[FFmpeg-cvslog] r11825 - trunk/libavcodec/i386/h264dsp_mmx.c
lorenm
subversion
Sun Feb 3 04:21:48 CET 2008
Author: lorenm
Date: Sun Feb 3 04:21:47 2008
New Revision: 11825
Log:
remove some movq in ff_h264_idct8_add_mmx. 288->280 cycles on core2.
Modified:
trunk/libavcodec/i386/h264dsp_mmx.c
Modified: trunk/libavcodec/i386/h264dsp_mmx.c
==============================================================================
--- trunk/libavcodec/i386/h264dsp_mmx.c (original)
+++ trunk/libavcodec/i386/h264dsp_mmx.c Sun Feb 3 04:21:47 2008
@@ -94,44 +94,38 @@ static inline void h264_idct8_1d(int16_t
{
asm volatile(
"movq 112(%0), %%mm7 \n\t"
- "movq 80(%0), %%mm5 \n\t"
+ "movq 80(%0), %%mm0 \n\t"
"movq 48(%0), %%mm3 \n\t"
- "movq 16(%0), %%mm1 \n\t"
-
- "movq %%mm7, %%mm4 \n\t"
- "movq %%mm3, %%mm6 \n\t"
- "movq %%mm5, %%mm0 \n\t"
- "movq %%mm7, %%mm2 \n\t"
- "psraw $1, %%mm4 \n\t"
- "psraw $1, %%mm6 \n\t"
- "psubw %%mm7, %%mm0 \n\t"
- "psubw %%mm6, %%mm2 \n\t"
- "psubw %%mm4, %%mm0 \n\t"
- "psubw %%mm3, %%mm2 \n\t"
- "psubw %%mm3, %%mm0 \n\t"
- "paddw %%mm1, %%mm2 \n\t"
+ "movq 16(%0), %%mm5 \n\t"
- "movq %%mm5, %%mm4 \n\t"
- "movq %%mm1, %%mm6 \n\t"
+ "movq %%mm0, %%mm4 \n\t"
+ "movq %%mm5, %%mm1 \n\t"
"psraw $1, %%mm4 \n\t"
- "psraw $1, %%mm6 \n\t"
- "paddw %%mm5, %%mm4 \n\t"
- "paddw %%mm1, %%mm6 \n\t"
+ "psraw $1, %%mm1 \n\t"
+ "paddw %%mm0, %%mm4 \n\t"
+ "paddw %%mm5, %%mm1 \n\t"
"paddw %%mm7, %%mm4 \n\t"
- "paddw %%mm5, %%mm6 \n\t"
- "psubw %%mm1, %%mm4 \n\t"
- "paddw %%mm3, %%mm6 \n\t"
+ "paddw %%mm0, %%mm1 \n\t"
+ "psubw %%mm5, %%mm4 \n\t"
+ "paddw %%mm3, %%mm1 \n\t"
+
+ "psubw %%mm3, %%mm5 \n\t"
+ "psubw %%mm3, %%mm0 \n\t"
+ "paddw %%mm7, %%mm5 \n\t"
+ "psubw %%mm7, %%mm0 \n\t"
+ "psraw $1, %%mm3 \n\t"
+ "psraw $1, %%mm7 \n\t"
+ "psubw %%mm3, %%mm5 \n\t"
+ "psubw %%mm7, %%mm0 \n\t"
- "movq %%mm0, %%mm1 \n\t"
"movq %%mm4, %%mm3 \n\t"
- "movq %%mm2, %%mm5 \n\t"
- "movq %%mm6, %%mm7 \n\t"
- "psraw $2, %%mm6 \n\t"
+ "movq %%mm1, %%mm7 \n\t"
+ "psraw $2, %%mm1 \n\t"
"psraw $2, %%mm3 \n\t"
+ "paddw %%mm5, %%mm3 \n\t"
"psraw $2, %%mm5 \n\t"
+ "paddw %%mm0, %%mm1 \n\t"
"psraw $2, %%mm0 \n\t"
- "paddw %%mm6, %%mm1 \n\t"
- "paddw %%mm2, %%mm3 \n\t"
"psubw %%mm4, %%mm5 \n\t"
"psubw %%mm0, %%mm7 \n\t"
More information about the ffmpeg-cvslog
mailing list