hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-24 02:24:07 +00:00
Code Issues Actions Projects Releases Wiki Activity

Converted some loads to load int8_t instead ints

Browse source
This commit is contained in:
Reima Hyvönen 2019-03-28 13:19:02 +02:00 committed by Pauli Oikkonen
parent 8b253209a8
commit 59dfb4570c
1 changed files with 7 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
src/strategies/avx2/sao-avx2.c
View file

@ -220,7 +220,6 @@ static void calc_sao_edge_dir_avx2(const kvz_pixel *orig_data,
// Don't sample the edge pixels because this function doesn't have access to // Don't sample the edge pixels because this function doesn't have access to
// their neighbours. // their neighbours.
__m256i zeros_epi32 = _mm256_setzero_si256(); __m256i zeros_epi32 = _mm256_setzero_si256();
__m256i ones_epi32 = _mm256_set1_epi32(1); __m256i ones_epi32 = _mm256_set1_epi32(1);
__m256i twos_epi32 = _mm256_set1_epi32(2); __m256i twos_epi32 = _mm256_set1_epi32(2);
@ -242,9 +241,9 @@ static void calc_sao_edge_dir_avx2(const kvz_pixel *orig_data,
for (x = 1; x < block_width - 8; x += 8) { for (x = 1; x < block_width - 8; x += 8) {
const kvz_pixel *c_data = &rec_data[y * block_width + x]; const kvz_pixel *c_data = &rec_data[y * block_width + x];
__m128i vector_a_epi8 = _mm_loadl_epi64((__m128i*)&c_data[a_ofs.y * block_width + a_ofs.x]); __m128i vector_a_epi8 = _mm_loadl_epi64((__m128i* __restrict)&c_data[a_ofs.y * block_width + a_ofs.x]);
__m128i vector_c_epi8 = _mm_loadl_epi64((__m128i*)c_data); __m128i vector_c_epi8 = _mm_loadl_epi64((__m128i* __restrict)c_data);
__m128i vector_b_epi8 = _mm_loadl_epi64((__m128i*)&c_data[b_ofs.y * block_width + b_ofs.x]); __m128i vector_b_epi8 = _mm_loadl_epi64((__m128i* __restrict)&c_data[b_ofs.y * block_width + b_ofs.x]);
__m256i v_cat_epi32 = sao_calc_eo_cat_avx2(&vector_a_epi8, &vector_b_epi8, &vector_c_epi8); __m256i v_cat_epi32 = sao_calc_eo_cat_avx2(&vector_a_epi8, &vector_b_epi8, &vector_c_epi8);
@ -258,7 +257,7 @@ static void calc_sao_edge_dir_avx2(const kvz_pixel *orig_data,
__m256i mask_epi32 = _mm256_cmpeq_epi32(zeros_epi32, v_cat_epi32); __m256i mask_epi32 = _mm256_cmpeq_epi32(zeros_epi32, v_cat_epi32);
int temp_cnt = _mm_popcnt_u32(_mm256_movemask_epi8(mask_epi32)) / 4; int temp_cnt = _mm_popcnt_u32(_mm256_movemask_epi8(mask_epi32)) / 4;
cat_sum_cnt[1][0] += temp_cnt; cat_sum_cnt[1][0] += temp_cnt;
temp_mem_epi32 = _mm256_sub_epi32(_mm256_load_si256((__m256i*)&orig_data[y * block_width + x]), _mm256_cvtepu8_epi32(vector_c_epi8)); temp_mem_epi32 = _mm256_sub_epi32(_mm256_cvtepu8_epi32(_mm_loadl_epi64((__m128i* __restrict)&(orig_data[y * block_width + x]))), _mm256_cvtepu8_epi32(vector_c_epi8));
temp_epi32 = _mm256_and_si256(mask_epi32, temp_mem_epi32); temp_epi32 = _mm256_and_si256(mask_epi32, temp_mem_epi32);
tmp_zero_values_epi32 = _mm256_add_epi32(tmp_zero_values_epi32, temp_epi32); tmp_zero_values_epi32 = _mm256_add_epi32(tmp_zero_values_epi32, temp_epi32);
//-------------------------------------------------------------------------- //--------------------------------------------------------------------------
@ -336,9 +335,9 @@ static void calc_sao_edge_dir_avx2(const kvz_pixel *orig_data,
__m256i v_cat_epi32 = sao_calc_eo_cat_avx2(&vector_a_epi8, &vector_b_epi8, &vector_c_epi8); __m256i v_cat_epi32 = sao_calc_eo_cat_avx2(&vector_a_epi8, &vector_b_epi8, &vector_c_epi8);
__m256i temp_mem_epi32 = _mm256_castsi128_si256(_mm_loadu_si128((__m128i*)&orig_data[y * block_width + x])); const kvz_pixel* orig_ptr = &(orig_data[y * block_width + x]);
__m128i temp_mem_upper_epi32 = _mm_loadl_epi64((__m128i*)&orig_data[y * block_width + x + 4]);
_mm256_inserti128_si256(temp_mem_epi32, temp_mem_upper_epi32, 1); __m256i temp_mem_epi32 = _mm256_cvtepu8_epi32(load_6_pixels(orig_ptr));
temp_mem_epi32 = _mm256_sub_epi32(temp_mem_epi32, _mm256_cvtepu8_epi32(vector_c_epi8)); temp_mem_epi32 = _mm256_sub_epi32(temp_mem_epi32, _mm256_cvtepu8_epi32(vector_c_epi8));