DOCSLIB.ORG

Machine Learning in Compiler Optimisation Zheng Wang and Michael O’Boyle

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Machine Learning in Compiler Optimisation Zheng Wang and Michael O’Boyle 1 Machine Learning in Compiler Optimisation Zheng Wang and Michael O’Boyle Abstract—In the last decade, machine learning based com- hope of improving performance but could in some instances pilation has moved from an an obscure research niche to damage it. a mainstream activity. In this article, we describe the rela- Machine learning predicts an outcome for a new data point tionship between machine learning and compiler optimisation and introduce the main concepts of features, models, training based on prior data. In its simplest guise it can be considered and deployment. We then provide a comprehensive survey and a from of interpolation. This ability to predict based on prior provide a road map for the wide variety of different research information can be used to find the data point with the best areas. We conclude with a discussion on open issues in the outcome and is closely tied to the area of optimisation. It is at area and potential research directions. This paper provides both this overlap of looking at code improvement as an optimisation an accessible introduction to the fast moving area of machine learning based compilation and a detailed bibliography of its problem and machine learning as a predictor of the optima main achievements. where we find machine-learning compilation. Optimisation as an area, machine-learning based or other- Index Terms—Compiler, Machine Learning, Code Optimisa- tion, Program Tuning wise, has been studied since the 1800s [8], [9]. An interesting question is therefore why has has the convergence of these two areas taken so long? There are two fundamental reasons. I. INTRODUCTION Firstly, despite the year-on year increasing potential perfor- “Why would anyone want to use machine learning to build a mance of hardware, software is increasingly unable to realise it compiler?” It’s a view expressed by many colleagues over the leading to a software-gap. This gap has yawned right open with last decade. Compilers translate programming languages writ- the advent of multi-cores (see also Section VI-B). Compiler ten by humans into binary executable by computer hardware. writers are looking for new ways to bridge this gap. It is a serious subject studied since the 50s [1], [2], [3] where Secondly, computer architecture evolves so quickly, that it correctness is critical and caution is a by-word. Machine- is difficult to keep up. Each generation has new quirks and learning on the other hand is an area of artificial intelligence compiler writers are always trying to play catch-up. Machine aimed at detecting and predicting patterns. It is a dynamic field learning has the desirable property of being automatic. Rather looking at subjects as diverse as galaxy classification [4] to than relying on expert compiler writers to develop clever predicting elections based on tweeter feeds [5]. When an open- heuristics to optimise the code, we can let the machine learn source machine learning compiler was announced by IBM in how to optimise a compiler to make the machine run faster, an 2009 [6], some wry slashdot commentators picked up on the approach sometimes referred to as auto-tuning [10], [11], [12], AI aspect, predicting the start of sentient computers, global net [13]. Machine learning is, therefore, ideally suited to making and the war with machines from the Terminator film series. any code optimization decision where the performance impact In fact as we will see in this article, compilers and machine depends on the underlying platform. As described later in this learning are a natural fit and have developed into an established paper, it can be used for topics ranging from selecting the research domain. best compiler flags to determining how to map parallelism to processors. A. It’s all about optimization Machine learning is part of a tradition in computer science arXiv:1805.03441v1 [cs.PL] 9 May 2018 and compilation in increasing automation The 50s to 70s were Compiler have two jobs – translation and optimisation. spent trying to automate compiler translation, e.g. lex for They must first translate programs into binary correctly. lexical analysis [14] and yacc for parsing [15], the last decade Secondly they have to find the most efficient translation by contrast has focussed on trying to automating compiler possible. There are many different correct translations whose optimisation. As we will see it is not “magic” or a panacea for performance varies significantly. The vast majority of research compiler writers, rather it is another tool allowing automation and engineering practices is focussed on this second goal of of tedious aspects of compilation providing new opportunities performance, traditionally misnamed optimisation. The goal for innovation. It also brings compilation nearer to the stan- was misnamed because in most cases, till recently finding dards of evidence based science. It introduces