changeset 6319:4089a1ae6558 libavcodec

remove some movq in ff_h264_idct8_add_mmx. 225->217 cycles on core2.
author lorenm
date Sun, 03 Feb 2008 03:21:47 +0000
parents 73c09e922744
children ffb2a7b80d6d
files i386/h264dsp_mmx.c
diffstat 1 files changed, 23 insertions(+), 29 deletions(-) [+]
line wrap: on
line diff
--- a/i386/h264dsp_mmx.c	Sat Feb 02 22:54:50 2008 +0000
+++ b/i386/h264dsp_mmx.c	Sun Feb 03 03:21:47 2008 +0000
@@ -94,44 +94,38 @@
 {
     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  16(%0), %%mm5  \n\t"
 
-        "movq   %%mm7, %%mm4  \n\t"
-        "movq   %%mm3, %%mm6  \n\t"
-        "movq   %%mm5, %%mm0  \n\t"
-        "movq   %%mm7, %%mm2  \n\t"
+        "movq   %%mm0, %%mm4  \n\t"
+        "movq   %%mm5, %%mm1  \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"
+        "psraw  $1,    %%mm1  \n\t"
+        "paddw  %%mm0, %%mm4  \n\t"
+        "paddw  %%mm5, %%mm1  \n\t"
+        "paddw  %%mm7, %%mm4  \n\t"
+        "paddw  %%mm0, %%mm1  \n\t"
+        "psubw  %%mm5, %%mm4  \n\t"
+        "paddw  %%mm3, %%mm1  \n\t"
 
-        "movq   %%mm5, %%mm4  \n\t"
-        "movq   %%mm1, %%mm6  \n\t"
-        "psraw  $1,    %%mm4  \n\t"
-        "psraw  $1,    %%mm6  \n\t"
-        "paddw  %%mm5, %%mm4  \n\t"
-        "paddw  %%mm1, %%mm6  \n\t"
-        "paddw  %%mm7, %%mm4  \n\t"
-        "paddw  %%mm5, %%mm6  \n\t"
-        "psubw  %%mm1, %%mm4  \n\t"
-        "paddw  %%mm3, %%mm6  \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"