Add SkMSAN.h

src/opts/SkBlitRow_opts_SSE4.cpp

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkBlitRow_opts_SSE4.h"
 
 // Some compilers can't compile SSSE3 or SSE4 intrinsics.  We give them stub methods.
 // The stubs should never be called, so we make them crash just to confirm that.
 #if SK_CPU_SSE_LEVEL < SK_CPU_SSE_LEVEL_SSE41
 void S32A_Opaque_BlitRow32_SSE4(SkPMColor* SK_RESTRICT, const SkPMColor* SK_RESTRICT, int, U8CPU) {
     sk_throw();
 }
 
 #else
 
 #include <smmintrin.h>      // SSE4.1 intrinsics
 #include "SkColorPriv.h"
 #include "SkColor_opts_SSE2.h"
+#include "SkMSAN.h"
 
 void S32A_Opaque_BlitRow32_SSE4(SkPMColor* SK_RESTRICT dst,
                                 const SkPMColor* SK_RESTRICT src,
                                 int count,
                                 U8CPU alpha) {
+    sk_msan_assert_initialized(src, src+count);
+
     SkASSERT(alpha == 255);
     // As long as we can, we'll work on 16 pixel pairs at once.
     int count16 = count / 16;
     __m128i* dst4 = (__m128i*)dst;
     const __m128i* src4 = (const __m128i*)src;
 
     for (int i = 0; i < count16 * 4; i += 4) {
         // Load 16 source pixels.
         __m128i s0 = _mm_loadu_si128(src4+i+0),
                 s1 = _mm_loadu_si128(src4+i+1),
@@ skipping 25 matching lines @@
     // Wrap up the last <= 15 pixels.
     for (int i = count16*16; i < count; i++) {
         // This check is not really necessarily, but it prevents pointless autovectorization.
         if (src[i] & 0xFF000000) {
             dst[i] = SkPMSrcOver(src[i], dst[i]);
         }
     }
 }
 
 #endif