path: root/media/libvpx/libvpx/vpx_dsp/x86/highbd_convolve_avx2.c

/*
 *  Copyright (c) 2017 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 <immintrin.h>
#include "./vpx_dsp_rtcd.h"
#include "vpx_dsp/x86/convolve.h"
#include "vpx_dsp/x86/convolve_avx2.h"

// -----------------------------------------------------------------------------
// Copy and average

void vpx_highbd_convolve_copy_avx2(const uint16_t *src, ptrdiff_t src_stride,
                                   uint16_t *dst, ptrdiff_t dst_stride,
                                   const InterpKernel *filter, int x0_q4,
                                   int x_step_q4, int y0_q4, int y_step_q4,
                                   int w, int h, int bd) {
  (void)filter;
  (void)x0_q4;
  (void)x_step_q4;
  (void)y0_q4;
  (void)y_step_q4;
  (void)bd;

  assert(w % 4 == 0);
  if (w > 32) {  // w = 64
    do {
      const __m256i p0 = _mm256_loadu_si256((const __m256i *)src);
      const __m256i p1 = _mm256_loadu_si256((const __m256i *)(src + 16));
      const __m256i p2 = _mm256_loadu_si256((const __m256i *)(src + 32));
      const __m256i p3 = _mm256_loadu_si256((const __m256i *)(src + 48));
      src += src_stride;
      _mm256_storeu_si256((__m256i *)dst, p0);
      _mm256_storeu_si256((__m256i *)(dst + 16), p1);
      _mm256_storeu_si256((__m256i *)(dst + 32), p2);
      _mm256_storeu_si256((__m256i *)(dst + 48), p3);
      dst += dst_stride;
      h--;
    } while (h > 0);
  } else if (w > 16) {  // w = 32
    do {
      const __m256i p0 = _mm256_loadu_si256((const __m256i *)src);
      const __m256i p1 = _mm256_loadu_si256((const __m256i *)(src + 16));
      src += src_stride;
      _mm256_storeu_si256((__m256i *)dst, p0);
      _mm256_storeu_si256((__m256i *)(dst + 16), p1);
      dst += dst_stride;
      h--;
    } while (h > 0);
  } else if (w > 8) {  // w = 16
    __m256i p0, p1;
    do {
      p0 = _mm256_loadu_si256((const __m256i *)src);
      src += src_stride;
      p1 = _mm256_loadu_si256((const __m256i *)src);
      src += src_stride;

      _mm256_storeu_si256((__m256i *)dst, p0);
      dst += dst_stride;
      _mm256_storeu_si256((__m256i *)dst, p1);
      dst += dst_stride;
      h -= 2;
    } while (h > 0);
  } else if (w > 4) {  // w = 8
    __m128i p0, p1;
    do {
      p0 = _mm_loadu_si128((const __m128i *)src);
      src += src_stride;
      p1 = _mm_loadu_si128((const __m128i *)src);
      src += src_stride;

      _mm_storeu_si128((__m128i *)dst, p0);
      dst += dst_stride;
      _mm_storeu_si128((__m128i *)dst, p1);
      dst += dst_stride;
      h -= 2;
    } while (h > 0);
  } else {  // w = 4
    __m128i p0, p1;
    do {
      p0 = _mm_loadl_epi64((const __m128i *)src);
      src += src_stride;
      p1 = _mm_loadl_epi64((const __m128i *)src);
      src += src_stride;

      _mm_storel_epi64((__m128i *)dst, p0);
      dst += dst_stride;
      _mm_storel_epi64((__m128i *)dst, p1);
      dst += dst_stride;
      h -= 2;
    } while (h > 0);
  }
}

void vpx_highbd_convolve_avg_avx2(const uint16_t *src, ptrdiff_t src_stride,
                                  uint16_t *dst, ptrdiff_t dst_stride,
                                  const InterpKernel *filter, int x0_q4,
                                  int x_step_q4, int y0_q4, int y_step_q4,
                                  int w, int h, int bd) {
  (void)filter;
  (void)x0_q4;
  (void)x_step_q4;
  (void)y0_q4;
  (void)y_step_q4;
  (void)bd;

  assert(w % 4 == 0);
  if (w > 32) {  // w = 64
    __m256i p0, p1, p2, p3, u0, u1, u2, u3;
    do {
      p0 = _mm256_loadu_si256((const __m256i *)src);
      p1 = _mm256_loadu_si256((const __m256i *)(src + 16));
      p2 = _mm256_loadu_si256((const __m256i *)(src + 32));
      p3 = _mm256_loadu_si256((const __m256i *)(src + 48));
      src += src_stride;
      u0 = _mm256_loadu_si256((const __m256i *)dst);
      u1 = _mm256_loadu_si256((const __m256i *)(dst + 16));
      u2 = _mm256_loadu_si256((const __m256i *)(dst + 32));
      u3 = _mm256_loadu_si256((const __m256i *)(dst + 48));
      _mm256_storeu_si256((__m256i *)dst, _mm256_avg_epu16(p0, u0));
      _mm256_storeu_si256((__m256i *)(dst + 16), _mm256_avg_epu16(p1, u1));
      _mm256_storeu_si256((__m256i *)(dst + 32), _mm256_avg_epu16(p2, u2));
      _mm256_storeu_si256((__m256i *)(dst + 48), _mm256_avg_epu16(p3, u3));
      dst += dst_stride;
      h--;
    } while (h > 0);
  } else if (w > 16) {  // w = 32
    __m256i p0, p1, u0, u1;
    do {
      p0 = _mm256_loadu_si256((const __m256i *)src);
      p1 = _mm256_loadu_si256((const __m256i *)(src + 16));
      src += src_stride;
      u0 = _mm256_loadu_si256((const __m256i *)dst);
      u1 = _mm256_loadu_si256((const __m256i *)(dst + 16));
      _mm256_storeu_si256((__m256i *)dst, _mm256_avg_epu16(p0, u0));
      _mm256_storeu_si256((__m256i *)(dst + 16), _mm256_avg_epu16(p1, u1));
      dst += dst_stride;
      h--;
    } while (h > 0);
  } else if (w > 8) {  // w = 16
    __m256i p0, p1, u0, u1;
    do {
      p0 = _mm256_loadu_si256((const __m256i *)src);
      p1 = _mm256_loadu_si256((const __m256i *)(src + src_stride));
      src += src_stride << 1;
      u0 = _mm256_loadu_si256((const __m256i *)dst);
      u1 = _mm256_loadu_si256((const __m256i *)(dst + dst_stride));

      _mm256_storeu_si256((__m256i *)dst, _mm256_avg_epu16(p0, u0));
      _mm256_storeu_si256((__m256i *)(dst + dst_stride),
                          _mm256_avg_epu16(p1, u1));
      dst += dst_stride << 1;
      h -= 2;
    } while (h > 0);
  } else if (w > 4) {  // w = 8
    __m128i p0, p1, u0, u1;
    do {
      p0 = _mm_loadu_si128((const __m128i *)src);
      p1 = _mm_loadu_si128((const __m128i *)(src + src_stride));
      src += src_stride << 1;
      u0 = _mm_loadu_si128((const __m128i *)dst);
      u1 = _mm_loadu_si128((const __m128i *)(dst + dst_stride));

      _mm_storeu_si128((__m128i *)dst, _mm_avg_epu16(p0, u0));
      _mm_storeu_si128((__m128i *)(dst + dst_stride), _mm_avg_epu16(p1, u1));
      dst += dst_stride << 1;
      h -= 2;
    } while (h > 0);
  } else {  // w = 4
    __m128i p0, p1, u0, u1;
    do {
      p0 = _mm_loadl_epi64((const __m128i *)src);
      p1 = _mm_loadl_epi64((const __m128i *)(src + src_stride));
      src += src_stride << 1;
      u0 = _mm_loadl_epi64((const __m128i *)dst);
      u1 = _mm_loadl_epi64((const __m128i *)(dst + dst_stride));

      _mm_storel_epi64((__m128i *)dst, _mm_avg_epu16(u0, p0));
      _mm_storel_epi64((__m128i *)(dst + dst_stride), _mm_avg_epu16(u1, p1));
      dst += dst_stride << 1;
      h -= 2;
    } while (h > 0);
  }
}

