add ConvertYUVAToARGBRow_MMX to media_perftests

media/base/yuv_convert_perftest.cc

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "base/base_paths.h"
 #include "base/cpu.h"
 #include "base/files/file_util.h"
 #include "base/logging.h"
 #include "base/path_service.h"
 #include "base/time/time.h"
 #include "media/base/simd/convert_yuv_to_rgb.h"
 #include "media/base/yuv_convert.h"
 #include "testing/gtest/include/gtest/gtest.h"
 #include "testing/perf/perf_test.h"
+#include "third_party/libyuv/include/libyuv/row.h"
 
 namespace media {
 #if !defined(ARCH_CPU_ARM_FAMILY) && !defined(ARCH_CPU_MIPS_FAMILY)
 // Size of raw image.
 static const int kSourceWidth = 640;
 static const int kSourceHeight = 360;
 static const int kSourceYSize = kSourceWidth * kSourceHeight;
 static const int kSourceUOffset = kSourceYSize;
 static const int kSourceVOffset = kSourceYSize * 5 / 4;
 static const int kBpp = 4;
@@ skipping 32 matching lines @@
     CHECK_EQ(bytes_read, kYUV12Size);
   }
 
   scoped_ptr<uint8[]> yuv_bytes_;
   scoped_ptr<uint8[]> rgb_bytes_converted_;
 
  private:
   DISALLOW_COPY_AND_ASSIGN(YUVConvertPerfTest);
 };
 
+TEST_F(YUVConvertPerfTest, I422ToARGBRow_SSSE3) {
+  ASSERT_TRUE(base::CPU().has_ssse3());

[fbarchard, 2015/08/12 23:27:05]

Is assert ok?

[DaleCurtis, 2015/08/13 00:46:40]

If the bots pass, then yes. It'll be obvious if they don't :)

+
+  base::TimeTicks start = base::TimeTicks::Now();
+  for (int i = 0; i < kPerfTestIterations; ++i) {
+    for (int row = 0; row < kSourceHeight; ++row) {
+      int chroma_row = row / 2;
+      libyuv::I422ToARGBRow_SSSE3(
+          yuv_bytes_.get() + row * kSourceWidth,
+          yuv_bytes_.get() + kSourceUOffset + (chroma_row * kSourceWidth / 2),
+          yuv_bytes_.get() + kSourceVOffset + (chroma_row * kSourceWidth / 2),
+          rgb_bytes_converted_.get(),
+          kWidth);
+    }
+  }
+  double total_time_seconds = (base::TimeTicks::Now() - start).InSecondsF();
+  perf_test::PrintResult(
+      "yuv_convert_perftest", "", "I422ToARGBRow_SSSE3",
+      kPerfTestIterations / total_time_seconds, "runs/s", true);
+}
+
 TEST_F(YUVConvertPerfTest, ConvertYUVToRGB32Row_SSE) {
   ASSERT_TRUE(base::CPU().has_sse());
 
   base::TimeTicks start = base::TimeTicks::Now();
   for (int i = 0; i < kPerfTestIterations; ++i) {
     for (int row = 0; row < kSourceHeight; ++row) {
       int chroma_row = row / 2;
       ConvertYUVToRGB32Row_SSE(
           yuv_bytes_.get() + row * kSourceWidth,
           yuv_bytes_.get() + kSourceUOffset + (chroma_row * kSourceWidth / 2),
           yuv_bytes_.get() + kSourceVOffset + (chroma_row * kSourceWidth / 2),
           rgb_bytes_converted_.get(),
           kWidth,
           GetLookupTable(YV12));
     }
   }
+  media::EmptyRegisterState();
   double total_time_seconds = (base::TimeTicks::Now() - start).InSecondsF();
   perf_test::PrintResult(
       "yuv_convert_perftest", "", "ConvertYUVToRGB32Row_SSE",
       kPerfTestIterations / total_time_seconds, "runs/s", true);
+}
+
+TEST_F(YUVConvertPerfTest, ConvertYUVAToARGBRow_MMX) {
+  ASSERT_TRUE(base::CPU().has_sse());
+
+  base::TimeTicks start = base::TimeTicks::Now();
+  for (int i = 0; i < kPerfTestIterations; ++i) {
+    for (int row = 0; row < kSourceHeight; ++row) {
+      int chroma_row = row / 2;
+      ConvertYUVAToARGBRow_MMX(
+          yuv_bytes_.get() + row * kSourceWidth,
+          yuv_bytes_.get() + kSourceUOffset + (chroma_row * kSourceWidth / 2),
+          yuv_bytes_.get() + kSourceVOffset + (chroma_row * kSourceWidth / 2),
+          yuv_bytes_.get() + row * kSourceWidth,  // hack: use luma for alpha
+          rgb_bytes_converted_.get(),
+          kWidth,
+          GetLookupTable(YV12));
+    }
+  }
   media::EmptyRegisterState();

