[SCM] libav/experimental: asm implementation of mid_pred. 20% faster huffyuv decoding, 4% faster ffv1.

[siretart at users.alioth.debian.org]

The following commit has been merged in the experimental branch:
commit 93a319f13ba94c783afe5eb832f4de8980e6baa5
Author: Loren Merritt <lorenm at u.washington.edu>
Date:   Fri Sep 15 00:36:49 2006 +0000

    asm implementation of mid_pred.
    20% faster huffyuv decoding, 4% faster ffv1.
    
    Originally committed as revision 6254 to svn://svn.ffmpeg.org/ffmpeg/trunk

diff --git a/libavutil/common.h b/libavutil/common.h
index 9d2d000..c9bdd56 100644
--- a/libavutil/common.h
+++ b/libavutil/common.h
@@ -235,7 +235,22 @@ static inline int av_log2_16bit(unsigned int v)
 /* median of 3 */
 static inline int mid_pred(int a, int b, int c)
 {
-#if 0
+#if (defined(ARCH_X86) && __CPU__ >= 686 || defined(ARCH_X86_64)) && !defined(RUNTIME_CPUDETECT)
+    int i=a, j=a;
+    asm volatile(
+        "cmp    %4, %2 \n\t"
+        "cmovg  %4, %0 \n\t"
+        "cmovl  %4, %1 \n\t"
+        "cmp    %4, %3 \n\t"
+        "cmovg  %3, %0 \n\t"
+        "cmovl  %3, %1 \n\t"
+        "cmp    %3, %2 \n\t"
+        "cmovl  %1, %0 \n\t"
+        :"+&r"(i), "+&r"(j)
+        :"r"(a), "r"(b), "r"(c)
+    );
+    return i;
+#elif 0
     int t= (a-b)&((a-b)>>31);
     a-=t;
     b+=t;

-- 
Libav/FFmpeg packaging