Install Nvidia HPC C, C++, Fortran compilers

November 20, 2024

The free-to-use Nvidia HPC SDK offers possible speed improvements and CUDA Fortran. A typical reason for using Nvidia HPC SDK is the Cuda GPU features. Nvidia HPC compilers support C11, C++23, and partial Fortran 2008 including submodule and error stop.

Download and install Nvidia HPC SDK. Create a script nvidia.sh:

To use NVIDIA HPC SDK, source the script:

source ~/nvidia.sh

CMake

In CMake, set NVIDIA HPC compiler-specific options in CMakeLists.txt like:

if(CMAKE_Fortran_COMPILER_ID STREQUAL "NVHPC")
  add_compile_options($<$<COMPILE_LANGUAGE:Fortran>:-Mdclchk;-Munixlogical>)
endif()

To use newer languages standard features ensure the underlying GCC toolchain is set to a new-enough compiler as per Nvidia HPC SDK documentation. The compiler path can be determined on RHEL-like Linux distros like:

scl enable gcc-toolset-12 "dirname $(which g++)"

If using a CMake toolchain file, instead of CXXFLAGS environment variable, one can set

set(CMAKE_CXX_COMPILER_EXTERNAL_TOOLCHAIN "/opt/rh/gcc-toolset-12/root/usr/")