michael@0: /* michael@0: * Copyright 2011 The LibYuv Project Authors. All rights reserved. michael@0: * michael@0: * Use of this source code is governed by a BSD-style license michael@0: * that can be found in the LICENSE file in the root of the source michael@0: * tree. An additional intellectual property rights grant can be found michael@0: * in the file PATENTS. All contributing project authors may michael@0: * be found in the AUTHORS file in the root of the source tree. michael@0: */ michael@0: michael@0: #include michael@0: #include michael@0: michael@0: #include "libyuv/compare.h" michael@0: #include "libyuv/convert.h" michael@0: #include "libyuv/convert_argb.h" michael@0: #include "libyuv/convert_from.h" michael@0: #include "libyuv/convert_from_argb.h" michael@0: #include "libyuv/cpu_id.h" michael@0: #include "libyuv/format_conversion.h" michael@0: #include "libyuv/planar_functions.h" michael@0: #include "libyuv/rotate.h" michael@0: #include "libyuv/row.h" // For Sobel michael@0: #include "../unit_test/unit_test.h" michael@0: michael@0: #if defined(_MSC_VER) michael@0: #define SIMD_ALIGNED(var) __declspec(align(16)) var michael@0: #else // __GNUC__ michael@0: #define SIMD_ALIGNED(var) var __attribute__((aligned(16))) michael@0: #endif michael@0: michael@0: namespace libyuv { michael@0: michael@0: TEST_F(libyuvTest, TestAttenuate) { michael@0: const int kSize = 1280 * 4; michael@0: align_buffer_64(orig_pixels, kSize); michael@0: align_buffer_64(atten_pixels, kSize); michael@0: align_buffer_64(unatten_pixels, kSize); michael@0: align_buffer_64(atten2_pixels, kSize); michael@0: michael@0: // Test unattenuation clamps michael@0: orig_pixels[0 * 4 + 0] = 200u; michael@0: orig_pixels[0 * 4 + 1] = 129u; michael@0: orig_pixels[0 * 4 + 2] = 127u; michael@0: orig_pixels[0 * 4 + 3] = 128u; michael@0: // Test unattenuation transparent and opaque are unaffected michael@0: orig_pixels[1 * 4 + 0] = 16u; michael@0: orig_pixels[1 * 4 + 1] = 64u; michael@0: orig_pixels[1 * 4 + 2] = 192u; michael@0: orig_pixels[1 * 4 + 3] = 0u; michael@0: orig_pixels[2 * 4 + 0] = 16u; michael@0: orig_pixels[2 * 4 + 1] = 64u; michael@0: orig_pixels[2 * 4 + 2] = 192u; michael@0: orig_pixels[2 * 4 + 3] = 255u; michael@0: orig_pixels[3 * 4 + 0] = 16u; michael@0: orig_pixels[3 * 4 + 1] = 64u; michael@0: orig_pixels[3 * 4 + 2] = 192u; michael@0: orig_pixels[3 * 4 + 3] = 128u; michael@0: ARGBUnattenuate(orig_pixels, 0, unatten_pixels, 0, 4, 1); michael@0: EXPECT_EQ(255u, unatten_pixels[0 * 4 + 0]); michael@0: EXPECT_EQ(255u, unatten_pixels[0 * 4 + 1]); michael@0: EXPECT_EQ(254u, unatten_pixels[0 * 4 + 2]); michael@0: EXPECT_EQ(128u, unatten_pixels[0 * 4 + 3]); michael@0: EXPECT_EQ(0u, unatten_pixels[1 * 4 + 0]); michael@0: EXPECT_EQ(0u, unatten_pixels[1 * 4 + 1]); michael@0: EXPECT_EQ(0u, unatten_pixels[1 * 4 + 2]); michael@0: EXPECT_EQ(0u, unatten_pixels[1 * 4 + 3]); michael@0: EXPECT_EQ(16u, unatten_pixels[2 * 4 + 0]); michael@0: EXPECT_EQ(64u, unatten_pixels[2 * 4 + 1]); michael@0: EXPECT_EQ(192u, unatten_pixels[2 * 4 + 2]); michael@0: EXPECT_EQ(255u, unatten_pixels[2 * 4 + 3]); michael@0: EXPECT_EQ(32u, unatten_pixels[3 * 4 + 0]); michael@0: EXPECT_EQ(128u, unatten_pixels[3 * 4 + 1]); michael@0: EXPECT_EQ(255u, unatten_pixels[3 * 4 + 2]); michael@0: EXPECT_EQ(128u, unatten_pixels[3 * 4 + 3]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i * 4 + 0] = i; michael@0: orig_pixels[i * 4 + 1] = i / 2; michael@0: orig_pixels[i * 4 + 2] = i / 3; michael@0: orig_pixels[i * 4 + 3] = i; michael@0: } michael@0: ARGBAttenuate(orig_pixels, 0, atten_pixels, 0, 1280, 1); michael@0: ARGBUnattenuate(atten_pixels, 0, unatten_pixels, 0, 1280, 1); michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBAttenuate(unatten_pixels, 0, atten2_pixels, 0, 1280, 1); michael@0: } michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_NEAR(atten_pixels[i * 4 + 0], atten2_pixels[i * 4 + 0], 2); michael@0: EXPECT_NEAR(atten_pixels[i * 4 + 1], atten2_pixels[i * 4 + 1], 2); michael@0: EXPECT_NEAR(atten_pixels[i * 4 + 2], atten2_pixels[i * 4 + 2], 2); michael@0: EXPECT_NEAR(atten_pixels[i * 4 + 3], atten2_pixels[i * 4 + 3], 2); michael@0: } michael@0: // Make sure transparent, 50% and opaque are fully accurate. michael@0: EXPECT_EQ(0, atten_pixels[0 * 4 + 0]); michael@0: EXPECT_EQ(0, atten_pixels[0 * 4 + 1]); michael@0: EXPECT_EQ(0, atten_pixels[0 * 4 + 2]); michael@0: EXPECT_EQ(0, atten_pixels[0 * 4 + 3]); michael@0: EXPECT_EQ(64, atten_pixels[128 * 4 + 0]); michael@0: EXPECT_EQ(32, atten_pixels[128 * 4 + 1]); michael@0: EXPECT_EQ(21, atten_pixels[128 * 4 + 2]); michael@0: EXPECT_EQ(128, atten_pixels[128 * 4 + 3]); michael@0: EXPECT_NEAR(255, atten_pixels[255 * 4 + 0], 1); michael@0: EXPECT_NEAR(127, atten_pixels[255 * 4 + 1], 1); michael@0: EXPECT_NEAR(85, atten_pixels[255 * 4 + 2], 1); michael@0: EXPECT_EQ(255, atten_pixels[255 * 4 + 3]); michael@0: michael@0: free_aligned_buffer_64(atten2_pixels); michael@0: free_aligned_buffer_64(unatten_pixels); michael@0: free_aligned_buffer_64(atten_pixels); michael@0: free_aligned_buffer_64(orig_pixels); michael@0: } michael@0: michael@0: static int TestAttenuateI(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = (width * kBpp + 15) & ~15; michael@0: align_buffer_64(src_argb, kStride * height + off); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb[i + off] = (random() & 0xff); michael@0: } michael@0: memset(dst_argb_c, 0, kStride * height); michael@0: memset(dst_argb_opt, 0, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBAttenuate(src_argb + off, kStride, michael@0: dst_argb_c, kStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBAttenuate(src_argb + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBAttenuate_Any) { michael@0: int max_diff = TestAttenuateI(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 2); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBAttenuate_Unaligned) { michael@0: int max_diff = TestAttenuateI(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_LE(max_diff, 2); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBAttenuate_Invert) { michael@0: int max_diff = TestAttenuateI(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_LE(max_diff, 2); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBAttenuate_Opt) { michael@0: int max_diff = TestAttenuateI(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 2); michael@0: } michael@0: michael@0: static int TestUnattenuateI(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = (width * kBpp + 15) & ~15; michael@0: align_buffer_64(src_argb, kStride * height + off); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb[i + off] = (random() & 0xff); michael@0: } michael@0: ARGBAttenuate(src_argb + off, kStride, michael@0: src_argb + off, kStride, michael@0: width, height); michael@0: memset(dst_argb_c, 0, kStride * height); michael@0: memset(dst_argb_opt, 0, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBUnattenuate(src_argb + off, kStride, michael@0: dst_argb_c, kStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBUnattenuate(src_argb + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBUnattenuate_Any) { michael@0: int max_diff = TestUnattenuateI(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 2); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBUnattenuate_Unaligned) { michael@0: int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_LE(max_diff, 2); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBUnattenuate_Invert) { michael@0: int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_LE(max_diff, 2); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBUnattenuate_Opt) { michael@0: int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 2); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBComputeCumulativeSum) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[16][16][4]); michael@0: SIMD_ALIGNED(int32 added_pixels[16][16][4]); michael@0: michael@0: for (int y = 0; y < 16; ++y) { michael@0: for (int x = 0; x < 16; ++x) { michael@0: orig_pixels[y][x][0] = 1u; michael@0: orig_pixels[y][x][1] = 2u; michael@0: orig_pixels[y][x][2] = 3u; michael@0: orig_pixels[y][x][3] = 255u; michael@0: } michael@0: } michael@0: michael@0: ARGBComputeCumulativeSum(&orig_pixels[0][0][0], 16 * 4, michael@0: &added_pixels[0][0][0], 16 * 4, michael@0: 16, 16); michael@0: michael@0: for (int y = 0; y < 16; ++y) { michael@0: for (int x = 0; x < 16; ++x) { michael@0: EXPECT_EQ((x + 1) * (y + 