[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Lapack routine sppedups

Hi Clint!  Thanks for the below.  My apologies, it was indeed a
miscompilation on my part of the tester programs.  lu and llt do
indeed show significant improvement in atlas.

Thanks for the good work!

R Clint Whaley <rwhaley@cs.utk.edu> writes:

> >Greetings!  What speedups do people typically find from the lapack
> >testers?  When using a fast blas in the lapack testers, I can't find
> >much difference at all between x?lutst and x?lutstF, for example.  I'm
> >just trying to see if I'm building the libraries incorrectly or
> >something. 
> x?lutst is typically better then x?lutstF for small problems, and very
> large problems.  If you have a good level 2 BLAS, it may not be better
> for small.  Asymtotically lutst should win, but by how much again depends
> on the platform.  Have you tried running large problems, say something like:
>    ./xdlutst -T 0 -N 1200 3000 200
> Cheers,
> Clint

Camm Maguire			     			camm@enhanced.com
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah