Commit 253209ac authored by Timothy Gu's avatar Timothy Gu

vf_blend: Add SSE2 optimization for multiply

5 times faster than C, 3 times overall.
parent a25c5dbb
......@@ -24,6 +24,7 @@
SECTION_RODATA
pw_1: times 8 dw 1
pw_128: times 8 dw 128
pw_255: times 8 dw 255
pb_127: times 16 db 127
......@@ -101,6 +102,34 @@ BLEND_INIT difference128, 4
jl .loop
BLEND_END
BLEND_INIT multiply, 4
pxor m2, m2
mova m3, [pw_1]
.nextrow:
mov xq, widthq
.loop:
; word
; |--|
movh m0, [topq + xq] ; 0000xxxx
movh m1, [bottomq + xq]
punpcklbw m0, m2 ; 00xx00xx
punpcklbw m1, m2
pmullw m0, m1 ; xxxxxxxx a * b
paddw m0, m3
mova m1, m0
psrlw m1, 8
paddw m0, m1
psrlw m0, 8 ; 00xx00xx a * b / 255
packuswb m0, m0 ; 0000xxxx
movh [dstq + xq], m0
add xq, mmsize / 2
jl .loop
BLEND_END
BLEND_INIT average, 3
pxor m2, m2
.nextrow:
......
......@@ -36,6 +36,7 @@ BLEND_FUNC(average, sse2)
BLEND_FUNC(and, sse2)
BLEND_FUNC(darken, sse2)
BLEND_FUNC(difference128, sse2)
BLEND_FUNC(multiply, sse2)
BLEND_FUNC(hardmix, sse2)
BLEND_FUNC(lighten, sse2)
BLEND_FUNC(or, sse2)
......@@ -61,6 +62,7 @@ av_cold void ff_blend_init_x86(FilterParams *param, int is_16bit)
case BLEND_DIFFERENCE128: param->blend = ff_blend_difference128_sse2; break;
case BLEND_HARDMIX: param->blend = ff_blend_hardmix_sse2; break;
case BLEND_LIGHTEN: param->blend = ff_blend_lighten_sse2; break;
case BLEND_MULTIPLY: param->blend = ff_blend_multiply_sse2; break;
case BLEND_OR: param->blend = ff_blend_or_sse2; break;
case BLEND_PHOENIX: param->blend = ff_blend_phoenix_sse2; break;
case BLEND_SUBTRACT: param->blend = ff_blend_subtract_sse2; break;
......
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