Efficient GPU synchronization without scopes: Saying no to complex consistency models

As GPUs have become increasingly general purpose, applications with more general sharing patterns and finegrained synchronization have started to emerge. Unfortunately, conventional GPU coherence protocols are fairly simplistic, with heavyweight requirements for synchronization accesses. Prior work has tried to resolve these inefficiencies by adding scoped synchronization to conventional GPU coherence protocols, but the resulting memory consistency model, heterogeneous-race-free (HRF), is more complex than the common data-race-free (DRF) model. This work applies the DeNovo coherence protocol to GPUs and compares it with conventional GPU coherence under the DRF and HRF consistency models. The results show that the complexity of the HRF model is neither necessary nor sufficient to obtain high performance. DeNovo with DRF provides a sweet spot in performance, energy, overhead, and memory consistency model complexity. Specifically, for benchmarks with globally scoped fine grained synchronization, compared to conventional GPU with HRF (GPU+HRF), DeNovo+DRF provides 28% lower execution time and 51% lower energy on average. For benchmarks with mostly locally scoped fine-grained synchronization, GPU+HRF is slightly better — however, this advantage requires a more complex consistency model and is eliminated with a modest enhancement to DeNovo+DRF. Further, if HRF's complexity is deemed acceptable, then DeNovo+HRF is the best protocol.