Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://smdc.sinp.msu.ru/doc/Fortran90UsersGuide.pdf Äàòà èçìåíåíèÿ: Fri Dec 15 06:58:45 2006 Äàòà èíäåêñèðîâàíèÿ: Mon Oct 1 19:36:10 2012 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: guide 8.0

Fortran 90 User 's Guide

A Sun Microsystems, Inc. Business

2550 Garcia Avenue Mountain View, CA 94043 U.S.A. Part No.: 801-5492-10 Revision A, March 1995


© 1995 Sun Microsystems, Inc. 2550 Garcia Avenue, Mountain View, California 94043-1100 U.S.A. All rights reserved. This product and related documentation are protected by copyright and distributed under licenses restricting its use, copying, distribution, and decompilation. No part of this product or related documentation 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® and Berkeley 4.3 BSD systems, licensed from UNIX System Laboratories, Inc., a wholly owned subsidiary of Novell, Inc., and the University of California, respectively. Third-party font software in this product is protected by copyright and licensed from Sun's font suppliers. RESTRICTED RIGHTS LEGEND: Use, duplication, or disclosure by the United States Government is subject to the restrictions set forth in DFARS 252.227-7013 (c)(1)(ii) and FAR 52.227-19. The product described in this manual may be protected by one or more U.S. patents, foreign patents, or pending applications. TRADEMARKS Sun, the Sun logo, Sun Microsystems, Solaris, are trademarks or registered trademarks of Sun Microsystems, Inc. in the U.S. and certain other countries. UNIX is a registered trademark in the United States and other countries, exclusively licensed through X/Open Company, Ltd. OPEN LOOK is a registered trademark of Novell, Inc. PostScript and Display PostScript are trademarks of Adobe Systems, Inc. CRAY is a registered trademark of Cray Research, Inc. All other product names mentioned herein are the trademarks of their respective owners. All SPARC trademarks, including the SCD Compliant Logo, are trademarks or registered trademarks of SPARC International, Inc. SPARCstation, SPARCserver, SPARCengine, SPARCstorage, SPARCware, SPARCcenter, SPARCclassic, SPARCcluster, SPARCdesign, SPARC811, SPARCprinter, UltraSPARC, microSPARC, SPARCworks, and SPARCompiler are licensed exclusively to Sun Microsystems, Inc. Products bearing SPARC trademarks are based upon an architecture developed by Sun Microsystems, Inc. The OPEN LOOK® and SunTM Graphical User Interfaces were developed by Sun Microsystems, Inc. for its users and licensees. Sun acknowledges the pioneering efforts of Xerox in researching and developing the concept of visual or graphical user interfaces for the computer industry. Sun holds a non-exclusive 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. X Window System is a product of the Massachusetts Institute of Technology. THIS PRODUCT IS DERIVED FROM CRAY CF90TM, A PRODUCT OF CRAY RESEARCH, 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. THIS PUBLICATION COULD INCLUDE TECHNICAL INACCURACIES OR TYPOGRAPHICAL ERRORS. CHANGES ARE PERIODICALLY ADDED TO THE INFORMATION HEREIN; THESE CHANGES WILL BE INCORPORATED IN NEW EDITIONS OF THE PUBLICATION. SUN MICROSYSTEMS, INC. MAY MAKE IMPROVEMENTS AND/OR CHANGES IN THE PRODUCT(S) AND/OR THE PROGRAM(S) DESCRIBED IN THIS PUBLICATION AT ANY TIME.

Please Recycle


Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1 Operating Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Text Editing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Program Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5 Licensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Getting Started . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Compiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Running . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 Renaming the Executables . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 2 3 3 3 5 5 6 6 6

iii


3. Using the Compiler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Compile Command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Compile Link Sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Command-line File Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . Unrecognized Arguments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Compiler Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Actions/Options Frequently Used . . . . . . . . . . . . . . . . . . . . . . Actions and What Options Invoke Them . . . . . . . . . . . . . . . . Options and What Actions They Do . . . . . . . . . . . . . . . . . . . . 3.3 Miscellaneous Tips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Floating-Point Hardware Type . . . . . . . . . . . . . . . . . . . . . . . . . Many Options on Short Commands . . . . . . . . . . . . . . . . . . . . . 4. File System and File I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Directories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 File Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Path Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Relative Path Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Absolute Path Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.5 Redirection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Piping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9 9 10 10 11 12 12 13 13 15 33 33 33 35 35 37 37 37 38 38 40 41

iv

Fortran 90 User 's Guide


4.7 Accessing Files from Fortran Programs . . . . . . . . . . . . . . . Accessing Named Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Accessing Unnamed Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Passing File Names to Programs . . . . . . . . . . . . . . . . . . . . . . . 4.8 Direct I/O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.9 Internal Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. Program Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Simple Program Builds . . . . . . . . . . . . . . . . . . . . . . . . . . . . Writing a Script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating an Alias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using a Script or Alias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Program Builds with the make Program . . . . . . . . . . . . . . The make File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using make . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Tracking and Controlling Changes with SCCS . . . . . . . . . Putting Files under SCCS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Checking Files Out and In . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Libraries in General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Load Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Advantages of Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

42 42 43 43 45 46 49 49 49 50 50 50 50 50 51 52 52 54 57 57 58 58

Contents

v


6.2 Static Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Disadvantages of Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sample Creation of a Static Library . . . . . . . . . . . . . . . . . . . . . Sample Replacement in a Static Library ................

58 58 59 61 61 62 62 63 65 65 65 66 66 67 67 69 69 70 70 71 73

6.3 Dynamic Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Performance Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Binding Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A Simple Dynamic Shared Library . . . . . . . . . . . . . . . . . . . . . . 6.4 Consistent Compile and Link . . . . . . . . . . . . . . . . . . . . . . . 6.5 Library Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Installation Directory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Building Executables: ld Search order. . . . . . . . . . . . . . . . . . . Running Executables: ld Search order . . . . . . . . . . . . . . . . . . Build Paths and Run Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . Finding Built-in Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Global Program Checking (-Xlist) . . . . . . . . . . . . . . . . . Errors in General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Details ............................................

Using Global Program Checking . . . . . . . . . . . . . . . . . . . . . . . Suboptions for Global Checking Across Routines . . . . . . . . .

vi

Fortran 90 User 's Guide


7.2 The dbx Debugger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sample Program for Debugging . . . . . . . . . . . . . . . . . . . . . . . . A Sample dbx Session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Segmentation Fault--Finding the Line Number. . . . . . . . . . . Exception--Finding the Line Number . . . . . . . . . . . . . . . . . . . Trace of Calls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pointer to a Scalar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pointer to an Array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . User-Defined Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pointer to User-Defined Type . . . . . . . . . . . . . . . . . . . . . . . . . . Allocated Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Print Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Print Array Slices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Generic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Miscellaneous Tips . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Main Features of the Debugger . . . . . . . . . . . . . . . . . . . . . . . . . Help ............................................

77 78 79 81 83 84 85 86 87 89 91 92 93 94 96 96 97 99 99 100 101

8. Floating Point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 The General Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 IEEE Solutions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Contents

vii


8.4 IEEE Exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Detecting a Floating-point Exception . . . . . . . . . . . . . . . . . . . . Generating a Signal for a Floating-point Exception . . . . . . . . Default Signal Handlers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.5 IEEE Routines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flags and ieee_flags() . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Values and ieee_values() . . . . . . . . . . . . . . . . . . . . . . . . . . Exception Handlers and ieee_handler() . . . . . . . . . . . . . . Retrospective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nonstandard Arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Messages about Floating-point Exceptions . . . . . . . . . . . . . . . 8.6 Debugging IEEE Exceptions . . . . . . . . . . . . . . . . . . . . . . . . 8.7 Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.8 Miscellaneous Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . Kinds of Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Simple Underflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Use Wrong Answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Excessive Underflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. C­Fortran Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1 Sample Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 How to Use this Chapter . . . . . . . . . . . . . . . . . . . . . . . . . . .

102 102 102 102 103 104 106 107 111 111 112 113 113 114 114 115 116 116 119 120 121

viii

Fortran 90 User 's Guide


9.3 Compatibility Requirements . . . . . . . . . . . . . . . . . . . . . . . Function or Subroutine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Underscore in Names of Routines . . . . . . . . . . . . . . . . . . . . . . Case Sensitivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data Type Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Passing Arguments by Reference or Value . . . . . . . . . . . . . . . Character Strings and Order . . . . . . . . . . . . . . . . . . . . . . . . . . . Array Indexing and Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Libraries and Linking with the f90 Command . . . . . . . . . . . File Descriptors and stdio . . . . . . . . . . . . . . . . . . . . . . . . . . . . File Permissions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.4 Fortran Calls C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Arguments Passed by Reference (f90 Calls C). . . . . . . . . . . . Arguments Passed by Value (f90 Calls C) . . . . . . . . . . . . . . . Function Return Values (f90 Calls C) . . . . . . . . . . . . . . . . . . . Labeled Common (f90 Calls C) . . . . . . . . . . . . . . . . . . . . . . . . Alternate Returns (f90 Calls C) - N/A . . . . . . . . . . . . . . . . . . 9.5 C Calls Fortran . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Arguments Passed by Reference (C Calls f90) . . . . . . . . . . . . Arguments Passed by Value (C Calls f90) - N/A . . . . . . . . . Function Return Values (C Calls f90) . . . . . . . . . . . . . . . . . . . Labeled Common (C Calls f90) . . . . . . . . . . . . . . . . . . . . . . . . Alternate Returns (C Calls f90) . . . . . . . . . . . . . . . . . . . . . . .

122 123 124 124 125 127 129 130 131 132 133 134 134 141 141 147 148 149 149 155 155 160 161

Contents

ix


A. Features and Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.1 Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.2 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tabs in the Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Continuation Line Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fixed-Form Source of 96 Characters. . . . . . . . . . . . . . . . . . . . . Directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Source Form Assumed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Boolean Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abbreviated Size Notation for Numeric Data Types . . . . . . . Cray Pointers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cray Character Pointers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Intrinsics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.3 Directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . General Directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Form of General Directive Lines . . . . . . . . . . . . . . . . . . . . . . . . FIXED and FREE Directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . Parallel Directives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Form of Parallel Directive Lines . . . . . . . . . . . . . . . . . . . . . . . .

163 163 164 164 165 165 165 165 167 170 171 176 178 179 179 180 181 182 182

x

Fortran 90 User 's Guide


A.4 Compatibility with FORTRAN 77 . . . . . . . . . . . . . . . . . . . Source ............................................

184 184 184 184 185 187 188 188 188 189 189 190 190 190 191 191 192 192 195 195 196 196 196 197

Executables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I/O ............................................

Intrinsics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.5 Forward Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.6 Mixing Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.7 Module Files . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. iMPact: Multiple Processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.1 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Automatic Parallelization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Explicit Parallelizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.3 Speed Gained or Lost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.4 Number of Processors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. iMPact: Automatic Parallelization . . . . . . . . . . . . . . . . . . . . . . . C.1 What You Do . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C.2 What the Compiler Does . . . . . . . . . . . . . . . . . . . . . . . . . . . Parallelize the Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Dependency Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Definitions: Array, Scalar, and Pure Scalar . . . . . . . . . . . . . . .

Contents

xi


C.3 Definition: Automatic Parallelizing . . . . . . . . . . . . . . . . . . General Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Details ............................................

197 197 197 198 201 201 202 203 203 205 206 206 206 208 208 209 209 210 211 213 227

Exceptions for Automatic Parallelizing . . . . . . . . . . . . . . . . . . D. iMPact: Explicit Parallelization . . . . . . . . . . . . . . . . . . . . . . . . . D.1 What You Do . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D.2 What the Compiler Does . . . . . . . . . . . . . . . . . . . . . . . . . . . D.3 Parallel Directives. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Form of Directive Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . DOALL Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D.4 DOALL Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Explicitly Parallelizing a DOALL Loop . . . . . . . . . . . . . . . . . . . CALL in a Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D.5 Exceptions for Explicit Parallelizing . . . . . . . . . . . . . . . . . D.6 Risk with Explicit: Nondeterministic Results . . . . . . . . . . Testing is not Enough . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . How Indeterminacy Arises . . . . . . . . . . . . . . . . . . . . . . . . . . . . D.7 Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Join the SunPro SIG Today . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xii

Fortran 90 User 's Guide


Figures
Figure 4-1 Figure 4-2 Figure 4-3 File System Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Relative Path Name . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Absolute Path Name . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 38 39

xiii


xiv

Fortran 90 User 's Guide


Tables
Table 3-1 Table 3-2 Table 3-3 Table 3-4 Table 7-1 Table 7-2 Table 8-1 Table 8-2 Table 9-1 Table 9-2 Table 9-3 Table A-1 Table A-2 Table A-3 Table A-4 File Name Suffixes Fortran 90 Recognizes . . . . . . . . . . . . . . . . . Options Frequently Used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Actions/Options Sorted by Action . . . . . . . . . . . . . . . . . . . . . . . Summary of -Xlist Suboptions . . . . . . . . . . . . . . . . . . . . . . . . . . . -Xlist Combination Special or A La Carte Suboptions . . . . -Xlist Suboptions Summary . . . . . . . . . . . . . . . . . . . . . . . . . . ieee_flags Argument Meanings . . . . . . . . . . . . . . . . . . . . . . Functions for Using IEEE Values . . . . . . . . . . . . . . . . . . . . . . . . C Data Type to Fortran 90 Data Type ................... 11 13 13 32 74 74 105 107 125 126 132 170 178 179 182

Fortran 90 Data Type to C Data Type . . . . . . . . . . . . . . . . . . . . Characteristics of Three I/O Systems . . . . . . . . . . . . . . . . . . . . . Size Notation for Numeric Data Types . . . . . . . . . . . . . . . . . . . Nonstandard Intrinsics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . General Directives Guaranteed Only in the Current Release . Parallel Directives Guaranteed Only in the Current Release .

xv


Table B-1 Table D-1 Table D-2

Parallelization Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . DOALL General Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . DOALL Scheduling Parameters . . . . . . . . . . . . . . . . . . . . . . . . . .

191 205 205

xvi

Fortran 90 User 's Guide


Preface
This preface is organized into the following sections.
Purpose and Audience Before You Read This Book How This Book is Organized Related Documentation Conventions in Text page xvii page xviii page xviii page xviii page xxii

Purpose and Audience
This guide shows how to use Sun Fortran 90 1.0. Major topics of the guide are:

· · · · · · · · · · ·

Using the compiler command and options Global program checking across routines Using iMPactTM multiprocessor Fortran 90 MP Making and using libraries Using some utilities and development tools Using IEEE floating point with Fortran 90 Using debuggers with Fortran 90 Mixing C and Fortran 90

The guide is intended for scientists and engineers with the following: Thorough knowledge of Fortran 90 General knowledge of some operating system (experience with some OS) Particular knowledge of the SunOSTM commands cd, pwd, ls, cat

xvii


Before You Read This Book
If you are not familiar with Fortran 90, you may want to consult the following.

· ·

Fortran 90 Handbook (Fortran 90 language definition, including intrinsics) Fortran 90 Explained (Text book introduction to Fortran 90)

See "Related Manuals" on page xix.

How This Book is Organized
This book is organized as follows.
Chapter 1, Introduction Chapter 2, Getting Started Chapter 3, Using the Compiler Chapter 4, File System and File I/O Chapter 5, Program Development Chapter 6, Libraries Chapter 7, Debugging Chapter 8, Floating Point Chapter 9, C­Fortran Interface Appendix A, Features and Differences Appendix B, iMPact: Multiple Processors Appendix C, iMPact: Automatic Parallelization Appendix D, iMPact: Explicit Parallelization page 1 page 5 page 9 page 35 page 49 page 57 page 69 page 99 page 119 page 163 page 189 page 195 page 201

Related Documentation
The related kinds of documentation included with Fortran 90 are as follow:

· · · · ·

Paper manuals (hard copy) On-line manuals in the AnswerBookTM viewing system On-line man pages f90 -help variations On-line READMEs directory of information files

xviii

Fortran 90 User 's Guide


AnswerBook
The AnswerBook system displays and searches the on-line copies of the paper manuals. The system and manuals are included on the CD-ROM and can be installed to hard disc during installation. Installing and starting AnswerBook are described in the manual Installing SunSoft Developer Products on Solaris.

Related Manuals
The following manuals are provided on-line or on paper, as indicated.
Title Part Number 801-5492-10 875-1202-10 802-2190-10 801-7105-10 801-7639-10 802-1561-10 800-7895-10 X X X X Paper X AnswerBook X X X X X X X

Fortran 90 User 's Guide Fortran 90 Handbook, by Adams, Brainerd, et al Fortran 90 Browser
Debugging a Program Numerical Computation Guide

Installing SunSoft Developer Products on Solaris
What Every Computer Scientist Should Know About Floating-Point Arithmetic

man Pages
Purpose
A man page is intended to answer "What does it do?" and "How do I use it?"

· ·

Memory Jogger-- A man page reminds the user of details, such as arguments and syntax. It assumes you knew and forgot. It is not a tutorial. Quick Reference--A man page helps find something fast. It is brief, covering major highlights. It is a quick reference, not a complete reference.

xix


Using man Pages
To display a man page, use the man command. Example: Display the f90 man page.
demo$ man f90

Example: Display the man page for the man command.
demo$ man man

The man command uses the MANPATH environment variable, which can effect which set of man pages are accessed. See man(1).

Related man Pages
The following man pages may be of interest to Fortran 90 users.
f90(1) asa(1) dbx(1) debugger(1) fpr(1) ieee_flags(3M) ieee_handler(3M) matherr(3M) Invoke the Fortran 90 compiler. Print files having Fortran carriage-control. Debug by a command-line-driven debugger. Debug by a graphical-user-interface . Print files having Fortran carriage-control. Examine, set, or clear floating-point exception bits. Handle exceptions. Handle errors.

xx

Fortran 90 User 's Guide


f90 -help Variations
The following variations are meant to suggest other possibilities.
f90 -help | more f90 -help | grep "par" f90 -help | grep "lib" f90 -help | lp f90 -help > MyWay f90 -help | tail The list does not scroll off the screen. Show only parallel options. Show only library options. Print a copy on paper. Put list onto a file, regroup, reorder, delete, ... Show how to send feedback to Sun.

READMEs
The READMEs directory has information files: bug descriptions, information discovered after the manuals were printed, feedback form, and so forth.
Standard Installation Location /opt/SUNWspro/READMEs/ Nonstandard Installation to /my/dir/ /my/dir/SUNWspro/READMEs/

Contents File Names feedback Sun programmers email template file: Send feedback comments to Sun fortran_90 Fortran 90 bugs, new features. behavior changes, documentation errata

SIG
Sun Programmers Special Interest Group membership entitles you to other documentation and software. A membership form is included at the very end of this book. See "Join the SunPro SIG Today," on page 215.

xxi


Conventions in Text
We use the following conventions in this manual to display information.

· · ·

We show code listing examples in boxes.
WRITE( *, * ) 'Hello world'

Plain typewriter font shows prompts and coding. In dialogs, boldface typewriter font shows text the user types in.
demo$ echo hello hello demo$ s

· ·

Italics indicate general arguments or parameters that you replace with the appropriate input. Italics also indicate emphasis. For Solaris 2.x, the default shell is sh and the default prompt is the dollar sign ($). Most systems have distinct host names, and you can read some of our examples more easily if we use a symbol longer than a dollar sign. Examples generally use "demo$" as the system prompt; where the csh shell is shown, we use "demo%" as the system prompt. A small clear triangle shows a blank space where that is significant. 36.001

· ·

We generally tag nonstandard features with a small black diamond (o). Wherever we indicate that a feature is nonstandard, that means a program using it does not conform to the ANSI X3.198-1992 standard, as described in American National Standard for Programming Language--Fortran--Extended, ANSI X3.198-1992, 1992, American National Standards Institute, Inc., informally abbreviated as the Fortran 90 Standard. We We We else usually show Fortran 90 examples in free form, not fixed form or tab. usually abbreviate "Sun Fortran 90" as "f90". usually show Fortran 90 keywords and intrinsics in uppercase, and all in lowercase or mixed case.

· · ·

xxii

Fortran 90 User 's Guide


Introduction
This chapter is organized into the following sections.
Operating Environment Text Editing Program Development Debugging Licensing page 2 page 2 page 3 page 3 page 3

1

Sun Fortran 90 comes with a programming environment, including certain operating system calls and support libraries. It integrates with powerful development tools, including SunSoftTM tools such as the Debugger, make, MakeToolTM, and TeamWareTM. Some examples assume you installed the Source Compatibility Package.

iMPactTM and WorkshopTM
The compiler is available in various packages and configurations:

· ·

Alone, or as part of a package, such as the Fortran 90 Workshop With or without the iMPact MT/MP multiple processor package

1


1
1.1 Operating Environment
Sun Fortran 90 runs in the Solaris® 2.x operating environments. The Solaris 2.x operating environment includes (among other components) the SunOSTM 5.x operating system. SunOS 5.x is based on the System V Release 4 (SVR4) UNIX operating system, and the ONC+TM family of published networking protocols and distributed services, including ToolTalkTM.

Abbreviations

· ·

Solaris 2.x is an abbreviation for "Solaris 2.3 and later." SunOS 5.x is an abbreviation for "SunOS 5.3 and later."

1.2 Text Editing
There are several text editors available.

vi

The major text editor for source programs is vi (vee-eye), the visual display editor. It has considerable power because it offers the capabilities of both a line and a screen editor. vi also provides several commands specifically for editing programs. These are options you can set in the editor. Two examples are the autoindent option, which supplies white space at the beginning of a line, and the showmatch option, which shows matching parentheses. For more information, read the vi section of the manual. The textedit editor and other editors are available, including ed and ex. For the emacs editor, and other editors not from Sun, read the Sun document CatalystTM, a Catalog of Third-Party Software and Hardware. Xemacs is an Emacs editor that provides interfaces to the selection service and to the ToolTalkTM service. The EOS package ("Era On Sparcworks") uses these two interfaces to provide simple yet useful editor integration with two SPARCworks tools: the SourceBrowser and the Debugger. Era is an earlier name of this editor. It is available through the University of Illinois, by anonymous ftp, at ftp.cs.uiuc.edu:/pub/era

textedit emacs

xemacs

2

Fortran 90 User 's Guide


1
1.3 Program Development
There are several development tools available. asa This utility is a Fortran output filter for printing files that have Fortran carriagecontrol characters in column one. The UNIX implementation on this system does not use carriage-control since UNIX systems provide no explicit printer files. You use asa when you want to transform files formatted with Fortran carriage-control conventions into files formatted according to UNIX lineprinter conventions. See asa(1) and fpr(1). This utility splits one Fortran file of several routines into several files, so that there is one routine per file.

fsplit

1.4 Debugging
There are two debugging tools. dbx An interactive symbolic debugger that understands Sun Fortran 90 programs (available with the SPARCworks set). A window, icon, mouse, and pointer interface to dbx (in SPARCworks set).

debugger

1.5 Licensing
This compiler uses network licensing. Before you use Sun Fortran 90, purchase and install a SunSoft Fortran 90 license. When you invoke the compiler, if a license is available, the compiler simply starts. If no license is available, your request for a license is put on a queue, and your compile continues when a license becomes available. See also ­noqueue and -xlicinfo. Licensing information is in the manual Installing SunSoft Developer Products on Solaris, including (among other items):

· · ·

Installing a license Starting a license daemon Restarting a license daemon after a license server crash

Introduction

3


1

4

Fortran 90 User 's Guide


Getting Started
This chapter is organized into the following sections.
Summary Compiling Running Renaming the Executables page 5 page 6 page 6 page 6

2

This chapter gives a bare minimum on how to compile and run Fortran 90 programs under Solaris. This chapter is for you if you know Fortran 90 thoroughly and need to start writing programs in this Fortran 90 immediately. Skip to Chapter 3, "Using the Compiler," to learn more about it first.

2.1 Summary
Before you use this release of f90, it must be installed and licensed. Read Installing SunSoft Developer Products on Solaris .

To use this Fortran 90 involves three steps:

· · ·

Write and save a Fortran 90 program; use .f90 or .f as file name suffix. Compile and link this file using the f90 command. Execute by typing the name of the executable file.

5


2
Example: This program displays a message on the screen.
demo$ cat hack.f90 PROGRAM Opinion PRINT *, 'Real programmers hack Fortran 90!' END PROGRAM Opinion demo$ s

2.2 Compiling
Compile and link using the f90 command as follows.
demo$ f90 hack demo$ s

In the example above, f90 compiles hack.f90 and puts the executable code in the a.out file.

2.3 Running
Run the program by typing a.out on the command line.
demo$ a.out Real programmers hack Fortran 90! demo$ s

2.4 Renaming the Executables
It is awkward to have the result of every compilation on a file called a.out, since if such a file exists, it is overwritten. You can avoid this in two ways.

·

After each compilation, use mv to change the name of a.out.
demo$ mv a.out maven demo$ s

·

On the command line, use -o to rename the output executable file.
demo$ f90 ­o maven hack demo$ s

The above command places the executable code in the maven file.

6

Fortran 90 User 's Guide


2
Either way, run the program by typing the name of the executable file.
demo$ maven Real programmers hack Fortran 90! demo$ s

At this point, read Chapter 3, "Using the Compiler for the compiler options and the summary of performance optimization. If you are not familiar with a UNIX file system, read Chapter 4, "File System and File I/O." or refer to any introductory UNIX book.

Getting Started

7


2

8

Fortran 90 User 's Guide


Using the Compiler
This chapter is organized into the following sections.
Compile Command Compiler Options Miscellaneous Tips page 9 page 12 page 33

3

3.1 Compile Command
The syntax of a simple compiler command is as follows.
Before you use this release of f90, it must be installed and licensed. Read Installing SunSoft Developer Products on Solaris . f90 [options] sfn ...

where sfn is a Fortran 90 source file name, and options is one or more of the compiler options. Example: A compile command with two files.
demo$ f90 growth.f90 fft.f90

Example: A compile command, same files, with some options.
demo$ f90 -g -P growth.f90 fft.f90

9


3
A more general form of the compiler command is as follows.
f90 [options] fn ... [-lx]

· · ·

The fn is a file name (not necessarily of a Fortran 90 source file). See "Command-line File Names" on page 11. The -lx is the option to link with library libx.a. The -lx is after the list of file names. Always safer. Not always required.

The files and the results of compilations are linked (in the order given) to make an executable program, named (by default) a.out or with a name specified by the -o option.

Purpose
The purpose of f90 is to translate source to an executable file. Other major uses:

· · · ·

Translate source code files to relocatable binary (.o) files Link .o files into an executable load module (a.out) file Show the commands built by the compiler, but do not execute Prepare for debugging

Compile Link Sequence
With the above commands, if you successfully compile the files growth.f90 and fft.f90, the object files growth.o and fft.o are generated, then an executable file is generated with the default name a.out. The files growth.o and fft.o are not removed. If there is more than one object file (.o file), then the object files are not removed. This allows easier relinking if there is a linking error. If the compile fails, you get an error message for each error, the a.out file is not generated, and the remaining .o files are not generated.

10

Fortran 90 User 's Guide


3
Compile and Link in Separate Steps
You can compile and link in separate steps. This is usually done if one of several source files was changed--that way you need not recompile all the other source files. Example: Compile and link in separate steps.
demo$ f90 demo$ f90 -c file1.f90 file2.f90 file3.f90 file1.o file2.o file3.o {make .o files} {make a.out file}

Of course, every file named in the first step (as a .f90 file) must also be named in the second step (as a .o file).

Consistent Compile and Link
Be consistent with compiling and linking. If you compile and link in separate steps, and you compile any subprogram with -dalign or ­fast, then be sure to link with the same options.

Command-line File Names
If a file name in the command line has any of the following suffixes, the compiler recognizes it; otherwise it is passed to the linker.
Table 3-1 Suffix .f90 Language Fortran 90 File Name Suffixes Fortran 90 Recognizes Form Free Action Compile Fortran 90 source files, put object files in current directory; default name of object file is that of the source but with .o suffix. Same as .f90, but different source form Same as .f Same as .f Assemble source files with the assembler. Pass object files through to the linker.

.f .for .ftn .s .o

Fortran 90 or standard FORTRAN 77 Same as .f Same as .f Assembler Object Files

Fixed Fixed Fixed

Fixed-form source and free-form source are explained in Section 3.3 Source Form, of the Fortran 90 Handbook.

Using the Compiler

11


3
Unrecognized Arguments
Any arguments f90 does not recognize are taken to be one of the following:

· · · · ·

Linker option arguments Names of f90-compatible object programs (maybe from a previous run) Libraries of f90-compatible routines

If an unrecognized argument: Has a "-", then it is an option, and generates a warning. Has no "-", then it generates no warnings; but if the linker does not recognize them, the linker issues error messages.

3.2 Compiler Options
This compiler has the power of many optional features, so this tends to produce a very long list of features. To help you use this long list, the options are organized from different perspectives--so you can look up an action to see which option does it, or you can look up an option to see what it does.
List/Perspective q Actions and What Invokes Them (Actions/Options Sorted by Action) Frequent--Actions/Options Frequently Used Summary--One-line descriptions (See also "compile action" in the Index.) q Options and What They Do (Options/Actions Sorted by Option) Summary--One-line descriptions Full Descriptions--Examples, risks, trade-offs, restrictions, interactions All risks, trade-offs, restrictions, interactions, and examples Some risks, trade-offs, restrictions, interactions, examples (See also: the option name in the Index.) See page 15, ... man f90 f90 -help See page 13 See page 13 How to Get It

12

Fortran 90 User 's Guide


3
Actions/Options Frequently Used
A few options are needed by almost every programmer.
Table 3-2 Action Debug--produce additional symbol table information for the debugger. Performance--make executable run faster using a selection of options. Performance--make executable run faster using the optimizer. Library--Allow or disallow dynamic libraries for the entire executable. -dy, -dn Compile only--suppress linking; make a .o file for each source file. Name the final output file nm instead of a.out. Display a list of compiler options. Options Frequently Used Option ­g ­fast ­O[n] ­dbinding ­c ­o nm ­help Details page 20 page 19 page 25 page 15 page 17 page 16 page 25 page 20

Bind as dynamic (or static) any library listed later in the command. -Bdynamic, -Bstatic ­Bbinding

Check "Details" for trade-offs, risks, restrictions, interactions, and examples.

Actions and What Options Invoke Them
Actions/Options Sorted by Action--This section groups related actions together. Check "Details" for risks, trade-offs, restrictions, interactions, and examples.
Table 3-3 Action Debug Compile for use with the debugger. Global checking--across routines (arguments, commons, parameters, ...). Version ID--show version ID along with name of each compiler pass. Library Bind as dynamic (or static) any library listed later in the command. Allow or disallow dynamic libraries for the entire executable. Build a dynamic shared library. Name a shared dynamic library. Directory--search this directory first. Link with library libx. ­Bbinding ­dbinding ­G ­hname -Ldir -lx page 15 page 17 page 20 page 20 page 22 page 21 ­g -Xlist -V page 20 page 32 page 30 Actions/Options Sorted by Action Option Details

Using the Compiler

13


3
Table 3-3 Action Library (continued) Multithread safe libraries, low level threads. Paths--store into object file. No automatic libraries. No run path in executable. Performance Faster execution--make executable run faster using a selection of options. ­fast Generate code to run on generic SPARC architecture. Generate code to run on SPARC V8 architecture. Use the best floating-point arithmetic for this machine. Optimize for execution time. Use selected math routines optimized for performance. Reset -fast so that it does not use -xlibmopt. Parallelization Parallelize automatically and with explicit directives. Parallelize explicitly. Stack local variables to allow better optimizing with parallelizing. Profile by Procedure for gprof. Procedure for prof. Information and Warnings Verbose--print name of each compiler pass. Version ID--show version ID. Warnings--suppress warnings. Licensing License information--display license server user ids. No license queue. Source Forms Fixed form source. Free form source. -fixed -free page 19 page 19 -xlicinfo -noqueue page 30 page 24 -v -V -w page 30 page 30 page 31 -pg -p page 27 page 26 -parallel -explicitpar -stackvar page 26 page 18 page 29 ­cg89 -cg92 -native -O[n] -xlibmopt -xnolibmopt page 19 page 16 page 16 page 23 page 25 page 31 page 30 -mt -R list -nolib -norunpath page 22 page 28 page 24 page 24 Actions/Options Sorted by Action (Continued) Option Details

14

Fortran 90 User 's Guide


3
Table 3-3 Action Miscellaneous ANSI conformance check--identify many non-ANSI extensions. Compile only, do not make a.out, do not execute CIF--generate a compiler information file. Command--show command line built by driver, but do not execute. Align on 8-byte boundaries. Do not trap floating-point exceptions. Options--display the list of options. Include path--add dir to the search path for INCLUDE statements. Module directory--look for Fortran 90 modules in the dir directory. Output--rename the output file. DO loops--use one trip DO loops. Pass option list to program. Assembly source--generate only assembly source code. Symbol table--strip executable of symbol table (prevents debugging). Temporary files--set directory to locate temporary files. Time for execution--display for each compilation pass. -ansi -c -db ­dryrun -f -fnonstop ­help -Idir -Mdir -o outfil -onetrip -Qoption pr ls -S -s -temp=dir -time page 15 page 16 page 17 page 17 page 19 page 19 page 20 page 21 page 23 page 25 page 25 page 27 page 30 page 29 page 30 page 30 Actions/Options Sorted by Action (Continued) Option Details

Options and What Actions They Do
Options/Actions Sorted by Option--This section shows all f90 options, with a full description, including risks, restrictions, caveats, interactions, examples, and other details.

