/* * 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 #include #include #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" #include "libyuv/scale_uv.h" #if !defined(LIBYUV_USING_SHARED_LIBRARY) && !defined(COMPONENT_BUILD) && \ !defined(ENABLE_ROW_TESTS) #define ENABLE_ROW_TESTS #endif #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(__has_feature) #if __has_feature(memory_sanitizer) #include #define MSAN_POISON(ptr, size) __msan_allocated_memory(ptr, size) #define MSAN_CHECK_INITIALIZED(ptr, size) \ __msan_check_mem_is_initialized(ptr, size) #endif #endif #ifndef MSAN_POISON #define MSAN_POISON(ptr, size) #endif #ifndef MSAN_CHECK_INITIALIZED #define MSAN_CHECK_INITIALIZED(ptr, size) #endif #define EXPECTED_UNATTENUATE_DIFF 0 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); ASSERT_EQ(255u, unatten_pixels[0 * 4 + 0]); ASSERT_EQ(255u, unatten_pixels[0 * 4 + 1]); ASSERT_EQ(254u, unatten_pixels[0 * 4 + 2]); ASSERT_EQ(128u, unatten_pixels[0 * 4 + 3]); ASSERT_EQ(0u, unatten_pixels[1 * 4 + 0]); ASSERT_EQ(0u, unatten_pixels[1 * 4 + 1]); ASSERT_EQ(0u, unatten_pixels[1 * 4 + 2]); ASSERT_EQ(0u, unatten_pixels[1 * 4 + 3]); ASSERT_EQ(32u, unatten_pixels[2 * 4 + 0]); ASSERT_EQ(128u, unatten_pixels[2 * 4 + 1]); ASSERT_EQ(255u, unatten_pixels[2 * 4 + 2]); ASSERT_EQ(128u, unatten_pixels[2 * 4 + 3]); ASSERT_EQ(16u, unatten_pixels[3 * 4 + 0]); ASSERT_EQ(64u, unatten_pixels[3 * 4 + 1]); ASSERT_EQ(192u, unatten_pixels[3 * 4 + 2]); ASSERT_EQ(255u, unatten_pixels[3 * 4 + 3]); ASSERT_EQ(255u, unatten_pixels[4 * 4 + 0]); ASSERT_EQ(255u, unatten_pixels[4 * 4 + 1]); ASSERT_EQ(255u, unatten_pixels[4 * 4 + 2]); ASSERT_EQ(255u, unatten_pixels[4 * 4 + 3]); ARGBAttenuate(orig_pixels, 0, atten_pixels, 0, 5, 1); ASSERT_EQ(100u, atten_pixels[0 * 4 + 0]); ASSERT_EQ(65u, atten_pixels[0 * 4 + 1]); ASSERT_EQ(64u, atten_pixels[0 * 4 + 2]); ASSERT_EQ(128u, atten_pixels[0 * 4 + 3]); ASSERT_EQ(0u, atten_pixels[1 * 4 + 0]); ASSERT_EQ(0u, atten_pixels[1 * 4 + 1]); ASSERT_EQ(0u, atten_pixels[1 * 4 + 2]); ASSERT_EQ(0u, atten_pixels[1 * 4 + 3]); ASSERT_EQ(8u, atten_pixels[2 * 4 + 0]); ASSERT_EQ(32u, atten_pixels[2 * 4 + 1]); ASSERT_EQ(96u, atten_pixels[2 * 4 + 2]); ASSERT_EQ(128u, atten_pixels[2 * 4 + 3]); ASSERT_EQ(16u, atten_pixels[3 * 4 + 0]); ASSERT_EQ(64u, atten_pixels[3 * 4 + 1]); ASSERT_EQ(192u, atten_pixels[3 * 4 + 2]); ASSERT_EQ(255u, atten_pixels[3 * 4 + 3]); ASSERT_EQ(255u, atten_pixels[4 * 4 + 0]); ASSERT_EQ(255u, atten_pixels[4 * 4 + 1]); ASSERT_EQ(255u, atten_pixels[4 * 4 + 2]); ASSERT_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) { ASSERT_EQ(orig_pixels[i * 4 + 0], atten_pixels[i * 4 + 0]); ASSERT_EQ(0, atten_pixels[i * 4 + 1]); ASSERT_EQ(0, atten_pixels[i * 4 + 2]); ASSERT_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) { ASSERT_NEAR(atten_pixels[i * 4 + 0], atten2_pixels[i * 4 + 0], 1); ASSERT_NEAR(atten_pixels[i * 4 + 1], atten2_pixels[i * 4 + 1], 1); ASSERT_NEAR(atten_pixels[i * 4 + 2], atten2_pixels[i * 4 + 2], 1); ASSERT_NEAR(atten_pixels[i * 4 + 3], atten2_pixels[i * 4 + 3], 1); } // Make sure transparent, 50% and opaque are fully accurate. ASSERT_EQ(0, atten_pixels[0 * 4 + 0]); ASSERT_EQ(0, atten_pixels[0 * 4 + 1]); ASSERT_EQ(0, atten_pixels[0 * 4 + 2]); ASSERT_EQ(0, atten_pixels[0 * 4 + 3]); ASSERT_EQ(64, atten_pixels[128 * 4 + 0]); ASSERT_EQ(32, atten_pixels[128 * 4 + 1]); ASSERT_EQ(21, atten_pixels[128 * 4 + 2]); ASSERT_EQ(128, atten_pixels[128 * 4 + 3]); ASSERT_EQ(255, atten_pixels[255 * 4 + 0]); ASSERT_EQ(127, atten_pixels[255 * 4 + 1]); ASSERT_EQ(85, atten_pixels[255 * 4 + 2]); ASSERT_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, (size_t)kStride * height + off); align_buffer_page_end(dst_argb_c, (size_t)kStride * height); align_buffer_page_end(dst_argb_opt, (size_t)kStride * height); for (size_t i = 0; i < (size_t)kStride * height; ++i) { src_argb[i + off] = (fastrand() & 0xff); } memset(dst_argb_c, 0, (size_t)kStride * height); memset(dst_argb_opt, 0, (size_t)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 (size_t i = 0; i < (size_t)kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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); ASSERT_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); ASSERT_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); ASSERT_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); ASSERT_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, (size_t)kStride * height + off); align_buffer_page_end(dst_argb_c, (size_t)kStride * height); align_buffer_page_end(dst_argb_opt, (size_t)kStride * height); for (size_t i = 0; i < (size_t)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, (size_t)kStride * height); memset(dst_argb_opt, 0, (size_t)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 (size_t i = 0; i < (size_t)kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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); ASSERT_EQ(max_diff, 0); } TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Unaligned) { int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_EQ(max_diff, 0); } TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Invert) { int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_EQ(max_diff, 0); } TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Opt) { int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(max_diff, 0); } 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) { ASSERT_EQ((x + 1) * (y + 1), added_pixels[y][x][0]); ASSERT_EQ((x + 1) * (y + 1) * 2, added_pixels[y][x][1]); ASSERT_EQ((x + 1) * (y + 1) * 3, added_pixels[y][x][2]); ASSERT_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); ASSERT_NEAR(29u, orig_pixels[0][0], 1); ASSERT_NEAR(29u, orig_pixels[0][1], 1); ASSERT_NEAR(29u, orig_pixels[0][2], 1); ASSERT_EQ(128u, orig_pixels[0][3]); ASSERT_EQ(149u, orig_pixels[1][0]); ASSERT_EQ(149u, orig_pixels[1][1]); ASSERT_EQ(149u, orig_pixels[1][2]); ASSERT_EQ(0u, orig_pixels[1][3]); ASSERT_NEAR(77u, orig_pixels[2][0], 1); ASSERT_NEAR(77u, orig_pixels[2][1], 1); ASSERT_NEAR(77u, orig_pixels[2][2], 1); ASSERT_EQ(255u, orig_pixels[2][3]); ASSERT_EQ(0u, orig_pixels[3][0]); ASSERT_EQ(0u, orig_pixels[3][1]); ASSERT_EQ(0u, orig_pixels[3][2]); ASSERT_EQ(255u, orig_pixels[3][3]); ASSERT_EQ(255u, orig_pixels[4][0]); ASSERT_EQ(255u, orig_pixels[4][1]); ASSERT_EQ(255u, orig_pixels[4][2]); ASSERT_EQ(255u, orig_pixels[4][3]); ASSERT_NEAR(97u, orig_pixels[5][0], 1); ASSERT_NEAR(97u, orig_pixels[5][1], 1); ASSERT_NEAR(97u, orig_pixels[5][2], 1); ASSERT_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); ASSERT_NEAR(30u, gray_pixels[0][0], 1); ASSERT_NEAR(30u, gray_pixels[0][1], 1); ASSERT_NEAR(30u, gray_pixels[0][2], 1); ASSERT_NEAR(128u, gray_pixels[0][3], 1); ASSERT_NEAR(149u, gray_pixels[1][0], 1); ASSERT_NEAR(149u, gray_pixels[1][1], 1); ASSERT_NEAR(149u, gray_pixels[1][2], 1); ASSERT_NEAR(0u, gray_pixels[1][3], 1); ASSERT_NEAR(76u, gray_pixels[2][0], 1); ASSERT_NEAR(76u, gray_pixels[2][1], 1); ASSERT_NEAR(76u, gray_pixels[2][2], 1); ASSERT_NEAR(255u, gray_pixels[2][3], 1); ASSERT_NEAR(0u, gray_pixels[3][0], 1); ASSERT_NEAR(0u, gray_pixels[3][1], 1); ASSERT_NEAR(0u, gray_pixels[3][2], 1); ASSERT_NEAR(255u, gray_pixels[3][3], 1); ASSERT_NEAR(255u, gray_pixels[4][0], 1); ASSERT_NEAR(255u, gray_pixels[4][1], 1); ASSERT_NEAR(255u, gray_pixels[4][2], 1); ASSERT_NEAR(255u, gray_pixels[4][3], 1); ASSERT_NEAR(96u, gray_pixels[5][0], 1); ASSERT_NEAR(96u, gray_pixels[5][1], 1); ASSERT_NEAR(96u, gray_pixels[5][2], 1); ASSERT_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) { ASSERT_EQ(i, orig_pixels[i][0]); ASSERT_EQ(i, orig_pixels[i][1]); ASSERT_EQ(i, orig_pixels[i][2]); ASSERT_EQ(i, orig_pixels[i][3]); } // Test RGBX input with uninitialized alpha channel under MSan. SIMD_ALIGNED(uint8_t gray_pixels_c[37][4]); for (int i = 0; i < 37; ++i) { orig_pixels[i][0] = static_cast(i * 7); orig_pixels[i][1] = static_cast(i * 13); orig_pixels[i][2] = static_cast(i * 23); orig_pixels[i][3] = 0u; } MaskCpuFlags(disable_cpu_flags_); ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels_c[0][0], 0, 37, 1); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < 37; ++i) { MSAN_POISON(&orig_pixels[i][3], 1); } MSAN_POISON(gray_pixels, sizeof(gray_pixels)); ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 37, 1); for (int i = 0; i < 37; ++i) { MSAN_CHECK_INITIALIZED(&gray_pixels[i][0], 3); ASSERT_EQ(gray_pixels_c[i][0], gray_pixels[i][0]); ASSERT_EQ(gray_pixels_c[i][1], gray_pixels[i][1]); ASSERT_EQ(gray_pixels_c[i][2], gray_pixels[i][2]); } ARGBGray(&orig_pixels[0][0], 0, 0, 0, 37, 1); for (int i = 0; i < 37; ++i) { MSAN_CHECK_INITIALIZED(&orig_pixels[i][0], 3); ASSERT_EQ(gray_pixels_c[i][0], orig_pixels[i][0]); ASSERT_EQ(gray_pixels_c[i][1], orig_pixels[i][1]); ASSERT_EQ(gray_pixels_c[i][2], orig_pixels[i][2]); } } 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); ASSERT_EQ(33u, orig_pixels[0][0]); ASSERT_EQ(43u, orig_pixels[0][1]); ASSERT_EQ(47u, orig_pixels[0][2]); ASSERT_EQ(128u, orig_pixels[0][3]); ASSERT_EQ(135u, orig_pixels[1][0]); ASSERT_EQ(175u, orig_pixels[1][1]); ASSERT_EQ(195u, orig_pixels[1][2]); ASSERT_EQ(0u, orig_pixels[1][3]); ASSERT_EQ(69u, orig_pixels[2][0]); ASSERT_EQ(89u, orig_pixels[2][1]); ASSERT_EQ(99u, orig_pixels[2][2]); ASSERT_EQ(255u, orig_pixels[2][3]); ASSERT_EQ(0u, orig_pixels[3][0]); ASSERT_EQ(0u, orig_pixels[3][1]); ASSERT_EQ(0u, orig_pixels[3][2]); ASSERT_EQ(255u, orig_pixels[3][3]); ASSERT_EQ(239u, orig_pixels[4][0]); ASSERT_EQ(255u, orig_pixels[4][1]); ASSERT_EQ(255u, orig_pixels[4][2]); ASSERT_EQ(255u, orig_pixels[4][3]); ASSERT_EQ(88u, orig_pixels[5][0]); ASSERT_EQ(114u, orig_pixels[5][1]); ASSERT_EQ(127u, orig_pixels[5][2]); ASSERT_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); ASSERT_EQ(31u, dst_pixels_opt[0][0]); ASSERT_EQ(43u, dst_pixels_opt[0][1]); ASSERT_EQ(47u, dst_pixels_opt[0][2]); ASSERT_EQ(128u, dst_pixels_opt[0][3]); ASSERT_EQ(135u, dst_pixels_opt[1][0]); ASSERT_EQ(175u, dst_pixels_opt[1][1]); ASSERT_EQ(195u, dst_pixels_opt[1][2]); ASSERT_EQ(0u, dst_pixels_opt[1][3]); ASSERT_EQ(67u, dst_pixels_opt[2][0]); ASSERT_EQ(87u, dst_pixels_opt[2][1]); ASSERT_EQ(99u, dst_pixels_opt[2][2]); ASSERT_EQ(255u, dst_pixels_opt[2][3]); ASSERT_EQ(87u, dst_pixels_opt[3][0]); ASSERT_EQ(112u, dst_pixels_opt[3][1]); ASSERT_EQ(127u, dst_pixels_opt[3][2]); ASSERT_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) { ASSERT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); ASSERT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); ASSERT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); ASSERT_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); ASSERT_EQ(31u, orig_pixels[0][0]); ASSERT_EQ(43u, orig_pixels[0][1]); ASSERT_EQ(47u, orig_pixels[0][2]); ASSERT_EQ(128u, orig_pixels[0][3]); ASSERT_EQ(135u, orig_pixels[1][0]); ASSERT_EQ(175u, orig_pixels[1][1]); ASSERT_EQ(195u, orig_pixels[1][2]); ASSERT_EQ(0u, orig_pixels[1][3]); ASSERT_EQ(67u, orig_pixels[2][0]); ASSERT_EQ(87u, orig_pixels[2][1]); ASSERT_EQ(99u, orig_pixels[2][2]); ASSERT_EQ(255u, orig_pixels[2][3]); ASSERT_EQ(87u, orig_pixels[3][0]); ASSERT_EQ(112u, orig_pixels[3][1]); ASSERT_EQ(127u, orig_pixels[3][2]); ASSERT_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); ASSERT_EQ(1u, orig_pixels[0][0]); ASSERT_EQ(2u, orig_pixels[0][1]); ASSERT_EQ(3u, orig_pixels[0][2]); ASSERT_EQ(4u, orig_pixels[0][3]); ASSERT_EQ(5u, orig_pixels[1][0]); ASSERT_EQ(6u, orig_pixels[1][1]); ASSERT_EQ(7u, orig_pixels[1][2]); ASSERT_EQ(8u, orig_pixels[1][3]); ASSERT_EQ(9u, orig_pixels[2][0]); ASSERT_EQ(10u, orig_pixels[2][1]); ASSERT_EQ(11u, orig_pixels[2][2]); ASSERT_EQ(12u, orig_pixels[2][3]); ASSERT_EQ(1u, orig_pixels[3][0]); ASSERT_EQ(6u, orig_pixels[3][1]); ASSERT_EQ(11u, orig_pixels[3][2]); ASSERT_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); ASSERT_EQ(1u, orig_pixels[0][0]); ASSERT_EQ(2u, orig_pixels[0][1]); ASSERT_EQ(3u, orig_pixels[0][2]); ASSERT_EQ(0u, orig_pixels[0][3]); // Alpha unchanged. ASSERT_EQ(5u, orig_pixels[1][0]); ASSERT_EQ(6u, orig_pixels[1][1]); ASSERT_EQ(7u, orig_pixels[1][2]); ASSERT_EQ(1u, orig_pixels[1][3]); // Alpha unchanged. ASSERT_EQ(9u, orig_pixels[2][0]); ASSERT_EQ(10u, orig_pixels[2][1]); ASSERT_EQ(11u, orig_pixels[2][2]); ASSERT_EQ(2u, orig_pixels[2][3]); // Alpha unchanged. ASSERT_EQ(1u, orig_pixels[3][0]); ASSERT_EQ(6u, orig_pixels[3][1]); ASSERT_EQ(11u, orig_pixels[3][2]); ASSERT_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) { ASSERT_EQ((i / 8 * 8 + 8 / 2) & 255, orig_pixels[i][0]); ASSERT_EQ((i / 2 / 8 * 8 + 8 / 2) & 255, orig_pixels[i][1]); ASSERT_EQ((i / 3 / 8 * 8 + 8 / 2) & 255, orig_pixels[i][2]); ASSERT_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) { ASSERT_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_16_Opt) { align_buffer_page_end(src_pixels_8, benchmark_width_ * benchmark_height_ * 2); align_buffer_page_end(dst_pixels_opt_8, benchmark_width_ * benchmark_height_ * 2); align_buffer_page_end(dst_pixels_c_8, benchmark_width_ * benchmark_height_ * 2); uint16_t* src_pixels = reinterpret_cast(src_pixels_8); uint16_t* dst_pixels_opt = reinterpret_cast(dst_pixels_opt_8); uint16_t* dst_pixels_c = reinterpret_cast(dst_pixels_c_8); MemRandomize(src_pixels_8, benchmark_width_ * benchmark_height_ * 2); MaskCpuFlags(disable_cpu_flags_); MirrorPlane_16(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_16(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_8); free_aligned_buffer_page_end(dst_pixels_opt_8); free_aligned_buffer_page_end(dst_pixels_c_8); } 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) { ASSERT_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) { ASSERT_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); ASSERT_EQ(10u, shade_pixels[0][0]); ASSERT_EQ(20u, shade_pixels[0][1]); ASSERT_EQ(40u, shade_pixels[0][2]); ASSERT_EQ(40u, shade_pixels[0][3]); ASSERT_EQ(0u, shade_pixels[1][0]); ASSERT_EQ(0u, shade_pixels[1][1]); ASSERT_EQ(0u, shade_pixels[1][2]); ASSERT_EQ(128u, shade_pixels[1][3]); ASSERT_EQ(0u, shade_pixels[2][0]); ASSERT_EQ(0u, shade_pixels[2][1]); ASSERT_EQ(0u, shade_pixels[2][2]); ASSERT_EQ(0u, shade_pixels[2][3]); ASSERT_EQ(0u, shade_pixels[3][0]); ASSERT_EQ(0u, shade_pixels[3][1]); ASSERT_EQ(0u, shade_pixels[3][2]); ASSERT_EQ(0u, shade_pixels[3][3]); ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x80808080); ASSERT_EQ(5u, shade_pixels[0][0]); ASSERT_EQ(10u, shade_pixels[0][1]); ASSERT_EQ(20u, shade_pixels[0][2]); ASSERT_EQ(40u, shade_pixels[0][3]); ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x10204080); ASSERT_EQ(5u, shade_pixels[0][0]); ASSERT_EQ(5u, shade_pixels[0][1]); ASSERT_EQ(5u, shade_pixels[0][2]); ASSERT_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); ASSERT_EQ(8u, interpolate_pixels[0][0]); ASSERT_EQ(16u, interpolate_pixels[0][1]); ASSERT_EQ(32u, interpolate_pixels[0][2]); ASSERT_EQ(64u, interpolate_pixels[0][3]); ASSERT_EQ(0u, interpolate_pixels[1][0]); ASSERT_EQ(0u, interpolate_pixels[1][1]); ASSERT_EQ(0u, interpolate_pixels[1][2]); ASSERT_EQ(128u, interpolate_pixels[1][3]); ASSERT_EQ(0u, interpolate_pixels[2][0]); ASSERT_EQ(0u, interpolate_pixels[2][1]); ASSERT_EQ(0u, interpolate_pixels[2][2]); ASSERT_EQ(0u, interpolate_pixels[2][3]); ASSERT_EQ(128u, interpolate_pixels[3][0]); ASSERT_EQ(128u, interpolate_pixels[3][1]); ASSERT_EQ(128u, interpolate_pixels[3][2]); ASSERT_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); ASSERT_EQ(16u, interpolate_pixels[0][0]); ASSERT_EQ(32u, interpolate_pixels[0][1]); ASSERT_EQ(64u, interpolate_pixels[0][2]); ASSERT_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); ASSERT_EQ(4u, interpolate_pixels[0][0]); ASSERT_EQ(8u, interpolate_pixels[0][1]); ASSERT_EQ(16u, interpolate_pixels[0][2]); ASSERT_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); ASSERT_EQ(8u, interpolate_pixels[0]); ASSERT_EQ(16u, interpolate_pixels[1]); ASSERT_EQ(32u, interpolate_pixels[2]); ASSERT_EQ(64u, interpolate_pixels[3]); ASSERT_EQ(0u, interpolate_pixels[4]); ASSERT_EQ(0u, interpolate_pixels[5]); ASSERT_EQ(0u, interpolate_pixels[6]); ASSERT_EQ(128u, interpolate_pixels[7]); ASSERT_EQ(0u, interpolate_pixels[8]); ASSERT_EQ(0u, interpolate_pixels[9]); ASSERT_EQ(0u, interpolate_pixels[10]); ASSERT_EQ(0u, interpolate_pixels[11]); ASSERT_EQ(128u, interpolate_pixels[12]); ASSERT_EQ(128u, interpolate_pixels[13]); ASSERT_EQ(128u, interpolate_pixels[14]); ASSERT_EQ(128u, interpolate_pixels[15]); InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0, &interpolate_pixels[0], 0, 16, 1, 0); ASSERT_EQ(16u, interpolate_pixels[0]); ASSERT_EQ(32u, interpolate_pixels[1]); ASSERT_EQ(64u, interpolate_pixels[2]); ASSERT_EQ(128u, interpolate_pixels[3]); InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0, &interpolate_pixels[0], 0, 16, 1, 192); ASSERT_EQ(4u, interpolate_pixels[0]); ASSERT_EQ(8u, interpolate_pixels[1]); ASSERT_EQ(16u, interpolate_pixels[2]); ASSERT_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); ASSERT_EQ(8u, interpolate_pixels[0]); ASSERT_EQ(16u, interpolate_pixels[1]); ASSERT_EQ(32u, interpolate_pixels[2]); ASSERT_EQ(64u, interpolate_pixels[3]); ASSERT_EQ(0u, interpolate_pixels[4]); ASSERT_EQ(0u, interpolate_pixels[5]); ASSERT_EQ(0u, interpolate_pixels[6]); ASSERT_EQ(128u, interpolate_pixels[7]); ASSERT_EQ(0u, interpolate_pixels[8]); ASSERT_EQ(0u, interpolate_pixels[9]); ASSERT_EQ(0u, interpolate_pixels[10]); ASSERT_EQ(0u, interpolate_pixels[11]); ASSERT_EQ(128u, interpolate_pixels[12]); ASSERT_EQ(128u, interpolate_pixels[13]); ASSERT_EQ(128u, interpolate_pixels[14]); ASSERT_EQ(128u, interpolate_pixels[15]); InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0, &interpolate_pixels[0], 0, 16, 1, 0); ASSERT_EQ(16u, interpolate_pixels[0]); ASSERT_EQ(32u, interpolate_pixels[1]); ASSERT_EQ(64u, interpolate_pixels[2]); ASSERT_EQ(128u, interpolate_pixels[3]); InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0, &interpolate_pixels[0], 0, 16, 1, 192); ASSERT_EQ(4u, interpolate_pixels[0]); ASSERT_EQ(8u, interpolate_pixels[1]); ASSERT_EQ(16u, interpolate_pixels[2]); ASSERT_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) { \ ASSERT_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(dst_argb_c[i]) - static_cast(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); ASSERT_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); ASSERT_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); ASSERT_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); ASSERT_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); ASSERT_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) { ASSERT_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) { ASSERT_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) { ASSERT_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) { ASSERT_EQ(dst_y_c[i + off], dst_y_opt[i + off]); } for (int i = 0; i < kSizeUV; ++i) { ASSERT_EQ(dst_u_c[i + off], dst_u_opt[i + off]); ASSERT_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); } #ifdef ENABLE_ROW_TESTS TEST_F(LibYUVPlanarTest, TestAffine) { SIMD_ALIGNED(uint8_t orig_pixels_0[2][1280][4]); SIMD_ALIGNED(uint8_t interpolate_pixels_C[1280][4]); for (int r = 0; r < 2; ++r) { for (int i = 0; i < 1280; ++i) { for (int j = 0; j < 4; ++j) { orig_pixels_0[r][i][j] = i + r * 17; } } } float uv_step[4] = {0.f, 0.f, 0.75f, 1.f / 1280.f}; const int stride = 1280 * 4; ARGBAffineRow_C(&orig_pixels_0[0][0][0], stride, &interpolate_pixels_C[0][0], uv_step, 1280); ASSERT_EQ(0u, interpolate_pixels_C[0][0]); ASSERT_EQ(96u, interpolate_pixels_C[128][0]); ASSERT_EQ(191u, interpolate_pixels_C[255][3]); #if defined(HAS_ARGBAFFINEROW_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); if (has_avx2) { SIMD_ALIGNED(uint8_t interpolate_pixels_Opt[1280][4]); ARGBAffineRow_AVX2(&orig_pixels_0[0][0][0], stride, &interpolate_pixels_Opt[0][0], uv_step, 