Quelle  planar_test.cc   Sprache: C

/*
 *  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.
 */

#include <math.h>
#include <stdlib.h>
#include <time.h>

#include "../unit_test/unit_test.h"
#include "libyuv/compare.h"
#include "libyuv/convert.h"
#include "libyuv/convert_argb.h"
#include "libyuv/convert_from.h"
#include "libyuv/convert_from_argb.h"
#include "libyuv/cpu_id.h"
#include "libyuv/planar_functions.h"
#include "libyuv/rotate.h"
#include "libyuv/scale.h"

#ifdef ENABLE_ROW_TESTS
// row.h defines SIMD_ALIGNED, overriding unit_test.h
// TODO(fbarchard): Remove row.h from unittests.  Test public functions.
#include "libyuv/row.h" /* For ScaleSumSamples_Neon */
#endif

#if defined(LIBYUV_BIT_EXACT)
#define EXPECTED_UNATTENUATE_DIFF 0
#else
#define EXPECTED_UNATTENUATE_DIFF 2
#endif

namespace libyuv {

TEST_F(LibYUVPlanarTest, TestAttenuate) {
  const int kSize = 1280 * 4;
  align_buffer_page_end(orig_pixels, kSize);
  align_buffer_page_end(atten_pixels, kSize);
  align_buffer_page_end(unatten_pixels, kSize);
  align_buffer_page_end(atten2_pixels, kSize);

  // Test unattenuation clamps
  orig_pixels[0 * 4 + 0] = 200u;
  orig_pixels[0 * 4 + 1] = 129u;
  orig_pixels[0 * 4 + 2] = 127u;
  orig_pixels[0 * 4 + 3] = 128u;
  // Test unattenuation transparent and opaque are unaffected
  orig_pixels[1 * 4 + 0] = 16u;
  orig_pixels[1 * 4 + 1] = 64u;
  orig_pixels[1 * 4 + 2] = 192u;
  orig_pixels[1 * 4 + 3] = 0u;
  orig_pixels[2 * 4 + 0] = 16u;
  orig_pixels[2 * 4 + 1] = 64u;
  orig_pixels[2 * 4 + 2] = 192u;
  orig_pixels[2 * 4 + 3] = 128u;
  orig_pixels[3 * 4 + 0] = 16u;
  orig_pixels[3 * 4 + 1] = 64u;
  orig_pixels[3 * 4 + 2] = 192u;
  orig_pixels[3 * 4 + 3] = 255u;
  orig_pixels[4 * 4 + 0] = 255u;
  orig_pixels[4 * 4 + 1] = 255u;
  orig_pixels[4 * 4 + 2] = 255u;
  orig_pixels[4 * 4 + 3] = 255u;

  ARGBUnattenuate(orig_pixels, 0, unatten_pixels, 0, 5, 1);
  EXPECT_EQ(255u, unatten_pixels[0 * 4 + 0]);
  EXPECT_EQ(255u, unatten_pixels[0 * 4 + 1]);
  EXPECT_EQ(254u, unatten_pixels[0 * 4 + 2]);
  EXPECT_EQ(128u, unatten_pixels[0 * 4 + 3]);
  EXPECT_EQ(0u, unatten_pixels[1 * 4 + 0]);
  EXPECT_EQ(0u, unatten_pixels[1 * 4 + 1]);
  EXPECT_EQ(0u, unatten_pixels[1 * 4 + 2]);
  EXPECT_EQ(0u, unatten_pixels[1 * 4 + 3]);
  EXPECT_EQ(32u, unatten_pixels[2 * 4 + 0]);
  EXPECT_EQ(128u, unatten_pixels[2 * 4 + 1]);
  EXPECT_EQ(255u, unatten_pixels[2 * 4 + 2]);
  EXPECT_EQ(128u, unatten_pixels[2 * 4 + 3]);
  EXPECT_EQ(16u, unatten_pixels[3 * 4 + 0]);
  EXPECT_EQ(64u, unatten_pixels[3 * 4 + 1]);
  EXPECT_EQ(192u, unatten_pixels[3 * 4 + 2]);
  EXPECT_EQ(255u, unatten_pixels[3 * 4 + 3]);
  EXPECT_EQ(255u, unatten_pixels[4 * 4 + 0]);
  EXPECT_EQ(255u, unatten_pixels[4 * 4 + 1]);
  EXPECT_EQ(255u, unatten_pixels[4 * 4 + 2]);
  EXPECT_EQ(255u, unatten_pixels[4 * 4 + 3]);

  ARGBAttenuate(orig_pixels, 0, atten_pixels, 0, 5, 1);
  EXPECT_EQ(100u, atten_pixels[0 * 4 + 0]);
  EXPECT_EQ(65u, atten_pixels[0 * 4 + 1]);
  EXPECT_EQ(64u, atten_pixels[0 * 4 + 2]);
  EXPECT_EQ(128u, atten_pixels[0 * 4 + 3]);
  EXPECT_EQ(0u, atten_pixels[1 * 4 + 0]);
  EXPECT_EQ(0u, atten_pixels[1 * 4 + 1]);
  EXPECT_EQ(0u, atten_pixels[1 * 4 + 2]);
  EXPECT_EQ(0u, atten_pixels[1 * 4 + 3]);
  EXPECT_EQ(8u, atten_pixels[2 * 4 + 0]);
  EXPECT_EQ(32u, atten_pixels[2 * 4 + 1]);
  EXPECT_EQ(96u, atten_pixels[2 * 4 + 2]);
  EXPECT_EQ(128u, atten_pixels[2 * 4 + 3]);
  EXPECT_EQ(16u, atten_pixels[3 * 4 + 0]);
  EXPECT_EQ(64u, atten_pixels[3 * 4 + 1]);
  EXPECT_EQ(192u, atten_pixels[3 * 4 + 2]);
  EXPECT_EQ(255u, atten_pixels[3 * 4 + 3]);
  EXPECT_EQ(255u, atten_pixels[4 * 4 + 0]);
  EXPECT_EQ(255u, atten_pixels[4 * 4 + 1]);
  EXPECT_EQ(255u, atten_pixels[4 * 4 + 2]);
  EXPECT_EQ(255u, atten_pixels[4 * 4 + 3]);

  // test 255
  for (int i = 0; i < 256; ++i) {
    orig_pixels[i * 4 + 0] = i;
    orig_pixels[i * 4 + 1] = 0;
    orig_pixels[i * 4 + 2] = 0;
    orig_pixels[i * 4 + 3] = 255;
  }
  ARGBAttenuate(orig_pixels, 0, atten_pixels, 0, 256, 1);
  for (int i = 0; i < 256; ++i) {
    EXPECT_EQ(orig_pixels[i * 4 + 0], atten_pixels[i * 4 + 0]);
    EXPECT_EQ(0, atten_pixels[i * 4 + 1]);
    EXPECT_EQ(0, atten_pixels[i * 4 + 2]);
    EXPECT_EQ(255, atten_pixels[i * 4 + 3]);
  }

  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i * 4 + 0] = i;
    orig_pixels[i * 4 + 1] = i / 2;
    orig_pixels[i * 4 + 2] = i / 3;
    orig_pixels[i * 4 + 3] = i;
  }
  ARGBAttenuate(orig_pixels, 0, atten_pixels, 0, 1280, 1);
  ARGBUnattenuate(atten_pixels, 0, unatten_pixels, 0, 1280, 1);
  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBAttenuate(unatten_pixels, 0, atten2_pixels, 0, 1280, 1);
  }
  for (int i = 0; i < 1280; ++i) {
    EXPECT_NEAR(atten_pixels[i * 4 + 0], atten2_pixels[i * 4 + 0], 1);
    EXPECT_NEAR(atten_pixels[i * 4 + 1], atten2_pixels[i * 4 + 1], 1);
    EXPECT_NEAR(atten_pixels[i * 4 + 2], atten2_pixels[i * 4 + 2], 1);
    EXPECT_NEAR(atten_pixels[i * 4 + 3], atten2_pixels[i * 4 + 3], 1);
  }
  // Make sure transparent, 50% and opaque are fully accurate.
  EXPECT_EQ(0, atten_pixels[0 * 4 + 0]);
  EXPECT_EQ(0, atten_pixels[0 * 4 + 1]);
  EXPECT_EQ(0, atten_pixels[0 * 4 + 2]);
  EXPECT_EQ(0, atten_pixels[0 * 4 + 3]);
  EXPECT_EQ(64, atten_pixels[128 * 4 + 0]);
  EXPECT_EQ(32, atten_pixels[128 * 4 + 1]);
  EXPECT_EQ(21, atten_pixels[128 * 4 + 2]);
  EXPECT_EQ(128, atten_pixels[128 * 4 + 3]);
  EXPECT_EQ(255, atten_pixels[255 * 4 + 0]);
  EXPECT_EQ(127, atten_pixels[255 * 4 + 1]);
  EXPECT_EQ(85, atten_pixels[255 * 4 + 2]);
  EXPECT_EQ(255, atten_pixels[255 * 4 + 3]);

  free_aligned_buffer_page_end(atten2_pixels);
  free_aligned_buffer_page_end(unatten_pixels);
  free_aligned_buffer_page_end(atten_pixels);
  free_aligned_buffer_page_end(orig_pixels);
}

static int TestAttenuateI(int width,
                          int height,
                          int benchmark_iterations,
                          int disable_cpu_flags,
                          int benchmark_cpu_info,
                          int invert,
                          int off) {
  if (width < 1) {
    width = 1;
  }
  const int kBpp = 4;
  const int kStride = width * kBpp;
  align_buffer_page_end(src_argb, kStride * height + off);
  align_buffer_page_end(dst_argb_c, kStride * height);
  align_buffer_page_end(dst_argb_opt, kStride * height);
  for (int i = 0; i < kStride * height; ++i) {
    src_argb[i + off] = (fastrand() & 0xff);
  }
  memset(dst_argb_c, 0, kStride * height);
  memset(dst_argb_opt, 0, kStride * height);

  MaskCpuFlags(disable_cpu_flags);
  ARGBAttenuate(src_argb + off, kStride, dst_argb_c, kStride, width,
                invert * height);
  MaskCpuFlags(benchmark_cpu_info);
  for (int i = 0; i < benchmark_iterations; ++i) {
    ARGBAttenuate(src_argb + off, kStride, dst_argb_opt, kStride, width,
                  invert * height);
  }
  int max_diff = 0;
  for (int i = 0; i < kStride * height; ++i) {
    int abs_diff = abs(static_cast<int>(dst_argb_c[i]) -
                       static_cast<int>(dst_argb_opt[i]));
    if (abs_diff > max_diff) {
      max_diff = abs_diff;
    }
  }
  free_aligned_buffer_page_end(src_argb);
  free_aligned_buffer_page_end(dst_argb_c);
  free_aligned_buffer_page_end(dst_argb_opt);
  return max_diff;
}

