libyuv/CMakeLists.txt
Frank Barchard fab11704cd [libyuv] Use Armv8.5-A architecture flags for SME and SVE2 build targets
Update BUILD.gn, CMakeLists.txt, and BUILD.bazel to specify Armv8.5-A
baseline architecture flags rather than Armv9:
- BUILD.gn: update :libyuv_sve to -march=armv8.5-a+i8mm+sve2 and
  :libyuv_sme to -march=armv8.5-a+i8mm+sme.
- CMakeLists.txt: update SME check and compile options to
  -march=armv8.5-a+i8mm+sme (SVE target already uses armv8.5-a).
- BUILD.bazel: update ENABLE_SVE_COPTS to -march=armv8.5-a+i8mm+sve2.

Using -march=armv9-a enabled baseline SVE in the compiler, causing
Clang to emit non-streaming SVE instructions (such as cntd in function
prologues prior to smstart) in SME translation units. On processors
such as Apple M4/M5 that support SME but not SVE, executing these
prologue instructions resulted in SIGILL crashes.

Test: ninja -C out/arm64.release libyuv libyuv_unittest
Bug: 558654023
Change-Id: Ic7f62842bc9bd71608cc78ad11fd5e2214ef5e8e
TAG=agy
CONV=a70f0e53-caf2-4fbe-a024-1616ed25ba9a
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/8376021
Commit-Queue: Wan-Teh Chang <wtc@google.com>
Reviewed-by: Wan-Teh Chang <wtc@google.com>
2026-09-09 09:17:24 -07:00

298 lines
11 KiB
CMake

# CMakeLists for libyuv
# Originally created for "roxlu build system" to compile libyuv on windows
# Run with -DUNIT_TEST=ON to build unit tests
cmake_minimum_required( VERSION 3.16 )
project ( YUV C CXX ) # "C" is required even for C++ projects
option( UNIT_TEST "Built unit tests" OFF )
option( LIBYUV_DISABLE_JPEG "Disable JPEG support" OFF )
include(CheckCSourceCompiles)
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 )
# We cannot use GLOB here since we want to be able to separate out files that
# need particular flags to enable architecture extensions like AArch64's SVE.
# TODO: More of these files could be separated out for different architectures.
set ( ly_common_source_files
${ly_src_dir}/compare.cc
${ly_src_dir}/compare_common.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_gcc.cc
${ly_src_dir}/rotate_win.cc
${ly_src_dir}/row_any.cc
${ly_src_dir}/row_common.cc
${ly_src_dir}/row_gcc.cc
${ly_src_dir}/row_rvv.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_gcc.cc
${ly_src_dir}/scale_rgb.cc
${ly_src_dir}/scale_rvv.cc
${ly_src_dir}/scale_uv.cc
${ly_src_dir}/scale_win.cc
${ly_src_dir}/video_common.cc)
file ( GLOB_RECURSE ly_unittest_sources ${ly_tst_dir}/*.cc )
list ( SORT ly_unittest_sources )
include_directories( BEFORE ${ly_inc_dir} )
if(MSVC)
add_definitions ( -D_CRT_SECURE_NO_WARNINGS )
endif()
# Need to set PIC to allow creating shared libraries from object file libraries.
set(CMAKE_POSITION_INDEPENDENT_CODE ON)
# Build the set of objects that do not need to be compiled with flags to enable
# particular architecture features.
add_library( ${ly_lib_name}_common_objects OBJECT ${ly_common_source_files} )
set(ly_lib_parts $<TARGET_OBJECTS:${ly_lib_name}_common_objects>)
string(TOLOWER "${CMAKE_SYSTEM_PROCESSOR}" SYSPROC)
set(LOONGARCH64_ALIASES loongarch64)
list(FIND LOONGARCH64_ALIASES "${SYSPROC}" LOONGARCH64MATCH)
if(LOONGARCH64MATCH GREATER "-1")
set(LOONGARCH64 1)
endif()
if(NOT MSVC)
string(TOLOWER "${CMAKE_SYSTEM_PROCESSOR}" arch_lowercase)
if(arch_lowercase MATCHES "^arm" AND NOT arch_lowercase STREQUAL "arm64")
# Enable Arm Neon kernels.
add_definitions(-DLIBYUV_NEON=1)
add_library(${ly_lib_name}_neon OBJECT
${ly_src_dir}/compare_neon.cc
${ly_src_dir}/rotate_neon.cc
${ly_src_dir}/row_neon.cc
${ly_src_dir}/scale_neon.cc)
target_compile_options(${ly_lib_name}_neon PRIVATE -mfpu=neon)
list(APPEND ly_lib_parts $<TARGET_OBJECTS:${ly_lib_name}_neon>)
endif()
if(arch_lowercase STREQUAL "aarch64" OR arch_lowercase STREQUAL "arm64")
