Does nvc support GPU offloading with OpenMP

Accelerated Computing HPC Compilers nvc, nvc++ and nvfortran
1

Hi,

Can we compile and run an OpenMP target offloading code with nvc?
I have NVIDIA HPC 20.7.0 installed. I tried building my code with “nvc -mp” option. It builds, but fails to run on GPU.
When run with nvprof it gives the error:

The path to CUPTI and CUDA Injection libraries might not be set in LD_LIBRARY_PATH.

I have set the library path in LD_LIBRARY_PATH, still this error.
Please advise.

Thank You.

2

The new 20.11 release is the first with Beta support for offloading OpenMP target regions. Since it’s beta, several features are missing and there are most likely compilation and performance issues, but please give it a try and provide feedback.

You can download 20.11 at: https://developer.nvidia.com/hpc-sdk

Also, the flag to enable GPU offload is “-mp=gpu”.

Thanks,
Mat

3

Thank you for the reply @MatColgrove. I tried with 20.11 and it works. Will submit performance issues once I find them.

Thank You.