Sun Microsystems FORTRAN 77 5.0, Fortran 90 2.0 Compilers Programming Guide

Fortran Programming Guide
FORTRAN 77 5.0 — Fortran 90
2.0
901 San Antonio Road
Palo Alto, , CA 94303-4900
USA 650 960-1300 fax 650 969-9131
Part No: 805-4940
Revision A, February 1999
Copyright Copyright 1999 Sun Microsystems, Inc. 901 San Antonio Road, Palo Alto, California 94303-4900 U.S.A. All rights reserved.
All rights reserved. This product or document is protected by copyright and distributed under licenses restricting its use, copying,
distribution, and decompilation. No part of this product or document may be reproduced in any form by any means without prior written
authorization of Sun and its licensors, if any.
Portions of this product may be derived from the UNIX® system, licensed from Novell, Inc., and from the Berkeley 4.3 BSD system,
licensed from the University of California. UNIX is a registered trademark in the United States and in other countries and is exclusively
licensed by X/Open Company Ltd. Third-party software, including font technology in this product, is protected by copyright and licensed
from Sun’s suppliers. RESTRICTED RIGHTS: Use, duplication, or disclosure by the U.S. Government is subject to restrictions of FAR
52.227-14(g)(2)(6/87) and FAR 52.227-19(6/87), or DFAR 252.227-7015(b)(6/95) and DFAR 227.7202-3(a).
Sun, Sun Microsystems, the Sun logo, SunDocs, SunExpress, Solaris, Sun Performance Library, Sun Performance WorkShop Fortran, Sun
Visual WorkShop C++, Sun WorkShop, and Sun WorkShop Professional are trademarks or registered trademarks of Sun Microsystems, Inc.
in the United States and in other countries. All SPARC trademarks are used under license and are trademarks or registered trademarks of
SPARC International, Inc. in the United States and in other countries. Products bearing SPARC trademarks are based upon an architecture
developed by Sun Microsystems, Inc.
TM
The OPEN LOOK® and Sun Graphical User Interfaces were developed by Sun Microsystems, Inc. for its users and licensees. Sun
acknowledges the pioneering efforts of Xerox Corporation in researching and developing the concept of visual or graphical user interfaces
for the computer industry. Sun holds a nonexclusive license from Xerox to the Xerox Graphical User Interface, which license also covers
Sun’s licensees who implement OPEN LOOK GUIs and otherwise comply with Sun’s written license agreements.
TM
f90 is derived from Cray CF90 , a product of Silicon Graphics, Inc.
THIS PUBLICATION IS PROVIDED “AS IS” WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING,
BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR
NON-INFRINGEMENT.
Copyright 1999 Sun Microsystems, Inc., 901 San Antonio Road, Palo Alto, Californie 94303-4900 U.S.A. Tous droits réservés.
Ce produit ou document est protégé par un copyright et distribué avec des licences qui en restreignent l’utilisation, la copie et la
décompilation. Aucune partie de ce produit ou de sa documentation associée ne peut être reproduite sous aucune forme, par quelque
moyen que ce soit, sans l’autorisation préalable et écrite de Sun et de ses bailleurs de licence, s’il y en a.
Des parties de ce produit pourront être derivées du système UNIX® licencié par Novell, Inc. et du système Berkeley 4.3 BSD licencié par
l’Université de Californie. UNIX est une marque enregistrée aux Etats-Unis et dans d’autres pays, et licenciée exclusivement par X/Open
Company Ltd. Le logiciel détenu par des tiers, et qui comprend la technologie relative aux polices de caractères, est protégé par un
copyright et licencié par des fournisseurs de Sun.
Sun, Sun Microsystems, le logo Sun, SunDocs, SunExpress, Solaris, Sun Performance Library, Sun Performance WorkShop Fortran, Sun
Visual WorkShop C++, Sun WorkShop, et Sun WorkShop Professional sont des marques déposées ou enregistrées de Sun Microsystems,
Inc. aux Etats-Unis et dans d’autres pays. Toutes les marques SPARC, utilisées sous licence, sont des marques déposées ou enregistrées de
SPARC International, Inc. aux Etats-Unis et dans d’autres pays. Les produits portant les marques SPARC sont basés sur une architecture
développée par Sun Microsystems, Inc.
TM
Les utilisateurs d’interfaces graphiques OPEN LOOK® et Sun ont été développés de Sun Microsystems, Inc. pour ses utilisateurs et
licenciés. Sun reconnaît les efforts de pionniers de Xerox Corporation pour la recherche et le développement du concept des interfaces
d’utilisation visuelle ou graphique pour l’industrie de l’informatique. Sun détient une licence non exclusive de Xerox sur l’interface
d’utilisation graphique, cette licence couvrant aussi les licenciés de Sun qui mettent en place les utilisateurs d’interfaces graphiques OPEN
LOOK et qui en outre se conforment aux licences écrites de Sun.
TM
f90 est derivé de CRAY CF90 , un produit de Silicon Graphics, Inc.
CETTE PUBLICATION EST FOURNIE "EN L’ETAT" SANS GARANTIE D’AUCUNE SORTE, NI EXPRESSE NI IMPLICITE, Y COMPRIS,
ET SANS QUE CETTE LISTE NE SOIT LIMITATIVE, DES GARANTIES CONCERNANT LA VALEUR MARCHANDE, L’APTITUDE DES
PRODUITS A REPONDRE A UNE UTILISATION PARTICULIERE OU LE FAIT QU’ILS NE SOIENT PAS CONTREFAISANTS DE
PRODUITS DE TIERS.
Please
Recycle
Contents
Preface
1.
xi
Introduction 1–1
Standards Conformance
1–1
Features of the Fortran Compilers 1–2
2.
Fortran Input/Output
2–1
Accessing Files From Within Fortran Programs 2–1
Accessing Named Files
2–1
Opening Files Without a Name
Preconnected Units
2–3
2–3
Opening Files Without an OPEN Statement
Passing File Names to Programs
2–4
f77: VAX / VMS Logical File Names
Direct I/O
Internal Files
2–3
2–7
2–8
2–9
Tape I/O (f77 Only)
2–11
Using TOPEN Routines
2–11
Fortran Formatted I/O for Tape
2–11
Fortran Unformatted I/O for Tape
Tape File Representation
2–11
2–12
Contents
iii
End-of-File
2–12
Multifile Tapes
2–13
Fortran 90 I/O Considerations
3.
Program Development
2–13
3–1
Facilitating Program Builds With the make Utility 3–1
The Makefile 3–2
make Command
Macros
3–3
3–3
Overriding of Macro Values
Suffix Rules in make
3–4
3–4
Version Tracking and Control With SCCS
Controlling Files With SCCS
3–5
3–5
Checking Files Out and In 3–7
4.
Libraries 4–1
Understanding Libraries
4–1
Specifying Linker Debugging Options
4–2
Generating a Load Map 4–2
Listing Other Information 4–3
Consistent Compiling and Linking 4–4
Setting Library Search Paths and Order
4–4
Search Order for Standard Library Paths 4–4
LD_LIBRARY_PATH Environment Variable
4–5
Library Search Path and Order—Static Linking
4–6
Library Search Path and Order—Dynamic Linking 4–7
Creating Static Libraries
4–8
Tradeoffs for Static Libraries
4–8
Creation of a Simple Static Library
Creating Dynamic Libraries
iv
4–12
Fortran Programming Guide ♦ Revision A, February 1999
4–9
Tradeoffs for Dynamic Libraries
4–12
Position-Independent Code and -pic
4–13
Binding Options 4–13
Naming Conventions 4–14
A Simple Dynamic Library 4–15
Libraries Provided with Sun Fortran Compilers 4–16
VMS Library
4–17
POSIX Library
4–17
Shippable Libraries 4–18
5.
Program Analysis and Debugging 5–1
Global Program Checking 5–1
GPC Overview 5–1
How to Invoke Global Program Checking 5–2
Some Examples of −-Xlist and Global Program Checking 5–4
Suboptions for Global Checking Across Routines
-Xlist Suboption Reference
5–7
5–9
Some Examples Using Suboptions 5–12
Special Compiler Options
5–14
Subscript Bounds (-C) 5–14
f77: Undeclared Variable Types (-u)
5–14
Version Checking (-V) 5–15
Interactive Debugging With dbx and Sun WorkShop 5–15
f77: Viewing Compiler Listing Diagnostics
6.
Floating-Point Arithmetic
Introduction
5–16
6–1
6–1
IEEE Floating-Point Arithmetic
6–2
-ftrap=mode Compiler Options 6–3
Floating-Point Exceptions and Fortran 6–3
Contents v
Handling Exceptions
6–4
Trapping a Floating-Point Exception
SPARC: Nonstandard Arithmetic
IEEE Routines
6–4
6–4
6–5
Flags and ieee_flags() 6–6
IEEE Extreme Value Functions 6–10
Exception Handlers and ieee_handler()
6–11
Retrospective Summary 6–15
Debugging IEEE Exceptions
6–16
Further Numerical Adventures 6–17
Avoiding Simple Underflow 6–18
Continuing With the Wrong Answer 6–19
SPARC: Excessive Underflow 6–19
7.
Porting 7–1
Time and Date Functions
Formats
7–1
7–4
Carriage-Control
7–4
Working With Files
7–5
Porting From Scientific Mainframes
Data Representation
7–5
7–6
Hollerith Data 7–7
Nonstandard Coding Practices 7–9
Uninitialized Variables
Aliasing Across Calls
Obscure Optimizations
Troubleshooting
7–9
7–10
7–10
7–11
Results Are Close, but Not Close Enough
Program Fails Without Warning 7–13
vi
Fortran Programming Guide ♦ Revision A, February 1999
7–12
8.
Performance Profiling 8–1
The time Command 8–1
Multiprocessor Interpretation of time Output 8–2
The gprof Profiling Command 8–2
Overhead Considerations 8–5
The tcov Profiling Command 8–5
“Old Style” tcov Coverage Analysis 8–5
“New Style” Enhanced tcov Analysis 8–6
f77 I/O Profiling
9.
8–7
Performance and Optimization 9–1
Choice of Compiler Options
9–1
Performance Option Reference
9–2
Other Performance Strategies 9–7
Using Optimized Libraries
9–7
Eliminating Performance Inhibitors 9–8
Further Reading
10.
9–9
SPARC: Parallelization 10–1
Essential Concepts
10–1
Speedups—What to Expect
10–2
Steps to Parallelizing a Program
Data Dependency Issues
10–3
10–3
Parallel Options and Directives Summary
Number of Processors
10–6
Stacks, Stack Sizes, and Parallelization
Automatic Parallelization
10–5
10–7
10–8
Loop Parallelization 10–8
Arrays, Scalars, and Pure Scalars
Automatic Parallelization Criteria
10–9
10–9
Contents vii
Automatic Parallelization With Reduction Operations
10–10
Explicit Parallelization 10–14
Parallelizable Loops
10–14
Sun-Style Parallelization Directives
10–16
Cray-Style Parallelization Directives
10–29
Debugging Parallelized Programs 10–32
Debugging Without dbx 10–32
Using dbx 10–34
11.
C-Fortran Interface 11–1
Compatibility Issues
11–1
Function or Subroutine?
11–2
Data Type Compatibility 11–2
Case Sensitivity
11–6
Underscores in Routine Names 11–6
Argument-Passing by Reference or Value
11–7
Argument Order 11–7
Array Indexing and Order
11–8
File Descriptors and stdio 11–8
File Permissions 11–9
Libraries and Linking With the f77 or f90 Command 11–10
Passing Data Arguments by Reference
Simple Data Types
COMPLEX Data
11–10
11–11
Character Strings 11–12
One-Dimensional Arrays
11–13
Two-Dimensional Arrays 11–14
Structures 11–15
Pointers 11–18
viii
Fortran Programming Guide ♦ Revision A, February 1999
11–10
Passing Data Arguments by Value
11–18
Functions That Return a Value 11–21
Returning a Simple Data Type
Returning COMPLEX Data
11–21
11–22
Returning a CHARACTER String
Labeled COMMON
11–23
11–26
Sharing I/O Between Fortran and C 11–27
Alternate Returns
11–27
Index 29
Contents ix
x
Fortran Programming Guide ♦ Revision A, February 1999
Preface
This guide combines the essential
information programmers need to develop efficient
TM
applications using the two Sun Fortran compilers, f77 (FORTRAN 77 version 5.0)
and f90 (Fortran 90 version 2.0). It presents issues relating to input/output, program
development, use and creation of software libraries, program analysis and
debugging, numerical accuracy, porting, performance, optimization, parallelization,
and the C/Fortran interface.
Discussion of the compiler command-line options and their use can be found in the
companion book, the Fortran User’s Guide.
Note - This guide covers the Sun FORTRAN 77 and Fortran 90 compilers. The text
uses "f77/f90" and “Fortran” to indicate information that is common to both
compilers.
Who Should Use This Book
This guide is intended for scientists, engineers, and programmers who have a
working knowledge of the Fortran language and wish toTMlearn how to use the Sun
Fortran compilers effectively. Familiarity with the Solaris operating environment or
UNIX®© in general is also assumed.
How This Book Is Organized
This guide is organized into the following chapters:
Preface xi
4 Chapter 1, "Introduction," briefly describes the features of the compilers.
4 Chapter 2, "Fortran Input/Output," discusses how to use I/O efficiently.
4 Chapter 3, "Program Development," demonstrates how program management tools
like SCCS, make, and Teamware can be helpful.
4 Chapter 4, "Libraries," explains use and creation of software libraries.
4 Chapter 5, "Program Analysis and Debugging," describes use of dbx and other
analysis tools.
4 Chapter 6, "Floating Point Arithmetic," introduces important issues regarding
numerical computation accuracy.
4 Chapter 7, "Porting," considers porting programs to Sun compilers.
4 Chapter 8, "Performance Profiling," describes techniques for performance
measurement.
4 Chapter 9, "Performance and Optimization," indicates ways to improve execution
performance of Fortran programs.
4 Chapter 10, "Parallelization," explains the multiprocessing features of the
compilers.
4 Chapter 11, "C-Fortran Interface," describes how C and Fortran routines can call
each other and pass data.
Multiplatform Release
Note - The name of the latest Solaris operating environment release is Solaris 7 but
some documentation and path or package path names may still use Solaris 2.7 or
SunOS 5.7.
The Sun Fortran documentation covers the release of the Fortran compilers on a
number of operating environments and hardware platforms:
FORTRAN 77 5.0 is released for:
4 Solaris 2.5.1, 2.6, and Solaris 7 environments on:
TM
4
architectures based on the SPARC
4
TM
x86-based architectures, where
x86 refers to
the Intel®TM
implementation of one
TM
TM
of the following: Intel 80386 , Intel 80486 , Pentium , or the equivalent
microprocessor
Fortran 90 2.0 is released for:
4 Solaris 2.5.1, 2.6, and Solaris 7 environments on SPARC processors only.
xii
Fortran Programming Guide ♦ Revision A, February 1999
Note - The term “x86” refers to the Intel 8086 family of microprocessor chips,
including the Pentium, Pentium Pro, and Pentium II processors and compatible
microprocessor chips made by AMD and Cyrix. In this document, the term “x86”
refers to the overall platform architecture. Features described in this book that are
particular to a specific platform are differentiated by the terms “SPARC” and
“x86” in the text.
Related Books
The following books augment this manual and provide essential information:
4 Fortran User’s Guide—provides information on command line options and how to
use the compilers.
4 FORTRAN 77 Language Reference. Complete language reference.
4 Fortran Library Reference-gives details on the language and routines.
4 Sun Performance WorkShop Fortran Overview gives a high-level outline of the
Fortran package suite.
Other Programming Books
4 C User’s Guide—describes compiler options, pragmas, and more.
4 Numerical Computation Guide—details floating-point computation and numerical
accuracy issues.
4 Sun WorkShop Performance Library Reference-discusses the library of subroutines
and functions to perform useful operations in computational linear algebra and
Fourier transforms.
Other Sun WorkShop Books
4 Sun WorkShop Quick Install-provides installation instructions.
4 Sun WorkShop Installation Reference-provides supporting installation and licensing
information.
4 Sun Visual WorkShop C++ Overview-gives a high-level outline of the C++
package suite.
4 Using Sun WorkShop—gives information on performing development operations
through Sun WorkShop.
xiii
4 Debugging a Program With dbx—provides information on using dbx commands
to debug a program.
4 Analyzing Program Performance with Sun WorkShop—describes the profiling tools;
LoopTool, LoopReport, LockLint utilities; and the Sampling Analyzer to enhance
program performance.
4 Sun WorkShop TeamWare User’s Guide—describes how to use the Sun WorkShop
TeamWare code management tools.
Solaris Books
The following Solaris manuals and guides provide additional useful information:
4 The Solaris Linker and Libraries Guide—gives information on linking and libraries.
4 The Solaris Programming Utilities Guide—provides information for developers
about the special built-in programming tools available in the SunOS system.
Ordering Sun Documents
The SunDocsSM program provides more than 250 manuals from Sun Microsystems,
Inc. If you live in the United States, Canada, Europe, or Japan, you can purchase
documentation sets or individual manuals using this program.
For a list of documents and how to order them, see the catalog section of the
SunExpressTM Internet site at http://www.sun.com/sunexpress.
Accessing Sun Documents Online
Sun WorkShop documentation is available online from several sources:
4 The docs.sun.com Web site
4 AnswerBook2TM collections
4 HTML documents
4 Online help and release notes
xiv
Fortran Programming Guide ♦ Revision A, February 1999
Using the docs.sun.com Web site
The docs.sun.com Web site enables you to access Sun technical documentation
online. You can browse the docs.sun.com archive or search for a specific book title
or subject. The URL is http://docs.sun.com.
Accessing AnswerBook2 Collections
The Sun WorkShop documentation is also available using AnswerBook2 software. To
access the AnswerBook2 collections, your system administrator must have installed
the AnswerBook2 documents during the installation process (if the documents are
not installed, see your system administrator or Chapter 3 of Sun WorkShop Quick
Install for installation instructions). For information about accessing AnswerBook2
documents, see Chapter 6 of Sun WorkShop Quick Install, Solaris installation
documentation, or your system administrator.
Note - To access AnswerBook2 documents, Solaris 2.5.1 users must first download
AnswerBook2 documentation server software from a Sun Web page. For more
information, see Chapter 6 of Sun WorkShop Quick Install.
Accessing HTML Documents
The following Sun Workshop documents are available online only in HTML format:
4 Tools.h++ Class Library Reference
4 Tools.h++ User’s Guide
4 Numerical Computation Guide
4 Standard C++ Library User’s Guide
4 Standard C++ Class Library Reference
4 Sun WorkShop Performance Library Reference Manual
4 Sun WorkShop Visual User’s Guide
4 Sun WorkShop Memory Monitor User’s Manual
To access these HTML documents:
1. Open the following file through your HTML browser:
install-directory/SUNWspro/DOC5.0/lib/locale/C/html/index.html
Replace install-directory with the name of the directory where your Sun WorkShop
software is installed (the default is /opt).
The browser displays an index of the HTML documents for the Sun WorkShop
products that are installed.
xv
2. Open a document in the index by clicking the document’s title.
Accessing Sun WorkShop Online Help and
Release Notes
This release of Sun WorkShop includes an online help system as well as online
manuals. To find out more see:
4 Online Help. A help system containing extensive task-oriented, context-sensitive
help. To access the help, choose Help Help Contents. Help menus are available in
all Sun WorkShop windows.
4 Release Notes. The Release Notes contain general information about Sun
WorkShop and specific information about software limitations and bugs. To access
the Release Notes, choose Help Release Notes.
4 You can view the latest release information regarding the Fortran compilers by
invoking the f77 or f90 compiler with the -xhelp=readme flag.
What Typographic Changes Mean
The following table describes the typographic changes used in this book.
TABLE P–1
Typographic Conventions
Typeface or
Symbol
AaBbCc123
Meaning
Example
The names of commands, files, and
directories; on-screen computer
output
Edit your .login file.
Use ls -a to list all files.
machine_name% You have mail.
AaBbCc123
What you type, contrasted with
on-screen computer output
machine_name% su
Password:
xvi
Fortran Programming Guide ♦ Revision A, February 1999
TABLE P–1
Typographic Conventions
(continued)
Typeface or
Symbol
Meaning
Example
AaBbCc123
Command-line placeholder:
To delete a file, type rm filename.
replace with a real name or value
Book titles, new words or terms, or
words to be emphasized
AaBbCc123
Read Chapter 6 in User’s Guide.
These are called class options.
You must be root to do this.
Shell Prompts in Command Examples
The following table shows the default system prompt and superuser prompt for the
C shell, Bourne shell, and Korn shell.
TABLE P–2
Shell Prompts
Shell
Prompt
C shell prompt
machine_name%
C shell superuser prompt
machine_name#
Bourne shell and Korn shell prompt
$
Bourne shell and Korn shell superuser prompt
#
Other Conventions Used in This Book
The following conventions appear in the text of this book:
4 Code listings and examples appear in boxes:
WRITE( *, * ) "Hello world"
xvii
4 The symbol ¤ stands for a blank space where a blank is significant:
¤¤36.001
4 FORTRAN 77 examples appear in tab format, while Fortran 90 examples appear in
free format. Examples common to both FORTRAN 77 and 90 use tab format except
where indicated.
4 Uppercase characters are generally used to show Fortran keywords and intrinsics
(PRINT), and lowercase or mixed case is used for variables (TbarX).
4 The Sun Fortran compilers are referred to by their command names, either f77 or
f90. "f77/f90" indicates information that is common to both the FORTRAN 77
and Fortran 90 compilers.
4 References to online man pages appear with the topic name and section number.
For example, a reference to GETENV will appear as getenv(3F), implying that the
man command to access this page would be: man -s 3F getenv
4 System Administrators may install the Sun Fortran compilers and supporting
material at: <install_point>/SUNWspro/SC5.0/ where <install_point> is usually
/opt for a standard install. This is the location assumed in this book.
xviii
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
1
Introduction
The Sun Fortran compilers, f77 and f90, described in this book (and the companion
Fortran User’s Guide) are available under the Solaris operating environment on the
various hardware platforms that Solaris supports. The compilers themselves conform
to published Fortran language standards, and provide many extended features,
including multiprocessor parallelization, sophisticated optimized code compilation,
and mixed C/Fortran language support.
Standards Conformance
4 f77 was designed to be compatible with the ANSI X3.9-1978 Fortran standard and
the corresponding International Organization for Standardization (ISO) 1539-1980,
as well as standards FIPS 69-1, BS 6832, and MIL-STD-1753.
4 f90 was designed to be compatible with the ANSI X3.198-1992 standard, and
ISO/IEC 1539:1991
4 Floating-point arithmetic for both compilers is based on IEEE standard 754-1985,
and international standard IEC 60559:1989.
4 On SPARC processors, both compilers provide support for the
optimization-exploiting
features of SPARC V8, and SPARC V9, including the
TM
UltraSPARC implementation. These features are defined in the SPARC
Architecture Manuals, Version 8 (ISBN 0-13-825001-4), and Version 9 (ISBN
0-13-099227-5), published by Prentice-Hall for SPARC International.
4 In this document, "Standard" means conforming to the versions of the standards
listed above. "Non-standard" or "Extension" refers to features that go beyond these
versions of these standards.
1-1
The responsible standards bodies may revise these standards from time to time. The
versions of the applicable standards to which these compilers conform may be
revised or replaced, resulting in features in future releases of the Sun Fortran
compilers that create incompatibilities with earlier releases.
Features of the Fortran Compilers
Sun Fortran compilers provide the following features or extensions:
4 f77: Global program checking across routines for consistency of arguments,
commons, parameters, and the like
4 SPARC: Multiprocessor support, including automatic and explicit loop
parallelization tightly integrated with optimization
TM
Note - Parallelization features
of the Fortran compilers are available only with Sun
TM
Performance WorkShop .
4 f77: Many VAX/VMS Fortran extensions, including:
4
4
4
4
4
4
4
4
4
NAMELIST
DO WHILE
Structures, records, unions, maps
Variable format expressions
Recursion
Pointers
Double-precision complex
SPARC: Quadruple-precision real
SPARC: Quadruple-precision complex
4 Cray-style parallelization directives, including TASKCOMMON, with extensions on
f90.
4 Global, peephole, and potential parallelization optimizations produce
high-performance applications. Benchmarks show that optimized applications can
run significantly faster when compared to unoptimized code.
4 Common calling conventions on Solaris systems permit routines written in C or
C++ to be combined with Fortran programs.
4 Support for 64-bit Solaris 7 environments on UltraSPARC processors.
4 Fortran 95 features in f90 include the attributes PURE and ELEMENTAL, and
enhanced forms of MAXVAL and MINVAL.
4 Call-by-value, %VAL, implemented in both f77 and f90.
1-2
Fortran Programming Guide ♦ Revision A, February 1999
4 Interoperability between FORTRAN 77 and Fortran 90 programs and object
binaries.
Introduction 1-3
1-4
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
2
Fortran Input/Output
This chapter discusses the input/output features provided by Sun Fortran compilers.
Accessing Files From Within Fortran
Programs
Data is transferred between the program and devices or files through a Fortran
logical unit. Logical units are identified in an I/O statement by a logical unit number,
a nonnegative integer from 0 to the maximum 4-byte integer value (2,147,483,647).
The character * can appear as a logical unit identifier. The asterisk stands for
standard input file when it appears in a READ statement; it stands for standard output
file when it appears in a WRITE or PRINT statement.
A Fortran logical unit can be associated with a specific, named file through the OPEN
statement. Also, certain "preconnected" units are automatically associated with
specific files at the start of program execution.
Accessing Named Files
The OPEN statement’s FILE= specifier establishes the association of a logical unit to a
named, physical file at runtime. This file can be pre-existing or created by the
program. See the Sun FORTRAN 77 Language Reference Manual for a full discussion
of the OPEN statement.
The FILE= specifier on an OPEN statement may specify a simple file name
(FILE=’myfile.out’) or a file name preceded by an absolute or relative directory
2-1
path (FILE=’../Amber/Qproj/myfile.out’). Also, the specifier may be a
character constant, variable, or character expression.
Library routines can be used to bring command-line arguments and environment
variables into the program as character variables for use as file names in OPEN
statements. (See man page entries for getarg(3F) and getenv(3F) for details; these and
other useful library routines are also described in the Fortran Library Reference).
The following example (GetFilNam.f) shows one way to construct an absolute path
file name from a typed-in name. The program uses the library routines GETENV,
LNBLNK, and GETCWD to return the value of the $HOME environment variable, find
the last non-blank in the string, and determine the current working directory:
CHARACTER F*128, FN*128, FULLNAME*128
PRINT*, "ENTER FILE NAME:"
READ *, F
FN = FULLNAME( F )
PRINT *, "PATH IS: ",FN
END
C
C
C
C
C
&
&
CHARACTER*128 FUNCTION FULLNAME( NAME )
CHARACTER NAME*(*), PREFIX*128
This assumes C shell.
Leave absolute path names unchanged.
If name starts with "~/", replace tilde with home
directory; otherwise prefix relative path name with
path to current directory.
IF ( NAME(1:1) .EQ. "/" ) THEN
FULLNAME = NAME
ELSE IF ( NAME(1:2) .EQ. "~/" ) THEN
CALL GETENV( "HOME", PREFIX )
FULLNAME = PREFIX(:LNBLNK(PREFIX)) //
NAME(2:LNBLNK(NAME))
ELSE
CALL GETCWD( PREFIX )
FULLNAME = PREFIX(:LNBLNK(PREFIX)) //
"/" // NAME(:LNBLNK(NAME))
ENDIF
RETURN
END
Compiling and running GetFilNam.f results in:
demo% pwd
/home/users/auser/subdir
demo% f77 -silent -o getfil GetFilNam.f
demo% getfil
anyfile
/home/users/auser/subdir/anyfile
demo%
2-2
Fortran Programming Guide ♦ Revision A, February 1999
Opening Files Without a Name
The OPEN statement need not specify a name; the runtime system supplies a file
name according to several conventions.
Opened as Scratch
Specifying STATUS=’SCRATCH’ in the OPEN statement opens a file with a name of
the form tmp.FAAAxnnnnn, where nnnnn is replaced by the current process ID, AAA
is a string of three characters, and x is a letter; the AAA and x make the file name
unique. This file is deleted upon termination of the program or execution of a CLOSE
statement, unless (with f77) STATUS=’KEEP’ is specified in the CLOSE statement.
Already Open
If the file has already been opened by the program, you can use a subsequent OPEN
statement to change some of the file’s characteristics; for example, BLANK and FORM.
In this case, you would specify only the file’s logical unit number and the parameters
to change.
Preconnected Units
Three unit numbers are automatically associated with specific standard I/O files at
the start of program execution. These preconnected units are standard input, standard
output, and standard error:
4 Standard input is logical unit 5 (also Fortran 90 unit 100)
4 Standard output is logical unit 6 (also Fortran 90 unit 101)
4 Standard error is logical unit 0 (also Fortran 90 unit 102)
Typically, standard input receives input from the workstation keyboard; standard
output and standard error display output on the workstation screen.
In all other cases where a logical unit number but no FILE= name is specified on an
OPEN statement, a file is opened with a name of the form fort.n, where n is the
logical unit number.
Opening Files Without an OPEN Statement
Use of the OPEN statement is optional in those cases where default conventions can
be assumed. If the first operation on a logical unit is an I/O statement other than
OPEN or INQUIRE, the file fort.n is referenced, where n is the logical unit number
(except for 0, 5, and 6, which have special meaning).
Fortran Input/Output
2-3
These files need not exist before program execution. If the first operation on the file
is not an OPEN or INQUIRE statement, they are created.
Example: The WRITE in the following code creates the file fort.25 if it is the first
input/output operation on that unit:
demo% cat TestUnit.f
IU=25
WRITE( IU, "(I4)" ) IU
END
demo%
The preceding program opens the file fort.25 and writes a single formatted record
onto that file:
demo% f77 -silent -o testunit TestUnit.f
demo% testunit
demo% cat fort.25
25
demo%
Passing File Names to Programs
The file system does not have any automatic facility to associate a logical unit
number in a Fortran program with a physical file.
However, there are several satisfactory ways to communicate file names to a Fortran
program.
Via Runtime Arguments and GETARG
The library routine getarg(3F) can be used to read the command-line arguments at
runtime into a character variable. The argument is interpreted as a file name and
used in the OPEN statement FILE= specifier:
demo% cat testarg.f
CHARACTER outfile*40
C Get first arg as output file name for unit 51
CALL getarg(1,outfile)
OPEN(51,FILE=outfile)
WRITE(51,*) "Writing to file: ", outfile
END
demo% f77 -silent -o tstarg testarg.f
demo% tstarg AnyFileName
demo% cat AnyFileName
Writing to file: AnyFileName
demo%
2-4
Fortran Programming Guide ♦ Revision A, February 1999
Via Environment Variables and GETENV
Similarly, the library routine getenv(3F) can be used to read the value of any
environment variable at runtime into a character variable that in turn is interpreted
as a file name:
demo% cat testenv.f
CHARACTER outfile*40
C Get $OUTFILE as output file name for unit 51
CALL getenv("OUTFILE",outfile)
OPEN(51,FILE=outfile)
WRITE(51,*) "Writing to file: ", outfile
END
demo% f77 -silent -o tstenv testenv.f
demo% setenv OUTFILE EnvFileName
demo% tstenv
demo% cat EnvFileName
Writing to file: EnvFileName
demo%
When using getarg or getenv, care should be taken regarding leading or trailing
blanks. (FORTRAN 77 programs can use the library function LNBLNK; Fortran 90
programs can use the intrinsic function TRIM.) Additional flexibility to accept relative
path names can be programmed along the lines of the FULLNAME function in the
example at the beginning of this chapter.
f77: Logical Unit Preattachment Using IOINIT
The library routine IOINIT can also be used with f77 to attach logical units to
specific files at runtime. IOINIT looks in the environment for names of a
user-specified form and then opens the corresponding logical unit for sequential
formatted I/O. Names must be of the general form PREFIXnn, where the particular
PREFIX is specified in the call to IOINIT, and nn is the logical unit to be opened.
Unit numbers less than 10 must include the leading 0. See the Sun Fortran Library
Reference, and the IOINIT(3F) man page. (The IOINIT facility is not implemented for
f90.)
