MTGP32 generator (Mersenne Twister) Problem with generator in v4.1

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello all,
I read about MTGP32 generator in docs CURAND_Library.pdf( http://developer.download.nvidia.com/compute/DevZone/docs/html/CUDALibraries/doc/CURAND_Library.pdf )on 14-th page.
But I didn’t find such functions (curandMakeMTGP32Constants() or curandMakeMTGP32KernelState()) in headers of Cuda toolkit v4.1.
So there is my question: is there any chance to use this functions and how?

2

Neither in toolkit 4.2 documentation I can find how to use [font=“Courier New”]curandMakeMTGP32Constants()[/font] or [font=“Courier New”]curandMakeMTGP32KernelState()[/font]… I mean, the CURAND Library 4.2 documentation mention them, but there’s no example on how to use them. I took a look on [font=“Courier New”]curand_mtgp32_kernel.h[/font] and [font=“Courier New”]curand_mtgp32.h[/font] codes and there is no mention to any of them.

How can I initialize MTGP32 within device? There is no [font=“Courier New”]curand_init()[/font] overload for [font=“Courier New”]curandStateMtgp32_t[/font].

I simply cannot believe there’s no way of using MTGP32 by Device API because of a missing initialization function… External Image

3

The CURAND team is aware of issues with MTGP32 in CUDA 4.2. Apparently a required header file somehow got excluded from the installation. This issue will be addressed in CUDA 5.0. Best I know the necessary fix did not make it into the currently available CUDA 5.0 preview. Sorry for the inconvenience.

4

Thanks, njuffa. I’ll wait for it. External Image

5

I have the same problem even though I’m using CUDA 5.0 and including curand_mtgp32_kernel.h. No overload for curand_init(int, int, int, curandStateMtgp32_t *).