C Language Reference Manual

007–0701–150
COPYRIGHT
Copyright © 1999, 2002–2003 Silicon Graphics, Inc. All rights reserved; provided portions may be copyright in third parties, as
indicated elsewhere herein. No permission is granted to copy, distribute, or create derivative works from the contents of this electronic
documentation in any manner, in whole or in part, without the prior written permission of Silicon Graphics, Inc.

LIMITED RIGHTS LEGEND

The electronic (software) version of this document was developed at private expense; if acquired under an agreement with the USA
government or any contractor thereto, it is acquired as "commercial computer software" subject to the provisions of its applicable
license agreement, as specified in (a) 48 CFR 12.212 of the FAR; or, if acquired for Department of Defense units, (b) 48 CFR 227-7202 of
the DoD FAR Supplement; or sections succeeding thereto. Contractor/manufacturer is Silicon Graphics, Inc., 1600 Amphitheatre Pkwy
2E, Mountain View, CA 94043-1351.

TRADEMARKS AND ATTRIBUTIONS

Silicon Graphics, SGI, the SGI logo, and IRIX are registered trademarks of Silicon Graphics, Inc. in the United States and/or other
countries worldwide.

Gaussian is a trademark of Gaussian, Inc. MIPSpro is a trademark of MIPS Technologies, Inc., and is used under license to Silicon
Graphics, Inc. UNIX and the X device are trademarks of The Open Group in the United States and other countries.

Cover Design By Sarah Bolles, Sarah Bolles Design, and Dany Galgani, SGI Technical Publications.
 New Features in This Manual

Information regarding the use of lint-style comments in macros has been added to
Appendix B, "lint-style Comments", page 167.

007–0701–150 iii
 Record of Revision

Version Description

7.3 April 1999

This document has been updated to support the MIPSpro 7.3
release.

140 September 2002

This document has been updated to support the MIPSpro 7.4
release which runs on IRIX operating systems version 6.5 and later.

150 June 2003

This document has been updated to support the MIPSpro 7.4.1m
release which runs on IRIX operating systems version 6.5 and later.

007–0701–150 v
 Contents

About This Manual . . . . . . . . . . . . . . . . . . . . xix

Related Publications . . . . . . . . . . . . . . . . . . . . . . xix
Obtaining Publications . . . . . . . . . . . . . . . . . . . . . . xx
Conventions . . . . . . . . . . . . . . . . . . . . . . . . . xxi
Reader Comments . . . . . . . . . . . . . . . . . . . . . . . xxi

1. An Overview of ANSI C . . . . . . . . . . . . . . . . . . 1
ANSI C . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Strictly Conforming Programs . . . . . . . . . . . . . . . . . . . 1
Name Spaces . . . . . . . . . . . . . . . . . . . . . . . . 1
Compiling ANSI Programs . . . . . . . . . . . . . . . . . . . 2
Guidelines for Using ANSI C . . . . . . . . . . . . . . . . . . . 2
Compiling Traditional C Programs . . . . . . . . . . . . . . . . . 3
Helpful Programming Hints . . . . . . . . . . . . . . . . . . . . 3
Recommended Practices . . . . . . . . . . . . . . . . . . . . 3
Practices to Avoid . . . . . . . . . . . . . . . . . . . . . . 4

2. C Language Changes . . . . . . . . . . . . . . . . . . . 7
Preprocessor Changes . . . . . . . . . . . . . . . . . . . . . . 7
Replacement of Macro Arguments in Strings . . . . . . . . . . . . . . 8
Token Concatenation . . . . . . . . . . . . . . . . . . . . . 9
Changes in Disambiguating Identifiers . . . . . . . . . . . . . . . . 10
Scoping Differences . . . . . . . . . . . . . . . . . . . . . . 10
Name Space Changes . . . . . . . . . . . . . . . . . . . . . 12
Changes in the Linkage of Identifiers . . . . . . . . . . . . . . . . 12

007–0701–150 vii
Contents

Types and Type Compatibility . . . . . . . . . . . . . . . . . . . 14

Type Promotion in Arithmetic Expressions . . . . . . . . . . . . . . . 14
Type Promotion and Floating Point Constants . . . . . . . . . . . . . . 15
Compatible Types . . . . . . . . . . . . . . . . . . . . . . 17
Argument Type Promotions . . . . . . . . . . . . . . . . . . . 17
Mixed Use of Functions . . . . . . . . . . . . . . . . . . . . 18
Function Prototypes . . . . . . . . . . . . . . . . . . . . . . 18
External Name Changes . . . . . . . . . . . . . . . . . . . . . 20
Changes in Function Names . . . . . . . . . . . . . . . . . . . 20
Changes in Linker-Defined Names . . . . . . . . . . . . . . . . . 20
Data Area Name Changes . . . . . . . . . . . . . . . . . . . . 21
Standard Headers . . . . . . . . . . . . . . . . . . . . . . . 22

3. Lexical Conventions . . . . . . . . . . . . . . . . . . . 23
Comments . . . . . . . . . . . . . . . . . . . . . . . . . 23
Identifiers . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Keywords . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Constants . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Integer Constants . . . . . . . . . . . . . . . . . . . . . . 24
Character Constants . . . . . . . . . . . . . . . . . . . . . . 25
Special Characters . . . . . . . . . . . . . . . . . . . . . . 25
Trigraph Sequences (ANSI C Only) . . . . . . . . . . . . . . . . 26
Floating Constants . . . . . . . . . . . . . . . . . . . . . . 27
Enumeration Constants . . . . . . . . . . . . . . . . . . . . . 28
String Literals . . . . . . . . . . . . . . . . . . . . . . . . 28
Operators . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Punctuators . . . . . . . . . . . . . . . . . . . . . . . . . 29

4. Meaning of Identifiers . . . . . . . . . . . . . . . . . . 31

viii 007–0701–150
 C Language Reference Manual

Disambiguating Names . . . . . . . . . . . . . . . . . . . . . 31
Scope . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Block Scope . . . . . . . . . . . . . . . . . . . . . . . 32
Function Scope . . . . . . . . . . . . . . . . . . . . . . 32
Function Prototype Scope . . . . . . . . . . . . . . . . . . . 32
File Scope . . . . . . . . . . . . . . . . . . . . . . . . 32
Name Spaces . . . . . . . . . . . . . . . . . . . . . . . . 33
Name Space Discrepancies Between Traditional and ANSI C . . . . . . . . 33
Linkage of Identifiers . . . . . . . . . . . . . . . . . . . . . 34
Linkage Discrepancies Between Traditional and ANSI C . . . . . . . . . . 36
Storage Duration . . . . . . . . . . . . . . . . . . . . . . . 37
Object Types . . . . . . . . . . . . . . . . . . . . . . . . . 38
Character Types . . . . . . . . . . . . . . . . . . . . . . . 38
Integer and Floating Point Types . . . . . . . . . . . . . . . . . . 38
Derived Types . . . . . . . . . . . . . . . . . . . . . . . 40
void Type . . . . . . . . . . . . . . . . . . . . . . . . 40
Objects and lvalues . . . . . . . . . . . . . . . . . . . . . . . 41

5. Operator Conversions . . . . . . . . . . . . . . . . . . . 43
Conversions of Characters and Integers . . . . . . . . . . . . . . . . 43
Conversions of Float and Double . . . . . . . . . . . . . . . . . . 43
Conversion of Floating and Integral Types . . . . . . . . . . . . . . . . 44
Conversion of Pointers and Integers . . . . . . . . . . . . . . . . . 44
Conversion of unsigned Integers . . . . . . . . . . . . . . . . . . 44
Arithmetic Conversions . . . . . . . . . . . . . . . . . . . . . 45
Integral Promotions . . . . . . . . . . . . . . . . . . . . . . 45
Usual Arithmetic Conversions . . . . . . . . . . . . . . . . . . 45

007–0701–150 ix
Contents

Traditional C Conversion Rules . . . . . . . . . . . . . . . . . 46

ANSI C Conversion Rules . . . . . . . . . . . . . . . . . . . 46
Conversion of Other Operands . . . . . . . . . . . . . . . . . . . 47
Conversion of lvalues and Function Designators . . . . . . . . . . . . . 47
Conversion of void Objects . . . . . . . . . . . . . . . . . . . 47
Conversion of Pointers . . . . . . . . . . . . . . . . . . . . . 48

6. Expressions and Operators . . . . . . . . . . . . . . . . . 49

Precedence and Associativity Rules in C . . . . . . . . . . . . . . . . 49
Primary Expressions . . . . . . . . . . . . . . . . . . . . . . 51
Postfix Expressions . . . . . . . . . . . . . . . . . . . . . . . 51
Subscripts . . . . . . . . . . . . . . . . . . . . . . . . . 52
Function Calls . . . . . . . . . . . . . . . . . . . . . . . 52
Structure and Union References . . . . . . . . . . . . . . . . . . 54
Indirect Structure and Union References . . . . . . . . . . . . . . . 54
postfix ++ and postfix - - . . . . . . . . . . . . . . . . . . 54
Unary Operators . . . . . . . . . . . . . . . . . . . . . . . 55
Address-of and Indirection Operators . . . . . . . . . . . . . . . . 55
Unary + and - Operators . . . . . . . . . . . . . . . . . . . . 56
Unary ! and ~ Operators . . . . . . . . . . . . . . . . . . . . 56
Prefix ++ and - - Operators . . . . . . . . . . . . . . . . . . . 56
sizeof Unary Operator . . . . . . . . . . . . . . . . . . . . 57
Cast Operators . . . . . . . . . . . . . . . . . . . . . . . . 57
Multiplicative Operators . . . . . . . . . . . . . . . . . . . . . 58
Additive Operators . . . . . . . . . . . . . . . . . . . . . . . 59
Shift Operators . . . . . . . . . . . . . . . . . . . . . . . . 60
Relational Operators . . . . . . . . . . . . . . . . . . . . . . 60
Equality Operators . . . . . . . . . . . . . . . . . . . . . . . 61

x 007–0701–150
 C Language Reference Manual

Bitwise AND Operator . . . . . . . . . . . . . . . . . . . . . . 62

Bitwise Exclusive OR Operator . . . . . . . . . . . . . . . . . . . 62
Bitwise Inclusive OR Operator . . . . . . . . . . . . . . . . . . . 63
Logical AND Operator . . . . . . . . . . . . . . . . . . . . . . 63
Logical OR Operator . . . . . . . . . . . . . . . . . . . . . . 63
Conditional Operator . . . . . . . . . . . . . . . . . . . . . . 64
Assignment Operators . . . . . . . . . . . . . . . . . . . . . . 65
Assignment Using = (Simple Assignment) . . . . . . . . . . . . . . . 65
Compound Assignment . . . . . . . . . . . . . . . . . . . . 66
Comma Operator . . . . . . . . . . . . . . . . . . . . . . . 66
Constant Expressions . . . . . . . . . . . . . . . . . . . . . . 66
Integer and Floating Point Exceptions . . . . . . . . . . . . . . . . . 67

7. Declarations . . . . . . . . . . . . . . . . . . . . . . 69
Storage Class Specifiers . . . . . . . . . . . . . . . . . . . . . 70
Type Specifiers . . . . . . . . . . . . . . . . . . . . . . . . 71
Structure and Union Declarations . . . . . . . . . . . . . . . . . . 72
Bitfields . . . . . . . . . . . . . . . . . . . . . . . . . . 75
Enumeration Declarations . . . . . . . . . . . . . . . . . . . . . 76
Type Qualifiers . . . . . . . . . . . . . . . . . . . . . . . . 77
Declarators . . . . . . . . . . . . . . . . . . . . . . . . . 78
Meaning of Declarators . . . . . . . . . . . . . . . . . . . . . 78
Pointer Declarators . . . . . . . . . . . . . . . . . . . . . . 79
Qualifiers and Pointers . . . . . . . . . . . . . . . . . . . . 79
Pointer-related Command Options . . . . . . . . . . . . . . . . 80
Array Declarators . . . . . . . . . . . . . . . . . . . . . . 81
Function Declarators and Prototypes . . . . . . . . . . . . . . . . 82
Prototyped Functions Summarized . . . . . . . . . . . . . . . . 84

007–0701–150 xi
Contents

Restrictions on Declarators . . . . . . . . . . . . . . . . . . . . 85
Type Names . . . . . . . . . . . . . . . . . . . . . . . . . 86
Implicit Declarations . . . . . . . . . . . . . . . . . . . . . . 87
typedef . . . . . . . . . . . . . . . . . . . . . . . . . . 87
Initialization . . . . . . . . . . . . . . . . . . . . . . . . . 88
Initialization of Aggregates . . . . . . . . . . . . . . . . . . . 89
Examples of Initialization . . . . . . . . . . . . . . . . . . . . 90

8. Statements . . . . . . . . . . . . . . . . . . . . . . . 93
Expression Statement . . . . . . . . . . . . . . . . . . . . . . 93
Compound Statement or Block . . . . . . . . . . . . . . . . . . . 93
Selection Statements . . . . . . . . . . . . . . . . . . . . . . 94
if Statement . . . . . . . . . . . . . . . . . . . . . . . . 94
switch Statement . . . . . . . . . . . . . . . . . . . . . . 95
Iteration Statements . . . . . . . . . . . . . . . . . . . . . . 95
while Statement . . . . . . . . . . . . . . . . . . . . . . 96
do Statement . . . . . . . . . . . . . . . . . . . . . . . . 96
for Statement . . . . . . . . . . . . . . . . . . . . . . . 96
Jump Statements . . . . . . . . . . . . . . . . . . . . . . . 97
goto Statement . . . . . . . . . . . . . . . . . . . . . . . 97
continue Statement . . . . . . . . . . . . . . . . . . . . . 97
break Statement . . . . . . . . . . . . . . . . . . . . . . 98
return Statement . . . . . . . . . . . . . . . . . . . . . . 98
Labeled Statements . . . . . . . . . . . . . . . . . . . . . . . 98

9. External Definitions . . . . . . . . . . . . . . . . . . . 101

External Function Definitions . . . . . . . . . . . . . . . . . . . . 101
External Object Definitions . . . . . . . . . . . . . . . . . . . . 102

10. Multiprocessing Directives . . . . . . . . . . . . . . . . 103

xii 007–0701–150
 C Language Reference Manual

OpenMP C/C++ API Multiprocessing Directives . . . . . . . . . . . . . . 104

Using Parallel Regions . . . . . . . . . . . . . . . . . . . . . . 104
Coding Rules of #pragma Directives . . . . . . . . . . . . . . . . 105
Parallel Regions . . . . . . . . . . . . . . . . . . . . . . . . 106
Parallel Reduction Operations in C and C++ . . . . . . . . . . . . . . . 107
Restrictions on the Reduction Clause . . . . . . . . . . . . . . . . 108
Performance Considerations . . . . . . . . . . . . . . . . . . . 109
Reduction on User-Defined Types in C++ . . . . . . . . . . . . . . . 110
Reduction Example . . . . . . . . . . . . . . . . . . . . . 110
Restrictions for the C++ Compiler . . . . . . . . . . . . . . . . . . 111
Restrictions on pfor . . . . . . . . . . . . . . . . . . . . . 111
Restrictions on Exception Handling . . . . . . . . . . . . . . . . . 112
Scoping Restrictions . . . . . . . . . . . . . . . . . . . . . . 113

11. Multiprocessing Advanced Features . . . . . . . . . . . . . 115

Run-time Library Routines . . . . . . . . . . . . . . . . . . . . 115
mp_block and mp_unblock . . . . . . . . . . . . . . . . . . . 115
mp_setup, mp_create, and mp_destroy . . . . . . . . . . . . . . 115
mp_blocktime . . . . . . . . . . . . . . . . . . . . . . . 116
mp_numthreads, mp_suggested_numthreads, mp_set_numthreads . . . . . 116
mp_my_threadnum . . . . . . . . . . . . . . . . . . . . . . 117
mp_setlock, mp_unsetlock, mp_barrier . . . . . . . . . . . . . . 117
mp_set_slave_stacksize . . . . . . . . . . . . . . . . . . . 117
Run-time Environment Variables . . . . . . . . . . . . . . . . . . . 118
MP_SET_NUMTHREADS, MP_BLOCKTIME, MP_SETUP . . . . . . . . . . . . 118
MP_SUGNUMTHD, MP_SUGNUMTHD_MIN, MP_SUGNUMTHD_MAX, MP_SUGNUMTHD_VERBOSE 119
MP_SCHEDTYPE, CHUNK . . . . . . . . . . . . . . . . . . . . 119
MP_SLAVE_STACKSIZE . . . . . . . . . . . . . . . . . . . . 120

007–0701–150 xiii
Contents

MPC_GANG . . . . . . . . . . . . . . . . . . . . . . . . . 120
Communicating Between Threads Through Thread Local Data . . . . . . . . . 120
Synchronization Intrinsics . . . . . . . . . . . . . . . . . . . . . 123
Atomic fetch-and-op Operations . . . . . . . . . . . . . . . . . . 124
Atomic op-and-fetch Operations . . . . . . . . . . . . . . . . . . 125
Atomic compare-and-swap Operation . . . . . . . . . . . . . . . . 126
Atomic synchronize Operation . . . . . . . . . . . . . . . . . . 127
Atomic lock and unlock Operations . . . . . . . . . . . . . . . . . 127
Atomic lock-test-and-set Operation . . . . . . . . . . . . . . . . 127
Atomic lock-release Operation . . . . . . . . . . . . . . . . . . 127
Example of Implementing a Pure Spin-Wait Lock . . . . . . . . . . . . 128

Appendix A. Implementation-Defined Behavior . . . . . . . . . . 129

Translation (F.3.1) . . . . . . . . . . . . . . . . . . . . . . . 129
Environment (F.3.2) . . . . . . . . . . . . . . . . . . . . . . . 130
Identifiers (F.3.3) . . . . . . . . . . . . . . . . . . . . . . . . 131
Characters (F.3.4) . . . . . . . . . . . . . . . . . . . . . . . 131
Integers (F.3.5) . . . . . . . . . . . . . . . . . . . . . . . . 133
Floating Point (F.3.6) . . . . . . . . . . . . . . . . . . . . . . 134
Arrays and Pointers (F.3.7) . . . . . . . . . . . . . . . . . . . . 135
Registers (F.3.8) . . . . . . . . . . . . . . . . . . . . . . . . 136
Structures, Unions, Enumerations, and Bitfields (F.3.9) . . . . . . . . . . . . 136
Qualifiers (F.3.10) . . . . . . . . . . . . . . . . . . . . . . . 138
Declarators (F.3.11) . . . . . . . . . . . . . . . . . . . . . . . 138
Statements (F.3.12) . . . . . . . . . . . . . . . . . . . . . . . 139
Preprocessing Directives (F.3.13) . . . . . . . . . . . . . . . . . . . 139
Library Functions (F.3.14) . . . . . . . . . . . . . . . . . . . . . 140

xiv 007–0701–150
 C Language Reference Manual

Signals . . . . . . . . . . . . . . . . . . . . . . . . . . 141
Signal Notes . . . . . . . . . . . . . . . . . . . . . . . 145
Diagnostics . . . . . . . . . . . . . . . . . . . . . . . 147
Streams and Files . . . . . . . . . . . . . . . . . . . . . . 148
Temporary Files . . . . . . . . . . . . . . . . . . . . . . . 150
errno and perror . . . . . . . . . . . . . . . . . . . . . . 150
Memory Allocation . . . . . . . . . . . . . . . . . . . . . . 158
abort Function . . . . . . . . . . . . . . . . . . . . . . . 158
exit Function . . . . . . . . . . . . . . . . . . . . . . . 158
getenv Function . . . . . . . . . . . . . . . . . . . . . . 158
system Function . . . . . . . . . . . . . . . . . . . . . . 159
strerror Function . . . . . . . . . . . . . . . . . . . . . . 159
Time Zones and the clock Function . . . . . . . . . . . . . . . . . 159
Locale-Specific Behavior (F.4) . . . . . . . . . . . . . . . . . . . . 160
Common Extensions (F.5) . . . . . . . . . . . . . . . . . . . . . 160
Environment Arguments (F.5.1) . . . . . . . . . . . . . . . . . . 161
Specialized Identifiers . . . . . . . . . . . . . . . . . . . . . 161
Lengths and Cases of Identifiers . . . . . . . . . . . . . . . . . . 161
Scopes of Identifiers (F.5.4) . . . . . . . . . . . . . . . . . . . . 161
Writable String Literals (F.5.5) . . . . . . . . . . . . . . . . . . . 162
Other Arithmetic Types (F.5.6) . . . . . . . . . . . . . . . . . . 162
Function Pointer Casts (F.5.7) . . . . . . . . . . . . . . . . . . . 162
Non-int Bit-Field Types (F.5.8) . . . . . . . . . . . . . . . . . . 162
fortran Keyword (F.5.9) . . . . . . . . . . . . . . . . . . . . 163
asm Keyword (F.5.10) . . . . . . . . . . . . . . . . . . . . . 163
Multiple External Definitions (F.5.11) . . . . . . . . . . . . . . . . 163
Empty Macro Arguments (F.5.12) . . . . . . . . . . . . . . . . . . 163

007–0701–150 xv
Contents

Predefined Macro Names (F.5.13) . . . . . . . . . . . . . . . . . . 164

Extra Arguments for Signal Handlers (F.5.14) . . . . . . . . . . . . . . 164
Additional Stream Types and File-Opening Modes (F.5.15) . . . . . . . . . . 164
Defined File Position Indicator (F.5.16) . . . . . . . . . . . . . . . . 165

Appendix B. lint-style Comments . . . . . . . . . . . . . . . 167

Appendix C. Built-in Functions . . . . . . . . . . . . . . . . 169

Index . . . . . . . . . . . . . . . . . . . . . . . . . . 171

xvi 007–0701–150
 Tables

Table 2-1 Effect of Compilation Options on Floating Point Conversions . . . . . . 16

Table 2-2 Using __STDC__ to Affect Floating Point Conversions . . . . . . . . 17
Table 2-3 Effect of Compilation Mode on Names . . . . . . . . . . . . . 21
Table 3-1 Reserved Keywords . . . . . . . . . . . . . . . . . . . 24
Table 3-2 Escape Sequences for Nongraphic Characters . . . . . . . . . . . 26
Table 3-3 Trigraph Sequences . . . . . . . . . . . . . . . . . . . 27
Table 4-1 Storage Class Sizes . . . . . . . . . . . . . . . . . . . 39
Table 6-1 Precedence and Associativity Examples . . . . . . . . . . . . . 49
Table 6-2 Operator Precedence and Associativity . . . . . . . . . . . . . 50
Table 7-1 Examples of Type Names . . . . . . . . . . . . . . . . . 86
Table 10-1 Multiprocessing C/C++ Compiler Directives . . . . . . . . . . . 103
Table A-1 Integer Types and Ranges . . . . . . . . . . . . . . . . . 133
Table A-2 Ranges of floating point Types . . . . . . . . . . . . . . . 135
Table A-3 Alignment of Structure Members . . . . . . . . . . . . . . . 137
Table A-4 Signals . . . . . . . . . . . . . . . . . . . . . . 142
Table A-5 Valid Codes in a Signal-Catching Function . . . . . . . . . . . . 144
Table B-1 lint–style Comments . . . . . . . . . . . . . . . . . . 168
Table C-1 Built-in Functions . . . . . . . . . . . . . . . . . . . 169

007–0701–150 xvii
 About This Manual

This manual contains a summary of the syntax and semantics of the C programming
language as implemented on SGI workstations. It documents previous releases of the
SGI C compiler as well as the American National Standards Institute (ANSI) C
compiler.
The SGI compiler system supports three modes of compilation: the old 32-bit mode
(-o32 or -32 option), the new 32-bit mode (-n32 option), and the 64-bit mode (-64
option).
For information on compilation modes and general compiler options for the old 32-bit
mode, see the o32(5) man page and the MIPS O32 Compiling and Performance Tuning
Guide.
For information on the new 32-bit mode and 64-bit mode, see the cc(1) man page and
the MIPSpro N32/64 Compiling and Performance Tuning Guide.
The term “traditional C” refers to the dialect of C described in the first edition of The
C Programming Language by Kernighan and Ritchie.

