest::~AV1HighbdJntConvolve2DTest() {}
+void AV1HighbdJntConvolve2DTest::SetUp() {
+  rnd_.Reset(ACMRandom::DeterministicSeed());
+}
+
+void AV1HighbdJntConvolve2DTest::TearDown() { libaom_test::ClearSystemState(); }
+
+void AV1HighbdJntConvolve2DTest::RunSpeedTest(
+    highbd_convolve_2d_func test_impl) {
+  const int w = kMaxSize, h = kMaxSize;
+  const int bd = GET_PARAM(0);
+  const int block_idx = GET_PARAM(4);
+  int hfilter, vfilter, subx, suby;
+  uint16_t input[kMaxSize * kMaxSize];
+  DECLARE_ALIGNED(32, CONV_BUF_TYPE, output[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint16_t, output16[MAX_SB_SQUARE]);
+
+  for (int i = 0; i < h; ++i)
+    for (int j = 0; j < w; ++j)
+      input[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) output[i] = rnd_.Rand16();
+  hfilter = EIGHTTAP_REGULAR;
+  vfilter = EIGHTTAP_REGULAR;
+  int do_average = 0;
+  const int out_w = block_size_wide[block_idx];
+  const int out_h = block_size_high[block_idx];
+
+  const InterpFilterParams *filter_params_x =
+      av1_get_interp_filter_params_with_block_size((InterpFilter)hfilter,
+                                                   out_w);
+  const InterpFilterParams *filter_params_y =
+      av1_get_interp_filter_params_with_block_size((InterpFilter)vfilter,
+                                                   out_h);
+
+  ConvolveParams conv_params =
+      get_conv_params_no_round(do_average, 0, output, MAX_SB_SIZE, 1, bd);
+
+  // Test special case where jnt_comp_avg is not used
+  conv_params.use_jnt_comp_avg = 0;
+
+  subx = 0;
+  suby = 0;
+  // Choose random locations within the source block
+  const int offset_r = 3;
+  const int offset_c = 3;
+
+  const int num_loops = 1000000000 / (out_w + out_h);
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < num_loops; ++i)
+    test_impl(input + offset_r * w + offset_c, w, output16, MAX_SB_SIZE, out_w,
+              out_h, filter_params_x, filter_params_y, subx, suby, &conv_params,
+              bd);
+
+  aom_usec_timer_mark(&timer);
+  const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+  printf("convolve %3dx%-3d: %7.2f us\n", out_w, out_h,
+         1000.0 * elapsed_time / num_loops);
+}
+
+void AV1HighbdJntConvolve2DTest::RunCheckOutput(
+    highbd_convolve_2d_func test_impl) {
+  const int w = kMaxSize, h = kMaxSize;
+  const int bd = GET_PARAM(0);
+  const int has_subx = GET_PARAM(2);
+  const int has_suby = GET_PARAM(3);
+  const int block_idx = GET_PARAM(4);
+  int hfilter, vfilter, subx, suby;
+  uint16_t input[kMaxSize * kMaxSize];
+  DECLARE_ALIGNED(32, CONV_BUF_TYPE, output1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, CONV_BUF_TYPE, output2[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint16_t, output16_1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint16_t, output16_2[MAX_SB_SQUARE]);
+
+  for (int i = 0; i < h; ++i)
+    for (int j = 0; j < w; ++j)
+      input[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    output1[i] = output2[i] = rnd_.Rand16();
+    output16_1[i] = output16_2[i] = rnd_.Rand16();
+  }
+
+  const int out_w = block_size_wide[block_idx];
+  const int out_h = block_size_high[block_idx];
+  for (hfilter = EIGHTTAP_REGULAR; hfilter < INTERP_FILTERS_ALL; ++hfilter) {
+    for (vfilter = EIGHTTAP_REGULAR; vfilter < INTERP_FILTERS_ALL; ++vfilter) {
+      const InterpFilterParams *filter_params_x =
+          av1_get_interp_filter_params_with_block_size((InterpFilter)hfilter,
+                                                       out_w);
+      const InterpFilterParams *filter_params_y =
+          av1_get_interp_filter_params_with_block_size((InterpFilter)vfilter,
+                                                       out_h);
+      for (int do_average = 0; do_average <= 1; ++do_average) {
+        ConvolveParams conv_params1 = get_conv_params_no_round(
+            do_average, 0, output1, MAX_SB_SIZE, 1, bd);
+        ConvolveParams conv_params2 = get_conv_params_no_round(
+            do_average, 0, output2, MAX_SB_SIZE, 1, bd);
+
+        // Test special case where jnt_comp_avg is not used
+        conv_params1.use_jnt_comp_avg = 0;
+        conv_params2.use_jnt_comp_avg = 0;
+
+        const int subx_range = has_subx ? 16 : 1;
+        const int suby_range = has_suby ? 16 : 1;
+        for (subx = 0; subx < subx_range; ++subx) {
+          for (suby = 0; suby < suby_range; ++suby) {
+            // Choose random locations within the source block
+            const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+            const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+            av1_highbd_jnt_convolve_2d_c(input + offset_r * w + offset_c, w,
+                                         output16_1, MAX_SB_SIZE, out_w, out_h,
+                                         filter_params_x, filter_params_y, subx,
+                                         suby, &conv_params1, bd);
+            test_impl(input + offset_r * w + offset_c, w, output16_2,
+                      MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                      filter_params_y, subx, suby, &conv_params2, bd);
+
+            for (int i = 0; i < out_h; ++i) {
+              for (int j = 0; j < out_w; ++j) {
+                int idx = i * MAX_SB_SIZE + j;
+                ASSERT_EQ(output1[idx], output2[idx])
+                    << out_w << "x" << out_h << " Pixel mismatch at index "
+                    << idx << " = (" << i << ", " << j
+                    << "), sub pixel offset = (" << suby << ", " << subx << ")";
+              }
+            }
+
+            if (memcmp(output16_1, output16_2, sizeof(output16_1))) {
+              for (int i = 0; i < MAX_SB_SIZE; ++i) {
+                for (int j = 0; j < MAX_SB_SIZE; ++j) {
+                  int idx = i * MAX_SB_SIZE + j;
+                  ASSERT_EQ(output16_1[idx], output16_2[idx])
+                      << out_w << "x" << out_h << " Pixel mismatch at index "
+                      << idx << " = (" << i << ", " << j
+                      << "), sub pixel offset = (" << suby << ", " << subx
+                      << ")";
+                }
+              }
+            }
+          }
+        }
+
+        // Test different combination of fwd and bck offset weights
+        for (int k = 0; k < 2; ++k) {
+          for (int l = 0; l < 4; ++l) {
+            conv_params1.use_jnt_comp_avg = 1;
+            conv_params2.use_jnt_comp_avg = 1;
+            conv_params1.fwd_offset = quant_dist_lookup_table[k][l][0];
+            conv_params1.bck_offset = quant_dist_lookup_table[k][l][1];
+            conv_params2.fwd_offset = quant_dist_lookup_table[k][l][0];
+            conv_params2.bck_offset = quant_dist_lookup_table[k][l][1];
+
+            const int subx_range = has_subx ? 16 : 1;
+            const int suby_range = has_suby ? 16 : 1;
+            for (subx = 0; subx < subx_range; ++subx) {
+              for (suby = 0; suby < suby_range; ++suby) {
+                // Choose random locations within the source block
+                const int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+                const int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+                av1_highbd_jnt_convolve_2d_c(
+                    input + offset_r * w + offset_c, w, output16_1, MAX_SB_SIZE,
+                    out_w, out_h, filter_params_x, filter_params_y, subx, suby,
+                    &conv_params1, bd);
+                test_impl(input + offset_r * w + offset_c, w, output16_2,
+                          MAX_SB_SIZE, out_w, out_h, filter_params_x,
+                          filter_params_y, subx, suby, &conv_params2, bd);
+
+                for (int i = 0; i < out_h; ++i) {
+                  for (int j = 0; j < out_w; ++j) {
+                    int idx = i * MAX_SB_SIZE + j;
+                    ASSERT_EQ(output1[idx], output2[idx])
+                        << out_w << "x" << out_h << " Pixel mismatch at index "
+                        << idx << " = (" << i << ", " << j
+                        << "), sub pixel offset = (" << suby << ", " << subx
+                        << ")";
+                  }
+                }
+
+                if (memcmp(output16_1, output16_2, sizeof(output16_1))) {
+                  for (int i = 0; i < MAX_SB_SIZE; ++i) {
+                    for (int j = 0; j < MAX_SB_SIZE; ++j) {
+                      int idx = i * MAX_SB_SIZE + j;
+                      ASSERT_EQ(output16_1[idx], output16_2[idx])
+                          << out_w << "x" << out_h
+                          << " Pixel mismatch at index " << idx << " = (" << i
+                          << ", " << j << "), sub pixel offset = (" << suby
+                          << ", " << subx << ")";
+                    }
+                  }
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+}
+}  // namespace AV1HighbdConvolve2D
+}  // namespace libaom_test
diff --git a/third_party/aom/test/av1_convolve_2d_test_util.h b/third_party/aom/test/av1_convolve_2d_test_util.h
new file mode 100644
index 0000000000..e0eb584108
--- /dev/null
+++ b/third_party/aom/test/av1_convolve_2d_test_util.h
@@ -0,0 +1,117 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_AV1_CONVOLVE_2D_TEST_UTIL_H_
+#define AOM_TEST_AV1_CONVOLVE_2D_TEST_UTIL_H_
+
+#include "config/av1_rtcd.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/util.h"
+
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+
+namespace libaom_test {
+
+namespace AV1Convolve2D {
+
+typedef void (*convolve_2d_func)(const uint8_t *src, int src_stride,
+                                 uint8_t *dst, int dst_stride, int w, int h,
+                                 const InterpFilterParams *filter_params_x,
+                                 const InterpFilterParams *filter_params_y,
+                                 const int subpel_x_q4, const int subpel_y_q4,
+                                 ConvolveParams *conv_params);
+
+typedef ::testing::tuple<convolve_2d_func, int, int, BLOCK_SIZE>
+    Convolve2DParam;
+
+::testing::internal::ParamGenerator<Convolve2DParam> BuildParams(
+    convolve_2d_func filter, int subx_exist, int suby_exist);
+
+class AV1Convolve2DSrTest : public ::testing::TestWithParam<Convolve2DParam> {
+ public:
+  virtual ~AV1Convolve2DSrTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput(convolve_2d_func test_impl);
+  void RunSpeedTest(convolve_2d_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+};
+
+class AV1JntConvolve2DTest : public ::testing::TestWithParam<Convolve2DParam> {
+ public:
+  virtual ~AV1JntConvolve2DTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput(convolve_2d_func test_impl);
+  void RunSpeedTest(convolve_2d_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+};
+}  // namespace AV1Convolve2D
+
+namespace AV1HighbdConvolve2D {
+typedef void (*highbd_convolve_2d_func)(
+    const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w,
+    int h, const InterpFilterParams *filter_params_x,
+    const InterpFilterParams *filter_params_y, const int subpel_x_q4,
+    const int subpel_y_q4, ConvolveParams *conv_params, int bd);
+
+typedef ::testing::tuple<int, highbd_convolve_2d_func, int, int, BLOCK_SIZE>
+    HighbdConvolve2DParam;
+
+::testing::internal::ParamGenerator<HighbdConvolve2DParam> BuildParams(
+    highbd_convolve_2d_func filter, int subx_exist, int suby_exist);
+
+class AV1HighbdConvolve2DSrTest
+    : public ::testing::TestWithParam<HighbdConvolve2DParam> {
+ public:
+  virtual ~AV1HighbdConvolve2DSrTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput(highbd_convolve_2d_func test_impl);
+  void RunSpeedTest(highbd_convolve_2d_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+};
+
+class AV1HighbdJntConvolve2DTest
+    : public ::testing::TestWithParam<HighbdConvolve2DParam> {
+ public:
+  virtual ~AV1HighbdJntConvolve2DTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput(highbd_convolve_2d_func test_impl);
+  void RunSpeedTest(highbd_convolve_2d_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+};
+}  // namespace AV1HighbdConvolve2D
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_AV1_CONVOLVE_2D_TEST_UTIL_H_
diff --git a/third_party/aom/test/av1_convolve_scale_test.cc b/third_party/aom/test/av1_convolve_scale_test.cc
new file mode 100644
index 0000000000..3b1698eeb4
--- /dev/null
+++ b/third_party/aom/test/av1_convolve_scale_test.cc
@@ -0,0 +1,529 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <vector>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/aom_timer.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+#include "av1/common/common_data.h"
+
+namespace {
+const int kTestIters = 10;
+const int kPerfIters = 1000;
+
+const int kVPad = 32;
+const int kHPad = 32;
+const int kXStepQn = 16;
+const int kYStepQn = 20;
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+
+enum NTaps { EIGHT_TAP, TEN_TAP, TWELVE_TAP };
+int NTapsToInt(NTaps ntaps) { return 8 + static_cast<int>(ntaps) * 2; }
+
+// A 16-bit filter with a configurable number of taps.
+class TestFilter {
+ public:
+  void set(NTaps ntaps, bool backwards);
+
+  InterpFilterParams params_;
+
+ private:
+  std::vector<int16_t> coeffs_;
+};
+
+void TestFilter::set(NTaps ntaps, bool backwards) {
+  const int n = NTapsToInt(ntaps);
+  assert(n >= 8 && n <= 12);
+
+  // The filter has n * SUBPEL_SHIFTS proper elements and an extra 8 bogus
+  // elements at the end so that convolutions can read off the end safely.
+  coeffs_.resize(n * SUBPEL_SHIFTS + 8);
+
+  // The coefficients are pretty much arbitrary, but convolutions shouldn't
+  // over or underflow. For the first filter (subpels = 0), we use an
+  // increasing or decreasing ramp (depending on the backwards parameter). We
+  // don't want any zero coefficients, so we make it have an x-intercept at -1
+  // or n. To ensure absence of under/overflow, we normalise the area under the
+  // ramp to be I = 1 << FILTER_BITS (so that convolving a constant function
+  // gives the identity).
+  //
+  // When increasing, the function has the form:
+  //
+  //   f(x) = A * (x + 1)
+  //
+  // Summing and rearranging for A gives A = 2 * I / (n * (n + 1)). If the
+  // filter is reversed, we have the same A but with formula
+  //
+  //   g(x) = A * (n - x)
+  const int I = 1 << FILTER_BITS;
+  const float A = 2.f * I / (n * (n + 1.f));
+  for (int i = 0; i < n; ++i) {
+    coeffs_[i] = static_cast<int16_t>(A * (backwards ? (n - i) : (i + 1)));
+  }
+
+  // For the other filters, make them slightly different by swapping two
+  // columns. Filter k will have the columns (k % n) and (7 * k) % n swapped.
+  const size_t filter_size = sizeof(coeffs_[0] * n);
+  int16_t *const filter0 = &coeffs_[0];
+  for (int k = 1; k < SUBPEL_SHIFTS; ++k) {
+    int16_t *filterk = &coeffs_[k * n];
+    memcpy(filterk, filter0, filter_size);
+
+    const int idx0 = k % n;
+    const int idx1 = (7 * k) % n;
+
+    const int16_t tmp = filterk[idx0];
+    filterk[idx0] = filterk[idx1];
+    filterk[idx1] = tmp;
+  }
+
+  // Finally, write some rubbish at the end to make sure we don't use it.
+  for (int i = 0; i < 8; ++i) coeffs_[n * SUBPEL_SHIFTS + i] = 123 + i;
+
+  // Fill in params
+  params_.filter_ptr = &coeffs_[0];
+  params_.taps = n;
+  // These are ignored by the functions being tested. Set them to whatever.
+  params_.subpel_shifts = SUBPEL_SHIFTS;
+  params_.interp_filter = EIGHTTAP_REGULAR;
+}
+
+template <typename SrcPixel>
+class TestImage {
+ public:
+  TestImage(int w, int h, int bd) : w_(w), h_(h), bd_(bd) {
+    assert(bd < 16);
+    assert(bd <= 8 * static_cast<int>(sizeof(SrcPixel)));
+
+    // Pad width by 2*kHPad and then round up to the next multiple of 16
+    // to get src_stride_. Add another 16 for dst_stride_ (to make sure
+    // something goes wrong if we use the wrong one)
+    src_stride_ = (w_ + 2 * kHPad + 15) & ~15;
+    dst_stride_ = src_stride_ + 16;
+
+    // Allocate image data
+    src_data_.resize(2 * src_block_size());
+    dst_data_.resize(2 * dst_block_size());
+    dst_16_data_.resize(2 * dst_block_size());
+  }
+
+  void Initialize(ACMRandom *rnd);
+  void Check() const;
+
+  int src_stride() const { return src_stride_; }
+  int dst_stride() const { return dst_stride_; }
+
+  int src_block_size() const { return (h_ + 2 * kVPad) * src_stride(); }
+  int dst_block_size() const { return (h_ + 2 * kVPad) * dst_stride(); }
+
+  const SrcPixel *GetSrcData(bool ref, bool borders) const {
+    const SrcPixel *block = &src_data_[ref ? 0 : src_block_size()];
+    return borders ? block : block + kHPad + src_stride_ * kVPad;
+  }
+
+  SrcPixel *GetDstData(bool ref, bool borders) {
+    SrcPixel *block = &dst_data_[ref ? 0 : dst_block_size()];
+    return borders ? block : block + kHPad + dst_stride_ * kVPad;
+  }
+
+  CONV_BUF_TYPE *GetDst16Data(bool ref, bool borders) {
+    CONV_BUF_TYPE *block = &dst_16_data_[ref ? 0 : dst_block_size()];
+    return borders ? block : block + kHPad + dst_stride_ * kVPad;
+  }
+
+ private:
+  int w_, h_, bd_;
+  int src_stride_, dst_stride_;
+
+  std::vector<SrcPixel> src_data_;
+  std::vector<SrcPixel> dst_data_;
+  std::vector<CONV_BUF_TYPE> dst_16_data_;
+};
+
+template <typename Pixel>
+void FillEdge(ACMRandom *rnd, int num_pixels, int bd, bool trash, Pixel *data) {
+  if (!trash) {
+    memset(data, 0, sizeof(*data) * num_pixels);
+    return;
+  }
+  const Pixel mask = (1 << bd) - 1;
+  for (int i = 0; i < num_pixels; ++i) data[i] = rnd->Rand16() & mask;
+}
+
+template <typename Pixel>
+void PrepBuffers(ACMRandom *rnd, int w, int h, int stride, int bd,
+                 bool trash_edges, Pixel *data) {
+  assert(rnd);
+  const Pixel mask = (1 << bd) - 1;
+
+  // Fill in the first buffer with random data
+  // Top border
+  FillEdge(rnd, stride * kVPad, bd, trash_edges, data);
+  for (int r = 0; r < h; ++r) {
+    Pixel *row_data = data + (kVPad + r) * stride;
+    // Left border, contents, right border
+    FillEdge(rnd, kHPad, bd, trash_edges, row_data);
+    for (int c = 0; c < w; ++c) row_data[kHPad + c] = rnd->Rand16() & mask;
+    FillEdge(rnd, kHPad, bd, trash_edges, row_data + kHPad + w);
+  }
+  // Bottom border
+  FillEdge(rnd, stride * kVPad, bd, trash_edges, data + stride * (kVPad + h));
+
+  const int bpp = sizeof(*data);
+  const int block_elts = stride * (h + 2 * kVPad);
+  const int block_size = bpp * block_elts;
+
+  // Now copy that to the second buffer
+  memcpy(data + block_elts, data, block_size);
+}
+
+template <typename SrcPixel>
+void TestImage<SrcPixel>::Initialize(ACMRandom *rnd) {
+  PrepBuffers(rnd, w_, h_, src_stride_, bd_, false, &src_data_[0]);
+  PrepBuffers(rnd, w_, h_, dst_stride_, bd_, true, &dst_data_[0]);
+  PrepBuffers(rnd, w_, h_, dst_stride_, bd_, true, &dst_16_data_[0]);
+}
+
+template <typename SrcPixel>
+void TestImage<SrcPixel>::Check() const {
+  // If memcmp returns 0, there's nothing to do.
+  const int num_pixels = dst_block_size();
+  const SrcPixel *ref_dst = &dst_data_[0];
+  const SrcPixel *tst_dst = &dst_data_[num_pixels];
+
+  const CONV_BUF_TYPE *ref_16_dst = &dst_16_data_[0];
+  const CONV_BUF_TYPE *tst_16_dst = &dst_16_data_[num_pixels];
+
+  if (0 == memcmp(ref_dst, tst_dst, sizeof(*ref_dst) * num_pixels)) {
+    if (0 == memcmp(ref_16_dst, tst_16_dst, sizeof(*ref_16_dst) * num_pixels))
+      return;
+  }
+  // Otherwise, iterate through the buffer looking for differences (including
+  // the edges)
+  const int stride = dst_stride_;
+  for (int r = 0; r < h_ + 2 * kVPad; ++r) {
+    for (int c = 0; c < w_ + 2 * kHPad; ++c) {
+      const int32_t ref_value = ref_dst[r * stride + c];
+      const int32_t tst_value = tst_dst[r * stride + c];
+
+      EXPECT_EQ(tst_value, ref_value)
+          << "Error at row: " << (r - kVPad) << ", col: " << (c - kHPad);
+    }
+  }
+
+  for (int r = 0; r < h_ + 2 * kVPad; ++r) {
+    for (int c = 0; c < w_ + 2 * kHPad; ++c) {
+      const int32_t ref_value = ref_16_dst[r * stride + c];
+      const int32_t tst_value = tst_16_dst[r * stride + c];
+
+      EXPECT_EQ(tst_value, ref_value)
+          << "Error in 16 bit buffer "
+          << "Error at row: " << (r - kVPad) << ", col: " << (c - kHPad);
+    }
+  }
+}
+
+typedef tuple<int, int> BlockDimension;
+
+struct BaseParams {
+  BaseParams(BlockDimension dims, NTaps ntaps_x, NTaps ntaps_y, bool avg)
+      : dims(dims), ntaps_x(ntaps_x), ntaps_y(ntaps_y), avg(avg) {}
+
+  BlockDimension dims;
+  NTaps ntaps_x, ntaps_y;
+  bool avg;
+};
+
+template <typename SrcPixel>
+class ConvolveScaleTestBase : public ::testing::Test {
+ public:
+  ConvolveScaleTestBase() : image_(NULL) {}
+  virtual ~ConvolveScaleTestBase() { delete image_; }
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+  // Implemented by subclasses (SetUp depends on the parameters passed
+  // in and RunOne depends on the function to be tested. These can't
+  // be templated for low/high bit depths because they have different
+  // numbers of parameters)
+  virtual void SetUp() = 0;
+  virtual void RunOne(bool ref) = 0;
+
+ protected:
+  void SetParams(const BaseParams &params, int bd) {
+    width_ = ::testing::get<0>(params.dims);
+    height_ = ::testing::get<1>(params.dims);
+    ntaps_x_ = params.ntaps_x;
+    ntaps_y_ = params.ntaps_y;
+    bd_ = bd;
+    avg_ = params.avg;
+
+    filter_x_.set(ntaps_x_, false);
+    filter_y_.set(ntaps_y_, true);
+    convolve_params_ =
+        get_conv_params_no_round(avg_ != false, 0, NULL, 0, 1, bd);
+
+    delete image_;
+    image_ = new TestImage<SrcPixel>(width_, height_, bd_);
+  }
+
+  void SetConvParamOffset(int i, int j, int is_compound, int do_average,
+                          int use_jnt_comp_avg) {
+    if (i == -1 && j == -1) {
+      convolve_params_.use_jnt_comp_avg = use_jnt_comp_avg;
+      convolve_params_.is_compound = is_compound;
+      convolve_params_.do_average = do_average;
+    } else {
+      convolve_params_.use_jnt_comp_avg = use_jnt_comp_avg;
+      convolve_params_.fwd_offset = quant_dist_lookup_table[i][j][0];
+      convolve_params_.bck_offset = quant_dist_lookup_table[i][j][1];
+      convolve_params_.is_compound = is_compound;
+      convolve_params_.do_average = do_average;
+    }
+  }
+
+  void Run() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    for (int i = 0; i < kTestIters; ++i) {
+      int is_compound = 0;
+      SetConvParamOffset(-1, -1, is_compound, 0, 0);
+      Prep(&rnd);
+      RunOne(true);
+      RunOne(false);
+      image_->Check();
+
+      is_compound = 1;
+      for (int do_average = 0; do_average < 2; do_average++) {
+        for (int use_jnt_comp_avg = 0; use_jnt_comp_avg < 2;
+             use_jnt_comp_avg++) {
+          for (int j = 0; j < 2; ++j) {
+            for (int k = 0; k < 4; ++k) {
+              SetConvParamOffset(j, k, is_compound, do_average,
+                                 use_jnt_comp_avg);
+              Prep(&rnd);
+              RunOne(true);
+              RunOne(false);
+              image_->Check();
+            }
+          }
+        }
+      }
+    }
+  }
+
+  void SpeedTest() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    Prep(&rnd);
+
+    aom_usec_timer ref_timer;
+    aom_usec_timer_start(&ref_timer);
+    for (int i = 0; i < kPerfIters; ++i) RunOne(true);
+    aom_usec_timer_mark(&ref_timer);
+    const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);
+
+    aom_usec_timer tst_timer;
+    aom_usec_timer_start(&tst_timer);
+    for (int i = 0; i < kPerfIters; ++i) RunOne(false);
+    aom_usec_timer_mark(&tst_timer);
+    const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);
+
+    std::cout << "[          ] C time = " << ref_time / 1000
+              << " ms, SIMD time = " << tst_time / 1000 << " ms\n";
+
+    EXPECT_GT(ref_time, tst_time)
+        << "Error: CDEFSpeedTest, SIMD slower than C.\n"
+        << "C time: " << ref_time << " us\n"
+        << "SIMD time: " << tst_time << " us\n";
+  }
+
+  static int RandomSubpel(ACMRandom *rnd) {
+    const uint8_t subpel_mode = rnd->Rand8();
+    if ((subpel_mode & 7) == 0) {
+      return 0;
+    } else if ((subpel_mode & 7) == 1) {
+      return SCALE_SUBPEL_SHIFTS - 1;
+    } else {
+      return 1 + rnd->PseudoUniform(SCALE_SUBPEL_SHIFTS - 2);
+    }
+  }
+
+  void Prep(ACMRandom *rnd) {
+    assert(rnd);
+
+    // Choose subpel_x_ and subpel_y_. They should be less than
+    // SCALE_SUBPEL_SHIFTS; we also want to add extra weight to "interesting"
+    // values: 0 and SCALE_SUBPEL_SHIFTS - 1
+    subpel_x_ = RandomSubpel(rnd);
+    subpel_y_ = RandomSubpel(rnd);
+
+    image_->Initialize(rnd);
+  }
+
+  int width_, height_, bd_;
+  NTaps ntaps_x_, ntaps_y_;
+  bool avg_;
+  int subpel_x_, subpel_y_;
+  TestFilter filter_x_, filter_y_;
+  TestImage<SrcPixel> *image_;
+  ConvolveParams convolve_params_;
+};
+
+typedef tuple<int, int> BlockDimension;
+
+typedef void (*LowbdConvolveFunc)(const uint8_t *src, int src_stride,
+                                  uint8_t *dst, int dst_stride, int w, int h,
+                                  const InterpFilterParams *filter_params_x,
+                                  const InterpFilterParams *filter_params_y,
+                                  const int subpel_x_qn, const int x_step_qn,
+                                  const int subpel_y_qn, const int y_step_qn,
+                                  ConvolveParams *conv_params);
+
+// Test parameter list:
+//  <tst_fun, dims, ntaps_x, ntaps_y, avg>
+typedef tuple<LowbdConvolveFunc, BlockDimension, NTaps, NTaps, bool>
+    LowBDParams;
+
+class LowBDConvolveScaleTest
+    : public ConvolveScaleTestBase<uint8_t>,
+      public ::testing::WithParamInterface<LowBDParams> {
+ public:
+  virtual ~LowBDConvolveScaleTest() {}
+
+  void SetUp() {
+    tst_fun_ = GET_PARAM(0);
+
+    const BlockDimension &block = GET_PARAM(1);
+    const NTaps ntaps_x = GET_PARAM(2);
+    const NTaps ntaps_y = GET_PARAM(3);
+    const int bd = 8;
+    const bool avg = GET_PARAM(4);
+
+    SetParams(BaseParams(block, ntaps_x, ntaps_y, avg), bd);
+  }
+
+  void RunOne(bool ref) {
+    const uint8_t *src = image_->GetSrcData(ref, false);
+    uint8_t *dst = image_->GetDstData(ref, false);
+    convolve_params_.dst = image_->GetDst16Data(ref, false);
+    const int src_stride = image_->src_stride();
+    const int dst_stride = image_->dst_stride();
+    if (ref) {
+      av1_convolve_2d_scale_c(src, src_stride, dst, dst_stride, width_, height_,
+                              &filter_x_.params_, &filter_y_.params_, subpel_x_,
+                              kXStepQn, subpel_y_, kYStepQn, &convolve_params_);
+    } else {
+      tst_fun_(src, src_stride, dst, dst_stride, width_, height_,
+               &filter_x_.params_, &filter_y_.params_, subpel_x_, kXStepQn,
+               subpel_y_, kYStepQn, &convolve_params_);
+    }
+  }
+
+ private:
+  LowbdConvolveFunc tst_fun_;
+};
+
+const BlockDimension kBlockDim[] = {
+  make_tuple(2, 2),    make_tuple(2, 4),    make_tuple(4, 4),
+  make_tuple(4, 8),    make_tuple(8, 4),    make_tuple(8, 8),
+  make_tuple(8, 16),   make_tuple(16, 8),   make_tuple(16, 16),
+  make_tuple(16, 32),  make_tuple(32, 16),  make_tuple(32, 32),
+  make_tuple(32, 64),  make_tuple(64, 32),  make_tuple(64, 64),
+  make_tuple(64, 128), make_tuple(128, 64), make_tuple(128, 128),
+};
+
+const NTaps kNTaps[] = { EIGHT_TAP };
+
+TEST_P(LowBDConvolveScaleTest, Check) { Run(); }
+TEST_P(LowBDConvolveScaleTest, DISABLED_Speed) { SpeedTest(); }
+
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, LowBDConvolveScaleTest,
+    ::testing::Combine(::testing::Values(av1_convolve_2d_scale_sse4_1),
+                       ::testing::ValuesIn(kBlockDim),
+                       ::testing::ValuesIn(kNTaps), ::testing::ValuesIn(kNTaps),
+                       ::testing::Bool()));
+
+typedef void (*HighbdConvolveFunc)(const uint16_t *src, int src_stride,
+                                   uint16_t *dst, int dst_stride, int w, int h,
+                                   const InterpFilterParams *filter_params_x,
+                                   const InterpFilterParams *filter_params_y,
+                                   const int subpel_x_qn, const int x_step_qn,
+                                   const int subpel_y_qn, const int y_step_qn,
+                                   ConvolveParams *conv_params, int bd);
+
+// Test parameter list:
+//  <tst_fun, dims, ntaps_x, ntaps_y, avg, bd>
+typedef tuple<HighbdConvolveFunc, BlockDimension, NTaps, NTaps, bool, int>
+    HighBDParams;
+
+class HighBDConvolveScaleTest
+    : public ConvolveScaleTestBase<uint16_t>,
+      public ::testing::WithParamInterface<HighBDParams> {
+ public:
+  virtual ~HighBDConvolveScaleTest() {}
+
+  void SetUp() {
+    tst_fun_ = GET_PARAM(0);
+
+    const BlockDimension &block = GET_PARAM(1);
+    const NTaps ntaps_x = GET_PARAM(2);
+    const NTaps ntaps_y = GET_PARAM(3);
+    const bool avg = GET_PARAM(4);
+    const int bd = GET_PARAM(5);
+
+    SetParams(BaseParams(block, ntaps_x, ntaps_y, avg), bd);
+  }
+
+  void RunOne(bool ref) {
+    const uint16_t *src = image_->GetSrcData(ref, false);
+    uint16_t *dst = image_->GetDstData(ref, false);
+    convolve_params_.dst = image_->GetDst16Data(ref, false);
+    const int src_stride = image_->src_stride();
+    const int dst_stride = image_->dst_stride();
+
+    if (ref) {
+      av1_highbd_convolve_2d_scale_c(
+          src, src_stride, dst, dst_stride, width_, height_, &filter_x_.params_,
+          &filter_y_.params_, subpel_x_, kXStepQn, subpel_y_, kYStepQn,
+          &convolve_params_, bd_);
+    } else {
+      tst_fun_(src, src_stride, dst, dst_stride, width_, height_,
+               &filter_x_.params_, &filter_y_.params_, subpel_x_, kXStepQn,
+               subpel_y_, kYStepQn, &convolve_params_, bd_);
+    }
+  }
+
+ private:
+  HighbdConvolveFunc tst_fun_;
+};
+
+const int kBDs[] = { 8, 10, 12 };
+
+TEST_P(HighBDConvolveScaleTest, Check) { Run(); }
+TEST_P(HighBDConvolveScaleTest, DISABLED_Speed) { SpeedTest(); }
+
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, HighBDConvolveScaleTest,
+    ::testing::Combine(::testing::Values(av1_highbd_convolve_2d_scale_sse4_1),
+                       ::testing::ValuesIn(kBlockDim),
+                       ::testing::ValuesIn(kNTaps), ::testing::ValuesIn(kNTaps),
+                       ::testing::Bool(), ::testing::ValuesIn(kBDs)));
+}  // namespace
diff --git a/third_party/aom/test/av1_encoder_parms_get_to_decoder.cc b/third_party/aom/test/av1_encoder_parms_get_to_decoder.cc
new file mode 100644
index 0000000000..e8470e5d50
--- /dev/null
+++ b/third_party/aom/test/av1_encoder_parms_get_to_decoder.cc
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+
+#include "aom/aom_decoder.h"
+#include "av1/decoder/decoder.h"
+
+namespace {
+
+const int kMaxPsnr = 100;
+
+struct ParamPassingTestVideo {
+  const char *name;
+  uint32_t width;
+  uint32_t height;
+  uint32_t bitrate;
+  int frames;
+};
+
+const ParamPassingTestVideo kAV1ParamPassingTestVector = {
+  "niklas_1280_720_30.y4m", 1280, 720, 600, 3
+};
+
+struct EncodeParameters {
+  int32_t lossless;
+  aom_color_primaries_t color_primaries;
+  aom_transfer_characteristics_t transfer_characteristics;
+  aom_matrix_coefficients_t matrix_coefficients;
+  aom_color_range_t color_range;
+  aom_chroma_sample_position_t chroma_sample_position;
+  int32_t render_size[2];
+};
+
+const EncodeParameters kAV1EncodeParameterSet[] = {
+  { 1,
+    AOM_CICP_CP_BT_709,
+    AOM_CICP_TC_BT_709,
+    AOM_CICP_MC_BT_709,
+    AOM_CR_STUDIO_RANGE,
+    AOM_CSP_UNKNOWN,
+    { 0, 0 } },
+  { 0,
+    AOM_CICP_CP_BT_470_M,
+    AOM_CICP_TC_BT_470_M,
+    AOM_CICP_MC_BT_470_B_G,
+    AOM_CR_FULL_RANGE,
+    AOM_CSP_VERTICAL,
+    { 0, 0 } },
+  { 1,
+    AOM_CICP_CP_BT_601,
+    AOM_CICP_TC_BT_601,
+    AOM_CICP_MC_BT_601,
+    AOM_CR_STUDIO_RANGE,
+    AOM_CSP_COLOCATED,
+    { 0, 0 } },
+  { 0,
+    AOM_CICP_CP_BT_2020,
+    AOM_CICP_TC_BT_2020_10_BIT,
+    AOM_CICP_MC_BT_2020_NCL,
+    AOM_CR_FULL_RANGE,
+    AOM_CSP_RESERVED,
+    { 640, 480 } },
+};
+
+class AVxEncoderParmsGetToDecoder
+    : public ::libaom_test::EncoderTest,
+      public ::libaom_test::CodecTestWithParam<EncodeParameters> {
+ protected:
+  AVxEncoderParmsGetToDecoder()
+      : EncoderTest(GET_PARAM(0)), encode_parms(GET_PARAM(1)) {}
+
+  virtual ~AVxEncoderParmsGetToDecoder() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(::libaom_test::kTwoPassGood);
+    cfg_.g_lag_in_frames = 25;
+    test_video_ = kAV1ParamPassingTestVector;
+    cfg_.rc_target_bitrate = test_video_.bitrate;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AV1E_SET_COLOR_PRIMARIES, encode_parms.color_primaries);
+      encoder->Control(AV1E_SET_TRANSFER_CHARACTERISTICS,
+                       encode_parms.transfer_characteristics);
+      encoder->Control(AV1E_SET_MATRIX_COEFFICIENTS,
+                       encode_parms.matrix_coefficients);
+      encoder->Control(AV1E_SET_COLOR_RANGE, encode_parms.color_range);
+      encoder->Control(AV1E_SET_CHROMA_SAMPLE_POSITION,
+                       encode_parms.chroma_sample_position);
+      encoder->Control(AV1E_SET_LOSSLESS, encode_parms.lossless);
+      if (encode_parms.render_size[0] > 0 && encode_parms.render_size[1] > 0) {
+        encoder->Control(AV1E_SET_RENDER_SIZE, encode_parms.render_size);
+      }
+    }
+  }
+
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     aom_codec_pts_t pts) {
+    (void)pts;
+    if (encode_parms.render_size[0] > 0 && encode_parms.render_size[1] > 0) {
+      EXPECT_EQ(encode_parms.render_size[0], (int)img.r_w);
+      EXPECT_EQ(encode_parms.render_size[1], (int)img.r_h);
+    }
+    EXPECT_EQ(encode_parms.color_primaries, img.cp);
+    EXPECT_EQ(encode_parms.transfer_characteristics, img.tc);
+    EXPECT_EQ(encode_parms.matrix_coefficients, img.mc);
+    EXPECT_EQ(encode_parms.color_range, img.range);
+    EXPECT_EQ(encode_parms.chroma_sample_position, img.csp);
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (encode_parms.lossless) {
+      EXPECT_EQ(kMaxPsnr, pkt->data.psnr.psnr[0]);
+    }
+  }
+
+  virtual bool HandleDecodeResult(const aom_codec_err_t res_dec,
+                                  libaom_test::Decoder *decoder) {
+    EXPECT_EQ(AOM_CODEC_OK, res_dec) << decoder->DecodeError();
+    return AOM_CODEC_OK == res_dec;
+  }
+
+  ParamPassingTestVideo test_video_;
+
+ private:
+  EncodeParameters encode_parms;
+};
+
+TEST_P(AVxEncoderParmsGetToDecoder, BitstreamParms) {
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  testing::internal::scoped_ptr<libaom_test::VideoSource> video(
+      new libaom_test::Y4mVideoSource(test_video_.name, 0, test_video_.frames));
+  ASSERT_TRUE(video.get() != NULL);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+}
+
+AV1_INSTANTIATE_TEST_CASE(AVxEncoderParmsGetToDecoder,
+                          ::testing::ValuesIn(kAV1EncodeParameterSet));
+}  // namespace
diff --git a/third_party/aom/test/av1_ext_tile_test.cc b/third_party/aom/test/av1_ext_tile_test.cc
new file mode 100644
index 0000000000..424d2f065f
--- /dev/null
+++ b/third_party/aom/test/av1_ext_tile_test.cc
@@ -0,0 +1,215 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <assert.h>
+#include <string>
+#include <vector>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/md5_helper.h"
+#include "test/util.h"
+
+namespace {
+// The number of frames to be encoded/decoded
+const int kLimit = 8;
+// Skip 1 frame to check the frame decoding independency.
+const int kSkip = 5;
+const int kTileSize = 1;
+const int kTIleSizeInPixels = (kTileSize << 6);
+// Fake width and height so that they can be multiples of the tile size.
+const int kImgWidth = 704;
+const int kImgHeight = 576;
+
+// This test tests large scale tile coding case. Non-large-scale tile coding
+// is tested by the tile_independence test.
+class AV1ExtTileTest
+    : public ::libaom_test::CodecTestWith2Params<libaom_test::TestMode, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  AV1ExtTileTest()
+      : EncoderTest(GET_PARAM(0)), encoding_mode_(GET_PARAM(1)),
+        set_cpu_used_(GET_PARAM(2)) {
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+    cfg.w = kImgWidth;
+    cfg.h = kImgHeight;
+    cfg.allow_lowbitdepth = 1;
+
+    decoder_ = codec_->CreateDecoder(cfg, 0);
+    decoder_->Control(AV1_SET_TILE_MODE, 1);
+    decoder_->Control(AV1D_EXT_TILE_DEBUG, 1);
+    decoder_->Control(AV1_SET_DECODE_TILE_ROW, -1);
+    decoder_->Control(AV1_SET_DECODE_TILE_COL, -1);
+
+    // Allocate buffer to store tile image.
+    aom_img_alloc(&tile_img_, AOM_IMG_FMT_I420, kImgWidth, kImgHeight, 32);
+
+    md5_.clear();
+    tile_md5_.clear();
+  }
+
+  virtual ~AV1ExtTileTest() {
+    aom_img_free(&tile_img_);
+    delete decoder_;
+  }
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+
+    cfg_.g_lag_in_frames = 0;
+    cfg_.rc_end_usage = AOM_VBR;
+    cfg_.g_error_resilient = 1;
+
+    cfg_.rc_max_quantizer = 56;
+    cfg_.rc_min_quantizer = 0;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 0) {
+      // Encode setting
+      encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 0);
+      encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
+
+      // TODO(yunqingwang): test single_tile_decoding = 0.
+      encoder->Control(AV1E_SET_SINGLE_TILE_DECODING, 1);
+      // Always use 64x64 max partition.
+      encoder->Control(AV1E_SET_SUPERBLOCK_SIZE, AOM_SUPERBLOCK_SIZE_64X64);
+      // Set tile_columns and tile_rows to MAX values, which guarantees the tile
+      // size of 64 x 64 pixels(i.e. 1 SB) for <= 4k resolution.
+      encoder->Control(AV1E_SET_TILE_COLUMNS, 6);
+      encoder->Control(AV1E_SET_TILE_ROWS, 6);
+    }
+
+    if (video->frame() == 1) {
+      frame_flags_ =
+          AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF;
+    }
+  }
+
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     aom_codec_pts_t pts) {
+    // Skip 1 already decoded frame to be consistent with the decoder in this
+    // test.
+    if (pts == (aom_codec_pts_t)kSkip) return;
+
+    // Calculate MD5 as the reference.
+    ::libaom_test::MD5 md5_res;
+    md5_res.Add(&img);
+    md5_.push_back(md5_res.Get());
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    // Skip decoding 1 frame.
+    if (pkt->data.frame.pts == (aom_codec_pts_t)kSkip) return;
+
+    bool IsLastFrame = (pkt->data.frame.pts == (aom_codec_pts_t)(kLimit - 1));
+
+    // Decode the first (kLimit - 1) frames as whole frame, and decode the last
+    // frame in single tiles.
+    for (int r = 0; r < kImgHeight / kTIleSizeInPixels; ++r) {
+      for (int c = 0; c < kImgWidth / kTIleSizeInPixels; ++c) {
+        if (!IsLastFrame) {
+          decoder_->Control(AV1_SET_DECODE_TILE_ROW, -1);
+          decoder_->Control(AV1_SET_DECODE_TILE_COL, -1);
+        } else {
+          decoder_->Control(AV1_SET_DECODE_TILE_ROW, r);
+          decoder_->Control(AV1_SET_DECODE_TILE_COL, c);
+        }
+
+        const aom_codec_err_t res = decoder_->DecodeFrame(
+            reinterpret_cast<uint8_t *>(pkt->data.frame.buf),
+            pkt->data.frame.sz);
+        if (res != AOM_CODEC_OK) {
+          abort_ = true;
+          ASSERT_EQ(AOM_CODEC_OK, res);
+        }
+        const aom_image_t *img = decoder_->GetDxData().Next();
+
+        if (!IsLastFrame) {
+          if (img) {
+            ::libaom_test::MD5 md5_res;
+            md5_res.Add(img);
+            tile_md5_.push_back(md5_res.Get());
+          }
+          break;
+        }
+
+        const int kMaxMBPlane = 3;
+        for (int plane = 0; plane < kMaxMBPlane; ++plane) {
+          const int shift = (plane == 0) ? 0 : 1;
+          int tile_height = kTIleSizeInPixels >> shift;
+          int tile_width = kTIleSizeInPixels >> shift;
+
+          for (int tr = 0; tr < tile_height; ++tr) {
+            memcpy(tile_img_.planes[plane] +
+                       tile_img_.stride[plane] * (r * tile_height + tr) +
+                       c * tile_width,
+                   img->planes[plane] + img->stride[plane] * tr, tile_width);
+          }
+        }
+      }
+
+      if (!IsLastFrame) break;
+    }
+
+    if (IsLastFrame) {
+      ::libaom_test::MD5 md5_res;
+      md5_res.Add(&tile_img_);
+      tile_md5_.push_back(md5_res.Get());
+    }
+  }
+
+  void TestRoundTrip() {
+    ::libaom_test::I420VideoSource video(
+        "hantro_collage_w352h288.yuv", kImgWidth, kImgHeight, 30, 1, 0, kLimit);
+    cfg_.rc_target_bitrate = 500;
+    cfg_.g_error_resilient = AOM_ERROR_RESILIENT_DEFAULT;
+    cfg_.large_scale_tile = 1;
+    cfg_.g_lag_in_frames = 0;
+    cfg_.g_threads = 1;
+
+    // Tile encoding
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+    // Compare to check if two vectors are equal.
+    ASSERT_EQ(md5_, tile_md5_);
+  }
+
+  ::libaom_test::TestMode encoding_mode_;
+  int set_cpu_used_;
+  ::libaom_test::Decoder *decoder_;
+  aom_image_t tile_img_;
+  std::vector<std::string> md5_;
+  std::vector<std::string> tile_md5_;
+};
+
+TEST_P(AV1ExtTileTest, DecoderResultTest) { TestRoundTrip(); }
+
+AV1_INSTANTIATE_TEST_CASE(
+    // Now only test 2-pass mode.
+    AV1ExtTileTest, ::testing::Values(::libaom_test::kTwoPassGood),
+    ::testing::Range(1, 4));
+
+class AV1ExtTileTestLarge : public AV1ExtTileTest {};
+
+TEST_P(AV1ExtTileTestLarge, DecoderResultTest) { TestRoundTrip(); }
+
+AV1_INSTANTIATE_TEST_CASE(
+    // Now only test 2-pass mode.
+    AV1ExtTileTestLarge, ::testing::Values(::libaom_test::kTwoPassGood),
+    ::testing::Range(0, 1));
+}  // namespace
diff --git a/third_party/aom/test/av1_fwd_txfm1d_test.cc b/third_party/aom/test/av1_fwd_txfm1d_test.cc
new file mode 100644
index 0000000000..49a6668793
--- /dev/null
+++ b/third_party/aom/test/av1_fwd_txfm1d_test.cc
@@ -0,0 +1,105 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "av1/encoder/av1_fwd_txfm1d.h"
+#include "test/av1_txfm_test.h"
+
+using libaom_test::ACMRandom;
+using libaom_test::TYPE_ADST;
+using libaom_test::TYPE_DCT;
+using libaom_test::TYPE_IDTX;
+using libaom_test::TYPE_TXFM;
+using libaom_test::input_base;
+using libaom_test::reference_hybrid_1d;
+
+namespace {
+const int txfm_type_num = 3;
+const TYPE_TXFM txfm_type_ls[txfm_type_num] = { TYPE_DCT, TYPE_ADST,
+                                                TYPE_IDTX };
+
+const int txfm_size_num = 5;
+
+const int txfm_size_ls[] = { 4, 8, 16, 32, 64 };
+
+const TxfmFunc fwd_txfm_func_ls[][txfm_type_num] = {
+  { av1_fdct4_new, av1_fadst4_new, av1_fidentity4_c },
+  { av1_fdct8_new, av1_fadst8_new, av1_fidentity8_c },
+  { av1_fdct16_new, av1_fadst16_new, av1_fidentity16_c },
+  { av1_fdct32_new, NULL, av1_fidentity32_c },
+  { av1_fdct64_new, NULL, NULL },
+};
+
+// the maximum stage number of fwd/inv 1d dct/adst txfm is 12
+const int8_t cos_bit = 14;
+const int8_t range_bit[12] = { 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20 };
+
+TEST(av1_fwd_txfm1d, round_shift) {
+  EXPECT_EQ(round_shift(7, 1), 4);
+  EXPECT_EQ(round_shift(-7, 1), -3);
+
+  EXPECT_EQ(round_shift(7, 2), 2);
+  EXPECT_EQ(round_shift(-7, 2), -2);
+
+  EXPECT_EQ(round_shift(8, 2), 2);
+  EXPECT_EQ(round_shift(-8, 2), -2);
+}
+
+TEST(av1_fwd_txfm1d, av1_cospi_arr_data) {
+  for (int i = 0; i < 7; i++) {
+    for (int j = 0; j < 64; j++) {
+      EXPECT_EQ(av1_cospi_arr_data[i][j],
+                (int32_t)round(cos(M_PI * j / 128) * (1 << (cos_bit_min + i))));
+    }
+  }
+}
+
+TEST(av1_fwd_txfm1d, accuracy) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  for (int si = 0; si < txfm_size_num; ++si) {
+    int txfm_size = txfm_size_ls[si];
+    int32_t *input = new int32_t[txfm_size];
+    int32_t *output = new int32_t[txfm_size];
+    double *ref_input = new double[txfm_size];
+    double *ref_output = new double[txfm_size];
+
+    for (int ti = 0; ti < txfm_type_num; ++ti) {
+      TYPE_TXFM txfm_type = txfm_type_ls[ti];
+      TxfmFunc fwd_txfm_func = fwd_txfm_func_ls[si][ti];
+      int max_error = 7;
+
+      const int count_test_block = 5000;
+      if (fwd_txfm_func != NULL) {
+        for (int ti = 0; ti < count_test_block; ++ti) {
+          for (int ni = 0; ni < txfm_size; ++ni) {
+            input[ni] = rnd.Rand16() % input_base - rnd.Rand16() % input_base;
+            ref_input[ni] = static_cast<double>(input[ni]);
+          }
+
+          fwd_txfm_func(input, output, cos_bit, range_bit);
+          reference_hybrid_1d(ref_input, ref_output, txfm_size, txfm_type);
+
+          for (int ni = 0; ni < txfm_size; ++ni) {
+            ASSERT_LE(
+                abs(output[ni] - static_cast<int32_t>(round(ref_output[ni]))),
+                max_error)
+                << "tx size = " << txfm_size << ", tx type = " << txfm_type;
+          }
+        }
+      }
+    }
+
+    delete[] input;
+    delete[] output;
+    delete[] ref_input;
+    delete[] ref_output;
+  }
+}
+}  // namespace
diff --git a/third_party/aom/test/av1_fwd_txfm2d_test.cc b/third_party/aom/test/av1_fwd_txfm2d_test.cc
new file mode 100644
index 0000000000..75f20536b2
--- /dev/null
+++ b/third_party/aom/test/av1_fwd_txfm2d_test.cc
@@ -0,0 +1,511 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <vector>
+
+#include "config/av1_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/util.h"
+#include "test/av1_txfm_test.h"
+#include "av1/common/av1_txfm.h"
+#include "av1/encoder/hybrid_fwd_txfm.h"
+
+using libaom_test::ACMRandom;
+using libaom_test::TYPE_TXFM;
+using libaom_test::bd;
+using libaom_test::compute_avg_abs_error;
+using libaom_test::input_base;
+
+using std::vector;
+
+namespace {
+// tx_type_, tx_size_, max_error_, max_avg_error_
+typedef ::testing::tuple<TX_TYPE, TX_SIZE, double, double> AV1FwdTxfm2dParam;
+
+class AV1FwdTxfm2d : public ::testing::TestWithParam<AV1FwdTxfm2dParam> {
+ public:
+  virtual void SetUp() {
+    tx_type_ = GET_PARAM(0);
+    tx_size_ = GET_PARAM(1);
+    max_error_ = GET_PARAM(2);
+    max_avg_error_ = GET_PARAM(3);
+    count_ = 500;
+    TXFM_2D_FLIP_CFG fwd_txfm_flip_cfg;
+    av1_get_fwd_txfm_cfg(tx_type_, tx_size_, &fwd_txfm_flip_cfg);
+    amplify_factor_ = libaom_test::get_amplification_factor(tx_type_, tx_size_);
+    tx_width_ = tx_size_wide[fwd_txfm_flip_cfg.tx_size];
+    tx_height_ = tx_size_high[fwd_txfm_flip_cfg.tx_size];
+    ud_flip_ = fwd_txfm_flip_cfg.ud_flip;
+    lr_flip_ = fwd_txfm_flip_cfg.lr_flip;
+
+    fwd_txfm_ = libaom_test::fwd_txfm_func_ls[tx_size_];
+    txfm2d_size_ = tx_width_ * tx_height_;
+    input_ = reinterpret_cast<int16_t *>(
+        aom_memalign(16, sizeof(input_[0]) * txfm2d_size_));
+    output_ = reinterpret_cast<int32_t *>(
+        aom_memalign(16, sizeof(output_[0]) * txfm2d_size_));
+    ref_input_ = reinterpret_cast<double *>(
+        aom_memalign(16, sizeof(ref_input_[0]) * txfm2d_size_));
+    ref_output_ = reinterpret_cast<double *>(
+        aom_memalign(16, sizeof(ref_output_[0]) * txfm2d_size_));
+  }
+
+  void RunFwdAccuracyCheck() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    double avg_abs_error = 0;
+    for (int ci = 0; ci < count_; ci++) {
+      for (int ni = 0; ni < txfm2d_size_; ++ni) {
+        input_[ni] = rnd.Rand16() % input_base;
+        ref_input_[ni] = static_cast<double>(input_[ni]);
+        output_[ni] = 0;
+        ref_output_[ni] = 0;
+      }
+
+      fwd_txfm_(input_, output_, tx_width_, tx_type_, bd);
+
+      if (lr_flip_ && ud_flip_) {
+        libaom_test::fliplrud(ref_input_, tx_width_, tx_height_, tx_width_);
+      } else if (lr_flip_) {
+        libaom_test::fliplr(ref_input_, tx_width_, tx_height_, tx_width_);
+      } else if (ud_flip_) {
+        libaom_test::flipud(ref_input_, tx_width_, tx_height_, tx_width_);
+      }
+
+      libaom_test::reference_hybrid_2d(ref_input_, ref_output_, tx_type_,
+                                       tx_size_);
+
+      double actual_max_error = 0;
+      for (int ni = 0; ni < txfm2d_size_; ++ni) {
+        ref_output_[ni] = round(ref_output_[ni]);
+        const double this_error =
+            fabs(output_[ni] - ref_output_[ni]) / amplify_factor_;
+        actual_max_error = AOMMAX(actual_max_error, this_error);
+      }
+      EXPECT_GE(max_error_, actual_max_error)
+          << "tx_size = " << tx_size_ << ", tx_type = " << tx_type_;
+      if (actual_max_error > max_error_) {  // exit early.
+        break;
+      }
+
+      avg_abs_error += compute_avg_abs_error<int32_t, double>(
+          output_, ref_output_, txfm2d_size_);
+    }
+
+    avg_abs_error /= amplify_factor_;
+    avg_abs_error /= count_;
+    EXPECT_GE(max_avg_error_, avg_abs_error)
+        << "tx_size = " << tx_size_ << ", tx_type = " << tx_type_;
+  }
+
+  virtual void TearDown() {
+    aom_free(input_);
+    aom_free(output_);
+    aom_free(ref_input_);
+    aom_free(ref_output_);
+  }
+
+ private:
+  double max_error_;
+  double max_avg_error_;
+  int count_;
+  double amplify_factor_;
+  TX_TYPE tx_type_;
+  TX_SIZE tx_size_;
+  int tx_width_;
+  int tx_height_;
+  int txfm2d_size_;
+  FwdTxfm2dFunc fwd_txfm_;
+  int16_t *input_;
+  int32_t *output_;
+  double *ref_input_;
+  double *ref_output_;
+  int ud_flip_;  // flip upside down
+  int lr_flip_;  // flip left to right
+};
+
+static double avg_error_ls[TX_SIZES_ALL] = {
+  0.5,   // 4x4 transform
+  0.5,   // 8x8 transform
+  1.2,   // 16x16 transform
+  6.1,   // 32x32 transform
+  3.4,   // 64x64 transform
+  0.57,  // 4x8 transform
+  0.68,  // 8x4 transform
+  0.92,  // 8x16 transform
+  1.1,   // 16x8 transform
+  4.1,   // 16x32 transform
+  6,     // 32x16 transform
+  3.5,   // 32x64 transform
+  5.7,   // 64x32 transform
+  0.6,   // 4x16 transform
+  0.9,   // 16x4 transform
+  1.2,   // 8x32 transform
+  1.7,   // 32x8 transform
+  2.0,   // 16x64 transform
+  4.7,   // 64x16 transform
+};
+
+static double max_error_ls[TX_SIZES_ALL] = {
+  3,    // 4x4 transform
+  5,    // 8x8 transform
+  11,   // 16x16 transform
+  70,   // 32x32 transform
+  64,   // 64x64 transform
+  3.9,  // 4x8 transform
+  4.3,  // 8x4 transform
+  12,   // 8x16 transform
+  12,   // 16x8 transform
+  32,   // 16x32 transform
+  46,   // 32x16 transform
+  136,  // 32x64 transform
+  136,  // 64x32 transform
+  5,    // 4x16 transform
+  6,    // 16x4 transform
+  21,   // 8x32 transform
+  13,   // 32x8 transform
+  30,   // 16x64 transform
+  36,   // 64x16 transform
+};
+
+vector<AV1FwdTxfm2dParam> GetTxfm2dParamList() {
+  vector<AV1FwdTxfm2dParam> param_list;
+  for (int s = 0; s < TX_SIZES; ++s) {
+    const double max_error = max_error_ls[s];
+    const double avg_error = avg_error_ls[s];
+    for (int t = 0; t < TX_TYPES; ++t) {
+      const TX_TYPE tx_type = static_cast<TX_TYPE>(t);
+      const TX_SIZE tx_size = static_cast<TX_SIZE>(s);
+      if (libaom_test::IsTxSizeTypeValid(tx_size, tx_type)) {
+        param_list.push_back(
+            AV1FwdTxfm2dParam(tx_type, tx_size, max_error, avg_error));
+      }
+    }
+  }
+  return param_list;
+}
+
+INSTANTIATE_TEST_CASE_P(C, AV1FwdTxfm2d,
+                        ::testing::ValuesIn(GetTxfm2dParamList()));
+
+TEST_P(AV1FwdTxfm2d, RunFwdAccuracyCheck) { RunFwdAccuracyCheck(); }
+
+TEST(AV1FwdTxfm2d, CfgTest) {
+  for (int bd_idx = 0; bd_idx < BD_NUM; ++bd_idx) {
+    int bd = libaom_test::bd_arr[bd_idx];
+    int8_t low_range = libaom_test::low_range_arr[bd_idx];
+    int8_t high_range = libaom_test::high_range_arr[bd_idx];
+    for (int tx_size = 0; tx_size < TX_SIZES_ALL; ++tx_size) {
+      for (int tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
+        if (libaom_test::IsTxSizeTypeValid(static_cast<TX_SIZE>(tx_size),
+                                           static_cast<TX_TYPE>(tx_type)) ==
+            false) {
+          continue;
+        }
+        TXFM_2D_FLIP_CFG cfg;
+        av1_get_fwd_txfm_cfg(static_cast<TX_TYPE>(tx_type),
+                             static_cast<TX_SIZE>(tx_size), &cfg);
+        int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
+        int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
+        av1_gen_fwd_stage_range(stage_range_col, stage_range_row, &cfg, bd);
+        libaom_test::txfm_stage_range_check(stage_range_col, cfg.stage_num_col,
+                                            cfg.cos_bit_col, low_range,
+                                            high_range);
+        libaom_test::txfm_stage_range_check(stage_range_row, cfg.stage_num_row,
+                                            cfg.cos_bit_row, low_range,
+                                            high_range);
+      }
+    }
+  }
+}
+
+typedef void (*lowbd_fwd_txfm_func)(const int16_t *src_diff, tran_low_t *coeff,
+                                    int diff_stride, TxfmParam *txfm_param);
+
+void AV1FwdTxfm2dMatchTest(TX_SIZE tx_size, lowbd_fwd_txfm_func target_func) {
+  const int bd = 8;
+  TxfmParam param;
+  memset(&param, 0, sizeof(param));
+  const int rows = tx_size_high[tx_size];
+  const int cols = tx_size_wide[tx_size];
+  // printf("%d x %d\n", cols, rows);
+  for (int tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
+    if (libaom_test::IsTxSizeTypeValid(
+            tx_size, static_cast<TX_TYPE>(tx_type)) == false) {
+      continue;
+    }
+
+    FwdTxfm2dFunc ref_func = libaom_test::fwd_txfm_func_ls[tx_size];
+    if (ref_func != NULL) {
+      DECLARE_ALIGNED(32, int16_t, input[64 * 64]) = { 0 };
+      DECLARE_ALIGNED(32, int32_t, output[64 * 64]);
+      DECLARE_ALIGNED(32, int32_t, ref_output[64 * 64]);
+      int input_stride = 64;
+      ACMRandom rnd(ACMRandom::DeterministicSeed());
+      for (int cnt = 0; cnt < 500; ++cnt) {
+        if (cnt == 0) {
+          for (int r = 0; r < rows; ++r) {
+            for (int c = 0; c < cols; ++c) {
+              input[r * input_stride + c] = (1 << bd) - 1;
+            }
+          }
+        } else {
+          for (int r = 0; r < rows; ++r) {
+            for (int c = 0; c < cols; ++c) {
+              input[r * input_stride + c] = rnd.Rand16() % (1 << bd);
+            }
+          }
+        }
+        param.tx_type = (TX_TYPE)tx_type;
+        param.tx_size = (TX_SIZE)tx_size;
+        param.tx_set_type = EXT_TX_SET_ALL16;
+        param.bd = bd;
+        ref_func(input, ref_output, input_stride, (TX_TYPE)tx_type, bd);
+        target_func(input, output, input_stride, &param);
+        const int check_rows = AOMMIN(32, rows);
+        const int check_cols = AOMMIN(32, rows * cols / check_rows);
+        for (int r = 0; r < check_rows; ++r) {
+          for (int c = 0; c < check_cols; ++c) {
+            ASSERT_EQ(ref_output[r * check_cols + c],
+                      output[r * check_cols + c])
+                << "[" << r << "," << c << "] cnt:" << cnt
+                << " tx_size: " << tx_size << " tx_type: " << tx_type;
+          }
+        }
+      }
+    }
+  }
+}
+
+typedef ::testing::tuple<TX_SIZE, lowbd_fwd_txfm_func> LbdFwdTxfm2dParam;
+
+class AV1FwdTxfm2dTest : public ::testing::TestWithParam<LbdFwdTxfm2dParam> {};
+
+TEST_P(AV1FwdTxfm2dTest, match) {
+  AV1FwdTxfm2dMatchTest(GET_PARAM(0), GET_PARAM(1));
+}
+
+using ::testing::Combine;
+using ::testing::Values;
+using ::testing::ValuesIn;
+
+#if HAVE_SSE2
+static TX_SIZE fwd_txfm_for_sse2[] = {
+  TX_4X4,
+  TX_8X8,
+  TX_16X16,
+  TX_32X32,
+  // TX_64X64,
+  TX_4X8,
+  TX_8X4,
+  TX_8X16,
+  TX_16X8,
+  TX_16X32,
+  TX_32X16,
+  // TX_32X64,
+  // TX_64X32,
+  TX_4X16,
+  TX_16X4,
+  TX_8X32,
+  TX_32X8,
+  TX_16X64,
+  TX_64X16,
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, AV1FwdTxfm2dTest,
+                        Combine(ValuesIn(fwd_txfm_for_sse2),
+                                Values(av1_lowbd_fwd_txfm_sse2)));
+#endif  // HAVE_SSE2
+
+#if HAVE_SSE4_1
+static TX_SIZE fwd_txfm_for_sse41[] = {
+  TX_4X4,
+  TX_64X64,
+  TX_32X64,
+  TX_64X32,
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1FwdTxfm2dTest,
+                        Combine(ValuesIn(fwd_txfm_for_sse41),
+                                Values(av1_lowbd_fwd_txfm_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+static TX_SIZE fwd_txfm_for_avx2[] = {
+  TX_4X4,  TX_8X8,  TX_16X16, TX_32X32, TX_64X64, TX_4X8,   TX_8X4,
+  TX_8X16, TX_16X8, TX_16X32, TX_32X16, TX_32X64, TX_64X32, TX_4X16,
+  TX_16X4, TX_8X32, TX_32X8,  TX_16X64, TX_64X16,
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, AV1FwdTxfm2dTest,
+                        Combine(ValuesIn(fwd_txfm_for_avx2),
+                                Values(av1_lowbd_fwd_txfm_avx2)));
+#endif  // HAVE_AVX2
+
+typedef void (*Highbd_fwd_txfm_func)(const int16_t *src_diff, tran_low_t *coeff,
+                                     int diff_stride, TxfmParam *txfm_param);
+
+void AV1HighbdFwdTxfm2dMatchTest(TX_SIZE tx_size,
+                                 Highbd_fwd_txfm_func target_func) {
+  const int bd_ar[2] = { 10, 12 };
+  TxfmParam param;
+  memset(&param, 0, sizeof(param));
+  const int rows = tx_size_high[tx_size];
+  const int cols = tx_size_wide[tx_size];
+  for (int i = 0; i < 2; ++i) {
+    const int bd = bd_ar[i];
+    for (int tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
+      if (libaom_test::IsTxSizeTypeValid(
+              tx_size, static_cast<TX_TYPE>(tx_type)) == false) {
+        continue;
+      }
+
+      FwdTxfm2dFunc ref_func = libaom_test::fwd_txfm_func_ls[tx_size];
+      if (ref_func != NULL) {
+        DECLARE_ALIGNED(32, int16_t, input[64 * 64]) = { 0 };
+        DECLARE_ALIGNED(32, int32_t, output[64 * 64]);
+        DECLARE_ALIGNED(32, int32_t, ref_output[64 * 64]);
+        int input_stride = 64;
+        ACMRandom rnd(ACMRandom::DeterministicSeed());
+        for (int cnt = 0; cnt < 500; ++cnt) {
+          if (cnt == 0) {
+            for (int r = 0; r < rows; ++r) {
+              for (int c = 0; c < cols; ++c) {
+                input[r * input_stride + c] = (1 << bd) - 1;
+              }
+            }
+          } else {
+            for (int r = 0; r < rows; ++r) {
+              for (int c = 0; c < cols; ++c) {
+                input[r * input_stride + c] = rnd.Rand16() % (1 << bd);
+              }
+            }
+          }
+          param.tx_type = (TX_TYPE)tx_type;
+          param.tx_size = (TX_SIZE)tx_size;
+          param.tx_set_type = EXT_TX_SET_ALL16;
+          param.bd = bd;
+
+          ref_func(input, ref_output, input_stride, (TX_TYPE)tx_type, bd);
+          target_func(input, output, input_stride, &param);
+          const int check_rows = AOMMIN(32, rows);
+          const int check_cols = AOMMIN(32, rows * cols / check_rows);
+          for (int r = 0; r < check_rows; ++r) {
+            for (int c = 0; c < check_cols; ++c) {
+              ASSERT_EQ(ref_output[r * check_cols + c],
+                        output[r * check_cols + c])
+                  << "[" << r << "," << c << "] cnt:" << cnt
+                  << " tx_size: " << tx_size << " tx_type: " << tx_type;
+            }
+          }
+        }
+      }
+    }
+  }
+}
+
+void AV1HighbdFwdTxfm2dSpeedTest(TX_SIZE tx_size,
+                                 Highbd_fwd_txfm_func target_func) {
+  const int bd_ar[2] = { 10, 12 };
+  TxfmParam param;
+  memset(&param, 0, sizeof(param));
+  const int rows = tx_size_high[tx_size];
+  const int cols = tx_size_wide[tx_size];
+  const int num_loops = 1000000 / (rows * cols);
+
+  for (int i = 0; i < 2; ++i) {
+    const int bd = bd_ar[i];
+    for (int tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
+      if (libaom_test::IsTxSizeTypeValid(
+              tx_size, static_cast<TX_TYPE>(tx_type)) == false) {
+        continue;
+      }
+
+      FwdTxfm2dFunc ref_func = libaom_test::fwd_txfm_func_ls[tx_size];
+      if (ref_func != NULL) {
+        DECLARE_ALIGNED(32, int16_t, input[64 * 64]) = { 0 };
+        DECLARE_ALIGNED(32, int32_t, output[64 * 64]);
+        DECLARE_ALIGNED(32, int32_t, ref_output[64 * 64]);
+        int input_stride = 64;
+        ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+        for (int r = 0; r < rows; ++r) {
+          for (int c = 0; c < cols; ++c) {
+            input[r * input_stride + c] = rnd.Rand16() % (1 << bd);
+          }
+        }
+
+        param.tx_type = (TX_TYPE)tx_type;
+        param.tx_size = (TX_SIZE)tx_size;
+        param.tx_set_type = EXT_TX_SET_ALL16;
+        param.bd = bd;
+
+        aom_usec_timer ref_timer, test_timer;
+
+        aom_usec_timer_start(&ref_timer);
+        for (int i = 0; i < num_loops; ++i) {
+          ref_func(input, ref_output, input_stride, (TX_TYPE)tx_type, bd);
+        }
+        aom_usec_timer_mark(&ref_timer);
+        const int elapsed_time_c =
+            static_cast<int>(aom_usec_timer_elapsed(&ref_timer));
+
+        aom_usec_timer_start(&test_timer);
+        for (int i = 0; i < num_loops; ++i) {
+          target_func(input, output, input_stride, &param);
+        }
+        aom_usec_timer_mark(&test_timer);
+        const int elapsed_time_simd =
+            static_cast<int>(aom_usec_timer_elapsed(&test_timer));
+
+        printf(
+            "txfm_size[%d] \t txfm_type[%d] \t c_time=%d \t simd_time=%d \t "
+            "gain=%d \n",
+            tx_size, tx_type, elapsed_time_c, elapsed_time_simd,
+            (elapsed_time_c / elapsed_time_simd));
+      }
+    }
+  }
+}
+
+typedef ::testing::tuple<TX_SIZE, Highbd_fwd_txfm_func> HighbdFwdTxfm2dParam;
+
+class AV1HighbdFwdTxfm2dTest
+    : public ::testing::TestWithParam<HighbdFwdTxfm2dParam> {};
+
+TEST_P(AV1HighbdFwdTxfm2dTest, match) {
+  AV1HighbdFwdTxfm2dMatchTest(GET_PARAM(0), GET_PARAM(1));
+}
+
+TEST_P(AV1HighbdFwdTxfm2dTest, DISABLED_Speed) {
+  AV1HighbdFwdTxfm2dSpeedTest(GET_PARAM(0), GET_PARAM(1));
+}
+
+using ::testing::Combine;
+using ::testing::Values;
+using ::testing::ValuesIn;
+
+#if HAVE_SSE4_1
+static TX_SIZE Highbd_fwd_txfm_for_sse4_1[] = {
+  TX_4X4,  TX_8X8,  TX_16X16, TX_32X32, TX_64X64, TX_4X8,   TX_8X4,
+  TX_8X16, TX_16X8, TX_16X32, TX_32X16, TX_32X64, TX_64X32, TX_4X16,
+  TX_16X4, TX_8X32, TX_32X8,  TX_16X64, TX_64X16,
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdFwdTxfm2dTest,
+                        Combine(ValuesIn(Highbd_fwd_txfm_for_sse4_1),
+                                Values(av1_highbd_fwd_txfm)));
+#endif  // HAVE_SSE4_1
+
+}  // namespace
diff --git a/third_party/aom/test/av1_highbd_iht_test.cc b/third_party/aom/test/av1_highbd_iht_test.cc
new file mode 100644
index 0000000000..2d6490c2ad
--- /dev/null
+++ b/third_party/aom/test/av1_highbd_iht_test.cc
@@ -0,0 +1,315 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/av1_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/av1_txfm_test.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "av1/common/enums.h"
+#include "av1/common/scan.h"
+#include "aom_dsp/aom_dsp_common.h"
+#include "aom_ports/mem.h"
+
+namespace {
+
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+
+typedef void (*HbdHtFunc)(const int16_t *input, int32_t *output, int stride,
+                          TX_TYPE tx_type, int bd);
+
+typedef void (*IHbdHtFunc)(const int32_t *coeff, uint16_t *output, int stride,
+                           TX_TYPE tx_type, int bd);
+
+// Test parameter argument list:
+//   <transform reference function,
+//    optimized inverse transform function,
+//    inverse transform reference function,
+//    num_coeffs,
+//    tx_type,
+//    bit_depth>
+typedef tuple<HbdHtFunc, IHbdHtFunc, IHbdHtFunc, int, TX_TYPE, int> IHbdHtParam;
+
+class AV1HighbdInvHTNxN : public ::testing::TestWithParam<IHbdHtParam> {
+ public:
+  virtual ~AV1HighbdInvHTNxN() {}
+
+  virtual void SetUp() {
+    txfm_ref_ = GET_PARAM(0);
+    inv_txfm_ = GET_PARAM(1);
+    inv_txfm_ref_ = GET_PARAM(2);
+    num_coeffs_ = GET_PARAM(3);
+    tx_type_ = GET_PARAM(4);
+    bit_depth_ = GET_PARAM(5);
+
+    input_ = reinterpret_cast<int16_t *>(
+        aom_memalign(16, sizeof(input_[0]) * num_coeffs_));
+
+    // Note:
+    // Inverse transform input buffer is 32-byte aligned
+    // Refer to <root>/av1/encoder/context_tree.c, function,
+    // void alloc_mode_context().
+    coeffs_ = reinterpret_cast<int32_t *>(
+        aom_memalign(32, sizeof(coeffs_[0]) * num_coeffs_));
+    output_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(32, sizeof(output_[0]) * num_coeffs_));
+    output_ref_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(32, sizeof(output_ref_[0]) * num_coeffs_));
+  }
+
+  virtual void TearDown() {
+    aom_free(input_);
+    aom_free(coeffs_);
+    aom_free(output_);
+    aom_free(output_ref_);
+    libaom_test::ClearSystemState();
+  }
+
+ protected:
+  void RunBitexactCheck();
+
+ private:
+  int GetStride() const {
+    if (16 == num_coeffs_) {
+      return 4;
+    } else if (64 == num_coeffs_) {
+      return 8;
+    } else if (256 == num_coeffs_) {
+      return 16;
+    } else if (1024 == num_coeffs_) {
+      return 32;
+    } else if (4096 == num_coeffs_) {
+      return 64;
+    } else {
+      return 0;
+    }
+  }
+
+  HbdHtFunc txfm_ref_;
+  IHbdHtFunc inv_txfm_;
+  IHbdHtFunc inv_txfm_ref_;
+  int num_coeffs_;
+  TX_TYPE tx_type_;
+  int bit_depth_;
+
+  int16_t *input_;
+  int32_t *coeffs_;
+  uint16_t *output_;
+  uint16_t *output_ref_;
+};
+
+void AV1HighbdInvHTNxN::RunBitexactCheck() {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int stride = GetStride();
+  const int num_tests = 20000;
+  const uint16_t mask = (1 << bit_depth_) - 1;
+
+  for (int i = 0; i < num_tests; ++i) {
+    for (int j = 0; j < num_coeffs_; ++j) {
+      input_[j] = (rnd.Rand16() & mask) - (rnd.Rand16() & mask);
+      output_ref_[j] = rnd.Rand16() & mask;
+      output_[j] = output_ref_[j];
+    }
+
+    txfm_ref_(input_, coeffs_, stride, tx_type_, bit_depth_);
+    inv_txfm_ref_(coeffs_, output_ref_, stride, tx_type_, bit_depth_);
+    ASM_REGISTER_STATE_CHECK(
+        inv_txfm_(coeffs_, output_, stride, tx_type_, bit_depth_));
+
+    for (int j = 0; j < num_coeffs_; ++j) {
+      EXPECT_EQ(output_ref_[j], output_[j])
+          << "Not bit-exact result at index: " << j << " At test block: " << i;
+    }
+  }
+}
+
+TEST_P(AV1HighbdInvHTNxN, InvTransResultCheck) { RunBitexactCheck(); }
+
+using ::testing::make_tuple;
+
+#if HAVE_SSE4_1
+#define PARAM_LIST_4X4                                   \
+  &av1_fwd_txfm2d_4x4_c, &av1_inv_txfm2d_add_4x4_sse4_1, \
+      &av1_inv_txfm2d_add_4x4_c, 16
+
+const IHbdHtParam kArrayIhtParam[] = {
+  // 4x4
+  make_tuple(PARAM_LIST_4X4, DCT_DCT, 10),
+  make_tuple(PARAM_LIST_4X4, DCT_DCT, 12),
+  make_tuple(PARAM_LIST_4X4, ADST_DCT, 10),
+  make_tuple(PARAM_LIST_4X4, ADST_DCT, 12),
+  make_tuple(PARAM_LIST_4X4, DCT_ADST, 10),
+  make_tuple(PARAM_LIST_4X4, DCT_ADST, 12),
+  make_tuple(PARAM_LIST_4X4, ADST_ADST, 10),
+  make_tuple(PARAM_LIST_4X4, ADST_ADST, 12),
+  make_tuple(PARAM_LIST_4X4, FLIPADST_DCT, 10),
+  make_tuple(PARAM_LIST_4X4, FLIPADST_DCT, 12),
+  make_tuple(PARAM_LIST_4X4, DCT_FLIPADST, 10),
+  make_tuple(PARAM_LIST_4X4, DCT_FLIPADST, 12),
+  make_tuple(PARAM_LIST_4X4, FLIPADST_FLIPADST, 10),
+  make_tuple(PARAM_LIST_4X4, FLIPADST_FLIPADST, 12),
+  make_tuple(PARAM_LIST_4X4, ADST_FLIPADST, 10),
+  make_tuple(PARAM_LIST_4X4, ADST_FLIPADST, 12),
+  make_tuple(PARAM_LIST_4X4, FLIPADST_ADST, 10),
+  make_tuple(PARAM_LIST_4X4, FLIPADST_ADST, 12),
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdInvHTNxN,
+                        ::testing::ValuesIn(kArrayIhtParam));
+#endif  // HAVE_SSE4_1
+
+typedef void (*HighbdInvTxfm2dFunc)(const int32_t *input, uint8_t *output,
+                                    int stride, const TxfmParam *txfm_param);
+
+typedef ::testing::tuple<const HighbdInvTxfm2dFunc> AV1HighbdInvTxfm2dParam;
+class AV1HighbdInvTxfm2d
+    : public ::testing::TestWithParam<AV1HighbdInvTxfm2dParam> {
+ public:
+  virtual void SetUp() { target_func_ = GET_PARAM(0); }
+  void RunAV1InvTxfm2dTest(TX_TYPE tx_type, TX_SIZE tx_size, int run_times,
+                           int bit_depth);
+
+ private:
+  HighbdInvTxfm2dFunc target_func_;
+};
+
+void AV1HighbdInvTxfm2d::RunAV1InvTxfm2dTest(TX_TYPE tx_type_, TX_SIZE tx_size_,
+                                             int run_times, int bit_depth_) {
+  FwdTxfm2dFunc fwd_func_ = libaom_test::fwd_txfm_func_ls[tx_size_];
+  TxfmParam txfm_param;
+  const int BLK_WIDTH = 64;
+  const int BLK_SIZE = BLK_WIDTH * BLK_WIDTH;
+  DECLARE_ALIGNED(16, int16_t, input[BLK_SIZE]) = { 0 };
+  DECLARE_ALIGNED(32, int32_t, inv_input[BLK_SIZE]) = { 0 };
+  DECLARE_ALIGNED(32, uint16_t, output[BLK_SIZE]) = { 0 };
+  DECLARE_ALIGNED(32, uint16_t, ref_output[BLK_SIZE]) = { 0 };
+  int stride = BLK_WIDTH;
+  int rows = tx_size_high[tx_size_];
+  int cols = tx_size_wide[tx_size_];
+  const int rows_nonezero = AOMMIN(32, rows);
+  const int cols_nonezero = AOMMIN(32, cols);
+  const uint16_t mask = (1 << bit_depth_) - 1;
+  run_times /= (rows * cols);
+  run_times = AOMMAX(1, run_times);
+  const SCAN_ORDER *scan_order = get_default_scan(tx_size_, tx_type_);
+  const int16_t *scan = scan_order->scan;
+  const int16_t eobmax = rows_nonezero * cols_nonezero;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int randTimes = run_times == 1 ? (eobmax) : 1;
+
+  txfm_param.tx_type = tx_type_;
+  txfm_param.tx_size = tx_size_;
+  txfm_param.lossless = 0;
+  txfm_param.bd = bit_depth_;
+  txfm_param.is_hbd = 1;
+  txfm_param.tx_set_type = EXT_TX_SET_ALL16;
+
+  for (int cnt = 0; cnt < randTimes; ++cnt) {
+    for (int r = 0; r < BLK_WIDTH; ++r) {
+      for (int c = 0; c < BLK_WIDTH; ++c) {
+        input[r * cols + c] = (rnd.Rand16() & mask) - (rnd.Rand16() & mask);
+        output[r * stride + c] = rnd.Rand16() & mask;
+
+        ref_output[r * stride + c] = output[r * stride + c];
+      }
+    }
+    fwd_func_(input, inv_input, stride, tx_type_, bit_depth_);
+
+    // produce eob input by setting high freq coeffs to zero
+    const int eob = AOMMIN(cnt + 1, eobmax);
+    for (int i = eob; i < eobmax; i++) {
+      inv_input[scan[i]] = 0;
+    }
+    txfm_param.eob = eob;
+    aom_usec_timer ref_timer, test_timer;
+
+    aom_usec_timer_start(&ref_timer);
+    for (int i = 0; i < run_times; ++i) {
+      av1_highbd_inv_txfm_add_c(inv_input, CONVERT_TO_BYTEPTR(ref_output),
+                                stride, &txfm_param);
+    }
+    aom_usec_timer_mark(&ref_timer);
+    const int elapsed_time_c =
+        static_cast<int>(aom_usec_timer_elapsed(&ref_timer));
+
+    aom_usec_timer_start(&test_timer);
+    for (int i = 0; i < run_times; ++i) {
+      target_func_(inv_input, CONVERT_TO_BYTEPTR(output), stride, &txfm_param);
+    }
+    aom_usec_timer_mark(&test_timer);
+    const int elapsed_time_simd =
+        static_cast<int>(aom_usec_timer_elapsed(&test_timer));
+    if (run_times > 10) {
+      printf(
+          "txfm_size[%d] \t txfm_type[%d] \t c_time=%d \t simd_time=%d \t "
+          "gain=%d \n",
+          tx_size_, tx_type_, elapsed_time_c, elapsed_time_simd,
+          (elapsed_time_c / elapsed_time_simd));
+    } else {
+      for (int r = 0; r < rows; ++r) {
+        for (int c = 0; c < cols; ++c) {
+          ASSERT_EQ(ref_output[r * stride + c], output[r * stride + c])
+              << "[" << r << "," << c << "] " << cnt
+              << " tx_size: " << static_cast<int>(tx_size_)
+              << " tx_type: " << tx_type_ << " eob " << eob;
+        }
+      }
+    }
+  }
+}
+
+TEST_P(AV1HighbdInvTxfm2d, match) {
+  int bitdepth_ar[2] = { 10, 12 };
+  for (int k = 0; k < 2; ++k) {
+    int bd = bitdepth_ar[k];
+    for (int j = 0; j < (int)(TX_SIZES_ALL); ++j) {
+      for (int i = 0; i < (int)TX_TYPES; ++i) {
+        if (libaom_test::IsTxSizeTypeValid(static_cast<TX_SIZE>(j),
+                                           static_cast<TX_TYPE>(i))) {
+          RunAV1InvTxfm2dTest(static_cast<TX_TYPE>(i), static_cast<TX_SIZE>(j),
+                              1, bd);
+        }
+      }
+    }
+  }
+}
+
+TEST_P(AV1HighbdInvTxfm2d, DISABLED_Speed) {
+  int bitdepth_ar[2] = { 10, 12 };
+  for (int k = 0; k < 2; ++k) {
+    int bd = bitdepth_ar[k];
+    for (int j = 0; j < (int)(TX_SIZES_ALL); ++j) {
+      for (int i = 0; i < (int)TX_TYPES; ++i) {
+        if (libaom_test::IsTxSizeTypeValid(static_cast<TX_SIZE>(j),
+                                           static_cast<TX_TYPE>(i))) {
+          RunAV1InvTxfm2dTest(static_cast<TX_TYPE>(i), static_cast<TX_SIZE>(j),
+                              1000000, bd);
+        }
+      }
+    }
+  }
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdInvTxfm2d,
+                        ::testing::Values(av1_highbd_inv_txfm_add_sse4_1));
+#endif
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, AV1HighbdInvTxfm2d,
+                        ::testing::Values(av1_highbd_inv_txfm_add_avx2));
+#endif
+}  // namespace
diff --git a/third_party/aom/test/av1_horz_only_frame_superres_test.cc b/third_party/aom/test/av1_horz_only_frame_superres_test.cc
new file mode 100644
index 0000000000..fd77ef35d7
--- /dev/null
+++ b/third_party/aom/test/av1_horz_only_frame_superres_test.cc
@@ -0,0 +1,362 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <vector>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/aom_timer.h"
+#include "av1/common/convolve.h"
+#include "av1/common/resize.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+namespace {
+const int kTestIters = 10;
+const int kPerfIters = 1000;
+
+const int kVPad = 32;
+const int kHPad = 32;
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+
+template <typename Pixel>
+class TestImage {
+ public:
+  TestImage(int w_src, int h, int superres_denom, int x0, int bd)
+      : w_src_(w_src), h_(h), superres_denom_(superres_denom), x0_(x0),
+        bd_(bd) {
+    assert(bd < 16);
+    assert(bd <= 8 * static_cast<int>(sizeof(Pixel)));
+    assert(9 <= superres_denom && superres_denom <= 16);
+    assert(SCALE_NUMERATOR == 8);
+    assert(0 <= x0_ && x0_ <= RS_SCALE_SUBPEL_MASK);
+
+    w_dst_ = w_src_;
+    av1_calculate_unscaled_superres_size(&w_dst_, NULL, superres_denom);
+
+    src_stride_ = ALIGN_POWER_OF_TWO(w_src_ + 2 * kHPad, 4);
+    dst_stride_ = ALIGN_POWER_OF_TWO(w_dst_ + 2 * kHPad, 4);
+
+    // Allocate image data
+    src_data_.resize(2 * src_block_size());
+    dst_data_.resize(2 * dst_block_size());
+  }
+
+  void Initialize(ACMRandom *rnd);
+  void Check() const;
+
+  int src_stride() const { return src_stride_; }
+  int dst_stride() const { return dst_stride_; }
+
+  int src_block_size() const { return (h_ + 2 * kVPad) * src_stride(); }
+  int dst_block_size() const { return (h_ + 2 * kVPad) * dst_stride(); }
+
+  int src_width() const { return w_src_; }
+  int dst_width() const { return w_dst_; }
+  int height() const { return h_; }
+  int x0() const { return x0_; }
+
+  const Pixel *GetSrcData(bool ref, bool borders) const {
+    const Pixel *block = &src_data_[ref ? 0 : src_block_size()];
+    return borders ? block : block + kHPad + src_stride_ * kVPad;
+  }
+
+  Pixel *GetDstData(bool ref, bool borders) {
+    Pixel *block = &dst_data_[ref ? 0 : dst_block_size()];
+    return borders ? block : block + kHPad + dst_stride_ * kVPad;
+  }
+
+ private:
+  int w_src_, w_dst_, h_, superres_denom_, x0_, bd_;
+  int src_stride_, dst_stride_;
+
+  std::vector<Pixel> src_data_;
+  std::vector<Pixel> dst_data_;
+};
+
+template <typename Pixel>
+void FillEdge(ACMRandom *rnd, int num_pixels, int bd, bool trash, Pixel *data) {
+  if (!trash) {
+    memset(data, 0, sizeof(*data) * num_pixels);
+    return;
+  }
+  const Pixel mask = (1 << bd) - 1;
+  for (int i = 0; i < num_pixels; ++i) data[i] = rnd->Rand16() & mask;
+}
+
+template <typename Pixel>
+void PrepBuffers(ACMRandom *rnd, int w, int h, int stride, int bd,
+                 bool trash_edges, Pixel *data) {
+  assert(rnd);
+  const Pixel mask = (1 << bd) - 1;
+
+  // Fill in the first buffer with random data
+  // Top border
+  FillEdge(rnd, stride * kVPad, bd, trash_edges, data);
+  for (int r = 0; r < h; ++r) {
+    Pixel *row_data = data + (kVPad + r) * stride;
+    // Left border, contents, right border
+    FillEdge(rnd, kHPad, bd, trash_edges, row_data);
+    for (int c = 0; c < w; ++c) row_data[kHPad + c] = rnd->Rand16() & mask;
+    FillEdge(rnd, kHPad, bd, trash_edges, row_data + kHPad + w);
+  }
+  // Bottom border
+  FillEdge(rnd, stride * kVPad, bd, trash_edges, data + stride * (kVPad + h));
+
+  const int bpp = sizeof(*data);
+  const int block_elts = stride * (h + 2 * kVPad);
+  const int block_size = bpp * block_elts;
+
+  // Now copy that to the second buffer
+  memcpy(data + block_elts, data, block_size);
+}
+
+template <typename Pixel>
+void TestImage<Pixel>::Initialize(ACMRandom *rnd) {
+  PrepBuffers(rnd, w_src_, h_, src_stride_, bd_, false, &src_data_[0]);
+  PrepBuffers(rnd, w_dst_, h_, dst_stride_, bd_, true, &dst_data_[0]);
+}
+
+template <typename Pixel>
+void TestImage<Pixel>::Check() const {
+  const int num_pixels = dst_block_size();
+  const Pixel *ref_dst = &dst_data_[0];
+  const Pixel *tst_dst = &dst_data_[num_pixels];
+
+  // If memcmp returns 0, there's nothing to do.
+  if (0 == memcmp(ref_dst, tst_dst, sizeof(*ref_dst) * num_pixels)) return;
+
+  // Otherwise, iterate through the buffer looking for differences, *ignoring
+  // the edges*
+  const int stride = dst_stride_;
+  for (int r = kVPad; r < h_ + kVPad; ++r) {
+    for (int c = kVPad; c < w_dst_ + kHPad; ++c) {
+      const int32_t ref_value = ref_dst[r * stride + c];
+      const int32_t tst_value = tst_dst[r * stride + c];
+
+      EXPECT_EQ(tst_value, ref_value)
+          << "Error at row: " << (r - kVPad) << ", col: " << (c - kHPad)
+          << ", superres_denom: " << superres_denom_ << ", height: " << h_
+          << ", src_width: " << w_src_ << ", dst_width: " << w_dst_
+          << ", x0: " << x0_;
+    }
+  }
+}
+
+template <typename Pixel>
+class ConvolveHorizRSTestBase : public ::testing::Test {
+ public:
+  ConvolveHorizRSTestBase() : image_(NULL) {}
+  virtual ~ConvolveHorizRSTestBase() {}
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+  // Implemented by subclasses (SetUp depends on the parameters passed
+  // in and RunOne depends on the function to be tested. These can't
+  // be templated for low/high bit depths because they have different
+  // numbers of parameters)
+  virtual void SetUp() = 0;
+  virtual void RunOne(bool ref) = 0;
+
+ protected:
+  void SetBitDepth(int bd) { bd_ = bd; }
+
+  void CorrectnessTest() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    for (int i = 0; i < kTestIters; ++i) {
+      for (int superres_denom = 9; superres_denom <= 16; superres_denom++) {
+        // Get a random height between 512 and 767
+        int height = rnd.Rand8() + 512;
+
+        // Get a random src width between 128 and 383
+        int width_src = rnd.Rand8() + 128;
+
+        // x0 is normally calculated by get_upscale_convolve_x0 in
+        // av1/common/resize.c. However, this test should work for
+        // any value of x0 between 0 and RS_SCALE_SUBPEL_MASK
+        // (inclusive), so we choose one at random.
+        int x0 = rnd.Rand16() % (RS_SCALE_SUBPEL_MASK + 1);
+
+        image_ =
+            new TestImage<Pixel>(width_src, height, superres_denom, x0, bd_);
+
+        Prep(&rnd);
+        RunOne(true);
+        RunOne(false);
+        image_->Check();
+
+        delete image_;
+      }
+    }
+  }
+
+  void SpeedTest() {
+    // Pick some specific parameters to test
+    int height = 767;
+    int width_src = 129;
+    int superres_denom = 13;
+    int x0 = RS_SCALE_SUBPEL_MASK >> 1;
+
+    image_ = new TestImage<Pixel>(width_src, height, superres_denom, x0, bd_);
+
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    Prep(&rnd);
+
+    aom_usec_timer ref_timer;
+    aom_usec_timer_start(&ref_timer);
+    for (int i = 0; i < kPerfIters; ++i) RunOne(true);
+    aom_usec_timer_mark(&ref_timer);
+    const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);
+
+    aom_usec_timer tst_timer;
+    aom_usec_timer_start(&tst_timer);
+    for (int i = 0; i < kPerfIters; ++i) RunOne(false);
+    aom_usec_timer_mark(&tst_timer);
+    const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);
+
+    std::cout << "[          ] C time = " << ref_time / 1000
+              << " ms, SIMD time = " << tst_time / 1000 << " ms\n";
+
+    EXPECT_GT(ref_time, tst_time)
+        << "Error: ConvolveHorizRSTest (Speed Test), SIMD slower than C.\n"
+        << "C time: " << ref_time << " us\n"
+        << "SIMD time: " << tst_time << " us\n";
+  }
+
+  void Prep(ACMRandom *rnd) {
+    assert(rnd);
+    image_->Initialize(rnd);
+  }
+
+  int bd_;
+  TestImage<Pixel> *image_;
+};
+
+typedef void (*LowBDConvolveHorizRsFunc)(const uint8_t *src, int src_stride,
+                                         uint8_t *dst, int dst_stride, int w,
+                                         int h, const int16_t *x_filters,
+                                         const int x0_qn, const int x_step_qn);
+
+// Test parameter list:
+//  <tst_fun_>
+typedef tuple<LowBDConvolveHorizRsFunc> LowBDParams;
+
+class LowBDConvolveHorizRSTest
+    : public ConvolveHorizRSTestBase<uint8_t>,
+      public ::testing::WithParamInterface<LowBDParams> {
+ public:
+  virtual ~LowBDConvolveHorizRSTest() {}
+
+  void SetUp() {
+    tst_fun_ = GET_PARAM(0);
+    const int bd = 8;
+    SetBitDepth(bd);
+  }
+
+  void RunOne(bool ref) {
+    const uint8_t *src = image_->GetSrcData(ref, false);
+    uint8_t *dst = image_->GetDstData(ref, false);
+    const int src_stride = image_->src_stride();
+    const int dst_stride = image_->dst_stride();
+    const int width_src = image_->src_width();
+    const int width_dst = image_->dst_width();
+    const int height = image_->height();
+    const int x0_qn = image_->x0();
+
+    const int32_t x_step_qn =
+        av1_get_upscale_convolve_step(width_src, width_dst);
+
+    if (ref) {
+      av1_convolve_horiz_rs_c(src, src_stride, dst, dst_stride, width_dst,
+                              height, &av1_resize_filter_normative[0][0], x0_qn,
+                              x_step_qn);
+    } else {
+      tst_fun_(src, src_stride, dst, dst_stride, width_dst, height,
+               &av1_resize_filter_normative[0][0], x0_qn, x_step_qn);
+    }
+  }
+
+ private:
+  LowBDConvolveHorizRsFunc tst_fun_;
+};
+
+TEST_P(LowBDConvolveHorizRSTest, Correctness) { CorrectnessTest(); }
+TEST_P(LowBDConvolveHorizRSTest, DISABLED_Speed) { SpeedTest(); }
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, LowBDConvolveHorizRSTest,
+                        ::testing::Values(av1_convolve_horiz_rs_sse4_1));
+
+typedef void (*HighBDConvolveHorizRsFunc)(const uint16_t *src, int src_stride,
+                                          uint16_t *dst, int dst_stride, int w,
+                                          int h, const int16_t *x_filters,
+                                          const int x0_qn, const int x_step_qn,
+                                          int bd);
+
+// Test parameter list:
+//  <tst_fun_, bd_>
+typedef tuple<HighBDConvolveHorizRsFunc, int> HighBDParams;
+
+class HighBDConvolveHorizRSTest
+    : public ConvolveHorizRSTestBase<uint16_t>,
+      public ::testing::WithParamInterface<HighBDParams> {
+ public:
+  virtual ~HighBDConvolveHorizRSTest() {}
+
+  void SetUp() {
+    tst_fun_ = GET_PARAM(0);
+    const int bd = GET_PARAM(1);
+    SetBitDepth(bd);
+  }
+
+  void RunOne(bool ref) {
+    const uint16_t *src = image_->GetSrcData(ref, false);
+    uint16_t *dst = image_->GetDstData(ref, false);
+    const int src_stride = image_->src_stride();
+    const int dst_stride = image_->dst_stride();
+    const int width_src = image_->src_width();
+    const int width_dst = image_->dst_width();
+    const int height = image_->height();
+    const int x0_qn = image_->x0();
+
+    const int32_t x_step_qn =
+        av1_get_upscale_convolve_step(width_src, width_dst);
+
+    if (ref) {
+      av1_highbd_convolve_horiz_rs_c(
+          src, src_stride, dst, dst_stride, width_dst, height,
+          &av1_resize_filter_normative[0][0], x0_qn, x_step_qn, bd_);
+    } else {
+      tst_fun_(src, src_stride, dst, dst_stride, width_dst, height,
+               &av1_resize_filter_normative[0][0], x0_qn, x_step_qn, bd_);
+    }
+  }
+
+ private:
+  HighBDConvolveHorizRsFunc tst_fun_;
+};
+
+const int kBDs[] = { 8, 10, 12 };
+
+TEST_P(HighBDConvolveHorizRSTest, Correctness) { CorrectnessTest(); }
+TEST_P(HighBDConvolveHorizRSTest, DISABLED_Speed) { SpeedTest(); }
+
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, HighBDConvolveHorizRSTest,
+    ::testing::Combine(::testing::Values(av1_highbd_convolve_horiz_rs_sse4_1),
+                       ::testing::ValuesIn(kBDs)));
+
+}  // namespace
diff --git a/third_party/aom/test/av1_inv_txfm1d_test.cc b/third_party/aom/test/av1_inv_txfm1d_test.cc
new file mode 100644
index 0000000000..bf3a44ed14
--- /dev/null
+++ b/third_party/aom/test/av1_inv_txfm1d_test.cc
@@ -0,0 +1,157 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+
+#include "test/av1_txfm_test.h"
+#include "test/util.h"
+#include "av1/common/av1_inv_txfm1d.h"
+#include "av1/encoder/av1_fwd_txfm1d.h"
+
+using libaom_test::ACMRandom;
+using libaom_test::input_base;
+
+namespace {
+const int txfm_type_num = 2;
+const int txfm_size_ls[] = { 4, 8, 16, 32, 64 };
+
+const TxfmFunc fwd_txfm_func_ls[][txfm_type_num] = {
+  { av1_fdct4_new, av1_fadst4_new },
+  { av1_fdct8_new, av1_fadst8_new },
+  { av1_fdct16_new, av1_fadst16_new },
+  { av1_fdct32_new, NULL },
+  { av1_fdct64_new, NULL },
+};
+
+const TxfmFunc inv_txfm_func_ls[][txfm_type_num] = {
+  { av1_idct4_new, av1_iadst4_new },
+  { av1_idct8_new, av1_iadst8_new },
+  { av1_idct16_new, av1_iadst16_new },
+  { av1_idct32_new, NULL },
+  { av1_idct64_new, NULL },
+};
+
+// the maximum stage number of fwd/inv 1d dct/adst txfm is 12
+const int8_t cos_bit = 13;
+const int8_t range_bit[12] = { 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20, 20 };
+
+void reference_idct_1d_int(const int32_t *in, int32_t *out, int size) {
+  double input[64];
+  for (int i = 0; i < size; ++i) input[i] = in[i];
+
+  double output[64];
+  libaom_test::reference_idct_1d(input, output, size);
+
+  for (int i = 0; i < size; ++i) {
+    ASSERT_GE(output[i], INT32_MIN);
+    ASSERT_LE(output[i], INT32_MAX);
+    out[i] = static_cast<int32_t>(round(output[i]));
+  }
+}
+
+void random_matrix(int32_t *dst, int len, ACMRandom *rnd) {
+  const int bits = 16;
+  const int maxVal = (1 << (bits - 1)) - 1;
+  const int minVal = -(1 << (bits - 1));
+  for (int i = 0; i < len; ++i) {
+    if (rnd->Rand8() % 10)
+      dst[i] = minVal + rnd->Rand16() % (1 << bits);
+    else
+      dst[i] = rnd->Rand8() % 2 ? minVal : maxVal;
+  }
+}
+
+TEST(av1_inv_txfm1d, InvAccuracyCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int count_test_block = 20000;
+  const int max_error[] = { 6, 10, 19, 31, 40 };
+  ASSERT_EQ(NELEMENTS(max_error), TX_SIZES);
+  ASSERT_EQ(NELEMENTS(inv_txfm_func_ls), TX_SIZES);
+  for (int k = 0; k < count_test_block; ++k) {
+    // choose a random transform to test
+    const TX_SIZE tx_size = static_cast<TX_SIZE>(rnd.Rand8() % TX_SIZES);
+    const int tx_size_pix = txfm_size_ls[tx_size];
+    const TxfmFunc inv_txfm_func = inv_txfm_func_ls[tx_size][0];
+
+    int32_t input[64];
+    random_matrix(input, tx_size_pix, &rnd);
+
+    // 64x64 transform assumes last 32 values are zero.
+    memset(input + 32, 0, 32 * sizeof(input[0]));
+
+    int32_t ref_output[64];
+    reference_idct_1d_int(input, ref_output, tx_size_pix);
+
+    int32_t output[64];
+    inv_txfm_func(input, output, cos_bit, range_bit);
+
+    for (int i = 0; i < tx_size_pix; ++i) {
+      EXPECT_LE(abs(output[i] - ref_output[i]), max_error[tx_size])
+          << "tx_size = " << tx_size << ", i = " << i
+          << ", output[i] = " << output[i]
+          << ", ref_output[i] = " << ref_output[i];
+    }
+  }
+}
+
+static INLINE int get_max_bit(int x) {
+  int max_bit = -1;
+  while (x) {
+    x = x >> 1;
+    max_bit++;
+  }
+  return max_bit;
+}
+
+TEST(av1_inv_txfm1d, get_max_bit) {
+  int max_bit = get_max_bit(8);
+  EXPECT_EQ(max_bit, 3);
+}
+
+TEST(av1_inv_txfm1d, round_trip) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  for (int si = 0; si < NELEMENTS(fwd_txfm_func_ls); ++si) {
+    int txfm_size = txfm_size_ls[si];
+
+    for (int ti = 0; ti < txfm_type_num; ++ti) {
+      TxfmFunc fwd_txfm_func = fwd_txfm_func_ls[si][ti];
+      TxfmFunc inv_txfm_func = inv_txfm_func_ls[si][ti];
+      int max_error = 2;
+
+      if (!fwd_txfm_func) continue;
+
+      const int count_test_block = 5000;
+      for (int ci = 0; ci < count_test_block; ++ci) {
+        int32_t input[64];
+        int32_t output[64];
+        int32_t round_trip_output[64];
+
+        ASSERT_LE(txfm_size, NELEMENTS(input));
+
+        for (int ni = 0; ni < txfm_size; ++ni) {
+          input[ni] = rnd.Rand16() % input_base - rnd.Rand16() % input_base;
+        }
+
+        fwd_txfm_func(input, output, cos_bit, range_bit);
+        inv_txfm_func(output, round_trip_output, cos_bit, range_bit);
+
+        for (int ni = 0; ni < txfm_size; ++ni) {
+          int node_err =
+              abs(input[ni] - round_shift(round_trip_output[ni],
+                                          get_max_bit(txfm_size) - 1));
+          EXPECT_LE(node_err, max_error);
+        }
+      }
+    }
+  }
+}
+
+}  // namespace
diff --git a/third_party/aom/test/av1_inv_txfm2d_test.cc b/third_party/aom/test/av1_inv_txfm2d_test.cc
new file mode 100644
index 0000000000..11e231ba64
--- /dev/null
+++ b/third_party/aom/test/av1_inv_txfm2d_test.cc
@@ -0,0 +1,378 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <vector>
+
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/aom_timer.h"
+#include "av1/common/av1_inv_txfm1d_cfg.h"
+#include "av1/common/scan.h"
+#include "test/acm_random.h"
+#include "test/av1_txfm_test.h"
+#include "test/util.h"
+
+using libaom_test::ACMRandom;
+using libaom_test::InvTxfm2dFunc;
+using libaom_test::LbdInvTxfm2dFunc;
+using libaom_test::bd;
+using libaom_test::compute_avg_abs_error;
+using libaom_test::input_base;
+
+using ::testing::Combine;
+using ::testing::Range;
+using ::testing::Values;
+
+using std::vector;
+
+namespace {
+
+// AV1InvTxfm2dParam argument list:
+// tx_type_, tx_size_, max_error_, max_avg_error_
+typedef ::testing::tuple<TX_TYPE, TX_SIZE, int, double> AV1InvTxfm2dParam;
+
+class AV1InvTxfm2d : public ::testing::TestWithParam<AV1InvTxfm2dParam> {
+ public:
+  virtual void SetUp() {
+    tx_type_ = GET_PARAM(0);
+    tx_size_ = GET_PARAM(1);
+    max_error_ = GET_PARAM(2);
+    max_avg_error_ = GET_PARAM(3);
+  }
+
+  void RunRoundtripCheck() {
+    int tx_w = tx_size_wide[tx_size_];
+    int tx_h = tx_size_high[tx_size_];
+    int txfm2d_size = tx_w * tx_h;
+    const FwdTxfm2dFunc fwd_txfm_func = libaom_test::fwd_txfm_func_ls[tx_size_];
+    const InvTxfm2dFunc inv_txfm_func = libaom_test::inv_txfm_func_ls[tx_size_];
+    double avg_abs_error = 0;
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+    const int count = 500;
+
+    for (int ci = 0; ci < count; ci++) {
+      DECLARE_ALIGNED(16, int16_t, input[64 * 64]) = { 0 };
+      ASSERT_LE(txfm2d_size, NELEMENTS(input));
+
+      for (int ni = 0; ni < txfm2d_size; ++ni) {
+        if (ci == 0) {
+          int extreme_input = input_base - 1;
+          input[ni] = extreme_input;  // extreme case
+        } else {
+          input[ni] = rnd.Rand16() % input_base;
+        }
+      }
+
+      DECLARE_ALIGNED(16, uint16_t, expected[64 * 64]) = { 0 };
+      ASSERT_LE(txfm2d_size, NELEMENTS(expected));
+      if (TxfmUsesApproximation()) {
+        // Compare reference forward HT + inverse HT vs forward HT + inverse HT.
+        double ref_input[64 * 64];
+        ASSERT_LE(txfm2d_size, NELEMENTS(ref_input));
+        for (int ni = 0; ni < txfm2d_size; ++ni) {
+          ref_input[ni] = input[ni];
+        }
+        double ref_coeffs[64 * 64] = { 0 };
+        ASSERT_LE(txfm2d_size, NELEMENTS(ref_coeffs));
+        ASSERT_EQ(tx_type_, DCT_DCT);
+        libaom_test::reference_hybrid_2d(ref_input, ref_coeffs, tx_type_,
+                                         tx_size_);
+        DECLARE_ALIGNED(16, int32_t, ref_coeffs_int[64 * 64]) = { 0 };
+        ASSERT_LE(txfm2d_size, NELEMENTS(ref_coeffs_int));
+        for (int ni = 0; ni < txfm2d_size; ++ni) {
+          ref_coeffs_int[ni] = (int32_t)round(ref_coeffs[ni]);
+        }
+        inv_txfm_func(ref_coeffs_int, expected, tx_w, tx_type_, bd);
+      } else {
+        // Compare original input vs forward HT + inverse HT.
+        for (int ni = 0; ni < txfm2d_size; ++ni) {
+          expected[ni] = input[ni];
+        }
+      }
+
+      DECLARE_ALIGNED(16, int32_t, coeffs[64 * 64]) = { 0 };
+      ASSERT_LE(txfm2d_size, NELEMENTS(coeffs));
+      fwd_txfm_func(input, coeffs, tx_w, tx_type_, bd);
+
+      DECLARE_ALIGNED(16, uint16_t, actual[64 * 64]) = { 0 };
+      ASSERT_LE(txfm2d_size, NELEMENTS(actual));
+      inv_txfm_func(coeffs, actual, tx_w, tx_type_, bd);
+
+      double actual_max_error = 0;
+      for (int ni = 0; ni < txfm2d_size; ++ni) {
+        const double this_error = abs(expected[ni] - actual[ni]);
+        actual_max_error = AOMMAX(actual_max_error, this_error);
+      }
+      EXPECT_GE(max_error_, actual_max_error)
+          << " tx_w: " << tx_w << " tx_h " << tx_h << " tx_type: " << tx_type_;
+      if (actual_max_error > max_error_) {  // exit early.
+        break;
+      }
+      avg_abs_error += compute_avg_abs_error<uint16_t, uint16_t>(
+          expected, actual, txfm2d_size);
+    }
+
+    avg_abs_error /= count;
+    EXPECT_GE(max_avg_error_, avg_abs_error)
+        << " tx_w: " << tx_w << " tx_h " << tx_h << " tx_type: " << tx_type_;
+  }
+
+ private:
+  bool TxfmUsesApproximation() {
+    if (tx_size_wide[tx_size_] == 64 || tx_size_high[tx_size_] == 64) {
+      return true;
+    }
+    return false;
+  }
+
+  int max_error_;
+  double max_avg_error_;
+  TX_TYPE tx_type_;
+  TX_SIZE tx_size_;
+};
+
+static int max_error_ls[TX_SIZES_ALL] = {
+  2,  // 4x4 transform
+  2,  // 8x8 transform
+  2,  // 16x16 transform
+  4,  // 32x32 transform
+  3,  // 64x64 transform
+  2,  // 4x8 transform
+  2,  // 8x4 transform
+  2,  // 8x16 transform
+  2,  // 16x8 transform
+  3,  // 16x32 transform
+  3,  // 32x16 transform
+  5,  // 32x64 transform
+  5,  // 64x32 transform
+  2,  // 4x16 transform
+  2,  // 16x4 transform
+  2,  // 8x32 transform
+  2,  // 32x8 transform
+  3,  // 16x64 transform
+  3,  // 64x16 transform
+};
+
+static double avg_error_ls[TX_SIZES_ALL] = {
+  0.002,  // 4x4 transform
+  0.05,   // 8x8 transform
+  0.07,   // 16x16 transform
+  0.4,    // 32x32 transform
+  0.3,    // 64x64 transform
+  0.02,   // 4x8 transform
+  0.02,   // 8x4 transform
+  0.04,   // 8x16 transform
+  0.07,   // 16x8 transform
+  0.4,    // 16x32 transform
+  0.5,    // 32x16 transform
+  0.38,   // 32x64 transform
+  0.39,   // 64x32 transform
+  0.2,    // 4x16 transform
+  0.2,    // 16x4 transform
+  0.2,    // 8x32 transform
+  0.2,    // 32x8 transform
+  0.38,   // 16x64 transform
+  0.38,   // 64x16 transform
+};
+
+vector<AV1InvTxfm2dParam> GetInvTxfm2dParamList() {
+  vector<AV1InvTxfm2dParam> param_list;
+  for (int s = 0; s < TX_SIZES; ++s) {
+    const int max_error = max_error_ls[s];
+    const double avg_error = avg_error_ls[s];
+    for (int t = 0; t < TX_TYPES; ++t) {
+      const TX_TYPE tx_type = static_cast<TX_TYPE>(t);
+      const TX_SIZE tx_size = static_cast<TX_SIZE>(s);
+      if (libaom_test::IsTxSizeTypeValid(tx_size, tx_type)) {
+        param_list.push_back(
+            AV1InvTxfm2dParam(tx_type, tx_size, max_error, avg_error));
+      }
+    }
+  }
+  return param_list;
+}
+
+INSTANTIATE_TEST_CASE_P(C, AV1InvTxfm2d,
+                        ::testing::ValuesIn(GetInvTxfm2dParamList()));
+
+TEST_P(AV1InvTxfm2d, RunRoundtripCheck) { RunRoundtripCheck(); }
+
+TEST(AV1InvTxfm2d, CfgTest) {
+  for (int bd_idx = 0; bd_idx < BD_NUM; ++bd_idx) {
+    int bd = libaom_test::bd_arr[bd_idx];
+    int8_t low_range = libaom_test::low_range_arr[bd_idx];
+    int8_t high_range = libaom_test::high_range_arr[bd_idx];
+    for (int tx_size = 0; tx_size < TX_SIZES_ALL; ++tx_size) {
+      for (int tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
+        if (libaom_test::IsTxSizeTypeValid(static_cast<TX_SIZE>(tx_size),
+                                           static_cast<TX_TYPE>(tx_type)) ==
+            false) {
+          continue;
+        }
+        TXFM_2D_FLIP_CFG cfg;
+        av1_get_inv_txfm_cfg(static_cast<TX_TYPE>(tx_type),
+                             static_cast<TX_SIZE>(tx_size), &cfg);
+        int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
+        int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
+        av1_gen_inv_stage_range(stage_range_col, stage_range_row, &cfg,
+                                (TX_SIZE)tx_size, bd);
+        libaom_test::txfm_stage_range_check(stage_range_col, cfg.stage_num_col,
+                                            cfg.cos_bit_col, low_range,
+                                            high_range);
+        libaom_test::txfm_stage_range_check(stage_range_row, cfg.stage_num_row,
+                                            cfg.cos_bit_row, low_range,
+                                            high_range);
+      }
+    }
+  }
+}
+
+typedef ::testing::tuple<const LbdInvTxfm2dFunc> AV1LbdInvTxfm2dParam;
+class AV1LbdInvTxfm2d : public ::testing::TestWithParam<AV1LbdInvTxfm2dParam> {
+ public:
+  virtual void SetUp() { target_func_ = GET_PARAM(0); }
+  void RunAV1InvTxfm2dTest(TX_TYPE tx_type, TX_SIZE tx_size, int run_times);
+
+ private:
+  LbdInvTxfm2dFunc target_func_;
+};
+
+void AV1LbdInvTxfm2d::RunAV1InvTxfm2dTest(TX_TYPE tx_type, TX_SIZE tx_size,
+                                          int run_times) {
+  FwdTxfm2dFunc fwd_func_ = libaom_test::fwd_txfm_func_ls[tx_size];
+  InvTxfm2dFunc ref_func_ = libaom_test::inv_txfm_func_ls[tx_size];
+  if (fwd_func_ == NULL || ref_func_ == NULL || target_func_ == NULL) {
+    return;
+  }
+  const int bd = 8;
+  const int BLK_WIDTH = 64;
+  const int BLK_SIZE = BLK_WIDTH * BLK_WIDTH;
+  DECLARE_ALIGNED(16, int16_t, input[BLK_SIZE]) = { 0 };
+  DECLARE_ALIGNED(32, int32_t, inv_input[BLK_SIZE]) = { 0 };
+  DECLARE_ALIGNED(16, uint8_t, output[BLK_SIZE]) = { 0 };
+  DECLARE_ALIGNED(16, uint16_t, ref_output[BLK_SIZE]) = { 0 };
+  int stride = BLK_WIDTH;
+  int rows = tx_size_high[tx_size];
+  int cols = tx_size_wide[tx_size];
+  const int rows_nonezero = AOMMIN(32, rows);
+  const int cols_nonezero = AOMMIN(32, cols);
+  run_times /= (rows * cols);
+  run_times = AOMMAX(1, run_times);
+  const SCAN_ORDER *scan_order = get_default_scan(tx_size, tx_type);
+  const int16_t *scan = scan_order->scan;
+  const int16_t eobmax = rows_nonezero * cols_nonezero;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  int randTimes = run_times == 1 ? (eobmax + 500) : 1;
+  for (int cnt = 0; cnt < randTimes; ++cnt) {
+    const int16_t max_in = (1 << (bd)) - 1;
+    for (int r = 0; r < BLK_WIDTH; ++r) {
+      for (int c = 0; c < BLK_WIDTH; ++c) {
+        input[r * cols + c] = (cnt == 0) ? max_in : rnd.Rand8Extremes();
+        output[r * stride + c] = (cnt == 0) ? 128 : rnd.Rand8();
+        ref_output[r * stride + c] = output[r * stride + c];
+      }
+    }
+    fwd_func_(input, inv_input, stride, tx_type, bd);
+
+    // produce eob input by setting high freq coeffs to zero
+    const int eob = AOMMIN(cnt + 1, eobmax);
+    for (int i = eob; i < eobmax; i++) {
+      inv_input[scan[i]] = 0;
+    }
+
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      ref_func_(inv_input, ref_output, stride, tx_type, bd);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      target_func_(inv_input, output, stride, tx_type, tx_size, eob);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    if (run_times > 10) {
+      printf("txfm[%d] %3dx%-3d:%7.2f/%7.2fns", tx_type, cols, rows, time1,
+             time2);
+      printf("(%3.2f)\n", time1 / time2);
+    }
+    for (int r = 0; r < rows; ++r) {
+      for (int c = 0; c < cols; ++c) {
+        uint8_t ref_value = static_cast<uint8_t>(ref_output[r * stride + c]);
+        ASSERT_EQ(ref_value, output[r * stride + c])
+            << "[" << r << "," << c << "] " << cnt
+            << " tx_size: " << static_cast<int>(tx_size)
+            << " tx_type: " << tx_type << " eob " << eob;
+      }
+    }
+  }
+}
+
+TEST_P(AV1LbdInvTxfm2d, match) {
+  for (int j = 0; j < (int)(TX_SIZES_ALL); ++j) {
+    for (int i = 0; i < (int)TX_TYPES; ++i) {
+      if (libaom_test::IsTxSizeTypeValid(static_cast<TX_SIZE>(j),
+                                         static_cast<TX_TYPE>(i))) {
+        RunAV1InvTxfm2dTest(static_cast<TX_TYPE>(i), static_cast<TX_SIZE>(j),
+                            1);
+      }
+    }
+  }
+}
+
+TEST_P(AV1LbdInvTxfm2d, DISABLED_Speed) {
+  for (int j = 0; j < (int)(TX_SIZES_ALL); ++j) {
+    for (int i = 0; i < (int)TX_TYPES; ++i) {
+      if (libaom_test::IsTxSizeTypeValid(static_cast<TX_SIZE>(j),
+                                         static_cast<TX_TYPE>(i))) {
+        RunAV1InvTxfm2dTest(static_cast<TX_TYPE>(i), static_cast<TX_SIZE>(j),
+                            10000000);
+      }
+    }
+  }
+}
+
+#if HAVE_SSSE3
+#if defined(_MSC_VER) || defined(__SSSE3__)
+#include "av1/common/x86/av1_inv_txfm_ssse3.h"
+INSTANTIATE_TEST_CASE_P(SSSE3, AV1LbdInvTxfm2d,
+                        ::testing::Values(av1_lowbd_inv_txfm2d_add_ssse3));
+#endif  // _MSC_VER || __SSSE3__
+#endif  // HAVE_SSSE3
+
+#if HAVE_AVX2
+extern "C" void av1_lowbd_inv_txfm2d_add_avx2(const int32_t *input,
+                                              uint8_t *output, int stride,
+                                              TX_TYPE tx_type, TX_SIZE tx_size,
+                                              int eob);
+
+INSTANTIATE_TEST_CASE_P(AVX2, AV1LbdInvTxfm2d,
+                        ::testing::Values(av1_lowbd_inv_txfm2d_add_avx2));
+#endif  // HAVE_AVX2
+
+#if HAVE_NEON
+
+extern "C" void av1_lowbd_inv_txfm2d_add_neon(const int32_t *input,
+                                              uint8_t *output, int stride,
+                                              TX_TYPE tx_type, TX_SIZE tx_size,
+                                              int eob);
+
+INSTANTIATE_TEST_CASE_P(NEON, AV1LbdInvTxfm2d,
+                        ::testing::Values(av1_lowbd_inv_txfm2d_add_neon));
+#endif  // HAVE_NEON
+
+}  // namespace
diff --git a/third_party/aom/test/av1_quantize_test.cc b/third_party/aom/test/av1_quantize_test.cc
new file mode 100644
index 0000000000..aaf0939181
--- /dev/null
+++ b/third_party/aom/test/av1_quantize_test.cc
@@ -0,0 +1,239 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include <stdlib.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/av1_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "av1/common/scan.h"
+
+namespace {
+
+typedef void (*QuantizeFpFunc)(
+    const tran_low_t *coeff_ptr, intptr_t count, const int16_t *zbin_ptr,
+    const int16_t *round_ptr, const int16_t *quant_ptr,
+    const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,
+    tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr,
+    const int16_t *scan, const int16_t *iscan, int log_scale);
+
+struct QuantizeFuncParams {
+  QuantizeFuncParams(QuantizeFpFunc qF = NULL, QuantizeFpFunc qRefF = NULL,
+                     int count = 16)
+      : qFunc(qF), qFuncRef(qRefF), coeffCount(count) {}
+  QuantizeFpFunc qFunc;
+  QuantizeFpFunc qFuncRef;
+  int coeffCount;
+};
+
+using libaom_test::ACMRandom;
+
+const int numTests = 1000;
+const int maxSize = 1024;
+const int roundFactorRange = 127;
+const int dequantRange = 32768;
+const int coeffRange = (1 << 20) - 1;
+
+class AV1QuantizeTest : public ::testing::TestWithParam<QuantizeFuncParams> {
+ public:
+  void RunQuantizeTest() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    DECLARE_ALIGNED(16, tran_low_t, coeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, int16_t, zbin_ptr[8]);
+    DECLARE_ALIGNED(16, int16_t, round_ptr[8]);
+    DECLARE_ALIGNED(16, int16_t, quant_ptr[8]);
+    DECLARE_ALIGNED(16, int16_t, quant_shift_ptr[8]);
+    DECLARE_ALIGNED(16, tran_low_t, qcoeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, tran_low_t, dqcoeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, tran_low_t, ref_qcoeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, tran_low_t, ref_dqcoeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, int16_t, dequant_ptr[8]);
+    uint16_t eob;
+    uint16_t ref_eob;
+    int err_count_total = 0;
+    int first_failure = -1;
+    int count = params_.coeffCount;
+    const TX_SIZE txSize = getTxSize(count);
+    int log_scale = (txSize == TX_32X32);
+    QuantizeFpFunc quanFunc = params_.qFunc;
+    QuantizeFpFunc quanFuncRef = params_.qFuncRef;
+
+    const SCAN_ORDER scanOrder = av1_default_scan_orders[txSize];
+    for (int i = 0; i < numTests; i++) {
+      int err_count = 0;
+      ref_eob = eob = -1;
+      for (int j = 0; j < count; j++) {
+        coeff_ptr[j] = rnd(coeffRange);
+      }
+
+      for (int j = 0; j < 2; j++) {
+        zbin_ptr[j] = rnd.Rand16();
+        quant_shift_ptr[j] = rnd.Rand16();
+        // int16_t positive
+        dequant_ptr[j] = abs(rnd(dequantRange));
+        quant_ptr[j] = (1 << 16) / dequant_ptr[j];
+        round_ptr[j] = (abs(rnd(roundFactorRange)) * dequant_ptr[j]) >> 7;
+      }
+      for (int j = 2; j < 8; ++j) {
+        zbin_ptr[j] = zbin_ptr[1];
+        quant_shift_ptr[j] = quant_shift_ptr[1];
+        dequant_ptr[j] = dequant_ptr[1];
+        quant_ptr[j] = quant_ptr[1];
+        round_ptr[j] = round_ptr[1];
+      }
+      quanFuncRef(coeff_ptr, count, zbin_ptr, round_ptr, quant_ptr,
+                  quant_shift_ptr, ref_qcoeff_ptr, ref_dqcoeff_ptr, dequant_ptr,
+                  &ref_eob, scanOrder.scan, scanOrder.iscan, log_scale);
+
+      ASM_REGISTER_STATE_CHECK(
+          quanFunc(coeff_ptr, count, zbin_ptr, round_ptr, quant_ptr,
+                   quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr, &eob,
+                   scanOrder.scan, scanOrder.iscan, log_scale));
+
+      for (int j = 0; j < count; ++j) {
+        err_count += (ref_qcoeff_ptr[j] != qcoeff_ptr[j]) |
+                     (ref_dqcoeff_ptr[j] != dqcoeff_ptr[j]);
+        ASSERT_EQ(ref_qcoeff_ptr[j], qcoeff_ptr[j])
+            << "qcoeff error: i = " << i << " j = " << j << "\n";
+        EXPECT_EQ(ref_dqcoeff_ptr[j], dqcoeff_ptr[j])
+            << "dqcoeff error: i = " << i << " j = " << j << "\n";
+      }
+      EXPECT_EQ(ref_eob, eob) << "eob error: "
+                              << "i = " << i << "\n";
+      err_count += (ref_eob != eob);
+      if (err_count && !err_count_total) {
+        first_failure = i;
+      }
+      err_count_total += err_count;
+    }
+    EXPECT_EQ(0, err_count_total)
+        << "Error: Quantization Test, C output doesn't match SSE2 output. "
+        << "First failed at test case " << first_failure;
+  }
+
+  void RunEobTest() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    DECLARE_ALIGNED(16, tran_low_t, coeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, int16_t, zbin_ptr[8]);
+    DECLARE_ALIGNED(16, int16_t, round_ptr[8]);
+    DECLARE_ALIGNED(16, int16_t, quant_ptr[8]);
+    DECLARE_ALIGNED(16, int16_t, quant_shift_ptr[8]);
+    DECLARE_ALIGNED(16, tran_low_t, qcoeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, tran_low_t, dqcoeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, tran_low_t, ref_qcoeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, tran_low_t, ref_dqcoeff_ptr[maxSize]);
+    DECLARE_ALIGNED(16, int16_t, dequant_ptr[8]);
+    uint16_t eob;
+    uint16_t ref_eob;
+    int count = params_.coeffCount;
+    const TX_SIZE txSize = getTxSize(count);
+    int log_scale = (txSize == TX_32X32);
+    QuantizeFpFunc quanFunc = params_.qFunc;
+    QuantizeFpFunc quanFuncRef = params_.qFuncRef;
+    const SCAN_ORDER scanOrder = av1_default_scan_orders[txSize];
+
+    for (int i = 0; i < numTests; i++) {
+      ref_eob = eob = -1;
+      for (int j = 0; j < count; j++) {
+        coeff_ptr[j] = 0;
+      }
+
+      coeff_ptr[rnd(count)] = rnd(coeffRange);
+      coeff_ptr[rnd(count)] = rnd(coeffRange);
+      coeff_ptr[rnd(count)] = rnd(coeffRange);
+
+      for (int j = 0; j < 2; j++) {
+        zbin_ptr[j] = rnd.Rand16();
+        quant_shift_ptr[j] = rnd.Rand16();
+        // int16_t positive
+        dequant_ptr[j] = abs(rnd(dequantRange));
+        quant_ptr[j] = (1 << 16) / dequant_ptr[j];
+        round_ptr[j] = (abs(rnd(roundFactorRange)) * dequant_ptr[j]) >> 7;
+      }
+      for (int j = 2; j < 8; ++j) {
+        zbin_ptr[j] = zbin_ptr[1];
+        quant_shift_ptr[j] = quant_shift_ptr[1];
+        dequant_ptr[j] = dequant_ptr[1];
+        quant_ptr[j] = quant_ptr[1];
+        round_ptr[j] = round_ptr[1];
+      }
+
+      quanFuncRef(coeff_ptr, count, zbin_ptr, round_ptr, quant_ptr,
+                  quant_shift_ptr, ref_qcoeff_ptr, ref_dqcoeff_ptr, dequant_ptr,
+                  &ref_eob, scanOrder.scan, scanOrder.iscan, log_scale);
+
+      ASM_REGISTER_STATE_CHECK(
+          quanFunc(coeff_ptr, count, zbin_ptr, round_ptr, quant_ptr,
+                   quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr, &eob,
+                   scanOrder.scan, scanOrder.iscan, log_scale));
+      EXPECT_EQ(ref_eob, eob) << "eob error: "
+                              << "i = " << i << "\n";
+    }
+  }
+
+  virtual void SetUp() { params_ = GetParam(); }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+  virtual ~AV1QuantizeTest() {}
+
+ private:
+  TX_SIZE getTxSize(int count) {
+    switch (count) {
+      case 16: return TX_4X4;
+      case 64: return TX_8X8;
+      case 256: return TX_16X16;
+      case 1024: return TX_32X32;
+      default: return TX_4X4;
+    }
+  }
+
+  QuantizeFuncParams params_;
+};
+
+TEST_P(AV1QuantizeTest, BitExactCheck) { RunQuantizeTest(); }
+TEST_P(AV1QuantizeTest, EobVerify) { RunEobTest(); }
+
+#if HAVE_SSE4_1
+const QuantizeFuncParams qfps[4] = {
+  QuantizeFuncParams(&av1_highbd_quantize_fp_sse4_1, &av1_highbd_quantize_fp_c,
+                     16),
+  QuantizeFuncParams(&av1_highbd_quantize_fp_sse4_1, &av1_highbd_quantize_fp_c,
+                     64),
+  QuantizeFuncParams(&av1_highbd_quantize_fp_sse4_1, &av1_highbd_quantize_fp_c,
+                     256),
+  QuantizeFuncParams(&av1_highbd_quantize_fp_sse4_1, &av1_highbd_quantize_fp_c,
+                     1024),
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1QuantizeTest, ::testing::ValuesIn(qfps));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+const QuantizeFuncParams qfps_avx2[4] = {
+  QuantizeFuncParams(&av1_highbd_quantize_fp_avx2, &av1_highbd_quantize_fp_c,
+                     16),
+  QuantizeFuncParams(&av1_highbd_quantize_fp_avx2, &av1_highbd_quantize_fp_c,
+                     64),
+  QuantizeFuncParams(&av1_highbd_quantize_fp_avx2, &av1_highbd_quantize_fp_c,
+                     256),
+  QuantizeFuncParams(&av1_highbd_quantize_fp_avx2, &av1_highbd_quantize_fp_c,
+                     1024),
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, AV1QuantizeTest, ::testing::ValuesIn(qfps_avx2));
+#endif  // HAVE_AVX2
+
+}  // namespace
diff --git a/third_party/aom/test/av1_round_shift_array_test.cc b/third_party/aom/test/av1_round_shift_array_test.cc
new file mode 100644
index 0000000000..181a394604
--- /dev/null
+++ b/third_party/aom/test/av1_round_shift_array_test.cc
@@ -0,0 +1,129 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom_mem/aom_mem.h"
+#include "aom_ports/aom_timer.h"
+#include "aom_ports/mem.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/util.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace AV1CompRoundShift {
+
+typedef void (*comp_round_shift_array_func)(int32_t *arr, int size, int bit);
+
+#if HAVE_SSE4_1 || HAVE_NEON
+const int kValidBitCheck[] = {
+  -4, -3, -2, -1, 0, 1, 2, 3, 4,
+};
+#endif  // HAVE_SSE4_1 || HAVE_NEON
+
+typedef ::testing::tuple<comp_round_shift_array_func, BLOCK_SIZE, int>
+    CompRoundShiftParam;
+
+class AV1CompRoundShiftTest
+    : public ::testing::TestWithParam<CompRoundShiftParam> {
+ public:
+  ~AV1CompRoundShiftTest();
+
+  void SetUp() { rnd_.Reset(libaom_test::ACMRandom::DeterministicSeed()); }
+  void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void RunCheckOutput(comp_round_shift_array_func test_impl, BLOCK_SIZE bsize,
+                      int bit);
+  void RunSpeedTest(comp_round_shift_array_func test_impl, BLOCK_SIZE bsize,
+                    int bit);
+
+  libaom_test::ACMRandom rnd_;
+};
+
+AV1CompRoundShiftTest::~AV1CompRoundShiftTest() { ; }
+
+void AV1CompRoundShiftTest::RunCheckOutput(
+    comp_round_shift_array_func test_impl, BLOCK_SIZE bsize, int bit) {
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+  const int blk_wd = 64;
+  DECLARE_ALIGNED(32, int32_t, pred_[blk_wd]);
+  DECLARE_ALIGNED(32, int32_t, ref_buffer_[blk_wd]);
+  for (int i = 0; i < (blk_wd); ++i) {
+    ref_buffer_[i] = pred_[i] = rnd_.Rand31() / 16;
+  }
+  av1_round_shift_array_c(ref_buffer_, w, bit);
+  test_impl(pred_, w, bit);
+  for (int x = 0; x < w; ++x) {
+    ASSERT_EQ(ref_buffer_[x], pred_[x]) << w << "x" << h << "mismatch @"
+                                        << "(" << x << ")";
+  }
+}
+
+void AV1CompRoundShiftTest::RunSpeedTest(comp_round_shift_array_func test_impl,
+                                         BLOCK_SIZE bsize, int bit) {
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+  const int blk_wd = 64;
+  DECLARE_ALIGNED(32, int32_t, ref_buffer_[blk_wd]);
+  for (int i = 0; i < (blk_wd); ++i) {
+    ref_buffer_[i] = rnd_.Rand31();
+  }
+
+  const int num_loops = 1000000000 / (w + h);
+  comp_round_shift_array_func funcs[2] = { av1_round_shift_array_c, test_impl };
+  double elapsed_time[2] = { 0 };
+  for (int i = 0; i < 2; ++i) {
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    comp_round_shift_array_func func = funcs[i];
+    for (int j = 0; j < num_loops; ++j) {
+      func(ref_buffer_, w, bit);
+    }
+    aom_usec_timer_mark(&timer);
+    double time = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    elapsed_time[i] = 1000.0 * time / num_loops;
+  }
+  printf("av1_round_shift_array %3dx%-3d: bit : %d %7.2f/%7.2fns", w, h, bit,
+         elapsed_time[0], elapsed_time[1]);
+  printf("(%3.2f)\n", elapsed_time[0] / elapsed_time[1]);
+}
+
+TEST_P(AV1CompRoundShiftTest, CheckOutput) {
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), GET_PARAM(2));
+}
+
+TEST_P(AV1CompRoundShiftTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(0), GET_PARAM(1), GET_PARAM(2));
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AV1CompRoundShiftTest,
+    ::testing::Combine(::testing::Values(&av1_round_shift_array_sse4_1),
+                       ::testing::ValuesIn(txsize_to_bsize),
+                       ::testing::ValuesIn(kValidBitCheck)));
+#endif
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(
+    NEON, AV1CompRoundShiftTest,
+    ::testing::Combine(::testing::Values(&av1_round_shift_array_neon),
+                       ::testing::ValuesIn(txsize_to_bsize),
+                       ::testing::ValuesIn(kValidBitCheck)));
+#endif
+
+};  // namespace AV1CompRoundShift
diff --git a/third_party/aom/test/av1_txfm_test.cc b/third_party/aom/test/av1_txfm_test.cc
new file mode 100644
index 0000000000..d5b0ce3255
--- /dev/null
+++ b/third_party/aom/test/av1_txfm_test.cc
@@ -0,0 +1,371 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdio.h>
+#include "test/av1_txfm_test.h"
+
+namespace libaom_test {
+
+int get_txfm1d_size(TX_SIZE tx_size) { return tx_size_wide[tx_size]; }
+
+void get_txfm1d_type(TX_TYPE txfm2d_type, TYPE_TXFM *type0, TYPE_TXFM *type1) {
+  switch (txfm2d_type) {
+    case DCT_DCT:
+      *type0 = TYPE_DCT;
+      *type1 = TYPE_DCT;
+      break;
+    case ADST_DCT:
+      *type0 = TYPE_ADST;
+      *type1 = TYPE_DCT;
+      break;
+    case DCT_ADST:
+      *type0 = TYPE_DCT;
+      *type1 = TYPE_ADST;
+      break;
+    case ADST_ADST:
+      *type0 = TYPE_ADST;
+      *type1 = TYPE_ADST;
+      break;
+    case FLIPADST_DCT:
+      *type0 = TYPE_ADST;
+      *type1 = TYPE_DCT;
+      break;
+    case DCT_FLIPADST:
+      *type0 = TYPE_DCT;
+      *type1 = TYPE_ADST;
+      break;
+    case FLIPADST_FLIPADST:
+      *type0 = TYPE_ADST;
+      *type1 = TYPE_ADST;
+      break;
+    case ADST_FLIPADST:
+      *type0 = TYPE_ADST;
+      *type1 = TYPE_ADST;
+      break;
+    case FLIPADST_ADST:
+      *type0 = TYPE_ADST;
+      *type1 = TYPE_ADST;
+      break;
+    case IDTX:
+      *type0 = TYPE_IDTX;
+      *type1 = TYPE_IDTX;
+      break;
+    case H_DCT:
+      *type0 = TYPE_IDTX;
+      *type1 = TYPE_DCT;
+      break;
+    case V_DCT:
+      *type0 = TYPE_DCT;
+      *type1 = TYPE_IDTX;
+      break;
+    case H_ADST:
+      *type0 = TYPE_IDTX;
+      *type1 = TYPE_ADST;
+      break;
+    case V_ADST:
+      *type0 = TYPE_ADST;
+      *type1 = TYPE_IDTX;
+      break;
+    case H_FLIPADST:
+      *type0 = TYPE_IDTX;
+      *type1 = TYPE_ADST;
+      break;
+    case V_FLIPADST:
+      *type0 = TYPE_ADST;
+      *type1 = TYPE_IDTX;
+      break;
+    default:
+      *type0 = TYPE_DCT;
+      *type1 = TYPE_DCT;
+      assert(0);
+      break;
+  }
+}
+
+double Sqrt2 = pow(2, 0.5);
+double invSqrt2 = 1 / pow(2, 0.5);
+
+double dct_matrix(double n, double k, int size) {
+  return cos(M_PI * (2 * n + 1) * k / (2 * size));
+}
+
+void reference_dct_1d(const double *in, double *out, int size) {
+  for (int k = 0; k < size; ++k) {
+    out[k] = 0;
+    for (int n = 0; n < size; ++n) {
+      out[k] += in[n] * dct_matrix(n, k, size);
+    }
+    if (k == 0) out[k] = out[k] * invSqrt2;
+  }
+}
+
+void reference_idct_1d(const double *in, double *out, int size) {
+  for (int k = 0; k < size; ++k) {
+    out[k] = 0;
+    for (int n = 0; n < size; ++n) {
+      if (n == 0)
+        out[k] += invSqrt2 * in[n] * dct_matrix(k, n, size);
+      else
+        out[k] += in[n] * dct_matrix(k, n, size);
+    }
+  }
+}
+
+// TODO(any): Copied from the old 'fadst4' (same as the new 'av1_fadst4_new'
+// function). Should be replaced by a proper reference function that takes
+// 'double' input & output.
+static void fadst4_new(const tran_low_t *input, tran_low_t *output) {
+  tran_high_t x0, x1, x2, x3;
+  tran_high_t s0, s1, s2, s3, s4, s5, s6, s7;
+
+  x0 = input[0];
+  x1 = input[1];
+  x2 = input[2];
+  x3 = input[3];
+
+  if (!(x0 | x1 | x2 | x3)) {
+    output[0] = output[1] = output[2] = output[3] = 0;
+    return;
+  }
+
+  s0 = sinpi_1_9 * x0;
+  s1 = sinpi_4_9 * x0;
+  s2 = sinpi_2_9 * x1;
+  s3 = sinpi_1_9 * x1;
+  s4 = sinpi_3_9 * x2;
+  s5 = sinpi_4_9 * x3;
+  s6 = sinpi_2_9 * x3;
+  s7 = x0 + x1 - x3;
+
+  x0 = s0 + s2 + s5;
+  x1 = sinpi_3_9 * s7;
+  x2 = s1 - s3 + s6;
+  x3 = s4;
+
+  s0 = x0 + x3;
+  s1 = x1;
+  s2 = x2 - x3;
+  s3 = x2 - x0 + x3;
+
+  // 1-D transform scaling factor is sqrt(2).
+  output[0] = (tran_low_t)fdct_round_shift(s0);
+  output[1] = (tran_low_t)fdct_round_shift(s1);
+  output[2] = (tran_low_t)fdct_round_shift(s2);
+  output[3] = (tran_low_t)fdct_round_shift(s3);
+}
+
+void reference_adst_1d(const double *in, double *out, int size) {
+  if (size == 4) {  // Special case.
+    tran_low_t int_input[4];
+    for (int i = 0; i < 4; ++i) {
+      int_input[i] = static_cast<tran_low_t>(round(in[i]));
+    }
+    tran_low_t int_output[4];
+    fadst4_new(int_input, int_output);
+    for (int i = 0; i < 4; ++i) {
+      out[i] = int_output[i];
+    }
+    return;
+  }
+
+  for (int k = 0; k < size; ++k) {
+    out[k] = 0;
+    for (int n = 0; n < size; ++n) {
+      out[k] += in[n] * sin(M_PI * (2 * n + 1) * (2 * k + 1) / (4 * size));
+    }
+  }
+}
+
+void reference_idtx_1d(const double *in, double *out, int size) {
+  double scale = 0;
+  if (size == 4)
+    scale = Sqrt2;
+  else if (size == 8)
+    scale = 2;
+  else if (size == 16)
+    scale = 2 * Sqrt2;
+  else if (size == 32)
+    scale = 4;
+  else if (size == 64)
+    scale = 4 * Sqrt2;
+  for (int k = 0; k < size; ++k) {
+    out[k] = in[k] * scale;
+  }
+}
+
+void reference_hybrid_1d(double *in, double *out, int size, int type) {
+  if (type == TYPE_DCT)
+    reference_dct_1d(in, out, size);
+  else if (type == TYPE_ADST)
+    reference_adst_1d(in, out, size);
+  else
+    reference_idtx_1d(in, out, size);
+}
+
+double get_amplification_factor(TX_TYPE tx_type, TX_SIZE tx_size) {
+  TXFM_2D_FLIP_CFG fwd_txfm_flip_cfg;
+  av1_get_fwd_txfm_cfg(tx_type, tx_size, &fwd_txfm_flip_cfg);
+  const int tx_width = tx_size_wide[fwd_txfm_flip_cfg.tx_size];
+  const int tx_height = tx_size_high[fwd_txfm_flip_cfg.tx_size];
+  const int8_t *shift = fwd_txfm_flip_cfg.shift;
+  const int amplify_bit = shift[0] + shift[1] + shift[2];
+  double amplify_factor =
+      amplify_bit >= 0 ? (1 << amplify_bit) : (1.0 / (1 << -amplify_bit));
+
+  // For rectangular transforms, we need to multiply by an extra factor.
+  const int rect_type = get_rect_tx_log_ratio(tx_width, tx_height);
+  if (abs(rect_type) == 1) {
+    amplify_factor *= pow(2, 0.5);
+  }
+  return amplify_factor;
+}
+
+void reference_hybrid_2d(double *in, double *out, TX_TYPE tx_type,
+                         TX_SIZE tx_size) {
+  // Get transform type and size of each dimension.
+  TYPE_TXFM type0;
+  TYPE_TXFM type1;
+  get_txfm1d_type(tx_type, &type0, &type1);
+  const int tx_width = tx_size_wide[tx_size];
+  const int tx_height = tx_size_high[tx_size];
+
+  double *const temp_in = new double[AOMMAX(tx_width, tx_height)];
+  double *const temp_out = new double[AOMMAX(tx_width, tx_height)];
+  double *const out_interm = new double[tx_width * tx_height];
+  const int stride = tx_width;
+
+  // Transform columns.
+  for (int c = 0; c < tx_width; ++c) {
+    for (int r = 0; r < tx_height; ++r) {
+      temp_in[r] = in[r * stride + c];
+    }
+    reference_hybrid_1d(temp_in, temp_out, tx_height, type0);
+    for (int r = 0; r < tx_height; ++r) {
+      out_interm[r * stride + c] = temp_out[r];
+    }
+  }
+
+  // Transform rows.
+  for (int r = 0; r < tx_height; ++r) {
+    reference_hybrid_1d(out_interm + r * stride, out + r * stride, tx_width,
+                        type1);
+  }
+
+  delete[] temp_in;
+  delete[] temp_out;
+  delete[] out_interm;
+
+  // These transforms use an approximate 2D DCT transform, by only keeping the
+  // top-left quarter of the coefficients, and repacking them in the first
+  // quarter indices.
+  // TODO(urvang): Refactor this code.
+  if (tx_width == 64 && tx_height == 64) {  // tx_size == TX_64X64
+    // Zero out top-right 32x32 area.
+    for (int row = 0; row < 32; ++row) {
+      memset(out + row * 64 + 32, 0, 32 * sizeof(*out));
+    }
+    // Zero out the bottom 64x32 area.
+    memset(out + 32 * 64, 0, 32 * 64 * sizeof(*out));
+    // Re-pack non-zero coeffs in the first 32x32 indices.
+    for (int row = 1; row < 32; ++row) {
+      memcpy(out + row * 32, out + row * 64, 32 * sizeof(*out));
+    }
+  } else if (tx_width == 32 && tx_height == 64) {  // tx_size == TX_32X64
+    // Zero out the bottom 32x32 area.
+    memset(out + 32 * 32, 0, 32 * 32 * sizeof(*out));
+    // Note: no repacking needed here.
+  } else if (tx_width == 64 && tx_height == 32) {  // tx_size == TX_64X32
+    // Zero out right 32x32 area.
+    for (int row = 0; row < 32; ++row) {
+      memset(out + row * 64 + 32, 0, 32 * sizeof(*out));
+    }
+    // Re-pack non-zero coeffs in the first 32x32 indices.
+    for (int row = 1; row < 32; ++row) {
+      memcpy(out + row * 32, out + row * 64, 32 * sizeof(*out));
+    }
+  } else if (tx_width == 16 && tx_height == 64) {  // tx_size == TX_16X64
+    // Zero out the bottom 16x32 area.
+    memset(out + 16 * 32, 0, 16 * 32 * sizeof(*out));
+    // Note: no repacking needed here.
+  } else if (tx_width == 64 && tx_height == 16) {  // tx_size == TX_64X16
+    // Zero out right 32x16 area.
+    for (int row = 0; row < 16; ++row) {
+      memset(out + row * 64 + 32, 0, 32 * sizeof(*out));
+    }
+    // Re-pack non-zero coeffs in the first 32x16 indices.
+    for (int row = 1; row < 16; ++row) {
+      memcpy(out + row * 32, out + row * 64, 32 * sizeof(*out));
+    }
+  }
+
+  // Apply appropriate scale.
+  const double amplify_factor = get_amplification_factor(tx_type, tx_size);
+  for (int c = 0; c < tx_width; ++c) {
+    for (int r = 0; r < tx_height; ++r) {
+      out[r * stride + c] *= amplify_factor;
+    }
+  }
+}
+
+template <typename Type>
+void fliplr(Type *dest, int width, int height, int stride) {
+  for (int r = 0; r < height; ++r) {
+    for (int c = 0; c < width / 2; ++c) {
+      const Type tmp = dest[r * stride + c];
+      dest[r * stride + c] = dest[r * stride + width - 1 - c];
+      dest[r * stride + width - 1 - c] = tmp;
+    }
+  }
+}
+
+template <typename Type>
+void flipud(Type *dest, int width, int height, int stride) {
+  for (int c = 0; c < width; ++c) {
+    for (int r = 0; r < height / 2; ++r) {
+      const Type tmp = dest[r * stride + c];
+      dest[r * stride + c] = dest[(height - 1 - r) * stride + c];
+      dest[(height - 1 - r) * stride + c] = tmp;
+    }
+  }
+}
+
+template <typename Type>
+void fliplrud(Type *dest, int width, int height, int stride) {
+  for (int r = 0; r < height / 2; ++r) {
+    for (int c = 0; c < width; ++c) {
+      const Type tmp = dest[r * stride + c];
+      dest[r * stride + c] = dest[(height - 1 - r) * stride + width - 1 - c];
+      dest[(height - 1 - r) * stride + width - 1 - c] = tmp;
+    }
+  }
+}
+
+template void fliplr<double>(double *dest, int width, int height, int stride);
+template void flipud<double>(double *dest, int width, int height, int stride);
+template void fliplrud<double>(double *dest, int width, int height, int stride);
+
+int bd_arr[BD_NUM] = { 8, 10, 12 };
+
+int8_t low_range_arr[BD_NUM] = { 18, 32, 32 };
+int8_t high_range_arr[BD_NUM] = { 32, 32, 32 };
+
+void txfm_stage_range_check(const int8_t *stage_range, int stage_num,
+                            int8_t cos_bit, int low_range, int high_range) {
+  for (int i = 0; i < stage_num; ++i) {
+    EXPECT_LE(stage_range[i], low_range);
+    ASSERT_LE(stage_range[i] + cos_bit, high_range) << "stage = " << i;
+  }
+  for (int i = 0; i < stage_num - 1; ++i) {
+    // make sure there is no overflow while doing half_btf()
+    ASSERT_LE(stage_range[i + 1] + cos_bit, high_range) << "stage = " << i;
+  }
+}
+}  // namespace libaom_test
diff --git a/third_party/aom/test/av1_txfm_test.h b/third_party/aom/test/av1_txfm_test.h
new file mode 100644
index 0000000000..a181647411
--- /dev/null
+++ b/third_party/aom/test/av1_txfm_test.h
@@ -0,0 +1,135 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_AV1_TXFM_TEST_H_
+#define AOM_TEST_AV1_TXFM_TEST_H_
+
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef _MSC_VER
+#define _USE_MATH_DEFINES
+#endif
+#include <math.h>
+
+#include "config/av1_rtcd.h"
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/acm_random.h"
+#include "av1/common/av1_txfm.h"
+#include "av1/common/blockd.h"
+#include "av1/common/enums.h"
+
+namespace libaom_test {
+typedef enum {
+  TYPE_DCT = 0,
+  TYPE_ADST,
+  TYPE_IDTX,
+  TYPE_IDCT,
+  TYPE_IADST,
+  TYPE_LAST
+} TYPE_TXFM;
+
+int get_txfm1d_size(TX_SIZE tx_size);
+
+void get_txfm1d_type(TX_TYPE txfm2d_type, TYPE_TXFM *type0, TYPE_TXFM *type1);
+
+void reference_dct_1d(const double *in, double *out, int size);
+void reference_idct_1d(const double *in, double *out, int size);
+
+void reference_adst_1d(const double *in, double *out, int size);
+
+void reference_hybrid_1d(double *in, double *out, int size, int type);
+
+double get_amplification_factor(TX_TYPE tx_type, TX_SIZE tx_size);
+
+void reference_hybrid_2d(double *in, double *out, TX_TYPE tx_type,
+                         TX_SIZE tx_size);
+template <typename Type1, typename Type2>
+static double compute_avg_abs_error(const Type1 *a, const Type2 *b,
+                                    const int size) {
+  double error = 0;
+  for (int i = 0; i < size; i++) {
+    error += fabs(static_cast<double>(a[i]) - static_cast<double>(b[i]));
+  }
+  error = error / size;
+  return error;
+}
+
+template <typename Type>
+void fliplr(Type *dest, int width, int height, int stride);
+
+template <typename Type>
+void flipud(Type *dest, int width, int height, int stride);
+
+template <typename Type>
+void fliplrud(Type *dest, int width, int height, int stride);
+
+typedef void (*TxfmFunc)(const int32_t *in, int32_t *out, const int8_t cos_bit,
+                         const int8_t *range_bit);
+
+typedef void (*InvTxfm2dFunc)(const int32_t *, uint16_t *, int, TX_TYPE, int);
+typedef void (*LbdInvTxfm2dFunc)(const int32_t *, uint8_t *, int, TX_TYPE,
+                                 TX_SIZE, int);
+
+static const int bd = 10;
+static const int input_base = (1 << bd);
+
+static INLINE bool IsTxSizeTypeValid(TX_SIZE tx_size, TX_TYPE tx_type) {
+  const TX_SIZE tx_size_sqr_up = txsize_sqr_up_map[tx_size];
+  TxSetType tx_set_type;
+  if (tx_size_sqr_up > TX_32X32) {
+    tx_set_type = EXT_TX_SET_DCTONLY;
+  } else if (tx_size_sqr_up == TX_32X32) {
+    tx_set_type = EXT_TX_SET_DCT_IDTX;
+  } else {
+    tx_set_type = EXT_TX_SET_ALL16;
+  }
+  return av1_ext_tx_used[tx_set_type][tx_type] != 0;
+}
+
+#if CONFIG_AV1_ENCODER
+
+static const FwdTxfm2dFunc fwd_txfm_func_ls[TX_SIZES_ALL] = {
+  av1_fwd_txfm2d_4x4_c,   av1_fwd_txfm2d_8x8_c,   av1_fwd_txfm2d_16x16_c,
+  av1_fwd_txfm2d_32x32_c, av1_fwd_txfm2d_64x64_c, av1_fwd_txfm2d_4x8_c,
+  av1_fwd_txfm2d_8x4_c,   av1_fwd_txfm2d_8x16_c,  av1_fwd_txfm2d_16x8_c,
+  av1_fwd_txfm2d_16x32_c, av1_fwd_txfm2d_32x16_c, av1_fwd_txfm2d_32x64_c,
+  av1_fwd_txfm2d_64x32_c, av1_fwd_txfm2d_4x16_c,  av1_fwd_txfm2d_16x4_c,
+  av1_fwd_txfm2d_8x32_c,  av1_fwd_txfm2d_32x8_c,  av1_fwd_txfm2d_16x64_c,
+  av1_fwd_txfm2d_64x16_c,
+};
+#endif
+
+static const InvTxfm2dFunc inv_txfm_func_ls[TX_SIZES_ALL] = {
+  av1_inv_txfm2d_add_4x4_c,   av1_inv_txfm2d_add_8x8_c,
+  av1_inv_txfm2d_add_16x16_c, av1_inv_txfm2d_add_32x32_c,
+  av1_inv_txfm2d_add_64x64_c, av1_inv_txfm2d_add_4x8_c,
+  av1_inv_txfm2d_add_8x4_c,   av1_inv_txfm2d_add_8x16_c,
+  av1_inv_txfm2d_add_16x8_c,  av1_inv_txfm2d_add_16x32_c,
+  av1_inv_txfm2d_add_32x16_c, av1_inv_txfm2d_add_32x64_c,
+  av1_inv_txfm2d_add_64x32_c, av1_inv_txfm2d_add_4x16_c,
+  av1_inv_txfm2d_add_16x4_c,  av1_inv_txfm2d_add_8x32_c,
+  av1_inv_txfm2d_add_32x8_c,  av1_inv_txfm2d_add_16x64_c,
+  av1_inv_txfm2d_add_64x16_c,
+};
+
+#define BD_NUM 3
+
+extern int bd_arr[];
+extern int8_t low_range_arr[];
+extern int8_t high_range_arr[];
+
+void txfm_stage_range_check(const int8_t *stage_range, int stage_num,
+                            const int8_t cos_bit, int low_range,
+                            int high_range);
+}  // namespace libaom_test
+#endif  // AOM_TEST_AV1_TXFM_TEST_H_
diff --git a/third_party/aom/test/av1_wedge_utils_test.cc b/third_party/aom/test/av1_wedge_utils_test.cc
new file mode 100644
index 0000000000..e8fbe69a4c
--- /dev/null
+++ b/third_party/aom/test/av1_wedge_utils_test.cc
@@ -0,0 +1,390 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+
+#include "aom_dsp/aom_dsp_common.h"
+
+#include "av1/common/enums.h"
+
+#include "test/acm_random.h"
+#include "test/function_equivalence_test.h"
+#include "test/register_state_check.h"
+
+#define WEDGE_WEIGHT_BITS 6
+#define MAX_MASK_VALUE (1 << (WEDGE_WEIGHT_BITS))
+
+using libaom_test::ACMRandom;
+using libaom_test::FunctionEquivalenceTest;
+
+namespace {
+
+static const int16_t kInt13Max = (1 << 12) - 1;
+
+//////////////////////////////////////////////////////////////////////////////
+// av1_wedge_sse_from_residuals - functionality
+//////////////////////////////////////////////////////////////////////////////
+
+class WedgeUtilsSSEFuncTest : public testing::Test {
+ protected:
+  WedgeUtilsSSEFuncTest() : rng_(ACMRandom::DeterministicSeed()) {}
+
+  static const int kIterations = 1000;
+
+  ACMRandom rng_;
+};
+
+static void equiv_blend_residuals(int16_t *r, const int16_t *r0,
+                                  const int16_t *r1, const uint8_t *m, int N) {
+  for (int i = 0; i < N; i++) {
+    const int32_t m0 = m[i];
+    const int32_t m1 = MAX_MASK_VALUE - m0;
+    const int16_t R = m0 * r0[i] + m1 * r1[i];
+    // Note that this rounding is designed to match the result
+    // you would get when actually blending the 2 predictors and computing
+    // the residuals.
+    r[i] = ROUND_POWER_OF_TWO(R - 1, WEDGE_WEIGHT_BITS);
+  }
+}
+
+static uint64_t equiv_sse_from_residuals(const int16_t *r0, const int16_t *r1,
+                                         const uint8_t *m, int N) {
+  uint64_t acc = 0;
+  for (int i = 0; i < N; i++) {
+    const int32_t m0 = m[i];
+    const int32_t m1 = MAX_MASK_VALUE - m0;
+    const int16_t R = m0 * r0[i] + m1 * r1[i];
+    const int32_t r = ROUND_POWER_OF_TWO(R - 1, WEDGE_WEIGHT_BITS);
+    acc += r * r;
+  }
+  return acc;
+}
+
+TEST_F(WedgeUtilsSSEFuncTest, ResidualBlendingEquiv) {
+  DECLARE_ALIGNED(32, uint8_t, s[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, p0[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, p1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, p[MAX_SB_SQUARE]);
+
+  DECLARE_ALIGNED(32, int16_t, r0[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, r_ref[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, r_tst[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      s[i] = rng_.Rand8();
+      m[i] = rng_(MAX_MASK_VALUE + 1);
+    }
+
+    const int w = 1 << (rng_(MAX_SB_SIZE_LOG2 + 1 - 3) + 3);
+    const int h = 1 << (rng_(MAX_SB_SIZE_LOG2 + 1 - 3) + 3);
+    const int N = w * h;
+
+    for (int j = 0; j < N; j++) {
+      p0[j] = clamp(s[j] + rng_(33) - 16, 0, UINT8_MAX);
+      p1[j] = clamp(s[j] + rng_(33) - 16, 0, UINT8_MAX);
+    }
+
+    aom_blend_a64_mask(p, w, p0, w, p1, w, m, w, w, h, 0, 0);
+
+    aom_subtract_block(h, w, r0, w, s, w, p0, w);
+    aom_subtract_block(h, w, r1, w, s, w, p1, w);
+
+    aom_subtract_block(h, w, r_ref, w, s, w, p, w);
+    equiv_blend_residuals(r_tst, r0, r1, m, N);
+
+    for (int i = 0; i < N; ++i) ASSERT_EQ(r_ref[i], r_tst[i]);
+
+    uint64_t ref_sse = aom_sum_squares_i16(r_ref, N);
+    uint64_t tst_sse = equiv_sse_from_residuals(r0, r1, m, N);
+
+    ASSERT_EQ(ref_sse, tst_sse);
+  }
+}
+
+static uint64_t sse_from_residuals(const int16_t *r0, const int16_t *r1,
+                                   const uint8_t *m, int N) {
+  uint64_t acc = 0;
+  for (int i = 0; i < N; i++) {
+    const int32_t m0 = m[i];
+    const int32_t m1 = MAX_MASK_VALUE - m0;
+    const int32_t r = m0 * r0[i] + m1 * r1[i];
+    acc += r * r;
+  }
+  return ROUND_POWER_OF_TWO(acc, 2 * WEDGE_WEIGHT_BITS);
+}
+
+TEST_F(WedgeUtilsSSEFuncTest, ResidualBlendingMethod) {
+  DECLARE_ALIGNED(32, int16_t, r0[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, d[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      r1[i] = rng_(2 * INT8_MAX - 2 * INT8_MIN + 1) + 2 * INT8_MIN;
+      d[i] = rng_(2 * INT8_MAX - 2 * INT8_MIN + 1) + 2 * INT8_MIN;
+      m[i] = rng_(MAX_MASK_VALUE + 1);
+    }
+
+    const int N = 64 * (rng_(MAX_SB_SQUARE / 64) + 1);
+
+    for (int i = 0; i < N; i++) r0[i] = r1[i] + d[i];
+
+    const uint64_t ref_res = sse_from_residuals(r0, r1, m, N);
+    const uint64_t tst_res = av1_wedge_sse_from_residuals(r1, d, m, N);
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// av1_wedge_sse_from_residuals - optimizations
+//////////////////////////////////////////////////////////////////////////////
+
+typedef uint64_t (*FSSE)(const int16_t *r1, const int16_t *d, const uint8_t *m,
+                         int N);
+typedef libaom_test::FuncParam<FSSE> TestFuncsFSSE;
+
+class WedgeUtilsSSEOptTest : public FunctionEquivalenceTest<FSSE> {
+ protected:
+  static const int kIterations = 10000;
+};
+
+TEST_P(WedgeUtilsSSEOptTest, RandomValues) {
+  DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, d[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      r1[i] = rng_(2 * kInt13Max + 1) - kInt13Max;
+      d[i] = rng_(2 * kInt13Max + 1) - kInt13Max;
+      m[i] = rng_(MAX_MASK_VALUE + 1);
+    }
+
+    const int N = 64 * (rng_(MAX_SB_SQUARE / 64) + 1);
+
+    const uint64_t ref_res = params_.ref_func(r1, d, m, N);
+    uint64_t tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(r1, d, m, N));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+TEST_P(WedgeUtilsSSEOptTest, ExtremeValues) {
+  DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, d[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    if (rng_(2)) {
+      for (int i = 0; i < MAX_SB_SQUARE; ++i) r1[i] = kInt13Max;
+    } else {
+      for (int i = 0; i < MAX_SB_SQUARE; ++i) r1[i] = -kInt13Max;
+    }
+
+    if (rng_(2)) {
+      for (int i = 0; i < MAX_SB_SQUARE; ++i) d[i] = kInt13Max;
+    } else {
+      for (int i = 0; i < MAX_SB_SQUARE; ++i) d[i] = -kInt13Max;
+    }
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) m[i] = MAX_MASK_VALUE;
+
+    const int N = 64 * (rng_(MAX_SB_SQUARE / 64) + 1);
+
+    const uint64_t ref_res = params_.ref_func(r1, d, m, N);
+    uint64_t tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(r1, d, m, N));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// av1_wedge_sign_from_residuals
+//////////////////////////////////////////////////////////////////////////////
+
+typedef int (*FSign)(const int16_t *ds, const uint8_t *m, int N, int64_t limit);
+typedef libaom_test::FuncParam<FSign> TestFuncsFSign;
+
+class WedgeUtilsSignOptTest : public FunctionEquivalenceTest<FSign> {
+ protected:
+  static const int kIterations = 10000;
+  static const int kMaxSize = 8196;  // Size limited by SIMD implementation.
+};
+
+TEST_P(WedgeUtilsSignOptTest, RandomValues) {
+  DECLARE_ALIGNED(32, int16_t, r0[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, ds[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      r0[i] = rng_(2 * kInt13Max + 1) - kInt13Max;
+      r1[i] = rng_(2 * kInt13Max + 1) - kInt13Max;
+      m[i] = rng_(MAX_MASK_VALUE + 1);
+    }
+
+    const int maxN = AOMMIN(kMaxSize, MAX_SB_SQUARE);
+    const int N = 64 * (rng_(maxN / 64 - 1) + 1);
+
+    int64_t limit;
+    limit = (int64_t)aom_sum_squares_i16(r0, N);
+    limit -= (int64_t)aom_sum_squares_i16(r1, N);
+    limit *= (1 << WEDGE_WEIGHT_BITS) / 2;
+
+    for (int i = 0; i < N; i++)
+      ds[i] = clamp(r0[i] * r0[i] - r1[i] * r1[i], INT16_MIN, INT16_MAX);
+
+    const int ref_res = params_.ref_func(ds, m, N, limit);
+    int tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(ds, m, N, limit));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+TEST_P(WedgeUtilsSignOptTest, ExtremeValues) {
+  DECLARE_ALIGNED(32, int16_t, r0[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, r1[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, ds[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint8_t, m[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    switch (rng_(4)) {
+      case 0:
+        for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+          r0[i] = 0;
+          r1[i] = kInt13Max;
+        }
+        break;
+      case 1:
+        for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+          r0[i] = kInt13Max;
+          r1[i] = 0;
+        }
+        break;
+      case 2:
+        for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+          r0[i] = 0;
+          r1[i] = -kInt13Max;
+        }
+        break;
+      default:
+        for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+          r0[i] = -kInt13Max;
+          r1[i] = 0;
+        }
+        break;
+    }
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) m[i] = MAX_MASK_VALUE;
+
+    const int maxN = AOMMIN(kMaxSize, MAX_SB_SQUARE);
+    const int N = 64 * (rng_(maxN / 64 - 1) + 1);
+
+    int64_t limit;
+    limit = (int64_t)aom_sum_squares_i16(r0, N);
+    limit -= (int64_t)aom_sum_squares_i16(r1, N);
+    limit *= (1 << WEDGE_WEIGHT_BITS) / 2;
+
+    for (int i = 0; i < N; i++)
+      ds[i] = clamp(r0[i] * r0[i] - r1[i] * r1[i], INT16_MIN, INT16_MAX);
+
+    const int ref_res = params_.ref_func(ds, m, N, limit);
+    int tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(ds, m, N, limit));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+//////////////////////////////////////////////////////////////////////////////
+// av1_wedge_compute_delta_squares
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*FDS)(int16_t *d, const int16_t *a, const int16_t *b, int N);
+typedef libaom_test::FuncParam<FDS> TestFuncsFDS;
+
+class WedgeUtilsDeltaSquaresOptTest : public FunctionEquivalenceTest<FDS> {
+ protected:
+  static const int kIterations = 10000;
+};
+
+TEST_P(WedgeUtilsDeltaSquaresOptTest, RandomValues) {
+  DECLARE_ALIGNED(32, int16_t, a[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, b[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, d_ref[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int16_t, d_tst[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      a[i] = rng_.Rand16();
+      b[i] = rng_(2 * INT16_MAX + 1) - INT16_MAX;
+    }
+
+    const int N = 64 * (rng_(MAX_SB_SQUARE / 64) + 1);
+
+    memset(&d_ref, INT16_MAX, sizeof(d_ref));
+    memset(&d_tst, INT16_MAX, sizeof(d_tst));
+
+    params_.ref_func(d_ref, a, b, N);
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(d_tst, a, b, N));
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) ASSERT_EQ(d_ref[i], d_tst[i]);
+  }
+}
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, WedgeUtilsSSEOptTest,
+    ::testing::Values(TestFuncsFSSE(av1_wedge_sse_from_residuals_c,
+                                    av1_wedge_sse_from_residuals_sse2)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2, WedgeUtilsSignOptTest,
+    ::testing::Values(TestFuncsFSign(av1_wedge_sign_from_residuals_c,
+                                     av1_wedge_sign_from_residuals_sse2)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2, WedgeUtilsDeltaSquaresOptTest,
+    ::testing::Values(TestFuncsFDS(av1_wedge_compute_delta_squares_c,
+                                   av1_wedge_compute_delta_squares_sse2)));
+#endif  // HAVE_SSE2
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, WedgeUtilsSSEOptTest,
+    ::testing::Values(TestFuncsFSSE(av1_wedge_sse_from_residuals_sse2,
+                                    av1_wedge_sse_from_residuals_avx2)));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, WedgeUtilsSignOptTest,
+    ::testing::Values(TestFuncsFSign(av1_wedge_sign_from_residuals_sse2,
+                                     av1_wedge_sign_from_residuals_avx2)));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, WedgeUtilsDeltaSquaresOptTest,
+    ::testing::Values(TestFuncsFDS(av1_wedge_compute_delta_squares_sse2,
+                                   av1_wedge_compute_delta_squares_avx2)));
+#endif  // HAVE_AVX2
+
+}  // namespace
diff --git a/third_party/aom/test/best_encode.sh b/third_party/aom/test/best_encode.sh
new file mode 100755
index 0000000000..fe31a01cb9
--- /dev/null
+++ b/third_party/aom/test/best_encode.sh
@@ -0,0 +1,103 @@
+#!/bin/bash
+#
+# Copyright (c) 2016, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and
+# the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+# was not distributed with this source code in the LICENSE file, you can
+# obtain it at www.aomedia.org/license/software. If the Alliance for Open
+# Media Patent License 1.0 was not distributed with this source code in the
+# PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+#
+# Author: jimbankoski@google.com (Jim Bankoski)
+
+if [[ $# -ne 2 ]]; then
+  echo "Encodes a file using best known settings (slow!)"
+  echo "  Usage:    be [FILE] [BITRATE]"
+  echo "  Example:  be akiyo_cif.y4m 200"
+  exit
+fi
+
+f=$1  # file is first parameter
+b=$2  # bitrate is second parameter
+
+if [[ -e $f.fpf ]]; then
+  # First-pass file found, do second pass only
+  aomenc \
+    $f \
+    -o $f-$b.av1.webm \
+    -p 2 \
+    --pass=2 \
+    --fpf=$f.fpf \
+    --best \
+    --cpu-used=0 \
+    --target-bitrate=$b \
+    --auto-alt-ref=1 \
+    -v \
+    --minsection-pct=0 \
+    --maxsection-pct=800 \
+    --lag-in-frames=25 \
+    --kf-min-dist=0 \
+    --kf-max-dist=99999 \
+    --static-thresh=0 \
+    --min-q=0 \
+    --max-q=63 \
+    --drop-frame=0 \
+    --bias-pct=50 \
+    --minsection-pct=0 \
+    --maxsection-pct=800 \
+    --psnr \
+    --arnr-maxframes=7 \
+    --arnr-strength=3 \
+    --arnr-type=3
+else
+  # No first-pass file found, do 2-pass encode
+  aomenc \
+    $f \
+    -o $f-$b.av1.webm \
+    -p 2 \
+    --pass=1 \
+    --fpf=$f.fpf \
+    --best \
+    --cpu-used=0 \
+    --target-bitrate=$b \
+    --auto-alt-ref=1 \
+    -v \
+    --minsection-pct=0 \
+    --maxsection-pct=800 \
+    --lag-in-frames=25 \
+    --kf-min-dist=0 \
+    --kf-max-dist=99999 \
+    --static-thresh=0 \
+    --min-q=0 \
+    --max-q=63 \
+    --drop-frame=0
+
+  aomenc \
+    $f \
+    -o $f-$b.av1.webm \
+    -p 2 \
+    --pass=2 \
+    --fpf=$f.fpf \
+    --best \
+    --cpu-used=0 \
+    --target-bitrate=$b \
+    --auto-alt-ref=1 \
+    -v \
+    --minsection-pct=0 \
+    --maxsection-pct=800 \
+    --lag-in-frames=25 \
+    --kf-min-dist=0 \
+    --kf-max-dist=99999 \
+    --static-thresh=0 \
+    --min-q=0 \
+    --max-q=63 \
+    --drop-frame=0 \
+    --bias-pct=50 \
+    --minsection-pct=0 \
+    --maxsection-pct=800 \
+    --psnr \
+    --arnr-maxframes=7 \
+    --arnr-strength=3 \
+    --arnr-type=3
+fi
diff --git a/third_party/aom/test/binary_codes_test.cc b/third_party/aom/test/binary_codes_test.cc
new file mode 100644
index 0000000000..45660cf853
--- /dev/null
+++ b/third_party/aom/test/binary_codes_test.cc
@@ -0,0 +1,83 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#include "test/acm_random.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/bitreader.h"
+#include "aom_dsp/bitwriter.h"
+#include "aom_dsp/binary_codes_reader.h"
+#include "aom_dsp/binary_codes_writer.h"
+
+#define ACCT_STR __func__
+
+using libaom_test::ACMRandom;
+
+namespace {
+
+// Test for Finite subexponential code with reference
+TEST(AV1, TestPrimitiveRefsubexpfin) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  const int kBufferSize = 65536;
+  aom_writer bw;
+  uint8_t bw_buffer[kBufferSize];
+  const uint16_t kRanges = 8;
+  const uint16_t kSubexpParams = 6;
+  const uint16_t kReferences = 8;
+  const uint16_t kValues = 16;
+  uint16_t enc_values[kRanges][kSubexpParams][kReferences][kValues][4];
+  const uint16_t range_vals[kRanges] = { 1, 13, 64, 120, 230, 420, 1100, 8000 };
+  aom_start_encode(&bw, bw_buffer);
+  for (int n = 0; n < kRanges; ++n) {
+    const uint16_t range = range_vals[n];
+    for (int k = 0; k < kSubexpParams; ++k) {
+      for (int r = 0; r < kReferences; ++r) {
+        const uint16_t ref = rnd(range);
+        for (int v = 0; v < kValues; ++v) {
+          const uint16_t value = rnd(range);
+          enc_values[n][k][r][v][0] = range;
+          enc_values[n][k][r][v][1] = k;
+          enc_values[n][k][r][v][2] = ref;
+          enc_values[n][k][r][v][3] = value;
+          aom_write_primitive_refsubexpfin(&bw, range, k, ref, value);
+        }
+      }
+    }
+  }
+  aom_stop_encode(&bw);
+  aom_reader br;
+  aom_reader_init(&br, bw_buffer, bw.pos);
+  GTEST_ASSERT_GE(aom_reader_tell(&br), 0u);
+  GTEST_ASSERT_LE(aom_reader_tell(&br), 1u);
+  for (int n = 0; n < kRanges; ++n) {
+    for (int k = 0; k < kSubexpParams; ++k) {
+      for (int r = 0; r < kReferences; ++r) {
+        for (int v = 0; v < kValues; ++v) {
+          const uint16_t range = enc_values[n][k][r][v][0];
+          assert(k == enc_values[n][k][r][v][1]);
+          const uint16_t ref = enc_values[n][k][r][v][2];
+          const uint16_t value =
+              aom_read_primitive_refsubexpfin(&br, range, k, ref, ACCT_STR);
+          GTEST_ASSERT_EQ(value, enc_values[n][k][r][v][3]);
+        }
+      }
+    }
+  }
+}
+// TODO(debargha): Adds tests for other primitives
+}  // namespace
diff --git a/third_party/aom/test/blend_a64_mask_1d_test.cc b/third_party/aom/test/blend_a64_mask_1d_test.cc
new file mode 100644
index 0000000000..f8844eef8b
--- /dev/null
+++ b/third_party/aom/test/blend_a64_mask_1d_test.cc
@@ -0,0 +1,339 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/register_state_check.h"
+#include "test/function_equivalence_test.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+
+#include "aom/aom_integer.h"
+
+#include "av1/common/enums.h"
+
+#include "aom_dsp/blend.h"
+
+using libaom_test::FunctionEquivalenceTest;
+
+namespace {
+
+template <typename F, typename T>
+class BlendA64Mask1DTest : public FunctionEquivalenceTest<F> {
+ public:
+  static const int kIterations = 10000;
+  static const int kMaxWidth = MAX_SB_SIZE * 5;  // * 5 to cover longer strides
+  static const int kMaxHeight = MAX_SB_SIZE;
+  static const int kBufSize = kMaxWidth * kMaxHeight;
+  static const int kMaxMaskWidth = 2 * MAX_SB_SIZE;
+  static const int kMaxMaskSize = kMaxMaskWidth;
+
+  virtual ~BlendA64Mask1DTest() {}
+
+  virtual void Execute(const T *p_src0, const T *p_src1) = 0;
+
+  void Common() {
+    w_ = 2 << this->rng_(MAX_SB_SIZE_LOG2);
+    h_ = 2 << this->rng_(MAX_SB_SIZE_LOG2);
+
+    dst_offset_ = this->rng_(33);
+    dst_stride_ = this->rng_(kMaxWidth + 1 - w_) + w_;
+
+    src0_offset_ = this->rng_(33);
+    src0_stride_ = this->rng_(kMaxWidth + 1 - w_) + w_;
+
+    src1_offset_ = this->rng_(33);
+    src1_stride_ = this->rng_(kMaxWidth + 1 - w_) + w_;
+
+    T *p_src0;
+    T *p_src1;
+
+    switch (this->rng_(3)) {
+      case 0:  // Separate sources
+        p_src0 = src0_;
+        p_src1 = src1_;
+        break;
+      case 1:  // src0 == dst
+        p_src0 = dst_tst_;
+        src0_stride_ = dst_stride_;
+        src0_offset_ = dst_offset_;
+        p_src1 = src1_;
+        break;
+      case 2:  // src1 == dst
+        p_src0 = src0_;
+        p_src1 = dst_tst_;
+        src1_stride_ = dst_stride_;
+        src1_offset_ = dst_offset_;
+        break;
+      default: FAIL();
+    }
+
+    Execute(p_src0, p_src1);
+
+    for (int r = 0; r < h_; ++r) {
+      for (int c = 0; c < w_; ++c) {
+        ASSERT_EQ(dst_ref_[dst_offset_ + r * dst_stride_ + c],
+                  dst_tst_[dst_offset_ + r * dst_stride_ + c]);
+      }
+    }
+  }
+
+  T dst_ref_[kBufSize];
+  T dst_tst_[kBufSize];
+  uint32_t dst_stride_;
+  uint32_t dst_offset_;
+
+  T src0_[kBufSize];
+  uint32_t src0_stride_;
+  uint32_t src0_offset_;
+
+  T src1_[kBufSize];
+  uint32_t src1_stride_;
+  uint32_t src1_offset_;
+
+  uint8_t mask_[kMaxMaskSize];
+
+  int w_;
+  int h_;
+};
+
+//////////////////////////////////////////////////////////////////////////////
+// 8 bit version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*F8B)(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
+                    uint32_t src0_stride, const uint8_t *src1,
+                    uint32_t src1_stride, const uint8_t *mask, int w, int h);
+typedef libaom_test::FuncParam<F8B> TestFuncs;
+
+class BlendA64Mask1DTest8B : public BlendA64Mask1DTest<F8B, uint8_t> {
+ protected:
+  void Execute(const uint8_t *p_src0, const uint8_t *p_src1) {
+    params_.ref_func(dst_ref_ + dst_offset_, dst_stride_, p_src0 + src0_offset_,
+                     src0_stride_, p_src1 + src1_offset_, src1_stride_, mask_,
+                     w_, h_);
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(
+        dst_tst_ + dst_offset_, dst_stride_, p_src0 + src0_offset_,
+        src0_stride_, p_src1 + src1_offset_, src1_stride_, mask_, w_, h_));
+  }
+};
+
+TEST_P(BlendA64Mask1DTest8B, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_.Rand8();
+      dst_tst_[i] = rng_.Rand8();
+
+      src0_[i] = rng_.Rand8();
+      src1_[i] = rng_.Rand8();
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(AOM_BLEND_A64_MAX_ALPHA + 1);
+
+    Common();
+  }
+}
+
+TEST_P(BlendA64Mask1DTest8B, ExtremeValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_(2) + 254;
+      dst_tst_[i] = rng_(2) + 254;
+      src0_[i] = rng_(2) + 254;
+      src1_[i] = rng_(2) + 254;
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(2) + AOM_BLEND_A64_MAX_ALPHA - 1;
+
+    Common();
+  }
+}
+
+static void blend_a64_hmask_ref(uint8_t *dst, uint32_t dst_stride,
+                                const uint8_t *src0, uint32_t src0_stride,
+                                const uint8_t *src1, uint32_t src1_stride,
+                                const uint8_t *mask, int w, int h) {
+  uint8_t mask2d[BlendA64Mask1DTest8B::kMaxMaskSize]
+                [BlendA64Mask1DTest8B::kMaxMaskSize];
+
+  for (int row = 0; row < h; ++row)
+    for (int col = 0; col < w; ++col) mask2d[row][col] = mask[col];
+
+  aom_blend_a64_mask_c(dst, dst_stride, src0, src0_stride, src1, src1_stride,
+                       &mask2d[0][0], BlendA64Mask1DTest8B::kMaxMaskSize, w, h,
+                       0, 0);
+}
+
+static void blend_a64_vmask_ref(uint8_t *dst, uint32_t dst_stride,
+                                const uint8_t *src0, uint32_t src0_stride,
+                                const uint8_t *src1, uint32_t src1_stride,
+                                const uint8_t *mask, int w, int h) {
+  uint8_t mask2d[BlendA64Mask1DTest8B::kMaxMaskSize]
+                [BlendA64Mask1DTest8B::kMaxMaskSize];
+
+  for (int row = 0; row < h; ++row)
+    for (int col = 0; col < w; ++col) mask2d[row][col] = mask[row];
+
+  aom_blend_a64_mask_c(dst, dst_stride, src0, src0_stride, src1, src1_stride,
+                       &mask2d[0][0], BlendA64Mask1DTest8B::kMaxMaskSize, w, h,
+                       0, 0);
+}
+
+INSTANTIATE_TEST_CASE_P(
+    C, BlendA64Mask1DTest8B,
+    ::testing::Values(TestFuncs(blend_a64_hmask_ref, aom_blend_a64_hmask_c),
+                      TestFuncs(blend_a64_vmask_ref, aom_blend_a64_vmask_c)));
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, BlendA64Mask1DTest8B,
+    ::testing::Values(
+        TestFuncs(blend_a64_hmask_ref, aom_blend_a64_hmask_sse4_1),
+        TestFuncs(blend_a64_vmask_ref, aom_blend_a64_vmask_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, BlendA64Mask1DTest8B,
+                        ::testing::Values(TestFuncs(blend_a64_hmask_ref,
+                                                    aom_blend_a64_hmask_neon),
+                                          TestFuncs(blend_a64_vmask_ref,
+                                                    aom_blend_a64_vmask_neon)));
+#endif  // HAVE_NEON
+
+//////////////////////////////////////////////////////////////////////////////
+// High bit-depth version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*FHBD)(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
+                     uint32_t src0_stride, const uint8_t *src1,
+                     uint32_t src1_stride, const uint8_t *mask, int w, int h,
+                     int bd);
+typedef libaom_test::FuncParam<FHBD> TestFuncsHBD;
+
+class BlendA64Mask1DTestHBD : public BlendA64Mask1DTest<FHBD, uint16_t> {
+ protected:
+  void Execute(const uint16_t *p_src0, const uint16_t *p_src1) {
+    params_.ref_func(CONVERT_TO_BYTEPTR(dst_ref_ + dst_offset_), dst_stride_,
+                     CONVERT_TO_BYTEPTR(p_src0 + src0_offset_), src0_stride_,
+                     CONVERT_TO_BYTEPTR(p_src1 + src1_offset_), src1_stride_,
+                     mask_, w_, h_, bit_depth_);
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(
+        CONVERT_TO_BYTEPTR(dst_tst_ + dst_offset_), dst_stride_,
+        CONVERT_TO_BYTEPTR(p_src0 + src0_offset_), src0_stride_,
+        CONVERT_TO_BYTEPTR(p_src1 + src1_offset_), src1_stride_, mask_, w_, h_,
+        bit_depth_));
+  }
+
+  int bit_depth_;
+};
+
+TEST_P(BlendA64Mask1DTestHBD, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    switch (rng_(3)) {
+      case 0: bit_depth_ = 8; break;
+      case 1: bit_depth_ = 10; break;
+      default: bit_depth_ = 12; break;
+    }
+
+    const int hi = 1 << bit_depth_;
+
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_(hi);
+      dst_tst_[i] = rng_(hi);
+      src0_[i] = rng_(hi);
+      src1_[i] = rng_(hi);
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(AOM_BLEND_A64_MAX_ALPHA + 1);
+
+    Common();
+  }
+}
+
+TEST_P(BlendA64Mask1DTestHBD, ExtremeValues) {
+  for (int iter = 0; iter < 1000 && !HasFatalFailure(); ++iter) {
+    switch (rng_(3)) {
+      case 0: bit_depth_ = 8; break;
+      case 1: bit_depth_ = 10; break;
+      default: bit_depth_ = 12; break;
+    }
+
+    const int hi = 1 << bit_depth_;
+    const int lo = hi - 2;
+
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_(hi - lo) + lo;
+      dst_tst_[i] = rng_(hi - lo) + lo;
+      src0_[i] = rng_(hi - lo) + lo;
+      src1_[i] = rng_(hi - lo) + lo;
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(2) + AOM_BLEND_A64_MAX_ALPHA - 1;
+
+    Common();
+  }
+}
+
+static void highbd_blend_a64_hmask_ref(
+    uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
+    uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
+    const uint8_t *mask, int w, int h, int bd) {
+  uint8_t mask2d[BlendA64Mask1DTestHBD::kMaxMaskSize]
+                [BlendA64Mask1DTestHBD::kMaxMaskSize];
+
+  for (int row = 0; row < h; ++row)
+    for (int col = 0; col < w; ++col) mask2d[row][col] = mask[col];
+
+  aom_highbd_blend_a64_mask_c(
+      dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask2d[0][0],
+      BlendA64Mask1DTestHBD::kMaxMaskSize, w, h, 0, 0, bd);
+}
+
+static void highbd_blend_a64_vmask_ref(
+    uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
+    uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
+    const uint8_t *mask, int w, int h, int bd) {
+  uint8_t mask2d[BlendA64Mask1DTestHBD::kMaxMaskSize]
+                [BlendA64Mask1DTestHBD::kMaxMaskSize];
+
+  for (int row = 0; row < h; ++row)
+    for (int col = 0; col < w; ++col) mask2d[row][col] = mask[row];
+
+  aom_highbd_blend_a64_mask_c(
+      dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask2d[0][0],
+      BlendA64Mask1DTestHBD::kMaxMaskSize, w, h, 0, 0, bd);
+}
+
+INSTANTIATE_TEST_CASE_P(
+    C, BlendA64Mask1DTestHBD,
+    ::testing::Values(TestFuncsHBD(highbd_blend_a64_hmask_ref,
+                                   aom_highbd_blend_a64_hmask_c),
+                      TestFuncsHBD(highbd_blend_a64_vmask_ref,
+                                   aom_highbd_blend_a64_vmask_c)));
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, BlendA64Mask1DTestHBD,
+    ::testing::Values(TestFuncsHBD(highbd_blend_a64_hmask_ref,
+                                   aom_highbd_blend_a64_hmask_sse4_1),
+                      TestFuncsHBD(highbd_blend_a64_vmask_ref,
+                                   aom_highbd_blend_a64_vmask_sse4_1)));
+#endif  // HAVE_SSE4_1
+}  // namespace
diff --git a/third_party/aom/test/blend_a64_mask_test.cc b/third_party/aom/test/blend_a64_mask_test.cc
new file mode 100644
index 0000000000..66ca6fc5f3
--- /dev/null
+++ b/third_party/aom/test/blend_a64_mask_test.cc
@@ -0,0 +1,583 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/register_state_check.h"
+#include "test/function_equivalence_test.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+
+#include "aom/aom_integer.h"
+
+#include "av1/common/enums.h"
+
+#include "aom_dsp/blend.h"
+
+using libaom_test::FunctionEquivalenceTest;
+
+namespace {
+
+template <typename BlendA64Func, typename SrcPixel, typename DstPixel>
+class BlendA64MaskTest : public FunctionEquivalenceTest<BlendA64Func> {
+ protected:
+  static const int kIterations = 10000;
+  static const int kMaxWidth = MAX_SB_SIZE * 5;  // * 5 to cover longer strides
+  static const int kMaxHeight = MAX_SB_SIZE;
+  static const int kBufSize = kMaxWidth * kMaxHeight;
+  static const int kMaxMaskWidth = 2 * MAX_SB_SIZE;
+  static const int kMaxMaskSize = kMaxMaskWidth * kMaxMaskWidth;
+
+  virtual ~BlendA64MaskTest() {}
+
+  virtual void Execute(const SrcPixel *p_src0, const SrcPixel *p_src1,
+                       int run_times) = 0;
+
+  template <typename Pixel>
+  void GetSources(Pixel **src0, Pixel **src1, Pixel * /*dst*/, int run_times) {
+    if (run_times > 1) {
+      *src0 = src0_;
+      *src1 = src1_;
+      return;
+    }
+    switch (this->rng_(3)) {
+      case 0:  // Separate sources
+        *src0 = src0_;
+        *src1 = src1_;
+        break;
+      case 1:  // src0 == dst
+        *src0 = dst_tst_;
+        src0_stride_ = dst_stride_;
+        src0_offset_ = dst_offset_;
+        *src1 = src1_;
+        break;
+      case 2:  // src1 == dst
+        *src0 = src0_;
+        *src1 = dst_tst_;
+        src1_stride_ = dst_stride_;
+        src1_offset_ = dst_offset_;
+        break;
+      default: FAIL();
+    }
+  }
+
+  void GetSources(uint16_t **src0, uint16_t **src1, uint8_t * /*dst*/,
+                  int /*run_times*/) {
+    *src0 = src0_;
+    *src1 = src1_;
+  }
+
+  uint8_t Rand1() { return this->rng_.Rand8() & 1; }
+
+  void RunOneTest(int block_size, int subx, int suby, int run_times) {
+    w_ = block_size_wide[block_size];
+    h_ = block_size_high[block_size];
+    run_times = run_times > 1 ? run_times / w_ : 1;
+    subx_ = subx;
+    suby_ = suby;
+
+    dst_offset_ = this->rng_(33);
+    dst_stride_ = this->rng_(kMaxWidth + 1 - w_) + w_;
+
+    src0_offset_ = this->rng_(33);
+    src0_stride_ = this->rng_(kMaxWidth + 1 - w_) + w_;
+
+    src1_offset_ = this->rng_(33);
+    src1_stride_ = this->rng_(kMaxWidth + 1 - w_) + w_;
+
+    mask_stride_ =
+        this->rng_(kMaxWidth + 1 - w_ * (subx_ ? 2 : 1)) + w_ * (subx_ ? 2 : 1);
+
+    SrcPixel *p_src0;
+    SrcPixel *p_src1;
+
+    p_src0 = src0_;
+    p_src1 = src1_;
+
+    GetSources(&p_src0, &p_src1, &dst_ref_[0], run_times);
+
+    Execute(p_src0, p_src1, run_times);
+
+    for (int r = 0; r < h_; ++r) {
+      for (int c = 0; c < w_; ++c) {
+        ASSERT_EQ(dst_ref_[dst_offset_ + r * dst_stride_ + c],
+                  dst_tst_[dst_offset_ + r * dst_stride_ + c])
+            << w_ << "x" << h_ << " subx " << subx_ << " suby " << suby_
+            << " r: " << r << " c: " << c;
+      }
+    }
+  }
+
+  void RunTest(int block_size, int run_times) {
+    subx_ = Rand1();
+    suby_ = Rand1();
+    RunOneTest(block_size, subx_, suby_, run_times);
+  }
+
+  DstPixel dst_ref_[kBufSize];
+  DstPixel dst_tst_[kBufSize];
+  uint32_t dst_stride_;
+  uint32_t dst_offset_;
+
+  SrcPixel src0_[kBufSize];
+  uint32_t src0_stride_;
+  uint32_t src0_offset_;
+
+  SrcPixel src1_[kBufSize];
+  uint32_t src1_stride_;
+  uint32_t src1_offset_;
+
+  uint8_t mask_[kMaxMaskSize];
+  size_t mask_stride_;
+
+  int w_;
+  int h_;
+
+  int suby_;
+  int subx_;
+};
+
+//////////////////////////////////////////////////////////////////////////////
+// 8 bit version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*F8B)(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
+                    uint32_t src0_stride, const uint8_t *src1,
+                    uint32_t src1_stride, const uint8_t *mask,
+                    uint32_t mask_stride, int w, int h, int subx, int suby);
+typedef libaom_test::FuncParam<F8B> TestFuncs;
+
+class BlendA64MaskTest8B : public BlendA64MaskTest<F8B, uint8_t, uint8_t> {
+ protected:
+  void Execute(const uint8_t *p_src0, const uint8_t *p_src1, int run_times) {
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      params_.ref_func(dst_ref_ + dst_offset_, dst_stride_,
+                       p_src0 + src0_offset_, src0_stride_,
+                       p_src1 + src1_offset_, src1_stride_, mask_,
+                       kMaxMaskWidth, w_, h_, subx_, suby_);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      params_.tst_func(dst_tst_ + dst_offset_, dst_stride_,
+                       p_src0 + src0_offset_, src0_stride_,
+                       p_src1 + src1_offset_, src1_stride_, mask_,
+                       kMaxMaskWidth, w_, h_, subx_, suby_);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    if (run_times > 1) {
+      printf("%3dx%-3d subx %d suby %d :%7.2f/%7.2fns", w_, h_, subx_, suby_,
+             time1, time2);
+      printf("(%3.2f)\n", time1 / time2);
+    }
+  }
+};
+
+TEST_P(BlendA64MaskTest8B, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    int bsize = rng_.Rand8() % BLOCK_SIZES_ALL;
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_.Rand8();
+      dst_tst_[i] = rng_.Rand8();
+
+      src0_[i] = rng_.Rand8();
+      src1_[i] = rng_.Rand8();
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(AOM_BLEND_A64_MAX_ALPHA + 1);
+
+    RunTest(bsize, 1);
+  }
+}
+
+TEST_P(BlendA64MaskTest8B, ExtremeValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    int bsize = rng_.Rand8() % BLOCK_SIZES_ALL;
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_(2) + 254;
+      dst_tst_[i] = rng_(2) + 254;
+      src0_[i] = rng_(2) + 254;
+      src1_[i] = rng_(2) + 254;
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(2) + AOM_BLEND_A64_MAX_ALPHA - 1;
+
+    RunTest(bsize, 1);
+  }
+}
+TEST_P(BlendA64MaskTest8B, DISABLED_Speed) {
+  const int kRunTimes = 10000000;
+  for (int bsize = 0; bsize < BLOCK_SIZES_ALL; ++bsize) {
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_.Rand8();
+      dst_tst_[i] = rng_.Rand8();
+
+      src0_[i] = rng_.Rand8();
+      src1_[i] = rng_.Rand8();
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(AOM_BLEND_A64_MAX_ALPHA + 1);
+
+    RunOneTest(bsize, 1, 1, kRunTimes);
+    RunOneTest(bsize, 1, 0, kRunTimes);
+    RunOneTest(bsize, 0, 1, kRunTimes);
+    RunOneTest(bsize, 0, 0, kRunTimes);
+  }
+}
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, BlendA64MaskTest8B,
+                        ::testing::Values(TestFuncs(
+                            aom_blend_a64_mask_c, aom_blend_a64_mask_sse4_1)));
+#endif  // HAVE_AVX2
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, BlendA64MaskTest8B,
+                        ::testing::Values(TestFuncs(aom_blend_a64_mask_sse4_1,
+                                                    aom_blend_a64_mask_avx2)));
+#endif  // HAVE_SSE4_1
+
+//////////////////////////////////////////////////////////////////////////////
+// 8 bit _d16 version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*F8B_D16)(uint8_t *dst, uint32_t dst_stride, const uint16_t *src0,
+                        uint32_t src0_stride, const uint16_t *src1,
+                        uint32_t src1_stride, const uint8_t *mask,
+                        uint32_t mask_stride, int w, int h, int subx, int suby,
+                        ConvolveParams *conv_params);
+typedef libaom_test::FuncParam<F8B_D16> TestFuncs_d16;
+
+class BlendA64MaskTest8B_d16
+    : public BlendA64MaskTest<F8B_D16, uint16_t, uint8_t> {
+ protected:
+  // max number of bits used by the source
+  static const int kSrcMaxBitsMask = 0x3fff;
+
+  void Execute(const uint16_t *p_src0, const uint16_t *p_src1, int run_times) {
+    ConvolveParams conv_params;
+    conv_params.round_0 = ROUND0_BITS;
+    conv_params.round_1 = COMPOUND_ROUND1_BITS;
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      params_.ref_func(dst_ref_ + dst_offset_, dst_stride_,
+                       p_src0 + src0_offset_, src0_stride_,
+                       p_src1 + src1_offset_, src1_stride_, mask_,
+                       kMaxMaskWidth, w_, h_, subx_, suby_, &conv_params);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      params_.tst_func(dst_tst_ + dst_offset_, dst_stride_,
+                       p_src0 + src0_offset_, src0_stride_,
+                       p_src1 + src1_offset_, src1_stride_, mask_,
+                       kMaxMaskWidth, w_, h_, subx_, suby_, &conv_params);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    if (run_times > 1) {
+      printf("%3dx%-3d subx %d suby %d :%7.2f/%7.2fns", w_, h_, subx_, suby_,
+             time1, time2);
+      printf("(%3.2f)\n", time1 / time2);
+    }
+  }
+};
+
+TEST_P(BlendA64MaskTest8B_d16, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    int bsize = rng_.Rand8() % BLOCK_SIZES_ALL;
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_.Rand8();
+      dst_tst_[i] = rng_.Rand8();
+
+      src0_[i] = rng_.Rand16() & kSrcMaxBitsMask;
+      src1_[i] = rng_.Rand16() & kSrcMaxBitsMask;
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(AOM_BLEND_A64_MAX_ALPHA + 1);
+
+    RunTest(bsize, 1);
+  }
+}
+
+TEST_P(BlendA64MaskTest8B_d16, ExtremeValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    int bsize = rng_.Rand8() % BLOCK_SIZES_ALL;
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = 255;
+      dst_tst_[i] = 255;
+
+      src0_[i] = kSrcMaxBitsMask;
+      src1_[i] = kSrcMaxBitsMask;
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = AOM_BLEND_A64_MAX_ALPHA - 1;
+
+    RunTest(bsize, 1);
+  }
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, BlendA64MaskTest8B_d16,
+    ::testing::Values(TestFuncs_d16(aom_lowbd_blend_a64_d16_mask_c,
+                                    aom_lowbd_blend_a64_d16_mask_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, BlendA64MaskTest8B_d16,
+    ::testing::Values(TestFuncs_d16(aom_lowbd_blend_a64_d16_mask_c,
+                                    aom_lowbd_blend_a64_d16_mask_avx2)));
+#endif  // HAVE_AVX2
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(
+    NEON, BlendA64MaskTest8B_d16,
+    ::testing::Values(TestFuncs_d16(aom_lowbd_blend_a64_d16_mask_c,
+                                    aom_lowbd_blend_a64_d16_mask_neon)));
+#endif  // HAVE_NEON
+
+//////////////////////////////////////////////////////////////////////////////
+// High bit-depth version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*FHBD)(uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
+                     uint32_t src0_stride, const uint8_t *src1,
+                     uint32_t src1_stride, const uint8_t *mask,
+                     uint32_t mask_stride, int w, int h, int subx, int suby,
+                     int bd);
+typedef libaom_test::FuncParam<FHBD> TestFuncsHBD;
+
+class BlendA64MaskTestHBD : public BlendA64MaskTest<FHBD, uint16_t, uint16_t> {
+ protected:
+  void Execute(const uint16_t *p_src0, const uint16_t *p_src1, int run_times) {
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      params_.ref_func(CONVERT_TO_BYTEPTR(dst_ref_ + dst_offset_), dst_stride_,
+                       CONVERT_TO_BYTEPTR(p_src0 + src0_offset_), src0_stride_,
+                       CONVERT_TO_BYTEPTR(p_src1 + src1_offset_), src1_stride_,
+                       mask_, kMaxMaskWidth, w_, h_, subx_, suby_, bit_depth_);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      params_.tst_func(CONVERT_TO_BYTEPTR(dst_tst_ + dst_offset_), dst_stride_,
+                       CONVERT_TO_BYTEPTR(p_src0 + src0_offset_), src0_stride_,
+                       CONVERT_TO_BYTEPTR(p_src1 + src1_offset_), src1_stride_,
+                       mask_, kMaxMaskWidth, w_, h_, subx_, suby_, bit_depth_);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    if (run_times > 1) {
+      printf("%3dx%-3d subx %d suby %d :%7.2f/%7.2fns", w_, h_, subx_, suby_,
+             time1, time2);
+      printf("(%3.2f)\n", time1 / time2);
+    }
+  }
+
+  int bit_depth_;
+};
+
+TEST_P(BlendA64MaskTestHBD, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    int bsize = rng_.Rand8() % BLOCK_SIZES_ALL;
+    switch (rng_(3)) {
+      case 0: bit_depth_ = 8; break;
+      case 1: bit_depth_ = 10; break;
+      default: bit_depth_ = 12; break;
+    }
+
+    const int hi = 1 << bit_depth_;
+
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_(hi);
+      dst_tst_[i] = rng_(hi);
+      src0_[i] = rng_(hi);
+      src1_[i] = rng_(hi);
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(AOM_BLEND_A64_MAX_ALPHA + 1);
+
+    RunTest(bsize, 1);
+  }
+}
+
+TEST_P(BlendA64MaskTestHBD, ExtremeValues) {
+  for (int iter = 0; iter < 1000 && !HasFatalFailure(); ++iter) {
+    int bsize = rng_.Rand8() % BLOCK_SIZES_ALL;
+    switch (rng_(3)) {
+      case 0: bit_depth_ = 8; break;
+      case 1: bit_depth_ = 10; break;
+      default: bit_depth_ = 12; break;
+    }
+
+    const int hi = 1 << bit_depth_;
+    const int lo = hi - 2;
+
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_(hi - lo) + lo;
+      dst_tst_[i] = rng_(hi - lo) + lo;
+      src0_[i] = rng_(hi - lo) + lo;
+      src1_[i] = rng_(hi - lo) + lo;
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(2) + AOM_BLEND_A64_MAX_ALPHA - 1;
+
+    RunTest(bsize, 1);
+  }
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, BlendA64MaskTestHBD,
+    ::testing::Values(TestFuncsHBD(aom_highbd_blend_a64_mask_c,
+                                   aom_highbd_blend_a64_mask_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+//////////////////////////////////////////////////////////////////////////////
+// HBD _d16 version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*FHBD_D16)(uint8_t *dst, uint32_t dst_stride,
+                         const CONV_BUF_TYPE *src0, uint32_t src0_stride,
+                         const CONV_BUF_TYPE *src1, uint32_t src1_stride,
+                         const uint8_t *mask, uint32_t mask_stride, int w,
+                         int h, int subx, int suby, ConvolveParams *conv_params,
+                         const int bd);
+typedef libaom_test::FuncParam<FHBD_D16> TestFuncsHBD_d16;
+
+class BlendA64MaskTestHBD_d16
+    : public BlendA64MaskTest<FHBD_D16, uint16_t, uint16_t> {
+ protected:
+  // max number of bits used by the source
+  static const int kSrcMaxBitsMask = (1 << 14) - 1;
+  static const int kSrcMaxBitsMaskHBD = (1 << 16) - 1;
+
+  void Execute(const uint16_t *p_src0, const uint16_t *p_src1, int run_times) {
+    ConvolveParams conv_params;
+    conv_params.round_0 = (bit_depth_ == 12) ? ROUND0_BITS + 2 : ROUND0_BITS;
+    conv_params.round_1 = COMPOUND_ROUND1_BITS;
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      params_.ref_func(CONVERT_TO_BYTEPTR(dst_ref_ + dst_offset_), dst_stride_,
+                       p_src0 + src0_offset_, src0_stride_,
+                       p_src1 + src1_offset_, src1_stride_, mask_,
+                       kMaxMaskWidth, w_, h_, subx_, suby_, &conv_params,
+                       bit_depth_);
+    }
+    if (params_.tst_func) {
+      aom_usec_timer_mark(&timer);
+      const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+      aom_usec_timer_start(&timer);
+      for (int i = 0; i < run_times; ++i) {
+        params_.tst_func(CONVERT_TO_BYTEPTR(dst_tst_ + dst_offset_),
+                         dst_stride_, p_src0 + src0_offset_, src0_stride_,
+                         p_src1 + src1_offset_, src1_stride_, mask_,
+                         kMaxMaskWidth, w_, h_, subx_, suby_, &conv_params,
+                         bit_depth_);
+      }
+      aom_usec_timer_mark(&timer);
+      const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+      if (run_times > 1) {
+        printf("%3dx%-3d subx %d suby %d :%7.2f/%7.2fns", w_, h_, subx_, suby_,
+               time1, time2);
+        printf("(%3.2f)\n", time1 / time2);
+      }
+    }
+  }
+
+  int bit_depth_;
+  int src_max_bits_mask_;
+};
+
+TEST_P(BlendA64MaskTestHBD_d16, RandomValues) {
+  if (params_.tst_func == NULL) return;
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    int bsize = rng_.Rand8() % BLOCK_SIZES_ALL;
+    switch (rng_(3)) {
+      case 0: bit_depth_ = 8; break;
+      case 1: bit_depth_ = 10; break;
+      default: bit_depth_ = 12; break;
+    }
+    src_max_bits_mask_ =
+        (bit_depth_ == 8) ? kSrcMaxBitsMask : kSrcMaxBitsMaskHBD;
+
+    for (int i = 0; i < kBufSize; ++i) {
+      dst_ref_[i] = rng_.Rand8();
+      dst_tst_[i] = rng_.Rand8();
+
+      src0_[i] = rng_.Rand16() & src_max_bits_mask_;
+      src1_[i] = rng_.Rand16() & src_max_bits_mask_;
+    }
+
+    for (int i = 0; i < kMaxMaskSize; ++i)
+      mask_[i] = rng_(AOM_BLEND_A64_MAX_ALPHA + 1);
+
+    RunTest(bsize, 1);
+  }
+}
+// TODO (Scott LaVarnway), fix this test
+TEST_P(BlendA64MaskTestHBD_d16, DISABLED_SaturatedValues) {
+  for (int bsize = 0; bsize < BLOCK_SIZES_ALL; ++bsize) {
+    for (bit_depth_ = 8; bit_depth_ <= 12; bit_depth_ += 2) {
+      src_max_bits_mask_ =
+          (bit_depth_ == 8) ? kSrcMaxBitsMask : kSrcMaxBitsMaskHBD;
+
+      for (int i = 0; i < kBufSize; ++i) {
+        dst_ref_[i] = 0;
+        dst_tst_[i] = (1 << bit_depth_) - 1;
+
+        src0_[i] = src_max_bits_mask_;
+        src1_[i] = src_max_bits_mask_;
+      }
+
+      for (int i = 0; i < kMaxMaskSize; ++i) mask_[i] = AOM_BLEND_A64_MAX_ALPHA;
+
+      RunTest(bsize, 1);
+    }
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(
+    C, BlendA64MaskTestHBD_d16,
+    ::testing::Values(TestFuncsHBD_d16(aom_highbd_blend_a64_d16_mask_c, NULL)));
+
+// TODO(slavarnway): Enable the following in the avx2 commit. (56501)
+#if 0
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, BlendA64MaskTestHBD,
+    ::testing::Values(TestFuncsHBD(aom_highbd_blend_a64_mask_c,
+                                   aom_highbd_blend_a64_mask_avx2)));
+#endif  // HAVE_AVX2
+#endif
+}  // namespace
diff --git a/third_party/aom/test/blockd_test.cc b/third_party/aom/test/blockd_test.cc
new file mode 100644
index 0000000000..ab624007ce
--- /dev/null
+++ b/third_party/aom/test/blockd_test.cc
@@ -0,0 +1,122 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "av1/common/blockd.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+// Verify the optimized implementation of get_partition_subsize() produces the
+// same results as the Partition_Subsize lookup table in the spec.
+TEST(BlockdTest, GetPartitionSubsize) {
+  // The Partition_Subsize table in the spec (Section 9.3. Conversion tables).
+  /* clang-format off */
+  static const BLOCK_SIZE kPartitionSubsize[10][BLOCK_SIZES_ALL] = {
+    {
+                                    BLOCK_4X4,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_128X128,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X4,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_128X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_4X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X128,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_4X4,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X4,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_128X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X4,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_128X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_4X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X128,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_4X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X128,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X4,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_32X8,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_64X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }, {
+                                    BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_4X16,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_8X32,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_16X64,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID,
+      BLOCK_INVALID, BLOCK_INVALID, BLOCK_INVALID
+    }
+  };
+  /* clang-format on */
+
+  for (int partition = 0; partition < 10; partition++) {
+    for (int bsize = BLOCK_4X4; bsize < BLOCK_SIZES_ALL; bsize++) {
+      EXPECT_EQ(kPartitionSubsize[partition][bsize],
+                get_partition_subsize(static_cast<BLOCK_SIZE>(bsize),
+                                      static_cast<PARTITION_TYPE>(partition)));
+    }
+  }
+}
diff --git a/third_party/aom/test/boolcoder_test.cc b/third_party/aom/test/boolcoder_test.cc
new file mode 100644
index 0000000000..680ec18774
--- /dev/null
+++ b/third_party/aom/test/boolcoder_test.cc
@@ -0,0 +1,173 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/acm_random.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/bitreader.h"
+#include "aom_dsp/bitwriter.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+const int num_tests = 10;
+}  // namespace
+
+TEST(AV1, TestBitIO) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  for (int n = 0; n < num_tests; ++n) {
+    for (int method = 0; method <= 7; ++method) {  // we generate various proba
+      const int kBitsToTest = 1000;
+      uint8_t probas[kBitsToTest];
+
+      for (int i = 0; i < kBitsToTest; ++i) {
+        const int parity = i & 1;
+        /* clang-format off */
+        probas[i] =
+          (method == 0) ? 0 : (method == 1) ? 255 :
+          (method == 2) ? 128 :
+          (method == 3) ? rnd.Rand8() :
+          (method == 4) ? (parity ? 0 : 255) :
+            // alternate between low and high proba:
+            (method == 5) ? (parity ? rnd(128) : 255 - rnd(128)) :
+            (method == 6) ?
+            (parity ? rnd(64) : 255 - rnd(64)) :
+            (parity ? rnd(32) : 255 - rnd(32));
+        /* clang-format on */
+      }
+      for (int bit_method = 0; bit_method <= 3; ++bit_method) {
+        const int random_seed = 6432;
+        const int kBufferSize = 10000;
+        ACMRandom bit_rnd(random_seed);
+        aom_writer bw;
+        uint8_t bw_buffer[kBufferSize];
+        aom_start_encode(&bw, bw_buffer);
+
+        int bit = (bit_method == 0) ? 0 : (bit_method == 1) ? 1 : 0;
+        for (int i = 0; i < kBitsToTest; ++i) {
+          if (bit_method == 2) {
+            bit = (i & 1);
+          } else if (bit_method == 3) {
+            bit = bit_rnd(2);
+          }
+          aom_write(&bw, bit, static_cast<int>(probas[i]));
+        }
+
+        aom_stop_encode(&bw);
+
+        aom_reader br;
+        aom_reader_init(&br, bw_buffer, bw.pos);
+        bit_rnd.Reset(random_seed);
+        for (int i = 0; i < kBitsToTest; ++i) {
+          if (bit_method == 2) {
+            bit = (i & 1);
+          } else if (bit_method == 3) {
+            bit = bit_rnd(2);
+          }
+          GTEST_ASSERT_EQ(aom_read(&br, probas[i], NULL), bit)
+              << "pos: " << i << " / " << kBitsToTest
+              << " bit_method: " << bit_method << " method: " << method;
+        }
+      }
+    }
+  }
+}
+
+#define FRAC_DIFF_TOTAL_ERROR 0.18
+
+TEST(AV1, TestTell) {
+  const int kBufferSize = 10000;
+  aom_writer bw;
+  uint8_t bw_buffer[kBufferSize];
+  const int kSymbols = 1024;
+  // Coders are noisier at low probabilities, so we start at p = 4.
+  for (int p = 4; p < 256; p++) {
+    double probability = p / 256.;
+    aom_start_encode(&bw, bw_buffer);
+    for (int i = 0; i < kSymbols; i++) {
+      aom_write(&bw, 0, p);
+    }
+    aom_stop_encode(&bw);
+    aom_reader br;
+    aom_reader_init(&br, bw_buffer, bw.pos);
+    uint32_t last_tell = aom_reader_tell(&br);
+    uint32_t last_tell_frac = aom_reader_tell_frac(&br);
+    double frac_diff_total = 0;
+    GTEST_ASSERT_GE(aom_reader_tell(&br), 0u);
+    GTEST_ASSERT_LE(aom_reader_tell(&br), 1u);
+    ASSERT_FALSE(aom_reader_has_overflowed(&br));
+    for (int i = 0; i < kSymbols; i++) {
+      aom_read(&br, p, NULL);
+      uint32_t tell = aom_reader_tell(&br);
+      uint32_t tell_frac = aom_reader_tell_frac(&br);
+      GTEST_ASSERT_GE(tell, last_tell)
+          << "tell: " << tell << ", last_tell: " << last_tell;
+      GTEST_ASSERT_GE(tell_frac, last_tell_frac)
+          << "tell_frac: " << tell_frac
+          << ", last_tell_frac: " << last_tell_frac;
+      // Frac tell should round up to tell.