// -----------------------------------------------------------------------------
// Horizontal and vertical filtering

static const uint8_t signal_pattern_0[32] = { 0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6,
                                              7, 6, 7, 8, 9, 0, 1, 2, 3, 2, 3,
                                              4, 5, 4, 5, 6, 7, 6, 7, 8, 9 };

static const uint8_t signal_pattern_1[32] = { 4, 5, 6,  7,  6,  7,  8,  9,
                                              8, 9, 10, 11, 10, 11, 12, 13,
                                              4, 5, 6,  7,  6,  7,  8,  9,
                                              8, 9, 10, 11, 10, 11, 12, 13 };

static const uint8_t signal_pattern_2[32] = { 6,  7,  8,  9,  8,  9,  10, 11,
                                              10, 11, 12, 13, 12, 13, 14, 15,
                                              6,  7,  8,  9,  8,  9,  10, 11,
                                              10, 11, 12, 13, 12, 13, 14, 15 };

static const uint32_t signal_index[8] = { 2, 3, 4, 5, 2, 3, 4, 5 };

#define CONV8_ROUNDING_BITS (7)
#define CONV8_ROUNDING_NUM (1 << (CONV8_ROUNDING_BITS - 1))

// -----------------------------------------------------------------------------
// Horizontal Filtering

static INLINE void pack_pixels(const __m256i *s, __m256i *p /*p[4]*/) {
  const __m256i idx = _mm256_loadu_si256((const __m256i *)signal_index);
  const __m256i sf0 = _mm256_loadu_si256((const __m256i *)signal_pattern_0);
  const __m256i sf1 = _mm256_loadu_si256((const __m256i *)signal_pattern_1);
  const __m256i c = _mm256_permutevar8x32_epi32(*s, idx);

  p[0] = _mm256_shuffle_epi8(*s, sf0);  // x0x6
  p[1] = _mm256_shuffle_epi8(*s, sf1);  // x1x7
  p[2] = _mm256_shuffle_epi8(c, sf0);   // x2x4
  p[3] = _mm256_shuffle_epi8(c, sf1);   // x3x5
}

// Note:
//  Shared by 8x2 and 16x1 block
static INLINE void pack_16_pixels(const __m256i *s0, const __m256i *s1,
                                  __m256i *x /*x[8]*/) {
  __m256i pp[8];
  pack_pixels(s0, pp);
  pack_pixels(s1, &pp[4]);
  x[0] = _mm256_permute2x128_si256(pp[0], pp[4], 0x20);
  x[1] = _mm256_permute2x128_si256(pp[1], pp[5], 0x20);
  x[2] = _mm256_permute2x128_si256(pp[2], pp[6], 0x20);
  x[3] = _mm256_permute2x128_si256(pp[3], pp[7], 0x20);
  x[4] = x[2];
  x[5] = x[3];
  x[6] = _mm256_permute2x128_si256(pp[0], pp[4], 0x31);
  x[7] = _mm256_permute2x128_si256(pp[1], pp[5], 0x31);
}

static INLINE void pack_8x1_pixels(const uint16_t *src, __m256i *x) {
  __m256i pp[8];
  __m256i s0;
  s0 = _mm256_loadu_si256((const __m256i *)src);
  pack_pixels(&s0, pp);
  x[0] = _mm256_permute2x128_si256(pp[0], pp[2], 0x30);
  x[1] = _mm256_permute2x128_si256(pp[1], pp[3], 0x30);
  x[2] = _mm256_permute2x128_si256(pp[2], pp[0], 0x30);
  x[3] = _mm256_permute2x128_si256(pp[3], pp[1], 0x30);
}

static INLINE void pack_8x2_pixels(const uint16_t *src, ptrdiff_t stride,
                                   __m256i *x) {
  __m256i s0, s1;
  s0 = _mm256_loadu_si256((const __m256i *)src);
  s1 = _mm256_loadu_si256((const __m256i *)(src + stride));
  pack_16_pixels(&s0, &s1, x);
}

static INLINE void pack_16x1_pixels(const uint16_t *src, __m256i *x) {
  __m256i s0, s1;
  s0 = _mm256_loadu_si256((const __m256i *)src);
  s1 = _mm256_loadu_si256((const __m256i *)(src + 8));
  pack_16_pixels(&s0, &s1, x);
}

// Note:
//  Shared by horizontal and vertical filtering
static INLINE void pack_filters(const int16_t *filter, __m256i *f /*f[4]*/) {
  const __m128i h = _mm_loadu_si128((const __m128i *)filter);
  const __m256i hh = _mm256_insertf128_si256(_mm256_castsi128_si256(h), h, 1);
  const __m256i p0 = _mm256_set1_epi32(0x03020100);
  const __m256i p1 = _mm256_set1_epi32(0x07060504);
  const __m256i p2 = _mm256_set1_epi32(0x0b0a0908);
  const __m256i p3 = _mm256_set1_epi32(0x0f0e0d0c);
  f[0] = _mm256_shuffle_epi8(hh, p0);
  f[1] = _mm256_shuffle_epi8(hh, p1);
  f[2] = _mm256_shuffle_epi8(hh, p2);
  f[3] = _mm256_shuffle_epi8(hh, p3);
}

static INLINE void filter_8x1_pixels(const __m256i *sig /*sig[4]*/,
                                     const __m256i *fil /*fil[4]*/,
                                     __m256i *y) {
  __m256i a, a0, a1;

  a0 = _mm256_madd_epi16(fil[0], sig[0]);
  a1 = _mm256_madd_epi16(fil[3], sig[3]);
  a = _mm256_add_epi32(a0, a1);

  a0 = _mm256_madd_epi16(fil[1], sig[1]);
  a1 = _mm256_madd_epi16(fil[2], sig[2]);

  {
    const __m256i min = _mm256_min_epi32(a0, a1);
    a = _mm256_add_epi32(a, min);
  }
  {
    const __m256i max = _mm256_max_epi32(a0, a1);
    a = _mm256_add_epi32(a, max);
  }
  {
    const __m256i rounding = _mm256_set1_epi32(1 << (CONV8_ROUNDING_BITS - 1));
    a = _mm256_add_epi32(a, rounding);
    *y = _mm256_srai_epi32(a, CONV8_ROUNDING_BITS);
  }
}

static INLINE void store_8x1_pixels(const __m256i *y, const __m256i *mask,
                                    uint16_t *dst) {
  const __m128i a0 = _mm256_castsi256_si128(*y);
  const __m128i a1 = _mm256_extractf128_si256(*y, 1);
  __m128i res = _mm_packus_epi32(a0, a1);
  res = _mm_min_epi16(res, _mm256_castsi256_si128(*mask));
  _mm_storeu_si128((__m128i *)dst, res);
}

static INLINE void store_8x2_pixels(const __m256i *y0, const __m256i *y1,
                                    const __m256i *mask, uint16_t *dst,
                                    ptrdiff_t pitch) {
  __m256i a = _mm256_packus_epi32(*y0, *y1);
  a = _mm256_min_epi16(a, *mask);
  _mm_storeu_si128((__m128i *)dst, _mm256_castsi256_si128(a));
  _mm_storeu_si128((__m128i *)(dst + pitch), _mm256_extractf128_si256(a, 1));
}

static INLINE void store_16x1_pixels(const __m256i *y0, const __m256i *y1,
                                     const __m256i *mask, uint16_t *dst) {
  __m256i a = _mm256_packus_epi32(*y0, *y1);
  a = _mm256_min_epi16(a, *mask);
  _mm256_storeu_si256((__m256i *)dst, a);
}

static void vpx_highbd_filter_block1d8_h8_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[8], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff[4];
  pack_filters(filter, ff);

  src_ptr -= 3;
  do {
    pack_8x2_pixels(src_ptr, src_pitch, signal);
    filter_8x1_pixels(signal, ff, &res0);
    filter_8x1_pixels(&signal[4], ff, &res1);
    store_8x2_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
    height -= 2;
    src_ptr += src_pitch << 1;
    dst_ptr += dst_pitch << 1;
  } while (height > 1);

  if (height > 0) {
    pack_8x1_pixels(src_ptr, signal);
    filter_8x1_pixels(signal, ff, &res0);
    store_8x1_pixels(&res0, &max, dst_ptr);
  }
}

static void vpx_highbd_filter_block1d16_h8_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[8], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff[4];
  pack_filters(filter, ff);

  src_ptr -= 3;
  do {
    pack_16x1_pixels(src_ptr, signal);
    filter_8x1_pixels(signal, ff, &res0);
    filter_8x1_pixels(&signal[4], ff, &res1);
    store_16x1_pixels(&res0, &res1, &max, dst_ptr);
    height -= 1;
    src_ptr += src_pitch;
    dst_ptr += dst_pitch;
  } while (height > 0);
}

