Re: A faster sum

[Victor Eijkhout]

On May 21, 2005, at 12:22 PM, Mike Miller wrote:

> It is clearer now that "sum" takes twice as long as "ones" and that, 
> for both, CPU times are porportional to n^2.

Harumpf. Matrix-matrix multiply is only n^3 if both matrices are 
square. I'm an idiot.

Damn. I thought the sum would be slower because it was hitting the 
cache lines the wrong way. Nope:

octave:27> n=4000; X=rand(n) ; t=cputime(); s=sum(X) ; cputime()-t
ans = 0.80000
octave:28> n=4000; X=rand(n) ; t=cputime(); s=sum(X,2) ; cputime()-t
ans = 3.2900

To show consistency:

octave:29> n=4000; X=rand(n) ; t=cputime(); s=ones(1,n)*X ; cputime()-t
ans = 0.21000
octave:30> n=4000; X=rand(n) ; t=cputime(); s=X*ones(n,1) ; cputime()-t
ans = 0.74000

Are matrices in octave stored by rows or by columns? If it's by columns 
(fortran style) these results make sense to me. Otherwise I'm truly 
stymied.

Victor.



-------------------------------------------------------------
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
-------------------------------------------------------------