文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > An Evaluation of Staged Run-Time Optimizations in DyC Abstract
侵权投诉

An Evaluation of Staged Run-Time Optimizations in DyC Abstract

Previous selective dynamic compilation systems have demonstrated that dynamic compilation can achieve performance improvements at low cost on small kernels, but they have had difficulty scaling to larger programs. To overcome this limitation, we developed

An Evaluation of Staged Run-Time Optimizations in DyC

Brian Grant, Matthai Philipose, Markus Mock, Craig Chambers, and Susan J. Eggers

Department of Computer Science and Engineering

University of Washington

Box 352350, Seattle WA 98195-2350

{grant,matthai,mock,chambers,eggers}@cs.washington.edu

Abstract

Previousthatatlargerlowdynamicselectivedynamiccompilationsystemshavedemonstratedcostoncompilationsmallkernels,canachieveperformanceimprovementsselectiveprograms.cateddynamicTocompilationovercomebutthistheyhavehaddif cultyscalingtosystemlimitation,thatwedevelopedDyC,aachieveand exibleanalysesandtransformations.includesDyCmoreissophisti-muchgoodperformanceimprovementsonprogramsthatablearetoindividuallargerandmorecomplexthanthekernels.Weanalyzethemance collectively and individually.

optimizationsofDyCandassesstheirimpactonperfor-1. Introduction

Selectivedynamiccompilationtransformsselectedpartsofprogramsatruntime,usinginformationavailableonlyatruntime,tooptimizethemmorefullythanstrictlystaticallycompiledcode.(Selectivedynamiccompilationisincontrasttocompletedynamiccompilationwhereallcompilationofaprogramisdelayeduntilrun-time;recent“justintime”compilersforJavaareexamplesofcompletedynamiccompilers.)Value-speci cselectivedynamiccompilersderivetheirbene tsbyoptimizingpartsofprogramsforparticularrun-time-computedvaluesofinvariantvariablesanddatastructures(calledrun-timeconstants),ineffectperformingakindofdynamicconstantpropagationandfolding.Proposedapplicationsforselective,value-speci cdynamiccompilationincludespecializingarchitecturalsimulatorsforthecon gurationbeingsimulated,languageinterpretersfortheprogrambeinginterpreted,renderingenginesforscene-speci cstatevariables,numericprogramsfordimensionsandvaluesoffrequentlyusedarrays,andcriticalpathsinoperatingsystemsforthetypeofdatabeingprocessedandthecurrentstateofthesystem.Trendsinsoftwareengineeringtowarddynamicrecon gurability,suchasparameterizationforreuseandportabilityacrossdifferenthardwarearchitectures, also imply a promising role for dynamic compilation.Recentresearcheffortshavemadeconsiderableprogresstowardsprovingtheviabilityofselectivedynamiccompilation.Inparticular,researchershavedemonstratedthatdynamiccompilationoverheadcanbequicklyamortizedbytheincreasedef ciencyofthedynamicallyoptimizedcode.Mostexperiments,however,werecon nedtosimplekernels,anddidnotdemonstratethatthedynamiccompilationsystemscouldcopereasonablywiththeincreasedsizeandcomplexityofapplicationsliketheinterpretersand simulators mentioned above.

Thereasonscurrentsystemshavenotmadebetterprogressonlarger,morecomplexapplicationsvary,dependingontheapproach

taken.Inimperativesystems,suchas`C[12,27,28],aprogrammerexplicitlyconstructs,composes,andcompilescodefragmentsatruntime.Imperativeapproachescanexpressawiderangeofoptimizations,butimposealargeburdenontheprogrammertomanuallyprogramtheoptimizations;theprogrammingburdenmakesitdif culttoapplyimperativeapproacheseffectivelytolargerapplications.Othersystems,suchasTempo[6,26],Fabius[21],andourprevioussystem[1],followadeclarativeapproach,wheresparseuserannotationstriggeranalysesandtransformationsoftheprogram(usingpartialevaluation-styletechniques)toexploitvalue-speci cdynamiccompilation.Tokeepdynamiccompilationcostslow,thesesystemspreplanthepossibleeffectsofdynamicoptimizationsstatically,producingaspecializeddynamiccompilertunedtotheparticularpartoftheprogrambeingdynamicallyoptimized;thissortofpreplanningwecallstagingtheoptimization.Declarativeapproachesarerelativelyeasyforprogrammerstouse,butareonlyaspowerfulastheoptimizationstheyapply.Thelimitationsofpreviousdeclarativesystemspreventedthemfromcopingeffectivelywiththemoreinvolvedpatternsofcontrolanddata owfoundinsomesmallandmostlargeapplications,causingthemtomissoptimizationopportunitiesorforcingsubstantialrewriting to t the limitations of the system.

DyC(pronounceddicey)[13,14]isaselective,value-speci cdynamiccompilationsystemthathasgoodpotentialforproducingspeedupsonlarger,morecomplexCprograms.Toreducetheprogrammingburden,DyCisadeclarative,annotation-basedsystem.Tosupporteffectiveoptimization,DyCcontainsasophisticatedformofpartial-evaluationbinding-timeanalysis,includingprogram-point-speci cpolyvariantdivisionandspecialization,1anddynamicversionsoftraditionalglobalandpeepholeoptimizations.Tokeepdynamiccompilationcostslow,nearlyallofDyC’sdynamicoptimizationsarestaged,withthebulkoftheworkoftheoptimizationoccurringatstaticcompiletimeandwithnorun-timeprogramrepresentationoriterativeanalysesrequired.DyCautomaticallycachesthedynamicallycompiledcodeandreusesitwherepossible,relievinganotherprogrammerburdenandreducingdynamiccompilationoverhead.Programmerscandeclarativelyspecifypoliciesthatgoverntheaggressivenessofspecializationandcaching,enablingprogrammerstoget nercontroloverthedynamiccompilationprocesswhilepreservingthedeclarativemodel.(Inthefuture,wehopetotreatDyCasaback-endforatoolthatautomaticallydecideswheretoapplyselective,value-speci cdynamicoptimizations,generatingannotationsthatDyC then carries out.)

Thispaperassessesthebene tsandapplicabilityofDyC’sanalysesandtransformations,bothindividuallyandwhenappliedtogether,

1

Polyvariantdivisionallowsthesamepieceofcodetobeanalyzedwithdifferentcombinationsofvariablesbeingtreatedasrun-timeconstants;eachcombinationiscalledadivision.Polyvariantspecializationallowsmultiplecompiledversionsofadivisiontobeproduced,eachspecializedfordifferentvaluesoftherun-time-constantvariables.Program-point-specificpolyvarianceenablespolyvariancetoariseatarbitrarypointsinprograms, not just at function entries.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email