The award honors the most outstanding research paper submitted to the conference’s Research Papers Committee. Their paper, titled “Remote OpenMP Offloading,” showcases how an OpenMP offloading-capable program can run on a remote CPU or remote accelerator (e.g., a graphics processing unit, or GPU) as if it were a local one; and, for applications that support multi-device offloading, it can use any combination of local and remote CPUs and accelerators simultaneously.
The researchers evaluated their work both on microbenchmarks and on high-performance computing proxy applications. The results show scaling across dozens of GPUs in multiple hosts.
“Unlike solutions such as MPI, our work demonstrates that OpenMP can efficiently use a distributed heterogeneous system – without source, compiler, or language modifications,” said Doerfert, an assistant computer scientist in Argonne’s Mathematics and Computer Science Division.
The authors of the award-winning paper presented a keynote talk on their work in a special session May 30, 2022. They received a cash prize of 3,000 Euros, along with an award certificate.
For the full paper, see Patel, A., Doerfert, J. (2022). Remote OpenMP Offloading. In: Varbanescu, AL., Bhatele, A., Luszczek, P., Marc, B. (eds) High Performance Computing. ISC High Performance 2022. Lecture Notes in Computer Science, vol. 13289. Springer, Cham.