Hi Bob and All,

In an effort to find ways to improve the processing speed of our lengthly deconvolutions, we've run into another possibility, and I am looking for some advice or an expert opinion. We have several Mac PowerPC G4's with the Motorola AltiVec processor (a.k.a Velocity Engine). My simple understanding of the AltiVec is that it contains four additional floating point processors. Do you think that getting AFNI to take advantage of this hardware, could be as simple as us using an AltiVec sensitive compiler on the current AFNI source code, say from Veridian (http://www.psrv.com/)? Alternatively, would we need to include special code or modify existing code to take advantage of AltiVec?

We've successfully run AFNI with the G4, but didn't get as much of a bump in speed as hoped. An 8 hour job on a 200MHz SGI Origin 2000 took 5 hours on the 1GHz G4.

The reason this seems to be such an interesting possibility is that Apple and Genentech teamed up to tune the BLAST bioinformatics tool to use the AltiVec processor on the G4 with impressive results (http://www.apple.com/pr/library/2002/feb/07blast.html).

Any suggestions or comments would be most welcome. Thanks,

-jim