diff --git a/README.chromium b/README.chromium
index 66d3ab3b3..3d87f9a51 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 666
+Version: 667
 License File: LICENSE
 
 Description:
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index 1ea91cb53..c91174421 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 666
+#define LIBYUV_VERSION 667
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT
diff --git a/source/row_common.cc b/source/row_common.cc
index c0ec0e3f4..e336d6557 100644
--- a/source/row_common.cc
+++ b/source/row_common.cc
@@ -737,14 +737,14 @@ void ARGBMultiplyRow_C(const uint8* src_argb0, const uint8* src_argb1,
 void ARGBAddRow_C(const uint8* src_argb0, const uint8* src_argb1,
                   uint8* dst_argb, int width) {
   for (int i = 0; i < width; ++i) {
-    const uint32 b = src_argb0[0];
-    const uint32 g = src_argb0[1];
-    const uint32 r = src_argb0[2];
-    const uint32 a = src_argb0[3];
-    const uint32 b_add = src_argb1[0];
-    const uint32 g_add = src_argb1[1];
-    const uint32 r_add = src_argb1[2];
-    const uint32 a_add = src_argb1[3];
+    const int b = src_argb0[0];
+    const int g = src_argb0[1];
+    const int r = src_argb0[2];
+    const int a = src_argb0[3];
+    const int b_add = src_argb1[0];
+    const int g_add = src_argb1[1];
+    const int r_add = src_argb1[2];
+    const int a_add = src_argb1[3];
     dst_argb[0] = SHADE(b, b_add);
     dst_argb[1] = SHADE(g, g_add);
     dst_argb[2] = SHADE(r, r_add);
@@ -756,19 +756,19 @@ void ARGBAddRow_C(const uint8* src_argb0, const uint8* src_argb1,
 }
 #undef SHADE
 
-#define SHADE(f, v) ((f - v) > f) ? 0 : (f - v)
+#define SHADE(f, v) ((f - v) < 0) ? 0 : (f - v)
 
 void ARGBSubtractRow_C(const uint8* src_argb0, const uint8* src_argb1,
                        uint8* dst_argb, int width) {
   for (int i = 0; i < width; ++i) {
-    const uint32 b = src_argb0[0];
-    const uint32 g = src_argb0[1];
-    const uint32 r = src_argb0[2];
-    const uint32 a = src_argb0[3];
-    const uint32 b_sub = src_argb1[0];
-    const uint32 g_sub = src_argb1[1];
-    const uint32 r_sub = src_argb1[2];
-    const uint32 a_sub = src_argb1[3];
+    const int b = src_argb0[0];
+    const int g = src_argb0[1];
+    const int r = src_argb0[2];
+    const int a = src_argb0[3];
+    const int b_sub = src_argb1[0];
+    const int g_sub = src_argb1[1];
+    const int r_sub = src_argb1[2];
+    const int a_sub = src_argb1[3];
     dst_argb[0] = SHADE(b, b_sub);
     dst_argb[1] = SHADE(g, g_sub);
     dst_argb[2] = SHADE(r, r_sub);
diff --git a/source/row_win.cc b/source/row_win.cc
index 6f71c942a..3e3c2666d 100644
--- a/source/row_win.cc
+++ b/source/row_win.cc
@@ -5350,9 +5350,9 @@ void ARGBMultiplyRow_AVX2(const uint8* src_argb0, const uint8* src_argb1,
     vpmulhuw   ymm0, ymm0, ymm2   // src_argb0 * src_argb1 low 4
     vpmulhuw   ymm1, ymm1, ymm3   // src_argb0 * src_argb1 high 4
     vpackuswb  ymm0, ymm0, ymm1
-    sub        ecx, 8
     vmovdqu    [eax + edx], ymm0
     lea        eax, [eax + 32]
+    sub        ecx, 8
     jg         convertloop
 
     pop        esi
@@ -5381,9 +5381,9 @@ void ARGBAddRow_AVX2(const uint8* src_argb0, const uint8* src_argb1,
  convertloop:
     vmovdqu    ymm0, [eax]              // read 8 pixels from src_argb0
     vpaddusb   ymm0, ymm0, [eax + esi]  // add 8 pixels from src_argb1
-    sub        ecx, 8
     vmovdqu    [eax + edx], ymm0
     lea        eax, [eax + 32]
+    sub        ecx, 8
     jg         convertloop
 
     pop        esi
@@ -5411,9 +5411,9 @@ void ARGBSubtractRow_AVX2(const uint8* src_argb0, const uint8* src_argb1,
  convertloop:
     vmovdqu    ymm0, [eax]              // read 8 pixels from src_argb0
     vpsubusb   ymm0, ymm0, [eax + esi]  // src_argb0 - src_argb1
-    sub        ecx, 8
     vmovdqu    [eax + edx], ymm0
     lea        eax, [eax + 32]
+    sub        ecx, 8
     jg         convertloop
 
     pop        esi