63 lines
2.1 KiB
CMake
63 lines
2.1 KiB
CMake
|
|
# SPDX-License-Identifier: Apache-2.0
|
|||
|
|
# ARM架构优化配置文件
|
|||
|
|
|
|||
|
|
# 检查是否为ARM架构
|
|||
|
|
if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm64" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm")
|
|||
|
|
|
|||
|
|
# 启用NEON指令集优化
|
|||
|
|
if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "aarch64" OR ${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm64")
|
|||
|
|
# ARM64架构
|
|||
|
|
add_compile_options(-march=armv8-a+simd)
|
|||
|
|
|
|||
|
|
# 如果支持 crypto 扩展,则启用
|
|||
|
|
include(CheckCSourceCompiles)
|
|||
|
|
check_c_source_compiles("
|
|||
|
|
#include <arm_neon.h>
|
|||
|
|
int main() {
|
|||
|
|
uint8x16_t a = vdupq_n_u8(0);
|
|||
|
|
uint8x16_t b = vaeseq_u8(a, vdupq_n_u8(0));
|
|||
|
|
return 0;
|
|||
|
|
}" HAVE_ARM64_CRYPTO)
|
|||
|
|
|
|||
|
|
if (HAVE_ARM64_CRYPTO)
|
|||
|
|
add_compile_options(-march=armv8-a+crypto)
|
|||
|
|
add_compile_definitions(HAVE_ARM64_CRYPTO)
|
|||
|
|
endif()
|
|||
|
|
else()
|
|||
|
|
# ARM32架构
|
|||
|
|
add_compile_options(-march=armv7-a -mfpu=neon)
|
|||
|
|
endif()
|
|||
|
|
|
|||
|
|
# 通用ARM优化选项
|
|||
|
|
# 启用循环展开和其他优化
|
|||
|
|
add_compile_options(-O3 -funroll-loops)
|
|||
|
|
|
|||
|
|
# 启用链接时优化(LTO)
|
|||
|
|
include(CheckIPOSupported)
|
|||
|
|
check_ipo_supported(RESULT result)
|
|||
|
|
if(result)
|
|||
|
|
set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
|
|||
|
|
# 检查编译器是否支持 thin LTO
|
|||
|
|
include(CheckCCompilerFlag)
|
|||
|
|
# check_c_compiler_flag("-flto=thin" HAS_THIN_LTO)
|
|||
|
|
# if(HAS_THIN_LTO)
|
|||
|
|
# add_compile_options(-flto=thin)
|
|||
|
|
# else()
|
|||
|
|
# # 回退到普通 LTO
|
|||
|
|
# add_compile_options(-flto)
|
|||
|
|
# endif()
|
|||
|
|
add_compile_options(-flto=auto)
|
|||
|
|
endif()
|
|||
|
|
|
|||
|
|
# 启用快速数学运算(可能影响精度)
|
|||
|
|
# add_compile_options(-ffast-math)
|
|||
|
|
|
|||
|
|
# 针对特定CPU的优化
|
|||
|
|
# 可以根据目标设备替换为具体的CPU型号,如"cortex-a72"等
|
|||
|
|
add_compile_options(-mtune=cortex-a76)
|
|||
|
|
|
|||
|
|
message(STATUS "ARM optimizations enabled for ${CMAKE_SYSTEM_PROCESSOR}")
|
|||
|
|
|
|||
|
|
# 添加NEON支持的定义
|
|||
|
|
add_compile_definitions(HAVE_NEON)
|
|||
|
|
endif()
|