ben.carrasco/freeswitch
1
0
Fork 0
mirror of https://github.com/signalwire/freeswitch.git synced 2026-07-04 19:31:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

update libyuv to hash 5a699df5 from https://chromium.googlesource.com/libyuv/libyuv/

Browse Source
This commit is contained in:
Anthony Minessale
2019-01-17 20:13:46 +00:00
committed by Andrey Volk
parent a714dacd7f
commit 5a924d5ef3
191 changed files with 59375 additions and 33474 deletions
Download Patch File Download Diff File Expand all files Collapse all files
libs/libyuv/Android.bp
+135
View File
@@ -0,0 +1,135 @@
cc_library {
name: "libyuv",
vendor_available: true,
vndk: {
enabled: true,
},
srcs: [
"source/compare.cc",
"source/compare_common.cc",
"source/compare_gcc.cc",
"source/compare_neon.cc",
"source/compare_neon64.cc",
"source/compare_mmi.cc",
"source/compare_msa.cc",
"source/convert.cc",
"source/convert_argb.cc",
"source/convert_from.cc",
"source/convert_from_argb.cc",
"source/convert_to_argb.cc",
"source/convert_to_i420.cc",
"source/cpu_id.cc",
"source/planar_functions.cc",
"source/rotate.cc",
"source/rotate_any.cc",
"source/rotate_argb.cc",
"source/rotate_common.cc",
"source/rotate_gcc.cc",
"source/rotate_mmi.cc",
"source/rotate_msa.cc",
"source/rotate_neon.cc",
"source/rotate_neon64.cc",
"source/row_any.cc",
"source/row_common.cc",
"source/row_gcc.cc",
"source/row_mmi.cc",
"source/row_msa.cc",
"source/row_neon.cc",
"source/row_neon64.cc",
"source/scale.cc",
"source/scale_any.cc",
"source/scale_argb.cc",
"source/scale_common.cc",
"source/scale_gcc.cc",
"source/scale_mmi.cc",
"source/scale_msa.cc",
"source/scale_neon.cc",
"source/scale_neon64.cc",
"source/video_common.cc",
"source/convert_jpeg.cc",
"source/mjpeg_decoder.cc",
"source/mjpeg_validate.cc",
],
cflags: [
"-Wall",
"-Werror",
"-Wno-unused-parameter",
"-fexceptions",
"-DHAVE_JPEG",
],
shared_libs: ["libjpeg"],
export_include_dirs: ["include"],
}
// compatibilty static library until all uses of libyuv_static are replaced
// with libyuv (b/37646797)
cc_library_static {
name: "libyuv_static",
whole_static_libs: ["libyuv"],
}
cc_test {
name: "libyuv_unittest",
static_libs: ["libyuv"],
shared_libs: ["libjpeg"],
cflags: ["-Wall", "-Werror"],
srcs: [
"unit_test/unit_test.cc",
"unit_test/basictypes_test.cc",
"unit_test/color_test.cc",
"unit_test/compare_test.cc",
"unit_test/convert_test.cc",
"unit_test/cpu_test.cc",
"unit_test/cpu_thread_test.cc",
"unit_test/math_test.cc",
"unit_test/planar_test.cc",
"unit_test/rotate_argb_test.cc",
"unit_test/rotate_test.cc",
"unit_test/scale_argb_test.cc",
"unit_test/scale_test.cc",
"unit_test/video_common_test.cc",
],
}
cc_test {
name: "compare",
gtest: false,
srcs: [
"util/compare.cc",
],
static_libs: ["libyuv"],
}
cc_test {
name: "cpuid",
gtest: false,
srcs: [
"util/cpuid.c",
],
static_libs: ["libyuv"],
}
cc_test {
name: "psnr",
gtest: false,
srcs: [
"util/psnr_main.cc",
"util/psnr.cc",
"util/ssim.cc",
],
static_libs: ["libyuv"],
}
cc_test {
name: "yuvconvert",
gtest: false,
srcs: [
"util/yuvconvert.cc",
],
static_libs: ["libyuv"],
shared_libs: ["libjpeg"],
}
libs/libyuv/Android.mk
+61 -26
View File
@@ -1,4 +1,4 @@
# This is the Android makefile for libyuv for both platform and NDK.
# This is the Android makefile for libyuv for NDK.
LOCAL_PATH:= $(call my-dir)
include $(CLEAR_VARS)
@@ -8,8 +8,11 @@ LOCAL_CPP_EXTENSION := .cc
LOCAL_SRC_FILES := \
source/compare.cc \
source/compare_common.cc \
source/compare_neon64.cc \
source/compare_gcc.cc \
source/compare_mmi.cc \
source/compare_msa.cc \
source/compare_neon.cc \
source/compare_neon64.cc \
source/convert.cc \
source/convert_argb.cc \
source/convert_from.cc \
@@ -22,48 +25,80 @@ LOCAL_SRC_FILES := \
source/rotate_any.cc \
source/rotate_argb.cc \
source/rotate_common.cc \
source/rotate_mips.cc \
source/rotate_neon64.cc \
source/rotate_gcc.cc \
source/rotate_mmi.cc \
source/rotate_msa.cc \
source/rotate_neon.cc \
source/rotate_neon64.cc \
source/row_any.cc \
source/row_common.cc \
source/row_mips.cc \
source/row_gcc.cc \
source/row_mmi.cc \
source/row_msa.cc \
source/row_neon.cc \
source/row_neon64.cc \
source/row_gcc.cc \
source/scale.cc \
source/scale_any.cc \
source/scale_argb.cc \
source/scale_common.cc \
source/scale_mips.cc \
source/scale_neon64.cc \
source/scale_gcc.cc \
source/scale_mmi.cc \
source/scale_msa.cc \
source/scale_neon.cc \
source/scale_neon64.cc \
source/video_common.cc
# TODO(fbarchard): Enable mjpeg encoder.
# source/mjpeg_decoder.cc
# source/convert_jpeg.cc
# source/mjpeg_validate.cc
ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)
LOCAL_CFLAGS += -DLIBYUV_NEON
LOCAL_SRC_FILES += \
source/compare_neon.cc.neon \
source/rotate_neon.cc.neon \
source/row_neon.cc.neon \
source/scale_neon.cc.neon
endif
ifeq ($(TARGET_ARCH_ABI),mips)
LOCAL_CFLAGS += -DLIBYUV_MSA
LOCAL_SRC_FILES += \
source/row_msa.cc
common_CFLAGS := -Wall -fexceptions
ifneq ($(LIBYUV_DISABLE_JPEG), "yes")
LOCAL_SRC_FILES += \
source/convert_jpeg.cc \
source/mjpeg_decoder.cc \
source/mjpeg_validate.cc
common_CFLAGS += -DHAVE_JPEG
LOCAL_SHARED_LIBRARIES := libjpeg
endif
LOCAL_CFLAGS += $(common_CFLAGS)
LOCAL_EXPORT_C_INCLUDES := $(LOCAL_PATH)/include
LOCAL_C_INCLUDES += $(LOCAL_PATH)/include
LOCAL_EXPORT_C_INCLUDE_DIRS := $(LOCAL_PATH)/include
LOCAL_MODULE := libyuv_static
LOCAL_MODULE_TAGS := optional
include $(BUILD_STATIC_LIBRARY)
include $(CLEAR_VARS)
LOCAL_WHOLE_STATIC_LIBRARIES := libyuv_static
LOCAL_MODULE := libyuv
ifneq ($(LIBYUV_DISABLE_JPEG), "yes")
LOCAL_SHARED_LIBRARIES := libjpeg
endif
include $(BUILD_SHARED_LIBRARY)
include $(CLEAR_VARS)
LOCAL_STATIC_LIBRARIES := libyuv_static
LOCAL_SHARED_LIBRARIES := libjpeg
LOCAL_MODULE_TAGS := tests
LOCAL_CPP_EXTENSION := .cc
LOCAL_C_INCLUDES += $(LOCAL_PATH)/include
LOCAL_SRC_FILES := \
unit_test/unit_test.cc \
unit_test/basictypes_test.cc \
unit_test/color_test.cc \
unit_test/compare_test.cc \
unit_test/convert_test.cc \
unit_test/cpu_test.cc \
unit_test/cpu_thread_test.cc \
unit_test/math_test.cc \
unit_test/planar_test.cc \
unit_test/rotate_argb_test.cc \
unit_test/rotate_test.cc \
unit_test/scale_argb_test.cc \
unit_test/scale_test.cc \
unit_test/video_common_test.cc
LOCAL_MODULE := libyuv_unittest
include $(BUILD_NATIVE_TEST)
libs/libyuv/BUILD.gn
+166 -41
View File
@@ -9,9 +9,19 @@
import("libyuv.gni")
import("//testing/test.gni")
declare_args() {
# Set to false to disable building with gflags.
libyuv_use_gflags = true
# When building a shared library using a target in WebRTC or
# Chromium projects that depends on libyuv, setting this flag
# to true makes libyuv symbols visible inside that library.
libyuv_symbols_visible = false
}
config("libyuv_config") {
include_dirs = [ "include" ]
if (is_android && current_cpu=="arm64") {
if (is_android && current_cpu == "arm64") {
ldflags = [ "-Wl,--dynamic-linker,/system/bin/linker64" ]
}
if (is_android && current_cpu != "arm64") {
@@ -19,7 +29,61 @@ config("libyuv_config") {
}
}
static_library("libyuv") {
# This target is built when no specific target is specified on the command line.
group("default") {
testonly = true
deps = [
":libyuv",
]
if (libyuv_include_tests) {
deps += [
":compare",
":cpuid",
":libyuv_unittest",
":psnr",
":yuvconvert",
]
}
}
group("libyuv") {
all_dependent_configs = [ ":libyuv_config" ]
deps = []
if (is_win && target_cpu == "x64") {
# Compile with clang in order to get inline assembly
public_deps = [
":libyuv_internal(//build/toolchain/win:win_clang_x64)",
]
} else {
public_deps = [
":libyuv_internal",
]
}
if (libyuv_use_neon) {
deps += [ ":libyuv_neon" ]
}
if (libyuv_use_msa) {
deps += [ ":libyuv_msa" ]
}
if (libyuv_use_mmi) {
deps += [ ":libyuv_mmi" ]
}
if (!is_ios) {
# Make sure that clients of libyuv link with libjpeg. This can't go in
# libyuv_internal because in Windows x64 builds that will generate a clang
# build of libjpeg, and we don't want two copies.
deps += [ "//third_party:jpeg" ]
}
}
static_library("libyuv_internal") {
visibility = [ ":*" ]
sources = [
# Headers
"include/libyuv.h",
@@ -62,47 +126,56 @@ static_library("libyuv") {
"source/rotate_any.cc",
"source/rotate_argb.cc",
"source/rotate_common.cc",
"source/rotate_mips.cc",
"source/rotate_gcc.cc",
"source/rotate_win.cc",
"source/row_any.cc",
"source/row_common.cc",
"source/row_mips.cc",
"source/row_gcc.cc",
"source/row_win.cc",
"source/scale.cc",
"source/scale_any.cc",
"source/scale_argb.cc",
"source/scale_common.cc",
"source/scale_mips.cc",
"source/scale_gcc.cc",
"source/scale_win.cc",
"source/video_common.cc",
]
public_configs = [ ":libyuv_config" ]
configs += [ ":libyuv_config" ]
defines = []
deps = []
if (libyuv_symbols_visible) {
configs -= [ "//build/config/gcc:symbol_visibility_hidden" ]
configs += [ "//build/config/gcc:symbol_visibility_default" ]
}
if (!is_ios) {
defines += [ "HAVE_JPEG" ]
deps += [ "//third_party:jpeg" ]
# Needed to pull in libjpeg headers. Can't add //third_party:jpeg to deps
# because in Windows x64 build it will get compiled with clang.
deps += [ "//third_party:jpeg_includes" ]
}
if (libyuv_use_neon) {
deps += [ ":libyuv_neon" ]
}
if (libyuv_use_msa) {
deps += [ ":libyuv_msa" ]
}
if (is_nacl) {
# Always enable optimization under NaCl to workaround crbug.com/538243 .
# Always enable optimization for Release and NaCl builds (to workaround
# crbug.com/538243).
if (!is_debug || is_nacl) {
configs -= [ "//build/config/compiler:default_optimization" ]
# Enable optimize for speed (-O2) over size (-Os).
configs += [ "//build/config/compiler:optimize_max" ]
}
# To enable AVX2 or other cpu optimization, pass flag here
if (!is_win) {
cflags = [
# "-mpopcnt",
# "-mavx2",
# "-mfma",
"-ffp-contract=fast", # Enable fma vectorization for NEON.
]
}
}
if (libyuv_use_neon) {
@@ -119,8 +192,22 @@ if (libyuv_use_neon) {
"source/scale_neon64.cc",
]
deps = [
":libyuv_internal",
]
public_configs = [ ":libyuv_config" ]
# Always enable optimization for Release and NaCl builds (to workaround
# crbug.com/538243).
if (!is_debug) {
configs -= [ "//build/config/compiler:default_optimization" ]
# Enable optimize for speed (-O2) over size (-Os).
# TODO(fbarchard): Consider optimize_speed which is O3.
configs += [ "//build/config/compiler:optimize_max" ]
}
if (current_cpu != "arm64") {
configs -= [ "//build/config/compiler:compiler_arm_fpu" ]
cflags = [ "-mfpu=neon" ]
@@ -132,7 +219,32 @@ if (libyuv_use_msa) {
static_library("libyuv_msa") {
sources = [
# MSA Source Files
"source/compare_msa.cc",
"source/rotate_msa.cc",
"source/row_msa.cc",
"source/scale_msa.cc",
]
deps = [
":libyuv_internal",
]
public_configs = [ ":libyuv_config" ]
}
}
if (libyuv_use_mmi) {
static_library("libyuv_mmi") {
sources = [
# MMI Source Files
"source/compare_mmi.cc",
"source/rotate_mmi.cc",
"source/row_mmi.cc",
"source/scale_mmi.cc",
]
deps = [
":libyuv_internal",
]
public_configs = [ ":libyuv_config" ]
@@ -145,13 +257,13 @@ if (libyuv_include_tests) {
cflags = [
# TODO(fbarchard): Fix sign and unused variable warnings.
"-Wno-sign-compare",
"-Wno-unused-variable"
"-Wno-unused-variable",
]
}
if (is_win) {
cflags = [
"/wd4245", # signed/unsigned mismatch
"/wd4189", # local variable is initialized but not referenced
"/wd4245", # signed/unsigned mismatch
"/wd4189", # local variable is initialized but not referenced
]
}
}
@@ -163,14 +275,14 @@ if (libyuv_include_tests) {
testonly = true
sources = [
# headers
"unit_test/unit_test.h",
# sources
# headers
"unit_test/basictypes_test.cc",
"unit_test/compare_test.cc",
"unit_test/color_test.cc",
"unit_test/compare_test.cc",
"unit_test/convert_test.cc",
"unit_test/cpu_test.cc",
"unit_test/cpu_thread_test.cc",
"unit_test/math_test.cc",
"unit_test/planar_test.cc",
"unit_test/rotate_argb_test.cc",
@@ -178,22 +290,28 @@ if (libyuv_include_tests) {
"unit_test/scale_argb_test.cc",
"unit_test/scale_test.cc",
"unit_test/unit_test.cc",
"unit_test/unit_test.h",
"unit_test/video_common_test.cc",
]
deps = [
":libyuv",
"//testing/gtest",
"//third_party/gflags",
]
defines = []
if (libyuv_use_gflags) {
defines += [ "LIBYUV_USE_GFLAGS" ]
deps += [ "//third_party/gflags" ]
}
configs += [ ":libyuv_unittest_warnings_config" ]
public_deps = [ "//testing/gtest" ]
public_deps = [
"//testing/gtest",
]
public_configs = [ ":libyuv_unittest_config" ]
defines = []
if (is_linux) {
cflags = [ "-fexceptions" ]
}
@@ -211,8 +329,8 @@ if (libyuv_include_tests) {
# TODO(YangZhang): These lines can be removed when high accuracy
# YUV to RGB to Neon is ported.
if ((target_cpu=="armv7" || target_cpu=="armv7s" ||
(target_cpu=="arm" && arm_version >= 7) || target_cpu=="arm64") &&
if ((target_cpu == "armv7" || target_cpu == "armv7s" ||
(target_cpu == "arm" && arm_version >= 7) || target_cpu == "arm64") &&
(arm_use_neon || arm_optionally_use_neon)) {
defines += [ "LIBYUV_NEON" ]
}
@@ -221,7 +339,6 @@ if (libyuv_include_tests) {
# Enable the following 3 macros to turn off assembly for specified CPU.
# "LIBYUV_DISABLE_X86",
# "LIBYUV_DISABLE_NEON",
# "LIBYUV_DISABLE_MIPS",
# Enable the following macro to build libyuv as a shared library (dll).
# "LIBYUV_USING_SHARED_LIBRARY"
]
@@ -230,20 +347,24 @@ if (libyuv_include_tests) {
executable("compare") {
sources = [
# sources
"util/compare.cc"
"util/compare.cc",
]
deps = [
":libyuv",
]
deps = [ ":libyuv" ]
if (is_linux) {
cflags = [ "-fexceptions" ]
}
}
executable("convert") {
executable("yuvconvert") {
sources = [
# sources
"util/convert.cc"
"util/yuvconvert.cc",
]
deps = [
":libyuv",
]
deps = [ ":libyuv" ]
if (is_linux) {
cflags = [ "-fexceptions" ]
}
@@ -252,11 +373,13 @@ if (libyuv_include_tests) {
executable("psnr") {
sources = [
# sources
"util/psnr_main.cc",
"util/psnr.cc",
"util/ssim.cc"
"util/psnr_main.cc",
"util/ssim.cc",
]
deps = [
":libyuv",
]
deps = [ ":libyuv" ]
if (!is_ios && !libyuv_disable_jpeg) {
defines = [ "HAVE_JPEG" ]
@@ -266,8 +389,10 @@ if (libyuv_include_tests) {
executable("cpuid") {
sources = [
# sources
"util/cpuid.c"
"util/cpuid.c",
]
deps = [
":libyuv",
]
deps = [ ":libyuv" ]
}
}
libs/libyuv/CM_linux_packages.cmake
+69
View File
@@ -0,0 +1,69 @@
# determine the version number from the #define in libyuv/version.h
EXECUTE_PROCESS (
COMMAND grep --perl-regex --only-matching "(?<=LIBYUV_VERSION )[0-9]+" include/libyuv/version.h
WORKING_DIRECTORY ${PROJECT_SOURCE_DIR}
OUTPUT_VARIABLE YUV_VERSION_NUMBER
OUTPUT_STRIP_TRAILING_WHITESPACE )
SET ( YUV_VER_MAJOR 0 )
SET ( YUV_VER_MINOR 0 )
SET ( YUV_VER_PATCH ${YUV_VERSION_NUMBER} )
SET ( YUV_VERSION ${YUV_VER_MAJOR}.${YUV_VER_MINOR}.${YUV_VER_PATCH} )
MESSAGE ( "Building ver.: ${YUV_VERSION}" )
# is this a 32-bit or 64-bit build?
IF ( CMAKE_SIZEOF_VOID_P EQUAL 8 )
SET ( YUV_BIT_SIZE 64 )
ELSEIF ( CMAKE_SIZEOF_VOID_P EQUAL 4 )
SET ( YUV_BIT_SIZE 32 )
ELSE ()
MESSAGE ( FATAL_ERROR "CMAKE_SIZEOF_VOID_P=${CMAKE_SIZEOF_VOID_P}" )
ENDIF ()
# detect if this is a ARM build
STRING (FIND "${CMAKE_CXX_COMPILER}" "arm-linux-gnueabihf-g++" pos)
IF ( ${pos} EQUAL -1 )
SET ( YUV_CROSS_COMPILE_FOR_ARM7 FALSE )
ELSE ()
MESSAGE ( "Cross compiling for ARM7" )
SET ( YUV_CROSS_COMPILE_FOR_ARM7 TRUE )
ENDIF ()
STRING (FIND "${CMAKE_SYSTEM_PROCESSOR}" "arm" pos)
IF ( ${pos} EQUAL -1 )
SET ( YUV_COMPILE_FOR_ARM7 FALSE )
ELSE ()
MESSAGE ( "Compiling for ARM" )
SET ( YUV_COMPILE_FOR_ARM7 TRUE )
ENDIF ()
# setup the sytem name, such as "x86-32", "amd-64", and "arm-32
IF ( ${YUV_CROSS_COMPILE_FOR_ARM7} OR ${YUV_COMPILE_FOR_ARM7} )
SET ( YUV_SYSTEM_NAME "armhf-${YUV_BIT_SIZE}" )
ELSE ()
IF ( YUV_BIT_SIZE EQUAL 32 )
SET ( YUV_SYSTEM_NAME "x86-${YUV_BIT_SIZE}" )
ELSE ()
SET ( YUV_SYSTEM_NAME "amd-${YUV_BIT_SIZE}" )
ENDIF ()
ENDIF ()
MESSAGE ( "Packaging for: ${YUV_SYSTEM_NAME}" )
# define all the variables needed by CPack to create .deb and .rpm packages
SET ( CPACK_PACKAGE_VENDOR "Frank Barchard" )
SET ( CPACK_PACKAGE_CONTACT "fbarchard@chromium.org" )
SET ( CPACK_PACKAGE_VERSION ${YUV_VERSION} )
SET ( CPACK_PACKAGE_VERSION_MAJOR ${YUV_VER_MAJOR} )
SET ( CPACK_PACKAGE_VERSION_MINOR ${YUV_VER_MINOR} )
SET ( CPACK_PACKAGE_VERSION_PATCH ${YUV_VER_PATCH} )
SET ( CPACK_RESOURCE_FILE_LICENSE ${PROJECT_SOURCE_DIR}/LICENSE )
SET ( CPACK_SYSTEM_NAME "linux-${YUV_SYSTEM_NAME}" )
SET ( CPACK_PACKAGE_NAME "libyuv" )
SET ( CPACK_PACKAGE_DESCRIPTION_SUMMARY "YUV library" )
SET ( CPACK_PACKAGE_DESCRIPTION "YUV library and YUV conversion tool" )
SET ( CPACK_DEBIAN_PACKAGE_SECTION "other" )
SET ( CPACK_DEBIAN_PACKAGE_PRIORITY "optional" )
SET ( CPACK_DEBIAN_PACKAGE_MAINTAINER "Frank Barchard <fbarchard@chromium.org>" )
SET ( CPACK_GENERATOR "DEB;RPM" )
# create the .deb and .rpm files (you'll need build-essential and rpm tools)
INCLUDE( CPack )
libs/libyuv/CMakeLists.txt
+51 -104
View File
@@ -1,112 +1,45 @@
cmake_minimum_required(VERSION 2.8)
# CMakeLists for libyuv
# Originally created for "roxlu build system" to compile libyuv on windows
# Run with -DTEST=ON to build unit tests
option(TEST "Built unit tests" OFF)
set(ly_base_dir ${CMAKE_CURRENT_LIST_DIR})
set(ly_src_dir ${ly_base_dir}/source/)
set(ly_inc_dir ${ly_base_dir}/include)
set(ly_lib_name "yuv")
PROJECT ( YUV C CXX ) # "C" is required even for C++ projects
CMAKE_MINIMUM_REQUIRED( VERSION 2.8 )
OPTION( TEST "Built unit tests" OFF )
set(ly_source_files
${ly_src_dir}/compare.cc
${ly_src_dir}/compare_common.cc
${ly_src_dir}/compare_neon.cc
${ly_src_dir}/compare_neon64.cc
${ly_src_dir}/compare_gcc.cc
${ly_src_dir}/compare_win.cc
${ly_src_dir}/convert.cc
${ly_src_dir}/convert_argb.cc
${ly_src_dir}/convert_from.cc
${ly_src_dir}/convert_from_argb.cc
${ly_src_dir}/convert_jpeg.cc
${ly_src_dir}/convert_to_argb.cc
${ly_src_dir}/convert_to_i420.cc
${ly_src_dir}/cpu_id.cc
${ly_src_dir}/mjpeg_decoder.cc
${ly_src_dir}/mjpeg_validate.cc
${ly_src_dir}/planar_functions.cc
${ly_src_dir}/rotate.cc
${ly_src_dir}/rotate_any.cc
${ly_src_dir}/rotate_argb.cc
${ly_src_dir}/rotate_common.cc
${ly_src_dir}/rotate_mips.cc
${ly_src_dir}/rotate_neon.cc
${ly_src_dir}/rotate_neon64.cc
${ly_src_dir}/rotate_gcc.cc
${ly_src_dir}/rotate_win.cc
${ly_src_dir}/row_any.cc
${ly_src_dir}/row_common.cc
${ly_src_dir}/row_mips.cc
${ly_src_dir}/row_msa.cc
${ly_src_dir}/row_neon.cc
${ly_src_dir}/row_neon64.cc
${ly_src_dir}/row_gcc.cc
${ly_src_dir}/row_win.cc
${ly_src_dir}/scale.cc
${ly_src_dir}/scale_any.cc
${ly_src_dir}/scale_argb.cc
${ly_src_dir}/scale_common.cc
${ly_src_dir}/scale_mips.cc
${ly_src_dir}/scale_neon.cc
${ly_src_dir}/scale_neon64.cc
${ly_src_dir}/scale_gcc.cc
${ly_src_dir}/scale_win.cc
${ly_src_dir}/video_common.cc
)
SET ( ly_base_dir ${PROJECT_SOURCE_DIR} )
SET ( ly_src_dir ${ly_base_dir}/source )
SET ( ly_inc_dir ${ly_base_dir}/include )
SET ( ly_tst_dir ${ly_base_dir}/unit_test )
SET ( ly_lib_name yuv )
SET ( ly_lib_static ${ly_lib_name} )
SET ( ly_lib_shared ${ly_lib_name}_shared )
set(ly_unittest_sources
${ly_base_dir}/unit_test/basictypes_test.cc
${ly_base_dir}/unit_test/color_test.cc
${ly_base_dir}/unit_test/compare_test.cc
${ly_base_dir}/unit_test/convert_test.cc
${ly_base_dir}/unit_test/cpu_test.cc
${ly_base_dir}/unit_test/math_test.cc
${ly_base_dir}/unit_test/planar_test.cc
${ly_base_dir}/unit_test/rotate_argb_test.cc
${ly_base_dir}/unit_test/rotate_test.cc
${ly_base_dir}/unit_test/scale_argb_test.cc
${ly_base_dir}/unit_test/scale_test.cc
${ly_base_dir}/unit_test/unit_test.cc
${ly_base_dir}/unit_test/video_common_test.cc
)
FILE ( GLOB_RECURSE ly_source_files ${ly_src_dir}/*.cc )
LIST ( SORT ly_source_files )
set(ly_header_files
${ly_inc_dir}/libyuv/basic_types.h
${ly_inc_dir}/libyuv/compare.h
${ly_inc_dir}/libyuv/convert.h
${ly_inc_dir}/libyuv/convert_argb.h
${ly_inc_dir}/libyuv/convert_from.h
${ly_inc_dir}/libyuv/convert_from_argb.h
${ly_inc_dir}/libyuv/cpu_id.h
${ly_inc_dir}/libyuv/macros_msa.h
${ly_inc_dir}/libyuv/planar_functions.h
${ly_inc_dir}/libyuv/rotate.h
${ly_inc_dir}/libyuv/rotate_argb.h
${ly_inc_dir}/libyuv/rotate_row.h
${ly_inc_dir}/libyuv/row.h
${ly_inc_dir}/libyuv/scale.h
${ly_inc_dir}/libyuv/scale_argb.h
${ly_inc_dir}/libyuv/scale_row.h
${ly_inc_dir}/libyuv/version.h
${ly_inc_dir}/libyuv/video_common.h
${ly_inc_dir}/libyuv/mjpeg_decoder.h
)
FILE ( GLOB_RECURSE ly_unittest_sources ${ly_tst_dir}/*.cc )
LIST ( SORT ly_unittest_sources )
include_directories(${ly_inc_dir})
INCLUDE_DIRECTORIES( BEFORE ${ly_inc_dir} )
add_library(${ly_lib_name} STATIC ${ly_source_files})
# this creates the static library (.a)
ADD_LIBRARY ( ${ly_lib_static} STATIC ${ly_source_files} )
add_executable(convert ${ly_base_dir}/util/convert.cc)
target_link_libraries(convert ${ly_lib_name})
# this creates the shared library (.so)
ADD_LIBRARY ( ${ly_lib_shared} SHARED ${ly_source_files} )
SET_TARGET_PROPERTIES ( ${ly_lib_shared} PROPERTIES OUTPUT_NAME "${ly_lib_name}" )
SET_TARGET_PROPERTIES ( ${ly_lib_shared} PROPERTIES PREFIX "lib" )
include(FindJPEG)
# this creates the conversion tool
ADD_EXECUTABLE ( yuvconvert ${ly_base_dir}/util/yuvconvert.cc )
TARGET_LINK_LIBRARIES ( yuvconvert ${ly_lib_static} )
INCLUDE ( FindJPEG )
if (JPEG_FOUND)
include_directories(${JPEG_INCLUDE_DIR})
target_link_libraries(convert ${JPEG_LIBRARY})
add_definitions(-DHAVE_JPEG)
include_directories( ${JPEG_INCLUDE_DIR} )
target_link_libraries( yuvconvert ${JPEG_LIBRARY} )
add_definitions( -DHAVE_JPEG )
endif()
if(TEST)
@@ -126,19 +59,33 @@ 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()
if(NACL AND NACL_LIBC STREQUAL "newlib")
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(TARGETS ${ly_lib_name} DESTINATION lib)
install(FILES ${ly_header_files} DESTINATION include/libyuv)
install(FILES ${ly_inc_dir}/libyuv.h DESTINATION include/)
# 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 RUNTIME DESTINATION bin )
INSTALL ( DIRECTORY ${PROJECT_SOURCE_DIR}/include/ DESTINATION include )
# create the .deb and .rpm packages using cpack
INCLUDE ( CM_linux_packages.cmake )
libs/libyuv/DEPS
+1007 -28
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/LICENSE_THIRD_PARTY
-8
View File
@@ -1,8 +0,0 @@
This source tree contains third party source code which is governed by third
party licenses. This file contains references to files which are under other
licenses than the one provided in the LICENSE file in the root of the source
tree.
Files governed by third party licenses:
source/x86inc.asm
libs/libyuv/OWNERS
+2 -7
View File
@@ -1,13 +1,8 @@
fbarchard@chromium.org
magjed@chromium.org
torbjorng@chromium.org
per-file *.gyp=kjellander@chromium.org
per-file *.gn=kjellander@chromium.org
per-file *.gn=phoglund@chromium.org
per-file .gitignore=*
per-file AUTHORS=*
per-file DEPS=*
per-file PRESUBMIT.py=kjellander@chromium.org
per-file gyp_libyuv.py=kjellander@chromium.org
per-file setup_links.py=*
per-file sync_chromium.py=kjellander@chromium.org
per-file PRESUBMIT.py=phoglund@chromium.org
libs/libyuv/PRESUBMIT.py
Executable → Regular
+61 -54
View File
@@ -1,4 +1,4 @@
# Copyright 2014 The LibYuv Project Authors. All rights reserved.
# Copyright 2017 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
@@ -6,60 +6,67 @@
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
import re
import sys
import os
def GetDefaultTryConfigs(bots=None):
"""Returns a list of ('bot', set(['tests']), optionally filtered by [bots].
def _RunPythonTests(input_api, output_api):
def join(*args):
return input_api.os_path.join(input_api.PresubmitLocalPath(), *args)
For WebRTC purposes, we always return an empty list of tests, since we want
to run all tests by default on all our trybots.
"""
return { 'tryserver.libyuv': dict((bot, []) for bot in bots)}
# pylint: disable=W0613
def GetPreferredTryMasters(project, change):
files = change.LocalPaths()
bots = [
'win',
'win_rel',
'win_x64_rel',
'win_x64_gn',
'win_x64_gn_rel',
'win_clang',
'win_clang_rel',
'win_x64_clang_rel',
'mac',
'mac_rel',
'mac_gn',
'mac_gn_rel',
'mac_asan',
'ios',
'ios_rel',
'ios_arm64',
'ios_arm64_rel',
'linux',
'linux_rel',
'linux_gn',
'linux_gn_rel',
'linux_memcheck',
'linux_tsan2',
'linux_asan',
'linux_msan',
'linux_ubsan',
'linux_ubsan_vptr',
'android',
'android_rel',
'android_clang',
'android_arm64',
'android_mips',
'android_x64',
'android_x86',
'android_gn',
'android_gn_rel',
test_directories = [
root for root, _, files in os.walk(join('tools_libyuv'))
if any(f.endswith('_test.py') for f in files)
]
if not files or all(re.search(r'[\\/]OWNERS$', f) for f in files):
return {}
return GetDefaultTryConfigs(bots)
tests = []
for directory in test_directories:
tests.extend(
input_api.canned_checks.GetUnitTestsInDirectory(
input_api,
output_api,
directory,
whitelist=[r'.+_test\.py$']))
return input_api.RunTests(tests, parallel=True)
def _CommonChecks(input_api, output_api):
"""Checks common to both upload and commit."""
results = []
results.extend(input_api.canned_checks.RunPylint(input_api, output_api,
black_list=(r'^base[\\\/].*\.py$',
r'^build[\\\/].*\.py$',
r'^buildtools[\\\/].*\.py$',
r'^ios[\\\/].*\.py$',
r'^out.*[\\\/].*\.py$',
r'^testing[\\\/].*\.py$',
r'^third_party[\\\/].*\.py$',
r'^tools[\\\/].*\.py$',
# TODO(kjellander): should arguably be checked.
r'^tools_libyuv[\\\/]valgrind[\\\/].*\.py$',
r'^xcodebuild.*[\\\/].*\.py$',),
disabled_warnings=['F0401', # Failed to import x
'E0611', # No package y in x
'W0232', # Class has no __init__ method
],
pylintrc='pylintrc'))
results.extend(_RunPythonTests(input_api, output_api))
return results
def CheckChangeOnUpload(input_api, output_api):
results = []
results.extend(_CommonChecks(input_api, output_api))
results.extend(
input_api.canned_checks.CheckGNFormatted(input_api, output_api))
return results
def CheckChangeOnCommit(input_api, output_api):
results = []
results.extend(_CommonChecks(input_api, output_api))
results.extend(input_api.canned_checks.CheckOwners(input_api, output_api))
results.extend(input_api.canned_checks.CheckChangeWasUploaded(
input_api, output_api))
results.extend(input_api.canned_checks.CheckChangeHasDescription(
input_api, output_api))
return results
libs/libyuv/README.chromium
+1 -1
View File
@@ -1,6 +1,6 @@
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 1620
Version: 1724
License: BSD
License File: LICENSE
libs/libyuv/README.md
+8 -8
View File
@@ -1,18 +1,18 @@
**libyuv** is an open source project that includes YUV scaling and conversion functionality.
* Scale YUV to prepare content for compression, with point, bilinear or box filter.
* Convert to YUV from webcam formats.
* Convert from YUV to formats for rendering/effects.
* Convert to YUV from webcam formats for compression.
* Convert to RGB formats for rendering/effects.
* Rotate by 90/180/270 degrees to adjust for mobile devices in portrait mode.
* Optimized for SSE2/SSSE3/AVX2 on x86/x64.
* Optimized for SSSE3/AVX2 on x86/x64.
* Optimized for Neon on Arm.
* Optimized for DSP R2 on Mips.
* Optimized for MSA on Mips.
### Development
See [Getting started] [1] for instructions on how to get started developing.
See [Getting started][1] for instructions on how to get started developing.
You can also browse the [docs directory] [2] for more documentation.
You can also browse the [docs directory][2] for more documentation.
[1]: https://chromium.googlesource.com/libyuv/libyuv/+/master/docs/getting_started.md
[2]: https://chromium.googlesource.com/libyuv/libyuv/+/master/docs/
[1]: ./docs/getting_started.md
[2]: ./docs/
libs/libyuv/build_overrides/build.gni
+31 -7
View File
@@ -6,13 +6,8 @@
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
# Using same overrides as WebRTC
# See https://bugs.chromium.org/p/webrtc/issues/detail?id=5453.
# Some WebRTC targets require the 10.7 deployment version of the Mac SDK and a
# 10.11 min SDK but those targets are only used in non-Chromium builds. We can
# remove this when Chromium drops 10.6 support and also requires 10.7.
mac_sdk_min_build_override = "10.11"
mac_deployment_target_build_override = "10.7"
# Some non-Chromium builds don't use Chromium's third_party/binutils.
linux_use_bundled_binutils_override = true
# Variable that can be used to support multiple build scenarios, like having
# Chromium specific targets in a client project's GN file etc.
@@ -20,3 +15,32 @@ build_with_chromium = false
# Some non-Chromium builds don't support building java targets.
enable_java_templates = true
# Allow using custom suppressions files (currently not used by libyuv).
asan_suppressions_file = "//build/sanitizers/asan_suppressions.cc"
lsan_suppressions_file = "//build/sanitizers/lsan_suppressions.cc"
tsan_suppressions_file = "//build/sanitizers/tsan_suppressions.cc"
msan_blacklist_path =
rebase_path("//tools_libyuv/msan/blacklist.txt", root_build_dir)
ubsan_blacklist_path =
rebase_path("//tools_libyuv/ubsan/blacklist.txt", root_build_dir)
ubsan_vptr_blacklist_path =
rebase_path("//tools_libyuv/ubsan/vptr_blacklist.txt", root_build_dir)
# For Chromium, Android 32-bit non-component, non-clang builds hit a 4GiB size
# limit, making them requiring symbol_level=2. WebRTC doesn't hit that problem
# so we just ignore that assert. See https://crbug.com/648948 for more info.
ignore_elf32_limitations = 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
}
libs/libyuv/chromium/.gclient
-20
View File
@@ -1,20 +0,0 @@
solutions = [{
'name': 'src',
'url': 'https://chromium.googlesource.com/chromium/src.git',
'deps_file': '.DEPS.git',
'managed': False,
'custom_deps': {
# Skip syncing some large dependencies Libyuv will never need.
'src/third_party/cld_2/src': None,
'src/third_party/ffmpeg': None,
'src/third_party/hunspell_dictionaries': None,
'src/third_party/liblouis/src': None,
'src/third_party/pdfium': None,
'src/third_party/skia': None,
'src/third_party/trace-viewer': None,
'src/third_party/webrtc': None,
},
'safesync_url': ''
}]
cache_dir = None
libs/libyuv/chromium/README
-5
View File
@@ -1,5 +0,0 @@
This .gclient file is used to do download a copy of Chromium.
Libyuv uses the Chromium build toolchain and a number of shared
dependencies by creating symlinks to folders in this checkout,
using the ../setup_links.py script.
libs/libyuv/cleanup_links.py
Executable
+107
View File
@@ -0,0 +1,107 @@
#!/usr/bin/env python
# Copyright 2017 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
# This is a copy of the file from WebRTC in:
# https://chromium.googlesource.com/external/webrtc/+/master/cleanup_links.py
"""Script to cleanup symlinks created from setup_links.py.
Before 177567c518b121731e507e9b9c4049c4dc96e4c8 (#15754) we had a Chromium
checkout which we created symlinks into. In order to do clean syncs after
landing that change, this script cleans up any old symlinks, avoiding annoying
manual cleanup needed in order to complete gclient sync.
"""
import logging
import optparse
import os
import shelve
import subprocess
import sys
ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
LINKS_DB = 'links'
# Version management to make future upgrades/downgrades easier to support.
SCHEMA_VERSION = 1
class WebRTCLinkSetup(object):
def __init__(self, links_db, dry_run=False):
self._dry_run = dry_run
self._links_db = links_db
def CleanupLinks(self):
logging.debug('CleanupLinks')
for source, link_path in self._links_db.iteritems():
if source == 'SCHEMA_VERSION':
continue
if os.path.islink(link_path) or sys.platform.startswith('win'):
# os.path.islink() always returns false on Windows
# See http://bugs.python.org/issue13143.
logging.debug('Removing link to %s at %s', source, link_path)
if not self._dry_run:
if os.path.exists(link_path):
if sys.platform.startswith('win') and os.path.isdir(link_path):
subprocess.check_call(['rmdir', '/q', '/s', link_path],
shell=True)
else:
os.remove(link_path)
del self._links_db[source]
def _initialize_database(filename):
links_database = shelve.open(filename)
# Wipe the database if this version of the script ends up looking at a
# newer (future) version of the links db, just to be sure.
version = links_database.get('SCHEMA_VERSION')
if version and version != SCHEMA_VERSION:
logging.info('Found database with schema version %s while this script only '
'supports %s. Wiping previous database contents.', version,
SCHEMA_VERSION)
links_database.clear()
links_database['SCHEMA_VERSION'] = SCHEMA_VERSION
return links_database
def main():
parser = optparse.OptionParser()
parser.add_option('-d', '--dry-run', action='store_true', default=False,
help='Print what would be done, but don\'t perform any '
'operations. This will automatically set logging to '
'verbose.')
parser.add_option('-v', '--verbose', action='store_const',
const=logging.DEBUG, default=logging.INFO,
help='Print verbose output for debugging.')
options, _ = parser.parse_args()
if options.dry_run:
options.verbose = logging.DEBUG
logging.basicConfig(format='%(message)s', level=options.verbose)
# Work from the root directory of the checkout.
script_dir = os.path.dirname(os.path.abspath(__file__))
os.chdir(script_dir)
# The database file gets .db appended on some platforms.
db_filenames = [LINKS_DB, LINKS_DB + '.db']
if any(os.path.isfile(f) for f in db_filenames):
links_database = _initialize_database(LINKS_DB)
try:
symlink_creator = WebRTCLinkSetup(links_database, options.dry_run)
symlink_creator.CleanupLinks()
finally:
for f in db_filenames:
if os.path.isfile(f):
os.remove(f)
return 0
if __name__ == '__main__':
sys.exit(main())
libs/libyuv/codereview.settings
+3 -8
View File
@@ -1,11 +1,6 @@
# This file is used by gcl to get repository specific information.
# This file is used by git cl to get repository specific information.
CODE_REVIEW_SERVER: codereview.chromium.org
#CC_LIST:
VIEW_VC: https://chromium.googlesource.com/libyuv/libyuv/+/
#STATUS:
FORCE_HTTPS_COMMIT_URL: True
GERRIT_HOST: True
PROJECT: libyuv
TRY_ON_UPLOAD: False
TRYSERVER_ROOT: src
#GITCL_PREUPLOAD:
#GITCL_PREDCOMMIT:
VIEW_VC: https://chromium.googlesource.com/libyuv/libyuv/+/
libs/libyuv/docs/deprecated_builds.md
+440
View File
@@ -0,0 +1,440 @@
# Deprecated Builds
Older documentation on build configs which are no longer supported.
## Pre-requisites
You'll need to have depot tools installed: https://www.chromium.org/developers/how-tos/install-depot-tools
Refer to chromium instructions for each platform for other prerequisites.
## Getting the Code
Create a working directory, enter it, and run:
gclient config https://chromium.googlesource.com/libyuv/libyuv
gclient sync
Then you'll get a .gclient file like:
solutions = [
{ "name" : "libyuv",
"url" : "https://chromium.googlesource.com/libyuv/libyuv",
"deps_file" : "DEPS",
"managed" : True,
"custom_deps" : {
},
"safesync_url": "",
},
];
For iOS add `;target_os=['ios'];` to your OSX .gclient and run `GYP_DEFINES="OS=ios" gclient sync.`
Browse the Git reprository: https://chromium.googlesource.com/libyuv/libyuv/+/master
### Android
For Android add `;target_os=['android'];` to your Linux .gclient
solutions = [
{ "name" : "libyuv",
"url" : "https://chromium.googlesource.com/libyuv/libyuv",
"deps_file" : "DEPS",
"managed" : True,
"custom_deps" : {
},
"safesync_url": "",
},
];
target_os = ["android", "unix"];
Then run:
export GYP_DEFINES="OS=android"
gclient sync
Caveat: Theres an error with Google Play services updates. If you get the error "Your version of the Google Play services library is not up to date", run the following:
cd chromium/src
./build/android/play_services/update.py download
cd ../..
For Windows the gclient sync must be done from an Administrator command prompt.
The sync will generate native build files for your environment using gyp (Windows: Visual Studio, OSX: XCode, Linux: make). This generation can also be forced manually: `gclient runhooks`
To get just the source (not buildable):
git clone https://chromium.googlesource.com/libyuv/libyuv
## Building the Library and Unittests
### Windows
set GYP_DEFINES=target_arch=ia32
call python gyp_libyuv -fninja -G msvs_version=2013
ninja -j7 -C out\Release
ninja -j7 -C out\Debug
set GYP_DEFINES=target_arch=x64
call python gyp_libyuv -fninja -G msvs_version=2013
ninja -C out\Debug_x64
ninja -C out\Release_x64
#### Building with clangcl
set GYP_DEFINES=clang=1 target_arch=ia32
call python tools\clang\scripts\update.py
call python gyp_libyuv -fninja libyuv_test.gyp
ninja -C out\Debug
ninja -C out\Release
### OSX
Clang 64 bit shown. Remove `clang=1` for GCC and change x64 to ia32 for 32 bit.
GYP_DEFINES="clang=1 target_arch=x64" ./gyp_libyuv
ninja -j7 -C out/Debug
ninja -j7 -C out/Release
GYP_DEFINES="clang=1 target_arch=ia32" ./gyp_libyuv
ninja -j7 -C out/Debug
ninja -j7 -C out/Release
### iOS
http://www.chromium.org/developers/how-tos/build-instructions-ios
Add to .gclient last line: `target_os=['ios'];`
armv7
GYP_DEFINES="OS=ios target_arch=armv7 target_subarch=arm32" GYP_CROSSCOMPILE=1 GYP_GENERATOR_FLAGS="output_dir=out_ios" ./gyp_libyuv
ninja -j7 -C out_ios/Debug-iphoneos libyuv_unittest
ninja -j7 -C out_ios/Release-iphoneos libyuv_unittest
arm64
GYP_DEFINES="OS=ios target_arch=arm64 target_subarch=arm64" GYP_CROSSCOMPILE=1 GYP_GENERATOR_FLAGS="output_dir=out_ios" ./gyp_libyuv
ninja -j7 -C out_ios/Debug-iphoneos libyuv_unittest
ninja -j7 -C out_ios/Release-iphoneos libyuv_unittest
both armv7 and arm64 (fat)
GYP_DEFINES="OS=ios target_arch=armv7 target_subarch=both" GYP_CROSSCOMPILE=1 GYP_GENERATOR_FLAGS="output_dir=out_ios" ./gyp_libyuv
ninja -j7 -C out_ios/Debug-iphoneos libyuv_unittest
ninja -j7 -C out_ios/Release-iphoneos libyuv_unittest
simulator
GYP_DEFINES="OS=ios target_arch=ia32 target_subarch=arm32" GYP_CROSSCOMPILE=1 GYP_GENERATOR_FLAGS="output_dir=out_sim" ./gyp_libyuv
ninja -j7 -C out_sim/Debug-iphonesimulator libyuv_unittest
ninja -j7 -C out_sim/Release-iphonesimulator libyuv_unittest
### Android
https://code.google.com/p/chromium/wiki/AndroidBuildInstructions
Add to .gclient last line: `target_os=['android'];`
armv7
GYP_DEFINES="OS=android" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
ninja -j7 -C out/Release yuv_unittest_apk
arm64
GYP_DEFINES="OS=android target_arch=arm64 target_subarch=arm64" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
ninja -j7 -C out/Release yuv_unittest_apk
ia32
GYP_DEFINES="OS=android target_arch=ia32" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
ninja -j7 -C out/Release yuv_unittest_apk
GYP_DEFINES="OS=android target_arch=ia32 android_full_debug=1" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
mipsel
GYP_DEFINES="OS=android target_arch=mipsel" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
ninja -j7 -C out/Release yuv_unittest_apk
arm32 disassembly:
third_party/android_ndk/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/bin/arm-linux-androideabi-objdump -d out/Release/obj/source/libyuv.row_neon.o
arm64 disassembly:
third_party/android_ndk/toolchains/aarch64-linux-android-4.9/prebuilt/linux-x86_64/bin/aarch64-linux-android-objdump -d out/Release/obj/source/libyuv.row_neon64.o
Running tests:
build/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=*
Running test as benchmark:
build/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=* -a "--libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=-1"
Running test with C code:
build/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=* -a "--libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=1 --libyuv_cpu_info=1"
#### Building with GN
gn gen out/Release "--args=is_debug=false target_cpu=\"x86\""
gn gen out/Debug "--args=is_debug=true target_cpu=\"x86\""
ninja -C out/Release
ninja -C out/Debug
### Building Offical with GN
gn gen out/Official "--args=is_debug=false is_official_build=true is_chrome_branded=true"
ninja -C out/Official
#### Building mips with GN
mipsel
gn gen out/Default "--args=is_debug=false target_cpu=\"mipsel\" target_os = \"android\" mips_arch_variant = \"r6\" mips_use_msa = true is_component_build = true is_clang = false"
ninja -C out/Default
mips64el
gn gen out/Default "--args=is_debug=false target_cpu=\"mips64el\" target_os = \"android\" mips_arch_variant = \"r6\" mips_use_msa = true is_component_build = true is_clang = false"
ninja -C out/Default
### Linux
GYP_DEFINES="target_arch=x64" ./gyp_libyuv
ninja -j7 -C out/Debug
ninja -j7 -C out/Release
GYP_DEFINES="target_arch=ia32" ./gyp_libyuv
ninja -j7 -C out/Debug
ninja -j7 -C out/Release
#### CentOS
On CentOS 32 bit the following work around allows a sync:
export GYP_DEFINES="host_arch=ia32"
gclient sync
### Windows Shared Library
Modify libyuv.gyp from 'static_library' to 'shared_library', and add 'LIBYUV_BUILDING_SHARED_LIBRARY' to 'defines'.
gclient runhooks
After this command follow the building the library instructions above.
If you get a compile error for atlthunk.lib on Windows, read http://www.chromium.org/developers/how-tos/build-instructions-windows
### Build targets
ninja -C out/Debug libyuv
ninja -C out/Debug libyuv_unittest
ninja -C out/Debug compare
ninja -C out/Debug yuvconvert
ninja -C out/Debug psnr
ninja -C out/Debug cpuid
## Building the Library with make
### Linux
make -j7 V=1 -f linux.mk
make -j7 V=1 -f linux.mk clean
make -j7 V=1 -f linux.mk CXX=clang++
## Building the Library with cmake
Install cmake: http://www.cmake.org/
Default debug build:
mkdir out
cd out
cmake ..
cmake --build .
Release build/install
mkdir out
cd out
cmake -DCMAKE_INSTALL_PREFIX="/usr/lib" -DCMAKE_BUILD_TYPE="Release" ..
cmake --build . --config Release
sudo cmake --build . --target install --config Release
### Windows 8 Phone
Pre-requisite:
* Install Visual Studio 2012 and Arm to your environment.<br>
Then:
call "c:\Program Files (x86)\Microsoft Visual Studio 11.0\VC\bin\x86_arm\vcvarsx86_arm.bat"
or with Visual Studio 2013:
call "c:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin\x86_arm\vcvarsx86_arm.bat"
nmake /f winarm.mk clean
nmake /f winarm.mk
### Windows Shared Library
Modify libyuv.gyp from 'static_library' to 'shared_library', and add 'LIBYUV_BUILDING_SHARED_LIBRARY' to 'defines'. Then run this.
gclient runhooks
After this command follow the building the library instructions above.
If you get a compile error for atlthunk.lib on Windows, read http://www.chromium.org/developers/how-tos/build-instructions-windows
### 64 bit Windows
set GYP_DEFINES=target_arch=x64
gclient runhooks V=1
### ARM Linux
export GYP_DEFINES="target_arch=arm"
export CROSSTOOL=`<path>`/arm-none-linux-gnueabi
export CXX=$CROSSTOOL-g++
export CC=$CROSSTOOL-gcc
export AR=$CROSSTOOL-ar
export AS=$CROSSTOOL-as
export RANLIB=$CROSSTOOL-ranlib
gclient runhooks
## Running Unittests
### Windows
out\Release\libyuv_unittest.exe --gtest_catch_exceptions=0 --gtest_filter="*"
### OSX
out/Release/libyuv_unittest --gtest_filter="*"
### Linux
out/Release/libyuv_unittest --gtest_filter="*"
Replace --gtest_filter="*" with specific unittest to run. May include wildcards. e.g.
out/Release/libyuv_unittest --gtest_filter=libyuvTest.I420ToARGB_Opt
## CPU Emulator tools
### Intel SDE (Software Development Emulator)
Pre-requisite: Install IntelSDE for Windows: http://software.intel.com/en-us/articles/intel-software-development-emulator
Then run:
c:\intelsde\sde -hsw -- out\release\libyuv_unittest.exe --gtest_filter=*
## Memory tools
### Running Dr Memory memcheck for Windows
Pre-requisite: Install Dr Memory for Windows and add it to your path: http://www.drmemory.org/docs/page_install_windows.html
set GYP_DEFINES=build_for_tool=drmemory target_arch=ia32
call python gyp_libyuv -fninja -G msvs_version=2013
ninja -C out\Debug
drmemory out\Debug\libyuv_unittest.exe --gtest_catch_exceptions=0 --gtest_filter=*
### Running UBSan
See Chromium instructions for sanitizers: https://www.chromium.org/developers/testing/undefinedbehaviorsanitizer
Sanitizers available: TSan, MSan, ASan, UBSan, LSan
GYP_DEFINES='ubsan=1' gclient runhooks
ninja -C out/Release
### Running Valgrind memcheck
Memory errors and race conditions can be found by running tests under special memory tools. [Valgrind] [1] is an instrumentation framework for building dynamic analysis tools. Various tests and profilers are built upon it to find memory handling errors and memory leaks, for instance.
[1]: http://valgrind.org
solutions = [
{ "name" : "libyuv",
"url" : "https://chromium.googlesource.com/libyuv/libyuv",
"deps_file" : "DEPS",
"managed" : True,
"custom_deps" : {
"libyuv/chromium/src/third_party/valgrind": "https://chromium.googlesource.com/chromium/deps/valgrind/binaries",
},
"safesync_url": "",
},
]
Then run:
GYP_DEFINES="clang=0 target_arch=x64 build_for_tool=memcheck" python gyp_libyuv
ninja -C out/Debug
valgrind out/Debug/libyuv_unittest
For more information, see http://www.chromium.org/developers/how-tos/using-valgrind
### Running Thread Sanitizer (TSan)
GYP_DEFINES="clang=0 target_arch=x64 build_for_tool=tsan" python gyp_libyuv
ninja -C out/Debug
valgrind out/Debug/libyuv_unittest
For more info, see http://www.chromium.org/developers/how-tos/using-valgrind/threadsanitizer
### Running Address Sanitizer (ASan)
GYP_DEFINES="clang=0 target_arch=x64 build_for_tool=asan" python gyp_libyuv
ninja -C out/Debug
valgrind out/Debug/libyuv_unittest
For more info, see http://dev.chromium.org/developers/testing/addresssanitizer
## Benchmarking
The unittests can be used to benchmark.
### Windows
set LIBYUV_WIDTH=1280
set LIBYUV_HEIGHT=720
set LIBYUV_REPEAT=999
set LIBYUV_FLAGS=-1
out\Release\libyuv_unittest.exe --gtest_filter=*I420ToARGB_Opt
### Linux and Mac
LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=1000 out/Release/libyuv_unittest --gtest_filter=*I420ToARGB_Opt
libyuvTest.I420ToARGB_Opt (547 ms)
Indicates 0.547 ms/frame for 1280 x 720.
## Making a change
gclient sync
git checkout -b mycl -t origin/master
git pull
<edit files>
git add -u
git commit -m "my change"
git cl lint
git cl try
git cl upload -r a-reviewer@chomium.org -s
<once approved..>
git cl land
libs/libyuv/docs/environment_variables.md
+18 -2
View File
@@ -6,7 +6,10 @@ For test purposes, environment variables can be set to control libyuv behavior.
By default the cpu is detected and the most advanced form of SIMD is used. But you can disable instruction sets selectively, or completely, falling back on C code. Set the variable to 1 to disable the specified instruction set.
## All CPUs
LIBYUV_DISABLE_ASM
## Intel CPUs
LIBYUV_DISABLE_X86
LIBYUV_DISABLE_SSE2
LIBYUV_DISABLE_SSSE3
@@ -14,12 +17,25 @@ By default the cpu is detected and the most advanced form of SIMD is used. But
LIBYUV_DISABLE_SSE42
LIBYUV_DISABLE_AVX
LIBYUV_DISABLE_AVX2
LIBYUV_DISABLE_AVX3
LIBYUV_DISABLE_ERMS
LIBYUV_DISABLE_FMA3
LIBYUV_DISABLE_DSPR2
LIBYUV_DISABLE_F16C
LIBYUV_DISABLE_AVX512BW
LIBYUV_DISABLE_AVX512VL
LIBYUV_DISABLE_AVX512VBMI
LIBYUV_DISABLE_AVX512VBMI2
LIBYUV_DISABLE_AVX512VBITALG
LIBYUV_DISABLE_AVX512VPOPCNTDQ
LIBYUV_DISABLE_GFNI
## ARM CPUs
LIBYUV_DISABLE_NEON
## MIPS CPUs
LIBYUV_DISABLE_MSA
LIBYUV_DISABLE_MMI
# Test Width/Height/Repeat
The unittests default to a small image (128x72) to run fast. This can be set by environment variable to test a specific resolutions.
libs/libyuv/docs/formats.md
+39 -18
View File
@@ -35,50 +35,46 @@ This is how OSX formats map to libyuv
# FOURCC (Four Charactacter Code) List
The following is extracted from video_common.h as a complete list of formats supported by libyuv.
enum FourCC {
// 9 Primary YUV formats: 5 planar, 2 biplanar, 2 packed.
FOURCC_I420 = FOURCC('I', '4', '2', '0'),
FOURCC_I422 = FOURCC('I', '4', '2', '2'),
FOURCC_I444 = FOURCC('I', '4', '4', '4'),
FOURCC_I411 = FOURCC('I', '4', '1', '1'),
FOURCC_I400 = FOURCC('I', '4', '0', '0'),
FOURCC_NV21 = FOURCC('N', 'V', '2', '1'),
FOURCC_NV12 = FOURCC('N', 'V', '1', '2'),
FOURCC_YUY2 = FOURCC('Y', 'U', 'Y', '2'),
FOURCC_UYVY = FOURCC('U', 'Y', 'V', 'Y'),
FOURCC_H010 = FOURCC('H', '0', '1', '0'), // unofficial fourcc. 10 bit lsb
// 2 Secondary YUV formats: row biplanar.
// 1 Secondary YUV format: row biplanar.
FOURCC_M420 = FOURCC('M', '4', '2', '0'),
FOURCC_Q420 = FOURCC('Q', '4', '2', '0'),
// 9 Primary RGB formats: 4 32 bpp, 2 24 bpp, 3 16 bpp.
// 11 Primary RGB formats: 4 32 bpp, 2 24 bpp, 3 16 bpp, 1 10 bpc
FOURCC_ARGB = FOURCC('A', 'R', 'G', 'B'),
FOURCC_BGRA = FOURCC('B', 'G', 'R', 'A'),
FOURCC_ABGR = FOURCC('A', 'B', 'G', 'R'),
FOURCC_AR30 = FOURCC('A', 'R', '3', '0'), // 10 bit per channel. 2101010.
FOURCC_AB30 = FOURCC('A', 'B', '3', '0'), // ABGR version of 10 bit
FOURCC_24BG = FOURCC('2', '4', 'B', 'G'),
FOURCC_RAW = FOURCC('r', 'a', 'w', ' '),
FOURCC_RAW = FOURCC('r', 'a', 'w', ' '),
FOURCC_RGBA = FOURCC('R', 'G', 'B', 'A'),
FOURCC_RGBP = FOURCC('R', 'G', 'B', 'P'), // rgb565 LE.
FOURCC_RGBO = FOURCC('R', 'G', 'B', 'O'), // argb1555 LE.
FOURCC_R444 = FOURCC('R', '4', '4', '4'), // argb4444 LE.
// 4 Secondary RGB formats: 4 Bayer Patterns.
FOURCC_RGGB = FOURCC('R', 'G', 'G', 'B'),
FOURCC_BGGR = FOURCC('B', 'G', 'G', 'R'),
FOURCC_GRBG = FOURCC('G', 'R', 'B', 'G'),
FOURCC_GBRG = FOURCC('G', 'B', 'R', 'G'),
// 1 Primary Compressed YUV format.
FOURCC_MJPG = FOURCC('M', 'J', 'P', 'G'),
// 5 Auxiliary YUV variations: 3 with U and V planes are swapped, 1 Alias.
// 8 Auxiliary YUV variations: 3 with U and V planes are swapped, 1 Alias.
FOURCC_YV12 = FOURCC('Y', 'V', '1', '2'),
FOURCC_YV16 = FOURCC('Y', 'V', '1', '6'),
FOURCC_YV24 = FOURCC('Y', 'V', '2', '4'),
FOURCC_YU12 = FOURCC('Y', 'U', '1', '2'), // Linux version of I420.
FOURCC_J420 = FOURCC('J', '4', '2', '0'),
FOURCC_J400 = FOURCC('J', '4', '0', '0'),
FOURCC_J400 = FOURCC('J', '4', '0', '0'), // unofficial fourcc
FOURCC_H420 = FOURCC('H', '4', '2', '0'), // unofficial fourcc
FOURCC_H422 = FOURCC('H', '4', '2', '2'), // unofficial fourcc
// 14 Auxiliary aliases. CanonicalFourCC() maps these to canonical fourcc.
FOURCC_IYUV = FOURCC('I', 'Y', 'U', 'V'), // Alias for I420.
@@ -99,12 +95,9 @@ The following is extracted from video_common.h as a complete list of formats sup
FOURCC_L565 = FOURCC('L', '5', '6', '5'), // Alias for RGBP.
FOURCC_5551 = FOURCC('5', '5', '5', '1'), // Alias for RGBO.
// 1 Auxiliary compressed YUV format set aside for capturer.
FOURCC_H264 = FOURCC('H', '2', '6', '4'),
# Planar YUV
The following formats contains a full size Y plane followed by 1 or 2
planes for UV: I420, I422, I444, I411, I400, NV21, NV12, I400
planes for UV: I420, I422, I444, I400, NV21, NV12, I400
The size (subsampling) of the UV varies.
I420, NV12 and NV21 are half width, half height
I422, NV16 and NV61 are half width, full height
@@ -140,3 +133,31 @@ Some are channel order agnostic (e.g. ARGBScale).
Some functions are symmetric (e.g. ARGBToBGRA is the same as BGRAToARGB, so its a macro).
ARGBBlend expects preattenuated ARGB. The R,G,B are premultiplied by alpha. Other functions don't care.
# RGB24 and RAW
There are 2 RGB layouts - RGB24 (aka 24BG) and RAW
RGB24 is B,G,R in memory
RAW is R,G,B in memory
# AR30 and XR30
AR30 is 2 10 10 10 ARGB stored in little endian order.
The 2 bit alpha has 4 values. Here are the comparable 8 bit alpha values.
0 - 0. 00000000b = 0x00 = 0
1 - 33%. 01010101b = 0x55 = 85
2 - 66%. 10101010b = 0xaa = 170
3 - 100%. 11111111b = 0xff = 255
The 10 bit RGB values range from 0 to 1023.
XR30 is the same as AR30 but with no alpha channel.
# NV12 and NV21
NV12 is a biplanar format with a full sized Y plane followed by a single
chroma plane with weaved U and V values.
NV21 is the same but with weaved V and U values.
The 12 in NV12 refers to 12 bits per pixel. NV12 has a half width and half
height chroma channel, and therefore is a 420 subsampling.
NV16 is 16 bits per pixel, with half width and full height. aka 422.
NV24 is 24 bits per pixel with full sized chroma channel. aka 444.
libs/libyuv/docs/getting_started.md
+142 -298
View File
@@ -11,14 +11,13 @@ Refer to chromium instructions for each platform for other prerequisites.
Create a working directory, enter it, and run:
gclient config https://chromium.googlesource.com/libyuv/libyuv
gclient config --name src https://chromium.googlesource.com/libyuv/libyuv
gclient sync
Then you'll get a .gclient file like:
solutions = [
{ "name" : "libyuv",
{ "name" : "src",
"url" : "https://chromium.googlesource.com/libyuv/libyuv",
"deps_file" : "DEPS",
"managed" : True,
@@ -28,17 +27,15 @@ Then you'll get a .gclient file like:
},
];
For iOS add `;target_os=['ios'];` to your OSX .gclient and run `GYP_DEFINES="OS=ios" gclient sync.`
For iOS add `;target_os=['ios'];` to your OSX .gclient and run `gclient sync.`
Browse the Git reprository: https://chromium.googlesource.com/libyuv/libyuv/+/master
### Android
For Android add `;target_os=['android'];` to your Linux .gclient
solutions = [
{ "name" : "libyuv",
{ "name" : "src",
"url" : "https://chromium.googlesource.com/libyuv/libyuv",
"deps_file" : "DEPS",
"managed" : True,
@@ -47,23 +44,12 @@ For Android add `;target_os=['android'];` to your Linux .gclient
"safesync_url": "",
},
];
target_os = ["android", "unix"];
target_os = ["android", "linux"];
Then run:
export GYP_DEFINES="OS=android"
gclient sync
Caveat: Theres an error with Google Play services updates. If you get the error "Your version of the Google Play services library is not up to date", run the following:
cd chromium/src
./build/android/play_services/update.py download
cd ../..