1280); ASSERT_EQ(0, memcmp(interpolate_pixels_Opt, interpolate_pixels_C, 1280 * 4)); ARGBAffineRow_AVX2(&orig_pixels_0[0][0][0], stride, &interpolate_pixels_Opt[0][0], uv_step, 1279); ASSERT_EQ(0, memcmp(interpolate_pixels_Opt, interpolate_pixels_C, 1279 * 4)); for (int i = 0; i < benchmark_pixels_div1280_; ++i) { ARGBAffineRow_AVX2(&orig_pixels_0[0][0][0], stride, &interpolate_pixels_Opt[0][0], uv_step, 1280); } } #endif #if defined(HAS_ARGBAFFINEROW_NEON) int has_neon = TestCpuFlag(kCpuHasNEON); if (has_neon) { SIMD_ALIGNED(uint8_t interpolate_pixels_Opt[1280][4]); ARGBAffineRow_NEON(&orig_pixels_0[0][0][0], stride, &interpolate_pixels_Opt[0][0], uv_step, 1280); ASSERT_EQ(0, memcmp(interpolate_pixels_Opt, interpolate_pixels_C, 1280 * 4)); ARGBAffineRow_NEON(&orig_pixels_0[0][0][0], stride, &interpolate_pixels_Opt[0][0], uv_step, 1279); ASSERT_EQ(0, memcmp(interpolate_pixels_Opt, interpolate_pixels_C, 1279 * 4)); for (int i = 0; i < benchmark_pixels_div1280_; ++i) { ARGBAffineRow_NEON(&orig_pixels_0[0][0][0], stride, &interpolate_pixels_Opt[0][0], uv_step, 1280); } } #endif } #endif // ENABLE_ROW_TESTS 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(dst_c[i]) - static_cast(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); ASSERT_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); ASSERT_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); ASSERT_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); ASSERT_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); ASSERT_EQ(src, 42); ASSERT_EQ(dst, 0); CopyPlane(&src, 1, &dst, 1, 1, 0); ASSERT_EQ(src, 42); ASSERT_EQ(dst, 0); CopyPlane(&src, 1, &dst, 1, 0, 1); ASSERT_EQ(src, 42); ASSERT_EQ(dst, 0); // Enable optimizations. MaskCpuFlags(benchmark_cpu_info_); CopyPlane(&src, 0, &dst, 0, 0, 0); ASSERT_EQ(src, 42); ASSERT_EQ(dst, 0); CopyPlane(&src, 1, &dst, 1, 1, 0); ASSERT_EQ(src, 42); ASSERT_EQ(dst, 0); CopyPlane(&src, 1, &dst, 1, 0, 1); ASSERT_EQ(src, 42); ASSERT_EQ(dst, 0); } static int TestCopyPlane_16(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) * 2); align_buffer_page_end(dst_c, y_plane_size * 2); align_buffer_page_end(dst_opt, y_plane_size * 2); uint16_t* orig_y_16 = reinterpret_cast(orig_y) + off; uint16_t* dst_c_16 = reinterpret_cast(dst_c); uint16_t* dst_opt_16 = reinterpret_cast(dst_opt); MemRandomize(orig_y, (y_plane_size + off) * 2); memset(dst_c, 1, y_plane_size * 2); memset(dst_opt, 2, y_plane_size * 2); // Disable all optimizations. MaskCpuFlags(disable_cpu_flags); for (int i = 0; i < benchmark_iterations; i++) { CopyPlane_16(orig_y_16, benchmark_width, dst_c_16, benchmark_width, benchmark_width, benchmark_height * invert); } // Enable optimizations. MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; i++) { CopyPlane_16(orig_y_16, benchmark_width, dst_opt_16, 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(dst_c_16[i]) - static_cast(dst_opt_16[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_16_Any) { int max_diff = TestCopyPlane_16(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_LE(max_diff, 0); } TEST_F(LibYUVPlanarTest, CopyPlane_16_Unaligned) { int max_diff = TestCopyPlane_16(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_LE(max_diff, 0); } TEST_F(LibYUVPlanarTest, CopyPlane_16_Invert) { int max_diff = TestCopyPlane_16(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_LE(max_diff, 0); } TEST_F(LibYUVPlanarTest, CopyPlane_16_Opt) { int max_diff = TestCopyPlane_16(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_LE(max_diff, 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) { ASSERT_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) { ASSERT_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 uv_width = (benchmark_width_ + 1) & ~1; int tile_width = (uv_width + 15) & ~15; int tile_height = (benchmark_height_ + 15) & ~15; int tile_plane_size = tile_width * tile_height; int uv_plane_size = (uv_width / 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, uv_width / 2, dst_v_opt, uv_width / 2, uv_width, benchmark_height_, 16); // Benchmark 2 step conversion for comparison. for (j = 0; j < benchmark_iterations_; j++) { DetilePlane(tile_uv, tile_width, detiled_uv, uv_width, uv_width, benchmark_height_, 16); SplitUVPlane(detiled_uv, uv_width, dst_u_two_stage, uv_width / 2, dst_v_two_stage, uv_width / 2, uv_width / 2, benchmark_height_); } for (i = 0; i < uv_plane_size; ++i) { ASSERT_EQ(dst_u_two_stage[i], dst_u_opt[i]); ASSERT_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 uv_width = (benchmark_width_ + 1) & ~1; int tile_width = (uv_width + 15) & ~15; int tile_height = (benchmark_height_ + 15) & ~15; int tile_plane_size = tile_width * tile_height; int uv_plane_size = (uv_width / 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, uv_width / 2, dst_v_c, uv_width / 2, uv_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, uv_width / 2, dst_v_opt, uv_width / 2, uv_width, benchmark_height_, 16); } for (i = 0; i < uv_plane_size; ++i) { ASSERT_EQ(dst_u_c[i], dst_u_opt[i]); ASSERT_EQ(dst_v_c[i], dst_v_opt[i]); } free_aligned_buffer_page_end(tile_uv); free_aligned_buffer_page_end(dst_u_c); free_aligned_buffer_page_end(dst_u_opt); free_aligned_buffer_page_end(dst_v_c); free_aligned_buffer_page_end(dst_v_opt); } static int TestMultiply(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_a, (size_t)kStride * height + off); align_buffer_page_end(src_argb_b, (size_t)kStride * height + off); align_buffer_page_end(dst_argb_c, (size_t)kStride * height); align_buffer_page_end(dst_argb_opt, (size_t)kStride * height); for (size_t i = 0; i < (size_t)kStride * height; ++i) { src_argb_a[i + off] = (fastrand() & 0xff); src_argb_b[i + off] = (fastrand() & 0xff); } memset(dst_argb_c, 0, (size_t)kStride * height); memset(dst_argb_opt, 0, (size_t)kStride * height); MaskCpuFlags(disable_cpu_flags); ARGBMultiply(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) { ARGBMultiply(src_argb_a + off, kStride, src_argb_b + off, kStride, dst_argb_opt, kStride, width, invert * height); } int max_diff = 0; for (size_t i = 0; i < (size_t)kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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, ARGBMultiply_Any) { int max_diff = TestMultiply(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_LE(max_diff, 0); } TEST_F(LibYUVPlanarTest, ARGBMultiply_Unaligned) { int max_diff = TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_LE(max_diff, 0); } TEST_F(LibYUVPlanarTest, ARGBMultiply_Invert) { int max_diff = TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_LE(max_diff, 0); } TEST_F(LibYUVPlanarTest, ARGBMultiply_Opt) { int max_diff = TestMultiply(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_LE(max_diff, 0); } static int TestAdd(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_a, (size_t)kStride * height + off); align_buffer_page_end(src_argb_b, (size_t)kStride * height + off); align_buffer_page_end(dst_argb_c, (size_t)kStride * height); align_buffer_page_end(dst_argb_opt, (size_t)kStride * height); for (size_t i = 0; i < (size_t)kStride * height; ++i) { src_argb_a[i + off] = (fastrand() & 0xff); src_argb_b[i + off] = (fastrand() & 0xff); } memset(dst_argb_c, 0, (size_t)kStride * height); memset(dst_argb_opt, 0, (size_t)kStride * height); MaskCpuFlags(disable_cpu_flags); ARGBAdd(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) { ARGBAdd(src_argb_a + off, kStride, src_argb_b + off, kStride, dst_argb_opt, kStride, width, invert * height); } int max_diff = 0; for (size_t i = 0; i < (size_t)kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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, ARGBAdd_Any) { int max_diff = TestAdd(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, ARGBAdd_Unaligned) { int max_diff = TestAdd(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, ARGBAdd_Invert) { int max_diff = TestAdd(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, ARGBAdd_Opt) { int max_diff = TestAdd(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_LE(max_diff, 1); } static int TestSubtract(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_a, (size_t)kStride * height + off); align_buffer_page_end(src_argb_b, (size_t)kStride * height + off); align_buffer_page_end(dst_argb_c, (size_t)kStride * height); align_buffer_page_end(dst_argb_opt, (size_t)kStride * height); for (size_t i = 0; i < (size_t)kStride * height; ++i) { src_argb_a[i + off] = (fastrand() & 0xff); src_argb_b[i + off] = (fastrand() & 0xff); } memset(dst_argb_c, 0, (size_t)kStride * height); memset(dst_argb_opt, 0, (size_t)kStride * height); MaskCpuFlags(disable_cpu_flags); ARGBSubtract(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) { ARGBSubtract(src_argb_a + off, kStride, src_argb_b + off, kStride, dst_argb_opt, kStride, width, invert * height); } int max_diff = 0; for (size_t i = 0; i < (size_t)kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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, ARGBSubtract_Any) { int max_diff = TestSubtract(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, ARGBSubtract_Unaligned) { int max_diff = TestSubtract(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, ARGBSubtract_Invert) { int max_diff = TestSubtract(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, ARGBSubtract_Opt) { int max_diff = TestSubtract(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_LE(max_diff, 1); } static int TestSobel(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_a, (size_t)kStride * height + off); align_buffer_page_end(dst_argb_c, (size_t)kStride * height); align_buffer_page_end(dst_argb_opt, (size_t)kStride * height); memset(src_argb_a, 0, (size_t)kStride * height + off); for (size_t i = 0; i < (size_t)kStride * height; ++i) { src_argb_a[i + off] = (fastrand() & 0xff); } memset(dst_argb_c, 0, (size_t)kStride * height); memset(dst_argb_opt, 0, (size_t)kStride * height); MaskCpuFlags(disable_cpu_flags); ARGBSobel(src_argb_a + off, kStride, dst_argb_c, kStride, width, invert * height); MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { ARGBSobel(src_argb_a + off, kStride, dst_argb_opt, kStride, width, invert * height); } int max_diff = 0; for (size_t i = 0; i < (size_t)kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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(dst_argb_c); free_aligned_buffer_page_end(dst_argb_opt); return max_diff; } TEST_F(LibYUVPlanarTest, ARGBSobel_Any) { int max_diff = TestSobel(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobel_Unaligned) { int max_diff = TestSobel(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobel_Invert) { int max_diff = TestSobel(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobel_Opt) { int max_diff = TestSobel(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } static int TestSobelToPlane(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 kSrcBpp = 4; const int kDstBpp = 1; const int kSrcStride = (width * kSrcBpp + 15) & ~15; const int kDstStride = (width * kDstBpp + 15) & ~15; align_buffer_page_end(src_argb_a, kSrcStride * height + off); align_buffer_page_end(dst_argb_c, kDstStride * height); align_buffer_page_end(dst_argb_opt, kDstStride * height); memset(src_argb_a, 0, kSrcStride * height + off); for (int i = 0; i < kSrcStride * height; ++i) { src_argb_a[i + off] = (fastrand() & 0xff); } memset(dst_argb_c, 0, kDstStride * height); memset(dst_argb_opt, 0, kDstStride * height); MaskCpuFlags(disable_cpu_flags); ARGBSobelToPlane(src_argb_a + off, kSrcStride, dst_argb_c, kDstStride, width, invert * height); MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { ARGBSobelToPlane(src_argb_a + off, kSrcStride, dst_argb_opt, kDstStride, width, invert * height); } int max_diff = 0; for (int i = 0; i < kDstStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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(dst_argb_c); free_aligned_buffer_page_end(dst_argb_opt); return max_diff; } TEST_F(LibYUVPlanarTest, ARGBSobelToPlane_Any) { int max_diff = TestSobelToPlane(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobelToPlane_Unaligned) { int max_diff = TestSobelToPlane(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobelToPlane_Invert) { int max_diff = TestSobelToPlane(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobelToPlane_Opt) { int max_diff = TestSobelToPlane(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } static int TestSobelXY(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_a, (size_t)kStride * height + off); align_buffer_page_end(dst_argb_c, (size_t)kStride * height); align_buffer_page_end(dst_argb_opt, (size_t)kStride * height); memset(src_argb_a, 0, (size_t)kStride * height + off); for (size_t i = 0; i < (size_t)kStride * height; ++i) { src_argb_a[i + off] = (fastrand() & 0xff); } memset(dst_argb_c, 0, (size_t)kStride * height); memset(dst_argb_opt, 0, (size_t)kStride * height); MaskCpuFlags(disable_cpu_flags); ARGBSobelXY(src_argb_a + off, kStride, dst_argb_c, kStride, width, invert * height); MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { ARGBSobelXY(src_argb_a + off, kStride, dst_argb_opt, kStride, width, invert * height); } int max_diff = 0; for (size_t i = 0; i < (size_t)kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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(dst_argb_c); free_aligned_buffer_page_end(dst_argb_opt); return max_diff; } TEST_F(LibYUVPlanarTest, ARGBSobelXY_Any) { int max_diff = TestSobelXY(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobelXY_Unaligned) { int max_diff = TestSobelXY(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobelXY_Invert) { int max_diff = TestSobelXY(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBSobelXY_Opt) { int max_diff = TestSobelXY(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } static int TestBlur(int width, int height, int benchmark_iterations, int disable_cpu_flags, int benchmark_cpu_info, int invert, int off, int radius) { if (width < 1) { width = 1; } const int kBpp = 4; const int kStride = width * kBpp; const int kCumsumHeight = radius * 2 + 2; align_buffer_page_end(src_argb_a, (size_t)kStride * height + off); align_buffer_page_end(dst_cumsum, (size_t)width * kCumsumHeight * 16); align_buffer_page_end(dst_argb_c, (size_t)kStride * height); align_buffer_page_end(dst_argb_opt, (size_t)kStride * height); for (size_t i = 0; i < (size_t)kStride * height; ++i) { src_argb_a[i + off] = (fastrand() & 0xff); } memset(dst_argb_c, 0, (size_t)kStride * height); memset(dst_argb_opt, 0, (size_t)kStride * height); MaskCpuFlags(disable_cpu_flags); ARGBBlur(src_argb_a + off, kStride, dst_argb_c, kStride, reinterpret_cast(dst_cumsum), width * 4, width, invert * height, radius); MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { ARGBBlur(src_argb_a + off, kStride, dst_argb_opt, kStride, reinterpret_cast(dst_cumsum), width * 4, width, invert * height, radius); } int max_diff = 0; for (size_t i = 0; i < (size_t)kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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(dst_cumsum); free_aligned_buffer_page_end(dst_argb_c); free_aligned_buffer_page_end(dst_argb_opt); return max_diff; } #if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__) #define DISABLED_ARM(name) name #else #define DISABLED_ARM(name) DISABLED_##name #endif static const int kBlurSize = 55; TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlur_Any)) { int max_diff = TestBlur(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, kBlurSize); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlur_Unaligned)) { int max_diff = TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1, kBlurSize); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlur_Invert)) { int max_diff = TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0, kBlurSize); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlur_Opt)) { int max_diff = TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, kBlurSize); ASSERT_LE(max_diff, 1); } static const int kBlurSmallSize = 1; TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlurSmall_Any)) { int max_diff = TestBlur((benchmark_width_ + 3) / 4 + 1, (benchmark_height_ + 3) / 4, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, kBlurSmallSize); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlurSmall_Unaligned)) { int max_diff = TestBlur((benchmark_width_ + 3) / 4, (benchmark_height_ + 3) / 4, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1, kBlurSmallSize); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlurSmall_Invert)) { int max_diff = TestBlur((benchmark_width_ + 3) / 4, (benchmark_height_ + 3) / 4, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0, kBlurSmallSize); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlurSmall_Opt)) { int max_diff = TestBlur((benchmark_width_ + 3) / 4, (benchmark_height_ + 3) / 4, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, kBlurSmallSize); ASSERT_LE(max_diff, 1); } TEST_F(LibYUVPlanarTest, TestARGBPolynomial) { 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]); memset(orig_pixels, 0, sizeof(orig_pixels)); SIMD_ALIGNED(static const float kWarmifyPolynomial[16]) = { 0.94230f, -3.03300f, -2.92500f, 0.f, // C0 0.584500f, 1.112000f, 1.535000f, 1.f, // C1 x 0.001313f, -0.002503f, -0.004496f, 0.f, // C2 x * x 0.0f, 0.000006965f, 0.000008781f, 0.f, // C3 x * x * x }; // 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 white orig_pixels[3][0] = 255u; orig_pixels[3][1] = 255u; orig_pixels[3][2] = 255u; orig_pixels[3][3] = 255u; // Test color orig_pixels[4][0] = 16u; orig_pixels[4][1] = 64u; orig_pixels[4][2] = 192u; orig_pixels[4][3] = 224u; // Do 16 to test asm version. ARGBPolynomial(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, &kWarmifyPolynomial[0], 16, 1); ASSERT_EQ(235u, dst_pixels_opt[0][0]); ASSERT_EQ(0u, dst_pixels_opt[0][1]); ASSERT_EQ(0u, dst_pixels_opt[0][2]); ASSERT_EQ(128u, dst_pixels_opt[0][3]); ASSERT_EQ(0u, dst_pixels_opt[1][0]); ASSERT_EQ(233u, dst_pixels_opt[1][1]); ASSERT_EQ(0u, dst_pixels_opt[1][2]); ASSERT_EQ(0u, dst_pixels_opt[1][3]); ASSERT_EQ(0u, dst_pixels_opt[2][0]); ASSERT_EQ(0u, dst_pixels_opt[2][1]); ASSERT_EQ(241u, dst_pixels_opt[2][2]); ASSERT_EQ(255u, dst_pixels_opt[2][3]); ASSERT_EQ(235u, dst_pixels_opt[3][0]); ASSERT_EQ(233u, dst_pixels_opt[3][1]); ASSERT_EQ(241u, dst_pixels_opt[3][2]); ASSERT_EQ(255u, dst_pixels_opt[3][3]); ASSERT_EQ(10u, dst_pixels_opt[4][0]); ASSERT_EQ(59u, dst_pixels_opt[4][1]); ASSERT_EQ(188u, dst_pixels_opt[4][2]); ASSERT_EQ(224u, dst_pixels_opt[4][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_); ARGBPolynomial(&orig_pixels[0][0], 0, &dst_pixels_c[0][0], 0, &kWarmifyPolynomial[0], 1280, 1); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_pixels_div1280_; ++i) { ARGBPolynomial(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, &kWarmifyPolynomial[0], 1280, 1); } for (int i = 0; i < 1280; ++i) { ASSERT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); ASSERT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); ASSERT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); ASSERT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]); } } static int TestHalfFloatPlane(int benchmark_width, int benchmark_height, int benchmark_iterations, int disable_cpu_flags, int benchmark_cpu_info, float scale, int mask, int invert, int off) { if (benchmark_width <= 0 || benchmark_height <= 0) { return 0; } const size_t y_plane_size = static_cast(benchmark_width) * static_cast(benchmark_height) * 2; 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); for (size_t i = 0; i < y_plane_size / 2; ++i) { reinterpret_cast(orig_y + off)[i] &= mask; } // Disable all optimizations. MaskCpuFlags(disable_cpu_flags); for (int j = 0; j < benchmark_iterations; j++) { HalfFloatPlane(reinterpret_cast(orig_y + off), benchmark_width * 2, reinterpret_cast(dst_c), benchmark_width * 2, scale, benchmark_width, benchmark_height * invert); } // Enable optimizations. MaskCpuFlags(benchmark_cpu_info); for (int j = 0; j < benchmark_iterations; j++) { HalfFloatPlane(reinterpret_cast(orig_y + off), benchmark_width * 2, reinterpret_cast(dst_opt), benchmark_width * 2, scale, benchmark_width, benchmark_height * invert); } int max_diff = 0; for (size_t i = 0; i < y_plane_size / 2; ++i) { int abs_diff = abs(static_cast(reinterpret_cast(dst_c)[i]) - static_cast(reinterpret_cast(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, TestHalfFloatPlane_16bit_One) { int diff = TestHalfFloatPlane(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f, 65535, +1, 0); ASSERT_LE(diff, 1); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_16bit_Opt) { int diff = TestHalfFloatPlane( benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 65535.0f, 65535, +1, 0); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_12bit_Opt) { int diff = TestHalfFloatPlane( benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 4095.0f, 4095, +1, 0); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_10bit_Opt) { int diff = TestHalfFloatPlane( benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 1023.0f, 1023, +1, 0); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_9bit_Opt) { int diff = TestHalfFloatPlane(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 511.0f, 511, +1, 0); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_Any) { int diff = TestHalfFloatPlane( benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 4096.0f, 4095, +1, 0); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_Unaligned) { int diff = TestHalfFloatPlane( benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 4096.0f, 4095, +1, 2); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_Invert) { int diff = TestHalfFloatPlane( benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 4096.0f, 4095, -1, 0); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_Opt) { int diff = TestHalfFloatPlane( benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 4096.0f, 4095, +1, 0); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_16bit_denormal) { int diff = TestHalfFloatPlane( benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f / 65535.0f, 65535, +1, 0); ASSERT_EQ(0, diff); } #if defined(__arm__) && !defined(__SOFTFP__) 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", "cc"); // Clobber List } static void DisableFlushDenormalToZero(void) { uint32_t cw; asm volatile( "vmrs %0, fpscr \n" "bic %0, %0, #0x1000000 \n" "vmsr fpscr, %0 \n" : "=r"(cw)::"memory", "cc"); // Clobber List } // 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<(dst_opt_mem); float* dst_c = reinterpret_cast(dst_c_mem); MemRandomize(orig_y, y_plane_size); memset(dst_c, 0, y_plane_size * 4); memset(dst_opt, 1, y_plane_size * 4); // Disable all optimizations. MaskCpuFlags(disable_cpu_flags); ByteToFloat(orig_y, dst_c, scale, y_plane_size); // Enable optimizations. MaskCpuFlags(benchmark_cpu_info); for (j = 0; j < benchmark_iterations; j++) { ByteToFloat(orig_y, dst_opt, scale, y_plane_size); } float max_diff = 0; for (i = 0; i < y_plane_size; ++i) { float abs_diff = fabs(dst_c[i] - 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_opt_mem); free_aligned_buffer_page_end(dst_c_mem); return max_diff; } TEST_F(LibYUVPlanarTest, TestByteToFloat) { float diff = TestByteToFloat(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, 1.0f); ASSERT_EQ(0.f, diff); } TEST_F(LibYUVPlanarTest, TestARGBLumaColorTable) { 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]); 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; v += 3; } // 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. ARGBLumaColorTable(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, &lumacolortable[0], 16, 1); ASSERT_EQ(253u, dst_pixels_opt[0][0]); ASSERT_EQ(0u, dst_pixels_opt[0][1]); ASSERT_EQ(0u, dst_pixels_opt[0][2]); ASSERT_EQ(128u, dst_pixels_opt[0][3]); ASSERT_EQ(0u, dst_pixels_opt[1][0]); ASSERT_EQ(253u, dst_pixels_opt[1][1]); ASSERT_EQ(0u, dst_pixels_opt[1][2]); ASSERT_EQ(0u, dst_pixels_opt[1][3]); ASSERT_EQ(0u, dst_pixels_opt[2][0]); ASSERT_EQ(0u, dst_pixels_opt[2][1]); ASSERT_EQ(253u, dst_pixels_opt[2][2]); ASSERT_EQ(255u, dst_pixels_opt[2][3]); ASSERT_EQ(48u, dst_pixels_opt[3][0]); ASSERT_EQ(192u, dst_pixels_opt[3][1]); ASSERT_EQ(64u, dst_pixels_opt[3][2]); ASSERT_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_); ARGBLumaColorTable(&orig_pixels[0][0], 0, &dst_pixels_c[0][0], 0, lumacolortable, 1280, 1); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_pixels_div1280_; ++i) { ARGBLumaColorTable(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, lumacolortable, 1280, 1); } for (int i = 0; i < 1280; ++i) { ASSERT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); ASSERT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); ASSERT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); ASSERT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]); } free_aligned_buffer_page_end(lumacolortable); } static int TestARGBCopyAlpha(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 kStride = width * 4; const int kSize = kStride * height; align_buffer_page_end(orig_pixels, kSize + off); align_buffer_page_end(dst_pixels_opt, kSize); align_buffer_page_end(dst_pixels_c, kSize); MemRandomize(orig_pixels + off, kSize); MemRandomize(dst_pixels_c, kSize); // Copy RGB bytes only, leaving alpha bytes in dst_pixels_opt uninitialized. MSAN_POISON(dst_pixels_opt, kSize); for (int i = 0; i < width * height; ++i) { dst_pixels_opt[i * 4 + 0] = dst_pixels_c[i * 4 + 0]; dst_pixels_opt[i * 4 + 1] = dst_pixels_c[i * 4 + 1]; dst_pixels_opt[i * 4 + 2] = dst_pixels_c[i * 4 + 2]; } MaskCpuFlags(disable_cpu_flags); ARGBCopyAlpha(orig_pixels + off, kStride, dst_pixels_c, kStride, width, invert * height); MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { ARGBCopyAlpha(orig_pixels + off, kStride, dst_pixels_opt, kStride, width, invert * height); } int max_diff = 0; for (int i = 0; i < kSize; ++i) { int abs_diff = abs(static_cast(dst_pixels_c[i]) - static_cast(dst_pixels_opt[i])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); free_aligned_buffer_page_end(orig_pixels); return max_diff; } TEST_F(LibYUVPlanarTest, ARGBCopyAlpha_Any) { int max_diff = TestARGBCopyAlpha(benchmark_width_ - 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBCopyAlpha_Unaligned) { int max_diff = TestARGBCopyAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBCopyAlpha_Invert) { int max_diff = TestARGBCopyAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBCopyAlpha_Opt) { int max_diff = TestARGBCopyAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } static int TestARGBExtractAlpha(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 kPixels = width * height; align_buffer_page_end(src_pixels, kPixels * 4 + off); align_buffer_page_end(dst_pixels_opt, kPixels); align_buffer_page_end(dst_pixels_c, kPixels); MemRandomize(src_pixels + off, kPixels * 4); // Leave dst_pixels_opt uninitialized to verify MSan unpoisons all written // alpha bytes (including masked tail stores). MSAN_POISON(dst_pixels_opt, kPixels); MaskCpuFlags(disable_cpu_flags); ARGBExtractAlpha(src_pixels + off, width * 4, dst_pixels_c, width, width, invert * height); MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { ARGBExtractAlpha(src_pixels + off, width * 4, dst_pixels_opt, width, width, invert * height); } int max_diff = 0; for (int i = 0; i < kPixels; ++i) { int abs_diff = abs(static_cast(dst_pixels_c[i]) - static_cast(dst_pixels_opt[i])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); free_aligned_buffer_page_end(src_pixels); return max_diff; } TEST_F(LibYUVPlanarTest, ARGBExtractAlpha_Any) { int max_diff = TestARGBExtractAlpha(benchmark_width_ - 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBExtractAlpha_Unaligned) { int max_diff = TestARGBExtractAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBExtractAlpha_Invert) { int max_diff = TestARGBExtractAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBExtractAlpha_Opt) { int max_diff = TestARGBExtractAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } static int TestARGBCopyYToAlpha(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 kPixels = width * height; align_buffer_page_end(orig_pixels, kPixels + off); align_buffer_page_end(dst_pixels_opt, kPixels * 4); align_buffer_page_end(dst_pixels_c, kPixels * 4); align_buffer_page_end(dst_uninit_opt, kPixels * 4); MemRandomize(orig_pixels + off, kPixels); MemRandomize(dst_pixels_c, kPixels * 4); // Copy RGB bytes only, leaving alpha bytes in dst_pixels_opt uninitialized. MSAN_POISON(dst_pixels_opt, kPixels * 4); for (int i = 0; i < kPixels; ++i) { dst_pixels_opt[i * 4 + 0] = dst_pixels_c[i * 4 + 0]; dst_pixels_opt[i * 4 + 1] = dst_pixels_c[i * 4 + 1]; dst_pixels_opt[i * 4 + 2] = dst_pixels_c[i * 4 + 2]; } // Also test with completely uninitialized destination buffer for MSan. MSAN_POISON(dst_uninit_opt, kPixels * 4); MaskCpuFlags(disable_cpu_flags); ARGBCopyYToAlpha(orig_pixels + off, width, dst_pixels_c, width * 4, width, invert * height); MaskCpuFlags(benchmark_cpu_info); ARGBCopyYToAlpha(orig_pixels + off, width, dst_uninit_opt, width * 4, width, invert * height); for (int i = 0; i < benchmark_iterations; ++i) { ARGBCopyYToAlpha(orig_pixels + off, width, dst_pixels_opt, width * 4, width, invert * height); } int max_diff = 0; for (int i = 0; i < kPixels * 4; ++i) { int abs_diff = abs(static_cast(dst_pixels_c[i]) - static_cast(dst_pixels_opt[i])); if (abs_diff > max_diff) { max_diff = abs_diff; } } for (int i = 0; i < kPixels; ++i) { int abs_diff = abs(static_cast(dst_pixels_c[i * 4 + 3]) - static_cast(dst_uninit_opt[i * 4 + 3])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(dst_uninit_opt); free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); free_aligned_buffer_page_end(orig_pixels); return max_diff; } TEST_F(LibYUVPlanarTest, ARGBCopyYToAlpha_Any) { int max_diff = TestARGBCopyYToAlpha(benchmark_width_ - 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBCopyYToAlpha_Unaligned) { int max_diff = TestARGBCopyYToAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBCopyYToAlpha_Invert) { int max_diff = TestARGBCopyYToAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBCopyYToAlpha_Opt) { int max_diff = TestARGBCopyYToAlpha(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); ASSERT_EQ(0, max_diff); } static int TestARGBRect(int width, int height, int benchmark_iterations, int disable_cpu_flags, int benchmark_cpu_info, int invert, int off, int bpp) { if (width < 1) { width = 1; } const int kStride = width * bpp; const int kSize = kStride * height; const uint32_t v32 = fastrand() & (bpp == 4 ? 0xffffffff : 0xff); align_buffer_page_end(dst_argb_c, kSize + off); align_buffer_page_end(dst_argb_opt, kSize + off); MemRandomize(dst_argb_c + off, kSize); memcpy(dst_argb_opt + off, dst_argb_c + off, kSize); MaskCpuFlags(disable_cpu_flags); if (bpp == 4) { ARGBRect(dst_argb_c + off, kStride, 0, 0, width, invert * height, v32); } else { SetPlane(dst_argb_c + off, kStride, width, invert * height, v32); } MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { if (bpp == 4) { ARGBRect(dst_argb_opt + off, kStride, 0, 0, width, invert * height, v32); } else { SetPlane(dst_argb_opt + off, kStride, width, invert * height, v32); } } int max_diff = 0; for (int i = 0; i < kStride * height; ++i) { int abs_diff = abs(static_cast(dst_argb_c[i + off]) - static_cast(dst_argb_opt[i + off])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(dst_argb_c); free_aligned_buffer_page_end(dst_argb_opt); return max_diff; } TEST_F(LibYUVPlanarTest, ARGBRect_Any) { int max_diff = TestARGBRect(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 4); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBRect_Unaligned) { int max_diff = TestARGBRect(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1, 4); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBRect_Invert) { int max_diff = TestARGBRect(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0, 4); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBRect_Opt) { int max_diff = TestARGBRect(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 4); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, SetPlane_Any) { int max_diff = TestARGBRect(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, SetPlane_Unaligned) { int max_diff = TestARGBRect(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, SetPlane_Invert) { int max_diff = TestARGBRect(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, SetPlane_Opt) { int max_diff = TestARGBRect(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 1); ASSERT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, MergeUVPlane_Opt) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels_u, kPixels); align_buffer_page_end(src_pixels_v, kPixels); align_buffer_page_end(dst_pixels_opt, kPixels * 2); align_buffer_page_end(dst_pixels_c, kPixels * 2); MemRandomize(src_pixels_u, kPixels); MemRandomize(src_pixels_v, kPixels); MemRandomize(dst_pixels_opt, kPixels * 2); MemRandomize(dst_pixels_c, kPixels * 2); MaskCpuFlags(disable_cpu_flags_); MergeUVPlane(src_pixels_u, benchmark_width_, src_pixels_v, benchmark_width_, dst_pixels_c, benchmark_width_ * 2, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { MergeUVPlane(src_pixels_u, benchmark_width_, src_pixels_v, benchmark_width_, dst_pixels_opt, benchmark_width_ * 2, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels * 2; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(src_pixels_u); free_aligned_buffer_page_end(src_pixels_v); free_aligned_buffer_page_end(dst_pixels_opt); free_aligned_buffer_page_end(dst_pixels_c); } // 16 bit channel split and merge TEST_F(LibYUVPlanarTest, MergeUVPlane_16_Opt) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels_u, kPixels * 2); align_buffer_page_end(src_pixels_v, kPixels * 2); align_buffer_page_end(dst_pixels_opt, kPixels * 2 * 2); align_buffer_page_end(dst_pixels_c, kPixels * 2 * 2); MemRandomize(src_pixels_u, kPixels * 2); MemRandomize(src_pixels_v, kPixels * 2); MemRandomize(dst_pixels_opt, kPixels * 2 * 2); MemRandomize(dst_pixels_c, kPixels * 2 * 2); MaskCpuFlags(disable_cpu_flags_); MergeUVPlane_16((const uint16_t*)src_pixels_u, benchmark_width_, (const uint16_t*)src_pixels_v, benchmark_width_, (uint16_t*)dst_pixels_c, benchmark_width_ * 2, benchmark_width_, benchmark_height_, 12); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { MergeUVPlane_16((const uint16_t*)src_pixels_u, benchmark_width_, (const uint16_t*)src_pixels_v, benchmark_width_, (uint16_t*)dst_pixels_opt, benchmark_width_ * 2, benchmark_width_, benchmark_height_, 12); } for (int i = 0; i < kPixels * 2 * 2; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(src_pixels_u); free_aligned_buffer_page_end(src_pixels_v); free_aligned_buffer_page_end(dst_pixels_opt); free_aligned_buffer_page_end(dst_pixels_c); } TEST_F(LibYUVPlanarTest, SplitUVPlane_Opt) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 2); align_buffer_page_end(dst_pixels_u_c, kPixels); align_buffer_page_end(dst_pixels_v_c, kPixels); align_buffer_page_end(dst_pixels_u_opt, kPixels); align_buffer_page_end(dst_pixels_v_opt, kPixels); MemRandomize(src_pixels, kPixels * 2); MemRandomize(dst_pixels_u_c, kPixels); MemRandomize(dst_pixels_v_c, kPixels); MemRandomize(dst_pixels_u_opt, kPixels); MemRandomize(dst_pixels_v_opt, kPixels); MaskCpuFlags(disable_cpu_flags_); SplitUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_u_c, benchmark_width_, dst_pixels_v_c, benchmark_width_, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { SplitUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_u_opt, benchmark_width_, dst_pixels_v_opt, benchmark_width_, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_u_c[i], dst_pixels_u_opt[i]); ASSERT_EQ(dst_pixels_v_c[i], dst_pixels_v_opt[i]); } free_aligned_buffer_page_end(src_pixels); free_aligned_buffer_page_end(dst_pixels_u_c); free_aligned_buffer_page_end(dst_pixels_v_c); free_aligned_buffer_page_end(dst_pixels_u_opt); free_aligned_buffer_page_end(dst_pixels_v_opt); } // 16 bit channel split TEST_F(LibYUVPlanarTest, SplitUVPlane_16_Opt) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 2 * 2); align_buffer_page_end(dst_pixels_u_c, kPixels * 2); align_buffer_page_end(dst_pixels_v_c, kPixels * 2); align_buffer_page_end(dst_pixels_u_opt, kPixels * 2); align_buffer_page_end(dst_pixels_v_opt, kPixels * 2); MemRandomize(src_pixels, kPixels * 2 * 2); MemRandomize(dst_pixels_u_c, kPixels * 2); MemRandomize(dst_pixels_v_c, kPixels * 2); MemRandomize(dst_pixels_u_opt, kPixels * 2); MemRandomize(dst_pixels_v_opt, kPixels * 2); MaskCpuFlags(disable_cpu_flags_); SplitUVPlane_16((const uint16_t*)src_pixels, benchmark_width_ * 2, (uint16_t*)dst_pixels_u_c, benchmark_width_, (uint16_t*)dst_pixels_v_c, benchmark_width_, benchmark_width_, benchmark_height_, 10); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { SplitUVPlane_16((const uint16_t*)src_pixels, benchmark_width_ * 2, (uint16_t*)dst_pixels_u_opt, benchmark_width_, (uint16_t*)dst_pixels_v_opt, benchmark_width_, benchmark_width_, benchmark_height_, 10); } for (int i = 0; i < kPixels * 2; ++i) { ASSERT_EQ(dst_pixels_u_c[i], dst_pixels_u_opt[i]); ASSERT_EQ(dst_pixels_v_c[i], dst_pixels_v_opt[i]); } free_aligned_buffer_page_end(src_pixels); free_aligned_buffer_page_end(dst_pixels_u_c); free_aligned_buffer_page_end(dst_pixels_v_c); free_aligned_buffer_page_end(dst_pixels_u_opt); free_aligned_buffer_page_end(dst_pixels_v_opt); } TEST_F(LibYUVPlanarTest, SwapUVPlane_Opt) { // Round count up to multiple of 16 const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 2); align_buffer_page_end(dst_pixels_opt, kPixels * 2); align_buffer_page_end(dst_pixels_c, kPixels * 2); MemRandomize(src_pixels, kPixels * 2); MemRandomize(dst_pixels_opt, kPixels * 2); MemRandomize(dst_pixels_c, kPixels * 2); MaskCpuFlags(disable_cpu_flags_); SwapUVPlane(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) { SwapUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_opt, benchmark_width_ * 2, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels * 2; ++i) { ASSERT_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, MergeRGBPlane_Opt) { // Round count up to multiple of 16 const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 3); align_buffer_page_end(tmp_pixels_c_r, kPixels); align_buffer_page_end(tmp_pixels_opt_r, kPixels); align_buffer_page_end(tmp_pixels_c_g, kPixels); align_buffer_page_end(tmp_pixels_opt_g, kPixels); align_buffer_page_end(tmp_pixels_c_b, kPixels); align_buffer_page_end(tmp_pixels_opt_b, kPixels); align_buffer_page_end(dst_pixels_c, kPixels * 3); align_buffer_page_end(dst_pixels_opt, kPixels * 3); MemRandomize(src_pixels, kPixels * 3); MemRandomize(tmp_pixels_c_r, kPixels); MemRandomize(tmp_pixels_opt_r, kPixels); MemRandomize(tmp_pixels_c_g, kPixels); MemRandomize(tmp_pixels_opt_g, kPixels); MemRandomize(tmp_pixels_c_b, kPixels); MemRandomize(tmp_pixels_opt_b, kPixels); MemRandomize(dst_pixels_c, kPixels * 3); MemRandomize(dst_pixels_opt, kPixels * 3); MaskCpuFlags(disable_cpu_flags_); SplitRGBPlane(src_pixels, benchmark_width_ * 3, tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, benchmark_width_, benchmark_height_); MergeRGBPlane(tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, dst_pixels_c, benchmark_width_ * 3, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); SplitRGBPlane(src_pixels, benchmark_width_ * 3, tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, benchmark_width_, benchmark_height_); for (int i = 0; i < benchmark_iterations_; ++i) { MergeRGBPlane(tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, dst_pixels_opt, benchmark_width_ * 3, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels * 3; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(src_pixels); free_aligned_buffer_page_end(tmp_pixels_c_r); free_aligned_buffer_page_end(tmp_pixels_opt_r); free_aligned_buffer_page_end(tmp_pixels_c_g); free_aligned_buffer_page_end(tmp_pixels_opt_g); free_aligned_buffer_page_end(tmp_pixels_c_b); free_aligned_buffer_page_end(tmp_pixels_opt_b); free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); } TEST_F(LibYUVPlanarTest, SplitRGBPlane_Opt) { // Round count up to multiple of 16 const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 3); align_buffer_page_end(tmp_pixels_c_r, kPixels); align_buffer_page_end(tmp_pixels_opt_r, kPixels); align_buffer_page_end(tmp_pixels_c_g, kPixels); align_buffer_page_end(tmp_pixels_opt_g, kPixels); align_buffer_page_end(tmp_pixels_c_b, kPixels); align_buffer_page_end(tmp_pixels_opt_b, kPixels); align_buffer_page_end(dst_pixels_c, kPixels * 3); align_buffer_page_end(dst_pixels_opt, kPixels * 3); MemRandomize(src_pixels, kPixels * 3); MemRandomize(tmp_pixels_c_r, kPixels); MemRandomize(tmp_pixels_opt_r, kPixels); MemRandomize(tmp_pixels_c_g, kPixels); MemRandomize(tmp_pixels_opt_g, kPixels); MemRandomize(tmp_pixels_c_b, kPixels); MemRandomize(tmp_pixels_opt_b, kPixels); MemRandomize(dst_pixels_c, kPixels * 3); MemRandomize(dst_pixels_opt, kPixels * 3); MaskCpuFlags(disable_cpu_flags_); SplitRGBPlane(src_pixels, benchmark_width_ * 3, tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, benchmark_width_, benchmark_height_); MergeRGBPlane(tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, dst_pixels_c, benchmark_width_ * 3, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { SplitRGBPlane(src_pixels, benchmark_width_ * 3, tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, benchmark_width_, benchmark_height_); } MergeRGBPlane(tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, dst_pixels_opt, benchmark_width_ * 3, benchmark_width_, benchmark_height_); for (int i = 0; i < kPixels * 3; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(src_pixels); free_aligned_buffer_page_end(tmp_pixels_c_r); free_aligned_buffer_page_end(tmp_pixels_opt_r); free_aligned_buffer_page_end(tmp_pixels_c_g); free_aligned_buffer_page_end(tmp_pixels_opt_g); free_aligned_buffer_page_end(tmp_pixels_c_b); free_aligned_buffer_page_end(tmp_pixels_opt_b); free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); } // Call SplitRGBPlane() with a width that is not a multiple of 32. Verify there // is no stack buffer overflow in SplitRGBRow_Any_AVX2(). TEST_F(LibYUVPlanarTest, SplitRGBPlaneOverflow) { int width = 1024 + 31; // 1055 const int height = 2; align_buffer_page_end(src_rgb, width * 3 * height); align_buffer_page_end(dst_r, width * height); align_buffer_page_end(dst_g, width * height); align_buffer_page_end(dst_b, width * height); memset(src_rgb, 0x5A, width * 3 * height); SplitRGBPlane(src_rgb, width * 3, dst_r, width, dst_g, width, dst_b, width, width, height); free_aligned_buffer_page_end(src_rgb); free_aligned_buffer_page_end(dst_r); free_aligned_buffer_page_end(dst_g); free_aligned_buffer_page_end(dst_b); } TEST_F(LibYUVPlanarTest, MergeARGBPlane_Opt) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 4); align_buffer_page_end(tmp_pixels_c_r, kPixels); align_buffer_page_end(tmp_pixels_opt_r, kPixels); align_buffer_page_end(tmp_pixels_c_g, kPixels); align_buffer_page_end(tmp_pixels_opt_g, kPixels); align_buffer_page_end(tmp_pixels_c_b, kPixels); align_buffer_page_end(tmp_pixels_opt_b, kPixels); align_buffer_page_end(tmp_pixels_c_a, kPixels); align_buffer_page_end(tmp_pixels_opt_a, kPixels); align_buffer_page_end(dst_pixels_c, kPixels * 4); align_buffer_page_end(dst_pixels_opt, kPixels * 4); MemRandomize(src_pixels, kPixels * 4); MemRandomize(tmp_pixels_c_r, kPixels); MemRandomize(tmp_pixels_opt_r, kPixels); MemRandomize(tmp_pixels_c_g, kPixels); MemRandomize(tmp_pixels_opt_g, kPixels); MemRandomize(tmp_pixels_c_b, kPixels); MemRandomize(tmp_pixels_opt_b, kPixels); MemRandomize(tmp_pixels_c_a, kPixels); MemRandomize(tmp_pixels_opt_a, kPixels); MemRandomize(dst_pixels_opt, kPixels * 4); MemRandomize(dst_pixels_c, kPixels * 4); MaskCpuFlags(disable_cpu_flags_); SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, tmp_pixels_c_a, benchmark_width_, benchmark_width_, benchmark_height_); MergeARGBPlane(tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, tmp_pixels_c_a, benchmark_width_, dst_pixels_c, benchmark_width_ * 4, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, tmp_pixels_opt_a, benchmark_width_, benchmark_width_, benchmark_height_); for (int i = 0; i < benchmark_iterations_; ++i) { MergeARGBPlane(tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, tmp_pixels_opt_a, benchmark_width_, dst_pixels_opt, benchmark_width_ * 4, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels * 4; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(src_pixels); free_aligned_buffer_page_end(tmp_pixels_c_r); free_aligned_buffer_page_end(tmp_pixels_opt_r); free_aligned_buffer_page_end(tmp_pixels_c_g); free_aligned_buffer_page_end(tmp_pixels_opt_g); free_aligned_buffer_page_end(tmp_pixels_c_b); free_aligned_buffer_page_end(tmp_pixels_opt_b); free_aligned_buffer_page_end(tmp_pixels_c_a); free_aligned_buffer_page_end(tmp_pixels_opt_a); free_aligned_buffer_page_end(dst_pixels_opt); free_aligned_buffer_page_end(dst_pixels_c); } TEST_F(LibYUVPlanarTest, SplitARGBPlane_Opt) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 4); align_buffer_page_end(tmp_pixels_c_r, kPixels); align_buffer_page_end(tmp_pixels_opt_r, kPixels); align_buffer_page_end(tmp_pixels_c_g, kPixels); align_buffer_page_end(tmp_pixels_opt_g, kPixels); align_buffer_page_end(tmp_pixels_c_b, kPixels); align_buffer_page_end(tmp_pixels_opt_b, kPixels); align_buffer_page_end(tmp_pixels_c_a, kPixels); align_buffer_page_end(tmp_pixels_opt_a, kPixels); align_buffer_page_end(dst_pixels_opt, kPixels * 4); align_buffer_page_end(dst_pixels_c, kPixels * 4); MemRandomize(src_pixels, kPixels * 4); MemRandomize(tmp_pixels_c_r, kPixels); MemRandomize(tmp_pixels_opt_r, kPixels); MemRandomize(tmp_pixels_c_g, kPixels); MemRandomize(tmp_pixels_opt_g, kPixels); MemRandomize(tmp_pixels_c_b, kPixels); MemRandomize(tmp_pixels_opt_b, kPixels); MemRandomize(tmp_pixels_c_a, kPixels); MemRandomize(tmp_pixels_opt_a, kPixels); MemRandomize(dst_pixels_opt, kPixels * 4); MemRandomize(dst_pixels_c, kPixels * 4); MaskCpuFlags(disable_cpu_flags_); SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, tmp_pixels_c_a, benchmark_width_, benchmark_width_, benchmark_height_); MergeARGBPlane(tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, tmp_pixels_c_a, benchmark_width_, dst_pixels_c, benchmark_width_ * 4, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, tmp_pixels_opt_a, benchmark_width_, benchmark_width_, benchmark_height_); } MergeARGBPlane(tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, tmp_pixels_opt_a, benchmark_width_, dst_pixels_opt, benchmark_width_ * 4, benchmark_width_, benchmark_height_); for (int i = 0; i < kPixels * 4; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(src_pixels); free_aligned_buffer_page_end(tmp_pixels_c_r); free_aligned_buffer_page_end(tmp_pixels_opt_r); free_aligned_buffer_page_end(tmp_pixels_c_g); free_aligned_buffer_page_end(tmp_pixels_opt_g); free_aligned_buffer_page_end(tmp_pixels_c_b); free_aligned_buffer_page_end(tmp_pixels_opt_b); free_aligned_buffer_page_end(tmp_pixels_c_a); free_aligned_buffer_page_end(tmp_pixels_opt_a); free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); } // Call SplitARGBPlane() with a width that is not a multiple of 16. Verify there // is no stack buffer overflow in SplitXRGBRow_Any_AVX2(). TEST_F(LibYUVPlanarTest, SplitARGBPlaneOverflow) { int width = 1024 + 15; // 1039 int height = 2; align_buffer_page_end(src_argb, width * 4 * height); align_buffer_page_end(dst_r, width * height); align_buffer_page_end(dst_g, width * height); align_buffer_page_end(dst_b, width * height); memset(src_argb, 0x5A, width * 4 * height); // dst_a == NULL selects SplitXRGBRow SplitARGBPlane(src_argb, width * 4, dst_r, width, dst_g, width, dst_b, width, NULL, 0, width, height); free_aligned_buffer_page_end(src_argb); free_aligned_buffer_page_end(dst_r); free_aligned_buffer_page_end(dst_g); free_aligned_buffer_page_end(dst_b); } TEST_F(LibYUVPlanarTest, MergeXRGBPlane_Opt) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 4); align_buffer_page_end(tmp_pixels_c_r, kPixels); align_buffer_page_end(tmp_pixels_opt_r, kPixels); align_buffer_page_end(tmp_pixels_c_g, kPixels); align_buffer_page_end(tmp_pixels_opt_g, kPixels); align_buffer_page_end(tmp_pixels_c_b, kPixels); align_buffer_page_end(tmp_pixels_opt_b, kPixels); align_buffer_page_end(dst_pixels_c, kPixels * 4); align_buffer_page_end(dst_pixels_opt, kPixels * 4); MemRandomize(src_pixels, kPixels * 4); MemRandomize(tmp_pixels_c_r, kPixels); MemRandomize(tmp_pixels_opt_r, kPixels); MemRandomize(tmp_pixels_c_g, kPixels); MemRandomize(tmp_pixels_opt_g, kPixels); MemRandomize(tmp_pixels_c_b, kPixels); MemRandomize(tmp_pixels_opt_b, kPixels); MemRandomize(dst_pixels_c, kPixels * 4); MemRandomize(dst_pixels_opt, kPixels * 4); MaskCpuFlags(disable_cpu_flags_); SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, NULL, 0, benchmark_width_, benchmark_height_); MergeARGBPlane(tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, NULL, 0, dst_pixels_c, benchmark_width_ * 4, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, NULL, 0, benchmark_width_, benchmark_height_); for (int i = 0; i < benchmark_iterations_; ++i) { MergeARGBPlane(tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, NULL, 0, dst_pixels_opt, benchmark_width_ * 4, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels * 4; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(src_pixels); free_aligned_buffer_page_end(tmp_pixels_c_r); free_aligned_buffer_page_end(tmp_pixels_opt_r); free_aligned_buffer_page_end(tmp_pixels_c_g); free_aligned_buffer_page_end(tmp_pixels_opt_g); free_aligned_buffer_page_end(tmp_pixels_c_b); free_aligned_buffer_page_end(tmp_pixels_opt_b); free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); } TEST_F(LibYUVPlanarTest, SplitXRGBPlane_Opt) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels, kPixels * 4); align_buffer_page_end(tmp_pixels_c_r, kPixels); align_buffer_page_end(tmp_pixels_opt_r, kPixels); align_buffer_page_end(tmp_pixels_c_g, kPixels); align_buffer_page_end(tmp_pixels_opt_g, kPixels); align_buffer_page_end(tmp_pixels_c_b, kPixels); align_buffer_page_end(tmp_pixels_opt_b, kPixels); align_buffer_page_end(dst_pixels_c, kPixels * 4); align_buffer_page_end(dst_pixels_opt, kPixels * 4); MemRandomize(src_pixels, kPixels * 4); MemRandomize(tmp_pixels_c_r, kPixels); MemRandomize(tmp_pixels_opt_r, kPixels); MemRandomize(tmp_pixels_c_g, kPixels); MemRandomize(tmp_pixels_opt_g, kPixels); MemRandomize(tmp_pixels_c_b, kPixels); MemRandomize(tmp_pixels_opt_b, kPixels); MemRandomize(dst_pixels_c, kPixels * 4); MemRandomize(dst_pixels_opt, kPixels * 4); MaskCpuFlags(disable_cpu_flags_); SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, NULL, 0, benchmark_width_, benchmark_height_); MergeARGBPlane(tmp_pixels_c_r, benchmark_width_, tmp_pixels_c_g, benchmark_width_, tmp_pixels_c_b, benchmark_width_, NULL, 0, dst_pixels_c, benchmark_width_ * 4, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, NULL, 0, benchmark_width_, benchmark_height_); } MergeARGBPlane(tmp_pixels_opt_r, benchmark_width_, tmp_pixels_opt_g, benchmark_width_, tmp_pixels_opt_b, benchmark_width_, NULL, 0, dst_pixels_opt, benchmark_width_ * 4, benchmark_width_, benchmark_height_); for (int i = 0; i < kPixels * 4; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(src_pixels); free_aligned_buffer_page_end(tmp_pixels_c_r); free_aligned_buffer_page_end(tmp_pixels_opt_r); free_aligned_buffer_page_end(tmp_pixels_c_g); free_aligned_buffer_page_end(tmp_pixels_opt_g); free_aligned_buffer_page_end(tmp_pixels_c_b); free_aligned_buffer_page_end(tmp_pixels_opt_b); free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); } // Merge 4 channels #define TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, W1280, N, NEG, OFF) \ TEST_F(LibYUVPlanarTest, FUNC##Plane_##DEPTH##N) { \ const int kWidth = W1280; \ const int kPixels = kWidth * benchmark_height_; \ align_buffer_page_end(src_memory_r, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(src_memory_g, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(src_memory_b, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(src_memory_a, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(dst_memory_c, kPixels * 4 * sizeof(DTYPE)); \ align_buffer_page_end(dst_memory_opt, kPixels * 4 * sizeof(DTYPE)); \ MemRandomize(src_memory_r, kPixels * sizeof(STYPE) + OFF); \ MemRandomize(src_memory_g, kPixels * sizeof(STYPE) + OFF); \ MemRandomize(src_memory_b, kPixels * sizeof(STYPE) + OFF); \ MemRandomize(src_memory_a, kPixels * sizeof(STYPE) + OFF); \ memset(dst_memory_c, 0, kPixels * 4 * sizeof(DTYPE)); \ memset(dst_memory_opt, 0, kPixels * 4 * sizeof(DTYPE)); \ STYPE* src_pixels_r = reinterpret_cast(src_memory_r + OFF); \ STYPE* src_pixels_g = reinterpret_cast(src_memory_g + OFF); \ STYPE* src_pixels_b = reinterpret_cast(src_memory_b + OFF); \ STYPE* src_pixels_a = reinterpret_cast(src_memory_a + OFF); \ DTYPE* dst_pixels_c = reinterpret_cast(dst_memory_c); \ DTYPE* dst_pixels_opt = reinterpret_cast(dst_memory_opt); \ MaskCpuFlags(disable_cpu_flags_); \ FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \ kWidth, src_pixels_a, kWidth, dst_pixels_c, kWidth * 4, \ kWidth, NEG benchmark_height_, DEPTH); \ MaskCpuFlags(benchmark_cpu_info_); \ for (int i = 0; i < benchmark_iterations_; ++i) { \ FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \ kWidth, src_pixels_a, kWidth, dst_pixels_opt, kWidth * 4, \ kWidth, NEG benchmark_height_, DEPTH); \ } \ for (int i = 0; i < kPixels * 4; ++i) { \ ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); \ } \ free_aligned_buffer_page_end(src_memory_r); \ free_aligned_buffer_page_end(src_memory_g); \ free_aligned_buffer_page_end(src_memory_b); \ free_aligned_buffer_page_end(src_memory_a); \ free_aligned_buffer_page_end(dst_memory_c); \ free_aligned_buffer_page_end(dst_memory_opt); \ } // Merge 3 channel RGB into 4 channel XRGB with opaque alpha #define TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, W1280, N, NEG, OFF) \ TEST_F(LibYUVPlanarTest, FUNC##Plane_Opaque_##DEPTH##N) { \ const int kWidth = W1280; \ const int kPixels = kWidth * benchmark_height_; \ align_buffer_page_end(src_memory_r, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(src_memory_g, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(src_memory_b, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(dst_memory_c, kPixels * 4 * sizeof(DTYPE)); \ align_buffer_page_end(dst_memory_opt, kPixels * 4 * sizeof(DTYPE)); \ MemRandomize(src_memory_r, kPixels * sizeof(STYPE) + OFF); \ MemRandomize(src_memory_g, kPixels * sizeof(STYPE) + OFF); \ MemRandomize(src_memory_b, kPixels * sizeof(STYPE) + OFF); \ memset(dst_memory_c, 0, kPixels * 4 * sizeof(DTYPE)); \ memset(dst_memory_opt, 0, kPixels * 4 * sizeof(DTYPE)); \ STYPE* src_pixels_r = reinterpret_cast(src_memory_r + OFF); \ STYPE* src_pixels_g = reinterpret_cast(src_memory_g + OFF); \ STYPE* src_pixels_b = reinterpret_cast(src_memory_b + OFF); \ DTYPE* dst_pixels_c = reinterpret_cast(dst_memory_c); \ DTYPE* dst_pixels_opt = reinterpret_cast(dst_memory_opt); \ MaskCpuFlags(disable_cpu_flags_); \ FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \ kWidth, NULL, 0, dst_pixels_c, kWidth * 4, kWidth, \ NEG benchmark_height_, DEPTH); \ MaskCpuFlags(benchmark_cpu_info_); \ for (int i = 0; i < benchmark_iterations_; ++i) { \ FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \ kWidth, NULL, 0, dst_pixels_opt, kWidth * 4, kWidth, \ NEG benchmark_height_, DEPTH); \ } \ for (int i = 0; i < kPixels * 4; ++i) { \ ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); \ } \ free_aligned_buffer_page_end(src_memory_r); \ free_aligned_buffer_page_end(src_memory_g); \ free_aligned_buffer_page_end(src_memory_b); \ free_aligned_buffer_page_end(dst_memory_c); \ free_aligned_buffer_page_end(dst_memory_opt); \ } #define TESTQPLANARTOP(FUNC, STYPE, DTYPE, DEPTH) \ TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_ + 1, _Any, +, 0) \ TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Unaligned, +, \ 2) \ TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Invert, -, 0) \ TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Opt, +, 0) \ TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_ + 1, _Any, +, \ 0) \ TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Unaligned, +, \ 2) \ TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Invert, -, 0) \ TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Opt, +, 0) TESTQPLANARTOP(MergeAR64, uint16_t, uint16_t, 10) TESTQPLANARTOP(MergeAR64, uint16_t, uint16_t, 12) TESTQPLANARTOP(MergeAR64, uint16_t, uint16_t, 16) TESTQPLANARTOP(MergeARGB16To8, uint16_t, uint8_t, 10) TESTQPLANARTOP(MergeARGB16To8, uint16_t, uint8_t, 12) TESTQPLANARTOP(MergeARGB16To8, uint16_t, uint8_t, 16) #define TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, W1280, N, NEG, OFF) \ TEST_F(LibYUVPlanarTest, FUNC##Plane_##DEPTH##N) { \ const int kWidth = W1280; \ const int kPixels = kWidth * benchmark_height_; \ align_buffer_page_end(src_memory_r, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(src_memory_g, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(src_memory_b, kPixels * sizeof(STYPE) + OFF); \ align_buffer_page_end(dst_memory_c, kPixels * 4 * sizeof(DTYPE)); \ align_buffer_page_end(dst_memory_opt, kPixels * 4 * sizeof(DTYPE)); \ MemRandomize(src_memory_r, kPixels * sizeof(STYPE) + OFF); \ MemRandomize(src_memory_g, kPixels * sizeof(STYPE) + OFF); \ MemRandomize(src_memory_b, kPixels * sizeof(STYPE) + OFF); \ STYPE* src_pixels_r = reinterpret_cast(src_memory_r + OFF); \ STYPE* src_pixels_g = reinterpret_cast(src_memory_g + OFF); \ STYPE* src_pixels_b = reinterpret_cast(src_memory_b + OFF); \ DTYPE* dst_pixels_c = reinterpret_cast(dst_memory_c); \ DTYPE* dst_pixels_opt = reinterpret_cast(dst_memory_opt); \ memset(dst_pixels_c, 1, kPixels * 4 * sizeof(DTYPE)); \ memset(dst_pixels_opt, 2, kPixels * 4 * sizeof(DTYPE)); \ MaskCpuFlags(disable_cpu_flags_); \ FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \ kWidth, dst_pixels_c, kWidth * 4, kWidth, \ NEG benchmark_height_, DEPTH); \ MaskCpuFlags(benchmark_cpu_info_); \ for (int i = 0; i < benchmark_iterations_; ++i) { \ FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \ kWidth, dst_pixels_opt, kWidth * 4, kWidth, \ NEG benchmark_height_, DEPTH); \ } \ for (int i = 0; i < kPixels * 4; ++i) { \ ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); \ } \ free_aligned_buffer_page_end(src_memory_r); \ free_aligned_buffer_page_end(src_memory_g); \ free_aligned_buffer_page_end(src_memory_b); \ free_aligned_buffer_page_end(dst_memory_c); \ free_aligned_buffer_page_end(dst_memory_opt); \ } #define TESTTPLANARTOP(FUNC, STYPE, DTYPE, DEPTH) \ TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_ + 1, _Any, +, 0) \ TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Unaligned, +, \ 2) \ TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Invert, -, 0) \ TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Opt, +, 0) TESTTPLANARTOP(MergeXR30, uint16_t, uint8_t, 