diff options
| author | Allan Sandfeld Jensen <allan.jensen@qt.io> | 2018-05-15 10:20:33 +0200 |
|---|---|---|
| committer | Allan Sandfeld Jensen <allan.jensen@qt.io> | 2018-05-15 10:28:57 +0000 |
| commit | d17ea114e5ef69ad5d5d7413280a13e6428098aa (patch) | |
| tree | 2c01a75df69f30d27b1432467cfe7c1467a498da /chromium/third_party/libyuv | |
| parent | 8c5c43c7b138c9b4b0bf56d946e61d3bbc111bec (diff) | |
| download | qtwebengine-chromium-d17ea114e5ef69ad5d5d7413280a13e6428098aa.tar.gz | |
BASELINE: Update Chromium to 67.0.3396.47
Change-Id: Idcb1341782e417561a2473eeecc82642dafda5b7
Reviewed-by: Michal Klocek <michal.klocek@qt.io>
Diffstat (limited to 'chromium/third_party/libyuv')
32 files changed, 1577 insertions, 679 deletions
diff --git a/chromium/third_party/libyuv/.gn b/chromium/third_party/libyuv/.gn index 015d0f0e5e8..fc66637f362 100644 --- a/chromium/third_party/libyuv/.gn +++ b/chromium/third_party/libyuv/.gn @@ -6,6 +6,8 @@ # in the file PATENTS. All contributing project authors may # be found in the AUTHORS file in the root of the source tree. +import("//build/dotfile_settings.gni") + # The location of the build configuration file. buildconfig = "//build/config/BUILDCONFIG.gn" @@ -23,32 +25,11 @@ check_targets = [ "//libyuv/*" ] # These are the list of GN files that run exec_script. This whitelist exists # to force additional review for new uses of exec_script, which is strongly # discouraged except for gypi_to_gn calls. -exec_script_whitelist = [ - "//build/config/android/BUILD.gn", - "//build/config/android/config.gni", - "//build/config/android/internal_rules.gni", - "//build/config/android/rules.gni", - "//build/config/BUILD.gn", - "//build/config/compiler/BUILD.gn", - "//build/config/gcc/gcc_version.gni", - "//build/config/ios/ios_sdk.gni", - "//build/config/linux/BUILD.gn", - "//build/config/linux/pkg_config.gni", - "//build/config/mac/mac_sdk.gni", - "//build/config/posix/BUILD.gn", - "//build/config/sysroot.gni", - "//build/config/win/BUILD.gn", - "//build/config/win/visual_studio_version.gni", - "//build/gn_helpers.py", - "//build/gypi_to_gn.py", - "//build/toolchain/concurrent_links.gni", - "//build/toolchain/gcc_toolchain.gni", - "//build/toolchain/mac/BUILD.gn", - "//build/toolchain/win/BUILD.gn", -] +exec_script_whitelist = build_dotfile_settings.exec_script_whitelist + + [ "//build_overrides/build.gni" ] default_args = { - mac_sdk_min = "10.11" + mac_sdk_min = "10.12" # LibYUV does not want to switch to C++14 yet. use_cxx11 = true diff --git a/chromium/third_party/libyuv/Android.bp b/chromium/third_party/libyuv/Android.bp index 7d95a7865ab..fc6a81fc66a 100644 --- a/chromium/third_party/libyuv/Android.bp +++ b/chromium/third_party/libyuv/Android.bp @@ -43,7 +43,6 @@ cc_library { "source/scale_neon.cc", "source/scale_neon64.cc", "source/video_common.cc", - "source/convert_jpeg.cc", "source/mjpeg_decoder.cc", "source/mjpeg_validate.cc", diff --git a/chromium/third_party/libyuv/CMakeLists.txt b/chromium/third_party/libyuv/CMakeLists.txt index 6420371de98..ed4948f066e 100644 --- a/chromium/third_party/libyuv/CMakeLists.txt +++ b/chromium/third_party/libyuv/CMakeLists.txt @@ -59,7 +59,11 @@ if(TEST) endif() add_executable(libyuv_unittest ${ly_unittest_sources}) - target_link_libraries(libyuv_unittest ${ly_lib_name} ${GTEST_LIBRARY} pthread) + target_link_libraries(libyuv_unittest ${ly_lib_name} ${GTEST_LIBRARY}) + find_library(PTHREAD_LIBRARY pthread) + if(NOT PTHREAD_LIBRARY STREQUAL "PTHREAD_LIBRARY-NOTFOUND") + target_link_libraries(libyuv_unittest pthread) + endif() if (JPEG_FOUND) target_link_libraries(libyuv_unittest ${JPEG_LIBRARY}) endif() @@ -68,14 +72,18 @@ if(TEST) target_link_libraries(libyuv_unittest glibc-compat) endif() - target_link_libraries(libyuv_unittest gflags) + find_library(GFLAGS_LIBRARY gflags) + if(NOT GFLAGS_LIBRARY STREQUAL "GFLAGS_LIBRARY-NOTFOUND") + target_link_libraries(libyuv_unittest gflags) + add_definitions(-DLIBYUV_USE_GFLAGS) + endif() endif() # install the conversion tool, .so, .a, and all the header files INSTALL ( PROGRAMS ${CMAKE_BINARY_DIR}/yuvconvert DESTINATION bin ) INSTALL ( TARGETS ${ly_lib_static} DESTINATION lib ) -INSTALL ( TARGETS ${ly_lib_shared} LIBRARY DESTINATION lib ) +INSTALL ( TARGETS ${ly_lib_shared} LIBRARY DESTINATION lib RUNTIME DESTINATION bin ) INSTALL ( DIRECTORY ${PROJECT_SOURCE_DIR}/include/ DESTINATION include ) # create the .deb and .rpm packages using cpack diff --git a/chromium/third_party/libyuv/DEPS b/chromium/third_party/libyuv/DEPS index ccc701c4f69..710a8a2b7d7 100644 --- a/chromium/third_party/libyuv/DEPS +++ b/chromium/third_party/libyuv/DEPS @@ -1,7 +1,7 @@ vars = { 'chromium_git': 'https://chromium.googlesource.com', - 'chromium_revision': 'ff3b31782d552b03104a6d831c7530605b52b13f', - 'swarming_revision': '5e8001d9a710121ce7a68efd0804430a34b4f9e4', + 'chromium_revision': '80f0c5570649c35a869429b2ab8c381a0a7246cb', + 'swarming_revision': '88229872dd17e71658fe96763feaa77915d8cbd6', # Three lines of non-changing comments so that # the commit queue can handle CLs rolling lss # and whatever else without interference from each other. @@ -9,30 +9,30 @@ vars = { # Three lines of non-changing comments so that # the commit queue can handle CLs rolling catapult # and whatever else without interference from each other. - 'catapult_revision': 'aa736cc76ee5e35215abcfb83a8c354f12d0c684', + 'catapult_revision': 'e7298f36f7912f2caa122086cfbe71734d04b73f', } deps = { 'src/build': - Var('chromium_git') + '/chromium/src/build' + '@' + '156ba982d749902e3403c242e23ded87fd316494', + Var('chromium_git') + '/chromium/src/build' + '@' + '39738e75b27f39d4c0030a0b11d5d2ddd34715f7', 'src/buildtools': - Var('chromium_git') + '/chromium/buildtools.git' + '@' + 'f6d165d9d842ddd29056c127a5f3a3c5d8e0d2e3', + Var('chromium_git') + '/chromium/buildtools.git' + '@' + 'a09e064635a49f08e585e3b173d5fbc3dd3f485e', 'src/testing': - Var('chromium_git') + '/chromium/src/testing' + '@' + 'cc96d3d66b5b9613fd0fe055509cfec5eb54b19c', + Var('chromium_git') + '/chromium/src/testing' + '@' + '5f7e36cad6434fd3d65674af96653a23ecc9f694', 'src/third_party': - Var('chromium_git') + '/chromium/src/third_party' + '@' + '72c52c224cdd3c377f7caff8ffed0f5749e79549', + Var('chromium_git') + '/chromium/src/third_party' + '@' + '2c0ced3ddbf84ce1c0759b277d9538da42f23650', 'src/third_party/catapult': Var('chromium_git') + '/catapult.git' + '@' + Var('catapult_revision'), 'src/third_party/colorama/src': Var('chromium_git') + '/external/colorama.git' + '@' + '799604a1041e9b3bc5d2789ecbd7e8db2e18e6b8', 'src/third_party/googletest/src': - Var('chromium_git') + '/external/github.com/google/googletest.git' + '@' + '7f8fefabedf2965980585be8c2bff97458f28e0b', + Var('chromium_git') + '/external/github.com/google/googletest.git' + '@' + '0062e4869f07a3ef235703ddf63af604b712446c', 'src/third_party/libjpeg_turbo': Var('chromium_git') + '/chromium/deps/libjpeg_turbo.git' + '@' + 'a1750dbc79a8792dde3d3f7d7d8ac28ba01ac9dd', 'src/third_party/yasm/source/patched-yasm': Var('chromium_git') + '/chromium/deps/yasm/patched-yasm.git' + '@' + 'b98114e18d8b9b84586b10d24353ab8616d4c5fc', 'src/tools': - Var('chromium_git') + '/chromium/src/tools' + '@' + 'eceb2c420b20350a2d2ba261953109280968647a', + Var('chromium_git') + '/chromium/src/tools' + '@' + '6202b67fc46a9984097caf237e12e3b8f7a9f7da', 'src/tools/gyp': Var('chromium_git') + '/external/gyp.git' + '@' + 'd61a9397e668fa9843c4aa7da9e79460fe590bfb', 'src/tools/swarming_client': @@ -52,8 +52,16 @@ deps = { }, # Android deps: + 'src/third_party/auto/src': { + 'url': Var('chromium_git') + '/external/github.com/google/auto.git' + '@' + '8a81a858ae7b78a1aef71ac3905fade0bbd64e82', + 'condition': 'checkout_android', + }, + 'src/third_party/auto/src': { + 'url': Var('chromium_git') + '/external/github.com/google/auto.git' + '@' + '8a81a858ae7b78a1aef71ac3905fade0bbd64e82', + 'condition': 'checkout_android', + }, 'src/base': { - 'url': Var('chromium_git') + '/chromium/src/base' + '@' + '9b543d487c7c38be191c6180001ff9ce186ae326', + 'url': Var('chromium_git') + '/chromium/src/base' + '@' + 'ac3d2b81181b085a9952cb83dba748420eefe691', 'condition': 'checkout_android', }, 'src/third_party/android_ndk': { @@ -61,17 +69,24 @@ deps = { 'condition': 'checkout_android', }, 'src/third_party/android_tools': { - 'url': Var('chromium_git') + '/android_tools.git' + '@' + 'aadb2fed04af8606545b0afe4e3060bc1a15fad7', + 'url': Var('chromium_git') + '/android_tools.git' + '@' + '9a70d48fcdd68cd0e7e968f342bd767ee6323bd1', 'condition': 'checkout_android', }, 'src/third_party/ced/src': { 'url': Var('chromium_git') + '/external/github.com/google/compact_enc_det.git' + '@' + '94c367a1fe3a13207f4b22604fcfd1d9f9ddf6d9', 'condition': 'checkout_android', }, - 'src/third_party/icu': { - 'url': Var('chromium_git') + '/chromium/deps/icu.git' + '@' + '08cb956852a5ccdba7f9c941728bb833529ba3c6', + 'src/third_party/errorprone/lib': { + 'url': Var('chromium_git') + '/chromium/third_party/errorprone.git' + '@' + 'ecc57c2b00627667874744b9ad8efe10734d97a8', + 'condition': 'checkout_android', + }, + 'src/third_party/findbugs': { + 'url': Var('chromium_git') + '/chromium/deps/findbugs.git' + '@' + '4275d9ac8610db6b1bc9a5e887f97e41b33fac67', 'condition': 'checkout_android', }, + 'src/third_party/icu': { + 'url': Var('chromium_git') + '/chromium/deps/icu.git' + '@' + 'd888fd2a1be890f4d35e43f68d6d79f42519a357', + }, 'src/third_party/jsr-305/src': { 'url': Var('chromium_git') + '/external/jsr-305.git' + '@' + '642c508235471f7220af6d5df2d3210e3bfc0919', 'condition': 'checkout_android', @@ -89,7 +104,7 @@ deps = { 'condition': 'checkout_android', }, 'src/third_party/robolectric/robolectric': { - 'url': Var('chromium_git') + '/external/robolectric.git' + '@' + 'b02c65cc6d7465f58f0de48a39914aa905692afa', + 'url': Var('chromium_git') + '/external/robolectric.git' + '@' + '7e067f1112e1502caa742f7be72d37b5678d3403', 'condition': 'checkout_android', }, 'src/third_party/ub-uiautomator/lib': { @@ -99,7 +114,7 @@ deps = { # iOS deps: 'src/ios': { - 'url': Var('chromium_git') + '/chromium/src/ios' + '@' + '39c4b2fcf73f5b1e82af3b9c57267c17217d6a30', + 'url': Var('chromium_git') + '/chromium/src/ios' + '@' + '299ef76e844a74a1f2f4ce7f06d101861fb49aba', 'condition': 'checkout_ios' }, @@ -140,200 +155,41 @@ hooks = [ 'src', ], }, - # Android dependencies. Many are downloaded using Google Storage these days. - # They're copied from https://cs.chromium.org/chromium/src/DEPS for all - # such dependencies we share with Chromium. - { - # This downloads SDK extras and puts them in the - # third_party/android_tools/sdk/extras directory. - 'name': 'sdkextras', - 'pattern': '.', - # When adding a new sdk extras package to download, add the package - # directory and zip file to .gitignore in third_party/android_tools. - 'action': ['python', - 'src/build/android/play_services/update.py', - 'download' - ], - }, - { - 'name': 'intellij', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-intellij', - '-l', 'third_party/intellij' - ], - }, - { - 'name': 'javax_inject', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-javax-inject', - '-l', 'third_party/javax_inject' - ], - }, - { - 'name': 'hamcrest', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-hamcrest', - '-l', 'third_party/hamcrest' - ], - }, - { - 'name': 'guava', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-guava', - '-l', 'third_party/guava' - ], - }, + # Downloads the current stable linux sysroot to build/linux/ if needed. { - 'name': 'android_support_test_runner', + 'name': 'sysroot_arm', 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-android-support-test-runner', - '-l', 'third_party/android_support_test_runner' - ], - }, - { - 'name': 'byte_buddy', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-byte-buddy', - '-l', 'third_party/byte_buddy' - ], - }, - { - 'name': 'espresso', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-espresso', - '-l', 'third_party/espresso' - ], - }, - { - 'name': 'robolectric_libs', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-robolectric', - '-l', 'third_party/robolectric' - ], - }, - { - 'name': 'apache_velocity', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-apache-velocity', - '-l', 'third_party/apache_velocity' - ], - }, - { - 'name': 'ow2_asm', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-ow2-asm', - '-l', 'third_party/ow2_asm' - ], - }, - { - 'name': 'desugar', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-android-tools/bazel/desugar', - '-l', 'third_party/bazel/desugar' - ], - }, - { - 'name': 'icu4j', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-icu4j', - '-l', 'third_party/icu4j' - ], - }, - { - 'name': 'accessibility_test_framework', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-accessibility-test-framework', - '-l', 'third_party/accessibility_test_framework' - ], - }, - { - 'name': 'bouncycastle', - 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-bouncycastle', - '-l', 'third_party/bouncycastle' - ], + 'condition': 'checkout_linux and checkout_arm', + 'action': ['python', 'src/build/linux/sysroot_scripts/install-sysroot.py', + '--arch=arm'], }, { - 'name': 'sqlite4java', + 'name': 'sysroot_arm64', 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-sqlite4java', - '-l', 'third_party/sqlite4java' - ], + 'condition': 'checkout_linux and checkout_arm64', + 'action': ['python', 'src/build/linux/sysroot_scripts/install-sysroot.py', + '--arch=arm64'], }, { - 'name': 'xstream', + 'name': 'sysroot_x86', 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-robolectric', - '-l', 'third_party/xstream' - ], + 'condition': 'checkout_linux and (checkout_x86 or checkout_x64)', + 'action': ['python', 'src/build/linux/sysroot_scripts/install-sysroot.py', + '--arch=x86'], }, { - 'name': 'objenesis', + 'name': 'sysroot_mips', 'pattern': '.', - 'action': ['python', - 'src/build/android/update_deps/update_third_party_deps.py', - 'download', - '-b', 'chromium-objenesis', - '-l', 'third_party/objenesis' - ], + 'condition': 'checkout_linux and checkout_mips', + 'action': ['python', 'src/build/linux/sysroot_scripts/install-sysroot.py', + '--arch=mips'], }, { - # Downloads the current stable linux sysroot to build/linux/ if needed. - # This sysroot updates at about the same rate that the chrome build deps - # change. This script is a no-op except for linux users who are doing - # official chrome builds or cross compiling. - 'name': 'sysroot', + 'name': 'sysroot_x64', 'pattern': '.', + 'condition': 'checkout_linux and checkout_x64', 'action': ['python', 'src/build/linux/sysroot_scripts/install-sysroot.py', - '--running-as-hook'], + '--arch=x64'], }, { # Update the Windows toolchain if necessary. @@ -341,6 +197,12 @@ hooks = [ 