# HG changeset patch # User michael # Date 1006910763 0 # Node ID 3a325e7f6e1df4c4a7b118b2f36b0c45176f7edc # Parent 59d8aea763414a68faeadd4a715c310fe756c7c7 faster brightness correcture in MMX2 diff -r 59d8aea76341 -r 3a325e7f6e1d postproc/postprocess_template.c --- a/postproc/postprocess_template.c Tue Nov 27 23:17:58 2001 +0000 +++ b/postproc/postprocess_template.c Wed Nov 28 01:26:03 2001 +0000 @@ -2464,7 +2464,30 @@ "movq packedYOffset, %%mm2 \n\t" "movq packedYScale, %%mm3 \n\t" "pxor %%mm4, %%mm4 \n\t" +#ifdef HAVE_MMX2 +#define SCALED_CPY(src1, src2, dst1, dst2) \ + "movq " #src1 ", %%mm0 \n\t"\ + "movq " #src1 ", %%mm5 \n\t"\ + "movq " #src2 ", %%mm1 \n\t"\ + "movq " #src2 ", %%mm6 \n\t"\ + "punpcklbw %%mm0, %%mm0 \n\t"\ + "punpckhbw %%mm5, %%mm5 \n\t"\ + "punpcklbw %%mm1, %%mm1 \n\t"\ + "punpckhbw %%mm6, %%mm6 \n\t"\ + "pmulhuw %%mm3, %%mm0 \n\t"\ + "pmulhuw %%mm3, %%mm5 \n\t"\ + "pmulhuw %%mm3, %%mm1 \n\t"\ + "pmulhuw %%mm3, %%mm6 \n\t"\ + "psubw %%mm2, %%mm0 \n\t"\ + "psubw %%mm2, %%mm5 \n\t"\ + "psubw %%mm2, %%mm1 \n\t"\ + "psubw %%mm2, %%mm6 \n\t"\ + "packuswb %%mm5, %%mm0 \n\t"\ + "packuswb %%mm6, %%mm1 \n\t"\ + "movq %%mm0, " #dst1 " \n\t"\ + "movq %%mm1, " #dst2 " \n\t"\ +#else //HAVE_MMX2 #define SCALED_CPY(src1, src2, dst1, dst2) \ "movq " #src1 ", %%mm0 \n\t"\ "movq " #src1 ", %%mm5 \n\t"\ @@ -2491,6 +2514,8 @@ "movq %%mm0, " #dst1 " \n\t"\ "movq %%mm1, " #dst2 " \n\t"\ +#endif //!HAVE_MMX2 + SCALED_CPY((%0) , (%0, %2) , (%1) , (%1, %3)) SCALED_CPY((%0, %2, 2), (%%eax, %2, 2), (%1, %3, 2), (%%ebx, %3, 2)) SCALED_CPY((%0, %2, 4), (%%eax, %2, 4), (%1, %3, 4), (%%ebx, %3, 4)) @@ -2678,13 +2703,19 @@ clipped-= yHistogram[white]; } + scale= (double)(maxAllowedY - minAllowedY) / (double)(white-black); + +#ifdef HAVE_MMX2 + packedYScale= (uint16_t)(scale*256.0 + 0.5); + packedYOffset= (((black*packedYScale)>>8) - minAllowedY) & 0xFFFF; +#else + packedYScale= (uint16_t)(scale*1024.0 + 0.5); packedYOffset= (black - minAllowedY) & 0xFFFF; +#endif + packedYOffset|= packedYOffset<<32; packedYOffset|= packedYOffset<<16; - scale= (double)(maxAllowedY - minAllowedY) / (double)(white-black); - - packedYScale= (uint16_t)(scale*1024.0 + 0.5); packedYScale|= packedYScale<<32; packedYScale|= packedYScale<<16; }