an experimental an optimal translation was dismissed as being too hard to methodology where we separate out evaluation from design find and an unrealistic endeavour1. Instead it focussed on and considers the robustness of solutions. Machine learning developing compiler heuristics to transform the code in the based schemes in general have the problem of relying on Z. Wang is with MetaLab, School of Computing and Communications, black-boxes whose working we do not understand and hence Lancaster University, U. K. E-mail: [email protected] trust. This problem is just as true for machine learning based M. O’Boyle is with School of Informatics, University of Edinburgh, U. K. compilers. In this paper we aim to demystify machine learning E-mail: [email protected] 1In fact the term superoptimizer [7] was coined to describe systems that based compilation and show it is a trustworthy and exciting tried to find the optimum direction for compiler research. 2 Features for new program #inst. Supervised … #load ... for( ) { Machine Model Model ... #branch Training programs Learner New Prediction } cache miss rate programs + program Features oftraining ... Optimal options (a) Feature engineering (b) Learning a model (c) Deployment Fig. 1: A generic view of supervised machine learning in compilers. The remainder of this article is structured as follows. We Standard machine learning algorithms typically work on first give an intuitive overview for machine learning in compil- fixed length inputs, so the selected properties will be sum- ers in Section II. We then describe how machine learning can marised into a fixed length feature vector. Each element of be used to search for or to directly predict good compiler opti- the vector can be an integer, real or Boolean value. The mizations in Section III. This is followed by a comprehensive process of feature selection and tuning is referred as feature discussion in Section IV for a wide range of machine learning engineering. This process may need to iteratively perform models that have been employed in prior work. Next, in multiple times to find a set of high-quality features to build a Section V, we review how previous work chooses quantifiable accurate machine learning model. In Section V, we provide a properties, or features, to represent programs. We discuss the comprehensive review of feature engineering for the topic of challenges and limitations for applying machine learning to program optimisation. compilation, as well as open research directions in Section VII before we summarise and conclude in Section VIII. B. Learning a model The second step is to use training data to derive a model II. OVERVIEW OF MACHINE LEARNING IN COMPILERS using a learning algorithm. This process is depicted in Fig- Given a program, compiler writers would like to know what ure 1b. Unlike other applications of machine learning, we compiler heuristic or optimisation to apply in order to make typically generate our own training data using existing ap- the code better. Better often means execute faster, but can plications or benchmarks. The compiler developer will select also mean smaller code footprint or reduced power. Machine training programs which are typical of the application domain. learning can be used to build a model used within the compiler, For each training program, we calculate the feature values, that makes such decisions for any given program. compiling the program with different optimisation options, and running and timing the compiled binaries to discover the best- There are two main stages involved: learning and deploy- performing option. This process produces, for each training ment. The first stage learns the model based on training data, program, a training instance that consists of the feature values while the second uses the model on new unseen programs. and the optimal compiler option for the program. Within the learning stage, we needs a way of representing The compiler developer then feeds these examples to a programs in a systematic way. This representation is known machine learning algorithm to automatically build a model. as the program features [16]. The learning algorithms job is to find from the training Figure 1 gives a intuitive view how machine learning can examples a correlation between the feature values and the be applied to compilers. This process which includes feature optimal optimisation decision. The learned model can then engineering, learning a model and deployment is described in be used to predict, for a new set of features, what the optimal the following sub-sections. optimisation option should be. Because the performance of the learned model strongly A. Feature engineering depends on how well the features and training programs are Before we can learn anything useful about programs, we chosen, so that the processes of featuring engineering and first need to be able to characterise them. Machine learn- training data generation often need to repeat multiple times. ing relies on a set of quantifiable properties, or features, to characterise the programs (Figure 1a).