For Windows the gclient sync must be done from an Administrator command prompt.
The sync will generate native build files for your environment using gyp (Windows: Visual Studio, OSX: XCode, Linux: make). This generation can also be forced manually: `gclient runhooks`
To get just the source (not buildable):
git clone https://chromium.googlesource.com/libyuv/libyuv
@@ -73,196 +59,151 @@ To get just the source (not buildable):
### Windows
set GYP_DEFINES=target_arch=ia32
call python gyp_libyuv -fninja -G msvs_version=2013
ninja -j7 -C out\Release
ninja -j7 -C out\Debug
call gn gen out\Release "--args=is_debug=false target_cpu=\"x64\""
call gn gen out\Debug "--args=is_debug=true target_cpu=\"x64\""
ninja -v -C out\Release
ninja -v -C out\Debug
set GYP_DEFINES=target_arch=x64
call python gyp_libyuv -fninja -G msvs_version=2013
ninja -C out\Debug_x64
ninja -C out\Release_x64
call gn gen out\Release "--args=is_debug=false target_cpu=\"x86\""
call gn gen out\Debug "--args=is_debug=true target_cpu=\"x86\""
ninja -v -C out\Release
ninja -v -C out\Debug
#### Building with clangcl
set GYP_DEFINES=clang=1 target_arch=ia32
call python tools\clang\scripts\update.py
call python gyp_libyuv -fninja libyuv_test.gyp
ninja -C out\Debug
ninja -C out\Release
### macOS and Linux
### OSX
Clang 64 bit shown. Remove `clang=1` for GCC and change x64 to ia32 for 32 bit.
GYP_DEFINES="clang=1 target_arch=x64" ./gyp_libyuv
ninja -j7 -C out/Debug
ninja -j7 -C out/Release
GYP_DEFINES="clang=1 target_arch=ia32" ./gyp_libyuv
ninja -j7 -C out/Debug
ninja -j7 -C out/Release
### iOS
http://www.chromium.org/developers/how-tos/build-instructions-ios
Add to .gclient last line: `target_os=['ios'];`
armv7
GYP_DEFINES="OS=ios target_arch=armv7 target_subarch=arm32" GYP_CROSSCOMPILE=1 GYP_GENERATOR_FLAGS="output_dir=out_ios" ./gyp_libyuv
ninja -j7 -C out_ios/Debug-iphoneos libyuv_unittest
ninja -j7 -C out_ios/Release-iphoneos libyuv_unittest
arm64
GYP_DEFINES="OS=ios target_arch=arm64 target_subarch=arm64" GYP_CROSSCOMPILE=1 GYP_GENERATOR_FLAGS="output_dir=out_ios" ./gyp_libyuv
ninja -j7 -C out_ios/Debug-iphoneos libyuv_unittest
ninja -j7 -C out_ios/Release-iphoneos libyuv_unittest
both armv7 and arm64 (fat)
GYP_DEFINES="OS=ios target_arch=armv7 target_subarch=both" GYP_CROSSCOMPILE=1 GYP_GENERATOR_FLAGS="output_dir=out_ios" ./gyp_libyuv
ninja -j7 -C out_ios/Debug-iphoneos libyuv_unittest
ninja -j7 -C out_ios/Release-iphoneos libyuv_unittest
simulator
GYP_DEFINES="OS=ios target_arch=ia32 target_subarch=arm32" GYP_CROSSCOMPILE=1 GYP_GENERATOR_FLAGS="output_dir=out_sim" ./gyp_libyuv
ninja -j7 -C out_sim/Debug-iphonesimulator libyuv_unittest
ninja -j7 -C out_sim/Release-iphonesimulator libyuv_unittest
### Android
https://code.google.com/p/chromium/wiki/AndroidBuildInstructions
Add to .gclient last line: `target_os=['android'];`
armv7
GYP_DEFINES="OS=android" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
ninja -j7 -C out/Release yuv_unittest_apk
arm64
GYP_DEFINES="OS=android target_arch=arm64 target_subarch=arm64" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
ninja -j7 -C out/Release yuv_unittest_apk
ia32
GYP_DEFINES="OS=android target_arch=ia32" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
ninja -j7 -C out/Release yuv_unittest_apk
GYP_DEFINES="OS=android target_arch=ia32 android_full_debug=1" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
mipsel
GYP_DEFINES="OS=android target_arch=mipsel" GYP_CROSSCOMPILE=1 ./gyp_libyuv
ninja -j7 -C out/Debug yuv_unittest_apk
ninja -j7 -C out/Release yuv_unittest_apk
arm32 disassembly:
third_party/android_tools/ndk/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/bin/arm-linux-androideabi-objdump -d out/Release/obj/source/libyuv.row_neon.o
arm64 disassembly:
third_party/android_tools/ndk/toolchains/aarch64-linux-android-4.9/prebuilt/linux-x86_64/bin/aarch64-linux-android-objdump -d out/Release/obj/source/libyuv.row_neon64.o
Running tests:
util/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=*
Running test as benchmark:
util/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=* -a "--libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=-1"
Running test with C code:
util/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release --gtest_filter=* -a "--libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=1 --libyuv_cpu_info=1"
#### Building with GN
gn gen out/Release "--args=is_debug=false target_cpu=\"x86\""
gn gen out/Debug "--args=is_debug=true target_cpu=\"x86\""
ninja -C out/Release
ninja -C out/Debug
gn gen out/Release "--args=is_debug=false"
gn gen out/Debug "--args=is_debug=true"
ninja -v -C out/Release
ninja -v -C out/Debug
### Building Offical with GN
gn gen out/Official "--args=is_debug=false is_official_build=true is_chrome_branded=true"
ninja -C out/Official
#### Building mips with GN
### iOS
http://www.chromium.org/developers/how-tos/build-instructions-ios
mipsel
gn gen out/Default "--args=is_debug=false target_cpu=\"mipsel\" target_os = \"android\" mips_arch_variant = \"r6\" mips_use_msa = true is_component_build = true is_clang = false"
ninja -C out/Default
Add to .gclient last line: `target_os=['ios'];`
mips64el
gn gen out/Default "--args=is_debug=false target_cpu=\"mips64el\" target_os = \"android\" mips_arch_variant = \"r6\" mips_use_msa = true is_component_build = true is_clang = false"
ninja -C out/Default
arm64
### Linux
gn gen out/Release "--args=is_debug=false target_os=\"ios\" ios_enable_code_signing=false target_cpu=\"arm64\""
gn gen out/Debug "--args=is_debug=true target_os=\"ios\" ios_enable_code_signing=false target_cpu=\"arm64\""
ninja -v -C out/Debug libyuv_unittest
ninja -v -C out/Release libyuv_unittest
GYP_DEFINES="target_arch=x64" ./gyp_libyuv
ninja -j7 -C out/Debug
ninja -j7 -C out/Release
ios simulator
GYP_DEFINES="target_arch=ia32" ./gyp_libyuv
ninja -j7 -C out/Debug
ninja -j7 -C out/Release
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
#### CentOS
ios disassembly
On CentOS 32 bit the following work around allows a sync:
otool -tV ./out/Release/obj/libyuv_neon/row_neon64.o >row_neon64.txt
export GYP_DEFINES="host_arch=ia32"
gclient sync
### Android
https://code.google.com/p/chromium/wiki/AndroidBuildInstructions
### Windows Shared Library
Add to .gclient last line: `target_os=['android'];`
Modify libyuv.gyp from 'static_library' to 'shared_library', and add 'LIBYUV_BUILDING_SHARED_LIBRARY' to 'defines'.
arm64
gclient runhooks
gn gen out/Release "--args=is_debug=false target_os=\"android\" target_cpu=\"arm64\""
gn gen out/Debug "--args=is_debug=true target_os=\"android\" target_cpu=\"arm64\""
ninja -v -C out/Debug libyuv_unittest
ninja -v -C out/Release libyuv_unittest
After this command follow the building the library instructions above.
armv7
If you get a compile error for atlthunk.lib on Windows, read http://www.chromium.org/developers/how-tos/build-instructions-windows
gn gen out/Release "--args=is_debug=false target_os=\"android\" target_cpu=\"arm\""
gn gen out/Debug "--args=is_debug=true target_os=\"android\" target_cpu=\"arm\""
ninja -v -C out/Debug libyuv_unittest
ninja -v -C out/Release libyuv_unittest
ia32
gn gen out/Release "--args=is_debug=false target_os=\"android\" target_cpu=\"x86\""
gn gen out/Debug "--args=is_debug=true target_os=\"android\" target_cpu=\"x86\""
ninja -v -C out/Debug libyuv_unittest
ninja -v -C out/Release libyuv_unittest
mips
gn gen out/Release "--args=is_debug=false target_os=\"android\" target_cpu=\"mips64el\" mips_arch_variant=\"r6\" mips_use_msa=true is_component_build=true"
gn gen out/Debug "--args=is_debug=true target_os=\"android\" target_cpu=\"mips64el\" mips_arch_variant=\"r6\" mips_use_msa=true is_component_build=true"
ninja -v -C out/Debug libyuv_unittest
ninja -v -C out/Release libyuv_unittest
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/row_common.o >row_common.txt
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_neon.o >row_neon.txt
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:
out/Release/bin/run_libyuv_unittest -vv --gtest_filter=*
Running test as benchmark:
out/Release/bin/run_libyuv_unittest -vv --gtest_filter=* --libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=-1 --libyuv_cpu_info=-1
Running test with C code:
out/Release/bin/run_libyuv_unittest -vv --gtest_filter=* --libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=999 --libyuv_flags=1 --libyuv_cpu_info=1
### Build targets
ninja -C out/Debug libyuv
ninja -C out/Debug libyuv_unittest
ninja -C out/Debug compare
ninja -C out/Debug convert
ninja -C out/Debug yuvconvert
ninja -C out/Debug psnr
ninja -C out/Debug cpuid
### ARM Linux
gn gen out/Release "--args=is_debug=false target_cpu=\"arm64\""
gn gen out/Debug "--args=is_debug=true target_cpu=\"arm64\""
ninja -v -C out/Debug libyuv_unittest
ninja -v -C out/Release libyuv_unittest
### MIPS Linux
mips
gn gen out/Release "--args=is_debug=false target_os=\"linux\" target_cpu=\"mips64el\" mips_arch_variant=\"loongson3\" mips_use_mmi=true is_component_build=false use_sysroot=false use_gold=false"
gn gen out/Debug "--args=is_debug=true target_os=\"linux\" target_cpu=\"mips64el\" mips_arch_variant=\"loongson3\" mips_use_mmi=true is_component_build=false use_sysroot=false use_gold=false"
ninja -v -C out/Debug libyuv_unittest
ninja -v -C out/Release libyuv_unittest
## Building the Library with make
### Linux
make -j7 V=1 -f linux.mk
make -j7 V=1 -f linux.mk clean
make -j7 V=1 -f linux.mk CXX=clang++
make V=1 -f linux.mk
make V=1 -f linux.mk clean
make V=1 -f linux.mk CXX=clang++
## Building the Library with cmake
## Building the library with cmake
Install cmake: http://www.cmake.org/
Default debug build:
### Default debug build:
mkdir out
cd out
cmake ..
cmake --build .
Release build/install
### Release build/install
mkdir out
cd out
@@ -270,47 +211,31 @@ Release build/install
cmake --build . --config Release
sudo cmake --build . --target install --config Release
### Windows 8 Phone
### Build RPM/DEB packages
Pre-requisite:
mkdir out
cd out
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j4
make package
* Install Visual Studio 2012 and Arm to your environment.<br>
## Setup for Arm Cross compile
Then:
See also https://www.ccoderun.ca/programming/2015-12-20_CrossCompiling/index.html
call "c:\Program Files (x86)\Microsoft Visual Studio 11.0\VC\bin\x86_arm\vcvarsx86_arm.bat"
sudo apt-get install ssh dkms build-essential linux-headers-generic
sudo apt-get install kdevelop cmake git subversion
sudo apt-get install graphviz doxygen doxygen-gui
sudo apt-get install manpages manpages-dev manpages-posix manpages-posix-dev
sudo apt-get install libboost-all-dev libboost-dev libssl-dev
sudo apt-get install rpm terminator fish
sudo apt-get install g++-arm-linux-gnueabihf gcc-arm-linux-gnueabihf
or with Visual Studio 2013:
### Build psnr tool
call "c:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin\x86_arm\vcvarsx86_arm.bat"
nmake /f winarm.mk clean
nmake /f winarm.mk
### Windows Shared Library
Modify libyuv.gyp from 'static_library' to 'shared_library', and add 'LIBYUV_BUILDING_SHARED_LIBRARY' to 'defines'. Then run this.
gclient runhooks
After this command follow the building the library instructions above.
If you get a compile error for atlthunk.lib on Windows, read http://www.chromium.org/developers/how-tos/build-instructions-windows
### 64 bit Windows
set GYP_DEFINES=target_arch=x64
gclient runhooks V=1
### ARM Linux
export GYP_DEFINES="target_arch=arm"
export CROSSTOOL=`<path>`/arm-none-linux-gnueabi
export CXX=$CROSSTOOL-g++
export CC=$CROSSTOOL-gcc
export AR=$CROSSTOOL-ar
export AS=$CROSSTOOL-as
export RANLIB=$CROSSTOOL-ranlib
gclient runhooks
cd util
arm-linux-gnueabihf-g++ psnr_main.cc psnr.cc ssim.cc -o psnr
arm-linux-gnueabihf-objdump -d psnr
## Running Unittests
@@ -318,123 +243,42 @@ If you get a compile error for atlthunk.lib on Windows, read http://www.chromium
out\Release\libyuv_unittest.exe --gtest_catch_exceptions=0 --gtest_filter="*"
### OSX
### macOS and Linux
out/Release/libyuv_unittest --gtest_filter="*"
### Linux
out/Release/libyuv_unittest --gtest_filter="*"
Replace --gtest_filter="*" with specific unittest to run. May include wildcards. e.g.
out/Release/libyuv_unittest --gtest_filter=libyuvTest.I420ToARGB_Opt
Replace --gtest_filter="*" with specific unittest to run. May include wildcards.
out/Release/libyuv_unittest --gtest_filter=*I420ToARGB_Opt
## CPU Emulator tools
### Intel SDE (Software Development Emulator)
Pre-requisite: Install IntelSDE for Windows: http://software.intel.com/en-us/articles/intel-software-development-emulator
Pre-requisite: Install IntelSDE: http://software.intel.com/en-us/articles/intel-software-development-emulator
Then run:
c:\intelsde\sde -hsw -- out\release\libyuv_unittest.exe --gtest_filter=*
c:\intelsde\sde -hsw -- out\Release\libyuv_unittest.exe --gtest_filter=*
~/intelsde/sde -skx -- out/Release/libyuv_unittest --gtest_filter=**I420ToARGB_Opt
## Memory tools
### Intel Architecture Code Analyzer
Inset these 2 macros into assembly code to be analyzed:
IACA_ASM_START
IACA_ASM_END
Build the code as usual, then run iaca on the object file.
~/iaca-lin64/bin/iaca.sh -reduceout -arch HSW out/Release/obj/libyuv_internal/compare_gcc.o
## Sanitizers
gn gen out/Release "--args=is_debug=false is_msan=true"
ninja -v -C out/Release
Sanitizers available: asan, msan, tsan, ubsan, lsan, ubsan_vptr
### Running Dr Memory memcheck for Windows
Pre-requisite: Install Dr Memory for Windows and add it to your path: http://www.drmemory.org/docs/page_install_windows.html
set GYP_DEFINES=build_for_tool=drmemory target_arch=ia32
call python gyp_libyuv -fninja -G msvs_version=2013
ninja -C out\Debug
drmemory out\Debug\libyuv_unittest.exe --gtest_catch_exceptions=0 --gtest_filter=*
### Running UBSan
See Chromium instructions for sanitizers: https://www.chromium.org/developers/testing/undefinedbehaviorsanitizer
Sanitizers available: TSan, MSan, ASan, UBSan, LSan
GYP_DEFINES='ubsan=1' gclient runhooks
ninja -C out/Release
### Running Valgrind memcheck
Memory errors and race conditions can be found by running tests under special memory tools. [Valgrind] [1] is an instrumentation framework for building dynamic analysis tools. Various tests and profilers are built upon it to find memory handling errors and memory leaks, for instance.
[1]: http://valgrind.org
solutions = [
{ "name" : "libyuv",
"url" : "https://chromium.googlesource.com/libyuv/libyuv",
"deps_file" : "DEPS",
"managed" : True,
"custom_deps" : {
"libyuv/chromium/src/third_party/valgrind": "https://chromium.googlesource.com/chromium/deps/valgrind/binaries",
},
"safesync_url": "",
},
]
Then run:
GYP_DEFINES="clang=0 target_arch=x64 build_for_tool=memcheck" python gyp_libyuv
ninja -C out/Debug
valgrind out/Debug/libyuv_unittest
For more information, see http://www.chromium.org/developers/how-tos/using-valgrind
### Running Thread Sanitizer (TSan)
GYP_DEFINES="clang=0 target_arch=x64 build_for_tool=tsan" python gyp_libyuv
ninja -C out/Debug
valgrind out/Debug/libyuv_unittest
For more info, see http://www.chromium.org/developers/how-tos/using-valgrind/threadsanitizer
### Running Address Sanitizer (ASan)
GYP_DEFINES="clang=0 target_arch=x64 build_for_tool=asan" python gyp_libyuv
ninja -C out/Debug
valgrind out/Debug/libyuv_unittest
For more info, see http://dev.chromium.org/developers/testing/addresssanitizer
## Benchmarking
The unittests can be used to benchmark.
### Windows
set LIBYUV_WIDTH=1280
set LIBYUV_HEIGHT=720
set LIBYUV_REPEAT=999
set LIBYUV_FLAGS=-1
out\Release\libyuv_unittest.exe --gtest_filter=*I420ToARGB_Opt
### Linux and Mac
LIBYUV_WIDTH=1280 LIBYUV_HEIGHT=720 LIBYUV_REPEAT=1000 out/Release/libyuv_unittest --gtest_filter=*I420ToARGB_Opt
libyuvTest.I420ToARGB_Opt (547 ms)
Indicates 0.547 ms/frame for 1280 x 720.
## Making a change
gclient sync
git checkout -b mycl -t origin/master
git pull
<edit files>
git add -u
git commit -m "my change"
git cl lint
git cl try
git cl upload -r a-reviewer@chomium.org -s
<once approved..>
git cl land
libs/libyuv/gyp_libyuv
-101
View File
@@ -1,101 +0,0 @@
#!/usr/bin/env python
#
# Copyright 2014 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
# This script is used to run GYP for libyuv. It contains selected parts of the
# main function from the src/build/gyp_chromium file.
import glob
import os
import shlex
import sys
checkout_root = os.path.dirname(os.path.realpath(__file__))
sys.path.insert(0, os.path.join(checkout_root, 'build'))
import gyp_chromium
import gyp_helper
import vs_toolchain
sys.path.insert(0, os.path.join(checkout_root, 'tools', 'gyp', 'pylib'))
import gyp
def GetSupplementalFiles():
"""Returns a list of the supplemental files that are included in all GYP
sources."""
# Can't use the one in gyp_chromium since the directory location of the root
# is different.
return glob.glob(os.path.join(checkout_root, '*', 'supplement.gypi'))
if __name__ == '__main__':
args = sys.argv[1:]
if int(os.environ.get('GYP_CHROMIUM_NO_ACTION', 0)):
print 'Skipping gyp_libyuv due to GYP_CHROMIUM_NO_ACTION env var.'
sys.exit(0)
# This could give false positives since it doesn't actually do real option
# parsing. Oh well.
gyp_file_specified = False
for arg in args:
if arg.endswith('.gyp'):
gyp_file_specified = True
break
# If we didn't get a file, assume 'all.gyp' in the root of the checkout.
if not gyp_file_specified:
# Because of a bug in gyp, simply adding the abspath to all.gyp doesn't
# work, but chdir'ing and adding the relative path does. Spooky :/
os.chdir(checkout_root)
args.append('all.gyp')
# There shouldn't be a circular dependency relationship between .gyp files,
args.append('--no-circular-check')
# Default to ninja unless GYP_GENERATORS is set.
if not os.environ.get('GYP_GENERATORS'):
os.environ['GYP_GENERATORS'] = 'ninja'
vs2013_runtime_dll_dirs = None
if int(os.environ.get('DEPOT_TOOLS_WIN_TOOLCHAIN', '1')):
vs2013_runtime_dll_dirs = vs_toolchain.SetEnvironmentAndGetRuntimeDllDirs()
# Enforce gyp syntax checking. This adds about 20% execution time.
args.append('--check')
supplemental_includes = gyp_chromium.GetSupplementalFiles()
gyp_vars_dict = gyp_chromium.GetGypVars(supplemental_includes)
# Automatically turn on crosscompile support for platforms that need it.
if all(('ninja' in os.environ.get('GYP_GENERATORS', ''),
gyp_vars_dict.get('OS') in ['android', 'ios'],
'GYP_CROSSCOMPILE' not in os.environ)):
os.environ['GYP_CROSSCOMPILE'] = '1'
args.extend(['-I' + i for i in
gyp_chromium.additional_include_files(supplemental_includes,
args)])
# Set the gyp depth variable to the root of the checkout.
args.append('--depth=' + os.path.relpath(checkout_root))
print 'Updating projects from gyp files...'
sys.stdout.flush()
# Off we go...
gyp_rc = gyp.main(args)
if vs2013_runtime_dll_dirs:
x64_runtime, x86_runtime = vs2013_runtime_dll_dirs
vs_toolchain.CopyVsRuntimeDlls(
os.path.join(checkout_root, gyp_chromium.GetOutputDirectory()),
(x86_runtime, x64_runtime))
sys.exit(gyp_rc)
libs/libyuv/gyp_libyuv.py
-28
View File
@@ -1,28 +0,0 @@
#!/usr/bin/env python
#
# Copyright 2014 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
# This script is a modified copy of the src/build/gyp_chromium.py file.
# It is needed for parallel processing.
# This file is (possibly, depending on python version) imported by
# gyp_libyuv when GYP_PARALLEL=1 and it creates sub-processes
# through the multiprocessing library.
# Importing in Python 2.6 (fixed in 2.7) on Windows doesn't search for
# imports that don't end in .py (and aren't directories with an
# __init__.py). This wrapper makes "import gyp_libyuv" work with
# those old versions and makes it possible to execute gyp_libyuv.py
# directly on Windows where the extension is useful.
import os
path = os.path.abspath(os.path.split(__file__)[0])
execfile(os.path.join(path, 'gyp_libyuv'))
libs/libyuv/include/libyuv/basic_types.h
+29 -79
View File
@@ -11,79 +11,36 @@
#ifndef INCLUDE_LIBYUV_BASIC_TYPES_H_
#define INCLUDE_LIBYUV_BASIC_TYPES_H_
#include <stddef.h> // for NULL, size_t
#include <stddef.h> // For size_t and NULL
#if !defined(INT_TYPES_DEFINED) && !defined(GG_LONGLONG)
#define INT_TYPES_DEFINED
#if defined(_MSC_VER) && (_MSC_VER < 1600)
#include <sys/types.h> // for uintptr_t on x86
typedef unsigned __int64 uint64_t;
typedef __int64 int64_t;
typedef unsigned int uint32_t;
typedef int int32_t;
typedef unsigned short uint16_t;
typedef short int16_t;
typedef unsigned char uint8_t;
typedef signed char int8_t;
#else
#include <stdint.h> // for uintptr_t
#endif
#ifndef GG_LONGLONG
#ifndef INT_TYPES_DEFINED
#define INT_TYPES_DEFINED
#ifdef COMPILER_MSVC
typedef unsigned __int64 uint64;
typedef __int64 int64;
#ifndef INT64_C
#define INT64_C(x) x ## I64
#endif
#ifndef UINT64_C
#define UINT64_C(x) x ## UI64
#endif
#define INT64_F "I64"
#else // COMPILER_MSVC
#if defined(__LP64__) && !defined(__OpenBSD__) && !defined(__APPLE__)
typedef unsigned long uint64; // NOLINT
typedef long int64; // NOLINT
#ifndef INT64_C
#define INT64_C(x) x ## L
#endif
#ifndef UINT64_C
#define UINT64_C(x) x ## UL
#endif
#define INT64_F "l"
#else // defined(__LP64__) && !defined(__OpenBSD__) && !defined(__APPLE__)
typedef unsigned long long uint64; // NOLINT
typedef long long int64; // NOLINT
#ifndef INT64_C
#define INT64_C(x) x ## LL
#endif
#ifndef UINT64_C
#define UINT64_C(x) x ## ULL
#endif
#define INT64_F "ll"
#endif // __LP64__
#endif // COMPILER_MSVC
typedef unsigned int uint32;
typedef int int32;
typedef unsigned short uint16; // NOLINT
typedef short int16; // NOLINT
typedef unsigned char uint8;
typedef signed char int8;
#include <stdint.h> // for uintptr_t and C99 types
#endif // defined(_MSC_VER) && (_MSC_VER < 1600)
// Types are deprecated. Enable this macro for legacy types.
#ifdef LIBYUV_LEGACY_TYPES
typedef uint64_t uint64;
typedef int64_t int64;
typedef uint32_t uint32;
typedef int32_t int32;
typedef uint16_t uint16;
typedef int16_t int16;
typedef uint8_t uint8;
typedef int8_t int8;
#endif // LIBYUV_LEGACY_TYPES
#endif // INT_TYPES_DEFINED
#endif // GG_LONGLONG
// Detect compiler is for x86 or x64.
#if defined(__x86_64__) || defined(_M_X64) || \
defined(__i386__) || defined(_M_IX86)
#define CPU_X86 1
#endif
// Detect compiler is for ARM.
#if defined(__arm__) || defined(_M_ARM)
#define CPU_ARM 1
#endif
#ifndef ALIGNP
#ifdef __cplusplus
#define ALIGNP(p, t) \
(reinterpret_cast<uint8*>(((reinterpret_cast<uintptr_t>(p) + \
((t) - 1)) & ~((t) - 1))))
#else
#define ALIGNP(p, t) \
((uint8*)((((uintptr_t)(p) + ((t) - 1)) & ~((t) - 1)))) /* NOLINT */
#endif
#endif
#if !defined(LIBYUV_API)
#if defined(_WIN32) || defined(__CYGWIN__)
@@ -95,24 +52,17 @@ typedef signed char int8;
#define LIBYUV_API
#endif // LIBYUV_BUILDING_SHARED_LIBRARY
#elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__APPLE__) && \
(defined(LIBYUV_BUILDING_SHARED_LIBRARY) || \
defined(LIBYUV_USING_SHARED_LIBRARY))
#define LIBYUV_API __attribute__ ((visibility ("default")))
(defined(LIBYUV_BUILDING_SHARED_LIBRARY) || \
defined(LIBYUV_USING_SHARED_LIBRARY))
#define LIBYUV_API __attribute__((visibility("default")))
#else
#define LIBYUV_API
#endif // __GNUC__
#endif // LIBYUV_API
// TODO(fbarchard): Remove bool macros.
#define LIBYUV_BOOL int
#define LIBYUV_FALSE 0
#define LIBYUV_TRUE 1
// Visual C x86 or GCC little endian.
#if defined(__x86_64__) || defined(_M_X64) || \
defined(__i386__) || defined(_M_IX86) || \
defined(__arm__) || defined(_M_ARM) || \
(defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
#define LIBYUV_LITTLE_ENDIAN
#endif
#endif // INCLUDE_LIBYUV_BASIC_TYPES_H_
libs/libyuv/include/libyuv/compare.h
+61 -28
View File
@@ -20,55 +20,88 @@ extern "C" {
// Compute a hash for specified memory. Seed of 5381 recommended.
LIBYUV_API
uint32 HashDjb2(const uint8* src, uint64 count, uint32 seed);
uint32_t HashDjb2(const uint8_t* src, uint64_t count, uint32_t seed);
// Hamming Distance
LIBYUV_API
uint64_t ComputeHammingDistance(const uint8_t* src_a,
const uint8_t* src_b,
int count);
// Scan an opaque argb image and return fourcc based on alpha offset.
// Returns FOURCC_ARGB, FOURCC_BGRA, or 0 if unknown.
LIBYUV_API
uint32 ARGBDetect(const uint8* argb, int stride_argb, int width, int height);
uint32_t ARGBDetect(const uint8_t* argb,
int stride_argb,
int width,
int height);
// Sum Square Error - used to compute Mean Square Error or PSNR.
LIBYUV_API
uint64 ComputeSumSquareError(const uint8* src_a,
const uint8* src_b, int count);
uint64_t ComputeSumSquareError(const uint8_t* src_a,
const uint8_t* src_b,
int count);
LIBYUV_API
uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height);
uint64_t ComputeSumSquareErrorPlane(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height);
static const int kMaxPsnr = 128;
LIBYUV_API
double SumSquareErrorToPsnr(uint64 sse, uint64 count);
double SumSquareErrorToPsnr(uint64_t sse, uint64_t count);
LIBYUV_API
double CalcFramePsnr(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height);
double CalcFramePsnr(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height);
LIBYUV_API
double I420Psnr(const uint8* src_y_a, int stride_y_a,
const uint8* src_u_a, int stride_u_a,
const uint8* src_v_a, int stride_v_a,
const uint8* src_y_b, int stride_y_b,
const uint8* src_u_b, int stride_u_b,
const uint8* src_v_b, int stride_v_b,
int width, int height);
double I420Psnr(const uint8_t* src_y_a,
int stride_y_a,
const uint8_t* src_u_a,
int stride_u_a,
const uint8_t* src_v_a,
int stride_v_a,
const uint8_t* src_y_b,
int stride_y_b,
const uint8_t* src_u_b,
int stride_u_b,
const uint8_t* src_v_b,
int stride_v_b,
int width,
int height);
LIBYUV_API
double CalcFrameSsim(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height);
double CalcFrameSsim(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height);
LIBYUV_API
double I420Ssim(const uint8* src_y_a, int stride_y_a,
const uint8* src_u_a, int stride_u_a,
const uint8* src_v_a, int stride_v_a,
const uint8* src_y_b, int stride_y_b,
const uint8* src_u_b, int stride_u_b,
const uint8* src_v_b, int stride_v_b,
int width, int height);
double I420Ssim(const uint8_t* src_y_a,
int stride_y_a,
const uint8_t* src_u_a,
int stride_u_a,
const uint8_t* src_v_a,
int stride_v_a,
const uint8_t* src_y_b,
int stride_y_b,
const uint8_t* src_u_b,
int stride_u_b,
const uint8_t* src_v_b,
int stride_v_b,
int width,
int height);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/compare_row.h
+76 -19
View File
@@ -18,20 +18,23 @@ namespace libyuv {
extern "C" {
#endif
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__i386__) && !defined(__SSE2__))
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__native_client__) && defined(__x86_64__)) || \
(defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
#define LIBYUV_DISABLE_X86
#endif
#if defined(__native_client__)
#define LIBYUV_DISABLE_NEON
#endif
// MemorySanitizer does not support assembly code yet. http://crbug.com/344505
#if defined(__has_feature)
#if __has_feature(memory_sanitizer)
#define LIBYUV_DISABLE_X86
#endif
#endif
// Visual C 2012 required for AVX2.
#if defined(_M_IX86) && !defined(__clang__) && \
defined(_MSC_VER) && _MSC_VER >= 1700
#if defined(_M_IX86) && !defined(__clang__) && defined(_MSC_VER) && \
_MSC_VER >= 1700
#define VISUALC_HAS_AVX2 1
#endif // VisualStudio >= 2012
@@ -42,39 +45,93 @@ extern "C" {
#endif // clang >= 3.4
#endif // __clang__
#if !defined(LIBYUV_DISABLE_X86) && \
defined(_M_IX86) && (defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2))
#define HAS_HASHDJB2_AVX2
#endif
// The following are available for Visual C and GCC:
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || (defined(__i386__) || defined(_M_IX86)))
(defined(__x86_64__) || defined(__i386__) || defined(_M_IX86))
#define HAS_HASHDJB2_SSE41
#define HAS_SUMSQUAREERROR_SSE2
#define HAS_HAMMINGDISTANCE_SSE42
#endif
// The following are available for Visual C and clangcl 32 bit:
#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
#define HAS_SUMSQUAREERROR_AVX2
#endif
// The following are available for GCC and clangcl 64 bit:
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
#define HAS_HAMMINGDISTANCE_SSSE3
#endif
// The following are available for GCC and clangcl 64 bit:
#if !defined(LIBYUV_DISABLE_X86) && defined(CLANG_HAS_AVX2) && \
(defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
#define HAS_HAMMINGDISTANCE_AVX2
#endif
// The following are available for Neon:
#if !defined(LIBYUV_DISABLE_NEON) && \
(defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))
#define HAS_SUMSQUAREERROR_NEON
#define HAS_HAMMINGDISTANCE_NEON
#endif
uint32 SumSquareError_C(const uint8* src_a, const uint8* src_b, int count);
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count);
uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count);
uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count);
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#define HAS_HAMMINGDISTANCE_MSA
#define HAS_SUMSQUAREERROR_MSA
#endif
uint32 HashDjb2_C(const uint8* src, int count, uint32 seed);
uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed);
uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed);
#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
#define HAS_HAMMINGDISTANCE_MMI
#define HAS_SUMSQUAREERROR_MMI
#endif
uint32_t HammingDistance_C(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t HammingDistance_SSSE3(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t HammingDistance_AVX2(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t HammingDistance_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t HammingDistance_MSA(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t HammingDistance_MMI(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t SumSquareError_C(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t SumSquareError_SSE2(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t SumSquareError_AVX2(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t SumSquareError_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t SumSquareError_MSA(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t SumSquareError_MMI(const uint8_t* src_a,
const uint8_t* src_b,
int count);
uint32_t HashDjb2_C(const uint8_t* src, int count, uint32_t seed);
uint32_t HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed);
uint32_t HashDjb2_AVX2(const uint8_t* src, int count, uint32_t seed);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/convert.h
+336 -133
View File
@@ -16,8 +16,8 @@
#include "libyuv/rotate.h" // For enum RotationMode.
// TODO(fbarchard): fix WebRTC source to include following libyuv headers:
#include "libyuv/convert_argb.h" // For WebRTC I420ToARGB. b/620
#include "libyuv/convert_from.h" // For WebRTC ConvertFromI420. b/620
#include "libyuv/convert_argb.h" // For WebRTC I420ToARGB. b/620
#include "libyuv/convert_from.h" // For WebRTC ConvertFromI420. b/620
#include "libyuv/planar_functions.h" // For WebRTC I420Rect, CopyPlane. b/618
#ifdef __cplusplus
@@ -27,195 +27,391 @@ extern "C" {
// Convert I444 to I420.
LIBYUV_API
int I444ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I444ToI420(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
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 width,
int height);
// Convert I444 to NV21.
LIBYUV_API
int I444ToNV21(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int width,
int height);
// Convert I422 to I420.
LIBYUV_API
int I422ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I422ToI420(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
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 width,
int height);
// Convert I411 to I420.
// Convert I422 to NV21.
LIBYUV_API
int I411ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I422ToNV21(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int width,
int height);
// Copy I420 to I420.
#define I420ToI420 I420Copy
LIBYUV_API
int I420Copy(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I420Copy(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
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 width,
int height);
// Copy I010 to I010
#define I010ToI010 I010Copy
#define H010ToH010 I010Copy
LIBYUV_API
int I010Copy(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint16_t* dst_y,
int dst_stride_y,
uint16_t* dst_u,
int dst_stride_u,
uint16_t* dst_v,
int dst_stride_v,
int width,
int height);
// Convert 10 bit YUV to 8 bit
#define H010ToH420 I010ToI420
LIBYUV_API
int I010ToI420(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
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 width,
int height);
// Convert I400 (grey) to I420.
LIBYUV_API
int I400ToI420(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I400ToI420(const uint8_t* src_y,
int src_stride_y,
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 width,
int height);
// Convert I400 (grey) to NV21.
LIBYUV_API
int I400ToNV21(const uint8_t* src_y,
int src_stride_y,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int width,
int height);
#define J400ToJ420 I400ToI420
// Convert NV12 to I420.
LIBYUV_API
int NV12ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int NV12ToI420(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_uv,
int src_stride_uv,
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 width,
int height);
// Convert NV21 to I420.
LIBYUV_API
int NV21ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_vu, int src_stride_vu,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int NV21ToI420(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_vu,
int src_stride_vu,
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 width,
int height);
// Convert YUY2 to I420.
LIBYUV_API
int YUY2ToI420(const uint8* src_yuy2, int src_stride_yuy2,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int YUY2ToI420(const uint8_t* src_yuy2,
int src_stride_yuy2,
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 width,
int height);
// Convert UYVY to I420.
LIBYUV_API
int UYVYToI420(const uint8* src_uyvy, int src_stride_uyvy,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int UYVYToI420(const uint8_t* src_uyvy,
int src_stride_uyvy,
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 width,
int height);
// Convert M420 to I420.
LIBYUV_API
int M420ToI420(const uint8* src_m420, int src_stride_m420,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int M420ToI420(const uint8_t* src_m420,
int src_stride_m420,
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 width,
int height);
// Convert Android420 to I420.
LIBYUV_API
int Android420ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
int pixel_stride_uv,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int Android420ToI420(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
int src_pixel_stride_uv,
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 width,
int height);
// ARGB little endian (bgra in memory) to I420.
LIBYUV_API
int ARGBToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToI420(const uint8_t* src_argb,
int src_stride_argb,
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 width,
int height);
// BGRA little endian (argb in memory) to I420.
LIBYUV_API
int BGRAToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int BGRAToI420(const uint8_t* src_bgra,
int src_stride_bgra,
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 width,
int height);
// ABGR little endian (rgba in memory) to I420.
LIBYUV_API
int ABGRToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ABGRToI420(const uint8_t* src_abgr,
int src_stride_abgr,
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 width,
int height);
// RGBA little endian (abgr in memory) to I420.
LIBYUV_API
int RGBAToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int RGBAToI420(const uint8_t* src_rgba,
int src_stride_rgba,
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 width,
int height);
// RGB little endian (bgr in memory) to I420.
LIBYUV_API
int RGB24ToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int RGB24ToI420(const uint8_t* src_rgb24,
int src_stride_rgb24,
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 width,
int height);
// RGB big endian (rgb in memory) to I420.
LIBYUV_API
int RAWToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int RAWToI420(const uint8_t* src_raw,
int src_stride_raw,
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 width,
int height);
// RGB16 (RGBP fourcc) little endian to I420.
LIBYUV_API
int RGB565ToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int RGB565ToI420(const uint8_t* src_rgb565,
int src_stride_rgb565,
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 width,
int height);
// RGB15 (RGBO fourcc) little endian to I420.
LIBYUV_API
int ARGB1555ToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGB1555ToI420(const uint8_t* src_argb1555,
int src_stride_argb1555,
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 width,
int height);
// RGB12 (R444 fourcc) little endian to I420.
LIBYUV_API
int ARGB4444ToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGB4444ToI420(const uint8_t* src_argb4444,
int src_stride_argb4444,
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 width,
int height);
#ifdef HAVE_JPEG
// src_mjpg is pointer to raw jpeg bytes in memory
// src_size_mjpg is size of jpeg in bytes
// src_width/height provided by capture.
// dst_width/height for clipping determine final size.
LIBYUV_API
int MJPGToI420(const uint8* sample, size_t sample_size,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int src_width, int src_height,
int dst_width, int dst_height);
int MJPGToI420(const uint8_t* src_mjpg,
size_t src_size_mjpg,
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);
// JPEG to NV21
LIBYUV_API
int MJPGToNV21(const uint8_t* src_mjpg,
size_t src_size_mjpg,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int src_width,
int src_height,
int dst_width,
int dst_height);
// Query size of MJPG in pixels.
LIBYUV_API
int MJPGSize(const uint8* sample, size_t sample_size,
int* width, int* height);
int MJPGSize(const uint8_t* src_mjpg,
size_t src_size_mjpg,
int* width,
int* height);
#endif
// Convert camera sample to I420 with cropping, rotation and vertical flip.
@@ -238,18 +434,25 @@ int MJPGSize(const uint8* sample, size_t sample_size,
// Must be less than or equal to src_width/src_height
// Cropping parameters are pre-rotation.
// "rotation" can be 0, 90, 180 or 270.
// "format" is a fourcc. ie 'I420', 'YUY2'
// "fourcc" is a fourcc. ie 'I420', 'YUY2'
// Returns 0 for successful; -1 for invalid parameter. Non-zero for failure.
LIBYUV_API
int ConvertToI420(const uint8* src_frame, size_t src_size,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int crop_x, int crop_y,
int src_width, int src_height,
int crop_width, int crop_height,
int ConvertToI420(const uint8_t* sample,
size_t sample_size,
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 crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32 format);
uint32_t fourcc);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/convert_argb.h
+542 -152
View File
@@ -30,258 +30,643 @@ extern "C" {
// Copy ARGB to ARGB.
LIBYUV_API
int ARGBCopy(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBCopy(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I420 to ARGB.
LIBYUV_API
int I420ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Duplicate prototype for function in convert_from.h for remoting.
LIBYUV_API
int I420ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert I010 to ARGB.
LIBYUV_API
int I010ToARGB(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I010 to ARGB.
LIBYUV_API
int I010ToARGB(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I010 to ABGR.
LIBYUV_API
int I010ToABGR(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert H010 to ARGB.
LIBYUV_API
int H010ToARGB(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert H010 to ABGR.
LIBYUV_API
int H010ToABGR(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert I422 to ARGB.
LIBYUV_API
int I422ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I422ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I444 to ARGB.
LIBYUV_API
int I444ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I444ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J444 to ARGB.
LIBYUV_API
int J444ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int J444ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I444 to ABGR.
LIBYUV_API
int I444ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
// Convert I411 to ARGB.
LIBYUV_API
int I411ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I444ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert I420 with Alpha to preattenuated ARGB.
LIBYUV_API
int I420AlphaToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
const uint8* src_a, int src_stride_a,
uint8* dst_argb, int dst_stride_argb,
int width, int height, int attenuate);
int I420AlphaToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
const uint8_t* src_a,
int src_stride_a,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height,
int attenuate);
// Convert I420 with Alpha to preattenuated ABGR.
LIBYUV_API
int I420AlphaToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
const uint8* src_a, int src_stride_a,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height, int attenuate);
int I420AlphaToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
const uint8_t* src_a,
int src_stride_a,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height,
int attenuate);
// Convert I400 (grey) to ARGB. Reverse of ARGBToI400.
LIBYUV_API
int I400ToARGB(const uint8* src_y, int src_stride_y,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I400ToARGB(const uint8_t* src_y,
int src_stride_y,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J400 (jpeg grey) to ARGB.
LIBYUV_API
int J400ToARGB(const uint8* src_y, int src_stride_y,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int J400ToARGB(const uint8_t* src_y,
int src_stride_y,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Alias.
#define YToARGB I400ToARGB
// Convert NV12 to ARGB.
LIBYUV_API
int NV12ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int NV12ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_uv,
int src_stride_uv,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert NV21 to ARGB.
LIBYUV_API
int NV21ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_vu, int src_stride_vu,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int NV21ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_vu,
int src_stride_vu,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert NV12 to ABGR.
int NV12ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_uv,
int src_stride_uv,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert NV21 to ABGR.
LIBYUV_API
int NV21ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_vu,
int src_stride_vu,
uint8_t* dst_abgr,
int dst_stride_abgr,
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 NV12 to RAW.
LIBYUV_API
int NV12ToRAW(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_uv,
int src_stride_uv,
uint8_t* dst_raw,
int dst_stride_raw,
int width,
int height);
// Convert NV21 to RAW.
LIBYUV_API
int NV21ToRAW(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_vu,
int src_stride_vu,
uint8_t* dst_raw,
int dst_stride_raw,
int width,
int height);
// Convert M420 to ARGB.
LIBYUV_API
int M420ToARGB(const uint8* src_m420, int src_stride_m420,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int M420ToARGB(const uint8_t* src_m420,
int src_stride_m420,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert YUY2 to ARGB.
LIBYUV_API
int YUY2ToARGB(const uint8* src_yuy2, int src_stride_yuy2,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int YUY2ToARGB(const uint8_t* src_yuy2,
int src_stride_yuy2,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert UYVY to ARGB.
LIBYUV_API
int UYVYToARGB(const uint8* src_uyvy, int src_stride_uyvy,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int UYVYToARGB(const uint8_t* src_uyvy,
int src_stride_uyvy,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J420 to ARGB.
LIBYUV_API
int J420ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int J420ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J422 to ARGB.
LIBYUV_API
int J422ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int J422ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J420 to ABGR.
LIBYUV_API
int J420ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int J420ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert J422 to ABGR.
LIBYUV_API
int J422ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int J422ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert H420 to ARGB.
LIBYUV_API
int H420ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int H420ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert H422 to ARGB.
LIBYUV_API
int H422ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int H422ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert H420 to ABGR.
LIBYUV_API
int H420ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int H420ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert H422 to ABGR.
LIBYUV_API
int H422ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int H422ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert H010 to ARGB.
LIBYUV_API
int H010ToARGB(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I010 to AR30.
LIBYUV_API
int I010ToAR30(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_ar30,
int dst_stride_ar30,
int width,
int height);
// Convert H010 to AR30.
LIBYUV_API
int H010ToAR30(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_ar30,
int dst_stride_ar30,
int width,
int height);
// Convert I010 to AB30.
LIBYUV_API
int I010ToAB30(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_ab30,
int dst_stride_ab30,
int width,
int height);
// Convert H010 to AB30.
LIBYUV_API
int H010ToAB30(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint8_t* dst_ab30,
int dst_stride_ab30,
int width,
int height);
// BGRA little endian (argb in memory) to ARGB.
LIBYUV_API
int BGRAToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int BGRAToARGB(const uint8_t* src_bgra,
int src_stride_bgra,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// ABGR little endian (rgba in memory) to ARGB.
LIBYUV_API
int ABGRToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ABGRToARGB(const uint8_t* src_abgr,
int src_stride_abgr,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGBA little endian (abgr in memory) to ARGB.
LIBYUV_API
int RGBAToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int RGBAToARGB(const uint8_t* src_rgba,
int src_stride_rgba,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Deprecated function name.
#define BG24ToARGB RGB24ToARGB
// RGB little endian (bgr in memory) to ARGB.
LIBYUV_API
int RGB24ToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int RGB24ToARGB(const uint8_t* src_rgb24,
int src_stride_rgb24,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGB big endian (rgb in memory) to ARGB.
LIBYUV_API
int RAWToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int RAWToARGB(const uint8_t* src_raw,
int src_stride_raw,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGB16 (RGBP fourcc) little endian to ARGB.
LIBYUV_API
int RGB565ToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int RGB565ToARGB(const uint8_t* src_rgb565,
int src_stride_rgb565,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGB15 (RGBO fourcc) little endian to ARGB.
LIBYUV_API
int ARGB1555ToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGB1555ToARGB(const uint8_t* src_argb1555,
int src_stride_argb1555,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGB12 (R444 fourcc) little endian to ARGB.
LIBYUV_API
int ARGB4444ToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGB4444ToARGB(const uint8_t* src_argb4444,
int src_stride_argb4444,
uint8_t* dst_argb,
int dst_stride_argb,
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,
int src_stride_ar30,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert AR30 To ABGR.
LIBYUV_API
int AR30ToABGR(const uint8_t* src_ar30,
int src_stride_ar30,
uint8_t* dst_abgr,
int dst_stride_abgr,
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.
LIBYUV_API
int MJPGToARGB(const uint8* sample, size_t sample_size,
uint8* dst_argb, int dst_stride_argb,
int src_width, int src_height,
int dst_width, int dst_height);
int MJPGToARGB(const uint8_t* src_mjpg,
size_t src_size_mjpg,
uint8_t* dst_argb,
int dst_stride_argb,
int src_width,
int src_height,
int dst_width,
int dst_height);
#endif
// Convert Android420 to ARGB.
LIBYUV_API
int Android420ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
int src_pixel_stride_uv,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert Android420 to ABGR.
LIBYUV_API
int Android420ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
int src_pixel_stride_uv,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert camera sample to ARGB with cropping, rotation and vertical flip.
// "src_size" is needed to parse MJPG.
// "sample_size" is needed to parse MJPG.
// "dst_stride_argb" number of bytes in a row of the dst_argb plane.
// Normally this would be the same as dst_width, with recommended alignment
// to 16 bytes for better efficiency.
@@ -300,16 +685,21 @@ int MJPGToARGB(const uint8* sample, size_t sample_size,
// Must be less than or equal to src_width/src_height
// Cropping parameters are pre-rotation.
// "rotation" can be 0, 90, 180 or 270.
// "format" is a fourcc. ie 'I420', 'YUY2'
// "fourcc" is a fourcc. ie 'I420', 'YUY2'
// Returns 0 for successful; -1 for invalid parameter. Non-zero for failure.
LIBYUV_API
int ConvertToARGB(const uint8* src_frame, size_t src_size,
uint8* dst_argb, int dst_stride_argb,
int crop_x, int crop_y,
int src_width, int src_height,
int crop_width, int crop_height,
int ConvertToARGB(const uint8_t* sample,
size_t sample_size,
uint8_t* dst_argb,
int dst_stride_argb,
int crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32 format);
uint32_t fourcc);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/convert_from.h
+292 -105
View File
@@ -21,155 +21,342 @@ extern "C" {
// See Also convert.h for conversions from formats to I420.
// I420Copy in convert to I420ToI420.
// Convert 8 bit YUV to 10 bit.
#define H420ToH010 I420ToI010
int I420ToI010(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint16_t* dst_y,
int dst_stride_y,
uint16_t* dst_u,
int dst_stride_u,
uint16_t* dst_v,
int dst_stride_v,
int width,
int height);
LIBYUV_API
int I420ToI422(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I420ToI422(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
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 width,
int height);
LIBYUV_API
int I420ToI444(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
LIBYUV_API
int I420ToI411(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I420ToI444(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
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 width,
int height);
// Copy to I400. Source can be I420, I422, I444, I400, NV12 or NV21.
LIBYUV_API
int I400Copy(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
int width, int height);
int I400Copy(const uint8_t* src_y,
int src_stride_y,
uint8_t* dst_y,
int dst_stride_y,
int width,
int height);
LIBYUV_API
int I420ToNV12(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_uv, int dst_stride_uv,
int width, int height);
int I420ToNV12(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int width,
int height);
LIBYUV_API
int I420ToNV21(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_vu, int dst_stride_vu,
int width, int height);
int I420ToNV21(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int width,
int height);
LIBYUV_API
int I420ToYUY2(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToYUY2(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_yuy2,
int dst_stride_yuy2,
int width,
int height);
LIBYUV_API
int I420ToUYVY(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToUYVY(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_uyvy,
int dst_stride_uyvy,
int width,
int height);
LIBYUV_API
int I420ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToARGB(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
LIBYUV_API
int I420ToBGRA(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToBGRA(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_bgra,
int dst_stride_bgra,
int width,
int height);
LIBYUV_API
int I420ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToABGR(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
LIBYUV_API
int I420ToRGBA(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_rgba, int dst_stride_rgba,
int width, int height);
int I420ToRGBA(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_rgba,
int dst_stride_rgba,
int width,
int height);
LIBYUV_API
int I420ToRGB24(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToRGB24(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_rgb24,
int dst_stride_rgb24,
int width,
int height);
LIBYUV_API
int I420ToRAW(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToRAW(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_raw,
int dst_stride_raw,
int width,
int height);
LIBYUV_API
int I420ToRGB565(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int H420ToRGB24(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_rgb24,
int dst_stride_rgb24,
int width,
int height);
LIBYUV_API
int H420ToRAW(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_raw,
int dst_stride_raw,
int width,
int height);
LIBYUV_API
int I420ToRGB565(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_rgb565,
int dst_stride_rgb565,
int width,
int height);
LIBYUV_API
int J420ToRGB565(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_rgb565,
int dst_stride_rgb565,
int width,
int height);
LIBYUV_API
int H420ToRGB565(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_rgb565,
int dst_stride_rgb565,
int width,
int height);
LIBYUV_API
int I422ToRGB565(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_rgb565,
int dst_stride_rgb565,
int width,
int height);
// Convert I420 To RGB565 with 4x4 dither matrix (16 bytes).
// Values in dither matrix from 0 to 7 recommended.
// The order of the dither matrix is first byte is upper left.
LIBYUV_API
int I420ToRGB565Dither(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
const uint8* dither4x4, int width, int height);
int I420ToRGB565Dither(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_rgb565,
int dst_stride_rgb565,
const uint8_t* dither4x4,
int width,
int height);
LIBYUV_API
int I420ToARGB1555(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToARGB1555(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb1555,
int dst_stride_argb1555,
int width,
int height);
LIBYUV_API
int I420ToARGB4444(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToARGB4444(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_argb4444,
int dst_stride_argb4444,
int width,
int height);
// Convert I420 to AR30.
LIBYUV_API
int I420ToAR30(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_ar30,
int dst_stride_ar30,
int width,
int height);
// Convert H420 to AR30.
LIBYUV_API
int H420ToAR30(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_ar30,
int dst_stride_ar30,
int width,
int height);
// Convert I420 to specified format.
// "dst_sample_stride" is bytes in a row for the destination. Pass 0 if the
// buffer has contiguous rows. Can be negative. A multiple of 16 is optimal.
LIBYUV_API
int ConvertFromI420(const uint8* y, int y_stride,
const uint8* u, int u_stride,
const uint8* v, int v_stride,
uint8* dst_sample, int dst_sample_stride,
int width, int height,
uint32 format);
int ConvertFromI420(const uint8_t* y,
int y_stride,
const uint8_t* u,
int u_stride,
const uint8_t* v,
int v_stride,
uint8_t* dst_sample,
int dst_sample_stride,
int width,
int height,
uint32_t fourcc);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/convert_from_argb.h
+188 -91
View File
@@ -21,166 +21,263 @@ extern "C" {
// Copy ARGB to ARGB.
#define ARGBToARGB ARGBCopy
LIBYUV_API
int ARGBCopy(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBCopy(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert ARGB To BGRA.
LIBYUV_API
int ARGBToBGRA(const uint8* src_argb, int src_stride_argb,
uint8* dst_bgra, int dst_stride_bgra,
int width, int height);
int ARGBToBGRA(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_bgra,
int dst_stride_bgra,
int width,
int height);
// Convert ARGB To ABGR.
LIBYUV_API
int ARGBToABGR(const uint8* src_argb, int src_stride_argb,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int ARGBToABGR(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert ARGB To RGBA.
LIBYUV_API
int ARGBToRGBA(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgba, int dst_stride_rgba,
int width, int height);
int ARGBToRGBA(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_rgba,
int dst_stride_rgba,
int width,
int height);
// Aliases
#define ARGBToAB30 ABGRToAR30
#define ABGRToAB30 ARGBToAR30
// Convert ABGR To AR30.
LIBYUV_API
int ABGRToAR30(const uint8_t* src_abgr,
int src_stride_abgr,
uint8_t* dst_ar30,
int dst_stride_ar30,
int width,
int height);
// Convert ARGB To AR30.
LIBYUV_API
int ARGBToAR30(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_ar30,
int dst_stride_ar30,
int width,
int height);
// Convert ARGB To RGB24.
LIBYUV_API
int ARGBToRGB24(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgb24, int dst_stride_rgb24,
int width, int height);
int ARGBToRGB24(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_rgb24,
int dst_stride_rgb24,
int width,
int height);
// Convert ARGB To RAW.
LIBYUV_API
int ARGBToRAW(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgb, int dst_stride_rgb,
int width, int height);
int ARGBToRAW(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_raw,
int dst_stride_raw,
int width,
int height);
// Convert ARGB To RGB565.
LIBYUV_API
int ARGBToRGB565(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgb565, int dst_stride_rgb565,
int width, int height);
int ARGBToRGB565(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_rgb565,
int dst_stride_rgb565,
int width,
int height);
// Convert ARGB To RGB565 with 4x4 dither matrix (16 bytes).
// Values in dither matrix from 0 to 7 recommended.
// The order of the dither matrix is first byte is upper left.
// TODO(fbarchard): Consider pointer to 2d array for dither4x4.
// const uint8(*dither)[4][4];
// const uint8_t(*dither)[4][4];
LIBYUV_API
int ARGBToRGB565Dither(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgb565, int dst_stride_rgb565,
const uint8* dither4x4, int width, int height);
int ARGBToRGB565Dither(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_rgb565,
int dst_stride_rgb565,
const uint8_t* dither4x4,
int width,
int height);
// Convert ARGB To ARGB1555.
LIBYUV_API
int ARGBToARGB1555(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb1555, int dst_stride_argb1555,
int width, int height);
int ARGBToARGB1555(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb1555,
int dst_stride_argb1555,
int width,
int height);
// Convert ARGB To ARGB4444.
LIBYUV_API
int ARGBToARGB4444(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb4444, int dst_stride_argb4444,
int width, int height);
int ARGBToARGB4444(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb4444,
int dst_stride_argb4444,
int width,
int height);
// Convert ARGB To I444.
LIBYUV_API
int ARGBToI444(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToI444(const uint8_t* src_argb,
int src_stride_argb,
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 width,
int height);
// Convert ARGB To I422.
LIBYUV_API
int ARGBToI422(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToI422(const uint8_t* src_argb,
int src_stride_argb,
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 width,
int height);
// Convert ARGB To I420. (also in convert.h)
LIBYUV_API
int ARGBToI420(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToI420(const uint8_t* src_argb,
int src_stride_argb,
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 width,
int height);
// Convert ARGB to J420. (JPeg full range I420).
LIBYUV_API
int ARGBToJ420(const uint8* src_argb, int src_stride_argb,
uint8* dst_yj, int dst_stride_yj,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToJ420(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_yj,
int dst_stride_yj,
uint8_t* dst_u,
int dst_stride_u,
uint8_t* dst_v,
int dst_stride_v,
int width,
int height);
// Convert ARGB to J422.
LIBYUV_API
int ARGBToJ422(const uint8* src_argb, int src_stride_argb,
uint8* dst_yj, int dst_stride_yj,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
// Convert ARGB To I411.
LIBYUV_API
int ARGBToI411(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToJ422(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_yj,
int dst_stride_yj,
uint8_t* dst_u,
int dst_stride_u,
uint8_t* dst_v,
int dst_stride_v,
int width,
int height);
// Convert ARGB to J400. (JPeg full range).
LIBYUV_API
int ARGBToJ400(const uint8* src_argb, int src_stride_argb,
uint8* dst_yj, int dst_stride_yj,
int width, int height);
int ARGBToJ400(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_yj,
int dst_stride_yj,
int width,
int height);
// Convert ARGB to I400.
LIBYUV_API
int ARGBToI400(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
int width, int height);
int ARGBToI400(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_y,
int dst_stride_y,
int width,
int height);
// Convert ARGB to G. (Reverse of J400toARGB, which replicates G back to ARGB)
LIBYUV_API
int ARGBToG(const uint8* src_argb, int src_stride_argb,
uint8* dst_g, int dst_stride_g,
int width, int height);
int ARGBToG(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_g,
int dst_stride_g,
int width,
int height);
// Convert ARGB To NV12.
LIBYUV_API
int ARGBToNV12(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_uv, int dst_stride_uv,
int width, int height);
int ARGBToNV12(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int width,
int height);
// Convert ARGB To NV21.
LIBYUV_API
int ARGBToNV21(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_vu, int dst_stride_vu,
int width, int height);
int ARGBToNV21(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int width,
int height);
// Convert ARGB To NV21.
LIBYUV_API
int ARGBToNV21(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_vu, int dst_stride_vu,
int width, int height);
int ARGBToNV21(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int width,
int height);
// Convert ARGB To YUY2.
LIBYUV_API
int ARGBToYUY2(const uint8* src_argb, int src_stride_argb,
uint8* dst_yuy2, int dst_stride_yuy2,
int width, int height);
int ARGBToYUY2(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_yuy2,
int dst_stride_yuy2,
int width,
int height);
// Convert ARGB To UYVY.
LIBYUV_API
int ARGBToUYVY(const uint8* src_argb, int src_stride_argb,
uint8* dst_uyvy, int dst_stride_uyvy,
int width, int height);
int ARGBToUYVY(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_uyvy,
int dst_stride_uyvy,
int width,
int height);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/cpu_id.h
+56 -17
View File
@@ -31,47 +31,86 @@ static const int kCpuHasX86 = 0x10;
static const int kCpuHasSSE2 = 0x20;
static const int kCpuHasSSSE3 = 0x40;
static const int kCpuHasSSE41 = 0x80;
static const int kCpuHasSSE42 = 0x100;
static const int kCpuHasSSE42 = 0x100; // unused at this time.
static const int kCpuHasAVX = 0x200;
static const int kCpuHasAVX2 = 0x400;
static const int kCpuHasERMS = 0x800;
static const int kCpuHasFMA3 = 0x1000;
static const int kCpuHasAVX3 = 0x2000;
// 0x2000, 0x4000, 0x8000 reserved for future X86 flags.
static const int kCpuHasF16C = 0x2000;
static const int kCpuHasGFNI = 0x4000;
static const int kCpuHasAVX512BW = 0x8000;
static const int kCpuHasAVX512VL = 0x10000;
static const int kCpuHasAVX512VBMI = 0x20000;
static const int kCpuHasAVX512VBMI2 = 0x40000;
static const int kCpuHasAVX512VBITALG = 0x80000;
static const int kCpuHasAVX512VPOPCNTDQ = 0x100000;
// These flags are only valid on MIPS processors.
static const int kCpuHasMIPS = 0x10000;
static const int kCpuHasDSPR2 = 0x20000;
static const int kCpuHasMSA = 0x40000;
static const int kCpuHasMIPS = 0x200000;
static const int kCpuHasMSA = 0x400000;
static const int kCpuHasMMI = 0x800000;
// Internal function used to auto-init.
// Optional init function. TestCpuFlag does an auto-init.
// Returns cpu_info flags.
LIBYUV_API
int InitCpuFlags(void);
// Detect CPU has SSE2 etc.
// Test_flag parameter should be one of kCpuHas constants above.
// Returns non-zero if instruction set is detected
static __inline int TestCpuFlag(int test_flag) {
LIBYUV_API extern int cpu_info_;
#ifdef __ATOMIC_RELAXED
int cpu_info = __atomic_load_n(&cpu_info_, __ATOMIC_RELAXED);
#else
int cpu_info = cpu_info_;
#endif
return (!cpu_info ? InitCpuFlags() : cpu_info) & test_flag;
}
// Internal function for parsing /proc/cpuinfo.
LIBYUV_API
int ArmCpuCaps(const char* cpuinfo_name);
// Detect CPU has SSE2 etc.
// Test_flag parameter should be one of kCpuHas constants above.
// returns non-zero if instruction set is detected
static __inline int TestCpuFlag(int test_flag) {
LIBYUV_API extern int cpu_info_;
return (!cpu_info_ ? InitCpuFlags() : cpu_info_) & test_flag;
}
// For testing, allow CPU flags to be disabled.
// ie MaskCpuFlags(~kCpuHasSSSE3) to disable SSSE3.
// MaskCpuFlags(-1) to enable all cpu specific optimizations.
// MaskCpuFlags(1) to disable all cpu specific optimizations.
// MaskCpuFlags(0) to reset state so next call will auto init.
// Returns cpu_info flags.
LIBYUV_API
void MaskCpuFlags(int enable_flags);
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.