Example: Associate physical files test.inp and test.out in the current directory
to logical units 1 and 2:
First, set the environment variables.
With ksh or sh:
demo$ TST01=ini1.inp
demo$ TST02=ini1.out
demo$ export TST01 TST02
Fortran Input/Output
2-5
With csh:
demo% setenv TST01 ini1.inp
demo% setenv TST02 ini1.out
demo% cat ini1.f
CHARACTER PRFX*8
LOGICAL CCTL, BZRO, APND, VRBOSE
DATA CCTL, BZRO, APND, PRFX, VRBOSE
&
/.TRUE.,.FALSE.,.FALSE., "TST",.FALSE. /
CALL IOINIT( CCTL, BZRO, APND, PRFX, VRBOSE )
READ(1, *) I, B, N
WRITE(2, *) I, B, N
END
demo%
The program ini1.f reads 1 and writes 2:
With environment variables and ioinit, ini1.f reads ini1.inp and writes to
ini1.out:
demo% cat ini1.inp
12 3.14159012 6
demo% f77 -silent -o tstinit ini1.f
demo% tstinit
demo% cat ini1.out
12
3.14159 6
demo%
IOINIT is adequate for most programs as written. However, it is written in Fortran
specifically to serve as an example for similar user-supplied routines. Retrieve a copy
from the following file, a part of the FORTRAN 77 package installation:
/opt/SUNWspro/SC5.0/src/ioinit.f
Command-Line I/O Redirection and Piping
Another way to associate a physical file with a program’s logical unit number is by
redirecting or piping the preconnected standard I/O files. Redirection or piping
occurs on the runtime execution command.
In this way, a program that reads standard input (unit 5) and writes to standard
output (unit 6) or standard error (unit 0) can, by redirection (using
<, >, >>, >&, |, |&, 2>, 2>&1 on the command line), read or write to any
other named file.
This is shown in the following table:
2-6
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 2–1
csh/sh/ksh Redirection and Piping on the Command Line
Action
Using C Shell
Using Bourne or Korn Shell
Standard input
—read from
mydata
myprog < mydata
myprog < mydata
Standard output
—write
(overwrite)
myoutput
myprog > myoutput
myprog > myoutput
Standard output
— write/
append to
myoutput
myprog >> myoutput
myprog >> myoutput
Redirect
standard error
to a file
myprog >& errorfile
myprog 2> errorfile
Pipe standard
output to input
of another
program
myprog1 | myprog2
myprog1 | myprog2
Pipe standard
error and
output to
another
program
myprog1 |& myprog2
myprog1 2>&1 | myprog2
See the csh, ksh,and sh man pages for details on redirection and piping on the
command line.
f77: VAX / VMS Logical File Names
If you are porting from VMS FORTRAN to FORTRAN 77, the VMS-style logical file
names in the INCLUDE statement are mapped to UNIX path names. The environment
variable LOGICALNAMEMAPPING defines the mapping between the logical names and
the UNIX path name. If the environment variable LOGICALNAMEMAPPING is set and
the -vax, -xl or -xld compiler options are used, the compiler interprets VMS
logical file names on the INCLUDE statement.
The compiler sets the environment variable to a string with the following syntax:
"lname1=path1; lname2=path2;
"
Fortran Input/Output
2-7
Each lname is a logical name, and each path is the path name of a directory (without
a trailing /). All blanks are ignored when parsing this string. Any trailing /list or
/nolist is stripped from the file name in the INCLUDE statement. Logical names in
a file name are delimited by the first colon in the VMS file name. The compiler
converts file names of the form:
lname1:file
to:
path1/file
Uppercase and lowercase are significant in logical names. If a logical name is
encountered on the INCLUDE statement that was not specified by
LOGICALNAMEMAPPING, the file name is used unchanged.
Direct I/O
Direct or random I/O allows you to access a file directly by record number. Record
numbers are assigned when a record is written. Unlike sequential I/O, direct I/O
records can be read and written in any order. However, in a direct access file, all
records must be the same fixed length. Direct access files are declared with the
ACCESS=’DIRECT’ specifier on the OPEN statement for the file.
A logical record in a direct access file is a string of bytes of a length specified by the
OPEN statement’s RECL= specifier. READ and WRITE statements must not specify
logical records larger than the defined record size. (Record sizes are specified in
bytes.) Shorter records are allowed. Unformatted, direct writes leave the unfilled part
of the record undefined. Formatted, direct writes cause the unfilled record to be
padded with blanks.
Direct access READ and WRITE statements have an extra argument, REC=n, to specify
the record number to be read or written.
Example: Direct access, unformatted:
&
2-8
OPEN( 2, FILE="data.db", ACCESS="DIRECT", RECL=200,
FORM="UNFORMATTED", ERR=90 )
READ( 2, REC=13, ERR=30 ) X, Y
Fortran Programming Guide ♦ Revision A, February 1999
This program opens a file for direct access, unformatted I/O, with a fixed record
length of 200 bytes, then reads the thirteenth record into X and Y.
Example: Direct access, formatted:
&
OPEN( 2, FILE="inven.db", ACCESS="DIRECT", RECL=200,
FORM="FORMATTED", ERR=90 )
READ( 2, FMT="(I10,F10.3)", REC=13, ERR=30 ) X, Y
This program opens a file for direct access, formatted I/O, with a fixed record length
of 200 bytes. It then reads the thirteenth record and converts it with the format
(I10,F10.3).
For formatted files, the size of the record written is determined by the FORMAT
statement. In the preceding example, the FORMAT statement defines a record of 20
characters or bytes. More than one record can be written by a single formatted write
if the amount of data on the list is larger than the record size specified in the FORMAT
statement. In such a case, each subsequent record is given successive record numbers.
Example: Direct access, formatted, multiple record write:
OPEN( 21, ACCESS="DIRECT", RECL=200, FORM="FORMATTED")
WRITE(21,"(10F10.3)",REC=11) (X(J),J=1,100)
The write to direct access unit 21 creates 10 records of 10 elements each (since the
format specifies 10 elements per record) these records are numbered 11 through 20.
Internal Files
An internal file is an object of type CHARACTER such as a variable, substring, array,
element of an array, or field of a structured record. Internal file READ can be from a
constant character string. I/O on internal files simulates formatted READ and WRITE
statements by transferring and converting data from one character object to another
data object. No file I/O is performed.
When using internal files:
4 The name of the character object receiving the data appears in place of the unit
number on a WRITE statement. On a READ statement, the name of the character
object source appears in place of the unit number.
4 A constant, variable, or substring object constitutes a single record in the file.
4 With an array object, each array element corresponds to a record.
4 f77: f77 extends direct I/O to internal files. (The ANSI standard includes only
sequential formatted I/O on internal files.) This is similar to direct I/O on external
Fortran Input/Output
2-9
files, except that the number of records in the file cannot be changed. In this case,
a record is a single element of an array of character strings.
4 Each sequential READ or WRITE statement starts at the beginning of an internal
file.
Example: Sequential formatted read from an internal file (one record only):
demo% cat intern1.f
CHARACTER X*80
READ( *, "(A)" ) X
READ( X, "(I3,I4)" ) N1, N2 ! This codeline reads the internal file X
WRITE( *, * ) N1, N2
END
demo% f77 -silent -o tstintern intern1.f
demo% tstintern
12 99
12 99
demo%
Example: Sequential formatted read from an internal file (three records):
demo% cat intern2.f
CHARACTER LINE(4)*16
! This is our "internal file"
*
12341234
DATA LINE(1) / " 81 81 " /
DATA LINE(2) / " 82 82 " /
DATA LINE(3) / " 83 83 " /
DATA LINE(4) / " 84 84 " /
READ( LINE,"(2I4)") I,J,K,L,M,N
PRINT *, I, J, K, L, M, N
END
demo% f77 -silent intern2.f
demo% a.out
81 81 82 82 83 83
demo%
Example: Direct access read from an internal file (one record) (f77 only):
demo% cat intern3.f
CHARACTER LINE(4)*16
! This is our "internal file"
*
12341234
DATA LINE(1) / " 81 81 " /
DATA LINE(2) / " 82 82 " /
DATA LINE(3) / " 83 83 " /
DATA LINE(4) / " 84 84 " /
READ ( LINE, FMT=20, REC=3 ) M, N
20
FORMAT( I4, I4 )
PRINT *, M, N
END
demo% f77 -silent intern3.f
demo% a.out
83 83
demo%
2-10
Fortran Programming Guide ♦ Revision A, February 1999
Tape I/O (f77 Only)
Most typical Fortran I/O is done to disk files. However, by associating a logical unit
number to a physically mounted tape drive via the OPEN statement, it is possible to
do I/O directly to tape.
It could be more efficient to use the TOPEN() routines rather than Fortran I/O
statements to do I/O on magnetic tape.
Using TOPEN Routines
With the nonstandard tape I/O package (see topen(3F)) you can transfer blocks
between the tape drive and buffers declared as Fortran character variables. You can
then use internal I/O to fill and empty these buffers. This facility does not integrate
with the rest of Fortran I/O and even has its own set of tape logical units. Refer to
the man pages for complete information.
Fortran Formatted I/O for Tape
The Fortran I/O statements provide facilities for transparent access to formatted,
sequential files on magnetic tape. There is no limit on formatted record size, and
records may span tape blocks.
Fortran Unformatted I/O for Tape
Using the Fortran I/O statements to connect a magnetic tape for unformatted access is
less satisfactory. The implementation of unformatted records implies that the size of
a record (plus eight characters of overhead) cannot be bigger than the buffer size.
As long as this restriction is complied with, the I/O system does not write records
that span physical tape blocks, writing short blocks when necessary. This
representation of unformatted records is preserved (even though it is inappropriate
for tapes) so that files can be freely copied between disk and tapes.
Since the block-spanning restriction does not apply to tape reads, files can be copied
from tape to disk without any special considerations.
Fortran Input/Output
2-11
Tape File Representation
A Fortran data file is represented on tape by a sequence of data records followed by
an endfile record. The data is grouped into blocks, with maximum block size
determined when the file is opened. The records are represented in the same way as
records in disk files: formatted records are followed by newlines; unformatted
records are preceded and followed by character counts. In general, there is no
relation between Fortran records and tape blocks; that is, records can span blocks,
which can contain parts of several records.
The only exception is that Fortran does not write an unformatted record that spans
blocks; thus, the size of the largest unformatted record is eight characters less than
the block size.
The dd Conversion Utility
An end-of-file record in Fortran maps directly into a tape mark. In this respect,
Fortran files are the same as tape system files. But since the representation of Fortran
files on tape is the same as that used in the rest of UNIX, naive Fortran programs
cannot read 80-column card images on tape. If you have an existing Fortran program
and an existing data tape to read with it, translate the tape using the dd(1) utility,
which adds newlines and strips trailing blanks.
Example: Convert a tape on mt0 and pipe that to the executable ftnprg:
demo% dd if=/dev/rmt0 ibs=20b cbs=80 conv=unblock | ftnprg
The getc Library Routine
As an alternative to dd, you can call the getc(3F) library routine to read characters
from the tape. You can then combine the characters into a character variable and use
internal I/O to transfer formatted data. See also TOPEN(3F).
End-of-File
The end-of-file condition is reached when an end-of-file record is encountered during
execution of a READ statement. The standard states that the file is positioned after
the end-of-file record. In real life, this means that the tape read head is poised at the
beginning of the next file on the tape. Although it seems as if you could read the
next file on the tape, this is not strictly true, and is not covered by the ANSI
FORTRAN 77 Language Standard.
The standard also says that a BACKSPACE or REWIND statement can be used to
reposition the file. Consequently, after reaching end-of-file, you can backspace over
2-12
Fortran Programming Guide ♦ Revision A, February 1999
the end-of-file record and further manipulate the file–for example, writing more
records at the end, rewinding the file, and rereading or rewriting it.
Multifile Tapes
The name used to open the tape file determines certain characteristics of the
connection, such as the recording density and whether the tape is automatically
rewound when opened and closed.
To access a file on a tape with multiple files, first use the mt(1) utility to position the
tape to the needed file. Then open the file as a no-rewind magnetic tape such as
/dev/nrmt0. Referencing the tape with this name prevents it from being
repositioned when it is closed. By reading the file until end-of-file and then
reopening it, a program can access the next file on the tape. Any program
subsequently referencing the same tape can access it where it was last left, preferably
at the beginning of a file, or past the end-of-file record.
However, if your program terminates prematurely, it may leave the tape positioned
anywhere. Use the SunOS mt(1) command to reposition the tape appropriately.
Fortran 90 I/O Considerations
Fortran 90 2.0 and FORTRAN 77 5.0 are I/O compatible. Executables containing
intermixed f77 and f90 compilations can do I/O to the same unit from both the
f77 and f90 parts of the program.
However, Fortran 90 provides some additional features:
4 A file opened with FORM=’BINARY’ enables nonstandard I/O of raw data
without record marks. This has nearly the same effect as FORM=’UNFORMATTED’,
except that no record lengths are embedded in the file. As a result, it is not
possible to backspace a FORM=’BINARY’ file. A READ statement on such a
’BINARY’ file reads as much data as needed to fill all the variables on the input
list.
4 ADVANCE=’NO’ enables nonadvancing I/O, as in:
write(*,"(a)",ADVANCE="NO")
read(*,*) n
"Enter size= "
4 NAMELIST input features:
4
f90 allows the group name to be preceded by $ or & on input. The Fortran 90
standard accepts only & and this is what a NAMELIST write outputs.
Fortran Input/Output
2-13
4 f90 accepts $ as the symbol terminating an input group unless the last data
item in the group is CHARACTER, in which case the $ is treated as input data.
4 f90 allows NAMELIST input to start in the first column of a record.
4 ENCODE and DECODE are recognized and implemented by f90 just as they are by
f77.
2-14
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
3
Program Development
This chapter briefly introduces two powerful program development tools, make and
SCCS, that can be used very successfully with Fortran programs.
A number of good, commercially published books on using make and SCCS are
currently available, including Managing Projects with make, by Andrew Oram and
Steve Talbott, and Applying RCS and SCCS, by Don Bolinger and Tan Bronson. Both
are from O’Reilly & Associates.
Facilitating Program Builds With the
make Utility
The make utility applies intelligence to the task of program compilation and linking.
Typically, a large application consists of a set of source files and INCLUDE files,
requiring linking with a number of libraries. Modifying any one or more of the
source files requires recompilation of that part of the program and relinking. You can
automate this process by specifying the interdependencies between files that make
up the application along with the commands needed to recompile and relink each
piece. With these specifications in a file of directives, make ensures that only the files
that need recompiling are recompiled and that relinking uses the options and
libraries you need to build the executable. The following discussion provides a
simple example of how to use make. For a summary, see make(1).
3-1
The Makefile
A file called makefile tells make in a structured manner which source and object
files depend on other files. It also defines the commands required to compile and
link the files.
For example, suppose you have a program of four source files and the makefile:
demo% ls
makefile
commonblock
computepts.f
pattern.f
startupcore.f
demo%
Assume both pattern.f and computepts.f have an INCLUDE of commonblock,
and you wish to compile each.f file and link the three relocatable files, along with a
series of libraries, into a program called pattern.
The makefile looks like this:
demo% cat makefile
pattern: pattern.o computepts.o startupcore.o
f77 pattern.o computepts.o startupcore.o --lcore77 \
--lcore --lsunwindow --lpixrect --o pattern
pattern.o: pattern.f commonblock
f77 --c --u pattern.f
computepts.o: computepts.f commonblock
f77 --c --u computepts.f
startupcore.o: startupcore.f
f77 --c --u startupcore.f
demo%
The first line of makefile indicates that making pattern depends on pattern.o,
computepts.o, and startupcore.o. The next line and its continuations give the
command for making pattern from the relocatable.o files and libraries.
Each entry in makefile is a rule expressing a target object’s dependencies and the
commands needed to make that object. The structure of a rule is:
target: dependencies-listTAB build-commands
4 Dependencies. Each entry starts with a line that names the target file, followed by
all the files the target depends on.
4 Commands. Each entry has one or more subsequent lines that specify the Bourne
shell commands that will build the target file for this entry. Each of these
command lines must be indented by a tab character.
3-2
Fortran Programming Guide ♦ Revision A, February 1999
make Command
The make command can be invoked with no arguments, simply:
demo% make
The make utility looks for a file named makefile or Makefile in the current
directory and takes its instructions from that file.
The make utility:
4 Reads makefile to determine all the target files it must process, the files they
depend on, and the commands needed to build them.
4 Finds the date and time each file was last changed.
4 Rebuilds any target file that is older than any of the files it depends on, using the
commands from makefile for that target.
Macros
The make utility’s macro facility allows simple, parameterless string substitutions. For
example, the list of relocatable files that make up the target program pattern can
be expressed as a single macro string, making it easier to change.
A macro string definition has the form:
NAME = string
Use of a macro string is indicated by:
$(NAME)
which is replaced by make with the actual value of the macro string.
This example adds a macro definition naming all the object files to the beginning of
makefile:
OBJ = pattern.o computepts.o startupcore.o
Now the macro can be used in both the list of dependencies as well as on the f77
link command for target pattern in makefile:
pattern: $(OBJ)
f77 $(OBJ) --lcore77 --lcore --lsunwindow \
--lpixrect --o pattern
For macro strings with single-letter names, the parentheses may be omitted.
Program Development
3-3
Overriding of Macro Values
The initial values of make macros can be overridden with command-line options to
make. For example:
FFLAGS=--u
OBJ = pattern.o computepts.o startupcore.o
pattern: $(OBJ)
f77 $(FFLAGS) $(OBJ) --lcore77 --lcore --lsunwindow \
--lpixrect --o pattern
pattern.o: pattern.f commonblock
f77 $(FFLAGS) -c pattern.f
computepts.o:
f77 $(FFLAGS) --c computepts.f
Now a simple make command without arguments uses the value of FFLAGS set
above. However, this can be overridden from the command line:
demo% make "FFLAGS=-u -O"
Here, the definition of the FFLAGS macro on the make command line overrides the
makefile initialization, and both the -O flag and the -u flag are passed to f77.
Note that "FFLAGS=" can also be used on the command to reset the macro to a null
string so that it has no effect.
Suffix Rules in make
To make writing a makefile easier, make will use its own default rules depending on
the suffix of a target file. Recognizing the .f suffix, make uses the f77 compiler,
passing as arguments any flags specified by the FFLAGS macro, the -c flag, and the
name of the source file to be compiled.
The example below demonstrates this rule twice:
OBJ = pattern.o computepts.o startupcore.o
FFLAGS=--u
pattern: $(OBJ)
f77 $(OBJ) --lcore77 --lcore --lsunwindow \
--lpixrect --o pattern
pattern.o: pattern.f commonblock
f77 $(FFLAGS) --c pattern.f
computepts.o: computepts.f commonblock
startupcore.o: startupcore.f
make uses default rules to compile computepts.f and startupcore.f.
Similarly, suffix rules for .f90 files will also invoke the f90 compiler.
3-4
Fortran Programming Guide ♦ Revision A, February 1999
Version Tracking and Control With SCCS
SCCS stands for Source Code Control System. SCCS provides a way to:
4 Keep track of the evolution of a source file—its change history
4 Prevent a source file from being simultaneously changed by other developers
4 Keep track of the version number by providing version stamps
The basic three operations of SCCS are:
4 Putting files under SCCS control
4 Checking out a file for editing
4 Checking in a file
This section shows you how to use SCCS to perform these tasks, using the previous
program as an example. Only basic SCCS is described and only three SCCS
commands are introduced: create, edit, and delget.
Controlling Files With SCCS
Putting files under SCCS control involves:
4 Making the SCCS directory
4 Inserting SCCS ID keywords into the files (this is optional)
4 Creating the SCCS files
Making the SCCS Directory
To begin, you must create the SCCS subdirectory in the directory in which your
program is being developed. Use this command:
demo% mkdir SCCS
SCCS must be in uppercase.
Inserting SCCS ID Keywords
Some developers put one or more SCCS ID keywords into each file, but that is
optional. These keywords are later identified with a version number each time the
files are checked in with an SCCS get or delget command. There are three likely
places to put these strings:
Program Development
3-5
4 Comment lines
4 Parameter statements
4 Initialized data
The advantage of using keywords is that the version information appears in the
source listing and compiled object program. If preceded by the string @(#), the
keywords in the object file can be printed using the what command.
Included header files that contain only parameter and data definition statements do
not generate any initialized data, so the keywords for those files usually are put in
comments or in parameter statements. In some files, like ASCII data files or
makefiles, the SCCS information will appear in comments.
SCCS keywords appear in the form %keyword% and are expanded into their values
by the SCCS get command. The most commonly used keywords are:
%Z% expands to the identifier string @(#) recognized by the what command. %M%
expands to the name of the source file. %I% expands to the version number of this
SCCS maintained file. %E% expands to the current date.
For example, you could identify the makefile with a make comment containing these
keywords:
#
%Z%%M%
%I%
%E%
The source files, startupcore.f, computepts.f, and pattern.f, can be
identified by initialized data of the form:
CHARACTER*50 SCCSID
DATA SCCSID/"%Z%%M%
%I%
%E%\n"/
When this file is processed by SCCS, compiled, and the object file processed by the
SCCS what command, the following is displayed:
demo% f77 -c pattern.f
...
demo% what pattern
pattern:
pattern.f 1.2 96/06/10
You can also create a PARAMETER named CTIME that is automatically updated
whenever the file is accessed with get.
CHARACTER*(*) CTIME
PARAMETER ( CTIME="%E%")
INCLUDE files can be annotated with a Fortran comment containing the SCCS stamp:
3-6
Fortran Programming Guide ♦ Revision A, February 1999
C
%Z%%M%
%I%
%E%
Creating SCCS Files
Now you can put these files under control of SCCS with the SCCS create command:
demo% sccs create makefile commonblock startupcore.f \
computepts.f pattern.f
demo%
Checking Files Out and In
Once your source code is under SCCS control, you use SCCS for two main tasks: to
check out a file so that you can edit it, and to check in a file you have finished editing.
Check out a file with the sccs edit command. For example:
demo% sccs edit computepts.f
SCCS then makes a writable copy of computepts.f in the current directory, and
records your login name. Other users cannot check the file out while you have it
checked out, but they can find out who has checked it out.
When you have completed your editing, check in the modified file with the sccs
delget command. For example:
demo% sccs delget computepts.f
This command causes the SCCS system to:
4 Make sure that you are the user who checked out the file by comparing login
names
4 Prompt for a comment from you on the changes
4 Make a record of what was changed in this editing session
4 Delete the writable copy of computepts.f from the current directory
4 Replace it by a read-only copy with the SCCS keywords expanded
The sccs delget command is a composite of two simpler SCCS commands, delta
and get. The delta command performs the first three tasks in the list above; the
get command performs the last two tasks.
Program Development
3-7
3-8
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
4
Libraries
This chapter describes how to use and create libraries of subprograms. Both static
and dynamic libraries are discussed.
Understanding Libraries
A software library is usually a set of subprograms that have been previously
compiled and organized into a single binary library file. Each member of the set is
called a library element or module. The linker searches the library files, loading object
modules referenced by the user program while building the executable binary
program. See ld(1) and the Solaris Linker and Libraries Guide for details.
There are two basic kinds of software libraries:
4 Static library. A library in which modules are bound into the executable file before
execution. Static libraries are commonly named libname.a. The .a suffix refers to
archive.
4 Dynamic library. A library in which modules can be bound into the executable
program at runtime. Dynamic libraries are commonly named libname.so. The
.so suffix refers to shared object.
Typical system libraries that have both static (.a) and dynamic (.so) versions are:
4 FORTRAN 77 libraries: libF77, libM77
4 Fortran 90 libraries: libfsu, libfui, libfai, libfai2, libfsunai,
libfprodai, libfminlai, libfmaxlai, libminvai, libmaxvai,
libf77compat
4 VMS Fortran libraries: libV77
4 C libraries: libc
4-1
There are two advantages to the use of libraries:
4 There is no need to have source code for the library routines that a program calls.
4 Only the needed modules are loaded.
Library files provide an easy way for programs to share commonly used subroutines.
You need only name the library when linking the program, and those library modules
that resolve references in the program are linked and merged into the executable file.
Specifying Linker Debugging Options
Summary information about library usage and loading can be obtained by passing
additional options to the linker through the LD_OPTIONS environment variable. The
compiler calls the linker with these options (and others it requires) when generating
object binary files.
Using the compiler to call the linker is always recommended over calling the linker
directly because many compiler options require specific linker options or library
references, and linking without these could produce unpredictable results.
demo% setenv LD_OPTIONS ’-m -Dfiles’
demo% f77 -o myprog myprog.f
Example: Using LD_OPTIONS to create a load map:
Some linker options do have compiler command-line equivalents that can appear
directly on the f77 or f90 command. These include
-Bx, -dx, -G, -hname, -Rpath, and -ztext. See the f77(1) and f90(1) man
pages or the Fortran User’s Guide for details.
More detailed examples and explanations of linker options and environment
variables can be found in the Solaris Linker and Libraries Guide.
Generating a Load Map
The linker -m option generates a load map that displays library linking information.
The routines linked during the building of the executable binary program are listed
together with the libraries that they come from.
Example: Using -m to generate a load map:
demo% setenv LD_OPTIONS ’-m’
demo% f77 any.f
any.f:
MAIN:
LINK EDITOR MEMORY MAP
4-2
Fortran Programming Guide ♦ Revision A, February 1999
output
section
.interp
.hash
.dynsym
.dynstr
.text
.text
.text
.text
.text
...
input
section
virtual
address
100d4
.interp 100d4
100e8
.hash
100e8
103d0
.dynsym 103d0
10a20
.dynstr 10a20
10c90
10c90
10c90
10d84
10d88
size
11
11 (null)
2e8
2e8 (null)
650
650 (null)
366
366 (null)
1e70
00 /opt/SUNWspro/SC5.0/lib/crti.o
f4 /opt/SUNWspro/SC5.0/lib/crt1.o
00 /opt/SUNWspro/SC5.0/lib/values-xi.o
d20 sparse.o
Listing Other Information
Additional linker debugging features are available through the linker’s -Dkeyword
option. A complete list can be displayed using -Dhelp.
Example: List linker debugging aid options using the -Dhelp option:
demo%
ld -Dhelp
…
debug: args
debug: bindings
debug: detail
debug: entry
…
demo%
display
display
provide
display
input argument processing
symbol binding;
more information
entrance criteria descriptors
For example, the -Dfiles linker option lists all the files and libraries referenced
during the link process:
demo% setenv LD_OPTIONS ’-Dfiles’
demo% f77 direct.f
direct.f:
MAIN direct:
debug: file=/opt/SUNWspro/SC5.0/lib/crti.o [ ET_REL ]
debug: file=/opt/SUNWspro/SC5.0/lib/crt1.o [ ET_REL ]
debug: file=/opt/SUNWspro/SC5.0/lib/values--xi.o [ ET_REL ]
debug: file=direct.o [ ET_REL ]
debug: file=/opt/SUNWspro/SC5.0/lib/libM77.a [ archive ]
debug: file=/opt/SUNWspro/lib/libF77.so [ ET_DYN ]
debug: file=/opt/SUNWspro/SC5.0/lib/libsunmath.a [ archive ]
…
Libraries 4-3
See the Linker and Libraries Guide for further information on these linker options.
Consistent Compiling and Linking
Ensuring a consistent choice of compiling and linking options is critical whenever
compilation and linking are done in separate steps. Compiling any part of a program
with any of the following options requires linking with the same options:
-a, -autopar, -Bx, -fast, -G, -Lpath, -lname,-mt, -nolib, -norunpath, -p, -pg, -xlibmopt
Example: Compiling sbr.f with -a and smain.f without it, then linking in
separate steps (-a invokes tcov old style profiling):
demo% f77 -c -a sbr.f
demo% f77 -c smain.f
demo% f77 -a sbr.o smain.o
{link step; pass -a to the linker}
Also, a number of options require that all source files be compiled with that option.
These include:
-autopar, -cg92, -dx, -dalign, -dbl, -explicitpar, -f, -misalign, -native, -parallel,
See the f77(1) and f90(1) man pages and the Fortran User’s Guide for details on all
compiler options.
Setting Library Search Paths and Order
The linker searches for libraries at several locations and in a certain prescribed order.
Some of these locations are standard paths, while others depend on the compiler
options -Rpath, -llibrary, and -Ldir and the environment variable
LD_LIBRARY_PATH.
Search Order for Standard Library Paths
The standard library search paths used by the linker are determined by the
installation path, and they differ for static and dynamic loading. <install-point> is the
path to where the Fortran compilers have been installed. In a standard install of the
software this is /opt .
4-4
Fortran Programming Guide ♦ Revision A, February 1999
Static Linking
While building the executable file, the static linker searches for any libraries in the
following paths (among others), in the specified order:
<install-point>/SUNWspro/lib
Sun shared libraries
/usr/ccs/lib/
Standard location for SVr4 software
/usr/lib
Standard location for UNIX software
These are the default paths used by the linker.
Dynamic Linking
The dynamic linker searches for shared libraries at runtime, in the specified order:
4 Paths specified by user with -Rpath
4 <install-point>/SUNWspro/lib/
4 /usr/lib standard UNIX default
The search paths are built into the executable.
LD_LIBRARY_PATH Environment Variable
Use the LD_LIBRARY_PATH environment variable to specify directory paths that the
linker should search for libraries specified with the -llibrary option.
Multiple directories can be specified, separated by a colon. Typically, the
LD_LIBRARY_PATH variable contains two lists of colon-separated directories
separated by a semicolon:
dirlist1;dirlist2
The directories in dirlist1 are searched first, followed by any explicit -Ldir directories
specified on the command line, followed by dirlist2 and the standard directories.
That is, if the compiler is called with any number of occurrences of -L, as in:
f77 ... -Lpath1 ... -Lpathn ...
then the search order is:
dirlist1 path1 ... pathn dirlist2 standard_paths
When the LD_LIBRARY_PATH variable contains only one colon-separated list of
directories, it is interpreted as dirlist2.
Libraries 4-5
In the Solaris 7 operating environment, a similar environment variable,
LD_LIBRARY_PATH_64 can be used to override LD_LIBRARY_PATH when searching
for 64-bit dependencies. See the Solaris Linker and Libraries Guide and the ld(1) man
page for details.
4 On a 32-bit SPARC processor and linking with Solaris 7, LD_LIBRARY_PATH_64
is ignored.
4 If only LD_LIBRARY_PATH is defined, it is used for both 32-bit and 64-bit linking.
4 If both LD_LIBRARY_PATH and LD_LIBRARY_PATH_64 are defined, 32-bit linking
will be done using LD_LIBRARY_PATH, and 64-bit linking with
LD_LIBRARY_PATH_64.
Note - Use of the LD_LIBRARY_PATH environment variable with production
software is strongly discouraged. Although useful as a temporary mechanism for
influencing the runtime linker’s search path, any dynamic executable that can
reference this environment variable will have its search paths altered. You might
see unexpected results or a degradation in performance.
Library Search Path and Order—Static Linking
Use the -llibrary compiler option to name additional libraries for the linker to search
when resolving external references. For example, the option -lmylib adds the
library libmylib.so or libmylib.a to the search list.
The linker looks in the standard directory paths to find the additional libmylib
library. The -L option (and the LD_LIBRARY_PATH environment variable) creates a
list of paths that tell the linker where to look for libraries outside the standard paths.
Were libmylib.a in directory /home/proj/libs, then the option
-L/home/proj/libs would tell the linker where to look when building the
executable:
demo% f77 -o pgram part1.o part2.o -L/home/proj/libs -lmylib
Command-Line Order for -llibrary Options
For any particular unresolved reference, libraries are searched only once, and only
for symbols that are undefined at that point in the search. If you list more than one
library on the command line, then the libraries are searched in the order in which
they are found on the command line. Place -llibrary options as follows:
4 Place the -llibrary option after any .f, .for, .F, .f90, or .o files.
4 If you call functions in libx, and they reference functions in liby, then place -lx
before -ly.
4-6
Fortran Programming Guide ♦ Revision A, February 1999
Command-Line Order for -Ldir Options
The -Ldir option adds the dir directory path to the library search list. The linker
searches for libraries first in any directories specified by the -L options and then in
the standard directories. This option is useful only if it is placed preceding the
-llibrary options to which it applies.