1), added_pixels[y][x][0]); michael@0: EXPECT_EQ((x + 1) * (y + 1) * 2, added_pixels[y][x][1]); michael@0: EXPECT_EQ((x + 1) * (y + 1) * 3, added_pixels[y][x][2]); michael@0: EXPECT_EQ((x + 1) * (y + 1) * 255, added_pixels[y][x][3]); michael@0: } michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBGray) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: // Test blue michael@0: orig_pixels[0][0] = 255u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 128u; michael@0: // Test green michael@0: orig_pixels[1][0] = 0u; michael@0: orig_pixels[1][1] = 255u; michael@0: orig_pixels[1][2] = 0u; michael@0: orig_pixels[1][3] = 0u; michael@0: // Test red michael@0: orig_pixels[2][0] = 0u; michael@0: orig_pixels[2][1] = 0u; michael@0: orig_pixels[2][2] = 255u; michael@0: orig_pixels[2][3] = 255u; michael@0: // Test black michael@0: orig_pixels[3][0] = 0u; michael@0: orig_pixels[3][1] = 0u; michael@0: orig_pixels[3][2] = 0u; michael@0: orig_pixels[3][3] = 255u; michael@0: // Test white michael@0: orig_pixels[4][0] = 255u; michael@0: orig_pixels[4][1] = 255u; michael@0: orig_pixels[4][2] = 255u; michael@0: orig_pixels[4][3] = 255u; michael@0: // Test color michael@0: orig_pixels[5][0] = 16u; michael@0: orig_pixels[5][1] = 64u; michael@0: orig_pixels[5][2] = 192u; michael@0: orig_pixels[5][3] = 224u; michael@0: // Do 16 to test asm version. michael@0: ARGBGray(&orig_pixels[0][0], 0, 0, 0, 16, 1); michael@0: EXPECT_EQ(30u, orig_pixels[0][0]); michael@0: EXPECT_EQ(30u, orig_pixels[0][1]); michael@0: EXPECT_EQ(30u, orig_pixels[0][2]); michael@0: EXPECT_EQ(128u, orig_pixels[0][3]); michael@0: EXPECT_EQ(149u, orig_pixels[1][0]); michael@0: EXPECT_EQ(149u, orig_pixels[1][1]); michael@0: EXPECT_EQ(149u, orig_pixels[1][2]); michael@0: EXPECT_EQ(0u, orig_pixels[1][3]); michael@0: EXPECT_EQ(76u, orig_pixels[2][0]); michael@0: EXPECT_EQ(76u, orig_pixels[2][1]); michael@0: EXPECT_EQ(76u, orig_pixels[2][2]); michael@0: EXPECT_EQ(255u, orig_pixels[2][3]); michael@0: EXPECT_EQ(0u, orig_pixels[3][0]); michael@0: EXPECT_EQ(0u, orig_pixels[3][1]); michael@0: EXPECT_EQ(0u, orig_pixels[3][2]); michael@0: EXPECT_EQ(255u, orig_pixels[3][3]); michael@0: EXPECT_EQ(255u, orig_pixels[4][0]); michael@0: EXPECT_EQ(255u, orig_pixels[4][1]); michael@0: EXPECT_EQ(255u, orig_pixels[4][2]); michael@0: EXPECT_EQ(255u, orig_pixels[4][3]); michael@0: EXPECT_EQ(96u, orig_pixels[5][0]); michael@0: EXPECT_EQ(96u, orig_pixels[5][1]); michael@0: EXPECT_EQ(96u, orig_pixels[5][2]); michael@0: EXPECT_EQ(224u, orig_pixels[5][3]); michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBGray(&orig_pixels[0][0], 0, 0, 0, 1280, 1); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBGrayTo) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: SIMD_ALIGNED(uint8 gray_pixels[1280][4]); michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: // Test blue michael@0: orig_pixels[0][0] = 255u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 128u; michael@0: // Test green michael@0: orig_pixels[1][0] = 0u; michael@0: orig_pixels[1][1] = 255u; michael@0: orig_pixels[1][2] = 0u; michael@0: orig_pixels[1][3] = 0u; michael@0: // Test red michael@0: orig_pixels[2][0] = 0u; michael@0: orig_pixels[2][1] = 0u; michael@0: orig_pixels[2][2] = 255u; michael@0: orig_pixels[2][3] = 255u; michael@0: // Test black michael@0: orig_pixels[3][0] = 0u; michael@0: orig_pixels[3][1] = 0u; michael@0: orig_pixels[3][2] = 0u; michael@0: orig_pixels[3][3] = 255u; michael@0: // Test white michael@0: orig_pixels[4][0] = 255u; michael@0: orig_pixels[4][1] = 255u; michael@0: orig_pixels[4][2] = 255u; michael@0: orig_pixels[4][3] = 255u; michael@0: // Test color michael@0: orig_pixels[5][0] = 16u; michael@0: orig_pixels[5][1] = 64u; michael@0: orig_pixels[5][2] = 192u; michael@0: orig_pixels[5][3] = 224u; michael@0: // Do 16 to test asm version. michael@0: ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 16, 1); michael@0: EXPECT_EQ(30u, gray_pixels[0][0]); michael@0: EXPECT_EQ(30u, gray_pixels[0][1]); michael@0: EXPECT_EQ(30u, gray_pixels[0][2]); michael@0: EXPECT_EQ(128u, gray_pixels[0][3]); michael@0: EXPECT_EQ(149u, gray_pixels[1][0]); michael@0: EXPECT_EQ(149u, gray_pixels[1][1]); michael@0: EXPECT_EQ(149u, gray_pixels[1][2]); michael@0: EXPECT_EQ(0u, gray_pixels[1][3]); michael@0: EXPECT_EQ(76u, gray_pixels[2][0]); michael@0: EXPECT_EQ(76u, gray_pixels[2][1]); michael@0: EXPECT_EQ(76u, gray_pixels[2][2]); michael@0: EXPECT_EQ(255u, gray_pixels[2][3]); michael@0: EXPECT_EQ(0u, gray_pixels[3][0]); michael@0: EXPECT_EQ(0u, gray_pixels[3][1]); michael@0: EXPECT_EQ(0u, gray_pixels[3][2]); michael@0: EXPECT_EQ(255u, gray_pixels[3][3]); michael@0: EXPECT_EQ(255u, gray_pixels[4][0]); michael@0: EXPECT_EQ(255u, gray_pixels[4][1]); michael@0: EXPECT_EQ(255u, gray_pixels[4][2]); michael@0: EXPECT_EQ(255u, gray_pixels[4][3]); michael@0: EXPECT_EQ(96u, gray_pixels[5][0]); michael@0: EXPECT_EQ(96u, gray_pixels[5][1]); michael@0: EXPECT_EQ(96u, gray_pixels[5][2]); michael@0: EXPECT_EQ(224u, gray_pixels[5][3]); michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 1280, 1); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBSepia) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: // Test blue michael@0: orig_pixels[0][0] = 255u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 128u; michael@0: // Test green michael@0: orig_pixels[1][0] = 0u; michael@0: orig_pixels[1][1] = 255u; michael@0: orig_pixels[1][2] = 0u; michael@0: orig_pixels[1][3] = 0u; michael@0: // Test red michael@0: orig_pixels[2][0] = 0u; michael@0: orig_pixels[2][1] = 0u; michael@0: orig_pixels[2][2] = 255u; michael@0: orig_pixels[2][3] = 255u; michael@0: // Test black michael@0: orig_pixels[3][0] = 0u; michael@0: orig_pixels[3][1] = 0u; michael@0: orig_pixels[3][2] = 0u; michael@0: orig_pixels[3][3] = 255u; michael@0: // Test white michael@0: orig_pixels[4][0] = 255u; michael@0: orig_pixels[4][1] = 255u; michael@0: orig_pixels[4][2] = 255u; michael@0: orig_pixels[4][3] = 255u; michael@0: // Test color michael@0: orig_pixels[5][0] = 16u; michael@0: orig_pixels[5][1] = 64u; michael@0: orig_pixels[5][2] = 192u; michael@0: orig_pixels[5][3] = 224u; michael@0: // Do 16 to test asm version. michael@0: ARGBSepia(&orig_pixels[0][0], 0, 0, 0, 16, 1); michael@0: EXPECT_EQ(33u, orig_pixels[0][0]); michael@0: EXPECT_EQ(43u, orig_pixels[0][1]); michael@0: EXPECT_EQ(47u, orig_pixels[0][2]); michael@0: EXPECT_EQ(128u, orig_pixels[0][3]); michael@0: EXPECT_EQ(135u, orig_pixels[1][0]); michael@0: EXPECT_EQ(175u, orig_pixels[1][1]); michael@0: EXPECT_EQ(195u, orig_pixels[1][2]); michael@0: EXPECT_EQ(0u, orig_pixels[1][3]); michael@0: EXPECT_EQ(69u, orig_pixels[2][0]); michael@0: EXPECT_EQ(89u, orig_pixels[2][1]); michael@0: EXPECT_EQ(99u, orig_pixels[2][2]); michael@0: EXPECT_EQ(255u, orig_pixels[2][3]); michael@0: EXPECT_EQ(0u, orig_pixels[3][0]); michael@0: EXPECT_EQ(0u, orig_pixels[3][1]); michael@0: EXPECT_EQ(0u, orig_pixels[3][2]); michael@0: EXPECT_EQ(255u, orig_pixels[3][3]); michael@0: EXPECT_EQ(239u, orig_pixels[4][0]); michael@0: EXPECT_EQ(255u, orig_pixels[4][1]); michael@0: EXPECT_EQ(255u, orig_pixels[4][2]); michael@0: EXPECT_EQ(255u, orig_pixels[4][3]); michael@0: EXPECT_EQ(88u, orig_pixels[5][0]); michael@0: EXPECT_EQ(114u, orig_pixels[5][1]); michael@0: EXPECT_EQ(127u, orig_pixels[5][2]); michael@0: EXPECT_EQ(224u, orig_pixels[5][3]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBSepia(&orig_pixels[0][0], 0, 0, 0, 1280, 1); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBColorMatrix) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: SIMD_ALIGNED(uint8 dst_pixels_opt[1280][4]); michael@0: SIMD_ALIGNED(uint8 dst_pixels_c[1280][4]); michael@0: michael@0: // Matrix for Sepia. michael@0: SIMD_ALIGNED(static const int8 kRGBToSepia[]) = { michael@0: 17 / 2, 68 / 2, 35 / 2, 0, michael@0: 22 / 2, 88 / 2, 45 / 2, 0, michael@0: 24 / 2, 98 / 2, 50 / 2, 0, michael@0: 0, 0, 0, 64, // Copy alpha. michael@0: }; michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: // Test blue michael@0: orig_pixels[0][0] = 255u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 128u; michael@0: // Test green michael@0: orig_pixels[1][0] = 0u; michael@0: orig_pixels[1][1] = 255u; michael@0: orig_pixels[1][2] = 0u; michael@0: orig_pixels[1][3] = 0u; michael@0: // Test red michael@0: orig_pixels[2][0] = 0u; michael@0: orig_pixels[2][1] = 0u; michael@0: orig_pixels[2][2] = 255u; michael@0: orig_pixels[2][3] = 255u; michael@0: // Test color michael@0: orig_pixels[3][0] = 16u; michael@0: orig_pixels[3][1] = 64u; michael@0: orig_pixels[3][2] = 192u; michael@0: orig_pixels[3][3] = 224u; michael@0: // Do 16 to test asm version. michael@0: ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, michael@0: &kRGBToSepia[0], 16, 1); michael@0: EXPECT_EQ(31u, dst_pixels_opt[0][0]); michael@0: EXPECT_EQ(43u, dst_pixels_opt[0][1]); michael@0: EXPECT_EQ(47u, dst_pixels_opt[0][2]); michael@0: EXPECT_EQ(128u, dst_pixels_opt[0][3]); michael@0: EXPECT_EQ(135u, dst_pixels_opt[1][0]); michael@0: EXPECT_EQ(175u, dst_pixels_opt[1][1]); michael@0: EXPECT_EQ(195u, dst_pixels_opt[1][2]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[1][3]); michael@0: EXPECT_EQ(67u, dst_pixels_opt[2][0]); michael@0: EXPECT_EQ(87u, dst_pixels_opt[2][1]); michael@0: EXPECT_EQ(99u, dst_pixels_opt[2][2]); michael@0: EXPECT_EQ(255u, dst_pixels_opt[2][3]); michael@0: EXPECT_EQ(87u, dst_pixels_opt[3][0]); michael@0: EXPECT_EQ(112u, dst_pixels_opt[3][1]); michael@0: EXPECT_EQ(127u, dst_pixels_opt[3][2]); michael@0: EXPECT_EQ(224u, dst_pixels_opt[3][3]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: MaskCpuFlags(0); michael@0: ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_c[0][0], 0, michael@0: &kRGBToSepia[0], 1280, 1); michael@0: MaskCpuFlags(-1); michael@0: michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, michael@0: &kRGBToSepia[0], 1280, 1); michael@0: } michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); michael@0: EXPECT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); michael@0: EXPECT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); michael@0: EXPECT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestRGBColorMatrix) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: michael@0: // Matrix for Sepia. michael@0: SIMD_ALIGNED(static const int8 kRGBToSepia[]) = { michael@0: 17, 68, 35, 0, michael@0: 22, 88, 45, 0, michael@0: 24, 98, 50, 0, michael@0: 0, 0, 0, 0, // Unused but makes matrix 16 bytes. michael@0: }; michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: // Test blue michael@0: orig_pixels[0][0] = 255u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 128u; michael@0: // Test green michael@0: orig_pixels[1][0] = 0u; michael@0: orig_pixels[1][1] = 255u; michael@0: orig_pixels[1][2] = 0u; michael@0: orig_pixels[1][3] = 0u; michael@0: // Test red michael@0: orig_pixels[2][0] = 0u; michael@0: orig_pixels[2][1] = 0u; michael@0: orig_pixels[2][2] = 255u; michael@0: orig_pixels[2][3] = 255u; michael@0: // Test color michael@0: orig_pixels[3][0] = 16u; michael@0: orig_pixels[3][1] = 64u; michael@0: orig_pixels[3][2] = 192u; michael@0: orig_pixels[3][3] = 224u; michael@0: // Do 16 to test asm version. michael@0: RGBColorMatrix(&orig_pixels[0][0], 0, &kRGBToSepia[0], 0, 0, 16, 1); michael@0: EXPECT_EQ(31u, orig_pixels[0][0]); michael@0: EXPECT_EQ(43u, orig_pixels[0][1]); michael@0: EXPECT_EQ(47u, orig_pixels[0][2]); michael@0: EXPECT_EQ(128u, orig_pixels[0][3]); michael@0: EXPECT_EQ(135u, orig_pixels[1][0]); michael@0: EXPECT_EQ(175u, orig_pixels[1][1]); michael@0: EXPECT_EQ(195u, orig_pixels[1][2]); michael@0: EXPECT_EQ(0u, orig_pixels[1][3]); michael@0: EXPECT_EQ(67u, orig_pixels[2][0]); michael@0: EXPECT_EQ(87u, orig_pixels[2][1]); michael@0: EXPECT_EQ(99u, orig_pixels[2][2]); michael@0: EXPECT_EQ(255u, orig_pixels[2][3]); michael@0: EXPECT_EQ(87u, orig_pixels[3][0]); michael@0: EXPECT_EQ(112u, orig_pixels[3][1]); michael@0: EXPECT_EQ(127u, orig_pixels[3][2]); michael@0: EXPECT_EQ(224u, orig_pixels[3][3]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: RGBColorMatrix(&orig_pixels[0][0], 0, &kRGBToSepia[0], 0, 0, 1280, 1); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBColorTable) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: // Matrix for Sepia. michael@0: static const uint8 kARGBTable[256 * 4] = { michael@0: 1u, 2u, 3u, 4u, michael@0: 5u, 6u, 7u, 8u, michael@0: 9u, 10u, 11u, 12u, michael@0: 13u, 14u, 15u, 16u, michael@0: }; michael@0: michael@0: orig_pixels[0][0] = 0u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 0u; michael@0: orig_pixels[1][0] = 1u; michael@0: orig_pixels[1][1] = 1u; michael@0: orig_pixels[1][2] = 1u; michael@0: orig_pixels[1][3] = 1u; michael@0: orig_pixels[2][0] = 2u; michael@0: orig_pixels[2][1] = 2u; michael@0: orig_pixels[2][2] = 2u; michael@0: orig_pixels[2][3] = 2u; michael@0: orig_pixels[3][0] = 0u; michael@0: orig_pixels[3][1] = 1u; michael@0: orig_pixels[3][2] = 2u; michael@0: orig_pixels[3][3] = 3u; michael@0: // Do 16 to test asm version. michael@0: ARGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 16, 1); michael@0: EXPECT_EQ(1u, orig_pixels[0][0]); michael@0: EXPECT_EQ(2u, orig_pixels[0][1]); michael@0: EXPECT_EQ(3u, orig_pixels[0][2]); michael@0: EXPECT_EQ(4u, orig_pixels[0][3]); michael@0: EXPECT_EQ(5u, orig_pixels[1][0]); michael@0: EXPECT_EQ(6u, orig_pixels[1][1]); michael@0: EXPECT_EQ(7u, orig_pixels[1][2]); michael@0: EXPECT_EQ(8u, orig_pixels[1][3]); michael@0: EXPECT_EQ(9u, orig_pixels[2][0]); michael@0: EXPECT_EQ(10u, orig_pixels[2][1]); michael@0: EXPECT_EQ(11u, orig_pixels[2][2]); michael@0: EXPECT_EQ(12u, orig_pixels[2][3]); michael@0: EXPECT_EQ(1u, orig_pixels[3][0]); michael@0: EXPECT_EQ(6u, orig_pixels[3][1]); michael@0: EXPECT_EQ(11u, orig_pixels[3][2]); michael@0: EXPECT_EQ(16u, orig_pixels[3][3]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 1280, 1); michael@0: } michael@0: } michael@0: michael@0: // Same as TestARGBColorTable except alpha does not change. michael@0: TEST_F(libyuvTest, TestRGBColorTable) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: // Matrix for Sepia. michael@0: static const uint8 kARGBTable[256 * 4] = { michael@0: 1u, 2u, 3u, 4u, michael@0: 5u, 6u, 7u, 8u, michael@0: 9u, 10u, 11u, 12u, michael@0: 13u, 14u, 15u, 16u, michael@0: }; michael@0: michael@0: orig_pixels[0][0] = 0u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 0u; michael@0: orig_pixels[1][0] = 1u; michael@0: orig_pixels[1][1] = 1u; michael@0: orig_pixels[1][2] = 1u; michael@0: orig_pixels[1][3] = 1u; michael@0: orig_pixels[2][0] = 2u; michael@0: orig_pixels[2][1] = 2u; michael@0: orig_pixels[2][2] = 2u; michael@0: orig_pixels[2][3] = 2u; michael@0: orig_pixels[3][0] = 0u; michael@0: orig_pixels[3][1] = 1u; michael@0: orig_pixels[3][2] = 2u; michael@0: orig_pixels[3][3] = 3u; michael@0: // Do 16 to test asm version. michael@0: RGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 16, 1); michael@0: EXPECT_EQ(1u, orig_pixels[0][0]); michael@0: EXPECT_EQ(2u, orig_pixels[0][1]); michael@0: EXPECT_EQ(3u, orig_pixels[0][2]); michael@0: EXPECT_EQ(0u, orig_pixels[0][3]); // Alpha unchanged. michael@0: EXPECT_EQ(5u, orig_pixels[1][0]); michael@0: EXPECT_EQ(6u, orig_pixels[1][1]); michael@0: EXPECT_EQ(7u, orig_pixels[1][2]); michael@0: EXPECT_EQ(1u, orig_pixels[1][3]); // Alpha unchanged. michael@0: EXPECT_EQ(9u, orig_pixels[2][0]); michael@0: EXPECT_EQ(10u, orig_pixels[2][1]); michael@0: EXPECT_EQ(11u, orig_pixels[2][2]); michael@0: EXPECT_EQ(2u, orig_pixels[2][3]); // Alpha unchanged. michael@0: EXPECT_EQ(1u, orig_pixels[3][0]); michael@0: EXPECT_EQ(6u, orig_pixels[3][1]); michael@0: EXPECT_EQ(11u, orig_pixels[3][2]); michael@0: EXPECT_EQ(3u, orig_pixels[3][3]); // Alpha unchanged. michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: RGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 1280, 1); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBQuantize) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: ARGBQuantize(&orig_pixels[0][0], 0, michael@0: (65536 + (8 / 2)) / 8, 8, 8 / 2, 0, 0, 1280, 1); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_EQ((i / 8 * 8 + 8 / 2) & 255, orig_pixels[i][0]); michael@0: EXPECT_EQ((i / 2 / 8 * 8 + 8 / 2) & 255, orig_pixels[i][1]); michael@0: EXPECT_EQ((i / 3 / 8 * 8 + 8 / 2) & 255, orig_pixels[i][2]); michael@0: EXPECT_EQ(i & 255, orig_pixels[i][3]); michael@0: } michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBQuantize(&orig_pixels[0][0], 0, michael@0: (65536 + (8 / 2)) / 8, 8, 8 / 2, 0, 0, 1280, 1); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBMirror) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: SIMD_ALIGNED(uint8 dst_pixels[1280][4]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i / 4; michael@0: } michael@0: ARGBMirror(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0, 1280, 1); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_EQ(i & 255, dst_pixels[1280 - 1 - i][0]); michael@0: EXPECT_EQ((i / 2) & 255, dst_pixels[1280 - 1 - i][1]); michael@0: EXPECT_EQ((i / 3) & 255, dst_pixels[1280 - 1 - i][2]); michael@0: EXPECT_EQ((i / 4) & 255, dst_pixels[1280 - 1 - i][3]); michael@0: } michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBMirror(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0, 1280, 1); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestShade) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: SIMD_ALIGNED(uint8 shade_pixels[1280][4]); michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: orig_pixels[0][0] = 10u; michael@0: orig_pixels[0][1] = 20u; michael@0: orig_pixels[0][2] = 40u; michael@0: orig_pixels[0][3] = 80u; michael@0: orig_pixels[1][0] = 0u; michael@0: orig_pixels[1][1] = 0u; michael@0: orig_pixels[1][2] = 0u; michael@0: orig_pixels[1][3] = 255u; michael@0: orig_pixels[2][0] = 0u; michael@0: orig_pixels[2][1] = 0u; michael@0: orig_pixels[2][2] = 0u; michael@0: orig_pixels[2][3] = 0u; michael@0: orig_pixels[3][0] = 0u; michael@0: orig_pixels[3][1] = 0u; michael@0: orig_pixels[3][2] = 0u; michael@0: orig_pixels[3][3] = 0u; michael@0: // Do 8 pixels to allow opt version to be used. michael@0: ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x80ffffff); michael@0: EXPECT_EQ(10u, shade_pixels[0][0]); michael@0: EXPECT_EQ(20u, shade_pixels[0][1]); michael@0: EXPECT_EQ(40u, shade_pixels[0][2]); michael@0: EXPECT_EQ(40u, shade_pixels[0][3]); michael@0: EXPECT_EQ(0u, shade_pixels[1][0]); michael@0: EXPECT_EQ(0u, shade_pixels[1][1]); michael@0: EXPECT_EQ(0u, shade_pixels[1][2]); michael@0: EXPECT_EQ(128u, shade_pixels[1][3]); michael@0: EXPECT_EQ(0u, shade_pixels[2][0]); michael@0: EXPECT_EQ(0u, shade_pixels[2][1]); michael@0: EXPECT_EQ(0u, shade_pixels[2][2]); michael@0: EXPECT_EQ(0u, shade_pixels[2][3]); michael@0: EXPECT_EQ(0u, shade_pixels[3][0]); michael@0: EXPECT_EQ(0u, shade_pixels[3][1]); michael@0: EXPECT_EQ(0u, shade_pixels[3][2]); michael@0: EXPECT_EQ(0u, shade_pixels[3][3]); michael@0: michael@0: ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x80808080); michael@0: EXPECT_EQ(5u, shade_pixels[0][0]); michael@0: EXPECT_EQ(10u, shade_pixels[0][1]); michael@0: EXPECT_EQ(20u, shade_pixels[0][2]); michael@0: EXPECT_EQ(40u, shade_pixels[0][3]); michael@0: michael@0: ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x10204080); michael@0: EXPECT_EQ(5u, shade_pixels[0][0]); michael@0: EXPECT_EQ(5u, shade_pixels[0][1]); michael@0: EXPECT_EQ(5u, shade_pixels[0][2]); michael@0: EXPECT_EQ(5u, shade_pixels[0][3]); michael@0: michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 1280, 1, michael@0: 0x80808080); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestInterpolate) { michael@0: SIMD_ALIGNED(uint8 orig_pixels_0[1280][4]); michael@0: SIMD_ALIGNED(uint8 orig_pixels_1[1280][4]); michael@0: SIMD_ALIGNED(uint8 interpolate_pixels[1280][4]); michael@0: memset(orig_pixels_0, 0, sizeof(orig_pixels_0)); michael@0: memset(orig_pixels_1, 0, sizeof(orig_pixels_1)); michael@0: michael@0: orig_pixels_0[0][0] = 16u; michael@0: orig_pixels_0[0][1] = 32u; michael@0: orig_pixels_0[0][2] = 64u; michael@0: orig_pixels_0[0][3] = 128u; michael@0: orig_pixels_0[1][0] = 0u; michael@0: orig_pixels_0[1][1] = 0u; michael@0: orig_pixels_0[1][2] = 0u; michael@0: orig_pixels_0[1][3] = 255u; michael@0: orig_pixels_0[2][0] = 0u; michael@0: orig_pixels_0[2][1] = 0u; michael@0: orig_pixels_0[2][2] = 0u; michael@0: orig_pixels_0[2][3] = 0u; michael@0: orig_pixels_0[3][0] = 0u; michael@0: orig_pixels_0[3][1] = 0u; michael@0: orig_pixels_0[3][2] = 0u; michael@0: orig_pixels_0[3][3] = 0u; michael@0: michael@0: orig_pixels_1[0][0] = 0u; michael@0: orig_pixels_1[0][1] = 0u; michael@0: orig_pixels_1[0][2] = 0u; michael@0: orig_pixels_1[0][3] = 0u; michael@0: orig_pixels_1[1][0] = 0u; michael@0: orig_pixels_1[1][1] = 0u; michael@0: orig_pixels_1[1][2] = 0u; michael@0: orig_pixels_1[1][3] = 0u; michael@0: orig_pixels_1[2][0] = 0u; michael@0: orig_pixels_1[2][1] = 0u; michael@0: orig_pixels_1[2][2] = 0u; michael@0: orig_pixels_1[2][3] = 0u; michael@0: orig_pixels_1[3][0] = 255u; michael@0: orig_pixels_1[3][1] = 255u; michael@0: orig_pixels_1[3][2] = 255u; michael@0: orig_pixels_1[3][3] = 255u; michael@0: michael@0: ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0, michael@0: &interpolate_pixels[0][0], 0, 4, 1, 128); michael@0: EXPECT_EQ(8u, interpolate_pixels[0][0]); michael@0: EXPECT_EQ(16u, interpolate_pixels[0][1]); michael@0: EXPECT_EQ(32u, interpolate_pixels[0][2]); michael@0: EXPECT_EQ(64u, interpolate_pixels[0][3]); michael@0: EXPECT_EQ(0u, interpolate_pixels[1][0]); michael@0: EXPECT_EQ(0u, interpolate_pixels[1][1]); michael@0: EXPECT_EQ(0u, interpolate_pixels[1][2]); michael@0: EXPECT_NEAR(128u, interpolate_pixels[1][3], 1); // C = 127, SSE = 128. michael@0: EXPECT_EQ(0u, interpolate_pixels[2][0]); michael@0: EXPECT_EQ(0u, interpolate_pixels[2][1]); michael@0: EXPECT_EQ(0u, interpolate_pixels[2][2]); michael@0: EXPECT_EQ(0u, interpolate_pixels[2][3]); michael@0: EXPECT_NEAR(128u, interpolate_pixels[3][0], 1); michael@0: EXPECT_NEAR(128u, interpolate_pixels[3][1], 1); michael@0: EXPECT_NEAR(128u, interpolate_pixels[3][2], 1); michael@0: EXPECT_NEAR(128u, interpolate_pixels[3][3], 1); michael@0: michael@0: ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0, michael@0: &interpolate_pixels[0][0], 0, 4, 1, 0); michael@0: EXPECT_EQ(16u, interpolate_pixels[0][0]); michael@0: EXPECT_EQ(32u, interpolate_pixels[0][1]); michael@0: EXPECT_EQ(64u, interpolate_pixels[0][2]); michael@0: EXPECT_EQ(128u, interpolate_pixels[0][3]); michael@0: michael@0: ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0, michael@0: &interpolate_pixels[0][0], 0, 4, 1, 192); michael@0: michael@0: EXPECT_EQ(4u, interpolate_pixels[0][0]); michael@0: EXPECT_EQ(8u, interpolate_pixels[0][1]); michael@0: EXPECT_EQ(16u, interpolate_pixels[0][2]); michael@0: EXPECT_EQ(32u, interpolate_pixels[0][3]); michael@0: michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0, michael@0: &interpolate_pixels[0][0], 0, 1280, 1, 128); michael@0: } michael@0: } michael@0: michael@0: #define TESTTERP(FMT_A, BPP_A, STRIDE_A, \ michael@0: FMT_B, BPP_B, STRIDE_B, \ michael@0: W1280, TERP, DIFF, N, NEG, OFF) \ michael@0: TEST_F(libyuvTest, ARGBInterpolate##TERP##N) { \ michael@0: const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ michael@0: const int kHeight = benchmark_height_; \ michael@0: const int kStrideA = (kWidth * BPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A; \ michael@0: const int kStrideB = (kWidth * BPP_B + STRIDE_B - 1) / STRIDE_B * STRIDE_B; \ michael@0: align_buffer_64(src_argb_a, kStrideA * kHeight + OFF); \ michael@0: align_buffer_64(src_argb_b, kStrideA * kHeight + OFF); \ michael@0: align_buffer_64(dst_argb_c, kStrideB * kHeight); \ michael@0: align_buffer_64(dst_argb_opt, kStrideB * kHeight); \ michael@0: srandom(time(NULL)); \ michael@0: for (int i = 0; i < kStrideA * kHeight; ++i) { \ michael@0: src_argb_a[i + OFF] = (random() & 0xff); \ michael@0: src_argb_b[i + OFF] = (random() & 0xff); \ michael@0: } \ michael@0: MaskCpuFlags(0); \ michael@0: ARGBInterpolate(src_argb_a + OFF, kStrideA, \ michael@0: src_argb_b + OFF, kStrideA, \ michael@0: dst_argb_c, kStrideB, \ michael@0: kWidth, NEG kHeight, TERP); \ michael@0: MaskCpuFlags(-1); \ michael@0: for (int i = 0; i < benchmark_iterations_; ++i) { \ michael@0: ARGBInterpolate(src_argb_a + OFF, kStrideA, \ michael@0: src_argb_b + OFF, kStrideA, \ michael@0: dst_argb_opt, kStrideB, \ michael@0: kWidth, NEG kHeight, TERP); \ michael@0: } \ michael@0: int max_diff = 0; \ michael@0: for (int i = 0; i < kStrideB * kHeight; ++i) { \ michael@0: int abs_diff = \ michael@0: abs(static_cast(dst_argb_c[i]) - \ michael@0: static_cast(dst_argb_opt[i])); \ michael@0: if (abs_diff > max_diff) { \ michael@0: max_diff = abs_diff; \ michael@0: } \ michael@0: } \ michael@0: EXPECT_LE(max_diff, DIFF); \ michael@0: free_aligned_buffer_64(src_argb_a); \ michael@0: free_aligned_buffer_64(src_argb_b); \ michael@0: free_aligned_buffer_64(dst_argb_c); \ michael@0: free_aligned_buffer_64(dst_argb_opt); \ michael@0: } michael@0: michael@0: #define TESTINTERPOLATE(TERP) \ michael@0: TESTTERP(ARGB, 4, 1, ARGB, 4, 1, \ michael@0: benchmark_width_ - 1, TERP, 1, _Any, +, 0) \ michael@0: TESTTERP(ARGB, 4, 1, ARGB, 4, 1, \ michael@0: benchmark_width_, TERP, 1, _Unaligned, +, 1) \ michael@0: