June 17, 2022 – Computer scientists working with the Exascale Computing Project (ECP) to advance the LLVM compiler infrastructure have shown that OpenMP is an effective tool for remote accelerator offloading with more than a single compute node . Their work aims to establish OpenMP as a unique high-performance portable parallel programming model combining CPU parallelism, accelerator offloading, and distributed computing to reduce complexity and eliminate porting tasks. The team’s paper received the Hans Meuer Award for Outstanding Research Paper at the 2022 ISC High Performance Conference, held in May. Their research was published in the conference proceedings.
The LLVM/Clang compiler, which mirrors the work of the scientists, provides a complete, new, production-ready OpenMP runtime system. Remote OpenMP offload allows users to use hardware in the cloud or on a compute cluster as if it were local to their machine and provides enhanced debugging tools. Users can run their programs wherever hardware is available while using their own machine’s files and CPU resources. The job also runs code offloaded in a separate process, which can help users identify memory placement issues.
Scientists experimented with scaling OpenMP to 120 GPUs, revealing limitations that inform future work. Plans include exploring the use of the UCX Framework Active Messaging API for more efficient use of network resources and the use of data compression to improve the overall performance of remote OpenMP offloading. The team is also working on additional prototype extensions to make OpenMP remote 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.
Source: Exascale Calculation Project
Comments are closed.