Blur use circular buffer of Cumulative Sums instead of full frame for better cache coherency.
BUG=none
TEST=none
Review URL: https://webrtc-codereview.appspot.com/646008

git-svn-id: http://libyuv.googlecode.com/svn/trunk@283 16f28f9a-4ce2-e073-06de-1de4eb20be90
diff --git a/source/row_posix.cc b/source/row_posix.cc
index a1d499a..0fdb092 100644
--- a/source/row_posix.cc
+++ b/source/row_posix.cc
@@ -2989,8 +2989,8 @@
   "10:                                         \n"
     "movd      (%0),%%xmm2                     \n"
     "lea       0x4(%0),%0                      \n"
-    "punpcklbw %%xmm4,%%xmm2                   \n"
-    "punpcklwd %%xmm4,%%xmm2                   \n"
+    "punpcklbw %%xmm1,%%xmm2                   \n"
+    "punpcklwd %%xmm1,%%xmm2                   \n"
     "paddd     %%xmm2,%%xmm0                   \n"
     "movdqu    (%1,%2,1),%%xmm2                \n"
     "paddd     %%xmm0,%%xmm2                   \n"