Load more

Recommended publications
  • Program Analysis and Optimization for Embedded Systems
    Program Analysis and Optimization for Embedded Systems Mike Jochen and Amie Souter jochen, souter ¡ @cis.udel.edu Computer and Information Sciences University of Delaware Newark, DE 19716 (302) 831-6339 1 Introduction ¢ have low cost A proliferation in use of embedded systems is giving cause ¢ have low power consumption for a rethinking of traditional methods and effects of pro- ¢ require as little physical space as possible gram optimization and how these methods and optimizations can be adapted for these highly specialized systems. This pa- ¢ meet rigid time constraints for computation completion per attempts to raise some of the issues unique to program These requirements place constraints on the underlying analysis and optimization for embedded system design and architecture’s design, which together affect compiler design to address some of the solutions that have been proposed to and program optimization techniques. Each of these crite- address these specialized issues. ria must be balanced against the other, as each will have an This paper is organized as follows: section 2 provides effect on the other (e.g. making a system smaller and faster pertinent background information on embedded systems, de- will typically increase its cost and power consumption). A lineating the special needs and problems inherent with em- typical embedded system, as illustrated in figure 1, consists bedded system design; section 3 addresses one such inherent of a processor core, a program ROM (Read Only Memory), problem, namely limited space for program code; section 4 RAM (Random Access Memory), and an ASIC (Applica- discusses current approaches and open issues for code min- tion Specific Integrated Circuit).
    [Show full text]
  • Stochastic Program Optimization by Eric Schkufza, Rahul Sharma, and Alex Aiken
    research highlights DOI:10.1145/2863701 Stochastic Program Optimization By Eric Schkufza, Rahul Sharma, and Alex Aiken Abstract Although the technique sacrifices completeness it pro- The optimization of short sequences of loop-free, fixed-point duces dramatic increases in the quality of the resulting code. assembly code sequences is an important problem in high- Figure 1 shows two versions of the Montgomery multiplica- performance computing. However, the competing constraints tion kernel used by the OpenSSL RSA encryption library. of transformation correctness and performance improve- Beginning from code compiled by llvm −O0 (116 lines, not ment often force even special purpose compilers to pro- shown), our prototype stochastic optimizer STOKE produces duce sub-optimal code. We show that by encoding these code (right) that is 16 lines shorter and 1.6 times faster than constraints as terms in a cost function, and using a Markov the code produced by gcc −O3 (left), and even slightly faster Chain Monte Carlo sampler to rapidly explore the space of than the expert handwritten assembly included in the all possible code sequences, we are able to generate aggres- OpenSSL repository. sively optimized versions of a given target code sequence. Beginning from binaries compiled by llvm −O0, we are able 2. RELATED WORK to produce provably correct code sequences that either Although techniques that preserve completeness are effec- match or outperform the code produced by gcc −O3, icc tive within certain domains, their general applicability −O3, and in some cases expert handwritten assembly. remains limited. The shortcomings are best highlighted in the context of the code sequence shown in Figure 1.
