OpenMP on ThetaGPU

Help Desk

Theta GPU Nodes

OpenMP threading on CPU

All the compilers available on ThetaGPU supports OpenMP threading.

OpenMP offload on A100 GPU

A few compilers which support OpenMP offload are accessible on ThetaGPU. They are made available via modules.

$ module avail llvm

--------- /lus/theta-fs0/software/environment/thetagpu/lmod/modulefiles ----------
   llvm/main-20210112    llvm/release-12.0.0 (D)

$ module avail nvhpc

--------- /lus/theta-fs0/software/environment/thetagpu/lmod/modulefiles ----------
   nvhpc-byo-compiler/20.9 (D)    nvhpc-nompi/20.9 (D)    nvhpc/20.9 (D)
   nvhpc-byo-compiler/21.2        nvhpc-nompi/21.2        nvhpc/21.2
   nvhpc-byo-compiler/21.3        nvhpc-nompi/21.3        nvhpc/21.3

 

LLVM Clang for C/C++

Clang OpenMP offload features More details about the OpenMP runtime

If there is an issue with the compiler, feel free to contact openmp-dev@lists.llvm.org

Warning message

clang-12: warning: Unknown CUDA version. version.txt: 11.0.205. Assuming the latest supported version 10.1 [-Wunknown-cuda-version]

means CUDA 11 language features are not supported. As long as these features are not used, the full CUDA 11.x toolchain works. This warning can be ignored or suppressed by -Wno-unknown-cuda-version compiler option.

Compiling example:

module load llvm/release-12.0.0

clang++ -fopenmp -fopenmp-targets=nvptx64 your_source.cpp

NVIDIA HPC SDK for C/C++/Fortran

OpenMP documentation

For compiler bugs, please file bug reports at https://developer.nvidia.com after login

Compiling example:

module load nvhpc-sdk/nvhpc/21.3

nvfortran -mp=gpu -gpu=cc80 your_source.f90

Systems
Topics