From 2b3b075aa9078fa94196d1a7f5382566cd353074 Mon Sep 17 00:00:00 2001 From: Martin Vignali Date: Tue, 16 Jan 2018 23:30:18 +0100 Subject: [PATCH] avfilter/x86/vf_blend : for testing with sse4 and avx --- libavfilter/x86/vf_blend.asm | 226 +++++++++++++++++++++++++++++----------- libavfilter/x86/vf_blend_init.c | 93 +++++++++++++++++ 2 files changed, 257 insertions(+), 62 deletions(-) diff --git a/libavfilter/x86/vf_blend.asm b/libavfilter/x86/vf_blend.asm index 4916aaf251..f8eb245eb7 100644 --- a/libavfilter/x86/vf_blend.asm +++ b/libavfilter/x86/vf_blend.asm @@ -34,6 +34,30 @@ pb_255: times 16 db 255 SECTION .text +;%1 dst, %2 src %3 xm fill by zero (only use in SSE2) +%macro PMOVZXBW 3 +%if cpuflag(avx2) + vpmovzxbw %1, %2 +%elif cpuflag(sse4) + pmovzxbw %1, %2 +%else; SSE2 + movh %1, %2 + punpcklbw %1, %3 +%endif +%endmacro + +; %1 dst, %2 src, %3 tmp (only use in avx2) : convert 16b to 8b and store +%macro PACKUSWB_AND_STORE 3 +%if mmsize == 32 + vextracti128 xm%3, m%2, 1 + packuswb xm%2, xm%3 + movu %1, xm%2 +%else + packuswb xm%2, xm%2 + movh %1, xm%2 +%endif +%endmacro + %macro BLEND_INIT 2 %if ARCH_X86_64 cglobal blend_%1, 6, 9, %2, top, top_linesize, bottom, bottom_linesize, dst, dst_linesize, width, end, x @@ -74,39 +98,32 @@ BLEND_INIT %1, 2 BLEND_END %endmacro -INIT_XMM sse2 -BLEND_SIMPLE xor, xor -BLEND_SIMPLE or, or -BLEND_SIMPLE and, and -BLEND_SIMPLE addition, addusb -BLEND_SIMPLE subtract, subusb -BLEND_SIMPLE darken, minub -BLEND_SIMPLE lighten, maxub - +%macro GRAINEXTRACT 0 +%if cpuflag(avx2)||cpuflag(sse4) +BLEND_INIT grainextract, 3 +%else ; SSE2 BLEND_INIT grainextract, 4 - pxor m2, m2 - mova m3, [pw_128] + pxor m3, m3 +%endif + + VBROADCASTI128 m2, [pw_128] .nextrow: mov xq, widthq - .loop: - movh m0, [topq + xq] - movh m1, [bottomq + xq] - punpcklbw m0, m2 - punpcklbw m1, m2 - paddw m0, m3 - psubw m0, m1 - packuswb m0, m0 - movh [dstq + xq], m0 + PMOVZXBW m0, [topq + xq], m3 + PMOVZXBW m1, [bottomq + xq], m3 + paddw m0, m2 + psubw m0, m1 + PACKUSWB_AND_STORE [dstq + xq], 0, 1 add xq, mmsize / 2 jl .loop BLEND_END +%endmacro %macro MULTIPLY 3 ; a, b, pw_1 pmullw %1, %2 ; xxxxxxxx a * b paddw %1, %3 - mova %2, %1 - psrlw %2, 8 + psrlw %2, %1, 8 paddw %1, %2 psrlw %1, 8 ; 00xx00xx a * b / 255 %endmacro @@ -118,92 +135,109 @@ BLEND_END pxor %1, %4 ; 00xx00xx 255 - x / 255 %endmacro +%macro BLEND_MULTIPLY 0 +%if cpuflag(avx2)||cpuflag(sse4) +BLEND_INIT multiply, 3 +%else BLEND_INIT multiply, 4 - pxor m2, m2 - mova m3, [pw_1] + pxor m3, m3 +%endif + + VBROADCASTI128 m2, [pw_1] .nextrow: mov xq, widthq .loop: ; word ; |--| - movh m0, [topq + xq] ; 0000xxxx - movh m1, [bottomq + xq] - punpcklbw m0, m2 ; 00xx00xx - punpcklbw m1, m2 + PMOVZXBW m0, [topq + xq], m3 ; 00xx00xx + PMOVZXBW m1, [bottomq + xq], m3 - MULTIPLY m0, m1, m3 + MULTIPLY m0, m1, m2 - packuswb m0, m0 ; 0000xxxx - movh [dstq + xq], m0 + PACKUSWB_AND_STORE [dstq + xq], 0, 1 ; 0000xxxx add xq, mmsize / 2 jl .loop BLEND_END +%endmacro +%macro BLEND_SCREEN 0 +%if