// -----------------------------------------------------------------------------
// 2-tap horizontal filtering

static INLINE void pack_2t_filter(const int16_t *filter, __m256i *f) {
  const __m128i h = _mm_loadu_si128((const __m128i *)filter);
  const __m256i hh = _mm256_insertf128_si256(_mm256_castsi128_si256(h), h, 1);
  const __m256i p = _mm256_set1_epi32(0x09080706);
  f[0] = _mm256_shuffle_epi8(hh, p);
}

// can be used by pack_8x2_2t_pixels() and pack_16x1_2t_pixels()
// the difference is s0/s1 specifies first and second rows or,
// first 16 samples and 8-sample shifted 16 samples
static INLINE void pack_16_2t_pixels(const __m256i *s0, const __m256i *s1,
                                     __m256i *sig) {
  const __m256i idx = _mm256_loadu_si256((const __m256i *)signal_index);
  const __m256i sf2 = _mm256_loadu_si256((const __m256i *)signal_pattern_2);
  __m256i x0 = _mm256_shuffle_epi8(*s0, sf2);
  __m256i x1 = _mm256_shuffle_epi8(*s1, sf2);
  __m256i r0 = _mm256_permutevar8x32_epi32(*s0, idx);
  __m256i r1 = _mm256_permutevar8x32_epi32(*s1, idx);
  r0 = _mm256_shuffle_epi8(r0, sf2);
  r1 = _mm256_shuffle_epi8(r1, sf2);
  sig[0] = _mm256_permute2x128_si256(x0, x1, 0x20);
  sig[1] = _mm256_permute2x128_si256(r0, r1, 0x20);
}

static INLINE void pack_8x2_2t_pixels(const uint16_t *src,
                                      const ptrdiff_t pitch, __m256i *sig) {
  const __m256i r0 = _mm256_loadu_si256((const __m256i *)src);
  const __m256i r1 = _mm256_loadu_si256((const __m256i *)(src + pitch));
  pack_16_2t_pixels(&r0, &r1, sig);
}

static INLINE void pack_16x1_2t_pixels(const uint16_t *src,
                                       __m256i *sig /*sig[2]*/) {
  const __m256i r0 = _mm256_loadu_si256((const __m256i *)src);
  const __m256i r1 = _mm256_loadu_si256((const __m256i *)(src + 8));
  pack_16_2t_pixels(&r0, &r1, sig);
}

static INLINE void pack_8x1_2t_pixels(const uint16_t *src,
                                      __m256i *sig /*sig[2]*/) {
  const __m256i idx = _mm256_loadu_si256((const __m256i *)signal_index);
  const __m256i sf2 = _mm256_loadu_si256((const __m256i *)signal_pattern_2);
  __m256i r0 = _mm256_loadu_si256((const __m256i *)src);
  __m256i x0 = _mm256_shuffle_epi8(r0, sf2);
  r0 = _mm256_permutevar8x32_epi32(r0, idx);
  r0 = _mm256_shuffle_epi8(r0, sf2);
  sig[0] = _mm256_permute2x128_si256(x0, r0, 0x20);
}

// can be used by filter_8x2_2t_pixels() and filter_16x1_2t_pixels()
static INLINE void filter_16_2t_pixels(const __m256i *sig, const __m256i *f,
                                       __m256i *y0, __m256i *y1) {
  const __m256i rounding = _mm256_set1_epi32(1 << (CONV8_ROUNDING_BITS - 1));
  __m256i x0 = _mm256_madd_epi16(sig[0], *f);
  __m256i x1 = _mm256_madd_epi16(sig[1], *f);
  x0 = _mm256_add_epi32(x0, rounding);
  x1 = _mm256_add_epi32(x1, rounding);
  *y0 = _mm256_srai_epi32(x0, CONV8_ROUNDING_BITS);
  *y1 = _mm256_srai_epi32(x1, CONV8_ROUNDING_BITS);
}

static INLINE void filter_8x1_2t_pixels(const __m256i *sig, const __m256i *f,
                                        __m256i *y0) {
  const __m256i rounding = _mm256_set1_epi32(1 << (CONV8_ROUNDING_BITS - 1));
  __m256i x0 = _mm256_madd_epi16(sig[0], *f);
  x0 = _mm256_add_epi32(x0, rounding);
  *y0 = _mm256_srai_epi32(x0, CONV8_ROUNDING_BITS);
}

static void vpx_highbd_filter_block1d8_h2_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[2], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff;
  pack_2t_filter(filter, &ff);

  src_ptr -= 3;
  do {
    pack_8x2_2t_pixels(src_ptr, src_pitch, signal);
    filter_16_2t_pixels(signal, &ff, &res0, &res1);
    store_8x2_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
    height -= 2;
    src_ptr += src_pitch << 1;
    dst_ptr += dst_pitch << 1;
  } while (height > 1);

  if (height > 0) {
    pack_8x1_2t_pixels(src_ptr, signal);
    filter_8x1_2t_pixels(signal, &ff, &res0);
    store_8x1_pixels(&res0, &max, dst_ptr);
  }
}

static void vpx_highbd_filter_block1d16_h2_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[2], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff;
  pack_2t_filter(filter, &ff);

  src_ptr -= 3;
  do {
    pack_16x1_2t_pixels(src_ptr, signal);
    filter_16_2t_pixels(signal, &ff, &res0, &res1);
    store_16x1_pixels(&res0, &res1, &max, dst_ptr);
    height -= 1;
    src_ptr += src_pitch;
    dst_ptr += dst_pitch;
  } while (height > 0);
}

// -----------------------------------------------------------------------------
// Vertical Filtering

static void pack_8x9_init(const uint16_t *src, ptrdiff_t pitch, __m256i *sig) {
  __m256i s0 = _mm256_castsi128_si256(_mm_loadu_si128((const __m128i *)src));
  __m256i s1 =
      _mm256_castsi128_si256(_mm_loadu_si128((const __m128i *)(src + pitch)));
  __m256i s2 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src + 2 * pitch)));
  __m256i s3 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src + 3 * pitch)));
  __m256i s4 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src + 4 * pitch)));
  __m256i s5 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src + 5 * pitch)));
  __m256i s6 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src + 6 * pitch)));

  s0 = _mm256_inserti128_si256(s0, _mm256_castsi256_si128(s1), 1);
  s1 = _mm256_inserti128_si256(s1, _mm256_castsi256_si128(s2), 1);
  s2 = _mm256_inserti128_si256(s2, _mm256_castsi256_si128(s3), 1);
  s3 = _mm256_inserti128_si256(s3, _mm256_castsi256_si128(s4), 1);
  s4 = _mm256_inserti128_si256(s4, _mm256_castsi256_si128(s5), 1);
  s5 = _mm256_inserti128_si256(s5, _mm256_castsi256_si128(s6), 1);

  sig[0] = _mm256_unpacklo_epi16(s0, s1);
  sig[4] = _mm256_unpackhi_epi16(s0, s1);
  sig[1] = _mm256_unpacklo_epi16(s2, s3);
  sig[5] = _mm256_unpackhi_epi16(s2, s3);
  sig[2] = _mm256_unpacklo_epi16(s4, s5);
  sig[6] = _mm256_unpackhi_epi16(s4, s5);
  sig[8] = s6;
}

static INLINE void pack_8x9_pixels(const uint16_t *src, ptrdiff_t pitch,
                                   __m256i *sig) {
  // base + 7th row
  __m256i s0 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src + 7 * pitch)));
  // base + 8th row
  __m256i s1 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src + 8 * pitch)));
  __m256i s2 = _mm256_inserti128_si256(sig[8], _mm256_castsi256_si128(s0), 1);
  __m256i s3 = _mm256_inserti128_si256(s0, _mm256_castsi256_si128(s1), 1);
  sig[3] = _mm256_unpacklo_epi16(s2, s3);
  sig[7] = _mm256_unpackhi_epi16(s2, s3);
  sig[8] = s1;
}

