hashirama/uvg266
1
0
Fork 0
mirror of https://github.com/ultravideo/uvg266.git synced 2024-11-27 11:24:05 +00:00
Code Issues Actions Projects Releases Wiki Activity

[avx2] Re-enable disabled avx2 functions that do not work with non-square blocks

Browse source
This commit is contained in:
Joose Sainio 2023-09-26 10:38:29 +03:00
parent ff77346527
commit e32cf4fb52
3 changed files with 13 additions and 13 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
src/strategies/avx2/intra-avx2.c
View file

@ -1075,10 +1075,10 @@ int uvg_strategy_register_intra_avx2(void* opaque, uint8_t bitdepth)
#if COMPILE_INTEL_AVX2 && defined X86_64 #if COMPILE_INTEL_AVX2 && defined X86_64
#if UVG_BIT_DEPTH == 8 #if UVG_BIT_DEPTH == 8
if (bitdepth == 8) { if (bitdepth == 8) {
//success &= uvg_strategyselector_register(opaque, "angular_pred", "avx2", 40, &uvg_angular_pred_avx2); success &= uvg_strategyselector_register(opaque, "angular_pred", "avx2", 40, &uvg_angular_pred_avx2);
//success &= uvg_strategyselector_register(opaque, "intra_pred_planar", "avx2", 40, &uvg_intra_pred_planar_avx2); success &= uvg_strategyselector_register(opaque, "intra_pred_planar", "avx2", 40, &uvg_intra_pred_planar_avx2);
//success &= uvg_strategyselector_register(opaque, "intra_pred_filtered_dc", "avx2", 40, &uvg_intra_pred_filtered_dc_avx2); success &= uvg_strategyselector_register(opaque, "intra_pred_filtered_dc", "avx2", 40, &uvg_intra_pred_filtered_dc_avx2);
//success &= uvg_strategyselector_register(opaque, "pdpc_planar_dc", "avx2", 40, &uvg_pdpc_planar_dc_avx2); success &= uvg_strategyselector_register(opaque, "pdpc_planar_dc", "avx2", 40, &uvg_pdpc_planar_dc_avx2);
} }
#endif //UVG_BIT_DEPTH == 8 #endif //UVG_BIT_DEPTH == 8
#endif //COMPILE_INTEL_AVX2 && defined X86_64 #endif //COMPILE_INTEL_AVX2 && defined X86_64

16
src/strategies/avx2/picture-avx2.c
View file

@ -1808,15 +1808,15 @@ int uvg_strategy_register_picture_avx2(void* opaque, uint8_t bitdepth)
success &= uvg_strategyselector_register(opaque, "satd_32x32", "avx2", 40, &satd_32x32_8bit_avx2); success &= uvg_strategyselector_register(opaque, "satd_32x32", "avx2", 40, &satd_32x32_8bit_avx2);
success &= uvg_strategyselector_register(opaque, "satd_64x64", "avx2", 40, &satd_64x64_8bit_avx2); success &= uvg_strategyselector_register(opaque, "satd_64x64", "avx2", 40, &satd_64x64_8bit_avx2);
//success &= uvg_strategyselector_register(opaque, "satd_4x4_dual", "avx2", 40, &satd_8bit_4x4_dual_avx2); success &= uvg_strategyselector_register(opaque, "satd_4x4_dual", "avx2", 40, &satd_8bit_4x4_dual_avx2);
//success &= uvg_strategyselector_register(opaque, "satd_8x8_dual", "avx2", 40, &satd_8bit_8x8_dual_avx2); success &= uvg_strategyselector_register(opaque, "satd_8x8_dual", "avx2", 40, &satd_8bit_8x8_dual_avx2);
//success &= uvg_strategyselector_register(opaque, "satd_16x16_dual", "avx2", 40, &satd_8bit_16x16_dual_avx2); success &= uvg_strategyselector_register(opaque, "satd_16x16_dual", "avx2", 40, &satd_8bit_16x16_dual_avx2);
//success &= uvg_strategyselector_register(opaque, "satd_32x32_dual", "avx2", 40, &satd_8bit_32x32_dual_avx2); success &= uvg_strategyselector_register(opaque, "satd_32x32_dual", "avx2", 40, &satd_8bit_32x32_dual_avx2);
//success &= uvg_strategyselector_register(opaque, "satd_64x64_dual", "avx2", 40, &satd_8bit_64x64_dual_avx2); success &= uvg_strategyselector_register(opaque, "satd_64x64_dual", "avx2", 40, &satd_8bit_64x64_dual_avx2);
//success &= uvg_strategyselector_register(opaque, "satd_any_size", "avx2", 40, &satd_any_size_8bit_avx2); success &= uvg_strategyselector_register(opaque, "satd_any_size", "avx2", 40, &satd_any_size_8bit_avx2);
//success &= uvg_strategyselector_register(opaque, "satd_any_size_quad", "avx2", 40, &satd_any_size_quad_avx2); success &= uvg_strategyselector_register(opaque, "satd_any_size_quad", "avx2", 40, &satd_any_size_quad_avx2);
//success &= uvg_strategyselector_register(opaque, "pixels_calc_ssd", "avx2", 40, &pixels_calc_ssd_avx2); success &= uvg_strategyselector_register(opaque, "pixels_calc_ssd", "avx2", 40, &pixels_calc_ssd_avx2);
success &= uvg_strategyselector_register(opaque, "bipred_average", "avx2", 40, &bipred_average_avx2); success &= uvg_strategyselector_register(opaque, "bipred_average", "avx2", 40, &bipred_average_avx2);
success &= uvg_strategyselector_register(opaque, "get_optimized_sad", "avx2", 40, &get_optimized_sad_avx2); success &= uvg_strategyselector_register(opaque, "get_optimized_sad", "avx2", 40, &get_optimized_sad_avx2);
success &= uvg_strategyselector_register(opaque, "ver_sad", "avx2", 40, &ver_sad_avx2); success &= uvg_strategyselector_register(opaque, "ver_sad", "avx2", 40, &ver_sad_avx2);

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

@ -960,7 +960,7 @@ int uvg_strategy_register_quant_avx2(void* opaque, uint8_t bitdepth)
#if COMPILE_INTEL_AVX2 && defined X86_64 #if COMPILE_INTEL_AVX2 && defined X86_64
#if UVG_BIT_DEPTH == 8 #if UVG_BIT_DEPTH == 8
if (bitdepth == 8) { if (bitdepth == 8) {
//success &= uvg_strategyselector_register(opaque, "quantize_residual", "avx2", 40, &uvg_quantize_residual_avx2); success &= uvg_strategyselector_register(opaque, "quantize_residual", "avx2", 40, &uvg_quantize_residual_avx2);
success &= uvg_strategyselector_register(opaque, "dequant", "avx2", 40, &uvg_dequant_avx2); success &= uvg_strategyselector_register(opaque, "dequant", "avx2", 40, &uvg_dequant_avx2);
} }
#endif // UVG_BIT_DEPTH == 8 #endif // UVG_BIT_DEPTH == 8