License

same as Numeric

Overview

This module provides BLAS-optimized replacements of Numeric.dot (== Numeric.matrixmultiply) and Numeric.innerproduct. The BLAS libraries supplied by machine vendors are highly optimised and very efficient. The ATLAS team ² provide very good machine optimised BLAS libraries which can be built or downloaded for most machines.

If you are concerned about speed, this dot() can give speed ups of about 80 times (!) on 1000x1000 matrix multiplication of single floats and 40 times for double floats. BLAS routines exist only for 32 & 64 bit float and complex types; so you will experience no speedup for integer types (which will be deferred to Numeric.dot).

from Numeric import * from dotblas import dot

where you formerly had:

from Numeric import *

To use the dotblas version for your whole program you can also do:

import Numeric import dotblas Numeric.dot = dotblas.dot

Or, possibly even simpler, you can just append:

from dotblas import dot

to <python-path>/site-packages/Numeric/Numeric.py, so that all code that imports Numeric will use it by default.

Finally, if you don't know whether dotblas will be available on all the machines where you whish to run your code, you can simply use a try-except statement like this:

import Numeric
try:
    import dotblas
    Numeric.dot = dotblas.dot
 if we can't find the fast dotblas, just use the normal dot
except ImportError: pass

Installation

Make sure you have a BLAS library. If you haven't got a vendor supplied version, get one from ATLAS ¹. They have a number of prebuilt ones, and you may well be able to find a Linux version to suit you with rpmfind ³. Obviously you also need Numeric (22.0 or later recommended). If all those are installed correctly:

python setup.py install

should be enough. Otherwise, you might have to edit setup.py, to change the default settings for library paths etc.

Timing

The script profileDot.py will do some (fairly rudimentary) timings of the new dot compared with the variants available in Numeric.

[1]	http://www.numpy.org

[2]	http://math-atlas.sourceforge.net/

[3]	http://rpmfind.net/linux/rpm2html/search.php?query=blas

Download

dotblas-1.0a.tar.gz