Wed 17 Jun 2015 09:40 - 10:05 at PLDI Main BLUE (Portland 254-255) - Performance Chair(s): Mary Hall

A daunting challenge faced by program performance autotuning is input sensitivity, where the best autotuned configuration may vary with different input sets. This paper presents a novel two-level input learning algorithm to tackle the challenge for an important class of autotuning problems, algorithmic autotuning. The new approach uses a two-level input clustering method to automatically refine input grouping, feature selection, and classifier construction. Its design solves a series of open issues that are particularly essential to algorithmic autotuning, including the enormous optimization space, complex influence by deep input features, high cost in feature extraction, and variable accuracy of algorithmic choices. Experimental results show that the new solution yields up to a 3x speedup over using a single configuration for all inputs, and a 34x speedup over a traditional one-level method for addressing input sensitivity in program optimizations.

Wed 17 Jun

pldi2015-papers
09:15 - 10:55: Research Papers - Performance at PLDI Main BLUE (Portland 254-255)
Chair(s): Mary HallUniversity of Utah
pldi2015-papers09:15 - 09:40
Talk
Oswaldo Olivo, Isil DilligUniversity of Texas, Austin, Calvin LinUT Austin
Media Attached
pldi2015-papers09:40 - 10:05
Talk
Yufei DingNorth Carolina State University, Jason AnselMassachusetts Institute of Technology, Kalyan VeeramachaneniMassachusetts Institute of Technology, Xipeng ShenNorth Carolina State University, Una-May O’ReillyMassachusetts Institute of Technology, Saman AmarasingheMIT
Link to publication Media Attached
pldi2015-papers10:05 - 10:30
Talk
Charith MendisMIT CSAIL, Jeffrey BosboomMIT CSAIL, Kevin WuMIT CSAIL, Shoaib KamilMIT CSAIL, USA, Jonathan Ragan-KelleyStanford, Sylvain ParisAdobe, Qin ZhaoGoogle, Saman AmarasingheMIT
Media Attached
pldi2015-papers10:30 - 10:55
Talk
William J. BowmanNortheastern University, Swaha MillerCisco Systems, Inc, Vincent St-AmourNortheastern University, R. Kent DybvigCisco Systems, Inc
Link to publication Media Attached