static INLINE void filter_8x9_pixels(const __m256i *sig, const __m256i *f,
                                     __m256i *y0, __m256i *y1) {
  filter_8x1_pixels(sig, f, y0);
  filter_8x1_pixels(&sig[4], f, y1);
}

static INLINE void update_pixels(__m256i *sig) {
  int i;
  for (i = 0; i < 3; ++i) {
    sig[i] = sig[i + 1];
    sig[i + 4] = sig[i + 5];
  }
}

static void vpx_highbd_filter_block1d8_v8_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[9], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff[4];
  pack_filters(filter, ff);

  pack_8x9_init(src_ptr, src_pitch, signal);

  do {
    pack_8x9_pixels(src_ptr, src_pitch, signal);

    filter_8x9_pixels(signal, ff, &res0, &res1);
    store_8x2_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
    update_pixels(signal);

    src_ptr += src_pitch << 1;
    dst_ptr += dst_pitch << 1;
    height -= 2;
  } while (height > 0);
}

static void pack_16x9_init(const uint16_t *src, ptrdiff_t pitch, __m256i *sig) {
  __m256i u0, u1, u2, u3;
  // load 0-6 rows
  const __m256i s0 = _mm256_loadu_si256((const __m256i *)src);
  const __m256i s1 = _mm256_loadu_si256((const __m256i *)(src + pitch));
  const __m256i s2 = _mm256_loadu_si256((const __m256i *)(src + 2 * pitch));
  const __m256i s3 = _mm256_loadu_si256((const __m256i *)(src + 3 * pitch));
  const __m256i s4 = _mm256_loadu_si256((const __m256i *)(src + 4 * pitch));
  const __m256i s5 = _mm256_loadu_si256((const __m256i *)(src + 5 * pitch));
  const __m256i s6 = _mm256_loadu_si256((const __m256i *)(src + 6 * pitch));

  u0 = _mm256_permute2x128_si256(s0, s1, 0x20);  // 0, 1 low
  u1 = _mm256_permute2x128_si256(s0, s1, 0x31);  // 0, 1 high

  u2 = _mm256_permute2x128_si256(s1, s2, 0x20);  // 1, 2 low
  u3 = _mm256_permute2x128_si256(s1, s2, 0x31);  // 1, 2 high

  sig[0] = _mm256_unpacklo_epi16(u0, u2);
  sig[4] = _mm256_unpackhi_epi16(u0, u2);

  sig[8] = _mm256_unpacklo_epi16(u1, u3);
  sig[12] = _mm256_unpackhi_epi16(u1, u3);

  u0 = _mm256_permute2x128_si256(s2, s3, 0x20);
  u1 = _mm256_permute2x128_si256(s2, s3, 0x31);

  u2 = _mm256_permute2x128_si256(s3, s4, 0x20);
  u3 = _mm256_permute2x128_si256(s3, s4, 0x31);

  sig[1] = _mm256_unpacklo_epi16(u0, u2);
  sig[5] = _mm256_unpackhi_epi16(u0, u2);

  sig[9] = _mm256_unpacklo_epi16(u1, u3);
  sig[13] = _mm256_unpackhi_epi16(u1, u3);

  u0 = _mm256_permute2x128_si256(s4, s5, 0x20);
  u1 = _mm256_permute2x128_si256(s4, s5, 0x31);

  u2 = _mm256_permute2x128_si256(s5, s6, 0x20);
  u3 = _mm256_permute2x128_si256(s5, s6, 0x31);

  sig[2] = _mm256_unpacklo_epi16(u0, u2);
  sig[6] = _mm256_unpackhi_epi16(u0, u2);

  sig[10] = _mm256_unpacklo_epi16(u1, u3);
  sig[14] = _mm256_unpackhi_epi16(u1, u3);

  sig[16] = s6;
}

static void pack_16x9_pixels(const uint16_t *src, ptrdiff_t pitch,
                             __m256i *sig) {
  // base + 7th row
  const __m256i s7 = _mm256_loadu_si256((const __m256i *)(src + 7 * pitch));
  // base + 8th row
  const __m256i s8 = _mm256_loadu_si256((const __m256i *)(src + 8 * pitch));

  __m256i u0, u1, u2, u3;
  u0 = _mm256_permute2x128_si256(sig[16], s7, 0x20);
  u1 = _mm256_permute2x128_si256(sig[16], s7, 0x31);

  u2 = _mm256_permute2x128_si256(s7, s8, 0x20);
  u3 = _mm256_permute2x128_si256(s7, s8, 0x31);

  sig[3] = _mm256_unpacklo_epi16(u0, u2);
  sig[7] = _mm256_unpackhi_epi16(u0, u2);

  sig[11] = _mm256_unpacklo_epi16(u1, u3);
  sig[15] = _mm256_unpackhi_epi16(u1, u3);

  sig[16] = s8;
}

static INLINE void filter_16x9_pixels(const __m256i *sig, const __m256i *f,
                                      __m256i *y0, __m256i *y1) {
  __m256i res[4];
  int i;
  for (i = 0; i < 4; ++i) {
    filter_8x1_pixels(&sig[i << 2], f, &res[i]);
  }

  {
    const __m256i l0l1 = _mm256_packus_epi32(res[0], res[1]);
    const __m256i h0h1 = _mm256_packus_epi32(res[2], res[3]);
    *y0 = _mm256_permute2x128_si256(l0l1, h0h1, 0x20);
    *y1 = _mm256_permute2x128_si256(l0l1, h0h1, 0x31);
  }
}

static INLINE void store_16x2_pixels(const __m256i *y0, const __m256i *y1,
                                     const __m256i *mask, uint16_t *dst,
                                     ptrdiff_t pitch) {
  __m256i p = _mm256_min_epi16(*y0, *mask);
  _mm256_storeu_si256((__m256i *)dst, p);
  p = _mm256_min_epi16(*y1, *mask);
  _mm256_storeu_si256((__m256i *)(dst + pitch), p);
}

static void update_16x9_pixels(__m256i *sig) {
  update_pixels(&sig[0]);
  update_pixels(&sig[8]);
}

static void vpx_highbd_filter_block1d16_v8_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[17], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff[4];
  pack_filters(filter, ff);

  pack_16x9_init(src_ptr, src_pitch, signal);

  do {
    pack_16x9_pixels(src_ptr, src_pitch, signal);
    filter_16x9_pixels(signal, ff, &res0, &res1);
    store_16x2_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
    update_16x9_pixels(signal);

    src_ptr += src_pitch << 1;
    dst_ptr += dst_pitch << 1;
    height -= 2;
  } while (height > 0);
}

// -----------------------------------------------------------------------------
// 2-tap vertical filtering

static void pack_16x2_init(const uint16_t *src, __m256i *sig) {
  sig[2] = _mm256_loadu_si256((const __m256i *)src);
}

static INLINE void pack_16x2_2t_pixels(const uint16_t *src, ptrdiff_t pitch,
                                       __m256i *sig) {
  // load the next row
  const __m256i u = _mm256_loadu_si256((const __m256i *)(src + pitch));
  sig[0] = _mm256_unpacklo_epi16(sig[2], u);
  sig[1] = _mm256_unpackhi_epi16(sig[2], u);
  sig[2] = u;
}

static INLINE void filter_16x2_2t_pixels(const __m256i *sig, const __m256i *f,
                                         __m256i *y0, __m256i *y1) {
  filter_16_2t_pixels(sig, f, y0, y1);
}

static void vpx_highbd_filter_block1d16_v2_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[3], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
  __m256i ff;

  pack_2t_filter(filter, &ff);
  pack_16x2_init(src_ptr, signal);

  do {
    pack_16x2_2t_pixels(src_ptr, src_pitch, signal);
    filter_16x2_2t_pixels(signal, &ff, &res0, &res1);
    store_16x1_pixels(&res0, &res1, &max, dst_ptr);

    src_ptr += src_pitch;
    dst_ptr += dst_pitch;
    height -= 1;
  } while (height > 0);
}

