apps/ffmpeg-rockchip
1
0
Fork 0
mirror of https://github.com/nyanmisaka/ffmpeg-rockchip.git synced 2026-04-27 17:51:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

swscale: convert rgb/bgr24ToY/UV_mmx functions from inline asm to yasm.

Browse Source 
Also implement sse2/ssse3/avx versions.
This commit is contained in:
Ronald S. Bultje
2012-01-27 10:52:55 +08:00
parent 3b15a6d742
commit b5d08c27c3
3 changed files with 300 additions and 178 deletions
Download Patch File Download Diff File Expand all files Collapse all files
libswscale/x86/swscale_template.c
+1 -158
View File
@@ -1361,148 +1361,6 @@ static void RENAME(yuv2yuyv422_1)(SwsContext *c, const int16_t *buf0,
}
}
static av_always_inline void RENAME(bgr24ToY_mmx)(uint8_t *dst, const uint8_t *src,
int width, enum PixelFormat srcFormat)
{
if(srcFormat == PIX_FMT_BGR24) {
__asm__ volatile(
"movq "MANGLE(ff_bgr24toY1Coeff)", %%mm5 \n\t"
"movq "MANGLE(ff_bgr24toY2Coeff)", %%mm6 \n\t"
:
);
} else {
__asm__ volatile(
"movq "MANGLE(ff_rgb24toY1Coeff)", %%mm5 \n\t"
"movq "MANGLE(ff_rgb24toY2Coeff)", %%mm6 \n\t"
:
);
}
__asm__ volatile(
"movq "MANGLE(ff_bgr24toYOffset)", %%mm4 \n\t"
"mov %2, %%"REG_a" \n\t"
"pxor %%mm7, %%mm7 \n\t"
"1: \n\t"
PREFETCH" 64(%0) \n\t"
"movd (%0), %%mm0 \n\t"
"movd 2(%0), %%mm1 \n\t"
"movd 6(%0), %%mm2 \n\t"
"movd 8(%0), %%mm3 \n\t"
"add $12, %0 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm2 \n\t"
"punpcklbw %%mm7, %%mm3 \n\t"
"pmaddwd %%mm5, %%mm0 \n\t"
"pmaddwd %%mm6, %%mm1 \n\t"
"pmaddwd %%mm5, %%mm2 \n\t"
"pmaddwd %%mm6, %%mm3 \n\t"
"paddd %%mm1, %%mm0 \n\t"
"paddd %%mm3, %%mm2 \n\t"
"paddd %%mm4, %%mm0 \n\t"
"paddd %%mm4, %%mm2 \n\t"
"psrad $15, %%mm0 \n\t"
"psrad $15, %%mm2 \n\t"
"packssdw %%mm2, %%mm0 \n\t"
"packuswb %%mm0, %%mm0 \n\t"
"movd %%mm0, (%1, %%"REG_a") \n\t"
"add $4, %%"REG_a" \n\t"
" js 1b \n\t"
: "+r" (src)
: "r" (dst+width), "g" ((x86_reg)-width)
: "%"REG_a
);
}
static void RENAME(bgr24ToY)(uint8_t *dst, const uint8_t *src,
int width, uint32_t *unused)
{
RENAME(bgr24ToY_mmx)(dst, src, width, PIX_FMT_BGR24);
}
static void RENAME(rgb24ToY)(uint8_t *dst, const uint8_t *src,
int width, uint32_t *unused)
{
RENAME(bgr24ToY_mmx)(dst, src, width, PIX_FMT_RGB24);
}
static av_always_inline void RENAME(bgr24ToUV_mmx)(uint8_t *dstU, uint8_t *dstV,
const uint8_t *src, int width,
enum PixelFormat srcFormat)
{
__asm__ volatile(
"movq 24(%4), %%mm6 \n\t"
"mov %3, %%"REG_a" \n\t"
"pxor %%mm7, %%mm7 \n\t"
"1: \n\t"
PREFETCH" 64(%0) \n\t"
"movd (%0), %%mm0 \n\t"
"movd 2(%0), %%mm1 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"movq %%mm0, %%mm2 \n\t"
