hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-24 10:34:05 +00:00
Code Issues Actions Projects Releases Wiki Activity

[alf] Fix incorrect conversion in alf_get_blk_stats_avx2

Browse source
This commit is contained in:
Marko Viitanen 2021-08-27 10:25:20 +03:00
parent 6714973264
commit fdf125f406
1 changed files with 1 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/strategies/avx2/alf-avx2.c
View file

@ -252,7 +252,7 @@ static void alf_get_blk_stats_avx2(encoder_state_t* const state,
}*/ }*/
__m128i e_local_1 = _mm_loadl_epi64((__m128i*) & e_local[k][0]); __m128i e_local_1 = _mm_loadl_epi64((__m128i*) & e_local[k][0]);
__m256i e_local_32 = _mm256_cvtepi16_epi64(e_local_1); __m256i e_local_32 = _mm256_cvtepi16_epi32(e_local_1);
__m256i multiplied = _mm256_mullo_epi32(y_local_32, e_local_32); __m256i multiplied = _mm256_mullo_epi32(y_local_32, e_local_32);
__m128i orig = _mm_loadu_si128((__m128i*) &alf_covariance[class_idx].y[k][0]); __m128i orig = _mm_loadu_si128((__m128i*) &alf_covariance[class_idx].y[k][0]);
_mm_storeu_si128((__m128i*)alf_covariance[class_idx].y[k], _mm_add_epi32(_mm256_castsi256_si128(multiplied),orig)); _mm_storeu_si128((__m128i*)alf_covariance[class_idx].y[k], _mm_add_epi32(_mm256_castsi256_si128(multiplied),orig));