+      GTEST_ASSERT_EQ(tell, (tell_frac + 7) >> 3);
+      last_tell = tell;
+      frac_diff_total +=
+          fabs(((tell_frac - last_tell_frac) / 8.0) + log2(probability));
+      last_tell_frac = tell_frac;
+    }
+    const uint32_t expected = (uint32_t)(-kSymbols * log2(probability));
+    // Last tell should be close to the expected value.
+    GTEST_ASSERT_LE(last_tell, expected + 20) << " last_tell: " << last_tell;
+    // The average frac_diff error should be pretty small.
+    GTEST_ASSERT_LE(frac_diff_total / kSymbols, FRAC_DIFF_TOTAL_ERROR)
+        << " frac_diff_total: " << frac_diff_total;
+    ASSERT_FALSE(aom_reader_has_overflowed(&br));
+  }
+}
+
+TEST(AV1, TestHasOverflowed) {
+  const int kBufferSize = 10000;
+  aom_writer bw;
+  uint8_t bw_buffer[kBufferSize];
+  const int kSymbols = 1024;
+  // Coders are noisier at low probabilities, so we start at p = 4.
+  for (int p = 4; p < 256; p++) {
+    aom_start_encode(&bw, bw_buffer);
+    for (int i = 0; i < kSymbols; i++) {
+      aom_write(&bw, 1, p);
+    }
+    aom_stop_encode(&bw);
+    aom_reader br;
+    aom_reader_init(&br, bw_buffer, bw.pos);
+    ASSERT_FALSE(aom_reader_has_overflowed(&br));
+    for (int i = 0; i < kSymbols; i++) {
+      GTEST_ASSERT_EQ(aom_read(&br, p, NULL), 1);
+      ASSERT_FALSE(aom_reader_has_overflowed(&br));
+    }
+    // In the worst case, the encoder uses just a tiny fraction of the last
+    // byte in the buffer. So to guarantee that aom_reader_has_overflowed()
+    // returns true, we have to consume very nearly 8 additional bits of data.
+    // In the worse case, one of the bits in that byte will be 1, and the rest
+    // will be zero. Once we are past that 1 bit, when the probability of
+    // reading zero symbol from aom_read() is high, each additional symbol read
+    // will consume very little additional data (in the case that p == 255,
+    // approximately -log_2(255/256) ~= 0.0056 bits). In that case it would
+    // take around 178 calls to consume more than 8 bits. That is only an upper
+    // bound. In practice we are not guaranteed to hit the worse case and can
+    // get away with 174 calls.
+    for (int i = 0; i < 174; i++) {
+      aom_read(&br, p, NULL);
+    }
+    ASSERT_TRUE(aom_reader_has_overflowed(&br));
+  }
+}
diff --git a/third_party/aom/test/borders_test.cc b/third_party/aom/test/borders_test.cc
new file mode 100644
index 0000000000..893237ef30
--- /dev/null
+++ b/third_party/aom/test/borders_test.cc
@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <climits>
+#include <vector>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+class BordersTestLarge
+    : public ::libaom_test::CodecTestWithParam<libaom_test::TestMode>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  BordersTestLarge() : EncoderTest(GET_PARAM(0)) {}
+  virtual ~BordersTestLarge() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_CPUUSED, 1);
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+      encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+      encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+    }
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (pkt->data.frame.flags & AOM_FRAME_IS_KEY) {
+    }
+  }
+};
+
+TEST_P(BordersTestLarge, TestEncodeHighBitrate) {
+  // Validate that this non multiple of 64 wide clip encodes and decodes
+  // without a mismatch when passing in a very low max q.  This pushes
+  // the encoder to producing lots of big partitions which will likely
+  // extend into the border and test the border condition.
+  cfg_.g_lag_in_frames = 25;
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_maxsection_pct = 2000;
+  cfg_.rc_target_bitrate = 2000;
+  cfg_.rc_max_quantizer = 10;
+
+  ::libaom_test::I420VideoSource video("hantro_odd.yuv", 208, 144, 30, 1, 0,
+                                       10);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+TEST_P(BordersTestLarge, TestLowBitrate) {
+  // Validate that this clip encodes and decodes without a mismatch
+  // when passing in a very high min q.  This pushes the encoder to producing
+  // lots of small partitions which might will test the other condition.
+
+  cfg_.g_lag_in_frames = 25;
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_maxsection_pct = 2000;
+  cfg_.rc_target_bitrate = 200;
+  cfg_.rc_min_quantizer = 40;
+
+  ::libaom_test::I420VideoSource video("hantro_odd.yuv", 208, 144, 30, 1, 0,
+                                       10);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
+AV1_INSTANTIATE_TEST_CASE(BordersTestLarge,
+                          ::testing::Values(::libaom_test::kTwoPassGood));
+}  // namespace
diff --git a/third_party/aom/test/cdef_test.cc b/third_party/aom/test/cdef_test.cc
new file mode 100644
index 0000000000..becc072918
--- /dev/null
+++ b/third_party/aom/test/cdef_test.cc
@@ -0,0 +1,425 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cstdlib>
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/aom_timer.h"
+#include "av1/common/cdef_block.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+
+typedef ::testing::tuple<cdef_filter_block_func, cdef_filter_block_func,
+                         BLOCK_SIZE, int, int>
+    cdef_dir_param_t;
+
+class CDEFBlockTest : public ::testing::TestWithParam<cdef_dir_param_t> {
+ public:
+  virtual ~CDEFBlockTest() {}
+  virtual void SetUp() {
+    cdef = GET_PARAM(0);
+    ref_cdef = GET_PARAM(1);
+    bsize = GET_PARAM(2);
+    boundary = GET_PARAM(3);
+    depth = GET_PARAM(4);
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  int bsize;
+  int boundary;
+  int depth;
+  cdef_filter_block_func cdef;
+  cdef_filter_block_func ref_cdef;
+};
+
+typedef CDEFBlockTest CDEFSpeedTest;
+
+void test_cdef(int bsize, int iterations, cdef_filter_block_func cdef,
+               cdef_filter_block_func ref_cdef, int boundary, int depth) {
+  const int size = 8;
+  const int ysize = size + 2 * CDEF_VBORDER;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint16_t, s[ysize * CDEF_BSTRIDE]);
+  DECLARE_ALIGNED(16, static uint16_t, d[size * size]);
+  DECLARE_ALIGNED(16, static uint16_t, ref_d[size * size]);
+  memset(ref_d, 0, sizeof(ref_d));
+  memset(d, 0, sizeof(d));
+
+  int error = 0, pristrength = 0, secstrength, dir;
+  int pridamping, secdamping, bits, level, count,
+      errdepth = 0, errpristrength = 0, errsecstrength = 0, errboundary = 0,
+      errpridamping = 0, errsecdamping = 0;
+  unsigned int pos = 0;
+
+  const unsigned int max_pos = size * size >> static_cast<int>(depth == 8);
+  for (pridamping = 3 + depth - 8; pridamping < 7 - 3 * !!boundary + depth - 8;
+       pridamping++) {
+    for (secdamping = 3 + depth - 8;
+         secdamping < 7 - 3 * !!boundary + depth - 8; secdamping++) {
+      for (count = 0; count < iterations; count++) {
+        for (level = 0; level < (1 << depth) && !error;
+             level += (2 + 6 * !!boundary) << (depth - 8)) {
+          for (bits = 1; bits <= depth && !error; bits += 1 + 3 * !!boundary) {
+            for (unsigned int i = 0; i < sizeof(s) / sizeof(*s); i++)
+              s[i] = clamp((rnd.Rand16() & ((1 << bits) - 1)) + level, 0,
+                           (1 << depth) - 1);
+            if (boundary) {
+              if (boundary & 1) {  // Left
+                for (int i = 0; i < ysize; i++)
+                  for (int j = 0; j < CDEF_HBORDER; j++)
+                    s[i * CDEF_BSTRIDE + j] = CDEF_VERY_LARGE;
+              }
+              if (boundary & 2) {  // Right
+                for (int i = 0; i < ysize; i++)
+                  for (int j = CDEF_HBORDER + size; j < CDEF_BSTRIDE; j++)
+                    s[i * CDEF_BSTRIDE + j] = CDEF_VERY_LARGE;
+              }
+              if (boundary & 4) {  // Above
+                for (int i = 0; i < CDEF_VBORDER; i++)
+                  for (int j = 0; j < CDEF_BSTRIDE; j++)
+                    s[i * CDEF_BSTRIDE + j] = CDEF_VERY_LARGE;
+              }
+              if (boundary & 8) {  // Below
+                for (int i = CDEF_VBORDER + size; i < ysize; i++)
+                  for (int j = 0; j < CDEF_BSTRIDE; j++)
+                    s[i * CDEF_BSTRIDE + j] = CDEF_VERY_LARGE;
+              }
+            }
+            for (dir = 0; dir < 8; dir++) {
+              for (pristrength = 0; pristrength <= 19 << (depth - 8) && !error;
+                   pristrength += (1 + 4 * !!boundary) << (depth - 8)) {
+                if (pristrength == 16) pristrength = 19;
+                for (secstrength = 0; secstrength <= 4 << (depth - 8) && !error;
+                     secstrength += 1 << (depth - 8)) {
+                  if (secstrength == 3 << (depth - 8)) continue;
+                  ref_cdef(depth == 8 ? (uint8_t *)ref_d : 0, ref_d, size,
+                           s + CDEF_HBORDER + CDEF_VBORDER * CDEF_BSTRIDE,
+                           pristrength, secstrength, dir, pridamping,
+                           secdamping, bsize, (1 << depth) - 1, depth - 8);
+                  // If cdef and ref_cdef are the same, we're just testing
+                  // speed
+                  if (cdef != ref_cdef)
+                    ASM_REGISTER_STATE_CHECK(
+                        cdef(depth == 8 ? (uint8_t *)d : 0, d, size,
+                             s + CDEF_HBORDER + CDEF_VBORDER * CDEF_BSTRIDE,
+                             pristrength, secstrength, dir, pridamping,
+                             secdamping, bsize, (1 << depth) - 1, depth - 8));
+                  if (ref_cdef != cdef) {
+                    for (pos = 0; pos < max_pos && !error; pos++) {
+                      error = ref_d[pos] != d[pos];
+                      errdepth = depth;
+                      errpristrength = pristrength;
+                      errsecstrength = secstrength;
+                      errboundary = boundary;
+                      errpridamping = pridamping;
+                      errsecdamping = secdamping;
+                    }
+                  }
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+
+  pos--;
+  EXPECT_EQ(0, error) << "Error: CDEFBlockTest, SIMD and C mismatch."
+                      << std::endl
+                      << "First error at " << pos % size << "," << pos / size
+                      << " (" << (int16_t)ref_d[pos] << " : " << (int16_t)d[pos]
+                      << ") " << std::endl
+                      << "pristrength: " << errpristrength << std::endl
+                      << "pridamping: " << errpridamping << std::endl
+                      << "secstrength: " << errsecstrength << std::endl
+                      << "secdamping: " << errsecdamping << std::endl
+                      << "depth: " << errdepth << std::endl
+                      << "size: " << bsize << std::endl
+                      << "boundary: " << errboundary << std::endl
+                      << std::endl;
+}
+
+void test_cdef_speed(int bsize, int iterations, cdef_filter_block_func cdef,
+                     cdef_filter_block_func ref_cdef, int boundary, int depth) {
+  aom_usec_timer ref_timer;
+  aom_usec_timer timer;
+
+  aom_usec_timer_start(&ref_timer);
+  test_cdef(bsize, iterations, ref_cdef, ref_cdef, boundary, depth);
+  aom_usec_timer_mark(&ref_timer);
+  int ref_elapsed_time = (int)aom_usec_timer_elapsed(&ref_timer);
+
+  aom_usec_timer_start(&timer);
+  test_cdef(bsize, iterations, cdef, cdef, boundary, depth);
+  aom_usec_timer_mark(&timer);
+  int elapsed_time = (int)aom_usec_timer_elapsed(&timer);
+
+  EXPECT_GT(ref_elapsed_time, elapsed_time)
+      << "Error: CDEFSpeedTest, SIMD slower than C." << std::endl
+      << "C time: " << ref_elapsed_time << " us" << std::endl
+      << "SIMD time: " << elapsed_time << " us" << std::endl;
+}
+
+typedef int (*find_dir_t)(const uint16_t *img, int stride, int32_t *var,
+                          int coeff_shift);
+
+typedef ::testing::tuple<find_dir_t, find_dir_t> find_dir_param_t;
+
+class CDEFFindDirTest : public ::testing::TestWithParam<find_dir_param_t> {
+ public:
+  virtual ~CDEFFindDirTest() {}
+  virtual void SetUp() {
+    finddir = GET_PARAM(0);
+    ref_finddir = GET_PARAM(1);
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  find_dir_t finddir;
+  find_dir_t ref_finddir;
+};
+
+typedef CDEFFindDirTest CDEFFindDirSpeedTest;
+
+void test_finddir(int (*finddir)(const uint16_t *img, int stride, int32_t *var,
+                                 int coeff_shift),
+                  int (*ref_finddir)(const uint16_t *img, int stride,
+                                     int32_t *var, int coeff_shift)) {
+  const int size = 8;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint16_t, s[size * size]);
+
+  int error = 0;
+  int depth, bits, level, count, errdepth = 0;
+  int ref_res = 0, res = 0;
+  int32_t ref_var = 0, var = 0;
+
+  for (depth = 8; depth <= 12 && !error; depth += 2) {
+    for (count = 0; count < 512 && !error; count++) {
+      for (level = 0; level < (1 << depth) && !error;
+           level += 1 << (depth - 8)) {
+        for (bits = 1; bits <= depth && !error; bits++) {
+          for (unsigned int i = 0; i < sizeof(s) / sizeof(*s); i++)
+            s[i] = clamp((rnd.Rand16() & ((1 << bits) - 1)) + level, 0,
+                         (1 << depth) - 1);
+          for (int c = 0; c < 1 + 9 * (finddir == ref_finddir); c++)
+            ref_res = ref_finddir(s, size, &ref_var, depth - 8);
+          if (finddir != ref_finddir)
+            ASM_REGISTER_STATE_CHECK(res = finddir(s, size, &var, depth - 8));
+          if (ref_finddir != finddir) {
+            if (res != ref_res || var != ref_var) error = 1;
+            errdepth = depth;
+          }
+        }
+      }
+    }
+  }
+
+  EXPECT_EQ(0, error) << "Error: CDEFFindDirTest, SIMD and C mismatch."
+                      << std::endl
+                      << "return: " << res << " : " << ref_res << std::endl
+                      << "var: " << var << " : " << ref_var << std::endl
+                      << "depth: " << errdepth << std::endl
+                      << std::endl;
+}
+
+void test_finddir_speed(int (*finddir)(const uint16_t *img, int stride,
+                                       int32_t *var, int coeff_shift),
+                        int (*ref_finddir)(const uint16_t *img, int stride,
+                                           int32_t *var, int coeff_shift)) {
+  aom_usec_timer ref_timer;
+  aom_usec_timer timer;
+
+  aom_usec_timer_start(&ref_timer);
+  test_finddir(ref_finddir, ref_finddir);
+  aom_usec_timer_mark(&ref_timer);
+  int ref_elapsed_time = (int)aom_usec_timer_elapsed(&ref_timer);
+
+  aom_usec_timer_start(&timer);
+  test_finddir(finddir, finddir);
+  aom_usec_timer_mark(&timer);
+  int elapsed_time = (int)aom_usec_timer_elapsed(&timer);
+
+  EXPECT_GT(ref_elapsed_time, elapsed_time)
+      << "Error: CDEFFindDirSpeedTest, SIMD slower than C." << std::endl
+      << "C time: " << ref_elapsed_time << " us" << std::endl
+      << "SIMD time: " << elapsed_time << " us" << std::endl;
+}
+
+TEST_P(CDEFBlockTest, TestSIMDNoMismatch) {
+  test_cdef(bsize, 1, cdef, ref_cdef, boundary, depth);
+}
+
+TEST_P(CDEFSpeedTest, DISABLED_TestSpeed) {
+  test_cdef_speed(bsize, 4, cdef, ref_cdef, boundary, depth);
+}
+
+TEST_P(CDEFFindDirTest, TestSIMDNoMismatch) {
+  test_finddir(finddir, ref_finddir);
+}
+
+TEST_P(CDEFFindDirSpeedTest, DISABLED_TestSpeed) {
+  test_finddir_speed(finddir, ref_finddir);
+}
+
+using ::testing::make_tuple;
+
+// VS compiling for 32 bit targets does not support vector types in
+// structs as arguments, which makes the v256 type of the intrinsics
+// hard to support, so optimizations for this target are disabled.
+#if defined(_WIN64) || !defined(_MSC_VER) || defined(__clang__)
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, CDEFBlockTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_sse2),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(SSE2, CDEFFindDirTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_sse2,
+                                                     &cdef_find_dir_c)));
+#endif
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, CDEFBlockTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_ssse3),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(SSSE3, CDEFFindDirTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_ssse3,
+                                                     &cdef_find_dir_c)));
+#endif
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, CDEFBlockTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_sse4_1),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(SSE4_1, CDEFFindDirTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_sse4_1,
+                                                     &cdef_find_dir_c)));
+#endif
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, CDEFBlockTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_avx2),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(AVX2, CDEFFindDirTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_avx2,
+                                                     &cdef_find_dir_c)));
+#endif
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(
+    NEON, CDEFBlockTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_neon),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(NEON, CDEFFindDirTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_neon,
+                                                     &cdef_find_dir_c)));
+#endif
+
+// Test speed for all supported architectures
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, CDEFSpeedTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_sse2),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(SSE2, CDEFFindDirSpeedTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_sse2,
+                                                     &cdef_find_dir_c)));
+#endif
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, CDEFSpeedTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_ssse3),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(SSSE3, CDEFFindDirSpeedTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_ssse3,
+                                                     &cdef_find_dir_c)));
+#endif
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, CDEFSpeedTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_sse4_1),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(SSE4_1, CDEFFindDirSpeedTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_sse4_1,
+                                                     &cdef_find_dir_c)));
+#endif
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, CDEFSpeedTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_avx2),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(AVX2, CDEFFindDirSpeedTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_avx2,
+                                                     &cdef_find_dir_c)));
+#endif
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(
+    NEON, CDEFSpeedTest,
+    ::testing::Combine(::testing::Values(&cdef_filter_block_neon),
+                       ::testing::Values(&cdef_filter_block_c),
+                       ::testing::Values(BLOCK_4X4, BLOCK_4X8, BLOCK_8X4,
+                                         BLOCK_8X8),
+                       ::testing::Range(0, 16), ::testing::Range(8, 13, 2)));
+INSTANTIATE_TEST_CASE_P(NEON, CDEFFindDirSpeedTest,
+                        ::testing::Values(make_tuple(&cdef_find_dir_neon,
+                                                     &cdef_find_dir_c)));
+#endif
+
+#endif  // defined(_WIN64) || !defined(_MSC_VER)
+}  // namespace
diff --git a/third_party/aom/test/cfl_test.cc b/third_party/aom/test/cfl_test.cc
new file mode 100644
index 0000000000..e4d438d6ab
--- /dev/null
+++ b/third_party/aom/test/cfl_test.cc
@@ -0,0 +1,567 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/aom_timer.h"
+#include "test/util.h"
+#include "test/acm_random.h"
+
+using ::testing::make_tuple;
+
+using libaom_test::ACMRandom;
+
+#define NUM_ITERATIONS (100)
+#define NUM_ITERATIONS_SPEED (INT16_MAX)
+
+#define ALL_CFL_TX_SIZES(function)                                     \
+  make_tuple(TX_4X4, &function), make_tuple(TX_4X8, &function),        \
+      make_tuple(TX_4X16, &function), make_tuple(TX_8X4, &function),   \
+      make_tuple(TX_8X8, &function), make_tuple(TX_8X16, &function),   \
+      make_tuple(TX_8X32, &function), make_tuple(TX_16X4, &function),  \
+      make_tuple(TX_16X8, &function), make_tuple(TX_16X16, &function), \
+      make_tuple(TX_16X32, &function), make_tuple(TX_32X8, &function), \
+      make_tuple(TX_32X16, &function), make_tuple(TX_32X32, &function)
+
+#define ALL_CFL_TX_SIZES_SUBSAMPLE(fun420, fun422, fun444) \
+  make_tuple(TX_4X4, &fun420, &fun422, &fun444),           \
+      make_tuple(TX_4X8, &fun420, &fun422, &fun444),       \
+      make_tuple(TX_4X16, &fun420, &fun422, &fun444),      \
+      make_tuple(TX_8X4, &fun420, &fun422, &fun444),       \
+      make_tuple(TX_8X8, &fun420, &fun422, &fun444),       \
+      make_tuple(TX_8X16, &fun420, &fun422, &fun444),      \
+      make_tuple(TX_8X32, &fun420, &fun422, &fun444),      \
+      make_tuple(TX_16X4, &fun420, &fun422, &fun444),      \
+      make_tuple(TX_16X8, &fun420, &fun422, &fun444),      \
+      make_tuple(TX_16X16, &fun420, &fun422, &fun444),     \
+      make_tuple(TX_16X32, &fun420, &fun422, &fun444),     \
+      make_tuple(TX_32X8, &fun420, &fun422, &fun444),      \
+      make_tuple(TX_32X16, &fun420, &fun422, &fun444),     \
+      make_tuple(TX_32X32, &fun420, &fun422, &fun444)
+
+namespace {
+
+template <typename A>
+static void assert_eq(const A *a, const A *b, int width, int height) {
+  for (int j = 0; j < height; j++) {
+    for (int i = 0; i < width; i++) {
+      ASSERT_EQ(a[j * CFL_BUF_LINE + i], b[j * CFL_BUF_LINE + i]);
+    }
+  }
+}
+
+static void assertFaster(int ref_elapsed_time, int elapsed_time) {
+  EXPECT_GT(ref_elapsed_time, elapsed_time)
+      << "Error: CFLSubtractSpeedTest, SIMD slower than C." << std::endl
+      << "C time: " << ref_elapsed_time << " us" << std::endl
+      << "SIMD time: " << elapsed_time << " us" << std::endl;
+}
+
+static void printSpeed(int ref_elapsed_time, int elapsed_time, int width,
+                       int height) {
+  std::cout.precision(2);
+  std::cout << "[          ] " << width << "x" << height
+            << ": C time = " << ref_elapsed_time
+            << " us, SIMD time = " << elapsed_time << " us"
+            << " (~" << ref_elapsed_time / (double)elapsed_time << "x) "
+            << std::endl;
+}
+
+class CFLTest {
+ public:
+  virtual ~CFLTest() {}
+  void init(TX_SIZE tx) {
+    tx_size = tx;
+    width = tx_size_wide[tx_size];
+    height = tx_size_high[tx_size];
+    rnd(ACMRandom::DeterministicSeed());
+  }
+
+ protected:
+  TX_SIZE tx_size;
+  int width;
+  int height;
+  ACMRandom rnd;
+};
+
+template <typename I>
+class CFLTestWithData : public CFLTest {
+ public:
+  virtual ~CFLTestWithData() {}
+
+ protected:
+  I data[CFL_BUF_SQUARE];
+  I data_ref[CFL_BUF_SQUARE];
+  void randData(I (ACMRandom::*random)()) {
+    for (int j = 0; j < this->height; j++) {
+      for (int i = 0; i < this->width; i++) {
+        const I d = (this->rnd.*random)();
+        data[j * CFL_BUF_LINE + i] = d;
+        data_ref[j * CFL_BUF_LINE + i] = d;
+      }
+    }
+  }
+};
+
+template <typename I>
+class CFLTestWithAlignedData : public CFLTest {
+ public:
+  CFLTestWithAlignedData() {
+    chroma_pels_ref =
+        reinterpret_cast<I *>(aom_memalign(32, sizeof(I) * CFL_BUF_SQUARE));
+    chroma_pels =
+        reinterpret_cast<I *>(aom_memalign(32, sizeof(I) * CFL_BUF_SQUARE));
+    sub_luma_pels_ref = reinterpret_cast<int16_t *>(
+        aom_memalign(32, sizeof(int16_t) * CFL_BUF_SQUARE));
+    sub_luma_pels = reinterpret_cast<int16_t *>(
+        aom_memalign(32, sizeof(int16_t) * CFL_BUF_SQUARE));
+    memset(chroma_pels_ref, 0, sizeof(I) * CFL_BUF_SQUARE);
+    memset(chroma_pels, 0, sizeof(I) * CFL_BUF_SQUARE);
+    memset(sub_luma_pels_ref, 0, sizeof(int16_t) * CFL_BUF_SQUARE);
+    memset(sub_luma_pels, 0, sizeof(int16_t) * CFL_BUF_SQUARE);
+  }
+  ~CFLTestWithAlignedData() {
+    aom_free(chroma_pels_ref);
+    aom_free(sub_luma_pels_ref);
+    aom_free(chroma_pels);
+    aom_free(sub_luma_pels);
+  }
+
+ protected:
+  I *chroma_pels_ref;
+  I *chroma_pels;
+  int16_t *sub_luma_pels_ref;
+  int16_t *sub_luma_pels;
+  int alpha_q3;
+  I dc;
+  void randData(int bd) {
+    alpha_q3 = this->rnd(33) - 16;
+    dc = this->rnd(1 << bd);
+    for (int j = 0; j < this->height; j++) {
+      for (int i = 0; i < this->width; i++) {
+        chroma_pels[j * CFL_BUF_LINE + i] = dc;
+        chroma_pels_ref[j * CFL_BUF_LINE + i] = dc;
+        sub_luma_pels_ref[j * CFL_BUF_LINE + i] =
+            sub_luma_pels[j * CFL_BUF_LINE + i] = this->rnd(1 << (bd + 3));
+      }
+    }
+  }
+};
+
+typedef cfl_subtract_average_fn (*sub_avg_fn)(TX_SIZE tx_size);
+typedef ::testing::tuple<TX_SIZE, sub_avg_fn> sub_avg_param;
+class CFLSubAvgTest : public ::testing::TestWithParam<sub_avg_param>,
+                      public CFLTestWithData<int16_t> {
+ public:
+  virtual void SetUp() {
+    CFLTest::init(::testing::get<0>(this->GetParam()));
+    sub_avg = ::testing::get<1>(this->GetParam())(tx_size);
+    sub_avg_ref = get_subtract_average_fn_c(tx_size);
+  }
+  virtual ~CFLSubAvgTest() {}
+
+ protected:
+  cfl_subtract_average_fn sub_avg;
+  cfl_subtract_average_fn sub_avg_ref;
+};
+
+TEST_P(CFLSubAvgTest, SubAvgTest) {
+  for (int it = 0; it < NUM_ITERATIONS; it++) {
+    randData(&ACMRandom::Rand15Signed);
+    sub_avg((uint16_t *)data, data);
+    sub_avg_ref((uint16_t *)data_ref, data_ref);
+    assert_eq<int16_t>(data, data_ref, width, height);
+  }
+}
+
+TEST_P(CFLSubAvgTest, DISABLED_SubAvgSpeedTest) {
+  aom_usec_timer ref_timer;
+  aom_usec_timer timer;
+  randData(&ACMRandom::Rand15Signed);
+  aom_usec_timer_start(&ref_timer);
+  for (int k = 0; k < NUM_ITERATIONS_SPEED; k++) {
+    sub_avg_ref((uint16_t *)data_ref, data_ref);
+  }
+  aom_usec_timer_mark(&ref_timer);
+  int ref_elapsed_time = (int)aom_usec_timer_elapsed(&ref_timer);
+  aom_usec_timer_start(&timer);
+  for (int k = 0; k < NUM_ITERATIONS_SPEED; k++) {
+    sub_avg((uint16_t *)data, data);
+  }
+  aom_usec_timer_mark(&timer);
+  int elapsed_time = (int)aom_usec_timer_elapsed(&timer);
+  printSpeed(ref_elapsed_time, elapsed_time, width, height);
+  assertFaster(ref_elapsed_time, elapsed_time);
+}
+
+template <typename S, typename T, typename I>
+class CFLSubsampleTest : public ::testing::TestWithParam<S>,
+                         public CFLTestWithData<I> {
+ public:
+  virtual void SetUp() {
+    CFLTest::init(::testing::get<0>(this->GetParam()));
+    fun_420 = ::testing::get<1>(this->GetParam())(this->tx_size);
+    fun_422 = ::testing::get<2>(this->GetParam())(this->tx_size);
+    fun_444 = ::testing::get<3>(this->GetParam())(this->tx_size);
+  }
+
+ protected:
+  T fun_420;
+  T fun_422;
+  T fun_444;
+  T fun_420_ref;
+  T fun_422_ref;
+  T fun_444_ref;
+
+  void subsampleTest(T fun, T fun_ref, int sub_width, int sub_height,
+                     I (ACMRandom::*random)()) {
+    uint16_t sub_luma_pels[CFL_BUF_SQUARE];
+    uint16_t sub_luma_pels_ref[CFL_BUF_SQUARE];
+
+    for (int it = 0; it < NUM_ITERATIONS; it++) {
+      CFLTestWithData<I>::randData(random);
+      fun(this->data, CFL_BUF_LINE, sub_luma_pels);
+      fun_ref(this->data_ref, CFL_BUF_LINE, sub_luma_pels_ref);
+      assert_eq<uint16_t>(sub_luma_pels, sub_luma_pels_ref, sub_width,
+                          sub_height);
+    }
+  }
+
+  void subsampleSpeedTest(T fun, T fun_ref, I (ACMRandom::*random)()) {
+    uint16_t sub_luma_pels[CFL_BUF_SQUARE];
+    uint16_t sub_luma_pels_ref[CFL_BUF_SQUARE];
+    aom_usec_timer ref_timer;
+    aom_usec_timer timer;
+
+    CFLTestWithData<I>::randData(random);
+    aom_usec_timer_start(&ref_timer);
+    for (int k = 0; k < NUM_ITERATIONS_SPEED; k++) {
+      fun_ref(this->data_ref, CFL_BUF_LINE, sub_luma_pels);
+    }
+    aom_usec_timer_mark(&ref_timer);
+    int ref_elapsed_time = (int)aom_usec_timer_elapsed(&ref_timer);
+    aom_usec_timer_start(&timer);
+    for (int k = 0; k < NUM_ITERATIONS_SPEED; k++) {
+      fun(this->data, CFL_BUF_LINE, sub_luma_pels_ref);
+    }
+    aom_usec_timer_mark(&timer);
+    int elapsed_time = (int)aom_usec_timer_elapsed(&timer);
+    printSpeed(ref_elapsed_time, elapsed_time, this->width, this->height);
+    assertFaster(ref_elapsed_time, elapsed_time);
+  }
+};
+
+typedef cfl_subsample_lbd_fn (*get_subsample_lbd_fn)(TX_SIZE tx_size);
+typedef ::testing::tuple<TX_SIZE, get_subsample_lbd_fn, get_subsample_lbd_fn,
+                         get_subsample_lbd_fn>
+    subsample_lbd_param;
+class CFLSubsampleLBDTest
+    : public CFLSubsampleTest<subsample_lbd_param, cfl_subsample_lbd_fn,
+                              uint8_t> {
+ public:
+  virtual ~CFLSubsampleLBDTest() {}
+  virtual void SetUp() {
+    CFLSubsampleTest::SetUp();
+    fun_420_ref = cfl_get_luma_subsampling_420_lbd_c(tx_size);
+    fun_422_ref = cfl_get_luma_subsampling_422_lbd_c(tx_size);
+    fun_444_ref = cfl_get_luma_subsampling_444_lbd_c(tx_size);
+  }
+};
+
+TEST_P(CFLSubsampleLBDTest, SubsampleLBD420Test) {
+  subsampleTest(fun_420, fun_420_ref, width >> 1, height >> 1,
+                &ACMRandom::Rand8);
+}
+
+TEST_P(CFLSubsampleLBDTest, DISABLED_SubsampleLBD420SpeedTest) {
+  subsampleSpeedTest(fun_420, fun_420_ref, &ACMRandom::Rand8);
+}
+
+TEST_P(CFLSubsampleLBDTest, SubsampleLBD422Test) {
+  subsampleTest(fun_422, fun_422_ref, width >> 1, height, &ACMRandom::Rand8);
+}
+
+TEST_P(CFLSubsampleLBDTest, DISABLED_SubsampleLBD422SpeedTest) {
+  subsampleSpeedTest(fun_422, fun_422_ref, &ACMRandom::Rand8);
+}
+
+TEST_P(CFLSubsampleLBDTest, SubsampleLBD444Test) {
+  subsampleTest(fun_444, fun_444_ref, width, height, &ACMRandom::Rand8);
+}
+
+TEST_P(CFLSubsampleLBDTest, DISABLED_SubsampleLBD444SpeedTest) {
+  subsampleSpeedTest(fun_444, fun_444_ref, &ACMRandom::Rand8);
+}
+
+typedef cfl_subsample_hbd_fn (*get_subsample_hbd_fn)(TX_SIZE tx_size);
+typedef ::testing::tuple<TX_SIZE, get_subsample_hbd_fn, get_subsample_hbd_fn,
+                         get_subsample_hbd_fn>
+    subsample_hbd_param;
+class CFLSubsampleHBDTest
+    : public CFLSubsampleTest<subsample_hbd_param, cfl_subsample_hbd_fn,
+                              uint16_t> {
+ public:
+  virtual ~CFLSubsampleHBDTest() {}
+  virtual void SetUp() {
+    CFLSubsampleTest::SetUp();
+    fun_420_ref = cfl_get_luma_subsampling_420_hbd_c(tx_size);
+    fun_422_ref = cfl_get_luma_subsampling_422_hbd_c(tx_size);
+    fun_444_ref = cfl_get_luma_subsampling_444_hbd_c(tx_size);
+  }
+};
+
+TEST_P(CFLSubsampleHBDTest, SubsampleHBD420Test) {
+  subsampleTest(fun_420, fun_420_ref, width >> 1, height >> 1,
+                &ACMRandom::Rand12);
+}
+
+TEST_P(CFLSubsampleHBDTest, DISABLED_SubsampleHBD420SpeedTest) {
+  subsampleSpeedTest(fun_420, fun_420_ref, &ACMRandom::Rand12);
+}
+
+TEST_P(CFLSubsampleHBDTest, SubsampleHBD422Test) {
+  subsampleTest(fun_422, fun_422_ref, width >> 1, height, &ACMRandom::Rand12);
+}
+
+TEST_P(CFLSubsampleHBDTest, DISABLED_SubsampleHBD422SpeedTest) {
+  subsampleSpeedTest(fun_422, fun_422_ref, &ACMRandom::Rand12);
+}
+
+TEST_P(CFLSubsampleHBDTest, SubsampleHBD444Test) {
+  subsampleTest(fun_444, fun_444_ref, width, height, &ACMRandom::Rand12);
+}
+
+TEST_P(CFLSubsampleHBDTest, DISABLED_SubsampleHBD444SpeedTest) {
+  subsampleSpeedTest(fun_444, fun_444_ref, &ACMRandom::Rand12);
+}
+
+typedef cfl_predict_lbd_fn (*get_predict_fn)(TX_SIZE tx_size);
+typedef ::testing::tuple<TX_SIZE, get_predict_fn> predict_param;
+class CFLPredictTest : public ::testing::TestWithParam<predict_param>,
+                       public CFLTestWithAlignedData<uint8_t> {
+ public:
+  virtual void SetUp() {
+    CFLTest::init(::testing::get<0>(this->GetParam()));
+    predict = ::testing::get<1>(this->GetParam())(tx_size);
+    predict_ref = get_predict_lbd_fn_c(tx_size);
+  }
+  virtual ~CFLPredictTest() {}
+
+ protected:
+  cfl_predict_lbd_fn predict;
+  cfl_predict_lbd_fn predict_ref;
+};
+
+TEST_P(CFLPredictTest, PredictTest) {
+  for (int it = 0; it < NUM_ITERATIONS; it++) {
+    randData(8);
+    predict(sub_luma_pels, chroma_pels, CFL_BUF_LINE, alpha_q3);
+    predict_ref(sub_luma_pels_ref, chroma_pels_ref, CFL_BUF_LINE, alpha_q3);
+    assert_eq<uint8_t>(chroma_pels, chroma_pels_ref, width, height);
+  }
+}
+TEST_P(CFLPredictTest, DISABLED_PredictSpeedTest) {
+  aom_usec_timer ref_timer;
+  aom_usec_timer timer;
+  randData(8);
+  aom_usec_timer_start(&ref_timer);
+  for (int k = 0; k < NUM_ITERATIONS_SPEED; k++) {
+    predict_ref(sub_luma_pels_ref, chroma_pels_ref, CFL_BUF_LINE, alpha_q3);
+  }
+  aom_usec_timer_mark(&ref_timer);
+  int ref_elapsed_time = (int)aom_usec_timer_elapsed(&ref_timer);
+
+  aom_usec_timer_start(&timer);
+  for (int k = 0; k < NUM_ITERATIONS_SPEED; k++) {
+    predict(sub_luma_pels, chroma_pels, CFL_BUF_LINE, alpha_q3);
+  }
+  aom_usec_timer_mark(&timer);
+  int elapsed_time = (int)aom_usec_timer_elapsed(&timer);
+  printSpeed(ref_elapsed_time, elapsed_time, width, height);
+  assertFaster(ref_elapsed_time, elapsed_time);
+}
+
+typedef cfl_predict_hbd_fn (*get_predict_fn_hbd)(TX_SIZE tx_size);
+typedef ::testing::tuple<TX_SIZE, get_predict_fn_hbd> predict_param_hbd;
+class CFLPredictHBDTest : public ::testing::TestWithParam<predict_param_hbd>,
+                          public CFLTestWithAlignedData<uint16_t> {
+ public:
+  virtual void SetUp() {
+    CFLTest::init(::testing::get<0>(this->GetParam()));
+    predict = ::testing::get<1>(this->GetParam())(tx_size);
+    predict_ref = get_predict_hbd_fn_c(tx_size);
+  }
+  virtual ~CFLPredictHBDTest() {}
+
+ protected:
+  cfl_predict_hbd_fn predict;
+  cfl_predict_hbd_fn predict_ref;
+};
+
+TEST_P(CFLPredictHBDTest, PredictHBDTest) {
+  int bd = 12;
+  for (int it = 0; it < NUM_ITERATIONS; it++) {
+    randData(bd);
+    predict(sub_luma_pels, chroma_pels, CFL_BUF_LINE, alpha_q3, bd);
+    predict_ref(sub_luma_pels_ref, chroma_pels_ref, CFL_BUF_LINE, alpha_q3, bd);
+    assert_eq<uint16_t>(chroma_pels, chroma_pels_ref, width, height);
+  }
+}
+TEST_P(CFLPredictHBDTest, DISABLED_PredictHBDSpeedTest) {
+  aom_usec_timer ref_timer;
+  aom_usec_timer timer;
+  const int bd = 12;
+  randData(bd);
+  aom_usec_timer_start(&ref_timer);
+  for (int k = 0; k < NUM_ITERATIONS_SPEED; k++) {
+    predict_ref(sub_luma_pels_ref, chroma_pels_ref, CFL_BUF_LINE, alpha_q3, bd);
+  }
+  aom_usec_timer_mark(&ref_timer);
+  int ref_elapsed_time = (int)aom_usec_timer_elapsed(&ref_timer);
+
+  aom_usec_timer_start(&timer);
+  for (int k = 0; k < NUM_ITERATIONS_SPEED; k++) {
+    predict(sub_luma_pels, chroma_pels, CFL_BUF_LINE, alpha_q3, bd);
+  }
+  aom_usec_timer_mark(&timer);
+  int elapsed_time = (int)aom_usec_timer_elapsed(&timer);
+  printSpeed(ref_elapsed_time, elapsed_time, width, height);
+  assertFaster(ref_elapsed_time, elapsed_time);
+}
+
+#if HAVE_SSE2
+const sub_avg_param sub_avg_sizes_sse2[] = { ALL_CFL_TX_SIZES(
+    get_subtract_average_fn_sse2) };
+
+INSTANTIATE_TEST_CASE_P(SSE2, CFLSubAvgTest,
+                        ::testing::ValuesIn(sub_avg_sizes_sse2));
+
+#endif
+
+#if HAVE_SSSE3
+const subsample_lbd_param subsample_lbd_sizes_ssse3[] = {
+  ALL_CFL_TX_SIZES_SUBSAMPLE(cfl_get_luma_subsampling_420_lbd_ssse3,
+                             cfl_get_luma_subsampling_422_lbd_ssse3,
+                             cfl_get_luma_subsampling_444_lbd_ssse3)
+};
+
+const subsample_hbd_param subsample_hbd_sizes_ssse3[] = {
+  ALL_CFL_TX_SIZES_SUBSAMPLE(cfl_get_luma_subsampling_420_hbd_ssse3,
+                             cfl_get_luma_subsampling_422_hbd_ssse3,
+                             cfl_get_luma_subsampling_444_hbd_ssse3)
+};
+
+const predict_param predict_sizes_ssse3[] = { ALL_CFL_TX_SIZES(
+    get_predict_lbd_fn_ssse3) };
+
+const predict_param_hbd predict_sizes_hbd_ssse3[] = { ALL_CFL_TX_SIZES(
+    get_predict_hbd_fn_ssse3) };
+
+INSTANTIATE_TEST_CASE_P(SSSE3, CFLSubsampleLBDTest,
+                        ::testing::ValuesIn(subsample_lbd_sizes_ssse3));
+
+INSTANTIATE_TEST_CASE_P(SSSE3, CFLSubsampleHBDTest,
+                        ::testing::ValuesIn(subsample_hbd_sizes_ssse3));
+
+INSTANTIATE_TEST_CASE_P(SSSE3, CFLPredictTest,
+                        ::testing::ValuesIn(predict_sizes_ssse3));
+
+INSTANTIATE_TEST_CASE_P(SSSE3, CFLPredictHBDTest,
+                        ::testing::ValuesIn(predict_sizes_hbd_ssse3));
+#endif
+
+#if HAVE_AVX2
+const sub_avg_param sub_avg_sizes_avx2[] = { ALL_CFL_TX_SIZES(
+    get_subtract_average_fn_avx2) };
+
+const subsample_lbd_param subsample_lbd_sizes_avx2[] = {
+  ALL_CFL_TX_SIZES_SUBSAMPLE(cfl_get_luma_subsampling_420_lbd_avx2,
+                             cfl_get_luma_subsampling_422_lbd_avx2,
+                             cfl_get_luma_subsampling_444_lbd_avx2)
+};
+
+const subsample_hbd_param subsample_hbd_sizes_avx2[] = {
+  ALL_CFL_TX_SIZES_SUBSAMPLE(cfl_get_luma_subsampling_420_hbd_avx2,
+                             cfl_get_luma_subsampling_422_hbd_avx2,
+                             cfl_get_luma_subsampling_444_hbd_avx2)
+};
+
+const predict_param predict_sizes_avx2[] = { ALL_CFL_TX_SIZES(
+    get_predict_lbd_fn_avx2) };
+
+const predict_param_hbd predict_sizes_hbd_avx2[] = { ALL_CFL_TX_SIZES(
+    get_predict_hbd_fn_avx2) };
+
+INSTANTIATE_TEST_CASE_P(AVX2, CFLSubAvgTest,
+                        ::testing::ValuesIn(sub_avg_sizes_avx2));
+
+INSTANTIATE_TEST_CASE_P(AVX2, CFLSubsampleLBDTest,
+                        ::testing::ValuesIn(subsample_lbd_sizes_avx2));
+
+INSTANTIATE_TEST_CASE_P(AVX2, CFLSubsampleHBDTest,
+                        ::testing::ValuesIn(subsample_hbd_sizes_avx2));
+
+INSTANTIATE_TEST_CASE_P(AVX2, CFLPredictTest,
+                        ::testing::ValuesIn(predict_sizes_avx2));
+
+INSTANTIATE_TEST_CASE_P(AVX2, CFLPredictHBDTest,
+                        ::testing::ValuesIn(predict_sizes_hbd_avx2));
+#endif
+
+#if HAVE_NEON
+
+const sub_avg_param sub_avg_sizes_neon[] = { ALL_CFL_TX_SIZES(
+    get_subtract_average_fn_neon) };
+
+const subsample_lbd_param subsample_lbd_sizes_neon[] = {
+  ALL_CFL_TX_SIZES_SUBSAMPLE(cfl_get_luma_subsampling_420_lbd_neon,
+                             cfl_get_luma_subsampling_422_lbd_neon,
+                             cfl_get_luma_subsampling_444_lbd_neon)
+};
+
+const subsample_hbd_param subsample_hbd_sizes_neon[] = {
+  ALL_CFL_TX_SIZES_SUBSAMPLE(cfl_get_luma_subsampling_420_hbd_neon,
+                             cfl_get_luma_subsampling_422_hbd_neon,
+                             cfl_get_luma_subsampling_444_hbd_neon)
+};
+
+const predict_param predict_sizes_neon[] = { ALL_CFL_TX_SIZES(
+    get_predict_lbd_fn_neon) };
+
+const predict_param_hbd predict_sizes_hbd_neon[] = { ALL_CFL_TX_SIZES(
+    get_predict_hbd_fn_neon) };
+
+INSTANTIATE_TEST_CASE_P(NEON, CFLSubAvgTest,
+                        ::testing::ValuesIn(sub_avg_sizes_neon));
+
+INSTANTIATE_TEST_CASE_P(NEON, CFLSubsampleLBDTest,
+                        ::testing::ValuesIn(subsample_lbd_sizes_neon));
+
+INSTANTIATE_TEST_CASE_P(NEON, CFLSubsampleHBDTest,
+                        ::testing::ValuesIn(subsample_hbd_sizes_neon));
+
+INSTANTIATE_TEST_CASE_P(NEON, CFLPredictTest,
+                        ::testing::ValuesIn(predict_sizes_neon));
+
+INSTANTIATE_TEST_CASE_P(NEON, CFLPredictHBDTest,
+                        ::testing::ValuesIn(predict_sizes_hbd_neon));
+#endif
+
+#if HAVE_VSX
+const sub_avg_param sub_avg_sizes_vsx[] = { ALL_CFL_TX_SIZES(
+    get_subtract_average_fn_vsx) };
+
+INSTANTIATE_TEST_CASE_P(VSX, CFLSubAvgTest,
+                        ::testing::ValuesIn(sub_avg_sizes_vsx));
+#endif
+}  // namespace
diff --git a/third_party/aom/test/clear_system_state.h b/third_party/aom/test/clear_system_state.h
new file mode 100644
index 0000000000..d38ff5dd51
--- /dev/null
+++ b/third_party/aom/test/clear_system_state.h
@@ -0,0 +1,31 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_CLEAR_SYSTEM_STATE_H_
+#define AOM_TEST_CLEAR_SYSTEM_STATE_H_
+
+#include "config/aom_config.h"
+
+#if ARCH_X86 || ARCH_X86_64
+#include "aom_ports/x86.h"
+#endif
+
+namespace libaom_test {
+
+// Reset system to a known state. This function should be used for all non-API
+// test cases.
+inline void ClearSystemState() {
+#if ARCH_X86 || ARCH_X86_64
+  aom_reset_mmx_state();
+#endif
+}
+
+}  // namespace libaom_test
+#endif  // AOM_TEST_CLEAR_SYSTEM_STATE_H_
diff --git a/third_party/aom/test/codec_factory.h b/third_party/aom/test/codec_factory.h
new file mode 100644
index 0000000000..dd99110ee6
--- /dev/null
+++ b/third_party/aom/test/codec_factory.h
@@ -0,0 +1,170 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_CODEC_FACTORY_H_
+#define AOM_TEST_CODEC_FACTORY_H_
+
+#include "config/aom_config.h"
+
+#include "aom/aom_decoder.h"
+#include "aom/aom_encoder.h"
+#if CONFIG_AV1_ENCODER
+#include "aom/aomcx.h"
+#endif
+#if CONFIG_AV1_DECODER
+#include "aom/aomdx.h"
+#endif
+
+#include "test/decode_test_driver.h"
+#include "test/encode_test_driver.h"
+namespace libaom_test {
+
+const int kCodecFactoryParam = 0;
+
+class CodecFactory {
+ public:
+  CodecFactory() {}
+
+  virtual ~CodecFactory() {}
+
+  virtual Decoder *CreateDecoder(aom_codec_dec_cfg_t cfg) const = 0;
+
+  virtual Decoder *CreateDecoder(aom_codec_dec_cfg_t cfg,
+                                 const aom_codec_flags_t flags) const = 0;
+
+  virtual Encoder *CreateEncoder(aom_codec_enc_cfg_t cfg,
+                                 const unsigned long init_flags,
+                                 TwopassStatsStore *stats) const = 0;
+
+  virtual aom_codec_err_t DefaultEncoderConfig(aom_codec_enc_cfg_t *cfg,
+                                               int usage) const = 0;
+};
+
+/* Provide CodecTestWith<n>Params classes for a variable number of parameters
+ * to avoid having to include a pointer to the CodecFactory in every test
+ * definition.
+ */
+template <class T1>
+class CodecTestWithParam
+    : public ::testing::TestWithParam<
+          ::testing::tuple<const libaom_test::CodecFactory *, T1> > {};
+
+template <class T1, class T2>
+class CodecTestWith2Params
+    : public ::testing::TestWithParam<
+          ::testing::tuple<const libaom_test::CodecFactory *, T1, T2> > {};
+
+template <class T1, class T2, class T3>
+class CodecTestWith3Params
+    : public ::testing::TestWithParam<
+          ::testing::tuple<const libaom_test::CodecFactory *, T1, T2, T3> > {};
+
+template <class T1, class T2, class T3, class T4>
+class CodecTestWith4Params
+    : public ::testing::TestWithParam< ::testing::tuple<
+          const libaom_test::CodecFactory *, T1, T2, T3, T4> > {};
+
+template <class T1, class T2, class T3, class T4, class T5>
+class CodecTestWith5Params
+    : public ::testing::TestWithParam< ::testing::tuple<
+          const libaom_test::CodecFactory *, T1, T2, T3, T4, T5> > {};
+
+/*
+ * AV1 Codec Definitions
+ */
+class AV1Decoder : public Decoder {
+ public:
+  explicit AV1Decoder(aom_codec_dec_cfg_t cfg) : Decoder(cfg) {}
+
+  AV1Decoder(aom_codec_dec_cfg_t cfg, const aom_codec_flags_t flag)
+      : Decoder(cfg, flag) {}
+
+ protected:
+  virtual aom_codec_iface_t *CodecInterface() const {
+#if CONFIG_AV1_DECODER
+    return aom_codec_av1_dx();
+#else
+    return NULL;
+#endif
+  }
+};
+
+class AV1Encoder : public Encoder {
+ public:
+  AV1Encoder(aom_codec_enc_cfg_t cfg, const uint32_t init_flags,
+             TwopassStatsStore *stats)
+      : Encoder(cfg, init_flags, stats) {}
+
+ protected:
+  virtual aom_codec_iface_t *CodecInterface() const {
+#if CONFIG_AV1_ENCODER
+    return aom_codec_av1_cx();
+#else
+    return NULL;
+#endif
+  }
+};
+
+class AV1CodecFactory : public CodecFactory {
+ public:
+  AV1CodecFactory() : CodecFactory() {}
+
+  virtual Decoder *CreateDecoder(aom_codec_dec_cfg_t cfg) const {
+    return CreateDecoder(cfg, 0);
+  }
+
+  virtual Decoder *CreateDecoder(aom_codec_dec_cfg_t cfg,
+                                 const aom_codec_flags_t flags) const {
+#if CONFIG_AV1_DECODER
+    return new AV1Decoder(cfg, flags);
+#else
+    (void)cfg;
+    (void)flags;
+    return NULL;
+#endif
+  }
+
+  virtual Encoder *CreateEncoder(aom_codec_enc_cfg_t cfg,
+                                 const unsigned long init_flags,
+                                 TwopassStatsStore *stats) const {
+#if CONFIG_AV1_ENCODER
+    return new AV1Encoder(cfg, init_flags, stats);
+#else
+    (void)cfg;
+    (void)init_flags;
+    (void)stats;
+    return NULL;
+#endif
+  }
+
+  virtual aom_codec_err_t DefaultEncoderConfig(aom_codec_enc_cfg_t *cfg,
+                                               int usage) const {
+#if CONFIG_AV1_ENCODER
+    return aom_codec_enc_config_default(aom_codec_av1_cx(), cfg, usage);
+#else
+    (void)cfg;
+    (void)usage;
+    return AOM_CODEC_INCAPABLE;
+#endif
+  }
+};
+
+const libaom_test::AV1CodecFactory kAV1;
+
+#define AV1_INSTANTIATE_TEST_CASE(test, ...)                                \
+  INSTANTIATE_TEST_CASE_P(                                                  \
+      AV1, test,                                                            \
+      ::testing::Combine(                                                   \
+          ::testing::Values(static_cast<const libaom_test::CodecFactory *>( \
+              &libaom_test::kAV1)),                                         \
+          __VA_ARGS__))
+
+}  // namespace libaom_test
+#endif  // AOM_TEST_CODEC_FACTORY_H_
diff --git a/third_party/aom/test/coding_path_sync.cc b/third_party/aom/test/coding_path_sync.cc
new file mode 100644
index 0000000000..6735236ccd
--- /dev/null
+++ b/third_party/aom/test/coding_path_sync.cc
@@ -0,0 +1,205 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <vector>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+
+#include "config/aom_config.h"
+
+#include "aom_ports/mem.h"  // ROUND_POWER_OF_TWO
+#include "aom/aomcx.h"
+#include "aom/aomdx.h"
+#include "aom/aom_encoder.h"
+#include "aom/aom_decoder.h"
+
+using libaom_test::ACMRandom;
+namespace {
+
+class CompressedSource {
+ public:
+  explicit CompressedSource(int seed) : rnd_(seed), frame_count_(0) {
+    aom_codec_iface_t *algo = aom_codec_av1_cx();
+
+    aom_codec_enc_cfg_t cfg;
+    aom_codec_enc_config_default(algo, &cfg, 0);
+
+    // force the quantizer, to reduce the sensitivity on encoding choices.
+    // e.g, we don't want this test to break when the rate control is modified.
+    {
+      const int max_q = cfg.rc_max_quantizer;
+      const int min_q = cfg.rc_min_quantizer;
+      const int q = rnd_.PseudoUniform(max_q - min_q + 1) + min_q;
+
+      cfg.rc_end_usage = AOM_Q;
+      cfg.rc_max_quantizer = q;
+      cfg.rc_min_quantizer = q;
+    }
+
+    // choose the picture size
+    {
+      width_ = rnd_.PseudoUniform(kWidth - 8) + 8;
+      height_ = rnd_.PseudoUniform(kHeight - 8) + 8;
+    }
+
+    // choose the chroma subsampling
+    {
+      const aom_img_fmt_t fmts[] = {
+        AOM_IMG_FMT_I420,
+        AOM_IMG_FMT_I422,
+        AOM_IMG_FMT_I444,
+      };
+
+      format_ = fmts[rnd_.PseudoUniform(NELEMENTS(fmts))];
+    }
+
+    cfg.g_w = width_;
+    cfg.g_h = height_;
+    cfg.g_lag_in_frames = 0;
+    if (format_ == AOM_IMG_FMT_I420)
+      cfg.g_profile = 0;
+    else if (format_ == AOM_IMG_FMT_I444)
+      cfg.g_profile = 1;
+    else if (format_ == AOM_IMG_FMT_I422)
+      cfg.g_profile = 2;
+
+    aom_codec_enc_init(&enc_, algo, &cfg, 0);
+  }
+
+  ~CompressedSource() { aom_codec_destroy(&enc_); }
+
+  const aom_codec_cx_pkt_t *ReadFrame() {
+    uint8_t buf[kWidth * kHeight * 3] = { 0 };
+
+    // render regular pattern
+    const int period = rnd_.Rand8() % 32 + 1;
+    const int phase = rnd_.Rand8() % period;
+
+    const int val_a = rnd_.Rand8();
+    const int val_b = rnd_.Rand8();
+
+    for (int i = 0; i < (int)sizeof buf; ++i)
+      buf[i] = (i + phase) % period < period / 2 ? val_a : val_b;
+
+    aom_image_t img;
+    aom_img_wrap(&img, format_, width_, height_, 0, buf);
+    aom_codec_encode(&enc_, &img, frame_count_++, 1, 0);
+
+    aom_codec_iter_t iter = NULL;
+
+    const aom_codec_cx_pkt_t *pkt = NULL;
+
+    do {
+      pkt = aom_codec_get_cx_data(&enc_, &iter);
+    } while (pkt && pkt->kind != AOM_CODEC_CX_FRAME_PKT);
+
+    return pkt;
+  }
+
+ private:
+  static const int kWidth = 128;
+  static const int kHeight = 128;
+
+  ACMRandom rnd_;
+  aom_img_fmt_t format_;
+  aom_codec_ctx_t enc_;
+  int frame_count_;
+  int width_, height_;
+};
+
+// lowers an aom_image_t to a easily comparable/printable form
+std::vector<int16_t> Serialize(const aom_image_t *img) {
+  std::vector<int16_t> bytes;
+  bytes.reserve(img->d_w * img->d_h * 3);
+  for (int plane = 0; plane < 3; ++plane) {
+    const int w = aom_img_plane_width(img, plane);
+    const int h = aom_img_plane_height(img, plane);
+
+    for (int r = 0; r < h; ++r) {
+      for (int c = 0; c < w; ++c) {
+        unsigned char *row = img->planes[plane] + r * img->stride[plane];
+        if (img->fmt & AOM_IMG_FMT_HIGHBITDEPTH)
+          bytes.push_back(row[c * 2]);
+        else
+          bytes.push_back(row[c]);
+      }
+    }
+  }
+
+  return bytes;
+}
+
+class Decoder {
+ public:
+  explicit Decoder(int allowLowbitdepth) {
+    aom_codec_iface_t *algo = aom_codec_av1_dx();
+
+    aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+    cfg.allow_lowbitdepth = allowLowbitdepth;
+
+    aom_codec_dec_init(&dec_, algo, &cfg, 0);
+  }
+
+  ~Decoder() { aom_codec_destroy(&dec_); }
+
+  std::vector<int16_t> decode(const aom_codec_cx_pkt_t *pkt) {
+    aom_codec_decode(&dec_, static_cast<uint8_t *>(pkt->data.frame.buf),
+                     pkt->data.frame.sz, NULL);
+
+    aom_codec_iter_t iter = NULL;
+    return Serialize(aom_codec_get_frame(&dec_, &iter));
+  }
+
+ private:
+  aom_codec_ctx_t dec_;
+};
+
+// Try to reveal a mismatch between LBD and HBD coding paths.
+TEST(CodingPathSync, SearchForHbdLbdMismatch) {
+  const int count_tests = 10;
+  for (int i = 0; i < count_tests; ++i) {
+    Decoder dec_hbd(0);
+    Decoder dec_lbd(1);
+
+    CompressedSource enc(i);
+
+    for (int k = 0; k < 3; ++k) {
+      const aom_codec_cx_pkt_t *frame = enc.ReadFrame();
+
+      std::vector<int16_t> lbd_yuv = dec_lbd.decode(frame);
+      std::vector<int16_t> hbd_yuv = dec_hbd.decode(frame);
+
+      ASSERT_EQ(lbd_yuv, hbd_yuv);
+    }
+  }
+}
+
+TEST(CodingPathSyncLarge, SearchForHbdLbdMismatchLarge) {
+  const int count_tests = 100;
+  const int seed = 1234;
+  for (int i = 0; i < count_tests; ++i) {
+    Decoder dec_hbd(0);
+    Decoder dec_lbd(1);
+
+    CompressedSource enc(seed + i);
+
+    for (int k = 0; k < 5; ++k) {
+      const aom_codec_cx_pkt_t *frame = enc.ReadFrame();
+
+      std::vector<int16_t> lbd_yuv = dec_lbd.decode(frame);
+      std::vector<int16_t> hbd_yuv = dec_hbd.decode(frame);
+
+      ASSERT_EQ(lbd_yuv, hbd_yuv);
+    }
+  }
+}
+
+}  // namespace
diff --git a/third_party/aom/test/comp_avg_pred_test.cc b/third_party/aom/test/comp_avg_pred_test.cc
new file mode 100644
index 0000000000..9ad8973f08
--- /dev/null
+++ b/third_party/aom/test/comp_avg_pred_test.cc
@@ -0,0 +1,72 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "test/comp_avg_pred_test.h"
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+using libaom_test::AV1JNTCOMPAVG::AV1HighBDJNTCOMPAVGTest;
+using libaom_test::AV1JNTCOMPAVG::AV1HighBDJNTCOMPAVGUPSAMPLEDTest;
+using libaom_test::AV1JNTCOMPAVG::AV1JNTCOMPAVGTest;
+using libaom_test::AV1JNTCOMPAVG::AV1JNTCOMPAVGUPSAMPLEDTest;
+
+namespace {
+
+TEST_P(AV1JNTCOMPAVGTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(0)); }
+
+TEST_P(AV1JNTCOMPAVGTest, CheckOutput) { RunCheckOutput(GET_PARAM(0)); }
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, AV1JNTCOMPAVGTest,
+    libaom_test::AV1JNTCOMPAVG::BuildParams(aom_jnt_comp_avg_pred_ssse3));
+#endif
+
+TEST_P(AV1JNTCOMPAVGUPSAMPLEDTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(0));
+}
+
+TEST_P(AV1JNTCOMPAVGUPSAMPLEDTest, CheckOutput) {
+  RunCheckOutput(GET_PARAM(0));
+}
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(SSSE3, AV1JNTCOMPAVGUPSAMPLEDTest,
+                        libaom_test::AV1JNTCOMPAVG::BuildParams(
+                            aom_jnt_comp_avg_upsampled_pred_ssse3));
+#endif
+
+TEST_P(AV1HighBDJNTCOMPAVGTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(1)); }
+
+TEST_P(AV1HighBDJNTCOMPAVGTest, CheckOutput) { RunCheckOutput(GET_PARAM(1)); }
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, AV1HighBDJNTCOMPAVGTest,
+                        libaom_test::AV1JNTCOMPAVG::BuildParams(
+                            aom_highbd_jnt_comp_avg_pred_sse2, 1));
+#endif
+
+TEST_P(AV1HighBDJNTCOMPAVGUPSAMPLEDTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(1));
+}
+
+TEST_P(AV1HighBDJNTCOMPAVGUPSAMPLEDTest, CheckOutput) {
+  RunCheckOutput(GET_PARAM(1));
+}
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, AV1HighBDJNTCOMPAVGUPSAMPLEDTest,
+                        libaom_test::AV1JNTCOMPAVG::BuildParams(
+                            aom_highbd_jnt_comp_avg_upsampled_pred_sse2));
+#endif
+
+}  // namespace
diff --git a/third_party/aom/test/comp_avg_pred_test.h b/third_party/aom/test/comp_avg_pred_test.h
new file mode 100644
index 0000000000..9661dd9f54
--- /dev/null
+++ b/third_party/aom/test/comp_avg_pred_test.h
@@ -0,0 +1,555 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_COMP_AVG_PRED_TEST_H_
+#define AOM_TEST_COMP_AVG_PRED_TEST_H_
+
+#include "config/aom_dsp_rtcd.h"
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/util.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "av1/common/common_data.h"
+#include "aom_ports/aom_timer.h"
+
+namespace libaom_test {
+const int kMaxSize = 128 + 32;  // padding
+
+namespace AV1JNTCOMPAVG {
+
+typedef void (*jntcompavg_func)(uint8_t *comp_pred, const uint8_t *pred,
+                                int width, int height, const uint8_t *ref,
+                                int ref_stride,
+                                const JNT_COMP_PARAMS *jcp_param);
+
+typedef void (*jntcompavgupsampled_func)(
+    MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
+    const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
+    int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
+    int ref_stride, const JNT_COMP_PARAMS *jcp_param, int subpel_search);
+
+typedef void (*highbdjntcompavgupsampled_func)(
+    MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
+    const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width,
+    int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8,
+    int ref_stride, int bd, const JNT_COMP_PARAMS *jcp_param,
+    int subpel_search);
+
+typedef ::testing::tuple<jntcompavg_func, BLOCK_SIZE> JNTCOMPAVGParam;
+
+typedef ::testing::tuple<jntcompavgupsampled_func, BLOCK_SIZE>
+    JNTCOMPAVGUPSAMPLEDParam;
+
+typedef ::testing::tuple<int, jntcompavg_func, BLOCK_SIZE>
+    HighbdJNTCOMPAVGParam;
+
+typedef ::testing::tuple<int, highbdjntcompavgupsampled_func, BLOCK_SIZE>
+    HighbdJNTCOMPAVGUPSAMPLEDParam;
+
+::testing::internal::ParamGenerator<JNTCOMPAVGParam> BuildParams(
+    jntcompavg_func filter) {
+  return ::testing::Combine(::testing::Values(filter),
+                            ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
+}
+
+::testing::internal::ParamGenerator<JNTCOMPAVGUPSAMPLEDParam> BuildParams(
+    jntcompavgupsampled_func filter) {
+  return ::testing::Combine(::testing::Values(filter),
+                            ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
+}
+
+::testing::internal::ParamGenerator<HighbdJNTCOMPAVGParam> BuildParams(
+    jntcompavg_func filter, int is_hbd) {
+  (void)is_hbd;
+  return ::testing::Combine(::testing::Range(8, 13, 2),
+                            ::testing::Values(filter),
+                            ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
+}
+
+::testing::internal::ParamGenerator<HighbdJNTCOMPAVGUPSAMPLEDParam> BuildParams(
+    highbdjntcompavgupsampled_func filter) {
+  return ::testing::Combine(::testing::Range(8, 13, 2),
+                            ::testing::Values(filter),
+                            ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
+}
+
+class AV1JNTCOMPAVGTest : public ::testing::TestWithParam<JNTCOMPAVGParam> {
+ public:
+  ~AV1JNTCOMPAVGTest() {}
+  void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+  void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void RunCheckOutput(jntcompavg_func test_impl) {
+    const int w = kMaxSize, h = kMaxSize;
+    const int block_idx = GET_PARAM(1);
+
+    uint8_t pred8[kMaxSize * kMaxSize];
+    uint8_t ref8[kMaxSize * kMaxSize];
+    uint8_t output[kMaxSize * kMaxSize];
+    uint8_t output2[kMaxSize * kMaxSize];
+
+    for (int i = 0; i < h; ++i)
+      for (int j = 0; j < w; ++j) {
+        pred8[i * w + j] = rnd_.Rand8();
+        ref8[i * w + j] = rnd_.Rand8();
+      }
+    const int in_w = block_size_wide[block_idx];
+    const int in_h = block_size_high[block_idx];
+
+    JNT_COMP_PARAMS jnt_comp_params;
+    jnt_comp_params.use_jnt_comp_avg = 1;
+
+    for (int ii = 0; ii < 2; ii++) {
+      for (int jj = 0; jj < 4; jj++) {
+        jnt_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+        jnt_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+
+        const int offset_r = 3 + rnd_.PseudoUniform(h - in_h - 7);
+        const int offset_c = 3 + rnd_.PseudoUniform(w - in_w - 7);
+        aom_jnt_comp_avg_pred_c(output, pred8 + offset_r * w + offset_c, in_w,
+                                in_h, ref8 + offset_r * w + offset_c, in_w,
+                                &jnt_comp_params);
+        test_impl(output2, pred8 + offset_r * w + offset_c, in_w, in_h,
+                  ref8 + offset_r * w + offset_c, in_w, &jnt_comp_params);
+
+        for (int i = 0; i < in_h; ++i) {
+          for (int j = 0; j < in_w; ++j) {
+            int idx = i * in_w + j;
+            ASSERT_EQ(output[idx], output2[idx])
+                << "Mismatch at unit tests for AV1JNTCOMPAVGTest\n"
+                << in_w << "x" << in_h << " Pixel mismatch at index " << idx
+                << " = (" << i << ", " << j << ")";
+          }
+        }
+      }
+    }
+  }
+  void RunSpeedTest(jntcompavg_func test_impl) {
+    const int w = kMaxSize, h = kMaxSize;
+    const int block_idx = GET_PARAM(1);
+
+    uint8_t pred8[kMaxSize * kMaxSize];
+    uint8_t ref8[kMaxSize * kMaxSize];
+    uint8_t output[kMaxSize * kMaxSize];
+    uint8_t output2[kMaxSize * kMaxSize];
+
+    for (int i = 0; i < h; ++i)
+      for (int j = 0; j < w; ++j) {
+        pred8[i * w + j] = rnd_.Rand8();
+        ref8[i * w + j] = rnd_.Rand8();
+      }
+    const int in_w = block_size_wide[block_idx];
+    const int in_h = block_size_high[block_idx];
+
+    JNT_COMP_PARAMS jnt_comp_params;
+    jnt_comp_params.use_jnt_comp_avg = 1;
+
+    jnt_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
+    jnt_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
+
+    const int num_loops = 1000000000 / (in_w + in_h);
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+
+    for (int i = 0; i < num_loops; ++i)
+      aom_jnt_comp_avg_pred_c(output, pred8, in_w, in_h, ref8, in_w,
+                              &jnt_comp_params);
+
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+    printf("jntcompavg c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+           1000.0 * elapsed_time / num_loops);
+
+    aom_usec_timer timer1;
+    aom_usec_timer_start(&timer1);
+
+    for (int i = 0; i < num_loops; ++i)
+      test_impl(output2, pred8, in_w, in_h, ref8, in_w, &jnt_comp_params);
+
+    aom_usec_timer_mark(&timer1);
+    const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
+    printf("jntcompavg test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+           1000.0 * elapsed_time1 / num_loops);
+  }
+
+  libaom_test::ACMRandom rnd_;
+};  // class AV1JNTCOMPAVGTest
+
+class AV1JNTCOMPAVGUPSAMPLEDTest
+    : public ::testing::TestWithParam<JNTCOMPAVGUPSAMPLEDParam> {
+ public:
+  ~AV1JNTCOMPAVGUPSAMPLEDTest() {}
+  void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+  void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void RunCheckOutput(jntcompavgupsampled_func test_impl) {
+    const int w = kMaxSize, h = kMaxSize;
+    const int block_idx = GET_PARAM(1);
+
+    uint8_t pred8[kMaxSize * kMaxSize];
+    uint8_t ref8[kMaxSize * kMaxSize];
+    DECLARE_ALIGNED(16, uint8_t, output[MAX_SB_SQUARE]);
+    DECLARE_ALIGNED(16, uint8_t, output2[MAX_SB_SQUARE]);
+
+    for (int i = 0; i < h; ++i)
+      for (int j = 0; j < w; ++j) {
+        pred8[i * w + j] = rnd_.Rand8();
+        ref8[i * w + j] = rnd_.Rand8();
+      }
+    const int in_w = block_size_wide[block_idx];
+    const int in_h = block_size_high[block_idx];
+
+    JNT_COMP_PARAMS jnt_comp_params;
+    jnt_comp_params.use_jnt_comp_avg = 1;
+    int sub_x_q3, sub_y_q3;
+    int subpel_search;
+    for (subpel_search = 1; subpel_search <= 2; ++subpel_search) {
+      for (sub_x_q3 = 0; sub_x_q3 < 8; ++sub_x_q3) {
+        for (sub_y_q3 = 0; sub_y_q3 < 8; ++sub_y_q3) {
+          for (int ii = 0; ii < 2; ii++) {
+            for (int jj = 0; jj < 4; jj++) {
+              jnt_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+              jnt_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+
+              const int offset_r = 3 + rnd_.PseudoUniform(h - in_h - 7);
+              const int offset_c = 3 + rnd_.PseudoUniform(w - in_w - 7);
+
+              aom_jnt_comp_avg_upsampled_pred_c(
+                  NULL, NULL, 0, 0, NULL, output,
+                  pred8 + offset_r * w + offset_c, in_w, in_h, sub_x_q3,
+                  sub_y_q3, ref8 + offset_r * w + offset_c, in_w,
+                  &jnt_comp_params, subpel_search);
+              test_impl(NULL, NULL, 0, 0, NULL, output2,
+                        pred8 + offset_r * w + offset_c, in_w, in_h, sub_x_q3,
+                        sub_y_q3, ref8 + offset_r * w + offset_c, in_w,
+                        &jnt_comp_params, subpel_search);
+
+              for (int i = 0; i < in_h; ++i) {
+                for (int j = 0; j < in_w; ++j) {
+                  int idx = i * in_w + j;
+                  ASSERT_EQ(output[idx], output2[idx])
+                      << "Mismatch at unit tests for "
+                         "AV1JNTCOMPAVGUPSAMPLEDTest\n"
+                      << in_w << "x" << in_h << " Pixel mismatch at index "
+                      << idx << " = (" << i << ", " << j
+                      << "), sub pixel offset = (" << sub_y_q3 << ", "
+                      << sub_x_q3 << ")";
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  void RunSpeedTest(jntcompavgupsampled_func test_impl) {
+    const int w = kMaxSize, h = kMaxSize;
+    const int block_idx = GET_PARAM(1);
+
+    uint8_t pred8[kMaxSize * kMaxSize];
+    uint8_t ref8[kMaxSize * kMaxSize];
+    DECLARE_ALIGNED(16, uint8_t, output[MAX_SB_SQUARE]);
+    DECLARE_ALIGNED(16, uint8_t, output2[MAX_SB_SQUARE]);
+
+    for (int i = 0; i < h; ++i)
+      for (int j = 0; j < w; ++j) {
+        pred8[i * w + j] = rnd_.Rand8();
+        ref8[i * w + j] = rnd_.Rand8();
+      }
+    const int in_w = block_size_wide[block_idx];
+    const int in_h = block_size_high[block_idx];
+
+    JNT_COMP_PARAMS jnt_comp_params;
+    jnt_comp_params.use_jnt_comp_avg = 1;
+
+    jnt_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
+    jnt_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
+
+    int sub_x_q3 = 0;
+    int sub_y_q3 = 0;
+
+    const int num_loops = 1000000000 / (in_w + in_h);
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    int subpel_search = 2;  // set to 1 to test 4-tap filter.
+
+    for (int i = 0; i < num_loops; ++i)
+      aom_jnt_comp_avg_upsampled_pred_c(NULL, NULL, 0, 0, NULL, output, pred8,
+                                        in_w, in_h, sub_x_q3, sub_y_q3, ref8,
+                                        in_w, &jnt_comp_params, subpel_search);
+
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+    printf("jntcompavgupsampled c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+           1000.0 * elapsed_time / num_loops);
+
+    aom_usec_timer timer1;
+    aom_usec_timer_start(&timer1);
+
+    for (int i = 0; i < num_loops; ++i)
+      test_impl(NULL, NULL, 0, 0, NULL, output2, pred8, in_w, in_h, sub_x_q3,
+                sub_y_q3, ref8, in_w, &jnt_comp_params, subpel_search);
+
+    aom_usec_timer_mark(&timer1);
+    const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
+    printf("jntcompavgupsampled test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+           1000.0 * elapsed_time1 / num_loops);
+  }
+
+  libaom_test::ACMRandom rnd_;
+};  // class AV1JNTCOMPAVGUPSAMPLEDTest
+
+class AV1HighBDJNTCOMPAVGTest
+    : public ::testing::TestWithParam<HighbdJNTCOMPAVGParam> {
+ public:
+  ~AV1HighBDJNTCOMPAVGTest() {}
+  void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+
+  void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void RunCheckOutput(jntcompavg_func test_impl) {
+    const int w = kMaxSize, h = kMaxSize;
+    const int block_idx = GET_PARAM(2);
+    const int bd = GET_PARAM(0);
+    uint16_t pred8[kMaxSize * kMaxSize];
+    uint16_t ref8[kMaxSize * kMaxSize];
+    uint16_t output[kMaxSize * kMaxSize];
+    uint16_t output2[kMaxSize * kMaxSize];
+
+    for (int i = 0; i < h; ++i)
+      for (int j = 0; j < w; ++j) {
+        pred8[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+        ref8[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+      }
+    const int in_w = block_size_wide[block_idx];
+    const int in_h = block_size_high[block_idx];
+
+    JNT_COMP_PARAMS jnt_comp_params;
+    jnt_comp_params.use_jnt_comp_avg = 1;
+
+    for (int ii = 0; ii < 2; ii++) {
+      for (int jj = 0; jj < 4; jj++) {
+        jnt_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+        jnt_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+
+        const int offset_r = 3 + rnd_.PseudoUniform(h - in_h - 7);
+        const int offset_c = 3 + rnd_.PseudoUniform(w - in_w - 7);
+        aom_highbd_jnt_comp_avg_pred_c(
+            CONVERT_TO_BYTEPTR(output),
+            CONVERT_TO_BYTEPTR(pred8) + offset_r * w + offset_c, in_w, in_h,
+            CONVERT_TO_BYTEPTR(ref8) + offset_r * w + offset_c, in_w,
+            &jnt_comp_params);
+        test_impl(CONVERT_TO_BYTEPTR(output2),
+                  CONVERT_TO_BYTEPTR(pred8) + offset_r * w + offset_c, in_w,
+                  in_h, CONVERT_TO_BYTEPTR(ref8) + offset_r * w + offset_c,
+                  in_w, &jnt_comp_params);
+
+        for (int i = 0; i < in_h; ++i) {
+          for (int j = 0; j < in_w; ++j) {
+            int idx = i * in_w + j;
+            ASSERT_EQ(output[idx], output2[idx])
+                << "Mismatch at unit tests for AV1HighBDJNTCOMPAVGTest\n"
+                << in_w << "x" << in_h << " Pixel mismatch at index " << idx
+                << " = (" << i << ", " << j << ")";
+          }
+        }
+      }
+    }
+  }
+  void RunSpeedTest(jntcompavg_func test_impl) {
+    const int w = kMaxSize, h = kMaxSize;
+    const int block_idx = GET_PARAM(2);
+    const int bd = GET_PARAM(0);
+    uint16_t pred8[kMaxSize * kMaxSize];
+    uint16_t ref8[kMaxSize * kMaxSize];
+    uint16_t output[kMaxSize * kMaxSize];
+    uint16_t output2[kMaxSize * kMaxSize];
+
+    for (int i = 0; i < h; ++i)
+      for (int j = 0; j < w; ++j) {
+        pred8[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+        ref8[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+      }
+    const int in_w = block_size_wide[block_idx];
+    const int in_h = block_size_high[block_idx];
+
+    JNT_COMP_PARAMS jnt_comp_params;
+    jnt_comp_params.use_jnt_comp_avg = 1;
+
+    jnt_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
+    jnt_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
+
+    const int num_loops = 1000000000 / (in_w + in_h);
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+
+    for (int i = 0; i < num_loops; ++i)
+      aom_highbd_jnt_comp_avg_pred_c(
+          CONVERT_TO_BYTEPTR(output), CONVERT_TO_BYTEPTR(pred8), in_w, in_h,
+          CONVERT_TO_BYTEPTR(ref8), in_w, &jnt_comp_params);
+
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+    printf("highbdjntcompavg c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+           1000.0 * elapsed_time / num_loops);
+
+    aom_usec_timer timer1;
+    aom_usec_timer_start(&timer1);
+
+    for (int i = 0; i < num_loops; ++i)
+      test_impl(CONVERT_TO_BYTEPTR(output2), CONVERT_TO_BYTEPTR(pred8), in_w,
+                in_h, CONVERT_TO_BYTEPTR(ref8), in_w, &jnt_comp_params);
+
+    aom_usec_timer_mark(&timer1);
+    const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
+    printf("highbdjntcompavg test_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+           1000.0 * elapsed_time1 / num_loops);
+  }
+
+  libaom_test::ACMRandom rnd_;
+};  // class AV1HighBDJNTCOMPAVGTest
+
+class AV1HighBDJNTCOMPAVGUPSAMPLEDTest
+    : public ::testing::TestWithParam<HighbdJNTCOMPAVGUPSAMPLEDParam> {
+ public:
+  ~AV1HighBDJNTCOMPAVGUPSAMPLEDTest() {}
+  void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+  void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void RunCheckOutput(highbdjntcompavgupsampled_func test_impl) {
+    const int w = kMaxSize, h = kMaxSize;
+    const int block_idx = GET_PARAM(2);
+    const int bd = GET_PARAM(0);
+    uint16_t pred8[kMaxSize * kMaxSize];
+    uint16_t ref8[kMaxSize * kMaxSize];
+    uint16_t output[kMaxSize * kMaxSize];
+    uint16_t output2[kMaxSize * kMaxSize];
+
+    for (int i = 0; i < h; ++i)
+      for (int j = 0; j < w; ++j) {
+        pred8[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+        ref8[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+      }
+    const int in_w = block_size_wide[block_idx];
+    const int in_h = block_size_high[block_idx];
+
+    JNT_COMP_PARAMS jnt_comp_params;
+    jnt_comp_params.use_jnt_comp_avg = 1;
+    int sub_x_q3, sub_y_q3;
+    int subpel_search;
+    for (subpel_search = 1; subpel_search <= 2; ++subpel_search) {
+      for (sub_x_q3 = 0; sub_x_q3 < 8; ++sub_x_q3) {
+        for (sub_y_q3 = 0; sub_y_q3 < 8; ++sub_y_q3) {
+          for (int ii = 0; ii < 2; ii++) {
+            for (int jj = 0; jj < 4; jj++) {
+              jnt_comp_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+              jnt_comp_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+
+              const int offset_r = 3 + rnd_.PseudoUniform(h - in_h - 7);
+              const int offset_c = 3 + rnd_.PseudoUniform(w - in_w - 7);
+
+              aom_highbd_jnt_comp_avg_upsampled_pred_c(
+                  NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(output),
+                  CONVERT_TO_BYTEPTR(pred8) + offset_r * w + offset_c, in_w,
+                  in_h, sub_x_q3, sub_y_q3,
+                  CONVERT_TO_BYTEPTR(ref8) + offset_r * w + offset_c, in_w, bd,
+                  &jnt_comp_params, subpel_search);
+              test_impl(NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(output2),
+                        CONVERT_TO_BYTEPTR(pred8) + offset_r * w + offset_c,
+                        in_w, in_h, sub_x_q3, sub_y_q3,
+                        CONVERT_TO_BYTEPTR(ref8) + offset_r * w + offset_c,
+                        in_w, bd, &jnt_comp_params, subpel_search);
+
+              for (int i = 0; i < in_h; ++i) {
+                for (int j = 0; j < in_w; ++j) {
+                  int idx = i * in_w + j;
+                  ASSERT_EQ(output[idx], output2[idx])
+                      << "Mismatch at unit tests for "
+                         "AV1HighBDJNTCOMPAVGUPSAMPLEDTest\n"
+                      << in_w << "x" << in_h << " Pixel mismatch at index "
+                      << idx << " = (" << i << ", " << j
+                      << "), sub pixel offset = (" << sub_y_q3 << ", "
+                      << sub_x_q3 << ")";
+                }
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  void RunSpeedTest(highbdjntcompavgupsampled_func test_impl) {
+    const int w = kMaxSize, h = kMaxSize;
+    const int block_idx = GET_PARAM(2);
+    const int bd = GET_PARAM(0);
+    uint16_t pred8[kMaxSize * kMaxSize];
+    uint16_t ref8[kMaxSize * kMaxSize];
+    uint16_t output[kMaxSize * kMaxSize];
+    uint16_t output2[kMaxSize * kMaxSize];
+
+    for (int i = 0; i < h; ++i)
+      for (int j = 0; j < w; ++j) {
+        pred8[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+        ref8[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+      }
+    const int in_w = block_size_wide[block_idx];
+    const int in_h = block_size_high[block_idx];
+
+    JNT_COMP_PARAMS jnt_comp_params;
+    jnt_comp_params.use_jnt_comp_avg = 1;
+
+    jnt_comp_params.fwd_offset = quant_dist_lookup_table[0][0][0];
+    jnt_comp_params.bck_offset = quant_dist_lookup_table[0][0][1];
+    int sub_x_q3 = 0;
+    int sub_y_q3 = 0;
+    const int num_loops = 1000000000 / (in_w + in_h);
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    int subpel_search = 2;  // set to 1 to test 4-tap filter.
+    for (int i = 0; i < num_loops; ++i)
+      aom_highbd_jnt_comp_avg_upsampled_pred_c(
+          NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(output),
+          CONVERT_TO_BYTEPTR(pred8), in_w, in_h, sub_x_q3, sub_y_q3,
+          CONVERT_TO_BYTEPTR(ref8), in_w, bd, &jnt_comp_params, subpel_search);
+
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+    printf("highbdjntcompavgupsampled c_code %3dx%-3d: %7.2f us\n", in_w, in_h,
+           1000.0 * elapsed_time / num_loops);
+
+    aom_usec_timer timer1;
+    aom_usec_timer_start(&timer1);
+
+    for (int i = 0; i < num_loops; ++i)
+      test_impl(NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(output2),
+                CONVERT_TO_BYTEPTR(pred8), in_w, in_h, sub_x_q3, sub_y_q3,
+                CONVERT_TO_BYTEPTR(ref8), in_w, bd, &jnt_comp_params,
+                subpel_search);
+
+    aom_usec_timer_mark(&timer1);
+    const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
+    printf("highbdjntcompavgupsampled test_code %3dx%-3d: %7.2f us\n", in_w,
+           in_h, 1000.0 * elapsed_time1 / num_loops);
+  }
+
+  libaom_test::ACMRandom rnd_;
+};  // class AV1HighBDJNTCOMPAVGUPSAMPLEDTest
+
+}  // namespace AV1JNTCOMPAVG
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_COMP_AVG_PRED_TEST_H_
diff --git a/third_party/aom/test/comp_mask_variance_test.cc b/third_party/aom/test/comp_mask_variance_test.cc
new file mode 100644
index 0000000000..34be2aa6db
--- /dev/null
+++ b/third_party/aom/test/comp_mask_variance_test.cc
@@ -0,0 +1,574 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cstdlib>
+#include <new>
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_codec.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/variance.h"
+#include "aom_mem/aom_mem.h"
+#include "aom_ports/aom_timer.h"
+#include "aom_ports/mem.h"
+#include "av1/common/reconinter.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace AV1CompMaskVariance {
+typedef void (*comp_mask_pred_func)(uint8_t *comp_pred, const uint8_t *pred,
+                                    int width, int height, const uint8_t *ref,
+                                    int ref_stride, const uint8_t *mask,
+                                    int mask_stride, int invert_mask);
+
+#if HAVE_SSSE3 || HAVE_AV2
+const BLOCK_SIZE kValidBlockSize[] = {
+  BLOCK_8X8,   BLOCK_8X16, BLOCK_8X32,  BLOCK_16X8,  BLOCK_16X16,
+  BLOCK_16X32, BLOCK_32X8, BLOCK_32X16, BLOCK_32X32,
+};
+#endif
+typedef ::testing::tuple<comp_mask_pred_func, BLOCK_SIZE> CompMaskPredParam;
+
+class AV1CompMaskVarianceTest
+    : public ::testing::TestWithParam<CompMaskPredParam> {
+ public:
+  ~AV1CompMaskVarianceTest();
+  void SetUp();
+
+  void TearDown();
+
+ protected:
+  void RunCheckOutput(comp_mask_pred_func test_impl, BLOCK_SIZE bsize, int inv);
+  void RunSpeedTest(comp_mask_pred_func test_impl, BLOCK_SIZE bsize);
+  bool CheckResult(int width, int height) {
+    for (int y = 0; y < height; ++y) {
+      for (int x = 0; x < width; ++x) {
+        const int idx = y * width + x;
+        if (comp_pred1_[idx] != comp_pred2_[idx]) {
+          printf("%dx%d mismatch @%d(%d,%d) ", width, height, idx, y, x);
+          printf("%d != %d ", comp_pred1_[idx], comp_pred2_[idx]);
+          return false;
+        }
+      }
+    }
+    return true;
+  }
+
+  libaom_test::ACMRandom rnd_;
+  uint8_t *comp_pred1_;
+  uint8_t *comp_pred2_;
+  uint8_t *pred_;
+  uint8_t *ref_buffer_;
+  uint8_t *ref_;
+};
+
+AV1CompMaskVarianceTest::~AV1CompMaskVarianceTest() { ; }
+
+void AV1CompMaskVarianceTest::SetUp() {
+  rnd_.Reset(libaom_test::ACMRandom::DeterministicSeed());
+  av1_init_wedge_masks();
+  comp_pred1_ = (uint8_t *)aom_memalign(16, MAX_SB_SQUARE);
+  comp_pred2_ = (uint8_t *)aom_memalign(16, MAX_SB_SQUARE);
+  pred_ = (uint8_t *)aom_memalign(16, MAX_SB_SQUARE);
+  ref_buffer_ = (uint8_t *)aom_memalign(16, MAX_SB_SQUARE + (8 * MAX_SB_SIZE));
+  ref_ = ref_buffer_ + (8 * MAX_SB_SIZE);
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    pred_[i] = rnd_.Rand8();
+  }
+  for (int i = 0; i < MAX_SB_SQUARE + (8 * MAX_SB_SIZE); ++i) {
+    ref_buffer_[i] = rnd_.Rand8();
+  }
+}
+
+void AV1CompMaskVarianceTest::TearDown() {
+  aom_free(comp_pred1_);
+  aom_free(comp_pred2_);
+  aom_free(pred_);
+  aom_free(ref_buffer_);
+  libaom_test::ClearSystemState();
+}
+
+void AV1CompMaskVarianceTest::RunCheckOutput(comp_mask_pred_func test_impl,
+                                             BLOCK_SIZE bsize, int inv) {
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+
+  int wedge_types = (1 << get_wedge_bits_lookup(bsize));
+  for (int wedge_index = 0; wedge_index < wedge_types; ++wedge_index) {
+    const uint8_t *mask = av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
+
+    aom_comp_mask_pred_c(comp_pred1_, pred_, w, h, ref_, MAX_SB_SIZE, mask, w,
+                         inv);
+    test_impl(comp_pred2_, pred_, w, h, ref_, MAX_SB_SIZE, mask, w, inv);
+
+    ASSERT_EQ(CheckResult(w, h), true)
+        << " wedge " << wedge_index << " inv " << inv;
+  }
+}
+
+void AV1CompMaskVarianceTest::RunSpeedTest(comp_mask_pred_func test_impl,
+                                           BLOCK_SIZE bsize) {
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+
+  int wedge_types = (1 << get_wedge_bits_lookup(bsize));
+  int wedge_index = wedge_types / 2;
+  const uint8_t *mask = av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
+  const int num_loops = 1000000000 / (w + h);
+
+  comp_mask_pred_func funcs[2] = { aom_comp_mask_pred_c, test_impl };
+  double elapsed_time[2] = { 0 };
+  for (int i = 0; i < 2; ++i) {
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    comp_mask_pred_func func = funcs[i];
+    for (int j = 0; j < num_loops; ++j) {
+      func(comp_pred1_, pred_, w, h, ref_, MAX_SB_SIZE, mask, w, 0);
+    }
+    aom_usec_timer_mark(&timer);
+    double time = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    elapsed_time[i] = 1000.0 * time / num_loops;
+  }
+  printf("compMask %3dx%-3d: %7.2f/%7.2fns", w, h, elapsed_time[0],
+         elapsed_time[1]);
+  printf("(%3.2f)\n", elapsed_time[0] / elapsed_time[1]);
+}
+
+TEST_P(AV1CompMaskVarianceTest, CheckOutput) {
+  // inv = 0, 1
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), 0);
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), 1);
+}
+
+TEST_P(AV1CompMaskVarianceTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(0), GET_PARAM(1));
+}
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, AV1CompMaskVarianceTest,
+    ::testing::Combine(::testing::Values(&aom_comp_mask_pred_ssse3),
+                       ::testing::ValuesIn(kValidBlockSize)));
+#endif
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AV1CompMaskVarianceTest,
+    ::testing::Combine(::testing::Values(&aom_comp_mask_pred_avx2),
+                       ::testing::ValuesIn(kValidBlockSize)));
+#endif
+
+#ifndef aom_comp_mask_pred
+// can't run this test if aom_comp_mask_pred is defined to aom_comp_mask_pred_c
+class AV1CompMaskUpVarianceTest : public AV1CompMaskVarianceTest {
+ public:
+  ~AV1CompMaskUpVarianceTest();
+
+ protected:
+  void RunCheckOutput(comp_mask_pred_func test_impl, BLOCK_SIZE bsize, int inv);
+  void RunSpeedTest(comp_mask_pred_func test_impl, BLOCK_SIZE bsize,
+                    int havSub);
+};
+
+AV1CompMaskUpVarianceTest::~AV1CompMaskUpVarianceTest() { ; }
+
+void AV1CompMaskUpVarianceTest::RunCheckOutput(comp_mask_pred_func test_impl,
+                                               BLOCK_SIZE bsize, int inv) {
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+  int wedge_types = (1 << get_wedge_bits_lookup(bsize));
+  int subpel_search;
+  for (subpel_search = 1; subpel_search <= 2; ++subpel_search) {
+    // loop through subx and suby
+    for (int sub = 0; sub < 8 * 8; ++sub) {
+      int subx = sub & 0x7;
+      int suby = (sub >> 3);
+      for (int wedge_index = 0; wedge_index < wedge_types; ++wedge_index) {
+        const uint8_t *mask =
+            av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
+
+        // ref
+        aom_comp_mask_upsampled_pred_c(
+            NULL, NULL, 0, 0, NULL, comp_pred1_, pred_, w, h, subx, suby, ref_,
+            MAX_SB_SIZE, mask, w, inv, subpel_search);
+
+        aom_comp_mask_pred = test_impl;  // test
+        aom_comp_mask_upsampled_pred(NULL, NULL, 0, 0, NULL, comp_pred2_, pred_,
+                                     w, h, subx, suby, ref_, MAX_SB_SIZE, mask,
+                                     w, inv, subpel_search);
+        ASSERT_EQ(CheckResult(w, h), true)
+            << " wedge " << wedge_index << " inv " << inv << "sub (" << subx
+            << "," << suby << ")";
+      }
+    }
+  }
+}
+
+void AV1CompMaskUpVarianceTest::RunSpeedTest(comp_mask_pred_func test_impl,
+                                             BLOCK_SIZE bsize, int havSub) {
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+  const int subx = havSub ? 3 : 0;
+  const int suby = havSub ? 4 : 0;
+
+  int wedge_types = (1 << get_wedge_bits_lookup(bsize));
+  int wedge_index = wedge_types / 2;
+  const uint8_t *mask = av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
+
+  const int num_loops = 1000000000 / (w + h);
+  comp_mask_pred_func funcs[2] = { &aom_comp_mask_pred_c, test_impl };
+  double elapsed_time[2] = { 0 };
+  int subpel_search = 2;  // set to 1 to test 4-tap filter.
+  for (int i = 0; i < 2; ++i) {
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    aom_comp_mask_pred = funcs[i];
+    for (int j = 0; j < num_loops; ++j) {
+      aom_comp_mask_upsampled_pred(NULL, NULL, 0, 0, NULL, comp_pred1_, pred_,
+                                   w, h, subx, suby, ref_, MAX_SB_SIZE, mask, w,
+                                   0, subpel_search);
+    }
+    aom_usec_timer_mark(&timer);
+    double time = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    elapsed_time[i] = 1000.0 * time / num_loops;
+  }
+  printf("CompMaskUp[%d] %3dx%-3d:%7.2f/%7.2fns", havSub, w, h, elapsed_time[0],
+         elapsed_time[1]);
+  printf("(%3.2f)\n", elapsed_time[0] / elapsed_time[1]);
+}
+
+TEST_P(AV1CompMaskUpVarianceTest, CheckOutput) {
+  // inv mask = 0, 1
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), 0);
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), 1);
+}
+
+TEST_P(AV1CompMaskUpVarianceTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(0), GET_PARAM(1), 1);
+}
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, AV1CompMaskUpVarianceTest,
+    ::testing::Combine(::testing::Values(&aom_comp_mask_pred_ssse3),
+                       ::testing::ValuesIn(kValidBlockSize)));
+#endif
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AV1CompMaskUpVarianceTest,
+    ::testing::Combine(::testing::Values(&aom_comp_mask_pred_avx2),
+                       ::testing::ValuesIn(kValidBlockSize)));
+#endif
+
+#endif  // ifndef aom_comp_mask_pred
+
+typedef void (*highbd_comp_mask_pred_func)(uint8_t *comp_pred8,
+                                           const uint8_t *pred8, int width,
+                                           int height, const uint8_t *ref8,
+                                           int ref_stride, const uint8_t *mask,
+                                           int mask_stride, int invert_mask);
+
+typedef ::testing::tuple<highbd_comp_mask_pred_func, BLOCK_SIZE, int>
+    HighbdCompMaskPredParam;
+
+class AV1HighbdCompMaskVarianceTest
+    : public ::testing::TestWithParam<HighbdCompMaskPredParam> {
+ public:
+  ~AV1HighbdCompMaskVarianceTest();
+  void SetUp();
+
+  void TearDown();
+
+ protected:
+  void RunCheckOutput(highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize,
+                      int inv);
+  void RunSpeedTest(highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize);
+  bool CheckResult(int width, int height) {
+    for (int y = 0; y < height; ++y) {
+      for (int x = 0; x < width; ++x) {
+        const int idx = y * width + x;
+        if (comp_pred1_[idx] != comp_pred2_[idx]) {
+          printf("%dx%d mismatch @%d(%d,%d) ", width, height, idx, y, x);
+          printf("%d != %d ", comp_pred1_[idx], comp_pred2_[idx]);
+          return false;
+        }
+      }
+    }
+    return true;
+  }
+
+  libaom_test::ACMRandom rnd_;
+  uint16_t *comp_pred1_;
+  uint16_t *comp_pred2_;
+  uint16_t *pred_;
+  uint16_t *ref_buffer_;
+  uint16_t *ref_;
+};
+
+AV1HighbdCompMaskVarianceTest::~AV1HighbdCompMaskVarianceTest() { ; }
+
+void AV1HighbdCompMaskVarianceTest::SetUp() {
+  rnd_.Reset(libaom_test::ACMRandom::DeterministicSeed());
+  av1_init_wedge_masks();
+
+  comp_pred1_ =
+      (uint16_t *)aom_memalign(16, MAX_SB_SQUARE * sizeof(*comp_pred1_));
+  comp_pred2_ =
+      (uint16_t *)aom_memalign(16, MAX_SB_SQUARE * sizeof(*comp_pred2_));
+  pred_ = (uint16_t *)aom_memalign(16, MAX_SB_SQUARE * sizeof(*pred_));
+  ref_buffer_ = (uint16_t *)aom_memalign(
+      16, (MAX_SB_SQUARE + (8 * MAX_SB_SIZE)) * sizeof(*ref_buffer_));
+  ref_ = ref_buffer_ + (8 * MAX_SB_SIZE);
+}
+
+void AV1HighbdCompMaskVarianceTest::TearDown() {
+  aom_free(comp_pred1_);
+  aom_free(comp_pred2_);
+  aom_free(pred_);
+  aom_free(ref_buffer_);
+  libaom_test::ClearSystemState();
+}
+
+void AV1HighbdCompMaskVarianceTest::RunCheckOutput(
+    highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize, int inv) {
+  int bd_ = GET_PARAM(2);
+
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+
+  int wedge_types = (1 << get_wedge_bits_lookup(bsize));
+
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    pred_[i] = rnd_.Rand16() & ((1 << bd_) - 1);
+  }
+  for (int i = 0; i < MAX_SB_SQUARE + (8 * MAX_SB_SIZE); ++i) {
+    ref_buffer_[i] = rnd_.Rand16() & ((1 << bd_) - 1);
+  }
+
+  for (int wedge_index = 0; wedge_index < wedge_types; ++wedge_index) {
+    const uint8_t *mask = av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
+
+    aom_highbd_comp_mask_pred_c(
+        CONVERT_TO_BYTEPTR(comp_pred1_), CONVERT_TO_BYTEPTR(pred_), w, h,
+        CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, mask, w, inv);
+
+    test_impl(CONVERT_TO_BYTEPTR(comp_pred2_), CONVERT_TO_BYTEPTR(pred_), w, h,
+              CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, mask, w, inv);
+
+    ASSERT_EQ(CheckResult(w, h), true)
+        << " wedge " << wedge_index << " inv " << inv;
+  }
+}
+
+void AV1HighbdCompMaskVarianceTest::RunSpeedTest(
+    highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize) {
+  int bd_ = GET_PARAM(2);
+
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+
+  int wedge_types = (1 << get_wedge_bits_lookup(bsize));
+  int wedge_index = wedge_types / 2;
+
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    pred_[i] = rnd_.Rand16() & ((1 << bd_) - 1);
+  }
+  for (int i = 0; i < MAX_SB_SQUARE + (8 * MAX_SB_SIZE); ++i) {
+    ref_buffer_[i] = rnd_.Rand16() & ((1 << bd_) - 1);
+  }
+
+  const uint8_t *mask = av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
+  const int num_loops = 1000000000 / (w + h);
+
+  highbd_comp_mask_pred_func funcs[2] = { aom_highbd_comp_mask_pred_c,
+                                          test_impl };
+  double elapsed_time[2] = { 0 };
+  for (int i = 0; i < 2; ++i) {
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    highbd_comp_mask_pred_func func = funcs[i];
+    for (int j = 0; j < num_loops; ++j) {
+      func(CONVERT_TO_BYTEPTR(comp_pred1_), CONVERT_TO_BYTEPTR(pred_), w, h,
+           CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, mask, w, 0);
+    }
+    aom_usec_timer_mark(&timer);
+    double time = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    elapsed_time[i] = 1000.0 * time / num_loops;
+  }
+  printf("compMask %3dx%-3d: %7.2f/%7.2fns", w, h, elapsed_time[0],
+         elapsed_time[1]);
+  printf("(%3.2f)\n", elapsed_time[0] / elapsed_time[1]);
+}
+
+TEST_P(AV1HighbdCompMaskVarianceTest, CheckOutput) {
+  // inv = 0, 1
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), 0);
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), 1);
+}
+
+TEST_P(AV1HighbdCompMaskVarianceTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(0), GET_PARAM(1));
+}
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AV1HighbdCompMaskVarianceTest,
+    ::testing::Combine(::testing::Values(&aom_highbd_comp_mask_pred_avx2),
+                       ::testing::ValuesIn(kValidBlockSize),
+                       ::testing::Range(8, 13, 2)));
+#endif
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, AV1HighbdCompMaskVarianceTest,
+    ::testing::Combine(::testing::Values(&aom_highbd_comp_mask_pred_sse2),
+                       ::testing::ValuesIn(kValidBlockSize),
+                       ::testing::Range(8, 13, 2)));
+#endif
+
+#ifndef aom_highbd_comp_mask_pred
+// can't run this test if aom_highbd_comp_mask_pred is defined to
+// aom_highbd_comp_mask_pred_c
+class AV1HighbdCompMaskUpVarianceTest : public AV1HighbdCompMaskVarianceTest {
+ public:
+  ~AV1HighbdCompMaskUpVarianceTest();
+
+ protected:
+  void RunCheckOutput(highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize,
+                      int inv);
+  void RunSpeedTest(highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize,
+                    int havSub);
+};
+
+AV1HighbdCompMaskUpVarianceTest::~AV1HighbdCompMaskUpVarianceTest() { ; }
+
+void AV1HighbdCompMaskUpVarianceTest::RunCheckOutput(
+    highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize, int inv) {
+  int bd_ = GET_PARAM(2);
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+  int wedge_types = (1 << get_wedge_bits_lookup(bsize));
+
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    pred_[i] = rnd_.Rand16() & ((1 << bd_) - 1);
+  }
+  for (int i = 0; i < MAX_SB_SQUARE + (8 * MAX_SB_SIZE); ++i) {
+    ref_buffer_[i] = rnd_.Rand16() & ((1 << bd_) - 1);
+  }
+
+  int subpel_search;
+  for (subpel_search = 1; subpel_search <= 2; ++subpel_search) {
+    // loop through subx and suby
+    for (int sub = 0; sub < 8 * 8; ++sub) {
+      int subx = sub & 0x7;
+      int suby = (sub >> 3);
+      for (int wedge_index = 0; wedge_index < wedge_types; ++wedge_index) {
+        const uint8_t *mask =
+            av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
+
+        aom_highbd_comp_mask_pred = aom_highbd_comp_mask_pred_c;  // ref
+        aom_highbd_comp_mask_upsampled_pred(
+            NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(comp_pred1_),
+            CONVERT_TO_BYTEPTR(pred_), w, h, subx, suby,
+            CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, mask, w, inv, bd_,
+            subpel_search);
+
+        aom_highbd_comp_mask_pred = test_impl;  // test
+        aom_highbd_comp_mask_upsampled_pred(
+            NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(comp_pred2_),
+            CONVERT_TO_BYTEPTR(pred_), w, h, subx, suby,
+            CONVERT_TO_BYTEPTR(ref_), MAX_SB_SIZE, mask, w, inv, bd_,
+            subpel_search);
+        ASSERT_EQ(CheckResult(w, h), true)
+            << " wedge " << wedge_index << " inv " << inv << "sub (" << subx
+            << "," << suby << ")";
+      }
+    }
+  }
+}
+
+void AV1HighbdCompMaskUpVarianceTest::RunSpeedTest(
+    highbd_comp_mask_pred_func test_impl, BLOCK_SIZE bsize, int havSub) {
+  int bd_ = GET_PARAM(2);
+  const int w = block_size_wide[bsize];
+  const int h = block_size_high[bsize];
+  const int subx = havSub ? 3 : 0;
+  const int suby = havSub ? 4 : 0;
+
+  int wedge_types = (1 << get_wedge_bits_lookup(bsize));
+  int wedge_index = wedge_types / 2;
+  const uint8_t *mask = av1_get_contiguous_soft_mask(wedge_index, 1, bsize);
+
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    pred_[i] = rnd_.Rand16() & ((1 << bd_) - 1);
+  }
+  for (int i = 0; i < MAX_SB_SQUARE + (8 * MAX_SB_SIZE); ++i) {
+    ref_buffer_[i] = rnd_.Rand16() & ((1 << bd_) - 1);
+  }
+
+  const int num_loops = 1000000000 / (w + h);
+  highbd_comp_mask_pred_func funcs[2] = { &aom_highbd_comp_mask_pred_c,
+                                          test_impl };
+  double elapsed_time[2] = { 0 };
+  for (int i = 0; i < 2; ++i) {
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    aom_highbd_comp_mask_pred = funcs[i];
+    int subpel_search = 2;  // set to 1 to test 4-tap filter.
+    for (int j = 0; j < num_loops; ++j) {
+      aom_highbd_comp_mask_upsampled_pred(
+          NULL, NULL, 0, 0, NULL, CONVERT_TO_BYTEPTR(comp_pred1_),
+          CONVERT_TO_BYTEPTR(pred_), w, h, subx, suby, CONVERT_TO_BYTEPTR(ref_),
+          MAX_SB_SIZE, mask, w, 0, bd_, subpel_search);
+    }
+    aom_usec_timer_mark(&timer);
+    double time = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    elapsed_time[i] = 1000.0 * time / num_loops;
+  }
+  printf("CompMaskUp[%d] %3dx%-3d:%7.2f/%7.2fns", havSub, w, h, elapsed_time[0],
+         elapsed_time[1]);
+  printf("(%3.2f)\n", elapsed_time[0] / elapsed_time[1]);
+}
+
+TEST_P(AV1HighbdCompMaskUpVarianceTest, CheckOutput) {
+  // inv mask = 0, 1
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), 0);
+  RunCheckOutput(GET_PARAM(0), GET_PARAM(1), 1);
+}
+
+TEST_P(AV1HighbdCompMaskUpVarianceTest, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(0), GET_PARAM(1), 1);
+}
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AV1HighbdCompMaskUpVarianceTest,
+    ::testing::Combine(::testing::Values(&aom_highbd_comp_mask_pred_avx2),
+                       ::testing::ValuesIn(kValidBlockSize),
+                       ::testing::Range(8, 13, 2)));
+#endif
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, AV1HighbdCompMaskUpVarianceTest,
+    ::testing::Combine(::testing::Values(&aom_highbd_comp_mask_pred_sse2),
+                       ::testing::ValuesIn(kValidBlockSize),
+                       ::testing::Range(8, 13, 2)));
+#endif
+
+#endif  // ifndef aom_highbd_comp_mask_pred
+}  // namespace AV1CompMaskVariance
diff --git a/third_party/aom/test/convolve_round_test.cc b/third_party/aom/test/convolve_round_test.cc
new file mode 100644
index 0000000000..2f801e7d46
--- /dev/null
+++ b/third_party/aom/test/convolve_round_test.cc
@@ -0,0 +1,183 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <assert.h>
+
+#include "config/av1_rtcd.h"
+
+#include "aom/aom_integer.h"
+#include "aom_ports/aom_timer.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+#define CONVOLVE_ROUNDING_PARAM                                            \
+  const int32_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, \
+      int h, int bits
+
+typedef void (*ConvolveRoundFunc)(CONVOLVE_ROUNDING_PARAM);
+
+typedef void (*ConvolveRoundFuncHbd)(CONVOLVE_ROUNDING_PARAM, int bd);
+
+template <ConvolveRoundFuncHbd fn>
+void highbd_convolve_rounding_8(CONVOLVE_ROUNDING_PARAM) {
+  const int bd = 8;
+  fn(src, src_stride, dst, dst_stride, w, h, bits, bd);
+}
+
+template <ConvolveRoundFuncHbd fn>
+void highbd_convolve_rounding_10(CONVOLVE_ROUNDING_PARAM) {
+  const int bd = 10;
+  fn(src, src_stride, dst, dst_stride, w, h, bits, bd);
+}
+
+template <ConvolveRoundFuncHbd fn>
+void highbd_convolve_rounding_12(CONVOLVE_ROUNDING_PARAM) {
+  const int bd = 12;
+  fn(src, src_stride, dst, dst_stride, w, h, bits, bd);
+}
+
+typedef enum { LOWBITDEPTH_TEST, HIGHBITDEPTH_TEST } DataPathType;
+
+using ::testing::tuple;
+
+typedef tuple<ConvolveRoundFunc, ConvolveRoundFunc, DataPathType>
+    ConvolveRoundParam;
+
+const int kTestNum = 5000;
+
+class ConvolveRoundTest : public ::testing::TestWithParam<ConvolveRoundParam> {
+ protected:
+  ConvolveRoundTest()
+      : func_ref_(GET_PARAM(0)), func_(GET_PARAM(1)), data_path_(GET_PARAM(2)) {
+  }
+  virtual ~ConvolveRoundTest() {}
+
+  virtual void SetUp() {
+    const size_t block_size = 128 * 128;
+    src_ = reinterpret_cast<int32_t *>(
+        aom_memalign(16, block_size * sizeof(*src_)));
+    dst_ref_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(16, block_size * sizeof(*dst_ref_)));
+    dst_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(16, block_size * sizeof(*dst_)));
+  }
+
+  virtual void TearDown() {
+    aom_free(src_);
+    aom_free(dst_ref_);
+    aom_free(dst_);
+  }
+
+  void ConvolveRoundingRun() {
+    int test_num = 0;
+    const int src_stride = 128;
+    const int dst_stride = 128;
+    int bits = 13;
+    uint8_t *dst = 0;
+    uint8_t *dst_ref = 0;
+
+    if (data_path_ == LOWBITDEPTH_TEST) {
+      dst = reinterpret_cast<uint8_t *>(dst_);
+      dst_ref = reinterpret_cast<uint8_t *>(dst_ref_);
+    } else if (data_path_ == HIGHBITDEPTH_TEST) {
+      dst = CONVERT_TO_BYTEPTR(dst_);
+      dst_ref = CONVERT_TO_BYTEPTR(dst_ref_);
+    } else {
+      assert(0);
+    }
+
+    while (test_num < kTestNum) {
+      int block_size = test_num % BLOCK_SIZES_ALL;
+      int w = block_size_wide[block_size];
+      int h = block_size_high[block_size];
+
+      if (test_num % 2 == 0)
+        bits -= 1;
+      else
+        bits += 1;
+
+      GenerateBufferWithRandom(src_, src_stride, bits, w, h);
+
+      func_ref_(src_, src_stride, dst_ref, dst_stride, w, h, bits);
+      ASM_REGISTER_STATE_CHECK(
+          func_(src_, src_stride, dst, dst_stride, w, h, bits));
+
+      if (data_path_ == LOWBITDEPTH_TEST) {
+        for (int r = 0; r < h; ++r) {
+          for (int c = 0; c < w; ++c) {
+            ASSERT_EQ(dst_ref[r * dst_stride + c], dst[r * dst_stride + c])
+                << "Mismatch at r: " << r << " c: " << c << " w: " << w
+                << " h: " << h << " test: " << test_num;
+          }
+        }
+      } else {
+        for (int r = 0; r < h; ++r) {
+          for (int c = 0; c < w; ++c) {
+            ASSERT_EQ(dst_ref_[r * dst_stride + c], dst_[r * dst_stride + c])
+                << "Mismatch at r: " << r << " c: " << c << " w: " << w
+                << " h: " << h << " test: " << test_num;
+          }
+        }
+      }
+
+      test_num++;
+    }
+  }
+
+  void GenerateBufferWithRandom(int32_t *src, int src_stride, int bits, int w,
+                                int h) {
+    int32_t number;
+    for (int r = 0; r < h; ++r) {
+      for (int c = 0; c < w; ++c) {
+        number = static_cast<int32_t>(rand_.Rand31());
+        number %= 1 << (bits + 9);
+        src[r * src_stride + c] = number;
+      }
+    }
+  }
+
+  ACMRandom rand_;
+  int32_t *src_;
+  uint16_t *dst_ref_;
+  uint16_t *dst_;
+
+  ConvolveRoundFunc func_ref_;
+  ConvolveRoundFunc func_;
+  DataPathType data_path_;
+};
+
+TEST_P(ConvolveRoundTest, BitExactCheck) { ConvolveRoundingRun(); }
+
+using ::testing::make_tuple;
+#if HAVE_AVX2
+const ConvolveRoundParam kConvRndParamArray[] = {
+  make_tuple(&av1_convolve_rounding_c, &av1_convolve_rounding_avx2,
+             LOWBITDEPTH_TEST),
+  make_tuple(&highbd_convolve_rounding_8<av1_highbd_convolve_rounding_c>,
+             &highbd_convolve_rounding_8<av1_highbd_convolve_rounding_avx2>,
+             HIGHBITDEPTH_TEST),
+  make_tuple(&highbd_convolve_rounding_10<av1_highbd_convolve_rounding_c>,
+             &highbd_convolve_rounding_10<av1_highbd_convolve_rounding_avx2>,
+             HIGHBITDEPTH_TEST),
+  make_tuple(&highbd_convolve_rounding_12<av1_highbd_convolve_rounding_c>,
+             &highbd_convolve_rounding_12<av1_highbd_convolve_rounding_avx2>,
+             HIGHBITDEPTH_TEST)
+};
+INSTANTIATE_TEST_CASE_P(AVX2, ConvolveRoundTest,
+                        ::testing::ValuesIn(kConvRndParamArray));
+#endif  // HAVE_AVX2
+}  // namespace
diff --git a/third_party/aom/test/convolve_test.cc b/third_party/aom/test/convolve_test.cc
new file mode 100644
index 0000000000..de3f47628a
--- /dev/null
+++ b/third_party/aom/test/convolve_test.cc
@@ -0,0 +1,856 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom_dsp/aom_dsp_common.h"
+#include "aom_dsp/aom_filter.h"
+#include "aom_mem/aom_mem.h"
+#include "aom_ports/aom_timer.h"
+#include "aom_ports/mem.h"
+#include "av1/common/filter.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+namespace {
+
+static const unsigned int kMaxDimension = MAX_SB_SIZE;
+
+typedef void (*ConvolveFunc)(const uint8_t *src, ptrdiff_t src_stride,
+                             uint8_t *dst, ptrdiff_t dst_stride,
+                             const int16_t *filter_x, int filter_x_stride,
+                             const int16_t *filter_y, int filter_y_stride,
+                             int w, int h);
+
+struct ConvolveFunctions {
+  ConvolveFunctions(ConvolveFunc copy, ConvolveFunc h8, ConvolveFunc v8, int bd)
+      : copy_(copy), h8_(h8), v8_(v8), use_highbd_(bd) {}
+
+  ConvolveFunc copy_;
+  ConvolveFunc h8_;
+  ConvolveFunc v8_;
+  int use_highbd_;  // 0 if high bitdepth not used, else the actual bit depth.
+};
+
+typedef ::testing::tuple<int, int, const ConvolveFunctions *> ConvolveParam;
+
+#define ALL_SIZES_64(convolve_fn)                                         \
+  make_tuple(4, 4, &convolve_fn), make_tuple(8, 4, &convolve_fn),         \
+      make_tuple(4, 8, &convolve_fn), make_tuple(8, 8, &convolve_fn),     \
+      make_tuple(16, 8, &convolve_fn), make_tuple(8, 16, &convolve_fn),   \
+      make_tuple(16, 16, &convolve_fn), make_tuple(32, 16, &convolve_fn), \
+      make_tuple(16, 32, &convolve_fn), make_tuple(32, 32, &convolve_fn), \
+      make_tuple(64, 32, &convolve_fn), make_tuple(32, 64, &convolve_fn), \
+      make_tuple(64, 64, &convolve_fn)
+
+#define ALL_SIZES(convolve_fn)                                          \
+  make_tuple(128, 64, &convolve_fn), make_tuple(64, 128, &convolve_fn), \
+      make_tuple(128, 128, &convolve_fn), ALL_SIZES_64(convolve_fn)
+
+// Reference 8-tap subpixel filter, slightly modified to fit into this test.
+#define AV1_FILTER_WEIGHT 128
+#define AV1_FILTER_SHIFT 7
+uint8_t clip_pixel(int x) { return x < 0 ? 0 : x > 255 ? 255 : x; }
+
+void filter_block2d_8_c(const uint8_t *src_ptr, unsigned int src_stride,
+                        const int16_t *HFilter, const int16_t *VFilter,
+                        uint8_t *dst_ptr, unsigned int dst_stride,
+                        unsigned int output_width, unsigned int output_height) {
+  // Between passes, we use an intermediate buffer whose height is extended to
+  // have enough horizontally filtered values as input for the vertical pass.
+  // This buffer is allocated to be big enough for the largest block type we
+  // support.
+  const int kInterp_Extend = 4;
+  const unsigned int intermediate_height =
+      (kInterp_Extend - 1) + output_height + kInterp_Extend;
+  unsigned int i, j;
+
+  assert(intermediate_height > 7);
+
+  // Size of intermediate_buffer is max_intermediate_height * filter_max_width,
+  // where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
+  //                                 + kInterp_Extend
+  //                               = 3 + 16 + 4
+  //                               = 23
+  // and filter_max_width          = 16
+  //
+  uint8_t intermediate_buffer[(kMaxDimension + 8) * kMaxDimension];
+  const int intermediate_next_stride =
+      1 - static_cast<int>(intermediate_height * output_width);
+
+  // Horizontal pass (src -> transposed intermediate).
+  uint8_t *output_ptr = intermediate_buffer;
+  const int src_next_row_stride = src_stride - output_width;
+  src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
+  for (i = 0; i < intermediate_height; ++i) {
+    for (j = 0; j < output_width; ++j) {
+      // Apply filter...
+      const int temp = (src_ptr[0] * HFilter[0]) + (src_ptr[1] * HFilter[1]) +
+                       (src_ptr[2] * HFilter[2]) + (src_ptr[3] * HFilter[3]) +
+                       (src_ptr[4] * HFilter[4]) + (src_ptr[5] * HFilter[5]) +
+                       (src_ptr[6] * HFilter[6]) + (src_ptr[7] * HFilter[7]) +
+                       (AV1_FILTER_WEIGHT >> 1);  // Rounding
+
+      // Normalize back to 0-255...
+      *output_ptr = clip_pixel(temp >> AV1_FILTER_SHIFT);
+      ++src_ptr;
+      output_ptr += intermediate_height;
+    }
+    src_ptr += src_next_row_stride;
+    output_ptr += intermediate_next_stride;
+  }
+
+  // Vertical pass (transposed intermediate -> dst).
+  src_ptr = intermediate_buffer;
+  const int dst_next_row_stride = dst_stride - output_width;
+  for (i = 0; i < output_height; ++i) {
+    for (j = 0; j < output_width; ++j) {
+      // Apply filter...
+      const int temp = (src_ptr[0] * VFilter[0]) + (src_ptr[1] * VFilter[1]) +
+                       (src_ptr[2] * VFilter[2]) + (src_ptr[3] * VFilter[3]) +
+                       (src_ptr[4] * VFilter[4]) + (src_ptr[5] * VFilter[5]) +
+                       (src_ptr[6] * VFilter[6]) + (src_ptr[7] * VFilter[7]) +
+                       (AV1_FILTER_WEIGHT >> 1);  // Rounding
+
+      // Normalize back to 0-255...
+      *dst_ptr++ = clip_pixel(temp >> AV1_FILTER_SHIFT);
+      src_ptr += intermediate_height;
+    }
+    src_ptr += intermediate_next_stride;
+    dst_ptr += dst_next_row_stride;
+  }
+}
+
+void block2d_average_c(uint8_t *src, unsigned int src_stride,
+                       uint8_t *output_ptr, unsigned int output_stride,
+                       unsigned int output_width, unsigned int output_height) {
+  unsigned int i, j;
+  for (i = 0; i < output_height; ++i) {
+    for (j = 0; j < output_width; ++j) {
+      output_ptr[j] = (output_ptr[j] + src[i * src_stride + j] + 1) >> 1;
+    }
+    output_ptr += output_stride;
+  }
+}
+
+void filter_average_block2d_8_c(const uint8_t *src_ptr,
+                                const unsigned int src_stride,
+                                const int16_t *HFilter, const int16_t *VFilter,
+                                uint8_t *dst_ptr, unsigned int dst_stride,
+                                unsigned int output_width,
+                                unsigned int output_height) {
+  uint8_t tmp[kMaxDimension * kMaxDimension];
+
+  assert(output_width <= kMaxDimension);
+  assert(output_height <= kMaxDimension);
+  filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, tmp, kMaxDimension,
+                     output_width, output_height);
+  block2d_average_c(tmp, kMaxDimension, dst_ptr, dst_stride, output_width,
+                    output_height);
+}
+
+void highbd_filter_block2d_8_c(const uint16_t *src_ptr,
+                               const unsigned int src_stride,
+                               const int16_t *HFilter, const int16_t *VFilter,
+                               uint16_t *dst_ptr, unsigned int dst_stride,
+                               unsigned int output_width,
+                               unsigned int output_height, int bd) {
+  // Between passes, we use an intermediate buffer whose height is extended to
+  // have enough horizontally filtered values as input for the vertical pass.
+  // This buffer is allocated to be big enough for the largest block type we
+  // support.
+  const int kInterp_Extend = 4;
+  const unsigned int intermediate_height =
+      (kInterp_Extend - 1) + output_height + kInterp_Extend;
+
+  /* Size of intermediate_buffer is max_intermediate_height * filter_max_width,
+   * where max_intermediate_height = (kInterp_Extend - 1) + filter_max_height
+   *                                 + kInterp_Extend
+   *                               = 3 + 16 + 4
+   *                               = 23
+   * and filter_max_width = 16
+   */
+  uint16_t intermediate_buffer[(kMaxDimension + 8) * kMaxDimension] = { 0 };
+  const int intermediate_next_stride =
+      1 - static_cast<int>(intermediate_height * output_width);
+
+  // Horizontal pass (src -> transposed intermediate).
+  {
+    uint16_t *output_ptr = intermediate_buffer;
+    const int src_next_row_stride = src_stride - output_width;
+    unsigned int i, j;
+    src_ptr -= (kInterp_Extend - 1) * src_stride + (kInterp_Extend - 1);
+    for (i = 0; i < intermediate_height; ++i) {
+      for (j = 0; j < output_width; ++j) {
+        // Apply filter...
+        const int temp = (src_ptr[0] * HFilter[0]) + (src_ptr[1] * HFilter[1]) +
+                         (src_ptr[2] * HFilter[2]) + (src_ptr[3] * HFilter[3]) +
+                         (src_ptr[4] * HFilter[4]) + (src_ptr[5] * HFilter[5]) +
+                         (src_ptr[6] * HFilter[6]) + (src_ptr[7] * HFilter[7]) +
+                         (AV1_FILTER_WEIGHT >> 1);  // Rounding
+
+        // Normalize back to 0-255...
+        *output_ptr = clip_pixel_highbd(temp >> AV1_FILTER_SHIFT, bd);
+        ++src_ptr;
+        output_ptr += intermediate_height;
+      }
+      src_ptr += src_next_row_stride;
+      output_ptr += intermediate_next_stride;
+    }
+  }
+
+  // Vertical pass (transposed intermediate -> dst).
+  {
+    const uint16_t *interm_ptr = intermediate_buffer;
+    const int dst_next_row_stride = dst_stride - output_width;
+    unsigned int i, j;
+    for (i = 0; i < output_height; ++i) {
+      for (j = 0; j < output_width; ++j) {
+        // Apply filter...
+        const int temp =
+            (interm_ptr[0] * VFilter[0]) + (interm_ptr[1] * VFilter[1]) +
+            (interm_ptr[2] * VFilter[2]) + (interm_ptr[3] * VFilter[3]) +
+            (interm_ptr[4] * VFilter[4]) + (interm_ptr[5] * VFilter[5]) +
+            (interm_ptr[6] * VFilter[6]) + (interm_ptr[7] * VFilter[7]) +
+            (AV1_FILTER_WEIGHT >> 1);  // Rounding
+
+        // Normalize back to 0-255...
+        *dst_ptr++ = clip_pixel_highbd(temp >> AV1_FILTER_SHIFT, bd);
+        interm_ptr += intermediate_height;
+      }
+      interm_ptr += intermediate_next_stride;
+      dst_ptr += dst_next_row_stride;
+    }
+  }
+}
+
+void highbd_block2d_average_c(uint16_t *src, unsigned int src_stride,
+                              uint16_t *output_ptr, unsigned int output_stride,
+                              unsigned int output_width,
+                              unsigned int output_height) {
+  unsigned int i, j;
+  for (i = 0; i < output_height; ++i) {
+    for (j = 0; j < output_width; ++j) {
+      output_ptr[j] = (output_ptr[j] + src[i * src_stride + j] + 1) >> 1;
+    }
+    output_ptr += output_stride;
+  }
+}
+
+void highbd_filter_average_block2d_8_c(
+    const uint16_t *src_ptr, unsigned int src_stride, const int16_t *HFilter,
+    const int16_t *VFilter, uint16_t *dst_ptr, unsigned int dst_stride,
+    unsigned int output_width, unsigned int output_height, int bd) {
+  uint16_t tmp[kMaxDimension * kMaxDimension];
+
+  assert(output_width <= kMaxDimension);
+  assert(output_height <= kMaxDimension);
+  highbd_filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, tmp,
+                            kMaxDimension, output_width, output_height, bd);
+  highbd_block2d_average_c(tmp, kMaxDimension, dst_ptr, dst_stride,
+                           output_width, output_height);
+}
+
+class ConvolveTest : public ::testing::TestWithParam<ConvolveParam> {
+ public:
+  static void SetUpTestCase() {
+    // Force input_ to be unaligned, output to be 16 byte aligned.
+    input_ = reinterpret_cast<uint8_t *>(
+                 aom_memalign(kDataAlignment, kInputBufferSize + 1)) +
+             1;
+    output_ = reinterpret_cast<uint8_t *>(
+        aom_memalign(kDataAlignment, kOutputBufferSize));
+    output_ref_ = reinterpret_cast<uint8_t *>(
+        aom_memalign(kDataAlignment, kOutputBufferSize));
+    input16_ = reinterpret_cast<uint16_t *>(aom_memalign(
+                   kDataAlignment, (kInputBufferSize + 1) * sizeof(uint16_t))) +
+               1;
+    output16_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(kDataAlignment, (kOutputBufferSize) * sizeof(uint16_t)));
+    output16_ref_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(kDataAlignment, (kOutputBufferSize) * sizeof(uint16_t)));
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+  static void TearDownTestCase() {
+    aom_free(input_ - 1);
+    input_ = NULL;
+    aom_free(output_);
+    output_ = NULL;
+    aom_free(output_ref_);
+    output_ref_ = NULL;
+    aom_free(input16_ - 1);
+    input16_ = NULL;
+    aom_free(output16_);
+    output16_ = NULL;
+    aom_free(output16_ref_);
+    output16_ref_ = NULL;
+  }
+
+ protected:
+  static const int kDataAlignment = 16;
+  static const int kOuterBlockSize = 4 * kMaxDimension;
+  static const int kInputStride = kOuterBlockSize;
+  static const int kOutputStride = kOuterBlockSize;
+  static const int kInputBufferSize = kOuterBlockSize * kOuterBlockSize;
+  static const int kOutputBufferSize = kOuterBlockSize * kOuterBlockSize;
+
+  int Width() const { return GET_PARAM(0); }
+  int Height() const { return GET_PARAM(1); }
+  int BorderLeft() const {
+    const int center = (kOuterBlockSize - Width()) / 2;
+    return (center + (kDataAlignment - 1)) & ~(kDataAlignment - 1);
+  }
+  int BorderTop() const { return (kOuterBlockSize - Height()) / 2; }
+
+  bool IsIndexInBorder(int i) {
+    return (i < BorderTop() * kOuterBlockSize ||
+            i >= (BorderTop() + Height()) * kOuterBlockSize ||
+            i % kOuterBlockSize < BorderLeft() ||
+            i % kOuterBlockSize >= (BorderLeft() + Width()));
+  }
+
+  virtual void SetUp() {
+    UUT_ = GET_PARAM(2);
+    if (UUT_->use_highbd_ != 0)
+      mask_ = (1 << UUT_->use_highbd_) - 1;
+    else
+      mask_ = 255;
+    /* Set up guard blocks for an inner block centered in the outer block */
+    for (int i = 0; i < kOutputBufferSize; ++i) {
+      if (IsIndexInBorder(i)) {
+        output_[i] = 255;
+        output16_[i] = mask_;
+      } else {
+        output_[i] = 0;
+        output16_[i] = 0;
+      }
+    }
+
+    ::libaom_test::ACMRandom prng;
+    for (int i = 0; i < kInputBufferSize; ++i) {
+      if (i & 1) {
+        input_[i] = 255;
+        input16_[i] = mask_;
+      } else {
+        input_[i] = prng.Rand8Extremes();
+        input16_[i] = prng.Rand16() & mask_;
+      }
+    }
+  }
+
+  void SetConstantInput(int value) {
+    memset(input_, value, kInputBufferSize);
+    aom_memset16(input16_, value, kInputBufferSize);
+  }
+
+  void CopyOutputToRef() {
+    memcpy(output_ref_, output_, kOutputBufferSize);
+    // Copy 16-bit pixels values. The effective number of bytes is double.
+    memcpy(output16_ref_, output16_, sizeof(output16_[0]) * kOutputBufferSize);
+  }
+
+  void CheckGuardBlocks() {
+    for (int i = 0; i < kOutputBufferSize; ++i) {
+      if (IsIndexInBorder(i)) {
+        EXPECT_EQ(255, output_[i]);
+      }
+    }
+  }
+
+  uint8_t *input() const {
+    const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
+    if (UUT_->use_highbd_ == 0) {
+      return input_ + offset;
+    } else {
+      return CONVERT_TO_BYTEPTR(input16_) + offset;
+    }
+  }
+
+  uint8_t *output() const {
+    const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
+    if (UUT_->use_highbd_ == 0) {
+      return output_ + offset;
+    } else {
+      return CONVERT_TO_BYTEPTR(output16_) + offset;
+    }
+  }
+
+  uint8_t *output_ref() const {
+    const int offset = BorderTop() * kOuterBlockSize + BorderLeft();
+    if (UUT_->use_highbd_ == 0) {
+      return output_ref_ + offset;
+    } else {
+      return CONVERT_TO_BYTEPTR(output16_ref_) + offset;
+    }
+  }
+
+  uint16_t lookup(uint8_t *list, int index) const {
+    if (UUT_->use_highbd_ == 0) {
+      return list[index];
+    } else {
+      return CONVERT_TO_SHORTPTR(list)[index];
+    }
+  }
+
+  void assign_val(uint8_t *list, int index, uint16_t val) const {
+    if (UUT_->use_highbd_ == 0) {
+      list[index] = (uint8_t)val;
+    } else {
+      CONVERT_TO_SHORTPTR(list)[index] = val;
+    }
+  }
+
+  void wrapper_filter_average_block2d_8_c(
+      const uint8_t *src_ptr, unsigned int src_stride, const int16_t *HFilter,
+      const int16_t *VFilter, uint8_t *dst_ptr, unsigned int dst_stride,
+      unsigned int output_width, unsigned int output_height) {
+    if (UUT_->use_highbd_ == 0) {
+      filter_average_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, dst_ptr,
+                                 dst_stride, output_width, output_height);
+    } else {
+      highbd_filter_average_block2d_8_c(
+          CONVERT_TO_SHORTPTR(src_ptr), src_stride, HFilter, VFilter,
+          CONVERT_TO_SHORTPTR(dst_ptr), dst_stride, output_width, output_height,
+          UUT_->use_highbd_);
+    }
+  }
+
+  void wrapper_filter_block2d_8_c(
+      const uint8_t *src_ptr, unsigned int src_stride, const int16_t *HFilter,
+      const int16_t *VFilter, uint8_t *dst_ptr, unsigned int dst_stride,
+      unsigned int output_width, unsigned int output_height) {
+    if (UUT_->use_highbd_ == 0) {
+      filter_block2d_8_c(src_ptr, src_stride, HFilter, VFilter, dst_ptr,
+                         dst_stride, output_width, output_height);
+    } else {
+      highbd_filter_block2d_8_c(CONVERT_TO_SHORTPTR(src_ptr), src_stride,
+                                HFilter, VFilter, CONVERT_TO_SHORTPTR(dst_ptr),
+                                dst_stride, output_width, output_height,
+                                UUT_->use_highbd_);
+    }
+  }
+
+  const ConvolveFunctions *UUT_;
+  static uint8_t *input_;
+  static uint8_t *output_;
+  static uint8_t *output_ref_;
+  static uint16_t *input16_;
+  static uint16_t *output16_;
+  static uint16_t *output16_ref_;
+  int mask_;
+};
+
+uint8_t *ConvolveTest::input_ = NULL;
+uint8_t *ConvolveTest::output_ = NULL;
+uint8_t *ConvolveTest::output_ref_ = NULL;
+uint16_t *ConvolveTest::input16_ = NULL;
+uint16_t *ConvolveTest::output16_ = NULL;
+uint16_t *ConvolveTest::output16_ref_ = NULL;
+
+TEST_P(ConvolveTest, GuardBlocks) { CheckGuardBlocks(); }
+
+TEST_P(ConvolveTest, Copy) {
+  uint8_t *const in = input();
+  uint8_t *const out = output();
+
+  ASM_REGISTER_STATE_CHECK(UUT_->copy_(in, kInputStride, out, kOutputStride,
+                                       NULL, 0, NULL, 0, Width(), Height()));
+
+  CheckGuardBlocks();
+
+  for (int y = 0; y < Height(); ++y)
+    for (int x = 0; x < Width(); ++x)
+      ASSERT_EQ(lookup(out, y * kOutputStride + x),
+                lookup(in, y * kInputStride + x))
+          << "(" << x << "," << y << ")";
+}
+
+const int kNumFilterBanks = SWITCHABLE_FILTERS;
+const int kNumFilters = 16;
+
+TEST(ConvolveTest, FiltersWontSaturateWhenAddedPairwise) {
+  for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+    const InterpFilter filter = (InterpFilter)filter_bank;
+    const InterpKernel *filters =
+        (const InterpKernel *)av1_get_interp_filter_kernel(filter);
+    const InterpFilterParams *filter_params =
+        av1_get_interp_filter_params_with_block_size(filter, 8);
+    if (filter_params->taps != SUBPEL_TAPS) continue;
+    for (int i = 0; i < kNumFilters; i++) {
+      const int p0 = filters[i][0] + filters[i][1];
+      const int p1 = filters[i][2] + filters[i][3];
+      const int p2 = filters[i][4] + filters[i][5];
+      const int p3 = filters[i][6] + filters[i][7];
+      EXPECT_LE(p0, 128);
+      EXPECT_LE(p1, 128);
+      EXPECT_LE(p2, 128);
+      EXPECT_LE(p3, 128);
+      EXPECT_LE(p0 + p3, 128);
+      EXPECT_LE(p0 + p3 + p1, 128);
+      EXPECT_LE(p0 + p3 + p1 + p2, 128);
+      EXPECT_EQ(p0 + p1 + p2 + p3, 128);
+    }
+  }
+}
+
+const int16_t kInvalidFilter[8] = { 0 };
+
+TEST_P(ConvolveTest, MatchesReferenceSubpixelFilter) {
+  uint8_t *const in = input();
+  uint8_t *const out = output();
+  uint8_t ref8[kOutputStride * kMaxDimension];
+  uint16_t ref16[kOutputStride * kMaxDimension];
+  uint8_t *ref;
+  if (UUT_->use_highbd_ == 0) {
+    ref = ref8;
+  } else {
+    ref = CONVERT_TO_BYTEPTR(ref16);
+  }
+
+  for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+    const InterpFilter filter = (InterpFilter)filter_bank;
+    const InterpKernel *filters =
+        (const InterpKernel *)av1_get_interp_filter_kernel(filter);
+    const InterpFilterParams *filter_params =
+        av1_get_interp_filter_params_with_block_size(filter, 8);
+    if (filter_params->taps != SUBPEL_TAPS) continue;
+
+    for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
+      for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
+        wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
+                                   filters[filter_y], ref, kOutputStride,
+                                   Width(), Height());
+
+        if (filter_x && filter_y)
+          continue;
+        else if (filter_y)
+          ASM_REGISTER_STATE_CHECK(
+              UUT_->v8_(in, kInputStride, out, kOutputStride, kInvalidFilter,
+                        16, filters[filter_y], 16, Width(), Height()));
+        else if (filter_x)
+          ASM_REGISTER_STATE_CHECK(
+              UUT_->h8_(in, kInputStride, out, kOutputStride, filters[filter_x],
+                        16, kInvalidFilter, 16, Width(), Height()));
+        else
+          ASM_REGISTER_STATE_CHECK(
+              UUT_->copy_(in, kInputStride, out, kOutputStride, kInvalidFilter,
+                          0, kInvalidFilter, 0, Width(), Height()));
+
+        CheckGuardBlocks();
+
+        for (int y = 0; y < Height(); ++y)
+          for (int x = 0; x < Width(); ++x)
+            ASSERT_EQ(lookup(ref, y * kOutputStride + x),
+                      lookup(out, y * kOutputStride + x))
+                << "mismatch at (" << x << "," << y << "), "
+                << "filters (" << filter_bank << "," << filter_x << ","
+                << filter_y << ")";
+      }
+    }
+  }
+}
+
+TEST_P(ConvolveTest, FilterExtremes) {
+  uint8_t *const in = input();
+  uint8_t *const out = output();
+  uint8_t ref8[kOutputStride * kMaxDimension];
+  uint16_t ref16[kOutputStride * kMaxDimension];
+  uint8_t *ref;
+  if (UUT_->use_highbd_ == 0) {
+    ref = ref8;
+  } else {
+    ref = CONVERT_TO_BYTEPTR(ref16);
+  }
+
+  // Populate ref and out with some random data
+  ::libaom_test::ACMRandom prng;
+  for (int y = 0; y < Height(); ++y) {
+    for (int x = 0; x < Width(); ++x) {
+      uint16_t r;
+      if (UUT_->use_highbd_ == 0 || UUT_->use_highbd_ == 8) {
+        r = prng.Rand8Extremes();
+      } else {
+        r = prng.Rand16() & mask_;
+      }
+      assign_val(out, y * kOutputStride + x, r);
+      assign_val(ref, y * kOutputStride + x, r);
+    }
+  }
+
+  for (int axis = 0; axis < 2; axis++) {
+    int seed_val = 0;
+    while (seed_val < 256) {
+      for (int y = 0; y < 8; ++y) {
+        for (int x = 0; x < 8; ++x) {
+          assign_val(in, y * kOutputStride + x - SUBPEL_TAPS / 2 + 1,
+                     ((seed_val >> (axis ? y : x)) & 1) * mask_);
+          if (axis) seed_val++;
+        }
+        if (axis)
+          seed_val -= 8;
+        else
+          seed_val++;
+      }
+      if (axis) seed_val += 8;
+
+      for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+        const InterpFilter filter = (InterpFilter)filter_bank;
+        const InterpKernel *filters =
+            (const InterpKernel *)av1_get_interp_filter_kernel(filter);
+        const InterpFilterParams *filter_params =
+            av1_get_interp_filter_params_with_block_size(filter, 8);
+        if (filter_params->taps != SUBPEL_TAPS) continue;
+        for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
+          for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
+            wrapper_filter_block2d_8_c(in, kInputStride, filters[filter_x],
+                                       filters[filter_y], ref, kOutputStride,
+                                       Width(), Height());
+            if (filter_x && filter_y)
+              continue;
+            else if (filter_y)
+              ASM_REGISTER_STATE_CHECK(UUT_->v8_(
+                  in, kInputStride, out, kOutputStride, kInvalidFilter, 16,
+                  filters[filter_y], 16, Width(), Height()));
+            else if (filter_x)
+              ASM_REGISTER_STATE_CHECK(UUT_->h8_(
+                  in, kInputStride, out, kOutputStride, filters[filter_x], 16,
+                  kInvalidFilter, 16, Width(), Height()));
+            else
+              ASM_REGISTER_STATE_CHECK(UUT_->copy_(
+                  in, kInputStride, out, kOutputStride, kInvalidFilter, 0,
+                  kInvalidFilter, 0, Width(), Height()));
+
+            for (int y = 0; y < Height(); ++y)
+              for (int x = 0; x < Width(); ++x)
+                ASSERT_EQ(lookup(ref, y * kOutputStride + x),
+                          lookup(out, y * kOutputStride + x))
+                    << "mismatch at (" << x << "," << y << "), "
+                    << "filters (" << filter_bank << "," << filter_x << ","
+                    << filter_y << ")";
+          }
+        }
+      }
+    }
+  }
+}
+
+TEST_P(ConvolveTest, DISABLED_Copy_Speed) {
+  const uint8_t *const in = input();
+  uint8_t *const out = output();
+  const int kNumTests = 5000000;
+  const int width = Width();
+  const int height = Height();
+  aom_usec_timer timer;
+
+  aom_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    UUT_->copy_(in, kInputStride, out, kOutputStride, NULL, 0, NULL, 0, width,
+                height);
+  }
+  aom_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+  printf("convolve_copy_%dx%d_%d: %d us\n", width, height,
+         UUT_->use_highbd_ ? UUT_->use_highbd_ : 8, elapsed_time);
+}
+
+TEST_P(ConvolveTest, DISABLED_Speed) {
+  uint8_t *const in = input();
+  uint8_t *const out = output();
+  uint8_t ref8[kOutputStride * kMaxDimension];
+  uint16_t ref16[kOutputStride * kMaxDimension];
+  uint8_t *ref;
+  if (UUT_->use_highbd_ == 0) {
+    ref = ref8;
+  } else {
+    ref = CONVERT_TO_BYTEPTR(ref16);
+  }
+
+  // Populate ref and out with some random data
+  ::libaom_test::ACMRandom prng;
+  for (int y = 0; y < Height(); ++y) {
+    for (int x = 0; x < Width(); ++x) {
+      uint16_t r;
+      if (UUT_->use_highbd_ == 0 || UUT_->use_highbd_ == 8) {
+        r = prng.Rand8Extremes();
+      } else {
+        r = prng.Rand16() & mask_;
+      }
+      assign_val(out, y * kOutputStride + x, r);
+      assign_val(ref, y * kOutputStride + x, r);
+    }
+  }
+
+  const InterpFilter filter = (InterpFilter)1;
+  const InterpKernel *filters =
+      (const InterpKernel *)av1_get_interp_filter_kernel(filter);
+  wrapper_filter_average_block2d_8_c(in, kInputStride, filters[1], filters[1],
+                                     out, kOutputStride, Width(), Height());
+
+  aom_usec_timer timer;
+  int tests_num = 1000;
+
+  aom_usec_timer_start(&timer);
+  while (tests_num > 0) {
+    for (int filter_bank = 0; filter_bank < kNumFilterBanks; ++filter_bank) {
+      const InterpFilter filter = (InterpFilter)filter_bank;
+      const InterpKernel *filters =
+          (const InterpKernel *)av1_get_interp_filter_kernel(filter);
+      const InterpFilterParams *filter_params =
+          av1_get_interp_filter_params_with_block_size(filter, 8);
+      if (filter_params->taps != SUBPEL_TAPS) continue;
+
+      for (int filter_x = 0; filter_x < kNumFilters; ++filter_x) {
+        for (int filter_y = 0; filter_y < kNumFilters; ++filter_y) {
+          if (filter_x && filter_y) continue;
+          if (filter_y)
+            ASM_REGISTER_STATE_CHECK(
+                UUT_->v8_(in, kInputStride, out, kOutputStride, kInvalidFilter,
+                          16, filters[filter_y], 16, Width(), Height()));
+          else if (filter_x)
+            ASM_REGISTER_STATE_CHECK(UUT_->h8_(
+                in, kInputStride, out, kOutputStride, filters[filter_x], 16,
+                kInvalidFilter, 16, Width(), Height()));
+        }
+      }
+    }
+    tests_num--;
+  }
+  aom_usec_timer_mark(&timer);
+
+  const int elapsed_time =
+      static_cast<int>(aom_usec_timer_elapsed(&timer) / 1000);
+  printf("%dx%d (bitdepth %d) time: %5d ms\n", Width(), Height(),
+         UUT_->use_highbd_, elapsed_time);
+}
+
+using ::testing::make_tuple;
+
+#define WRAP(func, bd)                                                       \
+  static void wrap_##func##_##bd(                                            \
+      const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,                \
+      ptrdiff_t dst_stride, const int16_t *filter_x, int filter_x_stride,    \
+      const int16_t *filter_y, int filter_y_stride, int w, int h) {          \
+    aom_highbd_##func(src, src_stride, dst, dst_stride, filter_x,            \
+                      filter_x_stride, filter_y, filter_y_stride, w, h, bd); \
+  }
+#if HAVE_SSE2 && ARCH_X86_64
+WRAP(convolve_copy_sse2, 8)
+WRAP(convolve_copy_sse2, 10)
+WRAP(convolve_copy_sse2, 12)
+WRAP(convolve8_horiz_sse2, 8)
+WRAP(convolve8_vert_sse2, 8)
+WRAP(convolve8_horiz_sse2, 10)
+WRAP(convolve8_vert_sse2, 10)
+WRAP(convolve8_horiz_sse2, 12)
+WRAP(convolve8_vert_sse2, 12)
+#endif  // HAVE_SSE2 && ARCH_X86_64
+
+WRAP(convolve_copy_c, 8)
+WRAP(convolve8_horiz_c, 8)
+WRAP(convolve8_vert_c, 8)
+WRAP(convolve_copy_c, 10)
+WRAP(convolve8_horiz_c, 10)
+WRAP(convolve8_vert_c, 10)
+WRAP(convolve_copy_c, 12)
+WRAP(convolve8_horiz_c, 12)
+WRAP(convolve8_vert_c, 12)
+
+#if HAVE_AVX2
+WRAP(convolve_copy_avx2, 8)
+WRAP(convolve8_horiz_avx2, 8)
+WRAP(convolve8_vert_avx2, 8)
+
+WRAP(convolve_copy_avx2, 10)
+WRAP(convolve8_horiz_avx2, 10)
+WRAP(convolve8_vert_avx2, 10)
+
+WRAP(convolve_copy_avx2, 12)
+WRAP(convolve8_horiz_avx2, 12)
+WRAP(convolve8_vert_avx2, 12)
+#endif  // HAVE_AVX2
+
+#undef WRAP
+
+const ConvolveFunctions convolve8_c(wrap_convolve_copy_c_8,
+                                    wrap_convolve8_horiz_c_8,
+                                    wrap_convolve8_vert_c_8, 8);
+const ConvolveFunctions convolve10_c(wrap_convolve_copy_c_10,
+                                     wrap_convolve8_horiz_c_10,
+                                     wrap_convolve8_vert_c_10, 10);
+const ConvolveFunctions convolve12_c(wrap_convolve_copy_c_12,
+                                     wrap_convolve8_horiz_c_12,
+                                     wrap_convolve8_vert_c_12, 12);
+const ConvolveParam kArrayConvolve_c[] = {
+  ALL_SIZES(convolve8_c), ALL_SIZES(convolve10_c), ALL_SIZES(convolve12_c)
+};
+
+INSTANTIATE_TEST_CASE_P(C, ConvolveTest, ::testing::ValuesIn(kArrayConvolve_c));
+
+#if HAVE_SSE2 && ARCH_X86_64
+const ConvolveFunctions convolve8_sse2(wrap_convolve_copy_sse2_8,
+                                       wrap_convolve8_horiz_sse2_8,
+                                       wrap_convolve8_vert_sse2_8, 8);
+const ConvolveFunctions convolve10_sse2(wrap_convolve_copy_sse2_10,
+                                        wrap_convolve8_horiz_sse2_10,
+                                        wrap_convolve8_vert_sse2_10, 10);
+const ConvolveFunctions convolve12_sse2(wrap_convolve_copy_sse2_12,
+                                        wrap_convolve8_horiz_sse2_12,
+                                        wrap_convolve8_vert_sse2_12, 12);
+const ConvolveParam kArrayConvolve_sse2[] = { ALL_SIZES(convolve8_sse2),
+                                              ALL_SIZES(convolve10_sse2),
+                                              ALL_SIZES(convolve12_sse2) };
+INSTANTIATE_TEST_CASE_P(SSE2, ConvolveTest,
+                        ::testing::ValuesIn(kArrayConvolve_sse2));
+#endif
+
+#if HAVE_SSSE3
+const ConvolveFunctions convolve8_ssse3(aom_convolve_copy_c,
+                                        aom_convolve8_horiz_ssse3,
+                                        aom_convolve8_vert_ssse3, 0);
+
+const ConvolveParam kArrayConvolve8_ssse3[] = { ALL_SIZES(convolve8_ssse3) };
+INSTANTIATE_TEST_CASE_P(SSSE3, ConvolveTest,
+                        ::testing::ValuesIn(kArrayConvolve8_ssse3));
+#endif
+
+#if HAVE_AVX2
+const ConvolveFunctions convolve8_avx2(aom_convolve_copy_c,
+                                       aom_convolve8_horiz_avx2,
+                                       aom_convolve8_vert_avx2, 0);
+
+const ConvolveFunctions wrap_convolve8_avx2(wrap_convolve_copy_avx2_8,
+                                            wrap_convolve8_horiz_avx2_8,
+                                            wrap_convolve8_vert_avx2_8, 8);
+const ConvolveFunctions wrap_convolve10_avx2(wrap_convolve_copy_avx2_10,
+                                             wrap_convolve8_horiz_avx2_10,
+                                             wrap_convolve8_vert_avx2_10, 10);
+const ConvolveFunctions wrap_convolve12_avx2(wrap_convolve_copy_avx2_12,
+                                             wrap_convolve8_horiz_avx2_12,
+                                             wrap_convolve8_vert_avx2_12, 12);
+const ConvolveParam kArray_Convolve8_avx2[] = {
+  ALL_SIZES_64(wrap_convolve8_avx2), ALL_SIZES_64(wrap_convolve10_avx2),
+  ALL_SIZES_64(wrap_convolve12_avx2), ALL_SIZES(convolve8_avx2)
+};
+INSTANTIATE_TEST_CASE_P(AVX2, ConvolveTest,
+                        ::testing::ValuesIn(kArray_Convolve8_avx2));
+#endif  // HAVE_AVX2
+
+}  // namespace
diff --git a/third_party/aom/test/corner_match_test.cc b/third_party/aom/test/corner_match_test.cc
new file mode 100644
index 0000000000..58e3139c5f
--- /dev/null
+++ b/third_party/aom/test/corner_match_test.cc
@@ -0,0 +1,100 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include "config/av1_rtcd.h"
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/util.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+
+#include "av1/encoder/corner_match.h"
+
+namespace test_libaom {
+
+namespace AV1CornerMatch {
+
+using libaom_test::ACMRandom;
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+typedef tuple<int> CornerMatchParam;
+
+class AV1CornerMatchTest : public ::testing::TestWithParam<CornerMatchParam> {
+ public:
+  virtual ~AV1CornerMatchTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput();
+
+  libaom_test::ACMRandom rnd_;
+};
+
+AV1CornerMatchTest::~AV1CornerMatchTest() {}
+void AV1CornerMatchTest::SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+void AV1CornerMatchTest::TearDown() { libaom_test::ClearSystemState(); }
+
+void AV1CornerMatchTest::RunCheckOutput() {
+  const int w = 128, h = 128;
+  const int num_iters = 10000;
+  int i, j;
+
+  uint8_t *input1 = new uint8_t[w * h];
+  uint8_t *input2 = new uint8_t[w * h];
+
+  // Test the two extreme cases:
+  // i) Random data, should have correlation close to 0
+  // ii) Linearly related data + noise, should have correlation close to 1
+  int mode = GET_PARAM(0);
+  if (mode == 0) {
+    for (i = 0; i < h; ++i)
+      for (j = 0; j < w; ++j) {
+        input1[i * w + j] = rnd_.Rand8();
+        input2[i * w + j] = rnd_.Rand8();
+      }
+  } else if (mode == 1) {
+    for (i = 0; i < h; ++i)
+      for (j = 0; j < w; ++j) {
+        int v = rnd_.Rand8();
+        input1[i * w + j] = v;
+        input2[i * w + j] = (v / 2) + (rnd_.Rand8() & 15);
+      }
+  }
+
+  for (i = 0; i < num_iters; ++i) {
+    int x1 = MATCH_SZ_BY2 + rnd_.PseudoUniform(w - 2 * MATCH_SZ_BY2);
+    int y1 = MATCH_SZ_BY2 + rnd_.PseudoUniform(h - 2 * MATCH_SZ_BY2);
+    int x2 = MATCH_SZ_BY2 + rnd_.PseudoUniform(w - 2 * MATCH_SZ_BY2);
+    int y2 = MATCH_SZ_BY2 + rnd_.PseudoUniform(h - 2 * MATCH_SZ_BY2);
+
+    double res_c =
+        compute_cross_correlation_c(input1, w, x1, y1, input2, w, x2, y2);
+    double res_sse4 =
+        compute_cross_correlation_sse4_1(input1, w, x1, y1, input2, w, x2, y2);
+
+    ASSERT_EQ(res_sse4, res_c);
+  }
+
+  delete[] input1;
+  delete[] input2;
+}
+
+TEST_P(AV1CornerMatchTest, CheckOutput) { RunCheckOutput(); }
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1CornerMatchTest,
+                        ::testing::Values(make_tuple(0), make_tuple(1)));
+
+}  // namespace AV1CornerMatch
+
+}  // namespace test_libaom
diff --git a/third_party/aom/test/cpu_speed_test.cc b/third_party/aom/test/cpu_speed_test.cc
new file mode 100644
index 0000000000..8ea3e69650
--- /dev/null
+++ b/third_party/aom/test/cpu_speed_test.cc
@@ -0,0 +1,180 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+
+namespace {
+
+const int kMaxPSNR = 100;
+
+class CpuSpeedTest
+    : public ::libaom_test::CodecTestWith2Params<libaom_test::TestMode, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  CpuSpeedTest()
+      : EncoderTest(GET_PARAM(0)), encoding_mode_(GET_PARAM(1)),
+        set_cpu_used_(GET_PARAM(2)), min_psnr_(kMaxPSNR),
+        tune_content_(AOM_CONTENT_DEFAULT) {}
+  virtual ~CpuSpeedTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+    if (encoding_mode_ != ::libaom_test::kRealTime) {
+      cfg_.g_lag_in_frames = 25;
+      cfg_.rc_end_usage = AOM_VBR;
+    } else {
+      cfg_.g_lag_in_frames = 0;
+      cfg_.rc_end_usage = AOM_CBR;
+    }
+  }
+
+  virtual void BeginPassHook(unsigned int /*pass*/) { min_psnr_ = kMaxPSNR; }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
+      encoder->Control(AV1E_SET_TUNE_CONTENT, tune_content_);
+      if (encoding_mode_ != ::libaom_test::kRealTime) {
+        encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+        encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+        encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+      }
+    }
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (pkt->data.psnr.psnr[0] < min_psnr_) min_psnr_ = pkt->data.psnr.psnr[0];
+  }
+
+  void TestQ0();
+  void TestScreencastQ0();
+  void TestTuneScreen();
+  void TestEncodeHighBitrate();
+  void TestLowBitrate();
+
+  ::libaom_test::TestMode encoding_mode_;
+  int set_cpu_used_;
+  double min_psnr_;
+  int tune_content_;
+};
+
+void CpuSpeedTest::TestQ0() {
+  // Validate that this non multiple of 64 wide clip encodes and decodes
+  // without a mismatch when passing in a very low max q.  This pushes
+  // the encoder to producing lots of big partitions which will likely
+  // extend into the border and test the border condition.
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_maxsection_pct = 2000;
+  cfg_.rc_target_bitrate = 400;
+  cfg_.rc_max_quantizer = 0;
+  cfg_.rc_min_quantizer = 0;
+
+  ::libaom_test::I420VideoSource video("hantro_odd.yuv", 208, 144, 30, 1, 0,
+                                       10);
+
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  EXPECT_GE(min_psnr_, kMaxPSNR);
+}
+
+void CpuSpeedTest::TestScreencastQ0() {
+  ::libaom_test::Y4mVideoSource video("screendata.y4m", 0, 3);
+  cfg_.g_timebase = video.timebase();
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_maxsection_pct = 2000;
+  cfg_.rc_target_bitrate = 400;
+  cfg_.rc_max_quantizer = 0;
+  cfg_.rc_min_quantizer = 0;
+
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  EXPECT_GE(min_psnr_, kMaxPSNR);
+}
+
+void CpuSpeedTest::TestTuneScreen() {
+  ::libaom_test::Y4mVideoSource video("screendata.y4m", 0, 3);
+  cfg_.g_timebase = video.timebase();
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_minsection_pct = 2000;
+  cfg_.rc_target_bitrate = 2000;
+  cfg_.rc_max_quantizer = 63;
+  cfg_.rc_min_quantizer = 0;
+  tune_content_ = AOM_CONTENT_SCREEN;
+
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
+void CpuSpeedTest::TestEncodeHighBitrate() {
+  // Validate that this non multiple of 64 wide clip encodes and decodes
+  // without a mismatch when passing in a very low max q.  This pushes
+  // the encoder to producing lots of big partitions which will likely
+  // extend into the border and test the border condition.
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_maxsection_pct = 2000;
+  cfg_.rc_target_bitrate = 12000;
+  cfg_.rc_max_quantizer = 10;
+  cfg_.rc_min_quantizer = 0;
+
+  ::libaom_test::I420VideoSource video("hantro_odd.yuv", 208, 144, 30, 1, 0,
+                                       10);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
+void CpuSpeedTest::TestLowBitrate() {
+  // Validate that this clip encodes and decodes without a mismatch
+  // when passing in a very high min q.  This pushes the encoder to producing
+  // lots of small partitions which might will test the other condition.
+  cfg_.rc_2pass_vbr_minsection_pct = 5;
+  cfg_.rc_2pass_vbr_maxsection_pct = 2000;
+  cfg_.rc_target_bitrate = 200;
+  cfg_.rc_min_quantizer = 40;
+
+  ::libaom_test::I420VideoSource video("hantro_odd.yuv", 208, 144, 30, 1, 0,
+                                       10);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
+TEST_P(CpuSpeedTest, TestQ0) { TestQ0(); }
+TEST_P(CpuSpeedTest, TestScreencastQ0) { TestScreencastQ0(); }
+TEST_P(CpuSpeedTest, TestTuneScreen) { TestTuneScreen(); }
+TEST_P(CpuSpeedTest, TestEncodeHighBitrate) { TestEncodeHighBitrate(); }
+TEST_P(CpuSpeedTest, TestLowBitrate) { TestLowBitrate(); }
+
+class CpuSpeedTestLarge : public CpuSpeedTest {};
+
+TEST_P(CpuSpeedTestLarge, TestQ0) { TestQ0(); }
+TEST_P(CpuSpeedTestLarge, TestScreencastQ0) { TestScreencastQ0(); }
+TEST_P(CpuSpeedTestLarge, TestTuneScreen) { TestTuneScreen(); }
+TEST_P(CpuSpeedTestLarge, TestEncodeHighBitrate) { TestEncodeHighBitrate(); }
+TEST_P(CpuSpeedTestLarge, TestLowBitrate) { TestLowBitrate(); }
+
+AV1_INSTANTIATE_TEST_CASE(CpuSpeedTest,
+                          ::testing::Values(::libaom_test::kTwoPassGood,
+                                            ::libaom_test::kOnePassGood),
+                          ::testing::Range(1, 3));
+AV1_INSTANTIATE_TEST_CASE(CpuSpeedTestLarge,
+                          ::testing::Values(::libaom_test::kTwoPassGood,
+                                            ::libaom_test::kOnePassGood),
+                          ::testing::Range(0, 1));
+}  // namespace
diff --git a/third_party/aom/test/datarate_test.cc b/third_party/aom/test/datarate_test.cc
new file mode 100644
index 0000000000..1588d3cc19
--- /dev/null
+++ b/third_party/aom/test/datarate_test.cc
@@ -0,0 +1,255 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "config/aom_config.h"
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+#include "aom/aom_codec.h"
+
+namespace {
+
+class DatarateTestLarge
+    : public ::libaom_test::CodecTestWith2Params<libaom_test::TestMode, int>,
+      public ::libaom_test::EncoderTest {
+ public:
+  DatarateTestLarge() : EncoderTest(GET_PARAM(0)) {}
+
+ protected:
+  virtual ~DatarateTestLarge() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+    set_cpu_used_ = GET_PARAM(2);
+    ResetModel();
+  }
+
+  virtual void ResetModel() {
+    last_pts_ = 0;
+    bits_in_buffer_model_ = cfg_.rc_target_bitrate * cfg_.rc_buf_initial_sz;
+    frame_number_ = 0;
+    tot_frame_number_ = 0;
+    first_drop_ = 0;
+    num_drops_ = 0;
+    // Denoiser is off by default.
+    denoiser_on_ = 0;
+    bits_total_ = 0;
+    denoiser_offon_test_ = 0;
+    denoiser_offon_period_ = -1;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 0) encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
+
+    if (denoiser_offon_test_) {
+      ASSERT_GT(denoiser_offon_period_, 0)
+          << "denoiser_offon_period_ is not positive.";
+      if ((video->frame() + 1) % denoiser_offon_period_ == 0) {
+        // Flip denoiser_on_ periodically
+        denoiser_on_ ^= 1;
+      }
+    }
+
+    encoder->Control(AV1E_SET_NOISE_SENSITIVITY, denoiser_on_);
+
+    const aom_rational_t tb = video->timebase();
+    timebase_ = static_cast<double>(tb.num) / tb.den;
+    duration_ = 0;
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    // Time since last timestamp = duration.
+    aom_codec_pts_t duration = pkt->data.frame.pts - last_pts_;
+
+    if (duration > 1) {
+      // If first drop not set and we have a drop set it to this time.
+      if (!first_drop_) first_drop_ = last_pts_ + 1;
+      // Update the number of frame drops.
+      num_drops_ += static_cast<int>(duration - 1);
+      // Update counter for total number of frames (#frames input to encoder).
+      // Needed for setting the proper layer_id below.
+      tot_frame_number_ += static_cast<int>(duration - 1);
+    }
+
+    // Add to the buffer the bits we'd expect from a constant bitrate server.
+    bits_in_buffer_model_ += static_cast<int64_t>(
+        duration * timebase_ * cfg_.rc_target_bitrate * 1000);
+
+    // Buffer should not go negative.
+    ASSERT_GE(bits_in_buffer_model_, 0)
+        << "Buffer Underrun at frame " << pkt->data.frame.pts;
+
+    const size_t frame_size_in_bits = pkt->data.frame.sz * 8;
+
+    // Update the total encoded bits.
+    bits_total_ += frame_size_in_bits;
+
+    // Update the most recent pts.
+    last_pts_ = pkt->data.frame.pts;
+    ++frame_number_;
+    ++tot_frame_number_;
+  }
+
+  virtual void EndPassHook(void) {
+    duration_ = (last_pts_ + 1) * timebase_;
+    // Effective file datarate:
+    effective_datarate_ = (bits_total_ / 1000.0) / duration_;
+  }
+
+  aom_codec_pts_t last_pts_;
+  double timebase_;
+  int frame_number_;      // Counter for number of non-dropped/encoded frames.
+  int tot_frame_number_;  // Counter for total number of input frames.
+  int64_t bits_total_;
+  double duration_;
+  double effective_datarate_;
+  int set_cpu_used_;
+  int64_t bits_in_buffer_model_;
+  aom_codec_pts_t first_drop_;
+  int num_drops_;
+  int denoiser_on_;
+  int denoiser_offon_test_;
+  int denoiser_offon_period_;
+};
+
+// Check basic rate targeting for VBR mode.
+TEST_P(DatarateTestLarge, BasicRateTargetingVBR) {
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 63;
+  cfg_.g_error_resilient = 0;
+  cfg_.rc_end_usage = AOM_VBR;
+  cfg_.g_lag_in_frames = 0;
+
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 140);
+  for (int i = 400; i <= 800; i += 400) {
+    cfg_.rc_target_bitrate = i;
+    ResetModel();
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    ASSERT_GE(effective_datarate_, cfg_.rc_target_bitrate * 0.75)
+        << " The datarate for the file is lower than target by too much!";
+    ASSERT_LE(effective_datarate_, cfg_.rc_target_bitrate * 1.25)
+        << " The datarate for the file is greater than target by too much!";
+  }
+}
+
+// Check basic rate targeting for CBR,
+TEST_P(DatarateTestLarge, BasicRateTargeting) {
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 500;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_dropframe_thresh = 1;
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 63;
+  cfg_.rc_end_usage = AOM_CBR;
+  cfg_.g_lag_in_frames = 0;
+
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 140);
+  for (int i = 150; i < 800; i += 400) {
+    cfg_.rc_target_bitrate = i;
+    ResetModel();
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    ASSERT_GE(effective_datarate_, cfg_.rc_target_bitrate * 0.85)
+        << " The datarate for the file is lower than target by too much!";
+    ASSERT_LE(effective_datarate_, cfg_.rc_target_bitrate * 1.15)
+        << " The datarate for the file is greater than target by too much!";
+  }
+}
+
+// Check basic rate targeting for CBR.
+TEST_P(DatarateTestLarge, BasicRateTargeting444) {
+  ::libaom_test::Y4mVideoSource video("rush_hour_444.y4m", 0, 140);
+
+  cfg_.g_profile = 1;
+  cfg_.g_timebase = video.timebase();
+
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 500;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_dropframe_thresh = 1;
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 63;
+  cfg_.rc_end_usage = AOM_CBR;
+
+  for (int i = 250; i < 900; i += 400) {
+    cfg_.rc_target_bitrate = i;
+    ResetModel();
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    ASSERT_GE(static_cast<double>(cfg_.rc_target_bitrate),
+              effective_datarate_ * 0.85)
+        << " The datarate for the file exceeds the target by too much!";
+    ASSERT_LE(static_cast<double>(cfg_.rc_target_bitrate),
+              effective_datarate_ * 1.15)
+        << " The datarate for the file missed the target!"
+        << cfg_.rc_target_bitrate << " " << effective_datarate_;
+  }
+}
+
+// Check that (1) the first dropped frame gets earlier and earlier
+// as the drop frame threshold is increased, and (2) that the total number of
+// frame drops does not decrease as we increase frame drop threshold.
+// Use a lower qp-max to force some frame drops.
+TEST_P(DatarateTestLarge, ChangingDropFrameThresh) {
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 500;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_undershoot_pct = 20;
+  cfg_.rc_undershoot_pct = 20;
+  cfg_.rc_dropframe_thresh = 10;
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 50;
+  cfg_.rc_end_usage = AOM_CBR;
+  cfg_.rc_target_bitrate = 200;
+  cfg_.g_lag_in_frames = 0;
+  cfg_.g_error_resilient = 1;
+  // TODO(marpan): Investigate datarate target failures with a smaller keyframe
+  // interval (128).
+  cfg_.kf_max_dist = 9999;
+
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 100);
+
+  const int kDropFrameThreshTestStep = 30;
+  aom_codec_pts_t last_drop = 140;
+  int last_num_drops = 0;
+  for (int i = 40; i < 100; i += kDropFrameThreshTestStep) {
+    cfg_.rc_dropframe_thresh = i;
+    ResetModel();
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    ASSERT_GE(effective_datarate_, cfg_.rc_target_bitrate * 0.85)
+        << " The datarate for the file is lower than target by too much!";
+    ASSERT_LE(effective_datarate_, cfg_.rc_target_bitrate * 1.15)
+        << " The datarate for the file is greater than target by too much!";
+    ASSERT_LE(first_drop_, last_drop)
+        << " The first dropped frame for drop_thresh " << i
+        << " > first dropped frame for drop_thresh "
+        << i - kDropFrameThreshTestStep;
+    ASSERT_GE(num_drops_, last_num_drops * 0.85)
+        << " The number of dropped frames for drop_thresh " << i
+        << " < number of dropped frames for drop_thresh "
+        << i - kDropFrameThreshTestStep;
+    last_drop = first_drop_;
+    last_num_drops = num_drops_;
+  }
+}
+
+AV1_INSTANTIATE_TEST_CASE(DatarateTestLarge,
+                          ::testing::Values(::libaom_test::kOnePassGood,
+                                            ::libaom_test::kRealTime),
+                          ::testing::Values(2, 5));
+}  // namespace
diff --git a/third_party/aom/test/decode_api_test.cc b/third_party/aom/test/decode_api_test.cc
new file mode 100644
index 0000000000..c1beacee17
--- /dev/null
+++ b/third_party/aom/test/decode_api_test.cc
@@ -0,0 +1,55 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#include "test/util.h"
+#include "aom/aomdx.h"
+#include "aom/aom_decoder.h"
+
+namespace {
+
+TEST(DecodeAPI, InvalidParams) {
+  static const aom_codec_iface_t *kCodecs[] = {
+#if CONFIG_AV1_DECODER
+    aom_codec_av1_dx(),
+#endif
+  };
+  uint8_t buf[1] = { 0 };
+  aom_codec_ctx_t dec;
+
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_dec_init(NULL, NULL, NULL, 0));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_dec_init(&dec, NULL, NULL, 0));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_decode(NULL, NULL, 0, NULL));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_decode(NULL, buf, 0, NULL));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+            aom_codec_decode(NULL, buf, NELEMENTS(buf), NULL));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+            aom_codec_decode(NULL, NULL, NELEMENTS(buf), NULL));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_destroy(NULL));
+  EXPECT_TRUE(aom_codec_error(NULL) != NULL);
+
+  for (int i = 0; i < NELEMENTS(kCodecs); ++i) {
+    EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+              aom_codec_dec_init(NULL, kCodecs[i], NULL, 0));
+
+    EXPECT_EQ(AOM_CODEC_OK, aom_codec_dec_init(&dec, kCodecs[i], NULL, 0));
+    EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+              aom_codec_decode(&dec, NULL, NELEMENTS(buf), NULL));
+    EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_decode(&dec, buf, 0, NULL));
+
+    EXPECT_EQ(AOM_CODEC_OK, aom_codec_destroy(&dec));
+  }
+}
+
+}  // namespace
diff --git a/third_party/aom/test/decode_multithreaded_test.cc b/third_party/aom/test/decode_multithreaded_test.cc
new file mode 100644
index 0000000000..cea1d144f1
--- /dev/null
+++ b/third_party/aom/test/decode_multithreaded_test.cc
@@ -0,0 +1,185 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+
+#include "aom_mem/aom_mem.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/md5_helper.h"
+#include "test/util.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+
+static const int kNumMultiThreadDecoders = 3;
+
+class AV1DecodeMultiThreadedTest
+    : public ::libaom_test::CodecTestWith5Params<int, int, int, int, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  AV1DecodeMultiThreadedTest()
+      : EncoderTest(GET_PARAM(0)), md5_single_thread_(), md5_multi_thread_(),
+        n_tile_cols_(GET_PARAM(1)), n_tile_rows_(GET_PARAM(2)),
+        n_tile_groups_(GET_PARAM(3)), set_cpu_used_(GET_PARAM(4)),
+        row_mt_(GET_PARAM(5)) {
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+    cfg.w = 704;
+    cfg.h = 576;
+    cfg.threads = 1;
+    cfg.allow_lowbitdepth = 1;
+    single_thread_dec_ = codec_->CreateDecoder(cfg, 0);
+
+    // Test cfg.threads == powers of 2.
+    for (int i = 0; i < kNumMultiThreadDecoders; ++i) {
+      cfg.threads <<= 1;
+      multi_thread_dec_[i] = codec_->CreateDecoder(cfg, 0);
+      multi_thread_dec_[i]->Control(AV1D_SET_ROW_MT, row_mt_);
+    }
+
+    if (single_thread_dec_->IsAV1()) {
+      single_thread_dec_->Control(AV1D_EXT_TILE_DEBUG, 1);
+      single_thread_dec_->Control(AV1_SET_DECODE_TILE_ROW, -1);
+      single_thread_dec_->Control(AV1_SET_DECODE_TILE_COL, -1);
+    }
+    for (int i = 0; i < kNumMultiThreadDecoders; ++i) {
+      if (multi_thread_dec_[i]->IsAV1()) {
+        multi_thread_dec_[i]->Control(AV1D_EXT_TILE_DEBUG, 1);
+        multi_thread_dec_[i]->Control(AV1_SET_DECODE_TILE_ROW, -1);
+        multi_thread_dec_[i]->Control(AV1_SET_DECODE_TILE_COL, -1);
+      }
+    }
+  }
+
+  virtual ~AV1DecodeMultiThreadedTest() {
+    delete single_thread_dec_;
+    for (int i = 0; i < kNumMultiThreadDecoders; ++i)
+      delete multi_thread_dec_[i];
+  }
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(libaom_test::kTwoPassGood);
+  }
+
+  virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
+                                  libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AV1E_SET_TILE_COLUMNS, n_tile_cols_);
+      encoder->Control(AV1E_SET_TILE_ROWS, n_tile_rows_);
+      encoder->Control(AV1E_SET_NUM_TG, n_tile_groups_);
+      encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
+    }
+  }
+
+  void UpdateMD5(::libaom_test::Decoder *dec, const aom_codec_cx_pkt_t *pkt,
+                 ::libaom_test::MD5 *md5) {
+    const aom_codec_err_t res = dec->DecodeFrame(
+        reinterpret_cast<uint8_t *>(pkt->data.frame.buf), pkt->data.frame.sz);
+    if (res != AOM_CODEC_OK) {
+      abort_ = true;
+      ASSERT_EQ(AOM_CODEC_OK, res);
+    }
+    const aom_image_t *img = dec->GetDxData().Next();
+    md5->Add(img);
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    UpdateMD5(single_thread_dec_, pkt, &md5_single_thread_);
+
+    for (int i = 0; i < kNumMultiThreadDecoders; ++i)
+      UpdateMD5(multi_thread_dec_[i], pkt, &md5_multi_thread_[i]);
+  }
+
+  void DoTest() {
+    const aom_rational timebase = { 33333333, 1000000000 };
+    cfg_.g_timebase = timebase;
+    cfg_.rc_target_bitrate = 500;
+    cfg_.g_lag_in_frames = 12;
+    cfg_.rc_end_usage = AOM_VBR;
+
+    libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 704, 576,
+                                       timebase.den, timebase.num, 0, 5);
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+    const char *md5_single_thread_str = md5_single_thread_.Get();
+
+    for (int i = 0; i < kNumMultiThreadDecoders; ++i) {
+      const char *md5_multi_thread_str = md5_multi_thread_[i].Get();
+      ASSERT_STREQ(md5_single_thread_str, md5_multi_thread_str);
+    }
+  }
+
+  ::libaom_test::MD5 md5_single_thread_;
+  ::libaom_test::MD5 md5_multi_thread_[kNumMultiThreadDecoders];
+  ::libaom_test::Decoder *single_thread_dec_;
+  ::libaom_test::Decoder *multi_thread_dec_[kNumMultiThreadDecoders];
+
+ private:
+  int n_tile_cols_;
+  int n_tile_rows_;
+  int n_tile_groups_;
+  int set_cpu_used_;
+  int row_mt_;
+};
+
+// run an encode and do the decode both in single thread
+// and multi thread. Ensure that the MD5 of the output in both cases
+// is identical. If so, the test passes.
+TEST_P(AV1DecodeMultiThreadedTest, MD5Match) {
+  cfg_.large_scale_tile = 0;
+  single_thread_dec_->Control(AV1_SET_TILE_MODE, 0);
+  for (int i = 0; i < kNumMultiThreadDecoders; ++i)
+    multi_thread_dec_[i]->Control(AV1_SET_TILE_MODE, 0);
+  DoTest();
+}
+
+class AV1DecodeMultiThreadedTestLarge : public AV1DecodeMultiThreadedTest {};
+
+TEST_P(AV1DecodeMultiThreadedTestLarge, MD5Match) {
+  cfg_.large_scale_tile = 0;
+  single_thread_dec_->Control(AV1_SET_TILE_MODE, 0);
+  for (int i = 0; i < kNumMultiThreadDecoders; ++i)
+    multi_thread_dec_[i]->Control(AV1_SET_TILE_MODE, 0);
+  DoTest();
+}
+
+// TODO(ranjit): More tests have to be added using pre-generated MD5.
+AV1_INSTANTIATE_TEST_CASE(AV1DecodeMultiThreadedTest, ::testing::Values(1, 2),
+                          ::testing::Values(1, 2), ::testing::Values(1),
+                          ::testing::Values(3), ::testing::Values(0, 1));
+AV1_INSTANTIATE_TEST_CASE(AV1DecodeMultiThreadedTestLarge,
+                          ::testing::Values(0, 1, 2, 6),
+                          ::testing::Values(0, 1, 2, 6),
+                          ::testing::Values(1, 4), ::testing::Values(0),
+                          ::testing::Values(0, 1));
+
+class AV1DecodeMultiThreadedLSTestLarge
+    : public AV1DecodeMultiThreadedTestLarge {};
+
+TEST_P(AV1DecodeMultiThreadedLSTestLarge, MD5Match) {
+  cfg_.large_scale_tile = 1;
+  single_thread_dec_->Control(AV1_SET_TILE_MODE, 1);
+  for (int i = 0; i < kNumMultiThreadDecoders; ++i)
+    multi_thread_dec_[i]->Control(AV1_SET_TILE_MODE, 1);
+  DoTest();
+}
+
+AV1_INSTANTIATE_TEST_CASE(AV1DecodeMultiThreadedLSTestLarge,
+                          ::testing::Values(6), ::testing::Values(6),
+                          ::testing::Values(1), ::testing::Values(0, 3),
+                          ::testing::Values(0, 1));
+
+}  // namespace
diff --git a/third_party/aom/test/decode_perf_test.cc b/third_party/aom/test/decode_perf_test.cc
new file mode 100644
index 0000000000..bb7b00032e
--- /dev/null
+++ b/third_party/aom/test/decode_perf_test.cc
@@ -0,0 +1,246 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string>
+
+#include "config/aom_version.h"
+
+#include "aom_ports/aom_timer.h"
+#include "common/ivfenc.h"
+#include "test/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/ivf_video_source.h"
+#include "test/md5_helper.h"
+#include "test/util.h"
+#include "test/webm_video_source.h"
+
+using ::testing::make_tuple;
+
+namespace {
+
+#define VIDEO_NAME 0
+#define THREADS 1
+
+const double kUsecsInSec = 1000000.0;
+const char kNewEncodeOutputFile[] = "new_encode.ivf";
+
+/*
+ DecodePerfTest takes a tuple of filename + number of threads to decode with
+ */
+typedef ::testing::tuple<const char *, unsigned> DecodePerfParam;
+
+// TODO(jimbankoski): Add actual test vectors here when available.
+// const DecodePerfParam kAV1DecodePerfVectors[] = {};
+
+/*
+ In order to reflect real world performance as much as possible, Perf tests
+ *DO NOT* do any correctness checks. Please run them alongside correctness
+ tests to ensure proper codec integrity. Furthermore, in this test we
+ deliberately limit the amount of system calls we make to avoid OS
+ preemption.
+
+ TODO(joshualitt) create a more detailed perf measurement test to collect
+   power/temp/min max frame decode times/etc
+ */
+
+class DecodePerfTest : public ::testing::TestWithParam<DecodePerfParam> {};
+
+TEST_P(DecodePerfTest, PerfTest) {
+  const char *const video_name = GET_PARAM(VIDEO_NAME);
+  const unsigned threads = GET_PARAM(THREADS);
+
+  libaom_test::WebMVideoSource video(video_name);
+  video.Init();
+
+  aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+  cfg.threads = threads;
+  cfg.allow_lowbitdepth = 1;
+  libaom_test::AV1Decoder decoder(cfg, 0);
+
+  aom_usec_timer t;
+  aom_usec_timer_start(&t);
+
+  for (video.Begin(); video.cxdata() != NULL; video.Next()) {
+    decoder.DecodeFrame(video.cxdata(), video.frame_size());
+  }
+
+  aom_usec_timer_mark(&t);
+  const double elapsed_secs = double(aom_usec_timer_elapsed(&t)) / kUsecsInSec;
+  const unsigned frames = video.frame_number();
+  const double fps = double(frames) / elapsed_secs;
+
+  printf("{\n");
+  printf("\t\"type\" : \"decode_perf_test\",\n");
+  printf("\t\"version\" : \"%s\",\n", VERSION_STRING_NOSP);
+  printf("\t\"videoName\" : \"%s\",\n", video_name);
+  printf("\t\"threadCount\" : %u,\n", threads);
+  printf("\t\"decodeTimeSecs\" : %f,\n", elapsed_secs);
+  printf("\t\"totalFrames\" : %u,\n", frames);
+  printf("\t\"framesPerSecond\" : %f\n", fps);
+  printf("}\n");
+}
+
+// TODO(jimbankoski): Enabled when we have actual AV1 Decode vectors.
+// INSTANTIATE_TEST_CASE_P(AV1, DecodePerfTest,
+//                        ::testing::ValuesIn(kAV1DecodePerfVectors));
+
+class AV1NewEncodeDecodePerfTest
+    : public ::libaom_test::CodecTestWithParam<libaom_test::TestMode>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  AV1NewEncodeDecodePerfTest()
+      : EncoderTest(GET_PARAM(0)), encoding_mode_(GET_PARAM(1)), speed_(0),
+        outfile_(0), out_frames_(0) {}
+
+  virtual ~AV1NewEncodeDecodePerfTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+
+    cfg_.g_lag_in_frames = 25;
+    cfg_.rc_min_quantizer = 2;
+    cfg_.rc_max_quantizer = 56;
+    cfg_.rc_dropframe_thresh = 0;
+    cfg_.rc_undershoot_pct = 50;
+    cfg_.rc_overshoot_pct = 50;
+    cfg_.rc_buf_sz = 1000;
+    cfg_.rc_buf_initial_sz = 500;
+    cfg_.rc_buf_optimal_sz = 600;
+    cfg_.rc_end_usage = AOM_VBR;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_CPUUSED, speed_);
+      encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
+      encoder->Control(AV1E_SET_TILE_COLUMNS, 2);
+    }
+  }
+
+  virtual void BeginPassHook(unsigned int /*pass*/) {
+    const char *const env = getenv("LIBAOM_TEST_DATA_PATH");
+    const std::string data_path(env ? env : ".");
+    const std::string path_to_source = data_path + "/" + kNewEncodeOutputFile;
+    outfile_ = fopen(path_to_source.c_str(), "wb");
+    ASSERT_TRUE(outfile_ != NULL);
+  }
+
+  virtual void EndPassHook() {
+    if (outfile_ != NULL) {
+      if (!fseek(outfile_, 0, SEEK_SET))
+        ivf_write_file_header(outfile_, &cfg_, AV1_FOURCC, out_frames_);
+      fclose(outfile_);
+      outfile_ = NULL;
+    }
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    ++out_frames_;
+
+    // Write initial file header if first frame.
+    if (pkt->data.frame.pts == 0)
+      ivf_write_file_header(outfile_, &cfg_, AV1_FOURCC, out_frames_);
+
+    // Write frame header and data.
+    ivf_write_frame_header(outfile_, out_frames_, pkt->data.frame.sz);
+    ASSERT_EQ(fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz, outfile_),
+              pkt->data.frame.sz);
+  }
+
+  virtual bool DoDecode() const { return false; }
+
+  void set_speed(unsigned int speed) { speed_ = speed; }
+
+ private:
+  libaom_test::TestMode encoding_mode_;
+  uint32_t speed_;
+  FILE *outfile_;
+  uint32_t out_frames_;
+};
+
+struct EncodePerfTestVideo {
+  EncodePerfTestVideo(const char *name_, uint32_t width_, uint32_t height_,
+                      uint32_t bitrate_, int frames_)
+      : name(name_), width(width_), height(height_), bitrate(bitrate_),
+        frames(frames_) {}
+  const char *name;
+  uint32_t width;
+  uint32_t height;
+  uint32_t bitrate;
+  int frames;
+};
+
+const EncodePerfTestVideo kAV1EncodePerfTestVectors[] = {
+  EncodePerfTestVideo("niklas_1280_720_30.yuv", 1280, 720, 600, 470),
+};
+
+TEST_P(AV1NewEncodeDecodePerfTest, PerfTest) {
+  SetUp();
+
+  // TODO(JBB): Make this work by going through the set of given files.
+  const int i = 0;
+  const aom_rational timebase = { 33333333, 1000000000 };
+  cfg_.g_timebase = timebase;
+  cfg_.rc_target_bitrate = kAV1EncodePerfTestVectors[i].bitrate;
+
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  const char *video_name = kAV1EncodePerfTestVectors[i].name;
+  libaom_test::I420VideoSource video(
+      video_name, kAV1EncodePerfTestVectors[i].width,
+      kAV1EncodePerfTestVectors[i].height, timebase.den, timebase.num, 0,
+      kAV1EncodePerfTestVectors[i].frames);
+  set_speed(2);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  const uint32_t threads = 4;
+
+  libaom_test::IVFVideoSource decode_video(kNewEncodeOutputFile);
+  decode_video.Init();
+
+  aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+  cfg.threads = threads;
+  cfg.allow_lowbitdepth = 1;
+  libaom_test::AV1Decoder decoder(cfg, 0);
+
+  aom_usec_timer t;
+  aom_usec_timer_start(&t);
+
+  for (decode_video.Begin(); decode_video.cxdata() != NULL;
+       decode_video.Next()) {
+    decoder.DecodeFrame(decode_video.cxdata(), decode_video.frame_size());
+  }
+
+  aom_usec_timer_mark(&t);
+  const double elapsed_secs =
+      static_cast<double>(aom_usec_timer_elapsed(&t)) / kUsecsInSec;
+  const unsigned decode_frames = decode_video.frame_number();
+  const double fps = static_cast<double>(decode_frames) / elapsed_secs;
+
+  printf("{\n");
+  printf("\t\"type\" : \"decode_perf_test\",\n");
+  printf("\t\"version\" : \"%s\",\n", VERSION_STRING_NOSP);
+  printf("\t\"videoName\" : \"%s\",\n", kNewEncodeOutputFile);
+  printf("\t\"threadCount\" : %u,\n", threads);
+  printf("\t\"decodeTimeSecs\" : %f,\n", elapsed_secs);
+  printf("\t\"totalFrames\" : %u,\n", decode_frames);
+  printf("\t\"framesPerSecond\" : %f\n", fps);
+  printf("}\n");
+}
+
+AV1_INSTANTIATE_TEST_CASE(AV1NewEncodeDecodePerfTest,
+                          ::testing::Values(::libaom_test::kTwoPassGood));
+}  // namespace
diff --git a/third_party/aom/test/decode_test_driver.cc b/third_party/aom/test/decode_test_driver.cc
new file mode 100644
index 0000000000..70de0cff69
--- /dev/null
+++ b/third_party/aom/test/decode_test_driver.cc
@@ -0,0 +1,114 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/register_state_check.h"
+#include "test/video_source.h"
+
+namespace libaom_test {
+
+const char kAV1Name[] = "AOMedia Project AV1 Decoder";
+
+aom_codec_err_t Decoder::PeekStream(const uint8_t *cxdata, size_t size,
+                                    aom_codec_stream_info_t *stream_info) {
+  return aom_codec_peek_stream_info(CodecInterface(), cxdata, size,
+                                    stream_info);
+}
+
+aom_codec_err_t Decoder::DecodeFrame(const uint8_t *cxdata, size_t size) {
+  return DecodeFrame(cxdata, size, NULL);
+}
+
+aom_codec_err_t Decoder::DecodeFrame(const uint8_t *cxdata, size_t size,
+                                     void *user_priv) {
+  aom_codec_err_t res_dec;
+  InitOnce();
+  API_REGISTER_STATE_CHECK(
+      res_dec = aom_codec_decode(&decoder_, cxdata, size, user_priv));
+  return res_dec;
+}
+
+bool Decoder::IsAV1() const {
+  const char *codec_name = GetDecoderName();
+  return strncmp(kAV1Name, codec_name, sizeof(kAV1Name) - 1) == 0;
+}
+
+void DecoderTest::HandlePeekResult(Decoder *const /*decoder*/,
+                                   CompressedVideoSource * /*video*/,
+                                   const aom_codec_err_t res_peek) {
+  /* The Av1 implementation of PeekStream returns an error only if the
+   * data passed to it isn't a valid Av1 chunk. */
+  ASSERT_EQ(AOM_CODEC_OK, res_peek)
+      << "Peek return failed: " << aom_codec_err_to_string(res_peek);
+}
+
+void DecoderTest::RunLoop(CompressedVideoSource *video,
+                          const aom_codec_dec_cfg_t &dec_cfg) {
+  Decoder *const decoder = codec_->CreateDecoder(dec_cfg, flags_);
+  ASSERT_TRUE(decoder != NULL);
+  bool end_of_file = false;
+  bool peeked_stream = false;
+
+  // Decode frames.
+  for (video->Begin(); !::testing::Test::HasFailure() && !end_of_file;
+       video->Next()) {
+    PreDecodeFrameHook(*video, decoder);
+
+    aom_codec_stream_info_t stream_info;
+    stream_info.is_annexb = 0;
+
+    if (video->cxdata() != NULL) {
+      if (!peeked_stream) {
+        // TODO(yaowu): PeekStream returns error for non-sequence_header_obu,
+        // therefore should only be tried once per sequence, this shall be fixed
+        // once PeekStream is updated to properly operate on other obus.
+        const aom_codec_err_t res_peek = decoder->PeekStream(
+            video->cxdata(), video->frame_size(), &stream_info);
+        HandlePeekResult(decoder, video, res_peek);
+        ASSERT_FALSE(::testing::Test::HasFailure());
+        peeked_stream = true;
+      }
+
+      aom_codec_err_t res_dec =
+          decoder->DecodeFrame(video->cxdata(), video->frame_size());
+      if (!HandleDecodeResult(res_dec, *video, decoder)) break;
+    } else {
+      // Signal end of the file to the decoder.
+      const aom_codec_err_t res_dec = decoder->DecodeFrame(NULL, 0);
+      ASSERT_EQ(AOM_CODEC_OK, res_dec) << decoder->DecodeError();
+      end_of_file = true;
+    }
+
+    DxDataIterator dec_iter = decoder->GetDxData();
+    const aom_image_t *img = NULL;
+
+    // Get decompressed data
+    while (!::testing::Test::HasFailure() && (img = dec_iter.Next()))
+      DecompressedFrameHook(*img, video->frame_number());
+  }
+  delete decoder;
+}
+
+void DecoderTest::RunLoop(CompressedVideoSource *video) {
+  aom_codec_dec_cfg_t dec_cfg = aom_codec_dec_cfg_t();
+  RunLoop(video, dec_cfg);
+}
+
+void DecoderTest::set_cfg(const aom_codec_dec_cfg_t &dec_cfg) {
+  memcpy(&cfg_, &dec_cfg, sizeof(cfg_));
+}
+
+void DecoderTest::set_flags(const aom_codec_flags_t flags) { flags_ = flags; }
+
+}  // namespace libaom_test
diff --git a/third_party/aom/test/decode_test_driver.h b/third_party/aom/test/decode_test_driver.h
new file mode 100644
index 0000000000..d13e13ea1f
--- /dev/null
+++ b/third_party/aom/test/decode_test_driver.h
@@ -0,0 +1,165 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_DECODE_TEST_DRIVER_H_
+#define AOM_TEST_DECODE_TEST_DRIVER_H_
+#include <cstring>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#include "aom/aom_decoder.h"
+
+namespace libaom_test {
+
+class CodecFactory;
+class CompressedVideoSource;
+
+// Provides an object to handle decoding output
+class DxDataIterator {
+ public:
+  explicit DxDataIterator(aom_codec_ctx_t *decoder)
+      : decoder_(decoder), iter_(NULL) {}
+
+  const aom_image_t *Next() { return aom_codec_get_frame(decoder_, &iter_); }
+
+ private:
+  aom_codec_ctx_t *decoder_;
+  aom_codec_iter_t iter_;
+};
+
+// Provides a simplified interface to manage one video decoding.
+// Similar to Encoder class, the exact services should be added
+// as more tests are added.
+class Decoder {
+ public:
+  explicit Decoder(aom_codec_dec_cfg_t cfg)
+      : cfg_(cfg), flags_(0), init_done_(false) {
+    memset(&decoder_, 0, sizeof(decoder_));
+  }
+
+  Decoder(aom_codec_dec_cfg_t cfg, const aom_codec_flags_t flag)
+      : cfg_(cfg), flags_(flag), init_done_(false) {
+    memset(&decoder_, 0, sizeof(decoder_));
+  }
+
+  virtual ~Decoder() { aom_codec_destroy(&decoder_); }
+
+  aom_codec_err_t PeekStream(const uint8_t *cxdata, size_t size,
+                             aom_codec_stream_info_t *stream_info);
+
+  aom_codec_err_t DecodeFrame(const uint8_t *cxdata, size_t size);
+
+  aom_codec_err_t DecodeFrame(const uint8_t *cxdata, size_t size,
+                              void *user_priv);
+
+  DxDataIterator GetDxData() { return DxDataIterator(&decoder_); }
+
+  void Control(int ctrl_id, int arg) { Control(ctrl_id, arg, AOM_CODEC_OK); }
+
+  void Control(int ctrl_id, const void *arg) {
+    InitOnce();
+    const aom_codec_err_t res = aom_codec_control_(&decoder_, ctrl_id, arg);
+    ASSERT_EQ(AOM_CODEC_OK, res) << DecodeError();
+  }
+
+  void Control(int ctrl_id, int arg, aom_codec_err_t expected_value) {
+    InitOnce();
+    const aom_codec_err_t res = aom_codec_control_(&decoder_, ctrl_id, arg);
+    ASSERT_EQ(expected_value, res) << DecodeError();
+  }
+
+  const char *DecodeError() {
+    const char *detail = aom_codec_error_detail(&decoder_);
+    return detail ? detail : aom_codec_error(&decoder_);
+  }
+
+  // Passes the external frame buffer information to libaom.
+  aom_codec_err_t SetFrameBufferFunctions(
+      aom_get_frame_buffer_cb_fn_t cb_get,
+      aom_release_frame_buffer_cb_fn_t cb_release, void *user_priv) {
+    InitOnce();
+    return aom_codec_set_frame_buffer_functions(&decoder_, cb_get, cb_release,
+                                                user_priv);
+  }
+
+  const char *GetDecoderName() const {
+    return aom_codec_iface_name(CodecInterface());
+  }
+
+  bool IsAV1() const;
+
+  aom_codec_ctx_t *GetDecoder() { return &decoder_; }
+
+ protected:
+  virtual aom_codec_iface_t *CodecInterface() const = 0;
+
+  void InitOnce() {
+    if (!init_done_) {
+      const aom_codec_err_t res =
+          aom_codec_dec_init(&decoder_, CodecInterface(), &cfg_, flags_);
+      ASSERT_EQ(AOM_CODEC_OK, res) << DecodeError();
+      init_done_ = true;
+    }
+  }
+
+  aom_codec_ctx_t decoder_;
+  aom_codec_dec_cfg_t cfg_;
+  aom_codec_flags_t flags_;
+  bool init_done_;
+};
+
+// Common test functionality for all Decoder tests.
+class DecoderTest {
+ public:
+  // Main decoding loop
+  virtual void RunLoop(CompressedVideoSource *video);
+  virtual void RunLoop(CompressedVideoSource *video,
+                       const aom_codec_dec_cfg_t &dec_cfg);
+
+  virtual void set_cfg(const aom_codec_dec_cfg_t &dec_cfg);
+  virtual void set_flags(const aom_codec_flags_t flags);
+
+  // Hook to be called before decompressing every frame.
+  virtual void PreDecodeFrameHook(const CompressedVideoSource & /*video*/,
+                                  Decoder * /*decoder*/) {}
+
+  // Hook to be called to handle decode result. Return true to continue.
+  virtual bool HandleDecodeResult(const aom_codec_err_t res_dec,
+                                  const CompressedVideoSource & /*video*/,
+                                  Decoder *decoder) {
+    EXPECT_EQ(AOM_CODEC_OK, res_dec) << decoder->DecodeError();
+    return AOM_CODEC_OK == res_dec;
+  }
+
+  // Hook to be called on every decompressed frame.
+  virtual void DecompressedFrameHook(const aom_image_t & /*img*/,
+                                     const unsigned int /*frame_number*/) {}
+
+  // Hook to be called on peek result
+  virtual void HandlePeekResult(Decoder *const decoder,
+                                CompressedVideoSource *video,
+                                const aom_codec_err_t res_peek);
+
+ protected:
+  explicit DecoderTest(const CodecFactory *codec)
+      : codec_(codec), cfg_(), flags_(0) {}
+
+  virtual ~DecoderTest() {}
+
+  const CodecFactory *codec_;
+  aom_codec_dec_cfg_t cfg_;
+  aom_codec_flags_t flags_;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_DECODE_TEST_DRIVER_H_
diff --git a/third_party/aom/test/decode_to_md5.sh b/third_party/aom/test/decode_to_md5.sh
new file mode 100755
index 0000000000..2edd1cb52b
--- /dev/null
+++ b/third_party/aom/test/decode_to_md5.sh
@@ -0,0 +1,77 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the libaom decode_to_md5 example. To add new tests to this
+## file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to decode_to_md5_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: Make sure input is available:
+#   $AV1_IVF_FILE is required.
+decode_to_md5_verify_environment() {
+  if [ "$(av1_encode_available)" != "yes" ] && [ ! -e "${AV1_IVF_FILE}" ]; then
+    return 1
+  fi
+}
+
+# Runs decode_to_md5 on $1 and captures the md5 sum for the final frame. $2 is
+# interpreted as codec name and used solely to name the output file. $3 is the
+# expected md5 sum: It must match that of the final frame.
+decode_to_md5() {
+  local decoder="$(aom_tool_path decode_to_md5)"
+  local input_file="$1"
+  local codec="$2"
+  local expected_md5="$3"
+  local output_file="${AOM_TEST_OUTPUT_DIR}/decode_to_md5_${codec}"
+
+  if [ ! -x "${decoder}" ]; then
+    elog "${decoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${decoder}" "${input_file}" "${output_file}" \
+      ${devnull}
+
+  [ -e "${output_file}" ] || return 1
+
+  local md5_last_frame="$(tail -n1 "${output_file}" | awk '{print $1}')"
+  local actual_md5="$(echo "${md5_last_frame}" | awk '{print $1}')"
+  if [ "${actual_md5}" = "${expected_md5}" ]; then
+    return 0
+  else
+    elog "MD5 mismatch:"
+    elog "Expected: ${expected_md5}"
+    elog "Actual: ${actual_md5}"
+    return 1
+  fi
+}
+
+DISABLED_decode_to_md5_av1() {
+  # expected MD5 sum for the last frame.
+  local expected_md5="567dd6d4b7a7170edddbf58bbcc3aff1"
+  local file="${AV1_IVF_FILE}"
+
+  # TODO(urvang): Check in the encoded file (like libvpx does) to avoid
+  # encoding every time.
+  if [ "$(av1_decode_available)" = "yes" ]; then
+    if [ ! -e "${AV1_IVF_FILE}" ]; then
+      file="${AOM_TEST_OUTPUT_DIR}/test_encode.ivf"
+      encode_yuv_raw_input_av1 "${file}" --ivf
+    fi
+    decode_to_md5 "${file}" "av1" "${expected_md5}"
+  fi
+}
+
+# TODO(tomfinegan): Enable when the bitstream stabilizes.
+decode_to_md5_tests="DISABLED_decode_to_md5_av1"
+
+run_tests decode_to_md5_verify_environment "${decode_to_md5_tests}"
diff --git a/third_party/aom/test/decode_with_drops.sh b/third_party/aom/test/decode_with_drops.sh
new file mode 100755
index 0000000000..155ee92077
--- /dev/null
+++ b/third_party/aom/test/decode_with_drops.sh
@@ -0,0 +1,68 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the libaom decode_with_drops example. To add new tests to
+## this file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to decode_with_drops_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: Make sure input is available:
+#   $AV1_IVF_FILE is required.
+decode_with_drops_verify_environment() {
+  if [ "$(av1_encode_available)" != "yes" ] && [ ! -e "${AV1_IVF_FILE}" ]; then
+    return 1
+  fi
+}
+
+# Runs decode_with_drops on $1, $2 is interpreted as codec name and used solely
+# to name the output file. $3 is the drop mode, and is passed directly to
+# decode_with_drops.
+decode_with_drops() {
+  local decoder="$(aom_tool_path decode_with_drops)"
+  local input_file="$1"
+  local codec="$2"
+  local output_file="${AOM_TEST_OUTPUT_DIR}/decode_with_drops_${codec}"
+  local drop_mode="$3"
+
+  if [ ! -x "${decoder}" ]; then
+    elog "${decoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${decoder}" "${input_file}" "${output_file}" \
+      "${drop_mode}" ${devnull}
+
+  [ -e "${output_file}" ] || return 1
+}
+
+
+# Decodes $AV1_IVF_FILE while dropping frames, twice: once in sequence mode,
+# and once in pattern mode.
+DISABLED_decode_with_drops_av1() {
+  if [ "$(av1_decode_available)" = "yes" ]; then
+    local file="${AV1_IVF_FILE}"
+    if [ ! -e "${AV1_IVF_FILE}" ]; then
+      file="${AOM_TEST_OUTPUT_DIR}/test_encode.ivf"
+      encode_yuv_raw_input_av1 "${file}" --ivf
+    fi
+    # Drop frames 3 and 4.
+    decode_with_drops "${file}" "av1" "3-4"
+
+    # Test pattern mode: Drop 3 of every 4 frames.
+    decode_with_drops "${file}" "av1" "3/4"
+  fi
+}
+
+# TODO(yaowu): Disable this test as trailing_bit check is expected to fail
+decode_with_drops_tests="DISABLED_decode_with_drops_av1"
+
+run_tests decode_with_drops_verify_environment "${decode_with_drops_tests}"
diff --git a/third_party/aom/test/divu_small_test.cc b/third_party/aom/test/divu_small_test.cc
new file mode 100644
index 0000000000..064f8ee454
--- /dev/null
+++ b/third_party/aom/test/divu_small_test.cc
@@ -0,0 +1,41 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdlib.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/acm_random.h"
+#include "av1/common/odintrin.h"
+
+using libaom_test::ACMRandom;
+
+TEST(Daala, TestDIVUuptoMAX) {
+  for (int d = 1; d <= OD_DIVU_DMAX; d++) {
+    for (uint32_t x = 1; x <= 1000000; x++) {
+      GTEST_ASSERT_EQ(x / d, OD_DIVU_SMALL(x, d))
+          << "x=" << x << " d=" << d << " x/d=" << (x / d)
+          << " != " << OD_DIVU_SMALL(x, d);
+    }
+  }
+}
+
+TEST(Daala, TestDIVUrandI31) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  for (int d = 1; d < OD_DIVU_DMAX; d++) {
+    for (int i = 0; i < 1000000; i++) {
+      uint32_t x = rnd.Rand31();
+      GTEST_ASSERT_EQ(x / d, OD_DIVU_SMALL(x, d))
+          << "x=" << x << " d=" << d << " x/d=" << (x / d)
+          << " != " << OD_DIVU_SMALL(x, d);
+    }
+  }
+}
diff --git a/third_party/aom/test/dr_prediction_test.cc b/third_party/aom/test/dr_prediction_test.cc
new file mode 100644
index 0000000000..ff2c1de4e9
--- /dev/null
+++ b/third_party/aom/test/dr_prediction_test.cc
@@ -0,0 +1,369 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom_mem/aom_mem.h"
+#include "aom_ports/aom_timer.h"
+#include "av1/common/blockd.h"
+#include "av1/common/pred_common.h"
+#include "av1/common/reconintra.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+namespace {
+
+const int kZ1Start = 0;
+const int kZ2Start = 90;
+const int kZ3Start = 180;
+
+const TX_SIZE kTxSize[] = { TX_4X4,   TX_8X8,   TX_16X16, TX_32X32, TX_64X64,
+                            TX_4X8,   TX_8X4,   TX_8X16,  TX_16X8,  TX_16X32,
+                            TX_32X16, TX_32X64, TX_64X32, TX_4X16,  TX_16X4,
+                            TX_8X32,  TX_32X8,  TX_16X64, TX_64X16 };
+
+const char *const kTxSizeStrings[] = {
+  "TX_4X4",   "TX_8X8",   "TX_16X16", "TX_32X32", "TX_64X64",
+  "TX_4X8",   "TX_8X4",   "TX_8X16",  "TX_16X8",  "TX_16X32",
+  "TX_32X16", "TX_32X64", "TX_64X32", "TX_4X16",  "TX_16X4",
+  "TX_8X32",  "TX_32X8",  "TX_16X64", "TX_64X16"
+};
+
+using libaom_test::ACMRandom;
+
+typedef void (*DrPred_Hbd)(uint16_t *dst, ptrdiff_t stride, int bw, int bh,
+                           const uint16_t *above, const uint16_t *left,
+                           int upsample_above, int upsample_left, int dx,
+                           int dy, int bd);
+
+typedef void (*DrPred)(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                       const uint8_t *above, const uint8_t *left,
+                       int upsample_above, int upsample_left, int dx, int dy,
+                       int bd);
+
+typedef void (*Z1_Lbd)(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                       const uint8_t *above, const uint8_t *left,
+                       int upsample_above, int dx, int dy);
+template <Z1_Lbd fn>
+void z1_wrapper(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                const uint8_t *above, const uint8_t *left, int upsample_above,
+                int /*upsample_left*/, int dx, int dy, int /*bd*/) {
+  fn(dst, stride, bw, bh, above, left, upsample_above, dx, dy);
+}
+
+typedef void (*Z2_Lbd)(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                       const uint8_t *above, const uint8_t *left,
+                       int upsample_above, int upsample_left, int dx, int dy);
+template <Z2_Lbd fn>
+void z2_wrapper(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                const uint8_t *above, const uint8_t *left, int upsample_above,
+                int upsample_left, int dx, int dy, int /*bd*/) {
+  fn(dst, stride, bw, bh, above, left, upsample_above, upsample_left, dx, dy);
+}
+
+typedef void (*Z3_Lbd)(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                       const uint8_t *above, const uint8_t *left,
+                       int upsample_left, int dx, int dy);
+template <Z3_Lbd fn>
+void z3_wrapper(uint8_t *dst, ptrdiff_t stride, int bw, int bh,
+                const uint8_t *above, const uint8_t *left,
+                int /*upsample_above*/, int upsample_left, int dx, int dy,
+                int /*bd*/) {
+  fn(dst, stride, bw, bh, above, left, upsample_left, dx, dy);
+}
+
+typedef void (*Z1_Hbd)(uint16_t *dst, ptrdiff_t stride, int bw, int bh,
+                       const uint16_t *above, const uint16_t *left,
+                       int upsample_above, int dx, int dy, int bd);
+template <Z1_Hbd fn>
+void z1_wrapper_hbd(uint16_t *dst, ptrdiff_t stride, int bw, int bh,
+                    const uint16_t *above, const uint16_t *left,
+                    int upsample_above, int /*upsample_left*/, int dx, int dy,
+                    int bd) {
+  fn(dst, stride, bw, bh, above, left, upsample_above, dx, dy, bd);
+}
+
+typedef void (*Z2_Hbd)(uint16_t *dst, ptrdiff_t stride, int bw, int bh,
+                       const uint16_t *above, const uint16_t *left,
+                       int upsample_above, int upsample_left, int dx, int dy,
+                       int bd);
+template <Z2_Hbd fn>
+void z2_wrapper_hbd(uint16_t *dst, ptrdiff_t stride, int bw, int bh,
+                    const uint16_t *above, const uint16_t *left,
+                    int upsample_above, int upsample_left, int dx, int dy,
+                    int bd) {
+  fn(dst, stride, bw, bh, above, left, upsample_above, upsample_left, dx, dy,
+     bd);
+}
+
+typedef void (*Z3_Hbd)(uint16_t *dst, ptrdiff_t stride, int bw, int bh,
+                       const uint16_t *above, const uint16_t *left,
+                       int upsample_left, int dx, int dy, int bd);
+template <Z3_Hbd fn>
+void z3_wrapper_hbd(uint16_t *dst, ptrdiff_t stride, int bw, int bh,
+                    const uint16_t *above, const uint16_t *left,
+                    int /*upsample_above*/, int upsample_left, int dx, int dy,
+                    int bd) {
+  fn(dst, stride, bw, bh, above, left, upsample_left, dx, dy, bd);
+}
+
+template <typename FuncType>
+struct DrPredFunc {
+  DrPredFunc(FuncType pred = NULL, FuncType tst = NULL, int bit_depth_value = 0,
+             int start_angle_value = 0)
+      : ref_fn(pred), tst_fn(tst), bit_depth(bit_depth_value),
+        start_angle(start_angle_value) {}
+
+  FuncType ref_fn;
+  FuncType tst_fn;
+  int bit_depth;
+  int start_angle;
+};
+
+template <typename Pixel, typename FuncType>
+class DrPredTest : public ::testing::TestWithParam<DrPredFunc<FuncType> > {
+ protected:
+  static const int kMaxNumTests = 100000;
+  static const int kIterations = 10;
+  static const int kDstStride = 64;
+  static const int kDstSize = kDstStride * kDstStride;
+  static const int kOffset = 16;
+  static const int kBufSize = ((2 * MAX_TX_SIZE) << 1) + 16;
+
+  DrPredTest()
+      : enable_upsample_(0), upsample_above_(0), upsample_left_(0), bw_(0),
+        bh_(0), dx_(1), dy_(1), bd_(8), txsize_(TX_4X4) {
+    params_ = this->GetParam();
+    start_angle_ = params_.start_angle;
+    stop_angle_ = start_angle_ + 90;
+
+    dst_ref_ = &dst_ref_data_[0];
+    dst_tst_ = &dst_tst_data_[0];
+    dst_stride_ = kDstStride;
+    above_ = &above_data_[kOffset];
+    left_ = &left_data_[kOffset];
+
+    for (int i = 0; i < kBufSize; ++i) {
+      above_data_[i] = rng_.Rand8();
+      left_data_[i] = rng_.Rand8();
+    }
+
+    for (int i = 0; i < kDstSize; ++i) {
+      dst_ref_[i] = 0;
+    }
+  }
+
+  virtual ~DrPredTest() {}
+
+  void Predict(bool speedtest, int tx) {
+    const int kNumTests = speedtest ? kMaxNumTests : 1;
+    aom_usec_timer timer;
+
+    aom_usec_timer_start(&timer);
+    for (int k = 0; k < kNumTests; ++k) {
+      params_.ref_fn(dst_ref_, dst_stride_, bw_, bh_, above_, left_,
+                     upsample_above_, upsample_left_, dx_, dy_, bd_);
+    }
+    aom_usec_timer_mark(&timer);
+    const int ref_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+
+    aom_usec_timer_start(&timer);
+    if (params_.tst_fn) {
+      for (int k = 0; k < kNumTests; ++k) {
+        ASM_REGISTER_STATE_CHECK(params_.tst_fn(dst_tst_, dst_stride_, bw_, bh_,
+                                                above_, left_, upsample_above_,
+                                                upsample_left_, dx_, dy_, bd_));
+      }
+    }
+    aom_usec_timer_mark(&timer);
+    const int tst_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+
+    OutputTimes(kNumTests, ref_time, tst_time, tx);
+  }
+
+  void RunTest(bool speedtest, int p_angle) {
+    for (int i = 0; i < kBufSize; ++i) {
+      above_data_[i] = left_data_[i] = (1 << bd_) - 1;
+    }
+
+    for (int tx = 0; tx < TX_SIZES_ALL; ++tx) {
+      if (params_.tst_fn == NULL) {
+        for (int i = 0; i < kDstSize; ++i) {
+          dst_tst_[i] = (1 << bd_) - 1;
+        }
+      } else {
+        for (int i = 0; i < kDstSize; ++i) {
+          dst_tst_[i] = 0;
+        }
+      }
+
+      bw_ = tx_size_wide[kTxSize[tx]];
+      bh_ = tx_size_high[kTxSize[tx]];
+
+      if (enable_upsample_) {
+        upsample_above_ =
+            av1_use_intra_edge_upsample(bw_, bh_, p_angle - 90, 0);
+        upsample_left_ =
+            av1_use_intra_edge_upsample(bw_, bh_, p_angle - 180, 0);
+      } else {
+        upsample_above_ = upsample_left_ = 0;
+      }
+
+      Predict(speedtest, tx);
+
+      for (int r = 0; r < bh_; ++r) {
+        for (int c = 0; c < bw_; ++c) {
+          ASSERT_EQ(dst_ref_[r * dst_stride_ + c],
+                    dst_tst_[r * dst_stride_ + c])
+              << bw_ << "x" << bh_ << " r: " << r << " c: " << c
+              << " dx: " << dx_ << " dy: " << dy_
+              << " upsample_above: " << upsample_above_
+              << " upsample_left: " << upsample_left_;
+        }
+      }
+    }
+  }
+
+  void OutputTimes(int num_tests, int ref_time, int tst_time, int tx) {
+    if (num_tests > 1) {
+      if (params_.tst_fn) {
+        const float x = static_cast<float>(ref_time) / tst_time;
+        printf("\t[%8s] :: ref time %6d, tst time %6d     %3.2f\n",
+               kTxSizeStrings[tx], ref_time, tst_time, x);
+      } else {
+        printf("\t[%8s] :: ref time %6d\n", kTxSizeStrings[tx], ref_time);
+      }
+    }
+  }
+
+  Pixel dst_ref_data_[kDstSize];
+  Pixel dst_tst_data_[kDstSize];
+
+  Pixel left_data_[kBufSize];
+  Pixel dummy_data_[kBufSize];
+  Pixel above_data_[kBufSize];
+
+  Pixel *dst_ref_;
+  Pixel *dst_tst_;
+  Pixel *above_;
+  Pixel *left_;
+  int dst_stride_;
+
+  int enable_upsample_;
+  int upsample_above_;
+  int upsample_left_;
+  int bw_;
+  int bh_;
+  int dx_;
+  int dy_;
+  int bd_;
+  TX_SIZE txsize_;
+
+  int start_angle_;
+  int stop_angle_;
+
+  ACMRandom rng_;
+
+  DrPredFunc<FuncType> params_;
+};
+
+class LowbdDrPredTest : public DrPredTest<uint8_t, DrPred> {};
+
+TEST_P(LowbdDrPredTest, SaturatedValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    enable_upsample_ = iter & 1;
+    for (int angle = start_angle_; angle < stop_angle_; ++angle) {
+      dx_ = av1_get_dx(angle);
+      dy_ = av1_get_dy(angle);
+      if (dx_ && dy_) RunTest(false, angle);
+    }
+  }
+}
+
+TEST_P(LowbdDrPredTest, DISABLED_Speed) {
+  const int angles[] = { 3, 45, 87 };
+  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
+    for (int i = 0; i < 3; ++i) {
+      const int angle = angles[i] + start_angle_;
+      dx_ = av1_get_dx(angle);
+      dy_ = av1_get_dy(angle);
+      printf("enable_upsample: %d angle: %d ~~~~~~~~~~~~~~~\n",
+             enable_upsample_, angle);
+      if (dx_ && dy_) RunTest(true, angle);
+    }
+  }
+}
+
+using ::testing::make_tuple;
+
+INSTANTIATE_TEST_CASE_P(
+    C, LowbdDrPredTest,
+    ::testing::Values(DrPredFunc<DrPred>(&z1_wrapper<av1_dr_prediction_z1_c>,
+                                         NULL, AOM_BITS_8, kZ1Start),
+                      DrPredFunc<DrPred>(&z2_wrapper<av1_dr_prediction_z2_c>,
+                                         NULL, AOM_BITS_8, kZ2Start),
+                      DrPredFunc<DrPred>(&z3_wrapper<av1_dr_prediction_z3_c>,
+                                         NULL, AOM_BITS_8, kZ3Start)));
+
+class HighbdDrPredTest : public DrPredTest<uint16_t, DrPred_Hbd> {};
+
+TEST_P(HighbdDrPredTest, SaturatedValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    enable_upsample_ = iter & 1;
+    for (int angle = start_angle_; angle < stop_angle_; ++angle) {
+      dx_ = av1_get_dx(angle);
+      dy_ = av1_get_dy(angle);
+      if (dx_ && dy_) RunTest(false, angle);
+    }
+  }
+}
+
+TEST_P(HighbdDrPredTest, DISABLED_Speed) {
+  const int angles[] = { 3, 45, 87 };
+  for (enable_upsample_ = 0; enable_upsample_ < 2; ++enable_upsample_) {
+    for (int i = 0; i < 3; ++i) {
+      const int angle = angles[i] + start_angle_;
+      dx_ = av1_get_dx(angle);
+      dy_ = av1_get_dy(angle);
+      printf("enable_upsample: %d angle: %d ~~~~~~~~~~~~~~~\n",
+             enable_upsample_, angle);
+      if (dx_ && dy_) RunTest(true, angle);
+    }
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(
+    C, HighbdDrPredTest,
+    ::testing::Values(
+        DrPredFunc<DrPred_Hbd>(&z1_wrapper_hbd<av1_highbd_dr_prediction_z1_c>,
+                               NULL, AOM_BITS_8, kZ1Start),
+        DrPredFunc<DrPred_Hbd>(&z1_wrapper_hbd<av1_highbd_dr_prediction_z1_c>,
+                               NULL, AOM_BITS_10, kZ1Start),
+        DrPredFunc<DrPred_Hbd>(&z1_wrapper_hbd<av1_highbd_dr_prediction_z1_c>,
+                               NULL, AOM_BITS_12, kZ1Start),
+        DrPredFunc<DrPred_Hbd>(&z2_wrapper_hbd<av1_highbd_dr_prediction_z2_c>,
+                               NULL, AOM_BITS_8, kZ2Start),
+        DrPredFunc<DrPred_Hbd>(&z2_wrapper_hbd<av1_highbd_dr_prediction_z2_c>,
+                               NULL, AOM_BITS_10, kZ2Start),
+        DrPredFunc<DrPred_Hbd>(&z2_wrapper_hbd<av1_highbd_dr_prediction_z2_c>,
+                               NULL, AOM_BITS_12, kZ2Start),
+        DrPredFunc<DrPred_Hbd>(&z3_wrapper_hbd<av1_highbd_dr_prediction_z3_c>,
+                               NULL, AOM_BITS_8, kZ3Start),
+        DrPredFunc<DrPred_Hbd>(&z3_wrapper_hbd<av1_highbd_dr_prediction_z3_c>,
+                               NULL, AOM_BITS_10, kZ3Start),
+        DrPredFunc<DrPred_Hbd>(&z3_wrapper_hbd<av1_highbd_dr_prediction_z3_c>,
+                               NULL, AOM_BITS_12, kZ3Start)));
+
+}  // namespace
diff --git a/third_party/aom/test/dump_obu.sh b/third_party/aom/test/dump_obu.sh
new file mode 100755
index 0000000000..da44dd7e67
--- /dev/null
+++ b/third_party/aom/test/dump_obu.sh
@@ -0,0 +1,70 @@
+#!/bin/sh
+## Copyright (c) 2018, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the libaom dump_obu tool. To add new tests to this
+## file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to dump_obu_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+readonly dump_obu_test_file="${AOM_TEST_OUTPUT_DIR}/av1_obu_test.ivf"
+
+dump_obu_verify_environment() {
+  if [ ! -e "${YUV_RAW_INPUT}" ]; then
+    elog "The file ${YUV_RAW_INPUT##*/} must exist in LIBAOM_TEST_DATA_PATH."
