Analysis of classic algorithms on highly-threaded many-core architectures

Lin Ma, Roger D. Chamberlain, Kunal Agrawal, Chen Tian, Ziang Hu

Abstract: Highlights•Analyze effect of memory latency hiding by threads and determine the algorithm bound.•A wide range of algorithms analyzed on a wide spectrum of architectures including both NVIDIA and AMD GPUs and XMT machines.•Analysis is accurate compared with our and other researchers’ experimental findings.•Predict important, non-trivial, and previously unexplained trends and artifacts in empirical data.•Verify the TMM model is effective at predicting effect of changing various parameters on diversified many-core machines.