'pattern': '.', 'action': ['python', 'src/build/vs_toolchain.py', 'update'], }, + { + # Update the Mac toolchain if necessary. + 'name': 'mac_toolchain', + 'pattern': '.', + 'action': ['python', 'src/build/mac_toolchain.py'], + }, # Pull binutils for linux, enabled debug fission for faster linking / # debugging when used with clang on Ubuntu Precise. # https://code.google.com/p/chromium/issues/detail?id=352046 @@ -468,6 +330,30 @@ hooks = [ '-d', 'src/tools/luci-go/linux64', ], }, + { + 'name': 'Android CIPD Ensure', + 'pattern': '.', + 'condition': 'checkout_android', + 'action': ['src/build/cipd/cipd_wrapper.py', + '--chromium-root', 'src', + '--ensure-file', 'src/build/cipd/android/android.ensure', + ], + }, + # Android dependencies. Many are downloaded using Google Storage these days. + # They're copied from https://cs.chromium.org/chromium/src/DEPS for all + # such dependencies we share with Chromium. + { + # This downloads SDK extras and puts them in the + # third_party/android_tools/sdk/extras directory. + 'name': 'sdkextras', + 'pattern': '.', + # When adding a new sdk extras package to download, add the package + # directory and zip file to .gitignore in third_party/android_tools. + 'action': ['python', + 'src/build/android/play_services/update.py', + 'download' + ], + }, ] recursedeps = [ diff --git a/chromium/third_party/libyuv/README.chromium b/chromium/third_party/libyuv/README.chromium index f38414b3a80..23c10d7950d 100644 --- a/chromium/third_party/libyuv/README.chromium +++ b/chromium/third_party/libyuv/README.chromium @@ -1,6 +1,6 @@ Name: libyuv URL: http://code.google.com/p/libyuv/ -Version: 1698 +Version: 1708 License: BSD License File: LICENSE diff --git a/chromium/third_party/libyuv/build_overrides/build.gni b/chromium/third_party/libyuv/build_overrides/build.gni index 9bb73ab023a..6d8319b965e 100644 --- a/chromium/third_party/libyuv/build_overrides/build.gni +++ b/chromium/third_party/libyuv/build_overrides/build.gni @@ -33,6 +33,14 @@ ubsan_vptr_blacklist_path = # so we just ignore that assert. See https://crbug.com/648948 for more info. ignore_elf32_limitations = true -# Use system Xcode installation instead of the Chromium bundled Mac toolchain, -# since it contains only SDK 10.11, not 10.12 which WebRTC needs. -use_system_xcode = true +# Use bundled hermetic Xcode installation maintained by Chromium, +# except for local iOS builds where it is unsupported. +if (host_os == "mac") { + _result = exec_script("//build/mac/should_use_hermetic_xcode.py", + [ target_os ], + "value") + assert(_result != 2, + "Do not allow building targets with the default" + + "hermetic toolchain if the minimum OS version is not met.") + use_system_xcode = _result == 0 +} diff --git a/chromium/third_party/libyuv/docs/getting_started.md b/chromium/third_party/libyuv/docs/getting_started.md index 09297b66a5c..f547c419d67 100644 --- a/chromium/third_party/libyuv/docs/getting_started.md +++ b/chromium/third_party/libyuv/docs/getting_started.md @@ -98,11 +98,15 @@ arm64 ios simulator - gn gen out/Release "--args=is_debug=false target_os=\"ios\" ios_enable_code_signing=false target_cpu=\"x86\"" - gn gen out/Debug "--args=is_debug=true target_os=\"ios\" ios_enable_code_signing=false target_cpu=\"x86\"" + gn gen out/Release "--args=is_debug=false target_os=\"ios\" ios_enable_code_signing=false use_xcode_clang=true target_cpu=\"x86\"" + gn gen out/Debug "--args=is_debug=true target_os=\"ios\" ios_enable_code_signing=false use_xcode_clang=true target_cpu=\"x86\"" ninja -v -C out/Debug libyuv_unittest ninja -v -C out/Release libyuv_unittest +ios disassembly + + otool -tV ./out/Release/obj/libyuv_neon/row_neon64.o >row_neon64.txt + ### Android https://code.google.com/p/chromium/wiki/AndroidBuildInstructions @@ -144,6 +148,8 @@ arm disassembly: third_party/android_ndk/toolchains/aarch64-linux-android-4.9/prebuilt/linux-x86_64/bin/aarch64-linux-android-objdump -d ./out/Release/obj/libyuv_neon/row_neon64.o >row_neon64.txt + Caveat: Disassembly may require optimize_max be disabled in BUILD.gn + Running tests: build/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=* diff --git a/chromium/third_party/libyuv/include/libyuv/compare_row.h b/chromium/third_party/libyuv/include/libyuv/compare_row.h index 72ee740600a..d38f9987dc6 100644 --- a/chromium/third_party/libyuv/include/libyuv/compare_row.h +++ b/chromium/third_party/libyuv/include/libyuv/compare_row.h @@ -46,7 +46,7 @@ extern "C" { #endif // __clang__ // The following are available for Visual C: -#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \ +#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER) && \ (defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2)) #define HAS_HASHDJB2_AVX2 #endif diff --git a/chromium/third_party/libyuv/include/libyuv/convert_argb.h b/chromium/third_party/libyuv/include/libyuv/convert_argb.h index cd4a611de51..ab772b6c323 100644 --- a/chromium/third_party/libyuv/include/libyuv/convert_argb.h +++ b/chromium/third_party/libyuv/include/libyuv/convert_argb.h @@ -276,6 +276,28 @@ int NV21ToABGR(const uint8_t* src_y, int width, int height); +// Convert NV12 to RGB24. +LIBYUV_API +int NV12ToRGB24(const uint8_t* src_y, + int src_stride_y, + const uint8_t* src_uv, + int src_stride_uv, + uint8_t* dst_rgb24, + int dst_stride_rgb24, + int width, + int height); + +// Convert NV21 to RGB24. +LIBYUV_API +int NV21ToRGB24(const uint8_t* src_y, + int src_stride_y, + const uint8_t* src_vu, + int src_stride_vu, + uint8_t* dst_rgb24, + int dst_stride_rgb24, + int width, + int height); + // Convert M420 to ARGB. LIBYUV_API int M420ToARGB(const uint8_t* src_m420, @@ -547,6 +569,11 @@ int ARGB4444ToARGB(const uint8_t* src_argb4444, int width, int height); +// Aliases +#define AB30ToARGB AR30ToABGR +#define AB30ToABGR AR30ToARGB +#define AB30ToAR30 AR30ToAB30 + // Convert AR30 To ARGB. LIBYUV_API int AR30ToARGB(const uint8_t* src_ar30, @@ -556,8 +583,6 @@ int AR30ToARGB(const uint8_t* src_ar30, int width, int height); -#define AB30ToABGR - // Convert AR30 To ABGR. LIBYUV_API int AR30ToABGR(const uint8_t* src_ar30, @@ -567,6 +592,15 @@ int AR30ToABGR(const uint8_t* src_ar30, int width, int height); +// Convert AR30 To AB30. +LIBYUV_API +int AR30ToAB30(const uint8_t* src_ar30, + int src_stride_ar30, + uint8_t* dst_ab30, + int dst_stride_ab30, + int width, + int height); + #ifdef HAVE_JPEG // src_width/height provided by capture // dst_width/height for clipping determine final size. diff --git a/chromium/third_party/libyuv/include/libyuv/convert_from_argb.h b/chromium/third_party/libyuv/include/libyuv/convert_from_argb.h index 857b46c5611..05c815a093e 100644 --- a/chromium/third_party/libyuv/include/libyuv/convert_from_argb.h +++ b/chromium/third_party/libyuv/include/libyuv/convert_from_argb.h @@ -55,6 +55,10 @@ int ARGBToRGBA(const uint8_t* src_argb, int width, int height); +// Aliases +#define ARGBToAB30 ABGRToAR30 +#define ABGRToAB30 ARGBToAR30 + // Convert ABGR To AR30. LIBYUV_API int ABGRToAR30(const uint8_t* src_abgr, diff --git a/chromium/third_party/libyuv/include/libyuv/cpu_id.h b/chromium/third_party/libyuv/include/libyuv/cpu_id.h index 91480c68b01..0229cb5e736 100644 --- a/chromium/third_party/libyuv/include/libyuv/cpu_id.h +++ b/chromium/third_party/libyuv/include/libyuv/cpu_id.h @@ -80,6 +80,31 @@ int ArmCpuCaps(const char* cpuinfo_name); LIBYUV_API int MaskCpuFlags(int enable_flags); +// Sets the CPU flags to |cpu_flags|, bypassing the detection code. |cpu_flags| +// should be a valid combination of the kCpuHas constants above and include +// kCpuInitialized. Use this method when running in a sandboxed process where +// the detection code might fail (as it might access /proc/cpuinfo). In such +// cases the cpu_info can be obtained from a non sandboxed process by calling +// InitCpuFlags() and passed to the sandboxed process (via command line +// parameters, IPC...) which can then call this method to initialize the CPU +// flags. +// Notes: +// - when specifying 0 for |cpu_flags|, the auto initialization is enabled +// again. +// - enabling CPU features that are not supported by the CPU will result in +// undefined behavior. +// TODO(fbarchard): consider writing a helper function that translates from +// other library CPU info to libyuv CPU info and add a .md doc that explains +// CPU detection. +static __inline void SetCpuFlags(int cpu_flags) { + LIBYUV_API extern int cpu_info_; +#ifdef __ATOMIC_RELAXED + __atomic_store_n(&cpu_info_, cpu_flags, __ATOMIC_RELAXED); +#else + cpu_info_ = cpu_flags; +#endif +} + // Low level cpuid for X86. Returns zeros on other CPUs. // eax is the info type that you want. // ecx is typically the cpu number, and should normally be zero. diff --git a/chromium/third_party/libyuv/include/libyuv/mjpeg_decoder.h b/chromium/third_party/libyuv/include/libyuv/mjpeg_decoder.h index 6c12633387f..275f8d4c185 100644 --- a/chromium/third_party/libyuv/include/libyuv/mjpeg_decoder.h +++ b/chromium/third_party/libyuv/include/libyuv/mjpeg_decoder.h @@ -168,8 +168,8 @@ class LIBYUV_API MJpegDecoder { int GetComponentScanlinePadding(int component); // A buffer holding the input data for a frame. - Buffer buf_{}; - BufferVector buf_vec_{}; + Buffer buf_; + BufferVector buf_vec_; jpeg_decompress_struct* decompress_struct_; jpeg_source_mgr* source_mgr_; @@ -181,12 +181,12 @@ class LIBYUV_API MJpegDecoder { // Temporaries used to point to scanline outputs. int num_outbufs_; // Outermost size of all arrays below. - uint8_t*** scanlines_{}; - int* scanlines_sizes_{}; + uint8_t*** scanlines_; + int* scanlines_sizes_; // Temporary buffer used for decoding when we can't decode directly to the // output buffers. Large enough for just one iMCU row. - uint8_t** databuf_{}; - int* databuf_strides_{}; + uint8_t** databuf_; + int* databuf_strides_; }; } // namespace libyuv diff --git a/chromium/third_party/libyuv/include/libyuv/planar_functions.h b/chromium/third_party/libyuv/include/libyuv/planar_functions.h index 324bb1ed0ea..91137baba25 100644 --- a/chromium/third_party/libyuv/include/libyuv/planar_functions.h +++ b/chromium/third_party/libyuv/include/libyuv/planar_functions.h @@ -498,6 +498,10 @@ int HalfFloatPlane(const uint16_t* src_y, int width, int height); +// Convert a buffer of bytes to floats, scale the values and store as floats. +LIBYUV_API +int ByteToFloat(const uint8_t* src_y, float* dst_y, float scale, int width); + // Quantize a rectangle of ARGB. Alpha unaffected. // scale is a 16 bit fractional fixed point scaler between 0 and 65535. // interval_size should be a value between 1 and 255. diff --git a/chromium/third_party/libyuv/include/libyuv/row.h b/chromium/third_party/libyuv/include/libyuv/row.h index 62ed119db7b..646a6abb10c 100644 --- a/chromium/third_party/libyuv/include/libyuv/row.h +++ b/chromium/third_party/libyuv/include/libyuv/row.h @@ -55,6 +55,14 @@ extern "C" { #endif // clang >= 3.4 #endif // __clang__ +// clang >= 6.0.0 required for AVX512. +// TODO(fbarchard): fix xcode 9 ios b/789. +#if defined(__clang__) && (defined(__x86_64__) || defined(__i386__)) +#if (__clang_major__ >= 7) && !defined(__APPLE_EMBEDDED_SIMULATOR__) +#define CLANG_HAS_AVX512 1 +#endif // clang >= 7 +#endif // __clang__ + // Visual C 2012 required for AVX2. #if defined(_M_IX86) && !defined(__clang__) && defined(_MSC_VER) && \ _MSC_VER >= 1700 @@ -105,8 +113,10 @@ extern "C" { #define HAS_MIRRORROW_SSSE3 #define HAS_MIRRORUVROW_SSSE3 #define HAS_NV12TOARGBROW_SSSE3 +#define HAS_NV12TORGB24ROW_SSSE3 #define HAS_NV12TORGB565ROW_SSSE3 #define HAS_NV21TOARGBROW_SSSE3 +#define HAS_NV21TORGB24ROW_SSSE3 #define HAS_RAWTOARGBROW_SSSE3 #define HAS_RAWTORGB24ROW_SSSE3 #define HAS_RAWTOYROW_SSSE3 @@ -200,8 +210,10 @@ extern "C" { #define HAS_MERGEUVROW_AVX2 #define HAS_MIRRORROW_AVX2 #define HAS_NV12TOARGBROW_AVX2 +#define HAS_NV12TORGB24ROW_AVX2 #define HAS_NV12TORGB565ROW_AVX2 #define HAS_NV21TOARGBROW_AVX2 +#define HAS_NV21TORGB24ROW_AVX2 #define HAS_SPLITUVROW_AVX2 #define HAS_UYVYTOARGBROW_AVX2 #define HAS_UYVYTOUV422ROW_AVX2 @@ -230,7 +242,7 @@ extern "C" { // The following are available for AVX2 Visual C and clangcl 32 bit: // TODO(fbarchard): Port to gcc. -#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \ +#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER) && \ (defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2)) #define HAS_ARGB1555TOARGBROW_AVX2 #define HAS_ARGB4444TOARGBROW_AVX2 @@ -271,10 +283,12 @@ extern "C" { (defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2)) #define HAS_ABGRTOAR30ROW_AVX2 #define HAS_ARGBTOAR30ROW_AVX2 +#define HAS_ARGBTORAWROW_AVX2 +#define HAS_ARGBTORGB24ROW_AVX2 #define HAS_CONVERT16TO8ROW_AVX2 #define HAS_CONVERT8TO16ROW_AVX2 -#define HAS_I210TOARGBROW_AVX2 #define HAS_I210TOAR30ROW_AVX2 +#define HAS_I210TOARGBROW_AVX2 #define HAS_I422TOAR30ROW_AVX2 #define HAS_I422TOUYVYROW_AVX2 #define HAS_I422TOYUY2ROW_AVX2 @@ -282,6 +296,15 @@ extern "C" { #define HAS_MULTIPLYROW_16_AVX2 #endif +// The following are available for AVX512 clang x86 platforms: +// TODO(fbarchard): Port to GCC and Visual C +// TODO(fbarchard): re-enable HAS_ARGBTORGB24ROW_AVX512VBMI. Issue libyuv:789 +#if !defined(LIBYUV_DISABLE_X86) && \ + (defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER))) && \ + (defined(CLANG_HAS_AVX512)) +#define HAS_ARGBTORGB24ROW_AVX512VBMI +#endif + // The following are available on Neon platforms: #if !defined(LIBYUV_DISABLE_NEON) && \ (defined(__aarch64__) || defined(__ARM_NEON__) || defined(LIBYUV_NEON)) @@ -308,6 +331,7 @@ extern "C" { #define HAS_ARGBTOYROW_NEON #define HAS_BGRATOUVROW_NEON #define HAS_BGRATOYROW_NEON +#define HAS_BYTETOFLOATROW_NEON #define HAS_COPYROW_NEON #define HAS_HALFFLOATROW_NEON #define HAS_I400TOARGBROW_NEON @@ -326,8 +350,10 @@ extern "C" { #define HAS_MIRRORROW_NEON #define HAS_MIRRORUVROW_NEON #define HAS_NV12TOARGBROW_NEON +#define HAS_NV12TORGB24ROW_NEON #define HAS_NV12TORGB565ROW_NEON #define HAS_NV21TOARGBROW_NEON +#define HAS_NV21TORGB24ROW_NEON #define HAS_RAWTOARGBROW_NEON #define HAS_RAWTORGB24ROW_NEON #define HAS_RAWTOUVROW_NEON @@ -703,6 +729,16 @@ void NV21ToARGBRow_NEON(const uint8_t* src_y, uint8_t* dst_argb, const struct YuvConstants* yuvconstants, int width); +void NV12ToRGB24Row_NEON(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width); +void NV21ToRGB24Row_NEON(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width); void YUY2ToARGBRow_NEON(const uint8_t* src_yuy2, uint8_t* dst_argb, const struct YuvConstants* yuvconstants, @@ -1612,6 +1648,9 @@ void ARGB4444ToARGBRow_C(const uint8_t* src_argb4444, int width); void AR30ToARGBRow_C(const uint8_t* src_ar30, uint8_t* dst_argb, int width); void AR30ToABGRRow_C(const uint8_t* src_ar30, uint8_t* dst_abgr, int width); +void