Related Publications
The following documents contain information that may be helpful in porting code to
the newer SGI compilers:
• MIPS O32 Compiling and Performance Tuning Guide
• MIPSpro N32/64 Compiling and Performance Tuning Guide
• MIPSpro N32 ABI Handbook
• MIPSpro 64-Bit Porting and Transition Guide
The following documents contain information about SGI’s implementation of C and
C++:
• C++ Programmer’s Guide
• MIPSpro C and C++ Pragmas

007–0701–150 xix
About This Manual

Several performance evaluation and debugging tools are available to help you
optimize and evaluate your code. See the ProDev WorkShop: Overview for a
description of the different tools that are available.
See the Guides to SGI Compilers and Compiling Tools for an overview of all SGI
compilers, compiler documentation, optimization tools, porting tools, and
performance tools.
In addition to the above SGI documentation, several third party documents contain
additional information which may be helpful. These books can be ordered from any
book vendor:
• Ellis, Margaret A., and Bjarne Stroustrup. The Annotated C++ Reference Manual.
Addison-Wesley Publishing Company, 1990. ISBN 0201514591.
• Josuttis, Nicolai. The C++ Standard Library: A Tutorial and Reference.
Addison-Wesley Publishing Company, 1999. ISBN 0201379260.
• Kernighan, Brian W. and Dennis M. Ritchie. The C Programming Language.
Prentice-Hall, 1988. ISBN 0131103628.

Obtaining Publications
You can obtain SGI documentation in the following ways:
• See the SGI Technical Publications Library at http://docs.sgi.com. Various
formats are available. This library contains the most recent and most
comprehensive set of online books, release notes, man pages, and other
information.
• If it is installed on your SGI system, you can use InfoSearch, an online tool that
provides a more limited set of online books, release notes, and man pages. With
an IRIX system, select Help from the Toolchest, and then select InfoSearch. Or
you can type infosearch on a command line.
• You can also view release notes by typing either grelnotes or relnotes on a
command line.
• You can also view man pages by typing man title on a command line.

xx 007–0701–150
 C Language Reference Manual

Conventions
The following conventions are used throughout this document:

Convention Meaning
command This fixed-space font denotes literal items such as
commands, files, routines, path names, signals,
messages, and programming language structures.
variable Italic typeface denotes variable entries and words or
concepts being defined.
user input This bold, fixed-space font denotes literal items that the
user enters in interactive sessions. (Output is shown in
nonbold, fixed-space font.)
[] Brackets enclose optional portions of a command or
directive line.
... Ellipses indicate that a preceding element can be
repeated.

Reader Comments
If you have comments about the technical accuracy, content, or organization of this
publication, contact SGI. Be sure to include the title and document number of the
publication with your comments. (Online, the document number is located in the
front matter of the publication. In printed publications, the document number is
located at the bottom of each page.)
You can contact SGI in any of the following ways:
• Send e-mail to the following address:
techpubs@sgi.com
• Use the Feedback option on the Technical Publications Library Web page:
http://docs.sgi.com
• Contact your customer service representative and ask that an incident be filed in
the SGI incident tracking system.

007–0701–150 xxi
About This Manual

• Send mail to the following address:

Technical Publications
SGI
1600 Amphitheatre Parkway, M/S 535
Mountain View, California 94043–1351
• Send a fax to the attention of “Technical Publications” at +1 650 932 0801.
SGI values your comments and will respond to them promptly.

xxii 007–0701–150
 Chapter 1

An Overview of ANSI C

This chapter briefly discusses the scope of the standard and lists some programming
practices to avoid and some practices to use.

ANSI C
The ANSI standard on the C programming language is designed to promote the
portability of C programs among a variety of data-processing systems. To accomplish
this, the standard covers three major areas: the environment in which the program
compiles and executes, the semantics and syntax of the language, and the content and
semantics of a set of library routines and header files.

Strictly Conforming Programs

Strictly conforming programs adhere to the following guidelines:
• They use only those features of the language defined in the standard.
• They do not produce output dependent on any ill-defined behavior. Ill-defined
behavior includes implementation-defined, undefined, and unspecified behavior
which refers to areas that the standard does not specify.
• They do not exceed any minimum limit.

This ANSI C environment is designed to be a conforming hosted implementation,

which will accept any strictly conforming program. Extensions are allowed only if the
behavior of strictly conforming programs is not altered.

Name Spaces
In addition to knowing which features of the language and library you can rely on
when writing portable programs, you must be able to avoid naming conflicts with
support routines used for the implementation of the library. To avoid such naming
conflicts, ANSI divides the space of available names into a set reserved for the user
and a set reserved for the implementation. Any name is in the user’s name space if it
meets these three requirements (this rule is given for simplicity; the space of names
reserved for the user is actually somewhat larger than this):

007–0701–150 1
1: An Overview of ANSI C

• It does not begin with an underscore

• It is not a keyword in the language
• It is not reserved for the ANSI library
Strictly conforming programs may not define any names unless they are in the user’s
namespace. New keywords as well as those names reserved for the ANSI library are
discussed in "Standard Headers", page 22.

Compiling ANSI Programs

To provide the portable clean environment dictated by ANSI while retaining the many
extensions available to SGI users, two modes of compilation are provided for ANSI
programs. Each of these switches to the cc command invokes the ANSI compiler:
-ansi Enforces a pure ANSI environment, eliminating SGI extensions. The
ANSI symbol indicating a pure environment (__STDC__) is defined to
be 1 for the preprocessor. Use this mode when compiling strictly
conforming programs, because it guarantees purity of the ANSI
namespace.
-xansi Adds SGI extensions to the environment. This mode is the default. The
ANSI preprocessor symbol (__STDC__) is defined to be 1. The symbol
to include extensions from standard headers (__EXTENSIONS__) is also
defined, as is the symbol to inline certain library routines that are
directly supported by the hardware (__INLINE_INTRINSICS.) Note
that when these library routines are made to be intrinsic, they may no
longer be strictly ANSI conforming (for example, errno may not be set
correctly).

Guidelines for Using ANSI C

The following are some key facts to keep in mind when you use ANSI C:
• Use only -lc and/or -lm to specify the C and/or math libraries. These switches
ensure the incorporation of the ANSI version of these libraries.
• Use the switch -fullwarn to receive additional diagnostic warnings that are
suppressed by default. SGI recommends using this option with the -woff option
to remove selected warnings during software development.

2 007–0701–150
 C Language Reference Manual

• Use the switch -wlint (-o32 mode only) to get lint-like warnings about the
compiled source. This option provides lint-like warnings for ANSI and -cckr
modes and can be used together with the other cc options and switches.
• Remember that the default compilation mode is shared and the libraries are shared.

Compiling Traditional C Programs

To compile code using traditional C (that is, non-ANSI), use the switch -cckr. The
dialect of C invoked by [-cckr] is referred to interchangeably as -cckr, “the
previous version of SGI C,” and “traditional C” in the remainder of this document.
You can find complete information concerning ANSI and non-ANSI compilation
modes in the cc(1) online reference page.

Helpful Programming Hints

Although the ANSI Standard has added only a few new features to the C language, it
has tightened the semantics of many areas. In some cases, constructs were removed
that were ambiguous, no longer used, or obvious hacks. The next two sections give
two lists of programming practices. The first section recommends practices that you
can use to ease your transition to this new environment. The second section lists
common C coding practices that cause problems when you use ANSI C.

Recommended Practices
Follow these recommendations as you code:
• Always use the appropriate header file when declaring standard external
functions. Avoid embedding the declaration in your code. This avoids inconsistent
declarations for the same function.
• Always use function prototypes, and write your function prologues in function
prototype form.
• Use the offsetof() macro to derive structure member offsets. The offsetof()
macro is in <stddef.h>.

007–0701–150 3
1: An Overview of ANSI C

• Always use casts when converting.

• Be strict with your use of qualified objects, such as with volatile and const.
Assign the addresses of these objects only to pointers that are so qualified.
• Return a value from all return points of all non-void functions.
• Use only structure designators of the appropriate type as the structure designator
in . and -> expressions (that is, ensure that the right side is a member of the
structure on the left side).
• Always specify the types of integer bitfields as signed or unsigned.

Practices to Avoid
Avoid the following as you code:
• Never mix prototyped and nonprototyped declarations of the same function.
• Never call a function before it has been declared. This may lead to an
incompatible implicit declaration for the function. In particular, this is unlikely to
work for prototyped functions that take a variable number of arguments.
• Never rely on the order in which arguments are evaluated. For example, what is
the result of the code fragment foo(a++, a, ...)?
• Avoid using expressions with side effects as arguments to a function.
• Avoid two side effects to the same data location between two successive sequence
points (for example, x=++x;).
• Avoid declaring functions in a local context, especially if they have prototypes.
• Never access parameters that are not specified in the argument list unless using
the stdarg facilities. Use the stdarg facilities only on a function with an
unbounded argument list (that is, an argument list terminated with …).
• Never cast a pointer type to anything other than another pointer type or an
integral type of the same size (unsigned long), and vice versa. Use a union
type to access the bit-pattern of a pointer as a nonintegral and nonpointer type
(that is, as an array of chars).
• Do not hack preprocessor tokens (for example, FOO/**/BAR).

4 007–0701–150
 C Language Reference Manual

• Never modify a string literal.

• Do not rely on search rules to locate include files that you specify with quotation
marks.

007–0701–150 5
 Chapter 2

C Language Changes

This chapter describes changes to the C language, which include the following:
• "Preprocessor Changes", page 7, discusses two changes in the way the
preprocessor handles string literals and tokens.
• "Changes in Disambiguating Identifiers ", page 10, covers the four characteristics
ANSI C uses to distinguish identifiers.
• "Types and Type Compatibility", page 14, describes ANSI C changes to type
promotions and type compatibility.
• "Function Prototypes", page 18, explains how ANSI C handles function
prototyping.
• "External Name Changes", page 20, discusses the changes in function,
linker-defined, and data area names.
• "Standard Headers", page 22, lists standard header files.

Preprocessor Changes
When compiling in an ANSI C mode (which is the default unless you specify
[-cckr]), ANSI-standard C preprocessing is used. The preprocessor is built into the
TM
compiler and is functionally unchanged from the version appearing on IRIX Release
3.10.
The 3.10 version of the compiler had no built-in preprocessor and used two
standalone preprocessors, for -cckr (cpp(1)) and ANSI C (acpp(5)) preprocessing,
respectively. If you compile using the -o32 option, you can activate acpp or cpp
instead of the built-in preprocessor by using the -oldcpp option, and acpp in -cckr
mode by using the -acpp option. SGI recommends that you always use the built-in
preprocessor, rather than cpp or acpp, because these standalone preprocessors may
not be supported in future releases of the compilers.
acpp is a public domain preprocessor and its source is included in
/usr/src/gnu/acpp.

007–0701–150 7
2: C Language Changes

Traditionally, the C preprocessor performed two functions that are now illegal under
ANSI C. These functions are the substitution of macro arguments within string literals
and the concatenation of tokens after removing a null comment sequence.

Replacement of Macro Arguments in Strings

Suppose you define two macros, IN and PLANT, as shown in this example:
#define IN(x) ‘x’
#define PLANT(y) "placing y in a string"

Later, you invoke them as follows:

IN(hi)
PLANT(foo)

Compiling with -cckr makes these substitutions:

‘hi’
"placing foo in a string"

However, because ANSI C considers a string literal to be an atomic unit, the expected
substitution does not occur. So, ANSI C adopted an explicit preprocessor sequence to
accomplish the substitution.
In ANSI C, adjacent string literals are concatenated. Therefore, this is the result:
"abc" "def" becomes "abcdef".
This concatenation led to a mechanism for quoting a macro argument. When a macro
definition contains one of its formal arguments preceded by a single #, the substituted
argument value is quoted in the output. The simplest example of this is as follows:

Macro: Invoked as: Yields:

#define STRING_LITERAL(a) # a STRING_LITERAL(foo) "foo"

In conjunction with the rule of concatenation of adjacent string literals, the following
macros can be defined:

8 007–0701–150
 C Language Reference Manual

Macro: Invoked as: Yields:

#define ARE(a,c) # a "are" # c ARE(trucks,big) "trucks"" are ""big" or
"trucks are big"

Blanks prepended and appended to the argument value are removed. If the value has
more than one word, each pair of words in the result is separated by a single blank.
Thus, the ARE macro could be invoked as follows:

Macro: Invoked as: Yields:

#define ARE(a,c) # a "are" # c ARE(fat cows, big) "fat cows are big"
or
ARE(fat cows, big)

Avoid enclosing your macro arguments in quotes, because these quotes are placed in
the output string. For example:
ARE ("fat cows", "big") becomes "\"fat cows\" are \"big\""
No obvious facility exists to enclose macro arguments with single quotes.

Token Concatenation
When compiling [-cckr], the value of macro arguments can be concatenated by
entering
#define glue(a,b) a/**/b
glue(FOO,BAR)

The result yields FOOBAR.

This concatenation does not occur under ANSI C, because null comments are replaced
by a blank. However, similar behavior can be obtained by using the ## operator in
-ansi and -xansi mode. ## instructs the precompiled to concatenate the value of a
macro argument with the adjacent token, as illustrated by the following example:

007–0701–150 9
2: C Language Changes

This code: Yields:

#define glue_left(a) GLUED ## a
#define glue_right(a) a ## GLUED
#define glue(a,b) a ## b
glue_left(LEFT) GLUEDLEFT
glue_right(RIGHT) RIGHTGLUED
glue(LEFT,RIGHT) LEFTRIGHT

Furthermore, the resulting token is a candidate for further replacement. Note what
happens in this example:

This code: Yields:

#define HELLO "hello"
#define glue(a,b) a ## b
glue(HEL,LO) "hello"

Changes in Disambiguating Identifiers

Under ANSI C, an identifier has four disambiguating characteristics: its scope,
linkage, name space, and storage duration. Each of these characteristics was used in
traditional C, either implicitly or explicitly. Except in the case of storage duration,
which is either static or automatic, the definitions of these characteristics chosen by
the standard differ in certain ways from those you may be accustomed to, as detailed
in "Scoping Differences", page 10, "Name Space Changes", page 12, and "Changes in
the Linkage of Identifiers", page 12. For a discussion of the same material with a
different focus, see "Disambiguating Names", page 31.

Scoping Differences
ANSI C recognizes four scopes of identifiers: the familiar file and block scopes and
the new function and function prototype scopes.
• Function scope includes only labels. As in traditional C, labels are valid until the
end of the current function.

10 007–0701–150
 C Language Reference Manual

• Block scope rules differ from traditional C in one significant instance: the
outermost block of a function and the block that contains the function arguments
are the same under ANSI C.
For example, when compiling the following code, ANSI C complains of a
redeclaration of x, whereas traditional C hides the argument x with the local
variable x, as if they were in distinct scopes:
int f(x);
int x;
{
int x;
x = 1;
}

• Function prototype scope is a new scope in ANSI C. If an identifier appears within

the list of parameter declarations in a function prototype that is not part of a
function definition, it has function prototype scope, which terminates at the end of
the prototype. This allows any dummy parameter names appearing in a function
prototype to disappear at the end of the prototype.
Consider the following example:
char * getenv (const char * name);
int name;

The int variable name does not conflict with the parameter name because the
parameter went out of scope at the end of the prototype. However, the prototype
is still in scope.
• File scope applies to identifiers appearing outside of any block, function, or
function prototype.
One last discrepancy in scoping rules between ANSI and traditional C concerns the
scope of the function foo() in the following example:
float f;
func0() {
extern float foo() ;
f = foo() ;
}
func1() {
f = foo() ;
}

007–0701–150 11
2: C Language Changes

In traditional C, the function foo() would be of type float when it is invoked in

the function func1(), because the declaration for foo() had file scope, even though
it occurred within a function. ANSI C dictates that the declaration for foo() has
block scope. Thus, there is no declaration for foo() in scope in func1(), and it is
implicitly typed int. This difference in typing between the explicitly and implicitly
declared versions of foo() results in a redeclaration error at compile time, because
they both are linked to the same external definition for foo() and the difference in
typing could otherwise produce unexpected behavior.

Name Space Changes

ANSI C recognizes four distinct name spaces: one for tags, one for labels, one for
members of a particular struct or union, and one for everything else. This division
creates two discrepancies with traditional C:
• In ANSI C, each struct or union has its own name space for its members. This is a
pointed departure from traditional C, in which these members were nothing more
than offsets, allowing you to use a member with a structure to which it does not
belong. This usage is illegal in ANSI C.
• Enumeration constants were special identifiers in versions of SGI C prior to IRIX
Release 3.3. In ANSI C, these constants are simply integer constants that can be
used wherever they are appropriate. Similarly, in ANSI C, other integer variables
can be assigned to a variable of an enumeration type with no error.

Changes in the Linkage of Identifiers

An identifier’s linkage determines which of the references to that identifier refer to
the same object. This terminology formalizes the familiar concept of variables
declared extern and variables declared static and is a necessary augmentation to
the concept of scope.
extern int mytime;
static int yourtime;

In the previous example, both mytime and yourtime have file scope. However,
mytime has external linkage, while yourtime has internal linkage. An object can
also have no linkage, as is the case of automatic variables.

12 007–0701–150
 C Language Reference Manual

The preceding example illustrates another implicit difference between the declarations
of mytime and yourtime. The declaration of yourtime allocates storage for the
object, whereas the declaration of mytime merely references it.
If mytime is initialized as follows, storage is allocated:
int mytime = 0;

In ANSI C terminology, a declaration that allocates storage is referred to as a

definition. This is different from traditional C.
In traditional C, neither of the following declarations was a definition:
extern int bert;
int bert;

In effect, the second declaration included an implicit extern specification. This is not
true in ANSI C.

Note: Objects with external linkage that are not specified as extern at the end of the
compilation unit are considered definitions, and, in effect, initialized to zero. (If
multiple declarations of the object are in the compilation unit, only one needs the
extern specification.)