"movq %%mm1, %%mm3 \n\t"
"pmaddwd (%4), %%mm0 \n\t"
"pmaddwd 8(%4), %%mm1 \n\t"
"pmaddwd 16(%4), %%mm2 \n\t"
"pmaddwd %%mm6, %%mm3 \n\t"
"paddd %%mm1, %%mm0 \n\t"
"paddd %%mm3, %%mm2 \n\t"
"movd 6(%0), %%mm1 \n\t"
"movd 8(%0), %%mm3 \n\t"
"add $12, %0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"movq %%mm3, %%mm5 \n\t"
"pmaddwd (%4), %%mm1 \n\t"
"pmaddwd 8(%4), %%mm3 \n\t"
"pmaddwd 16(%4), %%mm4 \n\t"
"pmaddwd %%mm6, %%mm5 \n\t"
"paddd %%mm3, %%mm1 \n\t"
"paddd %%mm5, %%mm4 \n\t"
"movq "MANGLE(ff_bgr24toUVOffset)", %%mm3 \n\t"
"paddd %%mm3, %%mm0 \n\t"
"paddd %%mm3, %%mm2 \n\t"
"paddd %%mm3, %%mm1 \n\t"
"paddd %%mm3, %%mm4 \n\t"
"psrad $15, %%mm0 \n\t"
"psrad $15, %%mm2 \n\t"
"psrad $15, %%mm1 \n\t"
"psrad $15, %%mm4 \n\t"
"packssdw %%mm1, %%mm0 \n\t"
"packssdw %%mm4, %%mm2 \n\t"
"packuswb %%mm0, %%mm0 \n\t"
"packuswb %%mm2, %%mm2 \n\t"
"movd %%mm0, (%1, %%"REG_a") \n\t"
"movd %%mm2, (%2, %%"REG_a") \n\t"
"add $4, %%"REG_a" \n\t"
" js 1b \n\t"
: "+r" (src)
: "r" (dstU+width), "r" (dstV+width), "g" ((x86_reg)-width), "r"(ff_bgr24toUV[srcFormat == PIX_FMT_RGB24])
: "%"REG_a
);
}
static void RENAME(bgr24ToUV)(uint8_t *dstU, uint8_t *dstV,
const uint8_t *src1, const uint8_t *src2,
int width, uint32_t *unused)
{
RENAME(bgr24ToUV_mmx)(dstU, dstV, src1, width, PIX_FMT_BGR24);
assert(src1 == src2);
}
static void RENAME(rgb24ToUV)(uint8_t *dstU, uint8_t *dstV,
const uint8_t *src1, const uint8_t *src2,
int width, uint32_t *unused)
{
assert(src1==src2);
RENAME(bgr24ToUV_mmx)(dstU, dstV, src1, width, PIX_FMT_RGB24);
}
#if COMPILE_TEMPLATE_MMX2
static void RENAME(hyscale_fast)(SwsContext *c, int16_t *dst,
int dstWidth, const uint8_t *src,
@@ -1689,8 +1547,7 @@ static void RENAME(hcscale_fast)(SwsContext *c, int16_t *dst1, int16_t *dst2,
static av_cold void RENAME(sws_init_swScale)(SwsContext *c)
{
enum PixelFormat srcFormat = c->srcFormat,
dstFormat = c->dstFormat;
enum PixelFormat dstFormat = c->dstFormat;
if (!is16BPS(dstFormat) && !is9_OR_10BPS(dstFormat) &&
dstFormat != PIX_FMT_NV12 && dstFormat != PIX_FMT_NV21) {
@@ -1762,18 +1619,4 @@ static av_cold void RENAME(sws_init_swScale)(SwsContext *c)
}
#endif /* COMPILE_TEMPLATE_MMX2 */
}
if (!c->chrSrcHSubSample) {
switch(srcFormat) {
case PIX_FMT_BGR24 : c->chrToYV12 = RENAME(bgr24ToUV); break;
case PIX_FMT_RGB24 : c->chrToYV12 = RENAME(rgb24ToUV); break;
default: break;
}
}
switch (srcFormat) {
case PIX_FMT_BGR24 : c->lumToYV12 = RENAME(bgr24ToY); break;
case PIX_FMT_RGB24 : c->lumToYV12 = RENAME(rgb24ToY); break;
default: break;
}
}