ARGBMirror for AVX had wrong loop counting. This fixes it to match windows, and…

ARGBMirror for AVX had wrong loop counting.  This fixes it to match windows, and reenables the function.
BUG=269
TESTED=try bots
R=tpsiaki@google.com

Review URL: https://webrtc-codereview.appspot.com/33639004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1196 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1195
+Version: 1196
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -186,7 +186,7 @@ extern "C" {
     defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2))
 #define HAS_ARGBCOPYALPHAROW_AVX2
 #define HAS_ARGBCOPYYTOALPHAROW_AVX2
-//#define HAS_ARGBMIRRORROW_AVX2
+#define HAS_ARGBMIRRORROW_AVX2
 #define HAS_ARGBPOLYNOMIALROW_AVX2
 #define HAS_ARGBSHUFFLEROW_AVX2
 //#define HAS_ARGBTOUVROW_AVX2

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1195
+#define LIBYUV_VERSION 1196
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_posix.cc

@@ -2667,7 +2667,7 @@ void ARGBMirrorRow_AVX2(const uint8* src, uint8* dst, int width) {
     VMEMOPREG(vpermd,-0x20,0,2,4,ymm5,ymm0) // vpermd -0x20(%0,%2,4),ymm5,ymm0
     "vmovdqu    %%ymm0," MEMACCESS(1) "        \n"
     "lea        " MEMLEA(0x20,1) ",%1          \n"
-    "sub        $0x20,%2                       \n"
+    "sub        $0x8,%2                        \n"
     "jg         1b                             \n"
     "vzeroupper                                \n"
   : "+r"(src),  // %0