Re: How does Octave shine?

[Fredrik Lingvall]

Cameron Laird wrote:

On Thu, Sep 21, 2006 at 08:39:45AM +0200, David Bateman wrote:
			.
			.
			.
  

If you have the matlab communications toolbox, one nasty example for
matlab I like is

N=5; for n=100:100:500, t=cputime(); inv(gf(randint(n,n,2^N), N));
cputime() - t, end


try this with octave-forge forges communications toolbox. And this is a
realistic thing to want to do in communications i you deal with LDPC codes.
    

			.
			.
			.
Many people have helped with their follow-ups; in most
cases, I've thanked them privately.  In this case, I'll
say a few words on the mailing list.

Yes, thank you, this is exactly the sort of specific
response I find helpful.  As it turns out, I do NOT
have the Matlab communications toolbox, and would
appreciate knowing what I'm getting into here.  I
suspect you're telling me that Octave handles the 
sequence above in a fraction of the time of Matlab; do
I have that right?  What sort of fraction have you
experienced?
  

On an RHEL4 Opteron 2 cpu dual-core machine:

1) Matlab Version 7.1.0.183 (R14) Service Pack 3

>> N=5; for n=100:100:500, t=cputime(); inv(gf(randint(n,n,2^N), N));cputime() - t, end

ans =

    1.0300


ans =

    4.8800


ans =

   13.0100


ans =

   26.7200


ans =

   48.2000

2) Octave version 2.9.8 (x86_64-unknown-linux-gnu)

octave:1> N=5; for n=100:100:500, t=cputime(); inv(gf(randint(n,n,2^N), N));cputime() - t, end
ans =  0.088987
ans =  0.38994
ans =  1.3278
ans =  3.0645
ans =  6.3920

Both Matlab and Octave uses Goto BLAS v 1.06 (libgoto_opteronp-r1.06.so)

/Fredrik