Library Search Path and Order—Dynamic Linking
With dynamic libraries, changing the library search path and order of loading differs
from the static case. Actual linking takes place at runtime rather than build time.
Specifying Dynamic Libraries at Build Time
When building the executable file, the linker records the paths to shared libraries in
the executable itself. These search paths can be specified using the -Rpath option.
This is in contrast to the -Ldir option which indicates at buildtime where to find the
library specified by a -llibrary option, but does not record this path into the binary
executable.
The directory paths that were built in when the executable was created can be
viewed using the dump command.
Example: List the directory paths built into a.out:
demo% f77 program.f -R/home/proj/libs -L/home/proj/libs -lmylib
demo% dump -Lv a.out | grep RPATH
[5]
RPATH
/home/proj/libs:/opt/SUNWspro/lib
Specifying Dynamic Libraries at Runtime
At runtime, the linker determines where to find the dynamic libraries that an
executable needs from:
4 The value of LD_LIBRARY_PATH at runtime
4 The paths that had been specified by -R at the time the executable file was built
As noted earlier, use of LD_LIBRARY_PATH can have unexpected side-effects and is
not recommended.
Fixing Errors During Dynamic Linking
When the dynamic linker cannot locate a needed library, it issues this error message:
Libraries 4-7
ld.so: prog: fatal: libmylib.so: can’t open file:
The message indicates that the libraries are not where they are supposed to be.
Perhaps you specified paths to shared libraries when the executable was built, but
the libraries have subsequently been moved. For example, you might have built
a.out with your own dynamic libraries in /my/libs/, and then later moved the
libraries to another directory.
Use ldd to determine where the executable expects to find the libraries:
demo% ldd a.out
libsolib.so => /export/home/proj/libsolib.so
libF77.so.4 => /opt/SUNWspro/lib/libF77.so.4
libc.so.1 => /usr/lib/libc.so.1
libdl.so.1 => /usr/lib/libdl.so.1
If possible, move or copy the libraries into the proper directory or make a soft link to
the directory (using ln -s) in the directory that the linker is searching. Or, it could
be that LD_LIBRARY_PATH is not set correctly. Check that LD_LIBRARY_PATH
includes the path to the needed libraries at runtime.
Creating Static Libraries
Static library files are built from precompiled object files (.o files) using the ar(1)
utility.
The linker extracts from the library any elements whose entry points are referenced
within the program it is linking, such as a subprogram, entry name, or COMMON block
initialized in a BLOCKDATA subprogram. These extracted elements (routines) are
bound permanently into the a.out executable file generated by the linker.
Tradeoffs for Static Libraries
There are three main issues to keep in mind regarding static, as compared to
dynamic, libraries and linking:
4 Static libraries are more self-contained but less adaptable.
If you bind an a.out executable file statically, the library routines it needs become
part of the executable binary. However, if it becomes necessary to update a static
library routine bound into the a.out executable, the entire a.out file must be
relinked and regenerated to take advantage of the updated library. With dynamic
libraries, the library is not part of the a.out file and linking is done at runtime. To
4-8
Fortran Programming Guide ♦ Revision A, February 1999
take advantage of an updated dynamic library, all that is required is that the new
library be installed on the system.
4 The “elements” in a static library are individual compilation units, .o files.
Since a single compilation unit (a source file) can contain more than one
subprogram, these routines when compiled together become a single module in
the static library. This means that all the routines in the compilation unit are loaded
together into the a.out executable, even though only one of those subprograms
was actually called. This situation can be improved by optimizing the way library
routines are distributed into compilable source files. (Still, only those library
modules actually referenced by the program are loaded into the executable.)
4 Order matters when linking static libraries.
The linker processes its input files in the order in which they appear on the
command line—left to right. When the linker decides whether or not to load an
element from a library, its decision is determined by the library elements that it
has already processed. This order is not only dependent on the order of the
elements as they appear in the library file but also on the order in which the
libraries are specified on the compile command line.
Example: If the Fortran program is in two files, main.f and crunch.f, and only
the latter accesses a library, it is an error to reference that library before crunch.f
or crunch.o:
demo% f77 main.f -lmylibrary crunch.f -o myprog
(Incorrect)
demo% f77 main.f crunch.f -lmylibrary -o myprog
(Correct)
Creation of a Simple Static Library
Suppose that you can distribute all the routines in a program over a group of source
files and that these files are wholly contained in the subdirectory test_lib/.
Suppose further that the files are organized in such a way that they each contain a
single principal subprogram that would be called by the user program, along with
any “helper” routines that the subprogram might call but that are called from no
other routine in the library. Also, any helper routines called from more than one
library routine are gathered together into a single source file. This gives a reasonably
well-organized set of source and object files.
Assume that the name of each source file is taken from the name of the first routine
in the file, which in most cases is one of the principal files in the library:
demo% cd test_lib
demo% ls
total 14
2 dropx.f
2 delte.f
2 etc.f
2 evalx.f
2 linkz.f
2 markx.f
2 point.f
Libraries 4-9
The lower-level “helper” routines are gathered together into the file etc.f. The
other files can contain one or more subprograms.
First, compile each of the library source files, using the -c option, to generate the
corresponding relocatable .o files:
demo% f77 -c *.f
delte.f:
delte:
q_fixx:
dropx.f:
dropx:
etc.f:
q_fill:
q_step:
q_node:
q_warn:
...etc
demo% ls
total 42
2 dropx.f
4 etc.o
2 delte.f
4 dropx.o
4 delte.o
2 etc.f
demo%
2 linkz.f
2 evalx.f
4 evalx.o
4 markx.o
4 linkz.o
2 markx.f
2 point.f
4 point.o
Now, create the static library testlib.a using ar:
demo% ar cr testlib.a *.o
To use this library, either include the library file on the compilation command or use
the -l and -L compilation options. The example uses the .a file directly:
demo% cat trylib.f
C
program to test testlib routines
x=21.998
call evalx(x)
call point(x)
print*, "value ",x
end
demo% f77 -o trylib trylib.f test_lib/testlib.a
trylib.f:
MAIN:
demo%
Notice that the main program calls only two of the routines in the library. You can
verify that the uncalled routines in the library were not loaded into the executable
file by looking for them in the list of names in the executable displayed by nm:
demo% nm trylib | grep FUNC
[146]
|
70016|
demo% nm trylib | grep FUNC
[165]
|
69848|
4-10
| grep point
152|FUNC |GLOB |0
| grep evalx
152|FUNC |GLOB |0
Fortran Programming Guide ♦ Revision A, February 1999
|8
|point_
|8
|evalx_
demo% nm trylib | grep FUNC | grep delte
demo% nm trylib | grep FUNC | grep markx
demo% ..etc
In the preceding example, grep finds entries in the list of names only for those
library routines that were actually called.
Another way to reference the library is through the -llibrary and -Lpath options.
Here, the library’s name would have to be changed to conform to the libname.a
convention:
demo% mv test_lib/testlib.a test_lib/libtestlib.a
demo% f77 -o trylib trylib.f -Ltest_lib -ltestlib
trylib.f:
MAIN:
The -llibrary and -Lpath options are used with libraries installed in a commonly
accessible directory on the system, like /usr/local/lib, so that other users can
reference it. For example, if you left libtestlib.a in /usr/local/lib, other
users could be informed to compile with the following command:
demo% f77 -o myprog myprog.f -L/usr/local/lib -ltestlib
Replacement in a Static Library
It is not necessary to recompile an entire library if only a few elements need
recompiling. The -r option of ar permits replacement of individual elements in a
static library.
Example: Recompile and replace a single routine in a static library:
demo% f77 --c point.f
demo% ar r testlib.a point.o
demo%
Ordering Routines in a Static Library
To order the elements in a static library when it is being built by ar, use the
commands lorder(1) and tsort(1):
demo% ar cr mylib.a ’lorder exg.o fofx.o diffz.o | tsort’
Libraries
4-11
Creating Dynamic Libraries
Dynamic library files are built by the linker ld from precompiled object modules
that can be bound into the executable file after execution begins.
Another feature of a dynamic library is that modules can be used by other executing
programs in the system without duplicating modules in each program’s memory. For
this reason, a dynamic library is also a shared library.
A dynamic library offers the following features:
4 The object modules are not bound into the executable file by the linker during the
compile-link sequence; such binding is deferred until runtime.
4 A shared library module is bound into system memory when the first running
program references it. If any subsequent running program references it, that
reference is mapped to this first copy.
4 Maintaining programs is easier with dynamic libraries. Installing an updated
dynamic library on a system immediately affects all the applications that use it
without requiring relinking of the executable.
Tradeoffs for Dynamic Libraries
Dynamic libraries introduce some additional tradeoff considerations:
4 Smaller a.out file
Deferring binding of the library routines until execution time means that the size
of the executable file is less than the equivalent executable calling a static version
of the library; the executable file does not contain the binaries for the library
routines.
4 Possibly smaller process memory utilization
When several processes using the library are active simultaneously, only one copy
of the memory resides in memory and is shared by all processes.
4 Possibly increased overhead
Additional processor time is needed to load and link-edit the library routines
during runtime. Also, the library’s position-independent coding might execute
more slowly than the relocatable coding in a static library.
4 Possible overall system performance improvement
Reduced memory utilization due to library sharing should result in better overall
system performance (reduced I/O access time from memory swapping).
Performance profiles among programs vary greatly from one to another. It is not
always possible to determine or estimate in advance the performance improvement
4-12
Fortran Programming Guide ♦ Revision A, February 1999
(or degradation) between dynamic versus static libraries. However, if both forms of a
needed library are available to you, it would be worthwhile to evaluate the
performance of your program with each.
Position-Independent Code and -pic
Position-independent code (PIC) can be bound to any address in a program without
requiring relocation by the link editor. Such code is inherently sharable between
simultaneous processes. Thus, if you are building a dynamic, shared library, you
must compile the component routines to be position-independent (by using compiler
options -pic or -PIC).
In position-independent code, each reference to a global item is compiled as a
reference through a pointer into a global offset table. Each function call is compiled
in a relative addressing mode through a procedure linkage table. The size of the
global offset table is limited to 8 Kbytes on SPARC processors. The -PIC compiler
option is similar to -pic, but -PIC allows the global offset table to span the range of
32–bit addresses.
Version 5.0 of f77 and version 2.0 of f90 introduce a more flexible compiler flag,
-xcode=v, for specifying the code address space of a binary object. With this
compiler flag, 32-, 44-, or 64-bit absolute addresses can be generated, as well as small
and large model position-independent code. -xcode=pic13 is equivalent to -pic,
and -xcode=pic32 is equivalent to -PIC. See the f77(1) and f90(1) man pages, or
the Fortran User’s Guide, for details.
Binding Options
You can specify dynamic or static library binding when you compile. These options
are actually linker options, but they are recognized by the compiler and passed on to
the linker.
-Bdynamic | -Bstatic
-Bdynamic sets the preference for shared, dynamic binding whenever possible.
-Bstatic restricts binding to static libraries only.
When both static and dynamic versions of a library are available, use this option to
toggle between preferences on the command line:
f77 prog.f -Bdynamic -lwells -Bstatic -lsurface
Libraries
4-13
-dy | -dn
Allows or disallows dynamic linking for the entire executable. (This option may
appear on the command line only once.)
-dy allows dynamic, shared libraries to be linked. -dn does not allow linking of
dynamic libraries.
Binding in 64-Bit Environments
Some static system libraries, such as libm.a and libc.a, are not available on 64-bit
environments with Solaris 7. These are supplied as dynamic libraries only. Use of
-dn in these environments will result in an error indicating that some static system
libraries are missing. Also, ending the compiler command line with -Bstatic will
have the same effect.
To link with static versions of specific libraries, use a command line that looks
something like:
f77 -o prog prog.f -Bstatic -labc -lxyz -Bdynamic
Here the user’s libabc.a and libxyz.a file are linked (rather than libabc.so or
libxyz.so), and the final -Bdynamic insures that the remaining libraries,
including system libraries, and dynamically linked.
In more complicated situations, it may be necessary to explicitly reference each
system and user library on the link step with the appropriate -Bstatic or
-Bdynamic as required. First use LD_OPTIONS set to ’-Dfiles’ to obtain a listing
of all the libraries needed. Then perform the link step with -nolib (to suppress
automatic linking of system libraries) and explicit references to the libraries you
need. For example:
f77 -xarch=v9 -o cdf -nolib cdf.o-Bstatic -lF77 -lM77 -lsunmath -Bdynamic -lm -lc
Naming Conventions
To conform to the dynamic library naming conventions assumed by the link loader
and the compilers, assign names to the dynamic libraries that you create with the
prefix lib and the suffix .so. For example, libmyfavs.so could be referenced by
the compiler option -lmyfavs.
The linker also accepts an optional version number suffix: for example,
libmyfavs.so.1 for version one of the library.
The compiler’s -hname option records name as the name of the dynamic library being
built.
4-14
Fortran Programming Guide ♦ Revision A, February 1999
A Simple Dynamic Library
Building a dynamic library requires a compilation of the source files with the -pic
or -PIC option and linker options -G, -ztext, and -hname. These linker options are
available through the compiler command line.
You can create a dynamic library with the same files used in the static library
example.
Example: Compile with -pic and other linker options:
demo% f77 -o libtestlib.so.1 -G -pic -ztext -hlibtestlib.so.1 *.f
delte.f:
delte:
q_fixx:
dropx.f:
dropx:
etc.f:
q_fill:
q_step:
q_node:
q_warn:
evalx.f:
evalx:
linkz.f:
linkz:
markx.f:
markx:
point.f:
point:
Linking:
-G tells the linker to build a dynamic library.
-ztext warns you if it finds anything other than position-independent code, such as
relocatable text.
Example: Make an executable file a.out using the dynamic library:
demo% f77 -o trylib -R‘pwd‘ trylib.f libtestlib.so.1
trylib.f:
MAIN main:
demo% file trylib
trylib:ELF 32--bit MSB executable SPARC Version 1, dynamically linked, not stripped
demo% ldd trylib
libtestlib.so.1 => /export/home/U/Tests/libtestlib.so.1
libF77.so.4 => /opt/SUNWspro/lib/libF77.so.4
libc.so.1 => /usr/lib/libc.so.1
libdl.so.1 => /usr/lib/libdl.so.1
Note that the example uses the -R option to bind into the executable the path (the
current directory) to the dynamic library.
The file command shows that the executable is dynamically linked.
Libraries
4-15
The ldd command shows that the executable, trylib, uses some shared libraries,
including libtestlib.so.1; libf77, libdl, and libc are included by default by
f77.
Libraries Provided with Sun Fortran
Compilers
The table shows the libraries installed with the compilers.
TABLE 4–1
4-16
Major Libraries Provided With the Compilers
Library
Name
Options Needed
f77 functions, nonmath
libF77
None
f77 functions, nonmath, multithread safe
libF77_mt
-parallel
f77 math library
libM77
None
f90 support intrinsics
libfsu
None
f90 interface
libfui
None
f90 array intrinsics libraries
libf*ai
None
f90/f77 I/O compatibility library
libf77compat
-lf77compat
VMS library
libV77
-lV77
Library used with Pascal, Fortran, and C
libpfc
None
Library of Sun math functions
libsunmath
None
POSIX bindings
libFposix
-lFposix
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 4–1
Major Libraries Provided With the Compilers
(continued)
f90 POSIX interface
libposix9
-lposix9
POSIX bindings for extra runtime checking
libFposix_c
-lFposix_c
See also the math_libraries README file for more information.
VMS Library
The libV77 library is the VMS library, which contains two special VMS routines,
idate and time.
To use either of these routines, include the -lV77 option.
For idate and time, there is a conflict between the VMS version and the version
that traditionally is available in UNIX environments. If you use the -lV77 option,
you get the VMS compatible versions of the idate and time routines.
See the Fortran Library Reference Manual and the FORTRAN 77 Language
Reference Manual for details on these routines.
POSIX Library
There are two versions of POSIX bindings provided with FORTRAN 77:
4 libFposix, which is just the bindings (-lFposix)
4 libFposix_c, which does some runtime checking to make sure you are passing
correct handles (-lFposix_c)
If you pass bad handles:
4 libFposix_c returns an error code (ENOHANDLE).
4 libFposix core dumps with a segmentation fault.
Of course, the checking is time-consuming, and libFposix_c is several times
slower.
Both POSIX libraries come in static and dynamic forms.
The POSIX bindings provided are for IEEE Standard 1003.9–1992.
IEEE 1003.9 is a binding of 1003.1–1990 to FORTRAN (X3.8–1978).
For more information, see these POSIX.1 documents:
4 ISO/IEC 9945–1:1990
Libraries
4-17
4 IEEE Standard 1003.1–1990
4 IEEE Order number SH13680
4 IEEE CS Catalog number 1019
To find out precisely what POSIX is, you need both the 1003.9 and the POSIX.1
documents.
The POSIX library for f90 is libposix9.
Shippable Libraries
If your executable uses a Sun dynamic library that is listed in the
runtime.libraries README file, your license includes the right to redistribute
the library to your customer.
This README file is located in the READMEs directory:
<install-point>/SUNWspro/READMEs/
Do not redistribute or otherwise disclose the header files, source code, object
modules, or static libraries of object modules in any form.
Refer to your software license for more details.
4-18
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
5
Program Analysis and Debugging
This chapter presents a number of Sun Fortran compiler features that facilitate
program analysis and debugging.
Global Program Checking
The f77 compiler’s -Xlistx options provide a valuable way to analyze a source
program for inconsistencies and possible runtime problems. The analysis performed
by the compiler is global, across subprograms.
-Xlistx reports errors in alignment, agreement in number and type for subprogram
arguments, common block, parameter, and various other kinds of errors.
-Xlistx also can be used to make detailed source code listings and cross-reference
tables.
Note - The f90 compiler provides only a subset of the −-Xlist options described
here. A conventional cross-reference map is produced, but complete global program
checking is not performed.
GPC Overview
Global program checking (GPC), invoked by the -Xlistx option, does the following:
4 Enforces type-checking rules of Fortran more stringently than usual, especially
between separately compiled routines
5-1
4 Enforces some portability restrictions needed to move programs between different
machines or operating systems
4 Detects legal constructions that nevertheless might be suboptimal or error-prone
4 Reveals other potential bugs and obscurities
In particular, global checking reports problems such as:
4 Interface problems
4
Conflicts in number and type of dummy and actual arguments
4
Wrong types of function values
4
Possible conflicts due to data type mismatches in common blocks between
different subprograms
4 Usage problems
4
Function used as a subroutine or subroutine used as a function
4
Declared but unused functions, subroutines, variables, and labels
4
Referenced but not declared functions, subroutines, variables, and labels
4
Usage of unset variables
4
Unreachable statements
4
Implicit type variables
4
Inconsistency of the named common block lengths, names, and layouts
How to Invoke Global Program Checking
The -Xlist option on the command line invokes the compiler’s global program
analyzer. There are a number of -Xlistx suboptions, as described in the sections
that follow.
Example: Compile three files for basic global program checking:
demo% f77 -Xlist
any1.f
any2.f
any3.f
In the preceding example, the compiler:
4 Produces output listings in the file any1.lst
5-2
Fortran Programming Guide ♦ Revision A, February 1999
4 Compiles and links the program if there are no errors
Screen Output
Normally, output listings produced by -Xlistx are written to a file. To display
directly to the screen, use -Xlisto to write the output file to /dev/tty.
Example: Display to terminal:
demo% f77 -Xlisto /dev/tty
any1.f
Default Output Features
The -Xlist option provides a combination of features available for output. With no
other -Xlist options, you get the following by default:
4 The listing file name is taken from the first input source or object file that appears,
with the extension replaced by .lst
4 A line-numbered source listing
4 Error messages (embedded in listing) for inconsistencies across routines
4 Cross-reference table of the identifiers
4 Pagination at 66 lines per page and 79 columns per line
4 No call graph
4 No expansion of include files
File Types
The checking process recognizes all the files in the compiler command line that end
in .f, .f90, .for, .F, or .o. The .o files supply the process with information
regarding only global names, such as subroutine and function names.
Analysis Files (.fln Files)
Programs compiled with −-Xlist options have their analysis data built into the
binary files automatically. This enables global program checking over programs in
libraries.
Alternatively, the compiler will save individual source file analysis results into files
with a .fln suffix if the -Xlistflndir option is also specified. dir indicates the
directory to receive these files.
Program Analysis and Debugging
5-3
demo% f77 -Xlistfln/tmp *.f
Some Examples of −-Xlist and Global Program
Checking
Here is a listing of the Repeat.f source code used in the following examples:
demo% cat Repeat.f
PROGRAM repeat
pn1 = REAL( LOC ( rp1 ) )
CALL subr1 ( pn1 )
CALL nwfrk ( pn1 )
PRINT *, pn1
END ! PROGRAM repeat
SUBROUTINE subr1 ( x )
IF ( x .GT. 1.0 ) THEN
CALL subr1 ( x * 0.5 )
END IF
END
SUBROUTINE nwfrk( ix )
EXTERNAL fork
INTEGER prnok, fork
PRINT *, prnok ( ix ), fork ( )
END
INTEGER FUNCTION prnok ( x )
prnok = INT ( x ) + LOC(x)
END
SUBROUTINE unreach_sub()
CALL sleep(1)
END
Example: Use -XlistE to show errors and warnings:
demo% f77 -XlistE -silent Repeat.f
demo% cat Repeat.lst
FILE "Repeat.f"
program repeat
4
CALL nwfrk ( pn1 )
^
**** ERR #418: argument "pn1" is real, but dummy argument is
integer*4
See: "Repeat.f" line #14
4
CALL nwfrk ( pn1 )
^
**** ERR #317: variable "pn1" referenced as integer*4 across
5-4
Fortran Programming Guide ♦ Revision A, February 1999
repeat/nwfrk//prnok in line #21 but set as real
by repeat in line #2
subroutine
10
subr1
CALL subr1 ( x * 0.5 )
^
**** WAR #348: recursive call for "subr1". See dynamic calls:
"Repeat.f" line #3
subroutine nwfrk
17
PRINT *, prnok ( ix ), fork ( )
^
**** ERR #418: argument "ix" is integer*4, but dummy argument
is real
See: "Repeat.f" line #20
subroutine unreach_sub
24
SUBROUTINE unreach_sub()
^
**** WAR #338: subroutine "unreach_sub" isn"t called from program
Date:
Wed Feb 24 10:40:32 1999
Files:
2 (Sources: 1; libraries: 1)
Lines:
26 (Sources: 26; Library subprograms:2)
Routines:
5 (MAIN: 1; Subroutines: 3; Functions: 1)
Messages:
5 (Errors: 3; Warnings: 2)
demo%
Compiling the same program with −-Xlist also produces a cross-reference table on
standard output:
Source
Legend:
D
U
M
A
C
I
E
N
C R O S S
R E F E R E N C E
file:
Repeat.f
T A B L E
Definition/Declaration
Simple use
Modified occurrence
Actual argument
Subroutine/Function call
Initialization: DATA or extended declaration
Occurrence in EQUIVALENCE
Occurrence in NAMELIST
P R O G R A M
F O R M
Program
-------------repeat
<repeat>
D
1:D
Functions and Subroutines
-------------------------------------------------fork
int*4 <nwfrk>
DC
15:D
16:D
int
loc
nwfrk
intrinsic
<prnok>
C
21:C
intrinsic
<repeat>
<prnok>
C
C
2:C
21:C
<repeat>
C
4:C
17:C
Program Analysis and Debugging
5-5
<nwfrk>
D
14:D
<nwfrk>
<prnok>
DC
DM
16:D
20:D
C
2:C
prnok
int*4
real
intrinsic
<repeat>
sleep
<unreach_sub>
subr1
<repeat>
<subr1>
<unreach_sub>
unreach_sub
C
DC
17:C
21:M
C
25:C
3:C
8:D
D
10:C
24:D
Output from compiling f77 -Xlist Repeat.f (Continued)
Variables and Arrays
---------------------------------------ix
int*4
dummy
<nwfrk>
DA
14:D
17:A
3:A
4:A
9:U
21:A
10:U
21:U
pn1
real*4 <repeat>
UMA
2:M
rp1
real*4 <repeat>
A
2:A
x
real*4 dummy
<subr1>
<prnok>
DU
DUA
8:D
20:D
5:U
--------------------------------------------------------------------------------------------------------Date:
Tue Feb 22 13:15:39 1995
Files:
2 (Sources: 1; libraries: 1)
Lines:
26 (Sources: 26; Library subprograms:2)
Routines:
5 (MAIN: 1; Subroutines: 3; Functions: 1)
Messages:
5 (Errors: 3; Warnings: 2)
demo%
In the cross-reference table in the preceding example:
4 ix is a 4-byte integer:
4
Used as an argument in the routine nwfrk
At line 14, used as a declaration of argument
4 At line 17, used as an actual argument
4
4 pn1 is a 4–byte real in the routine repeat:
4
At line 2, modified
4 At line 3, argument
4 At line 4, argument
5-6
Fortran Programming Guide ♦ Revision A, February 1999
4 At line 5, used
4 rp1 is a 4-byte real in the routine, repeat. At line 2, it is an argument.
4 x is a 4-byte real in the routines subr1 and prnok:
4 In subr1, at line 8, defined; used at lines 9 and 10
4 In prnok, at line 20, defined; at line 21, used as an argument
Suboptions for Global Checking Across Routines
The basic global cross-checking option is -Xlist with no suboption. It is a
combination of suboptions, each of which could have been specified separately.
The following sections describe options for producing the listing, errors, and
cross-reference table. Multiple suboptions may appear on the command line.
Suboption Syntax
Add suboptions according to the following rules:
4 Append the suboption to -Xlist.
4 Put no space between the -Xlist and the suboption.
4 Use only one suboption per -Xlist.
−-Xlist and its Suboptions
Combine suboptions according to the following rules:
4 The most general option is -Xlist (listing, errors, cross-reference table).
4 Specific features can be combined using -Xlistc, -XlistE, -XlistL, or
-XlistX.
4 Other suboptions specify further details.
Example: Each of these two command lines performs the same task:
demo% f77
-Xlistc
-Xlist
demo% f77
-Xlistc
any.f
any.f
The following table shows the reports generated by these basic −-Xlist suboptions
alone:
Program Analysis and Debugging
5-7
TABLE 5–1
Xlist Suboptions
Generated Report
Option
Errors, listing, cross-reference
-Xlist
Errors only
-XlistE
Errors and source listing only
-XlistL
Errors and cross-reference table only
-XlistX
Errors and call graph only
-Xlistc
The following table summarizes all -Xlist suboptions.
TABLE 5–2
Summary of −-Xlist Suboptions
Option
Action
-Xlist (no
suboption)
Shows errors, listing, and cross-reference table
-Xlistc
Shows call graphs and errors (f77 only)
-XlistE
Shows errors
Suppresses error nnn in the verification report
Xlisterr[nnn]
5-8
-Xlistf
Produces fast output (f77 only)
-Xlistflndir
Puts the .fln files in dir (f77 only)
-Xlisth
Shows errors from cross-checking stop compilation (f77 only)
-XlistI
Lists and cross-checks include files
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 5–2
Summary of −-Xlist Suboptions
(continued)
-XlistL
Shows the listing and errors
-Xlistln
Sets page breaks
-Xlisto name
Renames the -Xlist output report file
-Xlists
Suppresses unreferenced symbols from cross-reference (f77 only)
-Xlistvn
Sets checking “strictness” level (f77 only)
-Xlistw[nnn]
Sets the width of output lines
Xlistwar[nnn]
Suppresses warning nnn in the report
-XlistX
Shows just the cross-reference table and errors
-Xlist Suboption Reference
This section describes the −-Xlist suboptions. As noted, some are only available
with f77.
−f77: -Xlistc — Show call graphs and cross-routine errors
Used alone, −-Xlistc does not show a listing or cross-reference. It produces the call
graph in a tree form, using printable characters. If some subroutines are not called
from MAIN, more than one graph is shown. Each BLOCKDATA is printed separately
with no connection to MAIN.
The default is not to show the call graph.
−-XlistE – Show cross-routine errors
Used alone, −-XlistE shows only cross-routine errors and does not show a listing
or a cross-reference.
Program Analysis and Debugging
5-9
− -Xlisterr[nnn] – Suppress error nnn
Use −-Xlisterr to suppress a numbered error message from the listing or
cross-reference.
For example: -Xlisterr338 suppresses error message 338. If nnn is not specified,
all error messages are suppressed. To suppress additional specific errors, use this
option repeatedly.
f77: -Xlist−f – Produce faster output
Use −-Xlistf to produce source file listings and a cross-checking report and to
verify sources, but without generating object files.
The default without this option is to generate object files.
f77: –Xlist−flndir – Put .fln files into dir directory
Use −-Xlistfln to specify the directory to receive .fln source analysis files. The
directory specified (dir) must already exist. The default is to include the source
analysis information directly within the object .o files (and not generate .fln files).
f77: -Xlist−h – Halt on errors
With −-Xlisth, compilation stops if errors are detected while cross-checking the
program. In this case, the report is redirected to stdout instead of the *.lst file.
-Xlist−I – List and cross-check include files
If -XlistI is the only suboption used, include files are shown or scanned along
with the standard -Xlist output (line numbered listing, error messages, and a
cross-reference table).
4 Listing—If the listing is not suppressed, then the include files are listed in place.
Files are listed as often as they are included. The files are:
4
Source files
#include files
4 INCLUDE files
4
4 Cross-Reference Table—If the cross reference table is not suppressed, the
following files are all scanned while the cross reference table is generated:
4
4
5-10
Source files
#include files
Fortran Programming Guide ♦ Revision A, February 1999
4 INCLUDE files
The default is not to show include files.
-Xlist−L – Show listing and cross routine errors
Use −-XlistL to produce only a listing and a list of cross routine errors. This
suboption by itself does not show a cross reference table. The default is to show the
listing and cross reference table.
-Xlist−ln – Set the page length for pagination to n lines
Use −-Xlistl to set the page length to something other than the default page size.
For example, -Xlistl45 sets the page length to 45 lines. The default is 66.
With n=0 (-Xlistl0) this option shows listings and cross-references with no page
breaks for easier on-screen viewing.
-Xlist−o name – Rename the -Xlist output report file
Use −-Xlisto to rename the generated report output file. (A space between o and
name is required.) With −-Xlisto name, the output is to name.lst .
To display directly to the screen, use the command: -Xlisto /dev/tty
f77: -Xlist−s – Suppress unreferenced identifiers
Use −-Xlists to suppress from the cross reference table any identifiers defined in
the include files but not referenced in the source files.
This suboption has no effect if the suboption -XlistI is used.
The default is not to show the occurrences in #include or INCLUDE files.
f77: -Xlist−vn – Set level of checking strictness
n is 1,2, 3, or 4. The default is 2 (-Xlistv2):
4 -Xlistv1
Shows the cross-checked information of all names in summary form only, with no
line numbers. This is the lowest level of checking strictness—syntax errors only.
4 -Xlistv2
Shows cross-checked information with summaries and line numbers. This is the
default level of checking strictness and includes argument inconsistency errors and
variable usage errors.
Program Analysis and Debugging
5-11
4 -Xlistv3
Shows cross-checking with summaries, line numbers, and common block maps.
This is a high level of checking strictness and includes errors caused by incorrect
usage of data types in common blocks in different subprograms.
4 -Xlistv4
Shows cross-checking with summaries, line numbers, common block maps, and
equivalence block maps. This is the strictest level of checking with maximum error
detection.
-Xlist−w[nnn] – Set width of output line to n columns
Use −-Xlistw to set the width of the output line. For example, -Xlistw132 sets
the page width to 132 columns. The default is 79.
-Xlist−war[nnn] – Suppress warning nnn in the report
Use −-Xlistwar to suppress a specific warning message from the output reports. If
nnn is not specified, then all warning messages are suppressed from printing. For
example, -Xlistwar338 suppresses warning message number 338. To suppress
more than one, but not all warnings, use this option repeatedly.
-Xlist−X – Show cross-reference table and cross routine
errors
−-XlistX produces a cross reference table and cross routine error list but no source
listing.