LIBYUV_API
void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info);
void CpuId(int info_eax, int info_ecx, int* cpu_info);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/macros_msa.h
+186 -29
View File
@@ -12,15 +12,140 @@
#define INCLUDE_LIBYUV_MACROS_MSA_H_
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#include <stdint.h>
#include <msa.h>
#include <stdint.h>
#define LD_B(RTYPE, psrc) *((RTYPE*)(psrc)) /* NOLINT */
#define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
#if (__mips_isa_rev >= 6)
#define LW(psrc) \
({ \
const uint8_t* psrc_lw_m = (const uint8_t*)(psrc); \
uint32_t val_m; \
asm volatile("lw %[val_m], %[psrc_lw_m] \n" \
: [val_m] "=r"(val_m) \
: [psrc_lw_m] "m"(*psrc_lw_m)); \
val_m; \
})
#define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
#if (__mips == 64)
#define LD(psrc) \
({ \
const uint8_t* psrc_ld_m = (const uint8_t*)(psrc); \
uint64_t val_m = 0; \
asm volatile("ld %[val_m], %[psrc_ld_m] \n" \
: [val_m] "=r"(val_m) \
: [psrc_ld_m] "m"(*psrc_ld_m)); \
val_m; \
})
#else // !(__mips == 64)
#define LD(psrc) \
({ \
const uint8_t* psrc_ld_m = (const uint8_t*)(psrc); \
uint32_t val0_m, val1_m; \
uint64_t val_m = 0; \
val0_m = LW(psrc_ld_m); \
val1_m = LW(psrc_ld_m + 4); \
val_m = (uint64_t)(val1_m); /* NOLINT */ \
val_m = (uint64_t)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \
val_m = (uint64_t)(val_m | (uint64_t)val0_m); /* NOLINT */ \
val_m; \
})
#endif // (__mips == 64)
#define SW(val, pdst) \
({ \
uint8_t* pdst_sw_m = (uint8_t*)(pdst); /* NOLINT */ \
uint32_t val_m = (val); \
asm volatile("sw %[val_m], %[pdst_sw_m] \n" \
: [pdst_sw_m] "=m"(*pdst_sw_m) \
: [val_m] "r"(val_m)); \
})
#if (__mips == 64)
#define SD(val, pdst) \
({ \
uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */ \
uint64_t val_m = (val); \
asm volatile("sd %[val_m], %[pdst_sd_m] \n" \
: [pdst_sd_m] "=m"(*pdst_sd_m) \
: [val_m] "r"(val_m)); \
})
#else // !(__mips == 64)
#define SD(val, pdst) \
({ \
uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */ \
uint32_t val0_m, val1_m; \
val0_m = (uint32_t)((val)&0x00000000FFFFFFFF); \
val1_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
SW(val0_m, pdst_sd_m); \
SW(val1_m, pdst_sd_m + 4); \
})
#endif // !(__mips == 64)
#else // !(__mips_isa_rev >= 6)
#define LW(psrc) \
({ \
const uint8_t* psrc_lw_m = (const uint8_t*)(psrc); \
uint32_t val_m; \
asm volatile("ulw %[val_m], %[psrc_lw_m] \n" \
: [val_m] "=r"(val_m) \
: [psrc_lw_m] "m"(*psrc_lw_m)); \
val_m; \
})
#if (__mips == 64)
#define LD(psrc) \
({ \
const uint8_t* psrc_ld_m = (const uint8_t*)(psrc); \
uint64_t val_m = 0; \
asm volatile("uld %[val_m], %[psrc_ld_m] \n" \
: [val_m] "=r"(val_m) \
: [psrc_ld_m] "m"(*psrc_ld_m)); \
val_m; \
})
#else // !(__mips == 64)
#define LD(psrc) \
({ \
const uint8_t* psrc_ld_m = (const uint8_t*)(psrc); \
uint32_t val0_m, val1_m; \
uint64_t val_m = 0; \
val0_m = LW(psrc_ld_m); \
val1_m = LW(psrc_ld_m + 4); \
val_m = (uint64_t)(val1_m); /* NOLINT */ \
val_m = (uint64_t)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \
val_m = (uint64_t)(val_m | (uint64_t)val0_m); /* NOLINT */ \
val_m; \
})
#endif // (__mips == 64)
#define SW(val, pdst) \
({ \
uint8_t* pdst_sw_m = (uint8_t*)(pdst); /* NOLINT */ \
uint32_t val_m = (val); \
asm volatile("usw %[val_m], %[pdst_sw_m] \n" \
: [pdst_sw_m] "=m"(*pdst_sw_m) \
: [val_m] "r"(val_m)); \
})
#define SD(val, pdst) \
({ \
uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */ \
uint32_t val0_m, val1_m; \
val0_m = (uint32_t)((val)&0x00000000FFFFFFFF); \
val1_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
SW(val0_m, pdst_sd_m); \
SW(val1_m, pdst_sd_m + 4); \
})
#endif // (__mips_isa_rev >= 6)
// TODO(fbarchard): Consider removing __VAR_ARGS versions.
#define LD_B(RTYPE, psrc) *((RTYPE*)(psrc)) /* NOLINT */
#define LD_UB(...) LD_B(const v16u8, __VA_ARGS__)
#define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
#define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
#define ST_H(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
#define ST_UH(...) ST_H(v8u16, __VA_ARGS__)
/* Description : Load two vectors with 16 'byte' sized elements
Arguments : Inputs - psrc, stride
Outputs - out0, out1
@@ -28,17 +153,19 @@
Details : Load 16 byte elements in 'out0' from (psrc)
Load 16 byte elements in 'out1' from (psrc + stride)
*/
#define LD_B2(RTYPE, psrc, stride, out0, out1) { \
out0 = LD_B(RTYPE, (psrc)); \
out1 = LD_B(RTYPE, (psrc) + stride); \
}
#define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
#define LD_B2(RTYPE, psrc, stride, out0, out1) \
{ \
out0 = LD_B(RTYPE, (psrc)); \
out1 = LD_B(RTYPE, (psrc) + stride); \
}
#define LD_UB2(...) LD_B2(const v16u8, __VA_ARGS__)
#define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3) { \
LD_B2(RTYPE, (psrc), stride, out0, out1); \
LD_B2(RTYPE, (psrc) + 2 * stride , stride, out2, out3); \
}
#define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
#define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3) \
{ \
LD_B2(RTYPE, (psrc), stride, out0, out1); \
LD_B2(RTYPE, (psrc) + 2 * stride, stride, out2, out3); \
}
#define LD_UB4(...) LD_B4(const v16u8, __VA_ARGS__)
/* Description : Store two vectors with stride each having 16 'byte' sized
elements
@@ -46,18 +173,33 @@
Details : Store 16 byte elements from 'in0' to (pdst)
Store 16 byte elements from 'in1' to (pdst + stride)
*/
#define ST_B2(RTYPE, in0, in1, pdst, stride) { \
ST_B(RTYPE, in0, (pdst)); \
ST_B(RTYPE, in1, (pdst) + stride); \
}
#define ST_B2(RTYPE, in0, in1, pdst, stride) \
{ \
ST_B(RTYPE, in0, (pdst)); \
ST_B(RTYPE, in1, (pdst) + stride); \
}
#define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)
#
#define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride) { \
ST_B2(RTYPE, in0, in1, (pdst), stride); \
ST_B2(RTYPE, in2, in3, (pdst) + 2 * stride, stride); \
}
#define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride) \
{ \
ST_B2(RTYPE, in0, in1, (pdst), stride); \
ST_B2(RTYPE, in2, in3, (pdst) + 2 * stride, stride); \
}
#define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
#
/* Description : Store vectors of 8 halfword elements with stride
Arguments : Inputs - in0, in1, pdst, stride
Details : Store 8 halfword elements from 'in0' to (pdst)
Store 8 halfword elements from 'in1' to (pdst + stride)
*/
#define ST_H2(RTYPE, in0, in1, pdst, stride) \
{ \
ST_H(RTYPE, in0, (pdst)); \
ST_H(RTYPE, in1, (pdst) + stride); \
}
#define ST_UH2(...) ST_H2(v8u16, __VA_ARGS__)
// TODO(fbarchard): Consider using __msa_vshf_b and __msa_ilvr_b directly.
/* Description : Shuffle byte vector elements as per mask vector
Arguments : Inputs - in0, in1, in2, in3, mask0, mask1
Outputs - out0, out1
@@ -65,12 +207,27 @@
Details : Byte elements from 'in0' & 'in1' are copied selectively to
'out0' as per control vector 'mask0'
*/
#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) { \
out0 = (RTYPE) __msa_vshf_b((v16i8) mask0, (v16i8) in1, (v16i8) in0); \
out1 = (RTYPE) __msa_vshf_b((v16i8) mask1, (v16i8) in3, (v16i8) in2); \
}
#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) \
{ \
out0 = (RTYPE)__msa_vshf_b((v16i8)mask0, (v16i8)in1, (v16i8)in0); \
out1 = (RTYPE)__msa_vshf_b((v16i8)mask1, (v16i8)in3, (v16i8)in2); \
}
#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
#endif /* !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) */
/* Description : Interleave both left and right half of input vectors
Arguments : Inputs - in0, in1
Outputs - out0, out1
Return Type - as per RTYPE
Details : Right half of byte elements from 'in0' and 'in1' are
interleaved and written to 'out0'
*/
#define ILVRL_B2(RTYPE, in0, in1, out0, out1) \
{ \
out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1); \
out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1); \
}
#define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)
#endif /* !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) */
#endif // INCLUDE_LIBYUV_MACROS_MSA_H_
libs/libyuv/include/libyuv/mjpeg_decoder.h
+16 -13
View File
@@ -26,25 +26,24 @@ namespace libyuv {
extern "C" {
#endif
LIBYUV_BOOL ValidateJpeg(const uint8* sample, size_t sample_size);
LIBYUV_BOOL ValidateJpeg(const uint8_t* src_mjpg, size_t src_size_mjpg_size);
#ifdef __cplusplus
} // extern "C"
#endif
static const uint32 kUnknownDataSize = 0xFFFFFFFF;
static const uint32_t kUnknownDataSize = 0xFFFFFFFF;
enum JpegSubsamplingType {
kJpegYuv420,
kJpegYuv422,
kJpegYuv411,
kJpegYuv444,
kJpegYuv400,
kJpegUnknown
};
struct Buffer {
const uint8* data;
const uint8_t* data;
int len;
};
@@ -66,7 +65,7 @@ struct SetJmpErrorMgr;
class LIBYUV_API MJpegDecoder {
public:
typedef void (*CallbackFunction)(void* opaque,
const uint8* const* data,
const uint8_t* const* data,
const int* strides,
int rows);
@@ -86,7 +85,7 @@ class LIBYUV_API MJpegDecoder {
// If return value is LIBYUV_TRUE, then the values for all the following
// getters are populated.
// src_len is the size of the compressed mjpeg frame in bytes.
LIBYUV_BOOL LoadFrame(const uint8* src, size_t src_len);
LIBYUV_BOOL LoadFrame(const uint8_t* src, size_t src_len);
// Returns width of the last loaded frame in pixels.
int GetWidth();
@@ -139,18 +138,22 @@ class LIBYUV_API MJpegDecoder {
// at least GetComponentSize(i). The pointers in planes are incremented
// to point to after the end of the written data.
// TODO(fbarchard): Add dst_x, dst_y to allow specific rect to be decoded.
LIBYUV_BOOL DecodeToBuffers(uint8** planes, int dst_width, int dst_height);
LIBYUV_BOOL DecodeToBuffers(uint8_t** planes, int dst_width, int dst_height);
// Decodes the entire image and passes the data via repeated calls to a
// callback function. Each call will get the data for a whole number of
// image scanlines.
// TODO(fbarchard): Add dst_x, dst_y to allow specific rect to be decoded.
LIBYUV_BOOL DecodeToCallback(CallbackFunction fn, void* opaque,
int dst_width, int dst_height);
LIBYUV_BOOL DecodeToCallback(CallbackFunction fn,
void* opaque,
int dst_width,
int dst_height);
// The helper function which recognizes the jpeg sub-sampling type.
static JpegSubsamplingType JpegSubsamplingTypeHelper(
int* subsample_x, int* subsample_y, int number_of_components);
int* subsample_x,
int* subsample_y,
int number_of_components);
private:
void AllocOutputBuffers(int num_outbufs);
@@ -159,7 +162,7 @@ class LIBYUV_API MJpegDecoder {
LIBYUV_BOOL StartDecode();
LIBYUV_BOOL FinishDecode();
void SetScanlinePointers(uint8** data);
void SetScanlinePointers(uint8_t** data);
LIBYUV_BOOL DecodeImcuRow();
int GetComponentScanlinePadding(int component);
@@ -178,11 +181,11 @@ class LIBYUV_API MJpegDecoder {
// Temporaries used to point to scanline outputs.
int num_outbufs_; // Outermost size of all arrays below.
uint8*** scanlines_;
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** databuf_;
uint8_t** databuf_;
int* databuf_strides_;
};
libs/libyuv/include/libyuv/planar_functions.h
+785 -459
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/include/libyuv/rotate.h
+93 -46
View File
@@ -20,8 +20,8 @@ extern "C" {
// Supported rotation.
typedef enum RotationMode {
kRotate0 = 0, // No rotation.
kRotate90 = 90, // Rotate 90 degrees clockwise.
kRotate0 = 0, // No rotation.
kRotate90 = 90, // Rotate 90 degrees clockwise.
kRotate180 = 180, // Rotate 180 degrees.
kRotate270 = 270, // Rotate 270 degrees clockwise.
@@ -33,81 +33,128 @@ typedef enum RotationMode {
// Rotate I420 frame.
LIBYUV_API
int I420Rotate(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int src_width, int src_height, enum RotationMode mode);
int I420Rotate(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
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 width,
int height,
enum RotationMode mode);
// Rotate NV12 input and store in I420.
LIBYUV_API
int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int src_width, int src_height, enum RotationMode mode);
int NV12ToI420Rotate(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_uv,
int src_stride_uv,
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 width,
int height,
enum RotationMode mode);
// Rotate a plane by 0, 90, 180, or 270.
LIBYUV_API
int RotatePlane(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int src_width, int src_height, enum RotationMode mode);
int RotatePlane(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height,
enum RotationMode mode);
// Rotate planes by 90, 180, 270. Deprecated.
LIBYUV_API
void RotatePlane90(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height);
void RotatePlane90(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height);
LIBYUV_API
void RotatePlane180(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height);
void RotatePlane180(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height);
LIBYUV_API
void RotatePlane270(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height);
void RotatePlane270(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height);
LIBYUV_API
void RotateUV90(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void RotateUV90(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height);
// Rotations for when U and V are interleaved.
// These functions take one input pointer and
// split the data into two buffers while
// rotating them. Deprecated.
LIBYUV_API
void RotateUV180(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void RotateUV180(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height);
LIBYUV_API
void RotateUV270(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void RotateUV270(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height);
// The 90 and 270 functions are based on transposes.
// Doing a transpose with reversing the read/write
// order will result in a rotation by +- 90 degrees.
// Deprecated.
LIBYUV_API
void TransposePlane(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height);
void TransposePlane(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height);
LIBYUV_API
void TransposeUV(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void TransposeUV(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/rotate_argb.h
+7 -3
View File
@@ -21,9 +21,13 @@ extern "C" {
// Rotate ARGB frame
LIBYUV_API
int ARGBRotate(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int src_width, int src_height, enum RotationMode mode);
int ARGBRotate(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int src_width,
int src_height,
enum RotationMode mode);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/rotate_row.h
+165 -63
View File
@@ -18,10 +18,14 @@ namespace libyuv {
extern "C" {
#endif
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__i386__) && !defined(__SSE2__))
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__native_client__) && defined(__x86_64__)) || \
(defined(__i386__) && !defined(__SSE__) && !defined(__clang__))
#define LIBYUV_DISABLE_X86
#endif
#if defined(__native_client__)
#define LIBYUV_DISABLE_NEON
#endif
// MemorySanitizer does not support assembly code yet. http://crbug.com/344505
#if defined(__has_feature)
#if __has_feature(memory_sanitizer)
@@ -29,89 +33,187 @@ extern "C" {
#endif
#endif
// The following are available for Visual C and clangcl 32 bit:
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
#define HAS_TRANSPOSEWX8_SSSE3
#define HAS_TRANSPOSEUVWX8_SSE2
#endif
// The following are available for GCC 32 or 64 bit but not NaCL for 64 bit:
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__i386__) || (defined(__x86_64__) && !defined(__native_client__)))
// The following are available for GCC 32 or 64 bit:
#if !defined(LIBYUV_DISABLE_X86) && (defined(__i386__) || defined(__x86_64__))
#define HAS_TRANSPOSEWX8_SSSE3
#endif
// The following are available for 64 bit GCC but not NaCL:
#if !defined(LIBYUV_DISABLE_X86) && !defined(__native_client__) && \
defined(__x86_64__)
// The following are available for 64 bit GCC:
#if !defined(LIBYUV_DISABLE_X86) && defined(__x86_64__)
#define HAS_TRANSPOSEWX8_FAST_SSSE3
#define HAS_TRANSPOSEUVWX8_SSE2
#endif
#if !defined(LIBYUV_DISABLE_NEON) && !defined(__native_client__) && \
#if !defined(LIBYUV_DISABLE_NEON) && \
(defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))
#define HAS_TRANSPOSEWX8_NEON
#define HAS_TRANSPOSEUVWX8_NEON
#endif
#if !defined(LIBYUV_DISABLE_MIPS) && !defined(__native_client__) && \
defined(__mips__) && \
defined(__mips_dsp) && (__mips_dsp_rev >= 2)
#define HAS_TRANSPOSEWX8_DSPR2
#define HAS_TRANSPOSEUVWX8_DSPR2
#endif // defined(__mips__)
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#define HAS_TRANSPOSEWX16_MSA
#define HAS_TRANSPOSEUVWX16_MSA
#endif
void TransposeWxH_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width, int height);
#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
#define HAS_TRANSPOSEWX8_MMI
#define HAS_TRANSPOSEUVWX8_MMI
#endif
void TransposeWx8_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_NEON(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Fast_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Fast_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWxH_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height);
void TransposeWx8_Any_NEON(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Any_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Fast_Any_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Any_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx16_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx8_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx8_MMI(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx8_Fast_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx16_MSA(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeUVWxH_C(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void TransposeWx8_Any_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx8_Any_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx8_Any_MMI(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx8_Fast_Any_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeWx16_Any_MSA(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width);
void TransposeUVWx8_C(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_NEON(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_DSPR2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWxH_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height);
void TransposeUVWx8_Any_SSE2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_Any_NEON(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_Any_DSPR2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx16_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_SSE2(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_MMI(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx16_MSA(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_Any_SSE2(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_Any_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_Any_MMI(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx16_Any_MSA(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/include/libyuv/row.h
+3285 -1263
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/include/libyuv/scale.h
+67 -39
View File
@@ -20,25 +20,33 @@ extern "C" {
// Supported filtering.
typedef enum FilterMode {
kFilterNone = 0, // Point sample; Fastest.
kFilterLinear = 1, // Filter horizontally only.
kFilterNone = 0, // Point sample; Fastest.
kFilterLinear = 1, // Filter horizontally only.
kFilterBilinear = 2, // Faster than box, but lower quality scaling down.
kFilterBox = 3 // Highest quality.
kFilterBox = 3 // Highest quality.
} FilterModeEnum;
// Scale a YUV plane.
LIBYUV_API
void ScalePlane(const uint8* src, int src_stride,
int src_width, int src_height,
uint8* dst, int dst_stride,
int dst_width, int dst_height,
void ScalePlane(const uint8_t* src,
int src_stride,
int src_width,
int src_height,
uint8_t* dst,
int dst_stride,
int dst_width,
int dst_height,
enum FilterMode filtering);
LIBYUV_API
void ScalePlane_16(const uint16* src, int src_stride,
int src_width, int src_height,
uint16* dst, int dst_stride,
int dst_width, int dst_height,
void ScalePlane_16(const uint16_t* src,
int src_stride,
int src_width,
int src_height,
uint16_t* dst,
int dst_stride,
int dst_width,
int dst_height,
enum FilterMode filtering);
// Scales a YUV 4:2:0 image from the src width and height to the
@@ -52,44 +60,64 @@ void ScalePlane_16(const uint16* src, int src_stride,
// Returns 0 if successful.
LIBYUV_API
int I420Scale(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
int src_width, int src_height,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int dst_width, int dst_height,
int I420Scale(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
int src_width,
int src_height,
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 dst_width,
int dst_height,
enum FilterMode filtering);
LIBYUV_API
int I420Scale_16(const uint16* src_y, int src_stride_y,
const uint16* src_u, int src_stride_u,
const uint16* src_v, int src_stride_v,
int src_width, int src_height,
uint16* dst_y, int dst_stride_y,
uint16* dst_u, int dst_stride_u,
uint16* dst_v, int dst_stride_v,
int dst_width, int dst_height,
int I420Scale_16(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
int src_width,
int src_height,
uint16_t* dst_y,
int dst_stride_y,
uint16_t* dst_u,
int dst_stride_u,
uint16_t* dst_v,
int dst_stride_v,
int dst_width,
int dst_height,
enum FilterMode filtering);
#ifdef __cplusplus
// Legacy API. Deprecated.
LIBYUV_API
int Scale(const uint8* src_y, const uint8* src_u, const uint8* src_v,
int src_stride_y, int src_stride_u, int src_stride_v,
int src_width, int src_height,
uint8* dst_y, uint8* dst_u, uint8* dst_v,
int dst_stride_y, int dst_stride_u, int dst_stride_v,
int dst_width, int dst_height,
int Scale(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
int src_stride_y,
int src_stride_u,
int src_stride_v,
int src_width,
int src_height,
uint8_t* dst_y,
uint8_t* dst_u,
uint8_t* dst_v,
int dst_stride_y,
int dst_stride_u,
int dst_stride_v,
int dst_width,
int dst_height,
LIBYUV_BOOL interpolate);
// Legacy API. Deprecated.
LIBYUV_API
int ScaleOffset(const uint8* src_i420, int src_width, int src_height,
uint8* dst_i420, int dst_width, int dst_height, int dst_yoffset,
LIBYUV_BOOL interpolate);
// For testing, allow disabling of specialized scalers.
LIBYUV_API
void SetUseReferenceImpl(LIBYUV_BOOL use);
libs/libyuv/include/libyuv/scale_argb.h
+38 -18
View File
@@ -20,32 +20,52 @@ extern "C" {
#endif
LIBYUV_API
int ARGBScale(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
int ARGBScale(const uint8_t* src_argb,
int src_stride_argb,
int src_width,
int src_height,
uint8_t* dst_argb,
int dst_stride_argb,
int dst_width,
int dst_height,
enum FilterMode filtering);
// Clipped scale takes destination rectangle coordinates for clip values.
LIBYUV_API
int ARGBScaleClip(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
int ARGBScaleClip(const uint8_t* src_argb,
int src_stride_argb,
int src_width,
int src_height,
uint8_t* dst_argb,
int dst_stride_argb,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering);
// Scale with YUV conversion to ARGB and clipping.
LIBYUV_API
int YUVToARGBScaleClip(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint32 src_fourcc,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
uint32 dst_fourcc,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
int YUVToARGBScaleClip(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint32_t src_fourcc,
int src_width,
int src_height,
uint8_t* dst_argb,
int dst_stride_argb,
uint32_t dst_fourcc,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering);
#ifdef __cplusplus
libs/libyuv/include/libyuv/scale_row.h
+926 -322
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/include/libyuv/version.h
+1 -1
View File
@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1620
#define LIBYUV_VERSION 1724
#endif // INCLUDE_LIBYUV_VERSION_H_
libs/libyuv/include/libyuv/video_common.h
+29 -22
View File
@@ -28,13 +28,14 @@ extern "C" {
// Needs to be a macro otherwise the OS X compiler complains when the kFormat*
// constants are used in a switch.
#ifdef __cplusplus
#define FOURCC(a, b, c, d) ( \
(static_cast<uint32>(a)) | (static_cast<uint32>(b) << 8) | \
(static_cast<uint32>(c) << 16) | (static_cast<uint32>(d) << 24))
#define FOURCC(a, b, c, d) \
((static_cast<uint32_t>(a)) | (static_cast<uint32_t>(b) << 8) | \
(static_cast<uint32_t>(c) << 16) | /* NOLINT */ \
(static_cast<uint32_t>(d) << 24)) /* NOLINT */
#else
#define FOURCC(a, b, c, d) ( \
((uint32)(a)) | ((uint32)(b) << 8) | /* NOLINT */ \
((uint32)(c) << 16) | ((uint32)(d) << 24)) /* NOLINT */
#define FOURCC(a, b, c, d) \
(((uint32_t)(a)) | ((uint32_t)(b) << 8) | /* NOLINT */ \
((uint32_t)(c) << 16) | ((uint32_t)(d) << 24)) /* NOLINT */
#endif
// Some pages discussing FourCC codes:
@@ -53,38 +54,33 @@ enum FourCC {
FOURCC_I420 = FOURCC('I', '4', '2', '0'),
FOURCC_I422 = FOURCC('I', '4', '2', '2'),
FOURCC_I444 = FOURCC('I', '4', '4', '4'),
FOURCC_I411 = FOURCC('I', '4', '1', '1'),
FOURCC_I400 = FOURCC('I', '4', '0', '0'),
FOURCC_NV21 = FOURCC('N', 'V', '2', '1'),
FOURCC_NV12 = FOURCC('N', 'V', '1', '2'),
FOURCC_YUY2 = FOURCC('Y', 'U', 'Y', '2'),
FOURCC_UYVY = FOURCC('U', 'Y', 'V', 'Y'),
FOURCC_H010 = FOURCC('H', '0', '1', '0'), // unofficial fourcc. 10 bit lsb
// 2 Secondary YUV formats: row biplanar.
// 1 Secondary YUV format: row biplanar.
FOURCC_M420 = FOURCC('M', '4', '2', '0'),
FOURCC_Q420 = FOURCC('Q', '4', '2', '0'), // deprecated.
// 9 Primary RGB formats: 4 32 bpp, 2 24 bpp, 3 16 bpp.
// 11 Primary RGB formats: 4 32 bpp, 2 24 bpp, 3 16 bpp, 1 10 bpc
FOURCC_ARGB = FOURCC('A', 'R', 'G', 'B'),
FOURCC_BGRA = FOURCC('B', 'G', 'R', 'A'),
FOURCC_ABGR = FOURCC('A', 'B', 'G', 'R'),
FOURCC_AR30 = FOURCC('A', 'R', '3', '0'), // 10 bit per channel. 2101010.
FOURCC_AB30 = FOURCC('A', 'B', '3', '0'), // ABGR version of 10 bit
FOURCC_24BG = FOURCC('2', '4', 'B', 'G'),
FOURCC_RAW = FOURCC('r', 'a', 'w', ' '),
FOURCC_RAW = FOURCC('r', 'a', 'w', ' '),
FOURCC_RGBA = FOURCC('R', 'G', 'B', 'A'),
FOURCC_RGBP = FOURCC('R', 'G', 'B', 'P'), // rgb565 LE.
FOURCC_RGBO = FOURCC('R', 'G', 'B', 'O'), // argb1555 LE.
FOURCC_R444 = FOURCC('R', '4', '4', '4'), // argb4444 LE.
// 4 Secondary RGB formats: 4 Bayer Patterns. deprecated.
FOURCC_RGGB = FOURCC('R', 'G', 'G', 'B'),
FOURCC_BGGR = FOURCC('B', 'G', 'G', 'R'),
FOURCC_GRBG = FOURCC('G', 'R', 'B', 'G'),
FOURCC_GBRG = FOURCC('G', 'B', 'R', 'G'),
// 1 Primary Compressed YUV format.
FOURCC_MJPG = FOURCC('M', 'J', 'P', 'G'),
// 5 Auxiliary YUV variations: 3 with U and V planes are swapped, 1 Alias.
// 8 Auxiliary YUV variations: 3 with U and V planes are swapped, 1 Alias.
FOURCC_YV12 = FOURCC('Y', 'V', '1', '2'),
FOURCC_YV16 = FOURCC('Y', 'V', '1', '6'),
FOURCC_YV24 = FOURCC('Y', 'V', '2', '4'),
@@ -92,6 +88,7 @@ enum FourCC {
FOURCC_J420 = FOURCC('J', '4', '2', '0'),
FOURCC_J400 = FOURCC('J', '4', '0', '0'), // unofficial fourcc
FOURCC_H420 = FOURCC('H', '4', '2', '0'), // unofficial fourcc
FOURCC_H422 = FOURCC('H', '4', '2', '2'), // unofficial fourcc
// 14 Auxiliary aliases. CanonicalFourCC() maps these to canonical fourcc.
FOURCC_IYUV = FOURCC('I', 'Y', 'U', 'V'), // Alias for I420.
@@ -112,7 +109,13 @@ enum FourCC {
FOURCC_L565 = FOURCC('L', '5', '6', '5'), // Alias for RGBP.
FOURCC_5551 = FOURCC('5', '5', '5', '1'), // Alias for RGBO.
// 1 Auxiliary compressed YUV format set aside for capturer.
// deprecated formats. Not supported, but defined for backward compatibility.
FOURCC_I411 = FOURCC('I', '4', '1', '1'),
FOURCC_Q420 = FOURCC('Q', '4', '2', '0'),
FOURCC_RGGB = FOURCC('R', 'G', 'G', 'B'),
FOURCC_BGGR = FOURCC('B', 'G', 'G', 'R'),
FOURCC_GRBG = FOURCC('G', 'R', 'B', 'G'),
FOURCC_GBRG = FOURCC('G', 'B', 'R', 'G'),
FOURCC_H264 = FOURCC('H', '2', '6', '4'),
// Match any fourcc.
@@ -136,8 +139,10 @@ enum FourCCBpp {
FOURCC_BPP_BGRA = 32,
FOURCC_BPP_ABGR = 32,
FOURCC_BPP_RGBA = 32,
FOURCC_BPP_AR30 = 32,
FOURCC_BPP_AB30 = 32,
FOURCC_BPP_24BG = 24,
FOURCC_BPP_RAW = 24,
FOURCC_BPP_RAW = 24,
FOURCC_BPP_RGBP = 16,
FOURCC_BPP_RGBO = 16,
FOURCC_BPP_R444 = 16,
@@ -152,6 +157,8 @@ enum FourCCBpp {
FOURCC_BPP_J420 = 12,
FOURCC_BPP_J400 = 8,
FOURCC_BPP_H420 = 12,
FOURCC_BPP_H422 = 16,
FOURCC_BPP_H010 = 24,
FOURCC_BPP_MJPG = 0, // 0 means unknown.
FOURCC_BPP_H264 = 0,
FOURCC_BPP_IYUV = 12,
@@ -170,11 +177,11 @@ enum FourCCBpp {
FOURCC_BPP_CM24 = 24,
// Match any fourcc.
FOURCC_BPP_ANY = 0, // 0 means unknown.
FOURCC_BPP_ANY = 0, // 0 means unknown.
};
// Converts fourcc aliases into canonical ones.
LIBYUV_API uint32 CanonicalFourCC(uint32 fourcc);
LIBYUV_API uint32_t CanonicalFourCC(uint32_t fourcc);
#ifdef __cplusplus
} // extern "C"
libs/libyuv/infra/config/OWNERS
+3
View File
@@ -0,0 +1,3 @@
set noparent
agable@chromium.org
phoglund@chromium.org
libs/libyuv/all.gyp → libs/libyuv/infra/config/PRESUBMIT.py
+15 -21
View File
@@ -1,21 +1,15 @@
# Copyright 2013 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
# all.gyp and All target are for benefit of android gyp build.
{
'targets': [
{
'target_name': 'All',
'type': 'none',
'dependencies': [
'libyuv.gyp:*',
'libyuv_test.gyp:*',
],
},
],
}
# Copyright 2018 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
def CheckChangeOnUpload(input_api, output_api):
return input_api.canned_checks.CheckChangedLUCIConfigs(input_api, output_api)
def CheckChangeOnCommit(input_api, output_api):
return input_api.canned_checks.CheckChangedLUCIConfigs(input_api, output_api)
libs/libyuv/infra/config/README.md
+1
View File
@@ -0,0 +1 @@
This directory contains configuration files for infra services.
libs/libyuv/infra/config/cq.cfg
+50
View File
@@ -0,0 +1,50 @@
# Commit Queue configuration file. The documentation of the format can be found
# at http://luci-config.appspot.com/schemas/projects/refs:cq.cfg.
version: 1
cq_status_url: "https://chromium-cq-status.appspot.com"
git_repo_url: "https://chromium.googlesource.com/libyuv/libyuv.git"
gerrit {}
verifiers {
gerrit_cq_ability {
committer_list: "project-libyuv-committers"
dry_run_access_list: "project-libyuv-tryjob-access"
}
try_job {
buckets {
name: "luci.libyuv.try"
builders { name: "win" }
builders { name: "win_rel" }
builders { name: "win_x64_rel" }
builders { name: "win_clang" }
builders { name: "win_clang_rel" }
builders { name: "win_x64_clang_rel" }
builders { name: "mac" }
builders { name: "mac_rel" }
builders { name: "mac_asan" }
builders { name: "ios" }
builders { name: "ios_rel" }
builders { name: "ios_arm64" }
builders { name: "ios_arm64_rel" }
builders { name: "linux" }
builders { name: "linux_rel" }
builders {
name: "linux_gcc"
experiment_percentage: 100
}
builders { name: "linux_tsan2" }
builders { name: "linux_asan" }
builders { name: "linux_msan" }
builders { name: "linux_ubsan" }
builders { name: "linux_ubsan_vptr" }
builders { name: "android" }
builders { name: "android_rel" }
builders { name: "android_arm64" }
builders { name: "android_x86" }
builders { name: "android_x64" }
}
}
}
libs/libyuv/libyuv.gni
+7 -4
View File
@@ -13,8 +13,11 @@ import("//build/config/mips.gni")
declare_args() {
libyuv_include_tests = !build_with_chromium
libyuv_disable_jpeg = false
libyuv_use_neon = (current_cpu == "arm64" ||
(current_cpu == "arm" && (arm_use_neon || arm_optionally_use_neon)))
libyuv_use_msa = (current_cpu == "mips64el" || current_cpu == "mipsel") &&
mips_use_msa
libyuv_use_neon =
current_cpu == "arm64" ||
(current_cpu == "arm" && (arm_use_neon || arm_optionally_use_neon))
libyuv_use_msa =
(current_cpu == "mips64el" || current_cpu == "mipsel") && mips_use_msa
libyuv_use_mmi =
(current_cpu == "mips64el" || current_cpu == "mipsel") && mips_use_mmi
}
libs/libyuv/libyuv.gyp
-161
View File
@@ -1,161 +0,0 @@
# Copyright 2011 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
{
'includes': [
'libyuv.gypi',
],
# Make sure that if we are being compiled to an xcodeproj, nothing tries to
# include a .pch.
'xcode_settings': {
'GCC_PREFIX_HEADER': '',
'GCC_PRECOMPILE_PREFIX_HEADER': 'NO',
},
'variables': {
'use_system_libjpeg%': 0,
'libyuv_disable_jpeg%': 0,
# 'chromium_code' treats libyuv as internal and increases warning level.
'chromium_code': 1,
# clang compiler default variable usable by other apps that include libyuv.
'clang%': 0,
# Link-Time Optimizations.
'use_lto%': 0,
'mips_msa%': 0, # Default to msa off.
'build_neon': 0,
'build_msa': 0,
'conditions': [
['(target_arch == "armv7" or target_arch == "armv7s" or \
(target_arch == "arm" and arm_version >= 7) or target_arch == "arm64")\
and (arm_neon == 1 or arm_neon_optional == 1)', {
'build_neon': 1,
}],
['(target_arch == "mipsel" or target_arch == "mips64el")\
and (mips_msa == 1)',
{
'build_msa': 1,
}],
],
},
'targets': [
{
'target_name': 'libyuv',
# Change type to 'shared_library' to build .so or .dll files.
'type': 'static_library',
'variables': {
'optimize': 'max', # enable O2 and ltcg.
},
# Allows libyuv.a redistributable library without external dependencies.
'standalone_static_library': 1,
'conditions': [
# Disable -Wunused-parameter
['clang == 1', {
'cflags': [
'-Wno-unused-parameter',
],
}],
['build_neon != 0', {
'defines': [
'LIBYUV_NEON',
],
'cflags!': [
'-mfpu=vfp',
'-mfpu=vfpv3',
'-mfpu=vfpv3-d16',
# '-mthumb', # arm32 not thumb
],
'conditions': [
# Disable LTO in libyuv_neon target due to gcc 4.9 compiler bug.
['clang == 0 and use_lto == 1', {
'cflags!': [
'-flto',
'-ffat-lto-objects',
],
}],
# arm64 does not need -mfpu=neon option as neon is not optional
['target_arch != "arm64"', {
'cflags': [
'-mfpu=neon',
# '-marm', # arm32 not thumb
],
}],
],
}],
['build_msa != 0', {
'defines': [
'LIBYUV_MSA',
],
}],
['OS != "ios" and libyuv_disable_jpeg != 1', {
'defines': [
'HAVE_JPEG'
],
'conditions': [
# Caveat system jpeg support may not support motion jpeg
[ 'use_system_libjpeg == 1', {
'dependencies': [
'<(DEPTH)/third_party/libjpeg/libjpeg.gyp:libjpeg',
],
}, {
'dependencies': [
'<(DEPTH)/third_party/libjpeg_turbo/libjpeg.gyp:libjpeg',
],
}],
[ 'use_system_libjpeg == 1', {
'link_settings': {
'libraries': [
'-ljpeg',
],
}
}],
],
}],
], #conditions
'defines': [
# Enable the following 3 macros to turn off assembly for specified CPU.
# 'LIBYUV_DISABLE_X86',
# 'LIBYUV_DISABLE_NEON',
# 'LIBYUV_DISABLE_MIPS',
# Enable the following macro to build libyuv as a shared library (dll).
# 'LIBYUV_USING_SHARED_LIBRARY',
# TODO(fbarchard): Make these into gyp defines.
],
'include_dirs': [
'include',
'.',
],
'direct_dependent_settings': {
'include_dirs': [
'include',
'.',
],
'conditions': [
['OS == "android" and target_arch == "arm64"', {
'ldflags': [
'-Wl,--dynamic-linker,/system/bin/linker64',
],
}],
['OS == "android" and target_arch != "arm64"', {
'ldflags': [
'-Wl,--dynamic-linker,/system/bin/linker',
],
}],
], #conditions
},
'sources': [
'<@(libyuv_sources)',
],
},
], # targets.
}
# Local Variables:
# tab-width:2
# indent-tabs-mode:nil
# End:
# vim: set expandtab tabstop=2 shiftwidth=2:
libs/libyuv/libyuv.gypi
-81
View File
@@ -1,81 +0,0 @@
# Copyright 2014 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
{
'variables': {
'libyuv_sources': [
# includes.
'include/libyuv.h',
'include/libyuv/basic_types.h',
'include/libyuv/compare.h',
'include/libyuv/convert.h',
'include/libyuv/convert_argb.h',
'include/libyuv/convert_from.h',
'include/libyuv/convert_from_argb.h',
'include/libyuv/cpu_id.h',
'include/libyuv/macros_msa.h',
'include/libyuv/mjpeg_decoder.h',
'include/libyuv/planar_functions.h',
'include/libyuv/rotate.h',
'include/libyuv/rotate_argb.h',
'include/libyuv/rotate_row.h',
'include/libyuv/row.h',
'include/libyuv/scale.h',
'include/libyuv/scale_argb.h',
'include/libyuv/scale_row.h',
'include/libyuv/version.h',
'include/libyuv/video_common.h',
# sources.
'source/compare.cc',
'source/compare_common.cc',
'source/compare_gcc.cc',
'source/compare_neon.cc',
'source/compare_neon64.cc',
'source/compare_win.cc',
'source/convert.cc',
'source/convert_argb.cc',
'source/convert_from.cc',
'source/convert_from_argb.cc',
'source/convert_jpeg.cc',
'source/convert_to_argb.cc',
'source/convert_to_i420.cc',
'source/cpu_id.cc',
'source/mjpeg_decoder.cc',
'source/mjpeg_validate.cc',
'source/planar_functions.cc',
'source/rotate.cc',
'source/rotate_any.cc',
'source/rotate_argb.cc',
'source/rotate_common.cc',
'source/rotate_gcc.cc',
'source/rotate_mips.cc',
'source/rotate_neon.cc',
'source/rotate_neon64.cc',
'source/rotate_win.cc',
'source/row_any.cc',
'source/row_common.cc',
'source/row_gcc.cc',
'source/row_mips.cc',
'source/row_msa.cc',
'source/row_neon.cc',
'source/row_neon64.cc',
'source/row_win.cc',
'source/scale.cc',
'source/scale_any.cc',
'source/scale_argb.cc',
'source/scale_common.cc',
'source/scale_gcc.cc',
'source/scale_mips.cc',
'source/scale_neon.cc',
'source/scale_neon64.cc',
'source/scale_win.cc',
'source/video_common.cc',
],
}
}
libs/libyuv/libyuv_nacl.gyp
-37
View File
@@ -1,37 +0,0 @@
# Copyright 2014 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
{
'includes': [
'libyuv.gypi',
'../../native_client/build/untrusted.gypi',
],
'targets': [
{
'target_name': 'libyuv_nacl',
'type': 'none',
'variables': {
'nlib_target': 'libyuv_nacl.a',
'build_glibc': 0,
'build_newlib': 0,
'build_pnacl_newlib': 1,
},
'include_dirs': [
'include',
],
'direct_dependent_settings': {
'include_dirs': [
'include',
],
},
'sources': [
'<@(libyuv_sources)',
],
}, # target libyuv_nacl
]
}
libs/libyuv/libyuv_test.gyp
-202
View File
@@ -1,202 +0,0 @@
# Copyright 2011 The LibYuv Project Authors. All rights reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
{
'variables': {
'libyuv_disable_jpeg%': 0,
'mips_msa%': 0, # Default to msa off.
},
'targets': [
{
'target_name': 'libyuv_unittest',
'type': '<(gtest_target_type)',
'dependencies': [
'libyuv.gyp:libyuv',
'testing/gtest.gyp:gtest',
'third_party/gflags/gflags.gyp:gflags',
],
'direct_dependent_settings': {
'defines': [
'GTEST_RELATIVE_PATH',
],
},
'export_dependent_settings': [
'<(DEPTH)/testing/gtest.gyp:gtest',
],
'sources': [
# headers
'unit_test/unit_test.h',
# sources
'unit_test/basictypes_test.cc',
'unit_test/compare_test.cc',
'unit_test/color_test.cc',
'unit_test/convert_test.cc',
'unit_test/cpu_test.cc',
'unit_test/math_test.cc',
'unit_test/planar_test.cc',
'unit_test/rotate_argb_test.cc',
'unit_test/rotate_test.cc',
'unit_test/scale_argb_test.cc',
'unit_test/scale_test.cc',
'unit_test/unit_test.cc',
'unit_test/video_common_test.cc',
],
'conditions': [
['OS=="linux"', {
'cflags': [
'-fexceptions',
],
}],
[ 'OS == "ios"', {
'xcode_settings': {
'DEBUGGING_SYMBOLS': 'YES',
'DEBUG_INFORMATION_FORMAT' : 'dwarf-with-dsym',
# Work around compile issue with isosim.mm, see
# https://code.google.com/p/libyuv/issues/detail?id=548 for details.
'WARNING_CFLAGS': [
'-Wno-sometimes-uninitialized',
],
},
'cflags': [
'-Wno-sometimes-uninitialized',
],
}],
[ 'OS != "ios" and libyuv_disable_jpeg != 1', {
'defines': [
'HAVE_JPEG',
],
}],
['OS=="android"', {
'dependencies': [
'<(DEPTH)/testing/android/native_test.gyp:native_test_native_code',
],
}],
# TODO(YangZhang): These lines can be removed when high accuracy
# YUV to RGB to Neon is ported.
[ '(target_arch == "armv7" or target_arch == "armv7s" \
or (target_arch == "arm" and arm_version >= 7) \
or target_arch == "arm64") \
and (arm_neon == 1 or arm_neon_optional == 1)', {
'defines': [
'LIBYUV_NEON'
],
}],
[ '(target_arch == "mipsel" or target_arch == "mips64el") \
and (mips_msa == 1)', {
'defines': [
'LIBYUV_MSA'
],
}],
], # conditions
'defines': [
# Enable the following 3 macros to turn off assembly for specified CPU.
# 'LIBYUV_DISABLE_X86',
# 'LIBYUV_DISABLE_NEON',
# 'LIBYUV_DISABLE_MIPS',
# Enable the following macro to build libyuv as a shared library (dll).
# 'LIBYUV_USING_SHARED_LIBRARY',
],
},
{
'target_name': 'compare',
'type': 'executable',
'dependencies': [
'libyuv.gyp:libyuv',
],
'sources': [
# sources
'util/compare.cc',
],
'conditions': [
['OS=="linux"', {
'cflags': [
'-fexceptions',
],
}],
], # conditions
},
{
'target_name': 'convert',
'type': 'executable',
'dependencies': [
'libyuv.gyp:libyuv',
],
'sources': [
# sources
'util/convert.cc',
],
'conditions': [
['OS=="linux"', {
'cflags': [
'-fexceptions',
],
}],
], # conditions
},
# TODO(fbarchard): Enable SSE2 and OpenMP for better performance.
{
'target_name': 'psnr',
'type': 'executable',
'sources': [
# sources
'util/psnr_main.cc',
'util/psnr.cc',
'util/ssim.cc',
],
'dependencies': [
'libyuv.gyp:libyuv',
],
'conditions': [
[ 'OS != "ios" and libyuv_disable_jpeg != 1', {
'defines': [
'HAVE_JPEG',
],
}],
], # conditions
},
{
'target_name': 'cpuid',
'type': 'executable',
'sources': [
# sources
'util/cpuid.c',
],
'dependencies': [
'libyuv.gyp:libyuv',
],
},
], # targets
'conditions': [
['OS=="android"', {
'targets': [
{
'target_name': 'yuv_unittest_apk',
'type': 'none',
'variables': {
'test_suite_name': 'yuv_unittest',
'input_shlib_path': '<(SHARED_LIB_DIR)/(SHARED_LIB_PREFIX)libyuv_unittest<(SHARED_LIB_SUFFIX)',
},
'includes': [
'build/apk_test.gypi',
],
'dependencies': [
'libyuv_unittest',
],
},
],
}],
],
}
# Local Variables:
# tab-width:2
# indent-tabs-mode:nil
# End:
# vim: set expandtab tabstop=2 shiftwidth=2:
libs/libyuv/linux.mk
+12 -7
View File
@@ -13,6 +13,8 @@ LOCAL_OBJ_FILES := \
source/compare.o \
source/compare_common.o \
source/compare_gcc.o \
source/compare_mmi.o \
source/compare_msa.o \
source/compare_neon64.o \
source/compare_neon.o \
source/compare_win.o \
@@ -32,14 +34,16 @@ LOCAL_OBJ_FILES := \
source/rotate.o \
source/rotate_common.o \
source/rotate_gcc.o \
source/rotate_mips.o \
source/rotate_mmi.o \
source/rotate_msa.o \
source/rotate_neon64.o \
source/rotate_neon.o \
source/rotate_win.o \
source/row_any.o \
source/row_common.o \
source/row_gcc.o \
source/row_mips.o \
source/row_mmi.o \
source/row_msa.o \
source/row_neon64.o \
source/row_neon.o \
source/row_win.o \
@@ -48,7 +52,8 @@ LOCAL_OBJ_FILES := \
source/scale.o \
source/scale_common.o \
source/scale_gcc.o \
source/scale_mips.o \
source/scale_mmi.o \
source/scale_msa.o \
source/scale_neon64.o \
source/scale_neon.o \
source/scale_win.o \
@@ -60,14 +65,14 @@ LOCAL_OBJ_FILES := \
.c.o:
$(CC) -c $(CFLAGS) $*.c -o $*.o
all: libyuv.a convert cpuid psnr
all: libyuv.a yuvconvert cpuid psnr
libyuv.a: $(LOCAL_OBJ_FILES)
$(AR) $(ARFLAGS) $@ $(LOCAL_OBJ_FILES)
# A C++ test utility that uses libyuv conversion.
convert: util/convert.cc libyuv.a
$(CXX) $(CXXFLAGS) -Iutil/ -o $@ util/convert.cc libyuv.a
yuvconvert: util/yuvconvert.cc libyuv.a
$(CXX) $(CXXFLAGS) -Iutil/ -o $@ util/yuvconvert.cc libyuv.a
# A standalone test utility
psnr: util/psnr.cc
@@ -80,4 +85,4 @@ cpuid: util/cpuid.c libyuv.a
$(CC) $(CFLAGS) -o $@ util/cpuid.c libyuv.a
clean:
/bin/rm -f source/*.o *.ii *.s libyuv.a convert cpuid psnr
/bin/rm -f source/*.o *.ii *.s libyuv.a yuvconvert cpuid psnr
libs/libyuv/public.mk
+13 -13
View File
@@ -1,13 +1,13 @@
# This file contains all the common make variables which are useful for
# anyone depending on this library.
# Note that dependencies on NDK are not directly listed since NDK auto adds
# them.
LIBYUV_INCLUDES := $(LIBYUV_PATH)/include
LIBYUV_C_FLAGS :=
LIBYUV_CPP_FLAGS :=
LIBYUV_LDLIBS :=
LIBYUV_DEP_MODULES :=
# This file contains all the common make variables which are useful for
# anyone depending on this library.
# Note that dependencies on NDK are not directly listed since NDK auto adds
# them.
LIBYUV_INCLUDES := $(LIBYUV_PATH)/include
LIBYUV_C_FLAGS :=
LIBYUV_CPP_FLAGS :=
LIBYUV_LDLIBS :=
LIBYUV_DEP_MODULES :=
libs/libyuv/pylintrc
+17
View File
@@ -0,0 +1,17 @@
[MESSAGES CONTROL]
# Disable the message, report, category or checker with the given id(s).
# TODO(kjellander): Reduce this list to as small as possible.
disable=I0010,I0011,bad-continuation,broad-except,duplicate-code,eval-used,exec-used,fixme,invalid-name,missing-docstring,no-init,no-member,too-few-public-methods,too-many-ancestors,too-many-arguments,too-many-branches,too-many-function-args,too-many-instance-attributes,too-many-lines,too-many-locals,too-many-public-methods,too-many-return-statements,too-many-statements
[REPORTS]
# Don't write out full reports, just messages.
reports=no
[FORMAT]
# We use two spaces for indents, instead of the usual four spaces or tab.
indent-string=' '
libs/libyuv/setup_links.py
-499
View File
@@ -1,499 +0,0 @@
#!/usr/bin/env python
# Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
#
# Use of this source code is governed by a BSD-style license
# that can be found in the LICENSE file in the root of the source
# tree. An additional intellectual property rights grant can be found
# in the file PATENTS. All contributing project authors may
# be found in the AUTHORS file in the root of the source tree.
"""Setup links to a Chromium checkout for WebRTC.
WebRTC standalone shares a lot of dependencies and build tools with Chromium.
To do this, many of the paths of a Chromium checkout is emulated by creating
symlinks to files and directories. This script handles the setup of symlinks to
achieve this.
"""
import ctypes
import errno
import logging
import optparse
import os
import shelve
import shutil
import subprocess
import sys
import textwrap
DIRECTORIES = [
'build',
'buildtools',
'mojo', # TODO(kjellander): Remove, see webrtc:5629.
'native_client',
'net',
'testing',
'third_party/binutils',
'third_party/drmemory',
'third_party/instrumented_libraries',
'third_party/libjpeg',
'third_party/libjpeg_turbo',
'third_party/llvm-build',
'third_party/lss',
'third_party/proguard',
'third_party/tcmalloc',
'third_party/yasm',
'third_party/WebKit', # TODO(kjellander): Remove, see webrtc:5629.
'tools/clang',
'tools/gn',
'tools/gyp',
'tools/memory',
'tools/python',
'tools/swarming_client',
'tools/valgrind',
'tools/vim',
'tools/win',
]
from sync_chromium import get_target_os_list
target_os = get_target_os_list()
if 'android' in target_os:
DIRECTORIES += [
'base',
'third_party/accessibility_test_framework',
'third_party/android_platform',
'third_party/android_tools',
'third_party/apache_velocity',
'third_party/appurify-python',
'third_party/ashmem',
'third_party/bouncycastle',
'third_party/catapult',
'third_party/ced',
'third_party/closure_compiler',
'third_party/guava',
'third_party/hamcrest',
'third_party/icu',
'third_party/icu4j',
'third_party/ijar',
'third_party/intellij',
'third_party/jsr-305',
'third_party/junit',
'third_party/libxml',
'third_party/mockito',
'third_party/modp_b64',
'third_party/ow2_asm',
'third_party/protobuf',
'third_party/requests',
'third_party/robolectric',
'third_party/sqlite4java',
'third_party/zlib',
'tools/android',
'tools/grit',
]
if 'ios' in target_os:
DIRECTORIES.append('third_party/class-dump')
FILES = {
'tools/isolate_driver.py': None,
'third_party/BUILD.gn': None,
}
ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
CHROMIUM_CHECKOUT = os.path.join('chromium', 'src')
LINKS_DB = 'links'
# Version management to make future upgrades/downgrades easier to support.
SCHEMA_VERSION = 1
def query_yes_no(question, default=False):
"""Ask a yes/no question via raw_input() and return their answer.
Modified from http://stackoverflow.com/a/3041990.
"""
prompt = " [%s/%%s]: "
prompt = prompt % ('Y' if default is True else 'y')
prompt = prompt % ('N' if default is False else 'n')
if default is None:
default = 'INVALID'
while True:
sys.stdout.write(question + prompt)
choice = raw_input().lower()
if choice == '' and default != 'INVALID':
return default
if 'yes'.startswith(choice):
return True
elif 'no'.startswith(choice):
return False
print "Please respond with 'yes' or 'no' (or 'y' or 'n')."
# Actions
class Action(object):
def __init__(self, dangerous):
self.dangerous = dangerous
def announce(self, planning):
"""Log a description of this action.
Args:
planning - True iff we're in the planning stage, False if we're in the
doit stage.
"""
pass
def doit(self, links_db):
"""Execute the action, recording what we did to links_db, if necessary."""
pass
class Remove(Action):
def __init__(self, path, dangerous):
super(Remove, self).__init__(dangerous)
self._priority = 0
self._path = path
def announce(self, planning):
log = logging.warn
filesystem_type = 'file'
if not self.dangerous:
log = logging.info
filesystem_type = 'link'
if planning:
log('Planning to remove %s: %s', filesystem_type, self._path)
else:
log('Removing %s: %s', filesystem_type, self._path)
def doit(self, _):
os.remove(self._path)
class Rmtree(Action):
def __init__(self, path):
super(Rmtree, self).__init__(dangerous=True)
self._priority = 0
self._path = path
def announce(self, planning):
if planning:
logging.warn('Planning to remove directory: %s', self._path)
else:
logging.warn('Removing directory: %s', self._path)
def doit(self, _):
if sys.platform.startswith('win'):
# shutil.rmtree() doesn't work on Windows if any of the directories are
# read-only.
subprocess.check_call(['rd', '/q', '/s', self._path], shell=True)
else:
shutil.rmtree(self._path)
class Makedirs(Action):
def __init__(self, path):
super(Makedirs, self).__init__(dangerous=False)
self._priority = 1
self._path = path
def doit(self, _):
try:
os.makedirs(self._path)
except OSError as e:
if e.errno != errno.EEXIST:
raise
class Symlink(Action):
def __init__(self, source_path, link_path):
super(Symlink, self).__init__(dangerous=False)
self._priority = 2
self._source_path = source_path
self._link_path = link_path
def announce(self, planning):
if planning:
logging.info(
'Planning to create link from %s to %s', self._link_path,
self._source_path)
else:
logging.debug(
'Linking from %s to %s', self._link_path, self._source_path)
def doit(self, links_db):
# Files not in the root directory need relative path calculation.
# On Windows, use absolute paths instead since NTFS doesn't seem to support
# relative paths for symlinks.
if sys.platform.startswith('win'):
source_path = os.path.abspath(self._source_path)
else:
if os.path.dirname(self._link_path) != self._link_path:
source_path = os.path.relpath(self._source_path,
os.path.dirname(self._link_path))
os.symlink(source_path, os.path.abspath(self._link_path))
links_db[self._source_path] = self._link_path
class LinkError(IOError):
"""Failed to create a link."""
pass
# Use junctions instead of symlinks on the Windows platform.
if sys.platform.startswith('win'):
def symlink(source_path, link_path):
if os.path.isdir(source_path):
subprocess.check_call(['cmd.exe', '/c', 'mklink', '/J', link_path,
source_path])
else:
# Don't create symlinks to files on Windows, just copy the file instead
# (there's no way to create a link without administrator's privileges).
shutil.copy(source_path, link_path)
os.symlink = symlink
class WebRTCLinkSetup(object):
def __init__(self, links_db, force=False, dry_run=False, prompt=False):
self._force = force
self._dry_run = dry_run
self._prompt = prompt
self._links_db = links_db
def CreateLinks(self, on_bot):
logging.debug('CreateLinks')
# First, make a plan of action
actions = []
for source_path, link_path in FILES.iteritems():
actions += self._ActionForPath(
source_path, link_path, check_fn=os.path.isfile, check_msg='files')
for source_dir in DIRECTORIES:
actions += self._ActionForPath(
source_dir, None, check_fn=os.path.isdir,
check_msg='directories')
if not on_bot and self._force:
# When making the manual switch from legacy SVN checkouts to the new
# Git-based Chromium DEPS, the .gclient_entries file that contains cached
# URLs for all DEPS entries must be removed to avoid future sync problems.
entries_file = os.path.join(os.path.dirname(ROOT_DIR), '.gclient_entries')
if os.path.exists(entries_file):
actions.append(Remove(entries_file, dangerous=True))
actions.sort()
if self._dry_run:
for action in actions:
action.announce(planning=True)
logging.info('Not doing anything because dry-run was specified.')
sys.exit(0)
if any(a.dangerous for a in actions):
logging.warn('Dangerous actions:')
for action in (a for a in actions if a.dangerous):
action.announce(planning=True)
print
if not self._force:
logging.error(textwrap.dedent("""\
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
A C T I O N R E Q I R E D
@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
Setting up the checkout requires creating symlinks to directories in the
Chromium checkout inside chromium/src.
To avoid disrupting developers, we've chosen to not delete directories
forcibly, in case you have some work in progress in one of them :)
ACTION REQUIRED:
Before running `gclient sync|runhooks` again, you must run:
%s%s --force
Which will replace all directories which now must be symlinks, after
prompting with a summary of the work-to-be-done.
"""), 'python ' if sys.platform.startswith('win') else '', __file__)
sys.exit(1)
elif self._prompt:
if not query_yes_no('Would you like to perform the above plan?'):
sys.exit(1)
for action in actions:
action.announce(planning=False)
action.doit(self._links_db)
if not on_bot and self._force:
logging.info('Completed!\n\nNow run `gclient sync|runhooks` again to '
'let the remaining hooks (that probably were interrupted) '
'execute.')
def CleanupLinks(self):
logging.debug('CleanupLinks')
for source, link_path in self._links_db.iteritems():
if source == 'SCHEMA_VERSION':
continue
if os.path.islink(link_path) or sys.platform.startswith('win'):
# os.path.islink() always returns false on Windows
# See http://bugs.python.org/issue13143.
logging.debug('Removing link to %s at %s', source, link_path)
if not self._dry_run:
if os.path.exists(link_path):
if sys.platform.startswith('win') and os.path.isdir(link_path):
subprocess.check_call(['rmdir', '/q', '/s', link_path],
shell=True)
else:
os.remove(link_path)
del self._links_db[source]
@staticmethod
def _ActionForPath(source_path, link_path=None, check_fn=None,
check_msg=None):
"""Create zero or more Actions to link to a file or directory.
This will be a symlink on POSIX platforms. On Windows it will result in:
* a junction for directories
* a copied file for single files.
Args:
source_path: Path relative to the Chromium checkout root.
For readability, the path may contain slashes, which will
automatically be converted to the right path delimiter on Windows.
link_path: The location for the link to create. If omitted it will be the
same path as source_path.
check_fn: A function returning true if the type of filesystem object is
correct for the attempted call. Otherwise an error message with
check_msg will be printed.
check_msg: String used to inform the user of an invalid attempt to create
a file.
Returns:
A list of Action objects.
"""
def fix_separators(path):
if sys.platform.startswith('win'):
return path.replace(os.altsep, os.sep)
else:
return path
assert check_fn
assert check_msg
link_path = link_path or source_path
link_path = fix_separators(link_path)
source_path = fix_separators(source_path)
source_path = os.path.join(CHROMIUM_CHECKOUT, source_path)
if os.path.exists(source_path) and not check_fn:
raise LinkError('Can only to link to %s: tried to link to: %s' %
(check_msg, source_path))
if not os.path.exists(source_path):
logging.debug('Silently ignoring missing source: %s. This is to avoid '
'errors on platform-specific dependencies.', source_path)
return []
actions = []
if os.path.exists(link_path) or os.path.islink(link_path):
if os.path.islink(link_path):
actions.append(Remove(link_path, dangerous=False))
elif os.path.isfile(link_path):
actions.append(Remove(link_path, dangerous=True))
elif os.path.isdir(link_path):
actions.append(Rmtree(link_path))
else:
raise LinkError('Don\'t know how to plan: %s' % link_path)
# Create parent directories to the target link if needed.
target_parent_dirs = os.path.dirname(link_path)
if (target_parent_dirs and
target_parent_dirs != link_path and
not os.path.exists(target_parent_dirs)):
actions.append(Makedirs(target_parent_dirs))
actions.append(Symlink(source_path, link_path))
return actions
def _initialize_database(filename):
links_database = shelve.open(filename)
# Wipe the database if this version of the script ends up looking at a
# newer (future) version of the links db, just to be sure.
version = links_database.get('SCHEMA_VERSION')
if version and version != SCHEMA_VERSION:
logging.info('Found database with schema version %s while this script only '
'supports %s. Wiping previous database contents.', version,
SCHEMA_VERSION)
links_database.clear()
links_database['SCHEMA_VERSION'] = SCHEMA_VERSION
return links_database
def main():
on_bot = os.environ.get('CHROME_HEADLESS') == '1'
parser = optparse.OptionParser()
parser.add_option('-d', '--dry-run', action='store_true', default=False,
help='Print what would be done, but don\'t perform any '
'operations. This will automatically set logging to '
'verbose.')
parser.add_option('-c', '--clean-only', action='store_true', default=False,
help='Only clean previously created links, don\'t create '
'new ones. This will automatically set logging to '
'verbose.')
parser.add_option('-f', '--force', action='store_true', default=on_bot,
help='Force link creation. CAUTION: This deletes existing '
'folders and files in the locations where links are '
'about to be created.')
parser.add_option('-n', '--no-prompt', action='store_false', dest='prompt',
default=(not on_bot),
help='Prompt if we\'re planning to do a dangerous action')
parser.add_option('-v', '--verbose', action='store_const',
const=logging.DEBUG, default=logging.INFO,
help='Print verbose output for debugging.')
options, _ = parser.parse_args()
if options.dry_run or options.force or options.clean_only:
options.verbose = logging.DEBUG
logging.basicConfig(format='%(message)s', level=options.verbose)
# Work from the root directory of the checkout.
script_dir = os.path.dirname(os.path.abspath(__file__))
os.chdir(script_dir)
if sys.platform.startswith('win'):
def is_admin():
try:
return os.getuid() == 0
except AttributeError:
return ctypes.windll.shell32.IsUserAnAdmin() != 0
if is_admin():
logging.warning('WARNING: On Windows, you no longer need run as '
'administrator. Please run with user account privileges.')
if not os.path.exists(CHROMIUM_CHECKOUT):
logging.error('Cannot find a Chromium checkout at %s. Did you run "gclient '
'sync" before running this script?', CHROMIUM_CHECKOUT)
return 2
links_database = _initialize_database(LINKS_DB)
try:
symlink_creator = WebRTCLinkSetup(links_database, options.force,
options.dry_run, options.prompt)
symlink_creator.CleanupLinks()
if not options.clean_only:
symlink_creator.CreateLinks(on_bot)
except LinkError as e:
print >> sys.stderr, e.message
return 3
finally:
links_database.close()
return 0
if __name__ == '__main__':
sys.exit(main())
libs/libyuv/source/compare.cc
+189 -89
View File
@@ -29,10 +29,10 @@ extern "C" {
// hash seed of 5381 recommended.
LIBYUV_API
uint32 HashDjb2(const uint8* src, uint64 count, uint32 seed) {
uint32_t HashDjb2(const uint8_t* src, uint64_t count, uint32_t seed) {
const int kBlockSize = 1 << 15; // 32768;
int remainder;
uint32 (*HashDjb2_SSE)(const uint8* src, int count, uint32 seed) =
uint32_t (*HashDjb2_SSE)(const uint8_t* src, int count, uint32_t seed) =
HashDjb2_C;
#if defined(HAS_HASHDJB2_SSE41)
if (TestCpuFlag(kCpuHasSSE41)) {
@@ -45,25 +45,25 @@ uint32 HashDjb2(const uint8* src, uint64 count, uint32 seed) {
}
#endif
while (count >= (uint64)(kBlockSize)) {
while (count >= (uint64_t)(kBlockSize)) {
seed = HashDjb2_SSE(src, kBlockSize, seed);
src += kBlockSize;
count -= kBlockSize;
}
remainder = (int)(count) & ~15;
remainder = (int)count & ~15;
if (remainder) {
seed = HashDjb2_SSE(src, remainder, seed);
src += remainder;
count -= remainder;
}
remainder = (int)(count) & 15;
remainder = (int)count & 15;
if (remainder) {
seed = HashDjb2_C(src, remainder, seed);
}
return seed;
}
static uint32 ARGBDetectRow_C(const uint8* argb, int width) {
static uint32_t ARGBDetectRow_C(const uint8_t* argb, int width) {
int x;
for (x = 0; x < width - 1; x += 2) {
if (argb[0] != 255) { // First byte is not Alpha of 255, so not ARGB.
@@ -94,8 +94,11 @@ static uint32 ARGBDetectRow_C(const uint8* argb, int width) {
// Scan an opaque argb image and return fourcc based on alpha offset.
// Returns FOURCC_ARGB, FOURCC_BGRA, or 0 if unknown.
LIBYUV_API
uint32 ARGBDetect(const uint8* argb, int stride_argb, int width, int height) {
uint32 fourcc = 0;
uint32_t ARGBDetect(const uint8_t* argb,
int stride_argb,
int width,
int height) {
uint32_t fourcc = 0;
int h;
// Coalesce rows.
@@ -111,19 +114,86 @@ uint32 ARGBDetect(const uint8* argb, int stride_argb, int width, int height) {
return fourcc;
}
// NEON version accumulates in 16 bit shorts which overflow at 65536 bytes.
// So actual maximum is 1 less loop, which is 64436 - 32 bytes.
