Hi Gregory,

can you please send PyCUDA-related questions to the mailing list in the
future?

On Thu, 30 Sep 2010 18:46:55 +0000, Gregory Perry <[email protected]> 
wrote:
> How exactly does PyCuda work with Cuda?  Is it an overlay on top of
> the Cuda development tools or does it generate native object code for
> the GPU somehow?

Uses nvcc to generate .cubin GPU object code.

> Python natively supports multi precision bignums, would PyCuda then be
> able to generate GPU code that is multi precision, and more
> importantly, would it already be optimized to take advantage of Fermi
> 32-bit integer multipliers?  I just picked up a GTS 350 last week
> which is Fermi based and plan on getting it installed with PyCuda
> today or tomorrow hopefully.

PyCUDA doesn't support GPU bignums. This is nontrivial because of a
bignum's inherent variable size. Arbitrary-fixed-size integers should be
easy to do though.

Thanks,
Andreas

Attachment: pgpU2aJmveIad.pgp
Description: PGP signature

_______________________________________________
PyCUDA mailing list
[email protected]
http://lists.tiker.net/listinfo/pycuda

Reply via email to