Some Examples Using Suboptions
Example: Use -Xlistwarnnn to suppress two warnings from a preceding example:
demo% f77 -Xlistwar338 -Xlistwar348 -XlistE -silent Repeat.f
demo% cat Repeat.lst
FILE "Repeat.f"
program repeat
4
CALL nwfrk ( pn1 )
^
**** ERR #418: argument "pn1" is real, but dummy argument is
integer*4
See: "Repeat.f" line #14
4
CALL nwfrk ( pn1 )
^
**** ERR #317: variable "pn1" referenced as integer*4 across
repeat/nwfrk//prnok in line #21 but set as real
5-12
Fortran Programming Guide ♦ Revision A, February 1999
by repeat in line #2
subroutine
17
**** ERR
nwfrk
#418:
PRINT *, prnok ( ix ), fork ( )
^
argument "ix" is integer*4, but dummy argument
is real
See: "Repeat.f" line #20
Date:
Wed Feb 24 10:40:32 1999
Files:
2 (Sources: 1; libraries: 1)
Lines:
26 (Sources: 26; Library subprograms:2)
Routines:
5 (MAIN: 1; Subroutines: 3; Functions: 1)
Messages:
5 (Errors: 3; Warnings: 2)
demo%
Example: Explain a message and find a type mismatch in program ShoGetc.f:
demo% cat ShoGetc.f
CHARACTER*1 c
i = getc(c)
END
demo% f77 -silent ShoGetc.f
Compile program
demo% a.out
Program waits for input...
Z
Type "Z" on keyboard. This causes run-time message. Why?
Note: IEEE floating-point exception flags raised:
Invalid Operation;
See the Numerical Computation Guide, ieee_flags(3M)
demo% f77 -XlistE -silent ShoGetc.f Compile with Global Program Checking
demo% cat ShoGetc.lst
and view listing
FILE "ShoGetc.f"
program MAIN
2
i = getc(c)
^
**** WAR #320: variable "i" set but never referenced
2
i = getc(c)
^
**** ERR #412: function "getc" used as real but declared as
integer*4
Here is the error - function must be declared INTEGER.
2
i = getc(c)
^
**** WAR #320: variable "c" set but never referenced
demo% cat ShoGetc.f
Modify program to declare getc INTEGER and run again.
CHARACTER*1 c
INTEGER getc
i = getc(c)
END
demo% f77 -silent ShoGetc.f
demo% a.out
Z
Type "Z" on keyboard
demo%
Now no error.
Program Analysis and Debugging
5-13
Special Compiler Options
Some compiler options are useful for debugging. They check subscripts, spot
undeclared variables, show stages of the compile-link sequence, display versions of
software, and so on.
The Solaris linker has additional debugging aids. See ld(1), or run the command ld
-Dhelp at a shell prompt to see the online documentation.
Subscript Bounds (-C)
The -C option adds checks for out-of-bounds array subscripts.
If you compile with -C, the compiler adds checks at runtime for out-of-bounds
references on each array subscript. This action helps catch some situations that cause
segmentation faults.
Example: Index out of range:
demo% cat indrange.f
REAL a(10,10)
k = 11
a(k,2) = 1.0
END
demo% f77 --C --silent indrange.f
demo% a.out
Subscript out of range on file indrange.f, line 3, procedure MAIN.
Subscript number 1 has value 11 in array a.
Abort (core dumped)
demo%
f77: Undeclared Variable Types (-u)
The -u option checks for any undeclared variables. (Not available with f90.)
The -u option causes all variables to be initially identified as undeclared, so that all
variables that are not explicitly declared by type statements, or by an IMPLICIT
statement, are flagged with an error. The -u flag is useful for discovering mistyped
variables. If -u is set, all variables are treated as undeclared until explicitly declared.
Use of an undeclared variable is accompanied by an error message.
5-14
Fortran Programming Guide ♦ Revision A, February 1999
Version Checking (-V)
The -V option causes the name and version ID of each phase of the compiler to be
displayed. This option can be useful in tracking the origin of ambiguous error
messages and in reporting compiler failures, and to verify the level of installed
compiler patches.
Interactive Debugging With dbx and
Sun WorkShop
The Sun WorkShop provides a tightly integrated development environment for
building and browsing, as well as debugging applications written in Fortran, C, and
C++.
The Sun WorkShop debugging facility is a window-based interface to dbx, while dbx
itself is an interactive, line-oriented, source-level symbolic debugger. Either can be
used to determine where a program crashed, to view or trace the values of variables
and expressions in a running code, and to set breakpoints.
Sun WorkShop adds a sophisticated graphical environment to the debugging process
that is integrated with tools for editing, building, and source code version control. It
includes a data visualization capability to display and explore large and complex
datasets, simulate results, and interactively steer computations.
For details, see the Sun manuals Using Sun WorkShop and Debugging a Program
With Sun WorkShop, and the dbx(1) man pages.
The dbx program provides event management, process control, and data inspection.
You can watch what is happening during program execution, and perform the
following tasks:
4 Fix one routine, then continue executing without recompiling the others
4 Set watchpoints to stop or trace if a specified item changes
4 Collect data for performance tuning
4 Graphically monitor variables, structures, and arrays
4 Set breakpoints (set places to halt in the program) at lines or in functions
4 Show values—once halted, show or modify variables, arrays, structures
4 Step through a program, one source or assembly line at a time
4 Trace program flow—show sequence of calls taken
4 Invoke procedures in the program being debugged
4 Step over or into function calls; step up and out of a function call
Program Analysis and Debugging
5-15
4 Run, stop, and continue execution at the next line or at some other line
4 Save and then replay all or part of a debugging run
4 Examine the call stack, or move up and down the call stack
4 Program scripts in the embedded Korn shell
4 Follow programs as they fork(2) and exec(2)
To debug optimized programs, use the dbx fix command to recompile the routines
you want to debug:
1. Compile the program with the appropriate -On optimization level.
2. Start the execution under dbx.
3. Use fix -g any.f without optimization on the routine you want to debug.
4. Use continue with that routine compiled.
Some optimizations will be inhibited by the presence of −-g on the compilation
command. For example, −-g suppresses the automatic inlining usually obtained
with− -O4. −-g cancels any parallelization option (−-autopar, −-explicitpar,
−-parallel), as well as −-depend and −-reduction. Debugging is facilitated by
specifying −-g without any optimization options. See the dbx documentation for
details.
f77: Viewing Compiler Listing
Diagnostics
Use the error utility program to view compiler diagnostics merged with the source
code. error inserts compiler diagnostics above the relevant line in the source file.
The diagnostics include the standard compiler error and warning messages, but not
the -Xlist error and warning messages.
Note - The error utility rewrites your source files and does not work if the source
files are read-only, or are in a read only directory.
error(1) is included as part of a “developer” installation of the Solaris operating
environment; it can also be installed from the package, SUNWbtool.
Facilities also exist in the Sun WorkShop for viewing compiler diagnostics. Refer to
Using Sun WorkShop.
5-16
Fortran Programming Guide ♦ Revision A, February 1999
Program Analysis and Debugging
5-17
5-18
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
6
Floating-Point Arithmetic
This chapter considers floating-point arithmetic and suggests strategies for avoiding
and detecting numerical computation errors.
For a detailed examination of floating-point computation on SPARC and x86
processors, see the Sun Numerical Computation Guide.
Introduction
Sun’s floating-point environment on SPARC and x86 implements the arithmetic
model specified by the IEEE Standard 754 for Binary Floating Point Arithmetic. This
environment enables you to develop robust, high-performance, portable numerical
applications. It also provides tools to investigate any unusual behavior by a
numerical program.
In numerical programs, there are many potential sources for computational error:
4 The computational model could be wrong.
4 The algorithm used could be numerically unstable.
4 The data could be ill-conditioned.
4 The hardware could be producing unexpected results.
Finding the source of the errors in a numerical computation that has gone wrong can
be extremely difficult. The chance of coding errors can be reduced by using
commercially available and tested library packages whenever possible. Choice of
algorithms is another critical issue. Using the appropriate computer arithmetic is
another.
6-1
This chapter makes no attempt to teach or explain numerical error analysis. The
material presented here is intended to introduce the IEEE floating-point model as
implemented by Sun’s Fortran compilers.
IEEE Floating-Point Arithmetic
IEEE arithmetic is a relatively new way of dealing with arithmetic operations that
result in such problems as invalid, division by zero, overflow, underflow, or inexact.
The differences are in rounding, handling numbers near zero, and handling numbers
near the machine maximum.
The IEEE standard supports user handling of exceptions, rounding, and precision.
Consequently, the standard supports interval arithmetic and diagnosis of anomalies.
IEEE Standard 754 makes it possible to standardize elementary functions like exp and
cos, to create high precision arithmetic, and to couple numerical and symbolic
algebraic computation.
IEEE arithmetic offers users greater control over computation than does any other
kind of floating-point arithmetic. The standard simplifies the task of writing
numerically sophisticated, portable programs. Many questions about floating-point
arithmetic concern elementary operations on numbers. For example:
4 What is the result of an operation when the infinitely precise result is not
representable in the computer hardware?
4 Are elementary operations like multiplication and addition commutative?
Another class of questions concerns floating-point exceptions and exception
handling. What happens if you:
4 Multiply two very large numbers with the same sign?
4 Divide nonzero by zero?
4 Divide zero by zero?
In older arithmetic models, the first class of questions might not have the expected
answers, while the exceptional cases in the second class might all have the same
result: the program aborts on the spot or proceeds with garbage results.
The standard ensures that operations yield the mathematically expected results with
the expected properties. It also ensures that exceptional cases yield specified results,
unless the user specifically makes other choices.
For example, the exceptional values +Inf, -Inf, and NaN are introduced intuitively:
big*big = +Inf
big*(-big) = -Inf
num/0.0 = +Inf
num/0.0 = -Inf
6-2
Positive infinity
Negative infinity
Where num > 0.0
Where num < 0.0
Fortran Programming Guide ♦ Revision A, February 1999
0.0/0.0 = NaN
Not a Number
Also, five types of floating-point exception are identified:
4 Invalid. Operations with mathematically invalid operands—for example, 0.0/0.0,
sqrt(-1.0), and log(-37.8)
4 Division by zero. Divisor is zero and dividend is a finite nonzero number—for
example, 9.9/0.0
4 Overflow. Operation produces a result that exceeds the range of the exponent—
for example, MAXDOUBLE+0.0000000000001e308
4 Underflow. Operation produces a result that is too small to be represented as a
normal number—for example, MINDOUBLE * MINDOUBLE
4 Inexact. Operation produces a result that cannot be represented with infinite
precision—for example, 2.0 / 3.0, log(1.1) and 0.1 in input
The implementation of the IEEE standard is described in the Sun Numerical
Computation Guide.
-ftrap=mode Compiler Options
The -ftrap=mode option enables trapping for floating-point exceptions. If no signal
handler has been established by an ieee_handler() call, the exception terminates
the program with a memory dump core file. See Fortran User’s Guide for details on
this compiler option. For example, to enable trapping for overflow, division by zero,
and invalid operations, compile with -ftrap=common.
Note - You must compile the application’s main program with -ftrap= for
trapping to be enabled.
Floating-Point Exceptions and Fortran
Programs compiled by f77 automatically display a list of accrued floating-point
exceptions on program termination. In general, a message results if any one of the
invalid, division-by-zero, or overflow exceptions have occurred. Inexact exceptions
do not generate messages because they occur so frequently in real programs.
f90 programs do not automatically report on exceptions at program termination. An
explicit call to ieee_retrospective(3M) is required.
You can turn off any or all of these messages with ieee_flags() by clearing
exception status flags. Do this at the end of your program.
Floating-Point Arithmetic
6-3
Handling Exceptions
Exception handling according to the IEEE standard is the default on SPARC and x86
processors. However, there is a difference between detecting a floating-point
exception and generating a signal for a floating-point exception (SIGFPE).
Following the IEEE standard, two things happen when an untrapped exception
occurs during a floating-point operation:
4 The system returns a default result. For example, on 0/0 (invalid), the system
returns NaN as the result.
4 A flag is set to indicate that an exception is raised. For example, 0/0 (invalid), the
system sets the “invalid operation” flag.
Trapping a Floating-Point Exception
f77 and f90 differ significantly in the way they handle floating-point exceptions.
With f77, the default on SPARC and x86 systems is not to automatically generate a
signal to interrupt the running program for a floating-point exception. The
assumptions are that signals could degrade performance and that most exceptions
are not significant as long as expected values are returned.
The default with f90 is to automatically trap on division by zero, overflow, and
invalid operation.
The f77 and f90 command-line option -ftrap can be used to change the default.
In terms of -ftrap, the default for f77 is -ftrap=%none. The default for f90 is
-ftrap=common.
To enable exception trapping, compile the main program with one of the -ftrap
options—for example: -ftrap=common.
SPARC: Nonstandard Arithmetic
One aspect of standard IEEE arithmetic, called gradual underflow, can be manually
disabled. When disabled, the program is considered to be running with nonstandard
arithmetic.
The IEEE standard for arithmetic specifies a way of handling underflowed results
gradually by dynamically adjusting the radix point of the significand. In IEEE
floating-point format, the radix point occurs before the significand, and there is an
implicit leading bit of 1. Gradual underflow allows the implicit leading bit to be
cleared to 0 and shifts the radix point into the significant when the result of a
floating-point computation would otherwise underflow. With a SPARC processor this
result is not accomplished in hardware but in software. If your program generates
many underflows (perhaps a sign of a problem with your algorithm) and you run on
a SPARC processor, you may experience a performance loss.
6-4
Fortran Programming Guide ♦ Revision A, February 1999
Gradual underflow can be disabled either by compiling with the -fns option or by
calling the library routine nonstandard_arithmetic() from within the program
to turn it off. Call standard_arithmetic() to turn gradual underflow back on.
Note - To be effective, the application’s main program must be compiled with -fns.
See the Fortran User’s Guide.
For legacy applications, take note that:
4 The standard_arithmetic() subroutine replaces an earlier routine named
gradual_underflow().
4 The nonstandard_arithmetic() subroutine replaces an earlier routine named
abrupt_underflow().
Note - The -fns option and the nonstandard_arithmetic() library routine are
effective only on some SPARC systems. On x86 platforms, gradual underflow is
performed by the hardware.
IEEE Routines
The following interfaces help people use IEEE arithmetic. These are mostly in the
math library libsunmath and in several .h files.
4 ieee_flags(3m)—Controls rounding direction and rounding precision; query
exception status; clear exception status
4 ieee_handler(3m)—Establishes an exception handler routine
4 ieee_functions(3m)—Lists name and purpose of each IEEE function
4 ieee_values(3m)—Lists functions that return special values
4 Other libm functions described in this section:
4
ieee_retrospective
nonstandard_arithmetic
4 standard_arithmetic
4
The SPARC processors conform to the IEEE standard in a combination of hardware
and software support for different aspects. x86 processors conform to the IEEE
standard entirely through hardware support.
The newest SPARC processors contain floating-point units with integer multiply and
divide instructions and hardware square root.
Best performance is obtained when the compiled code properly matches the runtime
floating-point hardware. The compiler’s -xtarget= option permits specification of
Floating-Point Arithmetic
6-5
the runtime hardware. For example, -xtarget=ultra would inform the compiler
to generate object code that will perform best on an UltraSPARC processor.
On SPARC platforms: The utility fpversion displays which floating-point
hardware is installed and indicates the appropriate -xtarget value to specify. This
utility runs on all Sun SPARC architectures. See fpversion(1), the Sun Fortran User’s
Guide (regarding -xtarget) and the Numerical Computation Guide for details.
Flags and ieee_flags()
The ieee_flags function is used to query and clear exception status flags. It is part
of the libsunmath library shipped with Sun compilers and performs the following
tasks:
4 Controls rounding direction and rounding precision
4 Checks the status of the exception flags
4 Clears exception status flags
The general form of a call to ieee_flags is:
flags = ieee_flags( action, mode, in, out )
Each of the four arguments is a string. The input is action, mode, and in. The output is
out and flags. ieee_flags is an integer-valued function. Useful information is
returned in flags as a set of 1-bit flags. Refer to the man page for ieee_flags(3m) for
complete details.
Possible parameter values are shown in the following table:
6-6
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 6–1
ieee_flags( action, mode, in, out )Argument Values
action
mode
in, out
get
direction
nearest
set
precision
tozero
clear
exception
negative
positive
clearall
extended
double
single
inexact
division
underflow
overflow
invalid
all
common
The precision mode is available only on x86 platforms.
Note that these are literal character strings, and the output parameter out must be at
least CHARACTER*9. The meanings of the possible values for in and out depend on
the action and mode they are used with. These are summarized in the following table:
TABLE 6–2
ieee_flags Argument Meanings
Value of in and out
Refers to
nearest, tozero, negative, positive
Rounding direction
extended, double, single
Rounding precision
Floating-Point Arithmetic
6-7
TABLE 6–2
ieee_flags Argument Meanings
(continued)
Value of in and out
Refers to
inexact, division, underflow, overflow,
invalid
Exceptions
all
All five exceptions
common
Common exceptions: invalid,
division, overflow
For example, to determine what is the highest priority exception that has a flag
raised, pass the input argument in as the null string:
CHARACTER *9, out
ieeer = ieee_flags( "get", "exception", "", out )
PRINT *, out, " flag raised"
Also, to determine if the overflow exception flag is raised, set the input argument
in to overflow. On return, if out equals overflow, then the overflow exception flag
is raised; otherwise it is not raised.
ieeer = ieee_flags( "get", "exception", "overflow", out )
IF ( out.eq. "overflow") PRINT *,"overflow flag raised"
Example: Clear the invalid exception:
ieeer = ieee_flags( "clear", "exception", "invalid", out )
Example: Clear all exceptions:
ieeer = ieee_flags( "clear", "exception", "all", out )
Example: Set rounding direction to zero:
ieeer = ieee_flags( "set", "direction", "tozero", out )
Example: Set rounding precision to double:
ieeer = ieee_flags( "set", "precision", "double", out )
6-8
Fortran Programming Guide ♦ Revision A, February 1999
Turning Off All Warning Messages With ieee_flags
Calling ieee_flags with an action of clear, as shown in the following example, resets
any uncleared exceptions. Put this call before the program exits to suppress system
warning messages about floating-point exceptions at program termination.
Example: Clear all accrued exceptions with ieee_flags():
i = ieee_flags("clear", "exception", "all", out )
Detecting an Exception With ieee_flags
The following example demonstrates how to determine which floating-point
exceptions have been raised by earlier computations. Bit masks defined in the system
include file f77_floatingpoint.h are applied to the value returned by ieee_flags.
In this example, DetExcFlg.F, the include file is introduced using the #include
preprocessor directive, which requires us to name the source file with a .F suffix.
Underflow is caused by dividing the smallest double-precision number by 2.
Example: Detect an exception using ieee_flags and decode it:
#include "f77_floatingpoint.h"
CHARACTER*16 out
DOUBLE PRECISION d_max_subnormal, x
INTEGER div, flgs, inv, inx, over, under
x = d_max_subnormal() / 2.0
! Cause underflow
flgs=ieee_flags("get","exception","",out)
! Which are raised?
inx
div
under
over
inv
=
=
=
=
=
and(rshift(flgs,
and(rshift(flgs,
and(rshift(flgs,
and(rshift(flgs,
and(rshift(flgs,
fp_inexact) ,
fp_division) ,
fp_underflow),
fp_overflow) ,
fp_invalid) ,
1)
1)
1)
1)
1)
! Decode
! the value
! returned
! by
! ieee_flags
PRINT *, "Highest priority exception is: ", out
PRINT *, " invalid divide overflo underflo inexact"
PRINT "(5i8)", inv, div, over, under, inx
PRINT *, "(1 = exception is raised; 0 = it is not)"
i = ieee_flags("clear", "exception", "all", out)
! Clear all
END
Example: Compile and run the preceding example (DetExcFlg.F):
demo% f77 -silent DetExcFlg.F
demo% a.out
Floating-Point Arithmetic
6-9
Highest priority exception is: underflow
invalid divide overflo underflo inexact
0
0
0
1
1
(1 = exception is raised; 0 = it is not)
demo%
IEEE Extreme Value Functions
The compilers provide a set of functions that can be called to return a special IEEE
extreme value. These values, such as infinity or minimum normal, can be used directly
in an application program.
Example: A convergence test based on the smallest number supported by the
hardware would look like:
IF ( delta .LE. r_min_normal() ) RETURN
The values available are listed in the following table:
TABLE 6–3
6-10
Functions Returning IEEE Values
IEEE Value
Double Precision
Single Precision
infinity
d_infinity()
r_infinity()
quiet NaN
d_quiet_nan()
r_quiet_nan()
signaling
NaN
d_signaling_nan()
r_signaling_nan()
min normal
d_min_normal()
r_min_normal()
min
subnormal
d_min_subnormal()
r_min_subnormal()
max
subnormal
d_max_subnormal()
r_max_subnormal()
max normal
d_max_normal()
r_max_normal()
Fortran Programming Guide ♦ Revision A, February 1999
The two NaN values (quiet and signaling) are unordered and should not be used
in comparisons such as IF(X.ne.r_quiet_nan())THEN... To determine whether
some value is a NaN, use the function ir_isnan(r) or id_isnan(d).
The Fortran names for these functions are listed in these man pages:
4 libm_double(3f)
4 libm_single(3f)
4 ieee_functions(3m)
Also see:
4 ieee_values(3m)
4 The f77_floatingpoint.h header file
Exception Handlers and ieee_handler()
Typical concerns about IEEE exceptions are:
4 What happens when an exception occurs?
4 How do I use ieee_handler() to establish a user function as an exception
handler?
4 How do I write a function that can be used as an exception handler?
4 How do I locate the exception—where did it occur?
Exception trapping to a user routine begins with the system generating a signal on a
floating-point exception. The standard UNIX name for signal: floating-point exception is
SIGFPE. The default situation on SPARC and x86 platforms is not to generate a
SIGFPE when an exception occurs. For the system to generate a SIGFPE, exception
trapping must first be enabled, usually by a call to ieee_handler().
Establishing an Exception Handler Function
To establish a function as an exception handler, pass the name of the function to
ieee_handler(), together with the name of the exception to watch for and the
action to take. Once you establish a handler, a SIGFPE signal is generated whenever
the particular floating-point exception occurs, and the specified function is called.
The form for invoking ieee_handler() is shown in the following table:
Floating-Point Arithmetic
6-11
TABLE 6–4
Arguments for ieee_handler(action, exception, handler)
Argument
Type
Possible Values
action
character
get, set, or clear
exception
character
invalid, division, overflow, underflow, or
inexact
handler
Function name
The name of the user handler function or
SIGFPE_DEFAULT, SIGFPE_IGNORE, or
SIGFPE_ABORT
Return value
integer
0 =OK
The routine that calls ieee_handler() should also declare:
#include ’f77_floatingpoint.h’
For f90 programs, declare:
#include ’f90/floatingpoint.h’
The special arguments SIGFPE_DEFAULT, SIGFPE_IGNORE,and SIGFPE_ABORT
are defined in f77_floatingpoint.h and can be used to change the behavior of the
program for a specific exception:
SIGFPE_DEFAULT or
SIGFPE_IGNORE
No action taken when the specified exception occurs.
SIGFPE_ABORT
Program aborts, possibly with dump file, on exception.
Writing User Exception Handler Functions
The actions your exception handler takes are up to you. However, the routine must
be an integer function with three arguments specified as shown:
handler_name( sig, sip, uap )
4 handler_name is the name of the integer function.
4 sig is an integer.
4 sip is a record that has the structure siginfo.
4 uap is not used.
6-12
Fortran Programming Guide ♦ Revision A, February 1999
Example: An exception handler function:
INTEGER FUNCTION hand( sig, sip, uap )
INTEGER sig, location
STRUCTURE /fault/
INTEGER address
INTEGER trapno
END STRUCTURE
STRUCTURE /siginfo/
INTEGER si_signo
INTEGER si_code
INTEGER si_errno
RECORD /fault/ fault
END STRUCTURE
RECORD /siginfo/ sip
location = sip.fault.address
... actions you take ...
END
This f77 example would have to be modified to run on SPARC V9 architectures
(-xarch=v9 or v9a) by replacing all INTEGER declarations within each STRUCTURE
with INTEGER*8.
If the handler routine enabled by ieee_handler() is in Fortran as shown in the
example, the routine should not make any reference to its first argument (sig). This
first argument is passed by value to the routine and can only be referenced as
loc(sig). The value is the signal number.
Detecting an Exception by Handler
The following examples show how to create handler routines to detect floating-point
exceptions.
Example: Detect exception and abort:
demo% cat DetExcHan.f
EXTERNAL myhandler
REAL r / 14.2 /, s / 0.0 /
i = ieee_handler ("set", "division", myhandler )
t = r/s
END
INTEGER FUNCTION myhandler(sig,code,context)
INTEGER sig, code, context(5)
CALL abort()
END
demo% f77 -silent DetExcHan.f
demo% a.out
abort: called
Abort (core dumped)
demo%
Floating-Point Arithmetic
6-13
SIGFPE is generated whenever that floating-point exception occurs. When the
SIGFPE is detected, control passes to the myhandler function, which immediately
aborts. Compile with -g and use dbx to find the location of the exception.
Locating an Exception by Handler
Example: Locate an exception (print address) and abort:
demo% cat LocExcHan.F
#include "f77_floatingpoint.h"
EXTERNAL Exhandler
INTEGER Exhandler, i, ieee_handler
REAL r / 14.2 /, s / 0.0 /, t
C Detect division by zero
i = ieee_handler( "set", "division", Exhandler )
t = r/s
END
INTEGER FUNCTION Exhandler( sig, sip, uap)
INTEGER sig
STRUCTURE /fault/
INTEGER address
END STRUCTURE
STRUCTURE /siginfo/
INTEGER si_signo
INTEGER si_code
INTEGER si_errno
RECORD /fault/ fault
END STRUCTURE
RECORD /siginfo/ sip
WRITE (*,10) sip.si_signo, sip.si_code, sip.fault.address
10
FORMAT("Signal ",i4," code ",i4," at hex address ", Z8 )
CALL abort()
END
demo% f77 -silent -g LocExcHan.F
demo% a.out
Signal
8 code
3 at hex address
11230
abort: called
Abort (core dumped)
demo%
In SPARC V9 environments, replace the INTEGER declarations within each
STRUCTURE with INTEGER*8, and the i4 formats with i8.
In most cases, knowing the actual address of the exception is of little use, except with
dbx:
demo% dbx a.out
(dbx) stopi at 0x11230
Set breakpoint at address
(2) stopi at &MAIN+0x68
(dbx) run
Run program
Running: a.out
(process id 18803)
stopped in MAIN at 0x11230
MAIN+0x68:
fdivs
%f3, %f2, %f2
(dbx) where
Shows the line number of the exception
=>[1] MAIN(), line 7 in "LocExcHan.F"
6-14
Fortran Programming Guide ♦ Revision A, February 1999
(dbx) list 7
Displays the source code line
7
t = r/s
(dbx) cont
Continue after breakpoint, enter handler routine
Signal
8 code
3 at hex address
11230
abort: called
signal ABRT (Abort) in _kill at 0xef6e18a4
_kill+0x8:
bgeu
_kill+0x30
Current function is exhandler
24
CALL abort()
(dbx) quit
demo%
Of course, there are easier ways to determine the source line that caused the error.
However, this example does serve to show the basics of exception handling.
Disabling All Signal Handlers
With f77, some system signal handlers for trapping interrupts, bus errors,
segmentation violations, or illegal instructions are automatically enabled by default.
Although generally you would not want to turn off this default behavior, you can do
so by compiling a C program that sets the global C variable f77_no_handlers to 1
and linking into your executable program:
demo% cat NoHandlers.c
int f77_no_handlers=1 ;
demo% cc -c NoHandlers.c
demo% f77 NoHandlers.o MyProgram.f
Otherwise, by default, f77_no_handlers is 0. The setting takes effect just before
execution is transferred to the user program.
This variable is in the global name space of the program; do not use
f77_no_handlers as the name of a variable anywhere else in the program.
With f90, no signal handlers are on by default.
Retrospective Summary
The ieee_retrospective function queries the floating-point status registers to
find out which exceptions have accrued and a message is printed to standard error to
inform you which exceptions were raised but not cleared. This function is
automatically called by Fortran 77 programs at normal program termination (CALL
EXIT). The message typically looks like this; the format may vary with each compiler
release:
Floating-Point Arithmetic
6-15
Note: IEEE floating-point exception flags raised:
Division by Zero;
IEEE floating-point exception traps enabled:
inexact; underflow; overflow; invalid operation;
See the Numerical Computation Guide, ieee_flags(3M),
ieee_handler(3M)
Fortran 90 programs do not call ieee_retrospective automatically. A Fortran 90
program would need to call ieee_retrospective explicitly (and link with
-lf77compat).
Debugging IEEE Exceptions
In most cases, the only indication that any floating-point exceptions (such as overflow,
underflow, or invalid operation) have occurred is the retrospective summary message
at program termination. Locating where the exception occurred requires exception
trapping be enabled. This can be done by either compiling with the -ftrap=common
option or by establishing an exception handler routine with ieee_handler(). With
exception trapping enabled, run the program from dbx or the Sun WorkShop, using
the dbx catch FPE command to see where the error occurs.
The advantage of recompiling with -ftrap=common is that the source code need
not be modified to trap the exceptions. However, by calling ieee_handler() you
can be more selective as to which exceptions to look at.
Example: Recompiling with -ftrap=common and using dbx:
demo% f77 -g -ftrap=common -silent myprogram.f
demo% dbx a.out
Reading symbolic information for a.out
Reading symbolic information for rtld /usr/lib/ld.so.1
Reading symbolic information for libF77.so.3
Reading symbolic information for libc.so.1
Reading symbolic information for libdl.so.1
(dbx) catch FPE
(dbx) run
Running: a.out
(process id 19739)
signal FPE (floating point divide by zero) in MAIN at line 212 in file "myprogram.f"
212
Z = X/Y
(dbx) print Y
y = 0.0
(dbx)
If you find that the program terminates with overflow and other exceptions, you can
locate the first overflow specifically by calling ieee_handler() to trap just
6-16
Fortran Programming Guide ♦ Revision A, February 1999
overflows. This requires modifying the source code of at least the main program, as
shown in the following example.
Example: Locate an overflow when other exceptions occur:
demo% cat myprog.F
#include ‘‘f77_floatingpoint.h’’
program myprogram
...
ier = ieee_handler(‘set’,’overflow’,SIGFPE_ABORT)
...
demo% f77 -g -silent myprog.F
demo% dbx a.out
Reading symbolic information for a.out
Reading symbolic information for rtld /usr/lib/ld.so.1
Reading symbolic information for libF77.so.3
Reading symbolic information for libc.so.1
Reading symbolic information for libdl.so.1
(dbx) catch FPE
(dbx) run
Running: a.out
(process id 19793)
signal FPE (floating point overflow) in MAIN at line 55 in file "myprog.F"
55
w = rmax * 200.
! Cause of the overflow
(dbx) cont
! Continue execution to completion
Note: IEEE floating-point exception flags raised:
Inexact; Division by Zero; Underflow; ! There were other exceptions
IEEE floating-point exception traps enabled:
overflow;
See the Numerical Computation Guide...
execution completed, exit code is 0
(dbx)
To be selective, the example introduces the #include, which required renaming the
source file with a .F suffix and calling ieee_handler(). You could go further and
create your own handler function to be invoked on the overflow exception to do some
application-specific analysis and print intermediary or debug results before aborting.
Further Numerical Adventures
This section addresses some real world problems that involve arithmetic operations
that may unwittingly generate invalid, division by zero, overflow, underflow, or
inexact exceptions.
For instance, prior to the IEEE standard, if you multiplied two very small numbers
on a computer, you could get zero. Most mainframes and minicomputers behaved
that way. With IEEE arithmetic, gradual underflow expands the dynamic range of
computations.
Floating-Point Arithmetic
6-17
For example, consider a machine with 1.0E-38 as the machine’s epsilon, the smallest
representable value on the machine. Multiply two small numbers:
a = 1.0E-30
b = 1.0E-15
x = a * b
In older arithmetic, you would get 0.0, but with IEEE arithmetic and the same word
length, you get 1.40130E-45. Underflow tells you that you have an answer smaller
than the machine naturally represents. This result is accomplished by “stealing”
some bits from the mantissa and shifting them over to the exponent. The result, a
denormalized number, is less precise in some sense, but more precise in another. The
deep implications are beyond this discussion. If you are interested, consult Computer,
January 1980, Volume 13, Number 1, particularly J. Coonen’s article, “Underflow and
the Denormalized Numbers.”
