diff options
| author | Hangyu Kuang <hkuang@google.com> | 2016-07-06 14:21:45 -0700 |
|---|---|---|
| committer | Hangyu Kuang <hkuang@google.com> | 2016-07-08 09:51:10 -0700 |
| commit | f047e7ca6983218eed7703c7afd51fed7bd3b5c9 (patch) | |
| tree | 2667579566b6270c21ee4b495b4cd119af5ccf5b /files/unit_test/planar_test.cc | |
| parent | bb74e3e19b98261031216de8cadcef34cccd9e4a (diff) | |
| download | libyuv-f047e7ca6983218eed7703c7afd51fed7bd3b5c9.tar.gz | |
Update libyuv to r1602 version to get best performance.android-cts_7.1_r1android-cts-7.1_r9android-cts-7.1_r8android-cts-7.1_r7android-cts-7.1_r6android-cts-7.1_r5android-cts-7.1_r4android-cts-7.1_r3android-cts-7.1_r29android-cts-7.1_r28android-cts-7.1_r27android-cts-7.1_r26android-cts-7.1_r25android-cts-7.1_r24android-cts-7.1_r23android-cts-7.1_r22android-cts-7.1_r21android-cts-7.1_r20android-cts-7.1_r2android-cts-7.1_r19android-cts-7.1_r18android-cts-7.1_r17android-cts-7.1_r16android-cts-7.1_r15android-cts-7.1_r14android-cts-7.1_r13android-cts-7.1_r12android-cts-7.1_r11android-cts-7.1_r10android-cts-7.1_r1android-7.1.2_r9android-7.1.2_r8android-7.1.2_r6android-7.1.2_r5android-7.1.2_r4android-7.1.2_r39android-7.1.2_r38android-7.1.2_r37android-7.1.2_r36android-7.1.2_r33android-7.1.2_r32android-7.1.2_r30android-7.1.2_r3android-7.1.2_r29android-7.1.2_r28android-7.1.2_r27android-7.1.2_r25android-7.1.2_r24android-7.1.2_r23android-7.1.2_r2android-7.1.2_r19android-7.1.2_r18android-7.1.2_r17android-7.1.2_r16android-7.1.2_r15android-7.1.2_r14android-7.1.2_r13android-7.1.2_r12android-7.1.2_r11android-7.1.2_r10android-7.1.2_r1android-7.1.1_r9android-7.1.1_r8android-7.1.1_r7android-7.1.1_r61android-7.1.1_r60android-7.1.1_r6android-7.1.1_r59android-7.1.1_r58android-7.1.1_r57android-7.1.1_r56android-7.1.1_r55android-7.1.1_r54android-7.1.1_r53android-7.1.1_r52android-7.1.1_r51android-7.1.1_r50android-7.1.1_r49android-7.1.1_r48android-7.1.1_r47android-7.1.1_r46android-7.1.1_r45android-7.1.1_r44android-7.1.1_r43android-7.1.1_r42android-7.1.1_r41android-7.1.1_r40android-7.1.1_r4android-7.1.1_r39android-7.1.1_r38android-7.1.1_r35android-7.1.1_r33android-7.1.1_r32android-7.1.1_r31android-7.1.1_r3android-7.1.1_r28android-7.1.1_r27android-7.1.1_r26android-7.1.1_r25android-7.1.1_r24android-7.1.1_r23android-7.1.1_r22android-7.1.1_r21android-7.1.1_r20android-7.1.1_r2android-7.1.1_r17android-7.1.1_r16android-7.1.1_r15android-7.1.1_r14android-7.1.1_r13android-7.1.1_r12android-7.1.1_r11android-7.1.1_r10android-7.1.1_r1android-7.1.0_r7android-7.1.0_r6android-7.1.0_r5android-7.1.0_r4android-7.1.0_r3android-7.1.0_r2android-7.1.0_r1nougat-mr2.3-releasenougat-mr2.2-releasenougat-mr2.1-releasenougat-mr2-security-releasenougat-mr2-releasenougat-mr2-pixel-releasenougat-mr2-devnougat-mr1.8-releasenougat-mr1.7-releasenougat-mr1.6-releasenougat-mr1.5-releasenougat-mr1.4-releasenougat-mr1.3-releasenougat-mr1.2-releasenougat-mr1.1-releasenougat-mr1-volantis-releasenougat-mr1-security-releasenougat-mr1-releasenougat-mr1-flounder-releasenougat-mr1-devnougat-mr1-cts-releasenougat-dr1-release
Bug: 29870647
Change-Id: I8ec9fab7f55765fa33ebe7ba1c7ad2147f418de2
Diffstat (limited to 'files/unit_test/planar_test.cc')
| -rw-r--r-- | files/unit_test/planar_test.cc | 2520 |
1 files changed, 1938 insertions, 582 deletions
diff --git a/files/unit_test/planar_test.cc b/files/unit_test/planar_test.cc index e9053a35..bc0eebb5 100644 --- a/files/unit_test/planar_test.cc +++ b/files/unit_test/planar_test.cc @@ -4,460 +4,257 @@ * 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 + * in the file PATENTS. All contributing project authors may * be found in the AUTHORS file in the root of the source tree. */ #include <stdlib.h> #include <time.h> +#include "libyuv/compare.h" +#include "libyuv/convert.h" #include "libyuv/convert_argb.h" #include "libyuv/convert_from.h" -#include "libyuv/compare.h" +#include "libyuv/convert_from_argb.h" #include "libyuv/cpu_id.h" -#include "libyuv/format_conversion.h" #include "libyuv/planar_functions.h" #include "libyuv/rotate.h" #include "../unit_test/unit_test.h" -#if defined(_MSC_VER) -#define SIMD_ALIGNED(var) __declspec(align(16)) var -#else // __GNUC__ -#define SIMD_ALIGNED(var) var __attribute__((aligned(16))) -#endif - namespace libyuv { -#define TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, N, NEG) \ -TEST_F(libyuvTest, FMT_PLANAR##To##FMT_B##N##_OptVsC) { \ - const int kWidth = 1280; \ - const int kHeight = 720; \ - const int kStride = (kWidth * 8 * BPP_B + 7) / 8; \ - align_buffer_16(src_y, kWidth * kHeight); \ - align_buffer_16(src_u, kWidth / SUBSAMP_X * kHeight / SUBSAMP_Y); \ - align_buffer_16(src_v, kWidth / SUBSAMP_X * kHeight / SUBSAMP_Y); \ - align_buffer_16(dst_argb_c, kStride * kHeight); \ - align_buffer_16(dst_argb_opt, kStride * kHeight); \ - srandom(time(NULL)); \ - for (int i = 0; i < kHeight; ++i) \ - for (int j = 0; j < kWidth; ++j) \ - src_y[(i * kWidth) + j] = (random() & 0xff); \ - for (int i = 0; i < kHeight / SUBSAMP_Y; ++i) \ - for (int j = 0; j < kWidth / SUBSAMP_X; ++j) { \ - src_u[(i * kWidth / SUBSAMP_X) + j] = (random() & 0xff); \ - src_v[(i * kWidth / SUBSAMP_X) + j] = (random() & 0xff); \ - } \ - MaskCpuFlags(kCpuInitialized); \ - FMT_PLANAR##To##FMT_B(src_y, kWidth, \ - src_u, kWidth / SUBSAMP_X, \ - src_v, kWidth / SUBSAMP_X, \ - dst_argb_c, kStride, \ - kWidth, NEG kHeight); \ - MaskCpuFlags(-1); \ - for (int i = 0; i < benchmark_iterations_; ++i) { \ - FMT_PLANAR##To##FMT_B(src_y, kWidth, \ - src_u, kWidth / SUBSAMP_X, \ - src_v, kWidth / SUBSAMP_X, \ - dst_argb_opt, kStride, \ - kWidth, NEG kHeight); \ - } \ - int max_diff = 0; \ - for (int i = 0; i < kHeight; ++i) { \ - for (int j = 0; j < kWidth * BPP_B; ++j) { \ - int abs_diff = \ - abs(static_cast<int>(dst_argb_c[i * kWidth * BPP_B + j]) - \ - static_cast<int>(dst_argb_opt[i * kWidth * BPP_B + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 2); \ - free_aligned_buffer_16(src_y) \ - free_aligned_buffer_16(src_u) \ - free_aligned_buffer_16(src_v) \ - free_aligned_buffer_16(dst_argb_c) \ - free_aligned_buffer_16(dst_argb_opt) \ -} - -#define TESTPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B) \ - TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, , +) \ - TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, Invert, -) - -TESTPLANARTOB(I420, 2, 2, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, BGRA, 4) -TESTPLANARTOB(I420, 2, 2, ABGR, 4) -TESTPLANARTOB(I420, 2, 2, RGBA, 4) -TESTPLANARTOB(I420, 2, 2, RAW, 3) -TESTPLANARTOB(I420, 2, 2, RGB24, 3) -TESTPLANARTOB(I420, 2, 2, RGB565, 2) -TESTPLANARTOB(I420, 2, 2, ARGB1555, 2) -TESTPLANARTOB(I420, 2, 2, ARGB4444, 2) -TESTPLANARTOB(I422, 2, 1, ARGB, 4) -TESTPLANARTOB(I422, 2, 1, BGRA, 4) -TESTPLANARTOB(I422, 2, 1, ABGR, 4) -TESTPLANARTOB(I422, 2, 1, RGBA, 4) -TESTPLANARTOB(I411, 4, 1, ARGB, 4) -TESTPLANARTOB(I444, 1, 1, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, YUY2, 2) -TESTPLANARTOB(I420, 2, 2, UYVY, 2) -// TODO(fbarchard): Re-enable test and fix valgrind. -// TESTPLANARTOB(I420, 2, 2, V210, 16 / 6) -TESTPLANARTOB(I420, 2, 2, I400, 1) -TESTPLANARTOB(I420, 2, 2, BayerBGGR, 1) -TESTPLANARTOB(I420, 2, 2, BayerRGGB, 1) -TESTPLANARTOB(I420, 2, 2, BayerGBRG, 1) -TESTPLANARTOB(I420, 2, 2, BayerGRBG, 1) - -#define TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, \ - N, NEG) \ -TEST_F(libyuvTest, FMT_PLANAR##To##FMT_B##N##_OptVsC) { \ - const int kWidth = 1280; \ - const int kHeight = 720; \ - align_buffer_16(src_y, kWidth * kHeight); \ - align_buffer_16(src_uv, kWidth / SUBSAMP_X * kHeight / SUBSAMP_Y * 2); \ - align_buffer_16(dst_argb_c, (kWidth * BPP_B) * kHeight); \ - align_buffer_16(dst_argb_opt, (kWidth * BPP_B) * kHeight); \ - srandom(time(NULL)); \ - for (int i = 0; i < kHeight; ++i) \ - for (int j = 0; j < kWidth; ++j) \ - src_y[(i * kWidth) + j] = (random() & 0xff); \ - for (int i = 0; i < kHeight / SUBSAMP_Y; ++i) \ - for (int j = 0; j < kWidth / SUBSAMP_X * 2; ++j) { \ - src_uv[(i * kWidth / SUBSAMP_X) * 2 + j] = (random() & 0xff); \ - } \ - MaskCpuFlags(kCpuInitialized); \ - FMT_PLANAR##To##FMT_B(src_y, kWidth, \ - src_uv, kWidth / SUBSAMP_X * 2, \ - dst_argb_c, kWidth * BPP_B, \ - kWidth, NEG kHeight); \ - MaskCpuFlags(-1); \ - for (int i = 0; i < benchmark_iterations_; ++i) { \ - FMT_PLANAR##To##FMT_B(src_y, kWidth, \ - src_uv, kWidth / SUBSAMP_X * 2, \ - dst_argb_opt, kWidth * BPP_B, \ - kWidth, NEG kHeight); \ - } \ - int max_diff = 0; \ - for (int i = 0; i < kHeight; ++i) { \ - for (int j = 0; j < kWidth * BPP_B; ++j) { \ - int abs_diff = \ - abs(static_cast<int>(dst_argb_c[i * kWidth * BPP_B + j]) - \ - static_cast<int>(dst_argb_opt[i * kWidth * BPP_B + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 3); \ - free_aligned_buffer_16(src_y) \ - free_aligned_buffer_16(src_uv) \ - free_aligned_buffer_16(dst_argb_c) \ - free_aligned_buffer_16(dst_argb_opt) \ -} - -#define TESTBIPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B) \ - TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, , +) \ - TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, Invert, -) - -TESTBIPLANARTOB(NV12, 2, 2, ARGB, 4) -TESTBIPLANARTOB(NV21, 2, 2, ARGB, 4) -TESTBIPLANARTOB(NV12, 2, 2, RGB565, 2) -TESTBIPLANARTOB(NV21, 2, 2, RGB565, 2) - -#define TESTATOPLANARI(FMT_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, N, NEG) \ -TEST_F(libyuvTest, FMT_A##To##FMT_PLANAR##N##_OptVsC) { \ - const int kWidth = 1280; \ - const int kHeight = 720; \ - const int kStride = (kWidth * 8 * BPP_A + 7) / 8; \ - align_buffer_16(src_argb, kStride * kHeight); \ - align_buffer_16(dst_y_c, kWidth * kHeight); \ - align_buffer_16(dst_u_c, kWidth / SUBSAMP_X * kHeight / SUBSAMP_Y); \ - align_buffer_16(dst_v_c, kWidth / SUBSAMP_X * kHeight / SUBSAMP_Y); \ - align_buffer_16(dst_y_opt, kWidth * kHeight); \ - align_buffer_16(dst_u_opt, kWidth / SUBSAMP_X * kHeight / SUBSAMP_Y); \ - align_buffer_16(dst_v_opt, kWidth / SUBSAMP_X * kHeight / SUBSAMP_Y); \ - srandom(time(NULL)); \ - for (int i = 0; i < kHeight; ++i) \ - for (int j = 0; j < kStride; ++j) \ - src_argb[(i * kStride) + j] = (random() & 0xff); \ - MaskCpuFlags(kCpuInitialized); \ - FMT_A##To##FMT_PLANAR(src_argb, kStride, \ - dst_y_c, kWidth, \ - dst_u_c, kWidth / SUBSAMP_X, \ - dst_v_c, kWidth / SUBSAMP_X, \ - kWidth, NEG kHeight); \ - MaskCpuFlags(-1); \ - for (int i = 0; i < benchmark_iterations_; ++i) { \ - FMT_A##To##FMT_PLANAR(src_argb, kStride, \ - dst_y_opt, kWidth, \ - dst_u_opt, kWidth / SUBSAMP_X, \ - dst_v_opt, kWidth / SUBSAMP_X, \ - kWidth, NEG kHeight); \ - } \ - int max_diff = 0; \ - for (int i = 0; i < kHeight; ++i) { \ - for (int j = 0; j < kWidth; ++j) { \ - int abs_diff = \ - abs(static_cast<int>(dst_y_c[i * kWidth + j]) - \ - static_cast<int>(dst_y_opt[i * kWidth + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 2); \ - for (int i = 0; i < kHeight / SUBSAMP_Y; ++i) { \ - for (int j = 0; j < kWidth / SUBSAMP_X; ++j) { \ - int abs_diff = \ - abs(static_cast<int>(dst_u_c[i * kWidth / SUBSAMP_X + j]) - \ - static_cast<int>(dst_u_opt[i * kWidth / SUBSAMP_X + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 2); \ - for (int i = 0; i < kHeight / SUBSAMP_Y; ++i) { \ - for (int j = 0; j < kWidth / SUBSAMP_X; ++j) { \ - int abs_diff = \ - abs(static_cast<int>(dst_v_c[i * kWidth / SUBSAMP_X + j]) - \ - static_cast<int>(dst_v_opt[i * kWidth / SUBSAMP_X + j])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - } \ - EXPECT_LE(max_diff, 2); \ - free_aligned_buffer_16(dst_y_c) \ - free_aligned_buffer_16(dst_u_c) \ - free_aligned_buffer_16(dst_v_c) \ - free_aligned_buffer_16(dst_y_opt) \ - free_aligned_buffer_16(dst_u_opt) \ - free_aligned_buffer_16(dst_v_opt) \ - free_aligned_buffer_16(src_argb) \ -} - -#define TESTATOPLANAR(FMT_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y) \ - TESTATOPLANARI(FMT_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, , +) \ - TESTATOPLANARI(FMT_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, Invert, -) - -TESTATOPLANAR(ARGB, 4, I420, 2, 2) -TESTATOPLANAR(BGRA, 4, I420, 2, 2) -TESTATOPLANAR(ABGR, 4, I420, 2, 2) -TESTATOPLANAR(RGBA, 4, I420, 2, 2) -TESTATOPLANAR(RAW, 3, I420, 2, 2) -TESTATOPLANAR(RGB24, 3, I420, 2, 2) -TESTATOPLANAR(RGB565, 2, I420, 2, 2) -TESTATOPLANAR(ARGB1555, 2, I420, 2, 2) -TESTATOPLANAR(ARGB4444, 2, I420, 2, 2) -// TESTATOPLANAR(ARGB, 4, I411, 4, 1) -TESTATOPLANAR(ARGB, 4, I422, 2, 1) -// TESTATOPLANAR(ARGB, 4, I444, 1, 1) -// TODO(fbarchard): Implement and test 411 and 444 -TESTATOPLANAR(YUY2, 2, I420, 2, 2) -TESTATOPLANAR(UYVY, 2, I420, 2, 2) -TESTATOPLANAR(YUY2, 2, I422, 2, 1) -TESTATOPLANAR(UYVY, 2, I422, 2, 1) -TESTATOPLANAR(V210, 16 / 6, I420, 2, 2) -TESTATOPLANAR(I400, 1, I420, 2, 2) -TESTATOPLANAR(BayerBGGR, 1, I420, 2, 2) -TESTATOPLANAR(BayerRGGB, 1, I420, 2, 2) -TESTATOPLANAR(BayerGBRG, 1, I420, 2, 2) -TESTATOPLANAR(BayerGRBG, 1, I420, 2, 2) - -#define TESTATOBI(FMT_A, BPP_A, STRIDE_A, FMT_B, BPP_B, N, NEG) \ -TEST_F(libyuvTest, FMT_A##To##FMT_B##N##_OptVsC) { \ - const int kWidth = 1280; \ - const int kHeight = 720; \ - align_buffer_16(src_argb, (kWidth * BPP_A) * kHeight); \ - align_buffer_16(dst_argb_c, (kWidth * BPP_B) * kHeight); \ - align_buffer_16(dst_argb_opt, (kWidth * BPP_B) * kHeight); \ - srandom(time(NULL)); \ - for (int i = 0; i < kHeight * kWidth * BPP_A; ++i) { \ - src_argb[i] = (random() & 0xff); \ - } \ - MaskCpuFlags(kCpuInitialized); \ - FMT_A##To##FMT_B(src_argb, kWidth * STRIDE_A, \ - dst_argb_c, kWidth * BPP_B, \ - kWidth, NEG kHeight); \ - MaskCpuFlags(-1); \ - for (int i = 0; i < benchmark_iterations_; ++i) { \ - FMT_A##To##FMT_B(src_argb, kWidth * STRIDE_A, \ - dst_argb_opt, kWidth * BPP_B, \ - kWidth, NEG kHeight); \ - } \ - int max_diff = 0; \ - for (int i = 0; i < kHeight * kWidth * BPP_B; ++i) { \ - int abs_diff = \ - abs(static_cast<int>(dst_argb_c[i]) - \ - static_cast<int>(dst_argb_opt[i])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - EXPECT_LE(max_diff, 2); \ - free_aligned_buffer_16(src_argb) \ - free_aligned_buffer_16(dst_argb_c) \ - free_aligned_buffer_16(dst_argb_opt) \ -} -#define TESTATOB(FMT_A, BPP_A, STRIDE_A, FMT_B, BPP_B) \ - TESTATOBI(FMT_A, BPP_A, STRIDE_A, FMT_B, BPP_B, , +) \ - TESTATOBI(FMT_A, BPP_A, STRIDE_A, FMT_B, BPP_B, Invert, -) - -TESTATOB(I400, 1, 1, I400, 1) -TESTATOB(ARGB, 4, 4, ARGB, 4) -TESTATOB(ARGB, 4, 4, BGRA, 4) -TESTATOB(ARGB, 4, 4, ABGR, 4) -TESTATOB(ARGB, 4, 4, RGBA, 4) -TESTATOB(ARGB, 4, 4, RAW, 3) -TESTATOB(ARGB, 4, 4, RGB24, 3) -TESTATOB(ARGB, 4, 4, RGB565, 2) -TESTATOB(ARGB, 4, 4, ARGB1555, 2) -TESTATOB(ARGB, 4, 4, ARGB4444, 2) -TESTATOB(BGRA, 4, 4, ARGB, 4) -TESTATOB(ABGR, 4, 4, ARGB, 4) -TESTATOB(RGBA, 4, 4, ARGB, 4) -TESTATOB(RAW, 3, 3, ARGB, 4) -TESTATOB(RGB24, 3, 3, ARGB, 4) -TESTATOB(RGB565, 2, 2, ARGB, 4) -TESTATOB(ARGB1555, 2, 2, ARGB, 4) -TESTATOB(ARGB4444, 2, 2, ARGB, 4) -TESTATOB(YUY2, 2, 2, ARGB, 4) -TESTATOB(UYVY, 2, 2, ARGB, 4) -TESTATOB(M420, 3 / 2, 1, ARGB, 4) - -static const int kReadPad = 16; // Allow overread of 16 bytes. -#define TESTATOBRANDOM(FMT_A, BPP_A, STRIDE_A, FMT_B, BPP_B) \ -TEST_F(libyuvTest, FMT_A##To##FMT_B##_Random) { \ - srandom(time(NULL)); \ - for (int times = 0; times < benchmark_iterations_; ++times) { \ - const int kWidth = (random() & 63) + 1; \ - const int kHeight = (random() & 31) + 1; \ - align_buffer_page_end(src_argb, (kWidth * BPP_A) * kHeight + kReadPad); \ - align_buffer_page_end(dst_argb_c, (kWidth * BPP_B) * kHeight); \ - align_buffer_page_end(dst_argb_opt, (kWidth * BPP_B) * kHeight); \ - for (int i = 0; i < kHeight * kWidth * BPP_A; ++i) { \ - src_argb[i] = (random() & 0xff); \ - } \ - MaskCpuFlags(kCpuInitialized); \ - FMT_A##To##FMT_B(src_argb, kWidth * STRIDE_A, \ - dst_argb_c, kWidth * BPP_B, \ - kWidth, kHeight); \ - MaskCpuFlags(-1); \ - FMT_A##To##FMT_B(src_argb, kWidth * STRIDE_A, \ - dst_argb_opt, kWidth * BPP_B, \ - kWidth, kHeight); \ - int max_diff = 0; \ - for (int i = 0; i < kHeight * kWidth * BPP_B; ++i) { \ - int abs_diff = \ - abs(static_cast<int>(dst_argb_c[i]) - \ - static_cast<int>(dst_argb_opt[i])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - EXPECT_LE(max_diff, 2); \ - free_aligned_buffer_page_end(src_argb) \ - free_aligned_buffer_page_end(dst_argb_c) \ - free_aligned_buffer_page_end(dst_argb_opt) \ - } \ -} - -TESTATOBRANDOM(ARGB, 4, 4, ARGB, 4) -TESTATOBRANDOM(ARGB, 4, 4, BGRA, 4) -TESTATOBRANDOM(ARGB, 4, 4, ABGR, 4) -TESTATOBRANDOM(ARGB, 4, 4, RGBA, 4) -TESTATOBRANDOM(ARGB, 4, 4, RAW, 3) -TESTATOBRANDOM(ARGB, 4, 4, RGB24, 3) -TESTATOBRANDOM(ARGB, 4, 4, RGB565, 2) -TESTATOBRANDOM(ARGB, 4, 4, ARGB1555, 2) -TESTATOBRANDOM(ARGB, 4, 4, ARGB4444, 2) - -TESTATOBRANDOM(BGRA, 4, 4, ARGB, 4) -TESTATOBRANDOM(ABGR, 4, 4, ARGB, 4) -TESTATOBRANDOM(RGBA, 4, 4, ARGB, 4) -TESTATOBRANDOM(RAW, 3, 3, ARGB, 4) -TESTATOBRANDOM(RGB24, 3, 3, ARGB, 4) -TESTATOBRANDOM(RGB565, 2, 2, ARGB, 4) -TESTATOBRANDOM(ARGB1555, 2, 2, ARGB, 4) -TESTATOBRANDOM(ARGB4444, 2, 2, ARGB, 4) - -TEST_F(libyuvTest, TestAttenuate) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); - SIMD_ALIGNED(uint8 atten_pixels[256][4]); - SIMD_ALIGNED(uint8 unatten_pixels[256][4]); - SIMD_ALIGNED(uint8 atten2_pixels[256][4]); +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][0] = 200u; - orig_pixels[0][1] = 129u; - orig_pixels[0][2] = 127u; - orig_pixels[0][3] = 128u; + 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][0] = 16u; - orig_pixels[1][1] = 64u; - orig_pixels[1][2] = 192u; - orig_pixels[1][3] = 0u; - orig_pixels[2][0] = 16u; - orig_pixels[2][1] = 64u; - orig_pixels[2][2] = 192u; - orig_pixels[2][3] = 