diff options
Diffstat (limited to 'media/libvpx/libvpx/vp9/common/mips')
7 files changed, 626 insertions, 0 deletions
diff --git a/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans16_dspr2.c b/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans16_dspr2.c new file mode 100644 index 0000000000..e68d01e9fd --- /dev/null +++ b/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans16_dspr2.c @@ -0,0 +1,98 @@ +/* + * Copyright (c) 2013 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 <assert.h> +#include <stdio.h> + +#include "./vpx_config.h" +#include "./vp9_rtcd.h" +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_idct.h" +#include "vpx_dsp/mips/inv_txfm_dspr2.h" +#include "vpx_dsp/txfm_common.h" +#include "vpx_ports/mem.h" + +#if HAVE_DSPR2 +void vp9_iht16x16_256_add_dspr2(const int16_t *input, uint8_t *dest, int pitch, + int tx_type) { + int i, j; + DECLARE_ALIGNED(32, int16_t, out[16 * 16]); + int16_t *outptr = out; + int16_t temp_out[16]; + uint32_t pos = 45; + + /* bit positon for extract from acc */ + __asm__ __volatile__("wrdsp %[pos], 1 \n\t" : : [pos] "r"(pos)); + + switch (tx_type) { + case DCT_DCT: // DCT in both horizontal and vertical + idct16_rows_dspr2(input, outptr, 16); + idct16_cols_add_blk_dspr2(out, dest, pitch); + break; + case ADST_DCT: // ADST in vertical, DCT in horizontal + idct16_rows_dspr2(input, outptr, 16); + + outptr = out; + + for (i = 0; i < 16; ++i) { + iadst16_dspr2(outptr, temp_out); + + for (j = 0; j < 16; ++j) + dest[j * pitch + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6) + + dest[j * pitch + i]); + outptr += 16; + } + break; + case DCT_ADST: // DCT in vertical, ADST in horizontal + { + int16_t temp_in[16 * 16]; + + for (i = 0; i < 16; ++i) { + /* prefetch row */ + prefetch_load((const uint8_t *)(input + 16)); + + iadst16_dspr2(input, outptr); + input += 16; + outptr += 16; + } + + for (i = 0; i < 16; ++i) + for (j = 0; j < 16; ++j) temp_in[j * 16 + i] = out[i * 16 + j]; + + idct16_cols_add_blk_dspr2(temp_in, dest, pitch); + break; + } + case ADST_ADST: // ADST in both directions + { + int16_t temp_in[16]; + + for (i = 0; i < 16; ++i) { + /* prefetch row */ + prefetch_load((const uint8_t *)(input + 16)); + + iadst16_dspr2(input, outptr); + input += 16; + outptr += 16; + } + + for (i = 0; i < 16; ++i) { + for (j = 0; j < 16; ++j) temp_in[j] = out[j * 16 + i]; + iadst16_dspr2(temp_in, temp_out); + for (j = 0; j < 16; ++j) + dest[j * pitch + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 6) + + dest[j * pitch + i]); + } + break; + } + default: printf("vp9_short_iht16x16_add_dspr2 : Invalid tx_type\n"); break; + } +} +#endif // #if HAVE_DSPR2 diff --git a/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans4_dspr2.c b/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans4_dspr2.c new file mode 100644 index 0000000000..f6b29265e6 --- /dev/null +++ b/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans4_dspr2.c @@ -0,0 +1,90 @@ +/* + * Copyright (c) 2013 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 <assert.h> +#include <stdio.h> + +#include "./vpx_config.h" +#include "./vp9_rtcd.h" +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_blockd.h" +#include "vp9/common/vp9_idct.h" +#include "vpx_dsp/mips/inv_txfm_dspr2.h" +#include "vpx_dsp/txfm_common.h" +#include "vpx_ports/mem.h" + +#if HAVE_DSPR2 +void vp9_iht4x4_16_add_dspr2(const int16_t *input, uint8_t *dest, int stride, + int tx_type) { + int i, j; + DECLARE_ALIGNED(32, int16_t, out[4 * 4]); + int16_t *outptr = out; + int16_t temp_in[4 * 4], temp_out[4]; + uint32_t pos = 45; + + /* bit positon for extract from acc */ + __asm__ __volatile__("wrdsp %[pos], 1 \n\t" + : + : [pos] "r"(pos)); + + switch (tx_type) { + case DCT_DCT: // DCT in both horizontal and vertical + vpx_idct4_rows_dspr2(input, outptr); + vpx_idct4_columns_add_blk_dspr2(&out[0], dest, stride); + break; + case ADST_DCT: // ADST in vertical, DCT in horizontal + vpx_idct4_rows_dspr2(input, outptr); + + outptr = out; + + for (i = 0; i < 4; ++i) { + iadst4_dspr2(outptr, temp_out); + + for (j = 0; j < 4; ++j) + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4) + + dest[j * stride + i]); + + outptr += 4; + } + break; + case DCT_ADST: // DCT in vertical, ADST in horizontal + for (i = 0; i < 4; ++i) { + iadst4_dspr2(input, outptr); + input += 4; + outptr += 4; + } + + for (i = 0; i < 4; ++i) { + for (j = 0; j < 4; ++j) { + temp_in[i * 4 + j] = out[j * 4 + i]; + } + } + vpx_idct4_columns_add_blk_dspr2(&temp_in[0], dest, stride); + break; + case ADST_ADST: // ADST in both directions + for (i = 0; i < 4; ++i) { + iadst4_dspr2(input, outptr); + input += 4; + outptr += 4; + } + + for (i = 0; i < 4; ++i) { + for (j = 0; j < 4; ++j) temp_in[j] = out[j * 4 + i]; + iadst4_dspr2(temp_in, temp_out); + + for (j = 0; j < 4; ++j) + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4) + + dest[j * stride + i]); + } + break; + default: printf("vp9_short_iht4x4_add_dspr2 : Invalid tx_type\n"); break; + } +} +#endif // #if HAVE_DSPR2 diff --git a/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans8_dspr2.c b/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans8_dspr2.c new file mode 100644 index 0000000000..b945e307e6 --- /dev/null +++ b/media/libvpx/libvpx/vp9/common/mips/dspr2/vp9_itrans8_dspr2.c @@ -0,0 +1,84 @@ +/* + * Copyright (c) 2013 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 <assert.h> +#include <stdio.h> + +#include "./vpx_config.h" +#include "./vp9_rtcd.h" +#include "vp9/common/vp9_common.h" +#include "vp9/common/vp9_blockd.h" +#include "vpx_dsp/mips/inv_txfm_dspr2.h" +#include "vpx_dsp/txfm_common.h" +#include "vpx_ports/mem.h" + +#if HAVE_DSPR2 +void vp9_iht8x8_64_add_dspr2(const int16_t *input, uint8_t *dest, int stride, + int tx_type) { + int i, j; + DECLARE_ALIGNED(32, int16_t, out[8 * 8]); + int16_t *outptr = out; + int16_t temp_in[8 * 8], temp_out[8]; + uint32_t pos = 45; + + /* bit positon for extract from acc */ + __asm__ __volatile__("wrdsp %[pos], 1 \n\t" : : [pos] "r"(pos)); + + switch (tx_type) { + case DCT_DCT: // DCT in both horizontal and vertical + idct8_rows_dspr2(input, outptr, 8); + idct8_columns_add_blk_dspr2(&out[0], dest, stride); + break; + case ADST_DCT: // ADST in vertical, DCT in horizontal + idct8_rows_dspr2(input, outptr, 8); + + for (i = 0; i < 8; ++i) { + iadst8_dspr2(&out[i * 8], temp_out); + + for (j = 0; j < 8; ++j) + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5) + + dest[j * stride + i]); + } + break; + case DCT_ADST: // DCT in vertical, ADST in horizontal + for (i = 0; i < 8; ++i) { + iadst8_dspr2(input, outptr); + input += 8; + outptr += 8; + } + + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) { + temp_in[i * 8 + j] = out[j * 8 + i]; + } + } + idct8_columns_add_blk_dspr2(&temp_in[0], dest, stride); + break; + case ADST_ADST: // ADST in both directions + for (i = 0; i < 8; ++i) { + iadst8_dspr2(input, outptr); + input += 8; + outptr += 8; + } + + for (i = 0; i < 8; ++i) { + for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i]; + + iadst8_dspr2(temp_in, temp_out); + + for (j = 0; j < 8; ++j) + dest[j * stride + i] = clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 5) + + dest[j * stride + i]); + } + break; + default: printf("vp9_short_iht8x8_add_dspr2 : Invalid tx_type\n"); break; + } +} +#endif // #if HAVE_DSPR2 diff --git a/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct16x16_msa.c