如何让openBLAS与openMP协同工作？

Question

我从openBLAS那里得到了很多警告，比如

OpenBLAS Warning : Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option. OpenBLAS Warning : Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option. OpenBLAS Warning : Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option. OpenBLAS Warning : Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option. OpenBLAS Warning : Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option. OpenBLAS Warning : Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option. OpenBLAS Warning : Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option. OpenBLAS Warning : Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option.

下面是我的src/Makevars文件

PKG_CPPFLAGS = $(SHLIB_OPENMP_CXXFLAGS)
PKG_LIBS = $(SHLIB_OPENMP_CXXFLAGS) $(LAPACK_LIBS) $(BLAS_LIBS) $(FLIBS)

下面是我的头文件：

#define ARMA_NO_DEBUG

#ifdef _OPENMP
#include <omp.h>
#endif

//[[Rcpp::depends(RcppArmadillo)]]
#include <RcppArmadillo.h>

//[[Rcpp::depends(RcppProgress)]]
#include <progress.hpp>

#include <Rcpp.h>
#include <R.h>

在我的代码中使用openMP，如下所示

#pragma omp parallel for num_threads(n_threads) schedule(dynamic) private(W, mu)
for (unsigned int j = 0; j < m; j++)

其中n_threads是通过函数参数传递的。当n_threads = 1的时候一切都很好，但是当n_threads = 2的时候得到上面的警告。完整的代码可以在here找到，它也依赖于其他文件(对不起，我不能在这里粘贴，因为它有点长)。

我尝试将USE_OPENMP=1添加到src/Makevars中，但不起作用。任何人都有解决方案吗？谢谢!

Answer 1

在CPU模式下运行Torch时，我收到了相同的警告。用OPEN_MP 1重建OpenBLAS为我修复了这个问题。

如果要使用OPEN_MP 1重建OpenBLAS，则：

转到您克隆OpenBLAS存储库的文件夹(对于我来说，它是/home/brt/OpenBLAS/OpenBLAS)

cd /home/brt/code/OpenBLAS/
make clean
make USE_OPENMP=1
sudo make install

如果您还没有这样做，

sudo vi /etc/ld.so.conf.d/openblas.conf

添加行: /opt/OpenBLAS/lib

保存并关闭

sudo ldconfig

这将重新生成OpenBLAS，并将OPEN_MP设置为1