static INLINE void pack_8x1_2t_filter(const int16_t *filter, __m128i *f) {
  const __m128i h = _mm_loadu_si128((const __m128i *)filter);
  const __m128i p = _mm_set1_epi32(0x09080706);
  f[0] = _mm_shuffle_epi8(h, p);
}

static void pack_8x2_init(const uint16_t *src, __m128i *sig) {
  sig[2] = _mm_loadu_si128((const __m128i *)src);
}

static INLINE void pack_8x2_2t_pixels_ver(const uint16_t *src, ptrdiff_t pitch,
                                          __m128i *sig) {
  // load the next row
  const __m128i u = _mm_loadu_si128((const __m128i *)(src + pitch));
  sig[0] = _mm_unpacklo_epi16(sig[2], u);
  sig[1] = _mm_unpackhi_epi16(sig[2], u);
  sig[2] = u;
}

static INLINE void filter_8_2t_pixels(const __m128i *sig, const __m128i *f,
                                      __m128i *y0, __m128i *y1) {
  const __m128i rounding = _mm_set1_epi32(1 << (CONV8_ROUNDING_BITS - 1));
  __m128i x0 = _mm_madd_epi16(sig[0], *f);
  __m128i x1 = _mm_madd_epi16(sig[1], *f);
  x0 = _mm_add_epi32(x0, rounding);
  x1 = _mm_add_epi32(x1, rounding);
  *y0 = _mm_srai_epi32(x0, CONV8_ROUNDING_BITS);
  *y1 = _mm_srai_epi32(x1, CONV8_ROUNDING_BITS);
}

static INLINE void store_8x1_2t_pixels_ver(const __m128i *y0, const __m128i *y1,
                                           const __m128i *mask, uint16_t *dst) {
  __m128i res = _mm_packus_epi32(*y0, *y1);
  res = _mm_min_epi16(res, *mask);
  _mm_storeu_si128((__m128i *)dst, res);
}

static void vpx_highbd_filter_block1d8_v2_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m128i signal[3], res0, res1;
  const __m128i max = _mm_set1_epi16((1 << bd) - 1);
  __m128i ff;

  pack_8x1_2t_filter(filter, &ff);
  pack_8x2_init(src_ptr, signal);

  do {
    pack_8x2_2t_pixels_ver(src_ptr, src_pitch, signal);
    filter_8_2t_pixels(signal, &ff, &res0, &res1);
    store_8x1_2t_pixels_ver(&res0, &res1, &max, dst_ptr);

    src_ptr += src_pitch;
    dst_ptr += dst_pitch;
    height -= 1;
  } while (height > 0);
}

// Calculation with averaging the input pixels

static INLINE void store_8x1_avg_pixels(const __m256i *y0, const __m256i *mask,
                                        uint16_t *dst) {
  const __m128i a0 = _mm256_castsi256_si128(*y0);
  const __m128i a1 = _mm256_extractf128_si256(*y0, 1);
  __m128i res = _mm_packus_epi32(a0, a1);
  const __m128i pix = _mm_loadu_si128((const __m128i *)dst);
  res = _mm_min_epi16(res, _mm256_castsi256_si128(*mask));
  res = _mm_avg_epu16(res, pix);
  _mm_storeu_si128((__m128i *)dst, res);
}

static INLINE void store_8x2_avg_pixels(const __m256i *y0, const __m256i *y1,
                                        const __m256i *mask, uint16_t *dst,
                                        ptrdiff_t pitch) {
  __m256i a = _mm256_packus_epi32(*y0, *y1);
  const __m128i pix0 = _mm_loadu_si128((const __m128i *)dst);
  const __m128i pix1 = _mm_loadu_si128((const __m128i *)(dst + pitch));
  const __m256i pix =
      _mm256_insertf128_si256(_mm256_castsi128_si256(pix0), pix1, 1);
  a = _mm256_min_epi16(a, *mask);
  a = _mm256_avg_epu16(a, pix);
  _mm_storeu_si128((__m128i *)dst, _mm256_castsi256_si128(a));
  _mm_storeu_si128((__m128i *)(dst + pitch), _mm256_extractf128_si256(a, 1));
}

static INLINE void store_16x1_avg_pixels(const __m256i *y0, const __m256i *y1,
                                         const __m256i *mask, uint16_t *dst) {
  __m256i a = _mm256_packus_epi32(*y0, *y1);
  const __m256i pix = _mm256_loadu_si256((const __m256i *)dst);
  a = _mm256_min_epi16(a, *mask);
  a = _mm256_avg_epu16(a, pix);
  _mm256_storeu_si256((__m256i *)dst, a);
}

static INLINE void store_16x2_avg_pixels(const __m256i *y0, const __m256i *y1,
                                         const __m256i *mask, uint16_t *dst,
                                         ptrdiff_t pitch) {
  const __m256i pix0 = _mm256_loadu_si256((const __m256i *)dst);
  const __m256i pix1 = _mm256_loadu_si256((const __m256i *)(dst + pitch));
  __m256i p = _mm256_min_epi16(*y0, *mask);
  p = _mm256_avg_epu16(p, pix0);
  _mm256_storeu_si256((__m256i *)dst, p);

  p = _mm256_min_epi16(*y1, *mask);
  p = _mm256_avg_epu16(p, pix1);
  _mm256_storeu_si256((__m256i *)(dst + pitch), p);
}

static INLINE void store_8x1_2t_avg_pixels_ver(const __m128i *y0,
                                               const __m128i *y1,
                                               const __m128i *mask,
                                               uint16_t *dst) {
  __m128i res = _mm_packus_epi32(*y0, *y1);
  const __m128i pix = _mm_loadu_si128((const __m128i *)dst);
  res = _mm_min_epi16(res, *mask);
  res = _mm_avg_epu16(res, pix);
  _mm_storeu_si128((__m128i *)dst, res);
}

static void vpx_highbd_filter_block1d8_h8_avg_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[8], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff[4];
  pack_filters(filter, ff);

  src_ptr -= 3;
  do {
    pack_8x2_pixels(src_ptr, src_pitch, signal);
    filter_8x1_pixels(signal, ff, &res0);
    filter_8x1_pixels(&signal[4], ff, &res1);
    store_8x2_avg_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
    height -= 2;
    src_ptr += src_pitch << 1;
    dst_ptr += dst_pitch << 1;
  } while (height > 1);

  if (height > 0) {
    pack_8x1_pixels(src_ptr, signal);
    filter_8x1_pixels(signal, ff, &res0);
    store_8x1_avg_pixels(&res0, &max, dst_ptr);
  }
}

static void vpx_highbd_filter_block1d16_h8_avg_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[8], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff[4];
  pack_filters(filter, ff);

  src_ptr -= 3;
  do {
    pack_16x1_pixels(src_ptr, signal);
    filter_8x1_pixels(signal, ff, &res0);
    filter_8x1_pixels(&signal[4], ff, &res1);
    store_16x1_avg_pixels(&res0, &res1, &max, dst_ptr);
    height -= 1;
    src_ptr += src_pitch;
    dst_ptr += dst_pitch;
  } while (height > 0);
}

static void vpx_highbd_filter_block1d4_h4_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr,
    ptrdiff_t dst_stride, uint32_t height, const int16_t *kernel, int bd) {
  // We extract the middle four elements of the kernel into two registers in
  // the form
  // ... k[3] k[2] k[3] k[2]
  // ... k[5] k[4] k[5] k[4]
  // Then we shuffle the source into
  // ... s[1] s[0] s[0] s[-1]
  // ... s[3] s[2] s[2] s[1]
  // Calling multiply and add gives us half of the sum. Calling add on the two
  // halves gives us the output. Since avx2 allows us to use 256-bit buffer, we
  // can do this two rows at a time.

