SC22
This year’s “The International Conference for High Performance Computing, Networking, Storage, and Analysis” (SC22) takes place in Dallas, TX from 11/13 - 11/18. As part of SC, we’ll present a research poster on Tensor Processing Primitives (TPPs) in the computational sciences. Our work formulates key kernels of a demanding solver in terms of TPPs. As shown in a thorough performance analysis, this approach leads to performance portability enabled by the TPP-backend LIBXSMM. In particular, the poster presents competitive application performance when running on a wide range of server processors: Intel Cascade Lake, Intel Ice Lake, Intel Sapphire Rapids, AMD Rome, AMD Milan, Amazon Graviton2, Fujitsu A64FX and Amazon Graviton3.