diff options
| author | Henrik Gramner <gramner@twoorioles.com> | 2025-05-16 15:18:14 +0200 |
|---|---|---|
| committer | Henrik Gramner <henrik@gramner.com> | 2025-05-19 15:56:27 +0200 |
| commit | fd18ae88ae736b5aabff34e17394fcd103f9e5ad (patch) | |
| tree | 4263056347dad98d15b0f104fe128a67a69b9128 /libavcodec/x86/vp9dsp_init.c | |
| parent | b6803bf10419770737f8dd77497b9afb8ffebcff (diff) | |
| download | ffmpeg-fd18ae88ae736b5aabff34e17394fcd103f9e5ad.tar.gz | |
avcodec/x86/vp9: Add AVX-512ICL for 16x16 and 32x32 8bpc inverse transforms
Diffstat (limited to 'libavcodec/x86/vp9dsp_init.c')
| -rw-r--r-- | libavcodec/x86/vp9dsp_init.c | 15 |
1 files changed, 15 insertions, 0 deletions
diff --git a/libavcodec/x86/vp9dsp_init.c b/libavcodec/x86/vp9dsp_init.c index 8d11dbc348..4373fa3f04 100644 --- a/libavcodec/x86/vp9dsp_init.c +++ b/libavcodec/x86/vp9dsp_init.c @@ -114,7 +114,9 @@ itxfm_func(idct, idct, 32, ssse3); itxfm_func(idct, idct, 32, avx); itxfm_func(iwht, iwht, 4, mmx); itxfm_funcs(16, avx2); +itxfm_funcs(16, avx512icl); itxfm_func(idct, idct, 32, avx2); +itxfm_func(idct, idct, 32, avx512icl); #undef itxfm_func #undef itxfm_funcs @@ -406,6 +408,19 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact) init_ipred(32, avx2, tm, TM_VP8); } +#if ARCH_X86_64 + if (EXTERNAL_AVX512ICL(cpu_flags)) { + dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx512icl; + dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx512icl; + dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx512icl; + dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx512icl; + dsp->itxfm_add[TX_32X32][ADST_ADST] = + dsp->itxfm_add[TX_32X32][ADST_DCT] = + dsp->itxfm_add[TX_32X32][DCT_ADST] = + dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx512icl; + } +#endif + #undef init_fpel #undef init_subpel1 #undef init_subpel2 |