MTGP for Fermi?

Hi,

I’m currently working on a CUDA project that utilizes the MTGP random number generator.

http://www.math.sci.hiroshima-u.ac.jp/~m-m…MTGP/index.html

Apparently, the code is not optimized for Fermi architecture. Has anyone tried optimizing it for Fermi? If so, any pointers on what might need to be changed to accelerate the code? Thanks.

Any luck with this?