­ansi ­Bbinding

ANSI conformance check--identify many non-ANSI extensions. Bind as dynamic (or static) any library listed later in the command. No space is allowed between -B and dynamic or static, and either dynamic or static must be included.

· ·

­Bdynamic: Prefer dynamic binding (try for shared libraries). -Bstatic: Require static binding (no shared libraries).

If you have neither -Bdynamic nor -Bstatic, you get the default: dynamic.

Using the Compiler

15


3
For ­Bdynamic and -Bstatic, there is asymmetry besides prefer/require:

· ·

If you specify static, but it finds only a dynamic version, then the library is not linked, and you get a warning that the "library was not found." If you specify dynamic, but it finds only a static version, then the library is linked, and you get no warning.

You can toggle -Bstatic and -Bdynamic on the command line. That is, you can link some libraries statically and some dynamically by specifying -Bstatic and -Bdynamic any number of times on the command line. These are loader/linker options. If you compile and link in separate steps, and you need -Bbinding, then you need it in the link step also.

­c

Compile only, do not make a.out, do not execute. Suppress linking by the loader. Make a .o file for each source file. Do not make an executable file. You can name a single object file explicitly using the -o option.

­cg89

Generate code to run on generic SPARC architecture. Use a subset of the SPARC V8 instruction set. With -cg89 and optimization, the instructions are scheduled for faster executables on a generic SPARC machine. Code compiled with -cg89 does run on -cg92 hardware.

-cg92

Generate code to run on SPARC V8 architecture. Allow the use of the full SPARC V8 instruction set.

General Comments on -cg89 and -cg92

· · ·

For SPARC systems, the default code generation option is -cg89. You can mix routines compiled -cg89 with routines compiled -cg92; that is, you can have both kinds in one executable. Use fpversion(1) to tell which to use so the executables run faster: -cg89 or -cg92. It may take about a minute to start the full report.

16

Fortran 90 User 's Guide


3
­dalign
Allow f90 to use double load/store. Generate double load/store instructions wherever possible for faster execution. Using this option automatically triggers the ­f option, which causes all double-precision and quadruple-precision data types (both real and complex) to be double aligned. With -dalign, you may not get ANSI standard Fortran 90 alignment. It is a trade-off of portability for speed. If you compile one subprogram with -dalign, compile all subprograms of the program with ­dalign.

-db

Compiler information file. Generate a compiler information (CIF) file.

­dryrun ­d[y,n]

Commands--show commands built by driver, but do not execute. Allow or disallow dynamic libraries for the entire executable. No space is allowed between -d and y or n. Either y or n must be included.

· ·

-dy: Yes--allow dynamically bound libraries (allow shared libraries). -dn: No--do not allow dynamically bound libraries (no shared libraries).

If you have neither -dy nor -dn, you get the default: y. These apply to the whole executable. Use only once on the command line. If a.out uses only static libraries, then -dy causes a few seconds delay at runtime it makes the dynamic linker be invoked when a.out is run. This takes a few seconds to invoke and find that no dynamic libraries are needed. -dbinding is a loader/linker option. If you compile and link in separate steps, and you need -dbinding, then you need it in the link step.

­e

Extend the source line maximum length to 132 columns. Accept lines up to 132 characters long.

Using the Compiler

17


3
-explicitpar
Multiprocessor--parallelize explicitly. You do the dependency analysis: analyze and specify loops for inter-iteration data dependencies. The software parallelizes the specified loops. The -explicitpar option requires the Multiprocessor Fortran 90 multiprocessor enhancement package. To get faster code, this option requires a multiprocessor system. On a single-processor system the generated code usually runs slower. Before you parallelize explicitly, see Appendix B, "iMPact: Multiple Processors," Appendix C, "iMPact: Automatic Parallelization," and Appendix D, "iMPact: Explicit Parallelization." Summary: To parallelize explicitly, do the following.

· · ·

Analyze the loops to find those that are safe to parallelize. Insert !MIC$ DOALL to parallelize a loop. Use the -explicitpar option.

Example: Insert a parallel pragma immediately before the loop.
... !MIC$ DOALL DO i = 1, n a(i) = b(i) * c(i) END DO ...

Example: Compile to explicitly parallelize.
demo$ f90 -explicitpar any.f90

Restrictions: · -g turns off -explicitpar.

· · ·

Avoid -explicitpar if you do your own thread management. See ­mt. Do not mix parallelized f77 and parallelized f90. If you use -explicitpar and compile and link in separate steps, then link with -explicitpar.

18

Fortran 90 User 's Guide


3
­f
Align on 8-byte boundaries. Align all COMMON blocks and all double-precision and quadruple-precision local data on 8-byte boundaries. This applies to both real and complex data. If you compile with -f for any subprogram of a program, then compile all subprograms of that program with -f.

­fast

Faster execution--make executable run faster using a selection of options. Select the combination of compilation options that optimize for speed of execution without excessive compilation time. This should provide close to the maximum performance for most realistic applications. If you combine -fast with other options, the last specification applies. If you do not specify the level (as in -fast -O) you get -fast -O3. If you compile and link in separate steps, and you compile with -fast, then be sure to link with -fast.

-fixed

Fixed-form source. Interpret all Fortran 90 source files by fixed form rules. Overrides file suffix. See also, "FIXED and FREE Directives" on page 181.

-flags -fnonstop

Synonym for -help. Do not trap floating-point exceptions. Without -fnonstop, there is a trap on the invalid, overflow, and divide by zero floating-point exceptions.

-free

Free-form source. Interpret all Fortran 90 source files by free form rules. Overrides file suffix. See also "FIXED and FREE Directives" on page 181.

Using the Compiler

19


3
­g
Debug--produce additional symbol table information for the debugger. Produce a symbol table of information for the debuggers. You get much more debugging power if you compile with -g before using the debuggers.

· ·
­G

-g overrides -O. -g is turned off by -explicitpar, -parallel, or -reduction.

Library--build a dynamic library. Tell the linker to build a dynamic library. Without -G, the linker builds an executable file. With -G, it builds a dynamic library.

­hnm

Library--make nm be the name of the generated shared dynamic library. When generating a shared dynamic library, the compile-time linker records the specified name in the library file as the internal name of the library. If there is no -hnm option, then no internal name is recorded in the library file, and the path of the library is stored instead. Advantage--If the library has an internal name, then whenever the executable is run, the linker needs only a library with the exact same internal name--it can be in any path the linker is searching. If the library has no internal name, then whenever the executable is run, the linker must find the library in the exact same path used when the executable was created That is, the internal name method is more flexible.

Remarks

· · · · ·

A space between -h and nm is optional. In general, this name must be the same as what follows the -o. The -hnm option is meaningless without -G. Internal names facilitate versions of a dynamic library. This is a linker option.

See the Linker and Libraries Manual.

­help

Options--display the list of options. Display an equivalent of this list of options. This also shows how to send feedback comments to Sun.

20

Fortran 90 User 's Guide


3
Variations for -help:

· · ·

f90 -help | more f90 -help | grep "par" f90 -help | tail

Do not scroll the list off screen. Show only parallel options. Show how to send feedback to Sun.

See also "f90 -help Variations" on page xxi.

­Iloc

Include path--add to the search path for INCLUDE statements. Insert the path loc at the start of the list of directories in which to search for Fortran 90 INCLUDE files.

· · ·

No space is allowed between -I and loc. Invalid directories are just ignored with no warning message. The -Iloc applies to INCLUDE files with relative, not absolute, path names.

Example: f90 -I/usr/applib growth.f90 Above, f90 searches for INCLUDE files in the source file directory and then in the /usr/applib directory. Use -Iloc again to insert more paths. Example: f90 -Ipath1 -Ipath2 any.f90 Search Order: The search for INCLUDE files is in the following order:

· ·
­lx

The directory containing the source file Directories named in -Iloc options

Library--link with library libx. Pass -lx to the linker. ld links with object library libx. If shared library libx.so is available, ld uses it, otherwise, ld uses archive library libx.a. If it uses a shared library, the name is built in to a.out. No space is allowed between -l and x character strings. Example: Link with the library libabc.
demo$ f90 any.f90 ­labc

Use -lx again to link with more libraries.

Using the Compiler

21


3
Example: Link with the libraries liby and libz.
demo$ f90 any.f90 ­ly -lz

See also Section 6.5, "Library Paths," on page 65.

­Ldir

Library--search this directory first. Add dir at the start of the list of object-library search directories. While building the executable file,ld(1) searches dir for archive libraries (.a files) and shared libraries (.so files). A space between -L and dir is optional. The directory dir is not built in to the a.out file. See also ­lx. ld searches dir before the default directories. See "Building Executables: ld Search order" on page 66. For the relative order between LD_LIBRARY_PATH and -Ldir, see ld(1). Example: Use -Ldir to specify a library search directory.
demo$ f90 -Ldir1 any.f90

Example: Use -Ldir again to add more directories.
demo$ f90 -Ldir1 -Ldir2 any.f90

Restrictions

· ·

No -L/usr/lib: Do not use ­Ldir to specify /usr/lib. It is searched by default. Including it here may prevent using the unbundled libm. No -L/usr/ccs/lib: In Solaris 2.x, do not use ­Ldir to specify /usr/ccs/lib. It is searched by default. Including it here may prevent using the unbundled libm.

­mt

Multithread safe libraries--use for low level threads. Use multithread safe libraries. If you do your own low level thread management this helps prevent conflicts between threads. Use -mt if you mix C and Fortran 90 and you do your own thread management of multithread C coding using the libthread primitives. Before you use your own multithreaded coding, read the "Guide to Multi-Thread Programming."

22

Fortran 90 User 's Guide


3
The ­mt option does not require the Multiprocessor Fortran 90 multiprocessor enhancement package, but to compile and run it does require Solaris 2.2 or later. The equivalent of -mt is included automatically with -autopar, -explicitpar, or -parallel. On a single-processor system the generated code can run slower with the ­mt option, but not usually by a significant amount. The Fortran 90 library libf90 is multithread safe. The FORTRAN 77 library that is linked in if you use -mt, libF77_mt, is also multithread safe.

Restrictions for -mt

·

With -mt, if a function does I/O, do not name that function in an I/O list. Such I/O is called recursive I/O, and it causes the program to hang (deadlock). Recursive I/O is unreliable anyway, but is more apt to hang with -mt. In general, do not combine your own multi-threaded coding with -autopar, -explicitpar, or -parallel. Either do it all yourself or let the compiler do it. You may get conflicts and unexpected results if you and the compiler are both trying to manage threads with the same primitives.

·

-Mdir

Modules--look for Fortran 90 modules in the dir directory also. No space is allowed between -M and dir. By default, such files are sought in the current working directory. The -Mdir option allows you to keep them in some other location in addition. Background--If a file containing a Fortran 90 module is compiled, f90 generates a module file (.M file) in addition to the .o file.

­native

Native floating point--use what is best for this machine. Direct the compiler to decide which floating-point options are available on the machine the compiler is running on, and generate code for the best one. If you compile and link in separate steps, and you compile with the -native option, then be sure to link with -native.

Using the Compiler

23


3
-nolib
No automatic libraries. Do not automatically link with any system or language library; that is do not pass any -lx options on to ld. The default is to link such libraries into the executables automatically, without the user specifying them on the command line. The -nolib option makes it easier to link one of these libraries statically. The system and language libraries are required for final execution. It is the users responsibility to link them in manually. This provides complete control (and with control comes responsibility) for the user. For example, a program linked dynamically with libF77 fails on a machine that has no libF77. When you ship your program to your customer, you can ship libF77 or you can link it into your program statically. Example: Link libm statically and link libc dynamically.
demo$ f90 -nolib any.f90 -lf90 -Bstatic -lm -Bdynamic -lc

There is no dynamic libf90; it is always linked statically. Order for -lx options is important. Use the order shown in the example.

­noqueue

No license queue. If you use this option, and no license is available, the compiler returns without queueing your request and without doing your compile. A nonzero status is returned for testing in make files.

-norunpath

No run path in executable. If an executable file uses shared libraries, then the compiler normally builds in a path that tells the runtime linker where to find those shared libraries. The path depends on the directory where you installed the compiler. The -norunpath option prevents that path from being built in to the executable. This option is helpful if you have installed in some nonstandard location, and you ship an executable to your customers, but you do not want to make the customers deal with that nonstandard location. Compare with -Rlist.

24

Fortran 90 User 's Guide


3
­o nm
Output file--rename the output file. Name the final output file nm instead of a.out. There must be a blank between -O and nm.

­onetrip

DO loops--use one trip DO loops. Compile DO loops so that they are performed at least once if reached. DO loops in Fortran 90 ordinarily are not performed at all if the upper limit is less than the lower limit, unlike some implementations of FORTRAN 66 DO loops.

­O[n]

Optimize the object code for speed of execution. The n can be 1, 2, 3, or 4. No space is allowed between -O and n. If -O[n] is not specified, the compiler still performs a default level of optimization; that is, it executes a single iteration of local common subexpression elimination and live/dead analysis. ­g suppresses ­On.

­O

If you do not specify an n, f90 uses whatever n is most likely to yield the fastest performance for most reasonable applications. For the current release, this is 3. Do only conservative scalar optimization. This level usually results in moderate code size and compile time. If an optimization can create a false exception, then that optimization is not performed. At this level, f90 can analyze whether a variable is used before it is defined.

­O1

­O2 ­O3

Do moderate optimization. Do aggressive scalar optimization. Some of these optimizations can create false exceptions.This level can result in larger code size and compile time. At this level, f90 can analyze whether a variable is used before it is defined.

Using the Compiler

25


3
­O4 ­p
For this release, -O4 is equivalent to -O3. Profile by procedure for prof. Prepare object files for profiling, see prof (1). If you compile and link in separate steps, and if you compile with the -p option, then be sure to link with the -p option. -p with prof is provided mostly for compatibility with older systems. -pg with gprof does more.

-parallel

Multiprocessor--Parallelize automatically and parallelize explicitly indicated loops. Parallelize loops both automatically by the compiler and as explicitly specified by the programmer. With explicit parallelization of loops, there is a risk of producing incorrect results. If optimization is not at -O3, then it is raised to -O3. Restrictions:

· · · ·

-g turns off -parallel. Avoid -parallel if you do your own thread management. See ­mt. Do not mix parallelized f77 and parallelized f90. If you use -parallel and compile and link in separate steps, then link with -parallel.

The -parallel option requires the Multiprocessor Fortran 90 multiprocessor enhancement package. To get faster code, use this option on a multiprocessor SPARC system. On a single-processor system the generated code usually runs slower. Number of Processors--To request a number of processors, set the PARALLEL environment variable. The default is 1.

· · ·

Do not request more processors than are available. If N is the number of processors on the machine, then for a one-user, multiprocessor system, try PARALLEL=N-1. See Section B.4, "Number of Processors."

Before you use -parallel, see Appendix B, "iMPact: Multiple Processors," Appendix C, "iMPact: Automatic Parallelization," and Appendix D, "iMPact: Explicit Parallelization."

26

Fortran 90 User 's Guide


3
­pg
Profile by procedure for gprof. Produce counting code in the manner of ­p, but invoke a runtime recording mechanism that keeps more extensive statistics and produces a gmon.out file at normal termination. Then you can make an execution profile by running gprof (1). -pg and gprof are complementary to -a and tcov. Library options must be after the .f and .o files (-pg libraries are static). If you compile and link in separate steps, and you compile with -pg, then be sure to link with -pg. Compare this profiling method with the one described in the manual Performance Tuning an Application. For Solaris 2.x, when the operating system is installed, gprof is included if you do a Developer Install, rather than an End User Install; it is also included if you install the package SUNWbtool.

-Qoption pr op

Option--pass option list to specified program. Pass the option list op to the program pr. There must be a blank between Qoption and pr and op. The Q can be uppercase or lowercase. The list is a comma-delimited list of options, no blanks within the list. Each option must be appropriate to that program and can begin with a minus sign. The assembler used by the compiler is named fbe. Example: Pass the help option for help to the linker, ld.
demo$ f90 ­Qoption ld -Dhelp src.f90

Example: Pass the load map to the linker, ld.
demo$ f90 ­Qoption ld -m src.f90

-reduction

Multiprocessor--reduction loops: analyze loops for reduction. Analyze loops for reduction during automatic parallelization. There is potential for roundoff error with the reduction. The -reduction option requires the Multiprocessor Fortran 90 multiprocessor enhancement package. To get faster code, this option requires a multiprocessor system. On a single-processor system the generated code usually runs slower.

Using the Compiler

27


3
-reduction conflicts with -g. Before you use -reduction, see Appendix C, "iMPact: Multiple Processors" and Appendix D, "iMPact: Automatic Parallelization." Reduction works only during parallelization. If you specify -reduction without -parallel, the compiler does no reduction. If you have a directive explicitly specifying a loop, then there will be no reduction for that loop. Example: Automatically parallelize with reduction.
demo$ f90 -parallel -reduction any.f90

­R ls

Library paths--store paths into object file. While building the executable file, store a list of library search paths into it.

· ·

ls is a colon-separated list of directories for library search paths. The blank between -R and ls is optional.

Multiple instances of this option are concatenated together, with each list being separated by a colon. How this list is used--The list will be used at runtime by the runtime linker, ld.so. At runtime, dynamic libraries in the listed paths are scanned to satisfy any unresolved references. Why You would want to use it--Use this option to let your users run your executables without a special path option to find your dynamic libraries.

LD_RUN_PATH and -R
For f90, -R and the environment variable LD_RUN_PATH are not identical (for the runtime linker ld.so, they are).

28

Fortran 90 User 's Guide


3
If you build a.out with:

· · ·
­s ­stackvar

-R, then only the paths of -R are put in a.out. So -R is raw: it inserts only the paths you name, and no others.

LD_RUN_PATH, then the paths of LD_RUN_PATH are put in a.out, plus paths for Fortran 90 libraries. So LD_RUN_PATH is augmented: it inserts the
ones you name, plus various others. Both LD_RUN_PATH and -R, then only the paths of -R are put in a.out, and those of LD_RUN_PATH are ignored.

Strip the executable file of its symbol table (makes debugging impossible). Stack the local variables to allow better optimizing with parallelizing. Use the stack to allocate all local variables and arrays in a routine unless otherwise specified. This makes them automatic, rather than static. Purpose: More freedom to optimizer for parallelizing a CALL in a loop. Parallel CALL: This option gives more freedom to the optimizer for such tasks as parallelizing a loop that includes a CALL. Definition: Variables and arrays are local unless they are:

· · ·

Arguments in a SUBROUTINE or FUNCTION statement (already on stack) Global items in a COMMON or SAVE, or STATIC statement Initialized items in a type statement or a DATA statement, such as: REAL :: X=8.0 or DATA X /8.0/

Segmentation Fault and -stackvar: You can get a segmentation fault using -stackvar with large arrays. Putting large arrays onto the stack can overflow the stack, so you may need to increase the stack size. There are two stacks.

· ·

The whole program has a main stack. Each thread of a multi-threaded program has a thread stack.

The default stack size is about 8 MBytes for the main stack and 256 KBytes for each thread stack. The limit command (no parameters) shows the current main stack size. If you get a segmentation fault using -stackvar, you might try doubling the main stack size at least once.

Using the Compiler

29


3
Example: Stack size--show the current main stack size.
demo$ limit cputime filesize datasize stacksize coredumpsize descriptors memorysize demo$ s unlimited unlimited 523256 kbytes 8192 kbytes unlimited 64 unlimited

The main stack size



Example: Set the main stack size to 64 MBytes.
demo% limit stacksize 65536

Example: Set each thread stack size to 8 MBytes.
demo% setenv STACKSIZE 8192

See csh(1) for details on the limit command.

­S

Assembly source--generate only assembly source code. Compile the named programs and leave the assembly-language output on corresponding files suffixed with .s (no .o file is created).

­temp=dir

Temporary files--set directory to locate temporary files. Set directory for temporary files used by f90 to be dir. No space is allowed within this option string. Without this option, they go into /tmp/.

­time ­v

Time for execution--display for each compilation pass. Verbose--print name of each compiler pass. Print the name of each pass as the compiler executes, plus display in detail the options and environment variables used by the driver.

­V

Version ID--show version ID. Print the name and version ID of each pass as the compiler executes.

30

Fortran 90 User 's Guide


3
-w[n]
Warnings--suppress warnings. The n can be any one of 0, 1, 2, 3, or 4.

· · ·
-xlibmopt

-w0 suppresses the least warnings. -w4 suppresses the most warnings. -w with no n, is the same as -w0.

Use selected math routines optimized for performance. This usually generates faster code. It may produce slightly different results; if so, they usually differ in the last bit. The order on the command line for this library option is not significant.

-xlicinfo

License information--display license server user ids. Return license information about the licensing system. In particular, return the name of the license server and the user ID for each of the users who have licenses checked out. Generally, with this option no compilation is done and a license is not checked out; and generally this option is used with no other options. However, if a conflicting option is used, then the last one on the command line wins and there is a warning. Example: Report license information, do not compile (order counts).
demo$ f90 -c -xlicinfo any.f90

Example: Do not report license information, do compile (order counts).
demo$ f90 -xlicinfo -c any.f90

-xnolib -xnolibmopt

Synonym for -nolib. Reset -fast so that it does not use -xlibmopt. Reset -fast so that it does not use the library of selected math routines optimized for performance. Use this after the -fast option: f90 -fast -xnolibmopt ...

Using the Compiler

31


3
-xO[n] -xpg -xtime -Xlist
Synonym for -O[n]. Synonym for -pg. Synonym for -time. Global program checking--check across routines (arguments, commons, ...). This helps find a variety of bugs by checking across routines for consistency in arguments, common blocks, parameters, and so forth. In general, -Xlist also makes a line-numbered listing of the source, and a cross reference table of the identifiers. The errors found do not necessarily prevent the program from being compiled and linked.
Table 3-4 Option -Xlist (no suboption) -XlistE -Xlisterr -Xlisterr[nnn] -XlistI -XlistL -Xlistln -Xlisto name -Xlistwar -Xlistwar[nnn] -XlistX Summary of -Xlist Suboptions Action Errors, listing, and cross reference table. Errors. Suppress all error messages in the verification report. Suppress error nnn in the verification report. Include files. Listing (and errors). Page length is n lines. Rename the -Xlist output report file. Suppress all warning messages in the report. Suppress warning nnn in the report. Cross reference table (and errors).

For more information, see "Details of -Xlist Suboptions" on page 75.

32

Fortran 90 User 's Guide


3
3.3 Miscellaneous Tips
Floating-Point Hardware Type
Some compiler options are specific to particular hardware options. The utility fpversion tells which floating-point hardware is installed. The utility fpversion(1) takes 30 to 60 wall clock seconds before it returns, since it dynamically calculates hardware clock rates of the CPU and FPU. See fpversion(1). Also read the Numerical Computation Guide for details.

Many Options on Short Commands
Some users type long command lines--with many options. To avoid this, make a special alias or use environment variables.

Alias Method
Example: Define f90f.
demo$ alias f90f "f90 -fast -O4"

Example: Use f90f.
demo$ f90f any.f90

The above command is equivalent to "f90 -fast -O4 any.f90".

Environment Variable Method
Some users shorten command lines by using environment variables. The FFLAGS or OPTIONS variables are special variables for FORTRAN.

· ·

If you set FFLAGS or OPTIONS, they can be used in the command line. If you are compiling with make files, FFLAGS is used automatically if the make file uses only the implicit compilation rules.

Example: Set FFLAGS.
demo$ setenv FFLAGS '-fast -O4'

Using the Compiler

33


3
·
Example: Use FFLAGS explicitly.
demo$ f90 $FFLAGS any.f90

The above command is equivalent to "f90 -fast -O4 any.f90".

·

Example: Let make use FFLAGS implicitly. If both: · The compile in a make file is implicit (no explicit f90 compile line) · The FFLAGS variable is set as above Then invoking the make file results in a compile command equivalent to "f90 -fast -O4 any.f90".

34

Fortran 90 User 's Guide


File System and File I/O
This chapter is organized into the following sections.
Summary Directories File Names Path Names Redirection Piping Accessing Files from Fortran Programs Direct I/O Internal Files page 35 page 37 page 37 page 37 page 40 page 41 page 45 page 45 page 46

4

This chapter is a basic introduction to the file system and how it relates to the Fortran I/O system. If you understand these concepts, then skip this chapter.

4.1 Summary
The basic file system consists of a hierarchical file structure, established rules for file names and path names, and various commands for moving around in the file system, showing your current location in the file system, and making, deleting, or moving files or directories.

35


4
The system file structure of the UNIX operating system is analogous to an upside-down tree. The top of the file system is the root directory. Directories, subdirectories, and files all branch down from the root. Directories and subdirectories are considered nodes on the directory tree, and can have subdirectories or ordinary files branching down from them. The only directory that is not a subdirectory is the root directory, so except for this instance, you do not usually make a distinction between directories and subdirectories. A sequence of branching directory names and a file name in the file system tree describes a path. Files are at the ends of paths, and cannot have anything branching from them. When moving around in the file system, down means away from the root and up means toward the root. The figure below shows a diagram of a file system tree structure.

root directory

file

subdirectory

subdirectory

file

subdirectory

file

file

file

Figure 4-1

File System Hierarchy

36

Fortran 90 User 's Guide


4
4.2 Directories
All files branch from directories except the root directory. Directories are just files with special properties. While you are logged on, you are said to be in a directory. When you first log on, you are usually in your home directory. At any time, wherever you are, the directory you are in is called your current working directory. It is often useful to list your current working directory. The pwd command prints the current working directory name and the getcwd routine gets (returns) the current working directory name. You can change your current working directory simply by moving to another directory. The cd shell command and the chdir routine change the current working directory to a different directory.

4.3 File Names
All files have names, and you can use almost any character in a file name. The name can be up to 1024 characters long, but individual components can be only 512 characters long. However, to prevent the shell from misinterpreting certain special punctuation characters, restrict your use of punctuation in file names to the dot (.), underscore ( _ ), comma (,), plus (+), and minus (-). The slash ( / ) character has a specific meaning in a file name, and is only used to separate components of the path name (as described in the following section). Also, avoid using blanks in file names. Directories are just files with special properties and follow the same naming rules as files. The only exception is the root directory, named slash ( / ).

4.4 Path Names
To describe a file anywhere in the file system, you can list the sequence of names for the directory, subdirectory, and so forth, and file, separated by slash characters, down to the file you want to describe. If you show all the directories, starting at the root, that's called an absolute path name. If you show only the directories below the current directory, that's called a relative path name.

File System and File I/O

37


4
Relative Path Names
From anywhere in the directory structure, you can describe a relative path name of a file. Relative path names start with the directory you are in (the current directory) instead of the root. For example, if you are in the directory /usr/you, and you use the relative path name mail/record, that is equivalent to using the absolute path name /usr/you/mail/record. This is illustrated in the diagram below:
/usr/you

mail

record

Figure 4-2

Relative Path Name

Absolute Path Names
A list of directories and a file name, separated by slash characters, from the root to the file you want to describe, is called an absolute path name. It is also called the complete file specification or the complete path name. A complete file specification has the general form:
/directory/directory/.../directory/file

There can be any number of directory names between the root (/) and the file at the end of the path as long as the total number of characters in a given path name is less than or equal to 1024.

38

Fortran 90 User 's Guide


4
An absolute path name is illustrated in the diagram below: /usr/you/mail/record /

usr

you

mail

record

Figure 4-3

Absolute Path Name

File System and File I/O

39


4
4.5 Redirection
Redirection is a way of changing the files that a program uses without passing a file name to the program. Both input to and output from a program can be redirected. The symbol for redirecting standard input is the `<' sign, and for standard output is the ">" sign. File redirection is a function performed by the command interpreter or shell when a program is invoked by it.

Input
The shell command line:
demo$ myprog < mydata

The above command causes the file mydata (which must already exist) to be connected to the standard input of the program myprog when it is run. This means that if myprog is a Fortran 90 program and reads from unit 5, it reads from the mydata file.

Output/Truncate
The shell command line:
demo$ myprog > myoutput

The above command causes the file myoutput (which is created if it does not exist, or rewound and truncated if it does) to be connected to the standard output of the program myprog when it is run. So if the Fortran 90 program myprog writes to unit 6, it writes to the file myoutput.

Output/Append
The shell command line:
demo$ myprog >> myoutput

The above command causes the file myoutput (which must exist) to be connected for appending. So if the Fortran 90 program myprog writes to unit 6, it writes to the file myoutput but after wherever the file ended before.

40

Fortran 90 User 's Guide


4
Both standard input and standard output may be redirected to and from different files on the same command line. Standard error may also be redirected so it does not appear on your workstation display. In general, this is not a good idea, since you usually want to get error messages from the program immediately, rather than sending them to a file. The shell syntax to redirect standard error varies, depending on whether you are using sh or csh. Example: csh. Redirecting standard error and standard output.
demo% myprog1 |& myprog2

Example: sh. Redirecting standard error and standard output.
demo$ myprog1 2>&1 | myprog2

In each shell, the above command runs the program myprog1 and redirects the standard output and standard error to the program myprog2.

4.6 Piping
You can connect the standard output of one program directly to the standard input of another without using an intervening temporary file. The mechanism to accomplish this is called a pipe. Example: A shell command line using a pipe.
demo$ firstprog | secondprog

This causes the standard output (unit 6) of firstprog to be piped to the standard input (unit 5) of secondprog. Piping and file redirection can be combined in the same command line. Example: myprog reads mydata and pipes output to wc, wc writes datacnt.
demo$ myprog < mydata | wc > datacnt

The program myprog takes its standard input from the file mydata, and has its standard output piped into the standard input of the wc command, the standard output of wc is redirected into the file datacnt.

File System and File I/O

41


4
4.7 Accessing Files from Fortran Programs
Data are transferred to or from devices or files by specifying a logical unit number in an I/O statement. Logical unit numbers can be nonnegative integers or the character "*". The "*" stands for the standard input if it appears in a READ statement, or the standard output if it appears in a WRITE or PRINT statement. Standard input and standard output are explained in the section on preconnected units found later in this chapter.

Accessing Named Files
Before a program can access a file with a READ, WRITE, or PRINT statement, the file must be created and a connection established for communication between the program and the file. The file can already exist or be created at the time the program executes. The Fortran 90 OPEN statement establishes a connection between the program and file to be accessed. (For a description of OPEN, read the Fortran 90 Handbook.) File names can be simple expressions, as listed below.

· · ·

Quoted character constants
File = 'myfile.out'

Character variables
File = Filnam

Character expressions
File = LEN_TRIM(Prefix) // '/' // LEN_TRIM(Name)

A program can read file names from a file or terminal keyboard.
READ( 4, 401 ) Filnam

42

Fortran 90 User 's Guide


4
Accessing Unnamed Files
When a program opens a Fortran 90 file without a name, the runtime system supplies a file name. There are several ways it can do this.

Opened as Scratch
If you specify STATUS='SCRATCH' in the OPEN statement, then the system 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 STATUS='KEEP' is specified in the CLOSE statement.

Already Open
If a Fortran 90 program has a file already open, an OPEN statement that specifies only the file's logical unit number and the parameters to change can be used to change some of the file's parameters (specifically, BLANK and FORM). The system determines that it must not really OPEN a new file, but just change the parameter values. Thus, this looks like a case where the runtime system would make up a name, but it does not.

Other
In all other cases, the runtime system OPENs a file with a name of the form fort.n, where n is the logical unit number given in the OPEN statement.

Passing File Names to Programs
The file system does not have any notion of temporary file name binding (or file equating) as some other systems do. File name binding is the facility that is often used to associate a Fortran logical unit number with a physical file without changing the program. This mechanism evolved to communicate file names more easily to the running program, because in FORTRAN 66 there was no way to open files by name.

File System and File I/O

43


4
With this operating system the following ways communicate file names to a Fortran program.

·

Redirection and piping. Redirection and piping can change the names of program input and output files without changing the program. See the sections "Redirection" and "Piping" earlier in this chapter.

Preconnected Units
When a Fortran program begins execution under this operating system, there are usually three units already open. They are preconnected units. Their names are standard input, standard output, and standard error. In Fortran, the following are preconnected.

· · ·

Standard input is logical unit 5 Standard output is logical unit 6 Standard error is logical unit 0

All three are connected, unless file redirection or piping is done.