cpuflag(avx2)||cpuflag(sse4) +BLEND_INIT screen, 4 +%else BLEND_INIT screen, 5 - pxor m2, m2 - mova m3, [pw_1] - mova m4, [pw_255] + pxor m4, m4 +%endif + + VBROADCASTI128 m2, [pw_1] + VBROADCASTI128 m3, [pw_255] .nextrow: mov xq, widthq .loop: - movh m0, [topq + xq] ; 0000xxxx - movh m1, [bottomq + xq] - punpcklbw m0, m2 ; 00xx00xx - punpcklbw m1, m2 + PMOVZXBW m0, [topq + xq], m4 ; 00xx00xx + PMOVZXBW m1, [bottomq + xq], m4 ; 00xx00xx - SCREEN m0, m1, m3, m4 + SCREEN m0, m1, m2, m3 - packuswb m0, m0 ; 0000xxxx - movh [dstq + xq], m0 + PACKUSWB_AND_STORE [dstq + xq], 0, 1 ; 0000xxxx add xq, mmsize / 2 jl .loop BLEND_END +%endmacro +%macro AVERAGE 0 +%if cpuflag(avx2)||cpuflag(sse4) +BLEND_INIT average, 2 +%else BLEND_INIT average, 3 pxor m2, m2 +%endif .nextrow: mov xq, widthq .loop: - movh m0, [topq + xq] - movh m1, [bottomq + xq] - punpcklbw m0, m2 - punpcklbw m1, m2 + PMOVZXBW m0, [topq + xq], m2 + PMOVZXBW m1, [bottomq + xq], m2 paddw m0, m1 psrlw m0, 1 - packuswb m0, m0 - movh [dstq + xq], m0 + PACKUSWB_AND_STORE [dstq + xq], 0, 1 add xq, mmsize / 2 jl .loop BLEND_END +%endmacro + +%macro GRAINMERGE 0 +%if cpuflag(avx2)||cpuflag(sse4) +BLEND_INIT grainmerge, 3 +%else BLEND_INIT grainmerge, 4 - pxor m2, m2 - mova m3, [pw_128] + pxor m3, m3 +%endif + + VBROADCASTI128 m2, [pw_128] .nextrow: mov xq, widthq .loop: - movh m0, [topq + xq] - movh m1, [bottomq + xq] - punpcklbw m0, m2 - punpcklbw m1, m2 + PMOVZXBW m0, [topq + xq], m3 + PMOVZXBW m1, [bottomq + xq], m3 paddw m0, m1 - psubw m0, m3 - packuswb m0, m0 - movh [dstq + xq], m0 + psubw m0, m2 + PACKUSWB_AND_STORE [dstq + xq], 0, 1 add xq, mmsize / 2 jl .loop BLEND_END +%endmacro +%macro HARDMIX 0 BLEND_INIT hardmix, 5 - mova m2, [pb_255] - mova m3, [pb_128] - mova m4, [pb_127] + VBROADCASTI128 m2, [pb_255] + VBROADCASTI128 m3, [pb_128] + VBROADCASTI128 m4, [pb_127] .nextrow: mov xq, widthq @@ -218,7 +252,9 @@ BLEND_INIT hardmix, 5 add xq, mmsize jl .loop BLEND_END +%endmacro +%macro DIVIDE 0 BLEND_INIT divide, 4 pxor m2, m2 mova m3, [ps_255] @@ -247,9 +283,11 @@ BLEND_INIT divide, 4 jl .loop BLEND_END +%endmacro +%macro PHOENIX 0 BLEND_INIT phoenix, 4 - mova m3, [pb_255] + VBROADCASTI128 m3, [pb_255] .nextrow: mov xq, widthq @@ -266,6 +304,7 @@ BLEND_INIT phoenix, 4 add xq, mmsize jl .loop BLEND_END +%endmacro %macro BLEND_ABS 0 BLEND_INIT difference, 5 @@ -291,7 +330,7 @@ BLEND_END BLEND_INIT extremity, 8 pxor m2, m2 - mova m4, [pw_255] + VBROADCASTI128 m4, [pw_255] .nextrow: mov xq, widthq @@ -315,7 +354,7 @@ BLEND_END BLEND_INIT negation, 8 pxor m2, m2 - mova m4, [pw_255] + VBROADCASTI128 m4, [pw_255] .nextrow: mov xq, widthq @@ -341,6 +380,69 @@ BLEND_END %endmacro INIT_XMM sse2 +BLEND_SIMPLE xor, xor +BLEND_SIMPLE or, or +BLEND_SIMPLE and, and +BLEND_SIMPLE addition, addusb +BLEND_SIMPLE subtract, subusb +BLEND_SIMPLE darken, minub +BLEND_SIMPLE lighten, maxub +GRAINEXTRACT +BLEND_MULTIPLY +BLEND_SCREEN +AVERAGE +GRAINMERGE +HARDMIX +PHOENIX +DIVIDE + BLEND_ABS + INIT_XMM ssse3 