255u; - orig_pixels[3][0] = 16u; - orig_pixels[3][1] = 64u; - orig_pixels[3][2] = 192u; - orig_pixels[3][3] = 128u; - ARGBUnattenuate(&orig_pixels[0][0], 0, &unatten_pixels[0][0], 0, 4, 1); - EXPECT_EQ(255u, unatten_pixels[0][0]); - EXPECT_EQ(255u, unatten_pixels[0][1]); - EXPECT_EQ(254u, unatten_pixels[0][2]); - EXPECT_EQ(128u, unatten_pixels[0][3]); - EXPECT_EQ(16u, unatten_pixels[1][0]); - EXPECT_EQ(64u, unatten_pixels[1][1]); - EXPECT_EQ(192u, unatten_pixels[1][2]); - EXPECT_EQ(0u, unatten_pixels[1][3]); - EXPECT_EQ(16u, unatten_pixels[2][0]); - EXPECT_EQ(64u, unatten_pixels[2][1]); - EXPECT_EQ(192u, unatten_pixels[2][2]); - EXPECT_EQ(255u, unatten_pixels[2][3]); - EXPECT_EQ(32u, unatten_pixels[3][0]); - EXPECT_EQ(128u, unatten_pixels[3][1]); - EXPECT_EQ(255u, unatten_pixels[3][2]); - EXPECT_EQ(128u, unatten_pixels[3][3]); - - for (int i = 0; i < 256; ++i) { - orig_pixels[i][0] = i; - orig_pixels[i][1] = i / 2; - orig_pixels[i][2] = i / 3; - orig_pixels[i][3] = i; + 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] = 255u; + orig_pixels[3 * 4 + 0] = 16u; + orig_pixels[3 * 4 + 1] = 64u; + orig_pixels[3 * 4 + 2] = 192u; + orig_pixels[3 * 4 + 3] = 128u; + ARGBUnattenuate(orig_pixels, 0, unatten_pixels, 0, 4, 1); + EXPECT_EQ(255u, unatten_pixels[0 * 4 + 0]); + EXPECT_EQ(255u, unatten_pixels[0 * 4 + 1]); + EXPECT_EQ(254u, unatten_pixels[0 * 4 + 2]); + EXPECT_EQ(128u, unatten_pixels[0 * 4 + 3]); + EXPECT_EQ(0u, unatten_pixels[1 * 4 + 0]); + EXPECT_EQ(0u, unatten_pixels[1 * 4 + 1]); + EXPECT_EQ(0u, unatten_pixels[1 * 4 + 2]); + EXPECT_EQ(0u, unatten_pixels[1 * 4 + 3]); + EXPECT_EQ(16u, unatten_pixels[2 * 4 + 0]); + EXPECT_EQ(64u, unatten_pixels[2 * 4 + 1]); + EXPECT_EQ(192u, unatten_pixels[2 * 4 + 2]); + EXPECT_EQ(255u, unatten_pixels[2 * 4 + 3]); + EXPECT_EQ(32u, unatten_pixels[3 * 4 + 0]); + EXPECT_EQ(128u, unatten_pixels[3 * 4 + 1]); + EXPECT_EQ(255u, unatten_pixels[3 * 4 + 2]); + EXPECT_EQ(128u, unatten_pixels[3 * 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][0], 0, &atten_pixels[0][0], 0, 256, 1); - ARGBUnattenuate(&atten_pixels[0][0], 0, &unatten_pixels[0][0], 0, 256, 1); - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { - ARGBAttenuate(&unatten_pixels[0][0], 0, &atten2_pixels[0][0], 0, 256, 1); + 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 < 256; ++i) { - EXPECT_NEAR(atten_pixels[i][0], atten2_pixels[i][0], 2); - EXPECT_NEAR(atten_pixels[i][1], atten2_pixels[i][1], 2); - EXPECT_NEAR(atten_pixels[i][2], atten2_pixels[i][2], 2); - EXPECT_NEAR(atten_pixels[i][3], atten2_pixels[i][3], 2); + for (int i = 0; i < 1280; ++i) { + EXPECT_NEAR(atten_pixels[i * 4 + 0], atten2_pixels[i * 4 + 0], 2); + EXPECT_NEAR(atten_pixels[i * 4 + 1], atten2_pixels[i * 4 + 1], 2); + EXPECT_NEAR(atten_pixels[i * 4 + 2], atten2_pixels[i * 4 + 2], 2); + EXPECT_NEAR(atten_pixels[i * 4 + 3], atten2_pixels[i * 4 + 3], 2); } // Make sure transparent, 50% and opaque are fully accurate. - EXPECT_EQ(0, atten_pixels[0][0]); - EXPECT_EQ(0, atten_pixels[0][1]); - EXPECT_EQ(0, atten_pixels[0][2]); - EXPECT_EQ(0, atten_pixels[0][3]); - EXPECT_EQ(64, atten_pixels[128][0]); - EXPECT_EQ(32, atten_pixels[128][1]); - EXPECT_EQ(21, atten_pixels[128][2]); - EXPECT_EQ(128, atten_pixels[128][3]); - EXPECT_EQ(255, atten_pixels[255][0]); - EXPECT_EQ(127, atten_pixels[255][1]); - EXPECT_EQ(85, atten_pixels[255][2]); - EXPECT_EQ(255, atten_pixels[255][3]); -} - -TEST_F(libyuvTest, TestARGBComputeCumulativeSum) { + EXPECT_EQ(0, atten_pixels[0 * 4 + 0]); + EXPECT_EQ(0, atten_pixels[0 * 4 + 1]); + EXPECT_EQ(0, atten_pixels[0 * 4 + 2]); + EXPECT_EQ(0, atten_pixels[0 * 4 + 3]); + EXPECT_EQ(64, atten_pixels[128 * 4 + 0]); + EXPECT_EQ(32, atten_pixels[128 * 4 + 1]); + EXPECT_EQ(21, atten_pixels[128 * 4 + 2]); + EXPECT_EQ(128, atten_pixels[128 * 4 + 3]); + EXPECT_NEAR(255, atten_pixels[255 * 4 + 0], 1); + EXPECT_NEAR(127, atten_pixels[255 * 4 + 1], 1); + EXPECT_NEAR(85, atten_pixels[255 * 4 + 2], 1); + EXPECT_EQ(255, atten_pixels[255 * 4 + 3]); + + free_aligned_buffer_page_end(atten2_pixels); + free_aligned_buffer_page_end(unatten_pixels); + free_aligned_buffer_page_end(atten_pixels); + free_aligned_buffer_page_end(orig_pixels); +} + +static int TestAttenuateI(int width, int height, int benchmark_iterations, + int disable_cpu_flags, int benchmark_cpu_info, + int invert, int off) { + if (width < 1) { + width = 1; + } + const int kBpp = 4; + const int kStride = width * kBpp; + align_buffer_page_end(src_argb, kStride * height + off); + align_buffer_page_end(dst_argb_c, kStride * height); + align_buffer_page_end(dst_argb_opt, kStride * height); + for (int i = 0; i < kStride * height; ++i) { + src_argb[i + off] = (fastrand() & 0xff); + } + memset(dst_argb_c, 0, kStride * height); + memset(dst_argb_opt, 0, kStride * height); + + MaskCpuFlags(disable_cpu_flags); + ARGBAttenuate(src_argb + off, kStride, + dst_argb_c, kStride, + width, invert * height); + MaskCpuFlags(benchmark_cpu_info); + for (int i = 0; i < benchmark_iterations; ++i) { + ARGBAttenuate(src_argb + off, kStride, + dst_argb_opt, kStride, + width, invert * height); + } + int max_diff = 0; + for (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb); + free_aligned_buffer_page_end(dst_argb_c); + free_aligned_buffer_page_end(dst_argb_opt); + return max_diff; +} + +TEST_F(LibYUVPlanarTest, ARGBAttenuate_Any) { + int max_diff = TestAttenuateI(benchmark_width_ - 1, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 0); + EXPECT_LE(max_diff, 2); +} + +TEST_F(LibYUVPlanarTest, ARGBAttenuate_Unaligned) { + int max_diff = TestAttenuateI(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 1); + EXPECT_LE(max_diff, 2); +} + +TEST_F(LibYUVPlanarTest, ARGBAttenuate_Invert) { + int max_diff = TestAttenuateI(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + -1, 0); + EXPECT_LE(max_diff, 2); +} + +TEST_F(LibYUVPlanarTest, ARGBAttenuate_Opt) { + int max_diff = TestAttenuateI(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 0); + EXPECT_LE(max_diff, 2); +} + +static int TestUnattenuateI(int width, int height, int benchmark_iterations, + int disable_cpu_flags, int benchmark_cpu_info, + int invert, int off) { + if (width < 1) { + width = 1; + } + const int kBpp = 4; + const int kStride = width * kBpp; + align_buffer_page_end(src_argb, kStride * height + off); + align_buffer_page_end(dst_argb_c, kStride * height); + align_buffer_page_end(dst_argb_opt, kStride * height); + for (int i = 0; i < kStride * height; ++i) { + src_argb[i + off] = (fastrand() & 0xff); + } + ARGBAttenuate(src_argb + off, kStride, + src_argb + off, kStride, + width, height); + memset(dst_argb_c, 0, kStride * height); + memset(dst_argb_opt, 0, kStride * height); + + MaskCpuFlags(disable_cpu_flags); + ARGBUnattenuate(src_argb + off, kStride, + dst_argb_c, kStride, + width, invert * height); + MaskCpuFlags(benchmark_cpu_info); + for (int i = 0; i < benchmark_iterations; ++i) { + ARGBUnattenuate(src_argb + off, kStride, + dst_argb_opt, kStride, + width, invert * height); + } + int max_diff = 0; + for (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb); + free_aligned_buffer_page_end(dst_argb_c); + free_aligned_buffer_page_end(dst_argb_opt); + return max_diff; +} + +TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Any) { + int max_diff = TestUnattenuateI(benchmark_width_ - 1, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 0); + EXPECT_LE(max_diff, 2); +} + +TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Unaligned) { + int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 1); + EXPECT_LE(max_diff, 2); +} + +TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Invert) { + int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + -1, 0); + EXPECT_LE(max_diff, 2); +} + +TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Opt) { + int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 0); + EXPECT_LE(max_diff, 2); +} + +TEST_F(LibYUVPlanarTest, TestARGBComputeCumulativeSum) { SIMD_ALIGNED(uint8 orig_pixels[16][16][4]); SIMD_ALIGNED(int32 added_pixels[16][16][4]); @@ -484,8 +281,9 @@ TEST_F(libyuvTest, TestARGBComputeCumulativeSum) { } } -TEST_F(libyuvTest, TestARGBGray) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestARGBGray) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); + memset(orig_pixels, 0, sizeof(orig_pixels)); // Test blue orig_pixels[0][0] = 255u; @@ -502,45 +300,62 @@ TEST_F(libyuvTest, TestARGBGray) { 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[3][0] = 16u; - orig_pixels[3][1] = 64u; - orig_pixels[3][2] = 192u; - orig_pixels[3][3] = 224u; + orig_pixels[5][0] = 16u; + orig_pixels[5][1] = 64u; + orig_pixels[5][2] = 192u; + orig_pixels[5][3] = 224u; // Do 16 to test asm version. ARGBGray(&orig_pixels[0][0], 0, 0, 0, 16, 