The effect of this change is to produce “multiple definition” messages from the linker
when two modules contain definitions of the same identifier, even though neither is
explicitly initialized. This is often referred to as the strict ref/def model. A more
relaxed model can be achieved by using the -common compiler flag.
The ANSI C linker issues a warning when it finds redundant definitions, indicating
the modules that produced the conflict. However, the linker cannot determine
whether the definition of the object is explicit. If a definition is given with an explicit
initialization, and that definition is not the linker’s choice, the result may be
incorrectly initialized objects. This is illustrated in the following example:
module1.c:
int ernie;
module2.c:
int ernie = 5;

ANSI C implicitly initializes ernie in module1.c to zero. To the linker, ernie is

initialized in two different modules. The linker warns you of this situation, and

007–0701–150 13
2: C Language Changes

chooses the first such module it encounters as the true definition of ernie. This
module may or may not contain the explicitly initialized copy.

Types and Type Compatibility

Historically, C has allowed free mixing of arithmetic types in expressions and as
arguments to functions. (Arithmetic types include integral and floating point types.
Pointer types are not included.) C’s type promotion rules reduced the number of
actual types used in arithmetic expressions and as arguments to the following three:
int, unsigned, and double. This scheme allowed free mixing of types, but in some
cases forced unnecessary conversions and complexity in the generated code.
One ubiquitous example of unnecessary conversions is when float variables were
used as arguments to a function. C’s type promotion rules often caused two
unwanted, expensive conversions across a function boundary.
ANSI C has altered these rules somewhat to avoid the unnecessary overhead in many
C implementations. This alteration, however, may produce differences in arithmetic
and pointer expressions and in argument passing. For a complete discussion of
operator conversions and type promotions, see Chapter 5, "Operator Conversions",
page 43.

Type Promotion in Arithmetic Expressions

Two differences are noteworthy between ANSI and traditional C. First, ANSI C
relaxes the restriction that all floating point calculations must be performed in double
precision. In the following example, pre-ANSI C compilers are required to convert
each operand to double, perform the operation in double precision, and truncate the
result to float:
extern float f, f0, f1;
addf() {
f = f0 + f1;
}

These steps are not required in ANSI C. In ANSI C, the operation can be done
entirely in single-precision. (In traditional C, these operations were performed in
single-precision if the [-float] compiler option was selected.)
The second difference in arithmetic expression evaluation involves integral
promotions. ANSI C dictates that any integral promotions be “value-preserving.”

14 007–0701–150
 C Language Reference Manual

Traditional C used “unsignedness-preserving” promotions. Consider the following

example:
unsigned short us = 1, them = 2;
int i;
test() {
i = us - them;
}

ANSI C’s value-preserving rules cause each of us and them to be promoted to int,
which is the expression type. The unsignedness-preserving rules, in traditional C,
cause us and them to be promoted to unsigned. The latter case yields a large
unsigned number, whereas ANSI C yields -1. The discrepancy in this case is
inconsequential, because the same bit pattern is stored in the integer i in both cases,
and it is later interpreted as -1.
However, if the case is altered slightly, as in the following example, the result
assigned to f is quite different under the two schemes:
unsigned short us = 1, them = 2;
float f;
test() {
f = us - them;
}

If you use the -wlint option, the compiler will warn about the implicit conversions
from int or unsigned to float.
For more information on arithmetic conversions, see "Arithmetic Conversions", page
45.

Type Promotion and Floating Point Constants

The differences in behavior of ANSI C floating point constants and traditional C
floating point constants can cause numerical and performance differences in code
ported from the traditional C to the ANSI C compiler.
For example, consider the result type of the following computation:
#define PI 3.1415926
float a, b;

b = a * PI;

007–0701–150 15
2: C Language Changes

The result type of b depends on which compilation options you use. Table 2-1, page
16, lists the effects of various options.

Table 2-1 Effect of Compilation Options on Floating Point Conversions

Compilation Option PI Constant Type Promotion Behavior

-cckr double (float)((double)a * PI)
-cckr -float float a * PI
-xansi double (float)((double)a * PI)
-ansi double (float)((double)a * PI)

Each conversion incurs computational overhead.

The -float flag has no effect if you also specify -ansi or -xansi. To prevent the
promotion of floating constants to double (and promoting the computation to a
double precision multiply) you must specify the constant as a single precision floating
point constant. In the previous example, you would use the following statement:
#define PI 3.1415926f /* single precision float */

Traditional C (compiled with the -cckr option) does not recognize the float qualifier,
f, however. Instead, write the constant definition as follows:
#ifdef __STDC__
#define PI 3.1415926f
#else
#define PI 3.1415926
#endif

If you compile with the -ansi, -ansiposix or -xansi options, __STDC__ is

automatically defined, as though you used -D__STDC__= 1 on your compilation
line. Therefore, with the last form of constant definition noted above, the calculation
in the example is promoted as described in Table 2-2, page 17.

16 007–0701–150
 C Language Reference Manual

Table 2-2 Using __STDC__ to Affect Floating Point Conversions

Compilation Option PI Constant Type Promotion Behavior

-cckr double (float)((double)a * PI)
-cckr -float float a * PI
-xansi float a * PI
-ansi float a * PI

Compatible Types
To determine whether or not an implicit conversion is permissible, ANSI C
introduced the concept of compatible types. After promotion, using the appropriate
set of promotion rules, two non-pointer types are compatible if they have the same
size, signedness, and integer or float characteristic, or, in the case of aggregates, are of
the same structure or union type. Except as discussed in the previous section, no
surprises should result from these changes. You should not encounter unexpected
problems unless you are using pointers.
Pointers are compatible if they point to compatible types. No default promotion rules
apply to pointers. Under traditional C, the following code fragment compiled silently:
int *iptr;
unsigned int *uiptr;
foo() {
iptr = uiptr;
}

Under ANSI C, the pointers iptr and uiptr do not point to compatible types
(because they differ in unsignedness), which means that the assignment is illegal.
Insert the appropriate cast to alleviate the problem. When the underlying pointer
type is irrelevant or variable, use the wildcard type void *.

Argument Type Promotions

ANSI C rules for the promotion of arithmetic types when passing arguments to a
function depend on whether or not a prototype is in scope for the function at the
point of the call. If a prototype is not in scope, the arguments are converted using the
default argument promotion rules: short and char types (whether signed or

007–0701–150 17
2: C Language Changes

unsigned) are passed as ints, other integral quantities are not changed, and floating
point quantities are passed as doubles. These rules are also used for arguments in the
variable-argument portion of a function whose prototype ends in ellipses (…).
If a prototype is in scope, an attempt is made to convert each argument to the type
indicated in the prototype prior to the call. The types of conversions that succeed are
similar to those that succeed in expressions. Thus, an int is promoted to a float if
the prototype so indicates, but a pointer to unsigned is not converted to a pointer to
int. ANSI C also allows the implementation greater freedom when passing integral
arguments if a prototype is in scope. If it makes sense for an implementation to pass
short arguments as 16-bit quantities, it can do so.
Use of prototypes when calling functions allows greater ease in coding. However, due
to the differences in argument promotion rules, serious discrepancies can occur if a
function is called both with and without a prototype in scope. Make sure that you
use prototypes consistently and that any prototype is declared to be in scope for all
uses of the function identifier.

Mixed Use of Functions

To reduce the chances of problems occurring when calling a function with and
without a prototype in scope, limit the types of arithmetic arguments in function
declarations. In particular, avoid using short or char types for arguments; their use
rarely improves performance and may raise portability issues if you move your code
to a machine with a smaller word size. This is because function calls made with and
without a prototype in scope may promote the arguments differently. In addition, be
circumspect when typing a function argument float, because you can encounter
difficulties if the function is called without a prototype in scope. With these issues in
mind, you can quickly solve the few problems that may arise.

Function Prototypes
Function prototypes are not new to SGI C. In traditional C, however, the
implementation of prototypes was incomplete. In one case, a significant difference
still exists between the ANSI C and the traditional C implementations of prototypes.
You can prototype functions in two ways. The most common method is simply to
create a copy of the function declaration with the arguments typed, with or without
identifiers for each, such as either of the following:

18 007–0701–150
 C Language Reference Manual

int func(int, float, unsigned [2]);

int func(int i, float f, unsigned u[2]);

You can also prototype a function by writing the function definition in prototype form:
int func(int i, float f, unsigned u[2])
{
< code for func >
}

In each case, a prototype is created for func() that remains in scope for the rest of
the compilation unit.
One area of confusion about function prototypes is that you must write functions that
have prototypes in prototype form. Unless you do this, the default argument
promotion rules apply.
ANSI C elicits an error diagnostic for two incompatible types for the same parameter
in two declarations of the same function. Traditional C elicits an error diagnostic
when the incompatibility may lead to a difference between the bit-pattern of the value
passed in by the caller and the bit-pattern seen in the parameter by the callee.
In the following example, the function func() is declared twice with incompatible
parameter profiles:
int func (float);
int func (f)
float f;
{ … }

The parameter f in func() is assumed to be type double, because the default

argument promotions apply. Error diagnostics in traditional C and ANSI C are
elicited about the two incompatible declarations for func().
The following two situations produce diagnostics from the ANSI C compiler when
you use function prototypes:
• A prototyped function is called with one or more arguments of incompatible type.
(Incompatible types are discussed in "Types and Type Compatibility", page 14.)
• Two incompatible (explicit or implicit) declarations for the same function are
encountered. This version of the compiler scrutinizes duplicate declarations
carefully and catches inconsistencies.

007–0701–150 19
2: C Language Changes

Note: When you use -cckr you do not get warnings about prototyped functions,
unless you specify -prototypes.

External Name Changes

Many well-known UNIX external names that are not covered by the ANSI C standard
are in the user’s name space. These names fall into three categories:
1. Names of functions in the C library
2. Names defined by the linker
3. Names of data areas with external linkage

Changes in Function Names

Names of functions that are in the user’s name space and are referenced by ANSI
C functions in the C library are aliased to counterpart functions whose names are
reserved. In all cases, the new name is formed simply by prefixing an underbar to the
old name. Thus, although it was necessary to change the name of the familiar UNIX
C library function write() to _write(), the function write() remains in the
library as an alias.
The behavior of a program may change if you have written your own versions of
C library functions. If, for example, you have your own version of write(), the C
library continues to use its version of _write().

Changes in Linker-Defined Names

The linker is responsible for defining the standard UNIX symbols end, etext, and
edata, if these symbols are unresolved in the final phases of linking. (See the end(3c)
reference page for more information.) The ANSI C linker has been modified to satisfy
references for _etext, _edata, and _end as well. The ANSI C library reference to
end has been altered to _end.
This mechanism preserves the ANSI C name space, while providing for the definition
of the non-ANSI C forms of these names if they are referenced from existing code.

20 007–0701–150
 C Language Reference Manual

Data Area Name Changes

The names of several well-known data objects used in the ANSI C portion of the C
library were in the user’s name space. These objects are listed in Table 2-3, page 21.
These names were moved into the reserved name space by prefixing their old names
with an underscore. Whether these names are defined in your environment depends
on the compilation mode you are using (the default is -xansi).
Table 2-3, page 21, shows the effect of compilation mode on names and indicates
whether or not these well-known external names are visible when you compile code
in the various modes. The left column has three sets of names. Determine which
versions of these names are visible by examining the corresponding column under
your compilation mode.

Table 2-3 Effect of Compilation Mode on Names

Name -cckr -xansi -ansi

environ environ and environ and _environ only _environ visible
_environ aliased aliased
timezone, tzname, unchanged #define to ANSI C name if _timezone, _tzname,
altzone, daylight using <time.h> _altzone, _daylight
sys_nerr, unchanged identical copies with names identical copies with names
sys_errlist _sys_nerr, _sys_errlist _sys_nerr,
_sys_errlist

Definitions of some of the terms used in Table 2-3, page 21, are as follows:
• “aliased” means the two names access the same object.
• “unchanged” means the well-known version of the name is unaltered.
• “identical copies” means that two copies of the object exist—one with the
well-known name and one with the ANSI C name (prefixed with an underbar).
Applications should not alter these objects.
• “#define” means that a macro is provided in the indicated header to translate the
well-known name to the ANSI C counterpart. Only the ANSI C name exists. You
should include the indicated header if your code refers to the well-known name.
For example, the name tzname is:

007–0701–150 21
2: C Language Changes

– Unchanged when compiling -cckr

– Converted to the reserved ANSI C name (_tzname) by a macro if you include
<time.h> when compiling -xansi
– Available only as the ANSI C version (_tzname) if compiling -ansi (the
default is -xansi)

Standard Headers
Functions in the ANSI C library are declared in a set of standard headers. This set is
self-consistent and is free of name space pollution, when compiling in the pure ANSI
mode. Names that are normally elements of the user’s name space but are specifically
reserved by ANSI are described in the corresponding standard header. Refer to these
headers for information on both reserved names and ANSI library function
prototypes. The following list contains the set of standard headers:
<assert.h>
<ctype.h>
<errno.h>
<float.h>
<limits.h>
<locale.h>
<math.h>
<setjmp.h>
<signal.h>
<stdio.h>
<stddef.h>
<stdarg.h>
<string.h>
<stdlib.h>
<sys/errno.h>
<sys/signal.h>
<time.h>

22 007–0701–150
 Chapter 3

Lexical Conventions

This chapter covers the C lexical conventions including comments and tokens. A
token is a series of contiguous characters that the compiler treats as a unit.
Blanks, tabs, newlines, and comments are collectively known as “white space.” White
space is ignored except as it serves to separate tokens. Some white space is required
to separate otherwise adjacent identifiers, keywords, and constants.
If the input stream has been parsed into tokens up to a given character, the next token
is taken to include the longest string of characters that could possibly constitute a
token.

Comments
The /* characters introduce a comment; the */ characters terminate a comment.
They do not indicate a comment when occurring within a string literal. Comments do
not nest. Once the /* introducing a comment is seen, all other characters are ignored
until the ending */ is encountered.

Identifiers
An identifier, or name, is a sequence of letters, digits, and underscores (_). The first
character cannot be a digit. Uppercase and lowercase letters are distinct. Name length
is unlimited. The terms identifier and name are used interchangeably.

Keywords
The identifiers listed in Table 3-1, page 24, are reserved for use as keywords and
cannot be used for any other purpose.

007–0701–150 23
3: Lexical Conventions

Table 3-1 Reserved Keywords

Keywords
auto default float register struct volatile
break do for return switch while
case double goto short typedef
char else if signed union
const enum int sizeof unsigned
continue extern long static void

Traditional C reserves and ignores the fortran keyword.

Constants
The four types of constants are integer, character, floating, and enumeration. Each
constant has a type, determined by its form and value.
In this section’s discussions of the various types of constants, a unary operator
preceding the constant is not considered part of it. Rather, such a construct is a
constant-expression (see "Constant Expressions", page 66). Thus, the integer constant
0xff becomes an integral constant expression by prefixing a minus sign, for instance,
-0xff. The effect of the - operator is not considered in the discussion of integer
constants.
As an example, the integer constant 0xffffffff has type int in traditional C, with
value -1. It has type unsigned in ANSI C, with value 232 - 1. This discrepancy is
inconsequential if the constant is assigned to a variable of integral type (for example,
int or unsigned), as a conversion occurs. If it is assigned to a double, however, the
value differs as indicated between traditional and ANSI C.

Integer Constants
An integer constant consisting of a sequence of digits is considered octal if it begins
with 0 (zero). An octal constant consists of the digits 0 through 7 only. A sequence of
digits preceded by 0x or 0X is considered a hexadecimal integer. The hexadecimal
digits include [aA] through [fF], which have values of 10 through 15.

24 007–0701–150
 C Language Reference Manual

The suffixes [lL] traditionally indicate integer constants of type long. These suffixes
are allowed, but are superfluous, because int and long are the same size in -o32
and -n32 modes. The ll, LL, lL, and Ll suffixes indicate a long long constant (a
64-bit integral type). Note that long long is not a strict ANSI C type, and a warning
is given for long long constants in -ansi and -ansiposix modes. The following
are examples of long long:
12345LL
12345ll

In ANSI C, an integer constant can be suffixed with uU, in which case its type is
unsigned. (One or both of uU and lL can appear.) An integer constant also has type
unsigned if its value cannot be represented as an int. Otherwise, the type of an
integer constant is int. The following are examples of unsigned long long:
123456ULL
123456ull

Character Constants
A character constant is a character enclosed in single quotation marks, such as ’x’.
The value of a character constant is the numerical value of the character in the
machine’s character set. An explicit new-line character is illegal in a character
constant. The type of a character constant is int.
In ANSI C, a character constant can be prefixed by L, in which case it is a wide
character constant. For example, a wide character constant for ’z’ is written L’z’.
The type of a wide character constant is wchar_t, which is defined in the stddef.h
file.

Special Characters
Some special and nongraphic characters are represented by the escape sequences
shown in Table 3-2, page 26.

007–0701–150 25
3: Lexical Conventions

Table 3-2 Escape Sequences for Nongraphic Characters

Character Name Escape Sequence

newline \n
horizontal tab \t
vertical tab \v
backspace \b
carriage return \r
form feed \f
backslash \\
single quote \’
double quote \"
question mark \?
bell (ANSI C only) \a

The \ddd escape sequence consists of the backslash followed by 1, 2, or 3 octal digits
that specify the value of the desired character. A special case of this construction is
\0 (not followed by a digit), which indicates the ASCII character NUL.
In ANSI C, \x indicates the beginning of a hexadecimal escape sequence. The
sequence is assumed to continue until a character is encountered that is not a member
of the hexadecimal character set 0,1, … 9, [aA], [bB], … [fF]. The resulting unsigned
number cannot be larger than a character can accommodate (decimal 255).
If the character following a backslash is not one of those specified in this section, the
behavior is undefined.

Trigraph Sequences (ANSI C Only)

The character sets of some older machines lack certain members that have come into
common usage. To allow the machines to specify these characters, ANSI C defined an
alternate method for their specification, using sequences of characters that are
commonly available. These sequences are termed trigraph sequences. Nine sequences
are defined; each consists of three characters beginning with two question marks.
Each instance of one of these sequences is translated to the corresponding single
character. Other sequences of characters, perhaps including multiple question marks,

26 007–0701–150
 C Language Reference Manual

are unchanged. Each trigraph sequence with the single character it represents is listed
in the following table.

Table 3-3 Trigraph Sequences

Trigraph Sequence Single Character

??= #
??( [
??/ \
??) ]
??’ ^
??< {
??! |
??> }
??- ~

Floating Constants
A floating constant consists of an integer part, a decimal point, a fraction part, an
[eE], and an optionally signed integer exponent. The integer and fraction parts both
consist of a sequence of digits. Either the integer part or the fraction part (but not
both) can be missing. Either the decimal point or the [eE] and the exponent (not both)
can be missing.
In traditional C, every floating constant has type double.
In ANSI C, floating constants can be suffixed by either [fF] or [lL]. Floating constants
suffixed with [fF] have type float. Those suffixed with [lL] have type
long double, which has greater precision than double in -n32 and -64 modes
and a precision equal to double in -o32 mode.

007–0701–150 27
3: Lexical Conventions

Enumeration Constants
Names declared as enumerators have type int. For a discussion of enumerators, see
"Enumeration Declarations", page 76. For information on the use of enumerators in
expressions, see "Integer and Floating Point Types", page 38.

String Literals
A string literal is a sequence of characters surrounded by double quotation marks, as
in "...". A string literal has type array of char and is initialized with the given
characters. The compiler places a null byte (\0) at the end of each string literal so
that programs that scan the string literal can find its end. A double-quotation
character (") in a string literal must be preceded by a backslash (\). In addition, the
same escapes as those described for character constants can be used. (See "Character
Constants", page 25, for a list of escapes.) A backslash (\) and the immediately
following newline are ignored. Adjacent string literals are concatenated.
In traditional C, all string literals, even when written identically, are distinct.
In ANSI C, identical string literals are not necessarily distinct. Prefixing a string literal
with L specifies a wide string literal. Adjacent wide string literals are concatenated.
As an example, consider the sentence “He said, Hi there.” This sentence could be
written with three adjacent string literals:
"He said, " "Hi " "there.\’"

Operators
An operator specifies an operation to be performed. The operators [ ], ( ), and ?
: must occur in pairs, possibly separated by expressions. The operators # and ## can
occur only in preprocessing directives.
operator can be one of the following:
[ ]( ). ->
++ - - & * + - ~ ! sizeof
/ % << >> < > <= >= == != ^ | && ||
? :
= *= /= %= += -= <<= >>= &= ^= |=
, # ##

28 007–0701–150
 C Language Reference Manual

Individual operations are discussed in Chapter 6, "Expressions and Operators", page

49.

Punctuators
A punctuator is a symbol that has semantic significance but does not specify an
operation to be performed. The punctuators [ ], ( ), and { } must occur in pairs,
possibly separated by expressions, declarations, or statements. The punctuator # can
occur only in preprocessing directives.
punctuator; one of the [ ]( ){ } * , : = ; … #
following:
Some operators, determined by context, are also punctuators. For example, the array
index indicator [ ]is a punctuator in a declaration (see Chapter 7, "Declarations",
page 69), but an operator in an expression (see Chapter 6, "Expressions and
Operators", page 49).

