Does HPC SDK support pure OpenMP?

Accelerated Computing HPC Compilers Legacy PGI Compilers
#1

Hello everyone, I knew that NVIDIA HPC SDK shipped with the support of MPI, so I am wondering does it shipped to support pure OpenMP too.
Through my search I found two commands used to run OpenMP code on tesla and multi-core

pgcc -omp -ta=tesla -Minfo=accel -fast test.c
pgcc -Minfo -ta=multicore

but can I run pure OpenMP on multi-core with using acc accelerator and use only OpenMP
when I tried the commands above they gave me huge difference than running it using GCC compile with OpenMP flag, so does HPC SDK support pure OpenMP on multi-core CPU?

Thank you in advance.

#2

Hi hisham.M,

Yes, the NV HPC Compilers support OpenMP. The 20.11 release also includes Beta support for OpenMP Target directive targeting NVIDIA GPUs.

To enable OpenMP, please use the flag “-mp”, with “-mp=gpu” used to target the GPU.

Full documentation on our OpenMP support can be found at: https://docs.nvidia.com/hpc-sdk/compilers/hpc-compilers-user-guide/index.html#openmp-use

Hope this helps,
Mat

#3

Thank you, I already downloaded the 2.11 version but I am not trying to target NVIDIA GPU, can I compile pure OpenMP using HPC SDK 2.11 to compile OpenMP normally on CPU for example if I only used

#pragma omp parallel for

Thank you

#4

Yes, of course. We’ve supported OpenMP on the host for about 20 years. Again, just use the “-mp” flag to enable it.

#5

can you please give me the whole command that I use with c/c++ code.

#6

With the NV HPC Compiler Drivers:

C: nvc -mp -fast -Minfo=mp file.c
C++:: nvc++ -mp -fast -Minfo=mp file.cpp

With the old PGI Compiler drivers:

C: pgcc -mp -fast -Minfo=mp file.c
C++: pgc++ -mp -fast -Minfo=mp file.cpp