Publication Details
Polykarpos Thomadakis and Nikos Chrisochoides.
Published in The 28th International Conference on Parallel & Distributed Processing Techniques & Applications (PDPTA'22), March, 2022
Abstract
We present performance and ease-of-use improvements to a runtime for Domain-Specific Languages of irregular applications. Support for message-driven global address space is integrated with lightweight threads, and new abstractions for mapping high- level constructs are presented. Introducing features to handle control-flow and fine-grained concurrency improve load balancing in shared and distributed memory. We observe up to 100% difference in performance behavior of options for task creation and handling message-passing. Evaluations on an 1000- core distributed memory machine showed that the integration of fine-grained concurrency with the runtime achieves performance improvements of 12% on a seismic wave simulation benchmark, as opposed to 50% degradation with OpenMP. Moreover, a 3D mesh refinement application showed 50% improvement, exploiting multi-grain parallelism at domain and task level.