LIBYUV_API
uint64_t ComputeHammingDistance(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
const int kBlockSize = 1 << 15; // 32768;
const int kSimdSize = 64;
// SIMD for multiple of 64, and C for remainder
int remainder = count & (kBlockSize - 1) & ~(kSimdSize - 1);
uint64_t diff = 0;
int i;
uint32_t (*HammingDistance)(const uint8_t* src_a, const uint8_t* src_b,
int count) = HammingDistance_C;
#if defined(HAS_HAMMINGDISTANCE_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
HammingDistance = HammingDistance_NEON;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
HammingDistance = HammingDistance_SSSE3;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_SSE42)
if (TestCpuFlag(kCpuHasSSE42)) {
HammingDistance = HammingDistance_SSE42;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
HammingDistance = HammingDistance_AVX2;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
HammingDistance = HammingDistance_MSA;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
HammingDistance = HammingDistance_MMI;
}
#endif
#ifdef _OPENMP
#pragma omp parallel for reduction(+ : diff)
#endif
for (i = 0; i < (count - (kBlockSize - 1)); i += kBlockSize) {
diff += HammingDistance(src_a + i, src_b + i, kBlockSize);
}
src_a += count & ~(kBlockSize - 1);
src_b += count & ~(kBlockSize - 1);
if (remainder) {
diff += HammingDistance(src_a, src_b, remainder);
src_a += remainder;
src_b += remainder;
}
remainder = count & (kSimdSize - 1);
if (remainder) {
diff += HammingDistance_C(src_a, src_b, remainder);
}
return diff;
}
// TODO(fbarchard): Refactor into row function.
LIBYUV_API
uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
int count) {
uint64_t ComputeSumSquareError(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
// SumSquareError returns values 0 to 65535 for each squared difference.
// Up to 65536 of those can be summed and remain within a uint32.
// After each block of 65536 pixels, accumulate into a uint64.
// Up to 65536 of those can be summed and remain within a uint32_t.
// After each block of 65536 pixels, accumulate into a uint64_t.
const int kBlockSize = 65536;
int remainder = count & (kBlockSize - 1) & ~31;
uint64 sse = 0;
uint64_t sse = 0;
int i;
uint32 (*SumSquareError)(const uint8* src_a, const uint8* src_b, int count) =
SumSquareError_C;
uint32_t (*SumSquareError)(const uint8_t* src_a, const uint8_t* src_b,
int count) = SumSquareError_C;
#if defined(HAS_SUMSQUAREERROR_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
SumSquareError = SumSquareError_NEON;
@@ -141,8 +211,18 @@ uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
SumSquareError = SumSquareError_AVX2;
}
#endif
#if defined(HAS_SUMSQUAREERROR_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
SumSquareError = SumSquareError_MSA;
}
#endif
#if defined(HAS_SUMSQUAREERROR_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
SumSquareError = SumSquareError_MMI;
}
#endif
#ifdef _OPENMP
#pragma omp parallel for reduction(+: sse)
#pragma omp parallel for reduction(+ : sse)
#endif
for (i = 0; i < (count - (kBlockSize - 1)); i += kBlockSize) {
sse += SumSquareError(src_a + i, src_b + i, kBlockSize);
@@ -162,14 +242,16 @@ uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
}
LIBYUV_API
uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height) {
uint64 sse = 0;
uint64_t ComputeSumSquareErrorPlane(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height) {
uint64_t sse = 0;
int h;
// Coalesce rows.
if (stride_a == width &&
stride_b == width) {
if (stride_a == width && stride_b == width) {
width *= height;
height = 1;
stride_a = stride_b = 0;
@@ -183,66 +265,76 @@ uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
}
LIBYUV_API
double SumSquareErrorToPsnr(uint64 sse, uint64 count) {
double SumSquareErrorToPsnr(uint64_t sse, uint64_t count) {
double psnr;
if (sse > 0) {
double mse = (double)(count) / (double)(sse);
double mse = (double)count / (double)sse;
psnr = 10.0 * log10(255.0 * 255.0 * mse);
} else {
psnr = kMaxPsnr; // Limit to prevent divide by 0
psnr = kMaxPsnr; // Limit to prevent divide by 0
}
if (psnr > kMaxPsnr)
if (psnr > kMaxPsnr) {
psnr = kMaxPsnr;
}
return psnr;
}
LIBYUV_API
double CalcFramePsnr(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height) {
const uint64 samples = width * height;
const uint64 sse = ComputeSumSquareErrorPlane(src_a, stride_a,
src_b, stride_b,
width, height);
double CalcFramePsnr(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height) {
const uint64_t samples = (uint64_t)width * (uint64_t)height;
const uint64_t sse = ComputeSumSquareErrorPlane(src_a, stride_a, src_b,
stride_b, width, height);
return SumSquareErrorToPsnr(sse, samples);
}
LIBYUV_API
double I420Psnr(const uint8* src_y_a, int stride_y_a,
const uint8* src_u_a, int stride_u_a,
const uint8* src_v_a, int stride_v_a,
const uint8* src_y_b, int stride_y_b,
const uint8* src_u_b, int stride_u_b,
const uint8* src_v_b, int stride_v_b,
int width, int height) {
const uint64 sse_y = ComputeSumSquareErrorPlane(src_y_a, stride_y_a,
src_y_b, stride_y_b,
width, height);
double I420Psnr(const uint8_t* src_y_a,
int stride_y_a,
const uint8_t* src_u_a,
int stride_u_a,
const uint8_t* src_v_a,
int stride_v_a,
const uint8_t* src_y_b,
int stride_y_b,
const uint8_t* src_u_b,
int stride_u_b,
const uint8_t* src_v_b,
int stride_v_b,
int width,
int height) {
const uint64_t sse_y = ComputeSumSquareErrorPlane(
src_y_a, stride_y_a, src_y_b, stride_y_b, width, height);
const int width_uv = (width + 1) >> 1;
const int height_uv = (height + 1) >> 1;
const uint64 sse_u = ComputeSumSquareErrorPlane(src_u_a, stride_u_a,
src_u_b, stride_u_b,
width_uv, height_uv);
const uint64 sse_v = ComputeSumSquareErrorPlane(src_v_a, stride_v_a,
src_v_b, stride_v_b,
width_uv, height_uv);
const uint64 samples = width * height + 2 * (width_uv * height_uv);
const uint64 sse = sse_y + sse_u + sse_v;
const uint64_t sse_u = ComputeSumSquareErrorPlane(
src_u_a, stride_u_a, src_u_b, stride_u_b, width_uv, height_uv);
const uint64_t sse_v = ComputeSumSquareErrorPlane(
src_v_a, stride_v_a, src_v_b, stride_v_b, width_uv, height_uv);
const uint64_t samples = (uint64_t)width * (uint64_t)height +
2 * ((uint64_t)width_uv * (uint64_t)height_uv);
const uint64_t sse = sse_y + sse_u + sse_v;
return SumSquareErrorToPsnr(sse, samples);
}
static const int64 cc1 = 26634; // (64^2*(.01*255)^2
static const int64 cc2 = 239708; // (64^2*(.03*255)^2
static const int64_t cc1 = 26634; // (64^2*(.01*255)^2
static const int64_t cc2 = 239708; // (64^2*(.03*255)^2
static double Ssim8x8_C(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b) {
int64 sum_a = 0;
int64 sum_b = 0;
int64 sum_sq_a = 0;
int64 sum_sq_b = 0;
int64 sum_axb = 0;
static double Ssim8x8_C(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b) {
int64_t sum_a = 0;
int64_t sum_b = 0;
int64_t sum_sq_a = 0;
int64_t sum_sq_b = 0;
int64_t sum_axb = 0;
int i;
for (i = 0; i < 8; ++i) {
@@ -260,22 +352,22 @@ static double Ssim8x8_C(const uint8* src_a, int stride_a,
}
{
const int64 count = 64;
const int64_t count = 64;
// scale the constants by number of pixels
const int64 c1 = (cc1 * count * count) >> 12;
const int64 c2 = (cc2 * count * count) >> 12;
const int64_t c1 = (cc1 * count * count) >> 12;
const int64_t c2 = (cc2 * count * count) >> 12;
const int64 sum_a_x_sum_b = sum_a * sum_b;
const int64_t sum_a_x_sum_b = sum_a * sum_b;
const int64 ssim_n = (2 * sum_a_x_sum_b + c1) *
(2 * count * sum_axb - 2 * sum_a_x_sum_b + c2);
const int64_t ssim_n = (2 * sum_a_x_sum_b + c1) *
(2 * count * sum_axb - 2 * sum_a_x_sum_b + c2);
const int64 sum_a_sq = sum_a*sum_a;
const int64 sum_b_sq = sum_b*sum_b;
const int64_t sum_a_sq = sum_a * sum_a;
const int64_t sum_b_sq = sum_b * sum_b;
const int64 ssim_d = (sum_a_sq + sum_b_sq + c1) *
(count * sum_sq_a - sum_a_sq +
count * sum_sq_b - sum_b_sq + c2);
const int64_t ssim_d =
(sum_a_sq + sum_b_sq + c1) *
(count * sum_sq_a - sum_a_sq + count * sum_sq_b - sum_b_sq + c2);
if (ssim_d == 0.0) {
return DBL_MAX;
@@ -288,13 +380,16 @@ static double Ssim8x8_C(const uint8* src_a, int stride_a,
// on the 4x4 pixel grid. Such arrangement allows the windows to overlap
// block boundaries to penalize blocking artifacts.
LIBYUV_API
double CalcFrameSsim(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height) {
double CalcFrameSsim(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height) {
int samples = 0;
double ssim_total = 0;
double (*Ssim8x8)(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b) = Ssim8x8_C;
double (*Ssim8x8)(const uint8_t* src_a, int stride_a, const uint8_t* src_b,
int stride_b) = Ssim8x8_C;
// sample point start with each 4x4 location
int i;
@@ -314,22 +409,27 @@ double CalcFrameSsim(const uint8* src_a, int stride_a,
}
LIBYUV_API
double I420Ssim(const uint8* src_y_a, int stride_y_a,
const uint8* src_u_a, int stride_u_a,
const uint8* src_v_a, int stride_v_a,
const uint8* src_y_b, int stride_y_b,
const uint8* src_u_b, int stride_u_b,
const uint8* src_v_b, int stride_v_b,
int width, int height) {
const double ssim_y = CalcFrameSsim(src_y_a, stride_y_a,
src_y_b, stride_y_b, width, height);
double I420Ssim(const uint8_t* src_y_a,
int stride_y_a,
const uint8_t* src_u_a,
int stride_u_a,
const uint8_t* src_v_a,
int stride_v_a,
const uint8_t* src_y_b,
int stride_y_b,
const uint8_t* src_u_b,
int stride_u_b,
const uint8_t* src_v_b,
int stride_v_b,
int width,
int height) {
const double ssim_y =
CalcFrameSsim(src_y_a, stride_y_a, src_y_b, stride_y_b, width, height);
const int width_uv = (width + 1) >> 1;
const int height_uv = (height + 1) >> 1;
const double ssim_u = CalcFrameSsim(src_u_a, stride_u_a,
src_u_b, stride_u_b,
const double ssim_u = CalcFrameSsim(src_u_a, stride_u_a, src_u_b, stride_u_b,
width_uv, height_uv);
const double ssim_v = CalcFrameSsim(src_v_a, stride_v_a,
src_v_b, stride_v_b,
const double ssim_v = CalcFrameSsim(src_v_a, stride_v_a, src_v_b, stride_v_b,
width_uv, height_uv);
return ssim_y * 0.8 + 0.1 * (ssim_u + ssim_v);
}
libs/libyuv/source/compare_common.cc
+65 -5
View File
@@ -17,20 +17,80 @@ namespace libyuv {
extern "C" {
#endif
uint32 SumSquareError_C(const uint8* src_a, const uint8* src_b, int count) {
uint32 sse = 0u;
#if ORIGINAL_OPT
uint32_t HammingDistance_C1(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
int i;
for (i = 0; i < count; ++i) {
int x = src_a[i] ^ src_b[i];
if (x & 1)
++diff;
if (x & 2)
++diff;
if (x & 4)
++diff;
if (x & 8)
++diff;
if (x & 16)
++diff;
if (x & 32)
++diff;
if (x & 64)
++diff;
if (x & 128)
++diff;
}
return diff;
}
#endif
// Hakmem method for hamming distance.
uint32_t HammingDistance_C(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
int i;
for (i = 0; i < count - 3; i += 4) {
uint32_t x = *((const uint32_t*)src_a) ^ *((const uint32_t*)src_b);
uint32_t u = x - ((x >> 1) & 0x55555555);
u = ((u >> 2) & 0x33333333) + (u & 0x33333333);
diff += ((((u + (u >> 4)) & 0x0f0f0f0f) * 0x01010101) >> 24);
src_a += 4;
src_b += 4;
}
for (; i < count; ++i) {
uint32_t x = *src_a ^ *src_b;
uint32_t u = x - ((x >> 1) & 0x55);
u = ((u >> 2) & 0x33) + (u & 0x33);
diff += (u + (u >> 4)) & 0x0f;
src_a += 1;
src_b += 1;
}
return diff;
}
uint32_t SumSquareError_C(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse = 0u;
int i;
for (i = 0; i < count; ++i) {
int diff = src_a[i] - src_b[i];
sse += (uint32)(diff * diff);
sse += (uint32_t)(diff * diff);
}
return sse;
}
// hash seed of 5381 recommended.
// Internal C version of HashDjb2 with int sized count for efficiency.
uint32 HashDjb2_C(const uint8* src, int count, uint32 seed) {
uint32 hash = seed;
uint32_t HashDjb2_C(const uint8_t* src, int count, uint32_t seed) {
uint32_t hash = seed;
int i;
for (i = 0; i < count; ++i) {
hash += (hash << 5) + src[i];
libs/libyuv/source/compare_gcc.cc
+318 -109
View File
@@ -22,124 +22,334 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
uint32 sse;
asm volatile (
"pxor %%xmm0,%%xmm0 \n"
"pxor %%xmm5,%%xmm5 \n"
LABELALIGN
"1: \n"
"movdqu " MEMACCESS(0) ",%%xmm1 \n"
"lea " MEMLEA(0x10, 0) ",%0 \n"
"movdqu " MEMACCESS(1) ",%%xmm2 \n"
"lea " MEMLEA(0x10, 1) ",%1 \n"
"movdqa %%xmm1,%%xmm3 \n"
"psubusb %%xmm2,%%xmm1 \n"
"psubusb %%xmm3,%%xmm2 \n"
"por %%xmm2,%%xmm1 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklbw %%xmm5,%%xmm1 \n"
"punpckhbw %%xmm5,%%xmm2 \n"
"pmaddwd %%xmm1,%%xmm1 \n"
"pmaddwd %%xmm2,%%xmm2 \n"
"paddd %%xmm1,%%xmm0 \n"
"paddd %%xmm2,%%xmm0 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
#if defined(__x86_64__)
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint64_t diff = 0u;
"pshufd $0xee,%%xmm0,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"pshufd $0x1,%%xmm0,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"movd %%xmm0,%3 \n"
asm volatile(
"xor %3,%3 \n"
"xor %%r8,%%r8 \n"
"xor %%r9,%%r9 \n"
"xor %%r10,%%r10 \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=g"(sse) // %3
:: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
);
// Process 32 bytes per loop.
LABELALIGN
"1: \n"
"mov (%0),%%rcx \n"
"mov 0x8(%0),%%rdx \n"
"xor (%1),%%rcx \n"
"xor 0x8(%1),%%rdx \n"
"popcnt %%rcx,%%rcx \n"
"popcnt %%rdx,%%rdx \n"
"mov 0x10(%0),%%rsi \n"
"mov 0x18(%0),%%rdi \n"
"xor 0x10(%1),%%rsi \n"
"xor 0x18(%1),%%rdi \n"
"popcnt %%rsi,%%rsi \n"
"popcnt %%rdi,%%rdi \n"
"add $0x20,%0 \n"
"add $0x20,%1 \n"
"add %%rcx,%3 \n"
"add %%rdx,%%r8 \n"
"add %%rsi,%%r9 \n"
"add %%rdi,%%r10 \n"
"sub $0x20,%2 \n"
"jg 1b \n"
"add %%r8, %3 \n"
"add %%r9, %3 \n"
"add %%r10, %3 \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=r"(diff) // %3
:
: "memory", "cc", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10");
return static_cast<uint32_t>(diff);
}
#else
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
asm volatile(
// Process 16 bytes per loop.
LABELALIGN
"1: \n"
"mov (%0),%%ecx \n"
"mov 0x4(%0),%%edx \n"
"xor (%1),%%ecx \n"
"xor 0x4(%1),%%edx \n"
"popcnt %%ecx,%%ecx \n"
"add %%ecx,%3 \n"
"popcnt %%edx,%%edx \n"
"add %%edx,%3 \n"
"mov 0x8(%0),%%ecx \n"
"mov 0xc(%0),%%edx \n"
"xor 0x8(%1),%%ecx \n"
"xor 0xc(%1),%%edx \n"
"popcnt %%ecx,%%ecx \n"
"add %%ecx,%3 \n"
"popcnt %%edx,%%edx \n"
"add %%edx,%3 \n"
"add $0x10,%0 \n"
"add $0x10,%1 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"+r"(diff) // %3
:
: "memory", "cc", "ecx", "edx");
return diff;
}
#endif
static const vec8 kNibbleMask = {15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15};
static const vec8 kBitCount = {0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
uint32_t HammingDistance_SSSE3(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
asm volatile(
"movdqa %4,%%xmm2 \n"
"movdqa %5,%%xmm3 \n"
"pxor %%xmm0,%%xmm0 \n"
"pxor %%xmm1,%%xmm1 \n"
"sub %0,%1 \n"
LABELALIGN
"1: \n"
"movdqa (%0),%%xmm4 \n"
"movdqa 0x10(%0), %%xmm5 \n"
"pxor (%0,%1), %%xmm4 \n"
"movdqa %%xmm4,%%xmm6 \n"
"pand %%xmm2,%%xmm6 \n"
"psrlw $0x4,%%xmm4 \n"
"movdqa %%xmm3,%%xmm7 \n"
"pshufb %%xmm6,%%xmm7 \n"
"pand %%xmm2,%%xmm4 \n"
"movdqa %%xmm3,%%xmm6 \n"
"pshufb %%xmm4,%%xmm6 \n"
"paddb %%xmm7,%%xmm6 \n"
"pxor 0x10(%0,%1),%%xmm5 \n"
"add $0x20,%0 \n"
"movdqa %%xmm5,%%xmm4 \n"
"pand %%xmm2,%%xmm5 \n"
"psrlw $0x4,%%xmm4 \n"
"movdqa %%xmm3,%%xmm7 \n"
"pshufb %%xmm5,%%xmm7 \n"
"pand %%xmm2,%%xmm4 \n"
"movdqa %%xmm3,%%xmm5 \n"
"pshufb %%xmm4,%%xmm5 \n"
"paddb %%xmm7,%%xmm5 \n"
"paddb %%xmm5,%%xmm6 \n"
"psadbw %%xmm1,%%xmm6 \n"
"paddd %%xmm6,%%xmm0 \n"
"sub $0x20,%2 \n"
"jg 1b \n"
"pshufd $0xaa,%%xmm0,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"movd %%xmm0, %3 \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=r"(diff) // %3
: "m"(kNibbleMask), // %4
"m"(kBitCount) // %5
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
return diff;
}
#ifdef HAS_HAMMINGDISTANCE_AVX2
uint32_t HammingDistance_AVX2(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
asm volatile(
"vbroadcastf128 %4,%%ymm2 \n"
"vbroadcastf128 %5,%%ymm3 \n"
"vpxor %%ymm0,%%ymm0,%%ymm0 \n"
"vpxor %%ymm1,%%ymm1,%%ymm1 \n"
"sub %0,%1 \n"
LABELALIGN
"1: \n"
"vmovdqa (%0),%%ymm4 \n"
"vmovdqa 0x20(%0), %%ymm5 \n"
"vpxor (%0,%1), %%ymm4, %%ymm4 \n"
"vpand %%ymm2,%%ymm4,%%ymm6 \n"
"vpsrlw $0x4,%%ymm4,%%ymm4 \n"
"vpshufb %%ymm6,%%ymm3,%%ymm6 \n"
"vpand %%ymm2,%%ymm4,%%ymm4 \n"
"vpshufb %%ymm4,%%ymm3,%%ymm4 \n"
"vpaddb %%ymm4,%%ymm6,%%ymm6 \n"
"vpxor 0x20(%0,%1),%%ymm5,%%ymm4 \n"
"add $0x40,%0 \n"
"vpand %%ymm2,%%ymm4,%%ymm5 \n"
"vpsrlw $0x4,%%ymm4,%%ymm4 \n"
"vpshufb %%ymm5,%%ymm3,%%ymm5 \n"
"vpand %%ymm2,%%ymm4,%%ymm4 \n"
"vpshufb %%ymm4,%%ymm3,%%ymm4 \n"
"vpaddb %%ymm5,%%ymm4,%%ymm4 \n"
"vpaddb %%ymm6,%%ymm4,%%ymm4 \n"
"vpsadbw %%ymm1,%%ymm4,%%ymm4 \n"
"vpaddd %%ymm0,%%ymm4,%%ymm0 \n"
"sub $0x40,%2 \n"
"jg 1b \n"
"vpermq $0xb1,%%ymm0,%%ymm1 \n"
"vpaddd %%ymm1,%%ymm0,%%ymm0 \n"
"vpermq $0xaa,%%ymm0,%%ymm1 \n"
"vpaddd %%ymm1,%%ymm0,%%ymm0 \n"
"vmovd %%xmm0, %3 \n"
"vzeroupper \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=r"(diff) // %3
: "m"(kNibbleMask), // %4
"m"(kBitCount) // %5
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6");
return diff;
}
#endif // HAS_HAMMINGDISTANCE_AVX2
uint32_t SumSquareError_SSE2(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse;
asm volatile(
"pxor %%xmm0,%%xmm0 \n"
"pxor %%xmm5,%%xmm5 \n"
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm1 \n"
"lea 0x10(%0),%0 \n"
"movdqu (%1),%%xmm2 \n"
"lea 0x10(%1),%1 \n"
"movdqa %%xmm1,%%xmm3 \n"
"psubusb %%xmm2,%%xmm1 \n"
"psubusb %%xmm3,%%xmm2 \n"
"por %%xmm2,%%xmm1 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklbw %%xmm5,%%xmm1 \n"
"punpckhbw %%xmm5,%%xmm2 \n"
"pmaddwd %%xmm1,%%xmm1 \n"
"pmaddwd %%xmm2,%%xmm2 \n"
"paddd %%xmm1,%%xmm0 \n"
"paddd %%xmm2,%%xmm0 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
"pshufd $0xee,%%xmm0,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"pshufd $0x1,%%xmm0,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"movd %%xmm0,%3 \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=g"(sse) // %3
::"memory",
"cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5");
return sse;
}
static uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16
static uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
static const uvec32 kHash16x33 = {0x92d9e201, 0, 0, 0}; // 33 ^ 16
static const uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
};
static uvec32 kHashMul1 = {
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
static const uvec32 kHashMul1 = {
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
};
static uvec32 kHashMul2 = {
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
static const uvec32 kHashMul2 = {
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
};
static uvec32 kHashMul3 = {
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
static const uvec32 kHashMul3 = {
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
};
uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
uint32 hash;
asm volatile (
"movd %2,%%xmm0 \n"
"pxor %%xmm7,%%xmm7 \n"
"movdqa %4,%%xmm6 \n"
LABELALIGN
"1: \n"
"movdqu " MEMACCESS(0) ",%%xmm1 \n"
"lea " MEMLEA(0x10, 0) ",%0 \n"
"pmulld %%xmm6,%%xmm0 \n"
"movdqa %5,%%xmm5 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklbw %%xmm7,%%xmm2 \n"
"movdqa %%xmm2,%%xmm3 \n"
"punpcklwd %%xmm7,%%xmm3 \n"
"pmulld %%xmm5,%%xmm3 \n"
"movdqa %6,%%xmm5 \n"
"movdqa %%xmm2,%%xmm4 \n"
"punpckhwd %%xmm7,%%xmm4 \n"
"pmulld %%xmm5,%%xmm4 \n"
"movdqa %7,%%xmm5 \n"
"punpckhbw %%xmm7,%%xmm1 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklwd %%xmm7,%%xmm2 \n"
"pmulld %%xmm5,%%xmm2 \n"
"movdqa %8,%%xmm5 \n"
"punpckhwd %%xmm7,%%xmm1 \n"
"pmulld %%xmm5,%%xmm1 \n"
"paddd %%xmm4,%%xmm3 \n"
"paddd %%xmm2,%%xmm1 \n"
"paddd %%xmm3,%%xmm1 \n"
"pshufd $0xe,%%xmm1,%%xmm2 \n"
"paddd %%xmm2,%%xmm1 \n"
"pshufd $0x1,%%xmm1,%%xmm2 \n"
"paddd %%xmm2,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"sub $0x10,%1 \n"
"jg 1b \n"
"movd %%xmm0,%3 \n"
: "+r"(src), // %0
"+r"(count), // %1
"+rm"(seed), // %2
"=g"(hash) // %3
: "m"(kHash16x33), // %4
"m"(kHashMul0), // %5
"m"(kHashMul1), // %6
"m"(kHashMul2), // %7
"m"(kHashMul3) // %8
: "memory", "cc"
, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
);
uint32_t HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed) {
uint32_t hash;
asm volatile(
"movd %2,%%xmm0 \n"
"pxor %%xmm7,%%xmm7 \n"
"movdqa %4,%%xmm6 \n"
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm1 \n"
"lea 0x10(%0),%0 \n"
"pmulld %%xmm6,%%xmm0 \n"
"movdqa %5,%%xmm5 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklbw %%xmm7,%%xmm2 \n"
"movdqa %%xmm2,%%xmm3 \n"
"punpcklwd %%xmm7,%%xmm3 \n"
"pmulld %%xmm5,%%xmm3 \n"
"movdqa %6,%%xmm5 \n"
"movdqa %%xmm2,%%xmm4 \n"
"punpckhwd %%xmm7,%%xmm4 \n"
"pmulld %%xmm5,%%xmm4 \n"
"movdqa %7,%%xmm5 \n"
"punpckhbw %%xmm7,%%xmm1 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklwd %%xmm7,%%xmm2 \n"
"pmulld %%xmm5,%%xmm2 \n"
"movdqa %8,%%xmm5 \n"
"punpckhwd %%xmm7,%%xmm1 \n"
"pmulld %%xmm5,%%xmm1 \n"
"paddd %%xmm4,%%xmm3 \n"
"paddd %%xmm2,%%xmm1 \n"
"paddd %%xmm3,%%xmm1 \n"
"pshufd $0xe,%%xmm1,%%xmm2 \n"
"paddd %%xmm2,%%xmm1 \n"
"pshufd $0x1,%%xmm1,%%xmm2 \n"
"paddd %%xmm2,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"sub $0x10,%1 \n"
"jg 1b \n"
"movd %%xmm0,%3 \n"
: "+r"(src), // %0
"+r"(count), // %1
"+rm"(seed), // %2
"=g"(hash) // %3
: "m"(kHash16x33), // %4
"m"(kHashMul0), // %5
"m"(kHashMul1), // %6
"m"(kHashMul2), // %7
"m"(kHashMul3) // %8
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
return hash;
}
#endif // defined(__x86_64__) || (defined(__i386__) && !defined(__pic__)))
@@ -148,4 +358,3 @@ uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
} // extern "C"
} // namespace libyuv
#endif
libs/libyuv/source/compare_mmi.cc
+123
View File
@@ -0,0 +1,123 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for Mips MMI.
#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
// Hakmem method for hamming distance.
uint32_t HammingDistance_MMI(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
uint64_t temp = 0, temp1 = 0, ta = 0, tb = 0;
uint64_t c1 = 0x5555555555555555;
uint64_t c2 = 0x3333333333333333;
uint64_t c3 = 0x0f0f0f0f0f0f0f0f;
uint32_t c4 = 0x01010101;
uint64_t s1 = 1, s2 = 2, s3 = 4;
__asm__ volatile(
"1: \n\t"
"ldc1 %[ta], 0(%[src_a]) \n\t"
"ldc1 %[tb], 0(%[src_b]) \n\t"
"xor %[temp], %[ta], %[tb] \n\t"
"psrlw %[temp1], %[temp], %[s1] \n\t" // temp1=x>>1
"and %[temp1], %[temp1], %[c1] \n\t" // temp1&=c1
"psubw %[temp1], %[temp], %[temp1] \n\t" // x-temp1
"and %[temp], %[temp1], %[c2] \n\t" // t = (u&c2)
"psrlw %[temp1], %[temp1], %[s2] \n\t" // u>>2
"and %[temp1], %[temp1], %[c2] \n\t" // u>>2 & c2
"paddw %[temp1], %[temp1], %[temp] \n\t" // t1 = t1+t
"psrlw %[temp], %[temp1], %[s3] \n\t" // u>>4
"paddw %[temp1], %[temp1], %[temp] \n\t" // u+(u>>4)
"and %[temp1], %[temp1], %[c3] \n\t" //&c3
"dmfc1 $t0, %[temp1] \n\t"
"dsrl32 $t0, $t0, 0 \n\t "
"mul $t0, $t0, %[c4] \n\t"
"dsrl $t0, $t0, 24 \n\t"
"dadd %[diff], %[diff], $t0 \n\t"
"dmfc1 $t0, %[temp1] \n\t"
"mul $t0, $t0, %[c4] \n\t"
"dsrl $t0, $t0, 24 \n\t"
"dadd %[diff], %[diff], $t0 \n\t"
"daddiu %[src_a], %[src_a], 8 \n\t"
"daddiu %[src_b], %[src_b], 8 \n\t"
"addiu %[count], %[count], -8 \n\t"
"bgtz %[count], 1b \n\t"
"nop \n\t"
: [diff] "+r"(diff), [src_a] "+r"(src_a), [src_b] "+r"(src_b),
[count] "+r"(count), [ta] "+f"(ta), [tb] "+f"(tb), [temp] "+f"(temp),
[temp1] "+f"(temp1)
: [c1] "f"(c1), [c2] "f"(c2), [c3] "f"(c3), [c4] "r"(c4), [s1] "f"(s1),
[s2] "f"(s2), [s3] "f"(s3)
: "memory");
return diff;
}
uint32_t SumSquareError_MMI(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse = 0u;
uint32_t sse_hi = 0u, sse_lo = 0u;
uint64_t src1, src2;
uint64_t diff, diff_hi, diff_lo;
uint64_t sse_sum, sse_tmp;
const uint64_t mask = 0x0ULL;
__asm__ volatile(
"xor %[sse_sum], %[sse_sum], %[sse_sum] \n\t"
"1: \n\t"
"ldc1 %[src1], 0x00(%[src_a]) \n\t"
"ldc1 %[src2], 0x00(%[src_b]) \n\t"
"pasubub %[diff], %[src1], %[src2] \n\t"
"punpcklbh %[diff_lo], %[diff], %[mask] \n\t"
"punpckhbh %[diff_hi], %[diff], %[mask] \n\t"
"pmaddhw %[sse_tmp], %[diff_lo], %[diff_lo] \n\t"
"paddw %[sse_sum], %[sse_sum], %[sse_tmp] \n\t"
"pmaddhw %[sse_tmp], %[diff_hi], %[diff_hi] \n\t"
"paddw %[sse_sum], %[sse_sum], %[sse_tmp] \n\t"
"daddiu %[src_a], %[src_a], 0x08 \n\t"
"daddiu %[src_b], %[src_b], 0x08 \n\t"
"daddiu %[count], %[count], -0x08 \n\t"
"bnez %[count], 1b \n\t"
"mfc1 %[sse_lo], %[sse_sum] \n\t"
"mfhc1 %[sse_hi], %[sse_sum] \n\t"
"daddu %[sse], %[sse_hi], %[sse_lo] \n\t"
: [sse] "+&r"(sse), [diff] "=&f"(diff), [src1] "=&f"(src1),
[src2] "=&f"(src2), [diff_lo] "=&f"(diff_lo), [diff_hi] "=&f"(diff_hi),
[sse_sum] "=&f"(sse_sum), [sse_tmp] "=&f"(sse_tmp),
[sse_hi] "+&r"(sse_hi), [sse_lo] "+&r"(sse_lo)
: [src_a] "r"(src_a), [src_b] "r"(src_b), [count] "r"(count),
[mask] "f"(mask)
: "memory");
return sse;
}
#endif // !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
libs/libyuv/source/compare_msa.cc
+97
View File
@@ -0,0 +1,97 @@
/*
* Copyright 2017 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#include "libyuv/row.h"
// This module is for GCC MSA
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#include "libyuv/macros_msa.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
uint32_t HammingDistance_MSA(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
int i;
v16u8 src0, src1, src2, src3;
v2i64 vec0 = {0}, vec1 = {0};
for (i = 0; i < count; i += 32) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_a, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_a, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)src_b, 0);
src3 = (v16u8)__msa_ld_b((v16i8*)src_b, 16);
src0 ^= src2;
src1 ^= src3;
vec0 += __msa_pcnt_d((v2i64)src0);
vec1 += __msa_pcnt_d((v2i64)src1);
src_a += 32;
src_b += 32;
}
vec0 += vec1;
diff = (uint32_t)__msa_copy_u_w((v4i32)vec0, 0);
diff += (uint32_t)__msa_copy_u_w((v4i32)vec0, 2);
return diff;
}
uint32_t SumSquareError_MSA(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse = 0u;
int i;
v16u8 src0, src1, src2, src3;
v8i16 vec0, vec1, vec2, vec3;
v4i32 reg0 = {0}, reg1 = {0}, reg2 = {0}, reg3 = {0};
v2i64 tmp0;
for (i = 0; i < count; i += 32) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_a, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_a, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)src_b, 0);
src3 = (v16u8)__msa_ld_b((v16i8*)src_b, 16);
vec0 = (v8i16)__msa_ilvr_b((v16i8)src2, (v16i8)src0);
vec1 = (v8i16)__msa_ilvl_b((v16i8)src2, (v16i8)src0);
vec2 = (v8i16)__msa_ilvr_b((v16i8)src3, (v16i8)src1);
vec3 = (v8i16)__msa_ilvl_b((v16i8)src3, (v16i8)src1);
vec0 = __msa_hsub_u_h((v16u8)vec0, (v16u8)vec0);
vec1 = __msa_hsub_u_h((v16u8)vec1, (v16u8)vec1);
vec2 = __msa_hsub_u_h((v16u8)vec2, (v16u8)vec2);
vec3 = __msa_hsub_u_h((v16u8)vec3, (v16u8)vec3);
reg0 = __msa_dpadd_s_w(reg0, vec0, vec0);
reg1 = __msa_dpadd_s_w(reg1, vec1, vec1);
reg2 = __msa_dpadd_s_w(reg2, vec2, vec2);
reg3 = __msa_dpadd_s_w(reg3, vec3, vec3);
src_a += 32;
src_b += 32;
}
reg0 += reg1;
reg2 += reg3;
reg0 += reg2;
tmp0 = __msa_hadd_s_d(reg0, reg0);
sse = (uint32_t)__msa_copy_u_w((v4i32)tmp0, 0);
sse += (uint32_t)__msa_copy_u_w((v4i32)tmp0, 2);
return sse;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
libs/libyuv/source/compare_neon.cc
+62 -32
View File
@@ -21,40 +21,70 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__) && \
!defined(__aarch64__)
uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count) {
volatile uint32 sse;
asm volatile (
"vmov.u8 q8, #0 \n"
"vmov.u8 q10, #0 \n"
"vmov.u8 q9, #0 \n"
"vmov.u8 q11, #0 \n"
// 256 bits at a time
// uses short accumulator which restricts count to 131 KB
uint32_t HammingDistance_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff;
"1: \n"
MEMACCESS(0)
"vld1.8 {q0}, [%0]! \n"
MEMACCESS(1)
"vld1.8 {q1}, [%1]! \n"
"subs %2, %2, #16 \n"
"vsubl.u8 q2, d0, d2 \n"
"vsubl.u8 q3, d1, d3 \n"
"vmlal.s16 q8, d4, d4 \n"
"vmlal.s16 q9, d6, d6 \n"
"vmlal.s16 q10, d5, d5 \n"
"vmlal.s16 q11, d7, d7 \n"
"bgt 1b \n"
asm volatile(
"vmov.u16 q4, #0 \n" // accumulator
"vadd.u32 q8, q8, q9 \n"
"vadd.u32 q10, q10, q11 \n"
"vadd.u32 q11, q8, q10 \n"
"vpaddl.u32 q1, q11 \n"
"vadd.u64 d0, d2, d3 \n"
"vmov.32 %3, d0[0] \n"
: "+r"(src_a),
"+r"(src_b),
"+r"(count),
"=r"(sse)
:
: "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11");
"1: \n"
"vld1.8 {q0, q1}, [%0]! \n"
"vld1.8 {q2, q3}, [%1]! \n"
"veor.32 q0, q0, q2 \n"
"veor.32 q1, q1, q3 \n"
"vcnt.i8 q0, q0 \n"
"vcnt.i8 q1, q1 \n"
"subs %2, %2, #32 \n"
"vadd.u8 q0, q0, q1 \n" // 16 byte counts
"vpadal.u8 q4, q0 \n" // 8 shorts
"bgt 1b \n"
"vpaddl.u16 q0, q4 \n" // 4 ints
"vpadd.u32 d0, d0, d1 \n"
"vpadd.u32 d0, d0, d0 \n"
"vmov.32 %3, d0[0] \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(diff)
:
: "cc", "q0", "q1", "q2", "q3", "q4");
return diff;
}
uint32_t SumSquareError_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse;
asm volatile(
"vmov.u8 q8, #0 \n"
"vmov.u8 q10, #0 \n"
"vmov.u8 q9, #0 \n"
"vmov.u8 q11, #0 \n"
"1: \n"
"vld1.8 {q0}, [%0]! \n"
"vld1.8 {q1}, [%1]! \n"
"subs %2, %2, #16 \n"
"vsubl.u8 q2, d0, d2 \n"
"vsubl.u8 q3, d1, d3 \n"
"vmlal.s16 q8, d4, d4 \n"
"vmlal.s16 q9, d6, d6 \n"
"vmlal.s16 q10, d5, d5 \n"
"vmlal.s16 q11, d7, d7 \n"
"bgt 1b \n"
"vadd.u32 q8, q8, q9 \n"
"vadd.u32 q10, q10, q11 \n"
"vadd.u32 q11, q8, q10 \n"
"vpaddl.u32 q1, q11 \n"
"vadd.u64 d0, d2, d3 \n"
"vmov.32 %3, d0[0] \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(sse)
:
: "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11");
return sse;
}
libs/libyuv/source/compare_neon64.cc
+57 -31
View File
@@ -20,39 +20,65 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count) {
volatile uint32 sse;
asm volatile (
"eor v16.16b, v16.16b, v16.16b \n"
"eor v18.16b, v18.16b, v18.16b \n"
"eor v17.16b, v17.16b, v17.16b \n"
"eor v19.16b, v19.16b, v19.16b \n"
// 256 bits at a time
// uses short accumulator which restricts count to 131 KB
uint32_t HammingDistance_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff;
asm volatile(
"movi v4.8h, #0 \n"
"1: \n"
MEMACCESS(0)
"ld1 {v0.16b}, [%0], #16 \n"
MEMACCESS(1)
"ld1 {v1.16b}, [%1], #16 \n"
"subs %w2, %w2, #16 \n"
"usubl v2.8h, v0.8b, v1.8b \n"
"usubl2 v3.8h, v0.16b, v1.16b \n"
"smlal v16.4s, v2.4h, v2.4h \n"
"smlal v17.4s, v3.4h, v3.4h \n"
"smlal2 v18.4s, v2.8h, v2.8h \n"
"smlal2 v19.4s, v3.8h, v3.8h \n"
"b.gt 1b \n"
"1: \n"
"ld1 {v0.16b, v1.16b}, [%0], #32 \n"
"ld1 {v2.16b, v3.16b}, [%1], #32 \n"
"eor v0.16b, v0.16b, v2.16b \n"
"eor v1.16b, v1.16b, v3.16b \n"
"cnt v0.16b, v0.16b \n"
"cnt v1.16b, v1.16b \n"
"subs %w2, %w2, #32 \n"
"add v0.16b, v0.16b, v1.16b \n"
"uadalp v4.8h, v0.16b \n"
"b.gt 1b \n"
"add v16.4s, v16.4s, v17.4s \n"
"add v18.4s, v18.4s, v19.4s \n"
"add v19.4s, v16.4s, v18.4s \n"
"addv s0, v19.4s \n"
"fmov %w3, s0 \n"
: "+r"(src_a),
"+r"(src_b),
"+r"(count),
"=r"(sse)
:
: "cc", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19");
"uaddlv s4, v4.8h \n"
"fmov %w3, s4 \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(diff)
:
: "cc", "v0", "v1", "v2", "v3", "v4");
return diff;
}
uint32_t SumSquareError_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse;
asm volatile(
"eor v16.16b, v16.16b, v16.16b \n"
"eor v18.16b, v18.16b, v18.16b \n"
"eor v17.16b, v17.16b, v17.16b \n"
"eor v19.16b, v19.16b, v19.16b \n"
"1: \n"
"ld1 {v0.16b}, [%0], #16 \n"
"ld1 {v1.16b}, [%1], #16 \n"
"subs %w2, %w2, #16 \n"
"usubl v2.8h, v0.8b, v1.8b \n"
"usubl2 v3.8h, v0.16b, v1.16b \n"
"smlal v16.4s, v2.4h, v2.4h \n"
"smlal v17.4s, v3.4h, v3.4h \n"
"smlal2 v18.4s, v2.8h, v2.8h \n"
"smlal2 v19.4s, v3.8h, v3.8h \n"
"b.gt 1b \n"
"add v16.4s, v16.4s, v17.4s \n"
"add v18.4s, v18.4s, v19.4s \n"
"add v19.4s, v16.4s, v18.4s \n"
"addv s0, v19.4s \n"
"fmov %w3, s0 \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(sse)
:
: "cc", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19");
return sse;
}
libs/libyuv/source/compare_win.cc
+69 -50
View File
@@ -13,20 +13,39 @@
#include "libyuv/compare_row.h"
#include "libyuv/row.h"
#if defined(_MSC_VER)
#include <intrin.h> // For __popcnt
#endif
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for 32 bit Visual C x86 and clangcl
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
__declspec(naked)
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
int i;
for (i = 0; i < count - 3; i += 4) {
uint32_t x = *((uint32_t*)src_a) ^ *((uint32_t*)src_b); // NOLINT
src_a += 4;
src_b += 4;
diff += __popcnt(x);
}
return diff;
}
__declspec(naked) uint32_t
SumSquareError_SSE2(const uint8_t* src_a, const uint8_t* src_b, int count) {
__asm {
mov eax, [esp + 4] // src_a
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
mov eax, [esp + 4] // src_a
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
pxor xmm0, xmm0
pxor xmm5, xmm5
@@ -61,13 +80,13 @@ uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
// Visual C 2012 required for AVX2.
#if _MSC_VER >= 1700
// C4752: found Intel(R) Advanced Vector Extensions; consider using /arch:AVX.
#pragma warning(disable: 4752)
__declspec(naked)
uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
#pragma warning(disable : 4752)
__declspec(naked) uint32_t
SumSquareError_AVX2(const uint8_t* src_a, const uint8_t* src_b, int count) {
__asm {
mov eax, [esp + 4] // src_a
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
mov eax, [esp + 4] // src_a
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
vpxor ymm0, ymm0, ymm0 // sum
vpxor ymm5, ymm5, ymm5 // constant 0 for unpck
sub edx, eax
@@ -101,65 +120,65 @@ uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
}
#endif // _MSC_VER >= 1700
uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16
uvec32 kHash16x33 = {0x92d9e201, 0, 0, 0}; // 33 ^ 16
uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
};
uvec32 kHashMul1 = {
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
};
uvec32 kHashMul2 = {
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
};
uvec32 kHashMul3 = {
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
};
__declspec(naked)
uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
__declspec(naked) uint32_t
HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed) {
__asm {
mov eax, [esp + 4] // src
mov ecx, [esp + 8] // count
mov eax, [esp + 4] // src
mov ecx, [esp + 8] // count
movd xmm0, [esp + 12] // seed
pxor xmm7, xmm7 // constant 0 for unpck
pxor xmm7, xmm7 // constant 0 for unpck
movdqa xmm6, xmmword ptr kHash16x33
wloop:
movdqu xmm1, [eax] // src[0-15]
movdqu xmm1, [eax] // src[0-15]
lea eax, [eax + 16]
pmulld xmm0, xmm6 // hash *= 33 ^ 16
pmulld xmm0, xmm6 // hash *= 33 ^ 16
movdqa xmm5, xmmword ptr kHashMul0
movdqa xmm2, xmm1
punpcklbw xmm2, xmm7 // src[0-7]
punpcklbw xmm2, xmm7 // src[0-7]
movdqa xmm3, xmm2
punpcklwd xmm3, xmm7 // src[0-3]
punpcklwd xmm3, xmm7 // src[0-3]
pmulld xmm3, xmm5
movdqa xmm5, xmmword ptr kHashMul1
movdqa xmm4, xmm2
punpckhwd xmm4, xmm7 // src[4-7]
punpckhwd xmm4, xmm7 // src[4-7]
pmulld xmm4, xmm5
movdqa xmm5, xmmword ptr kHashMul2
punpckhbw xmm1, xmm7 // src[8-15]
punpckhbw xmm1, xmm7 // src[8-15]
movdqa xmm2, xmm1
punpcklwd xmm2, xmm7 // src[8-11]
punpcklwd xmm2, xmm7 // src[8-11]
pmulld xmm2, xmm5
movdqa xmm5, xmmword ptr kHashMul3
punpckhwd xmm1, xmm7 // src[12-15]
punpckhwd xmm1, xmm7 // src[12-15]
pmulld xmm1, xmm5
paddd xmm3, xmm4 // add 16 results
paddd xmm3, xmm4 // add 16 results
paddd xmm1, xmm2
paddd xmm1, xmm3
@@ -171,18 +190,18 @@ uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
sub ecx, 16
jg wloop
movd eax, xmm0 // return hash
movd eax, xmm0 // return hash
ret
}
}
// Visual C 2012 required for AVX2.
#if _MSC_VER >= 1700
__declspec(naked)
uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) {
__declspec(naked) uint32_t
HashDjb2_AVX2(const uint8_t* src, int count, uint32_t seed) {
__asm {
mov eax, [esp + 4] // src
mov ecx, [esp + 8] // count
mov eax, [esp + 4] // src
mov ecx, [esp + 8] // count
vmovd xmm0, [esp + 12] // seed
wloop:
@@ -196,7 +215,7 @@ uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) {
vpmulld xmm2, xmm2, xmmword ptr kHashMul2
lea eax, [eax + 16]
vpmulld xmm1, xmm1, xmmword ptr kHashMul3
vpaddd xmm3, xmm3, xmm4 // add 16 results
vpaddd xmm3, xmm3, xmm4 // add 16 results
vpaddd xmm1, xmm1, xmm2
vpaddd xmm1, xmm1, xmm3
vpshufd xmm2, xmm1, 0x0e // upper 2 dwords
@@ -207,7 +226,7 @@ uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) {
sub ecx, 16
jg wloop
vmovd eax, xmm0 // return hash
vmovd eax, xmm0 // return hash
vzeroupper
ret
}
libs/libyuv/source/convert.cc
+1070 -326
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/convert_argb.cc
+1367 -501
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/convert_from.cc
+819 -430
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/convert_from_argb.cc
+811 -254
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/convert_jpeg.cc
+241 -166
View File
@@ -22,28 +22,24 @@ extern "C" {
#ifdef HAVE_JPEG
struct I420Buffers {
uint8* y;
uint8_t* y;
int y_stride;
uint8* u;
uint8_t* u;
int u_stride;
uint8* v;
uint8_t* v;
int v_stride;
int w;
int h;
};
static void JpegCopyI420(void* opaque,
const uint8* const* data,
const uint8_t* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I420Copy(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
I420Copy(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
@@ -51,17 +47,13 @@ static void JpegCopyI420(void* opaque,
}
static void JpegI422ToI420(void* opaque,
const uint8* const* data,
const uint8_t* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I422ToI420(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
I422ToI420(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
@@ -69,35 +61,13 @@ static void JpegI422ToI420(void* opaque,
}
static void JpegI444ToI420(void* opaque,
const uint8* const* data,
const uint8_t* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I444ToI420(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
dest->h -= rows;
}
static void JpegI411ToI420(void* opaque,
const uint8* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I411ToI420(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
I444ToI420(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
@@ -105,15 +75,12 @@ static void JpegI411ToI420(void* opaque,
}
static void JpegI400ToI420(void* opaque,
const uint8* const* data,
const uint8_t* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I400ToI420(data[0], strides[0],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
I400ToI420(data[0], strides[0], dest->y, dest->y_stride, dest->u,
dest->u_stride, dest->v, dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
@@ -122,10 +89,12 @@ static void JpegI400ToI420(void* opaque,
// Query size of MJPG in pixels.
LIBYUV_API
int MJPGSize(const uint8* sample, size_t sample_size,
int* width, int* height) {
int MJPGSize(const uint8_t* src_mjpg,
size_t src_size_mjpg,
int* width,
int* height) {
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size);
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(src_mjpg, src_size_mjpg);
if (ret) {
*width = mjpeg_decoder.GetWidth();
*height = mjpeg_decoder.GetHeight();
@@ -135,34 +104,40 @@ int MJPGSize(const uint8* sample, size_t sample_size,
}
// 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* sample,
size_t sample_size,
uint8* y, int y_stride,
uint8* u, int u_stride,
uint8* v, int v_stride,
int w, int h,
int dw, int dh) {
if (sample_size == kUnknownDataSize) {
int MJPGToI420(const uint8_t* src_mjpg,
size_t src_size_mjpg,
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 (src_size_mjpg == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
}
// 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)) {
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(src_mjpg, src_size_mjpg);
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 &&
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
@@ -170,8 +145,9 @@ int MJPGToI420(const uint8* sample,
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegCopyI420, &bufs, dw, dh);
// YUV422
ret = mjpeg_decoder.DecodeToCallback(&JpegCopyI420, &bufs, dst_width,
dst_height);
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
@@ -181,8 +157,9 @@ int MJPGToI420(const uint8* sample,
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToI420, &bufs, dw, dh);
// YUV444
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToI420, &bufs, dst_width,
dst_height);
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
@@ -192,29 +169,158 @@ int MJPGToI420(const uint8* sample,
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToI420, &bufs, dw, dh);
// YUV411
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 4 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI411ToI420, &bufs, dw, dh);
// YUV400
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. 411 is supported by libjpeg
// ERROR: Unable to convert MJPEG frame because format is not supported
// TODO(fbarchard): Implement conversion for any other
// colorspace/subsample factors that occur in practice. ERROR: Unable to
// convert MJPEG frame because format is not supported
mjpeg_decoder.UnloadFrame();
return 1;
}
}
return ret ? 0 : 1;
}
struct NV21Buffers {
uint8_t* y;
int y_stride;
uint8_t* vu;
int vu_stride;
int w;
int h;
};
static void JpegI420ToNV21(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
I420ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI422ToNV21(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
I422ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI444ToNV21(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
I444ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI400ToNV21(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
I400ToNV21(data[0], strides[0], dest->y, dest->y_stride, dest->vu,
dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
// MJPG (Motion JPeg) to NV21
LIBYUV_API
int MJPGToNV21(const uint8_t* src_mjpg,
size_t src_size_mjpg,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int src_width,
int src_height,
int dst_width,
int dst_height) {
if (src_size_mjpg == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
}
// TODO(fbarchard): Port MJpeg to C.
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(src_mjpg, src_size_mjpg);
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) {
NV21Buffers bufs = {dst_y, dst_stride_y, dst_vu,
dst_stride_vu, dst_width, dst_height};
// YUV420
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToNV21, &bufs, dst_width,
dst_height);
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToNV21, &bufs, dst_width,
dst_height);
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToNV21, &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(&JpegI400ToNV21, &bufs, dst_width,
dst_height);
} else {
// Unknown colorspace.
mjpeg_decoder.UnloadFrame();
return 1;
}
@@ -222,109 +328,86 @@ int MJPGToI420(const uint8* sample,
return ret ? 0 : 1;
}
#ifdef HAVE_JPEG
struct ARGBBuffers {
uint8* argb;
uint8_t* argb;
int argb_stride;
int w;
int h;
};
static void JpegI420ToARGB(void* opaque,
const uint8* const* data,
const int* strides,
int rows) {
const uint8_t* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I420ToARGB(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->argb, dest->argb_stride,
dest->w, rows);
I420ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
static void JpegI422ToARGB(void* opaque,
const uint8* const* data,
const uint8_t* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I422ToARGB(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->argb, dest->argb_stride,
dest->w, rows);
I422ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
static void JpegI444ToARGB(void* opaque,
const uint8* const* data,
const uint8_t* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I444ToARGB(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->argb, dest->argb_stride,
dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
static void JpegI411ToARGB(void* opaque,
const uint8* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I411ToARGB(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->argb, dest->argb_stride,
dest->w, rows);
I444ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
static void JpegI400ToARGB(void* opaque,
const uint8* const* data,
const uint8_t* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I400ToARGB(data[0], strides[0],
dest->argb, dest->argb_stride,
dest->w, rows);
I400ToARGB(data[0], strides[0], dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
// 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* sample,
size_t sample_size,
uint8* argb, int argb_stride,
int w, int h,
int dw, int dh) {
if (sample_size == kUnknownDataSize) {
int MJPGToARGB(const uint8_t* src_mjpg,
size_t src_size_mjpg,
uint8_t* dst_argb,
int dst_stride_argb,
int src_width,
int src_height,
int dst_width,
int dst_height) {
if (src_size_mjpg == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
}
// 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)) {
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(src_mjpg, src_size_mjpg);
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 &&
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
@@ -332,8 +415,9 @@ int MJPGToARGB(const uint8* sample,
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToARGB, &bufs, dw, dh);
// YUV422
ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToARGB, &bufs, dst_width,
dst_height);
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
@@ -343,8 +427,9 @@ int MJPGToARGB(const uint8* sample,
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToARGB, &bufs, dw, dh);
// YUV444
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToARGB, &bufs, dst_width,
dst_height);
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
@@ -354,38 +439,28 @@ int MJPGToARGB(const uint8* sample,
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToARGB, &bufs, dw, dh);
// YUV411
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 4 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI411ToARGB, &bufs, dw, dh);
// YUV400
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. 411 is supported by libjpeg
// ERROR: Unable to convert MJPEG frame because format is not supported
// TODO(fbarchard): Implement conversion for any other
// colorspace/subsample factors that occur in practice. ERROR: Unable to
// convert MJPEG frame because format is not supported
mjpeg_decoder.UnloadFrame();
return 1;
}
}
return ret ? 0 : 1;
}
#endif
#endif
#endif // HAVE_JPEG
#ifdef __cplusplus
} // extern "C"
libs/libyuv/source/convert_to_argb.cc
+142 -132
View File
@@ -28,36 +28,48 @@ extern "C" {
// src_height is used to compute location of planes, and indicate inversion
// sample_size is measured in bytes and is the size of the frame.
// With MJPEG it is the compressed size of the frame.
// TODO(fbarchard): Add the following:
// H010ToARGB
// I010ToARGB
// J400ToARGB
// J422ToARGB
// J444ToARGB
LIBYUV_API
int ConvertToARGB(const uint8* sample, size_t sample_size,
uint8* crop_argb, int argb_stride,
int crop_x, int crop_y,
int src_width, int src_height,
int crop_width, int crop_height,
int ConvertToARGB(const uint8_t* sample,
size_t sample_size,
uint8_t* dst_argb,
int dst_stride_argb,
int crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32 fourcc) {
uint32 format = CanonicalFourCC(fourcc);
uint32_t fourcc) {
uint32_t format = CanonicalFourCC(fourcc);
int aligned_src_width = (src_width + 1) & ~1;
const uint8* src;
const uint8* src_uv;
const uint8_t* src;
const uint8_t* src_uv;
int abs_src_height = (src_height < 0) ? -src_height : src_height;
int inv_crop_height = (crop_height < 0) ? -crop_height : crop_height;
int r = 0;
// One pass rotation is available for some formats. For the rest, convert
// to I420 (with optional vertical flipping) into a temporary I420 buffer,
// and then rotate the I420 to the final destination buffer.
// For in-place conversion, if destination crop_argb is same as source sample,
// to ARGB (with optional vertical flipping) into a temporary ARGB buffer,
// and then rotate the ARGB to the final destination buffer.
// For in-place conversion, if destination dst_argb is same as source sample,
// also enable temporary buffer.
LIBYUV_BOOL need_buf = (rotation && format != FOURCC_ARGB) ||
crop_argb == sample;
uint8* dest_argb = crop_argb;
int dest_argb_stride = argb_stride;
uint8* rotate_buffer = NULL;
LIBYUV_BOOL need_buf =
(rotation && format != FOURCC_ARGB) || dst_argb == sample;
uint8_t* dest_argb = dst_argb;
int dest_dst_stride_argb = dst_stride_argb;
uint8_t* rotate_buffer = NULL;
int abs_crop_height = (crop_height < 0) ? -crop_height : crop_height;
if (crop_argb == NULL || sample == NULL ||
src_width <= 0 || crop_width <= 0 ||
if (dst_argb == NULL || sample == NULL || src_width <= 0 || crop_width <= 0 ||
src_height == 0 || crop_height == 0) {
return -1;
}
@@ -67,187 +79,200 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
if (need_buf) {
int argb_size = crop_width * 4 * abs_crop_height;
rotate_buffer = (uint8*)malloc(argb_size);
rotate_buffer = (uint8_t*)malloc(argb_size); /* NOLINT */
if (!rotate_buffer) {
return 1; // Out of memory runtime error.
}
crop_argb = rotate_buffer;
argb_stride = crop_width * 4;
dst_argb = rotate_buffer;
dst_stride_argb = crop_width * 4;
}
switch (format) {
// Single plane formats
case FOURCC_YUY2:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = YUY2ToARGB(src, aligned_src_width * 2,
crop_argb, argb_stride,
r = YUY2ToARGB(src, aligned_src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_UYVY:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = UYVYToARGB(src, aligned_src_width * 2,
crop_argb, argb_stride,
r = UYVYToARGB(src, aligned_src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_24BG:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RGB24ToARGB(src, src_width * 3,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = RGB24ToARGB(src, src_width * 3, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_RAW:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RAWToARGB(src, src_width * 3,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = RAWToARGB(src, src_width * 3, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_ARGB:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBToARGB(src, src_width * 4,
crop_argb, argb_stride,
crop_width, inv_crop_height);
if (!need_buf && !rotation) {
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBToARGB(src, src_width * 4, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
}
break;
case FOURCC_BGRA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = BGRAToARGB(src, src_width * 4,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = BGRAToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_ABGR:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ABGRToARGB(src, src_width * 4,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = ABGRToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_RGBA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = RGBAToARGB(src, src_width * 4,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = RGBAToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_AR30:
src = sample + (src_width * crop_y + crop_x) * 4;
r = AR30ToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_AB30:
src = sample + (src_width * crop_y + crop_x) * 4;
r = AB30ToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_RGBP:
src = sample + (src_width * crop_y + crop_x) * 2;
r = RGB565ToARGB(src, src_width * 2,
crop_argb, argb_stride,
r = RGB565ToARGB(src, src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_RGBO:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB1555ToARGB(src, src_width * 2,
crop_argb, argb_stride,
r = ARGB1555ToARGB(src, src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_R444:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB4444ToARGB(src, src_width * 2,
crop_argb, argb_stride,
r = ARGB4444ToARGB(src, src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_I400:
src = sample + src_width * crop_y + crop_x;
r = I400ToARGB(src, src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = I400ToARGB(src, src_width, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
// Biplanar formats
case FOURCC_NV12:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + aligned_src_width * (src_height + crop_y / 2) + crop_x;
r = NV12ToARGB(src, src_width,
src_uv, aligned_src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
src_uv =
sample + aligned_src_width * (abs_src_height + crop_y / 2) + crop_x;
r = NV12ToARGB(src, src_width, src_uv, aligned_src_width, dst_argb,
dst_stride_argb, crop_width, inv_crop_height);
break;
case FOURCC_NV21:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + aligned_src_width * (src_height + crop_y / 2) + crop_x;
src_uv =
sample + aligned_src_width * (abs_src_height + crop_y / 2) + crop_x;
// Call NV12 but with u and v parameters swapped.
r = NV21ToARGB(src, src_width,
src_uv, aligned_src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = NV21ToARGB(src, src_width, src_uv, aligned_src_width, dst_argb,
dst_stride_argb, crop_width, inv_crop_height);
break;
case FOURCC_M420:
src = sample + (src_width * crop_y) * 12 / 8 + crop_x;
r = M420ToARGB(src, src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = M420ToARGB(src, src_width, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
// Triplanar formats
case FOURCC_I420:
case FOURCC_YV12: {
const uint8* src_y = sample + (src_width * crop_y + crop_x);
const uint8* src_u;
const uint8* src_v;
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u;
const uint8_t* src_v;
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
if (format == FOURCC_YV12) {
src_v = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
(halfwidth * crop_y + crop_x) / 2;
src_u = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
} else {
src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
(halfwidth * crop_y + crop_x) / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
}
r = I420ToARGB(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = I420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_H420: {
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
r = H420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_J420: {
const uint8* src_y = sample + (src_width * crop_y + crop_x);
const uint8* src_u;
const uint8* src_v;
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
r = J420ToARGB(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
crop_argb, argb_stride,
crop_width, inv_crop_height);
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
r = J420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_I422:
case FOURCC_YV16: {
const uint8* src_y = sample + src_width * crop_y + crop_x;
const uint8* src_u;
const uint8* src_v;
int halfwidth = (src_width + 1) / 2;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
if (format == FOURCC_YV16) {
src_v = sample + src_width * abs_src_height +
halfwidth * crop_y + crop_x / 2;
src_v = sample + src_width * abs_src_height + halfwidth * crop_y +
crop_x / 2;
src_u = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
} else {
src_u = sample + src_width * abs_src_height +
halfwidth * crop_y + crop_x / 2;
src_u = sample + src_width * abs_src_height + halfwidth * crop_y +
crop_x / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
}
r = I422ToARGB(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = I422ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_H422: {
int halfwidth = (src_width + 1) / 2;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u =
sample + src_width * abs_src_height + halfwidth * crop_y + crop_x / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
r = H422ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_I444:
case FOURCC_YV24: {
const uint8* src_y = sample + src_width * crop_y + crop_x;
const uint8* src_u;
const uint8* src_v;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
if (format == FOURCC_YV24) {
src_v = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_u = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
@@ -255,32 +280,14 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
}
r = I444ToARGB(src_y, src_width,
src_u, src_width,
src_v, src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
break;
}
case FOURCC_I411: {
int quarterwidth = (src_width + 3) / 4;
const uint8* src_y = sample + src_width * crop_y + crop_x;
const uint8* src_u = sample + src_width * abs_src_height +
quarterwidth * crop_y + crop_x / 4;
const uint8* src_v = sample + src_width * abs_src_height +
quarterwidth * (abs_src_height + crop_y) + crop_x / 4;
r = I411ToARGB(src_y, src_width,
src_u, quarterwidth,
src_v, quarterwidth,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = I444ToARGB(src_y, src_width, src_u, src_width, src_v, src_width,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
#ifdef HAVE_JPEG
case FOURCC_MJPG:
r = MJPGToARGB(sample, sample_size,
crop_argb, argb_stride,
src_width, abs_src_height, crop_width, inv_crop_height);
r = MJPGToARGB(sample, sample_size, dst_argb, dst_stride_argb, src_width,
abs_src_height, crop_width, inv_crop_height);
break;
#endif
default:
@@ -289,11 +296,14 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
if (need_buf) {
if (!r) {
r = ARGBRotate(crop_argb, argb_stride,
dest_argb, dest_argb_stride,
r = ARGBRotate(dst_argb, dst_stride_argb, dest_argb, dest_dst_stride_argb,
crop_width, abs_crop_height, rotation);
}
free(rotate_buffer);
} else if (rotation) {
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBRotate(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height, rotation);
}
return r;
libs/libyuv/source/convert_to_i420.cc
+123 -183
View File
@@ -25,251 +25,216 @@ extern "C" {
// sample_size is measured in bytes and is the size of the frame.
// With MJPEG it is the compressed size of the frame.
LIBYUV_API
int ConvertToI420(const uint8* sample,
int ConvertToI420(const uint8_t* sample,
size_t sample_size,
uint8* y, int y_stride,
uint8* u, int u_stride,
uint8* v, int v_stride,
int crop_x, int crop_y,
int src_width, int src_height,
int crop_width, int crop_height,
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 crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32 fourcc) {
uint32 format = CanonicalFourCC(fourcc);
uint32_t fourcc) {
uint32_t format = CanonicalFourCC(fourcc);
int aligned_src_width = (src_width + 1) & ~1;
const uint8* src;
const uint8* src_uv;
const uint8_t* src;
const uint8_t* src_uv;
const int abs_src_height = (src_height < 0) ? -src_height : src_height;
// TODO(nisse): Why allow crop_height < 0?
const int abs_crop_height = (crop_height < 0) ? -crop_height : crop_height;
int r = 0;
LIBYUV_BOOL need_buf = (rotation && format != FOURCC_I420 &&
format != FOURCC_NV12 && format != FOURCC_NV21 &&
format != FOURCC_YV12) || y == sample;
uint8* tmp_y = y;
uint8* tmp_u = u;
uint8* tmp_v = v;
int tmp_y_stride = y_stride;
int tmp_u_stride = u_stride;
int tmp_v_stride = v_stride;
uint8* rotate_buffer = NULL;
LIBYUV_BOOL need_buf =
(rotation && format != FOURCC_I420 && format != FOURCC_NV12 &&
format != FOURCC_NV21 && format != FOURCC_YV12) ||
dst_y == sample;
uint8_t* tmp_y = dst_y;
uint8_t* tmp_u = dst_u;
uint8_t* tmp_v = dst_v;
int tmp_y_stride = dst_stride_y;
int tmp_u_stride = dst_stride_u;
int tmp_v_stride = dst_stride_v;
uint8_t* rotate_buffer = NULL;
const int inv_crop_height =
(src_height < 0) ? -abs_crop_height : abs_crop_height;
if (!y || !u || !v || !sample ||
src_width <= 0 || crop_width <= 0 ||
src_height == 0 || crop_height == 0) {
if (!dst_y || !dst_u || !dst_v || !sample || src_width <= 0 ||
crop_width <= 0 || src_height == 0 || crop_height == 0) {
return -1;
}
// One pass rotation is available for some formats. For the rest, convert
// to I420 (with optional vertical flipping) into a temporary I420 buffer,
// and then rotate the I420 to the final destination buffer.
