X-MAT: Exascale Mapping Tool

The hardware architecture of the modern supercomputers and future exascale systems is being designed to include novel custom and/or specialized processors to accelerate their exascale systems, resulting in increasingly heterogeneous and complex platforms- Dynamic task-based programming models and runtimes are emerging as dominant paradigms to address this complexity and exploit the computing power available in these heterogeneous environments- These programming paradigms are however complex, impose verbose expressions of algorithms, and demand expensive and rare expertise- The focus of this project will be to develop a key exascale software stack component in the form of a compiler-based optimization tool that helps the user to attain performance, programmability, and portability, and close critical gaps in the cooperative functioning of exascale hardware and software- Our approach will be to harden and extend an existing HPC compiler tool that can automatically parallelize and optimize programs and generate code for task-based runtime models- Specifically, we will extend the compiler to generate code for more task-based runtimes that are emerging from DOE-funded and non-DOE-funded efforts and support highly heterogeneous platforms, which is the evolving hardware trend for exascale systems-