Re: QUERY: what is status/recommendations for Octave + ATLAS compilation

[Steven G. Johnson]

Dennis Decoste <address@hidden> wrote:
> [...] Apparently MATLAB optimized their new numerics
> library a bit more than the standard BLAS/LAPACK
> available with the standard Octave distribution. [...]
> I'm also a bit surprised to find Octave's standard mult
> is more than twice as slow as MATLAB's old one [...]

Hi Dennis,

The results really aren't at all surprising, and one need not infer any
special hackery on Mathworks' part to explain them.

Octave currently uses the default *unoptimized* BLAS implementation; e.g.
the matrix multiplication is just a naive triple-loop.  The new MATLAB is
using an optimized BLAS, so of course they are going to be much faster than
Octave currently is.  Even the old MATLAB probably used somewhat-optimized
home-brewed routines, so it's not surprising that the classic MATLAB is
also somewhat faster than the current completely unoptimized routines
Octave uses (especially for simple things like matrix multiplication where
LAPACK does not give any advantage, and only BLAS matters).

I don't know if Mathworks has said whose BLAS implementation they use, but
considering that ATLAS is usually very close in performance even to
vendor-optimized codes, Octave will probably be roughly as fast as MATLAB
for dense matrix operations once you plug in ATLAS.

Steven




-----------------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.che.wisc.edu/octave/octave.html
How to fund new projects:  http://www.che.wisc.edu/octave/funding.html
Subscription information:  http://www.che.wisc.edu/octave/archive.html
-----------------------------------------------------------------------