007–0701–150 29
 Chapter 4

Meaning of Identifiers

Traditional C formally based the interpretation of an identifier on two of its attributes:

storage class and type. The storage class determined the location and lifetime of the
storage associated with an identifier; the type determined the meaning of the values
found in the identifier’s storage. Informally, name space, scope, and linkage were also
considered.
ANSI C formalizes the practices of traditional C. An ANSI C identifier is
disambiguated by four characteristics: its scope, name space, linkage, and storage
duration. The ANSI C definitions of these terms differ somewhat from their
interpretations in traditional C.
Storage-class specifiers and their meanings are described in Chapter 7, "Declarations",
page 69. Storage-class specifiers are discussed in this chapter only in terms of their
effect on an object’s storage duration and linkage.
You can find a discussion of focusing on changes to the language in "Changes in
Disambiguating Identifiers ", page 10, and "Types and Type Compatibility", page 14.

Disambiguating Names
This section discusses the ways C disambiguates names: scope, name space, linkage,
and storage class.

Scope
The region of a program in which a given instance of an identifier is visible is called
its scope. The scope of an identifier usually begins when its declaration is seen, or, in
the case of labels and functions, when it is implied by use. Although it is impossible
to have two declarations of the same identifier active in the same scope, no conflict
occurs if the instances are in different scopes. Of the four kinds of scope, two—file
and block—are traditional C scopes. Two other kinds of scope—function and function
prototype—are implied in traditional C and formalized in ANSI C.

007–0701–150 31
4: Meaning of Identifiers

Block Scope

Block scope is the scope of automatic variables (variables declared within a function).
Each block has its own scope. No conflict occurs if the same identifier is declared in
two blocks. If one block encloses the other, the declaration in the enclosed block hides
that in the enclosing block until the end of the enclosed block is reached. The
definition of a block is the same in ANSI C and traditional C, with one exception,
illustrated by the example below:
int f(x);
int x;
{
int x;
x = 1;
}

In ANSI C, the function arguments are in the function body block. Thus, ANSI C will
issue an error of a “redeclaration of x.”
In traditional C, the function arguments are in a separate block that encloses the
function body block. Thus, traditional C would quietly hide the argument x with the
local variable x, because they are in distinct blocks.
ANSI C and traditional C differ in the assignment of block and file scope in a few
instances. See "File Scope", page 32, for more details.

Function Scope

Only labels have function scope. Function scope continues until the end of the
current function.

Function Prototype Scope

If an identifier appears within the list of parameter declarations in a function

prototype that is not part of a function definition (see "Function Declarators and
Prototypes", page 82), it has function prototype scope, which terminates at the end of
the prototype. This termination allows any dummy parameter names appearing in a
function prototype to disappear at the end of the prototype.

File Scope

Identifiers appearing outside of any block, function, or function prototype, have file
scope. This scope continues to the end of the compilation unit. Unlike other scopes,

32 007–0701–150
 C Language Reference Manual

multiple declarations of the same identifier with file scope can exist in a compilation
unit, so long as the declarations are compatible.
Whereas ANSI C assigns block scope to all declarations occurring inside a function,
traditional C assigns file scope to such declarations if they have the storage class
extern. This storage class is implied in all function declarations, whether the
declaration is explicit (as in int foo();) or implicit (if there is no active declaration
for foo() when an invocation is encountered, as in f = foo();). For a further
discussion of this discrepancy, with examples, see "Scoping Differences", page 10.

Name Spaces
In certain cases, the purpose for which an identifier is used may disambiguate it from
other uses of the same identifier appearing in the same scope. This is true, for
example, for tags and is used in traditional C to avoid conflicts between identifiers
used as tags and those used in object or function declarations. ANSI C formalizes this
mechanism by defining certain name spaces. These name spaces are completely
independent. A member of one name space cannot conflict with a member of another.
ANSI C recognizes the following four distinct name spaces:
• Tags: struct, union, and enum tags have a single name space.
• Labels: labels are in their own name space.
• Members: each struct or union has its own name space for its members.
• Ordinary identifiers: ordinary identifiers, including function and object names as
well as user-defined type names, are placed in the last name space.

Name Space Discrepancies Between Traditional and ANSI C

The definition of name spaces causes discrepancies between traditional and ANSI C
in a few situations:
• Structure members in traditional C were nothing more than offsets, allowing the
use of a member with a structure to which it does not belong. This is illegal under
ANSI C.
• Enumeration constants were special identifiers in traditional C prior to IRIX
Release 3.3. In later releases of traditional C, as in ANSI C, these constants are
simply integer constants that can be used anywhere they are appropriate.

007–0701–150 33
4: Meaning of Identifiers

• Labels reside in the same name space as ordinary identifiers in traditional C. Thus,
the following example is legal in ANSI C but not in traditional C:
func() {
int lab;
if (lab) goto lab;
func1() ;
lab:
return;
}

Linkage of Identifiers
Two instances of the same identifier appearing in different scopes may, in fact, refer to
the same entity. For example, the references to a variable, counter, is declared with
file scope in the following example:
extern int counter;

In this example, two separate files refer to the same int object. The association
between the references to an identifier occurring in distinct scopes and the underlying
objects are determined by the identifier’s linkage.
The three kinds of linkage are as follows:
Internal linkage Within a file, all declarations of the same identifier with
internal linkage denote the same object.
External linkage Within an entire program, all declarations of an
identifier with external linkage denote the same object.
No linkage A unique entity, accessible only in its own scope, has
no linkage.
An identifier’s linkage is determined by whether it appears inside or outside a
function, whether it appears in a declaration of a function (as opposed to an object),
its storage-class specifier, and the linkage of any previous declarations of the same
identifier that have file scope. An identifier’s linkage is determined as follows:
1. If an identifier is declared with file scope and the storage-class specifier static, it
has internal linkage.
2. If the identifier is declared with the storage-class specifier extern, or is an
explicit or implicit function declaration with block scope, the identifier has the

34 007–0701–150
 C Language Reference Manual

same linkage as any previous declaration of the same identifier with file scope. If
no previous declaration exists, the identifier has external linkage.
3. If an identifier for an object is declared with file scope and no storage-class
specifier, it has external linkage. (See "Changes in the Linkage of Identifiers", page
12.)
4. All other identifiers have no linkage. This includes all identifiers that do not
denote an object or function, all objects with block scope declared without the
storage-class specifier extern, and all identifiers that are not members of the
ordinary variables name space.
Two declarations of the same identifier in a single file that have the same linkage,
either internal or external, refer to the same object. The same identifier cannot appear
in a file with both internal and external linkage.
This code gives an example where the linkage of each declaration is the same in both
traditional and ANSI C:

static int pete;

extern int bert;
int mom;
int func0() {
extern int mom;
extern int pete;
static int dad;
int bert;
...
}
int func1() {
static int mom;
extern int dad;
extern int bert;
...
}

The declaration of pete with file scope has internal linkage by rule 1 above. This
means that the declaration of pete in func0() also has internal linkage by rule 2
and refers to the same object.
By rule 2, the declaration of bert with file scope has external linkage, because there
is no previous declaration of bert with file scope. Thus, the declaration of bert in
func1() also has external linkage (again by rule 2) and refers to the same (external)

007–0701–150 35
4: Meaning of Identifiers

object. By rule 4, however, the declaration of bert in func0() has no linkage, and
refers to a unique object.
The declaration of mom with file scope has external linkage by rule 3, and, by rule 2,
so does the declaration of mom in func0(). (Again, two declarations of the same
identifier in a single file that both have either internal or external linkage refer to the
same object.) The declaration of mom in func1(), however, has no linkage by rule 4
and thus refers to a unique object.
Last, the declarations of dad in func0() and func1() refer to different objects, as
the former has no linkage and the latter, by rule 2, has external linkage.

Linkage Discrepancies Between Traditional and ANSI C

Traditional and ANSI C differ on the concept of linkage in the following important
ways:
• In traditional C, a function can be declared with block scope and the storage-class
specifier static. The declaration is given internal linkage. Only the storage class
extern can be specified in function declarations with block scope in ANSI C.
• In traditional C, if an object is declared with block scope and the storage-class
specifier static, and a declaration for the object with file scope and internal
linkage exists, the block scope declaration has internal linkage. In ANSI C, an
object declared with block scope and the storage-class specifier static has no
linkage.
Traditional and ANSI C handle the concepts of reference and definition differently.
For example:
extern int mytime;
static int yourtime;

In the preceding example, both mytime and yourtime have file scope. As discussed
previously, mytime has external linkage, while yourtime has internal linkage.
However, there is an implicit difference, which exists in both ANSI and traditional C,
between the declarations of mytime and yourtime in the preceding example. The
declaration of yourtime allocates storage for the object, whereas the declaration of
mytime merely references it. If mytime had been initialized, as in the following
example, it would also have allocated storage:
int mytime=0;

36 007–0701–150
 C Language Reference Manual

A declaration that allocates storage is referred to as a definition.

In traditional C, neither of the two declarations below is a definition:
extern int bert;
int bert;

In effect, the second declaration includes an implicit extern specification. ANSI C

does not include such an implicit specification.

Note: In ANSI C, objects with external linkage that are not specified as extern at the
end of the compilation unit are considered definitions, and, in effect, initialized to
zero. (If multiple declarations of the object occur in the compilation unit, only one
need have the extern specification.)

If two modules contain definitions of the same identifier, the linker complains of
“multiple definitions,” even though neither is explicitly initialized.
The ANSI C linker issues a warning when it finds redundant definitions, indicating
which modules produced the conflict. However, the linker cannot determine if the
initialization of the object is explicit. This may result in incorrectly initialized objects
if another module fails to tag the object with extern.
Thus, consider the following example:
module1.c:
int ernie;
module2.c:
int ernie = 5;

ANSI C implicitly initializes ernie in module1.c to zero. To the linker, ernie is

initialized in two different modules. The linker warns you of this situation, and
chooses the first such module it encountered as the true definition of ernie. This
module may or may not be the one containing the explicitly initialized copy.

Storage Duration
Storage duration denotes the lifetime of an object. Storage duration is of two types:
static and automatic.

007–0701–150 37
4: Meaning of Identifiers

Objects declared with external or internal linkage, or with the storage-class specifier
static, have static storage duration. If these objects are initialized, the initialization
occurs once, prior to any reference.
Other objects have automatic storage duration. Storage is newly allocated for these
objects each time the block that contains their declaration is entered, unless the object
has a variable length array type. If the object is variably modified, and the block is
entered by a jump to a labeled statement, then the behavior is undefined.
If an object with automatic storage duration is initialized, the initialization occurs
each time the block is entered at the top. This is not guaranteed to occur if the block
is entered by a jump to a labeled statement.

Object Types
The C language supports three fundamental types of objects: character, integer, and
floating point.

Character Types
Objects declared as characters (char) are large enough to store any member of the
implementation’s character set. If a genuine character from that character set is stored
in a char variable, its value is equivalent to the integer code for that character. Other
quantities may be stored into character variables, but the implementation is machine
dependent. In this implementation, char is unsigned by default.
The ANSI C standard has added multibyte and wide character types. In the initial
SGI release of ANSI C, wide characters are of type unsigned char, and multibyte
characters are of length one. (See the header files stddef.h and limits.h for more
information.)

Integer and Floating Point Types

Up to five sizes of integral types (signed and unsigned) are available: char,
short, int, long, and long long. Up to three sizes of floating point types are
available. The sizes are shown in Table 4-1, page 39. (The values in the table apply to
both ANSI and traditional C, with the exceptions noted in the subsequent discussion.)

38 007–0701–150
 C Language Reference Manual

Table 4-1 Storage Class Sizes

Size in Bits Size in Bits Size in Bits

Type (-o32) (-n32) (-64)
char 8 8 8
short 16 16 16
int 32 32 32
long 32 32 64
long long 64 64 64
float 32 32 32
double 64 64 64
long double 64 128 128
pointer 32 32 64

Although SGI supports long double as a type in -cckr mode, this is viewed as an
extension to traditional C and is ignored in subsequent discussions pertinent only to
traditional C.
Differences exist between -o32 mode, -n32 mode, and -64 mode compilations.
Types long and int have different sizes (and ranges) in 64-bit mode; type long always
has the same size as a pointer value. A pointer (or address) has a 64-bit
representation in 64-bit mode and a 32-bit representation in both 32-bit modes.
Therefore, an int object has a smaller size than a pointer object in 64-bit mode.
The long long type is not a valid ANSI C type, so a warning is elicited for every
occurrence of long long in the source program text in -ansi and -ansiposix modes.
The long double type has equal range in old 32-bit, new 32-bit, and 64-bit mode,
but it has increased precision in -n32 and -64 modes.
Characteristics of integer and floating point types are defined in the standard header
files <limits.h> and <float.h>. The range of a signed integral type of size n is
[(-2n-1)... (2n-1 -1)]. The range of an unsigned version of the type is [0... (2n -1)].
Enumeration constants were special identifiers under various versions of traditional
C, before IRIX Release 3.3. In ANSI C, these constants are simply integer constants
that may be used anywhere. Similarly, ANSI C allows the assignment of other integer
variables to variables of enumeration type, with no error.

007–0701–150 39
4: Meaning of Identifiers

You can find additional information on integers, floating points, and structures in the
following tables:
• For integer types and ranges, see Table A-1, page 133
• For floating point types and ranges, see Table A-2, page 135
• For structure alignment, see Table A-3, page 137

Derived Types
Because objects of the types mentioned in "Integer and Floating Point Types", page 38,
can be interpreted usefully as numbers, this manual refers to them as arithmetic
types. The types char, enum, and int of all sizes (whether unsigned or not) are
collectively called integral types. The float and double types are collectively called
floating types. Arithmetic types and pointers are collectively called scalar types.
The fundamental arithmetic types can be used to construct a conceptually infinite
class of derived types, such as the following:
• Arrays of objects of most types
• Functions that return objects of a given type
• Pointers to objects of a given type
• Structures that contain a sequence of objects of various types
• Unions capable of containing any one of several objects of various types
In general, these constructed objects can be used as building blocks for other
constructed objects.

void Type
The void type specifies an empty set of values. It is used as the type returned by
functions that generate no value. The void type never refers to an object and
therefore, is not included in any reference to object types.

40 007–0701–150
 C Language Reference Manual

Objects and lvalues

An object is a manipulatable region of storage. An lvalue is an expression referring to
an object. An obvious example of an lvalue expression is an identifier. Some
operators yield lvalues. For example, if E is an expression of pointer type, then *E is
an lvalue expression referring to the object to which E points. The term lvalue comes
from the term “left value.” In the assignment expression E1 = E2, the left operand
E1 must be an lvalue expression.

Most lvalues are modifiable, meaning that the lvalue may be used to modify the
object to which it refers. Examples of lvalues that are not modifiable include array
names, lvalues with incomplete type, and lvalues that refer to an object, part or all of
which is qualified with const (see "Type Qualifiers", page 77). Whether an lvalue
appearing in an expression must be modifiable is usually obvious. For example, in the
assignment expression E1 = E2, E1 must be modifiable. This document makes the
distinction between modifiable and unmodifiable lvalues only when it is not obvious.

007–0701–150 41
 Chapter 5

Operator Conversions

A number of operators can, depending on the types of their operands, cause an

implicit conversion of some operands from one type to another. The following
discussion explains the results you can expect from these conversions. The
conversions demanded by most operators are summarized in "Arithmetic
Conversions", page 45. When necessary, a discussion of the individual operators
supplements the summary.

Conversions of Characters and Integers

You can use a character or a short integer wherever you can use an integer. Characters
are unsigned by default. In all cases, the value is converted to an integer. Conversion
of a shorter integer to a longer integer preserves the sign. Traditional C uses
“unsigned preserving integer promotion” (unsigned short to unsigned int),
while ANSI C uses “value preserving integer promotion” (unsigned short to int).
A longer integer is truncated on the left when converted to a shorter integer or to a
char. Excess bits are discarded.

Conversions of Float and Double

Historically in C, expressions containing floating point operands (either float or
double) were calculated using double precision. This is also true of calculations in
traditional C, unless you have specified the compiler option -float. With the
-float option, calculations involving floating point operands and no double or
long double operands take place in single precision. The -float option has no
effect on argument promotion rules at function calls or on function prototypes.
ANSI C performs calculations involving floating point in the same precision as if
-float had been specified in traditional C, except when floating point constants are
involved.
In traditional C, specifying the -float option coerces floating point constants into
type float if all the other subexpressions are of type float. This is not the case in ANSI
C. ANSI C considers all floating point constants to be implicitly double precision, and
operations involving such constants therefore take place in double precision. To force
single precision arithmetic in ANSI C, use the f or F suffix on floating point

007–0701–150 43
5: Operator Conversions

constants. To force long double precision on constants, use the l or L suffix. For
example, 3.14l is long double precision, 3.14 is double precision, and 3.14f is
single precision in ANSI C.
For a complete discussion with examples, see "Type Promotion and Floating Point
Constants", page 15.

Conversion of Floating and Integral Types

Conversions between floating and integral values are machine-dependent. SGI uses
IEEE floating point, in which the default rounding mode is to nearest, or in case of a
tie, to even. Floating point rounding modes can be controlled using the facilities of
fpc. Floating point exception conditions are discussed in the introductory paragraph
of Chapter 6, "Expressions and Operators", page 49.
When a floating value is converted to an integral value, the rounded value is
preserved as long as it does not overflow. When an integral value is converted to a
floating value, the value is preserved unless a value of more than six significant digits
is being converted to single precision, or fifteen significant digits is being converted to
double precision.

Conversion of Pointers and Integers

An expression of integral type can be added to or subtracted from an object pointer.
In such a case, the integer expression is converted as specified in the discussion of the
addition operator in "Additive Operators", page 59. Two pointers to objects of the
same type can be subtracted. In this case, the result is converted to an integer as
specified in the discussion of the subtraction operator, in "Additive Operators", page
59.

Conversion of unsigned Integers

When an unsigned integer is converted to a longer unsigned or signed integer,
the value of the result is preserved. Thus, the conversion amounts to padding with
zeros on the left.

44 007–0701–150
 C Language Reference Manual

When an unsigned integer is converted to a shorter signed or unsigned integer,

the value is truncated on the left. If the result is signed, this truncation may produce
a negative value.

Arithmetic Conversions
Many types of operations in C require two operands to be converted to a common
type. Two sets of conversion rules are applied to accomplish this conversion. The first,
referred to as the integral promotions, defines how integral types are promoted to one
of several integral types that are at least as large as int. The second, called the usual
arithmetic conversions, derives a common type in which the operation is performed.
ANSI C and traditional C follow different sets of these rules.

Integral Promotions
The difference between the ANSI C and traditional versions of the conversion rules is
that the traditional C rules emphasize preservation of the (un)signedness of a
quantity, while ANSI C rules emphasize preservation of its value.

In traditional C, operands of types char, unsigned char, and unsigned short

are converted to unsigned int. Operands of types signed char and short are
converted to int.
ANSI C converts all char and short operands, whether signed or unsigned, to
int. Only operands of type unsigned int, unsigned long, and
unsigned long long may remain unsigned.

Usual Arithmetic Conversions

Besides differing in emphasis on signedness and value preservation, the usual
arithmetic conversion rules of ANSI C and traditional C also differ in the precision of
the chosen floating point type.
The following subsections describe two sets of conversion rules, one for traditional C,
and the other for ANSI C. Each set is ordered in decreasing precedence. In any
particular case, the rule that applies is the first whose conditions are met.

007–0701–150 45
5: Operator Conversions

Each rule specifies a type, referred to as the result type. Once a rule has been chosen,
each operand is converted to the result type, the operation is performed in that type,
and the result is of that type.

Traditional C Conversion Rules

The traditional C conversion rules are as follows:

• If any operand is of type double, the result type is double.
• If any operand is of type float, the result type is float if you have specified the
[-float] switch. Otherwise, the result type is double.
• The integral promotions are performed on each operand as follows:

If one of the operands is of type: The result is of type:

unsigned long long unsigned long long
long long long long
unsigned long unsigned long
long long
unsigned int unsigned int
otherwise int

ANSI C Conversion Rules

The ANSI C rules are as follows:

• If any operand is of type long double, the result type is long double.
• If any operand is of type double, the result type is double.
• If any operand is of type float, the result type is float.
• The integral promotions are performed on each operand as follows:

46 007–0701–150
 C Language Reference Manual

If one of the operands is of type: The result is of type:

unsigned long long unsigned long long
long long long long
unsigned long unsigned long
long long
unsigned int unsigned int
otherwise int

Conversion of Other Operands

The following three sections discuss conversion of lvalues, function designators, void
objects, and pointers.

Conversion of lvalues and Function Designators

Except as noted, if an lvalue that has type “array of <type>” appears as an operand,
it is converted to an expression of the type “pointer to <type>.” The resultant pointer
points to the initial element of the array. In this case, the resultant pointer ceases to be
an lvalue. (For a discussion of lvalues, see "Objects and lvalues", page 41.)
A function designator is an expression that has function type. Except as noted, a
function designator appearing as an operand is converted to an expression of type
“pointer to function.”

Conversion of void Objects

The (nonexistent) value of a void object cannot be used in any way, and neither
explicit nor implicit conversion can be applied. Because a void expression denotes a
nonexistent value, such an expression can be used only as an expression statement
(see "Expression Statement", page 93), or as the left operand of a comma expression
(see "Comma Operator", page 66).
An expression can be converted to type void by use of a cast. For example, this
makes explicit the discarding of the value of a function call used as an expression
statement.

