R and CUDA programming

Dear experts,

I was wondering if it is possible for the individual threads in a CUDA  
program to access the R API, e.g. to call it's random number  
generator. If this is possible, is it advisable or is there a  
recommended way of doing this?

If it is not possible/advisable, what is the best way to get the GPU  
device to generate random numbers?

Thanks in advance for any pointers,

Vik