Most scientific programs have sections of code that are sensitive to roundoff, often in
an equation solution or matrix factorization. Without gradual underflow,
programmers are left to implement their own methods of detecting the approach of
an inaccuracy threshold. Otherwise they must abandon the quest for a robust, stable
implementation of their algorithm.
For more details on these topics, see the Sun Numerical Computation Guide.
Avoiding Simple Underflow
Some applications actually do a lot of computation very near zero. This is common
in algorithms computing residuals or differential corrections. For maximum
numerically safe performance, perform the key computations in extended precision
arithmetic. If the application is a single-precision application, you can perform key
computations in double precision.
Example: A simple dot product computation in single precision:
sum = 0
DO i = 1, n
sum = sum + a(i) * b(i)
END DO
If a(i) and b(i) are very small, many underflows occur. By forcing the
computation to double precision, you compute the dot product with greater accuracy
and do not suffer underflows:
DOUBLE PRECISION sum
DO i = 1, n
sum = sum + dble(a(i)) * dble(b(i))
END DO
result = sum
6-18
Fortran Programming Guide ♦ Revision A, February 1999
On SPARC platforms: You can force a SPARC processor to behave like an older
system with respect to underflow (Store Zero) by adding a call to the library routine
nonstandard_arithmetic() or by compiling the application’s main program
with the -fns option.
Continuing With the Wrong Answer
You might wonder why you would continue a computation if the answer is clearly
wrong. IEEE arithmetic allows you to make distinctions about what kind of wrong
answers can be ignored, such as NaN or Inf. Then decisions can be made based on
such distinctions.
For an example, consider a circuit simulation. The only variable of interest (for the
sake of argument) from a particular 50-line computation is the voltage. Further,
assume that the only values that are possible are +5v, 0, -5v.
It is possible to carefully arrange each part of the calculation to coerce each
sub-result to the correct range:
if computed value is greater than 4.0, return 5.0
if computed value is between -4.0 and +4.0, return 0
if computed value is less than -4.0, return -5.0
Furthermore, since Inf is not an allowed value, you need special logic to ensure that
big numbers are not multiplied.
IEEE arithmetic allows the logic to be much simpler. The computation can be written
in the obvious fashion, and only the final result need be coerced to the correct
value—since Inf can occur and can be easily tested.
Furthermore, the special case of 0/0 can be detected and dealt with as you wish. The
result is easier to read and faster in executing, since you don’t do unneeded
comparisons.
SPARC: Excessive Underflow
If two very small numbers are multiplied, the result underflows.
If you know in advance that the operands in a multiplication (or subtraction) may be
small and underflow is likely, run the calculation in double precision and convert the
result to single precision later.
For example, a dot product loop like this:
real sum, a(maxn), b(maxn)
...
Floating-Point Arithmetic
6-19
do i =1, n
sum = sum + a(i)*b(i)
enddo
where the a(*) and b(*) are known to have small elements, should be run in
double precision to preserve numeric accuracy:
real a(maxn), b(maxn)
double sum
...
do i =1, n
sum = sum + a(i)*dble(b(i))
enddo
Doing so may also improve performance due to the software resolution of excessive
underflows caused by the original loop. However, there is no hard and fast rule here;
experiment with your intensely computational code to determine the most profitable
solutions.
6-20
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
7
Porting
This chapter discusses the porting of programs from other dialects of Fortran to Sun
compilers. VAX VMS Fortran programs compile almost exactly as is with Sun f77;
this is discussed further in the chapter on VMS extensions in the FORTRAN 77
Language Reference Manual.
Note - Porting issues bear mostly upon FORTRAN 77 programs. The Sun Fortran 90
compiler, f90, incorporates few nonstandard extensions, and these are described in
the Fortran User’s Guide.
Time and Date Functions
Library functions that return the time of day or elapsed CPU time vary from system
to system.
The following time functions are not supported directly in the Sun Fortran libraries,
but you can write subroutines to duplicate their functions:
4 Time-of-day in 10h format
4 Date in A10 format
4 Milliseconds of job CPU time
4 Julian date in ASCII
The time functions supported in the Sun Fortran library are listed in the following
table:
7-1
TABLE 7–1
Sun Fortran Time Functions
Name
Function
Man Page
time
Returns the number of seconds elapsed since January,
1, 1970
time(3F)
date
Returns date as a character string
date(3F)
fdate
Returns the current time and date as a character string
fdate(3F)
idate
Returns the current month, day, and year in an integer
array
idate(3F)
itime
Returns the current hour, minute, and second in an
integer array
itime(3F)
ctime
Converts the time returned by the time function to a
character string
ctime(3F)
ltime
Converts the time returned by the time function to
the local time
ltime(3F)
gmtime
Converts the time returned by the time function to
Greenwich time
gmtime(3F)
etime
Single processor: Returns elapsed user and system time
for program execution Multiple processors: Returns the
wall clock time
etime(3F)
dtime
Returns the elapsed user and system time since last
call to dtime
dtime(3F)
date_and_time Returns date and time in character and numeric form
date_and_time(3F)
For details, see Fortran Library Reference Manual or the individual man pages for
these functions. The routines listed in the following table provide compatibility with
VMS Fortran system routines idate and time. To use these routines, you must
include the -lV77 option on the f77 command line, in which case you also get
these VMS versions instead of the standard f77 versions.
7-2
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 7–2
Summary: Nonstandard VMS Fortran System Routines
Name
Definition
Calling Sequence
Argument Type
idate
Date as day,
month, year
call idate( d, m, y )
integer
time
Current time as
hhmmss
call time( t )
character*8
Note - The date(3F) routine and the VMS version of idate(3F) cannot be Year 2000
safe because they return 2-digit values for the year. Programs that compute time
duration by subtracting dates returned by these routines will compute erroneous
results after December 31, 1999. The Fortran 90 routine date_and_time(3F) is available
for both FORTRAN 77 and Fortran 90 programs, and should be used instead. See the
Fortran Library Reference Manual for details.
The error condition subroutine errsns is not provided, because it is totally specific
to the VMS operating system.
Here is a simple example of the use of these time functions (TestTim.f):
c
c
subroutine startclock
common / myclock / mytime
integer mytime, time
mytime = time()
return
end
function wallclock
integer wallclock
common / myclock / mytime
integer mytime, time, newtime
newtime = time()
wallclock = newtime -- mytime
mytime = newtime
return
end
integer wallclock, elapsed
character*24 greeting
real dtime, timediff, timearray(2)
print a heading
call fdate( greeting )
print*, "
Hello, Time Now Is: ", greeting
print*,
"See how long "sleep 4" takes, in seconds"
call startclock
call system( "sleep 4" )
elapsed = wallclock()
print*, "Elapsed time for sleep 4 was: ", elapsed," seconds"
now test the cpu time for some trivial computing
timediff = dtime( timearray )
q = 0.01
do 30 i = 1, 1000
Porting
7-3
q = atan( q )
continue
timediff = dtime( timearray )
print*, "atan(q) 1000 times took: ", timediff ," seconds"
end
30
Running this program produces the following results:
demo% TimeTest
Hello, Time Now Is: Mon Feb 12 11:53:54 1996
See how long "sleep 4" takes, in seconds
Elapsed time for sleep 4 was:
5 seconds
atan(q) 1000 times took:
2.26550E-03 seconds
demo%
Formats
Some f77 and f90 format edit descriptors can behave differently on other systems.
Here are some format specifiers that f77 treats differently than some other
implementations:
4 A – Alphanumeric conversion. Used with character type data elements. In
FORTRAN 77, this specifier worked with any variable type. f77 supports the
older usage, up to four characters to a word.
4 $ – Suppresses newline character output.
4 R – Sets an arbitrary radix for the I formats that follow in the descriptor.
4 SU – Selects unsigned output for following I formats. For example, you can
convert output to either hexadecimal or octal with the following formats, instead
of using the Z or O edit descriptors:
10
20
FORMAT( SU, 16R, I4 )
FORMAT( SU, 8R, I4 )
Carriage-Control
Fortran carriage-control grew out of the capabilities of the equipment used when
Fortran was originally developed. For similar historical reasons, an operating system
derived from the UNIX operating system, does not have Fortran carriage control, but
you can simulate it in two ways.
7-4
Fortran Programming Guide ♦ Revision A, February 1999
4 Use the asa filter to transform Fortran carriage-control conventions into the UNIX
carriage-control format (see the asa (1) man page) before printing files with the
lpr command.
4 f77: For simple jobs, use OPEN(N, FORM=’PRINT’) to enable single or double
spacing, formfeed, and stripping off of column one. It is legal to reopen unit 6 to
change the form parameter to PRINT. For example:
OPEN( 6, FORM="PRINT")
You can use lp(1) to print a file that is opened in this manner.
Working With Files
Early Fortran systems did not use named files, but did provide a command line
mechanism to equate actual file names with internal unit numbers. This facility can
be emulated in a number of ways, including standard UNIX redirection.
Example: Redirecting stdin to redir.data (using csh(1)):
demo% cat redir.data
The data file
9 9.9
demo% cat redir.f
The source file
read(*,*) i, z
print *, i, z
stop
end
The program reads standard input
demo% f77 -silent -o redir redir.f The compilation step
demo% redir < redir.data
Run with redirection reads data file
9 9.90000
demo%
Porting From Scientific Mainframes
If the application code was originally developed for 64-bit (or 60-bit) mainframes
such as CRAY or CDC, you might want to compile these codes with the
f77 -xtypemap option:“-xtypemap=real:64,double:128,integer:64 ”
Porting
7-5
to preserve the expected precision of the original. This option automatically
promotes all default REAL variables to REAL*8, default DOUBLE to REAL*16, and
COMPLEX to COMPLEX*16. Only undeclared variables or variables declared as simply
REAL or DOUBLE or DOUBLE PRECISION or COMPLEX are promoted; variables
declared explicitly (for example, REAL*4) are not promoted.
On x86, or for better performance on SPARC,
use:“-xtypemap=real:64,double:64,integer:64 ”
which does not promote default DOUBLE PRECISION.
The -xtypemap option, is preferred over -dbl and -r8 and -i2. See the Fortran
User’s Guide and the f77(1) man pages for details.
To further recreate the original mainframe environment, it is probably preferable to
stop on overflows, division by zero, and invalid operations. Compile the main
program with -ftrap=common to ensure this.
Data Representation
The FORTRAN 77 Language Reference Manual, Fortran User’s Guide, and the Sun
Numerical Computation Guide discuss in detail the hardware representation of data
objects in Fortran. Differences between data representations across systems and
hardware platforms usually generate the most significant portability problems.
The following issues should be noted:
4 Sun adheres to the IEEE Standard 754 for floating-point arithmetic. Therefore, the
first four bytes in a REAL*8 are not the same as in a REAL*4.
4 The default sizes for reals, integers, and logicals are described in the FORTRAN 77
standard, except when these default sizes are changed by the -xtypemap= option
(or by -i2, -dbl, or -r8).
4 Character variables can be freely mixed and equivalenced to variables of other
types, but be careful of potential alignment problems.
4 f77 IEEE floating-point arithmetic does raise exceptions on overflow or divide by
zero but does not signal SIGFPE or trap by default. It does deliver IEEE
indeterminate forms in cases where exceptions would otherwise be signaled. This
is explained in the Floating Point Arithmetic chapter of this Guide.
4 The extreme finite, normalized values can be determined. See libm_single(3F)
and libm_double(3F). The indeterminate forms can be written and read, using
formatted and list-directed I/O statements.
7-6
Fortran Programming Guide ♦ Revision A, February 1999
Hollerith Data
Many “dusty-deck” Fortran applications store Hollerith ASCII data into numerical
data objects. With the 1977 Fortran standard (and Fortran 90), the CHARACTER data
type was provided for this purpose and its use is recommended. You can still
initialize variables with the older Fortran Hollerith (nH) feature, but this is not
standard practice. The following table indicates the maximum number of characters
that will fit into certain data types. (In this table, boldfaced data types indicate
default types subject to promotion by the f77 command-line flags -dbl, -r8, or
-xtypemap= ).
TABLE 7–3
f77: Maximum Characters in Data Types
Maximum Number of Standard ASCII Characters
Data Type
No -i2,
-i4, -r8,
-dbl
-i2
-i4
-r8
-dbl
BYTE
1
1
1
1
1
COMPLEX
8
8
8
16
16
COMPLEX*16 16
16
16
16
16
COMPLEX*32 32
32
32
32
32
DOUBLE
COMPLEX
16
16
16
32
32
DOUBLE
PRECISION
8
8
8
16
16
INTEGER
4
2
4
4
8
INTEGER*2
2
2
2
2
2
INTEGER*4
4
4
4
4
4
INTEGER*8
8
8
8
8
8
LOGICAL
4
2
4
4
8
LOGICAL*1
1
1
1
1
1
LOGICAL*2
2
2
2
2
2
Porting
7-7
TABLE 7–3
f77: Maximum Characters in Data Types
(continued)
Maximum Number of Standard ASCII Characters
Data Type
No -i2,
-i4, -r8,
-dbl
-i2
-i4
-r8
-dbl
LOGICAL*4
4
4
4
4
4
LOGICAL*8
8
8
8
8
8
REAL
4
4
4
8
8
REAL*4
4
4
4
4
4
REAL*8
8
8
8
8
8
REAL*16
16
16
16
16
16
When storing standard ASCII characters with normal Fortran:
4 With -r8, unspecified size INTEGER and LOGICAL do not hold double.
4 With -dbl, unspecified size INTEGER and LOGICAL do hold double.
The storage is allocated with both options, but it is unavailable in normal Fortran
with -r8.
Options -i2, -r8, and -dbl are now considered obsolete; use -xtypemap instead.
Example: Initialize variables with Hollerith:
demo% cat FourA8.f
double complex x(2)
data x /16Habcdefghijklmnop, 16Hqrstuvwxyz012345/
write( 6, "(4A8, "!")" ) x
end
demo% f77 -silent -o FourA8 FourA8.f
demo% FourA8
abcdefghijklmnopqrstuvwxyz012345!
demo%
If you pass Hollerith constants as arguments, or if you use them in expressions or
comparisons, they are interpreted as character-type expressions.
If needed, you can initialize a data item of a compatible type with a Hollerith and
then pass it to other routines.
7-8
Fortran Programming Guide ♦ Revision A, February 1999
Example:
program respond
integer yes, no
integer ask
data yes, no / 3hyes, 2hno /
if ( ask() .eq. yes ) then
print *, "You may proceed!"
else
print *, "Request Rejected!"
endif
end
10
20
integer function ask()
double precision solaris, response
integer yes, no
data yes, no / 3hyes, 2hno /
data solaris/ 7hSOLARIS/
format( "What system? ", $ )
format( a8 )
write( 6, 10 )
read ( 5, 20 ) response
ask = no
if ( response .eq. solaris ) ask = yes
return
end
Nonstandard Coding Practices
As a general rule, porting an application program from one system and compiler to
another can be made easier by eliminating any nonstandard coding. Optimizations
or work-arounds that were successful on one system might only obscure and confuse
compilers on other systems. In particular, optimized hand-tuning for one particular
architecture can cause degradations in performance elsewhere. This is discussed later
in the chapters on performance and tuning. However, the following issues are worth
considering with regards to porting in general.
Uninitialized Variables
Some systems automatically initialize local and COMMON variables to zero or some
"not-a-number" (NaN) value. However, there is no standard practice, and programs
should not make assumptions regarding the initial value of any variable. To assure
maximum portability, a program should initialize all variables.
Porting
7-9
Aliasing Across Calls
Aliasing occurs when the same storage address is referenced by more than one
name. This happens when actual arguments to a subprogram overlap between
themselves or between COMMON variables within the subprogram. For example,
arguments X and Z refer to the same storage locations, as do B and H:
COMMON /INS/B(100)
REAL S(100), T(100)
...
CALL SUB(S,T,S,B,100)
...
SUBROUTINE SUB(X,Y,Z,H,N)
REAL X(N),Y(N),Z(N),H(N)
COMMON /INS/B(100)
...
Avoid aliasing in this manner in all portable code. The results on some systems and
with higher optimization levels could be unpredictable.
Obscure Optimizations
Legacy codes may contain source-code restructurings of ordinary computational DO
loops intended to cause older vectorizing compilers to generate optimal code for a
particular architecture. In most cases, these restructurings are no longer needed and
may degrade the portability of a program. Two common restructurings are
strip-mining and loop unrolling.
Strip-Mining
Fixed-length vector registers on some architectures led programmers to manually
“strip-mine” the array computations in a loop into segments:
REAL TX(0:63)
...
DO IOUTER = 1,NX,64
DO IINNER = 0,63
TX(IINNER) = AX(IOUTER+IINNER) * BX(IOUTER+IINNER)/2.
QX(IOUTER+IINNER) = TX(IINNER)**2
END DO
END DO
Strip-mining is no longer appropriate with modern compilers; the loop can be
written much less obscurely as:
DO IX = 1,N
TX = AX(I)*BX(I)/2.
QX(I) = TX**2
7-10
Fortran Programming Guide ♦ Revision A, February 1999
END DO
Loop Unrolling
Unrolling loops by hand was a typical source-code optimization technique before
compilers were available that could perform this restructuring automatically. A loop
written as:
DO
K = 1, N-5, 6
J = 1, N
DO I = 1,N
A(I,J) = A(I,J) +
*
+
*
+
*
+
*
+
*
+
END DO
END DO
END DO
DO
KK = K,N
DO
J =1,N
DO I =1,N
A(I,J) = A(I,J) +
END DO
END DO
END DO
DO
B(I,K )
B(I,K+1)
B(I,K+2)
B(I,K+3)
B(I,K+4)
B(I,K+5)
*
*
*
*
*
*
C(K ,J)
C(K+1,J)
C(K+2,J)
C(K+3,J)
C(K+4,J)
C(K+5,J)
B(I,KK) * C(KK,J)
should be rewritten the way it was originally intended:
DO
K = 1,N
J = 1,N
DO I = 1,N
A(I,J) = A(I,J) + B(I,K) * C(K,J)
END DO
END DO
END DO
DO
Troubleshooting
Here are a few suggestions for what to try when programs ported to Sun Fortran do
not run as expected.
Porting
7-11
Results Are Close, but Not Close Enough
Try the following:
4 Pay attention to the size and the engineering units. Numbers very close to zero
can appear to be different, but the difference is not significant, especially if this
number is the difference between two large numbers, such as the distance across
the continent in feet, as calculated on two different computers. For example,
1.9999999e-30 is very near -9.9992112e-33, even though they differ in sign.
VAX math is not as good as IEEE math, and even different IEEE processors may
differ. This is especially true if the mathematics involves many trigonometric
functions. These functions are much more complicated than one might think, and
the standard defines only the basic arithmetic functions. There can be subtle
differences, even between IEEE machines. Review the Floating-Point Arithmetic
chapter in this Guide.
4 Try running with a call nonstandard_arithmetic(). Doing so can also
improve performance considerably, and make your Sun workstation behave more
like a VAX system. If you have access to a VAX or some other system, run it there
also. It is quite common for many numerical applications to produce slightly
different results on each floating-point implementation.
4 Check for NaN, +Inf, and other signs of probable errors. See the Floating-Point
Arithmetic chapter in this Guide, or the man page ieee_handler(3m) for instructions
on how to trap the various exceptions. On most machines, these exceptions simply
abort the run.
4 Two numbers can differ by 6 x 1029 and still have the same floating-point form.
Here is an example of different numbers, with the same representation:
real*4 x,y
x=99999990e+29
y=99999996e+29
write (*,10), x, x
10
format("99,999,990 x 10^29 = ", e14.8, " = ", z8)
write(*,20) y, y
20
format("99,999,996 x 10^29 = ", e14.8, " = ", z8)
end
The output is:
99,999,990 x 10^29 = 0.99999993E+37 = 7cf0bdc1
99,999,996 x 10^29 = 0.99999993E+37 = 7cf0bdc1
In this example, the difference is 6 x 1029. The reason for this indistinguishable,
wide gap is that in IEEE single-precision arithmetic, you are guaranteed only six
decimal digits for any one decimal-to-binary conversion. You may be able to
convert seven or eight digits correctly, but it depends on the number.
7-12
Fortran Programming Guide ♦ Revision A, February 1999
Program Fails Without Warning
If the program fails without warning and runs different lengths of time between
failures, then:
4 Compile with minimal optimization (-O1). If the program then works, compile
only selective routines with higher optimization levels.
4 Understand that optimizers must make assumptions about the program.
Nonstandard coding or constructs can cause problems. Almost no optimizer
handles all programs at all levels of optimization.
Porting
7-13
7-14
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
8
Performance Profiling
This chapter describes how to measure and display program performance. Knowing
where a program is spending most of its compute cycles and how efficiently it uses
system resources is a prerequisite for performance tuning.
The time Command
The simplest way to gather basic data about program performance and resource
utilization is to use the time (1) command or, in csh, the set time command.
Running the program with the time command prints a line of timing information on
program termination.
demo% time myprog
The Answer is: 543.01
6.5u 17.1s 1:16 31% 11+21k 354+210io 135pf+0w
demo%
The interpretation is:
user system wallclock resources memory I/O paging
4 user – 6.5 seconds in user code, approximately
4 system – 17.1 seconds in system code for this task, approximately
4 wallclock – 1 minute 16 seconds to complete
4 resources – 31% of system resources dedicated to this program
4 memory – 11 kilobytes of shared program memory, 21 kilobytes of private data
memory
8-1
4 I/O – 354 reads, 210 writes
4 paging – 135 page faults, 0 swapouts
Multiprocessor Interpretation of time Output
Timing results are interpreted in a different way when the program is run in parallel
in a multiprocessor environment. Since /bin/time accumulates the user time on
different threads, only wall clock time is used.
Since the user time displayed includes the time spent on all the processors, it can be
quite large and is not a good measure of performance. A better measure is the real
time, which is the wall clock time. This also means that to get an accurate timing of a
parallelized program you must run it on a quiet system dedicated to just your
program.
The gprof Profiling Command
The gprof(1) command provides a detailed postmortem analysis of program timing
at the subprogram level, including how many times a subprogram was called, who
called it, whom it called, and how much time was spent in the routine and by the
routines it called.
To enable gprof profiling, compile and link the program with the -pg option:
demo% f77 -o Myprog -fast -pg Myprog.f ...
demo% Myprog
demo% gprof Myprog
The program must complete normally for gprof to obtain meaningful timing
information.
At program termination, the file gmon.out is automatically written in the working
directory. This file contains the profiling data that will be interpreted by gprof.
Invoking gprof produces a report on standard output. An example is shown on the
next pages. Not only the routines in your program are listed but also the library
procedures and the routines they call.
The report is mostly two profiles of how the total time is distributed across the
program procedures: the call graph and the flat profile. They are preceded by an
explanation of the column labels, followed by an index. (The gprof -b option
eliminates the explanatory text; see the gprof(1) man page for other options that can
be used to limit the amount of output generated.)
8-2
Fortran Programming Guide ♦ Revision A, February 1999
In the graph profile, each procedure (subprogram, procedure) is presented in a
call-tree representation. The line representing a procedure in its call-tree is called the
function line, and is identified by an index number in the leftmost column, within
square brackets; the lines above it are the parent lines; the lines below it, the
descendant lines.
--------------------------------------------------------------parent line
caller 1
parent line
caller 2
....
[n]
time
function line
function name
descendant line
called 1
descendant line
called 2
....
---------------------------------------------------------------
The call graph profile is followed by a flat profile that provides a routine-by-routine
overview. An (edited) example of gprof output follows.
Note - User-defined functions appear with their Fortran names followed by an
underscore. Library routines appear with leading underscores.
The call graph profile:
granularity: each sample hit covers 2 byte(s) for 0.08% of 12.78 seconds
index
[3]
%time
99.1
self descendents
0.00
0.00
0.92
0.62
0.11
0.02
0.00
0.00
0.00
12.66
12.66
10.99
0.00
0.00
0.00
0.00
0.00
0.00
called/total
called+self
called/total
1/1
1
1000/1000
2000/2001
1000/1000
1000/1000
1/1
2/2
1/1
parents
name
index
children
main [1]
MAIN_ [3]
diffr_ [4]
code_ [9]
shock_ [11]
bndry_ [14]
init_ [24]
output_ [40]
input_ [47]
-----------------------------------------------
[4]
93.2
0.92
0.92
1.11
1.29
1.17
10.99
10.99
4.52
2.91
0.00
1000/1000
1000
3000/3000
3000/6000
3000/3000
MAIN_ [3]
diffr_ [4]
deriv_ [7]
cheb1_ [5]
dissip_ [8]
-----------------------------------------------
[5]
65.7
1.29
1.29
2.58
5.81
0.00
2.91
2.91
5.81
0.00
0.00
3000/6000
3000/6000
6000
6000/6000
128/321
deriv_ [7]
diffr_ [4]
cheb1_ [5]
fftb_ [6]
cos [21]
Performance Profiling 8-3
0.00
0.00
128/192
__sin [279]
-----------------------------------------------
[6]
45.5
5.81
5.81
0.00
0.00
0.00
0.00
0.00
0.00
6000/6000
6000
64/321
64/192
cheb1_ [5]
fftb_ [6]
cos [21]
__sin [279]
----------------------------------------------...
The flat profile overview:
granularity: each sample hit covers 2 byte(s) for 0.08% of 12.84 seconds
% cumulative
time
seconds
45.2
5.81
20.1
8.39
9.1
9.56
8.6
10.67
7.1
11.58
4.8
12.20
2.5
12.53
0.9
12.64
...
self
seconds
5.81
2.5
1.17
1.11
0.92
0.62
0.33
0.11
calls
6000
6000
3000
3000
1000
2001
69000
1000
self
ms/call
0.97
0.43
0.39
0.37
0.92
0.31
0.00
0.11
total
ms/call
0.97
1.40
0.39
1.88
11.91
0.31
0.00
0.11
name
fftb_
cheb1_
dissip_
deriv_
diffr_
code_
__exp
shock_
[6]
[5]
[8]
[7]
[4]
[9]
[10]
[11]
4 Function Line.
The function line [5] in the preceding example reveals that:
4 cheb1 was called 6000 times— 3000 from deriv, 3000 from diffr.
4 2.58 seconds were spent in cheb1 itself.
4 5.81 seconds were spent in routines called by cheb1.
4 65.7% of the execution time of the program was within cheb1.
4 Parent Lines.
The parent lines above [5] indicate that cheb1 was called from two routines,
deriv and diffr. The timings on these lines show how much time was spent in
cheb1 when it was called from each of these routines.
4 Descendant Lines.
The lines below the function line indicate the routines called from cheb1,
fftb,sin, and cos. The library sine function is called indirectly.
4 Flat Profile.
Function names appear on the right. The profile is sorted by percentage of total
execution time.
8-4
Fortran Programming Guide ♦ Revision A, February 1999
Overhead Considerations
Profiling (compiling with the -pg option) may greatly increase the running time of a
program. This is due to the extra overhead required to clock program performance
and subprogram calls. Profiling tools like gprof attempt to subtract an approximate
overhead factor when computing relative runtime percentages. All other timings
shown may not be accurate due to UNIX and hardware timekeeping inaccuracies.
Programs with short execution times are the most difficult to profile because the
overhead may be a significant fraction of the total execution time. The best practice is
to choose input data for the profiling run that will result in a realistic test of the
program’s performance. If this is not possible, consider enclosing the main
computational part of the program within a loop that effectively runs the program N
times. Estimate actual performance by dividing the profile results by N.
The Fortran library includes two routines that return the total time used by the
calling process. See the man pages for dtime(3F) and etime(3F).
The tcov Profiling Command
The tcov(1) command, when used with programs compiled with the -a, -xa, or
-xprofile=tcov options, produces a statement-by-statement profile of the source
code showing which statements executed and how often. It also gives a summary of
information about the basic block structure of the program.
There are two implementations of tcov coverage analysis. The original tcov is
invoked by the -a or -xa compiler options. Enhanced statement level coverage is
invoked by the -xprofile=tcov compiler option and the -x tcov option. In
either case, the output is a copy of the source files annotated with statement
execution counts in the margin. Although these two versions of tcov are essentially
the same as far as the Fortran user is concerned (most of the enhancements apply to
C++ programs), there will be some performance improvement with the newer style.
“Old Style” tcov Coverage Analysis
Compile the program with the -a (or -xa) option. This produces the file
$TCOVDIR/file.d for each source .f file in the compilation. (If environment variable
$TCOVDIR is not set at compile time, the .d files are stored in the current directory.)
Run the program (execution must complete normally). This produces updated
information in the .d files. To view the coverage analysis merged with the individual
source files, run tcov on the source files. The annotated source files are named
$TCOVDIR/file.tcov for each source file.
Performance Profiling 8-5
The output produced by tcov shows the number of times each statement was
actually executed. Statements that were not executed are marked with ####-> to the
left of the statement.
Here is a simple example:
demo% f77 -a -o onetwo -silent one.f two.f
demo% onetwo
... output from program
demo% tcov one.f two.f
demo% cat one.tcov two.tcov
program one
1 ->
do i=1,10
10 ->
call two(i)
end do
1 ->
end
Top 10 Blocks
Line
Count
3
10
2
1
5
1
3
3
100.00
12
4.00
10 ->
Basic blocks in this file
Basic blocks executed
Percent of the file executed
Total basic block executions
Average executions per basic block
subroutine two(i)
print*, "two called", i
return
end
Top 10 Blocks
Line
Count
2
10
1
1
100.00
10
10.00
Basic blocks in this file
Basic blocks executed
Percent of the file executed
Total basic block executions
Average executions per basic block
demo%
“New Style” Enhanced tcov Analysis
To use new style tcov, compile with -xprofile=tcov. When the program is run,
coverage data is stored in program.profile/tcovd, where program is the name of
the executable file. (If the executable were a.out, a.out.profile/tcovd would
be created.)
8-6
Fortran Programming Guide ♦ Revision A, February 1999
Run tcov -x dirname source_files to create the coverage analysis merged with each
source file. The report is written to file.tcov in the current directory.
Running a simple example:
demo% f77 -o onetwo -silent -xprofile=tcov one.f two.f
demo% onetwo
... output from program
demo% tcov -x onetwo.profile one.f two.f
demo% cat one.f.tcov two.f.tcov
program one
1 ->
do i=1,10
10 ->
call two(i)
end do
1 ->
end
.....etc
demo%
Environment variables $SUN_PROFDATA and $SUN_PROFDATA_DIR can be used to
specify where the intermediary data collection files are kept. These are the *.d and
tcovd files created by old and new style tcov, respectively.
Each subsequent run accumulates more coverage data into the tcovd file. Data for
each object file is zeroed out the first time the program is executed after the
corresponding source file has been recompiled. Data for the entire program is zeroed
out by removing the tcovd file.
These environment variables can be used to separate the collected data from different
runs. With these variables set, the running program writes execution data to the files
in $SUN_PROFDATA_DIR/$SUN_PROFDATA/.
Similarly, the directory that tcov reads is specified by tcov -x $SUN_PROFDATA.
If $SUN_PROFDATA_DIR is set, tcov will prepend it, looking for files in
$SUN_PROFDATA_DIR/$SUN_PROFDATA/, and not in the working directory.
For the details, see the tcov(1) man page.
f77 I/O Profiling
You can obtain a report about how much data was transferred by your program. For
each Fortran unit, the report shows the file name, the number of I/O statements, the
number of bytes, and some statistics on these items.
To obtain an I/O profiling report, insert calls to the library routines start_iostats and
end_iostats around the parts of the program you wish to measure. (A call to
end_iostats is required if the program terminates with an END or STOP statement
rather than a CALL EXIT.)
Performance Profiling 8-7
Note - The I/O statements profiled are: READ, WRITE, PRINT, OPEN, CLOSE,
INQUIRE, BACKSPACE, ENDFILE, and REWIND. The runtime system opens stdin,
stdout, and stderr before the first executable statement of your program, so you
must explicitly reopen these units after the call to start_iostats.
Example: Profile stdin, stdout, and stderr:
EXTERNAL start_iostats
…
CALL start_iostats
OPEN(5)
OPEN(6)
OPEN(0)
If you want to measure only part of the program, call end_iostats to stop the
process. A call to end_iostats may also be required if your program terminates
with an END or STOP statement rather than CALL EXIT.
