Re: speeding up function

[c.]

On 12 Jul 2010, at 12:42, Jaroslav Hajek wrote:
> hardly, obviously.

still, as I need the function to be matlab-compatible this info is  
relevant to me

> > but
> > according to matlab's profiler, in the call
> >
> > aa = randn (4, 4, 100, 100, 25); res = matarrayfun (@inv , aa, [1,  
> > 2]);
> >
> > most of the time is consumed by 'num2cell'...
> >
> > c.
>
> If you don't work with the development version, my first advice is to
> upgrade.

I'd love to, unfortunately it's a few months (early march IIRC) since  
building tip on my system (OS X 10.5) last worked.
I pull, update and retry to build regularly (about every 2 weeks) but  
I consistently get the same problem:

after the octave binary is built and linked and the build system  
proceed to producing the docs, as soon as octave is run
it crashes with a "segmentation-fault"

I tried to debug this using gdb/valgrind but got no useful hints.
I think a similar problem was recently reported on the list from  
someone building on redhat,
in that case someone suggested a problem with atlas/blas/lapack, but  
I'm not sure there is any connection to my case.

> Then, I'd suggest you take a representative case and try to
> measure the actual time consumed by various parts using tic/toc or
> cputime.

If you beleive there is any hope to improve matarrayfun,
I'll look into this next week and report back.

> Your case for cell2mat seems rather special to me; perhaps it would be
> better handled by using cat() directly.

I couldn't find a way to this without for cycles, is there something  
obvious I am missing?

c.