10) TESTTPLANARTOP(MergeXR30, uint16_t, uint8_t, 12) TESTTPLANARTOP(MergeXR30, uint16_t, uint8_t, 16) // TODO(fbarchard): improve test for platforms and cpu detect #if defined(HAS_MERGEUVROW_16_AVX512BW) || defined(HAS_MERGEUVROW_16_AVX2) TEST_F(LibYUVPlanarTest, MergeUVRow_16_Opt) { // Round count up to multiple of 16 const int kPixels = (benchmark_width_ * benchmark_height_ + 15) & ~15; align_buffer_page_end(src_pixels_u, kPixels * 2); align_buffer_page_end(src_pixels_v, kPixels * 2); align_buffer_page_end(dst_pixels_uv_opt, kPixels * 2 * 2); align_buffer_page_end(dst_pixels_uv_c, kPixels * 2 * 2); MemRandomize(src_pixels_u, kPixels * 2); MemRandomize(src_pixels_v, kPixels * 2); memset(dst_pixels_uv_opt, 0, kPixels * 2 * 2); memset(dst_pixels_uv_c, 1, kPixels * 2 * 2); MergeUVRow_16_C(reinterpret_cast(src_pixels_u), reinterpret_cast(src_pixels_v), reinterpret_cast(dst_pixels_uv_c), 16, kPixels); #if defined(HAS_MERGEUVROW_16_AVX512BW) int has_avx512 = TestCpuFlag(kCpuHasAVX512BW); #endif #if defined(HAS_MERGEUVROW_16_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); #endif for (int i = 0; i < benchmark_iterations_; ++i) { #if defined(HAS_MERGEUVROW_16_AVX512BW) if (has_avx512) { MergeUVRow_16_AVX512BW(reinterpret_cast(src_pixels_u), reinterpret_cast(src_pixels_v), reinterpret_cast(dst_pixels_uv_opt), 16, kPixels); } else #endif #if defined(HAS_MERGEUVROW_16_AVX2) if (has_avx2) { MergeUVRow_16_AVX2(reinterpret_cast(src_pixels_u), reinterpret_cast(src_pixels_v), reinterpret_cast(dst_pixels_uv_opt), 16, kPixels); } else #endif { MergeUVRow_16_C(reinterpret_cast(src_pixels_u), reinterpret_cast(src_pixels_v), reinterpret_cast(dst_pixels_uv_opt), 16, kPixels); } } for (int i = 0; i < kPixels * 2 * 2; ++i) { ASSERT_EQ(dst_pixels_uv_opt[i], dst_pixels_uv_c[i]); } free_aligned_buffer_page_end(src_pixels_u); free_aligned_buffer_page_end(src_pixels_v); free_aligned_buffer_page_end(dst_pixels_uv_opt); free_aligned_buffer_page_end(dst_pixels_uv_c); } #endif #if defined(HAS_MULTIPLYROW_16_AVX512BW) || \ defined(HAS_MULTIPLYROW_16_AVX2) || defined(HAS_MULTIPLYROW_16_NEON) || \ defined(HAS_MULTIPLYROW_16_SME) || defined(HAS_MULTIPLYROW_16_RVV) || \ defined(HAS_MULTIPLYROW_16_HVX) TEST_F(LibYUVPlanarTest, MultiplyRow_16_Opt) { // Round count up to multiple of 64 const int kPixels = (benchmark_width_ * benchmark_height_ + 63) & ~63; align_buffer_page_end(src_pixels_y, kPixels * 2); align_buffer_page_end(dst_pixels_y_opt, kPixels * 2); align_buffer_page_end(dst_pixels_y_c, kPixels * 2); MemRandomize(src_pixels_y, kPixels * 2); memset(dst_pixels_y_opt, 0, kPixels * 2); memset(dst_pixels_y_c, 1, kPixels * 2); MultiplyRow_16_C(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_c), 64, kPixels); #if defined(HAS_MULTIPLYROW_16_AVX512BW) int has_avx512 = TestCpuFlag(kCpuHasAVX512BW); #endif #if defined(HAS_MULTIPLYROW_16_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); #endif #if defined(HAS_MULTIPLYROW_16_NEON) int has_neon = TestCpuFlag(kCpuHasNEON); #endif #if defined(HAS_MULTIPLYROW_16_SME) int has_sme = TestCpuFlag(kCpuHasSME); #endif #if defined(HAS_MULTIPLYROW_16_RVV) int has_rvv = TestCpuFlag(kCpuHasRVV); #endif #if defined(HAS_MULTIPLYROW_16_HVX) int has_hvx = TestCpuFlag(kCpuHasHVX); #endif for (int i = 0; i < benchmark_iterations_; ++i) { #if defined(HAS_MULTIPLYROW_16_AVX512BW) if (has_avx512) { MultiplyRow_16_AVX512BW(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, kPixels); } else #endif #if defined(HAS_MULTIPLYROW_16_AVX2) if (has_avx2) { MultiplyRow_16_AVX2(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, kPixels); } else #endif #if defined(HAS_MULTIPLYROW_16_SME) if (has_sme) { MultiplyRow_16_SME(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, kPixels); } else #endif #if defined(HAS_MULTIPLYROW_16_NEON) if (has_neon) { MultiplyRow_16_NEON(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, kPixels); } else #endif #if defined(HAS_MULTIPLYROW_16_RVV) if (has_rvv) { MultiplyRow_16_RVV(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, kPixels); } else #endif #if defined(HAS_MULTIPLYROW_16_HVX) if (has_hvx) { MultiplyRow_16_HVX(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, kPixels); } else #endif { MultiplyRow_16_C(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, kPixels); } } for (int i = 0; i < kPixels * 2; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } #if defined(HAS_MULTIPLYROW_16_AVX512BW) || \ defined(HAS_MULTIPLYROW_16_AVX2) || defined(HAS_MULTIPLYROW_16_NEON) TEST_F(LibYUVPlanarTest, MultiplyRow_16_Any) { const int kMaxPixels = 256; align_buffer_page_end(src_pixels_y, kMaxPixels * 2); align_buffer_page_end(dst_pixels_y_opt, kMaxPixels * 2); align_buffer_page_end(dst_pixels_y_c, kMaxPixels * 2); for (int i = 0; i < kMaxPixels; ++i) { reinterpret_cast(src_pixels_y)[i] = fastrand() & 1023; } #if defined(HAS_MULTIPLYROW_16_AVX512BW) int has_avx512 = TestCpuFlag(kCpuHasAVX512BW); #endif #if defined(HAS_MULTIPLYROW_16_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); #endif #if defined(HAS_MULTIPLYROW_16_NEON) int has_neon = TestCpuFlag(kCpuHasNEON); #endif for (int width = 1; width <= 129; ++width) { memset(dst_pixels_y_opt, 0, kMaxPixels * 2); memset(dst_pixels_y_c, 1, kMaxPixels * 2); MultiplyRow_16_C(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_c), 64, width); #if defined(HAS_MULTIPLYROW_16_AVX512BW) if (has_avx512) { MultiplyRow_16_AVX512BW(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, width); } else #endif #if defined(HAS_MULTIPLYROW_16_AVX2) if (has_avx2) { MultiplyRow_16_Any_AVX2(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, width); } else #endif #if defined(HAS_MULTIPLYROW_16_NEON) if (has_neon) { MultiplyRow_16_Any_NEON(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, width); } else #endif { MultiplyRow_16_C(reinterpret_cast(src_pixels_y), reinterpret_cast(dst_pixels_y_opt), 64, width); } for (int i = 0; i < width * 2; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } #endif // HAS_MULTIPLYROW_16_AVX512BW || HAS_MULTIPLYROW_16_AVX2 || // HAS_MULTIPLYROW_16_NEON #endif // HAS_MULTIPLYROW_16_... TEST_F(LibYUVPlanarTest, Convert16To8Plane) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels_y, kPixels * 2); align_buffer_page_end(dst_pixels_y_opt, kPixels); align_buffer_page_end(dst_pixels_y_c, kPixels); MemRandomize(src_pixels_y, kPixels * 2); memset(dst_pixels_y_opt, 0, kPixels); memset(dst_pixels_y_c, 1, kPixels); MaskCpuFlags(disable_cpu_flags_); Convert16To8Plane(reinterpret_cast(src_pixels_y), benchmark_width_, dst_pixels_y_c, benchmark_width_, 16384, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { Convert16To8Plane(reinterpret_cast(src_pixels_y), benchmark_width_, dst_pixels_y_opt, benchmark_width_, 16384, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } TEST_F(LibYUVPlanarTest, Convert8To8Plane) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels_y, kPixels); align_buffer_page_end(dst_pixels_y_opt, kPixels); align_buffer_page_end(dst_pixels_y_c, kPixels); MemRandomize(src_pixels_y, kPixels); memset(dst_pixels_y_opt, 0, kPixels); memset(dst_pixels_y_c, 1, kPixels); MaskCpuFlags(disable_cpu_flags_); Convert8To8Plane(src_pixels_y, benchmark_width_, dst_pixels_y_c, benchmark_width_, 220, 16, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { Convert8To8Plane(src_pixels_y, benchmark_width_, dst_pixels_y_opt, benchmark_width_, 220, 16, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } TEST_F(LibYUVPlanarTest, YUY2ToY) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels_y, kPixels * 2); align_buffer_page_end(dst_pixels_y_opt, kPixels); align_buffer_page_end(dst_pixels_y_c, kPixels); MemRandomize(src_pixels_y, kPixels * 2); memset(dst_pixels_y_opt, 0, kPixels); memset(dst_pixels_y_c, 1, kPixels); MaskCpuFlags(disable_cpu_flags_); YUY2ToY(src_pixels_y, benchmark_width_ * 2, dst_pixels_y_c, benchmark_width_, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { YUY2ToY(src_pixels_y, benchmark_width_ * 2, dst_pixels_y_opt, benchmark_width_, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } TEST_F(LibYUVPlanarTest, UYVYToY) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels_y, kPixels * 2); align_buffer_page_end(dst_pixels_y_opt, kPixels); align_buffer_page_end(dst_pixels_y_c, kPixels); MemRandomize(src_pixels_y, kPixels * 2); memset(dst_pixels_y_opt, 0, kPixels); memset(dst_pixels_y_c, 1, kPixels); MaskCpuFlags(disable_cpu_flags_); UYVYToY(src_pixels_y, benchmark_width_ * 2, dst_pixels_y_c, benchmark_width_, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { UYVYToY(src_pixels_y, benchmark_width_ * 2, dst_pixels_y_opt, benchmark_width_, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } #ifdef ENABLE_ROW_TESTS // TODO(fbarchard): Improve test for more platforms. #if defined(HAS_CONVERT16TO8ROW_AVX512BW) || defined(HAS_CONVERT16TO8ROW_AVX2) TEST_F(LibYUVPlanarTest, Convert16To8Row_Opt) { // AVX512 does multiple of 64, so round count up const int kPixels = (benchmark_width_ * benchmark_height_ + 63) & ~63; align_buffer_page_end(src_pixels_y, kPixels * 2); align_buffer_page_end(dst_pixels_y_opt, kPixels); align_buffer_page_end(dst_pixels_y_c, kPixels); MemRandomize(src_pixels_y, kPixels * 2); // clamp source range to 10 bits. for (int i = 0; i < kPixels; ++i) { reinterpret_cast(src_pixels_y)[i] &= 1023; } memset(dst_pixels_y_opt, 0, kPixels); memset(dst_pixels_y_c, 1, kPixels); Convert16To8Row_C(reinterpret_cast(src_pixels_y), dst_pixels_y_c, 16384, kPixels); #if defined(HAS_CONVERT16TO8ROW_AVX512BW) int has_avx512 = TestCpuFlag(kCpuHasAVX512BW); #endif #if defined(HAS_CONVERT16TO8ROW_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); #endif for (int i = 0; i < benchmark_iterations_; ++i) { #if defined(HAS_CONVERT16TO8ROW_AVX512BW) if (has_avx512) { Convert16To8Row_AVX512BW(reinterpret_cast(src_pixels_y), dst_pixels_y_opt, 16384, kPixels); } else #endif #if defined(HAS_CONVERT16TO8ROW_AVX2) if (has_avx2) { Convert16To8Row_AVX2(reinterpret_cast(src_pixels_y), dst_pixels_y_opt, 16384, kPixels); } else #endif { Convert16To8Row_C(reinterpret_cast(src_pixels_y), dst_pixels_y_opt, 16384, kPixels); } } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } #endif // HAS_CONVERT16TO8ROW_AVX512BW || HAS_CONVERT16TO8ROW_AVX2 #if defined(HAS_HALFWIDTHROW_16TO8_AVX2) || \ defined(HAS_HALFWIDTHROW_16TO8_SSSE3) || \ defined(HAS_HALFWIDTHROW_16TO8_AVX512BW) || \ defined(HAS_HALFWIDTHROW_16TO8_NEON) TEST_F(LibYUVPlanarTest, HalfWidthRow_16To8_Opt) { const int kPixels = (benchmark_width_ * benchmark_height_ + 63) & ~63; align_buffer_page_end_16(src_pixels_uv, kPixels * 4); align_buffer_page_end(dst_pixels_uv_opt, kPixels); align_buffer_page_end(dst_pixels_uv_c, kPixels); for (int i = 0; i < kPixels * 4; ++i) { src_pixels_uv[i] = fastrand() & 1023; } memset(dst_pixels_uv_opt, 0, kPixels); memset(dst_pixels_uv_c, 1, kPixels); HalfWidthRow_16To8_C(src_pixels_uv, kPixels * 2, dst_pixels_uv_c, 16384, kPixels); #if defined(HAS_HALFWIDTHROW_16TO8_AVX512BW) int has_avx512 = TestCpuFlag(kCpuHasAVX512BW); #endif #if defined(HAS_HALFWIDTHROW_16TO8_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); #endif #if defined(HAS_HALFWIDTHROW_16TO8_SSSE3) int has_ssse3 = TestCpuFlag(kCpuHasSSSE3); #endif #if defined(HAS_HALFWIDTHROW_16TO8_NEON) int has_neon = TestCpuFlag(kCpuHasNEON); #endif for (int i = 0; i < benchmark_iterations_; ++i) { #if defined(HAS_HALFWIDTHROW_16TO8_AVX512BW) if (has_avx512) { HalfWidthRow_16To8_AVX512BW(src_pixels_uv, kPixels * 2, dst_pixels_uv_opt, 16384, kPixels); } else #endif #if defined(HAS_HALFWIDTHROW_16TO8_AVX2) if (has_avx2) { HalfWidthRow_16To8_AVX2(src_pixels_uv, kPixels * 2, dst_pixels_uv_opt, 16384, kPixels); } else #endif #if defined(HAS_HALFWIDTHROW_16TO8_SSSE3) if (has_ssse3) { HalfWidthRow_16To8_SSSE3(src_pixels_uv, kPixels * 2, dst_pixels_uv_opt, 16384, kPixels); } else #endif #if defined(HAS_HALFWIDTHROW_16TO8_NEON) if (has_neon) { HalfWidthRow_16To8_NEON(src_pixels_uv, kPixels * 2, dst_pixels_uv_opt, 16384, kPixels); } else #endif { HalfWidthRow_16To8_C(src_pixels_uv, kPixels * 2, dst_pixels_uv_opt, 16384, kPixels); } } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_uv_opt[i], dst_pixels_uv_c[i]); } free_aligned_buffer_page_end_16(src_pixels_uv); free_aligned_buffer_page_end(dst_pixels_uv_opt); free_aligned_buffer_page_end(dst_pixels_uv_c); } TEST_F(LibYUVPlanarTest, HalfWidthRow_16To8_Any) { const int kMaxPixels = 256; align_buffer_page_end_16(src_pixels_uv, kMaxPixels * 4); align_buffer_page_end(dst_pixels_uv_opt, kMaxPixels); align_buffer_page_end(dst_pixels_uv_c, kMaxPixels); for (int i = 0; i < kMaxPixels * 4; ++i) { src_pixels_uv[i] = fastrand() & 1023; } #if defined(HAS_HALFWIDTHROW_16TO8_AVX512BW) int has_avx512 = TestCpuFlag(kCpuHasAVX512BW); #endif #if defined(HAS_HALFWIDTHROW_16TO8_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); #endif #if defined(HAS_HALFWIDTHROW_16TO8_SSSE3) int has_ssse3 = TestCpuFlag(kCpuHasSSSE3); #endif #if defined(HAS_HALFWIDTHROW_16TO8_NEON) int has_neon = TestCpuFlag(kCpuHasNEON); #endif for (int width = 1; width <= 129; ++width) { memset(dst_pixels_uv_opt, 0, kMaxPixels); memset(dst_pixels_uv_c, 1, kMaxPixels); HalfWidthRow_16To8_C(src_pixels_uv, kMaxPixels * 2, dst_pixels_uv_c, 16384, width); #if defined(HAS_HALFWIDTHROW_16TO8_AVX512BW) if (has_avx512) { HalfWidthRow_16To8_AVX512BW(src_pixels_uv, kMaxPixels * 2, dst_pixels_uv_opt, 16384, width); } else #endif #if defined(HAS_HALFWIDTHROW_16TO8_AVX2) if (has_avx2) { HalfWidthRow_16To8_Any_AVX2(src_pixels_uv, kMaxPixels * 2, dst_pixels_uv_opt, 16384, width); } else #endif #if defined(HAS_HALFWIDTHROW_16TO8_SSSE3) if (has_ssse3) { HalfWidthRow_16To8_Any_SSSE3(src_pixels_uv, kMaxPixels * 2, dst_pixels_uv_opt, 16384, width); } else #endif #if defined(HAS_HALFWIDTHROW_16TO8_NEON) if (has_neon) { HalfWidthRow_16To8_Any_NEON(src_pixels_uv, kMaxPixels * 2, dst_pixels_uv_opt, 16384, width); } else #endif { HalfWidthRow_16To8_C(src_pixels_uv, kMaxPixels * 2, dst_pixels_uv_opt, 16384, width); } for (int i = 0; i < width; ++i) { ASSERT_EQ(dst_pixels_uv_opt[i], dst_pixels_uv_c[i]); } } free_aligned_buffer_page_end_16(src_pixels_uv); free_aligned_buffer_page_end(dst_pixels_uv_opt); free_aligned_buffer_page_end(dst_pixels_uv_c); } #endif #if defined(HAS_UYVYTOYROW_AVX512BW) || defined(HAS_UYVYTOYROW_AVX2) || \ defined(HAS_UYVYTOYROW_NEON) TEST_F(LibYUVPlanarTest, UYVYToYRow_Opt) { // AVX2/AVX512 does multiple of 32, so round count up const int kPixels = (benchmark_width_ * benchmark_height_ + 31) & ~31; align_buffer_page_end(src_pixels_y, kPixels * 2); align_buffer_page_end(dst_pixels_y_opt, kPixels); align_buffer_page_end(dst_pixels_y_c, kPixels); MemRandomize(src_pixels_y, kPixels * 2); memset(dst_pixels_y_opt, 0, kPixels); memset(dst_pixels_y_c, 1, kPixels); UYVYToYRow_C(src_pixels_y, dst_pixels_y_c, kPixels); #if defined(HAS_UYVYTOYROW_AVX512BW) int has_avx512 = TestCpuFlag(kCpuHasAVX512BW); #endif #if defined(HAS_UYVYTOYROW_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); #endif #if defined(HAS_UYVYTOYROW_NEON) int has_neon = TestCpuFlag(kCpuHasNEON); #endif for (int i = 0; i < benchmark_iterations_; ++i) { #if defined(HAS_UYVYTOYROW_AVX512BW) if (has_avx512) { UYVYToYRow_AVX512BW(src_pixels_y, dst_pixels_y_opt, kPixels); } else #endif #if defined(HAS_UYVYTOYROW_AVX2) if (has_avx2) { UYVYToYRow_AVX2(src_pixels_y, dst_pixels_y_opt, kPixels); } else #endif #if defined(HAS_UYVYTOYROW_NEON) if (has_neon) { UYVYToYRow_NEON(src_pixels_y, dst_pixels_y_opt, kPixels); } else #endif { UYVYToYRow_C(src_pixels_y, dst_pixels_y_opt, kPixels); } } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } #endif // HAS_UYVYTOYROW_... #if defined(HAS_RGB565TOYROW_AVX2) && defined(HAS_ARGB1555TOYROW_AVX2) && \ defined(HAS_ARGB4444TOYROW_AVX2) TEST_F(LibYUVPlanarTest, RGB16ToYUVRow_Opt) { if (!TestCpuFlag(kCpuHasAVX2)) { return; } const int kMaxWidth = 128; SIMD_ALIGNED(uint8_t src0[kMaxWidth * 2]); SIMD_ALIGNED(uint8_t src1[kMaxWidth * 2]); SIMD_ALIGNED(uint8_t dst_y_c[kMaxWidth]); SIMD_ALIGNED(uint8_t dst_y_opt[kMaxWidth]); SIMD_ALIGNED(uint8_t dst_u_c[kMaxWidth]); SIMD_ALIGNED(uint8_t dst_v_c[kMaxWidth]); SIMD_ALIGNED(uint8_t dst_u_opt[kMaxWidth]); SIMD_ALIGNED(uint8_t dst_v_opt[kMaxWidth]); MemRandomize(src0, sizeof(src0)); MemRandomize(src1, sizeof(src1)); for (int width : {1, 2, 7, 15, 16, 17, 31, 32, 33, 64, 127, 128}) { const int halfwidth = (width + 1) / 2; const int stride = kMaxWidth * 2; RGB565ToYRow_C(src0, dst_y_c, width); if (IS_ALIGNED(width, 32)) { RGB565ToYRow_AVX2(src0, dst_y_opt, width); } else { RGB565ToYRow_Any_AVX2(src0, dst_y_opt, width); } for (int i = 0; i < width; ++i) { ASSERT_NEAR(dst_y_c[i], dst_y_opt[i], 1); } ARGB1555ToYRow_C(src0, dst_y_c, width); if (IS_ALIGNED(width, 32)) { ARGB1555ToYRow_AVX2(src0, dst_y_opt, width); } else { ARGB1555ToYRow_Any_AVX2(src0, dst_y_opt, width); } for (int i = 0; i < width; ++i) { ASSERT_NEAR(dst_y_c[i], dst_y_opt[i], 1); } ARGB4444ToYRow_C(src0, dst_y_c, width); if (IS_ALIGNED(width, 32)) { ARGB4444ToYRow_AVX2(src0, dst_y_opt, width); } else { ARGB4444ToYRow_Any_AVX2(src0, dst_y_opt, width); } for (int i = 0; i < width; ++i) { ASSERT_NEAR(dst_y_c[i], dst_y_opt[i], 1); } SIMD_ALIGNED(uint8_t src2rows[kMaxWidth * 2 * 2]); memcpy(src2rows, src0, stride); memcpy(src2rows + stride, src1, stride); RGB565ToUVRow_C(src2rows, stride, dst_u_c, dst_v_c, width); if (IS_ALIGNED(width, 32)) { RGB565ToUVRow_AVX2(src2rows, stride, dst_u_opt, dst_v_opt, width); } else { RGB565ToUVRow_Any_AVX2(src2rows, stride, dst_u_opt, dst_v_opt, width); } for (int i = 0; i < halfwidth; ++i) { ASSERT_NEAR(dst_u_c[i], dst_u_opt[i], 2); ASSERT_NEAR(dst_v_c[i], dst_v_opt[i], 2); } ARGB1555ToUVRow_C(src2rows, stride, dst_u_c, dst_v_c, width); if (IS_ALIGNED(width, 32)) { ARGB1555ToUVRow_AVX2(src2rows, stride, dst_u_opt, dst_v_opt, width); } else { ARGB1555ToUVRow_Any_AVX2(src2rows, stride, dst_u_opt, dst_v_opt, width); } for (int i = 0; i < halfwidth; ++i) { ASSERT_NEAR(dst_u_c[i], dst_u_opt[i], 2); ASSERT_NEAR(dst_v_c[i], dst_v_opt[i], 2); } ARGB4444ToUVRow_C(src2rows, stride, dst_u_c, dst_v_c, width); if (IS_ALIGNED(width, 32)) { ARGB4444ToUVRow_AVX2(src2rows, stride, dst_u_opt, dst_v_opt, width); } else { ARGB4444ToUVRow_Any_AVX2(src2rows, stride, dst_u_opt, dst_v_opt, width); } for (int i = 0; i < halfwidth; ++i) { ASSERT_NEAR(dst_u_c[i], dst_u_opt[i], 2); ASSERT_NEAR(dst_v_c[i], dst_v_opt[i], 2); } } } #endif #endif // ENABLE_ROW_TESTS TEST_F(LibYUVPlanarTest, Convert8To16Plane) { const int kPixels = benchmark_width_ * benchmark_height_; align_buffer_page_end(src_pixels_y, kPixels); align_buffer_page_end(dst_pixels_y_opt, kPixels * 2); align_buffer_page_end(dst_pixels_y_c, kPixels * 2); MemRandomize(src_pixels_y, kPixels); memset(dst_pixels_y_opt, 0, kPixels * 2); memset(dst_pixels_y_c, 1, kPixels * 2); MaskCpuFlags(disable_cpu_flags_); Convert8To16Plane(src_pixels_y, benchmark_width_, reinterpret_cast(dst_pixels_y_c), benchmark_width_, 10, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { Convert8To16Plane(src_pixels_y, benchmark_width_, reinterpret_cast(dst_pixels_y_opt), benchmark_width_, 1024, benchmark_width_, benchmark_height_); } for (int i = 0; i < kPixels * 2; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } #ifdef ENABLE_ROW_TESTS // TODO(fbarchard): Improve test for more platforms. #if defined(HAS_CONVERT8TO16ROW_AVX512BW) || defined(HAS_CONVERT8TO16ROW_AVX2) TEST_F(LibYUVPlanarTest, Convert8To16Row_Opt) { const int kPixels = (benchmark_width_ * benchmark_height_ + 63) & ~63; align_buffer_page_end(src_pixels_y, kPixels); align_buffer_page_end(dst_pixels_y_opt, kPixels * 2); align_buffer_page_end(dst_pixels_y_c, kPixels * 2); MemRandomize(src_pixels_y, kPixels); memset(dst_pixels_y_opt, 0, kPixels * 