+    return 1
+  fi
+  if [ "$(dump_obu_available)" = "yes" ]; then
+    if [ -z "$(aom_tool_path dump_obu)" ]; then
+      elog "dump_obu not found in LIBAOM_BIN_PATH, its parent, or child tools/."
+    fi
+  fi
+}
+
+dump_obu_available() {
+  if [ "$(av1_decode_available)" = "yes" ] && \
+     [ "$(av1_encode_available)" = "yes" ]; then
+    echo yes
+  fi
+}
+
+aomenc_available() {
+  if [ -x "$(aom_tool_path aomenc)" ]; then
+    echo yes
+  fi
+}
+
+encode_test_file() {
+  if [ "$(aomenc_available)" = "yes" ]; then
+    local encoder="$(aom_tool_path aomenc)"
+
+    eval "${encoder}" \
+      $(aomenc_encode_test_fast_params) \
+      $(yuv_raw_input) \
+      --ivf \
+      --output=${dump_obu_test_file} \
+      ${devnull}
+
+    if [ ! -e "${dump_obu_test_file}" ]; then
+      elog "dump_obu test input encode failed."
+      return 1
+    fi
+  fi
+}
+
+dump_obu() {
+  encode_test_file
+  eval $(aom_tool_path dump_obu) "${dump_obu_test_file}" ${devnull}
+}
+
+dump_obu_tests="dump_obu"
+
+run_tests dump_obu_verify_environment "${dump_obu_tests}"
diff --git a/third_party/aom/test/ec_test.cc b/third_party/aom/test/ec_test.cc
new file mode 100644
index 0000000000..e6a5ea63bf
--- /dev/null
+++ b/third_party/aom/test/ec_test.cc
@@ -0,0 +1,159 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include <cstdlib>
+
+#include "aom_dsp/entenc.h"
+#include "aom_dsp/entdec.h"
+
+TEST(EC_TEST, random_ec_test) {
+  od_ec_enc enc;
+  od_ec_dec dec;
+  int sz;
+  int i;
+  int ret;
+  unsigned int sym;
+  unsigned int seed;
+  unsigned char *ptr;
+  uint32_t ptr_sz;
+  char *seed_str;
+  ret = 0;
+  seed_str = getenv("EC_TEST_SEED");
+  if (seed_str) {
+    seed = atoi(seed_str);
+  } else {
+    seed = 0xdaa1a;
+  }
+  srand(seed);
+  od_ec_enc_init(&enc, 1);
+  /*Test compatibility between multiple different encode/decode routines.*/
+  for (i = 0; i < 409600; i++) {
+    unsigned *fz;
+    unsigned *fts;
+    unsigned *data;
+    unsigned *tell;
+    unsigned *enc_method;
+    int j;
+    sz = rand() / ((RAND_MAX >> (rand() % 9U)) + 1U);
+    fz = (unsigned *)malloc(sz * sizeof(*fz));
+    fts = (unsigned *)malloc(sz * sizeof(*fts));
+    data = (unsigned *)malloc(sz * sizeof(*data));
+    tell = (unsigned *)malloc((sz + 1) * sizeof(*tell));
+    enc_method = (unsigned *)malloc(sz * sizeof(*enc_method));
+    od_ec_enc_reset(&enc);
+    tell[0] = od_ec_enc_tell_frac(&enc);
+    for (j = 0; j < sz; j++) {
+      data[j] = rand() / ((RAND_MAX >> 1) + 1);
+
+      fts[j] = CDF_PROB_BITS;
+      fz[j] = (rand() % (CDF_PROB_TOP - 2)) >> (CDF_PROB_BITS - fts[j]);
+      fz[j] = OD_MAXI(fz[j], 1);
+      enc_method[j] = 3 + (rand() & 1);
+      switch (enc_method[j]) {
+        case 3: {
+          od_ec_encode_bool_q15(&enc, data[j],
+                                OD_ICDF(fz[j] << (CDF_PROB_BITS - fts[j])));
+          break;
+        }
+        case 4: {
+          uint16_t cdf[2];
+          cdf[0] = OD_ICDF(fz[j]);
+          cdf[1] = OD_ICDF(1U << fts[j]);
+          od_ec_encode_cdf_q15(&enc, data[j], cdf, 2);
+          break;
+        }
+      }
+
+      tell[j + 1] = od_ec_enc_tell_frac(&enc);
+    }
+    ptr = od_ec_enc_done(&enc, &ptr_sz);
+    EXPECT_GE(((od_ec_enc_tell(&enc) + 7U) >> 3), ptr_sz)
+        << "od_ec_enc_tell() lied: "
+           "there's "
+        << ptr_sz << " bytes instead of " << ((od_ec_enc_tell(&enc) + 7) >> 3)
+        << " (Random seed: " << seed << ")\n";
+    od_ec_dec_init(&dec, ptr, ptr_sz);
+    EXPECT_EQ(od_ec_dec_tell_frac(&dec), tell[0])
+        << "od_ec_dec_tell() mismatch between encoder and decoder "
+           "at symbol 0: "
+        << (unsigned)od_ec_dec_tell_frac(&dec) << " instead of " << tell[0]
+        << " (Random seed: " << seed << ").\n";
+    for (j = 0; j < sz; j++) {
+      int dec_method;
+      if (CDF_SHIFT == 0) {
+        dec_method = 3 + (rand() & 1);
+      } else {
+        dec_method = enc_method[j];
+      }
+      switch (dec_method) {
+        case 3: {
+          sym = od_ec_decode_bool_q15(
+              &dec, OD_ICDF(fz[j] << (CDF_PROB_BITS - fts[j])));
+          break;
+        }
+        case 4: {
+          uint16_t cdf[2];
+          cdf[0] = OD_ICDF(fz[j]);
+          cdf[1] = OD_ICDF(1U << fts[j]);
+          sym = od_ec_decode_cdf_q15(&dec, cdf, 2);
+          break;
+        }
+      }
+
+      EXPECT_EQ(sym, data[j])
+          << "Decoded " << sym << " instead of " << data[j]
+          << " with fz=" << fz[j] << " and ftb=" << fts[j] << "at position "
+          << j << " of " << sz << " (Random seed: " << seed << ").\n"
+          << "Encoding method: " << enc_method[j]
+          << " decoding method: " << dec_method << "\n";
+      EXPECT_EQ(od_ec_dec_tell_frac(&dec), tell[j + 1])
+          << "od_ec_dec_tell() mismatch between encoder and "
+             "decoder at symbol "
+          << j + 1 << ": " << (unsigned)od_ec_dec_tell_frac(&dec)
+          << " instead of " << tell[j + 1] << " (Random seed: " << seed
+          << ").\n";
+    }
+    free(enc_method);
+    free(tell);
+    free(data);
+    free(fts);
+    free(fz);
+  }
+  od_ec_enc_reset(&enc);
+  if (CDF_SHIFT == 0) {
+    od_ec_encode_bool_q15(&enc, 0, OD_ICDF(16384));
+    od_ec_encode_bool_q15(&enc, 0, OD_ICDF(16384));
+    od_ec_encode_bool_q15(&enc, 0, OD_ICDF(16384));
+    od_ec_encode_bool_q15(&enc, 0, OD_ICDF(16384));
+    od_ec_encode_bool_q15(&enc, 0, OD_ICDF(24576));
+    od_ec_enc_patch_initial_bits(&enc, 3, 2);
+    EXPECT_FALSE(enc.error) << "od_ec_enc_patch_initial_bits() failed.\n";
+    od_ec_enc_patch_initial_bits(&enc, 0, 5);
+    EXPECT_TRUE(enc.error)
+        << "od_ec_enc_patch_initial_bits() didn't fail when it should have.\n";
+    od_ec_enc_reset(&enc);
+    od_ec_encode_bool_q15(&enc, 0, OD_ICDF(16384));
+    od_ec_encode_bool_q15(&enc, 0, OD_ICDF(16384));
+    od_ec_encode_bool_q15(&enc, 1, OD_ICDF(32256));
+    od_ec_encode_bool_q15(&enc, 0, OD_ICDF(24576));
+    od_ec_enc_patch_initial_bits(&enc, 0, 2);
+    EXPECT_FALSE(enc.error) << "od_ec_enc_patch_initial_bits() failed.\n";
+    ptr = od_ec_enc_done(&enc, &ptr_sz);
+    EXPECT_EQ(ptr_sz, 2u);
+    EXPECT_EQ(ptr[0], 63)
+        << "Got " << ptr[0]
+        << " when expecting 63 for od_ec_enc_patch_initial_bits().\n";
+  }
+  od_ec_enc_clear(&enc);
+  EXPECT_EQ(ret, 0);
+}
diff --git a/third_party/aom/test/encode_api_test.cc b/third_party/aom/test/encode_api_test.cc
new file mode 100644
index 0000000000..c26f5720f2
--- /dev/null
+++ b/third_party/aom/test/encode_api_test.cc
@@ -0,0 +1,73 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#include "test/util.h"
+#include "aom/aomcx.h"
+#include "aom/aom_encoder.h"
+
+namespace {
+
+TEST(EncodeAPI, InvalidParams) {
+  static const aom_codec_iface_t *kCodecs[] = {
+#if CONFIG_AV1_ENCODER
+    aom_codec_av1_cx(),
+#endif
+  };
+  uint8_t buf[1] = { 0 };
+  aom_image_t img;
+  aom_codec_ctx_t enc;
+  aom_codec_enc_cfg_t cfg;
+
+  EXPECT_EQ(&img, aom_img_wrap(&img, AOM_IMG_FMT_I420, 1, 1, 1, buf));
+
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_enc_init(NULL, NULL, NULL, 0));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_enc_init(&enc, NULL, NULL, 0));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_encode(NULL, NULL, 0, 0, 0));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_encode(NULL, &img, 0, 0, 0));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM, aom_codec_destroy(NULL));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+            aom_codec_enc_config_default(NULL, NULL, 0));
+  EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+            aom_codec_enc_config_default(NULL, &cfg, 0));
+  EXPECT_TRUE(aom_codec_error(NULL) != NULL);
+
+  for (int i = 0; i < NELEMENTS(kCodecs); ++i) {
+    SCOPED_TRACE(aom_codec_iface_name(kCodecs[i]));
+    EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+              aom_codec_enc_init(NULL, kCodecs[i], NULL, 0));
+    EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+              aom_codec_enc_init(&enc, kCodecs[i], NULL, 0));
+    EXPECT_EQ(AOM_CODEC_INVALID_PARAM,
+              aom_codec_enc_config_default(kCodecs[i], &cfg, 1));
+
+    EXPECT_EQ(AOM_CODEC_OK, aom_codec_enc_config_default(kCodecs[i], &cfg, 0));
+    EXPECT_EQ(AOM_CODEC_OK, aom_codec_enc_init(&enc, kCodecs[i], &cfg, 0));
+
+    EXPECT_EQ(NULL, aom_codec_get_global_headers(NULL));
+
+    aom_fixed_buf_t *glob_headers = aom_codec_get_global_headers(&enc);
+    EXPECT_TRUE(glob_headers->buf != NULL);
+    if (glob_headers) {
+      free(glob_headers->buf);
+      free(glob_headers);
+    }
+
+    EXPECT_EQ(AOM_CODEC_OK, aom_codec_encode(&enc, NULL, 0, 0, 0));
+
+    EXPECT_EQ(AOM_CODEC_OK, aom_codec_destroy(&enc));
+  }
+}
+
+}  // namespace
diff --git a/third_party/aom/test/encode_perf_test.cc b/third_party/aom/test/encode_perf_test.cc
new file mode 100644
index 0000000000..fe649b1539
--- /dev/null
+++ b/third_party/aom/test/encode_perf_test.cc
@@ -0,0 +1,188 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_version.h"
+
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+#include "aom_ports/aom_timer.h"
+
+namespace {
+
+const int kMaxPsnr = 100;
+const double kUsecsInSec = 1000000.0;
+
+struct EncodePerfTestVideo {
+  EncodePerfTestVideo(const char *name_, uint32_t width_, uint32_t height_,
+                      uint32_t bitrate_, int frames_)
+      : name(name_), width(width_), height(height_), bitrate(bitrate_),
+        frames(frames_) {}
+  const char *name;
+  uint32_t width;
+  uint32_t height;
+  uint32_t bitrate;
+  int frames;
+};
+
+const EncodePerfTestVideo kAV1EncodePerfTestVectors[] = {
+  EncodePerfTestVideo("desktop_640_360_30.yuv", 640, 360, 200, 2484),
+  EncodePerfTestVideo("kirland_640_480_30.yuv", 640, 480, 200, 300),
+  EncodePerfTestVideo("macmarcomoving_640_480_30.yuv", 640, 480, 200, 987),
+  EncodePerfTestVideo("macmarcostationary_640_480_30.yuv", 640, 480, 200, 718),
+  EncodePerfTestVideo("niklas_640_480_30.yuv", 640, 480, 200, 471),
+  EncodePerfTestVideo("tacomanarrows_640_480_30.yuv", 640, 480, 200, 300),
+  EncodePerfTestVideo("tacomasmallcameramovement_640_480_30.yuv", 640, 480, 200,
+                      300),
+  EncodePerfTestVideo("thaloundeskmtg_640_480_30.yuv", 640, 480, 200, 300),
+  EncodePerfTestVideo("niklas_1280_720_30.yuv", 1280, 720, 600, 470),
+};
+
+const int kEncodePerfTestSpeeds[] = { 5, 6, 7, 8 };
+const int kEncodePerfTestThreads[] = { 1, 2, 4 };
+
+class AV1EncodePerfTest
+    : public ::libaom_test::CodecTestWithParam<libaom_test::TestMode>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  AV1EncodePerfTest()
+      : EncoderTest(GET_PARAM(0)), min_psnr_(kMaxPsnr), nframes_(0),
+        encoding_mode_(GET_PARAM(1)), speed_(0), threads_(1) {}
+
+  virtual ~AV1EncodePerfTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+
+    cfg_.g_lag_in_frames = 0;
+    cfg_.rc_min_quantizer = 2;
+    cfg_.rc_max_quantizer = 56;
+    cfg_.rc_dropframe_thresh = 0;
+    cfg_.rc_undershoot_pct = 50;
+    cfg_.rc_overshoot_pct = 50;
+    cfg_.rc_buf_sz = 1000;
+    cfg_.rc_buf_initial_sz = 500;
+    cfg_.rc_buf_optimal_sz = 600;
+    cfg_.rc_end_usage = AOM_CBR;
+    cfg_.g_error_resilient = 1;
+    cfg_.g_threads = threads_;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 0) {
+      const int log2_tile_columns = 3;
+      encoder->Control(AOME_SET_CPUUSED, speed_);
+      encoder->Control(AV1E_SET_TILE_COLUMNS, log2_tile_columns);
+      encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 0);
+    }
+  }
+
+  virtual void BeginPassHook(unsigned int /*pass*/) {
+    min_psnr_ = kMaxPsnr;
+    nframes_ = 0;
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (pkt->data.psnr.psnr[0] < min_psnr_) {
+      min_psnr_ = pkt->data.psnr.psnr[0];
+    }
+  }
+
+  // for performance reasons don't decode
+  virtual bool DoDecode() { return 0; }
+
+  double min_psnr() const { return min_psnr_; }
+
+  void set_speed(unsigned int speed) { speed_ = speed; }
+
+  void set_threads(unsigned int threads) { threads_ = threads; }
+
+ private:
+  double min_psnr_;
+  unsigned int nframes_;
+  libaom_test::TestMode encoding_mode_;
+  unsigned speed_;
+  unsigned int threads_;
+};
+
+TEST_P(AV1EncodePerfTest, PerfTest) {
+  for (size_t i = 0; i < NELEMENTS(kAV1EncodePerfTestVectors); ++i) {
+    for (size_t j = 0; j < NELEMENTS(kEncodePerfTestSpeeds); ++j) {
+      for (size_t k = 0; k < NELEMENTS(kEncodePerfTestThreads); ++k) {
+        if (kAV1EncodePerfTestVectors[i].width < 512 &&
+            kEncodePerfTestThreads[k] > 1)
+          continue;
+        else if (kAV1EncodePerfTestVectors[i].width < 1024 &&
+                 kEncodePerfTestThreads[k] > 2)
+          continue;
+
+        set_threads(kEncodePerfTestThreads[k]);
+        SetUp();
+
+        const aom_rational timebase = { 33333333, 1000000000 };
+        cfg_.g_timebase = timebase;
+        cfg_.rc_target_bitrate = kAV1EncodePerfTestVectors[i].bitrate;
+
+        init_flags_ = AOM_CODEC_USE_PSNR;
+
+        const unsigned frames = kAV1EncodePerfTestVectors[i].frames;
+        const char *video_name = kAV1EncodePerfTestVectors[i].name;
+        libaom_test::I420VideoSource video(
+            video_name, kAV1EncodePerfTestVectors[i].width,
+            kAV1EncodePerfTestVectors[i].height, timebase.den, timebase.num, 0,
+            kAV1EncodePerfTestVectors[i].frames);
+        set_speed(kEncodePerfTestSpeeds[j]);
+
+        aom_usec_timer t;
+        aom_usec_timer_start(&t);
+
+        ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+        aom_usec_timer_mark(&t);
+        const double elapsed_secs = aom_usec_timer_elapsed(&t) / kUsecsInSec;
+        const double fps = frames / elapsed_secs;
+        const double minimum_psnr = min_psnr();
+        std::string display_name(video_name);
+        if (kEncodePerfTestThreads[k] > 1) {
+          char thread_count[32];
+          snprintf(thread_count, sizeof(thread_count), "_t-%d",
+                   kEncodePerfTestThreads[k]);
+          display_name += thread_count;
+        }
+
+        printf("{\n");
+        printf("\t\"type\" : \"encode_perf_test\",\n");
+        printf("\t\"version\" : \"%s\",\n", VERSION_STRING_NOSP);
+        printf("\t\"videoName\" : \"%s\",\n", display_name.c_str());
+        printf("\t\"encodeTimeSecs\" : %f,\n", elapsed_secs);
+        printf("\t\"totalFrames\" : %u,\n", frames);
+        printf("\t\"framesPerSecond\" : %f,\n", fps);
+        printf("\t\"minPsnr\" : %f,\n", minimum_psnr);
+        printf("\t\"speed\" : %d,\n", kEncodePerfTestSpeeds[j]);
+        printf("\t\"threads\" : %d\n", kEncodePerfTestThreads[k]);
+        printf("}\n");
+      }
+    }
+  }
+}
+
+AV1_INSTANTIATE_TEST_CASE(AV1EncodePerfTest,
+                          ::testing::Values(::libaom_test::kRealTime));
+}  // namespace
diff --git a/third_party/aom/test/encode_test_driver.cc b/third_party/aom/test/encode_test_driver.cc
new file mode 100644
index 0000000000..f3d61dc36a
--- /dev/null
+++ b/third_party/aom/test/encode_test_driver.cc
@@ -0,0 +1,288 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#include "aom_ports/mem.h"
+#include "test/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/encode_test_driver.h"
+#include "test/register_state_check.h"
+#include "test/video_source.h"
+
+namespace libaom_test {
+void Encoder::InitEncoder(VideoSource *video) {
+  aom_codec_err_t res;
+  const aom_image_t *img = video->img();
+
+  if (video->img() && !encoder_.priv) {
+    cfg_.g_w = img->d_w;
+    cfg_.g_h = img->d_h;
+    cfg_.g_timebase = video->timebase();
+    cfg_.rc_twopass_stats_in = stats_->buf();
+
+    res = aom_codec_enc_init(&encoder_, CodecInterface(), &cfg_, init_flags_);
+    ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+  }
+}
+
+void Encoder::EncodeFrame(VideoSource *video, const unsigned long frame_flags) {
+  if (video->img())
+    EncodeFrameInternal(*video, frame_flags);
+  else
+    Flush();
+
+  // Handle twopass stats
+  CxDataIterator iter = GetCxData();
+
+  while (const aom_codec_cx_pkt_t *pkt = iter.Next()) {
+    if (pkt->kind != AOM_CODEC_STATS_PKT) continue;
+
+    stats_->Append(*pkt);
+  }
+}
+
+void Encoder::EncodeFrameInternal(const VideoSource &video,
+                                  const unsigned long frame_flags) {
+  aom_codec_err_t res;
+  const aom_image_t *img = video.img();
+
+  // Handle frame resizing
+  if (cfg_.g_w != img->d_w || cfg_.g_h != img->d_h) {
+    cfg_.g_w = img->d_w;
+    cfg_.g_h = img->d_h;
+    res = aom_codec_enc_config_set(&encoder_, &cfg_);
+    ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+  }
+
+  // Encode the frame
+  API_REGISTER_STATE_CHECK(res =
+                               aom_codec_encode(&encoder_, img, video.pts(),
+                                                video.duration(), frame_flags));
+  ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+}
+
+void Encoder::Flush() {
+  const aom_codec_err_t res = aom_codec_encode(&encoder_, NULL, 0, 0, 0);
+  if (!encoder_.priv)
+    ASSERT_EQ(AOM_CODEC_ERROR, res) << EncoderError();
+  else
+    ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+}
+
+void EncoderTest::InitializeConfig() {
+  const aom_codec_err_t res = codec_->DefaultEncoderConfig(&cfg_, 0);
+  ASSERT_EQ(AOM_CODEC_OK, res);
+}
+
+void EncoderTest::SetMode(TestMode mode) {
+  switch (mode) {
+    case kOnePassGood:
+    case kTwoPassGood: break;
+    case kRealTime: cfg_.g_lag_in_frames = 0; break;
+    default: ASSERT_TRUE(false) << "Unexpected mode " << mode;
+  }
+  mode_ = mode;
+  if (mode == kTwoPassGood)
+    passes_ = 2;
+  else
+    passes_ = 1;
+}
+
+static bool compare_plane(const uint8_t *const buf1, int stride1,
+                          const uint8_t *const buf2, int stride2, int w, int h,
+                          int *const mismatch_row, int *const mismatch_col,
+                          int *const mismatch_pix1, int *const mismatch_pix2) {
+  int r, c;
+
+  for (r = 0; r < h; ++r) {
+    for (c = 0; c < w; ++c) {
+      const int pix1 = buf1[r * stride1 + c];
+      const int pix2 = buf2[r * stride2 + c];
+
+      if (pix1 != pix2) {
+        if (mismatch_row != NULL) *mismatch_row = r;
+        if (mismatch_col != NULL) *mismatch_col = c;
+        if (mismatch_pix1 != NULL) *mismatch_pix1 = pix1;
+        if (mismatch_pix2 != NULL) *mismatch_pix2 = pix2;
+        return false;
+      }
+    }
+  }
+
+  return true;
+}
+
+// The function should return "true" most of the time, therefore no early
+// break-out is implemented within the match checking process.
+static bool compare_img(const aom_image_t *img1, const aom_image_t *img2,
+                        int *const mismatch_row, int *const mismatch_col,
+                        int *const mismatch_plane, int *const mismatch_pix1,
+                        int *const mismatch_pix2) {
+  if (img1->fmt != img2->fmt || img1->cp != img2->cp || img1->tc != img2->tc ||
+      img1->mc != img2->mc || img1->d_w != img2->d_w ||
+      img1->d_h != img2->d_h || img1->monochrome != img2->monochrome) {
+    if (mismatch_row != NULL) *mismatch_row = -1;
+    if (mismatch_col != NULL) *mismatch_col = -1;
+    return false;
+  }
+
+  const int num_planes = img1->monochrome ? 1 : 3;
+  for (int plane = 0; plane < num_planes; plane++) {
+    if (!compare_plane(img1->planes[plane], img1->stride[plane],
+                       img2->planes[plane], img2->stride[plane],
+                       aom_img_plane_width(img1, plane),
+                       aom_img_plane_height(img1, plane), mismatch_row,
+                       mismatch_col, mismatch_pix1, mismatch_pix2)) {
+      if (mismatch_plane != NULL) *mismatch_plane = plane;
+      return false;
+    }
+  }
+
+  return true;
+}
+
+void EncoderTest::MismatchHook(const aom_image_t *img_enc,
+                               const aom_image_t *img_dec) {
+  int mismatch_row = 0;
+  int mismatch_col = 0;
+  int mismatch_plane = 0;
+  int mismatch_pix_enc = 0;
+  int mismatch_pix_dec = 0;
+
+  ASSERT_FALSE(compare_img(img_enc, img_dec, &mismatch_row, &mismatch_col,
+                           &mismatch_plane, &mismatch_pix_enc,
+                           &mismatch_pix_dec));
+
+  GTEST_FAIL() << "Encode/Decode mismatch found:" << std::endl
+               << "  pixel value enc/dec: " << mismatch_pix_enc << "/"
+               << mismatch_pix_dec << std::endl
+               << "                plane: " << mismatch_plane << std::endl
+               << "              row/col: " << mismatch_row << "/"
+               << mismatch_col << std::endl;
+}
+
+void EncoderTest::RunLoop(VideoSource *video) {
+  aom_codec_dec_cfg_t dec_cfg = aom_codec_dec_cfg_t();
+  dec_cfg.allow_lowbitdepth = 1;
+
+  stats_.Reset();
+
+  ASSERT_TRUE(passes_ == 1 || passes_ == 2);
+  for (unsigned int pass = 0; pass < passes_; pass++) {
+    last_pts_ = 0;
+
+    if (passes_ == 1)
+      cfg_.g_pass = AOM_RC_ONE_PASS;
+    else if (pass == 0)
+      cfg_.g_pass = AOM_RC_FIRST_PASS;
+    else
+      cfg_.g_pass = AOM_RC_LAST_PASS;
+
+    BeginPassHook(pass);
+    testing::internal::scoped_ptr<Encoder> encoder(
+        codec_->CreateEncoder(cfg_, init_flags_, &stats_));
+    ASSERT_TRUE(encoder.get() != NULL);
+
+    ASSERT_NO_FATAL_FAILURE(video->Begin());
+    encoder->InitEncoder(video);
+
+    if (mode_ == kRealTime) {
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 0);
+    }
+
+    ASSERT_FALSE(::testing::Test::HasFatalFailure());
+
+    testing::internal::scoped_ptr<Decoder> decoder(
+        codec_->CreateDecoder(dec_cfg, 0 /* flags */));
+#if CONFIG_AV1_DECODER
+    if (decoder->IsAV1()) {
+      // Set dec_cfg.tile_row = -1 and dec_cfg.tile_col = -1 so that the whole
+      // frame is decoded.
+      decoder->Control(AV1_SET_TILE_MODE, cfg_.large_scale_tile);
+      decoder->Control(AV1D_EXT_TILE_DEBUG, 1);
+      decoder->Control(AV1_SET_DECODE_TILE_ROW, -1);
+      decoder->Control(AV1_SET_DECODE_TILE_COL, -1);
+    }
+#endif
+
+    bool again;
+    for (again = true; again; video->Next()) {
+      again = (video->img() != NULL);
+
+      PreEncodeFrameHook(video);
+      PreEncodeFrameHook(video, encoder.get());
+      encoder->EncodeFrame(video, frame_flags_);
+
+      CxDataIterator iter = encoder->GetCxData();
+
+      bool has_cxdata = false;
+      bool has_dxdata = false;
+      while (const aom_codec_cx_pkt_t *pkt = iter.Next()) {
+        pkt = MutateEncoderOutputHook(pkt);
+        again = true;
+        switch (pkt->kind) {
+          case AOM_CODEC_CX_FRAME_PKT:
+            has_cxdata = true;
+            if (decoder.get() != NULL && DoDecode()) {
+              aom_codec_err_t res_dec;
+              if (DoDecodeInvisible()) {
+                res_dec = decoder->DecodeFrame(
+                    (const uint8_t *)pkt->data.frame.buf, pkt->data.frame.sz);
+              } else {
+                res_dec = decoder->DecodeFrame(
+                    (const uint8_t *)pkt->data.frame.buf +
+                        (pkt->data.frame.sz - pkt->data.frame.vis_frame_size),
+                    pkt->data.frame.vis_frame_size);
+              }
+
+              if (!HandleDecodeResult(res_dec, decoder.get())) break;
+
+              has_dxdata = true;
+            }
+            ASSERT_GE(pkt->data.frame.pts, last_pts_);
+            last_pts_ = pkt->data.frame.pts;
+            FramePktHook(pkt);
+            break;
+
+          case AOM_CODEC_PSNR_PKT: PSNRPktHook(pkt); break;
+
+          default: break;
+        }
+      }
+
+      if (has_dxdata && has_cxdata) {
+        const aom_image_t *img_enc = encoder->GetPreviewFrame();
+        DxDataIterator dec_iter = decoder->GetDxData();
+        const aom_image_t *img_dec = dec_iter.Next();
+        if (img_enc && img_dec) {
+          const bool res =
+              compare_img(img_enc, img_dec, NULL, NULL, NULL, NULL, NULL);
+          if (!res) {  // Mismatch
+            MismatchHook(img_enc, img_dec);
+          }
+        }
+        if (img_dec) DecompressedFrameHook(*img_dec, video->pts());
+      }
+      if (!Continue()) break;
+    }
+
+    EndPassHook();
+
+    if (!Continue()) break;
+  }
+}
+
+}  // namespace libaom_test
diff --git a/third_party/aom/test/encode_test_driver.h b/third_party/aom/test/encode_test_driver.h
new file mode 100644
index 0000000000..4f3f855cf9
--- /dev/null
+++ b/third_party/aom/test/encode_test_driver.h
@@ -0,0 +1,249 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_ENCODE_TEST_DRIVER_H_
+#define AOM_TEST_ENCODE_TEST_DRIVER_H_
+
+#include <string>
+#include <vector>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#if CONFIG_AV1_ENCODER
+#include "aom/aomcx.h"
+#endif
+#include "aom/aom_encoder.h"
+
+namespace libaom_test {
+
+class CodecFactory;
+class VideoSource;
+
+enum TestMode { kRealTime, kOnePassGood, kTwoPassGood };
+#define ALL_TEST_MODES                                                     \
+  ::testing::Values(::libaom_test::kRealTime, ::libaom_test::kOnePassGood, \
+                    ::libaom_test::kTwoPassGood)
+
+#define ONE_PASS_TEST_MODES \
+  ::testing::Values(::libaom_test::kRealTime, ::libaom_test::kOnePassGood)
+
+#define TWO_PASS_TEST_MODES ::testing::Values(::libaom_test::kTwoPassGood)
+
+#define NONREALTIME_TEST_MODES \
+  ::testing::Values(::libaom_test::kOnePassGood, ::libaom_test::kTwoPassGood)
+
+// Provides an object to handle the libaom get_cx_data() iteration pattern
+class CxDataIterator {
+ public:
+  explicit CxDataIterator(aom_codec_ctx_t *encoder)
+      : encoder_(encoder), iter_(NULL) {}
+
+  const aom_codec_cx_pkt_t *Next() {
+    return aom_codec_get_cx_data(encoder_, &iter_);
+  }
+
+ private:
+  aom_codec_ctx_t *encoder_;
+  aom_codec_iter_t iter_;
+};
+
+// Implements an in-memory store for libaom twopass statistics
+class TwopassStatsStore {
+ public:
+  void Append(const aom_codec_cx_pkt_t &pkt) {
+    buffer_.append(reinterpret_cast<char *>(pkt.data.twopass_stats.buf),
+                   pkt.data.twopass_stats.sz);
+  }
+
+  aom_fixed_buf_t buf() {
+    const aom_fixed_buf_t buf = { &buffer_[0], buffer_.size() };
+    return buf;
+  }
+
+  void Reset() { buffer_.clear(); }
+
+ protected:
+  std::string buffer_;
+};
+
+// Provides a simplified interface to manage one video encoding pass, given
+// a configuration and video source.
+//
+// TODO(jkoleszar): The exact services it provides and the appropriate
+// level of abstraction will be fleshed out as more tests are written.
+class Encoder {
+ public:
+  Encoder(aom_codec_enc_cfg_t cfg, const uint32_t init_flags,
+          TwopassStatsStore *stats)
+      : cfg_(cfg), init_flags_(init_flags), stats_(stats) {
+    memset(&encoder_, 0, sizeof(encoder_));
+  }
+
+  virtual ~Encoder() { aom_codec_destroy(&encoder_); }
+
+  CxDataIterator GetCxData() { return CxDataIterator(&encoder_); }
+
+  void InitEncoder(VideoSource *video);
+
+  const aom_image_t *GetPreviewFrame() {
+    return aom_codec_get_preview_frame(&encoder_);
+  }
+  // This is a thin wrapper around aom_codec_encode(), so refer to
+  // aom_encoder.h for its semantics.
+  void EncodeFrame(VideoSource *video, const unsigned long frame_flags);
+
+  // Convenience wrapper for EncodeFrame()
+  void EncodeFrame(VideoSource *video) { EncodeFrame(video, 0); }
+
+  void Control(int ctrl_id, int arg) {
+    const aom_codec_err_t res = aom_codec_control_(&encoder_, ctrl_id, arg);
+    ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+  }
+
+  void Control(int ctrl_id, int *arg) {
+    const aom_codec_err_t res = aom_codec_control_(&encoder_, ctrl_id, arg);
+    ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+  }
+
+  void Control(int ctrl_id, struct aom_scaling_mode *arg) {
+    const aom_codec_err_t res = aom_codec_control_(&encoder_, ctrl_id, arg);
+    ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+  }
+
+#if CONFIG_AV1_ENCODER
+  void Control(int ctrl_id, aom_active_map_t *arg) {
+    const aom_codec_err_t res = aom_codec_control_(&encoder_, ctrl_id, arg);
+    ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+  }
+#endif
+
+  void Config(const aom_codec_enc_cfg_t *cfg) {
+    const aom_codec_err_t res = aom_codec_enc_config_set(&encoder_, cfg);
+    ASSERT_EQ(AOM_CODEC_OK, res) << EncoderError();
+    cfg_ = *cfg;
+  }
+
+ protected:
+  virtual aom_codec_iface_t *CodecInterface() const = 0;
+
+  const char *EncoderError() {
+    const char *detail = aom_codec_error_detail(&encoder_);
+    return detail ? detail : aom_codec_error(&encoder_);
+  }
+
+  // Encode an image
+  void EncodeFrameInternal(const VideoSource &video,
+                           const unsigned long frame_flags);
+
+  // Flush the encoder on EOS
+  void Flush();
+
+  aom_codec_ctx_t encoder_;
+  aom_codec_enc_cfg_t cfg_;
+  unsigned long init_flags_;
+  TwopassStatsStore *stats_;
+};
+
+// Common test functionality for all Encoder tests.
+//
+// This class is a mixin which provides the main loop common to all
+// encoder tests. It provides hooks which can be overridden by subclasses
+// to implement each test's specific behavior, while centralizing the bulk
+// of the boilerplate. Note that it doesn't inherit the gtest testing
+// classes directly, so that tests can be parameterized differently.
+class EncoderTest {
+ protected:
+  explicit EncoderTest(const CodecFactory *codec)
+      : codec_(codec), abort_(false), init_flags_(0), frame_flags_(0),
+        last_pts_(0), mode_(kRealTime) {
+    // Default to 1 thread.
+    cfg_.g_threads = 1;
+  }
+
+  virtual ~EncoderTest() {}
+
+  // Initialize the cfg_ member with the default configuration.
+  void InitializeConfig();
+
+  // Map the TestMode enum to the passes_ variables.
+  void SetMode(TestMode mode);
+
+  // Set encoder flag.
+  void set_init_flags(unsigned long flag) {  // NOLINT(runtime/int)
+    init_flags_ = flag;
+  }
+
+  // Main loop
+  virtual void RunLoop(VideoSource *video);
+
+  // Hook to be called at the beginning of a pass.
+  virtual void BeginPassHook(unsigned int /*pass*/) {}
+
+  // Hook to be called at the end of a pass.
+  virtual void EndPassHook() {}
+
+  // Hook to be called before encoding a frame.
+  virtual void PreEncodeFrameHook(VideoSource * /*video*/) {}
+  virtual void PreEncodeFrameHook(VideoSource * /*video*/,
+                                  Encoder * /*encoder*/) {}
+
+  // Hook to be called on every compressed data packet.
+  virtual void FramePktHook(const aom_codec_cx_pkt_t * /*pkt*/) {}
+
+  // Hook to be called on every PSNR packet.
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t * /*pkt*/) {}
+
+  // Hook to determine whether the encode loop should continue.
+  virtual bool Continue() const {
+    return !(::testing::Test::HasFatalFailure() || abort_);
+  }
+
+  // Hook to determine whether to decode frame after encoding
+  virtual bool DoDecode() const { return true; }
+
+  // Hook to determine whether to decode invisible frames after encoding
+  virtual bool DoDecodeInvisible() const { return true; }
+
+  // Hook to handle encode/decode mismatch
+  virtual void MismatchHook(const aom_image_t *img1, const aom_image_t *img2);
+
+  // Hook to be called on every decompressed frame.
+  virtual void DecompressedFrameHook(const aom_image_t & /*img*/,
+                                     aom_codec_pts_t /*pts*/) {}
+
+  // Hook to be called to handle decode result. Return true to continue.
+  virtual bool HandleDecodeResult(const aom_codec_err_t res_dec,
+                                  Decoder *decoder) {
+    EXPECT_EQ(AOM_CODEC_OK, res_dec) << decoder->DecodeError();
+    return AOM_CODEC_OK == res_dec;
+  }
+
+  // Hook that can modify the encoder's output data
+  virtual const aom_codec_cx_pkt_t *MutateEncoderOutputHook(
+      const aom_codec_cx_pkt_t *pkt) {
+    return pkt;
+  }
+
+  const CodecFactory *codec_;
+  bool abort_;
+  aom_codec_enc_cfg_t cfg_;
+  unsigned int passes_;
+  TwopassStatsStore stats_;
+  unsigned long init_flags_;
+  unsigned long frame_flags_;
+  aom_codec_pts_t last_pts_;
+  TestMode mode_;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_ENCODE_TEST_DRIVER_H_
diff --git a/third_party/aom/test/encodetxb_test.cc b/third_party/aom/test/encodetxb_test.cc
new file mode 100644
index 0000000000..11cc073682
--- /dev/null
+++ b/third_party/aom/test/encodetxb_test.cc
@@ -0,0 +1,262 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/aom_timer.h"
+#include "aom_ports/mem.h"
+#include "av1/common/idct.h"
+#include "av1/common/onyxc_int.h"
+#include "av1/common/scan.h"
+#include "av1/common/txb_common.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+namespace {
+using libaom_test::ACMRandom;
+
+typedef void (*GetNzMapContextsFunc)(const uint8_t *const levels,
+                                     const int16_t *const scan,
+                                     const uint16_t eob, const TX_SIZE tx_size,
+                                     const TX_CLASS tx_class,
+                                     int8_t *const coeff_contexts);
+
+class EncodeTxbTest : public ::testing::TestWithParam<GetNzMapContextsFunc> {
+ public:
+  EncodeTxbTest() : get_nz_map_contexts_func_(GetParam()) {}
+
+  virtual ~EncodeTxbTest() {}
+
+  virtual void SetUp() {
+    coeff_contexts_ref_ = reinterpret_cast<int8_t *>(
+        aom_memalign(16, sizeof(*coeff_contexts_ref_) * MAX_TX_SQUARE));
+    ASSERT_TRUE(coeff_contexts_ref_ != NULL);
+    coeff_contexts_ = reinterpret_cast<int8_t *>(
+        aom_memalign(16, sizeof(*coeff_contexts_) * MAX_TX_SQUARE));
+    ASSERT_TRUE(coeff_contexts_ != NULL);
+  }
+
+  virtual void TearDown() {
+    aom_free(coeff_contexts_ref_);
+    aom_free(coeff_contexts_);
+    libaom_test::ClearSystemState();
+  }
+
+  void GetNzMapContextsRun() {
+    const int kNumTests = 10;
+    int result = 0;
+
+    for (int is_inter = 0; is_inter < 2; ++is_inter) {
+      for (int tx_type = DCT_DCT; tx_type < TX_TYPES; ++tx_type) {
+        const TX_CLASS tx_class = tx_type_to_class[tx_type];
+        for (int tx_size = TX_4X4; tx_size < TX_SIZES_ALL; ++tx_size) {
+          const int bwl = get_txb_bwl((TX_SIZE)tx_size);
+          const int width = get_txb_wide((TX_SIZE)tx_size);
+          const int height = get_txb_high((TX_SIZE)tx_size);
+          const int real_width = tx_size_wide[tx_size];
+          const int real_height = tx_size_high[tx_size];
+          const int16_t *const scan = av1_scan_orders[tx_size][tx_type].scan;
+
+          levels_ = set_levels(levels_buf_, width);
+          for (int i = 0; i < kNumTests && !result; ++i) {
+            for (int eob = 1; eob <= width * height && !result; ++eob) {
+              InitDataWithEob(scan, bwl, eob);
+
+              av1_get_nz_map_contexts_c(levels_, scan, eob, (TX_SIZE)tx_size,
+                                        tx_class, coeff_contexts_ref_);
+              get_nz_map_contexts_func_(levels_, scan, eob, (TX_SIZE)tx_size,
+                                        tx_class, coeff_contexts_);
+
+              result = Compare(scan, eob);
+
+              EXPECT_EQ(result, 0)
+                  << " tx_class " << tx_class << " width " << real_width
+                  << " height " << real_height << " eob " << eob;
+            }
+          }
+        }
+      }
+    }
+  }
+
+  void SpeedTestGetNzMapContextsRun() {
+    const int kNumTests = 2000000000;
+    aom_usec_timer timer;
+
+    printf("Note: Only test the largest possible eob case!\n");
+    for (int tx_size = TX_4X4; tx_size < TX_SIZES_ALL; ++tx_size) {
+      const int bwl = get_txb_bwl((TX_SIZE)tx_size);
+      const int width = get_txb_wide((TX_SIZE)tx_size);
+      const int height = get_txb_high((TX_SIZE)tx_size);
+      const int real_width = tx_size_wide[tx_size];
+      const int real_height = tx_size_high[tx_size];
+      const TX_TYPE tx_type = DCT_DCT;
+      const TX_CLASS tx_class = tx_type_to_class[tx_type];
+      const int16_t *const scan = av1_scan_orders[tx_size][tx_type].scan;
+      const int eob = width * height;
+      const int numTests = kNumTests / (width * height);
+
+      levels_ = set_levels(levels_buf_, width);
+      InitDataWithEob(scan, bwl, eob);
+
+      aom_usec_timer_start(&timer);
+      for (int i = 0; i < numTests; ++i) {
+        get_nz_map_contexts_func_(levels_, scan, eob, (TX_SIZE)tx_size,
+                                  tx_class, coeff_contexts_);
+      }
+      aom_usec_timer_mark(&timer);
+
+      const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+      printf("get_nz_map_contexts_%2dx%2d: %7.1f ms\n", real_width, real_height,
+             elapsed_time / 1000.0);
+    }
+  }
+
+ private:
+  void InitDataWithEob(const int16_t *const scan, const int bwl,
+                       const int eob) {
+    memset(levels_buf_, 0, sizeof(levels_buf_));
+    memset(coeff_contexts_, 0, sizeof(*coeff_contexts_) * MAX_TX_SQUARE);
+
+    for (int c = 0; c < eob; ++c) {
+      levels_[get_padded_idx(scan[c], bwl)] =
+          static_cast<uint8_t>(clamp(rnd_.Rand8(), 0, INT8_MAX));
+      coeff_contexts_[scan[c]] = rnd_.Rand16() >> 1;
+    }
+
+    memcpy(coeff_contexts_ref_, coeff_contexts_,
+           sizeof(*coeff_contexts_) * MAX_TX_SQUARE);
+  }
+
+  bool Compare(const int16_t *const scan, const int eob) const {
+    bool result = false;
+    if (memcmp(coeff_contexts_, coeff_contexts_ref_,
+               sizeof(*coeff_contexts_ref_) * MAX_TX_SQUARE)) {
+      for (int i = 0; i < eob; i++) {
+        const int pos = scan[i];
+        if (coeff_contexts_ref_[pos] != coeff_contexts_[pos]) {
+          printf("coeff_contexts_[%d] diff:%6d (ref),%6d (opt)\n", pos,
+                 coeff_contexts_ref_[pos], coeff_contexts_[pos]);
+          result = true;
+          break;
+        }
+      }
+    }
+    return result;
+  }
+
+  GetNzMapContextsFunc get_nz_map_contexts_func_;
+  ACMRandom rnd_;
+  uint8_t levels_buf_[TX_PAD_2D];
+  uint8_t *levels_;
+  int8_t *coeff_contexts_ref_;
+  int8_t *coeff_contexts_;
+};
+
+TEST_P(EncodeTxbTest, GetNzMapContexts) { GetNzMapContextsRun(); }
+
+TEST_P(EncodeTxbTest, DISABLED_SpeedTestGetNzMapContexts) {
+  SpeedTestGetNzMapContextsRun();
+}
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, EncodeTxbTest,
+                        ::testing::Values(av1_get_nz_map_contexts_sse2));
+#endif
+
+typedef void (*av1_txb_init_levels_func)(const tran_low_t *const coeff,
+                                         const int width, const int height,
+                                         uint8_t *const levels);
+
+typedef ::testing::tuple<av1_txb_init_levels_func, int> TxbInitLevelParam;
+
+class EncodeTxbInitLevelTest
+    : public ::testing::TestWithParam<TxbInitLevelParam> {
+ public:
+  virtual ~EncodeTxbInitLevelTest() {}
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+  void RunTest(av1_txb_init_levels_func test_func, int tx_size, int is_speed);
+};
+
+void EncodeTxbInitLevelTest::RunTest(av1_txb_init_levels_func test_func,
+                                     int tx_size, int is_speed) {
+  const int width = get_txb_wide((TX_SIZE)tx_size);
+  const int height = get_txb_high((TX_SIZE)tx_size);
+  tran_low_t coeff[MAX_TX_SQUARE];
+
+  uint8_t levels_buf[2][TX_PAD_2D];
+  uint8_t *const levels0 = set_levels(levels_buf[0], width);
+  uint8_t *const levels1 = set_levels(levels_buf[1], width);
+
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  for (int i = 0; i < width * height; i++) {
+    coeff[i] = rnd.Rand15Signed() + rnd.Rand15Signed();
+  }
+  for (int i = 0; i < TX_PAD_2D; i++) {
+    levels_buf[0][i] = rnd.Rand8();
+    levels_buf[1][i] = rnd.Rand8();
+  }
+  const int run_times = is_speed ? (width * height) * 10000 : 1;
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < run_times; ++i) {
+    av1_txb_init_levels_c(coeff, width, height, levels0);
+  }
+  const double t1 = get_time_mark(&timer);
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < run_times; ++i) {
+    test_func(coeff, width, height, levels1);
+  }
+  const double t2 = get_time_mark(&timer);
+  if (is_speed) {
+    printf("init %3dx%-3d:%7.2f/%7.2fns", width, height, t1, t2);
+    printf("(%3.2f)\n", t1 / t2);
+  }
+  const int stride = width + TX_PAD_HOR;
+  for (int r = 0; r < height + TX_PAD_VER; ++r) {
+    for (int c = 0; c < stride; ++c) {
+      ASSERT_EQ(levels_buf[0][c + r * stride], levels_buf[1][c + r * stride])
+          << "[" << r << "," << c << "] " << run_times << width << "x"
+          << height;
+    }
+  }
+}
+
+TEST_P(EncodeTxbInitLevelTest, match) {
+  RunTest(GET_PARAM(0), GET_PARAM(1), 0);
+}
+
+TEST_P(EncodeTxbInitLevelTest, DISABLED_Speed) {
+  RunTest(GET_PARAM(0), GET_PARAM(1), 1);
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, EncodeTxbInitLevelTest,
+    ::testing::Combine(::testing::Values(&av1_txb_init_levels_sse4_1),
+                       ::testing::Range(0, static_cast<int>(TX_SIZES_ALL), 1)));
+#endif
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, EncodeTxbInitLevelTest,
+    ::testing::Combine(::testing::Values(&av1_txb_init_levels_avx2),
+                       ::testing::Range(0, static_cast<int>(TX_SIZES_ALL), 1)));
+#endif
+}  // namespace
diff --git a/third_party/aom/test/end_to_end_test.cc b/third_party/aom/test/end_to_end_test.cc
new file mode 100644
index 0000000000..1ac0ae9312
--- /dev/null
+++ b/third_party/aom/test/end_to_end_test.cc
@@ -0,0 +1,199 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+#include "test/yuv_video_source.h"
+
+namespace {
+
+const unsigned int kWidth = 160;
+const unsigned int kHeight = 90;
+const unsigned int kFramerate = 50;
+const unsigned int kFrames = 10;
+const int kBitrate = 500;
+// List of psnr thresholds for speed settings 0-7 and 5 encoding modes
+const double kPsnrThreshold[][5] = {
+// Note:
+// AV1 HBD average PSNR is slightly lower than AV1.
+// We make two cases here to enable the testing and
+// guard picture quality.
+#if CONFIG_AV1_ENCODER
+  { 36.0, 37.0, 37.0, 37.0, 37.0 }, { 31.0, 36.0, 36.0, 36.0, 36.0 },
+  { 31.0, 35.0, 35.0, 35.0, 35.0 }, { 31.0, 34.0, 34.0, 34.0, 34.0 },
+  { 31.0, 33.0, 33.0, 33.0, 33.0 }, { 31.0, 32.0, 32.0, 32.0, 32.0 },
+  { 30.0, 31.0, 31.0, 31.0, 31.0 }, { 29.0, 30.0, 30.0, 30.0, 30.0 },
+#else
+  { 36.0, 37.0, 37.0, 37.0, 37.0 }, { 35.0, 36.0, 36.0, 36.0, 36.0 },
+  { 34.0, 35.0, 35.0, 35.0, 35.0 }, { 33.0, 34.0, 34.0, 34.0, 34.0 },
+  { 32.0, 33.0, 33.0, 33.0, 33.0 }, { 31.0, 32.0, 32.0, 32.0, 32.0 },
+  { 30.0, 31.0, 31.0, 31.0, 31.0 }, { 29.0, 30.0, 30.0, 30.0, 30.0 },
+#endif  // CONFIG_AV1_ENCODER
+};
+
+typedef struct {
+  const char *filename;
+  unsigned int input_bit_depth;
+  aom_img_fmt fmt;
+  aom_bit_depth_t bit_depth;
+  unsigned int profile;
+} TestVideoParam;
+
+const TestVideoParam kTestVectors[] = {
+  { "park_joy_90p_8_420.y4m", 8, AOM_IMG_FMT_I420, AOM_BITS_8, 0 },
+  { "park_joy_90p_8_422.y4m", 8, AOM_IMG_FMT_I422, AOM_BITS_8, 2 },
+  { "park_joy_90p_8_444.y4m", 8, AOM_IMG_FMT_I444, AOM_BITS_8, 1 },
+  { "park_joy_90p_10_420.y4m", 10, AOM_IMG_FMT_I42016, AOM_BITS_10, 0 },
+  { "park_joy_90p_10_422.y4m", 10, AOM_IMG_FMT_I42216, AOM_BITS_10, 2 },
+  { "park_joy_90p_10_444.y4m", 10, AOM_IMG_FMT_I44416, AOM_BITS_10, 1 },
+  { "park_joy_90p_12_420.y4m", 12, AOM_IMG_FMT_I42016, AOM_BITS_12, 2 },
+  { "park_joy_90p_12_422.y4m", 12, AOM_IMG_FMT_I42216, AOM_BITS_12, 2 },
+  { "park_joy_90p_12_444.y4m", 12, AOM_IMG_FMT_I44416, AOM_BITS_12, 2 },
+};
+
+// Encoding modes tested
+const libaom_test::TestMode kEncodingModeVectors[] = {
+  ::libaom_test::kTwoPassGood,
+  ::libaom_test::kOnePassGood,
+  ::libaom_test::kRealTime,
+};
+
+// Speed settings tested
+const int kCpuUsedVectors[] = { 1, 2, 3, 5, 6 };
+
+int is_extension_y4m(const char *filename) {
+  const char *dot = strrchr(filename, '.');
+  if (!dot || dot == filename)
+    return 0;
+  else
+    return !strcmp(dot, ".y4m");
+}
+
+class EndToEndTest
+    : public ::libaom_test::CodecTestWith3Params<libaom_test::TestMode,
+                                                 TestVideoParam, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  EndToEndTest()
+      : EncoderTest(GET_PARAM(0)), test_video_param_(GET_PARAM(2)),
+        cpu_used_(GET_PARAM(3)), psnr_(0.0), nframes_(0),
+        encoding_mode_(GET_PARAM(1)) {}
+
+  virtual ~EndToEndTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+    if (encoding_mode_ != ::libaom_test::kRealTime) {
+      cfg_.g_lag_in_frames = 5;
+      cfg_.rc_end_usage = AOM_VBR;
+    } else {
+      cfg_.g_lag_in_frames = 0;
+      cfg_.rc_end_usage = AOM_CBR;
+      cfg_.rc_buf_sz = 1000;
+      cfg_.rc_buf_initial_sz = 500;
+      cfg_.rc_buf_optimal_sz = 600;
+    }
+  }
+
+  virtual void BeginPassHook(unsigned int) {
+    psnr_ = 0.0;
+    nframes_ = 0;
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    psnr_ += pkt->data.psnr.psnr[0];
+    nframes_++;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
+      encoder->Control(AV1E_SET_TILE_COLUMNS, 4);
+      encoder->Control(AOME_SET_CPUUSED, cpu_used_);
+      // Test screen coding tools at cpu_used = 1 && encoding mode is two-pass.
+      if (cpu_used_ == 1 && encoding_mode_ == ::libaom_test::kTwoPassGood)
+        encoder->Control(AV1E_SET_TUNE_CONTENT, AOM_CONTENT_SCREEN);
+      else
+        encoder->Control(AV1E_SET_TUNE_CONTENT, AOM_CONTENT_DEFAULT);
+      if (encoding_mode_ != ::libaom_test::kRealTime) {
+        encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+        encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+        encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+      }
+    }
+  }
+
+  double GetAveragePsnr() const {
+    if (nframes_) return psnr_ / nframes_;
+    return 0.0;
+  }
+
+  double GetPsnrThreshold() {
+    return kPsnrThreshold[cpu_used_][encoding_mode_];
+  }
+
+  void DoTest() {
+    cfg_.rc_target_bitrate = kBitrate;
+    cfg_.g_error_resilient = 0;
+    cfg_.g_profile = test_video_param_.profile;
+    cfg_.g_input_bit_depth = test_video_param_.input_bit_depth;
+    cfg_.g_bit_depth = test_video_param_.bit_depth;
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    if (cfg_.g_bit_depth > 8) init_flags_ |= AOM_CODEC_USE_HIGHBITDEPTH;
+
+    testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+    if (is_extension_y4m(test_video_param_.filename)) {
+      video.reset(new libaom_test::Y4mVideoSource(test_video_param_.filename, 0,
+                                                  kFrames));
+    } else {
+      video.reset(new libaom_test::YUVVideoSource(
+          test_video_param_.filename, test_video_param_.fmt, kWidth, kHeight,
+          kFramerate, 1, 0, kFrames));
+    }
+    ASSERT_TRUE(video.get() != NULL);
+
+    ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+    const double psnr = GetAveragePsnr();
+    EXPECT_GT(psnr, GetPsnrThreshold())
+        << "cpu used = " << cpu_used_ << ", encoding mode = " << encoding_mode_;
+  }
+
+  TestVideoParam test_video_param_;
+  int cpu_used_;
+
+ private:
+  double psnr_;
+  unsigned int nframes_;
+  libaom_test::TestMode encoding_mode_;
+};
+
+class EndToEndTestLarge : public EndToEndTest {};
+
+TEST_P(EndToEndTestLarge, EndtoEndPSNRTest) { DoTest(); }
+
+TEST_P(EndToEndTest, EndtoEndPSNRTest) { DoTest(); }
+
+AV1_INSTANTIATE_TEST_CASE(EndToEndTestLarge,
+                          ::testing::ValuesIn(kEncodingModeVectors),
+                          ::testing::ValuesIn(kTestVectors),
+                          ::testing::ValuesIn(kCpuUsedVectors));
+
+AV1_INSTANTIATE_TEST_CASE(EndToEndTest,
+                          ::testing::Values(kEncodingModeVectors[0]),
+                          ::testing::Values(kTestVectors[2]),  // 444
+                          ::testing::Values(kCpuUsedVectors[2]));
+}  // namespace
diff --git a/third_party/aom/test/error_block_test.cc b/third_party/aom/test/error_block_test.cc
new file mode 100644
index 0000000000..353947c3d7
--- /dev/null
+++ b/third_party/aom/test/error_block_test.cc
@@ -0,0 +1,171 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cmath>
+#include <cstdlib>
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/av1_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "av1/common/entropy.h"
+#include "aom/aom_codec.h"
+#include "aom/aom_integer.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+const int kNumIterations = 1000;
+
+typedef int64_t (*ErrorBlockFunc)(const tran_low_t *coeff,
+                                  const tran_low_t *dqcoeff,
+                                  intptr_t block_size, int64_t *ssz, int bps);
+
+typedef ::testing::tuple<ErrorBlockFunc, ErrorBlockFunc, aom_bit_depth_t>
+    ErrorBlockParam;
+
+class ErrorBlockTest : public ::testing::TestWithParam<ErrorBlockParam> {
+ public:
+  virtual ~ErrorBlockTest() {}
+  virtual void SetUp() {
+    error_block_op_ = GET_PARAM(0);
+    ref_error_block_op_ = GET_PARAM(1);
+    bit_depth_ = GET_PARAM(2);
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  aom_bit_depth_t bit_depth_;
+  ErrorBlockFunc error_block_op_;
+  ErrorBlockFunc ref_error_block_op_;
+};
+
+TEST_P(ErrorBlockTest, OperationCheck) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, tran_low_t, coeff[4096]);
+  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[4096]);
+  int err_count_total = 0;
+  int first_failure = -1;
+  intptr_t block_size;
+  int64_t ssz;
+  int64_t ret;
+  int64_t ref_ssz;
+  int64_t ref_ret;
+  const int msb = bit_depth_ + 8 - 1;
+  for (int i = 0; i < kNumIterations; ++i) {
+    int err_count = 0;
+    block_size = 16 << (i % 9);  // All block sizes from 4x4, 8x4 ..64x64
+    for (int j = 0; j < block_size; j++) {
+      // coeff and dqcoeff will always have at least the same sign, and this
+      // can be used for optimization, so generate test input precisely.
+      if (rnd(2)) {
+        // Positive number
+        coeff[j] = rnd(1 << msb);
+        dqcoeff[j] = rnd(1 << msb);
+      } else {
+        // Negative number
+        coeff[j] = -rnd(1 << msb);
+        dqcoeff[j] = -rnd(1 << msb);
+      }
+    }
+    ref_ret =
+        ref_error_block_op_(coeff, dqcoeff, block_size, &ref_ssz, bit_depth_);
+    ASM_REGISTER_STATE_CHECK(
+        ret = error_block_op_(coeff, dqcoeff, block_size, &ssz, bit_depth_));
+    err_count += (ref_ret != ret) | (ref_ssz != ssz);
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Error Block Test, C output doesn't match optimized output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(ErrorBlockTest, ExtremeValues) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, tran_low_t, coeff[4096]);
+  DECLARE_ALIGNED(16, tran_low_t, dqcoeff[4096]);
+  int err_count_total = 0;
+  int first_failure = -1;
+  intptr_t block_size;
+  int64_t ssz;
+  int64_t ret;
+  int64_t ref_ssz;
+  int64_t ref_ret;
+  const int msb = bit_depth_ + 8 - 1;
+  int max_val = ((1 << msb) - 1);
+  for (int i = 0; i < kNumIterations; ++i) {
+    int err_count = 0;
+    int k = (i / 9) % 9;
+
+    // Change the maximum coeff value, to test different bit boundaries
+    if (k == 8 && (i % 9) == 0) {
+      max_val >>= 1;
+    }
+    block_size = 16 << (i % 9);  // All block sizes from 4x4, 8x4 ..64x64
+    for (int j = 0; j < block_size; j++) {
+      if (k < 4) {
+        // Test at positive maximum values
+        coeff[j] = k % 2 ? max_val : 0;
+        dqcoeff[j] = (k >> 1) % 2 ? max_val : 0;
+      } else if (k < 8) {
+        // Test at negative maximum values
+        coeff[j] = k % 2 ? -max_val : 0;
+        dqcoeff[j] = (k >> 1) % 2 ? -max_val : 0;
+      } else {
+        if (rnd(2)) {
+          // Positive number
+          coeff[j] = rnd(1 << 14);
+          dqcoeff[j] = rnd(1 << 14);
+        } else {
+          // Negative number
+          coeff[j] = -rnd(1 << 14);
+          dqcoeff[j] = -rnd(1 << 14);
+        }
+      }
+    }
+    ref_ret =
+        ref_error_block_op_(coeff, dqcoeff, block_size, &ref_ssz, bit_depth_);
+    ASM_REGISTER_STATE_CHECK(
+        ret = error_block_op_(coeff, dqcoeff, block_size, &ssz, bit_depth_));
+    err_count += (ref_ret != ret) | (ref_ssz != ssz);
+    if (err_count && !err_count_total) {
+      first_failure = i;
+    }
+    err_count_total += err_count;
+  }
+  EXPECT_EQ(0, err_count_total)
+      << "Error: Error Block Test, C output doesn't match optimized output. "
+      << "First failed at test case " << first_failure;
+}
+
+#if (HAVE_SSE2 || HAVE_AVX)
+using ::testing::make_tuple;
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2, ErrorBlockTest,
+    ::testing::Values(make_tuple(&av1_highbd_block_error_sse2,
+                                 &av1_highbd_block_error_c, AOM_BITS_10),
+                      make_tuple(&av1_highbd_block_error_sse2,
+                                 &av1_highbd_block_error_c, AOM_BITS_12),
+                      make_tuple(&av1_highbd_block_error_sse2,
+                                 &av1_highbd_block_error_c, AOM_BITS_8)));
+#endif  // HAVE_SSE2
+}  // namespace
diff --git a/third_party/aom/test/error_resilience_test.cc b/third_party/aom/test/error_resilience_test.cc
new file mode 100644
index 0000000000..13ac0bf93d
--- /dev/null
+++ b/third_party/aom/test/error_resilience_test.cc
@@ -0,0 +1,438 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+const int kMaxErrorFrames = 12;
+const int kMaxInvisibleErrorFrames = 12;
+const int kMaxDroppableFrames = 12;
+const int kMaxErrorResilientFrames = 12;
+const int kMaxNoMFMVFrames = 12;
+const int kMaxPrimRefNoneFrames = 12;
+const int kMaxSFrames = 12;
+const int kCpuUsed = 1;
+
+class ErrorResilienceTestLarge
+    : public ::libaom_test::CodecTestWithParam<libaom_test::TestMode>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  ErrorResilienceTestLarge()
+      : EncoderTest(GET_PARAM(0)), psnr_(0.0), nframes_(0), mismatch_psnr_(0.0),
+        mismatch_nframes_(0), encoding_mode_(GET_PARAM(1)), allow_mismatch_(0) {
+    Reset();
+  }
+
+  virtual ~ErrorResilienceTestLarge() {}
+
+  void Reset() {
+    error_nframes_ = 0;
+    invisible_error_nframes_ = 0;
+    droppable_nframes_ = 0;
+    error_resilient_nframes_ = 0;
+    nomfmv_nframes_ = 0;
+    prim_ref_none_nframes_ = 0;
+    s_nframes_ = 0;
+  }
+
+  void SetupEncoder(int bitrate, int lag) {
+    const aom_rational timebase = { 33333333, 1000000000 };
+    cfg_.g_timebase = timebase;
+    cfg_.rc_target_bitrate = bitrate;
+    cfg_.kf_mode = AOM_KF_DISABLED;
+    cfg_.g_lag_in_frames = lag;
+    init_flags_ = AOM_CODEC_USE_PSNR;
+  }
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+  }
+
+  virtual void BeginPassHook(unsigned int /*pass*/) {
+    psnr_ = 0.0;
+    nframes_ = 0;
+    decoded_nframes_ = 0;
+    mismatch_psnr_ = 0.0;
+    mismatch_nframes_ = 0;
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    psnr_ += pkt->data.psnr.psnr[0];
+    nframes_++;
+  }
+
+  virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
+                                  libaom_test::Encoder *encoder) {
+    if (video->frame() == 0) encoder->Control(AOME_SET_CPUUSED, kCpuUsed);
+    frame_flags_ &=
+        ~(AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF |
+          AOM_EFLAG_NO_REF_FRAME_MVS | AOM_EFLAG_ERROR_RESILIENT |
+          AOM_EFLAG_SET_S_FRAME | AOM_EFLAG_SET_PRIMARY_REF_NONE);
+    if (droppable_nframes_ > 0 &&
+        (cfg_.g_pass == AOM_RC_LAST_PASS || cfg_.g_pass == AOM_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < droppable_nframes_; ++i) {
+        if (droppable_frames_[i] == video->frame()) {
+          std::cout << "             Encoding droppable frame: "
+                    << droppable_frames_[i] << "\n";
+          frame_flags_ |= (AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF |
+                           AOM_EFLAG_NO_UPD_ARF);
+          break;
+        }
+      }
+    }
+
+    if (error_resilient_nframes_ > 0 &&
+        (cfg_.g_pass == AOM_RC_LAST_PASS || cfg_.g_pass == AOM_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < error_resilient_nframes_; ++i) {
+        if (error_resilient_frames_[i] == video->frame()) {
+          std::cout << "             Encoding error_resilient frame: "
+                    << error_resilient_frames_[i] << "\n";
+          frame_flags_ |= AOM_EFLAG_ERROR_RESILIENT;
+          break;
+        }
+      }
+    }
+
+    if (nomfmv_nframes_ > 0 &&
+        (cfg_.g_pass == AOM_RC_LAST_PASS || cfg_.g_pass == AOM_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < nomfmv_nframes_; ++i) {
+        if (nomfmv_frames_[i] == video->frame()) {
+          std::cout << "             Encoding no mfmv frame: "
+                    << nomfmv_frames_[i] << "\n";
+          frame_flags_ |= AOM_EFLAG_NO_REF_FRAME_MVS;
+          break;
+        }
+      }
+    }
+
+    if (prim_ref_none_nframes_ > 0 &&
+        (cfg_.g_pass == AOM_RC_LAST_PASS || cfg_.g_pass == AOM_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < prim_ref_none_nframes_; ++i) {
+        if (prim_ref_none_frames_[i] == video->frame()) {
+          std::cout << "             Encoding no PRIMARY_REF_NONE frame: "
+                    << prim_ref_none_frames_[i] << "\n";
+          frame_flags_ |= AOM_EFLAG_SET_PRIMARY_REF_NONE;
+          break;
+        }
+      }
+    }
+
+    encoder->Control(AV1E_SET_S_FRAME_MODE, 0);
+    if (s_nframes_ > 0 &&
+        (cfg_.g_pass == AOM_RC_LAST_PASS || cfg_.g_pass == AOM_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < s_nframes_; ++i) {
+        if (s_frames_[i] == video->frame()) {
+          std::cout << "             Encoding S frame: " << s_frames_[i]
+                    << "\n";
+          frame_flags_ |= AOM_EFLAG_SET_S_FRAME;
+          break;
+        }
+      }
+    }
+  }
+
+  double GetAveragePsnr() const {
+    if (nframes_) return psnr_ / nframes_;
+    return 0.0;
+  }
+
+  double GetAverageMismatchPsnr() const {
+    if (mismatch_nframes_) return mismatch_psnr_ / mismatch_nframes_;
+    return 0.0;
+  }
+
+  virtual bool DoDecode() const {
+    if (error_nframes_ > 0 &&
+        (cfg_.g_pass == AOM_RC_LAST_PASS || cfg_.g_pass == AOM_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < error_nframes_; ++i) {
+        if (error_frames_[i] == nframes_ - 1) {
+          std::cout << "             Skipping decoding frame: "
+                    << error_frames_[i] << "\n";
+          return 0;
+        }
+      }
+    }
+    return 1;
+  }
+
+  virtual bool DoDecodeInvisible() const {
+    if (invisible_error_nframes_ > 0 &&
+        (cfg_.g_pass == AOM_RC_LAST_PASS || cfg_.g_pass == AOM_RC_ONE_PASS)) {
+      for (unsigned int i = 0; i < invisible_error_nframes_; ++i) {
+        if (invisible_error_frames_[i] == nframes_ - 1) {
+          std::cout << "             Skipping decoding all invisible frames in "
+                       "frame pkt: "
+                    << invisible_error_frames_[i] << "\n";
+          return 0;
+        }
+      }
+    }
+    return 1;
+  }
+
+  virtual void MismatchHook(const aom_image_t *img1, const aom_image_t *img2) {
+    if (allow_mismatch_) {
+      double mismatch_psnr = compute_psnr(img1, img2);
+      mismatch_psnr_ += mismatch_psnr;
+      ++mismatch_nframes_;
+      // std::cout << "Mismatch frame psnr: " << mismatch_psnr << "\n";
+    } else {
+      ::libaom_test::EncoderTest::MismatchHook(img1, img2);
+    }
+  }
+
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     aom_codec_pts_t pts) {
+    (void)img;
+    (void)pts;
+    ++decoded_nframes_;
+  }
+
+  void SetErrorFrames(int num, unsigned int *list) {
+    if (num > kMaxErrorFrames)
+      num = kMaxErrorFrames;
+    else if (num < 0)
+      num = 0;
+    error_nframes_ = num;
+    for (unsigned int i = 0; i < error_nframes_; ++i)
+      error_frames_[i] = list[i];
+  }
+
+  void SetInvisibleErrorFrames(int num, unsigned int *list) {
+    if (num > kMaxInvisibleErrorFrames)
+      num = kMaxInvisibleErrorFrames;
+    else if (num < 0)
+      num = 0;
+    invisible_error_nframes_ = num;
+    for (unsigned int i = 0; i < invisible_error_nframes_; ++i)
+      invisible_error_frames_[i] = list[i];
+  }
+
+  void SetDroppableFrames(int num, unsigned int *list) {
+    if (num > kMaxDroppableFrames)
+      num = kMaxDroppableFrames;
+    else if (num < 0)
+      num = 0;
+    droppable_nframes_ = num;
+    for (unsigned int i = 0; i < droppable_nframes_; ++i)
+      droppable_frames_[i] = list[i];
+  }
+
+  void SetErrorResilientFrames(int num, unsigned int *list) {
+    if (num > kMaxErrorResilientFrames)
+      num = kMaxErrorResilientFrames;
+    else if (num < 0)
+      num = 0;
+    error_resilient_nframes_ = num;
+    for (unsigned int i = 0; i < error_resilient_nframes_; ++i)
+      error_resilient_frames_[i] = list[i];
+  }
+
+  void SetNoMFMVFrames(int num, unsigned int *list) {
+    if (num > kMaxNoMFMVFrames)
+      num = kMaxNoMFMVFrames;
+    else if (num < 0)
+      num = 0;
+    nomfmv_nframes_ = num;
+    for (unsigned int i = 0; i < nomfmv_nframes_; ++i)
+      nomfmv_frames_[i] = list[i];
+  }
+
+  void SetPrimaryRefNoneFrames(int num, unsigned int *list) {
+    if (num > kMaxPrimRefNoneFrames)
+      num = kMaxPrimRefNoneFrames;
+    else if (num < 0)
+      num = 0;
+    prim_ref_none_nframes_ = num;
+    for (unsigned int i = 0; i < prim_ref_none_nframes_; ++i)
+      prim_ref_none_frames_[i] = list[i];
+  }
+
+  void SetSFrames(int num, unsigned int *list) {
+    if (num > kMaxSFrames)
+      num = kMaxSFrames;
+    else if (num < 0)
+      num = 0;
+    s_nframes_ = num;
+    for (unsigned int i = 0; i < s_nframes_; ++i) s_frames_[i] = list[i];
+  }
+
+  unsigned int GetMismatchFrames() { return mismatch_nframes_; }
+  unsigned int GetEncodedFrames() { return nframes_; }
+  unsigned int GetDecodedFrames() { return decoded_nframes_; }
+
+  void SetAllowMismatch(int allow) { allow_mismatch_ = allow; }
+
+ private:
+  double psnr_;
+  unsigned int nframes_;
+  unsigned int decoded_nframes_;
+  unsigned int error_nframes_;
+  unsigned int invisible_error_nframes_;
+  unsigned int droppable_nframes_;
+  unsigned int error_resilient_nframes_;
+  unsigned int nomfmv_nframes_;
+  unsigned int prim_ref_none_nframes_;
+  unsigned int s_nframes_;
+  double mismatch_psnr_;
+  unsigned int mismatch_nframes_;
+  unsigned int error_frames_[kMaxErrorFrames];
+  unsigned int invisible_error_frames_[kMaxInvisibleErrorFrames];
+  unsigned int droppable_frames_[kMaxDroppableFrames];
+  unsigned int error_resilient_frames_[kMaxErrorResilientFrames];
+  unsigned int nomfmv_frames_[kMaxNoMFMVFrames];
+  unsigned int prim_ref_none_frames_[kMaxPrimRefNoneFrames];
+  unsigned int s_frames_[kMaxSFrames];
+  libaom_test::TestMode encoding_mode_;
+  int allow_mismatch_;
+};
+
+TEST_P(ErrorResilienceTestLarge, OnVersusOff) {
+  SetupEncoder(2000, 10);
+  libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                     cfg_.g_timebase.den, cfg_.g_timebase.num,
+                                     0, 12);
+
+  // Global error resilient mode OFF.
+  cfg_.g_error_resilient = 0;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  const double psnr_resilience_off = GetAveragePsnr();
+  EXPECT_GT(psnr_resilience_off, 25.0);
+
+  Reset();
+  // Error resilient mode ON for certain frames
+  unsigned int num_error_resilient_frames = 5;
+  unsigned int error_resilient_frame_list[] = { 3, 5, 6, 9, 11 };
+  SetErrorResilientFrames(num_error_resilient_frames,
+                          error_resilient_frame_list);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  const double psnr_resilience_on = GetAveragePsnr();
+  EXPECT_GT(psnr_resilience_on, 25.0);
+
+  // Test that turning on error resilient mode hurts by 10% at most.
+  if (psnr_resilience_off > 0.0) {
+    const double psnr_ratio = psnr_resilience_on / psnr_resilience_off;
+    EXPECT_GE(psnr_ratio, 0.9);
+    EXPECT_LE(psnr_ratio, 1.1);
+  }
+}
+
+// Check for successful decoding and no encoder/decoder mismatch
+// if we lose (i.e., drop before decoding) a set of droppable
+// frames (i.e., frames that don't update any reference buffers).
+TEST_P(ErrorResilienceTestLarge, DropFramesWithoutRecovery) {
+  SetupEncoder(500, 10);
+  libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                     cfg_.g_timebase.den, cfg_.g_timebase.num,
+                                     0, 20);
+
+  // Set an arbitrary set of error frames same as droppable frames.
+  unsigned int num_droppable_frames = 3;
+  unsigned int droppable_frame_list[] = { 5, 10, 13 };
+  SetDroppableFrames(num_droppable_frames, droppable_frame_list);
+  SetErrorFrames(num_droppable_frames, droppable_frame_list);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  // Test that no mismatches have been found
+  std::cout << "             Encoded frames: " << GetEncodedFrames() << "\n";
+  std::cout << "             Decoded frames: " << GetDecodedFrames() << "\n";
+  std::cout << "             Mismatch frames: " << GetMismatchFrames() << "\n";
+  EXPECT_EQ(GetEncodedFrames() - GetDecodedFrames(), num_droppable_frames);
+}
+
+// Check for ParseAbility property of an error-resilient frame.
+// Encode a frame in error-resilient mode (E-frame), and disallow all
+// subsequent frames from using MFMV. If frames are dropped before the
+// E frame, all frames starting from the E frame should be parse-able.
+TEST_P(ErrorResilienceTestLarge, ParseAbilityTest) {
+  SetupEncoder(500, 10);
+
+  libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                     cfg_.g_timebase.den, cfg_.g_timebase.num,
+                                     0, 15);
+
+  SetAllowMismatch(1);
+
+  // Note that an E-frame cannot be forced on a frame that is a
+  // show_existing_frame, or a frame that comes directly after an invisible
+  // frame. Currently, this will cause an assertion failure.
+  // Set an arbitrary error resilient (E) frame
+  unsigned int num_error_resilient_frames = 1;
+  unsigned int error_resilient_frame_list[] = { 8 };
+  SetErrorResilientFrames(num_error_resilient_frames,
+                          error_resilient_frame_list);
+  // Ensure that any invisible frames before the E frame are dropped
+  SetInvisibleErrorFrames(num_error_resilient_frames,
+                          error_resilient_frame_list);
+  // Set all frames after the error resilient frame to not allow MFMV
+  unsigned int num_post_error_resilient_frames = 6;
+  unsigned int post_error_resilient_frame_list[] = { 9, 10, 11, 12, 13, 14 };
+  SetNoMFMVFrames(num_post_error_resilient_frames,
+                  post_error_resilient_frame_list);
+
+  // Set a few frames before the E frame that are lost (not decoded)
+  unsigned int num_error_frames = 5;
+  unsigned int error_frame_list[] = { 3, 4, 5, 6, 7 };
+  SetErrorFrames(num_error_frames, error_frame_list);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  std::cout << "             Encoded frames: " << GetEncodedFrames() << "\n";
+  std::cout << "             Decoded frames: " << GetDecodedFrames() << "\n";
+  std::cout << "             Mismatch frames: " << GetMismatchFrames() << "\n";
+  EXPECT_EQ(GetEncodedFrames() - GetDecodedFrames(), num_error_frames);
+  // All frames following the E-frame and the E-frame are expected to have
+  // mismatches, but still be parse-able.
+  EXPECT_LE(GetMismatchFrames(), num_post_error_resilient_frames + 1);
+}
+
+// Check for ParseAbility property of an S frame.
+// Encode an S-frame. If frames are dropped before the S-frame, all frames
+// starting from the S frame should be parse-able.
+TEST_P(ErrorResilienceTestLarge, SFrameTest) {
+  SetupEncoder(500, 10);
+
+  libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                     cfg_.g_timebase.den, cfg_.g_timebase.num,
+                                     0, 15);
+
+  SetAllowMismatch(1);
+
+  // Note that an S-frame cannot be forced on a frame that is a
+  // show_existing_frame. This issue still needs to be addressed.
+  // Set an arbitrary S-frame
+  unsigned int num_s_frames = 1;
+  unsigned int s_frame_list[] = { 6 };
+  SetSFrames(num_s_frames, s_frame_list);
+  // Ensure that any invisible frames before the S frame are dropped
+  SetInvisibleErrorFrames(num_s_frames, s_frame_list);
+
+  // Set a few frames before the S frame that are lost (not decoded)
+  unsigned int num_error_frames = 4;
+  unsigned int error_frame_list[] = { 2, 3, 4, 5 };
+  SetErrorFrames(num_error_frames, error_frame_list);
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  std::cout << "             Encoded frames: " << GetEncodedFrames() << "\n";
+  std::cout << "             Decoded frames: " << GetDecodedFrames() << "\n";
+  std::cout << "             Mismatch frames: " << GetMismatchFrames() << "\n";
+  EXPECT_EQ(GetEncodedFrames() - GetDecodedFrames(), num_error_frames);
+  // All frames following the S-frame and the S-frame are expected to have
+  // mismatches, but still be parse-able.
+  EXPECT_LE(GetMismatchFrames(), GetEncodedFrames() - s_frame_list[0]);
+}
+
+AV1_INSTANTIATE_TEST_CASE(ErrorResilienceTestLarge, NONREALTIME_TEST_MODES);
+}  // namespace
diff --git a/third_party/aom/test/ethread_test.cc b/third_party/aom/test/ethread_test.cc
new file mode 100644
index 0000000000..d9ac782828
--- /dev/null
+++ b/third_party/aom/test/ethread_test.cc
@@ -0,0 +1,273 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string>
+#include <vector>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/md5_helper.h"
+#include "test/util.h"
+#include "test/yuv_video_source.h"
+
+namespace {
+class AVxEncoderThreadTest
+    : public ::libaom_test::CodecTestWith4Params<libaom_test::TestMode, int,
+                                                 int, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  AVxEncoderThreadTest()
+      : EncoderTest(GET_PARAM(0)), encoder_initialized_(false),
+        encoding_mode_(GET_PARAM(1)), set_cpu_used_(GET_PARAM(2)),
+        tile_cols_(GET_PARAM(3)), tile_rows_(GET_PARAM(4)) {
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+    cfg.w = 1280;
+    cfg.h = 720;
+    cfg.allow_lowbitdepth = 1;
+    decoder_ = codec_->CreateDecoder(cfg, 0);
+    if (decoder_->IsAV1()) {
+      decoder_->Control(AV1_SET_DECODE_TILE_ROW, -1);
+      decoder_->Control(AV1_SET_DECODE_TILE_COL, -1);
+    }
+
+    size_enc_.clear();
+    md5_dec_.clear();
+    md5_enc_.clear();
+  }
+  virtual ~AVxEncoderThreadTest() { delete decoder_; }
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+
+    if (encoding_mode_ != ::libaom_test::kRealTime) {
+      cfg_.g_lag_in_frames = 5;
+      cfg_.rc_end_usage = AOM_VBR;
+      cfg_.rc_2pass_vbr_minsection_pct = 5;
+      cfg_.rc_2pass_vbr_maxsection_pct = 2000;
+    } else {
+      cfg_.g_lag_in_frames = 0;
+      cfg_.rc_end_usage = AOM_CBR;
+      cfg_.g_error_resilient = 1;
+    }
+    cfg_.rc_max_quantizer = 56;
+    cfg_.rc_min_quantizer = 0;
+  }
+
+  virtual void BeginPassHook(unsigned int /*pass*/) {
+    encoder_initialized_ = false;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource * /*video*/,
+                                  ::libaom_test::Encoder *encoder) {
+    if (!encoder_initialized_) {
+      SetTileSize(encoder);
+      encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
+      encoder->Control(AV1E_SET_ROW_MT, row_mt_);
+      if (encoding_mode_ != ::libaom_test::kRealTime) {
+        encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+        encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+        encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+        encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 0);
+      } else {
+        encoder->Control(AOME_SET_ENABLEAUTOALTREF, 0);
+        encoder->Control(AV1E_SET_AQ_MODE, 3);
+      }
+      encoder_initialized_ = true;
+    }
+  }
+
+  virtual void SetTileSize(libaom_test::Encoder *encoder) {
+    encoder->Control(AV1E_SET_TILE_COLUMNS, tile_cols_);
+    encoder->Control(AV1E_SET_TILE_ROWS, tile_rows_);
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    size_enc_.push_back(pkt->data.frame.sz);
+
+    ::libaom_test::MD5 md5_enc;
+    md5_enc.Add(reinterpret_cast<uint8_t *>(pkt->data.frame.buf),
+                pkt->data.frame.sz);
+    md5_enc_.push_back(md5_enc.Get());
+
+    const aom_codec_err_t res = decoder_->DecodeFrame(
+        reinterpret_cast<uint8_t *>(pkt->data.frame.buf), pkt->data.frame.sz);
+    if (res != AOM_CODEC_OK) {
+      abort_ = true;
+      ASSERT_EQ(AOM_CODEC_OK, res);
+    }
+    const aom_image_t *img = decoder_->GetDxData().Next();
+
+    if (img) {
+      ::libaom_test::MD5 md5_res;
+      md5_res.Add(img);
+      md5_dec_.push_back(md5_res.Get());
+    }
+  }
+
+  void DoTest() {
+    ::libaom_test::YUVVideoSource video(
+        "niklas_640_480_30.yuv", AOM_IMG_FMT_I420, 640, 480, 30, 1, 15, 21);
+    cfg_.rc_target_bitrate = 1000;
+
+    // Encode using single thread.
+    row_mt_ = 0;
+    cfg_.g_threads = 1;
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    std::vector<size_t> single_thr_size_enc;
+    std::vector<std::string> single_thr_md5_enc;
+    std::vector<std::string> single_thr_md5_dec;
+    single_thr_size_enc = size_enc_;
+    single_thr_md5_enc = md5_enc_;
+    single_thr_md5_dec = md5_dec_;
+    size_enc_.clear();
+    md5_enc_.clear();
+    md5_dec_.clear();
+
+    // Encode using multiple threads.
+    cfg_.g_threads = 4;
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    std::vector<size_t> multi_thr_size_enc;
+    std::vector<std::string> multi_thr_md5_enc;
+    std::vector<std::string> multi_thr_md5_dec;
+    multi_thr_size_enc = size_enc_;
+    multi_thr_md5_enc = md5_enc_;
+    multi_thr_md5_dec = md5_dec_;
+    size_enc_.clear();
+    md5_enc_.clear();
+    md5_dec_.clear();
+
+    // Check that the vectors are equal.
+    ASSERT_EQ(single_thr_size_enc, multi_thr_size_enc);
+    ASSERT_EQ(single_thr_md5_enc, multi_thr_md5_enc);
+    ASSERT_EQ(single_thr_md5_dec, multi_thr_md5_dec);
+
+    // Encode using multiple threads row-mt enabled.
+    row_mt_ = 1;
+    cfg_.g_threads = 2;
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    std::vector<size_t> multi_thr2_row_mt_size_enc;
+    std::vector<std::string> multi_thr2_row_mt_md5_enc;
+    std::vector<std::string> multi_thr2_row_mt_md5_dec;
+    multi_thr2_row_mt_size_enc = size_enc_;
+    multi_thr2_row_mt_md5_enc = md5_enc_;
+    multi_thr2_row_mt_md5_dec = md5_dec_;
+    size_enc_.clear();
+    md5_enc_.clear();
+    md5_dec_.clear();
+
+    // Disable threads=3 test for now to reduce the time so that the nightly
+    // test would not time out.
+    // cfg_.g_threads = 3;
+    // ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    // std::vector<size_t> multi_thr3_row_mt_size_enc;
+    // std::vector<std::string> multi_thr3_row_mt_md5_enc;
+    // std::vector<std::string> multi_thr3_row_mt_md5_dec;
+    // multi_thr3_row_mt_size_enc = size_enc_;
+    // multi_thr3_row_mt_md5_enc = md5_enc_;
+    // multi_thr3_row_mt_md5_dec = md5_dec_;
+    // size_enc_.clear();
+    // md5_enc_.clear();
+    // md5_dec_.clear();
+    // Check that the vectors are equal.
+    // ASSERT_EQ(multi_thr3_row_mt_size_enc, multi_thr2_row_mt_size_enc);
+    // ASSERT_EQ(multi_thr3_row_mt_md5_enc, multi_thr2_row_mt_md5_enc);
+    // ASSERT_EQ(multi_thr3_row_mt_md5_dec, multi_thr2_row_mt_md5_dec);
+
+    cfg_.g_threads = 4;
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+    std::vector<size_t> multi_thr4_row_mt_size_enc;
+    std::vector<std::string> multi_thr4_row_mt_md5_enc;
+    std::vector<std::string> multi_thr4_row_mt_md5_dec;
+    multi_thr4_row_mt_size_enc = size_enc_;
+    multi_thr4_row_mt_md5_enc = md5_enc_;
+    multi_thr4_row_mt_md5_dec = md5_dec_;
+    size_enc_.clear();
+    md5_enc_.clear();
+    md5_dec_.clear();
+
+    // Check that the vectors are equal.
+    ASSERT_EQ(multi_thr4_row_mt_size_enc, multi_thr2_row_mt_size_enc);
+    ASSERT_EQ(multi_thr4_row_mt_md5_enc, multi_thr2_row_mt_md5_enc);
+    ASSERT_EQ(multi_thr4_row_mt_md5_dec, multi_thr2_row_mt_md5_dec);
+  }
+
+  bool encoder_initialized_;
+  ::libaom_test::TestMode encoding_mode_;
+  int set_cpu_used_;
+  int tile_cols_;
+  int tile_rows_;
+  int row_mt_;
+  ::libaom_test::Decoder *decoder_;
+  std::vector<size_t> size_enc_;
+  std::vector<std::string> md5_enc_;
+  std::vector<std::string> md5_dec_;
+};
+
+TEST_P(AVxEncoderThreadTest, EncoderResultTest) {
+  cfg_.large_scale_tile = 0;
+  decoder_->Control(AV1_SET_TILE_MODE, 0);
+  DoTest();
+}
+
+class AVxEncoderThreadTestLarge : public AVxEncoderThreadTest {};
+
+TEST_P(AVxEncoderThreadTestLarge, EncoderResultTest) {
+  cfg_.large_scale_tile = 0;
+  decoder_->Control(AV1_SET_TILE_MODE, 0);
+  DoTest();
+}
+
+// For AV1, only test speed 0 to 3.
+// Here test cpu_used 2 and 3
+AV1_INSTANTIATE_TEST_CASE(AVxEncoderThreadTest,
+                          ::testing::Values(::libaom_test::kTwoPassGood),
+                          ::testing::Range(2, 4), ::testing::Values(0, 2),
+                          ::testing::Values(0, 1));
+
+// Test cpu_used 0 and 1.
+AV1_INSTANTIATE_TEST_CASE(AVxEncoderThreadTestLarge,
+                          ::testing::Values(::libaom_test::kTwoPassGood,
+                                            ::libaom_test::kOnePassGood),
+                          ::testing::Range(0, 2), ::testing::Values(0, 1, 2, 6),
+                          ::testing::Values(0, 1, 2, 6));
+
+class AVxEncoderThreadLSTest : public AVxEncoderThreadTest {
+  virtual void SetTileSize(libaom_test::Encoder *encoder) {
+    encoder->Control(AV1E_SET_TILE_COLUMNS, tile_cols_);
+    encoder->Control(AV1E_SET_TILE_ROWS, tile_rows_);
+  }
+};
+
+TEST_P(AVxEncoderThreadLSTest, EncoderResultTest) {
+  cfg_.large_scale_tile = 1;
+  decoder_->Control(AV1_SET_TILE_MODE, 1);
+  decoder_->Control(AV1D_EXT_TILE_DEBUG, 1);
+  DoTest();
+}
+
+class AVxEncoderThreadLSTestLarge : public AVxEncoderThreadLSTest {};
+
+TEST_P(AVxEncoderThreadLSTestLarge, EncoderResultTest) {
+  cfg_.large_scale_tile = 1;
+  decoder_->Control(AV1_SET_TILE_MODE, 1);
+  decoder_->Control(AV1D_EXT_TILE_DEBUG, 1);
+  DoTest();
+}
+
+AV1_INSTANTIATE_TEST_CASE(AVxEncoderThreadLSTestLarge,
+                          ::testing::Values(::libaom_test::kTwoPassGood,
+                                            ::libaom_test::kOnePassGood),
+                          ::testing::Range(0, 4), ::testing::Values(0, 6),
+                          ::testing::Values(0, 6));
+}  // namespace
diff --git a/third_party/aom/test/examples.sh b/third_party/aom/test/examples.sh
new file mode 100755
index 0000000000..2cdb89dd08
--- /dev/null
+++ b/third_party/aom/test/examples.sh
@@ -0,0 +1,29 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file runs all of the tests for the libaom examples.
+##
+. $(dirname $0)/tools_common.sh
+
+example_tests=$(ls -r $(dirname $0)/*.sh)
+
+# List of script names to exclude.
+exclude_list="best_encode examples run_encodes tools_common"
+
+# Filter out the scripts in $exclude_list.
+for word in ${exclude_list}; do
+  example_tests=$(filter_strings "${example_tests}" "${word}" exclude)
+done
+
+for test in ${example_tests}; do
+  # Source each test script so that exporting variables can be avoided.
+  AOM_TEST_NAME="$(basename ${test%.*})"
+  . "${test}"
+done
diff --git a/third_party/aom/test/external_frame_buffer_test.cc b/third_party/aom/test/external_frame_buffer_test.cc
new file mode 100644
index 0000000000..c2af059a4e
--- /dev/null
+++ b/third_party/aom/test/external_frame_buffer_test.cc
@@ -0,0 +1,512 @@
+/*
+ *  Copyright (c) 2014 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <string>
+
+#include "config/aom_config.h"
+#include "test/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/ivf_video_source.h"
+#include "test/md5_helper.h"
+#include "test/test_vectors.h"
+#include "test/util.h"
+#if CONFIG_WEBM_IO
+#include "test/webm_video_source.h"
+#endif
+
+namespace {
+
+const int kVideoNameParam = 1;
+
+struct ExternalFrameBuffer {
+  uint8_t *data;
+  size_t size;
+  int in_use;
+};
+
+// Class to manipulate a list of external frame buffers.
+class ExternalFrameBufferList {
+ public:
+  ExternalFrameBufferList()
+      : num_buffers_(0), num_used_buffers_(0), ext_fb_list_(NULL) {}
+
+  virtual ~ExternalFrameBufferList() {
+    for (int i = 0; i < num_buffers_; ++i) {
+      delete[] ext_fb_list_[i].data;
+    }
+    delete[] ext_fb_list_;
+  }
+
+  // Creates the list to hold the external buffers. Returns true on success.
+  bool CreateBufferList(int num_buffers) {
+    if (num_buffers < 0) return false;
+
+    num_buffers_ = num_buffers;
+    ext_fb_list_ = new ExternalFrameBuffer[num_buffers_];
+    EXPECT_TRUE(ext_fb_list_ != NULL);
+    memset(ext_fb_list_, 0, sizeof(ext_fb_list_[0]) * num_buffers_);
+    return true;
+  }
+
+  // Searches the frame buffer list for a free frame buffer. Makes sure
+  // that the frame buffer is at least |min_size| in bytes. Marks that the
+  // frame buffer is in use by libvpx. Finally sets |fb| to point to the
+  // external frame buffer. Returns < 0 on an error.
+  int GetFreeFrameBuffer(size_t min_size, aom_codec_frame_buffer_t *fb) {
+    EXPECT_TRUE(fb != NULL);
+    const int idx = FindFreeBufferIndex();
+    if (idx == num_buffers_) return -1;
+
+    if (ext_fb_list_[idx].size < min_size) {
+      delete[] ext_fb_list_[idx].data;
+      ext_fb_list_[idx].data = new uint8_t[min_size];
+      memset(ext_fb_list_[idx].data, 0, min_size);
+      ext_fb_list_[idx].size = min_size;
+    }
+
+    SetFrameBuffer(idx, fb);
+
+    num_used_buffers_++;
+    return 0;
+  }
+
+  // Test function that will not allocate any data for the frame buffer.
+  // Returns < 0 on an error.
+  int GetZeroFrameBuffer(size_t min_size, aom_codec_frame_buffer_t *fb) {
+    EXPECT_TRUE(fb != NULL);
+    const int idx = FindFreeBufferIndex();
+    if (idx == num_buffers_) return -1;
+
+    if (ext_fb_list_[idx].size < min_size) {
+      delete[] ext_fb_list_[idx].data;
+      ext_fb_list_[idx].data = NULL;
+      ext_fb_list_[idx].size = min_size;
+    }
+
+    SetFrameBuffer(idx, fb);
+    return 0;
+  }
+
+  // Marks the external frame buffer that |fb| is pointing to as free.
+  // Returns < 0 on an error.
+  int ReturnFrameBuffer(aom_codec_frame_buffer_t *fb) {
+    if (fb == NULL) {
+      EXPECT_TRUE(fb != NULL);
+      return -1;
+    }
+    ExternalFrameBuffer *const ext_fb =
+        reinterpret_cast<ExternalFrameBuffer *>(fb->priv);
+    if (ext_fb == NULL) {
+      EXPECT_TRUE(ext_fb != NULL);
+      return -1;
+    }
+    EXPECT_EQ(1, ext_fb->in_use);
+    ext_fb->in_use = 0;
+    num_used_buffers_--;
+    return 0;
+  }
+
+  // Checks that the ximage data is contained within the external frame buffer
+  // private data passed back in the ximage.
+  void CheckXImageFrameBuffer(const aom_image_t *img) {
+    if (img->fb_priv != NULL) {
+      const struct ExternalFrameBuffer *const ext_fb =
+          reinterpret_cast<ExternalFrameBuffer *>(img->fb_priv);
+
+      ASSERT_TRUE(img->planes[0] >= ext_fb->data &&
+                  img->planes[0] < (ext_fb->data + ext_fb->size));
+    }
+  }
+
+  int num_used_buffers() const { return num_used_buffers_; }
+
+ private:
+  // Returns the index of the first free frame buffer. Returns |num_buffers_|
+  // if there are no free frame buffers.
+  int FindFreeBufferIndex() {
+    int i;
+    // Find a free frame buffer.
+    for (i = 0; i < num_buffers_; ++i) {
+      if (!ext_fb_list_[i].in_use) break;
+    }
+    return i;
+  }
+
+  // Sets |fb| to an external frame buffer. idx is the index into the frame
+  // buffer list.
+  void SetFrameBuffer(int idx, aom_codec_frame_buffer_t *fb) {
+    ASSERT_TRUE(fb != NULL);
+    fb->data = ext_fb_list_[idx].data;
+    fb->size = ext_fb_list_[idx].size;
+    ASSERT_EQ(0, ext_fb_list_[idx].in_use);
+    ext_fb_list_[idx].in_use = 1;
+    fb->priv = &ext_fb_list_[idx];
+  }
+
+  int num_buffers_;
+  int num_used_buffers_;
+  ExternalFrameBuffer *ext_fb_list_;
+};
+
+#if CONFIG_WEBM_IO
+
+// Callback used by libvpx to request the application to return a frame
+// buffer of at least |min_size| in bytes.
+int get_aom_frame_buffer(void *user_priv, size_t min_size,
+                         aom_codec_frame_buffer_t *fb) {
+  ExternalFrameBufferList *const fb_list =
+      reinterpret_cast<ExternalFrameBufferList *>(user_priv);
+  return fb_list->GetFreeFrameBuffer(min_size, fb);
+}
+
+// Callback used by libvpx to tell the application that |fb| is not needed
+// anymore.
+int release_aom_frame_buffer(void *user_priv, aom_codec_frame_buffer_t *fb) {
+  ExternalFrameBufferList *const fb_list =
+      reinterpret_cast<ExternalFrameBufferList *>(user_priv);
+  return fb_list->ReturnFrameBuffer(fb);
+}
+
+// Callback will not allocate data for frame buffer.
+int get_aom_zero_frame_buffer(void *user_priv, size_t min_size,
+                              aom_codec_frame_buffer_t *fb) {
+  ExternalFrameBufferList *const fb_list =
+      reinterpret_cast<ExternalFrameBufferList *>(user_priv);
+  return fb_list->GetZeroFrameBuffer(min_size, fb);
+}
+
+// Callback will allocate one less byte than |min_size|.
+int get_aom_one_less_byte_frame_buffer(void *user_priv, size_t min_size,
+                                       aom_codec_frame_buffer_t *fb) {
+  ExternalFrameBufferList *const fb_list =
+      reinterpret_cast<ExternalFrameBufferList *>(user_priv);
+  return fb_list->GetFreeFrameBuffer(min_size - 1, fb);
+}
+
+// Callback will not release the external frame buffer.
+int do_not_release_aom_frame_buffer(void *user_priv,
+                                    aom_codec_frame_buffer_t *fb) {
+  (void)user_priv;
+  (void)fb;
+  return 0;
+}
+
+#endif  // CONFIG_WEBM_IO
+
+// Class for testing passing in external frame buffers to libaom.
+class ExternalFrameBufferMD5Test
+    : public ::libaom_test::DecoderTest,
+      public ::libaom_test::CodecTestWithParam<const char *> {
+ protected:
+  ExternalFrameBufferMD5Test()
+      : DecoderTest(GET_PARAM(::libaom_test::kCodecFactoryParam)),
+        md5_file_(NULL), num_buffers_(0) {}
+
+  virtual ~ExternalFrameBufferMD5Test() {
+    if (md5_file_ != NULL) fclose(md5_file_);
+  }
+
+  virtual void PreDecodeFrameHook(
+      const libaom_test::CompressedVideoSource &video,
+      libaom_test::Decoder *decoder) {
+    if (num_buffers_ > 0 && video.frame_number() == 0) {
+      // Have libvpx use frame buffers we create.
+      ASSERT_TRUE(fb_list_.CreateBufferList(num_buffers_));
+      ASSERT_EQ(AOM_CODEC_OK,
+                decoder->SetFrameBufferFunctions(GetAV1FrameBuffer,
+                                                 ReleaseAV1FrameBuffer, this));
+    }
+  }
+
+  void OpenMD5File(const std::string &md5_file_name_) {
+    md5_file_ = libaom_test::OpenTestDataFile(md5_file_name_);
+    ASSERT_TRUE(md5_file_ != NULL)
+        << "Md5 file open failed. Filename: " << md5_file_name_;
+  }
+
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     const unsigned int frame_number) {
+    ASSERT_TRUE(md5_file_ != NULL);
+    char expected_md5[33];
+    char junk[128];
+
+    // Read correct md5 checksums.
+    const int res = fscanf(md5_file_, "%s  %s", expected_md5, junk);
+    ASSERT_NE(EOF, res) << "Read md5 data failed";
+    expected_md5[32] = '\0';
+
+    ::libaom_test::MD5 md5_res;
+    md5_res.Add(&img);
+    const char *const actual_md5 = md5_res.Get();
+
+    // Check md5 match.
+    ASSERT_STREQ(expected_md5, actual_md5)
+        << "Md5 checksums don't match: frame number = " << frame_number;
+  }
+
+  // Callback to get a free external frame buffer. Return value < 0 is an
+  // error.
+  static int GetAV1FrameBuffer(void *user_priv, size_t min_size,
+                               aom_codec_frame_buffer_t *fb) {
+    ExternalFrameBufferMD5Test *const md5Test =
+        reinterpret_cast<ExternalFrameBufferMD5Test *>(user_priv);
+    return md5Test->fb_list_.GetFreeFrameBuffer(min_size, fb);
+  }
+
+  // Callback to release an external frame buffer. Return value < 0 is an
+  // error.
+  static int ReleaseAV1FrameBuffer(void *user_priv,
+                                   aom_codec_frame_buffer_t *fb) {
+    ExternalFrameBufferMD5Test *const md5Test =
+        reinterpret_cast<ExternalFrameBufferMD5Test *>(user_priv);
+    return md5Test->fb_list_.ReturnFrameBuffer(fb);
+  }
+
+  void set_num_buffers(int num_buffers) { num_buffers_ = num_buffers; }
+  int num_buffers() const { return num_buffers_; }
+
+ private:
+  FILE *md5_file_;
+  int num_buffers_;
+  ExternalFrameBufferList fb_list_;
+};
+
+#if CONFIG_WEBM_IO
+const char kAV1TestFile[] = "av1-1-b8-01-size-226x226.ivf";
+const char kAV1NonRefTestFile[] = "av1-1-b8-01-size-226x226.ivf";
+
+// Class for testing passing in external frame buffers to libvpx.
+class ExternalFrameBufferTest : public ::testing::Test {
+ protected:
+  ExternalFrameBufferTest() : video_(NULL), decoder_(NULL), num_buffers_(0) {}
+
+  virtual void SetUp() {
+    video_ = new libaom_test::IVFVideoSource(kAV1TestFile);
+    ASSERT_TRUE(video_ != NULL);
+    video_->Init();
+    video_->Begin();
+
+    aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+    decoder_ = new libaom_test::AV1Decoder(cfg, 0);
+    ASSERT_TRUE(decoder_ != NULL);
+  }
+
+  virtual void TearDown() {
+    delete decoder_;
+    decoder_ = NULL;
+    delete video_;
+    video_ = NULL;
+  }
+
+  // Passes the external frame buffer information to libvpx.
+  aom_codec_err_t SetFrameBufferFunctions(
+      int num_buffers, aom_get_frame_buffer_cb_fn_t cb_get,
+      aom_release_frame_buffer_cb_fn_t cb_release) {
+    if (num_buffers > 0) {
+      num_buffers_ = num_buffers;
+      EXPECT_TRUE(fb_list_.CreateBufferList(num_buffers_));
+    }
+
+    return decoder_->SetFrameBufferFunctions(cb_get, cb_release, &fb_list_);
+  }
+
+  aom_codec_err_t DecodeOneFrame() {
+    const aom_codec_err_t res =
+        decoder_->DecodeFrame(video_->cxdata(), video_->frame_size());
+    CheckDecodedFrames();
+    if (res == AOM_CODEC_OK) video_->Next();
+    return res;
+  }
+
+  aom_codec_err_t DecodeRemainingFrames() {
+    for (; video_->cxdata() != NULL; video_->Next()) {
+      const aom_codec_err_t res =
+          decoder_->DecodeFrame(video_->cxdata(), video_->frame_size());
+      if (res != AOM_CODEC_OK) return res;
+      CheckDecodedFrames();
+    }
+    return AOM_CODEC_OK;
+  }
+
+ protected:
+  void CheckDecodedFrames() {
+    libaom_test::DxDataIterator dec_iter = decoder_->GetDxData();
+    const aom_image_t *img = NULL;
+
+    // Get decompressed data
+    while ((img = dec_iter.Next()) != NULL) {
+      fb_list_.CheckXImageFrameBuffer(img);
+    }
+  }
+
+  libaom_test::IVFVideoSource *video_;
+  libaom_test::AV1Decoder *decoder_;
+  int num_buffers_;
+  ExternalFrameBufferList fb_list_;
+};
+
+class ExternalFrameBufferNonRefTest : public ExternalFrameBufferTest {
+ protected:
+  virtual void SetUp() {
+    video_ = new libaom_test::IVFVideoSource(kAV1NonRefTestFile);
+    ASSERT_TRUE(video_ != NULL);
+    video_->Init();
+    video_->Begin();
+
+    aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+    decoder_ = new libaom_test::AV1Decoder(cfg, 0);
+    ASSERT_TRUE(decoder_ != NULL);
+  }
+
+  virtual void CheckFrameBufferRelease() {
+    TearDown();
+    ASSERT_EQ(0, fb_list_.num_used_buffers());
+  }
+};
+#endif  // CONFIG_WEBM_IO
+
+// This test runs through the set of test vectors, and decodes them.
+// Libvpx will call into the application to allocate a frame buffer when
+// needed. The md5 checksums are computed for each frame in the video file.
+// If md5 checksums match the correct md5 data, then the test is passed.
+// Otherwise, the test failed.
+TEST_P(ExternalFrameBufferMD5Test, DISABLED_ExtFBMD5Match) {
+  const std::string filename = GET_PARAM(kVideoNameParam);
+
+  // Number of buffers equals #AOM_MAXIMUM_REF_BUFFERS +
+  // #AOM_MAXIMUM_WORK_BUFFERS + four jitter buffers.
+  const int jitter_buffers = 4;
+  const int num_buffers =
+      AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS + jitter_buffers;
+  set_num_buffers(num_buffers);
+
+  // Open compressed video file.
+  testing::internal::scoped_ptr<libaom_test::CompressedVideoSource> video;
+  if (filename.substr(filename.length() - 3, 3) == "ivf") {
+    video.reset(new libaom_test::IVFVideoSource(filename));
+  } else {
+#if CONFIG_WEBM_IO
+    video.reset(new libaom_test::WebMVideoSource(filename));
+#else
+    fprintf(stderr, "WebM IO is disabled, skipping test vector %s\n",
+            filename.c_str());
+    return;
+#endif
+  }
+  ASSERT_TRUE(video.get() != NULL);
+  video->Init();
+
+  // Construct md5 file name.
+  const std::string md5_filename = filename + ".md5";
+  OpenMD5File(md5_filename);
+
+  // Decode frame, and check the md5 matching.
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+}
+
+#if CONFIG_WEBM_IO
+TEST_F(ExternalFrameBufferTest, MinFrameBuffers) {
+  // Minimum number of external frame buffers for AV1 is
+  // #AOM_MAXIMUM_REF_BUFFERS + #AOM_MAXIMUM_WORK_BUFFERS.
+  const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(AOM_CODEC_OK,
+            SetFrameBufferFunctions(num_buffers, get_aom_frame_buffer,
+                                    release_aom_frame_buffer));
+  ASSERT_EQ(AOM_CODEC_OK, DecodeRemainingFrames());
+}
+
+TEST_F(ExternalFrameBufferTest, EightJitterBuffers) {
+  // Number of buffers equals #AOM_MAXIMUM_REF_BUFFERS +
+  // #AOM_MAXIMUM_WORK_BUFFERS + eight jitter buffers.
+  const int jitter_buffers = 8;
+  const int num_buffers =
+      AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS + jitter_buffers;
+  ASSERT_EQ(AOM_CODEC_OK,
+            SetFrameBufferFunctions(num_buffers, get_aom_frame_buffer,
+                                    release_aom_frame_buffer));
+  ASSERT_EQ(AOM_CODEC_OK, DecodeRemainingFrames());
+}
+
+TEST_F(ExternalFrameBufferTest, DISABLED_NotEnoughBuffers) {
+  // Minimum number of external frame buffers for AV1 is
+  // #AOM_MAXIMUM_REF_BUFFERS + #AOM_MAXIMUM_WORK_BUFFERS. Most files will
+  // only use 5 frame buffers at one time.
+  const int num_buffers = 2;
+  ASSERT_EQ(AOM_CODEC_OK,
+            SetFrameBufferFunctions(num_buffers, get_aom_frame_buffer,
+                                    release_aom_frame_buffer));
+  ASSERT_EQ(AOM_CODEC_OK, DecodeOneFrame());
+  // Only run this on long clips. Decoding a very short clip will return
+  // AOM_CODEC_OK even with only 2 buffers.
+  ASSERT_EQ(AOM_CODEC_MEM_ERROR, DecodeRemainingFrames());
+}
+
+TEST_F(ExternalFrameBufferTest, DISABLED_NoRelease) {
+  const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(AOM_CODEC_OK,
+            SetFrameBufferFunctions(num_buffers, get_aom_frame_buffer,
+                                    do_not_release_aom_frame_buffer));
+  ASSERT_EQ(AOM_CODEC_OK, DecodeOneFrame());
+  ASSERT_EQ(AOM_CODEC_MEM_ERROR, DecodeRemainingFrames());
+}
+
+TEST_F(ExternalFrameBufferTest, NullRealloc) {
+  const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(AOM_CODEC_OK,
+            SetFrameBufferFunctions(num_buffers, get_aom_zero_frame_buffer,
+                                    release_aom_frame_buffer));
+  ASSERT_EQ(AOM_CODEC_MEM_ERROR, DecodeOneFrame());
+}
+
+TEST_F(ExternalFrameBufferTest, ReallocOneLessByte) {
+  const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(AOM_CODEC_OK, SetFrameBufferFunctions(
+                              num_buffers, get_aom_one_less_byte_frame_buffer,
+                              release_aom_frame_buffer));
+  ASSERT_EQ(AOM_CODEC_MEM_ERROR, DecodeOneFrame());
+}
+
+TEST_F(ExternalFrameBufferTest, NullGetFunction) {
+  const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(
+      AOM_CODEC_INVALID_PARAM,
+      SetFrameBufferFunctions(num_buffers, NULL, release_aom_frame_buffer));
+}
+
+TEST_F(ExternalFrameBufferTest, NullReleaseFunction) {
+  const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(AOM_CODEC_INVALID_PARAM,
+            SetFrameBufferFunctions(num_buffers, get_aom_frame_buffer, NULL));
+}
+
+TEST_F(ExternalFrameBufferTest, SetAfterDecode) {
+  const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(AOM_CODEC_OK, DecodeOneFrame());
+  ASSERT_EQ(AOM_CODEC_ERROR,
+            SetFrameBufferFunctions(num_buffers, get_aom_frame_buffer,
+                                    release_aom_frame_buffer));
+}
+
+TEST_F(ExternalFrameBufferNonRefTest, ReleaseNonRefFrameBuffer) {
+  const int num_buffers = AOM_MAXIMUM_REF_BUFFERS + AOM_MAXIMUM_WORK_BUFFERS;
+  ASSERT_EQ(AOM_CODEC_OK,
+            SetFrameBufferFunctions(num_buffers, get_aom_frame_buffer,
+                                    release_aom_frame_buffer));
+  ASSERT_EQ(AOM_CODEC_OK, DecodeRemainingFrames());
+  CheckFrameBufferRelease();
+}
+#endif  // CONFIG_WEBM_IO
+
+AV1_INSTANTIATE_TEST_CASE(
+    ExternalFrameBufferMD5Test,
+    ::testing::ValuesIn(libaom_test::kAV1TestVectors,
+                        libaom_test::kAV1TestVectors +
+                            libaom_test::kNumAV1TestVectors));
+}  // namespace
diff --git a/third_party/aom/test/fft_test.cc b/third_party/aom/test/fft_test.cc
new file mode 100644
index 0000000000..e24e451a33
--- /dev/null
+++ b/third_party/aom/test/fft_test.cc
@@ -0,0 +1,256 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+
+#include <algorithm>
+#include <complex>
+#include <vector>
+
+#include "aom_dsp/fft_common.h"
+#include "aom_mem/aom_mem.h"
+#include "av1/common/common.h"
+#include "config/aom_dsp_rtcd.h"
+#include "test/acm_random.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+
+typedef void (*tform_fun_t)(const float *input, float *temp, float *output);
+
+// Simple 1D FFT implementation
+template <typename InputType>
+void fft(const InputType *data, std::complex<float> *result, int n) {
+  if (n == 1) {
+    result[0] = data[0];
+    return;
+  }
+  std::vector<InputType> temp(n);
+  for (int k = 0; k < n / 2; ++k) {
+    temp[k] = data[2 * k];
+    temp[n / 2 + k] = data[2 * k + 1];
+  }
+  fft(&temp[0], result, n / 2);
+  fft(&temp[n / 2], result + n / 2, n / 2);
+  for (int k = 0; k < n / 2; ++k) {
+    std::complex<float> w = std::complex<float>((float)cos(2. * PI * k / n),
+                                                (float)-sin(2. * PI * k / n));
+    std::complex<float> a = result[k];
+    std::complex<float> b = result[n / 2 + k];
+    result[k] = a + w * b;
+    result[n / 2 + k] = a - w * b;
+  }
+}
+
+void transpose(std::vector<std::complex<float> > *data, int n) {
+  for (int y = 0; y < n; ++y) {
+    for (int x = y + 1; x < n; ++x) {
+      std::swap((*data)[y * n + x], (*data)[x * n + y]);
+    }
+  }
+}
+
+// Simple 2D FFT implementation
+template <class InputType>
+std::vector<std::complex<float> > fft2d(const InputType *input, int n) {
+  std::vector<std::complex<float> > rowfft(n * n);
+  std::vector<std::complex<float> > result(n * n);
+  for (int y = 0; y < n; ++y) {
+    fft(input + y * n, &rowfft[y * n], n);
+  }
+  transpose(&rowfft, n);
+  for (int y = 0; y < n; ++y) {
+    fft(&rowfft[y * n], &result[y * n], n);
+  }
+  transpose(&result, n);
+  return result;
+}
+
+struct FFTTestArg {
+  int n;
+  void (*fft)(const float *input, float *temp, float *output);
+  FFTTestArg(int n_in, tform_fun_t fft_in) : n(n_in), fft(fft_in) {}
+};
+
+std::ostream &operator<<(std::ostream &os, const FFTTestArg &test_arg) {
+  return os << "fft_arg { n:" << test_arg.n << " fft:" << test_arg.fft << " }";
+}
+
+class FFT2DTest : public ::testing::TestWithParam<FFTTestArg> {
+ protected:
+  void SetUp() {
+    int n = GetParam().n;
+    input_ = (float *)aom_memalign(32, sizeof(*input_) * n * n);
+    temp_ = (float *)aom_memalign(32, sizeof(*temp_) * n * n);
+    output_ = (float *)aom_memalign(32, sizeof(*output_) * n * n * 2);
+    memset(input_, 0, sizeof(*input_) * n * n);
+    memset(temp_, 0, sizeof(*temp_) * n * n);
+    memset(output_, 0, sizeof(*output_) * n * n * 2);
+  }
+  void TearDown() {
+    aom_free(input_);
+    aom_free(temp_);
+    aom_free(output_);
+  }
+  float *input_;
+  float *temp_;
+  float *output_;
+};
+
+TEST_P(FFT2DTest, Correct) {
+  int n = GetParam().n;
+  for (int i = 0; i < n * n; ++i) {
+    input_[i] = 1;
+    std::vector<std::complex<float> > expected = fft2d<float>(&input_[0], n);
+    GetParam().fft(&input_[0], &temp_[0], &output_[0]);
+    for (int y = 0; y < n; ++y) {
+      for (int x = 0; x < (n / 2) + 1; ++x) {
+        EXPECT_NEAR(expected[y * n + x].real(), output_[2 * (y * n + x)], 1e-5);
+        EXPECT_NEAR(expected[y * n + x].imag(), output_[2 * (y * n + x) + 1],
+                    1e-5);
+      }
+    }
+    input_[i] = 0;
+  }
+}
+
+TEST_P(FFT2DTest, Benchmark) {
+  int n = GetParam().n;
+  float sum = 0;
+  for (int i = 0; i < 1000 * (64 - n); ++i) {
+    input_[i % (n * n)] = 1;
+    GetParam().fft(&input_[0], &temp_[0], &output_[0]);
+    sum += output_[0];
+    input_[i % (n * n)] = 0;
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(C, FFT2DTest,
+                        ::testing::Values(FFTTestArg(2, aom_fft2x2_float_c),
+                                          FFTTestArg(4, aom_fft4x4_float_c),
+                                          FFTTestArg(8, aom_fft8x8_float_c),
+                                          FFTTestArg(16, aom_fft16x16_float_c),
+                                          FFTTestArg(32,
+                                                     aom_fft32x32_float_c)));
+#if ARCH_X86 || ARCH_X86_64
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, FFT2DTest,
+    ::testing::Values(FFTTestArg(4, aom_fft4x4_float_sse2),
+                      FFTTestArg(8, aom_fft8x8_float_sse2),
+                      FFTTestArg(16, aom_fft16x16_float_sse2),
+                      FFTTestArg(32, aom_fft32x32_float_sse2)));
+#endif  // HAVE_SSE2
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, FFT2DTest,
+    ::testing::Values(FFTTestArg(8, aom_fft8x8_float_avx2),
+                      FFTTestArg(16, aom_fft16x16_float_avx2),
+                      FFTTestArg(32, aom_fft32x32_float_avx2)));
+#endif  // HAVE_AVX2
+#endif  // ARCH_X86 || ARCH_X86_64
+
+struct IFFTTestArg {
+  int n;
+  tform_fun_t ifft;
+  IFFTTestArg(int n_in, tform_fun_t ifft_in) : n(n_in), ifft(ifft_in) {}
+};
+
+std::ostream &operator<<(std::ostream &os, const IFFTTestArg &test_arg) {
+  return os << "ifft_arg { n:" << test_arg.n << " fft:" << test_arg.ifft
+            << " }";
+}
+
+class IFFT2DTest : public ::testing::TestWithParam<IFFTTestArg> {
+ protected:
+  void SetUp() {
+    int n = GetParam().n;
+    input_ = (float *)aom_memalign(32, sizeof(*input_) * n * n * 2);
+    temp_ = (float *)aom_memalign(32, sizeof(*temp_) * n * n * 2);
+    output_ = (float *)aom_memalign(32, sizeof(*output_) * n * n);
+    memset(input_, 0, sizeof(*input_) * n * n * 2);
+    memset(temp_, 0, sizeof(*temp_) * n * n * 2);
+    memset(output_, 0, sizeof(*output_) * n * n);
+  }
+  void TearDown() {
+    aom_free(input_);
+    aom_free(temp_);
+    aom_free(output_);
+  }
+  float *input_;
+  float *temp_;
+  float *output_;
+};
+
+TEST_P(IFFT2DTest, Correctness) {
+  int n = GetParam().n;
+  ASSERT_GE(n, 2);
+  std::vector<float> expected(n * n);
+  std::vector<float> actual(n * n);
+  // Do forward transform then invert to make sure we get back expected
+  for (int y = 0; y < n; ++y) {
+    for (int x = 0; x < n; ++x) {
+      expected[y * n + x] = 1;
+      std::vector<std::complex<float> > input_c = fft2d(&expected[0], n);
+      for (int i = 0; i < n * n; ++i) {
+        input_[2 * i + 0] = input_c[i].real();
+        input_[2 * i + 1] = input_c[i].imag();
+      }
+      GetParam().ifft(&input_[0], &temp_[0], &output_[0]);
+
+      for (int yy = 0; yy < n; ++yy) {
+        for (int xx = 0; xx < n; ++xx) {
+          EXPECT_NEAR(expected[yy * n + xx], output_[yy * n + xx] / (n * n),
+                      1e-5);
+        }
+      }
+      expected[y * n + x] = 0;
+    }
+  }
+};
+
+TEST_P(IFFT2DTest, Benchmark) {
+  int n = GetParam().n;
+  float sum = 0;
+  for (int i = 0; i < 1000 * (64 - n); ++i) {
+    input_[i % (n * n)] = 1;
+    GetParam().ifft(&input_[0], &temp_[0], &output_[0]);
+    sum += output_[0];
+    input_[i % (n * n)] = 0;
+  }
+}
+INSTANTIATE_TEST_CASE_P(
+    C, IFFT2DTest,
+    ::testing::Values(IFFTTestArg(2, aom_ifft2x2_float_c),
+                      IFFTTestArg(4, aom_ifft4x4_float_c),
+                      IFFTTestArg(8, aom_ifft8x8_float_c),
+                      IFFTTestArg(16, aom_ifft16x16_float_c),
+                      IFFTTestArg(32, aom_ifft32x32_float_c)));
+#if ARCH_X86 || ARCH_X86_64
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(
+    SSE2, IFFT2DTest,
+    ::testing::Values(IFFTTestArg(4, aom_ifft4x4_float_sse2),
+                      IFFTTestArg(8, aom_ifft8x8_float_sse2),
+                      IFFTTestArg(16, aom_ifft16x16_float_sse2),
+                      IFFTTestArg(32, aom_ifft32x32_float_sse2)));
+#endif  // HAVE_SSE2
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, IFFT2DTest,
+    ::testing::Values(IFFTTestArg(8, aom_ifft8x8_float_avx2),
+                      IFFTTestArg(16, aom_ifft16x16_float_avx2),
+                      IFFTTestArg(32, aom_ifft32x32_float_avx2)));
+#endif  // HAVE_AVX2
+#endif  // ARCH_X86 || ARCH_X86_64
+
+}  // namespace
diff --git a/third_party/aom/test/film_grain_table_test.cc b/third_party/aom/test/film_grain_table_test.cc
new file mode 100644
index 0000000000..524d67d7bc
--- /dev/null
+++ b/third_party/aom/test/film_grain_table_test.cc
@@ -0,0 +1,250 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "aom_dsp/grain_table.h"
+#include "aom/internal/aom_codec_internal.h"
+#include "av1/encoder/grain_test_vectors.h"
+#include "test/video_source.h"
+
+void grain_equal(const aom_film_grain_t *expected,
+                 const aom_film_grain_t *actual) {
+  EXPECT_EQ(expected->apply_grain, actual->apply_grain);
+  EXPECT_EQ(expected->update_parameters, actual->update_parameters);
+  if (!expected->update_parameters) return;
+  EXPECT_EQ(expected->num_y_points, actual->num_y_points);
+  EXPECT_EQ(expected->num_cb_points, actual->num_cb_points);
+  EXPECT_EQ(expected->num_cr_points, actual->num_cr_points);
+  EXPECT_EQ(0, memcmp(expected->scaling_points_y, actual->scaling_points_y,
+                      expected->num_y_points *
+                          sizeof(expected->scaling_points_y[0])));
+  EXPECT_EQ(0, memcmp(expected->scaling_points_cb, actual->scaling_points_cb,
+                      expected->num_cb_points *
+                          sizeof(expected->scaling_points_cb[0])));
+  EXPECT_EQ(0, memcmp(expected->scaling_points_cr, actual->scaling_points_cr,
+                      expected->num_cr_points *
+                          sizeof(expected->scaling_points_cr[0])));
+  EXPECT_EQ(expected->scaling_shift, actual->scaling_shift);
+  EXPECT_EQ(expected->ar_coeff_lag, actual->ar_coeff_lag);
+  EXPECT_EQ(expected->ar_coeff_shift, actual->ar_coeff_shift);
+
+  const int num_pos_luma =
+      2 * expected->ar_coeff_lag * (expected->ar_coeff_lag + 1);
+  const int num_pos_chroma = num_pos_luma;
+  EXPECT_EQ(0, memcmp(expected->ar_coeffs_y, actual->ar_coeffs_y,
+                      sizeof(expected->ar_coeffs_y[0]) * num_pos_luma));
+  if (actual->num_cb_points || actual->chroma_scaling_from_luma) {
+    EXPECT_EQ(0, memcmp(expected->ar_coeffs_cb, actual->ar_coeffs_cb,
+                        sizeof(expected->ar_coeffs_cb[0]) * num_pos_chroma));
+  }
+  if (actual->num_cr_points || actual->chroma_scaling_from_luma) {
+    EXPECT_EQ(0, memcmp(expected->ar_coeffs_cr, actual->ar_coeffs_cr,
+                        sizeof(expected->ar_coeffs_cr[0]) * num_pos_chroma));
+  }
+  EXPECT_EQ(expected->overlap_flag, actual->overlap_flag);
+  EXPECT_EQ(expected->chroma_scaling_from_luma,
+            actual->chroma_scaling_from_luma);
+  EXPECT_EQ(expected->grain_scale_shift, actual->grain_scale_shift);
+  // EXPECT_EQ(expected->random_seed, actual->random_seed);
+
+  // clip_to_restricted and bit_depth aren't written
+  if (expected->num_cb_points) {
+    EXPECT_EQ(expected->cb_mult, actual->cb_mult);
+    EXPECT_EQ(expected->cb_luma_mult, actual->cb_luma_mult);
+    EXPECT_EQ(expected->cb_offset, actual->cb_offset);
+  }
+  if (expected->num_cr_points) {
+    EXPECT_EQ(expected->cr_mult, actual->cr_mult);
+    EXPECT_EQ(expected->cr_luma_mult, actual->cr_luma_mult);
+    EXPECT_EQ(expected->cr_offset, actual->cr_offset);
+  }
+}
+
+TEST(FilmGrainTableTest, AddAndLookupSingleSegment) {
+  aom_film_grain_table_t table;
+  memset(&table, 0, sizeof(table));
+
+  aom_film_grain_t grain;
+  EXPECT_FALSE(aom_film_grain_table_lookup(&table, 0, 1000, false, &grain));
+
+  aom_film_grain_table_append(&table, 1000, 2000, film_grain_test_vectors + 0);
+  EXPECT_FALSE(aom_film_grain_table_lookup(&table, 0, 1000, false, &grain));
+  EXPECT_FALSE(aom_film_grain_table_lookup(&table, 2000, 3000, false, &grain));
+
+  EXPECT_TRUE(aom_film_grain_table_lookup(&table, 1000, 2000, false, &grain));
+
+  grain.bit_depth = film_grain_test_vectors[0].bit_depth;
+  EXPECT_EQ(0, memcmp(&grain, film_grain_test_vectors + 0, sizeof(table)));
+
+  // Extend the existing segment
+  aom_film_grain_table_append(&table, 2000, 3000, film_grain_test_vectors + 0);
+  EXPECT_EQ(0, table.head->next);
+
+  // Lookup and remove and check that the entry is no longer there
+  EXPECT_TRUE(aom_film_grain_table_lookup(&table, 1000, 2000, true, &grain));
+  EXPECT_FALSE(aom_film_grain_table_lookup(&table, 1000, 2000, false, &grain));
+
+  EXPECT_TRUE(aom_film_grain_table_lookup(&table, 2000, 3000, true, &grain));
+  EXPECT_FALSE(aom_film_grain_table_lookup(&table, 2000, 3000, false, &grain));
+
+  EXPECT_EQ(0, table.head);
+  EXPECT_EQ(0, table.tail);
+  aom_film_grain_table_free(&table);
+}
+
+TEST(FilmGrainTableTest, SplitSingleSegment) {
+  aom_film_grain_table_t table;
+  aom_film_grain_t grain;
+  memset(&table, 0, sizeof(table));
+
+  aom_film_grain_table_append(&table, 0, 1000, film_grain_test_vectors + 0);
+
+  // Test lookup and remove that adjusts start time
+  EXPECT_TRUE(aom_film_grain_table_lookup(&table, 0, 100, true, &grain));
+  EXPECT_EQ(NULL, table.head->next);
+  EXPECT_EQ(100, table.head->start_time);
+
+  // Test lookup and remove that adjusts end time
+  EXPECT_TRUE(aom_film_grain_table_lookup(&table, 900, 1000, true, &grain));
+  EXPECT_EQ(NULL, table.head->next);
+  EXPECT_EQ(100, table.head->start_time);
+  EXPECT_EQ(900, table.head->end_time);
+
+  // Test lookup and remove that splits the first entry
+  EXPECT_TRUE(aom_film_grain_table_lookup(&table, 400, 600, true, &grain));
+  EXPECT_EQ(100, table.head->start_time);
+  EXPECT_EQ(400, table.head->end_time);
+
+  ASSERT_NE((void *)NULL, table.head->next);
+  EXPECT_EQ(table.tail, table.head->next);
+  EXPECT_EQ(600, table.head->next->start_time);
+  EXPECT_EQ(900, table.head->next->end_time);
+
+  aom_film_grain_table_free(&table);
+}
+
+TEST(FilmGrainTableTest, AddAndLookupMultipleSegments) {
+  aom_film_grain_table_t table;
+  memset(&table, 0, sizeof(table));
+
+  aom_film_grain_t grain;
+  const int kNumTestVectors =
+      sizeof(film_grain_test_vectors) / sizeof(film_grain_test_vectors[0]);
+  for (int i = 0; i < kNumTestVectors; ++i) {
+    aom_film_grain_table_append(&table, i * 1000, (i + 1) * 1000,
+                                film_grain_test_vectors + i);
+  }
+
+  for (int i = kNumTestVectors - 1; i >= 0; --i) {
+    EXPECT_TRUE(aom_film_grain_table_lookup(&table, i * 1000, (i + 1) * 1000,
+                                            true, &grain));
+    grain_equal(film_grain_test_vectors + i, &grain);
+    EXPECT_FALSE(aom_film_grain_table_lookup(&table, i * 1000, (i + 1) * 1000,
+                                             true, &grain));
+  }
+
+  // Verify that all the data has been removed
+  for (int i = 0; i < kNumTestVectors; ++i) {
+    EXPECT_FALSE(aom_film_grain_table_lookup(&table, i * 1000, (i + 1) * 1000,
+                                             true, &grain));
+  }
+  aom_film_grain_table_free(&table);
+}
+
+class FilmGrainTableIOTest : public ::testing::Test {
+ protected:
+  void SetUp() { memset(&error_, 0, sizeof(error_)); }
+  struct aom_internal_error_info error_;
+};
+
+TEST_F(FilmGrainTableIOTest, ReadMissingFile) {
+  aom_film_grain_table_t table;
+  memset(&table, 0, sizeof(table));
+  ASSERT_EQ(AOM_CODEC_ERROR, aom_film_grain_table_read(
+                                 &table, "/path/to/missing/file", &error_));
+}
+
+TEST_F(FilmGrainTableIOTest, ReadTruncatedFile) {
+  aom_film_grain_table_t table;
+  memset(&table, 0, sizeof(table));
+
+  std::string grain_file;
+  FILE *file = libaom_test::GetTempOutFile(&grain_file);
+  fwrite("deadbeef", 8, 1, file);
+  fclose(file);
+  ASSERT_EQ(AOM_CODEC_ERROR,
+            aom_film_grain_table_read(&table, grain_file.c_str(), &error_));
+  EXPECT_EQ(0, remove(grain_file.c_str()));
+}
+
+TEST_F(FilmGrainTableIOTest, RoundTripReadWrite) {
+  aom_film_grain_table_t table;
+  memset(&table, 0, sizeof(table));
+
+  aom_film_grain_t expected_grain[16];
+  const int kNumTestVectors =
+      sizeof(film_grain_test_vectors) / sizeof(film_grain_test_vectors[0]);
+  for (int i = 0; i < kNumTestVectors; ++i) {
+    expected_grain[i] = film_grain_test_vectors[i];
+    expected_grain[i].random_seed = i;
+    expected_grain[i].update_parameters = i % 2;
+    expected_grain[i].apply_grain = (i + 1) % 2;
+    expected_grain[i].bit_depth = 0;
+    aom_film_grain_table_append(&table, i * 1000, (i + 1) * 1000,
+                                expected_grain + i);
+  }
+  std::string grain_file;
+  fclose(libaom_test::GetTempOutFile(&grain_file));
+  ASSERT_EQ(AOM_CODEC_OK,
+            aom_film_grain_table_write(&table, grain_file.c_str(), &error_));
+  aom_film_grain_table_free(&table);
+
+  memset(&table, 0, sizeof(table));
+  ASSERT_EQ(AOM_CODEC_OK,
+            aom_film_grain_table_read(&table, grain_file.c_str(), &error_));
+  for (int i = 0; i < kNumTestVectors; ++i) {
+    aom_film_grain_t grain;
+    EXPECT_TRUE(aom_film_grain_table_lookup(&table, i * 1000, (i + 1) * 1000,
+                                            true, &grain));
+    grain_equal(expected_grain + i, &grain);
+  }
+  aom_film_grain_table_free(&table);
+  EXPECT_EQ(0, remove(grain_file.c_str()));
+}
+
+TEST_F(FilmGrainTableIOTest, RoundTripSplit) {
+  std::string grain_file;
+  fclose(libaom_test::GetTempOutFile(&grain_file));
+
+  aom_film_grain_table_t table;
+  memset(&table, 0, sizeof(table));
+
+  aom_film_grain_t grain = film_grain_test_vectors[0];
+  aom_film_grain_table_append(&table, 0, 3000, &grain);
+  ASSERT_TRUE(aom_film_grain_table_lookup(&table, 1000, 2000, true, &grain));
+  ASSERT_TRUE(aom_film_grain_table_lookup(&table, 0, 1000, false, &grain));
+  EXPECT_FALSE(aom_film_grain_table_lookup(&table, 1000, 2000, false, &grain));
+  ASSERT_TRUE(aom_film_grain_table_lookup(&table, 2000, 3000, false, &grain));
+  ASSERT_EQ(AOM_CODEC_OK,
+            aom_film_grain_table_write(&table, grain_file.c_str(), &error_));
+  aom_film_grain_table_free(&table);
+
+  memset(&table, 0, sizeof(table));
+  ASSERT_EQ(AOM_CODEC_OK,
+            aom_film_grain_table_read(&table, grain_file.c_str(), &error_));
+  ASSERT_TRUE(aom_film_grain_table_lookup(&table, 0, 1000, false, &grain));
+  ASSERT_FALSE(aom_film_grain_table_lookup(&table, 1000, 2000, false, &grain));
+  ASSERT_TRUE(aom_film_grain_table_lookup(&table, 2000, 3000, false, &grain));
+  aom_film_grain_table_free(&table);
+
+  EXPECT_EQ(0, remove(grain_file.c_str()));
+}
diff --git a/third_party/aom/test/filterintra_test.cc b/third_party/aom/test/filterintra_test.cc
new file mode 100644
index 0000000000..5971349406
--- /dev/null
+++ b/third_party/aom/test/filterintra_test.cc
@@ -0,0 +1,134 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/av1_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "av1/common/enums.h"
+
+namespace {
+
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+
+typedef void (*Predictor)(uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size,
+                          const uint8_t *above, const uint8_t *left, int mode);
+
+// Note:
+//  Test parameter list:
+//  Reference predictor, optimized predictor, prediction mode, tx size
+//
+typedef tuple<Predictor, Predictor, int> PredFuncMode;
+typedef tuple<PredFuncMode, TX_SIZE> PredParams;
+
+const int MaxTxSize = 32;
+
+const int MaxTestNum = 100;
+
+class AV1FilterIntraPredTest : public ::testing::TestWithParam<PredParams> {
+ public:
+  virtual ~AV1FilterIntraPredTest() {}
+  virtual void SetUp() {
+    PredFuncMode funcMode = GET_PARAM(0);
+    predFuncRef_ = ::testing::get<0>(funcMode);
+    predFunc_ = ::testing::get<1>(funcMode);
+    mode_ = ::testing::get<2>(funcMode);
+    txSize_ = GET_PARAM(1);
+
+    alloc_ = new uint8_t[2 * MaxTxSize + 1];
+    predRef_ = new uint8_t[MaxTxSize * MaxTxSize];
+    pred_ = new uint8_t[MaxTxSize * MaxTxSize];
+  }
+
+  virtual void TearDown() {
+    delete[] alloc_;
+    delete[] predRef_;
+    delete[] pred_;
+    libaom_test::ClearSystemState();
+  }
+
+ protected:
+  void RunTest() const {
+    int tstIndex = 0;
+    int stride = tx_size_wide[txSize_];
+    uint8_t *left = alloc_;
+    uint8_t *above = alloc_ + MaxTxSize;
+    while (tstIndex < MaxTestNum) {
+      PrepareBuffer();
+      predFuncRef_(predRef_, stride, txSize_, &above[1], left, mode_);
+      ASM_REGISTER_STATE_CHECK(
+          predFunc_(pred_, stride, txSize_, &above[1], left, mode_));
+      DiffPred(tstIndex);
+      tstIndex += 1;
+    }
+  }
+
+ private:
+  void PrepareBuffer() const {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    int i = 0;
+    while (i < (2 * MaxTxSize + 1)) {
+      alloc_[i] = rnd.Rand8();
+      i++;
+    }
+  }
+
+  void DiffPred(int testNum) const {
+    int i = 0;
+    while (i < tx_size_wide[txSize_] * tx_size_high[txSize_]) {
+      EXPECT_EQ(predRef_[i], pred_[i]) << "Error at position: " << i << " "
+                                       << "Tx size: " << tx_size_wide[txSize_]
+                                       << "x" << tx_size_high[txSize_] << " "
+                                       << "Test number: " << testNum;
+      i++;
+    }
+  }
+
+  Predictor predFunc_;
+  Predictor predFuncRef_;
+  int mode_;
+  TX_SIZE txSize_;
+  uint8_t *alloc_;
+  uint8_t *pred_;
+  uint8_t *predRef_;
+};
+
+TEST_P(AV1FilterIntraPredTest, BitExactCheck) { RunTest(); }
+
+using ::testing::make_tuple;
+
+const PredFuncMode kPredFuncMdArray[] = {
+  make_tuple(&av1_filter_intra_predictor_c, &av1_filter_intra_predictor_sse4_1,
+             FILTER_DC_PRED),
+  make_tuple(&av1_filter_intra_predictor_c, &av1_filter_intra_predictor_sse4_1,
+             FILTER_V_PRED),
+  make_tuple(&av1_filter_intra_predictor_c, &av1_filter_intra_predictor_sse4_1,
+             FILTER_H_PRED),
+  make_tuple(&av1_filter_intra_predictor_c, &av1_filter_intra_predictor_sse4_1,
+             FILTER_D157_PRED),
+  make_tuple(&av1_filter_intra_predictor_c, &av1_filter_intra_predictor_sse4_1,
+             FILTER_PAETH_PRED),
+};
+
+const TX_SIZE kTxSize[] = { TX_4X4,  TX_8X8,  TX_16X16, TX_32X32, TX_4X8,
+                            TX_8X4,  TX_8X16, TX_16X8,  TX_16X32, TX_32X16,
+                            TX_4X16, TX_16X4, TX_8X32,  TX_32X8 };
+
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AV1FilterIntraPredTest,
+    ::testing::Combine(::testing::ValuesIn(kPredFuncMdArray),
+                       ::testing::ValuesIn(kTxSize)));
+}  // namespace
diff --git a/third_party/aom/test/frame_size_tests.cc b/third_party/aom/test/frame_size_tests.cc
new file mode 100644
index 0000000000..eaf0b8370a
--- /dev/null
+++ b/third_party/aom/test/frame_size_tests.cc
@@ -0,0 +1,78 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/video_source.h"
+
+namespace {
+
+class AV1FrameSizeTests : public ::testing::Test,
+                          public ::libaom_test::EncoderTest {
+ protected:
+  AV1FrameSizeTests()
+      : EncoderTest(&::libaom_test::kAV1), expected_res_(AOM_CODEC_OK) {}
+  virtual ~AV1FrameSizeTests() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(::libaom_test::kRealTime);
+  }
+
+  virtual bool HandleDecodeResult(const aom_codec_err_t res_dec,
+                                  libaom_test::Decoder *decoder) {
+    EXPECT_EQ(expected_res_, res_dec) << decoder->DecodeError();
+    return !::testing::Test::HasFailure();
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_CPUUSED, 7);
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+      encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+      encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+    }
+  }
+
+  int expected_res_;
+};
+
+#if CONFIG_SIZE_LIMIT
+TEST_F(AV1FrameSizeTests, TestInvalidSizes) {
+  ::libaom_test::RandomVideoSource video;
+
+  video.SetSize(DECODE_WIDTH_LIMIT + 16, DECODE_HEIGHT_LIMIT + 16);
+  video.set_limit(2);
+  expected_res_ = AOM_CODEC_CORRUPT_FRAME;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
+TEST_F(AV1FrameSizeTests, LargeValidSizes) {
+  ::libaom_test::RandomVideoSource video;
+
+  video.SetSize(DECODE_WIDTH_LIMIT, DECODE_HEIGHT_LIMIT);
+  video.set_limit(2);
+  expected_res_ = AOM_CODEC_OK;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+#endif
+
+TEST_F(AV1FrameSizeTests, OneByOneVideo) {
+  ::libaom_test::RandomVideoSource video;
+
+  video.SetSize(1, 1);
+  video.set_limit(2);
+  expected_res_ = AOM_CODEC_OK;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+#undef ONE_BY_ONE_VIDEO_NAME
+}  // namespace
diff --git a/third_party/aom/test/function_equivalence_test.h b/third_party/aom/test/function_equivalence_test.h
new file mode 100644
index 0000000000..f270689023
--- /dev/null
+++ b/third_party/aom/test/function_equivalence_test.h
@@ -0,0 +1,69 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_FUNCTION_EQUIVALENCE_TEST_H_
+#define AOM_TEST_FUNCTION_EQUIVALENCE_TEST_H_
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/util.h"
+
+using libaom_test::ACMRandom;
+
+namespace libaom_test {
+// Base class for tests that compare 2 implementations of the same function
+// for equivalence. The template parameter should be pointer to a function
+// that is being tested.
+//
+// The test takes a 3-parameters encapsulating struct 'FuncParam', containing:
+//   - Pointer to reference function
+//   - Pointer to tested function
+//   - Integer bit depth (default to 0).
+//
+// These values are then accessible in the tests as member of params_:
+// params_.ref_func, params_.tst_func, and params_.bit_depth.
+//
+
+template <typename T>
+struct FuncParam {
+  FuncParam(T ref = NULL, T tst = NULL, int bit_depth = 0)
+      : ref_func(ref), tst_func(tst), bit_depth(bit_depth) {}
+  T ref_func;
+  T tst_func;
+  int bit_depth;
+};
+
+template <typename T>
+std::ostream &operator<<(std::ostream &os, const FuncParam<T> &p) {
+  return os << "bit_depth:" << p.bit_depth
+            << " function:" << reinterpret_cast<const void *>(p.ref_func)
+            << " function:" << reinterpret_cast<const void *>(p.tst_func);
+}
+
+template <typename T>
+class FunctionEquivalenceTest : public ::testing::TestWithParam<FuncParam<T> > {
+ public:
+  FunctionEquivalenceTest() : rng_(ACMRandom::DeterministicSeed()) {}
+
+  virtual ~FunctionEquivalenceTest() {}
+
+  virtual void SetUp() { params_ = this->GetParam(); }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  ACMRandom rng_;
+  FuncParam<T> params_;
+};
+
+}  // namespace libaom_test
+#endif  // AOM_TEST_FUNCTION_EQUIVALENCE_TEST_H_
diff --git a/third_party/aom/test/fwht4x4_test.cc b/third_party/aom/test/fwht4x4_test.cc
new file mode 100644
index 0000000000..c8d98c5198
--- /dev/null
+++ b/third_party/aom/test/fwht4x4_test.cc
@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/av1_rtcd.h"
+#include "config/aom_dsp_rtcd.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/transform_test_base.h"
+#include "test/util.h"
+#include "av1/common/entropy.h"
+#include "aom/aom_codec.h"
+#include "aom/aom_integer.h"
+#include "aom_ports/mem.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+typedef void (*FdctFunc)(const int16_t *in, tran_low_t *out, int stride);
+typedef void (*IdctFunc)(const tran_low_t *in, uint8_t *out, int stride);
+
+using libaom_test::FhtFunc;
+
+typedef ::testing::tuple<FdctFunc, IdctFunc, TX_TYPE, aom_bit_depth_t, int>
+    Dct4x4Param;
+
+void fwht4x4_ref(const int16_t *in, tran_low_t *out, int stride,
+                 TxfmParam * /*txfm_param*/) {
+  av1_fwht4x4_c(in, out, stride);
+}
+
+void iwht4x4_10(const tran_low_t *in, uint8_t *out, int stride) {
+  av1_highbd_iwht4x4_16_add_c(in, out, stride, 10);
+}
+
+void iwht4x4_12(const tran_low_t *in, uint8_t *out, int stride) {
+  av1_highbd_iwht4x4_16_add_c(in, out, stride, 12);
+}
+
+class Trans4x4WHT : public libaom_test::TransformTestBase,
+                    public ::testing::TestWithParam<Dct4x4Param> {
+ public:
+  virtual ~Trans4x4WHT() {}
+
+  virtual void SetUp() {
+    fwd_txfm_ = GET_PARAM(0);
+    inv_txfm_ = GET_PARAM(1);
+    pitch_ = 4;
+    height_ = 4;
+    fwd_txfm_ref = fwht4x4_ref;
+    bit_depth_ = GET_PARAM(3);
+    mask_ = (1 << bit_depth_) - 1;
+    num_coeffs_ = GET_PARAM(4);
+  }
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) {
+    fwd_txfm_(in, out, stride);
+  }
+  void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) {
+    inv_txfm_(out, dst, stride);
+  }
+
+  FdctFunc fwd_txfm_;
+  IdctFunc inv_txfm_;
+};
+
+TEST_P(Trans4x4WHT, AccuracyCheck) { RunAccuracyCheck(0, 0.00001); }
+
+TEST_P(Trans4x4WHT, CoeffCheck) { RunCoeffCheck(); }
+
+TEST_P(Trans4x4WHT, MemCheck) { RunMemCheck(); }
+
+TEST_P(Trans4x4WHT, InvAccuracyCheck) { RunInvAccuracyCheck(0); }
+using ::testing::make_tuple;
+
+INSTANTIATE_TEST_CASE_P(
+    C, Trans4x4WHT,
+    ::testing::Values(make_tuple(&av1_highbd_fwht4x4_c, &iwht4x4_10, DCT_DCT,
+                                 AOM_BITS_10, 16),
+                      make_tuple(&av1_highbd_fwht4x4_c, &iwht4x4_12, DCT_DCT,
+                                 AOM_BITS_12, 16)));
+}  // namespace
diff --git a/third_party/aom/test/gviz_api.py b/third_party/aom/test/gviz_api.py
new file mode 100755
index 0000000000..d3a443dabf
--- /dev/null
+++ b/third_party/aom/test/gviz_api.py
@@ -0,0 +1,1087 @@
+#!/usr/bin/python
+#
+# Copyright (c) 2016, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and
+# the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+# was not distributed with this source code in the LICENSE file, you can
+# obtain it at www.aomedia.org/license/software. If the Alliance for Open
+# Media Patent License 1.0 was not distributed with this source code in the
+# PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+#
+
+"""Converts Python data into data for Google Visualization API clients.
+
+This library can be used to create a google.visualization.DataTable usable by
+visualizations built on the Google Visualization API. Output formats are raw
+JSON, JSON response, JavaScript, CSV, and HTML table.
+
+See http://code.google.com/apis/visualization/ for documentation on the
+Google Visualization API.
+"""
+
+__author__ = "Amit Weinstein, Misha Seltzer, Jacob Baskin"
+
+import cgi
+import cStringIO
+import csv
+import datetime
+try:
+  import json
+except ImportError:
+  import simplejson as json
+import types
+
+
+class DataTableException(Exception):
+  """The general exception object thrown by DataTable."""
+  pass
+
+
+class DataTableJSONEncoder(json.JSONEncoder):
+  """JSON encoder that handles date/time/datetime objects correctly."""
+
+  def __init__(self):
+    json.JSONEncoder.__init__(self,
+                              separators=(",", ":"),
+                              ensure_ascii=False)
+
+  def default(self, o):
+    if isinstance(o, datetime.datetime):
+      if o.microsecond == 0:
+        # If the time doesn't have ms-resolution, leave it out to keep
+        # things smaller.
+        return "Date(%d,%d,%d,%d,%d,%d)" % (
+            o.year, o.month - 1, o.day, o.hour, o.minute, o.second)
+      else:
+        return "Date(%d,%d,%d,%d,%d,%d,%d)" % (
+            o.year, o.month - 1, o.day, o.hour, o.minute, o.second,
+            o.microsecond / 1000)
+    elif isinstance(o, datetime.date):
+      return "Date(%d,%d,%d)" % (o.year, o.month - 1, o.day)
+    elif isinstance(o, datetime.time):
+      return [o.hour, o.minute, o.second]
+    else:
+      return super(DataTableJSONEncoder, self).default(o)
+
+
+class DataTable(object):
+  """Wraps the data to convert to a Google Visualization API DataTable.
+
+  Create this object, populate it with data, then call one of the ToJS...
+  methods to return a string representation of the data in the format described.
+
+  You can clear all data from the object to reuse it, but you cannot clear
+  individual cells, rows, or columns. You also cannot modify the table schema
+  specified in the class constructor.
+
+  You can add new data one or more rows at a time. All data added to an
+  instantiated DataTable must conform to the schema passed in to __init__().
+
+  You can reorder the columns in the output table, and also specify row sorting
+  order by column. The default column order is according to the original
+  table_description parameter. Default row sort order is ascending, by column
+  1 values. For a dictionary, we sort the keys for order.
+
+  The data and the table_description are closely tied, as described here:
+
+  The table schema is defined in the class constructor's table_description
+  parameter. The user defines each column using a tuple of
+  (id[, type[, label[, custom_properties]]]). The default value for type is
+  string, label is the same as ID if not specified, and custom properties is
+  an empty dictionary if not specified.
+
+  table_description is a dictionary or list, containing one or more column
+  descriptor tuples, nested dictionaries, and lists. Each dictionary key, list
+  element, or dictionary element must eventually be defined as
+  a column description tuple. Here's an example of a dictionary where the key
+  is a tuple, and the value is a list of two tuples:
+    {('a', 'number'): [('b', 'number'), ('c', 'string')]}
+
+  This flexibility in data entry enables you to build and manipulate your data
+  in a Python structure that makes sense for your program.
+
+  Add data to the table using the same nested design as the table's
+  table_description, replacing column descriptor tuples with cell data, and
+  each row is an element in the top level collection. This will be a bit
+  clearer after you look at the following examples showing the
+  table_description, matching data, and the resulting table:
+
+  Columns as list of tuples [col1, col2, col3]
+    table_description: [('a', 'number'), ('b', 'string')]
+    AppendData( [[1, 'z'], [2, 'w'], [4, 'o'], [5, 'k']] )
+    Table:
+    a  b   <--- these are column ids/labels
+    1  z
+    2  w
+    4  o
+    5  k
+
+  Dictionary of columns, where key is a column, and value is a list of
+  columns  {col1: [col2, col3]}
+    table_description: {('a', 'number'): [('b', 'number'), ('c', 'string')]}
+    AppendData( data: {1: [2, 'z'], 3: [4, 'w']}
+    Table:
+    a  b  c
+    1  2  z
+    3  4  w
+
+  Dictionary where key is a column, and the value is itself a dictionary of
+  columns {col1: {col2, col3}}
+    table_description: {('a', 'number'): {'b': 'number', 'c': 'string'}}
+    AppendData( data: {1: {'b': 2, 'c': 'z'}, 3: {'b': 4, 'c': 'w'}}
+    Table:
+    a  b  c
+    1  2  z
+    3  4  w
+  """
+
+  def __init__(self, table_description, data=None, custom_properties=None):
+    """Initialize the data table from a table schema and (optionally) data.
+
+    See the class documentation for more information on table schema and data
+    values.
+
+    Args:
+      table_description: A table schema, following one of the formats described
+                         in TableDescriptionParser(). Schemas describe the
+                         column names, data types, and labels. See
+                         TableDescriptionParser() for acceptable formats.
+      data: Optional. If given, fills the table with the given data. The data
+            structure must be consistent with schema in table_description. See
+            the class documentation for more information on acceptable data. You
+            can add data later by calling AppendData().
+      custom_properties: Optional. A dictionary from string to string that
+                         goes into the table's custom properties. This can be
+                         later changed by changing self.custom_properties.
+
+    Raises:
+      DataTableException: Raised if the data and the description did not match,
+                          or did not use the supported formats.
+    """
+    self.__columns = self.TableDescriptionParser(table_description)
+    self.__data = []
+    self.custom_properties = {}
+    if custom_properties is not None:
+      self.custom_properties = custom_properties
+    if data:
+      self.LoadData(data)
+
+  @staticmethod
+  def CoerceValue(value, value_type):
+    """Coerces a single value into the type expected for its column.
+
+    Internal helper method.
+
+    Args:
+      value: The value which should be converted
+      value_type: One of "string", "number", "boolean", "date", "datetime" or
+                  "timeofday".
+
+    Returns:
+      An item of the Python type appropriate to the given value_type. Strings
+      are also converted to Unicode using UTF-8 encoding if necessary.
+      If a tuple is given, it should be in one of the following forms:
+        - (value, formatted value)
+        - (value, formatted value, custom properties)
+      where the formatted value is a string, and custom properties is a
+      dictionary of the custom properties for this cell.
+      To specify custom properties without specifying formatted value, one can
+      pass None as the formatted value.
+      One can also have a null-valued cell with formatted value and/or custom
+      properties by specifying None for the value.
+      This method ignores the custom properties except for checking that it is a
+      dictionary. The custom properties are handled in the ToJSon and ToJSCode
+      methods.
+      The real type of the given value is not strictly checked. For example,
+      any type can be used for string - as we simply take its str( ) and for
+      boolean value we just check "if value".
+      Examples:
+        CoerceValue(None, "string") returns None
+        CoerceValue((5, "5$"), "number") returns (5, "5$")
+        CoerceValue(100, "string") returns "100"
+        CoerceValue(0, "boolean") returns False
+
+    Raises:
+      DataTableException: The value and type did not match in a not-recoverable
+                          way, for example given value 'abc' for type 'number'.
+    """
+    if isinstance(value, tuple):
+      # In case of a tuple, we run the same function on the value itself and
+      # add the formatted value.
+      if (len(value) not in [2, 3] or
+          (len(value) == 3 and not isinstance(value[2], dict))):
+        raise DataTableException("Wrong format for value and formatting - %s." %
+                                 str(value))
+      if not isinstance(value[1], types.StringTypes + (types.NoneType,)):
+        raise DataTableException("Formatted value is not string, given %s." %
+                                 type(value[1]))
+      js_value = DataTable.CoerceValue(value[0], value_type)
+      return (js_value,) + value[1:]
+
+    t_value = type(value)
+    if value is None:
+      return value
+    if value_type == "boolean":
+      return bool(value)
+
+    elif value_type == "number":
+      if isinstance(value, (int, long, float)):
+        return value
+      raise DataTableException("Wrong type %s when expected number" % t_value)
+
+    elif value_type == "string":
+      if isinstance(value, unicode):
+        return value
+      else:
+        return str(value).decode("utf-8")
+
+    elif value_type == "date":
+      if isinstance(value, datetime.datetime):
+        return datetime.date(value.year, value.month, value.day)
+      elif isinstance(value, datetime.date):
+        return value
+      else:
+        raise DataTableException("Wrong type %s when expected date" % t_value)
+
+    elif value_type == "timeofday":
+      if isinstance(value, datetime.datetime):
+        return datetime.time(value.hour, value.minute, value.second)
+      elif isinstance(value, datetime.time):
+        return value
+      else:
+        raise DataTableException("Wrong type %s when expected time" % t_value)
+
+    elif value_type == "datetime":
+      if isinstance(value, datetime.datetime):
+        return value
+      else:
+        raise DataTableException("Wrong type %s when expected datetime" %
+                                 t_value)
+    # If we got here, it means the given value_type was not one of the
+    # supported types.
+    raise DataTableException("Unsupported type %s" % value_type)
+
+  @staticmethod
+  def EscapeForJSCode(encoder, value):
+    if value is None:
+      return "null"
+    elif isinstance(value, datetime.datetime):
+      if value.microsecond == 0:
+        # If it's not ms-resolution, leave that out to save space.
+        return "new Date(%d,%d,%d,%d,%d,%d)" % (value.year,
+                                                value.month - 1,  # To match JS
+                                                value.day,
+                                                value.hour,
+                                                value.minute,
+                                                value.second)
+      else:
+        return "new Date(%d,%d,%d,%d,%d,%d,%d)" % (value.year,
+                                                   value.month - 1,  # match JS
+                                                   value.day,
+                                                   value.hour,
+                                                   value.minute,
+                                                   value.second,
+                                                   value.microsecond / 1000)
+    elif isinstance(value, datetime.date):
+      return "new Date(%d,%d,%d)" % (value.year, value.month - 1, value.day)
+    else:
+      return encoder.encode(value)
+
+  @staticmethod
+  def ToString(value):
+    if value is None:
+      return "(empty)"
+    elif isinstance(value, (datetime.datetime,
+                            datetime.date,
+                            datetime.time)):
+      return str(value)
+    elif isinstance(value, unicode):
+      return value
+    elif isinstance(value, bool):
+      return str(value).lower()
+    else:
+      return str(value).decode("utf-8")
+
+  @staticmethod
+  def ColumnTypeParser(description):
+    """Parses a single column description. Internal helper method.
+
+    Args:
+      description: a column description in the possible formats:
+       'id'
+       ('id',)
+       ('id', 'type')
+       ('id', 'type', 'label')
+       ('id', 'type', 'label', {'custom_prop1': 'custom_val1'})
+    Returns:
+      Dictionary with the following keys: id, label, type, and
+      custom_properties where:
+        - If label not given, it equals the id.
+        - If type not given, string is used by default.
+        - If custom properties are not given, an empty dictionary is used by
+          default.
+
+    Raises:
+      DataTableException: The column description did not match the RE, or
+          unsupported type was passed.
+    """
+    if not description:
+      raise DataTableException("Description error: empty description given")
+
+    if not isinstance(description, (types.StringTypes, tuple)):
+      raise DataTableException("Description error: expected either string or "
+                               "tuple, got %s." % type(description))
+
+    if isinstance(description, types.StringTypes):
+      description = (description,)
+
+    # According to the tuple's length, we fill the keys
+    # We verify everything is of type string
+    for elem in description[:3]:
+      if not isinstance(elem, types.StringTypes):
+        raise DataTableException("Description error: expected tuple of "
+                                 "strings, current element of type %s." %
+                                 type(elem))
+    desc_dict = {"id": description[0],
+                 "label": description[0],
+                 "type": "string",
+                 "custom_properties": {}}
+    if len(description) > 1:
+      desc_dict["type"] = description[1].lower()
+      if len(description) > 2:
+        desc_dict["label"] = description[2]
+        if len(description) > 3:
+          if not isinstance(description[3], dict):
+            raise DataTableException("Description error: expected custom "
+                                     "properties of type dict, current element "
+                                     "of type %s." % type(description[3]))
+          desc_dict["custom_properties"] = description[3]
+          if len(description) > 4:
+            raise DataTableException("Description error: tuple of length > 4")
+    if desc_dict["type"] not in ["string", "number", "boolean",
+                                 "date", "datetime", "timeofday"]:
+      raise DataTableException(
+          "Description error: unsupported type '%s'" % desc_dict["type"])
+    return desc_dict
+
+  @staticmethod
+  def TableDescriptionParser(table_description, depth=0):
+    """Parses the table_description object for internal use.
+
+    Parses the user-submitted table description into an internal format used
+    by the Python DataTable class. Returns the flat list of parsed columns.
+
+    Args:
+      table_description: A description of the table which should comply
+                         with one of the formats described below.
+      depth: Optional. The depth of the first level in the current description.
+             Used by recursive calls to this function.
+
+    Returns:
+      List of columns, where each column represented by a dictionary with the
+      keys: id, label, type, depth, container which means the following:
+      - id: the id of the column
+      - name: The name of the column
+      - type: The datatype of the elements in this column. Allowed types are
+              described in ColumnTypeParser().
+      - depth: The depth of this column in the table description
+      - container: 'dict', 'iter' or 'scalar' for parsing the format easily.
+      - custom_properties: The custom properties for this column.
+      The returned description is flattened regardless of how it was given.
+
+    Raises:
+      DataTableException: Error in a column description or in the description
+                          structure.
+
+    Examples:
+      A column description can be of the following forms:
+       'id'
+       ('id',)
+       ('id', 'type')
+       ('id', 'type', 'label')
+       ('id', 'type', 'label', {'custom_prop1': 'custom_val1'})
+       or as a dictionary:
+       'id': 'type'
+       'id': ('type',)
+       'id': ('type', 'label')
+       'id': ('type', 'label', {'custom_prop1': 'custom_val1'})
+      If the type is not specified, we treat it as string.
+      If no specific label is given, the label is simply the id.
+      If no custom properties are given, we use an empty dictionary.
+
+      input: [('a', 'date'), ('b', 'timeofday', 'b', {'foo': 'bar'})]
+      output: [{'id': 'a', 'label': 'a', 'type': 'date',
+                'depth': 0, 'container': 'iter', 'custom_properties': {}},
+               {'id': 'b', 'label': 'b', 'type': 'timeofday',
+                'depth': 0, 'container': 'iter',
+                'custom_properties': {'foo': 'bar'}}]
+
+      input: {'a': [('b', 'number'), ('c', 'string', 'column c')]}
+      output: [{'id': 'a', 'label': 'a', 'type': 'string',
+                'depth': 0, 'container': 'dict', 'custom_properties': {}},
+               {'id': 'b', 'label': 'b', 'type': 'number',
+                'depth': 1, 'container': 'iter', 'custom_properties': {}},
+               {'id': 'c', 'label': 'column c', 'type': 'string',
+                'depth': 1, 'container': 'iter', 'custom_properties': {}}]
+
+      input:  {('a', 'number', 'column a'): { 'b': 'number', 'c': 'string'}}
+      output: [{'id': 'a', 'label': 'column a', 'type': 'number',
+                'depth': 0, 'container': 'dict', 'custom_properties': {}},
+               {'id': 'b', 'label': 'b', 'type': 'number',
+                'depth': 1, 'container': 'dict', 'custom_properties': {}},
+               {'id': 'c', 'label': 'c', 'type': 'string',
+                'depth': 1, 'container': 'dict', 'custom_properties': {}}]
+
+      input: { ('w', 'string', 'word'): ('c', 'number', 'count') }
+      output: [{'id': 'w', 'label': 'word', 'type': 'string',
+                'depth': 0, 'container': 'dict', 'custom_properties': {}},
+               {'id': 'c', 'label': 'count', 'type': 'number',
+                'depth': 1, 'container': 'scalar', 'custom_properties': {}}]
+
+      input: {'a': ('number', 'column a'), 'b': ('string', 'column b')}
+      output: [{'id': 'a', 'label': 'column a', 'type': 'number', 'depth': 0,
+               'container': 'dict', 'custom_properties': {}},
+               {'id': 'b', 'label': 'column b', 'type': 'string', 'depth': 0,
+               'container': 'dict', 'custom_properties': {}}
+
+      NOTE: there might be ambiguity in the case of a dictionary representation
+      of a single column. For example, the following description can be parsed
+      in 2 different ways: {'a': ('b', 'c')} can be thought of a single column
+      with the id 'a', of type 'b' and the label 'c', or as 2 columns: one named
+      'a', and the other named 'b' of type 'c'. We choose the first option by
+      default, and in case the second option is the right one, it is possible to
+      make the key into a tuple (i.e. {('a',): ('b', 'c')}) or add more info
+      into the tuple, thus making it look like this: {'a': ('b', 'c', 'b', {})}
+      -- second 'b' is the label, and {} is the custom properties field.
+    """
+    # For the recursion step, we check for a scalar object (string or tuple)
+    if isinstance(table_description, (types.StringTypes, tuple)):
+      parsed_col = DataTable.ColumnTypeParser(table_description)
+      parsed_col["depth"] = depth
+      parsed_col["container"] = "scalar"
+      return [parsed_col]
+
+    # Since it is not scalar, table_description must be iterable.
+    if not hasattr(table_description, "__iter__"):
+      raise DataTableException("Expected an iterable object, got %s" %
+                               type(table_description))
+    if not isinstance(table_description, dict):
+      # We expects a non-dictionary iterable item.
+      columns = []
+      for desc in table_description:
+        parsed_col = DataTable.ColumnTypeParser(desc)
+        parsed_col["depth"] = depth
+        parsed_col["container"] = "iter"
+        columns.append(parsed_col)
+      if not columns:
+        raise DataTableException("Description iterable objects should not"
+                                 " be empty.")
+      return columns
+    # The other case is a dictionary
+    if not table_description:
+      raise DataTableException("Empty dictionaries are not allowed inside"
+                               " description")
+
+    # To differentiate between the two cases of more levels below or this is
+    # the most inner dictionary, we consider the number of keys (more then one
+    # key is indication for most inner dictionary) and the type of the key and
+    # value in case of only 1 key (if the type of key is string and the type of
+    # the value is a tuple of 0-3 items, we assume this is the most inner
+    # dictionary).
+    # NOTE: this way of differentiating might create ambiguity. See docs.
+    if (len(table_description) != 1 or
+        (isinstance(table_description.keys()[0], types.StringTypes) and
+         isinstance(table_description.values()[0], tuple) and
+         len(table_description.values()[0]) < 4)):
+      # This is the most inner dictionary. Parsing types.
+      columns = []
+      # We sort the items, equivalent to sort the keys since they are unique
+      for key, value in sorted(table_description.items()):
+        # We parse the column type as (key, type) or (key, type, label) using
+        # ColumnTypeParser.
+        if isinstance(value, tuple):
+          parsed_col = DataTable.ColumnTypeParser((key,) + value)
+        else:
+          parsed_col = DataTable.ColumnTypeParser((key, value))
+        parsed_col["depth"] = depth
+        parsed_col["container"] = "dict"
+        columns.append(parsed_col)
+      return columns
+    # This is an outer dictionary, must have at most one key.
+    parsed_col = DataTable.ColumnTypeParser(table_description.keys()[0])
+    parsed_col["depth"] = depth
+    parsed_col["container"] = "dict"
+    return ([parsed_col] +
+            DataTable.TableDescriptionParser(table_description.values()[0],
+                                             depth=depth + 1))
+
+  @property
+  def columns(self):
+    """Returns the parsed table description."""
+    return self.__columns
+
+  def NumberOfRows(self):
+    """Returns the number of rows in the current data stored in the table."""
+    return len(self.__data)
+
+  def SetRowsCustomProperties(self, rows, custom_properties):
+    """Sets the custom properties for given row(s).
+
+    Can accept a single row or an iterable of rows.
+    Sets the given custom properties for all specified rows.
+
+    Args:
+      rows: The row, or rows, to set the custom properties for.
+      custom_properties: A string to string dictionary of custom properties to
+      set for all rows.
+    """
+    if not hasattr(rows, "__iter__"):
+      rows = [rows]
+    for row in rows:
+      self.__data[row] = (self.__data[row][0], custom_properties)
+
+  def LoadData(self, data, custom_properties=None):
+    """Loads new rows to the data table, clearing existing rows.
+
+    May also set the custom_properties for the added rows. The given custom
+    properties dictionary specifies the dictionary that will be used for *all*
+    given rows.
+
+    Args:
+      data: The rows that the table will contain.
+      custom_properties: A dictionary of string to string to set as the custom
+                         properties for all rows.
+    """
+    self.__data = []
+    self.AppendData(data, custom_properties)
+
+  def AppendData(self, data, custom_properties=None):
+    """Appends new data to the table.
+
+    Data is appended in rows. Data must comply with
+    the table schema passed in to __init__(). See CoerceValue() for a list
+    of acceptable data types. See the class documentation for more information
+    and examples of schema and data values.
+
+    Args:
+      data: The row to add to the table. The data must conform to the table
+            description format.
+      custom_properties: A dictionary of string to string, representing the
+                         custom properties to add to all the rows.
+
+    Raises:
+      DataTableException: The data structure does not match the description.
+    """
+    # If the maximal depth is 0, we simply iterate over the data table
+    # lines and insert them using _InnerAppendData. Otherwise, we simply
+    # let the _InnerAppendData handle all the levels.
+    if not self.__columns[-1]["depth"]:
+      for row in data:
+        self._InnerAppendData(({}, custom_properties), row, 0)
+    else:
+      self._InnerAppendData(({}, custom_properties), data, 0)
+
+  def _InnerAppendData(self, prev_col_values, data, col_index):
+    """Inner function to assist LoadData."""
+    # We first check that col_index has not exceeded the columns size
+    if col_index >= len(self.__columns):
+      raise DataTableException("The data does not match description, too deep")
+
+    # Dealing with the scalar case, the data is the last value.
+    if self.__columns[col_index]["container"] == "scalar":
+      prev_col_values[0][self.__columns[col_index]["id"]] = data
+      self.__data.append(prev_col_values)
+      return
+
+    if self.__columns[col_index]["container"] == "iter":
+      if not hasattr(data, "__iter__") or isinstance(data, dict):
+        raise DataTableException("Expected iterable object, got %s" %
+                                 type(data))
+      # We only need to insert the rest of the columns
+      # If there are less items than expected, we only add what there is.
+      for value in data:
+        if col_index >= len(self.__columns):
+          raise DataTableException("Too many elements given in data")
+        prev_col_values[0][self.__columns[col_index]["id"]] = value
+        col_index += 1
+      self.__data.append(prev_col_values)
+      return
+
+    # We know the current level is a dictionary, we verify the type.
+    if not isinstance(data, dict):
+      raise DataTableException("Expected dictionary at current level, got %s" %
+                               type(data))
+    # We check if this is the last level
+    if self.__columns[col_index]["depth"] == self.__columns[-1]["depth"]:
+      # We need to add the keys in the dictionary as they are
+      for col in self.__columns[col_index:]:
+        if col["id"] in data:
+          prev_col_values[0][col["id"]] = data[col["id"]]
+      self.__data.append(prev_col_values)
+      return
+
+    # We have a dictionary in an inner depth level.
+    if not data.keys():
+      # In case this is an empty dictionary, we add a record with the columns
+      # filled only until this point.
+      self.__data.append(prev_col_values)
+    else:
+      for key in sorted(data):
+        col_values = dict(prev_col_values[0])
+        col_values[self.__columns[col_index]["id"]] = key
+        self._InnerAppendData((col_values, prev_col_values[1]),
+                              data[key], col_index + 1)
+
+  def _PreparedData(self, order_by=()):
+    """Prepares the data for enumeration - sorting it by order_by.
+
+    Args:
+      order_by: Optional. Specifies the name of the column(s) to sort by, and
+                (optionally) which direction to sort in. Default sort direction
+                is asc. Following formats are accepted:
+                "string_col_name"  -- For a single key in default (asc) order.
+                ("string_col_name", "asc|desc") -- For a single key.
+                [("col_1","asc|desc"), ("col_2","asc|desc")] -- For more than
+                    one column, an array of tuples of (col_name, "asc|desc").
+
+    Returns:
+      The data sorted by the keys given.
+
+    Raises:
+      DataTableException: Sort direction not in 'asc' or 'desc'
+    """
+    if not order_by:
+      return self.__data
+
+    proper_sort_keys = []
+    if isinstance(order_by, types.StringTypes) or (
+        isinstance(order_by, tuple) and len(order_by) == 2 and
+        order_by[1].lower() in ["asc", "desc"]):
+      order_by = (order_by,)
+    for key in order_by:
+      if isinstance(key, types.StringTypes):
+        proper_sort_keys.append((key, 1))
+      elif (isinstance(key, (list, tuple)) and len(key) == 2 and
+            key[1].lower() in ("asc", "desc")):
+        proper_sort_keys.append((key[0], key[1].lower() == "asc" and 1 or -1))
+      else:
+        raise DataTableException("Expected tuple with second value: "
+                                 "'asc' or 'desc'")
+
+    def SortCmpFunc(row1, row2):
+      """cmp function for sorted. Compares by keys and 'asc'/'desc' keywords."""
+      for key, asc_mult in proper_sort_keys:
+        cmp_result = asc_mult * cmp(row1[0].get(key), row2[0].get(key))
+        if cmp_result:
+          return cmp_result
+      return 0
+
+    return sorted(self.__data, cmp=SortCmpFunc)
+
+  def ToJSCode(self, name, columns_order=None, order_by=()):
+    """Writes the data table as a JS code string.
+
+    This method writes a string of JS code that can be run to
+    generate a DataTable with the specified data. Typically used for debugging
+    only.
+
+    Args:
+      name: The name of the table. The name would be used as the DataTable's
+            variable name in the created JS code.
+      columns_order: Optional. Specifies the order of columns in the
+                     output table. Specify a list of all column IDs in the order
+                     in which you want the table created.
+                     Note that you must list all column IDs in this parameter,
+                     if you use it.
+      order_by: Optional. Specifies the name of the column(s) to sort by.
+                Passed as is to _PreparedData.
+
+    Returns:
+      A string of JS code that, when run, generates a DataTable with the given
+      name and the data stored in the DataTable object.
+      Example result:
+        "var tab1 = new google.visualization.DataTable();
+         tab1.addColumn("string", "a", "a");
+         tab1.addColumn("number", "b", "b");
+         tab1.addColumn("boolean", "c", "c");
+         tab1.addRows(10);
+         tab1.setCell(0, 0, "a");
+         tab1.setCell(0, 1, 1, null, {"foo": "bar"});
+         tab1.setCell(0, 2, true);
+         ...
+         tab1.setCell(9, 0, "c");
+         tab1.setCell(9, 1, 3, "3$");
+         tab1.setCell(9, 2, false);"
+
+    Raises:
+      DataTableException: The data does not match the type.
+    """
+
+    encoder = DataTableJSONEncoder()
+
+    if columns_order is None:
+      columns_order = [col["id"] for col in self.__columns]
+    col_dict = dict([(col["id"], col) for col in self.__columns])
+
+    # We first create the table with the given name
+    jscode = "var %s = new google.visualization.DataTable();\n" % name
+    if self.custom_properties:
+      jscode += "%s.setTableProperties(%s);\n" % (
+          name, encoder.encode(self.custom_properties))
+
+    # We add the columns to the table
+    for i, col in enumerate(columns_order):
+      jscode += "%s.addColumn(%s, %s, %s);\n" % (
+          name,
+          encoder.encode(col_dict[col]["type"]),
+          encoder.encode(col_dict[col]["label"]),
+          encoder.encode(col_dict[col]["id"]))
+      if col_dict[col]["custom_properties"]:
+        jscode += "%s.setColumnProperties(%d, %s);\n" % (
+            name, i, encoder.encode(col_dict[col]["custom_properties"]))
+    jscode += "%s.addRows(%d);\n" % (name, len(self.__data))
+
+    # We now go over the data and add each row
+    for (i, (row, cp)) in enumerate(self._PreparedData(order_by)):
+      # We add all the elements of this row by their order
+      for (j, col) in enumerate(columns_order):
+        if col not in row or row[col] is None:
+          continue
+        value = self.CoerceValue(row[col], col_dict[col]["type"])
+        if isinstance(value, tuple):
+          cell_cp = ""
+          if len(value) == 3:
+            cell_cp = ", %s" % encoder.encode(row[col][2])
+          # We have a formatted value or custom property as well
+          jscode += ("%s.setCell(%d, %d, %s, %s%s);\n" %
+                     (name, i, j,
+                      self.EscapeForJSCode(encoder, value[0]),
+                      self.EscapeForJSCode(encoder, value[1]), cell_cp))
+        else:
+          jscode += "%s.setCell(%d, %d, %s);\n" % (
+              name, i, j, self.EscapeForJSCode(encoder, value))
+      if cp:
+        jscode += "%s.setRowProperties(%d, %s);\n" % (
+            name, i, encoder.encode(cp))
+    return jscode
+
+  def ToHtml(self, columns_order=None, order_by=()):
+    """Writes the data table as an HTML table code string.
+
+    Args:
+      columns_order: Optional. Specifies the order of columns in the
+                     output table. Specify a list of all column IDs in the order
+                     in which you want the table created.
+                     Note that you must list all column IDs in this parameter,
+                     if you use it.
+      order_by: Optional. Specifies the name of the column(s) to sort by.
+                Passed as is to _PreparedData.
+
+    Returns:
+      An HTML table code string.
+      Example result (the result is without the newlines):
+       <html><body><table border="1">
+        <thead><tr><th>a</th><th>b</th><th>c</th></tr></thead>
+        <tbody>
+         <tr><td>1</td><td>"z"</td><td>2</td></tr>
+         <tr><td>"3$"</td><td>"w"</td><td></td></tr>
+        </tbody>
+       </table></body></html>
+
+    Raises:
+      DataTableException: The data does not match the type.
+    """
+    table_template = "<html><body><table border=\"1\">%s</table></body></html>"
+    columns_template = "<thead><tr>%s</tr></thead>"
+    rows_template = "<tbody>%s</tbody>"
+    row_template = "<tr>%s</tr>"
+    header_cell_template = "<th>%s</th>"
+    cell_template = "<td>%s</td>"
+
+    if columns_order is None:
+      columns_order = [col["id"] for col in self.__columns]
+    col_dict = dict([(col["id"], col) for col in self.__columns])
+
+    columns_list = []
+    for col in columns_order:
+      columns_list.append(header_cell_template %
+                          cgi.escape(col_dict[col]["label"]))
+    columns_html = columns_template % "".join(columns_list)
+
+    rows_list = []
+    # We now go over the data and add each row
+    for row, unused_cp in self._PreparedData(order_by):
+      cells_list = []
+      # We add all the elements of this row by their order
+      for col in columns_order:
+        # For empty string we want empty quotes ("").
+        value = ""
+        if col in row and row[col] is not None:
+          value = self.CoerceValue(row[col], col_dict[col]["type"])
+        if isinstance(value, tuple):
+          # We have a formatted value and we're going to use it
+          cells_list.append(cell_template % cgi.escape(self.ToString(value[1])))
+        else:
+          cells_list.append(cell_template % cgi.escape(self.ToString(value)))
+      rows_list.append(row_template % "".join(cells_list))
+    rows_html = rows_template % "".join(rows_list)
+
+    return table_template % (columns_html + rows_html)
+
+  def ToCsv(self, columns_order=None, order_by=(), separator=","):
+    """Writes the data table as a CSV string.
+
+    Output is encoded in UTF-8 because the Python "csv" module can't handle
+    Unicode properly according to its documentation.
+
+    Args:
+      columns_order: Optional. Specifies the order of columns in the
+                     output table. Specify a list of all column IDs in the order
+                     in which you want the table created.
+                     Note that you must list all column IDs in this parameter,
+                     if you use it.
+      order_by: Optional. Specifies the name of the column(s) to sort by.
+                Passed as is to _PreparedData.
+      separator: Optional. The separator to use between the values.
+
+    Returns:
+      A CSV string representing the table.
+      Example result:
+       'a','b','c'
+       1,'z',2
+       3,'w',''
+
+    Raises:
+      DataTableException: The data does not match the type.
+    """
+
+    csv_buffer = cStringIO.StringIO()
+    writer = csv.writer(csv_buffer, delimiter=separator)
+
+    if columns_order is None:
+      columns_order = [col["id"] for col in self.__columns]
+    col_dict = dict([(col["id"], col) for col in self.__columns])
+
+    writer.writerow([col_dict[col]["label"].encode("utf-8")
+                     for col in columns_order])
+
+    # We now go over the data and add each row
+    for row, unused_cp in self._PreparedData(order_by):
+      cells_list = []
+      # We add all the elements of this row by their order
+      for col in columns_order:
+        value = ""
+        if col in row and row[col] is not None:
+          value = self.CoerceValue(row[col], col_dict[col]["type"])
+        if isinstance(value, tuple):
+          # We have a formatted value. Using it only for date/time types.
+          if col_dict[col]["type"] in ["date", "datetime", "timeofday"]:
+            cells_list.append(self.ToString(value[1]).encode("utf-8"))
+          else:
+            cells_list.append(self.ToString(value[0]).encode("utf-8"))
+        else:
+          cells_list.append(self.ToString(value).encode("utf-8"))
+      writer.writerow(cells_list)
+    return csv_buffer.getvalue()
+
+  def ToTsvExcel(self, columns_order=None, order_by=()):
+    """Returns a file in tab-separated-format readable by MS Excel.
+
+    Returns a file in UTF-16 little endian encoding, with tabs separating the
+    values.
+
+    Args:
+      columns_order: Delegated to ToCsv.
+      order_by: Delegated to ToCsv.
+
+    Returns:
+      A tab-separated little endian UTF16 file representing the table.
+    """
+    return (self.ToCsv(columns_order, order_by, separator="\t")
+            .decode("utf-8").encode("UTF-16LE"))
+
+  def _ToJSonObj(self, columns_order=None, order_by=()):
+    """Returns an object suitable to be converted to JSON.
+
+    Args:
+      columns_order: Optional. A list of all column IDs in the order in which
+                     you want them created in the output table. If specified,
+                     all column IDs must be present.
+      order_by: Optional. Specifies the name of the column(s) to sort by.
+                Passed as is to _PreparedData().
+
+    Returns:
+      A dictionary object for use by ToJSon or ToJSonResponse.
+    """
+    if columns_order is None:
+      columns_order = [col["id"] for col in self.__columns]
+    col_dict = dict([(col["id"], col) for col in self.__columns])
+
+    # Creating the column JSON objects
+    col_objs = []
+    for col_id in columns_order:
+      col_obj = {"id": col_dict[col_id]["id"],
+                 "label": col_dict[col_id]["label"],
+                 "type": col_dict[col_id]["type"]}
+      if col_dict[col_id]["custom_properties"]:
+        col_obj["p"] = col_dict[col_id]["custom_properties"]
+      col_objs.append(col_obj)
+
+    # Creating the rows jsons
+    row_objs = []
+    for row, cp in self._PreparedData(order_by):
+      cell_objs = []
+      for col in columns_order:
+        value = self.CoerceValue(row.get(col, None), col_dict[col]["type"])
+        if value is None:
+          cell_obj = None
+        elif isinstance(value, tuple):
+          cell_obj = {"v": value[0]}
+          if len(value) > 1 and value[1] is not None:
+            cell_obj["f"] = value[1]
+          if len(value) == 3:
+            cell_obj["p"] = value[2]
+        else:
+          cell_obj = {"v": value}
+        cell_objs.append(cell_obj)
+      row_obj = {"c": cell_objs}
+      if cp:
+        row_obj["p"] = cp
+      row_objs.append(row_obj)
+
+    json_obj = {"cols": col_objs, "rows": row_objs}
+    if self.custom_properties:
+      json_obj["p"] = self.custom_properties
+
+    return json_obj
+
+  def ToJSon(self, columns_order=None, order_by=()):
+    """Returns a string that can be used in a JS DataTable constructor.
+
+    This method writes a JSON string that can be passed directly into a Google
+    Visualization API DataTable constructor. Use this output if you are
+    hosting the visualization HTML on your site, and want to code the data
+    table in Python. Pass this string into the
+    google.visualization.DataTable constructor, e.g,:
+      ... on my page that hosts my visualization ...
+      google.setOnLoadCallback(drawTable);
+      function drawTable() {
+        var data = new google.visualization.DataTable(_my_JSon_string, 0.6);
+        myTable.draw(data);
+      }
+
+    Args:
+      columns_order: Optional. Specifies the order of columns in the
+                     output table. Specify a list of all column IDs in the order
+                     in which you want the table created.
+                     Note that you must list all column IDs in this parameter,
+                     if you use it.
+      order_by: Optional. Specifies the name of the column(s) to sort by.
+                Passed as is to _PreparedData().
+
+    Returns:
+      A JSon constructor string to generate a JS DataTable with the data
+      stored in the DataTable object.
+      Example result (the result is without the newlines):
+       {cols: [{id:"a",label:"a",type:"number"},
+               {id:"b",label:"b",type:"string"},
+              {id:"c",label:"c",type:"number"}],
+        rows: [{c:[{v:1},{v:"z"},{v:2}]}, c:{[{v:3,f:"3$"},{v:"w"},{v:null}]}],
+        p:    {'foo': 'bar'}}
+
+    Raises:
+      DataTableException: The data does not match the type.
+    """
+
+    encoder = DataTableJSONEncoder()
+    return encoder.encode(
+        self._ToJSonObj(columns_order, order_by)).encode("utf-8")
+
+  def ToJSonResponse(self, columns_order=None, order_by=(), req_id=0,
+                     response_handler="google.visualization.Query.setResponse"):
+    """Writes a table as a JSON response that can be returned as-is to a client.
+
+    This method writes a JSON response to return to a client in response to a
+    Google Visualization API query. This string can be processed by the calling
+    page, and is used to deliver a data table to a visualization hosted on
+    a different page.
+
+    Args:
+      columns_order: Optional. Passed straight to self.ToJSon().
+      order_by: Optional. Passed straight to self.ToJSon().
+      req_id: Optional. The response id, as retrieved by the request.
+      response_handler: Optional. The response handler, as retrieved by the
+          request.
+
+    Returns:
+      A JSON response string to be received by JS the visualization Query
+      object. This response would be translated into a DataTable on the
+      client side.
+      Example result (newlines added for readability):
+       google.visualization.Query.setResponse({
+          'version':'0.6', 'reqId':'0', 'status':'OK',
+          'table': {cols: [...], rows: [...]}});
+
+    Note: The URL returning this string can be used as a data source by Google
+          Visualization Gadgets or from JS code.
+    """
+
+    response_obj = {
+        "version": "0.6",
+        "reqId": str(req_id),
+        "table": self._ToJSonObj(columns_order, order_by),
+        "status": "ok"
+    }
+    encoder = DataTableJSONEncoder()
+    return "%s(%s);" % (response_handler,
+                        encoder.encode(response_obj).encode("utf-8"))
+
+  def ToResponse(self, columns_order=None, order_by=(), tqx=""):
+    """Writes the right response according to the request string passed in tqx.
+
+    This method parses the tqx request string (format of which is defined in
+    the documentation for implementing a data source of Google Visualization),
+    and returns the right response according to the request.
+    It parses out the "out" parameter of tqx, calls the relevant response
+    (ToJSonResponse() for "json", ToCsv() for "csv", ToHtml() for "html",
+    ToTsvExcel() for "tsv-excel") and passes the response function the rest of
+    the relevant request keys.
+
+    Args:
+      columns_order: Optional. Passed as is to the relevant response function.
+      order_by: Optional. Passed as is to the relevant response function.
+      tqx: Optional. The request string as received by HTTP GET. Should be in
+           the format "key1:value1;key2:value2...". All keys have a default
+           value, so an empty string will just do the default (which is calling
+           ToJSonResponse() with no extra parameters).
+
+    Returns:
+      A response string, as returned by the relevant response function.
+
+    Raises:
+      DataTableException: One of the parameters passed in tqx is not supported.
+    """
+    tqx_dict = {}
+    if tqx:
+      tqx_dict = dict(opt.split(":") for opt in tqx.split(";"))
+    if tqx_dict.get("version", "0.6") != "0.6":
+      raise DataTableException(
+          "Version (%s) passed by request is not supported."
+          % tqx_dict["version"])
+
+    if tqx_dict.get("out", "json") == "json":
+      response_handler = tqx_dict.get("responseHandler",
+                                      "google.visualization.Query.setResponse")
+      return self.ToJSonResponse(columns_order, order_by,
+                                 req_id=tqx_dict.get("reqId", 0),
+                                 response_handler=response_handler)
+    elif tqx_dict["out"] == "html":
+      return self.ToHtml(columns_order, order_by)
+    elif tqx_dict["out"] == "csv":
+      return self.ToCsv(columns_order, order_by)
+    elif tqx_dict["out"] == "tsv-excel":
+      return self.ToTsvExcel(columns_order, order_by)
+    else:
+      raise DataTableException(
+          "'out' parameter: '%s' is not supported" % tqx_dict["out"])
diff --git a/third_party/aom/test/hash_test.cc b/third_party/aom/test/hash_test.cc
new file mode 100644
index 0000000000..e9f7f63c9b
--- /dev/null
+++ b/third_party/aom/test/hash_test.cc
@@ -0,0 +1,133 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cstdlib>
+#include <new>
+
+#include "config/aom_config.h"
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/aom_timer.h"
+#include "av1/encoder/hash.h"
+#include "test/acm_random.h"
+#include "test/util.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+
+typedef uint32_t (*get_crc32c_value_func)(void *calculator, uint8_t *p,
+                                          int length);
+
+typedef ::testing::tuple<get_crc32c_value_func, int> HashParam;
+
+class AV1Crc32cHashTest : public ::testing::TestWithParam<HashParam> {
+ public:
+  ~AV1Crc32cHashTest();
+  void SetUp();
+
+  void TearDown();
+
+ protected:
+  void RunCheckOutput(get_crc32c_value_func test_impl);
+  void RunSpeedTest(get_crc32c_value_func test_impl);
+
+  void RunZeroTest(get_crc32c_value_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+  CRC32C calc_;
+  uint8_t *buffer_;
+  int bsize_;
+  int length_;
+};
+
+AV1Crc32cHashTest::~AV1Crc32cHashTest() { ; }
+
+void AV1Crc32cHashTest::SetUp() {
+  rnd_.Reset(libaom_test::ACMRandom::DeterministicSeed());
+  av1_crc32c_calculator_init(&calc_);
+
+  bsize_ = GET_PARAM(1);
+  length_ = bsize_ * bsize_ * sizeof(uint16_t);
+  buffer_ = new uint8_t[length_];
+  ASSERT_TRUE(buffer_ != NULL);
+  for (int i = 0; i < length_; ++i) {
+    buffer_[i] = rnd_.Rand8();
+  }
+}
+
+void AV1Crc32cHashTest::TearDown() { delete[] buffer_; }
+
+void AV1Crc32cHashTest::RunCheckOutput(get_crc32c_value_func test_impl) {
+  get_crc32c_value_func ref_impl = av1_get_crc32c_value_c;
+  // for the same buffer crc should be the same
+  uint32_t crc0 = test_impl(&calc_, buffer_, length_);
+  uint32_t crc1 = test_impl(&calc_, buffer_, length_);
+  uint32_t crc2 = ref_impl(&calc_, buffer_, length_);
+  ASSERT_EQ(crc0, crc1);
+  ASSERT_EQ(crc0, crc2);  // should equal to software version
+  // modify buffer
+  buffer_[0] += 1;
+  uint32_t crc3 = test_impl(&calc_, buffer_, length_);
+  uint32_t crc4 = ref_impl(&calc_, buffer_, length_);
+  ASSERT_NE(crc0, crc3);  // crc shoud not equal to previous one
+  ASSERT_EQ(crc3, crc4);
+}
+
+void AV1Crc32cHashTest::RunSpeedTest(get_crc32c_value_func test_impl) {
+  get_crc32c_value_func impls[] = { av1_get_crc32c_value_c, test_impl };
+  const int repeat = 10000000 / (bsize_ + bsize_);
+
+  aom_usec_timer timer;
+  double time[2];
+  for (int i = 0; i < 2; ++i) {
+    aom_usec_timer_start(&timer);
+    for (int j = 0; j < repeat; ++j) {
+      impls[i](&calc_, buffer_, length_);
+    }
+    aom_usec_timer_mark(&timer);
+    time[i] = static_cast<double>(aom_usec_timer_elapsed(&timer));
+  }
+  printf("hash %3dx%-3d:%7.2f/%7.2fus", bsize_, bsize_, time[0], time[1]);
+  printf("(%3.2f)\n", time[0] / time[1]);
+}
+
+void AV1Crc32cHashTest::RunZeroTest(get_crc32c_value_func test_impl) {
+  uint8_t buffer0[1024] = { 0 };
+  // for buffer with different size the crc should not be the same
+  const uint32_t crc0 = test_impl(&calc_, buffer0, 32);
+  const uint32_t crc1 = test_impl(&calc_, buffer0, 128);
+  const uint32_t crc2 = test_impl(&calc_, buffer0, 1024);
+  ASSERT_NE(crc0, crc1);
+  ASSERT_NE(crc0, crc2);
+  ASSERT_NE(crc1, crc2);
+}
+
+TEST_P(AV1Crc32cHashTest, CheckOutput) { RunCheckOutput(GET_PARAM(0)); }
+
+TEST_P(AV1Crc32cHashTest, CheckZero) { RunZeroTest(GET_PARAM(0)); }
+
+TEST_P(AV1Crc32cHashTest, DISABLED_Speed) { RunSpeedTest(GET_PARAM(0)); }
+
+const int kValidBlockSize[] = { 64, 32, 8, 4 };
+
+INSTANTIATE_TEST_CASE_P(
+    C, AV1Crc32cHashTest,
+    ::testing::Combine(::testing::Values(&av1_get_crc32c_value_c),
+                       ::testing::ValuesIn(kValidBlockSize)));
+
+#if HAVE_SSE4_2
+INSTANTIATE_TEST_CASE_P(
+    SSE4_2, AV1Crc32cHashTest,
+    ::testing::Combine(::testing::Values(&av1_get_crc32c_value_sse4_2),
+                       ::testing::ValuesIn(kValidBlockSize)));
+#endif
+
+}  // namespace
diff --git a/third_party/aom/test/hbd_metrics_test.cc b/third_party/aom/test/hbd_metrics_test.cc
new file mode 100644
index 0000000000..09df9bde47
--- /dev/null
+++ b/third_party/aom/test/hbd_metrics_test.cc
@@ -0,0 +1,239 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <new>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/util.h"
+
+#include "config/aom_config.h"
+
+#include "aom_dsp/psnr.h"
+#include "aom_dsp/ssim.h"
+#include "aom_ports/mem.h"
+#include "aom_ports/msvc.h"
+#include "aom_scale/yv12config.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+
+typedef double (*LBDMetricFunc)(const YV12_BUFFER_CONFIG *source,
+                                const YV12_BUFFER_CONFIG *dest);
+typedef double (*HBDMetricFunc)(const YV12_BUFFER_CONFIG *source,
+                                const YV12_BUFFER_CONFIG *dest, uint32_t in_bd,
+                                uint32_t bd);
+
+double compute_hbd_psnr(const YV12_BUFFER_CONFIG *source,
+                        const YV12_BUFFER_CONFIG *dest, uint32_t in_bd,
+                        uint32_t bd) {
+  PSNR_STATS psnr;
+  aom_calc_highbd_psnr(source, dest, &psnr, bd, in_bd);
+  return psnr.psnr[0];
+}
+
+double compute_psnr(const YV12_BUFFER_CONFIG *source,
+                    const YV12_BUFFER_CONFIG *dest) {
+  PSNR_STATS psnr;
+  aom_calc_psnr(source, dest, &psnr);
+  return psnr.psnr[0];
+}
+
+double compute_hbd_psnrhvs(const YV12_BUFFER_CONFIG *source,
+                           const YV12_BUFFER_CONFIG *dest, uint32_t in_bd,
+                           uint32_t bd) {
+  double tempy, tempu, tempv;
+  return aom_psnrhvs(source, dest, &tempy, &tempu, &tempv, bd, in_bd);
+}
+
+double compute_psnrhvs(const YV12_BUFFER_CONFIG *source,
+                       const YV12_BUFFER_CONFIG *dest) {
+  double tempy, tempu, tempv;
+  return aom_psnrhvs(source, dest, &tempy, &tempu, &tempv, 8, 8);
+}
+
+double compute_hbd_fastssim(const YV12_BUFFER_CONFIG *source,
+                            const YV12_BUFFER_CONFIG *dest, uint32_t in_bd,
+                            uint32_t bd) {
+  double tempy, tempu, tempv;
+  return aom_calc_fastssim(source, dest, &tempy, &tempu, &tempv, bd, in_bd);
+}
+
+double compute_fastssim(const YV12_BUFFER_CONFIG *source,
+                        const YV12_BUFFER_CONFIG *dest) {
+  double tempy, tempu, tempv;
+  return aom_calc_fastssim(source, dest, &tempy, &tempu, &tempv, 8, 8);
+}
+
+double compute_hbd_aomssim(const YV12_BUFFER_CONFIG *source,
+                           const YV12_BUFFER_CONFIG *dest, uint32_t in_bd,
+                           uint32_t bd) {
+  double ssim, weight;
+  ssim = aom_highbd_calc_ssim(source, dest, &weight, bd, in_bd);
+  return 100 * pow(ssim / weight, 8.0);
+}
+
+double compute_aomssim(const YV12_BUFFER_CONFIG *source,
+                       const YV12_BUFFER_CONFIG *dest) {
+  double ssim, weight;
+  ssim = aom_calc_ssim(source, dest, &weight);
+  return 100 * pow(ssim / weight, 8.0);
+}
+
+class HBDMetricsTestBase {
+ public:
+  virtual ~HBDMetricsTestBase() {}
+
+ protected:
+  void RunAccuracyCheck() {
+    const int width = 1920;
+    const int height = 1080;
+    size_t i = 0;
+    const uint8_t kPixFiller = 128;
+    YV12_BUFFER_CONFIG lbd_src, lbd_dst;
+    YV12_BUFFER_CONFIG hbd_src, hbd_dst;
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    double lbd_db, hbd_db;
+
+    memset(&lbd_src, 0, sizeof(lbd_src));
+    memset(&lbd_dst, 0, sizeof(lbd_dst));
+    memset(&hbd_src, 0, sizeof(hbd_src));
+    memset(&hbd_dst, 0, sizeof(hbd_dst));
+
+    aom_alloc_frame_buffer(&lbd_src, width, height, 1, 1, 0, 32, 16);
+    aom_alloc_frame_buffer(&lbd_dst, width, height, 1, 1, 0, 32, 16);
+    aom_alloc_frame_buffer(&hbd_src, width, height, 1, 1, 1, 32, 16);
+    aom_alloc_frame_buffer(&hbd_dst, width, height, 1, 1, 1, 32, 16);
+
+    memset(lbd_src.buffer_alloc, kPixFiller, lbd_src.buffer_alloc_sz);
+    while (i < lbd_src.buffer_alloc_sz) {
+      uint16_t spel, dpel;
+      spel = lbd_src.buffer_alloc[i];
+      // Create some distortion for dst buffer.
+      dpel = rnd.Rand8();
+      lbd_dst.buffer_alloc[i] = (uint8_t)dpel;
+      ((uint16_t *)(hbd_src.buffer_alloc))[i] = spel << (bit_depth_ - 8);
+      ((uint16_t *)(hbd_dst.buffer_alloc))[i] = dpel << (bit_depth_ - 8);
+      i++;
+    }
+
+    lbd_db = lbd_metric_(&lbd_src, &lbd_dst);
+    hbd_db = hbd_metric_(&hbd_src, &hbd_dst, input_bit_depth_, bit_depth_);
+    EXPECT_LE(fabs(lbd_db - hbd_db), threshold_);
+
+    i = 0;
+    while (i < lbd_src.buffer_alloc_sz) {
+      uint16_t dpel;
+      // Create some small distortion for dst buffer.
+      dpel = 120 + (rnd.Rand8() >> 4);
+      lbd_dst.buffer_alloc[i] = (uint8_t)dpel;
+      ((uint16_t *)(hbd_dst.buffer_alloc))[i] = dpel << (bit_depth_ - 8);
+      i++;
+    }
+
+    lbd_db = lbd_metric_(&lbd_src, &lbd_dst);
+    hbd_db = hbd_metric_(&hbd_src, &hbd_dst, input_bit_depth_, bit_depth_);
+    EXPECT_LE(fabs(lbd_db - hbd_db), threshold_);
+
+    i = 0;
+    while (i < lbd_src.buffer_alloc_sz) {
+      uint16_t dpel;
+      // Create some small distortion for dst buffer.
+      dpel = 126 + (rnd.Rand8() >> 6);
+      lbd_dst.buffer_alloc[i] = (uint8_t)dpel;
+      ((uint16_t *)(hbd_dst.buffer_alloc))[i] = dpel << (bit_depth_ - 8);
+      i++;
+    }
+
+    lbd_db = lbd_metric_(&lbd_src, &lbd_dst);
+    hbd_db = hbd_metric_(&hbd_src, &hbd_dst, input_bit_depth_, bit_depth_);
+    EXPECT_LE(fabs(lbd_db - hbd_db), threshold_);
+
+    aom_free_frame_buffer(&lbd_src);
+    aom_free_frame_buffer(&lbd_dst);
+    aom_free_frame_buffer(&hbd_src);
+    aom_free_frame_buffer(&hbd_dst);
+  }
+
+  int input_bit_depth_;
+  int bit_depth_;
+  double threshold_;
+  LBDMetricFunc lbd_metric_;
+  HBDMetricFunc hbd_metric_;
+};
+
+typedef ::testing::tuple<LBDMetricFunc, HBDMetricFunc, int, int, double>
+    MetricTestTParam;
+class HBDMetricsTest : public HBDMetricsTestBase,
+                       public ::testing::TestWithParam<MetricTestTParam> {
+ public:
+  virtual void SetUp() {
+    lbd_metric_ = GET_PARAM(0);
+    hbd_metric_ = GET_PARAM(1);
+    input_bit_depth_ = GET_PARAM(2);
+    bit_depth_ = GET_PARAM(3);
+    threshold_ = GET_PARAM(4);
+  }
+  virtual void TearDown() {}
+};
+
+TEST_P(HBDMetricsTest, RunAccuracyCheck) { RunAccuracyCheck(); }
+
+// Allow small variation due to floating point operations.
+static const double kSsim_thresh = 0.001;
+// Allow some additional errors accumulated in floating point operations.
+static const double kFSsim_thresh = 0.03;
+// Allow some extra variation due to rounding error accumulated in dct.
