RAJA: Portable Performance for Large-Scale Scientific Applications

Modern high-performance computing systems are diverse, with hardware designs ranging from homogeneous mult-icore CPUs to GPU or FPGA accelerated systems. Achieving desirable application performance often requires choosing a programming model best suited to a particular platform. For large codes used daily in production that are under continual development, architecture-specific ports are untenable. Maintainability requires single-source application code that is performance portable across a range of architectures and programming models. In this paper we describe RAJA, a portability layer that enables C++ applications to leverage various programming models, and thus architectures, with a single-source codebase. We describe preliminary results using RAJA in three large production codes at Lawrence Livermore National Laboratory, observing 17x, 13x and 12x speedup on GPU-only over CPUonly nodes with single-source application code in each case.