virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-11-23 21:54:53 +02:00
Code Issues Releases Activity

avfilter/x86/vf_noise: Port line_noise funcs to SSE2

Browse Source 
This avoids having to fix up ABI violations via emms_c and
also leads to a 73% speedup for the line noise average version
here.

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt
2025-10-13 15:35:35 +02:00
parent adfec0f52e
commit 3ba570de8b
2 changed files with 63 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
libavfilter/vf_noise.c
View File

@@ -24,7 +24,6 @@
* noise generator
*/
#include "libavutil/emms.h"
#include "libavutil/mem.h"
#include "libavutil/opt.h"
#include "libavutil/imgutils.h"
@@ -280,7 +279,6 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref)
ff_filter_execute(ctx, filter_slice, &td, NULL,
n->slice_threading_impossible ? 1 :
FFMIN(n->height[0], ff_filter_get_nb_threads(ctx)));
emms_c();
if (inpicref != out)
av_frame_free(&inpicref);

125
libavfilter/x86/vf_noise.c
View File

@@ -26,78 +26,81 @@
#if HAVE_INLINE_ASM
#if HAVE_6REGS
static void line_noise_avg_mmx(uint8_t *dst, const uint8_t *src,
int len, const int8_t * const *shift)
static void line_noise_avg_sse2(uint8_t *dst, const uint8_t *src,
int len, const int8_t * const *shift)
{
x86_reg mmx_len = len & (~7);
x86_reg xmm_len = len & (~15);
__asm__ volatile(
"mov %5, %%"FF_REG_a" \n\t"
"pxor %%mm4, %%mm4 \n\t"
".p2align 4 \n\t"
"1: \n\t"
"movq (%1, %%"FF_REG_a"), %%mm1 \n\t"
"movq (%0, %%"FF_REG_a"), %%mm0 \n\t"
"paddb (%2, %%"FF_REG_a"), %%mm1\n\t"
"paddb (%3, %%"FF_REG_a"), %%mm1\n\t"
"movq %%mm4, %%mm5 \n\t"
"pcmpgtb %%mm0, %%mm5 \n\t"
"movq %%mm0, %%mm6 \n\t"
"movq %%mm0, %%mm2 \n\t"
"punpcklbw %%mm5, %%mm0 \n\t"
"punpckhbw %%mm5, %%mm2 \n\t"
"movq %%mm4, %%mm5 \n\t"
"pcmpgtb %%mm1, %%mm5 \n\t"
"movq %%mm1, %%mm3 \n\t"
"punpcklbw %%mm5, %%mm1 \n\t"
"punpckhbw %%mm5, %%mm3 \n\t"
"pmullw %%mm0, %%mm1 \n\t"
"pmullw %%mm2, %%mm3 \n\t"
"psraw $7, %%mm1 \n\t"
"psraw $7, %%mm3 \n\t"
"packsswb %%mm3, %%mm1 \n\t"
"paddb %%mm6, %%mm1 \n\t"
"movq %%mm1, (%4, %%"FF_REG_a") \n\t"
"add $8, %%"FF_REG_a" \n\t"
"mov %5, %%"FF_REG_a" \n\t"
"pxor %%xmm4, %%xmm4 \n\t"
".p2align 4 \n\t"
"1: \n\t"
"movdqu (%1, %%"FF_REG_a"), %%xmm1 \n\t"
"movdqu (%2, %%"FF_REG_a"), %%xmm2 \n\t"
"movdqu (%3, %%"FF_REG_a"), %%xmm3 \n\t"
"movdqa (%0, %%"FF_REG_a"), %%xmm0 \n\t"
"paddb %%xmm2, %%xmm1 \n\t"
"paddb %%xmm3, %%xmm1 \n\t"
"movdqa %%xmm4, %%xmm5 \n\t"
"pcmpgtb %%xmm0, %%xmm5 \n\t"
"movdqa %%xmm0, %%xmm6 \n\t"
"movdqa %%xmm0, %%xmm2 \n\t"
"punpcklbw %%xmm5, %%xmm0 \n\t"
"punpckhbw %%xmm5, %%xmm2 \n\t"
"movdqa %%xmm4, %%xmm5 \n\t"
"pcmpgtb %%xmm1, %%xmm5 \n\t"
"movdqa %%xmm1, %%xmm3 \n\t"
"punpcklbw %%xmm5, %%xmm1 \n\t"
"punpckhbw %%xmm5, %%xmm3 \n\t"
"pmullw %%xmm0, %%xmm1 \n\t"
"pmullw %%xmm2, %%xmm3 \n\t"
"psraw $7, %%xmm1 \n\t"
"psraw $7, %%xmm3 \n\t"