    [Show full text]
  • The Effect of Code Expanding Optimizations on Instruction Cache Design
    May 1991 UILU-EN G-91-2227 CRHC-91-17 Center for Reliable and High-Performance Computing THE EFFECT OF CODE EXPANDING OPTIMIZATIONS ON INSTRUCTION CACHE DESIGN William Y. Chen Pohua P. Chang Thomas M. Conte Wen-mei W. Hwu Coordinated Science Laboratory College of Engineering UNIVERSITY OF ILLINOIS AT URBANA-CHAMPAIGN Approved for Public Release. Distribution Unlimited. UNCLASSIFIED____________ SECURITY ¿LASSIPlOvriON OF t h is PAGE REPORT DOCUMENTATION PAGE 1a. REPORT SECURITY CLASSIFICATION 1b. RESTRICTIVE MARKINGS Unclassified None 2a. SECURITY CLASSIFICATION AUTHORITY 3 DISTRIBUTION/AVAILABILITY OF REPORT 2b. DECLASSIFICATION/DOWNGRADING SCHEDULE Approved for public release; distribution unlimited 4. PERFORMING ORGANIZATION REPORT NUMBER(S) 5. MONITORING ORGANIZATION REPORT NUMBER(S) UILU-ENG-91-2227 CRHC-91-17 6a. NAME OF PERFORMING ORGANIZATION 6b. OFFICE SYMBOL 7a. NAME OF MONITORING ORGANIZATION Coordinated Science Lab (If applicable) NCR, NSF, AMD, NASA University of Illinois N/A 6c ADDRESS (G'ty, Staff, and ZIP Code) 7b. ADDRESS (Oty, Staff, and ZIP Code) 1101 W. Springfield Avenue Dayton, OH 45420 Urbana, IL 61801 Washington DC 20550 Langley VA 20200 8a. NAME OF FUNDING/SPONSORING 8b. OFFICE SYMBOL ORGANIZATION 9. PROCUREMENT INSTRUMENT IDENTIFICATION NUMBER 7a (If applicable) N00014-91-J-1283 NASA NAG 1-613 8c. ADDRESS (City, State, and ZIP Cod*) 10. SOURCE OF FUNDING NUMBERS PROGRAM PROJECT t a sk WORK UNIT 7b ELEMENT NO. NO. NO. ACCESSION NO. The Effect of Code Expanding Optimizations on Instruction Cache Design 12. PERSONAL AUTHOR(S) Chen, William, Pohua Chang, Thomas Conte and Wen-Mei Hwu 13a. TYPE OF REPORT 13b. TIME COVERED 14. OATE OF REPORT (Year, Month, Day) Jl5.
    [Show full text]
  • ARM Optimizing C/C++ Compiler V20.2.0.LTS User's Guide (Rev. V)
    ARM Optimizing C/C++ Compiler v20.2.0.LTS User's Guide Literature Number: SPNU151V January 1998–Revised February 2020 Contents Preface ........................................................................................................................................ 9 1 Introduction to the Software Development Tools.................................................................... 12 1.1 Software Development Tools Overview ................................................................................. 13 1.2 Compiler Interface.......................................................................................................... 15 1.3 ANSI/ISO Standard ........................................................................................................ 15 1.4 Output Files ................................................................................................................. 15 1.5 Utilities ....................................................................................................................... 16 2 Using the C/C++ Compiler ................................................................................................... 17 2.1 About the Compiler......................................................................................................... 18 2.2 Invoking the C/C++ Compiler ............................................................................................. 18 2.3 Changing the Compiler's Behavior with Options ......................................................................
    [Show full text]
  • Compiler Optimization-Space Exploration
    Journal of Instruction-Level Parallelism 7 (2005) 1-25 Submitted 10/04; published 2/05 Compiler Optimization-Space Exploration Spyridon Triantafyllis [email protected] Department of Computer Science Princeton University Princeton, NJ 08540 USA Manish Vachharajani [email protected] Department of Electrical and Computer Engineering University of Colorado at Boulder Boulder, CO 80309 USA David I. August [email protected] Department of Computer Science Princeton University Princeton, NJ 08540 USA Abstract To meet the performance demands of modern architectures, compilers incorporate an ever- increasing number of aggressive code transformations. Since most of these transformations are not universally beneficial, compilers traditionally control their application through predictive heuris- tics, which attempt to judge an optimization’s effect on final code quality a priori. However, com- plex target architectures and unpredictable optimization interactions severely limit the accuracy of these judgments, leading to performance degradation because of poor optimization decisions. This performance loss can be avoided through the iterative compilation approach, which ad- vocates exploring many optimization options and selecting the best one a posteriori. However, existing iterative compilation systems suffer from excessive compile times and narrow application domains. By overcoming these limitations, Optimization-Space Exploration (OSE) becomes the first iterative compilation technique suitable for general-purpose production compilers. OSE nar- rows down the space of optimization options explored through limited use of heuristics. A compiler tuning phase further limits the exploration space. At compile time, OSE prunes the remaining opti- mization configurations in the search space by exploiting feedback from earlier configurations tried. Finally, rather than measuring actual runtimes, OSE compares optimization outcomes through static performance estimation, further enhancing compilation speed.