TESTTERP(ARGB, 4, 1, ARGB, 4, 1, \ michael@0: benchmark_width_, TERP, 1, _Invert, -, 0) \ michael@0: TESTTERP(ARGB, 4, 1, ARGB, 4, 1, \ michael@0: benchmark_width_, TERP, 1, _Opt, +, 0) \ michael@0: TESTTERP(ARGB, 4, 1, ARGB, 4, 1, \ michael@0: benchmark_width_ - 1, TERP, 1, _Any_Invert, -, 0) michael@0: michael@0: TESTINTERPOLATE(0) michael@0: TESTINTERPOLATE(64) michael@0: TESTINTERPOLATE(128) michael@0: TESTINTERPOLATE(192) michael@0: TESTINTERPOLATE(255) michael@0: michael@0: static int TestBlend(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = width * kBpp; michael@0: align_buffer_64(src_argb_a, kStride * height + off); michael@0: align_buffer_64(src_argb_b, kStride * height + off); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb_a[i + off] = (random() & 0xff); michael@0: src_argb_b[i + off] = (random() & 0xff); michael@0: } michael@0: ARGBAttenuate(src_argb_a + off, kStride, src_argb_a + off, kStride, width, michael@0: height); michael@0: ARGBAttenuate(src_argb_b + off, kStride, src_argb_b + off, kStride, width, michael@0: height); michael@0: memset(dst_argb_c, 255, kStride * height); michael@0: memset(dst_argb_opt, 255, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBBlend(src_argb_a + off, kStride, michael@0: src_argb_b + off, kStride, michael@0: dst_argb_c, kStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBBlend(src_argb_a + off, kStride, michael@0: src_argb_b + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb_a); michael@0: free_aligned_buffer_64(src_argb_b); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlend_Any) { michael@0: int max_diff = TestBlend(benchmark_width_ - 4, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlend_Unaligned) { michael@0: int max_diff = TestBlend(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlend_Invert) { michael@0: int max_diff = TestBlend(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlend_Opt) { michael@0: int max_diff = TestBlend(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestAffine) { michael@0: SIMD_ALIGNED(uint8 orig_pixels_0[1280][4]); michael@0: SIMD_ALIGNED(uint8 interpolate_pixels_C[1280][4]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: for (int j = 0; j < 4; ++j) { michael@0: orig_pixels_0[i][j] = i; michael@0: } michael@0: } michael@0: michael@0: float uv_step[4] = { 0.f, 0.f, 0.75f, 0.f }; michael@0: michael@0: ARGBAffineRow_C(&orig_pixels_0[0][0], 0, &interpolate_pixels_C[0][0], michael@0: uv_step, 1280); michael@0: EXPECT_EQ(0u, interpolate_pixels_C[0][0]); michael@0: EXPECT_EQ(96u, interpolate_pixels_C[128][0]); michael@0: EXPECT_EQ(191u, interpolate_pixels_C[255][3]); michael@0: michael@0: #if defined(HAS_ARGBAFFINEROW_SSE2) michael@0: SIMD_ALIGNED(uint8 interpolate_pixels_Opt[1280][4]); michael@0: ARGBAffineRow_SSE2(&orig_pixels_0[0][0], 0, &interpolate_pixels_Opt[0][0], michael@0: uv_step, 1280); michael@0: EXPECT_EQ(0, memcmp(interpolate_pixels_Opt, interpolate_pixels_C, 1280 * 4)); michael@0: michael@0: int has_sse2 = TestCpuFlag(kCpuHasSSE2); michael@0: if (has_sse2) { michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBAffineRow_SSE2(&orig_pixels_0[0][0], 0, &interpolate_pixels_Opt[0][0], michael@0: uv_step, 1280); michael@0: } michael@0: } michael@0: #endif michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestSobelX) { michael@0: SIMD_ALIGNED(uint8 orig_pixels_0[1280 + 2]); michael@0: SIMD_ALIGNED(uint8 orig_pixels_1[1280 + 2]); michael@0: SIMD_ALIGNED(uint8 orig_pixels_2[1280 + 2]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_c[1280]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_opt[1280]); michael@0: michael@0: for (int i = 0; i < 1280 + 2; ++i) { michael@0: orig_pixels_0[i] = i; michael@0: orig_pixels_1[i] = i * 2; michael@0: orig_pixels_2[i] = i * 3; michael@0: } michael@0: michael@0: SobelXRow_C(orig_pixels_0, orig_pixels_1, orig_pixels_2, michael@0: sobel_pixels_c, 1280); michael@0: michael@0: EXPECT_EQ(16u, sobel_pixels_c[0]); michael@0: EXPECT_EQ(16u, sobel_pixels_c[100]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[255]); michael@0: michael@0: void (*SobelXRow)(const uint8* src_y0, const uint8* src_y1, michael@0: const uint8* src_y2, uint8* dst_sobely, int width) = michael@0: SobelXRow_C; michael@0: #if defined(HAS_SOBELXROW_SSE2) michael@0: if (TestCpuFlag(kCpuHasSSE2)) { michael@0: SobelXRow = SobelXRow_SSE2; michael@0: } michael@0: #endif michael@0: #if defined(HAS_SOBELXROW_NEON) michael@0: if (TestCpuFlag(kCpuHasNEON)) { michael@0: SobelXRow = SobelXRow_NEON; michael@0: } michael@0: #endif michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: SobelXRow(orig_pixels_0, orig_pixels_1, orig_pixels_2, michael@0: sobel_pixels_opt, 1280); michael@0: } michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_EQ(sobel_pixels_c[i], sobel_pixels_opt[i]); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestSobelY) { michael@0: SIMD_ALIGNED(uint8 orig_pixels_0[1280 + 2]); michael@0: SIMD_ALIGNED(uint8 orig_pixels_1[1280 + 2]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_c[1280]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_opt[1280]); michael@0: michael@0: for (int i = 0; i < 1280 + 2; ++i) { michael@0: orig_pixels_0[i] = i; michael@0: orig_pixels_1[i] = i * 2; michael@0: } michael@0: michael@0: SobelYRow_C(orig_pixels_0, orig_pixels_1, sobel_pixels_c, 1280); michael@0: michael@0: EXPECT_EQ(4u, sobel_pixels_c[0]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[100]); michael@0: EXPECT_EQ(0u, sobel_pixels_c[255]); michael@0: void (*SobelYRow)(const uint8* src_y0, const uint8* src_y1, michael@0: uint8* dst_sobely, int width) = SobelYRow_C; michael@0: #if defined(HAS_SOBELYROW_SSE2) michael@0: if (TestCpuFlag(kCpuHasSSE2)) { michael@0: SobelYRow = SobelYRow_SSE2; michael@0: } michael@0: #endif michael@0: #if defined(HAS_SOBELYROW_NEON) michael@0: if (TestCpuFlag(kCpuHasNEON)) { michael@0: SobelYRow = SobelYRow_NEON; michael@0: } michael@0: #endif michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: SobelYRow(orig_pixels_0, orig_pixels_1, sobel_pixels_opt, 1280); michael@0: } michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_EQ(sobel_pixels_c[i], sobel_pixels_opt[i]); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestSobel) { michael@0: SIMD_ALIGNED(uint8 orig_sobelx[1280]); michael@0: SIMD_ALIGNED(uint8 orig_sobely[1280]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_c[1280 * 4]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_opt[1280 * 4]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_sobelx[i] = i; michael@0: orig_sobely[i] = i * 2; michael@0: } michael@0: michael@0: SobelRow_C(orig_sobelx, orig_sobely, sobel_pixels_c, 1280); michael@0: michael@0: EXPECT_EQ(0u, sobel_pixels_c[0]); michael@0: EXPECT_EQ(3u, sobel_pixels_c[4]); michael@0: EXPECT_EQ(3u, sobel_pixels_c[5]); michael@0: EXPECT_EQ(3u, sobel_pixels_c[6]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[7]); michael@0: EXPECT_EQ(6u, sobel_pixels_c[8]); michael@0: EXPECT_EQ(6u, sobel_pixels_c[9]); michael@0: EXPECT_EQ(6u, sobel_pixels_c[10]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[7]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[100 * 4 + 1]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[255 * 4 + 1]); michael@0: void (*SobelRow)(const uint8* src_sobelx, const uint8* src_sobely, michael@0: uint8* dst_argb, int width) = SobelRow_C; michael@0: #if defined(HAS_SOBELROW_SSE2) michael@0: if (TestCpuFlag(kCpuHasSSE2)) { michael@0: SobelRow = SobelRow_SSE2; michael@0: } michael@0: #endif michael@0: #if defined(HAS_SOBELROW_NEON) michael@0: if (TestCpuFlag(kCpuHasNEON)) { michael@0: SobelRow = SobelRow_NEON; michael@0: } michael@0: #endif michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: SobelRow(orig_sobelx, orig_sobely, sobel_pixels_opt, 1280); michael@0: } michael@0: for (int i = 0; i < 1280 * 4; ++i) { michael@0: EXPECT_EQ(sobel_pixels_c[i], sobel_pixels_opt[i]); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestSobelToPlane) { michael@0: SIMD_ALIGNED(uint8 orig_sobelx[1280]); michael@0: SIMD_ALIGNED(uint8 orig_sobely[1280]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_c[1280]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_opt[1280]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_sobelx[i] = i; michael@0: orig_sobely[i] = i * 2; michael@0: } michael@0: michael@0: SobelToPlaneRow_C(orig_sobelx, orig_sobely, sobel_pixels_c, 1280); michael@0: michael@0: