ECO: an empirical-based compilation and optimization system

Baradaran, Nastaran; Chame, Jacqueline; Chen, Chun; Diniz, Pedro C.; Hall, Mary; Lee, Yoon-Ju; Liu, Bing; Lucas, Robert F.

doi:10.1109/ipdps.2003.1213377

Cited by 6 publications

(13 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…More recent research on iterative compilation has empirically modified the configurations of general-purpose compiler optimizations based on performance feedbacks [9], [16], [1], [14], [13]. None of these existing compilers support programmable control of optimizations by developers.…”

Section: Related Workmentioning

confidence: 99%

“…The auto-generated POET transformations are extensively parameterized so that each optimization can be turned on or off independently for each relevant array or code region, and arbitrary integers can be given as the blocking or unrolling factor for each loop being transformed. The granularity of external control is far beyond those supported by existing iterative compilation frameworks [9], [16], [1], [14], [13]. Independent search engines can be substituted with ease, and developers can easily interfere by modifying the auto-generated POET scripts.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Automated programmable control and parameterization of compiler optimizations

Yi¹

2011

International Symposium on Code Generation and Optimization (CGO 2011)

View full text Add to dashboard Cite

Abstract-We present a framework which effectively combines programmable control by developers, advanced optimization by compilers, and flexible parameterization of optimizations to achieve portable high performance. We have extended ROSE, a C/C++/Fortran source-to-source optimizing compiler, to automatically analyze scientific applications and discover optimization opportunities. Instead of directly generating optimized code, our optimizer produces parameterized scripts in POET, an interpreted program transformation language, so that developers can freely modify the optimization decisions by the compiler and add their own domain-specific optimizations if necessary. The auto-generated POET scripts support extra optimizations beyond those available in the ROSE optimizer. Additionally, all the optimizations are parameterized at an extremely fine granularity, so the scripts can be ported together with their input code and automatically tuned for different architectures. Our results show that this approach is highly effective, and the code optimized by the auto-generated POET scripts can significantly outperform those optimized using the ROSE optimizer alone. I. INTRODUCTIONCompiler optimization is critical for scientific applications to automatically achieve high performance on modern computers. However, due to the excessive complexity, compilers have lagged behind in precisely modeling the behaviors of applications running on diverse architectures. While computational specialists can adopt low-level programing in C or assembly to directly manage machine resources and can parameterize their algorithm implementations to accommodate architectural diversity [21], [2], [20], [7], [15], this approach is extremely error prone and time consuming, and compiler technology needs to be developed to automate the process.We present a framework, shown in Fig. 1, to reach a balance between direct control of resources by programmers and automation of optimization by compilers. This framework starts with a specialized source-to-source optimizing compiler (the ROSE analysis engine) which interacts with developers to automatically discover applicable optimizations and then produces output in a transformation scripting language, POET [25]. Developers can modify the auto-generated POET scripts as well as writing new POET transformations to directly control the optimization of applications. The POET output can then be ported together with an annotated input program to different machines, where an empirical transformation engine can dynamically interpret the POET scripts with different optimization configurations until satisfactory performance is achieved for the input code.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Automated programmable control and parameterization of compiler optimizations

Yi¹

2011

International Symposium on Code Generation and Optimization (CGO 2011)

View full text Add to dashboard Cite

show abstract

“…We have implemented the code isolator in the Stanford SUIF compiler as part of the ECO project, (3,8) with some modest manual intervention, and we describe its features in the remainder of this section. We first describe how to generate isolated code that can be compiled, followed by code that can be executed and subsequently describe how to initialize machine state of the isolated code.…”

Section: Isolating Code Fragmentsmentioning

confidence: 99%

“…Once we consider multiple levels of the memory hierarchy, our preliminary experience suggests enough constraints can be derived on variants and parameters to limit the search to something manageable, a subject of current and future work. (3,4) For algorithm variants, user guidance is required, both to implement the different variants, and to indicate that the two are equivalent. A linguistic mechanism called a selector for specifying variants and dynamic criteria for selection of variants is described elsewhere.…”

Section: Generating and Searching Variantsmentioning

confidence: 99%

“…This work is being performed in the context of the ECO project, which is based on the notion of model-guided empirical optimization, whereby code variants are generated and executed with representative input data sets so that performance can be measured and compare. (3,4) We begin by describing LS-DYNA's solver in more detail in Section 2. Section 3 describes a code isolator tool, which generates executable sub-programs from large applications, preserving the behavior of the sub-program when run in the context of the full program.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Empirical Optimization for a Sparse Linear Solver: A Case Study

Lee

Diniz

Hall

et al. 2005

Int J Parallel Prog

View full text Add to dashboard Cite

This paper describes initial experiences with semi-automated performance tuning of a sparse linear solver in LS-DYNA, a large, widely used engineering application. Through a collection of tools supporting empirical optimization, we alleviate the burden of performance tuning for mapping today's sophisticated engineering software to increasingly complex hardware platforms. We describe a tool that automatically isolates code segments to create benchmark subsets for the purposes of performance tuning. We present a collection of automatically generated empirical results that demonstrate the sensitivity of the application's performance to optimization parameters. Through this case study, we demonstrate the importance of developing automatic performance tuning support for performance-sensitive applications.

show abstract

A Code Isolator: Isolating Code Fragments from Large Programs

Lee

Hall

2005

Lecture Notes in Computer Science

View full text Add to dashboard Cite

ECO: an empirical-based compilation and optimization system

Cited by 6 publications

References 46 publications

Automated programmable control and parameterization of compiler optimizations

Automated programmable control and parameterization of compiler optimizations

Empirical Optimization for a Sparse Linear Solver: A Case Study

A Code Isolator: Isolating Code Fragments from Large Programs

Contact Info

Product

Resources

About