Other Units
All other units are preconnected to files named fort.n where n is the corresponding unit number, and can be 0, 1, 2, ..., with 0, 5, and 6 having the usual special meanings. These files need not exist, and are created only if the units are actually used, and if the first action to the unit is a WRITE or PRINT; that is, only if an OPEN statement does not override the preconnected name before any WRITE or PRINT is issued for that unit. Example: Preconnected Files. The program OtherUnit.f90.
WRITE( 25, '(I4)' ) 2 END

The above program preconnects the file fort.25 and writes a single formatted record onto that file.
demo$ f90 OtherUnit.f90 demo$ a.out demo$ cat fort.25 2 demo$

44

Fortran 90 User 's Guide


4
4.8 Direct I/O
Random access to files is also called direct access. A direct-access file contains a number of records that are written to or read from by referring to the record number. This record number is specified when the record is written. In a directaccess file, records must be all the same length and all the same type. A logical record in a direct access, external file is a string of bytes of a length specified when the file is opened. Read and write statements must not specify logical records longer than the original record size definition. Shorter logical 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. In using direct unformatted I/O, be careful with the number of values your program expects to read. Each READ operation acts on exactly one record; the number of values that the input list requires must be less than or equal to the number of values in that record. Direct access READ and WRITE statements have an extra argument, REC=n, which gives the record number to be read or written. Example: Direct-access, unformatted.
OPEN( 2, FILE='data.db', ACCESS='DIRECT', RECL=20, & FORM='UNFORMATTED', ERR=90 ) READ( 2, REC=13, ERR=30 ) x, y

This opens a file for direct-access, unformatted I/O, with a record length of 20 characters, then reads the thirteenth record as is. Example: Open, direct-access, formatted.
OPEN( 2, FILE='inven.db', ACCESS='DIRECT', RECL=20, & FORM='FORMATTED', ERR=90 ) READ( 2, FMT="(I10,F10.3)", REC=13, ERR=30 ) a, b

This opens a file for direct-access, formatted I/O, with a record length of 20 characters, then reads the thirteenth record and converts it according to the format "(I10,F10.3)".

File System and File I/O

45


4
4.9 Internal Files
An internal file is a variable of type default character. This means that an internal file can be one of the following:

· · · · · ·

Scalar Array Element of an array Section of an array Component of a structure Substring

To use an internal file, give the name or the designator of the character variable in place of the unit number. This is called I/O, because you use READ and WRITE statements to deal with internal files, although I/O is not a precise term to use here. f90 extends what can be an internal file: if you are reading from an internal file, the internal file can be a literal constant character string.

Rules and Restrictions

·

The variable must not be an array section with a vector subscript. · For a constant there is a single record in the file. o · For a variable or substring, there is a single record in the file. · For an array, each array element is a record. · Each sequential READ or WRITE starts at the beginning of an internal file.

46

Fortran 90 User 's Guide


4
Example: Scalar for internal file, sequential formatted read.
demo$ cat intern1.f90 CHARACTER x*80 WRITE(*,*) 'Enter two numbers' READ( *, '(A)' ) x ! Reads a character string from standard input to x READ( x, '(I3,I4)' ) n1, n2 ! Reads the internal file x WRITE( *, * ) n1, n2 END demo$ f90 intern1 demo$ a.out Enter two numbers 12 99 12 99 demo$ s

Example: Array for internal file, sequential formatted read.
demo$ cat intern3.f90 CHARACTER *16, Line(4) DATA Line / ' 81 81 ', ' 82 82 ', ' 83 83 ', ' 84 84 ' / READ( Line, '(2I4)' ) i, j, k, l, m, n ! Reads internal file Line PRINT *, i, j, k, l, m, n END demo$ f90 intern3 demo$ a.out 81 81 82 82 83 83 demo$ s

File System and File I/O

47


4

48

Fortran 90 User 's Guide


Program Development
This chapter is organized into the following sections.
Simple Program Builds Program Builds with the make Program Tracking and Controlling Changes with SCCS page 49 page 50 page 52

5

5.1 Simple Program Builds
For a program that depends on a few source files and some system libraries, you can easily compile all of the source files every time you change the program. Even in this simple case, the f90 command can involve much typing, and with options or libraries, a lot to remember. A script or alias can help.

Writing a Script
A shell script can save typing. For example, to compile a small program that is in the file example.f90, and that uses the Xview support library, you can save a one-line shell script onto a file, here called fex, that looks like this.
f90 example.f90 ­lFxview ­o example

You may need to put execution permissions on fex.
demo$ chmod +x fex

49


5
Creating an Alias
You can create an alias to do the same command.
demo$ alias fex "f90 example.f90 -lFxview ­o example"

Using a Script or Alias
Either way, to recompile example.f90, you type only fex.
demo$ fex

Limitations
With multiple source files, forgetting one compile makes the objects inconsistent with the source. Recompiling all files after every editing session wastes time, since not every source file needs recompiling. Forgetting an option or a library produces questionable executables.

5.2 Program Builds with the make Program
The make program recompiles only what needs recompiling, and it uses only the options and libraries you want. This section shows you how to use normal, basic make, and it provides a simple example. For a summary, see make (1).

The make File
The way you tell make what files depend on other files, and what processes to apply to which files, is to put this information into a file called the make file, in the directory where you are developing the program. Example: A program of four source files and a make file.
demo$ ls makefile commonblock computepts.f90 pattern.f90 startupcore.f90 demo$ s

50

Fortran 90 User 's Guide


5
Assume both pattern.f90 and computepts.f90 do an include of commonblock, and you wish to compile each .f90 file and link the three relocatable files (plus a series of libraries) into a program called pattern. The make file for this example is listed below.
demo$ cat makefile pattern: pattern.o computepts.o startupcore.o f90 pattern.o computepts.o startupcore.o ­Fxview ­o pattern pattern.o: pattern.f90 commonblock f90 ­c pattern.f90 computepts.o: computepts.f90 commonblock f90 ­c computepts.f90 startupcore.o: startupcore.f90 f90 ­c startupcore.f90 demo$ s

The first line of this make file says:

· ·

make pattern pattern depends on pattern.o, computepts.o, and startupcore.o

The second line is the command for making pattern. The third line is a continuation of the second (because it starts with a tab). There are four such paragraphs or entries in this make file. The structure of these entries is:

· ·

Dependencies -- Each entry starts with a line that names the file to make, and names all the files it depends on. Commands -- Each entry has one or more subsequent lines that contain Bourne shell commands, and that tell how to build the target file for this entry. These subsequent lines must each be indented by a tab.

Using make
The make command can be invoked with no arguments, such as this.
demo$ make

The make utility looks for a file named makefile or Makefile in the current directory, and takes its instructions from there.

Program Development

51


5
The make utility general actions are:

· · ·

From the make file, it gets all the target files it must make, and what files they depend on. It also gets the commands used to make the target files. It gets the date and time changed for each file. If any target file is not up to date with the files it depends on, then that target is rebuilt, using the commands from the make file for that target.

5.3 Tracking and Controlling Changes with SCCS
SCCS is Source Code Control System. It provides a way to:

· · ·

Keep track of the evolution of a source file (change history) Prevent different programmers from changing the same source file at the same time Keep track of the version number by providing version stamps

The basic three operations of SCCS are putting files under SCCS control, checking out a file for editing, and checking in a file. This section shows you how to use SCCS to do these things and provides a simple example, using the previous program. It describes normal, basic SCCS, and introduces only three SCCS commands: create, edit, and delget.

Putting Files under SCCS
Putting files under SCCS control involves making the SCCS directory, inserting SCCS ID keywords into the files (optional), and 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.
demo$ mkdir SCCS demo$ s

The `SCCS' must be uppercase.

52

Fortran 90 User 's Guide


5
Inserting SCCS ID Keywords
Some people put one or more SCCS ID keywords into each file, but that is optional. These will later be filled in with a version number each time the file is checked in with a get or delget SCCS command. There are three likely places to put such strings:

· · ·

Comment lines Parameter statements Initialized data

The advantage of the last is that the version information appears in the compiled object program, and can be printed using the what command. Included header files containing 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. Finally, in the case of some files, like ASCII data files or make files, the source is all there is, so the SCCS information can go in comments, if anywhere. Identify the make file with a make comment containing the keywords.
# %Z%%M% %I% %E%

The source files startupcore.f90 and computepts.f90 and pattern.f90 can be identified to SCCS by initialized data of the form.
CHARACTER (LEN=50) :: sccsid = "%Z%%M% %I% %E%\n"

Creating SCCS Files
Example: Put files under control of SCCS with an SCCS create command.
demo$ sccs create makefile commonblock startupcore.f90 \ computepts.f90 pattern.f90 demo$ s

Program Development

53


5
The make file looks like this after SCCS keyword expansion.
# @(#)makefile1.184/03/01 OBJ = pattern.o computepts.o startupcore.o pattern: $(OBJ) f90 $(OBJ) ­Fxview ­o pattern pattern.o: pattern.f90 commonblock computepts.o: computepts.f90 commonblock startupcore.o: startupcore.f90

Checking Files Out and In
Out-- 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 are done editing. A file is checked out using the SCCS edit command. Example: Check out a file using SCCS.
demo$ sccs edit computepts.f90 demo$s

In this example, SCCS makes a writable copy of computepts.f90 in the current directory, and records your login name. Other users cannot check it out while you have it checked out, but they can find who checked out which files. In-- Check in the file with the sccs delget command when you have completed your current editing. Example: Check in a file using SCCS.
demo$ sccs delget computepts.f90 demo$ s

This causes the SCCS system to do the following: 1. Make sure that you are the user who checked it out (compares login names). 2. Solicit a descriptive comment from the user on the changes. 3. Make a record of what was changed in this editing session. 4. Delete the writable copy of computepts.f90 from the current directory. 5. Replace it by a read-only copy with the SCCS keywords expanded.

54

Fortran 90 User 's Guide


5
The SCCS command delget is a composite of the two simpler SCCS commands, delta and get. The delta command does the first three items in the list above and the get command does the fourth and fifth.

Program Development

55


5

56

Fortran 90 User 's Guide


Libraries
This chapter is organized into the following sections.
Libraries in General Static Libraries Dynamic Libraries Consistent Compile and Link Library Paths page 57 page 58 page 61 page 65 page 65

6

See ld(1) for more details.

6.1 Libraries in General
A library can be a collection of subprograms. Each member of this collection is called a library element or module. A relocatable library is one whose elements are relocatable (.o) files. These object modules are inserted into the executable file by the linker during the compile/link sequence. Some examples of static libraries on the system are:

· · ·

Fortran 90 library: libf90.a Math library: libm.a C library: libc.a

57


6
Some examples of shared dynamic libraries on the system are:

· · Load Map

FORTRAN 77 library: libF77.so C library: libc.so

To display a load map, pass the load map option to the linker by -Qoption. This displays which libraries are linked and which routines are obtained from which libraries during the creation of the executable module. This is a very simple load map. Example: -m for load map.
demo$ f90 -Qoption ld -m any.f90

Advantages of Libraries
Relocatable libraries provide an easy way for commonly used subroutines to be used by several programs. The programmer need only name the library when linking the program, and those library modules that resolve references in the program are linked--copied into the executable file. This has two advantages.

· ·

Only the needed modules are loaded (at least, for static libraries). The programmer need not change the link command line as subroutine calls are added and removed during program development.

6.2 Static Libraries
Static libraries are built from object files (.o files) using the program ar.

Disadvantages of Libraries
When the linker searches a library, it extracts elements whose entry points are referenced in other parts of the program it is linking, such as subprogram or entry names or names of COMMON blocks initialized in BLOCKDATA subprograms. The nature of the elements and the nature of the search leads to some disadvantages.

58

Fortran 90 User 's Guide


6
·
The whole thing -- For static libraries, when the linker extracts a library element, it extracts the whole thing (not so for dynamic libraries). A library element corresponds to the result of a compilation, so routines that are compiled together are always linked together. This is a difference between this operating system and some other systems, and it may affect whether you chunk compilation files to many small files for your libraries. Order matters -- In linking static libraries, order really matters. The linker processes its input files in the order that they appear on the command line (that is, left to right). When the linker decides whether or not a library element is to be linked, its decision is based only on the relocatable modules it has already processed. You can use lorder and tsort to order static libraries. Example: Order matters. If a Fortran 90 program is in two files, main.f90 and graf.f90, and only the latter accesses the Xview library, it is an error to reference that library before graf.f90 or graf.o:
(Wrong) demo$f90 main.f ­lFxview graf.f90 ­o myprog (Right) demo$f90 main.f graf.f90 ­lFxview ­o myprog

·

Sample Creation of a Static Library
Base--source of four routines (for example on creating static library).
demo$ cat one.f SUBROUTINE REAL a, r r = a * 2.0 END SUBROUTINE REAL a, r r = a / 2.0 END SUBROUTINE REAL a, r r = a * 3.0 END SUBROUTINE REAL a, r r = a / 3.0 END demo$ s twice ( a, r )

half ( a, r )

thrice ( a, r )

third ( a, r )

Libraries

59


6
Main program to use one of the routines (for creating static library).
demo$ cat teslib.f READ(*,*) x CALL twice( x, z ) WRITE(*,*) z END demo$ s

·

Split the file, using fsplit, so there is one subroutine per file.
demo$ fsplit one.f demo$ s



fsplit assumes fixed form. It may not work on all f90 source.

·

Compile each with -c so it will compile only, and leave the .o object files.
demo$ demo$ demo$ demo$ demo$ f90 f90 f90 f90 s -c -c -c -c half.f third.f thrice.f twice.f

·
Create the library.

Use ar to create static library faclib.a from the four object files.
demo$ ar cr faclib.a half.o third.o thrice.o twice.o

Alternate: Specify the order using lorder and tsort.
demo$ ar cr faclib.a 'lorder half.third.o thrice.o twice.o | tsort'

·

Compile the main, using the new static library.
demo$ f90 teslib.f90 faclib.a demo$ s

·
Note that twice is here. Note that half, third, and thrice are not here (good).

Use nm to list the name of each object in a.out built from static library.
demo$ demo$ nm a.out | [260]| 77832| demo$ nm a.out | grep demo$ nm a.out | grep demo$ nm a.out | grep demo$ s grep twice 72|FUNC |GLOB |0 half third thrice |8 |twice_

60

Fortran 90 User 's Guide


6
·
Run a.out
demo$ a.out 6 12.0000 demo$ s

Sample Replacement in a Static Library
If you recompile an element of a static library (usually because you've changed the source), replace it in its library by running ar again. A library need not be specially flagged for the linker; the linker recognizes a library when it encounters one. Example: Recompile, replace. Give ar the r option; use cr only for creating.
demo$ f90 -c half.f demo$ ar r faclib.a half.o demo$ s

6.3 Dynamic Libraries
A dynamic library has the following features.

· · ·

It is a collection of object modules such that each is already in executable file format (the a.out format) but the collection has no main entry. The object modules are not bound into the executable file by the linker during the compile/link sequence; such binding is deferred until runtime. If you change a module of a shared library, then whenever any application using it starts to execute, it will get the changed version. The ability to modify and improve libraries independent of the executables that use it can be a significant advantage in maintaining programs.

If you have a a -pic option to use to generate position-independent code, then from the generated dynamic library, a module can be used by many executing programs without duplicating it in them all. In this release, f90 has no -pic to generate position-independent code, so the library is not truly shared. You still save disk space for storing the executable, and you can still get library changes into the executable without rebinding the executable.

Libraries

61


6
Performance Issues
The usual trade-off between space and time applies.

· ·

Less space In general, deferring the binding of the library module uses less disk space More time It · · · takes a little more CPU time to do the following: Load the library during runtime. Do the link editing operations. Execute the library position-independent code.

·

Programs Vary Because of these various performance issues, some programs are slower if they use nonshared libraries, and some if they use shared libraries. You might bind each way to tell whether one method is significantly better for your program.

Binding Options
You can specify the binding option when you compile, that is, dynamic or static libraries. These options are actually linker options, but they are recognized by the compiler and passed on to the linker.

-d[y,n]: Allow or disallow dynamic libraries for the entire executable

· ·

-dy: Yes--allow dynamically bound libraries (allow shared libraries). -dn: No--do not allow dynamically bound libraries (no shared libraries). These apply to the whole executable. Use only once on the command line. The default is y.

-B[dynamic,static]: Bind as dynamic (or static) libraries listed later
This applies to any library listed later in the command. The default is dynamic.

· ·

­Bdynamic: Prefer dynamic binding (try for shared libraries). -Bstatic: Require static binding (no shared libraries).

62

Fortran 90 User 's Guide


6
If you provide a library for your customers, then providing both a dynamic and a static version allows the customers the flexibility of binding whichever way is best for their application. For example, if the customer is doing some benchmarks, the ­dn option reduces one element of variability.

A Simple Dynamic Shared Library
You can build a shared library from the relocatable object (.o) files using the ld command. Be careful to avoid any need for reentrant code, since this release does not provide a way to guarantee that code is position independent.

Sample Create
Example: Create a dynamic shared library. Start with the same files used for the static library example: half.f90, third.f90, thrice.f90, twice.f90. Compile.
demo$ f90 -c demo$ s *.f90

Example: Create a dynamic shared library. Link, and specify the .so file, and the -h to get a version number.
demo$ ld -o libfac.so.1 -dy -G -h libfac.so.1 *.o demo$ s

The -G tells the linker to build a shared library. Note that there is no "-z text" because that would generate warnings. The "-z text" warns you if it finds anything other than position-independent code, such as relocatable text. It does not warn you if it finds writable data. Bind: Make the executable file a.out.
demo$ f90 teslib.f90 libfac.so.1 demo$ s

Libraries

63


6
Run.
demo$ a.out 6 12.0000 demo$ s

Inspect the a.out file for use of shared libraries. The file command shows that a.out is a dynamically linked executable -- programs that use shared libraries are completely link-edited during execution.
demo$ file a.out a.out: ELF 32-bit MSB executable SPARC Version 1 dynamically linked, not stripped demo$ s

The ldd command shows that a.out uses some shared libraries, including libfac.so.1 and libc (included by default by f90). It also shows exactly which files on the system will be used for these libraries.
demo$ ldd a.out libfac.so.1 => ./libfac.so.1 libF77.so.2 => /opt/SUNWspro/lib/libF90.so.2 libc.so.1 => /usr/lib/libc.so.1 libucb.so.1 => /usr/ucblib/libucb.so.1 libresolv.so.1 => /usr/lib/libresolv.so.1 libsocket.so.1 => /usr/lib/libsocket.so.1 libnsl.so.1 => /usr/lib/libnsl.so.1 libelf.so.1 => /usr/lib/libelf.so.1 libdl.so.1 => /usr/lib/libdl.so.1 libaio.so.1 => /usr/lib/libaio.so.1 libintl.so.1 => /usr/lib/libintl.so.1 libw.so.1 => /usr/lib/libw.so.1 demo$ s

Your path may vary.

64

Fortran 90 User 's Guide


6
6.4 Consistent Compile and Link
Be consistent with compiling and linking. Do not build libraries with inconsistent modules. Inconsistent compilation and linkage is not supported.

·

For compiling and linking as separate steps (separate commands), if you compile any module with -g, then be sure to link with the same option. Example. Compile sbr.f90 with -g and smain.f90 without it.
demo$ f90 -c -g sbr.f90 demo$ f90 -g sbr.o smain.f90



Pass the -g to the linker.

The above sequence is equivalent to the following commands.
demo$ f90 -c -g sbr.f90 demo$ f90 -c smain.f90 demo$ f90 -g sbr.o smain.o

·

If you compile any module under a major release of the operating system, then compile all modules of that program with the same major release.

6.5 Library Paths
The linker searches for libraries in several locations and it searches in certain prescribed orders. You can make some changes to the order and locations.

Installation Directory
Some library locations depend on the installation directory. These locations are described here in terms of a path called BasDir, defined as follows:
Installation Location Standard Nonstandard, for example, /my/dir/

BasDir
/opt/SUNWspro/lib/ /my/dir/

Libraries

65


6
Building Executables: ld Search order
During the build of the executable, ld searches for libraries in the following locations, in order:
Path /BasDir/lib/ /BasDir/SC3.0.1/lib/ /opt/SUNWspro/lib/ /usr/ccs/lib/ /usr/lib/ Comment Sun shared libraries here Sun libraries, shared or static, here Standard location for Sun libraries

The linker also searches in any directories specified in LD_LIBRARY_PATH or by the -Ldir option.

Running Executables: ld Search order
During the run of the executable, ld searches for shared libraries in the following locations, in order:
Path /BasDir/lib/ /opt/SUNWspro/lib Directories built in by -R or LD_RUN_PATH when executable was made /usr/ccs/lib/ /usr/lib/ Comment Built in by driver, unless -norunpath Built in by driver, unless -norunpath

The linker also searches in directories specified in LD_LIBRARY_PATH.

Summary of Environment Variables for Paths

· ·

LD_RUN_PATH: Value matters only when executable is created. LD_LIBRARY_PATH: Value matters when executable is created or run.

66

Fortran 90 User 's Guide


6
Build Paths and Run Paths
When you run the executable file, the runtime linker libraries again. The linker searches in any directories LD_LIBRARY_PATH environment variable, and that after the executable file has been created. Therefore it libraries were when the executable was created. locates the shared specified in the variable can change even doesn't matter where the

Finding Built-in Paths
Use dump to check which paths were built in when the executable was created. Example: Find which directories were built in to a.out.

demo$dump -Lv a.out | grep RPATH

Libraries

67


6

68

Fortran 90 User 's Guide


Debugging
This chapter is organized into the following sections.
Global Program Checking (-Xlist) The dbx Debugger page 69 page 77

7

7.1 Global Program Checking (-Xlist)
Purpose--Checking across routines helps find various kinds of bugs. With -Xlist, f90 reports errors of alignment, agreement in number and type for arguments, common blocks, parameters, plus many other kinds of errors (details follow). It also makes a listing and a cross reference table; combinations and variations of these are available using suboptions. An example follows. Example: Errors only--Use -XlistE to show errors only.
-XlistE Form of output varies. demo$ f90 -XlistE Repeat.f90 demo$ cat Repeat.lst FILE "Repeat.f90" ... demo$ s

69


7
Errors in General
Global program checking can do the following:

· · · · Details

Enforce type checking rules of Fortran 90 more stringently than usual, especially between separately compiled routines. Enforce some portability restrictions needed to move programs between different machines and/or operating systems. Detect legal constructions that are nevertheless wasteful or error-prone. Reveal other bugs and obscurities.

More particularly, global cross checking reports problems such as:

·

Interface problems · Checking number and type of dummy and actual arguments · Checking type of function values · Checking possible conflicts of incorrect usage of data types in common blocks of different subprograms Usage problems · Function used as a subroutine or subroutine used as a function · Declared but unused functions, subroutines, variables, and labels · Referenced but not declared functions, subroutines, variables, and labels · Usage of unset variables · Unreachable statements · Implicit type variables · Inconsistency of the named common block lengths, names, and layouts Syntax problems--syntax errors found in a Fortran 90 program Portability problems--codes that do not conform to ANSI Fortran 90, if the appropriate option is used

·

· ·

70

Fortran 90 User 's Guide


7
Using Global Program Checking
To cross check the named source files, use -Xlist on the command line. Example: Compile three files for global program checking.
demo$ f90 -Xlist any1.f90 any2.f90 any3.f90

In the above example, f90 does the following: · Saves the output in the file any1.lst · Compiles and links the program if there are no errors Example: Compile all Fortran 90 files for global program checking.
demo$ f90 -Xlist *.f90

Terminal Output
To display directly to the terminal, rename the output file to /dev/tty. Example: Display to terminal.
demo$ f90 -Xlisto /dev/tty any1.f90

See -Xlisto name, on page 76.

The Default Output Features
The simple -Xlist option (as shown in the example above) provides a combination of features available for output. That is, with no other -Xlist options on the f90 command line, the plain, simple -Xlist option provides the following:

· ·

The output file has the same name as the first file, with a .lst extension. The output content includes: · A line-numbered source listing (Default) · Error messages (embedded in listing) for inconsistencies across routines · Cross reference table of the identifiers (Default) · Pagination at 66 lines per page, 79 columns per line (Defaults) · No call graph (Default) · No expansion of include files (Default)

Debugging

71


7
Example: Using -Xlist--a program with inconsistencies between routines.
Repeat.f90 Form of output varies. demo$ cat Repeat.f90 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 EXTERNAL INTEGER PRINT *, END nwfrk( ix ) fork prnok, fork prnok ( ix ), fork ( )

INTEGER FUNCTION prnok ( x ) prnok = INT ( x ) + LOC(x) END SUBROUTINE unreach_sub() CALL sleep(1) END demo$ f90 -Xlist Repeat.f90 demo$ cat Repeat.lst

Compile with -Xlist. List the -Xlist output file.

72

Fortran 90 User 's Guide


7
Suboptions for Global Checking Across Routines
The standard global cross checking option is -Xlist (with no suboption). This shows the listing, errors, and cross reference table. For variations from this standard report, add one or more suboptions to the command line.

Suboption Syntax
Add suboptions according to the following rules:

· · ·

Append the suboption to -Xlist Put no space between the -Xlist and the suboption Put only one suboption per -Xlist

Combination Special and A La Carte Suboptions
Combine suboptions according to the following rules:

· · · ·

The combination special: -Xlist (listing, errors, and cross reference table) The a la carte options are: -XlistE, -XlistL, and -XlistX. All other options are detail options, not a la carte, not combination special. Once you start ordering a la carte, the three parts of the combination special are cancelled, and you get only what you specify. Example: Each of these two commands does the same thing.
demo$ f90 demo$ f90 -XlistL -XlistL -Xlist any.f90 any.f90

Debugging

73


7
Combination special or a la carte suboptions (with no other suboptions):
Table 7-1 Type/Amount of Output -Xlist Combination Special or A La Carte Suboptions Comment No suboptions Details page 71 Option

Errors, listing, cross reference table -Xlist Errors Errors and listing Errors and cross reference table -XlistE -XlistL -XlistX

By itself, does not show listing or cross reference table page 75 By itself, does not show cross reference table By itself, does not show listing page 76 page 76

Summary of -Xlist Suboptions
Table 7-2 Option -Xlist -XlistE -Xlisterr -Xlisterr[nnn] -XlistI -XlistL -Xlistln -Xlisto name -Xlistwar -Xlistwar[nnn] -XlistX (no suboption) -Xlist Suboptions Summary Details page 73 page 75 page 75 page 75 page 75 page 76 page 76 page 76 page 76 page 76 page 76

Action Errors, listing, and cross reference table. Errors. Suppress all error messages in the verification report. Suppress error nnn in the verification report. Include files. Listing (and errors). Page length is in n lines. Rename the -Xlist output report file. Suppress all warning messages in the report. Suppress warning nnn in the report. Cross reference table (and errors).

74

Fortran 90 User 's Guide


7
Details of -Xlist Suboptions
-Xlisterr -Xlisterr[nnn]
Suppress all error messages in the verification report. Suppress error nnn in the verification report. This is useful if you want a cross reference or a listing without the error messages. It is also useful if you do not consider certain practices to be real errors. To suppress more than one error, use this option repeatedly. Example: -Xlisterr338 suppresses error message 338. If nnn is not specified, then suppress all error messages. Global cross check errors. Show cross routine errors. This suboption by itself does not show a listing or a cross reference. Include files. List and cross check the include files. If -XlistI is the only -Xlist option/suboption used, then you get the standard -Xlist output of a line numbered listing, error messages, and a cross reference table--but include files are shown or scanned, as appropriate.

-XlistE

-XlistI

·

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 following are all listed: · Source files · INCLUDE files

·

Cross Reference Table If the cross reference table is not suppressed, then the following are all scanned while generating the cross reference table: · Source files · INCLUDE files

Default: No include files.

Debugging

75


7
-Xlistln
Page breaks. Set the page length for pagination to n lines. That is the letter ell for length, not the digit one. For example, -Xlistl45 sets the page length to 45 lines. Default: 66. No Page Breaks: The -Xlistl0 {that is a zero, not a letter oh} option shows listings and cross reference with no page breaks (easier for on-screen viewing).

-XlistL

Listing (and errors). Show cross check errors and listing. This suboption by itself does not show a cross reference. Default: Show listing, cross reference. Rename the -Xlist output report file. The space between o and name is required. Output is then to the name.lst file. To display directly to the terminal: -Xlisto /dev/tty

-Xlisto name

-Xlistwar -Xlistwar[nnn]

Suppress all warning messages in the report. Suppress warning nnn in the report. If nnn is not specified, then all warning messages will be suppressed from printing. To suppress more than one, but not all warnings, use this option repeatedly. For example, -Xlistwar338 suppresses warning message 338. Cross reference table (and errors). Show cross checking errors and cross reference. This suboption by itself does not show a listing. The cross reference table shows information about each identifier:

-XlistX

· · ·
-Xlistwarn Suppress specific warnings.

Is it an argument? Does it appear in a COMMON or EQUIVALENCE declaration? Is it set or used?

Example: Use -Xlistwarnnn to suppress two specific warnings.
demo$ f90 -Xlistwar338 demo$ cat Repeat.lst FILE "Repeat.f90" demo$ s -Xlistwar348 -XlistE Repeat.f90

76

Fortran 90 User 's Guide


7
7.2 The dbx Debugger
This section is organized as follows:
Sample Program for Debugging A Sample dbx Session Segmentation Fault--Finding the Line Number Exception--Finding the Line Number Trace of Calls Pointer to a Scalar Pointer to an Array User-Defined Types Pointer to User-Defined Type Allocated Arrays Print Arrays Print Array Slices Generic Functions Miscellaneous Tips Main Features of the Debugger Help (example) (example) (example) (example) (example) (example) (example) (example) (example) (example) (example) (example) (example) (example) page 78 page 79 page 81 page 83 page 84 page 85 page 86 page 87 page 89 page 91 page 92 page 93 page 94 page 96 page 96 page 97

This section introduces some dbx features likely to be used with Fortran. Use it as a quick start for debugging Fortran. Be sure to try the help feature. Note ­ Before you use the Debugger, you must install the appropriate Tools package--read Installing SunSoft Developer Products on Solaris for details. With dbx you can display and modify variables, set breakpoints, trace calls, and invoke procedures in the program being debugged without having to recompile. The Debugger program lets you make more effective use of dbx by replacing the original, terminal-oriented interface with a window- and mouse-based interface.

Debugging

77


7
Sample Program for Debugging
The following program, with bug, and consisting of files a1.f90, a2.f90, and a3.f90, is used in several examples of debugging. Example: Main for debugging.
a1.f90 PROGRAM TryDbx INTEGER, PARAMETER :: n=2 REAL, DIMENSION(n,n) :: twobytwo DATA (( twobytwo(k,j),k=1,n),j=1,n) / 4*-1 / CALL mkidentity( twobytwo, n ) PRINT *, determinant( twobytwo ) END

Example: Function for debugging.
a3.f90 REAL FUNCTION determinant ( a ) REAL a(2,2) determinant = a(1,1) * a(2,2) - a(1,2) / a(2,1) END

Example: Subroutine for debugging.
a2.f90 SUBROUTINE mkidentity ( array, m ) REAL, DIMENSION(m,m) :: array DO i = 1, m DO j = 1, m IF ( i .eq. j ) THEN array(i,j) = 1.0 ELSE array(i,j) = 0.0 END IF END DO END DO END

78

Fortran 90 User 's Guide


7
A Sample dbx Session
The following examples use the sample program above.

·

Compile--To use dbx or Debugger, compile and link with the -g flag. You can do this in one step or two, as shown in the examples below. Example Compile and link in one step, with -g.
demo$ f90 -o silly -g a1.f90 a2.f90 a3.f90

Example: Compile and link in separate steps.
demo$ f90 -c -g a1.f90 a2.f90 a3.f90 demo$ f90 -o silly a1.o a2.o a3.o

·

Start dbx--To start dbx, type dbx and the name of your executable file. Example: Start dbx on the executable named silly.
demo$ dbx silly Reading symbolic information... (dbx) s

·

Quit dbx--To quit dbx, enter the quit command. Example: Quit dbx.
(dbx)quit demo$ s {Skip this for now so you can do the next steps.}

·

Breakpoint--To set a breakpoint, at the dbx prompt; type "stop in subnam", where subnam names a subroutine or function, and subnam can be upper case or lower case. Example: A way to stop at the first executable statement in a main program.
(dbx) stop in main (2) stop in main (dbx) s

Debugging

79


7
·
Run Program--To run a program from within dbx, enter the run command. Example: Run a program from within dbx.
(dbx) run Running: silly (process id 8786) stopped in main at line 5 in file "a1.f90" 5 CALL mkidentity( twobytwo, n ) (dbx) s

When the breakpoint is reached, dbx displays a message showing where it stopped, in this case at line 5 of the a1.f90 file.

·

Print--To print a value, enter the print command. Example: Print the variable n. Note that dbx handles parameters.
(dbx) print n n=2 (dbx) s

Example: Print the matrix twobytwo (format may vary with release).
(dbx) print twobytwo twobytwo = (1,1) -1.0 (2,1) -1.0 (1,2) -1.0 (2,2) -1.0 (dbx) s

Example: Print the matrix array.
(dbx) dbx: dbx: (dbx) print array "array" is not defined in the current scope see `help scope' for details s

In the above example: · The print fails because array is not defined here--only in mkidentity. · The error message details may vary with the release, and translation.

