Skip to content

Argonne Leadership
Computing Facility



Kokkos Core implements a programming model in C++ for writing performance portable applications targeting all major HPC platforms. For that purpose it provides abstractions for both parallel execution of code and data management. Kokkos is designed to target complex node architectures with N-level memory hierarchies and multiple types of execution resources. It currently can use Serial and OpenMP (threads) for CPU execution spaces ("backends") and CUDA, HIP, SYCL, and OpenMPTarget for GPU execution spaces. By convention, Kokkos only allows one GPU backend at a time.

Kokkos Documentation

Kokkos on Polaris

The prebuilt Kokkos on polaris includes 3 backends: Serial and OpenMP for CPU execution and CUDA for GPU execution. To use it, run

module use /soft/modulefiles
module load kokkos

Configuring Your Own Kokkos Build on Polaris

Here are recommended environment settings and configuration to build your own kokkos libraries on Polaris:


To match what was done in the centrally-built kokkos associated with the modules discussed above, use the default programming environment PrgEnv-nvhpc, and use the Cray wrapper CC as the C++ compiler. To build Kokkos, you'll need cmake. You may also use PrgEnv-gnu to build kokkos (also using the Cray wrapper CC as the C++ compiler).

To use C++17, you'll need to work around a bug with the current PrgEnv-nvhpc/8.3.3 environment by loading a cudatoolkit-standalone module:

module load cmake cudatoolkit-standalone/11.6.2

CMake Configuration

This example builds three backends: OpenMP, Serial, and Cuda.

git clone
cd kokkos
mkdir build
cd build


make -j16 -l16 install