【海思】——HISI3559A交叉编译openblas与测试

摘要： openblas主要用于矩阵运算的加速

1. openblas交叉编译

1. 下载openblas，这里用0.3.7版本
   https://github.com/xianyi/OpenBLAS

2. 编译
   前提

• 需要安装好交叉编译工具链aarch64-linux-gnu

下载地址如下：
https://releases.linaro.org/components/toolchain/binaries/6.3-2017.05/aarch64-linux-gnu/
拷贝到编译服务器，海思交叉切换到该开源工具链，执行如下命令：

tar xf gcc-linaro-6.3.1-2017.05-x86_64_aarch64-linux-gnu.tar.xz
vi .profile
PATH="$HOME/bin:$PATH:$HOME/bin/gcc-linaro-6.3.1-2017.05-x86_64_aarch64-linux-gnu/bin"
source .profile

• 交叉编译器验证
  aarch64-linux-gnu-gcc --version

---

• 交叉编译openblas

tar -zxvf OpenBLAS-0.3.7.tar.gz
cd OpenBLAS-0.3.7
make BINARY=64 CC=aarch64-linux-gnu-gcc NOFORTRAN=1 HOSTCC=gcc TARGET=ARMV8
make PREFIX=../openblas_install install

• 结果
  到安装目录openblas_install，有lib, include, bin文件夹

.
├── bin
├── include
│   ├── cblas.h
│   ├── f77blas.h
│   ├── lapacke_config.h
│   ├── lapacke.h
│   ├── lapacke_mangling.h
│   ├── lapacke_utils.h
│   └── openblas_config.h
└── lib
    ├── cmake
    │   └── openblas
    │       ├── OpenBLASConfig.cmake
    │       └── OpenBLASConfigVersion.cmake
    ├── libopenblas.a -> libopenblas_armv8p-r0.3.7.a
    ├── libopenblas_armv8p-r0.3.7.a
    ├── libopenblas_armv8p-r0.3.7.so
    ├── libopenblas.so -> libopenblas_armv8p-r0.3.7.so
    ├── libopenblas.so.0 -> libopenblas_armv8p-r0.3.7.so
    └── pkgconfig
        └── openblas.pc

2. 使用测试

code

 
#include <cblas.h>
#include <stdio.h>
 
int main() {
 
    int i = 0;
    double A[6] = {1.0,3.0,1.0,-3.0,4.0,-1.0};         
    double B[6] = {1.0,4.0,1.0,-3.0,4.0,-1.0};  
    double C[9] = {.5,.5,.5,1.5,.5,2.5,.5,.5,.5}; 
 
    int M = 3; // row of A and C
    int N = 3; // col of B and C
    int K = 2; // col of A and row of B
 
    double alpha = 1.0;
    double beta = 0.0;
 
    cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, M, N, K, alpha, A, K, B, N, beta, C, N);
    //CblasRowMajor表示行优先
    //CblasNoTrans表示不转置
 
    for (i = 0; i < 9; i++) {
        printf("%lf ", C[i]);
    }
    printf("\n");

    return 1;
}

编译cmakelists.txt

cmake_minimum_required (VERSION 2.6)
project (TEST)
 
set (TEST_VERSION 0.1)
 
set(CMAKE_BUILD_TYPE "Debug")
set(CMAKE_CXX_FLAGS_DEBUG "$ENV{CXXFLAGS} -O0 -Wall -g -ggdb -DDEBUG")
set(CMAKE_CXX_FLAGS_RELEASE "$ENV{CXXFLAGS} -O3 -Wall")
 
aux_source_directory(${PROJECT_SOURCE_DIR} DIR_SRC)
 
include_directories(/home/yangna/chenjun/HISI3559a/openblas_install/include)        # 头文件
 
# link_directories(/home/yangna/chenjun/HISI3559a/openblas_install/lib)
find_library(Openblas_LIBS openblas /home/yangna/chenjun/HISI3559a/openblas_install/lib)    # 库文件
add_executable(main ${DIR_SRC})
 
target_link_libraries(main ${Openblas_LIBS})
target_link_libraries(main -lm)                 # 这两个选项是必须要的
target_link_libraries(main -lpthread)

• cmake的写法和其他动态库的使用是一样的。添加头文件，和库文件
• 链接的时候加上-lm -lpthread两项

2.1 cmake+cxx_compiler指定

• 我们平时编译c++的项目就是：

mkdir build && cd build
cmake ..
make

• 上面是采用系统默认的g++进行编译的。我们编译hisi3559a的项目时，应该要用aarch64-linux-gnu-g++交叉编译器。可以通过cmake的CMAKE_CXX_COMPILER进行指定。于是交叉编译的命令为:

mkdir build && cd build
cmake -DCMAKE_CXX_COMPILER=/home/yangna/Atlas500_DDK/toolchains/Euler_compile_env_cross/arm/cross_compile/install/bin/aarch64-linux-gnu-g++ ..
make

输出

Scanning dependencies of target main
[ 50%] Building CXX object CMakeFiles/main.dir/main.cpp.o
[100%] Linking CXX executable main
[100%] Built target main

2.2 cmake+CMAKE_TOOLCHAIN_FILE指定

• 上面的方法在命令行中，要输入很长的路径，还不能复用。可以用cmake的cmake_toolchain_file进行指定

• 新建一个hisi3559.cmake文件（一般应用开发文档都会给出的），写入：

set(EULER_CROSS_PATH /home/yangna/Atlas500_DDK/toolchains/Euler_compile_env_cross)

set(CMAKE_SYSTEM_NAME Linux)
set(CMAKE_SYSTEM_PROCESSOR arm)

set(tools ${EULER_CROSS_PATH}/arm/cross_compile/install/)

#set(CMAKE_SYSROOT ${tools}/sysroot)
set(CMAKE_C_COMPILER ${tools}/bin/aarch64-linux-gnu-gcc)
set(CMAKE_CXX_COMPILER ${tools}/bin/aarch64-linux-gnu-g++)
set(CMAKE_AR ${tools}/bin/aarch64-linux-gnu-ar)
set(CMAKE_RANLIB ${tools}/bin/aarch64-linux-gnu-ranlib)

set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)

• 编译。这样就简洁，复用。

mkdir build && cd build
cmake -DCMAKE_TOOLCHAIN_FILE=../atlas500_host.cmake ..
make

3. hisi3559上执行

• 将编译得到的可执行文件main传到hisi3559上
• 将openblas_install/lib文件上传hisi3559，并设置环境变量

vi /etc/profile
# for openblas lib , 2020-08-04
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/root/1_profile/openblas_lib
source /etc/profile

• 执行main
  ./main

• 结果

-8.000000 16.000000 -2.000000 10.000000 -8.000000 4.000000 7.000000 12.000000 5.000000

reference

1. https://blog.csdn.net/heliangbin87/article/details/103375571
2. https://blog.csdn.net/liqiming100/article/details/78841227