80

Fortran 90 User 's Guide


7
·
Next Line--To advance execution to the next line, enter the next command. Example: Advance execution to the next line.
(dbx) next stopped in main at line 6 in file "a1.f90" 6 PRINT *, determinant( twobytwo ) (dbx) print twobytwo twobytwo = (1,1) 1.0 (2,1) 0.0 (1,2) 0.0 (2,2) 1.0 (dbx) quit demo$ s

Note that print twobytwo now displays the unit matrix. The next command executes the current source line, then stops at the next line. It counts subprogram calls as single statements. Compare next with step. The step command executes the next source line, or the next step into a subprogram, and so forth. In general, if the next executable source statement is a subroutine or function call, then · step sets a breakpoint at the first source statement of the subprogram. · next sets the breakpoint at the first source statement after the call but still in the calling program.

Segmentation Fault--Finding the Line Number
If a program gets a segmentation fault (SIGSEGV), it referenced a memory address outside of the memory available to it.

Some Causes of SIGSEGV
The most frequent causes are the following:

· · · · · ·

An array index being outside the declared range The name of an array index is misspelled The calling routine has a REAL argument; called routine has it as INTEGER An array index is miscalculated The calling routine calls with fewer arguments than required A pointer is used before it is defined

Debugging

81


7
You can locate the offending source line using -Xlist or dbx.

· ·

Recompile with the -Xlist option to get global program checking Use dbx to find the source code line where a segmentation fault occurred

Example: Program to generate a segmentation fault.
demo 4% cat WhereSEGV.f90 INTEGER a(5) j = 2000000 DO i = 1,5 a(j) = (i * 10) END DO PRINT *, a END demo 5% s

Example: Use dbx to locate a segmentation fault.
demo 5% f90 -g WhereSEGV.f90 demo 6% a.out Segmentation fault (core dumped) demo 7% dbx a.out Reading symbolic information for a.out ... other messages ... (dbx)run Running: a.out (process id 8813) signal SEGV (no mapping at the fault address) in main at line 4 in file "WhereSEGV.f90" 4 a(j) = (i * 10) (dbx) quit demo 8% s

82

Fortran 90 User 's Guide


7
Exception--Finding the Line Number
Example: Find where an exception occurred.
WhereExcept.f90 EXTERNAL INTEGER REAL :: ieeer = PRINT *, END INTEGER ! INTEGER sig, code, context(5) CALL abort() END demo$ f90 -g WhereExcept.f90 demo$ dbx a.out Reading symbolic information for a.out ... (dbx) catch FPE (dbx) run Running: a.out signal FPE (floating point divide by zero) in main at line 5 in file "WhereExcept.f" 5 PRINT *, r/s (dbx) s myhandler ieeer, ieee_handler, myhandler r=14.2, s=0.0 ieee_handler('set', 'all', myhandler) r/s ! Main

You can find the source code line number where a floating-point exception occurred by using the ieee_handler routine with either dbx or Debugger.

FUNCTION myhandler(sig, code, context) ! Handler

Note the "catch FPE" dbx command



Debugging

83


7
Trace of Calls
Sometimes a program stops with a core dump, and you need to know the sequence of calls that brought it there (a stack trace). Example: Show the sequence of calls, starting at where the execution stopped.
ShowTrace.f90 is a program contrived to get a core dump a few levels deep in the call sequence--to show a stack trace. demo$ f90 -g ShowTrace.f90 demo$ a.out Segmentation Fault (core dumped) demo$ dbx a.out Reading symbolic information for a.out ... (dbx) run (process id 8939) Running: a.out (process id 1089) signal SEGV (no mapping at the fault address) in calcb at line 23 in file "ShowTrace.f" 23 v(j) = (i * 10) (dbx) where =>[1] calcb(v = ARRAY , m = 2), line 23 in "ShowTrace.f90" [2] calc(a = ARRAY , m = 2, d = 0), line 9 in "ShowTrace.f90" [3] main(), line 3 in "ShowTrace.f90" (dbx) s

Execution stopped, line 23 calcb called from calc, line 9 calc called from main, line 3 Note reverse order: main called calc, calc called calcb.

The where command shows where in the program flow execution stopped (how execution reached this point), that is, a stack trace of the called routines. This can be helpful, since you no longer get an automatic traceback, as bemoaned in the ode below.
Ode To Traceback O blinding core! File of death! Alone like Abel's brother, Seth. The demise of process I cannot face Without the aid of stackish trace. To see what by you must needs be done, Please see Example Twenty-One.1 © Mateo Burtch, 1992

1. Since trace be dead, or just not there, try dbx's better where. Seek not example twenty one, as it was cited just for fun.

84

Fortran 90 User 's Guide


7
Pointer to a Scalar
Example: Pointer to a scalar, in dbx.
demo% f90 -g PtrScal.f90 demo% dbx a.out (dbx) list 1,99 1 PROGRAM PtrScalar 2 REAL, POINTER :: p 3 REAL, TARGET :: r 4 p => r 5 r = 2.3 6 PRINT *, p 7 p = 3.2 8 PRINT *, r 9 END (dbx) stop at 8 (2) stop at "PtrScal.f90":8 (dbx) run Running: a.out (process id 12367) 2.29999995 stopped in main at line 8 in file "PtrScal.f90" 8 PRINT *, r (dbx) whatis p real*4 p ! f90 pointer (dbx) whatis r real*4 r (dbx) print p p = 3.2 (dbx) print r r = 3.2 (dbx) quit demo$ s

PtrScal.f90

Debugging

85


7
Pointer to an Array
Example: Pointer to an array, in dbx.
demo% f90 -g PtrArray.f90 demo% dbx a.out (dbx) list 1,99 1 PROGRAM PtrArray 2 INTEGER, TARGET :: a(5,5) 3 INTEGER, POINTER :: corners(:,:) 4 DO i = 1,5 5 a(i,:) = i 6 END DO 7 corners => a(1:5:4, 1:5:4) 8 PRINT *, corners 9 END (dbx) stop at 8 (2) stop at "PtrArray.f90":8 (dbx) run Running: a.out (process id 12397) stopped in main at line 8 in file "PtrArray.f90" 8 PRINT *, corners (dbx) whatis a integer*4 a(1:5,1:5) (dbx) whatis corners integer*4 , corners(1:2,1:2) ! f90 pointer (dbx) print corners corners = (1,1) 1 (2,1) 5 (1,2) 1 (2,2) 5 (dbx) quit demo$ s

PtrArray.f90

86

Fortran 90 User 's Guide


7
User-Defined Types
Example: Structures--user defined types, in dbx.
demo% f90 -g DebStruc.f90 demo% dbx debstr (dbx) list 1,99 1 PROGRAM Struct ! Debug a Structure 2 TYPE product 3 INTEGER id 4 CHARACTER*16 name 5 CHARACTER*8 model 6 REAL cost 7 REAL price 8 END TYPE product 9 10 TYPE(product) :: prod1 11 12 prod1%id = 82 13 prod1%name = "Schlepper" 14 prod1%model = "XL" 15 prod1%cost = 24.0 16 prod1%price = 104.0 17 WRITE ( *, * ) prod1%name 18 END (dbx) stop at 17 (2) stop at "Struct.f90":17 (dbx) run Running: a.out (process id 12326) stopped in main at line 17 in file "Struct.f90" 17 WRITE ( *, * ) prod1%name (dbx) whatis prod1 product prod1 (dbx) whatis -t product type product integer*4 id character*16 name character*8 model real*4 cost real*4 price end type product (dbx) s

DebStruc.f90

Debugging

87


7
Example: Structures--user-defined types, in dbx.
(dbx) print prod1 = ( id = name = model = cost = price = ) (dbx) quit (dbx) s prod1 82 'Schlepper' 'XL' 24.0 104.0

88

Fortran 90 User 's Guide


7
Pointer to User-Defined Type
Example: Structures--user defined types, and pointers, in dbx.
demo% f90 -o debstr -g DebStruc.f90 demo% dbx debstr (dbx) stop in main (2) stop in main (dbx) list 1,99 1 PROGRAM DebStruPtr ! Debug structures & pointers 2 TYPE product 3 INTEGER id 4 CHARACTER*16 name 5 CHARACTER*8 model 6 REAL cost 7 REAL price 8 END TYPE product 9 10 TYPE(product), TARGET :: prod1, prod2 11 TYPE(product), POINTER :: curr, prior 12 13 curr => prod2 14 prior => prod1 15 prior%id = 82 16 prior%name = "Schlepper" 17 prior%model = "XL" 18 prior%cost = 24.0 19 prior%price = 104.0 20 curr = prior 21 WRITE ( *, * ) curr%name, " ", prior%name 22 END PROGRAM DebStruPtr (dbx) s

DebStruc.f90

Declare a user-defined type.

Declare variables prod1 and prod2 to be of that type and targets. Declare variables curr and prior as pointers to the type. Make curr point to prod1. Make prior point to prod1. Initialize prior.

Set curr to prior. Print name from curr and prior.

Example: Structures--set a breakpoint, and run under dbx.
The exact layout and messages may vary with each release. (dbx) stop at 21 (1) stop at "DebStruc.f90":21 (dbx) run Running: debstr (process id 10972) stopped in main at line 21 in file "DebStruc.f90" 21 WRITE ( *, * ) curr%name, " ", prior%name (dbx) s

Debugging

89


7
Example: Structures--print an item of user-defined type.
(dbx) print prod1 prod1 = ( id = 82 name = "Schlepper " model = "XL " cost = 24.0 price = 104.0 ) (dbx) s

Above, dbx displays all fields of the user-defined type, including field names. Example: Structures--inquire about an item of user-defined type.
Ask about the variable. Ask about the type (-t). (dbx) whatis prod1 product prod1 (dbx) whatis -t product type product integer*4 id character*16 name character*8 model real cost real price end type product (dbx) s

Example: Structures--print a pointer, then quit dbx.
dbx displays the contents of a pointer, which is an address. This address can be different with every run. (dbx) print prior = ( id = name = model = cost = price = ) (dbx) quit demo$ s prior 82 'Schlepper' 'XL' 24.0 104.0

90

Fortran 90 User 's Guide


7
Allocated Arrays
Example: Allocated arrays in dbx.
The exact layout and messages may vary with each release. Alloc.f90 demo% f90 -g Alloc.f90 demo% dbx a.out (dbx) list 1,99 1 PROGRAM TestAllocate 2 INTEGER n, status 3 INTEGER, ALLOCATABLE :: buffer(:) 4 PRINT *, 'Size?' 5 READ *, n 6 ALLOCATE( buffer(n), STAT=status ) 7 IF ( status /= 0 ) STOP 'cannot allocate buffer' 8 buffer(n) = n 9 PRINT *, buffer(n) 10 DEALLOCATE( buffer, STAT=status) 11 END (dbx) stop at 6 (2) stop at "alloc.f90":6 (dbx) stop at 9 (3) stop at "alloc.f90":9 (dbx) run Running: a.out (process id 10749) Size? 1000 stopped in main at line 6 in file "alloc.f90" 6 ALLOCATE( buffer(n), STAT=status ) (dbx) whatis buffer integer*4 , allocatable::buffer(:) (dbx) next continuing stopped in main at line 7 in file "alloc.f90" 7 IF ( status /= 0 ) STOP 'cannot allocate buffer' (dbx) whatis buffer integer*4 buffer(1:1000) (dbx) cont stopped in main at line 9 in file "alloc.f90" 9 PRINT *, buffer(n) (dbx) print n n = 1000 (dbx) print buffer(n) buffer(n) = 1000 (dbx) s

Unknown size is at line 6

Known size is at line 9

buffer(1000) holds1000

Debugging

91


7
Print Arrays
Example: dbx recognizes arrays. It can print arrays.
demo$ dbx a.out (dbx) list 1,25 1 DIMENSION iarr(4,4) 2 DO i = 1,4 3 DO j = 1,4 4 iarr(i,j) = (i*10) + j 5 END DO 6 END DO 7 END (dbx) stop at 7 (1) stop at "Arraysdbx.f90":7 (dbx) run Running: a.out stopped in main at line 7 in file "Arraysdbx.f90" 7 END (dbx) print IARR iarr = (1,1) 11 (2,1) 21 (3,1) 31 (4,1) 41 (1,2) 12 (2,2) 22 (3,2) 32 (4,2) 42 (1,3) 13 (2,3) 23 (3,3) 33 (4,3) 43 (1,4) 14 (2,4) 24 (3,4) 34 (4,4) 44 (dbx) print IARR(2,3) iarr(2, 3) = 23 order of user-specified subscripts ok (dbx) quit demo$ s

Arraysdbx.f90

92

Fortran 90 User 's Guide


7
Print Array Slices
This section shows one way of printing portions of large arrays. Example: dbx prints array slices if you specify which rows and columns.
ShoSli.f90 demo$ f90 -g ShoSli.f90 demo$ dbx a.out (dbx) list 1,12 1 INTEGER*4 a(3,4), col, row 2 DO row = 1,3 3 DO col = 1,4 4 a(row,col) = (row*10) + col 5 END DO 6 END DO 7 DO row = 1, 3 8 write(*,'(4I3)') (A(row,col),col=1,4) 9 END DO 10 END (dbx) stop at 7 (1) stop at "ShoSli.f90":7 (dbx) run Running: a.out stopped in main at line 7 in file "ShoSli.f90" 7 DO row = 1, 3 (dbx)s

Example: Print row 3.
(dbx) print a(3:3,1:4) a(3:3, 1:4) = (3,1) 31 (3,2) 32 (3,3) 33 (3,4) 34 (dbx)s

Example: Print column 4.
(dbx) print a(1:3,4:4) a(3:3, 1:4) = (1,4) 14 (2,4) 24 (3,4) 34 (dbx)s

Debugging

93


7
Generic Functions
Example: Generic function, cube root.
Generic.f90 (dbx) list 1,99 1 MODULE cr 2 INTERFACE cube_root 3 FUNCTION s_cube_root(x) 4 REAL :: s_cube_root 5 REAL, INTENT(IN) :: x 6 END FUNCTION s_cube_root 7 FUNCTION d_cube_root(x) 8 DOUBLE PRECISION :: d_cube_root 9 DOUBLE PRECISION, INTENT(IN) :: x 10 END FUNCTION d_cube_root 11 END INTERFACE 12 END MODULE cr 13 FUNCTION s_cube_root(x) 14 REAL :: s_cube_root 15 REAL, INTENT(IN) :: x 16 s_cube_root = x ** (1.0/3.0) 17 END FUNCTION s_cube_root 18 FUNCTION d_cube_root(x) 19 DOUBLE PRECISION :: d_cube_root 20 DOUBLE PRECISION, INTENT(IN) :: x 21 d_cube_root = x ** (1.0d0/3.0d0) 22 END FUNCTION d_cube_root 23 USE cr 24 REAL :: x, cx 25 DOUBLE PRECISION :: y, cy 26 WRITE(*,"('Enter a SP number: ')") 27 READ (*,*) x 28 cx = cube_root(x) 29 y=x 30 cy = cube_root(y) 31 WRITE(*,'("Single: ",F10.4, ", ", F10.4)') 32 WRITE(*,'("Double: ",F12.6, ", ", F12.6)') 33 WRITE(*,"('Enter a DP number: ')") 34 READ (*,*) y 35 cy = cube_root(y) 36 x=y 37 cx = cube_root(x) 38 WRITE(*,'("Single: ",F10.4, ", ", F10.4)') 39 WRITE(*,'("Double: ",F12.6, ", ", F12.6)') 40 END

x, cx y, cy

x, cx y, cy

94

Fortran 90 User 's Guide


7
Example: dbx with a generic function, cube root.
(dbx) stop at 26 (2) stop at "Generic.f90":26 (dbx) run Running: Generic (process id 14633) stopped in main at line 26 in file "Generic.f90" 26 WRITE(*,"('Enter a SP number : ')") (dbx) whatis cube_root More than one identifier 'cube_root.' Select one of the following names: 1) `Generic.f90`cube_root s_cube_root ! real*4 s_cube_root 2) `Generic.f90`cube_root s_cube_root ! real*8 d_cube_root >1 real*4 function cube_root (x) (dummy argument) real*4 x (dbx) print cube_root(8.0) More than one identifier 'cube_root.' Select one of the following names: 1) `Generic.f90`cube_root ! real*4 s_cube_root 2) `Generic.f90`cube_root ! real*8 d_cube_root >1 cube_root(8) = 2.0 (dbx) stop in cube_root More than one identifier 'cube_root.' Select one of the following names: 1) `Generic.f90`cube_root ! real*4 s_cube_root 2) `Generic.f90`cube_root ! real*8 d_cube_root >1 (3) stop in cube_root (dbx) cont continuing Enter a SP number: 8 stopped in cube_root at line 16 in file "Generic.f90" 16 s_cube_root = x ** (1.0/3.0) (dbx) print x x = 8.0 (dbx)s

If asked "What is cube_root?", dbx tells you there are two, and asks you to select one.

If asked for cube_root(8) dbx tells you there are two, and asks you to select one.

If told to stop in cube_root, dbx tells you there are two, and asks you to select one.

From inside s_cube_root, show current value of x.

Debugging

95


7
Miscellaneous Tips
The following tips and background concepts can help.

Current Procedure and File
During a debug session, the Debugger defines a procedure and a source file as current. Requests to set breakpoints and to print or set variables are interpreted relative to the current function and file. Thus, "stop at 5" sets one of three different breakpoints, depending on whether the current file is a1.f90, a2.f90, or a3.f90.

Uppercase Letters
In general, if your program has uppercase letters in any identifiers, then the Debugger recognizes them. You do not need to give it any specific case sensitive/insensitive commands, as in some earlier versions. In fact, for f90 1.0, f90 and dbx must both be in the case insensitive mode; that is, do not set "dbxenv case sensitive". Note ­ Names of files or directories are always case sensitive in both Debugger and dbx. This is true even if you have set the "dbxenv case insensitive" environment attribute.

Main Features of the Debugger
Be sure to read Debugging a Program for the following:

· ·

The full range of features in the Debugger The window- and mouse-based interface

Overview of dbx Features Useful for Fortran 90
The Debugger provides event management, process control, and data inspection. It allows you to watch what is happening during program execution.

96

Fortran 90 User 's Guide


7
With dbx, you can do such things as the following:
Solaris 2.x

Solaris 2.x

· · · · · · · · · · · · · · ·

Set watchpoints to stop or trace if a specified item changes Collect data for the performance-tuning Analyzer Graphically monitor variables, structures, and arrays--Data Inspector Set breakpoints (set places to halt in the program) at lines or in functions Show values--once halted, show or modify variables, arrays, structures, ... Step through program, one source line at a time (or one assembly line) Trace program flow (show sequence of calls taken) Invoke procedures in the program being debugged Step over or into function calls; step up and out of a function call Run, stop, and continue execution (at the next line or at some other line) dbx-safe I/O in the command window--Program Input/Output Window Save and then replay all or part of a debugging run Stack--Examine the call stack; move up and down the call stack Program scripts by embedded Korn shell Follow programs as they fork(2) and exec(2)

Help
At the Debugger prompt, to get:

· · · ·

All commands--a list of commands, grouped by action, type help Details of one command--a command explanation, type help cmdname Changes--a list of the new and changed features, type help changes FAQ--answers to frequently asked questions, type help FAQ

Example: Command Summary (output varies with release).
(dbx) help Command Summary Execution and Tracing cancel catch clear cont delete fix fixed handler ignore intercept next pop replay rerun restore run save status step stop trace unintercept when whocatches Displaying and Naming Data assign call demangle dis display down dump examine exists frame hide inspect print undisplay unhide up whatis where whereami whereis which

Debugging

97


7
Example: "help cmdnam"--details for the where command.
(dbx) where where where where where where info) help where -f -h -q -v # # # # # # Print a procedure traceback Print the top frames in the traceback Start traceback from frame Include hidden frames Quick traceback (only function names) Verbose traceback (include function args and line

Any of the above forms may be followed by a thread or LWP ID to obtain the traceback for the specified entity. (dbx) s

98

Fortran 90 User 's Guide


Floating Point
This chapter is organized into the following sections.
Summary IEEE Solutions The General Problems IEEE Exceptions IEEE Routines Debugging IEEE Exceptions Guidelines Miscellaneous Examples page 99

8

page 101 page 100 page 102 page 103 page 113 page 113 page 114

8.1 Summary
This chapter introduces some floating-point issues. It focuses on IEEE floating point, and provides some reasons for it, some definitions, and some examples of how to use it. It lets you use IEEE floating point with some understanding. It is more tutorial than the other chapters, and deeper. This chapter is intended for scientists and engineers who use floating-point arithmetic in their work, but are not necessarily numerical analysts. Read the Numerical Computation Guide, where you will find more complete explanations, examples, and details. You might also want to read What Every Computer Scientist Should Know About Floating-point Arithmetic," by David

99


8
Goldberg, which is in the on-line READMEs directory. It is a PostScript file and can be printed by lpr on any PostScript-compatible printer that has Palatino font. It can be viewed on-line by pageview.

8.2 The General Problems
How can IEEE arithmetic help solve real problems? IEEE 754 standard floatingpoint arithmetic offers the user greater control over computation than is possible in any other type of floating point. In scientific research, there are many ways for errors to creep in.

· · · ·

The The ATA The The

model may be wrong. algorithm may be numerically unstable (solving equations by inverting for example). data may be ill-conditioned. computer may be doing something astonishing, or at least unexpected.

It is nearly impossible to separate these error sources. Using library packages which have been approved by the numerical analysis community reduces the chance of there being a code error. Using good algorithms is another must. Using good computer arithmetic is the next obvious step. The IEEE standard represents the work of many of the best arithmetic specialists in the world today. It was influenced by the mistakes of the past. It is, by construction, better than the arithmetic employed in the S/360 family, the VAX family, the CDC1, CRAY, and UNIVAC2 families (to name but a few). This is not because these vendors are not clever, but because the IEEE pundits came later and were able to evaluate the choices of the past, and their consequences. Does IEEE arithmetic solve all problems? No. But in general, the IEEE Standard makes it easier to write better numerical computation programs.

1. CDC is a registered trademark of Control Data Corporation. 2. UNIVAC is a registered trademark of UNISYS Corporation.

100

Fortran 90 User 's Guide


8
8.3 IEEE Solutions
IEEE arithmetic is a relatively new way of dealing with arithmetic operations where the result yields such problems as invalid, division by zero, overflow, underflow, or inexact. The big differences are in rounding, handling numbers near zero, and handling numbers near the machine maximum. For rounding, IEEE arithmetic defaults to doing the intuitive thing, and closely corresponds with old arithmetic. IEEE offers choices, which the expert can use to good effect, while old arithmetic did it just one way. What happens when we multiply two very large numbers with the same sign? Large numbers of different signs? Divide by zero? Divide zero by zero? In old arithmetic, all these cases are the same. The program aborts on the spot; or in some very old machines, the computation proceeds, but with garbage.IEEE provides choices. The default solution is to produce the following.
big*big = big*(-)big num/0.0 = num/0.0 = 0.0/0.0 = +Inf = -Inf +Inf Where num > 0.0 -Inf Where num < 0.0 NaN Not a Number

Above, +Inf, -Inf, and NaN are just introduced intuitively. More later. Also an exception of one of the following kinds is raised. Invalid -- Examples that yield invalid are 0.0/0.0, sqrt(-1.0), log(-37.8), ... Division by zero -- Examples that yield division by zero are 9.9/0.0, 1.0/0.0, ... Overflow -- Example with overflow: MAXDOUBLE+0.0000000000001e308 Underflow -- Example that yields underflow: MINDOUBLE * MINDOUBLE Inexact -- Examples that yield inexact are 2.0 / 3.0, log(1.1), read in 0.1, ... (no exact representation in binary for the precision involved) There are various reasons to care about how all this works.

· · ·

If you don't understand what you are using, you may not like the results. Poor arithmetic can produce poor results. This cannot easily be distinguished from other causes of poor results. Switching everything to double precision is no panacea.

Floating Point

101


8
8.4 IEEE Exceptions
IEEE exception handling is the default on a SPARC processor. However, there is a difference between detecting a floating-point exception, and generating a signal for a floating-point exception (SIGFPE).

Detecting a Floating-point Exception
In accordance with the IEEE standard, two things happen when a floatingpoint exception occurs in the course of an operation.

· ·

The handler returns a default result. For 0/0, return NaN as the result. A flag is set that an exception is raised. For 0/0, set "invalid operation" to 1.

Generating a Signal for a Floating-point Exception
The default on SPARC hardware systems is that they do not generate a signal for a floating-point exception. The assumption is that signals degrade performance, and that most users don't care about most exceptions. To generate a signal for a floating-point exception, you establish a signal handler. You use a predefined handler or write your own. See "Exception Handlers and ieee_handler()" later in this chapter for details.

Default Signal Handlers
By default, f90 sets up some signal handlers, mostly for dealing with such things as a floating-point exception, interrupt, bus error, segmentation violation, or illegal instruction. Although you would not generally want to turn off this default behavior, it is possible to do so by setting the global C variable f77_no_handlers to 1. Example: Get no default signal handlers (set f77_no_handlers to 1.) 1. Create the following C program.
demo$ cat NoHandlers.c int f90_no_handlers=1; demo$

102

Fortran 90 User 's Guide


8
2. Compile it and save the .o file.
demo$ cc -c -o NoHand NoHandlers.c demo$

3. Link the corresponding .o file into your executable file.
demo$ f90 NoHand.o Any.f90 demo$

Otherwise (by default) it is 0. The effect is felt just before execution is transferred to the user 's program so it does not make sense to set/unset it in the user 's program. Note ­ This variable is in the name space of the user 's program, so do not use f90_no_handlers as the name of a variable anywhere else other than in the above C program.

8.5 IEEE Routines
Many vendors support the IEEE standard. The SPARC processors conform to the IEEE standard in a combination of hardware and software support for different aspects. The older Sun-4 uses the Weitek 1164/5, and the Sun-4/110 has that as an option. The newer Sun-4 and the SPARCsystem series both use floating-point units with hardware square root. This is accessed if you compile with the -cg89 option. The newest SPARCsystem series uses new floating-point units, including SuperSPARC, with hardware integer multiply and divide instructions. These are accessed if you compile with the -cg92 option. The utility fpversion tells which floating-point hardware is installed. This utility runs on all Sun architectures. See fpversion(1), and read the Numerical Computation Guide for details. This replaces the older utility fpuversion4.

Floating Point

103


8
The the following interfaces help people use all the facets of IEEE arithmetic. These are mostly in the math library, in the libm.a and libm.il files, and in several .h files.

· · · · ·

ieee_flags(3m) Control rounding direction. Control rounding precision. Query exception status. Clear exception status. ieee_handler(3m) Establish exception handler. Remove exception handler. ieee_functions(3m) List name and purpose of each IEEE function. ieee_values(3m) A list of functions that return special values. Other libm functions: · ieee_retrospective · nonstandard_arithmetic · standard_arithmetic

Flags and ieee_flags()
The ieee_flags function is part of the libm shipped with the operating system. It allows the programmer to do the following.

· · ·

Control rounding direction and rounding precision Check the status of the exception flags Clear exception status flags

The ieee_flags function can be used to query and clear exception status flags. The general form of a call to ieee_flags is as follows.
i = ieee_flags ( action, mode, in, out )

· · · ·

Each of the four arguments is a string. Input: action, mode, and in Output: out and i ieee_flags is an integer-valued function. Useful information is returned in i. Refer to the man page for ieee_flags(3m) for complete details.

104

Fortran 90 User 's Guide


8
Possible parameter values are shown below.
action: mode: in,out: get, set, clear, clearall direction, precision, exception nearest, tozero, negative, positive, extended, double, single, inexact, division, underflow, overflow, invalid, all, common

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 8-1 ieee_flags Argument Meanings Refers to

Value of in and out

nearest, tozero, negative, positive extended, double, single inexact, division, underflow, overflow, invalid all common

Rounding direction Rounding precision Exceptions All 5 exceptions Common exceptions: invalid, division, overflow

Example: To determine what is the highest priority exception that has a flag raised, pass the input argument in as the null string.
ieeer = ieee_flags( 'get', 'exception', '', out ) PRINT *, out, ' flag raised'

Example: 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 )

Floating Point

105


8
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 )

Turn Off All Warning Messages with ieee_flags
Use this only if you are certain you don't want to know about the unrequited exceptions. To do this, clear all accrued exceptions by putting a call to ieee_flags() just before your program exits. Example: Clear all accrued exceptions with ieee_flags().
i = ieee_flags('clear', 'exception', 'all', out )

Calls of this form are used in the next two examples.

Values and ieee_values()
The ieee_values(3m) file is a collection of functions. Each function returns a special IEEE value. The Fortran names for these functions are listed in libm_double(3f) and libm_single(3f). You can use special IEEE entities, such as infinity or minimum normal, in a user program. See also the man page ieee_values(3m). Example: A convergence test might be like this.
IF ( delta .LE. r_min_normal() ) RETURN

106

Fortran 90 User 's Guide


8
The IEEE values available are listed in the table below.
Table 8-2 Functions for Using IEEE Values Double Precision d_infinity() d_quiet_nan() d_signaling_nan() d_min_normal() d_min_subnormal() d_max_subnormal() d_max_normal() Single Precision r_infinity() r_quiet_nan() r_signaling_nan() r_min_normal() r_min_subnormal() r_max_subnormal() r_max_normal()

IEEE Value infinity quiet NaN signaling NaN min normal min subnormal max subnormal max normal

For the two NaN functions, you can assign and/or print out the values, but comparisons using either of them always yield false. To determine whether some value is a NaN, use the function ir_isnan(r) or id_isnan(d); see libm_double(3f), libm_single(3f), and ieee_functions(3m).

Exception Handlers and ieee_handler()
A floating-point user may need to know the following about IEEE exceptions.

· · · · · ·

What How How How

happens when an exception occurs? to use ieee_handler() to establish a function as a signal handler to write a function that can be used as a signal handler to locate the exception (Where did it occur?)

To get information about an exception: Generate a signal for a floating-point exception. The official UNIX name for signal is floating-point exception is SIGFPE. To generate a SIGFPE, establish a signal handler. The default behavior on SPARC hardware systems is "do not generate a SIGFPE."

Establishing a Signal Handler Function with ieee_handler()
To establish a signal handler, pass the following to ieee_handler():

· · ·

Name of the function Exception to watch for Action to take

Floating Point

107


8
Once a handler is established, a signal is generated whenever the particular floating-point exception occurs. The form of invoking ieee_handler() is as follows.
i = ieee_handler( action, exception, handler )

action exception

character character

"get" or "set" or "clear" "invalid" or "division" or "overflow" or "underflow" or "inexact" The name of the function you wrote 0=OK

handler
return value

function name integer

Writing a Signal Handler Function
Actions taken by the function are up to you, but the form of the function is:

· ·

The function must be an integer function. The function must have three arguments, typed as follows: Pattern-- hand5x( sig, sip, uap ) · hand5x is your name for your integer function · sig is an integer · sip is a record which has the structure siginfo (see sample below) · uap is not used here

108

Fortran 90 User 's Guide


8
Form: Signal handler function.
5.x SunOS 5.x Form INTEGER FUNCTION hand5x( sig, sip, uap) INTEGER sig, location TYPE fault_typ INTEGER address END TYPE fault_typ TYPE siginfo INTEGER si_signo INTEGER SI_CODE INTEGER si_errno TYPE(fault_typ) fault END TYPE siginfo TYPE(siginfo) sip location = sip%fault%address ... actions you take ... END

Pattern-- hand4x( sig, code, context ) hand4x is your name for your integer functionForm: Signal handler function. Example: Detect an exception using a handler, and abort--SunOS 5.x or 4.x.
5.x and 4.x DetExcHan.f90 SunOS 5.x or 4.x SIGFPE is generated whenever a floating-point exception occurs Then the SIGFPE is detected and control is passed to the myhandler function. PROGRAM DetExcH 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) ! Handler, 5.x or 4.x ! OK in SunOS 5.x/4.x as all we do is abort INTEGER sig, code INTEGER, DIMENSION(5) :: context CALL abort() END demo% f90 DetExcHan.f90 demo% a.out abort: called Abort (core dumped) demo% s

Floating Point

109


8
Example: Locate an exception (get address) using a handler
5.x LocExcHan5x.f90 SunOS 5.x An addressis mostly for those who use such low-level debuggers as adb. For how get the line numbersee , Section 8.6, "Debugging IEEE Exceptions" for details. PROGRAM LocExcH5x ! Locate Exception, by Handler - SunOS 5.x EXTERNAL hand5x INTEGER hand5x REAL :: r = 14.2, s = 0.0 i = ieee_handler('set', 'division', hand5x) t = r/s END INTEGER FUNCTION hand5x(sig, sip, uap) ! Handler - SunOS5.x INTEGER location, sig TYPE fault_typ INTEGER address END TYPE fault_typ TYPE siginfo INTEGER si_signo INTEGER si_code INTEGER si_errno TYPE(fault_typ) fault END TYPE siginfo TYPE(siginfo) sip location = sip%fault%address WRITE (*,"('Exception at ',Z8)") location ! Risky in a handler CALL abort() ! Just to reduce risk. END

Caveat: I/O in a handler is risky. Calling abort() reduces the risk.