007–0701–150 47
5: Operator Conversions

Conversion of Pointers
A pointer to void can be converted to a pointer to any object type and back without
change in the underlying value.
The NULL pointer constant can be specified either as the integral value zero, or the
value zero cast to a pointer to void. If a NULL pointer constant is assigned or
compared to a pointer to any type, it is appropriately converted.

48 007–0701–150
 Chapter 6

Expressions and Operators

This chapter discusses the various expressions and operators available in C. The
sections describing expressions and operators are presented roughly in order of
precedence.

Precedence and Associativity Rules in C

Operators in C have rules of precedence and associativity that determine how
expressions are evaluated. Table 6-2, page 50, lists the operators and indicates the
precedence and associativity of each. Within each row, the operators have the same
precedence. Parentheses can be used to override these rules.
Table 6-1, page 49, shows some simple examples of precedence and associativity.

Table 6-1 Precedence and Associativity Examples

Expression Results Comments

3+ 2* 5 13 Multiplication is done before addition.
3 + (2 * 5) 13 Parentheses follow the precedence rules, but clarify the
expression for the reader.
(3 + 2) * 5 25 Parentheses override the precedence rules.
TRUE || TRUE && FALSE 1 (true) Logical AND has higher priority than logical OR.
TRUE || (TRUE && FALSE) 1 (true) Parentheses follow the precedence rules, but clarify the
expression for the reader.
(TRUE || TRUE) && FALSE 0 Parentheses override the precedence rules.
(false)

Except as indicated by the syntax or specified explicitly in this chapter, the order of
evaluation of expressions, as well as the order in which side-effects take place, is
unspecified. The compiler can arbitrarily rearrange expressions involving a
commutative and associative operator (*, +, &, |, ^).
Table 6-2, page 50, lists the precedence and associativity of all operators.

007–0701–150 49
6: Expressions and Operators

Table 6-2 Operator Precedence and Associativity

Tokens (From High to Low Priority) Operators Class Associativity

Identifiers, constants, string literal, Primary expression Primary
parenthesized expression
() [] -> . Function calls, subscripting, indirect Postfix L-R
selection, direct selection
++ -- Increment, decrement (postfix) Postfix L-R
++ -- Increment, decrement (prefix) Prefix R-L
! ~ + - & sizeof * Logical and bitwise NOT, unary plus Unary R-L
and minus, address, size, indirection
( type ) Cast Unary R-L
*/ % Multiplicative Binary L-R
+- Additive Binary L-R
<< >> Left shift, right shift Binary L-R
< <= > >= Relational comparisons Binary L-R
== != Equality comparisons Binary L-R
& Bitwise and Binary L-R
^ Bitwise exclusive or Binary L-R
| Bitwise inclusive or Binary L-R
&& Logical and Binary L-R
|| Logical or Binary L-R
? : conditional Ternary R-L
= += -= *= /= %= ^= &= |= <<= >>= Assignment Binary R-L
, Comma Binary L-R

50 007–0701–150
 C Language Reference Manual

Primary Expressions
The following are all considered “primary expressions:”
Identifiers An identifier referring to an object is an lvalue. An
identifier referring to a function is a function
designator. lvalues and function designators are
discussed in “Conversion of lvalues and Function
Designators” on page 59.
Constants A constant’s type is determined by its form and value,
as described in "Constants", page 24.
String literals A string literal’s type is “array of char,” subject to
modification, as described in "Conversions of
Characters and Integers", page 43.
Parenthesized A parenthesized expression’s type and value are
expressions identical to those of the unparenthesized expression.
The presence of parentheses does not affect whether the
expression is an lvalue, rvalue, or function designator.
For information on expressions, see “Constant
Expressions” on page 79.

Postfix Expressions
Postfix expressions involving ., ->, subscripting, and function calls associate left to
right. The syntax for these expressions is as follows:
postfix-expression: primary-expression
postfix-expression [expression]
postfix-expression (argument-expression-list opt)
postfix-expression. identifier
postfix-expression -> identifier
postfix-expression ++
postfix-expression - -

007–0701–150 51
6: Expressions and Operators

argument-expression-list: argument-expression
argument-expression-list, argument-expression

Subscripts
A postfix expression followed by an expression in square brackets is a subscript.
Usually, the postfix expression has type “pointer to <type>”, the expression within the
square brackets has type int, and the type of the result is <type>. However, it is
equally valid if the types of the postfix expression and the expression in brackets are
reversed. This is because the expression E1[E2] is identical (by definition) to
*((E1)+(E2)). Because addition is commutative, E1 and E2 can be interchanged.
You can find more information on this notation in the discussions on identifiers and
in the discussion of the * and + operators (in "Unary Operators", page 55, and
"Additive Operators", page 59), respectively.

Function Calls
The syntax of function call postfix expressions is as follows:
postfix-expression (argument-expression-listopt)
argument-expression-list: argument-expression
argument-expression-list, argument-expression
A function call is a postfix expression followed by parentheses containing a (possibly
empty) comma-separated list of expressions that are the arguments to the function.
The postfix expression must be of type “function returning <type>.” The result of the
function call is of type <type>, and is not an lvalue.
The behavior of function calls is as follows:
• If the function call consists solely of a previously unseen identifier foo, the call
produces an implicit declaration as if, in the innermost block containing the call,
the following declaration had appeared:
extern int foo();

• If a corresponding function prototype that specifies a type for the argument being
evaluated is in force, an attempt is made to convert the argument to that type.

52 007–0701–150
 C Language Reference Manual

• If the number of arguments does not agree with the number of parameters
specified in the prototype, the behavior is undefined.
• If the type returned by the function as specified in the prototype does not agree
with the type derived from the expression containing the called function, the
behavior is undefined. Such a scenario may occur for an external function
declared with conflicting prototypes in different files.
• If no corresponding prototype is in scope or if the argument is in the variable
argument section of a prototype that ends in ellipses (…), the argument is
converted according to the following default argument promotions:
– Type float is converted to double.
– Array and function names are converted to corresponding pointers.
– When using traditional C, types unsigned short and unsigned char are
converted to unsigned int, and types signed short and signed char
are converted to signed int.
– When using ANSI C, types short and char, whether signed or unsigned,
are converted to int.
• In preparing for the call to a function, a copy is made of each actual argument.
Thus, all argument passing in C is strictly by value. A function can change the
values of its parameters, but these changes cannot affect the values of the actual
arguments. It is possible to pass a pointer on the understanding that the function
can change the value of the object to which the pointer points. (Arguments that
are array names can be changed as well, because these arguments are converted to
pointer expressions.)
• Because the order of evaluation of arguments is unspecified, side effects may be
delayed until the next sequence point, which occurs at the point of the actual call
and after all arguments have been evaluated. (For example, in the function call
func(foo++), the incrementation of foo may be delayed.)
• Recursive calls to any function are permitted.
SGI recommends consistent use of prototypes for function declarations and
definitions. Do not mix prototyped and nonprototyped function declarations and
definitions. Even though the language allows it, never call functions before you
declare them. This results in an implicit nonprototyped declaration that may be
incompatible with the function definition.

007–0701–150 53
6: Expressions and Operators

Structure and Union References

A postfix expression followed by a dot followed by an identifier denotes a structure
or union reference. The syntax is as follows:

postfix-expression. identifier

The postfix expression must be a structure or a union, and the identifier must name a
member of the structure or union. The value is the value of the named member of the
structure or union, and is an lvalue if the first expression is an lvalue.The result has
the type of the indicated member and the qualifiers of the structure or union.

Indirect Structure and Union References

A postfix-expression followed by an arrow (built from – and >) followed by an
identifier is an indirect structure or union reference. The syntax is as follows:

postfix-expression -> identifier

The postfix expression must be a pointer to a structure or a union, and the identifier
must name a member of that structure or union. The result is an lvalue referring to
the named member of the structure or union to which the postfix expression points.
The result has the type of the selected member, and the qualifiers of the structure or
union to which the postfix expression points. Thus, the expression E1->MOS is the
same as (*E1).MOS.
Structures and unions are discussed in "Structure and Union Declarations", page 72.

postfix ++ and postfix - -

The syntax of postfix ++ and postfix -- is as follows:

postfix-expression ++

postfix-expression --

54 007–0701–150
 C Language Reference Manual

When postfix ++ is applied to a modifiable lvalue, the result is the value of the
object referred to by the lvalue. After the result is noted, the object is incremented by
1 (one). See the discussions in "Additive Operators", page 59, and "Assignment
Operators", page 65, for information on conversions. The type of the result is the
same as the type of the lvalue expression. The result is not an lvalue.
When postfix -- is applied to a modifiable lvalue, the result is the value of the
object referred to by the lvalue. After the result is noted, the object is decremented by
1 (one). See the discussions in "Additive Operators", page 59, and "Assignment
Operators", page 65, for information on conversions. The type of the result is the
same as the type of the lvalue expression. The result is not an lvalue.
For both postfix ++ and postfix -- operators, updating the stored value of the
operand may be delayed until the next sequence point.

Unary Operators
Expressions with unary operators associate from right to left. The syntax for unary
operators is as follows:
unary-expression: postfix-expression
++ unary-expression
- - unary-expression
unary-operator cast-expression
sizeof unary-expression
sizeof (type-name)
unary-operator: one of * & - ! ~ +
Except as noted, the operand of a unary operator must have arithmetic type.

Address-of and Indirection Operators

The unary * operator means “indirection”; the cast expression must be a pointer, and
the result is either an lvalue referring to the object to which the expression points, or
a function designator. If the type of the expression is “pointer to <type>”, the type of
the result is <type>.

007–0701–150 55
6: Expressions and Operators

The operand of the unary & operator can be either a function designator or an lvalue
that designates an object. If it is an lvalue, the object it designates cannot be a bitfield,
and it cannot be declared with the storage class register. The result of the unary &
operator is a pointer to the object or function referred to by the lvalue or function
designator. If the type of the lvalue is <type>, the type of the result is “pointer to
<type>”.

Unary + and - Operators

The result of the unary - operator is the negative of its operand. The integral
promotions are performed on the operand, and the result has the promoted type and
the value of the negative of the operand. Negation of unsigned quantities is
analogous to subtracting the value from 2n, where n is the number of bits in the
promoted type.
The unary + operator exists only in ANSI C. The integral promotions are used to
convert the operand. The result has the promoted type and the value of the operand.

Unary ! and ~ Operators

The result of the logical negation operator ! is 1 if the value of its operand is zero,
and 0 if the value of its operand is nonzero. The type of the result is int. The logical
negation operator is applicable to any arithmetic type and to pointers.
The ~ operator (bitwise not) yields the one’s complement of its operand. The usual
arithmetic conversions are performed. The type of the operand must be integral.

Prefix ++ and - - Operators

The prefix operators ++ and -- increment and decrement their operands. Their
syntax is as follows:

++ unary-expression

-- unary-expression

The object referred to by the modifiable lvalue operand of prefix ++ is incremented.

The expression value is the new value of the operand but is not an lvalue. The

56 007–0701–150
 C Language Reference Manual

expression ++x is equivalent to x += 1. See the discussions in "Additive Operators",

page 59, and "Assignment Operators", page 65, for information on conversions.
The prefix -- decrements its lvalue operand in the same way that prefix ++
increments it.

sizeof Unary Operator

The sizeof operator yields the size in bytes of its operand. The size of a char is 1
(one). Its major use is in communication with routines such as storage allocators and
I/O systems. The syntax of the sizeof operator is as follows:

sizeof unary-expression

sizeof (type-name)

The operand of sizeof cannot have function or incomplete type, or be an lvalue that
denotes a bitfield. It can be an object or a parenthesized type name. In traditional C,
the type of the result is unsigned. In ANSI C, the type of the result is size_t,
which is defined in <stddef.h> as unsigned int (in -o32 and -n32 modes) or as
unsigned long (in -64 mode). The result is a constant and can be used anywhere a
constant is required.
When applied to an array, sizeof returns the total number of bytes in the array. The
size is determined from the declaration of the object in the unary expression. For
variable length array types, the result is not a constant expression and is computed at
run time.
The sizeof operator can also be applied to a parenthesized type name. In that case,
it yields the size in bytes of an object of the indicated type.
When sizeof is applied to an aggregate, the result includes space used for padding,
if any.

Cast Operators
A cast expression preceded by a parenthesized type name causes the value of the
expression to convert to the indicated type. This construction is called a cast. Type

007–0701–150 57
6: Expressions and Operators

names are discussed in "Type Names", page 86. The syntax of a cast expression is as
follows:
cast-expression: unary-expression
(type-name) cast-expression
The type name specifies a scalar type or void, and the operand has scalar type.
Because a cast does not yield an lvalue, the effect of qualifiers attached to the type
name is inconsequential.
When an arithmetic value is cast to a pointer, and vice versa, the appropriate number
of bits are simply copied unchanged from one type of value to the other. Be aware of
the possible truncation of pointer values in 64-bit mode compilation, when a pointer
value is converted to an (unsigned) int.

Multiplicative Operators
The multiplicative operators *, /, and % group from left to right. The usual arithmetic
conversions are performed. The following is the syntax for the multiplicative
operators:
multiplicative expression: cast-expression
multiplicative-expression * cast-expression
multiplicative-expression / cast-expression
multiplicative-expression % cast-expression
Operands of * and / must have arithmetic type. Operands of % must have integral
type.
The binary * operator indicates multiplication, and its result is the product of the
operands.
The binary / operator indicates division of the first operator (dividend) by the second
(divisor). If the operands are integral and the value of the divisor is 0, SIGTRAP is
signalled. Integral division results in the integer quotient whose magnitude is less
than or equal to that of the true quotient, and with the same sign.

58 007–0701–150
 C Language Reference Manual

The binary % operator yields the remainder from the division of the first expression
(dividend) by the second (divisor). The operands must be integral. The remainder
has the same sign as the dividend, so that the equality below is true when the divisor
is nonzero:
(dividend / divisor) * divisor + dividend % divisor == dividend

If the value of the divisor is 0, SIGTRAP is signalled.

Additive Operators
The additive operators + and - associate from left to right. The usual arithmetic
conversions are performed.The syntax for the additive operators is as follows:
additive-expression: multiplicative-expression
additive-expression + multiplicative-expression
additive-expression - multiplicative-expression
In addition to arithmetic types, the following type combinations are acceptable for
additive expressions:
• For addition, one operand is a pointer to an object type and the other operand is
an integral type.
• For subtraction,
– Both operands are pointers to qualified or unqualified versions of compatible
object types.
– The left operand is a pointer to an object type, and the right operand has
integral type.
The result of the + operator is the sum of the operands. The result of the - operator is
the difference of the operands.
When an operand of integral type is added to or subtracted from a pointer to an
object type, the integral operand is first converted to an address offset by multiplying
it by the length of the object to which the pointer points. The result is a pointer of the
same type as the original pointer.

007–0701–150 59
6: Expressions and Operators

For instance, suppose a has type “array of <object>”, and p has type “pointer to
<object>” and points to the initial element of a. Then the result of p + n, where n is
an integral operand, is the same as &a[n].
If two pointers to objects of the same type are subtracted, the result is converted (by
division by the length of the object) to an integral quantity representing the number
of objects separating them. Unless the pointers point to objects in the same array, the
result is undefined. The actual type of the result is int in traditional C, and
ptrdiff_t (defined in <stddef.h> as int in -o32 and -n32 modes and as long
in -64 mode) in ANSI C.

Shift Operators
The shift operators << and >> associate from left to right. Each operand must be an
integral type. The integral promotions are performed on each operand. The syntax is
as follows:
shift-expression: additive-expression
shift-expression << additive-expression
shift-expression >> additive-expression
The type of the result is that of the promoted left operand. If the right operand is
negative, greater than, or equal to the length in bits of the promoted left operand, the
result is undefined.
The value of E1 << E2 is E1 (interpreted as a bit pattern) left-shifted E2 bits.
Vacated bits are filled with zeros.
The value of E1 >> E2 is E1 right-shifted E2 bit positions. If E1 is unsigned, or if
it is signed and its value is nonnegative, vacated bits are filled with zeros. If E1 is
signed and its value is negative, vacated bits are filled with ones.

Relational Operators
The relational operators associate from left to right. The syntax is as follows:
relational-expression: shift-expression
relational-expression < shift-expression

60 007–0701–150
 C Language Reference Manual

relational-expression > shift-expression

relational-expression <= shift-expression
relational-expression >= shift-expression
The operators < (less than), > (greater than), <= (less than or equal to), and >=
(greater than or equal to) all yield a result of type int with the value 0 if the
specified relation is false and 1 if it is true.
The operands must be one of the following:
• Both arithmetic, in which case the usual arithmetic conversions are performed on
them
• Both pointers to qualified or unqualified versions of compatible object types
• Both pointers to qualified or unqualified versions of compatible incomplete types
When two pointers are compared, the result depends on the relative locations in the
address space of the pointed-to objects. Pointer comparison is portable only when the
pointers point to objects in the same aggregate. In particular, no correlation is
guaranteed between the order in which objects are declared and their resulting
addresses.

Equality Operators
The == (equal to) and the != (not equal to) operators are exactly analogous to the
relational operators except for their lower precedence. (For example, a < b == c <
d is 1 whenever a < b and c < d have the same truth value.) The syntax of the
equality operators is as follows:
equality-expression: relational-expression
equality-expression == relational-expression
equality-expression != relational-expression
The operands must be one of the following:
• Both arithmetic, in which case the usual arithmetic conversions are performed on
them
• Both pointers to qualified or unqualified versions of compatible types

007–0701–150 61
6: Expressions and Operators

• A pointer to an object or incomplete type, and a pointer to qualified or unqualified

void type
• A pointer and a null pointer constant
The semantics detailed in "Relational Operators", page 60, apply if the operands have
types suitable for those operators. Combinations of other operands have the
following behavior:
• Two null pointers to object or incomplete types are equal. If two pointers to such
types are equal, they either are null, point to the same object, or point to one
object beyond the end of an array of such objects.
• Two pointers to the same function are equal, as are two null function pointers.
Two function pointers that are equal are either both null or both point to the same
function.

Bitwise AND Operator

Each operand of the bitwise AND operator must have integral type. The usual
arithmetic conversions are performed. The syntax is as follows:
AND-expression: equality-expression
AND-expression & equality-expression
The result is the bitwise AND function of the operands, that is, each bit in the result
is 0 unless the corresponding bit in each of the two operands is 1.

Bitwise Exclusive OR Operator

Each operand of the bitwise exclusive OR operator must have integral type. The
usual arithmetic conversions are performed. The syntax is as follows:
exclusive-OR-expression: AND-expression
exclusive-OR-expression ^ AND- expression
The result has type int, long, or long long, and the value is the bitwise exclusive
OR function of the operands. That is, each bit in the result is 0 unless the
corresponding bit in one of the operands is 1, and the corresponding bit in the other
operand is 0.

62 007–0701–150
 C Language Reference Manual

Bitwise Inclusive OR Operator

Each operand of the bitwise inclusive OR operator must have integral type. The usual
arithmetic conversions are performed. The syntax is as follows:
inclusive-OR-expression: exclusive-OR-expression
inclusive-OR-expression | exclusive-OR-expression
The result has type int, long, or long long, and the value is the bitwise inclusive
OR function of the operands. That is, each bit in the result is 0 unless the
corresponding bit in at least one of the operands is 1.

Logical AND Operator

Each of the operands of the logical AND operator must have scalar type. The &&
operator associates left to right. The syntax is as follows:
logical-AND-expression: inclusive-OR-expression
logical-AND-expression && inclusive-OR-expression
The result has type int. If neither of the operands evaluates to 0, the result has a
value of 1. Otherwise it has a value of 0.
Unlike &, && guarantees left-to-right evaluation; moreover, the second operand is not
evaluated if the first operand evaluates to zero. There is a sequence point after the
evaluation of the first operand.

Logical OR Operator
Each of the operands of the logical OR operator must have scalar type. The ||
operator associates left to right. The syntax is as follows:
logical-OR-expression: logical-AND-expression
logical-OR-expression || logical-AND-expression
The result has type int. If either of the operands evaluates to one, the result has a
value of 1. Otherwise it has a value of 0.

007–0701–150 63
6: Expressions and Operators

Unlike |, || guarantees left to right evaluation; moreover, the second operand is not
evaluated unless the first operand evaluates to zero. A sequence point occurs after
the evaluation of the first operand.

Conditional Operator
Conditional expressions associate from right to left. The syntax is as follows:
conditional-expression: logical-OR-expression
logical-OR-expression ? expression : conditional-expression
The type of the first operand must be scalar. Only certain combinations of types are
allowed for the second and third operands. These combinations are listed below,
along with the type of result that the combination yields:
• Both can be arithmetic types. In this case, the usual arithmetic conversions are
performed on them to derive a common type, which is the type of the result.
• Both are compatible structure or union objects. The result has the same type as the
operands.

• Both are void. The type of the result is void.

• One is a pointer, and the other a null pointer constant. The type of the result is the
type of the nonconstant pointer.
• One is a pointer to void, and the other is a pointer to an object or incomplete
type. The second operand is converted to a pointer to void. The result is also a
pointer to void.
• Both are pointers to qualified or unqualified versions of compatible types. The
result has a type compatible with each, qualified with all the qualifiers of the types
pointed to by both operands.
Evaluation of the conditional operator proceeds as follows:
• The first expression is evaluated, after which a sequence point occurs.
• If the value of the first expression is nonzero, the result is the value of the second
operand.
• If the value of the first expression is zero, the result is the value of the third
operand.