2); memset(dst_pixels_y_c, 1, kPixels * 2); Convert8To16Row_C(src_pixels_y, reinterpret_cast(dst_pixels_y_c), 10, kPixels); #if defined(HAS_CONVERT8TO16ROW_AVX512BW) int has_avx512 = TestCpuFlag(kCpuHasAVX512BW); #endif int has_avx2 = TestCpuFlag(kCpuHasAVX2); for (int i = 0; i < benchmark_iterations_; ++i) { #if defined(HAS_CONVERT8TO16ROW_AVX512BW) if (has_avx512) { Convert8To16Row_AVX512BW(src_pixels_y, reinterpret_cast(dst_pixels_y_opt), 10, kPixels); } else #endif if (has_avx2) { Convert8To16Row_AVX2(src_pixels_y, reinterpret_cast(dst_pixels_y_opt), 10, kPixels); } else { Convert8To16Row_C(src_pixels_y, reinterpret_cast(dst_pixels_y_opt), 10, kPixels); } } for (int i = 0; i < kPixels * 2; ++i) { ASSERT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]); } free_aligned_buffer_page_end(src_pixels_y); free_aligned_buffer_page_end(dst_pixels_y_opt); free_aligned_buffer_page_end(dst_pixels_y_c); } #endif // HAS_CONVERT8TO16ROW_AVX512BW || HAS_CONVERT8TO16ROW_AVX2 float TestScaleMaxSamples(int benchmark_width, int benchmark_height, int benchmark_iterations, float scale, bool opt) { int i, j; float max_c, max_opt = 0.f; // NEON does multiple of 8, so round count up const int kPixels = (benchmark_width * benchmark_height + 7) & ~7; align_buffer_page_end(orig_y, kPixels * 4 * 3 + 48); uint8_t* dst_c = orig_y + kPixels * 4 + 16; uint8_t* dst_opt = orig_y + kPixels * 4 * 2 + 32; // Randomize works but may contain some denormals affecting performance. // MemRandomize(orig_y, kPixels * 4); // large values are problematic. audio is really -1 to 1. for (i = 0; i < kPixels; ++i) { (reinterpret_cast(orig_y))[i] = sinf(static_cast(i) * 0.1f); } memset(dst_c, 0, kPixels * 4); memset(dst_opt, 1, kPixels * 4); max_c = ScaleMaxSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_c), scale, kPixels); for (j = 0; j < benchmark_iterations; j++) { if (opt) { #ifdef HAS_SCALESUMSAMPLES_NEON max_opt = ScaleMaxSamples_NEON(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); #elif defined(HAS_SCALEMAXSAMPLES_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { max_opt = ScaleMaxSamples_AVX2(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } else { max_opt = ScaleMaxSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } #else max_opt = ScaleMaxSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); #endif } else { max_opt = ScaleMaxSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } } float max_diff = FAbs(max_opt - max_c); for (i = 0; i < kPixels; ++i) { float abs_diff = FAbs((reinterpret_cast(dst_c)[i]) - (reinterpret_cast(dst_opt)[i])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(orig_y); return max_diff; } TEST_F(LibYUVPlanarTest, TestScaleMaxSamples_C) { float diff = TestScaleMaxSamples(benchmark_width_, benchmark_height_, benchmark_iterations_, 1.2f, false); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestScaleMaxSamples_Opt) { float diff = TestScaleMaxSamples(benchmark_width_, benchmark_height_, benchmark_iterations_, 1.2f, true); ASSERT_EQ(0, diff); } float TestScaleSumSamples(int benchmark_width, int benchmark_height, int benchmark_iterations, float scale, bool opt) { int i, j; float sum_c, sum_opt = 0.f; // NEON does multiple of 8, so round count up const int kPixels = (benchmark_width * benchmark_height + 7) & ~7; align_buffer_page_end(orig_y, kPixels * 4 * 3); uint8_t* dst_c = orig_y + kPixels * 4; uint8_t* dst_opt = orig_y + kPixels * 4 * 2; // Randomize works but may contain some denormals affecting performance. // MemRandomize(orig_y, kPixels * 4); // large values are problematic. audio is really -1 to 1. for (i = 0; i < kPixels; ++i) { (reinterpret_cast(orig_y))[i] = sinf(static_cast(i) * 0.1f); } memset(dst_c, 0, kPixels * 4); memset(dst_opt, 1, kPixels * 4); sum_c = ScaleSumSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_c), scale, kPixels); for (j = 0; j < benchmark_iterations; j++) { if (opt) { #ifdef HAS_SCALESUMSAMPLES_NEON sum_opt = ScaleSumSamples_NEON(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); #elif defined(HAS_SCALESUMSAMPLES_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { sum_opt = ScaleSumSamples_AVX2(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } else { sum_opt = ScaleSumSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } #else sum_opt = ScaleSumSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); #endif } else { sum_opt = ScaleSumSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } } float mse_opt = sum_opt / kPixels * 4; float mse_c = sum_c / kPixels * 4; float mse_error = FAbs(mse_opt - mse_c) / mse_c; // If the sum of a float is more than 4 million, small adds are round down on // float and produce different results with vectorized sum vs scalar sum. // Ignore the difference if the sum is large. float max_diff = 0.f; if (mse_error > 0.0001 && sum_c < 4000000) { // allow .01% difference of mse max_diff = mse_error; } for (i = 0; i < kPixels; ++i) { float abs_diff = FAbs((reinterpret_cast(dst_c)[i]) - (reinterpret_cast(dst_opt)[i])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(orig_y); return max_diff; } TEST_F(LibYUVPlanarTest, TestScaleSumSamples_C) { float diff = TestScaleSumSamples(benchmark_width_, benchmark_height_, benchmark_iterations_, 1.2f, false); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestScaleSumSamples_Opt) { float diff = TestScaleSumSamples(benchmark_width_, benchmark_height_, benchmark_iterations_, 1.2f, true); ASSERT_EQ(0, diff); } float TestScaleSamples(int benchmark_width, int benchmark_height, int benchmark_iterations, float scale, bool opt) { int i, j; // NEON does multiple of 8, so round count up const int kPixels = (benchmark_width * benchmark_height + 7) & ~7; align_buffer_page_end(orig_y, kPixels * 4 * 3); uint8_t* dst_c = orig_y + kPixels * 4; uint8_t* dst_opt = orig_y + kPixels * 4 * 2; // Randomize works but may contain some denormals affecting performance. // MemRandomize(orig_y, kPixels * 4); // large values are problematic. audio is really -1 to 1. for (i = 0; i < kPixels; ++i) { (reinterpret_cast(orig_y))[i] = sinf(static_cast(i) * 0.1f); } memset(dst_c, 0, kPixels * 4); memset(dst_opt, 1, kPixels * 4); ScaleSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_c), scale, kPixels); for (j = 0; j < benchmark_iterations; j++) { if (opt) { #ifdef HAS_SCALESUMSAMPLES_NEON ScaleSamples_NEON(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); #elif defined(HAS_SCALESAMPLES_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { ScaleSamples_AVX2(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } else { ScaleSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } #else ScaleSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); #endif } else { ScaleSamples_C(reinterpret_cast(orig_y), reinterpret_cast(dst_opt), scale, kPixels); } } float max_diff = 0.f; for (i = 0; i < kPixels; ++i) { float abs_diff = FAbs((reinterpret_cast(dst_c)[i]) - (reinterpret_cast(dst_opt)[i])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(orig_y); return max_diff; } TEST_F(LibYUVPlanarTest, TestScaleSamples_C) { float diff = TestScaleSamples(benchmark_width_, benchmark_height_, benchmark_iterations_, 1.2f, false); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestScaleSamples_Opt) { float diff = TestScaleSamples(benchmark_width_, benchmark_height_, benchmark_iterations_, 1.2f, true); ASSERT_EQ(0, diff); } float TestCopySamples(int benchmark_width, int benchmark_height, int benchmark_iterations, bool opt) { int i, j; // NEON does multiple of 16 floats, so round count up const int kPixels = (benchmark_width * benchmark_height + 15) & ~15; align_buffer_page_end(orig_y, kPixels * 4 * 3); uint8_t* dst_c = orig_y + kPixels * 4; uint8_t* dst_opt = orig_y + kPixels * 4 * 2; // Randomize works but may contain some denormals affecting performance. // MemRandomize(orig_y, kPixels * 4); // large values are problematic. audio is really -1 to 1. for (i = 0; i < kPixels; ++i) { (reinterpret_cast(orig_y))[i] = sinf(static_cast(i) * 0.1f); } memset(dst_c, 0, kPixels * 4); memset(dst_opt, 1, kPixels * 4); memcpy(reinterpret_cast(dst_c), reinterpret_cast(orig_y), kPixels * 4); for (j = 0; j < benchmark_iterations; j++) { if (opt) { #ifdef HAS_COPYROW_NEON CopyRow_NEON(orig_y, dst_opt, kPixels * 4); #else CopyRow_C(orig_y, dst_opt, kPixels * 4); #endif } else { CopyRow_C(orig_y, dst_opt, kPixels * 4); } } float max_diff = 0.f; for (i = 0; i < kPixels; ++i) { float abs_diff = FAbs((reinterpret_cast(dst_c)[i]) - (reinterpret_cast(dst_opt)[i])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(orig_y); return max_diff; } TEST_F(LibYUVPlanarTest, TestCopySamples_C) { float diff = TestCopySamples(benchmark_width_, benchmark_height_, benchmark_iterations_, false); ASSERT_EQ(0, diff); } TEST_F(LibYUVPlanarTest, TestCopySamples_Opt) { float diff = TestCopySamples(benchmark_width_, benchmark_height_, benchmark_iterations_, true); ASSERT_EQ(0, diff); } extern "C" void GaussRow_NEON(const uint32_t* src, uint16_t* dst, int width); extern "C" void GaussRow_C(const uint32_t* src, uint16_t* dst, int width); TEST_F(LibYUVPlanarTest, TestGaussRow_Opt) { SIMD_ALIGNED(uint32_t orig_pixels[1280 + 8]); SIMD_ALIGNED(uint16_t dst_pixels_c[1280]); SIMD_ALIGNED(uint16_t dst_pixels_opt[1280]); memset(orig_pixels, 0, sizeof(orig_pixels)); memset(dst_pixels_c, 1, sizeof(dst_pixels_c)); memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt)); for (int i = 0; i < 1280 + 8; ++i) { orig_pixels[i] = i * 256; } GaussRow_C(&orig_pixels[0], &dst_pixels_c[0], 1280); for (int i = 0; i < benchmark_pixels_div1280_; ++i) { #if defined(HAS_GAUSSROW_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); if (has_avx2) { GaussRow_AVX2(&orig_pixels[0], &dst_pixels_opt[0], 1280); } else { GaussRow_C(&orig_pixels[0], &dst_pixels_opt[0], 1280); } #elif !defined(LIBYUV_DISABLE_NEON) && \ (defined(__aarch64__) || defined(__ARM_NEON__) || defined(LIBYUV_NEON)) int has_neon = TestCpuFlag(kCpuHasNEON); if (has_neon) { GaussRow_NEON(&orig_pixels[0], &dst_pixels_opt[0], 1280); } else { GaussRow_C(&orig_pixels[0], &dst_pixels_opt[0], 1280); } #else GaussRow_C(&orig_pixels[0], &dst_pixels_opt[0], 1280); #endif } for (int i = 0; i < 1280; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } ASSERT_EQ(dst_pixels_c[0], static_cast(0 * 1 + 1 * 4 + 2 * 6 + 3 * 4 + 4 * 1)); ASSERT_EQ(dst_pixels_c[639], static_cast(10256)); } extern "C" void GaussCol_NEON(const uint16_t* src0, const uint16_t* src1, const uint16_t* src2, const uint16_t* src3, const uint16_t* src4, uint32_t* dst, int width); extern "C" void GaussCol_C(const uint16_t* src0, const uint16_t* src1, const uint16_t* src2, const uint16_t* src3, const uint16_t* src4, uint32_t* dst, int width); TEST_F(LibYUVPlanarTest, TestGaussCol_Opt) { SIMD_ALIGNED(uint16_t orig_pixels[1280 * 5]); SIMD_ALIGNED(uint32_t dst_pixels_c[1280]); SIMD_ALIGNED(uint32_t dst_pixels_opt[1280]); memset(orig_pixels, 0, sizeof(orig_pixels)); memset(dst_pixels_c, 1, sizeof(dst_pixels_c)); memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt)); for (int i = 0; i < 1280 * 5; ++i) { orig_pixels[i] = static_cast(i); } GaussCol_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_c[0], 1280); for (int i = 0; i < benchmark_pixels_div1280_; ++i) { #if defined(HAS_GAUSSCOL_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); if (has_avx2) { GaussCol_AVX2(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); } else { GaussCol_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); } #elif !defined(LIBYUV_DISABLE_NEON) && \ (defined(__aarch64__) || defined(__ARM_NEON__) || defined(LIBYUV_NEON)) int has_neon = TestCpuFlag(kCpuHasNEON); if (has_neon) { GaussCol_NEON(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); } else { GaussCol_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); } #else GaussCol_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); #endif } for (int i = 0; i < 1280; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } } TEST_F(LibYUVPlanarTest, TestGaussRow_F32_Opt) { SIMD_ALIGNED(float orig_pixels[1280 + 4]); SIMD_ALIGNED(float dst_pixels_c[1280]); SIMD_ALIGNED(float dst_pixels_opt[1280]); memset(orig_pixels, 0, sizeof(orig_pixels)); memset(dst_pixels_c, 1, sizeof(dst_pixels_c)); memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt)); for (int i = 0; i < 1280 + 4; ++i) { orig_pixels[i] = static_cast(i); } GaussRow_F32_C(&orig_pixels[0], &dst_pixels_c[0], 1280); for (int i = 0; i < benchmark_pixels_div1280_; ++i) { #if defined(HAS_GAUSSROW_F32_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); if (has_avx2) { GaussRow_F32_AVX2(&orig_pixels[0], &dst_pixels_opt[0], 1280); } else { GaussRow_F32_C(&orig_pixels[0], &dst_pixels_opt[0], 1280); } #elif !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__) int has_neon = TestCpuFlag(kCpuHasNEON); if (has_neon) { GaussRow_F32_NEON(&orig_pixels[0], &dst_pixels_opt[0], 1280); } else { GaussRow_F32_C(&orig_pixels[0], &dst_pixels_opt[0], 1280); } #else GaussRow_F32_C(&orig_pixels[0], &dst_pixels_opt[0], 1280); #endif } for (int i = 0; i < 1280; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } } TEST_F(LibYUVPlanarTest, TestGaussCol_F32_Opt) { SIMD_ALIGNED(float dst_pixels_c[1280]); SIMD_ALIGNED(float dst_pixels_opt[1280]); align_buffer_page_end(orig_pixels_buf, 1280 * 5 * 4); // 5 rows float* orig_pixels = reinterpret_cast(orig_pixels_buf); memset(orig_pixels, 0, 1280 * 5 * 4); memset(dst_pixels_c, 1, sizeof(dst_pixels_c)); memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt)); for (int i = 0; i < 1280 * 5; ++i) { orig_pixels[i] = static_cast(i); } GaussCol_F32_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_c[0], 1280); for (int i = 0; i < benchmark_pixels_div1280_; ++i) { #if defined(HAS_GAUSSCOL_F32_AVX2) int has_avx2 = TestCpuFlag(kCpuHasAVX2); if (has_avx2) { GaussCol_F32_AVX2(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); } else { GaussCol_F32_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); } #elif !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__) int has_neon = TestCpuFlag(kCpuHasNEON); if (has_neon) { GaussCol_F32_NEON(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); } else { GaussCol_F32_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); } #else GaussCol_F32_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2], &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280); #endif } for (int i = 0; i < 1280; ++i) { ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); } free_aligned_buffer_page_end(orig_pixels_buf); } TEST_F(LibYUVPlanarTest, SwapUVRow) { const int kPixels = benchmark_width_ * benchmark_height_; void (*SwapUVRow)(const uint8_t* src_uv, uint8_t* dst_vu, int width) = SwapUVRow_C; align_buffer_page_end(src_pixels_vu, kPixels * 2); align_buffer_page_end(dst_pixels_uv, kPixels * 2); MemRandomize(src_pixels_vu, kPixels * 2); memset(dst_pixels_uv, 1, kPixels * 2); #if defined(HAS_SWAPUVROW_SSSE3) if (TestCpuFlag(kCpuHasSSSE3)) { SwapUVRow = SwapUVRow_Any_SSSE3; if (IS_ALIGNED(kPixels, 16)) { SwapUVRow = SwapUVRow_SSSE3; } } #endif #if defined(HAS_SWAPUVROW_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { SwapUVRow = SwapUVRow_Any_AVX2; if (IS_ALIGNED(kPixels, 32)) { SwapUVRow = SwapUVRow_AVX2; } } #endif #if defined(HAS_SWAPUVROW_NEON) if (TestCpuFlag(kCpuHasNEON)) { SwapUVRow = SwapUVRow_Any_NEON; if (IS_ALIGNED(kPixels, 16)) { SwapUVRow = SwapUVRow_NEON; } } #endif #if defined(HAS_SWAPUVROW_RVV) if (TestCpuFlag(kCpuHasRVV)) { SwapUVRow = SwapUVRow_RVV; } #endif #if defined(HAS_SWAPUVROW_HVX) if (TestCpuFlag(kCpuHasHVX)) { SwapUVRow = SwapUVRow_HVX; } #endif for (int j = 0; j < benchmark_iterations_; j++) { SwapUVRow(src_pixels_vu, dst_pixels_uv, kPixels); } for (int i = 0; i < kPixels; ++i) { ASSERT_EQ(dst_pixels_uv[i * 2 + 0], src_pixels_vu[i * 2 + 1]); ASSERT_EQ(dst_pixels_uv[i * 2 + 1], src_pixels_vu[i * 2 + 0]); } free_aligned_buffer_page_end(src_pixels_vu); free_aligned_buffer_page_end(dst_pixels_uv); } TEST_F(LibYUVPlanarTest, HalfRow_Opt) { const int kMaxPixels = 160; align_buffer_page_end(src_pixels, kMaxPixels * 2); align_buffer_page_end(dst_pixels_opt, kMaxPixels); align_buffer_page_end(dst_pixels_c, kMaxPixels); MemRandomize(src_pixels, kMaxPixels * 2); for (ptrdiff_t stride : {ptrdiff_t{0}, ptrdiff_t{kMaxPixels}}) { for (int width = 1; width <= 129; ++width) { memset(dst_pixels_opt, 0, kMaxPixels); memset(dst_pixels_c, 1, kMaxPixels); HalfRow_C(src_pixels, stride, dst_pixels_c, width); #if defined(HAS_HALFROW_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { if (IS_ALIGNED(width, 32)) { HalfRow_AVX2(src_pixels, stride, dst_pixels_opt, width); } else { HalfRow_Any_AVX2(src_pixels, stride, dst_pixels_opt, width); } } else #endif { HalfRow_C(src_pixels, stride, dst_pixels_opt, width); } ASSERT_EQ(0, memcmp(dst_pixels_c, dst_pixels_opt, width)) << "stride=" << stride << " width=" << width; } } 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, HalfRow_16_Opt) { const int kMaxPixels = 160; align_buffer_page_end_16(src_pixels, kMaxPixels * 2); align_buffer_page_end_16(dst_pixels_opt, kMaxPixels); align_buffer_page_end_16(dst_pixels_c, kMaxPixels); for (int i = 0; i < kMaxPixels * 2; ++i) { src_pixels[i] = static_cast(fastrand()); } for (ptrdiff_t stride : {ptrdiff_t{0}, ptrdiff_t{kMaxPixels}}) { for (int width = 1; width <= 129; ++width) { memset(dst_pixels_opt, 0, kMaxPixels * 2); memset(dst_pixels_c, 1, kMaxPixels * 2); HalfRow_16_C(src_pixels, stride, dst_pixels_c, width); #if defined(HAS_HALFROW_16_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { if (IS_ALIGNED(width, 16)) { HalfRow_16_AVX2(src_pixels, stride, dst_pixels_opt, width); } else { HalfRow_16_Any_AVX2(src_pixels, stride, dst_pixels_opt, width); } } else #endif { HalfRow_16_C(src_pixels, stride, dst_pixels_opt, width); } ASSERT_EQ(0, memcmp(dst_pixels_c, dst_pixels_opt, width * 2)) << "stride=" << stride << " width=" << width; } } free_aligned_buffer_page_end_16(src_pixels); free_aligned_buffer_page_end_16(dst_pixels_opt); free_aligned_buffer_page_end_16(dst_pixels_c); } TEST_F(LibYUVPlanarTest, HalfWidthRow_16To8_Odd_Opt) { const int kMaxPixels = 160; align_buffer_page_end_16(src_pixels, kMaxPixels * 4); align_buffer_page_end(dst_pixels_opt, kMaxPixels); align_buffer_page_end(dst_pixels_c, kMaxPixels); for (int i = 0; i < kMaxPixels * 4; ++i) { src_pixels[i] = fastrand() & 1023; } for (ptrdiff_t stride : {ptrdiff_t{0}, ptrdiff_t{kMaxPixels * 2}}) { for (int width = 1; width <= 129; ++width) { memset(dst_pixels_opt, 0, kMaxPixels); memset(dst_pixels_c, 1, kMaxPixels); HalfWidthRow_16To8_Odd_C(src_pixels, stride, dst_pixels_c, 16384, width); #if defined(HAS_HALFWIDTHROW_16TO8_ODD_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { HalfWidthRow_16To8_Odd_AVX2(src_pixels, stride, dst_pixels_opt, 16384, width); } else #endif { HalfWidthRow_16To8_Odd_C(src_pixels, stride, dst_pixels_opt, 16384, width); } ASSERT_EQ(0, memcmp(dst_pixels_c, dst_pixels_opt, width)) << "stride=" << stride << " width=" << width; } } free_aligned_buffer_page_end_16(src_pixels); free_aligned_buffer_page_end(dst_pixels_opt); free_aligned_buffer_page_end(dst_pixels_c); } TEST_F(LibYUVPlanarTest, ARGBChannelReorderRow_C_Test) { const int kPixels = 65; align_buffer_page_end(src_argb, kPixels * 4); align_buffer_page_end(dst_row, kPixels * 4); align_buffer_page_end(dst_api, kPixels * 4); MemRandomize(src_argb, kPixels * 4); ARGBToABGRRow_C(src_argb, dst_row, kPixels); ARGBToABGR(src_argb, kPixels * 4, dst_api, kPixels * 4, kPixels, 1); ASSERT_EQ(0, memcmp(dst_row, dst_api, kPixels * 4)); ARGBToBGRARow_C(src_argb, dst_row, kPixels); ARGBToBGRA(src_argb, kPixels * 4, dst_api, kPixels * 4, kPixels, 1); ASSERT_EQ(0, memcmp(dst_row, dst_api, kPixels * 