  __m256i src_reg, src_reg_shift_0, src_reg_shift_2;
  __m256i res_reg;
  __m256i idx_shift_0 =
      _mm256_setr_epi8(0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8, 9, 0, 1, 2,
                       3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8, 9);
  __m256i idx_shift_2 =
      _mm256_setr_epi8(4, 5, 6, 7, 6, 7, 8, 9, 8, 9, 10, 11, 10, 11, 12, 13, 4,
                       5, 6, 7, 6, 7, 8, 9, 8, 9, 10, 11, 10, 11, 12, 13);

  __m128i kernel_reg_128;  // Kernel
  __m256i kernel_reg, kernel_reg_23,
      kernel_reg_45;  // Segments of the kernel used
  const __m256i reg_round =
      _mm256_set1_epi32(CONV8_ROUNDING_NUM);  // Used for rounding
  const __m256i reg_max = _mm256_set1_epi16((1 << bd) - 1);
  const ptrdiff_t unrolled_src_stride = src_stride << 1;
  const ptrdiff_t unrolled_dst_stride = dst_stride << 1;
  int h;

  // Start one pixel before as we need tap/2 - 1 = 1 sample from the past
  src_ptr -= 1;

  // Load Kernel
  kernel_reg_128 = _mm_loadu_si128((const __m128i *)kernel);
  kernel_reg = _mm256_broadcastsi128_si256(kernel_reg_128);
  kernel_reg_23 = _mm256_shuffle_epi32(kernel_reg, 0x55);
  kernel_reg_45 = _mm256_shuffle_epi32(kernel_reg, 0xaa);

  for (h = height; h >= 2; h -= 2) {
    // Load the source
    src_reg = mm256_loadu2_si128(src_ptr, src_ptr + src_stride);
    src_reg_shift_0 = _mm256_shuffle_epi8(src_reg, idx_shift_0);
    src_reg_shift_2 = _mm256_shuffle_epi8(src_reg, idx_shift_2);

    // Get the output
    res_reg = mm256_madd_add_epi32(&src_reg_shift_0, &src_reg_shift_2,
                                   &kernel_reg_23, &kernel_reg_45);

    // Round the result
    res_reg = mm256_round_epi32(&res_reg, &reg_round, CONV8_ROUNDING_BITS);

    // Finally combine to get the final dst
    res_reg = _mm256_packus_epi32(res_reg, res_reg);
    res_reg = _mm256_min_epi16(res_reg, reg_max);
    mm256_storeu2_epi64((__m128i *)dst_ptr, (__m128i *)(dst_ptr + dst_stride),
                        &res_reg);

    src_ptr += unrolled_src_stride;
    dst_ptr += unrolled_dst_stride;
  }

  // Repeat for the last row if needed
  if (h > 0) {
    // Load the source
    src_reg = mm256_loadu2_si128(src_ptr, src_ptr + 4);
    src_reg_shift_0 = _mm256_shuffle_epi8(src_reg, idx_shift_0);
    src_reg_shift_2 = _mm256_shuffle_epi8(src_reg, idx_shift_2);

    // Get the output
    res_reg = mm256_madd_add_epi32(&src_reg_shift_0, &src_reg_shift_2,
                                   &kernel_reg_23, &kernel_reg_45);

    // Round the result
    res_reg = mm256_round_epi32(&res_reg, &reg_round, CONV8_ROUNDING_BITS);

    // Finally combine to get the final dst
    res_reg = _mm256_packus_epi32(res_reg, res_reg);
    res_reg = _mm256_min_epi16(res_reg, reg_max);
    _mm_storel_epi64((__m128i *)dst_ptr, _mm256_castsi256_si128(res_reg));
  }
}

static void vpx_highbd_filter_block1d8_h4_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr,
    ptrdiff_t dst_stride, uint32_t height, const int16_t *kernel, int bd) {
  // We will extract the middle four elements of the kernel into two registers
  // in the form
  // ... k[3] k[2] k[3] k[2]
  // ... k[5] k[4] k[5] k[4]
  // Then we shuffle the source into
  // ... s[1] s[0] s[0] s[-1]
  // ... s[3] s[2] s[2] s[1]
  // Calling multiply and add gives us half of the sum of the first half.
  // Calling add gives us first half of the output. Repat again to get the whole
  // output. Since avx2 allows us to use 256-bit buffer, we can do this two rows
  // at a time.

  __m256i src_reg, src_reg_shift_0, src_reg_shift_2;
  __m256i res_reg, res_first, res_last;
  __m256i idx_shift_0 =
      _mm256_setr_epi8(0, 1, 2, 3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8, 9, 0, 1, 2,
                       3, 2, 3, 4, 5, 4, 5, 6, 7, 6, 7, 8, 9);
  __m256i idx_shift_2 =
      _mm256_setr_epi8(4, 5, 6, 7, 6, 7, 8, 9, 8, 9, 10, 11, 10, 11, 12, 13, 4,
                       5, 6, 7, 6, 7, 8, 9, 8, 9, 10, 11, 10, 11, 12, 13);

  __m128i kernel_reg_128;  // Kernel
  __m256i kernel_reg, kernel_reg_23,
      kernel_reg_45;  // Segments of the kernel used
  const __m256i reg_round =
      _mm256_set1_epi32(CONV8_ROUNDING_NUM);  // Used for rounding
  const __m256i reg_max = _mm256_set1_epi16((1 << bd) - 1);
  const ptrdiff_t unrolled_src_stride = src_stride << 1;
  const ptrdiff_t unrolled_dst_stride = dst_stride << 1;
  int h;

  // Start one pixel before as we need tap/2 - 1 = 1 sample from the past
  src_ptr -= 1;

  // Load Kernel
  kernel_reg_128 = _mm_loadu_si128((const __m128i *)kernel);
  kernel_reg = _mm256_broadcastsi128_si256(kernel_reg_128);
  kernel_reg_23 = _mm256_shuffle_epi32(kernel_reg, 0x55);
  kernel_reg_45 = _mm256_shuffle_epi32(kernel_reg, 0xaa);

  for (h = height; h >= 2; h -= 2) {
    // Load the source
    src_reg = mm256_loadu2_si128(src_ptr, src_ptr + src_stride);
    src_reg_shift_0 = _mm256_shuffle_epi8(src_reg, idx_shift_0);
    src_reg_shift_2 = _mm256_shuffle_epi8(src_reg, idx_shift_2);

    // Result for first half
    res_first = mm256_madd_add_epi32(&src_reg_shift_0, &src_reg_shift_2,
                                     &kernel_reg_23, &kernel_reg_45);

    // Do again to get the second half of dst
    // Load the source
    src_reg = mm256_loadu2_si128(src_ptr + 4, src_ptr + src_stride + 4);
    src_reg_shift_0 = _mm256_shuffle_epi8(src_reg, idx_shift_0);
    src_reg_shift_2 = _mm256_shuffle_epi8(src_reg, idx_shift_2);

    // Result for second half
    res_last = mm256_madd_add_epi32(&src_reg_shift_0, &src_reg_shift_2,
                                    &kernel_reg_23, &kernel_reg_45);

    // Round each result
    res_first = mm256_round_epi32(&res_first, &reg_round, CONV8_ROUNDING_BITS);
    res_last = mm256_round_epi32(&res_last, &reg_round, CONV8_ROUNDING_BITS);

    // Finally combine to get the final dst
    res_reg = _mm256_packus_epi32(res_first, res_last);
    res_reg = _mm256_min_epi16(res_reg, reg_max);
    mm256_store2_si128((__m128i *)dst_ptr, (__m128i *)(dst_ptr + dst_stride),
                       &res_reg);

    src_ptr += unrolled_src_stride;
    dst_ptr += unrolled_dst_stride;
  }

  // Repeat for the last row if needed
  if (h > 0) {
    src_reg = mm256_loadu2_si128(src_ptr, src_ptr + 4);
    src_reg_shift_0 = _mm256_shuffle_epi8(src_reg, idx_shift_0);
    src_reg_shift_2 = _mm256_shuffle_epi8(src_reg, idx_shift_2);

    res_reg = mm256_madd_add_epi32(&src_reg_shift_0, &src_reg_shift_2,
                                   &kernel_reg_23, &kernel_reg_45);

    res_reg = mm256_round_epi32(&res_reg, &reg_round, CONV8_ROUNDING_BITS);

    res_reg = _mm256_packus_epi32(res_reg, res_reg);
    res_reg = _mm256_min_epi16(res_reg, reg_max);

    mm256_storeu2_epi64((__m128i *)dst_ptr, (__m128i *)(dst_ptr + 4), &res_reg);
  }
}

static void vpx_highbd_filter_block1d16_h4_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr,
    ptrdiff_t dst_stride, uint32_t height, const int16_t *kernel, int bd) {
  vpx_highbd_filter_block1d8_h4_avx2(src_ptr, src_stride, dst_ptr, dst_stride,
                                     height, kernel, bd);
  vpx_highbd_filter_block1d8_h4_avx2(src_ptr + 8, src_stride, dst_ptr + 8,
                                     dst_stride, height, kernel, bd);
}

