Half float neon port, enable O2 for neon, and denormal test.

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 2103 matching lines @@
 
   for (i = 0; i < y_plane_size; ++i) {
     EXPECT_EQ(dst_c[i], dst_opt[i]);
   }
 
   free_aligned_buffer_page_end(orig_y);
   free_aligned_buffer_page_end(dst_c);
   free_aligned_buffer_page_end(dst_opt);
 }
 
+// 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 sicne scale is 1/(1<<bits) where bits is 9, 10 or 12.

[hubbe, 2016/10/18 22:40:42]

s/sicne/since/

Also, some people at intel are working on code where bits = 16

+TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_denormal) {
+  int i, j;
+  const int y_plane_size = benchmark_width_ * benchmark_height_ * 2;
+
+  align_buffer_page_end(orig_y, y_plane_size);
+  align_buffer_page_end(dst_c, y_plane_size);
+  align_buffer_page_end(dst_opt, y_plane_size);
+  MemRandomize(orig_y, y_plane_size);
+  memset(dst_c, 0, y_plane_size);
+  memset(dst_opt, 1, y_plane_size);
+
+  // Disable all optimizations.
+  MaskCpuFlags(disable_cpu_flags_);
+  double c_time = get_time();
+  for (j = 0; j < benchmark_iterations_; j++) {
+    HalfFloatPlane((uint16*)orig_y, benchmark_width_ * 2,
+                   (uint16*)dst_c, benchmark_width_ * 2,
+                   1.0f / 65536.0f, benchmark_width_, benchmark_height_);
+  }
+  c_time = (get_time() - c_time) / benchmark_iterations_;
+
+  // Enable optimizations.
+  MaskCpuFlags(benchmark_cpu_info_);
+  double opt_time = get_time();
+  for (j = 0; j < benchmark_iterations_; j++) {
+    HalfFloatPlane((uint16*)orig_y, benchmark_width_ * 2,
+                   (uint16*)dst_opt, benchmark_width_ * 2,
+                   1.0f / 65536.0f, benchmark_width_, benchmark_height_);
+  }
+  opt_time = (get_time() - opt_time) / benchmark_iterations_;
+
+  for (i = 0; i < y_plane_size; ++i) {
+    EXPECT_EQ(dst_c[i], dst_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(orig_y);
+  free_aligned_buffer_page_end(dst_c);
+  free_aligned_buffer_page_end(dst_opt);
+}
+
 TEST_F(LibYUVPlanarTest, TestARGBLumaColorTable) {
   SIMD_ALIGNED(uint8 orig_pixels[1280][4]);
   SIMD_ALIGNED(uint8 dst_pixels_opt[1280][4]);
   SIMD_ALIGNED(uint8 dst_pixels_c[1280][4]);
   memset(orig_pixels, 0, sizeof(orig_pixels));
 
   align_buffer_page_end(lumacolortable, 32768);
   int v = 0;
   for (int i = 0; i < 32768; ++i) {
     lumacolortable[i] = v;
@@ skipping 353 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