HalfFloat neon armv7 fix for destination pointer.

unit_test/planar_test.cc

 /*
  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS. All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 2112 matching lines @@
   int diff = 0;
   for (i = 0; i < y_plane_size; ++i) {
     diff = dst_c[i] - dst_opt[i];
     if (diff) break;
   }
 
   free_aligned_buffer_page_end(orig_y);
   return diff;
 }
 
+#if defined(__arm__)
+static void EnableFlushDenormalToZero(void) {
+  uint32_t cw;
+  __asm__ __volatile__ (
+    "vmrs   %0, fpscr         \n"
+    "orr    %0, %0, #0x1000000        \n"
+    "vmsr   fpscr, %0         \n"
+    : "=r"(cw) :: "memory");
+}
+#endif
+
 // 5 bit exponent with bias of 15 will underflow to a denormal if scale causes
 // exponent to be less than 0.  15 - log2(65536) = -1/  This shouldnt normally
 // happen since scale is 1/(1<<bits) where bits is 9, 10 or 12.
-#define MAXHALFDIFF 0
+
+#define MAXHALFDENORMALDIFF 0
 TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_16bit_denormal) {
+// 32 bit arm rounding on denormal is case is off by 1 compared to C.

[wangcheng, 2016/11/05 00:25:02]

remove "is" in the comment

[fbarchard1, 2016/11/07 17:43:51]

Done.

+#if defined(__arm__)
+  EnableFlushDenormalToZero();
+#endif
   int diff = TestHalfFloatPlane(benchmark_width_, benchmark_height_,
                                 benchmark_iterations_,
                                 disable_cpu_flags_, benchmark_cpu_info_,
                                 1.0f / 65536.0f, 65535);
+  EXPECT_LE(diff, MAXHALFDENORMALDIFF);
+}
+
+#define MAXHALFDIFF 0
+TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_16bit_One) {
+  int diff = TestHalfFloatPlane(benchmark_width_, benchmark_height_,
+                                benchmark_iterations_,
+                                disable_cpu_flags_, benchmark_cpu_info_,
+                                1.0f, 65535);
   EXPECT_LE(diff, MAXHALFDIFF);
 }
 
+TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_16bit_Opt) {
+  int diff = TestHalfFloatPlane(benchmark_width_, benchmark_height_,
+                                benchmark_iterations_,
+                                disable_cpu_flags_, benchmark_cpu_info_,
+                                1.0f / 4096.0f, 65535);
+  EXPECT_LE(diff, MAXHALFDIFF);
+}
+
 TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_10bit_Opt) {
   int diff = TestHalfFloatPlane(benchmark_width_, benchmark_height_,
                                 benchmark_iterations_,
                                 disable_cpu_flags_, benchmark_cpu_info_,
                                 1.0f / 1024.0f, 1023);
   EXPECT_LE(diff, MAXHALFDIFF);
 }
 
 TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_9bit_Opt) {
   int diff = TestHalfFloatPlane(benchmark_width_, benchmark_height_,
@@ skipping 394 matching lines @@
   }
 
   free_aligned_buffer_page_end(src_pixels);
   free_aligned_buffer_page_end(tmp_pixels_u);
   free_aligned_buffer_page_end(tmp_pixels_v);
   free_aligned_buffer_page_end(dst_pixels_opt);
   free_aligned_buffer_page_end(dst_pixels_c);
 }
 
 }  // namespace libyuv