OpenMP 4.5 GPU benchmarks

Hi,

I am working on a static analysis tool using LLVM for debugging usage of OpenMP 4.5 target pragma.

Can anyone point me to some GPU benchmarks that use the target offload feature. (Other than SPEC)

You should probably ask on one of the lists at http://OpenMP.org, since this is not specifically an LLVM request.

https://github.com/ParRes/Kernels/ has reference implementations of a few HPC oriented kernels but you will need to optimize them to use as benchmarks. They are in that you can compare OpenMP to other models fairly, assuming you put in the same optimization effort for all.

I haven’t checked if RAJA@OpenMPtarget works on GPUs but if it does, you can try RAJA benchmarks like RAJAperf and Kripke.

I have a version of NWChem CCSD(T) that can use OpenMP 4.5 target (we have KNC LEO in the past but this is new and different code targeting pure OpenMP 4.5). It comes with a standalone driver since building NWChem is not easy for some people.

Jeff