+static const double kPhvs_thresh = 0.3;
+
+INSTANTIATE_TEST_CASE_P(
+    AOMSSIM, HBDMetricsTest,
+    ::testing::Values(MetricTestTParam(&compute_aomssim, &compute_hbd_aomssim,
+                                       8, 10, kSsim_thresh),
+                      MetricTestTParam(&compute_aomssim, &compute_hbd_aomssim,
+                                       10, 10, kPhvs_thresh),
+                      MetricTestTParam(&compute_aomssim, &compute_hbd_aomssim,
+                                       8, 12, kSsim_thresh),
+                      MetricTestTParam(&compute_aomssim, &compute_hbd_aomssim,
+                                       12, 12, kPhvs_thresh)));
+INSTANTIATE_TEST_CASE_P(
+    FASTSSIM, HBDMetricsTest,
+    ::testing::Values(MetricTestTParam(&compute_fastssim, &compute_hbd_fastssim,
+                                       8, 10, kFSsim_thresh),
+                      MetricTestTParam(&compute_fastssim, &compute_hbd_fastssim,
+                                       10, 10, kFSsim_thresh),
+                      MetricTestTParam(&compute_fastssim, &compute_hbd_fastssim,
+                                       8, 12, kFSsim_thresh),
+                      MetricTestTParam(&compute_fastssim, &compute_hbd_fastssim,
+                                       12, 12, kFSsim_thresh)));
+INSTANTIATE_TEST_CASE_P(
+    PSNRHVS, HBDMetricsTest,
+    ::testing::Values(MetricTestTParam(&compute_psnrhvs, &compute_hbd_psnrhvs,
+                                       8, 10, kPhvs_thresh),
+                      MetricTestTParam(&compute_psnrhvs, &compute_hbd_psnrhvs,
+                                       10, 10, kPhvs_thresh),
+                      MetricTestTParam(&compute_psnrhvs, &compute_hbd_psnrhvs,
+                                       8, 12, kPhvs_thresh),
+                      MetricTestTParam(&compute_psnrhvs, &compute_hbd_psnrhvs,
+                                       12, 12, kPhvs_thresh)));
+INSTANTIATE_TEST_CASE_P(
+    PSNR, HBDMetricsTest,
+    ::testing::Values(
+        MetricTestTParam(&compute_psnr, &compute_hbd_psnr, 8, 10, kPhvs_thresh),
+        MetricTestTParam(&compute_psnr, &compute_hbd_psnr, 10, 10,
+                         kPhvs_thresh),
+        MetricTestTParam(&compute_psnr, &compute_hbd_psnr, 8, 12, kPhvs_thresh),
+        MetricTestTParam(&compute_psnr, &compute_hbd_psnr, 12, 12,
+                         kPhvs_thresh)));
+}  // namespace
diff --git a/third_party/aom/test/hiprec_convolve_test.cc b/third_party/aom/test/hiprec_convolve_test.cc
new file mode 100644
index 0000000000..f94a0730c1
--- /dev/null
+++ b/third_party/aom/test/hiprec_convolve_test.cc
@@ -0,0 +1,62 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/hiprec_convolve_test_util.h"
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+using libaom_test::AV1HighbdHiprecConvolve::AV1HighbdHiprecConvolveTest;
+using libaom_test::AV1HiprecConvolve::AV1HiprecConvolveTest;
+
+namespace {
+
+TEST_P(AV1HiprecConvolveTest, CheckOutput) { RunCheckOutput(GET_PARAM(3)); }
+TEST_P(AV1HiprecConvolveTest, DISABLED_SpeedTest) {
+  RunSpeedTest(GET_PARAM(3));
+}
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, AV1HiprecConvolveTest,
+                        libaom_test::AV1HiprecConvolve::BuildParams(
+                            av1_wiener_convolve_add_src_sse2));
+#endif
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, AV1HiprecConvolveTest,
+                        libaom_test::AV1HiprecConvolve::BuildParams(
+                            av1_wiener_convolve_add_src_avx2));
+#endif
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, AV1HiprecConvolveTest,
+                        libaom_test::AV1HiprecConvolve::BuildParams(
+                            av1_wiener_convolve_add_src_neon));
+#endif
+
+#if HAVE_SSSE3 || HAVE_AVX2
+TEST_P(AV1HighbdHiprecConvolveTest, CheckOutput) {
+  RunCheckOutput(GET_PARAM(4));
+}
+TEST_P(AV1HighbdHiprecConvolveTest, DISABLED_SpeedTest) {
+  RunSpeedTest(GET_PARAM(4));
+}
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(SSSE3, AV1HighbdHiprecConvolveTest,
+                        libaom_test::AV1HighbdHiprecConvolve::BuildParams(
+                            av1_highbd_wiener_convolve_add_src_ssse3));
+#endif
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, AV1HighbdHiprecConvolveTest,
+                        libaom_test::AV1HighbdHiprecConvolve::BuildParams(
+                            av1_highbd_wiener_convolve_add_src_avx2));
+#endif
+#endif
+
+}  // namespace
diff --git a/third_party/aom/test/hiprec_convolve_test_util.cc b/third_party/aom/test/hiprec_convolve_test_util.cc
new file mode 100644
index 0000000000..2672bcec3d
--- /dev/null
+++ b/third_party/aom/test/hiprec_convolve_test_util.cc
@@ -0,0 +1,331 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "test/hiprec_convolve_test_util.h"
+
+#include "av1/common/restoration.h"
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+
+namespace libaom_test {
+
+// Generate a random pair of filter kernels, using the ranges
+// of possible values from the loop-restoration experiment
+static void generate_kernels(ACMRandom *rnd, InterpKernel hkernel,
+                             InterpKernel vkernel) {
+  hkernel[0] = hkernel[6] =
+      WIENER_FILT_TAP0_MINV +
+      rnd->PseudoUniform(WIENER_FILT_TAP0_MAXV + 1 - WIENER_FILT_TAP0_MINV);
+  hkernel[1] = hkernel[5] =
+      WIENER_FILT_TAP1_MINV +
+      rnd->PseudoUniform(WIENER_FILT_TAP1_MAXV + 1 - WIENER_FILT_TAP1_MINV);
+  hkernel[2] = hkernel[4] =
+      WIENER_FILT_TAP2_MINV +
+      rnd->PseudoUniform(WIENER_FILT_TAP2_MAXV + 1 - WIENER_FILT_TAP2_MINV);
+  hkernel[3] = -(hkernel[0] + hkernel[1] + hkernel[2]);
+  hkernel[7] = 0;
+
+  vkernel[0] = vkernel[6] =
+      WIENER_FILT_TAP0_MINV +
+      rnd->PseudoUniform(WIENER_FILT_TAP0_MAXV + 1 - WIENER_FILT_TAP0_MINV);
+  vkernel[1] = vkernel[5] =
+      WIENER_FILT_TAP1_MINV +
+      rnd->PseudoUniform(WIENER_FILT_TAP1_MAXV + 1 - WIENER_FILT_TAP1_MINV);
+  vkernel[2] = vkernel[4] =
+      WIENER_FILT_TAP2_MINV +
+      rnd->PseudoUniform(WIENER_FILT_TAP2_MAXV + 1 - WIENER_FILT_TAP2_MINV);
+  vkernel[3] = -(vkernel[0] + vkernel[1] + vkernel[2]);
+  vkernel[7] = 0;
+}
+
+namespace AV1HiprecConvolve {
+
+::testing::internal::ParamGenerator<HiprecConvolveParam> BuildParams(
+    hiprec_convolve_func filter) {
+  const HiprecConvolveParam params[] = {
+    make_tuple(8, 8, 50000, filter),   make_tuple(8, 4, 50000, filter),
+    make_tuple(64, 24, 1000, filter),  make_tuple(64, 64, 1000, filter),
+    make_tuple(64, 56, 1000, filter),  make_tuple(32, 8, 10000, filter),
+    make_tuple(32, 28, 10000, filter), make_tuple(32, 32, 10000, filter),
+    make_tuple(16, 34, 10000, filter), make_tuple(32, 34, 10000, filter),
+    make_tuple(64, 34, 1000, filter),  make_tuple(8, 17, 10000, filter),
+    make_tuple(16, 17, 10000, filter), make_tuple(32, 17, 10000, filter)
+  };
+  return ::testing::ValuesIn(params);
+}
+
+AV1HiprecConvolveTest::~AV1HiprecConvolveTest() {}
+void AV1HiprecConvolveTest::SetUp() {
+  rnd_.Reset(ACMRandom::DeterministicSeed());
+}
+
+void AV1HiprecConvolveTest::TearDown() { libaom_test::ClearSystemState(); }
+
+void AV1HiprecConvolveTest::RunCheckOutput(hiprec_convolve_func test_impl) {
+  const int w = 128, h = 128;
+  const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
+  const int num_iters = GET_PARAM(2);
+  int i, j;
+  const ConvolveParams conv_params = get_conv_params_wiener(8);
+
+  uint8_t *input_ = new uint8_t[h * w];
+  uint8_t *input = input_;
+
+  // The AVX2 convolve functions always write rows with widths that are
+  // multiples of 16. So to avoid a buffer overflow, we may need to pad
+  // rows to a multiple of 16.
+  int output_n = ALIGN_POWER_OF_TWO(out_w, 4) * out_h;
+  uint8_t *output = new uint8_t[output_n];
+  uint8_t *output2 = new uint8_t[output_n];
+
+  // Generate random filter kernels
+  DECLARE_ALIGNED(16, InterpKernel, hkernel);
+  DECLARE_ALIGNED(16, InterpKernel, vkernel);
+
+  generate_kernels(&rnd_, hkernel, vkernel);
+
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand8();
+
+  for (i = 0; i < num_iters; ++i) {
+    // Choose random locations within the source block
+    int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+    int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+    av1_wiener_convolve_add_src_c(input + offset_r * w + offset_c, w, output,
+                                  out_w, hkernel, 16, vkernel, 16, out_w, out_h,
+                                  &conv_params);
+    test_impl(input + offset_r * w + offset_c, w, output2, out_w, hkernel, 16,
+              vkernel, 16, out_w, out_h, &conv_params);
+
+    for (j = 0; j < out_w * out_h; ++j)
+      ASSERT_EQ(output[j], output2[j])
+          << "Pixel mismatch at index " << j << " = (" << (j % out_w) << ", "
+          << (j / out_w) << ") on iteration " << i;
+  }
+  delete[] input_;
+  delete[] output;
+  delete[] output2;
+}
+
+void AV1HiprecConvolveTest::RunSpeedTest(hiprec_convolve_func test_impl) {
+  const int w = 128, h = 128;
+  const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
+  const int num_iters = GET_PARAM(2) / 500;
+  int i, j, k;
+  const ConvolveParams conv_params = get_conv_params_wiener(8);
+
+  uint8_t *input_ = new uint8_t[h * w];
+  uint8_t *input = input_;
+
+  // The AVX2 convolve functions always write rows with widths that are
+  // multiples of 16. So to avoid a buffer overflow, we may need to pad
+  // rows to a multiple of 16.
+  int output_n = ALIGN_POWER_OF_TWO(out_w, 4) * out_h;
+  uint8_t *output = new uint8_t[output_n];
+  uint8_t *output2 = new uint8_t[output_n];
+
+  // Generate random filter kernels
+  DECLARE_ALIGNED(16, InterpKernel, hkernel);
+  DECLARE_ALIGNED(16, InterpKernel, vkernel);
+
+  generate_kernels(&rnd_, hkernel, vkernel);
+
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand8();
+
+  aom_usec_timer ref_timer;
+  aom_usec_timer_start(&ref_timer);
+  for (i = 0; i < num_iters; ++i) {
+    for (j = 3; j < h - out_h - 4; j++) {
+      for (k = 3; k < w - out_w - 4; k++) {
+        av1_wiener_convolve_add_src_c(input + j * w + k, w, output, out_w,
+                                      hkernel, 16, vkernel, 16, out_w, out_h,
+                                      &conv_params);
+      }
+    }
+  }
+  aom_usec_timer_mark(&ref_timer);
+  const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);
+
+  aom_usec_timer tst_timer;
+  aom_usec_timer_start(&tst_timer);
+  for (i = 0; i < num_iters; ++i) {
+    for (j = 3; j < h - out_h - 4; j++) {
+      for (k = 3; k < w - out_w - 4; k++) {
+        test_impl(input + j * w + k, w, output2, out_w, hkernel, 16, vkernel,
+                  16, out_w, out_h, &conv_params);
+      }
+    }
+  }
+  aom_usec_timer_mark(&tst_timer);
+  const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);
+
+  std::cout << "[          ] C time = " << ref_time / 1000
+            << " ms, SIMD time = " << tst_time / 1000 << " ms\n";
+
+  EXPECT_GT(ref_time, tst_time)
+      << "Error: AV1HiprecConvolveTest.SpeedTest, SIMD slower than C.\n"
+      << "C time: " << ref_time << " us\n"
+      << "SIMD time: " << tst_time << " us\n";
+
+  delete[] input_;
+  delete[] output;
+  delete[] output2;
+}
+}  // namespace AV1HiprecConvolve
+
+namespace AV1HighbdHiprecConvolve {
+
+::testing::internal::ParamGenerator<HighbdHiprecConvolveParam> BuildParams(
+    highbd_hiprec_convolve_func filter) {
+  const HighbdHiprecConvolveParam params[] = {
+    make_tuple(8, 8, 50000, 8, filter),   make_tuple(64, 64, 1000, 8, filter),
+    make_tuple(32, 8, 10000, 8, filter),  make_tuple(8, 8, 50000, 10, filter),
+    make_tuple(64, 64, 1000, 10, filter), make_tuple(32, 8, 10000, 10, filter),
+    make_tuple(8, 8, 50000, 12, filter),  make_tuple(64, 64, 1000, 12, filter),
+    make_tuple(32, 8, 10000, 12, filter),
+  };
+  return ::testing::ValuesIn(params);
+}
+
+AV1HighbdHiprecConvolveTest::~AV1HighbdHiprecConvolveTest() {}
+void AV1HighbdHiprecConvolveTest::SetUp() {
+  rnd_.Reset(ACMRandom::DeterministicSeed());
+}
+
+void AV1HighbdHiprecConvolveTest::TearDown() {
+  libaom_test::ClearSystemState();
+}
+
+void AV1HighbdHiprecConvolveTest::RunCheckOutput(
+    highbd_hiprec_convolve_func test_impl) {
+  const int w = 128, h = 128;
+  const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
+  const int num_iters = GET_PARAM(2);
+  const int bd = GET_PARAM(3);
+  int i, j;
+  const ConvolveParams conv_params = get_conv_params_wiener(bd);
+
+  uint16_t *input = new uint16_t[h * w];
+
+  // The AVX2 convolve functions always write rows with widths that are
+  // multiples of 16. So to avoid a buffer overflow, we may need to pad
+  // rows to a multiple of 16.
+  int output_n = ALIGN_POWER_OF_TWO(out_w, 4) * out_h;
+  uint16_t *output = new uint16_t[output_n];
+  uint16_t *output2 = new uint16_t[output_n];
+
+  // Generate random filter kernels
+  DECLARE_ALIGNED(16, InterpKernel, hkernel);
+  DECLARE_ALIGNED(16, InterpKernel, vkernel);
+
+  generate_kernels(&rnd_, hkernel, vkernel);
+
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+
+  uint8_t *input_ptr = CONVERT_TO_BYTEPTR(input);
+  uint8_t *output_ptr = CONVERT_TO_BYTEPTR(output);
+  uint8_t *output2_ptr = CONVERT_TO_BYTEPTR(output2);
+
+  for (i = 0; i < num_iters; ++i) {
+    // Choose random locations within the source block
+    int offset_r = 3 + rnd_.PseudoUniform(h - out_h - 7);
+    int offset_c = 3 + rnd_.PseudoUniform(w - out_w - 7);
+    av1_highbd_wiener_convolve_add_src_c(
+        input_ptr + offset_r * w + offset_c, w, output_ptr, out_w, hkernel, 16,
+        vkernel, 16, out_w, out_h, &conv_params, bd);
+    test_impl(input_ptr + offset_r * w + offset_c, w, output2_ptr, out_w,
+              hkernel, 16, vkernel, 16, out_w, out_h, &conv_params, bd);
+
+    for (j = 0; j < out_w * out_h; ++j)
+      ASSERT_EQ(output[j], output2[j])
+          << "Pixel mismatch at index " << j << " = (" << (j % out_w) << ", "
+          << (j / out_w) << ") on iteration " << i;
+  }
+  delete[] input;
+  delete[] output;
+  delete[] output2;
+}
+
+void AV1HighbdHiprecConvolveTest::RunSpeedTest(
+    highbd_hiprec_convolve_func test_impl) {
+  const int w = 128, h = 128;
+  const int out_w = GET_PARAM(0), out_h = GET_PARAM(1);
+  const int num_iters = GET_PARAM(2) / 500;
+  const int bd = GET_PARAM(3);
+  int i, j, k;
+  const ConvolveParams conv_params = get_conv_params_wiener(bd);
+
+  uint16_t *input = new uint16_t[h * w];
+
+  // The AVX2 convolve functions always write rows with widths that are
+  // multiples of 16. So to avoid a buffer overflow, we may need to pad
+  // rows to a multiple of 16.
+  int output_n = ALIGN_POWER_OF_TWO(out_w, 4) * out_h;
+  uint16_t *output = new uint16_t[output_n];
+  uint16_t *output2 = new uint16_t[output_n];
+
+  // Generate random filter kernels
+  DECLARE_ALIGNED(16, InterpKernel, hkernel);
+  DECLARE_ALIGNED(16, InterpKernel, vkernel);
+
+  generate_kernels(&rnd_, hkernel, vkernel);
+
+  for (i = 0; i < h; ++i)
+    for (j = 0; j < w; ++j) input[i * w + j] = rnd_.Rand16() & ((1 << bd) - 1);
+
+  uint8_t *input_ptr = CONVERT_TO_BYTEPTR(input);
+  uint8_t *output_ptr = CONVERT_TO_BYTEPTR(output);
+  uint8_t *output2_ptr = CONVERT_TO_BYTEPTR(output2);
+
+  aom_usec_timer ref_timer;
+  aom_usec_timer_start(&ref_timer);
+  for (i = 0; i < num_iters; ++i) {
+    for (j = 3; j < h - out_h - 4; j++) {
+      for (k = 3; k < w - out_w - 4; k++) {
+        av1_highbd_wiener_convolve_add_src_c(
+            input_ptr + j * w + k, w, output_ptr, out_w, hkernel, 16, vkernel,
+            16, out_w, out_h, &conv_params, bd);
+      }
+    }
+  }
+  aom_usec_timer_mark(&ref_timer);
+  const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);
+
+  aom_usec_timer tst_timer;
+  aom_usec_timer_start(&tst_timer);
+  for (i = 0; i < num_iters; ++i) {
+    for (j = 3; j < h - out_h - 4; j++) {
+      for (k = 3; k < w - out_w - 4; k++) {
+        test_impl(input_ptr + j * w + k, w, output2_ptr, out_w, hkernel, 16,
+                  vkernel, 16, out_w, out_h, &conv_params, bd);
+      }
+    }
+  }
+  aom_usec_timer_mark(&tst_timer);
+  const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);
+
+  std::cout << "[          ] C time = " << ref_time / 1000
+            << " ms, SIMD time = " << tst_time / 1000 << " ms\n";
+
+  EXPECT_GT(ref_time, tst_time)
+      << "Error: AV1HighbdHiprecConvolveTest.SpeedTest, SIMD slower than C.\n"
+      << "C time: " << ref_time << " us\n"
+      << "SIMD time: " << tst_time << " us\n";
+
+  delete[] input;
+  delete[] output;
+  delete[] output2;
+}
+}  // namespace AV1HighbdHiprecConvolve
+}  // namespace libaom_test
diff --git a/third_party/aom/test/hiprec_convolve_test_util.h b/third_party/aom/test/hiprec_convolve_test_util.h
new file mode 100644
index 0000000000..2abe24b57e
--- /dev/null
+++ b/third_party/aom/test/hiprec_convolve_test_util.h
@@ -0,0 +1,93 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_HIPREC_CONVOLVE_TEST_UTIL_H_
+#define AOM_TEST_HIPREC_CONVOLVE_TEST_UTIL_H_
+
+#include "config/av1_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/util.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "aom_ports/aom_timer.h"
+#include "av1/common/convolve.h"
+#include "av1/common/mv.h"
+
+namespace libaom_test {
+
+namespace AV1HiprecConvolve {
+
+typedef void (*hiprec_convolve_func)(const uint8_t *src, ptrdiff_t src_stride,
+                                     uint8_t *dst, ptrdiff_t dst_stride,
+                                     const int16_t *filter_x, int x_step_q4,
+                                     const int16_t *filter_y, int y_step_q4,
+                                     int w, int h,
+                                     const ConvolveParams *conv_params);
+
+typedef ::testing::tuple<int, int, int, hiprec_convolve_func>
+    HiprecConvolveParam;
+
+::testing::internal::ParamGenerator<HiprecConvolveParam> BuildParams(
+    hiprec_convolve_func filter);
+
+class AV1HiprecConvolveTest
+    : public ::testing::TestWithParam<HiprecConvolveParam> {
+ public:
+  virtual ~AV1HiprecConvolveTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput(hiprec_convolve_func test_impl);
+  void RunSpeedTest(hiprec_convolve_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+};
+
+}  // namespace AV1HiprecConvolve
+
+namespace AV1HighbdHiprecConvolve {
+typedef void (*highbd_hiprec_convolve_func)(
+    const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
+    ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4,
+    const int16_t *filter_y, int y_step_q4, int w, int h,
+    const ConvolveParams *conv_params, int bps);
+
+typedef ::testing::tuple<int, int, int, int, highbd_hiprec_convolve_func>
+    HighbdHiprecConvolveParam;
+
+::testing::internal::ParamGenerator<HighbdHiprecConvolveParam> BuildParams(
+    highbd_hiprec_convolve_func filter);
+
+class AV1HighbdHiprecConvolveTest
+    : public ::testing::TestWithParam<HighbdHiprecConvolveParam> {
+ public:
+  virtual ~AV1HighbdHiprecConvolveTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput(highbd_hiprec_convolve_func test_impl);
+  void RunSpeedTest(highbd_hiprec_convolve_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+};
+
+}  // namespace AV1HighbdHiprecConvolve
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_HIPREC_CONVOLVE_TEST_UTIL_H_
diff --git a/third_party/aom/test/horz_superres_test.cc b/third_party/aom/test/horz_superres_test.cc
new file mode 100644
index 0000000000..973f55b668
--- /dev/null
+++ b/third_party/aom/test/horz_superres_test.cc
@@ -0,0 +1,322 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "av1/encoder/encoder.h"
+
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+#include "test/yuv_video_source.h"
+
+namespace {
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+
+/* TESTING PARAMETERS */
+
+#define NUM_TEST_VIDEOS 3
+
+const int kBitrate = 40;
+
+// PSNR thresholds found by experiment
+const double kPSNRThresholds[] = { 26.0, 28.0, 20.0 };
+
+typedef struct {
+  const char *filename;
+  aom_img_fmt fmt;
+  aom_bit_depth_t bit_depth;
+  unsigned int profile;
+  unsigned int limit;
+  unsigned int screen_content;
+} TestVideoParam;
+
+const TestVideoParam kTestVideoVectors[] = {
+  { "park_joy_90p_8_420.y4m", AOM_IMG_FMT_I420, AOM_BITS_8, 0, 5, 0 },
+  { "park_joy_90p_10_444.y4m", AOM_IMG_FMT_I44416, AOM_BITS_10, 1, 5, 0 },
+  { "screendata.y4m", AOM_IMG_FMT_I420, AOM_BITS_8, 0, 4, 1 },
+};
+
+// Superres modes tested
+// SUPERRES_QTHRESH is not included, as it has its own test
+const SUPERRES_MODE kSuperresModesNotQThresh[] = { SUPERRES_FIXED,
+                                                   SUPERRES_RANDOM };
+
+// Superres denominators and superres kf denominators to be tested
+typedef tuple<int, int> SuperresDenominatorPair;
+const SuperresDenominatorPair kSuperresDenominators[] = {
+  make_tuple(16, 9),  make_tuple(13, 11), make_tuple(9, 9),
+  make_tuple(13, 13), make_tuple(11, 16), make_tuple(8, 16),
+  make_tuple(16, 8),  make_tuple(8, 8),   make_tuple(9, 14),
+};
+
+// Superres q thresholds and superres kf q thresholds to be tested
+typedef tuple<int, int> SuperresQThresholdPair;
+const SuperresQThresholdPair kSuperresQThresholds[] = {
+  make_tuple(63, 63), make_tuple(63, 41), make_tuple(17, 63),
+  make_tuple(41, 11), make_tuple(1, 37),  make_tuple(11, 11),
+  make_tuple(1, 1),   make_tuple(17, 29), make_tuple(29, 11),
+};
+
+/* END (TESTING PARAMETERS) */
+
+// Test parameter list:
+//  <[needed for EncoderTest], test_video_idx_, superres_mode_,
+//  tuple(superres_denom_, superres_kf_denom_)>
+typedef tuple<const libaom_test::CodecFactory *, int, SUPERRES_MODE,
+              SuperresDenominatorPair>
+    HorzSuperresTestParam;
+
+class HorzSuperresEndToEndTest
+    : public ::testing::TestWithParam<HorzSuperresTestParam>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  HorzSuperresEndToEndTest()
+      : EncoderTest(GET_PARAM(0)), test_video_idx_(GET_PARAM(1)),
+        superres_mode_(GET_PARAM(2)), psnr_(0.0), frame_count_(0) {
+    test_video_param_ = kTestVideoVectors[test_video_idx_];
+
+    SuperresDenominatorPair denoms = GET_PARAM(3);
+    superres_denom_ = ::testing::get<0>(denoms);
+    superres_kf_denom_ = ::testing::get<1>(denoms);
+  }
+
+  virtual ~HorzSuperresEndToEndTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(::libaom_test::kTwoPassGood);
+    cfg_.g_lag_in_frames = 5;
+    cfg_.rc_end_usage = AOM_VBR;
+    cfg_.rc_target_bitrate = kBitrate;
+    cfg_.g_error_resilient = 0;
+    cfg_.g_profile = test_video_param_.profile;
+    cfg_.g_input_bit_depth = (unsigned int)test_video_param_.bit_depth;
+    cfg_.g_bit_depth = test_video_param_.bit_depth;
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    if (cfg_.g_bit_depth > 8) init_flags_ |= AOM_CODEC_USE_HIGHBITDEPTH;
+
+    // Set superres parameters
+    cfg_.rc_superres_mode = superres_mode_;
+    cfg_.rc_superres_denominator = superres_denom_;
+    cfg_.rc_superres_kf_denominator = superres_kf_denom_;
+  }
+
+  virtual void BeginPassHook(unsigned int) {
+    psnr_ = 0.0;
+    frame_count_ = 0;
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    psnr_ += pkt->data.psnr.psnr[0];
+    frame_count_++;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
+      encoder->Control(AV1E_SET_TILE_COLUMNS, 4);
+
+      // Set cpu-used = 8 for speed
+      encoder->Control(AOME_SET_CPUUSED, 8);
+
+      // Test screen coding tools
+      if (test_video_param_.screen_content)
+        encoder->Control(AV1E_SET_TUNE_CONTENT, AOM_CONTENT_SCREEN);
+      else
+        encoder->Control(AV1E_SET_TUNE_CONTENT, AOM_CONTENT_DEFAULT);
+
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+      encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+      encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+    }
+  }
+
+  double GetAveragePsnr() const {
+    if (frame_count_) return psnr_ / frame_count_;
+    return 0.0;
+  }
+
+  double GetPsnrThreshold() { return kPSNRThresholds[test_video_idx_]; }
+
+  void DoTest() {
+    testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+    video.reset(new libaom_test::Y4mVideoSource(test_video_param_.filename, 0,
+                                                test_video_param_.limit));
+    ASSERT_TRUE(video.get() != NULL);
+
+    ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+    const double psnr = GetAveragePsnr();
+    EXPECT_GT(psnr, GetPsnrThreshold())
+        << "superres_mode_ = " << superres_mode_
+        << ", superres_denom_ = " << superres_denom_
+        << ", superres_kf_denom_ = " << superres_kf_denom_;
+
+    EXPECT_EQ(test_video_param_.limit, frame_count_)
+        << "superres_mode_ = " << superres_mode_
+        << ", superres_denom_ = " << superres_denom_
+        << ", superres_kf_denom_ = " << superres_kf_denom_;
+  }
+
+  int test_video_idx_;
+  TestVideoParam test_video_param_;
+  SUPERRES_MODE superres_mode_;
+  int superres_denom_;
+  int superres_kf_denom_;
+
+ private:
+  double psnr_;
+  unsigned int frame_count_;
+};
+
+TEST_P(HorzSuperresEndToEndTest, HorzSuperresEndToEndPSNRTest) { DoTest(); }
+
+AV1_INSTANTIATE_TEST_CASE(HorzSuperresEndToEndTest,
+                          ::testing::Range(0, NUM_TEST_VIDEOS),
+                          ::testing::ValuesIn(kSuperresModesNotQThresh),
+                          ::testing::ValuesIn(kSuperresDenominators));
+
+// Test parameter list:
+//  <[needed for EncoderTest], test_video_idx_, tuple(superres_denom_,
+//  superres_kf_denom_), tuple(superres_qthresh_,superres_kf_qthresh_)>
+typedef tuple<const libaom_test::CodecFactory *, int, SuperresDenominatorPair,
+              SuperresQThresholdPair>
+    HorzSuperresQThreshTestParam;
+
+class HorzSuperresQThreshEndToEndTest
+    : public ::testing::TestWithParam<HorzSuperresQThreshTestParam>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  HorzSuperresQThreshEndToEndTest()
+      : EncoderTest(GET_PARAM(0)), test_video_idx_(GET_PARAM(1)),
+        superres_mode_(SUPERRES_QTHRESH), psnr_(0.0), frame_count_(0) {
+    test_video_param_ = kTestVideoVectors[test_video_idx_];
+
+    SuperresDenominatorPair denoms = GET_PARAM(2);
+    superres_denom_ = ::testing::get<0>(denoms);
+    superres_kf_denom_ = ::testing::get<1>(denoms);
+
+    SuperresQThresholdPair qthresholds = GET_PARAM(3);
+    superres_qthresh_ = ::testing::get<0>(qthresholds);
+    superres_kf_qthresh_ = ::testing::get<1>(qthresholds);
+  }
+
+  virtual ~HorzSuperresQThreshEndToEndTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(::libaom_test::kTwoPassGood);
+    cfg_.g_lag_in_frames = 5;
+    cfg_.rc_end_usage = AOM_VBR;
+    cfg_.rc_target_bitrate = kBitrate;
+    cfg_.g_error_resilient = 0;
+    cfg_.g_profile = test_video_param_.profile;
+    cfg_.g_input_bit_depth = (unsigned int)test_video_param_.bit_depth;
+    cfg_.g_bit_depth = test_video_param_.bit_depth;
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    if (cfg_.g_bit_depth > 8) init_flags_ |= AOM_CODEC_USE_HIGHBITDEPTH;
+
+    // Set superres parameters
+    cfg_.rc_superres_mode = superres_mode_;
+    cfg_.rc_superres_denominator = superres_denom_;
+    cfg_.rc_superres_kf_denominator = superres_kf_denom_;
+    cfg_.rc_superres_qthresh = superres_qthresh_;
+    cfg_.rc_superres_kf_qthresh = superres_kf_qthresh_;
+  }
+
+  virtual void BeginPassHook(unsigned int) {
+    psnr_ = 0.0;
+    frame_count_ = 0;
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    psnr_ += pkt->data.psnr.psnr[0];
+    frame_count_++;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AV1E_SET_FRAME_PARALLEL_DECODING, 1);
+      encoder->Control(AV1E_SET_TILE_COLUMNS, 0);
+
+      // Set cpu-used = 8 for speed
+      encoder->Control(AOME_SET_CPUUSED, 8);
+
+      // Test screen coding tools
+      if (test_video_param_.screen_content)
+        encoder->Control(AV1E_SET_TUNE_CONTENT, AOM_CONTENT_SCREEN);
+      else
+        encoder->Control(AV1E_SET_TUNE_CONTENT, AOM_CONTENT_DEFAULT);
+
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+      encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+      encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+    }
+  }
+
+  double GetAveragePsnr() const {
+    if (frame_count_) return psnr_ / frame_count_;
+    return 0.0;
+  }
+
+  double GetPsnrThreshold() { return kPSNRThresholds[test_video_idx_]; }
+
+  void DoTest() {
+    testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+    video.reset(new libaom_test::Y4mVideoSource(test_video_param_.filename, 0,
+                                                test_video_param_.limit));
+    ASSERT_TRUE(video.get() != NULL);
+
+    ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+    const double psnr = GetAveragePsnr();
+    EXPECT_GT(psnr, GetPsnrThreshold())
+        << "superres_mode_ = " << superres_mode_
+        << ", superres_denom_ = " << superres_denom_
+        << ", superres_kf_denom_ = " << superres_kf_denom_
+        << ", superres_qthresh_ = " << superres_qthresh_
+        << ", superres_kf_qthresh_ = " << superres_kf_qthresh_;
+
+    EXPECT_EQ(test_video_param_.limit, frame_count_)
+        << "superres_mode_ = " << superres_mode_
+        << ", superres_denom_ = " << superres_denom_
+        << ", superres_kf_denom_ = " << superres_kf_denom_
+        << ", superres_qthresh_ = " << superres_qthresh_
+        << ", superres_kf_qthresh_ = " << superres_kf_qthresh_;
+  }
+
+  int test_video_idx_;
+  TestVideoParam test_video_param_;
+  SUPERRES_MODE superres_mode_;
+  int superres_denom_;
+  int superres_kf_denom_;
+  int superres_qthresh_;
+  int superres_kf_qthresh_;
+
+ private:
+  double psnr_;
+  unsigned int frame_count_;
+};
+
+TEST_P(HorzSuperresQThreshEndToEndTest, HorzSuperresQThreshEndToEndPSNRTest) {
+  DoTest();
+}
+
+AV1_INSTANTIATE_TEST_CASE(HorzSuperresQThreshEndToEndTest,
+                          ::testing::Range(0, NUM_TEST_VIDEOS),
+                          ::testing::ValuesIn(kSuperresDenominators),
+                          ::testing::ValuesIn(kSuperresQThresholds));
+
+}  // namespace
diff --git a/third_party/aom/test/i420_video_source.h b/third_party/aom/test/i420_video_source.h
new file mode 100644
index 0000000000..233e7152b9
--- /dev/null
+++ b/third_party/aom/test/i420_video_source.h
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_I420_VIDEO_SOURCE_H_
+#define AOM_TEST_I420_VIDEO_SOURCE_H_
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+
+#include "test/yuv_video_source.h"
+
+namespace libaom_test {
+
+// This class extends VideoSource to allow parsing of raw yv12
+// so that we can do actual file encodes.
+class I420VideoSource : public YUVVideoSource {
+ public:
+  I420VideoSource(const std::string &file_name, unsigned int width,
+                  unsigned int height, int rate_numerator, int rate_denominator,
+                  unsigned int start, int limit)
+      : YUVVideoSource(file_name, AOM_IMG_FMT_I420, width, height,
+                       rate_numerator, rate_denominator, start, limit) {}
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_I420_VIDEO_SOURCE_H_
diff --git a/third_party/aom/test/intra_edge_test.cc b/third_party/aom/test/intra_edge_test.cc
new file mode 100644
index 0000000000..ce61402acb
--- /dev/null
+++ b/third_party/aom/test/intra_edge_test.cc
@@ -0,0 +1,337 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/register_state_check.h"
+#include "test/function_equivalence_test.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+
+#include "aom/aom_integer.h"
+#include "av1/common/enums.h"
+
+using libaom_test::FunctionEquivalenceTest;
+
+namespace {
+
+template <typename F, typename T>
+class UpsampleTest : public FunctionEquivalenceTest<F> {
+ protected:
+  static const int kIterations = 1000000;
+  static const int kMinEdge = 4;
+  static const int kMaxEdge = 24;
+  static const int kBufSize = 2 * 64 + 32;
+  static const int kOffset = 16;
+
+  virtual ~UpsampleTest() {}
+
+  virtual void Execute(T *edge_tst) = 0;
+
+  void Common() {
+    edge_ref_ = &edge_ref_data_[kOffset];
+    edge_tst_ = &edge_tst_data_[kOffset];
+
+    Execute(edge_tst_);
+
+    const int max_idx = (size_ - 1) * 2;
+    for (int r = -2; r <= max_idx; ++r) {
+      ASSERT_EQ(edge_ref_[r], edge_tst_[r]);
+    }
+  }
+
+  T edge_ref_data_[kBufSize];
+  T edge_tst_data_[kBufSize];
+
+  T *edge_ref_;
+  T *edge_tst_;
+
+  int size_;
+};
+
+//////////////////////////////////////////////////////////////////////////////
+// 8 bit version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*UP8B)(uint8_t *p, int size);
+typedef libaom_test::FuncParam<UP8B> TestFuncs;
+
+class UpsampleTest8B : public UpsampleTest<UP8B, uint8_t> {
+ protected:
+  void Execute(uint8_t *edge_tst) {
+    params_.ref_func(edge_ref_, size_);
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(edge_tst, size_));
+  }
+};
+
+TEST_P(UpsampleTest8B, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    size_ = 4 * (this->rng_(4) + 1);
+
+    int i, pix = 0;
+    for (i = 0; i < kOffset + size_; ++i) {
+      pix = rng_.Rand8();
+      edge_ref_data_[i] = pix;
+      edge_tst_data_[i] = edge_ref_data_[i];
+    }
+
+    // Extend final sample
+    while (i < kBufSize) {
+      edge_ref_data_[i] = pix;
+      edge_tst_data_[i] = pix;
+      i++;
+    }
+
+    Common();
+  }
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, UpsampleTest8B,
+    ::testing::Values(TestFuncs(av1_upsample_intra_edge_c,
+                                av1_upsample_intra_edge_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+//////////////////////////////////////////////////////////////////////////////
+// High bit-depth version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*UPHB)(uint16_t *p, int size, int bd);
+typedef libaom_test::FuncParam<UPHB> TestFuncsHBD;
+
+class UpsampleTestHB : public UpsampleTest<UPHB, uint16_t> {
+ protected:
+  void Execute(uint16_t *edge_tst) {
+    params_.ref_func(edge_ref_, size_, bit_depth_);
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(edge_tst, size_, bit_depth_));
+  }
+  int bit_depth_;
+};
+
+TEST_P(UpsampleTestHB, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    switch (rng_(3)) {
+      case 0: bit_depth_ = 8; break;
+      case 1: bit_depth_ = 10; break;
+      default: bit_depth_ = 12; break;
+    }
+    const int hi = 1 << bit_depth_;
+
+    size_ = 4 * (this->rng_(4) + 1);
+
+    int i, pix = 0;
+    for (i = 0; i < kOffset + size_; ++i) {
+      pix = rng_(hi);
+      edge_ref_data_[i] = pix;
+      edge_tst_data_[i] = pix;
+    }
+
+    // Extend final sample
+    while (i < kBufSize) {
+      edge_ref_data_[i] = pix;
+      edge_tst_data_[i] = pix;
+      i++;
+    }
+
+    Common();
+  }
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, UpsampleTestHB,
+    ::testing::Values(TestFuncsHBD(av1_upsample_intra_edge_high_c,
+                                   av1_upsample_intra_edge_high_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+template <typename F, typename T>
+class FilterEdgeTest : public FunctionEquivalenceTest<F> {
+ protected:
+  static const int kIterations = 1000000;
+  static const int kMaxEdge = 2 * 64;
+  static const int kBufSize = kMaxEdge + 32;
+  static const int kOffset = 15;
+
+  virtual ~FilterEdgeTest() {}
+
+  virtual void Execute(T *edge_tst) = 0;
+
+  void Common() {
+    edge_ref_ = &edge_ref_data_[kOffset];
+    edge_tst_ = &edge_tst_data_[kOffset];
+
+    Execute(edge_tst_);
+
+    for (int r = 0; r < size_; ++r) {
+      ASSERT_EQ(edge_ref_[r], edge_tst_[r]);
+    }
+  }
+
+  T edge_ref_data_[kBufSize];
+  T edge_tst_data_[kBufSize];
+
+  T *edge_ref_;
+  T *edge_tst_;
+
+  int size_;
+  int strength_;
+};
+
+//////////////////////////////////////////////////////////////////////////////
+// 8 bit version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*FE8B)(uint8_t *p, int size, int strength);
+typedef libaom_test::FuncParam<FE8B> FilterEdgeTestFuncs;
+
+class FilterEdgeTest8B : public FilterEdgeTest<FE8B, uint8_t> {
+ protected:
+  void Execute(uint8_t *edge_tst) {
+    params_.ref_func(edge_ref_, size_, strength_);
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(edge_tst, size_, strength_));
+  }
+};
+
+TEST_P(FilterEdgeTest8B, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    strength_ = this->rng_(4);
+    size_ = 4 * (this->rng_(128 / 4) + 1) + 1;
+
+    int i, pix = 0;
+    for (i = 0; i < kOffset + size_; ++i) {
+      pix = rng_.Rand8();
+      edge_ref_data_[i] = pix;
+      edge_tst_data_[i] = pix;
+    }
+
+    Common();
+  }
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, FilterEdgeTest8B,
+    ::testing::Values(FilterEdgeTestFuncs(av1_filter_intra_edge_c,
+                                          av1_filter_intra_edge_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+//////////////////////////////////////////////////////////////////////////////
+// High bit-depth version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef void (*FEHB)(uint16_t *p, int size, int strength);
+typedef libaom_test::FuncParam<FEHB> FilterEdgeTestFuncsHBD;
+
+class FilterEdgeTestHB : public FilterEdgeTest<FEHB, uint16_t> {
+ protected:
+  void Execute(uint16_t *edge_tst) {
+    params_.ref_func(edge_ref_, size_, strength_);
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(edge_tst, size_, strength_));
+  }
+  int bit_depth_;
+};
+
+TEST_P(FilterEdgeTestHB, RandomValues) {
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    switch (rng_(3)) {
+      case 0: bit_depth_ = 8; break;
+      case 1: bit_depth_ = 10; break;
+      default: bit_depth_ = 12; break;
+    }
+    const int hi = 1 << bit_depth_;
+    strength_ = this->rng_(4);
+    size_ = 4 * (this->rng_(128 / 4) + 1) + 1;
+
+    int i, pix = 0;
+    for (i = 0; i < kOffset + size_; ++i) {
+      pix = rng_(hi);
+      edge_ref_data_[i] = pix;
+      edge_tst_data_[i] = pix;
+    }
+
+    Common();
+  }
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, FilterEdgeTestHB,
+                        ::testing::Values(FilterEdgeTestFuncsHBD(
+                            av1_filter_intra_edge_high_c,
+                            av1_filter_intra_edge_high_sse4_1)));
+#endif  // HAVE_SSE4_1
+
+// Speed tests
+
+TEST_P(UpsampleTest8B, DISABLED_Speed) {
+  const int test_count = 10000000;
+  size_ = kMaxEdge;
+  for (int i = 0; i < kOffset + size_; ++i) {
+    edge_tst_data_[i] = rng_.Rand8();
+  }
+  edge_tst_ = &edge_tst_data_[kOffset];
+  for (int iter = 0; iter < test_count; ++iter) {
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(edge_tst_, size_));
+  }
+}
+
+TEST_P(UpsampleTestHB, DISABLED_Speed) {
+  const int test_count = 10000000;
+  size_ = kMaxEdge;
+  bit_depth_ = 12;
+  const int hi = 1 << bit_depth_;
+  for (int i = 0; i < kOffset + size_; ++i) {
+    edge_tst_data_[i] = rng_(hi);
+  }
+  edge_tst_ = &edge_tst_data_[kOffset];
+  for (int iter = 0; iter < test_count; ++iter) {
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(edge_tst_, size_, bit_depth_));
+  }
+}
+
+TEST_P(FilterEdgeTest8B, DISABLED_Speed) {
+  const int test_count = 10000000;
+  size_ = kMaxEdge;
+  strength_ = 1;
+  for (int i = 0; i < kOffset + size_; ++i) {
+    edge_tst_data_[i] = rng_.Rand8();
+  }
+  edge_tst_ = &edge_tst_data_[kOffset];
+  for (int iter = 0; iter < test_count; ++iter) {
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(edge_tst_, size_, strength_));
+    // iterate over filter strengths (1,2,3)
+    strength_ = (strength_ == 3) ? 1 : strength_ + 1;
+  }
+}
+
+TEST_P(FilterEdgeTestHB, DISABLED_Speed) {
+  const int test_count = 10000000;
+  size_ = kMaxEdge;
+  strength_ = 1;
+  bit_depth_ = 12;
+  const int hi = 1 << bit_depth_;
+  for (int i = 0; i < kOffset + size_; ++i) {
+    edge_tst_data_[i] = rng_(hi);
+  }
+  edge_tst_ = &edge_tst_data_[kOffset];
+  for (int iter = 0; iter < test_count; ++iter) {
+    ASM_REGISTER_STATE_CHECK(params_.tst_func(edge_tst_, size_, strength_));
+    // iterate over filter strengths (1,2,3)
+    strength_ = (strength_ == 3) ? 1 : strength_ + 1;
+  }
+}
+
+}  // namespace
diff --git a/third_party/aom/test/intrabc_test.cc b/third_party/aom/test/intrabc_test.cc
new file mode 100644
index 0000000000..3ea4217081
--- /dev/null
+++ b/third_party/aom/test/intrabc_test.cc
@@ -0,0 +1,168 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#include "av1/common/blockd.h"
+#include "av1/common/enums.h"
+#include "av1/common/mv.h"
+#include "av1/common/mvref_common.h"
+#include "av1/common/onyxc_int.h"
+#include "av1/common/tile_common.h"
+
+namespace {
+TEST(IntrabcTest, DvValidation) {
+  struct DvTestCase {
+    MV dv;
+    int mi_row_offset;
+    int mi_col_offset;
+    BLOCK_SIZE bsize;
+    bool valid;
+  };
+  const int kSubPelScale = 8;
+  const int kTileMaxMibWidth = 8;
+  const DvTestCase kDvCases[] = {
+    { { 0, 0 }, 0, 0, BLOCK_128X128, false },
+    { { 0, 0 }, 0, 0, BLOCK_64X64, false },
+    { { 0, 0 }, 0, 0, BLOCK_32X32, false },
+    { { 0, 0 }, 0, 0, BLOCK_16X16, false },
+    { { 0, 0 }, 0, 0, BLOCK_8X8, false },
+    { { 0, 0 }, 0, 0, BLOCK_4X4, false },
+    { { -MAX_SB_SIZE * kSubPelScale, -MAX_SB_SIZE * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_16X16,
+      true },
+    { { 0, -MAX_SB_SIZE * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_16X16,
+      false },
+    { { -MAX_SB_SIZE * kSubPelScale, 0 },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_16X16,
+      true },
+    { { MAX_SB_SIZE * kSubPelScale, 0 },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_16X16,
+      false },
+    { { 0, MAX_SB_SIZE * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_16X16,
+      false },
+    { { -32 * kSubPelScale, -32 * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_32X32,
+      true },
+    { { -32 * kSubPelScale, -32 * kSubPelScale },
+      32 / MI_SIZE,
+      32 / MI_SIZE,
+      BLOCK_32X32,
+      false },
+    { { -32 * kSubPelScale - kSubPelScale / 2, -32 * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_32X32,
+      false },
+    { { -33 * kSubPelScale, -32 * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_32X32,
+      true },
+    { { -32 * kSubPelScale, -32 * kSubPelScale - kSubPelScale / 2 },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_32X32,
+      false },
+    { { -32 * kSubPelScale, -33 * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_32X32,
+      true },
+    { { -MAX_SB_SIZE * kSubPelScale, -MAX_SB_SIZE * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      true },
+    { { -(MAX_SB_SIZE + 1) * kSubPelScale, -MAX_SB_SIZE * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      false },
+    { { -MAX_SB_SIZE * kSubPelScale, -(MAX_SB_SIZE + 1) * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      false },
+    { { -(MAX_SB_SIZE - 1) * kSubPelScale, -MAX_SB_SIZE * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      false },
+    { { -MAX_SB_SIZE * kSubPelScale, -(MAX_SB_SIZE - 1) * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      true },
+    { { -(MAX_SB_SIZE - 1) * kSubPelScale, -(MAX_SB_SIZE - 1) * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      false },
+    { { -MAX_SB_SIZE * kSubPelScale, MAX_SB_SIZE * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      false },
+    { { -MAX_SB_SIZE * kSubPelScale,
+        (kTileMaxMibWidth - 2) * MAX_SB_SIZE * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      false },
+    { { -MAX_SB_SIZE * kSubPelScale,
+        ((kTileMaxMibWidth - 2) * MAX_SB_SIZE + 1) * kSubPelScale },
+      MAX_SB_SIZE / MI_SIZE,
+      MAX_SB_SIZE / MI_SIZE,
+      BLOCK_LARGEST,
+      false },
+  };
+
+  MACROBLOCKD xd;
+  memset(&xd, 0, sizeof(xd));
+  xd.tile.mi_row_start = 8 * MAX_MIB_SIZE;
+  xd.tile.mi_row_end = 16 * MAX_MIB_SIZE;
+  xd.tile.mi_col_start = 24 * MAX_MIB_SIZE;
+  xd.tile.mi_col_end = xd.tile.mi_col_start + kTileMaxMibWidth * MAX_MIB_SIZE;
+  xd.plane[1].subsampling_x = 1;
+  xd.plane[1].subsampling_y = 1;
+  xd.plane[2].subsampling_x = 1;
+  xd.plane[2].subsampling_y = 1;
+
+  AV1_COMMON cm;
+  memset(&cm, 0, sizeof(cm));
+
+  for (int i = 0; i < static_cast<int>(GTEST_ARRAY_SIZE_(kDvCases)); ++i) {
+    EXPECT_EQ(static_cast<int>(kDvCases[i].valid),
+              av1_is_dv_valid(kDvCases[i].dv, &cm, &xd,
+                              xd.tile.mi_row_start + kDvCases[i].mi_row_offset,
+                              xd.tile.mi_col_start + kDvCases[i].mi_col_offset,
+                              kDvCases[i].bsize, MAX_MIB_SIZE_LOG2))
+        << "DvCases[" << i << "]";
+  }
+}
+}  // namespace
diff --git a/third_party/aom/test/intrapred_test.cc b/third_party/aom/test/intrapred_test.cc
new file mode 100644
index 0000000000..1a1c0fc421
--- /dev/null
+++ b/third_party/aom/test/intrapred_test.cc
@@ -0,0 +1,266 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "av1/common/blockd.h"
+#include "av1/common/common.h"
+#include "av1/common/pred_common.h"
+#include "aom_mem/aom_mem.h"
+
+namespace {
+
+using libaom_test::ACMRandom;
+
+const int count_test_block = 100000;
+
+typedef void (*HighbdIntraPred)(uint16_t *dst, ptrdiff_t stride,
+                                const uint16_t *above, const uint16_t *left,
+                                int bps);
+typedef void (*IntraPred)(uint8_t *dst, ptrdiff_t stride, const uint8_t *above,
+                          const uint8_t *left);
+
+}  // namespace
+
+// NOTE: Under gcc version 7.3.0 (Debian 7.3.0-5), if this template is in the
+// anonymous namespace, then we get a strange compiler warning in
+// the begin() and end() methods of the ParamGenerator template class in
+// gtest/internal/gtest-param-util.h:
+//   warning: ‘<anonymous>’ is used uninitialized in this function
+// As a workaround, put this template outside the anonymous namespace.
+// See bug aomedia:2003.
+template <typename FuncType>
+struct IntraPredFunc {
+  IntraPredFunc(FuncType pred = NULL, FuncType ref = NULL,
+                int block_width_value = 0, int block_height_value = 0,
+                int bit_depth_value = 0)
+      : pred_fn(pred), ref_fn(ref), block_width(block_width_value),
+        block_height(block_height_value), bit_depth(bit_depth_value) {}
+
+  FuncType pred_fn;
+  FuncType ref_fn;
+  int block_width;
+  int block_height;
+  int bit_depth;
+};
+
+namespace {
+
+template <typename FuncType, typename Pixel>
+class AV1IntraPredTest
+    : public ::testing::TestWithParam<IntraPredFunc<FuncType> > {
+ public:
+  void RunTest(Pixel *left_col, Pixel *above_data, Pixel *dst, Pixel *ref_dst) {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int block_width = params_.block_width;
+    const int block_height = params_.block_height;
+    above_row_ = above_data + 16;
+    left_col_ = left_col;
+    dst_ = dst;
+    ref_dst_ = ref_dst;
+    int error_count = 0;
+    for (int i = 0; i < count_test_block; ++i) {
+      // Fill edges with random data, try first with saturated values.
+      for (int x = -1; x <= block_width * 2; x++) {
+        if (i == 0) {
+          above_row_[x] = mask_;
+        } else {
+          above_row_[x] = rnd.Rand16() & mask_;
+        }
+      }
+      for (int y = 0; y < block_height; y++) {
+        if (i == 0) {
+          left_col_[y] = mask_;
+        } else {
+          left_col_[y] = rnd.Rand16() & mask_;
+        }
+      }
+      Predict();
+      CheckPrediction(i, &error_count);
+    }
+    ASSERT_EQ(0, error_count);
+  }
+
+ protected:
+  virtual void SetUp() {
+    params_ = this->GetParam();
+    stride_ = params_.block_width * 3;
+    mask_ = (1 << params_.bit_depth) - 1;
+  }
+
+  virtual void Predict() = 0;
+
+  void CheckPrediction(int test_case_number, int *error_count) const {
+    // For each pixel ensure that the calculated value is the same as reference.
+    const int block_width = params_.block_width;
+    const int block_height = params_.block_height;
+    for (int y = 0; y < block_height; y++) {
+      for (int x = 0; x < block_width; x++) {
+        *error_count += ref_dst_[x + y * stride_] != dst_[x + y * stride_];
+        if (*error_count == 1) {
+          ASSERT_EQ(ref_dst_[x + y * stride_], dst_[x + y * stride_])
+              << " Failed on Test Case Number " << test_case_number
+              << " location: x = " << x << " y = " << y;
+        }
+      }
+    }
+  }
+
+  Pixel *above_row_;
+  Pixel *left_col_;
+  Pixel *dst_;
+  Pixel *ref_dst_;
+  ptrdiff_t stride_;
+  int mask_;
+
+  IntraPredFunc<FuncType> params_;
+};
+
+class HighbdIntraPredTest : public AV1IntraPredTest<HighbdIntraPred, uint16_t> {
+ protected:
+  void Predict() {
+    const int bit_depth = params_.bit_depth;
+    params_.ref_fn(ref_dst_, stride_, above_row_, left_col_, bit_depth);
+    ASM_REGISTER_STATE_CHECK(
+        params_.pred_fn(dst_, stride_, above_row_, left_col_, bit_depth));
+  }
+};
+
+class LowbdIntraPredTest : public AV1IntraPredTest<IntraPred, uint8_t> {
+ protected:
+  void Predict() {
+    params_.ref_fn(ref_dst_, stride_, above_row_, left_col_);
+    ASM_REGISTER_STATE_CHECK(
+        params_.pred_fn(dst_, stride_, above_row_, left_col_));
+  }
+};
+
+// Suppress an unitialized warning. Once there are implementations to test then
+// this can be restored.
+TEST_P(HighbdIntraPredTest, Bitexact) {
+  // max block size is 64
+  DECLARE_ALIGNED(16, uint16_t, left_col[2 * 64]);
+  DECLARE_ALIGNED(16, uint16_t, above_data[2 * 64 + 64]);
+  DECLARE_ALIGNED(16, uint16_t, dst[3 * 64 * 64]);
+  DECLARE_ALIGNED(16, uint16_t, ref_dst[3 * 64 * 64]);
+  av1_zero(left_col);
+  av1_zero(above_data);
+  RunTest(left_col, above_data, dst, ref_dst);
+}
+
+// Same issue as above but for arm.
+#if !HAVE_NEON
+TEST_P(LowbdIntraPredTest, Bitexact) {
+  // max block size is 32
+  DECLARE_ALIGNED(16, uint8_t, left_col[2 * 32]);
+  DECLARE_ALIGNED(16, uint8_t, above_data[2 * 32 + 32]);
+  DECLARE_ALIGNED(16, uint8_t, dst[3 * 32 * 32]);
+  DECLARE_ALIGNED(16, uint8_t, ref_dst[3 * 32 * 32]);
+  av1_zero(left_col);
+  av1_zero(above_data);
+  RunTest(left_col, above_data, dst, ref_dst);
+}
+#endif  // !HAVE_NEON
+
+// -----------------------------------------------------------------------------
+// High Bit Depth Tests
+#define highbd_entry(type, width, height, opt, bd)                          \
+  IntraPredFunc<HighbdIntraPred>(                                           \
+      &aom_highbd_##type##_predictor_##width##x##height##_##opt,            \
+      &aom_highbd_##type##_predictor_##width##x##height##_c, width, height, \
+      bd)
+
+#if 0
+#define highbd_intrapred(type, opt, bd)                                       \
+  highbd_entry(type, 4, 4, opt, bd), highbd_entry(type, 4, 8, opt, bd),       \
+      highbd_entry(type, 8, 4, opt, bd), highbd_entry(type, 8, 8, opt, bd),   \
+      highbd_entry(type, 8, 16, opt, bd), highbd_entry(type, 16, 8, opt, bd), \
+      highbd_entry(type, 16, 16, opt, bd),                                    \
+      highbd_entry(type, 16, 32, opt, bd),                                    \
+      highbd_entry(type, 32, 16, opt, bd), highbd_entry(type, 32, 32, opt, bd)
+#endif
+
+  // ---------------------------------------------------------------------------
+  // Low Bit Depth Tests
+
+#define lowbd_entry(type, width, height, opt)                                  \
+  IntraPredFunc<IntraPred>(&aom_##type##_predictor_##width##x##height##_##opt, \
+                           &aom_##type##_predictor_##width##x##height##_c,     \
+                           width, height, 8)
+
+#define lowbd_intrapred(type, opt)                                    \
+  lowbd_entry(type, 4, 4, opt), lowbd_entry(type, 4, 8, opt),         \
+      lowbd_entry(type, 8, 4, opt), lowbd_entry(type, 8, 8, opt),     \
+      lowbd_entry(type, 8, 16, opt), lowbd_entry(type, 16, 8, opt),   \
+      lowbd_entry(type, 16, 16, opt), lowbd_entry(type, 16, 32, opt), \
+      lowbd_entry(type, 32, 16, opt), lowbd_entry(type, 32, 32, opt)
+
+#if HAVE_SSE2
+const IntraPredFunc<IntraPred> LowbdIntraPredTestVector[] = {
+  lowbd_intrapred(dc, sse2),      lowbd_intrapred(dc_top, sse2),
+  lowbd_intrapred(dc_left, sse2), lowbd_intrapred(dc_128, sse2),
+  lowbd_intrapred(v, sse2),       lowbd_intrapred(h, sse2),
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, LowbdIntraPredTest,
+                        ::testing::ValuesIn(LowbdIntraPredTestVector));
+
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3
+const IntraPredFunc<IntraPred> LowbdIntraPredTestVectorSsse3[] = {
+  lowbd_intrapred(paeth, ssse3),
+  lowbd_intrapred(smooth, ssse3),
+};
+
+INSTANTIATE_TEST_CASE_P(SSSE3, LowbdIntraPredTest,
+                        ::testing::ValuesIn(LowbdIntraPredTestVectorSsse3));
+
+#endif  // HAVE_SSSE3
+
+#if HAVE_AVX2
+const IntraPredFunc<IntraPred> LowbdIntraPredTestVectorAvx2[] = {
+  lowbd_entry(dc, 32, 32, avx2),      lowbd_entry(dc_top, 32, 32, avx2),
+  lowbd_entry(dc_left, 32, 32, avx2), lowbd_entry(dc_128, 32, 32, avx2),
+  lowbd_entry(v, 32, 32, avx2),       lowbd_entry(h, 32, 32, avx2),
+  lowbd_entry(dc, 32, 16, avx2),      lowbd_entry(dc_top, 32, 16, avx2),
+  lowbd_entry(dc_left, 32, 16, avx2), lowbd_entry(dc_128, 32, 16, avx2),
+  lowbd_entry(v, 32, 16, avx2),       lowbd_entry(paeth, 16, 8, avx2),
+  lowbd_entry(paeth, 16, 16, avx2),   lowbd_entry(paeth, 16, 32, avx2),
+  lowbd_entry(paeth, 32, 16, avx2),   lowbd_entry(paeth, 32, 32, avx2),
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, LowbdIntraPredTest,
+                        ::testing::ValuesIn(LowbdIntraPredTestVectorAvx2));
+
+#endif  // HAVE_AVX2
+
+#if HAVE_NEON
+const IntraPredFunc<HighbdIntraPred> HighbdIntraPredTestVectorNeon[] = {
+  highbd_entry(dc, 4, 4, neon, 8),   highbd_entry(dc, 8, 8, neon, 8),
+  highbd_entry(dc, 16, 16, neon, 8), highbd_entry(dc, 32, 32, neon, 8),
+  highbd_entry(dc, 64, 64, neon, 8),
+};
+
+INSTANTIATE_TEST_CASE_P(NEON, HighbdIntraPredTest,
+                        ::testing::ValuesIn(HighbdIntraPredTestVectorNeon));
+
+#endif  // HAVE_NEON
+}  // namespace
diff --git a/third_party/aom/test/invalid_file_test.cc b/third_party/aom/test/invalid_file_test.cc
new file mode 100644
index 0000000000..5b4f5a6c3d
--- /dev/null
+++ b/third_party/aom/test/invalid_file_test.cc
@@ -0,0 +1,122 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cstdio>
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/ivf_video_source.h"
+#include "test/util.h"
+#include "test/video_source.h"
+
+namespace {
+
+struct DecodeParam {
+  int threads;
+  const char *filename;
+};
+
+std::ostream &operator<<(std::ostream &os, const DecodeParam &dp) {
+  return os << "threads: " << dp.threads << " file: " << dp.filename;
+}
+
+class InvalidFileTest : public ::libaom_test::DecoderTest,
+                        public ::libaom_test::CodecTestWithParam<DecodeParam> {
+ protected:
+  InvalidFileTest() : DecoderTest(GET_PARAM(0)), res_file_(NULL) {}
+
+  virtual ~InvalidFileTest() {
+    if (res_file_ != NULL) fclose(res_file_);
+  }
+
+  void OpenResFile(const std::string &res_file_name) {
+    res_file_ = libaom_test::OpenTestDataFile(res_file_name);
+    ASSERT_TRUE(res_file_ != NULL)
+        << "Result file open failed. Filename: " << res_file_name;
+  }
+
+  virtual bool HandleDecodeResult(
+      const aom_codec_err_t res_dec,
+      const libaom_test::CompressedVideoSource &video,
+      libaom_test::Decoder *decoder) {
+    EXPECT_TRUE(res_file_ != NULL);
+    int expected_res_dec = -1;
+
+    // Read integer result.
+    const int res = fscanf(res_file_, "%d", &expected_res_dec);
+    EXPECT_NE(res, EOF) << "Read result data failed";
+
+    if (expected_res_dec != -1) {
+      // Check results match.
+      const DecodeParam input = GET_PARAM(1);
+      if (input.threads > 1) {
+        // The serial decode check is too strict for tile-threaded decoding as
+        // there is no guarantee on the decode order nor which specific error
+        // will take precedence. Currently a tile-level error is not forwarded
+        // so the frame will simply be marked corrupt.
+        EXPECT_TRUE(res_dec == expected_res_dec ||
+                    res_dec == AOM_CODEC_CORRUPT_FRAME)
+            << "Results don't match: frame number = " << video.frame_number()
+            << ". (" << decoder->DecodeError()
+            << "). Expected: " << expected_res_dec << " or "
+            << AOM_CODEC_CORRUPT_FRAME;
+      } else {
+        EXPECT_EQ(expected_res_dec, res_dec)
+            << "Results don't match: frame number = " << video.frame_number()
+            << ". (" << decoder->DecodeError() << ")";
+      }
+    }
+
+    return !HasFailure();
+  }
+
+  virtual void HandlePeekResult(libaom_test::Decoder *const /*decoder*/,
+                                libaom_test::CompressedVideoSource * /*video*/,
+                                const aom_codec_err_t /*res_peek*/) {}
+
+  void RunTest() {
+    const DecodeParam input = GET_PARAM(1);
+    aom_codec_dec_cfg_t cfg = { 0, 0, 0, CONFIG_LOWBITDEPTH, { 1 } };
+    cfg.threads = input.threads;
+    const std::string filename = input.filename;
+    libaom_test::IVFVideoSource decode_video(filename);
+    decode_video.Init();
+
+    // Construct result file name. The file holds a list of expected integer
+    // results, one for each decoded frame.  Any result that doesn't match
+    // the files list will cause a test failure.
+    const std::string res_filename = filename + ".res";
+    OpenResFile(res_filename);
+
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&decode_video, cfg));
+  }
+
+ private:
+  FILE *res_file_;
+};
+
+TEST_P(InvalidFileTest, ReturnCode) { RunTest(); }
+
+const DecodeParam kAV1InvalidFileTests[] = {
+  { 1, "invalid-bug-1814.ivf" },
+  { 4, "invalid-oss-fuzz-9463.ivf" },
+  { 1, "invalid-oss-fuzz-9482.ivf" },
+  { 1, "invalid-oss-fuzz-9720.ivf" },
+  { 1, "invalid-oss-fuzz-10061.ivf" },
+  { 1, "invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf" },
+  { 1, "invalid-oss-fuzz-10227.ivf" },
+};
+
+AV1_INSTANTIATE_TEST_CASE(InvalidFileTest,
+                          ::testing::ValuesIn(kAV1InvalidFileTests));
+
+}  // namespace
diff --git a/third_party/aom/test/ivf_video_source.h b/third_party/aom/test/ivf_video_source.h
new file mode 100644
index 0000000000..ff2841445e
--- /dev/null
+++ b/third_party/aom/test/ivf_video_source.h
@@ -0,0 +1,114 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_IVF_VIDEO_SOURCE_H_
+#define AOM_TEST_IVF_VIDEO_SOURCE_H_
+
+#include <cstdio>
+#include <cstdlib>
+#include <new>
+#include <string>
+
+#include "aom_ports/sanitizer.h"
+#include "test/video_source.h"
+
+namespace libaom_test {
+const unsigned int kCodeBufferSize = 256 * 1024 * 1024;
+const unsigned int kIvfFileHdrSize = 32;
+const unsigned int kIvfFrameHdrSize = 12;
+
+static unsigned int MemGetLe32(const uint8_t *mem) {
+  return (mem[3] << 24) | (mem[2] << 16) | (mem[1] << 8) | (mem[0]);
+}
+
+// This class extends VideoSource to allow parsing of ivf files,
+// so that we can do actual file decodes.
+class IVFVideoSource : public CompressedVideoSource {
+ public:
+  explicit IVFVideoSource(const std::string &file_name)
+      : file_name_(file_name), input_file_(NULL), compressed_frame_buf_(NULL),
+        frame_sz_(0), frame_(0), end_of_file_(false) {}
+
+  virtual ~IVFVideoSource() {
+    delete[] compressed_frame_buf_;
+
+    if (input_file_) fclose(input_file_);
+  }
+
+  virtual void Init() {
+    // Allocate a buffer for read in the compressed video frame.
+    compressed_frame_buf_ = new uint8_t[kCodeBufferSize];
+    ASSERT_TRUE(compressed_frame_buf_ != NULL)
+        << "Allocate frame buffer failed";
+    ASAN_POISON_MEMORY_REGION(compressed_frame_buf_, kCodeBufferSize);
+  }
+
+  virtual void Begin() {
+    input_file_ = OpenTestDataFile(file_name_);
+    ASSERT_TRUE(input_file_ != NULL)
+        << "Input file open failed. Filename: " << file_name_;
+
+    // Read file header
+    uint8_t file_hdr[kIvfFileHdrSize];
+    ASSERT_EQ(kIvfFileHdrSize, fread(file_hdr, 1, kIvfFileHdrSize, input_file_))
+        << "File header read failed.";
+    // Check file header
+    ASSERT_TRUE(file_hdr[0] == 'D' && file_hdr[1] == 'K' &&
+                file_hdr[2] == 'I' && file_hdr[3] == 'F')
+        << "Input is not an IVF file.";
+
+    FillFrame();
+  }
+
+  virtual void Next() {
+    ++frame_;
+    FillFrame();
+  }
+
+  void FillFrame() {
+    ASSERT_TRUE(input_file_ != NULL);
+    uint8_t frame_hdr[kIvfFrameHdrSize];
+    // Check frame header and read a frame from input_file.
+    if (fread(frame_hdr, 1, kIvfFrameHdrSize, input_file_) !=
+        kIvfFrameHdrSize) {
+      end_of_file_ = true;
+    } else {
+      end_of_file_ = false;
+
+      frame_sz_ = MemGetLe32(frame_hdr);
+      ASSERT_LE(frame_sz_, kCodeBufferSize)
+          << "Frame is too big for allocated code buffer";
+      ASAN_UNPOISON_MEMORY_REGION(compressed_frame_buf_, kCodeBufferSize);
+      ASSERT_EQ(frame_sz_,
+                fread(compressed_frame_buf_, 1, frame_sz_, input_file_))
+          << "Failed to read complete frame";
+      ASAN_POISON_MEMORY_REGION(compressed_frame_buf_ + frame_sz_,
+                                kCodeBufferSize - frame_sz_);
+    }
+  }
+
+  virtual const uint8_t *cxdata() const {
+    return end_of_file_ ? NULL : compressed_frame_buf_;
+  }
+  virtual size_t frame_size() const { return frame_sz_; }
+  virtual unsigned int frame_number() const { return frame_; }
+
+ protected:
+  std::string file_name_;
+  FILE *input_file_;
+  uint8_t *compressed_frame_buf_;
+  size_t frame_sz_;
+  unsigned int frame_;
+  bool end_of_file_;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_IVF_VIDEO_SOURCE_H_
diff --git a/third_party/aom/test/lightfield_test.sh b/third_party/aom/test/lightfield_test.sh
new file mode 100755
index 0000000000..b957a6b794
--- /dev/null
+++ b/third_party/aom/test/lightfield_test.sh
@@ -0,0 +1,98 @@
+#!/bin/sh
+## Copyright (c) 2018, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the lightfield example.
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: $infile is required.
+lightfield_test_verify_environment() {
+  local infile="${LIBAOM_TEST_DATA_PATH}/vase10x10.yuv"
+  if [ ! -e "${infile}" ]; then
+    echo "Libaom test data must exist in LIBAOM_TEST_DATA_PATH."
+    return 1
+  fi
+}
+
+# Run the lightfield example
+lightfield_test() {
+  local img_width=1024
+  local img_height=1024
+  local lf_width=10
+  local lf_height=10
+  local lf_blocksize=5
+  local num_references=4
+  local num_tile_lists=2
+
+  # Encode the lightfield.
+  local encoder="${LIBAOM_BIN_PATH}/lightfield_encoder${AOM_TEST_EXE_SUFFIX}"
+  local yuv_file="${LIBAOM_TEST_DATA_PATH}/vase10x10.yuv"
+  local lf_file="${AOM_TEST_OUTPUT_DIR}/vase10x10.ivf"
+  if [ ! -x "${encoder}" ]; then
+    elog "${encoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${encoder}" "${img_width}" "${img_height}" \
+      "${yuv_file}" "${lf_file}" "${lf_width}" \
+      "${lf_height}" "${lf_blocksize}" ${devnull}
+
+  [ -e "${lf_file}" ] || return 1
+
+  # Parse lightfield bitstream to construct and output a new bitstream that can
+  # be decoded by an AV1 decoder.
+  local bs_decoder="${LIBAOM_BIN_PATH}/lightfield_bitstream_parsing${AOM_TEST_EXE_SUFFIX}"
+  local tl_file="${AOM_TEST_OUTPUT_DIR}/vase_tile_list.ivf"
+  if [ ! -x "${bs_decoder}" ]; then
+    elog "${bs_decoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${bs_decoder}" "${lf_file}" "${tl_file}" \
+      "${num_references}" ${devnull}
+
+  [ -e "${tl_file}" ] || return 1
+
+  # Run lightfield tile list decoder
+  local tl_decoder="${LIBAOM_BIN_PATH}/lightfield_tile_list_decoder${AOM_TEST_EXE_SUFFIX}"
+  local tl_outfile="${AOM_TEST_OUTPUT_DIR}/vase_tile_list.yuv"
+  if [ ! -x "${tl_decoder}" ]; then
+    elog "${tl_decoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${tl_decoder}" "${tl_file}" "${tl_outfile}" \
+      "${num_references}" "${num_tile_lists}" ${devnull}
+
+  [ -e "${tl_outfile}" ] || return 1
+
+  # Run reference lightfield decoder
+  local ref_decoder="${LIBAOM_BIN_PATH}/lightfield_decoder${AOM_TEST_EXE_SUFFIX}"
+  local tl_reffile="${AOM_TEST_OUTPUT_DIR}/vase_reference.yuv"
+  if [ ! -x "${ref_decoder}" ]; then
+    elog "${ref_decoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${ref_decoder}" "${lf_file}" "${tl_reffile}" \
+      "${num_references}" ${devnull}
+
+  [ -e "${tl_reffile}" ] || return 1
+
+  # Check if tl_outfile and tl_reffile are identical. If not identical, this test fails.
+  diff ${tl_outfile} ${tl_reffile} > /dev/null
+  if [ $? -eq 1 ]; then
+    return 1
+  fi
+}
+
+lightfield_test_tests="lightfield_test"
+
+run_tests lightfield_test_verify_environment "${lightfield_test_tests}"
diff --git a/third_party/aom/test/log2_test.cc b/third_party/aom/test/log2_test.cc
new file mode 100644
index 0000000000..d7840c68b2
--- /dev/null
+++ b/third_party/aom/test/log2_test.cc
@@ -0,0 +1,50 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+
+#include "aom_ports/bitops.h"
+#include "av1/common/entropymode.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+TEST(Log2Test, GetMsb) {
+  // Test small numbers exhaustively.
+  for (unsigned int n = 1; n < 10000; n++) {
+    EXPECT_EQ(get_msb(n), static_cast<int>(floor(log2(n))));
+  }
+
+  // Test every power of 2 and the two adjacent numbers.
+  for (int exponent = 2; exponent < 32; exponent++) {
+    const unsigned int power_of_2 = 1U << exponent;
+    EXPECT_EQ(get_msb(power_of_2 - 1), exponent - 1);
+    EXPECT_EQ(get_msb(power_of_2), exponent);
+    EXPECT_EQ(get_msb(power_of_2 + 1), exponent);
+  }
+}
+
+TEST(Log2Test, Av1CeilLog2) {
+  // Test small numbers exhaustively.
+  EXPECT_EQ(av1_ceil_log2(0), 0);
+  for (int n = 1; n < 10000; n++) {
+    EXPECT_EQ(av1_ceil_log2(n), static_cast<int>(ceil(log2(n))));
+  }
+
+  // Test every power of 2 and the two adjacent numbers.
+  for (int exponent = 2; exponent < 31; exponent++) {
+    const int power_of_2 = 1 << exponent;
+    EXPECT_EQ(av1_ceil_log2(power_of_2 - 1), exponent);
+    EXPECT_EQ(av1_ceil_log2(power_of_2), exponent);
+    // The current implementation of av1_ceil_log2 only works up to 2^30.
+    if (exponent < 30) {
+      EXPECT_EQ(av1_ceil_log2(power_of_2 + 1), exponent + 1);
+    }
+  }
+}
diff --git a/third_party/aom/test/lossless_test.cc b/third_party/aom/test/lossless_test.cc
new file mode 100644
index 0000000000..3f8e89c815
--- /dev/null
+++ b/third_party/aom/test/lossless_test.cc
@@ -0,0 +1,126 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+
+namespace {
+
+const int kMaxPsnr = 100;
+
+class LosslessTestLarge
+    : public ::libaom_test::CodecTestWithParam<libaom_test::TestMode>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  LosslessTestLarge()
+      : EncoderTest(GET_PARAM(0)), psnr_(kMaxPsnr), nframes_(0),
+        encoding_mode_(GET_PARAM(1)) {}
+
+  virtual ~LosslessTestLarge() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      // Only call Control if quantizer > 0 to verify that using quantizer
+      // alone will activate lossless
+      if (cfg_.rc_max_quantizer > 0 || cfg_.rc_min_quantizer > 0) {
+        encoder->Control(AV1E_SET_LOSSLESS, 1);
+      }
+    }
+  }
+
+  virtual void BeginPassHook(unsigned int /*pass*/) {
+    psnr_ = kMaxPsnr;
+    nframes_ = 0;
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (pkt->data.psnr.psnr[0] < psnr_) psnr_ = pkt->data.psnr.psnr[0];
+  }
+
+  double GetMinPsnr() const { return psnr_; }
+
+ private:
+  double psnr_;
+  unsigned int nframes_;
+  libaom_test::TestMode encoding_mode_;
+};
+
+TEST_P(LosslessTestLarge, TestLossLessEncoding) {
+  const aom_rational timebase = { 33333333, 1000000000 };
+  cfg_.g_timebase = timebase;
+  cfg_.rc_target_bitrate = 2000;
+  cfg_.g_lag_in_frames = 25;
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 0;
+
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  // intentionally changed the dimension for better testing coverage
+  libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                     timebase.den, timebase.num, 0, 5);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  const double psnr_lossless = GetMinPsnr();
+  EXPECT_GE(psnr_lossless, kMaxPsnr);
+}
+
+TEST_P(LosslessTestLarge, TestLossLessEncoding444) {
+  libaom_test::Y4mVideoSource video("rush_hour_444.y4m", 0, 5);
+
+  cfg_.g_profile = 1;
+  cfg_.g_timebase = video.timebase();
+  cfg_.rc_target_bitrate = 2000;
+  cfg_.g_lag_in_frames = 25;
+  cfg_.rc_min_quantizer = 0;
+  cfg_.rc_max_quantizer = 0;
+
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  const double psnr_lossless = GetMinPsnr();
+  EXPECT_GE(psnr_lossless, kMaxPsnr);
+}
+
+TEST_P(LosslessTestLarge, TestLossLessEncodingCtrl) {
+  const aom_rational timebase = { 33333333, 1000000000 };
+  cfg_.g_timebase = timebase;
+  cfg_.rc_target_bitrate = 2000;
+  cfg_.g_lag_in_frames = 25;
+  // Intentionally set Q > 0, to make sure control can be used to activate
+  // lossless
+  cfg_.rc_min_quantizer = 10;
+  cfg_.rc_max_quantizer = 20;
+
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                     timebase.den, timebase.num, 0, 5);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  const double psnr_lossless = GetMinPsnr();
+  EXPECT_GE(psnr_lossless, kMaxPsnr);
+}
+
+AV1_INSTANTIATE_TEST_CASE(LosslessTestLarge,
+                          ::testing::Values(::libaom_test::kOnePassGood,
+                                            ::libaom_test::kTwoPassGood));
+}  // namespace
diff --git a/third_party/aom/test/lpf_test.cc b/third_party/aom/test/lpf_test.cc
new file mode 100644
index 0000000000..451bffd2a2
--- /dev/null
+++ b/third_party/aom/test/lpf_test.cc
@@ -0,0 +1,627 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cmath>
+#include <cstdlib>
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "av1/common/av1_loopfilter.h"
+#include "av1/common/entropy.h"
+#include "aom/aom_integer.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+// Horizontally and Vertically need 32x32: 8  Coeffs preceeding filtered section
+//                                         16 Coefs within filtered section
+//                                         8  Coeffs following filtered section
+const int kNumCoeffs = 1024;
+
+const int number_of_iterations = 10000;
+
+const int kSpeedTestNum = 500000;
+
+#define LOOP_PARAM \
+  int p, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh
+#define DUAL_LOOP_PARAM                                                      \
+  int p, const uint8_t *blimit0, const uint8_t *limit0,                      \
+      const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, \
+      const uint8_t *thresh1
+
+typedef void (*loop_op_t)(uint8_t *s, LOOP_PARAM);
+typedef void (*dual_loop_op_t)(uint8_t *s, DUAL_LOOP_PARAM);
+typedef void (*hbdloop_op_t)(uint16_t *s, LOOP_PARAM, int bd);
+typedef void (*hbddual_loop_op_t)(uint16_t *s, DUAL_LOOP_PARAM, int bd);
+
+typedef ::testing::tuple<hbdloop_op_t, hbdloop_op_t, int> hbdloop_param_t;
+typedef ::testing::tuple<hbddual_loop_op_t, hbddual_loop_op_t, int>
+    hbddual_loop_param_t;
+typedef ::testing::tuple<loop_op_t, loop_op_t, int> loop_param_t;
+typedef ::testing::tuple<dual_loop_op_t, dual_loop_op_t, int> dual_loop_param_t;
+
+template <typename Pixel_t, int PIXEL_WIDTH_t>
+void InitInput(Pixel_t *s, Pixel_t *ref_s, ACMRandom *rnd, const uint8_t limit,
+               const int mask, const int32_t p, const int i) {
+  uint16_t tmp_s[kNumCoeffs];
+
+  for (int j = 0; j < kNumCoeffs;) {
+    const uint8_t val = rnd->Rand8();
+    if (val & 0x80) {  // 50% chance to choose a new value.
+      tmp_s[j] = rnd->Rand16();
+      j++;
+    } else {  // 50% chance to repeat previous value in row X times.
+      int k = 0;
+      while (k++ < ((val & 0x1f) + 1) && j < kNumCoeffs) {
+        if (j < 1) {
+          tmp_s[j] = rnd->Rand16();
+        } else if (val & 0x20) {  // Increment by a value within the limit.
+          tmp_s[j] = tmp_s[j - 1] + (limit - 1);
+        } else {  // Decrement by a value within the limit.
+          tmp_s[j] = tmp_s[j - 1] - (limit - 1);
+        }
+        j++;
+      }
+    }
+  }
+
+  for (int j = 0; j < kNumCoeffs;) {
+    const uint8_t val = rnd->Rand8();
+    if (val & 0x80) {
+      j++;
+    } else {  // 50% chance to repeat previous value in column X times.
+      int k = 0;
+      while (k++ < ((val & 0x1f) + 1) && j < kNumCoeffs) {
+        if (j < 1) {
+          tmp_s[j] = rnd->Rand16();
+        } else if (val & 0x20) {  // Increment by a value within the limit.
+          tmp_s[(j % 32) * 32 + j / 32] =
+              tmp_s[((j - 1) % 32) * 32 + (j - 1) / 32] + (limit - 1);
+        } else {  // Decrement by a value within the limit.
+          tmp_s[(j % 32) * 32 + j / 32] =
+              tmp_s[((j - 1) % 32) * 32 + (j - 1) / 32] - (limit - 1);
+        }
+        j++;
+      }
+    }
+  }
+
+  for (int j = 0; j < kNumCoeffs; j++) {
+    if (i % 2) {
+      s[j] = tmp_s[j] & mask;
+    } else {
+      s[j] = tmp_s[p * (j % p) + j / p] & mask;
+    }
+    ref_s[j] = s[j];
+  }
+}
+
+uint8_t GetOuterThresh(ACMRandom *rnd) {
+  return static_cast<uint8_t>(rnd->PseudoUniform(3 * MAX_LOOP_FILTER + 5));
+}
+
+uint8_t GetInnerThresh(ACMRandom *rnd) {
+  return static_cast<uint8_t>(rnd->PseudoUniform(MAX_LOOP_FILTER + 1));
+}
+
+uint8_t GetHevThresh(ACMRandom *rnd) {
+  return static_cast<uint8_t>(rnd->PseudoUniform(MAX_LOOP_FILTER + 1) >> 4);
+}
+
+template <typename func_type_t, typename params_t>
+class LoopTestParam : public ::testing::TestWithParam<params_t> {
+ public:
+  virtual ~LoopTestParam() {}
+  virtual void SetUp() {
+    loopfilter_op_ = ::testing::get<0>(this->GetParam());
+    ref_loopfilter_op_ = ::testing::get<1>(this->GetParam());
+    bit_depth_ = ::testing::get<2>(this->GetParam());
+    mask_ = (1 << bit_depth_) - 1;
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  int bit_depth_;
+  int mask_;
+  func_type_t loopfilter_op_;
+  func_type_t ref_loopfilter_op_;
+};
+
+void call_filter(uint16_t *s, LOOP_PARAM, int bd, hbdloop_op_t op) {
+  op(s, p, blimit, limit, thresh, bd);
+}
+void call_filter(uint8_t *s, LOOP_PARAM, int bd, loop_op_t op) {
+  (void)bd;
+  op(s, p, blimit, limit, thresh);
+}
+void call_dualfilter(uint16_t *s, DUAL_LOOP_PARAM, int bd,
+                     hbddual_loop_op_t op) {
+  op(s, p, blimit0, limit0, thresh0, blimit1, limit1, thresh1, bd);
+}
+void call_dualfilter(uint8_t *s, DUAL_LOOP_PARAM, int bd, dual_loop_op_t op) {
+  (void)bd;
+  op(s, p, blimit0, limit0, thresh0, blimit1, limit1, thresh1);
+};
+
+typedef LoopTestParam<hbdloop_op_t, hbdloop_param_t> Loop8Test6Param_hbd;
+typedef LoopTestParam<loop_op_t, loop_param_t> Loop8Test6Param_lbd;
+typedef LoopTestParam<hbddual_loop_op_t, hbddual_loop_param_t>
+    Loop8Test9Param_hbd;
+typedef LoopTestParam<dual_loop_op_t, dual_loop_param_t> Loop8Test9Param_lbd;
+
+#define OPCHECK(a, b)                                                          \
+  ACMRandom rnd(ACMRandom::DeterministicSeed());                               \
+  const int count_test_block = number_of_iterations;                           \
+  const int32_t p = kNumCoeffs / 32;                                           \
+  DECLARE_ALIGNED(b, a, s[kNumCoeffs]);                                        \
+  DECLARE_ALIGNED(b, a, ref_s[kNumCoeffs]);                                    \
+  int err_count_total = 0;                                                     \
+  int first_failure = -1;                                                      \
+  for (int i = 0; i < count_test_block; ++i) {                                 \
+    int err_count = 0;                                                         \
+    uint8_t tmp = GetOuterThresh(&rnd);                                        \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    blimit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+    tmp = GetInnerThresh(&rnd);                                                \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    limit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,     \
+                                   tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };   \
+    tmp = GetHevThresh(&rnd);                                                  \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    thresh[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+    InitInput<a, b>(s, ref_s, &rnd, *limit, mask_, p, i);                      \
+    call_filter(ref_s + 8 + p * 8, p, blimit, limit, thresh, bit_depth_,       \
+                ref_loopfilter_op_);                                           \
+    ASM_REGISTER_STATE_CHECK(call_filter(s + 8 + p * 8, p, blimit, limit,      \
+                                         thresh, bit_depth_, loopfilter_op_)); \
+    for (int j = 0; j < kNumCoeffs; ++j) {                                     \
+      err_count += ref_s[j] != s[j];                                           \
+    }                                                                          \
+    if (err_count && !err_count_total) {                                       \
+      first_failure = i;                                                       \
+    }                                                                          \
+    err_count_total += err_count;                                              \
+  }                                                                            \
+  EXPECT_EQ(0, err_count_total)                                                \
+      << "Error: Loop8Test6Param, C output doesn't match SIMD "                \
+         "loopfilter output. "                                                 \
+      << "First failed at test case " << first_failure;
+
+TEST_P(Loop8Test6Param_hbd, OperationCheck) { OPCHECK(uint16_t, 16); }
+TEST_P(Loop8Test6Param_lbd, OperationCheck) { OPCHECK(uint8_t, 8); }
+
+#define VALCHECK(a, b)                                                         \
+  ACMRandom rnd(ACMRandom::DeterministicSeed());                               \
+  const int count_test_block = number_of_iterations;                           \
+  DECLARE_ALIGNED(b, a, s[kNumCoeffs]);                                        \
+  DECLARE_ALIGNED(b, a, ref_s[kNumCoeffs]);                                    \
+  int err_count_total = 0;                                                     \
+  int first_failure = -1;                                                      \
+  for (int i = 0; i < count_test_block; ++i) {                                 \
+    int err_count = 0;                                                         \
+    uint8_t tmp = GetOuterThresh(&rnd);                                        \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    blimit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+    tmp = GetInnerThresh(&rnd);                                                \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    limit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,     \
+                                   tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };   \
+    tmp = GetHevThresh(&rnd);                                                  \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    thresh[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+    int32_t p = kNumCoeffs / 32;                                               \
+    for (int j = 0; j < kNumCoeffs; ++j) {                                     \
+      s[j] = rnd.Rand16() & mask_;                                             \
+      ref_s[j] = s[j];                                                         \
+    }                                                                          \
+    call_filter(ref_s + 8 + p * 8, p, blimit, limit, thresh, bit_depth_,       \
+                ref_loopfilter_op_);                                           \
+    ASM_REGISTER_STATE_CHECK(call_filter(s + 8 + p * 8, p, blimit, limit,      \
+                                         thresh, bit_depth_, loopfilter_op_)); \
+    for (int j = 0; j < kNumCoeffs; ++j) {                                     \
+      err_count += ref_s[j] != s[j];                                           \
+    }                                                                          \
+    if (err_count && !err_count_total) {                                       \
+      first_failure = i;                                                       \
+    }                                                                          \
+    err_count_total += err_count;                                              \
+  }                                                                            \
+  EXPECT_EQ(0, err_count_total)                                                \
+      << "Error: Loop8Test6Param, C output doesn't match SIMD "                \
+         "loopfilter output. "                                                 \
+      << "First failed at test case " << first_failure;
+
+TEST_P(Loop8Test6Param_hbd, ValueCheck) { VALCHECK(uint16_t, 16); }
+TEST_P(Loop8Test6Param_lbd, ValueCheck) { VALCHECK(uint8_t, 8); }
+
+#define SPEEDCHECK(a, b)                                                      \
+  ACMRandom rnd(ACMRandom::DeterministicSeed());                              \
+  const int count_test_block = kSpeedTestNum;                                 \
+  const int32_t bd = bit_depth_;                                              \
+  DECLARE_ALIGNED(b, a, s[kNumCoeffs]);                                       \
+  uint8_t tmp = GetOuterThresh(&rnd);                                         \
+  DECLARE_ALIGNED(16, const uint8_t,                                          \
+                  blimit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,     \
+                                  tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };   \
+  tmp = GetInnerThresh(&rnd);                                                 \
+  DECLARE_ALIGNED(16, const uint8_t,                                          \
+                  limit[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,      \
+                                 tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };    \
+  tmp = GetHevThresh(&rnd);                                                   \
+  DECLARE_ALIGNED(16, const uint8_t,                                          \
+                  thresh[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,     \
+                                  tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };   \
+  int32_t p = kNumCoeffs / 32;                                                \
+  for (int j = 0; j < kNumCoeffs; ++j) {                                      \
+    s[j] = rnd.Rand16() & mask_;                                              \
+  }                                                                           \
+  for (int i = 0; i < count_test_block; ++i) {                                \
+    call_filter(s + 8 + p * 8, p, blimit, limit, thresh, bd, loopfilter_op_); \
+  }
+
+TEST_P(Loop8Test6Param_hbd, DISABLED_Speed) { SPEEDCHECK(uint16_t, 16); }
+TEST_P(Loop8Test6Param_lbd, DISABLED_Speed) { SPEEDCHECK(uint8_t, 8); }
+
+#define OPCHECKd(a, b)                                                         \
+  ACMRandom rnd(ACMRandom::DeterministicSeed());                               \
+  const int count_test_block = number_of_iterations;                           \
+  DECLARE_ALIGNED(b, a, s[kNumCoeffs]);                                        \
+  DECLARE_ALIGNED(b, a, ref_s[kNumCoeffs]);                                    \
+  int err_count_total = 0;                                                     \
+  int first_failure = -1;                                                      \
+  for (int i = 0; i < count_test_block; ++i) {                                 \
+    int err_count = 0;                                                         \
+    uint8_t tmp = GetOuterThresh(&rnd);                                        \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    blimit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+    tmp = GetInnerThresh(&rnd);                                                \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    limit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+    tmp = GetHevThresh(&rnd);                                                  \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    thresh0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+    tmp = GetOuterThresh(&rnd);                                                \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    blimit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+    tmp = GetInnerThresh(&rnd);                                                \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    limit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+    tmp = GetHevThresh(&rnd);                                                  \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    thresh1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+    int32_t p = kNumCoeffs / 32;                                               \
+    const uint8_t limit = *limit0 < *limit1 ? *limit0 : *limit1;               \
+    InitInput<a, b>(s, ref_s, &rnd, limit, mask_, p, i);                       \
+    call_dualfilter(ref_s + 8 + p * 8, p, blimit0, limit0, thresh0, blimit1,   \
+                    limit1, thresh1, bit_depth_, ref_loopfilter_op_);          \
+    ASM_REGISTER_STATE_CHECK(                                                  \
+        call_dualfilter(s + 8 + p * 8, p, blimit0, limit0, thresh0, blimit1,   \
+                        limit1, thresh1, bit_depth_, loopfilter_op_));         \
+    for (int j = 0; j < kNumCoeffs; ++j) {                                     \
+      err_count += ref_s[j] != s[j];                                           \
+    }                                                                          \
+    if (err_count && !err_count_total) {                                       \
+      first_failure = i;                                                       \
+    }                                                                          \
+    err_count_total += err_count;                                              \
+  }                                                                            \
+  EXPECT_EQ(0, err_count_total)                                                \
+      << "Error: Loop8Test9Param, C output doesn't match SIMD "                \
+         "loopfilter output. "                                                 \
+      << "First failed at test case " << first_failure;
+
+TEST_P(Loop8Test9Param_hbd, OperationCheck) { OPCHECKd(uint16_t, 16); }
+TEST_P(Loop8Test9Param_lbd, OperationCheck) { OPCHECKd(uint8_t, 8); }
+
+#define VALCHECKd(a, b)                                                        \
+  ACMRandom rnd(ACMRandom::DeterministicSeed());                               \
+  const int count_test_block = number_of_iterations;                           \
+  DECLARE_ALIGNED(b, a, s[kNumCoeffs]);                                        \
+  DECLARE_ALIGNED(b, a, ref_s[kNumCoeffs]);                                    \
+  int err_count_total = 0;                                                     \
+  int first_failure = -1;                                                      \
+  for (int i = 0; i < count_test_block; ++i) {                                 \
+    int err_count = 0;                                                         \
+    uint8_t tmp = GetOuterThresh(&rnd);                                        \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    blimit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+    tmp = GetInnerThresh(&rnd);                                                \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    limit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+    tmp = GetHevThresh(&rnd);                                                  \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    thresh0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+    tmp = GetOuterThresh(&rnd);                                                \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    blimit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+    tmp = GetInnerThresh(&rnd);                                                \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    limit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                    tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+    tmp = GetHevThresh(&rnd);                                                  \
+    DECLARE_ALIGNED(16, const uint8_t,                                         \
+                    thresh1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                     tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+    int32_t p = kNumCoeffs / 32;                                               \
+    for (int j = 0; j < kNumCoeffs; ++j) {                                     \
+      s[j] = rnd.Rand16() & mask_;                                             \
+      ref_s[j] = s[j];                                                         \
+    }                                                                          \
+    call_dualfilter(ref_s + 8 + p * 8, p, blimit0, limit0, thresh0, blimit1,   \
+                    limit1, thresh1, bit_depth_, ref_loopfilter_op_);          \
+    ASM_REGISTER_STATE_CHECK(                                                  \
+        call_dualfilter(s + 8 + p * 8, p, blimit0, limit0, thresh0, blimit1,   \
+                        limit1, thresh1, bit_depth_, loopfilter_op_));         \
+    for (int j = 0; j < kNumCoeffs; ++j) {                                     \
+      err_count += ref_s[j] != s[j];                                           \
+    }                                                                          \
+    if (err_count && !err_count_total) {                                       \
+      first_failure = i;                                                       \
+    }                                                                          \
+    err_count_total += err_count;                                              \
+  }                                                                            \
+  EXPECT_EQ(0, err_count_total)                                                \
+      << "Error: Loop8Test9Param, C output doesn't match SIMD "                \
+         "loopfilter output. "                                                 \
+      << "First failed at test case " << first_failure;
+
+TEST_P(Loop8Test9Param_hbd, ValueCheck) { VALCHECKd(uint16_t, 16); }
+TEST_P(Loop8Test9Param_lbd, ValueCheck) { VALCHECKd(uint8_t, 8); }
+
+#define SPEEDCHECKd(a, b)                                                    \
+  ACMRandom rnd(ACMRandom::DeterministicSeed());                             \
+  const int count_test_block = kSpeedTestNum;                                \
+  DECLARE_ALIGNED(b, a, s[kNumCoeffs]);                                      \
+  uint8_t tmp = GetOuterThresh(&rnd);                                        \
+  DECLARE_ALIGNED(16, const uint8_t,                                         \
+                  blimit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                   tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+  tmp = GetInnerThresh(&rnd);                                                \
+  DECLARE_ALIGNED(16, const uint8_t,                                         \
+                  limit0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                  tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+  tmp = GetHevThresh(&rnd);                                                  \
+  DECLARE_ALIGNED(16, const uint8_t,                                         \
+                  thresh0[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                   tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+  tmp = GetOuterThresh(&rnd);                                                \
+  DECLARE_ALIGNED(16, const uint8_t,                                         \
+                  blimit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                   tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+  tmp = GetInnerThresh(&rnd);                                                \
+  DECLARE_ALIGNED(16, const uint8_t,                                         \
+                  limit1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,    \
+                                  tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp };  \
+  tmp = GetHevThresh(&rnd);                                                  \
+  DECLARE_ALIGNED(16, const uint8_t,                                         \
+                  thresh1[16]) = { tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp,   \
+                                   tmp, tmp, tmp, tmp, tmp, tmp, tmp, tmp }; \
+  int32_t p = kNumCoeffs / 32;                                               \
+  for (int j = 0; j < kNumCoeffs; ++j) {                                     \
+    s[j] = rnd.Rand16() & mask_;                                             \
+  }                                                                          \
+  for (int i = 0; i < count_test_block; ++i) {                               \
+    call_dualfilter(s + 8 + p * 8, p, blimit0, limit0, thresh0, blimit1,     \
+                    limit1, thresh1, bit_depth_, loopfilter_op_);            \
+  }
+
+TEST_P(Loop8Test9Param_hbd, DISABLED_Speed) { SPEEDCHECKd(uint16_t, 16); }
+TEST_P(Loop8Test9Param_lbd, DISABLED_Speed) { SPEEDCHECKd(uint8_t, 8); }
+
+using ::testing::make_tuple;
+
+#if HAVE_SSE2
+
+const hbdloop_param_t kHbdLoop8Test6[] = {
+  make_tuple(&aom_highbd_lpf_horizontal_4_sse2, &aom_highbd_lpf_horizontal_4_c,
+             8),
+  make_tuple(&aom_highbd_lpf_vertical_4_sse2, &aom_highbd_lpf_vertical_4_c, 8),
+  make_tuple(&aom_highbd_lpf_horizontal_6_sse2, &aom_highbd_lpf_horizontal_6_c,
+             8),
+  make_tuple(&aom_highbd_lpf_horizontal_8_sse2, &aom_highbd_lpf_horizontal_8_c,
+             8),
+  make_tuple(&aom_highbd_lpf_horizontal_14_sse2,
+             &aom_highbd_lpf_horizontal_14_c, 8),
+  make_tuple(&aom_highbd_lpf_vertical_6_sse2, &aom_highbd_lpf_vertical_6_c, 8),
+  make_tuple(&aom_highbd_lpf_vertical_8_sse2, &aom_highbd_lpf_vertical_8_c, 8),
+
+  make_tuple(&aom_highbd_lpf_vertical_14_sse2, &aom_highbd_lpf_vertical_14_c,
+             8),
+  make_tuple(&aom_highbd_lpf_horizontal_4_sse2, &aom_highbd_lpf_horizontal_4_c,
+             10),
+  make_tuple(&aom_highbd_lpf_vertical_4_sse2, &aom_highbd_lpf_vertical_4_c, 10),
+  make_tuple(&aom_highbd_lpf_horizontal_6_sse2, &aom_highbd_lpf_horizontal_6_c,
+             10),
+  make_tuple(&aom_highbd_lpf_horizontal_8_sse2, &aom_highbd_lpf_horizontal_8_c,
+             10),
+  make_tuple(&aom_highbd_lpf_horizontal_14_sse2,
+             &aom_highbd_lpf_horizontal_14_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_6_sse2, &aom_highbd_lpf_vertical_6_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_8_sse2, &aom_highbd_lpf_vertical_8_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_14_sse2, &aom_highbd_lpf_vertical_14_c,
+             10),
+  make_tuple(&aom_highbd_lpf_horizontal_4_sse2, &aom_highbd_lpf_horizontal_4_c,
+             12),
+  make_tuple(&aom_highbd_lpf_vertical_4_sse2, &aom_highbd_lpf_vertical_4_c, 12),
+  make_tuple(&aom_highbd_lpf_horizontal_6_sse2, &aom_highbd_lpf_horizontal_6_c,
+             12),
+  make_tuple(&aom_highbd_lpf_horizontal_8_sse2, &aom_highbd_lpf_horizontal_8_c,
+             12),
+  make_tuple(&aom_highbd_lpf_horizontal_14_sse2,
+             &aom_highbd_lpf_horizontal_14_c, 12),
+  make_tuple(&aom_highbd_lpf_vertical_14_sse2, &aom_highbd_lpf_vertical_14_c,
+             12),
+  make_tuple(&aom_highbd_lpf_vertical_6_sse2, &aom_highbd_lpf_vertical_6_c, 12),
+  make_tuple(&aom_highbd_lpf_vertical_8_sse2, &aom_highbd_lpf_vertical_8_c, 12)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, Loop8Test6Param_hbd,
+                        ::testing::ValuesIn(kHbdLoop8Test6));
+
+const loop_param_t kLoop8Test6[] = {
+  make_tuple(&aom_lpf_horizontal_4_sse2, &aom_lpf_horizontal_4_c, 8),
+  make_tuple(&aom_lpf_horizontal_8_sse2, &aom_lpf_horizontal_8_c, 8),
+  make_tuple(&aom_lpf_horizontal_6_sse2, &aom_lpf_horizontal_6_c, 8),
+  make_tuple(&aom_lpf_vertical_6_sse2, &aom_lpf_vertical_6_c, 8),
+  make_tuple(&aom_lpf_horizontal_14_sse2, &aom_lpf_horizontal_14_c, 8),
+  make_tuple(&aom_lpf_vertical_4_sse2, &aom_lpf_vertical_4_c, 8),
+  make_tuple(&aom_lpf_vertical_8_sse2, &aom_lpf_vertical_8_c, 8),
+  make_tuple(&aom_lpf_vertical_14_sse2, &aom_lpf_vertical_14_c, 8),
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, Loop8Test6Param_lbd,
+                        ::testing::ValuesIn(kLoop8Test6));
+
+const dual_loop_param_t kLoop8Test9[] = {
+  make_tuple(&aom_lpf_horizontal_4_dual_sse2, &aom_lpf_horizontal_4_dual_c, 8),
+  make_tuple(&aom_lpf_vertical_4_dual_sse2, &aom_lpf_vertical_4_dual_c, 8),
+  make_tuple(&aom_lpf_horizontal_6_dual_sse2, &aom_lpf_horizontal_6_dual_c, 8),
+  make_tuple(&aom_lpf_vertical_6_dual_sse2, &aom_lpf_vertical_6_dual_c, 8),
+  make_tuple(&aom_lpf_horizontal_8_dual_sse2, &aom_lpf_horizontal_8_dual_c, 8),
+  make_tuple(&aom_lpf_vertical_8_dual_sse2, &aom_lpf_vertical_8_dual_c, 8),
+  make_tuple(&aom_lpf_horizontal_14_dual_sse2, &aom_lpf_horizontal_14_dual_c,
+             8),
+  make_tuple(&aom_lpf_vertical_14_dual_sse2, &aom_lpf_vertical_14_dual_c, 8)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, Loop8Test9Param_lbd,
+                        ::testing::ValuesIn(kLoop8Test9));
+
+#endif  // HAVE_SSE2
+
+#if HAVE_SSE2
+const hbddual_loop_param_t kHbdLoop8Test9[] = {
+  make_tuple(&aom_highbd_lpf_horizontal_4_dual_sse2,
+             &aom_highbd_lpf_horizontal_4_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_horizontal_6_dual_sse2,
+             &aom_highbd_lpf_horizontal_6_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_horizontal_8_dual_sse2,
+             &aom_highbd_lpf_horizontal_8_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_horizontal_14_dual_sse2,
+             &aom_highbd_lpf_horizontal_14_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_vertical_4_dual_sse2,
+             &aom_highbd_lpf_vertical_4_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_vertical_6_dual_sse2,
+             &aom_highbd_lpf_vertical_6_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_vertical_8_dual_sse2,
+             &aom_highbd_lpf_vertical_8_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_vertical_14_dual_sse2,
+             &aom_highbd_lpf_vertical_14_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_horizontal_4_dual_sse2,
+             &aom_highbd_lpf_horizontal_4_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_horizontal_6_dual_sse2,
+             &aom_highbd_lpf_horizontal_6_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_horizontal_8_dual_sse2,
+             &aom_highbd_lpf_horizontal_8_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_horizontal_14_dual_sse2,
+             &aom_highbd_lpf_horizontal_14_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_4_dual_sse2,
+             &aom_highbd_lpf_vertical_4_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_6_dual_sse2,
+             &aom_highbd_lpf_vertical_6_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_8_dual_sse2,
+             &aom_highbd_lpf_vertical_8_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_14_dual_sse2,
+             &aom_highbd_lpf_vertical_14_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_horizontal_4_dual_sse2,
+             &aom_highbd_lpf_horizontal_4_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_horizontal_6_dual_sse2,
+             &aom_highbd_lpf_horizontal_6_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_horizontal_8_dual_sse2,
+             &aom_highbd_lpf_horizontal_8_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_horizontal_14_dual_sse2,
+             &aom_highbd_lpf_horizontal_14_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_vertical_4_dual_sse2,
+             &aom_highbd_lpf_vertical_4_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_vertical_6_dual_sse2,
+             &aom_highbd_lpf_vertical_6_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_vertical_8_dual_sse2,
+             &aom_highbd_lpf_vertical_8_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_vertical_14_dual_sse2,
+             &aom_highbd_lpf_vertical_14_dual_c, 12),
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, Loop8Test9Param_hbd,
+                        ::testing::ValuesIn(kHbdLoop8Test9));
+
+#endif  // HAVE_SSE2
+
+#if HAVE_NEON
+const loop_param_t kLoop8Test6[] = {
+  make_tuple(&aom_lpf_vertical_14_neon, &aom_lpf_vertical_14_c, 8),
+  make_tuple(&aom_lpf_vertical_8_neon, &aom_lpf_vertical_8_c, 8),
+  make_tuple(&aom_lpf_vertical_6_neon, &aom_lpf_vertical_6_c, 8),
+  make_tuple(&aom_lpf_vertical_4_neon, &aom_lpf_vertical_4_c, 8),
+  make_tuple(&aom_lpf_horizontal_14_neon, &aom_lpf_horizontal_14_c, 8),
+  make_tuple(&aom_lpf_horizontal_8_neon, &aom_lpf_horizontal_8_c, 8),
+  make_tuple(&aom_lpf_horizontal_6_neon, &aom_lpf_horizontal_6_c, 8),
+  make_tuple(&aom_lpf_horizontal_4_neon, &aom_lpf_horizontal_4_c, 8)
+};
+
+INSTANTIATE_TEST_CASE_P(NEON, Loop8Test6Param_lbd,
+                        ::testing::ValuesIn(kLoop8Test6));
+#endif  // HAVE_NEON
+
+#if HAVE_AVX2
+const hbddual_loop_param_t kHbdLoop8Test9Avx2[] = {
+  make_tuple(&aom_highbd_lpf_horizontal_4_dual_avx2,
+             &aom_highbd_lpf_horizontal_4_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_horizontal_4_dual_avx2,
+             &aom_highbd_lpf_horizontal_4_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_horizontal_4_dual_avx2,
+             &aom_highbd_lpf_horizontal_4_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_horizontal_8_dual_avx2,
+             &aom_highbd_lpf_horizontal_8_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_horizontal_8_dual_avx2,
+             &aom_highbd_lpf_horizontal_8_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_horizontal_8_dual_avx2,
+             &aom_highbd_lpf_horizontal_8_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_vertical_4_dual_avx2,
+             &aom_highbd_lpf_vertical_4_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_vertical_4_dual_avx2,
+             &aom_highbd_lpf_vertical_4_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_4_dual_avx2,
+             &aom_highbd_lpf_vertical_4_dual_c, 12),
+  make_tuple(&aom_highbd_lpf_vertical_8_dual_avx2,
+             &aom_highbd_lpf_vertical_8_dual_c, 8),
+  make_tuple(&aom_highbd_lpf_vertical_8_dual_avx2,
+             &aom_highbd_lpf_vertical_8_dual_c, 10),
+  make_tuple(&aom_highbd_lpf_vertical_8_dual_avx2,
+             &aom_highbd_lpf_vertical_8_dual_c, 12),
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, Loop8Test9Param_hbd,
+                        ::testing::ValuesIn(kHbdLoop8Test9Avx2));
+#endif
+}  // namespace
diff --git a/third_party/aom/test/masked_sad_test.cc b/third_party/aom/test/masked_sad_test.cc
new file mode 100644
index 0000000000..311f1877d4
--- /dev/null
+++ b/third_party/aom/test/masked_sad_test.cc
@@ -0,0 +1,342 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_integer.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+const int number_of_iterations = 200;
+
+typedef unsigned int (*MaskedSADFunc)(const uint8_t *src, int src_stride,
+                                      const uint8_t *ref, int ref_stride,
+                                      const uint8_t *second_pred,
+                                      const uint8_t *msk, int msk_stride,
+                                      int invert_mask);
+typedef ::testing::tuple<MaskedSADFunc, MaskedSADFunc> MaskedSADParam;
+
+class MaskedSADTest : public ::testing::TestWithParam<MaskedSADParam> {
+ public:
+  virtual ~MaskedSADTest() {}
+  virtual void SetUp() {
+    maskedSAD_op_ = GET_PARAM(0);
+    ref_maskedSAD_op_ = GET_PARAM(1);
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+  void runMaskedSADTest(int run_times);
+
+ protected:
+  MaskedSADFunc maskedSAD_op_;
+  MaskedSADFunc ref_maskedSAD_op_;
+};
+void MaskedSADTest::runMaskedSADTest(int run_times) {
+  unsigned int ref_ret = 0, ret = 1;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint8_t, src_ptr[MAX_SB_SIZE * MAX_SB_SIZE]);
+  DECLARE_ALIGNED(16, uint8_t, ref_ptr[MAX_SB_SIZE * MAX_SB_SIZE]);
+  DECLARE_ALIGNED(16, uint8_t, second_pred_ptr[MAX_SB_SIZE * MAX_SB_SIZE]);
+  DECLARE_ALIGNED(16, uint8_t, msk_ptr[MAX_SB_SIZE * MAX_SB_SIZE]);
+  int err_count = 0;
+  int first_failure = -1;
+  int src_stride = MAX_SB_SIZE;
+  int ref_stride = MAX_SB_SIZE;
+  int msk_stride = MAX_SB_SIZE;
+  const int iters = run_times == 1 ? number_of_iterations : 1;
+  for (int i = 0; i < iters; ++i) {
+    for (int j = 0; j < MAX_SB_SIZE * MAX_SB_SIZE; j++) {
+      src_ptr[j] = rnd.Rand8();
+      ref_ptr[j] = rnd.Rand8();
+      second_pred_ptr[j] = rnd.Rand8();
+      msk_ptr[j] = ((rnd.Rand8() & 0x7f) > 64) ? rnd.Rand8() & 0x3f : 64;
+      assert(msk_ptr[j] <= 64);
+    }
+
+    for (int invert_mask = 0; invert_mask < 2; ++invert_mask) {
+      aom_usec_timer timer;
+      aom_usec_timer_start(&timer);
+      for (int repeat = 0; repeat < run_times; ++repeat) {
+        ref_ret = ref_maskedSAD_op_(src_ptr, src_stride, ref_ptr, ref_stride,
+                                    second_pred_ptr, msk_ptr, msk_stride,
+                                    invert_mask);
+      }
+      aom_usec_timer_mark(&timer);
+      const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+      aom_usec_timer_start(&timer);
+      if (run_times == 1) {
+        ASM_REGISTER_STATE_CHECK(ret = maskedSAD_op_(src_ptr, src_stride,
+                                                     ref_ptr, ref_stride,
+                                                     second_pred_ptr, msk_ptr,
+                                                     msk_stride, invert_mask));
+      } else {
+        for (int repeat = 0; repeat < run_times; ++repeat) {
+          ret =
+              maskedSAD_op_(src_ptr, src_stride, ref_ptr, ref_stride,
+                            second_pred_ptr, msk_ptr, msk_stride, invert_mask);
+        }
+      }
+      aom_usec_timer_mark(&timer);
+      const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+      if (run_times > 10) {
+        printf("%7.2f/%7.2fns", time1, time2);
+        printf("(%3.2f)\n", time1 / time2);
+      }
+      if (ret != ref_ret) {
+        err_count++;
+        if (first_failure == -1) first_failure = i;
+      }
+    }
+  }
+  EXPECT_EQ(0, err_count) << "Error: Masked SAD Test,  output doesn't match. "
+                          << "First failed at test case " << first_failure;
+}
+
+TEST_P(MaskedSADTest, OperationCheck) { runMaskedSADTest(1); }
+
+TEST_P(MaskedSADTest, DISABLED_Speed) { runMaskedSADTest(2000000); }
+
+typedef unsigned int (*HighbdMaskedSADFunc)(const uint8_t *src, int src_stride,
+                                            const uint8_t *ref, int ref_stride,
+                                            const uint8_t *second_pred,
+                                            const uint8_t *msk, int msk_stride,
+                                            int invert_mask);
+typedef ::testing::tuple<HighbdMaskedSADFunc, HighbdMaskedSADFunc>
+    HighbdMaskedSADParam;
+
+class HighbdMaskedSADTest
+    : public ::testing::TestWithParam<HighbdMaskedSADParam> {
+ public:
+  virtual ~HighbdMaskedSADTest() {}
+  virtual void SetUp() {
+    maskedSAD_op_ = GET_PARAM(0);
+    ref_maskedSAD_op_ = GET_PARAM(1);
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+  void runHighbdMaskedSADTest(int run_times);
+
+ protected:
+  HighbdMaskedSADFunc maskedSAD_op_;
+  HighbdMaskedSADFunc ref_maskedSAD_op_;
+};
+void HighbdMaskedSADTest::runHighbdMaskedSADTest(int run_times) {
+  unsigned int ref_ret = 0, ret = 1;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint16_t, src_ptr[MAX_SB_SIZE * MAX_SB_SIZE]);
+  DECLARE_ALIGNED(16, uint16_t, ref_ptr[MAX_SB_SIZE * MAX_SB_SIZE]);
+  DECLARE_ALIGNED(16, uint16_t, second_pred_ptr[MAX_SB_SIZE * MAX_SB_SIZE]);
+  DECLARE_ALIGNED(16, uint8_t, msk_ptr[MAX_SB_SIZE * MAX_SB_SIZE]);
+  uint8_t *src8_ptr = CONVERT_TO_BYTEPTR(src_ptr);
+  uint8_t *ref8_ptr = CONVERT_TO_BYTEPTR(ref_ptr);
+  uint8_t *second_pred8_ptr = CONVERT_TO_BYTEPTR(second_pred_ptr);
+  int err_count = 0;
+  int first_failure = -1;
+  int src_stride = MAX_SB_SIZE;
+  int ref_stride = MAX_SB_SIZE;
+  int msk_stride = MAX_SB_SIZE;
+  const int iters = run_times == 1 ? number_of_iterations : 1;
+  for (int i = 0; i < iters; ++i) {
+    for (int j = 0; j < MAX_SB_SIZE * MAX_SB_SIZE; j++) {
+      src_ptr[j] = rnd.Rand16() & 0xfff;
+      ref_ptr[j] = rnd.Rand16() & 0xfff;
+      second_pred_ptr[j] = rnd.Rand16() & 0xfff;
+      msk_ptr[j] = ((rnd.Rand8() & 0x7f) > 64) ? rnd.Rand8() & 0x3f : 64;
+    }
+
+    for (int invert_mask = 0; invert_mask < 2; ++invert_mask) {
+      aom_usec_timer timer;
+      aom_usec_timer_start(&timer);
+      for (int repeat = 0; repeat < run_times; ++repeat) {
+        ref_ret = ref_maskedSAD_op_(src8_ptr, src_stride, ref8_ptr, ref_stride,
+                                    second_pred8_ptr, msk_ptr, msk_stride,
+                                    invert_mask);
+      }
+      aom_usec_timer_mark(&timer);
+      const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+      aom_usec_timer_start(&timer);
+      if (run_times == 1) {
+        ASM_REGISTER_STATE_CHECK(ret = maskedSAD_op_(src8_ptr, src_stride,
+                                                     ref8_ptr, ref_stride,
+                                                     second_pred8_ptr, msk_ptr,
+                                                     msk_stride, invert_mask));
+      } else {
+        for (int repeat = 0; repeat < run_times; ++repeat) {
+          ret =
+              maskedSAD_op_(src8_ptr, src_stride, ref8_ptr, ref_stride,
+                            second_pred8_ptr, msk_ptr, msk_stride, invert_mask);
+        }
+      }
+      aom_usec_timer_mark(&timer);
+      const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+      if (run_times > 10) {
+        printf("%7.2f/%7.2fns", time1, time2);
+        printf("(%3.2f)\n", time1 / time2);
+      }
+      if (ret != ref_ret) {
+        err_count++;
+        if (first_failure == -1) first_failure = i;
+      }
+    }
+  }
+  EXPECT_EQ(0, err_count)
+      << "Error: High BD Masked SAD Test, output doesn't match. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(HighbdMaskedSADTest, OperationCheck) { runHighbdMaskedSADTest(1); }
+
+TEST_P(HighbdMaskedSADTest, DISABLED_Speed) { runHighbdMaskedSADTest(1000000); }
+
+using ::testing::make_tuple;
+
+#if HAVE_SSSE3
+const MaskedSADParam msad_test[] = {
+  make_tuple(&aom_masked_sad4x4_ssse3, &aom_masked_sad4x4_c),
+  make_tuple(&aom_masked_sad4x8_ssse3, &aom_masked_sad4x8_c),
+  make_tuple(&aom_masked_sad8x4_ssse3, &aom_masked_sad8x4_c),
+  make_tuple(&aom_masked_sad8x8_ssse3, &aom_masked_sad8x8_c),
+  make_tuple(&aom_masked_sad8x16_ssse3, &aom_masked_sad8x16_c),
+  make_tuple(&aom_masked_sad16x8_ssse3, &aom_masked_sad16x8_c),
+  make_tuple(&aom_masked_sad16x16_ssse3, &aom_masked_sad16x16_c),
+  make_tuple(&aom_masked_sad16x32_ssse3, &aom_masked_sad16x32_c),
+  make_tuple(&aom_masked_sad32x16_ssse3, &aom_masked_sad32x16_c),
+  make_tuple(&aom_masked_sad32x32_ssse3, &aom_masked_sad32x32_c),
+  make_tuple(&aom_masked_sad32x64_ssse3, &aom_masked_sad32x64_c),
+  make_tuple(&aom_masked_sad64x32_ssse3, &aom_masked_sad64x32_c),
+  make_tuple(&aom_masked_sad64x64_ssse3, &aom_masked_sad64x64_c),
+  make_tuple(&aom_masked_sad64x128_ssse3, &aom_masked_sad64x128_c),
+  make_tuple(&aom_masked_sad128x64_ssse3, &aom_masked_sad128x64_c),
+  make_tuple(&aom_masked_sad128x128_ssse3, &aom_masked_sad128x128_c),
+  make_tuple(&aom_masked_sad4x16_ssse3, &aom_masked_sad4x16_c),
+  make_tuple(&aom_masked_sad16x4_ssse3, &aom_masked_sad16x4_c),
+  make_tuple(&aom_masked_sad8x32_ssse3, &aom_masked_sad8x32_c),
+  make_tuple(&aom_masked_sad32x8_ssse3, &aom_masked_sad32x8_c),
+  make_tuple(&aom_masked_sad16x64_ssse3, &aom_masked_sad16x64_c),
+  make_tuple(&aom_masked_sad64x16_ssse3, &aom_masked_sad64x16_c),
+};
+
+INSTANTIATE_TEST_CASE_P(SSSE3, MaskedSADTest, ::testing::ValuesIn(msad_test));
+
+const HighbdMaskedSADParam hbd_msad_test[] = {
+  make_tuple(&aom_highbd_masked_sad4x4_ssse3, &aom_highbd_masked_sad4x4_c),
+  make_tuple(&aom_highbd_masked_sad4x8_ssse3, &aom_highbd_masked_sad4x8_c),
+  make_tuple(&aom_highbd_masked_sad8x4_ssse3, &aom_highbd_masked_sad8x4_c),
+  make_tuple(&aom_highbd_masked_sad8x8_ssse3, &aom_highbd_masked_sad8x8_c),
+  make_tuple(&aom_highbd_masked_sad8x16_ssse3, &aom_highbd_masked_sad8x16_c),
+  make_tuple(&aom_highbd_masked_sad16x8_ssse3, &aom_highbd_masked_sad16x8_c),
+  make_tuple(&aom_highbd_masked_sad16x16_ssse3, &aom_highbd_masked_sad16x16_c),
+  make_tuple(&aom_highbd_masked_sad16x32_ssse3, &aom_highbd_masked_sad16x32_c),
+  make_tuple(&aom_highbd_masked_sad32x16_ssse3, &aom_highbd_masked_sad32x16_c),
+  make_tuple(&aom_highbd_masked_sad32x32_ssse3, &aom_highbd_masked_sad32x32_c),
+  make_tuple(&aom_highbd_masked_sad32x64_ssse3, &aom_highbd_masked_sad32x64_c),
+  make_tuple(&aom_highbd_masked_sad64x32_ssse3, &aom_highbd_masked_sad64x32_c),
+  make_tuple(&aom_highbd_masked_sad64x64_ssse3, &aom_highbd_masked_sad64x64_c),
+  make_tuple(&aom_highbd_masked_sad64x128_ssse3,
+             &aom_highbd_masked_sad64x128_c),
+  make_tuple(&aom_highbd_masked_sad128x64_ssse3,
+             &aom_highbd_masked_sad128x64_c),
+  make_tuple(&aom_highbd_masked_sad128x128_ssse3,
+             &aom_highbd_masked_sad128x128_c),
+  make_tuple(&aom_highbd_masked_sad4x16_ssse3, &aom_highbd_masked_sad4x16_c),
+  make_tuple(&aom_highbd_masked_sad16x4_ssse3, &aom_highbd_masked_sad16x4_c),
+  make_tuple(&aom_highbd_masked_sad8x32_ssse3, &aom_highbd_masked_sad8x32_c),
+  make_tuple(&aom_highbd_masked_sad32x8_ssse3, &aom_highbd_masked_sad32x8_c),
+  make_tuple(&aom_highbd_masked_sad16x64_ssse3, &aom_highbd_masked_sad16x64_c),
+  make_tuple(&aom_highbd_masked_sad64x16_ssse3, &aom_highbd_masked_sad64x16_c),
+};
+
+INSTANTIATE_TEST_CASE_P(SSSE3, HighbdMaskedSADTest,
+                        ::testing::ValuesIn(hbd_msad_test));
+#endif  // HAVE_SSSE3
+
+#if HAVE_AVX2
+const MaskedSADParam msad_avx2_test[] = {
+  make_tuple(&aom_masked_sad4x4_avx2, &aom_masked_sad4x4_ssse3),
+  make_tuple(&aom_masked_sad4x8_avx2, &aom_masked_sad4x8_ssse3),
+  make_tuple(&aom_masked_sad8x4_avx2, &aom_masked_sad8x4_ssse3),
+  make_tuple(&aom_masked_sad8x8_avx2, &aom_masked_sad8x8_ssse3),
+  make_tuple(&aom_masked_sad8x16_avx2, &aom_masked_sad8x16_ssse3),
+  make_tuple(&aom_masked_sad16x8_avx2, &aom_masked_sad16x8_ssse3),
+  make_tuple(&aom_masked_sad16x16_avx2, &aom_masked_sad16x16_ssse3),
+  make_tuple(&aom_masked_sad16x32_avx2, &aom_masked_sad16x32_ssse3),
+  make_tuple(&aom_masked_sad32x16_avx2, &aom_masked_sad32x16_ssse3),
+  make_tuple(&aom_masked_sad32x32_avx2, &aom_masked_sad32x32_ssse3),
+  make_tuple(&aom_masked_sad32x64_avx2, &aom_masked_sad32x64_ssse3),
+  make_tuple(&aom_masked_sad64x32_avx2, &aom_masked_sad64x32_ssse3),
+  make_tuple(&aom_masked_sad64x64_avx2, &aom_masked_sad64x64_ssse3),
+  make_tuple(&aom_masked_sad64x128_avx2, &aom_masked_sad64x128_ssse3),
+  make_tuple(&aom_masked_sad128x64_avx2, &aom_masked_sad128x64_ssse3),
+  make_tuple(&aom_masked_sad128x128_avx2, &aom_masked_sad128x128_ssse3),
+  make_tuple(&aom_masked_sad4x16_avx2, &aom_masked_sad4x16_ssse3),
+  make_tuple(&aom_masked_sad16x4_avx2, &aom_masked_sad16x4_ssse3),
+  make_tuple(&aom_masked_sad8x32_avx2, &aom_masked_sad8x32_ssse3),
+  make_tuple(&aom_masked_sad32x8_avx2, &aom_masked_sad32x8_ssse3),
+  make_tuple(&aom_masked_sad16x64_avx2, &aom_masked_sad16x64_ssse3),
+  make_tuple(&aom_masked_sad64x16_avx2, &aom_masked_sad64x16_ssse3)
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, MaskedSADTest,
+                        ::testing::ValuesIn(msad_avx2_test));
+
+const HighbdMaskedSADParam hbd_msad_avx2_test[] = {
+  make_tuple(&aom_highbd_masked_sad4x4_avx2, &aom_highbd_masked_sad4x4_ssse3),
+  make_tuple(&aom_highbd_masked_sad4x8_avx2, &aom_highbd_masked_sad4x8_ssse3),
+  make_tuple(&aom_highbd_masked_sad8x4_avx2, &aom_highbd_masked_sad8x4_ssse3),
+  make_tuple(&aom_highbd_masked_sad8x8_avx2, &aom_highbd_masked_sad8x8_ssse3),
+  make_tuple(&aom_highbd_masked_sad8x16_avx2, &aom_highbd_masked_sad8x16_ssse3),
+  make_tuple(&aom_highbd_masked_sad16x8_avx2, &aom_highbd_masked_sad16x8_ssse3),
+  make_tuple(&aom_highbd_masked_sad16x16_avx2,
+             &aom_highbd_masked_sad16x16_ssse3),
+  make_tuple(&aom_highbd_masked_sad16x32_avx2,
+             &aom_highbd_masked_sad16x32_ssse3),
+  make_tuple(&aom_highbd_masked_sad32x16_avx2,
+             &aom_highbd_masked_sad32x16_ssse3),
+  make_tuple(&aom_highbd_masked_sad32x32_avx2,
+             &aom_highbd_masked_sad32x32_ssse3),
+  make_tuple(&aom_highbd_masked_sad32x64_avx2,
+             &aom_highbd_masked_sad32x64_ssse3),
+  make_tuple(&aom_highbd_masked_sad64x32_avx2,
+             &aom_highbd_masked_sad64x32_ssse3),
+  make_tuple(&aom_highbd_masked_sad64x64_avx2,
+             &aom_highbd_masked_sad64x64_ssse3),
+  make_tuple(&aom_highbd_masked_sad64x128_avx2,
+             &aom_highbd_masked_sad64x128_ssse3),
+  make_tuple(&aom_highbd_masked_sad128x64_avx2,
+             &aom_highbd_masked_sad128x64_ssse3),
+  make_tuple(&aom_highbd_masked_sad128x128_avx2,
+             &aom_highbd_masked_sad128x128_ssse3),
+  make_tuple(&aom_highbd_masked_sad4x16_avx2, &aom_highbd_masked_sad4x16_ssse3),
+  make_tuple(&aom_highbd_masked_sad16x4_avx2, &aom_highbd_masked_sad16x4_ssse3),
+  make_tuple(&aom_highbd_masked_sad8x32_avx2, &aom_highbd_masked_sad8x32_ssse3),
+  make_tuple(&aom_highbd_masked_sad32x8_avx2, &aom_highbd_masked_sad32x8_ssse3),
+  make_tuple(&aom_highbd_masked_sad16x64_avx2,
+             &aom_highbd_masked_sad16x64_ssse3),
+  make_tuple(&aom_highbd_masked_sad64x16_avx2,
+             &aom_highbd_masked_sad64x16_ssse3)
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, HighbdMaskedSADTest,
+                        ::testing::ValuesIn(hbd_msad_avx2_test));
+#endif  // HAVE_AVX2
+
+}  // namespace
diff --git a/third_party/aom/test/masked_variance_test.cc b/third_party/aom/test/masked_variance_test.cc
new file mode 100644
index 0000000000..275b9feb6a
--- /dev/null
+++ b/third_party/aom/test/masked_variance_test.cc
@@ -0,0 +1,459 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_codec.h"
+#include "aom/aom_integer.h"
+#include "aom_dsp/aom_filter.h"
+#include "aom_mem/aom_mem.h"
+
+using libaom_test::ACMRandom;
+
+namespace {
+const int number_of_iterations = 200;
+
+typedef unsigned int (*MaskedSubPixelVarianceFunc)(
+    const uint8_t *src, int src_stride, int xoffset, int yoffset,
+    const uint8_t *ref, int ref_stride, const uint8_t *second_pred,
+    const uint8_t *msk, int msk_stride, int invert_mask, unsigned int *sse);
+
+typedef ::testing::tuple<MaskedSubPixelVarianceFunc, MaskedSubPixelVarianceFunc>
+    MaskedSubPixelVarianceParam;
+
+class MaskedSubPixelVarianceTest
+    : public ::testing::TestWithParam<MaskedSubPixelVarianceParam> {
+ public:
+  virtual ~MaskedSubPixelVarianceTest() {}
+  virtual void SetUp() {
+    opt_func_ = GET_PARAM(0);
+    ref_func_ = GET_PARAM(1);
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  MaskedSubPixelVarianceFunc opt_func_;
+  MaskedSubPixelVarianceFunc ref_func_;
+};
+
+TEST_P(MaskedSubPixelVarianceTest, OperationCheck) {
+  unsigned int ref_ret, opt_ret;
+  unsigned int ref_sse, opt_sse;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  // Note: We pad out the input array to a multiple of 16 bytes wide, so that
+  // consecutive rows keep the 16-byte alignment.
+  DECLARE_ALIGNED(16, uint8_t, src_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16)]);
+  DECLARE_ALIGNED(16, uint8_t, ref_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16)]);
+  DECLARE_ALIGNED(16, uint8_t,
+                  second_pred_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16)]);
+  DECLARE_ALIGNED(16, uint8_t, msk_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16)]);
+  int err_count = 0;
+  int first_failure = -1;
+  int src_stride = (MAX_SB_SIZE + 16);
+  int ref_stride = (MAX_SB_SIZE + 16);
+  int msk_stride = (MAX_SB_SIZE + 16);
+  int xoffset;
+  int yoffset;
+
+  for (int i = 0; i < number_of_iterations; ++i) {
+    int xoffsets[] = { 0, 4, rnd(BIL_SUBPEL_SHIFTS) };
+    int yoffsets[] = { 0, 4, rnd(BIL_SUBPEL_SHIFTS) };
+    for (int j = 0; j < (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16); j++) {
+      src_ptr[j] = rnd.Rand8();
+      ref_ptr[j] = rnd.Rand8();
+      second_pred_ptr[j] = rnd.Rand8();
+      msk_ptr[j] = rnd(65);
+    }
+    for (int k = 0; k < 3; k++) {
+      for (int l = 0; l < 3; l++) {
+        xoffset = xoffsets[k];
+        yoffset = yoffsets[l];
+        for (int invert_mask = 0; invert_mask < 2; ++invert_mask) {
+          ref_ret = ref_func_(src_ptr, src_stride, xoffset, yoffset, ref_ptr,
+                              ref_stride, second_pred_ptr, msk_ptr, msk_stride,
+                              invert_mask, &ref_sse);
+          ASM_REGISTER_STATE_CHECK(
+              opt_ret = opt_func_(src_ptr, src_stride, xoffset, yoffset,
+                                  ref_ptr, ref_stride, second_pred_ptr, msk_ptr,
+                                  msk_stride, invert_mask, &opt_sse));
+
+          if (opt_ret != ref_ret || opt_sse != ref_sse) {
+            err_count++;
+            if (first_failure == -1) first_failure = i;
+          }
+        }
+      }
+    }
+  }
+
+  EXPECT_EQ(0, err_count)
+      << "Error: Masked Sub Pixel Variance Test OperationCheck,"
+      << "C output doesn't match SSSE3 output. "
+      << "First failed at test case " << first_failure;
+}
+
+TEST_P(MaskedSubPixelVarianceTest, ExtremeValues) {
+  unsigned int ref_ret, opt_ret;
+  unsigned int ref_sse, opt_sse;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint8_t, src_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16)]);
+  DECLARE_ALIGNED(16, uint8_t, ref_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16)]);
+  DECLARE_ALIGNED(16, uint8_t,
+                  second_pred_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16)]);
+  DECLARE_ALIGNED(16, uint8_t, msk_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16)]);
+  int first_failure_x = -1;
+  int first_failure_y = -1;
+  int err_count = 0;
+  int first_failure = -1;
+  int src_stride = (MAX_SB_SIZE + 16);
+  int ref_stride = (MAX_SB_SIZE + 16);
+  int msk_stride = (MAX_SB_SIZE + 16);
+
+  for (int xoffset = 0; xoffset < BIL_SUBPEL_SHIFTS; xoffset++) {
+    for (int yoffset = 0; yoffset < BIL_SUBPEL_SHIFTS; yoffset++) {
+      for (int i = 0; i < 16; ++i) {
+        memset(src_ptr, (i & 0x1) ? 255 : 0,
+               (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16));
+        memset(ref_ptr, (i & 0x2) ? 255 : 0,
+               (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16));
+        memset(second_pred_ptr, (i & 0x4) ? 255 : 0,
+               (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16));
+        memset(msk_ptr, (i & 0x8) ? 64 : 0,
+               (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 16));
+
+        for (int invert_mask = 0; invert_mask < 2; ++invert_mask) {
+          ref_ret = ref_func_(src_ptr, src_stride, xoffset, yoffset, ref_ptr,
+                              ref_stride, second_pred_ptr, msk_ptr, msk_stride,
+                              invert_mask, &ref_sse);
+          ASM_REGISTER_STATE_CHECK(
+              opt_ret = opt_func_(src_ptr, src_stride, xoffset, yoffset,
+                                  ref_ptr, ref_stride, second_pred_ptr, msk_ptr,
+                                  msk_stride, invert_mask, &opt_sse));
+
+          if (opt_ret != ref_ret || opt_sse != ref_sse) {
+            err_count++;
+            if (first_failure == -1) {
+              first_failure = i;
+              first_failure_x = xoffset;
+              first_failure_y = yoffset;
+            }
+          }
+        }
+      }
+    }
+  }
+
+  EXPECT_EQ(0, err_count) << "Error: Masked Variance Test ExtremeValues,"
+                          << "C output doesn't match SSSE3 output. "
+                          << "First failed at test case " << first_failure
+                          << " x_offset = " << first_failure_x
+                          << " y_offset = " << first_failure_y;
+}
+
+typedef ::testing::tuple<MaskedSubPixelVarianceFunc, MaskedSubPixelVarianceFunc,
+                         aom_bit_depth_t>
+    HighbdMaskedSubPixelVarianceParam;
+
+class HighbdMaskedSubPixelVarianceTest
+    : public ::testing::TestWithParam<HighbdMaskedSubPixelVarianceParam> {
+ public:
+  virtual ~HighbdMaskedSubPixelVarianceTest() {}
+  virtual void SetUp() {
+    opt_func_ = GET_PARAM(0);
+    ref_func_ = GET_PARAM(1);
+    bit_depth_ = GET_PARAM(2);
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  MaskedSubPixelVarianceFunc opt_func_;
+  MaskedSubPixelVarianceFunc ref_func_;
+  aom_bit_depth_t bit_depth_;
+};
+
+TEST_P(HighbdMaskedSubPixelVarianceTest, OperationCheck) {
+  unsigned int ref_ret, opt_ret;
+  unsigned int ref_sse, opt_sse;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint16_t, src_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8)]);
+  DECLARE_ALIGNED(16, uint16_t, ref_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8)]);
+  DECLARE_ALIGNED(16, uint16_t,
+                  second_pred_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8)]);
+  DECLARE_ALIGNED(16, uint8_t, msk_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8)]);
+  uint8_t *src8_ptr = CONVERT_TO_BYTEPTR(src_ptr);
+  uint8_t *ref8_ptr = CONVERT_TO_BYTEPTR(ref_ptr);
+  uint8_t *second_pred8_ptr = CONVERT_TO_BYTEPTR(second_pred_ptr);
+  int err_count = 0;
+  int first_failure = -1;
+  int first_failure_x = -1;
+  int first_failure_y = -1;
+  int src_stride = (MAX_SB_SIZE + 8);
+  int ref_stride = (MAX_SB_SIZE + 8);
+  int msk_stride = (MAX_SB_SIZE + 8);
+  int xoffset, yoffset;
+
+  for (int i = 0; i < number_of_iterations; ++i) {
+    for (int j = 0; j < (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8); j++) {
+      src_ptr[j] = rnd.Rand16() & ((1 << bit_depth_) - 1);
+      ref_ptr[j] = rnd.Rand16() & ((1 << bit_depth_) - 1);
+      second_pred_ptr[j] = rnd.Rand16() & ((1 << bit_depth_) - 1);
+      msk_ptr[j] = rnd(65);
+    }
+    for (xoffset = 0; xoffset < BIL_SUBPEL_SHIFTS; xoffset++) {
+      for (yoffset = 0; yoffset < BIL_SUBPEL_SHIFTS; yoffset++) {
+        for (int invert_mask = 0; invert_mask < 2; ++invert_mask) {
+          ref_ret = ref_func_(src8_ptr, src_stride, xoffset, yoffset, ref8_ptr,
+                              ref_stride, second_pred8_ptr, msk_ptr, msk_stride,
+                              invert_mask, &ref_sse);
+          ASM_REGISTER_STATE_CHECK(
+              opt_ret = opt_func_(src8_ptr, src_stride, xoffset, yoffset,
+                                  ref8_ptr, ref_stride, second_pred8_ptr,
+                                  msk_ptr, msk_stride, invert_mask, &opt_sse));
+
+          if (opt_ret != ref_ret || opt_sse != ref_sse) {
+            err_count++;
+            if (first_failure == -1) {
+              first_failure = i;
+              first_failure_x = xoffset;
+              first_failure_y = yoffset;
+            }
+          }
+        }
+      }
+    }
+  }
+
+  EXPECT_EQ(0, err_count)
+      << "Error: Masked Sub Pixel Variance Test OperationCheck,"
+      << "C output doesn't match SSSE3 output. "
+      << "First failed at test case " << first_failure
+      << " x_offset = " << first_failure_x << " y_offset = " << first_failure_y;
+}
+
+TEST_P(HighbdMaskedSubPixelVarianceTest, ExtremeValues) {
+  unsigned int ref_ret, opt_ret;
+  unsigned int ref_sse, opt_sse;
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  DECLARE_ALIGNED(16, uint16_t, src_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8)]);
+  DECLARE_ALIGNED(16, uint16_t, ref_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8)]);
+  DECLARE_ALIGNED(16, uint8_t, msk_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8)]);
+  DECLARE_ALIGNED(16, uint16_t,
+                  second_pred_ptr[(MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8)]);
+  uint8_t *src8_ptr = CONVERT_TO_BYTEPTR(src_ptr);
+  uint8_t *ref8_ptr = CONVERT_TO_BYTEPTR(ref_ptr);
+  uint8_t *second_pred8_ptr = CONVERT_TO_BYTEPTR(second_pred_ptr);
+  int first_failure_x = -1;
+  int first_failure_y = -1;
+  int err_count = 0;
+  int first_failure = -1;
+  int src_stride = (MAX_SB_SIZE + 8);
+  int ref_stride = (MAX_SB_SIZE + 8);
+  int msk_stride = (MAX_SB_SIZE + 8);
+
+  for (int xoffset = 0; xoffset < BIL_SUBPEL_SHIFTS; xoffset++) {
+    for (int yoffset = 0; yoffset < BIL_SUBPEL_SHIFTS; yoffset++) {
+      for (int i = 0; i < 16; ++i) {
+        aom_memset16(src_ptr, (i & 0x1) ? ((1 << bit_depth_) - 1) : 0,
+                     (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8));
+        aom_memset16(ref_ptr, (i & 0x2) ? ((1 << bit_depth_) - 1) : 0,
+                     (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8));
+        aom_memset16(second_pred_ptr, (i & 0x4) ? ((1 << bit_depth_) - 1) : 0,
+                     (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8));
+        memset(msk_ptr, (i & 0x8) ? 64 : 0,
+               (MAX_SB_SIZE + 1) * (MAX_SB_SIZE + 8));
+
+        for (int invert_mask = 0; invert_mask < 2; ++invert_mask) {
+          ref_ret = ref_func_(src8_ptr, src_stride, xoffset, yoffset, ref8_ptr,
+                              ref_stride, second_pred8_ptr, msk_ptr, msk_stride,
+                              invert_mask, &ref_sse);
+          ASM_REGISTER_STATE_CHECK(
+              opt_ret = opt_func_(src8_ptr, src_stride, xoffset, yoffset,
+                                  ref8_ptr, ref_stride, second_pred8_ptr,
+                                  msk_ptr, msk_stride, invert_mask, &opt_sse));
+
+          if (opt_ret != ref_ret || opt_sse != ref_sse) {
+            err_count++;
+            if (first_failure == -1) {
+              first_failure = i;
+              first_failure_x = xoffset;
+              first_failure_y = yoffset;
+            }
+          }
+        }
+      }
+    }
+  }
+
+  EXPECT_EQ(0, err_count) << "Error: Masked Variance Test ExtremeValues,"
+                          << "C output doesn't match SSSE3 output. "
+                          << "First failed at test case " << first_failure
+                          << " x_offset = " << first_failure_x
+                          << " y_offset = " << first_failure_y;
+}
+
+using ::testing::make_tuple;
+
+#if HAVE_SSSE3
+
+const MaskedSubPixelVarianceParam sub_pel_var_test[] = {
+  make_tuple(&aom_masked_sub_pixel_variance128x128_ssse3,
+             &aom_masked_sub_pixel_variance128x128_c),
+  make_tuple(&aom_masked_sub_pixel_variance128x64_ssse3,
+             &aom_masked_sub_pixel_variance128x64_c),
+  make_tuple(&aom_masked_sub_pixel_variance64x128_ssse3,
+             &aom_masked_sub_pixel_variance64x128_c),
+  make_tuple(&aom_masked_sub_pixel_variance64x64_ssse3,
+             &aom_masked_sub_pixel_variance64x64_c),
+  make_tuple(&aom_masked_sub_pixel_variance64x32_ssse3,
+             &aom_masked_sub_pixel_variance64x32_c),
+  make_tuple(&aom_masked_sub_pixel_variance32x64_ssse3,
+             &aom_masked_sub_pixel_variance32x64_c),
+  make_tuple(&aom_masked_sub_pixel_variance32x32_ssse3,
+             &aom_masked_sub_pixel_variance32x32_c),
+  make_tuple(&aom_masked_sub_pixel_variance32x16_ssse3,
+             &aom_masked_sub_pixel_variance32x16_c),
+  make_tuple(&aom_masked_sub_pixel_variance16x32_ssse3,
+             &aom_masked_sub_pixel_variance16x32_c),
+  make_tuple(&aom_masked_sub_pixel_variance16x16_ssse3,
+             &aom_masked_sub_pixel_variance16x16_c),
+  make_tuple(&aom_masked_sub_pixel_variance16x8_ssse3,
+             &aom_masked_sub_pixel_variance16x8_c),
+  make_tuple(&aom_masked_sub_pixel_variance8x16_ssse3,
+             &aom_masked_sub_pixel_variance8x16_c),
+  make_tuple(&aom_masked_sub_pixel_variance8x8_ssse3,
+             &aom_masked_sub_pixel_variance8x8_c),
+  make_tuple(&aom_masked_sub_pixel_variance8x4_ssse3,
+             &aom_masked_sub_pixel_variance8x4_c),
+  make_tuple(&aom_masked_sub_pixel_variance4x8_ssse3,
+             &aom_masked_sub_pixel_variance4x8_c),
+  make_tuple(&aom_masked_sub_pixel_variance4x4_ssse3,
+             &aom_masked_sub_pixel_variance4x4_c)
+};
+
+INSTANTIATE_TEST_CASE_P(SSSE3_C_COMPARE, MaskedSubPixelVarianceTest,
+                        ::testing::ValuesIn(sub_pel_var_test));
+
+const HighbdMaskedSubPixelVarianceParam hbd_sub_pel_var_test[] = {
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance128x128_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance128x128_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance128x64_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance128x64_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance64x128_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance64x128_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance64x64_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance64x64_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance64x32_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance64x32_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance32x64_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance32x64_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance32x32_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance32x32_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance32x16_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance32x16_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance16x32_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance16x32_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance16x16_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance16x16_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance16x8_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance16x8_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance8x16_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance8x16_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance8x8_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance8x8_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance8x4_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance8x4_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance4x8_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance4x8_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_8_masked_sub_pixel_variance4x4_ssse3,
+             &aom_highbd_8_masked_sub_pixel_variance4x4_c, AOM_BITS_8),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance128x128_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance128x128_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance128x64_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance128x64_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance64x128_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance64x128_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance64x64_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance64x64_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance64x32_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance64x32_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance32x64_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance32x64_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance32x32_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance32x32_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance32x16_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance32x16_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance16x32_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance16x32_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance16x16_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance16x16_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance16x8_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance16x8_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance8x16_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance8x16_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance8x8_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance8x8_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance8x4_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance8x4_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance4x8_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance4x8_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_10_masked_sub_pixel_variance4x4_ssse3,
+             &aom_highbd_10_masked_sub_pixel_variance4x4_c, AOM_BITS_10),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance128x128_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance128x128_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance128x64_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance128x64_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance64x128_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance64x128_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance64x64_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance64x64_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance64x32_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance64x32_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance32x64_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance32x64_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance32x32_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance32x32_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance32x16_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance32x16_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance16x32_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance16x32_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance16x16_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance16x16_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance16x8_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance16x8_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance8x16_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance8x16_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance8x8_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance8x8_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance8x4_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance8x4_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance4x8_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance4x8_c, AOM_BITS_12),
+  make_tuple(&aom_highbd_12_masked_sub_pixel_variance4x4_ssse3,
+             &aom_highbd_12_masked_sub_pixel_variance4x4_c, AOM_BITS_12)
+};
+
+INSTANTIATE_TEST_CASE_P(SSSE3_C_COMPARE, HighbdMaskedSubPixelVarianceTest,
+                        ::testing::ValuesIn(hbd_sub_pel_var_test));
+#endif  // HAVE_SSSE3
+}  // namespace
diff --git a/third_party/aom/test/md5_helper.h b/third_party/aom/test/md5_helper.h
new file mode 100644
index 0000000000..9443cb2624
--- /dev/null
+++ b/third_party/aom/test/md5_helper.h
@@ -0,0 +1,76 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_MD5_HELPER_H_
+#define AOM_TEST_MD5_HELPER_H_
+
+#include "aom/aom_decoder.h"
+#include "common/md5_utils.h"
+
+namespace libaom_test {
+class MD5 {
+ public:
+  MD5() { MD5Init(&md5_); }
+
+  void Add(const aom_image_t *img) {
+    for (int plane = 0; plane < 3; ++plane) {
+      const uint8_t *buf = img->planes[plane];
+      // Calculate the width and height to do the md5 check. For the chroma
+      // plane, we never want to round down and thus skip a pixel so if
+      // we are shifting by 1 (chroma_shift) we add 1 before doing the shift.
+      // This works only for chroma_shift of 0 and 1.
+      const int bytes_per_sample =
+          (img->fmt & AOM_IMG_FMT_HIGHBITDEPTH) ? 2 : 1;
+      const int h =
+          plane ? (img->d_h + img->y_chroma_shift) >> img->y_chroma_shift
+                : img->d_h;
+      const int w =
+          (plane ? (img->d_w + img->x_chroma_shift) >> img->x_chroma_shift
+                 : img->d_w) *
+          bytes_per_sample;
+
+      for (int y = 0; y < h; ++y) {
+        MD5Update(&md5_, buf, w);
+        buf += img->stride[plane];
+      }
+    }
+  }
+
+  void Add(const uint8_t *data, size_t size) {
+    MD5Update(&md5_, data, static_cast<uint32_t>(size));
+  }
+
+  const char *Get(void) {
+    static const char hex[16] = {
+      '0', '1', '2', '3', '4', '5', '6', '7',
+      '8', '9', 'a', 'b', 'c', 'd', 'e', 'f',
+    };
+    uint8_t tmp[16];
+    MD5Context ctx_tmp = md5_;
+
+    MD5Final(tmp, &ctx_tmp);
+    for (int i = 0; i < 16; i++) {
+      res_[i * 2 + 0] = hex[tmp[i] >> 4];
+      res_[i * 2 + 1] = hex[tmp[i] & 0xf];
+    }
+    res_[32] = 0;
+
+    return res_;
+  }
+
+ protected:
+  char res_[33];
+  MD5Context md5_;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_MD5_HELPER_H_
diff --git a/third_party/aom/test/metrics_template.html b/third_party/aom/test/metrics_template.html
new file mode 100644
index 0000000000..b57c62314a
--- /dev/null
+++ b/third_party/aom/test/metrics_template.html
@@ -0,0 +1,422 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<title>Video Codec Test Results</title>
+<style type="text/css">
+<!-- Begin 960 reset -->
+a,abbr,acronym,address,applet,article,aside,audio,b,big,blockquote,body,canvas,caption,center,cite,c
+ode,dd,del,details,dfn,dialog,div,dl,dt,em,embed,fieldset,figcaption,figure,font,footer,form,h1,h2,h
+3,h4,h5,h6,header,hgroup,hr,html,i,iframe,img,ins,kbd,label,legend,li,mark,menu,meter,nav,object,ol,
+output,p,pre,progress,q,rp,rt,ruby,s,samp,section,small,span,strike,strong,sub,summary,sup,table,tbo
+dy,td,tfoot,th,thead,time,tr,tt,u,ul,var,video,xmp{border:0;margin:0;padding:0;font-size:100%}html,b
+ody{height:100%}article,aside,details,figcaption,figure,footer,header,hgroup,menu,nav,section{displa
+y:block}b,strong{font-weight:bold}img{color:transparent;font-size:0;vertical-align:middle;-ms-interp
+olation-mode:bicubic}ol,ul{list-style:none}li{display:list-item}table{border-collapse:collapse;borde
+r-spacing:0}th,td,caption{font-weight:normal;vertical-align:top;text-align:left}q{quotes:none}q:befo
+re,q:after{content:'';content:none}sub,sup,small{font-size:75%}sub,sup{line-height:0;position:relati
+ve;vertical-align:baseline}sub{bottom:-0.25em}sup{top:-0.5em}svg{overflow:hidden}
+<!-- End 960 reset -->
+<!-- Begin 960 text -->
+body{font:13px/1.5 'Helvetica Neue',Arial,'Liberation Sans',FreeSans,sans-serif}pre,code{font-family
+:'DejaVu Sans Mono',Menlo,Consolas,monospace}hr{border:0 #ccc solid;border-top-width:1px;clear:both;
+height:0}h1{font-size:25px}h2{font-size:23px}h3{font-size:21px}h4{font-size:19px}h5{font-size:17px}h
+6{font-size:15px}ol{list-style:decimal}ul{list-style:disc}li{margin-left:30px}p,dl,hr,h1,h2,h3,h4,h5
+,h6,ol,ul,pre,table,address,fieldset,figure{margin-bottom:20px}
+<!-- End 960 text -->
+<!-- Begin 960 grid (fluid variant)
+     12 columns, 1152px total width
+     http://960.gs/ | http://grids.heroku.com/ -->
+.container_12{width:92%;margin-left:4%;margin-right:4%}.grid_1,.grid_2,.grid_3,.grid_4,.grid_5,.grid
+_6,.grid_7,.grid_8,.grid_9,.grid_10,.grid_11,.grid_12{display:inline;float:left;position:relative;ma
+rgin-left:1%;margin-right:1%}.alpha{margin-left:0}.omega{margin-right:0}.container_12 .grid_1{width:
+6.333%}.container_12 .grid_2{width:14.667%}.container_12 .grid_3{width:23.0%}.container_12 .grid_4{w
+idth:31.333%}.container_12 .grid_5{width:39.667%}.container_12 .grid_6{width:48.0%}.container_12 .gr
+id_7{width:56.333%}.container_12 .grid_8{width:64.667%}.container_12 .grid_9{width:73.0%}.container_
+12 .grid_10{width:81.333%}.container_12 .grid_11{width:89.667%}.container_12 .grid_12{width:98.0%}.c
+ontainer_12 .prefix_1{padding-left:8.333%}.container_12 .prefix_2{padding-left:16.667%}.container_12
+ .prefix_3{padding-left:25.0%}.container_12 .prefix_4{padding-left:33.333%}.container_12 .prefix_5{p
+adding-left:41.667%}.container_12 .prefix_6{padding-left:50.0%}.container_12 .prefix_7{padding-left:
+58.333%}.container_12 .prefix_8{padding-left:66.667%}.container_12 .prefix_9{padding-left:75.0%}.con
+tainer_12 .prefix_10{padding-left:83.333%}.container_12 .prefix_11{padding-left:91.667%}.container_1
+2 .suffix_1{padding-right:8.333%}.container_12 .suffix_2{padding-right:16.667%}.container_12 .suffix
+_3{padding-right:25.0%}.container_12 .suffix_4{padding-right:33.333%}.container_12 .suffix_5{padding
+-right:41.667%}.container_12 .suffix_6{padding-right:50.0%}.container_12 .suffix_7{padding-right:58.
+333%}.container_12 .suffix_8{padding-right:66.667%}.container_12 .suffix_9{padding-right:75.0%}.cont
+ainer_12 .suffix_10{padding-right:83.333%}.container_12 .suffix_11{padding-right:91.667%}.container_
+12 .push_1{left:8.333%}.container_12 .push_2{left:16.667%}.container_12 .push_3{left:25.0%}.containe
+r_12 .push_4{left:33.333%}.container_12 .push_5{left:41.667%}.container_12 .push_6{left:50.0%}.conta
+iner_12 .push_7{left:58.333%}.container_12 .push_8{left:66.667%}.container_12 .push_9{left:75.0%}.co
+ntainer_12 .push_10{left:83.333%}.container_12 .push_11{left:91.667%}.container_12 .pull_1{left:-8.3
+33%}.container_12 .pull_2{left:-16.667%}.container_12 .pull_3{left:-25.0%}.container_12 .pull_4{left
+:-33.333%}.container_12 .pull_5{left:-41.667%}.container_12 .pull_6{left:-50.0%}.container_12 .pull_
+7{left:-58.333%}.container_12 .pull_8{left:-66.667%}.container_12 .pull_9{left:-75.0%}.container_12
+.pull_10{left:-83.333%}.container_12 .pull_11{left:-91.667%}.clear{clear:both;display:block;overflow
+:hidden;visibility:hidden;width:0;height:0}.clearfix:after{clear:both;content:' ';display:block;font
+-size:0;line-height:0;visibility:hidden;width:0;height:0}.clearfix{display:inline-block}* html .clea
+rfix{height:1%}.clearfix{display:block}
+<!-- End 960 grid -->
+
+div.metricgraph {
+
+}
+
+body {
+
+}
+
+div.header {
+  font-family: Arial, sans-serif;
+}
+
+div.header h2 {
+  margin: .5em auto;
+}
+
+div.radio {
+  font-family: Arial, sans-serif;
+  margin-bottom: 1em;
+}
+
+div.main {
+
+}
+
+div.cliplist {
+  font-family: Arial, sans-serif;
+  margin-top: 6px;
+}
+
+div.chartarea {
+  font-family: Arial, sans-serif;
+}
+
+div.indicators {
+  font-family: Arial, sans-serif;
+  font-size: 13px;
+  margin-top: 6px;
+  min-height: 600px;
+  background-color: #f7f7f7;
+}
+
+div.indicators div.content {
+  margin: 1em;
+}
+
+div.indicators div.content h5 {
+  font-size: 13px;
+  text-align: center;
+  margin: 0;
+}
+
+div.indicators div.content ul {
+  margin-left: 0;
+  padding-left: 0;
+  margin-top: 0;
+}
+
+div.indicators div.content ul li {
+  margin-left: 1.5em;
+}
+
+div.indicators div.content p:first-child {
+  margin-bottom: .5em;
+}
+
+span.google-visualization-table-sortind {
+  color: #000;
+}
+.header-style {
+  font-weight: bold;
+  border: 1px solid #fff;
+  background-color: #ccc;
+}
+
+td.header-style+td {
+
+}
+
+.orange-background {
+  background-color: orange;
+}
+
+.light-gray-background {
+  background-color: #f0f0f0;
+}
+</style>
+<script type="text/javascript" src="https://www.google.com/jsapi"></script>
+<script type="text/javascript">
+var chart_left   = 40;
+var chart_top    = 6;
+var chart_height = document.documentElement.clientHeight-100;
+var chart_width  = "100%";
+ftable='filestable_avg'
+var snrs = [];
+var filestable_dsnr = [];
+var filestable_drate = [];
+var filestable_avg = [];
+
+// Python template code replaces the following 2 lines.
+//%%metrics_js%%//
+//%%filestable_dpsnr%%//
+//%%filestable_avg%%//
+//%%filestable_drate%%//
+//%%snrs%%//
+
+var selected = 0
+var imagestr = '';
+var bettertable=0;
+var chart=0;
+var better=0;
+var metricdata=0;
+var metricView=0;
+var column=1;
+var formatter=0;
+
+function changeColumn(col) {
+  column = col;
+  console.log(col)
+  draw_files();
+}
+
+function changeMetric(m) {
+  ftable=m
+  draw_files()
+}
+
+function setup_vis() {
+  chart = new google.visualization.ScatterChart(
+      document.getElementById("metricgraph"));
+
+  bettertable = new google.visualization.Table(
+      document.getElementById("bettertable"));
+
+  draw_files();
+  build_metrics_radio();
+}
+
+function build_metrics_radio() {
+  for (metric=1; metric < metrics.length; metric++) {
+    var rb = document.createElement('input');
+    var l = document.createElement('label');
+    rb.setAttribute('type','radio');
+    rb.setAttribute('name','metric');
+    rb.setAttribute('onClick', "changeColumn('"+metric.toString()+"')");
+    l.innerHTML = metrics[metric];
+    document.getElementById('metrics').appendChild(rb);
+    document.getElementById('metrics').appendChild(l);
+  }
+}
+
+function draw_files() {
+  var options = {'allowHtml': true, 'width': "100%", 'height': "50%"};
+  if (better != 0) delete better;
+
+  col=eval(ftable+'[column]')
+  better = new google.visualization.DataTable(col)
+
+  // Python Template code replaces the following line with a list of
+  // formatters.
+  if (ftable == 'filestable_dsnr')
+    formatter = new google.visualization.NumberFormat(
+      {fractionDigits: 4, suffix:" db"});
+  else
+    formatter = new google.visualization.NumberFormat(
+       {fractionDigits: 4, suffix:"%"});
+
+  //%%formatters%%//
+
+  bettertable.draw(better,options);
+  google.visualization.events.addListener(bettertable, 'select',
+                                          selectBetterHandler);
+  query_file()
+}
+
+function query_file() {
+  imagestr = better.getFormattedValue(selected, 0)
+  var metricjson = eval('(' + snrs[column][selected] + ')');
+  metricdata = new google.visualization.DataTable(metricjson, 0.6);
+  if( metricView != 0 ) delete metricView;
+  metricView = new google.visualization.DataView(metricdata);
+
+  chart.draw(metricView, {curveType:'function',
+      explorer: {},
+      chartArea:{left:chart_left, top:chart_top, width:chart_width,
+      height:chart_height-90},
+      hAxis:{title:"Datarate in kbps"},
+      vAxis:{title:"Quality in decibels", format: '##.0', textPosition: 'in'},
+      legend:{position:"in"}, title:imagestr, pointSize:2, lineWidth:1,
+      width:chart_width, height:chart_height-50 });
+
+  google.visualization.events.addListener(chart, 'select', chartSelect);
+  google.visualization.events.addListener(chart, 'onmouseover', chartMouseOver);
+  google.visualization.events.addListener(chart, 'onmouseout', chartMouseOut);
+}
+
+function chartMouseOut(e) {
+  statusbar = document.getElementById('status');
+  statusbar.style.display = 'none';
+}
+
+function chartMouseOver(e) {
+  pointDifference(e.row, e.column)
+}
+
+function pointDifference(row, col) {
+  if(!row || !col)
+    return;
+
+  var cols = metricdata.getNumberOfColumns();
+  var rows = metricdata.getNumberOfRows();
+
+  var sel_bitrate = metricView.getValue(row, 0 );
+  var sel_metric = metricView.getValue(row, col);
+
+  var message = '<ul>' + metricView.getColumnLabel(col) +
+     ' (' + sel_bitrate.toFixed(0) + ' kbps, ' + sel_metric.toFixed(2) + ')' + ' is ';
+
+
+  // col 0 is datarate
+  for( var i=1;i<cols;++i) {
+
+    var metric_greatest_thats_less = 0;
+    var rate_greatest_thats_less = 0;
+    var metric_smallest_thats_greater = 999;
+    var rate_smallest_thats_greater = 0;
+
+    if(i==col)
+      continue;
+
+    // Find the lowest metric for the column that's greater than sel_metric and
+    // the highest metric for this column that's less than the metric.
+    for(var line_count = 0; line_count < rows; ++line_count) {
+      this_metric = metricdata.getValue(line_count, i)
+      this_rate = metricdata.getValue(line_count, 0)
+      if(!this_metric)
+        continue;
+
+      if(this_metric > metric_greatest_thats_less &&
+         this_metric <= sel_metric) {
+        metric_greatest_thats_less = this_metric;
+        rate_greatest_thats_less = this_rate;
+      }
+      if(this_metric < metric_smallest_thats_greater &&
+        this_metric > sel_metric) {
+        metric_smallest_thats_greater = this_metric;
+        rate_smallest_thats_greater = this_rate;
+      }
+    }
+
+    if(rate_smallest_thats_greater == 0 || rate_greatest_thats_less == 0) {
+      message = message + " <li> Couldn't find a point on both sides.</li>"
+    } else {
+      metric_slope = ( rate_smallest_thats_greater - rate_greatest_thats_less) /
+          ( metric_smallest_thats_greater - metric_greatest_thats_less);
+
+      projected_rate = ( sel_metric - metric_greatest_thats_less) *
+          metric_slope + rate_greatest_thats_less;
+
+      difference = 100 * (projected_rate / sel_bitrate - 1);
+
+
+      if (difference > 0)
+        message = message + "<li>  " + difference.toFixed(2) +
+                  "% smaller than <em>" +
+                  metricdata.getColumnLabel(i) + "</em></li> "
+      else
+        message = message + "<li>  " + -difference.toFixed(2) +
+                  "% bigger than <em>" +
+                  metricdata.getColumnLabel(i) + "</em></li> "
+    }
+
+  }
+  message = message + "</ul>"
+  statusbar = document.getElementById('status');
+  statusbar.innerHTML = "<p>" + message + "</p>";
+  statusbar.style.display = 'block';
+}
+
+function chartSelect() {
+  var selection = chart.getSelection();
+  var message = '';
+  var min = metricView.getFormattedValue(selection[0].row, 0);
+  var max = metricView.getFormattedValue(selection[selection.length-1].row, 0);
+  var val = metricView.getFormattedValue(selection[0].row,selection[0].column);
+
+  pointDifference(selection[0].row, selection[0].column)
+  min = min / 3
+  max = max * 3
+  metricView.setRows(metricdata.getFilteredRows(
+      [{column: 0,minValue: min, maxValue:max}]));
+
+  chart.draw(metricView, {curveType:'function',
+      chartArea:{left:40, top:10, width:chart_width, height:chart_height - 110},
+      hAxis:{title:"datarate in kbps"}, vAxis:{title:"quality in decibels"},
+      legend:{position:"in"}, title:imagestr, pointSize:2, lineWidth:1,
+      width:chart_width, height:chart_height - 50});
+}
+
+function selectBetterHandler() {
+  var selection = bettertable.getSelection();
+  for (var i = 0; i < selection.length; i++) {
+    item = selection[i];
+  }
+  selected = item.row
+  query_file()
+}
+
+
+google.load('visualization', '1', {'packages' : ['corechart','table']});
+google.setOnLoadCallback(setup_vis);
+</script>
+</head>
+
+<body>
+
+  <div class="container_12">
+
+    <div class="grid_12 header">
+      <h2>Codec Comparison Results</h2>
+    </div>
+
+    <div class="grid_12 radio">
+
+      <form name="myform">
+        Method For Combining Points
+        <input type="radio" checked name="column" value="1"
+          onClick="changeMetric('filestable_avg')" />Average of bitrates difference
+        <input type="radio" name="column" value="2"
+          onClick="changeMetric('filestable_dsnr')" />BDSNR
+        <input type="radio" name="column" value="3"
+          onClick="changeMetric('filestable_drate')" />BDRATE
+      </form>
+
+      <form id="metrics" name="myform">
+      </form>
+
+    </div>
+
+    <div class="grid_12 main">
+
+      <div class="grid_5 alpha cliplist">
+        <div id="bettertable"></div>
+      </div>
+
+      <div class="grid_5 chartarea">
+        <div id="metricgraph"></div>
+      </div>
+
+      <div class="grid_2 omega indicators">
+        <div class="content">
+          <h5>Indicators</h5>
+          <hr>
+          <div id="status"></div>
+        </div>
+      </div>
+
+    </div>
+
+  </div>
+
+</body>
+</html>
diff --git a/third_party/aom/test/monochrome_test.cc b/third_party/aom/test/monochrome_test.cc
new file mode 100644
index 0000000000..ebccba5842
--- /dev/null
+++ b/third_party/aom/test/monochrome_test.cc
@@ -0,0 +1,130 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <climits>
+#include <vector>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/video_source.h"
+#include "test/util.h"
+
+namespace {
+
+class MonochromeTest
+    : public ::libaom_test::CodecTestWithParam<libaom_test::TestMode>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  MonochromeTest() : EncoderTest(GET_PARAM(0)), frame0_psnr_y_(0.) {}
+
+  virtual ~MonochromeTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+  }
+
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     aom_codec_pts_t pts) {
+    (void)pts;
+
+    // Get value of top-left corner pixel of U plane
+    int chroma_value = img.planes[AOM_PLANE_U][0];
+
+    bool is_chroma_constant =
+        ComparePlaneToValue(img, AOM_PLANE_U, chroma_value) &&
+        ComparePlaneToValue(img, AOM_PLANE_V, chroma_value);
+
+    // Chroma planes should be constant
+    EXPECT_TRUE(is_chroma_constant);
+
+    // Monochrome flag on image should be set
+    EXPECT_EQ(img.monochrome, 1);
+
+    chroma_value_list_.push_back(chroma_value);
+  }
+
+  // Returns true if all pixels on the plane are equal to value, and returns
+  // false otherwise.
+  bool ComparePlaneToValue(const aom_image_t &img, const int plane,
+                           const int value) {
+    const int w = aom_img_plane_width(&img, plane);
+    const int h = aom_img_plane_height(&img, plane);
+    const uint8_t *const buf = img.planes[plane];
+    const int stride = img.stride[plane];
+
+    for (int r = 0; r < h; ++r) {
+      for (int c = 0; c < w; ++c) {
+        if (buf[r * stride + c] != value) return false;
+      }
+    }
+    return true;
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    // Check that the initial Y PSNR value is 'high enough', and check that
+    // subsequent Y PSNR values are 'close' to this initial value.
+    if (frame0_psnr_y_ == 0.) {
+      frame0_psnr_y_ = pkt->data.psnr.psnr[1];
+      EXPECT_GT(frame0_psnr_y_, 29.);
+    }
+    EXPECT_NEAR(pkt->data.psnr.psnr[1], frame0_psnr_y_, 2.5);
+  }
+
+  std::vector<int> chroma_value_list_;
+  double frame0_psnr_y_;
+};
+
+TEST_P(MonochromeTest, TestMonochromeEncoding) {
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 5);
+
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  cfg_.g_w = 352;
+  cfg_.g_h = 288;
+
+  cfg_.rc_buf_initial_sz = 500;
+  cfg_.rc_buf_optimal_sz = 600;
+  cfg_.rc_buf_sz = 1000;
+  cfg_.rc_min_quantizer = 2;
+  cfg_.rc_max_quantizer = 56;
+  cfg_.rc_undershoot_pct = 50;
+  cfg_.rc_overshoot_pct = 50;
+  cfg_.rc_end_usage = AOM_CBR;
+  cfg_.kf_mode = AOM_KF_AUTO;
+  cfg_.g_lag_in_frames = 1;
+  cfg_.kf_min_dist = cfg_.kf_max_dist = 3000;
+  // Enable dropped frames.
+  cfg_.rc_dropframe_thresh = 1;
+  // Disable error_resilience mode.
+  cfg_.g_error_resilient = 0;
+  // Run at low bitrate.
+  cfg_.rc_target_bitrate = 40;
+  // Set monochrome encoding flag
+  cfg_.monochrome = 1;
+
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  // Check that the chroma planes are equal across all frames
+  std::vector<int>::const_iterator iter = chroma_value_list_.begin();
+  int initial_chroma_value = *iter;
+  for (; iter != chroma_value_list_.end(); ++iter) {
+    // Check that all decoded frames have the same constant chroma planes.
+    EXPECT_EQ(*iter, initial_chroma_value);
+  }
+}
+
+AV1_INSTANTIATE_TEST_CASE(MonochromeTest,
+                          ::testing::Values(::libaom_test::kTwoPassGood));
+
+}  // namespace
diff --git a/third_party/aom/test/motion_vector_test.cc b/third_party/aom/test/motion_vector_test.cc
new file mode 100644
index 0000000000..27eb938930
--- /dev/null
+++ b/third_party/aom/test/motion_vector_test.cc
@@ -0,0 +1,105 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/util.h"
+#include "test/yuv_video_source.h"
+
+namespace {
+#define MAX_EXTREME_MV 1
+#define MIN_EXTREME_MV 2
+
+// Encoding modes
+const libaom_test::TestMode kEncodingModeVectors[] = {
+  ::libaom_test::kTwoPassGood,
+  ::libaom_test::kOnePassGood,
+};
+
+// Encoding speeds
+const int kCpuUsedVectors[] = { 1, 5 };
+
+// MV test modes: 1 - always use maximum MV; 2 - always use minimum MV.
+const int kMVTestModes[] = { MAX_EXTREME_MV, MIN_EXTREME_MV };
+
+class MotionVectorTestLarge
+    : public ::libaom_test::CodecTestWith3Params<libaom_test::TestMode, int,
+                                                 int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  MotionVectorTestLarge()
+      : EncoderTest(GET_PARAM(0)), encoding_mode_(GET_PARAM(1)),
+        cpu_used_(GET_PARAM(2)), mv_test_mode_(GET_PARAM(3)) {}
+
+  virtual ~MotionVectorTestLarge() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(encoding_mode_);
+    if (encoding_mode_ != ::libaom_test::kRealTime) {
+      cfg_.g_lag_in_frames = 3;
+      cfg_.rc_end_usage = AOM_VBR;
+    } else {
+      cfg_.g_lag_in_frames = 0;
+      cfg_.rc_end_usage = AOM_CBR;
+      cfg_.rc_buf_sz = 1000;
+      cfg_.rc_buf_initial_sz = 500;
+      cfg_.rc_buf_optimal_sz = 600;
+    }
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_CPUUSED, cpu_used_);
+      encoder->Control(AV1E_ENABLE_MOTION_VECTOR_UNIT_TEST, mv_test_mode_);
+      if (encoding_mode_ != ::libaom_test::kRealTime) {
+        encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+        encoder->Control(AOME_SET_ARNR_MAXFRAMES, 7);
+        encoder->Control(AOME_SET_ARNR_STRENGTH, 5);
+      }
+    }
+  }
+
+  libaom_test::TestMode encoding_mode_;
+  int cpu_used_;
+  int mv_test_mode_;
+};
+
+TEST_P(MotionVectorTestLarge, OverallTest) {
+  int width = 3840;
+  int height = 2160;
+
+  // Reduce the test clip's resolution while testing on 32-bit system.
+  if (sizeof(void *) == 4) {
+    width = 2048;
+    height = 360;
+  }
+
+  cfg_.rc_target_bitrate = 24000;
+  cfg_.g_profile = 0;
+  init_flags_ = AOM_CODEC_USE_PSNR;
+
+  testing::internal::scoped_ptr<libaom_test::VideoSource> video;
+  video.reset(new libaom_test::YUVVideoSource(
+      "niklas_640_480_30.yuv", AOM_IMG_FMT_I420, width, height, 30, 1, 0, 3));
+
+  ASSERT_TRUE(video.get() != NULL);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get()));
+}
+
+AV1_INSTANTIATE_TEST_CASE(MotionVectorTestLarge,
+                          ::testing::ValuesIn(kEncodingModeVectors),
+                          ::testing::ValuesIn(kCpuUsedVectors),
+                          ::testing::ValuesIn(kMVTestModes));
+}  // namespace
diff --git a/third_party/aom/test/noise_model_test.cc b/third_party/aom/test/noise_model_test.cc
new file mode 100644
index 0000000000..b5b387e313
--- /dev/null
+++ b/third_party/aom/test/noise_model_test.cc
@@ -0,0 +1,1343 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <math.h>
+#include <algorithm>
+#include <vector>
+
+#include "aom_dsp/noise_model.h"
+#include "aom_dsp/noise_util.h"
+#include "config/aom_dsp_rtcd.h"
+#include "test/acm_random.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+
+// Return normally distrbuted values with standard deviation of sigma.
+double randn(libaom_test::ACMRandom *random, double sigma) {
+  while (1) {
+    const double u = 2.0 * ((double)random->Rand31() /
+                            testing::internal::Random::kMaxRange) -
+                     1.0;
+    const double v = 2.0 * ((double)random->Rand31() /
+                            testing::internal::Random::kMaxRange) -
+                     1.0;
+    const double s = u * u + v * v;
+    if (s > 0 && s < 1) {
+      return sigma * (u * sqrt(-2.0 * log(s) / s));
+    }
+  }
+  return 0;
+}
+
+// Synthesizes noise using the auto-regressive filter of the given lag,
+// with the provided n coefficients sampled at the given coords.
+void noise_synth(libaom_test::ACMRandom *random, int lag, int n,
+                 const int (*coords)[2], const double *coeffs, double *data,
+                 int w, int h) {
+  const int pad_size = 3 * lag;
+  const int padded_w = w + pad_size;
+  const int padded_h = h + pad_size;
+  int x = 0, y = 0;
+  std::vector<double> padded(padded_w * padded_h);
+
+  for (y = 0; y < padded_h; ++y) {
+    for (x = 0; x < padded_w; ++x) {
+      padded[y * padded_w + x] = randn(random, 1.0);
+    }
+  }
+  for (y = lag; y < padded_h; ++y) {
+    for (x = lag; x < padded_w; ++x) {
+      double sum = 0;
+      int i = 0;
+      for (i = 0; i < n; ++i) {
+        const int dx = coords[i][0];
+        const int dy = coords[i][1];
+        sum += padded[(y + dy) * padded_w + (x + dx)] * coeffs[i];
+      }
+      padded[y * padded_w + x] += sum;
+    }
+  }
+  // Copy over the padded rows to the output
+  for (y = 0; y < h; ++y) {
+    memcpy(data + y * w, &padded[0] + y * padded_w, sizeof(*data) * w);
+  }
+}
+
+std::vector<float> get_noise_psd(double *noise, int width, int height,
+                                 int block_size) {
+  float *block =
+      (float *)aom_memalign(32, block_size * block_size * sizeof(block));
+  std::vector<float> psd(block_size * block_size);
+  int num_blocks = 0;
+  struct aom_noise_tx_t *tx = aom_noise_tx_malloc(block_size);
+  for (int y = 0; y <= height - block_size; y += block_size / 2) {
+    for (int x = 0; x <= width - block_size; x += block_size / 2) {
+      for (int yy = 0; yy < block_size; ++yy) {
+        for (int xx = 0; xx < block_size; ++xx) {
+          block[yy * block_size + xx] = (float)noise[(y + yy) * width + x + xx];
+        }
+      }
+      aom_noise_tx_forward(tx, &block[0]);
+      aom_noise_tx_add_energy(tx, &psd[0]);
+      num_blocks++;
+    }
+  }
+  for (int yy = 0; yy < block_size; ++yy) {
+    for (int xx = 0; xx <= block_size / 2; ++xx) {
+      psd[yy * block_size + xx] /= num_blocks;
+    }
+  }
+  // Fill in the data that is missing due to symmetries
+  for (int xx = 1; xx < block_size / 2; ++xx) {
+    psd[(block_size - xx)] = psd[xx];
+  }
+  for (int yy = 1; yy < block_size; ++yy) {
+    for (int xx = 1; xx < block_size / 2; ++xx) {
+      psd[(block_size - yy) * block_size + (block_size - xx)] =
+          psd[yy * block_size + xx];
+    }
+  }
+  aom_noise_tx_free(tx);
+  aom_free(block);
+  return psd;
+}
+
+}  // namespace
+
+TEST(NoiseStrengthSolver, GetCentersTwoBins) {
+  aom_noise_strength_solver_t solver;
+  aom_noise_strength_solver_init(&solver, 2, 8);
+  EXPECT_NEAR(0, aom_noise_strength_solver_get_center(&solver, 0), 1e-5);
+  EXPECT_NEAR(255, aom_noise_strength_solver_get_center(&solver, 1), 1e-5);
+  aom_noise_strength_solver_free(&solver);
+}
+
+TEST(NoiseStrengthSolver, GetCentersTwoBins10bit) {
+  aom_noise_strength_solver_t solver;
+  aom_noise_strength_solver_init(&solver, 2, 10);
+  EXPECT_NEAR(0, aom_noise_strength_solver_get_center(&solver, 0), 1e-5);
+  EXPECT_NEAR(1023, aom_noise_strength_solver_get_center(&solver, 1), 1e-5);
+  aom_noise_strength_solver_free(&solver);
+}
+
+TEST(NoiseStrengthSolver, GetCenters256Bins) {
+  const int num_bins = 256;
+  aom_noise_strength_solver_t solver;
+  aom_noise_strength_solver_init(&solver, num_bins, 8);
+
+  for (int i = 0; i < 256; ++i) {
+    EXPECT_NEAR(i, aom_noise_strength_solver_get_center(&solver, i), 1e-5);
+  }
+  aom_noise_strength_solver_free(&solver);
+}
+
+// Tests that the noise strength solver returns the identity transform when
+// given identity-like constraints.
+TEST(NoiseStrengthSolver, ObserveIdentity) {
+  const int num_bins = 256;
+  aom_noise_strength_solver_t solver;
+  EXPECT_EQ(1, aom_noise_strength_solver_init(&solver, num_bins, 8));
+
+  // We have to add a big more strength to constraints at the boundary to
+  // overcome any regularization.
+  for (int j = 0; j < 5; ++j) {
+    aom_noise_strength_solver_add_measurement(&solver, 0, 0);
+    aom_noise_strength_solver_add_measurement(&solver, 255, 255);
+  }
+  for (int i = 0; i < 256; ++i) {
+    aom_noise_strength_solver_add_measurement(&solver, i, i);
+  }
+  EXPECT_EQ(1, aom_noise_strength_solver_solve(&solver));
+  for (int i = 2; i < num_bins - 2; ++i) {
+    EXPECT_NEAR(i, solver.eqns.x[i], 0.1);
+  }
+
+  aom_noise_strength_lut_t lut;
+  EXPECT_EQ(1, aom_noise_strength_solver_fit_piecewise(&solver, 2, &lut));
+
+  ASSERT_EQ(2, lut.num_points);
+  EXPECT_NEAR(0.0, lut.points[0][0], 1e-5);
+  EXPECT_NEAR(0.0, lut.points[0][1], 0.5);
+  EXPECT_NEAR(255.0, lut.points[1][0], 1e-5);
+  EXPECT_NEAR(255.0, lut.points[1][1], 0.5);
+
+  aom_noise_strength_lut_free(&lut);
+  aom_noise_strength_solver_free(&solver);
+}
+
+TEST(NoiseStrengthSolver, SimplifiesCurve) {
+  const int num_bins = 256;
+  aom_noise_strength_solver_t solver;
+  EXPECT_EQ(1, aom_noise_strength_solver_init(&solver, num_bins, 8));
+
+  // Create a parabolic input
+  for (int i = 0; i < 256; ++i) {
+    const double x = (i - 127.5) / 63.5;
+    aom_noise_strength_solver_add_measurement(&solver, i, x * x);
+  }
+  EXPECT_EQ(1, aom_noise_strength_solver_solve(&solver));
+
+  // First try to fit an unconstrained lut
+  aom_noise_strength_lut_t lut;
+  EXPECT_EQ(1, aom_noise_strength_solver_fit_piecewise(&solver, -1, &lut));
+  ASSERT_LE(20, lut.num_points);
+  aom_noise_strength_lut_free(&lut);
+
+  // Now constrain the maximum number of points
+  const int kMaxPoints = 9;
+  EXPECT_EQ(1,
+            aom_noise_strength_solver_fit_piecewise(&solver, kMaxPoints, &lut));
+  ASSERT_EQ(kMaxPoints, lut.num_points);
+
+  // Check that the input parabola is still well represented
+  EXPECT_NEAR(0.0, lut.points[0][0], 1e-5);
+  EXPECT_NEAR(4.0, lut.points[0][1], 0.1);
+  for (int i = 1; i < lut.num_points - 1; ++i) {
+    const double x = (lut.points[i][0] - 128.) / 64.;
+    EXPECT_NEAR(x * x, lut.points[i][1], 0.1);
+  }
+  EXPECT_NEAR(255.0, lut.points[kMaxPoints - 1][0], 1e-5);
+
+  EXPECT_NEAR(4.0, lut.points[kMaxPoints - 1][1], 0.1);
+  aom_noise_strength_lut_free(&lut);
+  aom_noise_strength_solver_free(&solver);
+}
+
+TEST(NoiseStrengthLut, LutEvalSinglePoint) {
+  aom_noise_strength_lut_t lut;
+  ASSERT_TRUE(aom_noise_strength_lut_init(&lut, 1));
+  ASSERT_EQ(1, lut.num_points);
+  lut.points[0][0] = 0;
+  lut.points[0][1] = 1;
+  EXPECT_EQ(1, aom_noise_strength_lut_eval(&lut, -1));
+  EXPECT_EQ(1, aom_noise_strength_lut_eval(&lut, 0));
+  EXPECT_EQ(1, aom_noise_strength_lut_eval(&lut, 1));
+  aom_noise_strength_lut_free(&lut);
+}
+
+TEST(NoiseStrengthLut, LutEvalMultiPointInterp) {
+  const double kEps = 1e-5;
+  aom_noise_strength_lut_t lut;
+  ASSERT_TRUE(aom_noise_strength_lut_init(&lut, 4));
+  ASSERT_EQ(4, lut.num_points);
+
+  lut.points[0][0] = 0;
+  lut.points[0][1] = 0;
+
+  lut.points[1][0] = 1;
+  lut.points[1][1] = 1;
+
+  lut.points[2][0] = 2;
+  lut.points[2][1] = 1;
+
+  lut.points[3][0] = 100;
+  lut.points[3][1] = 1001;
+
+  // Test lower boundary
+  EXPECT_EQ(0, aom_noise_strength_lut_eval(&lut, -1));
+  EXPECT_EQ(0, aom_noise_strength_lut_eval(&lut, 0));
+
+  // Test first part that should be identity
+  EXPECT_NEAR(0.25, aom_noise_strength_lut_eval(&lut, 0.25), kEps);
+  EXPECT_NEAR(0.75, aom_noise_strength_lut_eval(&lut, 0.75), kEps);
+
+  // This is a constant section (should evaluate to 1)
+  EXPECT_NEAR(1.0, aom_noise_strength_lut_eval(&lut, 1.25), kEps);
+  EXPECT_NEAR(1.0, aom_noise_strength_lut_eval(&lut, 1.75), kEps);
+
+  // Test interpolation between to non-zero y coords.
+  EXPECT_NEAR(1, aom_noise_strength_lut_eval(&lut, 2), kEps);
+  EXPECT_NEAR(251, aom_noise_strength_lut_eval(&lut, 26.5), kEps);
+  EXPECT_NEAR(751, aom_noise_strength_lut_eval(&lut, 75.5), kEps);
+
+  // Test upper boundary
+  EXPECT_EQ(1001, aom_noise_strength_lut_eval(&lut, 100));
+  EXPECT_EQ(1001, aom_noise_strength_lut_eval(&lut, 101));
+
+  aom_noise_strength_lut_free(&lut);
+}
+
+TEST(NoiseModel, InitSuccessWithValidSquareShape) {
+  aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, 2, 8, 0 };
+  aom_noise_model_t model;
+
+  EXPECT_TRUE(aom_noise_model_init(&model, params));
+
+  const int kNumCoords = 12;
+  const int kCoords[][2] = { { -2, -2 }, { -1, -2 }, { 0, -2 },  { 1, -2 },
+                             { 2, -2 },  { -2, -1 }, { -1, -1 }, { 0, -1 },
+                             { 1, -1 },  { 2, -1 },  { -2, 0 },  { -1, 0 } };
+  EXPECT_EQ(kNumCoords, model.n);
+  for (int i = 0; i < kNumCoords; ++i) {
+    const int *coord = kCoords[i];
+    EXPECT_EQ(coord[0], model.coords[i][0]);
+    EXPECT_EQ(coord[1], model.coords[i][1]);
+  }
+  aom_noise_model_free(&model);
+}
+
+TEST(NoiseModel, InitSuccessWithValidDiamondShape) {
+  aom_noise_model_t model;
+  aom_noise_model_params_t params = { AOM_NOISE_SHAPE_DIAMOND, 2, 8, 0 };
+  EXPECT_TRUE(aom_noise_model_init(&model, params));
+  EXPECT_EQ(6, model.n);
+  const int kNumCoords = 6;
+  const int kCoords[][2] = { { 0, -2 }, { -1, -1 }, { 0, -1 },
+                             { 1, -1 }, { -2, 0 },  { -1, 0 } };
+  EXPECT_EQ(kNumCoords, model.n);
+  for (int i = 0; i < kNumCoords; ++i) {
+    const int *coord = kCoords[i];
+    EXPECT_EQ(coord[0], model.coords[i][0]);
+    EXPECT_EQ(coord[1], model.coords[i][1]);
+  }
+  aom_noise_model_free(&model);
+}
+
+TEST(NoiseModel, InitFailsWithTooLargeLag) {
+  aom_noise_model_t model;
+  aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, 10, 8, 0 };
+  EXPECT_FALSE(aom_noise_model_init(&model, params));
+  aom_noise_model_free(&model);
+}
+
+TEST(NoiseModel, InitFailsWithTooSmallLag) {
+  aom_noise_model_t model;
+  aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, 0, 8, 0 };
+  EXPECT_FALSE(aom_noise_model_init(&model, params));
+  aom_noise_model_free(&model);
+}
+
+TEST(NoiseModel, InitFailsWithInvalidShape) {
+  aom_noise_model_t model;
+  aom_noise_model_params_t params = { aom_noise_shape(100), 3, 8, 0 };
+  EXPECT_FALSE(aom_noise_model_init(&model, params));
+  aom_noise_model_free(&model);
+}
+
+// A container template class to hold a data type and extra arguments.
+// All of these args are bundled into one struct so that we can use
+// parameterized tests on combinations of supported data types
+// (uint8_t and uint16_t) and bit depths (8, 10, 12).
+template <typename T, int bit_depth, bool use_highbd>
+struct BitDepthParams {
+  typedef T data_type_t;
+  static const int kBitDepth = bit_depth;
+  static const bool kUseHighBD = use_highbd;
+};
+
+template <typename T>
+class FlatBlockEstimatorTest : public ::testing::Test, public T {
+ public:
+  virtual void SetUp() { random_.Reset(171); }
+  typedef std::vector<typename T::data_type_t> VecType;
+  VecType data_;
+  libaom_test::ACMRandom random_;
+};
+
+TYPED_TEST_CASE_P(FlatBlockEstimatorTest);
+
+TYPED_TEST_P(FlatBlockEstimatorTest, ExtractBlock) {
+  const int kBlockSize = 16;
+  aom_flat_block_finder_t flat_block_finder;
+  ASSERT_EQ(1, aom_flat_block_finder_init(&flat_block_finder, kBlockSize,
+                                          this->kBitDepth, this->kUseHighBD));
+  const double normalization = flat_block_finder.normalization;
+
+  // Test with an image of more than one block.
+  const int h = 2 * kBlockSize;
+  const int w = 2 * kBlockSize;
+  const int stride = 2 * kBlockSize;
+  this->data_.resize(h * stride, 128);
+
+  // Set up the (0,0) block to be a plane and the (0,1) block to be a
+  // checkerboard
+  const int shift = this->kBitDepth - 8;
+  for (int y = 0; y < kBlockSize; ++y) {
+    for (int x = 0; x < kBlockSize; ++x) {
+      this->data_[y * stride + x] = (-y + x + 128) << shift;
+      this->data_[y * stride + x + kBlockSize] =
+          ((x % 2 + y % 2) % 2 ? 128 - 20 : 128 + 20) << shift;
+    }
+  }
+  std::vector<double> block(kBlockSize * kBlockSize, 1);
+  std::vector<double> plane(kBlockSize * kBlockSize, 1);
+
+  // The block data should be a constant (zero) and the rest of the plane
+  // trend is covered in the plane data.
+  aom_flat_block_finder_extract_block(&flat_block_finder,
+                                      (uint8_t *)&this->data_[0], w, h, stride,
+                                      0, 0, &plane[0], &block[0]);
+  for (int y = 0; y < kBlockSize; ++y) {
+    for (int x = 0; x < kBlockSize; ++x) {
+      EXPECT_NEAR(0, block[y * kBlockSize + x], 1e-5);
+      EXPECT_NEAR((double)(this->data_[y * stride + x]) / normalization,
+                  plane[y * kBlockSize + x], 1e-5);
+    }
+  }
+
+  // The plane trend is a constant, and the block is a zero mean checkerboard.
+  aom_flat_block_finder_extract_block(&flat_block_finder,
+                                      (uint8_t *)&this->data_[0], w, h, stride,
+                                      kBlockSize, 0, &plane[0], &block[0]);
+  const int mid = 128 << shift;
+  for (int y = 0; y < kBlockSize; ++y) {
+    for (int x = 0; x < kBlockSize; ++x) {
+      EXPECT_NEAR(((double)this->data_[y * stride + x + kBlockSize] - mid) /
+                      normalization,
+                  block[y * kBlockSize + x], 1e-5);
+      EXPECT_NEAR(mid / normalization, plane[y * kBlockSize + x], 1e-5);
+    }
+  }
+  aom_flat_block_finder_free(&flat_block_finder);
+}
+
+TYPED_TEST_P(FlatBlockEstimatorTest, FindFlatBlocks) {
+  const int kBlockSize = 32;
+  aom_flat_block_finder_t flat_block_finder;
+  ASSERT_EQ(1, aom_flat_block_finder_init(&flat_block_finder, kBlockSize,
+                                          this->kBitDepth, this->kUseHighBD));
+
+  const int num_blocks_w = 8;
+  const int h = kBlockSize;
+  const int w = kBlockSize * num_blocks_w;
+  const int stride = w;
+  this->data_.resize(h * stride, 128);
+  std::vector<uint8_t> flat_blocks(num_blocks_w, 0);
+
+  const int shift = this->kBitDepth - 8;
+  for (int y = 0; y < kBlockSize; ++y) {
+    for (int x = 0; x < kBlockSize; ++x) {
+      // Block 0 (not flat): constant doesn't have enough variance to qualify
+      this->data_[y * stride + x + 0 * kBlockSize] = 128 << shift;
+
+      // Block 1 (not flat): too high of variance is hard to validate as flat
+      this->data_[y * stride + x + 1 * kBlockSize] =
+          ((uint8_t)(128 + randn(&this->random_, 5))) << shift;
+
+      // Block 2 (flat): slight checkerboard added to constant
+      const int check = (x % 2 + y % 2) % 2 ? -2 : 2;
+      this->data_[y * stride + x + 2 * kBlockSize] = (128 + check) << shift;
+
+      // Block 3 (flat): planar block with checkerboard pattern is also flat
+      this->data_[y * stride + x + 3 * kBlockSize] =
+          (y * 2 - x / 2 + 128 + check) << shift;
+
+      // Block 4 (flat): gaussian random with standard deviation 1.
+      this->data_[y * stride + x + 4 * kBlockSize] =
+          ((uint8_t)(randn(&this->random_, 1) + x + 128.0)) << shift;
+
+      // Block 5 (flat): gaussian random with standard deviation 2.
+      this->data_[y * stride + x + 5 * kBlockSize] =
+          ((uint8_t)(randn(&this->random_, 2) + y + 128.0)) << shift;
+
+      // Block 6 (not flat): too high of directional gradient.
+      const int strong_edge = x > kBlockSize / 2 ? 64 : 0;
+      this->data_[y * stride + x + 6 * kBlockSize] =
+          ((uint8_t)(randn(&this->random_, 1) + strong_edge + 128.0)) << shift;
+
+      // Block 7 (not flat): too high gradient.
+      const int big_check = ((x >> 2) % 2 + (y >> 2) % 2) % 2 ? -16 : 16;
+      this->data_[y * stride + x + 7 * kBlockSize] =
+          ((uint8_t)(randn(&this->random_, 1) + big_check + 128.0)) << shift;
+    }
+  }
+
+  EXPECT_EQ(4, aom_flat_block_finder_run(&flat_block_finder,
+                                         (uint8_t *)&this->data_[0], w, h,
+                                         stride, &flat_blocks[0]));
+
+  // First two blocks are not flat
+  EXPECT_EQ(0, flat_blocks[0]);
+  EXPECT_EQ(0, flat_blocks[1]);
+
+  // Next 4 blocks are flat.
+  EXPECT_EQ(255, flat_blocks[2]);
+  EXPECT_EQ(255, flat_blocks[3]);
+  EXPECT_EQ(255, flat_blocks[4]);
+  EXPECT_EQ(255, flat_blocks[5]);
+
+  // Last 2 are not flat by threshold
+  EXPECT_EQ(0, flat_blocks[6]);
+  EXPECT_EQ(0, flat_blocks[7]);
+
+  // Add the noise from non-flat block 1 to every block.
+  for (int y = 0; y < kBlockSize; ++y) {
+    for (int x = 0; x < kBlockSize * num_blocks_w; ++x) {
+      this->data_[y * stride + x] +=
+          (this->data_[y * stride + x % kBlockSize + kBlockSize] -
+           (128 << shift));
+    }
+  }
+  // Now the scored selection will pick the one that is most likely flat (block
+  // 0)
+  EXPECT_EQ(1, aom_flat_block_finder_run(&flat_block_finder,
+                                         (uint8_t *)&this->data_[0], w, h,
+                                         stride, &flat_blocks[0]));
+  EXPECT_EQ(1, flat_blocks[0]);
+  EXPECT_EQ(0, flat_blocks[1]);
+  EXPECT_EQ(0, flat_blocks[2]);
+  EXPECT_EQ(0, flat_blocks[3]);
+  EXPECT_EQ(0, flat_blocks[4]);
+  EXPECT_EQ(0, flat_blocks[5]);
+  EXPECT_EQ(0, flat_blocks[6]);
+  EXPECT_EQ(0, flat_blocks[7]);
+
+  aom_flat_block_finder_free(&flat_block_finder);
+}
+
+REGISTER_TYPED_TEST_CASE_P(FlatBlockEstimatorTest, ExtractBlock,
+                           FindFlatBlocks);
+
+typedef ::testing::Types<BitDepthParams<uint8_t, 8, false>,   // lowbd
+                         BitDepthParams<uint16_t, 8, true>,   // lowbd in 16-bit
+                         BitDepthParams<uint16_t, 10, true>,  // highbd data
+                         BitDepthParams<uint16_t, 12, true> >
+    AllBitDepthParams;
+INSTANTIATE_TYPED_TEST_CASE_P(FlatBlockInstatiation, FlatBlockEstimatorTest,
+                              AllBitDepthParams);
+
+template <typename T>
+class NoiseModelUpdateTest : public ::testing::Test, public T {
+ public:
+  static const int kWidth = 128;
+  static const int kHeight = 128;
+  static const int kBlockSize = 16;
+  static const int kNumBlocksX = kWidth / kBlockSize;
+  static const int kNumBlocksY = kHeight / kBlockSize;
+
+  virtual void SetUp() {
+    const aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, 3,
+                                              T::kBitDepth, T::kUseHighBD };
+    ASSERT_TRUE(aom_noise_model_init(&model_, params));
+
+    random_.Reset(100171);
+
+    data_.resize(kWidth * kHeight * 3);
+    denoised_.resize(kWidth * kHeight * 3);
+    noise_.resize(kWidth * kHeight * 3);
+    renoise_.resize(kWidth * kHeight);
+    flat_blocks_.resize(kNumBlocksX * kNumBlocksY);
+
+    for (int c = 0, offset = 0; c < 3; ++c, offset += kWidth * kHeight) {
+      data_ptr_[c] = &data_[offset];
+      noise_ptr_[c] = &noise_[offset];
+      denoised_ptr_[c] = &denoised_[offset];
+      strides_[c] = kWidth;
+
+      data_ptr_raw_[c] = (uint8_t *)&data_[offset];
+      denoised_ptr_raw_[c] = (uint8_t *)&denoised_[offset];
+    }
+    chroma_sub_[0] = 0;
+    chroma_sub_[1] = 0;
+  }
+
+  int NoiseModelUpdate(int block_size = kBlockSize) {
+    return aom_noise_model_update(&model_, data_ptr_raw_, denoised_ptr_raw_,
+                                  kWidth, kHeight, strides_, chroma_sub_,
+                                  &flat_blocks_[0], block_size);
+  }
+
+  void TearDown() { aom_noise_model_free(&model_); }
+
+ protected:
+  aom_noise_model_t model_;
+  std::vector<typename T::data_type_t> data_;
+  std::vector<typename T::data_type_t> denoised_;
+
+  std::vector<double> noise_;
+  std::vector<double> renoise_;
+  std::vector<uint8_t> flat_blocks_;
+
+  typename T::data_type_t *data_ptr_[3];
+  typename T::data_type_t *denoised_ptr_[3];
+
+  double *noise_ptr_[3];
+  int strides_[3];
+  int chroma_sub_[2];
+  libaom_test::ACMRandom random_;
+
+ private:
+  uint8_t *data_ptr_raw_[3];
+  uint8_t *denoised_ptr_raw_[3];
+};
+
+TYPED_TEST_CASE_P(NoiseModelUpdateTest);
+
+TYPED_TEST_P(NoiseModelUpdateTest, UpdateFailsNoFlatBlocks) {
+  EXPECT_EQ(AOM_NOISE_STATUS_INSUFFICIENT_FLAT_BLOCKS,
+            this->NoiseModelUpdate());
+}
+
+TYPED_TEST_P(NoiseModelUpdateTest, UpdateSuccessForZeroNoiseAllFlat) {
+  this->flat_blocks_.assign(this->flat_blocks_.size(), 1);
+  this->denoised_.assign(this->denoised_.size(), 128);
+  this->data_.assign(this->denoised_.size(), 128);
+  EXPECT_EQ(AOM_NOISE_STATUS_INTERNAL_ERROR, this->NoiseModelUpdate());
+}
+
+TYPED_TEST_P(NoiseModelUpdateTest, UpdateFailsBlockSizeTooSmall) {
+  this->flat_blocks_.assign(this->flat_blocks_.size(), 1);
+  this->denoised_.assign(this->denoised_.size(), 128);
+  this->data_.assign(this->denoised_.size(), 128);
+  EXPECT_EQ(AOM_NOISE_STATUS_INVALID_ARGUMENT,
+            this->NoiseModelUpdate(6 /* block_size=6 is too small*/));
+}
+
+TYPED_TEST_P(NoiseModelUpdateTest, UpdateSuccessForWhiteRandomNoise) {
+  aom_noise_model_t &model = this->model_;
+  const int kWidth = this->kWidth;
+  const int kHeight = this->kHeight;
+
+  const int shift = this->kBitDepth - 8;
+  for (int y = 0; y < kHeight; ++y) {
+    for (int x = 0; x < kWidth; ++x) {
+      this->data_ptr_[0][y * kWidth + x] =
+          int(64 + y + randn(&this->random_, 1)) << shift;
+      this->denoised_ptr_[0][y * kWidth + x] = (64 + y) << shift;
+      // Make the chroma planes completely correlated with the Y plane
+      for (int c = 1; c < 3; ++c) {
+        this->data_ptr_[c][y * kWidth + x] = this->data_ptr_[0][y * kWidth + x];
+        this->denoised_ptr_[c][y * kWidth + x] =
+            this->denoised_ptr_[0][y * kWidth + x];
+      }
+    }
+  }
+  this->flat_blocks_.assign(this->flat_blocks_.size(), 1);
+  EXPECT_EQ(AOM_NOISE_STATUS_OK, this->NoiseModelUpdate());
+
+  const double kCoeffEps = 0.075;
+  const int n = model.n;
+  for (int c = 0; c < 3; ++c) {
+    for (int i = 0; i < n; ++i) {
+      EXPECT_NEAR(0, model.latest_state[c].eqns.x[i], kCoeffEps);
+      EXPECT_NEAR(0, model.combined_state[c].eqns.x[i], kCoeffEps);
+    }
+    // The second and third channels are highly correlated with the first.
+    if (c > 0) {
+      ASSERT_EQ(n + 1, model.latest_state[c].eqns.n);
+      ASSERT_EQ(n + 1, model.combined_state[c].eqns.n);
+
+      EXPECT_NEAR(1, model.latest_state[c].eqns.x[n], kCoeffEps);
+      EXPECT_NEAR(1, model.combined_state[c].eqns.x[n], kCoeffEps);
+    }
+  }
+
+  // The fitted noise strength should be close to the standard deviation
+  // for all intensity bins.
+  const double kStdEps = 0.1;
+  const double normalize = 1 << shift;
+
+  for (int i = 0; i < model.latest_state[0].strength_solver.eqns.n; ++i) {
+    EXPECT_NEAR(1.0,
+                model.latest_state[0].strength_solver.eqns.x[i] / normalize,
+                kStdEps);
+    EXPECT_NEAR(1.0,
+                model.combined_state[0].strength_solver.eqns.x[i] / normalize,
+                kStdEps);
+  }
+
+  aom_noise_strength_lut_t lut;
+  aom_noise_strength_solver_fit_piecewise(
+      &model.latest_state[0].strength_solver, -1, &lut);
+  ASSERT_EQ(2, lut.num_points);
+  EXPECT_NEAR(0.0, lut.points[0][0], 1e-5);
+  EXPECT_NEAR(1.0, lut.points[0][1] / normalize, kStdEps);
+  EXPECT_NEAR((1 << this->kBitDepth) - 1, lut.points[1][0], 1e-5);
+  EXPECT_NEAR(1.0, lut.points[1][1] / normalize, kStdEps);
+  aom_noise_strength_lut_free(&lut);
+}
+
+TYPED_TEST_P(NoiseModelUpdateTest, UpdateSuccessForScaledWhiteNoise) {
+  aom_noise_model_t &model = this->model_;
+  const int kWidth = this->kWidth;
+  const int kHeight = this->kHeight;
+
+  const double kCoeffEps = 0.055;
+  const double kLowStd = 1;
+  const double kHighStd = 4;
+  const int shift = this->kBitDepth - 8;
+  for (int y = 0; y < kHeight; ++y) {
+    for (int x = 0; x < kWidth; ++x) {
+      for (int c = 0; c < 3; ++c) {
+        // The image data is bimodal:
+        // Bottom half has low intensity and low noise strength
+        // Top half has high intensity and high noise strength
+        const int avg = (y < kHeight / 2) ? 4 : 245;
+        const double std = (y < kHeight / 2) ? kLowStd : kHighStd;
+        this->data_ptr_[c][y * kWidth + x] =
+            ((uint8_t)std::min((int)255,
+                               (int)(2 + avg + randn(&this->random_, std))))
+            << shift;
+        this->denoised_ptr_[c][y * kWidth + x] = (2 + avg) << shift;
+      }
+    }
+  }
+  // Label all blocks as flat for the update
+  this->flat_blocks_.assign(this->flat_blocks_.size(), 1);
+  EXPECT_EQ(AOM_NOISE_STATUS_OK, this->NoiseModelUpdate());
+
+  const int n = model.n;
+  // The noise is uncorrelated spatially and with the y channel.
+  // All coefficients should be reasonably close to zero.
+  for (int c = 0; c < 3; ++c) {
+    for (int i = 0; i < n; ++i) {
+      EXPECT_NEAR(0, model.latest_state[c].eqns.x[i], kCoeffEps);
+      EXPECT_NEAR(0, model.combined_state[c].eqns.x[i], kCoeffEps);
+    }
+    if (c > 0) {
+      ASSERT_EQ(n + 1, model.latest_state[c].eqns.n);
+      ASSERT_EQ(n + 1, model.combined_state[c].eqns.n);
+
+      // The correlation to the y channel should be low (near zero)
+      EXPECT_NEAR(0, model.latest_state[c].eqns.x[n], kCoeffEps);
+      EXPECT_NEAR(0, model.combined_state[c].eqns.x[n], kCoeffEps);
+    }
+  }
+
+  // Noise strength should vary between kLowStd and kHighStd.
+  const double kStdEps = 0.15;
+  // We have to normalize fitted standard deviation based on bit depth.
+  const double normalize = (1 << shift);
+
+  ASSERT_EQ(20, model.latest_state[0].strength_solver.eqns.n);
+  for (int i = 0; i < model.latest_state[0].strength_solver.eqns.n; ++i) {
+    const double a = i / 19.0;
+    const double expected = (kLowStd * (1.0 - a) + kHighStd * a);
+    EXPECT_NEAR(expected,
+                model.latest_state[0].strength_solver.eqns.x[i] / normalize,
+                kStdEps);
+    EXPECT_NEAR(expected,
+                model.combined_state[0].strength_solver.eqns.x[i] / normalize,
+                kStdEps);
+  }
+
+  // If we fit a piecewise linear model, there should be two points:
+  // one near kLowStd at 0, and the other near kHighStd and 255.
+  aom_noise_strength_lut_t lut;
+  aom_noise_strength_solver_fit_piecewise(
+      &model.latest_state[0].strength_solver, 2, &lut);
+  ASSERT_EQ(2, lut.num_points);
+  EXPECT_NEAR(0, lut.points[0][0], 1e-4);
+  EXPECT_NEAR(kLowStd, lut.points[0][1] / normalize, kStdEps);
+  EXPECT_NEAR((1 << this->kBitDepth) - 1, lut.points[1][0], 1e-5);
+  EXPECT_NEAR(kHighStd, lut.points[1][1] / normalize, kStdEps);
+  aom_noise_strength_lut_free(&lut);
+}
+
+TYPED_TEST_P(NoiseModelUpdateTest, UpdateSuccessForCorrelatedNoise) {
+  aom_noise_model_t &model = this->model_;
+  const int kWidth = this->kWidth;
+  const int kHeight = this->kHeight;
+  const int kNumCoeffs = 24;
+  const double kStd = 4;
+  const double kStdEps = 0.3;
+  const double kCoeffEps = 0.065;
+  // Use different coefficients for each channel
+  const double kCoeffs[3][24] = {
+    { 0.02884, -0.03356, 0.00633,  0.01757,  0.02849,  -0.04620,
+      0.02833, -0.07178, 0.07076,  -0.11603, -0.10413, -0.16571,
+      0.05158, -0.07969, 0.02640,  -0.07191, 0.02530,  0.41968,
+      0.21450, -0.00702, -0.01401, -0.03676, -0.08713, 0.44196 },
+    { 0.00269, -0.01291, -0.01513, 0.07234,  0.03208,   0.00477,
+      0.00226, -0.00254, 0.03533,  0.12841,  -0.25970,  -0.06336,
+      0.05238, -0.00845, -0.03118, 0.09043,  -0.36558,  0.48903,
+      0.00595, -0.11938, 0.02106,  0.095956, -0.350139, 0.59305 },
+    { -0.00643, -0.01080, -0.01466, 0.06951, 0.03707,  -0.00482,
+      0.00817,  -0.00909, 0.02949,  0.12181, -0.25210, -0.07886,
+      0.06083,  -0.01210, -0.03108, 0.08944, -0.35875, 0.49150,
+      0.00415,  -0.12905, 0.02870,  0.09740, -0.34610, 0.58824 },
+  };
+
+  ASSERT_EQ(model.n, kNumCoeffs);
+  this->chroma_sub_[0] = this->chroma_sub_[1] = 1;
+
+  this->flat_blocks_.assign(this->flat_blocks_.size(), 1);
+
+  // Add different noise onto each plane
+  const int shift = this->kBitDepth - 8;
+  for (int c = 0; c < 3; ++c) {
+    noise_synth(&this->random_, model.params.lag, model.n, model.coords,
+                kCoeffs[c], this->noise_ptr_[c], kWidth, kHeight);
+    const int x_shift = c > 0 ? this->chroma_sub_[0] : 0;
+    const int y_shift = c > 0 ? this->chroma_sub_[1] : 0;
+    for (int y = 0; y < (kHeight >> y_shift); ++y) {
+      for (int x = 0; x < (kWidth >> x_shift); ++x) {
+        const uint8_t value = 64 + x / 2 + y / 4;
+        this->data_ptr_[c][y * kWidth + x] =
+            (uint8_t(value + this->noise_ptr_[c][y * kWidth + x] * kStd))
+            << shift;
+        this->denoised_ptr_[c][y * kWidth + x] = value << shift;
+      }
+    }
+  }
+  EXPECT_EQ(AOM_NOISE_STATUS_OK, this->NoiseModelUpdate());
+
+  // For the Y plane, the solved coefficients should be close to the original
+  const int n = model.n;
+  for (int c = 0; c < 3; ++c) {
+    for (int i = 0; i < n; ++i) {
+      EXPECT_NEAR(kCoeffs[c][i], model.latest_state[c].eqns.x[i], kCoeffEps);
+      EXPECT_NEAR(kCoeffs[c][i], model.combined_state[c].eqns.x[i], kCoeffEps);
+    }
+    // The chroma planes should be uncorrelated with the luma plane
+    if (c > 0) {
+      EXPECT_NEAR(0, model.latest_state[c].eqns.x[n], kCoeffEps);
+      EXPECT_NEAR(0, model.combined_state[c].eqns.x[n], kCoeffEps);
+    }
+    // Correlation between the coefficient vector and the fitted coefficients
+    // should be close to 1.
+    EXPECT_LT(0.98, aom_normalized_cross_correlation(
+                        model.latest_state[c].eqns.x, kCoeffs[c], kNumCoeffs));
+
+    noise_synth(&this->random_, model.params.lag, model.n, model.coords,
+                model.latest_state[c].eqns.x, &this->renoise_[0], kWidth,
+                kHeight);
+
+    EXPECT_TRUE(aom_noise_data_validate(&this->renoise_[0], kWidth, kHeight));
+  }
+
+  // Check fitted noise strength
+  const double normalize = 1 << shift;
+  for (int c = 0; c < 3; ++c) {
+    for (int i = 0; i < model.latest_state[c].strength_solver.eqns.n; ++i) {
+      EXPECT_NEAR(kStd,
+                  model.latest_state[c].strength_solver.eqns.x[i] / normalize,
+                  kStdEps);
+    }
+  }
+}
+
+TYPED_TEST_P(NoiseModelUpdateTest,
+             NoiseStrengthChangeSignalsDifferentNoiseType) {
+  aom_noise_model_t &model = this->model_;
+  const int kWidth = this->kWidth;
+  const int kHeight = this->kHeight;
+  const int kBlockSize = this->kBlockSize;
+  // Create a gradient image with std = 2 uncorrelated noise
+  const double kStd = 2;
+  const int shift = this->kBitDepth - 8;
+
+  for (int i = 0; i < kWidth * kHeight; ++i) {
+    const uint8_t val = (i % kWidth) < kWidth / 2 ? 64 : 192;
+    for (int c = 0; c < 3; ++c) {
+      this->noise_ptr_[c][i] = randn(&this->random_, 1);
+      this->data_ptr_[c][i] = ((uint8_t)(this->noise_ptr_[c][i] * kStd + val))
+                              << shift;
+      this->denoised_ptr_[c][i] = val << shift;
+    }
+  }
+  this->flat_blocks_.assign(this->flat_blocks_.size(), 1);
+  EXPECT_EQ(AOM_NOISE_STATUS_OK, this->NoiseModelUpdate());
+
+  const int kNumBlocks = kWidth * kHeight / kBlockSize / kBlockSize;
+  EXPECT_EQ(kNumBlocks, model.latest_state[0].strength_solver.num_equations);
+  EXPECT_EQ(kNumBlocks, model.latest_state[1].strength_solver.num_equations);
+  EXPECT_EQ(kNumBlocks, model.latest_state[2].strength_solver.num_equations);
+  EXPECT_EQ(kNumBlocks, model.combined_state[0].strength_solver.num_equations);
+  EXPECT_EQ(kNumBlocks, model.combined_state[1].strength_solver.num_equations);
+  EXPECT_EQ(kNumBlocks, model.combined_state[2].strength_solver.num_equations);
+
+  // Bump up noise by an insignificant amount
+  for (int i = 0; i < kWidth * kHeight; ++i) {
+    const uint8_t val = (i % kWidth) < kWidth / 2 ? 64 : 192;
+    this->data_ptr_[0][i] =
+        ((uint8_t)(this->noise_ptr_[0][i] * (kStd + 0.085) + val)) << shift;
+  }
+  EXPECT_EQ(AOM_NOISE_STATUS_OK, this->NoiseModelUpdate());
+
+  const double kARGainTolerance = 0.02;
+  for (int c = 0; c < 3; ++c) {
+    EXPECT_EQ(kNumBlocks, model.latest_state[c].strength_solver.num_equations);
+    EXPECT_EQ(15250, model.latest_state[c].num_observations);
+    EXPECT_NEAR(1, model.latest_state[c].ar_gain, kARGainTolerance);
+
+    EXPECT_EQ(2 * kNumBlocks,
+              model.combined_state[c].strength_solver.num_equations);
+    EXPECT_EQ(2 * 15250, model.combined_state[c].num_observations);
+    EXPECT_NEAR(1, model.combined_state[c].ar_gain, kARGainTolerance);
+  }
+
+  // Bump up the noise strength on half the image for one channel by a
+  // significant amount.
+  for (int i = 0; i < kWidth * kHeight; ++i) {
+    const uint8_t val = (i % kWidth) < kWidth / 2 ? 64 : 128;
+    if (i % kWidth < kWidth / 2) {
+      this->data_ptr_[0][i] =
+          ((uint8_t)(randn(&this->random_, kStd + 0.5) + val)) << shift;
+    }
+  }
+  EXPECT_EQ(AOM_NOISE_STATUS_DIFFERENT_NOISE_TYPE, this->NoiseModelUpdate());
+
+  // Since we didn't update the combined state, it should still be at 2 *
+  // num_blocks
+  EXPECT_EQ(kNumBlocks, model.latest_state[0].strength_solver.num_equations);
+  EXPECT_EQ(2 * kNumBlocks,
+            model.combined_state[0].strength_solver.num_equations);
+
+  // In normal operation, the "latest" estimate can be saved to the "combined"
+  // state for continued updates.