BLEND_ABS + +INIT_XMM sse4 +GRAINMERGE +GRAINEXTRACT +BLEND_MULTIPLY +BLEND_SCREEN +AVERAGE + +INIT_XMM avx +BLEND_SIMPLE xor, xor +BLEND_SIMPLE or, or +BLEND_SIMPLE and, and +BLEND_SIMPLE addition, addusb +BLEND_SIMPLE subtract, subusb +BLEND_SIMPLE darken, minub +BLEND_SIMPLE lighten, maxub +GRAINEXTRACT +BLEND_MULTIPLY +BLEND_SCREEN +AVERAGE +GRAINMERGE +HARDMIX +PHOENIX + +BLEND_ABS + + +%if HAVE_AVX2_EXTERNAL +INIT_YMM avx2 +BLEND_SIMPLE xor, xor +BLEND_SIMPLE or, or +BLEND_SIMPLE and, and +BLEND_SIMPLE addition, addusb +BLEND_SIMPLE subtract, subusb +BLEND_SIMPLE darken, minub +BLEND_SIMPLE lighten, maxub +GRAINEXTRACT +BLEND_MULTIPLY +BLEND_SCREEN +AVERAGE +GRAINMERGE +HARDMIX +PHOENIX + +BLEND_ABS +%endif diff --git a/libavfilter/x86/vf_blend_init.c b/libavfilter/x86/vf_blend_init.c index a4fc9af246..de841d0b55 100644 --- a/libavfilter/x86/vf_blend_init.c +++ b/libavfilter/x86/vf_blend_init.c @@ -31,26 +31,65 @@ void ff_blend_##name##_##opt(const uint8_t *top, ptrdiff_t top_linesize, \ struct FilterParams *param, double *values, int starty); BLEND_FUNC(addition, sse2) +BLEND_FUNC(addition, avx)//TEST +BLEND_FUNC(addition, avx2) BLEND_FUNC(grainmerge, sse2) +BLEND_FUNC(grainmerge, sse4) +BLEND_FUNC(grainmerge, avx)//TEST +BLEND_FUNC(grainmerge, avx2) BLEND_FUNC(average, sse2) +BLEND_FUNC(average, sse4);//TEST +BLEND_FUNC(average, avx);//TEST +BLEND_FUNC(average, avx2) BLEND_FUNC(and, sse2) +BLEND_FUNC(and, avx)//TEST +BLEND_FUNC(and, avx2) BLEND_FUNC(darken, sse2) +BLEND_FUNC(darken, avx)//TEST +BLEND_FUNC(darken, avx2) BLEND_FUNC(grainextract, sse2) +BLEND_FUNC(grainextract, sse4);//TEST +BLEND_FUNC(grainextract, avx);//TEST +BLEND_FUNC(grainextract, avx2) BLEND_FUNC(multiply, sse2) +BLEND_FUNC(multiply, sse4);//TEST +BLEND_FUNC(multiply, avx);//TEST +BLEND_FUNC(multiply, avx2) BLEND_FUNC(screen, sse2) +BLEND_FUNC(screen, sse4);//TEST +BLEND_FUNC(screen, avx);//TEST +BLEND_FUNC(screen, avx2) BLEND_FUNC(hardmix, sse2) +BLEND_FUNC(hardmix, avx)//TEST +BLEND_FUNC(hardmix, avx2) BLEND_FUNC(divide, sse2) BLEND_FUNC(lighten, sse2) +BLEND_FUNC(lighten, avx)//TEST +BLEND_FUNC(lighten, avx2) BLEND_FUNC(or, sse2) +BLEND_FUNC(or, avx)//TEST +BLEND_FUNC(or, avx2) BLEND_FUNC(phoenix, sse2) +BLEND_FUNC(phoenix, avx)//TEST +BLEND_FUNC(phoenix, avx2) BLEND_FUNC(subtract, sse2) +BLEND_FUNC(subtract, avx)//TEST +BLEND_FUNC(subtract, avx2) BLEND_FUNC(xor, sse2) +BLEND_FUNC(xor, avx)//TEST +BLEND_FUNC(xor, avx2) BLEND_FUNC(difference, sse2) BLEND_FUNC(difference, ssse3) +BLEND_FUNC(difference, avx)//TEST +BLEND_FUNC(difference, avx2) BLEND_FUNC(extremity, sse2) BLEND_FUNC(extremity, ssse3) +BLEND_FUNC(extremity, avx)//TEST +BLEND_FUNC(extremity, avx2) BLEND_FUNC(negation, sse2) BLEND_FUNC(negation, ssse3) +BLEND_FUNC(negation, avx)//TEST +BLEND_FUNC(negation, avx2) av_cold void ff_blend_init_x86(FilterParams *param, int is_16bit) { @@ -85,4 +124,58 @@ av_cold void ff_blend_init_x86(FilterParams *param, int is_16bit) case BLEND_NEGATION: param->blend = ff_blend_negation_ssse3; break; } } + + if (EXTERNAL_SSE4(cpu_flags) && param->opacity == 1 && !is_16bit) { + switch (param->mode) { + case BLEND_GRAINMERGE: param->blend = ff_blend_grainmerge_sse4; break; + case BLEND_GRAINEXTRACT: param->blend = ff_blend_grainextract_sse4; break; + case BLEND_MULTIPLY: param->blend = ff_blend_multiply_sse4; break; + case BLEND_SCREEN: param->blend = ff_blend_screen_sse4; break; + case BLEND_AVERAGE: param->blend = ff_blend_average_sse4; break; + } + } + + if (EXTERNAL_AVX(cpu_flags) && param->opacity == 1 && !is_16bit) { + switch (param->mode) { + case BLEND_ADDITION: param->blend = ff_blend_addition_avx; break; + case BLEND_GRAINMERGE: param->blend = ff_blend_grainmerge_avx; break; + case BLEND_AND: param->blend = ff_blend_and_avx; break; + case BLEND_AVERAGE: param->blend = ff_blend_average_avx; break; + case BLEND_DARKEN: param->blend = ff_blend_darken_avx; break; + case BLEND_GRAINEXTRACT: param->blend = ff_blend_grainextract_avx; break; + case BLEND_HARDMIX: param->blend = ff_blend_hardmix_avx; break; + case BLEND_LIGHTEN: param->blend = ff_blend_lighten_avx; break; + case BLEND_MULTIPLY: param->blend = ff_blend_multiply_avx; break; + case BLEND_OR: param->blend = ff_blend_or_avx; break; + case BLEND_PHOENIX: param->blend = ff_blend_phoenix_avx; break; + case BLEND_SCREEN: param->blend = ff_blend_screen_avx; break; + case BLEND_SUBTRACT: param->blend = ff_blend_subtract_avx; break; + case BLEND_XOR: param->blend = ff_blend_xor_avx; break; + case BLEND_DIFFERENCE: param->blend = ff_blend_difference_avx; break; + case BLEND_EXTREMITY: param->blend = ff_blend_extremity_avx; break; + case BLEND_NEGATION: param->blend = ff_blend_negation_avx; break; + } + } + + if (EXTERNAL_AVX2_FAST(cpu_flags) && param->opacity == 1 && !is_16bit) { + switch (param->mode) { + case BLEND_ADDITION: param->blend = ff_blend_addition_avx2; break; + case BLEND_GRAINMERGE: param->blend = ff_blend_grainmerge_avx2; break; + case BLEND_AND: param->blend = ff_blend_and_avx2; break; + case BLEND_AVERAGE: param->blend = ff_blend_average_avx2; break; + case BLEND_DARKEN: param->blend = ff_blend_darken_avx2; break; + case BLEND_GRAINEXTRACT: param->blend = ff_blend_grainextract_avx2; break; + case BLEND_HARDMIX: param->blend = ff_blend_hardmix_avx2; break; + case BLEND_LIGHTEN: param->blend = ff_blend_lighten_avx2; break; + case BLEND_MULTIPLY: param->blend = ff_blend_multiply_avx2; break; + case BLEND_OR: param->blend = ff_blend_or_avx2; break; + case BLEND_PHOENIX: param->blend = ff_blend_phoenix_avx2; break; + case BLEND_SCREEN: param->blend = ff_blend_screen_avx2; break; + case BLEND_SUBTRACT: param->blend = ff_blend_subtract_avx2; break; + case BLEND_XOR: param->blend = ff_blend_xor_avx2; break; + case BLEND_DIFFERENCE: param->blend = ff_blend_difference_avx2; break; + case BLEND_EXTREMITY: param->blend = ff_blend_extremity_avx2; break; + case BLEND_NEGATION: param->blend = ff_blend_negation_avx2; break; + } + } } -- 2.14.3 (Apple Git-98)