A Task-Centric Memory Model for Scalable Accelerator Architectures

Abstract

This article presents a memory model for parallel compute accelerators with task-based programming models that uses a software protocol, working in collaboration with hardware caches, to maintain a coherent, single address space view of memory without requiring hardware cache coherence. The memory model supports visual computing applications, which are becoming an important class of workloads capable of exploiting 1,000-core processors.