Example: Locate an exception (get address) using a handler.
5.x Compile/Load/Run SunOS 5.x The actual address varies with installation and architecture. demo% f90 LocExcHan5x.f90 demo% a.out Exception at 11FA8 abort: called Abort (core dumped) Note: Following IEEE floating-point traps enabled;see ieee_handler(3M): Division by Zero; Sun's implementation of IEEE arithmetic is discussed in the Numerical Computation Guide. demo% s

110

Fortran 90 User 's Guide


8
Example: Locate an exception (get address) using a handler
4.x Compile/Load/Run SunOS 4.x The actual address varies with installation and architecture. demo% f90 LocExcHan4x.f90 demo% a.out Exception at pc 9172 abort: called Abort (core dumped) Note: Following IEEE floating-point traps enabled;see ieee_handler(3M): Division by Zero; Sun's implementation of IEEE arithmetic is discussed in the Numerical Computation Guide. demo% s

Retrospective
The ieee_retrospective function queries the floating-point status registers to find out which exceptions have accrued. If any exception has a raised accrued exception flag, a message is printed to standard error to inform the programmer which exceptions were raised but not cleared. For Fortran 90, this function is called automatically just before execution terminates. The message typically looks like this (may vary slightly with each release):
NOTE: The following IEEE floating-point arithmetic exceptions occurred and were never cleared: Inexact; Division by Zero; Underflow; Overflow; Invalid Operand; Sun's implementation of IEEE arithmetic is discussed in the Numerical Computation Guide

Nonstandard Arithmetic
Another useful math library function is nonstandard arithmetic. The IEEE standard for arithmetic specifies a way of handling underflowed results gradually, by dynamically adjusting the radix point of the significand. Recall that 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 to shift the radix point into the significand, when the result of a floating-point computation would otherwise underflow. This is not accomplished in

Floating Point

111


8
hardware on a SPARC processor, but in software. If your program happens to generate many underflows (perhaps a sign of a problem with your algorithm?), and you run on a SPARC processor, you may experience a performance loss. To turn off gradual underflow, compile with -fnonstd, or insert this.
CALL nonstandard_arithmetic()

To turn on gradual underflow (after you have turned it off), insert this.
CALL standard_arithmetic()

The standard_arithmetic() subroutine corresponds exactly to an earlier version named gradual_underflow(). The nonstandard_arithmetic() subroutine corresponds exactly to an earlier version named abrupt_underflow().

Messages about Floating-point Exceptions
For Fortran 90, the current default is to display a list of accrued floating-point exceptions at the end of execution. In general, you will get a message if any one of the invalid, division-by-zero, or overflow exceptions occur. Since most real programs raise underflow and inexact exceptions, you will get a message if any two of the underflow and inexact exceptions occur, in general. You can turn off any or all of these messages with ieee_flags() by clearing exception status flags. If this is done at all, it is usually done at the end of your program. Clearing all messages is not recommended. You can gain complete control with ieee_handler(). In your own exception handler routine you can specify actions, and you can turn off messages with ieee_flags() by clearing exception status flags.

112

Fortran 90 User 's Guide


8
8.6 Debugging IEEE Exceptions
You may want to debug programs that have worrisome messages like this.
NOTE: the following IEEE floating-point arithmetic exceptions occurred and were never cleared: Inexact; Division by Zero; Underflow; Overflow; Invalid Operand; Sun's implementation of IEEE arithmetic is discussed in the Numerical Computation Guide

You need to do these two things:

·

Establish a signal handler, using ieee_handler(3m). This will cause a SIGFPE to be generated when a floating-point exception occurs.

·

After you invoke dbx, enter the "catch FPE" command. This causes dbx to listen for any SIGFPE, and halt when it hears one. See the subsection on where the exception occurred in Section 7.2, "The dbx Debugger" for explicit examples.

8.7 Guidelines
To sum up, SPARC arithmetic is a state-of-the art implementation of IEEE arithmetic, optimized for the most common cases.

· ·

More problems can safely be solved in single precision, due to the clever design of IEEE arithmetic. To get the benefits of IEEE math for most applications, if your program gets one of the common exceptions, then you probably want to continue with a sensible result. That is, you do not want to use ieee_handler to abort on the common exceptions. If your system time is very large, over 50% of runtime, check into modifying your code, or using nonstandard_arithmetic.

·

Floating Point

113


8
8.8 Miscellaneous Examples
A miscellaneous collection of more or less realistic examples is provided here, as a possible additional aid.

Kinds of Problems
The problems in this chapter usually involve arithmetic operations with a result of invalid, division by zero, overflow, underflow, or inexact. For instance, Underflow -- In old arithmetic, that is, prior to IEEE, if you multiply two very small numbers on a computer, you get zero. Most mainframes and minicomputers behave that way. In IEEE arithmetic, there is gradual underflow; this expands the dynamic range of computations. For example, consider a machine with 1.0E-38 as the machine epsilon, the smallest representable value on the machine. Multiply two small numbers.
a = 1.0E-30 b = 1.0E-15 x=a*b

In old arithmetic you get 0.0, but with IEEE arithmetic (and the same word length) you get 1.40130E-45. With old arithmetic, if a result is near zero, it becomes zero. This can cause problems, especially when subtracting two numbers -- because this is a principal way accuracy is lost. You can also detect that the answer is inexact. The inexact exception is common, and means the calculated result cannot be represented exactly, at least not in the precision being used, but it is as good as can be delivered. Underflow tells us, as we can tell in this case, that we got an answer smaller than the machine naturally represents. This is accomplished by stealing some bits from the mantissa and shifting them over to the exponent. The result is less precise, in some sense, but more so in another. The deep implications are beyond this discussion. The interested reader may wish to consult Computer, January 1980, Volume 13, Number 1, particularly I. Coonen's article, "Underflow and the Denormalized Numbers."

114

Fortran 90 User 's Guide


8
Roundoff -- Most scientific programs have sections of code that are sensitive to roundoff, often in an equation solution or matrix factorization. So be concerned about numerical accuracy -- if your computer doesn't do a good job, your results will be tainted, and there is often no way to know that this has happened.

Simple Underflow
Some applications actually do a lot of work very near zero. This is common in algorithms which are computing residuals, or differential corrections. For maximum numerically safe performance, perform the key computations in extended precision. If the application is a single-precision application, this is easy, as we can perform key computations in double precision. Example: A simple dot product computation.
sum = 0 DO i = 1, n sum = sum + a(i) * b(i) END DO

If a(i) and b(i) are small, many underflows will occur. By forcing the computation to double precision, you compute the dot product with greater accuracy, and not suffer underflows.
REAL*8 sum DO i = 1, n sum = sum + dble(a(i)) * dble(b(i)) END DO result = sum

It may be advisable to have both versions, and to switch to the double precision version only when required. You can force a SPARC processor to behave like an older computer with respect to underflow. Add the following to your Fortran 90 main program.
CALL nonstandard_arithmetic()

But be aware that you are giving up the numerical safety belt that is the operating system default. You can get your answers faster, and you won't be any less safe than, say, a VAX -- but use at your own risk.

Floating Point

115


8
Use Wrong Answer
You might wonder why continue if the answer is clearly wrong. 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 which are possible are +5v, 0, -5v. It is possible to carefully arrange each part of the calculation to coerce each subresult to the correct range.
4.0 -4.0 -Inf < computed < Inf 5 volts computed 4.0 0 volts < computed -4.0 -5 volts

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, as 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 executing (since you don't do unneeded comparisons).

Excessive Underflow
If two very small numbers are multiplied, the result underflows. The hardware, being designed for the typical case, does not produce a result; instead software is employed to compute the correct IEEE complying result. As one might guess, this is much slower. In the majority of applications, this is invisible. When it is not, the symptom is that the system time component of your runtime (which can be determined by running your application with the time command) is much too large.

116

Fortran 90 User 's Guide


8
Example: Excessive underflow.
PROGRAM dotprod INTEGER maxn PARAMETER (maxn=10000) REAL a(maxn), b(maxn), eps /1.0e-37/, sum DO i = 1, maxn a(i) = 1.0e-30 b(i) = 1.0e-15 END DO sum = 0. DO i = 1, maxn sum = sum + a(i)*b(i) END DO END

After compiling and running dotprod, the results of the time command are:
real 0m4.50s user 0m0.11s sys 0m4.35s

So the real computation required about 0.1 seconds, but the software fix required four seconds. In a real application this can be hours. Clearly this is not desirable.

Solution 1: Change All of the Program
If you change the code to be double precision (by rewriting the code with double precision variables) you get vast improvement.
real 0m0.20s user 0m0.08s sys 0m0.11s

Now, of course, it may not be desirable to promote an entire program to double precision (though this is what is traditionally done to make up for the fact that old style arithmetic is less accurate).

Floating Point

117


8
Solution 2: Change One Double-Precision Variable
If you declare sum to be DOUBLE PRECISION, and change only the summation line of code as follows:
sum = sum + a(i)*dble(b(i))

then you get
real 0m0.18s user 0m0.06s sys 0m0.11s

By promoting one variable to double, you eliminate the software underflow problem. Note that in a real application, put the variable sum in double precision and coerce it to single precision only on output. This is not a performance issue, but a numeric one. Of course it may not be easy to tell which variables in a huge program need to be promoted. The effort is worthwhile, not only because of the performance (which, as you will learn, can be achieved in other ways), but because the numerics are enhanced as well.

Solution 3: Nonstandard Arithmetic
There is a quick and dirty solution, which is:
CALL nonstandard_arithmetic()

This tells the hardware to act like an old-style computer, and when underflow occurs just flush to zero. This results in a runtime like this.
real 0m0.18s user 0m0.01s sys 0m0.13s

Note that this time is very nearly the same as promoting one variable to double. The difference is that now the computed result is 0. This is bad because if this dot product is really the final result, there is probably nothing wrong with this solution. If, however, this result feeds into more elaborate computations, you have thrown away some information. This may be important. If the algorithm is stable, the input well conditioned, and the implementation careful, it won't matter. If there is anything else shaky, this may push it over.

118

Fortran 90 User 's Guide


C­Fortran Interface
This chapter is organized into the following sections.
Sample Interface How to Use this Chapter Compatibility Requirements Fortran Calls C C Calls Fortran

9

page 120 page 121 page 122 page 134 page 149

Glendower: I can call spirits from the vasty deep. Hotspur: Why, so can I, or so can any man; But will they come when you do call for them? Henry IV, Part I Purpose--This chapter shows how to write Fortran 90 routines that call C routines, and C routines that call Fortran 90 routines. A common reason to do such calls is to use existing libraries. Caveat--This subject requires more sophistication than most of this manual. To paraphrase Hotspur, any programmer can write such programs, but will they work when you do call for them? Approach--This chapter lists the compatibility rules and shows examples for item passable between C and Fortran. Examples show how to pass an item, not how it would be used in real applications.

119


9
9.1 Sample Interface
Example: A C function to be called by a Fortran main program.
Samp.c samp_ ( int *i, float *f ) /* both i and f are pointers */ { *i = 9; *f = 9.9; }

Example: A Fortran main program to call a C function.
Sampmain.f90 PROGRAM Sample INTEGER i REAL r CALL Samp ( i, r ) ! both i and r are passed by reference WRITE( *, "(I2, F4.1)") i, r END PROGRAM Sample

Example: A Fortran program calls a C function.
Compile/Link/Execute. demo$ cc -c Samp.c demo$ f90 Samp.o Sampmain.f90 demo$ a.out 9 9.9 demo$ s



This does the linking

120

Fortran 90 User 's Guide


9
9.2 How to Use this Chapter
1. Examine the previous section, "Sample Interface." 2. Examine the next section, "Compatibility Requirements." 3. Find what to do in the section "Fortran Calls C" or "C Calls Fortran." The following two tables help find the appropriate subsection. For a Fortran 90 main and a C function:
Fortran Calls C Arguments Passed by Reference (f90 Calls C) Simple Arguments Passed by Reference (f90 Calls C) Complex Arguments Passed by Reference (f90 Calls C) Character Arguments Passed by Reference (f90 Calls C) Vector Arguments Passed by Reference (f90 Calls C) Matrix Arguments Passed by Reference (f90 Calls C) Structure Arguments Passed by Reference (f90 Calls C) Pointer Arguments Passed by Reference (f90 Calls C)--N/A Arguments Passed by Value (f90 Calls C) Function Return Values (f90 Calls C) INTEGER Function Return Value (f90 Calls C) REAL Function Return Value (f90 Calls C) Pointer-to-a-REAL Function Return Value (f90 Calls C) DOUBLE PRECISION Function Return Value (f90 Calls C) LOGICAL Function Return Value (f90 Calls C) CHARACTER Function Return Value (f90 Calls C) N/A Labeled Common (f90 Calls C) Alternate Returns (f90 Calls C) - N/A page 134 page 134 page 134 page 135 page 136 page 138 page 139 page 140 page 140 page 141 page 141 page 141 page 142 page 143 page 144 page 145 page 146 page 147 page 148

C­Fortran Interface

121


9
For a C main and a Fortran 90 subprogram:)
C Calls Fortran Arguments Passed by Reference (C Calls f90) Simple Arguments Passed by Reference (C Calls f90) Complex Arguments Passed by Reference (C Calls f90) Character Arguments Passed by Reference (C Calls f90) Vector Arguments Passed by Reference (C Calls f90) Matrix Arguments Passed by Reference (C Calls f90) Structure Arguments Passed by Reference (C Calls f90) Pointer Arguments Passed by Reference (C Calls f90)--N/A Arguments Passed by Value (C Calls f90) - N/A Function Return Values (C Calls f90) INTEGER Function Return Value (C Calls f90) REAL Function Return Value (C Calls f90) DOUBLE PRECISION Function Return Value (C Calls f90) LOGICAL Function Return Value (C Calls f90) CHARACTER Function Return Value (C Calls f90) Labeled Common (C Calls f90) Alternate Returns (C Calls f90) page 149 page 149 page 149 page 150 page 151 page 152 page 152 page 153 page 155 page 155 page 155 page 155 page 156 page 156 page 158 page 159 page 160 page 161

9.3 Compatibility Requirements
Most C/Fortran interfaces must get all of these aspects right:

· · · · · · · · · ·
122

Function or subroutine Underscore in names of routines Upper and lowercase in identifiers Data type compatibility Passing arguments by reference or value String arguments and order Telling the linker to use Fortran libraries

Some C/Fortran interfaces must also get these right: Arrays: Indexing and order File descriptors and stdio File permissions

Fortran 90 User 's Guide


9
Function or Subroutine
The word function means different things in C and Fortran.

· ·

As far as C is concerned, all subprograms are functions, it is just that some of them return a null value. As far as Fortran is concerned, a function passes a return value and a subroutine does not.

Fortran Calls a C Function

· ·

If the called C function returns a value, call it from Fortran as a function. If the called C function does not return a value, call it as a subroutine.

C Calls a Fortran Subprogram

· ·

If the called Fortran subprogram is a function, call it from C as a function that returns a comparable data type. If the called Fortran subprogram is a subroutine, call it from C as a function that returns a value of int (comparable to Fortran INTEGER*4) or void. This return value is useful if the Fortran routine does a nonstandard return.

C­Fortran Interface

123


9
Underscore in Names of Routines
The Fortran compiler appends an underscore ( _ ) to the names of subprograms, for both a subprogram and a call to a subprogram. This distinguishes it from C procedures or external variables with the same userassigned name. Each subprogram name must have 31 or fewer characters. To avoid the underscore problem, in the C function definition, change the name of the C function by appending an underscore to that name.

Case Sensitivity
C and Fortran take opposite perspectives on case sensitivity.

· ·

C is case sensitive--uppercase or lowercase matters. Fortran ignores case.

The Fortran default is to ignore case by converting identifiers to lowercase. It converts all uppercase letters to lowercase letters, except within characterstring constants. To avoid the case sensitivity problem--in the C function definition, make the name of the C function all lowercase.

124

Fortran 90 User 's Guide


9
Data Type Compatibility
You may want to write Fortran 90 routines to interface with existing C routines, or C routines to interface with existing Fortran 90 routines.

Writing Fortran 90 Code for Existing C Routines
For any given C intrinsic data type, the following table provides a close corresponding Fortran 90 data type.
Table 9-1 C Intrinsic Type char x ; signed char x ; signed char x[n] ; unsigned char x[n] ; float x ; double x ; long double x ; int x ; signed x ; signed int x ; long x ; long int x ; signed long x ; signed long int x ; unsigned int x ; unsigned long x ; unsigned long int x ; short x ; short int x ; signed short int x ; unsigned short x ; unsigned short int x ; long long x ; unsigned long long x ; C Data Type to Fortran 90 Data Type Close Fortran 90 Type CHARACTER x INTEGER (KIND=1) x CHARACTER (LEN=n) x CHARACTER (LEN=n) x REAL x DOUBLE PRECISION x N/A INTEGER x INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER INTEGER N/A N/A x x x x x x x x x (KIND=2) (KIND=2) (KIND=2) (KIND=2) (KIND=2) x x x x x Size (Bytes) 1 1 n n 4 8 16 4 4 4 4 4 4 4 4 4 4 2 2 2 2 2 Alignment (Bytes) 1 4 1 1 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 -

C double aligns on 8-byte boundaries, unless in a common block, then on 4.

C­Fortran Interface

125


9
Writing C Code for Existing Fortran 90 Routines
For any given Fortran 90 intrinsic data type, the following table provides a close corresponding C data type.
Table 9-2 Fortran 90 Data Type to C Data Type Close C Data Type unsigned char x ; unsigned char x[n] ; unsigned char x[n] ; struct {float r,i;} x; struct {float r,i;} x; struct {double dr,di;} x; double x ; float x ; float x ; double x ; int x ; signed char x ; short x ; int x ; int x ; signed char x ; short x ; int x ; Size (Bytes) 1 n n 8 8 16 8 4 4 8 4 1 2 4 4 1 2 4 Alignment (Bytes) 1 1 1 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4

Fortran 90 Intrinsic Data Type CHARACTER x CHARACTER (LEN=n) x CHARACTER (LEN=n, KIND=1) x COMPLEX x COMPLEX (KIND=4) x COMPLEX (KIND=8) x DOUBLE PRECISION x REAL x REAL (KIND=4) x REAL (KIND=8) x INTEGER x INTEGER (KIND=1) x INTEGER (KIND=2) x INTEGER (KIND=4) x LOGICAL x LOGICAL (KIND=1) x LOGICAL (KIND=2) x LOGICAL (KIND=4) x

In the current release, with items of type INTEGER for KIND=1,2, or 4:

· · ·

Each uses 4 bytes of storage Each aligns on 4-byte boundaries Each involves 32 bits if any computations are involved

126

Fortran 90 User 's Guide


9
Passing Arguments by Reference or Value
C and Fortran 90 pass arguments using the following different basic rules.

· ·

Fortran 90 generally passes arguments by reference. C passes arguments by value.

Despite this seeming conflict, some compatibility can be achieved, at least for the Fortran 90 pass by reference. The compatibility is possible because a C program can specify that the value being passed is actually an address. For C, this is traditionally described as passing pointers.

Passing Arguments by Reference
Pass by reference in Fortran 90 can be made compatible with pass pointers in C. In C, passing pointers is specified in either of the following ways--one for defining new functions, and another for invoking existing functions.

·

Defining New Functions In a function, where data types of arguments are declared, if you precede a dummy argument by an asterisk (*), C passes a pointer to the item. Example: Define a C function--a dummy argument is a pointer to an int.
void simref1_ ( int * d ) { ... } /* d is a pointer to an int */

C­Fortran Interface

127


9
·
Invoking Existing Functions To make C pass the address of the argument, in the statement that invokes the function do the following: · If the item is not a character string or array, then precede the actual argument by an ampersand (&). Example: An actual argument is a pointer to an int.
int a ; ... simref2_ ( &a ) ...

/* &a is a pointer to an int */

· If the item is a character string or array, then do not precede the actual argument by an ampersand (&). Example: An actual argument is a pointer to a character string.
char s[9] ; ... simref3_ ( s ) ...

/* s is a pointer to a character string */

C always passes arrays and character strings using pointers. In fact, C universally promotes character strings and arrays to pointers. So in C, you cannot pass arrays or character strings by value.

Passing Arguments by Value (N/A)
If a C function passes an argument by value using no pointers, there is no compatible way of passing to Fortran 90. That is, Fortran 90, cannot interface with such a function. Example: Define a C function--dummy argument is an int (no pointer).
simval1_ ( int i ) { ... }

The above function cannot be called from Fortran 90.

128

Fortran 90 User 's Guide


9
Character Strings and Order
Passing strings between C and Fortran is nonstandard. It is not encouraged. The following compatibility rules are provided for those who need to do it, despite the complexities.

Rules for Passing Any Character Strings

· ·

If you make a string in Fortran and pass it to C, you must provide an explicit null terminator because Fortran does not automatically do that, and C expects (requires) it. All C character strings pass using pointers.

Arguments that are Character Strings

·

For a character · Contains the · Is equivalent · Is passed by

argument, Fortran 90 passes/needs an extra argument that: length of the character string to a C long int value

·

The order of arguments is as follows. · A list of the regular arguments · A list of hidden arguments, one for each character string argument The list of hidden arguments comes after the list of regular arguments.

· ·

For Fortran 90 calling C, if Fortran 90 passes a character string argument, the C function can ignore the extra argument or use it. For C calling Fortran 90, C must pass the extra argument because Fortran 90 expects (requires) it.

Example: Fortran string argument, passed by reference--a Fortran call.
CHARACTER*7 s INTEGER b ... CALL sam ( s, b )

C­Fortran Interface

129


9
Example: The above Fortran call is equivalent to the following C call.
char s[7] ; long b ; ... sam_ ( s, &b, 7L ) ;

In the above example: · s is passed by pointer because s is a character string. · b is passed by pointer because we explicitly use an ampersand (&). · 7, the length, is passed by value (without a pointer) as a literal 7 long.

Functions that are Character Strings
The returned character string is passed as two extra initial arguments, in the following order:

· ·

A pointer to the start of the string return value The length of the string return value

Array Indexing and Order
Array Indexing
C arrays always start at zero, but by default, Fortran arrays start at 1. There are two common ways of approaching this.

· ·

You can use the Fortran default, as in the above example. Then the Fortran element b(2) is equivalent to the C element b[1]. You can specify that the Fortran array b starts at 0. as follows.
INTEGER b(0:2)

This way the Fortran element b(1) is equivalent to the C element b[1].

130

Fortran 90 User 's Guide


9
Array Order
Fortran arrays are stored in column-major order, C arrays in row-major order. For one-dimensional arrays, this is no problem. For two-dimensional and higher arrays, switch subscripts in all references and declarations.

Tip
Many users tell us that it gets confusing, say, to triangularize in C and then pass the parts to Fortran. More generally, it may be confusing to do some of the matrix manipulation in C and some in Fortran. So if passing parts of arrays between C and Fortran does not work (or if it is confusing), try passing the whole array to the other language and do all the matrix manipulation there; avoid doing part in C and part in Fortran.

Libraries and Linking with the f90 Command
To get the proper Fortran libraries linked, use the f90 command to pass the.o files on to the linker. This usually shows up as a problem only if a C main calls Fortran. Dynamic linking is encouraged and made easy. Example 1: Use f90 to link.
demo$ f90 -c RetCmplx.f90 demo$ cc -c RetCmplxmain.c demo$ f90 RetCmplx.o RetCmplxmain.o This does the linking demo$ a.out 4.0 4.5 8.0 9.0 demo$ s

Example 2: Use cc to link. This fails. The libraries are not linked.
demo$ f90 -c RetCmplx.f90 demo$ cc RetCmplx.o RetCmplxmain.c ld: Undefined symbol missing routine _ _Fc_mult demo$ s wrong link command

C­Fortran Interface

131


9
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 don't have to know about file descriptors. 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 below.
Table 9-3 Characteristics of Three I/O Systems Fortran Units Standard I/O File Pointers Opened for reading; or Opened for writing; or Opened for both; or Opened for appending See OPEN(3S). Always unformatted, but can be read or written with format-interpreting routines Direct access if the physical file representation is direct access, but can always be read sequentially Character stream Pointers to structures in the user 's address space File Descriptors Opened for reading; or Opened for writing; or Opened for both

Files Open

Opened for reading and writing

Attributes

Formatted or unformatted Direct or sequential

Always unformatted

Access

Direct access if the physical file representation is direct access, but can always be read sequentially Character stream Integers from 0-63

Structure Form

Record Arbitrary nonnegative integers

132

Fortran 90 User 's Guide


9
File Permissions
C programmers traditionally open input files for reading and output files for writing, sometimes for both. In Fortran it's not possible for the system to foresee what use you will make of the file since there's no parameter to the OPEN statement that gives that information. Fortran tries to open a file with the maximum permissions possible, first for both reading and writing then for each separately. This occurs transparently and is of concern only if you try to perform a READ, WRITE, or ENDFILE but you don't have permission. Magnetic tape 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.

C­Fortran Interface

133


9
9.4 Fortran Calls C
Arguments Passed by Reference (f90 Calls C)
Simple Arguments Passed by Reference (f90 Calls C)
The simple arguments are of Fortran 90 data types INTEGER, REAL, DOUBLE PRECISION, or LOGICAL, without pointers, dimensions, or structures. Example: Simple arguments, C function--C arguments as pointers.
SimRef.c void simref_ ( b4, i4, r4, d8 ) /* Simple types, passed by reference, from f90 (f90 calls C)*/ int * b4 ; int * i4 ; float * r4 ; double * d8 ; { *b4 = 1 ; *i4 = 9 ; *r4 = 9.9f ; *d8 = 9.9F ; }

Example: Simple arguments--Fortran default way.
SimRefmain.f90 PROGRAM SimpleRef ! Pass some simple types, by reference, to C (f90 calls C) LOGICAL b4 ! Default kind is 4-byte INTEGER i4 ! Default kind is 4-byte REAL r4 ! Default kind is 4-byte DOUBLE PRECISION d8! This is 8-byte CALL SimRef ( b4, i4, r4, d8 ) WRITE( *, '(L4,I4,F6.1,F6.1)') b4, i4, r4, d8 END PROGRAM SimpleRef

Example: Simple arguments with Fortran and C.
Compile/Link/Execute demo$ demo$ demo$ T demo$ cc -c SimRef.c f90 SimRef.o SimRefmain.f90 a.out 9 9.9 9.9 s

134

Fortran 90 User 's Guide


9
Complex Arguments Passed by Reference (f90 Calls C)
Example: Complex arguments, C function--pointers to structures.
CmplxRef.c void cmplxref_ ( { w w z z } -> -> -> -> r i r i = = = = 6; 7; 8; 9; struct complex { float r, i; } *w struct dcomplex { double r, i; } *z )

Example: Complex arguments, Fortran main program calls C function.
CmplxRefmain.f90 PROGRAM ComplexRef ! Pass complex types, by reference, to C (f90 calls C) INTEGER, PARAMETER :: doublecomplex=8 COMPLEX w COMPLEX (KIND=doublecomplex) z CALL CmplxRef ( w, z ) WRITE(*,*) w WRITE(*,*) z END PROGRAM ComplexRef

Example: Complex arguments.
Compile/Link/Execute demo$ cc -c CmplxRef.c demo$ f90 CmplxRef.o CmplxRefmain.f90 demo$ a.out ( 6.0000000, 7.0000000) ( 8.0000000000000000, 9.0000000000000000) demo$ s

C­Fortran Interface

135


9
Character Arguments Passed by Reference (f90 Calls C)
Passing strings between C and Fortran is nonstandard. It is not encouraged. For Fortran 90 calling C, if Fortran 90 passes a character string argument, it always passes an extra, hidden argument. The C function can ignore these extra arguments or it can use them. For the detailed requirements of passing string arguments, see "Character Strings and Order" on page 129.

Ignoring Extra Arguments for Strings
A C function can ignore the extra arguments, since they are after the list of other arguments. Example: Character arguments--this C function ignores the extra arguments.
StrRefI.c void strrefi_ ( char *a, char *z ) { static char ax[11] = "abcdefghij" ; static char zx[31] = "abcdefghijklmnopqrstuvwxyz" ; strncpy ( a, ax, 10 ) ; strncpy ( z, zx, 26 ) ; }

Example: Character arguments--a Fortran call passes hidden extra arguments.
StrRefImain.f90 PROGRAM StringRefI CHARACTER a*10, z*30 a='' z='' CALL StrRefI( a, z ) WRITE (*, 1) a, z 1 FORMAT("a='", A, "'", /, "z='", A, "'") END PROGRAM StringRefI

Example: Character arguments, Fortran and C, C ignores the extra arguments.
Compile/Link/Execute demo$ cc -c StrRefI.c demo$ f90 StrRefI.o StrRefmain.f90 demo$ a.out s10='abcdefghij' s30='abcdefghijklmnopqrstuvwxyz ' demo$ s

136

Fortran 90 User 's Guide


9
Using Extra Arguments for Strings
A C function can use the extra arguments from Fortran; they are after the list of other arguments, and they are passed without pointers. Example: Character arguments--a C function uses the extra arguments.
StrRefU.c strrefu_ ( char *a, char *z, long L10, long L30 ) { static char ax[11] = "abcdefghij" ; static char zx[31] = "abcdefghijklmnopqrstuvwxyz" ; printf("%d %d \n", L10, L30 ) ; strncpy ( a, ax, 11 ) ; strncpy ( z, zx, 26 ) ; } /* Use L10 and L30, print them */

Example: Character arguments--a Fortran call makes hidden extra arguments.
StrRefUmain.f90 PROGRAM StringRefU CHARACTER a*10, z*30 a='' z='' CALL StrRefU( a, z ) WRITE (*, 1) a, z 1 FORMAT("a='", A, "'", /, "z='", A, "'") END PROGRAM StringRefU

Example: Character arguments, Fortran and C, C uses the extra arguments.
Compile/Link/Execute 10 30 s10='abcdefghij' s30='abcdefghijklmnopqrstuvwxyz

'

The above C function prints the extra arguments (the lengths); what you really do with them is up to you

C­Fortran Interface

137


9
Vector Arguments Passed by Reference (f90 Calls C)
Example: A C one-dimensional array argument, indexed from 0 to 8.
FixVec.c void fixvec_ ( int v[9], int *sum ) { int i; *sum = 0; for ( i = 0; i <= 8; i++ ) *sum = *sum + v[i]; }

Example: A Fortran vector argument, implicitly indexed from 1 to 9.
FixVecmain.f90 PROGRAM FixedVector INTEGER, DIMENSION(9) :: a = (/ 1,2,3,4,5,6,7,8,9 /) INTEGER i, sum CALL FixVec ( a, sum ) WRITE(*, '("a: ", 9I2, ", sum:" I3)') (a(i),i=1,9), sum END PROGRAM FixedVector

Example: A vector argument, Fortran and C, implicitly indexed from 1 to 9.
Compile/Link/Execute demo$ demo$ demo$ a: 1 demo$ cc -c FixVec.c f90 FixVec.o FixVecmain.f90 a.out 2 3 4 5 6 7 8 9, sum: 45 s

Example: A vector argument, Fortran and C, explicitly indexed from 0 to 8.
FixVecmainE.f90 PROGRAM FixedVectorE ! a is explicitly indexed 0 to 8 INTEGER, DIMENSION(0:8) :: a = (/ 1,2,3,4,5,6,7,8,9 /) INTEGER i, sum CALL FixVec ( a, sum ) WRITE(*, '("a: ", 9I2, ", sum:" I3)') (a(i),i=0,8), sum END PROGRAM FixedVectorE

Example: A vector argument, Fortran and C, explicitly indexed from 0 to 8.
Compile/Link/Execute demo$ demo$ demo$ a: 1 demo$ cc -c FixVec.c f90 FixVec.o FixVecmain#.f90 a.out 2 3 4 5 6 7 8 9, sum: 45 s

138

Fortran 90 User 's Guide


9
Matrix Arguments Passed by Reference (f90 Calls C)
In a two-dimensional array, the rows and columns are switched. Example: A 2 by 2 C array argument, indexed from 0 to 1.
FixMat.c fixmat_ ( int a[2][2] ) { a[0][1] = 99 ; /* C changes a[0][1] */ }

Example: A 2 by 2 Fortran array argument, explicitly indexed from 0 to 1.
FixMatmain.f90 PROGRAM FixedMatrix INTEGER c, r INTEGER, DIMENSION(0:1,0:1) :: m m(0,0)=00 ; m(0,1)=01 ; m(1,0)=10 ; m(1,1)=11 DO r = 0, 1 DO c = 0, 1 WRITE(*,'("m(",I1,",",I1,")=",I2.2)') r, c, m(r,c) END DO END DO CALL FixMat ( m ) DO r = 0, 1 DO c = 0, 1 WRITE(*,'("m(",I1,",",I1,")=",I2.2)') r, c, m(r,c) END DO END DO END PROGRAM FixedMatrix

Example: A 2 by 2 array argument--show m before and after the C call.
Compile/Link/Execut e Before demo$ cc -c FixMat.c demo$ f90 FixMat.o FixMatmain.f90 demo$ a.out m(0,0) = 00 m(0,1) = 01 m(1,0) = 10 m(1,1) = 11 m(0,0) = 00 m(0,1) = 01 m(1,0) = 99 Fortran shows that m(1,0) got changed m(1,1) = 11 demo$ s



After

C­Fortran Interface

139


9
Structure Arguments Passed by Reference (f90 Calls C)
Example: A C structure argument--an integer and a character string.
StruRef.c struct InCh { int nbytes ; char a[16] ; }; /* Define a structure */

void struref_ ( v ) /* Use the structure in definining a function */ struct InCh *v ; { bcopy( "oyvay", v->a, 5 ) ; /* Change the char component */ v -> nbytes = 5 ; /* Change the int component */ }

Example: A Fortran 90 structure argument, an integer and a character string.
StruRefmain.f90 PROGRAM StructureRef TYPE IntChr INTEGER n CHARACTER str*15 END TYPE IntChr TYPE(IntChr) vls ! Define the derived type IntChr

! Make vls an item of type IntChr

vls % n = 0 ! Initialize components vls % str = '123456789012345' CALL StruRef ( vls ) ! Change components WRITE ( *, 1 ) vls % n, vls % str ! Print components 1 FORMAT("n =", I2, ", str='", A, "'" ) END PROGRAM StructureRef

Example: A structure argument, Fortran 90 and C.
Compile/Link/Execute demo$ cc -c StruRef.c demo$ f90 StruRef.o StruRefmain.f90 demo$ a.out n = 5, str='oyvay6789012345' demo$ s

Pointer Arguments Passed by Reference (f90 Calls C)--N/A
These two kinds of pointers are not compatible.

