From 680ed8ab0e693ff0551c68b77d34893b31c53d91 Mon Sep 17 00:00:00 2001 From: Martin Vignali Date: Sat, 9 Dec 2017 18:56:05 +0100 Subject: [PATCH 2/2] avfilter/x86/vf_blend : add AVX2 version for each func except divide --- libavfilter/x86/vf_blend.asm | 139 +++++++++++++++++++++++++++++----------- libavfilter/x86/vf_blend_init.c | 38 +++++++++++ 2 files changed, 139 insertions(+), 38 deletions(-) diff --git a/libavfilter/x86/vf_blend.asm b/libavfilter/x86/vf_blend.asm index 4916aaf251..39f9107fdf 100644 --- a/libavfilter/x86/vf_blend.asm +++ b/libavfilter/x86/vf_blend.asm @@ -34,6 +34,27 @@ pb_255: times 16 db 255 SECTION .text +; %1 dst mm num, %2 src : load 64 bits in low of each lane +%macro LOAD_64B_BY_LANE 2 +%if mmsize == 32 + movu xm%1, %2 + vpermq m%1, m%1, q3120 ; mov 2 64b in xm, to low 64b in each lane +%else ; SSE + movh xm%1, %2 +%endif +%endmacro + +; %1 dst, %2 src : load 64 bits in low of each lane +%macro STORE_64B_BY_LANE 2 +%if mmsize == 32 + vpermq m%2, m%2, q3120 + movu %1, xm%2 +%else + 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 +95,31 @@ 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 BLEND_INIT grainextract, 4 pxor m2, m2 - mova m3, [pw_128] + VBROADCASTI128 m3, [pw_128] .nextrow: mov xq, widthq .loop: - movh m0, [topq + xq] - movh m1, [bottomq + xq] + LOAD_64B_BY_LANE 0, [topq + xq] + LOAD_64B_BY_LANE 1, [bottomq + xq] punpcklbw m0, m2 punpcklbw m1, m2 paddw m0, m3 psubw m0, m1 packuswb m0, m0 - movh [dstq + xq], m0 + STORE_64B_BY_LANE [dstq + xq], 0 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 +131,102 @@ BLEND_END pxor %1, %4 ; 00xx00xx 255 - x / 255 %endmacro +%macro BLEND_MULTIPLY 0 BLEND_INIT multiply, 4 pxor m2, m2 - mova m3, [pw_1] + VBROADCASTI128 m3, [pw_1] .nextrow: mov xq, widthq .loop: ; word ; |--| - movh m0, [topq + xq] ; 0000xxxx - movh m1, [bottomq + xq] + LOAD_64B_BY_LANE 0, [topq + xq] ; 0000xxxx + LOAD_64B_BY_LANE 1, [bottomq + xq] punpcklbw m0, m2 ; 00xx00xx punpcklbw m1, m2 MULTIPLY m0, m1, m3 packuswb m0, m0 ; 0000xxxx - movh [dstq + xq], m0 + STORE_64B_BY_LANE [dstq + xq], 0 add xq, mmsize / 2 jl .loop BLEND_END +%endmacro +%macro BLEND_SCREEN 0 BLEND_INIT screen, 5 pxor m2, m2 - mova m3, [pw_1] - mova m4, [pw_255] + VBROADCASTI128 m3, [pw_1] + VBROADCASTI128 m4, [pw_255] .nextrow: mov xq, widthq .loop: - movh m0, [topq + xq] ; 0000xxxx - movh m1, [bottomq + xq] + LOAD_64B_BY_LANE 0, [topq + xq] ; 0000xxxx + LOAD_64B_BY_LANE 1, [bottomq + xq] punpcklbw m0, m2 ; 00xx00xx punpcklbw m1, m2 SCREEN m0, m1, m3, m4 packuswb m0, m0 ; 0000xxxx - movh [dstq + xq], m0 + STORE_64B_BY_LANE [dstq + xq], 0 add xq, mmsize / 2 jl .loop BLEND_END +%endmacro +%macro AVERAGE 0 BLEND_INIT average, 3 pxor m2, m2 .nextrow: mov xq, widthq .loop: - movh m0, [topq + xq] - movh m1, [bottomq + xq] + LOAD_64B_BY_LANE 0, [topq + xq] + LOAD_64B_BY_LANE 1, [bottomq + xq] punpcklbw m0, m2 punpcklbw m1, m2 paddw m0, m1 psrlw m0, 1 packuswb m0, m0 - movh [dstq + xq], m0 + STORE_64B_BY_LANE [dstq + xq], 0 add xq, mmsize / 2 jl .loop BLEND_END +%endmacro + +%macro GRAINMERGE 0 BLEND_INIT grainmerge, 4 pxor m2, m2 - mova m3, [pw_128] + VBROADCASTI128 m3, [pw_128] .nextrow: mov xq, widthq .loop: - movh m0, [topq + xq] - movh m1, [bottomq + xq] + LOAD_64B_BY_LANE 0, [topq + xq] + LOAD_64B_BY_LANE 1, [bottomq + xq] punpcklbw m0, m2 punpcklbw m1, m2 paddw m0, m1 psubw m0, m3 packuswb m0, m0 - movh [dstq + xq], m0 + STORE_64B_BY_LANE [dstq + xq], 0 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 @@ -218,10 +241,12 @@ BLEND_INIT hardmix, 5 add xq, mmsize jl .loop BLEND_END +%endmacro +INIT_XMM sse2 BLEND_INIT divide, 4 pxor m2, m2 - mova m3, [ps_255] + VBROADCASTI128 m3, [ps_255] .nextrow: mov xq, widthq @@ -248,8 +273,9 @@ BLEND_INIT divide, 4 jl .loop BLEND_END +%macro PHOENIX 0 BLEND_INIT phoenix, 4 - mova m3, [pb_255] + VBROADCASTI128 m3, [pb_255] .nextrow: mov xq, widthq @@ -266,6 +292,7 @@ BLEND_INIT phoenix, 4 add xq, mmsize jl .loop BLEND_END +%endmacro %macro BLEND_ABS 0 BLEND_INIT difference, 5 @@ -291,7 +318,7 @@ BLEND_END BLEND_INIT extremity, 8 pxor m2, m2 - mova m4, [pw_255] + VBROADCASTI128 m4, [pw_255] .nextrow: mov xq, widthq @@ -315,7 +342,7 @@ BLEND_END BLEND_INIT negation, 8 pxor m2, m2 - mova m4, [pw_255] + VBROADCASTI128 m4, [pw_255] .nextrow: mov xq, widthq @@ -341,6 +368,42 @@ 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 + BLEND_ABS + INIT_XMM ssse3 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..afd014bc62 100644 --- a/libavfilter/x86/vf_blend_init.c +++ b/libavfilter/x86/vf_blend_init.c @@ -31,26 +31,43 @@ 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, avx2) BLEND_FUNC(grainmerge, sse2) +BLEND_FUNC(grainmerge, avx2) BLEND_FUNC(average, sse2) +BLEND_FUNC(average, avx2) BLEND_FUNC(and, sse2) +BLEND_FUNC(and, avx2) BLEND_FUNC(darken, sse2) +BLEND_FUNC(darken, avx2) BLEND_FUNC(grainextract, sse2) +BLEND_FUNC(grainextract, avx2) BLEND_FUNC(multiply, sse2) +BLEND_FUNC(multiply, avx2) BLEND_FUNC(screen, sse2) +BLEND_FUNC(screen, avx2) BLEND_FUNC(hardmix, sse2) +BLEND_FUNC(hardmix, avx2) BLEND_FUNC(divide, sse2) BLEND_FUNC(lighten, sse2) +BLEND_FUNC(lighten, avx2) BLEND_FUNC(or, sse2) +BLEND_FUNC(or, avx2) BLEND_FUNC(phoenix, sse2) +BLEND_FUNC(phoenix, avx2) BLEND_FUNC(subtract, sse2) +BLEND_FUNC(subtract, avx2) BLEND_FUNC(xor, sse2) +BLEND_FUNC(xor, avx2) BLEND_FUNC(difference, sse2) BLEND_FUNC(difference, ssse3) +BLEND_FUNC(difference, avx2) BLEND_FUNC(extremity, sse2) BLEND_FUNC(extremity, ssse3) +BLEND_FUNC(extremity, avx2) BLEND_FUNC(negation, sse2) BLEND_FUNC(negation, ssse3) +BLEND_FUNC(negation, avx2) av_cold void ff_blend_init_x86(FilterParams *param, int is_16bit) { @@ -85,4 +102,25 @@ av_cold void ff_blend_init_x86(FilterParams *param, int is_16bit) case BLEND_NEGATION: param->blend = ff_blend_negation_ssse3; 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; + } + } } -- 2.11.0 (Apple Git-81)