Optimizing SETI algorithms (possible parallelization)

[ziglar]

I have a colleague who has access to a PC card that contains two CSX600 processors on a PCI-X form factor that delivers 50 GFLOPS of sustained double-precision matrix multiply.

He would like to create some kind of demo that involves the SETI@Home client and the above card but don't know where to start. We can obviously download the source code but are there any sites that show the areas of the code that you would actually optimize?

Is this likely to involve the rewritting of FFT algorithms or is there some 'job scheduling' code that could benefit? The change would probably involve looking for areas of the code that could be parallelized - is anybody doing any work in this area?

Thanks

 

[Ken g6]

All the optimizations I've seen involve the FFT. I haven't done SETI recently, but I imagine it might be possible to do several frequency shifts at once, but somebody else here may know more. There should also be optimized source code somewhere to compare to.

Also, you might consider optimizing primality testing for Prime95 or the like. They use double-precision FFTs, too, and many more people have worked on that code. Source code here, and forums here.