genefer/CUDA

[msft]

Hi ,
I convert genefer.c to CUDA.

cudagenefer.0.1$ time ./CUDAGenefer input
GeneFer 1.3 (CUDA) Copyright (C) 2001-2002, Yves GALLOT
Copyright (C) 2010, Shoichiro Yamada
A program for finding large probable generalized Fermat primes.

GeneFer -t run test
GeneFer <filename> test <filename>
GeneFer use interactive mode

Start test of file 'input'.
1651902^16384+1 is a probable prime. (err = 3.75e-01)

real 2m8.700s
user 0m53.930s
sys 1m14.650s

[mdettweiler]

Very cool! Methinks I'll give this a try as soon as my friend with the GTX 460 gets it back online (still waiting for an OS reinstall...Ubuntu 9.04 is not your friend when it comes to installing CUDA drivers).

[rogue]

Originally Posted by msft

Hi ,
I convert genefer.c to CUDA.

I suggest that you add the improvements from genefer 2.2.0, which is used by PRPNet on PrimeGrid.

[msft]

Hi, mdettweiler, rouge

Originally Posted by rogue

I suggest that you add the improvements from genefer 2.2.0, which is used by PRPNet on PrimeGrid.

Can i have more information ?

[msft]

Hi ,

cudagenefer.0.2$ time ./CUDAGenefer input
GeneFer 1.3 (CUDA) Copyright (C) 2001-2002, Yves GALLOT
Copyright (C) 2010, Shoichiro Yamada
A program for finding large probable generalized Fermat primes.

GeneFer -t run test
GeneFer <filename> test <filename>
GeneFer use interactive mode

Start test of file 'input'.
48594^65536+1 is a probable prime. (err = 0.00e+00)

real 7m40.662s
user 3m35.520s
sys 4m5.030s

[frmky]

Originally Posted by msft

Can i have more information ?

The modifications were done by David Underbakke. The code is at http://www.underbakke.com/genefer/.

[msft]

Hi ,frmky

Originally Posted by frmky

The modifications were done by David Underbakke. The code is at http://www.underbakke.com/genefer/.

Good information.
Thank you,

[msft]

Convert complete.

cudagenefer.0.3$ time ./CUDAGenefer input
GeneFer 1.3 (CUDA) Copyright (C) 2001-2002, Yves GALLOT
Copyright (C) 2010, Shoichiro Yamada
A program for finding large probable generalized Fermat primes.

GeneFer -t run test
GeneFer <filename> test <filename>
GeneFer use interactive mode

Start test of file 'input'.
4^524288+1 is a probable prime. (err = 1.11e-16)

real 36m49.528s
user 13m36.050s
sys 23m13.320s

[rogue]

Originally Posted by msft

Hi, mdettweiler, rouge

Can i have more information ?

You can find the sourcesto the various versions of genefer 2.2.0 here.

genefer 2.2.0 has these enhancements over genefer 1.x:

• checkpointing - ability to stop and continue PRP test
• residue generation - ability to compare residue with another run on another computer to ensure PRP test didn't fail due to some problem with the hardware/software
• residue verification - ability to compare residue with known residues to ensure that have a good build
• quick option -use -q to pass number to test on the command line
• error checking - ability to stop a PRP test immediately if a roundoff error is detected

There are a few more, but you can see that many these modifications are very important when trying to ensure that the test result from genefer is valid.

To incorporate your changes, start with the 32-bit Windows with 64-bit x87 FPU sources. I recommend that you post your progress here because I think that what you are working on with be of great interest to PrimeGrid. AFAIK, none of the other GFN search projects are still running. Providing a means for CUDA users to PRP test GFNs will add a lot of excitement to that project.

[msft]

Hi, rogue
Now Post to PrimeGrid.

[frmky]

Originally Posted by rogue

You can find the sourcesto the various versions of genefer 2.2.0 here.

I don't see the source on that page, only the binaries. Is there a link I'm missing?