使用 MKL 编译 Eigen3

Question

在 TensorFlow 1.1 与我在 Ubuntu 16 上进行的C++项目的集成工作期间......我想包括对 MKL 和 64 位整数的支持。我在实例化直接调用 MKL 的模板结构时在 Eigen 库中遇到了编译问题：

 In file included from /usr/local/include/eigen3/unsupported/Eigen/CXX11/../../../Eigen/Core:526:0,
                 from /usr/local/include/eigen3/unsupported/Eigen/CXX11/Tensor:14,
                 from /home/drormeirovich/projects/tensorflow/third_party/eigen3/unsupported/Eigen/CXX11/Tensor:1,
                 from /home/drormeirovich/projects/tensorflow/tensorflow/core/framework/tensor.h:19,
                 from /home/drormeirovich/projects/tensorflow/tensorflow/cc/framework/ops.h:21,
                 from /home/drormeirovich/projects/tensorflow/tensorflow/cc/client/client_session.h:24,
                 from /home/drormeirovich/projects/my_project.cpp:10:
/usr/local/include/eigen3/unsupported/Eigen/CXX11/../../../Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h: In static member function ‘static void Eigen::internal::general_matrix_matrix_product<Index, double, LhsStorageOrder, ConjugateLhs, double, RhsStorageOrder, ConjugateRhs, 0>::run(Index, Index, Index, const double*, Index, const double*, Index, double*, Index, double, Eigen::internal::level3_blocking<double, double>&, Eigen::internal::GemmParallelInfo<Index>*)’:
/usr/local/include/eigen3/unsupported/Eigen/CXX11/../../../Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h:103:173: error: cannot convert ‘char*’ to ‘CBLAS_LAYOUT’ for argument ‘1’ to ‘void cblas_dgemm(CBLAS_LAYOUT, CBLAS_TRANSPOSE, CBLAS_TRANSPOSE, long long int, long long int, long long int, double, const double*, long long int, const double*, long long int, double, double*, long long int)’
   BLASPREFIX##gemm(&transa, &transb, &m, &n, &k, &numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, &numext::real_ref(beta), (BLASTYPE*)res, &ldc); 
                                                                                                                                                                             ^
/usr/local/include/eigen3/unsupported/Eigen/CXX11/../../../Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h:106:1: note: in expansion of macro ‘GEMM_SPECIALIZATION’
 GEMM_SPECIALIZATION(double,   d,  double, cblas_d)
 ^

欲了解更多详情...我在这个集成问题上的全部进展在这个链接中：

https://docs.google.com/document/d/1VFTdPJy59QTCTHO8NHMNmnO8AOoQhNXgWixas9KmLLM/edit?usp=drivesdk

我是否必须从 Eigen3 中删除对 MKL 的支持？

任何帮助将不胜感激...

Answer 1

免责声明：我曾经是一名EasyBuild开发人员。

在EasyBuild中，我们可以在MKL支持下构建Eigen3，所以这应该可以工作。

我们的一位贡献者似乎已经发现，对于 eigen3，您需要将"signature_of_eigen3_matrix_library"文件复制到用于包含的路径中，请参阅https://github.com/hpcugent/easybuild-easyblocks/blob/master/easybuild/easyblocks/e/eigen.py

https://github.com/RLovelett/eigen/blob/master/signature_of_eigen3_matrix_library