1); - EXPECT_EQ(27u, orig_pixels[0][0]); - EXPECT_EQ(27u, orig_pixels[0][1]); - EXPECT_EQ(27u, orig_pixels[0][2]); + EXPECT_EQ(30u, orig_pixels[0][0]); + EXPECT_EQ(30u, orig_pixels[0][1]); + EXPECT_EQ(30u, orig_pixels[0][2]); EXPECT_EQ(128u, orig_pixels[0][3]); - EXPECT_EQ(151u, orig_pixels[1][0]); - EXPECT_EQ(151u, orig_pixels[1][1]); - EXPECT_EQ(151u, orig_pixels[1][2]); + EXPECT_EQ(149u, orig_pixels[1][0]); + EXPECT_EQ(149u, orig_pixels[1][1]); + EXPECT_EQ(149u, orig_pixels[1][2]); EXPECT_EQ(0u, orig_pixels[1][3]); - EXPECT_EQ(75u, orig_pixels[2][0]); - EXPECT_EQ(75u, orig_pixels[2][1]); - EXPECT_EQ(75u, orig_pixels[2][2]); + EXPECT_EQ(76u, orig_pixels[2][0]); + EXPECT_EQ(76u, orig_pixels[2][1]); + EXPECT_EQ(76u, orig_pixels[2][2]); EXPECT_EQ(255u, orig_pixels[2][3]); - EXPECT_EQ(96u, orig_pixels[3][0]); - EXPECT_EQ(96u, orig_pixels[3][1]); - EXPECT_EQ(96u, orig_pixels[3][2]); - EXPECT_EQ(224u, orig_pixels[3][3]); - - for (int i = 0; i < 256; ++i) { + EXPECT_EQ(0u, orig_pixels[3][0]); + EXPECT_EQ(0u, orig_pixels[3][1]); + EXPECT_EQ(0u, orig_pixels[3][2]); + EXPECT_EQ(255u, orig_pixels[3][3]); + EXPECT_EQ(255u, orig_pixels[4][0]); + EXPECT_EQ(255u, orig_pixels[4][1]); + EXPECT_EQ(255u, orig_pixels[4][2]); + EXPECT_EQ(255u, orig_pixels[4][3]); + EXPECT_EQ(96u, orig_pixels[5][0]); + EXPECT_EQ(96u, orig_pixels[5][1]); + EXPECT_EQ(96u, orig_pixels[5][2]); + EXPECT_EQ(224u, orig_pixels[5][3]); + for (int i = 0; i < 1280; ++i) { orig_pixels[i][0] = i; orig_pixels[i][1] = i / 2; orig_pixels[i][2] = i / 3; orig_pixels[i][3] = i; } - - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { - ARGBGray(&orig_pixels[0][0], 0, 0, 0, 256, 1); + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { + ARGBGray(&orig_pixels[0][0], 0, 0, 0, 1280, 1); } } -TEST_F(libyuvTest, TestARGBGrayTo) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); - SIMD_ALIGNED(uint8 gray_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestARGBGrayTo) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); + SIMD_ALIGNED(uint8 gray_pixels[1280][4]); + memset(orig_pixels, 0, sizeof(orig_pixels)); // Test blue orig_pixels[0][0] = 255u; @@ -557,44 +372,61 @@ TEST_F(libyuvTest, TestARGBGrayTo) { 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[3][0] = 16u; - orig_pixels[3][1] = 64u; - orig_pixels[3][2] = 192u; - orig_pixels[3][3] = 224u; + orig_pixels[5][0] = 16u; + orig_pixels[5][1] = 64u; + orig_pixels[5][2] = 192u; + orig_pixels[5][3] = 224u; // Do 16 to test asm version. ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 16, 1); - EXPECT_EQ(27u, gray_pixels[0][0]); - EXPECT_EQ(27u, gray_pixels[0][1]); - EXPECT_EQ(27u, gray_pixels[0][2]); + EXPECT_EQ(30u, gray_pixels[0][0]); + EXPECT_EQ(30u, gray_pixels[0][1]); + EXPECT_EQ(30u, gray_pixels[0][2]); EXPECT_EQ(128u, gray_pixels[0][3]); - EXPECT_EQ(151u, gray_pixels[1][0]); - EXPECT_EQ(151u, gray_pixels[1][1]); - EXPECT_EQ(151u, gray_pixels[1][2]); + EXPECT_EQ(149u, gray_pixels[1][0]); + EXPECT_EQ(149u, gray_pixels[1][1]); + EXPECT_EQ(149u, gray_pixels[1][2]); EXPECT_EQ(0u, gray_pixels[1][3]); - EXPECT_EQ(75u, gray_pixels[2][0]); - EXPECT_EQ(75u, gray_pixels[2][1]); - EXPECT_EQ(75u, gray_pixels[2][2]); + EXPECT_EQ(76u, gray_pixels[2][0]); + EXPECT_EQ(76u, gray_pixels[2][1]); + EXPECT_EQ(76u, gray_pixels[2][2]); EXPECT_EQ(255u, gray_pixels[2][3]); - EXPECT_EQ(96u, gray_pixels[3][0]); - EXPECT_EQ(96u, gray_pixels[3][1]); - EXPECT_EQ(96u, gray_pixels[3][2]); - EXPECT_EQ(224u, gray_pixels[3][3]); - - for (int i = 0; i < 256; ++i) { + EXPECT_EQ(0u, gray_pixels[3][0]); + EXPECT_EQ(0u, gray_pixels[3][1]); + EXPECT_EQ(0u, gray_pixels[3][2]); + EXPECT_EQ(255u, gray_pixels[3][3]); + EXPECT_EQ(255u, gray_pixels[4][0]); + EXPECT_EQ(255u, gray_pixels[4][1]); + EXPECT_EQ(255u, gray_pixels[4][2]); + EXPECT_EQ(255u, gray_pixels[4][3]); + EXPECT_EQ(96u, gray_pixels[5][0]); + EXPECT_EQ(96u, gray_pixels[5][1]); + EXPECT_EQ(96u, gray_pixels[5][2]); + EXPECT_EQ(224u, gray_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_iterations_ * 1280 * 720 / 256; ++i) { - ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 256, 1); + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { + ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 1280, 1); } } -TEST_F(libyuvTest, TestARGBSepia) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestARGBSepia) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); + memset(orig_pixels, 0, sizeof(orig_pixels)); // Test blue orig_pixels[0][0] = 255u; @@ -611,11 +443,21 @@ TEST_F(libyuvTest, TestARGBSepia) { 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[3][0] = 16u; - orig_pixels[3][1] = 64u; - orig_pixels[3][2] = 192u; - orig_pixels[3][3] = 224u; + orig_pixels[5][0] = 16u; + orig_pixels[5][1] = 64u; + orig_pixels[5][2] = 192u; + orig_pixels[5][3] = 224u; // Do 16 to test asm version. ARGBSepia(&orig_pixels[0][0], 0, 0, 0, 16, 1); EXPECT_EQ(33u, orig_pixels[0][0]); @@ -630,32 +472,119 @@ TEST_F(libyuvTest, TestARGBSepia) { EXPECT_EQ(89u, orig_pixels[2][1]); EXPECT_EQ(99u, orig_pixels[2][2]); EXPECT_EQ(255u, orig_pixels[2][3]); - EXPECT_EQ(88u, orig_pixels[3][0]); - EXPECT_EQ(114u, orig_pixels[3][1]); - EXPECT_EQ(127u, orig_pixels[3][2]); - EXPECT_EQ(224u, orig_pixels[3][3]); + EXPECT_EQ(0u, orig_pixels[3][0]); + EXPECT_EQ(0u, orig_pixels[3][1]); + EXPECT_EQ(0u, orig_pixels[3][2]); + EXPECT_EQ(255u, orig_pixels[3][3]); + EXPECT_EQ(239u, orig_pixels[4][0]); + EXPECT_EQ(255u, orig_pixels[4][1]); + EXPECT_EQ(255u, orig_pixels[4][2]); + EXPECT_EQ(255u, orig_pixels[4][3]); + EXPECT_EQ(88u, orig_pixels[5][0]); + EXPECT_EQ(114u, orig_pixels[5][1]); + EXPECT_EQ(127u, orig_pixels[5][2]); + EXPECT_EQ(224u, orig_pixels[5][3]); + + for (int i = 0; i < 1280; ++i) { + orig_pixels[i][0] = i; + orig_pixels[i][1] = i / 2; + orig_pixels[i][2] = i / 3; + orig_pixels[i][3] = i; + } + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { + ARGBSepia(&orig_pixels[0][0], 0, 0, 0, 1280, 1); + } +} + +TEST_F(LibYUVPlanarTest, TestARGBColorMatrix) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); + SIMD_ALIGNED(uint8 dst_pixels_opt[1280][4]); + SIMD_ALIGNED(uint8 dst_pixels_c[1280][4]); + + // Matrix for Sepia. + SIMD_ALIGNED(static const int8 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)); - for (int i = 0; i < 256; ++i) { + // Test blue + orig_pixels[0][0] = 255u; + orig_pixels[0][1] = 0u; + orig_pixels[0][2] = 0u; + orig_pixels[0][3] = 128u; + // Test green + orig_pixels[1][0] = 0u; + orig_pixels[1][1] = 255u; + orig_pixels[1][2] = 0u; + orig_pixels[1][3] = 0u; + // Test red + orig_pixels[2][0] = 0u; + orig_pixels[2][1] = 0u; + orig_pixels[2][2] = 255u; + orig_pixels[2][3] = 255u; + // Test color + orig_pixels[3][0] = 16u; + orig_pixels[3][1] = 64u; + orig_pixels[3][2] = 192u; + orig_pixels[3][3] = 224u; + // Do 16 to test asm version. + ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, + &kRGBToSepia[0], 16, 1); + EXPECT_EQ(31u, dst_pixels_opt[0][0]); + EXPECT_EQ(43u, dst_pixels_opt[0][1]); + EXPECT_EQ(47u, dst_pixels_opt[0][2]); + EXPECT_EQ(128u, dst_pixels_opt[0][3]); + EXPECT_EQ(135u, dst_pixels_opt[1][0]); + EXPECT_EQ(175u, dst_pixels_opt[1][1]); + EXPECT_EQ(195u, dst_pixels_opt[1][2]); + EXPECT_EQ(0u, dst_pixels_opt[1][3]); + EXPECT_EQ(67u, dst_pixels_opt[2][0]); + EXPECT_EQ(87u, dst_pixels_opt[2][1]); + EXPECT_EQ(99u, dst_pixels_opt[2][2]); + EXPECT_EQ(255u, dst_pixels_opt[2][3]); + EXPECT_EQ(87u, dst_pixels_opt[3][0]); + EXPECT_EQ(112u, dst_pixels_opt[3][1]); + EXPECT_EQ(127u, dst_pixels_opt[3][2]); + EXPECT_EQ(224u, dst_pixels_opt[3][3]); + + for (int i = 0; i < 1280; ++i) { orig_pixels[i][0] = i; orig_pixels[i][1] = i / 2; orig_pixels[i][2] = i / 3; orig_pixels[i][3] = i; } + MaskCpuFlags(disable_cpu_flags_); + ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_c[0][0], 0, + &kRGBToSepia[0], 1280, 1); + MaskCpuFlags(benchmark_cpu_info_); + + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { + ARGBColorMatrix(&orig_pixels[0][0], 0, &dst_pixels_opt[0][0], 0, + &kRGBToSepia[0], 1280, 1); + } - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { - ARGBSepia(&orig_pixels[0][0], 0, 0, 0, 256, 1); + for (int i = 0; i < 1280; ++i) { + EXPECT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); + EXPECT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); + EXPECT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); + EXPECT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]); } } -TEST_F(libyuvTest, TestARGBColorMatrix) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestRGBColorMatrix) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); // Matrix for Sepia. - static const int8 kARGBToSepia[] = { + SIMD_ALIGNED(static const int8 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; @@ -678,8 +607,8 @@ TEST_F(libyuvTest, TestARGBColorMatrix) { orig_pixels[3][2] = 