64 007–0701–150
 C Language Reference Manual

• Only one of the second and third operands is evaluated.

Assignment Operators
All assignment operators associate from right to left. The syntax is as follows:
assignment-expression: conditional-expression
unary-expression assignment-operator assignment-expression
assignment operator: one = *= /= %= += -= <<= >>= &= ^= |=
of
Assignment operators require a modifiable lvalue as their left operand. The type of
an assignment expression is that of its unqualified left operand. The result is not an
lvalue. Its value is the value stored in the left operand after the assignment, but the
actual update of the stored value may be delayed until the next sequence point.
The order of evaluation of the operands is unspecified.

Assignment Using = (Simple Assignment)

The operands permissible in simple assignment must obey one of the following:
• Both have arithmetic type or are compatible structure or union types.
• Both are pointers, and the type pointed to by the left has all of the qualifiers of the
type pointed to by the right.
• One is a pointer to an object or incomplete type, and the other is a pointer to void.
The type pointed to by the left must have all of the qualifiers of the type pointed
to by the right.
• The left operand is a pointer, and the right is a null pointer constant.
In simple assignment, the value of the right operand is converted to the type of the
assignment expression and replaces the value of the object referred to by the left
operand. If the value being stored is accessed by another object that overlaps it, the
behavior is undefined unless the overlap is exact and the types of the two objects are
compatible.

007–0701–150 65
6: Expressions and Operators

Compound Assignment
For the operators += and -=, either both operators must have arithmetic types, or the
left operand must be a pointer and the right an operand integral. In the latter case,
the right operand is converted as explained in "Additive Operators", page 59. For all
other operators, each operand must have arithmetic type consistent with those
allowed for the corresponding binary operator.
The expression E1 op = E2is equivalent to the expression E1 = E1 op E2, except
that in the former, E1 is evaluated only once.

Comma Operator
A pair of expressions separated by a comma is evaluated left to right, and the value
of the left expression is discarded. This operator associates left to right. The syntax of
the comma operator is as follows:
expression: assignment-expression
expression, assignment-expression
The type and value of the result are the type and value of the right operand. In
contexts where the comma is given a special meaning, the comma operator as
described in this section can appear only in parentheses. Two such contexts are lists
of actual arguments to functions (described in "Primary Expressions", page 51) and
lists of initializers (see "Initialization", page 88). For example, the following code has
three arguments, the second of which has the value 5:
f(a, (t=3, t+2), c)

Constant Expressions
A constant expression can be used any place a constant can be used. The syntax is as
follows:
constant-expression: conditional-expression
A constant expression cannot contain assignment, increment, decrement, function-call,
or comma operators. It must evaluate to a constant that is in the range of
representable values for its type. Otherwise, the semantic rules for the evaluation of
nonconstant expressions apply.

66 007–0701–150
 C Language Reference Manual

Constant expressions are separated into three classes:

• An integral constant expression has integral type and is restricted to operands that
are integral constants, sizeof expressions (whose operands do not have variable
length array type or a parenthesized name of such a type), and floating constants
that are the immediate operands of integral casts.
• An arithmetic constant expression has arithmetic type and is restricted to operands
that are arithmetic constants, and sizeof expressions (whose operands do not have
variable length array type or a parenthesized name of such a type). Cast
expressions in arithmetic constant expressions can convert only between arithmetic
types.
• An address constant is a pointer to an lvalue designating an object of static storage
duration, or a pointer to a function designator. It can be created explicitly or
implicitly, as long as no attempt is made to access an object value.
Either address or arithmetic constant expressions can be used in initializers. In
addition, initializers can contain null pointer constants and address constants (for
object types), and plus or minus integral constant expressions.

Integer and Floating Point Exceptions

The following are a few points to keep in mind about integer and floating point
exceptions:
• Integer divide-by-zero results in a trap. Other integer exception conditions are
ignored.
• SGI floating point conforms to the IEEE standard. Floating point exceptions are
ignored by default, yielding the default IEEE results of infinity for divide-by-zero
and overflow, not-a-number for invalid operations, and zero for underflow.
• You can gain control over these exceptions and their results most easily by
<_newline> using the SGI IEEE floating point exception handler package (see the
handle_sigfpes(3c) reference page).
• You can also control these exceptions by implementing your own handler and
appropriately initializing the floating point unit (see the fpc(3c) reference page).

007–0701–150 67
 Chapter 7

Declarations

A declaration specifies the interpretation given to a set of identifiers. Declarations

have the following form:
declaration: declaration-specifiers init-declarator-listopt;
The init-declarator-list is a comma-separated sequence of declarators, each of which
can have an initializer.
In ANSI C, the init-declarator-list can also contain additional type information:
init-declarator-list: init-declarator
init-declarator-list , init-declarator
init-declarator: declarator
declarator = initializer
The declarators in the init-declarator list contain the identifiers being declared. The
declaration specifiers consist of a sequence of specifiers that determine the linkage,
storage duration, and part of the type of the identifiers indicated by the declarator.
Declaration specifiers have the following form:
declaration-specifiers: storage-class-specifier declaration-specifiersopt
type-specifier declaration-specifiersopt
type-qualifier declaration-specifiersopt
If an identifier that is not a tag has no linkage (see "Disambiguating Names", page
31), at most one declaration of the identifier can appear in the same scope and name
space. The type of an object that has no linkage must be complete by the end of its
declarator or initializer. Multiple declarations of tags and ordinary identifiers with
external or internal linkage can appear in the same scope so long as they specify
compatible types.
If a sequence of specifiers in a declarator contains a variable length array type, the
type specified by the declarator is said to be “variably modified.” All declarations of
variably modified types must be declared at either block or function prototype scope.
File scope identifiers cannot be declared with a variably modified type.
In traditional C, at most one declaration of an identifier with internal linkage can
appear in the same scope and name space, unless it is a tag.

007–0701–150 69
7: Declarations

In ANSI C, a declaration must declare at least one of the following:

• A declarator
• A tag
• The members of an enumeration
A declaration may reserve storage for the entities specified in the declarators. Such a
declaration is called a definition. (Function definitions have a different syntax and are
discussed in "Function Declarators and Prototypes", page 82, and Chapter 9, "External
Definitions", page 101.)

Storage Class Specifiers

The storage class specifier indicates linkage and storage duration. These attributes are
discussed in "Disambiguating Names", page 31. Storage class specifiers have the
following form:
storage-class-specifier: auto
static
extern
register
typedef
The typedef specifier does not reserve storage and is called a storage-class specifier
only for syntactic convenience. See "typedef", page 87, for more information.
The following rules apply to the use of storage class specifiers:
• A declaration can have at most one storage class specifier. If the storage class
specifier is missing from a declaration, it is assumed to be extern unless the
declaration is of an object and occurs inside a function, in which case it is assumed
to be auto. (See "Changes in Disambiguating Identifiers ", page 10.)
• Identifiers declared within a function with the storage class extern must have an
external definition (see Chapter 9, "External Definitions", page 101) somewhere
outside the function in which they are declared.
• Identifiers declared with the storage class static have static storage duration,
and either internal linkage (if declared outside a function) or no linkage (if

70 007–0701–150
 C Language Reference Manual

declared inside a function). If the identifiers are initialized, the initialization is

performed once before the beginning of execution. If no explicit initialization is
performed, static objects are implicitly initialized to zero.
• A register declaration is an auto declaration, with a hint to the compiler that the
objects declared will be heavily used. Whether the object is actually placed in fast
storage is implementation defined. You cannot take the address of any part of an
object declared with the register specifier.

Type Specifiers
Type specifiers are listed below. The syntax is as follows:
type-specifier: struct-or-union-specifier
typedef-name
enum-specifier
char
short
int
long
signed
unsigned
float
double
void
The following is a list of all valid combinations of type specifiers. These combinations
are organized into sets. The type specifiers in each set are equivalent in all
implementations. The arrangement of the type specifiers appearing in any set can be
altered without effect.
• void
• char

007–0701–150 71
7: Declarations

• signed char
• unsigned char
• short, signed short, short int, or signed short int
• unsigned short, or unsigned short int
• int, signed, signed int, or no type specifiers
• unsigned, or unsigned int
• long, signed long, long int, or signed long int
• unsigned long, or unsigned long int
• long long, signed long long, long long int, or signed long long
int
• unsigned long long, or unsigned long long int
• float
• double
• long double
In traditional C, the type long float is allowed and is equivalent to double; its
use is not recommended. It elicits a warning if you are not in -cckr mode. Use of
the type long double is not recommended in traditional C.
long long is not a valid ANSI C type, so a warning appears for every occurrence of
it in the source program text in -ansi and -ansiposix modes.
Specifiers for structures, unions, and enumerations are discussed in "Structure and
Union Declarations", page 72, and "Enumeration Declarations", page 76. Declarations
with typedef names are discussed in "typedef", page 87.

Structure and Union Declarations

A structure is an object consisting of a sequence of named members. Each member
can have any type. A union is an object that can, at a given time, contain any one of
several members. Structure and union specifiers have the same form. The syntax is as
follows:

72 007–0701–150
 C Language Reference Manual

struct-or-union-specifier: struct-or-union {struct-decl-list}

struct-or-union identifier {struct-decl-list}
struct-or-union identifier
struct-or-union: struct
union
The struct-decl-list is a sequence of declarations for the members of the structure or
union. The syntax, in three possible forms, is as follows:
struct-decl-list: struct-declaration
struct-decl-list struct-declaration
struct-declaration: specifier-qualifier-list struct-declarator-list;
struct-declarator-list: struct-declarator
struct-declarator-list , struct-declarator
In the usual case, a struct-declarator is just a declarator for a member of a structure or
union. A structure member can also consist of a specified number of bits. Such a
member is also called a bitfield. Its length, a non-negative constant expression, is
separated from the field name by a colon. "Bitfields", page 75, are discussed at the
end of this section.
The syntax for struct-declarator is as follows:
struct-declarator: declarator
declarator : constant-expression
: constant-expression
A struct or union cannot contain any of the following:
• A member with incomplete or function type.
• A member that is an instance of itself. It can, however, contain a member that is a
pointer to an instance of itself.
• A member that has a variable length array type.
• A member that is a pointer to a variable length array type.

007–0701–150 73
7: Declarations

Within a structure, the objects declared have addresses that increase as the
declarations are read left to right. Each non-field member of a structure begins on an
addressing boundary appropriate to its type; therefore, there may be unnamed holes
in a structure.
A union can be thought of as a structure whose members all begin at offset 0 and
whose size is sufficient to contain any of its members. At most, one of the members
can be stored in a union at any time.
A structure or union specifier of the second form declares the identifier to be the
structure tag (or union tag) of the structure specified by the list. This type of specifier
is one of the following:
struct identifier {struct-decl-list}
union identifier {struct-decl-list}

A subsequent declaration can use the third form of specifier, one of the following:
struct identifier
union identifier

Structure tags allow definition of self-referential structures. Structure tags also permit
the long part of the declaration to be given once and used several times.
The third form of a structure or union specifier can be used before a declaration that
gives the complete specification of the structure or union in situations in which the
size of the structure or union is unnecessary. The size is unnecessary in two situations:
when a pointer to a structure or union is being declared and when a typedef name
is declared to be a synonym for a structure or union. This, for example, allows the
declaration of a pair of structures that contain pointers to each other.
The names of members of each struct or union have their own name space, and do
not conflict with each other or with ordinary variables. A particular member name
cannot be used twice in the same structure, but it can be used in several different
structures in the same scope.
Names that are used for tags reside in a single name space. They do not conflict with
other names or with names used for tags in an enclosing scope. This tag name space,
however, consists of tag names used for all struct, union, or enum declarations.
Therefore, the tag name of an enum may conflict with the tag name of a struct in
the same scope. (See "Disambiguating Names", page 31.)
A simple but important example of a structure declaration is the following binary tree
structure:

74 007–0701–150
 C Language Reference Manual

struct tnode {
char tword[20];
int count;
struct tnode *left;
struct tnode *right;
};
struct tnode s, *sp;

This structure contains an array of 20 characters, an integer, and two pointers to

instances of itself. Once this structure has been declared, the next line declares a
structure of type struct tnode (s) and a pointer to a structure of type struct
tnode (sp).
With these declarations,
• The expression sp->count refers to the count field of the structure to which sp
points.
• The expression s.left refers to the left subtree pointer of the structure s.
• The expression s.right->tword[0] refers to the first character of the tword
member of the right subtree of s.

Bitfields
A structure member can consist of a specified number of bits, called a bitfield. In
strict ANSI C mode, bitfields should be of type int, signed int, or unsigned
int. SGI C allows bitfields of any integral type, but warns for non-int types in
-ansi and -ansiposix modes.
The default type of field members is int, but whether it is signed or unsigned
int is defined by the implementation. Therefore, you should specify the signedness
of bitfields when they are declared. In this implementation, the default type of a
bitfield is signed.
The constant expression that denotes the width of the bitfield must have a value no
greater than the width, in bits, of the type of the bitfield. An implementation can
allocate any addressable storage unit (referred to in this discussion as simply a
“unit”) large enough to hold a bitfield. If an adjacent bitfield will not fit into the
remainder of the unit, the implementation defines whether bitfields are allowed to
span units or whether another unit is allocated for the second bitfield. The ordering
of the bits within a unit is also implementation-defined.

007–0701–150 75
7: Declarations

A bitfield with no declarator, just a colon and a width, indicates an unnamed field
useful for padding. As a special case, a field with a width of zero (which cannot have
a declarator) specifies alignment of the next field at the next unit boundary.
These implementation-defined characteristics make the use of bitfields inherently
nonportable, particularly if they are used in situations where the underlying object
may be accessed by another data type (in a union, for example).
In the SGI implementation of C, the first bitfield encountered in a struct is not
necessarily allocated on a unit boundary and is packed into the current unit, if
possible. A bitfield cannot span a unit boundary. Bits for bitfields are allocated from
left (most significant) to right.
There are no arrays of bitfields. Because the address-of operator, &, cannot be applied
to bitfields, there are also no pointers to bitfields.

Enumeration Declarations
Enumeration variables and constants have integral type. The syntax is as follows:
enum-specifier: enum {enum-list}
enum {identifier enum-list}
enum identifier
enum-list: enumerator
enum-list , enumerator
enumerator: identifier
identifier = constant-expression
The identifiers in an enum-list are declared as int constants and can appear wherever
such constants are allowed. If no enumerators with = appear, then the values of the
corresponding constants begin at 0 and increase by 1 as the declaration is read from
left to right. An enumerator with = gives the associated identifier the value indicated;
subsequent identifiers continue the progression from the assigned value. Note that
the use of = may result in multiple enumeration constants having the same integral
value, even though they are declared in the same enumeration declaration.

76 007–0701–150
 C Language Reference Manual

Enumerators are in the ordinary identifiers name space (see "Name Spaces", page 33).
Thus, an identifier used as an enumerator may conflict with identifiers used for
objects, functions, and user-defined types in the same scope.
The role of the identifier in the enum-specifier is entirely analogous to that of the
structure tag in a struct-specifier; it names a particular enumeration. For example,
enum color { chartreuse, burgundy, claret = 20, winedark };
...
enum color *cp, col;
...
col = claret;
cp = &col;
...
if (*cp == burgundy) ...

This example makes color the enumeration-tag of a type describing various colors,
and then declares cp as a pointer to an object of that type, col. The possible values are
drawn from the set {0,1,20,21}. The tags of enumeration declarations are members of
the single tag name space, and thus must be distinct from tags of struct and union
declarations.

Type Qualifiers
Type qualifiers have the following syntax:
type-qualifier: const
volatile
__restrict
The same type qualifier cannot appear more than once in the same specifier list either
directly or indirectly (through typedefs).
The value of an object declared with the const type qualifier is constant. It cannot be
modified, although it can be initialized following the same rules as the initialization of
any other object. (See the discussion in "Initialization", page 88.) Implementations are
free to allocate const objects, that are not also declared volatile, in read-only storage.
An object declared with the volatile type qualifier may be accessed in unknown ways
or have unknown side effects. For example, a volatile object may be a special
hardware register. Expressions referring to objects qualified as volatile must be

007–0701–150 77
7: Declarations

evaluated strictly according to the semantics. When volatile objects are involved, an
implementation is not free to perform optimizations that would otherwise be valid.
At each sequence point, the value of all volatile objects must agree with that specified
by the semantics.
The __restrict qualifier applies only to pointers and is discussed in "Qualifiers
and Pointers", page 79.
If an array is specified with type qualifiers, the qualifiers are applied to the elements of
the array. If a struct or union is qualified, the qualification applies to each member.
Two qualified types are compatible if they are identically qualified versions of
compatible types. The order of qualifiers in a list has no effect on their semantics.
The syntax of pointers allows the specification of qualifiers that affect either the
pointer itself or the underlying object. Qualified pointers are covered in "Pointer
Declarators", page 79.

Declarators
Declarators have the syntax shown below:
declarator: pointeropt direct-declarator
direct-declarator: identifier
(declarator)
direct-declarator (parameter-type-listopt)
direct-declarator (identifier-listopt)
direct-declarator [constant-expressionopt]
The grouping is the same as in expressions.

Meaning of Declarators
Each declarator is an assertion that when a construction of the same form as the
declarator appears in an expression, it designates a function or object with the scope,
storage duration, and type indicated by the declaration.
Each declarator contains exactly one identifier; it is this identifier that is declared. If,
in the declaration “T D1;” D1 is simply an identifier, then the type of the identifier is

78 007–0701–150
 C Language Reference Manual

T. A declarator in parentheses is identical to the unparenthesized declarator. The

binding of complex declarators can, however, be altered by parentheses.

Pointer Declarators
Pointer declarators have the form:
pointer: * type-qualifier-listopt
* type-qualifier-listopt pointer
The following is an example of a declaration:
T D1

In this declaration, the identifier has type .. T, where the .. is empty if D1 is just
a plain identifier (so that the type of x in int x is just int). Then, if D1 has the form
*type-qualifier-listopt D, the type of the contained identifier is ”..
(possibly-qualified) pointer to T.”

Qualifiers and Pointers

It is important to be aware of the distinction between a qualified pointer to a type

and a pointer to a qualified type. In the declarations below, ptr_to_const is a
pointer to const long:
const long *ptr_to_const;
long * const const_ptr;
volatile int * const const_ptr_to_volatile;

The long pointed to by ptr_to_const in the first declaration, cannot be modified

by the pointer. The pointer itself, however, can be altered. In the second declaration,
const_ptr can be used to modify the long that it points to, but the pointer itself
cannot be modified. In the last declaration, const_ptr_to_volatile is a constant
pointer to a volatile int and can be used to modify it. The pointer itself, however,
cannot be modified.
The __restrict qualifier tells the compiler to assume that dereferencing the
qualified pointer is the only way the program can access the memory pointed to by
that pointer. Therefore, loads and stores through such a pointer are assumed not to
alias with any other loads and stores in the program, except other loads and stores
through the same pointer variable.

007–0701–150 79
7: Declarations

The following example illustrates the use of the __restrict qualifier:

float x[ARRAY_SIZE];
float *c = x;

void f4_opt(int n, float * __restrict a, float * __restrict b)

{
int i;
/* No data dependence across iterations because of __restrict */
for (i = 0; i < n; i++)
a[ i ] = b [ i ] + c [ i ] ;
}

Pointer-related Command Options

The SGI C compiler supports the following two alias-related command-line switches
that can be useful for improving performance:
-OPT:alias=restrict
Implements the following semantics: memory operations
dereferencing different named pointers in the program are assumed
not to alias with each other, nor with any named scalar in the
program.
For example, if p and q are pointers, this option means that *p does
not alias with *q, with p, or with any named scalar variable.
-OPT:alias=disjoint

Implements the following semantics: memory operations

dereferencing different named pointers in the program are assumed
not to alias with each other, and in addition, different dereferencing
depths of the same named pointer are assumed not to alias with each
other.
For example, if p and q are of type pointer to pointer, *p does not
alias with *q, with **p, or with **q.

Note: With either switch enabled, programs violating the corresponding aliasing
assumptions may be compiled incorrectly.

80 007–0701–150
 C Language Reference Manual

Array Declarators
If in the declaration T D1, D1 has the form D[expressionopt] or D[*], then the
contained identifier has type “array of T.” Starting with version 7.2, the SGI C
compiler now supports variable length arrays as well as fixed length arrays. A
variable length array is an array that has a size (at least one dimension) that is
determined at run time. The ability to use variable length arrays enhances the
compiler’s range of use for numerical programming.
The following rules apply to array declarations:
• If the array is a fixed length array, the expression enclosed in square brackets, if it
exists, must be an integral constant expression whose value is greater than zero.
(See "Primary Expressions", page 51.)
• When several “array of” specifications are adjacent, a multi-dimensional array is
created; the constant expressions that specify the bounds of the arrays can be
missing only for the first member of the sequence.
• The absence of the first array dimension is allowed if the array is external and the
actual definition (which allocates storage) is given elsewhere, or if the declarator is
followed by initialization. In the latter case, the size is calculated from the number
of elements supplied.

