On Donnerstag 29 Januar 2009, Jozef Vesely wrote:
> Hello,
>
> here is the alternative implementation of curandom.rand, the
> md5 crypto hash function of threadIdx, blockIdx, seed and
> other variables is used to generate random numbers (thus
> there should be no issues with combining results from
> several instances).
>
> It is about twice slower than present (not working) rand
> function. However 4x speed up could be achieved by using all
> 16 bytes returned by md5.
>
> Please CC me on the replies since I am not on pycuda list

I just replaced the old RNG with yours. I'm a little bit worried about the 
advertising clause in the RSA license, but I've added a blurb to the docs so 
that I hope we're ok.

Thanks for your contribution!

Andreas

Attachment: signature.asc
Description: This is a digitally signed message part.

_______________________________________________
PyCuda mailing list
[email protected]
http://tiker.net/mailman/listinfo/pycuda_tiker.net

Reply via email to