    [Show full text]
  • XL Fortran: Getting Started
    IBM XL Fortran for AIX, V12.1 Getting Started with XL Fortran Ve r s i o n 12.1 GC23-8898-00 IBM XL Fortran for AIX, V12.1 Getting Started with XL Fortran Ve r s i o n 12.1 GC23-8898-00 Note Before using this information and the product it supports, read the information in “Notices” on page 29. First edition This edition applies to IBM XL Fortran for AIX, V12.1 (Program number 5724-U82) and to all subsequent releases and modifications until otherwise indicated in new editions. Make sure you are using the correct edition for the level of the product. © Copyright International Business Machines Corporation 1996, 2008. All rights reserved. US Government Users Restricted Rights – Use, duplication or disclosure restricted by GSA ADP Schedule Contract with IBM Corp. Contents About this document . .v Architecture and processor support . .12 Conventions . .v Performance and optimization . .13 Related information . .ix Other new or changed compiler options . .15 IBM XL Fortran information . .ix Standards and specifications . .x Chapter 3. Setting up and customizing Other IBM information. .xi XL Fortran . .17 Technical support . .xi Using custom compiler configuration files . .17 How to send your comments . .xi Chapter 4. Developing applications Chapter 1. Introducing XL Fortran . .1 with XL Fortran . .19 Commonality with other IBM compilers . .1 The compiler phases . .19 Hardware and operating system support . .1 Editing Fortran source files . .19 A highly configurable compiler . .1 Compiling with XL Fortran . .20 Language standards compliance . .2 Invoking the compiler . .22 Source-code migration and conformance checking 3 Compiling parallelized XL Fortran applications 22 Tools and utilities .
    [Show full text]
  • An Optimization-Driven Incremental Inline Substitution Algorithm for Just-In-Time Compilers
    An Optimization-Driven Incremental Inline Substitution Algorithm for Just-in-Time Compilers Aleksandar Prokopec Gilles Duboscq David Leopoldseder Thomas Würthinger Oracle Labs, Switzerland Oracle Labs, Switzerland JKU Linz, Austria Oracle Labs, Switzerland [email protected] [email protected] [email protected] [email protected] Abstract—Inlining is one of the most important compiler mization prediction by incrementally exploring and specializing optimizations. It reduces call overheads and widens the scope of the program’s call tree, and clusters the callsites before inlining. other optimizations. But, inlining is somewhat of a black art of an We report the following new observations: optimizing compiler, and was characterized as a computationally intractable problem. Intricate heuristics, tuned during countless • In many programs, there is an impedance between the hours of compiler engineering, are often at the core of an logical units of code (i.e. subroutines) and the optimizable inliner implementation. And despite decades of research, well- units of code (groups of subroutines). It is therefore established inlining heuristics are still missing. beneficial to inline specific clusters of callsites, instead In this paper, we describe a novel inlining algorithm for of a single callsite at a time. We describe how to identify JIT compilers that incrementally explores a program’s call graph, and alternates between inlining and optimizations. We such callsite clusters. devise three novel heuristics that guide our inliner: adaptive • When deciding when to stop inlining, using an adaptive decision thresholds, callsite clustering, and deep inlining trials. threshold function (rather than a fixed threshold value) We implement the algorithm inside Graal, a dynamic JIT produces more efficient programs.
