diff --git a/README.chromium b/README.chromium
index 6a7da8db4..2ce0a625a 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 846
+Version: 847
 License: BSD
 License File: LICENSE
 
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index ad6b5253d..19fe8dbd3 100644
--- a/include/libyuv/version.h
+++ b/include/libyuv/version.h
@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 846
+#define LIBYUV_VERSION 847
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT
diff --git a/source/row_posix.cc b/source/row_posix.cc
index 2d7b63eac..7122f48f1 100644
--- a/source/row_posix.cc
+++ b/source/row_posix.cc
@@ -5090,19 +5090,20 @@ void CumulativeSumToAverageRow_SSE2(const int32* topleft, const int32* botleft,
                                     int width, int area, uint8* dst,
                                     int count) {
   asm volatile (
-    "movd      %5,%%xmm4                       \n"
-    "cvtdq2ps  %%xmm4,%%xmm4                   \n"
-    "rcpss     %%xmm4,%%xmm4                   \n"
+    "movd      %5,%%xmm5                       \n"
+    "cvtdq2ps  %%xmm5,%%xmm5                   \n"
+    "rcpss     %%xmm5,%%xmm4                   \n"
     "pshufd    $0x0,%%xmm4,%%xmm4              \n"
     "sub       $0x4,%3                         \n"
     "jl        49f                             \n"
     "cmpl      $0x80,%5                        \n"
     "ja        40f                             \n"
 
-    "pcmpeqb   %%xmm5,%%xmm5                   \n"
-    "psrld     $0x1f,%%xmm5                    \n"
-    "pslld     $0x10,%%xmm5                    \n"
-    "cvtdq2ps  %%xmm5,%%xmm5                   \n"
+    "pshufd    $0x0,%%xmm5,%%xmm5              \n"
+    "pcmpeqb   %%xmm6,%%xmm6                   \n"
+    "psrld     $0x10,%%xmm6                    \n"
+    "cvtdq2ps  %%xmm6,%%xmm6                   \n"
+    "addps     %%xmm6,%%xmm5                   \n"
     "mulps     %%xmm4,%%xmm5                   \n"
     "cvtps2dq  %%xmm5,%%xmm5                   \n"
     "packssdw  %%xmm5,%%xmm5                   \n"
@@ -5222,7 +5223,7 @@ void CumulativeSumToAverageRow_SSE2(const int32* topleft, const int32* botleft,
     , "r14"
 #endif
 #if defined(__SSE2__)
-    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"
+    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
 #endif
   );
 }
diff --git a/source/row_win.cc b/source/row_win.cc
index 0b8258b03..429f907b7 100644
--- a/source/row_win.cc
+++ b/source/row_win.cc
@@ -5763,11 +5763,11 @@ void CumulativeSumToAverageRow_SSE2(const int32* topleft, const int32* botleft,
     mov        eax, topleft  // eax topleft
     mov        esi, botleft  // esi botleft
     mov        edx, width
-    movd       xmm4, area
+    movd       xmm5, area
     mov        edi, dst
     mov        ecx, count
-    cvtdq2ps   xmm4, xmm4
-    rcpss      xmm4, xmm4  // 1.0f / area
+    cvtdq2ps   xmm5, xmm5
+    rcpss      xmm4, xmm5  // 1.0f / area
     pshufd     xmm4, xmm4, 0
     sub        ecx, 4
     jl         l4b
@@ -5775,13 +5775,14 @@ void CumulativeSumToAverageRow_SSE2(const int32* topleft, const int32* botleft,
     cmp        area, 128  // 128 pixels will not overflow 15 bits.
     ja         l4
 
-    pcmpeqb    xmm5, xmm5           // constant of 65536.0
-    psrld      xmm5, 31
-    pslld      xmm5, 16
-    cvtdq2ps   xmm5, xmm5
-    mulps      xmm5, xmm4           // 65536.0 * 1 / area
+    pshufd     xmm5, xmm5, 0        // area
+    pcmpeqb    xmm6, xmm6           // constant of 65536.0 - 1 = 65535.0
+    psrld      xmm6, 16
+    cvtdq2ps   xmm6, xmm6
+    addps      xmm5, xmm6           // (65536.0 + area - 1)
+    mulps      xmm5, xmm4           // (65536.0 + area - 1) * 1 / area
     cvtps2dq   xmm5, xmm5           // 0.16 fixed point
-    packssdw   xmm5, xmm5
+    packssdw   xmm5, xmm5           // 16 bit shorts
 
     // 4 pixel loop small blocks.
     align      4