TEST_F(LibYUVPlanarTest, ARGBAttenuate_Any) {
  int max_diff = TestAttenuateI(benchmark_width_ + 1, benchmark_height_,
                                benchmark_iterations_, disable_cpu_flags_,
                                benchmark_cpu_info_, +1, 0);

  EXPECT_EQ(max_diff, 0);
}

TEST_F(LibYUVPlanarTest, ARGBAttenuate_Unaligned) {
  int max_diff =
      TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
                     disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
  EXPECT_EQ(max_diff, 0);
}

TEST_F(LibYUVPlanarTest, ARGBAttenuate_Invert) {
  int max_diff =
      TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
                     disable_cpu_flags_, benchmark_cpu_info_, -1, 0);
  EXPECT_EQ(max_diff, 0);
}

TEST_F(LibYUVPlanarTest, ARGBAttenuate_Opt) {
  int max_diff =
      TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
                     disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
  EXPECT_EQ(max_diff, 0);
}

static int TestUnattenuateI(int width,
                            int height,
                            int benchmark_iterations,
                            int disable_cpu_flags,
                            int benchmark_cpu_info,
                            int invert,
                            int off) {
  if (width < 1) {
    width = 1;
  }
  const int kBpp = 4;
  const int kStride = width * kBpp;
  align_buffer_page_end(src_argb, kStride * height + off);
  align_buffer_page_end(dst_argb_c, kStride * height);
  align_buffer_page_end(dst_argb_opt, kStride * height);
  for (int i = 0; i < kStride * height; ++i) {
    src_argb[i + off] = (fastrand() & 0xff);
  }
  ARGBAttenuate(src_argb + off, kStride, src_argb + off, kStride, width,
                height);
  memset(dst_argb_c, 0, kStride * height);
  memset(dst_argb_opt, 0, kStride * height);

  MaskCpuFlags(disable_cpu_flags);
  ARGBUnattenuate(src_argb + off, kStride, dst_argb_c, kStride, width,
                  invert * height);
  MaskCpuFlags(benchmark_cpu_info);
  for (int i = 0; i < benchmark_iterations; ++i) {
    ARGBUnattenuate(src_argb + off, kStride, dst_argb_opt, kStride, width,
                    invert * height);
  }
  int max_diff = 0;
  for (int i = 0; i < kStride * height; ++i) {
    int abs_diff = abs(static_cast<int>(dst_argb_c[i]) -
                       static_cast<int>(dst_argb_opt[i]));
    if (abs_diff > max_diff) {
      max_diff = abs_diff;
    }
  }
  free_aligned_buffer_page_end(src_argb);
  free_aligned_buffer_page_end(dst_argb_c);
  free_aligned_buffer_page_end(dst_argb_opt);
  return max_diff;
}

TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Any) {
  int max_diff = TestUnattenuateI(benchmark_width_ + 1, benchmark_height_,
                                  benchmark_iterations_, disable_cpu_flags_,
                                  benchmark_cpu_info_, +1, 0);
  EXPECT_LE(max_diff, EXPECTED_UNATTENUATE_DIFF);
}

TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Unaligned) {
  int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
                                  benchmark_iterations_, disable_cpu_flags_,
                                  benchmark_cpu_info_, +1, 1);
  EXPECT_LE(max_diff, EXPECTED_UNATTENUATE_DIFF);
}

TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Invert) {
  int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
                                  benchmark_iterations_, disable_cpu_flags_,
                                  benchmark_cpu_info_, -1, 0);
  EXPECT_LE(max_diff, EXPECTED_UNATTENUATE_DIFF);
}

TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Opt) {
  int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
                                  benchmark_iterations_, disable_cpu_flags_,
                                  benchmark_cpu_info_, +1, 0);
  EXPECT_LE(max_diff, EXPECTED_UNATTENUATE_DIFF);
}

TEST_F(LibYUVPlanarTest, TestARGBComputeCumulativeSum) {
  SIMD_ALIGNED(uint8_t orig_pixels[16][16][4]);
  SIMD_ALIGNED(int32_t added_pixels[16][16][4]);

  for (int y = 0; y < 16; ++y) {
    for (int x = 0; x < 16; ++x) {
      orig_pixels[y][x][0] = 1u;
      orig_pixels[y][x][1] = 2u;
      orig_pixels[y][x][2] = 3u;
      orig_pixels[y][x][3] = 255u;
    }
  }

  ARGBComputeCumulativeSum(&orig_pixels[0][0][0], 16 * 4,
                           &added_pixels[0][0][0], 16 * 4, 16, 16);

  for (int y = 0; y < 16; ++y) {
    for (int x = 0; x < 16; ++x) {
      EXPECT_EQ((x + 1) * (y + 1), added_pixels[y][x][0]);
      EXPECT_EQ((x + 1) * (y + 1) * 2, added_pixels[y][x][1]);
      EXPECT_EQ((x + 1) * (y + 1) * 3, added_pixels[y][x][2]);
      EXPECT_EQ((x + 1) * (y + 1) * 255, added_pixels[y][x][3]);
    }
  }
}

// near is for legacy platforms.
TEST_F(LibYUVPlanarTest, TestARGBGray) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
  memset(orig_pixels, 0, sizeof(orig_pixels));

  // Test blue
  orig_pixels[0][0] = 255u;
  orig_pixels[0][1] = 0u;
  orig_pixels[0][2] = 0u;
  orig_pixels[0][3] = 128u;
  // Test green
  orig_pixels[1][0] = 0u;
  orig_pixels[1][1] = 255u;
  orig_pixels[1][2] = 0u;
  orig_pixels[1][3] = 0u;
  // Test red
  orig_pixels[2][0] = 0u;
  orig_pixels[2][1] = 0u;
  orig_pixels[2][2] = 255u;
  orig_pixels[2][3] = 255u;
  // Test black
  orig_pixels[3][0] = 0u;
  orig_pixels[3][1] = 0u;
  orig_pixels[3][2] = 0u;
  orig_pixels[3][3] = 255u;
  // Test white
  orig_pixels[4][0] = 255u;
  orig_pixels[4][1] = 255u;
  orig_pixels[4][2] = 255u;
  orig_pixels[4][3] = 255u;
  // Test color
  orig_pixels[5][0] = 16u;
  orig_pixels[5][1] = 64u;
  orig_pixels[5][2] = 192u;
  orig_pixels[5][3] = 224u;
  // Do 16 to test asm version.
  ARGBGray(&orig_pixels[0][0], 0, 0, 0, 16, 1);
  EXPECT_NEAR(29u, orig_pixels[0][0], 1);
  EXPECT_NEAR(29u, orig_pixels[0][1], 1);
  EXPECT_NEAR(29u, orig_pixels[0][2], 1);
  EXPECT_EQ(128u, orig_pixels[0][3]);
  EXPECT_EQ(149u, orig_pixels[1][0]);
  EXPECT_EQ(149u, orig_pixels[1][1]);
  EXPECT_EQ(149u, orig_pixels[1][2]);
  EXPECT_EQ(0u, orig_pixels[1][3]);
  EXPECT_NEAR(77u, orig_pixels[2][0], 1);
  EXPECT_NEAR(77u, orig_pixels[2][1], 1);
  EXPECT_NEAR(77u, orig_pixels[2][2], 1);
  EXPECT_EQ(255u, orig_pixels[2][3]);
  EXPECT_EQ(0u, orig_pixels[3][0]);
  EXPECT_EQ(0u, orig_pixels[3][1]);
  EXPECT_EQ(0u, orig_pixels[3][2]);
  EXPECT_EQ(255u, orig_pixels[3][3]);
  EXPECT_EQ(255u, orig_pixels[4][0]);
  EXPECT_EQ(255u, orig_pixels[4][1]);
  EXPECT_EQ(255u, orig_pixels[4][2]);
  EXPECT_EQ(255u, orig_pixels[4][3]);
  EXPECT_NEAR(97u, orig_pixels[5][0], 1);
  EXPECT_NEAR(97u, orig_pixels[5][1], 1);
  EXPECT_NEAR(97u, orig_pixels[5][2], 1);
  EXPECT_EQ(224u, orig_pixels[5][3]);
  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i / 2;
    orig_pixels[i][2] = i / 3;
    orig_pixels[i][3] = i;
  }
  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBGray(&orig_pixels[0][0], 0, 0, 0, 1280, 1);
  }
}

TEST_F(LibYUVPlanarTest, TestARGBGrayTo) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
  SIMD_ALIGNED(uint8_t gray_pixels[1280][4]);
  memset(orig_pixels, 0, sizeof(orig_pixels));

  // Test blue
  orig_pixels[0][0] = 255u;
  orig_pixels[0][1] = 0u;
  orig_pixels[0][2] = 0u;
  orig_pixels[0][3] = 128u;
  // Test green
  orig_pixels[1][0] = 0u;
  orig_pixels[1][1] = 255u;
  orig_pixels[1][2] = 0u;
  orig_pixels[1][3] = 0u;
  // Test red
  orig_pixels[2][0] = 0u;
  orig_pixels[2][1] = 0u;
  orig_pixels[2][2] = 255u;
  orig_pixels[2][3] = 255u;
  // Test black
  orig_pixels[3][0] = 0u;
  orig_pixels[3][1] = 0u;
  orig_pixels[3][2] = 0u;
  orig_pixels[3][3] = 255u;
  // Test white
  orig_pixels[4][0] = 255u;
  orig_pixels[4][1] = 255u;
  orig_pixels[4][2] = 255u;
  orig_pixels[4][3] = 255u;
  // Test color
  orig_pixels[5][0] = 16u;
  orig_pixels[5][1] = 64u;
  orig_pixels[5][2] = 192u;
  orig_pixels[5][3] = 224u;
  // Do 16 to test asm version.
  ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 16, 1);
  EXPECT_NEAR(30u, gray_pixels[0][0], 1);
  EXPECT_NEAR(30u, gray_pixels[0][1], 1);
  EXPECT_NEAR(30u, gray_pixels[0][2], 1);
  EXPECT_NEAR(128u, gray_pixels[0][3], 1);
  EXPECT_NEAR(149u, gray_pixels[1][0], 1);
  EXPECT_NEAR(149u, gray_pixels[1][1], 1);
  EXPECT_NEAR(149u, gray_pixels[1][2], 1);
  EXPECT_NEAR(0u, gray_pixels[1][3], 1);
  EXPECT_NEAR(76u, gray_pixels[2][0], 1);
  EXPECT_NEAR(76u, gray_pixels[2][1], 1);
  EXPECT_NEAR(76u, gray_pixels[2][2], 1);
  EXPECT_NEAR(255u, gray_pixels[2][3], 1);
  EXPECT_NEAR(0u, gray_pixels[3][0], 1);
  EXPECT_NEAR(0u, gray_pixels[3][1], 1);
  EXPECT_NEAR(0u, gray_pixels[3][2], 1);
  EXPECT_NEAR(255u, gray_pixels[3][3], 1);
  EXPECT_NEAR(255u, gray_pixels[4][0], 1);
  EXPECT_NEAR(255u, gray_pixels[4][1], 1);
  EXPECT_NEAR(255u, gray_pixels[4][2], 1);
  EXPECT_NEAR(255u, gray_pixels[4][3], 1);
  EXPECT_NEAR(96u, gray_pixels[5][0], 1);
  EXPECT_NEAR(96u, gray_pixels[5][1], 1);
  EXPECT_NEAR(96u, gray_pixels[5][2], 1);
  EXPECT_NEAR(224u, gray_pixels[5][3], 1);
  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i / 2;
    orig_pixels[i][2] = i / 3;
    orig_pixels[i][3] = i;
  }
  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 1280, 1);
  }

  for (int i = 0; i < 256; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i;
    orig_pixels[i][2] = i;
    orig_pixels[i][3] = i;
  }
  ARGBGray(&orig_pixels[0][0], 0, 0, 0, 256, 1);
  for (int i = 0; i < 256; ++i) {
    EXPECT_EQ(i, orig_pixels[i][0]);
    EXPECT_EQ(i, orig_pixels[i][1]);
    EXPECT_EQ(i, orig_pixels[i][2]);
    EXPECT_EQ(i, orig_pixels[i][3]);
  }
}

TEST_F(LibYUVPlanarTest, TestARGBSepia) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
  memset(orig_pixels, 0, sizeof(orig_pixels));

  // Test blue
  orig_pixels[0][0] = 255u;
  orig_pixels[0][1] = 0u;
  orig_pixels[0][2] = 0u;
  orig_pixels[0][3] = 128u;
  // Test green
  orig_pixels[1][0] = 0u;
  orig_pixels[1][1] = 255u;
  orig_pixels[1][2] = 0u;
  orig_pixels[1][3] = 0u;
  // Test red
  orig_pixels[2][0] = 0u;
  orig_pixels[2][1] = 0u;
  orig_pixels[2][2] = 255u;
  orig_pixels[2][3] = 255u;
  // Test black
  orig_pixels[3][0] = 0u;
  orig_pixels[3][1] = 0u;
  orig_pixels[3][2] = 0u;
  orig_pixels[3][3] = 255u;
  // Test white
  orig_pixels[4][0] = 255u;
  orig_pixels[4][1] = 255u;
  orig_pixels[4][2] = 255u;
  orig_pixels[4][3] = 255u;
  // Test color
  orig_pixels[5][0] = 16u;
  orig_pixels[5][1] = 64u;
  orig_pixels[5][2] = 192u;
  orig_pixels[5][3] = 224u;
  // Do 16 to test asm version.
  ARGBSepia(&orig_pixels[0][0], 0, 0, 0, 16, 1);
  EXPECT_EQ(33u, orig_pixels[0][0]);
  EXPECT_EQ(43u, orig_pixels[0][1]);
  EXPECT_EQ(47u, orig_pixels[0][2]);
  EXPECT_EQ(128u, orig_pixels[0][3]);
  EXPECT_EQ(135u, orig_pixels[1][0]);
  EXPECT_EQ(175u, orig_pixels[1][1]);
  EXPECT_EQ(195u, orig_pixels[1][2]);
  EXPECT_EQ(0u, orig_pixels[1][3]);
  EXPECT_EQ(69u, orig_pixels[2][0]);
  EXPECT_EQ(89u, orig_pixels[2][1]);
  EXPECT_EQ(99u, orig_pixels[2][2]);
  EXPECT_EQ(255u, orig_pixels[2][3]);
  EXPECT_EQ(0u, orig_pixels[3][0]);
  EXPECT_EQ(0u, orig_pixels[3][1]);
  EXPECT_EQ(0u, orig_pixels[3][2]);
  EXPECT_EQ(255u, orig_pixels[3][3]);
  EXPECT_EQ(239u, orig_pixels[4][0]);
  EXPECT_EQ(255u, orig_pixels[4][1]);
  EXPECT_EQ(255u, orig_pixels[4][2]);
  EXPECT_EQ(255u, orig_pixels[4][3]);
  EXPECT_EQ(88u, orig_pixels[5][0]);
  EXPECT_EQ(114u, orig_pixels[5][1]);
  EXPECT_EQ(127u, orig_pixels[5][2]);
  EXPECT_EQ(224u, orig_pixels[5][3]);

  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i / 2;
    orig_pixels[i][2] = i / 3;
    orig_pixels[i][3] = i;
  }
  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBSepia(&orig_pixels[0][0], 0, 0, 0, 1280, 1);
  }
}

TEST_F(LibYUVPlanarTest, TestARGBColorMatrix) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
  SIMD_ALIGNED(uint8_t dst_pixels_opt[1280][4]);
  SIMD_ALIGNED(uint8_t dst_pixels_c[1280][4]);

  // Matrix for Sepia.
  SIMD_ALIGNED(static const int8_t kRGBToSepia[]) = {
      17 / 2, 68 / 2, 35 / 2, 0, 22 / 2, 88 / 2, 45 / 2, 0,
      24 / 2, 98 / 2, 50 / 2, 0, 0,      0,      0,      64,  // Copy alpha.
  };
  memset(orig_pixels, 0, sizeof(orig_pixels));

  // Test blue
  orig_pixels[0][0] = 255u;
  orig_pixels[0][1] = 0u;
  orig_pixels[0][2] = 0u;
  orig_pixels[0][3] = 128u;
  // Test green
  orig_pixels[1][0] = 0u;
  orig_pixels[1][1] = 255u;
  orig_pixels[1][2] = 0u;
  orig_pixels[1][3] = 0u;
  // Test red
  orig_pixels[2][0] = 0u;
  orig_pixels[2][1] = 0u;
  orig_pixels[2][2] = 255u;
  orig_pixels[2][3] = 255u;
  // Test color
  orig_pixels[3][0] = 16u;
  orig_pixels[3][1] = 64u;
  orig_pixels[3][2] = 192u;
  orig_pixels[3][3] = 224u;
  // Do 16 to test asm version.
  ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0,
                  &kRGBToSepia[0], 16, 1);
  EXPECT_EQ(31u, dst_pixels_opt[0][0]);
  EXPECT_EQ(43u, dst_pixels_opt[0][1]);
  EXPECT_EQ(47u, dst_pixels_opt[0][2]);
  EXPECT_EQ(128u, dst_pixels_opt[0][3]);
  EXPECT_EQ(135u, dst_pixels_opt[1][0]);
  EXPECT_EQ(175u, dst_pixels_opt[1][1]);
  EXPECT_EQ(195u, dst_pixels_opt[1][2]);
  EXPECT_EQ(0u, dst_pixels_opt[1][3]);
  EXPECT_EQ(67u, dst_pixels_opt[2][0]);
  EXPECT_EQ(87u, dst_pixels_opt[2][1]);
  EXPECT_EQ(99u, dst_pixels_opt[2][2]);
  EXPECT_EQ(255u, dst_pixels_opt[2][3]);
  EXPECT_EQ(87u, dst_pixels_opt[3][0]);
  EXPECT_EQ(112u, dst_pixels_opt[3][1]);
  EXPECT_EQ(127u, dst_pixels_opt[3][2]);
  EXPECT_EQ(224u, dst_pixels_opt[3][3]);

  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i / 2;
    orig_pixels[i][2] = i / 3;
    orig_pixels[i][3] = i;
  }
  MaskCpuFlags(disable_cpu_flags_);
  ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_c[0][0], 0,
                  &kRGBToSepia[0], 1280, 1);
  MaskCpuFlags(benchmark_cpu_info_);

  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0,
                    &kRGBToSepia[0], 1280, 1);
  }

  for (int i = 0; i < 1280; ++i) {
    EXPECT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]);
    EXPECT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]);
    EXPECT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]);
    EXPECT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]);
  }
}

TEST_F(LibYUVPlanarTest, TestRGBColorMatrix) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);

  // Matrix for Sepia.
  SIMD_ALIGNED(static const int8_t kRGBToSepia[]) = {
      17, 68, 35, 0, 22, 88, 45, 0,
      24, 98, 50, 0, 0,  0,  0,  0,  // Unused but makes matrix 16 bytes.
  };
  memset(orig_pixels, 0, sizeof(orig_pixels));

  // Test blue
  orig_pixels[0][0] = 255u;
  orig_pixels[0][1] = 0u;
  orig_pixels[0][2] = 0u;
  orig_pixels[0][3] = 128u;
  // Test green
  orig_pixels[1][0] = 0u;
  orig_pixels[1][1] = 255u;
  orig_pixels[1][2] = 0u;
  orig_pixels[1][3] = 0u;
  // Test red
  orig_pixels[2][0] = 0u;
  orig_pixels[2][1] = 0u;
  orig_pixels[2][2] = 255u;
  orig_pixels[2][3] = 255u;
  // Test color
  orig_pixels[3][0] = 16u;
  orig_pixels[3][1] = 64u;
  orig_pixels[3][2] = 192u;
  orig_pixels[3][3] = 224u;
  // Do 16 to test asm version.
  RGBColorMatrix(&orig_pixels[0][0], 0, &kRGBToSepia[0], 0, 0, 16, 1);
  EXPECT_EQ(31u, orig_pixels[0][0]);
  EXPECT_EQ(43u, orig_pixels[0][1]);
  EXPECT_EQ(47u, orig_pixels[0][2]);
  EXPECT_EQ(128u, orig_pixels[0][3]);
  EXPECT_EQ(135u, orig_pixels[1][0]);
  EXPECT_EQ(175u, orig_pixels[1][1]);
  EXPECT_EQ(195u, orig_pixels[1][2]);
  EXPECT_EQ(0u, orig_pixels[1][3]);
  EXPECT_EQ(67u, orig_pixels[2][0]);
  EXPECT_EQ(87u, orig_pixels[2][1]);
  EXPECT_EQ(99u, orig_pixels[2][2]);
  EXPECT_EQ(255u, orig_pixels[2][3]);
  EXPECT_EQ(87u, orig_pixels[3][0]);
  EXPECT_EQ(112u, orig_pixels[3][1]);
  EXPECT_EQ(127u, orig_pixels[3][2]);
  EXPECT_EQ(224u, orig_pixels[3][3]);

