LRSCwait: Enabling Scalable and Efficient Synchronization in Manycore Systems

Explore a groundbreaking approach to synchronization in manycore systems through this conference talk presented at the Design, Automation and Test in Europe Conference (DATE 2024). Delve into the innovative LRSCwait technique, which addresses the challenges of extensive polling in shared-memory architectures. Discover how the proposed LRwait and SCwait synchronization pair eliminates polling by allowing contending cores to sleep while waiting for atomic access. Learn about Colibri, a scalable implementation of LRwait that manages reservations in a distributed manner. Examine the performance benefits of this approach through extensive benchmarking on a 256-core RISC-V platform, showcasing significant improvements in throughput, fairness, and energy efficiency compared to traditional LRSC-based implementations. Gain insights into the potential impact of this technology on the future of manycore system design and concurrent algorithm performance.