Do inputs matter?: using data-dependence profiling to evaluate thread level speculation in BG/Q

Authors:
Arnamoy Bhattacharyya
Affiliations:
University of Alberta, Edmonton, AB, Canada
Venue:
PACT '13 Proceedings of the 22nd international conference on Parallel architectures and compilation techniques
Year:
2013

Citing 2
Cited 0

Towards a holistic approach to auto-parallelization: integrating profile-driven parallelism detection and machine-learning based mapping

Proceedings of the 2009 ACM SIGPLAN conference on Programming language design and implementation
Combined profiling: A methodology to capture varied program behavior across multiple inputs

ISPASS '12 Proceedings of the 2012 IEEE International Symposium on Performance Analysis of Systems & Software

Quantified Score

Hi-index	0.00

Visualization

Abstract

Thread level speculation (TLS) is a hardware/software technique that guarantees correct execution of a loop even in the presence of a dependence. To reduce mispeculation overhead, data-dependence profiling is used to find out whether the may dependences materialize during runtime. Based on the probability of dependence, a cost model can be used to select candidate loops for speculative execution. But a single input profile is not sufficient to capture the dependence behaviour of a loop because Berube et al. showed that programs' behaviour may change based on input. Though previous work mentions that there is little variability in the dependence behaviour of loops based on inputs [1], there has not been an extensive study to support the claim.