Remote OpenMP Offloading Paper recognized by ISC

0

June 17, 2022 – Computer scientists working with the Exascale Computing Project (ECP) to advance the LLVM compiler infrastructure have demonstrated that OpenMP is an effective tool for remote accelerator offloading involving more than a single compute node. Their work aims to establish OpenMP as a single powerful parallel programming model that combines CPU parallelism, accelerator offloading, and distributed computing to reduce complexity and eliminate porting tasks. The team’s work was recognized with the Hans Meuer Prize for the most outstanding research work at the ISC High Performance 2022 conference, which took place in May. Their research was published in the conference proceedings.

The LLVM/Clang compiler, reflecting the scientists’ work, provides a rich, novel and production-ready OpenMP runtime system. Remote OpenMP offload allows users to leverage hardware in the cloud or on a compute cluster as if it were local to their computer and provides improved debugging tools. Users can run their programs wherever the hardware is available while using their own machine’s files and CPU resources. The work also runs paged code in a separate process, which can help users identify disk space issues.

Scientists experimented with scaling OpenMP to 120 GPUs and uncovered limitations affecting future work. Plans include exploring the use of the UCX framework’s Active Messaging API for more efficient use of network resources and using data compression to improve the overall performance of remote OpenMP offloading. The team is also working on additional prototype extensions to make remote OpenMP offloading more convenient and efficient.

Atmn Patel and Johannes Doerfert, “Remote OpenMP Offloading”. 2022. Proceedings of ISC High Performance 2022: High Performance Computing (May), https://doi.org/10.1007/978-3-031-07312-0_16.

Scientists collaborating with ECP aim to establish OpenMP as a single, powerful, parallel programming model that combines CPU parallelism, accelerator offloading, and distributed computing to reduce complexity and eliminate porting tasks.

Source: Exascale Computing Project

Share.

Comments are closed.