// For in-place conversion, if destination y is same as source sample,
// For in-place conversion, if destination dst_y is same as source sample,
// also enable temporary buffer.
if (need_buf) {
int y_size = crop_width * abs_crop_height;
int uv_size = ((crop_width + 1) / 2) * ((abs_crop_height + 1) / 2);
rotate_buffer = (uint8*)malloc(y_size + uv_size * 2);
rotate_buffer = (uint8_t*)malloc(y_size + uv_size * 2); /* NOLINT */
if (!rotate_buffer) {
return 1; // Out of memory runtime error.
}
y = rotate_buffer;
u = y + y_size;
v = u + uv_size;
y_stride = crop_width;
u_stride = v_stride = ((crop_width + 1) / 2);
dst_y = rotate_buffer;
dst_u = dst_y + y_size;
dst_v = dst_u + uv_size;
dst_stride_y = crop_width;
dst_stride_u = dst_stride_v = ((crop_width + 1) / 2);
}
switch (format) {
// Single plane formats
case FOURCC_YUY2:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = YUY2ToI420(src, aligned_src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = YUY2ToI420(src, aligned_src_width * 2, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_UYVY:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = UYVYToI420(src, aligned_src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = UYVYToI420(src, aligned_src_width * 2, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_RGBP:
src = sample + (src_width * crop_y + crop_x) * 2;
r = RGB565ToI420(src, src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = RGB565ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_RGBO:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB1555ToI420(src, src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = ARGB1555ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_R444:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB4444ToI420(src, src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = ARGB4444ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_24BG:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RGB24ToI420(src, src_width * 3,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = RGB24ToI420(src, src_width * 3, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_RAW:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RAWToI420(src, src_width * 3,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = RAWToI420(src, src_width * 3, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_ARGB:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBToI420(src, src_width * 4,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = ARGBToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_BGRA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = BGRAToI420(src, src_width * 4,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = BGRAToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_ABGR:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ABGRToI420(src, src_width * 4,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = ABGRToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_RGBA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = RGBAToI420(src, src_width * 4,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = RGBAToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
// TODO(fbarchard): Add AR30 and AB30
case FOURCC_I400:
src = sample + src_width * crop_y + crop_x;
r = I400ToI420(src, src_width,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = I400ToI420(src, src_width, dst_y, dst_stride_y, dst_u, dst_stride_u,
dst_v, dst_stride_v, crop_width, inv_crop_height);
break;
// Biplanar formats
case FOURCC_NV12:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + (src_width * src_height) +
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
r = NV12ToI420Rotate(src, src_width,
src_uv, aligned_src_width,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height, rotation);
src_uv = sample + (src_width * abs_src_height) +
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, dst_y,
dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, crop_width, inv_crop_height, rotation);
break;
case FOURCC_NV21:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + (src_width * src_height) +
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
// Call NV12 but with u and v parameters swapped.
r = NV12ToI420Rotate(src, src_width,
src_uv, aligned_src_width,
y, y_stride,
v, v_stride,
u, u_stride,
crop_width, inv_crop_height, rotation);
src_uv = sample + (src_width * abs_src_height) +
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
// Call NV12 but with dst_u and dst_v parameters swapped.
r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, dst_y,
dst_stride_y, dst_v, dst_stride_v, dst_u,
dst_stride_u, crop_width, inv_crop_height, rotation);
break;
case FOURCC_M420:
src = sample + (src_width * crop_y) * 12 / 8 + crop_x;
r = M420ToI420(src, src_width,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = M420ToI420(src, src_width, dst_y, dst_stride_y, dst_u, dst_stride_u,
dst_v, dst_stride_v, crop_width, inv_crop_height);
break;
// Triplanar formats
case FOURCC_I420:
case FOURCC_YV12: {
const uint8* src_y = sample + (src_width * crop_y + crop_x);
const uint8* src_u;
const uint8* src_v;
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u;
const uint8_t* src_v;
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
if (format == FOURCC_YV12) {
src_v = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
src_v = sample + src_width * abs_src_height + halfwidth * (crop_y / 2) +
(crop_x / 2);
src_u = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
halfwidth * (halfheight + (crop_y / 2)) + (crop_x / 2);
} else {
src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
src_u = sample + src_width * abs_src_height + halfwidth * (crop_y / 2) +
(crop_x / 2);
src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
halfwidth * (halfheight + (crop_y / 2)) + (crop_x / 2);
}
r = I420Rotate(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height, rotation);
r = I420Rotate(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, crop_width, inv_crop_height, rotation);
break;
}
case FOURCC_I422:
case FOURCC_YV16: {
const uint8* src_y = sample + src_width * crop_y + crop_x;
const uint8* src_u;
const uint8* src_v;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
int halfwidth = (src_width + 1) / 2;
if (format == FOURCC_YV16) {
src_v = sample + src_width * abs_src_height +
halfwidth * crop_y + crop_x / 2;
src_v = sample + src_width * abs_src_height + halfwidth * crop_y +
(crop_x / 2);
src_u = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
halfwidth * (abs_src_height + crop_y) + (crop_x / 2);
} else {
src_u = sample + src_width * abs_src_height +
halfwidth * crop_y + crop_x / 2;
src_u = sample + src_width * abs_src_height + halfwidth * crop_y +
(crop_x / 2);
src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
halfwidth * (abs_src_height + crop_y) + (crop_x / 2);
}
r = I422ToI420(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = I422ToI420(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, crop_width, inv_crop_height);
break;
}
case FOURCC_I444:
case FOURCC_YV24: {
const uint8* src_y = sample + src_width * crop_y + crop_x;
const uint8* src_u;
const uint8* src_v;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
if (format == FOURCC_YV24) {
src_v = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_u = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
@@ -277,38 +242,16 @@ int ConvertToI420(const uint8* sample,
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
}
r = I444ToI420(src_y, src_width,
src_u, src_width,
src_v, src_width,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
break;
}
case FOURCC_I411: {
int quarterwidth = (src_width + 3) / 4;
const uint8* src_y = sample + src_width * crop_y + crop_x;
const uint8* src_u = sample + src_width * abs_src_height +
quarterwidth * crop_y + crop_x / 4;
const uint8* src_v = sample + src_width * abs_src_height +
quarterwidth * (abs_src_height + crop_y) + crop_x / 4;
r = I411ToI420(src_y, src_width,
src_u, quarterwidth,
src_v, quarterwidth,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = I444ToI420(src_y, src_width, src_u, src_width, src_v, src_width,
dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, crop_width, inv_crop_height);
break;
}
#ifdef HAVE_JPEG
case FOURCC_MJPG:
r = MJPGToI420(sample, sample_size,
y, y_stride,
u, u_stride,
v, v_stride,
src_width, abs_src_height, crop_width, inv_crop_height);
r = MJPGToI420(sample, sample_size, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, src_width,
abs_src_height, crop_width, inv_crop_height);
break;
#endif
default:
@@ -317,13 +260,10 @@ int ConvertToI420(const uint8* sample,
if (need_buf) {
if (!r) {
r = I420Rotate(y, y_stride,
u, u_stride,
v, v_stride,
tmp_y, tmp_y_stride,
tmp_u, tmp_u_stride,
tmp_v, tmp_v_stride,
crop_width, abs_crop_height, rotation);
r = I420Rotate(dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, tmp_y, tmp_y_stride, tmp_u, tmp_u_stride,
tmp_v, tmp_v_stride, crop_width, abs_crop_height,
rotation);
}
free(rotate_buffer);
}
libs/libyuv/source/cpu_id.cc
+84 -132
View File
@@ -13,22 +13,16 @@
#if defined(_MSC_VER)
#include <intrin.h> // For __cpuidex()
#endif
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
!defined(__native_client__) && (defined(_M_IX86) || defined(_M_X64)) && \
defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
#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>
#include "libyuv/basic_types.h" // For CPU_X86
#ifdef __cplusplus
namespace libyuv {
extern "C" {
@@ -43,16 +37,20 @@ extern "C" {
#define SAFEBUFFERS
#endif
// cpu_info_ variable for SIMD instruction sets detected.
LIBYUV_API int cpu_info_ = 0;
// TODO(fbarchard): Consider using int for cpuid so casting is not needed.
// Low level cpuid for X86.
#if (defined(_M_IX86) || defined(_M_X64) || \
defined(__i386__) || defined(__x86_64__)) && \
#if (defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || \
defined(__x86_64__)) && \
!defined(__pnacl__) && !defined(__CLR_VER)
LIBYUV_API
void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
void CpuId(int info_eax, int info_ecx, int* cpu_info) {
#if defined(_MSC_VER)
// Visual C version uses intrinsic or inline x86 assembly.
#if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
__cpuidex((int*)(cpu_info), info_eax, info_ecx);
__cpuidex(cpu_info, info_eax, info_ecx);
#elif defined(_M_IX86)
__asm {
mov eax, info_eax
@@ -66,26 +64,26 @@ void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
}
#else // Visual C but not x86
if (info_ecx == 0) {
__cpuid((int*)(cpu_info), info_eax);
__cpuid(cpu_info, info_eax);
} else {
cpu_info[3] = cpu_info[2] = cpu_info[1] = cpu_info[0] = 0;
cpu_info[3] = cpu_info[2] = cpu_info[1] = cpu_info[0] = 0u;
}
#endif
// GCC version uses inline x86 assembly.
#else // defined(_MSC_VER)
uint32 info_ebx, info_edx;
asm volatile (
#if defined( __i386__) && defined(__PIC__)
// Preserve ebx for fpic 32 bit.
"mov %%ebx, %%edi \n"
"cpuid \n"
"xchg %%edi, %%ebx \n"
: "=D" (info_ebx),
int info_ebx, info_edx;
asm volatile(
#if defined(__i386__) && defined(__PIC__)
// Preserve ebx for fpic 32 bit.
"mov %%ebx, %%edi \n"
"cpuid \n"
"xchg %%edi, %%ebx \n"
: "=D"(info_ebx),
#else
"cpuid \n"
: "=b" (info_ebx),
"cpuid \n"
: "=b"(info_ebx),
#endif // defined( __i386__) && defined(__PIC__)
"+a" (info_eax), "+c" (info_ecx), "=d" (info_edx));
"+a"(info_eax), "+c"(info_ecx), "=d"(info_edx));
cpu_info[0] = info_eax;
cpu_info[1] = info_ebx;
cpu_info[2] = info_ecx;
@@ -94,7 +92,9 @@ void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
}
#else // (defined(_M_IX86) || defined(_M_X64) ...
LIBYUV_API
void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) {
void CpuId(int eax, int ecx, int* cpu_info) {
(void)eax;
(void)ecx;
cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0;
}
#endif
@@ -111,20 +111,22 @@ void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) {
#if defined(_M_IX86) && (_MSC_VER < 1900)
#pragma optimize("g", off)
#endif
#if (defined(_M_IX86) || defined(_M_X64) || \
defined(__i386__) || defined(__x86_64__)) && \
#if (defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || \
defined(__x86_64__)) && \
!defined(__pnacl__) && !defined(__CLR_VER) && !defined(__native_client__)
#define HAS_XGETBV
// X86 CPUs have xgetbv to detect OS saves high parts of ymm registers.
int GetXCR0() {
uint32 xcr0 = 0u;
int xcr0 = 0;
#if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
xcr0 = (uint32)(_xgetbv(0)); // VS2010 SP1 required.
xcr0 = (int)_xgetbv(0); // VS2010 SP1 required. NOLINT
#elif defined(__i386__) || defined(__x86_64__)
asm(".byte 0x0f, 0x01, 0xd0" : "=a" (xcr0) : "c" (0) : "%edx");
asm(".byte 0x0f, 0x01, 0xd0" : "=a"(xcr0) : "c"(0) : "%edx");
#endif // defined(__i386__) || defined(__x86_64__)
return xcr0;
}
#else
// xgetbv unavailable to query for OSSave support. Return 0.
#define GetXCR0() 0
#endif // defined(_M_IX86) || defined(_M_X64) ..
// Return optimization to previous setting.
#if defined(_M_IX86) && (_MSC_VER < 1900)
@@ -133,8 +135,7 @@ int GetXCR0() {
// based on libvpx arm_cpudetect.c
// For Arm, but public to allow testing on any CPU
LIBYUV_API SAFEBUFFERS
int ArmCpuCaps(const char* cpuinfo_name) {
LIBYUV_API SAFEBUFFERS int ArmCpuCaps(const char* cpuinfo_name) {
char cpuinfo_line[512];
FILE* f = fopen(cpuinfo_name, "r");
if (!f) {
@@ -151,7 +152,7 @@ int ArmCpuCaps(const char* cpuinfo_name) {
}
// aarch64 uses asimd for Neon.
p = strstr(cpuinfo_line, " asimd");
if (p && (p[6] == ' ' || p[6] == '\n')) {
if (p) {
fclose(f);
return kCpuHasNEON;
}
@@ -161,31 +162,40 @@ int ArmCpuCaps(const char* cpuinfo_name) {
return 0;
}
LIBYUV_API SAFEBUFFERS
int MipsCpuCaps(const char* cpuinfo_name, const char ase[]) {
// TODO(fbarchard): Consider read_msa_ir().
// TODO(fbarchard): Add unittest.
LIBYUV_API SAFEBUFFERS int MipsCpuCaps(const char* cpuinfo_name,
const char ase[]) {
char cpuinfo_line[512];
int len = (int)strlen(ase);
FILE* f = fopen(cpuinfo_name, "r");
if (!f) {
// ase enabled if /proc/cpuinfo is unavailable.
if (strcmp(ase, " msa") == 0) {
return kCpuHasMSA;
}
if (strcmp(ase, " dspr2") == 0) {
return kCpuHasDSPR2;
if (strcmp(ase, " mmi") == 0) {
return kCpuHasMMI;
}
return 0;
}
while (fgets(cpuinfo_line, sizeof(cpuinfo_line) - 1, f)) {
if (memcmp(cpuinfo_line, "ASEs implemented", 16) == 0) {
char* p = strstr(cpuinfo_line, ase);
if (p && (p[len] == ' ' || p[len] == '\n')) {
if (p) {
fclose(f);
if (strcmp(ase, " msa") == 0) {
return kCpuHasMSA;
}
if (strcmp(ase, " dspr2") == 0) {
return kCpuHasDSPR2;
return 0;
}
} else if (memcmp(cpuinfo_line, "cpu model", 9) == 0) {
char* p = strstr(cpuinfo_line, "Loongson-3");
if (p) {
fclose(f);
if (strcmp(ase, " mmi") == 0) {
return kCpuHasMMI;
}
return 0;
}
}
}
@@ -193,109 +203,51 @@ int MipsCpuCaps(const char* cpuinfo_name, const char ase[]) {
return 0;
}
// CPU detect function for SIMD instruction sets.
LIBYUV_API
int cpu_info_ = 0; // cpu_info is not initialized yet.
// 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
LIBYUV_API SAFEBUFFERS
int InitCpuFlags(void) {
// TODO(fbarchard): swap kCpuInit logic so 0 means uninitialized.
static SAFEBUFFERS int GetCpuFlags(void) {
int cpu_info = 0;
#if !defined(__pnacl__) && !defined(__CLR_VER) && defined(CPU_X86)
uint32 cpu_info0[4] = { 0, 0, 0, 0 };
uint32 cpu_info1[4] = { 0, 0, 0, 0 };
uint32 cpu_info7[4] = { 0, 0, 0, 0 };
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
(defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || \
defined(_M_IX86))
int cpu_info0[4] = {0, 0, 0, 0};
int cpu_info1[4] = {0, 0, 0, 0};
int cpu_info7[4] = {0, 0, 0, 0};
CpuId(0, 0, cpu_info0);
CpuId(1, 0, cpu_info1);
if (cpu_info0[0] >= 7) {
CpuId(7, 0, cpu_info7);
}
cpu_info = ((cpu_info1[3] & 0x04000000) ? kCpuHasSSE2 : 0) |
cpu_info = kCpuHasX86 | ((cpu_info1[3] & 0x04000000) ? kCpuHasSSE2 : 0) |
((cpu_info1[2] & 0x00000200) ? kCpuHasSSSE3 : 0) |
((cpu_info1[2] & 0x00080000) ? kCpuHasSSE41 : 0) |
((cpu_info1[2] & 0x00100000) ? kCpuHasSSE42 : 0) |
((cpu_info7[1] & 0x00000200) ? kCpuHasERMS : 0) |
((cpu_info1[2] & 0x00001000) ? kCpuHasFMA3 : 0) |
kCpuHasX86;
((cpu_info7[1] & 0x00000200) ? kCpuHasERMS : 0);
#ifdef HAS_XGETBV
// AVX requires CPU has AVX, XSAVE and OSXSave for xgetbv
// AVX requires OS saves YMM registers.
if (((cpu_info1[2] & 0x1c000000) == 0x1c000000) && // AVX and OSXSave
((GetXCR0() & 6) == 6)) { // Test OS saves YMM registers
cpu_info |= ((cpu_info7[1] & 0x00000020) ? kCpuHasAVX2 : 0) | kCpuHasAVX;
cpu_info |= kCpuHasAVX | ((cpu_info7[1] & 0x00000020) ? kCpuHasAVX2 : 0) |
((cpu_info1[2] & 0x00001000) ? kCpuHasFMA3 : 0) |
((cpu_info1[2] & 0x20000000) ? kCpuHasF16C : 0);
// Detect AVX512bw
if ((GetXCR0() & 0xe0) == 0xe0) {
cpu_info |= (cpu_info7[1] & 0x40000000) ? kCpuHasAVX3 : 0;
cpu_info |= (cpu_info7[1] & 0x40000000) ? kCpuHasAVX512BW : 0;
cpu_info |= (cpu_info7[1] & 0x80000000) ? kCpuHasAVX512VL : 0;
cpu_info |= (cpu_info7[2] & 0x00000002) ? kCpuHasAVX512VBMI : 0;
cpu_info |= (cpu_info7[2] & 0x00000040) ? kCpuHasAVX512VBMI2 : 0;
cpu_info |= (cpu_info7[2] & 0x00001000) ? kCpuHasAVX512VBITALG : 0;
cpu_info |= (cpu_info7[2] & 0x00004000) ? kCpuHasAVX512VPOPCNTDQ : 0;
cpu_info |= (cpu_info7[2] & 0x00000100) ? kCpuHasGFNI : 0;
}
}
#endif
// 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_AVX3")) {
cpu_info &= ~kCpuHasAVX3;
}
#endif
#if defined(__mips__) && defined(__linux__)
#if defined(__mips_dspr2)
cpu_info |= kCpuHasDSPR2;
#endif
#if defined(__mips_msa)
cpu_info = MipsCpuCaps("/proc/cpuinfo", " msa");
#elif defined(_MIPS_ARCH_LOONGSON3A)
cpu_info = MipsCpuCaps("/proc/cpuinfo", " mmi");
#endif
cpu_info |= kCpuHasMIPS;
if (getenv("LIBYUV_DISABLE_DSPR2")) {
cpu_info &= ~kCpuHasDSPR2;
}
if (getenv("LIBYUV_DISABLE_MSA")) {
cpu_info &= ~kCpuHasMSA;
}
#endif
#if defined(__arm__) || defined(__aarch64__)
// gcc -mfpu=neon defines __ARM_NEON__
@@ -314,22 +266,22 @@ int InitCpuFlags(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;
cpu_info_ = cpu_info;
cpu_info |= kCpuInitialized;
return cpu_info;
}
// Note that use of this function is not thread safe.
LIBYUV_API
void MaskCpuFlags(int enable_flags) {
cpu_info_ = InitCpuFlags() & enable_flags;
int MaskCpuFlags(int enable_flags) {
int cpu_info = GetCpuFlags() & enable_flags;
SetCpuFlags(cpu_info);
return cpu_info;
}
LIBYUV_API
int InitCpuFlags(void) {
return MaskCpuFlags(-1);
}
#ifdef __cplusplus
libs/libyuv/source/mjpeg_decoder.cc
+61 -65
View File
@@ -21,7 +21,7 @@
#if defined(_MSC_VER)
// disable warning 4324: structure was padded due to __declspec(align())
#pragma warning(disable:4324)
#pragma warning(disable : 4324)
#endif
#endif
@@ -102,7 +102,7 @@ MJpegDecoder::~MJpegDecoder() {
DestroyOutputBuffers();
}
LIBYUV_BOOL MJpegDecoder::LoadFrame(const uint8* src, size_t src_len) {
LIBYUV_BOOL MJpegDecoder::LoadFrame(const uint8_t* src, size_t src_len) {
if (!ValidateJpeg(src, src_len)) {
return LIBYUV_FALSE;
}
@@ -129,7 +129,7 @@ LIBYUV_BOOL MJpegDecoder::LoadFrame(const uint8* src, size_t src_len) {
if (scanlines_[i]) {
delete scanlines_[i];
}
scanlines_[i] = new uint8* [scanlines_size];
scanlines_[i] = new uint8_t*[scanlines_size];
scanlines_sizes_[i] = scanlines_size;
}
@@ -145,7 +145,7 @@ LIBYUV_BOOL MJpegDecoder::LoadFrame(const uint8* src, size_t src_len) {
if (databuf_[i]) {
delete databuf_[i];
}
databuf_[i] = new uint8[databuf_size];
databuf_[i] = new uint8_t[databuf_size];
databuf_strides_[i] = databuf_stride;
}
@@ -195,13 +195,11 @@ int MJpegDecoder::GetVertSampFactor(int component) {
}
int MJpegDecoder::GetHorizSubSampFactor(int component) {
return decompress_struct_->max_h_samp_factor /
GetHorizSampFactor(component);
return decompress_struct_->max_h_samp_factor / GetHorizSampFactor(component);
}
int MJpegDecoder::GetVertSubSampFactor(int component) {
return decompress_struct_->max_v_samp_factor /
GetVertSampFactor(component);
return decompress_struct_->max_v_samp_factor / GetVertSampFactor(component);
}
int MJpegDecoder::GetImageScanlinesPerImcuRow() {
@@ -245,10 +243,10 @@ LIBYUV_BOOL MJpegDecoder::UnloadFrame() {
}
// TODO(fbarchard): Allow rectangle to be specified: x, y, width, height.
LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(
uint8** planes, int dst_width, int dst_height) {
if (dst_width != GetWidth() ||
dst_height > GetHeight()) {
LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(uint8_t** planes,
int dst_width,
int dst_height) {
if (dst_width != GetWidth() || dst_height > GetHeight()) {
// ERROR: Bad dimensions
return LIBYUV_FALSE;
}
@@ -289,14 +287,13 @@ LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(
for (int i = 0; i < num_outbufs_; ++i) {
// TODO(fbarchard): Compute skip to avoid this
assert(skip % GetVertSubSampFactor(i) == 0);
int rows_to_skip =
DivideAndRoundDown(skip, GetVertSubSampFactor(i));
int scanlines_to_copy = GetComponentScanlinesPerImcuRow(i) -
rows_to_skip;
int rows_to_skip = DivideAndRoundDown(skip, GetVertSubSampFactor(i));
int scanlines_to_copy =
GetComponentScanlinesPerImcuRow(i) - rows_to_skip;
int data_to_skip = rows_to_skip * GetComponentStride(i);
CopyPlane(databuf_[i] + data_to_skip, GetComponentStride(i),
planes[i], GetComponentWidth(i),
GetComponentWidth(i), scanlines_to_copy);
CopyPlane(databuf_[i] + data_to_skip, GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i),
scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
lines_left -= (GetImageScanlinesPerImcuRow() - skip);
@@ -305,16 +302,15 @@ LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(
// Read full MCUs but cropped horizontally
for (; lines_left > GetImageScanlinesPerImcuRow();
lines_left -= GetImageScanlinesPerImcuRow()) {
lines_left -= GetImageScanlinesPerImcuRow()) {
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
for (int i = 0; i < num_outbufs_; ++i) {
int scanlines_to_copy = GetComponentScanlinesPerImcuRow(i);
CopyPlane(databuf_[i], GetComponentStride(i),
planes[i], GetComponentWidth(i),
GetComponentWidth(i), scanlines_to_copy);
CopyPlane(databuf_[i], GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i), scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
}
@@ -328,19 +324,19 @@ LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(
for (int i = 0; i < num_outbufs_; ++i) {
int scanlines_to_copy =
DivideAndRoundUp(lines_left, GetVertSubSampFactor(i));
CopyPlane(databuf_[i], GetComponentStride(i),
planes[i], GetComponentWidth(i),
GetComponentWidth(i), scanlines_to_copy);
CopyPlane(databuf_[i], GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i), scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
}
return FinishDecode();
}
LIBYUV_BOOL MJpegDecoder::DecodeToCallback(CallbackFunction fn, void* opaque,
int dst_width, int dst_height) {
if (dst_width != GetWidth() ||
dst_height > GetHeight()) {
LIBYUV_BOOL MJpegDecoder::DecodeToCallback(CallbackFunction fn,
void* opaque,
int dst_width,
int dst_height) {
if (dst_width != GetWidth() || dst_height > GetHeight()) {
// ERROR: Bad dimensions
return LIBYUV_FALSE;
}
@@ -395,7 +391,7 @@ LIBYUV_BOOL MJpegDecoder::DecodeToCallback(CallbackFunction fn, void* opaque,
}
// Read full MCUs until we get to the crop point.
for (; lines_left >= GetImageScanlinesPerImcuRow();
lines_left -= GetImageScanlinesPerImcuRow()) {
lines_left -= GetImageScanlinesPerImcuRow()) {
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
@@ -435,22 +431,22 @@ void skip_input_data(j_decompress_ptr cinfo, long num_bytes) { // NOLINT
}
void term_source(j_decompress_ptr cinfo) {
// Nothing to do.
(void)cinfo; // Nothing to do.
}
#ifdef HAVE_SETJMP
void ErrorHandler(j_common_ptr cinfo) {
// This is called when a jpeglib command experiences an error. Unfortunately
// jpeglib's error handling model is not very flexible, because it expects the
// error handler to not return--i.e., it wants the program to terminate. To
// recover from errors we use setjmp() as shown in their example. setjmp() is
// C's implementation for the "call with current continuation" functionality
// seen in some functional programming languages.
// A formatted message can be output, but is unsafe for release.
// This is called when a jpeglib command experiences an error. Unfortunately
// jpeglib's error handling model is not very flexible, because it expects the
// error handler to not return--i.e., it wants the program to terminate. To
// recover from errors we use setjmp() as shown in their example. setjmp() is
// C's implementation for the "call with current continuation" functionality
// seen in some functional programming languages.
// A formatted message can be output, but is unsafe for release.
#ifdef DEBUG
char buf[JMSG_LENGTH_MAX];
(*cinfo->err->format_message)(cinfo, buf);
// ERROR: Error in jpeglib: buf
// ERROR: Error in jpeglib: buf
#endif
SetJmpErrorMgr* mgr = reinterpret_cast<SetJmpErrorMgr*>(cinfo->err);
@@ -459,8 +455,9 @@ void ErrorHandler(j_common_ptr cinfo) {
longjmp(mgr->setjmp_buffer, 1);
}
// Suppress fprintf warnings.
void OutputHandler(j_common_ptr cinfo) {
// Suppress fprintf warnings.
(void)cinfo;
}
#endif // HAVE_SETJMP
@@ -472,9 +469,9 @@ void MJpegDecoder::AllocOutputBuffers(int num_outbufs) {
// it.
DestroyOutputBuffers();
scanlines_ = new uint8** [num_outbufs];
scanlines_ = new uint8_t**[num_outbufs];
scanlines_sizes_ = new int[num_outbufs];
databuf_ = new uint8* [num_outbufs];
databuf_ = new uint8_t*[num_outbufs];
databuf_strides_ = new int[num_outbufs];
for (int i = 0; i < num_outbufs; ++i) {
@@ -490,13 +487,13 @@ void MJpegDecoder::AllocOutputBuffers(int num_outbufs) {
void MJpegDecoder::DestroyOutputBuffers() {
for (int i = 0; i < num_outbufs_; ++i) {
delete [] scanlines_[i];
delete [] databuf_[i];
delete[] scanlines_[i];
delete[] databuf_[i];
}
delete [] scanlines_;
delete [] databuf_;
delete [] scanlines_sizes_;
delete [] databuf_strides_;
delete[] scanlines_;
delete[] databuf_;
delete[] scanlines_sizes_;
delete[] databuf_strides_;
scanlines_ = NULL;
databuf_ = NULL;
scanlines_sizes_ = NULL;
@@ -530,9 +527,9 @@ LIBYUV_BOOL MJpegDecoder::FinishDecode() {
return LIBYUV_TRUE;
}
void MJpegDecoder::SetScanlinePointers(uint8** data) {
void MJpegDecoder::SetScanlinePointers(uint8_t** data) {
for (int i = 0; i < num_outbufs_; ++i) {
uint8* data_i = data[i];
uint8_t* data_i = data[i];
for (int j = 0; j < scanlines_sizes_[i]; ++j) {
scanlines_[i][j] = data_i;
data_i += GetComponentStride(i);
@@ -542,26 +539,26 @@ void MJpegDecoder::SetScanlinePointers(uint8** data) {
inline LIBYUV_BOOL MJpegDecoder::DecodeImcuRow() {
return (unsigned int)(GetImageScanlinesPerImcuRow()) ==
jpeg_read_raw_data(decompress_struct_,
scanlines_,
GetImageScanlinesPerImcuRow());
jpeg_read_raw_data(decompress_struct_, scanlines_,
GetImageScanlinesPerImcuRow());
}
// The helper function which recognizes the jpeg sub-sampling type.
JpegSubsamplingType MJpegDecoder::JpegSubsamplingTypeHelper(
int* subsample_x, int* subsample_y, int number_of_components) {
int* subsample_x,
int* subsample_y,
int number_of_components) {
if (number_of_components == 3) { // Color images.
if (subsample_x[0] == 1 && subsample_y[0] == 1 &&
subsample_x[1] == 2 && subsample_y[1] == 2 &&
subsample_x[2] == 2 && subsample_y[2] == 2) {
if (subsample_x[0] == 1 && subsample_y[0] == 1 && subsample_x[1] == 2 &&
subsample_y[1] == 2 && subsample_x[2] == 2 && subsample_y[2] == 2) {
return kJpegYuv420;
} else if (subsample_x[0] == 1 && subsample_y[0] == 1 &&
subsample_x[1] == 2 && subsample_y[1] == 1 &&
subsample_x[2] == 2 && subsample_y[2] == 1) {
}
if (subsample_x[0] == 1 && subsample_y[0] == 1 && subsample_x[1] == 2 &&
subsample_y[1] == 1 && subsample_x[2] == 2 && subsample_y[2] == 1) {
return kJpegYuv422;
} else if (subsample_x[0] == 1 && subsample_y[0] == 1 &&
subsample_x[1] == 1 && subsample_y[1] == 1 &&
subsample_x[2] == 1 && subsample_y[2] == 1) {
}
if (subsample_x[0] == 1 && subsample_y[0] == 1 && subsample_x[1] == 1 &&
subsample_y[1] == 1 && subsample_x[2] == 1 && subsample_y[2] == 1) {
return kJpegYuv444;
}
} else if (number_of_components == 1) { // Grey-scale images.
@@ -574,4 +571,3 @@ JpegSubsamplingType MJpegDecoder::JpegSubsamplingTypeHelper(
} // namespace libyuv
#endif // HAVE_JPEG
libs/libyuv/source/mjpeg_validate.cc
+14 -15
View File
@@ -18,13 +18,13 @@ extern "C" {
#endif
// Helper function to scan for EOI marker (0xff 0xd9).
static LIBYUV_BOOL ScanEOI(const uint8* sample, size_t sample_size) {
if (sample_size >= 2) {
const uint8* end = sample + sample_size - 1;
const uint8* it = sample;
static LIBYUV_BOOL ScanEOI(const uint8_t* src_mjpg, size_t src_size_mjpg) {
if (src_size_mjpg >= 2) {
const uint8_t* end = src_mjpg + src_size_mjpg - 1;
const uint8_t* it = src_mjpg;
while (it < end) {
// TODO(fbarchard): scan for 0xd9 instead.
it = static_cast<const uint8 *>(memchr(it, 0xff, end - it));
it = (const uint8_t*)(memchr(it, 0xff, end - it));
if (it == NULL) {
break;
}
@@ -34,38 +34,37 @@ static LIBYUV_BOOL ScanEOI(const uint8* sample, size_t sample_size) {
++it; // Skip over current 0xff.
}
}
// ERROR: Invalid jpeg end code not found. Size sample_size
// ERROR: Invalid jpeg end code not found. Size src_size_mjpg
return LIBYUV_FALSE;
}
// Helper function to validate the jpeg appears intact.
LIBYUV_BOOL ValidateJpeg(const uint8* sample, size_t sample_size) {
LIBYUV_BOOL ValidateJpeg(const uint8_t* src_mjpg, size_t src_size_mjpg) {
// Maximum size that ValidateJpeg will consider valid.
const size_t kMaxJpegSize = 0x7fffffffull;
const size_t kBackSearchSize = 1024;
if (sample_size < 64 || sample_size > kMaxJpegSize || !sample) {
// ERROR: Invalid jpeg size: sample_size
if (src_size_mjpg < 64 || src_size_mjpg > kMaxJpegSize || !src_mjpg) {
// ERROR: Invalid jpeg size: src_size_mjpg
return LIBYUV_FALSE;
}
if (sample[0] != 0xff || sample[1] != 0xd8) { // SOI marker
if (src_mjpg[0] != 0xff || src_mjpg[1] != 0xd8) { // SOI marker
// ERROR: Invalid jpeg initial start code
return LIBYUV_FALSE;
}
// Look for the End Of Image (EOI) marker near the end of the buffer.
if (sample_size > kBackSearchSize) {
if (ScanEOI(sample + sample_size - kBackSearchSize, kBackSearchSize)) {
if (src_size_mjpg > kBackSearchSize) {
if (ScanEOI(src_mjpg + src_size_mjpg - kBackSearchSize, kBackSearchSize)) {
return LIBYUV_TRUE; // Success: Valid jpeg.
}
// Reduce search size for forward search.
sample_size = sample_size - kBackSearchSize + 1;
src_size_mjpg = src_size_mjpg - kBackSearchSize + 1;
}
// Step over SOI marker and scan for EOI.
return ScanEOI(sample + 2, sample_size - 2);
return ScanEOI(src_mjpg + 2, src_size_mjpg - 2);
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
libs/libyuv/source/planar_functions.cc
+1609 -553
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/rotate.cc
+224 -178
View File
@@ -10,8 +10,8 @@
#include "libyuv/rotate.h"
#include "libyuv/cpu_id.h"
#include "libyuv/convert.h"
#include "libyuv/cpu_id.h"
#include "libyuv/planar_functions.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
@@ -22,12 +22,20 @@ extern "C" {
#endif
LIBYUV_API
void TransposePlane(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void TransposePlane(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height) {
int i = height;
void (*TransposeWx8)(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) = TransposeWx8_C;
#if defined(HAS_TRANSPOSEWX16_MSA)
void (*TransposeWx16)(const uint8_t* src, int src_stride, uint8_t* dst,
int dst_stride, int width) = TransposeWx16_C;
#else
void (*TransposeWx8)(const uint8_t* src, int src_stride, uint8_t* dst,
int dst_stride, int width) = TransposeWx8_C;
#endif
#if defined(HAS_TRANSPOSEWX8_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
TransposeWx8 = TransposeWx8_NEON;
@@ -41,6 +49,11 @@ void TransposePlane(const uint8* src, int src_stride,
}
}
#endif
#if defined(HAS_TRANSPOSEWX8_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
TransposeWx8 = TransposeWx8_MMI;
}
#endif
#if defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
TransposeWx8 = TransposeWx8_Fast_Any_SSSE3;
@@ -49,24 +62,32 @@ void TransposePlane(const uint8* src, int src_stride,
}
}
#endif
#if defined(HAS_TRANSPOSEWX8_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2)) {
if (IS_ALIGNED(width, 4) &&
IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
TransposeWx8 = TransposeWx8_Fast_DSPR2;
} else {
TransposeWx8 = TransposeWx8_DSPR2;
#if defined(HAS_TRANSPOSEWX16_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
TransposeWx16 = TransposeWx16_Any_MSA;
if (IS_ALIGNED(width, 16)) {
TransposeWx16 = TransposeWx16_MSA;
}
}
#endif
#if defined(HAS_TRANSPOSEWX16_MSA)
// Work across the source in 16x16 tiles
while (i >= 16) {
TransposeWx16(src, src_stride, dst, dst_stride, width);
src += 16 * src_stride; // Go down 16 rows.
dst += 16; // Move over 16 columns.
i -= 16;
}
#else
// Work across the source in 8x8 tiles
while (i >= 8) {
TransposeWx8(src, src_stride, dst, dst_stride, width);
src += 8 * src_stride; // Go down 8 rows.
dst += 8; // Move over 8 columns.
src += 8 * src_stride; // Go down 8 rows.
dst += 8; // Move over 8 columns.
i -= 8;
}
#endif
if (i > 0) {
TransposeWxH_C(src, src_stride, dst, dst_stride, width, i);
@@ -74,9 +95,12 @@ void TransposePlane(const uint8* src, int src_stride,
}
LIBYUV_API
void RotatePlane90(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void RotatePlane90(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height) {
// Rotate by 90 is a transpose 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.
@@ -86,9 +110,12 @@ void RotatePlane90(const uint8* src, int src_stride,
}
LIBYUV_API
void RotatePlane270(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void RotatePlane270(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height) {
// Rotate by 270 is a transpose 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.
@@ -98,17 +125,20 @@ void RotatePlane270(const uint8* src, int src_stride,
}
LIBYUV_API
void RotatePlane180(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void RotatePlane180(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height) {
// Swap first and last row and mirror the content. Uses a temporary row.
align_buffer_64(row, width);
const uint8* src_bot = src + src_stride * (height - 1);
uint8* dst_bot = dst + dst_stride * (height - 1);
const uint8_t* src_bot = src + src_stride * (height - 1);
uint8_t* dst_bot = dst + dst_stride * (height - 1);
int half_height = (height + 1) >> 1;
int y;
void (*MirrorRow)(const uint8* src, uint8* dst, int width) = MirrorRow_C;
void (*CopyRow)(const uint8* src, uint8* dst, int width) = CopyRow_C;
void (*MirrorRow)(const uint8_t* src, uint8_t* dst, int width) = MirrorRow_C;
void (*CopyRow)(const uint8_t* src, uint8_t* dst, int width) = CopyRow_C;
#if defined(HAS_MIRRORROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
MirrorRow = MirrorRow_Any_NEON;
@@ -133,21 +163,21 @@ void RotatePlane180(const uint8* src, int src_stride,
}
}
#endif
// TODO(fbarchard): Mirror on mips handle unaligned memory.
#if defined(HAS_MIRRORROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4) &&
IS_ALIGNED(dst, 4) && IS_ALIGNED(dst_stride, 4)) {
MirrorRow = MirrorRow_DSPR2;
}
#endif
#if defined(HAS_MIRRORROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
MirrorRow = MirrorRow_Any_MSA;
if (IS_ALIGNED(width, 64)) {
MirrorRow = MirrorRow_MSA;
}
}
}
#endif
#if defined(HAS_MIRRORROW_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
MirrorRow = MirrorRow_Any_MMI;
if (IS_ALIGNED(width, 8)) {
MirrorRow = MirrorRow_MMI;
}
}
#endif
#if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
@@ -169,9 +199,9 @@ void RotatePlane180(const uint8* src, int src_stride,
CopyRow = IS_ALIGNED(width, 32) ? CopyRow_NEON : CopyRow_Any_NEON;
}
#endif
#if defined(HAS_COPYROW_MIPS)
if (TestCpuFlag(kCpuHasMIPS)) {
CopyRow = CopyRow_MIPS;
#if defined(HAS_COPYROW_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
CopyRow = IS_ALIGNED(width, 8) ? CopyRow_MMI : CopyRow_Any_MMI;
}
#endif
@@ -189,15 +219,24 @@ void RotatePlane180(const uint8* src, int src_stride,
}
LIBYUV_API
void TransposeUV(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void TransposeUV(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height) {
int i = height;
void (*TransposeUVWx8)(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
#if defined(HAS_TRANSPOSEUVWX16_MSA)
void (*TransposeUVWx16)(const uint8_t* src, int src_stride, uint8_t* dst_a,
int dst_stride_a, uint8_t* dst_b, int dst_stride_b,
int width) = TransposeUVWx16_C;
#else
void (*TransposeUVWx8)(const uint8_t* src, int src_stride, uint8_t* dst_a,
int dst_stride_a, uint8_t* dst_b, int dst_stride_b,
int width) = TransposeUVWx8_C;
#endif
#if defined(HAS_TRANSPOSEUVWX8_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
TransposeUVWx8 = TransposeUVWx8_NEON;
@@ -211,72 +250,98 @@ void TransposeUV(const uint8* src, int src_stride,
}
}
#endif
#if defined(HAS_TRANSPOSEUVWX8_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 2) &&
IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
TransposeUVWx8 = TransposeUVWx8_DSPR2;
#if defined(HAS_TRANSPOSEUVWX8_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
TransposeUVWx8 = TransposeUVWx8_Any_MMI;
if (IS_ALIGNED(width, 4)) {
TransposeUVWx8 = TransposeUVWx8_MMI;
}
}
#endif
#if defined(HAS_TRANSPOSEUVWX16_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
TransposeUVWx16 = TransposeUVWx16_Any_MSA;
if (IS_ALIGNED(width, 8)) {
TransposeUVWx16 = TransposeUVWx16_MSA;
}
}
#endif
#if defined(HAS_TRANSPOSEUVWX16_MSA)
// Work through the source in 8x8 tiles.
while (i >= 16) {
TransposeUVWx16(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width);
src += 16 * src_stride; // Go down 16 rows.
dst_a += 16; // Move over 8 columns.
dst_b += 16; // Move over 8 columns.
i -= 16;
}
#else
// Work through the source in 8x8 tiles.
while (i >= 8) {
TransposeUVWx8(src, src_stride,
dst_a, dst_stride_a,
dst_b, dst_stride_b,
TransposeUVWx8(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width);
src += 8 * src_stride; // Go down 8 rows.
dst_a += 8; // Move over 8 columns.
dst_b += 8; // Move over 8 columns.
src += 8 * src_stride; // Go down 8 rows.
dst_a += 8; // Move over 8 columns.
dst_b += 8; // Move over 8 columns.
i -= 8;
}
#endif
if (i > 0) {
TransposeUVWxH_C(src, src_stride,
dst_a, dst_stride_a,
dst_b, dst_stride_b,
TransposeUVWxH_C(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width, i);
}
}
LIBYUV_API
void RotateUV90(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void RotateUV90(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height) {
src += src_stride * (height - 1);
src_stride = -src_stride;
TransposeUV(src, src_stride,
dst_a, dst_stride_a,
dst_b, dst_stride_b,
width, height);
TransposeUV(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, width,
height);
}
LIBYUV_API
void RotateUV270(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void RotateUV270(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height) {
dst_a += dst_stride_a * (width - 1);
dst_b += dst_stride_b * (width - 1);
dst_stride_a = -dst_stride_a;
dst_stride_b = -dst_stride_b;
TransposeUV(src, src_stride,
dst_a, dst_stride_a,
dst_b, dst_stride_b,
width, height);
TransposeUV(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, width,
height);
}
// Rotate 180 is a horizontal and vertical flip.
LIBYUV_API
void RotateUV180(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void RotateUV180(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height) {
int i;
void (*MirrorUVRow)(const uint8* src, uint8* dst_u, uint8* dst_v, int width) =
MirrorUVRow_C;
void (*MirrorUVRow)(const uint8_t* src, uint8_t* dst_u, uint8_t* dst_v,
int width) = MirrorUVRow_C;
#if defined(HAS_MIRRORUVROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
MirrorUVRow = MirrorUVRow_NEON;
@@ -287,10 +352,14 @@ void RotateUV180(const uint8* src, int src_stride,
MirrorUVRow = MirrorUVRow_SSSE3;
}
#endif
#if defined(HAS_MIRRORUVROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
MirrorUVRow = MirrorUVRow_DSPR2;
#if defined(HAS_MIRRORUVROW_MSA)
if (TestCpuFlag(kCpuHasMSA) && IS_ALIGNED(width, 32)) {
MirrorUVRow = MirrorUVRow_MSA;
}
#endif
#if defined(HAS_MIRRORUVROW_MMI)
if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(width, 8)) {
MirrorUVRow = MirrorUVRow_MMI;
}
#endif
@@ -306,9 +375,12 @@ void RotateUV180(const uint8* src, int src_stride,
}
LIBYUV_API
int RotatePlane(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height,
int RotatePlane(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height,
enum RotationMode mode) {
if (!src || width <= 0 || height == 0 || !dst) {
return -1;
@@ -324,24 +396,16 @@ int RotatePlane(const uint8* src, int src_stride,
switch (mode) {
case kRotate0:
// copy frame
CopyPlane(src, src_stride,
dst, dst_stride,
width, height);
CopyPlane(src, src_stride, dst, dst_stride, width, height);
return 0;
case kRotate90:
RotatePlane90(src, src_stride,
dst, dst_stride,
width, height);
RotatePlane90(src, src_stride, dst, dst_stride, width, height);
return 0;
case kRotate270:
RotatePlane270(src, src_stride,
dst, dst_stride,
width, height);
RotatePlane270(src, src_stride, dst, dst_stride, width, height);
return 0;
case kRotate180:
RotatePlane180(src, src_stride,
dst, dst_stride,
width, height);
RotatePlane180(src, src_stride, dst, dst_stride, width, height);
return 0;
default:
break;
@@ -350,18 +414,25 @@ int RotatePlane(const uint8* src, int src_stride,
}
LIBYUV_API
int I420Rotate(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height,
int I420Rotate(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
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 width,
int height,
enum RotationMode mode) {
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
if (!src_y || !src_u || !src_v || width <= 0 || height == 0 ||
!dst_y || !dst_u || !dst_v) {
if (!src_y || !src_u || !src_v || width <= 0 || height == 0 || !dst_y ||
!dst_u || !dst_v) {
return -1;
}
@@ -380,45 +451,29 @@ int I420Rotate(const uint8* src_y, int src_stride_y,
switch (mode) {
case kRotate0:
// copy frame
return I420Copy(src_y, src_stride_y,
src_u, src_stride_u,
src_v, src_stride_v,
dst_y, dst_stride_y,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
width, height);
return I420Copy(src_y, src_stride_y, src_u, src_stride_u, src_v,
src_stride_v, dst_y, dst_stride_y, dst_u, dst_stride_u,
dst_v, dst_stride_v, width, height);
case kRotate90:
RotatePlane90(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotatePlane90(src_u, src_stride_u,
dst_u, dst_stride_u,
halfwidth, halfheight);
RotatePlane90(src_v, src_stride_v,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane90(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotatePlane90(src_u, src_stride_u, dst_u, dst_stride_u, halfwidth,
halfheight);
RotatePlane90(src_v, src_stride_v, dst_v, dst_stride_v, halfwidth,
halfheight);
return 0;
case kRotate270:
RotatePlane270(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotatePlane270(src_u, src_stride_u,
dst_u, dst_stride_u,
halfwidth, halfheight);
RotatePlane270(src_v, src_stride_v,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane270(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotatePlane270(src_u, src_stride_u, dst_u, dst_stride_u, halfwidth,
halfheight);
RotatePlane270(src_v, src_stride_v, dst_v, dst_stride_v, halfwidth,
halfheight);
return 0;
case kRotate180:
RotatePlane180(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotatePlane180(src_u, src_stride_u,
dst_u, dst_stride_u,
halfwidth, halfheight);
RotatePlane180(src_v, src_stride_v,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane180(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotatePlane180(src_u, src_stride_u, dst_u, dst_stride_u, halfwidth,
halfheight);
RotatePlane180(src_v, src_stride_v, dst_v, dst_stride_v, halfwidth,
halfheight);
return 0;
default:
break;
@@ -427,17 +482,23 @@ int I420Rotate(const uint8* src_y, int src_stride_y,
}
LIBYUV_API
int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height,
int NV12ToI420Rotate(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_uv,
int src_stride_uv,
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 width,
int height,
enum RotationMode mode) {
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
if (!src_y || !src_uv || width <= 0 || height == 0 ||
!dst_y || !dst_u || !dst_v) {
if (!src_y || !src_uv || width <= 0 || height == 0 || !dst_y || !dst_u ||
!dst_v) {
return -1;
}
@@ -454,38 +515,23 @@ int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
switch (mode) {
case kRotate0:
// copy frame
return NV12ToI420(src_y, src_stride_y,
src_uv, src_stride_uv,
dst_y, dst_stride_y,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
return NV12ToI420(src_y, src_stride_y, src_uv, src_stride_uv, dst_y,
dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v,
width, height);
case kRotate90:
RotatePlane90(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotateUV90(src_uv, src_stride_uv,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane90(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotateUV90(src_uv, src_stride_uv, dst_u, dst_stride_u, dst_v,
dst_stride_v, halfwidth, halfheight);
return 0;
case kRotate270:
RotatePlane270(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotateUV270(src_uv, src_stride_uv,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane270(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotateUV270(src_uv, src_stride_uv, dst_u, dst_stride_u, dst_v,
dst_stride_v, halfwidth, halfheight);
return 0;
case kRotate180:
RotatePlane180(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotateUV180(src_uv, src_stride_uv,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane180(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotateUV180(src_uv, src_stride_uv, dst_u, dst_stride_u, dst_v,
dst_stride_v, halfwidth, halfheight);
return 0;
default:
break;
libs/libyuv/source/rotate_any.cc
+31 -32
View File
@@ -18,16 +18,16 @@ namespace libyuv {
extern "C" {
#endif
#define TANY(NAMEANY, TPOS_SIMD, MASK) \
void NAMEANY(const uint8* src, int src_stride, \
uint8* dst, int dst_stride, int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst, dst_stride, n); \
} \
TransposeWx8_C(src + n, src_stride, dst + n * dst_stride, dst_stride, r);\
}
#define TANY(NAMEANY, TPOS_SIMD, MASK) \
void NAMEANY(const uint8_t* src, int src_stride, uint8_t* dst, \
int dst_stride, int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst, dst_stride, n); \
} \
TransposeWx8_C(src + n, src_stride, dst + n * dst_stride, dst_stride, r); \
}
#ifdef HAS_TRANSPOSEWX8_NEON
TANY(TransposeWx8_Any_NEON, TransposeWx8_NEON, 7)
@@ -35,28 +35,29 @@ TANY(TransposeWx8_Any_NEON, TransposeWx8_NEON, 7)
#ifdef HAS_TRANSPOSEWX8_SSSE3
TANY(TransposeWx8_Any_SSSE3, TransposeWx8_SSSE3, 7)
#endif
#ifdef HAS_TRANSPOSEWX8_MMI
TANY(TransposeWx8_Any_MMI, TransposeWx8_MMI, 7)
#endif
#ifdef HAS_TRANSPOSEWX8_FAST_SSSE3
TANY(TransposeWx8_Fast_Any_SSSE3, TransposeWx8_Fast_SSSE3, 15)
#endif
#ifdef HAS_TRANSPOSEWX8_DSPR2
TANY(TransposeWx8_Any_DSPR2, TransposeWx8_DSPR2, 7)
#ifdef HAS_TRANSPOSEWX16_MSA
TANY(TransposeWx16_Any_MSA, TransposeWx16_MSA, 15)
#endif
#undef TANY
#define TUVANY(NAMEANY, TPOS_SIMD, MASK) \
void NAMEANY(const uint8* src, int src_stride, \
uint8* dst_a, int dst_stride_a, \
uint8* dst_b, int dst_stride_b, int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, \
n); \
} \
TransposeUVWx8_C(src + n * 2, src_stride, \
dst_a + n * dst_stride_a, dst_stride_a, \
dst_b + n * dst_stride_b, dst_stride_b, r); \
}
void NAMEANY(const uint8_t* src, int src_stride, uint8_t* dst_a, \
int dst_stride_a, uint8_t* dst_b, int dst_stride_b, \
int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, n); \
} \
TransposeUVWx8_C(src + n * 2, src_stride, dst_a + n * dst_stride_a, \
dst_stride_a, dst_b + n * dst_stride_b, dst_stride_b, r); \
}
#ifdef HAS_TRANSPOSEUVWX8_NEON
TUVANY(TransposeUVWx8_Any_NEON, TransposeUVWx8_NEON, 7)
@@ -64,8 +65,11 @@ TUVANY(TransposeUVWx8_Any_NEON, TransposeUVWx8_NEON, 7)
#ifdef HAS_TRANSPOSEUVWX8_SSE2
TUVANY(TransposeUVWx8_Any_SSE2, TransposeUVWx8_SSE2, 7)
#endif
#ifdef HAS_TRANSPOSEUVWX8_DSPR2
TUVANY(TransposeUVWx8_Any_DSPR2, TransposeUVWx8_DSPR2, 7)
#ifdef HAS_TRANSPOSEUVWX8_MMI
TUVANY(TransposeUVWx8_Any_MMI, TransposeUVWx8_MMI, 7)
#endif
#ifdef HAS_TRANSPOSEUVWX16_MSA
TUVANY(TransposeUVWx16_Any_MSA, TransposeUVWx16_MSA, 7)
#endif
#undef TUVANY
@@ -73,8 +77,3 @@ TUVANY(TransposeUVWx8_Any_DSPR2, TransposeUVWx8_DSPR2, 7)
} // extern "C"
} // namespace libyuv
#endif
libs/libyuv/source/rotate_argb.cc
+99 -72
View File
@@ -10,90 +10,114 @@
#include "libyuv/rotate.h"
#include "libyuv/cpu_id.h"
#include "libyuv/convert.h"
#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(__native_client__)) || defined(__i386__))
#define HAS_SCALEARGBROWDOWNEVEN_SSE2
void ScaleARGBRowDownEven_SSE2(const uint8* src_ptr, int src_stride,
int src_stepx, uint8* dst_ptr, int dst_width);
#endif
#if !defined(LIBYUV_DISABLE_NEON) && !defined(__native_client__) && \
(defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))
#define HAS_SCALEARGBROWDOWNEVEN_NEON
void ScaleARGBRowDownEven_NEON(const uint8* src_ptr, int src_stride,
int src_stepx, uint8* dst_ptr, int dst_width);
#endif
void ScaleARGBRowDownEven_C(const uint8* src_ptr, int,
int src_stepx, uint8* dst_ptr, int dst_width);
static void ARGBTranspose(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width, int height) {
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* src_ptr, int src_stride,
int src_step, uint8* 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
#if defined(HAS_SCALEARGBROWDOWNEVEN_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
ScaleARGBRowDownEven = ScaleARGBRowDownEven_Any_MMI;
if (IS_ALIGNED(height, 4)) { // Width of dest.
ScaleARGBRowDownEven = ScaleARGBRowDownEven_MMI;
}
}
#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* src, int src_stride,
uint8* dst, int dst_stride, int width, int height) {
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* src, int src_stride,
uint8* dst, int dst_stride, int width, int height) {
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* src, int src_stride,
uint8* dst, int dst_stride, int width, int height) {
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* src_bot = src + src_stride * (height - 1);
uint8* 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* src, uint8* dst, int width) =
void (*ARGBMirrorRow)(const uint8_t* src_argb, uint8_t* dst_argb, int width) =
ARGBMirrorRow_C;
void (*CopyRow)(const uint8* src, uint8* 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;
@@ -126,6 +150,14 @@ void ARGBRotate180(const uint8* src, int src_stride,
}
}
#endif
#if defined(HAS_ARGBMIRRORROW_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
ARGBMirrorRow = ARGBMirrorRow_Any_MMI;
if (IS_ALIGNED(width, 2)) {
ARGBMirrorRow = ARGBMirrorRow_MMI;
}
}
#endif
#if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
CopyRow = IS_ALIGNED(width * 4, 32) ? CopyRow_SSE2 : CopyRow_Any_SSE2;
@@ -146,28 +178,27 @@ void ARGBRotate180(const uint8* src, int src_stride,
CopyRow = IS_ALIGNED(width * 4, 32) ? CopyRow_NEON : CopyRow_Any_NEON;
}
#endif
#if defined(HAS_COPYROW_MIPS)
if (TestCpuFlag(kCpuHasMIPS)) {
CopyRow = CopyRow_MIPS;
}
#endif
// 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
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 += src_stride;
dst += dst_stride;
src_bot -= src_stride;
dst_bot -= dst_stride;
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);
}
LIBYUV_API
int ARGBRotate(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb, int width, int height,
int ARGBRotate(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height,
enum RotationMode mode) {
if (!src_argb || width <= 0 || height == 0 || !dst_argb) {
return -1;
@@ -183,23 +214,19 @@ int ARGBRotate(const uint8* src_argb, int src_stride_argb,
switch (mode) {
case kRotate0:
// copy frame
return ARGBCopy(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
return ARGBCopy(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height);
case kRotate90:
ARGBRotate90(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
ARGBRotate90(src_argb, src_stride_argb, dst_argb, dst_stride_argb, width,
height);
return 0;
case kRotate270:
ARGBRotate270(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
ARGBRotate270(src_argb, src_stride_argb, dst_argb, dst_stride_argb, width,
height);
return 0;
case kRotate180:
ARGBRotate180(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
ARGBRotate180(src_argb, src_stride_argb, dst_argb, dst_stride_argb, width,
height);
return 0;
default:
break;
libs/libyuv/source/rotate_common.cc
+27 -13
View File
@@ -8,16 +8,19 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
void TransposeWx8_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
void TransposeWx8_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
int i;
for (i = 0; i < width; ++i) {
dst[0] = src[0 * src_stride];
@@ -33,9 +36,13 @@ void TransposeWx8_C(const uint8* src, int src_stride,
}
}
void TransposeUVWx8_C(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width) {
void TransposeUVWx8_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
int i;
for (i = 0; i < width; ++i) {
dst_a[0] = src[0 * src_stride + 0];
@@ -60,9 +67,12 @@ void TransposeUVWx8_C(const uint8* src, int src_stride,
}
}
void TransposeWxH_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void TransposeWxH_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height) {
int i;
for (i = 0; i < width; ++i) {
int j;
@@ -72,10 +82,14 @@ void TransposeWxH_C(const uint8* src, int src_stride,
}
}
void TransposeUVWxH_C(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void TransposeUVWxH_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height) {
int i;
for (i = 0; i < width * 2; i += 2) {
int j;
libs/libyuv/source/rotate_gcc.cc
+333 -327
View File
@@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
@@ -22,342 +22,348 @@ extern "C" {
// Transpose 8x8. 32 or 64 bit, but not NaCL for 64 bit.
#if defined(HAS_TRANSPOSEWX8_SSSE3)
void TransposeWx8_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
asm volatile (
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movq (%0),%%xmm0 \n"
"movq (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"movq (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"movq (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movq (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"movq (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movq (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"lea 0x8(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"sub $0x8,%2 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc",
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
);
void TransposeWx8_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movq (%0),%%xmm0 \n"
"movq (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"movq (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"movq (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movq (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"movq (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movq (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"lea 0x8(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"sub $0x8,%2 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
}
#endif // defined(HAS_TRANSPOSEWX8_SSSE3)
// Transpose 16x8. 64 bit
#if defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
void TransposeWx8_Fast_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
asm volatile (
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqu (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"movdqa %%xmm8,%%xmm9 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"palignr $0x8,%%xmm9,%%xmm9 \n"
"movdqu (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm2,%%xmm10 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm10 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movdqa %%xmm10,%%xmm11 \n"
"movdqu (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"movdqu (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm4,%%xmm12 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm12 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movdqa %%xmm12,%%xmm13 \n"
"movdqu (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movdqu (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm6,%%xmm14 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"punpckhbw %%xmm7,%%xmm14 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"movdqa %%xmm14,%%xmm15 \n"
"lea 0x10(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"punpcklwd %%xmm10,%%xmm8 \n"
"punpcklwd %%xmm11,%%xmm9 \n"
"movdqa %%xmm8,%%xmm10 \n"
"movdqa %%xmm9,%%xmm11 \n"
"palignr $0x8,%%xmm10,%%xmm10 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"punpcklwd %%xmm14,%%xmm12 \n"
"punpcklwd %%xmm15,%%xmm13 \n"
"movdqa %%xmm12,%%xmm14 \n"
"movdqa %%xmm13,%%xmm15 \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm12,%%xmm8 \n"
"movq %%xmm8,(%1) \n"
"movdqa %%xmm8,%%xmm12 \n"
"palignr $0x8,%%xmm12,%%xmm12 \n"
"movq %%xmm12,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm14,%%xmm10 \n"
"movdqa %%xmm10,%%xmm14 \n"
"movq %%xmm10,(%1) \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"punpckldq %%xmm13,%%xmm9 \n"
"movq %%xmm14,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm9,%%xmm13 \n"
"movq %%xmm9,(%1) \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movq %%xmm13,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm15,%%xmm11 \n"
"movq %%xmm11,(%1) \n"
"movdqa %%xmm11,%%xmm15 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"sub $0x10,%2 \n"
"movq %%xmm15,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc",
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
"xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15"
);
void TransposeWx8_Fast_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqu (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"movdqa %%xmm8,%%xmm9 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"palignr $0x8,%%xmm9,%%xmm9 \n"
"movdqu (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm2,%%xmm10 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm10 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movdqa %%xmm10,%%xmm11 \n"
"movdqu (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"movdqu (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm4,%%xmm12 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm12 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movdqa %%xmm12,%%xmm13 \n"
"movdqu (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movdqu (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm6,%%xmm14 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"punpckhbw %%xmm7,%%xmm14 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"movdqa %%xmm14,%%xmm15 \n"
"lea 0x10(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"punpcklwd %%xmm10,%%xmm8 \n"
"punpcklwd %%xmm11,%%xmm9 \n"
"movdqa %%xmm8,%%xmm10 \n"
"movdqa %%xmm9,%%xmm11 \n"
"palignr $0x8,%%xmm10,%%xmm10 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"punpcklwd %%xmm14,%%xmm12 \n"
"punpcklwd %%xmm15,%%xmm13 \n"
"movdqa %%xmm12,%%xmm14 \n"
"movdqa %%xmm13,%%xmm15 \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm12,%%xmm8 \n"
"movq %%xmm8,(%1) \n"
"movdqa %%xmm8,%%xmm12 \n"
"palignr $0x8,%%xmm12,%%xmm12 \n"
"movq %%xmm12,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm14,%%xmm10 \n"
"movdqa %%xmm10,%%xmm14 \n"
"movq %%xmm10,(%1) \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"punpckldq %%xmm13,%%xmm9 \n"
"movq %%xmm14,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm9,%%xmm13 \n"
"movq %%xmm9,(%1) \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movq %%xmm13,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm15,%%xmm11 \n"
"movq %%xmm11,(%1) \n"
"movdqa %%xmm11,%%xmm15 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"sub $0x10,%2 \n"
"movq %%xmm15,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7", "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14",
"xmm15");
}
#endif // defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
// Transpose UV 8x8. 64 bit.
#if defined(HAS_TRANSPOSEUVWX8_SSE2)
void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width) {
asm volatile (
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%4),%%xmm1 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqa %%xmm8,%%xmm1 \n"
"movdqu (%0),%%xmm2 \n"
"movdqu (%0,%4),%%xmm3 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm8 \n"
"movdqa %%xmm8,%%xmm3 \n"
"movdqu (%0),%%xmm4 \n"
"movdqu (%0,%4),%%xmm5 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm4,%%xmm8 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm8 \n"
"movdqa %%xmm8,%%xmm5 \n"
"movdqu (%0),%%xmm6 \n"
"movdqu (%0,%4),%%xmm7 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm6,%%xmm8 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %4 \n"
"lea 0x10(%0,%4,8),%0 \n"
"punpckhbw %%xmm7,%%xmm8 \n"
"movdqa %%xmm8,%%xmm7 \n"
"neg %4 \n"
// Second round of bit swap.
