Implementing Fast Barriers for a Shared-Memory Cluster of 1024 RISC-V Cores

Researchers from ETH Zürich and Università di Bologna have published a technical paper titled “Fast Shared-Memory Barrier Synchronization for a 1024-Cores RISC-V Many-Core Cluster.” The paper addresses the issue of synchronization in shared-memory parallel programs and focuses specifically on barrier synchronization for TeraPool, a cluster of 1024 RISC-V processors. The researchers compare different synchronization strategies and benchmark optimized barrier implementations in the Open-MP programming model. The results show that the optimized tree barriers achieve a 1.6x speed-up compared to a naive central counter barrier and only a 6.2% overhead on a 5G application. This work is significant as it is the first to use shared-memory barriers for the synchronization of a thousand processing elements tightly coupled to shared data memory.

To top