Fix for clobber registers on SumSquareError_SSE2 used for PSNR

BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/1410004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@683 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 682
+Version: 683
 License: BSD
 License File: LICENSE
 

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 682
+#define LIBYUV_VERSION 683
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/compare_posix.cc

@@ -56,7 +56,7 @@ uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
   :
   : "memory", "cc"
 #if defined(__SSE2__)
-    , "xmm0", "xmm1", "xmm2", "xmm5"
+    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
 #endif
   );
   return sse;

util/psnr.cc

@@ -159,7 +159,7 @@ static uint32 SumSquareError_SSE2(const uint8* src_a,
   :
   : "memory", "cc"
 #if defined(__SSE2__)
-    , "xmm0", "xmm1", "xmm2", "xmm5"
+    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
 #endif
   );  // NOLINT
   return sse;

util/ssim.cc

@@ -20,10 +20,11 @@ extern "C" {
 typedef unsigned int uint32;     // NOLINT
 typedef unsigned short uint16;   // NOLINT
 
-#if defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 2))
+#if !defined(LIBYUV_DISABLE_X86) && \
+  (defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)))
 #define __SSE2__
 #endif
-#ifdef __SSE2__
+#if !defined(LIBYUV_DISABLE_X86) && defined(__SSE2__)
 #include <emmintrin.h>
 #endif
 
@@ -48,7 +49,7 @@ static const double kiW[KERNEL + 1 + 1] = {
   1. / 726.,    // 1 / sum(i:0..3, j..6) K[i]*K[j]
 };
 
-#ifdef __SSE2__
+#if !defined(LIBYUV_DISABLE_X86) && defined(__SSE2__)
 
 #define PWEIGHT(A, B)  static_cast<uint16>(K[(A)] * K[(B)])   // weight product
 #define MAKE_WEIGHT(L)                                               \
@@ -130,7 +131,7 @@ double GetSSIMFullKernel(const uint8 *org, const uint8 *rec,
                          double area_weight) {
   uint32 xm = 0, ym = 0, xxm = 0, xym = 0, yym = 0;
 
-#ifndef __SSE2__
+#if defined(LIBYUV_DISABLE_X86) || !defined(__SSE2__)
 
   org += yo * stride + xo;
   rec += yo * stride + xo;