# Enable AArch64 Neon dot-product and i8mm kernels.
add_library(${ly_lib_name}_neon64 OBJECT
${ly_src_dir}/compare_neon64.cc
${ly_src_dir}/rotate_neon64.cc
${ly_src_dir}/row_neon64.cc
${ly_src_dir}/scale_neon64.cc)
target_compile_options(${ly_lib_name}_neon64 PRIVATE -march=armv8.2-a+dotprod+i8mm)
list(APPEND ly_lib_parts $<TARGET_OBJECTS:${ly_lib_name}_neon64>)
# Enable AArch64 SVE kernels.
add_library(${ly_lib_name}_sve OBJECT
${ly_src_dir}/row_sve.cc)
target_compile_options(${ly_lib_name}_sve PRIVATE -march=armv8.5-a+i8mm+sve2)
list(APPEND ly_lib_parts $<TARGET_OBJECTS:${ly_lib_name}_sve>)
# Check whether the compiler can compile SME functions; this fails
# with Clang for Windows.
set(OLD_CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS})
set(OLD_CMAKE_TRY_COMPILE_TARGET_TYPE ${CMAKE_TRY_COMPILE_TARGET_TYPE})
set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} -march=armv8.5-a+i8mm+sme")
set(CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY)
check_c_source_compiles("
__arm_locally_streaming void func(void) { }
int main(void) { return 0; }
" CAN_COMPILE_SME)
set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
set(CMAKE_TRY_COMPILE_TARGET_TYPE ${OLD_CMAKE_TRY_COMPILE_TARGET_TYPE})
if (CAN_COMPILE_SME)
# Enable AArch64 SME kernels.
add_library(${ly_lib_name}_sme OBJECT
${ly_src_dir}/rotate_sme.cc
${ly_src_dir}/row_sme.cc
${ly_src_dir}/scale_sme.cc)
target_compile_options(${ly_lib_name}_sme PRIVATE -march=armv8.5-a+i8mm+sme)
list(APPEND ly_lib_parts $<TARGET_OBJECTS:${ly_lib_name}_sme>)
else()
add_definitions(-DLIBYUV_DISABLE_SME)
endif()
endif()
if(arch_lowercase MATCHES "^riscv")
option(USE_RVV "Enable RISC-V Vector" ON)
if(USE_RVV)
if(NOT RISCV_COMPILER_FLAGS)
if(CMAKE_SIZEOF_VOID_P EQUAL 8 OR arch_lowercase STREQUAL "riscv64")
set(RISCV_COMPILER_FLAGS "-march=rv64gcv")
else()
set(RISCV_COMPILER_FLAGS "-march=rv32gcv")
endif()
endif()
include(CheckCSourceCompiles)
set(OLD_CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS})
set(OLD_CMAKE_TRY_COMPILE_TARGET_TYPE ${CMAKE_TRY_COMPILE_TARGET_TYPE})
set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} ${RISCV_COMPILER_FLAGS}")
set(CMAKE_TRY_COMPILE_TARGET_TYPE STATIC_LIBRARY)
check_c_source_compiles("
int main(void) { return 0; }
" CAN_COMPILE_RVV)
set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
set(CMAKE_TRY_COMPILE_TARGET_TYPE ${OLD_CMAKE_TRY_COMPILE_TARGET_TYPE})
if(CAN_COMPILE_RVV)
if(NOT CMAKE_C_FLAGS MATCHES "-march=" AND NOT CMAKE_C_FLAGS MATCHES "-mcpu=")
set(CMAKE_C_FLAGS "${RISCV_COMPILER_FLAGS} -fno-vectorize -fno-slp-vectorize ${CMAKE_C_FLAGS}")
set(CMAKE_CXX_FLAGS "${RISCV_COMPILER_FLAGS} -fno-vectorize -fno-slp-vectorize ${CMAKE_CXX_FLAGS}")
endif()
else()
add_definitions(-DLIBYUV_DISABLE_RVV)
endif()
else()
add_definitions(-DLIBYUV_DISABLE_RVV)
endif()
endif()
endif()
if(LOONGARCH64)
include(CheckCXXSourceCompiles)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wno-narrowing")
check_cxx_source_compiles("
int main(int argc, char **argv) {
__asm__ volatile (
\"vadd.w $vr0, $vr1, $vr1\"
);
return 0; }" SUPPORTS_LSX)
check_cxx_source_compiles("
int main(int argc, char **argv) {
__asm__ volatile (
\"xvadd.w $xr0, $xr1, $xr1\"
);
return 0; }" SUPPORTS_LASX)
if(SUPPORTS_LSX)