140

Fortran 90 User 's Guide


9
Arguments Passed by Value (f90 Calls C)
In general, if an existing C function passes arguments by value using no pointers, then Fortran 90 cannot use that function.

Function Return Values (f90 Calls C)
In general, for a C function return value:

· ·

If it has no pointer, then Fortran 90 can use it as is. If it returns a pointer, then Fortran 90 needs an interface block. See "Pointerto-a-REAL Function Return Value (f90 Calls C)" on page 143.

INTEGER Function Return Value (f90 Calls C)
Example: A C function with an int function return value (not a pointer).
RetInt.c int retint_ ( { int s s = *r s++ ; return } int *r ) ; ; (s);

Example: A Fortran program uses a C function that returns an int
RetIntmain.f90 PROGRAM ReturnInt INTEGER r, s, RetInt r=8 s = 100 + RetInt ( r ) ! The C function is invoked here WRITE( *, "(2I4)") r, s END PROGRAM ReturnInt

Example: Fortran and C with an INTEGER function return value.
Compile/Link/Execute demo$ cc -c RetInt.c demo$ f90 RetInt.o RetIntmain.f90 demo$ a.out 8 109 demo$s

C­Fortran Interface

141


9
REAL Function Return Value (f90 Calls C)
Example: A C function yields a float function return value (not a pointer).
RetFloat.c float retfloat ( float *pf ) { float f ; f = *pf ; f++ ; return ( f ) ; }

Example: A Fortran program uses a C function that returns a float.
RetFloatmain.f90 PROGRAM ReturnFloat REAL RetFloat, r, s r = 8.0 s = 100.0 + RetFloat ( r ) ! The C function is invoked here WRITE(*, '(2F6.1)' ) r, s END PROGRAM ReturnFloat

Example: Fortran and C with a REAL function return value.
Compile/Link/Execute demo$ cc -c RetFloat.c demo$ f90 RetFloat.o RetFloatmain.f demo$ a.out 8.0 109.0 demo$ s

In earlier versions of C, if C returned a function value that was a float, C promoted it to a double, and various tricks were needed to get around that.

142

Fortran 90 User 's Guide


9
Pointer-to-a-REAL Function Return Value (f90 Calls C)
In general, if an existing C function returns a pointer to an item, then Fortran 90 requires an interface block for the function. Example: A C function with a pointer-to-a-float function return value.
RetPtrF.c static float f; float *retptrf ( float *a ) { f = *a ; f++ ; return &f ; }

Example. A Fortran program uses a pointer-to-a-REAL function return value.
RetPtrFmain.f90 PROGRAM ReturnPtrFloat ! Use a C function return value that is a pointer to a real. INTERFACE FUNCTION RetPtrF ( x ) REAL x REAL, POINTER :: RetPtrF END FUNCTION RetPtrF END INTERFACE REAL a, b a = 8.0 b = 100.0 + RetPtrF(a) ! Uses C function here WRITE(*,'(F9.0)') b END PROGRAM ReturnPtrFloat

Example: Fortran and C with a pointer-to-a-REAL function return value.
Compile/Link/Execute demo$ cc -c RetPtrF.c demo$ f90 RetPtrF.o RetPtrFmain.f90 demo$ a.out 109. demo$ s

C­Fortran Interface

143


9
DOUBLE PRECISION Function Return Value (f90 Calls C)
Example: A C function with a double function return value (not a pointer).
RetDbl.c double retdbl_ { double s = *r s++ ; return } ( double *r ) s; ; (s);

Example: Fortran 90 uses a DOUBLE PRECISION function return value from C.
RetDblmain.f90 PROGRAM ReturnDbl DOUBLE PRECISION r, s, RetDbl r = 8.0 s = 100.0 + RetDbl(r) ! The C function is invoked here WRITE( *, "(2F6.1)") r, s END PROGRAM ReturnDbl

Example: Fortran and C with a DOUBLE PRECISION function return value.
Compile/Link/Execute demo$ cc -c RetDbl.c demo$ f90 RetDbl.o RetDblmain.f90 demo$ a.out 8.0 109.0 demo$ s

144

Fortran 90 User 's Guide


9
LOGICAL Function Return Value (f90 Calls C)
Example: A C function with an int function return value (not a pointer).
RetLog.c int retlog_ ( int *r ) { int s; s = *r; if ( s == 0 ) s = 1 ; else s = 0 ; return ( s ); }

Example: A Fortran program uses a C function as if it returns a LOGICAL.
RetLogmain.f90 PROGRAM TryRetLog LOGICAL r, s, RetLog r = .FALSE. s = .TRUE. .AND. RetLog(r) WRITE( *, "(2L4)") r, s END PROGRAM TryRetLog

Example: Fortran and C with a LOGICAL function return value.
Compile/Link/Execute demo$ cc -c RetLog.c demo$ f90 RetLog.o RetLogmain.f90 demo$ a.out F T demo$s

C­Fortran Interface

145


9
CHARACTER Function Return Value (f90 Calls C) N/A
Passing character strings between C and Fortran is not encouraged. See "Character Strings and Order" on page 129, for details of compatibility. Example. A C character string function for Fortran.
RetStr.c The function value is passed not as a "function return value ," but as these argument s: rval_ptr, pointer to string rval_len, length of string The normal string argument is passed as: &ch_ptr, pointer to string ch_len, length of string void retstr_ ( char *rval_ptr, /* pointer to returned string */ int rval_len, /* length of returned string */ char *ch_ptr, /* pointer to string argument */ int *n_ptr, /* pointer to number of copies */ int ch_len ) /* length of string argument */ { /* Return string: n_ptr copies of the character ch_ptr */ int count, i ; char *cp ; count = *n_ptr ; cp = rval_ptr ; for (i=0; i
Example. A Fortran program uses a C CHARACTER function.
RetStrmain.f90 PROGRAM TryRetStr CHARACTER String*16, RetStr*9 String = ' ' String = '1234' // RetStr('*',9) // '456' ! Use C function here WRITE(*,*) "'", String, "'" END PROGRAM TryRetStr

Example: Fortran and C with a character string function.
Compile/Link/Execute demo$ cc -c RetStr.c demo$ f90 RetStr.o RetStrmain.f90 demo$ a.out '1234*********456' demo$ s

146

Fortran 90 User 's Guide


9
Labeled Common (f90 Calls C)
C and Fortran can share values in labeled common. Example: A C function uses labeled common matching the Fortran one below.
UseCom.c extern struct comtype { float p ; float q ; float r ; }; extern struct comtype ilk_ ; void usecom_ () { ilk_.p = 1.0 ; ilk_.q = 2.0 ; ilk_.r = 3.0 ; }

Example: A Fortran main program uses a labeled common.
UseCommain.f90 PROGRAM TryUseCom REAL u, v, w COMMON / ilk / u, v, w u = 7.0 v = 8.0 w = 9.0 WRITE(*,*) u, v, w CALL UseCom ( u, v, w ) WRITE(*,*) u, v, w END PROGRAM TryUseCom

Example: Fortran and C share a labeled common.
Compile/Link/Execute demo$ cc -c CUseCom.c demo$ f90 CUseCom.o FUseCommain.f90 demo$ a.out 7.0000000 8.0000000 9.0000000 1.0000000 2.0000000 3.0000000 demo$ s

Note ­ Any of the options that change size or alignment (or any equivalences that change alignment) might invalidate such sharing.

C­Fortran Interface

147


9
Alternate Returns (f90 Calls C) - N/A
C does not have an alternate return. The work-around is to pass an argument and branch on that.

148

Fortran 90 User 's Guide


9
9.5 C Calls Fortran
Arguments Passed by Reference (C Calls f90)
Simple Arguments Passed by Reference (C Calls f90)
Example: Simple arguments, Fortran arguments by reference.
SimRef.f90 SUBROUTINE SimRef ( b4, i4, r4, d8 ) ! f90 gets passed some simple types, by reference, from C LOGICAL b4 ! Default kind is INTEGER i4 ! Default kind is REAL r4 ! Default kind is DOUBLE PRECISION d8 ! This is 8-byte b4 = .TRUE. i4 = 9 r4 = 9.9 d8 = 9.9 END SUBROUTINE SimRef (C calls f90 ) 4-byte 4-byte 4-byte

Example: Simple arguments, C passes the address of each.
SimRefmain.c void main () { /* Simple types passed by reference to f90 (C calls f90) */ int b4 ; /* f90: 4-byte LOGICAL */ int i4 ; /* f90: 4-byte INTEGER */ float r4 ; /* f90: 4-byte REAL */ double d8 ; /* f90: 8-byte DOUBLE PRECISION */ extern simref_ ( int *, int *, float *, double * ) ; simref_ ( &b4, &i4, &r4, &d8 ) ; printf ( "%08o %d %3.1f %3.1f \n", b4, i4, r4, d8 ) ; }

Example: Simple arguments, C and Fortran.
Compile/Link/Execute demo$ f90 -c SimRef.f90 demo$ cc -c SimRefmain.c demo$ f90 SimRef.o SimRefmain.o demo$ a.out 00000001 9 9.9 9.9 demo$ s

This does the linking

C­Fortran Interface

149


9
Complex Arguments Passed by Reference (C Calls f90)
Example: Complex arguments, Fortran 90 expects a simple structure.
CmplxRef.f90 SUBROUTINE CmplxRef ( w, z ) ! f90 gets passed complex arguments from C (C calls f90) INTEGER, PARAMETER :: doublecomplex=8 COMPLEX w COMPLEX (KIND=doublecomplex) :: z w = ( 6, 7 ) z = ( 8, 9 ) END SUBROUTINE CmplxRef

Example: Complex arguments--C passes pointers to structures.
CmplxRefmain.c main ( ) { struct complex { float r, i ; } ; struct complex d1 ; struct complex *w = &d1 ; struct dcomplex { double r, i ; } ; struct dcomplex d2 ; struct dcomplex *z = &d2 ; extern cmplxref_ ( struct complex *, struct dcomplex * ) ; cmplxref_ ( w, z ) ; /* w and z are pointers */ printf ( "%3.1f %3.1f \n%3.1f %3.1f \n", w->r, w->i, z->r, z->i ) ; }

Example: Complex arguments, C and Fortran.
Compile/Link/Execute demo$ f90 -c CmplxRef.f90 demo$ cc -c CmplxRefmain.c demo$ f90 CmplxRef.o CmplxRefmain.o demo$ a.out 6.0 7.0 8.0 9.0 demo$ s

This does the linking

150

Fortran 90 User 's Guide


9
Character Arguments Passed by Reference (C Calls f90)
Passing strings between C and Fortran is nonstandard. It is not encouraged. For C calling Fortran 90, if C passes a character string argument, C must pass the extra hidden argument because Fortran 90 expects (requires) it. For the detailed requirements of passing string arguments, see "Character Strings and Order" on page 129. Example: Character arguments--Fortran uses extra arguments from C.
StrRefU.f90 SUBROUTINE StrRefU ( a, s ) ! Character arguments -- use extra args passed from C. CHARACTER a*(*), s*(*) a = 'abcdefghi' // char(0) s = 'abcdefghijklmnopqrstuvwxyz' // char(0) END SUBROUTINE StrRefU

Example: Character arguments--C passes extra arguments.
StrRefUmain.c void strrefu ( char *, char *, long, long ) ; /* Declare fcn interface */ void main ( ) /* Pass string arguments to f90 with extra args */ { char s10[10], s80[80] ; /* Provide memory for the strings */ long L10, L80 ; L10 = 10 ; /* Initialize extra args */ L80 = 80 ; strrefu_ ( s10, s80, L10, L80 ) ; /* C strings pass by reference */ printf ( " s10='%s' \n s80='%s' \n", s10, s80 ) ; }

Example: Character arguments--C and Fortran.
Compile/Link/Execute demo$ f90 -c StrRef.f90 demo$ cc -c StrRefmain.c demo$ f90 StrRef.o StrRefmain.o demo$ a.out s10='abcdefghi' s80='abcdefghijklmnopqrstuvwxyz' demo$s

C­Fortran Interface

151


9
Vector Arguments Passed by Reference (C Calls f90)
Example: A Fortran one-dimensional array arg, implicitly indexed from 1 to 9.
VecRef.f90 SUBROUTINE VecRef ( v, total ) ! f90 gets passed a one-dimensional array argument, from C (C calls f90 ) INTEGER i, total, v(9) total = 0 DO i = 1, 9 total = total + v(i) END DO END SUBROUTINE VecRef

Example: A C one-dimensional array argument, indexed from 0 to 8.
VecRecmain.c void vecref_ ( int[ ], int * ) ; /* Declare fcn interface */ void main ( ) { /* A one-dimensional array argument passed to f90 */ int i, sum ; int v[9] = { 1,2,3,4,5,6,7,8,9 } ; vecref_ ( v, &sum ) ; /* Arrays pass by reference */ printf ( " %d \n", sum ) ; }

Example: A one-dimensional array argument, Fortran and C.
Compile/Link/Execute demo$ demo$ demo$ demo$ 45 demo$ f90 -c VecRef.f90 cc -c VecRefmain.c0 f90 VecRef.o VecRefmain.f90 a.out s

152

Fortran 90 User 's Guide


9
Matrix Arguments Passed by Reference (C Calls f90)
Example: A Fortran 2 by 2 array argument, explicitly indexed from 0 to 1.
MatRef.f90 In a two-dimensional array, the rows and columns are switched, comparing C and Fortran. SUBROUTINE MatRef ( a, total ) ! f90 gets passed a two-dimensional array from C (C calls f90 ) INTEGER c, r, total, a(0:1,0:1) a(0,1) = 99 ! Changes a(0,1) total = 0 DO r = 0, 1 ! Sums all of a DO c = 0, 1 total = total + a(r,c) END DO END DO END SUBROUTINE MatRef

Example: A 2 by 2 C array argument, indexed from 0 to 1.
MatRefmain.c void matref_ ( int[ ][2], int * ) ; /* Declare fcn interface */ void main ( ) { /* A C two-dimensional array argument passed to f90 */ int c, r, sum ; int m[2][2] = {{ 00, 01 }, { 10, 11 }} ; for ( c=0; c<2; c++ ) { for ( r=0; r<2; r++ ) printf ( "m(%d,%d)=%#02d \n", c, r, m[c][r] ) ; } matref_ ( m, &sum ) ; /* Arrays pass by reference */ for ( c=0; c<2; c++ ) { for ( r=0; r<2; r++ ) printf ( "m(%d,%d)=%#02d \n", c, r, m[c][r] ) ; } }

Such square arrays are either incompatible for C and Fortran, or awkward to do right, depending on your attitude or needs. Nonsquare arrays are worse.

Example: A 2 by 2 array argument--show m before and after Fortran call.
Compile/Link/Execut e Before m(0,0)=00 m(0,1)=01 m(1,0)=10 m(1,1)=11 m(0,0)=00 m(0,1)=01 m(1,0)=99 m(1,1)=11

After





C shows that m(1,0) got changed.

C­Fortran Interface

153


9
Structure Arguments Passed by Reference (C Calls f90)
Example: A Fortran 90 structure argument--received from C.
StruRef.f90 See also "Complex Arguments Passed by Reference (C Calls f90)" on page 150. SUBROUTINE StruRef ( n ) ! f90 gets passed structure argument from C (C calls f90) TYPE IntReal INTEGER i REAL r END TYPE IntReal TYPE (IntReal) n n%i=8 n % r = 9.0 END SUBROUTINE StruRef

Example: A C structure argument, passed to Fortran 90.
StruRefmain.c struct InRl { int i; float r ; }; /* Define a structure */

void struref_ ( struct InRl * ) ; /* Use structure, define function */ void main ( ) { struct InRl ir ; struct InRl *n = &ir ; n -> i = 1 ; n -> r = 2.0 struref_ printf ( } /* Initialize the structure */ ; (n); /* Uses Fortran routine here */ "n->i=%d, n->r=%3.1f \n", n->i, n->r ) ;

Example: A structure argument, Fortran 90 and C.
Compile/Link/Execute demo$ f90 -c StruRef.f90 demo$ cc -c StruRefmain.c demo$ f90 StruRef.o StruRefmain.o demo$ a.out n->i=8, n->r=9.0 demo$ s

154

Fortran 90 User 's Guide


9
Pointer Arguments Passed by Reference (C Calls f90)--N/A
These two kinds of pointers are not compatible.

Arguments Passed by Value (C Calls f90) - N/A
In general, Fortran 90 cannot pass an argument by value, whether Fortran 90 calls C or C calls Fortran 90. The work-around is to pass all arguments by reference.

Function Return Values (C Calls f90)
For function return values, a Fortran function of type BYTE, INTEGER, LOGICAL, DOUBLE PRECISION, or REAL*16 (quadruple precision) is equivalent to a C function that returns the corresponding type.

INTEGER Function Return Value (C Calls f90)
Example: A Fortran function with an INTEGER function return value.
RetInt.f90 FUNCTION RetInt ( k ) INTEGER k, RetInt RetInt = k + 1 END FUNCTION RetInt

Example: A C program uses a Fortran function that returns an INTEGER.
RetIntmain.c int retint_ ( int * ) ; /* Declare function interface */

void main() { int k, m ; k=8; m = 100 + retint_ ( &k ) ; printf( "%d %d\n", k, m ) ; }

C­Fortran Interface

155


9
Example Fortran and C with an INTEGER function return value.
Compile/Link/Execute demo$ demo$ demo$ demo$ 8 109 demo$ f90 -c RetInt.f cc -c RetIntmain.c f90 RetInt.o RetIntmain.o a.out s

REAL Function Return Value (C Calls f90)
Example: Fortran returns a REAL to a C float.
RetFloat.f90 FUNCTION RetFloat ( x ) REAL x, RetFloat RetFloat = x + 1.0 END FUNCTION RetFloat

Example: A C program uses a Fortran function that returns a REAL.
RetFloatmain.c float retfloat_ ( float * ) ; main ( ) { float r, s ; r = 8.0 ; s = 100.0 + retfloat_ ( &r ) ; printf( " %8.6f %8.6f \n", r, s ) ; } /* Declare function interface */

Example Fortran and C with a REAL function return value.
Compile/Link/Execute demo$ f90 -c RetFloat.f demo$ cc -c RetFloatmain.c demo$ f90 RetFloat.o RetFloatmain.o demo$ a.out 8.000000 109.000000 demo$ s

In earlier versions of C, if C returned a function value that was a float, C promoted it to a double, and various tricks were needed to get around that.

156

Fortran 90 User 's Guide


9
DOUBLE PRECISION Function Return Value (C Calls f90)
Example: Fortran function with a DOUBLE PRECISION function return value.
RetDbl.f90 FUNCTION RetDbl ( x ) DOUBLE PRECISION RetDbl, x RetDbl = x + 1.0 END

Example: A C main uses a Fortran function that returns a DOUBLE PRECISION.
RetDblmain.c double retdbl_ ( double * ) ; /* Declare function interface */ main() { double x, y ; x = 8.0 ; y = 100.0 + retdbl_ ( &x ) ; printf( "%8.6f %8.6f\n", x, y ) ; }

Example Fortran and C with a DOUBLE PRECISION function return value.
Compile/Link/Execute demo$ f90 -c RetDbl.f demo$ cc -c RetDblmain.c demo$ f90 RetDbl.o RetDblmain.o demo$ a.out 8.000000 109.000000 demo$ s

C­Fortran Interface

157


9
LOGICAL Function Return Value (C Calls f90)
Example: A Fortran function with a LOGICAL function return value.
RetLog.f90 FUNCTION RetLog ( b ) LOGICAL b, RetLog RetLog = .NOT. b END FUNCTION RetLog

Example: A C program uses a Fortran function that returns a LOGICAL.
RetLogmain.c int retlog_ ( int * ) ; /* Declare function interface */

void main() { int r, s ; r=0; s = retlog_ ( &r ) ; printf( "%d %d\n", r, s ) ; }

Example: Fortran and C with a LOGICAL function return value.
Compile/Link/Execute demo$ f90 -c RetLog.f90 demo$ cc -c RetLogmain.c demo$ f90 RetLog.o RetLogmain.o demo$ a.out 01 demo$s

158

Fortran 90 User 's Guide


9
CHARACTER Function Return Value (C Calls f90)
Passing strings between C and Fortran is not encouraged. See "Character Strings and Order" on page 129, for details of compatibility. Example: A Fortran character string function.
RetChr.f90 FUNCTION RetChr( c, n ) CHARACTER RetChr*(*), c RetChr = '' DO i = 1, n RetChr(i:i) = c END DO RetChr(n+1:n+1) = CHAR(0) ! Put in the null terminator for C END FUNCTION RetChr

Example: A C main uses a Fortran character function.
RetChrmain.c The function value is passed not as a "function return value ," but as these argument s: rval_ptr, pointer to string rval_len, length of string The normal string argument is passed as: &ch, pointer to string ch_len, length of string void retchr_ (char *, int , char *, int *, int) ; /* fcn interface */ main() { /* Use a Fortran 90 character function,(C calls f90) */ char strbuffer[9] = "123456789" ; char *rval_ptr = strbuffer ; /* int rval_len = sizeof(strbuffer) ; /* char ch = '*' ; /* int n = 4 ; /* int ch_len = sizeof(ch) ; /* printf( " '%s'\n", strbuffer ) ; retchr_ ( rval_ptr, rval_len, &ch, printf( " '%s'\n", strbuffer ) ; }

extra initial arg 1 */ extra initial arg 2 */ for normal arg 1 */ for normal arg 1 */ extra final arg */ &n, ch_len ) ;

Example: C and Fortran with a character string function.
Compile/Link/Execute demo$ f90 -c RetChr.f90 demo$ cc -c RetChrmain.c demo$ f90 RetChr.o RetChrmain.o demo$ a.out '123456789' '****' demo$ s

C­Fortran Interface

159


9
Labeled Common (C Calls f90)
C and Fortran can share values in labeled common. Example: A Fortran subroutine uses a labeled common.
UseCom.f90 SUBROUTINE UseCom REAL u, v, w COMMON / ilk / u, v, w u = 7.0 v = 8.0 w = 9.0 END SUBROUTINE UseCom

Example: A C main uses a labeled common matching the Fortran one above.
UseCommain.c extern struct comtype {/* Declare a structure */ float p ; float q ; float r ; }; extern struct comtype ilk_ ; /* Define an item using the structure */ void usecom_ ( ) ; /* Declare function interface */ void main() { ilk_.p = 1.0 ; ilk_.q = 2.0 ; ilk_.r = 3.0 ; usecom_ ( ) ; printf(" ilk_.p=%4.1f, ilk_.q=%4.1f, ilk_.r=%4.1f\n", ilk_.p, ilk_.q, ilk_.r ) ; }

Example: Fortran and C share a labeled common.
Compile/Link/Execute demo$ f90 -c UseCom.f90 demo$ cc -c UseCommain.c demo$ f90 UseCom.o UseCommain.o demo$ a.out ilk_.p= 7.0, ilk_.q= 8.0, ilk_.r= 9.0 demo$ s

Note ­ Any of the options that change size or alignment (or any equivalences that change alignment) might invalidate such sharing.

160

Fortran 90 User 's Guide


9
Alternate Returns (C Calls f90)
Some C programs may need to use a Fortran routine with nonstandard returns. No new Fortran 90 routine needs alternate returns--they are obsolete. Example: One regular argument and two alternate returns.
AltRet.f90 SUBROUTINE AltRet ( i, *, * ) ! Obsolete INTEGER i, k i=9 k = 20 IF ( k .eq. 10 ) RETURN 1 ! Obsolete IF ( k .eq. 20 ) RETURN 2 ! Obsolete RETURN END SUBROUTINE AltRet

Obsoletefeatures are candidates for removal from the next version of the standard. Fortran 90 has better ways of doing the same thing.

Example: Alternate returns--C invokes the subroutine as a function.
AltRetmain.c To C, a Fortran routine with nonstandard returns does return an int (INTEGER*4). The return value specifies which alternate return was used. If the routine has no entry points with alternate return arguments, the returned value is undefined. int altret_ ( int * ) ; /* Declare function interface */ main() { int k, m ; k=0; m = altret_ ( &k ) ; /* Use the Fortran subroutine */ printf( "%d %d\n", k, m ) ; }

Example: Alternate returns, C and Fortran.
Compile/Link/Execute demo$ f90 -c AltRet.f90 Obsolescent: The alternate RETURN construct is obsolete at line 6 Obsolescent: The alternate RETURN construct is obsolete at line 7 demo$ acc -c AltRetmain.c demo$ f90 AltRet.o AltRetmain.o demo$ a.out 92 demo$

In this example, the C main receives a 2 as the return value of the subroutine because the "RETURN 2" was executed.

C­Fortran Interface

161


9

162

Fortran 90 User 's Guide


Features and Differences
This appendix is organized into the following sections.
Standards Extensions Directives Compatibility with FORTRAN 77 Forward Compatibility Mixing Languages Module Files

A

page 163 page 164 page 179 page 184 page 188 page 188 page 188

This appendix shows some of the major features differences between:

· ·

Standard Fortran 90 and Sun Fortran 90 FORTRAN 77 and Fortran 90

A.1 Standards
This Fortran is an enhanced ANSI Standard Fortran 90 development system.

· ·

It conforms to the ANSI X3.198-1992 Fortran standard and the corresponding International Standards Organization ISO/IEC 1539:1991 (E) Fortran 90 standard. It provides an IEEE standard 754-1985 floating-point package.

163


A
·
On SPARC systems, it provides support for optimization exploiting features of SPARC V8, including the SuperSPARCTM implementation1. These features are defined in the SPARC Architecture Manual: Version 8.

A.2 Extensions
Sun Fortran 90 provides the following extensions.

Tabs in the Source
f90 allows the tab character in fixed-form source and in free-form source. Standard Fortran 90 does not allow tabs. The tab character is not converted to a blank, so the visual placement of tabbed statements depends on the utility you use to edit or display text.

Fixed-Form Source

·

For a tab in column one: If the next character is a nonzero digit, then the current line is a continuation line; otherwise, the current line is an initial line.

· ·

A tab cannot precede a statement label. A tab after column one is treated by f90 the same as a blank character, except in literal strings.

Free-Form Source
f90 treats a tab and a blank character as equivalent, except in literal strings.

1. SuperSPARC is a trademark of Texas Instruments, Inc.

164

Fortran 90 User 's Guide


A
Continuation Line Limits
f90 allows 99 continuation lines (1 initial and 98 continuation lines). Standard Fortran 90 allows 19 for fixed-form and 39 for free-form.

Fixed-Form Source of 96 Characters
In fixed-form source, lines can be 96 characters long. Columns 73 through 96 are ignored. Standard Fortran 90 allows 72-character lines.

Directives
f90 allows directive lines starting with CDIR$, !DIR$, CMIC$, or !MIC$. They look like comments but are not. For full details on directives, see "Directives" on page 179. Standard Fortran 90 has no directives.

Source Form Assumed
The source form assumed by f90 depends on options, directives, and suffixes.

·

Command-line options
Option -fixed -free Action Interpret all source files as Fortran 90 fixed form Interpret all source files as Fortran 90 free form

If the -free or -fixed option is used, that overrides the file name suffix.

Features and Differences

165


A
·
File name suffixes
Suffix .f90 .f Source Form Fortran 90 free-form source files Fortran 90 fixed-form source files or ANSI standard FORTRAN 77 source files .for .ftn other Same as .f. Same as .f. None--file name is passed to the linker

·

Directives
Directive !DIR$ FIXED !DIR$ FREE Action Interpret the rest of the source file as Fortran 90 fixed form Interpret the rest of the source file as Fortran 90 free form

If either a FREE or FIXED directive is used, that overrides the option and file name suffix.

Mixing Forms
Some mixing of source forms is allowed.

· ·

In the same f90 command, some source files can be fixed form, some free. In the same file, free form can be mixed with fixed form by using directives.

166

Fortran 90 User 's Guide


A
Boolean Type
f90 supports constants and expressions of Boolean type. There are no Boolean variables or arrays, and there is no Boolean type statement.

Miscellaneous Rules Governing Boolean Type

· ·

Masking--A bitwise logical expression has a Boolean result; each of its bits is the result of one or more logical operations on the corresponding bits of the operands. For binary arithmetic operators, and for relational operators: · If one operand is Boolean, the operation is performed with no conversion. · If both operands are Boolean, the operation is performed as if they were integers. No user-specified function can generate a Boolean result, although some (nonstandard) intrinsics can. Boolean and logical types differ as follows: · Variables, arrays, and functions can be of logical type, but they cannot be Boolean type. · There is a LOGICAL statement, but no BOOLEAN statement. · A logical variable or constant represents only one value. A Boolean constant can represent as many as 32 values. · A logical expression yields one value. A Boolean expression can yield as many as 32 values. · Logical entities are invalid in arithmetic, relational, or bitwise logical expressions. Boolean entities are valid in all three.

· ·

Features and Differences

167


A
Alternate Forms of Boolean Constants
f90 allows a Boolean constant (octal, hexadecimal, or Hollerith) in the following alternate forms (no binary). Variables cannot be declared Boolean. Standard Fortran 90 does not allow these forms.

Octal
ddddddB, where d is any octal digit

· · · · · · ·

You can use the letter B or b. There can be 1 to 11 octal digits (0 through 7). 11 octal digits represent a full 32-bit word, with the leftmost digit allowed to be 0, 1, 2, or 3. Each octal digit specifies three bit values. The last (rightmost) digit specifies the content of the rightmost three bit positions (bits 29, 30, and 31). If less than 11 digits are present, the value is right-justified--it represents the rightmost bits of a word: bits n through 31. The other bits are 0. Blanks are ignored.

Within an I/O format specification, the letter B indicates binary digits; elsewhere it indicates octal digits.

Hexadecimal
X'ddd' or X"ddd", where d is any hexadecimal digit

· · · · · · ·

There can be 1 to 8 hexadecimal digits (0 through 9, A-F). Any of the letters can be uppercase or lowercase (X, x, A-F, a-f). The digits must be enclosed in either apostrophes or quotes. Blanks are ignored. The hexadecimal digits may be preceded by a + or - sign. 8 hexadecimal digits represent a full 32-bit word and the binary equivalents correspond to the contents of each bit position in the 32-bit word. If less than 8 digits are present, the value is right-justified--it represents the rightmost bits of a word: bits n through 31. The other bits are 0.

168

Fortran 90 User 's Guide


A
Hollerith
nH... nL... nR...
'...'H '...'L '...'R "..."H "..."L "..."R

Above, "..." is a string of characters and n is the character count.

· · ·

A Hollerith constant is type Boolean. If any character constant is in a bitwise logical expression, the expression is evaluated as Hollerith. A Hollerith constant can have 1 to 4 characters.

Examples: Octal and hexadecimal constants.
Boolean Constant 0B 77740B X"ABE" X"-340" X'1 2 3' X'FFFFFFFFFFFFFFFF' Internal Octal for 32-bit word 00000000000 00000077740 00000005276 37777776300 00000000443 37777777777

Examples: Octal and hexadecimal in assignment statements.
i = 1357B j = X"28FF" k = X'-5A'

Alternate Contexts of Boolean Constants
f90 allows BOZ constants in the places other than DATA statements. B'bbb' B"bbb" O'ooo' O"ooo" Z'zzz' Z"zzz"

If these are assigned to a real variable, no type conversion occurs. Standard Fortran 90 allows these only in DATA statements.