  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i / 2;
    orig_pixels[i][2] = i / 3;
    orig_pixels[i][3] = i;
  }
  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    RGBColorMatrix(&orig_pixels[0][0], 0, &kRGBToSepia[0], 0, 0, 1280, 1);
  }
}

TEST_F(LibYUVPlanarTest, TestARGBColorTable) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
  memset(orig_pixels, 0, sizeof(orig_pixels));

  // Matrix for Sepia.
  static const uint8_t kARGBTable[256 * 4] = {
      1u, 2u, 3u, 4u, 5u, 6u, 7u, 8u, 9u, 10u, 11u, 12u, 13u, 14u, 15u, 16u,
  };

  orig_pixels[0][0] = 0u;
  orig_pixels[0][1] = 0u;
  orig_pixels[0][2] = 0u;
  orig_pixels[0][3] = 0u;
  orig_pixels[1][0] = 1u;
  orig_pixels[1][1] = 1u;
  orig_pixels[1][2] = 1u;
  orig_pixels[1][3] = 1u;
  orig_pixels[2][0] = 2u;
  orig_pixels[2][1] = 2u;
  orig_pixels[2][2] = 2u;
  orig_pixels[2][3] = 2u;
  orig_pixels[3][0] = 0u;
  orig_pixels[3][1] = 1u;
  orig_pixels[3][2] = 2u;
  orig_pixels[3][3] = 3u;
  // Do 16 to test asm version.
  ARGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 16, 1);
  EXPECT_EQ(1u, orig_pixels[0][0]);
  EXPECT_EQ(2u, orig_pixels[0][1]);
  EXPECT_EQ(3u, orig_pixels[0][2]);
  EXPECT_EQ(4u, orig_pixels[0][3]);
  EXPECT_EQ(5u, orig_pixels[1][0]);
  EXPECT_EQ(6u, orig_pixels[1][1]);
  EXPECT_EQ(7u, orig_pixels[1][2]);
  EXPECT_EQ(8u, orig_pixels[1][3]);
  EXPECT_EQ(9u, orig_pixels[2][0]);
  EXPECT_EQ(10u, orig_pixels[2][1]);
  EXPECT_EQ(11u, orig_pixels[2][2]);
  EXPECT_EQ(12u, orig_pixels[2][3]);
  EXPECT_EQ(1u, orig_pixels[3][0]);
  EXPECT_EQ(6u, orig_pixels[3][1]);
  EXPECT_EQ(11u, orig_pixels[3][2]);
  EXPECT_EQ(16u, orig_pixels[3][3]);

  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i / 2;
    orig_pixels[i][2] = i / 3;
    orig_pixels[i][3] = i;
  }
  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 1280, 1);
  }
}

// Same as TestARGBColorTable except alpha does not change.
TEST_F(LibYUVPlanarTest, TestRGBColorTable) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
  memset(orig_pixels, 0, sizeof(orig_pixels));

  // Matrix for Sepia.
  static const uint8_t kARGBTable[256 * 4] = {
      1u, 2u, 3u, 4u, 5u, 6u, 7u, 8u, 9u, 10u, 11u, 12u, 13u, 14u, 15u, 16u,
  };

  orig_pixels[0][0] = 0u;
  orig_pixels[0][1] = 0u;
  orig_pixels[0][2] = 0u;
  orig_pixels[0][3] = 0u;
  orig_pixels[1][0] = 1u;
  orig_pixels[1][1] = 1u;
  orig_pixels[1][2] = 1u;
  orig_pixels[1][3] = 1u;
  orig_pixels[2][0] = 2u;
  orig_pixels[2][1] = 2u;
  orig_pixels[2][2] = 2u;
  orig_pixels[2][3] = 2u;
  orig_pixels[3][0] = 0u;
  orig_pixels[3][1] = 1u;
  orig_pixels[3][2] = 2u;
  orig_pixels[3][3] = 3u;
  // Do 16 to test asm version.
  RGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 16, 1);
  EXPECT_EQ(1u, orig_pixels[0][0]);
  EXPECT_EQ(2u, orig_pixels[0][1]);
  EXPECT_EQ(3u, orig_pixels[0][2]);
  EXPECT_EQ(0u, orig_pixels[0][3]);  // Alpha unchanged.
  EXPECT_EQ(5u, orig_pixels[1][0]);
  EXPECT_EQ(6u, orig_pixels[1][1]);
  EXPECT_EQ(7u, orig_pixels[1][2]);
  EXPECT_EQ(1u, orig_pixels[1][3]);  // Alpha unchanged.
  EXPECT_EQ(9u, orig_pixels[2][0]);
  EXPECT_EQ(10u, orig_pixels[2][1]);
  EXPECT_EQ(11u, orig_pixels[2][2]);
  EXPECT_EQ(2u, orig_pixels[2][3]);  // Alpha unchanged.
  EXPECT_EQ(1u, orig_pixels[3][0]);
  EXPECT_EQ(6u, orig_pixels[3][1]);
  EXPECT_EQ(11u, orig_pixels[3][2]);
  EXPECT_EQ(3u, orig_pixels[3][3]);  // Alpha unchanged.

  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i / 2;
    orig_pixels[i][2] = i / 3;
    orig_pixels[i][3] = i;
  }
  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    RGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 1280, 1);
  }
}

TEST_F(LibYUVPlanarTest, TestARGBQuantize) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);

  for (int i = 0; i < 1280; ++i) {
    orig_pixels[i][0] = i;
    orig_pixels[i][1] = i / 2;
    orig_pixels[i][2] = i / 3;
    orig_pixels[i][3] = i;
  }
  ARGBQuantize(&orig_pixels[0][0], 0, (65536 + (8 / 2)) / 8, 8, 8 / 2, 0, 0,
               1280, 1);

  for (int i = 0; i < 1280; ++i) {
    EXPECT_EQ((i / 8 * 8 + 8 / 2) & 255, orig_pixels[i][0]);
    EXPECT_EQ((i / 2 / 8 * 8 + 8 / 2) & 255, orig_pixels[i][1]);
    EXPECT_EQ((i / 3 / 8 * 8 + 8 / 2) & 255, orig_pixels[i][2]);
    EXPECT_EQ(i & 255, orig_pixels[i][3]);
  }
  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBQuantize(&orig_pixels[0][0], 0, (65536 + (8 / 2)) / 8, 8, 8 / 2, 0, 0,
                 1280, 1);
  }
}

TEST_F(LibYUVPlanarTest, ARGBMirror_Opt) {
  align_buffer_page_end(src_pixels, benchmark_width_ * benchmark_height_ * 4);
  align_buffer_page_end(dst_pixels_opt,
                        benchmark_width_ * benchmark_height_ * 4);
  align_buffer_page_end(dst_pixels_c, benchmark_width_ * benchmark_height_ * 4);

  MemRandomize(src_pixels, benchmark_width_ * benchmark_height_ * 4);
  MaskCpuFlags(disable_cpu_flags_);
  ARGBMirror(src_pixels, benchmark_width_ * 4, dst_pixels_c,
             benchmark_width_ * 4, benchmark_width_, benchmark_height_);
  MaskCpuFlags(benchmark_cpu_info_);

  for (int i = 0; i < benchmark_iterations_; ++i) {
    ARGBMirror(src_pixels, benchmark_width_ * 4, dst_pixels_opt,
               benchmark_width_ * 4, benchmark_width_, benchmark_height_);
  }
  for (int i = 0; i < benchmark_width_ * benchmark_height_ * 4; ++i) {
    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
  }
  free_aligned_buffer_page_end(src_pixels);
  free_aligned_buffer_page_end(dst_pixels_opt);
  free_aligned_buffer_page_end(dst_pixels_c);
}

TEST_F(LibYUVPlanarTest, MirrorPlane_Opt) {
  align_buffer_page_end(src_pixels, benchmark_width_ * benchmark_height_);
  align_buffer_page_end(dst_pixels_opt, benchmark_width_ * benchmark_height_);
  align_buffer_page_end(dst_pixels_c, benchmark_width_ * benchmark_height_);

  MemRandomize(src_pixels, benchmark_width_ * benchmark_height_);
  MaskCpuFlags(disable_cpu_flags_);
  MirrorPlane(src_pixels, benchmark_width_, dst_pixels_c, benchmark_width_,
              benchmark_width_, benchmark_height_);
  MaskCpuFlags(benchmark_cpu_info_);

  for (int i = 0; i < benchmark_iterations_; ++i) {
    MirrorPlane(src_pixels, benchmark_width_, dst_pixels_opt, benchmark_width_,
                benchmark_width_, benchmark_height_);
  }
  for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) {
    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
  }
  free_aligned_buffer_page_end(src_pixels);
  free_aligned_buffer_page_end(dst_pixels_opt);
  free_aligned_buffer_page_end(dst_pixels_c);
}

TEST_F(LibYUVPlanarTest, MirrorUVPlane_Opt) {
  align_buffer_page_end(src_pixels, benchmark_width_ * benchmark_height_ * 2);
  align_buffer_page_end(dst_pixels_opt,
                        benchmark_width_ * benchmark_height_ * 2);
  align_buffer_page_end(dst_pixels_c, benchmark_width_ * benchmark_height_ * 2);

  MemRandomize(src_pixels, benchmark_width_ * benchmark_height_ * 2);
  MaskCpuFlags(disable_cpu_flags_);
  MirrorUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_c,
                benchmark_width_ * 2, benchmark_width_, benchmark_height_);
  MaskCpuFlags(benchmark_cpu_info_);

  for (int i = 0; i < benchmark_iterations_; ++i) {
    MirrorUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_opt,
                  benchmark_width_ * 2, benchmark_width_, benchmark_height_);
  }
  for (int i = 0; i < benchmark_width_ * benchmark_height_ * 2; ++i) {
    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
  }
  free_aligned_buffer_page_end(src_pixels);
  free_aligned_buffer_page_end(dst_pixels_opt);
  free_aligned_buffer_page_end(dst_pixels_c);
}