"movdqa %%xmm0,%%xmm8 \n"
"movdqa %%xmm1,%%xmm9 \n"
"punpckhwd %%xmm2,%%xmm8 \n"
"punpckhwd %%xmm3,%%xmm9 \n"
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm8,%%xmm2 \n"
"movdqa %%xmm9,%%xmm3 \n"
"movdqa %%xmm4,%%xmm8 \n"
"movdqa %%xmm5,%%xmm9 \n"
"punpckhwd %%xmm6,%%xmm8 \n"
"punpckhwd %%xmm7,%%xmm9 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm8,%%xmm6 \n"
"movdqa %%xmm9,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"movdqa %%xmm0,%%xmm8 \n"
"punpckldq %%xmm4,%%xmm0 \n"
"movlpd %%xmm0,(%1) \n" // Write back U channel
"movhpd %%xmm0,(%2) \n" // Write back V channel
"punpckhdq %%xmm4,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movlpd %%xmm2,(%1) \n"
"movhpd %%xmm2,(%2) \n"
"punpckhdq %%xmm6,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm1,%%xmm8 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movlpd %%xmm1,(%1) \n"
"movhpd %%xmm1,(%2) \n"
"punpckhdq %%xmm5,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm3,%%xmm8 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movlpd %%xmm3,(%1) \n"
"movhpd %%xmm3,(%2) \n"
"punpckhdq %%xmm7,%%xmm8 \n"
"sub $0x8,%3 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst_a), // %1
"+r"(dst_b), // %2
"+r"(width) // %3
: "r"((intptr_t)(src_stride)), // %4
"r"((intptr_t)(dst_stride_a)), // %5
"r"((intptr_t)(dst_stride_b)) // %6
: "memory", "cc",
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
"xmm8", "xmm9"
);
void TransposeUVWx8_SSE2(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%4),%%xmm1 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqa %%xmm8,%%xmm1 \n"
"movdqu (%0),%%xmm2 \n"
"movdqu (%0,%4),%%xmm3 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm8 \n"
"movdqa %%xmm8,%%xmm3 \n"
"movdqu (%0),%%xmm4 \n"
"movdqu (%0,%4),%%xmm5 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm4,%%xmm8 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm8 \n"
"movdqa %%xmm8,%%xmm5 \n"
"movdqu (%0),%%xmm6 \n"
"movdqu (%0,%4),%%xmm7 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm6,%%xmm8 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %4 \n"
"lea 0x10(%0,%4,8),%0 \n"
"punpckhbw %%xmm7,%%xmm8 \n"
"movdqa %%xmm8,%%xmm7 \n"
"neg %4 \n"
// Second round of bit swap.
"movdqa %%xmm0,%%xmm8 \n"
"movdqa %%xmm1,%%xmm9 \n"
"punpckhwd %%xmm2,%%xmm8 \n"
"punpckhwd %%xmm3,%%xmm9 \n"
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm8,%%xmm2 \n"
"movdqa %%xmm9,%%xmm3 \n"
"movdqa %%xmm4,%%xmm8 \n"
"movdqa %%xmm5,%%xmm9 \n"
"punpckhwd %%xmm6,%%xmm8 \n"
"punpckhwd %%xmm7,%%xmm9 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm8,%%xmm6 \n"
"movdqa %%xmm9,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"movdqa %%xmm0,%%xmm8 \n"
"punpckldq %%xmm4,%%xmm0 \n"
"movlpd %%xmm0,(%1) \n" // Write back U channel
"movhpd %%xmm0,(%2) \n" // Write back V channel
"punpckhdq %%xmm4,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movlpd %%xmm2,(%1) \n"
"movhpd %%xmm2,(%2) \n"
"punpckhdq %%xmm6,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm1,%%xmm8 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movlpd %%xmm1,(%1) \n"
"movhpd %%xmm1,(%2) \n"
"punpckhdq %%xmm5,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm3,%%xmm8 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movlpd %%xmm3,(%1) \n"
"movhpd %%xmm3,(%2) \n"
"punpckhdq %%xmm7,%%xmm8 \n"
"sub $0x8,%3 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst_a), // %1
"+r"(dst_b), // %2
"+r"(width) // %3
: "r"((intptr_t)(src_stride)), // %4
"r"((intptr_t)(dst_stride_a)), // %5
"r"((intptr_t)(dst_stride_b)) // %6
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7", "xmm8", "xmm9");
}
#endif // defined(HAS_TRANSPOSEUVWX8_SSE2)
#endif // defined(__x86_64__) || defined(__i386__)
libs/libyuv/source/rotate_mips.cc
-484
View File
@@ -1,484 +0,0 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/basic_types.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#if !defined(LIBYUV_DISABLE_MIPS) && \
defined(__mips_dsp) && (__mips_dsp_rev >= 2) && \
(_MIPS_SIM == _MIPS_SIM_ABI32)
void TransposeWx8_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
"addu $t3, $t2, %[src_stride] \n"
"addu $t5, $t4, %[src_stride] \n"
"addu $t6, $t2, $t4 \n"
"andi $t0, %[dst], 0x3 \n"
"andi $t1, %[dst_stride], 0x3 \n"
"or $t0, $t0, $t1 \n"
"bnez $t0, 11f \n"
" subu $t7, $t9, %[src_stride] \n"
//dst + dst_stride word aligned
"1: \n"
"lbu $t0, 0(%[src]) \n"
"lbux $t1, %[src_stride](%[src]) \n"
"lbux $t8, $t2(%[src]) \n"
"lbux $t9, $t3(%[src]) \n"
"sll $t1, $t1, 16 \n"
"sll $t9, $t9, 16 \n"
"or $t0, $t0, $t1 \n"
"or $t8, $t8, $t9 \n"
"precr.qb.ph $s0, $t8, $t0 \n"
"lbux $t0, $t4(%[src]) \n"
"lbux $t1, $t5(%[src]) \n"
"lbux $t8, $t6(%[src]) \n"
"lbux $t9, $t7(%[src]) \n"
"sll $t1, $t1, 16 \n"
"sll $t9, $t9, 16 \n"
"or $t0, $t0, $t1 \n"
"or $t8, $t8, $t9 \n"
"precr.qb.ph $s1, $t8, $t0 \n"
"sw $s0, 0(%[dst]) \n"
"addiu %[width], -1 \n"
"addiu %[src], 1 \n"
"sw $s1, 4(%[dst]) \n"
"bnez %[width], 1b \n"
" addu %[dst], %[dst], %[dst_stride] \n"
"b 2f \n"
//dst + dst_stride unaligned
"11: \n"
"lbu $t0, 0(%[src]) \n"
"lbux $t1, %[src_stride](%[src]) \n"
"lbux $t8, $t2(%[src]) \n"
"lbux $t9, $t3(%[src]) \n"
"sll $t1, $t1, 16 \n"
"sll $t9, $t9, 16 \n"
"or $t0, $t0, $t1 \n"
"or $t8, $t8, $t9 \n"
"precr.qb.ph $s0, $t8, $t0 \n"
"lbux $t0, $t4(%[src]) \n"
"lbux $t1, $t5(%[src]) \n"
"lbux $t8, $t6(%[src]) \n"
"lbux $t9, $t7(%[src]) \n"
"sll $t1, $t1, 16 \n"
"sll $t9, $t9, 16 \n"
"or $t0, $t0, $t1 \n"
"or $t8, $t8, $t9 \n"
"precr.qb.ph $s1, $t8, $t0 \n"
"swr $s0, 0(%[dst]) \n"
"swl $s0, 3(%[dst]) \n"
"addiu %[width], -1 \n"
"addiu %[src], 1 \n"
"swr $s1, 4(%[dst]) \n"
"swl $s1, 7(%[dst]) \n"
"bnez %[width], 11b \n"
"addu %[dst], %[dst], %[dst_stride] \n"
"2: \n"
".set pop \n"
:[src] "+r" (src),
[dst] "+r" (dst),
[width] "+r" (width)
:[src_stride] "r" (src_stride),
[dst_stride] "r" (dst_stride)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9",
"s0", "s1"
);
}
void TransposeWx8_Fast_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
__asm__ __volatile__ (
".set noat \n"
".set push \n"
".set noreorder \n"
"beqz %[width], 2f \n"
" sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
"addu $t3, $t2, %[src_stride] \n"
"addu $t5, $t4, %[src_stride] \n"
"addu $t6, $t2, $t4 \n"
"srl $AT, %[width], 0x2 \n"
"andi $t0, %[dst], 0x3 \n"
"andi $t1, %[dst_stride], 0x3 \n"
"or $t0, $t0, $t1 \n"
"bnez $t0, 11f \n"
" subu $t7, $t9, %[src_stride] \n"
//dst + dst_stride word aligned
"1: \n"
"lw $t0, 0(%[src]) \n"
"lwx $t1, %[src_stride](%[src]) \n"
"lwx $t8, $t2(%[src]) \n"
"lwx $t9, $t3(%[src]) \n"
// t0 = | 30 | 20 | 10 | 00 |
// t1 = | 31 | 21 | 11 | 01 |
// t8 = | 32 | 22 | 12 | 02 |
// t9 = | 33 | 23 | 13 | 03 |
"precr.qb.ph $s0, $t1, $t0 \n"
"precr.qb.ph $s1, $t9, $t8 \n"
"precrq.qb.ph $s2, $t1, $t0 \n"
"precrq.qb.ph $s3, $t9, $t8 \n"
// s0 = | 21 | 01 | 20 | 00 |
// s1 = | 23 | 03 | 22 | 02 |
// s2 = | 31 | 11 | 30 | 10 |
// s3 = | 33 | 13 | 32 | 12 |
"precr.qb.ph $s4, $s1, $s0 \n"
"precrq.qb.ph $s5, $s1, $s0 \n"
"precr.qb.ph $s6, $s3, $s2 \n"
"precrq.qb.ph $s7, $s3, $s2 \n"
// s4 = | 03 | 02 | 01 | 00 |
// s5 = | 23 | 22 | 21 | 20 |
// s6 = | 13 | 12 | 11 | 10 |
// s7 = | 33 | 32 | 31 | 30 |
"lwx $t0, $t4(%[src]) \n"
"lwx $t1, $t5(%[src]) \n"
"lwx $t8, $t6(%[src]) \n"
"lwx $t9, $t7(%[src]) \n"
// t0 = | 34 | 24 | 14 | 04 |
// t1 = | 35 | 25 | 15 | 05 |
// t8 = | 36 | 26 | 16 | 06 |
// t9 = | 37 | 27 | 17 | 07 |
"precr.qb.ph $s0, $t1, $t0 \n"
"precr.qb.ph $s1, $t9, $t8 \n"
"precrq.qb.ph $s2, $t1, $t0 \n"
"precrq.qb.ph $s3, $t9, $t8 \n"
// s0 = | 25 | 05 | 24 | 04 |
// s1 = | 27 | 07 | 26 | 06 |
// s2 = | 35 | 15 | 34 | 14 |
// s3 = | 37 | 17 | 36 | 16 |
"precr.qb.ph $t0, $s1, $s0 \n"
"precrq.qb.ph $t1, $s1, $s0 \n"
"precr.qb.ph $t8, $s3, $s2 \n"
"precrq.qb.ph $t9, $s3, $s2 \n"
// t0 = | 07 | 06 | 05 | 04 |
// t1 = | 27 | 26 | 25 | 24 |
// t8 = | 17 | 16 | 15 | 14 |
// t9 = | 37 | 36 | 35 | 34 |
"addu $s0, %[dst], %[dst_stride] \n"
"addu $s1, $s0, %[dst_stride] \n"
"addu $s2, $s1, %[dst_stride] \n"
"sw $s4, 0(%[dst]) \n"
"sw $t0, 4(%[dst]) \n"
"sw $s6, 0($s0) \n"
"sw $t8, 4($s0) \n"
"sw $s5, 0($s1) \n"
"sw $t1, 4($s1) \n"
"sw $s7, 0($s2) \n"
"sw $t9, 4($s2) \n"
"addiu $AT, -1 \n"
"addiu %[src], 4 \n"
"bnez $AT, 1b \n"
" addu %[dst], $s2, %[dst_stride] \n"
"b 2f \n"
//dst + dst_stride unaligned
"11: \n"
"lw $t0, 0(%[src]) \n"
"lwx $t1, %[src_stride](%[src]) \n"
"lwx $t8, $t2(%[src]) \n"
"lwx $t9, $t3(%[src]) \n"
// t0 = | 30 | 20 | 10 | 00 |
// t1 = | 31 | 21 | 11 | 01 |
// t8 = | 32 | 22 | 12 | 02 |
// t9 = | 33 | 23 | 13 | 03 |
"precr.qb.ph $s0, $t1, $t0 \n"
"precr.qb.ph $s1, $t9, $t8 \n"
"precrq.qb.ph $s2, $t1, $t0 \n"
"precrq.qb.ph $s3, $t9, $t8 \n"
// s0 = | 21 | 01 | 20 | 00 |
// s1 = | 23 | 03 | 22 | 02 |
// s2 = | 31 | 11 | 30 | 10 |
// s3 = | 33 | 13 | 32 | 12 |
"precr.qb.ph $s4, $s1, $s0 \n"
"precrq.qb.ph $s5, $s1, $s0 \n"
"precr.qb.ph $s6, $s3, $s2 \n"
"precrq.qb.ph $s7, $s3, $s2 \n"
// s4 = | 03 | 02 | 01 | 00 |
// s5 = | 23 | 22 | 21 | 20 |
// s6 = | 13 | 12 | 11 | 10 |
// s7 = | 33 | 32 | 31 | 30 |
"lwx $t0, $t4(%[src]) \n"
"lwx $t1, $t5(%[src]) \n"
"lwx $t8, $t6(%[src]) \n"
"lwx $t9, $t7(%[src]) \n"
// t0 = | 34 | 24 | 14 | 04 |
// t1 = | 35 | 25 | 15 | 05 |
// t8 = | 36 | 26 | 16 | 06 |
// t9 = | 37 | 27 | 17 | 07 |
"precr.qb.ph $s0, $t1, $t0 \n"
"precr.qb.ph $s1, $t9, $t8 \n"
"precrq.qb.ph $s2, $t1, $t0 \n"
"precrq.qb.ph $s3, $t9, $t8 \n"
// s0 = | 25 | 05 | 24 | 04 |
// s1 = | 27 | 07 | 26 | 06 |
// s2 = | 35 | 15 | 34 | 14 |
// s3 = | 37 | 17 | 36 | 16 |
"precr.qb.ph $t0, $s1, $s0 \n"
"precrq.qb.ph $t1, $s1, $s0 \n"
"precr.qb.ph $t8, $s3, $s2 \n"
"precrq.qb.ph $t9, $s3, $s2 \n"
// t0 = | 07 | 06 | 05 | 04 |
// t1 = | 27 | 26 | 25 | 24 |
// t8 = | 17 | 16 | 15 | 14 |
// t9 = | 37 | 36 | 35 | 34 |
"addu $s0, %[dst], %[dst_stride] \n"
"addu $s1, $s0, %[dst_stride] \n"
"addu $s2, $s1, %[dst_stride] \n"
"swr $s4, 0(%[dst]) \n"
"swl $s4, 3(%[dst]) \n"
"swr $t0, 4(%[dst]) \n"
"swl $t0, 7(%[dst]) \n"
"swr $s6, 0($s0) \n"
"swl $s6, 3($s0) \n"
"swr $t8, 4($s0) \n"
"swl $t8, 7($s0) \n"
"swr $s5, 0($s1) \n"
"swl $s5, 3($s1) \n"
"swr $t1, 4($s1) \n"
"swl $t1, 7($s1) \n"
"swr $s7, 0($s2) \n"
"swl $s7, 3($s2) \n"
"swr $t9, 4($s2) \n"
"swl $t9, 7($s2) \n"
"addiu $AT, -1 \n"
"addiu %[src], 4 \n"
"bnez $AT, 11b \n"
" addu %[dst], $s2, %[dst_stride] \n"
"2: \n"
".set pop \n"
".set at \n"
:[src] "+r" (src),
[dst] "+r" (dst),
[width] "+r" (width)
:[src_stride] "r" (src_stride),
[dst_stride] "r" (dst_stride)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9",
"s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7"
);
}
void TransposeUVWx8_DSPR2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"beqz %[width], 2f \n"
" sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
"addu $t3, $t2, %[src_stride] \n"
"addu $t5, $t4, %[src_stride] \n"
"addu $t6, $t2, $t4 \n"
"subu $t7, $t9, %[src_stride] \n"
"srl $t1, %[width], 1 \n"
// check word aligment for dst_a, dst_b, dst_stride_a and dst_stride_b
"andi $t0, %[dst_a], 0x3 \n"
"andi $t8, %[dst_b], 0x3 \n"
"or $t0, $t0, $t8 \n"
"andi $t8, %[dst_stride_a], 0x3 \n"
"andi $s5, %[dst_stride_b], 0x3 \n"
"or $t8, $t8, $s5 \n"
"or $t0, $t0, $t8 \n"
"bnez $t0, 11f \n"
" nop \n"
// dst + dst_stride word aligned (both, a & b dst addresses)
"1: \n"
"lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
"lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
"addu $s5, %[dst_a], %[dst_stride_a] \n"
"lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
"lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
"addu $s6, %[dst_b], %[dst_stride_b] \n"
"precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
"precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
"precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
"sll $t0, $t0, 16 \n"
"packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
"sll $t9, $t9, 16 \n"
"packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
"sw $s3, 0($s5) \n"
"sw $s4, 0($s6) \n"
"precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
"lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
"lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
"lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
"lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
"sw $s3, 0(%[dst_a]) \n"
"sw $s4, 0(%[dst_b]) \n"
"precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
"precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
"precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
"sll $t0, $t0, 16 \n"
"packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
"sll $t9, $t9, 16 \n"
"packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
"sw $s3, 4($s5) \n"
"sw $s4, 4($s6) \n"
"precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
"addiu %[src], 4 \n"
"addiu $t1, -1 \n"
"sll $t0, %[dst_stride_a], 1 \n"
"sll $t8, %[dst_stride_b], 1 \n"
"sw $s3, 4(%[dst_a]) \n"
"sw $s4, 4(%[dst_b]) \n"
"addu %[dst_a], %[dst_a], $t0 \n"
"bnez $t1, 1b \n"
" addu %[dst_b], %[dst_b], $t8 \n"
"b 2f \n"
" nop \n"
// dst_a or dst_b or dst_stride_a or dst_stride_b not word aligned
"11: \n"
"lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
"lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
"addu $s5, %[dst_a], %[dst_stride_a] \n"
"lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
"lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
"addu $s6, %[dst_b], %[dst_stride_b] \n"
"precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
"precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
"precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
"sll $t0, $t0, 16 \n"
"packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
"sll $t9, $t9, 16 \n"
"packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
"swr $s3, 0($s5) \n"
"swl $s3, 3($s5) \n"
"swr $s4, 0($s6) \n"
"swl $s4, 3($s6) \n"
"precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
"lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
"lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
"lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
"lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
"swr $s3, 0(%[dst_a]) \n"
"swl $s3, 3(%[dst_a]) \n"
"swr $s4, 0(%[dst_b]) \n"
"swl $s4, 3(%[dst_b]) \n"
"precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
"precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
"precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
"sll $t0, $t0, 16 \n"
"packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
"sll $t9, $t9, 16 \n"
"packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
"swr $s3, 4($s5) \n"
"swl $s3, 7($s5) \n"
"swr $s4, 4($s6) \n"
"swl $s4, 7($s6) \n"
"precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
"addiu %[src], 4 \n"
"addiu $t1, -1 \n"
"sll $t0, %[dst_stride_a], 1 \n"
"sll $t8, %[dst_stride_b], 1 \n"
"swr $s3, 4(%[dst_a]) \n"
"swl $s3, 7(%[dst_a]) \n"
"swr $s4, 4(%[dst_b]) \n"
"swl $s4, 7(%[dst_b]) \n"
"addu %[dst_a], %[dst_a], $t0 \n"
"bnez $t1, 11b \n"
" addu %[dst_b], %[dst_b], $t8 \n"
"2: \n"
".set pop \n"
: [src] "+r" (src),
[dst_a] "+r" (dst_a),
[dst_b] "+r" (dst_b),
[width] "+r" (width),
[src_stride] "+r" (src_stride)
: [dst_stride_a] "r" (dst_stride_a),
[dst_stride_b] "r" (dst_stride_b)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9",
"s0", "s1", "s2", "s3",
"s4", "s5", "s6"
);
}
#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
libs/libyuv/source/rotate_mmi.cc
+291
View File
@@ -0,0 +1,291 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for Mips MMI.
#if !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
void TransposeWx8_MMI(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
uint64_t tmp7, tmp8, tmp9, tmp10, tmp11, tmp12, tmp13;
uint8_t* src_tmp = nullptr;
__asm__ volatile(
"1: \n\t"
"ldc1 %[tmp12], 0x00(%[src]) \n\t"
"dadd %[src_tmp], %[src], %[src_stride] \n\t"
"ldc1 %[tmp13], 0x00(%[src_tmp]) \n\t"
/* tmp0 = (00 10 01 11 02 12 03 13) */
"punpcklbh %[tmp0], %[tmp12], %[tmp13] \n\t"
/* tmp1 = (04 14 05 15 06 16 07 17) */
"punpckhbh %[tmp1], %[tmp12], %[tmp13] \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
"ldc1 %[tmp12], 0x00(%[src_tmp]) \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
"ldc1 %[tmp13], 0x00(%[src_tmp]) \n\t"
/* tmp2 = (20 30 21 31 22 32 23 33) */
"punpcklbh %[tmp2], %[tmp12], %[tmp13] \n\t"
/* tmp3 = (24 34 25 35 26 36 27 37) */
"punpckhbh %[tmp3], %[tmp12], %[tmp13] \n\t"
/* tmp4 = (00 10 20 30 01 11 21 31) */
"punpcklhw %[tmp4], %[tmp0], %[tmp2] \n\t"
/* tmp5 = (02 12 22 32 03 13 23 33) */
"punpckhhw %[tmp5], %[tmp0], %[tmp2] \n\t"
/* tmp6 = (04 14 24 34 05 15 25 35) */
"punpcklhw %[tmp6], %[tmp1], %[tmp3] \n\t"
/* tmp7 = (06 16 26 36 07 17 27 37) */
"punpckhhw %[tmp7], %[tmp1], %[tmp3] \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
"ldc1 %[tmp12], 0x00(%[src_tmp]) \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
"ldc1 %[tmp13], 0x00(%[src_tmp]) \n\t"
/* tmp0 = (40 50 41 51 42 52 43 53) */
"punpcklbh %[tmp0], %[tmp12], %[tmp13] \n\t"
/* tmp1 = (44 54 45 55 46 56 47 57) */
"punpckhbh %[tmp1], %[tmp12], %[tmp13] \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
"ldc1 %[tmp12], 0x00(%[src_tmp]) \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
"ldc1 %[tmp13], 0x00(%[src_tmp]) \n\t"
/* tmp2 = (60 70 61 71 62 72 63 73) */
"punpcklbh %[tmp2], %[tmp12], %[tmp13] \n\t"
/* tmp3 = (64 74 65 75 66 76 67 77) */
"punpckhbh %[tmp3], %[tmp12], %[tmp13] \n\t"
/* tmp8 = (40 50 60 70 41 51 61 71) */
"punpcklhw %[tmp8], %[tmp0], %[tmp2] \n\t"
/* tmp9 = (42 52 62 72 43 53 63 73) */
"punpckhhw %[tmp9], %[tmp0], %[tmp2] \n\t"
/* tmp10 = (44 54 64 74 45 55 65 75) */
"punpcklhw %[tmp10], %[tmp1], %[tmp3] \n\t"
/* tmp11 = (46 56 66 76 47 57 67 77) */
"punpckhhw %[tmp11], %[tmp1], %[tmp3] \n\t"
/* tmp0 = (00 10 20 30 40 50 60 70) */
"punpcklwd %[tmp0], %[tmp4], %[tmp8] \n\t"
/* tmp1 = (01 11 21 31 41 51 61 71) */
"punpckhwd %[tmp1], %[tmp4], %[tmp8] \n\t"
"gssdlc1 %[tmp0], 0x07(%[dst]) \n\t"
"gssdrc1 %[tmp0], 0x00(%[dst]) \n\t"
"dadd %[dst], %[dst], %[dst_stride] \n\t"
"gssdlc1 %[tmp1], 0x07(%[dst]) \n\t"
"gssdrc1 %[tmp1], 0x00(%[dst]) \n\t"
/* tmp0 = (02 12 22 32 42 52 62 72) */
"punpcklwd %[tmp0], %[tmp5], %[tmp9] \n\t"
/* tmp1 = (03 13 23 33 43 53 63 73) */
"punpckhwd %[tmp1], %[tmp5], %[tmp9] \n\t"
"dadd %[dst], %[dst], %[dst_stride] \n\t"
"gssdlc1 %[tmp0], 0x07(%[dst]) \n\t"
"gssdrc1 %[tmp0], 0x00(%[dst]) \n\t"
"dadd %[dst], %[dst], %[dst_stride] \n\t"
"gssdlc1 %[tmp1], 0x07(%[dst]) \n\t"
"gssdrc1 %[tmp1], 0x00(%[dst]) \n\t"
/* tmp0 = (04 14 24 34 44 54 64 74) */
"punpcklwd %[tmp0], %[tmp6], %[tmp10] \n\t"
/* tmp1 = (05 15 25 35 45 55 65 75) */
"punpckhwd %[tmp1], %[tmp6], %[tmp10] \n\t"
"dadd %[dst], %[dst], %[dst_stride] \n\t"
"gssdlc1 %[tmp0], 0x07(%[dst]) \n\t"
"gssdrc1 %[tmp0], 0x00(%[dst]) \n\t"
"dadd %[dst], %[dst], %[dst_stride] \n\t"
"gssdlc1 %[tmp1], 0x07(%[dst]) \n\t"
"gssdrc1 %[tmp1], 0x00(%[dst]) \n\t"
/* tmp0 = (06 16 26 36 46 56 66 76) */
"punpcklwd %[tmp0], %[tmp7], %[tmp11] \n\t"
/* tmp1 = (07 17 27 37 47 57 67 77) */
"punpckhwd %[tmp1], %[tmp7], %[tmp11] \n\t"
"dadd %[dst], %[dst], %[dst_stride] \n\t"
"gssdlc1 %[tmp0], 0x07(%[dst]) \n\t"
"gssdrc1 %[tmp0], 0x00(%[dst]) \n\t"
"dadd %[dst], %[dst], %[dst_stride] \n\t"
"gssdlc1 %[tmp1], 0x07(%[dst]) \n\t"
"gssdrc1 %[tmp1], 0x00(%[dst]) \n\t"
"dadd %[dst], %[dst], %[dst_stride] \n\t"
"daddi %[src], %[src], 0x08 \n\t"
"daddi %[width], %[width], -0x08 \n\t"
"bnez %[width], 1b \n\t"
: [tmp0] "=&f"(tmp0), [tmp1] "=&f"(tmp1), [tmp2] "=&f"(tmp2),
[tmp3] "=&f"(tmp3), [tmp4] "=&f"(tmp4), [tmp5] "=&f"(tmp5),
[tmp6] "=&f"(tmp6), [tmp7] "=&f"(tmp7), [tmp8] "=&f"(tmp8),
[tmp9] "=&f"(tmp9), [tmp10] "=&f"(tmp10), [tmp11] "=&f"(tmp11),
[tmp12] "=&f"(tmp12), [tmp13] "=&f"(tmp13), [dst] "+&r"(dst),
[src_tmp] "+&r"(src_tmp)
: [src] "r"(src), [width] "r"(width), [src_stride] "r"(src_stride),
[dst_stride] "r"(dst_stride)
: "memory");
}
void TransposeUVWx8_MMI(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
uint64_t tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
uint64_t tmp7, tmp8, tmp9, tmp10, tmp11, tmp12, tmp13;
uint8_t* src_tmp = nullptr;
__asm__ volatile(
"1: \n\t"
/* tmp12 = (u00 v00 u01 v01 u02 v02 u03 v03) */
"ldc1 %[tmp12], 0x00(%[src]) \n\t"
"dadd %[src_tmp], %[src], %[src_stride] \n\t"
/* tmp13 = (u10 v10 u11 v11 u12 v12 u13 v13) */
"ldc1 %[tmp13], 0x00(%[src_tmp]) \n\t"
/* tmp0 = (u00 u10 v00 v10 u01 u11 v01 v11) */
"punpcklbh %[tmp0], %[tmp12], %[tmp13] \n\t"
/* tmp1 = (u02 u12 v02 v12 u03 u13 v03 v13) */
"punpckhbh %[tmp1], %[tmp12], %[tmp13] \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
/* tmp12 = (u20 v20 u21 v21 u22 v22 u23 v23) */
"ldc1 %[tmp12], 0x00(%[src_tmp]) \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
/* tmp13 = (u30 v30 u31 v31 u32 v32 u33 v33) */
"ldc1 %[tmp13], 0x00(%[src_tmp]) \n\t"
/* tmp2 = (u20 u30 v20 v30 u21 u31 v21 v31) */
"punpcklbh %[tmp2], %[tmp12], %[tmp13] \n\t"
/* tmp3 = (u22 u32 v22 v32 u23 u33 v23 v33) */
"punpckhbh %[tmp3], %[tmp12], %[tmp13] \n\t"
/* tmp4 = (u00 u10 u20 u30 v00 v10 v20 v30) */
"punpcklhw %[tmp4], %[tmp0], %[tmp2] \n\t"
/* tmp5 = (u01 u11 u21 u31 v01 v11 v21 v31) */
"punpckhhw %[tmp5], %[tmp0], %[tmp2] \n\t"
/* tmp6 = (u02 u12 u22 u32 v02 v12 v22 v32) */
"punpcklhw %[tmp6], %[tmp1], %[tmp3] \n\t"
/* tmp7 = (u03 u13 u23 u33 v03 v13 v23 v33) */
"punpckhhw %[tmp7], %[tmp1], %[tmp3] \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
/* tmp12 = (u40 v40 u41 v41 u42 v42 u43 v43) */
"ldc1 %[tmp12], 0x00(%[src_tmp]) \n\t"
/* tmp13 = (u50 v50 u51 v51 u52 v52 u53 v53) */
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
"ldc1 %[tmp13], 0x00(%[src_tmp]) \n\t"
/* tmp0 = (u40 u50 v40 v50 u41 u51 v41 v51) */
"punpcklbh %[tmp0], %[tmp12], %[tmp13] \n\t"
/* tmp1 = (u42 u52 v42 v52 u43 u53 v43 v53) */
"punpckhbh %[tmp1], %[tmp12], %[tmp13] \n\t"
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
/* tmp12 = (u60 v60 u61 v61 u62 v62 u63 v63) */
"ldc1 %[tmp12], 0x00(%[src_tmp]) \n\t"
/* tmp13 = (u70 v70 u71 v71 u72 v72 u73 v73) */
"dadd %[src_tmp], %[src_tmp], %[src_stride] \n\t"
"ldc1 %[tmp13], 0x00(%[src_tmp]) \n\t"
/* tmp2 = (u60 u70 v60 v70 u61 u71 v61 v71) */
"punpcklbh %[tmp2], %[tmp12], %[tmp13] \n\t"
/* tmp3 = (u62 u72 v62 v72 u63 u73 v63 v73) */
"punpckhbh %[tmp3], %[tmp12], %[tmp13] \n\t"
/* tmp8 = (u40 u50 u60 u70 v40 v50 v60 v70) */
"punpcklhw %[tmp8], %[tmp0], %[tmp2] \n\t"
/* tmp9 = (u41 u51 u61 u71 v41 v51 v61 v71) */
"punpckhhw %[tmp9], %[tmp0], %[tmp2] \n\t"
/* tmp10 = (u42 u52 u62 u72 v42 v52 v62 v72) */
"punpcklhw %[tmp10], %[tmp1], %[tmp3] \n\t"
/* tmp11 = (u43 u53 u63 u73 v43 v53 v63 v73) */
"punpckhhw %[tmp11], %[tmp1], %[tmp3] \n\t"
/* tmp0 = (u00 u10 u20 u30 u40 u50 u60 u70) */
"punpcklwd %[tmp0], %[tmp4], %[tmp8] \n\t"
/* tmp1 = (v00 v10 v20 v30 v40 v50 v60 v70) */
"punpckhwd %[tmp1], %[tmp4], %[tmp8] \n\t"
"gssdlc1 %[tmp0], 0x07(%[dst_a]) \n\t"
"gssdrc1 %[tmp0], 0x00(%[dst_a]) \n\t"
"gssdlc1 %[tmp1], 0x07(%[dst_b]) \n\t"
"gssdrc1 %[tmp1], 0x00(%[dst_b]) \n\t"
/* tmp0 = (u01 u11 u21 u31 u41 u51 u61 u71) */
"punpcklwd %[tmp0], %[tmp5], %[tmp9] \n\t"
/* tmp1 = (v01 v11 v21 v31 v41 v51 v61 v71) */
"punpckhwd %[tmp1], %[tmp5], %[tmp9] \n\t"
"dadd %[dst_a], %[dst_a], %[dst_stride_a] \n\t"
"gssdlc1 %[tmp0], 0x07(%[dst_a]) \n\t"
"gssdrc1 %[tmp0], 0x00(%[dst_a]) \n\t"
"dadd %[dst_b], %[dst_b], %[dst_stride_b] \n\t"
"gssdlc1 %[tmp1], 0x07(%[dst_b]) \n\t"
"gssdrc1 %[tmp1], 0x00(%[dst_b]) \n\t"
/* tmp0 = (u02 u12 u22 u32 u42 u52 u62 u72) */
"punpcklwd %[tmp0], %[tmp6], %[tmp10] \n\t"
/* tmp1 = (v02 v12 v22 v32 v42 v52 v62 v72) */
"punpckhwd %[tmp1], %[tmp6], %[tmp10] \n\t"
"dadd %[dst_a], %[dst_a], %[dst_stride_a] \n\t"
"gssdlc1 %[tmp0], 0x07(%[dst_a]) \n\t"
"gssdrc1 %[tmp0], 0x00(%[dst_a]) \n\t"
"dadd %[dst_b], %[dst_b], %[dst_stride_b] \n\t"
"gssdlc1 %[tmp1], 0x07(%[dst_b]) \n\t"
"gssdrc1 %[tmp1], 0x00(%[dst_b]) \n\t"
/* tmp0 = (u03 u13 u23 u33 u43 u53 u63 u73) */
"punpcklwd %[tmp0], %[tmp7], %[tmp11] \n\t"
/* tmp1 = (v03 v13 v23 v33 v43 v53 v63 v73) */
"punpckhwd %[tmp1], %[tmp7], %[tmp11] \n\t"
"dadd %[dst_a], %[dst_a], %[dst_stride_a] \n\t"
"gssdlc1 %[tmp0], 0x07(%[dst_a]) \n\t"
"gssdrc1 %[tmp0], 0x00(%[dst_a]) \n\t"
"dadd %[dst_b], %[dst_b], %[dst_stride_b] \n\t"
"gssdlc1 %[tmp1], 0x07(%[dst_b]) \n\t"
"gssdrc1 %[tmp1], 0x00(%[dst_b]) \n\t"
"dadd %[dst_a], %[dst_a], %[dst_stride_a] \n\t"
"dadd %[dst_b], %[dst_b], %[dst_stride_b] \n\t"
"daddiu %[src], %[src], 0x08 \n\t"
"daddi %[width], %[width], -0x04 \n\t"
"bnez %[width], 1b \n\t"
: [tmp0] "=&f"(tmp0), [tmp1] "=&f"(tmp1), [tmp2] "=&f"(tmp2),
[tmp3] "=&f"(tmp3), [tmp4] "=&f"(tmp4), [tmp5] "=&f"(tmp5),
[tmp6] "=&f"(tmp6), [tmp7] "=&f"(tmp7), [tmp8] "=&f"(tmp8),
[tmp9] "=&f"(tmp9), [tmp10] "=&f"(tmp10), [tmp11] "=&f"(tmp11),
[tmp12] "=&f"(tmp12), [tmp13] "=&f"(tmp13), [dst_a] "+&r"(dst_a),
[dst_b] "+&r"(dst_b), [src_tmp] "+&r"(src_tmp)
: [src] "r"(src), [width] "r"(width), [dst_stride_a] "r"(dst_stride_a),
[dst_stride_b] "r"(dst_stride_b), [src_stride] "r"(src_stride)
: "memory");
}
#endif // !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
libs/libyuv/source/rotate_msa.cc
+250
View File
@@ -0,0 +1,250 @@
/*
* Copyright 2016 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
// This module is for GCC MSA
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#include "libyuv/macros_msa.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#define ILVRL_B(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
out0 = (v16u8)__msa_ilvr_b((v16i8)in1, (v16i8)in0); \
out1 = (v16u8)__msa_ilvl_b((v16i8)in1, (v16i8)in0); \
out2 = (v16u8)__msa_ilvr_b((v16i8)in3, (v16i8)in2); \
out3 = (v16u8)__msa_ilvl_b((v16i8)in3, (v16i8)in2); \
}
#define ILVRL_H(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
out0 = (v16u8)__msa_ilvr_h((v8i16)in1, (v8i16)in0); \
out1 = (v16u8)__msa_ilvl_h((v8i16)in1, (v8i16)in0); \
out2 = (v16u8)__msa_ilvr_h((v8i16)in3, (v8i16)in2); \
out3 = (v16u8)__msa_ilvl_h((v8i16)in3, (v8i16)in2); \
}
#define ILVRL_W(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
out0 = (v16u8)__msa_ilvr_w((v4i32)in1, (v4i32)in0); \
out1 = (v16u8)__msa_ilvl_w((v4i32)in1, (v4i32)in0); \
out2 = (v16u8)__msa_ilvr_w((v4i32)in3, (v4i32)in2); \
out3 = (v16u8)__msa_ilvl_w((v4i32)in3, (v4i32)in2); \
}
#define ILVRL_D(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
out0 = (v16u8)__msa_ilvr_d((v2i64)in1, (v2i64)in0); \
out1 = (v16u8)__msa_ilvl_d((v2i64)in1, (v2i64)in0); \
out2 = (v16u8)__msa_ilvr_d((v2i64)in3, (v2i64)in2); \
out3 = (v16u8)__msa_ilvl_d((v2i64)in3, (v2i64)in2); \
}
void TransposeWx16_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
TransposeWx8_C(src, src_stride, dst, dst_stride, width);
TransposeWx8_C((src + 8 * src_stride), src_stride, (dst + 8), dst_stride,
width);
}
void TransposeUVWx16_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
TransposeUVWx8_C(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width);
TransposeUVWx8_C((src + 8 * src_stride), src_stride, (dst_a + 8),
dst_stride_a, (dst_b + 8), dst_stride_b, width);
}
void TransposeWx16_MSA(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
int x;
const uint8_t* s;
v16u8 src0, src1, src2, src3, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
v16u8 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
v16u8 res0, res1, res2, res3, res4, res5, res6, res7, res8, res9;
for (x = 0; x < width; x += 16) {
s = src;
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
ILVRL_W(reg0, reg4, reg1, reg5, res0, res1, res2, res3);
ILVRL_W(reg2, reg6, reg3, reg7, res4, res5, res6, res7);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
res8 = (v16u8)__msa_ilvr_w((v4i32)reg4, (v4i32)reg0);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg4, (v4i32)reg0);
ILVRL_D(res0, res8, res1, res9, dst0, dst1, dst2, dst3);
ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
dst += dst_stride * 4;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg5, (v4i32)reg1);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg5, (v4i32)reg1);
ILVRL_D(res2, res8, res3, res9, dst0, dst1, dst2, dst3);
ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
dst += dst_stride * 4;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg6, (v4i32)reg2);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg6, (v4i32)reg2);
ILVRL_D(res4, res8, res5, res9, dst0, dst1, dst2, dst3);
ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
dst += dst_stride * 4;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg7, (v4i32)reg3);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg7, (v4i32)reg3);
ILVRL_D(res6, res8, res7, res9, dst0, dst1, dst2, dst3);
ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
src += 16;
dst += dst_stride * 4;
}
}
void TransposeUVWx16_MSA(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
int x;
const uint8_t* s;
v16u8 src0, src1, src2, src3, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
v16u8 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
v16u8 res0, res1, res2, res3, res4, res5, res6, res7, res8, res9;
for (x = 0; x < width; x += 8) {
s = src;
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
ILVRL_W(reg0, reg4, reg1, reg5, res0, res1, res2, res3);
ILVRL_W(reg2, reg6, reg3, reg7, res4, res5, res6, res7);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
res8 = (v16u8)__msa_ilvr_w((v4i32)reg4, (v4i32)reg0);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg4, (v4i32)reg0);
ILVRL_D(res0, res8, res1, res9, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst2, dst_a, dst_stride_a);
ST_UB2(dst1, dst3, dst_b, dst_stride_b);
dst_a += dst_stride_a * 2;
dst_b += dst_stride_b * 2;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg5, (v4i32)reg1);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg5, (v4i32)reg1);
ILVRL_D(res2, res8, res3, res9, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst2, dst_a, dst_stride_a);
ST_UB2(dst1, dst3, dst_b, dst_stride_b);
dst_a += dst_stride_a * 2;
dst_b += dst_stride_b * 2;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg6, (v4i32)reg2);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg6, (v4i32)reg2);
ILVRL_D(res4, res8, res5, res9, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst2, dst_a, dst_stride_a);
ST_UB2(dst1, dst3, dst_b, dst_stride_b);
dst_a += dst_stride_a * 2;
dst_b += dst_stride_b * 2;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg7, (v4i32)reg3);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg7, (v4i32)reg3);
ILVRL_D(res6, res8, res7, res9, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst2, dst_a, dst_stride_a);
ST_UB2(dst1, dst3, dst_b, dst_stride_b);
src += 16;
dst_a += dst_stride_a * 2;
dst_b += dst_stride_b * 2;
}
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
libs/libyuv/source/rotate_neon.cc
+225 -342
View File
@@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#include "libyuv/basic_types.h"
@@ -21,38 +21,32 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__) && \
!defined(__aarch64__)
static uvec8 kVTbl4x4Transpose =
{ 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15 };
static const uvec8 kVTbl4x4Transpose = {0, 4, 8, 12, 1, 5, 9, 13,
2, 6, 10, 14, 3, 7, 11, 15};
void TransposeWx8_NEON(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
void TransposeWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
const uint8* src_temp;
asm volatile (
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %5, #8 \n"
const uint8_t* src_temp;
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %5, #8 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"vld1.8 {d0}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d1}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d2}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d3}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d4}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d5}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d6}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d7}, [%0] \n"
"vtrn.8 d1, d0 \n"
@@ -77,21 +71,13 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.8 {d1}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d0}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d3}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d2}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d5}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d4}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d7}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d6}, [%0] \n"
"add %1, #8 \n" // src += 8
@@ -99,180 +85,138 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
"subs %5, #8 \n" // w -= 8
"bge 1b \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %5, #8 \n"
"beq 4f \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %5, #8 \n"
"beq 4f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %5, #2 \n"
"blt 3f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %5, #2 \n"
"blt 3f \n"
"cmp %5, #4 \n"
"blt 2f \n"
"cmp %5, #4 \n"
"blt 2f \n"
// 4x8 block
"mov %0, %1 \n"
MEMACCESS(0)
"vld1.32 {d0[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d0[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d1[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d1[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d2[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d2[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d3[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d3[1]}, [%0] \n"
// 4x8 block
"mov %0, %1 \n"
"vld1.32 {d0[0]}, [%0], %2 \n"
"vld1.32 {d0[1]}, [%0], %2 \n"
"vld1.32 {d1[0]}, [%0], %2 \n"
"vld1.32 {d1[1]}, [%0], %2 \n"
"vld1.32 {d2[0]}, [%0], %2 \n"
"vld1.32 {d2[1]}, [%0], %2 \n"
"vld1.32 {d3[0]}, [%0], %2 \n"
"vld1.32 {d3[1]}, [%0] \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(6)
"vld1.8 {q3}, [%6] \n"
"vld1.8 {q3}, [%6] \n"
"vtbl.8 d4, {d0, d1}, d6 \n"
"vtbl.8 d5, {d0, d1}, d7 \n"
"vtbl.8 d0, {d2, d3}, d6 \n"
"vtbl.8 d1, {d2, d3}, d7 \n"
"vtbl.8 d4, {d0, d1}, d6 \n"
"vtbl.8 d5, {d0, d1}, d7 \n"
"vtbl.8 d0, {d2, d3}, d6 \n"
"vtbl.8 d1, {d2, d3}, d7 \n"
// TODO(frkoenig): Rework shuffle above to
// write out with 4 instead of 8 writes.
MEMACCESS(0)
"vst1.32 {d4[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d4[1]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d5[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d5[1]}, [%0] \n"
// TODO(frkoenig): Rework shuffle above to
// write out with 4 instead of 8 writes.
"vst1.32 {d4[0]}, [%0], %4 \n"
"vst1.32 {d4[1]}, [%0], %4 \n"
"vst1.32 {d5[0]}, [%0], %4 \n"
"vst1.32 {d5[1]}, [%0] \n"
"add %0, %3, #4 \n"
MEMACCESS(0)
"vst1.32 {d0[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d0[1]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d1[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d1[1]}, [%0] \n"
"add %0, %3, #4 \n"
"vst1.32 {d0[0]}, [%0], %4 \n"
"vst1.32 {d0[1]}, [%0], %4 \n"
"vst1.32 {d1[0]}, [%0], %4 \n"
"vst1.32 {d1[1]}, [%0] \n"
"add %1, #4 \n" // src += 4
"add %3, %3, %4, lsl #2 \n" // dst += 4 * dst_stride
"subs %5, #4 \n" // w -= 4
"beq 4f \n"
"add %1, #4 \n" // src += 4
"add %3, %3, %4, lsl #2 \n" // dst += 4 * dst_stride
"subs %5, #4 \n" // w -= 4
"beq 4f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %5, #2 \n"
"blt 3f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %5, #2 \n"
"blt 3f \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"vld1.16 {d0[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d1[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d0[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d1[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d0[2]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d1[2]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d0[3]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d1[3]}, [%0] \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
"vld1.16 {d0[0]}, [%0], %2 \n"
"vld1.16 {d1[0]}, [%0], %2 \n"
"vld1.16 {d0[1]}, [%0], %2 \n"
"vld1.16 {d1[1]}, [%0], %2 \n"
"vld1.16 {d0[2]}, [%0], %2 \n"
"vld1.16 {d1[2]}, [%0], %2 \n"
"vld1.16 {d0[3]}, [%0], %2 \n"
"vld1.16 {d1[3]}, [%0] \n"
"vtrn.8 d0, d1 \n"
"vtrn.8 d0, d1 \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.64 {d0}, [%0], %4 \n"
MEMACCESS(0)
"vst1.64 {d1}, [%0] \n"
"vst1.64 {d0}, [%0], %4 \n"
"vst1.64 {d1}, [%0] \n"
"add %1, #2 \n" // src += 2
"add %3, %3, %4, lsl #1 \n" // dst += 2 * dst_stride
"subs %5, #2 \n" // w -= 2
"beq 4f \n"
"add %1, #2 \n" // src += 2
"add %3, %3, %4, lsl #1 \n" // dst += 2 * dst_stride
"subs %5, #2 \n" // w -= 2
"beq 4f \n"
// 1x8 block
"3: \n"
MEMACCESS(1)
"vld1.8 {d0[0]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[1]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[2]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[3]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[4]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[5]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[6]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[7]}, [%1] \n"
// 1x8 block
"3: \n"
"vld1.8 {d0[0]}, [%1], %2 \n"
"vld1.8 {d0[1]}, [%1], %2 \n"
"vld1.8 {d0[2]}, [%1], %2 \n"
"vld1.8 {d0[3]}, [%1], %2 \n"
"vld1.8 {d0[4]}, [%1], %2 \n"
"vld1.8 {d0[5]}, [%1], %2 \n"
"vld1.8 {d0[6]}, [%1], %2 \n"
"vld1.8 {d0[7]}, [%1] \n"
MEMACCESS(3)
"vst1.64 {d0}, [%3] \n"
"vst1.64 {d0}, [%3] \n"
"4: \n"
"4: \n"
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(src_stride), // %2
"+r"(dst), // %3
"+r"(dst_stride), // %4
"+r"(width) // %5
: "r"(&kVTbl4x4Transpose) // %6
: "memory", "cc", "q0", "q1", "q2", "q3"
);
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(src_stride), // %2
"+r"(dst), // %3
"+r"(dst_stride), // %4
"+r"(width) // %5
: "r"(&kVTbl4x4Transpose) // %6
: "memory", "cc", "q0", "q1", "q2", "q3");
}
static uvec8 kVTbl4x4TransposeDi =
{ 0, 8, 1, 9, 2, 10, 3, 11, 4, 12, 5, 13, 6, 14, 7, 15 };
static const uvec8 kVTbl4x4TransposeDi = {0, 8, 1, 9, 2, 10, 3, 11,
4, 12, 5, 13, 6, 14, 7, 15};
void TransposeUVWx8_NEON(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
void TransposeUVWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
const uint8* src_temp;
asm volatile (
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %7, #8 \n"
const uint8_t* src_temp;
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %7, #8 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"vld2.8 {d0, d1}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d2, d3}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d4, d5}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d6, d7}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d16, d17}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d18, d19}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d20, d21}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d22, d23}, [%0] \n"
"vtrn.8 q1, q0 \n"
@@ -301,40 +245,24 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.8 {d2}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d0}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d6}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d4}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d18}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d16}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d22}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d20}, [%0] \n"
"mov %0, %5 \n"
MEMACCESS(0)
"vst1.8 {d3}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d1}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d7}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d5}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d19}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d17}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d23}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d21}, [%0] \n"
"add %1, #8*2 \n" // src += 8*2
@@ -343,187 +271,142 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
"subs %7, #8 \n" // w -= 8
"bge 1b \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %7, #8 \n"
"beq 4f \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %7, #8 \n"
"beq 4f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %7, #2 \n"
"blt 3f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %7, #2 \n"
"blt 3f \n"
"cmp %7, #4 \n"
"blt 2f \n"
"cmp %7, #4 \n"
"blt 2f \n"
// TODO(frkoenig): Clean this up
// 4x8 block
"mov %0, %1 \n"
MEMACCESS(0)
"vld1.64 {d0}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d1}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d2}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d3}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d4}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d5}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d6}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d7}, [%0] \n"
// TODO(frkoenig): Clean this up
// 4x8 block
"mov %0, %1 \n"
"vld1.64 {d0}, [%0], %2 \n"
"vld1.64 {d1}, [%0], %2 \n"
"vld1.64 {d2}, [%0], %2 \n"
"vld1.64 {d3}, [%0], %2 \n"
"vld1.64 {d4}, [%0], %2 \n"
"vld1.64 {d5}, [%0], %2 \n"
"vld1.64 {d6}, [%0], %2 \n"
"vld1.64 {d7}, [%0] \n"
MEMACCESS(8)
"vld1.8 {q15}, [%8] \n"
"vld1.8 {q15}, [%8] \n"
"vtrn.8 q0, q1 \n"
"vtrn.8 q2, q3 \n"
"vtrn.8 q0, q1 \n"
"vtrn.8 q2, q3 \n"
"vtbl.8 d16, {d0, d1}, d30 \n"
"vtbl.8 d17, {d0, d1}, d31 \n"
"vtbl.8 d18, {d2, d3}, d30 \n"
"vtbl.8 d19, {d2, d3}, d31 \n"
"vtbl.8 d20, {d4, d5}, d30 \n"
"vtbl.8 d21, {d4, d5}, d31 \n"
"vtbl.8 d22, {d6, d7}, d30 \n"
"vtbl.8 d23, {d6, d7}, d31 \n"
"vtbl.8 d16, {d0, d1}, d30 \n"
"vtbl.8 d17, {d0, d1}, d31 \n"
"vtbl.8 d18, {d2, d3}, d30 \n"
"vtbl.8 d19, {d2, d3}, d31 \n"
"vtbl.8 d20, {d4, d5}, d30 \n"
"vtbl.8 d21, {d4, d5}, d31 \n"
"vtbl.8 d22, {d6, d7}, d30 \n"
"vtbl.8 d23, {d6, d7}, d31 \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.32 {d16[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d16[1]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d17[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d17[1]}, [%0], %4 \n"
"vst1.32 {d16[0]}, [%0], %4 \n"
"vst1.32 {d16[1]}, [%0], %4 \n"
"vst1.32 {d17[0]}, [%0], %4 \n"
"vst1.32 {d17[1]}, [%0], %4 \n"
"add %0, %3, #4 \n"
MEMACCESS(0)
"vst1.32 {d20[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d20[1]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d21[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d21[1]}, [%0] \n"
"add %0, %3, #4 \n"
"vst1.32 {d20[0]}, [%0], %4 \n"
"vst1.32 {d20[1]}, [%0], %4 \n"
"vst1.32 {d21[0]}, [%0], %4 \n"
"vst1.32 {d21[1]}, [%0] \n"
"mov %0, %5 \n"
"mov %0, %5 \n"
MEMACCESS(0)
"vst1.32 {d18[0]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d18[1]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d19[0]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d19[1]}, [%0], %6 \n"
"vst1.32 {d18[0]}, [%0], %6 \n"
"vst1.32 {d18[1]}, [%0], %6 \n"
"vst1.32 {d19[0]}, [%0], %6 \n"
"vst1.32 {d19[1]}, [%0], %6 \n"
"add %0, %5, #4 \n"
MEMACCESS(0)
"vst1.32 {d22[0]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d22[1]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d23[0]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d23[1]}, [%0] \n"
"add %0, %5, #4 \n"
"vst1.32 {d22[0]}, [%0], %6 \n"
"vst1.32 {d22[1]}, [%0], %6 \n"
"vst1.32 {d23[0]}, [%0], %6 \n"
"vst1.32 {d23[1]}, [%0] \n"
"add %1, #4*2 \n" // src += 4 * 2
"add %3, %3, %4, lsl #2 \n" // dst_a += 4 * dst_stride_a
"add %5, %5, %6, lsl #2 \n" // dst_b += 4 * dst_stride_b
"subs %7, #4 \n" // w -= 4
"beq 4f \n"
"add %1, #4*2 \n" // src += 4 * 2
"add %3, %3, %4, lsl #2 \n" // dst_a += 4 *
// dst_stride_a
"add %5, %5, %6, lsl #2 \n" // dst_b += 4 *
// dst_stride_b
"subs %7, #4 \n" // w -= 4
"beq 4f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %7, #2 \n"
"blt 3f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %7, #2 \n"
"blt 3f \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"vld2.16 {d0[0], d2[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d1[0], d3[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d0[1], d2[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d1[1], d3[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d0[2], d2[2]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d1[2], d3[2]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d0[3], d2[3]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d1[3], d3[3]}, [%0] \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
"vld2.16 {d0[0], d2[0]}, [%0], %2 \n"
"vld2.16 {d1[0], d3[0]}, [%0], %2 \n"
"vld2.16 {d0[1], d2[1]}, [%0], %2 \n"
"vld2.16 {d1[1], d3[1]}, [%0], %2 \n"
"vld2.16 {d0[2], d2[2]}, [%0], %2 \n"
"vld2.16 {d1[2], d3[2]}, [%0], %2 \n"
"vld2.16 {d0[3], d2[3]}, [%0], %2 \n"
"vld2.16 {d1[3], d3[3]}, [%0] \n"
"vtrn.8 d0, d1 \n"
"vtrn.8 d2, d3 \n"
"vtrn.8 d0, d1 \n"
"vtrn.8 d2, d3 \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.64 {d0}, [%0], %4 \n"
MEMACCESS(0)
"vst1.64 {d2}, [%0] \n"
"vst1.64 {d0}, [%0], %4 \n"
"vst1.64 {d2}, [%0] \n"
"mov %0, %5 \n"
"mov %0, %5 \n"
MEMACCESS(0)
"vst1.64 {d1}, [%0], %6 \n"
MEMACCESS(0)
"vst1.64 {d3}, [%0] \n"
"vst1.64 {d1}, [%0], %6 \n"
"vst1.64 {d3}, [%0] \n"
"add %1, #2*2 \n" // src += 2 * 2
"add %3, %3, %4, lsl #1 \n" // dst_a += 2 * dst_stride_a
"add %5, %5, %6, lsl #1 \n" // dst_b += 2 * dst_stride_b
"subs %7, #2 \n" // w -= 2
"beq 4f \n"
"add %1, #2*2 \n" // src += 2 * 2
"add %3, %3, %4, lsl #1 \n" // dst_a += 2 *
// dst_stride_a
"add %5, %5, %6, lsl #1 \n" // dst_b += 2 *
// dst_stride_b
"subs %7, #2 \n" // w -= 2
"beq 4f \n"
// 1x8 block
"3: \n"
MEMACCESS(1)
"vld2.8 {d0[0], d1[0]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[1], d1[1]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[2], d1[2]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[3], d1[3]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[4], d1[4]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[5], d1[5]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[6], d1[6]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[7], d1[7]}, [%1] \n"
// 1x8 block
"3: \n"
"vld2.8 {d0[0], d1[0]}, [%1], %2 \n"
"vld2.8 {d0[1], d1[1]}, [%1], %2 \n"
"vld2.8 {d0[2], d1[2]}, [%1], %2 \n"
"vld2.8 {d0[3], d1[3]}, [%1], %2 \n"
"vld2.8 {d0[4], d1[4]}, [%1], %2 \n"
"vld2.8 {d0[5], d1[5]}, [%1], %2 \n"
"vld2.8 {d0[6], d1[6]}, [%1], %2 \n"
"vld2.8 {d0[7], d1[7]}, [%1] \n"
MEMACCESS(3)
"vst1.64 {d0}, [%3] \n"
MEMACCESS(5)
"vst1.64 {d1}, [%5] \n"
"vst1.64 {d0}, [%3] \n"
"vst1.64 {d1}, [%5] \n"
"4: \n"
"4: \n"
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(src_stride), // %2
"+r"(dst_a), // %3
"+r"(dst_stride_a), // %4
"+r"(dst_b), // %5
"+r"(dst_stride_b), // %6
"+r"(width) // %7
: "r"(&kVTbl4x4TransposeDi) // %8
: "memory", "cc",
"q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11"
);
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(src_stride), // %2
"+r"(dst_a), // %3
"+r"(dst_stride_a), // %4
"+r"(dst_b), // %5
"+r"(dst_stride_b), // %6
"+r"(width) // %7
: "r"(&kVTbl4x4TransposeDi) // %8
: "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11");
}
#endif // defined(__ARM_NEON__) && !defined(__aarch64__)
libs/libyuv/source/rotate_neon64.cc
+284 -401
View File
@@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#include "libyuv/basic_types.h"
@@ -21,38 +21,32 @@ extern "C" {
// This module is for GCC Neon armv8 64 bit.
#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
static uvec8 kVTbl4x4Transpose =
{ 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15 };
static const uvec8 kVTbl4x4Transpose = {0, 4, 8, 12, 1, 5, 9, 13,
2, 6, 10, 14, 3, 7, 11, 15};
void TransposeWx8_NEON(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
const uint8* src_temp;
int64 width64 = (int64) width; // Work around clang 3.4 warning.
asm volatile (
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %3, %3, #8 \n"
void TransposeWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
const uint8_t* src_temp;
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %w3, %w3, #8 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v2.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v3.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v4.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v5.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v6.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v7.8b}, [%0] \n"
"trn2 v16.8b, v0.8b, v1.8b \n"
@@ -84,456 +78,345 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v17.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v19.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v21.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v20.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v23.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v22.8b}, [%0] \n"
"add %1, %1, #8 \n" // src += 8
"add %2, %2, %6, lsl #3 \n" // dst += 8 * dst_stride
"subs %3, %3, #8 \n" // w -= 8
"subs %w3, %w3, #8 \n" // w -= 8
"b.ge 1b \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %3, %3, #8 \n"
"b.eq 4f \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %w3, %w3, #8 \n"
"b.eq 4f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %3, #2 \n"
"b.lt 3f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %w3, #2 \n"
"b.lt 3f \n"
"cmp %3, #4 \n"
"b.lt 2f \n"
"cmp %w3, #4 \n"
"b.lt 2f \n"
// 4x8 block
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.s}[0], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.s}[1], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.s}[2], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.s}[3], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.s}[0], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.s}[1], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.s}[2], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.s}[3], [%0] \n"
// 4x8 block
"mov %0, %1 \n"
"ld1 {v0.s}[0], [%0], %5 \n"
"ld1 {v0.s}[1], [%0], %5 \n"
"ld1 {v0.s}[2], [%0], %5 \n"
"ld1 {v0.s}[3], [%0], %5 \n"
"ld1 {v1.s}[0], [%0], %5 \n"
"ld1 {v1.s}[1], [%0], %5 \n"
"ld1 {v1.s}[2], [%0], %5 \n"
"ld1 {v1.s}[3], [%0] \n"
"mov %0, %2 \n"
"mov %0, %2 \n"
MEMACCESS(4)
"ld1 {v2.16b}, [%4] \n"
"ld1 {v2.16b}, [%4] \n"
"tbl v3.16b, {v0.16b}, v2.16b \n"
"tbl v0.16b, {v1.16b}, v2.16b \n"
"tbl v3.16b, {v0.16b}, v2.16b \n"
"tbl v0.16b, {v1.16b}, v2.16b \n"
// TODO(frkoenig): Rework shuffle above to
// write out with 4 instead of 8 writes.
MEMACCESS(0)
"st1 {v3.s}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v3.s}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v3.s}[2], [%0], %6 \n"
MEMACCESS(0)
"st1 {v3.s}[3], [%0] \n"
// TODO(frkoenig): Rework shuffle above to
// write out with 4 instead of 8 writes.
