diff options
Diffstat (limited to 'media/libaom/src/av1/common/x86/reconinter_avx2.c')
| -rw-r--r-- | media/libaom/src/av1/common/x86/reconinter_avx2.c | 48 |
1 files changed, 24 insertions, 24 deletions
diff --git a/media/libaom/src/av1/common/x86/reconinter_avx2.c b/media/libaom/src/av1/common/x86/reconinter_avx2.c index f645e0454..a38bd8317 100644 --- a/media/libaom/src/av1/common/x86/reconinter_avx2.c +++ b/media/libaom/src/av1/common/x86/reconinter_avx2.c @@ -28,8 +28,8 @@ static INLINE __m256i calc_mask_avx2(const __m256i mask_base, const __m256i s0, } void av1_build_compound_diffwtd_mask_avx2(uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, - const uint8_t *src0, int stride0, - const uint8_t *src1, int stride1, + const uint8_t *src0, int src0_stride, + const uint8_t *src1, int src1_stride, int h, int w) { const int mb = (mask_type == DIFFWTD_38_INV) ? AOM_BLEND_A64_MAX_ALPHA : 0; const __m256i y_mask_base = _mm256_set1_epi16(38 - mb); @@ -37,18 +37,18 @@ void av1_build_compound_diffwtd_mask_avx2(uint8_t *mask, if (4 == w) { do { const __m128i s0A = xx_loadl_32(src0); - const __m128i s0B = xx_loadl_32(src0 + stride0); - const __m128i s0C = xx_loadl_32(src0 + stride0 * 2); - const __m128i s0D = xx_loadl_32(src0 + stride0 * 3); + const __m128i s0B = xx_loadl_32(src0 + src0_stride); + const __m128i s0C = xx_loadl_32(src0 + src0_stride * 2); + const __m128i s0D = xx_loadl_32(src0 + src0_stride * 3); const __m128i s0AB = _mm_unpacklo_epi32(s0A, s0B); const __m128i s0CD = _mm_unpacklo_epi32(s0C, s0D); const __m128i s0ABCD = _mm_unpacklo_epi64(s0AB, s0CD); const __m256i s0ABCD_w = _mm256_cvtepu8_epi16(s0ABCD); const __m128i s1A = xx_loadl_32(src1); - const __m128i s1B = xx_loadl_32(src1 + stride1); - const __m128i s1C = xx_loadl_32(src1 + stride1 * 2); - const __m128i s1D = xx_loadl_32(src1 + stride1 * 3); + const __m128i s1B = xx_loadl_32(src1 + src1_stride); + const __m128i s1C = xx_loadl_32(src1 + src1_stride * 2); + const __m128i s1D = xx_loadl_32(src1 + src1_stride * 3); const __m128i s1AB = _mm_unpacklo_epi32(s1A, s1B); const __m128i s1CD = _mm_unpacklo_epi32(s1C, s1D); const __m128i s1ABCD = _mm_unpacklo_epi64(s1AB, s1CD); @@ -58,40 +58,40 @@ void av1_build_compound_diffwtd_mask_avx2(uint8_t *mask, const __m128i x_m8 = _mm256_castsi256_si128(_mm256_permute4x64_epi64(m8, 0xd8)); xx_storeu_128(mask, x_m8); - src0 += (stride0 << 2); - src1 += (stride1 << 2); + src0 += (src0_stride << 2); + src1 += (src1_stride << 2); mask += 16; i += 4; } while (i < h); } else if (8 == w) { do { const __m128i s0A = xx_loadl_64(src0); - const __m128i s0B = xx_loadl_64(src0 + stride0); - const __m128i s0C = xx_loadl_64(src0 + stride0 * 2); - const __m128i s0D = xx_loadl_64(src0 + stride0 * 3); + const __m128i s0B = xx_loadl_64(src0 + src0_stride); + const __m128i s0C = xx_loadl_64(src0 + src0_stride * 2); + const __m128i s0D = xx_loadl_64(src0 + src0_stride * 3); const __m256i s0AC_w = _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(s0A, s0C)); const __m256i s0BD_w = _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(s0B, s0D)); const __m128i s1A = xx_loadl_64(src1); - const __m128i s1B = xx_loadl_64(src1 + stride1); - const __m128i s1C = xx_loadl_64(src1 + stride1 * 2); - const __m128i s1D = xx_loadl_64(src1 + stride1 * 3); + const __m128i s1B = xx_loadl_64(src1 + src1_stride); + const __m128i s1C = xx_loadl_64(src1 + src1_stride * 2); + const __m128i s1D = xx_loadl_64(src1 + src1_stride * 3); const __m256i s1AB_w = _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(s1A, s1C)); const __m256i s1CD_w = _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(s1B, s1D)); const __m256i m16AC = calc_mask_avx2(y_mask_base, s0AC_w, s1AB_w); const __m256i m16BD = calc_mask_avx2(y_mask_base, s0BD_w, s1CD_w); const __m256i m8 = _mm256_packus_epi16(m16AC, m16BD); yy_storeu_256(mask, m8); - src0 += stride0 << 2; - src1 += stride1 << 2; + src0 += src0_stride << 2; + src1 += src1_stride << 2; mask += 32; i += 4; } while (i < h); } else if (16 == w) { do { const __m128i s0A = xx_load_128(src0); - const __m128i s0B = xx_load_128(src0 + stride0); + const __m128i s0B = xx_load_128(src0 + src0_stride); const __m128i s1A = xx_load_128(src1); - const __m128i s1B = xx_load_128(src1 + stride1); + const __m128i s1B = xx_load_128(src1 + src1_stride); const __m256i s0AL = _mm256_cvtepu8_epi16(s0A); const __m256i s0BL = _mm256_cvtepu8_epi16(s0B); const __m256i s1AL = _mm256_cvtepu8_epi16(s1A); @@ -103,8 +103,8 @@ void av1_build_compound_diffwtd_mask_avx2(uint8_t *mask, const __m256i m8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(m16AL, m16BL), 0xd8); yy_storeu_256(mask, m8); - src0 += stride0 << 1; - src1 += stride1 << 1; + src0 += src0_stride << 1; + src1 += src1_stride << 1; mask += 32; i += 2; } while (i < h); @@ -127,8 +127,8 @@ void av1_build_compound_diffwtd_mask_avx2(uint8_t *mask, yy_storeu_256(mask + j, m8); j += 32; } while (j < w); - src0 += stride0; - src1 += stride1; + src0 += src0_stride; + src1 += src1_stride; mask += w; i += 1; } while (i < h); |