The program must be compiled with the -pg option. When the program terminates,
the I/O profile report is produced on the file name.io_stats, where name is the
name of the executable file.
Here is an example:
demo% f77 -o myprog -pg -silent myprog.f
demo% myprog
... output from program
demo% cat myprog.io_stats
INPUT REPORT
1. unit
2. file name
3. input data
4. map
cnt
total
avg
std dev
(cnt)
-----------------------------------------------------------------------0
stderr
0
0
0
0
No
0
0
0
0
5
stdin
2
8
4
0
No
1
8
8
0
6
stdout
0
0
0
0
No
0
0
0
0
19
fort.19
8
48
6
4.276
No
4
48
12
0
20
fort.20
8
48
6
4.276
No
4
48
12
0
21
fort.21
8
48
6
4.276
No
4
48
12
0
22
fort.22
8
48
6
4.276
No
4
48
12
0
OUTPUT REPORT
5. output data
6. blk size 7. fmt 8. direct
cnt
total
avg
std dev
(rec len)
----------------------------------------------------------------------------0
4
40
10
0
-1
Yes
seq
1. unit
8-8
Fortran Programming Guide ♦ Revision A, February 1999
5
6
19
20
21
22
1
0
0
26
6
8
4
8
4
8
4
8
4
40
0
0
248
248
48
48
48
48
48
48
48
48
40
0
0
9.538
41.33
6
12
6
12
6
12
6
12
0
0
0
1.63
3.266
4.276
0
4.276
0
4.276
0
4.276
0
-1
Yes
seq
-1
Yes
seq
500548
Yes
seq
503116
No
seq
503116
Yes
dir
12)
dir
(
12)
(
503116
No
…
Each pair of lines in the report displays information about an I/O unit. One section
shows input operations and another shows output. The first line of a pair displays
statistics on the number of data elements transferred before the unit was closed. The
second row of statistics is based on the number of I/O statements processed.
In the example, there were 6 calls to write a total of 26 data elements to standard
output. A total of 248 bytes was transferred. The display also shows the average and
standard deviation in bytes transferred per I/O statement (9.538 and 1.63,
respectively), and the average and standard deviation per I/O statement call (42.33
and 3.266, respectively).
The input report also contains a column to indicate whether a unit was memory
mapped or not. If mapped, the number of mmap() calls is recorded in parentheses in
the second row of the pair.
The output report indicates block sizes, formatting, and access type. A file opened
for direct access shows its defined record length in parentheses in the second row of
the pair.
Note - Compiling with environment variable LD_LIBRARY_PATH set might disable
I/O profiling, which relies on its profiling I/O library being in a standard location.
Performance Profiling 8-9
8-10
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
9
Performance and Optimization
This chapter considers some optimization techniques that may improve the
performance of numerically intense Fortran programs. Proper use of algorithms,
compiler options, library routines, and coding practices can bring significant
performance gains. This discussion does not discuss cache, I/O, or system
environment tuning. Parallelization issues are treated in the next chapter.
Some of the issues considered here are:
4 Compiler options that may improve performance
4 Compiling with feedback from runtime performance profiles
4 Use of optimized library routines for common procedures
4 Coding strategies to improve performance of key loops
The subject of optimization and performance tuning is much too complex to be
treated exhaustively here. However, this discussion should provide the reader with a
useful introduction to these issues. A list of books that cover the subject much more
deeply appears at the end of the chapter.
Optimization and performance tuning is an art that depends heavily on being able to
determine what to optimize or tune.
Choice of Compiler Options
Choice of the proper compiler options is the first step in improving performance. Sun
compilers offer a wide range of options that affect the object code. In the default
case, where no options are explicitly stated on the compile command line, most
options are off. To improve performance, these options must be explicitly selected.
9-1
Performance options are normally off by default because most optimizations force
the compiler to make assumptions about a user’s source code. Programs that
conform to standard coding practices and do not introduce hidden side effects
should optimize correctly. However, programs that take liberties with standard
practices might run afoul of some of the compiler’s assumptions. The resulting code
might run faster, but the computational results might not be correct.
Recommended practice is to first compile with all options off, verify that the
computational results are correct and accurate, and use these initial results and
performance profile as a baseline. Then, proceed in steps—recompiling with
additional options and comparing execution results and performance against the
baseline. If numerical results change, the program might have questionable code,
which needs careful analysis to locate and reprogram.
If performance does not improve significantly, or degrades, as a result of adding
optimization options, the coding might not provide the compiler with opportunities
for further performance improvements. The next step would then be to analyze and
restructure the program at the source code level to achieve better performance.
Performance Option Reference
The compiler options listed in the following table provide the user with a repertoire
of strategies to improve the performance of a program over default compilation.
Only some of the compilers’ more potent performance options appear in the table. A
more complete list can be found in the Fortran User’s Guide.
TABLE 9–1
9-2
Some Effective Performance Options
Action
Option
Uses various optimization options together
-fast
Sets compiler optimization level to n
-On (-O = -O3)
Specifies target hardware
-xtarget=sys
Optimizes using performance profile data (with -O5)
xprofile=use
Unrolls loops by n
-unroll=n
Permits simplifications and optimization of floating-point
-fsimple=1|2
Performs dependency analysis to optimize loops
-depend
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 9–1
Some Effective Performance Options
(continued)
Some of these options increase compilation time because they invoke a deeper
analysis of the program. Some options work best when routines are collected into
files along with the routines that call them (rather than splitting each routine into its
own file); this allows the analysis to be global.
-fast
This single option selects a number of performance options that, working together,
produce object code optimized for execution speed without an excessive increase in
compilation time.
The options selected by -fast are subject to change from one release to another, and
not all are available on each platform:
4 -native generates code optimized for the host architecture.
4 -O4 sets optimization level.
4 -libmil inlines calls to some simple library functions.
4 -fsimple=1 simplifies floating-point code (SPARC only).
4 -dalign uses faster, double word loads and stores (SPARC only).
4 -xlibmopt use optimized libm math library (SPARC only).
4 -fns -ftrap=%none turns off all trapping.
4 -depend analyze loops for data dependencies (SPARC only).
4 -nofstore disables forcing precision on expressions (x86 only).
-fast provides a quick way to engage much of the optimizing power of the
compilers. Each of the composite options may be specified individually, and each
may have side effects to be aware of (discussed in the Fortran User’s Guide).
Following -fast with additional options adds further optimizations. For example:
f77 -fast -O5 ...
sets the optimization to level 5 instead of 4.
Note - -fast includes -dalign and -native. These options may have
unexpected side effects for some programs.
-On
No compiler optimizations are performed by the compilers unless a -O option is
specified explicitly (or implicitly with macro options like -fast). In nearly all cases,
Performance and Optimization 9-3
specifying an optimization level for compilation improves program execution
performance. On the other hand, higher levels of optimization increase compilation
time and may significantly increase code size.
For most cases, level -O3 is a good balance between performance gain, code size,
and compilation time. Level -O4 adds automatic inlining of calls to routines
contained in the same source file as the caller routine, among other things. Level -O5
adds more aggressive optimization techniques that would not be applied at lower
levels. In general, levels above -O3 should be specified only to those routines that
make up the most compute-intensive parts of the program and thereby have a high
certainty of improving performance. (There is no problem linking together parts of a
program compiled with different optimization levels.)
PRAGMA OPT=n
Use the C$ PRAGMA SUN OPT=n directive to set different optimization levels for
individual routines in a source file. This directive will override the -On flag on the
compiler command line, but must be used with the -xmaxopt=n flag to set a
maximum optimization level. See the f77(1) and f90(1) man pages for details.
Optimization With Runtime Profile Feedback
The compiler applies its optimization strategies at level O3 and above much more
efficiently if combined with -xprofile=use. With this option, the optimizer is
directed by a runtime execution profile produced by the program (compiled with
-xprofile=collect) with typical input data. The feedback profile indicates to the
compiler where optimization will have the greatest effect. This may be particularly
important with -O5. Here’s a typical example of profile collection with higher
optimization levels:
demo%
demo%
demo%
demo%
f77 -o prg -fast -xprofile=collect prg.f ...
prg
f77 -o prgx -fast -O5 -xprofile=use:prg.profile prg.f ...
prgx
The first compilation in the example generates an executable that produces statement
coverage statistics when run. The second compilation uses this performance data to
guide the optimization of the program.
(See the Fortran User’s Guide for details on -xprofile options.)
-dalign
With -dalign the compiler is able to generate double-word load/store instructions
whenever possible. Programs that do much data motion may benefit significantly
9-4
Fortran Programming Guide ♦ Revision A, February 1999
when compiled with this option. (It is one of the options selected by -fast.) The
double-word instructions are almost twice as fast as the equivalent single word
operations.
However, users should be aware that using -dalign (and therefore -fast) may
cause problems with some programs that have been coded expecting a specific
alignment of data in COMMON blocks. With -dalign, the compiler may add
padding to ensure that all double (and quad) precision data (either REAL or
COMPLEX) are aligned on double-word boundaries, with the result that:
4 COMMON blocks might be larger than expected due to added padding.
4 All program units sharing COMMON must be compiled with -dalign if any one
of them is compiled with -dalign.
For example, a program that writes data by aliasing an entire COMMON block of
mixed data types as a single array might not work properly with -dalign because
the block will be larger (due to padding of double and quad precision variables) than
the program expects.
SPARC: -depend
Adding -depend to optimization levels -O3 and higher (on the SPARC platform)
extends the compiler’s ability to optimize DO loops and loop nests. With this option,
the optimizer analyzes inter-iteration loop dependencies to determine whether or not
certain transformations of the loop structure can be performed. Only loops without
dependencies can be restructured. However, the added analysis might increase
compilation time.
-fsimple=2
Unless directed to, the compiler does not attempt to simplify floating-point
computations (the default is -fsimple=0). With the -fast option, -fsimple=1 is
used and some conservative assumptions are made. Adding -fsimple=2 enables
the optimizer to make further simplifications with the understanding that this might
cause some programs to produce slightly different results due to rounding effects. If
-fsimple level 1 or 2 is used, all program units should be similarly compiled to
ensure consistent numerical accuracy.
-unroll=n
Unrolling short loops with long iteration counts can be profitable for some routines.
However, unrolling can also increase program size and might even degrade
performance of other loops. With n=1, the default, no loops are unrolled
automatically by the optimizer. With n greater than 1, the optimizer attempts to
unroll loops up to a depth of n.
Performance and Optimization 9-5
The compiler’s code generator makes its decision to unroll loops depending on a
number of factors. The compiler might decline to unroll a loop even though this
option is specified with n>1.
If a DO loop with a variable loop limit can be unrolled, both an unrolled version and
the original loop are compiled. A runtime test on iteration count determines whether
or not executing the unrolled loop is inappropriate. Loop unrolling, especially with
simple one or two statement loops, increases the amount of computation done per
iteration and provides the optimizer with better opportunities to schedule registers
and simplify operations. The tradeoff between number of iterations, loop complexity,
and choice of unrolling depth is not easy to determine, and some experimentation
might be needed.
The example that follows shows how a simple loop might be unrolled to a depth of
four with -unroll=4 (the source code is not changed with this option):
Original Loop:
DO I=1,20000
X(I) = X(I) + Y(I)*A(I)
END DO
Unrolled by 4 compiles as if it were written:
DO I=1, 19997,4
TEMP1 = X(I) + Y(I)*A(I)
TEMP2 = X(I+1) + Y(I+1)*A(I+1)
TEMP3 = X(I+2) + Y(I+2)*A(I+2)
X(I+3) = X(I+3) + Y(I+3)*A(I+3)
X(I) = TEMP1
X(I+1) = TEMP2
X(I+2) = TEMP3
END DO
This example shows a simple loop with a fixed loop count. The restructuring is more
complex with variable loop counts.
-xtarget=platform
The performance of some programs might improve if the compiler has an accurate
description of the target computer hardware. When program performance is critical,
the proper specification of the target hardware could be very important. This is
especially true when running on the newer SPARC processors. However, for most
programs and older SPARC processors, the performance gain could be negligible and
a generic specification might be sufficient.
The Fortran User’s Guide lists all the system names recognized by -xtarget=. For
any given system name (for example, ss1000, for SPARCserver 1000), -xtarget
expands into a specific combination of -xarch, -xcache, and -xchip that properly
matches that system. The optimizer uses these specifications to determine strategies
to follow and instructions to generate.
9-6
Fortran Programming Guide ♦ Revision A, February 1999
The special setting -xtarget=native enables the optimizer to compile code
targeted at the host system (the system doing the compilation). This is obviously
useful when compilation and execution are done on the same system. When the
execution system is not known, it is desirable to compile for a generic architecture.
Therefore, -xtarget=generic is the default, even though it might produce
suboptimal performance.
Other Performance Strategies
Assuming that you have experimented with using a variety of optimization options,
compiling your program and measuring actual runtime performance, the next step
might be to look closely at the Fortran source program to see what further tuning
can be tried.
Focusing on just those parts of the program that use most of the compute time, you
might consider the following strategies:
4 Replace handwritten procedures with calls to equivalent optimized libraries.
4 Remove I/O, calls, and unnecessary conditional operations from key loops.
4 Eliminate aliasing that might inhibit optimization.
4 Rationalize tangled, spaghetti-like code to use block IF.
These are some of the good programming practices that tend to lead to better
performance. It is possible to go further, hand-tuning the source code for a specific
hardware configuration. However, these attempts might only further obscure the
code and make it even more difficult for the compiler’s optimizer to achieve
significant performance improvements. Excessive hand-tuning of the source code can
hide the original intent of the procedure and could have a significantly detrimental
effect on performance for different architectures.
Using Optimized Libraries
In most situations, optimized commercial or shareware libraries perform standard
computational procedures far more efficiently than you could by coding them by
hand.
TM
For example, the Sun Performance Library is a suite of highly optimized
mathematical subroutines based on the standard LAPACK, BLAS, FFTPACK,
VFFTPACK, and LINPACK libraries. Performance improvement using these routines
can be significant when compared with hand coding.
Performance and Optimization 9-7
Eliminating Performance Inhibitors
Use the profiling techniques described in Chapter 8 to identify the key computational
parts of the program. Then, carefully analyze the loop or loop nest to eliminate
coding that might either inhibit the optimizer from generating optimal code or
otherwise degrade performance. Many of the nonstandard coding practices that
make portability difficult might also inhibit optimization by the compiler.
Reprogramming techniques that improve performance are dealt with in more detail
in some of the reference books listed at the end of the chapter. Three major
approaches are worth mentioning here:
Removing I/O From Key Loops
I/O within a loop or loop nest enclosing the significant computational work of a
program will seriously degrade performance. The amount of CPU time spent in the
I/O library might be a major portion of the time spent in the loop. (I/O also causes
process interrupts, thereby degrading program throughput.) By moving I/O out of
the computation loop wherever possible, the number of calls to the I/O library can
be greatly reduced.
Eliminating Subprogram Calls
Subroutines called deep within a loop nest could be called thousands of times. Even
if the time spent in each routine per call is small, the total effect might be substantial.
Also, subprogram calls inhibit optimization of the loop that contains them because
the compiler cannot make assumptions about the state of registers over the call.
Automatic inlining of subprogram calls (using -inline=x,y,..z, or -O4) is one way
to let the compiler replace the actual call with the subprogram itself (pulling the
subprogram into the loop). The subprogram source code for the routines that are to
be inlined must be found in the same file as the calling routine.
There are other ways to eliminate subprogram calls:
4 Use statement functions. If the external function being called is a simple math
function, it might be possible to rewrite the function as a statement function or set
of statement functions. Statement functions are compiled in-line and can be
optimized.
4 Push the loop into the subprogram. That is, rewrite the subprogram so that it can
be called fewer times (outside the loop) and operate on a vector or array of values
per call.
9-8
Fortran Programming Guide ♦ Revision A, February 1999
Rationalizing Tangled Code
Complicated conditional operations within a computationally intensive loop can
dramatically inhibit the compiler’s attempt at optimization. In general, a good rule to
follow is to eliminate all arithmetic and logical IF’s, replacing them with block IF’s:
Original Code:
IF(A(I)-DELTA) 10,10,11
10 XA(I) = XB(I)*B(I,I)
XY(I) = XA(I) - A(I)
GOTO 13
11 XA(I) = Z(I)
XY(I) = Z(I)
IF(QZDATA.LT.0.) GOTO 12
ICNT = ICNT + 1
ROX(ICNT) = XA(I)-DELTA/2.
12 SUM = SUM + X(I)
13 SUM = SUM + XA(I)
Untangled Code:
IF(A(I).LE.DELTA) THEN
XA(I) = XB(I)*B(I,I)
XY(I) = XA(I) - A(I)
ELSE
XA(I) = Z(I)
XY(I) = Z(I)
IF(QZDATA.GE.0.) THEN
ICNT = ICNT + 1
ROX(ICNT) = XA(I)-DELTA/2.
ENDIF
SUM = SUM + X(I)
ENDIF
SUM = SUM + XA(I)
Using block IF not only improves the opportunities for the compiler to generate
optimal code, it also improves readability and assures portability.
Further Reading
The following reference books provide more details:
4 Numerical Computation Guide, Sun Microsystems, Inc.
4 Analyzing Program Performance with Sun WorkShop, Sun Microsystems, Inc.
4 FORTRAN Optimization, by Michael Metcalf, Academic Press 1985
4 High Performance Computing, by Kevin Dowd, O’Reilly & Associates, 1993
Performance and Optimization 9-9
9-10
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
10
SPARC: Parallelization
This chapter presents an overview of multiprocessor parallelization and describes the
capabilities of Sun’s Fortran compilers. Implementation differences between f77 and
f90 are noted.
Note - Parallelization features are available only on SPARC platforms with Solaris
2.5.1, 2.6, and Solaris 7 operating environments, and require a Sun WorkShop license.
Essential Concepts
Parallelizing (or multithreading) an application recasts the compiled program to run
on a multiprocessor system. Parallelization enables single tasks, such as a DO loop,
to run over multiple processors with a potentially significant execution speedup.
Before anTMapplication program
can be run efficiently on a multiprocessor system like
TM
the Ultra 60, Enterprise 450, or Ultra HPC 1000, it needs to be multithreaded.
That is, tasks that can be performed in parallel need to be identified and
reprogrammed to distribute their computations.
Multithreading an application can be done manually by making appropriate calls to
the libthread primitives. However, a significant amount of analysis and
reprogramming might be required. (See the Solaris Multithreaded Programming Guide
for more information.)
Sun compilers can automatically generate multithreaded object code to run on
multiprocessor systems. The Fortran compilers focus on DO loops as the primary
language element supporting parallelism. Parallelization distributes the
computational work of a loop over several processors without requiring modifications to
the Fortran source program.
10-1
The choice of which loops to parallelize and how to distribute them can be left
entirely up to the compiler (-autopar), determined explicitly by the programmer
with source code directives (-explicitpar), or done in combination (-parallel).
Note - Programs that do their own (explicit) thread management should not be
compiled with any of the compiler’s parallelization options. Explicit multithreading
(calls to libthread primitives) cannot be combined with routines compiled with
these parallelization options.
Not all loops in a program can be profitably parallelized. Loops containing only a
small amount of computational work (compared to the overhead spent starting and
synchronizing parallel tasks) may actually run more slowly when parallelized. Also,
some loops cannot be safely parallelized at all; they would compute different results
when run in parallel due to dependencies between statements or iterations.
Only explicit Fortran 90 DO loops are candidates for parallelization with f90.
Sun compilers can detect loops that might be safely and profitably parallelized
automatically. However, in most cases, the analysis is necessarily conservative, due to
the concern for possible hidden side effects. (A display of which loops were and
were not parallelized can be produced by the -loopinfo option.) By inserting
source code directives before loops, you can explicitly influence the analysis,
controlling how a specific loop is (or is not) to be parallelized. However, it then
becomes your responsibility to ensure that such explicit parallelization of a loop does
not lead to incorrect results.
Speedups—What to Expect
If you parallelize a program so that it runs over four processors, can you expect it to
take (roughly) one fourth the time that it did with a single processor (a fourfold
speedup)?
Probably not. It can be shown (by Amdahl’s law) that the overall speedup of a
program is strictly limited by the fraction of the execution time spent in code
running in parallel. This is true no matter how many processors are applied. In fact, if c is
the percentage of the execution time run in parallel, the theoretical speedup limit is
100/(100–c); therefore, if only 60% of a program runs in parallel, the maximum
increase in speed is 2.5, independent of the number of processors. And with just four
processors, the theoretical speedup for this program (assuming maximum efficiency)
would be just 1.8 and not 4. With overhead, the actual speedup would be less.
As with any optimization, choice of loops is critical. Parallelizing loops that
participate only minimally in the total program execution time has only minimal
effect. To be effective, the loops that consume the major part of the runtime must be
parallelized. The first step, therefore, is to determine which loops are significant and
to start from there.
10-2
Fortran Programming Guide ♦ Revision A, February 1999
Problem size also plays an important role in determining the fraction of the program
running in parallel and consequently the speedup. Increasing the problem size
increases the amount of work done in loops. A triply nested loop could see a cubic
increase in work. If the outer loop in the nest is parallelized, a small increase in
problem size could contribute to a significant performance improvement (compared
to the unparallelized performance).
Steps to Parallelizing a Program
Here is a very general outline of the steps needed to parallelize an application:
1. Optimize. Use the appropriate set of compiler options to get the best serial
performance on a single processor.
2. Profile. Using typical test data, determine the performance profile of the program.
Identify the most significant loops.
3. Benchmark. Determine that the serial test results are accurate. Use these results
and the performance profile as the benchmark.
4. Parallelize. Use a combination of options and directives to compile and build a
parallelized executable.
5. Verify. Run the parallelized program on a single processor and check results to
find instabilities and programming errors that might have crept in.
6. Test. Make various runs on several processors to check results.
7. Benchmark. Make performance measurements with various numbers of
processors on a dedicated system. Measure performance changes with changes in
problem size (scalability).
8. Repeat steps 4 to 7. Make improvements to parallelization scheme based on
performance.
Data Dependency Issues
Not all loops are parallelizable. Running a loop in parallel over a number of
processors may result in iterations executing out of order. Or the multiple processors
executing the loop in parallel may interfere with each other. These situations arise
whenever there are data dependencies in the loop.
Recurrence
Variables that are set in one iteration of a loop and used in a subsequent iteration
introduce cross-iteration dependencies, or recurrences. Recurrence in a loop requires
that the iterations to be executed in the proper order. For example:
DO I=2,N
A(I) = A(I-1)*B(I)+C(I)
END DO
SPARC: Parallelization
10-3
requires the value computed for A(I) in the previous iteration to be used (as A(I-1))
in the current iteration. To produce results running each iteration in parallel that are
the same as with single processor, iteration I must complete before iteration I+1 can
execute.
Reduction
Reduction operations reduce the elements of an array into a single value. For
example, summing the elements of an array into a single variable involves updating
that variable in each iteration:
DO K = 1,N
SUM = SUM + A(I)*B(I)
END DO
If each processor running this loop in parallel takes some subset of the iterations, the
processors will interfere with each other, overwriting the value in SUM. For this to
work, each processor must execute the summation one at a time, although the order
is not significant.
Certain common reduction operations are recognized and handled as special cases by
the compiler.
Indirect Addressing
Loop dependencies can result from stores into arrays that are indexed in the loop by
subscripts whose values are not known. For example, indirect addressing could be
order dependent if there are repeated values in the index array:
DO L = 1,NW
A(ID(L)) = A(L) + B(L)
END DO
In the preceding, repeated values in ID cause elements in A to be overwritten. In the
serial case, the last store is the final value. In the parallel case, the order is not
determined. The values of A(L) that are used, old or updated, are order dependent.
Data Dependent Loops
You might be able to rewrite a loop to eliminate data dependencies, making it
parallelizable. However, extensive restructuring could be needed.
Some general rules are:
10-4
Fortran Programming Guide ♦ Revision A, February 1999
4 A loop is data independent only if all iterations write to distinct memory locations.
4 Iterations may read from the same locations as long as no one iteration writes to
them.
These are general conditions for parallelization. The compilers’ automatic
parallelization analysis considers additional criteria when deciding whether to
parallelize a loop. However, you can use directives to explicitly force loops to be
parallelized, even loops that contain inhibitors and produce incorrect results.
Parallel Options and Directives Summary
The following table shows the f77 5.0 and f90 2.0 compilation options related to
parallelization.
TABLE 10–1
Parallelization Options
Option
Flag
Automatic (only)
-autopar
Automatic and Reduction
-autopar -reduction
Explicit (only)
-explicitpar
Automatic and Explicit
-parallel
Automatic and Reduction and Explicit
-parallel -reduction
Show which loops are parallelized
-loopinfo
Show warnings with explicit
-vpara
Allocate local variables on stack
-stackvar
Use Sun-style MP directives
-mp=sun
Use Cray-style MP directives
-mp=cray
Notes on these options:
4 -reduction requires -autopar.
4 -autopar includes -depend and loop structure optimization.
4 -parallel is equivalent to -autopar -explicitpar.
4 -noautopar, -noexplicitpar, -noreduction are the negations.
SPARC: Parallelization
10-5
4 Parallelization options can be in any order, but they must be all lowercase.
4 Reduction operations are not analyzed for explicitly parallelized loops.
4 Use of any of the parallelization options requires a WorkShop license.
The following table shows the f77/f90 and f90 parallel directives.
TABLE 10–2
Parallel Directives
Parallel Directive
Purpose
C$PAR TASKCOMMON
Declares a common block private
C$PAR DOALL optional qualifiers
Parallelizes next loop, if possible
C$PAR DOSERIAL
Inhibits parallelization of next loop
C$PAR DOSERIAL*
Inhibits parallelization of loop nest
Number of Processors
The PARALLEL environment variable controls the maximum number of processors
available to the program. The following example shows how to set it:
demo% setenv PARALLEL 4
C shell
-or-
demo$ PARALLEL=4
demo$ export PARALLEL
Bourne/Korn shell
In this example, setting PARALLEL to four enables the execution of a program using
at most four threads. If the target machine has four processors available, the threads
will map to independent processors. If there are fewer than four processors available,
some threads could run on the same processor as others, possibly degrading
performance.
The SunOS command psrinfo(1M) displays a list of the processors available on a
system:
demo% psrinfo
0
on-line
1
on-line
2
on-line
3
on-line
10-6
since
since
since
since
03/18/96
03/18/96
03/18/96
03/18/96
15:51:03
15:51:03
15:51:03
15:51:03
Fortran Programming Guide ♦ Revision A, February 1999
Stacks, Stack Sizes, and Parallelization
The executing program maintains a main memory stack for the parent program and
distinct stacks for each thread. Stacks are temporary memory address spaces used to
hold arguments and AUTOMATIC variables over subprogram invocations.
The default size of the main stack is about 8 megabytes. The Fortran compilers
normally allocate local variables and arrays as STATIC (not on the stack). However,
the -stackvar option forces allocation of all local variables and arrays on the stack
(as if they were AUTOMATIC variables). Use of -stackvar is recommended with
parallelization because it improves the optimizer’s ability to parallelize CALLs in
loops. -stackvar is required with explicitly parallelized loops containing
subprogram calls. (See the discussion of -stackvar in the Fortran User’s Guide.)
The limit command displays the current main stack size as well as setting it:
demo% limit
C shell example
cputime
unlimited
filesize
unlimited
datasize
2097148 kbytes
stacksize
8192 kbytes
<- current main stack size
coredumpsize
0 kbytes
descriptors
64
memorysize
unlimited
demo% limit stacksize 65536
<- set main stack to 64Mb
demo% limit stacksize
stacksize
65536 kbytes
demo$ >ulimit -a
Korn Shell example
time(seconds)
unlimited
file(blocks)
unlimited
data(kbytes)
2097148
stack(kbytes)
8192
coredump(blocks)
0
nofiles(descriptors) 64
vmemory(kbytes)
unlimited
demo$ ulimit -s 65536
demo$ ulimit -s
65536
Each thread of a multithreaded program has its own thread stack. This stack mimics
the main program stack but is unique to the thread. The thread’s PRIVATE arrays
and variables (local to the thread) are allocated on the thread stack. The default size
is 256 kilobytes. The size is set with the STACKSIZE environment variable:
demo% setenv STACKSIZE 8192
<- Set thread stack size to 8 Mb
-ordemo$ STACKSIZE=8192
Bourne/Korn Shell
demo$ export STACKSIZE
C shell
SPARC: Parallelization
10-7
Setting the thread stack size to a value larger than the default may be necessary for
most parallelized Fortran codes. However, it may not be possible to know just how
large to set it, except by trial and error, especially if private/local arrays are
involved. If the stack size is too small for a thread to run, the program will abort
with a segmentation fault.
Automatic Parallelization
With the -autopar and -parallel options, the compilers automatically find DO
loops that can be parallelized effectively. These loops are then transformed to
distribute their iterations evenly over the available processors. The compiler
generates the thread calls needed to make this happen.
Loop Parallelization
The compiler’s dependency analysis transforms a DO loop into a parallelizable task.
The compiler may restructure the loop to split out unparallelizable sections that will
run serially. It then distributes the work evenly over the available processors. Each
processor executes a different chunk of iterations.
For example, with four CPUs and a parallelized loop with 1000 iterations:
Processor 1 executing iterations
1
through
250
Processor 2 executing iterations
251
through
500
Processor 3 executing iterations
501
through
750
Processor 4 executing iterations
751
through
1000
Only loops that do not depend on the order in which the computations are performed
can be successfully parallelized. The compiler’s dependency analysis rejects loops
with inherent data dependencies. If it cannot fully determine the data flow in a loop,
the compiler acts conservatively and does not parallelize. Also, it may choose not to
parallelize a loop if it determines the performance gain does not justify the overhead.
Note that the compiler always chooses to parallelize loops using a chunk
distribution—simply dividing the work in the loop into equal blocks of iterations.
Other distribution schemes may be specified using explicit parallelization directives
described later in this chapter.
10-8
Fortran Programming Guide ♦ Revision A, February 1999
Arrays, Scalars, and Pure Scalars
A few definitions, from the point of view of automatic parallelization, are needed:
4 An array is a variable that is declared with at least one dimension.
4 A scalar is a variable that is not an array.
4 A pure scalar is a scalar variable that is not aliased—not referenced in an
EQUIVALENCE or POINTER statement.
Example: Array/scalar:
dimension a(10)
real m(100,10), s, u, x, z
equivalence ( u, z )
pointer ( px, x )
s = 0.0
...
Both m and a are array variables; s is pure scalar. The variables u, x, z, and px are
scalar variables, but not pure scalars.
Automatic Parallelization Criteria
DO loops that have no cross-iteration data dependencies are automatically parallelized
by -autopar or -parallel. The general criteria for automatic parallelization are:
4 DO loops are parallelized, but not DO WHILE or Fortran 90 array operations.
4 The values of array variables for each iteration of the loop must not depend on the
values of array variables for any other iteration of the loop.
4 Calculations within the loop must not conditionally change any pure scalar variable
that is referenced after the loop terminates.
4 Calculations within the loop must not change a scalar variable across iterations.
This is called a loop-carried dependency.
f77: Apparent Dependencies
The f77 compiler may automatically eliminate a reference that appears to create a
dependency transforming the compiled code. One of the many such transformations
makes use of private versions of some of the arrays. Typically, the compiler does this
if it can determine that such arrays are used in the original loops only as temporary
storage.
Example: Using -autopar, with dependencies eliminated by private arrays:
parameter (n=1000)
real a(n), b(n), c(n,n)
SPARC: Parallelization
10-9
do i = 1, 1000
do k = 1, n
a(k) = b(k) + 2.0
end do
do j = 1, n
c(i,j) = a(j) + 2.3
end do
end do
end
<--Parallelized
In the preceding example, the outer loop is parallelized and run on independent
processors. Although the inner loop references to array a(*) appear to result in a data
dependency, the compiler generates temporary private copies of the array to make
the outer loop iterations independent.
Inhibitors to Automatic Parallelization
Under automatic parallelization, the compilers do not parallelize a loop if:
4 The DO loop is nested inside another DO loop that is parallelized.
4 Flow control allows jumping out of the DO loop.
4 A user-level subprogram is invoked inside the loop.
4 An I/O statement is in the loop.
4 Calculations within the loop change an aliased scalar variable.
Nested Loops
On multiprocessor systems, it is most effective to parallelize the outermost loop in a
loop nest, rather than the innermost. Because parallel processing typically involves
relatively large loop overhead, parallelizing the outermost loop minimizes the
overhead and maximizes the work done for each processor. Under automatic
parallelization, the compilers start their loop analysis from the outermost loop in a
nest and work inward until a parallelizable loop is found. Once a loop within the
nest is parallelized, loops contained within the parallel loop are passed over.