• If * is used instead of a size expression, the array is of “variable length array”

type with unspecified size. This can only be used in declarations with function
prototype scope.
• The array type is “fixed length array” if the size expression is an integer constant
expression, and the element type has a fixed size. Otherwise the type is variable
length array.
• The size of a variable length array type does not change until the execution of the
block containing the declaration has finished.
• Array objects declared with either static or extern storage class specifiers cannot be
declared with a variable length array type. However, block scope pointers
declared with the static storage class specifier can be declared as pointers to
variable length array types.
• In order for two array types to be compatible, their element types must be
compatible. In addition, if both of their size specifications are present and are
integer constant expressions, they must have the same value. If either size
specifier is variable, the two sizes must evaluate to the same value at run time.

007–0701–150 81
7: Declarations

• An array can be constructed from one of the basic types, from a pointer, from a
structure or union, or from another array (to generate a multi-dimensional array).
The example below declares an array of float numbers and an array of pointers to
float numbers:
float fa[17], *afp[17];

The following example declares a static three-dimensional array of integers, with rank
3 2 5 2 7.
static int x3d[3][5][7];

In the above example, x3d is an array of three items; each item is an array of five
items, each of which is an array of seven integers. Any of the expressions x3d,
x3d[i], x3d[i][j], x3d[i][j][k] can reasonably appear in an expression. The
first three have type array and the last has type int.

Function Declarators and Prototypes

The syntax for function declarators is shown below:

direct-declarator (parameter-type-listopt)
direct-declarator (identifier-listopt)
parameter-type-list:
parameter-list
parameter-list , ...
parameter-list:
parameter-declaration
parameter-list , parameter-declaration
parameter-declaration:
declaration-specifiers declarator
declaration-specifiers abstract-declaratoropt
identifier-list:
identifier
identifier-list , identifier

Function declarators cannot specify a function or array type as the return type. In
addition, the only storage class specifier that can be used in a parameter declaration is
register. For example, in the declaration T D1, D1 has one of the following forms:

82 007–0701–150
 C Language Reference Manual

• D(parameter-type-listopt)
• D(identifier-listopt)
The contained identifier has the type ”.. function returning T,” and is possibly a
prototype, as discussed later in this section.
A parameter type list declares the types of, and can declare identifiers for, the formal
parameters of a function. A declared parameter that is a member of the parameter
type list that is not part of a function definition may use the [*] notation in its
sequence of declarator specifiers to specify a variable length array type.
The absence of a parameter type list indicates that no typing information is given for
the function. A parameter type list consisting only of the keyword void indicates that
the function takes zero parameters. If the parameter type list ends in ellipses (…), the
function can have one or more additional arguments of variable or unknown type.
(See <stdarg.h>.)
The semantics of a function declarator are determined by its form and context. The
possible combinations are as follows:
• The declarator is not part of the function definition. The function is defined
elsewhere. In this case, the declarator cannot have an identifier list.
– If the parameter type list is absent, the declarator is an old-style function
declaration. Only the return type is significant.
– If the parameter type list is present, the declarator is a function prototype.
• The declarator is part of the function definition:
– If the declarator has an identifier list, the declarator is an old-style function
definition. Only the return type is significant.
– If the declarator has a parameter type list, the definition is in prototype form.
If no previous declaration for this function has been encountered, a function
prototype is created for it that has file scope.
If two declarations (one of which can be a definition) of the same function in the
same scope are encountered, they must match, both in type of return value and in
parameter type list. If one and only one of the declarations has a parameter type list,
the behavior varies between ANSI C and Traditional C.
In traditional C, most combinations pass without any diagnostic messages. However,
an error message is emitted for cases where an incompatibility is likely to lead to a

007–0701–150 83
7: Declarations

run-time failure. For example, a float type in a parameter type list of a function
prototype is incompatible with any old-style declaration for the same function;
therefore, SGI considers such redeclarations erroneous.
In ANSI C, if the type of any parameter declared in the parameter type list is other
than that which would be derived using the default argument promotions, an error is
posted. Otherwise, a warning is posted and the function prototype remains in scope.
In all cases, the type of the return value of duplicate declarations of the same function
must match, as must the use of ellipses.
When a function is invoked for which a function prototype is in scope, an attempt is
made to convert each actual parameter to the type of the corresponding formal
parameter specified in the function prototype, superseding the default argument
promotions. In particular, floats specified in the type list are not converted to
double before the call. If the list terminates with an ellipsis (...), only the parameters
specified in the prototype have their types checked; additional parameters are
converted according to the default argument promotions (discussed in "Type
Qualifiers", page 77). Otherwise, the number of parameters appearing in the
parameter list at the point of call must agree in number with those in the function
prototype.
The following are two examples of function prototypes:
double foo(int *first, float second, ... );
int *fip(int a, long l, int (*ff)(float));

The first prototype declares a function foo() which returns a double and has (at
least) two parameters: a pointer to an int and a float. Further parameters can
appear in a call of the function and are unspecified. The default argument promotions
are applied to any unspecified arguments. The second prototype declares a function
fip(), which returns a pointer to an int. The function fip() has three parameters:
an int, a long, and a pointer to a function returning an int that has a single
(float) argument.

Prototyped Functions Summarized

When a function call occurs, each argument is converted using the default argument
promotions unless that argument has a type specified in a corresponding in-scope
prototype for the function being called. It is easy to envision situations that could
prove disastrous if some calls to a function are made with a prototype in-scope and
some are not. Unexpected results can also occur if a function is called with different

84 007–0701–150
 C Language Reference Manual

prototypes in scope. Therefore, if a function is prototyped, it is extremely important

to make sure that all invocations of the function use the prototype.
In addition to adding a new syntax for external declarations of functions, prototypes
have added a new syntax for external definitions of functions. This syntax is termed
“function prototype form.” It is highly important to define prototyped functions
using a parameter type list rather than a simple identifier list if the parameters are to
be received as intended.
In ANSI C, unless the function definition has a parameter type list, it is assumed that
arguments have been promoted according to the default argument promotions.
Specifically, an in-scope prototype for the function at the point of its definition has no
effect on the type of the arguments that the function expects.
The compilers issue error diagnostics when argument-type mismatches are likely to
result in faulty run-time behavior.

Restrictions on Declarators
Not all the possibilities allowed by the syntax of declarators are permitted. The
following restrictions apply:
• Functions cannot return arrays or functions although they can return pointers.
• No arrays of functions exist although arrays of pointers to functions can exist.
• A structure or union cannot contain a function, but it can contain a pointer to a
function.
As an example, the following declaration declares an integer i; a pointer to an
integer, ip; a function returning an integer, f(); a function returning a pointer to an
integer, fip(); and a pointer to a function that returns an integer, pfi:

int i, *ip, f(), *fip(), (*pfi)();

It is especially useful to compare the last two. The binding of *fip() is *(fip()).
The declaration suggests, and the same construction in an expression requires, the
calling of a function fip(), and then using indirection through the (pointer) result to
yield an integer. In the declarator *pfi)(), the extra parentheses are necessary,
because they are also in an expression, to indicate that indirection through a pointer
to a function yields a function, which is then called and returns an integer.

007–0701–150 85
7: Declarations

Type Names
In several contexts (for example, to specify type conversions explicitly by means of a
cast, in a function prototype, and as an argument of sizeof), it is best to supply the
name of a data type. This naming is accomplished using a “type name,” whose
syntax is a declaration for an object of that type without the identifier.
The syntax for type names is as follows:
type-name: specifier-qualifier-list abstract-declaratoropt
abstract-declarator: pointer
pointeropt direct-abstract-declarator
direct-abstract-declarator: (abstract-declarator)
direct-abstract-declaratoropt [constant-expressionopt]
direct-abstract-declaratoropt (parameter-type-listopt)
The type name created can be used as a synonym for the type of the omitted
identifier. The syntax indicates that a set of empty parentheses in a type name is
interpreted as function with no parameter information rather than as redundant
parentheses surrounding the (omitted) identifier.
Examples of type names are shown in Table 7-1, page 86.

Table 7-1 Examples of Type Names

Type Description
int Integer
int * Pointer to integer
int *[3] Array of three pointers to integers
int (*)[3] Pointer to an array of three integers
int *(void) Function with zero arguments returning pointer to integer

86 007–0701–150
 C Language Reference Manual

Type Description
int Pointer to function returning an integer, that has a variable
(*)(float, number of arguments the first of which is a float
...)
int (*[3])() Array of three pointers to functions returning an integer for
which no parameter type information is given

Implicit Declarations
It is not always necessary to specify both the storage class and the type of identifiers
in a declaration. The storage class is supplied by the context in external definitions,
and in declarations of formal parameters and structure members. Missing storage
class specifiers appearing in declarations outside of functions are assumed to be
extern (see "External Name Changes", page 20, for details. Missing type specifiers in
this context are assumed to be int. In a declaration inside a function, if a type but no
storage class is indicated, the identifier is assumed to be auto. An exception to the
latter rule is made for functions because auto functions do not exist. If the type of an
identifier is “function returning <type>”, it is implicitly declared to be extern.
In an expression, an identifier followed by a left parenthesis (indicating a function
call) that is not already declared is implicitly declared to be of type function
returning int.

typedef
Declarations with the storage class specifier typedef do not define storage. A
typedef has the following syntax:
typedef-name: identifier
An identifier appearing in a typedef declaration becomes a synonym for the type
rather than becoming an object with the given type. For example, if the int type
specifier in the following example were preceded with typedef, the identifier
declared as an object would instead be declared as a synonym for the array type:
int intarray[10]];

007–0701–150 87
7: Declarations

This can appear as shown below:

typedef int intarray[10];

intarray could then be used as if it were a basic type, as in the following:

intarray ia;

In the following example, the last three declarations are legal. The type of distance
is int, that of metricp is pointer to int, and that of z is the specified structure. The
zp is a pointer to such a structure:
typedef int MILES, *KLICKSP;
typedef struct {
double re, im;
}
complex;

MILES distance;
extern KLICKSP metricp;
complex z, *zp;

The typedef does not introduce brand-new types, only synonyms for types that
could be specified in another way. For instance, in the preceding example, distance
is considered to have the same type as any other int object.
typedef declarations that specify a variably modified type have block scope. The
array size specified by the variable length array type is evaluated at the time the type
definition is declared and not at the time it is used as a type specifier in an actual
declarator.

Initialization
A declaration of an object or of an array of unknown size can specify an initial value
for the identifier being declared. The initializer is preceded by = and consists of an
expression or a list of values enclosed in nested braces:
initializer: assignment-expression

{initializer-list}

88 007–0701–150
 C Language Reference Manual

initializer-list: initializer
initializer-list , initializer
There cannot be more initializers than there are objects to be initialized. All the
expressions in an initializer for an object of static storage duration must be constant
expressions (see "Primary Expressions", page 51). Objects with automatic storage
duration can be initialized by arbitrary expressions involving constants and
previously declared variables and functions, except for aggregate initialization, which
can include only constant expressions.
Identifiers declared with block scope and either external or internal linkage (that is,
objects declared in a function with the storage class specifier extern) cannot be
initialized.
Variables of static storage duration that are not explicitly initialized are implicitly
initialized to zero. The value of automatic and register variables that are not explicitly
initialized is undefined.
When an initializer applies to a scalar (a pointer or an object of arithmetic type; see
"Derived Types", page 40), it consists of a single expression, perhaps in braces. The
initial value of the object is taken from the expression. With the exception of type
qualifiers associated with the scalar, which are ignored during the initialization, the
same conversions as for assignment are performed.

Initialization of Aggregates
In traditional C, it is illegal to initialize a union. It is also illegal to initialize a struct
of automatic storage duration.
In ANSI C, objects that are struct or union types can be initialized, even if they
have automatic storage duration. unions are initialized using the type of the first
named element in their declaration. The initializers used for a struct or union of
automatic storage duration must be constant expressions if they are in an initializer
list. If the struct or union is initialized using an assignment expression, the
expression need not be constant.
When the declared variable is a struct or array, the initializer consists of a
brace-enclosed, comma-separated list of initializers for the members of the aggregate
written in increasing subscript or member order. If the aggregate contains
subaggregates, this rule applies recursively to the members of the aggregate.

007–0701–150 89
7: Declarations

If the initializer of a subaggregate or union begins with a left brace, its initializers
consist of all the initializers found between the left brace and the matching right
brace. If, however, the initializer does not begin with a left brace, then only enough
elements from the list are taken to account for the members of the subaggregate; any
remaining members are left to initialize the next member of the aggregate of which
the current subaggregate is a part.
Within any brace-enclosed list, there should not be more initializers than members. If
there are fewer initializers in the list than there are members of the aggregate, then
the aggregate is padded with zeros.
Unnamed struct or union members are ignored during initialization.
In ANSI C, if the variable is a union, the initializer consists of a brace-enclosed
initializer for the first member of the union. Initialization of struct or union objects
with automatic storage duration can be abbreviated as a simple assignment of a
compatible struct or union object.
A final abbreviation allows a char array to be initialized by a string literal. In this
case, successive characters of the string literal initialize the members of the array.
In ANSI C, an array of wide characters (that is, whose element type is compatible with
wchar_t) can be initialized with a wide string literal (see "String Literals", page 28).

Examples of Initialization
The following example declares and initializes x as a one-dimensional array that has
three members, because no size was specified and there are three initializers:
int x[] = { 1, 3, 5 };

The next example shows a completely bracketed initialization: 1, 3, and 5 initialize

the first row of the array y[0], namely y[0][0], y[0][1], and y[0][2]. Likewise,
the next two lines initialize y[1] and y[2]. The initializer ends early, and therefore,
y[3] is initialized with 0:
float y[4][3] =
{
{ 1, 3, 5 },
{ 2, 4, 6 },
{ 3, 5, 7 },
};

90 007–0701–150
 C Language Reference Manual

The next example achieves precisely the same effect. The initializer for y begins with
a left brace but that for y[0] does not; therefore, three elements from the list are
used. Likewise, the next three are taken successively for y[1] and y[2]:
float y[4][3] =
{
1, 3, 5, 2, 4, 6, 3, 5, 7
};

The next example initializes the first column of y (regarded as a two-dimensional

array) and leaves the rest 0:
float y[4][3] = {
{ 1 }, { 2 }, { 3 }, { 4 }
};

The following example demonstrates the ANSI C rules. A union object, dc_u, is
initialized by using the first element only:
union dc_u {
double d;
char *cptr;
};

union dc_u dc0 = { 4.0 };

The final example shows a character array whose members are initialized with a
string literal. The length of the string (or size of the array) includes the terminating
NULL character, \0:
char msg[] = "Syntax error on line %s\n";

007–0701–150 91
 Chapter 8

Statements

A statement is a complete instruction to the computer. Except as indicated, statements

are executed in sequence. Statements have the following form:
statement: expression-statement
compound-statement
selection-statement
iteration-statement
jump-statement
labeled-statement

Expression Statement
Most statements are expression statements, which have the following form:
expression-statement: expressionopt;
Usually expression statements are expressions evaluated for their side effects, such as
assignments or function calls. A special case is the null statement, which consists of
only a semicolon.

Compound Statement or Block

A compound statement (or block) groups a set of statements into a syntactic unit. The
set can have its own declarations and initializers, and has the following form:
compound-statement: {declaration-listopt statement-listopt}
declaration-list: declaration
declaration-list declaration

007–0701–150 93
8: Statements

statement-list: statement
statement-list statement
Declarations within compound statements have block scope. If any of the identifiers
in the declaration list were previously declared, the outer declaration is hidden for the
duration of the block, after which it resumes its force. In traditional C, however,
function declarations always have file scope whenever they appear.
Initialization of identifiers declared within the block is restricted to those that have no
linkage. Thus, the initialization of an identifier declared within the block using the
extern specifier is not allowed. These initializations are performed only once, prior to
the first entry into the block, for identifiers with static storage duration. For
identifiers with automatic storage duration, it is performed each time the block is
entered at the top. It is currently possible (but a bad practice) to transfer into a block;
in that case, no initializations are performed.

Selection Statements
Selection statements include if and switch statements and have the following form:
selection-statement: if (expression) statement
if (expression) statement else statement
switch (expression) statement
Selection statements choose one of a set of statements to execute, based on the
evaluation of the expression. The expression is referred to as the controlling
expression.

if Statement
The controlling expression of an if statement must have scalar type.
For both forms of the if statement, the first statement is executed if the controlling
expression evaluates to nonzero. For the second form, the second statement is
executed if the controlling expression evaluates to zero. An else clause that follows
multiple sequential else-less if statements is associated with the most recent if
statement in the same block (that is, not in an enclosed block).

94 007–0701–150
 C Language Reference Manual

switch Statement
The controlling expression of a switch statement must have integral type. The
statement is typically a compound statement, some of whose constituent statements
are labeled case statements (see "Labeled Statements", page 98, and "continue
Statement", page 97, respectively).
The following is a simple example of a complete switch statement:
switch (c) {
case ’o’:
oflag = TRUE;
break;
case ’p’:
pflag = TRUE;
break;
case ’r’:
rflag = TRUE;
break;
default :
(void) fprintf(stderr,
"Unknown option\n");
exit(2);
}

Iteration Statements
Iteration statements execute the attached statement (called the body) repeatedly until
the controlling expression evaluates to zero. In the for statement, the second
expression is the controlling expression. The format is as follows:
iteration-statement: while (expression) statement
do statement while (expression) ;
for (expressionopt ; expressionopt ; expressionopt) statement
The controlling expression must have scalar type.
The flow of control in an iteration statement can be altered by a jump statement (see
"Jump Statements", page 97).

007–0701–150 95
8: Statements

while Statement
The controlling expression of a while statement is evaluated before each execution of
the body.

do Statement
The controlling expression of a do statement is evaluated after each execution of the
body.

for Statement
The for statement has the following form:
for (expressionopt ; statement
expressionopt ;
expressionopt)
The first expression specifies initialization for the loop. The second expression is the
controlling expression, which is evaluated before each iteration. The third expression
often specifies incrementation. It is evaluated after each iteration.
This statement is equivalent to the following:
expression-1; while (expression-2)
{
statement
expression-3;
}
One exception exists, however. If a continue statement (see "continue Statement",
page 97 is encountered, expression-3 of the for statement is executed prior to the next
iteration.
Any or all of the expressions can be omitted. A missing expression-2 makes the
implied while clause equivalent to while. Other missing expressions are simply
dropped from the previous expansion.

96 007–0701–150
 C Language Reference Manual

Jump Statements
Jump statements cause unconditional transfer of control. The syntax is as follows:
jump-statement: goto identifier;
continue;
break;
return expressionopt;

goto Statement
Control can be transferred unconditionally by means of a goto statement:

goto identifier;

The identifier must name a label located in the enclosing function. If the label has not
yet appeared, it is implicitly declared. (See "Labeled Statements", page 98, for more
information.)

continue Statement
The continue statement can appear only in the body of an iteration statement. It
causes control to pass to the loop-continuation portion of the smallest enclosing
while, do, or for statement; that is, to the end of the loop. Consider each of the
following statements:
while (...)
{
..
contin: ;
}

do {
...
contin: ;
} while (...) ;

for (...) {

007–0701–150 97
8: Statements

...
contin: ;
}

A continue is equivalent to goto contin. Following the contin: is a null

statement.
A goto statement must not cause a block to be entered by a jump from outside the
block to a labeled statement in the block (or an enclosed block) if that block contains
the declaration of a variably modified object or variably modified typedef name.

break Statement
The break statement can appear only in the body of an iteration statement or code
attached to a switch statement. It transfers control to the statement immediately
following the smallest enclosing iteration or switch statement, terminating its
execution.

return Statement
A function returns to its caller by means of the return statement. The value of the
expression is returned to the caller (after conversion to the declared type of the
function), as the value of the function call expression. The return statement cannot
have an expression if the type of the current function is void.
If the end of a function is reached before the execution of an explicit return, an
implicit return (with no expression) is executed. If the value of the function call
expression is used when none is returned, the behavior is undefined.

Labeled Statements
Labeled statements have the following syntax:
labeled-statement: identifier : statement
case constant-expression : statement

default : statement
A case or default label can appear only on statements that are part of a switch.

98 007–0701–150
 C Language Reference Manual

Any statement can have a label attached as a simple identifier. The scope of such a
label is the current function. Thus, labels must be unique within a function. In
traditional C, identifiers used as labels and in object declarations share a name space.
Thus, use of an identifier as a label hides any declaration of that identifier in an
enclosing scope. In ANSI C, identifiers used as labels are placed in a different name
space from all other identifiers and do not conflict. Therefore, the following code
fragment is legal in ANSI C but not in traditional C:

{
int foo;
foo = 1;
…
goto foo;
…
foo: ;
}

007–0701–150 99
 Chapter 9

External Definitions

A C program consists of a sequence of external definitions. An external declaration

becomes an external definition when it reserves storage for the object or function
indicated. Within the entire program, all external declarations of the same identifier
with external linkage refer to the same object or function. Within a particular
translation unit, all external declarations of the same identifier with internal linkage
refer to the same object or function. External declarations have the following syntax:
external declaration: function-definition
declaration
The syntax for external definitions that are not functions is the same as the syntax for
the corresponding external declarations. The syntax for the corresponding external
function definition differs from that of the declaration, because the definition includes
the code for the function itself.