ARGBToAR30Row_C(const uint8_t* src_argb, uint8_t* dst_ar30, int width); +void AR30ToAB30Row_C(const uint8_t* src_ar30, uint8_t* dst_ab30, int width); + void RGB24ToARGBRow_Any_SSSE3(const uint8_t* src_ptr, uint8_t* dst_ptr, int width); @@ -1681,6 +1720,11 @@ void ARGBToARGB4444Row_SSE2(const uint8_t* src, uint8_t* dst, int width); void ABGRToAR30Row_SSSE3(const uint8_t* src, uint8_t* dst, int width); void ARGBToAR30Row_SSSE3(const uint8_t* src, uint8_t* dst, int width); +void ARGBToRAWRow_AVX2(const uint8_t* src, uint8_t* dst, int width); +void ARGBToRGB24Row_AVX2(const uint8_t* src, uint8_t* dst, int width); + +void ARGBToRGB24Row_AVX512VBMI(const uint8_t* src, uint8_t* dst, int width); + void ARGBToRGB565DitherRow_C(const uint8_t* src_argb, uint8_t* dst_rgb, const uint32_t dither4, @@ -1812,6 +1856,16 @@ void NV21ToARGBRow_C(const uint8_t* src_y, uint8_t* rgb_buf, const struct YuvConstants* yuvconstants, int width); +void NV12ToRGB24Row_C(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* rgb_buf, + const struct YuvConstants* yuvconstants, + int width); +void NV21ToRGB24Row_C(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* rgb_buf, + const struct YuvConstants* yuvconstants, + int width); void YUY2ToARGBRow_C(const uint8_t* src_yuy2, uint8_t* rgb_buf, const struct YuvConstants* yuvconstants, @@ -1953,11 +2007,31 @@ void NV12ToARGBRow_AVX2(const uint8_t* y_buf, uint8_t* dst_argb, const struct YuvConstants* yuvconstants, int width); +void NV12ToRGB24Row_SSSE3(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width); +void NV21ToRGB24Row_SSSE3(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width); void NV12ToRGB565Row_SSSE3(const uint8_t* src_y, const uint8_t* src_uv, uint8_t* dst_rgb565, const struct YuvConstants* yuvconstants, int width); +void NV12ToRGB24Row_AVX2(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width); +void NV21ToRGB24Row_AVX2(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width); void NV12ToRGB565Row_AVX2(const uint8_t* src_y, const uint8_t* src_uv, uint8_t* dst_rgb565, @@ -2143,6 +2217,26 @@ void NV21ToARGBRow_Any_AVX2(const uint8_t* y_buf, uint8_t* dst_ptr, const struct YuvConstants* yuvconstants, int width); +void NV12ToRGB24Row_Any_SSSE3(const uint8_t* y_buf, + const uint8_t* uv_buf, + uint8_t* dst_ptr, + const struct YuvConstants* yuvconstants, + int width); +void NV21ToRGB24Row_Any_SSSE3(const uint8_t* y_buf, + const uint8_t* uv_buf, + uint8_t* dst_ptr, + const struct YuvConstants* yuvconstants, + int width); +void NV12ToRGB24Row_Any_AVX2(const uint8_t* y_buf, + const uint8_t* uv_buf, + uint8_t* dst_ptr, + const struct YuvConstants* yuvconstants, + int width); +void NV21ToRGB24Row_Any_AVX2(const uint8_t* y_buf, + const uint8_t* uv_buf, + uint8_t* dst_ptr, + const struct YuvConstants* yuvconstants, + int width); void NV12ToRGB565Row_Any_SSSE3(const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* dst_ptr, @@ -2422,7 +2516,13 @@ void ABGRToAR30Row_Any_SSSE3(const uint8_t* src_ptr, void ARGBToAR30Row_Any_SSSE3(const uint8_t* src_ptr, uint8_t* dst_ptr, int width); - +void ARGBToRAWRow_Any_AVX2(const uint8_t* src_ptr, uint8_t* dst_ptr, int width); +void ARGBToRGB24Row_Any_AVX2(const uint8_t* src_ptr, + uint8_t* dst_ptr, + int width); +void ARGBToRGB24Row_Any_AVX512VBMI(const uint8_t* src_ptr, + uint8_t* dst_ptr, + int width); void ARGBToRGB565DitherRow_Any_SSE2(const uint8_t* src_ptr, uint8_t* dst_ptr, const uint32_t param, @@ -2542,6 +2642,16 @@ void NV21ToARGBRow_Any_NEON(const uint8_t* y_buf, uint8_t* dst_ptr, const struct YuvConstants* yuvconstants, int width); +void NV12ToRGB24Row_Any_NEON(const uint8_t* y_buf, + const uint8_t* uv_buf, + uint8_t* dst_ptr, + const struct YuvConstants* yuvconstants, + int width); +void NV21ToRGB24Row_Any_NEON(const uint8_t* y_buf, + const uint8_t* uv_buf, + uint8_t* dst_ptr, + const struct YuvConstants* yuvconstants, + int width); void NV12ToRGB565Row_Any_NEON(const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* dst_ptr, @@ -3317,6 +3427,15 @@ void HalfFloatRow_Any_MSA(const uint16_t* src_ptr, uint16_t* dst_ptr, float param, int width); +void ByteToFloatRow_C(const uint8_t* src, float* dst, float scale, int width); +void ByteToFloatRow_NEON(const uint8_t* src, + float* dst, + float scale, + int width); +void ByteToFloatRow_Any_NEON(const uint8_t* src_ptr, + float* dst_ptr, + float param, + int width); void ARGBLumaColorTableRow_C(const uint8_t* src_argb, uint8_t* dst_argb, diff --git a/chromium/third_party/libyuv/include/libyuv/version.h b/chromium/third_party/libyuv/include/libyuv/version.h index 683ac0482b4..fc5654115f9 100644 --- a/chromium/third_party/libyuv/include/libyuv/version.h +++ b/chromium/third_party/libyuv/include/libyuv/version.h @@ -11,6 +11,6 @@ #ifndef INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_ -#define LIBYUV_VERSION 1698 +#define LIBYUV_VERSION 1708 #endif // INCLUDE_LIBYUV_VERSION_H_ diff --git a/chromium/third_party/libyuv/source/convert_argb.cc b/chromium/third_party/libyuv/source/convert_argb.cc index e084f680680..f2fe474f704 100644 --- a/chromium/third_party/libyuv/source/convert_argb.cc +++ b/chromium/third_party/libyuv/source/convert_argb.cc @@ -512,8 +512,8 @@ int I010ToAB30(const uint16_t* src_y, int dst_stride_ab30, int width, int height) { - return I010ToAR30Matrix(src_y, src_stride_y, src_v, - src_stride_v, src_u, src_stride_u, dst_ab30, dst_stride_ab30, + return I010ToAR30Matrix(src_y, src_stride_y, src_v, src_stride_v, src_u, + src_stride_u, dst_ab30, dst_stride_ab30, &kYvuI601Constants, width, height); } @@ -529,8 +529,8 @@ int H010ToAB30(const uint16_t* src_y, int dst_stride_ab30, int width, int height) { - return I010ToAR30Matrix(src_y, src_stride_y, src_v, - src_stride_v, src_u, src_stride_u, dst_ab30, dst_stride_ab30, + return I010ToAR30Matrix(src_y, src_stride_y, src_v, src_stride_v, src_u, + src_stride_u, dst_ab30, dst_stride_ab30, &kYvuH709Constants, width, height); } @@ -1534,6 +1534,38 @@ int AR30ToABGR(const uint8_t* src_ar30, return 0; } +// Convert AR30 to AB30. +LIBYUV_API +int AR30ToAB30(const uint8_t* src_ar30, + int src_stride_ar30, + uint8_t* dst_ab30, + int dst_stride_ab30, + int width, + int height) { + int y; + if (!src_ar30 || !dst_ab30 || width <= 0 || height == 0) { + return -1; + } + // Negative height means invert the image. + if (height < 0) { + height = -height; + src_ar30 = src_ar30 + (height - 1) * src_stride_ar30; + src_stride_ar30 = -src_stride_ar30; + } + // Coalesce rows. + if (src_stride_ar30 == width * 4 && dst_stride_ab30 == width * 4) { + width *= height; + height = 1; + src_stride_ar30 = dst_stride_ab30 = 0; + } + for (y = 0; y < height; ++y) { + AR30ToAB30Row_C(src_ar30, dst_ab30, width); + src_ar30 += src_stride_ar30; + dst_ab30 += dst_stride_ab30; + } + return 0; +} + // Convert NV12 to ARGB with matrix static int NV12ToARGBMatrix(const uint8_t* src_y, int src_stride_y, @@ -1725,6 +1757,156 @@ int NV21ToABGR(const uint8_t* src_y, dst_stride_abgr, &kYvuI601Constants, width, height); } +// TODO(fbarchard): Consider SSSE3 2 step conversion. +// Convert NV12 to RGB24 with matrix +static int NV12ToRGB24Matrix(const uint8_t* src_y, + int src_stride_y, + const uint8_t* src_uv, + int src_stride_uv, + uint8_t* dst_rgb24, + int dst_stride_rgb24, + const struct YuvConstants* yuvconstants, + int width, + int height) { + int y; + void (*NV12ToRGB24Row)( + const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* rgb_buf, + const struct YuvConstants* yuvconstants, int width) = NV12ToRGB24Row_C; + if (!src_y || !src_uv || !dst_rgb24 || width <= 0 || height == 0) { + return -1; + } + // Negative height means invert the image. + if (height < 0) { + height = -height; + dst_rgb24 = dst_rgb24 + (height - 1) * dst_stride_rgb24; + dst_stride_rgb24 = -dst_stride_rgb24; + } +#if defined(HAS_NV12TORGB24ROW_NEON) + if (TestCpuFlag(kCpuHasNEON)) { + NV12ToRGB24Row = NV12ToRGB24Row_Any_NEON; + if (IS_ALIGNED(width, 8)) { + NV12ToRGB24Row = NV12ToRGB24Row_NEON; + } + } +#endif +#if defined(HAS_NV12TORGB24ROW_SSSE3) + if (TestCpuFlag(kCpuHasSSSE3)) { + NV12ToRGB24Row = NV12ToRGB24Row_Any_SSSE3; + if (IS_ALIGNED(width, 16)) { + NV12ToRGB24Row = NV12ToRGB24Row_SSSE3; + } + } +#endif +#if defined(HAS_NV12TORGB24ROW_AVX2) + if (TestCpuFlag(kCpuHasAVX2)) { + NV12ToRGB24Row = NV12ToRGB24Row_Any_AVX2; + if (IS_ALIGNED(width, 32)) { + NV12ToRGB24Row = NV12ToRGB24Row_AVX2; + } + } +#endif + + for (y = 0; y < height; ++y) { + NV12ToRGB24Row(src_y, src_uv, dst_rgb24, yuvconstants, width); + dst_rgb24 += dst_stride_rgb24; + src_y += src_stride_y; + if (y & 1) { + src_uv += src_stride_uv; + } + } + return 0; +} + +// Convert NV21 to RGB24 with matrix +static int NV21ToRGB24Matrix(const uint8_t* src_y, + int src_stride_y, + const uint8_t* src_vu, + int src_stride_vu, + uint8_t* dst_rgb24, + int dst_stride_rgb24, + const struct YuvConstants* yuvconstants, + int width, + int height) { + int y; + void (*NV21ToRGB24Row)( + const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* rgb_buf, + const struct YuvConstants* yuvconstants, int width) = NV21ToRGB24Row_C; + if (!src_y || !src_vu || !dst_rgb24 || width <= 0 || height == 0) { + return -1; + } + // Negative height means invert the image. + if (height < 0) { + height = -height; + dst_rgb24 = dst_rgb24 + (height - 1) * dst_stride_rgb24; + dst_stride_rgb24 = -dst_stride_rgb24; + } +#if defined(HAS_NV21TORGB24ROW_NEON) + if (TestCpuFlag(kCpuHasNEON)) { + NV21ToRGB24Row = NV21ToRGB24Row_Any_NEON; + if (IS_ALIGNED(width, 8)) { + NV21ToRGB24Row = NV21ToRGB24Row_NEON; + } + } +#endif +#if defined(HAS_NV21TORGB24ROW_SSSE3) + if (TestCpuFlag(kCpuHasSSSE3)) { + NV21ToRGB24Row = NV21ToRGB24Row_Any_SSSE3; + if (IS_ALIGNED(width, 16)) { + NV21ToRGB24Row = NV21ToRGB24Row_SSSE3; + } + } +#endif +#if defined(HAS_NV21TORGB24ROW_AVX2) + if (TestCpuFlag(kCpuHasAVX2)) { + NV21ToRGB24Row = NV21ToRGB24Row_Any_AVX2; + if (IS_ALIGNED(width, 32)) { + NV21ToRGB24Row = NV21ToRGB24Row_AVX2; + } + } +#endif + + for (y = 0; y < height; ++y) { + NV21ToRGB24Row(src_y, src_vu, dst_rgb24, yuvconstants, width); + dst_rgb24 += dst_stride_rgb24; + src_y += src_stride_y; + if (y & 1) { + src_vu += src_stride_vu; + } + } + return 0; +} + +// TODO(fbarchard): NV12ToRAW can be implemented by mirrored matrix. +// Convert NV12 to RGB24. +LIBYUV_API +int NV12ToRGB24(const uint8_t* src_y, + int src_stride_y, + const uint8_t* src_uv, + int src_stride_uv, + uint8_t* dst_rgb24, + int dst_stride_rgb24, + int width, + int height) { + return NV12ToRGB24Matrix(src_y, src_stride_y, src_uv, src_stride_uv, + dst_rgb24, dst_stride_rgb24, &kYuvI601Constants, + width, height); +} + +// Convert NV21 to RGB24. +LIBYUV_API +int NV21ToRGB24(const uint8_t* src_y, + int src_stride_y, + const uint8_t* src_vu, + int src_stride_vu, + uint8_t* dst_rgb24, + int dst_stride_rgb24, + int width, + int height) { + return NV21ToRGB24Matrix(src_y, src_stride_y, src_vu, src_stride_vu, + dst_rgb24, dst_stride_rgb24, &kYuvI601Constants, + width, height); +} + // Convert M420 to ARGB. LIBYUV_API int M420ToARGB(const uint8_t* src_m420, diff --git a/chromium/third_party/libyuv/source/convert_from_argb.cc b/chromium/third_party/libyuv/source/convert_from_argb.cc index 16b838458f0..2a36d6d83e3 100644 --- a/chromium/third_party/libyuv/source/convert_from_argb.cc +++ b/chromium/third_party/libyuv/source/convert_from_argb.cc @@ -879,6 +879,22 @@ int ARGBToRGB24(const uint8_t* src_argb, } } #endif +#if defined(HAS_ARGBTORGB24ROW_AVX2) + if (TestCpuFlag(kCpuHasAVX2)) { + ARGBToRGB24Row = ARGBToRGB24Row_Any_AVX2; + if (IS_ALIGNED(width, 32)) { + ARGBToRGB24Row = ARGBToRGB24Row_AVX2; + } + } +#endif +#if defined(HAS_ARGBTORGB24ROW_AVX512VBMI) + if (TestCpuFlag(kCpuHasAVX512VBMI)) { + ARGBToRGB24Row = ARGBToRGB24Row_Any_AVX512VBMI; + if (IS_ALIGNED(width, 32)) { + ARGBToRGB24Row = ARGBToRGB24Row_AVX512VBMI; + } + } +#endif #if defined(HAS_ARGBTORGB24ROW_NEON) if (TestCpuFlag(kCpuHasNEON)) { ARGBToRGB24Row = ARGBToRGB24Row_Any_NEON; @@ -937,6 +953,14 @@ int ARGBToRAW(const uint8_t* src_argb, } } #endif +#if defined(HAS_ARGBTORAWROW_AVX2) + if (TestCpuFlag(kCpuHasAVX2)) { + ARGBToRAWRow = ARGBToRAWRow_Any_AVX2; + if (IS_ALIGNED(width, 32)) { + ARGBToRAWRow = ARGBToRAWRow_AVX2; + } + } +#endif #if defined(HAS_ARGBTORAWROW_NEON) if (TestCpuFlag(kCpuHasNEON)) { ARGBToRAWRow = ARGBToRAWRow_Any_NEON; diff --git a/chromium/third_party/libyuv/source/convert_jpeg.cc b/chromium/third_party/libyuv/source/convert_jpeg.cc index c91b43dc226..ae3cc18cd24 100644 --- a/chromium/third_party/libyuv/source/convert_jpeg.cc +++ b/chromium/third_party/libyuv/source/convert_jpeg.cc @@ -104,20 +104,21 @@ int MJPGSize(const uint8_t* sample, } // MJPG (Motion JPeg) to I420 -// TODO(fbarchard): review w and h requirement. dw and dh may be enough. +// TODO(fbarchard): review src_width and src_height requirement. dst_width and +// dst_height may be enough. LIBYUV_API int MJPGToI420(const uint8_t* sample, size_t sample_size, - uint8_t* y, - int y_stride, - uint8_t* u, - int u_stride, - uint8_t* v, - int v_stride, - int w, - int h, - int dw, - int dh) { + uint8_t* dst_y, + int dst_stride_y, + uint8_t* dst_u, + int dst_stride_u, + uint8_t* dst_v, + int dst_stride_v, + int src_width, + int src_height, + int dst_width, + int dst_height) { if (sample_size == kUnknownDataSize) { // ERROR: MJPEG frame size unknown return -1; @@ -126,14 +127,15 @@ int MJPGToI420(const uint8_t* sample, // TODO(fbarchard): Port MJpeg to C. MJpegDecoder mjpeg_decoder; LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size); - if (ret && - (mjpeg_decoder.GetWidth() != w || mjpeg_decoder.GetHeight() != h)) { + if (ret && (mjpeg_decoder.GetWidth() != src_width || + mjpeg_decoder.GetHeight() != src_height)) { // ERROR: MJPEG frame has unexpected dimensions mjpeg_decoder.UnloadFrame(); return 1; // runtime failure } if (ret) { - I420Buffers bufs = {y, y_stride, u, u_stride, v, v_stride, dw, dh}; + I420Buffers bufs = {dst_y, dst_stride_y, dst_u, dst_stride_u, + dst_v, dst_stride_v, dst_width, dst_height}; // YUV420 if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr && mjpeg_decoder.GetNumComponents() == 3 && @@ -143,7 +145,8 @@ int MJPGToI420(const uint8_t* sample, mjpeg_decoder.GetHorizSampFactor(1) == 1 && mjpeg_decoder.GetVertSampFactor(2) == 1 && mjpeg_decoder.GetHorizSampFactor(2) == 1) { - ret = mjpeg_decoder.DecodeToCallback(&JpegCopyI420, &bufs, dw, dh); + ret = mjpeg_decoder.DecodeToCallback(&JpegCopyI420, &bufs, dst_width, + dst_height); // YUV422 } else if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr && @@ -154,7 +157,8 @@ int MJPGToI420(const uint8_t* sample, mjpeg_decoder.GetHorizSampFactor(1) == 1 && mjpeg_decoder.GetVertSampFactor(2) == 1 && mjpeg_decoder.GetHorizSampFactor(2) == 1) { - ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToI420, &bufs, dw, dh); + ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToI420, &bufs, dst_width, + dst_height); // YUV444 } else if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr && @@ -165,14 +169,16 @@ int MJPGToI420(const uint8_t* sample, mjpeg_decoder.GetHorizSampFactor(1) == 1 && mjpeg_decoder.GetVertSampFactor(2) == 1 && mjpeg_decoder.GetHorizSampFactor(2) == 1) { - ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToI420, &bufs, dw, dh); + ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToI420, &bufs, dst_width, + dst_height); // YUV400 } else if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceGrayscale && mjpeg_decoder.GetNumComponents() == 1 && mjpeg_decoder.GetVertSampFactor(0) == 1 && mjpeg_decoder.GetHorizSampFactor(0) == 1) { - ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToI420, &bufs, dw, dh); + ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToI420, &bufs, dst_width, + dst_height); } else { // TODO(fbarchard): Implement conversion for any other colorspace/sample // factors that occur in practice. @@ -236,16 +242,17 @@ static void JpegI400ToARGB(void* opaque, } // MJPG (Motion JPeg) to ARGB -// TODO(fbarchard): review w and h requirement. dw and dh may be enough. +// TODO(fbarchard): review src_width and src_height requirement. dst_width and +// dst_height may be enough. LIBYUV_API int MJPGToARGB(const uint8_t* sample, size_t sample_size, - uint8_t* argb, - int argb_stride, - int w, - int h, - int dw, - int dh) { + uint8_t* dst_argb, + int dst_stride_argb, + int src_width, + int src_height, + int dst_width, + int dst_height) { if (sample_size == kUnknownDataSize) { // ERROR: MJPEG frame size unknown return -1; @@ -254,14 +261,14 @@ int MJPGToARGB(const uint8_t* sample, // TODO(fbarchard): Port MJpeg to C. MJpegDecoder mjpeg_decoder; LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size); - if (ret && - (mjpeg_decoder.GetWidth() != w || mjpeg_decoder.GetHeight() != h)) { + if (ret && (mjpeg_decoder.GetWidth() != src_width || + mjpeg_decoder.GetHeight() != src_height)) { // ERROR: MJPEG frame has unexpected dimensions mjpeg_decoder.UnloadFrame(); return 1; // runtime failure } if (ret) { - ARGBBuffers bufs = {argb, argb_stride, dw, dh}; + ARGBBuffers bufs = {dst_argb, dst_stride_argb, dst_width, dst_height}; // YUV420 if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr && mjpeg_decoder.GetNumComponents() == 3 && @@ -271,7 +278,8 @@ int MJPGToARGB(const uint8_t* sample, mjpeg_decoder.GetHorizSampFactor(1) == 1 && mjpeg_decoder.GetVertSampFactor(2) == 1 && mjpeg_decoder.GetHorizSampFactor(2) == 1) { - ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToARGB, &bufs, dw, dh); + ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToARGB, &bufs, dst_width, + dst_height); // YUV422 } else if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr && @@ -282,7 +290,8 @@ int MJPGToARGB(const uint8_t* sample, mjpeg_decoder.GetHorizSampFactor(1) == 1 && mjpeg_decoder.GetVertSampFactor(2) == 1 && mjpeg_decoder.GetHorizSampFactor(2) == 1) { - ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToARGB, &bufs, dw, dh); + ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToARGB, &bufs, dst_width, + dst_height); // YUV444 } else if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr && @@ -293,14 +302,16 @@ int MJPGToARGB(const uint8_t* sample, mjpeg_decoder.GetHorizSampFactor(1) == 1 && mjpeg_decoder.GetVertSampFactor(2) == 1 && mjpeg_decoder.GetHorizSampFactor(2) == 1) { - ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToARGB, &bufs, dw, dh); + ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToARGB, &bufs, dst_width, + dst_height); // YUV400 } else if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceGrayscale && mjpeg_decoder.GetNumComponents() == 1 && mjpeg_decoder.GetVertSampFactor(0) == 1 && mjpeg_decoder.GetHorizSampFactor(0) == 1) { - ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToARGB, &bufs, dw, dh); + ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToARGB, &bufs, dst_width, + dst_height); } else { // TODO(fbarchard): Implement conversion for any other colorspace/sample // factors that occur in practice. diff --git a/chromium/third_party/libyuv/source/cpu_id.cc b/chromium/third_party/libyuv/source/cpu_id.cc index 446aad12078..31e24b6739b 100644 --- a/chromium/third_party/libyuv/source/cpu_id.cc +++ b/chromium/third_party/libyuv/source/cpu_id.cc @@ -19,10 +19,6 @@ #include <immintrin.h> // For _xgetbv() #endif -#if !defined(__native_client__) -#include <stdlib.h> // For getenv() -#endif - // For ArmCpuCaps() but unittested on all platforms #include <stdio.h> #include <string.h> @@ -195,25 +191,6 @@ LIBYUV_API SAFEBUFFERS int MipsCpuCaps(const char* cpuinfo_name, return 0; } -// Test environment variable for disabling CPU features. Any non-zero value -// to disable. Zero ignored to make it easy to set the variable on/off. -#if !defined(__native_client__) && !defined(_M_ARM) - -static LIBYUV_BOOL TestEnv(const char* name) { - const char* var = getenv(name); - if (var) { - if (var[0] != '0') { - return LIBYUV_TRUE; - } - } - return LIBYUV_FALSE; -} -#else // nacl does not support getenv(). -static LIBYUV_BOOL TestEnv(const char*) { - return LIBYUV_FALSE; -} -#endif - static SAFEBUFFERS int GetCpuFlags(void) { int cpu_info = 0; #if !defined(__pnacl__) && !defined(__CLR_VER) && \ @@ -251,52 +228,12 @@ static SAFEBUFFERS int GetCpuFlags(void) { cpu_info |= (cpu_info7[2] & 0x00000100) ? kCpuHasGFNI : 0; } } - - // TODO(fbarchard): Consider moving these to gtest - // Environment variable overrides for testing. - if (TestEnv("LIBYUV_DISABLE_X86")) { - cpu_info &= ~kCpuHasX86; - } - if (TestEnv("LIBYUV_DISABLE_SSE2")) { - cpu_info &= ~kCpuHasSSE2; - } - if (TestEnv("LIBYUV_DISABLE_SSSE3")) { - cpu_info &= ~kCpuHasSSSE3; - } - if (TestEnv("LIBYUV_DISABLE_SSE41")) { - cpu_info &= ~kCpuHasSSE41; - } - if (TestEnv("LIBYUV_DISABLE_SSE42")) { - cpu_info &= ~kCpuHasSSE42; - } - if (TestEnv("LIBYUV_DISABLE_AVX")) { - cpu_info &= ~kCpuHasAVX; - } - if (TestEnv("LIBYUV_DISABLE_AVX2")) { - cpu_info &= ~kCpuHasAVX2; - } - if (TestEnv("LIBYUV_DISABLE_ERMS")) { - cpu_info &= ~kCpuHasERMS; - } - if (TestEnv("LIBYUV_DISABLE_FMA3")) { - cpu_info &= ~kCpuHasFMA3; - } - if (TestEnv("LIBYUV_DISABLE_F16C")) { - cpu_info &= ~kCpuHasF16C; - } - if (TestEnv("LIBYUV_DISABLE_AVX512BW")) { - cpu_info &= ~kCpuHasAVX512BW; - } - #endif #if defined(__mips__) && defined(__linux__) #if defined(__mips_msa) cpu_info = MipsCpuCaps("/proc/cpuinfo", " msa"); #endif cpu_info |= kCpuHasMIPS; - if (getenv("LIBYUV_DISABLE_MSA")) { - cpu_info &= ~kCpuHasMSA; - } #endif #if defined(__arm__) || defined(__aarch64__) // gcc -mfpu=neon defines __ARM_NEON__ @@ -315,13 +252,7 @@ static SAFEBUFFERS int GetCpuFlags(void) { cpu_info = ArmCpuCaps("/proc/cpuinfo"); #endif cpu_info |= kCpuHasARM; - if (TestEnv("LIBYUV_DISABLE_NEON")) { - cpu_info &= ~kCpuHasNEON; - } #endif // __arm__ - if (TestEnv("LIBYUV_DISABLE_ASM")) { - cpu_info = 0; - } cpu_info |= kCpuInitialized; return cpu_info; } @@ -330,11 +261,7 @@ static SAFEBUFFERS int GetCpuFlags(void) { LIBYUV_API int MaskCpuFlags(int enable_flags) { int cpu_info = GetCpuFlags() & enable_flags; -#ifdef __ATOMIC_RELAXED - __atomic_store_n(&cpu_info_, cpu_info, __ATOMIC_RELAXED); -#else - cpu_info_ = cpu_info; -#endif + SetCpuFlags(cpu_info); return cpu_info; } diff --git a/chromium/third_party/libyuv/source/planar_functions.cc b/chromium/third_party/libyuv/source/planar_functions.cc index 77d71633f57..5eae3f763a7 100644 --- a/chromium/third_party/libyuv/source/planar_functions.cc +++ b/chromium/third_party/libyuv/source/planar_functions.cc @@ -3123,6 +3123,27 @@ int HalfFloatPlane(const uint16_t* src_y, return 0; } +// Convert a buffer of bytes to floats, scale the values and store as floats. +LIBYUV_API +int ByteToFloat(const uint8_t* src_y, float* dst_y, float scale, int width) { + void (*ByteToFloatRow)(const uint8_t* src, float* dst, float scale, + int width) = ByteToFloatRow_C; + if (!src_y || !dst_y || width <= 0) { + return -1; + } +#if defined(HAS_BYTETOFLOATROW_NEON) + if (TestCpuFlag(kCpuHasNEON)) { + ByteToFloatRow = ByteToFloatRow_Any_NEON; + if (IS_ALIGNED(width, 8)) { + ByteToFloatRow = ByteToFloatRow_NEON; + } + } +#endif + + ByteToFloatRow(src_y, dst_y, scale, width); + return 0; +} + // Apply a lumacolortable to each ARGB pixel. LIBYUV_API int ARGBLumaColorTable(const uint8_t* src_argb, diff --git a/chromium/third_party/libyuv/source/rotate_argb.cc b/chromium/third_party/libyuv/source/rotate_argb.cc index f13b041f88b..5a6e05376f1 100644 --- a/chromium/third_party/libyuv/source/rotate_argb.cc +++ b/chromium/third_party/libyuv/source/rotate_argb.cc @@ -14,111 +14,102 @@ #include "libyuv/cpu_id.h" #include "libyuv/planar_functions.h" #include "libyuv/row.h" +#include "libyuv/scale_row.h" /* for ScaleARGBRowDownEven_ */ #ifdef __cplusplus namespace libyuv { extern "C" { #endif -// ARGBScale has a function to copy pixels to a row, striding each source -// pixel by a constant. -#if !defined(LIBYUV_DISABLE_X86) && \ - (defined(_M_IX86) || defined(__x86_64__) || defined(__i386__)) -#define HAS_SCALEARGBROWDOWNEVEN_SSE2 -void ScaleARGBRowDownEven_SSE2(const uint8_t* src_ptr, - int src_stride, - int src_stepx, - uint8_t* dst_ptr, - int dst_width); -#endif -#if !defined(LIBYUV_DISABLE_NEON) && \ - (defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__)) -#define HAS_SCALEARGBROWDOWNEVEN_NEON -void ScaleARGBRowDownEven_NEON(const uint8_t* src_ptr, - int src_stride, - int src_stepx, - uint8_t* dst_ptr, - int dst_width); -#endif - -void ScaleARGBRowDownEven_C(const uint8_t* src_ptr, - int, - int src_stepx, - uint8_t* dst_ptr, - int dst_width); - -static void ARGBTranspose(const uint8_t* src, - int src_stride, - uint8_t* dst, - int dst_stride, +static void ARGBTranspose(const uint8_t* src_argb, + int src_stride_argb, + uint8_t* dst_argb, + int dst_stride_argb, int width, int height) { int i; - int src_pixel_step = src_stride >> 2; - void (*ScaleARGBRowDownEven)(const uint8_t* src_ptr, int src_stride, - int src_step, uint8_t* dst_ptr, int dst_width) = - ScaleARGBRowDownEven_C; + int src_pixel_step = src_stride_argb >> 2; + void (*ScaleARGBRowDownEven)( + const uint8_t* src_argb, ptrdiff_t src_stride_argb, int src_step, + uint8_t* dst_argb, int dst_width) = ScaleARGBRowDownEven_C; #if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2) - if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(height, 4)) { // Width of dest. - ScaleARGBRowDownEven = ScaleARGBRowDownEven_SSE2; + if (TestCpuFlag(kCpuHasSSE2)) { + ScaleARGBRowDownEven = ScaleARGBRowDownEven_Any_SSE2; + if (IS_ALIGNED(height, 4)) { // Width of dest. + ScaleARGBRowDownEven = ScaleARGBRowDownEven_SSE2; + } } #endif #if defined(HAS_SCALEARGBROWDOWNEVEN_NEON) - if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(height, 4)) { // Width of dest. - ScaleARGBRowDownEven = ScaleARGBRowDownEven_NEON; + if (TestCpuFlag(kCpuHasNEON)) { + ScaleARGBRowDownEven = ScaleARGBRowDownEven_Any_NEON; + if (IS_ALIGNED(height, 4)) { // Width of dest. + ScaleARGBRowDownEven = ScaleARGBRowDownEven_NEON; + } + } +#endif +#if defined(HAS_SCALEARGBROWDOWNEVEN_MSA) + if (TestCpuFlag(kCpuHasMSA)) { + ScaleARGBRowDownEven = ScaleARGBRowDownEven_Any_MSA; + if (IS_ALIGNED(height, 4)) { // Width of dest. + ScaleARGBRowDownEven = ScaleARGBRowDownEven_MSA; + } } #endif for (i = 0; i < width; ++i) { // column of source to row of dest. - ScaleARGBRowDownEven(src, 0, src_pixel_step, dst, height); - dst += dst_stride; - src += 4; + ScaleARGBRowDownEven(src_argb, 0, src_pixel_step, dst_argb, height); + dst_argb += dst_stride_argb; + src_argb += 4; } } -void ARGBRotate90(const uint8_t* src, - int src_stride, - uint8_t* dst, - int dst_stride, +void ARGBRotate90(const uint8_t* src_argb, + int src_stride_argb, + uint8_t* dst_argb, + int dst_stride_argb, int width, int height) { // Rotate by 90 is a ARGBTranspose with the source read // from bottom to top. So set the source pointer to the end // of the buffer and flip the sign of the source stride. - src += src_stride * (height - 1); - src_stride = -src_stride; - ARGBTranspose(src, src_stride, dst, dst_stride, width, height); + src_argb += src_stride_argb * (height - 1); + src_stride_argb = -src_stride_argb; + ARGBTranspose(src_argb, src_stride_argb, dst_argb, dst_stride_argb, width, + height); } -void ARGBRotate270(const uint8_t* src, - int src_stride, - uint8_t* dst, - int dst_stride, +void ARGBRotate270(const uint8_t* src_argb, + int src_stride_argb, + uint8_t* dst_argb, + int dst_stride_argb, int width, int height) { // Rotate by 270 is a ARGBTranspose with the destination written // from bottom to top. So set the destination pointer to the end // of the buffer and flip the sign of the destination stride. - dst += dst_stride * (width - 1); - dst_stride = -dst_stride; - ARGBTranspose(src, src_stride, dst, dst_stride, width, height); + dst_argb += dst_stride_argb * (width - 1); + dst_stride_argb = -dst_stride_argb; + ARGBTranspose(src_argb, src_stride_argb, dst_argb, dst_stride_argb, width, + height); } -void ARGBRotate180(const uint8_t* src, - int src_stride, - uint8_t* dst, - int dst_stride, +void ARGBRotate180(const uint8_t* src_argb, + int src_stride_argb, + uint8_t* dst_argb, + int dst_stride_argb, int width, int height) { // Swap first and last row and mirror the content. Uses a temporary row. align_buffer_64(row, width * 4); - const uint8_t* src_bot = src + src_stride * (height - 1); - uint8_t* dst_bot = dst + dst_stride * (height - 1); + const uint8_t* src_bot = src_argb + src_stride_argb * (height - 1); + uint8_t* dst_bot = dst_argb + dst_stride_argb * (height - 1); int half_height = (height + 1) >> 1; int y; - void (*ARGBMirrorRow)(const uint8_t* src, uint8_t* dst, int width) = + void (*ARGBMirrorRow)(const uint8_t* src_argb, uint8_t* dst_argb, int width) = ARGBMirrorRow_C; - void (*CopyRow)(const uint8_t* src, uint8_t* dst, int width) = CopyRow_C; + void (*CopyRow)(const uint8_t* src_argb, uint8_t* dst_argb, int width) = + CopyRow_C; #if defined(HAS_ARGBMIRRORROW_NEON) if (TestCpuFlag(kCpuHasNEON)) { ARGBMirrorRow = ARGBMirrorRow_Any_NEON; @@ -174,13 +165,13 @@ void ARGBRotate180(const uint8_t* src, // Odd height will harmlessly mirror the middle row twice. for (y = 0; y < half_height; ++y) { - ARGBMirrorRow(src, row, width); // Mirror first row into a buffer - ARGBMirrorRow(src_bot, dst, width); // Mirror last row into first row - CopyRow(row, dst_bot, width * 4); // Copy first mirrored row into last - src += src_stride; - dst += dst_stride; - src_bot -= src_stride; - dst_bot -= dst_stride; + ARGBMirrorRow(src_argb, row, width); // Mirror first row into