+  aom_noise_model_save_latest(&model);
+  for (int c = 0; c < 3; ++c) {
+    EXPECT_EQ(kNumBlocks, model.latest_state[c].strength_solver.num_equations);
+    EXPECT_EQ(15250, model.latest_state[c].num_observations);
+    EXPECT_NEAR(1, model.latest_state[c].ar_gain, kARGainTolerance);
+
+    EXPECT_EQ(kNumBlocks,
+              model.combined_state[c].strength_solver.num_equations);
+    EXPECT_EQ(15250, model.combined_state[c].num_observations);
+    EXPECT_NEAR(1, model.combined_state[c].ar_gain, kARGainTolerance);
+  }
+}
+
+TYPED_TEST_P(NoiseModelUpdateTest, NoiseCoeffsSignalsDifferentNoiseType) {
+  aom_noise_model_t &model = this->model_;
+  const int kWidth = this->kWidth;
+  const int kHeight = this->kHeight;
+  const double kCoeffs[2][24] = {
+    { 0.02884, -0.03356, 0.00633,  0.01757,  0.02849,  -0.04620,
+      0.02833, -0.07178, 0.07076,  -0.11603, -0.10413, -0.16571,
+      0.05158, -0.07969, 0.02640,  -0.07191, 0.02530,  0.41968,
+      0.21450, -0.00702, -0.01401, -0.03676, -0.08713, 0.44196 },
+    { 0.00269, -0.01291, -0.01513, 0.07234,  0.03208,   0.00477,
+      0.00226, -0.00254, 0.03533,  0.12841,  -0.25970,  -0.06336,
+      0.05238, -0.00845, -0.03118, 0.09043,  -0.36558,  0.48903,
+      0.00595, -0.11938, 0.02106,  0.095956, -0.350139, 0.59305 }
+  };
+
+  noise_synth(&this->random_, model.params.lag, model.n, model.coords,
+              kCoeffs[0], this->noise_ptr_[0], kWidth, kHeight);
+  for (int i = 0; i < kWidth * kHeight; ++i) {
+    this->data_ptr_[0][i] = (uint8_t)(128 + this->noise_ptr_[0][i]);
+  }
+  this->flat_blocks_.assign(this->flat_blocks_.size(), 1);
+  EXPECT_EQ(AOM_NOISE_STATUS_OK, this->NoiseModelUpdate());
+
+  // Now try with the second set of AR coefficients
+  noise_synth(&this->random_, model.params.lag, model.n, model.coords,
+              kCoeffs[1], this->noise_ptr_[0], kWidth, kHeight);
+  for (int i = 0; i < kWidth * kHeight; ++i) {
+    this->data_ptr_[0][i] = (uint8_t)(128 + this->noise_ptr_[0][i]);
+  }
+  EXPECT_EQ(AOM_NOISE_STATUS_DIFFERENT_NOISE_TYPE, this->NoiseModelUpdate());
+}
+REGISTER_TYPED_TEST_CASE_P(NoiseModelUpdateTest, UpdateFailsNoFlatBlocks,
+                           UpdateSuccessForZeroNoiseAllFlat,
+                           UpdateFailsBlockSizeTooSmall,
+                           UpdateSuccessForWhiteRandomNoise,
+                           UpdateSuccessForScaledWhiteNoise,
+                           UpdateSuccessForCorrelatedNoise,
+                           NoiseStrengthChangeSignalsDifferentNoiseType,
+                           NoiseCoeffsSignalsDifferentNoiseType);
+
+INSTANTIATE_TYPED_TEST_CASE_P(NoiseModelUpdateTestInstatiation,
+                              NoiseModelUpdateTest, AllBitDepthParams);
+
+TEST(NoiseModelGetGrainParameters, TestLagSize) {
+  aom_film_grain_t film_grain;
+  for (int lag = 1; lag <= 3; ++lag) {
+    aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, lag, 8, 0 };
+    aom_noise_model_t model;
+    EXPECT_TRUE(aom_noise_model_init(&model, params));
+    EXPECT_TRUE(aom_noise_model_get_grain_parameters(&model, &film_grain));
+    EXPECT_EQ(lag, film_grain.ar_coeff_lag);
+    aom_noise_model_free(&model);
+  }
+
+  aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, 4, 8, 0 };
+  aom_noise_model_t model;
+  EXPECT_TRUE(aom_noise_model_init(&model, params));
+  EXPECT_FALSE(aom_noise_model_get_grain_parameters(&model, &film_grain));
+  aom_noise_model_free(&model);
+}
+
+TEST(NoiseModelGetGrainParameters, TestARCoeffShiftBounds) {
+  struct TestCase {
+    double max_input_value;
+    int expected_ar_coeff_shift;
+    int expected_value;
+  };
+  const int lag = 1;
+  const int kNumTestCases = 19;
+  const TestCase test_cases[] = {
+    // Test cases for ar_coeff_shift = 9
+    { 0, 9, 0 },
+    { 0.125, 9, 64 },
+    { -0.125, 9, -64 },
+    { 0.2499, 9, 127 },
+    { -0.25, 9, -128 },
+    // Test cases for ar_coeff_shift = 8
+    { 0.25, 8, 64 },
+    { -0.2501, 8, -64 },
+    { 0.499, 8, 127 },
+    { -0.5, 8, -128 },
+    // Test cases for ar_coeff_shift = 7
+    { 0.5, 7, 64 },
+    { -0.5001, 7, -64 },
+    { 0.999, 7, 127 },
+    { -1, 7, -128 },
+    // Test cases for ar_coeff_shift = 6
+    { 1.0, 6, 64 },
+    { -1.0001, 6, -64 },
+    { 2.0, 6, 127 },
+    { -2.0, 6, -128 },
+    { 4, 6, 127 },
+    { -4, 6, -128 },
+  };
+  aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, lag, 8, 0 };
+  aom_noise_model_t model;
+  EXPECT_TRUE(aom_noise_model_init(&model, params));
+
+  for (int i = 0; i < kNumTestCases; ++i) {
+    const TestCase &test_case = test_cases[i];
+    model.combined_state[0].eqns.x[0] = test_case.max_input_value;
+
+    aom_film_grain_t film_grain;
+    EXPECT_TRUE(aom_noise_model_get_grain_parameters(&model, &film_grain));
+    EXPECT_EQ(1, film_grain.ar_coeff_lag);
+    EXPECT_EQ(test_case.expected_ar_coeff_shift, film_grain.ar_coeff_shift);
+    EXPECT_EQ(test_case.expected_value, film_grain.ar_coeffs_y[0]);
+  }
+  aom_noise_model_free(&model);
+}
+
+TEST(NoiseModelGetGrainParameters, TestNoiseStrengthShiftBounds) {
+  struct TestCase {
+    double max_input_value;
+    int expected_scaling_shift;
+    int expected_value;
+  };
+  const int kNumTestCases = 10;
+  const TestCase test_cases[] = {
+    { 0, 11, 0 },      { 1, 11, 64 },     { 2, 11, 128 }, { 3.99, 11, 255 },
+    { 4, 10, 128 },    { 7.99, 10, 255 }, { 8, 9, 128 },  { 16, 8, 128 },
+    { 31.99, 8, 255 }, { 64, 8, 255 },  // clipped
+  };
+  const int lag = 1;
+  aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, lag, 8, 0 };
+  aom_noise_model_t model;
+  EXPECT_TRUE(aom_noise_model_init(&model, params));
+
+  for (int i = 0; i < kNumTestCases; ++i) {
+    const TestCase &test_case = test_cases[i];
+    aom_equation_system_t &eqns = model.combined_state[0].strength_solver.eqns;
+    // Set the fitted scale parameters to be a constant value.
+    for (int j = 0; j < eqns.n; ++j) {
+      eqns.x[j] = test_case.max_input_value;
+    }
+    aom_film_grain_t film_grain;
+    EXPECT_TRUE(aom_noise_model_get_grain_parameters(&model, &film_grain));
+    // We expect a single constant segemnt
+    EXPECT_EQ(test_case.expected_scaling_shift, film_grain.scaling_shift);
+    EXPECT_EQ(test_case.expected_value, film_grain.scaling_points_y[0][1]);
+    EXPECT_EQ(test_case.expected_value, film_grain.scaling_points_y[1][1]);
+  }
+  aom_noise_model_free(&model);
+}
+
+// The AR coefficients are the same inputs used to generate "Test 2" in the test
+// vectors
+TEST(NoiseModelGetGrainParameters, GetGrainParametersReal) {
+  const double kInputCoeffsY[] = { 0.0315,  0.0073,  0.0218,  0.00235, 0.00511,
+                                   -0.0222, 0.0627,  -0.022,  0.05575, -0.1816,
+                                   0.0107,  -0.1966, 0.00065, -0.0809, 0.04934,
+                                   -0.1349, -0.0352, 0.41772, 0.27973, 0.04207,
+                                   -0.0429, -0.1372, 0.06193, 0.52032 };
+  const double kInputCoeffsCB[] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  0,
+                                    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.5 };
+  const double kInputCoeffsCR[] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,   0,
+                                    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -0.5 };
+  const int kExpectedARCoeffsY[] = { 4,  1,   3,  0,   1,  -3,  8, -3,
+                                     7,  -23, 1,  -25, 0,  -10, 6, -17,
+                                     -5, 53,  36, 5,   -5, -18, 8, 67 };
+  const int kExpectedARCoeffsCB[] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 84 };
+  const int kExpectedARCoeffsCR[] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,   0,
+                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -126 };
+  // Scaling function is initialized analytically with a sqrt function.
+  const int kNumScalingPointsY = 12;
+  const int kExpectedScalingPointsY[][2] = {
+    { 0, 0 },     { 13, 44 },   { 27, 62 },   { 40, 76 },
+    { 54, 88 },   { 67, 98 },   { 94, 117 },  { 121, 132 },
+    { 148, 146 }, { 174, 159 }, { 201, 171 }, { 255, 192 },
+  };
+
+  const int lag = 3;
+  aom_noise_model_params_t params = { AOM_NOISE_SHAPE_SQUARE, lag, 8, 0 };
+  aom_noise_model_t model;
+  EXPECT_TRUE(aom_noise_model_init(&model, params));
+
+  // Setup the AR coeffs
+  memcpy(model.combined_state[0].eqns.x, kInputCoeffsY, sizeof(kInputCoeffsY));
+  memcpy(model.combined_state[1].eqns.x, kInputCoeffsCB,
+         sizeof(kInputCoeffsCB));
+  memcpy(model.combined_state[2].eqns.x, kInputCoeffsCR,
+         sizeof(kInputCoeffsCR));
+  for (int i = 0; i < model.combined_state[0].strength_solver.num_bins; ++i) {
+    const double x =
+        ((double)i) / (model.combined_state[0].strength_solver.num_bins - 1.0);
+    model.combined_state[0].strength_solver.eqns.x[i] = 6 * sqrt(x);
+    model.combined_state[1].strength_solver.eqns.x[i] = 3;
+    model.combined_state[2].strength_solver.eqns.x[i] = 2;
+
+    // Inject some observations into the strength solver, as during film grain
+    // parameter extraction an estimate of the average strength will be used to
+    // adjust correlation.
+    const int n = model.combined_state[0].strength_solver.num_bins;
+    for (int j = 0; j < model.combined_state[0].strength_solver.num_bins; ++j) {
+      model.combined_state[0].strength_solver.eqns.A[i * n + j] = 1;
+      model.combined_state[1].strength_solver.eqns.A[i * n + j] = 1;
+      model.combined_state[2].strength_solver.eqns.A[i * n + j] = 1;
+    }
+  }
+
+  aom_film_grain_t film_grain;
+  EXPECT_TRUE(aom_noise_model_get_grain_parameters(&model, &film_grain));
+  EXPECT_EQ(lag, film_grain.ar_coeff_lag);
+  EXPECT_EQ(3, film_grain.ar_coeff_lag);
+  EXPECT_EQ(7, film_grain.ar_coeff_shift);
+  EXPECT_EQ(10, film_grain.scaling_shift);
+  EXPECT_EQ(kNumScalingPointsY, film_grain.num_y_points);
+  EXPECT_EQ(1, film_grain.update_parameters);
+  EXPECT_EQ(1, film_grain.apply_grain);
+
+  const int kNumARCoeffs = 24;
+  for (int i = 0; i < kNumARCoeffs; ++i) {
+    EXPECT_EQ(kExpectedARCoeffsY[i], film_grain.ar_coeffs_y[i]);
+  }
+  for (int i = 0; i < kNumARCoeffs + 1; ++i) {
+    EXPECT_EQ(kExpectedARCoeffsCB[i], film_grain.ar_coeffs_cb[i]);
+  }
+  for (int i = 0; i < kNumARCoeffs + 1; ++i) {
+    EXPECT_EQ(kExpectedARCoeffsCR[i], film_grain.ar_coeffs_cr[i]);
+  }
+  for (int i = 0; i < kNumScalingPointsY; ++i) {
+    EXPECT_EQ(kExpectedScalingPointsY[i][0], film_grain.scaling_points_y[i][0]);
+    EXPECT_EQ(kExpectedScalingPointsY[i][1], film_grain.scaling_points_y[i][1]);
+  }
+
+  // CB strength should just be a piecewise segment
+  EXPECT_EQ(2, film_grain.num_cb_points);
+  EXPECT_EQ(0, film_grain.scaling_points_cb[0][0]);
+  EXPECT_EQ(255, film_grain.scaling_points_cb[1][0]);
+  EXPECT_EQ(96, film_grain.scaling_points_cb[0][1]);
+  EXPECT_EQ(96, film_grain.scaling_points_cb[1][1]);
+
+  // CR strength should just be a piecewise segment
+  EXPECT_EQ(2, film_grain.num_cr_points);
+  EXPECT_EQ(0, film_grain.scaling_points_cr[0][0]);
+  EXPECT_EQ(255, film_grain.scaling_points_cr[1][0]);
+  EXPECT_EQ(64, film_grain.scaling_points_cr[0][1]);
+  EXPECT_EQ(64, film_grain.scaling_points_cr[1][1]);
+
+  EXPECT_EQ(128, film_grain.cb_mult);
+  EXPECT_EQ(192, film_grain.cb_luma_mult);
+  EXPECT_EQ(256, film_grain.cb_offset);
+  EXPECT_EQ(128, film_grain.cr_mult);
+  EXPECT_EQ(192, film_grain.cr_luma_mult);
+  EXPECT_EQ(256, film_grain.cr_offset);
+  EXPECT_EQ(0, film_grain.chroma_scaling_from_luma);
+  EXPECT_EQ(0, film_grain.grain_scale_shift);
+
+  aom_noise_model_free(&model);
+}
+
+template <typename T>
+class WienerDenoiseTest : public ::testing::Test, public T {
+ public:
+  static void SetUpTestCase() { aom_dsp_rtcd(); }
+
+ protected:
+  void SetUp() {
+    static const float kNoiseLevel = 5.f;
+    static const float kStd = 4.0;
+    static const double kMaxValue = (1 << T::kBitDepth) - 1;
+
+    chroma_sub_[0] = 1;
+    chroma_sub_[1] = 1;
+    stride_[0] = kWidth;
+    stride_[1] = kWidth / 2;
+    stride_[2] = kWidth / 2;
+    for (int k = 0; k < 3; ++k) {
+      data_[k].resize(kWidth * kHeight);
+      denoised_[k].resize(kWidth * kHeight);
+      noise_psd_[k].resize(kBlockSize * kBlockSize);
+    }
+
+    const double kCoeffsY[] = { 0.0406, -0.116, -0.078, -0.152, 0.0033, -0.093,
+                                0.048,  0.404,  0.2353, -0.035, -0.093, 0.441 };
+    const int kCoords[12][2] = {
+      { -2, -2 }, { -1, -2 }, { 0, -2 }, { 1, -2 }, { 2, -2 }, { -2, -1 },
+      { -1, -1 }, { 0, -1 },  { 1, -1 }, { 2, -1 }, { -2, 0 }, { -1, 0 }
+    };
+    const int kLag = 2;
+    const int kLength = 12;
+    libaom_test::ACMRandom random;
+    std::vector<double> noise(kWidth * kHeight);
+    noise_synth(&random, kLag, kLength, kCoords, kCoeffsY, &noise[0], kWidth,
+                kHeight);
+    noise_psd_[0] = get_noise_psd(&noise[0], kWidth, kHeight, kBlockSize);
+    for (int i = 0; i < kBlockSize * kBlockSize; ++i) {
+      noise_psd_[0][i] = (float)(noise_psd_[0][i] * kStd * kStd * kScaleNoise *
+                                 kScaleNoise / (kMaxValue * kMaxValue));
+    }
+
+    float psd_value =
+        aom_noise_psd_get_default_value(kBlockSizeChroma, kNoiseLevel);
+    for (int i = 0; i < kBlockSizeChroma * kBlockSizeChroma; ++i) {
+      noise_psd_[1][i] = psd_value;
+      noise_psd_[2][i] = psd_value;
+    }
+    for (int y = 0; y < kHeight; ++y) {
+      for (int x = 0; x < kWidth; ++x) {
+        data_[0][y * stride_[0] + x] = (typename T::data_type_t)fclamp(
+            (x + noise[y * stride_[0] + x] * kStd) * kScaleNoise, 0, kMaxValue);
+      }
+    }
+
+    for (int c = 1; c < 3; ++c) {
+      for (int y = 0; y < (kHeight >> 1); ++y) {
+        for (int x = 0; x < (kWidth >> 1); ++x) {
+          data_[c][y * stride_[c] + x] = (typename T::data_type_t)fclamp(
+              (x + randn(&random, kStd)) * kScaleNoise, 0, kMaxValue);
+        }
+      }
+    }
+    for (int k = 0; k < 3; ++k) {
+      noise_psd_ptrs_[k] = &noise_psd_[k][0];
+    }
+  }
+  static const int kBlockSize = 32;
+  static const int kBlockSizeChroma = 16;
+  static const int kWidth = 256;
+  static const int kHeight = 256;
+  static const int kScaleNoise = 1 << (T::kBitDepth - 8);
+
+  std::vector<typename T::data_type_t> data_[3];
+  std::vector<typename T::data_type_t> denoised_[3];
+  std::vector<float> noise_psd_[3];
+  int chroma_sub_[2];
+  float *noise_psd_ptrs_[3];
+  int stride_[3];
+};
+
+TYPED_TEST_CASE_P(WienerDenoiseTest);
+
+TYPED_TEST_P(WienerDenoiseTest, InvalidBlockSize) {
+  const uint8_t *const data_ptrs[3] = {
+    reinterpret_cast<uint8_t *>(&this->data_[0][0]),
+    reinterpret_cast<uint8_t *>(&this->data_[1][0]),
+    reinterpret_cast<uint8_t *>(&this->data_[2][0]),
+  };
+  uint8_t *denoised_ptrs[3] = {
+    reinterpret_cast<uint8_t *>(&this->denoised_[0][0]),
+    reinterpret_cast<uint8_t *>(&this->denoised_[1][0]),
+    reinterpret_cast<uint8_t *>(&this->denoised_[2][0]),
+  };
+  EXPECT_EQ(0, aom_wiener_denoise_2d(data_ptrs, denoised_ptrs, this->kWidth,
+                                     this->kHeight, this->stride_,
+                                     this->chroma_sub_, this->noise_psd_ptrs_,
+                                     18, this->kBitDepth, this->kUseHighBD));
+  EXPECT_EQ(0, aom_wiener_denoise_2d(data_ptrs, denoised_ptrs, this->kWidth,
+                                     this->kHeight, this->stride_,
+                                     this->chroma_sub_, this->noise_psd_ptrs_,
+                                     48, this->kBitDepth, this->kUseHighBD));
+  EXPECT_EQ(0, aom_wiener_denoise_2d(data_ptrs, denoised_ptrs, this->kWidth,
+                                     this->kHeight, this->stride_,
+                                     this->chroma_sub_, this->noise_psd_ptrs_,
+                                     64, this->kBitDepth, this->kUseHighBD));
+}
+
+TYPED_TEST_P(WienerDenoiseTest, InvalidChromaSubsampling) {
+  const uint8_t *const data_ptrs[3] = {
+    reinterpret_cast<uint8_t *>(&this->data_[0][0]),
+    reinterpret_cast<uint8_t *>(&this->data_[1][0]),
+    reinterpret_cast<uint8_t *>(&this->data_[2][0]),
+  };
+  uint8_t *denoised_ptrs[3] = {
+    reinterpret_cast<uint8_t *>(&this->denoised_[0][0]),
+    reinterpret_cast<uint8_t *>(&this->denoised_[1][0]),
+    reinterpret_cast<uint8_t *>(&this->denoised_[2][0]),
+  };
+  int chroma_sub[2] = { 1, 0 };
+  EXPECT_EQ(0, aom_wiener_denoise_2d(data_ptrs, denoised_ptrs, this->kWidth,
+                                     this->kHeight, this->stride_, chroma_sub,
+                                     this->noise_psd_ptrs_, 32, this->kBitDepth,
+                                     this->kUseHighBD));
+
+  chroma_sub[0] = 0;
+  chroma_sub[1] = 1;
+  EXPECT_EQ(0, aom_wiener_denoise_2d(data_ptrs, denoised_ptrs, this->kWidth,
+                                     this->kHeight, this->stride_, chroma_sub,
+                                     this->noise_psd_ptrs_, 32, this->kBitDepth,
+                                     this->kUseHighBD));
+}
+
+TYPED_TEST_P(WienerDenoiseTest, GradientTest) {
+  const int kWidth = this->kWidth;
+  const int kHeight = this->kHeight;
+  const int kBlockSize = this->kBlockSize;
+  const uint8_t *const data_ptrs[3] = {
+    reinterpret_cast<uint8_t *>(&this->data_[0][0]),
+    reinterpret_cast<uint8_t *>(&this->data_[1][0]),
+    reinterpret_cast<uint8_t *>(&this->data_[2][0]),
+  };
+  uint8_t *denoised_ptrs[3] = {
+    reinterpret_cast<uint8_t *>(&this->denoised_[0][0]),
+    reinterpret_cast<uint8_t *>(&this->denoised_[1][0]),
+    reinterpret_cast<uint8_t *>(&this->denoised_[2][0]),
+  };
+  const int ret = aom_wiener_denoise_2d(
+      data_ptrs, denoised_ptrs, kWidth, kHeight, this->stride_,
+      this->chroma_sub_, this->noise_psd_ptrs_, this->kBlockSize,
+      this->kBitDepth, this->kUseHighBD);
+  EXPECT_EQ(1, ret);
+
+  // Check the noise on the denoised image (from the analytical gradient)
+  // and make sure that it is less than what we added.
+  for (int c = 0; c < 3; ++c) {
+    std::vector<double> measured_noise(kWidth * kHeight);
+
+    double var = 0;
+    const int shift = (c > 0);
+    for (int x = 0; x < (kWidth >> shift); ++x) {
+      for (int y = 0; y < (kHeight >> shift); ++y) {
+        const double diff = this->denoised_[c][y * this->stride_[c] + x] -
+                            x * this->kScaleNoise;
+        var += diff * diff;
+        measured_noise[y * kWidth + x] = diff;
+      }
+    }
+    var /= (kWidth * kHeight);
+    const double std = sqrt(std::max(0.0, var));
+    EXPECT_LE(std, 1.25f * this->kScaleNoise);
+    if (c == 0) {
+      std::vector<float> measured_psd =
+          get_noise_psd(&measured_noise[0], kWidth, kHeight, kBlockSize);
+      std::vector<double> measured_psd_d(kBlockSize * kBlockSize);
+      std::vector<double> noise_psd_d(kBlockSize * kBlockSize);
+      std::copy(measured_psd.begin(), measured_psd.end(),
+                measured_psd_d.begin());
+      std::copy(this->noise_psd_[0].begin(), this->noise_psd_[0].end(),
+                noise_psd_d.begin());
+      EXPECT_LT(
+          aom_normalized_cross_correlation(&measured_psd_d[0], &noise_psd_d[0],
+                                           (int)(noise_psd_d.size())),
+          0.35);
+    }
+  }
+}
+
+REGISTER_TYPED_TEST_CASE_P(WienerDenoiseTest, InvalidBlockSize,
+                           InvalidChromaSubsampling, GradientTest);
+
+INSTANTIATE_TYPED_TEST_CASE_P(WienerDenoiseTestInstatiation, WienerDenoiseTest,
+                              AllBitDepthParams);
diff --git a/third_party/aom/test/obmc_sad_test.cc b/third_party/aom/test/obmc_sad_test.cc
new file mode 100644
index 0000000000..6cef869618
--- /dev/null
+++ b/third_party/aom/test/obmc_sad_test.cc
@@ -0,0 +1,237 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/function_equivalence_test.h"
+#include "test/register_state_check.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_integer.h"
+
+#define MAX_SB_SQUARE (MAX_SB_SIZE * MAX_SB_SIZE)
+
+using libaom_test::FunctionEquivalenceTest;
+
+namespace {
+
+static const int kIterations = 1000;
+static const int kMaskMax = 64;
+
+typedef unsigned int (*ObmcSadF)(const uint8_t *pre, int pre_stride,
+                                 const int32_t *wsrc, const int32_t *mask);
+typedef libaom_test::FuncParam<ObmcSadF> TestFuncs;
+
+////////////////////////////////////////////////////////////////////////////////
+// 8 bit
+////////////////////////////////////////////////////////////////////////////////
+
+class ObmcSadTest : public FunctionEquivalenceTest<ObmcSadF> {};
+
+TEST_P(ObmcSadTest, RandomValues) {
+  DECLARE_ALIGNED(32, uint8_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    const int pre_stride = rng_(MAX_SB_SIZE + 1);
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      pre[i] = rng_.Rand8();
+      wsrc[i] = rng_.Rand8() * rng_(kMaskMax * kMaskMax + 1);
+      mask[i] = rng_(kMaskMax * kMaskMax + 1);
+    }
+
+    const unsigned int ref_res = params_.ref_func(pre, pre_stride, wsrc, mask);
+    unsigned int tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res =
+                                 params_.tst_func(pre, pre_stride, wsrc, mask));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+TEST_P(ObmcSadTest, ExtremeValues) {
+  DECLARE_ALIGNED(32, uint8_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < MAX_SB_SIZE && !HasFatalFailure(); ++iter) {
+    const int pre_stride = iter;
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      pre[i] = UINT8_MAX;
+      wsrc[i] = UINT8_MAX * kMaskMax * kMaskMax;
+      mask[i] = kMaskMax * kMaskMax;
+    }
+
+    const unsigned int ref_res = params_.ref_func(pre, pre_stride, wsrc, mask);
+    unsigned int tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res =
+                                 params_.tst_func(pre, pre_stride, wsrc, mask));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+#if HAVE_SSE4_1
+const ObmcSadTest::ParamType sse4_functions[] = {
+  TestFuncs(aom_obmc_sad128x128_c, aom_obmc_sad128x128_sse4_1),
+  TestFuncs(aom_obmc_sad128x64_c, aom_obmc_sad128x64_sse4_1),
+  TestFuncs(aom_obmc_sad64x128_c, aom_obmc_sad64x128_sse4_1),
+  TestFuncs(aom_obmc_sad64x64_c, aom_obmc_sad64x64_sse4_1),
+  TestFuncs(aom_obmc_sad64x32_c, aom_obmc_sad64x32_sse4_1),
+  TestFuncs(aom_obmc_sad32x64_c, aom_obmc_sad32x64_sse4_1),
+  TestFuncs(aom_obmc_sad32x32_c, aom_obmc_sad32x32_sse4_1),
+  TestFuncs(aom_obmc_sad32x16_c, aom_obmc_sad32x16_sse4_1),
+  TestFuncs(aom_obmc_sad16x32_c, aom_obmc_sad16x32_sse4_1),
+  TestFuncs(aom_obmc_sad16x16_c, aom_obmc_sad16x16_sse4_1),
+  TestFuncs(aom_obmc_sad16x8_c, aom_obmc_sad16x8_sse4_1),
+  TestFuncs(aom_obmc_sad8x16_c, aom_obmc_sad8x16_sse4_1),
+  TestFuncs(aom_obmc_sad8x8_c, aom_obmc_sad8x8_sse4_1),
+  TestFuncs(aom_obmc_sad8x4_c, aom_obmc_sad8x4_sse4_1),
+  TestFuncs(aom_obmc_sad4x8_c, aom_obmc_sad4x8_sse4_1),
+  TestFuncs(aom_obmc_sad4x4_c, aom_obmc_sad4x4_sse4_1)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, ObmcSadTest,
+                        ::testing::ValuesIn(sse4_functions));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+const ObmcSadTest::ParamType avx2_functions[] = {
+  TestFuncs(aom_obmc_sad128x128_c, aom_obmc_sad128x128_avx2),
+  TestFuncs(aom_obmc_sad128x64_c, aom_obmc_sad128x64_avx2),
+  TestFuncs(aom_obmc_sad64x128_c, aom_obmc_sad64x128_avx2),
+  TestFuncs(aom_obmc_sad64x64_c, aom_obmc_sad64x64_avx2),
+  TestFuncs(aom_obmc_sad64x32_c, aom_obmc_sad64x32_avx2),
+  TestFuncs(aom_obmc_sad32x64_c, aom_obmc_sad32x64_avx2),
+  TestFuncs(aom_obmc_sad32x32_c, aom_obmc_sad32x32_avx2),
+  TestFuncs(aom_obmc_sad32x16_c, aom_obmc_sad32x16_avx2),
+  TestFuncs(aom_obmc_sad16x32_c, aom_obmc_sad16x32_avx2),
+  TestFuncs(aom_obmc_sad16x16_c, aom_obmc_sad16x16_avx2),
+  TestFuncs(aom_obmc_sad16x8_c, aom_obmc_sad16x8_avx2),
+  TestFuncs(aom_obmc_sad8x16_c, aom_obmc_sad8x16_avx2),
+  TestFuncs(aom_obmc_sad8x8_c, aom_obmc_sad8x8_avx2),
+  TestFuncs(aom_obmc_sad8x4_c, aom_obmc_sad8x4_avx2),
+  TestFuncs(aom_obmc_sad4x8_c, aom_obmc_sad4x8_avx2),
+  TestFuncs(aom_obmc_sad4x4_c, aom_obmc_sad4x4_avx2)
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, ObmcSadTest, ::testing::ValuesIn(avx2_functions));
+#endif  // HAVE_AVX2
+
+////////////////////////////////////////////////////////////////////////////////
+// High bit-depth
+////////////////////////////////////////////////////////////////////////////////
+
+class ObmcSadHBDTest : public FunctionEquivalenceTest<ObmcSadF> {};
+
+TEST_P(ObmcSadHBDTest, RandomValues) {
+  DECLARE_ALIGNED(32, uint16_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    const int pre_stride = rng_(MAX_SB_SIZE + 1);
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      pre[i] = rng_(1 << 12);
+      wsrc[i] = rng_(1 << 12) * rng_(kMaskMax * kMaskMax + 1);
+      mask[i] = rng_(kMaskMax * kMaskMax + 1);
+    }
+
+    const unsigned int ref_res =
+        params_.ref_func(CONVERT_TO_BYTEPTR(pre), pre_stride, wsrc, mask);
+    unsigned int tst_res;
+    ASM_REGISTER_STATE_CHECK(
+        tst_res =
+            params_.tst_func(CONVERT_TO_BYTEPTR(pre), pre_stride, wsrc, mask));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+TEST_P(ObmcSadHBDTest, ExtremeValues) {
+  DECLARE_ALIGNED(32, uint16_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < MAX_SB_SIZE && !HasFatalFailure(); ++iter) {
+    const int pre_stride = iter;
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      pre[i] = (1 << 12) - 1;
+      wsrc[i] = ((1 << 12) - 1) * kMaskMax * kMaskMax;
+      mask[i] = kMaskMax * kMaskMax;
+    }
+
+    const unsigned int ref_res =
+        params_.ref_func(CONVERT_TO_BYTEPTR(pre), pre_stride, wsrc, mask);
+    unsigned int tst_res;
+    ASM_REGISTER_STATE_CHECK(
+        tst_res =
+            params_.tst_func(CONVERT_TO_BYTEPTR(pre), pre_stride, wsrc, mask));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+#if HAVE_SSE4_1
+ObmcSadHBDTest::ParamType sse4_functions_hbd[] = {
+  TestFuncs(aom_highbd_obmc_sad128x128_c, aom_highbd_obmc_sad128x128_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad128x64_c, aom_highbd_obmc_sad128x64_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad64x128_c, aom_highbd_obmc_sad64x128_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad64x64_c, aom_highbd_obmc_sad64x64_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad64x32_c, aom_highbd_obmc_sad64x32_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad32x64_c, aom_highbd_obmc_sad32x64_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad32x32_c, aom_highbd_obmc_sad32x32_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad32x16_c, aom_highbd_obmc_sad32x16_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad16x32_c, aom_highbd_obmc_sad16x32_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad16x16_c, aom_highbd_obmc_sad16x16_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad16x8_c, aom_highbd_obmc_sad16x8_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad8x16_c, aom_highbd_obmc_sad8x16_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad8x8_c, aom_highbd_obmc_sad8x8_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad8x4_c, aom_highbd_obmc_sad8x4_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad4x8_c, aom_highbd_obmc_sad4x8_sse4_1),
+  TestFuncs(aom_highbd_obmc_sad4x4_c, aom_highbd_obmc_sad4x4_sse4_1)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, ObmcSadHBDTest,
+                        ::testing::ValuesIn(sse4_functions_hbd));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+ObmcSadHBDTest::ParamType avx2_functions_hbd[] = {
+  TestFuncs(aom_highbd_obmc_sad128x128_c, aom_highbd_obmc_sad128x128_avx2),
+  TestFuncs(aom_highbd_obmc_sad128x64_c, aom_highbd_obmc_sad128x64_avx2),
+  TestFuncs(aom_highbd_obmc_sad64x128_c, aom_highbd_obmc_sad64x128_avx2),
+  TestFuncs(aom_highbd_obmc_sad64x64_c, aom_highbd_obmc_sad64x64_avx2),
+  TestFuncs(aom_highbd_obmc_sad64x32_c, aom_highbd_obmc_sad64x32_avx2),
+  TestFuncs(aom_highbd_obmc_sad32x64_c, aom_highbd_obmc_sad32x64_avx2),
+  TestFuncs(aom_highbd_obmc_sad32x32_c, aom_highbd_obmc_sad32x32_avx2),
+  TestFuncs(aom_highbd_obmc_sad32x16_c, aom_highbd_obmc_sad32x16_avx2),
+  TestFuncs(aom_highbd_obmc_sad16x32_c, aom_highbd_obmc_sad16x32_avx2),
+  TestFuncs(aom_highbd_obmc_sad16x16_c, aom_highbd_obmc_sad16x16_avx2),
+  TestFuncs(aom_highbd_obmc_sad16x8_c, aom_highbd_obmc_sad16x8_avx2),
+  TestFuncs(aom_highbd_obmc_sad8x16_c, aom_highbd_obmc_sad8x16_avx2),
+  TestFuncs(aom_highbd_obmc_sad8x8_c, aom_highbd_obmc_sad8x8_avx2),
+  TestFuncs(aom_highbd_obmc_sad8x4_c, aom_highbd_obmc_sad8x4_avx2),
+  TestFuncs(aom_highbd_obmc_sad4x8_c, aom_highbd_obmc_sad4x8_avx2),
+  TestFuncs(aom_highbd_obmc_sad4x4_c, aom_highbd_obmc_sad4x4_avx2)
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, ObmcSadHBDTest,
+                        ::testing::ValuesIn(avx2_functions_hbd));
+#endif  // HAVE_AVX2
+}  // namespace
diff --git a/third_party/aom/test/obmc_variance_test.cc b/third_party/aom/test/obmc_variance_test.cc
new file mode 100644
index 0000000000..4563b964aa
--- /dev/null
+++ b/third_party/aom/test/obmc_variance_test.cc
@@ -0,0 +1,345 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+
+#include "test/function_equivalence_test.h"
+#include "test/register_state_check.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_integer.h"
+
+#define MAX_SB_SQUARE (MAX_SB_SIZE * MAX_SB_SIZE)
+
+using libaom_test::ACMRandom;
+using libaom_test::FunctionEquivalenceTest;
+
+namespace {
+
+static const int kIterations = 1000;
+static const int kMaskMax = 64;
+
+typedef unsigned int (*ObmcVarF)(const uint8_t *pre, int pre_stride,
+                                 const int32_t *wsrc, const int32_t *mask,
+                                 unsigned int *sse);
+typedef libaom_test::FuncParam<ObmcVarF> TestFuncs;
+
+////////////////////////////////////////////////////////////////////////////////
+// 8 bit
+////////////////////////////////////////////////////////////////////////////////
+
+class ObmcVarianceTest : public FunctionEquivalenceTest<ObmcVarF> {};
+
+TEST_P(ObmcVarianceTest, RandomValues) {
+  DECLARE_ALIGNED(32, uint8_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    const int pre_stride = this->rng_(MAX_SB_SIZE + 1);
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      pre[i] = this->rng_.Rand8();
+      wsrc[i] = this->rng_.Rand8() * this->rng_(kMaskMax * kMaskMax + 1);
+      mask[i] = this->rng_(kMaskMax * kMaskMax + 1);
+    }
+
+    unsigned int ref_sse, tst_sse;
+    const unsigned int ref_res =
+        params_.ref_func(pre, pre_stride, wsrc, mask, &ref_sse);
+    unsigned int tst_res;
+    ASM_REGISTER_STATE_CHECK(
+        tst_res = params_.tst_func(pre, pre_stride, wsrc, mask, &tst_sse));
+
+    ASSERT_EQ(ref_res, tst_res);
+    ASSERT_EQ(ref_sse, tst_sse);
+  }
+}
+
+TEST_P(ObmcVarianceTest, ExtremeValues) {
+  DECLARE_ALIGNED(32, uint8_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < MAX_SB_SIZE && !HasFatalFailure(); ++iter) {
+    const int pre_stride = iter;
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      pre[i] = UINT8_MAX;
+      wsrc[i] = UINT8_MAX * kMaskMax * kMaskMax;
+      mask[i] = kMaskMax * kMaskMax;
+    }
+
+    unsigned int ref_sse, tst_sse;
+    const unsigned int ref_res =
+        params_.ref_func(pre, pre_stride, wsrc, mask, &ref_sse);
+    unsigned int tst_res;
+    ASM_REGISTER_STATE_CHECK(
+        tst_res = params_.tst_func(pre, pre_stride, wsrc, mask, &tst_sse));
+
+    ASSERT_EQ(ref_res, tst_res);
+    ASSERT_EQ(ref_sse, tst_sse);
+  }
+}
+
+TEST_P(ObmcVarianceTest, DISABLED_Speed) {
+  DECLARE_ALIGNED(32, uint8_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  const int pre_stride = this->rng_(MAX_SB_SIZE + 1);
+
+  for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+    pre[i] = this->rng_.Rand8();
+    wsrc[i] = this->rng_.Rand8() * this->rng_(kMaskMax * kMaskMax + 1);
+    mask[i] = this->rng_(kMaskMax * kMaskMax + 1);
+  }
+
+  const int num_loops = 1000000;
+  unsigned int ref_sse, tst_sse;
+  aom_usec_timer ref_timer, test_timer;
+
+  aom_usec_timer_start(&ref_timer);
+  for (int i = 0; i < num_loops; ++i) {
+    params_.ref_func(pre, pre_stride, wsrc, mask, &ref_sse);
+  }
+  aom_usec_timer_mark(&ref_timer);
+  const int elapsed_time_c =
+      static_cast<int>(aom_usec_timer_elapsed(&ref_timer));
+
+  aom_usec_timer_start(&test_timer);
+  for (int i = 0; i < num_loops; ++i) {
+    params_.tst_func(pre, pre_stride, wsrc, mask, &tst_sse);
+  }
+  aom_usec_timer_mark(&test_timer);
+  const int elapsed_time_simd =
+      static_cast<int>(aom_usec_timer_elapsed(&test_timer));
+
+  printf("c_time=%d \t simd_time=%d \t gain=%d \n", elapsed_time_c,
+         elapsed_time_simd, (elapsed_time_c / elapsed_time_simd));
+}
+
+#if HAVE_SSE4_1
+const ObmcVarianceTest::ParamType sse4_functions[] = {
+  TestFuncs(aom_obmc_variance128x128_c, aom_obmc_variance128x128_sse4_1),
+  TestFuncs(aom_obmc_variance128x64_c, aom_obmc_variance128x64_sse4_1),
+  TestFuncs(aom_obmc_variance64x128_c, aom_obmc_variance64x128_sse4_1),
+  TestFuncs(aom_obmc_variance64x64_c, aom_obmc_variance64x64_sse4_1),
+  TestFuncs(aom_obmc_variance64x32_c, aom_obmc_variance64x32_sse4_1),
+  TestFuncs(aom_obmc_variance32x64_c, aom_obmc_variance32x64_sse4_1),
+  TestFuncs(aom_obmc_variance32x32_c, aom_obmc_variance32x32_sse4_1),
+  TestFuncs(aom_obmc_variance32x16_c, aom_obmc_variance32x16_sse4_1),
+  TestFuncs(aom_obmc_variance16x32_c, aom_obmc_variance16x32_sse4_1),
+  TestFuncs(aom_obmc_variance16x16_c, aom_obmc_variance16x16_sse4_1),
+  TestFuncs(aom_obmc_variance16x8_c, aom_obmc_variance16x8_sse4_1),
+  TestFuncs(aom_obmc_variance8x16_c, aom_obmc_variance8x16_sse4_1),
+  TestFuncs(aom_obmc_variance8x8_c, aom_obmc_variance8x8_sse4_1),
+  TestFuncs(aom_obmc_variance8x4_c, aom_obmc_variance8x4_sse4_1),
+  TestFuncs(aom_obmc_variance4x8_c, aom_obmc_variance4x8_sse4_1),
+  TestFuncs(aom_obmc_variance4x4_c, aom_obmc_variance4x4_sse4_1)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, ObmcVarianceTest,
+                        ::testing::ValuesIn(sse4_functions));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+const ObmcVarianceTest::ParamType avx2_functions[] = {
+  TestFuncs(aom_obmc_variance128x128_c, aom_obmc_variance128x128_avx2),
+  TestFuncs(aom_obmc_variance128x64_c, aom_obmc_variance128x64_avx2),
+  TestFuncs(aom_obmc_variance64x128_c, aom_obmc_variance64x128_avx2),
+  TestFuncs(aom_obmc_variance64x64_c, aom_obmc_variance64x64_avx2),
+  TestFuncs(aom_obmc_variance64x32_c, aom_obmc_variance64x32_avx2),
+  TestFuncs(aom_obmc_variance32x64_c, aom_obmc_variance32x64_avx2),
+  TestFuncs(aom_obmc_variance32x32_c, aom_obmc_variance32x32_avx2),
+  TestFuncs(aom_obmc_variance32x16_c, aom_obmc_variance32x16_avx2),
+  TestFuncs(aom_obmc_variance16x32_c, aom_obmc_variance16x32_avx2),
+  TestFuncs(aom_obmc_variance16x16_c, aom_obmc_variance16x16_avx2),
+  TestFuncs(aom_obmc_variance16x8_c, aom_obmc_variance16x8_avx2),
+  TestFuncs(aom_obmc_variance8x16_c, aom_obmc_variance8x16_avx2),
+  TestFuncs(aom_obmc_variance8x8_c, aom_obmc_variance8x8_avx2),
+  TestFuncs(aom_obmc_variance8x4_c, aom_obmc_variance8x4_avx2),
+  TestFuncs(aom_obmc_variance4x8_c, aom_obmc_variance4x8_sse4_1),
+  TestFuncs(aom_obmc_variance4x4_c, aom_obmc_variance4x4_sse4_1)
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, ObmcVarianceTest,
+                        ::testing::ValuesIn(avx2_functions));
+#endif  // HAVE_AVX2
+
+////////////////////////////////////////////////////////////////////////////////
+// High bit-depth
+////////////////////////////////////////////////////////////////////////////////
+
+class ObmcVarianceHBDTest : public FunctionEquivalenceTest<ObmcVarF> {};
+
+TEST_P(ObmcVarianceHBDTest, RandomValues) {
+  DECLARE_ALIGNED(32, uint16_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    const int pre_stride = this->rng_(MAX_SB_SIZE + 1);
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      pre[i] = this->rng_(1 << params_.bit_depth);
+      wsrc[i] = this->rng_(1 << params_.bit_depth) *
+                this->rng_(kMaskMax * kMaskMax + 1);
+      mask[i] = this->rng_(kMaskMax * kMaskMax + 1);
+    }
+
+    unsigned int ref_sse, tst_sse;
+    const unsigned int ref_res = params_.ref_func(
+        CONVERT_TO_BYTEPTR(pre), pre_stride, wsrc, mask, &ref_sse);
+    unsigned int tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(CONVERT_TO_BYTEPTR(pre),
+                                                        pre_stride, wsrc, mask,
+                                                        &tst_sse));
+
+    ASSERT_EQ(ref_res, tst_res);
+    ASSERT_EQ(ref_sse, tst_sse);
+  }
+}
+
+TEST_P(ObmcVarianceHBDTest, ExtremeValues) {
+  DECLARE_ALIGNED(32, uint16_t, pre[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, wsrc[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, int32_t, mask[MAX_SB_SQUARE]);
+
+  for (int iter = 0; iter < MAX_SB_SIZE && !HasFatalFailure(); ++iter) {
+    const int pre_stride = iter;
+
+    for (int i = 0; i < MAX_SB_SQUARE; ++i) {
+      pre[i] = (1 << params_.bit_depth) - 1;
+      wsrc[i] = ((1 << params_.bit_depth) - 1) * kMaskMax * kMaskMax;
+      mask[i] = kMaskMax * kMaskMax;
+    }
+
+    unsigned int ref_sse, tst_sse;
+    const unsigned int ref_res = params_.ref_func(
+        CONVERT_TO_BYTEPTR(pre), pre_stride, wsrc, mask, &ref_sse);
+    unsigned int tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(CONVERT_TO_BYTEPTR(pre),
+                                                        pre_stride, wsrc, mask,
+                                                        &tst_sse));
+
+    ASSERT_EQ(ref_res, tst_res);
+    ASSERT_EQ(ref_sse, tst_sse);
+  }
+}
+
+#if HAVE_SSE4_1
+ObmcVarianceHBDTest::ParamType sse4_functions_hbd[] = {
+  TestFuncs(aom_highbd_obmc_variance128x128_c,
+            aom_highbd_obmc_variance128x128_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance128x64_c,
+            aom_highbd_obmc_variance128x64_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance64x128_c,
+            aom_highbd_obmc_variance64x128_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance64x64_c,
+            aom_highbd_obmc_variance64x64_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance64x32_c,
+            aom_highbd_obmc_variance64x32_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance32x64_c,
+            aom_highbd_obmc_variance32x64_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance32x32_c,
+            aom_highbd_obmc_variance32x32_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance32x16_c,
+            aom_highbd_obmc_variance32x16_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance16x32_c,
+            aom_highbd_obmc_variance16x32_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance16x16_c,
+            aom_highbd_obmc_variance16x16_sse4_1, 8),
+  TestFuncs(aom_highbd_obmc_variance16x8_c, aom_highbd_obmc_variance16x8_sse4_1,
+            8),
+  TestFuncs(aom_highbd_obmc_variance8x16_c, aom_highbd_obmc_variance8x16_sse4_1,
+            8),
+  TestFuncs(aom_highbd_obmc_variance8x8_c, aom_highbd_obmc_variance8x8_sse4_1,
+            8),
+  TestFuncs(aom_highbd_obmc_variance8x4_c, aom_highbd_obmc_variance8x4_sse4_1,
+            8),
+  TestFuncs(aom_highbd_obmc_variance4x8_c, aom_highbd_obmc_variance4x8_sse4_1,
+            8),
+  TestFuncs(aom_highbd_obmc_variance4x4_c, aom_highbd_obmc_variance4x4_sse4_1,
+            8),
+  TestFuncs(aom_highbd_10_obmc_variance128x128_c,
+            aom_highbd_10_obmc_variance128x128_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance128x64_c,
+            aom_highbd_10_obmc_variance128x64_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance64x128_c,
+            aom_highbd_10_obmc_variance64x128_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance64x64_c,
+            aom_highbd_10_obmc_variance64x64_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance64x32_c,
+            aom_highbd_10_obmc_variance64x32_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance32x64_c,
+            aom_highbd_10_obmc_variance32x64_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance32x32_c,
+            aom_highbd_10_obmc_variance32x32_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance32x16_c,
+            aom_highbd_10_obmc_variance32x16_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance16x32_c,
+            aom_highbd_10_obmc_variance16x32_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance16x16_c,
+            aom_highbd_10_obmc_variance16x16_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance16x8_c,
+            aom_highbd_10_obmc_variance16x8_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance8x16_c,
+            aom_highbd_10_obmc_variance8x16_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance8x8_c,
+            aom_highbd_10_obmc_variance8x8_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance8x4_c,
+            aom_highbd_10_obmc_variance8x4_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance4x8_c,
+            aom_highbd_10_obmc_variance4x8_sse4_1, 10),
+  TestFuncs(aom_highbd_10_obmc_variance4x4_c,
+            aom_highbd_10_obmc_variance4x4_sse4_1, 10),
+  TestFuncs(aom_highbd_12_obmc_variance128x128_c,
+            aom_highbd_12_obmc_variance128x128_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance128x64_c,
+            aom_highbd_12_obmc_variance128x64_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance64x128_c,
+            aom_highbd_12_obmc_variance64x128_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance64x64_c,
+            aom_highbd_12_obmc_variance64x64_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance64x32_c,
+            aom_highbd_12_obmc_variance64x32_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance32x64_c,
+            aom_highbd_12_obmc_variance32x64_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance32x32_c,
+            aom_highbd_12_obmc_variance32x32_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance32x16_c,
+            aom_highbd_12_obmc_variance32x16_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance16x32_c,
+            aom_highbd_12_obmc_variance16x32_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance16x16_c,
+            aom_highbd_12_obmc_variance16x16_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance16x8_c,
+            aom_highbd_12_obmc_variance16x8_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance8x16_c,
+            aom_highbd_12_obmc_variance8x16_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance8x8_c,
+            aom_highbd_12_obmc_variance8x8_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance8x4_c,
+            aom_highbd_12_obmc_variance8x4_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance4x8_c,
+            aom_highbd_12_obmc_variance4x8_sse4_1, 12),
+  TestFuncs(aom_highbd_12_obmc_variance4x4_c,
+            aom_highbd_12_obmc_variance4x4_sse4_1, 12)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, ObmcVarianceHBDTest,
+                        ::testing::ValuesIn(sse4_functions_hbd));
+#endif  // HAVE_SSE4_1
+}  // namespace
diff --git a/third_party/aom/test/onyxc_int_test.cc b/third_party/aom/test/onyxc_int_test.cc
new file mode 100644
index 0000000000..3889595187
--- /dev/null
+++ b/third_party/aom/test/onyxc_int_test.cc
@@ -0,0 +1,22 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "av1/common/onyxc_int.h"
+
+TEST(OnyxcInt, TestGetTxSize) {
+  for (int t = TX_4X4; t < TX_SIZES_ALL; t++) {
+    TX_SIZE t2 = get_tx_size(tx_size_wide[t], tx_size_high[t]);
+    GTEST_ASSERT_EQ(tx_size_wide[t], tx_size_wide[t2]);
+    GTEST_ASSERT_EQ(tx_size_high[t], tx_size_high[t2]);
+  }
+}
diff --git a/third_party/aom/test/pickrst_test.cc b/third_party/aom/test/pickrst_test.cc
new file mode 100644
index 0000000000..040e8e8b76
--- /dev/null
+++ b/third_party/aom/test/pickrst_test.cc
@@ -0,0 +1,187 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/function_equivalence_test.h"
+#include "test/register_state_check.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_integer.h"
+#include "av1/encoder/pickrst.h"
+using libaom_test::FunctionEquivalenceTest;
+
+#define MAX_DATA_BLOCK 384
+
+namespace {
+static const int kIterations = 100;
+
+typedef int64_t (*lowbd_pixel_proj_error_func)(
+    const uint8_t *src8, int width, int height, int src_stride,
+    const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride,
+    int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params);
+
+typedef libaom_test::FuncParam<lowbd_pixel_proj_error_func> TestFuncs;
+
+////////////////////////////////////////////////////////////////////////////////
+// 8 bit
+////////////////////////////////////////////////////////////////////////////////
+
+typedef ::testing::tuple<const lowbd_pixel_proj_error_func>
+    PixelProjErrorTestParam;
+
+class PixelProjErrorTest
+    : public ::testing::TestWithParam<PixelProjErrorTestParam> {
+ public:
+  virtual void SetUp() {
+    target_func_ = GET_PARAM(0);
+    src_ = (uint8_t *)(aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK *
+                                  sizeof(uint8_t)));
+    dgd_ = (uint8_t *)(aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK *
+                                  sizeof(uint8_t)));
+    flt0_ = (int32_t *)(aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK *
+                                   sizeof(int32_t)));
+    flt1_ = (int32_t *)(aom_malloc(MAX_DATA_BLOCK * MAX_DATA_BLOCK *
+                                   sizeof(int32_t)));
+  }
+  virtual void TearDown() {
+    aom_free(src_);
+    aom_free(dgd_);
+    aom_free(flt0_);
+    aom_free(flt1_);
+  }
+  void runPixelProjErrorTest(int32_t run_times);
+  void runPixelProjErrorTest_ExtremeValues();
+
+ private:
+  lowbd_pixel_proj_error_func target_func_;
+  ACMRandom rng_;
+  uint8_t *src_;
+  uint8_t *dgd_;
+  int32_t *flt0_;
+  int32_t *flt1_;
+};
+
+void PixelProjErrorTest::runPixelProjErrorTest(int32_t run_times) {
+  int h_end = run_times != 1 ? 128 : (rng_.Rand16() % MAX_DATA_BLOCK) + 1;
+  int v_end = run_times != 1 ? 128 : (rng_.Rand16() % MAX_DATA_BLOCK) + 1;
+  const int dgd_stride = MAX_DATA_BLOCK;
+  const int src_stride = MAX_DATA_BLOCK;
+  const int flt0_stride = MAX_DATA_BLOCK;
+  const int flt1_stride = MAX_DATA_BLOCK;
+  sgr_params_type params;
+  int xq[2];
+  const int iters = run_times == 1 ? kIterations : 4;
+  for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) {
+    int64_t err_ref = 0, err_test = 1;
+    for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) {
+      dgd_[i] = rng_.Rand8();
+      src_[i] = rng_.Rand8();
+      flt0_[i] = rng_.Rand15Signed();
+      flt1_[i] = rng_.Rand15Signed();
+    }
+    xq[0] = rng_.Rand8() % (1 << SGRPROJ_PRJ_BITS);
+    xq[1] = rng_.Rand8() % (1 << SGRPROJ_PRJ_BITS);
+    params.r[0] = run_times == 1 ? (rng_.Rand8() % MAX_RADIUS) : (iter % 2);
+    params.r[1] = run_times == 1 ? (rng_.Rand8() % MAX_RADIUS) : (iter / 2);
+    params.s[0] = run_times == 1 ? (rng_.Rand8() % MAX_RADIUS) : (iter % 2);
+    params.s[1] = run_times == 1 ? (rng_.Rand8() % MAX_RADIUS) : (iter / 2);
+    uint8_t *dgd = dgd_;
+    uint8_t *src = src_;
+
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      err_ref = av1_lowbd_pixel_proj_error_c(src, h_end, v_end, src_stride, dgd,
+                                             dgd_stride, flt0_, flt0_stride,
+                                             flt1_, flt1_stride, xq, &params);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      err_test =
+          target_func_(src, h_end, v_end, src_stride, dgd, dgd_stride, flt0_,
+                       flt0_stride, flt1_, flt1_stride, xq, &params);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    if (run_times > 10) {
+      printf("r0 %d r1 %d %3dx%-3d:%7.2f/%7.2fns (%3.2f)\n", params.r[0],
+             params.r[1], h_end, v_end, time1, time2, time1 / time2);
+    }
+    ASSERT_EQ(err_ref, err_test);
+  }
+}
+
+void PixelProjErrorTest::runPixelProjErrorTest_ExtremeValues() {
+  const int h_start = 0;
+  int h_end = 192;
+  const int v_start = 0;
+  int v_end = 192;
+  const int dgd_stride = MAX_DATA_BLOCK;
+  const int src_stride = MAX_DATA_BLOCK;
+  const int flt0_stride = MAX_DATA_BLOCK;
+  const int flt1_stride = MAX_DATA_BLOCK;
+  sgr_params_type params;
+  int xq[2];
+  const int iters = kIterations;
+  for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) {
+    int64_t err_ref = 0, err_test = 1;
+    for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) {
+      dgd_[i] = 0;
+      src_[i] = 255;
+      flt0_[i] = rng_.Rand15Signed();
+      flt1_[i] = rng_.Rand15Signed();
+    }
+    xq[0] = rng_.Rand8() % (1 << SGRPROJ_PRJ_BITS);
+    xq[1] = rng_.Rand8() % (1 << SGRPROJ_PRJ_BITS);
+    params.r[0] = rng_.Rand8() % MAX_RADIUS;
+    params.r[1] = rng_.Rand8() % MAX_RADIUS;
+    params.s[0] = rng_.Rand8() % MAX_RADIUS;
+    params.s[1] = rng_.Rand8() % MAX_RADIUS;
+    uint8_t *dgd = dgd_;
+    uint8_t *src = src_;
+
+    err_ref = av1_lowbd_pixel_proj_error_c(
+        src, h_end - h_start, v_end - v_start, src_stride, dgd, dgd_stride,
+        flt0_, flt0_stride, flt1_, flt1_stride, xq, &params);
+
+    err_test = target_func_(src, h_end - h_start, v_end - v_start, src_stride,
+                            dgd, dgd_stride, flt0_, flt0_stride, flt1_,
+                            flt1_stride, xq, &params);
+
+    ASSERT_EQ(err_ref, err_test);
+  }
+}
+
+TEST_P(PixelProjErrorTest, RandomValues) { runPixelProjErrorTest(1); }
+
+TEST_P(PixelProjErrorTest, ExtremeValues) {
+  runPixelProjErrorTest_ExtremeValues();
+}
+
+TEST_P(PixelProjErrorTest, DISABLED_Speed) { runPixelProjErrorTest(200000); }
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, PixelProjErrorTest,
+                        ::testing::Values(av1_lowbd_pixel_proj_error_sse4_1));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+
+INSTANTIATE_TEST_CASE_P(AVX2, PixelProjErrorTest,
+                        ::testing::Values(av1_lowbd_pixel_proj_error_avx2));
+#endif  // HAVE_AVX2
+
+}  // namespace
diff --git a/third_party/aom/test/qm_test.cc b/third_party/aom/test/qm_test.cc
new file mode 100644
index 0000000000..c87506b41a
--- /dev/null
+++ b/third_party/aom/test/qm_test.cc
@@ -0,0 +1,81 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include "config/aom_config.h"
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+class QMTest
+    : public ::libaom_test::CodecTestWith2Params<libaom_test::TestMode, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  QMTest() : EncoderTest(GET_PARAM(0)) {}
+  virtual ~QMTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+    set_cpu_used_ = GET_PARAM(2);
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
+      encoder->Control(AV1E_SET_ENABLE_QM, 1);
+      encoder->Control(AV1E_SET_QM_MIN, qm_min_);
+      encoder->Control(AV1E_SET_QM_MAX, qm_max_);
+
+      encoder->Control(AOME_SET_MAX_INTRA_BITRATE_PCT, 100);
+    }
+  }
+
+  void DoTest(int qm_min, int qm_max) {
+    qm_min_ = qm_min;
+    qm_max_ = qm_max;
+    cfg_.kf_max_dist = 12;
+    cfg_.rc_min_quantizer = 8;
+    cfg_.rc_max_quantizer = 56;
+    cfg_.rc_end_usage = AOM_CBR;
+    cfg_.g_lag_in_frames = 6;
+    cfg_.rc_buf_initial_sz = 500;
+    cfg_.rc_buf_optimal_sz = 500;
+    cfg_.rc_buf_sz = 1000;
+    cfg_.rc_target_bitrate = 300;
+    ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352,
+                                         288, 30, 1, 0, 15);
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  }
+
+  int set_cpu_used_;
+  int qm_min_;
+  int qm_max_;
+};
+
+// encodes and decodes without a mismatch.
+TEST_P(QMTest, TestNoMisMatchQM1) { DoTest(5, 9); }
+
+// encodes and decodes without a mismatch.
+TEST_P(QMTest, TestNoMisMatchQM2) { DoTest(0, 8); }
+
+// encodes and decodes without a mismatch.
+TEST_P(QMTest, TestNoMisMatchQM3) { DoTest(9, 15); }
+
+AV1_INSTANTIATE_TEST_CASE(QMTest,
+                          ::testing::Values(::libaom_test::kRealTime,
+                                            ::libaom_test::kOnePassGood),
+                          ::testing::Range(5, 9));
+}  // namespace
diff --git a/third_party/aom/test/quantize_func_test.cc b/third_party/aom/test/quantize_func_test.cc
new file mode 100644
index 0000000000..554d0c7214
--- /dev/null
+++ b/third_party/aom/test/quantize_func_test.cc
@@ -0,0 +1,425 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+#include "config/av1_rtcd.h"
+
+#include "aom/aom_codec.h"
+#include "aom_ports/aom_timer.h"
+#include "av1/encoder/encoder.h"
+#include "av1/common/scan.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+namespace {
+using libaom_test::ACMRandom;
+
+#define QUAN_PARAM_LIST                                                       \
+  const tran_low_t *coeff_ptr, intptr_t n_coeffs, const int16_t *zbin_ptr,    \
+      const int16_t *round_ptr, const int16_t *quant_ptr,                     \
+      const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr,                 \
+      tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, \
+      const int16_t *scan, const int16_t *iscan
+
+typedef void (*QuantizeFunc)(QUAN_PARAM_LIST);
+typedef void (*QuantizeFuncHbd)(QUAN_PARAM_LIST, int log_scale);
+
+#define HBD_QUAN_FUNC                                                      \
+  fn(coeff_ptr, n_coeffs, zbin_ptr, round_ptr, quant_ptr, quant_shift_ptr, \
+     qcoeff_ptr, dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan, log_scale)
+
+#define LBD_QUAN_FUNC                                                      \
+  fn(coeff_ptr, n_coeffs, zbin_ptr, round_ptr, quant_ptr, quant_shift_ptr, \
+     qcoeff_ptr, dqcoeff_ptr, dequant_ptr, eob_ptr, scan, iscan)
+
+template <QuantizeFuncHbd fn>
+void highbd_quan16x16_wrapper(QUAN_PARAM_LIST) {
+  const int log_scale = 0;
+  HBD_QUAN_FUNC;
+}
+
+template <QuantizeFuncHbd fn>
+void highbd_quan32x32_wrapper(QUAN_PARAM_LIST) {
+  const int log_scale = 1;
+  HBD_QUAN_FUNC;
+}
+
+template <QuantizeFuncHbd fn>
+void highbd_quan64x64_wrapper(QUAN_PARAM_LIST) {
+  const int log_scale = 2;
+  HBD_QUAN_FUNC;
+}
+
+typedef enum { TYPE_B, TYPE_DC, TYPE_FP } QuantType;
+
+using ::testing::tuple;
+typedef tuple<QuantizeFunc, QuantizeFunc, TX_SIZE, QuantType, aom_bit_depth_t>
+    QuantizeParam;
+
+typedef struct {
+  QUANTS quant;
+  Dequants dequant;
+} QuanTable;
+
+const int kTestNum = 1000;
+
+class QuantizeTest : public ::testing::TestWithParam<QuantizeParam> {
+ protected:
+  QuantizeTest()
+      : quant_ref_(GET_PARAM(0)), quant_(GET_PARAM(1)), tx_size_(GET_PARAM(2)),
+        type_(GET_PARAM(3)), bd_(GET_PARAM(4)) {}
+
+  virtual ~QuantizeTest() {}
+
+  virtual void SetUp() {
+    qtab_ = reinterpret_cast<QuanTable *>(aom_memalign(32, sizeof(*qtab_)));
+    const int n_coeffs = coeff_num();
+    coeff_ = reinterpret_cast<tran_low_t *>(
+        aom_memalign(32, 6 * n_coeffs * sizeof(tran_low_t)));
+    InitQuantizer();
+  }
+
+  virtual void TearDown() {
+    aom_free(qtab_);
+    qtab_ = NULL;
+    aom_free(coeff_);
+    coeff_ = NULL;
+    libaom_test::ClearSystemState();
+  }
+
+  void InitQuantizer() {
+    av1_build_quantizer(bd_, 0, 0, 0, 0, 0, &qtab_->quant, &qtab_->dequant);
+  }
+
+  void QuantizeRun(bool is_loop, int q = 0, int test_num = 1) {
+    tran_low_t *coeff_ptr = coeff_;
+    const intptr_t n_coeffs = coeff_num();
+
+    tran_low_t *qcoeff_ref = coeff_ptr + n_coeffs;
+    tran_low_t *dqcoeff_ref = qcoeff_ref + n_coeffs;
+
+    tran_low_t *qcoeff = dqcoeff_ref + n_coeffs;
+    tran_low_t *dqcoeff = qcoeff + n_coeffs;
+    uint16_t *eob = (uint16_t *)(dqcoeff + n_coeffs);
+
+    // Testing uses 2-D DCT scan order table
+    const SCAN_ORDER *const sc = get_default_scan(tx_size_, DCT_DCT);
+
+    // Testing uses luminance quantization table
+    const int16_t *zbin = qtab_->quant.y_zbin[q];
+
+    const int16_t *round = 0;
+    const int16_t *quant = 0;
+    if (type_ == TYPE_B) {
+      round = qtab_->quant.y_round[q];
+      quant = qtab_->quant.y_quant[q];
+    } else if (type_ == TYPE_FP) {
+      round = qtab_->quant.y_round_fp[q];
+      quant = qtab_->quant.y_quant_fp[q];
+    }
+
+    const int16_t *quant_shift = qtab_->quant.y_quant_shift[q];
+    const int16_t *dequant = qtab_->dequant.y_dequant_QTX[q];
+
+    for (int i = 0; i < test_num; ++i) {
+      if (is_loop) FillCoeffRandom();
+
+      memset(qcoeff_ref, 0, 5 * n_coeffs * sizeof(*qcoeff_ref));
+
+      quant_ref_(coeff_ptr, n_coeffs, zbin, round, quant, quant_shift,
+                 qcoeff_ref, dqcoeff_ref, dequant, &eob[0], sc->scan,
+                 sc->iscan);
+
+      ASM_REGISTER_STATE_CHECK(quant_(coeff_ptr, n_coeffs, zbin, round, quant,
+                                      quant_shift, qcoeff, dqcoeff, dequant,
+                                      &eob[1], sc->scan, sc->iscan));
+
+      for (int j = 0; j < n_coeffs; ++j) {
+        ASSERT_EQ(qcoeff_ref[j], qcoeff[j])
+            << "Q mismatch on test: " << i << " at position: " << j
+            << " Q: " << q << " coeff: " << coeff_ptr[j];
+      }
+
+      for (int j = 0; j < n_coeffs; ++j) {
+        ASSERT_EQ(dqcoeff_ref[j], dqcoeff[j])
+            << "Dq mismatch on test: " << i << " at position: " << j
+            << " Q: " << q << " coeff: " << coeff_ptr[j];
+      }
+
+      ASSERT_EQ(eob[0], eob[1])
+          << "eobs mismatch on test: " << i << " Q: " << q;
+    }
+  }
+
+  void CompareResults(const tran_low_t *buf_ref, const tran_low_t *buf,
+                      int size, const char *text, int q, int number) {
+    int i;
+    for (i = 0; i < size; ++i) {
+      ASSERT_EQ(buf_ref[i], buf[i]) << text << " mismatch on test: " << number
+                                    << " at position: " << i << " Q: " << q;
+    }
+  }
+
+  int coeff_num() const { return av1_get_max_eob(tx_size_); }
+
+  void FillCoeff(tran_low_t c) {
+    const int n_coeffs = coeff_num();
+    for (int i = 0; i < n_coeffs; ++i) {
+      coeff_[i] = c;
+    }
+  }
+
+  void FillCoeffRandom() {
+    const int n_coeffs = coeff_num();
+    FillCoeffZero();
+    int num = rnd_.Rand16() % n_coeffs;
+    for (int i = 0; i < num; ++i) {
+      coeff_[i] = GetRandomCoeff();
+    }
+  }
+
+  void FillCoeffZero() { FillCoeff(0); }
+
+  void FillCoeffConstant() {
+    tran_low_t c = GetRandomCoeff();
+    FillCoeff(c);
+  }
+
+  void FillDcOnly() {
+    FillCoeffZero();
+    coeff_[0] = GetRandomCoeff();
+  }
+
+  void FillDcLargeNegative() {
+    FillCoeffZero();
+    // Generate a qcoeff which contains 512/-512 (0x0100/0xFE00) to catch issues
+    // like BUG=883 where the constant being compared was incorrectly
+    // initialized.
+    coeff_[0] = -8191;
+  }
+
+  tran_low_t GetRandomCoeff() {
+    tran_low_t coeff;
+    if (bd_ == AOM_BITS_8) {
+      coeff =
+          clamp(static_cast<int16_t>(rnd_.Rand16()), INT16_MIN + 1, INT16_MAX);
+    } else {
+      tran_low_t min = -(1 << (7 + bd_));
+      tran_low_t max = -min - 1;
+      coeff = clamp(static_cast<tran_low_t>(rnd_.Rand31()), min, max);
+    }
+    return coeff;
+  }
+
+  ACMRandom rnd_;
+  QuanTable *qtab_;
+  tran_low_t *coeff_;
+  QuantizeFunc quant_ref_;
+  QuantizeFunc quant_;
+  TX_SIZE tx_size_;
+  QuantType type_;
+  aom_bit_depth_t bd_;
+};
+
+TEST_P(QuantizeTest, ZeroInput) {
+  FillCoeffZero();
+  QuantizeRun(false);
+}
+
+TEST_P(QuantizeTest, LargeNegativeInput) {
+  FillDcLargeNegative();
+  QuantizeRun(false, 0, 1);
+}
+
+TEST_P(QuantizeTest, DcOnlyInput) {
+  FillDcOnly();
+  QuantizeRun(false, 0, 1);
+}
+
+TEST_P(QuantizeTest, RandomInput) { QuantizeRun(true, 0, kTestNum); }
+
+TEST_P(QuantizeTest, MultipleQ) {
+  for (int q = 0; q < QINDEX_RANGE; ++q) {
+    QuantizeRun(true, q, kTestNum);
+  }
+}
+
+// Force the coeff to be half the value of the dequant.  This exposes a
+// mismatch found in av1_quantize_fp_sse2().
+TEST_P(QuantizeTest, CoeffHalfDequant) {
+  FillCoeff(16);
+  QuantizeRun(false, 25, 1);
+}
+
+TEST_P(QuantizeTest, DISABLED_Speed) {
+  tran_low_t *coeff_ptr = coeff_;
+  const intptr_t n_coeffs = coeff_num();
+
+  tran_low_t *qcoeff_ref = coeff_ptr + n_coeffs;
+  tran_low_t *dqcoeff_ref = qcoeff_ref + n_coeffs;
+
+  tran_low_t *qcoeff = dqcoeff_ref + n_coeffs;
+  tran_low_t *dqcoeff = qcoeff + n_coeffs;
+  uint16_t *eob = (uint16_t *)(dqcoeff + n_coeffs);
+
+  // Testing uses 2-D DCT scan order table
+  const SCAN_ORDER *const sc = get_default_scan(tx_size_, DCT_DCT);
+
+  // Testing uses luminance quantization table
+  const int q = 22;
+  const int16_t *zbin = qtab_->quant.y_zbin[q];
+  const int16_t *round_fp = qtab_->quant.y_round_fp[q];
+  const int16_t *quant_fp = qtab_->quant.y_quant_fp[q];
+  const int16_t *quant_shift = qtab_->quant.y_quant_shift[q];
+  const int16_t *dequant = qtab_->dequant.y_dequant_QTX[q];
+  const int kNumTests = 5000000;
+  aom_usec_timer timer;
+
+  FillCoeffRandom();
+
+  aom_usec_timer_start(&timer);
+  for (int n = 0; n < kNumTests; ++n) {
+    quant_(coeff_ptr, n_coeffs, zbin, round_fp, quant_fp, quant_shift, qcoeff,
+           dqcoeff, dequant, eob, sc->scan, sc->iscan);
+  }
+  aom_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+  printf("Elapsed time: %d us\n", elapsed_time);
+}
+
+using ::testing::make_tuple;
+
+#if HAVE_AVX2
+const QuantizeParam kQParamArrayAvx2[] = {
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_avx2, TX_16X16, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_avx2, TX_4X16, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_avx2, TX_16X4, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_avx2, TX_32X8, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_avx2, TX_8X32, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_32x32_c, &av1_quantize_fp_32x32_avx2, TX_32X32,
+             TYPE_FP, AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_32x32_c, &av1_quantize_fp_32x32_avx2, TX_16X64,
+             TYPE_FP, AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_32x32_c, &av1_quantize_fp_32x32_avx2, TX_64X16,
+             TYPE_FP, AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_64x64_c, &av1_quantize_fp_64x64_avx2, TX_64X64,
+             TYPE_FP, AOM_BITS_8),
+  make_tuple(&highbd_quan16x16_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan16x16_wrapper<av1_highbd_quantize_fp_avx2>, TX_16X16,
+             TYPE_FP, AOM_BITS_8),
+  make_tuple(&highbd_quan16x16_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan16x16_wrapper<av1_highbd_quantize_fp_avx2>, TX_16X16,
+             TYPE_FP, AOM_BITS_10),
+  make_tuple(&highbd_quan16x16_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan16x16_wrapper<av1_highbd_quantize_fp_avx2>, TX_16X16,
+             TYPE_FP, AOM_BITS_12),
+  make_tuple(&highbd_quan32x32_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan32x32_wrapper<av1_highbd_quantize_fp_avx2>, TX_32X32,
+             TYPE_FP, AOM_BITS_8),
+  make_tuple(&highbd_quan32x32_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan32x32_wrapper<av1_highbd_quantize_fp_avx2>, TX_32X32,
+             TYPE_FP, AOM_BITS_10),
+  make_tuple(&highbd_quan32x32_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan32x32_wrapper<av1_highbd_quantize_fp_avx2>, TX_32X32,
+             TYPE_FP, AOM_BITS_12),
+  make_tuple(&highbd_quan64x64_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan64x64_wrapper<av1_highbd_quantize_fp_avx2>, TX_64X64,
+             TYPE_FP, AOM_BITS_8),
+  make_tuple(&highbd_quan64x64_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan64x64_wrapper<av1_highbd_quantize_fp_avx2>, TX_64X64,
+             TYPE_FP, AOM_BITS_10),
+  make_tuple(&highbd_quan64x64_wrapper<av1_highbd_quantize_fp_c>,
+             &highbd_quan64x64_wrapper<av1_highbd_quantize_fp_avx2>, TX_64X64,
+             TYPE_FP, AOM_BITS_12),
+  make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_avx2, TX_16X16,
+             TYPE_B, AOM_BITS_8),
+  make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_avx2, TX_16X16,
+             TYPE_B, AOM_BITS_10),
+  make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_avx2, TX_16X16,
+             TYPE_B, AOM_BITS_12),
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, QuantizeTest,
+                        ::testing::ValuesIn(kQParamArrayAvx2));
+#endif  // HAVE_AVX2
+
+#if HAVE_SSE2
+const QuantizeParam kQParamArraySSE2[] = {
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_sse2, TX_16X16, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_sse2, TX_4X16, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_sse2, TX_16X4, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_sse2, TX_8X32, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_sse2, TX_32X8, TYPE_FP,
+             AOM_BITS_8),
+  make_tuple(&aom_quantize_b_c, &aom_quantize_b_sse2, TX_16X16, TYPE_B,
+             AOM_BITS_8),
+  make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_sse2, TX_16X16,
+             TYPE_B, AOM_BITS_8),
+  make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_sse2, TX_16X16,
+             TYPE_B, AOM_BITS_10),
+  make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_sse2, TX_16X16,
+             TYPE_B, AOM_BITS_12),
+  make_tuple(&aom_highbd_quantize_b_32x32_c, &aom_highbd_quantize_b_32x32_sse2,
+             TX_32X32, TYPE_B, AOM_BITS_8),
+  make_tuple(&aom_highbd_quantize_b_32x32_c, &aom_highbd_quantize_b_32x32_sse2,
+             TX_32X32, TYPE_B, AOM_BITS_10),
+  make_tuple(&aom_highbd_quantize_b_32x32_c, &aom_highbd_quantize_b_32x32_sse2,
+             TX_32X32, TYPE_B, AOM_BITS_12),
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, QuantizeTest,
+                        ::testing::ValuesIn(kQParamArraySSE2));
+#endif
+
+#if HAVE_SSSE3 && ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, QuantizeTest,
+    ::testing::Values(make_tuple(&aom_quantize_b_c, &aom_quantize_b_ssse3,
+                                 TX_16X16, TYPE_B, AOM_BITS_8)));
+
+// Like libvpx, the ssse3 and avx quantize tests do not pass.
+// https://bugs.chromium.org/p/webm/issues/detail?id=1448
+INSTANTIATE_TEST_CASE_P(
+    DISABLED_SSSE3_32x32, QuantizeTest,
+    ::testing::Values(make_tuple(&aom_quantize_b_32x32_c,
+                                 &aom_quantize_b_32x32_ssse3, TX_16X16, TYPE_B,
+                                 AOM_BITS_8)));
+
+#endif  // HAVE_SSSE3 && ARCH_X86_64
+
+#if HAVE_AVX && ARCH_X86_64
+INSTANTIATE_TEST_CASE_P(
+    AVX, QuantizeTest,
+    ::testing::Values(
+        make_tuple(&aom_quantize_b_c, &aom_quantize_b_avx, TX_16X16, TYPE_B,
+                   AOM_BITS_8),
+        // Although these tests will not pass against _c, test them against each
+        // other so there is some minor checking.
+        make_tuple(&aom_quantize_b_32x32_ssse3, &aom_quantize_b_32x32_avx,
+                   TX_32X32, TYPE_B, AOM_BITS_8)));
+
+#endif  // HAVE_AVX && ARCH_X86_64
+}  // namespace
diff --git a/third_party/aom/test/reconinter_test.cc b/third_party/aom/test/reconinter_test.cc
new file mode 100644
index 0000000000..a8536e517c
--- /dev/null
+++ b/third_party/aom/test/reconinter_test.cc
@@ -0,0 +1,258 @@
+/*
+ * Copyright (c) 2017, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <stdint.h>
+#include <stdio.h>
+#include <string.h>
+
+#include "config/aom_config.h"
+#include "config/av1_rtcd.h"
+
+#include "aom_ports/mem.h"
+#include "av1/common/scan.h"
+#include "av1/common/txb_common.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+using libaom_test::ACMRandom;
+
+typedef void (*buildcompdiffwtdmaskd_func)(uint8_t *mask,
+                                           DIFFWTD_MASK_TYPE mask_type,
+                                           const uint8_t *src0, int src0_stride,
+                                           const uint8_t *src1, int src1_stride,
+                                           int h, int w);
+
+typedef ::testing::tuple<BLOCK_SIZE, buildcompdiffwtdmaskd_func>
+    BuildCompDiffwtdMaskDParam;
+
+#if HAVE_SSE4_1
+::testing::internal::ParamGenerator<BuildCompDiffwtdMaskDParam> BuildParams(
+    buildcompdiffwtdmaskd_func filter) {
+  return ::testing::Combine(::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL),
+                            ::testing::Values(filter));
+}
+#endif
+
+class BuildCompDiffwtdMaskTest
+    : public ::testing::TestWithParam<BuildCompDiffwtdMaskDParam> {
+ public:
+  virtual ~BuildCompDiffwtdMaskTest() {}
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+  void RunTest(buildcompdiffwtdmaskd_func test_impl, const int is_speed,
+               const DIFFWTD_MASK_TYPE type);
+
+ private:
+  ACMRandom rnd_;
+};
+
+typedef void (*buildcompdiffwtdmaskd16_func)(
+    uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const CONV_BUF_TYPE *src0,
+    int src0_stride, const CONV_BUF_TYPE *src1, int src1_stride, int h, int w,
+    ConvolveParams *conv_params, int bd);
+
+typedef ::testing::tuple<int, buildcompdiffwtdmaskd16_func, BLOCK_SIZE>
+    BuildCompDiffwtdMaskD16Param;
+
+#if HAVE_SSE4_1 || HAVE_NEON
+::testing::internal::ParamGenerator<BuildCompDiffwtdMaskD16Param> BuildParams(
+    buildcompdiffwtdmaskd16_func filter) {
+  return ::testing::Combine(::testing::Range(8, 13, 2),
+                            ::testing::Values(filter),
+                            ::testing::Range(BLOCK_4X4, BLOCK_SIZES_ALL));
+}
+#endif
+class BuildCompDiffwtdMaskD16Test
+    : public ::testing::TestWithParam<BuildCompDiffwtdMaskD16Param> {
+ public:
+  ~BuildCompDiffwtdMaskD16Test() {}
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+  void SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+
+ protected:
+  void RunCheckOutput(buildcompdiffwtdmaskd16_func test_impl);
+  void RunSpeedTest(buildcompdiffwtdmaskd16_func test_impl,
+                    DIFFWTD_MASK_TYPE mask_type);
+  libaom_test::ACMRandom rnd_;
+};  // class BuildCompDiffwtdMaskD16Test
+
+void BuildCompDiffwtdMaskD16Test::RunCheckOutput(
+    buildcompdiffwtdmaskd16_func test_impl) {
+  const int block_idx = GET_PARAM(2);
+  const int bd = GET_PARAM(0);
+  const int width = block_size_wide[block_idx];
+  const int height = block_size_high[block_idx];
+  DECLARE_ALIGNED(16, uint8_t, mask_ref[2 * MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(16, uint8_t, mask_test[2 * MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint16_t, src0[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint16_t, src1[MAX_SB_SQUARE]);
+
+  ConvolveParams conv_params = get_conv_params_no_round(0, 0, NULL, 0, 1, bd);
+
+  int in_precision =
+      bd + 2 * FILTER_BITS - conv_params.round_0 - conv_params.round_1 + 2;
+
+  for (int i = 0; i < MAX_SB_SQUARE; i++) {
+    src0[i] = rnd_.Rand16() & ((1 << in_precision) - 1);
+    src1[i] = rnd_.Rand16() & ((1 << in_precision) - 1);
+  }
+
+  for (int mask_type = 0; mask_type < DIFFWTD_MASK_TYPES; mask_type++) {
+    av1_build_compound_diffwtd_mask_d16_c(
+        mask_ref, (DIFFWTD_MASK_TYPE)mask_type, src0, width, src1, width,
+        height, width, &conv_params, bd);
+
+    test_impl(mask_test, (DIFFWTD_MASK_TYPE)mask_type, src0, width, src1, width,
+              height, width, &conv_params, bd);
+
+    for (int r = 0; r < height; ++r) {
+      for (int c = 0; c < width; ++c) {
+        ASSERT_EQ(mask_ref[c + r * width], mask_test[c + r * width])
+            << "Mismatch at unit tests for BuildCompDiffwtdMaskD16Test\n"
+            << " Pixel mismatch at index "
+            << "[" << r << "," << c << "] "
+            << " @ " << width << "x" << height << " inv " << mask_type;
+      }
+    }
+  }
+}
+
+void BuildCompDiffwtdMaskD16Test::RunSpeedTest(
+    buildcompdiffwtdmaskd16_func test_impl, DIFFWTD_MASK_TYPE mask_type) {
+  const int block_idx = GET_PARAM(2);
+  const int bd = GET_PARAM(0);
+  const int width = block_size_wide[block_idx];
+  const int height = block_size_high[block_idx];
+  DECLARE_ALIGNED(16, uint8_t, mask[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint16_t, src0[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(32, uint16_t, src1[MAX_SB_SQUARE]);
+
+  ConvolveParams conv_params = get_conv_params_no_round(0, 0, NULL, 0, 1, bd);
+
+  int in_precision =
+      bd + 2 * FILTER_BITS - conv_params.round_0 - conv_params.round_1 + 2;
+
+  for (int i = 0; i < MAX_SB_SQUARE; i++) {
+    src0[i] = rnd_.Rand16() & ((1 << in_precision) - 1);
+    src1[i] = rnd_.Rand16() & ((1 << in_precision) - 1);
+  }
+
+  const int num_loops = 10000000 / (width + height);
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+
+  for (int i = 0; i < num_loops; ++i)
+    av1_build_compound_diffwtd_mask_d16_c(mask, mask_type, src0, width, src1,
+                                          width, height, width, &conv_params,
+                                          bd);
+
+  aom_usec_timer_mark(&timer);
+  const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+
+  aom_usec_timer timer1;
+  aom_usec_timer_start(&timer1);
+
+  for (int i = 0; i < num_loops; ++i)
+    test_impl(mask, mask_type, src0, width, src1, width, height, width,
+              &conv_params, bd);
+
+  aom_usec_timer_mark(&timer1);
+  const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
+  printf("av1_build_compound_diffwtd_mask_d16  %3dx%-3d: %7.2f \n", width,
+         height, elapsed_time / double(elapsed_time1));
+}
+#if HAVE_SSE4_1
+void BuildCompDiffwtdMaskTest::RunTest(buildcompdiffwtdmaskd_func test_impl,
+                                       const int is_speed,
+                                       const DIFFWTD_MASK_TYPE type) {
+  const int sb_type = GET_PARAM(0);
+  const int width = block_size_wide[sb_type];
+  const int height = block_size_high[sb_type];
+  DECLARE_ALIGNED(16, uint8_t, mask_ref[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(16, uint8_t, mask_test[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(16, uint8_t, src0[MAX_SB_SQUARE]);
+  DECLARE_ALIGNED(16, uint8_t, src1[MAX_SB_SQUARE]);
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  for (int i = 0; i < width * height; i++) {
+    src0[i] = rnd.Rand8();
+    src1[i] = rnd.Rand8();
+  }
+  const int run_times = is_speed ? (10000000 / (width + height)) : 1;
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < run_times; ++i) {
+    av1_build_compound_diffwtd_mask_c(mask_ref, type, src0, width, src1, width,
+                                      height, width);
+  }
+  const double t1 = get_time_mark(&timer);
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < run_times; ++i) {
+    test_impl(mask_test, type, src0, width, src1, width, height, width);
+  }
+  const double t2 = get_time_mark(&timer);
+  if (is_speed) {
+    printf("mask %d %3dx%-3d:%7.2f/%7.2fns", type, width, height, t1, t2);
+    printf("(%3.2f)\n", t1 / t2);
+  }
+  for (int r = 0; r < height; ++r) {
+    for (int c = 0; c < width; ++c) {
+      ASSERT_EQ(mask_ref[c + r * width], mask_test[c + r * width])
+          << "[" << r << "," << c << "] " << run_times << " @ " << width << "x"
+          << height << " inv " << type;
+    }
+  }
+}
+
+TEST_P(BuildCompDiffwtdMaskTest, match) {
+  RunTest(GET_PARAM(1), 0, DIFFWTD_38);
+  RunTest(GET_PARAM(1), 0, DIFFWTD_38_INV);
+}
+TEST_P(BuildCompDiffwtdMaskTest, DISABLED_Speed) {
+  RunTest(GET_PARAM(1), 1, DIFFWTD_38);
+  RunTest(GET_PARAM(1), 1, DIFFWTD_38_INV);
+}
+#endif
+TEST_P(BuildCompDiffwtdMaskD16Test, CheckOutput) {
+  RunCheckOutput(GET_PARAM(1));
+}
+
+TEST_P(BuildCompDiffwtdMaskD16Test, DISABLED_Speed) {
+  RunSpeedTest(GET_PARAM(1), DIFFWTD_38);
+  RunSpeedTest(GET_PARAM(1), DIFFWTD_38_INV);
+}
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, BuildCompDiffwtdMaskTest,
+                        BuildParams(av1_build_compound_diffwtd_mask_sse4_1));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, BuildCompDiffwtdMaskD16Test,
+    BuildParams(av1_build_compound_diffwtd_mask_d16_sse4_1));
+#endif
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, BuildCompDiffwtdMaskTest,
+                        BuildParams(av1_build_compound_diffwtd_mask_avx2));
+
+INSTANTIATE_TEST_CASE_P(AVX2, BuildCompDiffwtdMaskD16Test,
+                        BuildParams(av1_build_compound_diffwtd_mask_d16_avx2));
+#endif
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, BuildCompDiffwtdMaskD16Test,
+                        BuildParams(av1_build_compound_diffwtd_mask_d16_neon));
+#endif
+
+}  // namespace
diff --git a/third_party/aom/test/register_state_check.h b/third_party/aom/test/register_state_check.h
new file mode 100644
index 0000000000..d404621dd7
--- /dev/null
+++ b/third_party/aom/test/register_state_check.h
@@ -0,0 +1,148 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_REGISTER_STATE_CHECK_H_
+#define AOM_TEST_REGISTER_STATE_CHECK_H_
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#include "aom/aom_integer.h"
+
+// ASM_REGISTER_STATE_CHECK(asm_function)
+//   Minimally validates the environment pre & post function execution. This
+//   variant should be used with assembly functions which are not expected to
+//   fully restore the system state. See platform implementations of
+//   RegisterStateCheck for details.
+//
+// API_REGISTER_STATE_CHECK(api_function)
+//   Performs all the checks done by ASM_REGISTER_STATE_CHECK() and any
+//   additional checks to ensure the environment is in a consistent state pre &
+//   post function execution. This variant should be used with API functions.
+//   See platform implementations of RegisterStateCheckXXX for details.
+//
+
+#if defined(_WIN64) && ARCH_X86_64
+
+#undef NOMINMAX
+#define NOMINMAX
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#include <winnt.h>
+
+inline bool operator==(const M128A &lhs, const M128A &rhs) {
+  return (lhs.Low == rhs.Low && lhs.High == rhs.High);
+}
+
+namespace libaom_test {
+
+// Compares the state of xmm[6-15] at construction with their state at
+// destruction. These registers should be preserved by the callee on
+// Windows x64.
+class RegisterStateCheck {
+ public:
+  RegisterStateCheck() { initialized_ = StoreRegisters(&pre_context_); }
+  ~RegisterStateCheck() { Check(); }
+
+ private:
+  static bool StoreRegisters(CONTEXT *const context) {
+    const HANDLE this_thread = GetCurrentThread();
+    EXPECT_TRUE(this_thread != NULL);
+    context->ContextFlags = CONTEXT_FLOATING_POINT;
+    const bool context_saved = GetThreadContext(this_thread, context) == TRUE;
+    EXPECT_TRUE(context_saved) << "GetLastError: " << GetLastError();
+    return context_saved;
+  }
+
+  // Compares the register state. Returns true if the states match.
+  void Check() const {
+    ASSERT_TRUE(initialized_);
+    CONTEXT post_context;
+    ASSERT_TRUE(StoreRegisters(&post_context));
+
+    const M128A *xmm_pre = &pre_context_.Xmm6;
+    const M128A *xmm_post = &post_context.Xmm6;
+    for (int i = 6; i <= 15; ++i) {
+      EXPECT_EQ(*xmm_pre, *xmm_post) << "xmm" << i << " has been modified!";
+      ++xmm_pre;
+      ++xmm_post;
+    }
+  }
+
+  bool initialized_;
+  CONTEXT pre_context_;
+};
+
+#define ASM_REGISTER_STATE_CHECK(statement)    \
+  do {                                         \
+    libaom_test::RegisterStateCheck reg_check; \
+    statement;                                 \
+  } while (false)
+
+}  // namespace libaom_test
+
+#else
+
+namespace libaom_test {
+
+class RegisterStateCheck {};
+#define ASM_REGISTER_STATE_CHECK(statement) statement
+
+}  // namespace libaom_test
+
+#endif  // _WIN64 && ARCH_X86_64
+
+#if ARCH_X86 || ARCH_X86_64
+#if defined(__GNUC__)
+
+namespace libaom_test {
+
+// Checks the FPU tag word pre/post execution to ensure emms has been called.
+class RegisterStateCheckMMX {
+ public:
+  RegisterStateCheckMMX() {
+    __asm__ volatile("fstenv %0" : "=rm"(pre_fpu_env_));
+  }
+  ~RegisterStateCheckMMX() { Check(); }
+
+ private:
+  // Checks the FPU tag word pre/post execution, returning false if not cleared
+  // to 0xffff.
+  void Check() const {
+    EXPECT_EQ(0xffff, pre_fpu_env_[4])
+        << "FPU was in an inconsistent state prior to call";
+
+    uint16_t post_fpu_env[14];
+    __asm__ volatile("fstenv %0" : "=rm"(post_fpu_env));
+    EXPECT_EQ(0xffff, post_fpu_env[4])
+        << "FPU was left in an inconsistent state after call";
+  }
+
+  uint16_t pre_fpu_env_[14];
+};
+
+#define API_REGISTER_STATE_CHECK(statement)       \
+  do {                                            \
+    libaom_test::RegisterStateCheckMMX reg_check; \
+    ASM_REGISTER_STATE_CHECK(statement);          \
+  } while (false)
+
+}  // namespace libaom_test
+
+#endif  // __GNUC__
+#endif  // ARCH_X86 || ARCH_X86_64
+
+#ifndef API_REGISTER_STATE_CHECK
+#define API_REGISTER_STATE_CHECK ASM_REGISTER_STATE_CHECK
+#endif
+
+#endif  // AOM_TEST_REGISTER_STATE_CHECK_H_
diff --git a/third_party/aom/test/resize_test.cc b/third_party/aom/test/resize_test.cc
new file mode 100644
index 0000000000..b270b8362f
--- /dev/null
+++ b/third_party/aom/test/resize_test.cc
@@ -0,0 +1,642 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <climits>
+#include <vector>
+#include "aom_dsp/aom_dsp_common.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/video_source.h"
+#include "test/util.h"
+
+// Enable(1) or Disable(0) writing of the compressed bitstream.
+#define WRITE_COMPRESSED_STREAM 0
+
+namespace {
+
+#if WRITE_COMPRESSED_STREAM
+static void mem_put_le16(char *const mem, unsigned int val) {
+  mem[0] = val;
+  mem[1] = val >> 8;
+}
+
+static void mem_put_le32(char *const mem, unsigned int val) {
+  mem[0] = val;
+  mem[1] = val >> 8;
+  mem[2] = val >> 16;
+  mem[3] = val >> 24;
+}
+
+static void write_ivf_file_header(const aom_codec_enc_cfg_t *const cfg,
+                                  int frame_cnt, FILE *const outfile) {
+  char header[32];
+
+  header[0] = 'D';
+  header[1] = 'K';
+  header[2] = 'I';
+  header[3] = 'F';
+  mem_put_le16(header + 4, 0);                    /* version */
+  mem_put_le16(header + 6, 32);                   /* headersize */
+  mem_put_le32(header + 8, 0x30395056);           /* fourcc (av1) */
+  mem_put_le16(header + 12, cfg->g_w);            /* width */
+  mem_put_le16(header + 14, cfg->g_h);            /* height */
+  mem_put_le32(header + 16, cfg->g_timebase.den); /* rate */
+  mem_put_le32(header + 20, cfg->g_timebase.num); /* scale */
+  mem_put_le32(header + 24, frame_cnt);           /* length */
+  mem_put_le32(header + 28, 0);                   /* unused */
+
+  (void)fwrite(header, 1, 32, outfile);
+}
+
+static void write_ivf_frame_size(FILE *const outfile, const size_t size) {
+  char header[4];
+  mem_put_le32(header, static_cast<unsigned int>(size));
+  (void)fwrite(header, 1, 4, outfile);
+}
+
+static void write_ivf_frame_header(const aom_codec_cx_pkt_t *const pkt,
+                                   FILE *const outfile) {
+  char header[12];
+  aom_codec_pts_t pts;
+
+  if (pkt->kind != AOM_CODEC_CX_FRAME_PKT) return;
+
+  pts = pkt->data.frame.pts;
+  mem_put_le32(header, static_cast<unsigned int>(pkt->data.frame.sz));
+  mem_put_le32(header + 4, pts & 0xFFFFFFFF);
+  mem_put_le32(header + 8, pts >> 32);
+
+  (void)fwrite(header, 1, 12, outfile);
+}
+#endif  // WRITE_COMPRESSED_STREAM
+
+const unsigned int kInitialWidth = 320;
+const unsigned int kInitialHeight = 240;
+
+struct FrameInfo {
+  FrameInfo(aom_codec_pts_t _pts, unsigned int _w, unsigned int _h)
+      : pts(_pts), w(_w), h(_h) {}
+
+  aom_codec_pts_t pts;
+  unsigned int w;
+  unsigned int h;
+};
+
+void ScaleForFrameNumber(unsigned int frame, unsigned int initial_w,
+                         unsigned int initial_h, unsigned int *w,
+                         unsigned int *h, int flag_codec) {
+  if (frame < 10) {
+    *w = initial_w;
+    *h = initial_h;
+    return;
+  }
+  if (frame < 20) {
+    *w = initial_w * 3 / 4;
+    *h = initial_h * 3 / 4;
+    return;
+  }
+  if (frame < 30) {
+    *w = initial_w / 2;
+    *h = initial_h / 2;
+    return;
+  }
+  if (frame < 40) {
+    *w = initial_w;
+    *h = initial_h;
+    return;
+  }
+  if (frame < 50) {
+    *w = initial_w * 3 / 4;
+    *h = initial_h * 3 / 4;
+    return;
+  }
+  if (frame < 60) {
+    *w = initial_w / 2;
+    *h = initial_h / 2;
+    return;
+  }
+  if (frame < 70) {
+    *w = initial_w;
+    *h = initial_h;
+    return;
+  }
+  if (frame < 80) {
+    *w = initial_w * 3 / 4;
+    *h = initial_h * 3 / 4;
+    return;
+  }
+  if (frame < 90) {
+    *w = initial_w / 2;
+    *h = initial_h / 2;
+    return;
+  }
+  if (frame < 100) {
+    *w = initial_w * 3 / 4;
+    *h = initial_h * 3 / 4;
+    return;
+  }
+  if (frame < 110) {
+    *w = initial_w;
+    *h = initial_h;
+    return;
+  }
+  // Go down very low
+  if (frame < 120) {
+    *w = initial_w / 4;
+    *h = initial_h / 4;
+    return;
+  }
+  if (flag_codec == 1) {
+    // Cases that only works for AV1.
+    // For AV1: Swap width and height of original.
+    if (frame < 140) {
+      *w = initial_h;
+      *h = initial_w;
+      return;
+    }
+  }
+  *w = initial_w;
+  *h = initial_h;
+}
+
+class ResizingVideoSource : public ::libaom_test::DummyVideoSource {
+ public:
+  ResizingVideoSource() {
+    SetSize(kInitialWidth, kInitialHeight);
+    limit_ = 150;
+  }
+  int flag_codec_;
+  virtual ~ResizingVideoSource() {}
+
+ protected:
+  virtual void Next() {
+    ++frame_;
+    unsigned int width;
+    unsigned int height;
+    ScaleForFrameNumber(frame_, kInitialWidth, kInitialHeight, &width, &height,
+                        flag_codec_);
+    SetSize(width, height);
+    FillFrame();
+  }
+};
+
+class ResizeTest
+    : public ::libaom_test::CodecTestWithParam<libaom_test::TestMode>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  ResizeTest() : EncoderTest(GET_PARAM(0)) {}
+
+  virtual ~ResizeTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+  }
+
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     aom_codec_pts_t pts) {
+    frame_info_list_.push_back(FrameInfo(pts, img.d_w, img.d_h));
+  }
+
+  std::vector<FrameInfo> frame_info_list_;
+};
+
+TEST_P(ResizeTest, TestExternalResizeWorks) {
+  ResizingVideoSource video;
+  video.flag_codec_ = 0;
+  cfg_.g_lag_in_frames = 0;
+  // We use max(kInitialWidth, kInitialHeight) because during the test
+  // the width and height of the frame are swapped
+  cfg_.g_forced_max_frame_width = cfg_.g_forced_max_frame_height =
+      AOMMAX(kInitialWidth, kInitialHeight);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  // Check we decoded the same number of frames as we attempted to encode
+  ASSERT_EQ(frame_info_list_.size(), video.limit());
+
+  for (std::vector<FrameInfo>::const_iterator info = frame_info_list_.begin();
+       info != frame_info_list_.end(); ++info) {
+    const unsigned int frame = static_cast<unsigned>(info->pts);
+    unsigned int expected_w;
+    unsigned int expected_h;
+    ScaleForFrameNumber(frame, kInitialWidth, kInitialHeight, &expected_w,
+                        &expected_h, 0);
+    EXPECT_EQ(expected_w, info->w)
+        << "Frame " << frame << " had unexpected width";
+    EXPECT_EQ(expected_h, info->h)
+        << "Frame " << frame << " had unexpected height";
+  }
+}
+
+const unsigned int kStepDownFrame = 3;
+const unsigned int kStepUpFrame = 6;
+
+class ResizeInternalTestLarge : public ResizeTest {
+ protected:
+#if WRITE_COMPRESSED_STREAM
+  ResizeInternalTestLarge()
+      : ResizeTest(), frame0_psnr_(0.0), outfile_(NULL), out_frames_(0) {}
+#else
+  ResizeInternalTestLarge() : ResizeTest(), frame0_psnr_(0.0) {}
+#endif
+
+  virtual ~ResizeInternalTestLarge() {}
+
+  virtual void BeginPassHook(unsigned int /*pass*/) {
+#if WRITE_COMPRESSED_STREAM
+    outfile_ = fopen("av10-2-05-resize.ivf", "wb");
+#endif
+  }
+
+  virtual void EndPassHook() {
+#if WRITE_COMPRESSED_STREAM
+    if (outfile_) {
+      if (!fseek(outfile_, 0, SEEK_SET))
+        write_ivf_file_header(&cfg_, out_frames_, outfile_);
+      fclose(outfile_);
+      outfile_ = NULL;
+    }
+#endif
+  }
+
+  virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
+                                  libaom_test::Encoder *encoder) {
+    if (change_config_) {
+      int new_q = 60;
+      if (video->frame() == 0) {
+        struct aom_scaling_mode mode = { AOME_ONETWO, AOME_ONETWO };
+        encoder->Control(AOME_SET_SCALEMODE, &mode);
+      }
+      if (video->frame() == 1) {
+        struct aom_scaling_mode mode = { AOME_NORMAL, AOME_NORMAL };
+        encoder->Control(AOME_SET_SCALEMODE, &mode);
+        cfg_.rc_min_quantizer = cfg_.rc_max_quantizer = new_q;
+        encoder->Config(&cfg_);
+      }
+    } else {
+      if (video->frame() >= kStepDownFrame && video->frame() < kStepUpFrame) {
+        struct aom_scaling_mode mode = { AOME_FOURFIVE, AOME_THREEFIVE };
+        encoder->Control(AOME_SET_SCALEMODE, &mode);
+      }
+      if (video->frame() >= kStepUpFrame) {
+        struct aom_scaling_mode mode = { AOME_NORMAL, AOME_NORMAL };
+        encoder->Control(AOME_SET_SCALEMODE, &mode);
+      }
+    }
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (frame0_psnr_ == 0.) frame0_psnr_ = pkt->data.psnr.psnr[0];
+    EXPECT_NEAR(pkt->data.psnr.psnr[0], frame0_psnr_, 2.5);
+  }
+
+#if WRITE_COMPRESSED_STREAM
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    ++out_frames_;
+
+    // Write initial file header if first frame.
+    if (pkt->data.frame.pts == 0) write_ivf_file_header(&cfg_, 0, outfile_);
+
+    // Write frame header and data.
+    write_ivf_frame_header(pkt, outfile_);
+    (void)fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz, outfile_);
+  }
+#endif
+
+  double frame0_psnr_;
+  bool change_config_;
+#if WRITE_COMPRESSED_STREAM
+  FILE *outfile_;
+  unsigned int out_frames_;
+#endif
+};
+
+TEST_P(ResizeInternalTestLarge, TestInternalResizeWorks) {
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 10);
+  init_flags_ = AOM_CODEC_USE_PSNR;
+  change_config_ = false;
+
+  // q picked such that initial keyframe on this clip is ~30dB PSNR
+  cfg_.rc_min_quantizer = cfg_.rc_max_quantizer = 48;
+
+  // If the number of frames being encoded is smaller than g_lag_in_frames
+  // the encoded frame is unavailable using the current API. Comparing
+  // frames to detect mismatch would then not be possible. Set
+  // g_lag_in_frames = 0 to get around this.
+  cfg_.g_lag_in_frames = 0;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  for (std::vector<FrameInfo>::const_iterator info = frame_info_list_.begin();
+       info != frame_info_list_.end(); ++info) {
+  }
+  for (std::vector<FrameInfo>::const_iterator info = frame_info_list_.begin();
+       info != frame_info_list_.end(); ++info) {
+    const aom_codec_pts_t pts = info->pts;
+    if (pts >= kStepDownFrame && pts < kStepUpFrame) {
+      ASSERT_EQ(282U, info->w) << "Frame " << pts << " had unexpected width";
+      ASSERT_EQ(173U, info->h) << "Frame " << pts << " had unexpected height";
+    } else {
+      EXPECT_EQ(352U, info->w) << "Frame " << pts << " had unexpected width";
+      EXPECT_EQ(288U, info->h) << "Frame " << pts << " had unexpected height";
+    }
+  }
+}
+
+TEST_P(ResizeInternalTestLarge, TestInternalResizeChangeConfig) {
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 10);
+  cfg_.g_w = 352;
+  cfg_.g_h = 288;
+  change_config_ = true;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+}
+
+class ResizeRealtimeTest
+    : public ::libaom_test::CodecTestWith2Params<libaom_test::TestMode, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  ResizeRealtimeTest() : EncoderTest(GET_PARAM(0)) {}
+  virtual ~ResizeRealtimeTest() {}
+
+  virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
+                                  libaom_test::Encoder *encoder) {
+    if (video->frame() == 0) {
+      encoder->Control(AV1E_SET_AQ_MODE, 3);
+      encoder->Control(AOME_SET_CPUUSED, set_cpu_used_);
+    }
+
+    if (change_bitrate_ && video->frame() == 120) {
+      change_bitrate_ = false;
+      cfg_.rc_target_bitrate = 500;
+      encoder->Config(&cfg_);
+    }
+  }
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+    set_cpu_used_ = GET_PARAM(2);
+  }
+
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     aom_codec_pts_t pts) {
+    frame_info_list_.push_back(FrameInfo(pts, img.d_w, img.d_h));
+  }
+
+  virtual void MismatchHook(const aom_image_t *img1, const aom_image_t *img2) {
+    double mismatch_psnr = compute_psnr(img1, img2);
+    mismatch_psnr_ += mismatch_psnr;
+    ++mismatch_nframes_;
+  }
+
+  unsigned int GetMismatchFrames() { return mismatch_nframes_; }
+
+  void DefaultConfig() {
+    cfg_.rc_buf_initial_sz = 500;
+    cfg_.rc_buf_optimal_sz = 600;
+    cfg_.rc_buf_sz = 1000;
+    cfg_.rc_min_quantizer = 2;
+    cfg_.rc_max_quantizer = 56;
+    cfg_.rc_undershoot_pct = 50;
+    cfg_.rc_overshoot_pct = 50;
+    cfg_.rc_end_usage = AOM_CBR;
+    cfg_.kf_mode = AOM_KF_AUTO;
+    cfg_.g_lag_in_frames = 0;
+    cfg_.kf_min_dist = cfg_.kf_max_dist = 3000;
+    // Enable dropped frames.
+    cfg_.rc_dropframe_thresh = 1;
+    // Disable error_resilience mode.
+    cfg_.g_error_resilient = 0;
+    // Run at low bitrate.
+    cfg_.rc_target_bitrate = 200;
+    // We use max(kInitialWidth, kInitialHeight) because during the test
+    // the width and height of the frame are swapped
+    cfg_.g_forced_max_frame_width = cfg_.g_forced_max_frame_height =
+        AOMMAX(kInitialWidth, kInitialHeight);
+  }
+
+  std::vector<FrameInfo> frame_info_list_;
+  int set_cpu_used_;
+  bool change_bitrate_;
+  double mismatch_psnr_;
+  int mismatch_nframes_;
+};
+
+TEST_P(ResizeRealtimeTest, TestExternalResizeWorks) {
+  ResizingVideoSource video;
+  video.flag_codec_ = 1;
+  DefaultConfig();
+  change_bitrate_ = false;
+  mismatch_psnr_ = 0.0;
+  mismatch_nframes_ = 0;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  // Check we decoded the same number of frames as we attempted to encode
+  ASSERT_EQ(frame_info_list_.size(), video.limit());
+
+  for (std::vector<FrameInfo>::const_iterator info = frame_info_list_.begin();
+       info != frame_info_list_.end(); ++info) {
+    const unsigned int frame = static_cast<unsigned>(info->pts);
+    unsigned int expected_w;
+    unsigned int expected_h;
+    ScaleForFrameNumber(frame, kInitialWidth, kInitialHeight, &expected_w,
+                        &expected_h, 1);
+    EXPECT_EQ(expected_w, info->w)
+        << "Frame " << frame << " had unexpected width";
+    EXPECT_EQ(expected_h, info->h)
+        << "Frame " << frame << " had unexpected height";
+    EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+  }
+}
+
+// Verify the dynamic resizer behavior for real time, 1 pass CBR mode.
+// Run at low bitrate, with resize_allowed = 1, and verify that we get
+// one resize down event.
+TEST_P(ResizeRealtimeTest, DISABLED_TestInternalResizeDown) {
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 299);
+  DefaultConfig();
+  cfg_.g_w = 352;
+  cfg_.g_h = 288;
+  change_bitrate_ = false;
+  mismatch_psnr_ = 0.0;
+  mismatch_nframes_ = 0;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  unsigned int last_w = cfg_.g_w;
+  unsigned int last_h = cfg_.g_h;
+  int resize_count = 0;
+  for (std::vector<FrameInfo>::const_iterator info = frame_info_list_.begin();
+       info != frame_info_list_.end(); ++info) {
+    if (info->w != last_w || info->h != last_h) {
+      // Verify that resize down occurs.
+      ASSERT_LT(info->w, last_w);
+      ASSERT_LT(info->h, last_h);
+      last_w = info->w;
+      last_h = info->h;
+      resize_count++;
+    }
+  }
+
+#if CONFIG_AV1_DECODER
+  // Verify that we get 1 resize down event in this test.
+  ASSERT_EQ(1, resize_count) << "Resizing should occur.";
+  EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+#else
+  printf("Warning: AV1 decoder unavailable, unable to check resize count!\n");
+#endif
+}
+
+// Verify the dynamic resizer behavior for real time, 1 pass CBR mode.
+// Start at low target bitrate, raise the bitrate in the middle of the clip,
+// scaling-up should occur after bitrate changed.
+TEST_P(ResizeRealtimeTest, DISABLED_TestInternalResizeDownUpChangeBitRate) {
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 359);
+  DefaultConfig();
+  cfg_.g_w = 352;
+  cfg_.g_h = 288;
+  change_bitrate_ = true;
+  mismatch_psnr_ = 0.0;
+  mismatch_nframes_ = 0;
+  // Disable dropped frames.
+  cfg_.rc_dropframe_thresh = 0;
+  // Starting bitrate low.
+  cfg_.rc_target_bitrate = 80;
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+  unsigned int last_w = cfg_.g_w;
+  unsigned int last_h = cfg_.g_h;
+  int resize_count = 0;
+  for (std::vector<FrameInfo>::const_iterator info = frame_info_list_.begin();
+       info != frame_info_list_.end(); ++info) {
+    if (info->w != last_w || info->h != last_h) {
+      resize_count++;
+      if (resize_count == 1) {
+        // Verify that resize down occurs.
+        ASSERT_LT(info->w, last_w);
+        ASSERT_LT(info->h, last_h);
+      } else if (resize_count == 2) {
+        // Verify that resize up occurs.
+        ASSERT_GT(info->w, last_w);
+        ASSERT_GT(info->h, last_h);
+      }
+      last_w = info->w;
+      last_h = info->h;
+    }
+  }
+
+#if CONFIG_AV1_DECODER
+  // Verify that we get 2 resize events in this test.
+  ASSERT_EQ(resize_count, 2) << "Resizing should occur twice.";
+  EXPECT_EQ(static_cast<unsigned int>(0), GetMismatchFrames());
+#else
+  printf("Warning: AV1 decoder unavailable, unable to check resize count!\n");
+#endif
+}
+
+class ResizeCspTest : public ResizeTest {
+ protected:
+#if WRITE_COMPRESSED_STREAM
+  ResizeCspTest()
+      : ResizeTest(), frame0_psnr_(0.0), outfile_(NULL), out_frames_(0) {}
+#else
+  ResizeCspTest() : ResizeTest(), frame0_psnr_(0.0) {}
+#endif
+
+  virtual ~ResizeCspTest() {}
+
+  virtual void BeginPassHook(unsigned int /*pass*/) {
+#if WRITE_COMPRESSED_STREAM
+    outfile_ = fopen("av11-2-05-cspchape.ivf", "wb");
+#endif
+  }
+
+  virtual void EndPassHook() {
+#if WRITE_COMPRESSED_STREAM
+    if (outfile_) {
+      if (!fseek(outfile_, 0, SEEK_SET))
+        write_ivf_file_header(&cfg_, out_frames_, outfile_);
+      fclose(outfile_);
+      outfile_ = NULL;
+    }
+#endif
+  }
+
+  virtual void PSNRPktHook(const aom_codec_cx_pkt_t *pkt) {
+    if (frame0_psnr_ == 0.) frame0_psnr_ = pkt->data.psnr.psnr[0];
+    EXPECT_NEAR(pkt->data.psnr.psnr[0], frame0_psnr_, 2.0);
+  }
+
+#if WRITE_COMPRESSED_STREAM
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    ++out_frames_;
+
+    // Write initial file header if first frame.
+    if (pkt->data.frame.pts == 0) write_ivf_file_header(&cfg_, 0, outfile_);
+
+    // Write frame header and data.
+    write_ivf_frame_header(pkt, outfile_);
+    (void)fwrite(pkt->data.frame.buf, 1, pkt->data.frame.sz, outfile_);
+  }
+#endif
+
+  double frame0_psnr_;
+#if WRITE_COMPRESSED_STREAM
+  FILE *outfile_;
+  unsigned int out_frames_;
+#endif
+};
+
+class ResizingCspVideoSource : public ::libaom_test::DummyVideoSource {
+ public:
+  explicit ResizingCspVideoSource(aom_img_fmt_t image_format) {
+    SetSize(kInitialWidth, kInitialHeight);
+    SetImageFormat(image_format);
+    limit_ = 30;
+  }
+
+  virtual ~ResizingCspVideoSource() {}
+};
+
+#if (defined(DISABLE_TRELLISQ_SEARCH) && DISABLE_TRELLISQ_SEARCH)
+TEST_P(ResizeCspTest, DISABLED_TestResizeCspWorks) {
+#else
+TEST_P(ResizeCspTest, TestResizeCspWorks) {
+#endif
+  const aom_img_fmt_t image_formats[] = { AOM_IMG_FMT_I420, AOM_IMG_FMT_I444 };
+  for (size_t i = 0; i < GTEST_ARRAY_SIZE_(image_formats); ++i) {
+    ResizingCspVideoSource video(image_formats[i]);
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    cfg_.rc_min_quantizer = cfg_.rc_max_quantizer = 48;
+    cfg_.g_lag_in_frames = 0;
+    cfg_.g_profile = (image_formats[i] == AOM_IMG_FMT_I420) ? 0 : 1;
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+    // Check we decoded the same number of frames as we attempted to encode
+    ASSERT_EQ(frame_info_list_.size(), video.limit());
+    frame_info_list_.clear();
+  }
+}
+
+AV1_INSTANTIATE_TEST_CASE(ResizeTest,
+                          ::testing::Values(::libaom_test::kRealTime));
+AV1_INSTANTIATE_TEST_CASE(ResizeInternalTestLarge,
+                          ::testing::Values(::libaom_test::kOnePassGood));
+AV1_INSTANTIATE_TEST_CASE(ResizeRealtimeTest,
+                          ::testing::Values(::libaom_test::kRealTime),
+                          ::testing::Range(5, 9));
+AV1_INSTANTIATE_TEST_CASE(ResizeCspTest,
+                          ::testing::Values(::libaom_test::kRealTime));
+}  // namespace
diff --git a/third_party/aom/test/run_encodes.sh b/third_party/aom/test/run_encodes.sh
new file mode 100755
index 0000000000..2096d8b158
--- /dev/null
+++ b/third_party/aom/test/run_encodes.sh
@@ -0,0 +1,39 @@
+#!/bin/bash
+#
+# Copyright (c) 2016, Alliance for Open Media. All rights reserved.
+#
+# This source code is subject to the terms of the BSD 2 Clause License and
+# the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+# was not distributed with this source code in the LICENSE file, you can
+# obtain it at www.aomedia.org/license/software. If the Alliance for Open
+# Media Patent License 1.0 was not distributed with this source code in the
+# PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+#
+# Author: jimbankoski@google.com (Jim Bankoski)
+
+if [[ $# -ne 4 ]]; then
+  echo Encodes all the y4m files in the directory at the bitrates specified by
+  echo the first 3 parameters and stores the results in a subdirectory named by
+  echo the 4th parameter:
+  echo
+  echo Usage:    run_encodes.sh start-kbps end-kbps step-kbps output-directory
+  echo Example:  run_encodes.sh 200 500 50 baseline
+  exit
+fi
+
+s=$1
+e=$2
+step=$3
+newdir=$4
+
+for i in ./*y4m; do
+  for (( b=$s; b<= $e; b+= $step ))
+  do
+    best_encode.sh $i $b
+  done
+  mv opsnr.stt $i.stt
+done
+
+mkdir $newdir
+mv *.stt $newdir
+mv *.webm $newdir
diff --git a/third_party/aom/test/sad_test.cc b/third_party/aom/test/sad_test.cc
new file mode 100644
index 0000000000..845fe79da1
--- /dev/null
+++ b/third_party/aom/test/sad_test.cc
@@ -0,0 +1,1528 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string.h>
+#include <limits.h>
+#include <stdio.h>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "aom/aom_codec.h"
+#include "aom_mem/aom_mem.h"
+#include "aom_ports/mem.h"
+
+typedef unsigned int (*SadMxNFunc)(const uint8_t *src_ptr, int src_stride,
+                                   const uint8_t *ref_ptr, int ref_stride);
+typedef ::testing::tuple<int, int, SadMxNFunc, int> SadMxNParam;
+
+typedef uint32_t (*SadMxNAvgFunc)(const uint8_t *src_ptr, int src_stride,
+                                  const uint8_t *ref_ptr, int ref_stride,
+                                  const uint8_t *second_pred);
+typedef ::testing::tuple<int, int, SadMxNAvgFunc, int> SadMxNAvgParam;
+
+typedef void (*JntCompAvgFunc)(uint8_t *comp_pred, const uint8_t *pred,
+                               int width, int height, const uint8_t *ref,
+                               int ref_stride,
+                               const JNT_COMP_PARAMS *jcp_param);
+typedef ::testing::tuple<int, int, JntCompAvgFunc, int> JntCompAvgParam;
+
+typedef unsigned int (*JntSadMxhFunc)(const uint8_t *src_ptr, int src_stride,
+                                      const uint8_t *ref_ptr, int ref_stride,
+                                      int width, int height);
+typedef ::testing::tuple<int, int, JntSadMxhFunc, int> JntSadMxhParam;
+
+typedef uint32_t (*JntSadMxNAvgFunc)(const uint8_t *src_ptr, int src_stride,
+                                     const uint8_t *ref_ptr, int ref_stride,
+                                     const uint8_t *second_pred,
+                                     const JNT_COMP_PARAMS *jcp_param);
+typedef ::testing::tuple<int, int, JntSadMxNAvgFunc, int> JntSadMxNAvgParam;
+
+typedef void (*SadMxNx4Func)(const uint8_t *src_ptr, int src_stride,
+                             const uint8_t *const ref_ptr[], int ref_stride,
+                             uint32_t *sad_array);
+typedef ::testing::tuple<int, int, SadMxNx4Func, int> SadMxNx4Param;
+
+using libaom_test::ACMRandom;
+
+namespace {
+class SADTestBase : public ::testing::Test {
+ public:
+  SADTestBase(int width, int height, int bit_depth)
+      : width_(width), height_(height), bd_(bit_depth) {}
+
+  static void SetUpTestCase() {
+    source_data8_ = reinterpret_cast<uint8_t *>(
+        aom_memalign(kDataAlignment, kDataBlockSize));
+    reference_data8_ = reinterpret_cast<uint8_t *>(
+        aom_memalign(kDataAlignment, kDataBufferSize));
+    second_pred8_ =
+        reinterpret_cast<uint8_t *>(aom_memalign(kDataAlignment, 128 * 128));
+    comp_pred8_ =
+        reinterpret_cast<uint8_t *>(aom_memalign(kDataAlignment, 128 * 128));
+    comp_pred8_test_ =
+        reinterpret_cast<uint8_t *>(aom_memalign(kDataAlignment, 128 * 128));
+    source_data16_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(kDataAlignment, kDataBlockSize * sizeof(uint16_t)));
+    reference_data16_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(kDataAlignment, kDataBufferSize * sizeof(uint16_t)));
+    second_pred16_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(kDataAlignment, 128 * 128 * sizeof(uint16_t)));
+    comp_pred16_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(kDataAlignment, 128 * 128 * sizeof(uint16_t)));
+    comp_pred16_test_ = reinterpret_cast<uint16_t *>(
+        aom_memalign(kDataAlignment, 128 * 128 * sizeof(uint16_t)));
+  }
+
+  static void TearDownTestCase() {
+    aom_free(source_data8_);
+    source_data8_ = NULL;
+    aom_free(reference_data8_);
+    reference_data8_ = NULL;
+    aom_free(second_pred8_);
+    second_pred8_ = NULL;
+    aom_free(comp_pred8_);
+    comp_pred8_ = NULL;
+    aom_free(comp_pred8_test_);
+    comp_pred8_test_ = NULL;
+    aom_free(source_data16_);
+    source_data16_ = NULL;
+    aom_free(reference_data16_);
+    reference_data16_ = NULL;
+    aom_free(second_pred16_);
+    second_pred16_ = NULL;
+    aom_free(comp_pred16_);
+    comp_pred16_ = NULL;
+    aom_free(comp_pred16_test_);
+    comp_pred16_test_ = NULL;
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  // Handle up to 4 128x128 blocks, with stride up to 256
+  static const int kDataAlignment = 16;
+  static const int kDataBlockSize = 128 * 256;
+  static const int kDataBufferSize = 4 * kDataBlockSize;
+
+  virtual void SetUp() {
+    if (bd_ == -1) {
+      use_high_bit_depth_ = false;
+      bit_depth_ = AOM_BITS_8;
+      source_data_ = source_data8_;
+      reference_data_ = reference_data8_;
+      second_pred_ = second_pred8_;
+      comp_pred_ = comp_pred8_;
+      comp_pred_test_ = comp_pred8_test_;
+    } else {
+      use_high_bit_depth_ = true;
+      bit_depth_ = static_cast<aom_bit_depth_t>(bd_);
+      source_data_ = CONVERT_TO_BYTEPTR(source_data16_);
+      reference_data_ = CONVERT_TO_BYTEPTR(reference_data16_);
+      second_pred_ = CONVERT_TO_BYTEPTR(second_pred16_);
+      comp_pred_ = CONVERT_TO_BYTEPTR(comp_pred16_);
+      comp_pred_test_ = CONVERT_TO_BYTEPTR(comp_pred16_test_);
+    }
+    mask_ = (1 << bit_depth_) - 1;
+    source_stride_ = (width_ + 31) & ~31;
+    reference_stride_ = width_ * 2;
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+  }
+
+  virtual uint8_t *GetReference(int block_idx) {
+    if (use_high_bit_depth_)
+      return CONVERT_TO_BYTEPTR(CONVERT_TO_SHORTPTR(reference_data_) +
+                                block_idx * kDataBlockSize);
+    return reference_data_ + block_idx * kDataBlockSize;
+  }
+
+  // Sum of Absolute Differences. Given two blocks, calculate the absolute
+  // difference between two pixels in the same relative location; accumulate.
+  unsigned int ReferenceSAD(int block_idx) {
+    unsigned int sad = 0;
+    const uint8_t *const reference8 = GetReference(block_idx);
+    const uint8_t *const source8 = source_data_;
+    const uint16_t *const reference16 =
+        CONVERT_TO_SHORTPTR(GetReference(block_idx));
+    const uint16_t *const source16 = CONVERT_TO_SHORTPTR(source_data_);
+    for (int h = 0; h < height_; ++h) {
+      for (int w = 0; w < width_; ++w) {
+        if (!use_high_bit_depth_) {
+          sad += abs(source8[h * source_stride_ + w] -
+                     reference8[h * reference_stride_ + w]);
+        } else {
+          sad += abs(source16[h * source_stride_ + w] -
+                     reference16[h * reference_stride_ + w]);
+        }
+      }
+    }
+    return sad;
+  }
+
+  // Sum of Absolute Differences Average. Given two blocks, and a prediction
+  // calculate the absolute difference between one pixel and average of the
+  // corresponding and predicted pixels; accumulate.
+  unsigned int ReferenceSADavg(int block_idx) {
+    unsigned int sad = 0;
+    const uint8_t *const reference8 = GetReference(block_idx);
+    const uint8_t *const source8 = source_data_;
+    const uint8_t *const second_pred8 = second_pred_;
+    const uint16_t *const reference16 =
+        CONVERT_TO_SHORTPTR(GetReference(block_idx));
+    const uint16_t *const source16 = CONVERT_TO_SHORTPTR(source_data_);
+    const uint16_t *const second_pred16 = CONVERT_TO_SHORTPTR(second_pred_);
+    for (int h = 0; h < height_; ++h) {
+      for (int w = 0; w < width_; ++w) {
+        if (!use_high_bit_depth_) {
+          const int tmp = second_pred8[h * width_ + w] +
+                          reference8[h * reference_stride_ + w];
+          const uint8_t comp_pred = ROUND_POWER_OF_TWO(tmp, 1);
+          sad += abs(source8[h * source_stride_ + w] - comp_pred);
+        } else {
+          const int tmp = second_pred16[h * width_ + w] +
+                          reference16[h * reference_stride_ + w];
+          const uint16_t comp_pred = ROUND_POWER_OF_TWO(tmp, 1);
+          sad += abs(source16[h * source_stride_ + w] - comp_pred);
+        }
+      }
+    }
+    return sad;
+  }
+
+  void ReferenceJntCompAvg(int block_idx) {
+    const uint8_t *const reference8 = GetReference(block_idx);
+    const uint8_t *const second_pred8 = second_pred_;
+    uint8_t *const comp_pred8 = comp_pred_;
+    const uint16_t *const reference16 =
+        CONVERT_TO_SHORTPTR(GetReference(block_idx));
+    const uint16_t *const second_pred16 = CONVERT_TO_SHORTPTR(second_pred_);
+    uint16_t *const comp_pred16 = CONVERT_TO_SHORTPTR(comp_pred_);
+    for (int h = 0; h < height_; ++h) {
+      for (int w = 0; w < width_; ++w) {
+        if (!use_high_bit_depth_) {
+          const int tmp =
+              second_pred8[h * width_ + w] * jcp_param_.bck_offset +
+              reference8[h * reference_stride_ + w] * jcp_param_.fwd_offset;
+          comp_pred8[h * width_ + w] = ROUND_POWER_OF_TWO(tmp, 4);
+        } else {
+          const int tmp =
+              second_pred16[h * width_ + w] * jcp_param_.bck_offset +
+              reference16[h * reference_stride_ + w] * jcp_param_.fwd_offset;
+          comp_pred16[h * width_ + w] = ROUND_POWER_OF_TWO(tmp, 4);
+        }
+      }
+    }
+  }
+
+  unsigned int ReferenceJntSADavg(int block_idx) {
+    unsigned int sad = 0;
+    const uint8_t *const reference8 = GetReference(block_idx);
+    const uint8_t *const source8 = source_data_;
+    const uint8_t *const second_pred8 = second_pred_;
+    const uint16_t *const reference16 =
+        CONVERT_TO_SHORTPTR(GetReference(block_idx));
+    const uint16_t *const source16 = CONVERT_TO_SHORTPTR(source_data_);
+    const uint16_t *const second_pred16 = CONVERT_TO_SHORTPTR(second_pred_);
+    for (int h = 0; h < height_; ++h) {
+      for (int w = 0; w < width_; ++w) {
+        if (!use_high_bit_depth_) {
+          const int tmp =
+              second_pred8[h * width_ + w] * jcp_param_.bck_offset +
+              reference8[h * reference_stride_ + w] * jcp_param_.fwd_offset;
+          const uint8_t comp_pred = ROUND_POWER_OF_TWO(tmp, 4);
+          sad += abs(source8[h * source_stride_ + w] - comp_pred);
+        } else {
+          const int tmp =
+              second_pred16[h * width_ + w] * jcp_param_.bck_offset +
+              reference16[h * reference_stride_ + w] * jcp_param_.fwd_offset;
+          const uint16_t comp_pred = ROUND_POWER_OF_TWO(tmp, 4);
+          sad += abs(source16[h * source_stride_ + w] - comp_pred);
+        }
+      }
+    }
+    return sad;
+  }
+
+  void FillConstant(uint8_t *data, int stride, uint16_t fill_constant) {
+    uint8_t *data8 = data;
+    uint16_t *data16 = CONVERT_TO_SHORTPTR(data);
+    for (int h = 0; h < height_; ++h) {
+      for (int w = 0; w < width_; ++w) {
+        if (!use_high_bit_depth_) {
+          data8[h * stride + w] = static_cast<uint8_t>(fill_constant);
+        } else {
+          data16[h * stride + w] = fill_constant;
+        }
+      }
+    }
+  }
+
+  void FillRandom(uint8_t *data, int stride) {
+    uint8_t *data8 = data;
+    uint16_t *data16 = CONVERT_TO_SHORTPTR(data);
+    for (int h = 0; h < height_; ++h) {
+      for (int w = 0; w < width_; ++w) {
+        if (!use_high_bit_depth_) {
+          data8[h * stride + w] = rnd_.Rand8();
+        } else {
+          data16[h * stride + w] = rnd_.Rand16() & mask_;
+        }
+      }
+    }
+  }
+
+  int width_, height_, mask_, bd_;
+  aom_bit_depth_t bit_depth_;
+  static uint8_t *source_data_;
+  static uint8_t *reference_data_;
+  static uint8_t *second_pred_;
+  int source_stride_;
+  bool use_high_bit_depth_;
+  static uint8_t *source_data8_;
+  static uint8_t *reference_data8_;
+  static uint8_t *second_pred8_;
+  static uint16_t *source_data16_;
+  static uint16_t *reference_data16_;
+  static uint16_t *second_pred16_;
+  int reference_stride_;
+  static uint8_t *comp_pred_;
+  static uint8_t *comp_pred8_;
+  static uint16_t *comp_pred16_;
+  static uint8_t *comp_pred_test_;
+  static uint8_t *comp_pred8_test_;
+  static uint16_t *comp_pred16_test_;
+  JNT_COMP_PARAMS jcp_param_;
+
+  ACMRandom rnd_;
+};
+
+class SADx4Test : public ::testing::WithParamInterface<SadMxNx4Param>,
+                  public SADTestBase {
+ public:
+  SADx4Test() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+
+ protected:
+  void SADs(unsigned int *results) {
+    const uint8_t *references[] = { GetReference(0), GetReference(1),
+                                    GetReference(2), GetReference(3) };
+
+    ASM_REGISTER_STATE_CHECK(GET_PARAM(2)(
+        source_data_, source_stride_, references, reference_stride_, results));
+  }
+
+  void CheckSADs() {
+    unsigned int reference_sad, exp_sad[4];
+
+    SADs(exp_sad);
+    for (int block = 0; block < 4; ++block) {
+      reference_sad = ReferenceSAD(block);
+
+      EXPECT_EQ(reference_sad, exp_sad[block]) << "block " << block;
+    }
+  }
+};
+
+class SADTest : public ::testing::WithParamInterface<SadMxNParam>,
+                public SADTestBase {
+ public:
+  SADTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+
+ protected:
+  unsigned int SAD(int block_idx) {
+    unsigned int ret;
+    const uint8_t *const reference = GetReference(block_idx);
+
+    ASM_REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
+                                                reference, reference_stride_));
+    return ret;
+  }
+
+  void CheckSAD() {
+    const unsigned int reference_sad = ReferenceSAD(0);
+    const unsigned int exp_sad = SAD(0);
+
+    ASSERT_EQ(reference_sad, exp_sad);
+  }
+
+  void SpeedSAD() {
+    int test_count = 20000000;
+    while (test_count > 0) {
+      SAD(0);
+      test_count -= 1;
+    }
+  }
+};
+
+class SADavgTest : public ::testing::WithParamInterface<SadMxNAvgParam>,
+                   public SADTestBase {
+ public:
+  SADavgTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+
+ protected:
+  unsigned int SAD_avg(int block_idx) {
+    unsigned int ret;
+    const uint8_t *const reference = GetReference(block_idx);
+
+    ASM_REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
+                                                reference, reference_stride_,
+                                                second_pred_));
+    return ret;
+  }
+
+  void CheckSAD() {
+    const unsigned int reference_sad = ReferenceSADavg(0);
+    const unsigned int exp_sad = SAD_avg(0);
+
+    ASSERT_EQ(reference_sad, exp_sad);
+  }
+};
+
+class JntCompAvgTest : public ::testing::WithParamInterface<JntCompAvgParam>,
+                       public SADTestBase {
+ public:
+  JntCompAvgTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+
+ protected:
+  void jnt_comp_avg(int block_idx) {
+    const uint8_t *const reference = GetReference(block_idx);
+
+    ASM_REGISTER_STATE_CHECK(GET_PARAM(2)(comp_pred_test_, second_pred_, width_,
+                                          height_, reference, reference_stride_,
+                                          &jcp_param_));
+  }
+
+  void CheckCompAvg() {
+    for (int j = 0; j < 2; ++j) {
+      for (int i = 0; i < 4; ++i) {
+        jcp_param_.fwd_offset = quant_dist_lookup_table[j][i][0];
+        jcp_param_.bck_offset = quant_dist_lookup_table[j][i][1];
+
+        ReferenceJntCompAvg(0);
+        jnt_comp_avg(0);
+
+        for (int y = 0; y < height_; ++y)
+          for (int x = 0; x < width_; ++x)
+            ASSERT_EQ(comp_pred_[y * width_ + x],
+                      comp_pred_test_[y * width_ + x]);
+      }
+    }
+  }
+};
+
+class JntSADTest : public ::testing::WithParamInterface<JntSadMxhParam>,
+                   public SADTestBase {
+ public:
+  JntSADTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+
+ protected:
+  unsigned int SAD(int block_idx) {
+    unsigned int ret;
+    const uint8_t *const reference = GetReference(block_idx);
+
+    ASM_REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
+                                                reference, reference_stride_,
+                                                GET_PARAM(0), GET_PARAM(1)));
+    return ret;
+  }
+
+  void CheckSAD() {
+    const unsigned int reference_sad = ReferenceSAD(0);
+    const unsigned int exp_sad = SAD(0);
+
+    ASSERT_EQ(reference_sad, exp_sad);
+  }
+
+  void SpeedSAD() {
+    int test_count = 20000000;
+    while (test_count > 0) {
+      SAD(0);
+      test_count -= 1;
+    }
+  }
+};
+
+class JntSADavgTest : public ::testing::WithParamInterface<JntSadMxNAvgParam>,
+                      public SADTestBase {
+ public:
+  JntSADavgTest() : SADTestBase(GET_PARAM(0), GET_PARAM(1), GET_PARAM(3)) {}
+
+ protected:
+  unsigned int jnt_SAD_avg(int block_idx) {
+    unsigned int ret;
+    const uint8_t *const reference = GetReference(block_idx);
+
+    ASM_REGISTER_STATE_CHECK(ret = GET_PARAM(2)(source_data_, source_stride_,
+                                                reference, reference_stride_,
+                                                second_pred_, &jcp_param_));
+    return ret;
+  }
+
+  void CheckSAD() {
+    for (int j = 0; j < 2; ++j) {
+      for (int i = 0; i < 4; ++i) {
+        jcp_param_.fwd_offset = quant_dist_lookup_table[j][i][0];
+        jcp_param_.bck_offset = quant_dist_lookup_table[j][i][1];
+
+        const unsigned int reference_sad = ReferenceJntSADavg(0);
+        const unsigned int exp_sad = jnt_SAD_avg(0);
+
+        ASSERT_EQ(reference_sad, exp_sad);
+      }
+    }
+  }
+};
+
+uint8_t *SADTestBase::source_data_ = NULL;
+uint8_t *SADTestBase::reference_data_ = NULL;
+uint8_t *SADTestBase::second_pred_ = NULL;
+uint8_t *SADTestBase::comp_pred_ = NULL;
+uint8_t *SADTestBase::comp_pred_test_ = NULL;
+uint8_t *SADTestBase::source_data8_ = NULL;
+uint8_t *SADTestBase::reference_data8_ = NULL;
+uint8_t *SADTestBase::second_pred8_ = NULL;
+uint8_t *SADTestBase::comp_pred8_ = NULL;
+uint8_t *SADTestBase::comp_pred8_test_ = NULL;
+uint16_t *SADTestBase::source_data16_ = NULL;
+uint16_t *SADTestBase::reference_data16_ = NULL;
+uint16_t *SADTestBase::second_pred16_ = NULL;
+uint16_t *SADTestBase::comp_pred16_ = NULL;
+uint16_t *SADTestBase::comp_pred16_test_ = NULL;
+
+TEST_P(SADTest, MaxRef) {
+  FillConstant(source_data_, source_stride_, 0);
+  FillConstant(reference_data_, reference_stride_, mask_);
+  CheckSAD();
+}
+
+TEST_P(SADTest, MaxSrc) {
+  FillConstant(source_data_, source_stride_, mask_);
+  FillConstant(reference_data_, reference_stride_, 0);
+  CheckSAD();
+}
+
+TEST_P(SADTest, ShortRef) {
+  const int tmp_stride = reference_stride_;
+  reference_stride_ >>= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  CheckSAD();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(SADTest, UnalignedRef) {
+  // The reference frame, but not the source frame, may be unaligned for
+  // certain types of searches.
+  const int tmp_stride = reference_stride_;
+  reference_stride_ -= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  CheckSAD();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(SADTest, ShortSrc) {
+  const int tmp_stride = source_stride_;
+  source_stride_ >>= 1;
+  int test_count = 2000;
+  while (test_count > 0) {
+    FillRandom(source_data_, source_stride_);
+    FillRandom(reference_data_, reference_stride_);
+    CheckSAD();
+    test_count -= 1;
+  }
+  source_stride_ = tmp_stride;
+}
+
+#define SPEED_TEST (0)
+#if SPEED_TEST
+TEST_P(SADTest, Speed) {
+  const int tmp_stride = source_stride_;
+  source_stride_ >>= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  SpeedSAD();
+  source_stride_ = tmp_stride;
+}
+#endif
+
+TEST_P(SADavgTest, MaxRef) {
+  FillConstant(source_data_, source_stride_, 0);
+  FillConstant(reference_data_, reference_stride_, mask_);
+  FillConstant(second_pred_, width_, 0);
+  CheckSAD();
+}
+TEST_P(SADavgTest, MaxSrc) {
+  FillConstant(source_data_, source_stride_, mask_);
+  FillConstant(reference_data_, reference_stride_, 0);
+  FillConstant(second_pred_, width_, 0);
+  CheckSAD();
+}
+
+TEST_P(SADavgTest, ShortRef) {
+  const int tmp_stride = reference_stride_;
+  reference_stride_ >>= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  FillRandom(second_pred_, width_);
+  CheckSAD();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(SADavgTest, UnalignedRef) {
+  // The reference frame, but not the source frame, may be unaligned for
+  // certain types of searches.
+  const int tmp_stride = reference_stride_;
+  reference_stride_ -= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  FillRandom(second_pred_, width_);
+  CheckSAD();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(SADavgTest, ShortSrc) {
+  const int tmp_stride = source_stride_;
+  source_stride_ >>= 1;
+  int test_count = 2000;
+  while (test_count > 0) {
+    FillRandom(source_data_, source_stride_);
+    FillRandom(reference_data_, reference_stride_);
+    FillRandom(second_pred_, width_);
+    CheckSAD();
+    test_count -= 1;
+  }
+  source_stride_ = tmp_stride;
+}
+
+TEST_P(JntCompAvgTest, MaxRef) {
+  FillConstant(reference_data_, reference_stride_, mask_);
+  FillConstant(second_pred_, width_, 0);
+  CheckCompAvg();
+}
+
+TEST_P(JntCompAvgTest, MaxSecondPred) {
+  FillConstant(reference_data_, reference_stride_, 0);
+  FillConstant(second_pred_, width_, mask_);
+  CheckCompAvg();
+}
+
+TEST_P(JntCompAvgTest, ShortRef) {
+  const int tmp_stride = reference_stride_;
+  reference_stride_ >>= 1;
+  FillRandom(reference_data_, reference_stride_);
+  FillRandom(second_pred_, width_);
+  CheckCompAvg();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(JntCompAvgTest, UnalignedRef) {
+  // The reference frame, but not the source frame, may be unaligned for
+  // certain types of searches.
+  const int tmp_stride = reference_stride_;
+  reference_stride_ -= 1;
+  FillRandom(reference_data_, reference_stride_);
+  FillRandom(second_pred_, width_);
+  CheckCompAvg();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(JntSADTest, MaxRef) {
+  FillConstant(source_data_, source_stride_, 0);
+  FillConstant(reference_data_, reference_stride_, mask_);
+  CheckSAD();
+}
+
+TEST_P(JntSADTest, MaxSrc) {
+  FillConstant(source_data_, source_stride_, mask_);
+  FillConstant(reference_data_, reference_stride_, 0);
+  CheckSAD();
+}
+
+TEST_P(JntSADTest, ShortRef) {
+  const int tmp_stride = reference_stride_;
+  reference_stride_ >>= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  CheckSAD();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(JntSADTest, UnalignedRef) {
+  // The reference frame, but not the source frame, may be unaligned for
+  // certain types of searches.
+  const int tmp_stride = reference_stride_;
+  reference_stride_ -= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  CheckSAD();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(JntSADTest, ShortSrc) {
+  const int tmp_stride = source_stride_;
+  source_stride_ >>= 1;
+  int test_count = 2000;
+  while (test_count > 0) {
+    FillRandom(source_data_, source_stride_);
+    FillRandom(reference_data_, reference_stride_);
+    CheckSAD();
+    test_count -= 1;
+  }
+  source_stride_ = tmp_stride;
+}
+
+TEST_P(JntSADavgTest, MaxRef) {
+  FillConstant(source_data_, source_stride_, 0);
+  FillConstant(reference_data_, reference_stride_, mask_);
+  FillConstant(second_pred_, width_, 0);
+  CheckSAD();
+}
+TEST_P(JntSADavgTest, MaxSrc) {
+  FillConstant(source_data_, source_stride_, mask_);
+  FillConstant(reference_data_, reference_stride_, 0);
+  FillConstant(second_pred_, width_, 0);
+  CheckSAD();
+}
+
+TEST_P(JntSADavgTest, ShortRef) {
+  const int tmp_stride = reference_stride_;
+  reference_stride_ >>= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  FillRandom(second_pred_, width_);
+  CheckSAD();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(JntSADavgTest, UnalignedRef) {
+  // The reference frame, but not the source frame, may be unaligned for
+  // certain types of searches.
+  const int tmp_stride = reference_stride_;
+  reference_stride_ -= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(reference_data_, reference_stride_);
+  FillRandom(second_pred_, width_);
+  CheckSAD();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(JntSADavgTest, ShortSrc) {
+  const int tmp_stride = source_stride_;
+  source_stride_ >>= 1;
+  int test_count = 2000;
+  while (test_count > 0) {
+    FillRandom(source_data_, source_stride_);
+    FillRandom(reference_data_, reference_stride_);
+    FillRandom(second_pred_, width_);
+    CheckSAD();
+    test_count -= 1;
+  }
+  source_stride_ = tmp_stride;
+}
+
+TEST_P(SADx4Test, MaxRef) {
+  FillConstant(source_data_, source_stride_, 0);
+  FillConstant(GetReference(0), reference_stride_, mask_);
+  FillConstant(GetReference(1), reference_stride_, mask_);
+  FillConstant(GetReference(2), reference_stride_, mask_);
+  FillConstant(GetReference(3), reference_stride_, mask_);
+  CheckSADs();
+}
+
+TEST_P(SADx4Test, MaxSrc) {
+  FillConstant(source_data_, source_stride_, mask_);
+  FillConstant(GetReference(0), reference_stride_, 0);
+  FillConstant(GetReference(1), reference_stride_, 0);
+  FillConstant(GetReference(2), reference_stride_, 0);
+  FillConstant(GetReference(3), reference_stride_, 0);
+  CheckSADs();
+}
+
+TEST_P(SADx4Test, ShortRef) {
+  int tmp_stride = reference_stride_;
+  reference_stride_ >>= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(GetReference(0), reference_stride_);
+  FillRandom(GetReference(1), reference_stride_);
+  FillRandom(GetReference(2), reference_stride_);
+  FillRandom(GetReference(3), reference_stride_);
+  CheckSADs();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(SADx4Test, UnalignedRef) {
+  // The reference frame, but not the source frame, may be unaligned for
+  // certain types of searches.
+  int tmp_stride = reference_stride_;
+  reference_stride_ -= 1;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(GetReference(0), reference_stride_);
+  FillRandom(GetReference(1), reference_stride_);
+  FillRandom(GetReference(2), reference_stride_);
+  FillRandom(GetReference(3), reference_stride_);
+  CheckSADs();
+  reference_stride_ = tmp_stride;
+}
+
+TEST_P(SADx4Test, ShortSrc) {
+  int tmp_stride = source_stride_;
+  source_stride_ >>= 1;
+  int test_count = 1000;
+  while (test_count > 0) {
+    FillRandom(source_data_, source_stride_);
+    FillRandom(GetReference(0), reference_stride_);
+    FillRandom(GetReference(1), reference_stride_);
+    FillRandom(GetReference(2), reference_stride_);
+    FillRandom(GetReference(3), reference_stride_);
+    CheckSADs();
+    test_count -= 1;
+  }
+  source_stride_ = tmp_stride;
+}
+
+TEST_P(SADx4Test, SrcAlignedByWidth) {
+  uint8_t *tmp_source_data = source_data_;
+  source_data_ += width_;
+  FillRandom(source_data_, source_stride_);
+  FillRandom(GetReference(0), reference_stride_);
+  FillRandom(GetReference(1), reference_stride_);
+  FillRandom(GetReference(2), reference_stride_);
+  FillRandom(GetReference(3), reference_stride_);
+  CheckSADs();
+  source_data_ = tmp_source_data;
+}
+
+using ::testing::make_tuple;
+
+//------------------------------------------------------------------------------
+// C functions
+const SadMxNParam c_tests[] = {
+  make_tuple(128, 128, &aom_sad128x128_c, -1),
+  make_tuple(128, 64, &aom_sad128x64_c, -1),
+  make_tuple(64, 128, &aom_sad64x128_c, -1),
+  make_tuple(64, 64, &aom_sad64x64_c, -1),
+  make_tuple(64, 32, &aom_sad64x32_c, -1),
+  make_tuple(32, 64, &aom_sad32x64_c, -1),
+  make_tuple(32, 32, &aom_sad32x32_c, -1),
+  make_tuple(32, 16, &aom_sad32x16_c, -1),
+  make_tuple(16, 32, &aom_sad16x32_c, -1),
+  make_tuple(16, 16, &aom_sad16x16_c, -1),
+  make_tuple(16, 8, &aom_sad16x8_c, -1),
+  make_tuple(8, 16, &aom_sad8x16_c, -1),
+  make_tuple(8, 8, &aom_sad8x8_c, -1),
+  make_tuple(8, 4, &aom_sad8x4_c, -1),
+  make_tuple(4, 8, &aom_sad4x8_c, -1),
+  make_tuple(4, 4, &aom_sad4x4_c, -1),
+  make_tuple(128, 128, &aom_highbd_sad128x128_c, 8),
+  make_tuple(128, 64, &aom_highbd_sad128x64_c, 8),
+  make_tuple(64, 128, &aom_highbd_sad64x128_c, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64_c, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32_c, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64_c, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32_c, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16_c, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32_c, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16_c, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8_c, 8),
+  make_tuple(8, 16, &aom_highbd_sad8x16_c, 8),
+  make_tuple(8, 8, &aom_highbd_sad8x8_c, 8),
+  make_tuple(8, 4, &aom_highbd_sad8x4_c, 8),
+  make_tuple(4, 8, &aom_highbd_sad4x8_c, 8),
+  make_tuple(4, 4, &aom_highbd_sad4x4_c, 8),
+  make_tuple(128, 128, &aom_highbd_sad128x128_c, 10),
+  make_tuple(128, 64, &aom_highbd_sad128x64_c, 10),
+  make_tuple(64, 128, &aom_highbd_sad64x128_c, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64_c, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32_c, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64_c, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32_c, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16_c, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32_c, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16_c, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8_c, 10),
+  make_tuple(8, 16, &aom_highbd_sad8x16_c, 10),
+  make_tuple(8, 8, &aom_highbd_sad8x8_c, 10),
+  make_tuple(8, 4, &aom_highbd_sad8x4_c, 10),
+  make_tuple(4, 8, &aom_highbd_sad4x8_c, 10),
+  make_tuple(4, 4, &aom_highbd_sad4x4_c, 10),
+  make_tuple(128, 128, &aom_highbd_sad128x128_c, 12),
+  make_tuple(128, 64, &aom_highbd_sad128x64_c, 12),
+  make_tuple(64, 128, &aom_highbd_sad64x128_c, 12),
+  make_tuple(64, 64, &aom_highbd_sad64x64_c, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32_c, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64_c, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32_c, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16_c, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32_c, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16_c, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8_c, 12),
+  make_tuple(8, 16, &aom_highbd_sad8x16_c, 12),
+  make_tuple(8, 8, &aom_highbd_sad8x8_c, 12),
+  make_tuple(8, 4, &aom_highbd_sad8x4_c, 12),
+  make_tuple(4, 8, &aom_highbd_sad4x8_c, 12),
+  make_tuple(4, 4, &aom_highbd_sad4x4_c, 12),
+};
+INSTANTIATE_TEST_CASE_P(C, SADTest, ::testing::ValuesIn(c_tests));
+
+const SadMxNAvgParam avg_c_tests[] = {
+  make_tuple(128, 128, &aom_sad128x128_avg_c, -1),
+  make_tuple(128, 64, &aom_sad128x64_avg_c, -1),
+  make_tuple(64, 128, &aom_sad64x128_avg_c, -1),
+  make_tuple(64, 64, &aom_sad64x64_avg_c, -1),
+  make_tuple(64, 32, &aom_sad64x32_avg_c, -1),
+  make_tuple(32, 64, &aom_sad32x64_avg_c, -1),
+  make_tuple(32, 32, &aom_sad32x32_avg_c, -1),
+  make_tuple(32, 16, &aom_sad32x16_avg_c, -1),
+  make_tuple(16, 32, &aom_sad16x32_avg_c, -1),
+  make_tuple(16, 16, &aom_sad16x16_avg_c, -1),
+  make_tuple(16, 8, &aom_sad16x8_avg_c, -1),
+  make_tuple(8, 16, &aom_sad8x16_avg_c, -1),
+  make_tuple(8, 8, &aom_sad8x8_avg_c, -1),
+  make_tuple(8, 4, &aom_sad8x4_avg_c, -1),
+  make_tuple(4, 8, &aom_sad4x8_avg_c, -1),
+  make_tuple(4, 4, &aom_sad4x4_avg_c, -1),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avg_c, 8),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avg_c, 8),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avg_c, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_c, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_c, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_c, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_c, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_c, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_c, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_c, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_c, 8),
+  make_tuple(8, 16, &aom_highbd_sad8x16_avg_c, 8),
+  make_tuple(8, 8, &aom_highbd_sad8x8_avg_c, 8),
+  make_tuple(8, 4, &aom_highbd_sad8x4_avg_c, 8),
+  make_tuple(4, 8, &aom_highbd_sad4x8_avg_c, 8),
+  make_tuple(4, 4, &aom_highbd_sad4x4_avg_c, 8),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avg_c, 10),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avg_c, 10),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avg_c, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_c, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_c, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_c, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_c, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_c, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_c, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_c, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_c, 10),
+  make_tuple(8, 16, &aom_highbd_sad8x16_avg_c, 10),
+  make_tuple(8, 8, &aom_highbd_sad8x8_avg_c, 10),
+  make_tuple(8, 4, &aom_highbd_sad8x4_avg_c, 10),
+  make_tuple(4, 8, &aom_highbd_sad4x8_avg_c, 10),
+  make_tuple(4, 4, &aom_highbd_sad4x4_avg_c, 10),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avg_c, 12),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avg_c, 12),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avg_c, 12),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_c, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_c, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_c, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_c, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_c, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_c, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_c, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_c, 12),
+  make_tuple(8, 16, &aom_highbd_sad8x16_avg_c, 12),
+  make_tuple(8, 8, &aom_highbd_sad8x8_avg_c, 12),
+  make_tuple(8, 4, &aom_highbd_sad8x4_avg_c, 12),
+  make_tuple(4, 8, &aom_highbd_sad4x8_avg_c, 12),
+  make_tuple(4, 4, &aom_highbd_sad4x4_avg_c, 12),
+};
+INSTANTIATE_TEST_CASE_P(C, SADavgTest, ::testing::ValuesIn(avg_c_tests));
+
+// TODO(chengchen): add highbd tests
+const JntCompAvgParam jnt_comp_avg_c_tests[] = {
+  make_tuple(128, 128, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(128, 64, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(64, 128, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(64, 64, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(64, 32, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(32, 64, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(32, 32, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(32, 16, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(16, 32, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(16, 16, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(16, 8, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(8, 16, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(8, 8, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(8, 4, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(4, 8, &aom_jnt_comp_avg_pred_c, -1),
+  make_tuple(4, 4, &aom_jnt_comp_avg_pred_c, -1),
+};
+
+INSTANTIATE_TEST_CASE_P(C, JntCompAvgTest,
+                        ::testing::ValuesIn(jnt_comp_avg_c_tests));
+
+const JntSadMxNAvgParam jnt_avg_c_tests[] = {
+  make_tuple(128, 128, &aom_jnt_sad128x128_avg_c, -1),
+  make_tuple(128, 64, &aom_jnt_sad128x64_avg_c, -1),
+  make_tuple(64, 128, &aom_jnt_sad64x128_avg_c, -1),
+  make_tuple(64, 64, &aom_jnt_sad64x64_avg_c, -1),
+  make_tuple(64, 32, &aom_jnt_sad64x32_avg_c, -1),
+  make_tuple(32, 64, &aom_jnt_sad32x64_avg_c, -1),
+  make_tuple(32, 32, &aom_jnt_sad32x32_avg_c, -1),
+  make_tuple(32, 16, &aom_jnt_sad32x16_avg_c, -1),
+  make_tuple(16, 32, &aom_jnt_sad16x32_avg_c, -1),
+  make_tuple(16, 16, &aom_jnt_sad16x16_avg_c, -1),
+  make_tuple(16, 8, &aom_jnt_sad16x8_avg_c, -1),
+  make_tuple(8, 16, &aom_jnt_sad8x16_avg_c, -1),
+  make_tuple(8, 8, &aom_jnt_sad8x8_avg_c, -1),
+  make_tuple(8, 4, &aom_jnt_sad8x4_avg_c, -1),
+  make_tuple(4, 8, &aom_jnt_sad4x8_avg_c, -1),
+  make_tuple(4, 4, &aom_jnt_sad4x4_avg_c, -1),
+};
+INSTANTIATE_TEST_CASE_P(C, JntSADavgTest, ::testing::ValuesIn(jnt_avg_c_tests));
+
+const SadMxNx4Param x4d_c_tests[] = {
+  make_tuple(128, 128, &aom_sad128x128x4d_c, -1),
+  make_tuple(128, 64, &aom_sad128x64x4d_c, -1),
+  make_tuple(64, 128, &aom_sad64x128x4d_c, -1),
+  make_tuple(64, 64, &aom_sad64x64x4d_c, -1),
+  make_tuple(64, 32, &aom_sad64x32x4d_c, -1),
+  make_tuple(32, 64, &aom_sad32x64x4d_c, -1),
+  make_tuple(32, 32, &aom_sad32x32x4d_c, -1),
+  make_tuple(32, 16, &aom_sad32x16x4d_c, -1),
+  make_tuple(16, 32, &aom_sad16x32x4d_c, -1),
+  make_tuple(16, 16, &aom_sad16x16x4d_c, -1),
+  make_tuple(16, 8, &aom_sad16x8x4d_c, -1),
+  make_tuple(8, 16, &aom_sad8x16x4d_c, -1),
+  make_tuple(8, 8, &aom_sad8x8x4d_c, -1),
+  make_tuple(8, 4, &aom_sad8x4x4d_c, -1),
+  make_tuple(4, 8, &aom_sad4x8x4d_c, -1),
+  make_tuple(4, 4, &aom_sad4x4x4d_c, -1),
+  make_tuple(128, 128, &aom_highbd_sad128x128x4d_c, 8),
+  make_tuple(128, 64, &aom_highbd_sad128x64x4d_c, 8),
+  make_tuple(64, 128, &aom_highbd_sad64x128x4d_c, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_c, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_c, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_c, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_c, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_c, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_c, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_c, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_c, 8),
+  make_tuple(8, 16, &aom_highbd_sad8x16x4d_c, 8),
+  make_tuple(8, 8, &aom_highbd_sad8x8x4d_c, 8),
+  make_tuple(8, 4, &aom_highbd_sad8x4x4d_c, 8),
+  make_tuple(4, 8, &aom_highbd_sad4x8x4d_c, 8),
+  make_tuple(4, 4, &aom_highbd_sad4x4x4d_c, 8),
+  make_tuple(128, 128, &aom_highbd_sad128x128x4d_c, 10),
+  make_tuple(128, 64, &aom_highbd_sad128x64x4d_c, 10),
+  make_tuple(64, 128, &aom_highbd_sad64x128x4d_c, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_c, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_c, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_c, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_c, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_c, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_c, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_c, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_c, 10),
+  make_tuple(8, 16, &aom_highbd_sad8x16x4d_c, 10),
+  make_tuple(8, 8, &aom_highbd_sad8x8x4d_c, 10),
+  make_tuple(8, 4, &aom_highbd_sad8x4x4d_c, 10),
+  make_tuple(4, 8, &aom_highbd_sad4x8x4d_c, 10),
+  make_tuple(4, 4, &aom_highbd_sad4x4x4d_c, 10),
+  make_tuple(128, 128, &aom_highbd_sad128x128x4d_c, 12),
+  make_tuple(128, 64, &aom_highbd_sad128x64x4d_c, 12),
+  make_tuple(64, 128, &aom_highbd_sad64x128x4d_c, 12),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_c, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_c, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_c, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_c, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_c, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_c, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_c, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_c, 12),
+  make_tuple(8, 16, &aom_highbd_sad8x16x4d_c, 12),
+  make_tuple(8, 8, &aom_highbd_sad8x8x4d_c, 12),
+  make_tuple(8, 4, &aom_highbd_sad8x4x4d_c, 12),
+  make_tuple(4, 8, &aom_highbd_sad4x8x4d_c, 12),
+  make_tuple(4, 4, &aom_highbd_sad4x4x4d_c, 12),
+};
+INSTANTIATE_TEST_CASE_P(C, SADx4Test, ::testing::ValuesIn(x4d_c_tests));
+
+//------------------------------------------------------------------------------
+// ARM functions
+#if HAVE_NEON
+const SadMxNParam neon_tests[] = {
+  make_tuple(64, 64, &aom_sad64x64_neon, -1),
+  make_tuple(32, 32, &aom_sad32x32_neon, -1),
+  make_tuple(16, 16, &aom_sad16x16_neon, -1),
+  make_tuple(16, 8, &aom_sad16x8_neon, -1),
+  make_tuple(8, 16, &aom_sad8x16_neon, -1),
+  make_tuple(8, 8, &aom_sad8x8_neon, -1),
+  make_tuple(4, 4, &aom_sad4x4_neon, -1),
+};
+INSTANTIATE_TEST_CASE_P(NEON, SADTest, ::testing::ValuesIn(neon_tests));
+
+const SadMxNx4Param x4d_neon_tests[] = {
+  make_tuple(64, 64, &aom_sad64x64x4d_neon, -1),
+  make_tuple(32, 32, &aom_sad32x32x4d_neon, -1),
+  make_tuple(16, 16, &aom_sad16x16x4d_neon, -1),
+};
+INSTANTIATE_TEST_CASE_P(NEON, SADx4Test, ::testing::ValuesIn(x4d_neon_tests));
+#endif  // HAVE_NEON
+
+//------------------------------------------------------------------------------
+// x86 functions
+#if HAVE_SSE2
+const SadMxNParam sse2_tests[] = {
+  make_tuple(128, 128, &aom_sad128x128_sse2, -1),
+  make_tuple(128, 64, &aom_sad128x64_sse2, -1),
+  make_tuple(64, 128, &aom_sad64x128_sse2, -1),
+  make_tuple(64, 64, &aom_sad64x64_sse2, -1),
+  make_tuple(64, 32, &aom_sad64x32_sse2, -1),
+  make_tuple(32, 64, &aom_sad32x64_sse2, -1),
+  make_tuple(32, 32, &aom_sad32x32_sse2, -1),
+  make_tuple(32, 16, &aom_sad32x16_sse2, -1),
+  make_tuple(16, 32, &aom_sad16x32_sse2, -1),
+  make_tuple(16, 16, &aom_sad16x16_sse2, -1),
+  make_tuple(16, 8, &aom_sad16x8_sse2, -1),
+  make_tuple(8, 16, &aom_sad8x16_sse2, -1),
+  make_tuple(8, 8, &aom_sad8x8_sse2, -1),
+  make_tuple(8, 4, &aom_sad8x4_sse2, -1),
+  make_tuple(4, 8, &aom_sad4x8_sse2, -1),
+  make_tuple(4, 4, &aom_sad4x4_sse2, -1),
+  make_tuple(64, 64, &aom_highbd_sad64x64_sse2, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32_sse2, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64_sse2, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32_sse2, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16_sse2, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32_sse2, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16_sse2, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8_sse2, 8),
+  make_tuple(8, 16, &aom_highbd_sad8x16_sse2, 8),
+  make_tuple(8, 8, &aom_highbd_sad8x8_sse2, 8),
+  make_tuple(8, 4, &aom_highbd_sad8x4_sse2, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64_sse2, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32_sse2, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64_sse2, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32_sse2, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16_sse2, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32_sse2, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16_sse2, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8_sse2, 10),
+  make_tuple(8, 16, &aom_highbd_sad8x16_sse2, 10),
+  make_tuple(8, 8, &aom_highbd_sad8x8_sse2, 10),
+  make_tuple(8, 4, &aom_highbd_sad8x4_sse2, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64_sse2, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32_sse2, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64_sse2, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32_sse2, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16_sse2, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32_sse2, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16_sse2, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8_sse2, 12),
+  make_tuple(8, 16, &aom_highbd_sad8x16_sse2, 12),
+  make_tuple(8, 8, &aom_highbd_sad8x8_sse2, 12),
+  make_tuple(8, 4, &aom_highbd_sad8x4_sse2, 12),
+};
+INSTANTIATE_TEST_CASE_P(SSE2, SADTest, ::testing::ValuesIn(sse2_tests));
+
+const SadMxNAvgParam avg_sse2_tests[] = {
+  make_tuple(128, 128, &aom_sad128x128_avg_sse2, -1),
+  make_tuple(128, 64, &aom_sad128x64_avg_sse2, -1),
+  make_tuple(64, 128, &aom_sad64x128_avg_sse2, -1),
+  make_tuple(64, 64, &aom_sad64x64_avg_sse2, -1),
+  make_tuple(64, 32, &aom_sad64x32_avg_sse2, -1),
+  make_tuple(32, 64, &aom_sad32x64_avg_sse2, -1),
+  make_tuple(32, 32, &aom_sad32x32_avg_sse2, -1),
+  make_tuple(32, 16, &aom_sad32x16_avg_sse2, -1),
+  make_tuple(16, 32, &aom_sad16x32_avg_sse2, -1),
+  make_tuple(16, 16, &aom_sad16x16_avg_sse2, -1),
+  make_tuple(16, 8, &aom_sad16x8_avg_sse2, -1),
+  make_tuple(8, 16, &aom_sad8x16_avg_sse2, -1),
+  make_tuple(8, 8, &aom_sad8x8_avg_sse2, -1),
+  make_tuple(8, 4, &aom_sad8x4_avg_sse2, -1),
+  make_tuple(4, 8, &aom_sad4x8_avg_sse2, -1),
+  make_tuple(4, 4, &aom_sad4x4_avg_sse2, -1),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_sse2, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_sse2, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_sse2, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_sse2, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_sse2, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_sse2, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_sse2, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_sse2, 8),
+  make_tuple(8, 16, &aom_highbd_sad8x16_avg_sse2, 8),
+  make_tuple(8, 8, &aom_highbd_sad8x8_avg_sse2, 8),
+  make_tuple(8, 4, &aom_highbd_sad8x4_avg_sse2, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_sse2, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_sse2, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_sse2, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_sse2, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_sse2, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_sse2, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_sse2, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_sse2, 10),
+  make_tuple(8, 16, &aom_highbd_sad8x16_avg_sse2, 10),
+  make_tuple(8, 8, &aom_highbd_sad8x8_avg_sse2, 10),
+  make_tuple(8, 4, &aom_highbd_sad8x4_avg_sse2, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_sse2, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_sse2, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_sse2, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_sse2, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_sse2, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_sse2, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_sse2, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_sse2, 12),
+  make_tuple(8, 16, &aom_highbd_sad8x16_avg_sse2, 12),
+  make_tuple(8, 8, &aom_highbd_sad8x8_avg_sse2, 12),
+  make_tuple(8, 4, &aom_highbd_sad8x4_avg_sse2, 12),
+};
+INSTANTIATE_TEST_CASE_P(SSE2, SADavgTest, ::testing::ValuesIn(avg_sse2_tests));
+
+const SadMxNx4Param x4d_sse2_tests[] = {
+  make_tuple(128, 128, &aom_sad128x128x4d_sse2, -1),
+  make_tuple(128, 64, &aom_sad128x64x4d_sse2, -1),
+  make_tuple(64, 128, &aom_sad64x128x4d_sse2, -1),
+  make_tuple(64, 64, &aom_sad64x64x4d_sse2, -1),
+  make_tuple(64, 32, &aom_sad64x32x4d_sse2, -1),
+  make_tuple(32, 64, &aom_sad32x64x4d_sse2, -1),
+  make_tuple(32, 32, &aom_sad32x32x4d_sse2, -1),
+  make_tuple(32, 16, &aom_sad32x16x4d_sse2, -1),
+  make_tuple(16, 32, &aom_sad16x32x4d_sse2, -1),
+  make_tuple(16, 16, &aom_sad16x16x4d_sse2, -1),
+  make_tuple(16, 8, &aom_sad16x8x4d_sse2, -1),
+  make_tuple(8, 16, &aom_sad8x16x4d_sse2, -1),
+  make_tuple(8, 8, &aom_sad8x8x4d_sse2, -1),
+  make_tuple(8, 4, &aom_sad8x4x4d_sse2, -1),
+  make_tuple(4, 8, &aom_sad4x8x4d_sse2, -1),
+  make_tuple(4, 4, &aom_sad4x4x4d_sse2, -1),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_sse2, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_sse2, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_sse2, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_sse2, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_sse2, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_sse2, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_sse2, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_sse2, 8),
+  make_tuple(8, 16, &aom_highbd_sad8x16x4d_sse2, 8),
+  make_tuple(8, 8, &aom_highbd_sad8x8x4d_sse2, 8),
+  make_tuple(8, 4, &aom_highbd_sad8x4x4d_sse2, 8),
+  make_tuple(4, 8, &aom_highbd_sad4x8x4d_sse2, 8),
+  make_tuple(4, 4, &aom_highbd_sad4x4x4d_sse2, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_sse2, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_sse2, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_sse2, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_sse2, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_sse2, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_sse2, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_sse2, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_sse2, 10),
+  make_tuple(8, 16, &aom_highbd_sad8x16x4d_sse2, 10),
+  make_tuple(8, 8, &aom_highbd_sad8x8x4d_sse2, 10),
+  make_tuple(8, 4, &aom_highbd_sad8x4x4d_sse2, 10),
+  make_tuple(4, 8, &aom_highbd_sad4x8x4d_sse2, 10),
+  make_tuple(4, 4, &aom_highbd_sad4x4x4d_sse2, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_sse2, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_sse2, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_sse2, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_sse2, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_sse2, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_sse2, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_sse2, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_sse2, 12),
+  make_tuple(8, 16, &aom_highbd_sad8x16x4d_sse2, 12),
+  make_tuple(8, 8, &aom_highbd_sad8x8x4d_sse2, 12),
+  make_tuple(8, 4, &aom_highbd_sad8x4x4d_sse2, 12),
+  make_tuple(4, 8, &aom_highbd_sad4x8x4d_sse2, 12),
+  make_tuple(4, 4, &aom_highbd_sad4x4x4d_sse2, 12),
+};
+INSTANTIATE_TEST_CASE_P(SSE2, SADx4Test, ::testing::ValuesIn(x4d_sse2_tests));
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3
+// Note: These are named sse2, but part of ssse3 file and only built and linked
+// when ssse3 is enabled.
+const JntSadMxhParam jnt_sad_sse2_tests[] = {
+  make_tuple(4, 4, &aom_sad4xh_sse2, -1),
+  make_tuple(4, 8, &aom_sad4xh_sse2, -1),
+  make_tuple(8, 4, &aom_sad8xh_sse2, -1),
+  make_tuple(8, 8, &aom_sad8xh_sse2, -1),
+  make_tuple(8, 16, &aom_sad8xh_sse2, -1),
+  make_tuple(16, 8, &aom_sad16xh_sse2, -1),
+  make_tuple(16, 16, &aom_sad16xh_sse2, -1),
+  make_tuple(16, 32, &aom_sad16xh_sse2, -1),
+  make_tuple(32, 16, &aom_sad32xh_sse2, -1),
+  make_tuple(32, 32, &aom_sad32xh_sse2, -1),
+  make_tuple(32, 64, &aom_sad32xh_sse2, -1),
+  make_tuple(64, 32, &aom_sad64xh_sse2, -1),
+  make_tuple(64, 64, &aom_sad64xh_sse2, -1),
+  make_tuple(128, 128, &aom_sad128xh_sse2, -1),
+  make_tuple(128, 64, &aom_sad128xh_sse2, -1),
+  make_tuple(64, 128, &aom_sad64xh_sse2, -1),
+  make_tuple(4, 16, &aom_sad4xh_sse2, -1),
+  make_tuple(16, 4, &aom_sad16xh_sse2, -1),
+  make_tuple(8, 32, &aom_sad8xh_sse2, -1),
+  make_tuple(32, 8, &aom_sad32xh_sse2, -1),
+  make_tuple(16, 64, &aom_sad16xh_sse2, -1),
+  make_tuple(64, 16, &aom_sad64xh_sse2, -1),
+};
+INSTANTIATE_TEST_CASE_P(SSE2, JntSADTest,
+                        ::testing::ValuesIn(jnt_sad_sse2_tests));
+
+#endif  // HAVE_SSSE3
+
+#if HAVE_SSE3
+// Only functions are x3, which do not have tests.
+#endif  // HAVE_SSE3
+
+#if HAVE_SSSE3
+const JntCompAvgParam jnt_comp_avg_ssse3_tests[] = {
+  make_tuple(128, 128, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(128, 64, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(64, 128, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(64, 64, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(64, 32, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(32, 64, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(32, 32, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(32, 16, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(16, 32, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(16, 16, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(16, 8, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(8, 16, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(8, 8, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(8, 4, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(4, 8, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(4, 4, &aom_jnt_comp_avg_pred_ssse3, -1),
+  make_tuple(16, 16, &aom_jnt_comp_avg_pred_ssse3, -1),
+};
+
+INSTANTIATE_TEST_CASE_P(SSSE3, JntCompAvgTest,
+                        ::testing::ValuesIn(jnt_comp_avg_ssse3_tests));
+
+const JntSadMxNAvgParam jnt_avg_ssse3_tests[] = {
+  make_tuple(128, 128, &aom_jnt_sad128x128_avg_ssse3, -1),
+  make_tuple(128, 64, &aom_jnt_sad128x64_avg_ssse3, -1),
+  make_tuple(64, 128, &aom_jnt_sad64x128_avg_ssse3, -1),
+  make_tuple(64, 64, &aom_jnt_sad64x64_avg_ssse3, -1),
+  make_tuple(64, 32, &aom_jnt_sad64x32_avg_ssse3, -1),
+  make_tuple(32, 64, &aom_jnt_sad32x64_avg_ssse3, -1),
+  make_tuple(32, 32, &aom_jnt_sad32x32_avg_ssse3, -1),
+  make_tuple(32, 16, &aom_jnt_sad32x16_avg_ssse3, -1),
+  make_tuple(16, 32, &aom_jnt_sad16x32_avg_ssse3, -1),
+  make_tuple(16, 16, &aom_jnt_sad16x16_avg_ssse3, -1),
+  make_tuple(16, 8, &aom_jnt_sad16x8_avg_ssse3, -1),
+  make_tuple(8, 16, &aom_jnt_sad8x16_avg_ssse3, -1),
+  make_tuple(8, 8, &aom_jnt_sad8x8_avg_ssse3, -1),
+  make_tuple(8, 4, &aom_jnt_sad8x4_avg_ssse3, -1),
+  make_tuple(4, 8, &aom_jnt_sad4x8_avg_ssse3, -1),
+  make_tuple(4, 4, &aom_jnt_sad4x4_avg_ssse3, -1),
+};
+INSTANTIATE_TEST_CASE_P(SSSE3, JntSADavgTest,
+                        ::testing::ValuesIn(jnt_avg_ssse3_tests));
+#endif  // HAVE_SSSE3
+
+#if HAVE_SSE4_1
+// Only functions are x8, which do not have tests.
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+const SadMxNParam avx2_tests[] = {
+  make_tuple(64, 128, &aom_sad64x128_avx2, -1),
+  make_tuple(128, 64, &aom_sad128x64_avx2, -1),
+  make_tuple(128, 128, &aom_sad128x128_avx2, -1),
+  make_tuple(64, 64, &aom_sad64x64_avx2, -1),
+  make_tuple(64, 32, &aom_sad64x32_avx2, -1),
+  make_tuple(32, 64, &aom_sad32x64_avx2, -1),
+  make_tuple(32, 32, &aom_sad32x32_avx2, -1),
+  make_tuple(32, 16, &aom_sad32x16_avx2, -1),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avx2, 8),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avx2, 10),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avx2, 12),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avx2, 8),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avx2, 10),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avx2, 12),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avx2, 8),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avx2, 10),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avx2, 12),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avx2, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avx2, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avx2, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avx2, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avx2, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avx2, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avx2, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avx2, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avx2, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avx2, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avx2, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avx2, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avx2, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avx2, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avx2, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avx2, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avx2, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avx2, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avx2, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avx2, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avx2, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avx2, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avx2, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avx2, 12),
+};
+INSTANTIATE_TEST_CASE_P(AVX2, SADTest, ::testing::ValuesIn(avx2_tests));
+
+const SadMxNAvgParam avg_avx2_tests[] = {
+  make_tuple(64, 128, &aom_sad64x128_avg_avx2, -1),
+  make_tuple(128, 64, &aom_sad128x64_avg_avx2, -1),
+  make_tuple(128, 128, &aom_sad128x128_avg_avx2, -1),
+  make_tuple(64, 64, &aom_sad64x64_avg_avx2, -1),
+  make_tuple(64, 32, &aom_sad64x32_avg_avx2, -1),
+  make_tuple(32, 64, &aom_sad32x64_avg_avx2, -1),
+  make_tuple(32, 32, &aom_sad32x32_avg_avx2, -1),
+  make_tuple(32, 16, &aom_sad32x16_avg_avx2, -1),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avg_avx2, 8),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avg_avx2, 10),
+  make_tuple(128, 128, &aom_highbd_sad128x128_avg_avx2, 12),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avg_avx2, 8),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avg_avx2, 10),
+  make_tuple(128, 64, &aom_highbd_sad128x64_avg_avx2, 12),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avg_avx2, 8),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avg_avx2, 10),
+  make_tuple(64, 128, &aom_highbd_sad64x128_avg_avx2, 12),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_avx2, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_avx2, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64_avg_avx2, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_avx2, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_avx2, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32_avg_avx2, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_avx2, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_avx2, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64_avg_avx2, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_avx2, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_avx2, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32_avg_avx2, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_avx2, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_avx2, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16_avg_avx2, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_avx2, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_avx2, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32_avg_avx2, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_avx2, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_avx2, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16_avg_avx2, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_avx2, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_avx2, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8_avg_avx2, 12),
+};
+INSTANTIATE_TEST_CASE_P(AVX2, SADavgTest, ::testing::ValuesIn(avg_avx2_tests));
+
+const SadMxNx4Param x4d_avx2_tests[] = {
+  make_tuple(64, 128, &aom_sad64x128x4d_avx2, -1),
+  make_tuple(128, 64, &aom_sad128x64x4d_avx2, -1),
+  make_tuple(128, 128, &aom_sad128x128x4d_avx2, -1),
+  make_tuple(64, 64, &aom_sad64x64x4d_avx2, -1),
+  make_tuple(32, 64, &aom_sad32x64x4d_avx2, -1),
+  make_tuple(64, 32, &aom_sad64x32x4d_avx2, -1),
+  make_tuple(32, 32, &aom_sad32x32x4d_avx2, -1),
+  make_tuple(128, 128, &aom_highbd_sad128x128x4d_avx2, 8),
+  make_tuple(128, 128, &aom_highbd_sad128x128x4d_avx2, 10),
+  make_tuple(128, 128, &aom_highbd_sad128x128x4d_avx2, 12),
+  make_tuple(128, 64, &aom_highbd_sad128x64x4d_avx2, 8),
+  make_tuple(128, 64, &aom_highbd_sad128x64x4d_avx2, 10),
+  make_tuple(128, 64, &aom_highbd_sad128x64x4d_avx2, 12),
+  make_tuple(64, 128, &aom_highbd_sad64x128x4d_avx2, 8),
+  make_tuple(64, 128, &aom_highbd_sad64x128x4d_avx2, 10),
+  make_tuple(64, 128, &aom_highbd_sad64x128x4d_avx2, 12),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_avx2, 8),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_avx2, 10),
+  make_tuple(64, 64, &aom_highbd_sad64x64x4d_avx2, 12),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_avx2, 8),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_avx2, 10),
+  make_tuple(64, 32, &aom_highbd_sad64x32x4d_avx2, 12),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_avx2, 8),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_avx2, 10),
+  make_tuple(32, 64, &aom_highbd_sad32x64x4d_avx2, 12),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_avx2, 8),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_avx2, 10),
+  make_tuple(32, 32, &aom_highbd_sad32x32x4d_avx2, 12),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_avx2, 8),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_avx2, 10),
+  make_tuple(32, 16, &aom_highbd_sad32x16x4d_avx2, 12),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_avx2, 8),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_avx2, 10),
+  make_tuple(16, 32, &aom_highbd_sad16x32x4d_avx2, 12),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_avx2, 8),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_avx2, 10),
+  make_tuple(16, 16, &aom_highbd_sad16x16x4d_avx2, 12),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_avx2, 8),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_avx2, 10),
+  make_tuple(16, 8, &aom_highbd_sad16x8x4d_avx2, 12),
+};
+INSTANTIATE_TEST_CASE_P(AVX2, SADx4Test, ::testing::ValuesIn(x4d_avx2_tests));
+#endif  // HAVE_AVX2
+
+//------------------------------------------------------------------------------
+// MIPS functions
+#if HAVE_MSA
+const SadMxNParam msa_tests[] = {
+  make_tuple(64, 64, &aom_sad64x64_msa, -1),
+  make_tuple(64, 32, &aom_sad64x32_msa, -1),
+  make_tuple(32, 64, &aom_sad32x64_msa, -1),
+  make_tuple(32, 32, &aom_sad32x32_msa, -1),
+  make_tuple(32, 16, &aom_sad32x16_msa, -1),
+  make_tuple(16, 32, &aom_sad16x32_msa, -1),
+  make_tuple(16, 16, &aom_sad16x16_msa, -1),
+  make_tuple(16, 8, &aom_sad16x8_msa, -1),
+  make_tuple(8, 16, &aom_sad8x16_msa, -1),
+  make_tuple(8, 8, &aom_sad8x8_msa, -1),
+  make_tuple(8, 4, &aom_sad8x4_msa, -1),
+  make_tuple(4, 8, &aom_sad4x8_msa, -1),
+  make_tuple(4, 4, &aom_sad4x4_msa, -1),
+};
+INSTANTIATE_TEST_CASE_P(MSA, SADTest, ::testing::ValuesIn(msa_tests));
+
+const SadMxNAvgParam avg_msa_tests[] = {
+  make_tuple(64, 64, &aom_sad64x64_avg_msa, -1),
+  make_tuple(64, 32, &aom_sad64x32_avg_msa, -1),
+  make_tuple(32, 64, &aom_sad32x64_avg_msa, -1),
+  make_tuple(32, 32, &aom_sad32x32_avg_msa, -1),
+  make_tuple(32, 16, &aom_sad32x16_avg_msa, -1),
+  make_tuple(16, 32, &aom_sad16x32_avg_msa, -1),
+  make_tuple(16, 16, &aom_sad16x16_avg_msa, -1),
+  make_tuple(16, 8, &aom_sad16x8_avg_msa, -1),
+  make_tuple(8, 16, &aom_sad8x16_avg_msa, -1),
+  make_tuple(8, 8, &aom_sad8x8_avg_msa, -1),
+  make_tuple(8, 4, &aom_sad8x4_avg_msa, -1),
+  make_tuple(4, 8, &aom_sad4x8_avg_msa, -1),
+  make_tuple(4, 4, &aom_sad4x4_avg_msa, -1),
+};
+INSTANTIATE_TEST_CASE_P(MSA, SADavgTest, ::testing::ValuesIn(avg_msa_tests));
+
+const SadMxNx4Param x4d_msa_tests[] = {
+  make_tuple(64, 64, &aom_sad64x64x4d_msa, -1),
+  make_tuple(64, 32, &aom_sad64x32x4d_msa, -1),
+  make_tuple(32, 64, &aom_sad32x64x4d_msa, -1),
+  make_tuple(32, 32, &aom_sad32x32x4d_msa, -1),
+  make_tuple(32, 16, &aom_sad32x16x4d_msa, -1),
+  make_tuple(16, 32, &aom_sad16x32x4d_msa, -1),
+  make_tuple(16, 16, &aom_sad16x16x4d_msa, -1),
+  make_tuple(16, 8, &aom_sad16x8x4d_msa, -1),
+  make_tuple(8, 16, &aom_sad8x16x4d_msa, -1),
+  make_tuple(8, 8, &aom_sad8x8x4d_msa, -1),
+  make_tuple(8, 4, &aom_sad8x4x4d_msa, -1),
+  make_tuple(4, 8, &aom_sad4x8x4d_msa, -1),
+  make_tuple(4, 4, &aom_sad4x4x4d_msa, -1),
+};
+INSTANTIATE_TEST_CASE_P(MSA, SADx4Test, ::testing::ValuesIn(x4d_msa_tests));
+#endif  // HAVE_MSA
+
+}  // namespace
diff --git a/third_party/aom/test/scalability_test.cc b/third_party/aom/test/scalability_test.cc
new file mode 100644
index 0000000000..b399188617
--- /dev/null
+++ b/third_party/aom/test/scalability_test.cc
@@ -0,0 +1,81 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+const int kCpuUsed = 8;
+const int kBaseLayerQp = 55;
+const int kEnhancementLayerQp = 20;
+
+class ScalabilityTest
+    : public ::libaom_test::CodecTestWithParam<libaom_test::TestMode>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  ScalabilityTest() : EncoderTest(GET_PARAM(0)) {}
+  virtual ~ScalabilityTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(GET_PARAM(1));
+    num_spatial_layers_ = 2;
+  }
+
+  virtual void PreEncodeFrameHook(::libaom_test::VideoSource *video,
+                                  ::libaom_test::Encoder *encoder) {
+    if (video->frame() == 0) {
+      encoder->Control(AOME_SET_CPUUSED, kCpuUsed);
+      encoder->Control(AOME_SET_NUMBER_SPATIAL_LAYERS, num_spatial_layers_);
+    } else if (video->frame() % num_spatial_layers_) {
+      frame_flags_ = AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 |
+                     AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF |
+                     AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2 |
+                     AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF |
+                     AOM_EFLAG_NO_UPD_ARF | AOM_EFLAG_NO_UPD_ENTROPY;
+      encoder->Control(AOME_SET_SPATIAL_LAYER_ID, 1);
+      encoder->Control(AOME_SET_CQ_LEVEL, kEnhancementLayerQp);
+    } else {
+      frame_flags_ = AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 |
+                     AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF |
+                     AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2 |
+                     AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF |
+                     AOM_EFLAG_NO_UPD_ENTROPY;
+      encoder->Control(AOME_SET_SPATIAL_LAYER_ID, 0);
+      encoder->Control(AOME_SET_CQ_LEVEL, kBaseLayerQp);
+    }
+  }
+
+  void DoTest(int num_spatial_layers) {
+    num_spatial_layers_ = num_spatial_layers;
+    cfg_.rc_end_usage = AOM_Q;
+    cfg_.g_lag_in_frames = 0;
+
+    ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352,
+                                         288, 30, 1, 0, 18);
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  }
+
+  int num_spatial_layers_;
+};
+
+TEST_P(ScalabilityTest, TestNoMismatch2SpatialLayers) { DoTest(2); }
+
+TEST_P(ScalabilityTest, TestNoMismatch3SpatialLayers) { DoTest(3); }
+
+AV1_INSTANTIATE_TEST_CASE(ScalabilityTest,
+                          ::testing::Values(::libaom_test::kRealTime));
+
+}  // namespace
diff --git a/third_party/aom/test/scan_test.cc b/third_party/aom/test/scan_test.cc
new file mode 100644
index 0000000000..dee2ab5a69
--- /dev/null
+++ b/third_party/aom/test/scan_test.cc
@@ -0,0 +1,133 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "av1/common/scan.h"
+#include "av1/common/txb_common.h"
+#include "test/av1_txfm_test.h"
+
+static int scan_test(const int16_t *scan, const int16_t *iscan, int si, int r,
+                     int c, int w) {
+  if (iscan[r * w + c] != si || scan[si] != r * w + c) {
+    printf("r %d c %d ref_iscan %d iscan %d ref_scan %d scan %d\n", r, c, si,
+           iscan[r * w + c], r * w + c, scan[si]);
+    return 1;
+  } else {
+    return 0;
+  }
+}
+
+int scan_order_test(const SCAN_ORDER *scan_order, int w, int h,
+                    SCAN_MODE mode) {
+  const int16_t *scan = scan_order->scan;
+  const int16_t *iscan = scan_order->iscan;
+  int dim = w + h - 1;
+  if (mode == SCAN_MODE_ZIG_ZAG) {
+    int si = 0;
+    for (int i = 0; i < dim; ++i) {
+      if (i % 2 == 0) {
+        for (int c = 0; c < w; ++c) {
+          int r = i - c;
+          if (r >= 0 && r < h) {
+            if (scan_test(scan, iscan, si, r, c, w)) return 1;
+            ++si;
+          }
+        }
+      } else {
+        for (int r = 0; r < h; ++r) {
+          int c = i - r;
+          if (c >= 0 && c < w) {
+            if (scan_test(scan, iscan, si, r, c, w)) return 1;
+            ++si;
+          }
+        }
+      }
+    }
+  } else if (mode == SCAN_MODE_COL_DIAG) {
+    int si = 0;
+    for (int i = 0; i < dim; ++i) {
+      for (int c = 0; c < w; ++c) {
+        int r = i - c;
+        if (r >= 0 && r < h) {
+          if (scan_test(scan, iscan, si, r, c, w)) return 1;
+          ++si;
+        }
+      }
+    }
+  } else if (mode == SCAN_MODE_ROW_DIAG) {
+    int si = 0;
+    for (int i = 0; i < dim; ++i) {
+      for (int r = 0; r < h; ++r) {
+        int c = i - r;
+        if (c >= 0 && c < w) {
+          if (scan_test(scan, iscan, si, r, c, w)) return 1;
+          ++si;
+        }
+      }
+    }
+  } else if (mode == SCAN_MODE_ROW_1D) {
+    int si = 0;
+    for (int r = 0; r < h; ++r) {
+      for (int c = 0; c < w; ++c) {
+        if (scan_test(scan, iscan, si, r, c, w)) return 1;
+        ++si;
+      }
+    }
+  } else {
+    assert(mode == SCAN_MODE_COL_1D);
+    int si = 0;
+    for (int c = 0; c < w; ++c) {
+      for (int r = 0; r < h; ++r) {
+        if (scan_test(scan, iscan, si, r, c, w)) return 1;
+        ++si;
+      }
+    }
+  }
+  return 0;
+}
+
+TEST(Av1ScanTest, Dependency) {
+  for (int tx_size = TX_4X4; tx_size < TX_SIZES_ALL; ++tx_size) {
+    const int org_rows = tx_size_high[(TX_SIZE)tx_size];
+    const int org_cols = tx_size_wide[(TX_SIZE)tx_size];
+    const int rows = get_txb_high((TX_SIZE)tx_size);
+    const int cols = get_txb_wide((TX_SIZE)tx_size);
+    for (int tx_type = 0; tx_type < TX_TYPES; ++tx_type) {
+      if (libaom_test::IsTxSizeTypeValid(static_cast<TX_SIZE>(tx_size),
+                                         static_cast<TX_TYPE>(tx_type)) ==
+          false) {
+        continue;
+      }
+      SCAN_MODE scan_mode;
+      TX_CLASS tx_class = tx_type_to_class[(TX_TYPE)tx_type];
+      if (tx_class == TX_CLASS_2D) {
+        if (rows == cols) {
+          scan_mode = SCAN_MODE_ZIG_ZAG;
+        } else if (rows > cols) {
+          scan_mode = SCAN_MODE_ROW_DIAG;
+        } else {
+          scan_mode = SCAN_MODE_COL_DIAG;
+        }
+      } else if (tx_class == TX_CLASS_VERT) {
+        scan_mode = SCAN_MODE_ROW_1D;
+      } else {
+        assert(tx_class == TX_CLASS_HORIZ);
+        scan_mode = SCAN_MODE_COL_1D;
+      }
+      const SCAN_ORDER *scan_order =
+          get_default_scan((TX_SIZE)tx_size, (TX_TYPE)tx_type);
+      ASSERT_EQ(scan_order_test(scan_order, cols, rows, scan_mode), 0)
+          << "scan mismatch tx_class " << tx_class << " tx_type " << tx_type
+          << " tx_w " << org_cols << " tx_h " << org_rows << " scan_mode "
+          << scan_mode << "\n";
+    }
+  }
+}
diff --git a/third_party/aom/test/segment_binarization_sync.cc b/third_party/aom/test/segment_binarization_sync.cc
new file mode 100644
index 0000000000..bd8cf11410
--- /dev/null
+++ b/third_party/aom/test/segment_binarization_sync.cc
@@ -0,0 +1,61 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+
+using libaom_test::ACMRandom;
+
+extern "C" {
+int av1_neg_interleave(int x, int ref, int max);
+int av1_neg_deinterleave(int diff, int ref, int max);
+}
+
+namespace {
+
+struct Segment {
+  int id;
+  int pred;
+  int last_id;
+};
+
+Segment GenerateSegment(int seed) {
+  static const int MAX_SEGMENTS = 8;
+
+  ACMRandom rnd_(seed);
+
+  Segment segment;
+  const int last_segid = rnd_.PseudoUniform(MAX_SEGMENTS);
+  segment.last_id = last_segid;
+  segment.pred = rnd_.PseudoUniform(MAX_SEGMENTS);
+  segment.id = rnd_.PseudoUniform(last_segid + 1);
+
+  return segment;
+}
+
+// Try to reveal a mismatch between segment binarization and debinarization
+TEST(SegmentBinarizationSync, SearchForBinarizationMismatch) {
+  const int count_tests = 1000;
+  const int seed_init = 4321;
+
+  for (int i = 0; i < count_tests; ++i) {
+    const Segment seg = GenerateSegment(seed_init + i);
+
+    const int max_segid = seg.last_id + 1;
+    const int seg_diff = av1_neg_interleave(seg.id, seg.pred, max_segid);
+    const int decoded_segid =
+        av1_neg_deinterleave(seg_diff, seg.pred, max_segid);
+
+    ASSERT_EQ(decoded_segid, seg.id);
+  }
+}
+
+}  // namespace
diff --git a/third_party/aom/test/selfguided_filter_test.cc b/third_party/aom/test/selfguided_filter_test.cc
new file mode 100644
index 0000000000..d2d5c6105d
--- /dev/null
+++ b/third_party/aom/test/selfguided_filter_test.cc
@@ -0,0 +1,410 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <ctime>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/av1_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+
+#include "aom_ports/aom_timer.h"
+#include "av1/common/mv.h"
+#include "av1/common/restoration.h"
+
+namespace {
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+
+typedef void (*SgrFunc)(const uint8_t *dat8, int width, int height, int stride,
+                        int eps, const int *xqd, uint8_t *dst8, int dst_stride,
+                        int32_t *tmpbuf, int bit_depth, int highbd);
+
+// Test parameter list:
+//  <tst_fun_>
+typedef tuple<SgrFunc> FilterTestParam;
+
+class AV1SelfguidedFilterTest
+    : public ::testing::TestWithParam<FilterTestParam> {
+ public:
+  virtual ~AV1SelfguidedFilterTest() {}
+  virtual void SetUp() {}
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void RunSpeedTest() {
+    tst_fun_ = GET_PARAM(0);
+    const int pu_width = RESTORATION_PROC_UNIT_SIZE;
+    const int pu_height = RESTORATION_PROC_UNIT_SIZE;
+    const int width = 256, height = 256, stride = 288, out_stride = 288;
+    const int NUM_ITERS = 2000;
+    int i, j, k;
+
+    uint8_t *input_ =
+        (uint8_t *)aom_memalign(32, stride * (height + 32) * sizeof(uint8_t));
+    uint8_t *output_ = (uint8_t *)aom_memalign(
+        32, out_stride * (height + 32) * sizeof(uint8_t));
+    int32_t *tmpbuf = (int32_t *)aom_memalign(32, RESTORATION_TMPBUF_SIZE);
+    uint8_t *input = input_ + stride * 16 + 16;
+    uint8_t *output = output_ + out_stride * 16 + 16;
+
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+    for (i = -16; i < height + 16; ++i)
+      for (j = -16; j < width + 16; ++j)
+        input[i * stride + j] = rnd.Rand16() & 0xFF;
+
+    int xqd[2] = { SGRPROJ_PRJ_MIN0 + rnd.PseudoUniform(SGRPROJ_PRJ_MAX0 + 1 -
+                                                        SGRPROJ_PRJ_MIN0),
+                   SGRPROJ_PRJ_MIN1 + rnd.PseudoUniform(SGRPROJ_PRJ_MAX1 + 1 -
+                                                        SGRPROJ_PRJ_MIN1) };
+    // Fix a parameter set, since the speed depends slightly on r.
+    // Change this to test different combinations of values of r.
+    int eps = 15;
+
+    av1_loop_restoration_precal();
+
+    aom_usec_timer ref_timer;
+    aom_usec_timer_start(&ref_timer);
+    for (i = 0; i < NUM_ITERS; ++i) {
+      for (k = 0; k < height; k += pu_height)
+        for (j = 0; j < width; j += pu_width) {
+          int w = AOMMIN(pu_width, width - j);
+          int h = AOMMIN(pu_height, height - k);
+          uint8_t *input_p = input + k * stride + j;
+          uint8_t *output_p = output + k * out_stride + j;
+          apply_selfguided_restoration_c(input_p, w, h, stride, eps, xqd,
+                                         output_p, out_stride, tmpbuf, 8, 0);
+        }
+    }
+    aom_usec_timer_mark(&ref_timer);
+    const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);
+
+    aom_usec_timer tst_timer;
+    aom_usec_timer_start(&tst_timer);
+    for (i = 0; i < NUM_ITERS; ++i) {
+      for (k = 0; k < height; k += pu_height)
+        for (j = 0; j < width; j += pu_width) {
+          int w = AOMMIN(pu_width, width - j);
+          int h = AOMMIN(pu_height, height - k);
+          uint8_t *input_p = input + k * stride + j;
+          uint8_t *output_p = output + k * out_stride + j;
+          tst_fun_(input_p, w, h, stride, eps, xqd, output_p, out_stride,
+                   tmpbuf, 8, 0);
+        }
+    }
+    aom_usec_timer_mark(&tst_timer);
+    const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);
+
+    std::cout << "[          ] C time = " << ref_time / 1000
+              << " ms, SIMD time = " << tst_time / 1000 << " ms\n";
+
+    EXPECT_GT(ref_time, tst_time)
+        << "Error: AV1SelfguidedFilterTest.SpeedTest, SIMD slower than C.\n"
+        << "C time: " << ref_time << " us\n"
+        << "SIMD time: " << tst_time << " us\n";
+
+    aom_free(input_);
+    aom_free(output_);
+    aom_free(tmpbuf);
+  }
+
+  void RunCorrectnessTest() {
+    tst_fun_ = GET_PARAM(0);
+    const int pu_width = RESTORATION_PROC_UNIT_SIZE;
+    const int pu_height = RESTORATION_PROC_UNIT_SIZE;
+    // Set the maximum width/height to test here. We actually test a small
+    // range of sizes *up to* this size, so that we can check, eg.,
+    // the behaviour on tiles which are not a multiple of 4 wide.
+    const int max_w = 260, max_h = 260, stride = 672, out_stride = 672;
+    const int NUM_ITERS = 81;
+    int i, j, k;
+
+    uint8_t *input_ =
+        (uint8_t *)aom_memalign(32, stride * (max_h + 32) * sizeof(uint8_t));
+    uint8_t *output_ = (uint8_t *)aom_memalign(
+        32, out_stride * (max_h + 32) * sizeof(uint8_t));
+    uint8_t *output2_ = (uint8_t *)aom_memalign(
+        32, out_stride * (max_h + 32) * sizeof(uint8_t));
+    int32_t *tmpbuf = (int32_t *)aom_memalign(32, RESTORATION_TMPBUF_SIZE);
+
+    uint8_t *input = input_ + stride * 16 + 16;
+    uint8_t *output = output_ + out_stride * 16 + 16;
+    uint8_t *output2 = output2_ + out_stride * 16 + 16;
+
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+    av1_loop_restoration_precal();
+
+    for (i = 0; i < NUM_ITERS; ++i) {
+      for (j = -16; j < max_h + 16; ++j)
+        for (k = -16; k < max_w + 16; ++k)
+          input[j * stride + k] = rnd.Rand16() & 0xFF;
+
+      int xqd[2] = { SGRPROJ_PRJ_MIN0 + rnd.PseudoUniform(SGRPROJ_PRJ_MAX0 + 1 -
+                                                          SGRPROJ_PRJ_MIN0),
+                     SGRPROJ_PRJ_MIN1 + rnd.PseudoUniform(SGRPROJ_PRJ_MAX1 + 1 -
+                                                          SGRPROJ_PRJ_MIN1) };
+      int eps = rnd.PseudoUniform(1 << SGRPROJ_PARAMS_BITS);
+
+      // Test various tile sizes around 256x256
+      int test_w = max_w - (i / 9);
+      int test_h = max_h - (i % 9);
+
+      for (k = 0; k < test_h; k += pu_height)
+        for (j = 0; j < test_w; j += pu_width) {
+          int w = AOMMIN(pu_width, test_w - j);
+          int h = AOMMIN(pu_height, test_h - k);
+          uint8_t *input_p = input + k * stride + j;
+          uint8_t *output_p = output + k * out_stride + j;
+          uint8_t *output2_p = output2 + k * out_stride + j;
+          tst_fun_(input_p, w, h, stride, eps, xqd, output_p, out_stride,
+                   tmpbuf, 8, 0);
+          apply_selfguided_restoration_c(input_p, w, h, stride, eps, xqd,
+                                         output2_p, out_stride, tmpbuf, 8, 0);
+        }
+
+      for (j = 0; j < test_h; ++j)
+        for (k = 0; k < test_w; ++k) {
+          ASSERT_EQ(output[j * out_stride + k], output2[j * out_stride + k]);
+        }
+    }
+
+    aom_free(input_);
+    aom_free(output_);
+    aom_free(output2_);
+    aom_free(tmpbuf);
+  }
+
+ private:
+  SgrFunc tst_fun_;
+};
+
+TEST_P(AV1SelfguidedFilterTest, DISABLED_SpeedTest) { RunSpeedTest(); }
+TEST_P(AV1SelfguidedFilterTest, CorrectnessTest) { RunCorrectnessTest(); }
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1SelfguidedFilterTest,
+                        ::testing::Values(apply_selfguided_restoration_sse4_1));
+#endif
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, AV1SelfguidedFilterTest,
+                        ::testing::Values(apply_selfguided_restoration_avx2));
+#endif
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, AV1SelfguidedFilterTest,
+                        ::testing::Values(apply_selfguided_restoration_neon));
+#endif
+
+// Test parameter list:
+//  <tst_fun_, bit_depth>
+typedef tuple<SgrFunc, int> HighbdFilterTestParam;
+
+class AV1HighbdSelfguidedFilterTest
+    : public ::testing::TestWithParam<HighbdFilterTestParam> {
+ public:
+  virtual ~AV1HighbdSelfguidedFilterTest() {}
+  virtual void SetUp() {}
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void RunSpeedTest() {
+    tst_fun_ = GET_PARAM(0);
+    const int pu_width = RESTORATION_PROC_UNIT_SIZE;
+    const int pu_height = RESTORATION_PROC_UNIT_SIZE;
+    const int width = 256, height = 256, stride = 288, out_stride = 288;
+    const int NUM_ITERS = 2000;
+    int i, j, k;
+    int bit_depth = GET_PARAM(1);
+    int mask = (1 << bit_depth) - 1;
+
+    uint16_t *input_ =
+        (uint16_t *)aom_memalign(32, stride * (height + 32) * sizeof(uint16_t));
+    uint16_t *output_ = (uint16_t *)aom_memalign(
+        32, out_stride * (height + 32) * sizeof(uint16_t));
+    int32_t *tmpbuf = (int32_t *)aom_memalign(32, RESTORATION_TMPBUF_SIZE);
+    uint16_t *input = input_ + stride * 16 + 16;
+    uint16_t *output = output_ + out_stride * 16 + 16;
+
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+    for (i = -16; i < height + 16; ++i)
+      for (j = -16; j < width + 16; ++j)
+        input[i * stride + j] = rnd.Rand16() & mask;
+
+    int xqd[2] = { SGRPROJ_PRJ_MIN0 + rnd.PseudoUniform(SGRPROJ_PRJ_MAX0 + 1 -
+                                                        SGRPROJ_PRJ_MIN0),
+                   SGRPROJ_PRJ_MIN1 + rnd.PseudoUniform(SGRPROJ_PRJ_MAX1 + 1 -
+                                                        SGRPROJ_PRJ_MIN1) };
+    // Fix a parameter set, since the speed depends slightly on r.
+    // Change this to test different combinations of values of r.
+    int eps = 15;
+
+    av1_loop_restoration_precal();
+
+    aom_usec_timer ref_timer;
+    aom_usec_timer_start(&ref_timer);
+    for (i = 0; i < NUM_ITERS; ++i) {
+      for (k = 0; k < height; k += pu_height)
+        for (j = 0; j < width; j += pu_width) {
+          int w = AOMMIN(pu_width, width - j);
+          int h = AOMMIN(pu_height, height - k);
+          uint16_t *input_p = input + k * stride + j;
+          uint16_t *output_p = output + k * out_stride + j;
+          apply_selfguided_restoration_c(
+              CONVERT_TO_BYTEPTR(input_p), w, h, stride, eps, xqd,
+              CONVERT_TO_BYTEPTR(output_p), out_stride, tmpbuf, bit_depth, 1);
+        }
+    }
+    aom_usec_timer_mark(&ref_timer);
+    const int64_t ref_time = aom_usec_timer_elapsed(&ref_timer);
+
+    aom_usec_timer tst_timer;
+    aom_usec_timer_start(&tst_timer);
+    for (i = 0; i < NUM_ITERS; ++i) {
+      for (k = 0; k < height; k += pu_height)
+        for (j = 0; j < width; j += pu_width) {
+          int w = AOMMIN(pu_width, width - j);
+          int h = AOMMIN(pu_height, height - k);
+          uint16_t *input_p = input + k * stride + j;
+          uint16_t *output_p = output + k * out_stride + j;
+          tst_fun_(CONVERT_TO_BYTEPTR(input_p), w, h, stride, eps, xqd,
+                   CONVERT_TO_BYTEPTR(output_p), out_stride, tmpbuf, bit_depth,
+                   1);
+        }
+    }
+    aom_usec_timer_mark(&tst_timer);
+    const int64_t tst_time = aom_usec_timer_elapsed(&tst_timer);
+
+    std::cout << "[          ] C time = " << ref_time / 1000
+              << " ms, SIMD time = " << tst_time / 1000 << " ms\n";
+
+    EXPECT_GT(ref_time, tst_time)
+        << "Error: AV1HighbdSelfguidedFilterTest.SpeedTest, SIMD slower than "
+           "C.\n"
+        << "C time: " << ref_time << " us\n"
+        << "SIMD time: " << tst_time << " us\n";
+
+    aom_free(input_);
+    aom_free(output_);
+    aom_free(tmpbuf);
+  }
+
+  void RunCorrectnessTest() {
+    tst_fun_ = GET_PARAM(0);
+    const int pu_width = RESTORATION_PROC_UNIT_SIZE;
+    const int pu_height = RESTORATION_PROC_UNIT_SIZE;
+    // Set the maximum width/height to test here. We actually test a small
+    // range of sizes *up to* this size, so that we can check, eg.,
+    // the behaviour on tiles which are not a multiple of 4 wide.
+    const int max_w = 260, max_h = 260, stride = 672, out_stride = 672;
+    const int NUM_ITERS = 81;
+    int i, j, k;
+    int bit_depth = GET_PARAM(1);
+    int mask = (1 << bit_depth) - 1;
+
+    uint16_t *input_ =
+        (uint16_t *)aom_memalign(32, stride * (max_h + 32) * sizeof(uint16_t));
+    uint16_t *output_ = (uint16_t *)aom_memalign(
+        32, out_stride * (max_h + 32) * sizeof(uint16_t));
+    uint16_t *output2_ = (uint16_t *)aom_memalign(
+        32, out_stride * (max_h + 32) * sizeof(uint16_t));
+    int32_t *tmpbuf = (int32_t *)aom_memalign(32, RESTORATION_TMPBUF_SIZE);
+
+    uint16_t *input = input_ + stride * 16 + 16;
+    uint16_t *output = output_ + out_stride * 16 + 16;
+    uint16_t *output2 = output2_ + out_stride * 16 + 16;
+
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+    av1_loop_restoration_precal();
+
+    for (i = 0; i < NUM_ITERS; ++i) {
+      for (j = -16; j < max_h + 16; ++j)
+        for (k = -16; k < max_w + 16; ++k)
+          input[j * stride + k] = rnd.Rand16() & mask;
+
+      int xqd[2] = { SGRPROJ_PRJ_MIN0 + rnd.PseudoUniform(SGRPROJ_PRJ_MAX0 + 1 -
+                                                          SGRPROJ_PRJ_MIN0),
+                     SGRPROJ_PRJ_MIN1 + rnd.PseudoUniform(SGRPROJ_PRJ_MAX1 + 1 -
+                                                          SGRPROJ_PRJ_MIN1) };
+      int eps = rnd.PseudoUniform(1 << SGRPROJ_PARAMS_BITS);
+
+      // Test various tile sizes around 256x256
+      int test_w = max_w - (i / 9);
+      int test_h = max_h - (i % 9);
+
+      for (k = 0; k < test_h; k += pu_height)
+        for (j = 0; j < test_w; j += pu_width) {
+          int w = AOMMIN(pu_width, test_w - j);
+          int h = AOMMIN(pu_height, test_h - k);
+          uint16_t *input_p = input + k * stride + j;
+          uint16_t *output_p = output + k * out_stride + j;
+          uint16_t *output2_p = output2 + k * out_stride + j;
+          tst_fun_(CONVERT_TO_BYTEPTR(input_p), w, h, stride, eps, xqd,
+                   CONVERT_TO_BYTEPTR(output_p), out_stride, tmpbuf, bit_depth,
+                   1);
+          apply_selfguided_restoration_c(
+              CONVERT_TO_BYTEPTR(input_p), w, h, stride, eps, xqd,
+              CONVERT_TO_BYTEPTR(output2_p), out_stride, tmpbuf, bit_depth, 1);
+        }
+
+      for (j = 0; j < test_h; ++j)
+        for (k = 0; k < test_w; ++k)
+          ASSERT_EQ(output[j * out_stride + k], output2[j * out_stride + k]);
+    }
+
+    aom_free(input_);
+    aom_free(output_);
+    aom_free(output2_);
+    aom_free(tmpbuf);
+  }
+
+ private:
+  SgrFunc tst_fun_;
+};
+
+TEST_P(AV1HighbdSelfguidedFilterTest, DISABLED_SpeedTest) { RunSpeedTest(); }
+TEST_P(AV1HighbdSelfguidedFilterTest, CorrectnessTest) { RunCorrectnessTest(); }
+
+#if HAVE_SSE4_1
+const int highbd_params_sse4_1[] = { 8, 10, 12 };
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AV1HighbdSelfguidedFilterTest,
+    ::testing::Combine(::testing::Values(apply_selfguided_restoration_sse4_1),
+                       ::testing::ValuesIn(highbd_params_sse4_1)));
+#endif
+
+#if HAVE_AVX2
+const int highbd_params_avx2[] = { 8, 10, 12 };
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AV1HighbdSelfguidedFilterTest,
+    ::testing::Combine(::testing::Values(apply_selfguided_restoration_avx2),
+                       ::testing::ValuesIn(highbd_params_avx2)));
+#endif
+#if HAVE_NEON
+const int highbd_params_neon[] = { 8, 10, 12 };
+INSTANTIATE_TEST_CASE_P(
+    NEON, AV1HighbdSelfguidedFilterTest,
+    ::testing::Combine(::testing::Values(apply_selfguided_restoration_neon),
+                       ::testing::ValuesIn(highbd_params_neon)));
+#endif
+}  // namespace
diff --git a/third_party/aom/test/set_maps.sh b/third_party/aom/test/set_maps.sh
new file mode 100755
index 0000000000..4f59b06d69
--- /dev/null
+++ b/third_party/aom/test/set_maps.sh
@@ -0,0 +1,52 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the libaom set_maps example. To add new tests to this file,
+## do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to set_maps_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: $YUV_RAW_INPUT is required, and set_maps must exist in
+# $LIBAOM_BIN_PATH.
+set_maps_verify_environment() {
+  if [ ! -e "${YUV_RAW_INPUT}" ]; then
+    echo "Libaom test data must exist in LIBAOM_TEST_DATA_PATH."
+    return 1
+  fi
+  if [ -z "$(aom_tool_path set_maps)" ]; then
+    elog "set_maps not found. It must exist in LIBAOM_BIN_PATH or its parent."
+    return 1
+  fi
+}
+
+# Runs set_maps using the codec specified by $1.
+set_maps() {
+  local encoder="$(aom_tool_path set_maps)"
+  local codec="$1"
+  local output_file="${AOM_TEST_OUTPUT_DIR}/set_maps_${codec}.ivf"
+
+  eval "${AOM_TEST_PREFIX}" "${encoder}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
+      "${YUV_RAW_INPUT_HEIGHT}" "${YUV_RAW_INPUT}" "${output_file}" \
+      ${devnull}
+
+  [ -e "${output_file}" ] || return 1
+}
+
+set_maps_av1() {
+  if [ "$(av1_encode_available)" = "yes" ]; then
+    set_maps av1 || return 1
+  fi
+}
+
+set_maps_tests="set_maps_av1"
+
+run_tests set_maps_verify_environment "${set_maps_tests}"
diff --git a/third_party/aom/test/simd_avx2_test.cc b/third_party/aom/test/simd_avx2_test.cc
new file mode 100644
index 0000000000..8a012bff88
--- /dev/null
+++ b/third_party/aom/test/simd_avx2_test.cc
@@ -0,0 +1,15 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#define ARCH AVX2
+#define ARCH_POSTFIX(name) name##_avx2
+#define SIMD_NAMESPACE simd_test_avx2
+#include "test/simd_impl.h"
diff --git a/third_party/aom/test/simd_cmp_avx2.cc b/third_party/aom/test/simd_cmp_avx2.cc
new file mode 100644
index 0000000000..cda632bcdf
--- /dev/null
+++ b/third_party/aom/test/simd_cmp_avx2.cc
@@ -0,0 +1,15 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#define ARCH AVX2
+#define ARCH_POSTFIX(name) name##_avx2
+#define SIMD_NAMESPACE simd_test_avx2
+#include "test/simd_cmp_impl.h"
diff --git a/third_party/aom/test/simd_cmp_impl.h b/third_party/aom/test/simd_cmp_impl.h
new file mode 100644
index 0000000000..b98af9aade
--- /dev/null
+++ b/third_party/aom/test/simd_cmp_impl.h
@@ -0,0 +1,2171 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <assert.h>
+#include <string>
+
+#include "config/aom_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "aom_dsp/aom_simd.h"
+#undef SIMD_INLINE
+#define SIMD_INLINE static  // Don't enforce inlining
+#include "aom_dsp/simd/v256_intrinsics_c.h"
+
+// Machine tuned code goes into this file. This file is included from
+// simd_cmp_sse2.cc, simd_cmp_ssse3.cc etc which define the macros
+// ARCH (=neon, sse2, ssse3, etc), SIMD_NAMESPACE and ARCH_POSTFIX().
