hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-28 03:34:06 +00:00
Code Issues Actions Projects Releases Wiki Activity

Clearified some sections, added _MM_SHUFFLE macro

Browse source
This commit is contained in:
Reima Hyvönen 2018-08-09 15:23:01 +03:00
parent dd04df8667
commit 28b165c971
1 changed files with 15 additions and 29 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

44
src/strategies/avx2/picture-avx2.c
View file

@ -729,6 +729,8 @@ static void inter_recon_bipred_no_mov_avx2(
kvz_pixel* temp_lcu_u, kvz_pixel* temp_lcu_u,
kvz_pixel* temp_lcu_v) { kvz_pixel* temp_lcu_v) {
// Doesn't pass test-asan in gitlab
int y_in_lcu, x_in_lcu; int y_in_lcu, x_in_lcu;
__m256i sample0_epi8, sample1_epi8, temp_y_epi8; __m256i sample0_epi8, sample1_epi8, temp_y_epi8;
@ -811,14 +813,8 @@ static void inter_recon_bipred_no_mov_avx2(
case 8: case 8:
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 0] = _mm256_extract_epi8(temp_u_epi8, 0); lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu] = _mm256_cvtsi256_si32(temp_u_epi8);
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 0] = _mm256_extract_epi8(temp_v_epi8, 0); lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu] = _mm256_cvtsi256_si32(temp_v_epi8);
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 1] = _mm256_extract_epi8(temp_u_epi8, 1);
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 1] = _mm256_extract_epi8(temp_v_epi8, 1);
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 2] = _mm256_extract_epi8(temp_u_epi8, 2);
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 2] = _mm256_extract_epi8(temp_v_epi8, 2);
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 3] = _mm256_extract_epi8(temp_u_epi8, 3);
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 3] = _mm256_extract_epi8(temp_v_epi8, 3);
break; break;
@ -925,12 +921,7 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
case 4: case 4:
temp_epi8 = _mm256_packus_epi16(temp_y_epi16, temp_y_epi16); temp_epi8 = _mm256_packus_epi16(temp_y_epi16, temp_y_epi16);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu] = _mm256_cvtsi256_si32(temp_epi8);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu + 0] = _mm256_extract_epi8(temp_epi8, 0);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu + 1] = _mm256_extract_epi8(temp_epi8, 1);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu + 2] = _mm256_extract_epi8(temp_epi8, 2);
lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu + 3] = _mm256_extract_epi8(temp_epi8, 3);
break; break;
case 8: case 8:
@ -945,8 +936,8 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
break; break;
case 16: case 16:
_MM_SHUFFLE
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_y_epi16, temp_y_epi16), 0b11011000); temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_y_epi16, temp_y_epi16), _MM_SHUFFLE(0, 2, 1, 3));
_mm_storeu_si128((__m128i*)&(lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu]), _mm256_castsi256_si128(temp_epi8)); _mm_storeu_si128((__m128i*)&(lcu->rec.y[(y_in_lcu)* LCU_WIDTH + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
break; break;
@ -962,7 +953,7 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
} }
else { else {
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_epi16_y, temp_y_epi16), 0b11011000); temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_epi16_y, temp_y_epi16), _MM_SHUFFLE(0, 2, 1, 3));
// Store 256-bits of integer data into memory // Store 256-bits of integer data into memory
_mm256_storeu_si256((__m256i*)&(lcu->rec.y[start_point]), temp_epi8); _mm256_storeu_si256((__m256i*)&(lcu->rec.y[start_point]), temp_epi8);
@ -1018,16 +1009,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
case 8: case 8:
temp_epi8 = _mm256_packus_epi16(temp_u_epi16, temp_u_epi16); temp_epi8 = _mm256_packus_epi16(temp_u_epi16, temp_u_epi16);
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 0] = _mm256_extract_epi8(temp_epi8, 0);
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 1] = _mm256_extract_epi8(temp_epi8, 1); lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu] = _mm256_cvtsi256_si32(temp_epi8);
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 2] = _mm256_extract_epi8(temp_epi8, 2);
lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 3] = _mm256_extract_epi8(temp_epi8, 3);
temp_epi8 = _mm256_packus_epi16(temp_v_epi16, temp_v_epi16); temp_epi8 = _mm256_packus_epi16(temp_v_epi16, temp_v_epi16);
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 0] = _mm256_extract_epi8(temp_epi8, 0); lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu] = _mm256_cvtsi256_si32(temp_epi8);
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 1] = _mm256_extract_epi8(temp_epi8, 1);
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 2] = _mm256_extract_epi8(temp_epi8, 2);
lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu + 3] = _mm256_extract_epi8(temp_epi8, 3);
break; break;
@ -1047,10 +1033,10 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
case 32: case 32:
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_u_epi16, temp_u_epi16), 0b11011000); temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_u_epi16, temp_u_epi16), _MM_SHUFFLE(0, 2, 1, 3));
_mm_storeu_si128((__m128i*)&(lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8)); _mm_storeu_si128((__m128i*)&(lcu->rec.u[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_v_epi16, temp_v_epi16), 0b11011000); temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_v_epi16, temp_v_epi16), _MM_SHUFFLE(0, 2, 1, 3));
_mm_storeu_si128((__m128i*)&(lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8)); _mm_storeu_si128((__m128i*)&(lcu->rec.v[(y_in_lcu)* LCU_WIDTH_C + x_in_lcu]), _mm256_castsi256_si128(temp_epi8));
break; break;
@ -1069,11 +1055,11 @@ static void inter_recon_bipred_avx2(const int hi_prec_luma_rec0,
} }
else { else {
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_epi16_u, temp_u_epi16), 0b11011000); temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_epi16_u, temp_u_epi16), _MM_SHUFFLE(0, 2, 1, 3));
_mm256_storeu_si256((__m256i*)&(lcu->rec.u[start_point_uv]), temp_epi8); _mm256_storeu_si256((__m256i*)&(lcu->rec.u[start_point_uv]), temp_epi8);
temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_epi16_v, temp_v_epi16), 0b11011000); temp_epi8 = _mm256_permute4x64_epi64(_mm256_packus_epi16(temp_epi16_v, temp_v_epi16), _MM_SHUFFLE(0, 2, 1, 3));
_mm256_storeu_si256((__m256i*)&(lcu->rec.v[start_point_uv]), temp_epi8); _mm256_storeu_si256((__m256i*)&(lcu->rec.v[start_point_uv]), temp_epi8);