Does PGI Workstation supports OpenMP to CUDA translation

Hello, everybody!

I am new with PGI and CUDA , now I’m evaluating PGI Workstation, so excuse me for may be silly question. My issue is that I have a code written with usual Fortran with use of OpenMP and I want to run it using parallelism of GPU’s processors. I supposed that PGI Workstation is a tool which exactly would make it possible but after looking through the manual I realized that PGI can’t perform something like automatic translation of OpenMP to CUDA and one has to manually tune the program according to CUDA Fortran specification…Am I correct?

Thanks in advance, Anton

Hi Anton,

You might want to take a look at the PGI Accelerator Model instead of CUDA Fortran. (See: http://www.pgroup.com/resources/accel.htm). The PGI Accelerator is a directive based approach similar to OpenMP. We did first look at re-purposing OpenMP for use with GPUs but programing for a GPU is very different then a SMP system and we thought it best to use a different approach. For more detail please refer to the following article: http://www.pgroup.com/lit/articles/insider/v2n2a5.htm.

Hope this helps,
Mat

Hi, Mat,
Thank you for reply and interesting links! Now it’s getting clear to me, I’ll probably going to learn your accelerator model and then try to compile my code)

Best wishes, Anton