"st1 {v3.s}[0], [%0], %6 \n"
"st1 {v3.s}[1], [%0], %6 \n"
"st1 {v3.s}[2], [%0], %6 \n"
"st1 {v3.s}[3], [%0] \n"
"add %0, %2, #4 \n"
MEMACCESS(0)
"st1 {v0.s}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v0.s}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v0.s}[2], [%0], %6 \n"
MEMACCESS(0)
"st1 {v0.s}[3], [%0] \n"
"add %0, %2, #4 \n"
"st1 {v0.s}[0], [%0], %6 \n"
"st1 {v0.s}[1], [%0], %6 \n"
"st1 {v0.s}[2], [%0], %6 \n"
"st1 {v0.s}[3], [%0] \n"
"add %1, %1, #4 \n" // src += 4
"add %2, %2, %6, lsl #2 \n" // dst += 4 * dst_stride
"subs %3, %3, #4 \n" // w -= 4
"b.eq 4f \n"
"add %1, %1, #4 \n" // src += 4
"add %2, %2, %6, lsl #2 \n" // dst += 4 * dst_stride
"subs %w3, %w3, #4 \n" // w -= 4
"b.eq 4f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %3, #2 \n"
"b.lt 3f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %w3, #2 \n"
"b.lt 3f \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.h}[0], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.h}[0], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.h}[1], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.h}[1], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.h}[2], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.h}[2], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.h}[3], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.h}[3], [%0] \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
"ld1 {v0.h}[0], [%0], %5 \n"
"ld1 {v1.h}[0], [%0], %5 \n"
"ld1 {v0.h}[1], [%0], %5 \n"
"ld1 {v1.h}[1], [%0], %5 \n"
"ld1 {v0.h}[2], [%0], %5 \n"
"ld1 {v1.h}[2], [%0], %5 \n"
"ld1 {v0.h}[3], [%0], %5 \n"
"ld1 {v1.h}[3], [%0] \n"
"trn2 v2.8b, v0.8b, v1.8b \n"
"trn1 v3.8b, v0.8b, v1.8b \n"
"trn2 v2.8b, v0.8b, v1.8b \n"
"trn1 v3.8b, v0.8b, v1.8b \n"
"mov %0, %2 \n"
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v3.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v2.8b}, [%0] \n"
"st1 {v3.8b}, [%0], %6 \n"
"st1 {v2.8b}, [%0] \n"
"add %1, %1, #2 \n" // src += 2
"add %2, %2, %6, lsl #1 \n" // dst += 2 * dst_stride
"subs %3, %3, #2 \n" // w -= 2
"b.eq 4f \n"
"add %1, %1, #2 \n" // src += 2
"add %2, %2, %6, lsl #1 \n" // dst += 2 * dst_stride
"subs %w3, %w3, #2 \n" // w -= 2
"b.eq 4f \n"
// 1x8 block
"3: \n"
MEMACCESS(1)
"ld1 {v0.b}[0], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[1], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[2], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[3], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[4], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[5], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[6], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[7], [%1] \n"
// 1x8 block
"3: \n"
"ld1 {v0.b}[0], [%1], %5 \n"
"ld1 {v0.b}[1], [%1], %5 \n"
"ld1 {v0.b}[2], [%1], %5 \n"
"ld1 {v0.b}[3], [%1], %5 \n"
"ld1 {v0.b}[4], [%1], %5 \n"
"ld1 {v0.b}[5], [%1], %5 \n"
"ld1 {v0.b}[6], [%1], %5 \n"
"ld1 {v0.b}[7], [%1] \n"
MEMACCESS(2)
"st1 {v0.8b}, [%2] \n"
"st1 {v0.8b}, [%2] \n"
"4: \n"
"4: \n"
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(dst), // %2
"+r"(width64) // %3
: "r"(&kVTbl4x4Transpose), // %4
"r"(static_cast<ptrdiff_t>(src_stride)), // %5
"r"(static_cast<ptrdiff_t>(dst_stride)) // %6
: "memory", "cc", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
"v17", "v18", "v19", "v20", "v21", "v22", "v23"
);
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(dst), // %2
"+r"(width) // %3
: "r"(&kVTbl4x4Transpose), // %4
"r"(static_cast<ptrdiff_t>(src_stride)), // %5
"r"(static_cast<ptrdiff_t>(dst_stride)) // %6
: "memory", "cc", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
"v17", "v18", "v19", "v20", "v21", "v22", "v23");
}
static uint8 kVTbl4x4TransposeDi[32] =
{ 0, 16, 32, 48, 2, 18, 34, 50, 4, 20, 36, 52, 6, 22, 38, 54,
1, 17, 33, 49, 3, 19, 35, 51, 5, 21, 37, 53, 7, 23, 39, 55};
static const uint8_t kVTbl4x4TransposeDi[32] = {
0, 16, 32, 48, 2, 18, 34, 50, 4, 20, 36, 52, 6, 22, 38, 54,
1, 17, 33, 49, 3, 19, 35, 51, 5, 21, 37, 53, 7, 23, 39, 55};
void TransposeUVWx8_NEON(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
void TransposeUVWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
const uint8* src_temp;
int64 width64 = (int64) width; // Work around clang 3.4 warning.
asm volatile (
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %4, %4, #8 \n"
const uint8_t* src_temp;
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %w4, %w4, #8 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v2.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v3.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v4.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v5.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v6.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v7.16b}, [%0] \n"
"ld1 {v0.16b}, [%0], %5 \n"
"ld1 {v1.16b}, [%0], %5 \n"
"ld1 {v2.16b}, [%0], %5 \n"
"ld1 {v3.16b}, [%0], %5 \n"
"ld1 {v4.16b}, [%0], %5 \n"
"ld1 {v5.16b}, [%0], %5 \n"
"ld1 {v6.16b}, [%0], %5 \n"
"ld1 {v7.16b}, [%0] \n"
"trn1 v16.16b, v0.16b, v1.16b \n"
"trn2 v17.16b, v0.16b, v1.16b \n"
"trn1 v18.16b, v2.16b, v3.16b \n"
"trn2 v19.16b, v2.16b, v3.16b \n"
"trn1 v20.16b, v4.16b, v5.16b \n"
"trn2 v21.16b, v4.16b, v5.16b \n"
"trn1 v22.16b, v6.16b, v7.16b \n"
"trn2 v23.16b, v6.16b, v7.16b \n"
"trn1 v16.16b, v0.16b, v1.16b \n"
"trn2 v17.16b, v0.16b, v1.16b \n"
"trn1 v18.16b, v2.16b, v3.16b \n"
"trn2 v19.16b, v2.16b, v3.16b \n"
"trn1 v20.16b, v4.16b, v5.16b \n"
"trn2 v21.16b, v4.16b, v5.16b \n"
"trn1 v22.16b, v6.16b, v7.16b \n"
"trn2 v23.16b, v6.16b, v7.16b \n"
"trn1 v0.8h, v16.8h, v18.8h \n"
"trn2 v1.8h, v16.8h, v18.8h \n"
"trn1 v2.8h, v20.8h, v22.8h \n"
"trn2 v3.8h, v20.8h, v22.8h \n"
"trn1 v4.8h, v17.8h, v19.8h \n"
"trn2 v5.8h, v17.8h, v19.8h \n"
"trn1 v6.8h, v21.8h, v23.8h \n"
"trn2 v7.8h, v21.8h, v23.8h \n"
"trn1 v0.8h, v16.8h, v18.8h \n"
"trn2 v1.8h, v16.8h, v18.8h \n"
"trn1 v2.8h, v20.8h, v22.8h \n"
"trn2 v3.8h, v20.8h, v22.8h \n"
"trn1 v4.8h, v17.8h, v19.8h \n"
"trn2 v5.8h, v17.8h, v19.8h \n"
"trn1 v6.8h, v21.8h, v23.8h \n"
"trn2 v7.8h, v21.8h, v23.8h \n"
"trn1 v16.4s, v0.4s, v2.4s \n"
"trn2 v17.4s, v0.4s, v2.4s \n"
"trn1 v18.4s, v1.4s, v3.4s \n"
"trn2 v19.4s, v1.4s, v3.4s \n"
"trn1 v20.4s, v4.4s, v6.4s \n"
"trn2 v21.4s, v4.4s, v6.4s \n"
"trn1 v22.4s, v5.4s, v7.4s \n"
"trn2 v23.4s, v5.4s, v7.4s \n"
"trn1 v16.4s, v0.4s, v2.4s \n"
"trn2 v17.4s, v0.4s, v2.4s \n"
"trn1 v18.4s, v1.4s, v3.4s \n"
"trn2 v19.4s, v1.4s, v3.4s \n"
"trn1 v20.4s, v4.4s, v6.4s \n"
"trn2 v21.4s, v4.4s, v6.4s \n"
"trn1 v22.4s, v5.4s, v7.4s \n"
"trn2 v23.4s, v5.4s, v7.4s \n"
"mov %0, %2 \n"
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v16.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v17.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v19.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.d}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.d}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v17.d}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v19.d}[1], [%0] \n"
"st1 {v16.d}[0], [%0], %6 \n"
"st1 {v18.d}[0], [%0], %6 \n"
"st1 {v17.d}[0], [%0], %6 \n"
"st1 {v19.d}[0], [%0], %6 \n"
"st1 {v16.d}[1], [%0], %6 \n"
"st1 {v18.d}[1], [%0], %6 \n"
"st1 {v17.d}[1], [%0], %6 \n"
"st1 {v19.d}[1], [%0] \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"st1 {v20.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v22.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v21.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v23.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v20.d}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v22.d}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v21.d}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v23.d}[1], [%0] \n"
"st1 {v20.d}[0], [%0], %7 \n"
"st1 {v22.d}[0], [%0], %7 \n"
"st1 {v21.d}[0], [%0], %7 \n"
"st1 {v23.d}[0], [%0], %7 \n"
"st1 {v20.d}[1], [%0], %7 \n"
"st1 {v22.d}[1], [%0], %7 \n"
"st1 {v21.d}[1], [%0], %7 \n"
"st1 {v23.d}[1], [%0] \n"
"add %1, %1, #16 \n" // src += 8*2
"add %2, %2, %6, lsl #3 \n" // dst_a += 8 * dst_stride_a
"add %3, %3, %7, lsl #3 \n" // dst_b += 8 * dst_stride_b
"subs %4, %4, #8 \n" // w -= 8
"b.ge 1b \n"
"add %1, %1, #16 \n" // src += 8*2
"add %2, %2, %6, lsl #3 \n" // dst_a += 8 *
// dst_stride_a
"add %3, %3, %7, lsl #3 \n" // dst_b += 8 *
// dst_stride_b
"subs %w4, %w4, #8 \n" // w -= 8
"b.ge 1b \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %4, %4, #8 \n"
"b.eq 4f \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %w4, %w4, #8 \n"
"b.eq 4f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %4, #2 \n"
"b.lt 3f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %w4, #2 \n"
"b.lt 3f \n"
"cmp %4, #4 \n"
"b.lt 2f \n"
"cmp %w4, #4 \n"
"b.lt 2f \n"
// TODO(frkoenig): Clean this up
// 4x8 block
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v2.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v3.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v4.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v5.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v6.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v7.8b}, [%0] \n"
// TODO(frkoenig): Clean this up
// 4x8 block
"mov %0, %1 \n"
"ld1 {v0.8b}, [%0], %5 \n"
"ld1 {v1.8b}, [%0], %5 \n"
"ld1 {v2.8b}, [%0], %5 \n"
"ld1 {v3.8b}, [%0], %5 \n"
"ld1 {v4.8b}, [%0], %5 \n"
"ld1 {v5.8b}, [%0], %5 \n"
"ld1 {v6.8b}, [%0], %5 \n"
"ld1 {v7.8b}, [%0] \n"
MEMACCESS(8)
"ld1 {v30.16b}, [%8], #16 \n"
"ld1 {v31.16b}, [%8] \n"
"ld1 {v30.16b}, [%8], #16 \n"
"ld1 {v31.16b}, [%8] \n"
"tbl v16.16b, {v0.16b, v1.16b, v2.16b, v3.16b}, v30.16b \n"
"tbl v17.16b, {v0.16b, v1.16b, v2.16b, v3.16b}, v31.16b \n"
"tbl v18.16b, {v4.16b, v5.16b, v6.16b, v7.16b}, v30.16b \n"
"tbl v19.16b, {v4.16b, v5.16b, v6.16b, v7.16b}, v31.16b \n"
"tbl v16.16b, {v0.16b, v1.16b, v2.16b, v3.16b}, v30.16b \n"
"tbl v17.16b, {v0.16b, v1.16b, v2.16b, v3.16b}, v31.16b \n"
"tbl v18.16b, {v4.16b, v5.16b, v6.16b, v7.16b}, v30.16b \n"
"tbl v19.16b, {v4.16b, v5.16b, v6.16b, v7.16b}, v31.16b \n"
"mov %0, %2 \n"
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v16.s}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.s}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.s}[2], [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.s}[3], [%0], %6 \n"
"st1 {v16.s}[0], [%0], %6 \n"
"st1 {v16.s}[1], [%0], %6 \n"
"st1 {v16.s}[2], [%0], %6 \n"
"st1 {v16.s}[3], [%0], %6 \n"
"add %0, %2, #4 \n"
MEMACCESS(0)
"st1 {v18.s}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.s}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.s}[2], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.s}[3], [%0] \n"
"add %0, %2, #4 \n"
"st1 {v18.s}[0], [%0], %6 \n"
"st1 {v18.s}[1], [%0], %6 \n"
"st1 {v18.s}[2], [%0], %6 \n"
"st1 {v18.s}[3], [%0] \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"st1 {v17.s}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v17.s}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v17.s}[2], [%0], %7 \n"
MEMACCESS(0)
"st1 {v17.s}[3], [%0], %7 \n"
"st1 {v17.s}[0], [%0], %7 \n"
"st1 {v17.s}[1], [%0], %7 \n"
"st1 {v17.s}[2], [%0], %7 \n"
"st1 {v17.s}[3], [%0], %7 \n"
"add %0, %3, #4 \n"
MEMACCESS(0)
"st1 {v19.s}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v19.s}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v19.s}[2], [%0], %7 \n"
MEMACCESS(0)
"st1 {v19.s}[3], [%0] \n"
"add %0, %3, #4 \n"
"st1 {v19.s}[0], [%0], %7 \n"
"st1 {v19.s}[1], [%0], %7 \n"
"st1 {v19.s}[2], [%0], %7 \n"
"st1 {v19.s}[3], [%0] \n"
"add %1, %1, #8 \n" // src += 4 * 2
"add %2, %2, %6, lsl #2 \n" // dst_a += 4 * dst_stride_a
"add %3, %3, %7, lsl #2 \n" // dst_b += 4 * dst_stride_b
"subs %4, %4, #4 \n" // w -= 4
"b.eq 4f \n"
"add %1, %1, #8 \n" // src += 4 * 2
"add %2, %2, %6, lsl #2 \n" // dst_a += 4 *
// dst_stride_a
"add %3, %3, %7, lsl #2 \n" // dst_b += 4 *
// dst_stride_b
"subs %w4, %w4, #4 \n" // w -= 4
"b.eq 4f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %4, #2 \n"
"b.lt 3f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %w4, #2 \n"
"b.lt 3f \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"ld2 {v0.h, v1.h}[0], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v2.h, v3.h}[0], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v0.h, v1.h}[1], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v2.h, v3.h}[1], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v0.h, v1.h}[2], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v2.h, v3.h}[2], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v0.h, v1.h}[3], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v2.h, v3.h}[3], [%0] \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
"ld2 {v0.h, v1.h}[0], [%0], %5 \n"
"ld2 {v2.h, v3.h}[0], [%0], %5 \n"
"ld2 {v0.h, v1.h}[1], [%0], %5 \n"
"ld2 {v2.h, v3.h}[1], [%0], %5 \n"
"ld2 {v0.h, v1.h}[2], [%0], %5 \n"
"ld2 {v2.h, v3.h}[2], [%0], %5 \n"
"ld2 {v0.h, v1.h}[3], [%0], %5 \n"
"ld2 {v2.h, v3.h}[3], [%0] \n"
"trn1 v4.8b, v0.8b, v2.8b \n"
"trn2 v5.8b, v0.8b, v2.8b \n"
"trn1 v6.8b, v1.8b, v3.8b \n"
"trn2 v7.8b, v1.8b, v3.8b \n"
"trn1 v4.8b, v0.8b, v2.8b \n"
"trn2 v5.8b, v0.8b, v2.8b \n"
"trn1 v6.8b, v1.8b, v3.8b \n"
"trn2 v7.8b, v1.8b, v3.8b \n"
"mov %0, %2 \n"
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v4.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v6.d}[0], [%0] \n"
"st1 {v4.d}[0], [%0], %6 \n"
"st1 {v6.d}[0], [%0] \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"st1 {v5.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v7.d}[0], [%0] \n"
"st1 {v5.d}[0], [%0], %7 \n"
"st1 {v7.d}[0], [%0] \n"
"add %1, %1, #4 \n" // src += 2 * 2
"add %2, %2, %6, lsl #1 \n" // dst_a += 2 * dst_stride_a
"add %3, %3, %7, lsl #1 \n" // dst_b += 2 * dst_stride_b
"subs %4, %4, #2 \n" // w -= 2
"b.eq 4f \n"
"add %1, %1, #4 \n" // src += 2 * 2
"add %2, %2, %6, lsl #1 \n" // dst_a += 2 *
// dst_stride_a
"add %3, %3, %7, lsl #1 \n" // dst_b += 2 *
// dst_stride_b
"subs %w4, %w4, #2 \n" // w -= 2
"b.eq 4f \n"
// 1x8 block
"3: \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[0], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[1], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[2], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[3], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[4], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[5], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[6], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[7], [%1] \n"
// 1x8 block
"3: \n"
"ld2 {v0.b, v1.b}[0], [%1], %5 \n"
"ld2 {v0.b, v1.b}[1], [%1], %5 \n"
"ld2 {v0.b, v1.b}[2], [%1], %5 \n"
"ld2 {v0.b, v1.b}[3], [%1], %5 \n"
"ld2 {v0.b, v1.b}[4], [%1], %5 \n"
"ld2 {v0.b, v1.b}[5], [%1], %5 \n"
"ld2 {v0.b, v1.b}[6], [%1], %5 \n"
"ld2 {v0.b, v1.b}[7], [%1] \n"
MEMACCESS(2)
"st1 {v0.d}[0], [%2] \n"
MEMACCESS(3)
"st1 {v1.d}[0], [%3] \n"
"st1 {v0.d}[0], [%2] \n"
"st1 {v1.d}[0], [%3] \n"
"4: \n"
"4: \n"
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(dst_a), // %2
"+r"(dst_b), // %3
"+r"(width64) // %4
: "r"(static_cast<ptrdiff_t>(src_stride)), // %5
"r"(static_cast<ptrdiff_t>(dst_stride_a)), // %6
"r"(static_cast<ptrdiff_t>(dst_stride_b)), // %7
"r"(&kVTbl4x4TransposeDi) // %8
: "memory", "cc",
"v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
"v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
"v30", "v31"
);
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(dst_a), // %2
"+r"(dst_b), // %3
"+r"(width) // %4
: "r"(static_cast<ptrdiff_t>(src_stride)), // %5
"r"(static_cast<ptrdiff_t>(dst_stride_a)), // %6
"r"(static_cast<ptrdiff_t>(dst_stride_b)), // %7
"r"(&kVTbl4x4TransposeDi) // %8
: "memory", "cc", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
"v17", "v18", "v19", "v20", "v21", "v22", "v23", "v30", "v31");
}
#endif // !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
libs/libyuv/source/rotate_win.cc
+28 -23
View File
@@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
@@ -17,17 +17,19 @@ extern "C" {
#endif
// This module is for 32 bit Visual C x86 and clangcl
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
__declspec(naked)
void TransposeWx8_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
__declspec(naked) void TransposeWx8_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
__asm {
push edi
push esi
push ebp
mov eax, [esp + 12 + 4] // src
mov edi, [esp + 12 + 8] // src_stride
mov eax, [esp + 12 + 4] // src
mov edi, [esp + 12 + 8] // src_stride
mov edx, [esp + 12 + 12] // dst
mov esi, [esp + 12 + 16] // dst_stride
mov ecx, [esp + 12 + 20] // width
@@ -110,18 +112,20 @@ void TransposeWx8_SSSE3(const uint8* src, int src_stride,
}
}
__declspec(naked)
void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int w) {
__declspec(naked) void TransposeUVWx8_SSE2(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int w) {
__asm {
push ebx
push esi
push edi
push ebp
mov eax, [esp + 16 + 4] // src
mov edi, [esp + 16 + 8] // src_stride
mov eax, [esp + 16 + 4] // src
mov edi, [esp + 16 + 8] // src_stride
mov edx, [esp + 16 + 12] // dst_a
mov esi, [esp + 16 + 16] // dst_stride_a
mov ebx, [esp + 16 + 20] // dst_b
@@ -133,9 +137,9 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
mov ecx, [ecx + 16 + 28] // w
align 4
convertloop:
// Read in the data from the source pointer.
// First round of bit swap.
convertloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + edi]
lea eax, [eax + 2 * edi]
@@ -162,13 +166,13 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
lea eax, [eax + 2 * edi]
movdqu [esp], xmm5 // backup xmm5
neg edi
movdqa xmm5, xmm6 // use xmm5 as temp register.
movdqa xmm5, xmm6 // use xmm5 as temp register.
punpcklbw xmm6, xmm7
punpckhbw xmm5, xmm7
movdqa xmm7, xmm5
lea eax, [eax + 8 * edi + 16]
neg edi
// Second round of bit swap.
// Second round of bit swap.
movdqa xmm5, xmm0
punpcklwd xmm0, xmm2
punpckhwd xmm5, xmm2
@@ -183,12 +187,13 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
movdqa xmm6, xmm5
movdqu xmm5, [esp] // restore xmm5
movdqu [esp], xmm6 // backup xmm6
movdqa xmm6, xmm5 // use xmm6 as temp register.
movdqa xmm6, xmm5 // use xmm6 as temp register.
punpcklwd xmm5, xmm7
punpckhwd xmm6, xmm7
movdqa xmm7, xmm6
// Third round of bit swap.
// Write to the destination pointer.
// Third round of bit swap.
// Write to the destination pointer.
movdqa xmm6, xmm0
punpckldq xmm0, xmm4
punpckhdq xmm6, xmm4
@@ -200,7 +205,7 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm4
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm2 // use xmm0 as the temp register.
movdqa xmm0, xmm2 // use xmm0 as the temp register.
punpckldq xmm2, xmm6
movlpd qword ptr [edx], xmm2
movhpd qword ptr [ebx], xmm2
@@ -209,7 +214,7 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm0
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm1 // use xmm0 as the temp register.
movdqa xmm0, xmm1 // use xmm0 as the temp register.
punpckldq xmm1, xmm5
movlpd qword ptr [edx], xmm1
movhpd qword ptr [ebx], xmm1
@@ -218,7 +223,7 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm0
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm3 // use xmm0 as the temp register.
movdqa xmm0, xmm3 // use xmm0 as the temp register.
punpckldq xmm3, xmm7
movlpd qword ptr [edx], xmm3
movhpd qword ptr [ebx], xmm3
libs/libyuv/source/row_any.cc
+832 -275
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/row_common.cc
+1562 -952
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/row_gcc.cc
+5565 -4422
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/row_mips.cc
-782
View File
@@ -1,782 +0,0 @@
/*
* Copyright (c) 2012 The LibYuv project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// The following are available on Mips platforms:
#if !defined(LIBYUV_DISABLE_MIPS) && defined(__mips__) && \
(_MIPS_SIM == _MIPS_SIM_ABI32)
#ifdef HAS_COPYROW_MIPS
void CopyRow_MIPS(const uint8* src, uint8* dst, int count) {
__asm__ __volatile__ (
".set noreorder \n"
".set noat \n"
"slti $at, %[count], 8 \n"
"bne $at ,$zero, $last8 \n"
"xor $t8, %[src], %[dst] \n"
"andi $t8, $t8, 0x3 \n"
"bne $t8, $zero, unaligned \n"
"negu $a3, %[dst] \n"
// make dst/src aligned
"andi $a3, $a3, 0x3 \n"
"beq $a3, $zero, $chk16w \n"
// word-aligned now count is the remining bytes count
"subu %[count], %[count], $a3 \n"
"lwr $t8, 0(%[src]) \n"
"addu %[src], %[src], $a3 \n"
"swr $t8, 0(%[dst]) \n"
"addu %[dst], %[dst], $a3 \n"
// Now the dst/src are mutually word-aligned with word-aligned addresses
"$chk16w: \n"
"andi $t8, %[count], 0x3f \n" // whole 64-B chunks?
// t8 is the byte count after 64-byte chunks
"beq %[count], $t8, chk8w \n"
// There will be at most 1 32-byte chunk after it
"subu $a3, %[count], $t8 \n" // the reminder
// Here a3 counts bytes in 16w chunks
"addu $a3, %[dst], $a3 \n"
// Now a3 is the final dst after 64-byte chunks
"addu $t0, %[dst], %[count] \n"
// t0 is the "past the end" address
// When in the loop we exercise "pref 30,x(a1)", the a1+x should not be past
// the "t0-32" address
// This means: for x=128 the last "safe" a1 address is "t0-160"
// Alternatively, for x=64 the last "safe" a1 address is "t0-96"
// we will use "pref 30,128(a1)", so "t0-160" is the limit
"subu $t9, $t0, 160 \n"
// t9 is the "last safe pref 30,128(a1)" address
"pref 0, 0(%[src]) \n" // first line of src
"pref 0, 32(%[src]) \n" // second line of src
"pref 0, 64(%[src]) \n"
"pref 30, 32(%[dst]) \n"
// In case the a1 > t9 don't use "pref 30" at all
"sgtu $v1, %[dst], $t9 \n"
"bgtz $v1, $loop16w \n"
"nop \n"
// otherwise, start with using pref30
"pref 30, 64(%[dst]) \n"
"$loop16w: \n"
"pref 0, 96(%[src]) \n"
"lw $t0, 0(%[src]) \n"
"bgtz $v1, $skip_pref30_96 \n" // skip
"lw $t1, 4(%[src]) \n"
"pref 30, 96(%[dst]) \n" // continue
"$skip_pref30_96: \n"
"lw $t2, 8(%[src]) \n"
"lw $t3, 12(%[src]) \n"
"lw $t4, 16(%[src]) \n"
"lw $t5, 20(%[src]) \n"
"lw $t6, 24(%[src]) \n"
"lw $t7, 28(%[src]) \n"
"pref 0, 128(%[src]) \n"
// bring the next lines of src, addr 128
"sw $t0, 0(%[dst]) \n"
"sw $t1, 4(%[dst]) \n"
"sw $t2, 8(%[dst]) \n"
"sw $t3, 12(%[dst]) \n"
"sw $t4, 16(%[dst]) \n"
"sw $t5, 20(%[dst]) \n"
"sw $t6, 24(%[dst]) \n"
"sw $t7, 28(%[dst]) \n"
"lw $t0, 32(%[src]) \n"
"bgtz $v1, $skip_pref30_128 \n" // skip pref 30,128(a1)
"lw $t1, 36(%[src]) \n"
"pref 30, 128(%[dst]) \n" // set dest, addr 128
"$skip_pref30_128: \n"
"lw $t2, 40(%[src]) \n"
"lw $t3, 44(%[src]) \n"
"lw $t4, 48(%[src]) \n"
"lw $t5, 52(%[src]) \n"
"lw $t6, 56(%[src]) \n"
"lw $t7, 60(%[src]) \n"
"pref 0, 160(%[src]) \n"
// bring the next lines of src, addr 160
"sw $t0, 32(%[dst]) \n"
"sw $t1, 36(%[dst]) \n"
"sw $t2, 40(%[dst]) \n"
"sw $t3, 44(%[dst]) \n"
"sw $t4, 48(%[dst]) \n"
"sw $t5, 52(%[dst]) \n"
"sw $t6, 56(%[dst]) \n"
"sw $t7, 60(%[dst]) \n"
"addiu %[dst], %[dst], 64 \n" // adding 64 to dest
"sgtu $v1, %[dst], $t9 \n"
"bne %[dst], $a3, $loop16w \n"
" addiu %[src], %[src], 64 \n" // adding 64 to src
"move %[count], $t8 \n"
// Here we have src and dest word-aligned but less than 64-bytes to go
"chk8w: \n"
"pref 0, 0x0(%[src]) \n"
"andi $t8, %[count], 0x1f \n" // 32-byte chunk?
// the t8 is the reminder count past 32-bytes
"beq %[count], $t8, chk1w \n"
// count=t8,no 32-byte chunk
" nop \n"
"lw $t0, 0(%[src]) \n"
"lw $t1, 4(%[src]) \n"
"lw $t2, 8(%[src]) \n"
"lw $t3, 12(%[src]) \n"
"lw $t4, 16(%[src]) \n"
"lw $t5, 20(%[src]) \n"
"lw $t6, 24(%[src]) \n"
"lw $t7, 28(%[src]) \n"
"addiu %[src], %[src], 32 \n"
"sw $t0, 0(%[dst]) \n"
"sw $t1, 4(%[dst]) \n"
"sw $t2, 8(%[dst]) \n"
"sw $t3, 12(%[dst]) \n"
"sw $t4, 16(%[dst]) \n"
"sw $t5, 20(%[dst]) \n"
"sw $t6, 24(%[dst]) \n"
"sw $t7, 28(%[dst]) \n"
"addiu %[dst], %[dst], 32 \n"
"chk1w: \n"
"andi %[count], $t8, 0x3 \n"
// now count is the reminder past 1w chunks
"beq %[count], $t8, $last8 \n"
" subu $a3, $t8, %[count] \n"
// a3 is count of bytes in 1w chunks
"addu $a3, %[dst], $a3 \n"
// now a3 is the dst address past the 1w chunks
// copying in words (4-byte chunks)
"$wordCopy_loop: \n"
"lw $t3, 0(%[src]) \n"
// the first t3 may be equal t0 ... optimize?
"addiu %[src], %[src],4 \n"
"addiu %[dst], %[dst],4 \n"
"bne %[dst], $a3,$wordCopy_loop \n"
" sw $t3, -4(%[dst]) \n"
// For the last (<8) bytes
"$last8: \n"
"blez %[count], leave \n"
" addu $a3, %[dst], %[count] \n" // a3 -last dst address
"$last8loop: \n"
"lb $v1, 0(%[src]) \n"
"addiu %[src], %[src], 1 \n"
"addiu %[dst], %[dst], 1 \n"
"bne %[dst], $a3, $last8loop \n"
" sb $v1, -1(%[dst]) \n"
"leave: \n"
" j $ra \n"
" nop \n"
//
// UNALIGNED case
//
"unaligned: \n"
// got here with a3="negu a1"
"andi $a3, $a3, 0x3 \n" // a1 is word aligned?
"beqz $a3, $ua_chk16w \n"
" subu %[count], %[count], $a3 \n"
// bytes left after initial a3 bytes
"lwr $v1, 0(%[src]) \n"
"lwl $v1, 3(%[src]) \n"
"addu %[src], %[src], $a3 \n" // a3 may be 1, 2 or 3
"swr $v1, 0(%[dst]) \n"
"addu %[dst], %[dst], $a3 \n"
// below the dst will be word aligned (NOTE1)
"$ua_chk16w: \n"
"andi $t8, %[count], 0x3f \n" // whole 64-B chunks?
// t8 is the byte count after 64-byte chunks
"beq %[count], $t8, ua_chk8w \n"
// if a2==t8, no 64-byte chunks
// There will be at most 1 32-byte chunk after it
"subu $a3, %[count], $t8 \n" // the reminder
// Here a3 counts bytes in 16w chunks
"addu $a3, %[dst], $a3 \n"
// Now a3 is the final dst after 64-byte chunks
"addu $t0, %[dst], %[count] \n" // t0 "past the end"
"subu $t9, $t0, 160 \n"
// t9 is the "last safe pref 30,128(a1)" address
"pref 0, 0(%[src]) \n" // first line of src
"pref 0, 32(%[src]) \n" // second line addr 32
"pref 0, 64(%[src]) \n"
"pref 30, 32(%[dst]) \n"
// safe, as we have at least 64 bytes ahead
// In case the a1 > t9 don't use "pref 30" at all
"sgtu $v1, %[dst], $t9 \n"
"bgtz $v1, $ua_loop16w \n"
// skip "pref 30,64(a1)" for too short arrays
" nop \n"
// otherwise, start with using pref30
"pref 30, 64(%[dst]) \n"
"$ua_loop16w: \n"
"pref 0, 96(%[src]) \n"
"lwr $t0, 0(%[src]) \n"
"lwl $t0, 3(%[src]) \n"
"lwr $t1, 4(%[src]) \n"
"bgtz $v1, $ua_skip_pref30_96 \n"
" lwl $t1, 7(%[src]) \n"
"pref 30, 96(%[dst]) \n"
// continue setting up the dest, addr 96
"$ua_skip_pref30_96: \n"
"lwr $t2, 8(%[src]) \n"
"lwl $t2, 11(%[src]) \n"
"lwr $t3, 12(%[src]) \n"
"lwl $t3, 15(%[src]) \n"
"lwr $t4, 16(%[src]) \n"
"lwl $t4, 19(%[src]) \n"
"lwr $t5, 20(%[src]) \n"
"lwl $t5, 23(%[src]) \n"
"lwr $t6, 24(%[src]) \n"
"lwl $t6, 27(%[src]) \n"
"lwr $t7, 28(%[src]) \n"
"lwl $t7, 31(%[src]) \n"
"pref 0, 128(%[src]) \n"
// bring the next lines of src, addr 128
"sw $t0, 0(%[dst]) \n"
"sw $t1, 4(%[dst]) \n"
"sw $t2, 8(%[dst]) \n"
"sw $t3, 12(%[dst]) \n"
"sw $t4, 16(%[dst]) \n"
"sw $t5, 20(%[dst]) \n"
"sw $t6, 24(%[dst]) \n"
"sw $t7, 28(%[dst]) \n"
"lwr $t0, 32(%[src]) \n"
"lwl $t0, 35(%[src]) \n"
"lwr $t1, 36(%[src]) \n"
"bgtz $v1, ua_skip_pref30_128 \n"
" lwl $t1, 39(%[src]) \n"
"pref 30, 128(%[dst]) \n"
// continue setting up the dest, addr 128
"ua_skip_pref30_128: \n"
"lwr $t2, 40(%[src]) \n"
"lwl $t2, 43(%[src]) \n"
"lwr $t3, 44(%[src]) \n"
"lwl $t3, 47(%[src]) \n"
"lwr $t4, 48(%[src]) \n"
"lwl $t4, 51(%[src]) \n"
"lwr $t5, 52(%[src]) \n"
"lwl $t5, 55(%[src]) \n"
"lwr $t6, 56(%[src]) \n"
"lwl $t6, 59(%[src]) \n"
"lwr $t7, 60(%[src]) \n"
"lwl $t7, 63(%[src]) \n"
"pref 0, 160(%[src]) \n"
// bring the next lines of src, addr 160
"sw $t0, 32(%[dst]) \n"
"sw $t1, 36(%[dst]) \n"
"sw $t2, 40(%[dst]) \n"
"sw $t3, 44(%[dst]) \n"
"sw $t4, 48(%[dst]) \n"
"sw $t5, 52(%[dst]) \n"
"sw $t6, 56(%[dst]) \n"
"sw $t7, 60(%[dst]) \n"
"addiu %[dst],%[dst],64 \n" // adding 64 to dest
"sgtu $v1,%[dst],$t9 \n"
"bne %[dst],$a3,$ua_loop16w \n"
" addiu %[src],%[src],64 \n" // adding 64 to src
"move %[count],$t8 \n"
// Here we have src and dest word-aligned but less than 64-bytes to go
"ua_chk8w: \n"
"pref 0, 0x0(%[src]) \n"
"andi $t8, %[count], 0x1f \n" // 32-byte chunk?
// the t8 is the reminder count
"beq %[count], $t8, $ua_chk1w \n"
// when count==t8, no 32-byte chunk
"lwr $t0, 0(%[src]) \n"
"lwl $t0, 3(%[src]) \n"
"lwr $t1, 4(%[src]) \n"
"lwl $t1, 7(%[src]) \n"
"lwr $t2, 8(%[src]) \n"
"lwl $t2, 11(%[src]) \n"
"lwr $t3, 12(%[src]) \n"
"lwl $t3, 15(%[src]) \n"
"lwr $t4, 16(%[src]) \n"
"lwl $t4, 19(%[src]) \n"
"lwr $t5, 20(%[src]) \n"
"lwl $t5, 23(%[src]) \n"
"lwr $t6, 24(%[src]) \n"
"lwl $t6, 27(%[src]) \n"
"lwr $t7, 28(%[src]) \n"
"lwl $t7, 31(%[src]) \n"
"addiu %[src], %[src], 32 \n"
"sw $t0, 0(%[dst]) \n"
"sw $t1, 4(%[dst]) \n"
"sw $t2, 8(%[dst]) \n"
"sw $t3, 12(%[dst]) \n"
"sw $t4, 16(%[dst]) \n"
"sw $t5, 20(%[dst]) \n"
"sw $t6, 24(%[dst]) \n"
"sw $t7, 28(%[dst]) \n"
"addiu %[dst], %[dst], 32 \n"
"$ua_chk1w: \n"
"andi %[count], $t8, 0x3 \n"
// now count is the reminder past 1w chunks
"beq %[count], $t8, ua_smallCopy \n"
"subu $a3, $t8, %[count] \n"
// a3 is count of bytes in 1w chunks
"addu $a3, %[dst], $a3 \n"
// now a3 is the dst address past the 1w chunks
// copying in words (4-byte chunks)
"$ua_wordCopy_loop: \n"
"lwr $v1, 0(%[src]) \n"
"lwl $v1, 3(%[src]) \n"
"addiu %[src], %[src], 4 \n"
"addiu %[dst], %[dst], 4 \n"
// note: dst=a1 is word aligned here, see NOTE1
"bne %[dst], $a3, $ua_wordCopy_loop \n"
" sw $v1,-4(%[dst]) \n"
// Now less than 4 bytes (value in count) left to copy
"ua_smallCopy: \n"
"beqz %[count], leave \n"
" addu $a3, %[dst], %[count] \n" // a3 = last dst address
"$ua_smallCopy_loop: \n"
"lb $v1, 0(%[src]) \n"
"addiu %[src], %[src], 1 \n"
"addiu %[dst], %[dst], 1 \n"
"bne %[dst],$a3,$ua_smallCopy_loop \n"
" sb $v1, -1(%[dst]) \n"
"j $ra \n"
" nop \n"
".set at \n"
".set reorder \n"
: [dst] "+r" (dst), [src] "+r" (src)
: [count] "r" (count)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7",
"t8", "t9", "a3", "v1", "at"
);
}
#endif // HAS_COPYROW_MIPS
// DSPR2 functions
#if !defined(LIBYUV_DISABLE_MIPS) && defined(__mips_dsp) && \
(__mips_dsp_rev >= 2) && \
(_MIPS_SIM == _MIPS_SIM_ABI32) && (__mips_isa_rev < 6)
void SplitUVRow_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
int width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"srl $t4, %[width], 4 \n" // multiplies of 16
"blez $t4, 2f \n"
" andi %[width], %[width], 0xf \n" // residual
"1: \n"
"addiu $t4, $t4, -1 \n"
"lw $t0, 0(%[src_uv]) \n" // V1 | U1 | V0 | U0
"lw $t1, 4(%[src_uv]) \n" // V3 | U3 | V2 | U2
"lw $t2, 8(%[src_uv]) \n" // V5 | U5 | V4 | U4
"lw $t3, 12(%[src_uv]) \n" // V7 | U7 | V6 | U6
"lw $t5, 16(%[src_uv]) \n" // V9 | U9 | V8 | U8
"lw $t6, 20(%[src_uv]) \n" // V11 | U11 | V10 | U10
"lw $t7, 24(%[src_uv]) \n" // V13 | U13 | V12 | U12
"lw $t8, 28(%[src_uv]) \n" // V15 | U15 | V14 | U14
"addiu %[src_uv], %[src_uv], 32 \n"
"precrq.qb.ph $t9, $t1, $t0 \n" // V3 | V2 | V1 | V0
"precr.qb.ph $t0, $t1, $t0 \n" // U3 | U2 | U1 | U0
"precrq.qb.ph $t1, $t3, $t2 \n" // V7 | V6 | V5 | V4
"precr.qb.ph $t2, $t3, $t2 \n" // U7 | U6 | U5 | U4
"precrq.qb.ph $t3, $t6, $t5 \n" // V11 | V10 | V9 | V8
"precr.qb.ph $t5, $t6, $t5 \n" // U11 | U10 | U9 | U8
"precrq.qb.ph $t6, $t8, $t7 \n" // V15 | V14 | V13 | V12
"precr.qb.ph $t7, $t8, $t7 \n" // U15 | U14 | U13 | U12
"sw $t9, 0(%[dst_v]) \n"
"sw $t0, 0(%[dst_u]) \n"
"sw $t1, 4(%[dst_v]) \n"
"sw $t2, 4(%[dst_u]) \n"
"sw $t3, 8(%[dst_v]) \n"
"sw $t5, 8(%[dst_u]) \n"
"sw $t6, 12(%[dst_v]) \n"
"sw $t7, 12(%[dst_u]) \n"
"addiu %[dst_v], %[dst_v], 16 \n"
"bgtz $t4, 1b \n"
" addiu %[dst_u], %[dst_u], 16 \n"
"beqz %[width], 3f \n"
" nop \n"
"2: \n"
"lbu $t0, 0(%[src_uv]) \n"
"lbu $t1, 1(%[src_uv]) \n"
"addiu %[src_uv], %[src_uv], 2 \n"
"addiu %[width], %[width], -1 \n"
"sb $t0, 0(%[dst_u]) \n"
"sb $t1, 0(%[dst_v]) \n"
"addiu %[dst_u], %[dst_u], 1 \n"
"bgtz %[width], 2b \n"
" addiu %[dst_v], %[dst_v], 1 \n"
"3: \n"
".set pop \n"
: [src_uv] "+r" (src_uv),
[width] "+r" (width),
[dst_u] "+r" (dst_u),
[dst_v] "+r" (dst_v)
:
: "t0", "t1", "t2", "t3",
"t4", "t5", "t6", "t7", "t8", "t9"
);
}
void MirrorRow_DSPR2(const uint8* src, uint8* dst, int width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"srl $t4, %[width], 4 \n" // multiplies of 16
"andi $t5, %[width], 0xf \n"
"blez $t4, 2f \n"
" addu %[src], %[src], %[width] \n" // src += width
"1: \n"
"lw $t0, -16(%[src]) \n" // |3|2|1|0|
"lw $t1, -12(%[src]) \n" // |7|6|5|4|
"lw $t2, -8(%[src]) \n" // |11|10|9|8|
"lw $t3, -4(%[src]) \n" // |15|14|13|12|
"wsbh $t0, $t0 \n" // |2|3|0|1|
"wsbh $t1, $t1 \n" // |6|7|4|5|
"wsbh $t2, $t2 \n" // |10|11|8|9|
"wsbh $t3, $t3 \n" // |14|15|12|13|
"rotr $t0, $t0, 16 \n" // |0|1|2|3|
"rotr $t1, $t1, 16 \n" // |4|5|6|7|
"rotr $t2, $t2, 16 \n" // |8|9|10|11|
"rotr $t3, $t3, 16 \n" // |12|13|14|15|
"addiu %[src], %[src], -16 \n"
"addiu $t4, $t4, -1 \n"
"sw $t3, 0(%[dst]) \n" // |15|14|13|12|
"sw $t2, 4(%[dst]) \n" // |11|10|9|8|
"sw $t1, 8(%[dst]) \n" // |7|6|5|4|
"sw $t0, 12(%[dst]) \n" // |3|2|1|0|
"bgtz $t4, 1b \n"
" addiu %[dst], %[dst], 16 \n"
"beqz $t5, 3f \n"
" nop \n"
"2: \n"
"lbu $t0, -1(%[src]) \n"
"addiu $t5, $t5, -1 \n"
"addiu %[src], %[src], -1 \n"
"sb $t0, 0(%[dst]) \n"
"bgez $t5, 2b \n"
" addiu %[dst], %[dst], 1 \n"
"3: \n"
".set pop \n"
: [src] "+r" (src), [dst] "+r" (dst)
: [width] "r" (width)
: "t0", "t1", "t2", "t3", "t4", "t5"
);
}
void MirrorUVRow_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
int width) {
int x;
int y;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"addu $t4, %[width], %[width] \n"
"srl %[x], %[width], 4 \n"
"andi %[y], %[width], 0xf \n"
"blez %[x], 2f \n"
" addu %[src_uv], %[src_uv], $t4 \n"
"1: \n"
"lw $t0, -32(%[src_uv]) \n" // |3|2|1|0|
"lw $t1, -28(%[src_uv]) \n" // |7|6|5|4|
"lw $t2, -24(%[src_uv]) \n" // |11|10|9|8|
"lw $t3, -20(%[src_uv]) \n" // |15|14|13|12|
"lw $t4, -16(%[src_uv]) \n" // |19|18|17|16|
"lw $t6, -12(%[src_uv]) \n" // |23|22|21|20|
"lw $t7, -8(%[src_uv]) \n" // |27|26|25|24|
"lw $t8, -4(%[src_uv]) \n" // |31|30|29|28|
"rotr $t0, $t0, 16 \n" // |1|0|3|2|
"rotr $t1, $t1, 16 \n" // |5|4|7|6|
"rotr $t2, $t2, 16 \n" // |9|8|11|10|
"rotr $t3, $t3, 16 \n" // |13|12|15|14|
"rotr $t4, $t4, 16 \n" // |17|16|19|18|
"rotr $t6, $t6, 16 \n" // |21|20|23|22|
"rotr $t7, $t7, 16 \n" // |25|24|27|26|
"rotr $t8, $t8, 16 \n" // |29|28|31|30|
"precr.qb.ph $t9, $t0, $t1 \n" // |0|2|4|6|
"precrq.qb.ph $t5, $t0, $t1 \n" // |1|3|5|7|
"precr.qb.ph $t0, $t2, $t3 \n" // |8|10|12|14|
"precrq.qb.ph $t1, $t2, $t3 \n" // |9|11|13|15|
"precr.qb.ph $t2, $t4, $t6 \n" // |16|18|20|22|
"precrq.qb.ph $t3, $t4, $t6 \n" // |17|19|21|23|
"precr.qb.ph $t4, $t7, $t8 \n" // |24|26|28|30|
"precrq.qb.ph $t6, $t7, $t8 \n" // |25|27|29|31|
"addiu %[src_uv], %[src_uv], -32 \n"
"addiu %[x], %[x], -1 \n"
"swr $t4, 0(%[dst_u]) \n"
"swl $t4, 3(%[dst_u]) \n" // |30|28|26|24|
"swr $t6, 0(%[dst_v]) \n"
"swl $t6, 3(%[dst_v]) \n" // |31|29|27|25|
"swr $t2, 4(%[dst_u]) \n"
"swl $t2, 7(%[dst_u]) \n" // |22|20|18|16|
"swr $t3, 4(%[dst_v]) \n"
"swl $t3, 7(%[dst_v]) \n" // |23|21|19|17|
"swr $t0, 8(%[dst_u]) \n"
"swl $t0, 11(%[dst_u]) \n" // |14|12|10|8|
"swr $t1, 8(%[dst_v]) \n"
"swl $t1, 11(%[dst_v]) \n" // |15|13|11|9|
"swr $t9, 12(%[dst_u]) \n"
"swl $t9, 15(%[dst_u]) \n" // |6|4|2|0|
"swr $t5, 12(%[dst_v]) \n"
"swl $t5, 15(%[dst_v]) \n" // |7|5|3|1|
"addiu %[dst_v], %[dst_v], 16 \n"
"bgtz %[x], 1b \n"
" addiu %[dst_u], %[dst_u], 16 \n"
"beqz %[y], 3f \n"
" nop \n"
"b 2f \n"
" nop \n"
"2: \n"
"lbu $t0, -2(%[src_uv]) \n"
"lbu $t1, -1(%[src_uv]) \n"
"addiu %[src_uv], %[src_uv], -2 \n"
"addiu %[y], %[y], -1 \n"
"sb $t0, 0(%[dst_u]) \n"
"sb $t1, 0(%[dst_v]) \n"
"addiu %[dst_u], %[dst_u], 1 \n"
"bgtz %[y], 2b \n"
" addiu %[dst_v], %[dst_v], 1 \n"
"3: \n"
".set pop \n"
: [src_uv] "+r" (src_uv),
[dst_u] "+r" (dst_u),
[dst_v] "+r" (dst_v),
[x] "=&r" (x),
[y] "=&r" (y)
: [width] "r" (width)
: "t0", "t1", "t2", "t3", "t4",
"t5", "t7", "t8", "t9"
);
}
// Convert (4 Y and 2 VU) I422 and arrange RGB values into
// t5 = | 0 | B0 | 0 | b0 |
// t4 = | 0 | B1 | 0 | b1 |
// t9 = | 0 | G0 | 0 | g0 |
// t8 = | 0 | G1 | 0 | g1 |
// t2 = | 0 | R0 | 0 | r0 |
// t1 = | 0 | R1 | 0 | r1 |
#define YUVTORGB \
"lw $t0, 0(%[y_buf]) \n" \
"lhu $t1, 0(%[u_buf]) \n" \
"lhu $t2, 0(%[v_buf]) \n" \
"preceu.ph.qbr $t1, $t1 \n" \
"preceu.ph.qbr $t2, $t2 \n" \
"preceu.ph.qbra $t3, $t0 \n" \
"preceu.ph.qbla $t0, $t0 \n" \
"subu.ph $t1, $t1, $s5 \n" \
"subu.ph $t2, $t2, $s5 \n" \
"subu.ph $t3, $t3, $s4 \n" \
"subu.ph $t0, $t0, $s4 \n" \
"mul.ph $t3, $t3, $s0 \n" \
"mul.ph $t0, $t0, $s0 \n" \
"shll.ph $t4, $t1, 0x7 \n" \
"subu.ph $t4, $t4, $t1 \n" \
"mul.ph $t6, $t1, $s1 \n" \
"mul.ph $t1, $t2, $s2 \n" \
"addq_s.ph $t5, $t4, $t3 \n" \
"addq_s.ph $t4, $t4, $t0 \n" \
"shra.ph $t5, $t5, 6 \n" \
"shra.ph $t4, $t4, 6 \n" \
"addiu %[u_buf], 2 \n" \
"addiu %[v_buf], 2 \n" \
"addu.ph $t6, $t6, $t1 \n" \
"mul.ph $t1, $t2, $s3 \n" \
"addu.ph $t9, $t6, $t3 \n" \
"addu.ph $t8, $t6, $t0 \n" \
"shra.ph $t9, $t9, 6 \n" \
"shra.ph $t8, $t8, 6 \n" \
"addu.ph $t2, $t1, $t3 \n" \
"addu.ph $t1, $t1, $t0 \n" \
"shra.ph $t2, $t2, 6 \n" \
"shra.ph $t1, $t1, 6 \n" \
"subu.ph $t5, $t5, $s5 \n" \
"subu.ph $t4, $t4, $s5 \n" \
"subu.ph $t9, $t9, $s5 \n" \
"subu.ph $t8, $t8, $s5 \n" \
"subu.ph $t2, $t2, $s5 \n" \
"subu.ph $t1, $t1, $s5 \n" \
"shll_s.ph $t5, $t5, 8 \n" \
"shll_s.ph $t4, $t4, 8 \n" \
"shll_s.ph $t9, $t9, 8 \n" \
"shll_s.ph $t8, $t8, 8 \n" \
"shll_s.ph $t2, $t2, 8 \n" \
"shll_s.ph $t1, $t1, 8 \n" \
"shra.ph $t5, $t5, 8 \n" \
"shra.ph $t4, $t4, 8 \n" \
"shra.ph $t9, $t9, 8 \n" \
"shra.ph $t8, $t8, 8 \n" \
"shra.ph $t2, $t2, 8 \n" \
"shra.ph $t1, $t1, 8 \n" \
"addu.ph $t5, $t5, $s5 \n" \
"addu.ph $t4, $t4, $s5 \n" \
"addu.ph $t9, $t9, $s5 \n" \
"addu.ph $t8, $t8, $s5 \n" \
"addu.ph $t2, $t2, $s5 \n" \
"addu.ph $t1, $t1, $s5 \n"
// TODO(fbarchard): accept yuv conversion constants.
void I422ToARGBRow_DSPR2(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"beqz %[width], 2f \n"
" repl.ph $s0, 74 \n" // |YG|YG| = |74|74|
"repl.ph $s1, -25 \n" // |UG|UG| = |-25|-25|
"repl.ph $s2, -52 \n" // |VG|VG| = |-52|-52|
"repl.ph $s3, 102 \n" // |VR|VR| = |102|102|
"repl.ph $s4, 16 \n" // |0|16|0|16|
"repl.ph $s5, 128 \n" // |128|128| // clipping
"lui $s6, 0xff00 \n"
"ori $s6, 0xff00 \n" // |ff|00|ff|00|ff|
"1: \n"
YUVTORGB
// Arranging into argb format
"precr.qb.ph $t4, $t8, $t4 \n" // |G1|g1|B1|b1|
"precr.qb.ph $t5, $t9, $t5 \n" // |G0|g0|B0|b0|
"addiu %[width], -4 \n"
"precrq.qb.ph $t8, $t4, $t5 \n" // |G1|B1|G0|B0|
"precr.qb.ph $t9, $t4, $t5 \n" // |g1|b1|g0|b0|
"precr.qb.ph $t2, $t1, $t2 \n" // |R1|r1|R0|r0|
"addiu %[y_buf], 4 \n"
"preceu.ph.qbla $t1, $t2 \n" // |0 |R1|0 |R0|
"preceu.ph.qbra $t2, $t2 \n" // |0 |r1|0 |r0|
"or $t1, $t1, $s6 \n" // |ff|R1|ff|R0|
"or $t2, $t2, $s6 \n" // |ff|r1|ff|r0|
"precrq.ph.w $t0, $t2, $t9 \n" // |ff|r1|g1|b1|
"precrq.ph.w $t3, $t1, $t8 \n" // |ff|R1|G1|B1|
"sll $t9, $t9, 16 \n"
"sll $t8, $t8, 16 \n"
"packrl.ph $t2, $t2, $t9 \n" // |ff|r0|g0|b0|
"packrl.ph $t1, $t1, $t8 \n" // |ff|R0|G0|B0|
// Store results.
"sw $t2, 0(%[rgb_buf]) \n"
"sw $t0, 4(%[rgb_buf]) \n"
"sw $t1, 8(%[rgb_buf]) \n"
"sw $t3, 12(%[rgb_buf]) \n"
"bnez %[width], 1b \n"
" addiu %[rgb_buf], 16 \n"
"2: \n"
".set pop \n"
:[y_buf] "+r" (y_buf),
[u_buf] "+r" (u_buf),
[v_buf] "+r" (v_buf),
[width] "+r" (width),
[rgb_buf] "+r" (rgb_buf)
:
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9",
"s0", "s1", "s2", "s3",
"s4", "s5", "s6"
);
}
// Bilinear filter 8x2 -> 8x1
void InterpolateRow_DSPR2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) {
int y0_fraction = 256 - source_y_fraction;
const uint8* src_ptr1 = src_ptr + src_stride;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"replv.ph $t0, %[y0_fraction] \n"
"replv.ph $t1, %[source_y_fraction] \n"
"1: \n"
"lw $t2, 0(%[src_ptr]) \n"
"lw $t3, 0(%[src_ptr1]) \n"
"lw $t4, 4(%[src_ptr]) \n"
"lw $t5, 4(%[src_ptr1]) \n"
"muleu_s.ph.qbl $t6, $t2, $t0 \n"
"muleu_s.ph.qbr $t7, $t2, $t0 \n"
"muleu_s.ph.qbl $t8, $t3, $t1 \n"
"muleu_s.ph.qbr $t9, $t3, $t1 \n"
"muleu_s.ph.qbl $t2, $t4, $t0 \n"
"muleu_s.ph.qbr $t3, $t4, $t0 \n"
"muleu_s.ph.qbl $t4, $t5, $t1 \n"
"muleu_s.ph.qbr $t5, $t5, $t1 \n"
"addq.ph $t6, $t6, $t8 \n"
"addq.ph $t7, $t7, $t9 \n"
"addq.ph $t2, $t2, $t4 \n"
"addq.ph $t3, $t3, $t5 \n"
"shra.ph $t6, $t6, 8 \n"
"shra.ph $t7, $t7, 8 \n"
"shra.ph $t2, $t2, 8 \n"
"shra.ph $t3, $t3, 8 \n"
"precr.qb.ph $t6, $t6, $t7 \n"
"precr.qb.ph $t2, $t2, $t3 \n"
"addiu %[src_ptr], %[src_ptr], 8 \n"
"addiu %[src_ptr1], %[src_ptr1], 8 \n"
"addiu %[dst_width], %[dst_width], -8 \n"
"sw $t6, 0(%[dst_ptr]) \n"
"sw $t2, 4(%[dst_ptr]) \n"
"bgtz %[dst_width], 1b \n"
" addiu %[dst_ptr], %[dst_ptr], 8 \n"
".set pop \n"
: [dst_ptr] "+r" (dst_ptr),
[src_ptr1] "+r" (src_ptr1),
[src_ptr] "+r" (src_ptr),
[dst_width] "+r" (dst_width)
: [source_y_fraction] "r" (source_y_fraction),
[y0_fraction] "r" (y0_fraction),
[src_stride] "r" (src_stride)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
}
#endif // __mips_dsp_rev >= 2
#endif // defined(__mips__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
libs/libyuv/source/row_mmi.cc
+6042
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/row_msa.cc
+3455 -4
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/row_neon.cc
+2112 -2262
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/row_neon64.cc
+2111 -2036
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/row_win.cc
+1954 -1989
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/scale.cc
+599 -449
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/scale_any.cc
+477 -123
View File
@@ -8,6 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include <string.h> // For memset/memcpy
#include "libyuv/scale.h"
#include "libyuv/scale_row.h"
@@ -20,184 +22,533 @@ extern "C" {
// Definition for ScaleFilterCols, ScaleARGBCols and ScaleARGBFilterCols
#define CANY(NAMEANY, TERP_SIMD, TERP_C, BPP, MASK) \
void NAMEANY(uint8* dst_ptr, const uint8* src_ptr, \
int dst_width, int x, int dx) { \
int n = dst_width & ~MASK; \
if (n > 0) { \
TERP_SIMD(dst_ptr, src_ptr, n, x, dx); \
} \
TERP_C(dst_ptr + n * BPP, src_ptr, \
dst_width & MASK, x + n * dx, dx); \
}
void NAMEANY(uint8_t* dst_ptr, const uint8_t* src_ptr, int dst_width, int x, \
int dx) { \
int r = dst_width & MASK; \
int n = dst_width & ~MASK; \
if (n > 0) { \
TERP_SIMD(dst_ptr, src_ptr, n, x, dx); \
} \
TERP_C(dst_ptr + n * BPP, src_ptr, r, x + n * dx, dx); \
}
#ifdef HAS_SCALEFILTERCOLS_NEON
CANY(ScaleFilterCols_Any_NEON, ScaleFilterCols_NEON, ScaleFilterCols_C, 1, 7)
#endif
#ifdef HAS_SCALEFILTERCOLS_MSA
CANY(ScaleFilterCols_Any_MSA, ScaleFilterCols_MSA, ScaleFilterCols_C, 1, 15)
#endif
#ifdef HAS_SCALEARGBCOLS_NEON
CANY(ScaleARGBCols_Any_NEON, ScaleARGBCols_NEON, ScaleARGBCols_C, 4, 7)
#endif
#ifdef HAS_SCALEARGBCOLS_MSA
CANY(ScaleARGBCols_Any_MSA, ScaleARGBCols_MSA, ScaleARGBCols_C, 4, 3)
#endif
#ifdef HAS_SCALEARGBCOLS_MMI
CANY(ScaleARGBCols_Any_MMI, ScaleARGBCols_MMI, ScaleARGBCols_C, 4, 0)
#endif
#ifdef HAS_SCALEARGBFILTERCOLS_NEON
CANY(ScaleARGBFilterCols_Any_NEON, ScaleARGBFilterCols_NEON,
ScaleARGBFilterCols_C, 4, 3)
CANY(ScaleARGBFilterCols_Any_NEON,
ScaleARGBFilterCols_NEON,
ScaleARGBFilterCols_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBFILTERCOLS_MSA
CANY(ScaleARGBFilterCols_Any_MSA,
ScaleARGBFilterCols_MSA,
ScaleARGBFilterCols_C,
4,
7)
#endif
#undef CANY
// Fixed scale down.
// Mask may be non-power of 2, so use MOD
#define SDANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, \
uint8* dst_ptr, int dst_width) { \
int r = (int)((unsigned int)dst_width % (MASK + 1)); \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r); \
}
void NAMEANY(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, \
int dst_width) { \
int r = (int)((unsigned int)dst_width % (MASK + 1)); /* NOLINT */ \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r); \
}
// Fixed scale down for odd source width. Used by I420Blend subsampling.
// Since dst_width is (width + 1) / 2, this function scales one less pixel
// and copies the last pixel.
#define SDODD(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, \
uint8* dst_ptr, int dst_width) { \
int r = (int)((unsigned int)(dst_width - 1) % (MASK + 1)); \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r); \
}
void NAMEANY(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, \
int dst_width) { \
int r = (int)((unsigned int)(dst_width - 1) % (MASK + 1)); /* NOLINT */ \
int n = (dst_width - 1) - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r + 1); \
}
#ifdef HAS_SCALEROWDOWN2_SSSE3
SDANY(ScaleRowDown2_Any_SSSE3, ScaleRowDown2_SSSE3, ScaleRowDown2_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_SSSE3, ScaleRowDown2Linear_SSSE3,
ScaleRowDown2Linear_C, 2, 1, 15)
SDANY(ScaleRowDown2Box_Any_SSSE3, ScaleRowDown2Box_SSSE3, ScaleRowDown2Box_C,
2, 1, 15)
SDODD(ScaleRowDown2Box_Odd_SSSE3, ScaleRowDown2Box_SSSE3,
ScaleRowDown2Box_Odd_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_SSSE3,
ScaleRowDown2Linear_SSSE3,
ScaleRowDown2Linear_C,
2,
1,
15)
SDANY(ScaleRowDown2Box_Any_SSSE3,
ScaleRowDown2Box_SSSE3,
ScaleRowDown2Box_C,
2,
1,
15)
SDODD(ScaleRowDown2Box_Odd_SSSE3,
ScaleRowDown2Box_SSSE3,
ScaleRowDown2Box_Odd_C,
2,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN2_AVX2
SDANY(ScaleRowDown2_Any_AVX2, ScaleRowDown2_AVX2, ScaleRowDown2_C, 2, 1, 31)
SDANY(ScaleRowDown2Linear_Any_AVX2, ScaleRowDown2Linear_AVX2,
ScaleRowDown2Linear_C, 2, 1, 31)
SDANY(ScaleRowDown2Box_Any_AVX2, ScaleRowDown2Box_AVX2, ScaleRowDown2Box_C,
2, 1, 31)
SDODD(ScaleRowDown2Box_Odd_AVX2, ScaleRowDown2Box_AVX2, ScaleRowDown2Box_Odd_C,
2, 1, 31)
SDANY(ScaleRowDown2Linear_Any_AVX2,
ScaleRowDown2Linear_AVX2,
ScaleRowDown2Linear_C,
2,
1,
31)
SDANY(ScaleRowDown2Box_Any_AVX2,
ScaleRowDown2Box_AVX2,
ScaleRowDown2Box_C,
2,
1,
31)
SDODD(ScaleRowDown2Box_Odd_AVX2,
ScaleRowDown2Box_AVX2,
ScaleRowDown2Box_Odd_C,
2,
1,
31)
#endif
#ifdef HAS_SCALEROWDOWN2_NEON
SDANY(ScaleRowDown2_Any_NEON, ScaleRowDown2_NEON, ScaleRowDown2_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_NEON, ScaleRowDown2Linear_NEON,
ScaleRowDown2Linear_C, 2, 1, 15)
SDANY(ScaleRowDown2Box_Any_NEON, ScaleRowDown2Box_NEON,
ScaleRowDown2Box_C, 2, 1, 15)
SDODD(ScaleRowDown2Box_Odd_NEON, ScaleRowDown2Box_NEON,
ScaleRowDown2Box_Odd_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_NEON,
ScaleRowDown2Linear_NEON,
ScaleRowDown2Linear_C,
2,
1,
15)
SDANY(ScaleRowDown2Box_Any_NEON,
ScaleRowDown2Box_NEON,
ScaleRowDown2Box_C,
2,
1,
15)
SDODD(ScaleRowDown2Box_Odd_NEON,
ScaleRowDown2Box_NEON,
ScaleRowDown2Box_Odd_C,
2,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN2_MSA
SDANY(ScaleRowDown2_Any_MSA, ScaleRowDown2_MSA, ScaleRowDown2_C, 2, 1, 31)
SDANY(ScaleRowDown2Linear_Any_MSA,
ScaleRowDown2Linear_MSA,
ScaleRowDown2Linear_C,
2,
1,
31)
SDANY(ScaleRowDown2Box_Any_MSA,
ScaleRowDown2Box_MSA,
ScaleRowDown2Box_C,
2,
1,
31)
#endif
#ifdef HAS_SCALEROWDOWN2_MMI
SDANY(ScaleRowDown2_Any_MMI, ScaleRowDown2_MMI, ScaleRowDown2_C, 2, 1, 7)
SDANY(ScaleRowDown2Linear_Any_MMI,
ScaleRowDown2Linear_MMI,
ScaleRowDown2Linear_C,
2,
1,
7)
SDANY(ScaleRowDown2Box_Any_MMI,
ScaleRowDown2Box_MMI,
ScaleRowDown2Box_C,
2,
1,
7)
SDODD(ScaleRowDown2Box_Odd_MMI,
ScaleRowDown2Box_MMI,
ScaleRowDown2Box_Odd_C,
2,
1,
7)
#endif
#ifdef HAS_SCALEROWDOWN4_SSSE3
SDANY(ScaleRowDown4_Any_SSSE3, ScaleRowDown4_SSSE3, ScaleRowDown4_C, 4, 1, 7)
SDANY(ScaleRowDown4Box_Any_SSSE3, ScaleRowDown4Box_SSSE3, ScaleRowDown4Box_C,
4, 1, 7)
SDANY(ScaleRowDown4Box_Any_SSSE3,
ScaleRowDown4Box_SSSE3,
ScaleRowDown4Box_C,
4,
1,
7)
#endif
#ifdef HAS_SCALEROWDOWN4_AVX2
SDANY(ScaleRowDown4_Any_AVX2, ScaleRowDown4_AVX2, ScaleRowDown4_C, 4, 1, 15)
SDANY(ScaleRowDown4Box_Any_AVX2, ScaleRowDown4Box_AVX2, ScaleRowDown4Box_C,
4, 1, 15)
SDANY(ScaleRowDown4Box_Any_AVX2,
ScaleRowDown4Box_AVX2,
ScaleRowDown4Box_C,
4,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN4_NEON
SDANY(ScaleRowDown4_Any_NEON, ScaleRowDown4_NEON, ScaleRowDown4_C, 4, 1, 7)
SDANY(ScaleRowDown4Box_Any_NEON, ScaleRowDown4Box_NEON, ScaleRowDown4Box_C,
4, 1, 7)
SDANY(ScaleRowDown4Box_Any_NEON,
ScaleRowDown4Box_NEON,
ScaleRowDown4Box_C,
4,
1,
7)
#endif
#ifdef HAS_SCALEROWDOWN4_MSA
SDANY(ScaleRowDown4_Any_MSA, ScaleRowDown4_MSA, ScaleRowDown4_C, 4, 1, 15)
SDANY(ScaleRowDown4Box_Any_MSA,
ScaleRowDown4Box_MSA,
ScaleRowDown4Box_C,
4,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN4_MMI
SDANY(ScaleRowDown4_Any_MMI, ScaleRowDown4_MMI, ScaleRowDown4_C, 4, 1, 7)
SDANY(ScaleRowDown4Box_Any_MMI,
ScaleRowDown4Box_MMI,
ScaleRowDown4Box_C,
4,
1,
7)
#endif
#ifdef HAS_SCALEROWDOWN34_SSSE3
SDANY(ScaleRowDown34_Any_SSSE3, ScaleRowDown34_SSSE3,
ScaleRowDown34_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_0_Box_Any_SSSE3, ScaleRowDown34_0_Box_SSSE3,
ScaleRowDown34_0_Box_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_1_Box_Any_SSSE3, ScaleRowDown34_1_Box_SSSE3,
ScaleRowDown34_1_Box_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_Any_SSSE3,
ScaleRowDown34_SSSE3,
ScaleRowDown34_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_0_Box_Any_SSSE3,
ScaleRowDown34_0_Box_SSSE3,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_1_Box_Any_SSSE3,
ScaleRowDown34_1_Box_SSSE3,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
23)
#endif
#ifdef HAS_SCALEROWDOWN34_NEON
SDANY(ScaleRowDown34_Any_NEON, ScaleRowDown34_NEON,
ScaleRowDown34_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_0_Box_Any_NEON, ScaleRowDown34_0_Box_NEON,
ScaleRowDown34_0_Box_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_1_Box_Any_NEON, ScaleRowDown34_1_Box_NEON,
ScaleRowDown34_1_Box_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_Any_NEON,
ScaleRowDown34_NEON,
ScaleRowDown34_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_0_Box_Any_NEON,
ScaleRowDown34_0_Box_NEON,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_1_Box_Any_NEON,
ScaleRowDown34_1_Box_NEON,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
23)
#endif
#ifdef HAS_SCALEROWDOWN34_MSA
SDANY(ScaleRowDown34_Any_MSA,
ScaleRowDown34_MSA,
ScaleRowDown34_C,
4 / 3,
1,
47)
SDANY(ScaleRowDown34_0_Box_Any_MSA,
ScaleRowDown34_0_Box_MSA,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
47)
SDANY(ScaleRowDown34_1_Box_Any_MSA,
ScaleRowDown34_1_Box_MSA,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
47)
#endif
#ifdef HAS_SCALEROWDOWN38_SSSE3
SDANY(ScaleRowDown38_Any_SSSE3, ScaleRowDown38_SSSE3,
ScaleRowDown38_C, 8 / 3, 1, 11)
SDANY(ScaleRowDown38_3_Box_Any_SSSE3, ScaleRowDown38_3_Box_SSSE3,
ScaleRowDown38_3_Box_C, 8 / 3, 1, 5)
SDANY(ScaleRowDown38_2_Box_Any_SSSE3, ScaleRowDown38_2_Box_SSSE3,
ScaleRowDown38_2_Box_C, 8 / 3, 1, 5)
SDANY(ScaleRowDown38_Any_SSSE3,
ScaleRowDown38_SSSE3,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_SSSE3,
ScaleRowDown38_3_Box_SSSE3,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
5)
SDANY(ScaleRowDown38_2_Box_Any_SSSE3,
ScaleRowDown38_2_Box_SSSE3,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
5)
#endif
#ifdef HAS_SCALEROWDOWN38_NEON
SDANY(ScaleRowDown38_Any_NEON, ScaleRowDown38_NEON,
ScaleRowDown38_C, 8 / 3, 1, 11)
SDANY(ScaleRowDown38_3_Box_Any_NEON, ScaleRowDown38_3_Box_NEON,
ScaleRowDown38_3_Box_C, 8 / 3, 1, 11)
SDANY(ScaleRowDown38_2_Box_Any_NEON, ScaleRowDown38_2_Box_NEON,
ScaleRowDown38_2_Box_C, 8 / 3, 1, 11)
SDANY(ScaleRowDown38_Any_NEON,
ScaleRowDown38_NEON,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_NEON,
ScaleRowDown38_3_Box_NEON,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_2_Box_Any_NEON,
ScaleRowDown38_2_Box_NEON,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
11)
#endif
#ifdef HAS_SCALEROWDOWN38_MSA
SDANY(ScaleRowDown38_Any_MSA,
ScaleRowDown38_MSA,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_MSA,
ScaleRowDown38_3_Box_MSA,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_2_Box_Any_MSA,
ScaleRowDown38_2_Box_MSA,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
11)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_SSE2
SDANY(ScaleARGBRowDown2_Any_SSE2, ScaleARGBRowDown2_SSE2,
ScaleARGBRowDown2_C, 2, 4, 3)
SDANY(ScaleARGBRowDown2Linear_Any_SSE2, ScaleARGBRowDown2Linear_SSE2,
ScaleARGBRowDown2Linear_C, 2, 4, 3)
SDANY(ScaleARGBRowDown2Box_Any_SSE2, ScaleARGBRowDown2Box_SSE2,
ScaleARGBRowDown2Box_C, 2, 4, 3)
SDANY(ScaleARGBRowDown2_Any_SSE2,
ScaleARGBRowDown2_SSE2,
ScaleARGBRowDown2_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Linear_Any_SSE2,
ScaleARGBRowDown2Linear_SSE2,
ScaleARGBRowDown2Linear_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Box_Any_SSE2,
ScaleARGBRowDown2Box_SSE2,
ScaleARGBRowDown2Box_C,
2,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_NEON
SDANY(ScaleARGBRowDown2_Any_NEON, ScaleARGBRowDown2_NEON,
ScaleARGBRowDown2_C, 2, 4, 7)
SDANY(ScaleARGBRowDown2Linear_Any_NEON, ScaleARGBRowDown2Linear_NEON,
ScaleARGBRowDown2Linear_C, 2, 4, 7)
SDANY(ScaleARGBRowDown2Box_Any_NEON, ScaleARGBRowDown2Box_NEON,
ScaleARGBRowDown2Box_C, 2, 4, 7)
SDANY(ScaleARGBRowDown2_Any_NEON,
ScaleARGBRowDown2_NEON,
ScaleARGBRowDown2_C,
2,
4,
7)
SDANY(ScaleARGBRowDown2Linear_Any_NEON,
ScaleARGBRowDown2Linear_NEON,
ScaleARGBRowDown2Linear_C,
2,
4,
7)
SDANY(ScaleARGBRowDown2Box_Any_NEON,
ScaleARGBRowDown2Box_NEON,
ScaleARGBRowDown2Box_C,
2,
4,
7)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_MSA
SDANY(ScaleARGBRowDown2_Any_MSA,
ScaleARGBRowDown2_MSA,
ScaleARGBRowDown2_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Linear_Any_MSA,
ScaleARGBRowDown2Linear_MSA,
ScaleARGBRowDown2Linear_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Box_Any_MSA,
ScaleARGBRowDown2Box_MSA,
ScaleARGBRowDown2Box_C,
2,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_MMI
SDANY(ScaleARGBRowDown2_Any_MMI,
ScaleARGBRowDown2_MMI,
ScaleARGBRowDown2_C,
2,
4,
1)
SDANY(ScaleARGBRowDown2Linear_Any_MMI,
ScaleARGBRowDown2Linear_MMI,
ScaleARGBRowDown2Linear_C,
2,
4,
1)
SDANY(ScaleARGBRowDown2Box_Any_MMI,
ScaleARGBRowDown2Box_MMI,
ScaleARGBRowDown2Box_C,
2,
4,
1)
#endif
#undef SDANY
// Scale down by even scale factor.
