Using Static Code Features for Intelligent Squash Prediction

Thread-level speculation (TLS) is widely accepted to be a realistic model for execution of sequential pro- grams on multi-core architectures. One problem with TLS occurs when large numbers of spawned threads are squashed due to data dependence violations. This can reduce or entirely obliterate the performance ben- efits of TLS. Until now, informal compiler heuristics or high-overhead runtime profiling have been used to identify likely squashes and suppress thread spawns at the appropriate program locations. In this paper, we adopt an alternative approach using machine learning to construct squash predictors based on static code features. On a set of standard Java benchmarks, we achieve a mean speedup that is 82% of the maximum speedup available from an oracle predictor.