hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-23 18:14:06 +00:00
Code Issues Actions Projects Releases Wiki Activity

[10bit] Enable AVX2 optimized DCT functions for all bit depths

Browse source
This commit is contained in:
Marko Viitanen 2024-09-05 16:55:46 +03:00
parent 862cc92e97
commit f6ef70adfa
1 changed files with 11 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
src/strategies/avx2/dct-avx2.c
View file

@ -54,7 +54,6 @@ extern const int16_t uvg_g_dct_32_t[32][32];
#if COMPILE_INTEL_AVX2 #if COMPILE_INTEL_AVX2
#include "uvg266.h" #include "uvg266.h"
#if UVG_BIT_DEPTH == 8
#include <immintrin.h> #include <immintrin.h>
#include "strategies/avx2/dct_avx2_tables.h" #include "strategies/avx2/dct_avx2_tables.h"
#define MAX_LOG2_TR_DYNAMIC_RANGE 15 #define MAX_LOG2_TR_DYNAMIC_RANGE 15
@ -8039,23 +8038,18 @@ static void mts_idct_avx2(
} }
} }
#endif // UVG_BIT_DEPTH == 8
#endif //COMPILE_INTEL_AVX2 #endif //COMPILE_INTEL_AVX2
int uvg_strategy_register_dct_avx2(void* opaque, uint8_t bitdepth) int uvg_strategy_register_dct_avx2(void* opaque, uint8_t bitdepth)
{ {
bool success = true; bool success = true;
#if COMPILE_INTEL_AVX2 #if COMPILE_INTEL_AVX2
#if UVG_BIT_DEPTH == 8
if (bitdepth == 8){
//success &= uvg_strategyselector_register(opaque, "fast_forward_dst_4x4", "avx2", 40, &matrix_dst_4x4_avx2);
success &= uvg_strategyselector_register(opaque, "dct_4x4", "avx2", 40, &matrix_dct_4x4_avx2); success &= uvg_strategyselector_register(opaque, "dct_4x4", "avx2", 40, &matrix_dct_4x4_avx2);
success &= uvg_strategyselector_register(opaque, "dct_8x8", "avx2", 40, &matrix_dct_8x8_avx2); success &= uvg_strategyselector_register(opaque, "dct_8x8", "avx2", 40, &matrix_dct_8x8_avx2);
success &= uvg_strategyselector_register(opaque, "dct_16x16", "avx2", 40, &matrix_dct_16x16_avx2); success &= uvg_strategyselector_register(opaque, "dct_16x16", "avx2", 40, &matrix_dct_16x16_avx2);
success &= uvg_strategyselector_register(opaque, "dct_32x32", "avx2", 40, &matrix_dct_32x32_avx2); success &= uvg_strategyselector_register(opaque, "dct_32x32", "avx2", 40, &matrix_dct_32x32_avx2);
// success &= uvg_strategyselector_register(opaque, "fast_inverse_dst_4x4", "avx2", 40, &matrix_idst_4x4_avx2);
success &= uvg_strategyselector_register(opaque, "idct_4x4", "avx2", 40, &matrix_idct_4x4_avx2); success &= uvg_strategyselector_register(opaque, "idct_4x4", "avx2", 40, &matrix_idct_4x4_avx2);
success &= uvg_strategyselector_register(opaque, "idct_8x8", "avx2", 40, &matrix_idct_8x8_avx2); success &= uvg_strategyselector_register(opaque, "idct_8x8", "avx2", 40, &matrix_idct_8x8_avx2);
@ -8065,8 +8059,7 @@ int uvg_strategy_register_dct_avx2(void* opaque, uint8_t bitdepth)
success &= uvg_strategyselector_register(opaque, "mts_dct", "avx2", 40, &mts_dct_avx2); success &= uvg_strategyselector_register(opaque, "mts_dct", "avx2", 40, &mts_dct_avx2);
success &= uvg_strategyselector_register(opaque, "mts_idct", "avx2", 40, &mts_idct_avx2); success &= uvg_strategyselector_register(opaque, "mts_idct", "avx2", 40, &mts_idct_avx2);
}
#endif // UVG_BIT_DEPTH == 8
#endif //COMPILE_INTEL_AVX2 #endif //COMPILE_INTEL_AVX2
return success; return success;
} }