4)); ARGBToRGBARow_C(src_argb, dst_row, kPixels); ARGBToRGBA(src_argb, kPixels * 4, dst_api, kPixels * 4, kPixels, 1); ASSERT_EQ(0, memcmp(dst_row, dst_api, kPixels * 4)); RGBAToARGBRow_C(src_argb, dst_row, kPixels); RGBAToARGB(src_argb, kPixels * 4, dst_api, kPixels * 4, kPixels, 1); ASSERT_EQ(0, memcmp(dst_row, dst_api, kPixels * 4)); align_buffer_page_end_16(src_ar64, kPixels * 4); align_buffer_page_end_16(dst_ab64_row, kPixels * 4); align_buffer_page_end_16(dst_ab64_api, kPixels * 4); for (int i = 0; i < kPixels * 4; ++i) { src_ar64[i] = static_cast(fastrand()); } AR64ToAB64Row_C(src_ar64, dst_ab64_row, kPixels); AR64ToAB64(src_ar64, kPixels * 4, dst_ab64_api, kPixels * 4, kPixels, 1); ASSERT_EQ(0, memcmp(dst_ab64_row, dst_ab64_api, kPixels * 8)); free_aligned_buffer_page_end_16(src_ar64); free_aligned_buffer_page_end_16(dst_ab64_row); free_aligned_buffer_page_end_16(dst_ab64_api); free_aligned_buffer_page_end(src_argb); free_aligned_buffer_page_end(dst_row); free_aligned_buffer_page_end(dst_api); } #endif // ENABLE_ROW_TESTS TEST_F(LibYUVPlanarTest, TestGaussPlane_F32) { const int kSize = benchmark_width_ * benchmark_height_ * 4; align_buffer_page_end(orig_pixels, kSize); align_buffer_page_end(dst_pixels_opt, kSize); align_buffer_page_end(dst_pixels_c, kSize); for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) { ((float*)(orig_pixels))[i] = (i & 1023) * 3.14f; } memset(dst_pixels_opt, 1, kSize); memset(dst_pixels_c, 2, kSize); MaskCpuFlags(disable_cpu_flags_); GaussPlane_F32((const float*)(orig_pixels), benchmark_width_, (float*)(dst_pixels_c), benchmark_width_, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { GaussPlane_F32((const float*)(orig_pixels), benchmark_width_, (float*)(dst_pixels_opt), benchmark_width_, benchmark_width_, benchmark_height_); } for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) { ASSERT_NEAR(((float*)(dst_pixels_c))[i], ((float*)(dst_pixels_opt))[i], 1.f) << i; } free_aligned_buffer_page_end(dst_pixels_c); free_aligned_buffer_page_end(dst_pixels_opt); free_aligned_buffer_page_end(orig_pixels); } TEST_F(LibYUVPlanarTest, HalfMergeUVPlane_Opt) { int dst_width = (benchmark_width_ + 1) / 2; int dst_height = (benchmark_height_ + 1) / 2; align_buffer_page_end(src_pixels_u, benchmark_width_ * benchmark_height_); align_buffer_page_end(src_pixels_v, benchmark_width_ * benchmark_height_); align_buffer_page_end(dst_pixels_uv_opt, dst_width * 2 * dst_height); align_buffer_page_end(dst_pixels_uv_c, dst_width * 2 * dst_height); MemRandomize(src_pixels_u, benchmark_width_ * benchmark_height_); MemRandomize(src_pixels_v, benchmark_width_ * benchmark_height_); MemRandomize(dst_pixels_uv_opt, dst_width * 2 * dst_height); MemRandomize(dst_pixels_uv_c, dst_width * 2 * dst_height); MaskCpuFlags(disable_cpu_flags_); HalfMergeUVPlane(src_pixels_u, benchmark_width_, src_pixels_v, benchmark_width_, dst_pixels_uv_c, dst_width * 2, benchmark_width_, benchmark_height_); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < benchmark_iterations_; ++i) { HalfMergeUVPlane(src_pixels_u, benchmark_width_, src_pixels_v, benchmark_width_, dst_pixels_uv_opt, dst_width * 2, benchmark_width_, benchmark_height_); } for (int i = 0; i < dst_width * 2 * dst_height; ++i) { ASSERT_EQ(dst_pixels_uv_c[i], dst_pixels_uv_opt[i]); } free_aligned_buffer_page_end(src_pixels_u); free_aligned_buffer_page_end(src_pixels_v); free_aligned_buffer_page_end(dst_pixels_uv_opt); free_aligned_buffer_page_end(dst_pixels_uv_c); } TEST_F(LibYUVPlanarTest, NV12Copy) { const int halfwidth = (benchmark_width_ + 1) >> 1; const int halfheight = (benchmark_height_ + 1) >> 1; align_buffer_page_end(src_y, benchmark_width_ * benchmark_height_); align_buffer_page_end(src_uv, halfwidth * 2 * halfheight); align_buffer_page_end(dst_y, benchmark_width_ * benchmark_height_); align_buffer_page_end(dst_uv, halfwidth * 2 * halfheight); MemRandomize(src_y, benchmark_width_ * benchmark_height_); MemRandomize(src_uv, halfwidth * 2 * halfheight); MemRandomize(dst_y, benchmark_width_ * benchmark_height_); MemRandomize(dst_uv, halfwidth * 2 * halfheight); for (int i = 0; i < benchmark_iterations_; ++i) { NV12Copy(src_y, benchmark_width_, src_uv, halfwidth * 2, dst_y, benchmark_width_, dst_uv, halfwidth * 2, benchmark_width_, benchmark_height_); } for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) { ASSERT_EQ(src_y[i], dst_y[i]); } for (int i = 0; i < halfwidth * 2 * halfheight; ++i) { ASSERT_EQ(src_uv[i], dst_uv[i]); } free_aligned_buffer_page_end(src_y); free_aligned_buffer_page_end(src_uv); free_aligned_buffer_page_end(dst_y); free_aligned_buffer_page_end(dst_uv); } TEST_F(LibYUVPlanarTest, NV21Copy) { const int halfwidth = (benchmark_width_ + 1) >> 1; const int halfheight = (benchmark_height_ + 1) >> 1; align_buffer_page_end(src_y, benchmark_width_ * benchmark_height_); align_buffer_page_end(src_vu, halfwidth * 2 * halfheight); align_buffer_page_end(dst_y, benchmark_width_ * benchmark_height_); align_buffer_page_end(dst_vu, halfwidth * 2 * halfheight); MemRandomize(src_y, benchmark_width_ * benchmark_height_); MemRandomize(src_vu, halfwidth * 2 * halfheight); MemRandomize(dst_y, benchmark_width_ * benchmark_height_); MemRandomize(dst_vu, halfwidth * 2 * halfheight); for (int i = 0; i < benchmark_iterations_; ++i) { NV21Copy(src_y, benchmark_width_, src_vu, halfwidth * 2, dst_y, benchmark_width_, dst_vu, halfwidth * 2, benchmark_width_, benchmark_height_); } for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) { ASSERT_EQ(src_y[i], dst_y[i]); } for (int i = 0; i < halfwidth * 2 * halfheight; ++i) { ASSERT_EQ(src_vu[i], dst_vu[i]); } free_aligned_buffer_page_end(src_y); free_aligned_buffer_page_end(src_vu); free_aligned_buffer_page_end(dst_y); free_aligned_buffer_page_end(dst_vu); } #if defined(ENABLE_ROW_TESTS) && !defined(LIBYUV_DISABLE_NEON) && \ defined(__aarch64__) TEST_F(LibYUVPlanarTest, TestConvertFP16ToFP32) { int i, j; const int y_plane_size = benchmark_width_ * benchmark_height_; align_buffer_page_end(orig_f, y_plane_size * 4); align_buffer_page_end(orig_y, y_plane_size * 2); align_buffer_page_end(dst_opt, y_plane_size * 4); align_buffer_page_end(rec_opt, y_plane_size * 2); for (i = 0; i < y_plane_size; ++i) { ((float*)orig_f)[i] = (float)(i % 10000) * 3.14f; } memset(orig_y, 1, y_plane_size * 2); memset(dst_opt, 2, y_plane_size * 4); memset(rec_opt, 3, y_plane_size * 2); ConvertFP32ToFP16Row_NEON((const float*)orig_f, (uint16_t*)orig_y, y_plane_size); for (j = 0; j < benchmark_iterations_; j++) { ConvertFP16ToFP32Row_NEON((const uint16_t*)orig_y, (float*)dst_opt, y_plane_size); } ConvertFP32ToFP16Row_NEON((const float*)dst_opt, (uint16_t*)rec_opt, y_plane_size); for (i = 0; i < y_plane_size; ++i) { ASSERT_EQ(((const uint16_t*)orig_y)[i], ((const uint16_t*)rec_opt)[i]); } free_aligned_buffer_page_end(orig_f); free_aligned_buffer_page_end(orig_y); free_aligned_buffer_page_end(dst_opt); free_aligned_buffer_page_end(rec_opt); } TEST_F(LibYUVPlanarTest, TestConvertFP16ToFP32Column) { int i, j; const int y_plane_size = benchmark_width_ * benchmark_height_; align_buffer_page_end(orig_f, y_plane_size * 4); align_buffer_page_end(orig_y, y_plane_size * 2); align_buffer_page_end(dst_opt, y_plane_size * 4); align_buffer_page_end(rec_opt, y_plane_size * 2); for (i = 0; i < y_plane_size; ++i) { ((float*)orig_f)[i] = (float)(i % 10000) * 3.14f; } memset(orig_y, 1, y_plane_size * 2); memset(dst_opt, 2, y_plane_size * 4); memset(rec_opt, 3, y_plane_size * 2); ConvertFP32ToFP16Row_NEON((const float*)orig_f, (uint16_t*)orig_y, y_plane_size); for (j = 0; j < benchmark_iterations_; j++) { ConvertFP16ToFP32Column_NEON((const uint16_t*)orig_y, 1, (float*)dst_opt, y_plane_size); } ConvertFP32ToFP16Row_NEON((const float*)dst_opt, (uint16_t*)rec_opt, y_plane_size); for (i = 0; i < y_plane_size; ++i) { ASSERT_EQ(((const uint16_t*)orig_y)[i], ((const uint16_t*)rec_opt)[i]); } free_aligned_buffer_page_end(orig_f); free_aligned_buffer_page_end(orig_y); free_aligned_buffer_page_end(dst_opt); free_aligned_buffer_page_end(rec_opt); } #endif // defined(ENABLE_ROW_TESTS) && defined(__aarch64__) static int TestARGBOr(int width, int height, int benchmark_iterations, int disable_cpu_flags, int benchmark_cpu_info, int invert, int off, uint32_t value) { if (width < 1) { width = 1; } const int kBpp = 4; const int kStride = width * kBpp; align_buffer_page_end(src_argb, kStride * abs(height) + off); align_buffer_page_end(dst_argb_c, kStride * abs(height)); align_buffer_page_end(dst_argb_opt, kStride * abs(height)); MemRandomize(src_argb, kStride * abs(height) + off); memset(dst_argb_c, 0, kStride * abs(height)); memset(dst_argb_opt, 1, kStride * abs(height)); MaskCpuFlags(disable_cpu_flags); ARGBOr(src_argb + off, kStride, dst_argb_c, kStride, width, invert * height, value); MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { ARGBOr(src_argb + off, kStride, dst_argb_opt, kStride, width, invert * height, value); } int max_diff = 0; for (int i = 0; i < kStride * abs(height); ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(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, ARGBOr_Any) { int max_diff = TestARGBOr(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 0xff000000u); EXPECT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBOr_Unaligned) { int max_diff = TestARGBOr(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1, 0xff000000u); EXPECT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBOr_Invert) { int max_diff = TestARGBOr(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0, 0xff000000u); EXPECT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBOr_Opt) { int max_diff = TestARGBOr(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 0xff000000u); EXPECT_EQ(0, max_diff); } static int TestARGBAlphaBlend(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_argb0, kStride * abs(height) + off); align_buffer_page_end(src_argb1, kStride * abs(height) + off); align_buffer_page_end(dst_argb_c, kStride * abs(height)); align_buffer_page_end(dst_argb_opt, kStride * abs(height)); MemRandomize(src_argb0, kStride * abs(height) + off); MemRandomize(src_argb1, kStride * abs(height) + off); memset(dst_argb_c, 0, kStride * abs(height)); memset(dst_argb_opt, 1, kStride * abs(height)); MaskCpuFlags(disable_cpu_flags); ARGBAlphaBlend(src_argb0 + off, kStride, src_argb1 + off, kStride, dst_argb_c, kStride, width, invert * height); MaskCpuFlags(benchmark_cpu_info); for (int i = 0; i < benchmark_iterations; ++i) { ARGBAlphaBlend(src_argb0 + off, kStride, src_argb1 + off, kStride, dst_argb_opt, kStride, width, invert * height); } int max_diff = 0; for (int i = 0; i < kStride * abs(height); ++i) { int abs_diff = abs(static_cast(dst_argb_c[i]) - static_cast(dst_argb_opt[i])); if (abs_diff > max_diff) { max_diff = abs_diff; } } free_aligned_buffer_page_end(src_argb0); free_aligned_buffer_page_end(src_argb1); free_aligned_buffer_page_end(dst_argb_c); free_aligned_buffer_page_end(dst_argb_opt); return max_diff; } TEST_F(LibYUVPlanarTest, ARGBAlphaBlend_Any) { int max_diff = TestARGBAlphaBlend(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); EXPECT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBAlphaBlend_Unaligned) { int max_diff = TestARGBAlphaBlend(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 1); EXPECT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBAlphaBlend_Invert) { int max_diff = TestARGBAlphaBlend(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, -1, 0); EXPECT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBAlphaBlend_Opt) { int max_diff = TestARGBAlphaBlend(benchmark_width_, benchmark_height_, benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_, +1, 0); EXPECT_EQ(0, max_diff); } TEST_F(LibYUVPlanarTest, ARGBAlphaBlend_AllAlphas) { const int kWidth = 256; const int kHeight = 256; const int kStride = kWidth * 4; align_buffer_page_end(src_argb0, kStride * kHeight); align_buffer_page_end(src_argb1, kStride * kHeight); align_buffer_page_end(dst_argb_c, kStride * kHeight); align_buffer_page_end(dst_argb_opt, kStride * kHeight); for (int da = 0; da < 256; ++da) { for (int sa = 0; sa < 256; ++sa) { int idx = (da * 256 + sa) * 4; src_argb0[idx + 0] = (sa * 7 + da * 3 + 13) & 255; src_argb0[idx + 1] = (sa * 13 + da * 5 + 29) & 255; src_argb0[idx + 2] = (sa * 29 + da * 11 + 47) & 255; src_argb0[idx + 3] = sa; src_argb1[idx + 0] = (da * 17 + sa * 19 + 53) & 255; src_argb1[idx + 1] = (da * 23 + sa * 31 + 71) & 255; src_argb1[idx + 2] = (da * 37 + sa * 41 + 89) & 255; src_argb1[idx + 3] = da; } } MaskCpuFlags(disable_cpu_flags_); ARGBAlphaBlend(src_argb0, kStride, src_argb1, kStride, dst_argb_c, kStride, kWidth, kHeight); MaskCpuFlags(benchmark_cpu_info_); ARGBAlphaBlend(src_argb0, kStride, src_argb1, kStride, dst_argb_opt, kStride, kWidth, kHeight); for (int i = 0; i < kStride * kHeight; ++i) { ASSERT_EQ(dst_argb_c[i], dst_argb_opt[i]) << "Mismatch at byte " << i; } free_aligned_buffer_page_end(src_argb0); free_aligned_buffer_page_end(src_argb1); free_aligned_buffer_page_end(dst_argb_c); free_aligned_buffer_page_end(dst_argb_opt); } TEST_F(LibYUVPlanarTest, MSanUninitializedInputsAndTails) { static const int kWidths[] = {1, 7, 15, 16, 17, 31, 32, 33, 63, 64, 65, 99}; const int kMaxW = 128; const int kMaxSrcUV = kMaxW * 3 * 2 + 128; align_buffer_page_end(src0, kMaxW * 4 + 128); align_buffer_page_end(src1, kMaxW * 4 + 128); align_buffer_page_end(src2, kMaxW * 4 + 128); align_buffer_page_end(dst_c0, kMaxW * 4 + 128); align_buffer_page_end(dst_c1, kMaxW * 4 + 128); align_buffer_page_end(dst_c2, kMaxW * 4 + 128); align_buffer_page_end(dst_opt0, kMaxW * 4 + 128); align_buffer_page_end(dst_opt1, kMaxW * 4 + 128); align_buffer_page_end(dst_opt2, kMaxW * 4 + 128); align_buffer_page_end(uv_src, kMaxSrcUV * 3); for (size_t wi = 0; wi < sizeof(kWidths) / sizeof(kWidths[0]); ++wi) { const int w = kWidths[wi]; const int uv_w = (w + 1) / 2; // 1. ARGBToI444 and ARGBToI420 with poisoned RGBX alpha and poisoned tail. for (int i = 0; i < w; ++i) { src0[i * 4 + 0] = static_cast((i * 11 + 3) & 255); src0[i * 4 + 1] = static_cast((i * 17 + 7) & 255); src0[i * 4 + 2] = static_cast((i * 29 + 13) & 255); src0[i * 4 + 3] = 0u; } MaskCpuFlags(disable_cpu_flags_); ARGBToI444(src0, w * 4, dst_c0, w, dst_c1, w, dst_c2, w, w, 1); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < w; ++i) { MSAN_POISON(&src0[i * 4 + 3], 1); } MSAN_POISON(src0 + w * 4, 128); MSAN_POISON(dst_opt0, kMaxW * 4); MSAN_POISON(dst_opt1, kMaxW * 4); MSAN_POISON(dst_opt2, kMaxW * 4); ARGBToI444(src0, w * 4, dst_opt0, w, dst_opt1, w, dst_opt2, w, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w); MSAN_CHECK_INITIALIZED(dst_opt1, w); MSAN_CHECK_INITIALIZED(dst_opt2, w); for (int i = 0; i < w; ++i) { EXPECT_EQ(dst_c0[i], dst_opt0[i]) << "ARGBToI444 Y w=" << w; EXPECT_EQ(dst_c1[i], dst_opt1[i]) << "ARGBToI444 U w=" << w; EXPECT_EQ(dst_c2[i], dst_opt2[i]) << "ARGBToI444 V w=" << w; } for (int i = 0; i < w; ++i) { src0[i * 4 + 3] = 0u; } MaskCpuFlags(disable_cpu_flags_); ARGBToI420(src0, w * 4, dst_c0, w, dst_c1, uv_w, dst_c2, uv_w, w, 1); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < w; ++i) { MSAN_POISON(&src0[i * 4 + 3], 1); } MSAN_POISON(src0 + w * 4, 128); MSAN_POISON(dst_opt0, kMaxW * 4); MSAN_POISON(dst_opt1, kMaxW * 4); MSAN_POISON(dst_opt2, kMaxW * 4); ARGBToI420(src0, w * 4, dst_opt0, w, dst_opt1, uv_w, dst_opt2, uv_w, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w); MSAN_CHECK_INITIALIZED(dst_opt1, uv_w); MSAN_CHECK_INITIALIZED(dst_opt2, uv_w); for (int i = 0; i < w; ++i) { EXPECT_EQ(dst_c0[i], dst_opt0[i]) << "ARGBToI420 Y w=" << w; } for (int i = 0; i < uv_w; ++i) { EXPECT_EQ(dst_c1[i], dst_opt1[i]) << "ARGBToI420 U w=" << w; EXPECT_EQ(dst_c2[i], dst_opt2[i]) << "ARGBToI420 V w=" << w; } // 2. ARGBMirror with poisoned alpha on even pixels (verifying per-lane // shadow permutation) and poisoned tail. for (int i = 0; i < w; ++i) { src0[i * 4 + 0] = static_cast((i * 5 + 1) & 255); src0[i * 4 + 1] = static_cast((i * 9 + 2) & 255); src0[i * 4 + 2] = static_cast((i * 15 + 3) & 255); src0[i * 4 + 3] = static_cast((i * 21 + 4) & 255); } MaskCpuFlags(disable_cpu_flags_); ARGBMirror(src0, w * 4, dst_c0, w * 4, w, 1); MaskCpuFlags(benchmark_cpu_info_); for (int i = 0; i < w; i += 2) { MSAN_POISON(&src0[i * 4 + 3], 1); } MSAN_POISON(src0 + w * 4, 128); MSAN_POISON(dst_opt0, kMaxW * 4); ARGBMirror(src0, w * 4, dst_opt0, w * 4, w, 1); for (int i = 0; i < w; ++i) { const int src_i = w - 1 - i; MSAN_CHECK_INITIALIZED(&dst_opt0[i * 4], 3); EXPECT_EQ(dst_c0[i * 4 + 0], dst_opt0[i * 4 + 0]); EXPECT_EQ(dst_c0[i * 4 + 1], dst_opt0[i * 4 + 1]); EXPECT_EQ(dst_c0[i * 4 + 2], dst_opt0[i * 4 + 2]); if (src_i & 1) { MSAN_CHECK_INITIALIZED(&dst_opt0[i * 4 + 3], 1); EXPECT_EQ(dst_c0[i * 4 + 3], dst_opt0[i * 4 + 3]); } } // 3. RGB24Mirror, RAWToRGBA, RAWToRGB24, MergeRGBPlane with poisoned tails. for (int i = 0; i < w * 3; ++i) { src0[i] = static_cast((i * 19 + 5) & 255); } MSAN_POISON(src0 + w * 3, 128); MaskCpuFlags(disable_cpu_flags_); RGB24Mirror(src0, w * 3, dst_c0, w * 3, w, 1); RAWToRGBA(src0, w * 3, dst_c1, w * 4, w, 1); RAWToRGB24(src0, w * 3, dst_c2, w * 3, w, 1); MaskCpuFlags(benchmark_cpu_info_); MSAN_POISON(dst_opt0, kMaxW * 4); MSAN_POISON(dst_opt1, kMaxW * 4); MSAN_POISON(dst_opt2, kMaxW * 4); RGB24Mirror(src0, w * 3, dst_opt0, w * 3, w, 1); RAWToRGBA(src0, w * 3, dst_opt1, w * 4, w, 1); RAWToRGB24(src0, w * 3, dst_opt2, w * 3, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w * 3); MSAN_CHECK_INITIALIZED(dst_opt1, w * 4); MSAN_CHECK_INITIALIZED(dst_opt2, w * 3); EXPECT_EQ(0, memcmp(dst_c0, dst_opt0, w * 3)); EXPECT_EQ(0, memcmp(dst_c1, dst_opt1, w * 4)); EXPECT_EQ(0, memcmp(dst_c2, dst_opt2, w * 3)); for (int i = 0; i < w; ++i) { src0[i] = static_cast((i * 3 + 11) & 255); src1[i] = static_cast((i * 7 + 23) & 255); src2[i] = static_cast((i * 13 + 37) & 255); } MSAN_POISON(src0 + w, 128); MSAN_POISON(src1 + w, 128); MSAN_POISON(src2 + w, 128); MaskCpuFlags(disable_cpu_flags_); MergeRGBPlane(src0, w, src1, w, src2, w, dst_c0, w * 3, w, 1); MirrorPlane(src0, w, dst_c1, w, w, 1); MaskCpuFlags(benchmark_cpu_info_); MSAN_POISON(dst_opt0, kMaxW * 4); MSAN_POISON(dst_opt1, kMaxW * 4); MergeRGBPlane(src0, w, src1, w, src2, w, dst_opt0, w * 3, w, 1); MirrorPlane(src0, w, dst_opt1, w, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w * 3); MSAN_CHECK_INITIALIZED(dst_opt1, w); EXPECT_EQ(0, memcmp(dst_c0, dst_opt0, w * 3)); EXPECT_EQ(0, memcmp(dst_c1, dst_opt1, w)); // 4. MirrorUVPlane and MirrorPlane_16 with poisoned tails. for (int i = 0; i < w * 2; ++i) { src0[i] = static_cast((i * 23 + 9) & 255); } MSAN_POISON(src0 + w * 2, 128); MaskCpuFlags(disable_cpu_flags_); MirrorUVPlane(src0, w * 2, dst_c0, w * 2, w, 1); MirrorPlane_16(reinterpret_cast(src0), w, reinterpret_cast(dst_c1), w, w, 1); MaskCpuFlags(benchmark_cpu_info_); MSAN_POISON(dst_opt0, kMaxW * 4); MSAN_POISON(dst_opt1, kMaxW * 4); MirrorUVPlane(src0, w * 2, dst_opt0, w * 2, w, 1); MirrorPlane_16(reinterpret_cast(src0), w, reinterpret_cast(dst_opt1), w, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w * 2); MSAN_CHECK_INITIALIZED(dst_opt1, w * 2); EXPECT_EQ(0, memcmp(dst_c0, dst_opt0, w * 2)); EXPECT_EQ(0, memcmp(dst_c1, dst_opt1, w * 2)); // 5. I420ToAR30, NV12ToARGB, NV12ToRGB24, I444ToARGB, I210ToARGB. for (int i = 0; i < w; ++i) { src0[i] = static_cast((i * 5 + 16) & 255); src1[i] = static_cast((i * 9 + 64) & 255); src2[i] = static_cast((i * 11 + 128) & 255); } MSAN_POISON(src0 + w, 128); MSAN_POISON(src1 + uv_w, 128); MSAN_POISON(src2 + uv_w, 128); MaskCpuFlags(disable_cpu_flags_); I420ToAR30(src0, w, src1, uv_w, src2, uv_w, dst_c0, w * 4, w, 1); MaskCpuFlags(benchmark_cpu_info_); MSAN_POISON(dst_opt0, kMaxW * 4); I420ToAR30(src0, w, src1, uv_w, src2, uv_w, dst_opt0, w * 4, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w * 4); EXPECT_EQ(0, memcmp(dst_c0, dst_opt0, w * 4)); for (int i = 0; i < uv_w * 2; ++i) { src1[i] = static_cast((i * 13 + 80) & 255); } MSAN_POISON(src1 + uv_w * 2, 128); MaskCpuFlags(disable_cpu_flags_); NV12ToARGB(src0, w, src1, uv_w * 2, dst_c0, w * 4, w, 1); NV12ToRGB24(src0, w, src1, uv_w * 2, dst_c1, w * 3, w, 1); MaskCpuFlags(benchmark_cpu_info_); MSAN_POISON(dst_opt0, kMaxW * 4); MSAN_POISON(dst_opt1, kMaxW * 4); NV12ToARGB(src0, w, src1, uv_w * 2, dst_opt0, w * 4, w, 1); NV12ToRGB24(src0, w, src1, uv_w * 2, dst_opt1, w * 3, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w * 4); MSAN_CHECK_INITIALIZED(dst_opt1, w * 3); EXPECT_EQ(0, memcmp(dst_c0, dst_opt0, w * 4)); EXPECT_EQ(0, memcmp(dst_c1, dst_opt1, w * 3)); for (int i = 0; i < w; ++i) { reinterpret_cast(src0)[i] = static_cast((i * 17 + 64) & 1023); } for (int i = 0; i < uv_w; ++i) { reinterpret_cast(src1)[i] = static_cast((i * 23 + 256) & 1023); reinterpret_cast(src2)[i] = static_cast((i * 31 + 512) & 1023); } MSAN_POISON(src0 + w * 2, 128); MSAN_POISON(src1 + uv_w * 2, 128); MSAN_POISON(src2 + uv_w * 2, 128); MaskCpuFlags(disable_cpu_flags_); I210ToARGB(reinterpret_cast(src0), w, reinterpret_cast(src1), uv_w, reinterpret_cast(src2), uv_w, dst_c0, w * 4, w, 1); MaskCpuFlags(benchmark_cpu_info_); MSAN_POISON(dst_opt0, kMaxW * 4); I210ToARGB(reinterpret_cast(src0), w, reinterpret_cast(src1), uv_w, reinterpret_cast(src2), uv_w, dst_opt0, w * 4, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w * 4); EXPECT_EQ(0, memcmp(dst_c0, dst_opt0, w * 4)); // 6. ARGBAlphaBlend with poisoned tails and poisoned destination. for (int i = 0; i < w * 4; ++i) { src0[i] = static_cast((i * 29 + 3) & 255); src1[i] = static_cast((i * 37 + 11) & 255); } MSAN_POISON(src0 + w * 4, 128); MSAN_POISON(src1 + w * 4, 128); MaskCpuFlags(disable_cpu_flags_); ARGBAlphaBlend(src0, w * 4, src1, w * 4, dst_c0, w * 4, w, 1); MaskCpuFlags(benchmark_cpu_info_); MSAN_POISON(dst_opt0, kMaxW * 4); ARGBAlphaBlend(src0, w * 4, src1, w * 4, dst_opt0, w * 4, w, 1); MSAN_CHECK_INITIALIZED(dst_opt0, w * 4); EXPECT_EQ(0, memcmp(dst_c0, dst_opt0, w * 4)); // 7. UVScale 3x box downscale (hits ScaleUVAddCols3_AVX512VBMI). const int src_uv_w = w * 3; const int src_uv_stride = src_uv_w * 2 + 64; MSAN_POISON(uv_src, kMaxSrcUV * 3); for (int row = 0; row < 3; ++row) { for (int i = 0; i < src_uv_w * 2; ++i) { uv_src[row * src_uv_stride + i] = static_cast((row * 53 + i * 17 + 7) & 255); } } MaskCpuFlags(disable_cpu_flags_); UVScale(uv_src, src_uv_stride, src_uv_w, 3, dst_c0, w * 2, w, 1, kFilterBox); MaskCpuFlags(benchmark_cpu_info_); MSAN_POISON(dst_opt0, kMaxW * 4); UVScale(uv_src, src_uv_stride, src_uv_w, 3, dst_opt0, w * 2, w, 1, kFilterBox); MSAN_CHECK_INITIALIZED(dst_opt0, w * 2); EXPECT_EQ(0, memcmp(dst_c0, dst_opt0, w * 2)); } free_aligned_buffer_page_end(src0); free_aligned_buffer_page_end(src1); free_aligned_buffer_page_end(src2); free_aligned_buffer_page_end(dst_c0); free_aligned_buffer_page_end(dst_c1); free_aligned_buffer_page_end(dst_c2); free_aligned_buffer_page_end(dst_opt0); free_aligned_buffer_page_end(dst_opt1); free_aligned_buffer_page_end(dst_opt2); free_aligned_buffer_page_end(uv_src); } // Regression test for RGBToUVMatrixRow_Any_SSSE3 MASK=15 (previously MASK=7 // overread 24 bytes past src_rgb24 at width=24). TEST_F(LibYUVPlanarTest, RGB24ToI420_Width24_OOB) { const int kWidth = 24; const int kHeight = 2; align_buffer_page_end(src_rgb24, kWidth * 3 * kHeight); align_buffer_page_end(dst_y_c, kWidth * kHeight); align_buffer_page_end(dst_u_c, (kWidth / 2) * (kHeight / 2)); align_buffer_page_end(dst_v_c, (kWidth / 2) * (kHeight / 2)); align_buffer_page_end(dst_y_opt, kWidth * kHeight); align_buffer_page_end(dst_u_opt, (kWidth / 2) * (kHeight / 2)); align_buffer_page_end(dst_v_opt, (kWidth / 2) * (kHeight / 2)); MemRandomize(src_rgb24, kWidth * 3 * kHeight); MaskCpuFlags(disable_cpu_flags_); RGB24ToI420(src_rgb24, kWidth * 3, dst_y_c, kWidth, dst_u_c, kWidth / 2, dst_v_c, kWidth / 2, kWidth, kHeight); const int kNoAvx512 = benchmark_cpu_info_ & ~(kCpuHasAVX512BW | kCpuHasAVX512VL | kCpuHasAVX512VBMI | kCpuHasAVX512VNNI); const int kSsse3Only = kNoAvx512 & ~kCpuHasAVX2; const int kModes[3] = {benchmark_cpu_info_, kNoAvx512, kSsse3Only}; for (int m = 0; m < 3; ++m) { MaskCpuFlags(kModes[m]); memset(dst_y_opt, 0, kWidth * kHeight); memset(dst_u_opt, 0, (kWidth / 2) * (kHeight / 2)); memset(dst_v_opt, 0, (kWidth / 2) * (kHeight / 2)); RGB24ToI420(src_rgb24, kWidth * 3, dst_y_opt, kWidth, dst_u_opt, kWidth / 2, dst_v_opt, kWidth / 2, kWidth, kHeight); for (int i = 0; i < kWidth * kHeight; ++i) { ASSERT_EQ(dst_y_c[i], dst_y_opt[i]); } for (int i = 0; i < (kWidth / 2) * (kHeight / 2); ++i) { ASSERT_EQ(dst_u_c[i], dst_u_opt[i]); ASSERT_EQ(dst_v_c[i], dst_v_opt[i]); } } MaskCpuFlags(benchmark_cpu_info_); free_aligned_buffer_page_end(src_rgb24); 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); } #ifdef ENABLE_ROW_TESTS #if defined(HAS_HALFWIDTHROW_16TO8_SSSE3) || \ defined(HAS_HALFWIDTHROW_16TO8_AVX2) || \ defined(HAS_HALFWIDTHROW_16TO8_AVX512BW) || \ defined(HAS_HALFROW_16TO8_AVX512BW) // Regression test for HalfWidthRow_16To8 (>= 0x8000 unsigned MSB samples and // exact page-end tail bounds) and HalfRow_16To8 (non-power-of-two scale). TEST_F(LibYUVPlanarTest, Half16To8_HighBitAndNonPow2Scale) { const int kWidths[] = {1, 7, 15, 24, 31, 33, 63, 65}; for (size_t w_idx = 0; w_idx < sizeof(kWidths) / sizeof(kWidths[0]); ++w_idx) { const int w = kWidths[w_idx]; align_buffer_page_end_16(src_uv_hw, w * 4); align_buffer_page_end(dst_c, w); align_buffer_page_end(dst_opt, w); for (int i = 0; i < w * 4; ++i) { src_uv_hw[i] = static_cast(0x8000u + ((i * 257u) & 0x7fffu)); } HalfWidthRow_16To8_C(src_uv_hw, w * 2, dst_c, 256, w); #if defined(HAS_HALFWIDTHROW_16TO8_AVX512BW) if (TestCpuFlag(kCpuHasAVX512BW)) { memset(dst_opt, 0x5a, w); HalfWidthRow_16To8_AVX512BW(src_uv_hw, w * 2, dst_opt, 256, w); for (int i = 0; i < w; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "AVX512BW w=" << w << " i=" << i; } } #endif #if defined(HAS_HALFWIDTHROW_16TO8_AVX2) if (TestCpuFlag(kCpuHasAVX2)) { memset(dst_opt, 0x5a, w); HalfWidthRow_16To8_Any_AVX2(src_uv_hw, w * 2, dst_opt, 256, w); for (int i = 0; i < w; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "AVX2 w=" << w << " i=" << i; } } #endif #if defined(HAS_HALFWIDTHROW_16TO8_SSSE3) if (TestCpuFlag(kCpuHasSSSE3)) { memset(dst_opt, 0x5a, w); HalfWidthRow_16To8_Any_SSSE3(src_uv_hw, w * 2, dst_opt, 256, w); for (int i = 0; i < w; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "SSSE3 w=" << w << " i=" << i; } } #endif // HalfRow_16To8 with non-power-of-two scale (e.g. 16320). align_buffer_page_end_16(src_uv_hr, w * 2); for (int i = 0; i < w * 2; ++i) { src_uv_hr[i] = static_cast((i * 73u) & 1023u); } HalfRow_16To8_C(src_uv_hr, w, dst_c, 16320, w); #if defined(HAS_HALFROW_16TO8_AVX512BW) if (TestCpuFlag(kCpuHasAVX512BW)) { memset(dst_opt, 0x5a, w); HalfRow_16To8_AVX512BW(src_uv_hr, w, dst_opt, 16320, w); for (int i = 0; i < w; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "HalfRow_16To8_AVX512BW w=" << w << " i=" << i; } } #endif free_aligned_buffer_page_end_16(src_uv_hr); free_aligned_buffer_page_end_16(src_uv_hw); free_aligned_buffer_page_end(dst_c); free_aligned_buffer_page_end(dst_opt); } } #endif // HAS_HALFWIDTHROW_16TO8_* || HAS_HALFROW_16TO8_AVX512BW #if defined(HAS_UNPACKMT2T_AVX2) // Regression test for UnpackMT2T_C and UnpackMT2T_AVX2 when size is not a // multiple of the 80-byte (64-pixel) MT2T block size. TEST_F(LibYUVPlanarTest, UnpackMT2T_PartialTail) { if (!TestCpuFlag(kCpuHasAVX2)) { return; } const size_t kSrcSize = 120; // 1 full 80-byte block + 40 trailing bytes const size_t kDstCount = (kSrcSize / 80) * 64; align_buffer_page_end(src, kSrcSize); align_buffer_page_end_16(dst_c, kDstCount); align_buffer_page_end_16(dst_opt, kDstCount); MemRandomize(src, kSrcSize); UnpackMT2T_C(src, dst_c, kSrcSize); UnpackMT2T_AVX2(src, dst_opt, kSrcSize); for (size_t i = 0; i < kDstCount; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "i=" << i; } // Verify size < 80 performs zero out-of-bounds reads or writes. UnpackMT2T_C(src + 80, dst_c, 40); UnpackMT2T_AVX2(src + 80, dst_opt, 40); for (size_t i = 0; i < kDstCount; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "i=" << i; } free_aligned_buffer_page_end(src); free_aligned_buffer_page_end_16(dst_c); free_aligned_buffer_page_end_16(dst_opt); } #endif // HAS_UNPACKMT2T_AVX2 #endif // ENABLE_ROW_TESTS // Regression test for ConvertToLSBPlane_16 with depth=16 (scale=65536). TEST_F(LibYUVPlanarTest, ConvertToLSBPlane_16_Depth16) { const int kWidth = 37; const int kHeight = 4; align_buffer_page_end_16(src_y, kWidth * kHeight); align_buffer_page_end_16(dst_c, kWidth * kHeight); align_buffer_page_end_16(dst_opt, kWidth * kHeight); for (int i = 0; i < kWidth * kHeight; ++i) { src_y[i] = static_cast(0x1234u + i * 251u); } MaskCpuFlags(disable_cpu_flags_); ConvertToLSBPlane_16(src_y, kWidth, dst_c, kWidth, kWidth, kHeight, 16); const int kNoAvx512 = benchmark_cpu_info_ & ~(kCpuHasAVX512BW | kCpuHasAVX512VL); MaskCpuFlags(benchmark_cpu_info_); memset(dst_opt, 0, kWidth * kHeight * 2); ConvertToLSBPlane_16(src_y, kWidth, dst_opt, kWidth, kWidth, kHeight, 16); for (int i = 0; i < kWidth * kHeight; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]); } MaskCpuFlags(kNoAvx512); memset(dst_opt, 0, kWidth * kHeight * 2); ConvertToLSBPlane_16(src_y, kWidth, dst_opt, kWidth, kWidth, kHeight, 16); for (int i = 0; i < kWidth * kHeight; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]); } MaskCpuFlags(benchmark_cpu_info_); free_aligned_buffer_page_end_16(src_y); free_aligned_buffer_page_end_16(dst_c); free_aligned_buffer_page_end_16(dst_opt); } // Regression test for ARGBPolynomial saturation on outputs >= 32768 / 65536, // odd widths, and 4-byte-aligned (non-16-byte-aligned) polynomial arrays. TEST_F(LibYUVPlanarTest, ARGBPolynomial_LargeOutputAndUnalignedPoly) { const int kWidth = 37; const int kHeight = 3; align_buffer_page_end(src_argb, kWidth * 4 * kHeight); align_buffer_page_end(dst_c, kWidth * 4 * kHeight); align_buffer_page_end(dst_opt, kWidth * 4 * kHeight); align_buffer_page_end(poly_buf, 17 * sizeof(float)); // Offset by +1 float so poly is 4-byte aligned, not 16-byte aligned. float* poly = reinterpret_cast(poly_buf) + 1; for (int c = 0; c < 4; ++c) { poly[0 + c] = 0.0f; poly[4 + c] = 300.0f; // 255 * 300 = 76500 >= 65536 -> must clamp to 255 poly[8 + c] = 1.0f; // 200^2 = 40000 in [32768, 65535] -> clamp to 255 poly[12 + c] = 0.0f; } for (int i = 0; i < kWidth * 4 * kHeight; ++i) { src_argb[i] = static_cast(i & 255); } MaskCpuFlags(disable_cpu_flags_); ARGBPolynomial(src_argb, kWidth * 4, dst_c, kWidth * 4, poly, kWidth, kHeight); const int kSse2Only = benchmark_cpu_info_ & ~(kCpuHasAVX2 | kCpuHasFMA3 | kCpuHasAVX512BW); const int kModes[2] = {benchmark_cpu_info_, kSse2Only}; for (int m = 0; m < 2; ++m) { MaskCpuFlags(kModes[m]); memset(dst_opt, 0, kWidth * 4 * kHeight); ARGBPolynomial(src_argb, kWidth * 4, dst_opt, kWidth * 4, poly, kWidth, kHeight); for (int i = 0; i < kWidth * 4 * kHeight; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "mode=" << m << " i=" << i; } } MaskCpuFlags(benchmark_cpu_info_); free_aligned_buffer_page_end(src_argb); free_aligned_buffer_page_end(dst_c); free_aligned_buffer_page_end(dst_opt); free_aligned_buffer_page_end(poly_buf); } // Regression test for ARGBBlur / CumulativeSumToAverageRow_AVX2 when radius=0 // or radius=1 produces small window areas (area == 1 or area == 2) on 0xff. TEST_F(LibYUVPlanarTest, ARGBBlur_SmallRadiusWhite) { const int kWidth = 37; const int kHeight = 4; const int kCumSumStride32 = (kWidth + 16) * 4; align_buffer_page_end(src_argb, kWidth * 4 * kHeight); align_buffer_page_end(dst_c, kWidth * 4 * kHeight); align_buffer_page_end(dst_opt, kWidth * 4 * kHeight); align_buffer_page_end(cumsum, kCumSumStride32 * (kHeight + 2) * 4); memset(src_argb, 0xff, kWidth * 4 * kHeight); for (int radius = 1; radius <= 2; ++radius) { MaskCpuFlags(disable_cpu_flags_); ARGBBlur(src_argb, kWidth * 4, dst_c, kWidth * 4, reinterpret_cast(cumsum), kCumSumStride32, kWidth, kHeight, radius); MaskCpuFlags(benchmark_cpu_info_); memset(dst_opt, 0, kWidth * 4 * kHeight); ARGBBlur(src_argb, kWidth * 4, dst_opt, kWidth * 4, reinterpret_cast(cumsum), kCumSumStride32, kWidth, kHeight, radius); for (int i = 0; i < kWidth * 4 * kHeight; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "radius=" << radius << " i=" << i; } } free_aligned_buffer_page_end(src_argb); free_aligned_buffer_page_end(dst_c); free_aligned_buffer_page_end(dst_opt); free_aligned_buffer_page_end(cumsum); } // Regression test for ARGBLumaColorTable with standard BT.601 green weight // 150 (> 127) and bright pixels where R*77 + G*150 + B*29 > 32767. TEST_F(LibYUVPlanarTest, ARGBLumaColorTable_BrightPixels) { const int kWidth = 37; const int kHeight = 3; align_buffer_page_end(src_argb, kWidth * 4 * kHeight); align_buffer_page_end(dst_c, kWidth * 4 * kHeight); align_buffer_page_end(dst_opt, kWidth * 4 * kHeight); align_buffer_page_end(luma, 32768); for (int i = 0; i < 32768; ++i) { luma[i] = static_cast(((i >> 8) * 3 + (i & 255)) & 255); } for (int i = 0; i < kWidth * kHeight; ++i) { src_argb[i * 4 + 0] = static_cast(200 + (i % 56)); src_argb[i * 4 + 1] = static_cast(200 + ((i * 3) % 56)); src_argb[i * 4 + 2] = static_cast(200 + ((i * 7) % 56)); src_argb[i * 4 + 3] = static_cast(i & 255); } MaskCpuFlags(disable_cpu_flags_); ARGBLumaColorTable(src_argb, kWidth * 4, dst_c, kWidth * 4, luma, kWidth, kHeight); MaskCpuFlags(benchmark_cpu_info_); ARGBLumaColorTable(src_argb, kWidth * 4, dst_opt, kWidth * 4, luma, kWidth, kHeight); for (int i = 0; i < kWidth * 4 * kHeight; ++i) { ASSERT_EQ(dst_c[i], dst_opt[i]) << "i=" << i; } free_aligned_buffer_page_end(src_argb); free_aligned_buffer_page_end(dst_c); free_aligned_buffer_page_end(dst_opt); free_aligned_buffer_page_end(luma); } #if defined(ENABLE_ROW_TESTS) && !defined(LIBYUV_DISABLE_X86) && \ (defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64) || \ defined(__i386__)) // Regression test for 2-byte-aligned (non-16/32-byte-aligned) Intel // YuvConstants and ArgbConstants structs and custom non-0x8000 kAddUV offsets. TEST_F(LibYUVPlanarTest, UnalignedMatrixConstantsAndCustomKAddUV) { const int kWidth = 38; const int kHeight = 4; align_buffer_page_end(yuvc_buf, sizeof(YuvConstants) + 16); align_buffer_page_end(argbc_buf, sizeof(ArgbConstants) + 16); // Offset by +2 bytes (matches alignof == 2, misaligned for movdqa/vmovdqa). YuvConstants* yuvc = reinterpret_cast(yuvc_buf + 2); ArgbConstants* argbc = reinterpret_cast(argbc_buf + 2); memcpy(yuvc, &kYuvI601Constants, sizeof(YuvConstants)); memcpy(argbc, &kArgbI601Constants, sizeof(ArgbConstants)); for (int i = 0; i < 16; ++i) { argbc->kAddUV[i] = 0x7200u; // Non-default UV bias (!= 0x8000) } align_buffer_page_end(src_argb, kWidth * 4 * kHeight); align_buffer_page_end(y_c, kWidth * kHeight); align_buffer_page_end(u_c, kWidth * kHeight); align_buffer_page_end(v_c, kWidth * kHeight); align_buffer_page_end(y_opt, kWidth * kHeight); align_buffer_page_end(u_opt, kWidth * kHeight); align_buffer_page_end(v_opt, kWidth * kHeight); align_buffer_page_end(rgb_c, kWidth * 4 * kHeight); align_buffer_page_end(rgb_opt, kWidth * 4 * kHeight); MemRandomize(src_argb, kWidth * 4 * kHeight); MaskCpuFlags(disable_cpu_flags_); ARGBToI444Matrix(src_argb, kWidth * 4, y_c, kWidth, u_c, kWidth, v_c, kWidth, argbc, kWidth, kHeight); I420ToARGBMatrix(y_c, kWidth, u_c, kWidth / 2, v_c, kWidth / 2, rgb_c, kWidth * 4, yuvc, kWidth, kHeight); const int kNoAvx512 = benchmark_cpu_info_ & ~(kCpuHasAVX512BW | kCpuHasAVX512VL | kCpuHasAVX512VBMI | kCpuHasAVX512VNNI); const int kSsse3Only = kNoAvx512 & ~kCpuHasAVX2; const int kModes[3] = {benchmark_cpu_info_, kNoAvx512, kSsse3Only}; for (int m = 0; m < 3; ++m) { MaskCpuFlags(kModes[m]); ARGBToI444Matrix(src_argb, kWidth * 4, y_opt, kWidth, u_opt, kWidth, v_opt, kWidth, argbc, kWidth, kHeight); I420ToARGBMatrix(y_c, kWidth, u_c, kWidth / 2, v_c, kWidth / 2, rgb_opt, kWidth * 4, yuvc, kWidth, kHeight); for (int i = 0; i < kWidth * kHeight; ++i) { ASSERT_EQ(y_c[i], y_opt[i]); ASSERT_EQ(u_c[i], u_opt[i]) << "mode=" << m << " i=" << i; ASSERT_EQ(v_c[i], v_opt[i]) << "mode=" << m << " i=" << i; } for (int i = 0; i < kWidth * 4 * kHeight; ++i) { ASSERT_EQ(rgb_c[i], rgb_opt[i]) << "mode=" << m << " i=" << i; } } // Also test ARGBToI420Matrix with custom kAddUV and I400ToARGBMatrix with // unaligned YuvConstants. MaskCpuFlags(disable_cpu_flags_); ARGBToI420Matrix(src_argb, kWidth * 4, y_c, kWidth, u_c, kWidth / 2, v_c, kWidth / 2, argbc, kWidth, kHeight); I400ToARGBMatrix(y_c, kWidth, rgb_c, kWidth * 4, yuvc, kWidth, kHeight); for (int m = 0; m < 3; ++m) { MaskCpuFlags(kModes[m]); ARGBToI420Matrix(src_argb, kWidth * 4, y_opt, kWidth, u_opt, kWidth / 2, v_opt, kWidth / 2, argbc, kWidth, kHeight); I400ToARGBMatrix(y_c, kWidth, rgb_opt, kWidth * 4, yuvc, kWidth, kHeight); for (int i = 0; i < (kWidth / 2) * (kHeight / 2); ++i) { ASSERT_NEAR(u_c[i], u_opt[i], 1) << "mode=" << m << " i=" << i; ASSERT_NEAR(v_c[i], v_opt[i], 1) << "mode=" << m << " i=" << i; } for (int i = 0; i < kWidth * 4 * kHeight; ++i) { ASSERT_EQ(rgb_c[i], rgb_opt[i]) << "mode=" << m << " i=" << i; } } MaskCpuFlags(benchmark_cpu_info_); free_aligned_buffer_page_end(yuvc_buf); free_aligned_buffer_page_end(argbc_buf); free_aligned_buffer_page_end(src_argb); free_aligned_buffer_page_end(y_c); free_aligned_buffer_page_end(u_c); free_aligned_buffer_page_end(v_c); free_aligned_buffer_page_end(y_opt); free_aligned_buffer_page_end(u_opt); free_aligned_buffer_page_end(v_opt); free_aligned_buffer_page_end(rgb_c); free_aligned_buffer_page_end(rgb_opt); } #endif // ENABLE_ROW_TESTS // Regression test for MergeAR64Plane depth=16 (UBSan shift) and // MergeARGB16To8Plane depth=8 saturation on out-of-range >255 samples. TEST_F(LibYUVPlanarTest, MergeAR64Depth16AndMergeARGB16To8Depth8Clamp) { const int kWidth = 37; const int kHeight = 2; align_buffer_page_end_16(src_r, kWidth * kHeight); align_buffer_page_end_16(src_g, kWidth * kHeight); align_buffer_page_end_16(src_b, kWidth * kHeight); align_buffer_page_end_16(src_a, kWidth * kHeight); align_buffer_page_end_16(ar64_c, kWidth * 4 * kHeight); align_buffer_page_end_16(ar64_opt, kWidth * 4 * kHeight); align_buffer_page_end(argb_c, kWidth * 4 * kHeight); align_buffer_page_end(argb_opt, kWidth * 4 * kHeight); for (int i = 0; i < kWidth * kHeight; ++i) { src_r[i] = static_cast(200u + (i * 19u)); src_g[i] = static_cast(100u + (i * 37u)); src_b[i] = static_cast(50u + (i * 53u)); src_a[i] = static_cast(0x8000u + (i * 101u)); } MaskCpuFlags(disable_cpu_flags_); MergeAR64Plane(src_r, kWidth, src_g, kWidth, src_b, kWidth, src_a, kWidth, ar64_c, kWidth * 4, kWidth, kHeight, 16); MergeARGB16To8Plane(src_r, kWidth, src_g, kWidth, src_b, kWidth, src_a, kWidth, argb_c, kWidth * 4, kWidth, kHeight, 8); MaskCpuFlags(benchmark_cpu_info_); MergeAR64Plane(src_r, kWidth, src_g, kWidth, src_b, kWidth, src_a, kWidth, ar64_opt, kWidth * 4, kWidth, kHeight, 16); MergeARGB16To8Plane(src_r, kWidth, src_g, kWidth, src_b, kWidth, src_a, kWidth, argb_opt, kWidth * 4, kWidth, kHeight, 8); for (int i = 0; i < kWidth * 4 * kHeight; ++i) { ASSERT_EQ(ar64_c[i], ar64_opt[i]); ASSERT_EQ(argb_c[i], argb_opt[i]); } free_aligned_buffer_page_end_16(src_r); free_aligned_buffer_page_end_16(src_g); free_aligned_buffer_page_end_16(src_b); free_aligned_buffer_page_end_16(src_a); free_aligned_buffer_page_end_16(ar64_c); free_aligned_buffer_page_end_16(ar64_opt); free_aligned_buffer_page_end(argb_c); free_aligned_buffer_page_end(argb_opt); } } // namespace libyuv