Automatic Parallelization With Reduction
Operations
A computation that transforms an array into a scalar is called a reduction operation.
Typical reduction operations are the sum or product of the elements of a vector.
Reduction operations violate the criterion that calculations within a loop not change
a scalar variable in a cumulative way across iterations.
10-10
Fortran Programming Guide ♦ Revision A, February 1999
Example: Reduction summation of the elements of a vector:
s = 0.0
do i = 1, 1000
s = s + v(i)
end do
t(k) = s
However, for some operations, if the reduction is the only factor that prevents
parallelization, it is still possible to parallelize the loop. Common reduction
operations occur so frequently that the compilers are capable of recognizing and
parallelizing them as special cases.
Recognition of reduction operations is not included in the automatic parallelization
analysis unless the -reduction compiler option is specified along with -autopar
or -parallel.
If a parallelizable loop contains one of the reduction operations listed in Table 10–3,
the compiler will parallelize it if -reduction is specified.
Recognized Reduction Operations
The following table lists the reduction operations that are recognized by f77 and f90.
TABLE 10–3
Recognized Reduction Operations
Mathematical Operations
Fortran Statement Templates
Sum of the elements
s = s + v(i)
Product of the elements
s = s * v(i)
Dot product of two vectors
s = s + v(i) * u(i)
Minimum of the elements
s = amin( s, v(i))
Maximum of the elements
s = amax( s, v(i))
OR of the elements
do i = 1, n
b = b .or. v(i)
end do
SPARC: Parallelization 10-11
TABLE 10–3
Recognized Reduction Operations
(continued)
Mathematical Operations
Fortran Statement Templates
AND of nonpositive elements
b = .true.
do i = 1, n
if (v(i) .le. 0) b=b .and. v(i)
end do
Count nonzero elements
k=0
do i = 1, n
if ( v(i) .ne. 0 )k = k + 1
end do
All forms of the MIN and MAX function are recognized.
Numerical Accuracy and Reduction Operations
Floating-point sum or product reduction operations may be inaccurate due to the
following conditions:
4 The order in which the calculations were performed in parallel was not the same
as when performed serially on a single processor.
4 The order of calculation affected the sum or product of floating-point numbers.
Hardware floating-point addition and multiplication are not associative. Roundoff,
overflow, or underflow errors may result depending on how the operands
associate. For example, (X*Y)*Z and X*(Y*Z) may not have the same numerical
significance.
In some situations, the error may not be acceptable.
Example: Overflow and underflow, with and without reduction:
demo% cat t3.f
real A(10002), result, MAXFLOAT
MAXFLOAT = r_max_normal()
do 10 i = 1 , 10000, 2
A(i) = MAXFLOAT
A(i+1) = -MAXFLOAT
10
continue
A(5001)=-MAXFLOAT
A(5002)=MAXFLOAT
10-12
Fortran Programming Guide ♦ Revision A, February 1999
do 20 i = 1 ,10002
!Add up the array
RESULT = RESULT + A(i)
20
continue
write(6,*) RESULT
end
demo% setenv PARALLEL 2
{Number of processors is 2}
demo% f77 -silent -autopar t3.f
demo% a.out
0.
{Without reduction, 0. is correct}
demo% f77 -silent -autopar -reduction t3.f
demo% a.out
Inf
{With reduction, Inf. is not correct}
demo%
Example: Roundoff, get the sum of 100,000 random numbers between –1 and +1:
demo% cat t4.f
parameter ( n = 100000 )
double precision d_lcrans, lb / -1.0 /, s, ub / +1.0 /, v(n)
s = d_lcrans ( v, n, lb, ub ) ! Get n random nos. between -1 and +1
s = 0.0
do i = 1, n
s = s + v(i)
end do
write(*, ’(" s = ", e21.15)’) s
end
demo% f77 -autopar -reduction t4.f
Results vary with the number of processors. The following table shows the sum of
100,000 random numbers between –1 and +1.
Number of Processors
Output
1
s = 0.568582080884714E+02
2
s = 0.568582080884722E+02
3
s = 0.568582080884721E+02
4
s = 0.568582080884724E+02
In this situation, roundoff error on the order of 10-14 is acceptable for data that is
random to begin with. For more information, see the Sun Numerical Computation
Guide.
SPARC: Parallelization 10-13
Explicit Parallelization
This section describes the source code directives recognized by f77 5.0 and f90 2.0
to explicitly indicate which loops to parallelize and what strategy to use.
Explicit parallelization of a program requires prior analysis and deep understanding
of the application code as well as the concepts of shared-memory parallelization.
DO loops are marked for parallelization by directives placed immediately before
them. The compiler options -parallel and -explicitpar must be used for DO
loops to be recognized and parallel code generated. Take care when choosing which
loops to mark for parallelization. The compiler generates threaded, parallel code for
all loops marked with DOALL directives, even if there are data dependencies that
will cause the loop to compute incorrect results when run in parallel.
If you do your own multithreaded coding using the libthread primitives, do not use
any of the compilers’ parallelization options—the compilers cannot parallelize code
that has already been parallelized with user calls to the threads library.
Parallelizable Loops
A loop is appropriate for explicit parallelization if:
4 It is a DO loop, but not a DO WHILE or Fortran 90 array syntax.
4 The values of array variables for each iteration of the loop do not depend on the
values of array variables for any other iteration of the loop.
4 If the loop changes a scalar, that scalar is not referenced after the loop terminates.
Such scalar variables are not guaranteed to have a defined value after the loop
terminates, since the compiler does not automatically ensure a proper storeback
for them.
4 For each iteration, any subprogram that is invoked inside the loop does not
reference or change values of array variables for any other iteration.
4 The DO loop index must be an integer.
Scoping Rules: Private and Shared
A private variable or array is private to a single iteration of a loop. The value assigned
to a private variable or array in one iteration is not propagated to any other iteration
of the loop.
A shared variable or array is shared with all other iterations. The value assigned to a
shared variable or array in an iteration is seen by other iterations of the loop.
10-14
Fortran Programming Guide ♦ Revision A, February 1999
If an explicitly parallelized loop contains shared references, then you must ensure
that sharing does not cause correctness problems. The compiler does no
synchronization on updates or accesses to shared variables.
If you specify a variable as private in one loop, and its only initialization is within
some other loop, the value of that variable may be left undefined in the loop.
Default Scoping Rules for Sun-Style Directives
For Sun-style (C$PAR) explicit directives, the compiler uses default rules to
determine whether a scalar or array is shared or private. You can override the
default rules to specify the attributes of scalars or arrays referenced inside a loop.
(With Cray-style !MIC$ directives, all variables that appear in the loop must be
explicitly declared either shared or private on the DOALL directive.)
The compiler applies these default rules:
4 All scalars are treated as private. A processor local copy of the scalar is made in
each processor, and that local copy is used within that process.
4 All array references are treated as shared references. Any write of an array element
by one processor is visible to all processors. No synchronization is performed on
accesses to shared variables.
If inter-iteration dependencies exist in a loop, then the execution may result in
erroneous results. You must ensure that these cases do not arise. The compiler may
sometimes be able to detect such a situation at compile time and issue a warning,
but it does not disable parallelization of such loops.
Example: Potential problem through equivalence:
equivalence (a(1),y)
C$PAR DOALL
do i = 1,n
y = i
a(i) = y
end do
In the preceding example, since the scalar variable y has been equivalenced to a(1),
it is no longer a private variable, even though the compiler treats it as such by the
default scoping rule. Thus, the presence of the DOALL directive might lead to
erroneous results when the parallelized i loop is executed.
You can fix the example by using C$PAR DOALL PRIVATE(y).
SPARC: Parallelization 10-15
Sun-Style Parallelization Directives
Parallelization directives are comment lines that tell the compiler to parallelize (or
not to parallelize) the DO loop that follows the directive. Directives are also called
pragmas.
A parallelization directive consists of one or more directive lines.
Sun-style directives are recognized by f77 and f90 by default (or with the -mp=sun
option). A Sun-style directive line is defined as follows:
C$PAR Directive
[ Qualifiers ]
C$PAR& [More_Qualifiers]
<- Initial directive line
<- Optional continuation lines
4 The letters of a directive line are case-insensitive.
4 The first five characters are C$PAR, *$PAR, or !$PAR.
4 An initial directive line has a blank in column 6.
4 A continuation directive line has a nonblank in column 6.
4 Directives are listed in columns 7 and beyond.
4 Qualifiers, if any, follow directives—on the same line or continuation lines.
4 Multiple qualifiers on one line are separated by commas.
4 Spaces before, after, or within a directive or qualifier are ignored.
4 Columns beyond 72 are ignored unless the -e option is specified.
The parallel directives and their actions are as follows:
Directive
Action
TASKCOMMON
Declares COMMON block private
DOALL
Parallelizes the next loop
DOSERIAL
Does not parallelize the next loop
DOSERIAL*
Does not parallelize the next nest of loops
Examples of f77 parallel directives:
C$PAR TASKCOMMON ALPHA
COMMON /ALPHA/BZ,BY(100)
Declare block private
C$PAR DOALL
No qualifiers
C$PAR DOSERIAL
10-16
Fortran Programming Guide ♦ Revision A, February 1999
C$PAR DOALL SHARED(I,K,X,V), PRIVATE(A)
This one-line directive is equivalent to the three-line directive that follows.
C$PAR DOALL
C$PAR& SHARED(I,K,X,V)
C$PAR& PRIVATE(A)
TASKCOMMON Directive
The TASKCOMMON directive declares variables in a global COMMON block as
private. Every variable declared in a task common block becomes a private variable.
Only named COMMON blocks can be declared TASK COMMON.
The syntax of the directive is:
C$PAR TASKCOMMON common_block_name
The directive must appear immediately after the defining COMMON declaration.
This directive is effective only when compiled with -explicitpar or -parallel.
Otherwise, the directive is ignored and the block is treated as a regular common
block.
Variables declared in task common blocks are treated as private variables in all the
DOALL loops they appear in explicitly, and in the routines called from a loop where
the specified common block is in its scope.
It is an error to declare a common block as task common in some but not all
compilation units where the block is defined. A check at runtime for task common
consistency can be enabled by compiling the program with the -xcommonchk=yes
flag. (Enable the runtime check only during program development, as it can degrade
performance.)
DOALL Directive
The compilers will parallelize the DO loop following a DOALL directive (if compiled
with the -parallel or -explicitpar options).
Note - Analysis and transformation of reduction operations within loops is not done
if they are explicitly parallelized.
Example: Explicit parallelization of a loop:
demo% cat t4.f
...
C$PAR DOALL
do i = 1, n
a(i) = b(i) * c(i)
end do
SPARC: Parallelization 10-17
do k = 1, m
x(k) = x(k) * z(k,k)
end do
...
demo% f77 -explicitpar t4.f
Subprogram Call in a Loop
A subprogram call in a loop (or in any subprograms called from within the called
routine) may introduce data dependencies that could go unnoticed without a deep
analysis of the data and control flow through the chain of calls. While it is best to
parallelize outermost loops that do a significant amount of the work, these tend to be
the very loops that involve subprogram calls.
Because such an interprocedural analysis is difficult and could greatly increase
compilation time, automatic parallelization modes do not attempt it. With explicit
parallelization, the compiler generates parallelized code for a loop marked with a
DOALL directive that contains calls to subprograms. It is still the programmer’s
responsibility toeinsure that no data dependencies exist within the loop and all that
the loop encloses, including called subprograms.
Multiple invocations of a routine from different processors can cause problems
resulting from references to local static variables that interfere with each other.
Making all the local variables in a routine automatic rather than static prevents this.
Each invocation of a subprogram then has its own unique store of local variables
maintained on the stack, and no two invocations will interfere with each other.
Local subprogram variables can be made automatic variables that reside on the stack
either by listing them on an AUTOMATIC statement or by compiling the subprogram
with the -stackvar option. However, local variables initialized in DATA statements
must be rewritten to be initialized in actual assignments.
Note - Allocating local variables to the stack can cause stack overflow. See “Stacks,
Stack Sizes, and Parallelization” on page 10-7 about increasing the size of the stack.
Data dependencies can still be introduced through the data passed down the call tree
as arguments or through COMMON blocks. This data flow should be analyzed
carefully before parallelizing a loop with subprogram calls.
DOALL Qualifiers
All qualifiers on the DOALL directive are optional. The following table summarizes
them:
10-18
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 10–4
DOALL Qualifiers
Qualifier
Assertion
Syntax
PRIVATE
Do not share variables u1, …
between iterations
DOALL PRIVATE(u1,u2,)
SHARED
Share variables v1, v2, … between
iterations
DOALL SHARED(v1,v2,)
MAXCPUS
Use no more than n CPUs
DOALL MAXCPUS(n)
READONLY
The listed variables are not
modified in the DOALL loop
DOALL READONLY(v1,v2,)
SAVELAST
Save the last DO iteration values of
all private variables
DOALL SAVELAST
STOREBACK
Save the last DO iteration values of
variables v1, …
DOALL STOREBACK(v1,v2,)
REDUCTION
Treat the variables v1, v2, … as
reduction variables.
DOALL REDUCTION(v1,v2,)
SCHEDTYPE
Set the scheduling type to t.
DOALL SCHEDTYPE(t)
PRIVATE(varlist)
The PRIVATE(varlist)qualifier specifies that all scalars and arrays in the list varlist
are private for the DOALL loop. Both arrays and scalars can be specified as private. In
the case of an array, each thread of the DOALL loop gets a copy of the entire array.
All other scalars and arrays referenced in the DOALL loop, but not contained in the
private list, conform to their appropriate default scoping rules.
Example: Specify a private array:
C$PAR DOALL PRIVATE(a)
do i = 1, n
a(1) = b(i)
do j = 2, n
a(j) = a(j-1) + b(j) * c(j)
end do
x(i) = f(a)
end do
In the preceding example, the array a is specified as private to the i loop.
SPARC: Parallelization 10-19
SHARED(varlist)
The SHARED(varlist) qualifier specifies that all scalars and arrays in the list varlist are
shared for the DOALL loop. Both arrays and scalars can be specified as shared. Shared
scalars and arrays are common to all the iterations of a DOALL loop. All other scalars
and arrays referenced in the DOALL loop, but not contained in the shared list,
conform to their appropriate default scoping rules.
Example: Specify a shared variable:
equivalence (a(1),y)
C$PAR DOALL SHARED(y)
do i = 1,n
a(i) = y
end do
In the preceding example, the variable y has been specified as a variable whose
value should be shared among the iterations of the i loop.
READONLY(varlist)
The READONLY(varlist) qualifier specifies that all scalars and arrays in the list varlist
are read-only for the DOALL loop. Read-only scalars and arrays are a special class of
shared scalars and arrays that are not modified in any iteration of the DOALL loop.
Specifying scalars and arrays as READONLY indicates to the compiler that it does not
need to use a separate copy of that variable or array for each thread of the DOALL
loop.
Example: Specify a read-only variable:
x = 3
C$PAR DOALL SHARED(x),READONLY(x)
do i = 1, n
b(i) = x + 1
end do
In the preceding example, x is a shared variable, but the compiler can rely on the
fact that it will not change over each iteration of the i loop because of its READONLY
specification.
STOREBACK(varlist)
A STOREBACK variable or array is one whose value is computed in a DOALL loop.
The computed value can be used after the termination of the loop. In other words,
the last loop iteration values of storeback scalars and arrays may be visible outside of
the DOALL loop.
10-20
Fortran Programming Guide ♦ Revision A, February 1999
Example: Specify the loop index variable as storeback:
C$PAR DOALL PRIVATE(x), STOREBACK(x,i)
do i = 1, n
x = ...
end do
... = i
... = x
In the preceding example, both the variables x and i are STOREBACK variables,
even though both variables are private to the i loop.
There are some potential problems for STOREBACK, however.
The STOREBACK operation occurs at the last iteration of the explicitly parallelized
loop, even if this is the same iteration that last updates the value of the STOREBACK
variable or array.
Example: STOREBACK variable potentially different from the serial version:
C$PAR DOALL PRIVATE(x), STOREBACK(x)
do i = 1, n
if (...) then
x = ...
end if
end do
print *,x
In the preceding example, the value of the STOREBACK variable x that is printed out
might not be the same as that printed out by a serial version of the i loop. In the
explicitly parallelized case, the processor that processes the last iteration of the i
loop (when i = n) and performs the STOREBACK operation for x, might not be the
same processor that currently contains the last updated value of x. The compiler
issues a warning message about these potential problems.
In an explicitly parallelized loop, arrays are not treated by default as STOREBACK, so
include them in the list varlist if such a storeback operation is desired—for example,
if the arrays have been declared as private.
SAVELAST
The SAVELAST qualifier specifies that all private scalars and arrays are STOREBACK
for the DOALL loop. A STOREBACK variable or array is one whose value is computed
in a DOALL loop; this computed value can be used after the termination of the loop.
In other words, the last loop iteration values of STOREBACK scalars and arrays may
be visible outside of the DOALL loop.
Example: Specify SAVELAST:
SPARC: Parallelization 10-21
C$PAR DOALL PRIVATE(x,y), SAVELAST
do i = 1, n
x = ...
y = ...
end do
... = i
... = x
... = y
In the preceding example, variables x, y, and i are STOREBACK variables.
REDUCTION(varlist)
The REDUCTION(varlist) qualifier specifies that all variables in the list varlist are
reduction variables for the DOALL loop. A reduction variable (or array) is one whose
partial values can be individually computed on various processors, and whose final
value can be computed from all its partial values.
The presence of a list of reduction variables can aid the compiler in identifying if a
DOALL loop is a reduction loop and in generating parallel reduction code for it.
Example: Specify a reduction variable:
C$PAR DOALL REDUCTION(x)
do i = 1, n
x = x + a(i)
end do
In the preceding example, the variable x is a (sum) reduction variable; the i loop is a
(sum) reduction loop.
SCHEDTYPE(t)
The SCHEDTYPE(t) qualifier specifies that the specific scheduling type t be used to
schedule the DOALL loop.
10-22
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 10–5
DOALL SCHEDTYPE Qualifiers
Scheduling
Type
Action
STATIC
Use static scheduling for this DO loop.
Distribute all iterations uniformly to all available processors.
Example: With 1000 iterations and 4 CPUs each CPU gets a single
iteration in turn until all the iterations have been distributed.
SELF[(chunksize)]Use self-scheduling for this DO loop.
Distribute chunksize iterations to each available processor:
• Repeat with the remaining iterations until all the iterations have been
processed. • If chunksize is not provided, f77 selects a value.
Example: With 1000 iterations and chunksize of 4, distribute 4 iterations to
each CPU.
FACTORING[(
m )]
Use factoring scheduling for this DO loop.
With n iterations initially and k CPUs, distribute n/(2k) iterations
uniformly to each processor until all iterations have been processed.
• At least m iterations must be assigned to each processor.
• There can be one final smaller residual chunk.
• If m is not provided, f77 selects a value.
Example: With 1000 iterations and FACTORING(4), and 4 CPUs, distribute
125 iterations to each CPU, then 62 iterations, then 31 iterations, and so
on.
GSS[( m )]
Use guided self-scheduling for this DO loop.
With n iterations initially, and k CPUs, then:
• Assign n/k iterations to the first processor.
• Assign the remaining iterations divided by k to the second processor,
and so on until all iterations have been processed.
Note:
• At least m iterations must be assigned to each CPU.
• There can be one final smaller residual chunk.
• If m is not provided, f77 selects a value.
Example: With 1000 iterations and GSS(10), and 4 CPUs, distribute 250
iterations to the first CPU, then 187 to the second CPU, then 140 to the
third CPU, and so on.
SPARC: Parallelization 10-23
Multiple Qualifiers
Qualifiers can appear multiple times with cumulative effect. In the case of conflicting
qualifiers, the compiler issues a warning message, and the qualifier appearing last
prevails.
Example: A three-line Sun-style directive:
C$PAR DOALL MAXCPUS(4), READONLY(S), PRIVATE(A,B,X), MAXCPUS(2)
C$PAR DOALL SHARED(B,X,Y), PRIVATE(Y,Z)
C$PAR DOALL READONLY(T)
Example: A one-line equivalent of the preceding three lines (note duplicate MAXCPUS
and conflicting SHARED/PRIVATE):
C$PAR DOALL MAXCPUS(2), PRIVATE(A,Y,Z), SHARED(B,X), READONLY(S,T)
DOSERIAL Directive
The DOSERIAL directive disables parallelization of the specified loop. This directive
applies to the one loop immediately following it (if you compile it with
-explicitpar or -parallel).
Example: Exclude one loop from parallelization:
do i = 1, n
C$PAR DOSERIAL
do j = 1, n
do k = 1, n
...
end do
end do
end do
In the preceding example, the j loop is not parallelized, but the i or k loop can be.
DOSERIAL* Directive
The DOSERIAL* directive disables parallelization the specified nest of loops. This
directive applies to the whole nest of loops immediately following it (if you compile
with -explicitpar or -parallel).
Example: Exclude a whole nest of loops from parallelization:
do i = 1, n
C$PAR DOSERIAL*
do j = 1, n
10-24
Fortran Programming Guide ♦ Revision A, February 1999
do k = 1, n
...
end do
end do
end do
In the preceding loops, the j and k loops are not parallelized; the i loop could be.
Interaction Between DOSERIAL* and DOALL
If both DOSERIAL and DOALL are specified, the last one prevails.
Example: Specifying both DOSERIAL and DOALL:
C$PAR DOSERIAL*
do i = 1, 1000
C$PAR DOALL
do j = 1, 1000
...
end do
end do
In the preceding example, the i loop is not parallelized, but the j loop is.
Also, the scope of the DOSERIAL* directive does not extend beyond the textual loop
nest immediately following it. The directive is limited to the same function or
subroutine that it is in.
Example: DOSERIAL* does not extend to a loop of a called subroutine:
program caller
common /block/ a(10,10)
C$PAR DOSERIAL*
do i = 1, 10
call callee(i)
end do
end
subroutine callee(k)
common /block/a(10,10)
do j = 1, 10
a(j,k) = j + k
end do
return
end
In the preceding example, DOSERIAL* applies only to the i loop and not to the j
loop, regardless of whether the call to the subroutine callee is inlined.
SPARC: Parallelization 10-25
Inhibitors to Explicit Parallelization
In general, the compiler parallelizes a loop if you explicitly direct it to. There are
exceptions—some loops the compiler just cannot parallelize.
The following are the primary detectable inhibitors that might prevent explicitly
parallelizing a DO loop:
4 The DO loop is nested inside another DO loop that is parallelized.
This exception holds for indirect nesting, too. If you explicitly parallelize a loop
that includes a call to a subroutine, then even if you parallelize loops in that
subroutine, those loops are not run in parallel at runtime.
4 A flow control statement allows jumping out of the DO loop.
4 The index variable of the loop is subject to side effects, such as being equivalenced.
If you compile with -vpara, you may get a warning message if f77/f90 detects a
problem with explicitly parallelizing a loop. f77/f90 may still parallelize the loop.
The following list of typical parallelization problems shows those that are ignored by
the compiler
TABLE 10–6
Explicit Parallelization Problems
Problem
Parallelized
Message
Loop is nested inside another loop that is parallelized.
No
No
Loop is in a subroutine, and a call to the subroutine is
in a parallelized loop.
No
No
Jumping out of loop is allowed by a flow control
statement.
No
Yes
Index variable of loop is subject to side effects.
Yes
No
Some variable in the loop keeps a loop-carried
dependency.
Yes
Yes
I/O statement in the loop—usually unwise, because the
order of the output is not predictable.
Yes
No
and those that generate messages with -vpara.
Example: Nested loops:
...
C$PAR DOALL
do 900 i = 1, 1000
10-26
!
Parallelized (outer loop)
Fortran Programming Guide ♦ Revision A, February 1999
do 200 j = 1, 1000
! Not parallelized, no warning
...
200
continue
900
continue
...
demo% f77 -explicitpar -vpara t6.f
Example: A parallelized loop in a subroutine:
C$PAR DOALL
subroutine calc ( b, y )
do 100 i = 1, 200
...
...
C$PAR DOALL
call calc (a, x)
...
100 continue
...
demo% f77 -explicitpar -vpara t.f
At runtime, the loop could run in parallel.
do 1 m = 1, 1000
...
1 continue
return
end
At runtime, both loops do not run in
parallel.
In the preceding example, the loop within the subroutine is not parallelized because
the subroutine itself is run in parallel.
Example: Jumping out of a loop:
C$PAR DOALL
do i = 1, 1000
! ¨
Not parallelized, with warning
...
if (a(i) .gt. min_threshold ) go to 20
...
end do
20
continue
...
demo% f77 -explicitpar -vpara t9.f
Example: An index variable subject to side effects:
SPARC: Parallelization 10-27
equivalence ( a(1), y )
!
¨ Source of possible side effects
...
C$PAR DOALL
do i = 1, 2000
!
¨ Parallelized: no warning, but not safe
y = i
a(i) = y
end do
...
demo% f77 -explicitpar -vpara t11.f
Example: A variable in a loop has a loop-carried dependency:
C$PAR DOALL
do 100 i = 1, 200
! Parallelized, with warning
y = y * i
! y has a loop-carried dependency
a(i) = y
100
continue
...
demo% f77 -explicitpar -vpara t12.f
I/O With Explicit Parallelization
You can do I/O in a loop that executes in parallel, provided that:
4 It does not matter that the output from different threads is interleaved (program
output is nondeterministic.)
4 You can ensure the safety of executing the loop in parallel.
Example: I/O statement in loop
C$PAR DOALL
do i = 1, 10
! Parallelized with no warning (not advisable)
k = i
call show ( k )
end do
end
subroutine show( j )
write(6,1) j
1
format("Line number ", i3, ".")
end
demo% f77 -silent -explicitpar -vpara t13.f
demo% setenv PARALLEL 2
demo% a.out
(The output displays the numbers 1 through 10, but in a different order each time.)
Example: Recursive I/O:
10-28
Fortran Programming Guide ♦ Revision A, February 1999
do i = 1, 10
Parallelized with no warning —unsafe
k = i
print *, list( k )
<-- list is a function that does I/O
end do
end
function list( j )
write(6,"(’Line number ’, i3, ’.’)") j
list = j
end
demo% f77 -silent -mt t14.f
demo% setenv PARALLEL 2
demo% a.out
<--
In the preceding example, the program may deadlock in libF77_mt and hang. Press
Control-C to regain keyboard control.
There are situations where the programmer might not be aware that I/O could take
place within a parallelized loop. Consider a user-supplied exception handler that
prints output when it catches an arithmetic exception (like divide by zero). If a
parallelized loop provokes an exception, the implicit I/O from the handler may
cause I/O deadlocks and a system hang.
In general:
4 The library libF77_mt is MT safe, but mostly not MT hot.
4 You cannot do recursive (nested) I/O if you compile with -mt.
As an informal definition, an interface is MT safe if:
4 It can be simultaneously invoked by more than one thread of control.
4 The caller is not required to do any explicit synchronization before calling the
function.
4 The interface is free of data races.
A data race occurs when the content of memory is being updated by more than one
thread, and that bit of memory is not protected by a lock. The value of that bit of
memory is nondeterministic—the two threads race to see who gets to update the
thread (but in this case, the one who gets there last, wins).
An interface is colloquially called MT hot if the implementation has been tuned for
performance advantage, using the techniques of multithreading. For some formal
definitions of multithreading technology, read the Solaris Multithreaded Programming
Guide.
Cray-Style Parallelization Directives
Parallel directives have two forms: Sun style and Cray style. The f77 and f90
default is Sun style (-mp=sun). To use Cray-style directives, you must compile with
-mp=cray.
SPARC: Parallelization 10-29
Mixing program units compiled with both Sun and Cray directives can produce
different results.
A major difference between Sun and Cray directives is that Cray style requires explicit
scoping of every scalar and array in the loop as either SHARED or PRIVATE.
The following table shows Cray-style directive syntax.
!MIC$ DOALL
!MIC$& SHARED( v1
, v2,
)
!MIC$& PRIVATE( u1
, u2,
)
...optional qualifiers
Cray Directive Syntax
A parallel directive consists of one or more directive lines. A directive line is defined
as follows:
4 The directive line is case insensitive.
4 The first five characters are CMIC$, *MIC$, or !MIC$.
4 An initial directive line has a blank in column 6.
4 A continuation directive line has a nonblank in column 6.
4 Directives are listed in columns 7 and beyond.
4 Qualifiers, if any, follow directives—on the same line or continuation lines.
4 Multiple qualifiers on a line are separated by commas.
4 All variables and arrays are in qualifiers SHARED or PRIVATE.
4 Spaces before, after, or within a directive or qualifier are ignored.
4 Columns beyond 72 are ignored.
With f90 -free free-format, leading blanks can appear before !MIC$.
Qualifiers (Cray Style)
For Cray-style directives, the PRIVATE qualifier is required. Each variable within the
DO loop must be qualified as private or shared, and the DO loop index must always
be private. The following table summarizes available Cray-style qualifiers.
10-30
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 10–7
DOALL Qualifiers (Cray Style)
Qualifier
Assertion
SHARED( v1, v2,
…)
Share the variables v1, v2, … between parallel processes. That is, they are
accessible to all the tasks.
PRIVATE( x1,
x2, … )
Do not share the variables x1, x2, … between parallel processes. That is,
each task has its own private copy of these variables.
SAVELAST
Save the values of private variables from the last DO iteration.
MAXCPUS( n )
Use no more than n CPUs.
For Cray-style directives, the DOALL directive allows a single scheduling qualifier, for
example, !MIC$& CHUNKSIZE(100). Table 10–8 shows the Cray-style DOALL
directive
TABLE 10–8
DOALL Cray Scheduling
Qualifier
Assertion
GUIDED
Distribute the iterations by use of guided self-scheduling.
This distribution minimizes synchronization overhead, with acceptable
dynamic load balancing.
SINGLE
Distribute one iteration to each available processor.
CHUNKSIZE( n
)
Distribute n iterations to each available processor.
NUMCHUNKS( m
)
n may be an expression. For best performance, n must be an integer
constant. Example: With 100 iterations and CHUNKSIZE(4), distribute 4
iterations to each CPU.
If there are n iterations, distribute n/m iterations to each available
processor. There can be one smaller residual chunk.
m is an expression. For best performance, m must be an integer constant.
Example: With 100 iterations and NUMCHUNKS(4), distribute 25 iterations
to each CPU.
scheduling qualifiers:
The f77 default scheduling type is the Sun-style STATIC. The f90 default is
GUIDED.
SPARC: Parallelization 10-31
Inhibitors to f90 Explicit Parallelization
With the explicit parallelization situations listed in “Inhibitors to Explicit
Parallelization” on page 10-26, the additional parallelization inhibitors for f90
include:
4 The DO increment parameter, if specified, is a variable.
4 There is an I/O statement in the loop.
4 Parallelized loops in subprograms called from parallelized loops are, in fact, not
run in parallel.
Debugging Parallelized Programs
Compiling with the -g option cancels any of the parallelization options -autopar,
-explicitpar, and -parallel, as well as -reduction and -depend. Some
alternative ways to debug parallelized code are suggested in the following section.
Debugging Without dbx
Debugging parallelized programs requires some cleverness. The following schemes
suggest ways to approach the problem:
4 Turn off parallelization.
You can do one of the following:
4
Turn off the parallelization options—Verify that the program works correctly
by compiling with -O3 or -O4, but without any parallelization.
4
Set the CPUs to one—run the program with the environment variable
PARALLEL=1.
If the problem disappears, then you know it was due to parallelization.
Check also for out of bounds array references by compiling with -C.
Problems using -autopar may indicate that the compiler is parallelizing
something it should not.
4 Turn off -reduction.
If you are using the -reduction option, summation reduction may be occurring
and yielding slightly different answers. Try running without this option.
4 Reduce the number of compile options.
10-32
Fortran Programming Guide ♦ Revision A, February 1999
Compile with just -parallel -O3 and check the results.