Features and Differences

169


A
Abbreviated Size Notation for Numeric Data Types
f90 allows the following nonstandard type declaration forms in declaration statements, function statements, and IMPLICIT statements.
Table A-1 Size Notation for Numeric Data Types

Nonstandard
INTEGER*1 INTEGER*2 INTEGER*4 LOGICAL*1 LOGICAL*2 LOGICAL*4 REAL*4 REAL*8 COMPLEX*8 COMPLEX*16

Declarator

Short Form

Meaning
One-byte signed integers Two-byte signed integers Four-byte signed integers One-byte logicals Two-byte logicals Four-byte logicals IEEE single-precision floating-point (Four-byte) IEEE double-precision floating-point (Eight-byte) Single-precision complex (Four-bytes each part) Double-precision complex (Eight-bytes each part)

INTEGER(KIND=1) INTEGER(1) INTEGER(KIND=2) INTEGER(2) INTEGER(KIND=4) INTEGER(4) LOGICAL(KIND=1) LOGICAL(1) LOGICAL(KIND=2) LOGICAL(2) LOGICAL(KIND=4) LOGICAL(4) REAL(KIND=4) REAL(KIND=8) REAL(4) REAL(8)

COMPLEX(KIND=4) COMPLEX(4) COMPLEX(KIND=8) COMPLEX(8)

The form in column one is nonstandard. The kind number can vary by vendor.

170

Fortran 90 User 's Guide


A
Cray Pointers
A Cray pointer is a variable whose value is the address of another entity, which is called the pointee. f90 supports Cray pointers. Standard Fortran 90 does not support them.

Syntax
The Cray POINTER statement has the following format:
POINTER ( pointer_name, pointee_name [array_spec] ), ...

Where pointer_name, pointee_name, and array_spec are as follows: pointer_name
Pointer to the corresponding pointee_name. pointer_name contains the address of pointee_name. Must be: a scalar variable name (but not a structure) Cannot be: a constant, a name of a structure, an array, or a function Pointee of the corresponding pointer_name Must be: a variable name, array declarator, or array name If array_spec is present, it must be explicit shape, (constant or nonconstant bounds), or assumed-size.

pointee_name array_spec

Example: Declare Cray pointers to two pointees.
POINTER ( p, b ), ( q, c )

The above example declares Cray pointer p and its pointee b, and Cray pointer q and its pointee c. Example: Declare a Cray pointer to an array.
POINTER ( ix, x(n, 0:m) )

The above example declares Cray pointer ix and its pointee x; and declares x to be an array of dimensions n by m-1.

Features and Differences

171


A
Purpose of Cray Pointers
You can use pointers to access user-managed storage by dynamically associating variables to particular locations in a block of storage. Cray pointers allow accessing absolute memory locations. Cray pointers do not provide convenient manipulation of linked lists because (for optimization purposes) it is assumed that no two pointers have the same value.

Cray Pointers and Fortran 90 Pointers
Cray pointers are declared as follows: POINTER ( pointer_name, pointee_name [array_spec] ) Fortran 90 pointers are declared as follows: POINTER :: object_name The two kinds of pointers cannot be mixed.

Features of Cray Pointers

· · · · · · ·

Whenever the pointee is referenced, f90 uses the current value of the pointer as the address of the pointee. The Cray pointer type statement declares both the pointer and the pointee. The Cray pointer is of type Cray pointer. The value of a Cray pointer occupies one storage unit. Its range of values depends on the size of memory for the machine in use. The Cray pointer can appear in a COMMON list or as a dummy argument. The Cray pointee has no address until the value of the Cray pointer is defined. If an array is named as a pointee, it is called a pointee array. Its array declarator can appear in: · A separate type statement · A separate DIMENSION statement · The pointer statement itself

172

Fortran 90 User 's Guide


A
· ·
If the array declarator is in a subprogram, the dimensioning can refer to: · Variables in a common block, or · Variables that are dummy arguments The size of each dimension is evaluated on entrance to the subprogram, not when the pointee is referenced.

Restrictions on Cray Pointers

· · · ·

If pointee_name is of character type, it must be a variable typed CHARACTER*(*). If pointee_name is an array declarator, it must be explicit shape, (constant or nonconstant bounds), or assumed-size. An array of Cray pointers is not allowed. A · · · Cray pointer cannot be: Pointed to by another Cray pointer or by a Fortran 90 pointer. A component of a structure. Declared to be any other data type.

·

A Cray pointer cannot appear in: · A PARAMETER statement or in a type declaration statement that includes the PARAMETER attribute. · A DATA statement.

Restrictions on Cray Pointees

· · · · ·

A Cray pointee cannot appear in a SAVE, DATA, EQUIVALENCE, COMMON, or PARAMETER statement. A Cray pointee cannot be a dummy argument. A Cray pointee cannot be a function value. A Cray pointee cannot be a structure or a structure component. A Cray pointee cannot be of a derived type.

Note ­ Cray pointees can be of type character, but their Cray pointers are different from other Cray pointers. The two kinds cannot be mixed in the same expression.

Features and Differences

173


A
Usage of Cray Pointers
Cray pointers can be assigned values as follows:

· · · ·

Set to an absolute address Example: q = 0 Assigned to or from integer variables, plus or minus expressions Example: p = q + 100 Cray pointers are not integers. You cannot assign them to a real variable. The LOC function (nonstandard) can be used to define a Cray pointer. Example: p = LOC( x )

Example: Use Cray pointers as described above.
SUBROUTINE sub ( n ) COMMON pool(100000) INTEGER blk(128), word64 REAL a(1000), b(n), c(100000-n-1000) POINTER ( pblk, blk ), (ia, a ), ( ib, b ), & ( ic, c ), ( address, word64 ) DATA address / 64 / pblk = 0 ia = LOC( pool ) ib = ia + 1000 ic = ib + n ...

Remarks about the above example:

· · · · · · ·

word64 refers to the contents of absolute address 64 blk is an array that occupies the first 128 words of memory a is an array of length 1000 located in blank common b follows a and is of length n c follows b a, b, and c are associated with pool word64 is the same as blk(17) because Cray pointers are byte address and the integer elements of blk are each 4 bytes long

174

Fortran 90 User 's Guide


A
Optimization and Cray Pointers
For purposes of optimization, f90 assumes the storage of a pointee is never overlaid on the storage of another variable--it assumes that a pointee is not associated with another variable. Such association could occur in either of two ways:

· ·

A Cray pointer has two pointees, or Two Cray pointers are given the same value

Note ­ You are responsible for preventing such association. These kinds of association are sometimes done deliberately, such as for equivalencing arrays, but then results can differ depending on whether optimization is turned on or off. Example: b and c have the same pointer.
POINTER REAL x, p = LOC( b = 1.0 c = 2.0 PRINT *, ... ( p, b ), b, c x) ( p, c )

b

Above, because b and c have the same pointer, assigning 2.0 to c gives the same value to b. Therefore b prints out as 2.0, even though it was assigned 1.0.

Features and Differences

175


A
Cray Character Pointers
If a pointee is declared as a character type, its Cray pointer is a Cray character pointer.

Purpose of Cray Character Pointers
A Cray character pointer is a special data type that allows f90 to maintain character strings by keeping track of the following:

· · ·

Byte address of the first character of the string Length Offset

An assignment to a Cray character pointer alters all three. That is, when you change what it points to, all three change.

Declaration of Cray Character Pointers
For a pointee that has been declared with an assumed length character type, the Cray pointer declaration statement declares the pointer to be a Cray character pointer. 1. Before the Cray pointer declaration statement, declare the pointee as a character type with an assumed length. 2. Declare a Cray pointer to that pointee. 3. Assign a value to the Cray character pointer. You can use functions CLOC or FCD, both nonstandard intrinsics. Example: Declare Ccp to be a Cray character pointer and use CLOC to make it point to character string s.
CHARACTER*(*) POINTER ( Ccp, CHARACTER*80 Ccp = CLOC( s a a) :: s = "abcdefgskooterwxyz" )

176

Fortran 90 User 's Guide


A
Operations on Cray Character Pointers
You can do the following operations with Cray character pointers:
Ccp1 + i Ccp1 - i i + Ccp1 Ccp1 = Ccp2 Ccp1 relational_operator Ccp2

where Ccp1 and Ccp2 are Cray character pointers and i is an integer.

Restrictions on Cray Character Pointers and Pointees
All restrictions to Cray pointers also apply to Cray character pointers. In addition, the following apply:

· · · · · · ·

A Cray character pointee cannot be an array. In a relational operation, a Cray character pointer can be mixed with only another Cray character pointer--not with a Cray pointer, not with an integer. A relational operation applies only to the character address and the bit offset; the length field is not involved. Cray character pointers must not appear in EQUIVALENCE statements, or any storage association statements. (The size can vary with the platform.) Cray character pointers are not optimized. Code containing Cray character pointers is not parallelized. A Cray character pointer in a list of an I/O statement is treated as an integer.

Features and Differences

177


A
Intrinsics
f90 supports some intrinsic procedures which are extensions beyond the standard.
Table A-2 Nonstandard Intrinsics Type Name CLOC Definition Get Fortran character descriptor (FCD) Cotangent Positive difference Function Cray character pointer real double precision Arguments character Arguments ([C=]c) Remark Notes NP, I

COT DDIM FCD

real double precision i: integer or Cray pointer j: integer

([X=]x) ([X=]x,[Y=]y) ([I=]i,[J=]j) i: word address of first character j: character length

P, E P, E NP, I

Create Cray character pointer Cray in Fortran character pointer descriptor (FCD) format

LEADZ POPCNT POPPAR

Get the number of leading 0 bits Get the number of set bits Calculate bit population parity

integer integer integer

Boolean, integer, real, or pointer Boolean, integer, real, or pointer Boolean, integer, real, or pointer

([I=]i) ([I=]i) ([X=]x)

NP, I NP, I NP, I

The notes in the above table are explained as follows:
Note P NP E I Meaning The name can be passed as an argument. The name cannot be passed as an argument. External code for the intrinsic is called at run time. f90 generates inline code for the intrinsic procedure.

178

Fortran 90 User 's Guide


A
A.3 Directives
A compiler directive directs the compiler to do some special action. Directives are also called pragmas. A compiler directive is inserted into the source program as one or more lines of text. Each line looks like a comment, but has additional characters that identify it as more than a comment for this compiler. For most other compilers, it is treated as a comment, so there is some code portability.

General Directives
Currently there are only two general directives, FREE and FIXED. These directives tell the compiler to assume free-form source or fixed-form source.

Other General Directives
Some other parallel directives are included which are not described in detail because they are not guaranteed to be in the next release.
Table A-3 General Directives Guaranteed Only in the Current Release Directive TASK, NOTASK SUPPRESS( var1, var2, ... ) TASKCOMMON( cb1, cb2, ... )

Features and Differences

179


A
Form of General Directive Lines
General directives have the following syntax.
!DIR$ d1, d2, ...

A general directive line is defined as follows.

· · ·

A directive line starts with the 5 characters CDIR$ or !DIR$, followed by: · A space · A directive Spaces before, after, or within a directive are ignored. Letters of a directive line can be in uppercase, lowercase, or mixed.

The form varies for fixed-form and free-form source as follows.

Fixed-Form Source

· · · · ·

Put CDIR$ or !DIR$ in columns 1 through 5. Directives are listed in columns 7 and beyond. Columns beyond 72 are ignored. An initial directive line has a blank in column 6. A continuation directive line has a nonblank in column 6.

Free-Form Source

· · · ·

Put !DIR$ followed by a space anywhere in the line. The !DIR$ characters are the first nonblank characters in the line (actually, non-whitespace). Directives are listed after the space. An initial directive line has a blank, tab, or newline in the position immediately after the !DIR$. A continuation directive line has a character other than a blank, tab, or newline in the position immediately after the !DIR$.

Thus, !DIR$ in columns 1 through 5 works for both free-form source and fixed-form source.

180

Fortran 90 User 's Guide


A
FIXED and FREE Directives
These directives specify the source form of lines following the directive line.

Scope
They apply to the rest of the file in which they appear, or until the next FREE or FIXED directive is encountered.

Uses

· ·

They allow you to switch source forms within a source file. They allow you to switch source forms for an INCLUDE file. You insert the directive at the start of the INCLUDE file. After the INCLUDE file has been processed, the source form reverts back to the form being used prior to processing the INCLUDE file.

Restrictions
The FREE/FIXED directives:

· ·

Each must appear alone on a compiler directive line (not continued). Each can appear anywhere in your source code. Other directives must appear within the program unit they affect.

Example: A FREE directive.
!DIR$ FREE DO i = 1, n a(i) = b(i) * c(i) END DO

Features and Differences

181


A
Parallel Directives
A parallel directive is a special comment that directs the compiler to do some parallelizing. Currently there is only one parallel directive, DOALL.

DOALL Directive
The DOALL directive tells the compiler to parallelize the next loop it finds, if possible.

Other Parallel Directives
Some other parallel directives are included which are not described in detail because they are not guaranteed to be in the next release.
Table A-4 Parallel Directives Guaranteed Only in the Current Release Directive CASE, END CASE PARALLEL, END PARALLEL DO PARALLEL, END DO GUARD, END GUARD

Form of Parallel Directive Lines
Parallel directives have the following syntax.
!MIC$ DOALL [general parameters] [scheduling parameter]

A parallel directive line is defined as follows.

·

A · · ·

parallel directive starts with the CMIC$ or !MIC$, followed by: A space A directive For some directives, one or more parameters

· ·

Spaces before, after, or within a directive are ignored. Letters of a parallel directive line can be in uppercase, lowercase, or mixed.

182

Fortran 90 User 's Guide


A
The form varies for fixed-form and free-form source as follows.

Fixed

· · · · ·

Put CMIC$ or !MIC$ in columns 1 through 5. Directives are listed in columns 7 and beyond. Columns beyond 72 are ignored. An initial directive line has a blank in column 6. A continuation directive line has a nonblank in column 6.

Free

· · · ·

Put !MIC$ followed by a space anywhere in the line. The !MIC$ characters are the first nonblank characters in the line (actually, non-whitespace). Directives are listed after the space. An initial directive line has a blank, tab, or newline in the position immediately after the !MIC$. A continuation directive line has a character other than a blank, tab, or newline in the position immediately after the !MIC$.

Thus, !MIC$ in columns 1 through 5 works for both free and fixed. Example: Directive with continuation lines (DOALL directive and parameters.)
!MIC$ DOALL !MIC$& SHARED( a, b, c, n ) !MIC$& PRIVATE( i ) DO i = 1, n a(i) = b(i) * c(i) END DO

Example: Same directive and parameters, with no continuation lines.
!MIC$ DOALL SHARED( a, b, c, n ) DO i = 1, n a(i) = b(i) * c(i) END DO PRIVATE( i )

Features and Differences

183


A
A.4 Compatibility with FORTRAN 77
Source
Source from Sun FORTRAN 77 is not generally compatible with Fortran 90, unless it strictly follows the FORTRAN 77 standard. In general, i f it uses no extensions, then it is compatible.

Executables
Libraries compiled and linked in FORTRAN 77 under Solaris 2.x run in the Fortran 90 1.0 environment.

Libraries ·
Libraries (.a) and object files (.o) compiled and linked in FORTRAN 77 under Solaris 2.x are compatible with Fortran 90 1.0. You can check the /usr/4lib directory on your SunOS 5.x system for the libF77.so.2.0 and libV77.so.2.0 library files. Example: f90 main and f77 subroutine.
demo$ cat m.f90 CHARACTER*74 :: c = 'This is a test.' CALL echo1( c ) END demo$ cat s.f SUBROUTINE echo1( a ) CHARACTER*74 a PRINT*, a RETURN END demo$ f77 -c -silent s.f demo$ f90 m.f90 s.o demo$ a.out This is a test. demo$ s

184

Fortran 90 User 's Guide


A
·
The library libF77 is generally compatible with f90. Example: f90 main calls a routine from the libF77 library.
demo$ cat tdtime.f90 REAL e, dtime, t(2) e = dtime( t ) DO i = 1, 10000 k = k+1 END DO e = dtime( t ) PRINT *, 'elapsed:', e, ', user:', t(1), ', sys:', t(2) END demo$ f90 tdtime.f90 demo$ a.out elapsed:6.405999884E-3, user:5.943499971E-3, sys:4.625000001E-4 demo$ s

See dtime(3f).

I/O
f77 and f90 are generally I/O compatible for binary I/O, since f90 loads the f77 I/O compatibility library. Such compatibility includes the following two situations:

· · · ·

In the same program, you can write some records in f90, then read them in f77. An f90 program can write a file. Then an f77 program can read it.

The numbers read back in may or may not equal the numbers written out. Unformatted The numbers read back in do equal the numbers written out. Floating-point formatted The numbers read back in can be different from the numbers written out. This is caused by slightly different base conversion routines, or by different conventions for uppercase/lowercase, spaces, plus or minus signs, and so forth. Examples: 1.0e12, 1.0E12, 1.0E+12

Features and Differences

185


A
·
List-directed The numbers read back in can be different from the numbers written out. This can be caused by various layout conventions with commas, spaces, zeros, repeat factors, and so forth. Example: '0.0' as compared to '.0' Example: ' 7' as compared to '7' Example: '3, 4, 5' as compared to '3 4 5' Example: '3*0' as compared to '0 0 0' The above results are from: integer::v(3)=(/0,0,0/); print *,v Example: '0.333333343' as compared to '0.333333' The above results are from PRINT *, 1.0/3.0

186

Fortran 90 User 's Guide


A
Intrinsics
The Fortran 90 standard supports the following new intrinsic functions that FORTRAN 77 does not have. If you use one of these names in your program, you must add an EXTERNAL statement to make f90 use your function rather than the intrinsic one.
ADJUSTL ADJUSTR ALLOCATED ASSOCIATED BIT_SIZE DIGITS EPSILON EXPONENT FRACTION HUGE KIND LBOUND LEN_TRIM MAXEPONENT MINEXPONENT NEAREST PRECISION PRESENT RADIX RANGE REPEAT RRSPACING SCALE SCAN SELECTED_INT_KIND SELECTED_REAL_KIND SET_EXPONENT SHAPE SIZE SPACING TINY TRANSFER TRIM UBOUND VERIFY

The Fortran 90 standard supports the following new array intrinsic functions.
ALL ANY COUNT CSHIFT DOT_PRODUCT EOSHIFT MATMUL MAXLOC MAXVAL MERGE MINLOC MINVAL PACK PRODUCT RESHAPE SPREAD SUM TRANSPOSE UNPACK

Features and Differences

187


A
A.5 Forward Compatibility
This next release of f90 is intended to be source code compatible with this release. If you generate any libraries with this release of f90, they are not guaranteed to be compatible with the next release.

A.6 Mixing Languages
On Solaris systems, routines written in C can be combined with Fortran 90 programs, since these languages have common calling conventions.

A.7 Module Files
If a file containing a Fortran 90 module is compiled, f90 generates a module file (.M file) in addition to the .o file. By default, such files are usually sought in the current working directory. The -Mdir option allows you to tell f90 to seek them in an additional location. The .M files cannot be stored into an archive file. If you have many .M files in some directory, and you want to reduce the number of such files (to reduce clutter), you can concatenate them into one large .M file.

188

Fortran 90 User 's Guide


iMPact: Multiple Processors
This appendix is organized into the following sections.
Requirements Overview Speed Gained or Lost Number of Processors

B

page 189 page 190 page 192 page 192

This appendix introduces ways to spread a set of programming instructions over a multiple-processor system so they execute in parallel. The process is called parallelizing. The goal is speed. It is assumed that you are familiar with parallel processing and with Sun Fortran and the SunOS or UNIX operating system.

B.1 Requirements
Multiprocessor Fortran 90 requires the following.

· ·

SPARC multiple processor system Solaris 2.3 Operating Environment, or later Solaris 2.3 and later supports the libthread multi-thread library and running many processors simultaneously. Fortran 90 MP has features that exploit multiple processors using Solaris 2.3 and later.

·

The iMPact MT/MP multiple processor package

189


B
B.2 Overview
In general, this compiler can parallelize certain kinds of loops that include arrays. You can let the compiler determine which loops to parallelize (automatic parallelizing) or you can specify each loop yourself (explicit parallelizing).

Automatic Parallelization
Automatic parallelization is both fast and safe. To automatically parallelize loops, use the -parallel option. With this option, the software determines which loops are appropriate to parallelize. Example: Automatic parallelization. Do all appropriate loops.
demo$ f90 -parallel any.f90

Explicit Parallelizing
Explicit parallelization may yield extra performance at some risk of producing incorrect results. To explicitly parallelize all user-specified loops, do the following.

· · ·

Determine which loops are appropriate to parallelize. Insert a special directive just before each loop that you want to parallelize. Use the -explicitpar option on the compile command line.

Example: Explicit parallelization. Do only the "DO I=1, N" loop.
demo$ cat t1.f90 ... !MIC$ DOALL See Appendix D, "iMPact: Explicit Parallelization. !MIC$& SHARED( a, b, c, n ) !MIC$& PRIVATE( i ) DO i = 1, n ! This loop gets parallelized. a(i) = b(i) * c(i) END DO DO k = 1, m ! This loop does not get parallelized. x(k) = y(k) * z(k) END DO ... demo$ f90 -explicitpar t1.f90

190

Fortran 90 User 's Guide


B
Summary
The following table summarizes the parallel options and the directive.
Table B-1 Parallelization Summary Options Explicit (only) Automatic and Explicit Automatic with reduction Directive DOALL !MIC$ DOALL !MIC$& SHARED( v1, v2, ... ) !MIC$& PRIVATE( u1, u2, ... ) other parameters High Syntax -explicitpar -parallel -parallel -reduction Risk of Incorrect Results Note directive, below. Note directive, below. Note directive, below.

Notes on the Parallel Options and the Directive

· · · · · ·

-parallel includes -explicitpar and does automatic parallelization. The parallelization options can be in any order but must be all lower case. All require a Fortran MP enhancement package and Solaris 2.2, or later. To get faster code, all require a multiprocessor system; on a single-processor system the code usually runs slower. Using -explicitpar or -parallel has high risk as soon as you insert a directive. You get automatic and explicit parallelization with the -parallel option. · The compiler automatically parallelizes all appropriate loops. · It also parallelizes any appropriate loops that you explicitly identify by a directive (still a risk with directives of producing incorrect results). · A loop with an explicit directive gets no reductions.

Standards
Multiprocessing is an evolving concept. When standards for multiprocessing are established, the above features may be superceded.

iMPact: Multiple Processors

191


B
B.3 Speed Gained or Lost
The speed gained varies widely with the application. Some programs are inherently parallel and show great speedup. Many have no parallel potential and show no speedup. There is such a wide range of improvement that it is hard to predict what speedup any one program will get.

Variations in Speedups
To illustrate the range of possible speedups, the following hypothetical scenario is presented.

Assume 4 Processors
With parallelization the following variations occur. The normal upper limit (with 4 processors) is about 3 times as fast.

· · · · · ·

Many perfectly good programs, tuned for single-processor computation, and with the overhead of the parallelization, actually run slower. Many perfectly good programs (tuned for single-processor computation) get absolutely no speedup. Some programs run 10% faster A few less run 50% faster Even fewer run 100% faster A few have so much parallelism that they run 3 or 4 times faster.

Vectorization Comparison
If you have good speedup on vector machines (with an autovectorizing compiler) a first-order rough approximation may be performed as follows. speedup = vectorization * (number of CPUS - 1) Remember that this is only a first-order rough approximation.

B.4 Number of Processors
To set the number of processors, set the environment variable PARALLEL.

192

Fortran 90 User 's Guide


B
Setting environment variables varies with the shell, csh(1) or sh(1). Example: Set PARALLEL to 4.

·

sh:
demo$ PARALLEL=4 demo$ export PARALLEL

·

csh:
demo% setenv PARALLEL 4

Guidelines for Number of Processors
The following are general guidelines, not hard and fast rules. It usually helps to be flexible and experimental with number of processors. For these guidelines, let N be the number of processors on the machine.

· · ·

Do not set PARALLEL to more than N (usually degrades performance) Try PARALLEL set to the number of processors wanted and expected to get. In general, allow at least one processor for activities other than the program you are parallelizing (for overhead, other users, and so forth). · For a one-user system, try PARALLEL=N-1 and try PARALLEL=N. · For a multiple-user system, if the machine is overloaded with users it may help to try PARALLEL set to much less than N. For example, with a 10 user machine, it may help to try PARALLEL at 4, or 6, or 8. If you ask for 10 and cannot get 10, then you may end up time-sharing some CPU's with other users.

iMPact: Multiple Processors

193


B

194

Fortran 90 User 's Guide


iMPact: Automatic Parallelization
This appendix is organized into the following sections.
What You Do What the Compiler Does Definition: Automatic Parallelizing

C

page 195 page 196 page 197

This appendix shows an easy way to parallelize programs for multiple processors. This is called automatic parallelizing. This is a "how to" guide.

C.1 What You Do
See Appendix B, "iMPact: Multiple Processors" for required background.

To tell the compiler to parallelize automatically, use the -parallel option. Example: Parallelize automatically, some loops get parallelized, some do not.
demo$ cat t2.f90 ... DO i = 1, 1000 a(i) = b(i) * c(i) END DO

!



Parallelized

DO k = 3, 1000 ! Not parallelized -- dependency x(k) = x(k-1) * x(k-2) ! See page 196, under Dependency Analysis. END Do ... demo$ f90 -parallel t2.f90

195


C
To determine which programs benefit from automatic parallelization, study the rules the compiler uses to detect parallelizable constructs. Alternatively, compile the programs with automatic parallelization then time the executions.

C.2 What the Compiler Does
For automatic parallelization, the compiler does two things:

· ·

Dependency analysis to detect loops that are parallelizable Parallelization of those loops

This is similar to the analysis and transformations of a vectorizing compiler.

Parallelize the Loop
The compiler applies appropriate dependence-based restructuring transformations. It then distributes the work evenly over the available processors. Each processor executes a different chunk of iterations. Example: 4 processors, 1000 iterations; the following occur simultaneously. Processor Processor Processor Processor 1 2 3 4 executing executing executing executing iterations iterations iterations iterations 1 251 501 751 through through through through 250 500 750 1000

Dependency Analysis
A set of operations can be executed in parallel only if the computational result does not depend on the order of execution. The compiler does a dependency analysis to detect loops with no order-dependence. If it errs, it does so on the side of caution. Also, it may not parallelize a loop that could be parallelized because the gain in performance does not justify the overhead. Example: Automatic parallelizing skips this loop; it has data dependencies.
DO k = 3, 1000 x(k) = x(k-1) * x(k-2) END DO

You cannot calculate x(k) until two previous elements are ready.

196

Fortran 90 User 's Guide


C
Definitions: Array, Scalar, and Pure Scalar · An array variable is one that is declared · ·
with dimensioning in a DIMENSION statement or a type statement (examples below).

A scalar variable is a variable that is not an array variable. A pure scalar variable is a scalar variable that is not aliased (not referenced in an equivalence statement and not in a pointer statement).

Examples: Array/scalar, both m and a are array variables; s is pure scalar.
DIMENSION a(10) REAL m(100,10), REAL, TARGET :: REAL, POINTER :: EQUIVALENCE ( u, s = 0.0 ... s, u, x, z x px z)

The variables u, x, z, and px are scalar variables, but not pure scalar.

C.3 Definition: Automatic Parallelizing
General Definition
Automatic parallelization parallelizes DO loops that have no inter-iteration data dependencies.

Details
This compiler finds and parallelizes any loop that meets the following criteria (but note exceptions below).

· · · ·

The construct is a DO loop (uses the DO statement, but not DO WHILE). 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. Calculations within the loop do not conditionally change any pure scalar variable that is referenced after the loop terminates. Calculations within the loop do not change a scalar variable across iterations. This is called loop-carried dependency.

iMPact: Automatic Parallelization

197


C
There are slight differences from vendor to vendor, since no two vendors have compilers with precisely the same criteria. Example: Using the -parallel option.
... DO i = 1, n a(i) = b(i) * c(i) END DO ... demo$ f90 -parallel t.f90 !



Parallelized

Exceptions for Automatic Parallelizing
For automatic parallelization, the compiler does not parallelize a loop if any of the following occur: · The DO loop is nested inside another DO loop that is parallelized. · Flow control allows jumping out of the DO loop. · There is a user-level subprogram invoked inside the loop. · There is an I/O statement in the loop. · Calculations within the loop change an aliased scalar variable.

Examples
The following examples illustrate the definition of what gets done with automatic parallelization, plus the exceptions. Example: Using -parallel, a call inside a loop.
... DO 40 kb = 1, n ! Not parallelized k = n + 1 - kb b(k) = b(k)/a(k,k) t = -b(k) call daxpy(k-1,t,a(1,k),1,b(1),1) CONTINUE ...

40

198

Fortran 90 User 's Guide


C
Example: Using -parallel, a constant step size loop.
INTEGER, PARAMETER :: del = 2 ... DO k = 3, 1000, del ! x(k) = x(k) * z(k,k) END DO ...



Parallelized

Example: Using -parallel, a variable step size loop.
INTEGER :: del = 2 ... DO k = 3, 1000, del x(k) = x(k) * z(k,k) END DO ...

!



Not parallelized

Example: Using -parallel, nested loops.
DO 900 i = 1, 1000 do 200 j = 1, 1000 ... CONTINUE CONTINUE ! !

Parallelized (outer loop) Not parallelized (inner

loop)

200 900

Example: Using -parallel, a jump out of loop.
DO i = 1, 1000 ! Not parallelized ... IF (a(i) .gt. min_threshold ) GO TO 20 ... END DO CONTINUE ...

20

iMPact: Automatic Parallelization

199


C
Example: Using -parallel, a loop that conditionally changes a scalar variable referenced after a loop.
... DO i = 1, 1000 ! Not parallelized ... IF ( whatever ) s = v(i) END DO t(k) = s ...

200

Fortran 90 User 's Guide


iMPact: Explicit Parallelization
The appendix is organized into the following sections.
What You Do See Appendix B, "iMPact: Multiple Processors for required background. What the Compiler Does Parallel Directives DOALL Loops Exceptions for Explicit Parallelizing Risk with Explicit: Nondeterministic Results Signals

D

page 201 page 202 page 203 page 206 page 208 page 209 page 211

This appendix shows an advanced way to parallelize programs for multiple processors. This is called explicit parallelizing. It may be faster, with some risk of incorrect results. This is a "how to" guide.

D.1 What You Do
To parallelize explicit loops, do the following.

· · · ·

Analyze loops to detect those with no order-dependence. This requires far more analysis and sophistication than using automatic parallelization. Insert a special directive just before each loop that you want parallelized. Use the -explicitpar option on the f90 command line. Check results very carefully.

201


D
The special directive is described later, but first it is illustrated in the following example. Example: Parallelize the "DO I=1, N" loop explicitly.
!MIC$ DOALL !MIC$& SHARED( a, b, !MIC$& PRIVATE( i ) DO i = 1, n a(i) = b(i) END DO DO k = 1, m x(k) = y(k) END DO demo$ f90 -explicitpar c, n ) ! This loop gets parallelized. * c(i) ! This loop does not get parallelized. * z(k) t1.f90

The "!MIC$ DOALL" is explained later.

D.2 What the Compiler Does
For explicit parallelization, the compiler parallelizes those loops that you have specified. This is similar to the transformations of a vectorizing compiler. The compiler applies appropriate dependence-based restructuring transformations. It then distributes the work evenly over the available processors. Each processor executes a different chunk of iterations. Example: 4 processors, 1000 iterations; the following occur simultaneously.
Processor 1 executing iterations Processor 2 executing iterations Processor 3 executing iterations Processor 4 executing iterations 1 251 501 751 through through through through 250 500 750 1000

202

Fortran 90 User 's Guide


D
D.3 Parallel Directives
Explicitly parallelizing loops requires using both of the following:

· ·

A parallel directive One or more command-line options

A parallel directive is a special comment that directs the compiler to do some parallelizing. Directives are also called pragmas.

DOALL--Currently there is one parallel directive, DOALL. The compiler
parallelizes the next loop it finds, if possible.

Form of Directive Lines
Parallel directives have the following syntax.
!MIC$ DOALL [general parameters] [scheduling parameter]

A directive line is defined as follows.

·

It · · ·

starts with the 5 characters CMIC$ or !MIC$, followed by: A space A directive For some directives, one or more parameters

· ·

Spaces before, after, or within a directive are ignored. Letters of a directive line can be in uppercase, lowercase, or mixed.

The form varies for fixed and free form source as follows.

Fixed

· · · · ·

Put CMIC$ or !MIC$ in columns 1 through 5. Directives are listed in columns 7 and beyond. Columns beyond 72 are ignored An initial directive line has a blank in column 6. A continuation directive line has a nonblank in column 6.

iMPact: Explicit Parallelization

203


D
Free

· · · ·

Put !MIC$ followed by a space anywhere in the line. The !MIC$ characters are the first nonblank characters in the line (actually, non-whitespace). Directives are listed after the space. An initial directive line has a blank, tab, or newline in the position immediately after the !MIC$. A continuation directive line has a character other than a blank, tab, or newline in the position immediately after the !MIC$.