TEST_F(LibYUVPlanarTest, TestShade) {
  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
  SIMD_ALIGNED(uint8_t shade_pixels[1280][4]);
  memset(orig_pixels, 0, sizeof(orig_pixels));

  orig_pixels[0][0] = 10u;
  orig_pixels[0][1] = 20u;
  orig_pixels[0][2] = 40u;
  orig_pixels[0][3] = 80u;
  orig_pixels[1][0] = 0u;
  orig_pixels[1][1] = 0u;
  orig_pixels[1][2] = 0u;
  orig_pixels[1][3] = 255u;
  orig_pixels[2][0] = 0u;
  orig_pixels[2][1] = 0u;
  orig_pixels[2][2] = 0u;
  orig_pixels[2][3] = 0u;
  orig_pixels[3][0] = 0u;
  orig_pixels[3][1] = 0u;
  orig_pixels[3][2] = 0u;
  orig_pixels[3][3] = 0u;
  // Do 8 pixels to allow opt version to be used.
  ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x80ffffff);
  EXPECT_EQ(10u, shade_pixels[0][0]);
  EXPECT_EQ(20u, shade_pixels[0][1]);
  EXPECT_EQ(40u, shade_pixels[0][2]);
  EXPECT_EQ(40u, shade_pixels[0][3]);
  EXPECT_EQ(0u, shade_pixels[1][0]);
  EXPECT_EQ(0u, shade_pixels[1][1]);
  EXPECT_EQ(0u, shade_pixels[1][2]);
  EXPECT_EQ(128u, shade_pixels[1][3]);
  EXPECT_EQ(0u, shade_pixels[2][0]);
  EXPECT_EQ(0u, shade_pixels[2][1]);
  EXPECT_EQ(0u, shade_pixels[2][2]);
  EXPECT_EQ(0u, shade_pixels[2][3]);
  EXPECT_EQ(0u, shade_pixels[3][0]);
  EXPECT_EQ(0u, shade_pixels[3][1]);
  EXPECT_EQ(0u, shade_pixels[3][2]);
  EXPECT_EQ(0u, shade_pixels[3][3]);

  ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x80808080);
  EXPECT_EQ(5u, shade_pixels[0][0]);
  EXPECT_EQ(10u, shade_pixels[0][1]);
  EXPECT_EQ(20u, shade_pixels[0][2]);
  EXPECT_EQ(40u, shade_pixels[0][3]);

  ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x10204080);
  EXPECT_EQ(5u, shade_pixels[0][0]);
  EXPECT_EQ(5u, shade_pixels[0][1]);
  EXPECT_EQ(5u, shade_pixels[0][2]);
  EXPECT_EQ(5u, shade_pixels[0][3]);

  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 1280, 1,
              0x80808080);
  }
}

TEST_F(LibYUVPlanarTest, TestARGBInterpolate) {
  SIMD_ALIGNED(uint8_t orig_pixels_0[1280][4]);
  SIMD_ALIGNED(uint8_t orig_pixels_1[1280][4]);
  SIMD_ALIGNED(uint8_t interpolate_pixels[1280][4]);
  memset(orig_pixels_0, 0, sizeof(orig_pixels_0));
  memset(orig_pixels_1, 0, sizeof(orig_pixels_1));

  orig_pixels_0[0][0] = 16u;
  orig_pixels_0[0][1] = 32u;
  orig_pixels_0[0][2] = 64u;
  orig_pixels_0[0][3] = 128u;
  orig_pixels_0[1][0] = 0u;
  orig_pixels_0[1][1] = 0u;
  orig_pixels_0[1][2] = 0u;
  orig_pixels_0[1][3] = 255u;
  orig_pixels_0[2][0] = 0u;
  orig_pixels_0[2][1] = 0u;
  orig_pixels_0[2][2] = 0u;
  orig_pixels_0[2][3] = 0u;
  orig_pixels_0[3][0] = 0u;
  orig_pixels_0[3][1] = 0u;
  orig_pixels_0[3][2] = 0u;
  orig_pixels_0[3][3] = 0u;

  orig_pixels_1[0][0] = 0u;
  orig_pixels_1[0][1] = 0u;
  orig_pixels_1[0][2] = 0u;
  orig_pixels_1[0][3] = 0u;
  orig_pixels_1[1][0] = 0u;
  orig_pixels_1[1][1] = 0u;
  orig_pixels_1[1][2] = 0u;
  orig_pixels_1[1][3] = 0u;
  orig_pixels_1[2][0] = 0u;
  orig_pixels_1[2][1] = 0u;
  orig_pixels_1[2][2] = 0u;
  orig_pixels_1[2][3] = 0u;
  orig_pixels_1[3][0] = 255u;
  orig_pixels_1[3][1] = 255u;
  orig_pixels_1[3][2] = 255u;
  orig_pixels_1[3][3] = 255u;

  ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0,
                  &interpolate_pixels[0][0], 0, 4, 1, 128);
  EXPECT_EQ(8u, interpolate_pixels[0][0]);
  EXPECT_EQ(16u, interpolate_pixels[0][1]);
  EXPECT_EQ(32u, interpolate_pixels[0][2]);
  EXPECT_EQ(64u, interpolate_pixels[0][3]);
  EXPECT_EQ(0u, interpolate_pixels[1][0]);
  EXPECT_EQ(0u, interpolate_pixels[1][1]);
  EXPECT_EQ(0u, interpolate_pixels[1][2]);
  EXPECT_EQ(128u, interpolate_pixels[1][3]);
  EXPECT_EQ(0u, interpolate_pixels[2][0]);
  EXPECT_EQ(0u, interpolate_pixels[2][1]);
  EXPECT_EQ(0u, interpolate_pixels[2][2]);
  EXPECT_EQ(0u, interpolate_pixels[2][3]);
  EXPECT_EQ(128u, interpolate_pixels[3][0]);
  EXPECT_EQ(128u, interpolate_pixels[3][1]);
  EXPECT_EQ(128u, interpolate_pixels[3][2]);
  EXPECT_EQ(128u, interpolate_pixels[3][3]);

  ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0,
                  &interpolate_pixels[0][0], 0, 4, 1, 0);
  EXPECT_EQ(16u, interpolate_pixels[0][0]);
  EXPECT_EQ(32u, interpolate_pixels[0][1]);
  EXPECT_EQ(64u, interpolate_pixels[0][2]);
  EXPECT_EQ(128u, interpolate_pixels[0][3]);

  ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0,
                  &interpolate_pixels[0][0], 0, 4, 1, 192);

  EXPECT_EQ(4u, interpolate_pixels[0][0]);
  EXPECT_EQ(8u, interpolate_pixels[0][1]);
  EXPECT_EQ(16u, interpolate_pixels[0][2]);
  EXPECT_EQ(32u, interpolate_pixels[0][3]);

  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0,
                    &interpolate_pixels[0][0], 0, 1280, 1, 128);
  }
}

TEST_F(LibYUVPlanarTest, TestInterpolatePlane) {
  SIMD_ALIGNED(uint8_t orig_pixels_0[1280]);
  SIMD_ALIGNED(uint8_t orig_pixels_1[1280]);
  SIMD_ALIGNED(uint8_t interpolate_pixels[1280]);
  memset(orig_pixels_0, 0, sizeof(orig_pixels_0));
  memset(orig_pixels_1, 0, sizeof(orig_pixels_1));

  orig_pixels_0[0] = 16u;
  orig_pixels_0[1] = 32u;
  orig_pixels_0[2] = 64u;
  orig_pixels_0[3] = 128u;
  orig_pixels_0[4] = 0u;
  orig_pixels_0[5] = 0u;
  orig_pixels_0[6] = 0u;
  orig_pixels_0[7] = 255u;
  orig_pixels_0[8] = 0u;
  orig_pixels_0[9] = 0u;
  orig_pixels_0[10] = 0u;
  orig_pixels_0[11] = 0u;
  orig_pixels_0[12] = 0u;
  orig_pixels_0[13] = 0u;
  orig_pixels_0[14] = 0u;
  orig_pixels_0[15] = 0u;

  orig_pixels_1[0] = 0u;
  orig_pixels_1[1] = 0u;
  orig_pixels_1[2] = 0u;
  orig_pixels_1[3] = 0u;
  orig_pixels_1[4] = 0u;
  orig_pixels_1[5] = 0u;
  orig_pixels_1[6] = 0u;
  orig_pixels_1[7] = 0u;
  orig_pixels_1[8] = 0u;
  orig_pixels_1[9] = 0u;
  orig_pixels_1[10] = 0u;
  orig_pixels_1[11] = 0u;
  orig_pixels_1[12] = 255u;
  orig_pixels_1[13] = 255u;
  orig_pixels_1[14] = 255u;
  orig_pixels_1[15] = 255u;

  InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
                   &interpolate_pixels[0], 0, 16, 1, 128);
  EXPECT_EQ(8u, interpolate_pixels[0]);
  EXPECT_EQ(16u, interpolate_pixels[1]);
  EXPECT_EQ(32u, interpolate_pixels[2]);
  EXPECT_EQ(64u, interpolate_pixels[3]);
  EXPECT_EQ(0u, interpolate_pixels[4]);
  EXPECT_EQ(0u, interpolate_pixels[5]);
  EXPECT_EQ(0u, interpolate_pixels[6]);
  EXPECT_EQ(128u, interpolate_pixels[7]);
  EXPECT_EQ(0u, interpolate_pixels[8]);
  EXPECT_EQ(0u, interpolate_pixels[9]);
  EXPECT_EQ(0u, interpolate_pixels[10]);
  EXPECT_EQ(0u, interpolate_pixels[11]);
  EXPECT_EQ(128u, interpolate_pixels[12]);
  EXPECT_EQ(128u, interpolate_pixels[13]);
  EXPECT_EQ(128u, interpolate_pixels[14]);
  EXPECT_EQ(128u, interpolate_pixels[15]);

  InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
                   &interpolate_pixels[0], 0, 16, 1, 0);
  EXPECT_EQ(16u, interpolate_pixels[0]);
  EXPECT_EQ(32u, interpolate_pixels[1]);
  EXPECT_EQ(64u, interpolate_pixels[2]);
  EXPECT_EQ(128u, interpolate_pixels[3]);

  InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
                   &interpolate_pixels[0], 0, 16, 1, 192);

  EXPECT_EQ(4u, interpolate_pixels[0]);
  EXPECT_EQ(8u, interpolate_pixels[1]);
  EXPECT_EQ(16u, interpolate_pixels[2]);
  EXPECT_EQ(32u, interpolate_pixels[3]);

  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
                     &interpolate_pixels[0], 0, 1280, 1, 123);
  }
}

