can't execute CUDALucas on GNU/Linux

[emily]

I've a P4 machine with an nVidia GT440 card and I want to run CUDALucas 1.4.2 (from the PDF) using 64bit GNU/Linux. I download the executable but it does nothing :'( It just says permission denied. I tried to chmod u+x

I've also got an i7 machine with ATI/AMD HD-6770, is there any way to do LL primality test on the AMD GPU?

Also, can I run mfakto for the AMD GPU using the free/open-source ATI driver? I don't use Catalyst on GNU/Linux.

[Dubslow]

Originally Posted by emily

I've also got an i7 machine with ATI/AMD HD-6770, is there any way to do LL primality test on the AMD GPU?

Not as of yet.

As to the first question, I assume the chmod appeared to work fine?

If you have a Windows partition, and the software is in said partition, everything defaults to 'root' in Linux because NTFS permissions don't work very well. If that's the case, then sudo should do the trick. If not, and chmod doesn't work, then I'm clueless.

[moebius]

Are you sure that you are the owner of the file? If not try the 'chown' command.

[Dubslow]

Originally Posted by moebius

Are you sure that you are the owner of the file? If not try the 'chown' command.

Note that the command will still not affect anything on a NTFS partition (and presumably most others) as mentioned above.

[emily]

Oops, correct about the NTFS!

And about the LL support for AMD/ATI GPUs, is there any particular reason (e.g. technical) why it hasn't been implemented, or are we just waiting for someone to do it? I assume it's all about an OpenCL re-write of what CUDALucas does and then some debugging/optimization, no? Or is AMD/ATI architecture not suitable for LL?

[fivemack]

Originally Posted by emily

Oops, correct about the NTFS!

And about the LL support for AMD/ATI GPUs, is there any particular reason (e.g. technical) why it hasn't been implemented, or are we just waiting for someone to do it? I assume it's all about an OpenCL re-write of what CUDALucas does and then some debugging/optimization, no? Or is AMD/ATI architecture not suitable for LL?

What CUDALucas does at its heart is a call to a pre-written CUDA FFT library; as far as I know there isn't an equivalent OpenCL FFT library, and designing and implementing a competent one of those would be a hard one-year masters project.

[ckdo]

Originally Posted by Dubslow

Note that the command will still not affect anything on a NTFS partition (and presumably most others) as mentioned above.

Take note that e.g. Ubuntu will auto-set +x for files on NTFS if their name ends in ".exe" (remounting the file system might be necessary after renaming).

[emily]

Originally Posted by fivemack

as far as I know there isn't an equivalent OpenCL FFT library.

Googling for "OpenCL FFT" does return some results, though, so apparently other people also need this functionality or are working on it or on something similar.

When there will be an OpenCL FFT library, would we be correct to assume that the implementation for Mersenne primes and the development time would be equivalent to the effort spent for CUDALucas? Was it too hard to code CUDALucas?