make unittest allocator align to 64 bytes.

blur requires memory be aligned.  change the unittest allocator to guarantee 64 byte alignment.
re-enable blur any test that fails if memory is unaligned.

TBR=harryjin@google.com
BUG=libyuv:596,libyuv:594
TESTED=local build passes with row.h removed from tests.

Review URL: https://codereview.chromium.org/2019753002 .

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1591
+Version: 1592
 License: BSD
 License File: LICENSE
 

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1591
+#define LIBYUV_VERSION 1592
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/planar_functions.cc

@@ -1744,9 +1744,7 @@ int ARGBBlur(const uint8* src_argb, int src_stride_argb,
 #if defined(HAS_CUMULATIVESUMTOAVERAGEROW_SSE2)
   if (TestCpuFlag(kCpuHasSSE2)) {
     ComputeCumulativeSumRow = ComputeCumulativeSumRow_SSE2;
-    if (IS_ALIGNED(dst_cumsum, 16)) {
-      CumulativeSumToAverageRow = CumulativeSumToAverageRow_SSE2;
-    }
+    CumulativeSumToAverageRow = CumulativeSumToAverageRow_SSE2;
   }
 #endif
   // Compute enough CumulativeSum for first row to be blurred. After this

source/row_win.cc

@@ -5275,6 +5275,7 @@ void SobelXYRow_SSE2(const uint8* src_sobelx, const uint8* src_sobely,
 // dst points to pixel to store result to.
 // count is number of averaged pixels to produce.
 // Does 4 pixels at a time.
+// This function requires alignment on accumulation buffer pointers.
 void CumulativeSumToAverageRow_SSE2(const int32* topleft, const int32* botleft,
                                     int width, int area, uint8* dst,
                                     int count) {

unit_test/planar_test.cc

@@ -1929,7 +1929,7 @@ static int TestBlur(int width, int height, int benchmark_iterations,
 }
 
 static const int kBlurSize = 55;
-TEST_F(LibYUVPlanarTest, DISABLED_ARGBBlur_Any) {
+TEST_F(LibYUVPlanarTest, ARGBBlur_Any) {
   int max_diff = TestBlur(benchmark_width_ - 1, benchmark_height_,
                           benchmark_iterations_,
                           disable_cpu_flags_, benchmark_cpu_info_,

unit_test/unit_test.cc

@@ -34,7 +34,7 @@ DEFINE_int32(libyuv_cpu_info, -1,
 // Set flags to -1 for benchmarking to avoid slower C code.
 
 LibYUVConvertTest::LibYUVConvertTest() :
-    benchmark_iterations_(BENCHMARK_ITERATIONS), benchmark_width_(130),
+    benchmark_iterations_(BENCHMARK_ITERATIONS), benchmark_width_(128),
     benchmark_height_(72), disable_cpu_flags_(1), benchmark_cpu_info_(-1) {
   const char* repeat = getenv("LIBYUV_REPEAT");
   if (repeat) {

unit_test/unit_test.h

@@ -67,9 +67,9 @@ static inline bool SizeValid(int src_width, int src_height,
 #define align_buffer_page_end(var, size)                                       \
   uint8* var;                                                                  \
   uint8* var##_mem;                                                            \
-  var##_mem = reinterpret_cast<uint8*>(malloc((((size) + 4095) & ~4095) +      \
-      OFFBY));                                                                 \
-  var = var##_mem + (-(size) & 4095) + OFFBY;
+  var##_mem = reinterpret_cast<uint8*>(malloc(((size) + 4095 + 63) & ~4095));  \
+  var = (uint8*)((intptr_t)(var##_mem + (((size) + 4095 + 63) & ~4095) -       \
+      (size)) & ~63);
 
 #define free_aligned_buffer_page_end(var) \
   free(var##_mem);  \