TEST_F(LibYUVPlanarTest, TestInterpolatePlane_16) {
  SIMD_ALIGNED(uint16_t orig_pixels_0[1280]);
  SIMD_ALIGNED(uint16_t orig_pixels_1[1280]);
  SIMD_ALIGNED(uint16_t interpolate_pixels[1280]);
  memset(orig_pixels_0, 0, sizeof(orig_pixels_0));
  memset(orig_pixels_1, 0, sizeof(orig_pixels_1));

  orig_pixels_0[0] = 16u;
  orig_pixels_0[1] = 32u;
  orig_pixels_0[2] = 64u;
  orig_pixels_0[3] = 128u;
  orig_pixels_0[4] = 0u;
  orig_pixels_0[5] = 0u;
  orig_pixels_0[6] = 0u;
  orig_pixels_0[7] = 255u;
  orig_pixels_0[8] = 0u;
  orig_pixels_0[9] = 0u;
  orig_pixels_0[10] = 0u;
  orig_pixels_0[11] = 0u;
  orig_pixels_0[12] = 0u;
  orig_pixels_0[13] = 0u;
  orig_pixels_0[14] = 0u;
  orig_pixels_0[15] = 0u;

  orig_pixels_1[0] = 0u;
  orig_pixels_1[1] = 0u;
  orig_pixels_1[2] = 0u;
  orig_pixels_1[3] = 0u;
  orig_pixels_1[4] = 0u;
  orig_pixels_1[5] = 0u;
  orig_pixels_1[6] = 0u;
  orig_pixels_1[7] = 0u;
  orig_pixels_1[8] = 0u;
  orig_pixels_1[9] = 0u;
  orig_pixels_1[10] = 0u;
  orig_pixels_1[11] = 0u;
  orig_pixels_1[12] = 255u;
  orig_pixels_1[13] = 255u;
  orig_pixels_1[14] = 255u;
  orig_pixels_1[15] = 255u;

  InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
                      &interpolate_pixels[0], 0, 16, 1, 128);
  EXPECT_EQ(8u, interpolate_pixels[0]);
  EXPECT_EQ(16u, interpolate_pixels[1]);
  EXPECT_EQ(32u, interpolate_pixels[2]);
  EXPECT_EQ(64u, interpolate_pixels[3]);
  EXPECT_EQ(0u, interpolate_pixels[4]);
  EXPECT_EQ(0u, interpolate_pixels[5]);
  EXPECT_EQ(0u, interpolate_pixels[6]);
  EXPECT_EQ(128u, interpolate_pixels[7]);
  EXPECT_EQ(0u, interpolate_pixels[8]);
  EXPECT_EQ(0u, interpolate_pixels[9]);
  EXPECT_EQ(0u, interpolate_pixels[10]);
  EXPECT_EQ(0u, interpolate_pixels[11]);
  EXPECT_EQ(128u, interpolate_pixels[12]);
  EXPECT_EQ(128u, interpolate_pixels[13]);
  EXPECT_EQ(128u, interpolate_pixels[14]);
  EXPECT_EQ(128u, interpolate_pixels[15]);

  InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
                      &interpolate_pixels[0], 0, 16, 1, 0);
  EXPECT_EQ(16u, interpolate_pixels[0]);
  EXPECT_EQ(32u, interpolate_pixels[1]);
  EXPECT_EQ(64u, interpolate_pixels[2]);
  EXPECT_EQ(128u, interpolate_pixels[3]);

  InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
                      &interpolate_pixels[0], 0, 16, 1, 192);

  EXPECT_EQ(4u, interpolate_pixels[0]);
  EXPECT_EQ(8u, interpolate_pixels[1]);
  EXPECT_EQ(16u, interpolate_pixels[2]);
  EXPECT_EQ(32u, interpolate_pixels[3]);

  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
    InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
                        &interpolate_pixels[0], 0, 1280, 1, 123);
  }
}

#define TESTTERP(FMT_A, BPP_A, STRIDE_A, FMT_B, BPP_B, STRIDE_B, W1280, TERP, \
                 N, NEG, OFF)                                                 \
  TEST_F(LibYUVPlanarTest, ARGBInterpolate##TERP##N) {                        \
    const int kWidth = W1280;                                                 \
    const int kHeight = benchmark_height_;                                    \
    const int kStrideA =                                                      \
        (kWidth * BPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A;                \
    const int kStrideB =                                                      \
        (kWidth * BPP_B + STRIDE_B - 1) / STRIDE_B * STRIDE_B;                \
    align_buffer_page_end(src_argb_a, kStrideA* kHeight + OFF);               \
    align_buffer_page_end(src_argb_b, kStrideA* kHeight + OFF);               \
    align_buffer_page_end(dst_argb_c, kStrideB* kHeight);                     \
    align_buffer_page_end(dst_argb_opt, kStrideB* kHeight);                   \
    for (int i = 0; i < kStrideA * kHeight; ++i) {                            \
      src_argb_a[i + OFF] = (fastrand() & 0xff);                              \
      src_argb_b[i + OFF] = (fastrand() & 0xff);                              \
    }                                                                         \
    MaskCpuFlags(disable_cpu_flags_);                                         \
    ARGBInterpolate(src_argb_a + OFF, kStrideA, src_argb_b + OFF, kStrideA,   \
                    dst_argb_c, kStrideB, kWidth, NEG kHeight, TERP);         \
    MaskCpuFlags(benchmark_cpu_info_);                                        \
    for (int i = 0; i < benchmark_iterations_; ++i) {                         \
      ARGBInterpolate(src_argb_a + OFF, kStrideA, src_argb_b + OFF, kStrideA, \
                      dst_argb_opt, kStrideB, kWidth, NEG kHeight, TERP);     \
    }                                                                         \
    for (int i = 0; i < kStrideB * kHeight; ++i) {                            \
      EXPECT_EQ(dst_argb_c[i], dst_argb_opt[i]);                              \
    }                                                                         \
    free_aligned_buffer_page_end(src_argb_a);                                 \
    free_aligned_buffer_page_end(src_argb_b);                                 \
    free_aligned_buffer_page_end(dst_argb_c);                                 \
    free_aligned_buffer_page_end(dst_argb_opt);                               \
  }

#define TESTINTERPOLATE(TERP)                                                \
  TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_ + 1, TERP, _Any, +, 0)   \
  TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Unaligned, +, 1) \
  TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Invert, -, 0)    \
  TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Opt, +, 0)

TESTINTERPOLATE(0)
TESTINTERPOLATE(64)
TESTINTERPOLATE(128)
TESTINTERPOLATE(192)
TESTINTERPOLATE(255)

static int TestBlend(int width,
                     int height,
                     int benchmark_iterations,
                     int disable_cpu_flags,
                     int benchmark_cpu_info,
                     int invert,
                     int off,
                     int attenuate) {
  if (width < 1) {
    width = 1;
  }
  const int kBpp = 4;
  const int kStride = width * kBpp;
  align_buffer_page_end(src_argb_a, kStride * height + off);
  align_buffer_page_end(src_argb_b, kStride * height + off);
  align_buffer_page_end(dst_argb_c, kStride * height);
  align_buffer_page_end(dst_argb_opt, kStride * height);
  for (int i = 0; i < kStride * height; ++i) {
    src_argb_a[i + off] = (fastrand() & 0xff);
    src_argb_b[i + off] = (fastrand() & 0xff);
  }
  MemRandomize(src_argb_a, kStride * height + off);
  MemRandomize(src_argb_b, kStride * height + off);
  if (attenuate) {
    ARGBAttenuate(src_argb_a + off, kStride, src_argb_a + off, kStride, width,
                  height);
  }
  memset(dst_argb_c, 255, kStride * height);
  memset(dst_argb_opt, 255, kStride * height);

  MaskCpuFlags(disable_cpu_flags);
  ARGBBlend(src_argb_a + off, kStride, src_argb_b + off, kStride, dst_argb_c,
            kStride, width, invert * height);
  MaskCpuFlags(benchmark_cpu_info);
  for (int i = 0; i < benchmark_iterations; ++i) {
    ARGBBlend(src_argb_a + off, kStride, src_argb_b + off, kStride,
              dst_argb_opt, kStride, width, invert * height);
  }
  int max_diff = 0;
  for (int i = 0; i < kStride * height; ++i) {
    int abs_diff = abs(static_cast<int>(dst_argb_c[i]) -
                       static_cast<int>(dst_argb_opt[i]));
    if (abs_diff > max_diff) {
      max_diff = abs_diff;
    }
  }
  free_aligned_buffer_page_end(src_argb_a);
  free_aligned_buffer_page_end(src_argb_b);
  free_aligned_buffer_page_end(dst_argb_c);
  free_aligned_buffer_page_end(dst_argb_opt);
  return max_diff;
}

TEST_F(LibYUVPlanarTest, ARGBBlend_Any) {
  int max_diff =
      TestBlend(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 1);
  EXPECT_LE(max_diff, 1);
}

TEST_F(LibYUVPlanarTest, ARGBBlend_Unaligned) {
  int max_diff =
      TestBlend(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 1, 1);
  EXPECT_LE(max_diff, 1);
}

TEST_F(LibYUVPlanarTest, ARGBBlend_Invert) {
  int max_diff =
      TestBlend(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, -1, 0, 1);
  EXPECT_LE(max_diff, 1);
}

TEST_F(LibYUVPlanarTest, ARGBBlend_Unattenuated) {
  int max_diff =
      TestBlend(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 0);
  EXPECT_LE(max_diff, 1);
}

TEST_F(LibYUVPlanarTest, ARGBBlend_Opt) {
  int max_diff =
      TestBlend(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 1);
  EXPECT_LE(max_diff, 1);
}

static void TestBlendPlane(int width,
                           int height,
                           int benchmark_iterations,
                           int disable_cpu_flags,
                           int benchmark_cpu_info,
                           int invert,
                           int off) {
  if (width < 1) {
    width = 1;
  }
  const int kBpp = 1;
  const int kStride = width * kBpp;
  align_buffer_page_end(src_argb_a, kStride * height + off);
  align_buffer_page_end(src_argb_b, kStride * height + off);
  align_buffer_page_end(src_argb_alpha, kStride * height + off);
  align_buffer_page_end(dst_argb_c, kStride * height + off);
  align_buffer_page_end(dst_argb_opt, kStride * height + off);
  memset(dst_argb_c, 255, kStride * height + off);
  memset(dst_argb_opt, 255, kStride * height + off);

