Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

fixed a sig4 bug an non mmx2 cpus (in case of more sig4 errors please send me a "disassemble $eip-16 $eip+16" from gdb)

Browse Source 
Originally committed as revision 2182 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
pull/126/head
Michael Niedermayer 24 years ago
parent 79cccf7038
commit 5b65f0df86
2 changed files with 40 additions and 34 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 37
      postproc/postprocess.c
  2. 37
      postproc/postprocess_template.c

37
postproc/postprocess.c
Unescape View File

@ -1909,28 +1909,31 @@ static inline void blockCopy(uint8_t dst[], int dstStride, uint8_t src[], int sr
"leal (%3,%3), %%ebx \n\t"
"movq packedYOffset, %%mm2 \n\t"
"movq packedYScale, %%mm3 \n\t"
"pxor %%mm4, %%mm4 \n\t"
#define SCALED_CPY \
"movq (%0), %%mm0 \n\t"\
"movq (%0,%2), %%mm1 \n\t"\
"psubusb %%mm2, %%mm0 \n\t"\
"psubusb %%mm2, %%mm1 \n\t"\
"pxor %%mm4, %%mm4 \n\t"\
"pxor %%mm5, %%mm5 \n\t"\
"punpcklbw %%mm0, %%mm4 \n\t"\
"punpckhbw %%mm0, %%mm5 \n\t"\
"pmulhuw %%mm3, %%mm4 \n\t"\
"pmulhuw %%mm3, %%mm5 \n\t"\
"packuswb %%mm5, %%mm4 \n\t"\
"movq %%mm4, (%1) \n\t"\
"pxor %%mm4, %%mm4 \n\t"\
"pxor %%mm5, %%mm5 \n\t"\
"punpcklbw %%mm1, %%mm4 \n\t"\
"punpckhbw %%mm1, %%mm5 \n\t"\
"pmulhuw %%mm3, %%mm4 \n\t"\
"pmulhuw %%mm3, %%mm5 \n\t"\
"packuswb %%mm5, %%mm4 \n\t"\
"movq %%mm4, (%1, %3) \n\t"\
"movq %%mm0, %%mm5 \n\t"\
"punpcklbw %%mm4, %%mm0 \n\t"\
"punpckhbw %%mm4, %%mm5 \n\t"\
"psllw $7, %%mm0 \n\t"\
"psllw $7, %%mm5 \n\t"\
"pmulhw %%mm3, %%mm0 \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"\
"psllw $7, %%mm1 \n\t"\
"psllw $7, %%mm5 \n\t"\
"pmulhw %%mm3, %%mm1 \n\t"\
"pmulhw %%mm3, %%mm5 \n\t"\
"packuswb %%mm5, %%mm1 \n\t"\
"movq %%mm1, (%1, %3) \n\t"\
"1: \n\t"
SCALED_CPY
@ -2073,7 +2076,7 @@ void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int
scale= (double)(maxAllowedY - minAllowedY) / (double)(white-black);
packedYScale= (uint16_t)(scale*256.0 + 0.5);
packedYScale= (uint16_t)(scale*512.0 + 0.5);
packedYScale|= packedYScale<<32;
packedYScale|= packedYScale<<16;
}

37
postproc/postprocess_template.c
Unescape View File

@ -1909,28 +1909,31 @@ static inline void blockCopy(uint8_t dst[], int dstStride, uint8_t src[], int sr
"leal (%3,%3), %%ebx \n\t"
"movq packedYOffset, %%mm2 \n\t"
"movq packedYScale, %%mm3 \n\t"
"pxor %%mm4, %%mm4 \n\t"
#define SCALED_CPY \
"movq (%0), %%mm0 \n\t"\
"movq (%0,%2), %%mm1 \n\t"\
"psubusb %%mm2, %%mm0 \n\t"\
"psubusb %%mm2, %%mm1 \n\t"\
"pxor %%mm4, %%mm4 \n\t"\
"pxor %%mm5, %%mm5 \n\t"\
"punpcklbw %%mm0, %%mm4 \n\t"\
"punpckhbw %%mm0, %%mm5 \n\t"\
"pmulhuw %%mm3, %%mm4 \n\t"\
"pmulhuw %%mm3, %%mm5 \n\t"\
"packuswb %%mm5, %%mm4 \n\t"\
"movq %%mm4, (%1) \n\t"\
"pxor %%mm4, %%mm4 \n\t"\
"pxor %%mm5, %%mm5 \n\t"\
"punpcklbw %%mm1, %%mm4 \n\t"\
"punpckhbw %%mm1, %%mm5 \n\t"\
"pmulhuw %%mm3, %%mm4 \n\t"\
"pmulhuw %%mm3, %%mm5 \n\t"\
"packuswb %%mm5, %%mm4 \n\t"\
"movq %%mm4, (%1, %3) \n\t"\
"movq %%mm0, %%mm5 \n\t"\
"punpcklbw %%mm4, %%mm0 \n\t"\
"punpckhbw %%mm4, %%mm5 \n\t"\
"psllw $7, %%mm0 \n\t"\
"psllw $7, %%mm5 \n\t"\
"pmulhw %%mm3, %%mm0 \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"\
"psllw $7, %%mm1 \n\t"\
"psllw $7, %%mm5 \n\t"\
"pmulhw %%mm3, %%mm1 \n\t"\
"pmulhw %%mm3, %%mm5 \n\t"\
"packuswb %%mm5, %%mm1 \n\t"\
"movq %%mm1, (%1, %3) \n\t"\
"1: \n\t"
SCALED_CPY
@ -2073,7 +2076,7 @@ void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int
scale= (double)(maxAllowedY - minAllowedY) / (double)(white-black);
packedYScale= (uint16_t)(scale*256.0 + 0.5);
packedYScale= (uint16_t)(scale*512.0 + 0.5);
packedYScale|= packedYScale<<32;
packedYScale|= packedYScale<<16;
}

Write Preview
Loading…
Cancel
Save