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

Fine-tune 16x16 AVX2 iDCT

Browse source
This commit is contained in:
Pauli Oikkonen 2019-07-09 13:43:48 +03:00
parent 2b95d9cdd6
commit 36a8f89025
1 changed files with 10 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/strategies/avx2/dct-avx2.c
View file

@ -719,18 +719,18 @@ static void partial_butterfly_inverse_16_avx2(const int16_t *src, int16_t *dst,
8, 9, 10, 11, 12, 13, 14, 15, 8, 9, 10, 11, 12, 13, 14, 15,
6, 7, 4, 5, 2, 3, 0, 1, 6, 7, 4, 5, 2, 3, 0, 1,
14, 15, 12, 13, 10, 11, 8, 9); 14, 15, 12, 13, 10, 11, 8, 9);
// TODO: this seems stoopid, fix 16x16 tp? :D
for (uint32_t i = 0; i < width; i++) {
__m256i v = _mm256_load_si256((const __m256i *)src + i);
_mm256_store_si256((__m256i *)tsrc + i, v);
}
transpose_16x16(src, (int16_t *)tsrc); transpose_16x16(src, (int16_t *)tsrc);
__m256i dct_cols[8]; const __m256i dct_cols[8] = {
for (uint32_t j = 0; j < 8; j++) { _mm256_load_si256((const __m256i *)tdct + 0),
dct_cols[j] = _mm256_load_si256((const __m256i *)tdct + j); _mm256_load_si256((const __m256i *)tdct + 1),
} _mm256_load_si256((const __m256i *)tdct + 2),
_mm256_load_si256((const __m256i *)tdct + 3),
_mm256_load_si256((const __m256i *)tdct + 4),
_mm256_load_si256((const __m256i *)tdct + 5),
_mm256_load_si256((const __m256i *)tdct + 6),
_mm256_load_si256((const __m256i *)tdct + 7),
};
// These contain: D1,0 D3,0 D5,0 D7,0 D9,0 Db,0 Dd,0 Df,0 | D1,4 D3,4 D5,4 D7,4 D9,4 Db,4 Dd,4 Df,4 // These contain: D1,0 D3,0 D5,0 D7,0 D9,0 Db,0 Dd,0 Df,0 | D1,4 D3,4 D5,4 D7,4 D9,4 Db,4 Dd,4 Df,4
// D1,1 D3,1 D5,1 D7,1 D9,1 Db,1 Dd,1 Df,1 | D1,5 D3,5 D5,5 D7,5 D9,5 Db,5 Dd,5 Df,5 // D1,1 D3,1 D5,1 D7,1 D9,1 Db,1 Dd,1 Df,1 | D1,5 D3,5 D5,5 D7,5 D9,5 Db,5 Dd,5 Df,5