Fast and efficient searches for effective optimization-phase sequences

Kulkarni, Prasad A.; Hines, Stephen; Whalley, David; Hiser, Jason D.; Davidson, Jack W.; Jones, Douglas L.

doi:10.1145/1071604.1071607

Cited by 38 publications

(28 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Identifying the best combination of optimization flags to activate is difficult. Iterative compilation techniques search the space of optimizations to optimize metrics, such as performance, energy, and code size [Cooper et al 1999;Kulkarni et al 2005;Triantafyllis et al 2005]. Statistical machine learning further improves search efficiency [Cavazos and O'Boyle 2006].…”

Section: Discussionmentioning

confidence: 99%

Applied inference

Lee

Brooks

2010

ACM Trans. Archit. Code Optim.

View full text Add to dashboard Cite

We propose and apply a new simulation paradigm for microarchitectural design evaluation and optimization. This paradigm enables more comprehensive design studies by combining spatial sampling and statistical inference. Specifically, this paradigm (1) defines a large, comprehensive design space, (2) samples points from the space for simulation, and (3) constructs regression models based on sparse simulations. This approach greatly improves the computational efficiency of microarchitectural simulation and enables new capabilities in design space exploration.We illustrate new capabilities in three case studies for a large design space of approximately 260,000 points: (1) Pareto frontier, (2) pipeline depth, and (3) multiprocessor heterogeneity analyses. In particular, regression models are exhaustively evaluated to identify Pareto optimal designs that maximize performance for given power budgets. These models enable pipeline depth studies in which all parameters vary simultaneously with depth, thereby more effectively revealing interactions with non-depth parameters. Heterogeneity analysis combines regression based optimization with clustering heuristics to identify efficient design compromises between similar optimal architectures. These compromises are potential core designs in a heterogeneous multicore architecture. Increasing heterogeneity can improve bips 3 /w efficiency by as much as 2.4x, a theoretical upper bound on heterogeneity benefits that neglects contention between shared resources as well as design complexity. Collectively these studies demonstrate regression models' ability to expose trends and identify optima in diverse design regions, motivating the application of such models in statistical inference for more effective use of modern simulator infrastructure.

show abstract

Section: Discussionmentioning

confidence: 99%

Applied inference

Lee

Brooks

2010

ACM Trans. Archit. Code Optim.

View full text Add to dashboard Cite

show abstract

“…Agakov et al characterized programs using static features and developed adaptive mechanisms using statistical correlation models to focus their iterative search to those areas of the phase ordering space most likely to provide the greatest performance benefit [1]. A separate study by Kulkarni et al employed several innovative search space pruning techniques to avoid over 84% of program executions, thereby considerably reducing the search overhead [20,23]. Fursin et al devised a novel strategy to evaluate the relative benefit of multiple per-function phase sequences during a single program execution by maintaining different versions of the same function in one executable and switching between them during execution [10].…”

Section: Related Workmentioning

confidence: 99%

Improving both the performance benefits and speed of optimization phase sequence searches

Kulkarni

Jantz

Whalley

2010

Proceedings of the ACM SIGPLAN/SIGBED 2010 Conference on Languages, Compilers, and Tools for Embedded Systems

Self Cite

View full text Add to dashboard Cite

The issues of compiler optimization phase ordering and selection present important challenges to compiler developers in several domains, and in particular to the speed, code size, power, and costconstrained domain of embedded systems. Different sequences of optimization phases have been observed to provide the best performance for different applications. Compiler writers and embedded systems developers have recently addressed this problem by conducting iterative empirical searches using machine-learning based heuristic algorithms in an attempt to find the phase sequences that are most effective for each application. Such searches are generally performed at the program level, although a few studies have been performed at the function level. The finer granularity of functionlevel searches has the potential to provide greater overall performance benefits, but only at the cost of slower searches caused by a greater number of performance evaluations that often require expensive program simulations. In this paper, we evaluate the performance benefits and search time increases of functionlevel approaches as compared to their program-level counterparts. We, then, present a novel search algorithm that conducts distinct function-level searches simultaneously, but requires only a single program simulation for evaluating the performance of potentially unique sequences for each function. Thus, our new hybrid search strategy provides the enhanced performance benefits of functionlevel searches with a search-time cost that is comparable to or less than program-level searches.

show abstract

“…In other attempts to reduce the cost of iterative optimizations, researchers have used predictive modeling and code context information to focus search on the most fruitful areas of the phase order space for the program being compiled for static compilers [Agakov et al 2006] as well as for dynamic compilers . In our past research, we used genetic algorithms with aggressive pruning of the search space [Kulkarni et al 2004;Kulkarni et al 2005] to make searches for effective optimization phase sequences faster and more efficient. During this work we realized that a significant portion of typical phase order search spaces is redundant because many different orderings of optimization phases produce the same code.…”

Section: Related Workmentioning

confidence: 99%

“…From previous studies we have realized that it is possible for different function instances to be identical except for register numbers used in the instructions [Kulkarni et al 2004;Kulkarni et al 2005]. This situation can occur since different optimization phases compete for registers.…”

Section: Detecting Equivalent Function Instancesmentioning

confidence: 99%

Practical exhaustive optimization phase order exploration and evaluation

Kulkarni

Whalley

Tyson

et al. 2009

ACM Trans. Archit. Code Optim.

Self Cite

View full text Add to dashboard Cite

Choosing the most appropriate optimization phase ordering has been a long standing problem in compiler optimizations. Exhaustive evaluation of all possible orderings of optimization phases for each function is generally dismissed as infeasible for production-quality compilers targeting accepted benchmarks. In this paper we show that it is possible to exhaustively evaluate the optimization phase order space for each function in a reasonable amount of time for most of the functions in our benchmark suite. To achieve this goal we used various techniques to significantly prune the optimization phase order search space so that it can be inexpensively enumerated in most cases, and to reduce the number of program simulations required to evaluate program performance for each distinct phase ordering. The techniques described are applicable to other compilers in which it is desirable to find the best phase ordering for most functions in a reasonable amount of time. We also describe some interesting properties of the optimization phase order space, which will prove useful for further studies of related problems in compilers. (1) We introduce a completely new search algorithm (Section 5) to more efficiently find the successive phase order sequences to evaluate. We also note the trade-offs in choosing any one algorithm over the other. (2) We have doubled our benchmark set from 6 to 12 benchmarks, and more than doubled the number of studied functions, from 111 to 244. Many of the newer functions added are significantly larger, making our switch to the new search algorithm more critical. (3) A new section (Section 8) presents interesting results from analyzing the exhaustive phase order space over the entire set of functions. These results, which are shown in Figures 11 through 19, required a significant amount of time to collect. (4) All the remaining sections in this paper also go into more depth about the issues involved in the development of the search algorithm, pruning techniques, and other implementation issues.

show abstract

Fast and efficient searches for effective optimization-phase sequences

Cited by 38 publications

References 20 publications

Applied inference

Applied inference

Improving both the performance benefits and speed of optimization phase sequence searches

Practical exhaustive optimization phase order exploration and evaluation

Contact Info

Product

Resources

About