"packsswb %%xmm3, %%xmm1 \n\t"
"paddb %%xmm6, %%xmm1 \n\t"
"movdqa %%xmm1, (%4, %%"FF_REG_a") \n\t"
"add $16, %%"FF_REG_a" \n\t"
" js 1b \n\t"
:: "r" (src+mmx_len), "r" (shift[0]+mmx_len), "r" (shift[1]+mmx_len), "r" (shift[2]+mmx_len),
"r" (dst+mmx_len), "g" (-mmx_len)
: "%"FF_REG_a
:: "r" (src+xmm_len), "r" (shift[0]+xmm_len), "r" (shift[1]+xmm_len), "r" (shift[2]+xmm_len),
"r" (dst+xmm_len), "g" (-xmm_len)
: XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3",
"%xmm4", "%xmm5", "%xmm6",) "%"FF_REG_a
);
if (mmx_len != len){
const int8_t *shift2[3] = { shift[0]+mmx_len, shift[1]+mmx_len, shift[2]+mmx_len };
ff_line_noise_avg_c(dst+mmx_len, src+mmx_len, len-mmx_len, shift2);
if (xmm_len != len){
const int8_t *shift2[3] = { shift[0]+xmm_len, shift[1]+xmm_len, shift[2]+xmm_len };
ff_line_noise_avg_c(dst + xmm_len, src + xmm_len, len - xmm_len, shift2);
}
}
#endif /* HAVE_6REGS */
static void line_noise_mmxext(uint8_t *dst, const uint8_t *src,
const int8_t *noise, int len, int shift)
static void line_noise_sse2(uint8_t *dst, const uint8_t *src,
const int8_t *noise, int len, int shift)
{
x86_reg mmx_len = len & (~7);
x86_reg xmm_len = len & (~15);
noise += shift;
__asm__ volatile(
"mov %3, %%"FF_REG_a" \n\t"
"pcmpeqb %%mm7, %%mm7 \n\t"
"psllw $15, %%mm7 \n\t"
"packsswb %%mm7, %%mm7 \n\t"
".p2align 4 \n\t"
"1: \n\t"
"movq (%0, %%"FF_REG_a"), %%mm0 \n\t"
"movq (%1, %%"FF_REG_a"), %%mm1 \n\t"
"pxor %%mm7, %%mm0 \n\t"
"paddsb %%mm1, %%mm0 \n\t"
"pxor %%mm7, %%mm0 \n\t"
"movntq %%mm0, (%2, %%"FF_REG_a") \n\t"
"add $8, %%"FF_REG_a" \n\t"
" js 1b \n\t"
:: "r" (src+mmx_len), "r" (noise+mmx_len), "r" (dst+mmx_len), "g" (-mmx_len)
: "%"FF_REG_a
"mov %3, %%"FF_REG_a" \n\t"
"pcmpeqb %%xmm2, %%xmm2 \n\t"
"psllw $15, %%xmm2 \n\t"
"packsswb %%xmm2, %%xmm2 \n\t"
".p2align 4 \n\t"
"1: \n\t"
"movdqa (%0, %%"FF_REG_a"), %%xmm0 \n\t"
"movdqu (%1, %%"FF_REG_a"), %%xmm1 \n\t"
"pxor %%xmm2, %%xmm0 \n\t"
"paddsb %%xmm1, %%xmm0 \n\t"
"pxor %%xmm2, %%xmm0 \n\t"
"movntdq %%xmm0, (%2, %%"FF_REG_a") \n\t"
"add $16, %%"FF_REG_a" \n\t"
" js 1b \n\t"
:: "r" (src+xmm_len), "r" (noise+xmm_len), "r" (dst+xmm_len), "g" (-xmm_len)
: XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2",) "%"FF_REG_a
);
if (mmx_len != len)
ff_line_noise_c(dst+mmx_len, src+mmx_len, noise+mmx_len, len-mmx_len, 0);
if (xmm_len != len)
ff_line_noise_c(dst+xmm_len, src + xmm_len, noise + xmm_len, len - xmm_len, 0);
}
#endif /* HAVE_INLINE_ASM */
@@ -106,13 +109,11 @@ av_cold void ff_noise_init_x86(NoiseContext *n)
#if HAVE_INLINE_ASM
int cpu_flags = av_get_cpu_flags();
if (INLINE_MMX(cpu_flags)) {
if (INLINE_SSE2(cpu_flags)) {
#if HAVE_6REGS
n->line_noise_avg = line_noise_avg_mmx;
n->line_noise_avg = line_noise_avg_sse2;
#endif
}
if (INLINE_MMXEXT(cpu_flags)) {
n->line_noise = line_noise_mmxext;
n->line_noise = line_noise_sse2;
}
#endif
}