# These kernels use only LSX (128-bit) instructions and must not be
# compiled with -mlasx, which would let the compiler emit 256-bit LASX
# code (e.g. for stack-array copies) and crash on LSX-only CPUs.
add_library(${ly_lib_name}_lsx OBJECT
${ly_src_dir}/rotate_lsx.cc
${ly_src_dir}/row_lsx.cc
${ly_src_dir}/scale_lsx.cc)
target_compile_options(${ly_lib_name}_lsx PRIVATE -mlsx)
list(APPEND ly_lib_parts $<TARGET_OBJECTS:${ly_lib_name}_lsx>)
endif()
if(SUPPORTS_LASX)
add_library(${ly_lib_name}_lasx OBJECT
${ly_src_dir}/row_lasx.cc)
target_compile_options(${ly_lib_name}_lasx PRIVATE -mlasx)
list(APPEND ly_lib_parts $<TARGET_OBJECTS:${ly_lib_name}_lasx>)
endif()
endif()
# this creates the static library (.a)
add_library( ${ly_lib_static} STATIC ${ly_lib_parts})
# this creates the shared library (.so)
add_library( ${ly_lib_shared} SHARED ${ly_lib_parts})
set_target_properties( ${ly_lib_shared} PROPERTIES OUTPUT_NAME "${ly_lib_name}" )
set_target_properties( ${ly_lib_shared} PROPERTIES PREFIX "lib" )
if(WIN32)
set_target_properties( ${ly_lib_shared} PROPERTIES IMPORT_PREFIX "lib" )
endif()
# this creates the cpuid tool
add_executable ( cpuid ${ly_base_dir}/util/cpuid.c )
target_link_libraries ( cpuid ${ly_lib_static} )
# this creates the conversion tool
add_executable ( yuvconvert ${ly_base_dir}/util/yuvconvert.cc )
target_link_libraries ( yuvconvert ${ly_lib_static} )
# this creates the yuvconstants tool
add_executable ( yuvconstants ${ly_base_dir}/util/yuvconstants.c )
target_link_libraries ( yuvconstants ${ly_lib_static} )
if (NOT LIBYUV_DISABLE_JPEG)
find_package ( JPEG )
if (JPEG_FOUND)
include_directories( ${JPEG_INCLUDE_DIR} )
target_link_libraries( ${ly_lib_shared} ${JPEG_LIBRARY} )
add_definitions( -DHAVE_JPEG )
endif()
endif()
if(UNIT_TEST)
find_library(GTEST_LIBRARY gtest)
if(GTEST_LIBRARY STREQUAL "GTEST_LIBRARY-NOTFOUND")
set(GTEST_SRC_DIR_DEFAULT /usr/src/gtest)
if (CMAKE_CROSSCOMPILING)
set(GTEST_SRC_DIR_DEFAULT ${CMAKE_SOURCE_DIR}/third_party/googletest/src/googletest)
endif()
set(GTEST_SRC_DIR ${GTEST_SRC_DIR_DEFAULT} CACHE STRING "Location of gtest sources")
if(EXISTS ${GTEST_SRC_DIR}/src/gtest-all.cc)
message(STATUS "building gtest from sources in ${GTEST_SRC_DIR}")
set(gtest_sources ${GTEST_SRC_DIR}/src/gtest-all.cc)
add_library(gtest STATIC ${gtest_sources})
include_directories(${GTEST_SRC_DIR})
include_directories(${GTEST_SRC_DIR}/include)
set(GTEST_LIBRARY gtest)
else()
message(FATAL_ERROR "UNIT_TEST is set but unable to find gtest library")
endif()
endif()
add_executable(libyuv_unittest ${ly_unittest_sources})
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()
find_library(GFLAGS_LIBRARY gflags)
if(NOT GFLAGS_LIBRARY STREQUAL "GFLAGS_LIBRARY-NOTFOUND")
target_link_libraries(libyuv_unittest gflags)
add_definitions(-DLIBYUV_USE_GFLAGS)
endif()
endif()
# install the conversion tool, .so, .a, and all the header files
install ( TARGETS yuvconvert DESTINATION bin )
install ( TARGETS ${ly_lib_static} DESTINATION lib )
install ( TARGETS ${ly_lib_shared} LIBRARY DESTINATION lib RUNTIME DESTINATION bin ARCHIVE DESTINATION lib )
install ( DIRECTORY ${PROJECT_SOURCE_DIR}/include/ DESTINATION include )
# create the .deb and .rpm packages using cpack
include ( CM_linux_packages.cmake )