+
+#ifdef _MSC_VER
+// Disable "value of intrinsic immediate argument 'value' is out of range
+// 'lowerbound - upperbound'" warning. Visual Studio emits this warning though
+// the parameters are conditionally checked in e.g., v256_shr_n_byte. Adding a
+// mask doesn't always appear to be sufficient.
+#pragma warning(disable : 4556)
+#endif
+
+using libaom_test::ACMRandom;
+
+namespace SIMD_NAMESPACE {
+
+// Wrap templates around intrinsics using immediate values
+template <int shift>
+v64 imm_v64_shl_n_byte(v64 a) {
+  return v64_shl_n_byte(a, shift);
+}
+template <int shift>
+v64 imm_v64_shr_n_byte(v64 a) {
+  return v64_shr_n_byte(a, shift);
+}
+template <int shift>
+v64 imm_v64_shl_n_8(v64 a) {
+  return v64_shl_n_8(a, shift);
+}
+template <int shift>
+v64 imm_v64_shr_n_u8(v64 a) {
+  return v64_shr_n_u8(a, shift);
+}
+template <int shift>
+v64 imm_v64_shr_n_s8(v64 a) {
+  return v64_shr_n_s8(a, shift);
+}
+template <int shift>
+v64 imm_v64_shl_n_16(v64 a) {
+  return v64_shl_n_16(a, shift);
+}
+template <int shift>
+v64 imm_v64_shr_n_u16(v64 a) {
+  return v64_shr_n_u16(a, shift);
+}
+template <int shift>
+v64 imm_v64_shr_n_s16(v64 a) {
+  return v64_shr_n_s16(a, shift);
+}
+template <int shift>
+v64 imm_v64_shl_n_32(v64 a) {
+  return v64_shl_n_32(a, shift);
+}
+template <int shift>
+v64 imm_v64_shr_n_u32(v64 a) {
+  return v64_shr_n_u32(a, shift);
+}
+template <int shift>
+v64 imm_v64_shr_n_s32(v64 a) {
+  return v64_shr_n_s32(a, shift);
+}
+template <int shift>
+v64 imm_v64_align(v64 a, v64 b) {
+  return v64_align(a, b, shift);
+}
+
+// Wrap templates around corresponding C implementations of the above
+template <int shift>
+c_v64 c_imm_v64_shl_n_byte(c_v64 a) {
+  return c_v64_shl_n_byte(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shr_n_byte(c_v64 a) {
+  return c_v64_shr_n_byte(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shl_n_8(c_v64 a) {
+  return c_v64_shl_n_8(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shr_n_u8(c_v64 a) {
+  return c_v64_shr_n_u8(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shr_n_s8(c_v64 a) {
+  return c_v64_shr_n_s8(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shl_n_16(c_v64 a) {
+  return c_v64_shl_n_16(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shr_n_u16(c_v64 a) {
+  return c_v64_shr_n_u16(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shr_n_s16(c_v64 a) {
+  return c_v64_shr_n_s16(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shl_n_32(c_v64 a) {
+  return c_v64_shl_n_32(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shr_n_u32(c_v64 a) {
+  return c_v64_shr_n_u32(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_shr_n_s32(c_v64 a) {
+  return c_v64_shr_n_s32(a, shift);
+}
+template <int shift>
+c_v64 c_imm_v64_align(c_v64 a, c_v64 b) {
+  return c_v64_align(a, b, shift);
+}
+
+template <int shift>
+v128 imm_v128_shl_n_byte(v128 a) {
+  return v128_shl_n_byte(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_byte(v128 a) {
+  return v128_shr_n_byte(a, shift);
+}
+template <int shift>
+v128 imm_v128_shl_n_8(v128 a) {
+  return v128_shl_n_8(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_u8(v128 a) {
+  return v128_shr_n_u8(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_s8(v128 a) {
+  return v128_shr_n_s8(a, shift);
+}
+template <int shift>
+v128 imm_v128_shl_n_16(v128 a) {
+  return v128_shl_n_16(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_u16(v128 a) {
+  return v128_shr_n_u16(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_s16(v128 a) {
+  return v128_shr_n_s16(a, shift);
+}
+template <int shift>
+v128 imm_v128_shl_n_32(v128 a) {
+  return v128_shl_n_32(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_u32(v128 a) {
+  return v128_shr_n_u32(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_s32(v128 a) {
+  return v128_shr_n_s32(a, shift);
+}
+template <int shift>
+v128 imm_v128_shl_n_64(v128 a) {
+  return v128_shl_n_64(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_u64(v128 a) {
+  return v128_shr_n_u64(a, shift);
+}
+template <int shift>
+v128 imm_v128_shr_n_s64(v128 a) {
+  return v128_shr_n_s64(a, shift);
+}
+template <int shift>
+v128 imm_v128_align(v128 a, v128 b) {
+  return v128_align(a, b, shift);
+}
+
+template <int shift>
+c_v128 c_imm_v128_shl_n_byte(c_v128 a) {
+  return c_v128_shl_n_byte(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_byte(c_v128 a) {
+  return c_v128_shr_n_byte(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shl_n_8(c_v128 a) {
+  return c_v128_shl_n_8(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_u8(c_v128 a) {
+  return c_v128_shr_n_u8(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_s8(c_v128 a) {
+  return c_v128_shr_n_s8(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shl_n_16(c_v128 a) {
+  return c_v128_shl_n_16(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_u16(c_v128 a) {
+  return c_v128_shr_n_u16(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_s16(c_v128 a) {
+  return c_v128_shr_n_s16(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shl_n_32(c_v128 a) {
+  return c_v128_shl_n_32(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_u32(c_v128 a) {
+  return c_v128_shr_n_u32(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_s32(c_v128 a) {
+  return c_v128_shr_n_s32(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shl_n_64(c_v128 a) {
+  return c_v128_shl_n_64(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_u64(c_v128 a) {
+  return c_v128_shr_n_u64(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_shr_n_s64(c_v128 a) {
+  return c_v128_shr_n_s64(a, shift);
+}
+template <int shift>
+c_v128 c_imm_v128_align(c_v128 a, c_v128 b) {
+  return c_v128_align(a, b, shift);
+}
+
+template <int shift>
+v256 imm_v256_shl_n_word(v256 a) {
+  return v256_shl_n_word(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_word(v256 a) {
+  return v256_shr_n_word(a, shift);
+}
+template <int shift>
+v256 imm_v256_shl_n_byte(v256 a) {
+  return v256_shl_n_byte(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_byte(v256 a) {
+  return v256_shr_n_byte(a, shift);
+}
+template <int shift>
+v256 imm_v256_shl_n_8(v256 a) {
+  return v256_shl_n_8(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_u8(v256 a) {
+  return v256_shr_n_u8(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_s8(v256 a) {
+  return v256_shr_n_s8(a, shift);
+}
+template <int shift>
+v256 imm_v256_shl_n_16(v256 a) {
+  return v256_shl_n_16(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_u16(v256 a) {
+  return v256_shr_n_u16(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_s16(v256 a) {
+  return v256_shr_n_s16(a, shift);
+}
+template <int shift>
+v256 imm_v256_shl_n_32(v256 a) {
+  return v256_shl_n_32(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_u32(v256 a) {
+  return v256_shr_n_u32(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_s32(v256 a) {
+  return v256_shr_n_s32(a, shift);
+}
+template <int shift>
+v256 imm_v256_shl_n_64(v256 a) {
+  return v256_shl_n_64(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_u64(v256 a) {
+  return v256_shr_n_u64(a, shift);
+}
+template <int shift>
+v256 imm_v256_shr_n_s64(v256 a) {
+  return v256_shr_n_s64(a, shift);
+}
+template <int shift>
+v256 imm_v256_align(v256 a, v256 b) {
+  return v256_align(a, b, shift);
+}
+
+template <int shift>
+c_v256 c_imm_v256_shl_n_word(c_v256 a) {
+  return c_v256_shl_n_word(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_word(c_v256 a) {
+  return c_v256_shr_n_word(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shl_n_byte(c_v256 a) {
+  return c_v256_shl_n_byte(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_byte(c_v256 a) {
+  return c_v256_shr_n_byte(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shl_n_8(c_v256 a) {
+  return c_v256_shl_n_8(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_u8(c_v256 a) {
+  return c_v256_shr_n_u8(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_s8(c_v256 a) {
+  return c_v256_shr_n_s8(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shl_n_16(c_v256 a) {
+  return c_v256_shl_n_16(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_u16(c_v256 a) {
+  return c_v256_shr_n_u16(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_s16(c_v256 a) {
+  return c_v256_shr_n_s16(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shl_n_32(c_v256 a) {
+  return c_v256_shl_n_32(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_u32(c_v256 a) {
+  return c_v256_shr_n_u32(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_s32(c_v256 a) {
+  return c_v256_shr_n_s32(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shl_n_64(c_v256 a) {
+  return c_v256_shl_n_64(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_u64(c_v256 a) {
+  return c_v256_shr_n_u64(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_shr_n_s64(c_v256 a) {
+  return c_v256_shr_n_s64(a, shift);
+}
+template <int shift>
+c_v256 c_imm_v256_align(c_v256 a, c_v256 b) {
+  return c_v256_align(a, b, shift);
+}
+
+// Wrappers around the the SAD and SSD functions
+uint32_t v64_sad_u8(v64 a, v64 b) {
+  return v64_sad_u8_sum(::v64_sad_u8(v64_sad_u8_init(), a, b));
+}
+uint32_t v64_ssd_u8(v64 a, v64 b) {
+  return v64_ssd_u8_sum(::v64_ssd_u8(v64_ssd_u8_init(), a, b));
+}
+
+uint32_t c_v64_sad_u8(c_v64 a, c_v64 b) {
+  return c_v64_sad_u8_sum(::c_v64_sad_u8(c_v64_sad_u8_init(), a, b));
+}
+uint32_t c_v64_ssd_u8(c_v64 a, c_v64 b) {
+  return c_v64_ssd_u8_sum(::c_v64_ssd_u8(c_v64_ssd_u8_init(), a, b));
+}
+uint32_t v128_sad_u8(v128 a, v128 b) {
+  return v128_sad_u8_sum(::v128_sad_u8(v128_sad_u8_init(), a, b));
+}
+uint32_t v128_ssd_u8(v128 a, v128 b) {
+  return v128_ssd_u8_sum(::v128_ssd_u8(v128_ssd_u8_init(), a, b));
+}
+uint32_t c_v128_sad_u8(c_v128 a, c_v128 b) {
+  return c_v128_sad_u8_sum(::c_v128_sad_u8(c_v128_sad_u8_init(), a, b));
+}
+uint32_t c_v128_ssd_u8(c_v128 a, c_v128 b) {
+  return c_v128_ssd_u8_sum(::c_v128_ssd_u8(c_v128_ssd_u8_init(), a, b));
+}
+uint32_t v128_sad_u16(v128 a, v128 b) {
+  return v128_sad_u16_sum(::v128_sad_u16(v128_sad_u16_init(), a, b));
+}
+uint64_t v128_ssd_s16(v128 a, v128 b) {
+  return v128_ssd_s16_sum(::v128_ssd_s16(v128_ssd_s16_init(), a, b));
+}
+uint32_t c_v128_sad_u16(c_v128 a, c_v128 b) {
+  return c_v128_sad_u16_sum(::c_v128_sad_u16(c_v128_sad_u16_init(), a, b));
+}
+uint64_t c_v128_ssd_s16(c_v128 a, c_v128 b) {
+  return c_v128_ssd_s16_sum(::c_v128_ssd_s16(c_v128_ssd_s16_init(), a, b));
+}
+uint32_t v256_sad_u8(v256 a, v256 b) {
+  return v256_sad_u8_sum(::v256_sad_u8(v256_sad_u8_init(), a, b));
+}
+uint32_t v256_ssd_u8(v256 a, v256 b) {
+  return v256_ssd_u8_sum(::v256_ssd_u8(v256_ssd_u8_init(), a, b));
+}
+uint32_t c_v256_sad_u8(c_v256 a, c_v256 b) {
+  return c_v256_sad_u8_sum(::c_v256_sad_u8(c_v256_sad_u8_init(), a, b));
+}
+uint32_t c_v256_ssd_u8(c_v256 a, c_v256 b) {
+  return c_v256_ssd_u8_sum(::c_v256_ssd_u8(c_v256_ssd_u8_init(), a, b));
+}
+uint32_t v256_sad_u16(v256 a, v256 b) {
+  return v256_sad_u16_sum(::v256_sad_u16(v256_sad_u16_init(), a, b));
+}
+uint64_t v256_ssd_s16(v256 a, v256 b) {
+  return v256_ssd_s16_sum(::v256_ssd_s16(v256_ssd_s16_init(), a, b));
+}
+uint32_t c_v256_sad_u16(c_v256 a, c_v256 b) {
+  return c_v256_sad_u16_sum(::c_v256_sad_u16(c_v256_sad_u16_init(), a, b));
+}
+uint64_t c_v256_ssd_s16(c_v256 a, c_v256 b) {
+  return c_v256_ssd_s16_sum(::c_v256_ssd_s16(c_v256_ssd_s16_init(), a, b));
+}
+
+namespace {
+
+typedef void (*fptr)();
+
+typedef struct {
+  const char *name;
+  fptr ref;
+  fptr simd;
+} mapping;
+
+#define MAP(name)                                \
+  {                                              \
+    #name, reinterpret_cast < fptr > (c_##name), \
+        reinterpret_cast < fptr > (name)         \
+  }
+
+const mapping m[] = { MAP(v64_sad_u8),
+                      MAP(v64_ssd_u8),
+                      MAP(v64_add_8),
+                      MAP(v64_add_16),
+                      MAP(v64_sadd_s8),
+                      MAP(v64_sadd_u8),
+                      MAP(v64_sadd_s16),
+                      MAP(v64_add_32),
+                      MAP(v64_sub_8),
+                      MAP(v64_ssub_u8),
+                      MAP(v64_ssub_s8),
+                      MAP(v64_sub_16),
+                      MAP(v64_ssub_s16),
+                      MAP(v64_ssub_u16),
+                      MAP(v64_sub_32),
+                      MAP(v64_ziplo_8),
+                      MAP(v64_ziphi_8),
+                      MAP(v64_ziplo_16),
+                      MAP(v64_ziphi_16),
+                      MAP(v64_ziplo_32),
+                      MAP(v64_ziphi_32),
+                      MAP(v64_pack_s32_u16),
+                      MAP(v64_pack_s32_s16),
+                      MAP(v64_pack_s16_u8),
+                      MAP(v64_pack_s16_s8),
+                      MAP(v64_unziphi_8),
+                      MAP(v64_unziplo_8),
+                      MAP(v64_unziphi_16),
+                      MAP(v64_unziplo_16),
+                      MAP(v64_or),
+                      MAP(v64_xor),
+                      MAP(v64_and),
+                      MAP(v64_andn),
+                      MAP(v64_mullo_s16),
+                      MAP(v64_mulhi_s16),
+                      MAP(v64_mullo_s32),
+                      MAP(v64_madd_s16),
+                      MAP(v64_madd_us8),
+                      MAP(v64_avg_u8),
+                      MAP(v64_rdavg_u8),
+                      MAP(v64_rdavg_u16),
+                      MAP(v64_avg_u16),
+                      MAP(v64_min_u8),
+                      MAP(v64_max_u8),
+                      MAP(v64_min_s8),
+                      MAP(v64_max_s8),
+                      MAP(v64_min_s16),
+                      MAP(v64_max_s16),
+                      MAP(v64_cmpgt_s8),
+                      MAP(v64_cmplt_s8),
+                      MAP(v64_cmpeq_8),
+                      MAP(v64_cmpgt_s16),
+                      MAP(v64_cmplt_s16),
+                      MAP(v64_cmpeq_16),
+                      MAP(v64_shuffle_8),
+                      MAP(imm_v64_align<1>),
+                      MAP(imm_v64_align<2>),
+                      MAP(imm_v64_align<3>),
+                      MAP(imm_v64_align<4>),
+                      MAP(imm_v64_align<5>),
+                      MAP(imm_v64_align<6>),
+                      MAP(imm_v64_align<7>),
+                      MAP(v64_abs_s8),
+                      MAP(v64_abs_s16),
+                      MAP(v64_unpacklo_u8_s16),
+                      MAP(v64_unpackhi_u8_s16),
+                      MAP(v64_unpacklo_s8_s16),
+                      MAP(v64_unpackhi_s8_s16),
+                      MAP(v64_unpacklo_u16_s32),
+                      MAP(v64_unpacklo_s16_s32),
+                      MAP(v64_unpackhi_u16_s32),
+                      MAP(v64_unpackhi_s16_s32),
+                      MAP(imm_v64_shr_n_byte<1>),
+                      MAP(imm_v64_shr_n_byte<2>),
+                      MAP(imm_v64_shr_n_byte<3>),
+                      MAP(imm_v64_shr_n_byte<4>),
+                      MAP(imm_v64_shr_n_byte<5>),
+                      MAP(imm_v64_shr_n_byte<6>),
+                      MAP(imm_v64_shr_n_byte<7>),
+                      MAP(imm_v64_shl_n_byte<1>),
+                      MAP(imm_v64_shl_n_byte<2>),
+                      MAP(imm_v64_shl_n_byte<3>),
+                      MAP(imm_v64_shl_n_byte<4>),
+                      MAP(imm_v64_shl_n_byte<5>),
+                      MAP(imm_v64_shl_n_byte<6>),
+                      MAP(imm_v64_shl_n_byte<7>),
+                      MAP(imm_v64_shl_n_8<1>),
+                      MAP(imm_v64_shl_n_8<2>),
+                      MAP(imm_v64_shl_n_8<3>),
+                      MAP(imm_v64_shl_n_8<4>),
+                      MAP(imm_v64_shl_n_8<5>),
+                      MAP(imm_v64_shl_n_8<6>),
+                      MAP(imm_v64_shl_n_8<7>),
+                      MAP(imm_v64_shr_n_u8<1>),
+                      MAP(imm_v64_shr_n_u8<2>),
+                      MAP(imm_v64_shr_n_u8<3>),
+                      MAP(imm_v64_shr_n_u8<4>),
+                      MAP(imm_v64_shr_n_u8<5>),
+                      MAP(imm_v64_shr_n_u8<6>),
+                      MAP(imm_v64_shr_n_u8<7>),
+                      MAP(imm_v64_shr_n_s8<1>),
+                      MAP(imm_v64_shr_n_s8<2>),
+                      MAP(imm_v64_shr_n_s8<3>),
+                      MAP(imm_v64_shr_n_s8<4>),
+                      MAP(imm_v64_shr_n_s8<5>),
+                      MAP(imm_v64_shr_n_s8<6>),
+                      MAP(imm_v64_shr_n_s8<7>),
+                      MAP(imm_v64_shl_n_16<1>),
+                      MAP(imm_v64_shl_n_16<2>),
+                      MAP(imm_v64_shl_n_16<4>),
+                      MAP(imm_v64_shl_n_16<6>),
+                      MAP(imm_v64_shl_n_16<8>),
+                      MAP(imm_v64_shl_n_16<10>),
+                      MAP(imm_v64_shl_n_16<12>),
+                      MAP(imm_v64_shl_n_16<14>),
+                      MAP(imm_v64_shr_n_u16<1>),
+                      MAP(imm_v64_shr_n_u16<2>),
+                      MAP(imm_v64_shr_n_u16<4>),
+                      MAP(imm_v64_shr_n_u16<6>),
+                      MAP(imm_v64_shr_n_u16<8>),
+                      MAP(imm_v64_shr_n_u16<10>),
+                      MAP(imm_v64_shr_n_u16<12>),
+                      MAP(imm_v64_shr_n_u16<14>),
+                      MAP(imm_v64_shr_n_s16<1>),
+                      MAP(imm_v64_shr_n_s16<2>),
+                      MAP(imm_v64_shr_n_s16<4>),
+                      MAP(imm_v64_shr_n_s16<6>),
+                      MAP(imm_v64_shr_n_s16<8>),
+                      MAP(imm_v64_shr_n_s16<10>),
+                      MAP(imm_v64_shr_n_s16<12>),
+                      MAP(imm_v64_shr_n_s16<14>),
+                      MAP(imm_v64_shl_n_32<1>),
+                      MAP(imm_v64_shl_n_32<4>),
+                      MAP(imm_v64_shl_n_32<8>),
+                      MAP(imm_v64_shl_n_32<12>),
+                      MAP(imm_v64_shl_n_32<16>),
+                      MAP(imm_v64_shl_n_32<20>),
+                      MAP(imm_v64_shl_n_32<24>),
+                      MAP(imm_v64_shl_n_32<28>),
+                      MAP(imm_v64_shr_n_u32<1>),
+                      MAP(imm_v64_shr_n_u32<4>),
+                      MAP(imm_v64_shr_n_u32<8>),
+                      MAP(imm_v64_shr_n_u32<12>),
+                      MAP(imm_v64_shr_n_u32<16>),
+                      MAP(imm_v64_shr_n_u32<20>),
+                      MAP(imm_v64_shr_n_u32<24>),
+                      MAP(imm_v64_shr_n_u32<28>),
+                      MAP(imm_v64_shr_n_s32<1>),
+                      MAP(imm_v64_shr_n_s32<4>),
+                      MAP(imm_v64_shr_n_s32<8>),
+                      MAP(imm_v64_shr_n_s32<12>),
+                      MAP(imm_v64_shr_n_s32<16>),
+                      MAP(imm_v64_shr_n_s32<20>),
+                      MAP(imm_v64_shr_n_s32<24>),
+                      MAP(imm_v64_shr_n_s32<28>),
+                      MAP(v64_shl_8),
+                      MAP(v64_shr_u8),
+                      MAP(v64_shr_s8),
+                      MAP(v64_shl_16),
+                      MAP(v64_shr_u16),
+                      MAP(v64_shr_s16),
+                      MAP(v64_shl_32),
+                      MAP(v64_shr_u32),
+                      MAP(v64_shr_s32),
+                      MAP(v64_hadd_u8),
+                      MAP(v64_hadd_s16),
+                      MAP(v64_dotp_s16),
+                      MAP(v64_dotp_su8),
+                      MAP(v64_u64),
+                      MAP(v64_low_u32),
+                      MAP(v64_high_u32),
+                      MAP(v64_low_s32),
+                      MAP(v64_high_s32),
+                      MAP(v64_dup_8),
+                      MAP(v64_dup_16),
+                      MAP(v64_dup_32),
+                      MAP(v64_from_32),
+                      MAP(v64_zero),
+                      MAP(v64_from_16),
+                      MAP(v128_sad_u8),
+                      MAP(v128_ssd_u8),
+                      MAP(v128_sad_u16),
+                      MAP(v128_ssd_s16),
+                      MAP(v128_add_8),
+                      MAP(v128_add_16),
+                      MAP(v128_sadd_s8),
+                      MAP(v128_sadd_u8),
+                      MAP(v128_sadd_s16),
+                      MAP(v128_add_32),
+                      MAP(v128_add_64),
+                      MAP(v128_sub_8),
+                      MAP(v128_ssub_u8),
+                      MAP(v128_ssub_s8),
+                      MAP(v128_sub_16),
+                      MAP(v128_ssub_s16),
+                      MAP(v128_ssub_u16),
+                      MAP(v128_sub_32),
+                      MAP(v128_sub_64),
+                      MAP(v128_ziplo_8),
+                      MAP(v128_ziphi_8),
+                      MAP(v128_ziplo_16),
+                      MAP(v128_ziphi_16),
+                      MAP(v128_ziplo_32),
+                      MAP(v128_ziphi_32),
+                      MAP(v128_ziplo_64),
+                      MAP(v128_ziphi_64),
+                      MAP(v128_unziphi_8),
+                      MAP(v128_unziplo_8),
+                      MAP(v128_unziphi_16),
+                      MAP(v128_unziplo_16),
+                      MAP(v128_unziphi_32),
+                      MAP(v128_unziplo_32),
+                      MAP(v128_pack_s32_u16),
+                      MAP(v128_pack_s32_s16),
+                      MAP(v128_pack_s16_u8),
+                      MAP(v128_pack_s16_s8),
+                      MAP(v128_or),
+                      MAP(v128_xor),
+                      MAP(v128_and),
+                      MAP(v128_andn),
+                      MAP(v128_mullo_s16),
+                      MAP(v128_mulhi_s16),
+                      MAP(v128_mullo_s32),
+                      MAP(v128_madd_s16),
+                      MAP(v128_madd_us8),
+                      MAP(v128_avg_u8),
+                      MAP(v128_rdavg_u8),
+                      MAP(v128_rdavg_u16),
+                      MAP(v128_avg_u16),
+                      MAP(v128_min_u8),
+                      MAP(v128_max_u8),
+                      MAP(v128_min_s8),
+                      MAP(v128_max_s8),
+                      MAP(v128_min_s16),
+                      MAP(v128_max_s16),
+                      MAP(v128_min_s32),
+                      MAP(v128_max_s32),
+                      MAP(v128_cmpgt_s8),
+                      MAP(v128_cmplt_s8),
+                      MAP(v128_cmpeq_8),
+                      MAP(v128_cmpgt_s16),
+                      MAP(v128_cmpeq_16),
+                      MAP(v128_cmplt_s16),
+                      MAP(v128_cmpgt_s32),
+                      MAP(v128_cmpeq_32),
+                      MAP(v128_cmplt_s32),
+                      MAP(v128_shuffle_8),
+                      MAP(imm_v128_align<1>),
+                      MAP(imm_v128_align<2>),
+                      MAP(imm_v128_align<3>),
+                      MAP(imm_v128_align<4>),
+                      MAP(imm_v128_align<5>),
+                      MAP(imm_v128_align<6>),
+                      MAP(imm_v128_align<7>),
+                      MAP(imm_v128_align<8>),
+                      MAP(imm_v128_align<9>),
+                      MAP(imm_v128_align<10>),
+                      MAP(imm_v128_align<11>),
+                      MAP(imm_v128_align<12>),
+                      MAP(imm_v128_align<13>),
+                      MAP(imm_v128_align<14>),
+                      MAP(imm_v128_align<15>),
+                      MAP(v128_abs_s8),
+                      MAP(v128_abs_s16),
+                      MAP(v128_padd_u8),
+                      MAP(v128_padd_s16),
+                      MAP(v128_unpacklo_u16_s32),
+                      MAP(v128_unpacklo_s16_s32),
+                      MAP(v128_unpackhi_u16_s32),
+                      MAP(v128_unpackhi_s16_s32),
+                      MAP(imm_v128_shr_n_byte<1>),
+                      MAP(imm_v128_shr_n_byte<2>),
+                      MAP(imm_v128_shr_n_byte<3>),
+                      MAP(imm_v128_shr_n_byte<4>),
+                      MAP(imm_v128_shr_n_byte<5>),
+                      MAP(imm_v128_shr_n_byte<6>),
+                      MAP(imm_v128_shr_n_byte<7>),
+                      MAP(imm_v128_shr_n_byte<8>),
+                      MAP(imm_v128_shr_n_byte<9>),
+                      MAP(imm_v128_shr_n_byte<10>),
+                      MAP(imm_v128_shr_n_byte<11>),
+                      MAP(imm_v128_shr_n_byte<12>),
+                      MAP(imm_v128_shr_n_byte<13>),
+                      MAP(imm_v128_shr_n_byte<14>),
+                      MAP(imm_v128_shr_n_byte<15>),
+                      MAP(imm_v128_shl_n_byte<1>),
+                      MAP(imm_v128_shl_n_byte<2>),
+                      MAP(imm_v128_shl_n_byte<3>),
+                      MAP(imm_v128_shl_n_byte<4>),
+                      MAP(imm_v128_shl_n_byte<5>),
+                      MAP(imm_v128_shl_n_byte<6>),
+                      MAP(imm_v128_shl_n_byte<7>),
+                      MAP(imm_v128_shl_n_byte<8>),
+                      MAP(imm_v128_shl_n_byte<9>),
+                      MAP(imm_v128_shl_n_byte<10>),
+                      MAP(imm_v128_shl_n_byte<11>),
+                      MAP(imm_v128_shl_n_byte<12>),
+                      MAP(imm_v128_shl_n_byte<13>),
+                      MAP(imm_v128_shl_n_byte<14>),
+                      MAP(imm_v128_shl_n_byte<15>),
+                      MAP(imm_v128_shl_n_8<1>),
+                      MAP(imm_v128_shl_n_8<2>),
+                      MAP(imm_v128_shl_n_8<3>),
+                      MAP(imm_v128_shl_n_8<4>),
+                      MAP(imm_v128_shl_n_8<5>),
+                      MAP(imm_v128_shl_n_8<6>),
+                      MAP(imm_v128_shl_n_8<7>),
+                      MAP(imm_v128_shr_n_u8<1>),
+                      MAP(imm_v128_shr_n_u8<2>),
+                      MAP(imm_v128_shr_n_u8<3>),
+                      MAP(imm_v128_shr_n_u8<4>),
+                      MAP(imm_v128_shr_n_u8<5>),
+                      MAP(imm_v128_shr_n_u8<6>),
+                      MAP(imm_v128_shr_n_u8<7>),
+                      MAP(imm_v128_shr_n_s8<1>),
+                      MAP(imm_v128_shr_n_s8<2>),
+                      MAP(imm_v128_shr_n_s8<3>),
+                      MAP(imm_v128_shr_n_s8<4>),
+                      MAP(imm_v128_shr_n_s8<5>),
+                      MAP(imm_v128_shr_n_s8<6>),
+                      MAP(imm_v128_shr_n_s8<7>),
+                      MAP(imm_v128_shl_n_16<1>),
+                      MAP(imm_v128_shl_n_16<2>),
+                      MAP(imm_v128_shl_n_16<4>),
+                      MAP(imm_v128_shl_n_16<6>),
+                      MAP(imm_v128_shl_n_16<8>),
+                      MAP(imm_v128_shl_n_16<10>),
+                      MAP(imm_v128_shl_n_16<12>),
+                      MAP(imm_v128_shl_n_16<14>),
+                      MAP(imm_v128_shr_n_u16<1>),
+                      MAP(imm_v128_shr_n_u16<2>),
+                      MAP(imm_v128_shr_n_u16<4>),
+                      MAP(imm_v128_shr_n_u16<6>),
+                      MAP(imm_v128_shr_n_u16<8>),
+                      MAP(imm_v128_shr_n_u16<10>),
+                      MAP(imm_v128_shr_n_u16<12>),
+                      MAP(imm_v128_shr_n_u16<14>),
+                      MAP(imm_v128_shr_n_s16<1>),
+                      MAP(imm_v128_shr_n_s16<2>),
+                      MAP(imm_v128_shr_n_s16<4>),
+                      MAP(imm_v128_shr_n_s16<6>),
+                      MAP(imm_v128_shr_n_s16<8>),
+                      MAP(imm_v128_shr_n_s16<10>),
+                      MAP(imm_v128_shr_n_s16<12>),
+                      MAP(imm_v128_shr_n_s16<14>),
+                      MAP(imm_v128_shl_n_32<1>),
+                      MAP(imm_v128_shl_n_32<4>),
+                      MAP(imm_v128_shl_n_32<8>),
+                      MAP(imm_v128_shl_n_32<12>),
+                      MAP(imm_v128_shl_n_32<16>),
+                      MAP(imm_v128_shl_n_32<20>),
+                      MAP(imm_v128_shl_n_32<24>),
+                      MAP(imm_v128_shl_n_32<28>),
+                      MAP(imm_v128_shr_n_u32<1>),
+                      MAP(imm_v128_shr_n_u32<4>),
+                      MAP(imm_v128_shr_n_u32<8>),
+                      MAP(imm_v128_shr_n_u32<12>),
+                      MAP(imm_v128_shr_n_u32<16>),
+                      MAP(imm_v128_shr_n_u32<20>),
+                      MAP(imm_v128_shr_n_u32<24>),
+                      MAP(imm_v128_shr_n_u32<28>),
+                      MAP(imm_v128_shr_n_s32<1>),
+                      MAP(imm_v128_shr_n_s32<4>),
+                      MAP(imm_v128_shr_n_s32<8>),
+                      MAP(imm_v128_shr_n_s32<12>),
+                      MAP(imm_v128_shr_n_s32<16>),
+                      MAP(imm_v128_shr_n_s32<20>),
+                      MAP(imm_v128_shr_n_s32<24>),
+                      MAP(imm_v128_shr_n_s32<28>),
+                      MAP(imm_v128_shl_n_64<1>),
+                      MAP(imm_v128_shl_n_64<4>),
+                      MAP(imm_v128_shl_n_64<8>),
+                      MAP(imm_v128_shl_n_64<12>),
+                      MAP(imm_v128_shl_n_64<16>),
+                      MAP(imm_v128_shl_n_64<20>),
+                      MAP(imm_v128_shl_n_64<24>),
+                      MAP(imm_v128_shl_n_64<28>),
+                      MAP(imm_v128_shl_n_64<32>),
+                      MAP(imm_v128_shl_n_64<36>),
+                      MAP(imm_v128_shl_n_64<40>),
+                      MAP(imm_v128_shl_n_64<44>),
+                      MAP(imm_v128_shl_n_64<48>),
+                      MAP(imm_v128_shl_n_64<52>),
+                      MAP(imm_v128_shl_n_64<56>),
+                      MAP(imm_v128_shl_n_64<60>),
+                      MAP(imm_v128_shr_n_u64<1>),
+                      MAP(imm_v128_shr_n_u64<4>),
+                      MAP(imm_v128_shr_n_u64<8>),
+                      MAP(imm_v128_shr_n_u64<12>),
+                      MAP(imm_v128_shr_n_u64<16>),
+                      MAP(imm_v128_shr_n_u64<20>),
+                      MAP(imm_v128_shr_n_u64<24>),
+                      MAP(imm_v128_shr_n_u64<28>),
+                      MAP(imm_v128_shr_n_u64<32>),
+                      MAP(imm_v128_shr_n_u64<36>),
+                      MAP(imm_v128_shr_n_u64<40>),
+                      MAP(imm_v128_shr_n_u64<44>),
+                      MAP(imm_v128_shr_n_u64<48>),
+                      MAP(imm_v128_shr_n_u64<52>),
+                      MAP(imm_v128_shr_n_u64<56>),
+                      MAP(imm_v128_shr_n_u64<60>),
+                      MAP(imm_v128_shr_n_s64<1>),
+                      MAP(imm_v128_shr_n_s64<4>),
+                      MAP(imm_v128_shr_n_s64<8>),
+                      MAP(imm_v128_shr_n_s64<12>),
+                      MAP(imm_v128_shr_n_s64<16>),
+                      MAP(imm_v128_shr_n_s64<20>),
+                      MAP(imm_v128_shr_n_s64<24>),
+                      MAP(imm_v128_shr_n_s64<28>),
+                      MAP(imm_v128_shr_n_s64<32>),
+                      MAP(imm_v128_shr_n_s64<36>),
+                      MAP(imm_v128_shr_n_s64<40>),
+                      MAP(imm_v128_shr_n_s64<44>),
+                      MAP(imm_v128_shr_n_s64<48>),
+                      MAP(imm_v128_shr_n_s64<52>),
+                      MAP(imm_v128_shr_n_s64<56>),
+                      MAP(imm_v128_shr_n_s64<60>),
+                      MAP(v128_from_v64),
+                      MAP(v128_zip_8),
+                      MAP(v128_zip_16),
+                      MAP(v128_zip_32),
+                      MAP(v128_mul_s16),
+                      MAP(v128_unpack_u8_s16),
+                      MAP(v128_unpack_s8_s16),
+                      MAP(v128_unpack_u16_s32),
+                      MAP(v128_unpack_s16_s32),
+                      MAP(v128_shl_8),
+                      MAP(v128_shr_u8),
+                      MAP(v128_shr_s8),
+                      MAP(v128_shl_16),
+                      MAP(v128_shr_u16),
+                      MAP(v128_shr_s16),
+                      MAP(v128_shl_32),
+                      MAP(v128_shr_u32),
+                      MAP(v128_shr_s32),
+                      MAP(v128_shl_64),
+                      MAP(v128_shr_u64),
+                      MAP(v128_shr_s64),
+                      MAP(v128_hadd_u8),
+                      MAP(v128_dotp_su8),
+                      MAP(v128_dotp_s16),
+                      MAP(v128_dotp_s32),
+                      MAP(v128_low_u32),
+                      MAP(v128_low_v64),
+                      MAP(v128_high_v64),
+                      MAP(v128_from_64),
+                      MAP(v128_from_32),
+                      MAP(v128_movemask_8),
+                      MAP(v128_zero),
+                      MAP(v128_dup_8),
+                      MAP(v128_dup_16),
+                      MAP(v128_dup_32),
+                      MAP(v128_dup_64),
+                      MAP(v128_unpacklo_u8_s16),
+                      MAP(v128_unpackhi_u8_s16),
+                      MAP(v128_unpacklo_s8_s16),
+                      MAP(v128_unpackhi_s8_s16),
+                      MAP(v128_blend_8),
+                      MAP(u32_load_unaligned),
+                      MAP(u32_store_unaligned),
+                      MAP(v64_load_unaligned),
+                      MAP(v64_store_unaligned),
+                      MAP(v128_load_unaligned),
+                      MAP(v128_store_unaligned),
+                      MAP(v256_sad_u8),
+                      MAP(v256_ssd_u8),
+                      MAP(v256_sad_u16),
+                      MAP(v256_ssd_s16),
+                      MAP(v256_hadd_u8),
+                      MAP(v256_low_u64),
+                      MAP(v256_dotp_su8),
+                      MAP(v256_dotp_s16),
+                      MAP(v256_dotp_s32),
+                      MAP(v256_add_8),
+                      MAP(v256_add_16),
+                      MAP(v256_sadd_s8),
+                      MAP(v256_sadd_u8),
+                      MAP(v256_sadd_s16),
+                      MAP(v256_add_32),
+                      MAP(v256_add_64),
+                      MAP(v256_sub_8),
+                      MAP(v256_ssub_u8),
+                      MAP(v256_ssub_s8),
+                      MAP(v256_sub_16),
+                      MAP(v256_ssub_u16),
+                      MAP(v256_ssub_s16),
+                      MAP(v256_sub_32),
+                      MAP(v256_sub_64),
+                      MAP(v256_ziplo_8),
+                      MAP(v256_ziphi_8),
+                      MAP(v256_ziplo_16),
+                      MAP(v256_ziphi_16),
+                      MAP(v256_ziplo_32),
+                      MAP(v256_ziphi_32),
+                      MAP(v256_ziplo_64),
+                      MAP(v256_ziphi_64),
+                      MAP(v256_unziphi_8),
+                      MAP(v256_unziplo_8),
+                      MAP(v256_unziphi_16),
+                      MAP(v256_unziplo_16),
+                      MAP(v256_unziphi_32),
+                      MAP(v256_unziplo_32),
+                      MAP(v256_unziphi_64),
+                      MAP(v256_unziplo_64),
+                      MAP(v256_pack_s32_u16),
+                      MAP(v256_pack_s32_s16),
+                      MAP(v256_pack_s16_u8),
+                      MAP(v256_pack_s16_s8),
+                      MAP(v256_or),
+                      MAP(v256_xor),
+                      MAP(v256_and),
+                      MAP(v256_andn),
+                      MAP(v256_mullo_s16),
+                      MAP(v256_mulhi_s16),
+                      MAP(v256_mullo_s32),
+                      MAP(v256_madd_s16),
+                      MAP(v256_madd_us8),
+                      MAP(v256_avg_u8),
+                      MAP(v256_rdavg_u8),
+                      MAP(v256_rdavg_u16),
+                      MAP(v256_avg_u16),
+                      MAP(v256_min_u8),
+                      MAP(v256_max_u8),
+                      MAP(v256_min_s8),
+                      MAP(v256_max_s8),
+                      MAP(v256_min_s16),
+                      MAP(v256_max_s16),
+                      MAP(v256_min_s32),
+                      MAP(v256_max_s32),
+                      MAP(v256_cmpgt_s8),
+                      MAP(v256_cmplt_s8),
+                      MAP(v256_cmpeq_8),
+                      MAP(v256_cmpgt_s16),
+                      MAP(v256_cmplt_s16),
+                      MAP(v256_cmpeq_16),
+                      MAP(v256_cmpgt_s32),
+                      MAP(v256_cmplt_s32),
+                      MAP(v256_cmpeq_32),
+                      MAP(v256_shuffle_8),
+                      MAP(v256_pshuffle_8),
+                      MAP(v256_wideshuffle_8),
+                      MAP(imm_v256_align<1>),
+                      MAP(imm_v256_align<2>),
+                      MAP(imm_v256_align<3>),
+                      MAP(imm_v256_align<4>),
+                      MAP(imm_v256_align<5>),
+                      MAP(imm_v256_align<6>),
+                      MAP(imm_v256_align<7>),
+                      MAP(imm_v256_align<8>),
+                      MAP(imm_v256_align<9>),
+                      MAP(imm_v256_align<10>),
+                      MAP(imm_v256_align<11>),
+                      MAP(imm_v256_align<12>),
+                      MAP(imm_v256_align<13>),
+                      MAP(imm_v256_align<14>),
+                      MAP(imm_v256_align<15>),
+                      MAP(imm_v256_align<16>),
+                      MAP(imm_v256_align<17>),
+                      MAP(imm_v256_align<18>),
+                      MAP(imm_v256_align<19>),
+                      MAP(imm_v256_align<20>),
+                      MAP(imm_v256_align<21>),
+                      MAP(imm_v256_align<22>),
+                      MAP(imm_v256_align<23>),
+                      MAP(imm_v256_align<24>),
+                      MAP(imm_v256_align<25>),
+                      MAP(imm_v256_align<26>),
+                      MAP(imm_v256_align<27>),
+                      MAP(imm_v256_align<28>),
+                      MAP(imm_v256_align<29>),
+                      MAP(imm_v256_align<30>),
+                      MAP(imm_v256_align<31>),
+                      MAP(v256_from_v128),
+                      MAP(v256_zip_8),
+                      MAP(v256_zip_16),
+                      MAP(v256_zip_32),
+                      MAP(v256_mul_s16),
+                      MAP(v256_unpack_u8_s16),
+                      MAP(v256_unpack_s8_s16),
+                      MAP(v256_unpack_u16_s32),
+                      MAP(v256_unpack_s16_s32),
+                      MAP(v256_shl_8),
+                      MAP(v256_shr_u8),
+                      MAP(v256_shr_s8),
+                      MAP(v256_shl_16),
+                      MAP(v256_shr_u16),
+                      MAP(v256_shr_s16),
+                      MAP(v256_shl_32),
+                      MAP(v256_shr_u32),
+                      MAP(v256_shr_s32),
+                      MAP(v256_shl_64),
+                      MAP(v256_shr_u64),
+                      MAP(v256_shr_s64),
+                      MAP(v256_abs_s8),
+                      MAP(v256_abs_s16),
+                      MAP(v256_padd_u8),
+                      MAP(v256_padd_s16),
+                      MAP(v256_unpacklo_u16_s32),
+                      MAP(v256_unpacklo_s16_s32),
+                      MAP(v256_unpackhi_u16_s32),
+                      MAP(v256_unpackhi_s16_s32),
+                      MAP(imm_v256_shr_n_word<1>),
+                      MAP(imm_v256_shr_n_word<2>),
+                      MAP(imm_v256_shr_n_word<3>),
+                      MAP(imm_v256_shr_n_word<4>),
+                      MAP(imm_v256_shr_n_word<5>),
+                      MAP(imm_v256_shr_n_word<6>),
+                      MAP(imm_v256_shr_n_word<7>),
+                      MAP(imm_v256_shr_n_word<8>),
+                      MAP(imm_v256_shr_n_word<9>),
+                      MAP(imm_v256_shr_n_word<10>),
+                      MAP(imm_v256_shr_n_word<11>),
+                      MAP(imm_v256_shr_n_word<12>),
+                      MAP(imm_v256_shr_n_word<13>),
+                      MAP(imm_v256_shr_n_word<14>),
+                      MAP(imm_v256_shr_n_word<15>),
+                      MAP(imm_v256_shl_n_word<1>),
+                      MAP(imm_v256_shl_n_word<2>),
+                      MAP(imm_v256_shl_n_word<3>),
+                      MAP(imm_v256_shl_n_word<4>),
+                      MAP(imm_v256_shl_n_word<5>),
+                      MAP(imm_v256_shl_n_word<6>),
+                      MAP(imm_v256_shl_n_word<7>),
+                      MAP(imm_v256_shl_n_word<8>),
+                      MAP(imm_v256_shl_n_word<9>),
+                      MAP(imm_v256_shl_n_word<10>),
+                      MAP(imm_v256_shl_n_word<11>),
+                      MAP(imm_v256_shl_n_word<12>),
+                      MAP(imm_v256_shl_n_word<13>),
+                      MAP(imm_v256_shl_n_word<14>),
+                      MAP(imm_v256_shl_n_word<15>),
+                      MAP(imm_v256_shr_n_byte<1>),
+                      MAP(imm_v256_shr_n_byte<2>),
+                      MAP(imm_v256_shr_n_byte<3>),
+                      MAP(imm_v256_shr_n_byte<4>),
+                      MAP(imm_v256_shr_n_byte<5>),
+                      MAP(imm_v256_shr_n_byte<6>),
+                      MAP(imm_v256_shr_n_byte<7>),
+                      MAP(imm_v256_shr_n_byte<8>),
+                      MAP(imm_v256_shr_n_byte<9>),
+                      MAP(imm_v256_shr_n_byte<10>),
+                      MAP(imm_v256_shr_n_byte<11>),
+                      MAP(imm_v256_shr_n_byte<12>),
+                      MAP(imm_v256_shr_n_byte<13>),
+                      MAP(imm_v256_shr_n_byte<14>),
+                      MAP(imm_v256_shr_n_byte<15>),
+                      MAP(imm_v256_shr_n_byte<16>),
+                      MAP(imm_v256_shr_n_byte<17>),
+                      MAP(imm_v256_shr_n_byte<18>),
+                      MAP(imm_v256_shr_n_byte<19>),
+                      MAP(imm_v256_shr_n_byte<20>),
+                      MAP(imm_v256_shr_n_byte<21>),
+                      MAP(imm_v256_shr_n_byte<22>),
+                      MAP(imm_v256_shr_n_byte<23>),
+                      MAP(imm_v256_shr_n_byte<24>),
+                      MAP(imm_v256_shr_n_byte<25>),
+                      MAP(imm_v256_shr_n_byte<26>),
+                      MAP(imm_v256_shr_n_byte<27>),
+                      MAP(imm_v256_shr_n_byte<28>),
+                      MAP(imm_v256_shr_n_byte<29>),
+                      MAP(imm_v256_shr_n_byte<30>),
+                      MAP(imm_v256_shr_n_byte<31>),
+                      MAP(imm_v256_shl_n_byte<1>),
+                      MAP(imm_v256_shl_n_byte<2>),
+                      MAP(imm_v256_shl_n_byte<3>),
+                      MAP(imm_v256_shl_n_byte<4>),
+                      MAP(imm_v256_shl_n_byte<5>),
+                      MAP(imm_v256_shl_n_byte<6>),
+                      MAP(imm_v256_shl_n_byte<7>),
+                      MAP(imm_v256_shl_n_byte<8>),
+                      MAP(imm_v256_shl_n_byte<9>),
+                      MAP(imm_v256_shl_n_byte<10>),
+                      MAP(imm_v256_shl_n_byte<11>),
+                      MAP(imm_v256_shl_n_byte<12>),
+                      MAP(imm_v256_shl_n_byte<13>),
+                      MAP(imm_v256_shl_n_byte<14>),
+                      MAP(imm_v256_shl_n_byte<15>),
+                      MAP(imm_v256_shl_n_byte<16>),
+                      MAP(imm_v256_shl_n_byte<17>),
+                      MAP(imm_v256_shl_n_byte<18>),
+                      MAP(imm_v256_shl_n_byte<19>),
+                      MAP(imm_v256_shl_n_byte<20>),
+                      MAP(imm_v256_shl_n_byte<21>),
+                      MAP(imm_v256_shl_n_byte<22>),
+                      MAP(imm_v256_shl_n_byte<23>),
+                      MAP(imm_v256_shl_n_byte<24>),
+                      MAP(imm_v256_shl_n_byte<25>),
+                      MAP(imm_v256_shl_n_byte<26>),
+                      MAP(imm_v256_shl_n_byte<27>),
+                      MAP(imm_v256_shl_n_byte<28>),
+                      MAP(imm_v256_shl_n_byte<29>),
+                      MAP(imm_v256_shl_n_byte<30>),
+                      MAP(imm_v256_shl_n_byte<31>),
+                      MAP(imm_v256_shl_n_8<1>),
+                      MAP(imm_v256_shl_n_8<2>),
+                      MAP(imm_v256_shl_n_8<3>),
+                      MAP(imm_v256_shl_n_8<4>),
+                      MAP(imm_v256_shl_n_8<5>),
+                      MAP(imm_v256_shl_n_8<6>),
+                      MAP(imm_v256_shl_n_8<7>),
+                      MAP(imm_v256_shr_n_u8<1>),
+                      MAP(imm_v256_shr_n_u8<2>),
+                      MAP(imm_v256_shr_n_u8<3>),
+                      MAP(imm_v256_shr_n_u8<4>),
+                      MAP(imm_v256_shr_n_u8<5>),
+                      MAP(imm_v256_shr_n_u8<6>),
+                      MAP(imm_v256_shr_n_u8<7>),
+                      MAP(imm_v256_shr_n_s8<1>),
+                      MAP(imm_v256_shr_n_s8<2>),
+                      MAP(imm_v256_shr_n_s8<3>),
+                      MAP(imm_v256_shr_n_s8<4>),
+                      MAP(imm_v256_shr_n_s8<5>),
+                      MAP(imm_v256_shr_n_s8<6>),
+                      MAP(imm_v256_shr_n_s8<7>),
+                      MAP(imm_v256_shl_n_16<1>),
+                      MAP(imm_v256_shl_n_16<2>),
+                      MAP(imm_v256_shl_n_16<4>),
+                      MAP(imm_v256_shl_n_16<6>),
+                      MAP(imm_v256_shl_n_16<8>),
+                      MAP(imm_v256_shl_n_16<10>),
+                      MAP(imm_v256_shl_n_16<12>),
+                      MAP(imm_v256_shl_n_16<14>),
+                      MAP(imm_v256_shr_n_u16<1>),
+                      MAP(imm_v256_shr_n_u16<2>),
+                      MAP(imm_v256_shr_n_u16<4>),
+                      MAP(imm_v256_shr_n_u16<6>),
+                      MAP(imm_v256_shr_n_u16<8>),
+                      MAP(imm_v256_shr_n_u16<10>),
+                      MAP(imm_v256_shr_n_u16<12>),
+                      MAP(imm_v256_shr_n_u16<14>),
+                      MAP(imm_v256_shr_n_s16<1>),
+                      MAP(imm_v256_shr_n_s16<2>),
+                      MAP(imm_v256_shr_n_s16<4>),
+                      MAP(imm_v256_shr_n_s16<6>),
+                      MAP(imm_v256_shr_n_s16<8>),
+                      MAP(imm_v256_shr_n_s16<10>),
+                      MAP(imm_v256_shr_n_s16<12>),
+                      MAP(imm_v256_shr_n_s16<14>),
+                      MAP(imm_v256_shl_n_32<1>),
+                      MAP(imm_v256_shl_n_32<4>),
+                      MAP(imm_v256_shl_n_32<8>),
+                      MAP(imm_v256_shl_n_32<12>),
+                      MAP(imm_v256_shl_n_32<16>),
+                      MAP(imm_v256_shl_n_32<20>),
+                      MAP(imm_v256_shl_n_32<24>),
+                      MAP(imm_v256_shl_n_32<28>),
+                      MAP(imm_v256_shr_n_u32<1>),
+                      MAP(imm_v256_shr_n_u32<4>),
+                      MAP(imm_v256_shr_n_u32<8>),
+                      MAP(imm_v256_shr_n_u32<12>),
+                      MAP(imm_v256_shr_n_u32<16>),
+                      MAP(imm_v256_shr_n_u32<20>),
+                      MAP(imm_v256_shr_n_u32<24>),
+                      MAP(imm_v256_shr_n_u32<28>),
+                      MAP(imm_v256_shr_n_s32<1>),
+                      MAP(imm_v256_shr_n_s32<4>),
+                      MAP(imm_v256_shr_n_s32<8>),
+                      MAP(imm_v256_shr_n_s32<12>),
+                      MAP(imm_v256_shr_n_s32<16>),
+                      MAP(imm_v256_shr_n_s32<20>),
+                      MAP(imm_v256_shr_n_s32<24>),
+                      MAP(imm_v256_shr_n_s32<28>),
+                      MAP(imm_v256_shl_n_64<1>),
+                      MAP(imm_v256_shl_n_64<4>),
+                      MAP(imm_v256_shl_n_64<8>),
+                      MAP(imm_v256_shl_n_64<12>),
+                      MAP(imm_v256_shl_n_64<16>),
+                      MAP(imm_v256_shl_n_64<20>),
+                      MAP(imm_v256_shl_n_64<24>),
+                      MAP(imm_v256_shl_n_64<28>),
+                      MAP(imm_v256_shl_n_64<32>),
+                      MAP(imm_v256_shl_n_64<36>),
+                      MAP(imm_v256_shl_n_64<40>),
+                      MAP(imm_v256_shl_n_64<44>),
+                      MAP(imm_v256_shl_n_64<48>),
+                      MAP(imm_v256_shl_n_64<52>),
+                      MAP(imm_v256_shl_n_64<56>),
+                      MAP(imm_v256_shl_n_64<60>),
+                      MAP(imm_v256_shr_n_u64<1>),
+                      MAP(imm_v256_shr_n_u64<4>),
+                      MAP(imm_v256_shr_n_u64<8>),
+                      MAP(imm_v256_shr_n_u64<12>),
+                      MAP(imm_v256_shr_n_u64<16>),
+                      MAP(imm_v256_shr_n_u64<20>),
+                      MAP(imm_v256_shr_n_u64<24>),
+                      MAP(imm_v256_shr_n_u64<28>),
+                      MAP(imm_v256_shr_n_u64<32>),
+                      MAP(imm_v256_shr_n_u64<36>),
+                      MAP(imm_v256_shr_n_u64<40>),
+                      MAP(imm_v256_shr_n_u64<44>),
+                      MAP(imm_v256_shr_n_u64<48>),
+                      MAP(imm_v256_shr_n_u64<52>),
+                      MAP(imm_v256_shr_n_u64<56>),
+                      MAP(imm_v256_shr_n_u64<60>),
+                      MAP(imm_v256_shr_n_s64<1>),
+                      MAP(imm_v256_shr_n_s64<4>),
+                      MAP(imm_v256_shr_n_s64<8>),
+                      MAP(imm_v256_shr_n_s64<12>),
+                      MAP(imm_v256_shr_n_s64<16>),
+                      MAP(imm_v256_shr_n_s64<20>),
+                      MAP(imm_v256_shr_n_s64<24>),
+                      MAP(imm_v256_shr_n_s64<28>),
+                      MAP(imm_v256_shr_n_s64<32>),
+                      MAP(imm_v256_shr_n_s64<36>),
+                      MAP(imm_v256_shr_n_s64<40>),
+                      MAP(imm_v256_shr_n_s64<44>),
+                      MAP(imm_v256_shr_n_s64<48>),
+                      MAP(imm_v256_shr_n_s64<52>),
+                      MAP(imm_v256_shr_n_s64<56>),
+                      MAP(imm_v256_shr_n_s64<60>),
+                      MAP(v256_movemask_8),
+                      MAP(v256_zero),
+                      MAP(v256_dup_8),
+                      MAP(v256_dup_16),
+                      MAP(v256_dup_32),
+                      MAP(v256_dup_64),
+                      MAP(v256_low_u32),
+                      MAP(v256_low_v64),
+                      MAP(v256_from_64),
+                      MAP(v256_from_v64),
+                      MAP(v256_ziplo_128),
+                      MAP(v256_ziphi_128),
+                      MAP(v256_unpacklo_u8_s16),
+                      MAP(v256_unpackhi_u8_s16),
+                      MAP(v256_unpacklo_s8_s16),
+                      MAP(v256_unpackhi_s8_s16),
+                      MAP(v256_blend_8),
+                      { NULL, NULL, NULL } };
+#undef MAP
+
+// Map reference functions to machine tuned functions. Since the
+// functions depend on machine tuned types, the non-machine tuned
+// instantiations of the test can't refer to these functions directly,
+// so we refer to them by name and do the mapping here.
+void Map(const char *name, fptr *ref, fptr *simd) {
+  unsigned int i;
+  for (i = 0; m[i].name && strcmp(name, m[i].name); i++) {
+  }
+
+  *ref = m[i].ref;
+  *simd = m[i].simd;
+}
+
+// Used for printing errors in TestSimd1Arg, TestSimd2Args and TestSimd3Args
+std::string Print(const uint8_t *a, int size) {
+  std::string text = "0x";
+  for (int i = 0; i < size; i++) {
+    const uint8_t c = a[!CONFIG_BIG_ENDIAN ? size - 1 - i : i];
+    // Same as snprintf(..., ..., "%02x", c)
+    text += (c >> 4) + '0' + ((c >> 4) > 9) * ('a' - '0' - 10);
+    text += (c & 15) + '0' + ((c & 15) > 9) * ('a' - '0' - 10);
+  }
+
+  return text;
+}
+
+// Used in TestSimd1Arg, TestSimd2Args and TestSimd3Args to restrict argument
+// ranges
+void SetMask(uint8_t *s, int size, uint32_t mask, uint32_t maskwidth) {
+  switch (maskwidth) {
+    case 0: {
+      break;
+    }
+    case 8: {
+      for (int i = 0; i < size; i++) s[i] &= mask;
+      break;
+    }
+    case 16: {
+      uint16_t *t = reinterpret_cast<uint16_t *>(s);
+      assert(!(reinterpret_cast<uintptr_t>(s) & 1));
+      for (int i = 0; i < size / 2; i++) t[i] &= mask;
+      break;
+    }
+    case 32: {
+      uint32_t *t = reinterpret_cast<uint32_t *>(s);
+      assert(!(reinterpret_cast<uintptr_t>(s) & 3));
+      for (int i = 0; i < size / 4; i++) t[i] &= mask;
+      break;
+    }
+    case 64: {
+      uint64_t *t = reinterpret_cast<uint64_t *>(s);
+      assert(!(reinterpret_cast<uintptr_t>(s) & 7));
+      for (int i = 0; i < size / 8; i++) t[i] &= mask;
+      break;
+    }
+    default: {
+      FAIL() << "Unsupported mask width";
+      break;
+    }
+  }
+}
+
+// We need some extra load/store functions
+void u64_store_aligned(void *p, uint64_t a) {
+  v64_store_aligned(p, v64_from_64(a));
+}
+void s32_store_aligned(void *p, int32_t a) {
+  u32_store_aligned(p, static_cast<uint32_t>(a));
+}
+void s64_store_aligned(void *p, int64_t a) {
+  v64_store_aligned(p, v64_from_64(static_cast<uint64_t>(a)));
+}
+
+void c_u64_store_aligned(void *p, uint64_t a) {
+  c_v64_store_aligned(p, c_v64_from_64(a));
+}
+
+void c_s32_store_aligned(void *p, int32_t a) {
+  c_u32_store_aligned(p, static_cast<uint32_t>(a));
+}
+
+void c_s64_store_aligned(void *p, int64_t a) {
+  c_v64_store_aligned(p, c_v64_from_64(static_cast<uint64_t>(a)));
+}
+
+uint64_t u64_load_aligned(const void *p) {
+  return v64_u64(v64_load_aligned(p));
+}
+uint16_t u16_load_aligned(const void *p) {
+  return *(reinterpret_cast<const uint16_t *>(p));
+}
+uint8_t u8_load_aligned(const void *p) {
+  return *(reinterpret_cast<const uint8_t *>(p));
+}
+
+uint64_t c_u64_load_aligned(const void *p) {
+  return c_v64_u64(c_v64_load_aligned(p));
+}
+uint16_t c_u16_load_aligned(const void *p) {
+  return *(reinterpret_cast<const uint16_t *>(p));
+}
+uint8_t c_u8_load_aligned(const void *p) {
+  return *(reinterpret_cast<const uint8_t *>(p));
+}
+
+// CompareSimd1Arg, CompareSimd2Args and CompareSimd3Args compare
+// intrinsics taking 1, 2 or 3 arguments respectively with their
+// corresponding C reference.  Ideally, the loads and stores should
+// have gone into the template parameter list, but v64 and v128 could
+// be typedef'ed to the same type (which is the case on x86) and then
+// we can't instantiate both v64 and v128, so the function return and
+// argument types, including the always differing types in the C
+// equivalent are used instead.  The function arguments must be void
+// pointers and then go through a cast to avoid matching errors in the
+// branches eliminated by the typeid tests in the calling function.
+template <typename Ret, typename Arg, typename CRet, typename CArg>
+int CompareSimd1Arg(fptr store, fptr load, fptr simd, void *d, fptr c_store,
+                    fptr c_load, fptr c_simd, void *ref_d, const void *a) {
+  void (*const my_store)(void *, Ret) = (void (*const)(void *, Ret))store;
+  Arg (*const my_load)(const void *) = (Arg(*const)(const void *))load;
+  Ret (*const my_simd)(Arg) = (Ret(*const)(Arg))simd;
+  void (*const my_c_store)(void *, CRet) = (void (*const)(void *, CRet))c_store;
+  CArg (*const my_c_load)(const void *) = (CArg(*const)(const void *))c_load;
+  CRet (*const my_c_simd)(CArg) = (CRet(*const)(CArg))c_simd;
+
+  // Call reference and intrinsic
+  my_c_store(ref_d, my_c_simd(my_c_load(a)));
+  my_store(d, my_simd(my_load(a)));
+
+  // Compare results
+  return memcmp(ref_d, d, sizeof(CRet));
+}
+
+template <typename Ret, typename Arg1, typename Arg2, typename CRet,
+          typename CArg1, typename CArg2>
+int CompareSimd2Args(fptr store, fptr load1, fptr load2, fptr simd, void *d,
+                     fptr c_store, fptr c_load1, fptr c_load2, fptr c_simd,
+                     void *ref_d, const void *a, const void *b) {
+  void (*const my_store)(void *, Ret) = (void (*const)(void *, Ret))store;
+  Arg1 (*const my_load1)(const void *) = (Arg1(*const)(const void *))load1;
+  Arg2 (*const my_load2)(const void *) = (Arg2(*const)(const void *))load2;
+  Ret (*const my_simd)(Arg1, Arg2) = (Ret(*const)(Arg1, Arg2))simd;
+  void (*const my_c_store)(void *, CRet) = (void (*const)(void *, CRet))c_store;
+  CArg1 (*const my_c_load1)(const void *) =
+      (CArg1(*const)(const void *))c_load1;
+  CArg2 (*const my_c_load2)(const void *) =
+      (CArg2(*const)(const void *))c_load2;
+  CRet (*const my_c_simd)(CArg1, CArg2) = (CRet(*const)(CArg1, CArg2))c_simd;
+
+  // Call reference and intrinsic
+  my_c_store(ref_d, my_c_simd(my_c_load1(a), my_c_load2(b)));
+  my_store(d, my_simd(my_load1(a), my_load2(b)));
+
+  // Compare results
+  return memcmp(ref_d, d, sizeof(CRet));
+}
+
+template <typename Ret, typename Arg1, typename Arg2, typename Arg3,
+          typename CRet, typename CArg1, typename CArg2, typename CArg3>
+int CompareSimd3Args(fptr store, fptr load1, fptr load2, fptr load3, fptr simd,
+                     void *d, fptr c_store, fptr c_load1, fptr c_load2,
+                     fptr c_load3, fptr c_simd, void *ref_d, const void *a,
+                     const void *b, const void *c) {
+  void (*const my_store)(void *, Ret) = (void (*const)(void *, Ret))store;
+  Arg1 (*const my_load1)(const void *) = (Arg1(*const)(const void *))load1;
+  Arg2 (*const my_load2)(const void *) = (Arg2(*const)(const void *))load2;
+  Arg3 (*const my_load3)(const void *) = (Arg3(*const)(const void *))load3;
+  Ret (*const my_simd)(Arg1, Arg2, Arg3) = (Ret(*const)(Arg1, Arg2, Arg3))simd;
+  void (*const my_c_store)(void *, CRet) = (void (*const)(void *, CRet))c_store;
+  CArg1 (*const my_c_load1)(const void *) =
+      (CArg1(*const)(const void *))c_load1;
+  CArg2 (*const my_c_load2)(const void *) =
+      (CArg2(*const)(const void *))c_load2;
+  CArg2 (*const my_c_load3)(const void *) =
+      (CArg2(*const)(const void *))c_load3;
+  CRet (*const my_c_simd)(CArg1, CArg2, CArg3) =
+      (CRet(*const)(CArg1, CArg2, CArg3))c_simd;
+
+  // Call reference and intrinsic
+  my_c_store(ref_d, my_c_simd(my_c_load1(a), my_c_load2(b), my_c_load3(c)));
+  my_store(d, my_simd(my_load1(a), my_load2(b), my_load3(c)));
+
+  // Compare results
+  return memcmp(ref_d, d, sizeof(CRet));
+}
+
+}  // namespace
+
+template <typename CRet, typename CArg>
+void TestSimd1Arg(uint32_t iterations, uint32_t mask, uint32_t maskwidth,
+                  const char *name) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  fptr ref_simd;
+  fptr simd;
+  int error = 0;
+  DECLARE_ALIGNED(32, uint8_t, s[32]);
+  DECLARE_ALIGNED(32, uint8_t, d[32]);
+  DECLARE_ALIGNED(32, uint8_t, ref_d[32]);
+  assert(sizeof(CArg) <= 32 && sizeof(CRet) <= 32);
+  memset(ref_d, 0, sizeof(ref_d));
+  memset(d, 0, sizeof(d));
+
+  Map(name, &ref_simd, &simd);
+  if (simd == NULL || ref_simd == NULL) {
+    FAIL() << "Internal error: Unknown intrinsic function " << name;
+  }
+  for (unsigned int count = 0;
+       count < iterations && !error && !testing::Test::HasFailure(); count++) {
+    for (unsigned int c = 0; c < sizeof(CArg); c++) s[c] = rnd.Rand8();
+
+    if (maskwidth) {
+      SetMask(s, sizeof(CArg), mask, maskwidth);
+    }
+
+    if (typeid(CRet) == typeid(c_v64) && typeid(CArg) == typeid(c_v64)) {
+      // V64_V64
+      error = CompareSimd1Arg<v64, v64, CRet, CArg>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v64) &&
+               typeid(CArg) == typeid(uint8_t)) {
+      // V64_U8
+      error = CompareSimd1Arg<v64, uint8_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(u8_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_u8_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v64) &&
+               typeid(CArg) == typeid(uint16_t)) {
+      // V64_U16
+      error = CompareSimd1Arg<v64, uint16_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(u16_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_u16_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v64) &&
+               typeid(CArg) == typeid(uint32_t)) {
+      // V64_U32
+      error = CompareSimd1Arg<v64, uint32_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(u32_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_u32_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(uint64_t) &&
+               typeid(CArg) == typeid(c_v64)) {
+      // U64_V64
+      error = CompareSimd1Arg<uint64_t, v64, CRet, CArg>(
+          reinterpret_cast<fptr>(u64_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u64_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(int64_t) &&
+               typeid(CArg) == typeid(c_v64)) {
+      // S64_V64
+      error = CompareSimd1Arg<int64_t, v64, CRet, CArg>(
+          reinterpret_cast<fptr>(s64_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_s64_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(uint32_t) &&
+               typeid(CArg) == typeid(c_v64)) {
+      // U32_V64
+      error = CompareSimd1Arg<uint32_t, v64, CRet, CArg>(
+          reinterpret_cast<fptr>(u32_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u32_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(int32_t) &&
+               typeid(CArg) == typeid(c_v64)) {
+      // S32_V64
+      error = CompareSimd1Arg<int32_t, v64, CRet, CArg>(
+          reinterpret_cast<fptr>(s32_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_s32_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(uint32_t) &&
+               typeid(CArg) == typeid(c_v128)) {
+      // U32_V128
+      error = CompareSimd1Arg<uint32_t, v128, CRet, CArg>(
+          reinterpret_cast<fptr>(u32_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u32_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(uint64_t) &&
+               typeid(CArg) == typeid(c_v128)) {
+      // U64_V128
+      error = CompareSimd1Arg<uint64_t, v128, CRet, CArg>(
+          reinterpret_cast<fptr>(u64_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u64_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(uint64_t) &&
+               typeid(CArg) == typeid(c_v256)) {
+      // U64_V256
+      error = CompareSimd1Arg<uint64_t, v256, CRet, CArg>(
+          reinterpret_cast<fptr>(u64_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u64_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v64) &&
+               typeid(CArg) == typeid(c_v128)) {
+      // V64_V128
+      error = CompareSimd1Arg<v64, v128, CRet, CArg>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg) == typeid(c_v128)) {
+      // V128_V128
+      error = CompareSimd1Arg<v128, v128, CRet, CArg>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg) == typeid(c_v64)) {
+      // V128_V64
+      error = CompareSimd1Arg<v128, v64, CRet, CArg>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg) == typeid(uint8_t)) {
+      // V128_U8
+      error = CompareSimd1Arg<v128, uint8_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(u8_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_u8_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg) == typeid(uint16_t)) {
+      // V128_U16
+      error = CompareSimd1Arg<v128, uint16_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(u16_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_u16_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg) == typeid(uint32_t)) {
+      // V128_U32
+      error = CompareSimd1Arg<v128, uint32_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(u32_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_u32_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg) == typeid(uint64_t)) {
+      // V128_U64
+      error = CompareSimd1Arg<v128, uint64_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(u64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_u64_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg) == typeid(c_v256)) {
+      // V256_V256
+      error = CompareSimd1Arg<v256, v256, CRet, CArg>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg) == typeid(c_v128)) {
+      // V256_V128
+      error = CompareSimd1Arg<v256, v128, CRet, CArg>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg) == typeid(uint8_t)) {
+      // V256_U8
+      error = CompareSimd1Arg<v256, uint8_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(u8_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_u8_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg) == typeid(uint16_t)) {
+      // V256_U16
+      error = CompareSimd1Arg<v256, uint16_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(u16_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_u16_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg) == typeid(uint32_t)) {
+      // V256_U32
+      error = CompareSimd1Arg<v256, uint32_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(u32_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_u32_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg) == typeid(uint64_t)) {
+      // V256_U64
+      error = CompareSimd1Arg<v256, uint64_t, CRet, CArg>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(u64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_u64_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(uint32_t) &&
+               typeid(CArg) == typeid(c_v256)) {
+      // U32_V256
+      error = CompareSimd1Arg<uint32_t, v256, CRet, CArg>(
+          reinterpret_cast<fptr>(u32_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u32_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned), ref_simd, ref_d, s);
+    } else if (typeid(CRet) == typeid(c_v64) &&
+               typeid(CArg) == typeid(c_v256)) {
+      // V64_V256
+      error = CompareSimd1Arg<v64, v256, CRet, CArg>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned), ref_simd, ref_d, s);
+    } else {
+      FAIL() << "Internal error: Unknown intrinsic function "
+             << typeid(CRet).name() << " " << name << "(" << typeid(CArg).name()
+             << ")";
+    }
+  }
+
+  EXPECT_EQ(0, error) << "Error: mismatch for " << name << "("
+                      << Print(s, sizeof(s)) << ") -> " << Print(d, sizeof(d))
+                      << " (simd), " << Print(ref_d, sizeof(ref_d)) << " (ref)";
+}
+
+template <typename CRet, typename CArg1, typename CArg2>
+void TestSimd2Args(uint32_t iterations, uint32_t mask, uint32_t maskwidth,
+                   const char *name) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  fptr ref_simd;
+  fptr simd;
+  int error = 0;
+  DECLARE_ALIGNED(32, uint8_t, s1[32]);
+  DECLARE_ALIGNED(32, uint8_t, s2[32]);
+  DECLARE_ALIGNED(32, uint8_t, d[32]);
+  DECLARE_ALIGNED(32, uint8_t, ref_d[32]);
+  assert(sizeof(CArg1) <= 32 && sizeof(CArg2) <= 32 && sizeof(CRet) <= 32);
+  memset(ref_d, 0, sizeof(ref_d));
+  memset(d, 0, sizeof(d));
+
+  Map(name, &ref_simd, &simd);
+  if (simd == NULL || ref_simd == NULL) {
+    FAIL() << "Internal error: Unknown intrinsic function " << name;
+  }
+
+  for (unsigned int count = 0;
+       count < iterations && !error && !testing::Test::HasFailure(); count++) {
+    for (unsigned int c = 0; c < sizeof(CArg1); c++) s1[c] = rnd.Rand8();
+
+    for (unsigned int c = 0; c < sizeof(CArg2); c++) s2[c] = rnd.Rand8();
+
+    if (maskwidth) SetMask(s2, sizeof(CArg2), mask, maskwidth);
+
+    if (typeid(CRet) == typeid(c_v64) && typeid(CArg1) == typeid(c_v64) &&
+        typeid(CArg2) == typeid(c_v64)) {
+      // V64_V64V64
+      error = CompareSimd2Args<v64, v64, v64, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v64) &&
+               typeid(CArg1) == typeid(uint32_t) &&
+               typeid(CArg2) == typeid(uint32_t)) {
+      // V64_U32U32
+      error = CompareSimd2Args<v64, uint32_t, uint32_t, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(u32_load_aligned),
+          reinterpret_cast<fptr>(u32_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_u32_load_aligned),
+          reinterpret_cast<fptr>(c_u32_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(uint32_t) &&
+               typeid(CArg1) == typeid(c_v64) &&
+               typeid(CArg2) == typeid(c_v64)) {
+      // U32_V64V64
+      error = CompareSimd2Args<uint32_t, v64, v64, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(u32_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u32_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(int64_t) &&
+               typeid(CArg1) == typeid(c_v64) &&
+               typeid(CArg2) == typeid(c_v64)) {
+      // S64_V64V64
+      error = CompareSimd2Args<int64_t, v64, v64, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(s64_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_s64_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v64) &&
+               typeid(CArg1) == typeid(c_v64) &&
+               typeid(CArg2) == typeid(uint32_t)) {
+      // V64_V64U32
+      error = CompareSimd2Args<v64, v64, uint32_t, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v64_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned),
+          reinterpret_cast<fptr>(u32_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v64_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(c_u32_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg1) == typeid(c_v128) &&
+               typeid(CArg2) == typeid(c_v128)) {
+      // V128_V128V128
+      error = CompareSimd2Args<v128, v128, v128, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(uint32_t) &&
+               typeid(CArg1) == typeid(c_v128) &&
+               typeid(CArg2) == typeid(c_v128)) {
+      // U32_V128V128
+      error = CompareSimd2Args<uint32_t, v128, v128, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(u32_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u32_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(uint64_t) &&
+               typeid(CArg1) == typeid(c_v128) &&
+               typeid(CArg2) == typeid(c_v128)) {
+      // U64_V128V128
+      error = CompareSimd2Args<uint64_t, v128, v128, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(u64_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u64_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(int64_t) &&
+               typeid(CArg1) == typeid(c_v128) &&
+               typeid(CArg2) == typeid(c_v128)) {
+      // S64_V128V128
+      error = CompareSimd2Args<int64_t, v128, v128, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(s64_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_s64_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg1) == typeid(uint64_t) &&
+               typeid(CArg2) == typeid(uint64_t)) {
+      // V128_U64U64
+      error = CompareSimd2Args<v128, uint64_t, uint64_t, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(u64_load_aligned),
+          reinterpret_cast<fptr>(u64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_u64_load_aligned),
+          reinterpret_cast<fptr>(c_u64_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg1) == typeid(c_v64) &&
+               typeid(CArg2) == typeid(c_v64)) {
+      // V128_V64V64
+      error = CompareSimd2Args<v128, v64, v64, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned),
+          reinterpret_cast<fptr>(v64_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(c_v64_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v128) &&
+               typeid(CArg1) == typeid(c_v128) &&
+               typeid(CArg2) == typeid(uint32_t)) {
+      // V128_V128U32
+      error = CompareSimd2Args<v128, v128, uint32_t, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v128_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned),
+          reinterpret_cast<fptr>(u32_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v128_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(c_u32_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg1) == typeid(c_v256) &&
+               typeid(CArg2) == typeid(c_v256)) {
+      // V256_V256V256
+      error = CompareSimd2Args<v256, v256, v256, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(uint64_t) &&
+               typeid(CArg1) == typeid(c_v256) &&
+               typeid(CArg2) == typeid(c_v256)) {
+      // U64_V256V256
+      error = CompareSimd2Args<uint64_t, v256, v256, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(u64_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u64_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(int64_t) &&
+               typeid(CArg1) == typeid(c_v256) &&
+               typeid(CArg2) == typeid(c_v256)) {
+      // S64_V256V256
+      error = CompareSimd2Args<int64_t, v256, v256, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(s64_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_s64_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(uint32_t) &&
+               typeid(CArg1) == typeid(c_v256) &&
+               typeid(CArg2) == typeid(c_v256)) {
+      // U32_V256V256
+      error = CompareSimd2Args<uint32_t, v256, v256, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(u32_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_u32_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg1) == typeid(c_v128) &&
+               typeid(CArg2) == typeid(c_v128)) {
+      // V256_V128V128
+      error = CompareSimd2Args<v256, v128, v128, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned),
+          reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(c_v128_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg1) == typeid(c_v256) &&
+               typeid(CArg2) == typeid(uint32_t)) {
+      // V256_V256U32
+      error = CompareSimd2Args<v256, v256, uint32_t, CRet, CArg1, CArg2>(
+          reinterpret_cast<fptr>(v256_store_aligned),
+          reinterpret_cast<fptr>(v256_load_aligned),
+          reinterpret_cast<fptr>(u32_load_aligned), simd, d,
+          reinterpret_cast<fptr>(c_v256_store_aligned),
+          reinterpret_cast<fptr>(c_v256_load_aligned),
+          reinterpret_cast<fptr>(c_u32_load_aligned),
+          reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2);
+
+    } else {
+      FAIL() << "Internal error: Unknown intrinsic function "
+             << typeid(CRet).name() << " " << name << "("
+             << typeid(CArg1).name() << ", " << typeid(CArg2).name() << ")";
+    }
+  }
+
+  EXPECT_EQ(0, error) << "Error: mismatch for " << name << "("
+                      << Print(s1, sizeof(s1)) << ", " << Print(s2, sizeof(s2))
+                      << ") -> " << Print(d, sizeof(d)) << " (simd), "
+                      << Print(ref_d, sizeof(ref_d)) << " (ref)";
+}
+
+template <typename CRet, typename CArg1, typename CArg2, typename CArg3>
+void TestSimd3Args(uint32_t iterations, uint32_t mask, uint32_t maskwidth,
+                   const char *name) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+  fptr ref_simd;
+  fptr simd;
+  int error = 0;
+  DECLARE_ALIGNED(32, uint8_t, s1[32]);
+  DECLARE_ALIGNED(32, uint8_t, s2[32]);
+  DECLARE_ALIGNED(32, uint8_t, s3[32]);
+  DECLARE_ALIGNED(32, uint8_t, d[32]);
+  DECLARE_ALIGNED(32, uint8_t, ref_d[32]);
+  assert(sizeof(CArg1) <= 32 && sizeof(CArg2) <= 32 && sizeof(CArg3) <= 32 &&
+         sizeof(CRet) <= 32);
+  memset(ref_d, 0, sizeof(ref_d));
+  memset(d, 0, sizeof(d));
+
+  Map(name, &ref_simd, &simd);
+  if (simd == NULL || ref_simd == NULL) {
+    FAIL() << "Internal error: Unknown intrinsic function " << name;
+  }
+
+  for (unsigned int count = 0;
+       count < iterations && !error && !testing::Test::HasFailure(); count++) {
+    for (unsigned int c = 0; c < sizeof(CArg1); c++) s1[c] = rnd.Rand8();
+
+    for (unsigned int c = 0; c < sizeof(CArg2); c++) s2[c] = rnd.Rand8();
+
+    for (unsigned int c = 0; c < sizeof(CArg3); c++) s3[c] = rnd.Rand8();
+
+    if (maskwidth) SetMask(s3, sizeof(CArg3), mask, maskwidth);
+
+    if (typeid(CRet) == typeid(c_v128) && typeid(CArg1) == typeid(c_v128) &&
+        typeid(CArg2) == typeid(c_v128) && typeid(CArg3) == typeid(c_v128)) {
+      // V128_V128V128V128
+      error =
+          CompareSimd3Args<v128, v128, v128, v128, CRet, CArg1, CArg2, CArg3>(
+              reinterpret_cast<fptr>(v128_store_aligned),
+              reinterpret_cast<fptr>(v128_load_aligned),
+              reinterpret_cast<fptr>(v128_load_aligned),
+              reinterpret_cast<fptr>(v128_load_aligned), simd, d,
+              reinterpret_cast<fptr>(c_v128_store_aligned),
+              reinterpret_cast<fptr>(c_v128_load_aligned),
+              reinterpret_cast<fptr>(c_v128_load_aligned),
+              reinterpret_cast<fptr>(c_v128_load_aligned),
+              reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2, s3);
+    } else if (typeid(CRet) == typeid(c_v256) &&
+               typeid(CArg1) == typeid(c_v256) &&
+               typeid(CArg2) == typeid(c_v256) &&
+               typeid(CArg3) == typeid(c_v256)) {
+      // V256_V256V256V256
+      error =
+          CompareSimd3Args<v256, v256, v256, v256, CRet, CArg1, CArg2, CArg3>(
+              reinterpret_cast<fptr>(v256_store_aligned),
+              reinterpret_cast<fptr>(v256_load_aligned),
+              reinterpret_cast<fptr>(v256_load_aligned),
+              reinterpret_cast<fptr>(v256_load_aligned), simd, d,
+              reinterpret_cast<fptr>(c_v256_store_aligned),
+              reinterpret_cast<fptr>(c_v256_load_aligned),
+              reinterpret_cast<fptr>(c_v256_load_aligned),
+              reinterpret_cast<fptr>(c_v256_load_aligned),
+              reinterpret_cast<fptr>(ref_simd), ref_d, s1, s2, s3);
+    } else {
+      FAIL() << "Internal error: Unknown intrinsic function "
+             << typeid(CRet).name() << " " << name << "("
+             << typeid(CArg1).name() << ", " << typeid(CArg2).name() << ", "
+             << typeid(CArg3).name() << ")";
+    }
+  }
+
+  EXPECT_EQ(0, error) << "Error: mismatch for " << name << "("
+                      << Print(s1, sizeof(s1)) << ", " << Print(s2, sizeof(s2))
+                      << ", " << Print(s3, sizeof(s3)) << ") -> "
+                      << Print(d, sizeof(d)) << " (simd), "
+                      << Print(ref_d, sizeof(ref_d)) << " (ref)";
+}
+
+// Instantiations to make the functions callable from another files
+template void TestSimd1Arg<c_v64, uint8_t>(uint32_t, uint32_t, uint32_t,
+                                           const char *);
+template void TestSimd1Arg<c_v64, uint16_t>(uint32_t, uint32_t, uint32_t,
+                                            const char *);
+template void TestSimd1Arg<c_v64, uint32_t>(uint32_t, uint32_t, uint32_t,
+                                            const char *);
+template void TestSimd1Arg<c_v64, c_v64>(uint32_t, uint32_t, uint32_t,
+                                         const char *);
+template void TestSimd1Arg<uint32_t, c_v64>(uint32_t, uint32_t, uint32_t,
+                                            const char *);
+template void TestSimd1Arg<int32_t, c_v64>(uint32_t, uint32_t, uint32_t,
+                                           const char *);
+template void TestSimd1Arg<uint64_t, c_v64>(uint32_t, uint32_t, uint32_t,
+                                            const char *);
+template void TestSimd1Arg<int64_t, c_v64>(uint32_t, uint32_t, uint32_t,
+                                           const char *);
+template void TestSimd2Args<c_v64, uint32_t, uint32_t>(uint32_t, uint32_t,
+                                                       uint32_t, const char *);
+template void TestSimd2Args<c_v64, c_v64, c_v64>(uint32_t, uint32_t, uint32_t,
+                                                 const char *);
+template void TestSimd2Args<c_v64, c_v64, uint32_t>(uint32_t, uint32_t,
+                                                    uint32_t, const char *);
+template void TestSimd2Args<int64_t, c_v64, c_v64>(uint32_t, uint32_t, uint32_t,
+                                                   const char *);
+template void TestSimd2Args<uint32_t, c_v64, c_v64>(uint32_t, uint32_t,
+                                                    uint32_t, const char *);
+template void TestSimd1Arg<c_v128, c_v128>(uint32_t, uint32_t, uint32_t,
+                                           const char *);
+template void TestSimd1Arg<c_v128, uint8_t>(uint32_t, uint32_t, uint32_t,
+                                            const char *);
+template void TestSimd1Arg<c_v128, uint16_t>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<c_v128, uint32_t>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<c_v128, uint64_t>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<c_v128, c_v64>(uint32_t, uint32_t, uint32_t,
+                                          const char *);
+template void TestSimd1Arg<uint32_t, c_v128>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<uint64_t, c_v128>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<c_v64, c_v128>(uint32_t, uint32_t, uint32_t,
+                                          const char *);
+template void TestSimd2Args<c_v128, c_v128, c_v128>(uint32_t, uint32_t,
+                                                    uint32_t, const char *);
+template void TestSimd2Args<c_v128, c_v128, uint32_t>(uint32_t, uint32_t,
+                                                      uint32_t, const char *);
+template void TestSimd2Args<c_v128, uint64_t, uint64_t>(uint32_t, uint32_t,
+                                                        uint32_t, const char *);
+template void TestSimd2Args<c_v128, c_v64, c_v64>(uint32_t, uint32_t, uint32_t,
+                                                  const char *);
+template void TestSimd2Args<uint64_t, c_v128, c_v128>(uint32_t, uint32_t,
+                                                      uint32_t, const char *);
+template void TestSimd2Args<int64_t, c_v128, c_v128>(uint32_t, uint32_t,
+                                                     uint32_t, const char *);
+template void TestSimd2Args<uint32_t, c_v128, c_v128>(uint32_t, uint32_t,
+                                                      uint32_t, const char *);
+template void TestSimd3Args<c_v128, c_v128, c_v128, c_v128>(uint32_t, uint32_t,
+                                                            uint32_t,
+                                                            const char *);
+template void TestSimd1Arg<c_v256, c_v128>(uint32_t, uint32_t, uint32_t,
+                                           const char *);
+template void TestSimd1Arg<c_v256, c_v256>(uint32_t, uint32_t, uint32_t,
+                                           const char *);
+template void TestSimd1Arg<uint64_t, c_v256>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<c_v256, uint8_t>(uint32_t, uint32_t, uint32_t,
+                                            const char *);
+template void TestSimd1Arg<c_v256, uint16_t>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<c_v256, uint32_t>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<c_v256, uint64_t>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<uint32_t, c_v256>(uint32_t, uint32_t, uint32_t,
+                                             const char *);
+template void TestSimd1Arg<c_v64, c_v256>(uint32_t, uint32_t, uint32_t,
+                                          const char *);
+template void TestSimd2Args<c_v256, c_v128, c_v128>(uint32_t, uint32_t,
+                                                    uint32_t, const char *);
+template void TestSimd2Args<c_v256, c_v256, c_v256>(uint32_t, uint32_t,
+                                                    uint32_t, const char *);
+template void TestSimd2Args<c_v256, c_v256, uint32_t>(uint32_t, uint32_t,
+                                                      uint32_t, const char *);
+template void TestSimd2Args<uint64_t, c_v256, c_v256>(uint32_t, uint32_t,
+                                                      uint32_t, const char *);
+template void TestSimd2Args<int64_t, c_v256, c_v256>(uint32_t, uint32_t,
+                                                     uint32_t, const char *);
+template void TestSimd2Args<uint32_t, c_v256, c_v256>(uint32_t, uint32_t,
+                                                      uint32_t, const char *);
+template void TestSimd3Args<c_v256, c_v256, c_v256, c_v256>(uint32_t, uint32_t,
+                                                            uint32_t,
+                                                            const char *);
+
+}  // namespace SIMD_NAMESPACE
diff --git a/third_party/aom/test/simd_cmp_neon.cc b/third_party/aom/test/simd_cmp_neon.cc
new file mode 100644
index 0000000000..53c1e2a07f
--- /dev/null
+++ b/third_party/aom/test/simd_cmp_neon.cc
@@ -0,0 +1,17 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#if defined(__OPTIMIZE__) && __OPTIMIZE__
+#define ARCH NEON
+#define ARCH_POSTFIX(name) name##_neon
+#define SIMD_NAMESPACE simd_test_neon
+#include "test/simd_cmp_impl.h"
+#endif
diff --git a/third_party/aom/test/simd_cmp_sse2.cc b/third_party/aom/test/simd_cmp_sse2.cc
new file mode 100644
index 0000000000..f7827a7fa1
--- /dev/null
+++ b/third_party/aom/test/simd_cmp_sse2.cc
@@ -0,0 +1,18 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#if (defined(__OPTIMIZE__) && __OPTIMIZE__) || \
+    (!defined(__GNUC__) && !defined(_DEBUG))
+#define ARCH SSE2
+#define ARCH_POSTFIX(name) name##_sse2
+#define SIMD_NAMESPACE simd_test_sse2
+#include "test/simd_cmp_impl.h"
+#endif
diff --git a/third_party/aom/test/simd_cmp_sse4.cc b/third_party/aom/test/simd_cmp_sse4.cc
new file mode 100644
index 0000000000..3566764b64
--- /dev/null
+++ b/third_party/aom/test/simd_cmp_sse4.cc
@@ -0,0 +1,18 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#if (defined(__OPTIMIZE__) && __OPTIMIZE__) || \
+    (!defined(__GNUC__) && !defined(_DEBUG))
+#define ARCH SSE4_1
+#define ARCH_POSTFIX(name) name##_sse4_1
+#define SIMD_NAMESPACE simd_test_sse4_1
+#include "test/simd_cmp_impl.h"
+#endif
diff --git a/third_party/aom/test/simd_cmp_ssse3.cc b/third_party/aom/test/simd_cmp_ssse3.cc
new file mode 100644
index 0000000000..57bf135ddb
--- /dev/null
+++ b/third_party/aom/test/simd_cmp_ssse3.cc
@@ -0,0 +1,18 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#if (defined(__OPTIMIZE__) && __OPTIMIZE__) || \
+    (!defined(__GNUC__) && !defined(_DEBUG))
+#define ARCH SSSE3
+#define ARCH_POSTFIX(name) name##_ssse3
+#define SIMD_NAMESPACE simd_test_ssse3
+#include "test/simd_cmp_impl.h"
+#endif
diff --git a/third_party/aom/test/simd_impl.h b/third_party/aom/test/simd_impl.h
new file mode 100644
index 0000000000..fd06f67fdf
--- /dev/null
+++ b/third_party/aom/test/simd_impl.h
@@ -0,0 +1,1141 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#define SIMD_CHECK 1
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "aom_dsp/aom_simd_inline.h"
+#include "aom_dsp/simd/v256_intrinsics_c.h"
+
+namespace SIMD_NAMESPACE {
+
+template <typename param_signature>
+class TestIntrinsic : public ::testing::TestWithParam<param_signature> {
+ public:
+  virtual ~TestIntrinsic() {}
+  virtual void SetUp() {
+    mask = ::testing::get<0>(this->GetParam());
+    maskwidth = ::testing::get<1>(this->GetParam());
+    name = ::testing::get<2>(this->GetParam());
+  }
+
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+
+ protected:
+  uint32_t mask, maskwidth;
+  const char *name;
+};
+
+// Create one typedef for each function signature
+#define TYPEDEF_SIMD(name)                                                    \
+  typedef TestIntrinsic< ::testing::tuple<uint32_t, uint32_t, const char *> > \
+      ARCH_POSTFIX(name)
+
+TYPEDEF_SIMD(V64_U8);
+TYPEDEF_SIMD(V64_U16);
+TYPEDEF_SIMD(V64_U32);
+TYPEDEF_SIMD(V64_V64);
+TYPEDEF_SIMD(U32_V64);
+TYPEDEF_SIMD(S32_V64);
+TYPEDEF_SIMD(U64_V64);
+TYPEDEF_SIMD(S64_V64);
+TYPEDEF_SIMD(V64_U32U32);
+TYPEDEF_SIMD(V64_V64V64);
+TYPEDEF_SIMD(S64_V64V64);
+TYPEDEF_SIMD(V64_V64U32);
+TYPEDEF_SIMD(U32_V64V64);
+TYPEDEF_SIMD(V128_V64);
+TYPEDEF_SIMD(V128_V128);
+TYPEDEF_SIMD(U32_V128);
+TYPEDEF_SIMD(U64_V128);
+TYPEDEF_SIMD(V64_V128);
+TYPEDEF_SIMD(V128_U8);
+TYPEDEF_SIMD(V128_U16);
+TYPEDEF_SIMD(V128_U32);
+TYPEDEF_SIMD(V128_U64);
+TYPEDEF_SIMD(V128_U64U64);
+TYPEDEF_SIMD(V128_V64V64);
+TYPEDEF_SIMD(V128_V128V128);
+TYPEDEF_SIMD(V128_V128V128V128);
+TYPEDEF_SIMD(S64_V128V128);
+TYPEDEF_SIMD(V128_V128U32);
+TYPEDEF_SIMD(U32_V128V128);
+TYPEDEF_SIMD(U64_V128V128);
+TYPEDEF_SIMD(V256_V128);
+TYPEDEF_SIMD(V256_V256);
+TYPEDEF_SIMD(U64_V256);
+TYPEDEF_SIMD(V256_V128V128);
+TYPEDEF_SIMD(V256_V256V256);
+TYPEDEF_SIMD(V256_V256V256V256);
+TYPEDEF_SIMD(U64_V256V256);
+TYPEDEF_SIMD(S64_V256V256);
+TYPEDEF_SIMD(V256_V256U32);
+TYPEDEF_SIMD(U32_V256V256);
+TYPEDEF_SIMD(V256_U8);
+TYPEDEF_SIMD(V256_U16);
+TYPEDEF_SIMD(V256_U32);
+TYPEDEF_SIMD(V256_U64);
+TYPEDEF_SIMD(U32_V256);
+TYPEDEF_SIMD(V64_V256);
+
+// Google Test allows up to 50 tests per case, so split the largest
+typedef ARCH_POSTFIX(V64_V64) ARCH_POSTFIX(V64_V64_Part2);
+typedef ARCH_POSTFIX(V64_V64V64) ARCH_POSTFIX(V64_V64V64_Part2);
+typedef ARCH_POSTFIX(V128_V128) ARCH_POSTFIX(V128_V128_Part2);
+typedef ARCH_POSTFIX(V128_V128) ARCH_POSTFIX(V128_V128_Part3);
+typedef ARCH_POSTFIX(V128_V128) ARCH_POSTFIX(V128_V128_Part4);
+typedef ARCH_POSTFIX(V128_V128V128) ARCH_POSTFIX(V128_V128V128_Part2);
+typedef ARCH_POSTFIX(V256_V256) ARCH_POSTFIX(V256_V256_Part2);
+typedef ARCH_POSTFIX(V256_V256) ARCH_POSTFIX(V256_V256_Part3);
+typedef ARCH_POSTFIX(V256_V256) ARCH_POSTFIX(V256_V256_Part4);
+typedef ARCH_POSTFIX(V256_V256) ARCH_POSTFIX(V256_V256_Part5);
+typedef ARCH_POSTFIX(V256_V256V256) ARCH_POSTFIX(V256_V256V256_Part2);
+
+// These functions are machine tuned located elsewhere
+template <typename c_ret, typename c_arg>
+void TestSimd1Arg(uint32_t iterations, uint32_t mask, uint32_t maskwidth,
+                  const char *name);
+
+template <typename c_ret, typename c_arg1, typename c_arg2>
+void TestSimd2Args(uint32_t iterations, uint32_t mask, uint32_t maskwidth,
+                   const char *name);
+
+template <typename c_ret, typename c_arg1, typename c_arg2, typename c_arg3>
+void TestSimd3Args(uint32_t iterations, uint32_t mask, uint32_t maskwidth,
+                   const char *name);
+
+const int kIterations = 65536;
+
+// Add a macro layer since TEST_P will quote the name so we need to
+// expand it first with the prefix.
+#define MY_TEST_P(name, test) TEST_P(name, test)
+
+MY_TEST_P(ARCH_POSTFIX(V64_U8), TestIntrinsics) {
+  TestSimd1Arg<c_v64, uint8_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_U16), TestIntrinsics) {
+  TestSimd1Arg<c_v64, uint16_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_U32), TestIntrinsics) {
+  TestSimd1Arg<c_v64, uint32_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_V64), TestIntrinsics) {
+  TestSimd1Arg<c_v64, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U64_V64), TestIntrinsics) {
+  TestSimd1Arg<uint64_t, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(S64_V64), TestIntrinsics) {
+  TestSimd1Arg<int64_t, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U32_V64), TestIntrinsics) {
+  TestSimd1Arg<uint32_t, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(S32_V64), TestIntrinsics) {
+  TestSimd1Arg<int32_t, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_U32U32), TestIntrinsics) {
+  TestSimd2Args<c_v64, uint32_t, uint32_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_V64V64), TestIntrinsics) {
+  TestSimd2Args<c_v64, c_v64, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(S64_V64V64), TestIntrinsics) {
+  TestSimd2Args<int64_t, c_v64, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U32_V64V64), TestIntrinsics) {
+  TestSimd2Args<uint32_t, c_v64, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_V64U32), TestIntrinsics) {
+  TestSimd2Args<c_v64, c_v64, uint32_t>(kIterations, mask, maskwidth, name);
+}
+
+// Google Test allows up to 50 tests per case, so split the largest
+MY_TEST_P(ARCH_POSTFIX(V64_V64_Part2), TestIntrinsics) {
+  TestSimd1Arg<c_v64, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_V64V64_Part2), TestIntrinsics) {
+  TestSimd2Args<c_v64, c_v64, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U32_V128), TestIntrinsics) {
+  TestSimd1Arg<uint32_t, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U64_V128), TestIntrinsics) {
+  TestSimd1Arg<uint64_t, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_V128), TestIntrinsics) {
+  TestSimd1Arg<c_v64, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V128), TestIntrinsics) {
+  TestSimd1Arg<c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_U8), TestIntrinsics) {
+  TestSimd1Arg<c_v128, uint8_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_U16), TestIntrinsics) {
+  TestSimd1Arg<c_v128, uint16_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_U32), TestIntrinsics) {
+  TestSimd1Arg<c_v128, uint32_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_U64), TestIntrinsics) {
+  TestSimd1Arg<c_v128, uint64_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V64), TestIntrinsics) {
+  TestSimd1Arg<c_v128, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V128V128), TestIntrinsics) {
+  TestSimd2Args<c_v128, c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V128V128V128), TestIntrinsics) {
+  TestSimd3Args<c_v128, c_v128, c_v128, c_v128>(kIterations, mask, maskwidth,
+                                                name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U32_V128V128), TestIntrinsics) {
+  TestSimd2Args<uint32_t, c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U64_V128V128), TestIntrinsics) {
+  TestSimd2Args<uint64_t, c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(S64_V128V128), TestIntrinsics) {
+  TestSimd2Args<int64_t, c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_U64U64), TestIntrinsics) {
+  TestSimd2Args<c_v128, uint64_t, uint64_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V64V64), TestIntrinsics) {
+  TestSimd2Args<c_v128, c_v64, c_v64>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V128U32), TestIntrinsics) {
+  TestSimd2Args<c_v128, c_v128, uint32_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V128V128_Part2), TestIntrinsics) {
+  TestSimd2Args<c_v128, c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V128_Part2), TestIntrinsics) {
+  TestSimd1Arg<c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V128_Part3), TestIntrinsics) {
+  TestSimd1Arg<c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V128_V128_Part4), TestIntrinsics) {
+  TestSimd1Arg<c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U64_V256), TestIntrinsics) {
+  TestSimd1Arg<uint64_t, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256), TestIntrinsics) {
+  TestSimd1Arg<c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V128), TestIntrinsics) {
+  TestSimd1Arg<c_v256, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256V256), TestIntrinsics) {
+  TestSimd2Args<c_v256, c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256V256V256), TestIntrinsics) {
+  TestSimd3Args<c_v256, c_v256, c_v256, c_v256>(kIterations, mask, maskwidth,
+                                                name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V128V128), TestIntrinsics) {
+  TestSimd2Args<c_v256, c_v128, c_v128>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U32_V256V256), TestIntrinsics) {
+  TestSimd2Args<uint32_t, c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U64_V256V256), TestIntrinsics) {
+  TestSimd2Args<uint64_t, c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(S64_V256V256), TestIntrinsics) {
+  TestSimd2Args<int64_t, c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256V256_Part2), TestIntrinsics) {
+  TestSimd2Args<c_v256, c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256U32), TestIntrinsics) {
+  TestSimd2Args<c_v256, c_v256, uint32_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256_Part2), TestIntrinsics) {
+  TestSimd1Arg<c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256_Part3), TestIntrinsics) {
+  TestSimd1Arg<c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256_Part4), TestIntrinsics) {
+  TestSimd1Arg<c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_V256_Part5), TestIntrinsics) {
+  TestSimd1Arg<c_v256, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_U8), TestIntrinsics) {
+  TestSimd1Arg<c_v256, uint8_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_U16), TestIntrinsics) {
+  TestSimd1Arg<c_v256, uint16_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_U32), TestIntrinsics) {
+  TestSimd1Arg<c_v256, uint32_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V256_U64), TestIntrinsics) {
+  TestSimd1Arg<c_v256, uint64_t>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(U32_V256), TestIntrinsics) {
+  TestSimd1Arg<uint32_t, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+MY_TEST_P(ARCH_POSTFIX(V64_V256), TestIntrinsics) {
+  TestSimd1Arg<c_v64, c_v256>(kIterations, mask, maskwidth, name);
+}
+
+// Add a macro layer since INSTANTIATE_TEST_CASE_P will quote the name
+// so we need to expand it first with the prefix
+#define INSTANTIATE(name, type, ...) \
+  INSTANTIATE_TEST_CASE_P(name, type, ::testing::Values(__VA_ARGS__))
+
+#define SIMD_TUPLE(name, mask, maskwidth) \
+  ::testing::make_tuple(mask, maskwidth, static_cast<const char *>(#name))
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U32_V64V64),
+            (SIMD_TUPLE(v64_sad_u8, 0U, 0U), SIMD_TUPLE(v64_ssd_u8, 0U, 0U)));
+
+INSTANTIATE(
+    ARCH, ARCH_POSTFIX(V64_V64V64), SIMD_TUPLE(v64_add_8, 0U, 0U),
+    SIMD_TUPLE(v64_add_16, 0U, 0U), SIMD_TUPLE(v64_sadd_s16, 0U, 0U),
+    SIMD_TUPLE(v64_add_32, 0U, 0U), SIMD_TUPLE(v64_sub_8, 0U, 0U),
+    SIMD_TUPLE(v64_ssub_u8, 0U, 0U), SIMD_TUPLE(v64_ssub_s8, 0U, 0U),
+    SIMD_TUPLE(v64_sub_16, 0U, 0U), SIMD_TUPLE(v64_ssub_s16, 0U, 0U),
+    SIMD_TUPLE(v64_ssub_u16, 0U, 0U), SIMD_TUPLE(v64_sub_32, 0U, 0U),
+    SIMD_TUPLE(v64_ziplo_8, 0U, 0U), SIMD_TUPLE(v64_ziphi_8, 0U, 0U),
+    SIMD_TUPLE(v64_ziplo_16, 0U, 0U), SIMD_TUPLE(v64_ziphi_16, 0U, 0U),
+    SIMD_TUPLE(v64_ziplo_32, 0U, 0U), SIMD_TUPLE(v64_ziphi_32, 0U, 0U),
+    SIMD_TUPLE(v64_pack_s32_s16, 0U, 0U), SIMD_TUPLE(v64_pack_s16_u8, 0U, 0U),
+    SIMD_TUPLE(v64_pack_s16_s8, 0U, 0U), SIMD_TUPLE(v64_unziphi_8, 0U, 0U),
+    SIMD_TUPLE(v64_unziplo_8, 0U, 0U), SIMD_TUPLE(v64_unziphi_16, 0U, 0U),
+    SIMD_TUPLE(v64_unziplo_16, 0U, 0U), SIMD_TUPLE(v64_or, 0U, 0U),
+    SIMD_TUPLE(v64_xor, 0U, 0U), SIMD_TUPLE(v64_and, 0U, 0U),
+    SIMD_TUPLE(v64_andn, 0U, 0U), SIMD_TUPLE(v64_mullo_s16, 0U, 0U),
+    SIMD_TUPLE(v64_mulhi_s16, 0U, 0U), SIMD_TUPLE(v64_mullo_s32, 0U, 0U),
+    SIMD_TUPLE(v64_madd_s16, 0U, 0U), SIMD_TUPLE(v64_madd_us8, 0U, 0U),
+    SIMD_TUPLE(v64_avg_u8, 0U, 0U), SIMD_TUPLE(v64_rdavg_u8, 0U, 0U),
+    SIMD_TUPLE(v64_avg_u16, 0U, 0U), SIMD_TUPLE(v64_min_u8, 0U, 0U),
+    SIMD_TUPLE(v64_max_u8, 0U, 0U), SIMD_TUPLE(v64_min_s8, 0U, 0U),
+    SIMD_TUPLE(v64_max_s8, 0U, 0U), SIMD_TUPLE(v64_min_s16, 0U, 0U),
+    SIMD_TUPLE(v64_max_s16, 0U, 0U), SIMD_TUPLE(v64_cmpgt_s8, 0U, 0U),
+    SIMD_TUPLE(v64_cmplt_s8, 0U, 0U), SIMD_TUPLE(v64_cmpeq_8, 0U, 0U),
+    SIMD_TUPLE(v64_cmpgt_s16, 0U, 0U), SIMD_TUPLE(v64_cmplt_s16, 0U, 0U),
+    SIMD_TUPLE(v64_cmpeq_16, 0U, 0U));
+
+INSTANTIATE(
+    ARCH, ARCH_POSTFIX(V64_V64V64_Part2), SIMD_TUPLE(v64_shuffle_8, 7U, 8U),
+    SIMD_TUPLE(v64_pack_s32_u16, 0U, 0U), SIMD_TUPLE(v64_rdavg_u16, 0U, 0U),
+    SIMD_TUPLE(v64_sadd_s8, 0U, 0U), SIMD_TUPLE(v64_sadd_u8, 0U, 0U),
+    SIMD_TUPLE(imm_v64_align<1>, 0U, 0U), SIMD_TUPLE(imm_v64_align<2>, 0U, 0U),
+    SIMD_TUPLE(imm_v64_align<3>, 0U, 0U), SIMD_TUPLE(imm_v64_align<4>, 0U, 0U),
+    SIMD_TUPLE(imm_v64_align<5>, 0U, 0U), SIMD_TUPLE(imm_v64_align<6>, 0U, 0U),
+    SIMD_TUPLE(imm_v64_align<7>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_V64), SIMD_TUPLE(v64_abs_s8, 0U, 0U),
+            SIMD_TUPLE(v64_abs_s16, 0U, 0U),
+            SIMD_TUPLE(v64_unpacklo_u8_s16, 0U, 0U),
+            SIMD_TUPLE(v64_unpackhi_u8_s16, 0U, 0U),
+            SIMD_TUPLE(v64_unpacklo_s8_s16, 0U, 0U),
+            SIMD_TUPLE(v64_unpackhi_s8_s16, 0U, 0U),
+            SIMD_TUPLE(v64_unpacklo_u16_s32, 0U, 0U),
+            SIMD_TUPLE(v64_unpacklo_s16_s32, 0U, 0U),
+            SIMD_TUPLE(v64_unpackhi_u16_s32, 0U, 0U),
+            SIMD_TUPLE(v64_unpackhi_s16_s32, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_byte<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_byte<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_byte<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_byte<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_byte<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_byte<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_byte<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_byte<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_byte<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_byte<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_byte<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_byte<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_byte<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_byte<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_8<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u8<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s8<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_16<8>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_V64_Part2),
+            SIMD_TUPLE(imm_v64_shl_n_16<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u16<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u16<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s16<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s16<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_32<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shl_n_32<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u32<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_u32<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s32<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v64_shr_n_s32<28>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_V64U32), SIMD_TUPLE(v64_shl_8, 7U, 32U),
+            SIMD_TUPLE(v64_shr_u8, 7U, 32U), SIMD_TUPLE(v64_shr_s8, 7U, 32U),
+            SIMD_TUPLE(v64_shl_16, 15U, 32U), SIMD_TUPLE(v64_shr_u16, 15U, 32U),
+            SIMD_TUPLE(v64_shr_s16, 15U, 32U), SIMD_TUPLE(v64_shl_32, 31U, 32U),
+            SIMD_TUPLE(v64_shr_u32, 31U, 32U),
+            SIMD_TUPLE(v64_shr_s32, 31U, 32U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U64_V64), SIMD_TUPLE(v64_hadd_u8, 0U, 0U),
+            SIMD_TUPLE(v64_u64, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(S64_V64), SIMD_TUPLE(v64_hadd_s16, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U32_V64), SIMD_TUPLE(v64_low_u32, 0U, 0U),
+            SIMD_TUPLE(v64_high_u32, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(S32_V64), SIMD_TUPLE(v64_low_s32, 0U, 0U),
+            SIMD_TUPLE(v64_high_s32, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(S64_V64V64), SIMD_TUPLE(v64_dotp_s16, 0U, 0U),
+            SIMD_TUPLE(v64_dotp_su8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_U8), SIMD_TUPLE(v64_dup_8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_U16), SIMD_TUPLE(v64_dup_16, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_U32), SIMD_TUPLE(v64_dup_32, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_U32U32), SIMD_TUPLE(v64_from_32, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U32_V128V128), SIMD_TUPLE(v128_sad_u8, 0U, 0U),
+            SIMD_TUPLE(v128_ssd_u8, 0U, 0U), SIMD_TUPLE(v128_sad_u16, 0U, 0U));
+INSTANTIATE(ARCH, ARCH_POSTFIX(U64_V128V128), SIMD_TUPLE(v128_ssd_s16, 0U, 0U));
+
+INSTANTIATE(
+    ARCH, ARCH_POSTFIX(V128_V128V128), SIMD_TUPLE(v128_add_8, 0U, 0U),
+    SIMD_TUPLE(v128_add_16, 0U, 0U), SIMD_TUPLE(v128_sadd_s16, 0U, 0U),
+    SIMD_TUPLE(v128_add_32, 0U, 0U), SIMD_TUPLE(v128_sub_8, 0U, 0U),
+    SIMD_TUPLE(v128_ssub_u8, 0U, 0U), SIMD_TUPLE(v128_ssub_s8, 0U, 0U),
+    SIMD_TUPLE(v128_sub_16, 0U, 0U), SIMD_TUPLE(v128_ssub_s16, 0U, 0U),
+    SIMD_TUPLE(v128_ssub_u16, 0U, 0U), SIMD_TUPLE(v128_sub_32, 0U, 0U),
+    SIMD_TUPLE(v128_ziplo_8, 0U, 0U), SIMD_TUPLE(v128_ziphi_8, 0U, 0U),
+    SIMD_TUPLE(v128_ziplo_16, 0U, 0U), SIMD_TUPLE(v128_ziphi_16, 0U, 0U),
+    SIMD_TUPLE(v128_ziplo_32, 0U, 0U), SIMD_TUPLE(v128_ziphi_32, 0U, 0U),
+    SIMD_TUPLE(v128_ziplo_64, 0U, 0U), SIMD_TUPLE(v128_ziphi_64, 0U, 0U),
+    SIMD_TUPLE(v128_unziphi_8, 0U, 0U), SIMD_TUPLE(v128_unziplo_8, 0U, 0U),
+    SIMD_TUPLE(v128_unziphi_16, 0U, 0U), SIMD_TUPLE(v128_unziplo_16, 0U, 0U),
+    SIMD_TUPLE(v128_unziphi_32, 0U, 0U), SIMD_TUPLE(v128_unziplo_32, 0U, 0U),
+    SIMD_TUPLE(v128_pack_s32_s16, 0U, 0U), SIMD_TUPLE(v128_pack_s16_u8, 0U, 0U),
+    SIMD_TUPLE(v128_pack_s16_s8, 0U, 0U), SIMD_TUPLE(v128_or, 0U, 0U),
+    SIMD_TUPLE(v128_xor, 0U, 0U), SIMD_TUPLE(v128_and, 0U, 0U),
+    SIMD_TUPLE(v128_andn, 0U, 0U), SIMD_TUPLE(v128_mullo_s16, 0U, 0U),
+    SIMD_TUPLE(v128_mulhi_s16, 0U, 0U), SIMD_TUPLE(v128_mullo_s32, 0U, 0U),
+    SIMD_TUPLE(v128_madd_s16, 0U, 0U), SIMD_TUPLE(v128_madd_us8, 0U, 0U),
+    SIMD_TUPLE(v128_avg_u8, 0U, 0U), SIMD_TUPLE(v128_rdavg_u8, 0U, 0U),
+    SIMD_TUPLE(v128_avg_u16, 0U, 0U), SIMD_TUPLE(v128_min_u8, 0U, 0U),
+    SIMD_TUPLE(v128_max_u8, 0U, 0U), SIMD_TUPLE(v128_min_s8, 0U, 0U),
+    SIMD_TUPLE(v128_max_s8, 0U, 0U), SIMD_TUPLE(v128_min_s16, 0U, 0U),
+    SIMD_TUPLE(v128_max_s16, 0U, 0U), SIMD_TUPLE(v128_cmpgt_s8, 0U, 0U),
+    SIMD_TUPLE(v128_cmplt_s8, 0U, 0U), SIMD_TUPLE(v128_cmpeq_8, 0U, 0U),
+    SIMD_TUPLE(v128_cmpgt_s16, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_V128V128_Part2),
+            SIMD_TUPLE(v128_pack_s32_u16, 0U, 0U),
+            SIMD_TUPLE(v128_rdavg_u16, 0U, 0U), SIMD_TUPLE(v128_add_64, 0U, 0U),
+            SIMD_TUPLE(v128_sub_64, 0U, 0U), SIMD_TUPLE(v128_sadd_s8, 0U, 0U),
+            SIMD_TUPLE(v128_sadd_u8, 0U, 0U), SIMD_TUPLE(v128_cmpeq_16, 0U, 0U),
+            SIMD_TUPLE(v128_cmplt_s16, 0U, 0U),
+            SIMD_TUPLE(v128_cmplt_s32, 0U, 0U),
+            SIMD_TUPLE(v128_cmpeq_32, 0U, 0U),
+            SIMD_TUPLE(v128_cmpgt_s32, 0U, 0U),
+            SIMD_TUPLE(v128_shuffle_8, 15U, 8U),
+            SIMD_TUPLE(v128_min_s32, 0U, 0U), SIMD_TUPLE(v128_max_s32, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<9>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<11>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<13>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_align<15>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_V128V128V128),
+            SIMD_TUPLE(v128_blend_8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_V128), SIMD_TUPLE(v128_abs_s8, 0U, 0U),
+            SIMD_TUPLE(v128_abs_s16, 0U, 0U), SIMD_TUPLE(v128_padd_s16, 0U, 0U),
+            SIMD_TUPLE(v128_unpacklo_u8_s16, 0U, 0U),
+            SIMD_TUPLE(v128_unpacklo_s8_s16, 0U, 0U),
+            SIMD_TUPLE(v128_unpacklo_u16_s32, 0U, 0U),
+            SIMD_TUPLE(v128_unpacklo_s16_s32, 0U, 0U),
+            SIMD_TUPLE(v128_unpackhi_u8_s16, 0U, 0U),
+            SIMD_TUPLE(v128_unpackhi_s8_s16, 0U, 0U),
+            SIMD_TUPLE(v128_unpackhi_u16_s32, 0U, 0U),
+            SIMD_TUPLE(v128_unpackhi_s16_s32, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<9>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<11>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<13>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_byte<15>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<9>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<11>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<13>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_byte<15>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_8<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u8<1>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_V128_Part2),
+            SIMD_TUPLE(imm_v128_shr_n_u8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s8<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_16<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_16<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u16<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u16<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s16<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s16<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_32<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_32<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u32<4>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_V128_Part3),
+            SIMD_TUPLE(imm_v128_shr_n_u32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u32<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s32<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s32<28>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_V128_Part4),
+            SIMD_TUPLE(imm_v128_shl_n_64<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<32>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<36>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<40>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<44>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<48>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<52>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<56>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shl_n_64<60>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<32>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<36>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<40>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<44>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<48>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<52>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<56>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_u64<60>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<32>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<36>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<40>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<44>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<48>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<52>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<56>, 0U, 0U),
+            SIMD_TUPLE(imm_v128_shr_n_s64<60>, 0U, 0U),
+            SIMD_TUPLE(v128_padd_u8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_V64V64), SIMD_TUPLE(v128_from_v64, 0U, 0U),
+            SIMD_TUPLE(v128_zip_8, 0U, 0U), SIMD_TUPLE(v128_zip_16, 0U, 0U),
+            SIMD_TUPLE(v128_zip_32, 0U, 0U), SIMD_TUPLE(v128_mul_s16, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_U64U64), SIMD_TUPLE(v128_from_64, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_V64),
+            SIMD_TUPLE(v128_unpack_u8_s16, 0U, 0U),
+            SIMD_TUPLE(v128_unpack_s8_s16, 0U, 0U),
+            SIMD_TUPLE(v128_unpack_u16_s32, 0U, 0U),
+            SIMD_TUPLE(v128_unpack_s16_s32, 0U, 0U));
+
+INSTANTIATE(
+    ARCH, ARCH_POSTFIX(V128_V128U32), SIMD_TUPLE(v128_shl_8, 7U, 32U),
+    SIMD_TUPLE(v128_shr_u8, 7U, 32U), SIMD_TUPLE(v128_shr_s8, 7U, 32U),
+    SIMD_TUPLE(v128_shl_16, 15U, 32U), SIMD_TUPLE(v128_shr_u16, 15U, 32U),
+    SIMD_TUPLE(v128_shr_s16, 15U, 32U), SIMD_TUPLE(v128_shl_32, 31U, 32U),
+    SIMD_TUPLE(v128_shr_u32, 31U, 32U), SIMD_TUPLE(v128_shr_s32, 31U, 32U),
+    SIMD_TUPLE(v128_shl_64, 63U, 32U), SIMD_TUPLE(v128_shr_u64, 63U, 32U),
+    SIMD_TUPLE(v128_shr_s64, 63U, 32U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U32_V128), SIMD_TUPLE(v128_low_u32, 0U, 0U),
+            SIMD_TUPLE(v128_movemask_8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U64_V128), SIMD_TUPLE(v128_hadd_u8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_V128), SIMD_TUPLE(v128_low_v64, 0U, 0U),
+            SIMD_TUPLE(v128_high_v64, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_U8), SIMD_TUPLE(v128_dup_8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_U16), SIMD_TUPLE(v128_dup_16, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_U32), SIMD_TUPLE(v128_dup_32, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V128_U64), SIMD_TUPLE(v128_dup_64, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(S64_V128V128), SIMD_TUPLE(v128_dotp_s16, 0U, 0U),
+            SIMD_TUPLE(v128_dotp_s32, 0U, 0U),
+            SIMD_TUPLE(v128_dotp_su8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U32_V256V256), SIMD_TUPLE(v256_sad_u8, 0U, 0U),
+            SIMD_TUPLE(v256_ssd_u8, 0U, 0U), SIMD_TUPLE(v256_sad_u16, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U64_V256), SIMD_TUPLE(v256_hadd_u8, 0U, 0U),
+            SIMD_TUPLE(v256_low_u64, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(S64_V256V256), SIMD_TUPLE(v256_dotp_s16, 0U, 0U),
+            SIMD_TUPLE(v256_dotp_s32, 0U, 0U),
+            SIMD_TUPLE(v256_dotp_su8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U64_V256V256), SIMD_TUPLE(v256_ssd_s16, 0U, 0U));
+
+INSTANTIATE(
+    ARCH, ARCH_POSTFIX(V256_V256V256), SIMD_TUPLE(v256_add_8, 0U, 0U),
+    SIMD_TUPLE(v256_add_16, 0U, 0U), SIMD_TUPLE(v256_sadd_s16, 0U, 0U),
+    SIMD_TUPLE(v256_add_32, 0U, 0U), SIMD_TUPLE(v256_sub_8, 0U, 0U),
+    SIMD_TUPLE(v256_ssub_u8, 0U, 0U), SIMD_TUPLE(v256_ssub_s8, 0U, 0U),
+    SIMD_TUPLE(v256_sub_16, 0U, 0U), SIMD_TUPLE(v256_ssub_s16, 0U, 0U),
+    SIMD_TUPLE(v256_ssub_u16, 0U, 0U), SIMD_TUPLE(v256_sub_32, 0U, 0U),
+    SIMD_TUPLE(v256_ziplo_8, 0U, 0U), SIMD_TUPLE(v256_ziphi_8, 0U, 0U),
+    SIMD_TUPLE(v256_ziplo_16, 0U, 0U), SIMD_TUPLE(v256_ziphi_16, 0U, 0U),
+    SIMD_TUPLE(v256_ziplo_32, 0U, 0U), SIMD_TUPLE(v256_ziphi_32, 0U, 0U),
+    SIMD_TUPLE(v256_ziplo_64, 0U, 0U), SIMD_TUPLE(v256_ziphi_64, 0U, 0U),
+    SIMD_TUPLE(v256_ziplo_128, 0U, 0U), SIMD_TUPLE(v256_ziphi_128, 0U, 0U),
+    SIMD_TUPLE(v256_unziphi_8, 0U, 0U), SIMD_TUPLE(v256_unziplo_8, 0U, 0U),
+    SIMD_TUPLE(v256_unziphi_16, 0U, 0U), SIMD_TUPLE(v256_unziplo_16, 0U, 0U),
+    SIMD_TUPLE(v256_unziphi_32, 0U, 0U), SIMD_TUPLE(v256_unziplo_32, 0U, 0U),
+    SIMD_TUPLE(v256_pack_s32_s16, 0U, 0U), SIMD_TUPLE(v256_pack_s16_u8, 0U, 0U),
+    SIMD_TUPLE(v256_pack_s16_s8, 0U, 0U), SIMD_TUPLE(v256_or, 0U, 0U),
+    SIMD_TUPLE(v256_xor, 0U, 0U), SIMD_TUPLE(v256_and, 0U, 0U),
+    SIMD_TUPLE(v256_andn, 0U, 0U), SIMD_TUPLE(v256_mullo_s16, 0U, 0U),
+    SIMD_TUPLE(v256_mulhi_s16, 0U, 0U), SIMD_TUPLE(v256_mullo_s32, 0U, 0U),
+    SIMD_TUPLE(v256_madd_s16, 0U, 0U), SIMD_TUPLE(v256_madd_us8, 0U, 0U),
+    SIMD_TUPLE(v256_avg_u8, 0U, 0U), SIMD_TUPLE(v256_rdavg_u8, 0U, 0U),
+    SIMD_TUPLE(v256_avg_u16, 0U, 0U), SIMD_TUPLE(v256_min_u8, 0U, 0U),
+    SIMD_TUPLE(v256_max_u8, 0U, 0U), SIMD_TUPLE(v256_min_s8, 0U, 0U),
+    SIMD_TUPLE(v256_max_s8, 0U, 0U), SIMD_TUPLE(v256_min_s16, 0U, 0U),
+    SIMD_TUPLE(v256_max_s16, 0U, 0U), SIMD_TUPLE(v256_cmpgt_s8, 0U, 0U),
+    SIMD_TUPLE(v256_cmplt_s8, 0U, 0U));
+
+INSTANTIATE(
+    ARCH, ARCH_POSTFIX(V256_V256V256_Part2), SIMD_TUPLE(v256_cmpeq_8, 0U, 0U),
+    SIMD_TUPLE(v256_min_s32, 0U, 0U), SIMD_TUPLE(v256_max_s32, 0U, 0U),
+    SIMD_TUPLE(v256_add_64, 0U, 0U), SIMD_TUPLE(v256_sub_64, 0U, 0U),
+    SIMD_TUPLE(v256_cmpgt_s16, 0U, 0U), SIMD_TUPLE(v256_cmplt_s16, 0U, 0U),
+    SIMD_TUPLE(v256_cmpeq_16, 0U, 0U), SIMD_TUPLE(v256_cmpgt_s32, 0U, 0U),
+    SIMD_TUPLE(v256_cmplt_s32, 0U, 0U), SIMD_TUPLE(v256_cmpeq_32, 0U, 0U),
+    SIMD_TUPLE(v256_shuffle_8, 31U, 8U), SIMD_TUPLE(v256_pshuffle_8, 15U, 8U),
+    SIMD_TUPLE(imm_v256_align<1>, 0U, 0U), SIMD_TUPLE(v256_sadd_s8, 0U, 0U),
+    SIMD_TUPLE(v256_sadd_u8, 0U, 0U), SIMD_TUPLE(v256_pack_s32_u16, 0U, 0U),
+    SIMD_TUPLE(v256_rdavg_u16, 0U, 0U), SIMD_TUPLE(imm_v256_align<2>, 0U, 0U),
+    SIMD_TUPLE(v256_unziphi_64, 0U, 0U), SIMD_TUPLE(v256_unziplo_64, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<3>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<4>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<5>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<6>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<7>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<8>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<9>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<10>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<11>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<12>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<13>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<14>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<15>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<16>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<17>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<18>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<19>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<20>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<21>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<22>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<23>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<24>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<25>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<26>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<27>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<28>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<29>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<30>, 0U, 0U),
+    SIMD_TUPLE(imm_v256_align<31>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_V128V128),
+            SIMD_TUPLE(v256_from_v128, 0U, 0U), SIMD_TUPLE(v256_zip_8, 0U, 0U),
+            SIMD_TUPLE(v256_zip_16, 0U, 0U), SIMD_TUPLE(v256_zip_32, 0U, 0U),
+            SIMD_TUPLE(v256_mul_s16, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_V128),
+            SIMD_TUPLE(v256_unpack_u8_s16, 0U, 0U),
+            SIMD_TUPLE(v256_unpack_s8_s16, 0U, 0U),
+            SIMD_TUPLE(v256_unpack_u16_s32, 0U, 0U),
+            SIMD_TUPLE(v256_unpack_s16_s32, 0U, 0U));
+
+INSTANTIATE(
+    ARCH, ARCH_POSTFIX(V256_V256U32), SIMD_TUPLE(v256_shl_8, 7U, 32U),
+    SIMD_TUPLE(v256_shr_u8, 7U, 32U), SIMD_TUPLE(v256_shr_s8, 7U, 32U),
+    SIMD_TUPLE(v256_shl_16, 15U, 32U), SIMD_TUPLE(v256_shr_u16, 15U, 32U),
+    SIMD_TUPLE(v256_shr_s16, 15U, 32U), SIMD_TUPLE(v256_shl_32, 31U, 32U),
+    SIMD_TUPLE(v256_shr_u32, 31U, 32U), SIMD_TUPLE(v256_shr_s32, 31U, 32U),
+    SIMD_TUPLE(v256_shl_64, 63U, 32U), SIMD_TUPLE(v256_shr_u64, 63U, 32U),
+    SIMD_TUPLE(v256_shr_s64, 63U, 32U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_V256), SIMD_TUPLE(v256_abs_s8, 0U, 0U),
+            SIMD_TUPLE(v256_abs_s16, 0U, 0U), SIMD_TUPLE(v256_padd_s16, 0U, 0U),
+            SIMD_TUPLE(v256_unpacklo_u8_s16, 0U, 0U),
+            SIMD_TUPLE(v256_unpacklo_s8_s16, 0U, 0U),
+            SIMD_TUPLE(v256_unpacklo_u16_s32, 0U, 0U),
+            SIMD_TUPLE(v256_unpacklo_s16_s32, 0U, 0U),
+            SIMD_TUPLE(v256_unpackhi_u8_s16, 0U, 0U),
+            SIMD_TUPLE(v256_unpackhi_s8_s16, 0U, 0U),
+            SIMD_TUPLE(v256_unpackhi_u16_s32, 0U, 0U),
+            SIMD_TUPLE(v256_unpackhi_s16_s32, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<9>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<11>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<13>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<15>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<17>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<18>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<19>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<21>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<22>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<23>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<25>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<26>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<27>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<29>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<30>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_byte<31>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<8>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_V256_Part2),
+            SIMD_TUPLE(imm_v256_shl_n_byte<9>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<11>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<13>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<15>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<17>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<18>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<19>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<21>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<22>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<23>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<25>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<26>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<27>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<29>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<30>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_byte<31>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_8<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u8<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s8<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s8<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s8<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s8<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s8<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s8<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s8<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_16<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_16<10>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_V256_Part3),
+            SIMD_TUPLE(imm_v256_shl_n_16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u16<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u16<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s16<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s16<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s16<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s16<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s16<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s16<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s16<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s16<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_32<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_32<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u32<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u32<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s32<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s32<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s32<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s32<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s32<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s32<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s32<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s32<28>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_V256_Part4),
+            SIMD_TUPLE(imm_v256_shl_n_64<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<32>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<36>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<40>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<44>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<48>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<52>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<56>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_64<60>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<32>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<36>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<40>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<44>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<48>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<52>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<56>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_u64<60>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<16>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<20>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<24>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<28>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<32>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<36>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<40>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<44>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<48>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<52>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<56>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_s64<60>, 0U, 0U),
+            SIMD_TUPLE(v256_padd_u8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_V256_Part5),
+            SIMD_TUPLE(imm_v256_shr_n_word<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<9>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<11>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<13>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shr_n_word<15>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<1>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<2>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<3>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<4>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<5>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<6>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<7>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<8>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<9>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<10>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<11>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<12>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<13>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<14>, 0U, 0U),
+            SIMD_TUPLE(imm_v256_shl_n_word<15>, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_V256V256V256),
+            SIMD_TUPLE(v256_blend_8, 0U, 0U),
+            SIMD_TUPLE(v256_wideshuffle_8, 63U, 8U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_U8), SIMD_TUPLE(v256_dup_8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_U16), SIMD_TUPLE(v256_dup_16, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_U32), SIMD_TUPLE(v256_dup_32, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V256_U64), SIMD_TUPLE(v256_dup_64, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(U32_V256), SIMD_TUPLE(v256_low_u32, 0U, 0U),
+            SIMD_TUPLE(v256_movemask_8, 0U, 0U));
+
+INSTANTIATE(ARCH, ARCH_POSTFIX(V64_V256), SIMD_TUPLE(v256_low_v64, 0U, 0U));
+
+}  // namespace SIMD_NAMESPACE
diff --git a/third_party/aom/test/simd_neon_test.cc b/third_party/aom/test/simd_neon_test.cc
new file mode 100644
index 0000000000..b67b188959
--- /dev/null
+++ b/third_party/aom/test/simd_neon_test.cc
@@ -0,0 +1,17 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#if defined(__OPTIMIZE__) && __OPTIMIZE__
+#define ARCH NEON
+#define ARCH_POSTFIX(name) name##_neon
+#define SIMD_NAMESPACE simd_test_neon
+#include "test/simd_impl.h"
+#endif
diff --git a/third_party/aom/test/simd_sse2_test.cc b/third_party/aom/test/simd_sse2_test.cc
new file mode 100644
index 0000000000..b37a931b38
--- /dev/null
+++ b/third_party/aom/test/simd_sse2_test.cc
@@ -0,0 +1,18 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#if (defined(__OPTIMIZE__) && __OPTIMIZE__) || \
+    (!defined(__GNUC__) && !defined(_DEBUG))
+#define ARCH SSE2
+#define ARCH_POSTFIX(name) name##_sse2
+#define SIMD_NAMESPACE simd_test_sse2
+#include "test/simd_impl.h"
+#endif
diff --git a/third_party/aom/test/simd_sse4_test.cc b/third_party/aom/test/simd_sse4_test.cc
new file mode 100644
index 0000000000..b1c9d5cd88
--- /dev/null
+++ b/third_party/aom/test/simd_sse4_test.cc
@@ -0,0 +1,18 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#if (defined(__OPTIMIZE__) && __OPTIMIZE__) || \
+    (!defined(__GNUC__) && !defined(_DEBUG))
+#define ARCH SSE4_1
+#define ARCH_POSTFIX(name) name##_sse4_1
+#define SIMD_NAMESPACE simd_test_sse4_1
+#include "test/simd_impl.h"
+#endif
diff --git a/third_party/aom/test/simd_ssse3_test.cc b/third_party/aom/test/simd_ssse3_test.cc
new file mode 100644
index 0000000000..d95c26fb5e
--- /dev/null
+++ b/third_party/aom/test/simd_ssse3_test.cc
@@ -0,0 +1,18 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#if (defined(__OPTIMIZE__) && __OPTIMIZE__) || \
+    (!defined(__GNUC__) && !defined(_DEBUG))
+#define ARCH SSSE3
+#define ARCH_POSTFIX(name) name##_ssse3
+#define SIMD_NAMESPACE simd_test_ssse3
+#include "test/simd_impl.h"
+#endif
diff --git a/third_party/aom/test/simple_decoder.sh b/third_party/aom/test/simple_decoder.sh
new file mode 100755
index 0000000000..5f39ad206e
--- /dev/null
+++ b/third_party/aom/test/simple_decoder.sh
@@ -0,0 +1,58 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the libaom simple_decoder example code. To add new tests to
+## this file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to simple_decoder_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: Make sure input is available:
+simple_decoder_verify_environment() {
+  if [ ! "$(av1_encode_available)" = "yes" ] && [ ! -e "${AV1_IVF_FILE}" ]; then
+    return 1
+  fi
+}
+
+# Runs simple_decoder using $1 as input file. $2 is the codec name, and is used
+# solely to name the output file.
+simple_decoder() {
+  local decoder="$(aom_tool_path simple_decoder)"
+  local input_file="$1"
+  local codec="$2"
+  local output_file="${AOM_TEST_OUTPUT_DIR}/simple_decoder_${codec}.raw"
+
+  if [ ! -x "${decoder}" ]; then
+    elog "${decoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${decoder}" "${input_file}" "${output_file}" \
+      ${devnull}
+
+  [ -e "${output_file}" ] || return 1
+}
+
+simple_decoder_av1() {
+  if [ "$(av1_decode_available)" = "yes" ]; then
+    if [ ! -e "${AV1_IVF_FILE}" ]; then
+      local file="${AOM_TEST_OUTPUT_DIR}/test_encode.ivf"
+      encode_yuv_raw_input_av1 "${file}" --ivf
+      simple_decoder "${file}" av1 || return 1
+    else
+      simple_decoder "${AV1_IVF_FILE}" av1 || return 1
+    fi
+  fi
+}
+
+simple_decoder_tests="simple_decoder_av1"
+
+run_tests simple_decoder_verify_environment "${simple_decoder_tests}"
diff --git a/third_party/aom/test/simple_encoder.sh b/third_party/aom/test/simple_encoder.sh
new file mode 100755
index 0000000000..5cd6b46a10
--- /dev/null
+++ b/third_party/aom/test/simple_encoder.sh
@@ -0,0 +1,53 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the libaom simple_encoder example. To add new tests to this
+## file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to simple_encoder_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: $YUV_RAW_INPUT is required.
+simple_encoder_verify_environment() {
+  if [ ! -e "${YUV_RAW_INPUT}" ]; then
+    echo "Libaom test data must exist in LIBAOM_TEST_DATA_PATH."
+    return 1
+  fi
+}
+
+# Runs simple_encoder using the codec specified by $1 with a frame limit of 100.
+simple_encoder() {
+  local encoder="${LIBAOM_BIN_PATH}/simple_encoder${AOM_TEST_EXE_SUFFIX}"
+  local codec="$1"
+  local output_file="${AOM_TEST_OUTPUT_DIR}/simple_encoder_${codec}.ivf"
+
+  if [ ! -x "${encoder}" ]; then
+    elog "${encoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${encoder}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
+      "${YUV_RAW_INPUT_HEIGHT}" "${YUV_RAW_INPUT}" "${output_file}" 9999 0 5 \
+      ${devnull}
+
+  [ -e "${output_file}" ] || return 1
+}
+
+
+simple_encoder_av1() {
+  if [ "$(av1_encode_available)" = "yes" ]; then
+    simple_encoder av1 || return 1
+  fi
+}
+
+simple_encoder_tests="simple_encoder_av1"
+
+run_tests simple_encoder_verify_environment "${simple_encoder_tests}"
diff --git a/third_party/aom/test/subtract_test.cc b/third_party/aom/test/subtract_test.cc
new file mode 100644
index 0000000000..7dcedf56de
--- /dev/null
+++ b/third_party/aom/test/subtract_test.cc
@@ -0,0 +1,249 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "av1/common/blockd.h"
+#include "aom_mem/aom_mem.h"
+#include "aom_ports/mem.h"
+
+typedef void (*SubtractFunc)(int rows, int cols, int16_t *diff_ptr,
+                             ptrdiff_t diff_stride, const uint8_t *src_ptr,
+                             ptrdiff_t src_stride, const uint8_t *pred_ptr,
+                             ptrdiff_t pred_stride);
+
+namespace {
+
+class AV1SubtractBlockTest : public ::testing::TestWithParam<SubtractFunc> {
+ public:
+  virtual void TearDown() { libaom_test::ClearSystemState(); }
+};
+
+using libaom_test::ACMRandom;
+
+TEST_P(AV1SubtractBlockTest, SimpleSubtract) {
+  ACMRandom rnd(ACMRandom::DeterministicSeed());
+
+  // FIXME(rbultje) split in its own file
+  for (BLOCK_SIZE bsize = BLOCK_4X4; bsize < BLOCK_SIZES;
+       bsize = static_cast<BLOCK_SIZE>(static_cast<int>(bsize) + 1)) {
+    const int block_width = block_size_wide[bsize];
+    const int block_height = block_size_high[bsize];
+    int16_t *diff = reinterpret_cast<int16_t *>(
+        aom_memalign(16, sizeof(*diff) * block_width * block_height * 2));
+    uint8_t *pred = reinterpret_cast<uint8_t *>(
+        aom_memalign(16, block_width * block_height * 2));
+    uint8_t *src = reinterpret_cast<uint8_t *>(
+        aom_memalign(16, block_width * block_height * 2));
+
+    for (int n = 0; n < 100; n++) {
+      for (int r = 0; r < block_height; ++r) {
+        for (int c = 0; c < block_width * 2; ++c) {
+          src[r * block_width * 2 + c] = rnd.Rand8();
+          pred[r * block_width * 2 + c] = rnd.Rand8();
+        }
+      }
+
+      GetParam()(block_height, block_width, diff, block_width, src, block_width,
+                 pred, block_width);
+
+      for (int r = 0; r < block_height; ++r) {
+        for (int c = 0; c < block_width; ++c) {
+          EXPECT_EQ(diff[r * block_width + c],
+                    (src[r * block_width + c] - pred[r * block_width + c]))
+              << "r = " << r << ", c = " << c << ", bs = " << bsize;
+        }
+      }
+
+      GetParam()(block_height, block_width, diff, block_width * 2, src,
+                 block_width * 2, pred, block_width * 2);
+
+      for (int r = 0; r < block_height; ++r) {
+        for (int c = 0; c < block_width; ++c) {
+          EXPECT_EQ(
+              diff[r * block_width * 2 + c],
+              (src[r * block_width * 2 + c] - pred[r * block_width * 2 + c]))
+              << "r = " << r << ", c = " << c << ", bs = " << bsize;
+        }
+      }
+    }
+    aom_free(diff);
+    aom_free(pred);
+    aom_free(src);
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(C, AV1SubtractBlockTest,
+                        ::testing::Values(aom_subtract_block_c));
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, AV1SubtractBlockTest,
+                        ::testing::Values(aom_subtract_block_sse2));
+#endif
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, AV1SubtractBlockTest,
+                        ::testing::Values(aom_subtract_block_neon));
+#endif
+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(MSA, AV1SubtractBlockTest,
+                        ::testing::Values(aom_subtract_block_msa));
+#endif
+
+typedef void (*HBDSubtractFunc)(int rows, int cols, int16_t *diff_ptr,
+                                ptrdiff_t diff_stride, const uint8_t *src_ptr,
+                                ptrdiff_t src_stride, const uint8_t *pred_ptr,
+                                ptrdiff_t pred_stride, int bd);
+
+using ::testing::get;
+using ::testing::make_tuple;
+using ::testing::tuple;
+
+// <width, height, bit_dpeth, subtract>
+typedef tuple<int, int, int, HBDSubtractFunc> Params;
+
+class AV1HBDSubtractBlockTest : public ::testing::TestWithParam<Params> {
+ public:
+  virtual void SetUp() {
+    block_width_ = GET_PARAM(0);
+    block_height_ = GET_PARAM(1);
+    bit_depth_ = static_cast<aom_bit_depth_t>(GET_PARAM(2));
+    func_ = GET_PARAM(3);
+
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+
+    const size_t max_width = 128;
+    const size_t max_block_size = max_width * max_width;
+    src_ = CONVERT_TO_BYTEPTR(reinterpret_cast<uint16_t *>(
+        aom_memalign(16, max_block_size * sizeof(uint16_t))));
+    pred_ = CONVERT_TO_BYTEPTR(reinterpret_cast<uint16_t *>(
+        aom_memalign(16, max_block_size * sizeof(uint16_t))));
+    diff_ = reinterpret_cast<int16_t *>(
+        aom_memalign(16, max_block_size * sizeof(int16_t)));
+  }
+
+  virtual void TearDown() {
+    aom_free(CONVERT_TO_SHORTPTR(src_));
+    aom_free(CONVERT_TO_SHORTPTR(pred_));
+    aom_free(diff_);
+  }
+
+ protected:
+  void CheckResult();
+  void RunForSpeed();
+
+ private:
+  ACMRandom rnd_;
+  int block_height_;
+  int block_width_;
+  aom_bit_depth_t bit_depth_;
+  HBDSubtractFunc func_;
+  uint8_t *src_;
+  uint8_t *pred_;
+  int16_t *diff_;
+};
+
+void AV1HBDSubtractBlockTest::CheckResult() {
+  const int test_num = 100;
+  const size_t max_width = 128;
+  const int max_block_size = max_width * max_width;
+  const int mask = (1 << bit_depth_) - 1;
+  int i, j;
+
+  for (i = 0; i < test_num; ++i) {
+    for (j = 0; j < max_block_size; ++j) {
+      CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask;
+      CONVERT_TO_SHORTPTR(pred_)[j] = rnd_.Rand16() & mask;
+    }
+
+    func_(block_height_, block_width_, diff_, block_width_, src_, block_width_,
+          pred_, block_width_, bit_depth_);
+
+    for (int r = 0; r < block_height_; ++r) {
+      for (int c = 0; c < block_width_; ++c) {
+        EXPECT_EQ(diff_[r * block_width_ + c],
+                  (CONVERT_TO_SHORTPTR(src_)[r * block_width_ + c] -
+                   CONVERT_TO_SHORTPTR(pred_)[r * block_width_ + c]))
+            << "r = " << r << ", c = " << c << ", test: " << i;
+      }
+    }
+  }
+}
+
+TEST_P(AV1HBDSubtractBlockTest, CheckResult) { CheckResult(); }
+
+void AV1HBDSubtractBlockTest::RunForSpeed() {
+  const int test_num = 200000;
+  const size_t max_width = 128;
+  const int max_block_size = max_width * max_width;
+  const int mask = (1 << bit_depth_) - 1;
+  int i, j;
+
+  for (j = 0; j < max_block_size; ++j) {
+    CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask;
+    CONVERT_TO_SHORTPTR(pred_)[j] = rnd_.Rand16() & mask;
+  }
+
+  for (i = 0; i < test_num; ++i) {
+    func_(block_height_, block_width_, diff_, block_width_, src_, block_width_,
+          pred_, block_width_, bit_depth_);
+  }
+}
+
+TEST_P(AV1HBDSubtractBlockTest, DISABLED_Speed) { RunForSpeed(); }
+
+#if HAVE_SSE2
+
+const Params kAV1HBDSubtractBlock_sse2[] = {
+  make_tuple(4, 4, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(4, 4, 12, &aom_highbd_subtract_block_c),
+  make_tuple(4, 8, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(4, 8, 12, &aom_highbd_subtract_block_c),
+  make_tuple(8, 4, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(8, 4, 12, &aom_highbd_subtract_block_c),
+  make_tuple(8, 8, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(8, 8, 12, &aom_highbd_subtract_block_c),
+  make_tuple(8, 16, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(8, 16, 12, &aom_highbd_subtract_block_c),
+  make_tuple(16, 8, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(16, 8, 12, &aom_highbd_subtract_block_c),
+  make_tuple(16, 16, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(16, 16, 12, &aom_highbd_subtract_block_c),
+  make_tuple(16, 32, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(16, 32, 12, &aom_highbd_subtract_block_c),
+  make_tuple(32, 16, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(32, 16, 12, &aom_highbd_subtract_block_c),
+  make_tuple(32, 32, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(32, 32, 12, &aom_highbd_subtract_block_c),
+  make_tuple(32, 64, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(32, 64, 12, &aom_highbd_subtract_block_c),
+  make_tuple(64, 32, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(64, 32, 12, &aom_highbd_subtract_block_c),
+  make_tuple(64, 64, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(64, 64, 12, &aom_highbd_subtract_block_c),
+  make_tuple(64, 128, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(64, 128, 12, &aom_highbd_subtract_block_c),
+  make_tuple(128, 64, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(128, 64, 12, &aom_highbd_subtract_block_c),
+  make_tuple(128, 128, 12, &aom_highbd_subtract_block_sse2),
+  make_tuple(128, 128, 12, &aom_highbd_subtract_block_c)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, AV1HBDSubtractBlockTest,
+                        ::testing::ValuesIn(kAV1HBDSubtractBlock_sse2));
+#endif  // HAVE_SSE2
+}  // namespace
diff --git a/third_party/aom/test/sum_squares_test.cc b/third_party/aom/test/sum_squares_test.cc
new file mode 100644
index 0000000000..f109984986
--- /dev/null
+++ b/third_party/aom/test/sum_squares_test.cc
@@ -0,0 +1,228 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cmath>
+#include <cstdlib>
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom_ports/mem.h"
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "test/util.h"
+#include "test/function_equivalence_test.h"
+
+using libaom_test::ACMRandom;
+using libaom_test::FunctionEquivalenceTest;
+
+namespace {
+const int kNumIterations = 10000;
+
+static const int16_t kInt13Max = (1 << 12) - 1;
+
+typedef uint64_t (*SSI16Func)(const int16_t *src, int stride, int width,
+                              int height);
+typedef libaom_test::FuncParam<SSI16Func> TestFuncs;
+
+class SumSquaresTest : public ::testing::TestWithParam<TestFuncs> {
+ public:
+  virtual ~SumSquaresTest() {}
+  virtual void SetUp() {
+    params_ = this->GetParam();
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+    src_ = reinterpret_cast<int16_t *>(aom_memalign(16, 256 * 256 * 2));
+    ASSERT_TRUE(src_ != NULL);
+  }
+
+  virtual void TearDown() {
+    libaom_test::ClearSystemState();
+    aom_free(src_);
+  }
+  void RunTest(int isRandom);
+  void RunSpeedTest();
+
+  void GenRandomData(int width, int height, int stride) {
+    const int msb = 11;  // Up to 12 bit input
+    const int limit = 1 << (msb + 1);
+    for (int ii = 0; ii < height; ii++) {
+      for (int jj = 0; jj < width; jj++) {
+        src_[ii * stride + jj] = rnd_(2) ? rnd_(limit) : -rnd_(limit);
+      }
+    }
+  }
+
+  void GenExtremeData(int width, int height, int stride) {
+    const int msb = 11;  // Up to 12 bit input
+    const int limit = 1 << (msb + 1);
+    const int val = rnd_(2) ? limit - 1 : -(limit - 1);
+    for (int ii = 0; ii < height; ii++) {
+      for (int jj = 0; jj < width; jj++) {
+        src_[ii * stride + jj] = val;
+      }
+    }
+  }
+
+ protected:
+  TestFuncs params_;
+  int16_t *src_;
+  ACMRandom rnd_;
+};
+
+void SumSquaresTest::RunTest(int isRandom) {
+  int failed = 0;
+  for (int k = 0; k < kNumIterations; k++) {
+    const int width = 4 * (rnd_(31) + 1);   // Up to 128x128
+    const int height = 4 * (rnd_(31) + 1);  // Up to 128x128
+    int stride = 4 << rnd_(7);              // Up to 256 stride
+    while (stride < width) {                // Make sure it's valid
+      stride = 4 << rnd_(7);
+    }
+    if (isRandom) {
+      GenRandomData(width, height, stride);
+    } else {
+      GenExtremeData(width, height, stride);
+    }
+    const uint64_t res_ref = params_.ref_func(src_, stride, width, height);
+    uint64_t res_tst;
+    ASM_REGISTER_STATE_CHECK(res_tst =
+                                 params_.tst_func(src_, stride, width, height));
+
+    if (!failed) {
+      failed = res_ref != res_tst;
+      EXPECT_EQ(res_ref, res_tst)
+          << "Error: Sum Squares Test [" << width << "x" << height
+          << "] C output does not match optimized output.";
+    }
+  }
+}
+
+void SumSquaresTest::RunSpeedTest() {
+  for (int block = BLOCK_4X4; block < BLOCK_SIZES_ALL; block++) {
+    const int width = block_size_wide[block];   // Up to 128x128
+    const int height = block_size_high[block];  // Up to 128x128
+    int stride = 4 << rnd_(7);                  // Up to 256 stride
+    while (stride < width) {                    // Make sure it's valid
+      stride = 4 << rnd_(7);
+    }
+    GenExtremeData(width, height, stride);
+    const int num_loops = 1000000000 / (width + height);
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+
+    for (int i = 0; i < num_loops; ++i)
+      params_.ref_func(src_, stride, width, height);
+
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+    printf("SumSquaresTest C %3dx%-3d: %7.2f ns\n", width, height,
+           1000.0 * elapsed_time / num_loops);
+
+    aom_usec_timer timer1;
+    aom_usec_timer_start(&timer1);
+    for (int i = 0; i < num_loops; ++i)
+      params_.tst_func(src_, stride, width, height);
+    aom_usec_timer_mark(&timer1);
+    const int elapsed_time1 = static_cast<int>(aom_usec_timer_elapsed(&timer1));
+    printf("SumSquaresTest Test %3dx%-3d: %7.2f ns\n", width, height,
+           1000.0 * elapsed_time1 / num_loops);
+  }
+}
+
+TEST_P(SumSquaresTest, OperationCheck) {
+  RunTest(1);  // GenRandomData
+}
+
+TEST_P(SumSquaresTest, ExtremeValues) {
+  RunTest(0);  // GenExtremeData
+}
+
+TEST_P(SumSquaresTest, DISABLED_Speed) { RunSpeedTest(); }
+
+#if HAVE_SSE2
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2, SumSquaresTest,
+    ::testing::Values(TestFuncs(&aom_sum_squares_2d_i16_c,
+                                &aom_sum_squares_2d_i16_sse2)));
+
+#endif  // HAVE_SSE2
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(
+    AVX2, SumSquaresTest,
+    ::testing::Values(TestFuncs(&aom_sum_squares_2d_i16_c,
+                                &aom_sum_squares_2d_i16_avx2)));
+#endif  // HAVE_AVX2
+
+//////////////////////////////////////////////////////////////////////////////
+// 1D version
+//////////////////////////////////////////////////////////////////////////////
+
+typedef uint64_t (*F1D)(const int16_t *src, uint32_t N);
+typedef libaom_test::FuncParam<F1D> TestFuncs1D;
+
+class SumSquares1DTest : public FunctionEquivalenceTest<F1D> {
+ protected:
+  static const int kIterations = 1000;
+  static const int kMaxSize = 256;
+};
+
+TEST_P(SumSquares1DTest, RandomValues) {
+  DECLARE_ALIGNED(16, int16_t, src[kMaxSize * kMaxSize]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < kMaxSize * kMaxSize; ++i)
+      src[i] = rng_(kInt13Max * 2 + 1) - kInt13Max;
+
+    const int N = rng_(2) ? rng_(kMaxSize * kMaxSize + 1 - kMaxSize) + kMaxSize
+                          : rng_(kMaxSize) + 1;
+
+    const uint64_t ref_res = params_.ref_func(src, N);
+    uint64_t tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(src, N));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+TEST_P(SumSquares1DTest, ExtremeValues) {
+  DECLARE_ALIGNED(16, int16_t, src[kMaxSize * kMaxSize]);
+
+  for (int iter = 0; iter < kIterations && !HasFatalFailure(); ++iter) {
+    if (rng_(2)) {
+      for (int i = 0; i < kMaxSize * kMaxSize; ++i) src[i] = kInt13Max;
+    } else {
+      for (int i = 0; i < kMaxSize * kMaxSize; ++i) src[i] = -kInt13Max;
+    }
+
+    const int N = rng_(2) ? rng_(kMaxSize * kMaxSize + 1 - kMaxSize) + kMaxSize
+                          : rng_(kMaxSize) + 1;
+
+    const uint64_t ref_res = params_.ref_func(src, N);
+    uint64_t tst_res;
+    ASM_REGISTER_STATE_CHECK(tst_res = params_.tst_func(src, N));
+
+    ASSERT_EQ(ref_res, tst_res);
+  }
+}
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, SumSquares1DTest,
+                        ::testing::Values(TestFuncs1D(
+                            aom_sum_squares_i16_c, aom_sum_squares_i16_sse2)));
+
+#endif  // HAVE_SSE2
+}  // namespace
diff --git a/third_party/aom/test/superframe_test.cc b/third_party/aom/test/superframe_test.cc
new file mode 100644
index 0000000000..7be18f72a3
--- /dev/null
+++ b/third_party/aom/test/superframe_test.cc
@@ -0,0 +1,109 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <climits>
+#include <vector>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+
+namespace {
+
+const int kTestMode = 0;
+const int kTileCols = 1;
+const int kTileRows = 2;
+
+typedef ::testing::tuple<libaom_test::TestMode, int, int> SuperframeTestParam;
+
+class SuperframeTest
+    : public ::libaom_test::CodecTestWithParam<SuperframeTestParam>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  SuperframeTest() : EncoderTest(GET_PARAM(0)), last_sf_pts_(0) {}
+  virtual ~SuperframeTest() {}
+
+  virtual void SetUp() {
+    InitializeConfig();
+    const SuperframeTestParam input = GET_PARAM(1);
+    const libaom_test::TestMode mode = ::testing::get<kTestMode>(input);
+    SetMode(mode);
+    sf_count_ = 0;
+    sf_count_max_ = INT_MAX;
+    n_tile_cols_ = ::testing::get<kTileCols>(input);
+    n_tile_rows_ = ::testing::get<kTileRows>(input);
+  }
+
+  virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
+                                  libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AOME_SET_ENABLEAUTOALTREF, 1);
+      encoder->Control(AOME_SET_CPUUSED, 2);
+      encoder->Control(AV1E_SET_TILE_COLUMNS, n_tile_cols_);
+      encoder->Control(AV1E_SET_TILE_ROWS, n_tile_rows_);
+    }
+  }
+
+  virtual const aom_codec_cx_pkt_t *MutateEncoderOutputHook(
+      const aom_codec_cx_pkt_t *pkt) {
+    if (pkt->kind != AOM_CODEC_CX_FRAME_PKT) return pkt;
+
+    const uint8_t *buffer = reinterpret_cast<uint8_t *>(pkt->data.frame.buf);
+    const uint8_t marker = buffer[0];
+    const int frames = (marker & 0x7) + 1;
+    const int mag = ((marker >> 3) & 3) + 1;
+    const unsigned int index_sz = 2 + mag * (frames - 1);
+    if ((marker & 0xe0) == 0xc0 && pkt->data.frame.sz >= index_sz &&
+        buffer[index_sz - 1] == marker) {
+      // frame is a superframe. strip off the index.
+      modified_buf_.resize(pkt->data.frame.sz - index_sz);
+      memcpy(&modified_buf_[0], (uint8_t *)pkt->data.frame.buf + index_sz,
+             pkt->data.frame.sz - index_sz);
+      modified_pkt_ = *pkt;
+      modified_pkt_.data.frame.buf = &modified_buf_[0];
+      modified_pkt_.data.frame.sz -= index_sz;
+
+      sf_count_++;
+      last_sf_pts_ = pkt->data.frame.pts;
+      return &modified_pkt_;
+    }
+
+    // Make sure we do a few frames after the last SF
+    abort_ |=
+        sf_count_ > sf_count_max_ && pkt->data.frame.pts - last_sf_pts_ >= 5;
+    return pkt;
+  }
+
+  int sf_count_;
+  int sf_count_max_;
+  aom_codec_cx_pkt_t modified_pkt_;
+  std::vector<uint8_t> modified_buf_;
+  aom_codec_pts_t last_sf_pts_;
+
+ private:
+  int n_tile_cols_;
+  int n_tile_rows_;
+};
+
+TEST_P(SuperframeTest, TestSuperframeIndexIsOptional) {
+  sf_count_max_ = 0;  // early exit on successful test.
+  cfg_.g_lag_in_frames = 25;
+  cfg_.large_scale_tile = 1;
+  ::libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 352, 288,
+                                       30, 1, 0, 40);
+  ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+  // NOTE: The use of BWDREF_FRAME will enable the coding of more non-show
+  //       frames besides ALTREF_FRAME.
+  EXPECT_GE(sf_count_, 1);
+}
+
+}  // namespace
diff --git a/third_party/aom/test/test-data.sha1 b/third_party/aom/test/test-data.sha1
new file mode 100644
index 0000000000..b6ee34701a
--- /dev/null
+++ b/third_party/aom/test/test-data.sha1
@@ -0,0 +1,507 @@
+d5dfb0151c9051f8c85999255645d7a23916d3c0 *hantro_collage_w352h288.yuv
+b87815bf86020c592ccc7a846ba2e28ec8043902 *hantro_odd.yuv
+26b7f64399b84db4b4c9c915d743ec5c2619d4b9 *invalid-bug-1814.ivf
+d3964f9dad9f60363c81b688324d95b4ec7c8038 *invalid-bug-1814.ivf.res
+fa06784f23751d8c37be94160fb821e855199af4 *invalid-oss-fuzz-10061.ivf
+b055f06b9a95aaa5697fa26497b592a47843a7c8 *invalid-oss-fuzz-10061.ivf.res
+c9e06c4c7fb7d69fd635a1f606a5e478d60e99cf *invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf
+88e18e61bd2b7457b4c71ebefbdff0029c41cc04 *invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf.res
+91a5bedeb4832c1c2900736cc0f644bb63971bbc *invalid-oss-fuzz-10227.ivf
+b055f06b9a95aaa5697fa26497b592a47843a7c8 *invalid-oss-fuzz-10227.ivf.res
+c0960f032484579f967881cc025b71cfd7a79ee1 *invalid-oss-fuzz-9463.ivf
+d3964f9dad9f60363c81b688324d95b4ec7c8038 *invalid-oss-fuzz-9463.ivf.res
+f448caf378e250b7eea4fa2d1c3cd7ef4a3211ce *invalid-oss-fuzz-9482.ivf
+b055f06b9a95aaa5697fa26497b592a47843a7c8 *invalid-oss-fuzz-9482.ivf.res
+a686989de79af89136f631fd630df639c7861851 *invalid-oss-fuzz-9720.ivf
+d3964f9dad9f60363c81b688324d95b4ec7c8038 *invalid-oss-fuzz-9720.ivf.res
+a432f96ff0a787268e2f94a8092ab161a18d1b06 *park_joy_90p_10_420.y4m
+0b194cc312c3a2e84d156a221b0a5eb615dfddc5 *park_joy_90p_10_422.y4m
+ff0e0a21dc2adc95b8c1b37902713700655ced17 *park_joy_90p_10_444.y4m
+c934da6fb8cc54ee2a8c17c54cf6076dac37ead0 *park_joy_90p_10_440.yuv
+614c32ae1eca391e867c70d19974f0d62664dd99 *park_joy_90p_12_420.y4m
+c92825f1ea25c5c37855083a69faac6ac4641a9e *park_joy_90p_12_422.y4m
+b592189b885b6cc85db55cc98512a197d73d3b34 *park_joy_90p_12_444.y4m
+82c1bfcca368c2f22bad7d693d690d5499ecdd11 *park_joy_90p_12_440.yuv
+b9e1e90aece2be6e2c90d89e6ab2372d5f8c792d *park_joy_90p_8_420_a10-1.y4m
+4e0eb61e76f0684188d9bc9f3ce61f6b6b77bb2c *park_joy_90p_8_420.y4m
+7a193ff7dfeb96ba5f82b2afd7afa9e1fe83d947 *park_joy_90p_8_422.y4m
+bdb7856e6bc93599bdda05c2e773a9f22b6c6d03 *park_joy_90p_8_444.y4m
+81e1f3843748438b8f2e71db484eb22daf72e939 *park_joy_90p_8_440.yuv
+b1f1c3ec79114b9a0651af24ce634afb44a9a419 *rush_hour_444.y4m
+eb438c6540eb429f74404eedfa3228d409c57874 *desktop_640_360_30.yuv
+89e70ebd22c27d275fe14dc2f1a41841a6d8b9ab *kirland_640_480_30.yuv
+33c533192759e5bb4f07abfbac389dc259db4686 *macmarcomoving_640_480_30.yuv
+8bfaab121080821b8f03b23467911e59ec59b8fe *macmarcostationary_640_480_30.yuv
+70894878d916a599842d9ad0dcd24e10c13e5467 *niklas_640_480_30.yuv
+8784b6df2d8cc946195a90ac00540500d2e522e4 *tacomanarrows_640_480_30.yuv
+edd86a1f5e62fd9da9a9d46078247759c2638009 *tacomasmallcameramovement_640_480_30.yuv
+9a70e8b7d14fba9234d0e51dce876635413ce444 *thaloundeskmtg_640_480_30.yuv
+e7d315dbf4f3928779e0dc624311196d44491d32 *niklas_1280_720_30.yuv
+717da707afcaa1f692ff1946f291054eb75a4f06 *screendata.y4m
+9cfc855459e7549fd015c79e8eca512b2f2cb7e3 *niklas_1280_720_30.y4m
+5b5763b388b1b52a81bb82b39f7ec25c4bd3d0e1 *desktop_credits.y4m
+36ddab9b99eb7545aa0bf362d6f498212d596516 *vase10x10.yuv
+c2e1ec9936b95254187a359e94aa32a9f3dad1b7 *av1-1-b8-00-quantizer-00.ivf
+26cd2a0321d01d9db5f6dace8b43a40cd5b9d58d *av1-1-b8-00-quantizer-00.ivf.md5
+a56dd02c0258d4afea1ee358a22b54e99e39d5e1 *av1-1-b8-00-quantizer-01.ivf
+b3d24124d81f1fbb26f5eb0036accb54f3ec69b2 *av1-1-b8-00-quantizer-01.ivf.md5
+3466327cb842a91d69839b11ef930a74f086f4c6 *av1-1-b8-00-quantizer-02.ivf
+c111dce946100efeaad34203080eee1d55464df6 *av1-1-b8-00-quantizer-02.ivf.md5
+d3f1f32de5e2c0c19a58bb8ef096108388c6a820 *av1-1-b8-00-quantizer-03.ivf
+6265321b31130545b4454982ca93e412a56845b8 *av1-1-b8-00-quantizer-03.ivf.md5
+f37c393ebe73266a5ec8508a2ca33c586ff28e64 *av1-1-b8-00-quantizer-04.ivf
+c6e979da71aecc593c0abb40135dd304152b00dd *av1-1-b8-00-quantizer-04.ivf.md5
+ac9c5e93cb19942a9be259d0567ec96c54dcdc7c *av1-1-b8-00-quantizer-05.ivf
+49e35a7399568a0e4f015ce323d5a45ea780ca87 *av1-1-b8-00-quantizer-05.ivf.md5
+461142b1b50ae74c6b698d23f5ed3b764eadfb89 *av1-1-b8-00-quantizer-06.ivf
+6477ff260624e0f76c94ac872d1e7d5576af4177 *av1-1-b8-00-quantizer-06.ivf.md5
+7f8113cd13d8faaa06fdbaaa50dc328daf037e6d *av1-1-b8-00-quantizer-07.ivf
+b26795c6cb408487c20737977cd6b77311772bf7 *av1-1-b8-00-quantizer-07.ivf.md5
+4218f7945a172e1fe4f9e77ec35085a394eda9f4 *av1-1-b8-00-quantizer-08.ivf
+ea5d7d501e9a69d805251e4871515d28468d8676 *av1-1-b8-00-quantizer-08.ivf.md5
+837f3bcadfe56cf302db2ebaf9a990446fb35801 *av1-1-b8-00-quantizer-09.ivf
+eede995cdac5fd01a411da2e74e86e8394138be1 *av1-1-b8-00-quantizer-09.ivf.md5
+adc229b3780a4968c18ded1bcbe72e3f04643833 *av1-1-b8-00-quantizer-10.ivf
+0799b7e54e54ee97bf0e8aad2b75509ce59c7097 *av1-1-b8-00-quantizer-10.ivf.md5
+44bac8247160a8d9a0ab19f890fc89cc9298de1d *av1-1-b8-00-quantizer-11.ivf
+cc6b2bf167e114599b242aba574e8c6f1fa2f047 *av1-1-b8-00-quantizer-11.ivf.md5
+ebb3af7dfc15567188bcb617021cdc95ebc560e3 *av1-1-b8-00-quantizer-12.ivf
+b716ae29d56cd0c052dbfa1b5dcf850cd0fa8ca7 *av1-1-b8-00-quantizer-12.ivf.md5
+46159641f981a26fb9c374a5ca41e44f0ce0a9f0 *av1-1-b8-00-quantizer-13.ivf
+c6db1b8b4a74f83e4a0647e053cea0fc00f6abab *av1-1-b8-00-quantizer-13.ivf.md5
+fadc909d18eb640760fbb075f922fb050e715470 *av1-1-b8-00-quantizer-14.ivf
+e36bb6b23273633ba3ef7d28160a7258840a1476 *av1-1-b8-00-quantizer-14.ivf.md5
+8befbd9cc1601dcd36ec6911613855f68e6fd40e *av1-1-b8-00-quantizer-15.ivf
+cfc2334b76fb5e7aa9d8607e89d37cbc7716d62e *av1-1-b8-00-quantizer-15.ivf.md5
+ca42e00ae27c6b7f684fe3d2a787d50d2827cb3f *av1-1-b8-00-quantizer-16.ivf
+f11278218a7c3c73cfaab2332bab55f06cedcc81 *av1-1-b8-00-quantizer-16.ivf.md5
+05270d365bdc067f9446eda3029a6f41571a5229 *av1-1-b8-00-quantizer-17.ivf
+fb6482f35e7ad04bf231ea1806226760abcb3c26 *av1-1-b8-00-quantizer-17.ivf.md5
+617bc72037165efbff478d5a0d342b3c20ffcafd *av1-1-b8-00-quantizer-18.ivf
+1ff68d5424f91322123fe0d58f436b8e49cfa99d *av1-1-b8-00-quantizer-18.ivf.md5
+821c3b1ae6054c7a91b2f64428806e57f1157ca6 *av1-1-b8-00-quantizer-19.ivf
+f2fd118e786697553d6987f786660a2bb9f00680 *av1-1-b8-00-quantizer-19.ivf.md5
+48bcf17c27d9a4eb73632a68c09f42eff9f9af99 *av1-1-b8-00-quantizer-20.ivf
+64d55e4c858414bc2837c9c3e2d5fb6d2208c4b8 *av1-1-b8-00-quantizer-20.ivf.md5
+d61ecdd4f0950bc5c8bae1270b22e711bdd22763 *av1-1-b8-00-quantizer-21.ivf
+9d447938596096704fd5f4d41bcdf6fabf9cdfb9 *av1-1-b8-00-quantizer-21.ivf.md5
+59b4b65d8e56ccdd1bddff26a03e991a63409334 *av1-1-b8-00-quantizer-22.ivf
+aa1be0c7c7622d612af85f9bf96a212f6fe5ab56 *av1-1-b8-00-quantizer-22.ivf.md5
+95ed96988eb9916cad956db9b929718769de49f1 *av1-1-b8-00-quantizer-23.ivf
+596b8a3aea468996d609624367465c412751f52b *av1-1-b8-00-quantizer-23.ivf.md5
+e6c2dc4ce725003152797b3d7b34d7eb34da50c8 *av1-1-b8-00-quantizer-24.ivf
+1cd3d7e8b3813a9e5591b94eaeb72d471780e64a *av1-1-b8-00-quantizer-24.ivf.md5
+6734e353008824e523939d1a18daa3f2ab2d8ec6 *av1-1-b8-00-quantizer-25.ivf
+c45cf440a05802c1f9e29472175ed397d130d988 *av1-1-b8-00-quantizer-25.ivf.md5
+3372b1c69fb39811156adcea4f6dba802c0918c2 *av1-1-b8-00-quantizer-26.ivf
+b1751d55bb3fb788751fe28fb7434bee153bda68 *av1-1-b8-00-quantizer-26.ivf.md5
+e7ddb19a6e2a798d6a4e7dfdfc10b4df777b60e3 *av1-1-b8-00-quantizer-27.ivf
+0e19d6b79cd71de69d03e0455349568af979b170 *av1-1-b8-00-quantizer-27.ivf.md5
+7f1c90a35543d6b673e353b3702baf3aa1caeaa7 *av1-1-b8-00-quantizer-28.ivf
+d9a4f9cb88103249a05a7e6aa616bf0c16bf9c95 *av1-1-b8-00-quantizer-28.ivf.md5
+28d741b923011c7fcc50a7318256a638d3110a07 *av1-1-b8-00-quantizer-29.ivf
+c68cacf2b2ff2694945a99ad836dcf1ee3961c09 *av1-1-b8-00-quantizer-29.ivf.md5
+9a5d9ea4bc76dd40d04e92f33f45e9c2e120e85d *av1-1-b8-00-quantizer-30.ivf
+eb02bb8c16c4c0368ddff83e05e516e84ec9eaf3 *av1-1-b8-00-quantizer-30.ivf.md5
+20193c372f44f522e094c2c05fc7e4aaa0717fa8 *av1-1-b8-00-quantizer-31.ivf
+a4c1a4ac332f4911f0d5abbd826ebecfb8432d6c *av1-1-b8-00-quantizer-31.ivf.md5
+9617bbd691f093d259dbc8a642a57a153c1fc00c *av1-1-b8-00-quantizer-32.ivf
+73d60a348454b126ea6368ea604954bc23f210ae *av1-1-b8-00-quantizer-32.ivf.md5
+d9aea9d72a686c59b60584d827f60ca1ee8eee26 *av1-1-b8-00-quantizer-33.ivf
+fbf64de376a63d2d3051da83b0e4e56579b55c0a *av1-1-b8-00-quantizer-33.ivf.md5
+791aaf067f125e5cf4a247cf06a2e29ab071ec90 *av1-1-b8-00-quantizer-34.ivf
+8e2e6efe4c069e54844da19125c4280b95990c69 *av1-1-b8-00-quantizer-34.ivf.md5
+01ba67bba5cbf7c94c65da8f4c9bd6e7db24cf3a *av1-1-b8-00-quantizer-35.ivf
+0c5e60704a4a6bd27e67b6fd72ca7d2cf7fff50f *av1-1-b8-00-quantizer-35.ivf.md5
+3e255b4a320c9522dcec539fef770b6920b9a102 *av1-1-b8-00-quantizer-36.ivf
+1241aab865fd7b4bae73736cbeec1866ea9c90ec *av1-1-b8-00-quantizer-36.ivf.md5
+44fa6fca109747d8f43f6c6aa46d782e5d476d54 *av1-1-b8-00-quantizer-37.ivf
+947f0f887c5ac9149cf85e8114a709d6f410fc32 *av1-1-b8-00-quantizer-37.ivf.md5
+8319ac1ddd6ce3279da5780175dff7a3a5fa1054 *av1-1-b8-00-quantizer-38.ivf
+5f571b7f88678eab9e54f162cc9898f14e437770 *av1-1-b8-00-quantizer-38.ivf.md5
+5975e7056e17608593a8c40619b68e6576d373d9 *av1-1-b8-00-quantizer-39.ivf
+7c870192d6eb70ce5367147a3d2c6a52e11f7bec *av1-1-b8-00-quantizer-39.ivf.md5
+47da942f1e455f1422fc65f06dd57304541d16ac *av1-1-b8-00-quantizer-40.ivf
+6ea7116c9ce3a1641c7060bab2f5e06fd0910d61 *av1-1-b8-00-quantizer-40.ivf.md5
+ab35c15dfde21c2572b14e04dbfd5fac1adae449 *av1-1-b8-00-quantizer-41.ivf
+19596f9849653b913186b9d6b7072984ede96177 *av1-1-b8-00-quantizer-41.ivf.md5
+23a5fa6c3d0eaffaf13f6402465f5dd33d8ea7f1 *av1-1-b8-00-quantizer-42.ivf
+5a2726f0d1b1799d4f70883f1bfe5c9d976c6cf5 *av1-1-b8-00-quantizer-42.ivf.md5
+86cddfc463d2b186ec5a1aa25c4562c05201e3c3 *av1-1-b8-00-quantizer-43.ivf
+674c64ec8487ee774ad09350380fa6ac43815807 *av1-1-b8-00-quantizer-43.ivf.md5
+6894c154eb56c4f3fe44d54fc4f9af468b03d175 *av1-1-b8-00-quantizer-44.ivf
+eca679a2781eb894d18b3d578e3aaf4f48019a15 *av1-1-b8-00-quantizer-44.ivf.md5
+0960bf018ada4224b8344519cf091850d50a57bd *av1-1-b8-00-quantizer-45.ivf
+291bb43b9e1ab167040b51019daf1ccf94fd1e50 *av1-1-b8-00-quantizer-45.ivf.md5
+ea644a4732f1a2534332802c2fa5073344f3c356 *av1-1-b8-00-quantizer-46.ivf
+4c7915382b1d6d08709c95525b04ab8830f20ca1 *av1-1-b8-00-quantizer-46.ivf.md5
+d1f8832d33234e2c74a2280090850153ea24ea82 *av1-1-b8-00-quantizer-47.ivf
+90eb9959e612602934dcc512fe6f54abf0c88d9c *av1-1-b8-00-quantizer-47.ivf.md5
+69c93f760e8b666eb5b98f510e09d90f9230ac9b *av1-1-b8-00-quantizer-48.ivf
+931f869e14bd455de9dac2101b383c29e7d6f04c *av1-1-b8-00-quantizer-48.ivf.md5
+8b660c577d95c031d6711c1134b8d115097f8d7e *av1-1-b8-00-quantizer-49.ivf
+0e3fe8b49d497050dc1a0eac5f3ad60f5fe068fe *av1-1-b8-00-quantizer-49.ivf.md5
+d40bb21448a6da0fc9b88cbcf76d2f4226573acb *av1-1-b8-00-quantizer-50.ivf
+bcd2a9c9a021ba44fc5dc74ae02194fe49ca76a4 *av1-1-b8-00-quantizer-50.ivf.md5
+3b5a1d464aa89b0f1a6ad4f5a03602292b826172 *av1-1-b8-00-quantizer-51.ivf
+49bcde0c56cf8b7fbe429336981be22d39025b74 *av1-1-b8-00-quantizer-51.ivf.md5
+38970a02fb38ddb4954fe4240164cb75de5fc744 *av1-1-b8-00-quantizer-52.ivf
+fd02b034d79d4be150efb02bd4349edfd0e41311 *av1-1-b8-00-quantizer-52.ivf.md5
+2fde7a7cf3014d5196d011c47de4a144227ed122 *av1-1-b8-00-quantizer-53.ivf
+0cb66e6d8fbb29962a69ae1703e22da50db2c92b *av1-1-b8-00-quantizer-53.ivf.md5
+89a69e9b9a601e40cb491ac3a1d32491f2468ac8 *av1-1-b8-00-quantizer-54.ivf
+2f8af51acc73c99b5af81db2bdd1883b611ad311 *av1-1-b8-00-quantizer-54.ivf.md5
+31ee4f56fcb0043e95fff7af49e4ef82aafa5543 *av1-1-b8-00-quantizer-55.ivf
+04a7104e02bdd0fa38c118202dbbecdbd11ace02 *av1-1-b8-00-quantizer-55.ivf.md5
+f262f0b234006a2652fceb77b1a8711aa53abb54 *av1-1-b8-00-quantizer-56.ivf
+bdd54dc25bc5a147c76163af0bced45c56435d79 *av1-1-b8-00-quantizer-56.ivf.md5
+1ef00617091db4b2b839de623bd6b4fb0b2f5f83 *av1-1-b8-00-quantizer-57.ivf
+714c65363a87ed5e6e4ad75c79ddb6af57d41fd9 *av1-1-b8-00-quantizer-57.ivf.md5
+43c9b02feccbb3c709d96015f126b7e3d4c24c64 *av1-1-b8-00-quantizer-58.ivf
+bae22b8d6377862bff8219470c0d87205d186a68 *av1-1-b8-00-quantizer-58.ivf.md5
+ca5f780abe4c02e48cceb9c804f3625723c359bf *av1-1-b8-00-quantizer-59.ivf
+c60a20bbf60b0b0a442ef3f7b682979053909d6e *av1-1-b8-00-quantizer-59.ivf.md5
+1f6f047e9f0e1da22fb514370d92c3c7c66dcf89 *av1-1-b8-00-quantizer-60.ivf
+86dc7fa59d363cf1ae4b027a57b119bda893c1c1 *av1-1-b8-00-quantizer-60.ivf.md5
+bcf0c3353568c47a043f2dc34c9abd3fc04eebd4 *av1-1-b8-00-quantizer-61.ivf
+66fc4f729c5915aa19939d1b6e28e5b398e747bb *av1-1-b8-00-quantizer-61.ivf.md5
+ac8d3c54451b52cf557ef435d33e7638088d66df *av1-1-b8-00-quantizer-62.ivf
+b57f4e1276ead626a3662339a86111ae6fda49d2 *av1-1-b8-00-quantizer-62.ivf.md5
+2a8aa33513d8e01ae9410c4bf5fe1e471b775482 *av1-1-b8-00-quantizer-63.ivf
+9f646ec35a168f495e144c64ba7ce9aeb41cd0a2 *av1-1-b8-00-quantizer-63.ivf.md5
+838388fbda4a1d91be81ff62694c3bf13c460d38 *av1-1-b8-01-size-16x16.ivf
+4229c1caf8e25eb3073456fb90ceed206753901e *av1-1-b8-01-size-16x16.ivf.md5
+23f4253bf71e02b2e8ead66da4b3de875e879ef2 *av1-1-b8-01-size-18x16.ivf
+af125644436d4b6897dade68336cedad663b6610 *av1-1-b8-01-size-18x16.ivf.md5
+94e4a75bd93052f79998e9e08e6b5dd73dc27e50 *av1-1-b8-01-size-32x16.ivf
+e7b3fbc5e4b2469838e7ae36512bd3ce0a81040c *av1-1-b8-01-size-32x16.ivf.md5
+f297bde01c05ec5c07ff8118a0280bd36c52b246 *av1-1-b8-01-size-34x16.ivf
+f6bbd94d6063c689de3c7cf94afa2c68b969d12c *av1-1-b8-01-size-34x16.ivf.md5
+1e18bdf68bab7e7282aacc77e423bc7d93d04a8e *av1-1-b8-01-size-64x16.ivf
+de75732fccfb385294b23c17f0f1a57b455edcf7 *av1-1-b8-01-size-64x16.ivf.md5
+26b1f6ae80b161e971468085778cc1ece502b330 *av1-1-b8-01-size-66x16.ivf
+48bd99813557c314d398e6952da78da07c79d416 *av1-1-b8-01-size-66x16.ivf.md5
+ff213ecf31b982a3a7f009c9739f64e066e1ffe9 *av1-1-b8-01-size-16x18.ivf
+86b20a13b1939dc5f678e80491f190d376233d58 *av1-1-b8-01-size-16x18.ivf.md5
+c90bd878c59263a15c6a6f515d1c7e071f141559 *av1-1-b8-01-size-18x18.ivf
+6f659036ffcd3dd380cf970cf1a06f7755e0b2de *av1-1-b8-01-size-18x18.ivf.md5
+e16a1411381b34817a4c0d8e5eeaeb8cddcc9c46 *av1-1-b8-01-size-32x18.ivf
+fdb1c4ec56f5aa690eadbe897340fee86a06ae2f *av1-1-b8-01-size-32x18.ivf.md5
+fac7052b39bd2d0ae107e0e94050226712c770c2 *av1-1-b8-01-size-34x18.ivf
+adb0d5a99228027eaa3b016963df447c9818c447 *av1-1-b8-01-size-34x18.ivf.md5
+b8be5e55d9be42746c2b547d0e26e80b21c9802a *av1-1-b8-01-size-64x18.ivf
+8f8f6da34cdf78c5a6551c637e1afe279cc3884e *av1-1-b8-01-size-64x18.ivf.md5
+9e066bdcc2cd789cdf551bd4c9c85c178887b880 *av1-1-b8-01-size-66x18.ivf
+e8ec6effa936423ae2eec2b60a3160720d2de912 *av1-1-b8-01-size-66x18.ivf.md5
+6ebe45085cdeebc2acd6da5abd542a59312c0ff4 *av1-1-b8-01-size-16x32.ivf
+044695669103dbf158591dce9c649317a177d5f6 *av1-1-b8-01-size-16x32.ivf.md5
+9fabb4f60641b8c7995d1dc451419165d41258ff *av1-1-b8-01-size-18x32.ivf
+7263764680dfec864c3fad5df824ab1973489a14 *av1-1-b8-01-size-18x32.ivf.md5
+3f72841a24a13e601d79cf029aa1fdb02970ce0b *av1-1-b8-01-size-32x32.ivf
+bbe1ae2888d291ec6bc98cd0784937580c554103 *av1-1-b8-01-size-32x32.ivf.md5
+392131a7c7609acd0dba88fee14f1ed042d23ab1 *av1-1-b8-01-size-34x32.ivf
+eea68165ebe9acd28693374bf2266374b9c77786 *av1-1-b8-01-size-34x32.ivf.md5
+78afdd96265811ab9466e906347b57161e5c010d *av1-1-b8-01-size-64x32.ivf
+47b317af582700b67f6e77659db1dfaa26c8cde6 *av1-1-b8-01-size-64x32.ivf.md5
+2b4d01f2c9f23044c0d886482c7073bd4d5d37d1 *av1-1-b8-01-size-66x32.ivf
+3ad5a58a0ee5086af370b22ab2b5b7592a4f33e7 *av1-1-b8-01-size-66x32.ivf.md5
+78ddae04eb8277ae605bd7017ad7ad27bfc82d39 *av1-1-b8-01-size-16x34.ivf
+d0c18e679f1fc51e4f7409831321eed9c4858f6f *av1-1-b8-01-size-16x34.ivf.md5
+38d8ed885f46aead6ec1271d8a5d4aee79b8eb68 *av1-1-b8-01-size-18x34.ivf
+097ddbd69b8f54826a35efeb0b8b07ec198bba6b *av1-1-b8-01-size-18x34.ivf.md5
+91a42720bc2e7ba701f4d97b463a098b6707cdbd *av1-1-b8-01-size-32x34.ivf
+c590d43d37095bd2e8f8d12c9278477419b72d1a *av1-1-b8-01-size-32x34.ivf.md5
+4cc2a437dba56e8878113d9b390b980522542028 *av1-1-b8-01-size-34x34.ivf
+57eeb971f00e64abde25be69dbcb4e3ce5065a57 *av1-1-b8-01-size-34x34.ivf.md5
+b36fee1b6ad69d1206466615d69c05e0a4407939 *av1-1-b8-01-size-64x34.ivf
+a78aea0250d0b32657dc0eaf2d8394bc766c0e35 *av1-1-b8-01-size-64x34.ivf.md5
+10e441209262e082e31fef8c15b51579c9e81509 *av1-1-b8-01-size-66x34.ivf
+558b46f6ef1662c208012d0b66d1857eeff3244e *av1-1-b8-01-size-66x34.ivf.md5
+dd44aad500c7ca0fc97e3d8f0abed3c83b24c79c *av1-1-b8-01-size-16x64.ivf
+a5b64e8063abcf3e4872dc4baf1c32384dc5cf83 *av1-1-b8-01-size-16x64.ivf.md5
+aa849f0d09bcb2ead44719d63043536932d5c9f2 *av1-1-b8-01-size-18x64.ivf
+bcdf2dea3590c7031158ffe7b907d9ee35e2fe57 *av1-1-b8-01-size-18x64.ivf.md5
+36e856d30e160ba2fbb00510296202f61afaae49 *av1-1-b8-01-size-32x64.ivf
+99299f75b82c40c13f168adf2d124f57044a39a2 *av1-1-b8-01-size-32x64.ivf.md5
+e3e03ec5d38eb25e97e4ec3adc6ed40ecdebd278 *av1-1-b8-01-size-34x64.ivf
+84625abf8a200a7d20dd3dd3b277b50b3d62ce32 *av1-1-b8-01-size-34x64.ivf.md5
+7d017daebef2d39ed42a505a8e6103ab0c0988c1 *av1-1-b8-01-size-64x64.ivf
+1ff38d5ecba82fb2e6ac3b09c29c9fe74885ac29 *av1-1-b8-01-size-64x64.ivf.md5
+e1b58ba0b462508593399a2ed84db5f1c59ffcd2 *av1-1-b8-01-size-66x64.ivf
+a6b2c84c94fe79ab0373d157d1203f8d66de0706 *av1-1-b8-01-size-66x64.ivf.md5
+7b4faa7eb7b73392b62de6613282a98dddc13bb6 *av1-1-b8-01-size-16x66.ivf
+a2dacf2bae3c4ab352af66a9600946d29ab9a6ee *av1-1-b8-01-size-16x66.ivf.md5
+0f97805fa30497d4cf39665150f00dfdea52d862 *av1-1-b8-01-size-18x66.ivf
+33d8ea0765953250f998da3fe161f2a8cfca2353 *av1-1-b8-01-size-18x66.ivf.md5
+c8bb00256de973e3b3ee31b924f554336d310cdb *av1-1-b8-01-size-32x66.ivf
+6a6588e6edc68ff7739968a9e7cc6d9eaaeed356 *av1-1-b8-01-size-32x66.ivf.md5
+75ec54fec5c36eecde6d0a16e0389a5f7ad8ec22 *av1-1-b8-01-size-34x66.ivf
+36101dfa9495c18696c0d7d61f25e748f4de7425 *av1-1-b8-01-size-34x66.ivf.md5
+7e5491716e70f8199156b8843513c935667b281e *av1-1-b8-01-size-64x66.ivf
+da38755bb0c9ef56b81617835ddf1340242c6dce *av1-1-b8-01-size-64x66.ivf.md5
+68b47b386f61d67cb5b824a7e6bf87c8b9c2bf7b *av1-1-b8-01-size-66x66.ivf
+25974893956ebd92df474325946130c34f880ea7 *av1-1-b8-01-size-66x66.ivf.md5
+9f386d19c87dbfd6ac84a06d2393dd88863ac003 *av1-1-b8-01-size-196x196.ivf
+788f77f655f55de3db94dd69870316134c149116 *av1-1-b8-01-size-196x196.ivf.md5
+ed3bb2bb52a9d1786e233ef38142b15b85097875 *av1-1-b8-01-size-198x196.ivf
+3bb6b6721ad9b2838b2d07e47b29d6c0117526b1 *av1-1-b8-01-size-198x196.ivf.md5
+49461772caaaa7b824d48f4e9c77a906b0dc02d5 *av1-1-b8-01-size-200x196.ivf
+f1cba00c36909c56097c8785df476d42bc91f259 *av1-1-b8-01-size-200x196.ivf.md5
+44a656a22958e26ed169a69deb8f373117224f06 *av1-1-b8-01-size-202x196.ivf
+69be876b52fe42811bba52d36d0bcc88d6c25b3f *av1-1-b8-01-size-202x196.ivf.md5
+0a6fe9b478363faedbfd465a75790b4c2661b9ba *av1-1-b8-01-size-208x196.ivf
+fc8e95a6860a8a37ccdf1dfe49828502fcf96a08 *av1-1-b8-01-size-208x196.ivf.md5
+8e05b5a20ec95afd92bb615a7daa2e17a7ef55a8 *av1-1-b8-01-size-210x196.ivf
+0add512bffbda3300d8f684a53b13b996fe2e46d *av1-1-b8-01-size-210x196.ivf.md5
+a15f12652c6b4d0c30f13a439c941bfc4a431d1a *av1-1-b8-01-size-224x196.ivf
+b904b93252175f79e0e2b28896131ce93d5fc925 *av1-1-b8-01-size-224x196.ivf.md5
+1a57b913443b267f4a31a6925c39f5b58022f550 *av1-1-b8-01-size-226x196.ivf
+7cf3087de5804763a82d2a798243a66459664772 *av1-1-b8-01-size-226x196.ivf.md5
+2cc28541a2a72e8b45a368f71e70fc294e2de3ab *av1-1-b8-01-size-196x198.ivf
+bb736eedb4bd1e39bf9d60435b4b27a12842e112 *av1-1-b8-01-size-196x198.ivf.md5
+c4ebf93fbf3ae52108fd7b39ddef3afae48188ea *av1-1-b8-01-size-198x198.ivf
+fa4de6881511728bafa15b5f441a0cfdf683cc75 *av1-1-b8-01-size-198x198.ivf.md5
+55fce983186d454b0eb15527393bb2465ba41c6b *av1-1-b8-01-size-200x198.ivf
+1ac8fb1ee622cbc4aa1b83cb46b4731c85efae62 *av1-1-b8-01-size-200x198.ivf.md5
+67d276c67886f0a91a7ee06751a64f95eeb7bc1f *av1-1-b8-01-size-202x198.ivf
+1633b62d9e4ea41737c42f70cbde9a5671da0cef *av1-1-b8-01-size-202x198.ivf.md5
+081cb3f29d3956d4d858d9661fd3d62c94b68867 *av1-1-b8-01-size-208x198.ivf
+871d1c99167408dd32fa7603a7296c9b99ccda15 *av1-1-b8-01-size-208x198.ivf.md5
+b2d80b42468d5f296ae240cfb1fc0b3dd3d96bbc *av1-1-b8-01-size-210x198.ivf
+6a3382656cb17b532a97b1061697f9a878fc58d1 *av1-1-b8-01-size-210x198.ivf.md5
+84d7994fa20fcf6c1d8dbd4c2060c988a6fce831 *av1-1-b8-01-size-224x198.ivf
+42ea12e15de81f2e8617b6de7bae76de2da4d648 *av1-1-b8-01-size-224x198.ivf.md5
+c74a9281cf98c597121df6bff0ac5312b887f969 *av1-1-b8-01-size-226x198.ivf
+4133aae0001804e2bbc7928fc065517a6dd8b288 *av1-1-b8-01-size-226x198.ivf.md5
+27adbf148c63f807bd617cfd78aeaedb8b0f2304 *av1-1-b8-01-size-196x200.ivf
+9253e525e6207ef1ce0839b8f88ea781e9abe41e *av1-1-b8-01-size-196x200.ivf.md5
+21c9ea4d882e48353d3df66fcde0e4746168163f *av1-1-b8-01-size-198x200.ivf
+3d5ee59fde9194f0eaff736051cfd1d7b7daeff1 *av1-1-b8-01-size-198x200.ivf.md5
+c27b0b57667910847122a0309c703315e444110f *av1-1-b8-01-size-200x200.ivf
+7b2a15a17b421ef07e285ca4e8a224f0512c434d *av1-1-b8-01-size-200x200.ivf.md5
+780de549e4163a52590f7c0f488e027a8a4aa053 *av1-1-b8-01-size-202x200.ivf
+cb0ec0969522ca60d79a639e9b9509363468ffd0 *av1-1-b8-01-size-202x200.ivf.md5
+2c59821904863e264ae61401cbd494a79bc04f13 *av1-1-b8-01-size-208x200.ivf
+9963955966a52b65cdd13465c9fb2ba3b5356755 *av1-1-b8-01-size-208x200.ivf.md5
+ff63121611ea9c0628c7e5af13de5e7786611ca6 *av1-1-b8-01-size-210x200.ivf
+2a5993be234e3af2af6d185b2a6f3aaf1979b83a *av1-1-b8-01-size-210x200.ivf.md5
+b8485ada95440d78b51153227231b1aced1a8273 *av1-1-b8-01-size-224x200.ivf
+9c3cd32ea6c006a91eb37d69dbeccf878de5d214 *av1-1-b8-01-size-224x200.ivf.md5
+1aa0ce3e3a74f9b600a146e98b05547a0b454c48 *av1-1-b8-01-size-226x200.ivf
+e045be96c3af16a9ddc10a9933e8ddfb3319d716 *av1-1-b8-01-size-226x200.ivf.md5
+e92b76480f4339855d998b97182f36b28deadcfa *av1-1-b8-01-size-196x202.ivf
+480c707abcd2a650e2160ec397f8348cecb45770 *av1-1-b8-01-size-196x202.ivf.md5
+137b9c0d10a3bdbdf6f97b3e6331f3e8acaf8f91 *av1-1-b8-01-size-198x202.ivf
+7429642146d0da55161ab13024a261094ee2ce87 *av1-1-b8-01-size-198x202.ivf.md5
+9cea71c44ad015ac702d675bacca17876e65cb1a *av1-1-b8-01-size-200x202.ivf
+76b1ec6c42da55f47e389a561590d1a7c713e495 *av1-1-b8-01-size-200x202.ivf.md5
+26dffdcd0dac9becf68d12e31fcd91eddf1f7154 *av1-1-b8-01-size-202x202.ivf
+ddb75e99123fed4ef05d9b85200cefd8985bc84c *av1-1-b8-01-size-202x202.ivf.md5
+04007e83bb66ba547d09f8926ea5bfc7fd9e4b2a *av1-1-b8-01-size-208x202.ivf
+5b72eb58db22087ad416c499119f41e718395b52 *av1-1-b8-01-size-208x202.ivf.md5
+721ff7c0ae0e2ed896b5acac230113f1404e769c *av1-1-b8-01-size-210x202.ivf
+187d2ef939fc26e1a1c7de65abe8e058d8aae17a *av1-1-b8-01-size-210x202.ivf.md5
+dba41421cc938bcf0234254f96be0325ab66186e *av1-1-b8-01-size-224x202.ivf
+58856038c1eb13a7bf0353a30b1affe844cd31b1 *av1-1-b8-01-size-224x202.ivf.md5
+55eba14878d25dcc351ee5e92fa06e559035b409 *av1-1-b8-01-size-226x202.ivf
+e295b3d791d40d7c1fff2c40a260078dccaef24a *av1-1-b8-01-size-226x202.ivf.md5
+6c777223990ddfd92040a8526646ed0f39299b0d *av1-1-b8-01-size-196x208.ivf
+5210daff766cddaf3945610ee05ff242aef8175a *av1-1-b8-01-size-196x208.ivf.md5
+252831abfb9f4a9a8556c21cc3bf60adfe88210f *av1-1-b8-01-size-198x208.ivf
+35ed9601e608a829980cec81e41b7bd3e5f4c2ce *av1-1-b8-01-size-198x208.ivf.md5
+e800ed893a88704a4576d4984957f3664560daa9 *av1-1-b8-01-size-200x208.ivf
+82c038f9072a2fcf8d55fb4a474fdd791ba9a290 *av1-1-b8-01-size-200x208.ivf.md5
+9ce7bb932dd99f86da8ff2ab89fa4d3089a78da8 *av1-1-b8-01-size-202x208.ivf
+0611bf0179abe3c820a447a2bd3a04c3790f3a87 *av1-1-b8-01-size-202x208.ivf.md5
+e5900d9150c8bebc49776227afd3b0a21f5a6ac6 *av1-1-b8-01-size-208x208.ivf
+86d6b9a3840aa0a77938547c905bd6f45d069681 *av1-1-b8-01-size-208x208.ivf.md5
+2758ba5dad16f4a91334f2ed07a4a037201bb873 *av1-1-b8-01-size-210x208.ivf
+78453b1fda2ccc6f35e0d762567807757bcddb16 *av1-1-b8-01-size-210x208.ivf.md5
+fff88fb8e833f6b4ad64cb591b219c7cceb7f2d2 *av1-1-b8-01-size-224x208.ivf
+87266fc34aaed82cdb98cbc309b221ad52eccd81 *av1-1-b8-01-size-224x208.ivf.md5
+dec839fe64046461015b56cda191835284f42a52 *av1-1-b8-01-size-226x208.ivf
+d7a15264fc3fd55d3aec0ccfaa7c434c6d90969f *av1-1-b8-01-size-226x208.ivf.md5
+584782e93ed1cb7797a90fece44becdd1e23bf0d *av1-1-b8-01-size-196x210.ivf
+ed76ec841b18a457853e368576967c4768fc2730 *av1-1-b8-01-size-196x210.ivf.md5
+dab625599b9f01398b593e865d9a4a95a029d60f *av1-1-b8-01-size-198x210.ivf
+b90e8d96a1f5b329b088b467a11fed2d055d74ca *av1-1-b8-01-size-198x210.ivf.md5
+6774bee17b9e50d2d8630e2e1afc30ded67e662d *av1-1-b8-01-size-200x210.ivf
+343a86bd54eb3dd5e9902eb62a3d776dcff2f4f3 *av1-1-b8-01-size-200x210.ivf.md5
+0456c3b8e242eeee019ca97d155f81124de62c90 *av1-1-b8-01-size-202x210.ivf
+5a6a6428c9858a0d3561db42ceaf981c143fe479 *av1-1-b8-01-size-202x210.ivf.md5
+6a3a8f65bf806b1be7726b983427880f772c9986 *av1-1-b8-01-size-208x210.ivf
+5563ea6d8c65887553ff3000addc6418913f1650 *av1-1-b8-01-size-208x210.ivf.md5
+5a8b69489f8e9b917ea7718ad2645101cdbe5644 *av1-1-b8-01-size-210x210.ivf
+f4b01604036fa23000d44fbf42097ae1181bcd62 *av1-1-b8-01-size-210x210.ivf.md5
+fb6f5b08a048698cfe324557ee8cd840c4a3f6ce *av1-1-b8-01-size-224x210.ivf
+3ce5c404e3ca09c8e994b3043bad42cd555b00c0 *av1-1-b8-01-size-224x210.ivf.md5
+2e9fc8510d2131b2f3c9a93bececac985e4426d2 *av1-1-b8-01-size-226x210.ivf
+897c537e259331ca86cdd6e4d2bd343f8538402e *av1-1-b8-01-size-226x210.ivf.md5
+8300512106fce3424eb74b5d4bc0f4f19f7c9af8 *av1-1-b8-01-size-196x224.ivf
+43662ea025ea79afe4964fd4d12a77f4aa4e565e *av1-1-b8-01-size-196x224.ivf.md5
+640f8fda7ade8f2850e2275a9f5e233e33a0ba8d *av1-1-b8-01-size-198x224.ivf
+9ac690bdbbce47d7b169128b568f955e70076f8c *av1-1-b8-01-size-198x224.ivf.md5
+ce2e9379c72fc924e364d5727605394a1438a211 *av1-1-b8-01-size-200x224.ivf
+1ec35a53d88072b96b255202f678178bc7e5bb20 *av1-1-b8-01-size-200x224.ivf.md5
+5d3af7921623deccb578115c8ce207c019f97f50 *av1-1-b8-01-size-202x224.ivf
+14eafd55b0cda3a3476cae7ad500dbd5ee899dd5 *av1-1-b8-01-size-202x224.ivf.md5
+6b6d78e466cf94a5ef8dfe252caa0948dd2ec175 *av1-1-b8-01-size-208x224.ivf
+e178b0c272dfcfe614c6b49cb28dad11781af0b6 *av1-1-b8-01-size-208x224.ivf.md5
+dd2232b9e18971d7e19650a1e3218aef1010247f *av1-1-b8-01-size-210x224.ivf
+40a66198c47820f5fa2d2e389ec0c1191ea4ffcc *av1-1-b8-01-size-210x224.ivf.md5
+9ec028b81a5ea311683328d856f436e6d0b0e6a0 *av1-1-b8-01-size-224x224.ivf
+143b9530ce722385db2c2d883daa649ed42b8d40 *av1-1-b8-01-size-224x224.ivf.md5
+bf833947e62935c54e1e727ccb36157f7c1e9e5d *av1-1-b8-01-size-226x224.ivf
+ca4f3b44463106e4f0bb54e490c3bd457d7d780b *av1-1-b8-01-size-226x224.ivf.md5
+5525f7e312ec073f480ed5a2be5bdc4f0ce51a09 *av1-1-b8-01-size-196x226.ivf
+062d4b240741184458d2d2abd243ed7877631de8 *av1-1-b8-01-size-196x226.ivf.md5
+e6b911142394b94c23191eaa63c9eb41a00f80b0 *av1-1-b8-01-size-198x226.ivf
+3b580d903dddf47082f5e055bfb01a4f05c09b7d *av1-1-b8-01-size-198x226.ivf.md5
+70feb5efeb28df25f7d1a661c73bf013c5ada9b4 *av1-1-b8-01-size-200x226.ivf
+f0b894e7f787e62f1492be62f3dedeb065062160 *av1-1-b8-01-size-200x226.ivf.md5
+7f9a10831e2389b31497fad50080b4d5452d6e91 *av1-1-b8-01-size-202x226.ivf
+45b7194eba9367c8059403c23ca4ae49e988dfaf *av1-1-b8-01-size-202x226.ivf.md5
+967837a2cfbf9aa3131f73aec6a52dcdd82926c7 *av1-1-b8-01-size-208x226.ivf
+c8baedb48fd5d4c956aa8d73fd957370f718f047 *av1-1-b8-01-size-208x226.ivf.md5
+9c926226b9f6b015501d8ac1e3f95e8570283a05 *av1-1-b8-01-size-210x226.ivf
+57d4837667fd4c5a7aeb908626d701b632852c60 *av1-1-b8-01-size-210x226.ivf.md5
+25a4940922761239809d82c45c2be1c5e4f48785 *av1-1-b8-01-size-224x226.ivf
+87ae7e7558241bf3575a333f56fbad4dfdade8ff *av1-1-b8-01-size-224x226.ivf.md5
+40dd208eb525cd90d7c0674cf787097fb909afae *av1-1-b8-01-size-226x226.ivf
+34bdef682a4eae0e0a05e4486a968af1df8b220a *av1-1-b8-01-size-226x226.ivf.md5
+9bbe8499796aa588ff02e313fb0d4349940d2fea *av1-1-b10-00-quantizer-00.ivf
+36b402eedad2bacee8ac09acce44e2fc356dd80b *av1-1-b10-00-quantizer-00.ivf.md5
+1d5e1d2827624f328020bf123df213bb175577e0 *av1-1-b10-00-quantizer-01.ivf
+16c529be5502369e43ce9c6fe99a9709968e3daf *av1-1-b10-00-quantizer-01.ivf.md5
+39abc20739242a8f05efd4b35d7603c8ad7ff45d *av1-1-b10-00-quantizer-02.ivf
+81faa72c3d43b003966fe09ffaae51b07b1059be *av1-1-b10-00-quantizer-02.ivf.md5
+92ebf349b803333a43824a83d997b8cf76f656f9 *av1-1-b10-00-quantizer-03.ivf
+5e7556dc998cb8b506a43cc078e30802d7e600e6 *av1-1-b10-00-quantizer-03.ivf.md5
+1c496177c66e49f2e3556af87ec67afb5060170b *av1-1-b10-00-quantizer-04.ivf
+560fea4800a44fe19ed8d3e74f425bdbf1fb8abd *av1-1-b10-00-quantizer-04.ivf.md5
+7de864b8475ce0acd0ecb01827f2c9add815352b *av1-1-b10-00-quantizer-05.ivf
+1c1aea3db3f54a91866d89fd3b1a0d285ca10310 *av1-1-b10-00-quantizer-05.ivf.md5
+b6501c165619b036d0f7864fd4739973d2d18970 *av1-1-b10-00-quantizer-06.ivf
+d758c8eff275651006c41e7dd447cac13b489ad7 *av1-1-b10-00-quantizer-06.ivf.md5
+e4df6f588f156dffaafd9517b64f753cfc9ccf05 *av1-1-b10-00-quantizer-07.ivf
+3c577f67dade4537de642fd457ea2b367424f336 *av1-1-b10-00-quantizer-07.ivf.md5
+07e9c4c18abb36c8699c1c12bebcc727f090b525 *av1-1-b10-00-quantizer-08.ivf
+4981568ade3170f311cb114fa2689edc4bc35e67 *av1-1-b10-00-quantizer-08.ivf.md5
+2268ecd2899f1b41ae9898925b1d62cfefa30282 *av1-1-b10-00-quantizer-09.ivf
+029b03029b65b7c4c208961f0820467ad42fd3d6 *av1-1-b10-00-quantizer-09.ivf.md5
+3d2adaf6441cfa9585dcbf7d19d65bf6992a29a3 *av1-1-b10-00-quantizer-10.ivf
+017b7fb4c3ba0747c2d5688d493da33ef993d110 *av1-1-b10-00-quantizer-10.ivf.md5
+006535760bd7dc1cfc95e648b05215954a2e76c2 *av1-1-b10-00-quantizer-11.ivf
+c0ae083deb8e820aa49034af4d100944dd977018 *av1-1-b10-00-quantizer-11.ivf.md5
+840e0cbfe1acc8a7a45c823dc55ab44a0b6b553e *av1-1-b10-00-quantizer-12.ivf
+49232ea38bdef650c94808f53834f1137cd4bf39 *av1-1-b10-00-quantizer-12.ivf.md5
+04b0e5a7387e07474f51be4b2c3e05211b40f0d0 *av1-1-b10-00-quantizer-13.ivf
+a51b5ec4b890df3a64f9f0d866b8c41296c9e081 *av1-1-b10-00-quantizer-13.ivf.md5
+5dc47a140fbcbf08bf91481ee3585e9e067561ab *av1-1-b10-00-quantizer-14.ivf
+2625319eef69d6225e6ab6e5ce7790491406cb5d *av1-1-b10-00-quantizer-14.ivf.md5
+f866be86d8d8aa08ded30e42988b0936c1a16064 *av1-1-b10-00-quantizer-15.ivf
+03b7c1eefb54d99e30051c7123c0453f04a6579d *av1-1-b10-00-quantizer-15.ivf.md5
+548df2371dfb485419ed9baf28e3f495c64f364a *av1-1-b10-00-quantizer-16.ivf
+8a0d6bf1626b05b65c77331305414fe9be54e8c6 *av1-1-b10-00-quantizer-16.ivf.md5
+0077c82f96a2e095a3cb8de9bfa63715e3c9f438 *av1-1-b10-00-quantizer-17.ivf
+5d85f77f3087f4b206930722a945c60039262be4 *av1-1-b10-00-quantizer-17.ivf.md5
+1e0f1245ecb4c903b5dc7072d959fc43a7bba381 *av1-1-b10-00-quantizer-18.ivf
+06316ae2b45f2359a70cc3855ffd6ab81048b41a *av1-1-b10-00-quantizer-18.ivf.md5
+f197198f7ec058110185fda5297a1a43993654df *av1-1-b10-00-quantizer-19.ivf
+bac522c7f234d506c75b5495d74b3fa57c83a4df *av1-1-b10-00-quantizer-19.ivf.md5
+c2f57324d000b349323f37d5ebebde8c2b861f30 *av1-1-b10-00-quantizer-20.ivf
+999c6110786cbc25e67792234a5a02f2cb4553c7 *av1-1-b10-00-quantizer-20.ivf.md5
+2ffad9adfd19286fe2166ba877289d201c9a634f *av1-1-b10-00-quantizer-21.ivf
+d55713eaa791cfd7bf69b6c26d5032029d9a0f06 *av1-1-b10-00-quantizer-21.ivf.md5
+382528db53328c1a38976f5d9b579eef35d839f4 *av1-1-b10-00-quantizer-22.ivf
+cb5bd459e1a90126da9264cff4281515f95755b2 *av1-1-b10-00-quantizer-22.ivf.md5
+b52cc6160fc66f72ad66c198d275a1c73f925022 *av1-1-b10-00-quantizer-23.ivf
+c0f9d6659e1f283e9356fd7b4ac9f7cc5544cdc2 *av1-1-b10-00-quantizer-23.ivf.md5
+e11f15e3b63e7606b1122bb3670ee77c09c04840 *av1-1-b10-00-quantizer-24.ivf
+e9f141b924440e044270c81a68458fe498599a8e *av1-1-b10-00-quantizer-24.ivf.md5
+fb91793b69824c99b0218788dcea0a74ebd7e84e *av1-1-b10-00-quantizer-25.ivf
+434e33d609b2683c3cfbcc3a2cdfc26339590fb6 *av1-1-b10-00-quantizer-25.ivf.md5
+d82e38f31cdcf8b43479e6ddaa83373de38f70a2 *av1-1-b10-00-quantizer-26.ivf
+183943b851ba383a536f13c83b93f61ac8961ad5 *av1-1-b10-00-quantizer-26.ivf.md5
+6bf5e4e8e0aca699e493b9eb3672d2117494d74d *av1-1-b10-00-quantizer-27.ivf
+f0fb7e0a99180828b0e38b2cfe0622eecc2d26b8 *av1-1-b10-00-quantizer-27.ivf.md5
+d5adee2567544c3ae4223b3f3528a770377878d2 *av1-1-b10-00-quantizer-28.ivf
+14edf588efc67570e529b0ff8aeb8e7a0c69238b *av1-1-b10-00-quantizer-28.ivf.md5
+e6dcdc106847956035e3f00aabf4470f97e1887e *av1-1-b10-00-quantizer-29.ivf
+413c5cb778611c7c1a810b53861b9ab1fb391f17 *av1-1-b10-00-quantizer-29.ivf.md5
+b5e98b3f6b1db04d46bf43064c6ac64f797aff00 *av1-1-b10-00-quantizer-30.ivf
+d1a603661d76c28658c7cd2892b408e91d77893e *av1-1-b10-00-quantizer-30.ivf.md5
+80168371d1150e82e3f46bcbbcabba458b835b19 *av1-1-b10-00-quantizer-31.ivf
+904ecd033d4af5239c4d5b3f86e51ed5c3c2e3fb *av1-1-b10-00-quantizer-31.ivf.md5
+96291f6ace85980892d135a5b74188cd629c325f *av1-1-b10-00-quantizer-32.ivf
+a5ceace390d4a75d48281fe29060c21557e4f5ae *av1-1-b10-00-quantizer-32.ivf.md5
+0f80495de34eae07c4905b72573a315a879390ec *av1-1-b10-00-quantizer-33.ivf
+72b8f662973a660412946687dff878b276ae518e *av1-1-b10-00-quantizer-33.ivf.md5
+24905e3be7db320994b7fb8311dfd50a7c9e54da *av1-1-b10-00-quantizer-34.ivf
+cea514bb1b7b064c4d31914a2cb266611c278577 *av1-1-b10-00-quantizer-34.ivf.md5
+083012960dd7c17d3b00fa0e807759c98faded8f *av1-1-b10-00-quantizer-35.ivf
+de5fdb9e1e581484af1cc7d2dd3c3e84c90cebb2 *av1-1-b10-00-quantizer-35.ivf.md5
+f725f179aeee5b413620c0dd81b007b245c2a7ed *av1-1-b10-00-quantizer-36.ivf
+246b1931c04c02df1f168090e2650827cd5dbabd *av1-1-b10-00-quantizer-36.ivf.md5
+f6aa824156e9848f237481889a8103eb6130f31d *av1-1-b10-00-quantizer-37.ivf
+a8f78dd15fc2994369a08c2ddddcd0760c62ea5b *av1-1-b10-00-quantizer-37.ivf.md5
+a8dd662338c493aea266b99203e70af25982633f *av1-1-b10-00-quantizer-38.ivf
+09f36d998e85d0450060f540e50b075ae1432fc6 *av1-1-b10-00-quantizer-38.ivf.md5
+d97428871720ed658da6ed0e3f7c15da83387e4c *av1-1-b10-00-quantizer-39.ivf
+8c5230048909ee8f86f87c116f153cd910d0141f *av1-1-b10-00-quantizer-39.ivf.md5
+86e754e55e9b63c6e0a4fef01761414f8a6b61ca *av1-1-b10-00-quantizer-40.ivf
+99a71accf6457264e45ca80d3b1f082ee5acdecc *av1-1-b10-00-quantizer-40.ivf.md5
+9d18b7236506ab7e107c062620b64096ec0cf423 *av1-1-b10-00-quantizer-41.ivf
+5771159a9a7c7b66c9e13bb13ec3d53b37860208 *av1-1-b10-00-quantizer-41.ivf.md5
+54b72bc879a80e66613f421e67db62bba1c0041b *av1-1-b10-00-quantizer-42.ivf
+bf958236883ee7209ef4cb0b7503b430634a291e *av1-1-b10-00-quantizer-42.ivf.md5
+a06d5321a51d90404dd7085ae511d7df5d5e1e05 *av1-1-b10-00-quantizer-43.ivf
+ddb25723d976043d863634b9dc3b5fb84a245803 *av1-1-b10-00-quantizer-43.ivf.md5
+2ea0b64c170d7299dae1c14a8a49349aee8e0d08 *av1-1-b10-00-quantizer-44.ivf
+d18bde1b4893792173fa2014665e9364395ad5e9 *av1-1-b10-00-quantizer-44.ivf.md5
+73e506a32d3518e23424f231c7b5323d7a34a3d6 *av1-1-b10-00-quantizer-45.ivf
+be6224ebc77a3e5fb9c1645b876007e584a09d89 *av1-1-b10-00-quantizer-45.ivf.md5
+841223871374464194edc739c48dc7cefd1ff255 *av1-1-b10-00-quantizer-46.ivf
+4766d616f923496a8dc113c9b7f875f0c0735f9a *av1-1-b10-00-quantizer-46.ivf.md5
+8bbbbea130aaea453f7b826956a5520d10a0eccf *av1-1-b10-00-quantizer-47.ivf
+3ea21fac0c492b03d8ec25e4ee0971cd57e5f71a *av1-1-b10-00-quantizer-47.ivf.md5
+3ce83e0f1e1835b9a6c10fe502a16fd3650839e0 *av1-1-b10-00-quantizer-48.ivf
+b468de2c09fca5a6b2bb7a20bab4afd8d192c31d *av1-1-b10-00-quantizer-48.ivf.md5
+f3a757c678aa00f9a9c4c4658d37733fd935925a *av1-1-b10-00-quantizer-49.ivf
+f888dc88db576122695d4eb41c486aacd28a2d1d *av1-1-b10-00-quantizer-49.ivf.md5
+a9d78aaef105cc5a95b7ebb54783f37e75673123 *av1-1-b10-00-quantizer-50.ivf
+06d0c5e79cc794030c4be022089b1d12c1383f71 *av1-1-b10-00-quantizer-50.ivf.md5
+165c20ee372f83682d094541097e375227353239 *av1-1-b10-00-quantizer-51.ivf
+b3d90214b8c6e6f6d9357bb5784d10081325c356 *av1-1-b10-00-quantizer-51.ivf.md5
+5b3ea7a18654d943065f5c176974c3960b56664e *av1-1-b10-00-quantizer-52.ivf
+dc61a6e4e2549074130023b14b137fb4fe442ce3 *av1-1-b10-00-quantizer-52.ivf.md5
+74c3b5851b6a94d33b575a689eb8d34592e95d5f *av1-1-b10-00-quantizer-53.ivf
+a80e43a0fb2b852426bd941b8d4b8f56690e9bc9 *av1-1-b10-00-quantizer-53.ivf.md5
+d05b8dea2cddd4f0d9e792f42f71afbd29f7811c *av1-1-b10-00-quantizer-54.ivf
+432937893321f4bd25fa400b8988c5788cb06ecf *av1-1-b10-00-quantizer-54.ivf.md5
+4eaee0f1970426be0bbeb7d4fccdc7e804e9bea4 *av1-1-b10-00-quantizer-55.ivf
+710ab95ce1dcd2540db4477ff4ee6ab771fe0759 *av1-1-b10-00-quantizer-55.ivf.md5
+fe637930c9faa8744cba37effc4cb5510315d1c0 *av1-1-b10-00-quantizer-56.ivf
+2f9431b30523fb6a3e4122f22c6c3ff7b96a7987 *av1-1-b10-00-quantizer-56.ivf.md5
+ed54fc7fcec194eef1f50adbbe12a6a36ab6836b *av1-1-b10-00-quantizer-57.ivf
+43bccac7800b399210cf15520a83739c23a5d9c7 *av1-1-b10-00-quantizer-57.ivf.md5
+a7b8d628ba3e4c5f37aa6a3d7b82afda73ac89dc *av1-1-b10-00-quantizer-58.ivf
+b26638272b787df54f45a46629b852acbcb73e3d *av1-1-b10-00-quantizer-58.ivf.md5
+c077f22ff547fb5ffd020e8dac91d05942fb52df *av1-1-b10-00-quantizer-59.ivf
+4efd99cc0891bf345b8cd2ae8e21709d61be497b *av1-1-b10-00-quantizer-59.ivf.md5
+301ab53039d75e1ffa8cc6a0874d9ea94e4a6a0d *av1-1-b10-00-quantizer-60.ivf
+4729bd734a6edd2d8d0432a3f66b3d91d565050e *av1-1-b10-00-quantizer-60.ivf.md5
+c78640d3211034df9fcb273bdfc18625819652f2 *av1-1-b10-00-quantizer-61.ivf
+3d823eb2b33ccfea68db506626bcbecf49b0f167 *av1-1-b10-00-quantizer-61.ivf.md5
+bf241a449a28773b93e6e529a06dfc28109577e4 *av1-1-b10-00-quantizer-62.ivf
+75457d8476f1927f737d089dcf3d0f7f99f3c4fb *av1-1-b10-00-quantizer-62.ivf.md5
+8b6eb3fff2e0db7eac775b08c745250ca591e2d9 *av1-1-b10-00-quantizer-63.ivf
+63ea689d025593e5d91760785b8e446d04d4671e *av1-1-b10-00-quantizer-63.ivf.md5
+a9f7ea6312a533cc6426a6145edd190d45813c37 *av1-1-b8-02-allintra.ivf
+8fd8f789cfee1069d20f3e2c241f5cad7292239e *av1-1-b8-02-allintra.ivf.md5
+e69e41fee40b408b6eebcc79f266a95f2ee24f9e *av1-1-b8-03-sizedown.mkv
+8c528fb3ccda959a29721566e132f730935ca32b *av1-1-b8-03-sizedown.mkv.md5
+1889da5ee1708007e47bb887470ac477e1d7ba01 *av1-1-b8-03-sizeup.mkv
+8de81b170635d456602dc8923a8b39c534d01fa8 *av1-1-b8-03-sizeup.mkv.md5
+d3ed7de0aa8c155fe35e0f5f4203240710d31383 *park_joy_90p_8_420_monochrome.y4m
+5b3f0907407b809aa66b62cb080feda8c92454ca *park_joy_90p_8_420_vertical_csp.y4m
diff --git a/third_party/aom/test/test.cmake b/third_party/aom/test/test.cmake
new file mode 100644
index 0000000000..b16ae14c3d
--- /dev/null
+++ b/third_party/aom/test/test.cmake
@@ -0,0 +1,438 @@
+#
+# Copyright (c) 2017, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and the
+# Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License was
+# not distributed with this source code in the LICENSE file, you can obtain it
+# at www.aomedia.org/license/software. If the Alliance for Open Media Patent
+# License 1.0 was not distributed with this source code in the PATENTS file, you
+# can obtain it at www.aomedia.org/license/patent.
