diff --git a/README.chromium b/README.chromium
index 2263bf9c1..5beb92e5d 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,6 +1,6 @@
 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 572
+Version: 573
 License: BSD
 License File: LICENSE
 
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index cf4b72a5b..f5f4ad4c6 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 572
+#define LIBYUV_VERSION 573
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT
diff --git a/source/compare.cc b/source/compare.cc
index 2dbf311b5..b829eb035 100644
--- a/source/compare.cc
+++ b/source/compare.cc
@@ -102,7 +102,9 @@ uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
   }
 #endif
 #if defined(HAS_SUMSQUAREERROR_AVX2)
+  bool clear = false;
   if (TestCpuFlag(kCpuHasAVX2)) {
+    clear = true;
     // Note only used for multiples of 32 so count is not checked.
     SumSquareError = SumSquareError_AVX2;
   }
@@ -130,6 +132,12 @@ uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
   if (remainder) {
     sse += SumSquareError_C(src_a, src_b, remainder);
   }
+
+#if defined(HAS_SUMSQUAREERROR_AVX2)
+  if (clear) {
+    __asm vzeroupper;
+  }
+#endif
   return sse;
 }
 
@@ -157,7 +165,9 @@ uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
   }
 #endif
 #if defined(HAS_SUMSQUAREERROR_AVX2)
+  bool clear = false;
   if (TestCpuFlag(kCpuHasAVX2) && IS_ALIGNED(width, 32)) {
+    clear = true;
     SumSquareError = SumSquareError_AVX2;
   }
 #endif
@@ -168,6 +178,11 @@ uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
     src_b += stride_b;
   }
 
+#if defined(HAS_SUMSQUAREERROR_AVX2)
+  if (clear) {
+    __asm vzeroupper;
+  }
+#endif
   return sse;
 }
 
diff --git a/source/compare_win.cc b/source/compare_win.cc
index 1a4ad1985..720152fd4 100644
--- a/source/compare_win.cc
+++ b/source/compare_win.cc
@@ -47,9 +47,9 @@ uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
     paddd      xmm0, xmm2
     jg         wloop
 
-    pshufd     xmm1, xmm0, 0EEh
+    pshufd     xmm1, xmm0, 0xee
     paddd      xmm0, xmm1
-    pshufd     xmm1, xmm0, 01h
+    pshufd     xmm1, xmm0, 0x01
     paddd      xmm0, xmm1
     movd       eax, xmm0
     ret
@@ -67,7 +67,7 @@ uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
     mov        edx, [esp + 8]    // src_b
     mov        ecx, [esp + 12]   // count
     vpxor      ymm0, ymm0, ymm0  // sum
-    vpxor      ymm5, ymm5, ymm5  // for unpack.
+    vpxor      ymm5, ymm5, ymm5  // constant 0 for unpck
     sub        edx, eax
 
     align      16
@@ -92,9 +92,8 @@ uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
     vpshufd    ymm1, ymm0, 0x01  // 1 + 0 both lanes.
     vpaddd     ymm0, ymm0, ymm1
     vpermq     ymm1, ymm0, 0x02  // high + low lane.
-    vpaddd     ymm4, ymm0, ymm1
-    vzeroupper                   // TODO(fbarchard): Remove.
-    movd       eax, xmm4
+    vpaddd     ymm0, ymm0, ymm1
+    vmovd      eax, xmm0
     ret
   }
 }
@@ -173,14 +172,14 @@ uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
     sub        ecx, 16
     paddd      xmm1, xmm3
 
-    pshufd     xmm2, xmm1, 14    // upper 2 dwords
+    pshufd     xmm2, xmm1, 0x0e  // upper 2 dwords
     paddd      xmm1, xmm2
-    pshufd     xmm2, xmm1, 1
+    pshufd     xmm2, xmm1, 0x01
     paddd      xmm1, xmm2
     paddd      xmm0, xmm1
     jg         wloop
 
-    movd       eax, xmm0        // return hash
+    movd       eax, xmm0         // return hash
     ret
   }
 }