192u; orig_pixels[3][3] = 224u; // Do 16 to test asm version. - ARGBColorMatrix(&orig_pixels[0][0], 0, &kARGBToSepia[0], 0, 0, 16, 1); - EXPECT_EQ(33u, orig_pixels[0][0]); + RGBColorMatrix(&orig_pixels[0][0], 0, &kRGBToSepia[0], 0, 0, 16, 1); + EXPECT_EQ(31u, orig_pixels[0][0]); EXPECT_EQ(43u, orig_pixels[0][1]); EXPECT_EQ(47u, orig_pixels[0][2]); EXPECT_EQ(128u, orig_pixels[0][3]); @@ -687,29 +616,28 @@ TEST_F(libyuvTest, TestARGBColorMatrix) { EXPECT_EQ(175u, orig_pixels[1][1]); EXPECT_EQ(195u, orig_pixels[1][2]); EXPECT_EQ(0u, orig_pixels[1][3]); - EXPECT_EQ(69u, orig_pixels[2][0]); - EXPECT_EQ(89u, orig_pixels[2][1]); + EXPECT_EQ(67u, orig_pixels[2][0]); + EXPECT_EQ(87u, orig_pixels[2][1]); EXPECT_EQ(99u, orig_pixels[2][2]); EXPECT_EQ(255u, orig_pixels[2][3]); - EXPECT_EQ(88u, orig_pixels[3][0]); - EXPECT_EQ(114u, orig_pixels[3][1]); + EXPECT_EQ(87u, orig_pixels[3][0]); + EXPECT_EQ(112u, orig_pixels[3][1]); EXPECT_EQ(127u, orig_pixels[3][2]); EXPECT_EQ(224u, orig_pixels[3][3]); - for (int i = 0; i < 256; ++i) { + 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_iterations_ * 1280 * 720 / 256; ++i) { - ARGBColorMatrix(&orig_pixels[0][0], 0, &kARGBToSepia[0], 0, 0, 256, 1); + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { + RGBColorMatrix(&orig_pixels[0][0], 0, &kRGBToSepia[0], 0, 0, 1280, 1); } } -TEST_F(libyuvTest, TestARGBColorTable) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestARGBColorTable) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); memset(orig_pixels, 0, sizeof(orig_pixels)); // Matrix for Sepia. @@ -755,68 +683,127 @@ TEST_F(libyuvTest, TestARGBColorTable) { EXPECT_EQ(11u, orig_pixels[3][2]); EXPECT_EQ(16u, orig_pixels[3][3]); - for (int i = 0; i < 256; ++i) { + 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 orig_pixels[1280][4]); + memset(orig_pixels, 0, sizeof(orig_pixels)); + + // Matrix for Sepia. + static const uint8 kARGBTable[256 * 4] = { + 1u, 2u, 3u, 4u, + 5u, 6u, 7u, 8u, + 9u, 10u, 11u, 12u, + 13u, 14u, 15u, 16u, + }; + + orig_pixels[0][0] = 0u; + orig_pixels[0][1] = 0u; + orig_pixels[0][2] = 0u; + orig_pixels[0][3] = 0u; + orig_pixels[1][0] = 1u; + orig_pixels[1][1] = 1u; + orig_pixels[1][2] = 1u; + orig_pixels[1][3] = 1u; + orig_pixels[2][0] = 2u; + orig_pixels[2][1] = 2u; + orig_pixels[2][2] = 2u; + orig_pixels[2][3] = 2u; + orig_pixels[3][0] = 0u; + orig_pixels[3][1] = 1u; + orig_pixels[3][2] = 2u; + orig_pixels[3][3] = 3u; + // Do 16 to test asm version. + RGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 16, 1); + EXPECT_EQ(1u, orig_pixels[0][0]); + EXPECT_EQ(2u, orig_pixels[0][1]); + EXPECT_EQ(3u, orig_pixels[0][2]); + EXPECT_EQ(0u, orig_pixels[0][3]); // Alpha unchanged. + EXPECT_EQ(5u, orig_pixels[1][0]); + EXPECT_EQ(6u, orig_pixels[1][1]); + EXPECT_EQ(7u, orig_pixels[1][2]); + EXPECT_EQ(1u, orig_pixels[1][3]); // Alpha unchanged. + EXPECT_EQ(9u, orig_pixels[2][0]); + EXPECT_EQ(10u, orig_pixels[2][1]); + EXPECT_EQ(11u, orig_pixels[2][2]); + EXPECT_EQ(2u, orig_pixels[2][3]); // Alpha unchanged. + EXPECT_EQ(1u, orig_pixels[3][0]); + EXPECT_EQ(6u, orig_pixels[3][1]); + EXPECT_EQ(11u, orig_pixels[3][2]); + EXPECT_EQ(3u, orig_pixels[3][3]); // Alpha unchanged. - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { - ARGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 256, 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) { + RGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 1280, 1); } } -TEST_F(libyuvTest, TestARGBQuantize) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestARGBQuantize) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); - for (int i = 0; i < 256; ++i) { + 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, 256, 1); + (65536 + (8 / 2)) / 8, 8, 8 / 2, 0, 0, 1280, 1); - for (int i = 0; i < 256; ++i) { - EXPECT_EQ(i / 8 * 8 + 8 / 2, orig_pixels[i][0]); - EXPECT_EQ(i / 2 / 8 * 8 + 8 / 2, orig_pixels[i][1]); - EXPECT_EQ(i / 3 / 8 * 8 + 8 / 2, orig_pixels[i][2]); - EXPECT_EQ(i, orig_pixels[i][3]); + for (int i = 0; i < 1280; ++i) { + EXPECT_EQ((i / 8 * 8 + 8 / 2) & 255, orig_pixels[i][0]); + EXPECT_EQ((i / 2 / 8 * 8 + 8 / 2) & 255, orig_pixels[i][1]); + EXPECT_EQ((i / 3 / 8 * 8 + 8 / 2) & 255, orig_pixels[i][2]); + EXPECT_EQ(i & 255, orig_pixels[i][3]); } - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { ARGBQuantize(&orig_pixels[0][0], 0, - (65536 + (8 / 2)) / 8, 8, 8 / 2, 0, 0, 256, 1); + (65536 + (8 / 2)) / 8, 8, 8 / 2, 0, 0, 1280, 1); } } -TEST_F(libyuvTest, TestARGBMirror) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); - SIMD_ALIGNED(uint8 dst_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestARGBMirror) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); + SIMD_ALIGNED(uint8 dst_pixels[1280][4]); - for (int i = 0; i < 256; ++i) { + 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 / 4; } - ARGBMirror(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0, 256, 1); + ARGBMirror(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0, 1280, 1); - for (int i = 0; i < 256; ++i) { - EXPECT_EQ(i, dst_pixels[255 - i][0]); - EXPECT_EQ(i / 2, dst_pixels[255 - i][1]); - EXPECT_EQ(i / 3, dst_pixels[255 - i][2]); - EXPECT_EQ(i / 4, dst_pixels[255 - i][3]); + for (int i = 0; i < 1280; ++i) { + EXPECT_EQ(i & 255, dst_pixels[1280 - 1 - i][0]); + EXPECT_EQ((i / 2) & 255, dst_pixels[1280 - 1 - i][1]); + EXPECT_EQ((i / 3) & 255, dst_pixels[1280 - 1 - i][2]); + EXPECT_EQ((i / 4) & 255, dst_pixels[1280 - 1 - i][3]); } - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { - ARGBMirror(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0, 256, 1); + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { + ARGBMirror(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0, 1280, 1); } } -TEST_F(libyuvTest, TestShade) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); - SIMD_ALIGNED(uint8 shade_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestShade) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); + SIMD_ALIGNED(uint8 shade_pixels[1280][4]); + memset(orig_pixels, 0, sizeof(orig_pixels)); orig_pixels[0][0] = 10u; orig_pixels[0][1] = 20u; @@ -834,7 +821,8 @@ TEST_F(libyuvTest, TestShade) { orig_pixels[3][1] = 0u; orig_pixels[3][2] = 0u; orig_pixels[3][3] = 0u; - ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 4, 1, 0x80ffffff); + // Do 8 pixels to allow opt version to be used. + ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x80ffffff); EXPECT_EQ(10u, shade_pixels[0][0]); EXPECT_EQ(20u, shade_pixels[0][1]); EXPECT_EQ(40u, shade_pixels[0][2]); @@ -852,22 +840,30 @@ TEST_F(libyuvTest, TestShade) { EXPECT_EQ(0u, shade_pixels[3][2]); EXPECT_EQ(0u, shade_pixels[3][3]); - ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 4, 1, 0x80808080); + ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x80808080); EXPECT_EQ(5u, shade_pixels[0][0]); EXPECT_EQ(10u, shade_pixels[0][1]); EXPECT_EQ(20u, shade_pixels[0][2]); EXPECT_EQ(40u, shade_pixels[0][3]); - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { - ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 256, 1, + ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 8, 1, 0x10204080); + EXPECT_EQ(5u, shade_pixels[0][0]); + EXPECT_EQ(5u, shade_pixels[0][1]); + EXPECT_EQ(5u, shade_pixels[0][2]); + EXPECT_EQ(5u, shade_pixels[0][3]); + + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { + ARGBShade(&orig_pixels[0][0], 0, &shade_pixels[0][0], 0, 1280, 1, 0x80808080); } } -TEST_F(libyuvTest, TestInterpolate) { - SIMD_ALIGNED(uint8 orig_pixels_0[256][4]); - SIMD_ALIGNED(uint8 orig_pixels_1[256][4]); - SIMD_ALIGNED(uint8 interpolate_pixels[256][4]); +TEST_F(LibYUVPlanarTest, TestARGBInterpolate) { + SIMD_ALIGNED(uint8 orig_pixels_0[1280][4]); + SIMD_ALIGNED(uint8 orig_pixels_1[1280][4]); + SIMD_ALIGNED(uint8 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; @@ -912,15 +908,15 @@ TEST_F(libyuvTest, TestInterpolate) { EXPECT_EQ(0u, interpolate_pixels[1][0]); EXPECT_EQ(0u, interpolate_pixels[1][1]); EXPECT_EQ(0u, interpolate_pixels[1][2]); - EXPECT_NEAR(128u, interpolate_pixels[1][3], 1); // C = 127, SSE = 128. + EXPECT_EQ(128u, interpolate_pixels[1][3]); EXPECT_EQ(0u, interpolate_pixels[2][0]); EXPECT_EQ(0u, interpolate_pixels[2][1]); EXPECT_EQ(0u, interpolate_pixels[2][2]); EXPECT_EQ(0u, interpolate_pixels[2][3]); - EXPECT_NEAR(128u, interpolate_pixels[3][0], 1); - EXPECT_NEAR(128u, interpolate_pixels[3][1], 1); - EXPECT_NEAR(128u, interpolate_pixels[3][2], 1); - EXPECT_NEAR(128u, interpolate_pixels[3][3], 1); + EXPECT_EQ(128u, interpolate_pixels[3][0]); + EXPECT_EQ(128u, interpolate_pixels[3][1]); + EXPECT_EQ(128u, interpolate_pixels[3][2]); + EXPECT_EQ(128u, interpolate_pixels[3][3]); ARGBInterpolate(&orig_pixels_0[0][0], 0, &orig_pixels_1[0][0], 0, &interpolate_pixels[0][0], 0, 