Managing performance vs. accuracy trade-offs with loop perforation

Abstract

Many modern computations (such as video and audio encoders, Monte Carlo simulations, and machine learning algorithms) are designed to trade off accuracy in return for increased performance. To date, such computations typically use ad-hoc, domain-specific techniques developed specifically for the computation at hand.Loop perforation provides a general technique to trade accuracy for performance by transforming loops to execute a subset of their iterations. A criticality testing phase filters out critical loops (whose perforation produces unacceptable behavior) to identify tunable loops (whose perforation produces more efficient and still acceptably accurate computations). A perforation space exploration algorithm perforates combinations of tunable loops to find Pareto-optimal perforation policies. Our results indicate that, for a range of applications, this approach typically delivers performance increases of over a factor of two (and up to a factor of seven) while changing the result that the application produces by less than 10%.

Load Value Approximation

“…Profiling tools can be use to determine where the bulk of cache misses occurs. We use profiling and previous work [39] with PARSEC, to find frequently visited regions of code and then determine if the data being used can be approximated.…”

Automated Transplantation of Call Graph and Layout Features into Kate

“…Our work is closely related to recent achievements in genetic improvement, which have been able to dramatically speed up real world systems [9,14,15], port between languages [8], balance memory consumption and execution time [16], reduced energy consumption [3,13] and fix bugs [10]. Most closely related to our approach is work on auto-specialisation using transplantation [11] and grow and graft genetic improvement [6].…”

DIVERSIFY: Ecology-inspired software evolution for diversity emergence

“…R. Feldt [9] used genetic programming to automatically synthesize variants of an aircraft controller in order to achieve failure diversity. M. Rinard and colleagues [28], [26] have developed unsound program transformations that support the runtime production of diversity and handle changes in quality of service. Forrest and colleagues have explored genetic programming for automatic bug fixing [15] and neutral mutation [27].…”