Re: BLAS lib performance (was Re: MPI (was Re: BIG libraries ....))

[John W. Eaton]

On  6-Oct-2005, Quentin Spencer wrote:

| I haven't looked at the source code, but I think Octave doesn't do this 
| automatically. I was recently comparing Octave with ATLAS libraries to 
| Matlab, and I discovered that in Octave, Z=X*Y takes the same amount of 
| time as Z=X*X', while in Matlab, the second is faster than the first.

It would be a nice project for someone to add these optimizations.
There are a couple of ways you could do it.  Probably the simplest
would be to modify the code that handles the evaluation of the binary
op so that it checks to see whether either or both of the operands are
tanspose (or hermitian) operators.  Then you could call a function to
handle the combined operations.  It would be slightly better to
perform this optimization just once, when an expression (or function)
is first parsed.  The idea would be the same, but you would modify the
parse tree so the evaluator would not have to check each time it runs.

jwe



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

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