+#
+if(AOM_TEST_TEST_CMAKE_)
+  return()
+endif() # AOM_TEST_TEST_CMAKE_
+set(AOM_TEST_TEST_CMAKE_ 1)
+
+include(FindPythonInterp)
+include(ProcessorCount)
+
+include("${AOM_ROOT}/test/test_data_util.cmake")
+
+set(AOM_UNIT_TEST_DATA_LIST_FILE "${AOM_ROOT}/test/test-data.sha1")
+
+list(APPEND AOM_UNIT_TEST_WRAPPER_SOURCES "${AOM_GEN_SRC_DIR}/usage_exit.c"
+            "${AOM_ROOT}/test/test_libaom.cc")
+
+list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+            "${AOM_ROOT}/test/acm_random.h"
+            "${AOM_ROOT}/test/aom_integer_test.cc"
+            "${AOM_ROOT}/test/av1_config_test.cc"
+            "${AOM_ROOT}/test/blockd_test.cc"
+            "${AOM_ROOT}/test/clear_system_state.h"
+            "${AOM_ROOT}/test/codec_factory.h"
+            "${AOM_ROOT}/test/decode_test_driver.cc"
+            "${AOM_ROOT}/test/decode_test_driver.h"
+            "${AOM_ROOT}/test/function_equivalence_test.h"
+            "${AOM_ROOT}/test/log2_test.cc"
+            "${AOM_ROOT}/test/md5_helper.h"
+            "${AOM_ROOT}/test/register_state_check.h"
+            "${AOM_ROOT}/test/test_vectors.cc"
+            "${AOM_ROOT}/test/test_vectors.h"
+            "${AOM_ROOT}/test/transform_test_base.h"
+            "${AOM_ROOT}/test/util.h"
+            "${AOM_ROOT}/test/video_source.h")
+
+if(CONFIG_INTERNAL_STATS)
+  list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+              "${AOM_ROOT}/test/hbd_metrics_test.cc")
+endif()
+
+list(APPEND AOM_UNIT_TEST_DECODER_SOURCES "${AOM_ROOT}/test/decode_api_test.cc"
+            "${AOM_ROOT}/test/external_frame_buffer_test.cc"
+            "${AOM_ROOT}/test/invalid_file_test.cc"
+            "${AOM_ROOT}/test/test_vector_test.cc"
+            "${AOM_ROOT}/test/ivf_video_source.h")
+
+list(APPEND AOM_UNIT_TEST_ENCODER_SOURCES
+            "${AOM_ROOT}/test/active_map_test.cc"
+            "${AOM_ROOT}/test/altref_test.cc"
+            "${AOM_ROOT}/test/aq_segment_test.cc"
+            "${AOM_ROOT}/test/borders_test.cc"
+            "${AOM_ROOT}/test/cpu_speed_test.cc"
+            "${AOM_ROOT}/test/datarate_test.cc"
+            "${AOM_ROOT}/test/encode_api_test.cc"
+            "${AOM_ROOT}/test/encode_test_driver.cc"
+            "${AOM_ROOT}/test/encode_test_driver.h"
+            "${AOM_ROOT}/test/end_to_end_test.cc"
+            "${AOM_ROOT}/test/error_resilience_test.cc"
+            "${AOM_ROOT}/test/frame_size_tests.cc"
+            "${AOM_ROOT}/test/horz_superres_test.cc"
+            "${AOM_ROOT}/test/i420_video_source.h"
+            "${AOM_ROOT}/test/lossless_test.cc"
+            "${AOM_ROOT}/test/monochrome_test.cc"
+            "${AOM_ROOT}/test/qm_test.cc"
+            "${AOM_ROOT}/test/resize_test.cc"
+            "${AOM_ROOT}/test/scalability_test.cc"
+            "${AOM_ROOT}/test/y4m_test.cc"
+            "${AOM_ROOT}/test/y4m_video_source.h"
+            "${AOM_ROOT}/test/yuv_video_source.h")
+
+list(APPEND AOM_DECODE_PERF_TEST_SOURCES "${AOM_ROOT}/test/decode_perf_test.cc")
+list(APPEND AOM_ENCODE_PERF_TEST_SOURCES "${AOM_ROOT}/test/encode_perf_test.cc")
+list(APPEND AOM_UNIT_TEST_WEBM_SOURCES "${AOM_ROOT}/test/webm_video_source.h")
+list(APPEND AOM_TEST_INTRA_PRED_SPEED_SOURCES "${AOM_GEN_SRC_DIR}/usage_exit.c"
+            "${AOM_ROOT}/test/test_intra_pred_speed.cc")
+
+if(NOT BUILD_SHARED_LIBS)
+  list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+              "${AOM_ROOT}/test/cdef_test.cc"
+              "${AOM_ROOT}/test/cfl_test.cc"
+              "${AOM_ROOT}/test/convolve_test.cc"
+              "${AOM_ROOT}/test/hiprec_convolve_test.cc"
+              "${AOM_ROOT}/test/hiprec_convolve_test_util.cc"
+              "${AOM_ROOT}/test/hiprec_convolve_test_util.h"
+              "${AOM_ROOT}/test/intrabc_test.cc"
+              "${AOM_ROOT}/test/intrapred_test.cc"
+              "${AOM_ROOT}/test/lpf_test.cc"
+              "${AOM_ROOT}/test/onyxc_int_test.cc"
+              "${AOM_ROOT}/test/scan_test.cc"
+              "${AOM_ROOT}/test/selfguided_filter_test.cc"
+              "${AOM_ROOT}/test/simd_cmp_impl.h"
+              "${AOM_ROOT}/test/simd_impl.h")
+
+  if(CONFIG_ACCOUNTING)
+    list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+                "${AOM_ROOT}/test/accounting_test.cc")
+  endif()
+
+  if(CONFIG_AV1_DECODER AND CONFIG_AV1_ENCODER)
+    list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+                "${AOM_ROOT}/test/av1_encoder_parms_get_to_decoder.cc"
+                "${AOM_ROOT}/test/av1_ext_tile_test.cc"
+                "${AOM_ROOT}/test/binary_codes_test.cc"
+                "${AOM_ROOT}/test/boolcoder_test.cc"
+                "${AOM_ROOT}/test/coding_path_sync.cc"
+                "${AOM_ROOT}/test/decode_multithreaded_test.cc"
+                "${AOM_ROOT}/test/divu_small_test.cc"
+                "${AOM_ROOT}/test/dr_prediction_test.cc"
+                "${AOM_ROOT}/test/ec_test.cc"
+                "${AOM_ROOT}/test/ethread_test.cc"
+                "${AOM_ROOT}/test/film_grain_table_test.cc"
+                "${AOM_ROOT}/test/segment_binarization_sync.cc"
+                "${AOM_ROOT}/test/superframe_test.cc"
+                "${AOM_ROOT}/test/tile_independence_test.cc")
+  endif()
+
+  list(APPEND AOM_UNIT_TEST_COMMON_INTRIN_NEON
+              "${AOM_ROOT}/test/simd_cmp_neon.cc")
+  if(HAVE_NEON)
+    list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+                "${AOM_ROOT}/test/simd_neon_test.cc")
+  endif()
+
+  list(APPEND AOM_UNIT_TEST_COMMON_INTRIN_SSE2
+              "${AOM_ROOT}/test/simd_cmp_sse2.cc")
+  if(HAVE_SSE2)
+    list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+                "${AOM_ROOT}/test/simd_sse2_test.cc")
+  endif()
+
+  list(APPEND AOM_UNIT_TEST_COMMON_INTRIN_SSSE3
+              "${AOM_ROOT}/test/simd_cmp_ssse3.cc")
+  if(HAVE_SSSE3)
+    list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+                "${AOM_ROOT}/test/simd_ssse3_test.cc")
+  endif()
+
+  if(HAVE_SSE4)
+    list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+                "${AOM_ROOT}/test/simd_sse4_test.cc")
+  endif()
+
+  if(HAVE_SSE4_1)
+    list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+                "${AOM_ROOT}/test/filterintra_test.cc")
+  endif()
+
+  list(APPEND AOM_UNIT_TEST_COMMON_INTRIN_AVX2
+              "${AOM_ROOT}/test/simd_cmp_avx2.cc")
+  if(HAVE_AVX2)
+    list(APPEND AOM_UNIT_TEST_COMMON_SOURCES
+                "${AOM_ROOT}/test/simd_avx2_test.cc")
+  endif()
+
+  list(APPEND AOM_UNIT_TEST_ENCODER_SOURCES
+              "${AOM_ROOT}/test/arf_freq_test.cc"
+              "${AOM_ROOT}/test/av1_convolve_2d_test.cc"
+              "${AOM_ROOT}/test/av1_convolve_2d_test_util.cc"
+              "${AOM_ROOT}/test/av1_convolve_2d_test_util.h"
+              "${AOM_ROOT}/test/av1_fwd_txfm1d_test.cc"
+              "${AOM_ROOT}/test/av1_fwd_txfm2d_test.cc"
+              "${AOM_ROOT}/test/av1_inv_txfm1d_test.cc"
+              "${AOM_ROOT}/test/av1_inv_txfm2d_test.cc"
+              "${AOM_ROOT}/test/av1_round_shift_array_test.cc"
+              "${AOM_ROOT}/test/av1_txfm_test.cc"
+              "${AOM_ROOT}/test/av1_txfm_test.h"
+              "${AOM_ROOT}/test/av1_wedge_utils_test.cc"
+              "${AOM_ROOT}/test/blend_a64_mask_1d_test.cc"
+              "${AOM_ROOT}/test/blend_a64_mask_test.cc"
+              "${AOM_ROOT}/test/comp_avg_pred_test.cc"
+              "${AOM_ROOT}/test/comp_avg_pred_test.h"
+              "${AOM_ROOT}/test/comp_mask_variance_test.cc"
+              "${AOM_ROOT}/test/encodetxb_test.cc"
+              "${AOM_ROOT}/test/error_block_test.cc"
+              "${AOM_ROOT}/test/fft_test.cc"
+              "${AOM_ROOT}/test/fwht4x4_test.cc"
+              "${AOM_ROOT}/test/masked_sad_test.cc"
+              "${AOM_ROOT}/test/masked_variance_test.cc"
+              "${AOM_ROOT}/test/motion_vector_test.cc"
+              "${AOM_ROOT}/test/noise_model_test.cc"
+              "${AOM_ROOT}/test/obmc_sad_test.cc"
+              "${AOM_ROOT}/test/obmc_variance_test.cc"
+              "${AOM_ROOT}/test/pickrst_test.cc"
+              "${AOM_ROOT}/test/sad_test.cc"
+              "${AOM_ROOT}/test/subtract_test.cc"
+              "${AOM_ROOT}/test/reconinter_test.cc"
+              "${AOM_ROOT}/test/sum_squares_test.cc"
+              "${AOM_ROOT}/test/variance_test.cc"
+              "${AOM_ROOT}/test/wiener_test.cc"
+              "${AOM_ROOT}/test/warp_filter_test.cc"
+              "${AOM_ROOT}/test/warp_filter_test_util.cc"
+              "${AOM_ROOT}/test/warp_filter_test_util.h")
+
+  list(APPEND AOM_UNIT_TEST_ENCODER_INTRIN_SSE4_1
+              "${AOM_ROOT}/test/av1_highbd_iht_test.cc"
+              "${AOM_ROOT}/test/av1_quantize_test.cc"
+              "${AOM_ROOT}/test/corner_match_test.cc"
+              "${AOM_ROOT}/test/quantize_func_test.cc"
+              "${AOM_ROOT}/test/simd_cmp_sse4.cc")
+
+  if(HAVE_SSE4_1)
+    list(APPEND AOM_UNIT_TEST_ENCODER_SOURCES
+                "${AOM_ROOT}/test/av1_convolve_scale_test.cc"
+                "${AOM_ROOT}/test/av1_horz_only_frame_superres_test.cc"
+                "${AOM_ROOT}/test/intra_edge_test.cc")
+
+  endif()
+
+  if(HAVE_SSE4_2)
+    list(APPEND AOM_UNIT_TEST_ENCODER_SOURCES "${AOM_ROOT}/test/hash_test.cc")
+  endif()
+
+endif()
+
+if(ENABLE_TESTS)
+  find_package(PythonInterp)
+  if(NOT PYTHONINTERP_FOUND)
+    message(FATAL_ERROR
+              "--- Unit tests require Python, rerun cmake with "
+              "-DENABLE_TESTS=0 to avoid this error, or install Python and "
+              "make sure it's in your PATH.")
+  endif()
+
+  if(MSVC) # Force static run time to avoid collisions with googletest.
+    include("${AOM_ROOT}/build/cmake/msvc_runtime.cmake")
+    if(BUILD_SHARED_LIBS)
+      set(AOM_DISABLE_GTEST_CMAKE 1)
+    endif()
+  endif()
+
+  if(BUILD_SHARED_LIBS AND APPLE) # Silence an RPATH warning.
+    set(CMAKE_MACOSX_RPATH 1)
+  endif()
+
+  include_directories(
+    "${AOM_ROOT}/third_party/googletest/src/googletest/include")
+
+  if(AOM_DISABLE_GTEST_CMAKE)
+    include_directories("${AOM_ROOT}/third_party/googletest/src/googletest")
+    add_library(
+      gtest
+      STATIC
+      "${AOM_ROOT}/third_party/googletest/src/googletest/src/gtest-all.cc")
+  else()
+    add_subdirectory("${AOM_ROOT}/third_party/googletest/src/googletest"
+                     EXCLUDE_FROM_ALL)
+  endif()
+endif()
+
+# Setup testdata download targets, test build targets, and test run targets. The
+# libaom and app util targets must exist before this function is called.
+function(setup_aom_test_targets)
+
+  # TODO(tomfinegan): Build speed optimization. $AOM_UNIT_TEST_COMMON_SOURCES
+  # and $AOM_UNIT_TEST_ENCODER_SOURCES are very large. The build of test targets
+  # could be sped up (on multicore build machines) by compiling sources in each
+  # list into separate object library targets, and then linking them into
+  # test_libaom.
+  add_library(test_aom_common OBJECT ${AOM_UNIT_TEST_COMMON_SOURCES})
+  add_dependencies(test_aom_common aom)
+
+  if(CONFIG_AV1_DECODER)
+    add_library(test_aom_decoder OBJECT ${AOM_UNIT_TEST_DECODER_SOURCES})
+    add_dependencies(test_aom_decoder aom)
+  endif()
+
+  if(CONFIG_AV1_ENCODER)
+    add_library(test_aom_encoder OBJECT ${AOM_UNIT_TEST_ENCODER_SOURCES})
+    add_dependencies(test_aom_encoder aom)
+  endif()
+
+  add_executable(test_libaom ${AOM_UNIT_TEST_WRAPPER_SOURCES}
+                 $<TARGET_OBJECTS:aom_common_app_util>
+                 $<TARGET_OBJECTS:test_aom_common>)
+  list(APPEND AOM_APP_TARGETS test_libaom)
+
+  if(CONFIG_AV1_DECODER)
+    target_sources(test_libaom PRIVATE $<TARGET_OBJECTS:aom_decoder_app_util>
+                   $<TARGET_OBJECTS:test_aom_decoder>)
+
+    if(ENABLE_DECODE_PERF_TESTS AND CONFIG_WEBM_IO)
+      target_sources(test_libaom PRIVATE ${AOM_DECODE_PERF_TEST_SOURCES})
+    endif()
+  endif()
+
+  if(CONFIG_AV1_ENCODER)
+    target_sources(test_libaom PRIVATE $<TARGET_OBJECTS:test_aom_encoder>
+                   $<TARGET_OBJECTS:aom_encoder_app_util>)
+
+    if(ENABLE_ENCODE_PERF_TESTS)
+      target_sources(test_libaom PRIVATE ${AOM_ENCODE_PERF_TEST_SOURCES})
+    endif()
+
+    if(NOT BUILD_SHARED_LIBS)
+      add_executable(test_intra_pred_speed ${AOM_TEST_INTRA_PRED_SPEED_SOURCES}
+                     $<TARGET_OBJECTS:aom_common_app_util>)
+      target_link_libraries(test_intra_pred_speed ${AOM_LIB_LINK_TYPE} aom
+                            gtest)
+      list(APPEND AOM_APP_TARGETS test_intra_pred_speed)
+    endif()
+  endif()
+
+  target_link_libraries(test_libaom ${AOM_LIB_LINK_TYPE} aom gtest)
+
+  if(CONFIG_LIBYUV)
+    target_sources(test_libaom PRIVATE $<TARGET_OBJECTS:yuv>)
+  endif()
+  if(CONFIG_WEBM_IO)
+    target_sources(test_libaom PRIVATE $<TARGET_OBJECTS:webm>)
+  endif()
+  if(HAVE_SSE2)
+    add_intrinsics_source_to_target("-msse2" "test_libaom"
+                                    "AOM_UNIT_TEST_COMMON_INTRIN_SSE2")
+  endif()
+  if(HAVE_SSSE3)
+    add_intrinsics_source_to_target("-mssse3" "test_libaom"
+                                    "AOM_UNIT_TEST_COMMON_INTRIN_SSSE3")
+  endif()
+  if(HAVE_SSE4_1)
+    add_intrinsics_source_to_target("-msse4.1" "test_libaom"
+                                    "AOM_UNIT_TEST_COMMON_INTRIN_SSE4_1")
+    if(CONFIG_AV1_ENCODER)
+      if(AOM_UNIT_TEST_ENCODER_INTRIN_SSE4_1)
+        add_intrinsics_source_to_target("-msse4.1" "test_libaom"
+                                        "AOM_UNIT_TEST_ENCODER_INTRIN_SSE4_1")
+      endif()
+    endif()
+  endif()
+  if(HAVE_AVX2)
+    add_intrinsics_source_to_target("-mavx2" "test_libaom"
+                                    "AOM_UNIT_TEST_COMMON_INTRIN_AVX2")
+  endif()
+  if(HAVE_NEON)
+    add_intrinsics_source_to_target("${AOM_NEON_INTRIN_FLAG}" "test_libaom"
+                                    "AOM_UNIT_TEST_COMMON_INTRIN_NEON")
+  endif()
+
+  if(ENABLE_TESTDATA)
+    make_test_data_lists("${AOM_UNIT_TEST_DATA_LIST_FILE}" test_files
+                         test_file_checksums)
+    list(LENGTH test_files num_test_files)
+    list(LENGTH test_file_checksums num_test_file_checksums)
+
+    math(EXPR max_file_index "${num_test_files} - 1")
+    foreach(test_index RANGE ${max_file_index})
+      list(GET test_files ${test_index} test_file)
+      list(GET test_file_checksums ${test_index} test_file_checksum)
+      add_custom_target(testdata_${test_index}
+                        COMMAND
+                          ${CMAKE_COMMAND} -DAOM_CONFIG_DIR="${AOM_CONFIG_DIR}"
+                          -DAOM_ROOT="${AOM_ROOT}"
+                          -DAOM_TEST_FILE="${test_file}"
+                          -DAOM_TEST_CHECKSUM=${test_file_checksum} -P
+                          "${AOM_ROOT}/test/test_data_download_worker.cmake")
+      list(APPEND testdata_targets testdata_${test_index})
+    endforeach()
+
+    # Create a custom build target for running each test data download target.
+    add_custom_target(testdata)
+    add_dependencies(testdata ${testdata_targets})
+
+    # Skip creation of test run targets when generating for Visual Studio and
+    # Xcode unless the user explicitly requests IDE test hosting. This is done
+    # to make build cycles in the IDE tolerable when the IDE command for build
+    # project is used to build AOM. Default behavior in IDEs is to build all
+    # targets, and the test run takes hours.
+    if(((NOT MSVC) AND (NOT XCODE)) OR ENABLE_IDE_TEST_HOSTING)
+
+      # Pick a reasonable number of targets (this controls parallelization).
+      processorcount(num_test_targets)
+      if(num_test_targets EQUAL 0) # Just default to 10 targets when there's no
+                                   # processor count available.
+        set(num_test_targets 10)
+      endif()
+
+      math(EXPR max_shard_index "${num_test_targets} - 1")
+      foreach(shard_index RANGE ${max_shard_index})
+        set(test_name "test_${shard_index}")
+        add_custom_target(${test_name}
+                          COMMAND ${CMAKE_COMMAND}
+                                  -DGTEST_SHARD_INDEX=${shard_index}
+                                  -DGTEST_TOTAL_SHARDS=${num_test_targets}
+                                  -DTEST_LIBAOM=$<TARGET_FILE:test_libaom> -P
+                                  "${AOM_ROOT}/test/test_runner.cmake"
+                          DEPENDS testdata test_libaom)
+        list(APPEND test_targets ${test_name})
+      endforeach()
+      add_custom_target(runtests)
+      add_dependencies(runtests ${test_targets})
+    endif()
+  endif()
+
+  # Collect all variables containing libaom test source files.
+  get_cmake_property(all_cmake_vars VARIABLES)
+  foreach(var ${all_cmake_vars})
+
+    # https://github.com/cheshirekow/cmake_format/issues/34
+# cmake-format: off
+    if (("${var}" MATCHES "_TEST_" AND NOT
+         "${var}" MATCHES
+         "_DATA_\|_CMAKE_\|INTRA_PRED\|_COMPILED\|_HOSTING\|_PERF_\|CODER_")
+        OR (CONFIG_AV1_ENCODER AND ENABLE_ENCODE_PERF_TESTS AND
+            "${var}" MATCHES "_ENCODE_PERF_TEST_")
+        OR (CONFIG_AV1_DECODER AND ENABLE_DECODE_PERF_TESTS AND
+            "${var}" MATCHES "_DECODE_PERF_TEST_")
+        OR (CONFIG_AV1_ENCODER AND "${var}" MATCHES "_TEST_ENCODER_")
+        OR (CONFIG_AV1_DECODER AND  "${var}" MATCHES "_TEST_DECODER_"))
+      list(APPEND aom_test_source_vars ${var})
+    endif()
+    # cmake-format: on
+  endforeach()
+
+  # Libaom_test_srcs.txt generation.
+  set(libaom_test_srcs_txt_file "${AOM_CONFIG_DIR}/libaom_test_srcs.txt")
+  file(WRITE "${libaom_test_srcs_txt_file}"
+             "# This file is generated. DO NOT EDIT.\n")
+
+  # Static source file list first.
+  foreach(aom_test_source_var ${aom_test_source_vars})
+    foreach(file ${${aom_test_source_var}})
+      if(NOT "${file}" MATCHES "${AOM_CONFIG_DIR}")
+        string(REPLACE "${AOM_ROOT}/" "" file "${file}")
+        file(APPEND "${libaom_test_srcs_txt_file}" "${file}\n")
+      endif()
+    endforeach()
+  endforeach()
+
+  set(AOM_APP_TARGETS ${AOM_APP_TARGETS} PARENT_SCOPE)
+endfunction()
diff --git a/third_party/aom/test/test_data_download_worker.cmake b/third_party/aom/test/test_data_download_worker.cmake
new file mode 100644
index 0000000000..dc803497da
--- /dev/null
+++ b/third_party/aom/test/test_data_download_worker.cmake
@@ -0,0 +1,46 @@
+#
+# Copyright (c) 2017, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and the
+# Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License was
+# not distributed with this source code in the LICENSE file, you can obtain it
+# at www.aomedia.org/license/software. If the Alliance for Open Media Patent
+# License 1.0 was not distributed with this source code in the PATENTS file, you
+# can obtain it at www.aomedia.org/license/patent.
+#
+include("${AOM_ROOT}/test/test_data_util.cmake")
+
+# https://github.com/cheshirekow/cmake_format/issues/34
+# cmake-format: off
+if (NOT AOM_ROOT OR NOT AOM_CONFIG_DIR OR NOT AOM_TEST_FILE
+    OR NOT AOM_TEST_CHECKSUM)
+  message(FATAL_ERROR
+          "AOM_ROOT, AOM_CONFIG_DIR, AOM_TEST_FILE and AOM_TEST_CHECKSUM must be
+          defined.")
+endif ()
+# cmake-format: on
+
+set(AOM_TEST_DATA_URL "http://storage.googleapis.com/aom-test-data")
+
+if(NOT AOM_TEST_DATA_PATH)
+  set(AOM_TEST_DATA_PATH "$ENV{LIBAOM_TEST_DATA_PATH}")
+endif()
+
+if("${AOM_TEST_DATA_PATH}" STREQUAL "")
+  message(WARNING
+            "Writing test data to ${AOM_CONFIG_DIR}, set "
+            "$LIBAOM_TEST_DATA_PATH in your environment to avoid this warning.")
+  set(AOM_TEST_DATA_PATH "${AOM_CONFIG_DIR}")
+endif()
+
+if(NOT EXISTS "${AOM_TEST_DATA_PATH}")
+  file(MAKE_DIRECTORY "${AOM_TEST_DATA_PATH}")
+endif()
+
+expand_test_file_paths("AOM_TEST_FILE" "${AOM_TEST_DATA_PATH}" "filepath")
+expand_test_file_paths("AOM_TEST_FILE" "${AOM_TEST_DATA_URL}" "url")
+
+check_file("${filepath}" "${AOM_TEST_CHECKSUM}" "needs_download")
+if(needs_download)
+  download_test_file("${url}" "${AOM_TEST_CHECKSUM}" "${filepath}")
+endif()
diff --git a/third_party/aom/test/test_data_util.cmake b/third_party/aom/test/test_data_util.cmake
new file mode 100644
index 0000000000..45c951478a
--- /dev/null
+++ b/third_party/aom/test/test_data_util.cmake
@@ -0,0 +1,598 @@
+#
+# Copyright (c) 2017, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and the
+# Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License was
+# not distributed with this source code in the LICENSE file, you can obtain it
+# at www.aomedia.org/license/software. If the Alliance for Open Media Patent
+# License 1.0 was not distributed with this source code in the PATENTS file, you
+# can obtain it at www.aomedia.org/license/patent.
+#
+
+list(APPEND AOM_TEST_DATA_FILE_NAMES
+            "hantro_collage_w352h288.yuv"
+            "hantro_odd.yuv"
+            "invalid-bug-1814.ivf"
+            "invalid-bug-1814.ivf.res"
+            "invalid-oss-fuzz-10061.ivf"
+            "invalid-oss-fuzz-10061.ivf.res"
+            "invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf"
+            "invalid-oss-fuzz-10117-mc-buf-use-highbd.ivf.res"
+            "invalid-oss-fuzz-10227.ivf"
+            "invalid-oss-fuzz-10227.ivf.res"
+            "invalid-oss-fuzz-9463.ivf"
+            "invalid-oss-fuzz-9463.ivf.res"
+            "invalid-oss-fuzz-9482.ivf"
+            "invalid-oss-fuzz-9482.ivf.res"
+            "invalid-oss-fuzz-9720.ivf"
+            "invalid-oss-fuzz-9720.ivf.res"
+            "park_joy_90p_10_420.y4m"
+            "park_joy_90p_10_422.y4m"
+            "park_joy_90p_10_444.y4m"
+            "park_joy_90p_12_420.y4m"
+            "park_joy_90p_12_422.y4m"
+            "park_joy_90p_12_444.y4m"
+            "park_joy_90p_8_420_a10-1.y4m"
+            "park_joy_90p_8_420.y4m"
+            "park_joy_90p_8_420_monochrome.y4m"
+            "park_joy_90p_8_420_vertical_csp.y4m"
+            "park_joy_90p_8_422.y4m"
+            "park_joy_90p_8_444.y4m"
+            "desktop_credits.y4m"
+            "niklas_1280_720_30.y4m"
+            "rush_hour_444.y4m"
+            "screendata.y4m"
+            "niklas_640_480_30.yuv"
+            "vase10x10.yuv")
+
+if(ENABLE_DECODE_PERF_TESTS AND CONFIG_AV1_ENCODER)
+  list(APPEND AOM_TEST_DATA_FILE_NAMES "niklas_1280_720_30.yuv")
+endif()
+
+if(CONFIG_AV1_DECODER)
+  list(APPEND AOM_TEST_DATA_FILE_NAMES
+              "av1-1-b8-00-quantizer-00.ivf"
+              "av1-1-b8-00-quantizer-00.ivf.md5"
+              "av1-1-b8-00-quantizer-01.ivf"
+              "av1-1-b8-00-quantizer-01.ivf.md5"
+              "av1-1-b8-00-quantizer-02.ivf"
+              "av1-1-b8-00-quantizer-02.ivf.md5"
+              "av1-1-b8-00-quantizer-03.ivf"
+              "av1-1-b8-00-quantizer-03.ivf.md5"
+              "av1-1-b8-00-quantizer-04.ivf"
+              "av1-1-b8-00-quantizer-04.ivf.md5"
+              "av1-1-b8-00-quantizer-05.ivf"
+              "av1-1-b8-00-quantizer-05.ivf.md5"
+              "av1-1-b8-00-quantizer-06.ivf"
+              "av1-1-b8-00-quantizer-06.ivf.md5"
+              "av1-1-b8-00-quantizer-07.ivf"
+              "av1-1-b8-00-quantizer-07.ivf.md5"
+              "av1-1-b8-00-quantizer-08.ivf"
+              "av1-1-b8-00-quantizer-08.ivf.md5"
+              "av1-1-b8-00-quantizer-09.ivf"
+              "av1-1-b8-00-quantizer-09.ivf.md5"
+              "av1-1-b8-00-quantizer-10.ivf"
+              "av1-1-b8-00-quantizer-10.ivf.md5"
+              "av1-1-b8-00-quantizer-11.ivf"
+              "av1-1-b8-00-quantizer-11.ivf.md5"
+              "av1-1-b8-00-quantizer-12.ivf"
+              "av1-1-b8-00-quantizer-12.ivf.md5"
+              "av1-1-b8-00-quantizer-13.ivf"
+              "av1-1-b8-00-quantizer-13.ivf.md5"
+              "av1-1-b8-00-quantizer-14.ivf"
+              "av1-1-b8-00-quantizer-14.ivf.md5"
+              "av1-1-b8-00-quantizer-15.ivf"
+              "av1-1-b8-00-quantizer-15.ivf.md5"
+              "av1-1-b8-00-quantizer-16.ivf"
+              "av1-1-b8-00-quantizer-16.ivf.md5"
+              "av1-1-b8-00-quantizer-17.ivf"
+              "av1-1-b8-00-quantizer-17.ivf.md5"
+              "av1-1-b8-00-quantizer-18.ivf"
+              "av1-1-b8-00-quantizer-18.ivf.md5"
+              "av1-1-b8-00-quantizer-19.ivf"
+              "av1-1-b8-00-quantizer-19.ivf.md5"
+              "av1-1-b8-00-quantizer-20.ivf"
+              "av1-1-b8-00-quantizer-20.ivf.md5"
+              "av1-1-b8-00-quantizer-21.ivf"
+              "av1-1-b8-00-quantizer-21.ivf.md5"
+              "av1-1-b8-00-quantizer-22.ivf"
+              "av1-1-b8-00-quantizer-22.ivf.md5"
+              "av1-1-b8-00-quantizer-23.ivf"
+              "av1-1-b8-00-quantizer-23.ivf.md5"
+              "av1-1-b8-00-quantizer-24.ivf"
+              "av1-1-b8-00-quantizer-24.ivf.md5"
+              "av1-1-b8-00-quantizer-25.ivf"
+              "av1-1-b8-00-quantizer-25.ivf.md5"
+              "av1-1-b8-00-quantizer-26.ivf"
+              "av1-1-b8-00-quantizer-26.ivf.md5"
+              "av1-1-b8-00-quantizer-27.ivf"
+              "av1-1-b8-00-quantizer-27.ivf.md5"
+              "av1-1-b8-00-quantizer-28.ivf"
+              "av1-1-b8-00-quantizer-28.ivf.md5"
+              "av1-1-b8-00-quantizer-29.ivf"
+              "av1-1-b8-00-quantizer-29.ivf.md5"
+              "av1-1-b8-00-quantizer-30.ivf"
+              "av1-1-b8-00-quantizer-30.ivf.md5"
+              "av1-1-b8-00-quantizer-31.ivf"
+              "av1-1-b8-00-quantizer-31.ivf.md5"
+              "av1-1-b8-00-quantizer-32.ivf"
+              "av1-1-b8-00-quantizer-32.ivf.md5"
+              "av1-1-b8-00-quantizer-33.ivf"
+              "av1-1-b8-00-quantizer-33.ivf.md5"
+              "av1-1-b8-00-quantizer-34.ivf"
+              "av1-1-b8-00-quantizer-34.ivf.md5"
+              "av1-1-b8-00-quantizer-35.ivf"
+              "av1-1-b8-00-quantizer-35.ivf.md5"
+              "av1-1-b8-00-quantizer-36.ivf"
+              "av1-1-b8-00-quantizer-36.ivf.md5"
+              "av1-1-b8-00-quantizer-37.ivf"
+              "av1-1-b8-00-quantizer-37.ivf.md5"
+              "av1-1-b8-00-quantizer-38.ivf"
+              "av1-1-b8-00-quantizer-38.ivf.md5"
+              "av1-1-b8-00-quantizer-39.ivf"
+              "av1-1-b8-00-quantizer-39.ivf.md5"
+              "av1-1-b8-00-quantizer-40.ivf"
+              "av1-1-b8-00-quantizer-40.ivf.md5"
+              "av1-1-b8-00-quantizer-41.ivf"
+              "av1-1-b8-00-quantizer-41.ivf.md5"
+              "av1-1-b8-00-quantizer-42.ivf"
+              "av1-1-b8-00-quantizer-42.ivf.md5"
+              "av1-1-b8-00-quantizer-43.ivf"
+              "av1-1-b8-00-quantizer-43.ivf.md5"
+              "av1-1-b8-00-quantizer-44.ivf"
+              "av1-1-b8-00-quantizer-44.ivf.md5"
+              "av1-1-b8-00-quantizer-45.ivf"
+              "av1-1-b8-00-quantizer-45.ivf.md5"
+              "av1-1-b8-00-quantizer-46.ivf"
+              "av1-1-b8-00-quantizer-46.ivf.md5"
+              "av1-1-b8-00-quantizer-47.ivf"
+              "av1-1-b8-00-quantizer-47.ivf.md5"
+              "av1-1-b8-00-quantizer-48.ivf"
+              "av1-1-b8-00-quantizer-48.ivf.md5"
+              "av1-1-b8-00-quantizer-49.ivf"
+              "av1-1-b8-00-quantizer-49.ivf.md5"
+              "av1-1-b8-00-quantizer-50.ivf"
+              "av1-1-b8-00-quantizer-50.ivf.md5"
+              "av1-1-b8-00-quantizer-51.ivf"
+              "av1-1-b8-00-quantizer-51.ivf.md5"
+              "av1-1-b8-00-quantizer-52.ivf"
+              "av1-1-b8-00-quantizer-52.ivf.md5"
+              "av1-1-b8-00-quantizer-53.ivf"
+              "av1-1-b8-00-quantizer-53.ivf.md5"
+              "av1-1-b8-00-quantizer-54.ivf"
+              "av1-1-b8-00-quantizer-54.ivf.md5"
+              "av1-1-b8-00-quantizer-55.ivf"
+              "av1-1-b8-00-quantizer-55.ivf.md5"
+              "av1-1-b8-00-quantizer-56.ivf"
+              "av1-1-b8-00-quantizer-56.ivf.md5"
+              "av1-1-b8-00-quantizer-57.ivf"
+              "av1-1-b8-00-quantizer-57.ivf.md5"
+              "av1-1-b8-00-quantizer-58.ivf"
+              "av1-1-b8-00-quantizer-58.ivf.md5"
+              "av1-1-b8-00-quantizer-59.ivf"
+              "av1-1-b8-00-quantizer-59.ivf.md5"
+              "av1-1-b8-00-quantizer-60.ivf"
+              "av1-1-b8-00-quantizer-60.ivf.md5"
+              "av1-1-b8-00-quantizer-61.ivf"
+              "av1-1-b8-00-quantizer-61.ivf.md5"
+              "av1-1-b8-00-quantizer-62.ivf"
+              "av1-1-b8-00-quantizer-62.ivf.md5"
+              "av1-1-b8-00-quantizer-63.ivf"
+              "av1-1-b8-00-quantizer-63.ivf.md5"
+              "av1-1-b10-00-quantizer-00.ivf"
+              "av1-1-b10-00-quantizer-00.ivf.md5"
+              "av1-1-b10-00-quantizer-01.ivf"
+              "av1-1-b10-00-quantizer-01.ivf.md5"
+              "av1-1-b10-00-quantizer-02.ivf"
+              "av1-1-b10-00-quantizer-02.ivf.md5"
+              "av1-1-b10-00-quantizer-03.ivf"
+              "av1-1-b10-00-quantizer-03.ivf.md5"
+              "av1-1-b10-00-quantizer-04.ivf"
+              "av1-1-b10-00-quantizer-04.ivf.md5"
+              "av1-1-b10-00-quantizer-05.ivf"
+              "av1-1-b10-00-quantizer-05.ivf.md5"
+              "av1-1-b10-00-quantizer-06.ivf"
+              "av1-1-b10-00-quantizer-06.ivf.md5"
+              "av1-1-b10-00-quantizer-07.ivf"
+              "av1-1-b10-00-quantizer-07.ivf.md5"
+              "av1-1-b10-00-quantizer-08.ivf"
+              "av1-1-b10-00-quantizer-08.ivf.md5"
+              "av1-1-b10-00-quantizer-09.ivf"
+              "av1-1-b10-00-quantizer-09.ivf.md5"
+              "av1-1-b10-00-quantizer-10.ivf"
+              "av1-1-b10-00-quantizer-10.ivf.md5"
+              "av1-1-b10-00-quantizer-11.ivf"
+              "av1-1-b10-00-quantizer-11.ivf.md5"
+              "av1-1-b10-00-quantizer-12.ivf"
+              "av1-1-b10-00-quantizer-12.ivf.md5"
+              "av1-1-b10-00-quantizer-13.ivf"
+              "av1-1-b10-00-quantizer-13.ivf.md5"
+              "av1-1-b10-00-quantizer-14.ivf"
+              "av1-1-b10-00-quantizer-14.ivf.md5"
+              "av1-1-b10-00-quantizer-15.ivf"
+              "av1-1-b10-00-quantizer-15.ivf.md5"
+              "av1-1-b10-00-quantizer-16.ivf"
+              "av1-1-b10-00-quantizer-16.ivf.md5"
+              "av1-1-b10-00-quantizer-17.ivf"
+              "av1-1-b10-00-quantizer-17.ivf.md5"
+              "av1-1-b10-00-quantizer-18.ivf"
+              "av1-1-b10-00-quantizer-18.ivf.md5"
+              "av1-1-b10-00-quantizer-19.ivf"
+              "av1-1-b10-00-quantizer-19.ivf.md5"
+              "av1-1-b10-00-quantizer-20.ivf"
+              "av1-1-b10-00-quantizer-20.ivf.md5"
+              "av1-1-b10-00-quantizer-21.ivf"
+              "av1-1-b10-00-quantizer-21.ivf.md5"
+              "av1-1-b10-00-quantizer-22.ivf"
+              "av1-1-b10-00-quantizer-22.ivf.md5"
+              "av1-1-b10-00-quantizer-23.ivf"
+              "av1-1-b10-00-quantizer-23.ivf.md5"
+              "av1-1-b10-00-quantizer-24.ivf"
+              "av1-1-b10-00-quantizer-24.ivf.md5"
+              "av1-1-b10-00-quantizer-25.ivf"
+              "av1-1-b10-00-quantizer-25.ivf.md5"
+              "av1-1-b10-00-quantizer-26.ivf"
+              "av1-1-b10-00-quantizer-26.ivf.md5"
+              "av1-1-b10-00-quantizer-27.ivf"
+              "av1-1-b10-00-quantizer-27.ivf.md5"
+              "av1-1-b10-00-quantizer-28.ivf"
+              "av1-1-b10-00-quantizer-28.ivf.md5"
+              "av1-1-b10-00-quantizer-29.ivf"
+              "av1-1-b10-00-quantizer-29.ivf.md5"
+              "av1-1-b10-00-quantizer-30.ivf"
+              "av1-1-b10-00-quantizer-30.ivf.md5"
+              "av1-1-b10-00-quantizer-31.ivf"
+              "av1-1-b10-00-quantizer-31.ivf.md5"
+              "av1-1-b10-00-quantizer-32.ivf"
+              "av1-1-b10-00-quantizer-32.ivf.md5"
+              "av1-1-b10-00-quantizer-33.ivf"
+              "av1-1-b10-00-quantizer-33.ivf.md5"
+              "av1-1-b10-00-quantizer-34.ivf"
+              "av1-1-b10-00-quantizer-34.ivf.md5"
+              "av1-1-b10-00-quantizer-35.ivf"
+              "av1-1-b10-00-quantizer-35.ivf.md5"
+              "av1-1-b10-00-quantizer-36.ivf"
+              "av1-1-b10-00-quantizer-36.ivf.md5"
+              "av1-1-b10-00-quantizer-37.ivf"
+              "av1-1-b10-00-quantizer-37.ivf.md5"
+              "av1-1-b10-00-quantizer-38.ivf"
+              "av1-1-b10-00-quantizer-38.ivf.md5"
+              "av1-1-b10-00-quantizer-39.ivf"
+              "av1-1-b10-00-quantizer-39.ivf.md5"
+              "av1-1-b10-00-quantizer-40.ivf"
+              "av1-1-b10-00-quantizer-40.ivf.md5"
+              "av1-1-b10-00-quantizer-41.ivf"
+              "av1-1-b10-00-quantizer-41.ivf.md5"
+              "av1-1-b10-00-quantizer-42.ivf"
+              "av1-1-b10-00-quantizer-42.ivf.md5"
+              "av1-1-b10-00-quantizer-43.ivf"
+              "av1-1-b10-00-quantizer-43.ivf.md5"
+              "av1-1-b10-00-quantizer-44.ivf"
+              "av1-1-b10-00-quantizer-44.ivf.md5"
+              "av1-1-b10-00-quantizer-45.ivf"
+              "av1-1-b10-00-quantizer-45.ivf.md5"
+              "av1-1-b10-00-quantizer-46.ivf"
+              "av1-1-b10-00-quantizer-46.ivf.md5"
+              "av1-1-b10-00-quantizer-47.ivf"
+              "av1-1-b10-00-quantizer-47.ivf.md5"
+              "av1-1-b10-00-quantizer-48.ivf"
+              "av1-1-b10-00-quantizer-48.ivf.md5"
+              "av1-1-b10-00-quantizer-49.ivf"
+              "av1-1-b10-00-quantizer-49.ivf.md5"
+              "av1-1-b10-00-quantizer-50.ivf"
+              "av1-1-b10-00-quantizer-50.ivf.md5"
+              "av1-1-b10-00-quantizer-51.ivf"
+              "av1-1-b10-00-quantizer-51.ivf.md5"
+              "av1-1-b10-00-quantizer-52.ivf"
+              "av1-1-b10-00-quantizer-52.ivf.md5"
+              "av1-1-b10-00-quantizer-53.ivf"
+              "av1-1-b10-00-quantizer-53.ivf.md5"
+              "av1-1-b10-00-quantizer-54.ivf"
+              "av1-1-b10-00-quantizer-54.ivf.md5"
+              "av1-1-b10-00-quantizer-55.ivf"
+              "av1-1-b10-00-quantizer-55.ivf.md5"
+              "av1-1-b10-00-quantizer-56.ivf"
+              "av1-1-b10-00-quantizer-56.ivf.md5"
+              "av1-1-b10-00-quantizer-57.ivf"
+              "av1-1-b10-00-quantizer-57.ivf.md5"
+              "av1-1-b10-00-quantizer-58.ivf"
+              "av1-1-b10-00-quantizer-58.ivf.md5"
+              "av1-1-b10-00-quantizer-59.ivf"
+              "av1-1-b10-00-quantizer-59.ivf.md5"
+              "av1-1-b10-00-quantizer-60.ivf"
+              "av1-1-b10-00-quantizer-60.ivf.md5"
+              "av1-1-b10-00-quantizer-61.ivf"
+              "av1-1-b10-00-quantizer-61.ivf.md5"
+              "av1-1-b10-00-quantizer-62.ivf"
+              "av1-1-b10-00-quantizer-62.ivf.md5"
+              "av1-1-b10-00-quantizer-63.ivf"
+              "av1-1-b10-00-quantizer-63.ivf.md5"
+              "av1-1-b8-01-size-16x16.ivf"
+              "av1-1-b8-01-size-16x16.ivf.md5"
+              "av1-1-b8-01-size-16x18.ivf"
+              "av1-1-b8-01-size-16x18.ivf.md5"
+              "av1-1-b8-01-size-16x32.ivf"
+              "av1-1-b8-01-size-16x32.ivf.md5"
+              "av1-1-b8-01-size-16x34.ivf"
+              "av1-1-b8-01-size-16x34.ivf.md5"
+              "av1-1-b8-01-size-16x64.ivf"
+              "av1-1-b8-01-size-16x64.ivf.md5"
+              "av1-1-b8-01-size-16x66.ivf"
+              "av1-1-b8-01-size-16x66.ivf.md5"
+              "av1-1-b8-01-size-18x16.ivf"
+              "av1-1-b8-01-size-18x16.ivf.md5"
+              "av1-1-b8-01-size-18x18.ivf"
+              "av1-1-b8-01-size-18x18.ivf.md5"
+              "av1-1-b8-01-size-18x32.ivf"
+              "av1-1-b8-01-size-18x32.ivf.md5"
+              "av1-1-b8-01-size-18x34.ivf"
+              "av1-1-b8-01-size-18x34.ivf.md5"
+              "av1-1-b8-01-size-18x64.ivf"
+              "av1-1-b8-01-size-18x64.ivf.md5"
+              "av1-1-b8-01-size-18x66.ivf"
+              "av1-1-b8-01-size-18x66.ivf.md5"
+              "av1-1-b8-01-size-196x196.ivf"
+              "av1-1-b8-01-size-196x196.ivf.md5"
+              "av1-1-b8-01-size-196x198.ivf"
+              "av1-1-b8-01-size-196x198.ivf.md5"
+              "av1-1-b8-01-size-196x200.ivf"
+              "av1-1-b8-01-size-196x200.ivf.md5"
+              "av1-1-b8-01-size-196x202.ivf"
+              "av1-1-b8-01-size-196x202.ivf.md5"
+              "av1-1-b8-01-size-196x208.ivf"
+              "av1-1-b8-01-size-196x208.ivf.md5"
+              "av1-1-b8-01-size-196x210.ivf"
+              "av1-1-b8-01-size-196x210.ivf.md5"
+              "av1-1-b8-01-size-196x224.ivf"
+              "av1-1-b8-01-size-196x224.ivf.md5"
+              "av1-1-b8-01-size-196x226.ivf"
+              "av1-1-b8-01-size-196x226.ivf.md5"
+              "av1-1-b8-01-size-198x196.ivf"
+              "av1-1-b8-01-size-198x196.ivf.md5"
+              "av1-1-b8-01-size-198x198.ivf"
+              "av1-1-b8-01-size-198x198.ivf.md5"
+              "av1-1-b8-01-size-198x200.ivf"
+              "av1-1-b8-01-size-198x200.ivf.md5"
+              "av1-1-b8-01-size-198x202.ivf"
+              "av1-1-b8-01-size-198x202.ivf.md5"
+              "av1-1-b8-01-size-198x208.ivf"
+              "av1-1-b8-01-size-198x208.ivf.md5"
+              "av1-1-b8-01-size-198x210.ivf"
+              "av1-1-b8-01-size-198x210.ivf.md5"
+              "av1-1-b8-01-size-198x224.ivf"
+              "av1-1-b8-01-size-198x224.ivf.md5"
+              "av1-1-b8-01-size-198x226.ivf"
+              "av1-1-b8-01-size-198x226.ivf.md5"
+              "av1-1-b8-01-size-200x196.ivf"
+              "av1-1-b8-01-size-200x196.ivf.md5"
+              "av1-1-b8-01-size-200x198.ivf"
+              "av1-1-b8-01-size-200x198.ivf.md5"
+              "av1-1-b8-01-size-200x200.ivf"
+              "av1-1-b8-01-size-200x200.ivf.md5"
+              "av1-1-b8-01-size-200x202.ivf"
+              "av1-1-b8-01-size-200x202.ivf.md5"
+              "av1-1-b8-01-size-200x208.ivf"
+              "av1-1-b8-01-size-200x208.ivf.md5"
+              "av1-1-b8-01-size-200x210.ivf"
+              "av1-1-b8-01-size-200x210.ivf.md5"
+              "av1-1-b8-01-size-200x224.ivf"
+              "av1-1-b8-01-size-200x224.ivf.md5"
+              "av1-1-b8-01-size-200x226.ivf"
+              "av1-1-b8-01-size-200x226.ivf.md5"
+              "av1-1-b8-01-size-202x196.ivf"
+              "av1-1-b8-01-size-202x196.ivf.md5"
+              "av1-1-b8-01-size-202x198.ivf"
+              "av1-1-b8-01-size-202x198.ivf.md5"
+              "av1-1-b8-01-size-202x200.ivf"
+              "av1-1-b8-01-size-202x200.ivf.md5"
+              "av1-1-b8-01-size-202x202.ivf"
+              "av1-1-b8-01-size-202x202.ivf.md5"
+              "av1-1-b8-01-size-202x208.ivf"
+              "av1-1-b8-01-size-202x208.ivf.md5"
+              "av1-1-b8-01-size-202x210.ivf"
+              "av1-1-b8-01-size-202x210.ivf.md5"
+              "av1-1-b8-01-size-202x224.ivf"
+              "av1-1-b8-01-size-202x224.ivf.md5"
+              "av1-1-b8-01-size-202x226.ivf"
+              "av1-1-b8-01-size-202x226.ivf.md5"
+              "av1-1-b8-01-size-208x196.ivf"
+              "av1-1-b8-01-size-208x196.ivf.md5"
+              "av1-1-b8-01-size-208x198.ivf"
+              "av1-1-b8-01-size-208x198.ivf.md5"
+              "av1-1-b8-01-size-208x200.ivf"
+              "av1-1-b8-01-size-208x200.ivf.md5"
+              "av1-1-b8-01-size-208x202.ivf"
+              "av1-1-b8-01-size-208x202.ivf.md5"
+              "av1-1-b8-01-size-208x208.ivf"
+              "av1-1-b8-01-size-208x208.ivf.md5"
+              "av1-1-b8-01-size-208x210.ivf"
+              "av1-1-b8-01-size-208x210.ivf.md5"
+              "av1-1-b8-01-size-208x224.ivf"
+              "av1-1-b8-01-size-208x224.ivf.md5"
+              "av1-1-b8-01-size-208x226.ivf"
+              "av1-1-b8-01-size-208x226.ivf.md5"
+              "av1-1-b8-01-size-210x196.ivf"
+              "av1-1-b8-01-size-210x196.ivf.md5"
+              "av1-1-b8-01-size-210x198.ivf"
+              "av1-1-b8-01-size-210x198.ivf.md5"
+              "av1-1-b8-01-size-210x200.ivf"
+              "av1-1-b8-01-size-210x200.ivf.md5"
+              "av1-1-b8-01-size-210x202.ivf"
+              "av1-1-b8-01-size-210x202.ivf.md5"
+              "av1-1-b8-01-size-210x208.ivf"
+              "av1-1-b8-01-size-210x208.ivf.md5"
+              "av1-1-b8-01-size-210x210.ivf"
+              "av1-1-b8-01-size-210x210.ivf.md5"
+              "av1-1-b8-01-size-210x224.ivf"
+              "av1-1-b8-01-size-210x224.ivf.md5"
+              "av1-1-b8-01-size-210x226.ivf"
+              "av1-1-b8-01-size-210x226.ivf.md5"
+              "av1-1-b8-01-size-224x196.ivf"
+              "av1-1-b8-01-size-224x196.ivf.md5"
+              "av1-1-b8-01-size-224x198.ivf"
+              "av1-1-b8-01-size-224x198.ivf.md5"
+              "av1-1-b8-01-size-224x200.ivf"
+              "av1-1-b8-01-size-224x200.ivf.md5"
+              "av1-1-b8-01-size-224x202.ivf"
+              "av1-1-b8-01-size-224x202.ivf.md5"
+              "av1-1-b8-01-size-224x208.ivf"
+              "av1-1-b8-01-size-224x208.ivf.md5"
+              "av1-1-b8-01-size-224x210.ivf"
+              "av1-1-b8-01-size-224x210.ivf.md5"
+              "av1-1-b8-01-size-224x224.ivf"
+              "av1-1-b8-01-size-224x224.ivf.md5"
+              "av1-1-b8-01-size-224x226.ivf"
+              "av1-1-b8-01-size-224x226.ivf.md5"
+              "av1-1-b8-01-size-226x196.ivf"
+              "av1-1-b8-01-size-226x196.ivf.md5"
+              "av1-1-b8-01-size-226x198.ivf"
+              "av1-1-b8-01-size-226x198.ivf.md5"
+              "av1-1-b8-01-size-226x200.ivf"
+              "av1-1-b8-01-size-226x200.ivf.md5"
+              "av1-1-b8-01-size-226x202.ivf"
+              "av1-1-b8-01-size-226x202.ivf.md5"
+              "av1-1-b8-01-size-226x208.ivf"
+              "av1-1-b8-01-size-226x208.ivf.md5"
+              "av1-1-b8-01-size-226x210.ivf"
+              "av1-1-b8-01-size-226x210.ivf.md5"
+              "av1-1-b8-01-size-226x224.ivf"
+              "av1-1-b8-01-size-226x224.ivf.md5"
+              "av1-1-b8-01-size-226x226.ivf"
+              "av1-1-b8-01-size-226x226.ivf.md5"
+              "av1-1-b8-01-size-32x16.ivf"
+              "av1-1-b8-01-size-32x16.ivf.md5"
+              "av1-1-b8-01-size-32x18.ivf"
+              "av1-1-b8-01-size-32x18.ivf.md5"
+              "av1-1-b8-01-size-32x32.ivf"
+              "av1-1-b8-01-size-32x32.ivf.md5"
+              "av1-1-b8-01-size-32x34.ivf"
+              "av1-1-b8-01-size-32x34.ivf.md5"
+              "av1-1-b8-01-size-32x64.ivf"
+              "av1-1-b8-01-size-32x64.ivf.md5"
+              "av1-1-b8-01-size-32x66.ivf"
+              "av1-1-b8-01-size-32x66.ivf.md5"
+              "av1-1-b8-01-size-34x16.ivf"
+              "av1-1-b8-01-size-34x16.ivf.md5"
+              "av1-1-b8-01-size-34x18.ivf"
+              "av1-1-b8-01-size-34x18.ivf.md5"
+              "av1-1-b8-01-size-34x32.ivf"
+              "av1-1-b8-01-size-34x32.ivf.md5"
+              "av1-1-b8-01-size-34x34.ivf"
+              "av1-1-b8-01-size-34x34.ivf.md5"
+              "av1-1-b8-01-size-34x64.ivf"
+              "av1-1-b8-01-size-34x64.ivf.md5"
+              "av1-1-b8-01-size-34x66.ivf"
+              "av1-1-b8-01-size-34x66.ivf.md5"
+              "av1-1-b8-01-size-64x16.ivf"
+              "av1-1-b8-01-size-64x16.ivf.md5"
+              "av1-1-b8-01-size-64x18.ivf"
+              "av1-1-b8-01-size-64x18.ivf.md5"
+              "av1-1-b8-01-size-64x32.ivf"
+              "av1-1-b8-01-size-64x32.ivf.md5"
+              "av1-1-b8-01-size-64x34.ivf"
+              "av1-1-b8-01-size-64x34.ivf.md5"
+              "av1-1-b8-01-size-64x64.ivf"
+              "av1-1-b8-01-size-64x64.ivf.md5"
+              "av1-1-b8-01-size-64x66.ivf"
+              "av1-1-b8-01-size-64x66.ivf.md5"
+              "av1-1-b8-01-size-66x16.ivf"
+              "av1-1-b8-01-size-66x16.ivf.md5"
+              "av1-1-b8-01-size-66x18.ivf"
+              "av1-1-b8-01-size-66x18.ivf.md5"
+              "av1-1-b8-01-size-66x32.ivf"
+              "av1-1-b8-01-size-66x32.ivf.md5"
+              "av1-1-b8-01-size-66x34.ivf"
+              "av1-1-b8-01-size-66x34.ivf.md5"
+              "av1-1-b8-01-size-66x64.ivf"
+              "av1-1-b8-01-size-66x64.ivf.md5"
+              "av1-1-b8-01-size-66x66.ivf"
+              "av1-1-b8-01-size-66x66.ivf.md5"
+              "av1-1-b8-02-allintra.ivf"
+              "av1-1-b8-02-allintra.ivf.md5"
+              "av1-1-b8-03-sizeup.mkv"
+              "av1-1-b8-03-sizeup.mkv.md5"
+              "av1-1-b8-03-sizedown.mkv"
+              "av1-1-b8-03-sizedown.mkv.md5")
+endif()
+
+if(ENABLE_ENCODE_PERF_TESTS AND CONFIG_AV1_ENCODER)
+  list(APPEND AOM_TEST_DATA_FILE_NAMES "desktop_640_360_30.yuv"
+              "kirland_640_480_30.yuv" "macmarcomoving_640_480_30.yuv"
+              "macmarcostationary_640_480_30.yuv" "niklas_1280_720_30.yuv"
+              "tacomanarrows_640_480_30.yuv"
+              "tacomasmallcameramovement_640_480_30.yuv"
+              "thaloundeskmtg_640_480_30.yuv")
+endif()
+
+# Parses test/test-data.sha1 and writes captured file names and checksums to
+# $out_files and $out_checksums as lists.
+function(make_test_data_lists test_data_file out_files out_checksums)
+  if(NOT test_data_file OR NOT EXISTS "${test_data_file}")
+    message(FATAL_ERROR "Test info file missing or empty (${test_data_file})")
+  endif()
+
+  # Read $test_data_file into $files_and_checksums. $files_and_checksums becomes
+  # a list with an entry for each line from $test_data_file.
+  file(STRINGS "${test_data_file}" files_and_checksums)
+
+  # Iterate over the list of lines and split it into $checksums and $filenames.
+  foreach(line ${files_and_checksums})
+    string(FIND "${line}" " *" delim_pos)
+
+    math(EXPR filename_pos "${delim_pos} + 2")
+    string(SUBSTRING "${line}" 0 ${delim_pos} checksum)
+    string(SUBSTRING "${line}" ${filename_pos} -1 filename)
+
+    list(FIND AOM_TEST_DATA_FILE_NAMES ${filename} list_index)
+    if(NOT ${list_index} EQUAL -1)
+
+      # Include the name and checksum in output only when the file is needed.
+      set(checksums ${checksums} ${checksum})
+      set(filenames ${filenames} ${filename})
+    endif()
+  endforeach()
+
+  list(LENGTH filenames num_files)
+  list(LENGTH checksums num_checksums)
+  if(NOT checksums OR NOT filenames OR NOT num_files EQUAL num_checksums)
+    message(FATAL_ERROR "Parsing of ${test_data_file} failed.")
+  endif()
+
+  set(${out_checksums} ${checksums} PARENT_SCOPE)
+  set(${out_files} ${filenames} PARENT_SCOPE)
+endfunction()
+
+# Appends each file name in $test_files to $test_dir and adds the result path to
+# $out_path_list.
+function(expand_test_file_paths test_files test_dir out_path_list)
+  foreach(filename ${${test_files}})
+    set(path_list ${path_list} "${test_dir}/${filename}")
+  endforeach()
+  set(${out_path_list} ${path_list} PARENT_SCOPE)
+endfunction()
+
+function(check_file local_path expected_checksum out_needs_update)
+  if(EXISTS "${local_path}")
+    file(SHA1 "${local_path}" file_checksum)
+  else()
+    set(${out_needs_update} 1 PARENT_SCOPE)
+    return()
+  endif()
+
+  if("${file_checksum}" STREQUAL "${expected_checksum}")
+    unset(${out_needs_update} PARENT_SCOPE)
+  else()
+    set(${out_needs_update} 1 PARENT_SCOPE)
+    return()
+  endif()
+  message("${local_path} up to date.")
+endfunction()
+
+# Downloads data from $file_url, confirms that $file_checksum matches, and
+# writes it to $local_path.
+function(download_test_file file_url file_checksum local_path)
+  message("Downloading ${file_url} ...")
+  file(DOWNLOAD "${file_url}" "${local_path}" SHOW_PROGRESS
+       EXPECTED_HASH SHA1=${file_checksum})
+  message("Download of ${file_url} complete.")
+endfunction()
diff --git a/third_party/aom/test/test_intra_pred_speed.cc b/third_party/aom/test/test_intra_pred_speed.cc
new file mode 100644
index 0000000000..b72ac11674
--- /dev/null
+++ b/third_party/aom/test/test_intra_pred_speed.cc
@@ -0,0 +1,1464 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+//  Test and time AOM intra-predictor functions
+
+#include <stdio.h>
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/md5_helper.h"
+#include "aom/aom_integer.h"
+#include "aom_ports/mem.h"
+#include "aom_ports/aom_timer.h"
+#include "av1/common/common_data.h"
+
+// -----------------------------------------------------------------------------
+
+namespace {
+
+// Note:
+// APPLY_UNIT_TESTS
+// 1: Do unit tests
+// 0: Generate MD5 array as required
+#define APPLY_UNIT_TESTS 1
+
+typedef void (*AvxPredFunc)(uint8_t *dst, ptrdiff_t y_stride,
+                            const uint8_t *above, const uint8_t *left);
+
+const int kBPS = 64;
+const int kTotalPixels = kBPS * kBPS;
+// 4 DC variants, V, H, PAETH, SMOOTH, SMOOTH_V, SMOOTH_H
+const int kNumAv1IntraFuncs = 10;
+
+#if APPLY_UNIT_TESTS
+const char *kAv1IntraPredNames[kNumAv1IntraFuncs] = {
+  "DC_PRED", "DC_LEFT_PRED", "DC_TOP_PRED", "DC_128_PRED",   "V_PRED",
+  "H_PRED",  "PAETH_PRED",   "SMOOTH_PRED", "SMOOTH_V_PRED", "SMOOTH_H_PRED",
+};
+#endif  // APPLY_UNIT_TESTS
+
+template <typename Pixel>
+struct IntraPredTestMem {
+  void Init(int block_width, int block_height, int bd) {
+    ASSERT_LE(block_width, kBPS);
+    ASSERT_LE(block_height, kBPS);
+    // Note: for blocks having width <= 32 and height <= 32, we generate 32x32
+    // random pixels as before to avoid having to recalculate all hashes again.
+    const int block_size_upto_32 = (block_width <= 32) && (block_height <= 32);
+    stride = block_size_upto_32 ? 32 : kBPS;
+    num_pixels = stride * stride;
+    libaom_test::ACMRandom rnd(libaom_test::ACMRandom::DeterministicSeed());
+    above = above_mem + 16;
+    const int mask = (1 << bd) - 1;
+    for (int i = 0; i < num_pixels; ++i) ref_src[i] = rnd.Rand16() & mask;
+    for (int i = 0; i < stride; ++i) left[i] = rnd.Rand16() & mask;
+    for (int i = -1; i < stride; ++i) above[i] = rnd.Rand16() & mask;
+
+    for (int i = stride; i < 2 * stride; ++i) {
+      left[i] = rnd.Rand16() & mask;
+      above[i] = rnd.Rand16() & mask;
+    }
+  }
+
+  DECLARE_ALIGNED(16, Pixel, src[kTotalPixels]);
+  DECLARE_ALIGNED(16, Pixel, ref_src[kTotalPixels]);
+  DECLARE_ALIGNED(16, Pixel, left[2 * kBPS]);
+  Pixel *above;
+  int stride;
+  int num_pixels;
+
+ private:
+  DECLARE_ALIGNED(16, Pixel, above_mem[2 * kBPS + 16]);
+};
+
+// -----------------------------------------------------------------------------
+// Low Bittdepth
+
+typedef IntraPredTestMem<uint8_t> Av1IntraPredTestMem;
+
+static const char *const kTxSizeStrings[TX_SIZES_ALL] = {
+  "4X4",  "8X8",  "16X16", "32X32", "64X64", "4X8",   "8X4",
+  "8X16", "16X8", "16X32", "32X16", "32X64", "64X32", "4X16",
+  "16X4", "8X32", "32X8",  "16X64", "64X16",
+};
+
+void CheckMd5Signature(TX_SIZE tx_size, bool is_hbd,
+                       const char *const signatures[], const void *data,
+                       size_t data_size, int elapsed_time, int idx) {
+  const std::string hbd_str = is_hbd ? "Hbd " : "";
+  const std::string name_str = hbd_str + "Intra" + kTxSizeStrings[tx_size];
+  libaom_test::MD5 md5;
+  md5.Add(reinterpret_cast<const uint8_t *>(data), data_size);
+#if APPLY_UNIT_TESTS
+  printf("Mode %s[%13s]: %5d ms     MD5: %s\n", name_str.c_str(),
+         kAv1IntraPredNames[idx], elapsed_time, md5.Get());
+  EXPECT_STREQ(signatures[idx], md5.Get());
+#else
+  (void)signatures;
+  (void)elapsed_time;
+  (void)idx;
+  printf("\"%s\",\n", md5.Get());
+#endif
+}
+
+void TestIntraPred(TX_SIZE tx_size, AvxPredFunc const *pred_funcs,
+                   const char *const signatures[]) {
+  const int block_width = tx_size_wide[tx_size];
+  const int block_height = tx_size_high[tx_size];
+  const int num_pixels_per_test =
+      block_width * block_height * kNumAv1IntraFuncs;
+  const int kNumTests = static_cast<int>(2.e10 / num_pixels_per_test);
+  Av1IntraPredTestMem intra_pred_test_mem;
+  intra_pred_test_mem.Init(block_width, block_height, 8);
+
+  for (int k = 0; k < kNumAv1IntraFuncs; ++k) {
+    if (pred_funcs[k] == NULL) continue;
+    memcpy(intra_pred_test_mem.src, intra_pred_test_mem.ref_src,
+           sizeof(intra_pred_test_mem.src));
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int num_tests = 0; num_tests < kNumTests; ++num_tests) {
+      pred_funcs[k](intra_pred_test_mem.src, intra_pred_test_mem.stride,
+                    intra_pred_test_mem.above, intra_pred_test_mem.left);
+    }
+    libaom_test::ClearSystemState();
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time =
+        static_cast<int>(aom_usec_timer_elapsed(&timer) / 1000);
+    CheckMd5Signature(
+        tx_size, false, signatures, intra_pred_test_mem.src,
+        intra_pred_test_mem.num_pixels * sizeof(*intra_pred_test_mem.src),
+        elapsed_time, k);
+  }
+}
+
+static const char *const kSignatures[TX_SIZES_ALL][kNumAv1IntraFuncs] = {
+  {
+      // 4X4
+      "e7ed7353c3383fff942e500e9bfe82fe",
+      "2a4a26fcc6ce005eadc08354d196c8a9",
+      "269d92eff86f315d9c38fe7640d85b15",
+      "ae2960eea9f71ee3dabe08b282ec1773",
+      "6c1abcc44e90148998b51acd11144e9c",
+      "f7bb3186e1ef8a2b326037ff898cad8e",
+      "59fc0e923a08cfac0a493fb38988e2bb",
+      "9ff8bb37d9c830e6ab8ecb0c435d3c91",
+      "de6937fca02354f2874dbc5dbec5d5b3",
+      "723cf948137f7d8c7860d814e55ae67d",
+  },
+  {
+      // 8X8
+      "d8bbae5d6547cfc17e4f5f44c8730e88",
+      "373bab6d931868d41a601d9d88ce9ac3",
+      "6fdd5ff4ff79656c14747598ca9e3706",
+      "d9661c2811d6a73674f40ffb2b841847",
+      "7c722d10b19ccff0b8c171868e747385",
+      "f81dd986eb2b50f750d3a7da716b7e27",
+      "064404361748dd111a890a1470d7f0ea",
+      "dc29b7e1f78cc8e7525d5ea4c0ab9b78",
+      "97111eb1bc26bade6272015df829f1ae",
+      "d19a8a73cc46b807f2c5e817576cc1e1",
+  },
+  {
+      // 16X16
+      "50971c07ce26977d30298538fffec619",
+      "527a6b9e0dc5b21b98cf276305432bef",
+      "7eff2868f80ebc2c43a4f367281d80f7",
+      "67cd60512b54964ef6aff1bd4816d922",
+      "48371c87dc95c08a33b2048f89cf6468",
+      "b0acf2872ee411d7530af6d2625a7084",
+      "93d6b5352b571805ab16a55e1bbed86a",
+      "03764e4c0aebbc180e4e2c68fb06df2b",
+      "bb6c74c9076c9f266ab11fb57060d8e6",
+      "0c5162bc28489756ddb847b5678e6f07",
+  },
+  {
+      // 32X32
+      "a0a618c900e65ae521ccc8af789729f2",
+      "985aaa7c72b4a6c2fb431d32100cf13a",
+      "10662d09febc3ca13ee4e700120daeb5",
+      "b3b01379ba08916ef6b1b35f7d9ad51c",
+      "9f4261755795af97e34679c333ec7004",
+      "bc2c9da91ad97ef0d1610fb0a9041657",
+      "ef1653982b69e1f64bee3759f3e1ec45",
+      "1a51a675deba2c83282142eb48d3dc3d",
+      "866c224746dc260cda861a7b1b383fb3",
+      "cea23799fc3526e1b6a6ff02b42b82af",
+  },
+  {
+      // 64X64
+      "6e1094fa7b50bc813aa2ba29f5df8755",
+      "afe020786b83b793c2bbd9468097ff6e",
+      "be91585259bc37bf4dc1651936e90b3e",
+      "a1650dbcd56e10288c3e269eca37967d",
+      "9e5c34f3797e0cdd3cd9d4c05b0d8950",
+      "bc87be7ac899cc6a28f399d7516c49fe",
+      "9811fd0d2dd515f06122f5d1bd18b784",
+      "3c140e466f2c2c0d9cb7d2157ab8dc27",
+      "9543de76c925a8f6adc884cc7f98dc91",
+      "df1df0376cc944afe7e74e94f53e575a",
+  },
+  {
+      // 4X8
+      "d9fbebdc85f71ab1e18461b2db4a2adc",
+      "5ccb2a68284bc9714d94b8a06ccadbb2",
+      "735d059abc2744f3ff3f9590f7191b37",
+      "d9fbebdc85f71ab1e18461b2db4a2adc",
+      "6819497c44cd0ace120add83672996ee",
+      "7e3244f5a2d3edf81c7e962a842b97f9",
+      "809350f164cd4d1650850bb0f59c3260",
+      "1b60a394331eeab6927a6f8aaff57040",
+      "5307de1bd7329ba6b281d2c1b0b457f9",
+      "24c58a8138339846d95568efb91751db",
+  },
+  {
+      // 8X4
+      "23f9fc11344426c9bee2e06d57dfd628",
+      "2d71a26d1bae1fb34734de7b42fc5eb7",
+      "5af9c1b2fd9d5721fad67b67b3f7c816",
+      "00d71b17be662753813d515f197d145e",
+      "bef10ec984427e28f4390f43809d10af",
+      "77773cdfb7ed6bc882ab202a64b0a470",
+      "2cc48bd66d6b0121b5221d52ccd732af",
+      "b302155e1c9eeeafe2ba2bf68e807a46",
+      "561bc8d0e76d5041ebd5168fc6a115e1",
+      "81d0113fb1d0a9a24ffd6f1987b77948",
+  },
+  {
+      // 8X16
+      "c849de88b24f773dfcdd1d48d1209796",
+      "6cb807c1897b94866a0f3d3c56ed8695",
+      "d56db05a8ac7981762f5b877f486c4ef",
+      "b4bc01eb6e59a40922ad17715cafb04b",
+      "09d178439534f4062ae687c351f66d64",
+      "644501399cf73080ac606e5cef7ca09b",
+      "278076495180e17c065a95ab7278539a",
+      "9dd7f324816f242be408ffeb0c673732",
+      "f520c4a20acfa0bea1d253c6f0f040fd",
+      "85f38df809df2c2d7c8b4a157a65cd44",
+  },
+  {
+      // 16X8
+      "b4cbdbdf10ce13300b4063a3daf99e04",
+      "3731e1e6202064a9d0604d7c293ecee4",
+      "6c856188c4256a06452f0d5d70cac436",
+      "1f2192b4c8c497589484ea7bf9c944e8",
+      "84011bd4b7f565119d06787840e333a0",
+      "0e48949f7a6aa36f0d76b5d01f91124a",
+      "60eff8064634b6c73b10681356baeee9",
+      "1559aeb081a9c0c71111d6093c2ff9fd",
+      "c15479b739713773e5cabb748451987b",
+      "72e33ec12c9b67aea26d8d005fb82de2",
+  },
+  {
+      // 16X32
+      "abe5233d189cdbf79424721571bbaa7b",
+      "282759f81e3cfb2e2d396fe406b72a8b",
+      "e2224926c264f6f174cbc3167a233168",
+      "6814e85c2b33f8c9415d62e80394b47b",
+      "99cbbb60459c08a3061d72c4e4f6276a",
+      "1d1567d40b8e816f8c1f71e576fe0f87",
+      "36fdd371b624a075814d497c4832ec85",
+      "8ab8da61b727442b6ff692b40d0df018",
+      "e35a10ad7fdf2327e821504a90f6a6eb",
+      "1f7211e727dc1de7d6a55d082fbdd821",
+  },
+  {
+      // 32X16
+      "d1aeb8d5fdcfd3307922af01a798a4dc",
+      "b0bcb514ebfbee065faea9d34c12ae75",
+      "d6a18c63b4e909871c0137ca652fad23",
+      "fd047f2fc1b8ffb95d0eeef3e8796a45",
+      "645ab60779ea348fd93c81561c31bab9",
+      "4409633c9db8dff41ade4292a3a56e7f",
+      "5e36a11e069b31c2a739f3a9c7b37c24",
+      "e83b9483d702cfae496991c3c7fa92c0",
+      "12f6ddf98c7f30a277307f1ea935b030",
+      "354321d6c32bbdb0739e4fa2acbf41e1",
+  },
+  {
+      // 32X64
+      "0ce332b343934b34cd4417725faa85cb",
+      "4e2a2cfd8f56f15939bdfc753145b303",
+      "0f46d124ba9f48cdd5d5290acf786d6d",
+      "e1e8ed803236367821981500a3d9eebe",
+      "1d2f8e48e3adb7c448be05d9f66f4954",
+      "9fb2e176636a5689b26f73ca73fcc512",
+      "e720ebccae7e25e36f23da53ae5b5d6a",
+      "86fe4364734169aaa4520d799890d530",
+      "b1870290764bb1b100d1974e2bd70f1d",
+      "ce5b238e19d85ef69d85badfab4e63ae",
+  },
+  {
+      // 64X32
+      "a6c5aeb722615089efbca80b02951ceb",
+      "538424b24bd0830f21788e7238ca762f",
+      "80c15b303235f9bc2259027bb92dfdc4",
+      "e48e1ac15e97191a8fda08d62fff343e",
+      "12604b37875533665078405ef4582e35",
+      "0048afa17bd3e1632d68b96048836530",
+      "07a0cfcb56a5eed50c4bd6c26814336b",
+      "529d8a070de5bc6531fa3ee8f450c233",
+      "33c50a11c7d78f72434064f634305e95",
+      "e0ef7f0559c1a50ec5a8c12011b962f7",
+  },
+  {
+      // 4X16
+      "750491056568eb8fe15387b86bdf06b8",
+      "3a52dae9f599f08cfb3bd1b910dc0e11",
+      "af79f71e3e03dbeca44e2e13561f70c7",
+      "ca7dfd7624afc0c06fb5552f44398535",
+      "b591af115444bf43140c29c269f68fb2",
+      "483d942ae36e69e62f31eb215331416f",
+      "f14b58525e81870bc5d95c7ac71a347f",
+      "371208bb4027d9badb04095d1590bbc4",
+      "c7049c21b2924d70c7c12784d6b6b796",
+      "7d87233f4b5b0f12086045e5d7b2d4c2",
+  },
+  {
+      // 16X4
+      "7c6e325a65e77e732b3adbe237e045e4",
+      "24478f93ffcec47852e004d0fe948464",
+      "258d042c67d4ba3ecfa667f0adc9aebf",
+      "b2cd21d06959f159a1f3c4d9768ee7fb",
+      "b4e1f38157bf8410e7c3da02f687a343",
+      "869e703729eb0fc0711c254944ff5d5a",
+      "9638dd77105a640b146a8201ea7a0801",
+      "919d932c6af8a1cc7486e8ce996dd487",
+      "e1c9be493b6714c7ae48f30044c43140",
+      "bf0fe3889d654b2f6eb98c8fc751f9e4",
+  },
+  {
+      // 8X32
+      "8dfac4319fe0bd40013ffb3102da8c72",
+      "feb46b6dc4e2ca0a09533bfc51d4dcb0",
+      "850837ec714c37262216527aaf4cbbe9",
+      "4603c7800fb08361f163daca876e8bda",
+      "1ff95e7d2debc27b05806fb25abfd624",
+      "d81b9a51a062b23ca7823804cb7bec22",
+      "f1d8978158766f46335203608cb807e7",
+      "f3527096256258c0878d644a9d7d53ca",
+      "cbde98ac8b009953eb112807ad2ea29e",
+      "654fb1153415747feae599f538122af5",
+  },
+  {
+      // 32X8
+      "3d4ee16fab374357474f60b845327bc7",
+      "bc17c5059473a476df4e85f56395ad55",
+      "3d4ee16fab374357474f60b845327bc7",
+      "c14b8db34dc2355b84e3735c9ba16c7f",
+      "a71d25b5d47a92a8b9223c98f18458ee",
+      "6c1cfe2b1893f4576a80675687cb6426",
+      "92d11bbef8b85bb48d799bb055de3514",
+      "bcf81d1db8ae5cc03360467f44f498ec",
+      "79f8c564163555592e808e145eaf5c60",
+      "46fff139cef2ef773938bcc8b0e5abb8",
+  },
+  {
+      // 16X64
+      "3b2a053ee8b05a8ac35ad23b0422a151",
+      "12b0c69595328c465e0b25e0c9e3e9fc",
+      "f77c544ac8035e01920deae40cee7b07",
+      "727797ef15ccd8d325476fe8f12006a3",
+      "f3be77c0fe67eb5d9d515e92bec21eb7",
+      "f1ece6409e01e9dd98b800d49628247d",
+      "efd2ec9bfbbd4fd1f6604ea369df1894",
+      "ec703de918422b9e03197ba0ed60a199",
+      "739418efb89c07f700895deaa5d0b3e3",
+      "9943ae1bbeeebfe1d3a92dc39e049d63",
+  },
+  {
+      // 64X16
+      "821b76b1494d4f84d20817840f719a1a",
+      "69e462c3338a9aaf993c3f7cfbc15649",
+      "516d8f6eb054d74d150e7b444185b6b9",
+      "de1b736e9d99129609d6ef3a491507a0",
+      "fd9b4276e7affe1e0e4ce4f428058994",
+      "cd82fd361a4767ac29a9f406b480b8f3",
+      "2792c2f810157a4a6cb13c28529ff779",
+      "1220442d90c4255ba0969d28b91e93a6",
+      "c7253e10b45f7f67dfee3256c9b94825",
+      "879792198071c7e0b50b9b5010d8c18f",
+  },
+};
+
+}  // namespace
+
+// Defines a test case for |arch| (e.g., C, SSE2, ...) passing the predictors
+// to TestIntraPred. The test name is 'arch.TestIntraPred_tx_size', e.g.,
+// C.TestIntraPred.0
+#define INTRA_PRED_TEST(arch, tx_size, dc, dc_left, dc_top, dc_128, v, h,  \
+                        paeth, smooth, smooth_v, smooth_h)                 \
+  TEST(arch, DISABLED_##TestIntraPred_##tx_size) {                         \
+    static const AvxPredFunc aom_intra_pred[] = {                          \
+      dc, dc_left, dc_top, dc_128, v, h, paeth, smooth, smooth_v, smooth_h \
+    };                                                                     \
+    TestIntraPred(tx_size, aom_intra_pred, kSignatures[tx_size]);          \
+  }
+
+// -----------------------------------------------------------------------------
+// 4x4, 4x8, 4x16
+
+INTRA_PRED_TEST(C_1, TX_4X4, aom_dc_predictor_4x4_c,
+                aom_dc_left_predictor_4x4_c, aom_dc_top_predictor_4x4_c,
+                aom_dc_128_predictor_4x4_c, aom_v_predictor_4x4_c,
+                aom_h_predictor_4x4_c, aom_paeth_predictor_4x4_c,
+                aom_smooth_predictor_4x4_c, aom_smooth_v_predictor_4x4_c,
+                aom_smooth_h_predictor_4x4_c)
+
+INTRA_PRED_TEST(C_2, TX_4X8, aom_dc_predictor_4x8_c,
+                aom_dc_left_predictor_4x8_c, aom_dc_top_predictor_4x8_c,
+                aom_dc_128_predictor_4x8_c, aom_v_predictor_4x8_c,
+                aom_h_predictor_4x8_c, aom_paeth_predictor_4x8_c,
+                aom_smooth_predictor_4x8_c, aom_smooth_v_predictor_4x8_c,
+                aom_smooth_h_predictor_4x8_c)
+
+INTRA_PRED_TEST(C_3, TX_4X16, aom_dc_predictor_4x16_c,
+                aom_dc_left_predictor_4x16_c, aom_dc_top_predictor_4x16_c,
+                aom_dc_128_predictor_4x16_c, aom_v_predictor_4x16_c,
+                aom_h_predictor_4x16_c, aom_paeth_predictor_4x16_c,
+                aom_smooth_predictor_4x16_c, aom_smooth_v_predictor_4x16_c,
+                aom_smooth_h_predictor_4x16_c)
+
+#if HAVE_SSE2
+INTRA_PRED_TEST(SSE2_1, TX_4X4, aom_dc_predictor_4x4_sse2,
+                aom_dc_left_predictor_4x4_sse2, aom_dc_top_predictor_4x4_sse2,
+                aom_dc_128_predictor_4x4_sse2, aom_v_predictor_4x4_sse2,
+                aom_h_predictor_4x4_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_2, TX_4X8, aom_dc_predictor_4x8_sse2,
+                aom_dc_left_predictor_4x8_sse2, aom_dc_top_predictor_4x8_sse2,
+                aom_dc_128_predictor_4x8_sse2, aom_v_predictor_4x8_sse2,
+                aom_h_predictor_4x8_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_3, TX_4X16, aom_dc_predictor_4x16_sse2,
+                aom_dc_left_predictor_4x16_sse2, aom_dc_top_predictor_4x16_sse2,
+                aom_dc_128_predictor_4x16_sse2, aom_v_predictor_4x16_sse2,
+                aom_h_predictor_4x16_sse2, NULL, NULL, NULL, NULL)
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3
+INTRA_PRED_TEST(SSSE3_1, TX_4X4, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_4x4_ssse3, aom_smooth_predictor_4x4_ssse3,
+                aom_smooth_v_predictor_4x4_ssse3,
+                aom_smooth_h_predictor_4x4_ssse3)
+INTRA_PRED_TEST(SSSE3_2, TX_4X8, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_4x8_ssse3, aom_smooth_predictor_4x8_ssse3,
+                aom_smooth_v_predictor_4x8_ssse3,
+                aom_smooth_h_predictor_4x8_ssse3)
+INTRA_PRED_TEST(SSSE3_3, TX_4X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_4x16_ssse3, aom_smooth_predictor_4x16_ssse3,
+                aom_smooth_v_predictor_4x16_ssse3,
+                aom_smooth_h_predictor_4x16_ssse3)
+#endif  // HAVE_SSSE3
+
+#if HAVE_DSPR2
+INTRA_PRED_TEST(DSPR2, TX_4X4, aom_dc_predictor_4x4_dspr2, NULL, NULL, NULL,
+                NULL, aom_h_predictor_4x4_dspr2, NULL, NULL, NULL, NULL)
+#endif  // HAVE_DSPR2
+
+#if HAVE_NEON
+INTRA_PRED_TEST(NEON, TX_4X4, aom_dc_predictor_4x4_neon,
+                aom_dc_left_predictor_4x4_neon, aom_dc_top_predictor_4x4_neon,
+                aom_dc_128_predictor_4x4_neon, aom_v_predictor_4x4_neon,
+                aom_h_predictor_4x4_neon, NULL, NULL, NULL, NULL)
+#endif  // HAVE_NEON
+
+#if HAVE_MSA
+INTRA_PRED_TEST(MSA, TX_4X4, aom_dc_predictor_4x4_msa,
+                aom_dc_left_predictor_4x4_msa, aom_dc_top_predictor_4x4_msa,
+                aom_dc_128_predictor_4x4_msa, aom_v_predictor_4x4_msa,
+                aom_h_predictor_4x4_msa, NULL, NULL, NULL, NULL)
+#endif  // HAVE_MSA
+
+// -----------------------------------------------------------------------------
+// 8x8, 8x4, 8x16, 8x32
+
+INTRA_PRED_TEST(C_1, TX_8X8, aom_dc_predictor_8x8_c,
+                aom_dc_left_predictor_8x8_c, aom_dc_top_predictor_8x8_c,
+                aom_dc_128_predictor_8x8_c, aom_v_predictor_8x8_c,
+                aom_h_predictor_8x8_c, aom_paeth_predictor_8x8_c,
+                aom_smooth_predictor_8x8_c, aom_smooth_v_predictor_8x8_c,
+                aom_smooth_h_predictor_8x8_c)
+
+INTRA_PRED_TEST(C_2, TX_8X4, aom_dc_predictor_8x4_c,
+                aom_dc_left_predictor_8x4_c, aom_dc_top_predictor_8x4_c,
+                aom_dc_128_predictor_8x4_c, aom_v_predictor_8x4_c,
+                aom_h_predictor_8x4_c, aom_paeth_predictor_8x4_c,
+                aom_smooth_predictor_8x4_c, aom_smooth_v_predictor_8x4_c,
+                aom_smooth_h_predictor_8x4_c)
+
+INTRA_PRED_TEST(C_3, TX_8X16, aom_dc_predictor_8x16_c,
+                aom_dc_left_predictor_8x16_c, aom_dc_top_predictor_8x16_c,
+                aom_dc_128_predictor_8x16_c, aom_v_predictor_8x16_c,
+                aom_h_predictor_8x16_c, aom_paeth_predictor_8x16_c,
+                aom_smooth_predictor_8x16_c, aom_smooth_v_predictor_8x16_c,
+                aom_smooth_h_predictor_8x16_c)
+
+INTRA_PRED_TEST(C_4, TX_8X32, aom_dc_predictor_8x32_c,
+                aom_dc_left_predictor_8x32_c, aom_dc_top_predictor_8x32_c,
+                aom_dc_128_predictor_8x32_c, aom_v_predictor_8x32_c,
+                aom_h_predictor_8x32_c, aom_paeth_predictor_8x32_c,
+                aom_smooth_predictor_8x32_c, aom_smooth_v_predictor_8x32_c,
+                aom_smooth_h_predictor_8x32_c)
+
+#if HAVE_SSE2
+INTRA_PRED_TEST(SSE2_1, TX_8X8, aom_dc_predictor_8x8_sse2,
+                aom_dc_left_predictor_8x8_sse2, aom_dc_top_predictor_8x8_sse2,
+                aom_dc_128_predictor_8x8_sse2, aom_v_predictor_8x8_sse2,
+                aom_h_predictor_8x8_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_2, TX_8X4, aom_dc_predictor_8x4_sse2,
+                aom_dc_left_predictor_8x4_sse2, aom_dc_top_predictor_8x4_sse2,
+                aom_dc_128_predictor_8x4_sse2, aom_v_predictor_8x4_sse2,
+                aom_h_predictor_8x4_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_3, TX_8X16, aom_dc_predictor_8x16_sse2,
+                aom_dc_left_predictor_8x16_sse2, aom_dc_top_predictor_8x16_sse2,
+                aom_dc_128_predictor_8x16_sse2, aom_v_predictor_8x16_sse2,
+                aom_h_predictor_8x16_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_4, TX_8X32, aom_dc_predictor_8x32_sse2,
+                aom_dc_left_predictor_8x32_sse2, aom_dc_top_predictor_8x32_sse2,
+                aom_dc_128_predictor_8x32_sse2, aom_v_predictor_8x32_sse2,
+                aom_h_predictor_8x32_sse2, NULL, NULL, NULL, NULL)
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3
+INTRA_PRED_TEST(SSSE3_1, TX_8X8, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_8x8_ssse3, aom_smooth_predictor_8x8_ssse3,
+                aom_smooth_v_predictor_8x8_ssse3,
+                aom_smooth_h_predictor_8x8_ssse3)
+INTRA_PRED_TEST(SSSE3_2, TX_8X4, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_8x4_ssse3, aom_smooth_predictor_8x4_ssse3,
+                aom_smooth_v_predictor_8x4_ssse3,
+                aom_smooth_h_predictor_8x4_ssse3)
+INTRA_PRED_TEST(SSSE3_3, TX_8X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_8x16_ssse3, aom_smooth_predictor_8x16_ssse3,
+                aom_smooth_v_predictor_8x16_ssse3,
+                aom_smooth_h_predictor_8x16_ssse3)
+INTRA_PRED_TEST(SSSE3_4, TX_8X32, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_8x32_ssse3, aom_smooth_predictor_8x32_ssse3,
+                aom_smooth_v_predictor_8x32_ssse3,
+                aom_smooth_h_predictor_8x32_ssse3)
+#endif  // HAVE_SSSE3
+
+#if HAVE_DSPR2
+INTRA_PRED_TEST(DSPR2, TX_8X8, aom_dc_predictor_8x8_dspr2, NULL, NULL, NULL,
+                NULL, aom_h_predictor_8x8_dspr2, NULL, NULL, NULL, NULL)
+#endif  // HAVE_DSPR2
+
+#if HAVE_NEON
+INTRA_PRED_TEST(NEON, TX_8X8, aom_dc_predictor_8x8_neon,
+                aom_dc_left_predictor_8x8_neon, aom_dc_top_predictor_8x8_neon,
+                aom_dc_128_predictor_8x8_neon, aom_v_predictor_8x8_neon,
+                aom_h_predictor_8x8_neon, NULL, NULL, NULL, NULL)
+#endif  // HAVE_NEON
+
+#if HAVE_MSA
+INTRA_PRED_TEST(MSA, TX_8X8, aom_dc_predictor_8x8_msa,
+                aom_dc_left_predictor_8x8_msa, aom_dc_top_predictor_8x8_msa,
+                aom_dc_128_predictor_8x8_msa, aom_v_predictor_8x8_msa,
+                aom_h_predictor_8x8_msa, NULL, NULL, NULL, NULL)
+#endif  // HAVE_MSA
+
+// -----------------------------------------------------------------------------
+// 16x16, 16x8, 16x32, 16x4, 16x64
+
+INTRA_PRED_TEST(C_1, TX_16X16, aom_dc_predictor_16x16_c,
+                aom_dc_left_predictor_16x16_c, aom_dc_top_predictor_16x16_c,
+                aom_dc_128_predictor_16x16_c, aom_v_predictor_16x16_c,
+                aom_h_predictor_16x16_c, aom_paeth_predictor_16x16_c,
+                aom_smooth_predictor_16x16_c, aom_smooth_v_predictor_16x16_c,
+                aom_smooth_h_predictor_16x16_c)
+
+INTRA_PRED_TEST(C_2, TX_16X8, aom_dc_predictor_16x8_c,
+                aom_dc_left_predictor_16x8_c, aom_dc_top_predictor_16x8_c,
+                aom_dc_128_predictor_16x8_c, aom_v_predictor_16x8_c,
+                aom_h_predictor_16x8_c, aom_paeth_predictor_16x8_c,
+                aom_smooth_predictor_16x8_c, aom_smooth_v_predictor_16x8_c,
+                aom_smooth_h_predictor_16x8_c)
+
+INTRA_PRED_TEST(C_3, TX_16X32, aom_dc_predictor_16x32_c,
+                aom_dc_left_predictor_16x32_c, aom_dc_top_predictor_16x32_c,
+                aom_dc_128_predictor_16x32_c, aom_v_predictor_16x32_c,
+                aom_h_predictor_16x32_c, aom_paeth_predictor_16x32_c,
+                aom_smooth_predictor_16x32_c, aom_smooth_v_predictor_16x32_c,
+                aom_smooth_h_predictor_16x32_c)
+
+INTRA_PRED_TEST(C_4, TX_16X4, aom_dc_predictor_16x4_c,
+                aom_dc_left_predictor_16x4_c, aom_dc_top_predictor_16x4_c,
+                aom_dc_128_predictor_16x4_c, aom_v_predictor_16x4_c,
+                aom_h_predictor_16x4_c, aom_paeth_predictor_16x4_c,
+                aom_smooth_predictor_16x4_c, aom_smooth_v_predictor_16x4_c,
+                aom_smooth_h_predictor_16x4_c)
+
+INTRA_PRED_TEST(C_5, TX_16X64, aom_dc_predictor_16x64_c,
+                aom_dc_left_predictor_16x64_c, aom_dc_top_predictor_16x64_c,
+                aom_dc_128_predictor_16x64_c, aom_v_predictor_16x64_c,
+                aom_h_predictor_16x64_c, aom_paeth_predictor_16x64_c,
+                aom_smooth_predictor_16x64_c, aom_smooth_v_predictor_16x64_c,
+                aom_smooth_h_predictor_16x64_c)
+
+#if HAVE_SSE2
+INTRA_PRED_TEST(SSE2_1, TX_16X16, aom_dc_predictor_16x16_sse2,
+                aom_dc_left_predictor_16x16_sse2,
+                aom_dc_top_predictor_16x16_sse2,
+                aom_dc_128_predictor_16x16_sse2, aom_v_predictor_16x16_sse2,
+                aom_h_predictor_16x16_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_2, TX_16X8, aom_dc_predictor_16x8_sse2,
+                aom_dc_left_predictor_16x8_sse2, aom_dc_top_predictor_16x8_sse2,
+                aom_dc_128_predictor_16x8_sse2, aom_v_predictor_16x8_sse2,
+                aom_h_predictor_16x8_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_3, TX_16X32, aom_dc_predictor_16x32_sse2,
+                aom_dc_left_predictor_16x32_sse2,
+                aom_dc_top_predictor_16x32_sse2,
+                aom_dc_128_predictor_16x32_sse2, aom_v_predictor_16x32_sse2,
+                aom_h_predictor_16x32_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_4, TX_16X64, aom_dc_predictor_16x64_sse2,
+                aom_dc_left_predictor_16x64_sse2,
+                aom_dc_top_predictor_16x64_sse2,
+                aom_dc_128_predictor_16x64_sse2, aom_v_predictor_16x64_sse2,
+                aom_h_predictor_16x64_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_5, TX_16X4, aom_dc_predictor_16x4_sse2,
+                aom_dc_left_predictor_16x4_sse2, aom_dc_top_predictor_16x4_sse2,
+                aom_dc_128_predictor_16x4_sse2, aom_v_predictor_16x4_sse2,
+                aom_h_predictor_16x4_sse2, NULL, NULL, NULL, NULL)
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3
+INTRA_PRED_TEST(SSSE3_1, TX_16X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x16_ssse3,
+                aom_smooth_predictor_16x16_ssse3,
+                aom_smooth_v_predictor_16x16_ssse3,
+                aom_smooth_h_predictor_16x16_ssse3)
+INTRA_PRED_TEST(SSSE3_2, TX_16X8, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x8_ssse3, aom_smooth_predictor_16x8_ssse3,
+                aom_smooth_v_predictor_16x8_ssse3,
+                aom_smooth_h_predictor_16x8_ssse3)
+INTRA_PRED_TEST(SSSE3_3, TX_16X32, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x32_ssse3,
+                aom_smooth_predictor_16x32_ssse3,
+                aom_smooth_v_predictor_16x32_ssse3,
+                aom_smooth_h_predictor_16x32_ssse3)
+INTRA_PRED_TEST(SSSE3_4, TX_16X64, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x64_ssse3,
+                aom_smooth_predictor_16x64_ssse3,
+                aom_smooth_v_predictor_16x64_ssse3,
+                aom_smooth_h_predictor_16x64_ssse3)
+INTRA_PRED_TEST(SSSE3_5, TX_16X4, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x4_ssse3, aom_smooth_predictor_16x4_ssse3,
+                aom_smooth_v_predictor_16x4_ssse3,
+                aom_smooth_h_predictor_16x4_ssse3)
+#endif  // HAVE_SSSE3
+
+#if HAVE_AVX2
+INTRA_PRED_TEST(AVX2_1, TX_16X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x16_avx2, NULL, NULL, NULL)
+INTRA_PRED_TEST(AVX2_2, TX_16X8, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x8_avx2, NULL, NULL, NULL)
+INTRA_PRED_TEST(AVX2_3, TX_16X32, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x32_avx2, NULL, NULL, NULL)
+INTRA_PRED_TEST(AVX2_4, TX_16X64, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_16x64_avx2, NULL, NULL, NULL)
+#endif  // HAVE_AVX2
+
+#if HAVE_DSPR2
+INTRA_PRED_TEST(DSPR2, TX_16X16, aom_dc_predictor_16x16_dspr2, NULL, NULL, NULL,
+                NULL, aom_h_predictor_16x16_dspr2, NULL, NULL, NULL, NULL)
+#endif  // HAVE_DSPR2
+
+#if HAVE_NEON
+INTRA_PRED_TEST(NEON, TX_16X16, aom_dc_predictor_16x16_neon,
+                aom_dc_left_predictor_16x16_neon,
+                aom_dc_top_predictor_16x16_neon,
+                aom_dc_128_predictor_16x16_neon, aom_v_predictor_16x16_neon,
+                aom_h_predictor_16x16_neon, NULL, NULL, NULL, NULL)
+#endif  // HAVE_NEON
+
+#if HAVE_MSA
+INTRA_PRED_TEST(MSA, TX_16X16, aom_dc_predictor_16x16_msa,
+                aom_dc_left_predictor_16x16_msa, aom_dc_top_predictor_16x16_msa,
+                aom_dc_128_predictor_16x16_msa, aom_v_predictor_16x16_msa,
+                aom_h_predictor_16x16_msa, NULL, NULL, NULL, NULL)
+#endif  // HAVE_MSA
+
+// -----------------------------------------------------------------------------
+// 32x32, 32x16, 32x64, 32x8
+
+INTRA_PRED_TEST(C_1, TX_32X32, aom_dc_predictor_32x32_c,
+                aom_dc_left_predictor_32x32_c, aom_dc_top_predictor_32x32_c,
+                aom_dc_128_predictor_32x32_c, aom_v_predictor_32x32_c,
+                aom_h_predictor_32x32_c, aom_paeth_predictor_32x32_c,
+                aom_smooth_predictor_32x32_c, aom_smooth_v_predictor_32x32_c,
+                aom_smooth_h_predictor_32x32_c)
+
+INTRA_PRED_TEST(C_2, TX_32X16, aom_dc_predictor_32x16_c,
+                aom_dc_left_predictor_32x16_c, aom_dc_top_predictor_32x16_c,
+                aom_dc_128_predictor_32x16_c, aom_v_predictor_32x16_c,
+                aom_h_predictor_32x16_c, aom_paeth_predictor_32x16_c,
+                aom_smooth_predictor_32x16_c, aom_smooth_v_predictor_32x16_c,
+                aom_smooth_h_predictor_32x16_c)
+
+INTRA_PRED_TEST(C_3, TX_32X64, aom_dc_predictor_32x64_c,
+                aom_dc_left_predictor_32x64_c, aom_dc_top_predictor_32x64_c,
+                aom_dc_128_predictor_32x64_c, aom_v_predictor_32x64_c,
+                aom_h_predictor_32x64_c, aom_paeth_predictor_32x64_c,
+                aom_smooth_predictor_32x64_c, aom_smooth_v_predictor_32x64_c,
+                aom_smooth_h_predictor_32x64_c)
+
+INTRA_PRED_TEST(C_4, TX_32X8, aom_dc_predictor_32x8_c,
+                aom_dc_left_predictor_32x8_c, aom_dc_top_predictor_32x8_c,
+                aom_dc_128_predictor_32x8_c, aom_v_predictor_32x8_c,
+                aom_h_predictor_32x8_c, aom_paeth_predictor_32x8_c,
+                aom_smooth_predictor_32x8_c, aom_smooth_v_predictor_32x8_c,
+                aom_smooth_h_predictor_32x8_c)
+
+#if HAVE_SSE2
+INTRA_PRED_TEST(SSE2_1, TX_32X32, aom_dc_predictor_32x32_sse2,
+                aom_dc_left_predictor_32x32_sse2,
+                aom_dc_top_predictor_32x32_sse2,
+                aom_dc_128_predictor_32x32_sse2, aom_v_predictor_32x32_sse2,
+                aom_h_predictor_32x32_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_2, TX_32X16, aom_dc_predictor_32x16_sse2,
+                aom_dc_left_predictor_32x16_sse2,
+                aom_dc_top_predictor_32x16_sse2,
+                aom_dc_128_predictor_32x16_sse2, aom_v_predictor_32x16_sse2,
+                aom_h_predictor_32x16_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_3, TX_32X64, aom_dc_predictor_32x64_sse2,
+                aom_dc_left_predictor_32x64_sse2,
+                aom_dc_top_predictor_32x64_sse2,
+                aom_dc_128_predictor_32x64_sse2, aom_v_predictor_32x64_sse2,
+                aom_h_predictor_32x64_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_4, TX_32X8, aom_dc_predictor_32x8_sse2,
+                aom_dc_left_predictor_32x8_sse2, aom_dc_top_predictor_32x8_sse2,
+                aom_dc_128_predictor_32x8_sse2, aom_v_predictor_32x8_sse2,
+                aom_h_predictor_32x8_sse2, NULL, NULL, NULL, NULL)
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3
+INTRA_PRED_TEST(SSSE3_1, TX_32X32, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_32x32_ssse3,
+                aom_smooth_predictor_32x32_ssse3,
+                aom_smooth_v_predictor_32x32_ssse3,
+                aom_smooth_h_predictor_32x32_ssse3)
+INTRA_PRED_TEST(SSSE3_2, TX_32X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_32x16_ssse3,
+                aom_smooth_predictor_32x16_ssse3,
+                aom_smooth_v_predictor_32x16_ssse3,
+                aom_smooth_h_predictor_32x16_ssse3)
+INTRA_PRED_TEST(SSSE3_3, TX_32X64, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_32x64_ssse3,
+                aom_smooth_predictor_32x64_ssse3,
+                aom_smooth_v_predictor_32x64_ssse3,
+                aom_smooth_h_predictor_32x64_ssse3)
+INTRA_PRED_TEST(SSSE3_4, TX_32X8, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_32x8_ssse3, aom_smooth_predictor_32x8_ssse3,
+                aom_smooth_v_predictor_32x8_ssse3,
+                aom_smooth_h_predictor_32x8_ssse3)
+#endif  // HAVE_SSSE3
+
+#if HAVE_AVX2
+INTRA_PRED_TEST(AVX2_1, TX_32X32, aom_dc_predictor_32x32_avx2,
+                aom_dc_left_predictor_32x32_avx2,
+                aom_dc_top_predictor_32x32_avx2,
+                aom_dc_128_predictor_32x32_avx2, aom_v_predictor_32x32_avx2,
+                aom_h_predictor_32x32_avx2, aom_paeth_predictor_32x32_avx2,
+                NULL, NULL, NULL)
+INTRA_PRED_TEST(AVX2_2, TX_32X16, aom_dc_predictor_32x16_avx2,
+                aom_dc_left_predictor_32x16_avx2,
+                aom_dc_top_predictor_32x16_avx2,
+                aom_dc_128_predictor_32x16_avx2, aom_v_predictor_32x16_avx2,
+                NULL, aom_paeth_predictor_32x16_avx2, NULL, NULL, NULL)
+INTRA_PRED_TEST(AVX2_3, TX_32X64, aom_dc_predictor_32x64_avx2,
+                aom_dc_left_predictor_32x64_avx2,
+                aom_dc_top_predictor_32x64_avx2,
+                aom_dc_128_predictor_32x64_avx2, aom_v_predictor_32x64_avx2,
+                NULL, aom_paeth_predictor_32x64_avx2, NULL, NULL, NULL)
+#endif  // HAVE_AVX2
+
+#if HAVE_NEON
+INTRA_PRED_TEST(NEON, TX_32X32, aom_dc_predictor_32x32_neon,
+                aom_dc_left_predictor_32x32_neon,
+                aom_dc_top_predictor_32x32_neon,
+                aom_dc_128_predictor_32x32_neon, aom_v_predictor_32x32_neon,
+                aom_h_predictor_32x32_neon, NULL, NULL, NULL, NULL)
+#endif  // HAVE_NEON
+
+#if HAVE_MSA
+INTRA_PRED_TEST(MSA, TX_32X32, aom_dc_predictor_32x32_msa,
+                aom_dc_left_predictor_32x32_msa, aom_dc_top_predictor_32x32_msa,
+                aom_dc_128_predictor_32x32_msa, aom_v_predictor_32x32_msa,
+                aom_h_predictor_32x32_msa, NULL, NULL, NULL, NULL)
+#endif  // HAVE_MSA
+
+// -----------------------------------------------------------------------------
+// 64x64, 64x32, 64x16
+
+INTRA_PRED_TEST(C_1, TX_64X64, aom_dc_predictor_64x64_c,
+                aom_dc_left_predictor_64x64_c, aom_dc_top_predictor_64x64_c,
+                aom_dc_128_predictor_64x64_c, aom_v_predictor_64x64_c,
+                aom_h_predictor_64x64_c, aom_paeth_predictor_64x64_c,
+                aom_smooth_predictor_64x64_c, aom_smooth_v_predictor_64x64_c,
+                aom_smooth_h_predictor_64x64_c)
+
+INTRA_PRED_TEST(C_2, TX_64X32, aom_dc_predictor_64x32_c,
+                aom_dc_left_predictor_64x32_c, aom_dc_top_predictor_64x32_c,
+                aom_dc_128_predictor_64x32_c, aom_v_predictor_64x32_c,
+                aom_h_predictor_64x32_c, aom_paeth_predictor_64x32_c,
+                aom_smooth_predictor_64x32_c, aom_smooth_v_predictor_64x32_c,
+                aom_smooth_h_predictor_64x32_c)
+
+INTRA_PRED_TEST(C_3, TX_64X16, aom_dc_predictor_64x16_c,
+                aom_dc_left_predictor_64x16_c, aom_dc_top_predictor_64x16_c,
+                aom_dc_128_predictor_64x16_c, aom_v_predictor_64x16_c,
+                aom_h_predictor_64x16_c, aom_paeth_predictor_64x16_c,
+                aom_smooth_predictor_64x16_c, aom_smooth_v_predictor_64x16_c,
+                aom_smooth_h_predictor_64x16_c)
+
+#if HAVE_SSE2
+INTRA_PRED_TEST(SSE2_4, TX_64X64, aom_dc_predictor_64x64_sse2,
+                aom_dc_left_predictor_64x64_sse2,
+                aom_dc_top_predictor_64x64_sse2,
+                aom_dc_128_predictor_64x64_sse2, aom_v_predictor_64x64_sse2,
+                aom_h_predictor_64x64_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_5, TX_64X32, aom_dc_predictor_64x32_sse2,
+                aom_dc_left_predictor_64x32_sse2,
+                aom_dc_top_predictor_64x32_sse2,
+                aom_dc_128_predictor_64x32_sse2, aom_v_predictor_64x32_sse2,
+                aom_h_predictor_64x32_sse2, NULL, NULL, NULL, NULL)
+INTRA_PRED_TEST(SSE2_6, TX_64X16, aom_dc_predictor_64x16_sse2,
+                aom_dc_left_predictor_64x16_sse2,
+                aom_dc_top_predictor_64x16_sse2,
+                aom_dc_128_predictor_64x16_sse2, aom_v_predictor_64x16_sse2,
+                aom_h_predictor_64x16_sse2, NULL, NULL, NULL, NULL)
+#endif
+
+#if HAVE_SSSE3
+INTRA_PRED_TEST(SSSE3_4, TX_64X64, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_64x64_ssse3,
+                aom_smooth_predictor_64x64_ssse3,
+                aom_smooth_v_predictor_64x64_ssse3,
+                aom_smooth_h_predictor_64x64_ssse3)
+INTRA_PRED_TEST(SSSE3_5, TX_64X32, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_64x32_ssse3,
+                aom_smooth_predictor_64x32_ssse3,
+                aom_smooth_v_predictor_64x32_ssse3,
+                aom_smooth_h_predictor_64x32_ssse3)
+INTRA_PRED_TEST(SSSE3_6, TX_64X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                aom_paeth_predictor_64x16_ssse3,
+                aom_smooth_predictor_64x16_ssse3,
+                aom_smooth_v_predictor_64x16_ssse3,
+                aom_smooth_h_predictor_64x16_ssse3)
+#endif
+
+#if HAVE_AVX2
+INTRA_PRED_TEST(AVX2_4, TX_64X64, aom_dc_predictor_64x64_avx2,
+                aom_dc_left_predictor_64x64_avx2,
+                aom_dc_top_predictor_64x64_avx2,
+                aom_dc_128_predictor_64x64_avx2, aom_v_predictor_64x64_avx2,
+                NULL, aom_paeth_predictor_64x64_avx2, NULL, NULL, NULL)
+INTRA_PRED_TEST(AVX2_5, TX_64X32, aom_dc_predictor_64x32_avx2,
+                aom_dc_left_predictor_64x32_avx2,
+                aom_dc_top_predictor_64x32_avx2,
+                aom_dc_128_predictor_64x32_avx2, aom_v_predictor_64x32_avx2,
+                NULL, aom_paeth_predictor_64x32_avx2, NULL, NULL, NULL)
+INTRA_PRED_TEST(AVX2_6, TX_64X16, aom_dc_predictor_64x16_avx2,
+                aom_dc_left_predictor_64x16_avx2,
+                aom_dc_top_predictor_64x16_avx2,
+                aom_dc_128_predictor_64x16_avx2, aom_v_predictor_64x16_avx2,
+                NULL, aom_paeth_predictor_64x16_avx2, NULL, NULL, NULL)
+#endif
+// -----------------------------------------------------------------------------
+// High Bitdepth
+namespace {
+
+typedef void (*AvxHighbdPredFunc)(uint16_t *dst, ptrdiff_t y_stride,
+                                  const uint16_t *above, const uint16_t *left,
+                                  int bd);
+
+typedef IntraPredTestMem<uint16_t> Av1HighbdIntraPredTestMem;
+
+void TestHighbdIntraPred(TX_SIZE tx_size, AvxHighbdPredFunc const *pred_funcs,
+                         const char *const signatures[]) {
+  const int block_width = tx_size_wide[tx_size];
+  const int block_height = tx_size_high[tx_size];
+  const int num_pixels_per_test =
+      block_width * block_height * kNumAv1IntraFuncs;
+  const int kNumTests = static_cast<int>(2.e10 / num_pixels_per_test);
+  Av1HighbdIntraPredTestMem intra_pred_test_mem;
+  const int bd = 12;
+  intra_pred_test_mem.Init(block_width, block_height, bd);
+
+  for (int k = 0; k < kNumAv1IntraFuncs; ++k) {
+    if (pred_funcs[k] == NULL) continue;
+    memcpy(intra_pred_test_mem.src, intra_pred_test_mem.ref_src,
+           sizeof(intra_pred_test_mem.src));
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int num_tests = 0; num_tests < kNumTests; ++num_tests) {
+      pred_funcs[k](intra_pred_test_mem.src, intra_pred_test_mem.stride,
+                    intra_pred_test_mem.above, intra_pred_test_mem.left, bd);
+    }
+    libaom_test::ClearSystemState();
+    aom_usec_timer_mark(&timer);
+    const int elapsed_time =
+        static_cast<int>(aom_usec_timer_elapsed(&timer) / 1000);
+    CheckMd5Signature(
+        tx_size, true, signatures, intra_pred_test_mem.src,
+        intra_pred_test_mem.num_pixels * sizeof(*intra_pred_test_mem.src),
+        elapsed_time, k);
+  }
+}
+
+static const char *const kHighbdSignatures[TX_SIZES_ALL][kNumAv1IntraFuncs] = {
+  {
+      // 4X4
+      "11f74af6c5737df472f3275cbde062fa",
+      "51bea056b6447c93f6eb8f6b7e8f6f71",
+      "27e97f946766331795886f4de04c5594",
+      "53ab15974b049111fb596c5168ec7e3f",
+      "f0b640bb176fbe4584cf3d32a9b0320a",
+      "729783ca909e03afd4b47111c80d967b",
+      "6e30009c45474a22032678b1bd579c8f",
+      "e57cba016d808aa8a35619df2a65f049",
+      "55a6c37f39afcbbf5abca4a985b96459",
+      "a623d45b37dafec1f8a75c4c5218913d",
+  },
+  {
+      // 8X8
+      "03da8829fe94663047fd108c5fcaa71d",
+      "ecdb37b8120a2d3a4c706b016bd1bfd7",
+      "1d4543ed8d2b9368cb96898095fe8a75",
+      "f791c9a67b913cbd82d9da8ecede30e2",
+      "065c70646f4dbaff913282f55a45a441",
+      "51f87123616662ef7c35691497dfd0ba",
+      "85c01ba03df68f9ece7bd3fa0f8980e6",
+      "ad19b7dac092f56df6d054e1f67f21e7",
+      "0edc415b5dd7299f7a34fb9f71d31d78",
+      "2bc8ec19e9f4b77a64b8a0a1f6aec7e7",
+  },
+  {
+      // 16X16
+      "e33cb3f56a878e2fddb1b2fc51cdd275",
+      "c7bff6f04b6052c8ab335d726dbbd52d",
+      "d0b0b47b654a9bcc5c6008110a44589b",
+      "78f5da7b10b2b9ab39f114a33b6254e9",
+      "c78e31d23831abb40d6271a318fdd6f3",
+      "90d1347f4ec9198a0320daecb6ff90b8",
+      "e63ded54ab3d0e8728b6f24d4f01e53f",
+      "35ce21fbe0ea114c089fc3489a78155d",
+      "f277f6ef8e4d717f1f0dfe2706ac197d",
+      "e8014d3f41256976c02e0f1e622ba2b9",
+  },
+  {
+      // 32X32
+      "a3e8056ba7e36628cce4917cd956fedd",
+      "cc7d3024fe8748b512407edee045377e",
+      "2aab0a0f330a1d3e19b8ecb8f06387a3",
+      "a547bc3fb7b06910bf3973122a426661",
+      "26f712514da95042f93d6e8dc8e431dc",
+      "bb08c6e16177081daa3d936538dbc2e3",
+      "84bf83f94a51b33654ca940c6f8bc057",
+      "7168b03fc31bf29596a344d6a35d007c",
+      "b073a70d3672f1282236994f5d12e94b",
+      "c51607aebad5dcb3c1e3b58ef9e5b84e",
+  },
+  {
+      // 64X64
+      "a6baa0d4bfb2269a94c7a38f86a4bccf",
+      "3f1ef5f473a49eba743f17a3324adf9d",
+      "12ac11889ae5f55b7781454efd706a6a",
+      "d9a906c0e692b22e1b4414e71a704b7e",
+      "47d4cadd56f70c11ff8f3e5d8df81161",
+      "de997744cf24c16c5ac2a36b02b351cc",
+      "23781211ae178ddeb6c4bb97a6bd7d83",
+      "a79d2e28340ca34b9e37daabbf030f63",
+      "0372bd3ddfc258750a6ac106b70587f4",
+      "228ef625d9460cbf6fa253a16a730976",
+  },
+  {
+      // 4X8
+      "22d519b796d59644043466320e4ccd14",
+      "09513a738c49b3f9542d27f34abbe1d5",
+      "807ae5e8813443ff01e71be6efacfb69",
+      "cbfa18d0293430b6e9708b0be1fd2394",
+      "346c354c34ec7fa780b576db355dab88",
+      "f97dae85c35359632380b09ca98d611e",
+      "698ae351d8896d89ed9e4e67b6e53eda",
+      "dcc197034a9c45a3d8238bf085835f4e",
+      "7a35e2c42ffdc2efc2d6d1d75a100fc7",
+      "41ab6cebd4516c87a91b2a593e2c2506",
+  },
+  {
+      // 8X4
+      "d58cd4c4bf3b7bbaa5db5e1a5622ec78",
+      "6e572c35aa782d00cafcb99e9ea047ea",
+      "e8c22a3702b416dc9ab974505afbed09",
+      "aaa4e4762a795aad7ad74de0c662c4e4",
+      "a19f9101967383c3dcbd516dc317a291",
+      "9ab8cb91f1a595b9ebe3fe8de58031aa",
+      "2cf9021d5f1169268699807ee118b65f",
+      "ee9605fcbd6fb871f1c5cd81a6989327",
+      "b4871af8316089e3e23522175df7e93f",
+      "d33301e1c2cb173be46792a22d19881a",
+  },
+  {
+      // 8X16
+      "4562de1d0336610880fdd5685498a9ec",
+      "16310fa7076394f16fc85c4b149d89c9",
+      "0e94af88e1dc573b6f0f499cddd1f530",
+      "dfd245ee20d091c67809160340365aa9",
+      "d3562504327f70c096c5be23fd8a3747",
+      "601b853558502acbb5135eadd2da117a",
+      "3c624345a723a1b2b1bea05a6a08bc99",
+      "2a9c781de609e0184cc7ab442050f4e5",
+      "0ddc5035c22252747126b61fc238c74d",
+      "e43f5d83bab759af69c7b6773fc8f9b2",
+  },
+  {
+      // 16X8
+      "a57d6b5a9bfd30c29591d8717ace9c51",
+      "f5907ba97ee6c53e339e953fc8d845ee",
+      "ea3aa727913ce45af06f89dd1808db5f",
+      "408af4f23e48d14b48ee35ae094fcd18",
+      "85c41cbcb5d744f7961e8950026fbffe",
+      "8a4e588a837638887ba671f8d4910485",
+      "b792d8826b67a21757ea7097cff9e05b",
+      "f94ce7101bb87fd3bb9312112527dbf4",
+      "688c6660a6dc6fa61fa1aa38e708c209",
+      "0cdf641b4f81d69509c92ae0b93ef5ff",
+  },
+  {
+      // 16X32
+      "aee4b3b0e3cc02d48e2c40d77f807927",
+      "8baef2b2e789f79c8df9d90ad10f34a4",
+      "038c38ee3c4f090bb8d736eab136aafc",
+      "1a3de2aaeaffd68a9fd6c7f6557b83f3",
+      "385c6e0ea29421dd81011a2934641e26",
+      "6cf96c285d1a2d4787f955dad715b08c",
+      "2d7f75dcd73b9528c8396279ff09ff3a",
+      "5a63cd1841e4ed470e4ca5ef845f2281",
+      "610d899ca945fbead33287d4335a8b32",
+      "6bafaad81fce37be46730187e78d8b11",
+  },
+  {
+      // 32X16
+      "290b23c9f5a1de7905bfa71a942da29b",
+      "701e7b82593c66da5052fc4b6afd79ce",
+      "4da828c5455cd246735a663fbb204989",
+      "e3fbeaf234efece8dbd752b77226200c",
+      "4d1d8c969f05155a7e7e84cf7aad021b",
+      "c22e4877c2c946d5bdc0d542e29e70cf",
+      "8ac1ce815e7780500f842b0beb0bb980",
+      "9fee2e2502b507f25bfad30a55b0b610",
+      "4ced9c212ec6f9956e27f68a91b59fef",
+      "4a7a0b93f138bb0863e4e465b01ec0b1",
+  },
+  {
+      // 32X64
+      "ad9cfc395a5c5644a21d958c7274ac14",
+      "f29d6d03c143ddf96fef04c19f2c8333",
+      "a8bdc852ef704dd4975c61893e8fbc3f",
+      "7d0bd7dea26226741dbca9a97f27fa74",
+      "45c27c5cca9a91b6ae8379feb0881c9f",
+      "8a0b78df1e001b85c874d686eac4aa1b",
+      "ce9fa75fac54a3f6c0cc3f2083b938f1",
+      "c0dca10d88762c954af18dc9e3791a39",
+      "61df229eddfccab913b8fda4bb02f9ac",
+      "4f4df6bc8d50a5600b573f0e44d70e66",
+  },
+  {
+      // 64X32
+      "db9d82921fd88b24fdff6f849f2f9c87",
+      "5ecc7fdc52d2f575ad4f2d0e9e6b1e11",
+      "b4581311a0a73d95dfac7f8f44591032",
+      "68bd283cfd1a125f6b2ee47cee874d36",
+      "804179f05c032908a5e36077bb87c994",
+      "fc5fd041a8ee779015394d0c066ee43c",
+      "68f5579ccadfe9a1baafb158334a3db2",
+      "fe237e45e215ab06d79046da9ad71e84",
+      "9a8a938a6824551bf7d21b8fd1d70ea1",
+      "eb7332f2017cd96882c76e7136aeaf53",
+  },
+  {
+      // 4X16
+      "7bafa307d507747b8132e7735b7f1c73",
+      "e58bc2d8213a97d1fea9cfb73d7a9633",
+      "435f8a8e8bbf14dbf2fe16b2be9e97aa",
+      "1d0e767b68d84acbfb50b7a04e633836",
+      "5f713bd7b324fe73bb7063e35ee14e5e",
+      "0dac4e1fa3d59814202715468c01ed56",
+      "47709d1db4a330c7a8900f450e6fddd1",
+      "258e0b930bb27db28f05da9cf7d1ee7c",
+      "36cf030fbae767912593efea045bfff5",
+      "248d7aceabb7499febae663fae41a920",
+  },
+  {
+      // 16X4
+      "04dde98e632670e393704742c89f9067",
+      "8c72543f1664651ae1fa08e2ac0adb9b",
+      "2354a2cdc2773aa2df8ab4010db1be39",
+      "6300ad3221c26da39b10e0e6d87ee3be",
+      "8ea30b661c6ba60b28d3167f19e449b8",
+      "fb6c1e4ff101a371cede63c2955cdb7e",
+      "a517c06433d6d7927b16a72184a23e92",
+      "393828be5d62ab6c48668bea5e2f801a",
+      "b1e510c542013eb9d6fb188dea2ce90a",
+      "569a8f2fe01679ca216535ecbcdccb62",
+  },
+  {
+      // 8X32
+      "9d541865c185ca7607852852613ac1fc",
+      "b96be67f08c6b5fa5ebd3411299c2f7c",
+      "75a2dcf50004b9d188849b048239767e",
+      "429492ff415c9fd9b050d73b2ad500f8",
+      "64b3606c1ccd036bd766bd5711392cf4",
+      "cb59844a0f01660ac955bae3511f1100",
+      "3e076155b7a70e8828618e3f33b51e3d",
+      "ed2d1f597ab7c50beff690f737cf9726",
+      "7909c6a26aaf20c59d996d3e5b5f9c29",
+      "965798807240c98c6f7cc9b457ed0773",
+  },
+  {
+      // 32X8
+      "36f391aa31619eec1f4d9ee95ea454cc",
+      "b82648f14eeba2527357cb50bc3223cb",
+      "7a7b2adf429125e8bee9d1d00a66e13f",
+      "4198e4d6ba503b7cc2d7e96bb845f661",
+      "96c160d2ec1be9fe0cdea9682f14d257",
+      "19a450bcebaa75afb4fc6bd1fd6434af",
+      "2bd2e35967d43d0ec1c6587a36f204d5",
+      "49799a99aa4ccfbd989bee92a99422f1",
+      "955530e99813812a74659edeac3f5475",
+      "f0316b84e378a19cd11b19a6e40b2914",
+  },
+  {
+      // 16X64
+      "8cba1b70a0bde29e8ef235cedc5faa7d",
+      "96d00ddc7537bf7f196006591b733b4e",
+      "cbf69d5d157c9f3355a4757b1d6e3414",
+      "3ac1f642019493dec1b737d7a3a1b4e5",
+      "35f9ee300d7fa3c97338e81a6f21dcd4",
+      "aae335442e77c8ebc280f16ea50ba9c7",
+      "a6140fdac2278644328be094d88731db",
+      "2df93621b6ff100f7008432d509f4161",
+      "c77bf5aee39e7ed4a3dd715f816f452a",
+      "02109bd63557d90225c32a8f1338258e",
+  },
+  {
+      // 64X16
+      "a5e2f9fb685d5f4a048e9a96affd25a4",
+      "1348f249690d9eefe09d9ad7ead2c801",
+      "525da4b187acd81b1ff1116b60461141",
+      "e99d072de858094c98b01bd4a6772634",
+      "873bfa9dc24693f19721f7c8d527f7d3",
+      "0acfc6507bd3468e9679efc127d6e4b9",
+      "57d03f8d079c7264854e22ac1157cfae",
+      "6c2c4036f70c7d957a9399b5436c0774",
+      "42b8e4a97b7f8416c72a5148c031c0b1",
+      "a38a2c5f79993dfae8530e9e25800893",
+  },
+};
+
+}  // namespace
+
+#define HIGHBD_INTRA_PRED_TEST(arch, tx_size, dc, dc_left, dc_top, dc_128, v, \
+                               h, paeth, smooth, smooth_v, smooth_h)          \
+  TEST(arch, DISABLED_##TestHighbdIntraPred_##tx_size) {                      \
+    static const AvxHighbdPredFunc aom_intra_pred[] = {                       \
+      dc, dc_left, dc_top, dc_128, v, h, paeth, smooth, smooth_v, smooth_h    \
+    };                                                                        \
+    TestHighbdIntraPred(tx_size, aom_intra_pred, kHighbdSignatures[tx_size]); \
+  }
+
+// -----------------------------------------------------------------------------
+// 4x4, 4x8, 4x16
+
+HIGHBD_INTRA_PRED_TEST(
+    C_1, TX_4X4, aom_highbd_dc_predictor_4x4_c,
+    aom_highbd_dc_left_predictor_4x4_c, aom_highbd_dc_top_predictor_4x4_c,
+    aom_highbd_dc_128_predictor_4x4_c, aom_highbd_v_predictor_4x4_c,
+    aom_highbd_h_predictor_4x4_c, aom_highbd_paeth_predictor_4x4_c,
+    aom_highbd_smooth_predictor_4x4_c, aom_highbd_smooth_v_predictor_4x4_c,
+    aom_highbd_smooth_h_predictor_4x4_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_2, TX_4X8, aom_highbd_dc_predictor_4x8_c,
+    aom_highbd_dc_left_predictor_4x8_c, aom_highbd_dc_top_predictor_4x8_c,
+    aom_highbd_dc_128_predictor_4x8_c, aom_highbd_v_predictor_4x8_c,
+    aom_highbd_h_predictor_4x8_c, aom_highbd_paeth_predictor_4x8_c,
+    aom_highbd_smooth_predictor_4x8_c, aom_highbd_smooth_v_predictor_4x8_c,
+    aom_highbd_smooth_h_predictor_4x8_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_3, TX_4X16, aom_highbd_dc_predictor_4x16_c,
+    aom_highbd_dc_left_predictor_4x16_c, aom_highbd_dc_top_predictor_4x16_c,
+    aom_highbd_dc_128_predictor_4x16_c, aom_highbd_v_predictor_4x16_c,
+    aom_highbd_h_predictor_4x16_c, aom_highbd_paeth_predictor_4x16_c,
+    aom_highbd_smooth_predictor_4x16_c, aom_highbd_smooth_v_predictor_4x16_c,
+    aom_highbd_smooth_h_predictor_4x16_c)
+
+#if HAVE_SSE2
+HIGHBD_INTRA_PRED_TEST(SSE2_1, TX_4X4, aom_highbd_dc_predictor_4x4_sse2,
+                       aom_highbd_dc_left_predictor_4x4_sse2,
+                       aom_highbd_dc_top_predictor_4x4_sse2,
+                       aom_highbd_dc_128_predictor_4x4_sse2,
+                       aom_highbd_v_predictor_4x4_sse2,
+                       aom_highbd_h_predictor_4x4_sse2, NULL, NULL, NULL, NULL)
+
+HIGHBD_INTRA_PRED_TEST(SSE2_2, TX_4X8, aom_highbd_dc_predictor_4x8_sse2,
+                       aom_highbd_dc_left_predictor_4x8_sse2,
+                       aom_highbd_dc_top_predictor_4x8_sse2,
+                       aom_highbd_dc_128_predictor_4x8_sse2,
+                       aom_highbd_v_predictor_4x8_sse2,
+                       aom_highbd_h_predictor_4x8_sse2, NULL, NULL, NULL, NULL)
+#endif
+
+// -----------------------------------------------------------------------------
+// 8x8, 8x4, 8x16, 8x32
+
+HIGHBD_INTRA_PRED_TEST(
+    C_1, TX_8X8, aom_highbd_dc_predictor_8x8_c,
+    aom_highbd_dc_left_predictor_8x8_c, aom_highbd_dc_top_predictor_8x8_c,
+    aom_highbd_dc_128_predictor_8x8_c, aom_highbd_v_predictor_8x8_c,
+    aom_highbd_h_predictor_8x8_c, aom_highbd_paeth_predictor_8x8_c,
+    aom_highbd_smooth_predictor_8x8_c, aom_highbd_smooth_v_predictor_8x8_c,
+    aom_highbd_smooth_h_predictor_8x8_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_2, TX_8X4, aom_highbd_dc_predictor_8x4_c,
+    aom_highbd_dc_left_predictor_8x4_c, aom_highbd_dc_top_predictor_8x4_c,
+    aom_highbd_dc_128_predictor_8x4_c, aom_highbd_v_predictor_8x4_c,
+    aom_highbd_h_predictor_8x4_c, aom_highbd_paeth_predictor_8x4_c,
+    aom_highbd_smooth_predictor_8x4_c, aom_highbd_smooth_v_predictor_8x4_c,
+    aom_highbd_smooth_h_predictor_8x4_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_3, TX_8X16, aom_highbd_dc_predictor_8x16_c,
+    aom_highbd_dc_left_predictor_8x16_c, aom_highbd_dc_top_predictor_8x16_c,
+    aom_highbd_dc_128_predictor_8x16_c, aom_highbd_v_predictor_8x16_c,
+    aom_highbd_h_predictor_8x16_c, aom_highbd_paeth_predictor_8x16_c,
+    aom_highbd_smooth_predictor_8x16_c, aom_highbd_smooth_v_predictor_8x16_c,
+    aom_highbd_smooth_h_predictor_8x16_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_4, TX_8X32, aom_highbd_dc_predictor_8x32_c,
+    aom_highbd_dc_left_predictor_8x32_c, aom_highbd_dc_top_predictor_8x32_c,
+    aom_highbd_dc_128_predictor_8x32_c, aom_highbd_v_predictor_8x32_c,
+    aom_highbd_h_predictor_8x32_c, aom_highbd_paeth_predictor_8x32_c,
+    aom_highbd_smooth_predictor_8x32_c, aom_highbd_smooth_v_predictor_8x32_c,
+    aom_highbd_smooth_h_predictor_8x32_c)
+
+#if HAVE_SSE2
+HIGHBD_INTRA_PRED_TEST(SSE2_1, TX_8X8, aom_highbd_dc_predictor_8x8_sse2,
+                       aom_highbd_dc_left_predictor_8x8_sse2,
+                       aom_highbd_dc_top_predictor_8x8_sse2,
+                       aom_highbd_dc_128_predictor_8x8_sse2,
+                       aom_highbd_v_predictor_8x8_sse2,
+                       aom_highbd_h_predictor_8x8_sse2, NULL, NULL, NULL, NULL)
+HIGHBD_INTRA_PRED_TEST(SSE2_2, TX_8X4, aom_highbd_dc_predictor_8x4_sse2,
+                       aom_highbd_dc_left_predictor_8x4_sse2,
+                       aom_highbd_dc_top_predictor_8x4_sse2,
+                       aom_highbd_dc_128_predictor_8x4_sse2,
+                       aom_highbd_v_predictor_8x4_sse2,
+                       aom_highbd_h_predictor_8x4_sse2, NULL, NULL, NULL, NULL)
+HIGHBD_INTRA_PRED_TEST(SSE2_3, TX_8X16, aom_highbd_dc_predictor_8x16_sse2,
+                       aom_highbd_dc_left_predictor_8x16_sse2,
+                       aom_highbd_dc_top_predictor_8x16_sse2,
+                       aom_highbd_dc_128_predictor_8x16_sse2,
+                       aom_highbd_v_predictor_8x16_sse2,
+                       aom_highbd_h_predictor_8x16_sse2, NULL, NULL, NULL, NULL)
+#endif
+
+#if HAVE_SSSE3
+HIGHBD_INTRA_PRED_TEST(SSSE3, TX_8X8, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
+                       NULL, NULL, NULL)
+#endif
+
+// -----------------------------------------------------------------------------
+// 16x16, 16x8, 16x32, 16x4, 16x64
+
+HIGHBD_INTRA_PRED_TEST(
+    C_1, TX_16X16, aom_highbd_dc_predictor_16x16_c,
+    aom_highbd_dc_left_predictor_16x16_c, aom_highbd_dc_top_predictor_16x16_c,
+    aom_highbd_dc_128_predictor_16x16_c, aom_highbd_v_predictor_16x16_c,
+    aom_highbd_h_predictor_16x16_c, aom_highbd_paeth_predictor_16x16_c,
+    aom_highbd_smooth_predictor_16x16_c, aom_highbd_smooth_v_predictor_16x16_c,
+    aom_highbd_smooth_h_predictor_16x16_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_2, TX_16X8, aom_highbd_dc_predictor_16x8_c,
+    aom_highbd_dc_left_predictor_16x8_c, aom_highbd_dc_top_predictor_16x8_c,
+    aom_highbd_dc_128_predictor_16x8_c, aom_highbd_v_predictor_16x8_c,
+    aom_highbd_h_predictor_16x8_c, aom_highbd_paeth_predictor_16x8_c,
+    aom_highbd_smooth_predictor_16x8_c, aom_highbd_smooth_v_predictor_16x8_c,
+    aom_highbd_smooth_h_predictor_16x8_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_3, TX_16X32, aom_highbd_dc_predictor_16x32_c,
+    aom_highbd_dc_left_predictor_16x32_c, aom_highbd_dc_top_predictor_16x32_c,
+    aom_highbd_dc_128_predictor_16x32_c, aom_highbd_v_predictor_16x32_c,
+    aom_highbd_h_predictor_16x32_c, aom_highbd_paeth_predictor_16x32_c,
+    aom_highbd_smooth_predictor_16x32_c, aom_highbd_smooth_v_predictor_16x32_c,
+    aom_highbd_smooth_h_predictor_16x32_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_4, TX_16X4, aom_highbd_dc_predictor_16x4_c,
+    aom_highbd_dc_left_predictor_16x4_c, aom_highbd_dc_top_predictor_16x4_c,
+    aom_highbd_dc_128_predictor_16x4_c, aom_highbd_v_predictor_16x4_c,
+    aom_highbd_h_predictor_16x4_c, aom_highbd_paeth_predictor_16x4_c,
+    aom_highbd_smooth_predictor_16x4_c, aom_highbd_smooth_v_predictor_16x4_c,
+    aom_highbd_smooth_h_predictor_16x4_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_5, TX_16X64, aom_highbd_dc_predictor_16x64_c,
+    aom_highbd_dc_left_predictor_16x64_c, aom_highbd_dc_top_predictor_16x64_c,
+    aom_highbd_dc_128_predictor_16x64_c, aom_highbd_v_predictor_16x64_c,
+    aom_highbd_h_predictor_16x64_c, aom_highbd_paeth_predictor_16x64_c,
+    aom_highbd_smooth_predictor_16x64_c, aom_highbd_smooth_v_predictor_16x64_c,
+    aom_highbd_smooth_h_predictor_16x64_c)
+
+#if HAVE_SSE2
+HIGHBD_INTRA_PRED_TEST(SSE2_1, TX_16X16, aom_highbd_dc_predictor_16x16_sse2,
+                       aom_highbd_dc_left_predictor_16x16_sse2,
+                       aom_highbd_dc_top_predictor_16x16_sse2,
+                       aom_highbd_dc_128_predictor_16x16_sse2,
+                       aom_highbd_v_predictor_16x16_sse2,
+                       aom_highbd_h_predictor_16x16_sse2, NULL, NULL, NULL,
+                       NULL)
+HIGHBD_INTRA_PRED_TEST(SSE2_2, TX_16X8, aom_highbd_dc_predictor_16x8_sse2,
+                       aom_highbd_dc_left_predictor_16x8_sse2,
+                       aom_highbd_dc_top_predictor_16x8_sse2,
+                       aom_highbd_dc_128_predictor_16x8_sse2,
+                       aom_highbd_v_predictor_16x8_sse2,
+                       aom_highbd_h_predictor_16x8_sse2, NULL, NULL, NULL, NULL)
+HIGHBD_INTRA_PRED_TEST(SSE2_3, TX_16X32, aom_highbd_dc_predictor_16x32_sse2,
+                       aom_highbd_dc_left_predictor_16x32_sse2,
+                       aom_highbd_dc_top_predictor_16x32_sse2,
+                       aom_highbd_dc_128_predictor_16x32_sse2,
+                       aom_highbd_v_predictor_16x32_sse2,
+                       aom_highbd_h_predictor_16x32_sse2, NULL, NULL, NULL,
+                       NULL)
+#endif
+
+#if HAVE_SSSE3
+HIGHBD_INTRA_PRED_TEST(SSSE3_1, TX_16X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                       NULL, NULL, NULL, NULL)
+#endif
+
+#if HAVE_AVX2
+HIGHBD_INTRA_PRED_TEST(AVX2_1, TX_16X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                       NULL, NULL, NULL, NULL)
+
+HIGHBD_INTRA_PRED_TEST(AVX2_2, TX_16X8, NULL, NULL, NULL, NULL, NULL, NULL,
+                       NULL, NULL, NULL, NULL)
+
+HIGHBD_INTRA_PRED_TEST(AVX2_3, TX_16X32, NULL, NULL, NULL, NULL, NULL, NULL,
+                       NULL, NULL, NULL, NULL)
+#endif
+
+// -----------------------------------------------------------------------------
+// 32x32, 32x16, 32x64, 32x8
+
+HIGHBD_INTRA_PRED_TEST(
+    C_1, TX_32X32, aom_highbd_dc_predictor_32x32_c,
+    aom_highbd_dc_left_predictor_32x32_c, aom_highbd_dc_top_predictor_32x32_c,
+    aom_highbd_dc_128_predictor_32x32_c, aom_highbd_v_predictor_32x32_c,
+    aom_highbd_h_predictor_32x32_c, aom_highbd_paeth_predictor_32x32_c,
+    aom_highbd_smooth_predictor_32x32_c, aom_highbd_smooth_v_predictor_32x32_c,
+    aom_highbd_smooth_h_predictor_32x32_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_2, TX_32X16, aom_highbd_dc_predictor_32x16_c,
+    aom_highbd_dc_left_predictor_32x16_c, aom_highbd_dc_top_predictor_32x16_c,
+    aom_highbd_dc_128_predictor_32x16_c, aom_highbd_v_predictor_32x16_c,
+    aom_highbd_h_predictor_32x16_c, aom_highbd_paeth_predictor_32x16_c,
+    aom_highbd_smooth_predictor_32x16_c, aom_highbd_smooth_v_predictor_32x16_c,
+    aom_highbd_smooth_h_predictor_32x16_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_3, TX_32X64, aom_highbd_dc_predictor_32x64_c,
+    aom_highbd_dc_left_predictor_32x64_c, aom_highbd_dc_top_predictor_32x64_c,
+    aom_highbd_dc_128_predictor_32x64_c, aom_highbd_v_predictor_32x64_c,
+    aom_highbd_h_predictor_32x64_c, aom_highbd_paeth_predictor_32x64_c,
+    aom_highbd_smooth_predictor_32x64_c, aom_highbd_smooth_v_predictor_32x64_c,
+    aom_highbd_smooth_h_predictor_32x64_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_4, TX_32X8, aom_highbd_dc_predictor_32x8_c,
+    aom_highbd_dc_left_predictor_32x8_c, aom_highbd_dc_top_predictor_32x8_c,
+    aom_highbd_dc_128_predictor_32x8_c, aom_highbd_v_predictor_32x8_c,
+    aom_highbd_h_predictor_32x8_c, aom_highbd_paeth_predictor_32x8_c,
+    aom_highbd_smooth_predictor_32x8_c, aom_highbd_smooth_v_predictor_32x8_c,
+    aom_highbd_smooth_h_predictor_32x8_c)
+
+#if HAVE_SSE2
+HIGHBD_INTRA_PRED_TEST(SSE2_1, TX_32X32, aom_highbd_dc_predictor_32x32_sse2,
+                       aom_highbd_dc_left_predictor_32x32_sse2,
+                       aom_highbd_dc_top_predictor_32x32_sse2,
+                       aom_highbd_dc_128_predictor_32x32_sse2,
+                       aom_highbd_v_predictor_32x32_sse2,
+                       aom_highbd_h_predictor_32x32_sse2, NULL, NULL, NULL,
+                       NULL)
+HIGHBD_INTRA_PRED_TEST(SSE2_2, TX_32X16, aom_highbd_dc_predictor_32x16_sse2,
+                       aom_highbd_dc_left_predictor_32x16_sse2,
+                       aom_highbd_dc_top_predictor_32x16_sse2,
+                       aom_highbd_dc_128_predictor_32x16_sse2,
+                       aom_highbd_v_predictor_32x16_sse2,
+                       aom_highbd_h_predictor_32x16_sse2, NULL, NULL, NULL,
+                       NULL)
+#endif
+
+#if HAVE_SSSE3
+HIGHBD_INTRA_PRED_TEST(SSSE3_1, TX_32X32, NULL, NULL, NULL, NULL, NULL, NULL,
+                       NULL, NULL, NULL, NULL)
+#endif
+
+#if HAVE_AVX2
+HIGHBD_INTRA_PRED_TEST(AVX2_1, TX_32X32, NULL, NULL, NULL, NULL, NULL, NULL,
+                       NULL, NULL, NULL, NULL)
+
+HIGHBD_INTRA_PRED_TEST(AVX2_2, TX_32X16, NULL, NULL, NULL, NULL, NULL, NULL,
+                       NULL, NULL, NULL, NULL)
+#endif
+
+// -----------------------------------------------------------------------------
+// 64x64, 64x32, 64x16
+
+HIGHBD_INTRA_PRED_TEST(
+    C_1, TX_64X64, aom_highbd_dc_predictor_64x64_c,
+    aom_highbd_dc_left_predictor_64x64_c, aom_highbd_dc_top_predictor_64x64_c,
+    aom_highbd_dc_128_predictor_64x64_c, aom_highbd_v_predictor_64x64_c,
+    aom_highbd_h_predictor_64x64_c, aom_highbd_paeth_predictor_64x64_c,
+    aom_highbd_smooth_predictor_64x64_c, aom_highbd_smooth_v_predictor_64x64_c,
+    aom_highbd_smooth_h_predictor_64x64_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_2, TX_64X32, aom_highbd_dc_predictor_64x32_c,
+    aom_highbd_dc_left_predictor_64x32_c, aom_highbd_dc_top_predictor_64x32_c,
+    aom_highbd_dc_128_predictor_64x32_c, aom_highbd_v_predictor_64x32_c,
+    aom_highbd_h_predictor_64x32_c, aom_highbd_paeth_predictor_64x32_c,
+    aom_highbd_smooth_predictor_64x32_c, aom_highbd_smooth_v_predictor_64x32_c,
+    aom_highbd_smooth_h_predictor_64x32_c)
+
+HIGHBD_INTRA_PRED_TEST(
+    C_3, TX_64X16, aom_highbd_dc_predictor_64x16_c,
+    aom_highbd_dc_left_predictor_64x16_c, aom_highbd_dc_top_predictor_64x16_c,
+    aom_highbd_dc_128_predictor_64x16_c, aom_highbd_v_predictor_64x16_c,
+    aom_highbd_h_predictor_64x16_c, aom_highbd_paeth_predictor_64x16_c,
+    aom_highbd_smooth_predictor_64x16_c, aom_highbd_smooth_v_predictor_64x16_c,
+    aom_highbd_smooth_h_predictor_64x16_c)
+
+// -----------------------------------------------------------------------------
+
+#include "test/test_libaom.cc"
diff --git a/third_party/aom/test/test_libaom.cc b/third_party/aom/test/test_libaom.cc
new file mode 100644
index 0000000000..b55d762373
--- /dev/null
+++ b/third_party/aom/test/test_libaom.cc
@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string.h>
+
+#include <string>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+
+#if ARCH_X86 || ARCH_X86_64
+#include "aom_ports/x86.h"
+#endif
+extern "C" {
+extern void av1_rtcd();
+extern void aom_dsp_rtcd();
+extern void aom_scale_rtcd();
+}
+
+#if ARCH_X86 || ARCH_X86_64
+static void append_negative_gtest_filter(const char *str) {
+  std::string filter = ::testing::FLAGS_gtest_filter;
+  // Negative patterns begin with one '-' followed by a ':' separated list.
+  if (filter.find('-') == std::string::npos) filter += '-';
+  // OPT.* matches TEST() functions
+  // OPT/* matches TEST_P() functions
+  // OPT_* matches tests which have been manually sharded.
+  // We do not match OPT* because of SSE/SSE2 collisions.
+  const char *search_terminators = "./_";
+  for (size_t pos = 0; pos < strlen(search_terminators); ++pos) {
+    filter += ":";
+    filter += str;
+    filter += search_terminators[pos];
+    filter += "*";
+  }
+  ::testing::FLAGS_gtest_filter = filter;
+}
+#endif  // ARCH_X86 || ARCH_X86_64
+
+int main(int argc, char **argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+
+#if ARCH_X86 || ARCH_X86_64
+  const int simd_caps = x86_simd_caps();
+  if (!(simd_caps & HAS_MMX)) append_negative_gtest_filter("MMX");
+  if (!(simd_caps & HAS_SSE)) append_negative_gtest_filter("SSE");
+  if (!(simd_caps & HAS_SSE2)) append_negative_gtest_filter("SSE2");
+  if (!(simd_caps & HAS_SSE3)) append_negative_gtest_filter("SSE3");
+  if (!(simd_caps & HAS_SSSE3)) append_negative_gtest_filter("SSSE3");
+  if (!(simd_caps & HAS_SSE4_1)) append_negative_gtest_filter("SSE4_1");
+  if (!(simd_caps & HAS_SSE4_2)) append_negative_gtest_filter("SSE4_2");
+  if (!(simd_caps & HAS_AVX)) append_negative_gtest_filter("AVX");
+  if (!(simd_caps & HAS_AVX2)) append_negative_gtest_filter("AVX2");
+#endif  // ARCH_X86 || ARCH_X86_64
+
+// Shared library builds don't support whitebox tests that exercise internal
+// symbols.
+#if !CONFIG_SHARED
+  av1_rtcd();
+  aom_dsp_rtcd();
+  aom_scale_rtcd();
+#endif  // !CONFIG_SHARED
+
+  return RUN_ALL_TESTS();
+}
diff --git a/third_party/aom/test/test_runner.cmake b/third_party/aom/test/test_runner.cmake
new file mode 100644
index 0000000000..d3747b1e30
--- /dev/null
+++ b/third_party/aom/test/test_runner.cmake
@@ -0,0 +1,28 @@
+#
+# Copyright (c) 2017, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and the
+# Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License was
+# not distributed with this source code in the LICENSE file, you can obtain it
+# at www.aomedia.org/license/software. If the Alliance for Open Media Patent
+# License 1.0 was not distributed with this source code in the PATENTS file, you
+# can obtain it at www.aomedia.org/license/patent.
+#
+if(NOT GTEST_TOTAL_SHARDS OR "${GTEST_SHARD_INDEX}" STREQUAL "" OR NOT
+   TEST_LIBAOM)
+  message(
+    FATAL_ERROR
+      "The variables GTEST_SHARD_INDEX, GTEST_TOTAL_SHARDS and TEST_LIBAOM
+          must be defined."
+    )
+endif()
+
+set($ENV{GTEST_SHARD_INDEX} ${GTEST_SHARD_INDEX})
+set($ENV{GTEST_TOTAL_SHARDS} ${GTEST_TOTAL_SHARDS})
+execute_process(COMMAND ${TEST_LIBAOM} RESULT_VARIABLE test_result)
+set(test_message "Test shard ${GTEST_SHARD_INDEX}/${GTEST_TOTAL_SHARDS} result")
+message("${test_message}: ${test_result}")
+
+if(NOT "${test_result}" STREQUAL "0")
+  message(FATAL_ERROR "${test_message}: FAILED, non-zero exit code.")
+endif()
diff --git a/third_party/aom/test/test_vector_test.cc b/third_party/aom/test/test_vector_test.cc
new file mode 100644
index 0000000000..286988b170
--- /dev/null
+++ b/third_party/aom/test/test_vector_test.cc
@@ -0,0 +1,172 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+#include <set>
+#include <string>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "common/tools_common.h"
+#include "config/aom_config.h"
+#include "test/codec_factory.h"
+#include "test/decode_test_driver.h"
+#include "test/ivf_video_source.h"
+#include "test/md5_helper.h"
+#include "test/test_vectors.h"
+#include "test/util.h"
+#if CONFIG_WEBM_IO
+#include "test/webm_video_source.h"
+#endif
+
+namespace {
+
+const int kThreads = 0;
+const int kFileName = 1;
+const int kRowMT = 2;
+
+typedef ::testing::tuple<int, const char *, int> DecodeParam;
+
+class TestVectorTest : public ::libaom_test::DecoderTest,
+                       public ::libaom_test::CodecTestWithParam<DecodeParam> {
+ protected:
+  TestVectorTest() : DecoderTest(GET_PARAM(0)), md5_file_(NULL) {}
+
+  virtual ~TestVectorTest() {
+    if (md5_file_) fclose(md5_file_);
+  }
+
+  void OpenMD5File(const std::string &md5_file_name_) {
+    md5_file_ = libaom_test::OpenTestDataFile(md5_file_name_);
+    ASSERT_TRUE(md5_file_ != NULL)
+        << "Md5 file open failed. Filename: " << md5_file_name_;
+  }
+
+  virtual void PreDecodeFrameHook(
+      const libaom_test::CompressedVideoSource &video,
+      libaom_test::Decoder *decoder) {
+    if (video.frame_number() == 0) decoder->Control(AV1D_SET_ROW_MT, row_mt_);
+  }
+
+  virtual void DecompressedFrameHook(const aom_image_t &img,
+                                     const unsigned int frame_number) {
+    ASSERT_TRUE(md5_file_ != NULL);
+    char expected_md5[33];
+    char junk[128];
+
+    // Read correct md5 checksums.
+    const int res = fscanf(md5_file_, "%s  %s", expected_md5, junk);
+    ASSERT_NE(res, EOF) << "Read md5 data failed";
+    expected_md5[32] = '\0';
+
+    ::libaom_test::MD5 md5_res;
+#if !CONFIG_LOWBITDEPTH
+    const aom_img_fmt_t shifted_fmt =
+        (aom_img_fmt)(img.fmt & ~AOM_IMG_FMT_HIGHBITDEPTH);
+    if (img.bit_depth == 8 && shifted_fmt != img.fmt) {
+      aom_image_t *img_shifted =
+          aom_img_alloc(NULL, shifted_fmt, img.d_w, img.d_h, 16);
+      img_shifted->bit_depth = img.bit_depth;
+      img_shifted->monochrome = img.monochrome;
+      aom_img_downshift(img_shifted, &img, 0);
+      md5_res.Add(img_shifted);
+      aom_img_free(img_shifted);
+    } else {
+#endif
+      md5_res.Add(&img);
+#if !CONFIG_LOWBITDEPTH
+    }
+#endif
+
+    const char *actual_md5 = md5_res.Get();
+    // Check md5 match.
+    ASSERT_STREQ(expected_md5, actual_md5)
+        << "Md5 checksums don't match: frame number = " << frame_number;
+  }
+
+  unsigned int row_mt_;
+
+ private:
+  FILE *md5_file_;
+};
+
+// This test runs through the whole set of test vectors, and decodes them.
+// The md5 checksums are computed for each frame in the video file. If md5
+// checksums match the correct md5 data, then the test is passed. Otherwise,
+// the test failed.
+TEST_P(TestVectorTest, MD5Match) {
+  const DecodeParam input = GET_PARAM(1);
+  const std::string filename = ::testing::get<kFileName>(input);
+  aom_codec_flags_t flags = 0;
+  aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+  char str[256];
+
+  cfg.threads = ::testing::get<kThreads>(input);
+  row_mt_ = ::testing::get<kRowMT>(input);
+
+  snprintf(str, sizeof(str) / sizeof(str[0]) - 1, "file: %s threads: %d",
+           filename.c_str(), cfg.threads);
+  SCOPED_TRACE(str);
+
+  // Open compressed video file.
+  testing::internal::scoped_ptr<libaom_test::CompressedVideoSource> video;
+  if (filename.substr(filename.length() - 3, 3) == "ivf") {
+    video.reset(new libaom_test::IVFVideoSource(filename));
+  } else if (filename.substr(filename.length() - 4, 4) == "webm" ||
+             filename.substr(filename.length() - 3, 3) == "mkv") {
+#if CONFIG_WEBM_IO
+    video.reset(new libaom_test::WebMVideoSource(filename));
+#else
+    fprintf(stderr, "WebM IO is disabled, skipping test vector %s\n",
+            filename.c_str());
+    return;
+#endif
+  }
+  ASSERT_TRUE(video.get() != NULL);
+  video->Init();
+
+  // Construct md5 file name.
+  const std::string md5_filename = filename + ".md5";
+  OpenMD5File(md5_filename);
+
+  // Set decode config and flags.
+  cfg.allow_lowbitdepth = CONFIG_LOWBITDEPTH;
+  set_cfg(cfg);
+  set_flags(flags);
+
+  // Decode frame, and check the md5 matching.
+  ASSERT_NO_FATAL_FAILURE(RunLoop(video.get(), cfg));
+}
+
+#if CONFIG_AV1_DECODER
+AV1_INSTANTIATE_TEST_CASE(
+    TestVectorTest,
+    ::testing::Combine(::testing::Values(1),  // Single thread.
+                       ::testing::ValuesIn(libaom_test::kAV1TestVectors,
+                                           libaom_test::kAV1TestVectors +
+                                               libaom_test::kNumAV1TestVectors),
+                       ::testing::Values(0)));
+
+// Test AV1 decode in with different numbers of threads.
+INSTANTIATE_TEST_CASE_P(
+    AV1MultiThreaded, TestVectorTest,
+    ::testing::Combine(
+        ::testing::Values(
+            static_cast<const libaom_test::CodecFactory *>(&libaom_test::kAV1)),
+        ::testing::Combine(
+            ::testing::Range(2, 9),  // With 2 ~ 8 threads.
+            ::testing::ValuesIn(libaom_test::kAV1TestVectors,
+                                libaom_test::kAV1TestVectors +
+                                    libaom_test::kNumAV1TestVectors),
+            ::testing::Range(0, 2))));
+
+#endif  // CONFIG_AV1_DECODER
+
+}  // namespace
diff --git a/third_party/aom/test/test_vectors.cc b/third_party/aom/test/test_vectors.cc
new file mode 100644
index 0000000000..71e431e18b
--- /dev/null
+++ b/third_party/aom/test/test_vectors.cc
@@ -0,0 +1,140 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include "test/test_vectors.h"
+
+namespace libaom_test {
+
+#define NELEMENTS(x) static_cast<int>(sizeof(x) / sizeof(x[0]))
+
+#if CONFIG_AV1_DECODER
+const char *const kAV1TestVectors[] = {
+  "av1-1-b8-00-quantizer-00.ivf",  "av1-1-b8-00-quantizer-01.ivf",
+  "av1-1-b8-00-quantizer-02.ivf",  "av1-1-b8-00-quantizer-03.ivf",
+  "av1-1-b8-00-quantizer-04.ivf",  "av1-1-b8-00-quantizer-05.ivf",
+  "av1-1-b8-00-quantizer-06.ivf",  "av1-1-b8-00-quantizer-07.ivf",
+  "av1-1-b8-00-quantizer-08.ivf",  "av1-1-b8-00-quantizer-09.ivf",
+  "av1-1-b8-00-quantizer-10.ivf",  "av1-1-b8-00-quantizer-11.ivf",
+  "av1-1-b8-00-quantizer-12.ivf",  "av1-1-b8-00-quantizer-13.ivf",
+  "av1-1-b8-00-quantizer-14.ivf",  "av1-1-b8-00-quantizer-15.ivf",
+  "av1-1-b8-00-quantizer-16.ivf",  "av1-1-b8-00-quantizer-17.ivf",
+  "av1-1-b8-00-quantizer-18.ivf",  "av1-1-b8-00-quantizer-19.ivf",
+  "av1-1-b8-00-quantizer-20.ivf",  "av1-1-b8-00-quantizer-21.ivf",
+  "av1-1-b8-00-quantizer-22.ivf",  "av1-1-b8-00-quantizer-23.ivf",
+  "av1-1-b8-00-quantizer-24.ivf",  "av1-1-b8-00-quantizer-25.ivf",
+  "av1-1-b8-00-quantizer-26.ivf",  "av1-1-b8-00-quantizer-27.ivf",
+  "av1-1-b8-00-quantizer-28.ivf",  "av1-1-b8-00-quantizer-29.ivf",
+  "av1-1-b8-00-quantizer-30.ivf",  "av1-1-b8-00-quantizer-31.ivf",
+  "av1-1-b8-00-quantizer-32.ivf",  "av1-1-b8-00-quantizer-33.ivf",
+  "av1-1-b8-00-quantizer-34.ivf",  "av1-1-b8-00-quantizer-35.ivf",
+  "av1-1-b8-00-quantizer-36.ivf",  "av1-1-b8-00-quantizer-37.ivf",
+  "av1-1-b8-00-quantizer-38.ivf",  "av1-1-b8-00-quantizer-39.ivf",
+  "av1-1-b8-00-quantizer-40.ivf",  "av1-1-b8-00-quantizer-41.ivf",
+  "av1-1-b8-00-quantizer-42.ivf",  "av1-1-b8-00-quantizer-43.ivf",
+  "av1-1-b8-00-quantizer-44.ivf",  "av1-1-b8-00-quantizer-45.ivf",
+  "av1-1-b8-00-quantizer-46.ivf",  "av1-1-b8-00-quantizer-47.ivf",
+  "av1-1-b8-00-quantizer-48.ivf",  "av1-1-b8-00-quantizer-49.ivf",
+  "av1-1-b8-00-quantizer-50.ivf",  "av1-1-b8-00-quantizer-51.ivf",
+  "av1-1-b8-00-quantizer-52.ivf",  "av1-1-b8-00-quantizer-53.ivf",
+  "av1-1-b8-00-quantizer-54.ivf",  "av1-1-b8-00-quantizer-55.ivf",
+  "av1-1-b8-00-quantizer-56.ivf",  "av1-1-b8-00-quantizer-57.ivf",
+  "av1-1-b8-00-quantizer-58.ivf",  "av1-1-b8-00-quantizer-59.ivf",
+  "av1-1-b8-00-quantizer-60.ivf",  "av1-1-b8-00-quantizer-61.ivf",
+  "av1-1-b8-00-quantizer-62.ivf",  "av1-1-b8-00-quantizer-63.ivf",
+  "av1-1-b10-00-quantizer-00.ivf", "av1-1-b10-00-quantizer-01.ivf",
+  "av1-1-b10-00-quantizer-02.ivf", "av1-1-b10-00-quantizer-03.ivf",
+  "av1-1-b10-00-quantizer-04.ivf", "av1-1-b10-00-quantizer-05.ivf",
+  "av1-1-b10-00-quantizer-06.ivf", "av1-1-b10-00-quantizer-07.ivf",
+  "av1-1-b10-00-quantizer-08.ivf", "av1-1-b10-00-quantizer-09.ivf",
+  "av1-1-b10-00-quantizer-10.ivf", "av1-1-b10-00-quantizer-11.ivf",
+  "av1-1-b10-00-quantizer-12.ivf", "av1-1-b10-00-quantizer-13.ivf",
+  "av1-1-b10-00-quantizer-14.ivf", "av1-1-b10-00-quantizer-15.ivf",
+  "av1-1-b10-00-quantizer-16.ivf", "av1-1-b10-00-quantizer-17.ivf",
+  "av1-1-b10-00-quantizer-18.ivf", "av1-1-b10-00-quantizer-19.ivf",
+  "av1-1-b10-00-quantizer-20.ivf", "av1-1-b10-00-quantizer-21.ivf",
+  "av1-1-b10-00-quantizer-22.ivf", "av1-1-b10-00-quantizer-23.ivf",
+  "av1-1-b10-00-quantizer-24.ivf", "av1-1-b10-00-quantizer-25.ivf",
+  "av1-1-b10-00-quantizer-26.ivf", "av1-1-b10-00-quantizer-27.ivf",
+  "av1-1-b10-00-quantizer-28.ivf", "av1-1-b10-00-quantizer-29.ivf",
+  "av1-1-b10-00-quantizer-30.ivf", "av1-1-b10-00-quantizer-31.ivf",
+  "av1-1-b10-00-quantizer-32.ivf", "av1-1-b10-00-quantizer-33.ivf",
+  "av1-1-b10-00-quantizer-34.ivf", "av1-1-b10-00-quantizer-35.ivf",
+  "av1-1-b10-00-quantizer-36.ivf", "av1-1-b10-00-quantizer-37.ivf",
+  "av1-1-b10-00-quantizer-38.ivf", "av1-1-b10-00-quantizer-39.ivf",
+  "av1-1-b10-00-quantizer-40.ivf", "av1-1-b10-00-quantizer-41.ivf",
+  "av1-1-b10-00-quantizer-42.ivf", "av1-1-b10-00-quantizer-43.ivf",
+  "av1-1-b10-00-quantizer-44.ivf", "av1-1-b10-00-quantizer-45.ivf",
+  "av1-1-b10-00-quantizer-46.ivf", "av1-1-b10-00-quantizer-47.ivf",
+  "av1-1-b10-00-quantizer-48.ivf", "av1-1-b10-00-quantizer-49.ivf",
+  "av1-1-b10-00-quantizer-50.ivf", "av1-1-b10-00-quantizer-51.ivf",
+  "av1-1-b10-00-quantizer-52.ivf", "av1-1-b10-00-quantizer-53.ivf",
+  "av1-1-b10-00-quantizer-54.ivf", "av1-1-b10-00-quantizer-55.ivf",
+  "av1-1-b10-00-quantizer-56.ivf", "av1-1-b10-00-quantizer-57.ivf",
+  "av1-1-b10-00-quantizer-58.ivf", "av1-1-b10-00-quantizer-59.ivf",
+  "av1-1-b10-00-quantizer-60.ivf", "av1-1-b10-00-quantizer-61.ivf",
+  "av1-1-b10-00-quantizer-62.ivf", "av1-1-b10-00-quantizer-63.ivf",
+  "av1-1-b8-01-size-16x16.ivf",    "av1-1-b8-01-size-16x18.ivf",
+  "av1-1-b8-01-size-16x32.ivf",    "av1-1-b8-01-size-16x34.ivf",
+  "av1-1-b8-01-size-16x64.ivf",    "av1-1-b8-01-size-16x66.ivf",
+  "av1-1-b8-01-size-18x16.ivf",    "av1-1-b8-01-size-18x18.ivf",
+  "av1-1-b8-01-size-18x32.ivf",    "av1-1-b8-01-size-18x34.ivf",
+  "av1-1-b8-01-size-18x64.ivf",    "av1-1-b8-01-size-18x66.ivf",
+  "av1-1-b8-01-size-196x196.ivf",  "av1-1-b8-01-size-196x198.ivf",
+  "av1-1-b8-01-size-196x200.ivf",  "av1-1-b8-01-size-196x202.ivf",
+  "av1-1-b8-01-size-196x208.ivf",  "av1-1-b8-01-size-196x210.ivf",
+  "av1-1-b8-01-size-196x224.ivf",  "av1-1-b8-01-size-196x226.ivf",
+  "av1-1-b8-01-size-198x196.ivf",  "av1-1-b8-01-size-198x198.ivf",
+  "av1-1-b8-01-size-198x200.ivf",  "av1-1-b8-01-size-198x202.ivf",
+  "av1-1-b8-01-size-198x208.ivf",  "av1-1-b8-01-size-198x210.ivf",
+  "av1-1-b8-01-size-198x224.ivf",  "av1-1-b8-01-size-198x226.ivf",
+  "av1-1-b8-01-size-200x196.ivf",  "av1-1-b8-01-size-200x198.ivf",
+  "av1-1-b8-01-size-200x200.ivf",  "av1-1-b8-01-size-200x202.ivf",
+  "av1-1-b8-01-size-200x208.ivf",  "av1-1-b8-01-size-200x210.ivf",
+  "av1-1-b8-01-size-200x224.ivf",  "av1-1-b8-01-size-200x226.ivf",
+  "av1-1-b8-01-size-202x196.ivf",  "av1-1-b8-01-size-202x198.ivf",
+  "av1-1-b8-01-size-202x200.ivf",  "av1-1-b8-01-size-202x202.ivf",
+  "av1-1-b8-01-size-202x208.ivf",  "av1-1-b8-01-size-202x210.ivf",
+  "av1-1-b8-01-size-202x224.ivf",  "av1-1-b8-01-size-202x226.ivf",
+  "av1-1-b8-01-size-208x196.ivf",  "av1-1-b8-01-size-208x198.ivf",
+  "av1-1-b8-01-size-208x200.ivf",  "av1-1-b8-01-size-208x202.ivf",
+  "av1-1-b8-01-size-208x208.ivf",  "av1-1-b8-01-size-208x210.ivf",
+  "av1-1-b8-01-size-208x224.ivf",  "av1-1-b8-01-size-208x226.ivf",
+  "av1-1-b8-01-size-210x196.ivf",  "av1-1-b8-01-size-210x198.ivf",
+  "av1-1-b8-01-size-210x200.ivf",  "av1-1-b8-01-size-210x202.ivf",
+  "av1-1-b8-01-size-210x208.ivf",  "av1-1-b8-01-size-210x210.ivf",
+  "av1-1-b8-01-size-210x224.ivf",  "av1-1-b8-01-size-210x226.ivf",
+  "av1-1-b8-01-size-224x196.ivf",  "av1-1-b8-01-size-224x198.ivf",
+  "av1-1-b8-01-size-224x200.ivf",  "av1-1-b8-01-size-224x202.ivf",
+  "av1-1-b8-01-size-224x208.ivf",  "av1-1-b8-01-size-224x210.ivf",
+  "av1-1-b8-01-size-224x224.ivf",  "av1-1-b8-01-size-224x226.ivf",
+  "av1-1-b8-01-size-226x196.ivf",  "av1-1-b8-01-size-226x198.ivf",
+  "av1-1-b8-01-size-226x200.ivf",  "av1-1-b8-01-size-226x202.ivf",
+  "av1-1-b8-01-size-226x208.ivf",  "av1-1-b8-01-size-226x210.ivf",
+  "av1-1-b8-01-size-226x224.ivf",  "av1-1-b8-01-size-226x226.ivf",
+  "av1-1-b8-01-size-32x16.ivf",    "av1-1-b8-01-size-32x18.ivf",
+  "av1-1-b8-01-size-32x32.ivf",    "av1-1-b8-01-size-32x34.ivf",
+  "av1-1-b8-01-size-32x64.ivf",    "av1-1-b8-01-size-32x66.ivf",
+  "av1-1-b8-01-size-34x16.ivf",    "av1-1-b8-01-size-34x18.ivf",
+  "av1-1-b8-01-size-34x32.ivf",    "av1-1-b8-01-size-34x34.ivf",
+  "av1-1-b8-01-size-34x64.ivf",    "av1-1-b8-01-size-34x66.ivf",
+  "av1-1-b8-01-size-64x16.ivf",    "av1-1-b8-01-size-64x18.ivf",
+  "av1-1-b8-01-size-64x32.ivf",    "av1-1-b8-01-size-64x34.ivf",
+  "av1-1-b8-01-size-64x64.ivf",    "av1-1-b8-01-size-64x66.ivf",
+  "av1-1-b8-01-size-66x16.ivf",    "av1-1-b8-01-size-66x18.ivf",
+  "av1-1-b8-01-size-66x32.ivf",    "av1-1-b8-01-size-66x34.ivf",
+  "av1-1-b8-01-size-66x64.ivf",    "av1-1-b8-01-size-66x66.ivf",
+  "av1-1-b8-02-allintra.ivf",      "av1-1-b8-03-sizedown.mkv",
+  "av1-1-b8-03-sizeup.mkv"
+};
+const int kNumAV1TestVectors = NELEMENTS(kAV1TestVectors);
+#endif  // CONFIG_AV1_DECODER
+
+}  // namespace libaom_test
diff --git a/third_party/aom/test/test_vectors.h b/third_party/aom/test/test_vectors.h
new file mode 100644
index 0000000000..be37f6e377
--- /dev/null
+++ b/third_party/aom/test/test_vectors.h
@@ -0,0 +1,26 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_TEST_VECTORS_H_
+#define AOM_TEST_TEST_VECTORS_H_
+
+#include "config/aom_config.h"
+
+namespace libaom_test {
+
+#if CONFIG_AV1_DECODER
+extern const int kNumAV1TestVectors;
+extern const char *const kAV1TestVectors[];
+#endif
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_TEST_VECTORS_H_
diff --git a/third_party/aom/test/tile_independence_test.cc b/third_party/aom/test/tile_independence_test.cc
new file mode 100644
index 0000000000..cf534c0c59
--- /dev/null
+++ b/third_party/aom/test/tile_independence_test.cc
@@ -0,0 +1,173 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/codec_factory.h"
+#include "test/encode_test_driver.h"
+#include "test/i420_video_source.h"
+#include "test/util.h"
+#include "test/md5_helper.h"
+#include "aom_mem/aom_mem.h"
+
+namespace {
+class TileIndependenceTest
+    : public ::libaom_test::CodecTestWith3Params<int, int, int>,
+      public ::libaom_test::EncoderTest {
+ protected:
+  TileIndependenceTest()
+      : EncoderTest(GET_PARAM(0)), md5_fw_order_(), md5_inv_order_(),
+        n_tile_cols_(GET_PARAM(1)), n_tile_rows_(GET_PARAM(2)),
+        n_tile_groups_(GET_PARAM(3)) {
+    init_flags_ = AOM_CODEC_USE_PSNR;
+    aom_codec_dec_cfg_t cfg = aom_codec_dec_cfg_t();
+    cfg.w = 704;
+    cfg.h = 576;
+    cfg.threads = 1;
+    cfg.allow_lowbitdepth = 1;
+    fw_dec_ = codec_->CreateDecoder(cfg, 0);
+    inv_dec_ = codec_->CreateDecoder(cfg, 0);
+    inv_dec_->Control(AV1_INVERT_TILE_DECODE_ORDER, 1);
+
+    if (fw_dec_->IsAV1() && inv_dec_->IsAV1()) {
+      fw_dec_->Control(AV1_SET_DECODE_TILE_ROW, -1);
+      fw_dec_->Control(AV1_SET_DECODE_TILE_COL, -1);
+      inv_dec_->Control(AV1_SET_DECODE_TILE_ROW, -1);
+      inv_dec_->Control(AV1_SET_DECODE_TILE_COL, -1);
+    }
+  }
+
+  virtual ~TileIndependenceTest() {
+    delete fw_dec_;
+    delete inv_dec_;
+  }
+
+  virtual void SetUp() {
+    InitializeConfig();
+    SetMode(libaom_test::kTwoPassGood);
+  }
+
+  virtual void PreEncodeFrameHook(libaom_test::VideoSource *video,
+                                  libaom_test::Encoder *encoder) {
+    if (video->frame() == 1) {
+      encoder->Control(AV1E_SET_TILE_COLUMNS, n_tile_cols_);
+      encoder->Control(AV1E_SET_TILE_ROWS, n_tile_rows_);
+      SetCpuUsed(encoder);
+    } else if (video->frame() == 3) {
+      encoder->Control(AV1E_SET_NUM_TG, n_tile_groups_);
+    }
+  }
+
+  virtual void SetCpuUsed(libaom_test::Encoder *encoder) {
+    static const int kCpuUsed = 3;
+    encoder->Control(AOME_SET_CPUUSED, kCpuUsed);
+  }
+
+  void UpdateMD5(::libaom_test::Decoder *dec, const aom_codec_cx_pkt_t *pkt,
+                 ::libaom_test::MD5 *md5) {
+    const aom_codec_err_t res = dec->DecodeFrame(
+        reinterpret_cast<uint8_t *>(pkt->data.frame.buf), pkt->data.frame.sz);
+    if (res != AOM_CODEC_OK) {
+      abort_ = true;
+      ASSERT_EQ(AOM_CODEC_OK, res);
+    }
+    const aom_image_t *img = dec->GetDxData().Next();
+    md5->Add(img);
+  }
+
+  virtual void FramePktHook(const aom_codec_cx_pkt_t *pkt) {
+    UpdateMD5(fw_dec_, pkt, &md5_fw_order_);
+    UpdateMD5(inv_dec_, pkt, &md5_inv_order_);
+  }
+
+  void DoTest() {
+    const aom_rational timebase = { 33333333, 1000000000 };
+    cfg_.g_timebase = timebase;
+    cfg_.rc_target_bitrate = 500;
+    cfg_.g_lag_in_frames = 12;
+    cfg_.rc_end_usage = AOM_VBR;
+
+    libaom_test::I420VideoSource video("hantro_collage_w352h288.yuv", 704, 576,
+                                       timebase.den, timebase.num, 0, 5);
+    ASSERT_NO_FATAL_FAILURE(RunLoop(&video));
+
+    const char *md5_fw_str = md5_fw_order_.Get();
+    const char *md5_inv_str = md5_inv_order_.Get();
+    ASSERT_STREQ(md5_fw_str, md5_inv_str);
+  }
+
+  ::libaom_test::MD5 md5_fw_order_, md5_inv_order_;
+  ::libaom_test::Decoder *fw_dec_, *inv_dec_;
+
+ private:
+  int n_tile_cols_;
+  int n_tile_rows_;
+  int n_tile_groups_;
+};
+
+// run an encode with 2 or 4 tiles, and do the decode both in normal and
+// inverted tile ordering. Ensure that the MD5 of the output in both cases
+// is identical. If so, tiles are considered independent and the test passes.
+TEST_P(TileIndependenceTest, MD5Match) {
+  cfg_.large_scale_tile = 0;
+  fw_dec_->Control(AV1_SET_TILE_MODE, 0);
+  inv_dec_->Control(AV1_SET_TILE_MODE, 0);
+  DoTest();
+}
+
+class TileIndependenceTestLarge : public TileIndependenceTest {
+  virtual void SetCpuUsed(libaom_test::Encoder *encoder) {
+    static const int kCpuUsed = 0;
+    encoder->Control(AOME_SET_CPUUSED, kCpuUsed);
+  }
+};
+
+TEST_P(TileIndependenceTestLarge, MD5Match) {
+  cfg_.large_scale_tile = 0;
+  fw_dec_->Control(AV1_SET_TILE_MODE, 0);
+  inv_dec_->Control(AV1_SET_TILE_MODE, 0);
+  DoTest();
+}
+
+AV1_INSTANTIATE_TEST_CASE(TileIndependenceTest, ::testing::Values(0, 1),
+                          ::testing::Values(0, 1), ::testing::Values(1, 2, 4));
+AV1_INSTANTIATE_TEST_CASE(TileIndependenceTestLarge, ::testing::Values(0, 1),
+                          ::testing::Values(0, 1), ::testing::Values(1, 2, 4));
+
+class TileIndependenceLSTest : public TileIndependenceTest {};
+
+TEST_P(TileIndependenceLSTest, MD5Match) {
+  cfg_.large_scale_tile = 1;
+  fw_dec_->Control(AV1_SET_TILE_MODE, 1);
+  fw_dec_->Control(AV1D_EXT_TILE_DEBUG, 1);
+  inv_dec_->Control(AV1_SET_TILE_MODE, 1);
+  inv_dec_->Control(AV1D_EXT_TILE_DEBUG, 1);
+  DoTest();
+}
+
+class TileIndependenceLSTestLarge : public TileIndependenceTestLarge {};
+
+TEST_P(TileIndependenceLSTestLarge, MD5Match) {
+  cfg_.large_scale_tile = 1;
+  fw_dec_->Control(AV1_SET_TILE_MODE, 1);
+  fw_dec_->Control(AV1D_EXT_TILE_DEBUG, 1);
+  inv_dec_->Control(AV1_SET_TILE_MODE, 1);
+  inv_dec_->Control(AV1D_EXT_TILE_DEBUG, 1);
+  DoTest();
+}
+
+AV1_INSTANTIATE_TEST_CASE(TileIndependenceLSTest, ::testing::Values(6),
+                          ::testing::Values(6), ::testing::Values(1));
+AV1_INSTANTIATE_TEST_CASE(TileIndependenceLSTestLarge, ::testing::Values(6),
+                          ::testing::Values(6), ::testing::Values(1));
+}  // namespace
diff --git a/third_party/aom/test/tools_common.sh b/third_party/aom/test/tools_common.sh
new file mode 100755
index 0000000000..c087106060
--- /dev/null
+++ b/third_party/aom/test/tools_common.sh
@@ -0,0 +1,477 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+##  This file contains shell code shared by test scripts for libaom tools.
+
+# Use $AOM_TEST_TOOLS_COMMON_SH as a pseudo include guard.
+if [ -z "${AOM_TEST_TOOLS_COMMON_SH}" ]; then
+AOM_TEST_TOOLS_COMMON_SH=included
+
+set -e
+devnull='> /dev/null 2>&1'
+AOM_TEST_PREFIX=""
+
+elog() {
+  echo "$@" 1>&2
+}
+
+vlog() {
+  if [ "${AOM_TEST_VERBOSE_OUTPUT}" = "yes" ]; then
+    echo "$@"
+  fi
+}
+
+# Sets $AOM_TOOL_TEST to the name specified by positional parameter one.
+test_begin() {
+  AOM_TOOL_TEST="${1}"
+}
+
+# Clears the AOM_TOOL_TEST variable after confirming that $AOM_TOOL_TEST matches
+# positional parameter one.
+test_end() {
+  if [ "$1" != "${AOM_TOOL_TEST}" ]; then
+    echo "FAIL completed test mismatch!."
+    echo "  completed test: ${1}"
+    echo "  active test: ${AOM_TOOL_TEST}."
+    return 1
+  fi
+  AOM_TOOL_TEST='<unset>'
+}
+
+# Echoes the target configuration being tested.
+test_configuration_target() {
+  aom_config_c="${LIBAOM_CONFIG_PATH}/config/aom_config.c"
+  # Clean up the cfg pointer line from aom_config.c for easier re-use by
+  # someone examining a failure in the example tests.
+  # 1. Run grep on aom_config.c for cfg and limit the results to 1.
+  # 2. Split the line using ' = ' as separator.
+  # 3. Abuse sed to consume the leading " and trailing "; from the assignment
+  #    to the cfg pointer.
+  cmake_config=$(awk -F ' = ' '/cfg/ { print $NF; exit }' "${aom_config_c}" \
+    | sed -e s/\"// -e s/\"\;//)
+  echo cmake generated via command: cmake path/to/aom ${cmake_config}
+}
+
+# Trap function used for failure reports and tool output directory removal.
+# When the contents of $AOM_TOOL_TEST do not match the string '<unset>', reports
+# failure of test stored in $AOM_TOOL_TEST.
+cleanup() {
+  if [ -n "${AOM_TOOL_TEST}" ] && [ "${AOM_TOOL_TEST}" != '<unset>' ]; then
+    echo "FAIL: $AOM_TOOL_TEST"
+  fi
+  if [ "${AOM_TEST_PRESERVE_OUTPUT}" = "yes" ]; then
+    return
+  fi
+  if [ -n "${AOM_TEST_OUTPUT_DIR}" ] && [ -d "${AOM_TEST_OUTPUT_DIR}" ]; then
+    rm -rf "${AOM_TEST_OUTPUT_DIR}"
+  fi
+}
+
+# Echoes the version string assigned to the VERSION_STRING_NOSP variable defined
+# in $LIBAOM_CONFIG_PATH/config/aom_version.h to stdout.
+cmake_version() {
+  aom_version_h="${LIBAOM_CONFIG_PATH}/config/aom_version.h"
+
+  # Find VERSION_STRING_NOSP line, split it with '"' and print the next to last
+  # field to output the version string to stdout.
+  aom_version=$(awk -F \" '/VERSION_STRING_NOSP/ {print $(NF-1)}' \
+    "${aom_version_h}")
+  echo "v${aom_version}"
+}
+
+# Echoes current git version as reported by running 'git describe', or the
+# version used by the cmake build when git is unavailable.
+source_version() {
+  if git --version > /dev/null 2>&1; then
+    (cd "$(dirname "${0}")"
+    git describe)
+  else
+    cmake_version
+  fi
+}
+
+# Echoes warnings to stdout when source version and CMake build generated
+# version are out of sync.
+check_version_strings() {
+  cmake_version=$(cmake_version)
+  source_version=$(source_version)
+
+  if [ "${cmake_version}" != "${source_version}" ]; then
+    echo "Warning: version has changed since last cmake run."
+    vlog "  cmake version: ${cmake_version} version now: ${source_version}"
+  fi
+}
+
+# $1 is the name of an environment variable containing a directory name to
+# test.
+test_env_var_dir() {
+  local dir=$(eval echo "\${$1}")
+  if [ ! -d "${dir}" ]; then
+    elog "'${dir}': No such directory"
+    elog "The $1 environment variable must be set to a valid directory."
+    return 1
+  fi
+}
+
+# This script requires that the LIBAOM_BIN_PATH, LIBAOM_CONFIG_PATH, and
+# LIBAOM_TEST_DATA_PATH variables are in the environment: Confirm that
+# the variables are set and that they all evaluate to directory paths.
+verify_aom_test_environment() {
+  test_env_var_dir "LIBAOM_BIN_PATH" \
+    && test_env_var_dir "LIBAOM_CONFIG_PATH" \
+    && test_env_var_dir "LIBAOM_TEST_DATA_PATH"
+}
+
+# Greps aom_config.h in LIBAOM_CONFIG_PATH for positional parameter one, which
+# should be a LIBAOM preprocessor flag. Echoes yes to stdout when the feature
+# is available.
+aom_config_option_enabled() {
+  aom_config_option="${1}"
+  aom_config_file="${LIBAOM_CONFIG_PATH}/config/aom_config.h"
+  config_line=$(grep "${aom_config_option}" "${aom_config_file}")
+  if echo "${config_line}" | egrep -q '1$'; then
+    echo yes
+  fi
+}
+
+# Echoes yes when output of test_configuration_target() contains win32 or win64.
+is_windows_target() {
+  if test_configuration_target \
+     | grep -q -e win32 -e win64 > /dev/null 2>&1; then
+    echo yes
+  fi
+}
+
+# Echoes path to $1 when it's executable and exists in one of the directories
+# included in $tool_paths, or an empty string. Caller is responsible for testing
+# the string once the function returns.
+aom_tool_path() {
+  local tool_name="$1"
+  local root_path="${LIBAOM_BIN_PATH}"
+  local suffix="${AOM_TEST_EXE_SUFFIX}"
+  local tool_paths="\
+    ${root_path}/${tool_name}${suffix} \
+    ${root_path}/../${tool_name}${suffix} \
+    ${root_path}/tools/${tool_name}${suffix} \
+    ${root_path}/../tools/${tool_name}${suffix}"
+
+  local toolpath=""
+
+  for tool_path in ${tool_paths}; do
+    if [ -x "${tool_path}" ] && [ -f "${tool_path}" ]; then
+      echo "${tool_path}"
+      return 0
+    fi
+  done
+
+  return 1
+}
+
+# Echoes yes to stdout when the file named by positional parameter one exists
+# in LIBAOM_BIN_PATH, and is executable.
+aom_tool_available() {
+  local tool_name="$1"
+  local tool="${LIBAOM_BIN_PATH}/${tool_name}${AOM_TEST_EXE_SUFFIX}"
+  [ -x "${tool}" ] && echo yes
+}
+
+# Echoes yes to stdout when aom_config_option_enabled() reports yes for
+# CONFIG_AV1_DECODER.
+av1_decode_available() {
+  [ "$(aom_config_option_enabled CONFIG_AV1_DECODER)" = "yes" ] && echo yes
+}
+
+# Echoes yes to stdout when aom_config_option_enabled() reports yes for
+# CONFIG_AV1_ENCODER.
+av1_encode_available() {
+  [ "$(aom_config_option_enabled CONFIG_AV1_ENCODER)" = "yes" ] && echo yes
+}
+
+# Echoes "fast" encode params for use with aomenc.
+aomenc_encode_test_fast_params() {
+  echo "--cpu-used=1
+        --limit=${AV1_ENCODE_TEST_FRAME_LIMIT}
+        --lag-in-frames=0
+        --test-decode=fatal"
+}
+
+# Echoes yes to stdout when aom_config_option_enabled() reports yes for
+# CONFIG_WEBM_IO.
+webm_io_available() {
+  [ "$(aom_config_option_enabled CONFIG_WEBM_IO)" = "yes" ] && echo yes
+}
+
+# Filters strings from $1 using the filter specified by $2. Filter behavior
+# depends on the presence of $3. When $3 is present, strings that match the
+# filter are excluded. When $3 is omitted, strings matching the filter are
+# included.
+# The filtered result is echoed to stdout.
+filter_strings() {
+  strings=${1}
+  filter=${2}
+  exclude=${3}
+
+  if [ -n "${exclude}" ]; then
+    # When positional parameter three exists the caller wants to remove strings.
+    # Tell grep to invert matches using the -v argument.
+    exclude='-v'
+  else
+    unset exclude
+  fi
+
+  if [ -n "${filter}" ]; then
+    for s in ${strings}; do
+      if echo "${s}" | egrep -q ${exclude} "${filter}" > /dev/null 2>&1; then
+        filtered_strings="${filtered_strings} ${s}"
+      fi
+    done
+  else
+    filtered_strings="${strings}"
+  fi
+  echo "${filtered_strings}"
+}
+
+# Runs user test functions passed via positional parameters one and two.
+# Functions in positional parameter one are treated as environment verification
+# functions and are run unconditionally. Functions in positional parameter two
+# are run according to the rules specified in aom_test_usage().
+run_tests() {
+  local env_tests="verify_aom_test_environment $1"
+  local tests_to_filter="$2"
+  local test_name="${AOM_TEST_NAME}"
+
+  if [ -z "${test_name}" ]; then
+    test_name="$(basename "${0%.*}")"
+  fi
+
+  if [ "${AOM_TEST_RUN_DISABLED_TESTS}" != "yes" ]; then
+    # Filter out DISABLED tests.
+    tests_to_filter=$(filter_strings "${tests_to_filter}" ^DISABLED exclude)
+  fi
+
+  if [ -n "${AOM_TEST_FILTER}" ]; then
+    # Remove tests not matching the user's filter.
+    tests_to_filter=$(filter_strings "${tests_to_filter}" ${AOM_TEST_FILTER})
+  fi
+
+  # User requested test listing: Dump test names and return.
+  if [ "${AOM_TEST_LIST_TESTS}" = "yes" ]; then
+    for test_name in $tests_to_filter; do
+      echo ${test_name}
+    done
+    return
+  fi
+
+  # Don't bother with the environment tests if everything else was disabled.
+  [ -z "${tests_to_filter}" ] && return
+
+  # Combine environment and actual tests.
+  local tests_to_run="${env_tests} ${tests_to_filter}"
+
+  check_version_strings
+
+  # Run tests.
+  for test in ${tests_to_run}; do
+    test_begin "${test}"
+    vlog "  RUN  ${test}"
+    "${test}"
+    vlog "  PASS ${test}"
+    test_end "${test}"
+  done
+
+  local tested_config="$(test_configuration_target) @ $(source_version)"
+  echo "${test_name}: Done, all tests pass for ${tested_config}."
+}
+
+aom_test_usage() {
+cat << EOF
+  Usage: ${0##*/} [arguments]
+    --bin-path <path to libaom binaries directory>
+    --config-path <path to libaom config directory>
+    --filter <filter>: User test filter. Only tests matching filter are run.
+    --run-disabled-tests: Run disabled tests.
+    --help: Display this message and exit.
+    --test-data-path <path to libaom test data directory>
+    --show-program-output: Shows output from all programs being tested.
+    --prefix: Allows for a user specified prefix to be inserted before all test
+              programs. Grants the ability, for example, to run test programs
+              within valgrind.
+    --list-tests: List all test names and exit without actually running tests.
+    --verbose: Verbose output.
+
+    When the --bin-path option is not specified the script attempts to use
+    \$LIBAOM_BIN_PATH and then the current directory.
+
+    When the --config-path option is not specified the script attempts to use
+    \$LIBAOM_CONFIG_PATH and then the current directory.
+
+    When the -test-data-path option is not specified the script attempts to use
+    \$LIBAOM_TEST_DATA_PATH and then the current directory.
+EOF
+}
+
+# Returns non-zero (failure) when required environment variables are empty
+# strings.
+aom_test_check_environment() {
+  if [ -z "${LIBAOM_BIN_PATH}" ] || \
+     [ -z "${LIBAOM_CONFIG_PATH}" ] || \
+     [ -z "${LIBAOM_TEST_DATA_PATH}" ]; then
+    return 1
+  fi
+}
+
+# Echo aomenc command line parameters allowing use of a raw yuv file as
+# input to aomenc.
+yuv_raw_input() {
+  echo ""${YUV_RAW_INPUT}"
+       --width="${YUV_RAW_INPUT_WIDTH}"
+       --height="${YUV_RAW_INPUT_HEIGHT}""
+}
+
+# Do a small encode for testing decoders.
+encode_yuv_raw_input_av1() {
+  if [ "$(av1_encode_available)" = "yes" ]; then
+    local output="$1"
+    local encoder="$(aom_tool_path aomenc)"
+    shift
+    eval "${encoder}" $(yuv_raw_input) \
+      $(aomenc_encode_test_fast_params) \
+      --output="${output}" \
+      $@ \
+      ${devnull}
+
+    if [ ! -e "${output}" ]; then
+      elog "Output file does not exist."
+      return 1
+    fi
+  fi
+}
+
+# Parse the command line.
+while [ -n "$1" ]; do
+  case "$1" in
+    --bin-path)
+      LIBAOM_BIN_PATH="$2"
+      shift
+      ;;
+    --config-path)
+      LIBAOM_CONFIG_PATH="$2"
+      shift
+      ;;
+    --filter)
+      AOM_TEST_FILTER="$2"
+      shift
+      ;;
+    --run-disabled-tests)
+      AOM_TEST_RUN_DISABLED_TESTS=yes
+      ;;
+    --help)
+      aom_test_usage
+      exit
+      ;;
+    --test-data-path)
+      LIBAOM_TEST_DATA_PATH="$2"
+      shift
+      ;;
+    --prefix)
+      AOM_TEST_PREFIX="$2"
+      shift
+      ;;
+    --verbose)
+      AOM_TEST_VERBOSE_OUTPUT=yes
+      ;;
+    --show-program-output)
+      devnull=
+      ;;
+    --list-tests)
+      AOM_TEST_LIST_TESTS=yes
+      ;;
+    *)
+      aom_test_usage
+      exit 1
+      ;;
+  esac
+  shift
+done
+
+# Handle running the tests from a build directory without arguments when running
+# the tests on *nix/macosx.
+LIBAOM_BIN_PATH="${LIBAOM_BIN_PATH:-.}"
+LIBAOM_CONFIG_PATH="${LIBAOM_CONFIG_PATH:-.}"
+LIBAOM_TEST_DATA_PATH="${LIBAOM_TEST_DATA_PATH:-.}"
+
+# Create a temporary directory for output files, and a trap to clean it up.
+if [ -n "${TMPDIR}" ]; then
+  AOM_TEST_TEMP_ROOT="${TMPDIR}"
+elif [ -n "${TEMPDIR}" ]; then
+  AOM_TEST_TEMP_ROOT="${TEMPDIR}"
+else
+  AOM_TEST_TEMP_ROOT=/tmp
+fi
+
+AOM_TEST_OUTPUT_DIR="${AOM_TEST_OUTPUT_DIR:-${AOM_TEST_TEMP_ROOT}/aom_test_$$}"
+
+if ! mkdir -p "${AOM_TEST_OUTPUT_DIR}" || \
+   [ ! -d "${AOM_TEST_OUTPUT_DIR}" ]; then
+  echo "${0##*/}: Cannot create output directory, giving up."
+  echo "${0##*/}:   AOM_TEST_OUTPUT_DIR=${AOM_TEST_OUTPUT_DIR}"
+  exit 1
+fi
+
+AOM_TEST_PRESERVE_OUTPUT=${AOM_TEST_PRESERVE_OUTPUT:-no}
+
+if [ "$(is_windows_target)" = "yes" ]; then
+  AOM_TEST_EXE_SUFFIX=".exe"
+fi
+
+# Variables shared by tests.
+AV1_ENCODE_CPU_USED=${AV1_ENCODE_CPU_USED:-1}
+AV1_ENCODE_TEST_FRAME_LIMIT=${AV1_ENCODE_TEST_FRAME_LIMIT:-5}
+AV1_IVF_FILE="${AV1_IVF_FILE:-${AOM_TEST_OUTPUT_DIR}/av1.ivf}"
+AV1_OBU_ANNEXB_FILE="${AV1_OBU_ANNEXB_FILE:-${AOM_TEST_OUTPUT_DIR}/av1.annexb.obu}"
+AV1_OBU_SEC5_FILE="${AV1_OBU_SEC5_FILE:-${AOM_TEST_OUTPUT_DIR}/av1.section5.obu}"
+AV1_WEBM_FILE="${AV1_WEBM_FILE:-${AOM_TEST_OUTPUT_DIR}/av1.webm}"
+
+YUV_RAW_INPUT="${LIBAOM_TEST_DATA_PATH}/hantro_collage_w352h288.yuv"
+YUV_RAW_INPUT_WIDTH=352
+YUV_RAW_INPUT_HEIGHT=288
+
+Y4M_NOSQ_PAR_INPUT="${LIBAOM_TEST_DATA_PATH}/park_joy_90p_8_420_a10-1.y4m"
+Y4M_720P_INPUT="${LIBAOM_TEST_DATA_PATH}/niklas_1280_720_30.y4m"
+
+# Setup a trap function to clean up after tests complete.
+trap cleanup EXIT
+
+vlog "$(basename "${0%.*}") test configuration:
+  LIBAOM_BIN_PATH=${LIBAOM_BIN_PATH}
+  LIBAOM_CONFIG_PATH=${LIBAOM_CONFIG_PATH}
+  LIBAOM_TEST_DATA_PATH=${LIBAOM_TEST_DATA_PATH}
+  AOM_TEST_EXE_SUFFIX=${AOM_TEST_EXE_SUFFIX}
+  AOM_TEST_FILTER=${AOM_TEST_FILTER}
+  AOM_TEST_LIST_TESTS=${AOM_TEST_LIST_TESTS}
+  AOM_TEST_OUTPUT_DIR=${AOM_TEST_OUTPUT_DIR}
+  AOM_TEST_PREFIX=${AOM_TEST_PREFIX}
+  AOM_TEST_PRESERVE_OUTPUT=${AOM_TEST_PRESERVE_OUTPUT}
+  AOM_TEST_RUN_DISABLED_TESTS=${AOM_TEST_RUN_DISABLED_TESTS}
+  AOM_TEST_SHOW_PROGRAM_OUTPUT=${AOM_TEST_SHOW_PROGRAM_OUTPUT}
+  AOM_TEST_TEMP_ROOT=${AOM_TEST_TEMP_ROOT}
+  AOM_TEST_VERBOSE_OUTPUT=${AOM_TEST_VERBOSE_OUTPUT}
+  AV1_ENCODE_CPU_USED=${AV1_ENCODE_CPU_USED}
+  AV1_ENCODE_TEST_FRAME_LIMIT=${AV1_ENCODE_TEST_FRAME_LIMIT}
+  AV1_IVF_FILE=${AV1_IVF_FILE}
+  AV1_OBU_ANNEXB_FILE=${AV1_OBU_ANNEXB_FILE}
+  AV1_OBU_SEC5_FILE=${AV1_OBU_SEC5_FILE}
+  AV1_WEBM_FILE=${AV1_WEBM_FILE}
+  YUV_RAW_INPUT=${YUV_RAW_INPUT}
+  YUV_RAW_INPUT_WIDTH=${YUV_RAW_INPUT_WIDTH}
+  YUV_RAW_INPUT_HEIGHT=${YUV_RAW_INPUT_HEIGHT}
+  Y4M_NOSQ_PAR_INPUT=${Y4M_NOSQ_PAR_INPUT}"
+
+fi  # End $AOM_TEST_TOOLS_COMMON_SH pseudo include guard.
diff --git a/third_party/aom/test/transform_test_base.h b/third_party/aom/test/transform_test_base.h
new file mode 100644
index 0000000000..8ebcf5ff7c
--- /dev/null
+++ b/third_party/aom/test/transform_test_base.h
@@ -0,0 +1,342 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_TRANSFORM_TEST_BASE_H_
+#define AOM_TEST_TRANSFORM_TEST_BASE_H_
+
+#include "config/aom_config.h"
+
+#include "aom_mem/aom_mem.h"
+#include "aom/aom_codec.h"
+#include "aom_dsp/txfm_common.h"
+
+namespace libaom_test {
+
+//  Note:
+//   Same constant are defined in av1/common/av1_entropy.h and
+//   av1/common/entropy.h.  Goal is to make this base class
+//   to use for future codec transform testing.  But including
+//   either of them would lead to compiling error when we do
+//   unit test for another codec. Suggest to move the definition
+//   to a aom header file.
+const int kDctMaxValue = 16384;
+
+typedef void (*FhtFunc)(const int16_t *in, tran_low_t *out, int stride,
+                        TxfmParam *txfm_param);
+
+typedef void (*IhtFunc)(const tran_low_t *in, uint8_t *out, int stride,
+                        const TxfmParam *txfm_param);
+
+class TransformTestBase {
+ public:
+  virtual ~TransformTestBase() {}
+
+ protected:
+  virtual void RunFwdTxfm(const int16_t *in, tran_low_t *out, int stride) = 0;
+
+  virtual void RunInvTxfm(const tran_low_t *out, uint8_t *dst, int stride) = 0;
+
+  void RunAccuracyCheck(uint32_t ref_max_error, double ref_avg_error) {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    uint32_t max_error = 0;
+    int64_t total_error = 0;
+    const int count_test_block = 10000;
+
+    int16_t *test_input_block = reinterpret_cast<int16_t *>(
+        aom_memalign(16, sizeof(int16_t) * num_coeffs_));
+    tran_low_t *test_temp_block = reinterpret_cast<tran_low_t *>(
+        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
+    uint8_t *dst = reinterpret_cast<uint8_t *>(
+        aom_memalign(16, sizeof(uint8_t) * num_coeffs_));
+    uint8_t *src = reinterpret_cast<uint8_t *>(
+        aom_memalign(16, sizeof(uint8_t) * num_coeffs_));
+    uint16_t *dst16 = reinterpret_cast<uint16_t *>(
+        aom_memalign(16, sizeof(uint16_t) * num_coeffs_));
+    uint16_t *src16 = reinterpret_cast<uint16_t *>(
+        aom_memalign(16, sizeof(uint16_t) * num_coeffs_));
+
+    for (int i = 0; i < count_test_block; ++i) {
+      // Initialize a test block with input range [-255, 255].
+      for (int j = 0; j < num_coeffs_; ++j) {
+        if (bit_depth_ == AOM_BITS_8) {
+          src[j] = rnd.Rand8();
+          dst[j] = rnd.Rand8();
+          test_input_block[j] = src[j] - dst[j];
+        } else {
+          src16[j] = rnd.Rand16() & mask_;
+          dst16[j] = rnd.Rand16() & mask_;
+          test_input_block[j] = src16[j] - dst16[j];
+        }
+      }
+
+      ASM_REGISTER_STATE_CHECK(
+          RunFwdTxfm(test_input_block, test_temp_block, pitch_));
+      if (bit_depth_ == AOM_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(test_temp_block, dst, pitch_));
+      } else {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(test_temp_block, CONVERT_TO_BYTEPTR(dst16), pitch_));
+      }
+
+      for (int j = 0; j < num_coeffs_; ++j) {
+        const int diff =
+            bit_depth_ == AOM_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+        const uint32_t error = diff * diff;
+        if (max_error < error) max_error = error;
+        total_error += error;
+      }
+    }
+
+    double avg_error = total_error * 1. / count_test_block / num_coeffs_;
+
+    EXPECT_GE(ref_max_error, max_error)
+        << "Error: FHT/IHT has an individual round trip error > "
+        << ref_max_error;
+
+    EXPECT_GE(ref_avg_error, avg_error)
+        << "Error: FHT/IHT has average round trip error > " << ref_avg_error
+        << " per block";
+
+    aom_free(test_input_block);
+    aom_free(test_temp_block);
+    aom_free(dst);
+    aom_free(src);
+    aom_free(dst16);
+    aom_free(src16);
+  }
+
+  void RunCoeffCheck() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 5000;
+
+    // Use a stride value which is not the width of any transform, to catch
+    // cases where the transforms use the stride incorrectly.
+    int stride = 96;
+
+    int16_t *input_block = reinterpret_cast<int16_t *>(
+        aom_memalign(16, sizeof(int16_t) * stride * height_));
+    tran_low_t *output_ref_block = reinterpret_cast<tran_low_t *>(
+        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
+    tran_low_t *output_block = reinterpret_cast<tran_low_t *>(
+        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
+
+    for (int i = 0; i < count_test_block; ++i) {
+      int j, k;
+      for (j = 0; j < height_; ++j) {
+        for (k = 0; k < pitch_; ++k) {
+          int in_idx = j * stride + k;
+          int out_idx = j * pitch_ + k;
+          input_block[in_idx] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
+          if (bit_depth_ == AOM_BITS_8) {
+            output_block[out_idx] = output_ref_block[out_idx] = rnd.Rand8();
+          } else {
+            output_block[out_idx] = output_ref_block[out_idx] =
+                rnd.Rand16() & mask_;
+          }
+        }
+      }
+
+      fwd_txfm_ref(input_block, output_ref_block, stride, &txfm_param_);
+      ASM_REGISTER_STATE_CHECK(RunFwdTxfm(input_block, output_block, stride));
+
+      // The minimum quant value is 4.
+      for (j = 0; j < height_; ++j) {
+        for (k = 0; k < pitch_; ++k) {
+          int out_idx = j * pitch_ + k;
+          ASSERT_EQ(output_block[out_idx], output_ref_block[out_idx])
+              << "Error: not bit-exact result at index: " << out_idx
+              << " at test block: " << i;
+        }
+      }
+    }
+    aom_free(input_block);
+    aom_free(output_ref_block);
+    aom_free(output_block);
+  }
+
+  void RunInvCoeffCheck() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 5000;
+
+    // Use a stride value which is not the width of any transform, to catch
+    // cases where the transforms use the stride incorrectly.
+    int stride = 96;
+
+    int16_t *input_block = reinterpret_cast<int16_t *>(
+        aom_memalign(16, sizeof(int16_t) * num_coeffs_));
+    tran_low_t *trans_block = reinterpret_cast<tran_low_t *>(
+        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
+    uint8_t *output_block = reinterpret_cast<uint8_t *>(
+        aom_memalign(16, sizeof(uint8_t) * stride * height_));
+    uint8_t *output_ref_block = reinterpret_cast<uint8_t *>(
+        aom_memalign(16, sizeof(uint8_t) * stride * height_));
+
+    for (int i = 0; i < count_test_block; ++i) {
+      // Initialize a test block with input range [-mask_, mask_].
+      int j, k;
+      for (j = 0; j < height_; ++j) {
+        for (k = 0; k < pitch_; ++k) {
+          int in_idx = j * pitch_ + k;
+          int out_idx = j * stride + k;
+          input_block[in_idx] = (rnd.Rand16() & mask_) - (rnd.Rand16() & mask_);
+          output_ref_block[out_idx] = rnd.Rand16() & mask_;
+          output_block[out_idx] = output_ref_block[out_idx];
+        }
+      }
+
+      fwd_txfm_ref(input_block, trans_block, pitch_, &txfm_param_);
+
+      inv_txfm_ref(trans_block, output_ref_block, stride, &txfm_param_);
+      ASM_REGISTER_STATE_CHECK(RunInvTxfm(trans_block, output_block, stride));
+
+      for (j = 0; j < height_; ++j) {
+        for (k = 0; k < pitch_; ++k) {
+          int out_idx = j * stride + k;
+          ASSERT_EQ(output_block[out_idx], output_ref_block[out_idx])
+              << "Error: not bit-exact result at index: " << out_idx
+              << " j = " << j << " k = " << k << " at test block: " << i;
+        }
+      }
+    }
+    aom_free(input_block);
+    aom_free(trans_block);
+    aom_free(output_ref_block);
+    aom_free(output_block);
+  }
+
+  void RunMemCheck() {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 5000;
+
+    int16_t *input_extreme_block = reinterpret_cast<int16_t *>(
+        aom_memalign(16, sizeof(int16_t) * num_coeffs_));
+    tran_low_t *output_ref_block = reinterpret_cast<tran_low_t *>(
+        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
+    tran_low_t *output_block = reinterpret_cast<tran_low_t *>(
+        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
+
+    for (int i = 0; i < count_test_block; ++i) {
+      // Initialize a test block with input range [-mask_, mask_].
+      for (int j = 0; j < num_coeffs_; ++j) {
+        input_extreme_block[j] = rnd.Rand8() % 2 ? mask_ : -mask_;
+      }
+      if (i == 0) {
+        for (int j = 0; j < num_coeffs_; ++j) input_extreme_block[j] = mask_;
+      } else if (i == 1) {
+        for (int j = 0; j < num_coeffs_; ++j) input_extreme_block[j] = -mask_;
+      }
+
+      fwd_txfm_ref(input_extreme_block, output_ref_block, pitch_, &txfm_param_);
+      ASM_REGISTER_STATE_CHECK(
+          RunFwdTxfm(input_extreme_block, output_block, pitch_));
+
+      int row_length = FindRowLength();
+      // The minimum quant value is 4.
