Fast, automatic, procedure-level performance tuning

Abstract

This paper presents an automated performance tuning solution, which partitions a program into a number of tuning sections and finds the best combination of compiler options for each section. Our solution builds on prior work on feedback-driven optimization, which tuned the whole program, instead of each section. Our key novel algorithm partitions a program into appropriate tuning sections. We also present the architecture of a system that automates the tuning process; it includes several pre-tuning steps that partition and instrument the program, followed by the actual tuning and the post-tuning assembly of the individuallyoptimized parts. Our system, called PEAK, achieves fast tuning speed by measuring a small number of invocations of each code section, instead of the whole-program execution time, as in common solutions. Compared to these solutions PEAK reduces tuning time from 2.19 hours to 5.85 minutes on average, while achieving similar program performance. PEAK improves the performance of SPEC CPU2000 FP benchmarks by 12% on average over GCC O3, the highest optimization level, on a Pentium IV machine.

Hardware Performance Monitoring for the Rest of Us: A Position and Survey

“…An increasingly popular solution is to simply guess a lot and choose the best solution [8][9][10][11], though the guessing process (also known as search or learning) can take a long time, and defining the parameters to use may still require an understanding of the underlying microarchitecture.…”

Fast compiler optimisation evaluation using code-feature based performance prediction

“…All of these approaches are successful in automatically generating compiler heuristics for code segments rather than in predicting the eventual performance of the selected optimisations for whole programs. Rather than predicting the impact of a single transformation, others have looked at searching [2,6,9,10,11,16,19,22] for the best set or sequence of optimisations for a particular program. Cooper et al [6] propose a number of algorithms to solve the compilation phase ordering problem.…”

“…Pan et al [19] partitioned a program into tuning sections and then developed fast techniques to find the best combination of optimisations for each of these tuning section. They are able to reduce the time to find good optimisation settings from hours to minutes.…”

“…There has been much work in the area of iterative optimisation aimed at automating this process [7,9,10,11,16,19,22]. Such approaches focus on choosing good program modifications or transformations so that the number of modify-compile-execute cycles is reduced.…”

Automatically Tuning Parallel and Parallelized Programs

“…An important un-met need is to customize compiler techniques on a section-by-section basis. Typical tuning systems choose a set of compilation flags, which are then applied to the entire program, even though some approaches have tuned individual subroutines [8]. Finer-grain tuning would allow many additional techniques and parameters to be made amenable to runtime tuning; for example, the most suitable register allocation method could be selected for each basic block.…”