Skip to content

F
ffmpeg.wasm-core
Commit fd77fbb3 authored by Ronald S. Bultje's avatar Ronald S. Bultje Committed by Michael Niedermayer
Browse files
Options

vp9/x86: 32bit and sse2 support for vp9 inverse transform assembly 

Signed-off-by: 's avatarMichael Niedermayer <michaelni@gmx.at>
parent 4efdb29c
Hide whitespace changes
Inline Side-by-side
Showing with 1737 additions and 719 deletions
libavcodec/x86/vp9dsp_init.c
View file @ fd77fbb3
...@@ -186,11 +186,18 @@ itxfm_func(iadst, idct, size, opt); \ ...@@ -186,11 +186,18 @@ itxfm_func(iadst, idct, size, opt); \
itxfm_func(idct, iadst, size, opt); \ itxfm_func(idct, iadst, size, opt); \
itxfm_func(iadst, iadst, size, opt) itxfm_func(iadst, iadst, size, opt)
itxfm_func(idct, idct, 4, mmxext);
itxfm_func(idct, iadst, 4, sse2);
itxfm_func(iadst, idct, 4, sse2);
itxfm_func(iadst, iadst, 4, sse2);
itxfm_funcs(4, ssse3); itxfm_funcs(4, ssse3);
itxfm_funcs(8, sse2);
itxfm_funcs(8, ssse3); itxfm_funcs(8, ssse3);
itxfm_funcs(8, avx); itxfm_funcs(8, avx);
itxfm_funcs(16, sse2);
itxfm_funcs(16, ssse3); itxfm_funcs(16, ssse3);
itxfm_funcs(16, avx); itxfm_funcs(16, avx);
itxfm_func(idct, idct, 32, sse2);
itxfm_func(idct, idct, 32, ssse3); itxfm_func(idct, idct, 32, ssse3);
itxfm_func(idct, idct, 32, avx); itxfm_func(idct, idct, 32, avx);
itxfm_func(iwht, iwht, 4, mmx); itxfm_func(iwht, iwht, 4, mmx);
...@@ -352,6 +359,7 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp) ...@@ -352,6 +359,7 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
if (EXTERNAL_MMXEXT(cpu_flags)) { if (EXTERNAL_MMXEXT(cpu_flags)) {
init_fpel(4, 1, 4, avg, mmxext); init_fpel(4, 1, 4, avg, mmxext);
init_fpel(3, 1, 8, avg, mmxext); init_fpel(3, 1, 8, avg, mmxext);
dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_mmxext;
} }
if (EXTERNAL_SSE(cpu_flags)) { if (EXTERNAL_SSE(cpu_flags)) {
...@@ -365,6 +373,21 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp) ...@@ -365,6 +373,21 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
init_fpel(1, 1, 32, avg, sse2); init_fpel(1, 1, 32, avg, sse2);
init_fpel(0, 1, 64, avg, sse2); init_fpel(0, 1, 64, avg, sse2);
init_lpf(sse2); init_lpf(sse2);
dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_sse2;
dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_sse2;
dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_sse2;
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_sse2;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_sse2;
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_sse2;
dsp->intra_pred[TX_16X16][VERT_PRED] = ff_vp9_ipred_v_16x16_sse2; dsp->intra_pred[TX_16X16][VERT_PRED] = ff_vp9_ipred_v_16x16_sse2;
dsp->intra_pred[TX_32X32][VERT_PRED] = ff_vp9_ipred_v_32x32_sse2; dsp->intra_pred[TX_32X32][VERT_PRED] = ff_vp9_ipred_v_32x32_sse2;
} }
...@@ -376,20 +399,18 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp) ...@@ -376,20 +399,18 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_ssse3; dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_ssse3; dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_ssse3; dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_ssse3;
if (ARCH_X86_64) { dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_ssse3; dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_ssse3; dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_ssse3; dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_ssse3; dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_ssse3; dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_ssse3; dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_ssse3; dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_ssse3; dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_ADST] = dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][ADST_DCT] = dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_ADST] = dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_ssse3;
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_ssse3;
}
init_lpf(ssse3); init_lpf(ssse3);
init_dc_ipred(TX_4X4, 4, ssse3); init_dc_ipred(TX_4X4, 4, ssse3);
init_dc_ipred(TX_8X8, 8, ssse3); init_dc_ipred(TX_8X8, 8, ssse3);
...@@ -398,20 +419,18 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp) ...@@ -398,20 +419,18 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
} }
if (EXTERNAL_AVX(cpu_flags)) { if (EXTERNAL_AVX(cpu_flags)) {
if (ARCH_X86_64) { dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_avx;
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_avx; dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_avx;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_avx; dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_avx;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_avx; dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_avx;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_avx; dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx; dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx; dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx; dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx; dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_ADST] = dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][ADST_DCT] = dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_ADST] = dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx;
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx;
}
init_fpel(1, 0, 32, put, avx); init_fpel(1, 0, 32, put, avx);
init_fpel(0, 0, 64, put, avx); init_fpel(0, 0, 64, put, avx);
init_lpf(avx); init_lpf(avx);
......
libavcodec/x86/vp9itxfm.asm
View file @ fd77fbb3
This source diff could not be displayed because it is too large. You can view the blob instead.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment