CUDA-based GNU Scientific Library (GSL)

Accelerated Computing GPU-Accelerated Libraries
#1

Hi，

Does anyone know whether the CUDA-based GNU scientific library is available now?

I am working on moving some CPU-based computing to GPU, but found the CUDA-based mathematical and statistical functions are limited. For example, there seems no CUDA-based student’s t inverse distribution function available in any GPU-accelerated libraries.

If you knows such CUDA-based functions and libraries, could you please leave me an link? Many thanks in advance.

#2

I would encourage filing enhancement requests for any desired functions, using the bug reporting form linked from the registered developer website. Please prefix the synopsis of such requests with “RFE:” to mark them as an enhancement request rather than a bug. It is helpful if enhancement requests are accompanied by a brief description of the use case.

I am not very knowledgable about statistical distributions, would it be possible to approximate the Student’s t distribution using the standard normal distribution which is supported by CUDA (normcdf and normcdfinv)?

Otherwise, for now, you could try porting the stdtri() code from the well-regarded Cephes library: http://www.netlib.org/cephes/

#3

Hi, njufaa,

The student’s t inverse distribution function cannot be directly approximated using the normcdf or normcdfinv, but there is an indirect way. I will have a try.

Thanks for your suggestions. I really appreciate them.