4, 1, 0); @@ -937,20 +933,418 @@ TEST_F(libyuvTest, TestInterpolate) { EXPECT_EQ(16u, interpolate_pixels[0][2]); EXPECT_EQ(32u, interpolate_pixels[0][3]); - for (int i = 0; i < benchmark_iterations_ * (1280 * 720 / 256); ++i) { + 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, 256, 1, 128); + &interpolate_pixels[0][0], 0, 1280, 1, 128); } } -TEST_F(libyuvTest, TestAffine) { - SIMD_ALIGNED(uint8 orig_pixels_0[256][4]); - SIMD_ALIGNED(uint8 interpolate_pixels_C[256][4]); -#if defined(HAS_ARGBAFFINEROW_SSE2) - SIMD_ALIGNED(uint8 interpolate_pixels_Opt[256][4]); -#endif +TEST_F(LibYUVPlanarTest, TestInterpolatePlane) { + SIMD_ALIGNED(uint8 orig_pixels_0[1280]); + SIMD_ALIGNED(uint8 orig_pixels_1[1280]); + SIMD_ALIGNED(uint8 interpolate_pixels[1280]); + memset(orig_pixels_0, 0, sizeof(orig_pixels_0)); + memset(orig_pixels_1, 0, sizeof(orig_pixels_1)); + + orig_pixels_0[0] = 16u; + orig_pixels_0[1] = 32u; + orig_pixels_0[2] = 64u; + orig_pixels_0[3] = 128u; + orig_pixels_0[4] = 0u; + orig_pixels_0[5] = 0u; + orig_pixels_0[6] = 0u; + orig_pixels_0[7] = 255u; + orig_pixels_0[8] = 0u; + orig_pixels_0[9] = 0u; + orig_pixels_0[10] = 0u; + orig_pixels_0[11] = 0u; + orig_pixels_0[12] = 0u; + orig_pixels_0[13] = 0u; + orig_pixels_0[14] = 0u; + orig_pixels_0[15] = 0u; + + orig_pixels_1[0] = 0u; + orig_pixels_1[1] = 0u; + orig_pixels_1[2] = 0u; + orig_pixels_1[3] = 0u; + orig_pixels_1[4] = 0u; + orig_pixels_1[5] = 0u; + orig_pixels_1[6] = 0u; + orig_pixels_1[7] = 0u; + orig_pixels_1[8] = 0u; + orig_pixels_1[9] = 0u; + orig_pixels_1[10] = 0u; + orig_pixels_1[11] = 0u; + orig_pixels_1[12] = 255u; + orig_pixels_1[13] = 255u; + orig_pixels_1[14] = 255u; + orig_pixels_1[15] = 255u; + + InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0, + &interpolate_pixels[0], 0, 16, 1, 128); + EXPECT_EQ(8u, interpolate_pixels[0]); + EXPECT_EQ(16u, interpolate_pixels[1]); + EXPECT_EQ(32u, interpolate_pixels[2]); + EXPECT_EQ(64u, interpolate_pixels[3]); + EXPECT_EQ(0u, interpolate_pixels[4]); + EXPECT_EQ(0u, interpolate_pixels[5]); + EXPECT_EQ(0u, interpolate_pixels[6]); + EXPECT_EQ(128u, interpolate_pixels[7]); + EXPECT_EQ(0u, interpolate_pixels[8]); + EXPECT_EQ(0u, interpolate_pixels[9]); + EXPECT_EQ(0u, interpolate_pixels[10]); + EXPECT_EQ(0u, interpolate_pixels[11]); + EXPECT_EQ(128u, interpolate_pixels[12]); + EXPECT_EQ(128u, interpolate_pixels[13]); + EXPECT_EQ(128u, interpolate_pixels[14]); + EXPECT_EQ(128u, interpolate_pixels[15]); + + InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0, + &interpolate_pixels[0], 0, 16, 1, 0); + EXPECT_EQ(16u, interpolate_pixels[0]); + EXPECT_EQ(32u, interpolate_pixels[1]); + EXPECT_EQ(64u, interpolate_pixels[2]); + EXPECT_EQ(128u, interpolate_pixels[3]); + + InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0, + &interpolate_pixels[0], 0, 16, 1, 192); + + EXPECT_EQ(4u, interpolate_pixels[0]); + EXPECT_EQ(8u, interpolate_pixels[1]); + EXPECT_EQ(16u, interpolate_pixels[2]); + EXPECT_EQ(32u, interpolate_pixels[3]); + + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { + InterpolatePlane(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0, + &interpolate_pixels[0], 0, 1280, 1, 123); + } +} + +#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) > 0) ? (W1280) : 1; \ + const int kHeight = benchmark_height_; \ + const int kStrideA = (kWidth * BPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A; \ + const int kStrideB = (kWidth * BPP_B + STRIDE_B - 1) / STRIDE_B * STRIDE_B; \ + align_buffer_page_end(src_argb_a, kStrideA * kHeight + OFF); \ + align_buffer_page_end(src_argb_b, kStrideA * kHeight + OFF); \ + align_buffer_page_end(dst_argb_c, kStrideB * kHeight); \ + align_buffer_page_end(dst_argb_opt, kStrideB * kHeight); \ + for (int i = 0; i < kStrideA * kHeight; ++i) { \ + src_argb_a[i + OFF] = (fastrand() & 0xff); \ + src_argb_b[i + OFF] = (fastrand() & 0xff); \ + } \ + MaskCpuFlags(disable_cpu_flags_); \ + ARGBInterpolate(src_argb_a + OFF, kStrideA, \ + src_argb_b + OFF, kStrideA, \ + dst_argb_c, kStrideB, \ + kWidth, NEG kHeight, TERP); \ + MaskCpuFlags(benchmark_cpu_info_); \ + for (int i = 0; i < benchmark_iterations_; ++i) { \ + ARGBInterpolate(src_argb_a + OFF, kStrideA, \ + src_argb_b + OFF, kStrideA, \ + dst_argb_opt, kStrideB, \ + kWidth, NEG kHeight, TERP); \ + } \ + for (int i = 0; i < kStrideB * kHeight; ++i) { \ + EXPECT_EQ(dst_argb_c[i], dst_argb_opt[i]); \ + } \ + free_aligned_buffer_page_end(src_argb_a); \ + free_aligned_buffer_page_end(src_argb_b); \ + free_aligned_buffer_page_end(dst_argb_c); \ + free_aligned_buffer_page_end(dst_argb_opt); \ +} + +#define TESTINTERPOLATE(TERP) \ + TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_ - 1, TERP, _Any, +, 0) \ + TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Unaligned, +, 1) \ + TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Invert, -, 0) \ + TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Opt, +, 0) + +TESTINTERPOLATE(0) +TESTINTERPOLATE(64) +TESTINTERPOLATE(128) +TESTINTERPOLATE(192) +TESTINTERPOLATE(255) + +static int TestBlend(int width, int height, int benchmark_iterations, + int disable_cpu_flags, int benchmark_cpu_info, + int invert, int off) { + 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); + } + ARGBAttenuate(src_argb_a + off, kStride, src_argb_a + off, kStride, width, + height); + ARGBAttenuate(src_argb_b + off, kStride, src_argb_b + off, kStride, width, + height); + memset(dst_argb_c, 255, kStride * height); + memset(dst_argb_opt, 255, kStride * height); + + MaskCpuFlags(disable_cpu_flags); + ARGBBlend(src_argb_a + off, kStride, + src_argb_b + off, kStride, + dst_argb_c, kStride, + width, invert * height); + MaskCpuFlags(benchmark_cpu_info); + for (int i = 0; i < benchmark_iterations; ++i) { + ARGBBlend(src_argb_a + off, kStride, + src_argb_b + off, kStride, + dst_argb_opt, kStride, + width, invert * height); + } + int max_diff = 0; + for (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb_a); + free_aligned_buffer_page_end(src_argb_b); + free_aligned_buffer_page_end(dst_argb_c); + free_aligned_buffer_page_end(dst_argb_opt); + return max_diff; +} + +TEST_F(LibYUVPlanarTest, ARGBBlend_Any) { + int max_diff = TestBlend(benchmark_width_ - 4, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 0); + EXPECT_LE(max_diff, 1); +} - for (int i = 0; i < 256; ++i) { +TEST_F(LibYUVPlanarTest, ARGBBlend_Unaligned) { + int max_diff = TestBlend(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 1); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBBlend_Invert) { + int max_diff = TestBlend(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, -1, 0); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBBlend_Opt) { + int max_diff = TestBlend(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 0); + EXPECT_LE(max_diff, 1); +} + +static void TestBlendPlane(int width, int height, int benchmark_iterations, + int disable_cpu_flags, int benchmark_cpu_info, + int invert, int off) { + if (width < 1) { + width = 1; + } + const int kBpp = 1; + const int kStride = width * kBpp; + align_buffer_page_end(src_argb_a, kStride * height + off); + align_buffer_page_end(src_argb_b, kStride * height + off); + align_buffer_page_end(src_argb_alpha, kStride * height + off); + align_buffer_page_end(dst_argb_c, kStride * height + off); + align_buffer_page_end(dst_argb_opt, kStride * height + off); + memset(dst_argb_c, 255, kStride * height + off); + memset(dst_argb_opt, 255, kStride * height + off); + + // Test source is maintained exactly if alpha is 255. + for (int i = 0; i < width; ++i) { + src_argb_a[i + off] = i & 255; + src_argb_b[i + off] = 255 - (i & 255); + } + memset(src_argb_alpha + off, 255, width); + BlendPlane(src_argb_a + off, width, + src_argb_b + off, width, + src_argb_alpha + off, width, + dst_argb_opt + off, width, + width, 1); + for (int i = 0; i < width; ++i) { + EXPECT_EQ(src_argb_a[i + off], dst_argb_opt[i + off]); + } + // Test destination is maintained exactly if alpha is 0. + memset(src_argb_alpha + off, 0, width); + BlendPlane(src_argb_a + off, width, + src_argb_b + off, width, + src_argb_alpha + off, width, + dst_argb_opt + off, width, + width, 1); + for (int i = 0; i < width; ++i) { + EXPECT_EQ(src_argb_b[i + off], dst_argb_opt[i + off]); + } + for (int i = 0; i < kStride * height; ++i) { + src_argb_a[i + off] = (fastrand() & 0xff); + src_argb_b[i + off] = (fastrand() & 0xff); + src_argb_alpha[i + off] = (fastrand() & 0xff); + } + + MaskCpuFlags(disable_cpu_flags); + BlendPlane(src_argb_a + off, width, + src_argb_b + off, width, + src_argb_alpha + off, width, + dst_argb_c + off, width, + width, 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, height); + } + for (int i = 0; i < kStride * height; ++i) { + EXPECT_EQ(dst_argb_c[i + off], dst_argb_opt[i + off]); + } + free_aligned_buffer_page_end(src_argb_a); + free_aligned_buffer_page_end(src_argb_b); + free_aligned_buffer_page_end(src_argb_alpha); + free_aligned_buffer_page_end(dst_argb_c); + free_aligned_buffer_page_end(dst_argb_opt); + return; +} + +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_ - 4, 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, 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, height); + } + for (int i = 0; i < width * height; ++i) { + EXPECT_EQ(dst_y_c[i + off], dst_y_opt[i + off]); + } + for (int i = 0; i < kSizeUV; ++i) { + EXPECT_EQ(dst_u_c[i + off], dst_u_opt[i + off]); + EXPECT_EQ(dst_v_c[i + off], dst_v_opt[i + off]); + } + free_aligned_buffer_page_end(src_y0); + free_aligned_buffer_page_end(src_u0); + free_aligned_buffer_page_end(src_v0); + free_aligned_buffer_page_end(src_y1); + free_aligned_buffer_page_end(src_u1); + free_aligned_buffer_page_end(src_v1); + free_aligned_buffer_page_end(src_a); + free_aligned_buffer_page_end(dst_y_c); + free_aligned_buffer_page_end(dst_u_c); + free_aligned_buffer_page_end(dst_v_c); + free_aligned_buffer_page_end(dst_y_opt); + free_aligned_buffer_page_end(dst_u_opt); + free_aligned_buffer_page_end(dst_v_opt); + return; +} + +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_ - 4, benchmark_height_, benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 0); +} +TEST_F(LibYUVPlanarTest, I420Blend_Invert) { + TestI420Blend(benchmark_width_, benchmark_height_, benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, -1, 0); +} + +TEST_F(LibYUVPlanarTest, TestAffine) { + SIMD_ALIGNED(uint8 orig_pixels_0[1280][4]); + SIMD_ALIGNED(uint8 interpolate_pixels_C[1280][4]); + + for (int i = 0; i < 1280; ++i) { for (int j = 0; j < 4; ++j) { orig_pixels_0[i][j] = i; } @@ -959,47 +1353,1009 @@ TEST_F(libyuvTest, TestAffine) { float uv_step[4] = { 0.f, 0.f, 0.75f, 0.f }; ARGBAffineRow_C(&orig_pixels_0[0][0], 0, &interpolate_pixels_C[0][0], - uv_step, 256); + uv_step, 1280); EXPECT_EQ(0u, interpolate_pixels_C[0][0]); EXPECT_EQ(96u, interpolate_pixels_C[128][0]); EXPECT_EQ(191u, interpolate_pixels_C[255][3]); #if defined(HAS_ARGBAFFINEROW_SSE2) + SIMD_ALIGNED(uint8 interpolate_pixels_Opt[1280][4]); ARGBAffineRow_SSE2(&orig_pixels_0[0][0], 0, &interpolate_pixels_Opt[0][0], - uv_step, 256); - EXPECT_EQ(0, memcmp(interpolate_pixels_Opt, interpolate_pixels_C, 256 * 4)); -#endif + uv_step, 1280); + EXPECT_EQ(0, memcmp(interpolate_pixels_Opt, interpolate_pixels_C, 1280 * 4)); -#if defined(HAS_ARGBAFFINEROW_SSE2) int has_sse2 = TestCpuFlag(kCpuHasSSE2); if (has_sse2) { - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { + for (int i = 0; i < benchmark_pixels_div1280_; ++i) { ARGBAffineRow_SSE2(&orig_pixels_0[0][0], 0, &interpolate_pixels_Opt[0][0], - uv_step, 256); + uv_step, 1280); } - } else { + } #endif - for (int i = 0; i < benchmark_iterations_ * 1280 * 720 / 256; ++i) { - ARGBAffineRow_C(&orig_pixels_0[0][0], 0, &interpolate_pixels_C[0][0], - uv_step, 256); +} + +TEST_F(LibYUVPlanarTest, TestCopyPlane) { + int err = 0; + int yw = benchmark_width_; + int yh = benchmark_height_; + int b = 12; + int i, j; + + int y_plane_size = (yw + b * 2) * (yh + b * 2); + align_buffer_page_end(orig_y, y_plane_size); + align_buffer_page_end(dst_c, y_plane_size); + align_buffer_page_end(dst_opt, y_plane_size); + + memset(orig_y, 0, y_plane_size); + memset(dst_c, 0, y_plane_size); + memset(dst_opt, 0, y_plane_size); + + // Fill image buffers with random data. + for (i = b; i < (yh + b); ++i) { + for (j = b; j < (yw + b); ++j) { + orig_y[i * (yw + b * 2) + j] = fastrand() & 0xff; } -#if defined(HAS_ARGBAFFINEROW_SSE2) } -#endif + + // Fill destination buffers with random data. + for (i = 0; i < y_plane_size; ++i) { + uint8 random_number = fastrand() & 0x7f; + dst_c[i] = random_number; + dst_opt[i] = dst_c[i]; + } + + int y_off = b * (yw + b * 2) + b; + + int y_st = yw + b * 2; + int stride = 8; + + // Disable all optimizations. + MaskCpuFlags(disable_cpu_flags_); + double c_time = get_time(); + for (j = 0; j < benchmark_iterations_; j++) { + CopyPlane(orig_y + y_off, y_st, dst_c + y_off, stride, yw, yh); + } + c_time = (get_time() - c_time) / benchmark_iterations_; + + // Enable optimizations. + MaskCpuFlags(benchmark_cpu_info_); + double opt_time = get_time(); + for (j = 0; j < benchmark_iterations_; j++) { + CopyPlane(orig_y + y_off, y_st, dst_opt + y_off, stride, yw, yh); + } + opt_time = (get_time() - opt_time) / benchmark_iterations_; + + for (i = 0; i < y_plane_size; ++i) { + if (dst_c[i] != dst_opt[i]) + ++err; + } + + free_aligned_buffer_page_end(orig_y); + free_aligned_buffer_page_end(dst_c); + free_aligned_buffer_page_end(dst_opt); + + EXPECT_EQ(0, err); } -TEST_F(libyuvTest, Test565) { - SIMD_ALIGNED(uint8 orig_pixels[256][4]); - SIMD_ALIGNED(uint8 pixels565[256][2]); +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, 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); + } + memset(dst_argb_c, 0, kStride * height); + memset(dst_argb_opt, 0, 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 (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb_a); + free_aligned_buffer_page_end(src_argb_b); + free_aligned_buffer_page_end(dst_argb_c); + free_aligned_buffer_page_end(dst_argb_opt); + return max_diff; +} - for (int i = 0; i < 256; ++i) { - for (int j = 0; j < 4; ++j) { - orig_pixels[i][j] = i; +TEST_F(LibYUVPlanarTest, ARGBMultiply_Any) { + int max_diff = TestMultiply(benchmark_width_ - 1, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 0); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBMultiply_Unaligned) { + int max_diff = TestMultiply(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 1); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBMultiply_Invert) { + int max_diff = TestMultiply(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, -1, 0); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBMultiply_Opt) { + int max_diff = TestMultiply(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 0); + EXPECT_LE(max_diff, 1); +} + +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, 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); + } + memset(dst_argb_c, 0, kStride * height); + memset(dst_argb_opt, 0, 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 (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb_a); + free_aligned_buffer_page_end(src_argb_b); + free_aligned_buffer_page_end(dst_argb_c); + free_aligned_buffer_page_end(dst_argb_opt); + return max_diff; +} + +TEST_F(LibYUVPlanarTest, ARGBAdd_Any) { + int max_diff = TestAdd(benchmark_width_ - 1, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 0); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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, 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); + } + memset(dst_argb_c, 0, kStride * height); + memset(dst_argb_opt, 0, 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 (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb_a); + free_aligned_buffer_page_end(src_argb_b); + free_aligned_buffer_page_end(dst_argb_c); + free_aligned_buffer_page_end(dst_argb_opt); + return max_diff; +} + +TEST_F(LibYUVPlanarTest, ARGBSubtract_Any) { + int max_diff = TestSubtract(benchmark_width_ - 1, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, +1, 0); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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, kStride * height + off); + align_buffer_page_end(dst_argb_c, kStride * height); + align_buffer_page_end(dst_argb_opt, kStride * height); + memset(src_argb_a, 0, kStride * height + off); + for (int i = 0; i < kStride * height; ++i) { + src_argb_a[i + off] = (fastrand() & 0xff); + } + memset(dst_argb_c, 0, kStride * height); + memset(dst_argb_opt, 0, 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 (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb_a); + free_aligned_buffer_page_end(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); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb_a); + free_aligned_buffer_page_end(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); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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, kStride * height + off); + align_buffer_page_end(dst_argb_c, kStride * height); + align_buffer_page_end(dst_argb_opt, kStride * height); + memset(src_argb_a, 0, kStride * height + off); + for (int i = 0; i < kStride * height; ++i) { + src_argb_a[i + off] = (fastrand() & 0xff); + } + memset(dst_argb_c, 0, kStride * height); + memset(dst_argb_opt, 0, 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 (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + