diff options
| author | michael <michael@b3059339-0415-0410-9bf9-f77b7e298cf2> | 2001-10-23 01:46:50 +0000 |
|---|---|---|
| committer | michael <michael@b3059339-0415-0410-9bf9-f77b7e298cf2> | 2001-10-23 01:46:50 +0000 |
| commit | 3c0fdf13164623978f4a7db18f7ff6767990f9e8 (patch) | |
| tree | c296f4296ef57629d698f6254b8f2616eb617630 /postproc | |
| parent | 67850eee3b2e74884412ea015661dfcb3fd8111b (diff) | |
| download | mpv-3c0fdf13164623978f4a7db18f7ff6767990f9e8.tar.bz2 mpv-3c0fdf13164623978f4a7db18f7ff6767990f9e8.tar.xz | |
brightness / contrast fix/copy optimizations +2% speedup
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@2402 b3059339-0415-0410-9bf9-f77b7e298cf2
Diffstat (limited to 'postproc')
| -rw-r--r-- | postproc/postprocess.c | 36 | ||||
| -rw-r--r-- | postproc/postprocess_template.c | 36 |
2 files changed, 30 insertions, 42 deletions
diff --git a/postproc/postprocess.c b/postproc/postprocess.c index 3ea0031d00..df2cd3cb3b 100644 --- a/postproc/postprocess.c +++ b/postproc/postprocess.c @@ -2478,8 +2478,6 @@ int getPpModeForQuality(int quality){ return modes[quality]; } -//} // extern "C" - /** * Copies a block from src to dst and fixes the blacklevel * numLines must be a multiple of 4 @@ -2493,8 +2491,6 @@ static inline void blockCopy(uint8_t dst[], int dstStride, uint8_t src[], int sr { #ifdef HAVE_MMX asm volatile( - "movl %4, %%eax \n\t" - "movl %%eax, temp0\n\t" "pushl %0 \n\t" "pushl %1 \n\t" "leal (%2,%2), %%eax \n\t" @@ -2505,47 +2501,45 @@ static inline void blockCopy(uint8_t dst[], int dstStride, uint8_t src[], int sr #define SCALED_CPY \ "movq (%0), %%mm0 \n\t"\ - "movq (%0,%2), %%mm1 \n\t"\ - "movq %%mm0, %%mm5 \n\t"\ + "movq (%0), %%mm5 \n\t"\ "punpcklbw %%mm4, %%mm0 \n\t"\ "punpckhbw %%mm4, %%mm5 \n\t"\ "psubw %%mm2, %%mm0 \n\t"\ "psubw %%mm2, %%mm5 \n\t"\ + "movq (%0,%2), %%mm1 \n\t"\ "psllw $6, %%mm0 \n\t"\ "psllw $6, %%mm5 \n\t"\ "pmulhw %%mm3, %%mm0 \n\t"\ + "movq (%0,%2), %%mm6 \n\t"\ "pmulhw %%mm3, %%mm5 \n\t"\ - "packuswb %%mm5, %%mm0 \n\t"\ - "movq %%mm0, (%1) \n\t"\ - "movq %%mm1, %%mm5 \n\t"\ "punpcklbw %%mm4, %%mm1 \n\t"\ - "punpckhbw %%mm4, %%mm5 \n\t"\ + "punpckhbw %%mm4, %%mm6 \n\t"\ "psubw %%mm2, %%mm1 \n\t"\ - "psubw %%mm2, %%mm5 \n\t"\ + "psubw %%mm2, %%mm6 \n\t"\ "psllw $6, %%mm1 \n\t"\ - "psllw $6, %%mm5 \n\t"\ + "psllw $6, %%mm6 \n\t"\ "pmulhw %%mm3, %%mm1 \n\t"\ - "pmulhw %%mm3, %%mm5 \n\t"\ - "packuswb %%mm5, %%mm1 \n\t"\ + "pmulhw %%mm3, %%mm6 \n\t"\ + "addl %%eax, %0 \n\t"\ + "packuswb %%mm5, %%mm0 \n\t"\ + "packuswb %%mm6, %%mm1 \n\t"\ + "movq %%mm0, (%1) \n\t"\ "movq %%mm1, (%1, %3) \n\t"\ - "1: \n\t" SCALED_CPY - "addl %%eax, %0 \n\t" "addl %%ebx, %1 \n\t" SCALED_CPY - "addl %%eax, %0 \n\t" "addl %%ebx, %1 \n\t" - "decl temp0 \n\t" - "jnz 1b \n\t" +SCALED_CPY + "addl %%ebx, %1 \n\t" +SCALED_CPY "popl %1 \n\t" "popl %0 \n\t" : : "r" (src), "r" (dst), "r" (srcStride), - "r" (dstStride), - "m" (numLines>>2) + "r" (dstStride) : "%eax", "%ebx" ); #else diff --git a/postproc/postprocess_template.c b/postproc/postprocess_template.c index 3ea0031d00..df2cd3cb3b 100644 --- a/postproc/postprocess_template.c +++ b/postproc/postprocess_template.c @@ -2478,8 +2478,6 @@ int getPpModeForQuality(int quality){ return modes[quality]; } -//} // extern "C" - /** * Copies a block from src to dst and fixes the blacklevel * numLines must be a multiple of 4 @@ -2493,8 +2491,6 @@ static inline void blockCopy(uint8_t dst[], int dstStride, uint8_t src[], int sr { #ifdef HAVE_MMX asm volatile( - "movl %4, %%eax \n\t" - "movl %%eax, temp0\n\t" "pushl %0 \n\t" "pushl %1 \n\t" "leal (%2,%2), %%eax \n\t" @@ -2505,47 +2501,45 @@ static inline void blockCopy(uint8_t dst[], int dstStride, uint8_t src[], int sr #define SCALED_CPY \ "movq (%0), %%mm0 \n\t"\ - "movq (%0,%2), %%mm1 \n\t"\ - "movq %%mm0, %%mm5 \n\t"\ + "movq (%0), %%mm5 \n\t"\ "punpcklbw %%mm4, %%mm0 \n\t"\ "punpckhbw %%mm4, %%mm5 \n\t"\ "psubw %%mm2, %%mm0 \n\t"\ "psubw %%mm2, %%mm5 \n\t"\ + "movq (%0,%2), %%mm1 \n\t"\ "psllw $6, %%mm0 \n\t"\ "psllw $6, %%mm5 \n\t"\ "pmulhw %%mm3, %%mm0 \n\t"\ + "movq (%0,%2), %%mm6 \n\t"\ "pmulhw %%mm3, %%mm5 \n\t"\ - "packuswb %%mm5, %%mm0 \n\t"\ - "movq %%mm0, (%1) \n\t"\ - "movq %%mm1, %%mm5 \n\t"\ "punpcklbw %%mm4, %%mm1 \n\t"\ - "punpckhbw %%mm4, %%mm5 \n\t"\ + "punpckhbw %%mm4, %%mm6 \n\t"\ "psubw %%mm2, %%mm1 \n\t"\ - "psubw %%mm2, %%mm5 \n\t"\ + "psubw %%mm2, %%mm6 \n\t"\ "psllw $6, %%mm1 \n\t"\ - "psllw $6, %%mm5 \n\t"\ + "psllw $6, %%mm6 \n\t"\ "pmulhw %%mm3, %%mm1 \n\t"\ - "pmulhw %%mm3, %%mm5 \n\t"\ - "packuswb %%mm5, %%mm1 \n\t"\ + "pmulhw %%mm3, %%mm6 \n\t"\ + "addl %%eax, %0 \n\t"\ + "packuswb %%mm5, %%mm0 \n\t"\ + "packuswb %%mm6, %%mm1 \n\t"\ + "movq %%mm0, (%1) \n\t"\ "movq %%mm1, (%1, %3) \n\t"\ - "1: \n\t" SCALED_CPY - "addl %%eax, %0 \n\t" "addl %%ebx, %1 \n\t" SCALED_CPY - "addl %%eax, %0 \n\t" "addl %%ebx, %1 \n\t" - "decl temp0 \n\t" - "jnz 1b \n\t" +SCALED_CPY + "addl %%ebx, %1 \n\t" +SCALED_CPY "popl %1 \n\t" "popl %0 \n\t" : : "r" (src), "r" (dst), "r" (srcStride), - "r" (dstStride), - "m" (numLines>>2) + "r" (dstStride) : "%eax", "%ebx" ); #else |