Work Assisting - Linking Task-Parallel Work Stealing with Data-Parallel Self Scheduling
Explore a novel scheduling strategy called work assisting in this 25-minute conference talk from ARRAY 2024. Discover how this approach combines data parallelism with task parallelism, allowing threads to share their current data-parallel activity in a shared array for assistance from other threads. Learn about the advantages of preserving data parallelism structure instead of implementing all parallelism as task parallelism, enabling the use of self-scheduling for data-parallel algorithms. Understand how this flexible scheduling algorithm adapts to different scenarios, potentially outperforming schedulers based purely on task parallelism. Examine benchmarks demonstrating the performance of this innovative approach in various problem domains.