    [Show full text]
  • A Unified Approach to Global Program Optimization
    A UNIFIED APPROACH TO GLOBAL PROGRAM OPTIMIZATION Gary A. Kildall Computer Science Group Naval Postgraduate School Monterey, California Abstract A technique is presented for global analysie of program structure in order to perform compile time optimization of object code generated for expressions. The global expression optimization presented includes constant propagation, common subexpression elimination, elimination of redundant register load operations, and live expression analysis. A general purpose program flow analysis algorithm is developed which depends upon the existence of an “optimizing function.” The algorithm is defined formally using a directed graph model of program flow structure, and is shown to be correct, Several optimizing functions are defined which, when used in conjunction with the flow analysis algorithm, provide the various forms of code optimization. The flow analysis algorithm is sufficiently general that additional functions can easily be defined for other forms of globa~ cod: optimization. 1. INTRODUCTION of the graph represent program control flow Dossi- bilities between the nodes at execution–time. A number of techniques have evolved for the compile-time analysis of program structure in order to locate redundant computations, perform constant computations, and reduce the number of store–load sequences between memory and high-speed registers. Some of these techniques provide analysis of only entry’ ~:=j straight-line sequences of instructions [5,6,9,14, o 17,18,19,20,27,29,34,36,38,39,43,45 ,46], while B others take the program branching structure into &() account [2,3,4,10,11,12,13,15,23,30, 32,33,35]. The purpose here is to describe a single program c flow analysis algorithm which extends all of b=2 these straight-line optimizing techniques to in- D clude branching structure.
    [Show full text]
  • XL C/C++: Optimization and Programming Guide About This Document
    IBM XL C/C++ for Linux, V16.1.1 IBM Optimization and Programming Guide Version 16.1.1 SC27-8046-01 IBM XL C/C++ for Linux, V16.1.1 IBM Optimization and Programming Guide Version 16.1.1 SC27-8046-01 Note Before using this information and the product it supports, read the information in “Notices” on page 121. First edition This edition applies to IBM XL C/C++ for Linux, V16.1.1 (Program 5765-J13, 5725-C73) and to all subsequent releases and modifications until otherwise indicated in new editions. Make sure you are using the correct edition for the level of the product. © Copyright IBM Corporation 1996, 2018. US Government Users Restricted Rights – Use, duplication or disclosure restricted by GSA ADP Schedule Contract with IBM Corp. Contents About this document ......... v Optimizing at level 4 .......... 26 Who should read this document........ v Optimizing at level 5 .......... 27 How to use this document.......... v Tuning for your system architecture ...... 28 How this document is organized ....... v Getting the most out of target machine options 28 Conventions .............. vi Using high-order loop analysis and transformations 29 Related information ............ ix Getting the most out of -qhot ....... 30 Available help information ........ ix Using shared-memory parallelism (SMP) .... 31 Standards and specifications ........ xi Getting the most out of -qsmp ....... 32 Technical support ............ xi Using interprocedural analysis ........ 33 How to send your comments ........ xii Getting the most from -qipa ........ 35 Using profile-directed feedback ........ 36 Chapter 1. Using XL C/C++ with Fortran 1 Compiling with -qpdf1 ......... 37 Training with typical data sets ......
    [Show full text]
  • Demand-Driven Inlining Heuristics in a Region-Based Optimizing Compiler for ILP Architectures
    Demand-driven Inlining Heuristics in a Region-based Optimizing Compiler for ILP Architectures Tom Way, Ben Breech, Wei Du and Lori Pollock Department of Computer and Information Sciences University of Delaware, Newark, DE 19716, USA ¡ email: way,breech,wei,pollock ¢ @cis.udel.edu ABSTRACT compilation unit size and content. In addition to expos- While scientists and engineers have been taking on the ing opportunities for interprocedural instruction schedul- challenge of converting their applications into parallel ing and optimization, the size of the resulting regions (i.e., programs, compilers are charged with identifying op- compilation units) has been shown to be smaller than func- portunities and transforming codes to exploit the avail- tions, and can be limited to any desired size, in order to able instruction-level parallelism (ILP) offered by today’s reduce the impact of the quadratic algorithmic complexity uniprocessors, even within a parallel computing environ- of the applied optimization transformations [14]. ment. Region-based compilation repartitions a program The key component of a region-based compiler is the into more desirable compilation units for optimization and region formation phase which partitions the program into ILP scheduling. With region-based compilation, the com- regions using heuristics with the intent that the optimizer piler can control problem size and complexity by control- will be invoked with a scope that is limited to a single re- ling region size and contents, and expose interprocedu- gion at a time. Thus, the quality of the generated code ral scheduling and optimization opportunities without in- depends greatly upon the ability of the region formation terprocedural analysis or large function bodies.