free_aligned_buffer_page_end(src_argb_a); + free_aligned_buffer_page_end(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); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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; + align_buffer_page_end(src_argb_a, kStride * height + off); + align_buffer_page_end(dst_cumsum, width * height * 16); + 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); + } + memset(dst_cumsum, 0, width * height * 16); + memset(dst_argb_c, 0, kStride * height); + memset(dst_argb_opt, 0, kStride * height); + + MaskCpuFlags(disable_cpu_flags); + ARGBBlur(src_argb_a + off, kStride, + dst_argb_c, kStride, + reinterpret_cast<int32*>(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<int32*>(dst_cumsum), width * 4, + width, invert * height, radius); + } + int max_diff = 0; + for (int i = 0; i < kStride * height; ++i) { + int abs_diff = + abs(static_cast<int>(dst_argb_c[i]) - + static_cast<int>(dst_argb_opt[i])); + if (abs_diff > max_diff) { + max_diff = abs_diff; } } - ARGBToRGB565(&orig_pixels[0][0], 0, &pixels565[0][0], 0, 256, 1); - uint32 checksum = HashDjb2(&pixels565[0][0], sizeof(pixels565), 5381); - EXPECT_EQ(610919429u, checksum); + 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; +} + +static const int kBlurSize = 55; +TEST_F(LibYUVPlanarTest, ARGBBlur_Any) { + int max_diff = TestBlur(benchmark_width_ - 1, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 0, kBlurSize); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBBlur_Unaligned) { + int max_diff = TestBlur(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 1, kBlurSize); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBBlur_Invert) { + int max_diff = TestBlur(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + -1, 0, kBlurSize); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBBlur_Opt) { + int max_diff = TestBlur(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 0, kBlurSize); + EXPECT_LE(max_diff, 1); +} + +static const int kBlurSmallSize = 5; +TEST_F(LibYUVPlanarTest, ARGBBlurSmall_Any) { + int max_diff = TestBlur(benchmark_width_ - 1, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 0, kBlurSmallSize); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBBlurSmall_Unaligned) { + int max_diff = TestBlur(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 1, kBlurSmallSize); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBBlurSmall_Invert) { + int max_diff = TestBlur(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + -1, 0, kBlurSmallSize); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, ARGBBlurSmall_Opt) { + int max_diff = TestBlur(benchmark_width_, benchmark_height_, + benchmark_iterations_, + disable_cpu_flags_, benchmark_cpu_info_, + +1, 0, kBlurSmallSize); + EXPECT_LE(max_diff, 1); +} + +TEST_F(LibYUVPlanarTest, TestARGBPolynomial) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); + SIMD_ALIGNED(uint8 dst_pixels_opt[1280][4]); + SIMD_ALIGNED(uint8 dst_pixels_c[1280][4]); + memset(orig_pixels, 0, sizeof(orig_pixels)); + + 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); + EXPECT_EQ(235u, dst_pixels_opt[0][0]); + EXPECT_EQ(0u, dst_pixels_opt[0][1]); + EXPECT_EQ(0u, dst_pixels_opt[0][2]); + EXPECT_EQ(128u, dst_pixels_opt[0][3]); + EXPECT_EQ(0u, dst_pixels_opt[1][0]); + EXPECT_EQ(233u, dst_pixels_opt[1][1]); + EXPECT_EQ(0u, dst_pixels_opt[1][2]); + EXPECT_EQ(0u, dst_pixels_opt[1][3]); + EXPECT_EQ(0u, dst_pixels_opt[2][0]); + EXPECT_EQ(0u, dst_pixels_opt[2][1]); + EXPECT_EQ(241u, dst_pixels_opt[2][2]); + EXPECT_EQ(255u, dst_pixels_opt[2][3]); + EXPECT_EQ(235u, dst_pixels_opt[3][0]); + EXPECT_EQ(233u, dst_pixels_opt[3][1]); + EXPECT_EQ(241u, dst_pixels_opt[3][2]); + EXPECT_EQ(255u, dst_pixels_opt[3][3]); + EXPECT_EQ(10u, dst_pixels_opt[4][0]); + EXPECT_EQ(59u, dst_pixels_opt[4][1]); + EXPECT_EQ(188u, dst_pixels_opt[4][2]); + EXPECT_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) { + EXPECT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); + EXPECT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); + EXPECT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); + EXPECT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]); + } +} + +TEST_F(LibYUVPlanarTest, TestARGBLumaColorTable) { + SIMD_ALIGNED(uint8 orig_pixels[1280][4]); + SIMD_ALIGNED(uint8 dst_pixels_opt[1280][4]); + SIMD_ALIGNED(uint8 dst_pixels_c[1280][4]); + memset(orig_pixels, 0, sizeof(orig_pixels)); + + align_buffer_page_end(lumacolortable, 32768); + int v = 0; + for (int i = 0; i < 32768; ++i) { + lumacolortable[i] = v; + 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); + EXPECT_EQ(253u, dst_pixels_opt[0][0]); + EXPECT_EQ(0u, dst_pixels_opt[0][1]); + EXPECT_EQ(0u, dst_pixels_opt[0][2]); + EXPECT_EQ(128u, dst_pixels_opt[0][3]); + EXPECT_EQ(0u, dst_pixels_opt[1][0]); + EXPECT_EQ(253u, dst_pixels_opt[1][1]); + EXPECT_EQ(0u, dst_pixels_opt[1][2]); + EXPECT_EQ(0u, dst_pixels_opt[1][3]); + EXPECT_EQ(0u, dst_pixels_opt[2][0]); + EXPECT_EQ(0u, dst_pixels_opt[2][1]); + EXPECT_EQ(253u, dst_pixels_opt[2][2]); + EXPECT_EQ(255u, dst_pixels_opt[2][3]); + EXPECT_EQ(48u, dst_pixels_opt[3][0]); + EXPECT_EQ(192u, dst_pixels_opt[3][1]); + EXPECT_EQ(64u, dst_pixels_opt[3][2]); + EXPECT_EQ(224u, dst_pixels_opt[3][3]); + + for (int i = 0; i < 1280; ++i) { + orig_pixels[i][0] = i; + orig_pixels[i][1] = i / 2; + orig_pixels[i][2] = i / 3; + orig_pixels[i][3] = i; + } + + MaskCpuFlags(disable_cpu_flags_); + 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) { + EXPECT_EQ(dst_pixels_c[i][0], dst_pixels_opt[i][0]); + EXPECT_EQ(dst_pixels_c[i][1], dst_pixels_opt[i][1]); + EXPECT_EQ(dst_pixels_c[i][2], dst_pixels_opt[i][2]); + EXPECT_EQ(dst_pixels_c[i][3], dst_pixels_opt[i][3]); + } + + free_aligned_buffer_page_end(lumacolortable); +} + +TEST_F(LibYUVPlanarTest, TestARGBCopyAlpha) { + 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); + + MemRandomize(orig_pixels, kSize); + MemRandomize(dst_pixels_opt, kSize); + memcpy(dst_pixels_c, dst_pixels_opt, kSize); + + MaskCpuFlags(disable_cpu_flags_); + ARGBCopyAlpha(orig_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) { + ARGBCopyAlpha(orig_pixels, benchmark_width_ * 4, + dst_pixels_opt, benchmark_width_ * 4, + benchmark_width_, benchmark_height_); + } + for (int i = 0; i < kSize; ++i) { + EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[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, TestARGBExtractAlpha) { + const int kPixels = benchmark_width_ * benchmark_height_; + align_buffer_page_end(src_pixels, kPixels * 4); + align_buffer_page_end(dst_pixels_opt, kPixels); + align_buffer_page_end(dst_pixels_c, kPixels); + + MemRandomize(src_pixels, kPixels * 4); + MemRandomize(dst_pixels_opt, kPixels); + memcpy(dst_pixels_c, dst_pixels_opt, kPixels); + + MaskCpuFlags(disable_cpu_flags_); + ARGBExtractAlpha(src_pixels, benchmark_width_ * 4, + dst_pixels_c, benchmark_width_, + benchmark_width_, benchmark_height_); + MaskCpuFlags(benchmark_cpu_info_); + + for (int i = 0; i < benchmark_iterations_; ++i) { + ARGBExtractAlpha(src_pixels, benchmark_width_ * 4, + dst_pixels_opt, benchmark_width_, + benchmark_width_, benchmark_height_); + } + for (int i = 0; i < kPixels; ++i) { + EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); + } + + free_aligned_buffer_page_end(dst_pixels_c); + free_aligned_buffer_page_end(dst_pixels_opt); + free_aligned_buffer_page_end(src_pixels); +} + +TEST_F(LibYUVPlanarTest, TestARGBCopyYToAlpha) { + const int kPixels = benchmark_width_ * benchmark_height_; + align_buffer_page_end(orig_pixels, kPixels); + align_buffer_page_end(dst_pixels_opt, kPixels * 4); + align_buffer_page_end(dst_pixels_c, kPixels * 4); + + MemRandomize(orig_pixels, kPixels); + MemRandomize(dst_pixels_opt, kPixels * 4); + memcpy(dst_pixels_c, dst_pixels_opt, kPixels * 4); + + MaskCpuFlags(disable_cpu_flags_); + ARGBCopyYToAlpha(orig_pixels, benchmark_width_, + dst_pixels_c, benchmark_width_ * 4, + benchmark_width_, benchmark_height_); + MaskCpuFlags(benchmark_cpu_info_); + + for (int i = 0; i < benchmark_iterations_; ++i) { + ARGBCopyYToAlpha(orig_pixels, benchmark_width_, + dst_pixels_opt, benchmark_width_ * 4, + benchmark_width_, benchmark_height_); + } + for (int i = 0; i < kPixels * 4; ++i) { + EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]); + } + + free_aligned_buffer_page_end(dst_pixels_c); + free_aligned_buffer_page_end(dst_pixels_opt); + free_aligned_buffer_page_end(orig_pixels); +} + +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 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<int>(dst_argb_c[i + off]) - + static_cast<int>(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); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_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); + EXPECT_EQ(0, max_diff); } } // namespace libyuv |