Thus, !MIC$ in columns 1 through 5 works for both free and fixed. Example: Directive with continuation lines (DOALL directive and parameters.)
!MIC$ DOALL !MIC$& SHARED( a, b, c, n ) !MIC$& PRIVATE( i ) DO i = 1, n a(i) = b(i) * c(i) END DO

Example: Same directive and parameters, with no continuation lines.
!MIC$ DOALL SHARED( a, b, c, n ) DO i = 1, n a(i) = b(i) * c(i) END DO PRIVATE( i )

204

Fortran 90 User 's Guide


D
DOALL Parameters
The DOALL directive allows general parameters and a scheduling parameter.
Table D-1 DOALL General Parameters Parameter IF ( expr ) Action At runtime, if the expression expr is true, use multiprocessing. If this parameter is not specified, and the loop was not called from a parallel region, then use multiprocessing. Share the variables v1, v2, ... between parallel processes. That is, they are accessible to all the tasks.

SHARED( v1, v2, ...)

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 MAXCPUS( n ) Save the values of private variables from the last DO iteration. Use no more than n CPUs.

Table D-2 DOALL Scheduling Parameters Parameter SINGLE CHUNKSIZE( n ) Action Distribute one iteration to each available processor. Distribute n iterations to each available processor. · n is 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 i iterations, then distribute i/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.

NUMCHUNKS( m )

GUIDED VECTOR

Distribute the iterations by use of guided self-scheduling. · This minimizes synchronization overhead, with acceptable dynamic load balancing. Distribute 64 iterations to each available processor. · If stripmining an inner loop, unrolling is used to automatically improve scheduling.

iMPact: Explicit Parallelization

205


D
Restrictions on DOALL Parameters

· · · · ·

No one variable can be declared both shared and private. The loop control variable of the DOALL loop must be declared private. These variables cannot be array elements or components of derived types. A directive can have many general parameters. A directive can have at most one scheduling parameter.

D.4 DOALL Loops
To use explicit parallelization safely, you must understand the rules for explicit parallelizing. Explicit parallelization of a DOALL loop requires more analysis and sophistication than automatic parallelization. There is far more risk of indeterminate results. This is not only roundoff, but inter-iteration interference.

Definition
For explicit parallelization the DOALL loop is defined as follows:

· · ·

The construct is a DO loop (uses the DO statement, but not DO WHILE). 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. Calculations within the loop do not change any scalar variable that is 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 ensure a proper storeback for them. For each iteration, any subprogram invoked inside the loop does not reference or change values of array variables for any other iteration.

·

Explicitly Parallelizing a DOALL Loop
To explicitly parallelize a DOALL loop, do the following.

· ·

Use the -explicitpar option on the f90 command line. Insert a DOALL parallel directive immediately before the loop, including specifying each variable in the loop as shared or private.

206

Fortran 90 User 's Guide


D
Example: Explicit, DOALL loop.
demo$ cat t4.f90 ... !MIC$ DOALL !MIC$& SHARED( a, b, c, n ) !MIC$& PRIVATE( i ) DO i = 1, n a(i) = b(i) * c(i) END DO DO k = 1, m x(k) = x(k) * z(k,k) END DO ... demo$ f90 -explicitpar t4.f90

!



Parallelized

!

Not parallelized

Example: Explicit, DOALL, some calls can make dependencies.
demo$ cat t5.f90 ... !MIC$ DOALL !MIC$& SHARED( a, b, n ) !MIC$& PRIVATE( kb, k, t ) DO 40 kb = 1, n ! Parallelized k = n + 1 - kb b(k) = b(k)/a(k,k) t = -b(k) CALL daxpy(k-1,t,a(1,k),1,b(1),1) 40 CONTINUE ... demo$ f90 -explicitpar t5.f90

The code is taken from linpack. The subroutine daxpy was analyzed by some software engineer for iteration dependencies and found to not have any. It is a nontrivial analysis. This example is an instance where explicit parallelization is useful over automatic parallelization.

iMPact: Explicit Parallelization

207


D
CALL in a Loop
It is sometimes difficult to determine if there are any inter-iteration dependencies. A subprogram invoked from within the loop requires advanced dependency analysis. Since such a case works only under explicit parallelization, it is you who must do the advanced dependency analysis, not the compiler. The following rule sometimes helps with subprogram calls in a loop: Within a subprogram, if all local variables are automatic, rather than static, then the subprogram does not have iteration dependencies. Note that the above rule is sufficient, but it is by no means necessary. For instance, the daxpy() routine in the previous example does not satisfy this rule, and it does not have iteration dependencies, although that is not obvious. You can make all local variables of a subprogram automatic as follows:

·

List them in an automatic statement. However, then you cannot initialize them in a data statement.

D.5 Exceptions for Explicit Parallelizing
The following are the primary exceptions that prevent the compiler from explicitly parallelizing a DO loop. The compiler issues error messages that the loops are not parallelized, except for a DO loop nested inside another DO loop, which is so common that messages would be distracting.

·

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, and that loop includes a call to a subroutine, then even if you parallelize loops in that subroutine, still, at runtime, those loops are not run in parallel.

· · ·

A flow control statement allows jumping out of the DO loop. The index variable of the loop is subject to side effects, such as being equivalenced. There is an I/O statement in the loop.

For the following exception, the compiler issues no error message.

208

Fortran 90 User 's Guide


D
· ·
If you explicitly parallelize a loop, and that loop includes a call to a subroutine, then even if you parallelize loops in that subroutine, still, at runtime, those loops are not run in parallel. Example: A parallelized loop with a call to a routine that also has a parallelized loop.
... !MIC$ DOALL !MIC$& SHARED( a, x ) !MIC$& PRIVATE( i ) DO 100 i = 1, 200 ... CALL calc (a, x) ... 100 CONTINUE ... SUBROUTINE calc ( b, y ) ... !MIC$ DOALL !MIC$& SHARED( ... ) !MIC$& PRIVATE( m ) DO 1 m = 1, 1000 ... 1 CONTINUE RETURN END



At runtime, loops within this subroutine do not run in parallel.

D.6 Risk with Explicit: Nondeterministic Results
A set of operations can be safely executed in parallel only if the computational result does not depend on the order of execution. For explicit parallelizing, you (rather than the compiler) specify which constructs to parallelize, and then the compiler parallelizes the specified constructs. You do your own dependency analysis. If you force parallelization where dependencies are real, then the results depend on the order of execution; they are nondeterministic; you can get incorrect results.

Testing is not Enough
An entire test suite can produce correct results over and over again, and then produce incorrect results. What happens is that the number of processors (or the system load, or some other parameter) changed. So you must test with different numbers of processors, different system loads, and so forth. But this means you cannot be exhaustive in your test cases.

iMPact: Explicit Parallelization

209


D
The problem is not roundoff but interference between iterations. An example of this is one iteration referencing an element of an array that is calculated in another iteration, but the reference happens before the calculation. One approach is systematic analysis of every explicitly parallelized loop. To be sure of correct results, you must be certain there are no dependencies. Example: Loop with dependency: parallelize explicitly, nondeterministic result
REAL a(1001), s / 0.0 / DO i = 1, 1001 ! Initialize array a. a(i) = i END DO !MIC$ DOALL !MIC$& SHARED( a ) !MIC$& PRIVATE( i ) DO i = 1, 1000 ! This loop has dependencies. a(i) = a(i+1) END DO DO i = 1, 1000 ! Get the sum of all a(i). s = s + a(i) END DO PRINT *, s ! Print the sum. END demo$ f90 -explicitpar t1.f90

How Indeterminacy Arises
In a simpler example, 4 processors, 8 iterations, same kind of initialization:

· · ·

The first 2 iterations run on processor 1 The next 2 iterations run on processor 2 ...

All processors run simultaneously, and usually finish at about the same time. But the compiler provides no synchronization for arrays, and for many reasons, one processor can finish before others; you cannot know the finishing order in advance.
Processor 1 a(1) = a(2) a(2) = a(3) Processor 2 a(3) = a(4) a(4) = a(5) Processor 3 a(5) = a(6) a(6) = a(7) Processor 4 a(7) = a(8) a(8) = a(9)

210

Fortran 90 User 's Guide


D
When processor 1 does a(2) = a(3):

· ·

If processor 2 has done a(3) = a(4), then a(2) gets 4 If processor 2 has not yet done a(3) = a(4), then a(2) gets 3

Therefore the values in a(2) depend on which processor finishes first. After completion of the parallelized loop, the values in array a depend on which processor finishes first. And which finishes second, ... So the sum depends on events you cannot determine. The major variables in the runtime environment that cause this kind of trouble are the number of processors in the system, the system load, interrupts, and so forth. However, you usually cannot know them all, much less control them all.

D.7 Signals
In general, if the loop you are parallelizing does any signal handling, then there is a risk of unpredictable behavior, including a system hang, getting hosed, and other generic bad juju. In particular, if

· ·

The I/O statement raises an exception The signal handler you provide does I/O

then your system can lock up. This causes problems even on single-processor machines. Two common ways of doing signal handling without being explicitly aware of it are the following.

· ·

Input/Output statements (WRITE, PRINT, and so forth) that raise exceptions Requesting Exception Handling

Example: Output that can raise exceptions.
REAL :: x = 1.0, y = 0.0 PRINT *, x/y END

Input/Output statements do locking, and if an exception is raised then there may be an attempt to lock an already locked item, resulting in a deadlock. One (possibly overly cautious) approach: If you are parallelizing, do not have I/O in that loop, and do not request exception handling.

iMPact: Explicit Parallelization

211


D
Example: Using a signal handler which breaks the rules.
CHARACTER string*5, out*20 DOUBLE PRECISION value EXTERNAL exception_handler PRINT *, ' ' PRINT *, 'output' i = ieee_handler('set', 'all', exception_handler) READ(5, '(E5)') value string = '1e310' READ(string, '(E5)') value PRINT *, 'Input string ', string, ' becomes: ', value PRINT *, 'Value of 1e300 * 1e10 is:', 1e300 * 1e10 i = ieee_flags('clear', 'exception', 'all', out) END INTEGER FUNCTION exception_handler(sig, code, sigcontext) INTEGER sig, code, sigcontext(5) PRINT *, '*** IEEE exception raised!' RETURN END

212

Fortran 90 User 's Guide


Index
Symbols
!DIR$ in directives, 180 !MIC$ in directives, 183 .M files, 23 /usr/ccs/lib, error to specify it, 22 /usr/lib, error to specify it, 22 align data types, 125, 126 double word, -dalign, 17 errors across routines, -Xlist, 69 allocated array, 91 ANSI conformance check, -Xlist, 70 X3.198-1992 Fortran standard, 163 -ansi extensions, 15 AnswerBook, documents in, xix ar, 58 create static library, 60 arithmetic nonstandard, 111 standard, 111 array allocated, 91 bounds, exceeding, 81 C FORTRAN differences, 130 dbx, 92, 93 slices in dbx, 93 asa FORTRAN print, 3 audience, xvii

Numerics
132-column lines, -e, 17

A
a.out file, 10 abrupt underflow, 112 access named files, 42 unnamed files, 43 accrued exceptions, do not warn, 106 actions actions/options sorted by action, 13 and what options invoke them, 13 frequently used actions/options, 13 addenda for manuals, read me file, xxi agreement across routines, -Xlist, 69 alias, 50 many options, short commands, 33

213


automatic parallelization definition, 197 exceptions, 198 overview, 190 usage, 195 what the compiler does, 196 automatic variables, 29 autovectorizing compiler, comparison, 192

B
-Bdynamic, 15, 62 best floating point -native, 23 performance, 25 binding dynamic, 15, 17, 62 static, 15, 62 boldface font conventions, xxii Boolean constant, alternate forms, 168 type, constants, 167 bounds of arrays checking, 81 box, clear, xxii BS 6832, xviii -Bstatic, 15, 62

C
C, 155 calls FORTRAN, 149 function return value, 141 is called by FORTRAN, 134 C FORTRAN function compared to subroutine, 123 key aspects of calls, 122 labeled common, 147, 160 -c, compile only, 16 call C from FORTRAN, 134 FORTRAN from C, 149 CALL in a loop, parallelize, 29

case preserving, 124 catalog, 2 Catalyst, 2 catch FPE, 83, 113 CDIR$ in directives, 180 -cg89, 16 -cg92, 16 CIF file, -db, 17 clear box, xxii CMIC$ in directives, 183 code generator option, -cgyr, 16 command ar, create static library, 60 asa, 3 compiler, 9 f90, 9 comments as directives, 179 to Sun, xxi, 20 compatibility C, 188 FORTRAN 77, 184 forward, 188 with f77 I/O, 185 with f77 libraries, 185 with f77 object files, 184 compile check across routines, -Xlist, 71 fails, message, 10 link for a dynamic shared library, 20 link sequence, 10 link, consistent, 65 make assembler source files only, 30 only, -c, 16 passes, times for, 30 compile action align on 8-byte boundaries, -f, 19 ANSI, show non-ANSI extensions, -ansi, 15 assembly-language output files, keep, -S, 30 check across routines, -Xlist, 32

214

Fortran 90 User 's Guide


compile action (continued) compile only, -c, 16 debug, -g, 20 DO loops for one trip min, -onetrip, 25 do not trap on floating-point exceptions, -fnonstop, 19 dynamic binding -Bdynamic, 15, 62 -dy, 17, 62 executable file, name the, -o outfil, 25 explicit parallelization, -explicitpar, 18 extend lines to 132 columns, -e, 17 fast execution, -fast, 19 feedback to Sun, -help, 20 -fixed-form source, -fixed, 19 floating point best, -native, 23 free-form source, -free, 19 generate a CIF file, -db, 17 generate code for generic SPARC, -cg89, 16 SPARC, V8 -cg92, 16 generate double load/store instructions, -dalign, 17 global program checking, -Xlist, 32 library add to search path for, -Ldir, 22 build shared library, -G, 20 name a shared dynamic, -hname, 20 license do not queue request, -noqueue, 24 information, -xlicinfo, 31 link with library x, -lx, 21 list of options, -help, 20 multi-thread safe libraries, -mt, 22 no automatic libraries, -nolib, 24 no run path, norunpath, 24 optimize object code, -On, 25 parallelize, -parallel, 26 pass option to other program, -Qoption, 27 paths, store into object file, -R ls, 28

compile action (continued) print name of each pass as compiler executes, -v, 30 version id of each pass as compiler executes, -V, 30 profile by procedure, -p, 26 procedure, -pg, 27 reduction, analyze loops for reduction, -reduction, 27 report execution times for compilation passes, -time, 30 reset -fast so that it does not use -xlibmopt, 31 set directory for temporary files, -temp=dir, 30 INCLUDE path, -Ipath, 21 module files path, -Mdir, 23 show commands, -dryrun, 17 show compile flags, -flags, 19 stack the local variables, stackvar, 29 static binding -Bstatic, 15, 62 strip executable file of symbol table, s, 29 use fast math routines, xlibmopt, 31 verbose -v, 30 compiler command, 9 frequently used options, 13 passes, 30 recognizes files by types, 11 complete path name, 38 consistent across routines, -Xlist, 69 arguments, commons, parameters, etc., 32 compile and link, 11, 65 continuation lines, 165

Index

215


conventions in text, xxii Courier font, xxii Cray character pointer, 176 pointer, 171 pointer and Fortran 90 pointer, 172 pointer and optimization, 175 create library, 59 SCCS files, 53 cross reference table, -Xlist, 32, 76 current working directory, 37

D
-dalign, double-word align, 17 data inspection, dbx, 96 -db CIF file, 17 dbx, 77 allocated arrays, 91 arrays, 92 catch FPE, 82, 83 commands, 96 current procedure and file, 96 debug, 3 f90 -g, 20 -g, 79 locate exception by line number, 82, 83 next, 81 print, 80 quit, 79 run, 80 set breakpoint, 79 structures, 85, 86, 87, 88, 89 debug, 113

allocated arrays, 91 arguments, agree in number and type, 69 array slices, 93 arrays, 92 checking across routines for global consistency, 69 column print, 93 common blocks, agree in size and type, 69 dbx, 3 debugger, 3 generic function, 94 IEEE exceptions, 113 locating exception by line number, 83 option, -g, 20 parameters, agree globally, 69 pointer, 90 to a scalar, 85 to an array, 86 to user defined type, 89 record, 90 row print, 93 slices of arrays, 93 stack trace, 84 structure, 85, 86, 87, 88, 89 trace of calls, 84 uppercase, lowercase, 96 user defined type, 87 debugger, main features, 96 declared but unused, checking, -Xlist, 70 deep, vasty, 119 dependency analysis, 196 with explicit parallelization, 210 diamond indicates nonstandard, xxii differences Fortran 90, standard, Sun, FORTRAN 77, 163 direct I/O, 45

216

Fortran 90 User 's Guide


directive, 165, 179, 182, 203 DOALL, 182 explicit parallelization, 180, 182, 203 line defined, 180 directory, 37 current working, 37 object library search, 22 temporary files, 30 display to terminal, -Xlist, 71 division by zero IEEE, 101 -dn, 17, 62 DO loops executed once, -onetrip, 25 DOALL directive, 182 doall loop, 206 double-word align, -dalign, 17 -dryrun, 17 -dy, 17, 62 dynamic binding, 17, 62 library, 61 build, -G, 20 name a dynamic library, 20 path in executables, 28

E
-e, extended source lines, 17 ed, 2 emacs, 2 email alias, Sun Programmers SIG, 227 send feedback comments to Sun, xxi environment variable, shorten command line, 33 EOS package, 2 era, 2 errata and addenda for manuals, read me file, xxi error standard error, 41, 44 standard error, accrued exceptions, 111 errors only, -XlistE, 75

establish a signal handler, 109 event management, dbx, 96 ex, 2 exceptions debugging, 113 explicit parallelization, 208 handlers, 102, 107 ieee_handler, 107 location in dbx by line number, 83 unrequited, 111 executable file built-in path to dynamic libraries, 28 names in, nm command, 60 naming it, 25 strip symbol table from, -s, 29 execution time compilation passes, 30 optimization, 25 explicit parallelization, 201 exceptions, 208 overview, 190 risk, 209 -explicitpar, parallelize explicitly, 18 extended lines, -e, 17 syntax check, -Xlist, 70 extensions non-ANSI, 15 to Fortran 90, 164

F
-f, align on 8-byte boundaries, 19 f90 command, 9 -fast, fast execution, 19 features debugger, 96 Fortran 90, standard, Sun, FORTRAN 77, 163 feedback file for email to Sun, xxi feedback to Sun, -help, 20 FFLAGS shorten command line, 33

Index

217


file a.out, 10 directory, 37 executable, 10 information files, xxi internal, 46 object, 10 permissions C FORTRAN, 133 pipe, 41 redirection, 40 split by fsplit, 3 standard error, 44 standard input, 44 standard output, 44 system, 35 file names, 42 passing to programs, 43 recognized by the compiler, 11, 165, 166 FIPS 69-1, xviii fixed form source, 180 form source and tabs, 164 FIXED directive, 179 -fixed form source, 19 -flags synonym for -help, 19 floating-point exceptions, -fnonstop, 19 Goldberg paper, xix hardware, 33 option, -native, 23 -fnonstop no stop on floating-point exceptions, 19 font boldface, xxii conventions, xxii Courier, xxii italic, xxii FORTRAN calls C, 134 is called by C, 149 read me file, bugs, new/changed features, xxi Fortran print, fpr, 3

FPE catch in dbx, 83 fpr FORTRAN print, 3 fpversion, show floating-point version, 33 free form source, 180 form source and tabs, 164 FREE directive, 179, 181 -free, free-form source, 19 fsplit FORTRAN file split, 3 function called within a loop, parallelization, 208 compared to subroutine, C FORTRAN, 123 data type of, checking, -Xlist, 70 names, 124 return values from C, 141 return values to C, 155 unused, checking, -Xlist, 70 used as a subroutine, checking, -Xlist, 70

G
-g, debug, 20 -G, generate a dynamic library, 20 generic functions, debug, 94 getcwd, 37 Glendower, 119 global program checking, 69 Goldberg, floating-point white paper, xix gprof -pg, profile by procedure, 27 gradual underflow, 112 graphically monitor variables, dbx, 97 GSA validated, xviii guidelines for number of processors, 193

218

Fortran 90 User 's Guide


H
-h name, 20 handlers, exception, 102, 107 hardware floating-point fpversion, 33 -help, 20 -help, list of options, 20 Henry IV, 119 hexadecimal, 168 hierarchical file system, 35 Hollerith, 169 Hotspur, 119

interface for C and FORTRAN, 119 problems, checking for, -Xlist, 70 internal files, 46 intrinsic procedures, extensions, 178 invalid, IEEE exception, 101 -Ipath, INCLUDE files, 21 italic font conventions, xxii

L
labeled common C FORTRAN, 147, 160 labels, unused, -Xlist, 70 LD_LIBRARY_PATH, 66 LD_RUN_PATH, 66 and -R, not identical, 28 -Ldir, 22 libm, user error making it unavailable, 22 libraries C FORTRAN, 131 paths in executables, 28 search order, 65 library, 57 build, -G, 20 create, 59 load, 21 loaded, 58 name a shared library, 20 paths in executables, 28 replace module, 61 static, 58 license information, 31 no queue, 24 licensing, 3 limit stack size, 29 line length, 165 line number of exception, 83 segmentation fault (SIGSEGV), 81 line-numbered listing, -Xlist, 71 lines extended -e, 17

I
I/O, 40 identifiers and lowercase, 124 IEEE, 101, 111, 113 754, xviii exceptions, 102 signal handler, 109 standard 754-1985, 163 warning messages off, 106 ieee_flags, 104, 105 ieee_functions, 104 ieee_handler, 104, 107 ieee_values, 104, 106 impatient user 's guide, 5 INCLUDE path, 21 inconsistency arguments, checking, -Xlist, 70 named common blocks, checking, -Xlist, 70 increase stack size, 30 indeterminacy, how it arises, 210 index check of arrays, 81 information files, xxi input redirection, 40 standard, 44 inserting SCCS ID keywords, 53 installation directory, 65

Index

219


link options, 65 sequence, 10 suppress, 16 linker, search order, 65 lint-like checking across routines, -Xlist, 69 list of options, -help, 20 listing line numbered with diagnostics, -Xlist, 69 -Xlist, 76 load library, 21 map, 58 loaded library, 58 local variables, 29 locating exception by line number, 83 segmentation fault by line number, 82 long command lines, 33 loop parallelizing a CALL in a loop, 29 lowercase, do not convert to, 124

-Mdir modules directory, 23 membership in SunPro SIG, Sun Programmers Special Interest Group, 227 MIL-STD-1753, xviii miscellaneous tips alias, many options, short commands, 33 environment variable, many options, short commands, 33 floating-point version, 33 mixing form of source lines, 166 monitor variables graphically, dbx, 97 MP FORTRAN, 189 ­mt, multi-thread safe libraries, 22 multiprocessing standards, 191 multiprocessor FORTRAN, 189

N
name compiler pass, show each, 30 executable file, 25 names in executable, nm command, 60 -native floating point, 23 NBS validated, xviii nesting parallelized loops, 198, 208 network licensing, 3 NIST validated, xviii nm, names in executable, 60 no license queue, 24 -nolib, 24 non-ANSI extensions, 15 nondeterministic results, explicit parallelization, 209 nonstandard arithmetic, 111 indicated by diamond, xxii -noqueue, 24 -norunpath, 24

M
M files, .M files, 23 -m linker option for load map, 58 main stack, 29 make, 50 making SCCS directory, 52 man pages, xix manuals, xix many options, short commands, 33 map, load, 58 mateo, 84 math library, user error making it unavailable, 22

220

Fortran 90 User 's Guide


number of processors for parallelization, 192

O
-O, 25 with -g, 25 -o, output file, 25 -O1, 25, 26 -O2, 25 -O3, 25 object library search directories, 22 obscurities, checking for -Xlist, 70 octal, 168 ode to trace, 84 off license queue, 24 link system libraries, 24 linking, 16 trap for floating-point exceptions, 19 warnings IEEE accrued exceptions, 106 -xlibmopt, 31 -onetrip, 25 on-line documents, xviii, xix optimization object code, 25 performance, 19 options, 12 and what actions they do, 15 frequently used, 13 list available options, -help, 20 listed by option name, 15 what they do, 13 most frequently used, 12 options/actions sorted by option, 15 show list of, -help, 20 OPTIONS variable for command line, 33 order of linker search, 66

output file, naming it, 25 redirection, 40 standard, 44 to terminal, -Xlist, 71 overflow IEEE, 101 stack, 29

P
-p, profile by procedure, 26 parallel directive, 182 PARALLEL, number of processors, 193 -parallel, parallelize loops, 26 parallelization automatic, 195 CALL in a loop, 29 explicit, 18, 201 general requirements, 189 number of processors, 193 overview, 190 reduction, 27 speed gained or lost, 192 summary table, 191 part numbers for manuals, xix parts of large arrays in dbx, 93 pass arguments by value, 127 file names to programs, 43 passes of the compiler, 30 path, 36 .M files, 23 built in during build of a.out, 67 INCLUDE files, 21 modules files, 23 path name, 38 absolute, 38 complete, 38 relative, 38 performance optimization, 19 -pg, profile by procedure, 27 pipes, 41

Index

221


pointee, 171 pointer, 171 pointer, debug, 90 porting problems, checking, -Xlist, 70 position-independent code, 61 pragma, 179, 203 preconnected units, 44 prerequisites, xvii preserve case, 124 print array parts of large, in dbx, 93 slices in dbx, 93 asa, 3 fpr, 3 PRIVATE parameter of DOALL, 183 procedure profile -pg gprof, 27 process control, dbx, 96 processors number for parallelization, 192 prof, -p, 26 profile by procedure, -p, prof, 26 procedure, -pg, gprof, 27 prompt conventions, xxii pure scalar variable, 197 purpose of manual, xvii pwd, 37

record debug, 90 recursive I/O, 23 redirection, 40 standard error, 41 -reduction, parallelize automatically, with reduction, 27 reference versus value, C/FORTRAN, 127 referenced but not declared, checking, -Xlist, 70 relative path name, 38 rename executable file, 6 replace library module, 61 retrospective of accrued exceptions, 111 return function values to C, 155 risk with explicit parallelization, 209 root, 36 run path in executable, 24 running FORTRAN, 6

S
-S, 30 -s, 29 safe libraries for multi-thread programming, 22 sample interface C FORTRAN, 120 SCCS, 52 checking in files, 54 checking out files, 54 creating files, 53 inserting keywords, 53 making directory, 52 putting files under SCCS, 52 search object library directories, 22 segmentation fault, 29, 82 some causes, 81

Q
-Qoption, 27

R
-R and LD_RUN_PATH, not identical, 28 -R list, store lib paths, 28 -r option for ar, 61 random I/O, 45 READMEs directory, xxi

222

Fortran 90 User 's Guide


set directory for temporary files, 30 INCLUDE path, 21 number of processors for parallelization, 192 Shakespeare, 119 shared library name a shared library, 20 SHARED parameter of DOALL, 183 shell script, 49 shorten command lines alias, 33 environment variable, 33 show commands, 17 SIG, Sun Programmers Special Interest Group, xxi, 227 SIGFPE debugging, 113 definition, 102, 107 detect in dbx, 113 generate, 107 when generated, 109, 113 signal handler, 109 with explicit parallelization, 211 SIGSEGV, some causes, 81 size of data types, 125, 126 slices of arrays in dbx, 93 Solaris, 2 source lines -e, 17 source form directives, 166 options, 165 suffixes, 166 speed gained or lost from parallelization, 192 spirits, 119

stack overflow, 29 variables, 29 stack trace, 84 -stackvar, 29 standard arithmetic, 112 conformance to standards, xviii error, 41 error, accrued exceptions, 111 Fortran 90, 163 input, 40, 44 output, 40, 44 statement unreachable, checking, -Xlist, 70 static binding, 17, 62 library, 58 strip executable of symbol table, -s, 29 structure debug, 87, 88, 89 stupid UNIX tricks shorten command line, alias, 33 shorten command line, environment variable, 33 subprogram in loop, explicit parallelization, 208 subroutine compared to function, C FORTRAN, 123 unused, checking, -Xlist, 70 used as a function, checking, -Xlist, 70 suffix of file names recognized by compiler, 11, 165, 166 Sun Programmer Quarterly Newsletter, 227 Sun, sending feedback to, xxi, 20 SunOS 5.x, 2

Index

223


suppress error nnn, -Xlist, 75 license queue, 24 linking, 16 trap for floating-point exceptions, 19 warnings -Xlist, 76 SVR4, 2 symbol table for dbx, -g, 20 strip executable of, 29 syntax compiler, 9 errors, -Xlist, 70 f90, 9 System V Release 4 (SVR4), 2

U
-U do not convert to lowercase, 124 underflow abrupt, 112 gradual, 112 IEEE, 101 units, preconnected, 44 unrecognized options, 12 unrequited exceptions, 111 unused functions, subroutines, variables, labels, -Xlist, 70 uppercase debug, 96 external names, 124 usage automatic parallelization, 195 compiler, 9 explicit parallelization, 201

T
tab character in source, 164 -temp, 30 templates inline, 21 temporary files, directory for, 30 terminal display, -Xlist, 71, 76 textedit, 2 third-party software and hardware, 2 thread stack, 29 -time, timing compilation passes, 30 traceback dbx, 84 ode, 84 tree, 36 triangle as blank space, xxii turn off warnings about IEEE accrued exceptions, 106 type checking across routines, -Xlist, 70 type declaration alternate form, 170 typewriter font, xxii

V
-V, 30 -v, 30 variable unused, checking, -Xlist, 70 used but unset, checking, -Xlist, 70 vasty deep, 119 verify agreement across routines, -Xlist, 69 version id of each compiler pass, 30 vi, 2

W
-w, 31 watchpoints, dbx, 97 where exception occurred, by line number, 83 execution stopped, 84 wimp, 77 interface to dbx, 3, 96

224

Fortran 90 User 's Guide


X
xemacs, 2 -xlibmopt, use fast math routines, 31 -xlicinfo, 31 -Xlist, 71 a la carte options, 73 combination special, 73 defaults, 71 display directly to terminal, 71 errors and cross reference, -XlistX, 74 listing, -XlistL, 74 sample usage, 72 suboptions, 73 details, 75 summary, 32, 74 -Xlist, global program checking, 32, 69 -XlistE, 74, 75 -Xlisterr, 75 -XlistI, 75 -Xlistln, 76 -Xlisto, 76 -Xlistw, 76 -Xlistwar, 76 -XlistX, 76 -xnolib, 31, 32 -xnolibmopt, 31 -xOn, 32 -xpg, 32

Z
zero, division by, 101

Index

225


226

Fortran 90 User 's Guide


Join the SunPro SIG Today
Sun Programmer Special Interest Group The benefits are SIGnificant
At SunSoft, in the Software Development Products business of Sun Microsystems, our goal is to meet the needs of professional software developers by providing the most useful line of software development products for the Solaris platform. We've also recently formed a special interest group, SunPro SIG, designed to provide a worldwide forum for exchanging software development information. This is your invitation to join our world-class organization for professional programmers. For a nominal annual fee of $20, your SunPro SIG membership automatically entitles you to:

·

Membership on an International SunPro SIG Email Alias Share tips on performance tuning, product feedback, or anything you wish; available as a UUNET address and a dial-up number

·

Subscription to the SunProgrammer Quarterly Newsletter Includes advice on getting the most out of your code, regular features, guest columns, product previews and the latest industry gossip

·

Access to a Repository of Free Software SunSoft will collect software related to application development and make it available for downloading

·

Free SunSoft Best-of-Repository CD-ROM You receive one free CD-ROM for joining, plus we'll take the cream of the crop from the depository and distribute it to members annually

·
SPECIAL OFFER

Free Access to SIG Events Including national events, like SIG seminars held at the SUG conference, and regional SunPro SIG seminars

Sign up today, and receive a SunPro SIG Tote Bag A spiffy 15" x 12" black nylon Cordura tote with the SIG logo proof positive of your Power Programmer status


So join the SunPro SIG today. And plug into what's happening in SPARC and Solaris development world-wide. Simply complete the form below. Mail to: SunPro SIG, 2550 Garcia Avenue MS UMPK 03-205, Mountain View, CA,94043-1100 TEL: (415) 688-9862 or FAX: (415) 968-6396 Unfortunately we cannot accept credit card orders via Email since we need to have your signature on file.

Sign me up for SunPro SIG! Sun Programmer Special Interest Group
Date Name Title Company Email Address Address City ZIP Phone Fax State Country Card # _______________ Expiration Date: _________ Signature: I'd like to pay for my one-year membership fee of $20 by:

ê ê

VISA

MASTERCARD

ALL INFO MUST BE FILLED OUT

ê

_____________________ Check made payable to SunSoft

SunSoft, A Sun Microsystems, Inc. Business