4 Use fsplit or f90split.
If you have a lot of subroutines in your program, use fsplit(1) to break them into
separate files. (Use f90split(1) on Fortran 90 source codes.) Then compile some files
with and without -parallel, and use f77 or f90 to link the .o files. You must
specify -parallel on this link step as well. (See Fortran User’s Guide section on
consistent compiling and linking.)
Execute the binary and verify results.
Repeat this process until the problem is narrowed down to one subroutine.
You can proceed using a dummy subroutine or explicit parallelization to track
down the loop that causes the problem.
4 Use -loopinfo.
Check which loops are being parallelized and which loops are not.
4 Use a dummy subroutine.
Create a dummy subroutine or function that does nothing. Put calls to this
subroutine in a few of the loops that are being parallelized. Recompile and
execute. Use -loopinfo to see which loops are being parallelized.
Continue this process until you start getting the correct results.
Then remove the calls from the other loops, compile, and execute to verify that
you are getting the correct results.
4 Use explicit parallelization.
Add the C$PAR DOALL directive to a couple of the loops that are being
parallelized. Compile with -explicitpar, then execute and verify the results.
Use -loopinfo to see which loops are being parallelized. This method permits
the addition of I/O statements to the parallelized loop.
Repeat this process until you find the loop that causes the wrong results.
Note - If you need -explicitpar only (without -autopar), do not compile with
-explicitpar and -depend. This method is the same as compiling with
-parallel, which, of course, includes -autopar.
4 Run loops backward serially.
Replace DO I=1,N with DO I=N,1,-1. Different results point to data
dependencies.
4 Avoid using the loop index. It is safer to do so in the loop body, especially if the
index is used as an argument in a call.
Replace:
DO I=1,N
...
CALL SNUBBER(I)
SPARC: Parallelization 10-33
...
ENDDO
With:
DO I1=1,N
I=I1
...
CALL SNUBBER(I)
...
ENDDO
Using dbx
To use dbx on a parallel loop, temporarily rewrite the program as follows:
4 Isolate the body of the loop in a file and subroutine of its own.
4 In the original routine, replace loop body with a call to the new subroutine.
4 Compile the new subroutine with -g and no parallelization options.
4 Compile the changed original routine with parallelization and no -g.
Example: Manually transform a loop to allow using dbx in parallel:
Original code:
demo% cat loop.f
C$PAR DOALL
DO i = 1,10
WRITE(0,*) "Iteration ", i
END DO
END
Split into two parts: caller loop and loop body as a subroutine
demo% cat loop1.f
C$PAR DOALL
DO i = 1,10
k = i
CALL loop_body ( k )
END DO
END
demo% cat loop2.f
SUBROUTINE loop_body ( k )
WRITE(0,*) "Iteration ", k
RETURN
END
Compile caller loop with parallelization but no debugging
demo% f77 -O3 -c -explicitpar loop1.f
Compile the subprogram with debugging but not parallelized
demo% f77 -c -g loop2.f
Link together both parts into a.out
demo% f77 loop1.o loop2.o -explicitpar
Run a.out under dbx and put breakpoint into loop body subroutine
demo% dbx a.out
¨ Various
dbx messages not shown
(dbx) stop in loop_body
(2) stop in loop_body
(dbx) run
Running: a.out
10-34
Fortran Programming Guide ♦ Revision A, February 1999
(process id 28163)
dbx stops at breakpoint
t@1 (l@1) stopped in loop_body at line 2 in file
"loop2.f"
2
write(0,*) "Iteration ", k
Now show value of k
(dbx) print k
k = 1
¨ Various values other than 1 are possible
(dbx)
SPARC: Parallelization 10-35
10-36
Fortran Programming Guide ♦ Revision A, February 1999
CHAPTER
11
C-Fortran Interface
This chapter treats issues regarding Fortran and C interoperability.
The discussion is inherently limited to the specifics of the Sun FORTRAN 77, Fortran
90, and C compilers.
Note - Material common to both Sun FORTRAN 77 and Fortran 90 is presented in
examples that use FORTRAN 77.
Compatibility Issues
Most C-Fortran interfaces must agree in all of these aspects:
4 Function/subroutine: definition and call
4 Data types: compatibility of types
4 Arguments: passing by reference or value
4 Arguments: order
4 Procedure name: uppercase and lowercase and trailing underscore (_)
4 Libraries: telling the linker to use Fortran libraries
Some C-Fortran interfaces must also agree on:
4 Arrays: indexing and order
4 File descriptors and stdio
4 File permissions
11-1
Function or Subroutine?
The word function has different meanings in C and Fortran. Depending on the
situation, the choice is important:
4 In C, all subprograms are functions; however, some may return a null (void) value.
4 In Fortran, a function passes a return value, but a subroutine does not.
When a Fortran routine calls a C function:
4 If the called C function returns a value, call it from Fortran as a function.
4 If the called C function does not return a value, call it as a subroutine.
When a C function calls a Fortran subprogram:
4 If the called Fortran subprogram is a function, call it from C as a function that
returns a compatible data type.
4 If the called Fortran subprogram is a subroutine, call it from C as a function that
returns a value of int (compatible to Fortran INTEGER*4) or void. A value is
returned if the Fortran subroutine uses alternate returns, in which case it is the
value of the expression on the RETURN statement. If no expression appears on the
RETURN statement, and alternate returns are declared on SUBROUTINE statement,
a zero is returned.
Data Type Compatibility
The tables below summarize the data sizes and default alignments for FORTRAN 77
and Fortran 90 data types. In both tables, note the following:
4 C data types int, long int, and long are equivalent (4 bytes). In a 64-bit
environment and compiling with -xarch=v9 or v9a, long and pointers are 8
bytes. This is referred to as "LP64".
4 REAL*16 and COMPLEX*32, (REAL(KIND=16) and COMPLEX(KIND=16)), are
available only on SPARC platforms. In a 64-bit environment and compiling with
-xarch=v9 or v9a, alignment is on 16-byte boundaries.
4 Alignments marked 4/8 for SPARC indicate that alignment is 8-bytes by default,
but on 4-byte boundaries in COMMON blocks. The maximum alignment in
COMMON is 4-bytes.
4 The elements and fields of arrays and structures must be compatible.
4 You cannot pass arrays, character strings, or structures by value.
4 You can pass arguments by value from f77 to C, but not from C to f77, since
%VAL() is not allowed in a Fortran dummy argument list.
11-2
Fortran Programming Guide ♦ Revision A, February 1999
FORTRAN 77 and C Data Types
Table 11–1 shows the sizes and allowable alignments for FORTRAN 77 data types. It
assumes no compilation options affecting alignment or promoting default data sizes
are applied. (See also the FORTRAN 77 Language Reference Manual).
TABLE 11–1
Data Sizes and Alignments—(in Bytes) Pass by Reference (f77 and cc)
FORTRAN 77
Data Type
C Data Type
Size
Default
Alignment
SPARC x86
BYTE X
char x
1
1
1
CHARACTER X
unsigned char x
1
1
1
n
1
1
CHARACTER*n
X
unsigned char x[n]
COMPLEX X
struct {float r,i;} x;
8
4
4
COMPLEX*8 X
struct {float r,i;} x;
8
4
4
16
4/8
4
DOUBLE
COMPLEX X
struct {double dr,di;}x;
16
4/8
4
struct {double dr,di;}x;
32
4/8/16
—
8
4/8
4
4
4
4
COMPLEX*16 X
COMPLEX*32 X
DOUBLE
PRECISION X
struct {long double dr,di;}
x;
double x
float x
4
4
4
float x
8
4/8
4
double x
16
4/8/16
—
REAL X
REAL*4 X
REAL*8 X
long double x
REAL*16 X
C-Fortran Interface
11-3
TABLE 11–1
cc)
Data Sizes and Alignments—(in Bytes) Pass by Reference (f77 and
(continued)
FORTRAN 77
Data Type
C Data Type
Size
Default
Alignment
SPARC x86
INTEGER X
int x
4
4
4
INTEGER*2 X
short x
2
2
2
4
4
4
INTEGER*4 X
int x
8
4
4
INTEGER*8 X
long long int x
LOGICAL X
int x
4
4
4
LOGICAL*1 X
char x
1
1
1
2
2
2
LOGICAL*2 X
short x
4
4
4
LOGICAL*4 X
int x
8
4
4
LOGICAL*8 X
long long int x
SPARC: Fortran 90 and C Data Types
The following table similarly compares the Fortran 90 data types with C.
TABLE 11–2
11-4
Data Sizes and Alignment—(in Bytes) Pass by Reference (f90 and cc)
Fortran 90 Data Type
C Data Type
Alignment
Size
CHARACTER x
unsigned char x ;
1
CHARACTER (LEN=n) x
unsigned char x[n] ;
1
n
COMPLEX x
struct {float r,i;} x;
48
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 11–2
Data Sizes and Alignment—(in Bytes) Pass by Reference (f90 and cc)
(continued)
Fortran 90 Data Type
C Data Type
Alignment
Size
COMPLEX (KIND=4) x
struct {float r,i;} x;
48
COMPLEX (KIND=8) x
struct {double dr,di;} x;
COMPLEX (KIND=16) x
struct {long double,
dr,di;} x;
DOUBLE COMPLEX x
struct {double dr, di;} x;
16
4/8
DOUBLE PRECISION x
double x ;
84
REAL x
float x ;
4
REAL (KIND=4) x
float x ;
4
REAL (KIND=8) x
double x ;
REAL (KIND=16) x
long double x ;
INTEGER x
int x ;
4
INTEGER (KIND=1) x
signed char x ;
41
INTEGER (KIND=2) x
short x ;
INTEGER (KIND=4) x
int x ;
INTEGER (KIND=8) x
long long int x;
LOGICAL x
int x ;
4
LOGICAL (KIND=1) x
signed char x ;
41
LOGICAL (KIND=2) x
short x ;
LOGICAL (KIND=4) x
int x ;
LOGICAL (KIND=8) x
long long int x;
16
4/8
32
4/8/16
8
4/8
16
4/8/16
42
4
48
42
4
48
C-Fortran Interface
11-5
Case Sensitivity
C and Fortran take opposite perspectives on case sensitivity:
4 C is case sensitive—case matters.
4 Fortran ignores case.
The f77 and f90 default is to ignore case by converting subprogram names to
lowercase. It converts all uppercase letters to lowercase letters, except within
character-string constants.
There are two usual solutions to the uppercase/lowercase problem:
4 In the C subprogram, make the name of the C function all lowercase.
4 Compile the f77 program with the -U option, which tells f77 to preserve existing
uppercase/lowercase distinctions on function/subprogram names.
Use one of these two solutions, but not both.
Most examples in this chapter use all lowercase letters for the name in the C function,
and do not use the f77 -U compiler option. (f90 does not have an equivalent option.)
Underscores in Routine Names
The Fortran compiler normally appends an underscore (_) to the names of
subprograms appearing both at entry point definition and in calls. This convention
differs from C procedures or external variables with the same user-assigned name. If
the name has exactly 32 characters, the underscore is not appended. All Fortran
library procedure names have double leading underscores to reduce clashes with
user-assigned subroutine names.
There are three usual solutions to the underscore problem:
4 In the C function, change the name of the function by appending an underscore to
that name.
4 Use the f77 C() pragma to tell the FORTRAN 77 compiler to omit those trailing
underscores.
4 Use the f77 -ext_names option to make external names without underscores.
Use only one of these solutions.
The examples in this chapter could use the FORTRAN 77 C() compiler pragma to
avoid underscores. The C() pragma directive takes the names of external functions
as arguments. It specifies that these functions are written in the C language, so the
Fortran compiler does not append an underscore as it ordinarily does with external
names. The C()directive for a particular function must appear before the first
reference to that function. It must also appear in each subprogram that contains such
a reference. The conventional usage is:
EXTERNAL ABC, XYZ
11-6
!$PRAGMA C( ABC, XYZ )
Fortran Programming Guide ♦ Revision A, February 1999
If you use this pragma, the C function does not need an underscore appended to the
function name.
Fortran 90 does not have equivalent methods for avoiding underscores. Trailing
underscores are required in the names of C routines called from Fortran 90 routines.
Argument-Passing by Reference or Value
In general, Fortran routines pass arguments by reference. In a call, if you enclose an
argument with the f77 and f90 nonstandard function %VAL(), the calling routine
passes it by value.
In general, C passes arguments by value. If you precede an argument by the
ampersand operator (&), C passes the argument by reference using a pointer. C
always passes arrays and character strings by reference.
Argument Order
Except for arguments that are character strings, Fortran and C pass arguments in the
same order. However, for every argument of character type, the Fortran routine
passes an additional argument giving the length of the string. These are long int
quantities in C, passed by value.
The order of arguments is:
4 Address for each argument (datum or function)
4 A long int for each character argument (the whole list of string lengths comes
after the whole list of other arguments)
Example:
This Fortran code fragment:
Is equivalent to this in C:
CHARACTER*7 S
INTEGER B(3)
…
CALL SAM( S, B(2) )
char s[7];
long b[3];
…
sam_( s, &b[1], 7L ) ;
C-Fortran Interface
11-7
Array Indexing and Order
Array indexing and order differ between Fortran and C.
Array Indexing
C arrays always start at zero, but by default Fortran arrays start at 1. There are two
usual ways of approaching indexing.
4 You can use the Fortran default, as in the preceding example. Then the Fortran
element B(2) is equivalent to the C element b[1].
4 You can specify that the Fortran array B starts at B(0) as follows:
INTEGER B(0:2)
This way, the Fortran element B(1) is equivalent to the C element b[1].
Array Order
Fortran arrays are stored in column-major order: A(3,2)
A(1,1) A(2,1) A(3,1) A(1,2) A(2,2) A(3,2) A(1,3) A(2,3) A(3,3)
C arrays in row-major order: A[3][2]
A[0][0] A[0][1] A[0][2] A[1][0] A[1][1] A[1][2] A[2][0] A[2][1] A[2][2]
For one-dimensional arrays, this is no problem. For two-dimensional and higher
arrays, be aware of how subscripts appear and are used in all references and
declarations—some adjustments might be necessary.
For example, it may be confusing to do part of a matrix manipulation in C and the
rest in Fortran. It might be preferable to pass an entire array to a routine in the other
language and perform all the matrix manipulation in that routine to avoid doing part
in C and part in Fortran.
File Descriptors and stdio
Fortran I/O channels are in terms of unit numbers. The I/O system does not deal
with unit numbers but with file descriptors. The Fortran runtime system translates
from one to the other, so most Fortran programs do not have to recognize file
descriptors.
11-8
Fortran Programming Guide ♦ Revision A, February 1999
Many C programs use a set of subroutines, called standard I/O (or stdio). Many
functions of Fortran I/O use standard I/O, which in turn uses operating system I/O
calls. Some of the characteristics of these I/O systems are listed in in the following
table.
TABLE 11–3
Comparing I/O Between Fortran and C
Fortran Units
Standard I/O File Pointers
File Descriptors
Files Open
Opened for
reading and
writing
Opened for reading; or Opened for
writing; or Opened for both; or
Opened for appending; See
OPEN(3S)
Opened for
reading; or
Opened for
writing; or
Opened for both
Attributes
Formatted or
unformatted
Always unformatted, but can be
read or written with
format-interpreting routines
Always
unformatted
Access
Direct or
sequential
Direct access if the physical file
representation is direct access, but
can always be read sequentially
Direct access if
the physical file
representation is
direct access,
but can always
be read
sequentially
Structure
Record
Byte stream
Byte stream
Form
Arbitrary
nonnegative
integers from
0-2147483647
Pointers to structures in the user’s
address space
Integers from
0-1023
File Permissions
C programs typically open input files for reading and output files for writing or for
reading and writing. A f77 program can OPEN a file READONLY or with
READWRITE=’READ’ or ’WRITE’ or ’READWRITE’. f90 supports the READWRITE
specifier, but not READONLY.
Fortran tries to open a file with the maximum permissions possible, first for both
reading and writing, then for each separately.
This event occurs transparently and is of concern only if you try to perform a READ,
WRITE, or ENDFILE operation but you do not have permission. Magnetic tape
C-Fortran Interface
11-9
operations are an exception to this general freedom, since you can have write
permissions on a file, but not have a write ring on the tape.
Libraries and Linking With the f77 or f90
Command
To link the proper Fortran and C libraries, use the f77 or f90 command to invoke
the linker.
Example 1: Use f77 to link:
demo% cc -c RetCmplxmain.c
demo% f77 RetCmplx.f RetCmplxmain.o
¨ This command line does the linking.
demo% a.out
4.0 4.5
8.0 9.0
demo%
Passing Data Arguments by Reference
The standard method for passing data between Fortran routines and C procedures is
by reference. To a C procedure, a Fortran subroutine or function call looks like a
procedure call with all arguments represented by pointers. The only peculiarity is the
way Fortran handles character strings and functions as arguments and as the return
value from a CHARACTER*n function.
Simple Data Types
For simple data types (not COMPLEX or CHARACTER strings), define or pass each
associated argument in the C routine as a pointer:.
11-10
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 11–4
Passing Simple Data Types
Fortran calls C
C calls Fortran
integer i
int i=100;
real r
float r;
external CSim
extern void fsim_(int *i, float *r);
i = 100
fsim_(&i, &r);
call CSim(i,r)
...
...
------------------------------
---------------------------
subroutine FSim(i,r)
integer i
void csim_(int *i, float *r)
real r
{
r=i
*r = *i;
return
}
end
COMPLEX Data
Pass a Fortran COMPLEX data item as a pointer to a C struct of two float or two
double data types:
C-Fortran Interface 11-11
TABLE 11–5
Passing COMPLEX Data Types
Fortran calls C
C calls Fortran
complex w
struct cpx {float r, i;};
double complex z
struct cpx d1;
external CCmplx
struct cpx *w = &d1;
call CCmplx(w,z)
struct dpx {double r, i;};
...
struct dpx d2;
------------------------------
struct dpx *z = &d2;
struct cpx {float r, i;};
fcmplx_( w, z );
struct dpx {double r,i;};
...
void ccmplx_(
---------------
struct cpx *w,
subroutine FCmplx( w, z )
struct dpx *z)
complex w
{
double complex z
w -> r = 32.;
w = (32., .007)
w -> i = .007;
z = (66.67, 94.1)
z -> r = 66.67;
return
z -> i = 94.1;
end
}
In 64-bit environments and compiling with -xarch=v9, COMPLEX values are
returned in registers.
Character Strings
Passing strings between C and Fortran routines is not recommended because there is
no standard interface. However, note the following:
4 All C strings are passed by reference.
11-12
Fortran Programming Guide ♦ Revision A, February 1999
4 Fortran calls pass an additional argument for every argument with character type
in the argument list. The extra argument gives the length of the string and is
equivalent to a C long int passed by value. (This is implementation dependent.)
The extra string-length arguments appear after the explicit arguments in the call.
A Fortran call with a character string argument is shown in the next example with its
C equivalent:
TABLE 11–6
Passing a CHARACTER string
Fortran call:
C equivalent:
CHARACTER*7 S
char s[7];
INTEGER B(3)
long b[3];
...
...
CALL CSTRNG( S, B(2) )
cstrng_( s, &b[1], 7L );
...
...
If the length of the string is not needed in the called routine, the extra arguments may
be ignored. However, note that Fortran does not automatically terminate strings with
the explicit null character that C expects. This must be added by the calling program.
One-Dimensional Arrays
Array subscripts in C start with 0.
C-Fortran Interface 11-13
TABLE 11–7
Passing a One-Dimensional Array
Fortran calls C
C calls Fortran
integer i, Sum
extern void vecref_
integer a(9)
( int[], int * );
external FixVec
...
...
int i, sum;
call FixVec ( a, Sum )
int v[9] = ...
...
vecref_( v, &sum );
------------------------------
...
void fixvec_ (
------------------------------
int v[9], int *sum )
subroutine VecRef( v, total)
{
integer i, total, v(9)
int i;
total = 0
*sum = 0;
do i = 1,9
for ( i = 0; i <= 8; i++ )
total = total + v(i)
*sum = *sum + v[i];
end do
}
...
Two-Dimensional Arrays
Rows and columns between C and Fortran are switched.
11-14
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 11–8
Passing a Two-Dimensional Array
Fortran calls C
C calls Fortran
REAL Q(10,20)
extern void
...
qref_( int[][10], int *);
Q(3,5) = 1.0
...
CALL FIXQ(Q)
int m[20][10] = ... ;
...
int sum;
------------------------------
...
void fixq_( float a[20][10] )
qref_( m, &sum );
{
...
...
------------------------------
a[5][3] = a[5][3] + 1.;
SUBROUTINE QREF(A,TOTAL)
...
INTEGER A(10,20), TOTAL
}
DO I = 1,10
DO J = 1,20
TOTAL = TOTAL + A(I,J)
END DO
END DO
...
Structures
C and FORTRAN 77 structures and Fortran 90 derived types can be passed to each
other’s routines as long as the corresponding elements are compatible.
C-Fortran Interface 11-15
TABLE 11–9
Passing FORTRAN 77 STRUCTURE Records
Fortran calls C
C calls Fortran
STRUCTURE /POINT/
struct point {
REAL X, Y, Z
float x,y,z;
END STRUCTURE
};
RECORD /POINT/ BASE
void fflip_ ( struct point *) ;
EXTERNAL FLIP
...
...
struct point d;
CALL FLIP( BASE )
struct point *ptx = &d;
...
...
------------------------------
fflip_ (ptx);
struct point {
...
float x,y,z;
------------------------------
};
SUBROUTINE FFLIP(P)
void flip_( struct point *v; )
STRUCTURE /POINT/
{
REAL X,Y,Z
float t;
END STRUCTURE
t = v -> x;
RECORD /POINT/ P
v -> x = v -> y;
REAL T
v -> y = t;
T = P.X
v -> z = -2.*(v -> z);
P.X = P.Y
}
P.Y = T
P.Z = -2.*P.Z
...
11-16
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 11–10
Passing Fortran 90 Derived Types
Fortran 90 calls C
C calls Fortran 90
TYPE point
struct point {
SEQUENCE
float x,y,z;
REAL :: x, y, z
};
END TYPE point
extern void fflip_ (
TYPE (point) base
struct point *) ;
EXTERNAL flip
...
...
struct point d;
CALL flip( base)
struct point *ptx = &d;
...
...
------------------------------
fflip_ (ptx);
struct point {
...
float x,y,z;
------------------------------
};
SUBROUTINE FFLIP( P )
void flip_( struct point *v; )
TYPE POINT
{
REAL :: X, Y, Z
float t;
END TYPE POINT
t = v -> x;
TYPE (POINT) P
v -> x = v -> y;
REAL :: T
v -> y = t;
T = P%X
v -> z = -2.*(v -> z);
P%X = P%Y
}
P%Y = T
P%Z = -2.*P%Z
...
C-Fortran Interface 11-17
Pointers
A FORTRAN 77 pointer can be passed to a C routine as a pointer to a pointer
because the Fortran routine passes arguments by reference.
TABLE 11–11
Passing a FORTRAN 77 POINTER
Fortran calls C
C calls Fortran
REAL X
extern void fpass_;
POINTER (P2X, X)
...
EXTERNAL PASS
float *x;
P2X = MALLOC(4)
float **p2x;
X = 0.
fpass_(p2x) ;
CALL PASS(X)
...
...
------------------------------
------------------------------
SUBROUTINE FPASS (P2X)
void pass_(x)
REAL X
int **x;
POINTER (P2X, X)
{
X = 0.
**x = 100.1;
...
}
C pointers are compatible with Fortran 90 (release 2.0) scalar pointers, but not array
pointers.
Passing Data Arguments by Value
Call by value is available only for simple data with FORTRAN 77, and only by
Fortran routines calling C routines. There is no way for a C routine to call a Fortran
routine and pass arguments by value. It is not possible to pass arrays, character
strings, or structures by value. These are best passed by reference.
11-18
Fortran Programming Guide ♦ Revision A, February 1999
Use the nonstandard Fortran function %VAL(arg) as an argument in the call.
In the following example, the Fortran routine passes x by value and y by reference.
The C routine incremented both x and y, but only y is changed.
C-Fortran Interface 11-19
TABLE 11–12
Passing Simple Data Arguments by Value: FORTRAN 77 Calling C
Fortran 77 calls C
REAL x, y
x = 1.
y = 0.
PRINT *, x,y
CALL value( %VAL(x), y)
PRINT *, x,y
END
———————————————————–
void value_( float x, float *y)
{
printf("%f, %f\n",x,*y);
x = x + 1.;
y = y + 1.;
printf("%f, %f\n",x,*y);
}
———————————————————–
Compiling and running produces output:
1.00000 0. x and y from Fortran
1.000000, 0.000000 x and y from C
2.000000, 1.000000 new x and y from C
1.00000 1.00000 new x and y from Fortran
11-20
Fortran Programming Guide ♦ Revision A, February 1999
Functions That Return a Value
A Fortran function that returns a value of type BYTE (FORTRAN 77 only), INTEGER,
REAL, LOGICAL, DOUBLE PRECISION, or REAL*16 (SPARC only) is equivalent to a
C function that returns a compatible type (see Table 11–1 and Table 11–2). There are
two extra arguments for the return values of character functions, and one extra
argument for the return values of complex functions.
Returning a Simple Data Type
The following example returns a REAL or float value. BYTE, INTEGER, LOGICAL,
DOUBLE PRECISION, and REAL*16 are treated in a similar way:
TABLE 11–13
Functions Returning a REAL or float Value
Fortran calls C
C calls Fortran
real ADD1, R, S
float r, s;
external ADD1
extern float fadd1_() ;
R = 8.0
r = 8.0;
S = ADD1( R )
s = fadd1_( &r );
...
...
------------------------------
------------------------------
float add1_( pf )
real function fadd1 (p)
float *pf;
real p
{
fadd1 = p + 1.0
float f ;
return
f = *pf;
end
f++;
return ( f );
}
C-Fortran Interface 11-21
TABLE 11–13
Functions Returning a REAL or float Value
(continued)
Returning COMPLEX Data
A Fortran function returning COMPLEX or DOUBLE COMPLEX is equivalent to a C
function with an additional first argument that points to the return value in memory.
The general pattern for the Fortran function and its corresponding C function is:
Fortran function
C function
COMPLEX FUNCTION CF(a1, a2, ..., an)
cf_ ()return, a1, a2, ..., an
struct { float r, i; } *;return
11-22
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 11–14
Function Returning COMPLEX Data
Fortran calls C
C calls Fortran
COMPLEX U, V, RETCPX
struct complex { float r, i; };
EXTERNAL RETCPX
struct complex c1, c2;
U = ( 7.0, -8.0)
struct complex *u=&c1, *v=&c2;
V = RETCPX(U)
extern retfpx_();
...
u -> r = 7.0;
------------------------------
u -> i = -8.0;
struct complex { float r, i; };
retfpx_( v, u );
void retcpx_( temp, w )
...
struct complex *temp, *w;
------------------------------
{
COMPLEX FUNCTION RETFPX(Z)
temp->r = w->r + 1.0;
COMPLEX Z
temp->i = w->i + 1.0;
RETFPX = Z + (1.0, 1.0)
return;
RETURN
}
END
In 64-bit environments and compiling with -xarch=v9, COMPLEX values are
returned in floating-point registers: COMPLEX and DOUBLE COMPLEX in %f0 and %f1,
and COMPLEX*32 in %f0, %f1, %f2, and %f3.
Returning a CHARACTER String
Passing strings between C and Fortran routines is not encouraged. However, a
Fortran character-string-valued function is equivalent to a C function with two
additional first arguments—data address and string length. The general pattern for
the Fortran function and its corresponding C function is:
C-Fortran Interface 11-23
Fortran function
C function
CHARACTER*n FUNCTION C(a1, ..., an)
void c_ (result, length, a1, ..., an)
char result[ ];
long length;
Here is an example:
11-24
Fortran Programming Guide ♦ Revision A, February 1999
TABLE 11–15
A Function Returning a CHARACTER String
Fortran calls C
C calls Fortran
CHARACTER STRING*16, CSTR*9
void fstr_( char *, int,
STRING = ’’
char *, int *, int );
STRING = ’123’ // CSTR(’*’,9)
char sbf[9] = "123456789";
...
char *p2rslt = sbf;
------------------------------
int rslt_len = sizeof(sbf);
void cstr_( char *p2rslt,
char ch = ’*’;
int rslt_len,
int n = 4;
char *p2arg,
int ch_len = sizeof(ch);
int *p2n,
/* make n copies of ch in sbf
int arg_len )
*/
{ /* return n copies of arg */
fstr_( p2rslt, rslt_len,
int count, i;
&ch, &n, ch_len );
char *cp;
...
count = *p2n;
------------------------------
cp = p2rslt;
FUNCTION FSTR( C, N)
for (i=0; i<count; i++) {
CHARACTER FSTR*(*), C
*cp++ = *p2arg ;
FSTR = ’’
}
DO I = 1,N
}
FSTR(I:I) = C
END DO
FSTR(N+1:N+1) = CHAR(0)
END
In this example, the C function and calling C routine must accommodate two initial
extra arguments (a pointer to the result string and the length of the string) and one
C-Fortran Interface 11-25
additional argument at the end of the list (length of character argument). Note that
in the Fortran routine called from C, it is necessary to explicitly add a final null
character.
Labeled COMMON
Fortran labeled COMMON can be emulated in C by using a global struct.
TABLE 11–16
Emulating Labeled COMMON
Fortran COMMON Definition
C "COMMON" Definition
COMMON /BLOCK/ ALPHA,NUM
extern struct block {
...
float alpha;
int num;
};
extern struct block block_ ;
main ()
{
...
block_.alpha = 32.;
block_.num += 1;
...
}
Note that the external name established by the C routine must end in an underscore
to link with the block created by the Fortran program. Note also that the C directive
#pragma pack may be needed to get the same padding as with Fortran. Both f77
and f90 align data in COMMON blocks to at most 4-byte boundaries.
11-26
Fortran Programming Guide ♦ Revision A, February 1999
Sharing I/O Between Fortran and C
Mixing Fortran I/O with C I/O (issuing I/O calls from both C and Fortran routines)
is not recommended. It is better to do all Fortran I/O or all C I/O, not both.
The Fortran I/O library is implemented largely on top of the C standard I/O library.
Every open unit in a Fortran program has an associated standard I/O file structure.
For the stdin, stdout, and stderr streams, the file structure need not be explicitly
referenced, so it is possible to share them.
If a Fortran main program calls C to do I/O, the Fortran I/O library must be
initialized at program startup to connect units 0, 5, and 6 to stderr, stdin, and
stdout, respectively. The C function must take the Fortran I/O environment into
consideration to perform I/O on open file descriptors.
However, if a C main program calls a Fortran subprogram to do I/O, the automatic
initialization of the Fortran I/O library to connect units 0, 5, and 6 to stderr,
stdin, and stdout is lacking. This connection is normally made by a Fortran main
program. If a Fortran function attempts to reference the stderr stream (unit 0)
without the normal Fortran main program I/O initialization, output will be written
to fort.0 instead of to the stderr stream.
The C main program can initialize Fortran I/O and establish the preconnection of
units 0, 5, and 6 by calling the f_init() FORTRAN 77 library routine at the start
of the program and, optionally, f_exit() at termination.
Remember: even though the main program is in C, you should link with f77.
Alternate Returns
Fortran’s alternate returns mechanism is obsolescent and should not be used if
portability is an issue. There is no equivalent in C to alternate returns, so the only
concern would be for a C routine calling a Fortran routine with alternate returns.
The Sun Fortran implementation returns the int value of the expression on the
RETURN statement. This is implementation dependent and its use should be avoided.
C-Fortran Interface 11-27
TABLE 11–17
Alternate Returns
C calls Fortran
Running the Example
int altret_ ( int * );
demo% cc -c tst.c
main ()
demo% f77 -o alt alt.f tst.o
{
alt.f:
int k, m ;
altret:
k =0;
demo% alt
m = altret_( &k );
12
printf( "%d %d\n", k, m);
The C routine receives the return value 2 from
}
the Fortran routine because it executed the
-----------------------------RETURN 2 statement.
SUBROUTINE ALTRET( I, *, *)
INTEGER I
I=I +1
IF(I .EQ. 0) RETURN 1
IF(I .GT. 0) RETURN 2
RETURN
END
11-28
Fortran Programming Guide ♦ Revision A, February 1999
Index
Index-29
">