static void vpx_highbd_filter_block1d8_v8_avg_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[9], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff[4];
  pack_filters(filter, ff);

  pack_8x9_init(src_ptr, src_pitch, signal);

  do {
    pack_8x9_pixels(src_ptr, src_pitch, signal);

    filter_8x9_pixels(signal, ff, &res0, &res1);
    store_8x2_avg_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
    update_pixels(signal);

    src_ptr += src_pitch << 1;
    dst_ptr += dst_pitch << 1;
    height -= 2;
  } while (height > 0);
}

static void vpx_highbd_filter_block1d16_v8_avg_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[17], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff[4];
  pack_filters(filter, ff);

  pack_16x9_init(src_ptr, src_pitch, signal);

  do {
    pack_16x9_pixels(src_ptr, src_pitch, signal);
    filter_16x9_pixels(signal, ff, &res0, &res1);
    store_16x2_avg_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
    update_16x9_pixels(signal);

    src_ptr += src_pitch << 1;
    dst_ptr += dst_pitch << 1;
    height -= 2;
  } while (height > 0);
}

static void vpx_highbd_filter_block1d8_h2_avg_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[2], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff;
  pack_2t_filter(filter, &ff);

  src_ptr -= 3;
  do {
    pack_8x2_2t_pixels(src_ptr, src_pitch, signal);
    filter_16_2t_pixels(signal, &ff, &res0, &res1);
    store_8x2_avg_pixels(&res0, &res1, &max, dst_ptr, dst_pitch);
    height -= 2;
    src_ptr += src_pitch << 1;
    dst_ptr += dst_pitch << 1;
  } while (height > 1);

  if (height > 0) {
    pack_8x1_2t_pixels(src_ptr, signal);
    filter_8x1_2t_pixels(signal, &ff, &res0);
    store_8x1_avg_pixels(&res0, &max, dst_ptr);
  }
}

static void vpx_highbd_filter_block1d16_h2_avg_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[2], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);

  __m256i ff;
  pack_2t_filter(filter, &ff);

  src_ptr -= 3;
  do {
    pack_16x1_2t_pixels(src_ptr, signal);
    filter_16_2t_pixels(signal, &ff, &res0, &res1);
    store_16x1_avg_pixels(&res0, &res1, &max, dst_ptr);
    height -= 1;
    src_ptr += src_pitch;
    dst_ptr += dst_pitch;
  } while (height > 0);
}

static void vpx_highbd_filter_block1d16_v2_avg_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m256i signal[3], res0, res1;
  const __m256i max = _mm256_set1_epi16((1 << bd) - 1);
  __m256i ff;

  pack_2t_filter(filter, &ff);
  pack_16x2_init(src_ptr, signal);

  do {
    pack_16x2_2t_pixels(src_ptr, src_pitch, signal);
    filter_16x2_2t_pixels(signal, &ff, &res0, &res1);
    store_16x1_avg_pixels(&res0, &res1, &max, dst_ptr);

    src_ptr += src_pitch;
    dst_ptr += dst_pitch;
    height -= 1;
  } while (height > 0);
}

static void vpx_highbd_filter_block1d8_v2_avg_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_pitch, uint16_t *dst_ptr,
    ptrdiff_t dst_pitch, uint32_t height, const int16_t *filter, int bd) {
  __m128i signal[3], res0, res1;
  const __m128i max = _mm_set1_epi16((1 << bd) - 1);
  __m128i ff;

  pack_8x1_2t_filter(filter, &ff);
  pack_8x2_init(src_ptr, signal);

  do {
    pack_8x2_2t_pixels_ver(src_ptr, src_pitch, signal);
    filter_8_2t_pixels(signal, &ff, &res0, &res1);
    store_8x1_2t_avg_pixels_ver(&res0, &res1, &max, dst_ptr);

    src_ptr += src_pitch;
    dst_ptr += dst_pitch;
    height -= 1;
  } while (height > 0);
}

static void vpx_highbd_filter_block1d4_v4_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr,
    ptrdiff_t dst_stride, uint32_t height, const int16_t *kernel, int bd) {
  // We will load two rows of pixels and rearrange them into the form
  // ... s[1,0] s[0,0] s[0,0] s[-1,0]
  // so that we can call multiply and add with the kernel partial output. Then
  // we can call add with another row to get the output.

  // Register for source s[-1:3, :]
  __m256i src_reg_1, src_reg_2, src_reg_3;
  // Interleaved rows of the source. lo is first half, hi second
  __m256i src_reg_m10, src_reg_01, src_reg_12, src_reg_23;
  __m256i src_reg_m1001, src_reg_1223;

  // Result after multiply and add
  __m256i res_reg;

  __m128i kernel_reg_128;                            // Kernel
  __m256i kernel_reg, kernel_reg_23, kernel_reg_45;  // Segments of kernel used

  const __m256i reg_round =
      _mm256_set1_epi32(CONV8_ROUNDING_NUM);  // Used for rounding
  const __m256i reg_max = _mm256_set1_epi16((1 << bd) - 1);
  const ptrdiff_t src_stride_unrolled = src_stride << 1;
  const ptrdiff_t dst_stride_unrolled = dst_stride << 1;
  int h;

  // Load Kernel
  kernel_reg_128 = _mm_loadu_si128((const __m128i *)kernel);
  kernel_reg = _mm256_broadcastsi128_si256(kernel_reg_128);
  kernel_reg_23 = _mm256_shuffle_epi32(kernel_reg, 0x55);
  kernel_reg_45 = _mm256_shuffle_epi32(kernel_reg, 0xaa);

  // Row -1 to row 0
  src_reg_m10 = mm256_loadu2_epi64((const __m128i *)src_ptr,
                                   (const __m128i *)(src_ptr + src_stride));

  // Row 0 to row 1
  src_reg_1 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src_ptr + src_stride * 2)));
  src_reg_01 = _mm256_permute2x128_si256(src_reg_m10, src_reg_1, 0x21);

  // First three rows
  src_reg_m1001 = _mm256_unpacklo_epi16(src_reg_m10, src_reg_01);

  for (h = height; h > 1; h -= 2) {
    src_reg_2 = _mm256_castsi128_si256(
        _mm_loadl_epi64((const __m128i *)(src_ptr + src_stride * 3)));

    src_reg_12 = _mm256_inserti128_si256(src_reg_1,
                                         _mm256_castsi256_si128(src_reg_2), 1);

    src_reg_3 = _mm256_castsi128_si256(
        _mm_loadl_epi64((const __m128i *)(src_ptr + src_stride * 4)));

    src_reg_23 = _mm256_inserti128_si256(src_reg_2,
                                         _mm256_castsi256_si128(src_reg_3), 1);

    // Last three rows
    src_reg_1223 = _mm256_unpacklo_epi16(src_reg_12, src_reg_23);

    // Output
    res_reg = mm256_madd_add_epi32(&src_reg_m1001, &src_reg_1223,
                                   &kernel_reg_23, &kernel_reg_45);

    // Round the words
    res_reg = mm256_round_epi32(&res_reg, &reg_round, CONV8_ROUNDING_BITS);

    // Combine to get the result
    res_reg = _mm256_packus_epi32(res_reg, res_reg);
    res_reg = _mm256_min_epi16(res_reg, reg_max);

    // Save the result
    mm256_storeu2_epi64((__m128i *)dst_ptr, (__m128i *)(dst_ptr + dst_stride),
                        &res_reg);

    // Update the source by two rows
    src_ptr += src_stride_unrolled;
    dst_ptr += dst_stride_unrolled;

    src_reg_m1001 = src_reg_1223;
    src_reg_1 = src_reg_3;
  }
}

static void vpx_highbd_filter_block1d8_v4_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr,
    ptrdiff_t dst_stride, uint32_t height, const int16_t *kernel, int bd) {
  // We will load two rows of pixels and rearrange them into the form
  // ... s[1,0] s[0,0] s[0,0] s[-1,0]
  // so that we can call multiply and add with the kernel partial output. Then
  // we can call add with another row to get the output.

