标签：python lapack altas numpy install
上篇笔记介绍了不依赖lapack和atlas库的NumPy包源码编译/安装方法，但“纯净版”的NumPy会损失性能，故本篇笔记说明如何源码编译安装依赖lapack和atlas库的NumPy包。
1. GCC版本要求
使用较新版本的GCC工具集（尽量不低于v4.7）且集成有gfortran编译器。
备注1：这里大写的"GCC"是指GNU Compiler Collection，它除包含C语言编译器gcc外，还包含很多其它语言的编译器（如g++/gfortran等）
备注2：3.x版的的C语言编译器gcc会由于某些头文件缺失导致编译atlas库报错
备注3：若GCC工具集中没有gfortran编译器，则编译lapack库时会遇到一些莫名其妙的错误（因为lapack是用fortran编写的），好在GCC4.7及以上版本中已经集成了gfortran编译器
在GCC版本符合要求的前提下，临时将其加入环境变量PATH并设置动态库查找路径：
$ export PATH=/home/slvher/tools/gcc48/bin/:$PATH $ export LD_LIBRARY_PATH=/home/slvher/tools/gcc48/lib64:/home/slvher/tools/gcc48/lib备注4：在当前shell会话中临时设置LD_LIBRARY_PATH可以保证编译过程中正确搜索到GCC库，但最好不要设置到.bash_profile中，因为那样会影响其它程序的查找路径，可能会踩到坑。
$ ../configure --shared -b 64 --prefix=/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs --with-netlib-lapack-tarfile=/home/slvher/tools/scikit-learn-virtualenv/dep-libs/lapack-3.5.0.tgz其中，--shared表明要编译atlas共享库（configure会自动在编译命令中插入"-fPIC"参数，无需在这里显式指定）；--prefix指定编译结果的安装路径；--with-netlib-lapack-tarfile表明编译atlas库时会用相同的编译器及编译/链接参数自动编译lapack库，这里指定lapack源码包的路径后，configure运行后会自动解压lapack源码并将其拷贝至BLDdir/src/lapack/reference/这个目录下。
3. 编译优化版NumPy包
前提：官网下载NumPy源码包并解压，这里以目前最新版numpy-1.9.2.tar.gz为例进行说明。
1) cd至解压目录numpy-1.9.2
2) cp site.cfg.example site.cfg
3) 在site.cfg中配置atlas项，其中include_dirs和library_dirs是atlas库安装路径下的include和lib目录
[atlas] atlas_libs = lapack,f77blas,cblas,atlas library_dirs = /home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/lib include_dirs = /home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/include4) python setup.py config
>>> import numpy as np
>>> np.__config__.show()
atlas_3_10_blas_threads_info:
libraries = ['lapack', 'f77blas', 'cblas', 'atlas']
library_dirs = ['/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/lib']
define_macros = [('HAVE_CBLAS', None), ('ATLAS_INFO', '"\\"3.10.2\\""')]
language = c
include_dirs = ['/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/include']
lapack_opt_info:
libraries = ['tatlas', 'lapack', 'f77blas', 'cblas', 'atlas']
library_dirs = ['/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/lib']
define_macros = [('ATLAS_INFO', '"\\"3.10.2\\""')]
language = f77
include_dirs = ['/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/include']
blas_opt_info:
libraries = ['lapack', 'f77blas', 'cblas', 'atlas']
library_dirs = ['/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/lib']
define_macros = [('HAVE_CBLAS', None), ('ATLAS_INFO', '"\\"3.10.2\\""')]
language = c
include_dirs = ['/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/include']
openblas_info:
NOT AVAILABLE
openblas_lapack_info:
NOT AVAILABLE
atlas_3_10_threads_info:
libraries = ['tatlas', 'lapack', 'f77blas', 'cblas', 'atlas']
library_dirs = ['/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/lib']
define_macros = [('ATLAS_INFO', '"\\"3.10.2\\""')]
language = f77
include_dirs = ['/home/slvher/tools/scikit-learn-virtualenv/dep-libs/sklearn-libs/include']
lapack_mkl_info:
NOT AVAILABLE
blas_mkl_info:
NOT AVAILABLE
mkl_info:
NOT AVAILABLE也可以用具体的例子来验证其功能是否正常：>>> import numpy as np
>>> np.arange(15).reshape(3, 5)
array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14]])
>>>
>>> a = np.arange(15).reshape(3, 5)
>>> a
array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14]])
>>> type(a)
<type 'numpy.ndarray'>
>>>
>>>
>>> from numpy.linalg import *
>>> b = np.array([[1.0, 2.0], [3.0, 4.0]])
>>> b
array([[ 1., 2.],
[ 3., 4.]])
>>> b.transpose()
array([[ 1., 3.],
[ 2., 4.]])
>>> inv(b)
array([[-2. , 1. ],
[ 1.5, -0.5]])
>>>
【参考资料】======================= EOF ======================
【Python笔记】如何源码编译依赖LAPACK和ATLAS库的NumPy包
标签：python lapack altas numpy install
原文地址：http://blog.csdn.net/slvher/article/details/44853721
