Re: Porting Octave to IBM Cell Processor

[Christopher Hulbert]

On 4/7/07, Shai Ayal <address@hidden> wrote:
> For most of the operations you talk about, octave uses external
> libraries (e.g. fftw for fft, lapack for matrix ops), so the place to
> start is to optimise these libraries for the Cell. I don't think this
> is an easy task. However chances are that this was already done by
> someone (these are pretty popular libraries), so your task will
> degenerate into getting octave to compile on the Cell with these
> libraries.

codesourcery is working on porting VSIPL++ to the Cell processor. I
think they have the FFT and some other routines implemented. I don't
think they have the cell code in a release as of yet, but you may
contact them about obtaining a development version.

http://www.codesourcery.com/vsiplplusplus/
http://hpec-si.com/

> In any case I would start bu downloading the source and trying to
> compile it -- ust to see where you're standing
> Good luck
> Shai
>
> On 4/7/07, Balaji Rao <address@hidden> wrote:
> > Hello ppl,
> >
> > I am an undergraduate student from India. I want to port Octave to CELL
> > Processor(
> > http://en.wikipedia.org/wiki/Cell_microprocessor). As you
> > might be aware of, the CELL processor has 8 vector co processors. Hence, a
> > huge amount of parallelism can be achieved. Also, since this particular
> > processor is highly optimised for Single  Precision Floating Point and
> > Integer Operations, it can be used by octave to achieve higher number
> > crunching speed. Certain algorithms like FFT, Matrix Multiplication etc can
> > be easily parallelized and vectorized. Hence i feel that Octave will gain
> > huge advantage by porting it to CELL.
> >
> > If you are happy with this idea, please tell me where to start and how to go
> > about it.
> >
> > Thank you in advance.
> >
> > --
> > warm regards
> >
> > Balaji Rao