  // Register for source s[-1:3, :]
  __m256i src_reg_1, src_reg_2, src_reg_3;
  // Interleaved rows of the source. lo is first half, hi second
  __m256i src_reg_m10, src_reg_01, src_reg_12, src_reg_23;
  __m256i src_reg_m1001_lo, src_reg_m1001_hi, src_reg_1223_lo, src_reg_1223_hi;

  __m128i kernel_reg_128;                            // Kernel
  __m256i kernel_reg, kernel_reg_23, kernel_reg_45;  // Segments of kernel

  // Result after multiply and add
  __m256i res_reg, res_reg_lo, res_reg_hi;

  const __m256i reg_round =
      _mm256_set1_epi32(CONV8_ROUNDING_NUM);  // Used for rounding
  const __m256i reg_max = _mm256_set1_epi16((1 << bd) - 1);
  const ptrdiff_t src_stride_unrolled = src_stride << 1;
  const ptrdiff_t dst_stride_unrolled = dst_stride << 1;
  int h;

  // Load Kernel
  kernel_reg_128 = _mm_loadu_si128((const __m128i *)kernel);
  kernel_reg = _mm256_broadcastsi128_si256(kernel_reg_128);
  kernel_reg_23 = _mm256_shuffle_epi32(kernel_reg, 0x55);
  kernel_reg_45 = _mm256_shuffle_epi32(kernel_reg, 0xaa);

  // Row -1 to row 0
  src_reg_m10 = mm256_loadu2_si128((const __m128i *)src_ptr,
                                   (const __m128i *)(src_ptr + src_stride));

  // Row 0 to row 1
  src_reg_1 = _mm256_castsi128_si256(
      _mm_loadu_si128((const __m128i *)(src_ptr + src_stride * 2)));
  src_reg_01 = _mm256_permute2x128_si256(src_reg_m10, src_reg_1, 0x21);

  // First three rows
  src_reg_m1001_lo = _mm256_unpacklo_epi16(src_reg_m10, src_reg_01);
  src_reg_m1001_hi = _mm256_unpackhi_epi16(src_reg_m10, src_reg_01);

  for (h = height; h > 1; h -= 2) {
    src_reg_2 = _mm256_castsi128_si256(
        _mm_loadu_si128((const __m128i *)(src_ptr + src_stride * 3)));

    src_reg_12 = _mm256_inserti128_si256(src_reg_1,
                                         _mm256_castsi256_si128(src_reg_2), 1);

    src_reg_3 = _mm256_castsi128_si256(
        _mm_loadu_si128((const __m128i *)(src_ptr + src_stride * 4)));

    src_reg_23 = _mm256_inserti128_si256(src_reg_2,
                                         _mm256_castsi256_si128(src_reg_3), 1);

    // Last three rows
    src_reg_1223_lo = _mm256_unpacklo_epi16(src_reg_12, src_reg_23);
    src_reg_1223_hi = _mm256_unpackhi_epi16(src_reg_12, src_reg_23);

    // Output from first half
    res_reg_lo = mm256_madd_add_epi32(&src_reg_m1001_lo, &src_reg_1223_lo,
                                      &kernel_reg_23, &kernel_reg_45);

    // Output from second half
    res_reg_hi = mm256_madd_add_epi32(&src_reg_m1001_hi, &src_reg_1223_hi,
                                      &kernel_reg_23, &kernel_reg_45);

    // Round the words
    res_reg_lo =
        mm256_round_epi32(&res_reg_lo, &reg_round, CONV8_ROUNDING_BITS);
    res_reg_hi =
        mm256_round_epi32(&res_reg_hi, &reg_round, CONV8_ROUNDING_BITS);

    // Combine to get the result
    res_reg = _mm256_packus_epi32(res_reg_lo, res_reg_hi);
    res_reg = _mm256_min_epi16(res_reg, reg_max);

    // Save the result
    mm256_store2_si128((__m128i *)dst_ptr, (__m128i *)(dst_ptr + dst_stride),
                       &res_reg);

    // Update the source by two rows
    src_ptr += src_stride_unrolled;
    dst_ptr += dst_stride_unrolled;

    src_reg_m1001_lo = src_reg_1223_lo;
    src_reg_m1001_hi = src_reg_1223_hi;
    src_reg_1 = src_reg_3;
  }
}

static void vpx_highbd_filter_block1d16_v4_avx2(
    const uint16_t *src_ptr, ptrdiff_t src_stride, uint16_t *dst_ptr,
    ptrdiff_t dst_stride, uint32_t height, const int16_t *kernel, int bd) {
  vpx_highbd_filter_block1d8_v4_avx2(src_ptr, src_stride, dst_ptr, dst_stride,
                                     height, kernel, bd);
  vpx_highbd_filter_block1d8_v4_avx2(src_ptr + 8, src_stride, dst_ptr + 8,
                                     dst_stride, height, kernel, bd);
}

// From vpx_dsp/x86/vpx_high_subpixel_8t_sse2.asm.
highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h8_sse2;
highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v8_sse2;

// From vpx_dsp/x86/vpx_high_subpixel_bilinear_sse2.asm.
highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h2_sse2;
highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v2_sse2;

#define vpx_highbd_filter_block1d4_h8_avx2 vpx_highbd_filter_block1d4_h8_sse2
#define vpx_highbd_filter_block1d4_h2_avx2 vpx_highbd_filter_block1d4_h2_sse2
#define vpx_highbd_filter_block1d4_v8_avx2 vpx_highbd_filter_block1d4_v8_sse2
#define vpx_highbd_filter_block1d4_v2_avx2 vpx_highbd_filter_block1d4_v2_sse2

// Use the [vh]8 version because there is no [vh]4 implementation.
#define vpx_highbd_filter_block1d16_v4_avg_avx2 \
  vpx_highbd_filter_block1d16_v8_avg_avx2
#define vpx_highbd_filter_block1d16_h4_avg_avx2 \
  vpx_highbd_filter_block1d16_h8_avg_avx2
#define vpx_highbd_filter_block1d8_v4_avg_avx2 \
  vpx_highbd_filter_block1d8_v8_avg_avx2
#define vpx_highbd_filter_block1d8_h4_avg_avx2 \
  vpx_highbd_filter_block1d8_h8_avg_avx2
#define vpx_highbd_filter_block1d4_v4_avg_avx2 \
  vpx_highbd_filter_block1d4_v8_avg_avx2
#define vpx_highbd_filter_block1d4_h4_avg_avx2 \
  vpx_highbd_filter_block1d4_h8_avg_avx2

HIGH_FUN_CONV_1D(horiz, x0_q4, x_step_q4, h, src, , avx2, 0)
HIGH_FUN_CONV_1D(vert, y0_q4, y_step_q4, v,
                 src - src_stride * (num_taps / 2 - 1), , avx2, 0)
HIGH_FUN_CONV_2D(, avx2, 0)

// From vpx_dsp/x86/vpx_high_subpixel_8t_sse2.asm.
highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h8_avg_sse2;
highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v8_avg_sse2;

// From vpx_dsp/x86/vpx_high_subpixel_bilinear_sse2.asm.
highbd_filter8_1dfunction vpx_highbd_filter_block1d4_h2_avg_sse2;
highbd_filter8_1dfunction vpx_highbd_filter_block1d4_v2_avg_sse2;

#define vpx_highbd_filter_block1d4_h8_avg_avx2 \
  vpx_highbd_filter_block1d4_h8_avg_sse2
#define vpx_highbd_filter_block1d4_h2_avg_avx2 \
  vpx_highbd_filter_block1d4_h2_avg_sse2
#define vpx_highbd_filter_block1d4_v8_avg_avx2 \
  vpx_highbd_filter_block1d4_v8_avg_sse2
#define vpx_highbd_filter_block1d4_v2_avg_avx2 \
  vpx_highbd_filter_block1d4_v2_avg_sse2

HIGH_FUN_CONV_1D(avg_horiz, x0_q4, x_step_q4, h, src, avg_, avx2, 1)
HIGH_FUN_CONV_1D(avg_vert, y0_q4, y_step_q4, v,
                 src - src_stride * (num_taps / 2 - 1), avg_, avx2, 1)
HIGH_FUN_CONV_2D(avg_, avx2, 1)

#undef HIGHBD_FUNC