+      for (int j = 0; j < num_coeffs_; ++j) {
+        ASSERT_EQ(output_block[j], output_ref_block[j])
+            << "Not bit-exact at test index: " << i << ", "
+            << "j = " << j << std::endl;
+        EXPECT_GE(row_length * kDctMaxValue << (bit_depth_ - 8),
+                  abs(output_block[j]))
+            << "Error: NxN FDCT has coefficient larger than N*DCT_MAX_VALUE";
+      }
+    }
+    aom_free(input_extreme_block);
+    aom_free(output_ref_block);
+    aom_free(output_block);
+  }
+
+  void RunInvAccuracyCheck(int limit) {
+    ACMRandom rnd(ACMRandom::DeterministicSeed());
+    const int count_test_block = 1000;
+
+    int16_t *in = reinterpret_cast<int16_t *>(
+        aom_memalign(16, sizeof(int16_t) * num_coeffs_));
+    tran_low_t *coeff = reinterpret_cast<tran_low_t *>(
+        aom_memalign(16, sizeof(tran_low_t) * num_coeffs_));
+    uint8_t *dst = reinterpret_cast<uint8_t *>(
+        aom_memalign(16, sizeof(uint8_t) * num_coeffs_));
+    uint8_t *src = reinterpret_cast<uint8_t *>(
+        aom_memalign(16, sizeof(uint8_t) * num_coeffs_));
+
+    uint16_t *dst16 = reinterpret_cast<uint16_t *>(
+        aom_memalign(16, sizeof(uint16_t) * num_coeffs_));
+    uint16_t *src16 = reinterpret_cast<uint16_t *>(
+        aom_memalign(16, sizeof(uint16_t) * num_coeffs_));
+
+    for (int i = 0; i < count_test_block; ++i) {
+      // Initialize a test block with input range [-mask_, mask_].
+      for (int j = 0; j < num_coeffs_; ++j) {
+        if (bit_depth_ == AOM_BITS_8) {
+          src[j] = rnd.Rand8();
+          dst[j] = rnd.Rand8();
+          in[j] = src[j] - dst[j];
+        } else {
+          src16[j] = rnd.Rand16() & mask_;
+          dst16[j] = rnd.Rand16() & mask_;
+          in[j] = src16[j] - dst16[j];
+        }
+      }
+
+      fwd_txfm_ref(in, coeff, pitch_, &txfm_param_);
+
+      if (bit_depth_ == AOM_BITS_8) {
+        ASM_REGISTER_STATE_CHECK(RunInvTxfm(coeff, dst, pitch_));
+      } else {
+        ASM_REGISTER_STATE_CHECK(
+            RunInvTxfm(coeff, CONVERT_TO_BYTEPTR(dst16), pitch_));
+      }
+
+      for (int j = 0; j < num_coeffs_; ++j) {
+        const int diff =
+            bit_depth_ == AOM_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
+        const uint32_t error = diff * diff;
+        ASSERT_GE(static_cast<uint32_t>(limit), error)
+            << "Error: 4x4 IDCT has error " << error << " at index " << j;
+      }
+    }
+    aom_free(in);
+    aom_free(coeff);
+    aom_free(dst);
+    aom_free(src);
+    aom_free(src16);
+    aom_free(dst16);
+  }
+
+  int pitch_;
+  int height_;
+  FhtFunc fwd_txfm_ref;
+  IhtFunc inv_txfm_ref;
+  aom_bit_depth_t bit_depth_;
+  int mask_;
+  int num_coeffs_;
+  TxfmParam txfm_param_;
+
+ private:
+  //  Assume transform size is 4x4, 8x8, 16x16,...
+  int FindRowLength() const {
+    int row = 4;
+    if (16 == num_coeffs_) {
+      row = 4;
+    } else if (64 == num_coeffs_) {
+      row = 8;
+    } else if (256 == num_coeffs_) {
+      row = 16;
+    } else if (1024 == num_coeffs_) {
+      row = 32;
+    }
+    return row;
+  }
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_TRANSFORM_TEST_BASE_H_
diff --git a/third_party/aom/test/twopass_encoder.sh b/third_party/aom/test/twopass_encoder.sh
new file mode 100755
index 0000000000..cca44ced8a
--- /dev/null
+++ b/third_party/aom/test/twopass_encoder.sh
@@ -0,0 +1,54 @@
+#!/bin/sh
+## Copyright (c) 2016, Alliance for Open Media. All rights reserved
+##
+## This source code is subject to the terms of the BSD 2 Clause License and
+## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+## was not distributed with this source code in the LICENSE file, you can
+## obtain it at www.aomedia.org/license/software. If the Alliance for Open
+## Media Patent License 1.0 was not distributed with this source code in the
+## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+##
+## This file tests the libaom twopass_encoder example. To add new tests to this
+## file, do the following:
+##   1. Write a shell function (this is your test).
+##   2. Add the function to twopass_encoder_tests (on a new line).
+##
+. $(dirname $0)/tools_common.sh
+
+# Environment check: $YUV_RAW_INPUT is required.
+twopass_encoder_verify_environment() {
+  if [ ! -e "${YUV_RAW_INPUT}" ]; then
+    echo "Libaom test data must exist in LIBAOM_TEST_DATA_PATH."
+    return 1
+  fi
+}
+
+# Runs twopass_encoder using the codec specified by $1 with a frame limit of
+# 100.
+twopass_encoder() {
+  local encoder="$(aom_tool_path twopass_encoder)"
+  local codec="$1"
+  local output_file="${AOM_TEST_OUTPUT_DIR}/twopass_encoder_${codec}.ivf"
+  local limit=7
+
+  if [ ! -x "${encoder}" ]; then
+    elog "${encoder} does not exist or is not executable."
+    return 1
+  fi
+
+  eval "${AOM_TEST_PREFIX}" "${encoder}" "${codec}" "${YUV_RAW_INPUT_WIDTH}" \
+      "${YUV_RAW_INPUT_HEIGHT}" "${YUV_RAW_INPUT}" "${output_file}" "${limit}" \
+      ${devnull}
+
+  [ -e "${output_file}" ] || return 1
+}
+
+twopass_encoder_av1() {
+  if [ "$(av1_encode_available)" = "yes" ]; then
+    twopass_encoder av1 || return 1
+  fi
+}
+
+twopass_encoder_tests="twopass_encoder_av1"
+
+run_tests twopass_encoder_verify_environment "${twopass_encoder_tests}"
diff --git a/third_party/aom/test/util.h b/third_party/aom/test/util.h
new file mode 100644
index 0000000000..c3f4e44421
--- /dev/null
+++ b/third_party/aom/test/util.h
@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_UTIL_H_
+#define AOM_TEST_UTIL_H_
+
+#include <stdio.h>
+#include <math.h>
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "aom/aom_integer.h"
+#include "aom/aom_image.h"
+#include "aom_ports/aom_timer.h"
+
+// Macros
+#define GET_PARAM(k) ::testing::get<k>(GetParam())
+
+inline double compute_psnr(const aom_image_t *img1, const aom_image_t *img2) {
+  assert((img1->fmt == img2->fmt) && (img1->d_w == img2->d_w) &&
+         (img1->d_h == img2->d_h));
+
+  const unsigned int width_y = img1->d_w;
+  const unsigned int height_y = img1->d_h;
+  unsigned int i, j;
+
+  int64_t sqrerr = 0;
+  for (i = 0; i < height_y; ++i)
+    for (j = 0; j < width_y; ++j) {
+      int64_t d = img1->planes[AOM_PLANE_Y][i * img1->stride[AOM_PLANE_Y] + j] -
+                  img2->planes[AOM_PLANE_Y][i * img2->stride[AOM_PLANE_Y] + j];
+      sqrerr += d * d;
+    }
+  double mse = static_cast<double>(sqrerr) / (width_y * height_y);
+  double psnr = 100.0;
+  if (mse > 0.0) {
+    psnr = 10 * log10(255.0 * 255.0 / mse);
+  }
+  return psnr;
+}
+
+static INLINE double get_time_mark(aom_usec_timer *t) {
+  aom_usec_timer_mark(t);
+  return static_cast<double>(aom_usec_timer_elapsed(t));
+}
+
+#endif  // AOM_TEST_UTIL_H_
diff --git a/third_party/aom/test/variance_test.cc b/third_party/aom/test/variance_test.cc
new file mode 100644
index 0000000000..0df314b0fd
--- /dev/null
+++ b/third_party/aom/test/variance_test.cc
@@ -0,0 +1,2064 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <cstdlib>
+#include <new>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "test/acm_random.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+#include "aom/aom_codec.h"
+#include "aom/aom_integer.h"
+#include "aom_mem/aom_mem.h"
+#include "aom_ports/aom_timer.h"
+#include "aom_ports/mem.h"
+
+namespace {
+
+typedef unsigned int (*VarianceMxNFunc)(const uint8_t *a, int a_stride,
+                                        const uint8_t *b, int b_stride,
+                                        unsigned int *sse);
+typedef unsigned int (*SubpixVarMxNFunc)(const uint8_t *a, int a_stride,
+                                         int xoffset, int yoffset,
+                                         const uint8_t *b, int b_stride,
+                                         unsigned int *sse);
+typedef unsigned int (*SubpixAvgVarMxNFunc)(const uint8_t *a, int a_stride,
+                                            int xoffset, int yoffset,
+                                            const uint8_t *b, int b_stride,
+                                            uint32_t *sse,
+                                            const uint8_t *second_pred);
+typedef unsigned int (*Get4x4SseFunc)(const uint8_t *a, int a_stride,
+                                      const uint8_t *b, int b_stride);
+typedef unsigned int (*SumOfSquaresFunction)(const int16_t *src);
+typedef unsigned int (*JntSubpixAvgVarMxNFunc)(
+    const uint8_t *a, int a_stride, int xoffset, int yoffset, const uint8_t *b,
+    int b_stride, uint32_t *sse, const uint8_t *second_pred,
+    const JNT_COMP_PARAMS *jcp_param);
+typedef uint32_t (*ObmcSubpelVarFunc)(const uint8_t *pre, int pre_stride,
+                                      int xoffset, int yoffset,
+                                      const int32_t *wsrc, const int32_t *mask,
+                                      unsigned int *sse);
+
+using libaom_test::ACMRandom;
+
+// Truncate high bit depth results by downshifting (with rounding) by:
+// 2 * (bit_depth - 8) for sse
+// (bit_depth - 8) for se
+static void RoundHighBitDepth(int bit_depth, int64_t *se, uint64_t *sse) {
+  switch (bit_depth) {
+    case AOM_BITS_12:
+      *sse = (*sse + 128) >> 8;
+      *se = (*se + 8) >> 4;
+      break;
+    case AOM_BITS_10:
+      *sse = (*sse + 8) >> 4;
+      *se = (*se + 2) >> 2;
+      break;
+    case AOM_BITS_8:
+    default: break;
+  }
+}
+
+static unsigned int mb_ss_ref(const int16_t *src) {
+  unsigned int res = 0;
+  for (int i = 0; i < 256; ++i) {
+    res += src[i] * src[i];
+  }
+  return res;
+}
+
+/* Note:
+ *  Our codebase calculates the "diff" value in the variance algorithm by
+ *  (src - ref).
+ */
+static uint32_t variance_ref(const uint8_t *src, const uint8_t *ref, int l2w,
+                             int l2h, int src_stride, int ref_stride,
+                             uint32_t *sse_ptr, bool use_high_bit_depth_,
+                             aom_bit_depth_t bit_depth) {
+  int64_t se = 0;
+  uint64_t sse = 0;
+  const int w = 1 << l2w;
+  const int h = 1 << l2h;
+  for (int y = 0; y < h; y++) {
+    for (int x = 0; x < w; x++) {
+      int diff;
+      if (!use_high_bit_depth_) {
+        diff = src[y * src_stride + x] - ref[y * ref_stride + x];
+        se += diff;
+        sse += diff * diff;
+      } else {
+        diff = CONVERT_TO_SHORTPTR(src)[y * src_stride + x] -
+               CONVERT_TO_SHORTPTR(ref)[y * ref_stride + x];
+        se += diff;
+        sse += diff * diff;
+      }
+    }
+  }
+  RoundHighBitDepth(bit_depth, &se, &sse);
+  *sse_ptr = static_cast<uint32_t>(sse);
+  return static_cast<uint32_t>(sse - ((se * se) >> (l2w + l2h)));
+}
+
+/* The subpel reference functions differ from the codec version in one aspect:
+ * they calculate the bilinear factors directly instead of using a lookup table
+ * and therefore upshift xoff and yoff by 1. Only every other calculated value
+ * is used so the codec version shrinks the table to save space and maintain
+ * compatibility with vp8.
+ */
+static uint32_t subpel_variance_ref(const uint8_t *ref, const uint8_t *src,
+                                    int l2w, int l2h, int xoff, int yoff,
+                                    uint32_t *sse_ptr, bool use_high_bit_depth_,
+                                    aom_bit_depth_t bit_depth) {
+  int64_t se = 0;
+  uint64_t sse = 0;
+  const int w = 1 << l2w;
+  const int h = 1 << l2h;
+
+  xoff <<= 1;
+  yoff <<= 1;
+
+  for (int y = 0; y < h; y++) {
+    for (int x = 0; x < w; x++) {
+      // Bilinear interpolation at a 16th pel step.
+      if (!use_high_bit_depth_) {
+        const int a1 = ref[(w + 1) * (y + 0) + x + 0];
+        const int a2 = ref[(w + 1) * (y + 0) + x + 1];
+        const int b1 = ref[(w + 1) * (y + 1) + x + 0];
+        const int b2 = ref[(w + 1) * (y + 1) + x + 1];
+        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
+        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
+        const int r = a + (((b - a) * yoff + 8) >> 4);
+        const int diff = r - src[w * y + x];
+        se += diff;
+        sse += diff * diff;
+      } else {
+        uint16_t *ref16 = CONVERT_TO_SHORTPTR(ref);
+        uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
+        const int a1 = ref16[(w + 1) * (y + 0) + x + 0];
+        const int a2 = ref16[(w + 1) * (y + 0) + x + 1];
+        const int b1 = ref16[(w + 1) * (y + 1) + x + 0];
+        const int b2 = ref16[(w + 1) * (y + 1) + x + 1];
+        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
+        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
+        const int r = a + (((b - a) * yoff + 8) >> 4);
+        const int diff = r - src16[w * y + x];
+        se += diff;
+        sse += diff * diff;
+      }
+    }
+  }
+  RoundHighBitDepth(bit_depth, &se, &sse);
+  *sse_ptr = static_cast<uint32_t>(sse);
+  return static_cast<uint32_t>(sse - ((se * se) >> (l2w + l2h)));
+}
+
+static uint32_t subpel_avg_variance_ref(const uint8_t *ref, const uint8_t *src,
+                                        const uint8_t *second_pred, int l2w,
+                                        int l2h, int xoff, int yoff,
+                                        uint32_t *sse_ptr,
+                                        bool use_high_bit_depth,
+                                        aom_bit_depth_t bit_depth) {
+  int64_t se = 0;
+  uint64_t sse = 0;
+  const int w = 1 << l2w;
+  const int h = 1 << l2h;
+
+  xoff <<= 1;
+  yoff <<= 1;
+
+  for (int y = 0; y < h; y++) {
+    for (int x = 0; x < w; x++) {
+      // bilinear interpolation at a 16th pel step
+      if (!use_high_bit_depth) {
+        const int a1 = ref[(w + 1) * (y + 0) + x + 0];
+        const int a2 = ref[(w + 1) * (y + 0) + x + 1];
+        const int b1 = ref[(w + 1) * (y + 1) + x + 0];
+        const int b2 = ref[(w + 1) * (y + 1) + x + 1];
+        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
+        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
+        const int r = a + (((b - a) * yoff + 8) >> 4);
+        const int diff =
+            ((r + second_pred[w * y + x] + 1) >> 1) - src[w * y + x];
+        se += diff;
+        sse += diff * diff;
+      } else {
+        const uint16_t *ref16 = CONVERT_TO_SHORTPTR(ref);
+        const uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
+        const uint16_t *sec16 = CONVERT_TO_SHORTPTR(second_pred);
+        const int a1 = ref16[(w + 1) * (y + 0) + x + 0];
+        const int a2 = ref16[(w + 1) * (y + 0) + x + 1];
+        const int b1 = ref16[(w + 1) * (y + 1) + x + 0];
+        const int b2 = ref16[(w + 1) * (y + 1) + x + 1];
+        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
+        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
+        const int r = a + (((b - a) * yoff + 8) >> 4);
+        const int diff = ((r + sec16[w * y + x] + 1) >> 1) - src16[w * y + x];
+        se += diff;
+        sse += diff * diff;
+      }
+    }
+  }
+  RoundHighBitDepth(bit_depth, &se, &sse);
+  *sse_ptr = static_cast<uint32_t>(sse);
+  return static_cast<uint32_t>(sse - ((se * se) >> (l2w + l2h)));
+}
+
+static uint32_t jnt_subpel_avg_variance_ref(
+    const uint8_t *ref, const uint8_t *src, const uint8_t *second_pred, int l2w,
+    int l2h, int xoff, int yoff, uint32_t *sse_ptr, bool use_high_bit_depth,
+    aom_bit_depth_t bit_depth, JNT_COMP_PARAMS *jcp_param) {
+  int64_t se = 0;
+  uint64_t sse = 0;
+  const int w = 1 << l2w;
+  const int h = 1 << l2h;
+
+  xoff <<= 1;
+  yoff <<= 1;
+
+  for (int y = 0; y < h; y++) {
+    for (int x = 0; x < w; x++) {
+      // bilinear interpolation at a 16th pel step
+      if (!use_high_bit_depth) {
+        const int a1 = ref[(w + 0) * (y + 0) + x + 0];
+        const int a2 = ref[(w + 0) * (y + 0) + x + 1];
+        const int b1 = ref[(w + 0) * (y + 1) + x + 0];
+        const int b2 = ref[(w + 0) * (y + 1) + x + 1];
+        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
+        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
+        const int r = a + (((b - a) * yoff + 8) >> 4);
+        const int avg = ROUND_POWER_OF_TWO(
+            r * jcp_param->fwd_offset +
+                second_pred[w * y + x] * jcp_param->bck_offset,
+            DIST_PRECISION_BITS);
+        const int diff = avg - src[w * y + x];
+
+        se += diff;
+        sse += diff * diff;
+      } else {
+        const uint16_t *ref16 = CONVERT_TO_SHORTPTR(ref);
+        const uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
+        const uint16_t *sec16 = CONVERT_TO_SHORTPTR(second_pred);
+        const int a1 = ref16[(w + 0) * (y + 0) + x + 0];
+        const int a2 = ref16[(w + 0) * (y + 0) + x + 1];
+        const int b1 = ref16[(w + 0) * (y + 1) + x + 0];
+        const int b2 = ref16[(w + 0) * (y + 1) + x + 1];
+        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
+        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
+        const int r = a + (((b - a) * yoff + 8) >> 4);
+        const int avg =
+            ROUND_POWER_OF_TWO(r * jcp_param->fwd_offset +
+                                   sec16[w * y + x] * jcp_param->bck_offset,
+                               DIST_PRECISION_BITS);
+        const int diff = avg - src16[w * y + x];
+
+        se += diff;
+        sse += diff * diff;
+      }
+    }
+  }
+  RoundHighBitDepth(bit_depth, &se, &sse);
+  *sse_ptr = static_cast<uint32_t>(sse);
+  return static_cast<uint32_t>(sse - ((se * se) >> (l2w + l2h)));
+}
+
+static uint32_t obmc_subpel_variance_ref(const uint8_t *pre, int l2w, int l2h,
+                                         int xoff, int yoff,
+                                         const int32_t *wsrc,
+                                         const int32_t *mask, uint32_t *sse_ptr,
+                                         bool use_high_bit_depth_,
+                                         aom_bit_depth_t bit_depth) {
+  int64_t se = 0;
+  uint64_t sse = 0;
+  const int w = 1 << l2w;
+  const int h = 1 << l2h;
+
+  xoff <<= 1;
+  yoff <<= 1;
+
+  for (int y = 0; y < h; y++) {
+    for (int x = 0; x < w; x++) {
+      // Bilinear interpolation at a 16th pel step.
+      if (!use_high_bit_depth_) {
+        const int a1 = pre[(w + 1) * (y + 0) + x + 0];
+        const int a2 = pre[(w + 1) * (y + 0) + x + 1];
+        const int b1 = pre[(w + 1) * (y + 1) + x + 0];
+        const int b2 = pre[(w + 1) * (y + 1) + x + 1];
+        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
+        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
+        const int r = a + (((b - a) * yoff + 8) >> 4);
+        const int diff = ROUND_POWER_OF_TWO_SIGNED(
+            wsrc[w * y + x] - r * mask[w * y + x], 12);
+        se += diff;
+        sse += diff * diff;
+      } else {
+        uint16_t *pre16 = CONVERT_TO_SHORTPTR(pre);
+        const int a1 = pre16[(w + 1) * (y + 0) + x + 0];
+        const int a2 = pre16[(w + 1) * (y + 0) + x + 1];
+        const int b1 = pre16[(w + 1) * (y + 1) + x + 0];
+        const int b2 = pre16[(w + 1) * (y + 1) + x + 1];
+        const int a = a1 + (((a2 - a1) * xoff + 8) >> 4);
+        const int b = b1 + (((b2 - b1) * xoff + 8) >> 4);
+        const int r = a + (((b - a) * yoff + 8) >> 4);
+        const int diff = ROUND_POWER_OF_TWO_SIGNED(
+            wsrc[w * y + x] - r * mask[w * y + x], 12);
+        se += diff;
+        sse += diff * diff;
+      }
+    }
+  }
+  RoundHighBitDepth(bit_depth, &se, &sse);
+  *sse_ptr = static_cast<uint32_t>(sse);
+  return static_cast<uint32_t>(sse - ((se * se) >> (l2w + l2h)));
+}
+
+////////////////////////////////////////////////////////////////////////////////
+
+class SumOfSquaresTest : public ::testing::TestWithParam<SumOfSquaresFunction> {
+ public:
+  SumOfSquaresTest() : func_(GetParam()) {}
+
+  virtual ~SumOfSquaresTest() { libaom_test::ClearSystemState(); }
+
+ protected:
+  void ConstTest();
+  void RefTest();
+
+  SumOfSquaresFunction func_;
+  ACMRandom rnd_;
+};
+
+void SumOfSquaresTest::ConstTest() {
+  int16_t mem[256];
+  unsigned int res;
+  for (int v = 0; v < 256; ++v) {
+    for (int i = 0; i < 256; ++i) {
+      mem[i] = v;
+    }
+    ASM_REGISTER_STATE_CHECK(res = func_(mem));
+    EXPECT_EQ(256u * (v * v), res);
+  }
+}
+
+void SumOfSquaresTest::RefTest() {
+  int16_t mem[256];
+  for (int i = 0; i < 100; ++i) {
+    for (int j = 0; j < 256; ++j) {
+      mem[j] = rnd_.Rand8() - rnd_.Rand8();
+    }
+
+    const unsigned int expected = mb_ss_ref(mem);
+    unsigned int res;
+    ASM_REGISTER_STATE_CHECK(res = func_(mem));
+    EXPECT_EQ(expected, res);
+  }
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// Encapsulating struct to store the function to test along with
+// some testing context.
+// Can be used for MSE, SSE, Variance, etc.
+
+template <typename Func>
+struct TestParams {
+  TestParams(int log2w = 0, int log2h = 0, Func function = NULL,
+             int bit_depth_value = 0)
+      : log2width(log2w), log2height(log2h), func(function) {
+    use_high_bit_depth = (bit_depth_value > 0);
+    if (use_high_bit_depth) {
+      bit_depth = static_cast<aom_bit_depth_t>(bit_depth_value);
+    } else {
+      bit_depth = AOM_BITS_8;
+    }
+    width = 1 << log2width;
+    height = 1 << log2height;
+    block_size = width * height;
+    mask = (1u << bit_depth) - 1;
+  }
+
+  int log2width, log2height;
+  int width, height;
+  int block_size;
+  Func func;
+  aom_bit_depth_t bit_depth;
+  bool use_high_bit_depth;
+  uint32_t mask;
+};
+
+template <typename Func>
+std::ostream &operator<<(std::ostream &os, const TestParams<Func> &p) {
+  return os << "width/height:" << p.width << "/" << p.height
+            << " function:" << reinterpret_cast<const void *>(p.func)
+            << " bit-depth:" << p.bit_depth;
+}
+
+// Main class for testing a function type
+template <typename FunctionType>
+class MainTestClass
+    : public ::testing::TestWithParam<TestParams<FunctionType> > {
+ public:
+  virtual void SetUp() {
+    params_ = this->GetParam();
+
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+    const size_t unit =
+        use_high_bit_depth() ? sizeof(uint16_t) : sizeof(uint8_t);
+    src_ = reinterpret_cast<uint8_t *>(aom_memalign(16, block_size() * unit));
+    ref_ = new uint8_t[block_size() * unit];
+    ASSERT_TRUE(src_ != NULL);
+    ASSERT_TRUE(ref_ != NULL);
+    if (use_high_bit_depth()) {
+      // TODO(skal): remove!
+      src_ = CONVERT_TO_BYTEPTR(src_);
+      ref_ = CONVERT_TO_BYTEPTR(ref_);
+    }
+  }
+
+  virtual void TearDown() {
+    if (use_high_bit_depth()) {
+      // TODO(skal): remove!
+      src_ = reinterpret_cast<uint8_t *>(CONVERT_TO_SHORTPTR(src_));
+      ref_ = reinterpret_cast<uint8_t *>(CONVERT_TO_SHORTPTR(ref_));
+    }
+
+    aom_free(src_);
+    delete[] ref_;
+    src_ = NULL;
+    ref_ = NULL;
+    libaom_test::ClearSystemState();
+  }
+
+ protected:
+  // We could sub-class MainTestClass into dedicated class for Variance
+  // and MSE/SSE, but it involves a lot of 'this->xxx' dereferencing
+  // to access top class fields xxx. That's cumbersome, so for now we'll just
+  // implement the testing methods here:
+
+  // Variance tests
+  void ZeroTest();
+  void RefTest();
+  void RefStrideTest();
+  void OneQuarterTest();
+  void SpeedTest();
+
+  // MSE/SSE tests
+  void RefTestMse();
+  void RefTestSse();
+  void MaxTestMse();
+  void MaxTestSse();
+
+ protected:
+  ACMRandom rnd_;
+  uint8_t *src_;
+  uint8_t *ref_;
+  TestParams<FunctionType> params_;
+
+  // some relay helpers
+  bool use_high_bit_depth() const { return params_.use_high_bit_depth; }
+  int byte_shift() const { return params_.bit_depth - 8; }
+  int block_size() const { return params_.block_size; }
+  int width() const { return params_.width; }
+  int height() const { return params_.height; }
+  uint32_t mask() const { return params_.mask; }
+};
+
+////////////////////////////////////////////////////////////////////////////////
+// Tests related to variance.
+
+template <typename VarianceFunctionType>
+void MainTestClass<VarianceFunctionType>::ZeroTest() {
+  for (int i = 0; i <= 255; ++i) {
+    if (!use_high_bit_depth()) {
+      memset(src_, i, block_size());
+    } else {
+      uint16_t *const src16 = CONVERT_TO_SHORTPTR(src_);
+      for (int k = 0; k < block_size(); ++k) src16[k] = i << byte_shift();
+    }
+    for (int j = 0; j <= 255; ++j) {
+      if (!use_high_bit_depth()) {
+        memset(ref_, j, block_size());
+      } else {
+        uint16_t *const ref16 = CONVERT_TO_SHORTPTR(ref_);
+        for (int k = 0; k < block_size(); ++k) ref16[k] = j << byte_shift();
+      }
+      unsigned int sse, var;
+      ASM_REGISTER_STATE_CHECK(
+          var = params_.func(src_, width(), ref_, width(), &sse));
+      EXPECT_EQ(0u, var) << "src values: " << i << " ref values: " << j;
+    }
+  }
+}
+
+template <typename VarianceFunctionType>
+void MainTestClass<VarianceFunctionType>::RefTest() {
+  for (int i = 0; i < 10; ++i) {
+    for (int j = 0; j < block_size(); j++) {
+      if (!use_high_bit_depth()) {
+        src_[j] = rnd_.Rand8();
+        ref_[j] = rnd_.Rand8();
+      } else {
+        CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask();
+        CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() & mask();
+      }
+    }
+    unsigned int sse1, sse2, var1, var2;
+    const int stride = width();
+    ASM_REGISTER_STATE_CHECK(
+        var1 = params_.func(src_, stride, ref_, stride, &sse1));
+    var2 =
+        variance_ref(src_, ref_, params_.log2width, params_.log2height, stride,
+                     stride, &sse2, use_high_bit_depth(), params_.bit_depth);
+    EXPECT_EQ(sse1, sse2) << "Error at test index: " << i;
+    EXPECT_EQ(var1, var2) << "Error at test index: " << i;
+  }
+}
+
+template <typename VarianceFunctionType>
+void MainTestClass<VarianceFunctionType>::RefStrideTest() {
+  for (int i = 0; i < 10; ++i) {
+    const int ref_stride = (i & 1) * width();
+    const int src_stride = ((i >> 1) & 1) * width();
+    for (int j = 0; j < block_size(); j++) {
+      const int ref_ind = (j / width()) * ref_stride + j % width();
+      const int src_ind = (j / width()) * src_stride + j % width();
+      if (!use_high_bit_depth()) {
+        src_[src_ind] = rnd_.Rand8();
+        ref_[ref_ind] = rnd_.Rand8();
+      } else {
+        CONVERT_TO_SHORTPTR(src_)[src_ind] = rnd_.Rand16() & mask();
+        CONVERT_TO_SHORTPTR(ref_)[ref_ind] = rnd_.Rand16() & mask();
+      }
+    }
+    unsigned int sse1, sse2;
+    unsigned int var1, var2;
+
+    ASM_REGISTER_STATE_CHECK(
+        var1 = params_.func(src_, src_stride, ref_, ref_stride, &sse1));
+    var2 = variance_ref(src_, ref_, params_.log2width, params_.log2height,
+                        src_stride, ref_stride, &sse2, use_high_bit_depth(),
+                        params_.bit_depth);
+    EXPECT_EQ(sse1, sse2) << "Error at test index: " << i;
+    EXPECT_EQ(var1, var2) << "Error at test index: " << i;
+  }
+}
+
+template <typename VarianceFunctionType>
+void MainTestClass<VarianceFunctionType>::OneQuarterTest() {
+  const int half = block_size() / 2;
+  if (!use_high_bit_depth()) {
+    memset(src_, 255, block_size());
+    memset(ref_, 255, half);
+    memset(ref_ + half, 0, half);
+  } else {
+    aom_memset16(CONVERT_TO_SHORTPTR(src_), 255 << byte_shift(), block_size());
+    aom_memset16(CONVERT_TO_SHORTPTR(ref_), 255 << byte_shift(), half);
+    aom_memset16(CONVERT_TO_SHORTPTR(ref_) + half, 0, half);
+  }
+  unsigned int sse, var, expected;
+  ASM_REGISTER_STATE_CHECK(
+      var = params_.func(src_, width(), ref_, width(), &sse));
+  expected = block_size() * 255 * 255 / 4;
+  EXPECT_EQ(expected, var);
+}
+
+template <typename VarianceFunctionType>
+void MainTestClass<VarianceFunctionType>::SpeedTest() {
+  for (int j = 0; j < block_size(); j++) {
+    if (!use_high_bit_depth()) {
+      src_[j] = rnd_.Rand8();
+      ref_[j] = rnd_.Rand8();
+    } else {
+      CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask();
+      CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() & mask();
+    }
+  }
+  unsigned int sse;
+  const int stride = width();
+  int run_time = 1000000000 / block_size();
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < run_time; ++i) {
+    params_.func(src_, stride, ref_, stride, &sse);
+  }
+
+  aom_usec_timer_mark(&timer);
+  const double elapsed_time =
+      static_cast<double>(aom_usec_timer_elapsed(&timer));
+  printf("Variance %dx%d : %7.2fns\n", width(), height(), elapsed_time);
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// Tests related to MSE / SSE.
+
+template <typename FunctionType>
+void MainTestClass<FunctionType>::RefTestMse() {
+  for (int i = 0; i < 10; ++i) {
+    for (int j = 0; j < block_size(); ++j) {
+      src_[j] = rnd_.Rand8();
+      ref_[j] = rnd_.Rand8();
+    }
+    unsigned int sse1, sse2;
+    const int stride = width();
+    ASM_REGISTER_STATE_CHECK(params_.func(src_, stride, ref_, stride, &sse1));
+    variance_ref(src_, ref_, params_.log2width, params_.log2height, stride,
+                 stride, &sse2, false, AOM_BITS_8);
+    EXPECT_EQ(sse1, sse2);
+  }
+}
+
+template <typename FunctionType>
+void MainTestClass<FunctionType>::RefTestSse() {
+  for (int i = 0; i < 10; ++i) {
+    for (int j = 0; j < block_size(); ++j) {
+      src_[j] = rnd_.Rand8();
+      ref_[j] = rnd_.Rand8();
+    }
+    unsigned int sse2;
+    unsigned int var1;
+    const int stride = width();
+    ASM_REGISTER_STATE_CHECK(var1 = params_.func(src_, stride, ref_, stride));
+    variance_ref(src_, ref_, params_.log2width, params_.log2height, stride,
+                 stride, &sse2, false, AOM_BITS_8);
+    EXPECT_EQ(var1, sse2);
+  }
+}
+
+template <typename FunctionType>
+void MainTestClass<FunctionType>::MaxTestMse() {
+  memset(src_, 255, block_size());
+  memset(ref_, 0, block_size());
+  unsigned int sse;
+  ASM_REGISTER_STATE_CHECK(params_.func(src_, width(), ref_, width(), &sse));
+  const unsigned int expected = block_size() * 255 * 255;
+  EXPECT_EQ(expected, sse);
+}
+
+template <typename FunctionType>
+void MainTestClass<FunctionType>::MaxTestSse() {
+  memset(src_, 255, block_size());
+  memset(ref_, 0, block_size());
+  unsigned int var;
+  ASM_REGISTER_STATE_CHECK(var = params_.func(src_, width(), ref_, width()));
+  const unsigned int expected = block_size() * 255 * 255;
+  EXPECT_EQ(expected, var);
+}
+
+////////////////////////////////////////////////////////////////////////////////
+
+using ::testing::get;
+using ::testing::make_tuple;
+using ::testing::tuple;
+
+template <typename FunctionType>
+class SubpelVarianceTest
+    : public ::testing::TestWithParam<TestParams<FunctionType> > {
+ public:
+  virtual void SetUp() {
+    params_ = this->GetParam();
+
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+    if (!use_high_bit_depth()) {
+      src_ = reinterpret_cast<uint8_t *>(aom_memalign(32, block_size()));
+      sec_ = reinterpret_cast<uint8_t *>(aom_memalign(32, block_size()));
+      ref_ = reinterpret_cast<uint8_t *>(
+          aom_memalign(32, block_size() + width() + height() + 1));
+    } else {
+      src_ = CONVERT_TO_BYTEPTR(reinterpret_cast<uint16_t *>(
+          aom_memalign(32, block_size() * sizeof(uint16_t))));
+      sec_ = CONVERT_TO_BYTEPTR(reinterpret_cast<uint16_t *>(
+          aom_memalign(32, block_size() * sizeof(uint16_t))));
+      ref_ = CONVERT_TO_BYTEPTR(aom_memalign(
+          32, (block_size() + width() + height() + 1) * sizeof(uint16_t)));
+    }
+    ASSERT_TRUE(src_ != NULL);
+    ASSERT_TRUE(sec_ != NULL);
+    ASSERT_TRUE(ref_ != NULL);
+  }
+
+  virtual void TearDown() {
+    if (!use_high_bit_depth()) {
+      aom_free(src_);
+      aom_free(ref_);
+      aom_free(sec_);
+    } else {
+      aom_free(CONVERT_TO_SHORTPTR(src_));
+      aom_free(CONVERT_TO_SHORTPTR(ref_));
+      aom_free(CONVERT_TO_SHORTPTR(sec_));
+    }
+    libaom_test::ClearSystemState();
+  }
+
+ protected:
+  void RefTest();
+  void ExtremeRefTest();
+
+  ACMRandom rnd_;
+  uint8_t *src_;
+  uint8_t *ref_;
+  uint8_t *sec_;
+  TestParams<FunctionType> params_;
+  JNT_COMP_PARAMS jcp_param_;
+
+  // some relay helpers
+  bool use_high_bit_depth() const { return params_.use_high_bit_depth; }
+  int byte_shift() const { return params_.bit_depth - 8; }
+  int block_size() const { return params_.block_size; }
+  int width() const { return params_.width; }
+  int height() const { return params_.height; }
+  uint32_t mask() const { return params_.mask; }
+};
+
+template <typename SubpelVarianceFunctionType>
+void SubpelVarianceTest<SubpelVarianceFunctionType>::RefTest() {
+  for (int x = 0; x < 8; ++x) {
+    for (int y = 0; y < 8; ++y) {
+      if (!use_high_bit_depth()) {
+        for (int j = 0; j < block_size(); j++) {
+          src_[j] = rnd_.Rand8();
+        }
+        for (int j = 0; j < block_size() + width() + height() + 1; j++) {
+          ref_[j] = rnd_.Rand8();
+        }
+      } else {
+        for (int j = 0; j < block_size(); j++) {
+          CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask();
+        }
+        for (int j = 0; j < block_size() + width() + height() + 1; j++) {
+          CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() & mask();
+        }
+      }
+      unsigned int sse1, sse2;
+      unsigned int var1;
+      ASM_REGISTER_STATE_CHECK(
+          var1 = params_.func(ref_, width() + 1, x, y, src_, width(), &sse1));
+      const unsigned int var2 = subpel_variance_ref(
+          ref_, src_, params_.log2width, params_.log2height, x, y, &sse2,
+          use_high_bit_depth(), params_.bit_depth);
+      EXPECT_EQ(sse1, sse2) << "at position " << x << ", " << y;
+      EXPECT_EQ(var1, var2) << "at position " << x << ", " << y;
+    }
+  }
+}
+
+template <typename SubpelVarianceFunctionType>
+void SubpelVarianceTest<SubpelVarianceFunctionType>::ExtremeRefTest() {
+  // Compare against reference.
+  // Src: Set the first half of values to 0, the second half to the maximum.
+  // Ref: Set the first half of values to the maximum, the second half to 0.
+  for (int x = 0; x < 8; ++x) {
+    for (int y = 0; y < 8; ++y) {
+      const int half = block_size() / 2;
+      if (!use_high_bit_depth()) {
+        memset(src_, 0, half);
+        memset(src_ + half, 255, half);
+        memset(ref_, 255, half);
+        memset(ref_ + half, 0, half + width() + height() + 1);
+      } else {
+        aom_memset16(CONVERT_TO_SHORTPTR(src_), mask(), half);
+        aom_memset16(CONVERT_TO_SHORTPTR(src_) + half, 0, half);
+        aom_memset16(CONVERT_TO_SHORTPTR(ref_), 0, half);
+        aom_memset16(CONVERT_TO_SHORTPTR(ref_) + half, mask(),
+                     half + width() + height() + 1);
+      }
+      unsigned int sse1, sse2;
+      unsigned int var1;
+      ASM_REGISTER_STATE_CHECK(
+          var1 = params_.func(ref_, width() + 1, x, y, src_, width(), &sse1));
+      const unsigned int var2 = subpel_variance_ref(
+          ref_, src_, params_.log2width, params_.log2height, x, y, &sse2,
+          use_high_bit_depth(), params_.bit_depth);
+      EXPECT_EQ(sse1, sse2) << "for xoffset " << x << " and yoffset " << y;
+      EXPECT_EQ(var1, var2) << "for xoffset " << x << " and yoffset " << y;
+    }
+  }
+}
+
+template <>
+void SubpelVarianceTest<SubpixAvgVarMxNFunc>::RefTest() {
+  for (int x = 0; x < 8; ++x) {
+    for (int y = 0; y < 8; ++y) {
+      if (!use_high_bit_depth()) {
+        for (int j = 0; j < block_size(); j++) {
+          src_[j] = rnd_.Rand8();
+          sec_[j] = rnd_.Rand8();
+        }
+        for (int j = 0; j < block_size() + width() + height() + 1; j++) {
+          ref_[j] = rnd_.Rand8();
+        }
+      } else {
+        for (int j = 0; j < block_size(); j++) {
+          CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask();
+          CONVERT_TO_SHORTPTR(sec_)[j] = rnd_.Rand16() & mask();
+        }
+        for (int j = 0; j < block_size() + width() + height() + 1; j++) {
+          CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() & mask();
+        }
+      }
+      uint32_t sse1, sse2;
+      uint32_t var1, var2;
+      ASM_REGISTER_STATE_CHECK(var1 = params_.func(ref_, width() + 1, x, y,
+                                                   src_, width(), &sse1, sec_));
+      var2 = subpel_avg_variance_ref(ref_, src_, sec_, params_.log2width,
+                                     params_.log2height, x, y, &sse2,
+                                     use_high_bit_depth(), params_.bit_depth);
+      EXPECT_EQ(sse1, sse2) << "at position " << x << ", " << y;
+      EXPECT_EQ(var1, var2) << "at position " << x << ", " << y;
+    }
+  }
+}
+
+template <>
+void SubpelVarianceTest<JntSubpixAvgVarMxNFunc>::RefTest() {
+  for (int x = 0; x < 8; ++x) {
+    for (int y = 0; y < 8; ++y) {
+      if (!use_high_bit_depth()) {
+        for (int j = 0; j < block_size(); j++) {
+          src_[j] = rnd_.Rand8();
+          sec_[j] = rnd_.Rand8();
+        }
+        for (int j = 0; j < block_size() + width() + height() + 1; j++) {
+          ref_[j] = rnd_.Rand8();
+        }
+      } else {
+        for (int j = 0; j < block_size(); j++) {
+          CONVERT_TO_SHORTPTR(src_)[j] = rnd_.Rand16() & mask();
+          CONVERT_TO_SHORTPTR(sec_)[j] = rnd_.Rand16() & mask();
+        }
+        for (int j = 0; j < block_size() + width() + height() + 1; j++) {
+          CONVERT_TO_SHORTPTR(ref_)[j] = rnd_.Rand16() & mask();
+        }
+      }
+      for (int x0 = 0; x0 < 2; ++x0) {
+        for (int y0 = 0; y0 < 4; ++y0) {
+          uint32_t sse1, sse2;
+          uint32_t var1, var2;
+          jcp_param_.fwd_offset = quant_dist_lookup_table[x0][y0][0];
+          jcp_param_.bck_offset = quant_dist_lookup_table[x0][y0][1];
+          ASM_REGISTER_STATE_CHECK(var1 = params_.func(ref_, width() + 0, x, y,
+                                                       src_, width(), &sse1,
+                                                       sec_, &jcp_param_));
+          var2 = jnt_subpel_avg_variance_ref(
+              ref_, src_, sec_, params_.log2width, params_.log2height, x, y,
+              &sse2, use_high_bit_depth(), params_.bit_depth, &jcp_param_);
+          EXPECT_EQ(sse1, sse2) << "at position " << x << ", " << y;
+          EXPECT_EQ(var1, var2) << "at position " << x << ", " << y;
+        }
+      }
+    }
+  }
+}
+
+////////////////////////////////////////////////////////////////////////////////
+
+static const int kMaskMax = 64;
+
+typedef TestParams<ObmcSubpelVarFunc> ObmcSubpelVarianceParams;
+
+template <typename FunctionType>
+class ObmcVarianceTest
+    : public ::testing::TestWithParam<TestParams<FunctionType> > {
+ public:
+  virtual void SetUp() {
+    params_ = this->GetParam();
+
+    rnd_.Reset(ACMRandom::DeterministicSeed());
+    if (!use_high_bit_depth()) {
+      pre_ = reinterpret_cast<uint8_t *>(
+          aom_memalign(32, block_size() + width() + height() + 1));
+    } else {
+      pre_ = CONVERT_TO_BYTEPTR(reinterpret_cast<uint16_t *>(aom_memalign(
+          32, block_size() + width() + height() + 1 * sizeof(uint16_t))));
+    }
+    wsrc_ = reinterpret_cast<int32_t *>(
+        aom_memalign(32, block_size() * sizeof(uint32_t)));
+    mask_ = reinterpret_cast<int32_t *>(
+        aom_memalign(32, block_size() * sizeof(uint32_t)));
+    ASSERT_TRUE(pre_ != NULL);
+    ASSERT_TRUE(wsrc_ != NULL);
+    ASSERT_TRUE(mask_ != NULL);
+  }
+
+  virtual void TearDown() {
+    if (!use_high_bit_depth()) {
+      aom_free(pre_);
+    } else {
+      aom_free(CONVERT_TO_SHORTPTR(pre_));
+    }
+    aom_free(wsrc_);
+    aom_free(mask_);
+    libaom_test::ClearSystemState();
+  }
+
+ protected:
+  void RefTest();
+  void ExtremeRefTest();
+  void SpeedTest();
+
+  ACMRandom rnd_;
+  uint8_t *pre_;
+  int32_t *wsrc_;
+  int32_t *mask_;
+  TestParams<FunctionType> params_;
+
+  // some relay helpers
+  bool use_high_bit_depth() const { return params_.use_high_bit_depth; }
+  int byte_shift() const { return params_.bit_depth - 8; }
+  int block_size() const { return params_.block_size; }
+  int width() const { return params_.width; }
+  int height() const { return params_.height; }
+  uint32_t bd_mask() const { return params_.mask; }
+};
+
+template <>
+void ObmcVarianceTest<ObmcSubpelVarFunc>::RefTest() {
+  for (int x = 0; x < 8; ++x) {
+    for (int y = 0; y < 8; ++y) {
+      if (!use_high_bit_depth())
+        for (int j = 0; j < block_size() + width() + height() + 1; j++)
+          pre_[j] = rnd_.Rand8();
+      else
+        for (int j = 0; j < block_size() + width() + height() + 1; j++)
+          CONVERT_TO_SHORTPTR(pre_)[j] = rnd_.Rand16() & bd_mask();
+      for (int j = 0; j < block_size(); j++) {
+        wsrc_[j] = (rnd_.Rand16() & bd_mask()) * rnd_(kMaskMax * kMaskMax + 1);
+        mask_[j] = rnd_(kMaskMax * kMaskMax + 1);
+      }
+
+      uint32_t sse1, sse2;
+      uint32_t var1, var2;
+      ASM_REGISTER_STATE_CHECK(
+          var1 = params_.func(pre_, width() + 1, x, y, wsrc_, mask_, &sse1));
+      var2 = obmc_subpel_variance_ref(
+          pre_, params_.log2width, params_.log2height, x, y, wsrc_, mask_,
+          &sse2, use_high_bit_depth(), params_.bit_depth);
+      EXPECT_EQ(sse1, sse2) << "for xoffset " << x << " and yoffset " << y;
+      EXPECT_EQ(var1, var2) << "for xoffset " << x << " and yoffset " << y;
+    }
+  }
+}
+
+template <>
+void ObmcVarianceTest<ObmcSubpelVarFunc>::ExtremeRefTest() {
+  // Pre: Set the first half of values to the maximum, the second half to 0.
+  // Mask: same as above
+  // WSrc: Set the first half of values to 0, the second half to the maximum.
+  for (int x = 0; x < 8; ++x) {
+    for (int y = 0; y < 8; ++y) {
+      const int half = block_size() / 2;
+      if (!use_high_bit_depth()) {
+        memset(pre_, 255, half);
+        memset(pre_ + half, 0, half + width() + height() + 1);
+      } else {
+        aom_memset16(CONVERT_TO_SHORTPTR(pre_), bd_mask(), half);
+        aom_memset16(CONVERT_TO_SHORTPTR(pre_) + half, 0, half);
+      }
+      for (int j = 0; j < half; j++) {
+        wsrc_[j] = bd_mask() * kMaskMax * kMaskMax;
+        mask_[j] = 0;
+      }
+      for (int j = half; j < block_size(); j++) {
+        wsrc_[j] = 0;
+        mask_[j] = kMaskMax * kMaskMax;
+      }
+
+      uint32_t sse1, sse2;
+      uint32_t var1, var2;
+      ASM_REGISTER_STATE_CHECK(
+          var1 = params_.func(pre_, width() + 1, x, y, wsrc_, mask_, &sse1));
+      var2 = obmc_subpel_variance_ref(
+          pre_, params_.log2width, params_.log2height, x, y, wsrc_, mask_,
+          &sse2, use_high_bit_depth(), params_.bit_depth);
+      EXPECT_EQ(sse1, sse2) << "for xoffset " << x << " and yoffset " << y;
+      EXPECT_EQ(var1, var2) << "for xoffset " << x << " and yoffset " << y;
+    }
+  }
+}
+
+template <>
+void ObmcVarianceTest<ObmcSubpelVarFunc>::SpeedTest() {
+  if (!use_high_bit_depth())
+    for (int j = 0; j < block_size() + width() + height() + 1; j++)
+      pre_[j] = rnd_.Rand8();
+  else
+    for (int j = 0; j < block_size() + width() + height() + 1; j++)
+      CONVERT_TO_SHORTPTR(pre_)[j] = rnd_.Rand16() & bd_mask();
+  for (int j = 0; j < block_size(); j++) {
+    wsrc_[j] = (rnd_.Rand16() & bd_mask()) * rnd_(kMaskMax * kMaskMax + 1);
+    mask_[j] = rnd_(kMaskMax * kMaskMax + 1);
+  }
+  unsigned int sse1;
+  const int stride = width() + 1;
+  int run_time = 1000000000 / block_size();
+  aom_usec_timer timer;
+
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < run_time; ++i) {
+    int x = rnd_(8);
+    int y = rnd_(8);
+    ASM_REGISTER_STATE_CHECK(
+        params_.func(pre_, stride, x, y, wsrc_, mask_, &sse1));
+  }
+  aom_usec_timer_mark(&timer);
+
+  const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+  printf("obmc_sub_pixel_variance_%dx%d_%d: %d us\n", width(), height(),
+         params_.bit_depth, elapsed_time);
+}
+
+typedef MainTestClass<Get4x4SseFunc> AvxSseTest;
+typedef MainTestClass<VarianceMxNFunc> AvxMseTest;
+typedef MainTestClass<VarianceMxNFunc> AvxVarianceTest;
+typedef SubpelVarianceTest<SubpixVarMxNFunc> AvxSubpelVarianceTest;
+typedef SubpelVarianceTest<SubpixAvgVarMxNFunc> AvxSubpelAvgVarianceTest;
+typedef SubpelVarianceTest<JntSubpixAvgVarMxNFunc> AvxJntSubpelAvgVarianceTest;
+typedef ObmcVarianceTest<ObmcSubpelVarFunc> AvxObmcSubpelVarianceTest;
+
+TEST_P(AvxSseTest, RefSse) { RefTestSse(); }
+TEST_P(AvxSseTest, MaxSse) { MaxTestSse(); }
+TEST_P(AvxMseTest, RefMse) { RefTestMse(); }
+TEST_P(AvxMseTest, MaxMse) { MaxTestMse(); }
+TEST_P(AvxVarianceTest, Zero) { ZeroTest(); }
+TEST_P(AvxVarianceTest, Ref) { RefTest(); }
+TEST_P(AvxVarianceTest, RefStride) { RefStrideTest(); }
+TEST_P(AvxVarianceTest, OneQuarter) { OneQuarterTest(); }
+TEST_P(AvxVarianceTest, DISABLED_Speed) { SpeedTest(); }
+TEST_P(SumOfSquaresTest, Const) { ConstTest(); }
+TEST_P(SumOfSquaresTest, Ref) { RefTest(); }
+TEST_P(AvxSubpelVarianceTest, Ref) { RefTest(); }
+TEST_P(AvxSubpelVarianceTest, ExtremeRef) { ExtremeRefTest(); }
+TEST_P(AvxSubpelAvgVarianceTest, Ref) { RefTest(); }
+TEST_P(AvxJntSubpelAvgVarianceTest, Ref) { RefTest(); }
+TEST_P(AvxObmcSubpelVarianceTest, Ref) { RefTest(); }
+TEST_P(AvxObmcSubpelVarianceTest, ExtremeRef) { ExtremeRefTest(); }
+TEST_P(AvxObmcSubpelVarianceTest, DISABLED_Speed) { SpeedTest(); }
+
+INSTANTIATE_TEST_CASE_P(C, SumOfSquaresTest,
+                        ::testing::Values(aom_get_mb_ss_c));
+
+typedef TestParams<Get4x4SseFunc> SseParams;
+INSTANTIATE_TEST_CASE_P(C, AvxSseTest,
+                        ::testing::Values(SseParams(2, 2,
+                                                    &aom_get4x4sse_cs_c)));
+
+typedef TestParams<VarianceMxNFunc> MseParams;
+INSTANTIATE_TEST_CASE_P(C, AvxMseTest,
+                        ::testing::Values(MseParams(4, 4, &aom_mse16x16_c),
+                                          MseParams(4, 3, &aom_mse16x8_c),
+                                          MseParams(3, 4, &aom_mse8x16_c),
+                                          MseParams(3, 3, &aom_mse8x8_c)));
+
+typedef TestParams<VarianceMxNFunc> VarianceParams;
+INSTANTIATE_TEST_CASE_P(
+    C, AvxVarianceTest,
+    ::testing::Values(VarianceParams(7, 7, &aom_variance128x128_c),
+                      VarianceParams(7, 6, &aom_variance128x64_c),
+                      VarianceParams(6, 7, &aom_variance64x128_c),
+                      VarianceParams(6, 6, &aom_variance64x64_c),
+                      VarianceParams(6, 5, &aom_variance64x32_c),
+                      VarianceParams(5, 6, &aom_variance32x64_c),
+                      VarianceParams(5, 5, &aom_variance32x32_c),
+                      VarianceParams(5, 4, &aom_variance32x16_c),
+                      VarianceParams(4, 5, &aom_variance16x32_c),
+                      VarianceParams(4, 4, &aom_variance16x16_c),
+                      VarianceParams(4, 3, &aom_variance16x8_c),
+                      VarianceParams(3, 4, &aom_variance8x16_c),
+                      VarianceParams(3, 3, &aom_variance8x8_c),
+                      VarianceParams(3, 2, &aom_variance8x4_c),
+                      VarianceParams(2, 3, &aom_variance4x8_c),
+                      VarianceParams(2, 2, &aom_variance4x4_c)));
+
+typedef TestParams<SubpixVarMxNFunc> SubpelVarianceParams;
+INSTANTIATE_TEST_CASE_P(
+    C, AvxSubpelVarianceTest,
+    ::testing::Values(
+        SubpelVarianceParams(7, 7, &aom_sub_pixel_variance128x128_c, 0),
+        SubpelVarianceParams(7, 6, &aom_sub_pixel_variance128x64_c, 0),
+        SubpelVarianceParams(6, 7, &aom_sub_pixel_variance64x128_c, 0),
+        SubpelVarianceParams(6, 6, &aom_sub_pixel_variance64x64_c, 0),
+        SubpelVarianceParams(6, 5, &aom_sub_pixel_variance64x32_c, 0),
+        SubpelVarianceParams(5, 6, &aom_sub_pixel_variance32x64_c, 0),
+        SubpelVarianceParams(5, 5, &aom_sub_pixel_variance32x32_c, 0),
+        SubpelVarianceParams(5, 4, &aom_sub_pixel_variance32x16_c, 0),
+        SubpelVarianceParams(4, 5, &aom_sub_pixel_variance16x32_c, 0),
+        SubpelVarianceParams(4, 4, &aom_sub_pixel_variance16x16_c, 0),
+        SubpelVarianceParams(4, 3, &aom_sub_pixel_variance16x8_c, 0),
+        SubpelVarianceParams(3, 4, &aom_sub_pixel_variance8x16_c, 0),
+        SubpelVarianceParams(3, 3, &aom_sub_pixel_variance8x8_c, 0),
+        SubpelVarianceParams(3, 2, &aom_sub_pixel_variance8x4_c, 0),
+        SubpelVarianceParams(2, 3, &aom_sub_pixel_variance4x8_c, 0),
+        SubpelVarianceParams(2, 2, &aom_sub_pixel_variance4x4_c, 0)));
+
+typedef TestParams<SubpixAvgVarMxNFunc> SubpelAvgVarianceParams;
+INSTANTIATE_TEST_CASE_P(
+    C, AvxSubpelAvgVarianceTest,
+    ::testing::Values(
+        SubpelAvgVarianceParams(7, 7, &aom_sub_pixel_avg_variance128x128_c, 0),
+        SubpelAvgVarianceParams(7, 6, &aom_sub_pixel_avg_variance128x64_c, 0),
+        SubpelAvgVarianceParams(6, 7, &aom_sub_pixel_avg_variance64x128_c, 0),
+        SubpelAvgVarianceParams(6, 6, &aom_sub_pixel_avg_variance64x64_c, 0),
+        SubpelAvgVarianceParams(6, 5, &aom_sub_pixel_avg_variance64x32_c, 0),
+        SubpelAvgVarianceParams(5, 6, &aom_sub_pixel_avg_variance32x64_c, 0),
+        SubpelAvgVarianceParams(5, 5, &aom_sub_pixel_avg_variance32x32_c, 0),
+        SubpelAvgVarianceParams(5, 4, &aom_sub_pixel_avg_variance32x16_c, 0),
+        SubpelAvgVarianceParams(4, 5, &aom_sub_pixel_avg_variance16x32_c, 0),
+        SubpelAvgVarianceParams(4, 4, &aom_sub_pixel_avg_variance16x16_c, 0),
+        SubpelAvgVarianceParams(4, 3, &aom_sub_pixel_avg_variance16x8_c, 0),
+        SubpelAvgVarianceParams(3, 4, &aom_sub_pixel_avg_variance8x16_c, 0),
+        SubpelAvgVarianceParams(3, 3, &aom_sub_pixel_avg_variance8x8_c, 0),
+        SubpelAvgVarianceParams(3, 2, &aom_sub_pixel_avg_variance8x4_c, 0),
+        SubpelAvgVarianceParams(2, 3, &aom_sub_pixel_avg_variance4x8_c, 0),
+        SubpelAvgVarianceParams(2, 2, &aom_sub_pixel_avg_variance4x4_c, 0)));
+
+typedef TestParams<JntSubpixAvgVarMxNFunc> JntSubpelAvgVarianceParams;
+INSTANTIATE_TEST_CASE_P(
+    C, AvxJntSubpelAvgVarianceTest,
+    ::testing::Values(
+        JntSubpelAvgVarianceParams(6, 6, &aom_jnt_sub_pixel_avg_variance64x64_c,
+                                   0),
+        JntSubpelAvgVarianceParams(6, 5, &aom_jnt_sub_pixel_avg_variance64x32_c,
+                                   0),
+        JntSubpelAvgVarianceParams(5, 6, &aom_jnt_sub_pixel_avg_variance32x64_c,
+                                   0),
+        JntSubpelAvgVarianceParams(5, 5, &aom_jnt_sub_pixel_avg_variance32x32_c,
+                                   0),
+        JntSubpelAvgVarianceParams(5, 4, &aom_jnt_sub_pixel_avg_variance32x16_c,
+                                   0),
+        JntSubpelAvgVarianceParams(4, 5, &aom_jnt_sub_pixel_avg_variance16x32_c,
+                                   0),
+        JntSubpelAvgVarianceParams(4, 4, &aom_jnt_sub_pixel_avg_variance16x16_c,
+                                   0),
+        JntSubpelAvgVarianceParams(4, 3, &aom_jnt_sub_pixel_avg_variance16x8_c,
+                                   0),
+        JntSubpelAvgVarianceParams(3, 4, &aom_jnt_sub_pixel_avg_variance8x16_c,
+                                   0),
+        JntSubpelAvgVarianceParams(3, 3, &aom_jnt_sub_pixel_avg_variance8x8_c,
+                                   0),
+        JntSubpelAvgVarianceParams(3, 2, &aom_jnt_sub_pixel_avg_variance8x4_c,
+                                   0),
+        JntSubpelAvgVarianceParams(2, 3, &aom_jnt_sub_pixel_avg_variance4x8_c,
+                                   0),
+        JntSubpelAvgVarianceParams(2, 2, &aom_jnt_sub_pixel_avg_variance4x4_c,
+                                   0)));
+
+INSTANTIATE_TEST_CASE_P(
+    C, AvxObmcSubpelVarianceTest,
+    ::testing::Values(
+        ObmcSubpelVarianceParams(7, 7, &aom_obmc_sub_pixel_variance128x128_c,
+                                 0),
+        ObmcSubpelVarianceParams(7, 6, &aom_obmc_sub_pixel_variance128x64_c, 0),
+        ObmcSubpelVarianceParams(6, 7, &aom_obmc_sub_pixel_variance64x128_c, 0),
+        ObmcSubpelVarianceParams(6, 6, &aom_obmc_sub_pixel_variance64x64_c, 0),
+        ObmcSubpelVarianceParams(6, 5, &aom_obmc_sub_pixel_variance64x32_c, 0),
+        ObmcSubpelVarianceParams(5, 6, &aom_obmc_sub_pixel_variance32x64_c, 0),
+        ObmcSubpelVarianceParams(5, 5, &aom_obmc_sub_pixel_variance32x32_c, 0),
+        ObmcSubpelVarianceParams(5, 4, &aom_obmc_sub_pixel_variance32x16_c, 0),
+        ObmcSubpelVarianceParams(4, 5, &aom_obmc_sub_pixel_variance16x32_c, 0),
+        ObmcSubpelVarianceParams(4, 4, &aom_obmc_sub_pixel_variance16x16_c, 0),
+        ObmcSubpelVarianceParams(4, 3, &aom_obmc_sub_pixel_variance16x8_c, 0),
+        ObmcSubpelVarianceParams(3, 4, &aom_obmc_sub_pixel_variance8x16_c, 0),
+        ObmcSubpelVarianceParams(3, 3, &aom_obmc_sub_pixel_variance8x8_c, 0),
+        ObmcSubpelVarianceParams(3, 2, &aom_obmc_sub_pixel_variance8x4_c, 0),
+        ObmcSubpelVarianceParams(2, 3, &aom_obmc_sub_pixel_variance4x8_c, 0),
+        ObmcSubpelVarianceParams(2, 2, &aom_obmc_sub_pixel_variance4x4_c, 0)));
+
+typedef MainTestClass<VarianceMxNFunc> AvxHBDMseTest;
+typedef MainTestClass<VarianceMxNFunc> AvxHBDVarianceTest;
+typedef SubpelVarianceTest<SubpixVarMxNFunc> AvxHBDSubpelVarianceTest;
+typedef SubpelVarianceTest<SubpixAvgVarMxNFunc> AvxHBDSubpelAvgVarianceTest;
+typedef ObmcVarianceTest<ObmcSubpelVarFunc> AvxHBDObmcSubpelVarianceTest;
+
+TEST_P(AvxHBDMseTest, RefMse) { RefTestMse(); }
+TEST_P(AvxHBDMseTest, MaxMse) { MaxTestMse(); }
+TEST_P(AvxHBDVarianceTest, Zero) { ZeroTest(); }
+TEST_P(AvxHBDVarianceTest, Ref) { RefTest(); }
+TEST_P(AvxHBDVarianceTest, RefStride) { RefStrideTest(); }
+TEST_P(AvxHBDVarianceTest, OneQuarter) { OneQuarterTest(); }
+TEST_P(AvxHBDVarianceTest, DISABLED_Speed) { SpeedTest(); }
+TEST_P(AvxHBDSubpelVarianceTest, Ref) { RefTest(); }
+TEST_P(AvxHBDSubpelVarianceTest, ExtremeRef) { ExtremeRefTest(); }
+TEST_P(AvxHBDSubpelAvgVarianceTest, Ref) { RefTest(); }
+
+/* TODO(debargha): This test does not support the highbd version
+INSTANTIATE_TEST_CASE_P(
+    C, AvxHBDMseTest,
+    ::testing::Values(make_tuple(4, 4, &aom_highbd_12_mse16x16_c),
+                      make_tuple(4, 4, &aom_highbd_12_mse16x8_c),
+                      make_tuple(4, 4, &aom_highbd_12_mse8x16_c),
+                      make_tuple(4, 4, &aom_highbd_12_mse8x8_c),
+                      make_tuple(4, 4, &aom_highbd_10_mse16x16_c),
+                      make_tuple(4, 4, &aom_highbd_10_mse16x8_c),
+                      make_tuple(4, 4, &aom_highbd_10_mse8x16_c),
+                      make_tuple(4, 4, &aom_highbd_10_mse8x8_c),
+                      make_tuple(4, 4, &aom_highbd_8_mse16x16_c),
+                      make_tuple(4, 4, &aom_highbd_8_mse16x8_c),
+                      make_tuple(4, 4, &aom_highbd_8_mse8x16_c),
+                      make_tuple(4, 4, &aom_highbd_8_mse8x8_c)));
+*/
+
+const VarianceParams kArrayHBDVariance_c[] = {
+  VarianceParams(7, 7, &aom_highbd_12_variance128x128_c, 12),
+  VarianceParams(7, 6, &aom_highbd_12_variance128x64_c, 12),
+  VarianceParams(6, 7, &aom_highbd_12_variance64x128_c, 12),
+  VarianceParams(6, 6, &aom_highbd_12_variance64x64_c, 12),
+  VarianceParams(6, 5, &aom_highbd_12_variance64x32_c, 12),
+  VarianceParams(5, 6, &aom_highbd_12_variance32x64_c, 12),
+  VarianceParams(5, 5, &aom_highbd_12_variance32x32_c, 12),
+  VarianceParams(5, 4, &aom_highbd_12_variance32x16_c, 12),
+  VarianceParams(4, 5, &aom_highbd_12_variance16x32_c, 12),
+  VarianceParams(4, 4, &aom_highbd_12_variance16x16_c, 12),
+  VarianceParams(4, 3, &aom_highbd_12_variance16x8_c, 12),
+  VarianceParams(3, 4, &aom_highbd_12_variance8x16_c, 12),
+  VarianceParams(3, 3, &aom_highbd_12_variance8x8_c, 12),
+  VarianceParams(3, 2, &aom_highbd_12_variance8x4_c, 12),
+  VarianceParams(2, 3, &aom_highbd_12_variance4x8_c, 12),
+  VarianceParams(2, 2, &aom_highbd_12_variance4x4_c, 12),
+  VarianceParams(7, 7, &aom_highbd_10_variance128x128_c, 10),
+  VarianceParams(7, 6, &aom_highbd_10_variance128x64_c, 10),
+  VarianceParams(6, 7, &aom_highbd_10_variance64x128_c, 10),
+  VarianceParams(6, 6, &aom_highbd_10_variance64x64_c, 10),
+  VarianceParams(6, 5, &aom_highbd_10_variance64x32_c, 10),
+  VarianceParams(5, 6, &aom_highbd_10_variance32x64_c, 10),
+  VarianceParams(5, 5, &aom_highbd_10_variance32x32_c, 10),
+  VarianceParams(5, 4, &aom_highbd_10_variance32x16_c, 10),
+  VarianceParams(4, 5, &aom_highbd_10_variance16x32_c, 10),
+  VarianceParams(4, 4, &aom_highbd_10_variance16x16_c, 10),
+  VarianceParams(4, 3, &aom_highbd_10_variance16x8_c, 10),
+  VarianceParams(3, 4, &aom_highbd_10_variance8x16_c, 10),
+  VarianceParams(3, 3, &aom_highbd_10_variance8x8_c, 10),
+  VarianceParams(3, 2, &aom_highbd_10_variance8x4_c, 10),
+  VarianceParams(2, 3, &aom_highbd_10_variance4x8_c, 10),
+  VarianceParams(2, 2, &aom_highbd_10_variance4x4_c, 10),
+  VarianceParams(7, 7, &aom_highbd_8_variance128x128_c, 8),
+  VarianceParams(7, 6, &aom_highbd_8_variance128x64_c, 8),
+  VarianceParams(6, 7, &aom_highbd_8_variance64x128_c, 8),
+  VarianceParams(6, 6, &aom_highbd_8_variance64x64_c, 8),
+  VarianceParams(6, 5, &aom_highbd_8_variance64x32_c, 8),
+  VarianceParams(5, 6, &aom_highbd_8_variance32x64_c, 8),
+  VarianceParams(5, 5, &aom_highbd_8_variance32x32_c, 8),
+  VarianceParams(5, 4, &aom_highbd_8_variance32x16_c, 8),
+  VarianceParams(4, 5, &aom_highbd_8_variance16x32_c, 8),
+  VarianceParams(4, 4, &aom_highbd_8_variance16x16_c, 8),
+  VarianceParams(4, 3, &aom_highbd_8_variance16x8_c, 8),
+  VarianceParams(3, 4, &aom_highbd_8_variance8x16_c, 8),
+  VarianceParams(3, 3, &aom_highbd_8_variance8x8_c, 8),
+  VarianceParams(3, 2, &aom_highbd_8_variance8x4_c, 8),
+  VarianceParams(2, 3, &aom_highbd_8_variance4x8_c, 8),
+  VarianceParams(2, 2, &aom_highbd_8_variance4x4_c, 8)
+};
+INSTANTIATE_TEST_CASE_P(C, AvxHBDVarianceTest,
+                        ::testing::ValuesIn(kArrayHBDVariance_c));
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AvxHBDVarianceTest,
+    ::testing::Values(
+        VarianceParams(2, 2, &aom_highbd_8_variance4x4_sse4_1, 8),
+        VarianceParams(2, 2, &aom_highbd_10_variance4x4_sse4_1, 10),
+        VarianceParams(2, 2, &aom_highbd_12_variance4x4_sse4_1, 12)));
+#endif  // HAVE_SSE4_1
+
+const SubpelVarianceParams kArrayHBDSubpelVariance_c[] = {
+  SubpelVarianceParams(7, 7, &aom_highbd_8_sub_pixel_variance128x128_c, 8),
+  SubpelVarianceParams(7, 6, &aom_highbd_8_sub_pixel_variance128x64_c, 8),
+  SubpelVarianceParams(6, 7, &aom_highbd_8_sub_pixel_variance64x128_c, 8),
+  SubpelVarianceParams(6, 6, &aom_highbd_8_sub_pixel_variance64x64_c, 8),
+  SubpelVarianceParams(6, 5, &aom_highbd_8_sub_pixel_variance64x32_c, 8),
+  SubpelVarianceParams(5, 6, &aom_highbd_8_sub_pixel_variance32x64_c, 8),
+  SubpelVarianceParams(5, 5, &aom_highbd_8_sub_pixel_variance32x32_c, 8),
+  SubpelVarianceParams(5, 4, &aom_highbd_8_sub_pixel_variance32x16_c, 8),
+  SubpelVarianceParams(4, 5, &aom_highbd_8_sub_pixel_variance16x32_c, 8),
+  SubpelVarianceParams(4, 4, &aom_highbd_8_sub_pixel_variance16x16_c, 8),
+  SubpelVarianceParams(4, 3, &aom_highbd_8_sub_pixel_variance16x8_c, 8),
+  SubpelVarianceParams(3, 4, &aom_highbd_8_sub_pixel_variance8x16_c, 8),
+  SubpelVarianceParams(3, 3, &aom_highbd_8_sub_pixel_variance8x8_c, 8),
+  SubpelVarianceParams(3, 2, &aom_highbd_8_sub_pixel_variance8x4_c, 8),
+  SubpelVarianceParams(2, 3, &aom_highbd_8_sub_pixel_variance4x8_c, 8),
+  SubpelVarianceParams(2, 2, &aom_highbd_8_sub_pixel_variance4x4_c, 8),
+  SubpelVarianceParams(7, 7, &aom_highbd_10_sub_pixel_variance128x128_c, 10),
+  SubpelVarianceParams(7, 6, &aom_highbd_10_sub_pixel_variance128x64_c, 10),
+  SubpelVarianceParams(6, 7, &aom_highbd_10_sub_pixel_variance64x128_c, 10),
+  SubpelVarianceParams(6, 6, &aom_highbd_10_sub_pixel_variance64x64_c, 10),
+  SubpelVarianceParams(6, 5, &aom_highbd_10_sub_pixel_variance64x32_c, 10),
+  SubpelVarianceParams(5, 6, &aom_highbd_10_sub_pixel_variance32x64_c, 10),
+  SubpelVarianceParams(5, 5, &aom_highbd_10_sub_pixel_variance32x32_c, 10),
+  SubpelVarianceParams(5, 4, &aom_highbd_10_sub_pixel_variance32x16_c, 10),
+  SubpelVarianceParams(4, 5, &aom_highbd_10_sub_pixel_variance16x32_c, 10),
+  SubpelVarianceParams(4, 4, &aom_highbd_10_sub_pixel_variance16x16_c, 10),
+  SubpelVarianceParams(4, 3, &aom_highbd_10_sub_pixel_variance16x8_c, 10),
+  SubpelVarianceParams(3, 4, &aom_highbd_10_sub_pixel_variance8x16_c, 10),
+  SubpelVarianceParams(3, 3, &aom_highbd_10_sub_pixel_variance8x8_c, 10),
+  SubpelVarianceParams(3, 2, &aom_highbd_10_sub_pixel_variance8x4_c, 10),
+  SubpelVarianceParams(2, 3, &aom_highbd_10_sub_pixel_variance4x8_c, 10),
+  SubpelVarianceParams(2, 2, &aom_highbd_10_sub_pixel_variance4x4_c, 10),
+  SubpelVarianceParams(7, 7, &aom_highbd_12_sub_pixel_variance128x128_c, 12),
+  SubpelVarianceParams(7, 6, &aom_highbd_12_sub_pixel_variance128x64_c, 12),
+  SubpelVarianceParams(6, 7, &aom_highbd_12_sub_pixel_variance64x128_c, 12),
+  SubpelVarianceParams(6, 6, &aom_highbd_12_sub_pixel_variance64x64_c, 12),
+  SubpelVarianceParams(6, 5, &aom_highbd_12_sub_pixel_variance64x32_c, 12),
+  SubpelVarianceParams(5, 6, &aom_highbd_12_sub_pixel_variance32x64_c, 12),
+  SubpelVarianceParams(5, 5, &aom_highbd_12_sub_pixel_variance32x32_c, 12),
+  SubpelVarianceParams(5, 4, &aom_highbd_12_sub_pixel_variance32x16_c, 12),
+  SubpelVarianceParams(4, 5, &aom_highbd_12_sub_pixel_variance16x32_c, 12),
+  SubpelVarianceParams(4, 4, &aom_highbd_12_sub_pixel_variance16x16_c, 12),
+  SubpelVarianceParams(4, 3, &aom_highbd_12_sub_pixel_variance16x8_c, 12),
+  SubpelVarianceParams(3, 4, &aom_highbd_12_sub_pixel_variance8x16_c, 12),
+  SubpelVarianceParams(3, 3, &aom_highbd_12_sub_pixel_variance8x8_c, 12),
+  SubpelVarianceParams(3, 2, &aom_highbd_12_sub_pixel_variance8x4_c, 12),
+  SubpelVarianceParams(2, 3, &aom_highbd_12_sub_pixel_variance4x8_c, 12),
+  SubpelVarianceParams(2, 2, &aom_highbd_12_sub_pixel_variance4x4_c, 12),
+};
+INSTANTIATE_TEST_CASE_P(C, AvxHBDSubpelVarianceTest,
+                        ::testing::ValuesIn(kArrayHBDSubpelVariance_c));
+
+const SubpelAvgVarianceParams kArrayHBDSubpelAvgVariance_c[] = {
+  SubpelAvgVarianceParams(7, 7, &aom_highbd_8_sub_pixel_avg_variance128x128_c,
+                          8),
+  SubpelAvgVarianceParams(7, 6, &aom_highbd_8_sub_pixel_avg_variance128x64_c,
+                          8),
+  SubpelAvgVarianceParams(6, 7, &aom_highbd_8_sub_pixel_avg_variance64x128_c,
+                          8),
+  SubpelAvgVarianceParams(6, 6, &aom_highbd_8_sub_pixel_avg_variance64x64_c, 8),
+  SubpelAvgVarianceParams(6, 5, &aom_highbd_8_sub_pixel_avg_variance64x32_c, 8),
+  SubpelAvgVarianceParams(5, 6, &aom_highbd_8_sub_pixel_avg_variance32x64_c, 8),
+  SubpelAvgVarianceParams(5, 5, &aom_highbd_8_sub_pixel_avg_variance32x32_c, 8),
+  SubpelAvgVarianceParams(5, 4, &aom_highbd_8_sub_pixel_avg_variance32x16_c, 8),
+  SubpelAvgVarianceParams(4, 5, &aom_highbd_8_sub_pixel_avg_variance16x32_c, 8),
+  SubpelAvgVarianceParams(4, 4, &aom_highbd_8_sub_pixel_avg_variance16x16_c, 8),
+  SubpelAvgVarianceParams(4, 3, &aom_highbd_8_sub_pixel_avg_variance16x8_c, 8),
+  SubpelAvgVarianceParams(3, 4, &aom_highbd_8_sub_pixel_avg_variance8x16_c, 8),
+  SubpelAvgVarianceParams(3, 3, &aom_highbd_8_sub_pixel_avg_variance8x8_c, 8),
+  SubpelAvgVarianceParams(3, 2, &aom_highbd_8_sub_pixel_avg_variance8x4_c, 8),
+  SubpelAvgVarianceParams(2, 3, &aom_highbd_8_sub_pixel_avg_variance4x8_c, 8),
+  SubpelAvgVarianceParams(2, 2, &aom_highbd_8_sub_pixel_avg_variance4x4_c, 8),
+  SubpelAvgVarianceParams(7, 7, &aom_highbd_10_sub_pixel_avg_variance128x128_c,
+                          10),
+  SubpelAvgVarianceParams(7, 6, &aom_highbd_10_sub_pixel_avg_variance128x64_c,
+                          10),
+  SubpelAvgVarianceParams(6, 7, &aom_highbd_10_sub_pixel_avg_variance64x128_c,
+                          10),
+  SubpelAvgVarianceParams(6, 6, &aom_highbd_10_sub_pixel_avg_variance64x64_c,
+                          10),
+  SubpelAvgVarianceParams(6, 5, &aom_highbd_10_sub_pixel_avg_variance64x32_c,
+                          10),
+  SubpelAvgVarianceParams(5, 6, &aom_highbd_10_sub_pixel_avg_variance32x64_c,
+                          10),
+  SubpelAvgVarianceParams(5, 5, &aom_highbd_10_sub_pixel_avg_variance32x32_c,
+                          10),
+  SubpelAvgVarianceParams(5, 4, &aom_highbd_10_sub_pixel_avg_variance32x16_c,
+                          10),
+  SubpelAvgVarianceParams(4, 5, &aom_highbd_10_sub_pixel_avg_variance16x32_c,
+                          10),
+  SubpelAvgVarianceParams(4, 4, &aom_highbd_10_sub_pixel_avg_variance16x16_c,
+                          10),
+  SubpelAvgVarianceParams(4, 3, &aom_highbd_10_sub_pixel_avg_variance16x8_c,
+                          10),
+  SubpelAvgVarianceParams(3, 4, &aom_highbd_10_sub_pixel_avg_variance8x16_c,
+                          10),
+  SubpelAvgVarianceParams(3, 3, &aom_highbd_10_sub_pixel_avg_variance8x8_c, 10),
+  SubpelAvgVarianceParams(3, 2, &aom_highbd_10_sub_pixel_avg_variance8x4_c, 10),
+  SubpelAvgVarianceParams(2, 3, &aom_highbd_10_sub_pixel_avg_variance4x8_c, 10),
+  SubpelAvgVarianceParams(2, 2, &aom_highbd_10_sub_pixel_avg_variance4x4_c, 10),
+  SubpelAvgVarianceParams(7, 7, &aom_highbd_12_sub_pixel_avg_variance128x128_c,
+                          12),
+  SubpelAvgVarianceParams(7, 6, &aom_highbd_12_sub_pixel_avg_variance128x64_c,
+                          12),
+  SubpelAvgVarianceParams(6, 7, &aom_highbd_12_sub_pixel_avg_variance64x128_c,
+                          12),
+  SubpelAvgVarianceParams(6, 6, &aom_highbd_12_sub_pixel_avg_variance64x64_c,
+                          12),
+  SubpelAvgVarianceParams(6, 5, &aom_highbd_12_sub_pixel_avg_variance64x32_c,
+                          12),
+  SubpelAvgVarianceParams(5, 6, &aom_highbd_12_sub_pixel_avg_variance32x64_c,
+                          12),
+  SubpelAvgVarianceParams(5, 5, &aom_highbd_12_sub_pixel_avg_variance32x32_c,
+                          12),
+  SubpelAvgVarianceParams(5, 4, &aom_highbd_12_sub_pixel_avg_variance32x16_c,
+                          12),
+  SubpelAvgVarianceParams(4, 5, &aom_highbd_12_sub_pixel_avg_variance16x32_c,
+                          12),
+  SubpelAvgVarianceParams(4, 4, &aom_highbd_12_sub_pixel_avg_variance16x16_c,
+                          12),
+  SubpelAvgVarianceParams(4, 3, &aom_highbd_12_sub_pixel_avg_variance16x8_c,
+                          12),
+  SubpelAvgVarianceParams(3, 4, &aom_highbd_12_sub_pixel_avg_variance8x16_c,
+                          12),
+  SubpelAvgVarianceParams(3, 3, &aom_highbd_12_sub_pixel_avg_variance8x8_c, 12),
+  SubpelAvgVarianceParams(3, 2, &aom_highbd_12_sub_pixel_avg_variance8x4_c, 12),
+  SubpelAvgVarianceParams(2, 3, &aom_highbd_12_sub_pixel_avg_variance4x8_c, 12),
+  SubpelAvgVarianceParams(2, 2, &aom_highbd_12_sub_pixel_avg_variance4x4_c, 12)
+};
+INSTANTIATE_TEST_CASE_P(C, AvxHBDSubpelAvgVarianceTest,
+                        ::testing::ValuesIn(kArrayHBDSubpelAvgVariance_c));
+
+const ObmcSubpelVarianceParams kArrayHBDObmcSubpelVariance_c[] = {
+  ObmcSubpelVarianceParams(7, 7, &aom_highbd_obmc_sub_pixel_variance128x128_c,
+                           8),
+  ObmcSubpelVarianceParams(7, 6, &aom_highbd_obmc_sub_pixel_variance128x64_c,
+                           8),
+  ObmcSubpelVarianceParams(6, 7, &aom_highbd_obmc_sub_pixel_variance64x128_c,
+                           8),
+  ObmcSubpelVarianceParams(6, 6, &aom_highbd_obmc_sub_pixel_variance64x64_c, 8),
+  ObmcSubpelVarianceParams(6, 5, &aom_highbd_obmc_sub_pixel_variance64x32_c, 8),
+  ObmcSubpelVarianceParams(5, 6, &aom_highbd_obmc_sub_pixel_variance32x64_c, 8),
+  ObmcSubpelVarianceParams(5, 5, &aom_highbd_obmc_sub_pixel_variance32x32_c, 8),
+  ObmcSubpelVarianceParams(5, 4, &aom_highbd_obmc_sub_pixel_variance32x16_c, 8),
+  ObmcSubpelVarianceParams(4, 5, &aom_highbd_obmc_sub_pixel_variance16x32_c, 8),
+  ObmcSubpelVarianceParams(4, 4, &aom_highbd_obmc_sub_pixel_variance16x16_c, 8),
+  ObmcSubpelVarianceParams(4, 3, &aom_highbd_obmc_sub_pixel_variance16x8_c, 8),
+  ObmcSubpelVarianceParams(3, 4, &aom_highbd_obmc_sub_pixel_variance8x16_c, 8),
+  ObmcSubpelVarianceParams(3, 3, &aom_highbd_obmc_sub_pixel_variance8x8_c, 8),
+  ObmcSubpelVarianceParams(3, 2, &aom_highbd_obmc_sub_pixel_variance8x4_c, 8),
+  ObmcSubpelVarianceParams(2, 3, &aom_highbd_obmc_sub_pixel_variance4x8_c, 8),
+  ObmcSubpelVarianceParams(2, 2, &aom_highbd_obmc_sub_pixel_variance4x4_c, 8),
+  ObmcSubpelVarianceParams(7, 7,
+                           &aom_highbd_10_obmc_sub_pixel_variance128x128_c, 10),
+  ObmcSubpelVarianceParams(7, 6, &aom_highbd_10_obmc_sub_pixel_variance128x64_c,
+                           10),
+  ObmcSubpelVarianceParams(6, 7, &aom_highbd_10_obmc_sub_pixel_variance64x128_c,
+                           10),
+  ObmcSubpelVarianceParams(6, 6, &aom_highbd_10_obmc_sub_pixel_variance64x64_c,
+                           10),
+  ObmcSubpelVarianceParams(6, 5, &aom_highbd_10_obmc_sub_pixel_variance64x32_c,
+                           10),
+  ObmcSubpelVarianceParams(5, 6, &aom_highbd_10_obmc_sub_pixel_variance32x64_c,
+                           10),
+  ObmcSubpelVarianceParams(5, 5, &aom_highbd_10_obmc_sub_pixel_variance32x32_c,
+                           10),
+  ObmcSubpelVarianceParams(5, 4, &aom_highbd_10_obmc_sub_pixel_variance32x16_c,
+                           10),
+  ObmcSubpelVarianceParams(4, 5, &aom_highbd_10_obmc_sub_pixel_variance16x32_c,
+                           10),
+  ObmcSubpelVarianceParams(4, 4, &aom_highbd_10_obmc_sub_pixel_variance16x16_c,
+                           10),
+  ObmcSubpelVarianceParams(4, 3, &aom_highbd_10_obmc_sub_pixel_variance16x8_c,
+                           10),
+  ObmcSubpelVarianceParams(3, 4, &aom_highbd_10_obmc_sub_pixel_variance8x16_c,
+                           10),
+  ObmcSubpelVarianceParams(3, 3, &aom_highbd_10_obmc_sub_pixel_variance8x8_c,
+                           10),
+  ObmcSubpelVarianceParams(3, 2, &aom_highbd_10_obmc_sub_pixel_variance8x4_c,
+                           10),
+  ObmcSubpelVarianceParams(2, 3, &aom_highbd_10_obmc_sub_pixel_variance4x8_c,
+                           10),
+  ObmcSubpelVarianceParams(2, 2, &aom_highbd_10_obmc_sub_pixel_variance4x4_c,
+                           10),
+  ObmcSubpelVarianceParams(7, 7,
+                           &aom_highbd_12_obmc_sub_pixel_variance128x128_c, 12),
+  ObmcSubpelVarianceParams(7, 6, &aom_highbd_12_obmc_sub_pixel_variance128x64_c,
+                           12),
+  ObmcSubpelVarianceParams(6, 7, &aom_highbd_12_obmc_sub_pixel_variance64x128_c,
+                           12),
+  ObmcSubpelVarianceParams(6, 6, &aom_highbd_12_obmc_sub_pixel_variance64x64_c,
+                           12),
+  ObmcSubpelVarianceParams(6, 5, &aom_highbd_12_obmc_sub_pixel_variance64x32_c,
+                           12),
+  ObmcSubpelVarianceParams(5, 6, &aom_highbd_12_obmc_sub_pixel_variance32x64_c,
+                           12),
+  ObmcSubpelVarianceParams(5, 5, &aom_highbd_12_obmc_sub_pixel_variance32x32_c,
+                           12),
+  ObmcSubpelVarianceParams(5, 4, &aom_highbd_12_obmc_sub_pixel_variance32x16_c,
+                           12),
+  ObmcSubpelVarianceParams(4, 5, &aom_highbd_12_obmc_sub_pixel_variance16x32_c,
+                           12),
+  ObmcSubpelVarianceParams(4, 4, &aom_highbd_12_obmc_sub_pixel_variance16x16_c,
+                           12),
+  ObmcSubpelVarianceParams(4, 3, &aom_highbd_12_obmc_sub_pixel_variance16x8_c,
+                           12),
+  ObmcSubpelVarianceParams(3, 4, &aom_highbd_12_obmc_sub_pixel_variance8x16_c,
+                           12),
+  ObmcSubpelVarianceParams(3, 3, &aom_highbd_12_obmc_sub_pixel_variance8x8_c,
+                           12),
+  ObmcSubpelVarianceParams(3, 2, &aom_highbd_12_obmc_sub_pixel_variance8x4_c,
+                           12),
+  ObmcSubpelVarianceParams(2, 3, &aom_highbd_12_obmc_sub_pixel_variance4x8_c,
+                           12),
+  ObmcSubpelVarianceParams(2, 2, &aom_highbd_12_obmc_sub_pixel_variance4x4_c,
+                           12)
+};
+INSTANTIATE_TEST_CASE_P(C, AvxHBDObmcSubpelVarianceTest,
+                        ::testing::ValuesIn(kArrayHBDObmcSubpelVariance_c));
+
+#if HAVE_SSE2
+INSTANTIATE_TEST_CASE_P(SSE2, SumOfSquaresTest,
+                        ::testing::Values(aom_get_mb_ss_sse2));
+
+INSTANTIATE_TEST_CASE_P(SSE2, AvxMseTest,
+                        ::testing::Values(MseParams(4, 4, &aom_mse16x16_sse2),
+                                          MseParams(4, 3, &aom_mse16x8_sse2),
+                                          MseParams(3, 4, &aom_mse8x16_sse2),
+                                          MseParams(3, 3, &aom_mse8x8_sse2)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2, AvxVarianceTest,
+    ::testing::Values(VarianceParams(7, 7, &aom_variance128x128_sse2),
+                      VarianceParams(7, 6, &aom_variance128x64_sse2),
+                      VarianceParams(6, 7, &aom_variance64x128_sse2),
+                      VarianceParams(6, 6, &aom_variance64x64_sse2),
+                      VarianceParams(6, 5, &aom_variance64x32_sse2),
+                      VarianceParams(6, 4, &aom_variance64x16_sse2),
+                      VarianceParams(5, 6, &aom_variance32x64_sse2),
+                      VarianceParams(5, 5, &aom_variance32x32_sse2),
+                      VarianceParams(5, 4, &aom_variance32x16_sse2),
+                      VarianceParams(5, 3, &aom_variance32x8_sse2),
+                      VarianceParams(4, 6, &aom_variance16x64_sse2),
+                      VarianceParams(4, 5, &aom_variance16x32_sse2),
+                      VarianceParams(4, 4, &aom_variance16x16_sse2),
+                      VarianceParams(4, 3, &aom_variance16x8_sse2),
+                      VarianceParams(4, 2, &aom_variance16x4_sse2),
+                      VarianceParams(3, 5, &aom_variance8x32_sse2),
+                      VarianceParams(3, 4, &aom_variance8x16_sse2),
+                      VarianceParams(3, 3, &aom_variance8x8_sse2),
+                      VarianceParams(3, 2, &aom_variance8x4_sse2),
+                      VarianceParams(2, 4, &aom_variance4x16_sse2),
+                      VarianceParams(2, 3, &aom_variance4x8_sse2),
+                      VarianceParams(2, 2, &aom_variance4x4_sse2)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2, AvxSubpelVarianceTest,
+    ::testing::Values(
+        SubpelVarianceParams(7, 7, &aom_sub_pixel_variance128x128_sse2, 0),
+        SubpelVarianceParams(7, 6, &aom_sub_pixel_variance128x64_sse2, 0),
+        SubpelVarianceParams(6, 7, &aom_sub_pixel_variance64x128_sse2, 0),
+        SubpelVarianceParams(6, 6, &aom_sub_pixel_variance64x64_sse2, 0),
+        SubpelVarianceParams(6, 5, &aom_sub_pixel_variance64x32_sse2, 0),
+        SubpelVarianceParams(5, 6, &aom_sub_pixel_variance32x64_sse2, 0),
+        SubpelVarianceParams(5, 5, &aom_sub_pixel_variance32x32_sse2, 0),
+        SubpelVarianceParams(5, 4, &aom_sub_pixel_variance32x16_sse2, 0),
+        SubpelVarianceParams(4, 5, &aom_sub_pixel_variance16x32_sse2, 0),
+        SubpelVarianceParams(4, 4, &aom_sub_pixel_variance16x16_sse2, 0),
+        SubpelVarianceParams(4, 3, &aom_sub_pixel_variance16x8_sse2, 0),
+        SubpelVarianceParams(3, 4, &aom_sub_pixel_variance8x16_sse2, 0),
+        SubpelVarianceParams(3, 3, &aom_sub_pixel_variance8x8_sse2, 0),
+        SubpelVarianceParams(3, 2, &aom_sub_pixel_variance8x4_sse2, 0),
+        SubpelVarianceParams(2, 3, &aom_sub_pixel_variance4x8_sse2, 0),
+        SubpelVarianceParams(2, 2, &aom_sub_pixel_variance4x4_sse2, 0)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE2, AvxSubpelAvgVarianceTest,
+    ::testing::Values(
+        SubpelAvgVarianceParams(7, 7, &aom_sub_pixel_avg_variance128x128_sse2,
+                                0),
+        SubpelAvgVarianceParams(7, 6, &aom_sub_pixel_avg_variance128x64_sse2,
+                                0),
+        SubpelAvgVarianceParams(6, 7, &aom_sub_pixel_avg_variance64x128_sse2,
+                                0),
+        SubpelAvgVarianceParams(6, 6, &aom_sub_pixel_avg_variance64x64_sse2, 0),
+        SubpelAvgVarianceParams(6, 5, &aom_sub_pixel_avg_variance64x32_sse2, 0),
+        SubpelAvgVarianceParams(5, 6, &aom_sub_pixel_avg_variance32x64_sse2, 0),
+        SubpelAvgVarianceParams(5, 5, &aom_sub_pixel_avg_variance32x32_sse2, 0),
+        SubpelAvgVarianceParams(5, 4, &aom_sub_pixel_avg_variance32x16_sse2, 0),
+        SubpelAvgVarianceParams(4, 5, &aom_sub_pixel_avg_variance16x32_sse2, 0),
+        SubpelAvgVarianceParams(4, 4, &aom_sub_pixel_avg_variance16x16_sse2, 0),
+        SubpelAvgVarianceParams(4, 3, &aom_sub_pixel_avg_variance16x8_sse2, 0),
+        SubpelAvgVarianceParams(3, 4, &aom_sub_pixel_avg_variance8x16_sse2, 0),
+        SubpelAvgVarianceParams(3, 3, &aom_sub_pixel_avg_variance8x8_sse2, 0),
+        SubpelAvgVarianceParams(3, 2, &aom_sub_pixel_avg_variance8x4_sse2, 0),
+        SubpelAvgVarianceParams(2, 3, &aom_sub_pixel_avg_variance4x8_sse2, 0),
+        SubpelAvgVarianceParams(2, 2, &aom_sub_pixel_avg_variance4x4_sse2, 0)));
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AvxSubpelVarianceTest,
+    ::testing::Values(
+        SubpelVarianceParams(2, 2, &aom_highbd_8_sub_pixel_variance4x4_sse4_1,
+                             8),
+        SubpelVarianceParams(2, 2, &aom_highbd_10_sub_pixel_variance4x4_sse4_1,
+                             10),
+        SubpelVarianceParams(2, 2, &aom_highbd_12_sub_pixel_variance4x4_sse4_1,
+                             12)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AvxSubpelAvgVarianceTest,
+    ::testing::Values(
+        SubpelAvgVarianceParams(2, 2,
+                                &aom_highbd_8_sub_pixel_avg_variance4x4_sse4_1,
+                                8),
+        SubpelAvgVarianceParams(2, 2,
+                                &aom_highbd_10_sub_pixel_avg_variance4x4_sse4_1,
+                                10),
+        SubpelAvgVarianceParams(2, 2,
+                                &aom_highbd_12_sub_pixel_avg_variance4x4_sse4_1,
+                                12)));
+#endif  // HAVE_SSE4_1
+
+/* TODO(debargha): This test does not support the highbd version
+INSTANTIATE_TEST_CASE_P(
+    SSE2, AvxHBDMseTest,
+    ::testing::Values(MseParams(4, 4, &aom_highbd_12_mse16x16_sse2),
+                      MseParams(4, 3, &aom_highbd_12_mse16x8_sse2),
+                      MseParams(3, 4, &aom_highbd_12_mse8x16_sse2),
+                      MseParams(3, 3, &aom_highbd_12_mse8x8_sse2),
+                      MseParams(4, 4, &aom_highbd_10_mse16x16_sse2),
+                      MseParams(4, 3, &aom_highbd_10_mse16x8_sse2),
+                      MseParams(3, 4, &aom_highbd_10_mse8x16_sse2),
+                      MseParams(3, 3, &aom_highbd_10_mse8x8_sse2),
+                      MseParams(4, 4, &aom_highbd_8_mse16x16_sse2),
+                      MseParams(4, 3, &aom_highbd_8_mse16x8_sse2),
+                      MseParams(3, 4, &aom_highbd_8_mse8x16_sse2),
+                      MseParams(3, 3, &aom_highbd_8_mse8x8_sse2)));
+*/
+
+const VarianceParams kArrayHBDVariance_sse2[] = {
+  VarianceParams(7, 7, &aom_highbd_12_variance128x128_sse2, 12),
+  VarianceParams(7, 6, &aom_highbd_12_variance128x64_sse2, 12),
+  VarianceParams(6, 7, &aom_highbd_12_variance64x128_sse2, 12),
+  VarianceParams(6, 6, &aom_highbd_12_variance64x64_sse2, 12),
+  VarianceParams(6, 5, &aom_highbd_12_variance64x32_sse2, 12),
+  VarianceParams(5, 6, &aom_highbd_12_variance32x64_sse2, 12),
+  VarianceParams(5, 5, &aom_highbd_12_variance32x32_sse2, 12),
+  VarianceParams(5, 4, &aom_highbd_12_variance32x16_sse2, 12),
+  VarianceParams(4, 5, &aom_highbd_12_variance16x32_sse2, 12),
+  VarianceParams(4, 4, &aom_highbd_12_variance16x16_sse2, 12),
+  VarianceParams(4, 3, &aom_highbd_12_variance16x8_sse2, 12),
+  VarianceParams(3, 4, &aom_highbd_12_variance8x16_sse2, 12),
+  VarianceParams(3, 3, &aom_highbd_12_variance8x8_sse2, 12),
+  VarianceParams(7, 7, &aom_highbd_10_variance128x128_sse2, 10),
+  VarianceParams(7, 6, &aom_highbd_10_variance128x64_sse2, 10),
+  VarianceParams(6, 7, &aom_highbd_10_variance64x128_sse2, 10),
+  VarianceParams(6, 6, &aom_highbd_10_variance64x64_sse2, 10),
+  VarianceParams(6, 5, &aom_highbd_10_variance64x32_sse2, 10),
+  VarianceParams(5, 6, &aom_highbd_10_variance32x64_sse2, 10),
+  VarianceParams(5, 5, &aom_highbd_10_variance32x32_sse2, 10),
+  VarianceParams(5, 4, &aom_highbd_10_variance32x16_sse2, 10),
+  VarianceParams(4, 5, &aom_highbd_10_variance16x32_sse2, 10),
+  VarianceParams(4, 4, &aom_highbd_10_variance16x16_sse2, 10),
+  VarianceParams(4, 3, &aom_highbd_10_variance16x8_sse2, 10),
+  VarianceParams(3, 4, &aom_highbd_10_variance8x16_sse2, 10),
+  VarianceParams(3, 3, &aom_highbd_10_variance8x8_sse2, 10),
+  VarianceParams(7, 7, &aom_highbd_8_variance128x128_sse2, 8),
+  VarianceParams(7, 6, &aom_highbd_8_variance128x64_sse2, 8),
+  VarianceParams(6, 7, &aom_highbd_8_variance64x128_sse2, 8),
+  VarianceParams(6, 6, &aom_highbd_8_variance64x64_sse2, 8),
+  VarianceParams(6, 5, &aom_highbd_8_variance64x32_sse2, 8),
+  VarianceParams(5, 6, &aom_highbd_8_variance32x64_sse2, 8),
+  VarianceParams(5, 5, &aom_highbd_8_variance32x32_sse2, 8),
+  VarianceParams(5, 4, &aom_highbd_8_variance32x16_sse2, 8),
+  VarianceParams(4, 5, &aom_highbd_8_variance16x32_sse2, 8),
+  VarianceParams(4, 4, &aom_highbd_8_variance16x16_sse2, 8),
+  VarianceParams(4, 3, &aom_highbd_8_variance16x8_sse2, 8),
+  VarianceParams(3, 4, &aom_highbd_8_variance8x16_sse2, 8),
+  VarianceParams(3, 3, &aom_highbd_8_variance8x8_sse2, 8)
+};
+INSTANTIATE_TEST_CASE_P(SSE2, AvxHBDVarianceTest,
+                        ::testing::ValuesIn(kArrayHBDVariance_sse2));
+
+#if HAVE_AVX2
+
+const VarianceParams kArrayHBDVariance_avx2[] = {
+  VarianceParams(7, 7, &aom_highbd_10_variance128x128_avx2, 10),
+  VarianceParams(7, 6, &aom_highbd_10_variance128x64_avx2, 10),
+  VarianceParams(6, 7, &aom_highbd_10_variance64x128_avx2, 10),
+  VarianceParams(6, 6, &aom_highbd_10_variance64x64_avx2, 10),
+  VarianceParams(6, 5, &aom_highbd_10_variance64x32_avx2, 10),
+  VarianceParams(5, 6, &aom_highbd_10_variance32x64_avx2, 10),
+  VarianceParams(5, 5, &aom_highbd_10_variance32x32_avx2, 10),
+  VarianceParams(5, 4, &aom_highbd_10_variance32x16_avx2, 10),
+  VarianceParams(4, 5, &aom_highbd_10_variance16x32_avx2, 10),
+  VarianceParams(4, 4, &aom_highbd_10_variance16x16_avx2, 10),
+  VarianceParams(4, 3, &aom_highbd_10_variance16x8_avx2, 10),
+  VarianceParams(3, 4, &aom_highbd_10_variance8x16_avx2, 10),
+  VarianceParams(3, 3, &aom_highbd_10_variance8x8_avx2, 10)
+};
+
+INSTANTIATE_TEST_CASE_P(AVX2, AvxHBDVarianceTest,
+                        ::testing::ValuesIn(kArrayHBDVariance_avx2));
+#endif  // HAVE_AVX2
+
+const SubpelVarianceParams kArrayHBDSubpelVariance_sse2[] = {
+  SubpelVarianceParams(6, 6, &aom_highbd_12_sub_pixel_variance64x64_sse2, 12),
+  SubpelVarianceParams(6, 5, &aom_highbd_12_sub_pixel_variance64x32_sse2, 12),
+  SubpelVarianceParams(5, 6, &aom_highbd_12_sub_pixel_variance32x64_sse2, 12),
+  SubpelVarianceParams(5, 5, &aom_highbd_12_sub_pixel_variance32x32_sse2, 12),
+  SubpelVarianceParams(5, 4, &aom_highbd_12_sub_pixel_variance32x16_sse2, 12),
+  SubpelVarianceParams(4, 5, &aom_highbd_12_sub_pixel_variance16x32_sse2, 12),
+  SubpelVarianceParams(4, 4, &aom_highbd_12_sub_pixel_variance16x16_sse2, 12),
+  SubpelVarianceParams(4, 3, &aom_highbd_12_sub_pixel_variance16x8_sse2, 12),
+  SubpelVarianceParams(3, 4, &aom_highbd_12_sub_pixel_variance8x16_sse2, 12),
+  SubpelVarianceParams(3, 3, &aom_highbd_12_sub_pixel_variance8x8_sse2, 12),
+  SubpelVarianceParams(3, 2, &aom_highbd_12_sub_pixel_variance8x4_sse2, 12),
+  SubpelVarianceParams(6, 6, &aom_highbd_10_sub_pixel_variance64x64_sse2, 10),
+  SubpelVarianceParams(6, 5, &aom_highbd_10_sub_pixel_variance64x32_sse2, 10),
+  SubpelVarianceParams(5, 6, &aom_highbd_10_sub_pixel_variance32x64_sse2, 10),
+  SubpelVarianceParams(5, 5, &aom_highbd_10_sub_pixel_variance32x32_sse2, 10),
+  SubpelVarianceParams(5, 4, &aom_highbd_10_sub_pixel_variance32x16_sse2, 10),
+  SubpelVarianceParams(4, 5, &aom_highbd_10_sub_pixel_variance16x32_sse2, 10),
+  SubpelVarianceParams(4, 4, &aom_highbd_10_sub_pixel_variance16x16_sse2, 10),
+  SubpelVarianceParams(4, 3, &aom_highbd_10_sub_pixel_variance16x8_sse2, 10),
+  SubpelVarianceParams(3, 4, &aom_highbd_10_sub_pixel_variance8x16_sse2, 10),
+  SubpelVarianceParams(3, 3, &aom_highbd_10_sub_pixel_variance8x8_sse2, 10),
+  SubpelVarianceParams(3, 2, &aom_highbd_10_sub_pixel_variance8x4_sse2, 10),
+  SubpelVarianceParams(6, 6, &aom_highbd_8_sub_pixel_variance64x64_sse2, 8),
+  SubpelVarianceParams(6, 5, &aom_highbd_8_sub_pixel_variance64x32_sse2, 8),
+  SubpelVarianceParams(5, 6, &aom_highbd_8_sub_pixel_variance32x64_sse2, 8),
+  SubpelVarianceParams(5, 5, &aom_highbd_8_sub_pixel_variance32x32_sse2, 8),
+  SubpelVarianceParams(5, 4, &aom_highbd_8_sub_pixel_variance32x16_sse2, 8),
+  SubpelVarianceParams(4, 5, &aom_highbd_8_sub_pixel_variance16x32_sse2, 8),
+  SubpelVarianceParams(4, 4, &aom_highbd_8_sub_pixel_variance16x16_sse2, 8),
+  SubpelVarianceParams(4, 3, &aom_highbd_8_sub_pixel_variance16x8_sse2, 8),
+  SubpelVarianceParams(3, 4, &aom_highbd_8_sub_pixel_variance8x16_sse2, 8),
+  SubpelVarianceParams(3, 3, &aom_highbd_8_sub_pixel_variance8x8_sse2, 8),
+  SubpelVarianceParams(3, 2, &aom_highbd_8_sub_pixel_variance8x4_sse2, 8)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, AvxHBDSubpelVarianceTest,
+                        ::testing::ValuesIn(kArrayHBDSubpelVariance_sse2));
+
+const SubpelAvgVarianceParams kArrayHBDSubpelAvgVariance_sse2[] = {
+  SubpelAvgVarianceParams(6, 6, &aom_highbd_12_sub_pixel_avg_variance64x64_sse2,
+                          12),
+  SubpelAvgVarianceParams(6, 5, &aom_highbd_12_sub_pixel_avg_variance64x32_sse2,
+                          12),
+  SubpelAvgVarianceParams(5, 6, &aom_highbd_12_sub_pixel_avg_variance32x64_sse2,
+                          12),
+  SubpelAvgVarianceParams(5, 5, &aom_highbd_12_sub_pixel_avg_variance32x32_sse2,
+                          12),
+  SubpelAvgVarianceParams(5, 4, &aom_highbd_12_sub_pixel_avg_variance32x16_sse2,
+                          12),
+  SubpelAvgVarianceParams(4, 5, &aom_highbd_12_sub_pixel_avg_variance16x32_sse2,
+                          12),
+  SubpelAvgVarianceParams(4, 4, &aom_highbd_12_sub_pixel_avg_variance16x16_sse2,
+                          12),
+  SubpelAvgVarianceParams(4, 3, &aom_highbd_12_sub_pixel_avg_variance16x8_sse2,
+                          12),
+  SubpelAvgVarianceParams(3, 4, &aom_highbd_12_sub_pixel_avg_variance8x16_sse2,
+                          12),
+  SubpelAvgVarianceParams(3, 3, &aom_highbd_12_sub_pixel_avg_variance8x8_sse2,
+                          12),
+  SubpelAvgVarianceParams(3, 2, &aom_highbd_12_sub_pixel_avg_variance8x4_sse2,
+                          12),
+  SubpelAvgVarianceParams(6, 6, &aom_highbd_10_sub_pixel_avg_variance64x64_sse2,
+                          10),
+  SubpelAvgVarianceParams(6, 5, &aom_highbd_10_sub_pixel_avg_variance64x32_sse2,
+                          10),
+  SubpelAvgVarianceParams(5, 6, &aom_highbd_10_sub_pixel_avg_variance32x64_sse2,
+                          10),
+  SubpelAvgVarianceParams(5, 5, &aom_highbd_10_sub_pixel_avg_variance32x32_sse2,
+                          10),
+  SubpelAvgVarianceParams(5, 4, &aom_highbd_10_sub_pixel_avg_variance32x16_sse2,
+                          10),
+  SubpelAvgVarianceParams(4, 5, &aom_highbd_10_sub_pixel_avg_variance16x32_sse2,
+                          10),
+  SubpelAvgVarianceParams(4, 4, &aom_highbd_10_sub_pixel_avg_variance16x16_sse2,
+                          10),
+  SubpelAvgVarianceParams(4, 3, &aom_highbd_10_sub_pixel_avg_variance16x8_sse2,
+                          10),
+  SubpelAvgVarianceParams(3, 4, &aom_highbd_10_sub_pixel_avg_variance8x16_sse2,
+                          10),
+  SubpelAvgVarianceParams(3, 3, &aom_highbd_10_sub_pixel_avg_variance8x8_sse2,
+                          10),
+  SubpelAvgVarianceParams(3, 2, &aom_highbd_10_sub_pixel_avg_variance8x4_sse2,
+                          10),
+  SubpelAvgVarianceParams(6, 6, &aom_highbd_8_sub_pixel_avg_variance64x64_sse2,
+                          8),
+  SubpelAvgVarianceParams(6, 5, &aom_highbd_8_sub_pixel_avg_variance64x32_sse2,
+                          8),
+  SubpelAvgVarianceParams(5, 6, &aom_highbd_8_sub_pixel_avg_variance32x64_sse2,
+                          8),
+  SubpelAvgVarianceParams(5, 5, &aom_highbd_8_sub_pixel_avg_variance32x32_sse2,
+                          8),
+  SubpelAvgVarianceParams(5, 4, &aom_highbd_8_sub_pixel_avg_variance32x16_sse2,
+                          8),
+  SubpelAvgVarianceParams(4, 5, &aom_highbd_8_sub_pixel_avg_variance16x32_sse2,
+                          8),
+  SubpelAvgVarianceParams(4, 4, &aom_highbd_8_sub_pixel_avg_variance16x16_sse2,
+                          8),
+  SubpelAvgVarianceParams(4, 3, &aom_highbd_8_sub_pixel_avg_variance16x8_sse2,
+                          8),
+  SubpelAvgVarianceParams(3, 4, &aom_highbd_8_sub_pixel_avg_variance8x16_sse2,
+                          8),
+  SubpelAvgVarianceParams(3, 3, &aom_highbd_8_sub_pixel_avg_variance8x8_sse2,
+                          8),
+  SubpelAvgVarianceParams(3, 2, &aom_highbd_8_sub_pixel_avg_variance8x4_sse2, 8)
+};
+
+INSTANTIATE_TEST_CASE_P(SSE2, AvxHBDSubpelAvgVarianceTest,
+                        ::testing::ValuesIn(kArrayHBDSubpelAvgVariance_sse2));
+#endif  // HAVE_SSE2
+
+#if HAVE_SSSE3
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, AvxSubpelVarianceTest,
+    ::testing::Values(
+        SubpelVarianceParams(7, 7, &aom_sub_pixel_variance128x128_ssse3, 0),
+        SubpelVarianceParams(7, 6, &aom_sub_pixel_variance128x64_ssse3, 0),
+        SubpelVarianceParams(6, 7, &aom_sub_pixel_variance64x128_ssse3, 0),
+        SubpelVarianceParams(6, 6, &aom_sub_pixel_variance64x64_ssse3, 0),
+        SubpelVarianceParams(6, 5, &aom_sub_pixel_variance64x32_ssse3, 0),
+        SubpelVarianceParams(5, 6, &aom_sub_pixel_variance32x64_ssse3, 0),
+        SubpelVarianceParams(5, 5, &aom_sub_pixel_variance32x32_ssse3, 0),
+        SubpelVarianceParams(5, 4, &aom_sub_pixel_variance32x16_ssse3, 0),
+        SubpelVarianceParams(4, 5, &aom_sub_pixel_variance16x32_ssse3, 0),
+        SubpelVarianceParams(4, 4, &aom_sub_pixel_variance16x16_ssse3, 0),
+        SubpelVarianceParams(4, 3, &aom_sub_pixel_variance16x8_ssse3, 0),
+        SubpelVarianceParams(3, 4, &aom_sub_pixel_variance8x16_ssse3, 0),
+        SubpelVarianceParams(3, 3, &aom_sub_pixel_variance8x8_ssse3, 0),
+        SubpelVarianceParams(3, 2, &aom_sub_pixel_variance8x4_ssse3, 0),
+        SubpelVarianceParams(2, 3, &aom_sub_pixel_variance4x8_ssse3, 0),
+        SubpelVarianceParams(2, 2, &aom_sub_pixel_variance4x4_ssse3, 0)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, AvxSubpelAvgVarianceTest,
+    ::testing::Values(
+        SubpelAvgVarianceParams(7, 7, &aom_sub_pixel_avg_variance128x128_ssse3,
+                                0),
+        SubpelAvgVarianceParams(7, 6, &aom_sub_pixel_avg_variance128x64_ssse3,
+                                0),
+        SubpelAvgVarianceParams(6, 7, &aom_sub_pixel_avg_variance64x128_ssse3,
+                                0),
+        SubpelAvgVarianceParams(6, 6, &aom_sub_pixel_avg_variance64x64_ssse3,
+                                0),
+        SubpelAvgVarianceParams(6, 5, &aom_sub_pixel_avg_variance64x32_ssse3,
+                                0),
+        SubpelAvgVarianceParams(5, 6, &aom_sub_pixel_avg_variance32x64_ssse3,
+                                0),
+        SubpelAvgVarianceParams(5, 5, &aom_sub_pixel_avg_variance32x32_ssse3,
+                                0),
+        SubpelAvgVarianceParams(5, 4, &aom_sub_pixel_avg_variance32x16_ssse3,
+                                0),
+        SubpelAvgVarianceParams(4, 5, &aom_sub_pixel_avg_variance16x32_ssse3,
+                                0),
+        SubpelAvgVarianceParams(4, 4, &aom_sub_pixel_avg_variance16x16_ssse3,
+                                0),
+        SubpelAvgVarianceParams(4, 3, &aom_sub_pixel_avg_variance16x8_ssse3, 0),
+        SubpelAvgVarianceParams(3, 4, &aom_sub_pixel_avg_variance8x16_ssse3, 0),
+        SubpelAvgVarianceParams(3, 3, &aom_sub_pixel_avg_variance8x8_ssse3, 0),
+        SubpelAvgVarianceParams(3, 2, &aom_sub_pixel_avg_variance8x4_ssse3, 0),
+        SubpelAvgVarianceParams(2, 3, &aom_sub_pixel_avg_variance4x8_ssse3, 0),
+        SubpelAvgVarianceParams(2, 2, &aom_sub_pixel_avg_variance4x4_ssse3,
+                                0)));
+
+INSTANTIATE_TEST_CASE_P(
+    SSSE3, AvxJntSubpelAvgVarianceTest,
+    ::testing::Values(
+        JntSubpelAvgVarianceParams(6, 6,
+                                   &aom_jnt_sub_pixel_avg_variance64x64_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(6, 5,
+                                   &aom_jnt_sub_pixel_avg_variance64x32_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(5, 6,
+                                   &aom_jnt_sub_pixel_avg_variance32x64_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(5, 5,
+                                   &aom_jnt_sub_pixel_avg_variance32x32_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(5, 4,
+                                   &aom_jnt_sub_pixel_avg_variance32x16_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(4, 5,
+                                   &aom_jnt_sub_pixel_avg_variance16x32_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(4, 4,
+                                   &aom_jnt_sub_pixel_avg_variance16x16_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(4, 3,
+                                   &aom_jnt_sub_pixel_avg_variance16x8_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(3, 4,
+                                   &aom_jnt_sub_pixel_avg_variance8x16_ssse3,
+                                   0),
+        JntSubpelAvgVarianceParams(3, 3,
+                                   &aom_jnt_sub_pixel_avg_variance8x8_ssse3, 0),
+        JntSubpelAvgVarianceParams(3, 2,
+                                   &aom_jnt_sub_pixel_avg_variance8x4_ssse3, 0),
+        JntSubpelAvgVarianceParams(2, 3,
+                                   &aom_jnt_sub_pixel_avg_variance4x8_ssse3, 0),
+        JntSubpelAvgVarianceParams(2, 2,
+                                   &aom_jnt_sub_pixel_avg_variance4x4_ssse3,
+                                   0)));
+#endif  // HAVE_SSSE3
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AvxObmcSubpelVarianceTest,
+    ::testing::Values(
+        ObmcSubpelVarianceParams(7, 7,
+                                 &aom_obmc_sub_pixel_variance128x128_sse4_1, 0),
+        ObmcSubpelVarianceParams(7, 6,
+                                 &aom_obmc_sub_pixel_variance128x64_sse4_1, 0),
+        ObmcSubpelVarianceParams(6, 7,
+                                 &aom_obmc_sub_pixel_variance64x128_sse4_1, 0),
+        ObmcSubpelVarianceParams(6, 6, &aom_obmc_sub_pixel_variance64x64_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(6, 5, &aom_obmc_sub_pixel_variance64x32_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(5, 6, &aom_obmc_sub_pixel_variance32x64_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(5, 5, &aom_obmc_sub_pixel_variance32x32_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(5, 4, &aom_obmc_sub_pixel_variance32x16_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(4, 5, &aom_obmc_sub_pixel_variance16x32_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(4, 4, &aom_obmc_sub_pixel_variance16x16_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(4, 3, &aom_obmc_sub_pixel_variance16x8_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(3, 4, &aom_obmc_sub_pixel_variance8x16_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(3, 3, &aom_obmc_sub_pixel_variance8x8_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(3, 2, &aom_obmc_sub_pixel_variance8x4_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(2, 3, &aom_obmc_sub_pixel_variance4x8_sse4_1,
+                                 0),
+        ObmcSubpelVarianceParams(2, 2, &aom_obmc_sub_pixel_variance4x4_sse4_1,
+                                 0)));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+INSTANTIATE_TEST_CASE_P(AVX2, AvxMseTest,
+                        ::testing::Values(MseParams(4, 4, &aom_mse16x16_avx2)));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AvxVarianceTest,
+    ::testing::Values(VarianceParams(7, 7, &aom_variance128x128_avx2),
+                      VarianceParams(7, 6, &aom_variance128x64_avx2),
+                      VarianceParams(6, 7, &aom_variance64x128_avx2),
+                      VarianceParams(6, 6, &aom_variance64x64_avx2),
+                      VarianceParams(6, 5, &aom_variance64x32_avx2),
+                      VarianceParams(6, 4, &aom_variance64x16_avx2),
+                      VarianceParams(5, 6, &aom_variance32x64_avx2),
+                      VarianceParams(5, 5, &aom_variance32x32_avx2),
+                      VarianceParams(5, 4, &aom_variance32x16_avx2),
+                      VarianceParams(5, 3, &aom_variance32x8_avx2),
+                      VarianceParams(4, 6, &aom_variance16x64_avx2),
+                      VarianceParams(4, 5, &aom_variance16x32_avx2),
+                      VarianceParams(4, 4, &aom_variance16x16_avx2),
+                      VarianceParams(4, 3, &aom_variance16x8_avx2),
+                      VarianceParams(4, 2, &aom_variance16x4_avx2)));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AvxSubpelVarianceTest,
+    ::testing::Values(
+        SubpelVarianceParams(7, 7, &aom_sub_pixel_variance128x128_avx2, 0),
+        SubpelVarianceParams(7, 6, &aom_sub_pixel_variance128x64_avx2, 0),
+        SubpelVarianceParams(6, 7, &aom_sub_pixel_variance64x128_avx2, 0),
+        SubpelVarianceParams(6, 6, &aom_sub_pixel_variance64x64_avx2, 0),
+        SubpelVarianceParams(6, 5, &aom_sub_pixel_variance64x32_avx2, 0),
+        SubpelVarianceParams(5, 6, &aom_sub_pixel_variance32x64_avx2, 0),
+        SubpelVarianceParams(5, 5, &aom_sub_pixel_variance32x32_avx2, 0),
+        SubpelVarianceParams(5, 4, &aom_sub_pixel_variance32x16_avx2, 0)));
+
+INSTANTIATE_TEST_CASE_P(
+    AVX2, AvxSubpelAvgVarianceTest,
+    ::testing::Values(
+        SubpelAvgVarianceParams(7, 7, &aom_sub_pixel_avg_variance128x128_avx2,
+                                0),
+        SubpelAvgVarianceParams(7, 6, &aom_sub_pixel_avg_variance128x64_avx2,
+                                0),
+        SubpelAvgVarianceParams(6, 7, &aom_sub_pixel_avg_variance64x128_avx2,
+                                0),
+        SubpelAvgVarianceParams(6, 6, &aom_sub_pixel_avg_variance64x64_avx2, 0),
+        SubpelAvgVarianceParams(6, 5, &aom_sub_pixel_avg_variance64x32_avx2, 0),
+        SubpelAvgVarianceParams(5, 6, &aom_sub_pixel_avg_variance32x64_avx2, 0),
+        SubpelAvgVarianceParams(5, 5, &aom_sub_pixel_avg_variance32x32_avx2, 0),
+        SubpelAvgVarianceParams(5, 4, &aom_sub_pixel_avg_variance32x16_avx2,
+                                0)));
+#endif  // HAVE_AVX2
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(NEON, AvxSseTest,
+                        ::testing::Values(SseParams(2, 2,
+                                                    &aom_get4x4sse_cs_neon)));
+
+INSTANTIATE_TEST_CASE_P(NEON, AvxMseTest,
+                        ::testing::Values(MseParams(4, 4, &aom_mse16x16_neon)));
+
+INSTANTIATE_TEST_CASE_P(
+    NEON, AvxVarianceTest,
+    ::testing::Values(VarianceParams(6, 6, &aom_variance64x64_neon),
+                      VarianceParams(6, 5, &aom_variance64x32_neon),
+                      VarianceParams(5, 6, &aom_variance32x64_neon),
+                      VarianceParams(5, 5, &aom_variance32x32_neon),
+                      VarianceParams(4, 4, &aom_variance16x16_neon),
+                      VarianceParams(4, 3, &aom_variance16x8_neon),
+                      VarianceParams(3, 4, &aom_variance8x16_neon),
+                      VarianceParams(3, 3, &aom_variance8x8_neon)));
+
+INSTANTIATE_TEST_CASE_P(
+    NEON, AvxSubpelVarianceTest,
+    ::testing::Values(
+        SubpelVarianceParams(6, 6, &aom_sub_pixel_variance64x64_neon, 0),
+        SubpelVarianceParams(5, 5, &aom_sub_pixel_variance32x32_neon, 0),
+        SubpelVarianceParams(4, 4, &aom_sub_pixel_variance16x16_neon, 0),
+        SubpelVarianceParams(3, 3, &aom_sub_pixel_variance8x8_neon, 0)));
+#endif  // HAVE_NEON
+
+#if HAVE_MSA
+INSTANTIATE_TEST_CASE_P(MSA, SumOfSquaresTest,
+                        ::testing::Values(aom_get_mb_ss_msa));
+
+INSTANTIATE_TEST_CASE_P(MSA, AvxSseTest,
+                        ::testing::Values(SseParams(2, 2,
+                                                    &aom_get4x4sse_cs_msa)));
+
+INSTANTIATE_TEST_CASE_P(MSA, AvxMseTest,
+                        ::testing::Values(MseParams(4, 4, &aom_mse16x16_msa),
+                                          MseParams(4, 3, &aom_mse16x8_msa),
+                                          MseParams(3, 4, &aom_mse8x16_msa),
+                                          MseParams(3, 3, &aom_mse8x8_msa)));
+
+INSTANTIATE_TEST_CASE_P(
+    MSA, AvxVarianceTest,
+    ::testing::Values(VarianceParams(6, 6, &aom_variance64x64_msa),
+                      VarianceParams(6, 5, &aom_variance64x32_msa),
+                      VarianceParams(5, 6, &aom_variance32x64_msa),
+                      VarianceParams(5, 5, &aom_variance32x32_msa),
+                      VarianceParams(5, 4, &aom_variance32x16_msa),
+                      VarianceParams(4, 5, &aom_variance16x32_msa),
+                      VarianceParams(4, 4, &aom_variance16x16_msa),
+                      VarianceParams(4, 3, &aom_variance16x8_msa),
+                      VarianceParams(3, 4, &aom_variance8x16_msa),
+                      VarianceParams(3, 3, &aom_variance8x8_msa),
+                      VarianceParams(3, 2, &aom_variance8x4_msa),
+                      VarianceParams(2, 3, &aom_variance4x8_msa),
+                      VarianceParams(2, 2, &aom_variance4x4_msa)));
+
+INSTANTIATE_TEST_CASE_P(
+    MSA, AvxSubpelVarianceTest,
+    ::testing::Values(
+        SubpelVarianceParams(2, 2, &aom_sub_pixel_variance4x4_msa, 0),
+        SubpelVarianceParams(2, 3, &aom_sub_pixel_variance4x8_msa, 0),
+        SubpelVarianceParams(3, 2, &aom_sub_pixel_variance8x4_msa, 0),
+        SubpelVarianceParams(3, 3, &aom_sub_pixel_variance8x8_msa, 0),
+        SubpelVarianceParams(3, 4, &aom_sub_pixel_variance8x16_msa, 0),
+        SubpelVarianceParams(4, 3, &aom_sub_pixel_variance16x8_msa, 0),
+        SubpelVarianceParams(4, 4, &aom_sub_pixel_variance16x16_msa, 0),
+        SubpelVarianceParams(4, 5, &aom_sub_pixel_variance16x32_msa, 0),
+        SubpelVarianceParams(5, 4, &aom_sub_pixel_variance32x16_msa, 0),
+        SubpelVarianceParams(5, 5, &aom_sub_pixel_variance32x32_msa, 0),
+        SubpelVarianceParams(5, 6, &aom_sub_pixel_variance32x64_msa, 0),
+        SubpelVarianceParams(6, 5, &aom_sub_pixel_variance64x32_msa, 0),
+        SubpelVarianceParams(6, 6, &aom_sub_pixel_variance64x64_msa, 0)));
+
+INSTANTIATE_TEST_CASE_P(
+    MSA, AvxSubpelAvgVarianceTest,
+    ::testing::Values(
+        SubpelAvgVarianceParams(6, 6, &aom_sub_pixel_avg_variance64x64_msa, 0),
+        SubpelAvgVarianceParams(6, 5, &aom_sub_pixel_avg_variance64x32_msa, 0),
+        SubpelAvgVarianceParams(5, 6, &aom_sub_pixel_avg_variance32x64_msa, 0),
+        SubpelAvgVarianceParams(5, 5, &aom_sub_pixel_avg_variance32x32_msa, 0),
+        SubpelAvgVarianceParams(5, 4, &aom_sub_pixel_avg_variance32x16_msa, 0),
+        SubpelAvgVarianceParams(4, 5, &aom_sub_pixel_avg_variance16x32_msa, 0),
+        SubpelAvgVarianceParams(4, 4, &aom_sub_pixel_avg_variance16x16_msa, 0),
+        SubpelAvgVarianceParams(4, 3, &aom_sub_pixel_avg_variance16x8_msa, 0),
+        SubpelAvgVarianceParams(3, 4, &aom_sub_pixel_avg_variance8x16_msa, 0),
+        SubpelAvgVarianceParams(3, 3, &aom_sub_pixel_avg_variance8x8_msa, 0),
+        SubpelAvgVarianceParams(3, 2, &aom_sub_pixel_avg_variance8x4_msa, 0),
+        SubpelAvgVarianceParams(2, 3, &aom_sub_pixel_avg_variance4x8_msa, 0),
+        SubpelAvgVarianceParams(2, 2, &aom_sub_pixel_avg_variance4x4_msa, 0)));
+#endif  // HAVE_MSA
+}  // namespace
diff --git a/third_party/aom/test/video_source.h b/third_party/aom/test/video_source.h
new file mode 100644
index 0000000000..3c1c5e559e
--- /dev/null
+++ b/third_party/aom/test/video_source.h
@@ -0,0 +1,259 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_VIDEO_SOURCE_H_
+#define AOM_TEST_VIDEO_SOURCE_H_
+
+#if defined(_WIN32)
+#undef NOMINMAX
+#define NOMINMAX
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+#endif
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+#include "test/acm_random.h"
+#include "aom/aom_encoder.h"
+
+namespace libaom_test {
+
+// Helper macros to ensure LIBAOM_TEST_DATA_PATH is a quoted string.
+// These are undefined right below GetDataPath
+// NOTE: LIBAOM_TEST_DATA_PATH MUST NOT be a quoted string before
+// Stringification or the GetDataPath will fail at runtime
+#define TO_STRING(S) #S
+#define STRINGIFY(S) TO_STRING(S)
+
+// A simple function to encapsulate cross platform retrieval of test data path
+static std::string GetDataPath() {
+  const char *const data_path = getenv("LIBAOM_TEST_DATA_PATH");
+  if (data_path == NULL) {
+#ifdef LIBAOM_TEST_DATA_PATH
+    // In some environments, we cannot set environment variables
+    // Instead, we set the data path by using a preprocessor symbol
+    // which can be set from make files
+    return STRINGIFY(LIBAOM_TEST_DATA_PATH);
+#else
+    return ".";
+#endif
+  }
+  return data_path;
+}
+
+// Undefining stringification macros because they are not used elsewhere
+#undef TO_STRING
+#undef STRINGIFY
+
+inline FILE *OpenTestDataFile(const std::string &file_name) {
+  const std::string path_to_source = GetDataPath() + "/" + file_name;
+  return fopen(path_to_source.c_str(), "rb");
+}
+
+static FILE *GetTempOutFile(std::string *file_name) {
+  file_name->clear();
+#if defined(_WIN32)
+  char fname[MAX_PATH];
+  char tmppath[MAX_PATH];
+  if (GetTempPathA(MAX_PATH, tmppath)) {
+    // Assume for now that the filename generated is unique per process
+    if (GetTempFileNameA(tmppath, "lvx", 0, fname)) {
+      file_name->assign(fname);
+      return fopen(fname, "wb+");
+    }
+  }
+  return NULL;
+#else
+  char name_template[] = "/tmp/libaomtest.XXXXXX";
+  const int fd = mkstemp(name_template);
+  *file_name = name_template;
+  return fdopen(fd, "wb+");
+#endif
+}
+
+class TempOutFile {
+ public:
+  TempOutFile() { file_ = GetTempOutFile(&file_name_); }
+  ~TempOutFile() {
+    CloseFile();
+    if (!file_name_.empty()) {
+      EXPECT_EQ(0, remove(file_name_.c_str()));
+    }
+  }
+  FILE *file() { return file_; }
+  const std::string &file_name() { return file_name_; }
+
+ protected:
+  void CloseFile() {
+    if (file_) {
+      fclose(file_);
+      file_ = NULL;
+    }
+  }
+  FILE *file_;
+  std::string file_name_;
+};
+
+// Abstract base class for test video sources, which provide a stream of
+// aom_image_t images with associated timestamps and duration.
+class VideoSource {
+ public:
+  virtual ~VideoSource() {}
+
+  // Prepare the stream for reading, rewind/open as necessary.
+  virtual void Begin() = 0;
+
+  // Advance the cursor to the next frame
+  virtual void Next() = 0;
+
+  // Get the current video frame, or NULL on End-Of-Stream.
+  virtual aom_image_t *img() const = 0;
+
+  // Get the presentation timestamp of the current frame.
+  virtual aom_codec_pts_t pts() const = 0;
+
+  // Get the current frame's duration
+  virtual unsigned long duration() const = 0;
+
+  // Get the timebase for the stream
+  virtual aom_rational_t timebase() const = 0;
+
+  // Get the current frame counter, starting at 0.
+  virtual unsigned int frame() const = 0;
+
+  // Get the current file limit.
+  virtual unsigned int limit() const = 0;
+};
+
+class DummyVideoSource : public VideoSource {
+ public:
+  DummyVideoSource()
+      : img_(NULL), limit_(100), width_(80), height_(64),
+        format_(AOM_IMG_FMT_I420) {
+    ReallocImage();
+  }
+
+  virtual ~DummyVideoSource() { aom_img_free(img_); }
+
+  virtual void Begin() {
+    frame_ = 0;
+    FillFrame();
+  }
+
+  virtual void Next() {
+    ++frame_;
+    FillFrame();
+  }
+
+  virtual aom_image_t *img() const { return (frame_ < limit_) ? img_ : NULL; }
+
+  // Models a stream where Timebase = 1/FPS, so pts == frame.
+  virtual aom_codec_pts_t pts() const { return frame_; }
+
+  virtual unsigned long duration() const { return 1; }
+
+  virtual aom_rational_t timebase() const {
+    const aom_rational_t t = { 1, 30 };
+    return t;
+  }
+
+  virtual unsigned int frame() const { return frame_; }
+
+  virtual unsigned int limit() const { return limit_; }
+
+  void set_limit(unsigned int limit) { limit_ = limit; }
+
+  void SetSize(unsigned int width, unsigned int height) {
+    if (width != width_ || height != height_) {
+      width_ = width;
+      height_ = height;
+      ReallocImage();
+    }
+  }
+
+  void SetImageFormat(aom_img_fmt_t format) {
+    if (format_ != format) {
+      format_ = format;
+      ReallocImage();
+    }
+  }
+
+ protected:
+  virtual void FillFrame() {
+    if (img_) memset(img_->img_data, 0, raw_sz_);
+  }
+
+  void ReallocImage() {
+    aom_img_free(img_);
+    img_ = aom_img_alloc(NULL, format_, width_, height_, 32);
+    raw_sz_ = ((img_->w + 31) & ~31) * img_->h * img_->bps / 8;
+  }
+
+  aom_image_t *img_;
+  size_t raw_sz_;
+  unsigned int limit_;
+  unsigned int frame_;
+  unsigned int width_;
+  unsigned int height_;
+  aom_img_fmt_t format_;
+};
+
+class RandomVideoSource : public DummyVideoSource {
+ public:
+  RandomVideoSource(int seed = ACMRandom::DeterministicSeed())
+      : rnd_(seed), seed_(seed) {}
+
+ protected:
+  // Reset the RNG to get a matching stream for the second pass
+  virtual void Begin() {
+    frame_ = 0;
+    rnd_.Reset(seed_);
+    FillFrame();
+  }
+
+  // 15 frames of noise, followed by 15 static frames. Reset to 0 rather
+  // than holding previous frames to encourage keyframes to be thrown.
+  virtual void FillFrame() {
+    if (img_) {
+      if (frame_ % 30 < 15)
+        for (size_t i = 0; i < raw_sz_; ++i) img_->img_data[i] = rnd_.Rand8();
+      else
+        memset(img_->img_data, 0, raw_sz_);
+    }
+  }
+
+  ACMRandom rnd_;
+  int seed_;
+};
+
+// Abstract base class for test video sources, which provide a stream of
+// decompressed images to the decoder.
+class CompressedVideoSource {
+ public:
+  virtual ~CompressedVideoSource() {}
+
+  virtual void Init() = 0;
+
+  // Prepare the stream for reading, rewind/open as necessary.
+  virtual void Begin() = 0;
+
+  // Advance the cursor to the next frame
+  virtual void Next() = 0;
+
+  virtual const uint8_t *cxdata() const = 0;
+
+  virtual size_t frame_size() const = 0;
+
+  virtual unsigned int frame_number() const = 0;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_VIDEO_SOURCE_H_
diff --git a/third_party/aom/test/visual_metrics.py b/third_party/aom/test/visual_metrics.py
new file mode 100755
index 0000000000..9055feb334
--- /dev/null
+++ b/third_party/aom/test/visual_metrics.py
@@ -0,0 +1,466 @@
+#!/usr/bin/python
+#
+# Copyright (c) 2016, Alliance for Open Media. All rights reserved
+#
+# This source code is subject to the terms of the BSD 2 Clause License and
+# the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+# was not distributed with this source code in the LICENSE file, you can
+# obtain it at www.aomedia.org/license/software. If the Alliance for Open
+# Media Patent License 1.0 was not distributed with this source code in the
+# PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+#
+
+"""Converts video encoding result data from text files to visualization
+data source."""
+
+__author__ = "jzern@google.com (James Zern),"
+__author__ += "jimbankoski@google.com (Jim Bankoski)"
+
+import fnmatch
+import numpy as np
+import scipy as sp
+import scipy.interpolate
+import os
+import re
+import string
+import sys
+import math
+import warnings
+
+import gviz_api
+
+from os.path import basename
+from os.path import splitext
+
+warnings.simplefilter('ignore', np.RankWarning)
+warnings.simplefilter('ignore', RuntimeWarning)
+
+def bdsnr2(metric_set1, metric_set2):
+  """
+  BJONTEGAARD    Bjontegaard metric calculation adapted
+  Bjontegaard's snr metric allows to compute the average % saving in decibels
+  between two rate-distortion curves [1].  This is an adaptation of that
+  method that fixes inconsistencies when the curve fit operation goes awry
+  by replacing the curve fit function with a Piecewise Cubic Hermite
+  Interpolating Polynomial and then integrating that by evaluating that
+  function at small intervals using the trapezoid method to calculate
+  the integral.
+
+  metric_set1 - list of tuples ( bitrate,  metric ) for first graph
+  metric_set2 - list of tuples ( bitrate,  metric ) for second graph
+  """
+
+  if not metric_set1 or not metric_set2:
+    return 0.0
+
+  try:
+
+    # pchip_interlopate requires keys sorted by x axis. x-axis will
+    # be our metric not the bitrate so sort by metric.
+    metric_set1.sort()
+    metric_set2.sort()
+
+    # Pull the log of the rate and clamped psnr from metric_sets.
+    log_rate1 = [math.log(x[0]) for x in metric_set1]
+    metric1 = [100.0 if x[1] == float('inf') else x[1] for x in metric_set1]
+    log_rate2 = [math.log(x[0]) for x in metric_set2]
+    metric2 = [100.0 if x[1] == float('inf') else x[1] for x in metric_set2]
+
+    # Integration interval.  This metric only works on the area that's
+    # overlapping.   Extrapolation of these things is sketchy so we avoid.
+    min_int = max([min(log_rate1), min(log_rate2)])
+    max_int = min([max(log_rate1), max(log_rate2)])
+
+    # No overlap means no sensible metric possible.
+    if max_int <= min_int:
+      return 0.0
+
+    # Use Piecewise Cubic Hermite Interpolating Polynomial interpolation to
+    # create 100 new samples points separated by interval.
+    lin = np.linspace(min_int, max_int, num=100, retstep=True)
+    interval = lin[1]
+    samples = lin[0]
+    v1 = scipy.interpolate.pchip_interpolate(log_rate1, metric1, samples)
+    v2 = scipy.interpolate.pchip_interpolate(log_rate2, metric2, samples)
+
+    # Calculate the integral using the trapezoid method on the samples.
+    int_v1 = np.trapz(v1, dx=interval)
+    int_v2 = np.trapz(v2, dx=interval)
+
+    # Calculate the average improvement.
+    avg_exp_diff = (int_v2 - int_v1) / (max_int - min_int)
+
+  except (TypeError, ZeroDivisionError, ValueError, np.RankWarning) as e:
+    return 0
+
+  return avg_exp_diff
+
+def bdrate2(metric_set1, metric_set2):
+  """
+  BJONTEGAARD    Bjontegaard metric calculation adapted
+  Bjontegaard's metric allows to compute the average % saving in bitrate
+  between two rate-distortion curves [1].  This is an adaptation of that
+  method that fixes inconsistencies when the curve fit operation goes awry
+  by replacing the curve fit function with a Piecewise Cubic Hermite
+  Interpolating Polynomial and then integrating that by evaluating that
+  function at small intervals using the trapezoid method to calculate
+  the integral.
+
+  metric_set1 - list of tuples ( bitrate,  metric ) for first graph
+  metric_set2 - list of tuples ( bitrate,  metric ) for second graph
+  """
+
+  if not metric_set1 or not metric_set2:
+    return 0.0
+
+  try:
+
+    # pchip_interlopate requires keys sorted by x axis. x-axis will
+    # be our metric not the bitrate so sort by metric.
+    metric_set1.sort(key=lambda tup: tup[1])
+    metric_set2.sort(key=lambda tup: tup[1])
+
+    # Pull the log of the rate and clamped psnr from metric_sets.
+    log_rate1 = [math.log(x[0]) for x in metric_set1]
+    metric1 = [100.0 if x[1] == float('inf') else x[1] for x in metric_set1]
+    log_rate2 = [math.log(x[0]) for x in metric_set2]
+    metric2 = [100.0 if x[1] == float('inf') else x[1] for x in metric_set2]
+
+    # Integration interval.  This metric only works on the area that's
+    # overlapping.   Extrapolation of these things is sketchy so we avoid.
+    min_int = max([min(metric1), min(metric2)])
+    max_int = min([max(metric1), max(metric2)])
+
+    # No overlap means no sensible metric possible.
+    if max_int <= min_int:
+      return 0.0
+
+    # Use Piecewise Cubic Hermite Interpolating Polynomial interpolation to
+    # create 100 new samples points separated by interval.
+    lin = np.linspace(min_int, max_int, num=100, retstep=True)
+    interval = lin[1]
+    samples = lin[0]
+    v1 = scipy.interpolate.pchip_interpolate(metric1, log_rate1, samples)
+    v2 = scipy.interpolate.pchip_interpolate(metric2, log_rate2, samples)
+
+    # Calculate the integral using the trapezoid method on the samples.
+    int_v1 = np.trapz(v1, dx=interval)
+    int_v2 = np.trapz(v2, dx=interval)
+
+    # Calculate the average improvement.
+    avg_exp_diff = (int_v2 - int_v1) / (max_int - min_int)
+
+  except (TypeError, ZeroDivisionError, ValueError, np.RankWarning) as e:
+    return 0
+
+  # Convert to a percentage.
+  avg_diff = (math.exp(avg_exp_diff) - 1) * 100
+
+  return avg_diff
+
+
+
+def FillForm(string_for_substitution, dictionary_of_vars):
+  """
+  This function substitutes all matches of the command string //%% ... %%//
+  with the variable represented by ...  .
+  """
+  return_string = string_for_substitution
+  for i in re.findall("//%%(.*)%%//", string_for_substitution):
+    return_string = re.sub("//%%" + i + "%%//", dictionary_of_vars[i],
+                           return_string)
+  return return_string
+
+
+def HasMetrics(line):
+  """
+  The metrics files produced by aomenc are started with a B for headers.
+  """
+  # If the first char of the first word on the line is a digit
+  if len(line) == 0:
+    return False
+  if len(line.split()) == 0:
+    return False
+  if line.split()[0][0:1].isdigit():
+    return True
+  return False
+
+def GetMetrics(file_name):
+  metric_file = open(file_name, "r")
+  return metric_file.readline().split();
+
+def ParseMetricFile(file_name, metric_column):
+  metric_set1 = set([])
+  metric_file = open(file_name, "r")
+  for line in metric_file:
+    metrics = string.split(line)
+    if HasMetrics(line):
+      if metric_column < len(metrics):
+        try:
+          tuple = float(metrics[0]), float(metrics[metric_column])
+        except:
+          tuple = float(metrics[0]), 0
+      else:
+        tuple = float(metrics[0]), 0
+      metric_set1.add(tuple)
+  metric_set1_sorted = sorted(metric_set1)
+  return metric_set1_sorted
+
+
+def FileBetter(file_name_1, file_name_2, metric_column, method):
+  """
+  Compares two data files and determines which is better and by how
+  much. Also produces a histogram of how much better, by PSNR.
+  metric_column is the metric.
+  """
+  # Store and parse our two files into lists of unique tuples.
+
+  # Read the two files, parsing out lines starting with bitrate.
+  metric_set1_sorted = ParseMetricFile(file_name_1, metric_column)
+  metric_set2_sorted = ParseMetricFile(file_name_2, metric_column)
+
+
+  def GraphBetter(metric_set1_sorted, metric_set2_sorted, base_is_set_2):
+    """
+    Search through the sorted metric file for metrics on either side of
+    the metric from file 1.  Since both lists are sorted we really
+    should not have to search through the entire range, but these
+    are small files."""
+    total_bitrate_difference_ratio = 0.0
+    count = 0
+    for bitrate, metric in metric_set1_sorted:
+      if bitrate == 0:
+        continue
+      for i in range(len(metric_set2_sorted) - 1):
+        s2_bitrate_0, s2_metric_0 = metric_set2_sorted[i]
+        s2_bitrate_1, s2_metric_1 = metric_set2_sorted[i + 1]
+        # We have a point on either side of our metric range.
+        if metric > s2_metric_0 and metric <= s2_metric_1:
+
+          # Calculate a slope.
+          if s2_metric_1 - s2_metric_0 != 0:
+            metric_slope = ((s2_bitrate_1 - s2_bitrate_0) /
+                            (s2_metric_1 - s2_metric_0))
+          else:
+            metric_slope = 0
+
+          estimated_s2_bitrate = (s2_bitrate_0 + (metric - s2_metric_0) *
+                                  metric_slope)
+
+          if estimated_s2_bitrate == 0:
+            continue
+          # Calculate percentage difference as given by base.
+          if base_is_set_2 == 0:
+            bitrate_difference_ratio = ((bitrate - estimated_s2_bitrate) /
+                                        bitrate)
+          else:
+            bitrate_difference_ratio = ((bitrate - estimated_s2_bitrate) /
+                                        estimated_s2_bitrate)
+
+          total_bitrate_difference_ratio += bitrate_difference_ratio
+          count += 1
+          break
+
+    # Calculate the average improvement between graphs.
+    if count != 0:
+      avg = total_bitrate_difference_ratio / count
+
+    else:
+      avg = 0.0
+
+    return avg
+
+  # Be fair to both graphs by testing all the points in each.
+  if method == 'avg':
+    avg_improvement = 50 * (
+                       GraphBetter(metric_set1_sorted, metric_set2_sorted, 1) -
+                       GraphBetter(metric_set2_sorted, metric_set1_sorted, 0))
+  elif method == 'dsnr':
+      avg_improvement = bdsnr2(metric_set1_sorted, metric_set2_sorted)
+  else:
+      avg_improvement = bdrate2(metric_set2_sorted, metric_set1_sorted)
+
+  return avg_improvement
+
+
+def HandleFiles(variables):
+  """
+  This script creates html for displaying metric data produced from data
+  in a video stats file,  as created by the AOM project when enable_psnr
+  is turned on:
+
+  Usage: visual_metrics.py template.html pattern base_dir sub_dir [ sub_dir2 ..]
+
+  The script parses each metrics file [see below] that matches the
+  statfile_pattern  in the baseline directory and looks for the file that
+  matches that same file in each of the sub_dirs, and compares the resultant
+  metrics bitrate, avg psnr, glb psnr, and ssim. "
+
+  It provides a table in which each row is a file in the line directory,
+  and a column for each subdir, with the cells representing how that clip
+  compares to baseline for that subdir.   A graph is given for each which
+  compares filesize to that metric.  If you click on a point in the graph it
+  zooms in on that point.
+
+  a SAMPLE metrics file:
+
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+   25.911   38.242   38.104   38.258   38.121   75.790    14103
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+   49.982   41.264   41.129   41.255   41.122   83.993    19817
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+   74.967   42.911   42.767   42.899   42.756   87.928    17332
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+  100.012   43.983   43.838   43.881   43.738   89.695    25389
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+  149.980   45.338   45.203   45.184   45.043   91.591    25438
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+  199.852   46.225   46.123   46.113   45.999   92.679    28302
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+  249.922   46.864   46.773   46.777   46.673   93.334    27244
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+  299.998   47.366   47.281   47.317   47.220   93.844    27137
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+  349.769   47.746   47.677   47.722   47.648   94.178    32226
+  Bitrate  AVGPsnr  GLBPsnr  AVPsnrP  GLPsnrP  VPXSSIM    Time(us)
+  399.773   48.032   47.971   48.013   47.946   94.362    36203
+
+  sample use:
+  visual_metrics.py template.html "*stt" aom aom_b aom_c > metrics.html
+  """
+
+  # The template file is the html file into which we will write the
+  # data from the stats file, formatted correctly for the gviz_api.
+  template_file = open(variables[1], "r")
+  page_template = template_file.read()
+  template_file.close()
+
+  # This is the path match pattern for finding stats files amongst
+  # all the other files it could be.  eg: *.stt
+  file_pattern = variables[2]
+
+  # This is the directory with files that we will use to do the comparison
+  # against.
+  baseline_dir = variables[3]
+  snrs = ''
+  filestable = {}
+
+  filestable['dsnr'] = ''
+  filestable['drate'] = ''
+  filestable['avg'] = ''
+
+  # Dirs is directories after the baseline to compare to the base.
+  dirs = variables[4:len(variables)]
+
+  # Find the metric files in the baseline directory.
+  dir_list = sorted(fnmatch.filter(os.listdir(baseline_dir), file_pattern))
+
+  metrics = GetMetrics(baseline_dir + "/" + dir_list[0])
+
+  metrics_js = 'metrics = ["' + '", "'.join(metrics) + '"];'
+
+  for column in range(1, len(metrics)):
+
+    for metric in ['avg','dsnr','drate']:
+      description = {"file": ("string", "File")}
+
+      # Go through each directory and add a column header to our description.
+      countoverall = {}
+      sumoverall = {}
+
+      for directory in dirs:
+        description[directory] = ("number", directory)
+        countoverall[directory] = 0
+        sumoverall[directory] = 0
+
+      # Data holds the data for the visualization, name given comes from
+      # gviz_api sample code.
+      data = []
+      for filename in dir_list:
+        row = {'file': splitext(basename(filename))[0] }
+        baseline_file_name = baseline_dir + "/" + filename
+
+        # Read the metric file from each of the directories in our list.
+        for directory in dirs:
+          metric_file_name = directory + "/" + filename
+
+          # If there is a metric file in the current directory, open it
+          # and calculate its overall difference between it and the baseline
+          # directory's metric file.
+          if os.path.isfile(metric_file_name):
+            overall = FileBetter(baseline_file_name, metric_file_name,
+                                 column, metric)
+            row[directory] = overall
+
+            sumoverall[directory] += overall
+            countoverall[directory] += 1
+
+        data.append(row)
+
+      # Add the overall numbers.
+      row = {"file": "OVERALL" }
+      for directory in dirs:
+        row[directory] = sumoverall[directory] / countoverall[directory]
+      data.append(row)
+
+      # write the tables out
+      data_table = gviz_api.DataTable(description)
+      data_table.LoadData(data)
+
+      filestable[metric] = ( filestable[metric] + "filestable_" + metric +
+                             "[" + str(column) + "]=" +
+                             data_table.ToJSon(columns_order=["file"]+dirs) + "\n" )
+
+    filestable_avg = filestable['avg']
+    filestable_dpsnr = filestable['dsnr']
+    filestable_drate = filestable['drate']
+
+    # Now we collect all the data for all the graphs.  First the column
+    # headers which will be Datarate and then each directory.
+    columns = ("datarate",baseline_dir)
+    description = {"datarate":("number", "Datarate")}
+    for directory in dirs:
+      description[directory] = ("number", directory)
+
+    description[baseline_dir] = ("number", baseline_dir)
+
+    snrs = snrs + "snrs[" + str(column) + "] = ["
+
+    # Now collect the data for the graphs, file by file.
+    for filename in dir_list:
+
+      data = []
+
+      # Collect the file in each directory and store all of its metrics
+      # in the associated gviz metrics table.
+      all_dirs = dirs + [baseline_dir]
+      for directory in all_dirs:
+
+        metric_file_name = directory + "/" + filename
+        if not os.path.isfile(metric_file_name):
+          continue
+
+        # Read and parse the metrics file storing it to the data we'll
+        # use for the gviz_api.Datatable.
+        metrics = ParseMetricFile(metric_file_name, column)
+        for bitrate, metric in metrics:
+          data.append({"datarate": bitrate, directory: metric})
+
+      data_table = gviz_api.DataTable(description)
+      data_table.LoadData(data)
+      snrs = snrs + "'" + data_table.ToJSon(
+         columns_order=tuple(["datarate",baseline_dir]+dirs)) + "',"
+
+    snrs = snrs + "]\n"
+
+    formatters = ""
+    for i in range(len(dirs)):
+      formatters = "%s   formatter.format(better, %d);" % (formatters, i+1)
+
+  print FillForm(page_template, vars())
+  return
+
+if len(sys.argv) < 3:
+  print HandleFiles.__doc__
+else:
+  HandleFiles(sys.argv)
diff --git a/third_party/aom/test/warp_filter_test.cc b/third_party/aom/test/warp_filter_test.cc
new file mode 100644
index 0000000000..19a4e8b6a1
--- /dev/null
+++ b/third_party/aom/test/warp_filter_test.cc
@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/warp_filter_test_util.h"
+using ::testing::make_tuple;
+using ::testing::tuple;
+using libaom_test::ACMRandom;
+using libaom_test::AV1HighbdWarpFilter::AV1HighbdWarpFilterTest;
+using libaom_test::AV1WarpFilter::AV1WarpFilterTest;
+
+namespace {
+
+TEST_P(AV1WarpFilterTest, CheckOutput) {
+  RunCheckOutput(::testing::get<3>(GET_PARAM(0)));
+}
+TEST_P(AV1WarpFilterTest, DISABLED_Speed) {
+  RunSpeedTest(::testing::get<3>(GET_PARAM(0)));
+}
+
+INSTANTIATE_TEST_CASE_P(
+    C, AV1WarpFilterTest,
+    libaom_test::AV1WarpFilter::BuildParams(av1_warp_affine_c));
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(
+    SSE4_1, AV1WarpFilterTest,
+    libaom_test::AV1WarpFilter::BuildParams(av1_warp_affine_sse4_1));
+
+TEST_P(AV1HighbdWarpFilterTest, CheckOutput) {
+  RunCheckOutput(::testing::get<4>(GET_PARAM(0)));
+}
+TEST_P(AV1HighbdWarpFilterTest, DISABLED_Speed) {
+  RunSpeedTest(::testing::get<4>(GET_PARAM(0)));
+}
+
+INSTANTIATE_TEST_CASE_P(SSE4_1, AV1HighbdWarpFilterTest,
+                        libaom_test::AV1HighbdWarpFilter::BuildParams(
+                            av1_highbd_warp_affine_sse4_1));
+
+#endif  // HAVE_SSE4_1
+
+#if HAVE_NEON
+INSTANTIATE_TEST_CASE_P(
+    NEON, AV1WarpFilterTest,
+    libaom_test::AV1WarpFilter::BuildParams(av1_warp_affine_neon));
+#endif  // HAVE_NEON
+
+}  // namespace
diff --git a/third_party/aom/test/warp_filter_test_util.cc b/third_party/aom/test/warp_filter_test_util.cc
new file mode 100644
index 0000000000..69b2ed4afe
--- /dev/null
+++ b/third_party/aom/test/warp_filter_test_util.cc
@@ -0,0 +1,480 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#include "aom_ports/aom_timer.h"
+#include "test/warp_filter_test_util.h"
+
+using ::testing::make_tuple;
+using ::testing::tuple;
+
+namespace libaom_test {
+
+int32_t random_warped_param(libaom_test::ACMRandom *rnd, int bits) {
+  // 1 in 8 chance of generating zero (arbitrarily chosen)
+  if (((rnd->Rand8()) & 7) == 0) return 0;
+  // Otherwise, enerate uniform values in the range
+  // [-(1 << bits), 1] U [1, 1<<bits]
+  int32_t v = 1 + (rnd->Rand16() & ((1 << bits) - 1));
+  if ((rnd->Rand8()) & 1) return -v;
+  return v;
+}
+
+void generate_warped_model(libaom_test::ACMRandom *rnd, int32_t *mat,
+                           int16_t *alpha, int16_t *beta, int16_t *gamma,
+                           int16_t *delta, const int is_alpha_zero,
+                           const int is_beta_zero, const int is_gamma_zero,
+                           const int is_delta_zero) {
+  while (1) {
+    int rnd8 = rnd->Rand8() & 3;
+    mat[0] = random_warped_param(rnd, WARPEDMODEL_PREC_BITS + 6);
+    mat[1] = random_warped_param(rnd, WARPEDMODEL_PREC_BITS + 6);
+    mat[2] = (random_warped_param(rnd, WARPEDMODEL_PREC_BITS - 3)) +
+             (1 << WARPEDMODEL_PREC_BITS);
+    mat[3] = random_warped_param(rnd, WARPEDMODEL_PREC_BITS - 3);
+
+    if (rnd8 <= 1) {
+      // AFFINE
+      mat[4] = random_warped_param(rnd, WARPEDMODEL_PREC_BITS - 3);
+      mat[5] = (random_warped_param(rnd, WARPEDMODEL_PREC_BITS - 3)) +
+               (1 << WARPEDMODEL_PREC_BITS);
+    } else if (rnd8 == 2) {
+      mat[4] = -mat[3];
+      mat[5] = mat[2];
+    } else {
+      mat[4] = random_warped_param(rnd, WARPEDMODEL_PREC_BITS - 3);
+      mat[5] = (random_warped_param(rnd, WARPEDMODEL_PREC_BITS - 3)) +
+               (1 << WARPEDMODEL_PREC_BITS);
+      if (is_alpha_zero == 1) mat[2] = 1 << WARPEDMODEL_PREC_BITS;
+      if (is_beta_zero == 1) mat[3] = 0;
+      if (is_gamma_zero == 1) mat[4] = 0;
+      if (is_delta_zero == 1)
+        mat[5] = (((int64_t)mat[3] * mat[4] + (mat[2] / 2)) / mat[2]) +
+                 (1 << WARPEDMODEL_PREC_BITS);
+    }
+
+    // Calculate the derived parameters and check that they are suitable
+    // for the warp filter.
+    assert(mat[2] != 0);
+
+    *alpha = clamp(mat[2] - (1 << WARPEDMODEL_PREC_BITS), INT16_MIN, INT16_MAX);
+    *beta = clamp(mat[3], INT16_MIN, INT16_MAX);
+    *gamma = clamp(((int64_t)mat[4] * (1 << WARPEDMODEL_PREC_BITS)) / mat[2],
+                   INT16_MIN, INT16_MAX);
+    *delta =
+        clamp(mat[5] - (((int64_t)mat[3] * mat[4] + (mat[2] / 2)) / mat[2]) -
+                  (1 << WARPEDMODEL_PREC_BITS),
+              INT16_MIN, INT16_MAX);
+
+    if ((4 * abs(*alpha) + 7 * abs(*beta) >= (1 << WARPEDMODEL_PREC_BITS)) ||
+        (4 * abs(*gamma) + 4 * abs(*delta) >= (1 << WARPEDMODEL_PREC_BITS)))
+      continue;
+
+    *alpha = ROUND_POWER_OF_TWO_SIGNED(*alpha, WARP_PARAM_REDUCE_BITS) *
+             (1 << WARP_PARAM_REDUCE_BITS);
+    *beta = ROUND_POWER_OF_TWO_SIGNED(*beta, WARP_PARAM_REDUCE_BITS) *
+            (1 << WARP_PARAM_REDUCE_BITS);
+    *gamma = ROUND_POWER_OF_TWO_SIGNED(*gamma, WARP_PARAM_REDUCE_BITS) *
+             (1 << WARP_PARAM_REDUCE_BITS);
+    *delta = ROUND_POWER_OF_TWO_SIGNED(*delta, WARP_PARAM_REDUCE_BITS) *
+             (1 << WARP_PARAM_REDUCE_BITS);
+
+    // We have a valid model, so finish
+    return;
+  }
+}
+
+namespace AV1WarpFilter {
+::testing::internal::ParamGenerator<WarpTestParams> BuildParams(
+    warp_affine_func filter) {
+  WarpTestParam params[] = {
+    make_tuple(4, 4, 50000, filter),  make_tuple(8, 8, 50000, filter),
+    make_tuple(64, 64, 1000, filter), make_tuple(4, 16, 20000, filter),
+    make_tuple(32, 8, 10000, filter),
+  };
+  return ::testing::Combine(::testing::ValuesIn(params),
+                            ::testing::Values(0, 1), ::testing::Values(0, 1),
+                            ::testing::Values(0, 1), ::testing::Values(0, 1));
+}
+
+AV1WarpFilterTest::~AV1WarpFilterTest() {}
+void AV1WarpFilterTest::SetUp() { rnd_.Reset(ACMRandom::DeterministicSeed()); }
+
+void AV1WarpFilterTest::TearDown() { libaom_test::ClearSystemState(); }
+
+void AV1WarpFilterTest::RunSpeedTest(warp_affine_func test_impl) {
+  const int w = 128, h = 128;
+  const int border = 16;
+  const int stride = w + 2 * border;
+  WarpTestParam params = GET_PARAM(0);
+  const int out_w = ::testing::get<0>(params),
+            out_h = ::testing::get<1>(params);
+  const int is_alpha_zero = GET_PARAM(1);
+  const int is_beta_zero = GET_PARAM(2);
+  const int is_gamma_zero = GET_PARAM(3);
+  const int is_delta_zero = GET_PARAM(4);
+  int sub_x, sub_y;
+  const int bd = 8;
+
+  uint8_t *input_ = new uint8_t[h * stride];
+  uint8_t *input = input_ + border;
+
+  // The warp functions always write rows with widths that are multiples of 8.
+  // So to avoid a buffer overflow, we may need to pad rows to a multiple of 8.
+  int output_n = ((out_w + 7) & ~7) * out_h;
+  uint8_t *output = new uint8_t[output_n];
+  int32_t mat[8];
+  int16_t alpha, beta, gamma, delta;
+  ConvolveParams conv_params = get_conv_params(0, 0, bd);
+  CONV_BUF_TYPE *dsta = new CONV_BUF_TYPE[output_n];
+  generate_warped_model(&rnd_, mat, &alpha, &beta, &gamma, &delta,
+                        is_alpha_zero, is_beta_zero, is_gamma_zero,
+                        is_delta_zero);
+
+  for (int r = 0; r < h; ++r)
+    for (int c = 0; c < w; ++c) input[r * stride + c] = rnd_.Rand8();
+  for (int r = 0; r < h; ++r) {
+    memset(input + r * stride - border, input[r * stride], border);
+    memset(input + r * stride + w, input[r * stride + (w - 1)], border);
+  }
+
+  sub_x = 0;
+  sub_y = 0;
+  int do_average = 0;
+
+  conv_params = get_conv_params_no_round(do_average, 0, dsta, out_w, 1, bd);
+  conv_params.use_jnt_comp_avg = 0;
+
+  const int num_loops = 1000000000 / (out_w + out_h);
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+  for (int i = 0; i < num_loops; ++i)
+    test_impl(mat, input, w, h, stride, output, 32, 32, out_w, out_h, out_w,
+              sub_x, sub_y, &conv_params, alpha, beta, gamma, delta);
+
+  aom_usec_timer_mark(&timer);
+  const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+  printf("warp %3dx%-3d: %7.2f ns\n", out_w, out_h,
+         1000.0 * elapsed_time / num_loops);
+
+  delete[] input_;
+  delete[] output;
+  delete[] dsta;
+}
+
+void AV1WarpFilterTest::RunCheckOutput(warp_affine_func test_impl) {
+  const int w = 128, h = 128;
+  const int border = 16;
+  const int stride = w + 2 * border;
+  WarpTestParam params = GET_PARAM(0);
+  const int is_alpha_zero = GET_PARAM(1);
+  const int is_beta_zero = GET_PARAM(2);
+  const int is_gamma_zero = GET_PARAM(3);
+  const int is_delta_zero = GET_PARAM(4);
+  const int out_w = ::testing::get<0>(params),
+            out_h = ::testing::get<1>(params);
+  const int num_iters = ::testing::get<2>(params);
+  int i, j, sub_x, sub_y;
+  const int bd = 8;
+
+  // The warp functions always write rows with widths that are multiples of 8.
+  // So to avoid a buffer overflow, we may need to pad rows to a multiple of 8.
+  int output_n = ((out_w + 7) & ~7) * out_h;
+  uint8_t *input_ = new uint8_t[h * stride];
+  uint8_t *input = input_ + border;
+  uint8_t *output = new uint8_t[output_n];
+  uint8_t *output2 = new uint8_t[output_n];
+  int32_t mat[8];
+  int16_t alpha, beta, gamma, delta;
+  ConvolveParams conv_params = get_conv_params(0, 0, bd);
+  CONV_BUF_TYPE *dsta = new CONV_BUF_TYPE[output_n];
+  CONV_BUF_TYPE *dstb = new CONV_BUF_TYPE[output_n];
+  for (int i = 0; i < output_n; ++i) output[i] = output2[i] = rnd_.Rand8();
+
+  for (i = 0; i < num_iters; ++i) {
+    // Generate an input block and extend its borders horizontally
+    for (int r = 0; r < h; ++r)
+      for (int c = 0; c < w; ++c) input[r * stride + c] = rnd_.Rand8();
+    for (int r = 0; r < h; ++r) {
+      memset(input + r * stride - border, input[r * stride], border);
+      memset(input + r * stride + w, input[r * stride + (w - 1)], border);
+    }
+    const int use_no_round = rnd_.Rand8() & 1;
+    for (sub_x = 0; sub_x < 2; ++sub_x)
+      for (sub_y = 0; sub_y < 2; ++sub_y) {
+        generate_warped_model(&rnd_, mat, &alpha, &beta, &gamma, &delta,
+                              is_alpha_zero, is_beta_zero, is_gamma_zero,
+                              is_delta_zero);
+
+        for (int ii = 0; ii < 2; ++ii) {
+          for (int jj = 0; jj < 5; ++jj) {
+            for (int do_average = 0; do_average <= 1; ++do_average) {
+              if (use_no_round) {
+                conv_params =
+                    get_conv_params_no_round(do_average, 0, dsta, out_w, 1, bd);
+              } else {
+                conv_params = get_conv_params(0, 0, bd);
+              }
+              if (jj >= 4) {
+                conv_params.use_jnt_comp_avg = 0;
+              } else {
+                conv_params.use_jnt_comp_avg = 1;
+                conv_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+                conv_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+              }
+              av1_warp_affine_c(mat, input, w, h, stride, output, 32, 32, out_w,
+                                out_h, out_w, sub_x, sub_y, &conv_params, alpha,
+                                beta, gamma, delta);
+              if (use_no_round) {
+                conv_params =
+                    get_conv_params_no_round(do_average, 0, dstb, out_w, 1, bd);
+              }
+              if (jj >= 4) {
+                conv_params.use_jnt_comp_avg = 0;
+              } else {
+                conv_params.use_jnt_comp_avg = 1;
+                conv_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+                conv_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+              }
+              test_impl(mat, input, w, h, stride, output2, 32, 32, out_w, out_h,
+                        out_w, sub_x, sub_y, &conv_params, alpha, beta, gamma,
+                        delta);
+              if (use_no_round) {
+                for (j = 0; j < out_w * out_h; ++j)
+                  ASSERT_EQ(dsta[j], dstb[j])
+                      << "Pixel mismatch at index " << j << " = ("
+                      << (j % out_w) << ", " << (j / out_w) << ") on iteration "
+                      << i;
+                for (j = 0; j < out_w * out_h; ++j)
+                  ASSERT_EQ(output[j], output2[j])
+                      << "Pixel mismatch at index " << j << " = ("
+                      << (j % out_w) << ", " << (j / out_w) << ") on iteration "
+                      << i;
+              } else {
+                for (j = 0; j < out_w * out_h; ++j)
+                  ASSERT_EQ(output[j], output2[j])
+                      << "Pixel mismatch at index " << j << " = ("
+                      << (j % out_w) << ", " << (j / out_w) << ") on iteration "
+                      << i;
+              }
+            }
+          }
+        }
+      }
+  }
+  delete[] input_;
+  delete[] output;
+  delete[] output2;
+  delete[] dsta;
+  delete[] dstb;
+}
+}  // namespace AV1WarpFilter
+
+namespace AV1HighbdWarpFilter {
+::testing::internal::ParamGenerator<HighbdWarpTestParams> BuildParams(
+    highbd_warp_affine_func filter) {
+  const HighbdWarpTestParam params[] = {
+    make_tuple(4, 4, 100, 8, filter),    make_tuple(8, 8, 100, 8, filter),
+    make_tuple(64, 64, 100, 8, filter),  make_tuple(4, 16, 100, 8, filter),
+    make_tuple(32, 8, 100, 8, filter),   make_tuple(4, 4, 100, 10, filter),
+    make_tuple(8, 8, 100, 10, filter),   make_tuple(64, 64, 100, 10, filter),
+    make_tuple(4, 16, 100, 10, filter),  make_tuple(32, 8, 100, 10, filter),
+    make_tuple(4, 4, 100, 12, filter),   make_tuple(8, 8, 100, 12, filter),
+    make_tuple(64, 64, 100, 12, filter), make_tuple(4, 16, 100, 12, filter),
+    make_tuple(32, 8, 100, 12, filter),
+  };
+  return ::testing::Combine(::testing::ValuesIn(params),
+                            ::testing::Values(0, 1), ::testing::Values(0, 1),
+                            ::testing::Values(0, 1), ::testing::Values(0, 1));
+}
+
+AV1HighbdWarpFilterTest::~AV1HighbdWarpFilterTest() {}
+void AV1HighbdWarpFilterTest::SetUp() {
+  rnd_.Reset(ACMRandom::DeterministicSeed());
+}
+
+void AV1HighbdWarpFilterTest::TearDown() { libaom_test::ClearSystemState(); }
+
+void AV1HighbdWarpFilterTest::RunSpeedTest(highbd_warp_affine_func test_impl) {
+  const int w = 128, h = 128;
+  const int border = 16;
+  const int stride = w + 2 * border;
+  HighbdWarpTestParam param = GET_PARAM(0);
+  const int is_alpha_zero = GET_PARAM(1);
+  const int is_beta_zero = GET_PARAM(2);
+  const int is_gamma_zero = GET_PARAM(3);
+  const int is_delta_zero = GET_PARAM(4);
+  const int out_w = ::testing::get<0>(param), out_h = ::testing::get<1>(param);
+  const int bd = ::testing::get<3>(param);
+  const int mask = (1 << bd) - 1;
+  int sub_x, sub_y;
+
+  // The warp functions always write rows with widths that are multiples of 8.
+  // So to avoid a buffer overflow, we may need to pad rows to a multiple of 8.
+  int output_n = ((out_w + 7) & ~7) * out_h;
+  uint16_t *input_ = new uint16_t[h * stride];
+  uint16_t *input = input_ + border;
+  uint16_t *output = new uint16_t[output_n];
+  int32_t mat[8];
+  int16_t alpha, beta, gamma, delta;
+  ConvolveParams conv_params = get_conv_params(0, 0, bd);
+  CONV_BUF_TYPE *dsta = new CONV_BUF_TYPE[output_n];
+
+  generate_warped_model(&rnd_, mat, &alpha, &beta, &gamma, &delta,
+                        is_alpha_zero, is_beta_zero, is_gamma_zero,
+                        is_delta_zero);
+  // Generate an input block and extend its borders horizontally
+  for (int r = 0; r < h; ++r)
+    for (int c = 0; c < w; ++c) input[r * stride + c] = rnd_.Rand16() & mask;
+  for (int r = 0; r < h; ++r) {
+    for (int c = 0; c < border; ++c) {
+      input[r * stride - border + c] = input[r * stride];
+      input[r * stride + w + c] = input[r * stride + (w - 1)];
+    }
+  }
+
+  sub_x = 0;
+  sub_y = 0;
+  int do_average = 0;
+  conv_params.use_jnt_comp_avg = 0;
+  conv_params = get_conv_params_no_round(do_average, 0, dsta, out_w, 1, bd);
+
+  const int num_loops = 1000000000 / (out_w + out_h);
+  aom_usec_timer timer;
+  aom_usec_timer_start(&timer);
+
+  for (int i = 0; i < num_loops; ++i)
+    test_impl(mat, input, w, h, stride, output, 32, 32, out_w, out_h, out_w,
+              sub_x, sub_y, bd, &conv_params, alpha, beta, gamma, delta);
+
+  aom_usec_timer_mark(&timer);
+  const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer));
+  printf("highbd warp %3dx%-3d: %7.2f ns\n", out_w, out_h,
+         1000.0 * elapsed_time / num_loops);
+
+  delete[] input_;
+  delete[] output;
+  delete[] dsta;
+}
+
+void AV1HighbdWarpFilterTest::RunCheckOutput(
+    highbd_warp_affine_func test_impl) {
+  const int w = 128, h = 128;
+  const int border = 16;
+  const int stride = w + 2 * border;
+  HighbdWarpTestParam param = GET_PARAM(0);
+  const int is_alpha_zero = GET_PARAM(1);
+  const int is_beta_zero = GET_PARAM(2);
+  const int is_gamma_zero = GET_PARAM(3);
+  const int is_delta_zero = GET_PARAM(4);
+  const int out_w = ::testing::get<0>(param), out_h = ::testing::get<1>(param);
+  const int bd = ::testing::get<3>(param);
+  const int num_iters = ::testing::get<2>(param);
+  const int mask = (1 << bd) - 1;
+  int i, j, sub_x, sub_y;
+
+  // The warp functions always write rows with widths that are multiples of 8.
+  // So to avoid a buffer overflow, we may need to pad rows to a multiple of 8.
+  int output_n = ((out_w + 7) & ~7) * out_h;
+  uint16_t *input_ = new uint16_t[h * stride];
+  uint16_t *input = input_ + border;
+  uint16_t *output = new uint16_t[output_n];
+  uint16_t *output2 = new uint16_t[output_n];
+  int32_t mat[8];
+  int16_t alpha, beta, gamma, delta;
+  ConvolveParams conv_params = get_conv_params(0, 0, bd);
+  CONV_BUF_TYPE *dsta = new CONV_BUF_TYPE[output_n];
+  CONV_BUF_TYPE *dstb = new CONV_BUF_TYPE[output_n];
+  for (int i = 0; i < output_n; ++i) output[i] = output2[i] = rnd_.Rand16();
+
+  for (i = 0; i < num_iters; ++i) {
+    // Generate an input block and extend its borders horizontally
+    for (int r = 0; r < h; ++r)
+      for (int c = 0; c < w; ++c) input[r * stride + c] = rnd_.Rand16() & mask;
+    for (int r = 0; r < h; ++r) {
+      for (int c = 0; c < border; ++c) {
+        input[r * stride - border + c] = input[r * stride];
+        input[r * stride + w + c] = input[r * stride + (w - 1)];
+      }
+    }
+    const int use_no_round = rnd_.Rand8() & 1;
+    for (sub_x = 0; sub_x < 2; ++sub_x)
+      for (sub_y = 0; sub_y < 2; ++sub_y) {
+        generate_warped_model(&rnd_, mat, &alpha, &beta, &gamma, &delta,
+                              is_alpha_zero, is_beta_zero, is_gamma_zero,
+                              is_delta_zero);
+        for (int ii = 0; ii < 2; ++ii) {
+          for (int jj = 0; jj < 5; ++jj) {
+            for (int do_average = 0; do_average <= 1; ++do_average) {
+              if (use_no_round) {
+                conv_params =
+                    get_conv_params_no_round(do_average, 0, dsta, out_w, 1, bd);
+              } else {
+                conv_params = get_conv_params(0, 0, bd);
+              }
+              if (jj >= 4) {
+                conv_params.use_jnt_comp_avg = 0;
+              } else {
+                conv_params.use_jnt_comp_avg = 1;
+                conv_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+                conv_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+              }
+
+              av1_highbd_warp_affine_c(mat, input, w, h, stride, output, 32, 32,
+                                       out_w, out_h, out_w, sub_x, sub_y, bd,
+                                       &conv_params, alpha, beta, gamma, delta);
+              if (use_no_round) {
+                // TODO(angiebird): Change this to test_impl once we have SIMD
+                // implementation
+                conv_params =
+                    get_conv_params_no_round(do_average, 0, dstb, out_w, 1, bd);
+              }
+              if (jj >= 4) {
+                conv_params.use_jnt_comp_avg = 0;
+              } else {
+                conv_params.use_jnt_comp_avg = 1;
+                conv_params.fwd_offset = quant_dist_lookup_table[ii][jj][0];
+                conv_params.bck_offset = quant_dist_lookup_table[ii][jj][1];
+              }
+              test_impl(mat, input, w, h, stride, output2, 32, 32, out_w, out_h,
+                        out_w, sub_x, sub_y, bd, &conv_params, alpha, beta,
+                        gamma, delta);
+
+              if (use_no_round) {
+                for (j = 0; j < out_w * out_h; ++j)
+                  ASSERT_EQ(dsta[j], dstb[j])
+                      << "Pixel mismatch at index " << j << " = ("
+                      << (j % out_w) << ", " << (j / out_w) << ") on iteration "
+                      << i;
+                for (j = 0; j < out_w * out_h; ++j)
+                  ASSERT_EQ(output[j], output2[j])
+                      << "Pixel mismatch at index " << j << " = ("
+                      << (j % out_w) << ", " << (j / out_w) << ") on iteration "
+                      << i;
+              } else {
+                for (j = 0; j < out_w * out_h; ++j)
+                  ASSERT_EQ(output[j], output2[j])
+                      << "Pixel mismatch at index " << j << " = ("
+                      << (j % out_w) << ", " << (j / out_w) << ") on iteration "
+                      << i;
+              }
+            }
+          }
+        }
+      }
+  }
+
+  delete[] input_;
+  delete[] output;
+  delete[] output2;
+  delete[] dsta;
+  delete[] dstb;
+}
+}  // namespace AV1HighbdWarpFilter
+}  // namespace libaom_test
diff --git a/third_party/aom/test/warp_filter_test_util.h b/third_party/aom/test/warp_filter_test_util.h
new file mode 100644
index 0000000000..b8998e5c8e
--- /dev/null
+++ b/third_party/aom/test/warp_filter_test_util.h
@@ -0,0 +1,103 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#ifndef AOM_TEST_WARP_FILTER_TEST_UTIL_H_
+#define AOM_TEST_WARP_FILTER_TEST_UTIL_H_
+
+#include "config/av1_rtcd.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+#include "test/acm_random.h"
+#include "test/util.h"
+#include "test/clear_system_state.h"
+#include "test/register_state_check.h"
+
+#include "av1/common/mv.h"
+#include "av1/common/common_data.h"
+
+namespace libaom_test {
+
+void generate_warped_model(libaom_test::ACMRandom *rnd, int32_t *mat,
+                           int16_t *alpha, int16_t *beta, int16_t *gamma,
+                           int16_t *delta, int is_alpha_zero, int is_beta_zero,
+                           int is_gamma_zero, int is_delta_zero);
+
+namespace AV1WarpFilter {
+
+typedef void (*warp_affine_func)(const int32_t *mat, const uint8_t *ref,
+                                 int width, int height, int stride,
+                                 uint8_t *pred, int p_col, int p_row,
+                                 int p_width, int p_height, int p_stride,
+                                 int subsampling_x, int subsampling_y,
+                                 ConvolveParams *conv_params, int16_t alpha,
+                                 int16_t beta, int16_t gamma, int16_t delta);
+
+typedef ::testing::tuple<int, int, int, warp_affine_func> WarpTestParam;
+typedef ::testing::tuple<WarpTestParam, int, int, int, int> WarpTestParams;
+
+::testing::internal::ParamGenerator<WarpTestParams> BuildParams(
+    warp_affine_func filter);
+
+class AV1WarpFilterTest : public ::testing::TestWithParam<WarpTestParams> {
+ public:
+  virtual ~AV1WarpFilterTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput(warp_affine_func test_impl);
+  void RunSpeedTest(warp_affine_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+};
+
+}  // namespace AV1WarpFilter
+
+namespace AV1HighbdWarpFilter {
+typedef void (*highbd_warp_affine_func)(const int32_t *mat, const uint16_t *ref,
+                                        int width, int height, int stride,
+                                        uint16_t *pred, int p_col, int p_row,
+                                        int p_width, int p_height, int p_stride,
+                                        int subsampling_x, int subsampling_y,
+                                        int bd, ConvolveParams *conv_params,
+                                        int16_t alpha, int16_t beta,
+                                        int16_t gamma, int16_t delta);
+
+typedef ::testing::tuple<int, int, int, int, highbd_warp_affine_func>
+    HighbdWarpTestParam;
+typedef ::testing::tuple<HighbdWarpTestParam, int, int, int, int>
+    HighbdWarpTestParams;
+
+::testing::internal::ParamGenerator<HighbdWarpTestParams> BuildParams(
+    highbd_warp_affine_func filter);
+
+class AV1HighbdWarpFilterTest
+    : public ::testing::TestWithParam<HighbdWarpTestParams> {
+ public:
+  virtual ~AV1HighbdWarpFilterTest();
+  virtual void SetUp();
+
+  virtual void TearDown();
+
+ protected:
+  void RunCheckOutput(highbd_warp_affine_func test_impl);
+  void RunSpeedTest(highbd_warp_affine_func test_impl);
+
+  libaom_test::ACMRandom rnd_;
+};
+
+}  // namespace AV1HighbdWarpFilter
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_WARP_FILTER_TEST_UTIL_H_
diff --git a/third_party/aom/test/webm_video_source.h b/third_party/aom/test/webm_video_source.h
new file mode 100644
index 0000000000..bb3d117355
--- /dev/null
+++ b/third_party/aom/test/webm_video_source.h
@@ -0,0 +1,96 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_WEBM_VIDEO_SOURCE_H_
+#define AOM_TEST_WEBM_VIDEO_SOURCE_H_
+#include <cstdarg>
+#include <cstdio>
+#include <cstdlib>
+#include <new>
+#include <string>
+#include "common/tools_common.h"
+#include "common/webmdec.h"
+#include "test/video_source.h"
+
+namespace libaom_test {
+
+// This class extends VideoSource to allow parsing of WebM files,
+// so that we can do actual file decodes.
+class WebMVideoSource : public CompressedVideoSource {
+ public:
+  explicit WebMVideoSource(const std::string &file_name)
+      : file_name_(file_name), aom_ctx_(new AvxInputContext()),
+        webm_ctx_(new WebmInputContext()), buf_(NULL), buf_sz_(0), frame_sz_(0),
+        frame_number_(0), end_of_file_(false) {}
+
+  virtual ~WebMVideoSource() {
+    if (aom_ctx_->file != NULL) fclose(aom_ctx_->file);
+    webm_free(webm_ctx_);
+    delete aom_ctx_;
+    delete webm_ctx_;
+  }
+
+  virtual void Init() {}
+
+  virtual void Begin() {
+    aom_ctx_->file = OpenTestDataFile(file_name_);
+    ASSERT_TRUE(aom_ctx_->file != NULL)
+        << "Input file open failed. Filename: " << file_name_;
+
+    ASSERT_EQ(file_is_webm(webm_ctx_, aom_ctx_), 1) << "file is not WebM";
+
+    FillFrame();
+  }
+
+  virtual void Next() {
+    ++frame_number_;
+    FillFrame();
+  }
+
+  void FillFrame() {
+    ASSERT_TRUE(aom_ctx_->file != NULL);
+    const int status = webm_read_frame(webm_ctx_, &buf_, &frame_sz_, &buf_sz_);
+    ASSERT_GE(status, 0) << "webm_read_frame failed";
+    if (status == 1) {
+      end_of_file_ = true;
+    }
+  }
+
+  void SeekToNextKeyFrame() {
+    ASSERT_TRUE(aom_ctx_->file != NULL);
+    do {
+      const int status =
+          webm_read_frame(webm_ctx_, &buf_, &frame_sz_, &buf_sz_);
+      ASSERT_GE(status, 0) << "webm_read_frame failed";
+      ++frame_number_;
+      if (status == 1) {
+        end_of_file_ = true;
+      }
+    } while (!webm_ctx_->is_key_frame && !end_of_file_);
+  }
+
+  virtual const uint8_t *cxdata() const { return end_of_file_ ? NULL : buf_; }
+  virtual size_t frame_size() const { return frame_sz_; }
+  virtual unsigned int frame_number() const { return frame_number_; }
+
+ protected:
+  std::string file_name_;
+  AvxInputContext *aom_ctx_;
+  WebmInputContext *webm_ctx_;
+  uint8_t *buf_;  // Owned by webm_ctx_ and freed when webm_ctx_ is freed.
+  size_t buf_sz_;
+  size_t frame_sz_;
+  unsigned int frame_number_;
+  bool end_of_file_;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_WEBM_VIDEO_SOURCE_H_
diff --git a/third_party/aom/test/wiener_test.cc b/third_party/aom/test/wiener_test.cc
new file mode 100644
index 0000000000..dfec09119d
--- /dev/null
+++ b/third_party/aom/test/wiener_test.cc
@@ -0,0 +1,280 @@
+/*
+ * Copyright (c) 2018, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <vector>
+
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+#include "test/function_equivalence_test.h"
+#include "test/register_state_check.h"
+
+#include "config/aom_config.h"
+#include "config/aom_dsp_rtcd.h"
+
+#include "aom/aom_integer.h"
+#include "av1/encoder/pickrst.h"
+
+#define MAX_WIENER_BLOCK 384
+#define MAX_DATA_BLOCK (MAX_WIENER_BLOCK + WIENER_WIN)
+using libaom_test::FunctionEquivalenceTest;
+
+namespace {
+
+static void compute_stats_win_opt_c(int wiener_win, const uint8_t *dgd,
+                                    const uint8_t *src, int h_start, int h_end,
+                                    int v_start, int v_end, int dgd_stride,
+                                    int src_stride, double *M, double *H) {
+  ASSERT_TRUE(wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA);
+  int i, j, k, l, m, n;
+  const int pixel_count = (h_end - h_start) * (v_end - v_start);
+  const int wiener_win2 = wiener_win * wiener_win;
+  const int wiener_halfwin = (wiener_win >> 1);
+  const double avg =
+      find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride);
+
+  std::vector<std::vector<int64_t> > M_int(wiener_win,
+                                           std::vector<int64_t>(wiener_win, 0));
+  std::vector<std::vector<int64_t> > H_int(
+      wiener_win * wiener_win, std::vector<int64_t>(wiener_win * 8, 0));
+  std::vector<std::vector<int32_t> > sumY(wiener_win,
+                                          std::vector<int32_t>(wiener_win, 0));
+  int32_t sumX = 0;
+  const uint8_t *dgd_win = dgd - wiener_halfwin * dgd_stride - wiener_halfwin;
+
+  for (i = v_start; i < v_end; i++) {
+    for (j = h_start; j < h_end; j += 2) {
+      const uint8_t X1 = src[i * src_stride + j];
+      const uint8_t X2 = src[i * src_stride + j + 1];
+      sumX += X1 + X2;
+
+      const uint8_t *dgd_ij = dgd_win + i * dgd_stride + j;
+      for (k = 0; k < wiener_win; k++) {
+        for (l = 0; l < wiener_win; l++) {
+          const uint8_t *dgd_ijkl = dgd_ij + k * dgd_stride + l;
+          int64_t *H_int_temp = &H_int[(l * wiener_win + k)][0];
+          const uint8_t D1 = dgd_ijkl[0];
+          const uint8_t D2 = dgd_ijkl[1];
+          sumY[k][l] += D1 + D2;
+          M_int[l][k] += D1 * X1 + D2 * X2;
+          for (m = 0; m < wiener_win; m++) {
+            for (n = 0; n < wiener_win; n++) {
+              H_int_temp[m * 8 + n] += D1 * dgd_ij[n + dgd_stride * m] +
+                                       D2 * dgd_ij[n + dgd_stride * m + 1];
+            }
+          }
+        }
+      }
+    }
+  }
+
+  const double avg_square_sum = avg * avg * pixel_count;
+  for (k = 0; k < wiener_win; k++) {
+    for (l = 0; l < wiener_win; l++) {
+      M[l * wiener_win + k] =
+          M_int[l][k] + avg_square_sum - avg * (sumX + sumY[k][l]);
+      for (m = 0; m < wiener_win; m++) {
+        for (n = 0; n < wiener_win; n++) {
+          H[(l * wiener_win + k) * wiener_win2 + m * wiener_win + n] =
+              H_int[(l * wiener_win + k)][n * 8 + m] + avg_square_sum -
+              avg * (sumY[k][l] + sumY[n][m]);
+        }
+      }
+    }
+  }
+}
+
+void compute_stats_opt_c(int wiener_win, const uint8_t *dgd, const uint8_t *src,
+                         int h_start, int h_end, int v_start, int v_end,
+                         int dgd_stride, int src_stride, double *M, double *H) {
+  if (wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA) {
+    compute_stats_win_opt_c(wiener_win, dgd, src, h_start, h_end, v_start,
+                            v_end, dgd_stride, src_stride, M, H);
+  } else {
+    av1_compute_stats_c(wiener_win, dgd, src, h_start, h_end, v_start, v_end,
+                        dgd_stride, src_stride, M, H);
+  }
+}
+
+static const int kIterations = 100;
+static const double min_error = (double)(0.01);
+typedef void (*compute_stats_Func)(int wiener_win, const uint8_t *dgd,
+                                   const uint8_t *src, int h_start, int h_end,
+                                   int v_start, int v_end, int dgd_stride,
+                                   int src_stride, double *M, double *H);
+
+typedef libaom_test::FuncParam<compute_stats_Func> TestFuncs;
+
+////////////////////////////////////////////////////////////////////////////////
+// 8 bit
+////////////////////////////////////////////////////////////////////////////////
+
+typedef ::testing::tuple<const compute_stats_Func> WienerTestParam;
+
+class WienerTest : public ::testing::TestWithParam<WienerTestParam> {
+ public:
+  virtual void SetUp() { target_func_ = GET_PARAM(0); }
+  void runWienerTest(const int32_t wiener_win, int32_t run_times);
+  void runWienerTest_ExtremeValues(const int32_t wiener_win);
+
+ private:
+  compute_stats_Func target_func_;
+  ACMRandom rng_;
+};
+
+void WienerTest::runWienerTest(const int32_t wiener_win, int32_t run_times) {
+  const int32_t wiener_halfwin = wiener_win >> 1;
+  const int32_t wiener_win2 = wiener_win * wiener_win;
+  DECLARE_ALIGNED(32, uint8_t, dgd_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]);
+  DECLARE_ALIGNED(32, uint8_t, src_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]);
+  DECLARE_ALIGNED(32, double, M_ref[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, double, H_ref[WIENER_WIN2 * WIENER_WIN2]);
+  DECLARE_ALIGNED(32, double, M_test[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, double, H_test[WIENER_WIN2 * WIENER_WIN2]);
+  const int h_start = ((rng_.Rand16() % (MAX_WIENER_BLOCK / 2)) & (~7));
+  int h_end =
+      run_times != 1 ? 256 : ((rng_.Rand16() % MAX_WIENER_BLOCK) & (~7)) + 8;
+  const int v_start = ((rng_.Rand16() % (MAX_WIENER_BLOCK / 2)) & (~7));
+  int v_end =
+      run_times != 1 ? 256 : ((rng_.Rand16() % MAX_WIENER_BLOCK) & (~7)) + 8;
+  const int dgd_stride = h_end;
+  const int src_stride = MAX_DATA_BLOCK;
+  const int iters = run_times == 1 ? kIterations : 2;
+  for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) {
+      dgd_buf[i] = rng_.Rand8();
+      src_buf[i] = rng_.Rand8();
+    }
+    uint8_t *dgd = dgd_buf + wiener_halfwin * MAX_DATA_BLOCK + wiener_halfwin;
+    uint8_t *src = src_buf;
+
+    aom_usec_timer timer;
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      av1_compute_stats_c(wiener_win, dgd, src, h_start, h_end, v_start, v_end,
+                          dgd_stride, src_stride, M_ref, H_ref);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    aom_usec_timer_start(&timer);
+    for (int i = 0; i < run_times; ++i) {
+      target_func_(wiener_win, dgd, src, h_start, h_end, v_start, v_end,
+                   dgd_stride, src_stride, M_test, H_test);
+    }
+    aom_usec_timer_mark(&timer);
+    const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer));
+    if (run_times > 10) {
+      printf("win %d %3dx%-3d:%7.2f/%7.2fns", wiener_win, h_end, v_end, time1,
+             time2);
+      printf("(%3.2f)\n", time1 / time2);
+    }
+    int failed = 0;
+    for (int i = 0; i < wiener_win2; ++i) {
+      if (fabs(M_ref[i] - M_test[i]) > min_error) {
+        failed = 1;
+        printf("win %d M iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win,
+               iter, i, M_ref[i], M_test[i]);
+        break;
+      }
+    }
+    // ASSERT_EQ(failed, 0);
+    for (int i = 0; i < wiener_win2 * wiener_win2; ++i) {
+      if (fabs(H_ref[i] - H_test[i]) > min_error) {
+        failed = 1;
+        printf("win %d H iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win,
+               iter, i, H_ref[i], H_test[i]);
+        break;
+      }
+    }
+    ASSERT_EQ(failed, 0);
+  }
+}
+
+void WienerTest::runWienerTest_ExtremeValues(const int32_t wiener_win) {
+  const int32_t wiener_halfwin = wiener_win >> 1;
+  const int32_t wiener_win2 = wiener_win * wiener_win;
+  DECLARE_ALIGNED(32, uint8_t, dgd_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]);
+  DECLARE_ALIGNED(32, uint8_t, src_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]);
+  DECLARE_ALIGNED(32, double, M_ref[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, double, H_ref[WIENER_WIN2 * WIENER_WIN2]);
+  DECLARE_ALIGNED(32, double, M_test[WIENER_WIN2]);
+  DECLARE_ALIGNED(32, double, H_test[WIENER_WIN2 * WIENER_WIN2]);
+  const int h_start = 16;
+  const int h_end = MAX_WIENER_BLOCK;
+  const int v_start = 16;
+  const int v_end = MAX_WIENER_BLOCK;
+  const int dgd_stride = h_end;
+  const int src_stride = MAX_DATA_BLOCK;
+  const int iters = 1;
+  for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) {
+    for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) {
+      dgd_buf[i] = 255;
+      src_buf[i] = 255;
+    }
+    uint8_t *dgd = dgd_buf + wiener_halfwin * MAX_DATA_BLOCK + wiener_halfwin;
+    uint8_t *src = src_buf;
+
+    av1_compute_stats_c(wiener_win, dgd, src, h_start, h_end, v_start, v_end,
+                        dgd_stride, src_stride, M_ref, H_ref);
+
+    target_func_(wiener_win, dgd, src, h_start, h_end, v_start, v_end,
+                 dgd_stride, src_stride, M_test, H_test);
+
+    int failed = 0;
+    for (int i = 0; i < wiener_win2; ++i) {
+      if (fabs(M_ref[i] - M_test[i]) > min_error) {
+        failed = 1;
+        printf("win %d M iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win,
+               iter, i, M_ref[i], M_test[i]);
+        break;
+      }
+    }
+    // ASSERT_EQ(failed, 0);
+    for (int i = 0; i < wiener_win2 * wiener_win2; ++i) {
+      if (fabs(H_ref[i] - H_test[i]) > min_error) {
+        failed = 1;
+        printf("win %d H iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win,
+               iter, i, H_ref[i], H_test[i]);
+        break;
+      }
+    }
+    ASSERT_EQ(failed, 0);
+  }
+}
+
+TEST_P(WienerTest, RandomValues) {
+  runWienerTest(WIENER_WIN, 1);
+  runWienerTest(WIENER_WIN_CHROMA, 1);
+}
+
+TEST_P(WienerTest, ExtremeValues) {
+  runWienerTest_ExtremeValues(WIENER_WIN);
+  runWienerTest_ExtremeValues(WIENER_WIN_CHROMA);
+}
+
+TEST_P(WienerTest, DISABLED_Speed) {
+  runWienerTest(WIENER_WIN, 200);
+  runWienerTest(WIENER_WIN_CHROMA, 200);
+}
+
+INSTANTIATE_TEST_CASE_P(C, WienerTest, ::testing::Values(compute_stats_opt_c));
+
+#if HAVE_SSE4_1
+INSTANTIATE_TEST_CASE_P(SSE4_1, WienerTest,
+                        ::testing::Values(av1_compute_stats_sse4_1));
+#endif  // HAVE_SSE4_1
+
+#if HAVE_AVX2
+
+INSTANTIATE_TEST_CASE_P(AVX2, WienerTest,
+                        ::testing::Values(av1_compute_stats_avx2));
+#endif  // HAVE_AVX2
+
+}  // namespace
diff --git a/third_party/aom/test/y4m_test.cc b/third_party/aom/test/y4m_test.cc
new file mode 100644
index 0000000000..6cc75ef5b0
--- /dev/null
+++ b/third_party/aom/test/y4m_test.cc
@@ -0,0 +1,180 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+
+#include <string>
+
+#include "config/aom_config.h"
+
+#include "common/y4menc.h"
+#include "test/md5_helper.h"
+#include "test/util.h"
+#include "test/y4m_video_source.h"
+#include "third_party/googletest/src/googletest/include/gtest/gtest.h"
+
+namespace {
+
+using std::string;
+
+static const unsigned int kWidth = 160;
+static const unsigned int kHeight = 90;
+static const unsigned int kFrames = 10;
+
+struct Y4mTestParam {
+  const char *filename;
+  unsigned int bit_depth;
+  aom_img_fmt format;
+  const char *md5raw;
+};
+
+const Y4mTestParam kY4mTestVectors[] = {
+  { "park_joy_90p_8_420.y4m", 8, AOM_IMG_FMT_I420,
+    "e5406275b9fc6bb3436c31d4a05c1cab" },
+  { "park_joy_90p_8_420_monochrome.y4m", 8, AOM_IMG_FMT_I420,
+    "95ef5bf6218580588be24a5271bb6a7f" },
+  { "park_joy_90p_8_420_vertical_csp.y4m", 8, AOM_IMG_FMT_I420,
+    "f53a40fec15254ac312527339d9c686b" },
+  { "park_joy_90p_8_422.y4m", 8, AOM_IMG_FMT_I422,
+    "284a47a47133b12884ec3a14e959a0b6" },
+  { "park_joy_90p_8_444.y4m", 8, AOM_IMG_FMT_I444,
+    "90517ff33843d85de712fd4fe60dbed0" },
+  { "park_joy_90p_10_420.y4m", 10, AOM_IMG_FMT_I42016,
+    "63f21f9f717d8b8631bd2288ee87137b" },
+  { "park_joy_90p_10_422.y4m", 10, AOM_IMG_FMT_I42216,
+    "48ab51fb540aed07f7ff5af130c9b605" },
+  { "park_joy_90p_10_444.y4m", 10, AOM_IMG_FMT_I44416,
+    "067bfd75aa85ff9bae91fa3e0edd1e3e" },
+  { "park_joy_90p_12_420.y4m", 12, AOM_IMG_FMT_I42016,
+    "9e6d8f6508c6e55625f6b697bc461cef" },
+  { "park_joy_90p_12_422.y4m", 12, AOM_IMG_FMT_I42216,
+    "b239c6b301c0b835485be349ca83a7e3" },
+  { "park_joy_90p_12_444.y4m", 12, AOM_IMG_FMT_I44416,
+    "5a6481a550821dab6d0192f5c63845e9" },
+};
+
+static const int PLANES_YUV[] = { AOM_PLANE_Y, AOM_PLANE_U, AOM_PLANE_V };
+
+class Y4mVideoSourceTest : public ::testing::TestWithParam<Y4mTestParam>,
+                           public ::libaom_test::Y4mVideoSource {
+ protected:
+  Y4mVideoSourceTest() : Y4mVideoSource("", 0, 0) {}
+
+  virtual ~Y4mVideoSourceTest() { CloseSource(); }
+
+  virtual void Init(const std::string &file_name, int limit) {
+    file_name_ = file_name;
+    start_ = 0;
+    limit_ = limit;
+    frame_ = 0;
+    Begin();
+  }
+
+  // Checks y4m header information
+  void HeaderChecks(unsigned int bit_depth, aom_img_fmt_t fmt) {
+    ASSERT_TRUE(input_file_ != NULL);
+    ASSERT_EQ(y4m_.pic_w, (int)kWidth);
+    ASSERT_EQ(y4m_.pic_h, (int)kHeight);
+    ASSERT_EQ(img()->d_w, kWidth);
+    ASSERT_EQ(img()->d_h, kHeight);
+    ASSERT_EQ(y4m_.bit_depth, bit_depth);
+    ASSERT_EQ(y4m_.aom_fmt, fmt);
+    if (fmt == AOM_IMG_FMT_I420 || fmt == AOM_IMG_FMT_I42016) {
+      ASSERT_EQ(y4m_.bps, (int)y4m_.bit_depth * 3 / 2);
+      ASSERT_EQ(img()->x_chroma_shift, 1U);
+      ASSERT_EQ(img()->y_chroma_shift, 1U);
+    }
+    if (fmt == AOM_IMG_FMT_I422 || fmt == AOM_IMG_FMT_I42216) {
+      ASSERT_EQ(y4m_.bps, (int)y4m_.bit_depth * 2);
+      ASSERT_EQ(img()->x_chroma_shift, 1U);
+      ASSERT_EQ(img()->y_chroma_shift, 0U);
+    }
+    if (fmt == AOM_IMG_FMT_I444 || fmt == AOM_IMG_FMT_I44416) {
+      ASSERT_EQ(y4m_.bps, (int)y4m_.bit_depth * 3);
+      ASSERT_EQ(img()->x_chroma_shift, 0U);
+      ASSERT_EQ(img()->y_chroma_shift, 0U);
+    }
+  }
+
+  // Checks MD5 of the raw frame data
+  void Md5Check(const string &expected_md5) {
+    ASSERT_TRUE(input_file_ != NULL);
+    libaom_test::MD5 md5;
+    for (unsigned int i = start_; i < limit_; i++) {
+      md5.Add(img());
+      Next();
+    }
+    ASSERT_EQ(string(md5.Get()), expected_md5);
+  }
+};
+
+TEST_P(Y4mVideoSourceTest, SourceTest) {
+  const Y4mTestParam t = GetParam();
+  Init(t.filename, kFrames);
+  HeaderChecks(t.bit_depth, t.format);
+  Md5Check(t.md5raw);
+}
+
+INSTANTIATE_TEST_CASE_P(C, Y4mVideoSourceTest,
+                        ::testing::ValuesIn(kY4mTestVectors));
+
+class Y4mVideoWriteTest : public Y4mVideoSourceTest {
+ protected:
+  Y4mVideoWriteTest() : tmpfile_(NULL) {}
+
+  virtual ~Y4mVideoWriteTest() {
+    delete tmpfile_;
+    input_file_ = NULL;
+  }
+
+  void ReplaceInputFile(FILE *input_file) {
+    CloseSource();
+    frame_ = 0;
+    input_file_ = input_file;
+    rewind(input_file_);
+    ReadSourceToStart();
+  }
+
+  // Writes out a y4m file and then reads it back
+  void WriteY4mAndReadBack() {
+    ASSERT_TRUE(input_file_ != NULL);
+    char buf[Y4M_BUFFER_SIZE] = { 0 };
+    const struct AvxRational framerate = { y4m_.fps_n, y4m_.fps_d };
+    tmpfile_ = new libaom_test::TempOutFile;
+    ASSERT_TRUE(tmpfile_->file() != NULL);
+    y4m_write_file_header(buf, sizeof(buf), kWidth, kHeight, &framerate,
+                          img()->monochrome, img()->csp, y4m_.aom_fmt,
+                          y4m_.bit_depth);
+    fputs(buf, tmpfile_->file());
+    for (unsigned int i = start_; i < limit_; i++) {
+      y4m_write_frame_header(buf, sizeof(buf));
+      fputs(buf, tmpfile_->file());
+      y4m_write_image_file(img(), PLANES_YUV, tmpfile_->file());
+      Next();
+    }
+    ReplaceInputFile(tmpfile_->file());
+  }
+
+  virtual void Init(const std::string &file_name, int limit) {
+    Y4mVideoSourceTest::Init(file_name, limit);
+    WriteY4mAndReadBack();
+  }
+  libaom_test::TempOutFile *tmpfile_;
+};
+
+TEST_P(Y4mVideoWriteTest, WriteTest) {
+  const Y4mTestParam t = GetParam();
+  Init(t.filename, kFrames);
+  HeaderChecks(t.bit_depth, t.format);
+  Md5Check(t.md5raw);
+}
+
+INSTANTIATE_TEST_CASE_P(C, Y4mVideoWriteTest,
+                        ::testing::ValuesIn(kY4mTestVectors));
+}  // namespace
diff --git a/third_party/aom/test/y4m_video_source.h b/third_party/aom/test/y4m_video_source.h
new file mode 100644
index 0000000000..3dea901e6d
--- /dev/null
+++ b/third_party/aom/test/y4m_video_source.h
@@ -0,0 +1,123 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_Y4M_VIDEO_SOURCE_H_
+#define AOM_TEST_Y4M_VIDEO_SOURCE_H_
+#include <algorithm>
+#include <string>
+
+#include "common/y4minput.h"
+#include "test/video_source.h"
+
+namespace libaom_test {
+
+// This class extends VideoSource to allow parsing of raw yv12
+// so that we can do actual file encodes.
+class Y4mVideoSource : public VideoSource {
+ public:
+  Y4mVideoSource(const std::string &file_name, unsigned int start, int limit)
+      : file_name_(file_name), input_file_(NULL), img_(new aom_image_t()),
+        start_(start), limit_(limit), frame_(0), framerate_numerator_(0),
+        framerate_denominator_(0), y4m_() {}
+
+  virtual ~Y4mVideoSource() {
+    aom_img_free(img_.get());
+    CloseSource();
+  }
+
+  virtual void OpenSource() {
+    CloseSource();
+    input_file_ = OpenTestDataFile(file_name_);
+    ASSERT_TRUE(input_file_ != NULL)
+        << "Input file open failed. Filename: " << file_name_;
+  }
+
+  virtual void ReadSourceToStart() {
+    ASSERT_TRUE(input_file_ != NULL);
+    ASSERT_FALSE(y4m_input_open(&y4m_, input_file_, NULL, 0, 0));
+    framerate_numerator_ = y4m_.fps_n;
+    framerate_denominator_ = y4m_.fps_d;
+    frame_ = 0;
+    for (unsigned int i = 0; i < start_; i++) {
+      Next();
+    }
+    FillFrame();
+  }
+
+  virtual void Begin() {
+    OpenSource();
+    ReadSourceToStart();
+  }
+
+  virtual void Next() {
+    ++frame_;
+    FillFrame();
+  }
+
+  virtual aom_image_t *img() const {
+    return (frame_ < limit_) ? img_.get() : NULL;
+  }
+
+  // Models a stream where Timebase = 1/FPS, so pts == frame.
+  virtual aom_codec_pts_t pts() const { return frame_; }
+
+  virtual unsigned long duration() const { return 1; }
+
+  virtual aom_rational_t timebase() const {
+    const aom_rational_t t = { framerate_denominator_, framerate_numerator_ };
+    return t;
+  }
+
+  virtual unsigned int frame() const { return frame_; }
+
+  virtual unsigned int limit() const { return limit_; }
+
+  virtual void FillFrame() {
+    ASSERT_TRUE(input_file_ != NULL);
+    // Read a frame from input_file.
+    y4m_input_fetch_frame(&y4m_, input_file_, img_.get());
+  }
+
+  // Swap buffers with another y4m source. This allows reading a new frame
+  // while keeping the old frame around. A whole Y4mSource is required and
+  // not just a aom_image_t because of how the y4m reader manipulates
+  // aom_image_t internals,
+  void SwapBuffers(Y4mVideoSource *other) {
+    std::swap(other->y4m_.dst_buf, y4m_.dst_buf);
+    aom_image_t *tmp;
+    tmp = other->img_.release();
+    other->img_.reset(img_.release());
+    img_.reset(tmp);
+  }
+
+ protected:
+  void CloseSource() {
+    y4m_input_close(&y4m_);
+    y4m_ = y4m_input();
+    if (input_file_ != NULL) {
+      fclose(input_file_);
+      input_file_ = NULL;
+    }
+  }
+
+  std::string file_name_;
+  FILE *input_file_;
+  testing::internal::scoped_ptr<aom_image_t> img_;
+  unsigned int start_;
+  unsigned int limit_;
+  unsigned int frame_;
+  int framerate_numerator_;
+  int framerate_denominator_;
+  y4m_input y4m_;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_Y4M_VIDEO_SOURCE_H_
diff --git a/third_party/aom/test/yuv_video_source.h b/third_party/aom/test/yuv_video_source.h
new file mode 100644
index 0000000000..774ecc0086
--- /dev/null
+++ b/third_party/aom/test/yuv_video_source.h
@@ -0,0 +1,123 @@
+/*
+ * Copyright (c) 2016, Alliance for Open Media. All rights reserved
+ *
+ * This source code is subject to the terms of the BSD 2 Clause License and
+ * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
+ * was not distributed with this source code in the LICENSE file, you can
+ * obtain it at www.aomedia.org/license/software. If the Alliance for Open
+ * Media Patent License 1.0 was not distributed with this source code in the
+ * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
+ */
+#ifndef AOM_TEST_YUV_VIDEO_SOURCE_H_
+#define AOM_TEST_YUV_VIDEO_SOURCE_H_
+
+#include <cstdio>
+#include <cstdlib>
+#include <string>
+
+#include "test/video_source.h"
+#include "aom/aom_image.h"
+
+namespace libaom_test {
+
+// This class extends VideoSource to allow parsing of raw YUV
+// formats of various color sampling and bit-depths so that we can
+// do actual file encodes.
+class YUVVideoSource : public VideoSource {
+ public:
+  YUVVideoSource(const std::string &file_name, aom_img_fmt format,
+                 unsigned int width, unsigned int height, int rate_numerator,
+                 int rate_denominator, unsigned int start, int limit)
+      : file_name_(file_name), input_file_(NULL), img_(NULL), start_(start),
+        limit_(limit), frame_(0), width_(0), height_(0),
+        format_(AOM_IMG_FMT_NONE), framerate_numerator_(rate_numerator),
+        framerate_denominator_(rate_denominator) {
+    // This initializes format_, raw_size_, width_, height_ and allocates img.
+    SetSize(width, height, format);
+  }
+
+  virtual ~YUVVideoSource() {
+    aom_img_free(img_);
+    if (input_file_) fclose(input_file_);
+  }
+
+  virtual void Begin() {
+    if (input_file_) fclose(input_file_);
+    input_file_ = OpenTestDataFile(file_name_);
+    ASSERT_TRUE(input_file_ != NULL)
+        << "Input file open failed. Filename: " << file_name_;
+    if (start_)
+      fseek(input_file_, static_cast<unsigned>(raw_size_) * start_, SEEK_SET);
+
+    frame_ = start_;
+    FillFrame();
+  }
+
+  virtual void Next() {
+    ++frame_;
+    FillFrame();
+  }
+
+  virtual aom_image_t *img() const { return (frame_ < limit_) ? img_ : NULL; }
+
+  // Models a stream where Timebase = 1/FPS, so pts == frame.
+  virtual aom_codec_pts_t pts() const { return frame_; }
+
+  virtual unsigned long duration() const { return 1; }
+
+  virtual aom_rational_t timebase() const {
+    const aom_rational_t t = { framerate_denominator_, framerate_numerator_ };
+    return t;
+  }
+
+  virtual unsigned int frame() const { return frame_; }
+
+  virtual unsigned int limit() const { return limit_; }
+
+  virtual void SetSize(unsigned int width, unsigned int height,
+                       aom_img_fmt format) {
+    if (width != width_ || height != height_ || format != format_) {
+      aom_img_free(img_);
+      img_ = aom_img_alloc(NULL, format, width, height, 1);
+      ASSERT_TRUE(img_ != NULL);
+      width_ = width;
+      height_ = height;
+      format_ = format;
+      switch (format) {
+        case AOM_IMG_FMT_I420: raw_size_ = width * height * 3 / 2; break;
+        case AOM_IMG_FMT_I422: raw_size_ = width * height * 2; break;
+        case AOM_IMG_FMT_I444: raw_size_ = width * height * 3; break;
+        case AOM_IMG_FMT_I42016: raw_size_ = width * height * 3; break;
+        case AOM_IMG_FMT_I42216: raw_size_ = width * height * 4; break;
+        case AOM_IMG_FMT_I44416: raw_size_ = width * height * 6; break;
+        default: ASSERT_TRUE(0);
+      }
+    }
+  }
+
+  virtual void FillFrame() {
+    ASSERT_TRUE(input_file_ != NULL);
+    // Read a frame from input_file.
+    if (fread(img_->img_data, raw_size_, 1, input_file_) == 0) {
+      limit_ = frame_;
+    }
+  }
+
+ protected:
+  std::string file_name_;
+  FILE *input_file_;
+  aom_image_t *img_;
+  size_t raw_size_;
+  unsigned int start_;
+  unsigned int limit_;
+  unsigned int frame_;
+  unsigned int width_;
+  unsigned int height_;
+  aom_img_fmt format_;
+  int framerate_numerator_;
+  int framerate_denominator_;
+};
+
+}  // namespace libaom_test
+
+#endif  // AOM_TEST_YUV_VIDEO_SOURCE_H_