a buffer + ARGBMirrorRow(src_bot, dst_argb, width); // Mirror last row into first row + CopyRow(row, dst_bot, width * 4); // Copy first mirrored row into last + src_argb += src_stride_argb; + dst_argb += dst_stride_argb; + src_bot -= src_stride_argb; + dst_bot -= dst_stride_argb; } free_aligned_buffer_64(row); } diff --git a/chromium/third_party/libyuv/source/row_any.cc b/chromium/third_party/libyuv/source/row_any.cc index 9343992b1e9..0983e2a1a83 100644 --- a/chromium/third_party/libyuv/source/row_any.cc +++ b/chromium/third_party/libyuv/source/row_any.cc @@ -123,27 +123,28 @@ ANY31(BlendPlaneRow_Any_SSSE3, BlendPlaneRow_SSSE3, 0, 0, 1, 7) // Note that odd width replication includes 444 due to implementation // on arm that subsamples 444 to 422 internally. // Any 3 planes to 1 with yuvconstants -#define ANY31C(NAMEANY, ANY_SIMD, UVSHIFT, DUVSHIFT, BPP, MASK) \ - void NAMEANY(const uint8_t* y_buf, const uint8_t* u_buf, \ - const uint8_t* v_buf, uint8_t* dst_ptr, \ - const struct YuvConstants* yuvconstants, int width) { \ - SIMD_ALIGNED(uint8_t temp[64 * 4]); \ - memset(temp, 0, 64 * 3); /* for YUY2 and msan */ \ - int r = width & MASK; \ - int n = width & ~MASK; \ - if (n > 0) { \ - ANY_SIMD(y_buf, u_buf, v_buf, dst_ptr, yuvconstants, n); \ - } \ - memcpy(temp, y_buf + n, r); \ - memcpy(temp + 64, u_buf + (n >> UVSHIFT), SS(r, UVSHIFT)); \ - memcpy(temp + 128, v_buf + (n >> UVSHIFT), SS(r, UVSHIFT)); \ - if (width & 1) { \ - temp[64 + SS(r, UVSHIFT)] = temp[64 + SS(r, UVSHIFT) - 1]; \ - temp[128 + SS(r, UVSHIFT)] = temp[128 + SS(r, UVSHIFT) - 1]; \ - } \ - ANY_SIMD(temp, temp + 64, temp + 128, temp + 192, yuvconstants, MASK + 1); \ - memcpy(dst_ptr + (n >> DUVSHIFT) * BPP, temp + 192, \ - SS(r, DUVSHIFT) * BPP); \ +#define ANY31C(NAMEANY, ANY_SIMD, UVSHIFT, DUVSHIFT, BPP, MASK) \ + void NAMEANY(const uint8_t* y_buf, const uint8_t* u_buf, \ + const uint8_t* v_buf, uint8_t* dst_ptr, \ + const struct YuvConstants* yuvconstants, int width) { \ + SIMD_ALIGNED(uint8_t temp[128 * 4]); \ + memset(temp, 0, 128 * 3); /* for YUY2 and msan */ \ + int r = width & MASK; \ + int n = width & ~MASK; \ + if (n > 0) { \ + ANY_SIMD(y_buf, u_buf, v_buf, dst_ptr, yuvconstants, n); \ + } \ + memcpy(temp, y_buf + n, r); \ + memcpy(temp + 128, u_buf + (n >> UVSHIFT), SS(r, UVSHIFT)); \ + memcpy(temp + 256, v_buf + (n >> UVSHIFT), SS(r, UVSHIFT)); \ + if (width & 1) { \ + temp[128 + SS(r, UVSHIFT)] = temp[128 + SS(r, UVSHIFT) - 1]; \ + temp[256 + SS(r, UVSHIFT)] = temp[256 + SS(r, UVSHIFT) - 1]; \ + } \ + ANY_SIMD(temp, temp + 128, temp + 256, temp + 384, yuvconstants, \ + MASK + 1); \ + memcpy(dst_ptr + (n >> DUVSHIFT) * BPP, temp + 384, \ + SS(r, DUVSHIFT) * BPP); \ } #ifdef HAS_I422TOARGBROW_SSSE3 @@ -161,10 +162,10 @@ ANY31C(I422ToRGBARow_Any_SSSE3, I422ToRGBARow_SSSE3, 1, 0, 4, 7) ANY31C(I422ToARGB4444Row_Any_SSSE3, I422ToARGB4444Row_SSSE3, 1, 0, 2, 7) ANY31C(I422ToARGB1555Row_Any_SSSE3, I422ToARGB1555Row_SSSE3, 1, 0, 2, 7) ANY31C(I422ToRGB565Row_Any_SSSE3, I422ToRGB565Row_SSSE3, 1, 0, 2, 7) -ANY31C(I422ToRGB24Row_Any_SSSE3, I422ToRGB24Row_SSSE3, 1, 0, 3, 7) +ANY31C(I422ToRGB24Row_Any_SSSE3, I422ToRGB24Row_SSSE3, 1, 0, 3, 15) #endif // HAS_I444TOARGBROW_SSSE3 #ifdef HAS_I422TORGB24ROW_AVX2 -ANY31C(I422ToRGB24Row_Any_AVX2, I422ToRGB24Row_AVX2, 1, 0, 3, 15) +ANY31C(I422ToRGB24Row_Any_AVX2, I422ToRGB24Row_AVX2, 1, 0, 3, 31) #endif #ifdef HAS_I422TOARGBROW_AVX2 ANY31C(I422ToARGBRow_Any_AVX2, I422ToARGBRow_AVX2, 1, 0, 4, 15) @@ -341,18 +342,18 @@ ANY21(SobelXYRow_Any_MSA, SobelXYRow_MSA, 0, 1, 1, 4, 15) #define ANY21C(NAMEANY, ANY_SIMD, UVSHIFT, SBPP, SBPP2, BPP, MASK) \ void NAMEANY(const uint8_t* y_buf, const uint8_t* uv_buf, uint8_t* dst_ptr, \ const struct YuvConstants* yuvconstants, int width) { \ - SIMD_ALIGNED(uint8_t temp[64 * 3]); \ - memset(temp, 0, 64 * 2); /* for msan */ \ + SIMD_ALIGNED(uint8_t temp[128 * 3]); \ + memset(temp, 0, 128 * 2); /* for msan */ \ int r = width & MASK; \ int n = width & ~MASK; \ if (n > 0) { \ ANY_SIMD(y_buf, uv_buf, dst_ptr, yuvconstants, n); \ } \ memcpy(temp, y_buf + n * SBPP, r * SBPP); \ - memcpy(temp + 64, uv_buf + (n >> UVSHIFT) * SBPP2, \ + memcpy(temp + 128, uv_buf + (n >> UVSHIFT) * SBPP2, \ SS(r, UVSHIFT) * SBPP2); \ - ANY_SIMD(temp, temp + 64, temp + 128, yuvconstants, MASK + 1); \ - memcpy(dst_ptr + n * BPP, temp + 128, r * BPP); \ + ANY_SIMD(temp, temp + 128, temp + 256, yuvconstants, MASK + 1); \ + memcpy(dst_ptr + n * BPP, temp + 256, r * BPP); \ } // Biplanar to RGB. @@ -380,6 +381,24 @@ ANY21C(NV21ToARGBRow_Any_NEON, NV21ToARGBRow_NEON, 1, 1, 2, 4, 7) #ifdef HAS_NV21TOARGBROW_MSA ANY21C(NV21ToARGBRow_Any_MSA, NV21ToARGBRow_MSA, 1, 1, 2, 4, 7) #endif +#ifdef HAS_NV12TORGB24ROW_NEON +ANY21C(NV12ToRGB24Row_Any_NEON, NV12ToRGB24Row_NEON, 1, 1, 2, 3, 7) +#endif +#ifdef HAS_NV21TORGB24ROW_NEON +ANY21C(NV21ToRGB24Row_Any_NEON, NV21ToRGB24Row_NEON, 1, 1, 2, 3, 7) +#endif +#ifdef HAS_NV12TORGB24ROW_SSSE3 +ANY21C(NV12ToRGB24Row_Any_SSSE3, NV12ToRGB24Row_SSSE3, 1, 1, 2, 3, 15) +#endif +#ifdef HAS_NV21TORGB24ROW_SSSE3 +ANY21C(NV21ToRGB24Row_Any_SSSE3, NV21ToRGB24Row_SSSE3, 1, 1, 2, 3, 15) +#endif +#ifdef HAS_NV12TORGB24ROW_AVX2 +ANY21C(NV12ToRGB24Row_Any_AVX2, NV12ToRGB24Row_AVX2, 1, 1, 2, 3, 31) +#endif +#ifdef HAS_NV21TORGB24ROW_AVX2 +ANY21C(NV21ToRGB24Row_Any_AVX2, NV21ToRGB24Row_AVX2, 1, 1, 2, 3, 31) +#endif #ifdef HAS_NV12TORGB565ROW_SSSE3 ANY21C(NV12ToRGB565Row_Any_SSSE3, NV12ToRGB565Row_SSSE3, 1, 1, 2, 2, 7) #endif @@ -425,6 +444,15 @@ ANY11(ARGBToRGB565Row_Any_SSE2, ARGBToRGB565Row_SSE2, 0, 4, 2, 3) ANY11(ARGBToARGB1555Row_Any_SSE2, ARGBToARGB1555Row_SSE2, 0, 4, 2, 3) ANY11(ARGBToARGB4444Row_Any_SSE2, ARGBToARGB4444Row_SSE2, 0, 4, 2, 3) #endif +#if defined(HAS_ARGBTORGB24ROW_AVX2) +ANY11(ARGBToRGB24Row_Any_AVX2, ARGBToRGB24Row_AVX2, 0, 4, 3, 31) +#endif +#if defined(HAS_ARGBTORGB24ROW_AVX512VBMI) +ANY11(ARGBToRGB24Row_Any_AVX512VBMI, ARGBToRGB24Row_AVX512VBMI, 0, 4, 3, 31) +#endif +#if defined(HAS_ARGBTORAWROW_AVX2) +ANY11(ARGBToRAWRow_Any_AVX2, ARGBToRAWRow_AVX2, 0, 4, 3, 31) +#endif #if defined(HAS_ARGBTORGB565ROW_AVX2) ANY11(ARGBToRGB565Row_Any_AVX2, ARGBToRGB565Row_AVX2, 0, 4, 2, 7) #endif @@ -801,37 +829,52 @@ ANY11C(Convert8To16Row_Any_AVX2, #undef ANY11C // Any 1 to 1 with parameter and shorts to byte. BPP measures in shorts. -#define ANY11P16(NAMEANY, ANY_SIMD, T, SBPP, BPP, MASK) \ - void NAMEANY(const uint16_t* src_ptr, uint16_t* dst_ptr, T param, \ - int width) { \ - SIMD_ALIGNED(uint16_t temp[32 * 2]); \ - memset(temp, 0, 64); /* for msan */ \ - int r = width & MASK; \ - int n = width & ~MASK; \ - if (n > 0) { \ - ANY_SIMD(src_ptr, dst_ptr, param, n); \ - } \ - memcpy(temp, src_ptr + n, r * SBPP); \ - ANY_SIMD(temp, temp + 16, param, MASK + 1); \ - memcpy(dst_ptr + n, temp + 16, r * BPP); \ +#define ANY11P16(NAMEANY, ANY_SIMD, ST, T, SBPP, BPP, MASK) \ + void NAMEANY(const ST* src_ptr, T* dst_ptr, float param, int width) { \ + SIMD_ALIGNED(ST temp[32]); \ + SIMD_ALIGNED(T out[32]); \ + memset(temp, 0, SBPP * 32); /* for msan */ \ + int r = width & MASK; \ + int n = width & ~MASK; \ + if (n > 0) { \ + ANY_SIMD(src_ptr, dst_ptr, param, n); \ + } \ + memcpy(temp, src_ptr + n, r * SBPP); \ + ANY_SIMD(temp, out, param, MASK + 1); \ + memcpy(dst_ptr + n, out, r * BPP); \ } #ifdef HAS_HALFFLOATROW_SSE2 -ANY11P16(HalfFloatRow_Any_SSE2, HalfFloatRow_SSE2, float, 2, 2, 7) +ANY11P16(HalfFloatRow_Any_SSE2, HalfFloatRow_SSE2, uint16_t, uint16_t, 2, 2, 7) #endif #ifdef HAS_HALFFLOATROW_AVX2 -ANY11P16(HalfFloatRow_Any_AVX2, HalfFloatRow_AVX2, float, 2, 2, 15) +ANY11P16(HalfFloatRow_Any_AVX2, HalfFloatRow_AVX2, uint16_t, uint16_t, 2, 2, 15) #endif #ifdef HAS_HALFFLOATROW_F16C -ANY11P16(HalfFloatRow_Any_F16C, HalfFloatRow_F16C, float, 2, 2, 15) -ANY11P16(HalfFloat1Row_Any_F16C, HalfFloat1Row_F16C, float, 2, 2, 15) +ANY11P16(HalfFloatRow_Any_F16C, HalfFloatRow_F16C, uint16_t, uint16_t, 2, 2, 15) +ANY11P16(HalfFloat1Row_Any_F16C, + HalfFloat1Row_F16C, + uint16_t, + uint16_t, + 2, + 2, + 15) #endif #ifdef HAS_HALFFLOATROW_NEON -ANY11P16(HalfFloatRow_Any_NEON, HalfFloatRow_NEON, float, 2, 2, 7) -ANY11P16(HalfFloat1Row_Any_NEON, HalfFloat1Row_NEON, float, 2, 2, 7) +ANY11P16(HalfFloatRow_Any_NEON, HalfFloatRow_NEON, uint16_t, uint16_t, 2, 2, 7) +ANY11P16(HalfFloat1Row_Any_NEON, + HalfFloat1Row_NEON, + uint16_t, + uint16_t, + 2, + 2, + 7) #endif #ifdef HAS_HALFFLOATROW_MSA -ANY11P16(HalfFloatRow_Any_MSA, HalfFloatRow_MSA, float, 2, 2, 31) +ANY11P16(HalfFloatRow_Any_MSA, HalfFloatRow_MSA, uint16_t, uint16_t, 2, 2, 31) +#endif +#ifdef HAS_BYTETOFLOATROW_NEON +ANY11P16(ByteToFloatRow_Any_NEON, ByteToFloatRow_NEON, uint8_t, float, 1, 3, 7) #endif #undef ANY11P16 diff --git a/chromium/third_party/libyuv/source/row_common.cc b/chromium/third_party/libyuv/source/row_common.cc index 297d87e01db..6ffaab72133 100644 --- a/chromium/third_party/libyuv/source/row_common.cc +++ b/chromium/third_party/libyuv/source/row_common.cc @@ -182,14 +182,11 @@ void AR30ToARGBRow_C(const uint8_t* src_ar30, uint8_t* dst_argb, int width) { int x; for (x = 0; x < width; ++x) { uint32_t ar30 = *(uint32_t*)src_ar30; - uint32_t b = ar30 & 0x3ff; - uint32_t g = (ar30 >> 10) & 0x3ff; - uint32_t r = (ar30 >> 20) & 0x3ff; - uint32_t a = (ar30 >> 30) & 0x3; - dst_argb[0] = b >> 2; - dst_argb[1] = g >> 2; - dst_argb[2] = r >> 2; - dst_argb[3] = a * 0x55; + uint32_t b = (ar30 >> 2) & 0xff; + uint32_t g = (ar30 >> 12) & 0xff; + uint32_t r = (ar30 >> 22) & 0xff; + uint32_t a = (ar30 >> 30) * 0x55; // Replicate 2 bits to 8 bits. + *(uint32_t*)(dst_argb) = b | (g << 8) | (r << 16) | (a << 24); dst_argb += 4; src_ar30 += 4; } @@ -199,15 +196,25 @@ void AR30ToABGRRow_C(const uint8_t* src_ar30, uint8_t* dst_abgr, int width) { int x; for (x = 0; x < width; ++x) { uint32_t ar30 = *(uint32_t*)src_ar30; + uint32_t b = (ar30 >> 2) & 0xff; + uint32_t g = (ar30 >> 12) & 0xff; + uint32_t r = (ar30 >> 22) & 0xff; + uint32_t a = (ar30 >> 30) * 0x55; // Replicate 2 bits to 8 bits. + *(uint32_t*)(dst_abgr) = r | (g << 8) | (b << 16) | (a << 24); + dst_abgr += 4; + src_ar30 += 4; + } +} + +void AR30ToAB30Row_C(const uint8_t* src_ar30, uint8_t* dst_ab30, int width) { + int x; + for (x = 0; x < width; ++x) { + uint32_t ar30 = *(uint32_t*)src_ar30; uint32_t b = ar30 & 0x3ff; - uint32_t g = (ar30 >> 10) & 0x3ff; + uint32_t ga = ar30 & 0xc00ffc00; uint32_t r = (ar30 >> 20) & 0x3ff; - uint32_t a = (ar30 >> 30) & 0x3; - dst_abgr[0] = r >> 2; - dst_abgr[1] = g >> 2; - dst_abgr[2] = b >> 2; - dst_abgr[3] = a * 0x55; - dst_abgr += 4; + *(uint32_t*)(dst_ab30) = r | ga | (b << 20); + dst_ab30 += 4; src_ar30 += 4; } } @@ -1851,6 +1858,48 @@ void NV21ToARGBRow_C(const uint8_t* src_y, } } +void NV12ToRGB24Row_C(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* rgb_buf, + const struct YuvConstants* yuvconstants, + int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_uv[0], src_uv[1], rgb_buf + 0, rgb_buf + 1, + rgb_buf + 2, yuvconstants); + YuvPixel(src_y[1], src_uv[0], src_uv[1], rgb_buf + 3, rgb_buf + 4, + rgb_buf + 5, yuvconstants); + src_y += 2; + src_uv += 2; + rgb_buf += 6; // Advance 2 pixels. + } + if (width & 1) { + YuvPixel(src_y[0], src_uv[0], src_uv[1], rgb_buf + 0, rgb_buf + 1, + rgb_buf + 2, yuvconstants); + } +} + +void NV21ToRGB24Row_C(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* rgb_buf, + const struct YuvConstants* yuvconstants, + int width) { + int x; + for (x = 0; x < width - 1; x += 2) { + YuvPixel(src_y[0], src_vu[1], src_vu[0], rgb_buf + 0, rgb_buf + 1, + rgb_buf + 2, yuvconstants); + YuvPixel(src_y[1], src_vu[1], src_vu[0], rgb_buf + 3, rgb_buf + 4, + rgb_buf + 5, yuvconstants); + src_y += 2; + src_vu += 2; + rgb_buf += 6; // Advance 2 pixels. + } + if (width & 1) { + YuvPixel(src_y[0], src_vu[1], src_vu[0], rgb_buf + 0, rgb_buf + 1, + rgb_buf + 2, yuvconstants); + } +} + void NV12ToRGB565Row_C(const uint8_t* src_y, const uint8_t* src_uv, uint8_t* dst_rgb565, @@ -2725,6 +2774,14 @@ void HalfFloatRow_C(const uint16_t* src, } } +void ByteToFloatRow_C(const uint8_t* src, float* dst, float scale, int width) { + int i; + for (i = 0; i < width; ++i) { + float value = src[i] * scale; + dst[i] = value; + } +} + void ARGBLumaColorTableRow_C(const uint8_t* src_argb, uint8_t* dst_argb, int width, @@ -2896,6 +2953,94 @@ void NV12ToRGB565Row_SSSE3(const uint8_t* src_y, } #endif +#if defined(HAS_NV12TORGB24ROW_SSSE3) +void NV12ToRGB24Row_SSSE3(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED(uint8_t row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + NV12ToARGBRow_SSSE3(src_y, src_uv, row, yuvconstants, twidth); + ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth); + src_y += twidth; + src_uv += twidth; + dst_rgb24 += twidth * 3; + width -= twidth; + } +} +#endif + +#if defined(HAS_NV21TORGB24ROW_SSSE3) +void NV21ToRGB24Row_SSSE3(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED(uint8_t row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + NV21ToARGBRow_SSSE3(src_y, src_vu, row, yuvconstants, twidth); + ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth); + src_y += twidth; + src_vu += twidth; + dst_rgb24 += twidth * 3; + width -= twidth; + } +} +#endif + +#if defined(HAS_NV12TORGB24ROW_AVX2) +void NV12ToRGB24Row_AVX2(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED(uint8_t row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + NV12ToARGBRow_AVX2(src_y, src_uv, row, yuvconstants, twidth); +#if defined(HAS_ARGBTORGB24ROW_AVX2) + ARGBToRGB24Row_AVX2(row, dst_rgb24, twidth); +#else + ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth); +#endif + src_y += twidth; + src_uv += twidth; + dst_rgb24 += twidth * 3; + width -= twidth; + } +} +#endif + +#if defined(HAS_NV21TORGB24ROW_AVX2) +void NV21ToRGB24Row_AVX2(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + // Row buffer for intermediate ARGB pixels. + SIMD_ALIGNED(uint8_t row[MAXTWIDTH * 4]); + while (width > 0) { + int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; + NV21ToARGBRow_AVX2(src_y, src_vu, row, yuvconstants, twidth); +#if defined(HAS_ARGBTORGB24ROW_AVX2) + ARGBToRGB24Row_AVX2(row, dst_rgb24, twidth); +#else + ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth); +#endif + src_y += twidth; + src_vu += twidth; + dst_rgb24 += twidth * 3; + width -= twidth; + } +} +#endif + #if defined(HAS_I422TORGB565ROW_AVX2) void I422ToRGB565Row_AVX2(const uint8_t* src_y, const uint8_t* src_u, @@ -2985,8 +3130,11 @@ void I422ToRGB24Row_AVX2(const uint8_t* src_y, while (width > 0) { int twidth = width > MAXTWIDTH ? MAXTWIDTH : width; I422ToARGBRow_AVX2(src_y, src_u, src_v, row, yuvconstants, twidth); - // TODO(fbarchard): ARGBToRGB24Row_AVX2 +#if defined(HAS_ARGBTORGB24ROW_AVX2) + ARGBToRGB24Row_AVX2(row, dst_rgb24, twidth); +#else ARGBToRGB24Row_SSSE3(row, dst_rgb24, twidth); +#endif