  // Test source is maintained exactly if alpha is 255.
  for (int i = 0; i < width; ++i) {
    src_argb_a[i + off] = i & 255;
    src_argb_b[i + off] = 255 - (i & 255);
  }
  memset(src_argb_alpha + off, 255, width);
  BlendPlane(src_argb_a + off, width, src_argb_b + off, width,
             src_argb_alpha + off, width, dst_argb_opt + off, width, width, 1);
  for (int i = 0; i < width; ++i) {
    EXPECT_EQ(src_argb_a[i + off], dst_argb_opt[i + off]);
  }
  // Test destination is maintained exactly if alpha is 0.
  memset(src_argb_alpha + off, 0, width);
  BlendPlane(src_argb_a + off, width, src_argb_b + off, width,
             src_argb_alpha + off, width, dst_argb_opt + off, width, width, 1);
  for (int i = 0; i < width; ++i) {
    EXPECT_EQ(src_argb_b[i + off], dst_argb_opt[i + off]);
  }
  for (int i = 0; i < kStride * height; ++i) {
    src_argb_a[i + off] = (fastrand() & 0xff);
    src_argb_b[i + off] = (fastrand() & 0xff);
    src_argb_alpha[i + off] = (fastrand() & 0xff);
  }

  MaskCpuFlags(disable_cpu_flags);
  BlendPlane(src_argb_a + off, width, src_argb_b + off, width,
             src_argb_alpha + off, width, dst_argb_c + off, width, width,
             invert * height);
  MaskCpuFlags(benchmark_cpu_info);
  for (int i = 0; i < benchmark_iterations; ++i) {
    BlendPlane(src_argb_a + off, width, src_argb_b + off, width,
               src_argb_alpha + off, width, dst_argb_opt + off, width, width,
               invert * height);
  }
  for (int i = 0; i < kStride * height; ++i) {
    EXPECT_EQ(dst_argb_c[i + off], dst_argb_opt[i + off]);
  }
  free_aligned_buffer_page_end(src_argb_a);
  free_aligned_buffer_page_end(src_argb_b);
  free_aligned_buffer_page_end(src_argb_alpha);
  free_aligned_buffer_page_end(dst_argb_c);
  free_aligned_buffer_page_end(dst_argb_opt);
}

TEST_F(LibYUVPlanarTest, BlendPlane_Opt) {
  TestBlendPlane(benchmark_width_, benchmark_height_, benchmark_iterations_,
                 disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
}
TEST_F(LibYUVPlanarTest, BlendPlane_Unaligned) {
  TestBlendPlane(benchmark_width_, benchmark_height_, benchmark_iterations_,
                 disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
}
TEST_F(LibYUVPlanarTest, BlendPlane_Any) {
  TestBlendPlane(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
                 disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
}
TEST_F(LibYUVPlanarTest, BlendPlane_Invert) {
  TestBlendPlane(benchmark_width_, benchmark_height_, benchmark_iterations_,
                 disable_cpu_flags_, benchmark_cpu_info_, -1, 1);
}

#define SUBSAMPLE(v, a) ((((v) + (a)-1)) / (a))

static void TestI420Blend(int width,
                          int height,
                          int benchmark_iterations,
                          int disable_cpu_flags,
                          int benchmark_cpu_info,
                          int invert,
                          int off) {
  width = ((width) > 0) ? (width) : 1;
  const int kStrideUV = SUBSAMPLE(width, 2);
  const int kSizeUV = kStrideUV * SUBSAMPLE(height, 2);
  align_buffer_page_end(src_y0, width * height + off);
  align_buffer_page_end(src_u0, kSizeUV + off);
  align_buffer_page_end(src_v0, kSizeUV + off);
  align_buffer_page_end(src_y1, width * height + off);
  align_buffer_page_end(src_u1, kSizeUV + off);
  align_buffer_page_end(src_v1, kSizeUV + off);
  align_buffer_page_end(src_a, width * height + off);
  align_buffer_page_end(dst_y_c, width * height + off);
  align_buffer_page_end(dst_u_c, kSizeUV + off);
  align_buffer_page_end(dst_v_c, kSizeUV + off);
  align_buffer_page_end(dst_y_opt, width * height + off);
  align_buffer_page_end(dst_u_opt, kSizeUV + off);
  align_buffer_page_end(dst_v_opt, kSizeUV + off);

  MemRandomize(src_y0, width * height + off);
  MemRandomize(src_u0, kSizeUV + off);
  MemRandomize(src_v0, kSizeUV + off);
  MemRandomize(src_y1, width * height + off);
  MemRandomize(src_u1, kSizeUV + off);
  MemRandomize(src_v1, kSizeUV + off);
  MemRandomize(src_a, width * height + off);
  memset(dst_y_c, 255, width * height + off);
  memset(dst_u_c, 255, kSizeUV + off);
  memset(dst_v_c, 255, kSizeUV + off);
  memset(dst_y_opt, 255, width * height + off);
  memset(dst_u_opt, 255, kSizeUV + off);
  memset(dst_v_opt, 255, kSizeUV + off);

  MaskCpuFlags(disable_cpu_flags);
  I420Blend(src_y0 + off, width, src_u0 + off, kStrideUV, src_v0 + off,
            kStrideUV, src_y1 + off, width, src_u1 + off, kStrideUV,
            src_v1 + off, kStrideUV, src_a + off, width, dst_y_c + off, width,
            dst_u_c + off, kStrideUV, dst_v_c + off, kStrideUV, width,
            invert * height);
  MaskCpuFlags(benchmark_cpu_info);
  for (int i = 0; i < benchmark_iterations; ++i) {
    I420Blend(src_y0 + off, width, src_u0 + off, kStrideUV, src_v0 + off,
              kStrideUV, src_y1 + off, width, src_u1 + off, kStrideUV,
              src_v1 + off, kStrideUV, src_a + off, width, dst_y_opt + off,
              width, dst_u_opt + off, kStrideUV, dst_v_opt + off, kStrideUV,
              width, invert * height);
  }
  for (int i = 0; i < width * height; ++i) {
    EXPECT_EQ(dst_y_c[i + off], dst_y_opt[i + off]);
  }
  for (int i = 0; i < kSizeUV; ++i) {
    EXPECT_EQ(dst_u_c[i + off], dst_u_opt[i + off]);
    EXPECT_EQ(dst_v_c[i + off], dst_v_opt[i + off]);
  }
  free_aligned_buffer_page_end(src_y0);
  free_aligned_buffer_page_end(src_u0);
  free_aligned_buffer_page_end(src_v0);
  free_aligned_buffer_page_end(src_y1);
  free_aligned_buffer_page_end(src_u1);
  free_aligned_buffer_page_end(src_v1);
  free_aligned_buffer_page_end(src_a);
  free_aligned_buffer_page_end(dst_y_c);
  free_aligned_buffer_page_end(dst_u_c);
  free_aligned_buffer_page_end(dst_v_c);
  free_aligned_buffer_page_end(dst_y_opt);
  free_aligned_buffer_page_end(dst_u_opt);
  free_aligned_buffer_page_end(dst_v_opt);
}

TEST_F(LibYUVPlanarTest, I420Blend_Opt) {
  TestI420Blend(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
}
TEST_F(LibYUVPlanarTest, I420Blend_Unaligned) {
  TestI420Blend(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
}

// TODO(fbarchard): DISABLED because _Any uses C.  Avoid C and re-enable.
TEST_F(LibYUVPlanarTest, DISABLED_I420Blend_Any) {
  TestI420Blend(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
}
TEST_F(LibYUVPlanarTest, I420Blend_Invert) {
  TestI420Blend(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, -1, 0);
}

TEST_F(LibYUVPlanarTest, TestAffine) {
  SIMD_ALIGNED(uint8_t orig_pixels_0[1280][4]);
  SIMD_ALIGNED(uint8_t interpolate_pixels_C[1280][4]);

  for (int i = 0; i < 1280; ++i) {
    for (int j = 0; j < 4; ++j) {
      orig_pixels_0[i][j] = i;
    }
  }

  float uv_step[4] = {0.f, 0.f, 0.75f, 0.f};

  ARGBAffineRow_C(&orig_pixels_0[0][0], 0, &interpolate_pixels_C[0][0], uv_step,
                  1280);
  EXPECT_EQ(0u, interpolate_pixels_C[0][0]);
  EXPECT_EQ(96u, interpolate_pixels_C[128][0]);
  EXPECT_EQ(191u, interpolate_pixels_C[255][3]);

#if defined(HAS_ARGBAFFINEROW_SSE2)
  SIMD_ALIGNED(uint8_t interpolate_pixels_Opt[1280][4]);
  ARGBAffineRow_SSE2(&orig_pixels_0[0][0], 0, &interpolate_pixels_Opt[0][0],
                     uv_step, 1280);
  EXPECT_EQ(0, memcmp(interpolate_pixels_Opt, interpolate_pixels_C, 1280 * 4));

  int has_sse2 = TestCpuFlag(kCpuHasSSE2);
  if (has_sse2) {
    for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
      ARGBAffineRow_SSE2(&orig_pixels_0[0][0], 0, &interpolate_pixels_Opt[0][0],
                         uv_step, 1280);
    }
  }
#endif
}

static int TestCopyPlane(int benchmark_width,
                         int benchmark_height,
                         int benchmark_iterations,
                         int disable_cpu_flags,
                         int benchmark_cpu_info,
                         int invert,
                         int off) {
  const int y_plane_size = benchmark_width * benchmark_height;
  align_buffer_page_end(orig_y, y_plane_size + off);
  align_buffer_page_end(dst_c, y_plane_size);
  align_buffer_page_end(dst_opt, y_plane_size);

  MemRandomize(orig_y + off, y_plane_size);
  memset(dst_c, 1, y_plane_size);
  memset(dst_opt, 2, y_plane_size);

  // Disable all optimizations.
  MaskCpuFlags(disable_cpu_flags);
  for (int i = 0; i < benchmark_iterations; i++) {
    CopyPlane(orig_y + off, benchmark_width, dst_c, benchmark_width,
              benchmark_width, benchmark_height * invert);
  }

  // Enable optimizations.
  MaskCpuFlags(benchmark_cpu_info);
  for (int i = 0; i < benchmark_iterations; i++) {
    CopyPlane(orig_y + off, benchmark_width, dst_opt, benchmark_width,
              benchmark_width, benchmark_height * invert);
  }

  int max_diff = 0;
  for (int i = 0; i < y_plane_size; ++i) {
    int abs_diff =
        abs(static_cast<int>(dst_c[i]) - static_cast<int>(dst_opt[i]));
    if (abs_diff > max_diff) {
      max_diff = abs_diff;
    }
  }

  free_aligned_buffer_page_end(orig_y);
  free_aligned_buffer_page_end(dst_c);
  free_aligned_buffer_page_end(dst_opt);

  return max_diff;
}

TEST_F(LibYUVPlanarTest, CopyPlane_Any) {
  int max_diff = TestCopyPlane(benchmark_width_ + 1, benchmark_height_,
                               benchmark_iterations_, disable_cpu_flags_,
                               benchmark_cpu_info_, +1, 0);
  EXPECT_LE(max_diff, 0);
}

