The goal of this project is to conduct research on improving parallel programming models and languages for performance, portability and correctness. The team's activities will include:

• Tooling for parallel programming models, including Level Zero, CUDA, HIP, OpenMP and MPI using the THAPI framework
• Extending the GCC and LLVM compiler for OpenMP
• Implementing a distributed-shared memory system relying on MPI with a focus on multi-GPU programs

The activities will be about equally split between development (i.e., programming on single core, compilation on multiple cores, and testing on multiple cores and GPUs) and evaluations (i.e., running jobs to generate scientific results on multiple cores + GPUs).