src_y += twidth; src_u += twidth / 2; src_v += twidth / 2; diff --git a/chromium/third_party/libyuv/source/row_gcc.cc b/chromium/third_party/libyuv/source/row_gcc.cc index 95845c2592f..8d3cb81cec2 100644 --- a/chromium/third_party/libyuv/source/row_gcc.cc +++ b/chromium/third_party/libyuv/source/row_gcc.cc @@ -505,6 +505,149 @@ void ARGBToRAWRow_SSSE3(const uint8_t* src, uint8_t* dst, int width) { : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"); } +#ifdef HAS_ARGBTORGB24ROW_AVX2 +// vpermd for 12+12 to 24 +static const lvec32 kPermdRGB24_AVX = {0, 1, 2, 4, 5, 6, 3, 7}; + +void ARGBToRGB24Row_AVX2(const uint8_t* src, uint8_t* dst, int width) { + asm volatile( + "vbroadcastf128 %3,%%ymm6 \n" + "vmovdqa %4,%%ymm7 \n" + + LABELALIGN + "1: \n" + "vmovdqu (%0),%%ymm0 \n" + "vmovdqu 0x20(%0),%%ymm1 \n" + "vmovdqu 0x40(%0),%%ymm2 \n" + "vmovdqu 0x60(%0),%%ymm3 \n" + "lea 0x80(%0),%0 \n" + "vpshufb %%ymm6,%%ymm0,%%ymm0 \n" // xxx0yyy0 + "vpshufb %%ymm6,%%ymm1,%%ymm1 \n" + "vpshufb %%ymm6,%%ymm2,%%ymm2 \n" + "vpshufb %%ymm6,%%ymm3,%%ymm3 \n" + "vpermd %%ymm0,%%ymm7,%%ymm0 \n" // pack to 24 bytes + "vpermd %%ymm1,%%ymm7,%%ymm1 \n" + "vpermd %%ymm2,%%ymm7,%%ymm2 \n" + "vpermd %%ymm3,%%ymm7,%%ymm3 \n" + "vpermq $0x3f,%%ymm1,%%ymm4 \n" // combine 24 + 8 + "vpor %%ymm4,%%ymm0,%%ymm0 \n" + "vmovdqu %%ymm0,(%1) \n" + "vpermq $0xf9,%%ymm1,%%ymm1 \n" // combine 16 + 16 + "vpermq $0x4f,%%ymm2,%%ymm4 \n" + "vpor %%ymm4,%%ymm1,%%ymm1 \n" + "vmovdqu %%ymm1,0x20(%1) \n" + "vpermq $0xfe,%%ymm2,%%ymm2 \n" // combine 8 + 24 + "vpermq $0x93,%%ymm3,%%ymm3 \n" + "vpor %%ymm3,%%ymm2,%%ymm2 \n" + "vmovdqu %%ymm2,0x40(%1) \n" + "lea 0x60(%1),%1 \n" + "sub $0x20,%2 \n" + "jg 1b \n" + "vzeroupper \n" + : "+r"(src), // %0 + "+r"(dst), // %1 + "+r"(width) // %2 + : "m"(kShuffleMaskARGBToRGB24), // %3 + "m"(kPermdRGB24_AVX) // %4 + : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", + "xmm7"); +} +#endif + +#ifdef HAS_ARGBTORGB24ROW_AVX512VBMI +// Shuffle table for converting ARGBToRGB24 +static const ulvec8 kPermARGBToRGB24_0 = { + 0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 10u, 12u, 13u, + 14u, 16u, 17u, 18u, 20u, 21u, 22u, 24u, 25u, 26u, 28u, + 29u, 30u, 32u, 33u, 34u, 36u, 37u, 38u, 40u, 41u}; +static const ulvec8 kPermARGBToRGB24_1 = { + 10u, 12u, 13u, 14u, 16u, 17u, 18u, 20u, 21u, 22u, 24u, + 25u, 26u, 28u, 29u, 30u, 32u, 33u, 34u, 36u, 37u, 38u, + 40u, 41u, 42u, 44u, 45u, 46u, 48u, 49u, 50u, 52u}; +static const ulvec8 kPermARGBToRGB24_2 = { + 21u, 22u, 24u, 25u, 26u, 28u, 29u, 30u, 32u, 33u, 34u, + 36u, 37u, 38u, 40u, 41u, 42u, 44u, 45u, 46u, 48u, 49u, + 50u, 52u, 53u, 54u, 56u, 57u, 58u, 60u, 61u, 62u}; + +void ARGBToRGB24Row_AVX512VBMI(const uint8_t* src, uint8_t* dst, int width) { + asm volatile( + "vmovdqa %3,%%ymm5 \n" + "vmovdqa %4,%%ymm6 \n" + "vmovdqa %5,%%ymm7 \n" + + LABELALIGN + "1: \n" + "vmovdqu (%0),%%ymm0 \n" + "vmovdqu 0x20(%0),%%ymm1 \n" + "vmovdqu 0x40(%0),%%ymm2 \n" + "vmovdqu 0x60(%0),%%ymm3 \n" + "lea 0x80(%0),%0 \n" + "vpermt2b %%ymm1,%%ymm5,%%ymm0 \n" + "vpermt2b %%ymm2,%%ymm6,%%ymm1 \n" + "vpermt2b %%ymm3,%%ymm7,%%ymm2 \n" + "vmovdqu %%ymm0,(%1) \n" + "vmovdqu %%ymm1,0x20(%1) \n" + "vmovdqu %%ymm2,0x40(%1) \n" + "lea 0x60(%1),%1 \n" + "sub $0x20,%2 \n" + "jg 1b \n" + "vzeroupper \n" + : "+r"(src), // %0 + "+r"(dst), // %1 + "+r"(width) // %2 + : "m"(kPermARGBToRGB24_0), // %3 + "m"(kPermARGBToRGB24_1), // %4 + "m"(kPermARGBToRGB24_2) // %5 + : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5", "xmm6", "xmm7"); +} +#endif + +#ifdef HAS_ARGBTORAWROW_AVX2 +void ARGBToRAWRow_AVX2(const uint8_t* src, uint8_t* dst, int width) { + asm volatile( + "vbroadcastf128 %3,%%ymm6 \n" + "vmovdqa %4,%%ymm7 \n" + + LABELALIGN + "1: \n" + "vmovdqu (%0),%%ymm0 \n" + "vmovdqu 0x20(%0),%%ymm1 \n" + "vmovdqu 0x40(%0),%%ymm2 \n" + "vmovdqu 0x60(%0),%%ymm3 \n" + "lea 0x80(%0),%0 \n" + "vpshufb %%ymm6,%%ymm0,%%ymm0 \n" // xxx0yyy0 + "vpshufb %%ymm6,%%ymm1,%%ymm1 \n" + "vpshufb %%ymm6,%%ymm2,%%ymm2 \n" + "vpshufb %%ymm6,%%ymm3,%%ymm3 \n" + "vpermd %%ymm0,%%ymm7,%%ymm0 \n" // pack to 24 bytes + "vpermd %%ymm1,%%ymm7,%%ymm1 \n" + "vpermd %%ymm2,%%ymm7,%%ymm2 \n" + "vpermd %%ymm3,%%ymm7,%%ymm3 \n" + "vpermq $0x3f,%%ymm1,%%ymm4 \n" // combine 24 + 8 + "vpor %%ymm4,%%ymm0,%%ymm0 \n" + "vmovdqu %%ymm0,(%1) \n" + "vpermq $0xf9,%%ymm1,%%ymm1 \n" // combine 16 + 16 + "vpermq $0x4f,%%ymm2,%%ymm4 \n" + "vpor %%ymm4,%%ymm1,%%ymm1 \n" + "vmovdqu %%ymm1,0x20(%1) \n" + "vpermq $0xfe,%%ymm2,%%ymm2 \n" // combine 8 + 24 + "vpermq $0x93,%%ymm3,%%ymm3 \n" + "vpor %%ymm3,%%ymm2,%%ymm2 \n" + "vmovdqu %%ymm2,0x40(%1) \n" + "lea 0x60(%1),%1 \n" + "sub $0x20,%2 \n" + "jg 1b \n" + "vzeroupper \n" + : "+r"(src), // %0 + "+r"(dst), // %1 + "+r"(width) // %2 + : "m"(kShuffleMaskARGBToRAW), // %3 + "m"(kPermdRGB24_AVX) // %4 + : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", + "xmm7"); +} +#endif + void ARGBToRGB565Row_SSE2(const uint8_t* src, uint8_t* dst, int width) { asm volatile( "pcmpeqb %%xmm3,%%xmm3 \n" diff --git a/chromium/third_party/libyuv/source/row_neon.cc b/chromium/third_party/libyuv/source/row_neon.cc index 93a3497d275..ff87e74c62c 100644 --- a/chromium/third_party/libyuv/source/row_neon.cc +++ b/chromium/third_party/libyuv/source/row_neon.cc @@ -196,8 +196,7 @@ void I422ToRGBARow_NEON(const uint8_t* src_y, YUVTORGB_SETUP "1: \n" READYUV422 YUVTORGB "subs %4, %4, #8 \n" - "vmov.u8 d19, #255 \n" // d19 modified by - // YUVTORGB + "vmov.u8 d19, #255 \n" // YUVTORGB modified d19 "vst4.8 {d19, d20, d21, d22}, [%3]! \n" "bgt 1b \n" : "+r"(src_y), // %0 @@ -291,7 +290,6 @@ void I422ToARGB1555Row_NEON(const uint8_t* src_y, "subs %4, %4, #8 \n" "vmov.u8 d23, #255 \n" ARGBTOARGB1555 "vst1.8 {q0}, [%3]! \n" // store 8 pixels - // ARGB1555. "bgt 1b \n" : "+r"(src_y), // %0 "+r"(src_u), // %1 @@ -323,13 +321,13 @@ void I422ToARGB4444Row_NEON(const uint8_t* src_y, int width) { asm volatile( YUVTORGB_SETUP - "vmov.u8 d4, #0x0f \n" // bits to clear with - // vbic. - "1: \n" READYUV422 YUVTORGB + "vmov.u8 d4, #0x0f \n" // vbic bits to clear + "1: \n" + + READYUV422 YUVTORGB "subs %4, %4, #8 \n" "vmov.u8 d23, #255 \n" ARGBTOARGB4444 "vst1.8 {q0}, [%3]! \n" // store 8 pixels - // ARGB4444. "bgt 1b \n" : "+r"(src_y), // %0 "+r"(src_u), // %1 @@ -426,6 +424,60 @@ void NV21ToARGBRow_NEON(const uint8_t* src_y, "q10", "q11", "q12", "q13", "q14", "q15"); } +void NV12ToRGB24Row_NEON(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + asm volatile( + + YUVTORGB_SETUP + + "1: \n" + + READNV12 YUVTORGB + "subs %3, %3, #8 \n" + "vst3.8 {d20, d21, d22}, [%2]! \n" + "bgt 1b \n" + : "+r"(src_y), // %0 + "+r"(src_uv), // %1 + "+r"(dst_rgb24), // %2 + "+r"(width) // %3 + : [kUVToRB] "r"(&yuvconstants->kUVToRB), + [kUVToG] "r"(&yuvconstants->kUVToG), + [kUVBiasBGR] "r"(&yuvconstants->kUVBiasBGR), + [kYToRgb] "r"(&yuvconstants->kYToRgb) + : "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q8", "q9", "q10", "q11", + "q12", "q13", "q14", "q15"); +} + +void NV21ToRGB24Row_NEON(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + asm volatile( + + YUVTORGB_SETUP + + "1: \n" + + READNV21 YUVTORGB + "subs %3, %3, #8 \n" + "vst3.8 {d20, d21, d22}, [%2]! \n" + "bgt 1b \n" + : "+r"(src_y), // %0 + "+r"(src_vu), // %1 + "+r"(dst_rgb24), // %2 + "+r"(width) // %3 + : [kUVToRB] "r"(&yuvconstants->kUVToRB), + [kUVToG] "r"(&yuvconstants->kUVToG), + [kUVBiasBGR] "r"(&yuvconstants->kUVBiasBGR), + [kYToRgb] "r"(&yuvconstants->kYToRgb) + : "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q8", "q9", "q10", "q11", + "q12", "q13", "q14", "q15"); +} + void NV12ToRGB565Row_NEON(const uint8_t* src_y, const uint8_t* src_uv, uint8_t* dst_rgb565, @@ -2552,12 +2604,15 @@ void SobelYRow_NEON(const uint8_t* src_y0, ); } +// %y passes a float as a scalar vector for vector * scalar multiply. +// the regoster must be d0 to d15 and indexed with [0] or [1] to access +// the float in the first or second float of the d-reg + void HalfFloat1Row_NEON(const uint16_t* src, uint16_t* dst, float /*unused*/, int width) { asm volatile( - "vdup.32 q0, %3 \n" "1: \n" "vld1.8 {q1}, [%0]! \n" // load 8 shorts @@ -2566,8 +2621,8 @@ void HalfFloat1Row_NEON(const uint16_t* src, "vmovl.u16 q3, d3 \n" "vcvt.f32.u32 q2, q2 \n" // 8 floats "vcvt.f32.u32 q3, q3 \n" - "vmul.f32 q2, q2, q0 \n" // adjust exponent - "vmul.f32 q3, q3, q0 \n" + "vmul.f32 q2, q2, %y3 \n" // adjust exponent + "vmul.f32 q3, q3, %y3 \n" "vqshrn.u32 d2, q2, #13 \n" // isolate halffloat "vqshrn.u32 d3, q3, #13 \n" "vst1.8 {q1}, [%1]! \n" @@ -2575,17 +2630,15 @@ void HalfFloat1Row_NEON(const uint16_t* src, : "+r"(src), // %0 "+r"(dst), // %1 "+r"(width) // %2 - : "r"(1.9259299444e-34f) // %3 - : "cc", "memory", "q0", "q1", "q2", "q3"); + : "w"(1.9259299444e-34f) // %3 + : "cc", "memory", "q1", "q2", "q3"); } -// TODO(fbarchard): multiply by element. void HalfFloatRow_NEON(const uint16_t* src, uint16_t* dst, float scale, int width) { asm volatile( - "vdup.32 q0, %3 \n" "1: \n" "vld1.8 {q1}, [%0]! \n" // load 8 shorts @@ -2594,8 +2647,8 @@ void HalfFloatRow_NEON(const uint16_t* src, "vmovl.u16 q3, d3 \n" "vcvt.f32.u32 q2, q2 \n" // 8 floats "vcvt.f32.u32 q3, q3 \n" - "vmul.f32 q2, q2, q0 \n" // adjust exponent - "vmul.f32 q3, q3, q0 \n" + "vmul.f32 q2, q2, %y3 \n" // adjust exponent + "vmul.f32 q3, q3, %y3 \n" "vqshrn.u32 d2, q2, #13 \n" // isolate halffloat "vqshrn.u32 d3, q3, #13 \n" "vst1.8 {q1}, [%1]! \n" @@ -2603,8 +2656,33 @@ void HalfFloatRow_NEON(const uint16_t* src, : "+r"(src), // %0 "+r"(dst), // %1 "+r"(width) // %2 - : "r"(scale * 1.9259299444e-34f) // %3 - : "cc", "memory", "q0", "q1", "q2", "q3"); + : "w"(scale * 1.9259299444e-34f) // %3 + : "cc", "memory", "q1", "q2", "q3"); +} + +void ByteToFloatRow_NEON(const uint8_t* src, + float* dst, + float scale, + int width) { + asm volatile( + + "1: \n" + "vld1.8 {d2}, [%0]! \n" // load 8 bytes + "subs %2, %2, #8 \n" // 8 pixels per loop + "vmovl.u8 q1, d2 \n" // 8 shorts + "vmovl.u16 q2, d2 \n" // 8 ints + "vmovl.u16 q3, d3 \n" + "vcvt.f32.u32 q2, q2 \n" // 8 floats + "vcvt.f32.u32 q3, q3 \n" + "vmul.f32 q2, q2, %y3 \n" // scale + "vmul.f32 q3, q3, %y3 \n" + "vst1.8 {q2, q3}, [%1]! \n" // store 8 floats + "bgt 1b \n" + : "+r"(src), // %0 + "+r"(dst), // %1 + "+r"(width) // %2 + : "w"(scale) // %3 + : "cc", "memory", "q1", "q2", "q3"); } #endif // !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__).. diff --git a/chromium/third_party/libyuv/source/row_neon64.cc b/chromium/third_party/libyuv/source/row_neon64.cc index e7b8b5c1dd9..24b4520babc 100644 --- a/chromium/third_party/libyuv/source/row_neon64.cc +++ b/chromium/third_party/libyuv/source/row_neon64.cc @@ -463,6 +463,58 @@ void NV21ToARGBRow_NEON(const uint8_t* src_y, ); } +void NV12ToRGB24Row_NEON(const uint8_t* src_y, + const uint8_t* src_uv, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + asm volatile ( + YUVTORGB_SETUP + "1: \n" + READNV12 + YUVTORGB(v22, v21, v20) + "subs %w3, %w3, #8 \n" + "st3 {v20.8b,v21.8b,v22.8b}, [%2], #24 \n" + "b.gt 1b \n" + : "+r"(src_y), // %0 + "+r"(src_uv), // %1 + "+r"(dst_rgb24), // %2 + "+r"(width) // %3 + : [kUVToRB]"r"(&yuvconstants->kUVToRB), + [kUVToG]"r"(&yuvconstants->kUVToG), + [kUVBiasBGR]"r"(&yuvconstants->kUVBiasBGR), + [kYToRgb]"r"(&yuvconstants->kYToRgb) + : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20", + "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30" + ); +} + +void NV21ToRGB24Row_NEON(const uint8_t* src_y, + const uint8_t* src_vu, + uint8_t* dst_rgb24, + const struct YuvConstants* yuvconstants, + int width) { + asm volatile ( + YUVTORGB_SETUP + "1: \n" + READNV21 + YUVTORGB(v22, v21, v20) + "subs %w3, %w3, #8 \n" + "st3 {v20.8b,v21.8b,v22.8b}, [%2], #24 \n" + "b.gt 1b \n" + : "+r"(src_y), // %0 + "+r"(src_vu), // %1 + "+r"(dst_rgb24), // %2 + "+r"(width) // %3 + : [kUVToRB]"r"(&yuvconstants->kUVToRB), + [kUVToG]"r"(&yuvconstants->kUVToG), + [kUVBiasBGR]"r"(&yuvconstants->kUVBiasBGR), + [kYToRgb]"r"(&yuvconstants->kYToRgb) + : "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v20", + "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30" + ); +} + void NV12ToRGB565Row_NEON(const uint8_t* src_y, const uint8_t* src_uv, uint8_t* dst_rgb565, @@ -2648,6 +2700,30 @@ void HalfFloatRow_NEON(const uint16_t* src, : "cc", "memory", "v1", "v2", "v3"); } +void ByteToFloatRow_NEON(const uint8_t* src, + float* dst, + float scale, + int width) { + asm volatile( + "1: \n" + "ld1 {v1.8b}, [%0], #8 \n" // load 8 bytes + "subs %w2, %w2, #8 \n" // 8 pixels per loop + "uxtl v1.8h, v1.8b \n" // 8 shorts + "uxtl v2.4s, v1.4h \n" // 8 ints + "uxtl2 v3.4s, v1.8h \n" + "scvtf v2.4s, v2.4s \n" // 8 floats + "scvtf v3.4s, v3.4s \n" + "fmul v2.4s, v2.4s, %3.s[0] \n" // scale + "fmul v3.4s, v3.4s, %3.s[0] \n" + "st1 {v2.16b, v3.16b}, [%1], #32 \n" // store 8 floats + "b.gt 1b \n" + : "+r"(src), // %0 + "+r"(dst), // %1 + "+r"(width) // %2 + : "w"(scale) // %3 + : "cc", "memory", "v1", "v2", "v3"); +} + float ScaleMaxSamples_NEON(const float* src, float* dst, float scale, diff --git a/chromium/third_party/libyuv/tools_libyuv/get_landmines.py b/chromium/third_party/libyuv/tools_libyuv/get_landmines.py index 3dc78bb973f..c554f04a39a 100755 --- a/chromium/third_party/libyuv/tools_libyuv/get_landmines.py +++ b/chromium/third_party/libyuv/tools_libyuv/get_landmines.py @@ -12,20 +12,8 @@ This file emits the list of reasons why a particular build needs to be clobbered (or a list of 'landmines'). """ -import os import sys -script_dir = os.path.dirname(os.path.realpath(__file__)) -checkout_root = os.path.abspath(os.path.join(script_dir, os.pardir)) -sys.path.insert(0, os.path.join(checkout_root, 'build')) -import landmine_utils - - -distributor = landmine_utils.distributor -gyp_defines = landmine_utils.gyp_defines -gyp_msvs_version = landmine_utils.gyp_msvs_version -platform = landmine_utils.platform - def print_landmines(): """ diff --git a/chromium/third_party/libyuv/unit_test/convert_test.cc b/chromium/third_party/libyuv/unit_test/convert_test.cc index 750bd871992..e11b101fca2 100644 --- a/chromium/third_party/libyuv/unit_test/convert_test.cc +++ b/chromium/third_party/libyuv/unit_test/convert_test.cc @@ -51,12 +51,16 @@ namespace libyuv { SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, \ DST_SUBSAMP_X, DST_SUBSAMP_Y, W1280, N, NEG, OFF) \ TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) { \ - assert(SRC_BPC == 1 || SRC_BPC == 2); \ - assert(DST_BPC == 1 || DST_BPC == 2); \ - assert(SRC_SUBSAMP_X == 1 || SRC_SUBSAMP_X == 2); \ - assert(SRC_SUBSAMP_Y == 1 || SRC_SUBSAMP_Y == 2); \ - assert(DST_SUBSAMP_X == 1 || DST_SUBSAMP_X == 2); \ - assert(DST_SUBSAMP_Y == 1 || DST_SUBSAMP_Y == 2); \ + static_assert(SRC_BPC == 1 || SRC_BPC == 2, "SRC BPC unsupported"); \ + static_assert(DST_BPC == 1 || DST_BPC == 