EXPECT_EQ(0u, sobel_pixels_c[0]); michael@0: EXPECT_EQ(3u, sobel_pixels_c[1]); michael@0: EXPECT_EQ(6u, sobel_pixels_c[2]); michael@0: EXPECT_EQ(99u, sobel_pixels_c[33]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[100]); michael@0: void (*SobelToPlaneRow)(const uint8* src_sobelx, const uint8* src_sobely, michael@0: uint8* dst_y, int width) = SobelToPlaneRow_C; michael@0: #if defined(HAS_SOBELTOPLANEROW_SSE2) michael@0: if (TestCpuFlag(kCpuHasSSE2)) { michael@0: SobelToPlaneRow = SobelToPlaneRow_SSE2; michael@0: } michael@0: #endif michael@0: #if defined(HAS_SOBELTOPLANEROW_NEON) michael@0: if (TestCpuFlag(kCpuHasNEON)) { michael@0: SobelToPlaneRow = SobelToPlaneRow_NEON; michael@0: } michael@0: #endif michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: SobelToPlaneRow(orig_sobelx, orig_sobely, sobel_pixels_opt, 1280); michael@0: } michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_EQ(sobel_pixels_c[i], sobel_pixels_opt[i]); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestSobelXY) { michael@0: SIMD_ALIGNED(uint8 orig_sobelx[1280]); michael@0: SIMD_ALIGNED(uint8 orig_sobely[1280]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_c[1280 * 4]); michael@0: SIMD_ALIGNED(uint8 sobel_pixels_opt[1280 * 4]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_sobelx[i] = i; michael@0: orig_sobely[i] = i * 2; michael@0: } michael@0: michael@0: SobelXYRow_C(orig_sobelx, orig_sobely, sobel_pixels_c, 1280); michael@0: michael@0: EXPECT_EQ(0u, sobel_pixels_c[0]); michael@0: EXPECT_EQ(2u, sobel_pixels_c[4]); michael@0: EXPECT_EQ(3u, sobel_pixels_c[5]); michael@0: EXPECT_EQ(1u, sobel_pixels_c[6]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[7]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[100 * 4 + 1]); michael@0: EXPECT_EQ(255u, sobel_pixels_c[255 * 4 + 1]); michael@0: void (*SobelXYRow)(const uint8* src_sobelx, const uint8* src_sobely, michael@0: uint8* dst_argb, int width) = SobelXYRow_C; michael@0: #if defined(HAS_SOBELXYROW_SSE2) michael@0: if (TestCpuFlag(kCpuHasSSE2)) { michael@0: SobelXYRow = SobelXYRow_SSE2; michael@0: } michael@0: #endif michael@0: #if defined(HAS_SOBELXYROW_NEON) michael@0: if (TestCpuFlag(kCpuHasNEON)) { michael@0: SobelXYRow = SobelXYRow_NEON; michael@0: } michael@0: #endif michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: SobelXYRow(orig_sobelx, orig_sobely, sobel_pixels_opt, 1280); michael@0: } michael@0: for (int i = 0; i < 1280 * 4; ++i) { michael@0: EXPECT_EQ(sobel_pixels_c[i], sobel_pixels_opt[i]); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestCopyPlane) { michael@0: int err = 0; michael@0: int yw = benchmark_width_; michael@0: int yh = benchmark_height_; michael@0: int b = 12; michael@0: int i, j; michael@0: michael@0: int y_plane_size = (yw + b * 2) * (yh + b * 2); michael@0: srandom(time(NULL)); michael@0: align_buffer_64(orig_y, y_plane_size); michael@0: align_buffer_64(dst_c, y_plane_size); michael@0: align_buffer_64(dst_opt, y_plane_size); michael@0: michael@0: memset(orig_y, 0, y_plane_size); michael@0: memset(dst_c, 0, y_plane_size); michael@0: memset(dst_opt, 0, y_plane_size); michael@0: michael@0: // Fill image buffers with random data. michael@0: for (i = b; i < (yh + b); ++i) { michael@0: for (j = b; j < (yw + b); ++j) { michael@0: orig_y[i * (yw + b * 2) + j] = random() & 0xff; michael@0: } michael@0: } michael@0: michael@0: // Fill destination buffers with random data. michael@0: for (i = 0; i < y_plane_size; ++i) { michael@0: uint8 random_number = random() & 0x7f; michael@0: dst_c[i] = random_number; michael@0: dst_opt[i] = dst_c[i]; michael@0: } michael@0: michael@0: int y_off = b * (yw + b * 2) + b; michael@0: michael@0: int y_st = yw + b * 2; michael@0: int stride = 8; michael@0: michael@0: // Disable all optimizations. michael@0: MaskCpuFlags(0); michael@0: double c_time = get_time(); michael@0: for (j = 0; j < benchmark_iterations_; j++) { michael@0: CopyPlane(orig_y + y_off, y_st, dst_c + y_off, stride, yw, yh); michael@0: } michael@0: c_time = (get_time() - c_time) / benchmark_iterations_; michael@0: michael@0: // Enable optimizations. michael@0: MaskCpuFlags(-1); michael@0: double opt_time = get_time(); michael@0: for (j = 0; j < benchmark_iterations_; j++) { michael@0: CopyPlane(orig_y + y_off, y_st, dst_opt + y_off, stride, yw, yh); michael@0: } michael@0: opt_time = (get_time() - opt_time) / benchmark_iterations_; michael@0: michael@0: for (i = 0; i < y_plane_size; ++i) { michael@0: if (dst_c[i] != dst_opt[i]) michael@0: ++err; michael@0: } michael@0: michael@0: free_aligned_buffer_64(orig_y); michael@0: free_aligned_buffer_64(dst_c); michael@0: free_aligned_buffer_64(dst_opt); michael@0: michael@0: EXPECT_EQ(0, err); michael@0: } michael@0: michael@0: static int TestMultiply(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = (width * kBpp + 15) & ~15; michael@0: align_buffer_64(src_argb_a, kStride * height + off); michael@0: align_buffer_64(src_argb_b, kStride * height + off); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb_a[i + off] = (random() & 0xff); michael@0: src_argb_b[i + off] = (random() & 0xff); michael@0: } michael@0: memset(dst_argb_c, 0, kStride * height); michael@0: memset(dst_argb_opt, 0, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBMultiply(src_argb_a + off, kStride, michael@0: src_argb_b + off, kStride, michael@0: dst_argb_c, kStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBMultiply(src_argb_a + off, kStride, michael@0: src_argb_b + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb_a); michael@0: free_aligned_buffer_64(src_argb_b); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBMultiply_Any) { michael@0: int max_diff = TestMultiply(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBMultiply_Unaligned) { michael@0: int max_diff = TestMultiply(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBMultiply_Invert) { michael@0: int max_diff = TestMultiply(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBMultiply_Opt) { michael@0: int max_diff = TestMultiply(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: static int TestAdd(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = (width * kBpp + 15) & ~15; michael@0: align_buffer_64(src_argb_a, kStride * height + off); michael@0: align_buffer_64(src_argb_b, kStride * height + off); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb_a[i + off] = (random() & 0xff); michael@0: src_argb_b[i + off] = (random() & 0xff); michael@0: } michael@0: memset(dst_argb_c, 0, kStride * height); michael@0: memset(dst_argb_opt, 0, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBAdd(src_argb_a + off, kStride, michael@0: src_argb_b + off, kStride, michael@0: dst_argb_c, kStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBAdd(src_argb_a + off, kStride, michael@0: src_argb_b + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb_a); michael@0: free_aligned_buffer_64(src_argb_b); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBAdd_Any) { michael@0: int max_diff = TestAdd(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBAdd_Unaligned) { michael@0: int max_diff = TestAdd(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBAdd_Invert) { michael@0: int max_diff = TestAdd(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBAdd_Opt) { michael@0: int max_diff = TestAdd(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: static int TestSubtract(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = (width * kBpp + 15) & ~15; michael@0: align_buffer_64(src_argb_a, kStride * height + off); michael@0: align_buffer_64(src_argb_b, kStride * height + off); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb_a[i + off] = (random() & 0xff); michael@0: src_argb_b[i + off] = (random() & 0xff); michael@0: } michael@0: memset(dst_argb_c, 0, kStride * height); michael@0: memset(dst_argb_opt, 0, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBSubtract(src_argb_a + off, kStride, michael@0: src_argb_b + off, kStride, michael@0: dst_argb_c, kStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBSubtract(src_argb_a + off, kStride, michael@0: src_argb_b + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb_a); michael@0: free_aligned_buffer_64(src_argb_b); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSubtract_Any) { michael@0: int max_diff = TestSubtract(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSubtract_Unaligned) { michael@0: int max_diff = TestSubtract(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSubtract_Invert) { michael@0: int max_diff = TestSubtract(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSubtract_Opt) { michael@0: int max_diff = TestSubtract(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: static int TestSobel(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = (width * kBpp + 15) & ~15; michael@0: align_buffer_64(src_argb_a, kStride * height + off); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: memset(src_argb_a, 0, kStride * height + off); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb_a[i + off] = (random() & 0xff); michael@0: } michael@0: memset(dst_argb_c, 0, kStride * height); michael@0: memset(dst_argb_opt, 0, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBSobel(src_argb_a + off, kStride, michael@0: dst_argb_c, kStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBSobel(src_argb_a + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb_a); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobel_Any) { michael@0: int max_diff = TestSobel(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobel_Unaligned) { michael@0: int max_diff = TestSobel(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobel_Invert) { michael@0: int max_diff = TestSobel(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobel_Opt) { michael@0: int max_diff = TestSobel(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: static int TestSobelToPlane(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kSrcBpp = 4; michael@0: const int kDstBpp = 1; michael@0: const int kSrcStride = (width * kSrcBpp + 15) & ~15; michael@0: const int kDstStride = (width * kDstBpp + 15) & ~15; michael@0: align_buffer_64(src_argb_a, kSrcStride * height + off); michael@0: align_buffer_64(dst_argb_c, kDstStride * height); michael@0: align_buffer_64(dst_argb_opt, kDstStride * height); michael@0: memset(src_argb_a, 0, kSrcStride * height + off); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kSrcStride * height; ++i) { michael@0: src_argb_a[i + off] = (random() & 0xff); michael@0: } michael@0: memset(dst_argb_c, 0, kDstStride * height); michael@0: memset(dst_argb_opt, 0, kDstStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBSobelToPlane(src_argb_a + off, kSrcStride, michael@0: dst_argb_c, kDstStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBSobelToPlane(src_argb_a + off, kSrcStride, michael@0: dst_argb_opt, kDstStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kDstStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb_a); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobelToPlane_Any) { michael@0: int max_diff = TestSobelToPlane(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobelToPlane_Unaligned) { michael@0: int max_diff = TestSobelToPlane(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobelToPlane_Invert) { michael@0: int max_diff = TestSobelToPlane(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobelToPlane_Opt) { michael@0: int max_diff = TestSobelToPlane(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: static int TestSobelXY(int width, int height, int benchmark_iterations, michael@0: int invert, int off) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = (width * kBpp + 15) & ~15; michael@0: align_buffer_64(src_argb_a, kStride * height + off); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: memset(src_argb_a, 0, kStride * height + off); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb_a[i + off] = (random() & 0xff); michael@0: } michael@0: memset(dst_argb_c, 0, kStride * height); michael@0: memset(dst_argb_opt, 0, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBSobelXY(src_argb_a + off, kStride, michael@0: dst_argb_c, kStride, michael@0: width, invert * height); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBSobelXY(src_argb_a + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: width, invert * height); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb_a); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobelXY_Any) { michael@0: int max_diff = TestSobelXY(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobelXY_Unaligned) { michael@0: int max_diff = TestSobelXY(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobelXY_Invert) { michael@0: int max_diff = TestSobelXY(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBSobelXY_Opt) { michael@0: int max_diff = TestSobelXY(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0); michael@0: EXPECT_EQ(0, max_diff); michael@0: } michael@0: michael@0: static int TestBlur(int width, int height, int benchmark_iterations, michael@0: int invert, int off, int radius) { michael@0: if (width < 1) { michael@0: width = 1; michael@0: } michael@0: const int kBpp = 4; michael@0: const int kStride = (width * kBpp + 15) & ~15; michael@0: align_buffer_64(src_argb_a, kStride * height + off); michael@0: align_buffer_64(dst_cumsum, width * height * 16); michael@0: align_buffer_64(dst_argb_c, kStride * height); michael@0: align_buffer_64(dst_argb_opt, kStride * height); michael@0: srandom(time(NULL)); michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: src_argb_a[i + off] = (random() & 0xff); michael@0: } michael@0: memset(dst_cumsum, 0, width * height * 16); michael@0: memset(dst_argb_c, 0, kStride * height); michael@0: memset(dst_argb_opt, 0, kStride * height); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBBlur(src_argb_a + off, kStride, michael@0: dst_argb_c, kStride, michael@0: reinterpret_cast(dst_cumsum), width * 4, michael@0: width, invert * height, radius); michael@0: MaskCpuFlags(-1); michael@0: for (int i = 0; i < benchmark_iterations; ++i) { michael@0: ARGBBlur(src_argb_a + off, kStride, michael@0: dst_argb_opt, kStride, michael@0: reinterpret_cast(dst_cumsum), width * 4, michael@0: width, invert * height, radius); michael@0: } michael@0: int max_diff = 0; michael@0: for (int i = 0; i < kStride * height; ++i) { michael@0: int abs_diff = michael@0: abs(static_cast(dst_argb_c[i]) - michael@0: static_cast(dst_argb_opt[i])); michael@0: if (abs_diff > max_diff) { michael@0: max_diff = abs_diff; michael@0: } michael@0: } michael@0: free_aligned_buffer_64(src_argb_a); michael@0: free_aligned_buffer_64(dst_cumsum); michael@0: free_aligned_buffer_64(dst_argb_c); michael@0: free_aligned_buffer_64(dst_argb_opt); michael@0: return max_diff; michael@0: } michael@0: michael@0: static const int kBlurSize = 55; michael@0: TEST_F(libyuvTest, ARGBBlur_Any) { michael@0: int max_diff = TestBlur(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0, kBlurSize); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlur_Unaligned) { michael@0: int max_diff = TestBlur(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1, kBlurSize); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlur_Invert) { michael@0: int max_diff = TestBlur(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0, kBlurSize); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlur_Opt) { michael@0: int max_diff = TestBlur(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0, kBlurSize); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: static const int kBlurSmallSize = 5; michael@0: TEST_F(libyuvTest, ARGBBlurSmall_Any) { michael@0: int max_diff = TestBlur(benchmark_width_ - 1, benchmark_height_, michael@0: benchmark_iterations_, +1, 0, kBlurSmallSize); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlurSmall_Unaligned) { michael@0: int max_diff = TestBlur(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 1, kBlurSmallSize); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlurSmall_Invert) { michael@0: int max_diff = TestBlur(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, -1, 0, kBlurSmallSize); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, ARGBBlurSmall_Opt) { michael@0: int max_diff = TestBlur(benchmark_width_, benchmark_height_, michael@0: benchmark_iterations_, +1, 0, kBlurSmallSize); michael@0: EXPECT_LE(max_diff, 1); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBPolynomial) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: SIMD_ALIGNED(uint8 dst_pixels_opt[1280][4]); michael@0: SIMD_ALIGNED(uint8 dst_pixels_c[1280][4]); michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: SIMD_ALIGNED(static const float kWarmifyPolynomial[16]) = { michael@0: 0.94230f, -3.03300f, -2.92500f, 0.f, // C0 michael@0: 0.584500f, 1.112000f, 1.535000f, 1.f, // C1 x michael@0: 0.001313f, -0.002503f, -0.004496f, 0.f, // C2 x * x michael@0: 