eessppeelllloo/ffmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2025-12-05 14:30:00 +01:00
Code Issues Packages Projects Releases Wiki Activity

avfilter/x86/vf_spp: Port store_slice to SSE2

Browse Source 
This allows to remove an emms_c from the filter. It also gives
25% speedup here (when timing the calls to store_slice using
START/STOP_TIMER).

Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This commit is contained in:
Andreas Rheinhardt
2025-10-12 17:19:26 +02:00
parent f4a87d8ca4
commit dcb28ed860
2 changed files with 24 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
libavfilter/vf_spp.c
View File

@@ -31,7 +31,6 @@
* ported by Clément Bœsch for FFmpeg.
*/
#include "libavutil/emms.h"
#include "libavutil/imgutils.h"
#include "libavutil/mem.h"
#include "libavutil/mem_internal.h"
@@ -425,7 +424,6 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
filter(s, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw, ch, qp_table, qp_stride, 0, depth);
filter(s, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw, ch, qp_table, qp_stride, 0, depth);
}
emms_c();
}
}

49
libavfilter/x86/vf_spp.c
View File

@@ -18,16 +18,20 @@
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdint.h>
#include "config.h"
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/x86/asm.h"
#include "libavutil/x86/cpu.h"
#include "libavfilter/vf_spp.h"
#if HAVE_MMX_INLINE
static void store_slice_mmx(uint8_t *dst, const int16_t *src,
int dst_stride, int src_stride,
int width, int height, int log2_scale,
const uint8_t dither[8][8])
#if HAVE_SSE2_INLINE
static void store_slice_sse2(uint8_t *dst, const int16_t *src,
int dst_stride, int src_stride,
int width, int height, int log2_scale,
const uint8_t dither[8][8])
{
int y;
@@ -35,30 +39,25 @@ static void store_slice_mmx(uint8_t *dst, const int16_t *src,
uint8_t *dst1 = dst;
const int16_t *src1 = src;
__asm__ volatile(
"movq (%3), %%mm3 \n"
"movq (%3), %%mm4 \n"
"movd %4, %%mm2 \n"
"pxor %%mm0, %%mm0 \n"
"punpcklbw %%mm0, %%mm3 \n"
"punpckhbw %%mm0, %%mm4 \n"
"psraw %%mm2, %%mm3 \n"
"psraw %%mm2, %%mm4 \n"
"movd %5, %%mm2 \n"
"movq (%3), %%xmm1 \n"
"movd %4, %%xmm2 \n"
"pxor %%xmm0, %%xmm0 \n"
"punpcklbw %%xmm0, %%xmm1 \n"
"psraw %%xmm2, %%xmm1 \n"
"movd %5, %%xmm2 \n"
"1: \n"
"movq (%0), %%mm0 \n"
"movq 8(%0), %%mm1 \n"
"paddw %%mm3, %%mm0 \n"
"paddw %%mm4, %%mm1 \n"
"psraw %%mm2, %%mm0 \n"
"psraw %%mm2, %%mm1 \n"
"packuswb %%mm1, %%mm0 \n"
"movq %%mm0, (%1) \n"
"movdqa (%0), %%xmm0 \n"
"paddw %%xmm1, %%xmm0 \n"
"psraw %%xmm2, %%xmm0 \n"
"packuswb %%xmm0, %%xmm0 \n"
"movq %%xmm0, (%1) \n"
"add $16, %0 \n"
"add $8, %1 \n"
"cmp %2, %1 \n"
" jb 1b \n"
: "+r" (src1), "+r"(dst1)
: "r"(dst + width), "r"(dither[y]), "g"(log2_scale), "g"(MAX_LEVEL - log2_scale)
XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm2")
);
src += src_stride;
dst += dst_stride;
@@ -69,11 +68,11 @@ static void store_slice_mmx(uint8_t *dst, const int16_t *src,
av_cold void ff_spp_init_x86(SPPContext *s)
{
#if HAVE_MMX_INLINE
#if HAVE_SSE2_INLINE
int cpu_flags = av_get_cpu_flags();
if (cpu_flags & AV_CPU_FLAG_MMX) {
s->store_slice = store_slice_mmx;
if (INLINE_SSE2(cpu_flags)) {
s->store_slice = store_slice_sse2;
}
#endif
}