b/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct16x16_msa.c new file mode 100644 index 0000000000..c031322806 --- /dev/null +++ b/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct16x16_msa.c @@ -0,0 +1,80 @@ +/* + * Copyright (c) 2015 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 <assert.h> + +#include "./vp9_rtcd.h" +#include "vp9/common/vp9_enums.h" +#include "vpx_dsp/mips/inv_txfm_msa.h" + +void vp9_iht16x16_256_add_msa(const int16_t *input, uint8_t *dst, + int32_t dst_stride, int32_t tx_type) { + int32_t i; + DECLARE_ALIGNED(32, int16_t, out[16 * 16]); + int16_t *out_ptr = &out[0]; + + switch (tx_type) { + case DCT_DCT: + /* transform rows */ + for (i = 0; i < 2; ++i) { + /* process 16 * 8 block */ + vpx_idct16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7))); + } + + /* transform columns */ + for (i = 0; i < 2; ++i) { + /* process 8 * 16 block */ + vpx_idct16_1d_columns_addblk_msa((out_ptr + (i << 3)), (dst + (i << 3)), + dst_stride); + } + break; + case ADST_DCT: + /* transform rows */ + for (i = 0; i < 2; ++i) { + /* process 16 * 8 block */ + vpx_idct16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7))); + } + + /* transform columns */ + for (i = 0; i < 2; ++i) { + vpx_iadst16_1d_columns_addblk_msa((out_ptr + (i << 3)), + (dst + (i << 3)), dst_stride); + } + break; + case DCT_ADST: + /* transform rows */ + for (i = 0; i < 2; ++i) { + /* process 16 * 8 block */ + vpx_iadst16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7))); + } + + /* transform columns */ + for (i = 0; i < 2; ++i) { + /* process 8 * 16 block */ + vpx_idct16_1d_columns_addblk_msa((out_ptr + (i << 3)), (dst + (i << 3)), + dst_stride); + } + break; + case ADST_ADST: + /* transform rows */ + for (i = 0; i < 2; ++i) { + /* process 16 * 8 block */ + vpx_iadst16_1d_rows_msa((input + (i << 7)), (out_ptr + (i << 7))); + } + + /* transform columns */ + for (i = 0; i < 2; ++i) { + vpx_iadst16_1d_columns_addblk_msa((out_ptr + (i << 3)), + (dst + (i << 3)), dst_stride); + } + break; + default: assert(0); break; + } +} diff --git a/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct4x4_msa.c b/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct4x4_msa.c new file mode 100644 index 0000000000..aaccd5ca7b --- /dev/null +++ b/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct4x4_msa.c @@ -0,0 +1,61 @@ +/* + * Copyright (c) 2015 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 <assert.h> + +#include "./vp9_rtcd.h" +#include "vp9/common/vp9_enums.h" +#include "vpx_dsp/mips/inv_txfm_msa.h" + +void vp9_iht4x4_16_add_msa(const int16_t *input, uint8_t *dst, + int32_t dst_stride, int32_t tx_type) { + v8i16 in0, in1, in2, in3; + + /* load vector elements of 4x4 block */ + LD4x4_SH(input, in0, in1, in2, in3); + TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3); + + switch (tx_type) { + case DCT_DCT: + /* DCT in horizontal */ + VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3); + /* DCT in vertical */ + TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3); + VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3); + break; + case ADST_DCT: + /* DCT in horizontal */ + VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3); + /* ADST in vertical */ + TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3); + VP9_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3); + break; + case DCT_ADST: + /* ADST in horizontal */ + VP9_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3); + /* DCT in vertical */ + TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3); + VP9_IDCT4x4(in0, in1, in2, in3, in0, in1, in2, in3); + break; + case ADST_ADST: + /* ADST in horizontal */ + VP9_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3); + /* ADST in vertical */ + TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3); + VP9_IADST4x4(in0, in1, in2, in3, in0, in1, in2, in3); + break; + default: assert(0); break; + } + + /* final rounding (add 2^3, divide by 2^4) and shift */ + SRARI_H4_SH(in0, in1, in2, in3, 4); + /* add block and store 4x4 */ + ADDBLK_ST4x4_UB(in0, in1, in2, in3, dst, dst_stride); +} diff --git a/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct8x8_msa.c b/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct8x8_msa.c new file mode 100644 index 0000000000..76d15ff8c0 --- /dev/null +++ b/media/libvpx/libvpx/vp9/common/mips/msa/vp9_idct8x8_msa.c @@ -0,0 +1,79 @@ +/* + * Copyright (c) 2015 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 <assert.h> + +#include "./vp9_rtcd.h" +#include "vp9/common/vp9_enums.h" +#include "vpx_dsp/mips/inv_txfm_msa.h" + +void vp9_iht8x8_64_add_msa(const int16_t *input, uint8_t *dst, + int32_t dst_stride, int32_t tx_type) { + v8i16 in0, in1, in2, in3, in4, in5, in6, in7; + + /* load vector elements of 8x8 block */ + LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7); + + TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, + in4, in5, in6, in7); + + switch (tx_type) { + case DCT_DCT: + /* DCT in horizontal */ + VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, + in4, in5, in6, in7); + /* DCT in vertical */ + TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, + in3, in4, in5, in6, in7); + VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, + in4, in5, in6, in7); + break; + case ADST_DCT: + /* DCT in horizontal */ + VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, + in4, in5, in6, in7); + /* ADST in vertical */ + TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, + in3, in4, in5, in6, in7); + VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4, + in5, in6, in7); + break; + case DCT_ADST: + /* ADST in horizontal */ + VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4, + in5, in6, in7); + /* DCT in vertical */ + TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, + in3, in4, in5, in6, in7); + VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, + in4, in5, in6, in7); + break; + case ADST_ADST: + /* ADST in horizontal */ + VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4, + in5, in6, in7); + /* ADST in vertical */ + TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, + in3, in4, in5, in6, in7); + VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3, in4, + in5, in6, in7); + break; + default: assert(0); break; + } + + /* final rounding (add 2^4, divide by 2^5) and shift */ + SRARI_H4_SH(in0, in1, in2, in3, 5); + SRARI_H4_SH(in4, in5, in6, in7, 5); + + /* add block and store 8x8 */ + VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in0, in1, in2, in3); + dst += (4 * dst_stride); + VP9_ADDBLK_ST8x4_UB(dst, dst_stride, in4, in5, in6, in7); +} diff --git a/media/libvpx/libvpx/vp9/common/mips/msa/vp9_mfqe_msa.c b/media/libvpx/libvpx/vp9/common/mips/msa/vp9_mfqe_msa.c new file mode 100644 index 0000000000..2c3840958e --- /dev/null +++ b/media/libvpx/libvpx/vp9/common/mips/msa/vp9_mfqe_msa.c @@ -0,0 +1,134 @@ +/* + * Copyright (c) 2015 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 "./vp9_rtcd.h" +#include "vp9/common/vp9_onyxc_int.h" +#include "vpx_dsp/mips/macros_msa.h" + +static void filter_by_weight8x8_msa(const uint8_t *src_ptr, int32_t src_stride, + uint8_t *dst_ptr, int32_t dst_stride, + int32_t src_weight) { + int32_t dst_weight = (1 << MFQE_PRECISION) - src_weight; + int32_t row; + uint64_t src0_d, src1_d, dst0_d, dst1_d; + v16i8 src0 = { 0 }; + v16i8 src1 = { 0 }; + v16i8 dst0 = { 0 }; + v16i8 dst1 = { 0 }; + v8i16 src_wt, dst_wt, res_h_r, res_h_l, src_r, src_l, dst_r, dst_l; + + src_wt = __msa_fill_h(src_weight); + dst_wt = __msa_fill_h(dst_weight); + + for (row = 2; row--;) { + LD2(src_ptr, src_stride, src0_d, src1_d); + src_ptr += (2 * src_stride); + LD2(dst_ptr, dst_stride, dst0_d, dst1_d); + INSERT_D2_SB(src0_d, src1_d, src0); + INSERT_D2_SB(dst0_d, dst1_d, dst0); + + LD2(src_ptr, src_stride, src0_d, src1_d); + src_ptr += (2 * src_stride); + LD2((dst_ptr + 2 * dst_stride), dst_stride, dst0_d, dst1_d); + INSERT_D2_SB(src0_d, src1_d, src1); + INSERT_D2_SB(dst0_d, dst1_d, dst1); + + UNPCK_UB_SH(src0, src_r, src_l); + UNPCK_UB_SH(dst0, dst_r, dst_l); + res_h_r = (src_r * src_wt); + res_h_r += (dst_r * dst_wt); + res_h_l = (src_l * src_wt); + res_h_l += (dst_l * dst_wt); + SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION); + dst0 = (v16i8)__msa_pckev_b((v16i8)res_h_l, (v16i8)res_h_r); + ST8x2_UB(dst0, dst_ptr, dst_stride); + dst_ptr += (2 * dst_stride); + + UNPCK_UB_SH(src1, src_r, src_l); + UNPCK_UB_SH(dst1, dst_r, dst_l); + res_h_r = (src_r * src_wt); + res_h_r += (dst_r * dst_wt); + res_h_l = (src_l * src_wt); + res_h_l += (dst_l * dst_wt); + SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION); + dst1 = (v16i8)__msa_pckev_b((v16i8)res_h_l, (v16i8)res_h_r); + ST8x2_UB(dst1, dst_ptr, dst_stride); + dst_ptr += (2 * dst_stride); + } +} + +static void filter_by_weight16x16_msa(const uint8_t *src_ptr, + int32_t src_stride, uint8_t *dst_ptr, + int32_t dst_stride, int32_t src_weight) { + int32_t dst_weight = (1 << MFQE_PRECISION) - src_weight; + int32_t row; + v16i8 src0, src1, src2, src3, dst0, dst1, dst2, dst3; + v8i16 src_wt, dst_wt, res_h_r, res_h_l, src_r, src_l, dst_r, dst_l; + + src_wt = __msa_fill_h(src_weight); + dst_wt = __msa_fill_h(dst_weight); + + for (row = 4; row--;) { + LD_SB4(src_ptr, src_stride, src0, src1, src2, src3); + src_ptr += (4 * src_stride); + LD_SB4(dst_ptr, dst_stride, dst0, dst1, dst2, dst3); + + UNPCK_UB_SH(src0, src_r, src_l); + UNPCK_UB_SH(dst0, dst_r, dst_l); + res_h_r = (src_r * src_wt); + res_h_r += (dst_r * dst_wt); + res_h_l = (src_l * src_wt); + res_h_l += (dst_l * dst_wt); + SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION); + PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr); + dst_ptr += dst_stride; + + UNPCK_UB_SH(src1, src_r, src_l); + UNPCK_UB_SH(dst1, dst_r, dst_l); + res_h_r = (src_r * src_wt); + res_h_r += (dst_r * dst_wt); + res_h_l = (src_l * src_wt); + res_h_l += (dst_l * dst_wt); + SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION); + PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr); + dst_ptr += dst_stride; + + UNPCK_UB_SH(src2, src_r, src_l); + UNPCK_UB_SH(dst2, dst_r, dst_l); + res_h_r = (src_r * src_wt); + res_h_r += (dst_r * dst_wt); + res_h_l = (src_l * src_wt); + res_h_l += (dst_l * dst_wt); + SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION); + PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr); + dst_ptr += dst_stride; + + UNPCK_UB_SH(src3, src_r, src_l); + UNPCK_UB_SH(dst3, dst_r, dst_l); + res_h_r = (src_r * src_wt); + res_h_r += (dst_r * dst_wt); + res_h_l = (src_l * src_wt); + res_h_l += (dst_l * dst_wt); + SRARI_H2_SH(res_h_r, res_h_l, MFQE_PRECISION); + PCKEV_ST_SB(res_h_r, res_h_l, dst_ptr); + dst_ptr += dst_stride; + } +} + +void vp9_filter_by_weight8x8_msa(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, int src_weight) { + filter_by_weight8x8_msa(src, src_stride, dst, dst_stride, src_weight); +} + +void vp9_filter_by_weight16x16_msa(const uint8_t *src, int src_stride, + uint8_t *dst, int dst_stride, + int src_weight) { + filter_by_weight16x16_msa(src, src_stride, dst, dst_stride, src_weight); +} |