    [Show full text]
  • Analysis and Optimization of Scientific Applications Through Set and Relation Abstractions M
    Louisiana State University LSU Digital Commons LSU Doctoral Dissertations Graduate School 2017 Analysis and Optimization of Scientific Applications through Set and Relation Abstractions M. Tohid (Rastegar Tohid, Mohammed) Louisiana State University and Agricultural and Mechanical College Follow this and additional works at: https://digitalcommons.lsu.edu/gradschool_dissertations Part of the Electrical and Computer Engineering Commons Recommended Citation Tohid (Rastegar Tohid, Mohammed), M., "Analysis and Optimization of Scientific Applications through Set and Relation Abstractions" (2017). LSU Doctoral Dissertations. 4404. https://digitalcommons.lsu.edu/gradschool_dissertations/4404 This Dissertation is brought to you for free and open access by the Graduate School at LSU Digital Commons. It has been accepted for inclusion in LSU Doctoral Dissertations by an authorized graduate school editor of LSU Digital Commons. For more information, please [email protected]. ANALYSIS AND OPTIMIZATION OF SCIENTIFIC APPLICATIONS THROUGH SET AND RELATION ABSTRACTIONS A Dissertation Submitted to the Graduate Faculty of the Louisiana State University and Agricultural and Mechanical College in partial fulfillment of the requirements for the degree of Doctor of Philosophy in The School of Electrical Engineering and Computer Science by Mohammad Rastegar Tohid M.Sc., Newcastle University, 2008 B.Sc., Azad University (S. Tehran Campus), 2006 May 2017 Dedication Dedicated to my parents who have dedicated their lives to their children. ii Acknowledgments I thank my advisor Dr. Ramanujam for valueing independent research and allowing me to work on topics that geniunely intrigued me while never lettig me go off the rail. His optimism and words of encouragement always aspired me. It has been a great pleasure to have known and worked with him.
    [Show full text]
  • Object Code Optimization Analysis of Source Code and Produces the Most Efficient Object Code of Any Compiler Presently Known to the Writers
    R. M. McCLURE, Editor Object Code Optimization analysis of source code and produces the most efficient object code of any compiler presently known to the writers. For small loops of a few statements, it very often EDWARD S. LOWRY AND C. W. MEDLOCK* produces perfect code. The efficiency is limited mainly by International Business Machines Corp., Poughkeepsie, N. Y. the (rather unnecessary) restrictions on the types of data organization that can be described in FORTRAN and the inability to combine subprograms in a single compilation. Methods of analyzing the control flow and data flow of pro- Of course the optimization does not compensate for in- grams during compilation are applied to transforming the pro- efficient source coding except in minor ways. The methods gram to improve object time efficiency. Dominance relation- will apply to almost any other procedural language, ships, indicating which statements are necessarily executed but they are not effective for interpretive systems or for before others, are used to do global common expression object code that relies heavily on precoded library sub- elimination and loop identification. Implementation of these routines. and other optimizations in OS/360 FORTRAN H are described. Most optimization can be optionally bypassed to gain a 40 percent reduction in compilation time. However, KEY WORDS AND PHRASES: compilers, data flow analysis, dominancee this expands the object code by about 25 percent, and efficiency, FORTRAN, graph theory, loop structure, machine instructions, execution times are increased threefold. object code, optimization, redundancy elimination, register assignment, System/360 Compiler Organization CR CATEGORIES: 4.12, 5.24, 5.32 The optimization within the FORTRAN I-I compiler Introduction proceeds as follows.
    [Show full text]