Project Description
This project is a direct extension to some earlier work on performance portability, where we’ve demonstrated the ability to run the same codes on AMD, Intel and NVIDIA hardware using respective third-party libraries with SYCL interoperability. The goal is to implement existing open-source RNG and FFT libraries into SYCL directly, as opposed to using interoperability. This is to investigate both performance portability and reproducibility in scientific codes across architectures and platforms. Adding ARM into the mix would be impactful. Moreover, this work will serve as a means for continued support of ARM and NVIDIA in the open-source Intel DPC++ (intel/llvm).
Testbed
Fujitsu A64FX, NVIDIA ARM Dev Kit