#define SDAANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, int src_stepx, \
uint8* dst_ptr, int dst_width) { \
int r = (int)((unsigned int)dst_width % (MASK + 1)); \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, src_stepx, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * src_stepx) * BPP, src_stride, \
src_stepx, dst_ptr + n * BPP, r); \
}
#define SDAANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, BPP, MASK) \
void NAMEANY(const uint8_t* src_ptr, ptrdiff_t src_stride, int src_stepx, \
uint8_t* dst_ptr, int dst_width) { \
int r = dst_width & MASK; \
int n = dst_width & ~MASK; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, src_stepx, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * src_stepx) * BPP, src_stride, src_stepx, \
dst_ptr + n * BPP, r); \
}
#ifdef HAS_SCALEARGBROWDOWNEVEN_SSE2
SDAANY(ScaleARGBRowDownEven_Any_SSE2, ScaleARGBRowDownEven_SSE2,
ScaleARGBRowDownEven_C, 4, 3)
SDAANY(ScaleARGBRowDownEvenBox_Any_SSE2, ScaleARGBRowDownEvenBox_SSE2,
ScaleARGBRowDownEvenBox_C, 4, 3)
SDAANY(ScaleARGBRowDownEven_Any_SSE2,
ScaleARGBRowDownEven_SSE2,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_SSE2,
ScaleARGBRowDownEvenBox_SSE2,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWNEVEN_NEON
SDAANY(ScaleARGBRowDownEven_Any_NEON, ScaleARGBRowDownEven_NEON,
ScaleARGBRowDownEven_C, 4, 3)
SDAANY(ScaleARGBRowDownEvenBox_Any_NEON, ScaleARGBRowDownEvenBox_NEON,
ScaleARGBRowDownEvenBox_C, 4, 3)
SDAANY(ScaleARGBRowDownEven_Any_NEON,
ScaleARGBRowDownEven_NEON,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_NEON,
ScaleARGBRowDownEvenBox_NEON,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWNEVEN_MSA
SDAANY(ScaleARGBRowDownEven_Any_MSA,
ScaleARGBRowDownEven_MSA,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_MSA,
ScaleARGBRowDownEvenBox_MSA,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWNEVEN_MMI
SDAANY(ScaleARGBRowDownEven_Any_MMI,
ScaleARGBRowDownEven_MMI,
ScaleARGBRowDownEven_C,
4,
1)
SDAANY(ScaleARGBRowDownEvenBox_Any_MMI,
ScaleARGBRowDownEvenBox_MMI,
ScaleARGBRowDownEvenBox_C,
4,
1)
#endif
#ifdef SASIMDONLY
// This also works and uses memcpy and SIMD instead of C, but is slower on ARM
// Add rows box filter scale down. Using macro from row_any
#define SAROW(NAMEANY, ANY_SIMD, SBPP, BPP, MASK) \
void NAMEANY(const uint8_t* src_ptr, uint16_t* dst_ptr, int width) { \
SIMD_ALIGNED(uint16_t dst_temp[32]); \
SIMD_ALIGNED(uint8_t src_temp[32]); \
memset(dst_temp, 0, 32 * 2); /* for msan */ \
int r = width & MASK; \
int n = width & ~MASK; \
if (n > 0) { \
ANY_SIMD(src_ptr, dst_ptr, n); \
} \
memcpy(src_temp, src_ptr + n * SBPP, r * SBPP); \
memcpy(dst_temp, dst_ptr + n * BPP, r * BPP); \
ANY_SIMD(src_temp, dst_temp, MASK + 1); \
memcpy(dst_ptr + n * BPP, dst_temp, r * BPP); \
}
#ifdef HAS_SCALEADDROW_SSE2
SAROW(ScaleAddRow_Any_SSE2, ScaleAddRow_SSE2, 1, 2, 15)
#endif
#ifdef HAS_SCALEADDROW_AVX2
SAROW(ScaleAddRow_Any_AVX2, ScaleAddRow_AVX2, 1, 2, 31)
#endif
#ifdef HAS_SCALEADDROW_NEON
SAROW(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, 1, 2, 15)
#endif
#ifdef HAS_SCALEADDROW_MSA
SAROW(ScaleAddRow_Any_MSA, ScaleAddRow_MSA, 1, 2, 15)
#endif
#ifdef HAS_SCALEADDROW_MMI
SAROW(ScaleAddRow_Any_MMI, ScaleAddRow_MMI, 1, 2, 7)
#endif
#undef SAANY
#else
// Add rows box filter scale down.
#define SAANY(NAMEANY, SCALEADDROW_SIMD, SCALEADDROW_C, MASK) \
void NAMEANY(const uint8* src_ptr, uint16* dst_ptr, int src_width) { \
int n = src_width & ~MASK; \
if (n > 0) { \
SCALEADDROW_SIMD(src_ptr, dst_ptr, n); \
} \
SCALEADDROW_C(src_ptr + n, dst_ptr + n, src_width & MASK); \
}
#define SAANY(NAMEANY, SCALEADDROW_SIMD, SCALEADDROW_C, MASK) \
void NAMEANY(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width) { \
int n = src_width & ~MASK; \
if (n > 0) { \
SCALEADDROW_SIMD(src_ptr, dst_ptr, n); \
} \
SCALEADDROW_C(src_ptr + n, dst_ptr + n, src_width & MASK); \
}
#ifdef HAS_SCALEADDROW_SSE2
SAANY(ScaleAddRow_Any_SSE2, ScaleAddRow_SSE2, ScaleAddRow_C, 15)
@@ -208,14 +559,17 @@ SAANY(ScaleAddRow_Any_AVX2, ScaleAddRow_AVX2, ScaleAddRow_C, 31)
#ifdef HAS_SCALEADDROW_NEON
SAANY(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, ScaleAddRow_C, 15)
#endif
#ifdef HAS_SCALEADDROW_MSA
SAANY(ScaleAddRow_Any_MSA, ScaleAddRow_MSA, ScaleAddRow_C, 15)
#endif
#ifdef HAS_SCALEADDROW_MMI
SAANY(ScaleAddRow_Any_MMI, ScaleAddRow_MMI, ScaleAddRow_C, 7)
#endif
#undef SAANY
#endif // SASIMDONLY
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
libs/libyuv/source/scale_argb.cc
+435 -211
View File
@@ -30,20 +30,31 @@ static __inline int Abs(int v) {
// ScaleARGB ARGB, 1/2
// This is an optimized version for scaling down a ARGB to 1/2 of
// its original size.
static void ScaleARGBDown2(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
static void ScaleARGBDown2(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8_t* src_argb,
uint8_t* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
int row_stride = src_stride * (dy >> 16);
void (*ScaleARGBRowDown2)(const uint8* src_argb, ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) =
filtering == kFilterNone ? ScaleARGBRowDown2_C :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_C :
ScaleARGBRowDown2Box_C);
assert(dx == 65536 * 2); // Test scale factor of 2.
void (*ScaleARGBRowDown2)(const uint8_t* src_argb, ptrdiff_t src_stride,
uint8_t* dst_argb, int dst_width) =
filtering == kFilterNone
? ScaleARGBRowDown2_C
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_C
: ScaleARGBRowDown2Box_C);
(void)src_width;
(void)src_height;
(void)dx;
assert(dx == 65536 * 2); // Test scale factor of 2.
assert((dy & 0x1ffff) == 0); // Test vertical scale is multiple of 2.
// Advance to odd row, even column.
if (filtering == kFilterBilinear) {
@@ -54,25 +65,65 @@ static void ScaleARGBDown2(int src_width, int src_height,
#if defined(HAS_SCALEARGBROWDOWN2_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_Any_SSE2 :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_SSE2 :
ScaleARGBRowDown2Box_Any_SSE2);
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_Any_SSE2
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_SSE2
: ScaleARGBRowDown2Box_Any_SSE2);
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_SSE2 :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_SSE2 :
ScaleARGBRowDown2Box_SSE2);
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_SSE2
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_SSE2
: ScaleARGBRowDown2Box_SSE2);
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWN2_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_Any_NEON :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_NEON :
ScaleARGBRowDown2Box_Any_NEON);
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_Any_NEON
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_NEON
: ScaleARGBRowDown2Box_Any_NEON);
if (IS_ALIGNED(dst_width, 8)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_NEON :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_NEON :
ScaleARGBRowDown2Box_NEON);
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_NEON
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_NEON
: ScaleARGBRowDown2Box_NEON);
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWN2_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_Any_MSA
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_MSA
: ScaleARGBRowDown2Box_Any_MSA);
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_MSA
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_MSA
: ScaleARGBRowDown2Box_MSA);
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWN2_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_Any_MMI
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_MMI
: ScaleARGBRowDown2Box_Any_MMI);
if (IS_ALIGNED(dst_width, 2)) {
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_MMI
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_MMI
: ScaleARGBRowDown2Box_MMI);
}
}
#endif
@@ -90,21 +141,32 @@ static void ScaleARGBDown2(int src_width, int src_height,
// ScaleARGB ARGB, 1/4
// This is an optimized version for scaling down a ARGB to 1/4 of
// its original size.
static void ScaleARGBDown4Box(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy) {
static void ScaleARGBDown4Box(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8_t* src_argb,
uint8_t* dst_argb,
int x,
int dx,
int y,
int dy) {
int j;
// Allocate 2 rows of ARGB.
const int kRowSize = (dst_width * 2 * 4 + 31) & ~31;
align_buffer_64(row, kRowSize * 2);
int row_stride = src_stride * (dy >> 16);
void (*ScaleARGBRowDown2)(const uint8* src_argb, ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) = ScaleARGBRowDown2Box_C;
void (*ScaleARGBRowDown2)(const uint8_t* src_argb, ptrdiff_t src_stride,
uint8_t* dst_argb, int dst_width) =
ScaleARGBRowDown2Box_C;
// Advance to odd row, even column.
src_argb += (y >> 16) * src_stride + (x >> 16) * 4;
assert(dx == 65536 * 4); // Test scale factor of 4.
(void)src_width;
(void)src_height;
(void)dx;
assert(dx == 65536 * 4); // Test scale factor of 4.
assert((dy & 0x3ffff) == 0); // Test vertical scale is multiple of 4.
#if defined(HAS_SCALEARGBROWDOWN2_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
@@ -125,8 +187,8 @@ static void ScaleARGBDown4Box(int src_width, int src_height,
for (j = 0; j < dst_height; ++j) {
ScaleARGBRowDown2(src_argb, src_stride, row, dst_width * 2);
ScaleARGBRowDown2(src_argb + src_stride * 2, src_stride,
row + kRowSize, dst_width * 2);
ScaleARGBRowDown2(src_argb + src_stride * 2, src_stride, row + kRowSize,
dst_width * 2);
ScaleARGBRowDown2(row, kRowSize, dst_argb, dst_width);
src_argb += row_stride;
dst_argb += dst_stride;
@@ -137,38 +199,67 @@ static void ScaleARGBDown4Box(int src_width, int src_height,
// ScaleARGB ARGB Even
// This is an optimized version for scaling down a ARGB to even
// multiple of its original size.
static void ScaleARGBDownEven(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
static void ScaleARGBDownEven(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8_t* src_argb,
uint8_t* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
int col_step = dx >> 16;
int row_stride = (dy >> 16) * src_stride;
void (*ScaleARGBRowDownEven)(const uint8* src_argb, ptrdiff_t src_stride,
int src_step, uint8* dst_argb, int dst_width) =
void (*ScaleARGBRowDownEven)(const uint8_t* src_argb, ptrdiff_t src_stride,
int src_step, uint8_t* dst_argb, int dst_width) =
filtering ? ScaleARGBRowDownEvenBox_C : ScaleARGBRowDownEven_C;
(void)src_width;
(void)src_height;
assert(IS_ALIGNED(src_width, 2));
assert(IS_ALIGNED(src_height, 2));
src_argb += (y >> 16) * src_stride + (x >> 16) * 4;
#if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_SSE2 :
ScaleARGBRowDownEven_Any_SSE2;
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_SSE2
: ScaleARGBRowDownEven_Any_SSE2;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_SSE2 :
ScaleARGBRowDownEven_SSE2;
ScaleARGBRowDownEven =
filtering ? ScaleARGBRowDownEvenBox_SSE2 : ScaleARGBRowDownEven_SSE2;
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWNEVEN_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_NEON :
ScaleARGBRowDownEven_Any_NEON;
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_NEON
: ScaleARGBRowDownEven_Any_NEON;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_NEON :
ScaleARGBRowDownEven_NEON;
ScaleARGBRowDownEven =
filtering ? ScaleARGBRowDownEvenBox_NEON : ScaleARGBRowDownEven_NEON;
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWNEVEN_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_MSA
: ScaleARGBRowDownEven_Any_MSA;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDownEven =
filtering ? ScaleARGBRowDownEvenBox_MSA : ScaleARGBRowDownEven_MSA;
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWNEVEN_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_MMI
: ScaleARGBRowDownEven_Any_MMI;
if (IS_ALIGNED(dst_width, 2)) {
ScaleARGBRowDownEven =
filtering ? ScaleARGBRowDownEvenBox_MMI : ScaleARGBRowDownEven_MMI;
}
}
#endif
@@ -184,25 +275,32 @@ static void ScaleARGBDownEven(int src_width, int src_height,
}
// Scale ARGB down with bilinear interpolation.
static void ScaleARGBBilinearDown(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
static void ScaleARGBBilinearDown(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8_t* src_argb,
uint8_t* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
InterpolateRow_C;
void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
void (*InterpolateRow)(uint8_t * dst_argb, const uint8_t* src_argb,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_C;
void (*ScaleARGBFilterCols)(uint8_t * dst_argb, const uint8_t* src_argb,
int dst_width, int x, int dx) =
(src_width >= 32768) ? ScaleARGBFilterCols64_C : ScaleARGBFilterCols_C;
int64 xlast = x + (int64)(dst_width - 1) * dx;
int64 xl = (dx >= 0) ? x : xlast;
int64 xr = (dx >= 0) ? xlast : x;
int64_t xlast = x + (int64_t)(dst_width - 1) * dx;
int64_t xl = (dx >= 0) ? x : xlast;
int64_t xr = (dx >= 0) ? xlast : x;
int clip_src_width;
xl = (xl >> 16) & ~3; // Left edge aligned.
xr = (xr >> 16) + 1; // Right most pixel used. Bilinear uses 2 pixels.
xl = (xl >> 16) & ~3; // Left edge aligned.
xr = (xr >> 16) + 1; // Right most pixel used. Bilinear uses 2 pixels.
xr = (xr + 1 + 3) & ~3; // 1 beyond 4 pixel aligned right most pixel.
if (xr > src_width) {
xr = src_width;
@@ -234,12 +332,11 @@ static void ScaleARGBBilinearDown(int src_width, int src_height,
}
}
#endif
#if defined(HAS_INTERPOLATEROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4)) {
InterpolateRow = InterpolateRow_Any_DSPR2;
if (IS_ALIGNED(clip_src_width, 4)) {
InterpolateRow = InterpolateRow_DSPR2;
#if defined(HAS_INTERPOLATEROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
InterpolateRow = InterpolateRow_Any_MSA;
if (IS_ALIGNED(clip_src_width, 32)) {
InterpolateRow = InterpolateRow_MSA;
}
}
#endif
@@ -255,6 +352,14 @@ static void ScaleARGBBilinearDown(int src_width, int src_height,
ScaleARGBFilterCols = ScaleARGBFilterCols_NEON;
}
}
#endif
#if defined(HAS_SCALEARGBFILTERCOLS_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ScaleARGBFilterCols = ScaleARGBFilterCols_Any_MSA;
if (IS_ALIGNED(dst_width, 8)) {
ScaleARGBFilterCols = ScaleARGBFilterCols_MSA;
}
}
#endif
// TODO(fbarchard): Consider not allocating row buffer for kFilterLinear.
// Allocate a row of ARGB.
@@ -267,7 +372,7 @@ static void ScaleARGBBilinearDown(int src_width, int src_height,
}
for (j = 0; j < dst_height; ++j) {
int yi = y >> 16;
const uint8* src = src_argb + yi * src_stride;
const uint8_t* src = src_argb + yi * src_stride;
if (filtering == kFilterLinear) {
ScaleARGBFilterCols(dst_argb, src, dst_width, x, dx);
} else {
@@ -286,18 +391,25 @@ static void ScaleARGBBilinearDown(int src_width, int src_height,
}
// Scale ARGB up with bilinear interpolation.
static void ScaleARGBBilinearUp(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
static void ScaleARGBBilinearUp(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8_t* src_argb,
uint8_t* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
InterpolateRow_C;
void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
void (*InterpolateRow)(uint8_t * dst_argb, const uint8_t* src_argb,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_C;
void (*ScaleARGBFilterCols)(uint8_t * dst_argb, const uint8_t* src_argb,
int dst_width, int x, int dx) =
filtering ? ScaleARGBFilterCols_C : ScaleARGBCols_C;
const int max_y = (src_height - 1) << 16;
#if defined(HAS_INTERPOLATEROW_SSSE3)
@@ -324,15 +436,25 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
}
}
#endif
#if defined(HAS_INTERPOLATEROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) {
InterpolateRow = InterpolateRow_DSPR2;
#if defined(HAS_INTERPOLATEROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
InterpolateRow = InterpolateRow_Any_MSA;
if (IS_ALIGNED(dst_width, 8)) {
InterpolateRow = InterpolateRow_MSA;
}
}
#endif
#if defined(HAS_INTERPOLATEROW_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
InterpolateRow = InterpolateRow_Any_MMI;
if (IS_ALIGNED(dst_width, 2)) {
InterpolateRow = InterpolateRow_MMI;
}
}
#endif
if (src_width >= 32768) {
ScaleARGBFilterCols = filtering ?
ScaleARGBFilterCols64_C : ScaleARGBCols64_C;
ScaleARGBFilterCols =
filtering ? ScaleARGBFilterCols64_C : ScaleARGBCols64_C;
}
#if defined(HAS_SCALEARGBFILTERCOLS_SSSE3)
if (filtering && TestCpuFlag(kCpuHasSSSE3) && src_width < 32768) {
@@ -347,6 +469,14 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
}
}
#endif
#if defined(HAS_SCALEARGBFILTERCOLS_MSA)
if (filtering && TestCpuFlag(kCpuHasMSA)) {
ScaleARGBFilterCols = ScaleARGBFilterCols_Any_MSA;
if (IS_ALIGNED(dst_width, 8)) {
ScaleARGBFilterCols = ScaleARGBFilterCols_MSA;
}
}
#endif
#if defined(HAS_SCALEARGBCOLS_SSE2)
if (!filtering && TestCpuFlag(kCpuHasSSE2) && src_width < 32768) {
ScaleARGBFilterCols = ScaleARGBCols_SSE2;
@@ -359,6 +489,22 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
ScaleARGBFilterCols = ScaleARGBCols_NEON;
}
}
#endif
#if defined(HAS_SCALEARGBCOLS_MSA)
if (!filtering && TestCpuFlag(kCpuHasMSA)) {
ScaleARGBFilterCols = ScaleARGBCols_Any_MSA;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBFilterCols = ScaleARGBCols_MSA;
}
}
#endif
#if defined(HAS_SCALEARGBCOLS_MMI)
if (!filtering && TestCpuFlag(kCpuHasMMI)) {
ScaleARGBFilterCols = ScaleARGBCols_Any_MMI;
if (IS_ALIGNED(dst_width, 1)) {
ScaleARGBFilterCols = ScaleARGBCols_MMI;
}
}
#endif
if (!filtering && src_width * 2 == dst_width && x < 0x8000) {
ScaleARGBFilterCols = ScaleARGBColsUp2_C;
@@ -366,6 +512,11 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
ScaleARGBFilterCols = ScaleARGBColsUp2_SSE2;
}
#endif
#if defined(HAS_SCALEARGBCOLSUP2_MMI)
if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 4)) {
ScaleARGBFilterCols = ScaleARGBColsUp2_MMI;
}
#endif
}
@@ -375,13 +526,13 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
{
int yi = y >> 16;
const uint8* src = src_argb + yi * src_stride;
const uint8_t* src = src_argb + yi * src_stride;
// Allocate 2 rows of ARGB.
const int kRowSize = (dst_width * 4 + 31) & ~31;
align_buffer_64(row, kRowSize * 2);
uint8* rowptr = row;
uint8_t* rowptr = row;
int rowstride = kRowSize;
int lasty = yi;
@@ -423,24 +574,27 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
#ifdef YUVSCALEUP
// Scale YUV to ARGB up with bilinear interpolation.
static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
int dst_width, int dst_height,
static void ScaleYUVToARGBBilinearUp(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride_y,
int src_stride_u,
int src_stride_v,
int dst_stride_argb,
const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb,
int x, int dx, int y, int dy,
const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
uint8_t* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
void (*I422ToARGBRow)(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
int width) = I422ToARGBRow_C;
void (*I422ToARGBRow)(const uint8_t* y_buf, const uint8_t* u_buf,
const uint8_t* v_buf, uint8_t* rgb_buf, int width) =
I422ToARGBRow_C;
#if defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
I422ToARGBRow = I422ToARGBRow_Any_SSSE3;
@@ -465,19 +619,18 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
}
}
#endif
#if defined(HAS_I422TOARGBROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_width, 4) &&
IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) &&
IS_ALIGNED(src_u, 2) && IS_ALIGNED(src_stride_u, 2) &&
IS_ALIGNED(src_v, 2) && IS_ALIGNED(src_stride_v, 2) &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) {
I422ToARGBRow = I422ToARGBRow_DSPR2;
#if defined(HAS_I422TOARGBROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
I422ToARGBRow = I422ToARGBRow_Any_MSA;
if (IS_ALIGNED(src_width, 8)) {
I422ToARGBRow = I422ToARGBRow_MSA;
}
}
#endif
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
InterpolateRow_C;
void (*InterpolateRow)(uint8_t * dst_argb, const uint8_t* src_argb,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_C;
#if defined(HAS_INTERPOLATEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
InterpolateRow = InterpolateRow_Any_SSSE3;
@@ -502,19 +655,21 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
}
}
#endif
#if defined(HAS_INTERPOLATEROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) {
InterpolateRow = InterpolateRow_DSPR2;
#if defined(HAS_INTERPOLATEROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
InterpolateRow = InterpolateRow_Any_MSA;
if (IS_ALIGNED(dst_width, 8)) {
InterpolateRow = InterpolateRow_MSA;
}
}
#endif
void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
void (*ScaleARGBFilterCols)(uint8_t * dst_argb, const uint8_t* src_argb,
int dst_width, int x, int dx) =
filtering ? ScaleARGBFilterCols_C : ScaleARGBCols_C;
if (src_width >= 32768) {
ScaleARGBFilterCols = filtering ?
ScaleARGBFilterCols64_C : ScaleARGBCols64_C;
ScaleARGBFilterCols =
filtering ? ScaleARGBFilterCols64_C : ScaleARGBCols64_C;
}
#if defined(HAS_SCALEARGBFILTERCOLS_SSSE3)
if (filtering && TestCpuFlag(kCpuHasSSSE3) && src_width < 32768) {
@@ -529,6 +684,14 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
}
}
#endif
#if defined(HAS_SCALEARGBFILTERCOLS_MSA)
if (filtering && TestCpuFlag(kCpuHasMSA)) {
ScaleARGBFilterCols = ScaleARGBFilterCols_Any_MSA;
if (IS_ALIGNED(dst_width, 8)) {
ScaleARGBFilterCols = ScaleARGBFilterCols_MSA;
}
}
#endif
#if defined(HAS_SCALEARGBCOLS_SSE2)
if (!filtering && TestCpuFlag(kCpuHasSSE2) && src_width < 32768) {
ScaleARGBFilterCols = ScaleARGBCols_SSE2;
@@ -541,6 +704,22 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
ScaleARGBFilterCols = ScaleARGBCols_NEON;
}
}
#endif
#if defined(HAS_SCALEARGBCOLS_MSA)
if (!filtering && TestCpuFlag(kCpuHasMSA)) {
ScaleARGBFilterCols = ScaleARGBCols_Any_MSA;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBFilterCols = ScaleARGBCols_MSA;
}
}
#endif
#if defined(HAS_SCALEARGBCOLS_MMI)
if (!filtering && TestCpuFlag(kCpuHasMMI)) {
ScaleARGBFilterCols = ScaleARGBCols_Any_MMI;
if (IS_ALIGNED(dst_width, 1)) {
ScaleARGBFilterCols = ScaleARGBCols_MMI;
}
}
#endif
if (!filtering && src_width * 2 == dst_width && x < 0x8000) {
ScaleARGBFilterCols = ScaleARGBColsUp2_C;
@@ -548,6 +727,11 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
ScaleARGBFilterCols = ScaleARGBColsUp2_SSE2;
}
#endif
#if defined(HAS_SCALEARGBCOLSUP2_MMI)
if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 4)) {
ScaleARGBFilterCols = ScaleARGBColsUp2_MMI;
}
#endif
}
@@ -558,9 +742,9 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
const int kYShift = 1; // Shift Y by 1 to convert Y plane to UV coordinate.
int yi = y >> 16;
int uv_yi = yi >> kYShift;
const uint8* src_row_y = src_y + yi * src_stride_y;
const uint8* src_row_u = src_u + uv_yi * src_stride_u;
const uint8* src_row_v = src_v + uv_yi * src_stride_v;
const uint8_t* src_row_y = src_y + yi * src_stride_y;
const uint8_t* src_row_u = src_u + uv_yi * src_stride_u;
const uint8_t* src_row_v = src_v + uv_yi * src_stride_v;
// Allocate 2 rows of ARGB.
const int kRowSize = (dst_width * 4 + 31) & ~31;
@@ -569,7 +753,7 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
// Allocate 1 row of ARGB for source conversion.
align_buffer_64(argb_row, src_width * 4);
uint8* rowptr = row;
uint8_t* rowptr = row;
int rowstride = kRowSize;
int lasty = yi;
@@ -635,15 +819,23 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
// of x and dx is the integer part of the source position and
// the lower 16 bits are the fixed decimal part.
static void ScaleARGBSimple(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy) {
static void ScaleARGBSimple(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8_t* src_argb,
uint8_t* dst_argb,
int x,
int dx,
int y,
int dy) {
int j;
void (*ScaleARGBCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
void (*ScaleARGBCols)(uint8_t * dst_argb, const uint8_t* src_argb,
int dst_width, int x, int dx) =
(src_width >= 32768) ? ScaleARGBCols64_C : ScaleARGBCols_C;
(void)src_height;
#if defined(HAS_SCALEARGBCOLS_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && src_width < 32768) {
ScaleARGBCols = ScaleARGBCols_SSE2;
@@ -656,6 +848,22 @@ static void ScaleARGBSimple(int src_width, int src_height,
ScaleARGBCols = ScaleARGBCols_NEON;
}
}
#endif
#if defined(HAS_SCALEARGBCOLS_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ScaleARGBCols = ScaleARGBCols_Any_MSA;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBCols = ScaleARGBCols_MSA;
}
}
#endif
#if defined(HAS_SCALEARGBCOLS_MMI)
if (TestCpuFlag(kCpuHasMMI)) {
ScaleARGBCols = ScaleARGBCols_Any_MMI;
if (IS_ALIGNED(dst_width, 1)) {
ScaleARGBCols = ScaleARGBCols_MMI;
}
}
#endif
if (src_width * 2 == dst_width && x < 0x8000) {
ScaleARGBCols = ScaleARGBColsUp2_C;
@@ -663,12 +871,17 @@ static void ScaleARGBSimple(int src_width, int src_height,
if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(dst_width, 8)) {
ScaleARGBCols = ScaleARGBColsUp2_SSE2;
}
#endif
#if defined(HAS_SCALEARGBCOLSUP2_MMI)
if (TestCpuFlag(kCpuHasMMI) && IS_ALIGNED(dst_width, 4)) {
ScaleARGBCols = ScaleARGBColsUp2_MMI;
}
#endif
}
for (j = 0; j < dst_height; ++j) {
ScaleARGBCols(dst_argb, src_argb + (y >> 16) * src_stride,
dst_width, x, dx);
ScaleARGBCols(dst_argb, src_argb + (y >> 16) * src_stride, dst_width, x,
dx);
dst_argb += dst_stride;
y += dy;
}
@@ -677,11 +890,18 @@ static void ScaleARGBSimple(int src_width, int src_height,
// ScaleARGB a ARGB.
// This function in turn calls a scaling function
// suitable for handling the desired resolutions.
static void ScaleARGB(const uint8* src, int src_stride,
int src_width, int src_height,
uint8* dst, int dst_stride,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
static void ScaleARGB(const uint8_t* src,
int src_stride,
int src_width,
int src_height,
uint8_t* dst,
int dst_stride,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering) {
// Initial source x/y coordinate and step values as 16.16 fixed point.
int x = 0;
@@ -690,8 +910,7 @@ static void ScaleARGB(const uint8* src, int src_stride,
int dy = 0;
// ARGB does not support box filter yet, but allow the user to pass it.
// Simplify filtering when possible.
filtering = ScaleFilterReduce(src_width, src_height,
dst_width, dst_height,
filtering = ScaleFilterReduce(src_width, src_height, dst_width, dst_height,
filtering);
// Negative src_height means invert the image.
@@ -700,17 +919,17 @@ static void ScaleARGB(const uint8* src, int src_stride,
src = src + (src_height - 1) * src_stride;
src_stride = -src_stride;
}
ScaleSlope(src_width, src_height, dst_width, dst_height, filtering,
&x, &y, &dx, &dy);
ScaleSlope(src_width, src_height, dst_width, dst_height, filtering, &x, &y,
&dx, &dy);
src_width = Abs(src_width);
if (clip_x) {
int64 clipf = (int64)(clip_x) * dx;
int64_t clipf = (int64_t)(clip_x)*dx;
x += (clipf & 0xffff);
src += (clipf >> 16) * 4;
dst += clip_x * 4;
}
if (clip_y) {
int64 clipf = (int64)(clip_y) * dy;
int64_t clipf = (int64_t)(clip_y)*dy;
y += (clipf & 0xffff);
src += (clipf >> 16) * src_stride;
dst += clip_y * dst_stride;
@@ -725,24 +944,20 @@ static void ScaleARGB(const uint8* src, int src_stride,
if (!(dx & 0x10000) && !(dy & 0x10000)) {
if (dx == 0x20000) {
// Optimized 1/2 downsample.
ScaleARGBDown2(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy, filtering);
ScaleARGBDown2(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy,
filtering);
return;
}
if (dx == 0x40000 && filtering == kFilterBox) {
// Optimized 1/4 box downsample.
ScaleARGBDown4Box(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy);
ScaleARGBDown4Box(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy);
return;
}
ScaleARGBDownEven(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy, filtering);
ScaleARGBDownEven(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy,
filtering);
return;
}
// Optimized odd scale down. ie 3, 5, 7, 9x.
@@ -759,96 +974,105 @@ static void ScaleARGB(const uint8* src, int src_stride,
}
if (dx == 0x10000 && (x & 0xffff) == 0) {
// Arbitrary scale vertically, but unscaled vertically.
ScalePlaneVertical(src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, y, dy, 4, filtering);
ScalePlaneVertical(src_height, clip_width, clip_height, src_stride,
dst_stride, src, dst, x, y, dy, 4, filtering);
return;
}
if (filtering && dy < 65536) {
ScaleARGBBilinearUp(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy, filtering);
ScaleARGBBilinearUp(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy,
filtering);
return;
}
if (filtering) {
ScaleARGBBilinearDown(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy, filtering);
ScaleARGBBilinearDown(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy,
filtering);
return;
}
ScaleARGBSimple(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy);
ScaleARGBSimple(src_width, src_height, clip_width, clip_height, src_stride,
dst_stride, src, dst, x, dx, y, dy);
}
LIBYUV_API
int ARGBScaleClip(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
int ARGBScaleClip(const uint8_t* src_argb,
int src_stride_argb,
int src_width,
int src_height,
uint8_t* dst_argb,
int dst_stride_argb,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering) {
if (!src_argb || src_width == 0 || src_height == 0 ||
!dst_argb || dst_width <= 0 || dst_height <= 0 ||
clip_x < 0 || clip_y < 0 ||
if (!src_argb || src_width == 0 || src_height == 0 || !dst_argb ||
dst_width <= 0 || dst_height <= 0 || clip_x < 0 || clip_y < 0 ||
clip_width > 32768 || clip_height > 32768 ||
(clip_x + clip_width) > dst_width ||
(clip_y + clip_height) > dst_height) {
return -1;
}
ScaleARGB(src_argb, src_stride_argb, src_width, src_height,
dst_argb, dst_stride_argb, dst_width, dst_height,
clip_x, clip_y, clip_width, clip_height, filtering);
ScaleARGB(src_argb, src_stride_argb, src_width, src_height, dst_argb,
dst_stride_argb, dst_width, dst_height, clip_x, clip_y, clip_width,
clip_height, filtering);
return 0;
}
// Scale an ARGB image.
LIBYUV_API
int ARGBScale(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
int ARGBScale(const uint8_t* src_argb,
int src_stride_argb,
int src_width,
int src_height,
uint8_t* dst_argb,
int dst_stride_argb,
int dst_width,
int dst_height,
enum FilterMode filtering) {
if (!src_argb || src_width == 0 || src_height == 0 ||
src_width > 32768 || src_height > 32768 ||
!dst_argb || dst_width <= 0 || dst_height <= 0) {
if (!src_argb || src_width == 0 || src_height == 0 || src_width > 32768 ||
src_height > 32768 || !dst_argb || dst_width <= 0 || dst_height <= 0) {
return -1;
}
ScaleARGB(src_argb, src_stride_argb, src_width, src_height,
dst_argb, dst_stride_argb, dst_width, dst_height,
0, 0, dst_width, dst_height, filtering);
ScaleARGB(src_argb, src_stride_argb, src_width, src_height, dst_argb,
dst_stride_argb, dst_width, dst_height, 0, 0, dst_width, dst_height,
filtering);
return 0;
}
// Scale with YUV conversion to ARGB and clipping.
LIBYUV_API
int YUVToARGBScaleClip(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint32 src_fourcc,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
uint32 dst_fourcc,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
int YUVToARGBScaleClip(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint32_t src_fourcc,
int src_width,
int src_height,
uint8_t* dst_argb,
int dst_stride_argb,
uint32_t dst_fourcc,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering) {
uint8* argb_buffer = (uint8*)malloc(src_width * src_height * 4);
uint8_t* argb_buffer = (uint8_t*)malloc(src_width * src_height * 4);
int r;
I420ToARGB(src_y, src_stride_y,
src_u, src_stride_u,
src_v, src_stride_v,
argb_buffer, src_width * 4,
src_width, src_height);
(void)src_fourcc; // TODO(fbarchard): implement and/or assert.
(void)dst_fourcc;
I420ToARGB(src_y, src_stride_y, src_u, src_stride_u, src_v, src_stride_v,
argb_buffer, src_width * 4, src_width, src_height);
r = ARGBScaleClip(argb_buffer, src_width * 4,
src_width, src_height,
dst_argb, dst_stride_argb,
dst_width, dst_height,
clip_x, clip_y, clip_width, clip_height,
filtering);
r = ARGBScaleClip(argb_buffer, src_width * 4, src_width, src_height, dst_argb,
dst_stride_argb, dst_width, dst_height, clip_x, clip_y,
clip_width, clip_height, filtering);
free(argb_buffer);
return r;
}
libs/libyuv/source/scale_common.cc
+495 -321
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/scale_gcc.cc
+1166 -1114
View File
File diff suppressed because it is too large Load Diff
libs/libyuv/source/scale_mips.cc
-644
View File
@@ -1,644 +0,0 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for GCC MIPS DSPR2
#if !defined(LIBYUV_DISABLE_MIPS) && \
defined(__mips_dsp) && (__mips_dsp_rev >= 2) && \
(_MIPS_SIM == _MIPS_SIM_ABI32)
void ScaleRowDown2_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"srl $t9, %[dst_width], 4 \n" // iterations -> by 16
"beqz $t9, 2f \n"
" nop \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
"lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
"lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
"lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
// TODO(fbarchard): Use odd pixels instead of even.
"precr.qb.ph $t8, $t1, $t0 \n" // |6|4|2|0|
"precr.qb.ph $t0, $t3, $t2 \n" // |14|12|10|8|
"precr.qb.ph $t1, $t5, $t4 \n" // |22|20|18|16|
"precr.qb.ph $t2, $t7, $t6 \n" // |30|28|26|24|
"addiu %[src_ptr], %[src_ptr], 32 \n"
"addiu $t9, $t9, -1 \n"
"sw $t8, 0(%[dst]) \n"
"sw $t0, 4(%[dst]) \n"
"sw $t1, 8(%[dst]) \n"
"sw $t2, 12(%[dst]) \n"
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 16 \n"
"2: \n"
"andi $t9, %[dst_width], 0xf \n" // residue
"beqz $t9, 3f \n"
" nop \n"
"21: \n"
"lbu $t0, 0(%[src_ptr]) \n"
"addiu %[src_ptr], %[src_ptr], 2 \n"
"addiu $t9, $t9, -1 \n"
"sb $t0, 0(%[dst]) \n"
"bgtz $t9, 21b \n"
" addiu %[dst], %[dst], 1 \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst)
: [dst_width] "r" (dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
}
void ScaleRowDown2Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
const uint8* t = src_ptr + src_stride;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"srl $t9, %[dst_width], 3 \n" // iterations -> step 8
"bltz $t9, 2f \n"
" nop \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t4, 0(%[t]) \n" // |19|18|17|16|
"lw $t5, 4(%[t]) \n" // |23|22|21|20|
"lw $t6, 8(%[t]) \n" // |27|26|25|24|
"lw $t7, 12(%[t]) \n" // |31|30|29|28|
"addiu $t9, $t9, -1 \n"
"srl $t8, $t0, 16 \n" // |X|X|3|2|
"ins $t0, $t4, 16, 16 \n" // |17|16|1|0|
"ins $t4, $t8, 0, 16 \n" // |19|18|3|2|
"raddu.w.qb $t0, $t0 \n" // |17+16+1+0|
"raddu.w.qb $t4, $t4 \n" // |19+18+3+2|
"shra_r.w $t0, $t0, 2 \n" // |t0+2|>>2
"shra_r.w $t4, $t4, 2 \n" // |t4+2|>>2
"srl $t8, $t1, 16 \n" // |X|X|7|6|
"ins $t1, $t5, 16, 16 \n" // |21|20|5|4|
"ins $t5, $t8, 0, 16 \n" // |22|23|7|6|
"raddu.w.qb $t1, $t1 \n" // |21+20+5+4|
"raddu.w.qb $t5, $t5 \n" // |23+22+7+6|
"shra_r.w $t1, $t1, 2 \n" // |t1+2|>>2
"shra_r.w $t5, $t5, 2 \n" // |t5+2|>>2
"srl $t8, $t2, 16 \n" // |X|X|11|10|
"ins $t2, $t6, 16, 16 \n" // |25|24|9|8|
"ins $t6, $t8, 0, 16 \n" // |27|26|11|10|
"raddu.w.qb $t2, $t2 \n" // |25+24+9+8|
"raddu.w.qb $t6, $t6 \n" // |27+26+11+10|
"shra_r.w $t2, $t2, 2 \n" // |t2+2|>>2
"shra_r.w $t6, $t6, 2 \n" // |t5+2|>>2
"srl $t8, $t3, 16 \n" // |X|X|15|14|
"ins $t3, $t7, 16, 16 \n" // |29|28|13|12|
"ins $t7, $t8, 0, 16 \n" // |31|30|15|14|
"raddu.w.qb $t3, $t3 \n" // |29+28+13+12|
"raddu.w.qb $t7, $t7 \n" // |31+30+15+14|
"shra_r.w $t3, $t3, 2 \n" // |t3+2|>>2
"shra_r.w $t7, $t7, 2 \n" // |t7+2|>>2
"addiu %[src_ptr], %[src_ptr], 16 \n"
"addiu %[t], %[t], 16 \n"
"sb $t0, 0(%[dst]) \n"
"sb $t4, 1(%[dst]) \n"
"sb $t1, 2(%[dst]) \n"
"sb $t5, 3(%[dst]) \n"
"sb $t2, 4(%[dst]) \n"
"sb $t6, 5(%[dst]) \n"
"sb $t3, 6(%[dst]) \n"
"sb $t7, 7(%[dst]) \n"
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 8 \n"
"2: \n"
"andi $t9, %[dst_width], 0x7 \n" // x = residue
"beqz $t9, 3f \n"
" nop \n"
"21: \n"
"lwr $t1, 0(%[src_ptr]) \n"
"lwl $t1, 3(%[src_ptr]) \n"
"lwr $t2, 0(%[t]) \n"
"lwl $t2, 3(%[t]) \n"
"srl $t8, $t1, 16 \n"
"ins $t1, $t2, 16, 16 \n"
"ins $t2, $t8, 0, 16 \n"
"raddu.w.qb $t1, $t1 \n"
"raddu.w.qb $t2, $t2 \n"
"shra_r.w $t1, $t1, 2 \n"
"shra_r.w $t2, $t2, 2 \n"
"sb $t1, 0(%[dst]) \n"
"sb $t2, 1(%[dst]) \n"
"addiu %[src_ptr], %[src_ptr], 4 \n"
"addiu $t9, $t9, -2 \n"
"addiu %[t], %[t], 4 \n"
"bgtz $t9, 21b \n"
" addiu %[dst], %[dst], 2 \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst), [t] "+r" (t)
: [dst_width] "r" (dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
}
void ScaleRowDown4_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"srl $t9, %[dst_width], 3 \n"
"beqz $t9, 2f \n"
" nop \n"
"1: \n"
"lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t4, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t5, 16(%[src_ptr]) \n" // |19|18|17|16|
"lw $t6, 20(%[src_ptr]) \n" // |23|22|21|20|
"lw $t7, 24(%[src_ptr]) \n" // |27|26|25|24|
"lw $t8, 28(%[src_ptr]) \n" // |31|30|29|28|
"precr.qb.ph $t1, $t2, $t1 \n" // |6|4|2|0|
"precr.qb.ph $t2, $t4, $t3 \n" // |14|12|10|8|
"precr.qb.ph $t5, $t6, $t5 \n" // |22|20|18|16|
"precr.qb.ph $t6, $t8, $t7 \n" // |30|28|26|24|
"precr.qb.ph $t1, $t2, $t1 \n" // |12|8|4|0|
"precr.qb.ph $t5, $t6, $t5 \n" // |28|24|20|16|
"addiu %[src_ptr], %[src_ptr], 32 \n"
"addiu $t9, $t9, -1 \n"
"sw $t1, 0(%[dst]) \n"
"sw $t5, 4(%[dst]) \n"
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 8 \n"
"2: \n"
"andi $t9, %[dst_width], 7 \n" // residue
"beqz $t9, 3f \n"
" nop \n"
"21: \n"
"lbu $t1, 0(%[src_ptr]) \n"
"addiu %[src_ptr], %[src_ptr], 4 \n"
"addiu $t9, $t9, -1 \n"
"sb $t1, 0(%[dst]) \n"
"bgtz $t9, 21b \n"
" addiu %[dst], %[dst], 1 \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst)
: [dst_width] "r" (dst_width)
: "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
}
void ScaleRowDown4Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
intptr_t stride = src_stride;
const uint8* s1 = src_ptr + stride;
const uint8* s2 = s1 + stride;
const uint8* s3 = s2 + stride;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"srl $t9, %[dst_width], 1 \n"
"andi $t8, %[dst_width], 1 \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 0(%[s1]) \n" // |7|6|5|4|
"lw $t2, 0(%[s2]) \n" // |11|10|9|8|
"lw $t3, 0(%[s3]) \n" // |15|14|13|12|
"lw $t4, 4(%[src_ptr]) \n" // |19|18|17|16|
"lw $t5, 4(%[s1]) \n" // |23|22|21|20|
"lw $t6, 4(%[s2]) \n" // |27|26|25|24|
"lw $t7, 4(%[s3]) \n" // |31|30|29|28|
"raddu.w.qb $t0, $t0 \n" // |3 + 2 + 1 + 0|
"raddu.w.qb $t1, $t1 \n" // |7 + 6 + 5 + 4|
"raddu.w.qb $t2, $t2 \n" // |11 + 10 + 9 + 8|
"raddu.w.qb $t3, $t3 \n" // |15 + 14 + 13 + 12|
"raddu.w.qb $t4, $t4 \n" // |19 + 18 + 17 + 16|
"raddu.w.qb $t5, $t5 \n" // |23 + 22 + 21 + 20|
"raddu.w.qb $t6, $t6 \n" // |27 + 26 + 25 + 24|
"raddu.w.qb $t7, $t7 \n" // |31 + 30 + 29 + 28|
"add $t0, $t0, $t1 \n"
"add $t1, $t2, $t3 \n"
"add $t0, $t0, $t1 \n"
"add $t4, $t4, $t5 \n"
"add $t6, $t6, $t7 \n"
"add $t4, $t4, $t6 \n"
"shra_r.w $t0, $t0, 4 \n"
"shra_r.w $t4, $t4, 4 \n"
"sb $t0, 0(%[dst]) \n"
"sb $t4, 1(%[dst]) \n"
"addiu %[src_ptr], %[src_ptr], 8 \n"
"addiu %[s1], %[s1], 8 \n"
"addiu %[s2], %[s2], 8 \n"
"addiu %[s3], %[s3], 8 \n"
"addiu $t9, $t9, -1 \n"
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 2 \n"
"beqz $t8, 2f \n"
" nop \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 0(%[s1]) \n" // |7|6|5|4|
"lw $t2, 0(%[s2]) \n" // |11|10|9|8|
"lw $t3, 0(%[s3]) \n" // |15|14|13|12|
"raddu.w.qb $t0, $t0 \n" // |3 + 2 + 1 + 0|
"raddu.w.qb $t1, $t1 \n" // |7 + 6 + 5 + 4|
"raddu.w.qb $t2, $t2 \n" // |11 + 10 + 9 + 8|
"raddu.w.qb $t3, $t3 \n" // |15 + 14 + 13 + 12|
"add $t0, $t0, $t1 \n"
"add $t1, $t2, $t3 \n"
"add $t0, $t0, $t1 \n"
"shra_r.w $t0, $t0, 4 \n"
"sb $t0, 0(%[dst]) \n"
"2: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst),
[s1] "+r" (s1),
[s2] "+r" (s2),
[s3] "+r" (s3)
: [dst_width] "r" (dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6","t7", "t8", "t9"
);
}
void ScaleRowDown34_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"1: \n"
"lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t4, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t5, 16(%[src_ptr]) \n" // |19|18|17|16|
"lw $t6, 20(%[src_ptr]) \n" // |23|22|21|20|
"lw $t7, 24(%[src_ptr]) \n" // |27|26|25|24|
"lw $t8, 28(%[src_ptr]) \n" // |31|30|29|28|
"precrq.qb.ph $t0, $t2, $t4 \n" // |7|5|15|13|
"precrq.qb.ph $t9, $t6, $t8 \n" // |23|21|31|30|
"addiu %[dst_width], %[dst_width], -24 \n"
"ins $t1, $t1, 8, 16 \n" // |3|1|0|X|
"ins $t4, $t0, 8, 16 \n" // |X|15|13|12|
"ins $t5, $t5, 8, 16 \n" // |19|17|16|X|
"ins $t8, $t9, 8, 16 \n" // |X|31|29|28|
"addiu %[src_ptr], %[src_ptr], 32 \n"
"packrl.ph $t0, $t3, $t0 \n" // |9|8|7|5|
"packrl.ph $t9, $t7, $t9 \n" // |25|24|23|21|
"prepend $t1, $t2, 8 \n" // |4|3|1|0|
"prepend $t3, $t4, 24 \n" // |15|13|12|11|
"prepend $t5, $t6, 8 \n" // |20|19|17|16|
"prepend $t7, $t8, 24 \n" // |31|29|28|27|
"sw $t1, 0(%[dst]) \n"
"sw $t0, 4(%[dst]) \n"
"sw $t3, 8(%[dst]) \n"
"sw $t5, 12(%[dst]) \n"
"sw $t9, 16(%[dst]) \n"
"sw $t7, 20(%[dst]) \n"
"bnez %[dst_width], 1b \n"
" addiu %[dst], %[dst], 24 \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst),
[dst_width] "+r" (dst_width)
:
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6","t7", "t8", "t9"
);
}
void ScaleRowDown34_0_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* d, int dst_width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"repl.ph $t3, 3 \n" // 0x00030003
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
"lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
"rotr $t2, $t0, 8 \n" // |S0|S3|S2|S1|
"rotr $t6, $t1, 8 \n" // |T0|T3|T2|T1|
"muleu_s.ph.qbl $t4, $t2, $t3 \n" // |S0*3|S3*3|
"muleu_s.ph.qbl $t5, $t6, $t3 \n" // |T0*3|T3*3|
"andi $t0, $t2, 0xFFFF \n" // |0|0|S2|S1|
"andi $t1, $t6, 0xFFFF \n" // |0|0|T2|T1|
"raddu.w.qb $t0, $t0 \n"
"raddu.w.qb $t1, $t1 \n"
"shra_r.w $t0, $t0, 1 \n"
"shra_r.w $t1, $t1, 1 \n"
"preceu.ph.qbr $t2, $t2 \n" // |0|S2|0|S1|
"preceu.ph.qbr $t6, $t6 \n" // |0|T2|0|T1|
"rotr $t2, $t2, 16 \n" // |0|S1|0|S2|
"rotr $t6, $t6, 16 \n" // |0|T1|0|T2|
"addu.ph $t2, $t2, $t4 \n"
"addu.ph $t6, $t6, $t5 \n"
"sll $t5, $t0, 1 \n"
"add $t0, $t5, $t0 \n"
"shra_r.ph $t2, $t2, 2 \n"
"shra_r.ph $t6, $t6, 2 \n"
"shll.ph $t4, $t2, 1 \n"
"addq.ph $t4, $t4, $t2 \n"
"addu $t0, $t0, $t1 \n"
"addiu %[src_ptr], %[src_ptr], 4 \n"
"shra_r.w $t0, $t0, 2 \n"
"addu.ph $t6, $t6, $t4 \n"
"shra_r.ph $t6, $t6, 2 \n"
"srl $t1, $t6, 16 \n"
"addiu %[dst_width], %[dst_width], -3 \n"
"sb $t1, 0(%[d]) \n"
"sb $t0, 1(%[d]) \n"
"sb $t6, 2(%[d]) \n"
"bgtz %[dst_width], 1b \n"
" addiu %[d], %[d], 3 \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[src_stride] "+r" (src_stride),
[d] "+r" (d),
[dst_width] "+r" (dst_width)
:
: "t0", "t1", "t2", "t3",
"t4", "t5", "t6"
);
}
void ScaleRowDown34_1_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* d, int dst_width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"repl.ph $t2, 3 \n" // 0x00030003
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
"lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
"rotr $t4, $t0, 8 \n" // |S0|S3|S2|S1|
"rotr $t6, $t1, 8 \n" // |T0|T3|T2|T1|
"muleu_s.ph.qbl $t3, $t4, $t2 \n" // |S0*3|S3*3|
"muleu_s.ph.qbl $t5, $t6, $t2 \n" // |T0*3|T3*3|
"andi $t0, $t4, 0xFFFF \n" // |0|0|S2|S1|
"andi $t1, $t6, 0xFFFF \n" // |0|0|T2|T1|
"raddu.w.qb $t0, $t0 \n"
"raddu.w.qb $t1, $t1 \n"
"shra_r.w $t0, $t0, 1 \n"
"shra_r.w $t1, $t1, 1 \n"
"preceu.ph.qbr $t4, $t4 \n" // |0|S2|0|S1|
"preceu.ph.qbr $t6, $t6 \n" // |0|T2|0|T1|
"rotr $t4, $t4, 16 \n" // |0|S1|0|S2|
"rotr $t6, $t6, 16 \n" // |0|T1|0|T2|
"addu.ph $t4, $t4, $t3 \n"
"addu.ph $t6, $t6, $t5 \n"
"shra_r.ph $t6, $t6, 2 \n"
"shra_r.ph $t4, $t4, 2 \n"
"addu.ph $t6, $t6, $t4 \n"
"addiu %[src_ptr], %[src_ptr], 4 \n"
"shra_r.ph $t6, $t6, 1 \n"
"addu $t0, $t0, $t1 \n"
"addiu %[dst_width], %[dst_width], -3 \n"
"shra_r.w $t0, $t0, 1 \n"
"srl $t1, $t6, 16 \n"
"sb $t1, 0(%[d]) \n"
"sb $t0, 1(%[d]) \n"
"sb $t6, 2(%[d]) \n"
"bgtz %[dst_width], 1b \n"
" addiu %[d], %[d], 3 \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[src_stride] "+r" (src_stride),
[d] "+r" (d),
[dst_width] "+r" (dst_width)
:
: "t0", "t1", "t2", "t3",
"t4", "t5", "t6"
);
}
void ScaleRowDown38_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
"lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
"lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
"lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
"wsbh $t0, $t0 \n" // |2|3|0|1|
"wsbh $t6, $t6 \n" // |26|27|24|25|
"srl $t0, $t0, 8 \n" // |X|2|3|0|
"srl $t3, $t3, 16 \n" // |X|X|15|14|
"srl $t5, $t5, 16 \n" // |X|X|23|22|
"srl $t7, $t7, 16 \n" // |X|X|31|30|
"ins $t1, $t2, 24, 8 \n" // |8|6|5|4|
"ins $t6, $t5, 0, 8 \n" // |26|27|24|22|
"ins $t1, $t0, 0, 16 \n" // |8|6|3|0|
"ins $t6, $t7, 24, 8 \n" // |30|27|24|22|
"prepend $t2, $t3, 24 \n" // |X|15|14|11|
"ins $t4, $t4, 16, 8 \n" // |19|16|17|X|
"ins $t4, $t2, 0, 16 \n" // |19|16|14|11|
"addiu %[src_ptr], %[src_ptr], 32 \n"
"addiu %[dst_width], %[dst_width], -12 \n"
"addiu $t8,%[dst_width], -12 \n"
"sw $t1, 0(%[dst]) \n"
"sw $t4, 4(%[dst]) \n"
"sw $t6, 8(%[dst]) \n"
"bgez $t8, 1b \n"
" addiu %[dst], %[dst], 12 \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst),
[dst_width] "+r" (dst_width)
:
: "t0", "t1", "t2", "t3", "t4",
"t5", "t6", "t7", "t8"
);
}
void ScaleRowDown38_2_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
intptr_t stride = src_stride;
const uint8* t = src_ptr + stride;
const int c = 0x2AAA;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
"lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
"lw $t2, 0(%[t]) \n" // |T3|T2|T1|T0|
"lw $t3, 4(%[t]) \n" // |T7|T6|T5|T4|
"rotr $t1, $t1, 16 \n" // |S5|S4|S7|S6|
"packrl.ph $t4, $t1, $t3 \n" // |S7|S6|T7|T6|
"packrl.ph $t5, $t3, $t1 \n" // |T5|T4|S5|S4|
"raddu.w.qb $t4, $t4 \n" // S7+S6+T7+T6
"raddu.w.qb $t5, $t5 \n" // T5+T4+S5+S4
"precrq.qb.ph $t6, $t0, $t2 \n" // |S3|S1|T3|T1|
"precrq.qb.ph $t6, $t6, $t6 \n" // |S3|T3|S3|T3|
"srl $t4, $t4, 2 \n" // t4 / 4
"srl $t6, $t6, 16 \n" // |0|0|S3|T3|
"raddu.w.qb $t6, $t6 \n" // 0+0+S3+T3
"addu $t6, $t5, $t6 \n"
"mul $t6, $t6, %[c] \n" // t6 * 0x2AAA
"sll $t0, $t0, 8 \n" // |S2|S1|S0|0|
"sll $t2, $t2, 8 \n" // |T2|T1|T0|0|
"raddu.w.qb $t0, $t0 \n" // S2+S1+S0+0
"raddu.w.qb $t2, $t2 \n" // T2+T1+T0+0
"addu $t0, $t0, $t2 \n"
"mul $t0, $t0, %[c] \n" // t0 * 0x2AAA
"addiu %[src_ptr], %[src_ptr], 8 \n"
"addiu %[t], %[t], 8 \n"
"addiu %[dst_width], %[dst_width], -3 \n"
"addiu %[dst_ptr], %[dst_ptr], 3 \n"
"srl $t6, $t6, 16 \n"
"srl $t0, $t0, 16 \n"
"sb $t4, -1(%[dst_ptr]) \n"
"sb $t6, -2(%[dst_ptr]) \n"
"bgtz %[dst_width], 1b \n"
" sb $t0, -3(%[dst_ptr]) \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst_ptr] "+r" (dst_ptr),
[t] "+r" (t),
[dst_width] "+r" (dst_width)
: [c] "r" (c)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6"
);
}
void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
intptr_t stride = src_stride;
const uint8* s1 = src_ptr + stride;
stride += stride;
const uint8* s2 = src_ptr + stride;
const int c1 = 0x1C71;
const int c2 = 0x2AAA;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
"lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
"lw $t2, 0(%[s1]) \n" // |T3|T2|T1|T0|
"lw $t3, 4(%[s1]) \n" // |T7|T6|T5|T4|
"lw $t4, 0(%[s2]) \n" // |R3|R2|R1|R0|
"lw $t5, 4(%[s2]) \n" // |R7|R6|R5|R4|
"rotr $t1, $t1, 16 \n" // |S5|S4|S7|S6|
"packrl.ph $t6, $t1, $t3 \n" // |S7|S6|T7|T6|
"raddu.w.qb $t6, $t6 \n" // S7+S6+T7+T6
"packrl.ph $t7, $t3, $t1 \n" // |T5|T4|S5|S4|
"raddu.w.qb $t7, $t7 \n" // T5+T4+S5+S4
"sll $t8, $t5, 16 \n" // |R5|R4|0|0|
"raddu.w.qb $t8, $t8 \n" // R5+R4
"addu $t7, $t7, $t8 \n"
"srl $t8, $t5, 16 \n" // |0|0|R7|R6|
"raddu.w.qb $t8, $t8 \n" // R7 + R6
"addu $t6, $t6, $t8 \n"
"mul $t6, $t6, %[c2] \n" // t6 * 0x2AAA
"precrq.qb.ph $t8, $t0, $t2 \n" // |S3|S1|T3|T1|
"precrq.qb.ph $t8, $t8, $t4 \n" // |S3|T3|R3|R1|
"srl $t8, $t8, 8 \n" // |0|S3|T3|R3|
"raddu.w.qb $t8, $t8 \n" // S3 + T3 + R3
"addu $t7, $t7, $t8 \n"
"mul $t7, $t7, %[c1] \n" // t7 * 0x1C71
"sll $t0, $t0, 8 \n" // |S2|S1|S0|0|
"sll $t2, $t2, 8 \n" // |T2|T1|T0|0|
"sll $t4, $t4, 8 \n" // |R2|R1|R0|0|
"raddu.w.qb $t0, $t0 \n"
"raddu.w.qb $t2, $t2 \n"
"raddu.w.qb $t4, $t4 \n"
"addu $t0, $t0, $t2 \n"
"addu $t0, $t0, $t4 \n"
"mul $t0, $t0, %[c1] \n" // t0 * 0x1C71
"addiu %[src_ptr], %[src_ptr], 8 \n"
"addiu %[s1], %[s1], 8 \n"
"addiu %[s2], %[s2], 8 \n"
"addiu %[dst_width], %[dst_width], -3 \n"
"addiu %[dst_ptr], %[dst_ptr], 3 \n"
"srl $t6, $t6, 16 \n"
"srl $t7, $t7, 16 \n"
"srl $t0, $t0, 16 \n"
"sb $t6, -1(%[dst_ptr]) \n"
"sb $t7, -2(%[dst_ptr]) \n"
"bgtz %[dst_width], 1b \n"
" sb $t0, -3(%[dst_ptr]) \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst_ptr] "+r" (dst_ptr),
[s1] "+r" (s1),
[s2] "+r" (s2),
[dst_width] "+r" (dst_width)
: [c1] "r" (c1), [c2] "r" (c2)
: "t0", "t1", "t2", "t3", "t4",
"t5", "t6", "t7", "t8"
);
}
#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

Some files were not shown because too many files have changed in this diff Show More