Anaconda 4.0.2 是否已经在 MKL 上运行 numpy

Question

我的系统中有 Ananconda4.0.2。我尝试在 python 控制台中通过以下命令找出我的 numpy 配置：

import numpy as np
np.__config__.show()

它返回以下输出

lapack_opt_info:
libraries = ['mkl_lapack95_lp64', 'mkl_intel_lp64', 'mkl_intel_thread', 'mkl_core', 'iomp5', 'pthread']
library_dirs = ['/home/<username>/anaconda2/lib']
define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
include_dirs = ['/home/<username>/anaconda2/include']
blas_opt_info:
libraries = ['mkl_intel_lp64', 'mkl_intel_thread', 'mkl_core', 'iomp5', 'pthread']
library_dirs = ['/home/<username>/anaconda2/lib']
define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
include_dirs = ['/home/<username>/anaconda2/include']
openblas_lapack_info:
NOT AVAILABLE
lapack_mkl_info:
libraries = ['mkl_lapack95_lp64', 'mkl_intel_lp64', 'mkl_intel_thread','mkl_core', 'iomp5', 'pthread']
library_dirs = ['/home/<username>/anaconda2/lib']
define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
include_dirs = ['/home/<username>/anaconda2/include']
blas_mkl_info:
libraries = ['mkl_intel_lp64', 'mkl_intel_thread', 'mkl_core', 'iomp5', 'pthread']
library_dirs = ['/home/<username>/anaconda2/lib']
define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
include_dirs = ['/home/<username>/anaconda2/include']
mkl_info:
libraries = ['mkl_intel_lp64', 'mkl_intel_thread', 'mkl_core', 'iomp5', 'pthread']
library_dirs = ['/home/<username>/anaconda2/lib']
define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
include_dirs = ['/home/<username>/anaconda2/include']

这是否意味着我的 numpy 已经使用了 mkl？

Answer 1

来自 numpy 1.15.0 OpenBLAS seems to be the default。我在 1.16.2 上，这成立：

In [2]:np.__version__
Out[2]: '1.16.2'

In [3]: np.__config__.show()
blas_mkl_info:
  NOT AVAILABLE
blis_info:
    libraries = ['blas', 'cblas', 'lapack', 'blas', 'cblas', 'lapack']
    library_dirs = ['/home/tams00/anaconda3/envs/course2019/lib']
    define_macros = [('HAVE_CBLAS', None)]
    include_dirs = ['/home/tams00/anaconda3/envs/course2019/include']
    language = c
blas_opt_info:
    libraries = ['blas', 'cblas', 'lapack', 'blas', 'cblas', 'lapack', 'blas', 'cblas', 'lapack']
    library_dirs = ['/home/tams00/anaconda3/envs/course2019/lib']
    define_macros = [('HAVE_CBLAS', None)]
    include_dirs = ['/home/tams00/anaconda3/envs/course2019/include']
    language = c
lapack_mkl_info:
  NOT AVAILABLE
openblas_lapack_info:
    libraries = ['blas', 'cblas', 'lapack', 'blas', 'cblas', 'lapack']
    library_dirs = ['/home/tams00/anaconda3/envs/course2019/lib']
    language = c
    define_macros = [('HAVE_CBLAS', None)]
lapack_opt_info:
    libraries = ['blas', 'cblas', 'lapack', 'blas', 'cblas', 'lapack', 'blas', 'cblas', 'lapack']
    library_dirs = ['/home/tams00/anaconda3/envs/course2019/lib']
    language = c
    define_macros = [('HAVE_CBLAS', None)]

Answer 2

是的，从 Anaconda 2.5 MKL support is the default 开始。

要确定，请输入：

conda search numpy

你会看到类似的东西

*  1.10.4                   py27_0  defaults        
   1.10.4             py27_nomkl_0  defaults        [nomkl]

或

.  1.10.4                   py35_0  defaults        
   1.10.4             py35_nomkl_0  defaults        [nomkl]

因此，没有 MKL 的版本被明确标记为 nomkl，而安装的版本（* 或 .）包括 MKL。