.. -*- rest -*- .. vim:syntax=rest .. NB! Keep this document a valid restructured document. Building and installing NumPy +++++++++++++++++++++++++++++ :Authors: Numpy Developers :Discussions to: numpy-discussion@scipy.org .. Contents:: PREREQUISITES ============= Building NumPy requires the following software installed: 1) Python__ 2.4.x or newer On Debian and derivative (Ubuntu): python python-dev On Windows: the official python installer on Python__ is enough Make sure that the Python package distutils is installed before continuing. For example, in Debian GNU/Linux, distutils is included in the python-dev package. Python must also be compiled with the zlib module enabled. 2) nose__ (optional) 0.10.3 or later This is required for testing numpy, but not for using it. Python__ http://www.python.org nose__ http://somethingaboutorange.com/mrl/projects/nose/ Fortran ABI mismatch ==================== The two most popular open source fortran compilers are g77 and gfortran. Unfortunately, they are not ABI compatible, which means that concretely you should avoid mixing libraries built with one with another. In particular, if your blas/lapack/atlas is built with g77, you *must* use g77 when building numpy and scipy; on the contrary, if your atlas is built with gfortran, you *must* build numpy/scipy with gfortran. Choosing the fortran compiler ----------------------------- To build with g77: python setup.py build --fcompiler=gnu To build with gfortran: python setup.py build --fcompiler=gnu95 How to check the ABI of blas/lapack/atlas ----------------------------------------- One relatively simple and reliable way to check for the compiler used to build a library is to use ldd on the library. If libg2c.so is a dependency, this means that g77 has been used. If libgfortran.so is a a dependency, gfortran has been used. If both are dependencies, this means both have been used, which is almost always a very bad idea. Building with ATLAS support =========================== Ubuntu 8.10 (Intrepid) ---------------------- You can install the necessary packages for optimized ATLAS with this command: sudo apt-get install libatlas-base-dev If you have a recent CPU with SIMD suppport (SSE, SSE2, etc...), you should also install the corresponding package for optimal performances. For example, for SSE2: sudo apt-get install libatlas3gf-sse2 *NOTE*: if you build your own atlas, Intrepid changed its default fortran compiler to gfortran. So you should rebuild everything from scratch, including lapack, to use it on Intrepid. Ubuntu 8.04 and lower --------------------- You can install the necessary packages for optimized ATLAS with this command: sudo apt-get install atlas3-base-dev If you have a recent CPU with SIMD suppport (SSE, SSE2, etc...), you should also install the corresponding package for optimal performances. For example, for SSE2: sudo apt-get install atlas3-sse2 Windows 64 bits notes ===================== Note: only AMD64 is supported (IA64 is not) - AMD64 is the version most people want. Free compilers (mingw-w64) -------------------------- http://mingw-w64.sourceforge.net/ To use the free compilers (mingw-w64), you need to build your own toolchain, as the mingw project only distribute cross-compilers (cross-compilation is not supported by numpy). Since this toolchain is still being worked on, serious compilers bugs can be expected. binutil 2.19 + gcc 4.3.3 + mingw-w64 runtime gives you a working C compiler (but the C++ is broken). gcc 4.4 will hopefully be able to run natively. This is the only tested way to get a numpy with a FULL blas/lapack (scipy does not work because of C++). MS compilers ------------ If you are familiar with MS tools, that's obviously the easiest path, and the compilers are hopefully more mature (although in my experience, they are quite fragile, and often segfault on invalid C code). The main drawback is that no fortran compiler + MS compiler combination has been tested - mingw-w64 gfortran + MS compiler does not work at all (it is unclear whether it ever will). For python 2.5, you need VS 2005 (MS compiler version 14) targetting AMD64 bits, or the Platform SDK v6.0 or below (which gives command line versions of 64 bits target compilers). The PSDK is free. For python 2.6, you need VS 2008. The freely available version does not contains 64 bits compilers (you also need the PSDK, v6.1). It is *crucial* to use the right version: python 2.5 -> version 14, python 2.6, version 15. You can check the compiler version with cl.exe /?. Note also that for python 2.5, 64 bits and 32 bits versions use a different compiler version.