[fbarchard, 2015/08/12 23:27:05]

media::EmptyRegisterState() does emms, which allows the float math that follows
to work.

+  double total_time_seconds = (base::TimeTicks::Now() - start).InSecondsF();
+  perf_test::PrintResult(
+      "yuv_convert_perftest", "", "ConvertYUVAToARGBRow_MMX",
+      kPerfTestIterations / total_time_seconds, "runs/s", true);
 }
 
 // 64-bit release + component builds on Windows are too smart and optimizes
 // away the function being tested.
 #if defined(OS_WIN) && (defined(ARCH_CPU_X86) || !defined(COMPONENT_BUILD))
 TEST_F(YUVConvertPerfTest, ScaleYUVToRGB32Row_SSE) {
   ASSERT_TRUE(base::CPU().has_sse());
 
   const int kSourceDx = 80000;  // This value means a scale down.
 
   base::TimeTicks start = base::TimeTicks::Now();
   for (int i = 0; i < kPerfTestIterations; ++i) {
     for (int row = 0; row < kSourceHeight; ++row) {
       int chroma_row = row / 2;
       ScaleYUVToRGB32Row_SSE(
           yuv_bytes_.get() + row * kSourceWidth,
           yuv_bytes_.get() + kSourceUOffset + (chroma_row * kSourceWidth / 2),
           yuv_bytes_.get() + kSourceVOffset + (chroma_row * kSourceWidth / 2),
           rgb_bytes_converted_.get(),
           kWidth,
           kSourceDx,
           GetLookupTable(YV12));
     }
   }
+  media::EmptyRegisterState();
   double total_time_seconds = (base::TimeTicks::Now() - start).InSecondsF();
   perf_test::PrintResult(
       "yuv_convert_perftest", "", "ScaleYUVToRGB32Row_SSE",
       kPerfTestIterations / total_time_seconds, "runs/s", true);
-  media::EmptyRegisterState();
 }
 
 TEST_F(YUVConvertPerfTest, LinearScaleYUVToRGB32Row_SSE) {
   ASSERT_TRUE(base::CPU().has_sse());
 
   const int kSourceDx = 80000;  // This value means a scale down.
 
   base::TimeTicks start = base::TimeTicks::Now();
   for (int i = 0; i < kPerfTestIterations; ++i) {
     for (int row = 0; row < kSourceHeight; ++row) {
       int chroma_row = row / 2;
       LinearScaleYUVToRGB32Row_SSE(
           yuv_bytes_.get() + row * kSourceWidth,
           yuv_bytes_.get() + kSourceUOffset + (chroma_row * kSourceWidth / 2),
           yuv_bytes_.get() + kSourceVOffset + (chroma_row * kSourceWidth / 2),
           rgb_bytes_converted_.get(),
           kWidth,
           kSourceDx,
           GetLookupTable(YV12));
     }
   }
+  media::EmptyRegisterState();
   double total_time_seconds = (base::TimeTicks::Now() - start).InSecondsF();
   perf_test::PrintResult(
       "yuv_convert_perftest", "", "LinearScaleYUVToRGB32Row_SSE",
       kPerfTestIterations / total_time_seconds, "runs/s", true);
-  media::EmptyRegisterState();
 }
 #endif  // defined(OS_WIN) && (ARCH_CPU_X86 || COMPONENT_BUILD)
 
 #endif  // !defined(ARCH_CPU_ARM_FAMILY) && !defined(ARCH_CPU_MIPS_FAMILY)
 
 }  // namespace media