0.0f, 0.000006965f, 0.000008781f, 0.f, // C3 x * x * x michael@0: }; michael@0: michael@0: // Test blue michael@0: orig_pixels[0][0] = 255u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 128u; michael@0: // Test green michael@0: orig_pixels[1][0] = 0u; michael@0: orig_pixels[1][1] = 255u; michael@0: orig_pixels[1][2] = 0u; michael@0: orig_pixels[1][3] = 0u; michael@0: // Test red michael@0: orig_pixels[2][0] = 0u; michael@0: orig_pixels[2][1] = 0u; michael@0: orig_pixels[2][2] = 255u; michael@0: orig_pixels[2][3] = 255u; michael@0: // Test white michael@0: orig_pixels[3][0] = 255u; michael@0: orig_pixels[3][1] = 255u; michael@0: orig_pixels[3][2] = 255u; michael@0: orig_pixels[3][3] = 255u; michael@0: // Test color michael@0: orig_pixels[4][0] = 16u; michael@0: orig_pixels[4][1] = 64u; michael@0: orig_pixels[4][2] = 192u; michael@0: orig_pixels[4][3] = 224u; michael@0: // Do 16 to test asm version. michael@0: ARGBPolynomial(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, michael@0: &kWarmifyPolynomial[0], 16, 1); michael@0: EXPECT_EQ(235u, dst_pixels_opt[0][0]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[0][1]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[0][2]); michael@0: EXPECT_EQ(128u, dst_pixels_opt[0][3]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[1][0]); michael@0: EXPECT_EQ(233u, dst_pixels_opt[1][1]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[1][2]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[1][3]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[2][0]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[2][1]); michael@0: EXPECT_EQ(241u, dst_pixels_opt[2][2]); michael@0: EXPECT_EQ(255u, dst_pixels_opt[2][3]); michael@0: EXPECT_EQ(235u, dst_pixels_opt[3][0]); michael@0: EXPECT_EQ(233u, dst_pixels_opt[3][1]); michael@0: EXPECT_EQ(241u, dst_pixels_opt[3][2]); michael@0: EXPECT_EQ(255u, dst_pixels_opt[3][3]); michael@0: EXPECT_EQ(10u, dst_pixels_opt[4][0]); michael@0: EXPECT_EQ(59u, dst_pixels_opt[4][1]); michael@0: EXPECT_EQ(188u, dst_pixels_opt[4][2]); michael@0: EXPECT_EQ(224u, dst_pixels_opt[4][3]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBPolynomial(&orig_pixels[0][0], 0, &dst_pixels_c[0][0], 0, michael@0: &kWarmifyPolynomial[0], 1280, 1); michael@0: MaskCpuFlags(-1); michael@0: michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBPolynomial(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, michael@0: &kWarmifyPolynomial[0], 1280, 1); michael@0: } michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); michael@0: EXPECT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); michael@0: EXPECT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); michael@0: EXPECT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]); michael@0: } michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBLumaColorTable) { michael@0: SIMD_ALIGNED(uint8 orig_pixels[1280][4]); michael@0: SIMD_ALIGNED(uint8 dst_pixels_opt[1280][4]); michael@0: SIMD_ALIGNED(uint8 dst_pixels_c[1280][4]); michael@0: memset(orig_pixels, 0, sizeof(orig_pixels)); michael@0: michael@0: align_buffer_64(lumacolortable, 32768); michael@0: int v = 0; michael@0: for (int i = 0; i < 32768; ++i) { michael@0: lumacolortable[i] = v; michael@0: v += 3; michael@0: } michael@0: // Test blue michael@0: orig_pixels[0][0] = 255u; michael@0: orig_pixels[0][1] = 0u; michael@0: orig_pixels[0][2] = 0u; michael@0: orig_pixels[0][3] = 128u; michael@0: // Test green michael@0: orig_pixels[1][0] = 0u; michael@0: orig_pixels[1][1] = 255u; michael@0: orig_pixels[1][2] = 0u; michael@0: orig_pixels[1][3] = 0u; michael@0: // Test red michael@0: orig_pixels[2][0] = 0u; michael@0: orig_pixels[2][1] = 0u; michael@0: orig_pixels[2][2] = 255u; michael@0: orig_pixels[2][3] = 255u; michael@0: // Test color michael@0: orig_pixels[3][0] = 16u; michael@0: orig_pixels[3][1] = 64u; michael@0: orig_pixels[3][2] = 192u; michael@0: orig_pixels[3][3] = 224u; michael@0: // Do 16 to test asm version. michael@0: ARGBLumaColorTable(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, michael@0: &lumacolortable[0], 16, 1); michael@0: EXPECT_EQ(253u, dst_pixels_opt[0][0]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[0][1]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[0][2]); michael@0: EXPECT_EQ(128u, dst_pixels_opt[0][3]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[1][0]); michael@0: EXPECT_EQ(253u, dst_pixels_opt[1][1]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[1][2]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[1][3]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[2][0]); michael@0: EXPECT_EQ(0u, dst_pixels_opt[2][1]); michael@0: EXPECT_EQ(253u, dst_pixels_opt[2][2]); michael@0: EXPECT_EQ(255u, dst_pixels_opt[2][3]); michael@0: EXPECT_EQ(48u, dst_pixels_opt[3][0]); michael@0: EXPECT_EQ(192u, dst_pixels_opt[3][1]); michael@0: EXPECT_EQ(64u, dst_pixels_opt[3][2]); michael@0: EXPECT_EQ(224u, dst_pixels_opt[3][3]); michael@0: michael@0: for (int i = 0; i < 1280; ++i) { michael@0: orig_pixels[i][0] = i; michael@0: orig_pixels[i][1] = i / 2; michael@0: orig_pixels[i][2] = i / 3; michael@0: orig_pixels[i][3] = i; michael@0: } michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBLumaColorTable(&orig_pixels[0][0], 0, &dst_pixels_c[0][0], 0, michael@0: lumacolortable, 1280, 1); michael@0: MaskCpuFlags(-1); michael@0: michael@0: for (int i = 0; i < benchmark_pixels_div1280_; ++i) { michael@0: ARGBLumaColorTable(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, michael@0: lumacolortable, 1280, 1); michael@0: } michael@0: for (int i = 0; i < 1280; ++i) { michael@0: EXPECT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); michael@0: EXPECT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); michael@0: EXPECT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); michael@0: EXPECT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]); michael@0: } michael@0: michael@0: free_aligned_buffer_64(lumacolortable); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBCopyAlpha) { michael@0: const int kSize = benchmark_width_ * benchmark_height_ * 4; michael@0: align_buffer_64(orig_pixels, kSize); michael@0: align_buffer_64(dst_pixels_opt, kSize); michael@0: align_buffer_64(dst_pixels_c, kSize); michael@0: michael@0: MemRandomize(orig_pixels, kSize); michael@0: MemRandomize(dst_pixels_opt, kSize); michael@0: memcpy(dst_pixels_c, dst_pixels_opt, kSize); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBCopyAlpha(orig_pixels, benchmark_width_ * 4, michael@0: dst_pixels_c, benchmark_width_ * 4, michael@0: benchmark_width_, benchmark_height_); michael@0: MaskCpuFlags(-1); michael@0: michael@0: for (int i = 0; i < benchmark_iterations_; ++i) { michael@0: ARGBCopyAlpha(orig_pixels, benchmark_width_ * 4, michael@0: dst_pixels_opt, benchmark_width_ * 4, michael@0: benchmark_width_, benchmark_height_); michael@0: } michael@0: for (int i = 0; i < kSize; ++i) { michael@0: EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); michael@0: } michael@0: michael@0: free_aligned_buffer_64(dst_pixels_c); michael@0: free_aligned_buffer_64(dst_pixels_opt); michael@0: free_aligned_buffer_64(orig_pixels); michael@0: } michael@0: michael@0: TEST_F(libyuvTest, TestARGBCopyYToAlpha) { michael@0: const int kPixels = benchmark_width_ * benchmark_height_; michael@0: align_buffer_64(orig_pixels, kPixels); michael@0: align_buffer_64(dst_pixels_opt, kPixels * 4); michael@0: align_buffer_64(dst_pixels_c, kPixels * 4); michael@0: michael@0: MemRandomize(orig_pixels, kPixels); michael@0: MemRandomize(dst_pixels_opt, kPixels * 4); michael@0: memcpy(dst_pixels_c, dst_pixels_opt, kPixels * 4); michael@0: michael@0: MaskCpuFlags(0); michael@0: ARGBCopyYToAlpha(orig_pixels, benchmark_width_, michael@0: dst_pixels_c, benchmark_width_ * 4, michael@0: benchmark_width_, benchmark_height_); michael@0: MaskCpuFlags(-1); michael@0: michael@0: for (int i = 0; i < benchmark_iterations_; ++i) { michael@0: ARGBCopyYToAlpha(orig_pixels, benchmark_width_, michael@0: dst_pixels_opt, benchmark_width_ * 4, michael@0: benchmark_width_, benchmark_height_); michael@0: } michael@0: for (int i = 0; i < kPixels * 4; ++i) { michael@0: EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); michael@0: } michael@0: michael@0: free_aligned_buffer_64(dst_pixels_c); michael@0: free_aligned_buffer_64(dst_pixels_opt); michael@0: free_aligned_buffer_64(orig_pixels); michael@0: } michael@0: michael@0: } // namespace libyuv