2, "DST BPC unsupported"); \ + static_assert(SRC_SUBSAMP_X == 1 || SRC_SUBSAMP_X == 2, \ + "DST SRC_SUBSAMP_X unsupported"); \ + static_assert(SRC_SUBSAMP_Y == 1 || SRC_SUBSAMP_Y == 2, \ + "DST SRC_SUBSAMP_Y unsupported"); \ + static_assert(DST_SUBSAMP_X == 1 || DST_SUBSAMP_X == 2, \ + "DST DST_SUBSAMP_X unsupported"); \ + static_assert(DST_SUBSAMP_Y == 1 || DST_SUBSAMP_Y == 2, \ + "DST DST_SUBSAMP_Y unsupported"); \ const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ const int kHeight = benchmark_height_; \ const int kSrcHalfWidth = SUBSAMPLE(kWidth, SRC_SUBSAMP_X); \ @@ -490,113 +494,100 @@ TESTBIPLANARTOP(NV21, 2, 2, I420, 2, 2) #define ALIGNINT(V, ALIGN) (((V) + (ALIGN)-1) / (ALIGN) * (ALIGN)) -#define TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ - YALIGN, W1280, DIFF, N, NEG, OFF, FMT_C, BPP_C) \ - TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) { \ - const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ - const int kHeight = ALIGNINT(benchmark_height_, YALIGN); \ - const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN); \ - const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X); \ - const int kSizeUV = kStrideUV * SUBSAMPLE(kHeight, SUBSAMP_Y); \ - align_buffer_page_end(src_y, kWidth* kHeight + OFF); \ - align_buffer_page_end(src_u, kSizeUV + OFF); \ - align_buffer_page_end(src_v, kSizeUV + OFF); \ - align_buffer_page_end(dst_argb_c, kStrideB* kHeight + OFF); \ - align_buffer_page_end(dst_argb_opt, kStrideB* kHeight + OFF); \ - for (int i = 0; i < kWidth * kHeight; ++i) { \ - src_y[i + OFF] = (fastrand() & 0xff); \ - } \ - for (int i = 0; i < kSizeUV; ++i) { \ - src_u[i + OFF] = (fastrand() & 0xff); \ - src_v[i + OFF] = (fastrand() & 0xff); \ - } \ - memset(dst_argb_c + OFF, 1, kStrideB * kHeight); \ - memset(dst_argb_opt + OFF, 101, kStrideB * kHeight); \ - MaskCpuFlags(disable_cpu_flags_); \ - FMT_PLANAR##To##FMT_B(src_y + OFF, kWidth, src_u + OFF, kStrideUV, \ - src_v + OFF, kStrideUV, dst_argb_c + OFF, kStrideB, \ - kWidth, NEG kHeight); \ - MaskCpuFlags(benchmark_cpu_info_); \ - for (int i = 0; i < benchmark_iterations_; ++i) { \ - FMT_PLANAR##To##FMT_B(src_y + OFF, kWidth, src_u + OFF, kStrideUV, \ - src_v + OFF, kStrideUV, dst_argb_opt + OFF, \ - kStrideB, kWidth, NEG kHeight); \ - } \ - int max_diff = 0; \ - /* Convert to ARGB so 565 is expanded to bytes that can be compared. */ \ - align_buffer_page_end(dst_argb32_c, kWidth* BPP_C* kHeight); \ - align_buffer_page_end(dst_argb32_opt, kWidth* BPP_C* kHeight); \ - memset(dst_argb32_c, 2, kWidth* BPP_C* kHeight); \ - memset(dst_argb32_opt, 102, kWidth* BPP_C* kHeight); \ - FMT_B##To##FMT_C(dst_argb_c + OFF, kStrideB, dst_argb32_c, kWidth * BPP_C, \ - kWidth, kHeight); \ - FMT_B##To##FMT_C(dst_argb_opt + OFF, kStrideB, dst_argb32_opt, \ - kWidth * BPP_C, kWidth, kHeight); \ - for (int i = 0; i < kWidth * BPP_C * kHeight; ++i) { \ - int abs_diff = abs(static_cast<int>(dst_argb32_c[i]) - \ - static_cast<int>(dst_argb32_opt[i])); \ - if (abs_diff > max_diff) { \ - max_diff = abs_diff; \ - } \ - } \ - EXPECT_LE(max_diff, DIFF); \ - free_aligned_buffer_page_end(src_y); \ - free_aligned_buffer_page_end(src_u); \ - free_aligned_buffer_page_end(src_v); \ - free_aligned_buffer_page_end(dst_argb_c); \ - free_aligned_buffer_page_end(dst_argb_opt); \ - free_aligned_buffer_page_end(dst_argb32_c); \ - free_aligned_buffer_page_end(dst_argb32_opt); \ - } - -#define TESTPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ - YALIGN, DIFF, FMT_C, BPP_C) \ - TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ - YALIGN, benchmark_width_ - 4, DIFF, _Any, +, 0, FMT_C, BPP_C) \ - TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ - YALIGN, benchmark_width_, DIFF, _Unaligned, +, 1, FMT_C, \ - BPP_C) \ - TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ - YALIGN, benchmark_width_, DIFF, _Invert, -, 0, FMT_C, BPP_C) \ - TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ - YALIGN, benchmark_width_, DIFF, _Opt, +, 0, FMT_C, BPP_C) - -TESTPLANARTOB(I420, 2, 2, ARGB, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(J420, 2, 2, ARGB, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(J420, 2, 2, ABGR, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(H420, 2, 2, ARGB, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(H420, 2, 2, ABGR, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, BGRA, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, ABGR, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, RGBA, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, RAW, 3, 3, 1, 2, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, RGB24, 3, 3, 1, 2, ARGB, 4) -TESTPLANARTOB(H420, 2, 2, RAW, 3, 3, 1, 2, ARGB, 4) -TESTPLANARTOB(H420, 2, 2, RGB24, 3, 3, 1, 2, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, RGB565, 2, 2, 1, 9, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, ARGB1555, 2, 2, 1, 9, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, ARGB4444, 2, 2, 1, 17, ARGB, 4) -TESTPLANARTOB(I422, 2, 1, ARGB, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I422, 2, 1, RGB565, 2, 2, 1, 9, ARGB, 4) -TESTPLANARTOB(J422, 2, 1, ARGB, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(J422, 2, 1, ABGR, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(H422, 2, 1, ARGB, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(H422, 2, 1, ABGR, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I422, 2, 1, BGRA, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I422, 2, 1, ABGR, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I422, 2, 1, RGBA, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I444, 1, 1, ARGB, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(J444, 1, 1, ARGB, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I444, 1, 1, ABGR, 4, 4, 1, 2, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, YUY2, 2, 4, 1, 1, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, UYVY, 2, 4, 1, 1, ARGB, 4) -TESTPLANARTOB(I422, 2, 1, YUY2, 2, 4, 1, 0, ARGB, 4) -TESTPLANARTOB(I422, 2, 1, UYVY, 2, 4, 1, 0, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, I400, 1, 1, 1, 0, ARGB, 4) -TESTPLANARTOB(J420, 2, 2, J400, 1, 1, 1, 0, ARGB, 4) -TESTPLANARTOB(I420, 2, 2, AR30, 4, 4, 1, 0, ARGB, 4) -TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1, 0, ARGB, 4) -// TESTPLANARTOB(I420, 2, 2, AR30, 4, 4, 1, 0, ABGR, 4) +#define TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ + YALIGN, W1280, N, NEG, OFF) \ + TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) { \ + const int kWidth = ((W1280) > 0) ? (W1280) : 1; \ + const int kHeight = ALIGNINT(benchmark_height_, YALIGN); \ + const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN); \ + const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X); \ + const int kSizeUV = kStrideUV * SUBSAMPLE(kHeight, SUBSAMP_Y); \ + align_buffer_page_end(src_y, kWidth* kHeight + OFF); \ + align_buffer_page_end(src_u, kSizeUV + OFF); \ + align_buffer_page_end(src_v, kSizeUV + OFF); \ + align_buffer_page_end(dst_argb_c, kStrideB* kHeight + OFF); \ + align_buffer_page_end(dst_argb_opt, kStrideB* kHeight + OFF); \ + for (int i = 0; i < kWidth * kHeight; ++i) { \ + src_y[i + OFF] = (fastrand() & 0xff); \ + } \ + for (int i = 0; i < kSizeUV; ++i) { \ + src_u[i + OFF] = (fastrand() & 0xff); \ + src_v[i + OFF] = (fastrand() & 0xff); \ + } \ + memset(dst_argb_c + OFF, 1, kStrideB * kHeight); \ + memset(dst_argb_opt + OFF, 101, kStrideB * kHeight); \ + MaskCpuFlags(disable_cpu_flags_); \ + double time0 = get_time(); \ + FMT_PLANAR##To##FMT_B(src_y + OFF, kWidth, src_u + OFF, kStrideUV, \ + src_v + OFF, kStrideUV, dst_argb_c + OFF, kStrideB, \ + kWidth, NEG kHeight); \ + double time1 = get_time(); \ + MaskCpuFlags(benchmark_cpu_info_); \ + for (int i = 0; i < benchmark_iterations_; ++i) { \ + FMT_PLANAR##To##FMT_B(src_y + OFF, kWidth, src_u + OFF, kStrideUV, \ + src_v + OFF, kStrideUV, dst_argb_opt + OFF, \ + kStrideB, kWidth, NEG kHeight); \ + } \ + double time2 = get_time(); \ + printf(" %8d us C - %8d us OPT\n", \ + static_cast<int>((time1 - time0) * 1e6), \ + static_cast<int>((time2 - time1) * 1e6 / benchmark_iterations_)); \ + for (int i = 0; i < kWidth * BPP_B * kHeight; ++i) { \ + EXPECT_EQ(dst_argb_c[i + OFF], dst_argb_opt[i + OFF]); \ + } \ + free_aligned_buffer_page_end(src_y); \ + free_aligned_buffer_page_end(src_u); \ + free_aligned_buffer_page_end(src_v); \ + free_aligned_buffer_page_end(dst_argb_c); \ + free_aligned_buffer_page_end(dst_argb_opt); \ + } + +#define TESTPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ + YALIGN) \ + TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ + YALIGN, benchmark_width_ - 4, _Any, +, 0) \ + TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ + YALIGN, benchmark_width_, _Unaligned, +, 1) \ + TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ + YALIGN, benchmark_width_, _Invert, -, 0) \ + TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ + YALIGN, benchmark_width_, _Opt, +, 0) + +TESTPLANARTOB(I420, 2, 2, ARGB, 4, 4, 1) +TESTPLANARTOB(J420, 2, 2, ARGB, 4, 4, 1) +TESTPLANARTOB(J420, 2, 2, ABGR, 4, 4, 1) +TESTPLANARTOB(H420, 2, 2, ARGB, 4, 4, 1) +TESTPLANARTOB(H420, 2, 2, ABGR, 4, 4, 1) +TESTPLANARTOB(I420, 2, 2, BGRA, 4, 4, 1) +TESTPLANARTOB(I420, 2, 2, ABGR, 4, 4, 1) +TESTPLANARTOB(I420, 2, 2, RGBA, 4, 4, 1) +TESTPLANARTOB(I420, 2, 2, RAW, 3, 3, 1) +TESTPLANARTOB(I420, 2, 2, RGB24, 3, 3, 1) +TESTPLANARTOB(H420, 2, 2, RAW, 3, 3, 1) +TESTPLANARTOB(H420, 2, 2, RGB24, 3, 3, 1) +TESTPLANARTOB(I420, 2, 2, RGB565, 2, 2, 1) +TESTPLANARTOB(I420, 2, 2, ARGB1555, 2, 2, 1) +TESTPLANARTOB(I420, 2, 2, ARGB4444, 2, 2, 1) +TESTPLANARTOB(I422, 2, 1, ARGB, 4, 4, 1) +TESTPLANARTOB(I422, 2, 1, RGB565, 2, 2, 1) +TESTPLANARTOB(J422, 2, 1, ARGB, 4, 4, 1) +TESTPLANARTOB(J422, 2, 1, ABGR, 4, 4, 1) +TESTPLANARTOB(H422, 2, 1, ARGB, 4, 4, 1) +TESTPLANARTOB(H422, 2, 1, ABGR, 4, 4, 1) +TESTPLANARTOB(I422, 2, 1, BGRA, 4, 4, 1) +TESTPLANARTOB(I422, 2, 1, ABGR, 4, 4, 1) +TESTPLANARTOB(I422, 2, 1, RGBA, 4, 4, 1) +TESTPLANARTOB(I444, 1, 1, ARGB, 4, 4, 1) +TESTPLANARTOB(J444, 1, 1, ARGB, 4, 4, 1) +TESTPLANARTOB(I444, 1, 1, ABGR, 4, 4, 1) +TESTPLANARTOB(I420, 2, 2, YUY2, 2, 4, 1) +TESTPLANARTOB(I420, 2, 2, UYVY, 2, 4, 1) +TESTPLANARTOB(I422, 2, 1, YUY2, 2, 4, 1) +TESTPLANARTOB(I422, 2, 1, UYVY, 2, 4, 1) +TESTPLANARTOB(I420, 2, 2, I400, 1, 1, 1) +TESTPLANARTOB(J420, 2, 2, J400, 1, 1, 1) +TESTPLANARTOB(I420, 2, 2, AR30, 4, 4, 1) +TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1) #define TESTQPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \ YALIGN, W1280, DIFF, N, NEG, OFF, ATTEN) \ @@ -741,6 +732,8 @@ TESTBIPLANARTOB(NV12, 2, 2, ARGB, 4, 2) TESTBIPLANARTOB(NV21, 2, 2, ARGB, 4, 2) TESTBIPLANARTOB(NV12, 2, 2, ABGR, 4, 2) TESTBIPLANARTOB(NV21, 2, 2, ABGR, 4, 2) +TESTBIPLANARTOB(NV12, 2, 2, RGB24, 3, 2) +TESTBIPLANARTOB(NV21, 2, 2, RGB24, 3, 2) TESTBIPLANARTOB(NV12, 2, 2, RGB565, 2, 9) #ifdef DO_THREE_PLANES @@ -1029,15 +1022,9 @@ TESTATOBIPLANAR(UYVY, 2, 4, NV12, 2, 2) MaskCpuFlags(benchmark_cpu_info_); \ FMT_A##To##FMT_B(src_argb, kStrideA, dst_argb_opt, kStrideB, kWidth, \ kHeight); \ - int max_diff = 0; \ for (int i = 0; i < kStrideB * kHeightB; ++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_NEAR(dst_argb_c[i], dst_argb_opt[i], DIFF); \ } \ - EXPECT_LE(max_diff, DIFF); \ free_aligned_buffer_page_end(src_argb); \ free_aligned_buffer_page_end(dst_argb_c); \ free_aligned_buffer_page_end(dst_argb_opt); \ @@ -1057,6 +1044,7 @@ TESTATOBIPLANAR(UYVY, 2, 4, NV12, 2, 2) TESTATOBRANDOM(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B, \ HEIGHT_B, DIFF) +// TODO(fbarchard): make ARM version of C code that matches NEON. TESTATOB(ARGB, 4, 4, 1, ARGB, 4, 4, 1, 0) TESTATOB(ARGB, 4, 4, 1, BGRA, 4, 4, 1, 0) TESTATOB(ARGB, 4, 4, 1, ABGR, 4, 4, 1, 0) @@ -1084,6 +1072,9 @@ TESTATOB(ARGB1555, 2, 2, 1, ARGB, 4, 4, 1, 0) TESTATOB(ARGB4444, 2, 2, 1, ARGB, 4, 4, 1, 0) TESTATOB(AR30, 4, 4, 1, ARGB, 4, 4, 1, 0) TESTATOB(AR30, 4, 4, 1, ABGR, 4, 4, 1, 0) +TESTATOB(AB30, 4, 4, 1, ARGB, 4, 4, 1, 0) +TESTATOB(AB30, 4, 4, 1, ABGR, 4, 4, 1, 0) +TESTATOB(AR30, 4, 4, 1, AB30, 4, 4, 1, 0) TESTATOB(YUY2, 2, 4, 1, ARGB, 4, 4, 1, ARM_YUV_ERROR) TESTATOB(UYVY, 2, 4, 1, ARGB, 4, 4, 1, ARM_YUV_ERROR) TESTATOB(YUY2, 2, 4, 1, Y, 1, 1, 1, 0) @@ -1950,6 +1941,10 @@ TESTPLANETOE(ARGB, 1, 4, AR30, 1, 4, ARGB, 4) TESTPLANETOE(ABGR, 1, 4, AR30, 1, 4, ABGR, 4) TESTPLANETOE(AR30, 1, 4, ARGB, 1, 4, ABGR, 4) TESTPLANETOE(AR30, 1, 4, ABGR, 1, 4, ARGB, 4) +TESTPLANETOE(ARGB, 1, 4, AB30, 1, 4, ARGB, 4) +TESTPLANETOE(ABGR, 1, 4, AB30, 1, 4, ABGR, 4) +TESTPLANETOE(AB30, 1, 4, ARGB, 1, 4, ABGR, 4) +TESTPLANETOE(AB30, 1, 4, ABGR, 1, 4, ARGB, 4) TEST_F(LibYUVConvertTest, RotateWithARGBSource) { // 2x2 frames @@ -2458,4 +2453,28 @@ TEST_F(LibYUVConvertTest, TestH420ToAR30) { free_aligned_buffer_page_end(ar30_pixels); } +// Test RGB24 to ARGB and back to RGB24 +TEST_F(LibYUVConvertTest, TestARGBToRGB24) { + const int kSize = 256; + align_buffer_page_end(orig_rgb24, kSize * 3); + align_buffer_page_end(argb_pixels, kSize * 4); + align_buffer_page_end(dest_rgb24, kSize * 3); + + // Test grey scale + for (int i = 0; i < kSize * 3; ++i) { + orig_rgb24[i] = i; + } + + RGB24ToARGB(orig_rgb24, 0, argb_pixels, 0, kSize, 1); + ARGBToRGB24(argb_pixels, 0, dest_rgb24, 0, kSize, 1); + + for (int i = 0; i < kSize * 3; ++i) { + EXPECT_EQ(orig_rgb24[i], dest_rgb24[i]); + } + + free_aligned_buffer_page_end(orig_rgb24); + free_aligned_buffer_page_end(argb_pixels); + free_aligned_buffer_page_end(dest_rgb24); +} + } // namespace libyuv diff --git a/chromium/third_party/libyuv/unit_test/cpu_test.cc b/chromium/third_party/libyuv/unit_test/cpu_test.cc index a8fb4b4ac01..c4648bb949f 100644 --- a/chromium/third_party/libyuv/unit_test/cpu_test.cc +++ b/chromium/third_party/libyuv/unit_test/cpu_test.cc @@ -20,12 +20,12 @@ namespace libyuv { TEST_F(LibYUVBaseTest, TestCpuHas) { int cpu_flags = TestCpuFlag(-1); - printf("Cpu Flags %x\n", cpu_flags); + printf("Cpu Flags %d\n", cpu_flags); #if defined(__arm__) || defined(__aarch64__) int has_arm = TestCpuFlag(kCpuHasARM); - printf("Has ARM %x\n", has_arm); + printf("Has ARM %d\n", has_arm); int has_neon = TestCpuFlag(kCpuHasNEON); - printf("Has NEON %x\n", has_neon); + printf("Has NEON %d\n", has_neon); #endif int has_x86 = TestCpuFlag(kCpuHasX86); int has_sse2 = TestCpuFlag(kCpuHasSSE2); @@ -44,29 +44,29 @@ TEST_F(LibYUVBaseTest, TestCpuHas) { int has_avx512vbmi2 = TestCpuFlag(kCpuHasAVX512VBMI2); int has_avx512vbitalg = TestCpuFlag(kCpuHasAVX512VBITALG); int has_avx512vpopcntdq = TestCpuFlag(kCpuHasAVX512VPOPCNTDQ); - printf("Has X86 %x\n", has_x86); - printf("Has SSE2 %x\n", has_sse2); - printf("Has SSSE3 %x\n", has_ssse3); - printf("Has SSE4.1 %x\n", has_sse41); - printf("Has SSE4.2 %x\n", has_sse42); - printf("Has AVX %x\n", has_avx); - printf("Has AVX2 %x\n", has_avx2); - printf("Has ERMS %x\n", has_erms); - printf("Has FMA3 %x\n", has_fma3); - printf("Has F16C %x\n", has_f16c); - printf("Has GFNI %x\n", has_gfni); - printf("Has AVX512BW %x\n", has_avx512bw); - printf("Has AVX512VL %x\n", has_avx512vl); - printf("Has AVX512VBMI %x\n", has_avx512vbmi); - printf("Has AVX512VBMI2 %x\n", has_avx512vbmi2); - printf("Has AVX512VBITALG %x\n", has_avx512vbitalg); - printf("Has AVX512VPOPCNTDQ %x\n", has_avx512vpopcntdq); + printf("Has X86 %d\n", has_x86); + printf("Has SSE2 %d\n", has_sse2); + printf("Has SSSE3 %d\n", has_ssse3); + printf("Has SSE41 %d\n", has_sse41); + printf("Has SSE42 %d\n", has_sse42); + printf("Has AVX %d\n", has_avx); + printf("Has AVX2 %d\n", has_avx2); + printf("Has ERMS %d\n", has_erms); + printf("Has FMA3 %d\n", has_fma3); + printf("Has F16C %d\n", has_f16c); + printf("Has GFNI %d\n", has_gfni); + printf("Has AVX512BW %d\n", has_avx512bw); + printf("Has AVX512VL %d\n", has_avx512vl); + printf("Has AVX512VBMI %d\n", has_avx512vbmi); + printf("Has AVX512VBMI2 %d\n", has_avx512vbmi2); + printf("Has AVX512VBITALG %d\n", has_avx512vbitalg); + printf("Has AVX512VPOPCNTDQ %d\n", has_avx512vpopcntdq); #if defined(__mips__) int has_mips = TestCpuFlag(kCpuHasMIPS); - printf("Has MIPS %x\n", has_mips); + printf("Has MIPS %d\n", has_mips); int has_msa = TestCpuFlag(kCpuHasMSA); - printf("Has MSA %x\n", has_msa); + printf("Has MSA %d\n", has_msa); #endif } @@ -158,4 +158,27 @@ TEST_F(LibYUVBaseTest, TestLinuxNeon) { #endif } +TEST_F(LibYUVBaseTest, TestSetCpuFlags) { + // Reset any masked flags that may have been set so auto init is enabled. + MaskCpuFlags(0); + + int original_cpu_flags = TestCpuFlag(-1); + + // Test setting different CPU configurations. + int cpu_flags = kCpuHasARM | kCpuHasNEON | kCpuInitialized; + SetCpuFlags(cpu_flags); + EXPECT_EQ(cpu_flags, TestCpuFlag(-1)); + + cpu_flags = kCpuHasX86 | kCpuInitialized; + SetCpuFlags(cpu_flags); + EXPECT_EQ(cpu_flags, TestCpuFlag(-1)); + + // Test that setting 0 turns auto-init back on. + SetCpuFlags(0); + EXPECT_EQ(original_cpu_flags, TestCpuFlag(-1)); + + // Restore the CPU flag mask. + MaskCpuFlags(benchmark_cpu_info_); +} + } // namespace libyuv diff --git a/chromium/third_party/libyuv/unit_test/planar_test.cc b/chromium/third_party/libyuv/unit_test/planar_test.cc index 9f95941ce03..756089558f7 100644 --- a/chromium/third_party/libyuv/unit_test/planar_test.cc +++ b/chromium/third_party/libyuv/unit_test/planar_test.cc @@ -2168,6 +2168,52 @@ TEST_F(LibYUVPlanarTest, TestHalfFloatPlane_12bit_One) { EXPECT_LE(diff, 1); } +float TestByteToFloat(int benchmark_width, + int benchmark_height, + int benchmark_iterations, + int disable_cpu_flags, + int benchmark_cpu_info, + float scale) { + int i, j; + const int y_plane_size = benchmark_width * benchmark_height; + + align_buffer_page_end(orig_y, y_plane_size * (1 + 4 + 4)); + float* dst_opt = reinterpret_cast<float*>(orig_y + y_plane_size); + float* dst_c = reinterpret_cast<float*>(orig_y + y_plane_size * 5); + + MemRandomize(orig_y, y_plane_size); + memset(dst_c, 0, y_plane_size * 4); + memset(dst_opt, 1, y_plane_size * 4); + + // Disable all optimizations. + MaskCpuFlags(disable_cpu_flags); + ByteToFloat(orig_y, dst_c, scale, y_plane_size); + + // Enable optimizations. + MaskCpuFlags(benchmark_cpu_info); + for (j = 0; j < benchmark_iterations; j++) { + ByteToFloat(orig_y, dst_opt, scale, y_plane_size); + } + + float max_diff = 0; + for (i = 0; i < y_plane_size; ++i) { + float abs_diff = fabs(dst_c[i] - dst_opt[i]); + if (abs_diff > max_diff) { + max_diff = abs_diff; + } + } + + free_aligned_buffer_page_end(orig_y); + return max_diff; +} + +TEST_F(LibYUVPlanarTest, TestByteToFloat) { + float diff = TestByteToFloat(benchmark_width_, benchmark_height_, + benchmark_iterations_, disable_cpu_flags_, + benchmark_cpu_info_, 1.0f); + EXPECT_EQ(0.f, diff); +} + TEST_F(LibYUVPlanarTest, TestARGBLumaColorTable) { SIMD_ALIGNED(uint8_t orig_pixels[1280][4]); SIMD_ALIGNED(uint8_t dst_pixels_opt[1280][4]); diff --git a/chromium/third_party/libyuv/unit_test/unit_test.cc b/chromium/third_party/libyuv/unit_test/unit_test.cc index 20aadb44e2f..7d662706aaa 100644 --- a/chromium/third_party/libyuv/unit_test/unit_test.cc +++ b/chromium/third_party/libyuv/unit_test/unit_test.cc @@ -38,6 +38,96 @@ static const int32_t FLAGS_libyuv_flags = 0; static const int32_t FLAGS_libyuv_cpu_info = 0; #endif +// Test environment variable for disabling CPU features. Any non-zero value +// to disable. Zero ignored to make it easy to set the variable on/off. +#if !defined(__native_client__) && !defined(_M_ARM) +static LIBYUV_BOOL TestEnv(const char* name) { + const char* var = getenv(name); + if (var) { + if (var[0] != '0') { + return LIBYUV_TRUE; + } + } + return LIBYUV_FALSE; +} +#else // nacl does not support getenv(). +static LIBYUV_BOOL TestEnv(const char*) { + return LIBYUV_FALSE; +} +#endif + +int TestCpuEnv(int cpu_info) { +#if defined(__arm__) || defined(__aarch64__) + if (TestEnv("LIBYUV_DISABLE_NEON")) { + cpu_info &= ~libyuv::kCpuHasNEON; + } +#endif +#if defined(__mips__) && defined(__linux__) + if (TestEnv("LIBYUV_DISABLE_MSA")) { + cpu_info &= ~libyuv::kCpuHasMSA; + } +#endif +#if !defined(__pnacl__) && !defined(__CLR_VER) && \ + (defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || \ + defined(_M_IX86)) + if (TestEnv("LIBYUV_DISABLE_X86")) { + cpu_info &= ~libyuv::kCpuHasX86; + } + if (TestEnv("LIBYUV_DISABLE_SSE2")) { + cpu_info &= ~libyuv::kCpuHasSSE2; + } + if (TestEnv("LIBYUV_DISABLE_SSSE3")) { + cpu_info &= ~libyuv::kCpuHasSSSE3; + } + if (TestEnv("LIBYUV_DISABLE_SSE41")) { + cpu_info &= ~libyuv::kCpuHasSSE41; + } + if (TestEnv("LIBYUV_DISABLE_SSE42")) { + cpu_info &= ~libyuv::kCpuHasSSE42; + } + if (TestEnv("LIBYUV_DISABLE_AVX")) { + cpu_info &= ~libyuv::kCpuHasAVX; + } + if (TestEnv("LIBYUV_DISABLE_AVX2")) { + cpu_info &= ~libyuv::kCpuHasAVX2; + } + if (TestEnv("LIBYUV_DISABLE_ERMS")) { + cpu_info &= ~libyuv::kCpuHasERMS; + } + if (TestEnv("LIBYUV_DISABLE_FMA3")) { + cpu_info &= ~libyuv::kCpuHasFMA3; + } + if (TestEnv("LIBYUV_DISABLE_F16C")) { + cpu_info &= ~libyuv::kCpuHasF16C; + } + if (TestEnv("LIBYUV_DISABLE_AVX512BW")) { + cpu_info &= ~libyuv::kCpuHasAVX512BW; + } + if (TestEnv("LIBYUV_DISABLE_AVX512VL")) { + cpu_info &= ~libyuv::kCpuHasAVX512VL; + } + if (TestEnv("LIBYUV_DISABLE_AVX512VBMI")) { + cpu_info &= ~libyuv::kCpuHasAVX512VBMI; + } + if (TestEnv("LIBYUV_DISABLE_AVX512VBMI2")) { + cpu_info &= ~libyuv::kCpuHasAVX512VBMI2; + } + if (TestEnv("LIBYUV_DISABLE_AVX512VBITALG")) { + cpu_info &= ~libyuv::kCpuHasAVX512VBITALG; + } + if (TestEnv("LIBYUV_DISABLE_AVX512VPOPCNTDQ")) { + cpu_info &= ~libyuv::kCpuHasAVX512VPOPCNTDQ; + } + if (TestEnv("LIBYUV_DISABLE_GFNI")) { + cpu_info &= ~libyuv::kCpuHasGFNI; + } +#endif + if (TestEnv("LIBYUV_DISABLE_ASM")) { + cpu_info = libyuv::kCpuInitialized; + } + return cpu_info; +} + // For quicker unittests, default is 128 x 72. But when benchmarking, // default to 720p. Allow size to specify. // Set flags to -1 for benchmarking to avoid slower C code. @@ -87,6 +177,8 @@ LibYUVConvertTest::LibYUVConvertTest() if (FLAGS_libyuv_cpu_info) { benchmark_cpu_info_ = FLAGS_libyuv_cpu_info; } + disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_); + benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_); libyuv::MaskCpuFlags(benchmark_cpu_info_); benchmark_pixels_div1280_ = static_cast<int>((static_cast<double>(Abs(benchmark_width_)) * @@ -141,6 +233,8 @@ LibYUVColorTest::LibYUVColorTest() if (FLAGS_libyuv_cpu_info) { benchmark_cpu_info_ = FLAGS_libyuv_cpu_info; } + disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_); + benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_); libyuv::MaskCpuFlags(benchmark_cpu_info_); benchmark_pixels_div1280_ = static_cast<int>((static_cast<double>(Abs(benchmark_width_)) * @@ -195,6 +289,8 @@ LibYUVScaleTest::LibYUVScaleTest() if (FLAGS_libyuv_cpu_info) { benchmark_cpu_info_ = FLAGS_libyuv_cpu_info; } + disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_); + benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_); libyuv::MaskCpuFlags(benchmark_cpu_info_); benchmark_pixels_div1280_ = static_cast<int>((static_cast<double>(Abs(benchmark_width_)) * @@ -249,6 +345,8 @@ LibYUVRotateTest::LibYUVRotateTest() if (FLAGS_libyuv_cpu_info) { benchmark_cpu_info_ = FLAGS_libyuv_cpu_info; } + disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_); + benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_); libyuv::MaskCpuFlags(benchmark_cpu_info_); benchmark_pixels_div1280_ = static_cast<int>((static_cast<double>(Abs(benchmark_width_)) * @@ -303,6 +401,8 @@ LibYUVPlanarTest::LibYUVPlanarTest() if (FLAGS_libyuv_cpu_info) { benchmark_cpu_info_ = FLAGS_libyuv_cpu_info; } + disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_); + benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_); libyuv::MaskCpuFlags(benchmark_cpu_info_); benchmark_pixels_div1280_ = static_cast<int>((static_cast<double>(Abs(benchmark_width_)) * @@ -357,6 +457,8 @@ LibYUVBaseTest::LibYUVBaseTest() if (FLAGS_libyuv_cpu_info) { benchmark_cpu_info_ = FLAGS_libyuv_cpu_info; } + disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_); + benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_); libyuv::MaskCpuFlags(benchmark_cpu_info_); benchmark_pixels_div1280_ = static_cast<int>((static_cast<double>(Abs(benchmark_width_)) * @@ -411,6 +513,8 @@ LibYUVCompareTest::LibYUVCompareTest() if (FLAGS_libyuv_cpu_info) { benchmark_cpu_info_ = FLAGS_libyuv_cpu_info; } + disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_); + benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_); libyuv::MaskCpuFlags(benchmark_cpu_info_); benchmark_pixels_div1280_ = static_cast<int>((static_cast<double>(Abs(benchmark_width_)) * diff --git a/chromium/third_party/libyuv/unit_test/unit_test.h b/chromium/third_party/libyuv/unit_test/unit_test.h index dee3952fdc8..87907fa1603 100644 --- a/chromium/third_party/libyuv/unit_test/unit_test.h +++ b/chromium/third_party/libyuv/unit_test/unit_test.h @@ -126,9 +126,9 @@ class LibYUVColorTest : public ::testing::Test { protected: LibYUVColorTest(); - int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. - int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. - int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. + int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. + int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. + int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. int benchmark_pixels_div1280_; // Total pixels to benchmark / 1280. int disable_cpu_flags_; // Default 1. Use -1 for benchmarking. int benchmark_cpu_info_; // Default -1. Use 1 to disable SIMD. @@ -138,9 +138,9 @@ class LibYUVConvertTest : public ::testing::Test { protected: LibYUVConvertTest(); - int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. - int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. - int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. + int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. + int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. + int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. int benchmark_pixels_div1280_; // Total pixels to benchmark / 1280. int disable_cpu_flags_; // Default 1. Use -1 for benchmarking. int benchmark_cpu_info_; // Default -1. Use 1 to disable SIMD. @@ -150,9 +150,9 @@ class LibYUVScaleTest : public ::testing::Test { protected: LibYUVScaleTest(); - int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. - int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. - int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. + int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. + int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. + int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. int benchmark_pixels_div1280_; // Total pixels to benchmark / 1280. int disable_cpu_flags_; // Default 1. Use -1 for benchmarking. int benchmark_cpu_info_; // Default -1. Use 1 to disable SIMD. @@ -162,9 +162,9 @@ class LibYUVRotateTest : public ::testing::Test { protected: LibYUVRotateTest(); - int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. - int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. - int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. + int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. + int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. + int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. int benchmark_pixels_div1280_; // Total pixels to benchmark / 1280. int disable_cpu_flags_; // Default 1. Use -1 for benchmarking. int benchmark_cpu_info_; // Default -1. Use 1 to disable SIMD. @@ -174,9 +174,9 @@ class LibYUVPlanarTest : public ::testing::Test { protected: LibYUVPlanarTest(); - int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. - int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. - int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. + int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. + int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. + int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. int benchmark_pixels_div1280_; // Total pixels to benchmark / 1280. int disable_cpu_flags_; // Default 1. Use -1 for benchmarking. int benchmark_cpu_info_; // Default -1. Use 1 to disable SIMD. @@ -186,9 +186,9 @@ class LibYUVBaseTest : public ::testing::Test { protected: LibYUVBaseTest(); - int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. - int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. - int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. + int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. + int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. + int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. int benchmark_pixels_div1280_; // Total pixels to benchmark / 1280. int disable_cpu_flags_; // Default 1. Use -1 for benchmarking. int benchmark_cpu_info_; // Default -1. Use 1 to disable SIMD. @@ -198,9 +198,9 @@ class LibYUVCompareTest : public ::testing::Test { protected: LibYUVCompareTest(); - int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. - int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. - int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. + int benchmark_iterations_; // Default 1. Use 1000 for benchmarking. + int benchmark_width_; // Default 1280. Use 640 for benchmarking VGA. + int benchmark_height_; // Default 720. Use 360 for benchmarking VGA. int benchmark_pixels_div1280_; // Total pixels to benchmark / 1280. int disable_cpu_flags_; // Default 1. Use -1 for benchmarking. int benchmark_cpu_info_; // Default -1. Use 1 to disable SIMD. |