TEST_F(LibYUVPlanarTest, CopyPlane_Unaligned) {
  int max_diff =
      TestCopyPlane(benchmark_width_, benchmark_height_, benchmark_iterations_,
                    disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
  EXPECT_LE(max_diff, 0);
}

TEST_F(LibYUVPlanarTest, CopyPlane_Invert) {
  int max_diff =
      TestCopyPlane(benchmark_width_, benchmark_height_, benchmark_iterations_,
                    disable_cpu_flags_, benchmark_cpu_info_, -1, 0);
  EXPECT_LE(max_diff, 0);
}

TEST_F(LibYUVPlanarTest, CopyPlane_Opt) {
  int max_diff =
      TestCopyPlane(benchmark_width_, benchmark_height_, benchmark_iterations_,
                    disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
  EXPECT_LE(max_diff, 0);
}

TEST_F(LibYUVPlanarTest, TestCopyPlaneZero) {
  // Test to verify copying a rect with a zero height or width does
  // not touch destination memory.
  uint8_t src = 42;
  uint8_t dst = 0;

  // Disable all optimizations.
  MaskCpuFlags(disable_cpu_flags_);
  CopyPlane(&src, 0, &dst, 0, 0, 0);
  EXPECT_EQ(src, 42);
  EXPECT_EQ(dst, 0);

  CopyPlane(&src, 1, &dst, 1, 1, 0);
  EXPECT_EQ(src, 42);
  EXPECT_EQ(dst, 0);

  CopyPlane(&src, 1, &dst, 1, 0, 1);
  EXPECT_EQ(src, 42);
  EXPECT_EQ(dst, 0);

  // Enable optimizations.
  MaskCpuFlags(benchmark_cpu_info_);
  CopyPlane(&src, 0, &dst, 0, 0, 0);
  EXPECT_EQ(src, 42);
  EXPECT_EQ(dst, 0);

  CopyPlane(&src, 1, &dst, 1, 1, 0);
  EXPECT_EQ(src, 42);
  EXPECT_EQ(dst, 0);

  CopyPlane(&src, 1, &dst, 1, 0, 1);
  EXPECT_EQ(src, 42);
  EXPECT_EQ(dst, 0);
}

TEST_F(LibYUVPlanarTest, TestDetilePlane) {
  int i, j;

  // orig is tiled.  Allocate enough memory for tiles.
  int tile_width = (benchmark_width_ + 15) & ~15;
  int tile_height = (benchmark_height_ + 15) & ~15;
  int tile_plane_size = tile_width * tile_height;
  int y_plane_size = benchmark_width_ * benchmark_height_;
  align_buffer_page_end(tile_y, tile_plane_size);
  align_buffer_page_end(dst_c, y_plane_size);
  align_buffer_page_end(dst_opt, y_plane_size);

  MemRandomize(tile_y, tile_plane_size);
  memset(dst_c, 0, y_plane_size);
  memset(dst_opt, 0, y_plane_size);

  // Disable all optimizations.
  MaskCpuFlags(disable_cpu_flags_);
  for (j = 0; j < benchmark_iterations_; j++) {
    DetilePlane(tile_y, tile_width, dst_c, benchmark_width_, benchmark_width_,
                benchmark_height_, 16);
  }

  // Enable optimizations.
  MaskCpuFlags(benchmark_cpu_info_);
  for (j = 0; j < benchmark_iterations_; j++) {
    DetilePlane(tile_y, tile_width, dst_opt, benchmark_width_, benchmark_width_,
                benchmark_height_, 16);
  }

  for (i = 0; i < y_plane_size; ++i) {
    EXPECT_EQ(dst_c[i], dst_opt[i]);
  }

  free_aligned_buffer_page_end(tile_y);
  free_aligned_buffer_page_end(dst_c);
  free_aligned_buffer_page_end(dst_opt);
}

TEST_F(LibYUVPlanarTest, TestDetilePlane_16) {
  int i, j;

  // orig is tiled.  Allocate enough memory for tiles.
  int tile_width = (benchmark_width_ + 15) & ~15;
  int tile_height = (benchmark_height_ + 15) & ~15;
  int tile_plane_size = tile_width * tile_height * 2;
  int y_plane_size = benchmark_width_ * benchmark_height_ * 2;
  align_buffer_page_end(tile_y, tile_plane_size);
  align_buffer_page_end(dst_c, y_plane_size);
  align_buffer_page_end(dst_opt, y_plane_size);

  MemRandomize(tile_y, tile_plane_size);
  memset(dst_c, 0, y_plane_size);
  memset(dst_opt, 0, y_plane_size);

  // Disable all optimizations.
  MaskCpuFlags(disable_cpu_flags_);
  for (j = 0; j < benchmark_iterations_; j++) {
    DetilePlane_16((const uint16_t*)tile_y, tile_width, (uint16_t*)dst_c,
                   benchmark_width_, benchmark_width_, benchmark_height_, 16);
  }

  // Enable optimizations.
  MaskCpuFlags(benchmark_cpu_info_);
  for (j = 0; j < benchmark_iterations_; j++) {
    DetilePlane_16((const uint16_t*)tile_y, tile_width, (uint16_t*)dst_opt,
                   benchmark_width_, benchmark_width_, benchmark_height_, 16);
  }

  for (i = 0; i < y_plane_size; ++i) {
    EXPECT_EQ(dst_c[i], dst_opt[i]);
  }

  free_aligned_buffer_page_end(tile_y);
  free_aligned_buffer_page_end(dst_c);
  free_aligned_buffer_page_end(dst_opt);
}

// Compares DetileSplitUV to 2 step Detile + SplitUV
TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Correctness) {
  int i, j;

  // orig is tiled.  Allocate enough memory for tiles.
  int tile_width = (benchmark_width_ + 15) & ~15;
  int tile_height = (benchmark_height_ + 15) & ~15;
  int tile_plane_size = tile_width * tile_height;
  int uv_plane_size = ((benchmark_width_ + 1) / 2) * benchmark_height_;
  align_buffer_page_end(tile_uv, tile_plane_size);
  align_buffer_page_end(detiled_uv, tile_plane_size);
  align_buffer_page_end(dst_u_two_stage, uv_plane_size);
  align_buffer_page_end(dst_u_opt, uv_plane_size);
  align_buffer_page_end(dst_v_two_stage, uv_plane_size);
  align_buffer_page_end(dst_v_opt, uv_plane_size);

  MemRandomize(tile_uv, tile_plane_size);
  memset(detiled_uv, 0, tile_plane_size);
  memset(dst_u_two_stage, 0, uv_plane_size);
  memset(dst_u_opt, 0, uv_plane_size);
  memset(dst_v_two_stage, 0, uv_plane_size);
  memset(dst_v_opt, 0, uv_plane_size);

  DetileSplitUVPlane(tile_uv, tile_width, dst_u_opt, (benchmark_width_ + 1) / 2,
                     dst_v_opt, (benchmark_width_ + 1) / 2, benchmark_width_,
                     benchmark_height_, 16);

  // Benchmark 2 step conversion for comparison.
  for (j = 0; j < benchmark_iterations_; j++) {
    DetilePlane(tile_uv, tile_width, detiled_uv, benchmark_width_,
                benchmark_width_, benchmark_height_, 16);
    SplitUVPlane(detiled_uv, tile_width, dst_u_two_stage,
                 (benchmark_width_ + 1) / 2, dst_v_two_stage,
                 (benchmark_width_ + 1) / 2, (benchmark_width_ + 1) / 2,
                 benchmark_height_);
  }

  for (i = 0; i < uv_plane_size; ++i) {
    EXPECT_EQ(dst_u_two_stage[i], dst_u_opt[i]);
    EXPECT_EQ(dst_v_two_stage[i], dst_v_opt[i]);
  }

  free_aligned_buffer_page_end(tile_uv);
  free_aligned_buffer_page_end(detiled_uv);
  free_aligned_buffer_page_end(dst_u_two_stage);
  free_aligned_buffer_page_end(dst_u_opt);
  free_aligned_buffer_page_end(dst_v_two_stage);
  free_aligned_buffer_page_end(dst_v_opt);
}

TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Benchmark) {
  int i, j;

  // orig is tiled.  Allocate enough memory for tiles.
  int tile_width = (benchmark_width_ + 15) & ~15;
  int tile_height = (benchmark_height_ + 15) & ~15;
  int tile_plane_size = tile_width * tile_height;
  int uv_plane_size = ((benchmark_width_ + 1) / 2) * benchmark_height_;
  align_buffer_page_end(tile_uv, tile_plane_size);
  align_buffer_page_end(dst_u_c, uv_plane_size);
  align_buffer_page_end(dst_u_opt, uv_plane_size);
  align_buffer_page_end(dst_v_c, uv_plane_size);
  align_buffer_page_end(dst_v_opt, uv_plane_size);

  MemRandomize(tile_uv, tile_plane_size);
  memset(dst_u_c, 0, uv_plane_size);
  memset(dst_u_opt, 0, uv_plane_size);
  memset(dst_v_c, 0, uv_plane_size);
  memset(dst_v_opt, 0, uv_plane_size);

  // Disable all optimizations.
  MaskCpuFlags(disable_cpu_flags_);

  DetileSplitUVPlane(tile_uv, tile_width, dst_u_c, (benchmark_width_ + 1) / 2,
                     dst_v_c, (benchmark_width_ + 1) / 2, benchmark_width_,
                     benchmark_height_, 16);

  // Enable optimizations.
  MaskCpuFlags(benchmark_cpu_info_);

  for (j = 0; j < benchmark_iterations_; j++) {
    DetileSplitUVPlane(
        tile_uv, tile_width, dst_u_opt, (benchmark_width_ + 1) / 2, dst_v_opt,
        (benchmark_width_ + 1) / 2, benchmark_width_, benchmark_height_, 16);
  }

  for (i = 0; i < uv_plane_size; ++i) {
    EXPECT_EQ(dst_u_c[i], dst_u_opt[i]);
    EXPECT_EQ(dst_v_c[i], dst_v_opt[i]);
  }

  free_aligned_buffer_page_end(tile_uv);
--> --------------------

--> maximum size reached

--> --------------------