RAJA is a collection of C++ software abstractions, being developed at Lawrence Livermore National Laboratory (LLNL), that enable architecture portability for HPC applications. The overarching goals of RAJA are to:
- Make existing (production) applications portable with minimal disruption
- Provide a model for new applications so that they are portable from inception.
RAJA targets portable, parallel loop execution by providing building blocks that extend the generally-accepted parallel for idiom.
Additional information can be found at RAJA User Guide.
RAJA provides a project template for how to use RAJA in an application project that uses CMake or Make. This is located at RAJA Project Template.
How to get the source code
The RAJA source code lives at RAJA github.
It can be cloned with
git clone --recursive https://github.com/llnl/raja.git. The recursive clone will also clone RAJA's dependencies in the proper locations.
Building on Theta
RAJA requires a compiler with C++11 support and CMake version 3.9 or greater RAJA includes an example build script for Theta that you can use. A quick example is below.
module load cmake/3.14.5 cd raja cp scripts/alcf-builds/theta_intel18.sh . ./theta_intel18.sh cd build_alcf-theta-intel18.0/ make -j 4
The build script makes use of a CMake configuration file
raja/host-configs/alcf-builds/theta_intel18_0.cmake. The RAJA and compiler options can be adjusted in this configuration file.