External Function Definitions

Function definitions have the following form:
function-definition: declaration-specifiersopt declarator declaration-listopt
compound statement
The form of a declarator used for a function definition can be as follows:

pointeropt direct-declarator ( parameter-type-listopt)

pointeropt direct-declarator ( identifier-listopt)

In this syntax, the simplest instance of a direct-declarator is an identifier. (For the

exact syntax, see "Declarators", page 78.)
The only storage-class specifiers allowed in a function definition are extern and
static.
If the function declarator has a parameter type list (see "Declarators", page 78), it is in
function prototype form (as discussed in "Function Declarators and Prototypes", page
82), and the function definition cannot have a declaration list. Otherwise, the function

007–0701–150 101
9: External Definitions

declarator has a possibly empty identifier list, and the declaration list declares the
types of the formal parameters. register is the only storage class specifier
permitted in declarations that are in the declaration list. Any identifiers in the
identifier list of the function declarator that do not have their types specified in the
declaration list are assumed to have type int.
Each parameter has block scope and automatic storage duration. ANSI C and
traditional C place parameters in different blocks. See "Scope", page 31, for details.
Each parameter is also an lvalue, but because function calls in C are by value, the
modification of a parameter of arithmetic type cannot affect the corresponding
argument. Pointer parameters, while unmodifiable for this reason, can be used to
modify the objects to which they point.
Argument promotion rules are discussed in "Function Calls", page 52.
The type of a function must be either void or an object type that is not an array.

External Object Definitions

A declaration of an object with file scope that has either an initializer or static linkage
is an external object definition.
In ANSI C, a file-scope object declaration with external linkage that is declared
without the storage-class specifier extern, and also without an initializer, results in a
definition of the object at the end of the translation unit. See the discussion in
"Preprocessor Changes", page 7, for more information.

102 007–0701–150
 Chapter 10

Multiprocessing Directives

In addition to the usual interpretation performed by a C/C++ compiler, the

multiprocessing C/C++ compiler can process explicit multiprocessing directives to
produce code that can run concurrently on multiple processors.
Table 10-1lists the multiprocessing #pragma directives to use when processing code
in parallel regions. The multiprocessing compiler does not know whether an
automatic parallelization tool, you the user, or a combination of the two put the
directives in the code. The multiprocessing C/C++ compiler does not check for or
warn against data dependencies or other restrictions that have been violated.
See the MIPSpro C and C++ Pragmas for more details.

Table 10-1 Multiprocessing C/C++ Compiler Directives

#pragma Description
#pragma copyin Copies the value from the master thread’s version of an -Xlocal-linked global
variable into the slave thread’s version.
#pragma critical Protects access to critical statements.
#pragma enter gate Indicates the point that all threads must clear before any threads are allowed to
pass the corresponding exit gate.
#pragma exit gate Stops threads from passing this point until all threads have cleared the
corresponding enter gate.
#pragma independent Starts an independent code section that executes in parallel with other code in
the parallel region.
#pragma local Tells the compiler the names of all the variables that must be local to each thread.
#pragma no side Tells the compiler to assume that all of the named functions are safe to execute
effects concurrently.
#pragma one processor Causes the next statement to be executed on only one processor.
#pragma parallel Marks the start of a parallel region.
#pragma pfor Marks a for loop to run in parallel.

007–0701–150 103
10: Multiprocessing Directives

#pragma Description
#pragma set chunksize Tells the compiler which values to use for chunksize.
#pragma set Tells the compiler which values to use for numthreads.
numthreads
#pragma set schedtype Tells the compiler which values to use for schedtype.
#pragma shared Tells the compiler the names of all the variables that the threads must share.
#pragma synchronize Stops threads until all threads reach here.

OpenMP C/C++ API Multiprocessing Directives

The MIPSpro C and C++ compilers support OpenMP multiprocessing directives.
These directives are based on the OpenMP C/C++ Application Program Interface
(API) standard. Programs that use these directives are portable and can be compiled
by other compilers that support the OpenMP standard.
To enable recognition of the OpenMP directives, specify -mp on the cc or CC
command line.
For more information on how to use these directives, see the MIPSpro C and C++
Pragmas manual.

Using Parallel Regions

To understand how most of the multiprocessing C/C++ compiler directives work,
consider the concept of a parallel region. On some systems, a parallel region is a
single loop that runs in parallel. However, with the multi-processing C/C++
compiler, a parallel region can include several loops and/or independent code
segments that execute in parallel.
A simple parallel region consists of only one work-sharing construct, usually a loop.
(A parallel region consisting of only a serial section or independent code is a waste of
processing resources.)
A parallel region of code can contain sections that execute sequentially as well as
sections that execute concurrently. A single large parallel region has a number of
advantages over a series of isolated parallel regions: each isolated region executes a
single loop in parallel. At the very least, the single large parallel region can help

104 007–0701–150
 C Language Reference Manual

reduce the overhead associated with moving from serial execution to parallel
execution.
Large mixed parallel regions avoid the forced synchronization that occurs at the end
of each parallel region. The large mixed parallel region also allows you to use
#pragma directives that execute independent code sections that run concurrently.
Thus, if a thread finishes its work early, it can go on to execute the next section of
code–provided that the next section of code is not dependent on the completion of the
previous section. However, when you create parallel regions, you need more
sophisticated synchronization methods than you need for isolated parallel loops.

Coding Rules of #pragma Directives

The #pragma directives are modeled after the Parallel Computing Forum (PCF)
directives for parallel FORTRAN. The PCF directives define a broad range of parallel
execution modes and provide a framework for defining corresponding C/C++
#pragma directives.
The following changes have been made to make the #pragma directives more C-like:
• Each #pragma directive starts with #pragma and follows the conventions of
ANSI C for compiler directives. You can use white space before and after the #,
and you must sometimes use white space to separate the words in a #pragma
directive, as with C syntax. A line that contains a #pragma directive can contain
nothing else (code or comments).
• #pragma directives apply to only one succeeding statement. If a directive applies
to more than one statement, you must make a compound statement. C/C++
syntax lets you use curly braces, { }, to do this. Because of the differences between
this syntax and FORTRAN, C/C++ can omit the PCF directives that indicate the
end of a range (for example, END PSECTIONS).
• The #pragma pfor directive replaces the PARALLEL DO directive because the
for statement in C is more loosely defined than the FORTRAN DO statement.
To make it easier to use #pragma directives, you can put several keywords on a
single #pragma directive line, or spread the keywords over several lines. In either
case, you must put the keywords in the correct order, and each directive must contain
an initial keyword. For example, the following two code samples do the same thing:

007–0701–150 105
10: Multiprocessing Directives

Example 1:
#pragma parallel shared(a,b,c, n) local(i) pfor
for (i=0; i<n; i++) a[i]=b[i]+c[i];

Example 2:
#pragma parallel
#pragma shared( a )
#pragma shared( b, c, n )
#pragma local( i )
#pragma pfor
for (i=0; i<n; i++) a[i]=b[i]+c[i];

Parallel Regions
A parallel region consists of a number of work-sharing constructs. The C/C++
compiler supports the following work-sharing constructs:
• A loop executed in parallel
• “Local” code run (identically) by all threads
• An independent code section executed in parallel with the rest of the code in the
parallel region
• Code executed by only one thread
• Code run in “protected mode” by all threads
In addition, the C/C++ compiler supports three types of explicit synchronization. To
account for data dependencies, it is sometimes necessary for threads to wait for all
other threads to complete executing an earlier section of code. Three sets of directives
implement this coordination: #pragma critical, #pragma synchronize, and
#pragma enter gate and #pragma exit gate.
The parallel region should have a single entry at the top and a single exit at the
bottom.
To start a parallel region, use the #pragma parallel directive. To mark a for loop
to run in parallel, use the #pragma pfor directive. To start an independent code
section that executes in parallel with the rest of the code in the parallel region, use the
#pragma independent.

106 007–0701–150
 C Language Reference Manual

When you execute a program, nothing actually runs in parallel until it reaches a
parallel region. Then multiple threads begin (or continue, if this is not the first
parallel region), and the program runs in parallel mode. When the program exits a
parallel region, only a single thread continues (sequential mode) until the program
again enters a parallel region and the process repeats.

Parallel Reduction Operations in C and C++

A reduction operation applies to an array of values and reduces (combines) the array
values into a single value.
Consider the following example:
int a[10000];
int i;
int sum = 0;
for (i = 0; i < 10000; i++)
sum = sum + a[i];

The loop computes the cumulative sum of the elements of the array. Because the
value of a sum computed in one iteration is used in the next iteration, the loop as
written cannot be executed in parallel directly on multiprocessors.
However, you can rewrite the above loop to compute the local sum on each processor
by introducing a local variable. This breaks the iteration dependency of sum and the
loop can be executed in parallel on multiprocessors. This loop computes the local
sum of the elements on each processor, which can subsequently be serially added to
yield the final sum.
The multiprocessing C/C++ compiler provides a reduction clause as a modifier for
a pfor statement. Using this clause, the above loop can be parallelized as follows:
int a[10000];
int i;
int sum = 0
#pragma parallel shared(a, sum) local(i)
#pragma pfor reduction(sum)
for i=0; i<10000; i++)
sum = sum + a[i];

007–0701–150 107
10: Multiprocessing Directives

Restrictions on the Reduction Clause

The following restrictions are imposed on the reduction clause:
• You can specify only variables of integer types (int, short, and so forth) or of
floating point types (float, double, and so forth).
• You can use the reduction clause only with the primitive operations plus (+),
and times (*), which satisfy the associativity property as illustrated in the
following example:
a op (b op c) == (a op b) op c.

The preceding example that uses a reduction clause has the same semantics as the
following code that uses local variables and explicit synchronization. In this code,
because sum is shared, the computation of the final sum has to be done in a critical
region to allow each processor exclusive access to sum:
int a[10000];
int i;
int sum,localsum;
sum = 0;
#pragma parallel shared(a,sum) local(i,localsum)
{
localsum = 0;
#pragma pfor iterate(;;)
for (i = 0; i < 10000; i++) localsum +=a[i];
#pragma critical
sum = sum + localsum;
}

The general case of reduction of a binary operation, op, on an array a1,a2,a3,...an

involves the following computation:
a1 op a2 op a3 op.... op an

When the various operations are performed in parallel, they can be invoked in any
order. In order for the reduction to produce a unique result, the binary operation, op,
must therefore satisfy the associativity property, as shown below:
a op (b op c) == (a op b) op c

108 007–0701–150
 C Language Reference Manual

Performance Considerations
The reduction example in "Restrictions on the Reduction Clause", page 108, has the
drawback that when the number of processors increases, there is more contention for
the lock in the critical region.
The following example uses a shared array to record the result on individual
processors. The array entries are CacheLine apart to prevent write contention on the
cache line (128 bytes in this example. The array permits recording results for up to
NumProcs processors. Both these variables CacheLine and NumProcs can be tuned for a
specific platform:
#define CacheLine 128
int a[10000];
int i, sum;
int *localsum = malloc(NumProcs * CacheLine);

for (i = 0; i < NumProcs; i++)

localsum [i] = 0;

sum = 0;
#pragma parallel shared (a, sum, localsum) local (i) local (myid)
{
myid = mp_my_threadnum();

#pragma pfor
for (i = 0; i < 10000; i++)
localsum [myid] += a [i];
}
for (i = 0; i < numprocs; i++)
sum += localsum[i];

The only operation in the critical region is the computation of the final result from the
local results on individual processors.
In the case when the reduction applies to an array of integers, the reduction function
can be specialized by using an intrinsic operation __fetch_and_<op> rather than
the more expensive critical region. (See "Synchronization Intrinsics", page 123">.)
For example, to add an array of integers, the critical region can be replaced by the
following call:
__fetch_and_add(&sum, localsum);

007–0701–150 109
10: Multiprocessing Directives

The intrinsic __fetch_and_<op> is defined only for the following operations: add,
sub, or, xor, nand, mpy, min, and max; and for the type integers together with their
size and signed variants. Therefore, it cannot be used in the general case.

Reduction on User-Defined Types in C++

In C++ a generalized reduction function can be written for any user-defined binary
operator op that satisfies the associative property.

Reduction Example

The following generic function performs reduction on an array of elements of type

ElemType, with array indices of type IndexType, and a binary operation op that
takes two arguments of type ElemType and produces a result of type ElemType.
The type IndexType is assumed to have operators <, -, and ++ defined on it. The
use of a function object plus is in keeping with the spirit of generic programming as
in the Standard Template Library (STL). A function object is preferred over a function
pointer because it permits inlining:
template <class ElemType, class IndexType, class BinaryOp>
ElemType reduction(IndexType first, IndexType last,
ElemType zero, ElemType ar[],
BinaryOp op) {

ElemType result = zero;

IndexType i;
#pragma parallel shared (result, ar) local (i) byvalue(zero, first, last)
{
ElemType localresult = zero;

#pragma pfor
{
for (i = first; i < last - first; i++)
localresult = op(localresult,ar[i]);
}

110 007–0701–150
 C Language Reference Manual

#pragma critical
result = op(result,localresult);
}

return result;
}

With the preceding definition of reduction, you can perform the following reduction:
adsum = reduction(0,size,0,ad,plus<double>());
acsum = reduction(0,size,czero,ac,plus<Complex>());

Restrictions for the C++ Compiler

This section summarizes some restrictions that are relevant only for the C++ compiler.
It also lists some restrictions that result from the interaction between pragmas and
C++ semantics.

Restrictions on pfor
If you are writing a pfor loop for the multiprocessing C++ compiler, the index
variable i can be declared within the for statement using the following:
int i = 0;

The ANSI C++ standard states that the scope of the index variable declared in a for
statement extends to the end of the for statement, as in this example:
#pragma pfor
for (int i = 0, ...) { ... }

The MIPSpro 7.2 C++ compiler does not enforce this rule. By default, the scope
extends to the end of the enclosing block. The default behavior can be changed by
using the command line option -LANG:ansi-for-init-scope=on which enforces
the ANSI C++ standard.
To avoid future problems, write for loops in accordance with the ANSI standard, so a
subsequent change in the compiler implementation of the default scope rules does not
break your code.

007–0701–150 111
10: Multiprocessing Directives

Restrictions on Exception Handling

The following restrictions apply to exception handling by the multiprocessing C++
compiler:
• A throw cannot cross an multiprocessing parallel region boundary; it must be
caught within the multiprocessing region.
A thread that throws an exception must catch the exception as well. For example,
the following program is valid. Each thread throws and catches an exception:
extern ‘‘C’’ printf(char *,...);
extern ‘‘C’’ int mp_my_threadnum();
main() {
int localmax,n;

#pragma parallel local (localmax,n)

{
localmax = 0;

try {
throw 10;
}
/* .... */
catch (int) {
printf(‘‘!!!!exception caught in process \n’’);
printf(‘‘My thread number is %d\n’’,mp_my_threadnum());
} /* end of try block */
} /* end of parallel region */
}

• An attempt to throw (propagate) an exception past the end of a parallel program

region results in a runtime abort. All other threads abort.
For example, if the following program is executed, all threads abort:
extern ‘‘C’’ printf(char *,...);
void ehfn() {
try {
throw 10;
}
catch (double) // not a handler for throw 10
{

112 007–0701–150
 C Language Reference Manual

printf(‘‘exception caught in process \n’’);

}
}

main() {
#pragma parallel
{
ehfn();
}
}

The program aborts even if a handler is present in main(), as in the following

example:
main() {
#pragma parallel
{
try {
ehfn();
}
catch (...) {};
}

The reason this program aborts is that the throw propagates past the
multiprocessing region.

Scoping Restrictions
The following default scope rules apply for the C++ multiprocessing compiler.
• Class objects or structures that have constructors [that is, non-pods (plain old data
structures)] cannot be placed on the local list of #pragma parallel.
The following is invalid:
class C {
....
};

main() {

C c;

007–0701–150 113
10: Multiprocessing Directives

#pragma parallel local (c) // Class object c cannot be in local list

{
....
}
}

Instead, declaring such objects within the parallel region allows the default rule to
be used to indicate that they are local (as the following example illustrates):
main() {
#pragma parallel
{
C c;
....
}
}

• Structure fields and class object members cannot be placed on the local list.
Instead, the entire class object must be made local.
• Values of variables in the local list are not copied into each processor’s local
variables; instead, initialize locals within the parallel program text. For example,
main() {
int i;

i = 0;
#pragma parallel local(i)
{
// Here i is not 0.
// Explicit initialization of i within the parallel region
// is necessary
}
}

114 007–0701–150
 Chapter 11

Multiprocessing Advanced Features

A number of features are provided so that you can override the multiprocessing
defaults and customize the parallelism to your particular applications. The following
sections provide brief explanations of these features.

Run-time Library Routines

The SGI multiprocessing C and C++ compiler provides the following routines for
customizing your program.

mp_block and mp_unblock

The mp_block routine puts the slave threads into a blocked state using the
blockproc system call. The slave threads stay blocked until a call is made to the
mp_unblock routine. These routines are useful if the job has bursts of parallelism
separated by long stretches of single processing, as with an interactive program. You
can block the slave processes so they consume CPU cycles only as needed, thus
freeing the machine for other users. The system automatically unblocks the slaves on
entering a parallel region if you neglect to do so.

mp_setup, mp_create, and mp_destroy

The mp_setup, mp_create, and mp_destroy subroutine calls create and destroy
threads of execution. This can be useful if the job has only one parallel portion or if
the parallel parts are widely scattered. When you destroy the extra execution threads,
they cannot consume system resources; they must be recreated when needed. Use of
these routines is discouraged because they degrade performance; the mp_block and
mp_unblock routines should be used in almost all cases.
mp_setup takes no arguments. It creates the default number of processes as defined
by previous calls to mp_set_numthreads, by the MP_SET_NUMTHREADS
environment variable, or by the number of CPUs on the current hardware platform.
mp_setup is called automatically when the first parallel loop is entered to initialize
the slave threads.

007–0701–150 115
11: Multiprocessing Advanced Features

mp_create takes a single integer argument, the total number of execution threads
desired. Note that the total number of threads includes the master thread. Thus,
mp_create(n) creates one thread less than the value of its argument. mp_destroy
takes no arguments; it destroys all the slave execution threads, leaving the master
untouched.
When the slave threads die, they generate a SIGCLD signal. If your program has
changed the signal handler to catch SIGCLD, it must be prepared to deal with this
signal when mp_destroy is executed. This signal also occurs when the program
exits; mp_destroy is called as part of normal cleanup when a parallel job terminates.

mp_blocktime
The slave threads spin wait until there is work to do. This makes them immediately
available when a parallel region is reached. However, this consumes CPU resources.
After enough wait time has passed, the slaves block themselves through blockproc.
Once the slaves are blocked, it requires a system call to unblockproc to activate the
slaves again (refer to the unblockproc(2) man page for details). This makes the
response time much longer when starting up a parallel region.
This trade-off between response time and CPU usage can be adjusted with the
mp_blocktime call. The mp_blocktime routine takes a single integer argument
that specifies the number of times to spin before blocking. By default, it is set to
10,000,000; this takes roughly one second. If called with an argument of 0, the slave
threads will not block themselves no matter how much time has passed. Explicit calls
to mp_block, however, will still block the threads.
This automatic blocking is transparent to the user’s program; blocked threads are
automatically unblocked when a parallel region is reached.

mp_numthreads, mp_suggested_numthreads, mp_set_numthreads

Occasionally, you may want to know how many execution threads are available. The
mp_numthreads routine is a zero-argument integer function that returns the total
number of execution threads for this job. The count includes the master thread. In
addition, this routine has the side effect of freezing (for eternity) the number of threads
to the returned value, so this routine should be used sparingly. To determine the
number of threads without this freeze property, use mp_suggested_numthreads.
mp_suggested_numthreads takes an unsigned integer and uses the supplied value
as a hint about how many threads to use in subsequent parallel regions. It returns the

116 007–0701–150
 C Language Reference Manual

previous value of the number of threads to be employed in parallel regions. It does

not affect currently executing parallel regions, if any. The implementation may ignore
this hint depending on factors such as overall system load. This routine may also be
called with the value 0, in which case it simply returns the number of threads to be
employed in parallel regions.
mp_set_numthreads takes a single integer argument. It changes the default number
of threads to the specified value. A subsequent call to mp_setup will use the specified
value rather than the original defaults. If the slave threads have already been created,
this call will not change their number. It has an effect only when mp_setup is called.

mp_my_threadnum
The mp_my_threadnum routine is a zero-argument function that allows a thread to
differentiate itself while in a parallel region. If there are n execution threads, the
function call returns a value between zero and n – 1. The master thread is always
thread zero. This function can be useful when parallelizing certain kinds of loops.
Most of the time the loop index variable can be used for the same purpose.
Occasionally, the loop index may not be accessible, as, for example, when an external
routine is called from within the parallel loop. This routine provides a mechanism for
those cases.

mp_setlock, mp_unsetlock, mp_barrier

The mp_setlock, mp_unsetlock, and mp_barrier zero-argument subroutines
provide convenient (although limited) access to the locking and barrier functions
provided by ussetlock, usunsetlock, and barrier. These subroutines are
convenient because you do not need to initialize them; calls such as usconfig and
usinit are done automatically. The limitation is that there is only one lock and one
barrier. For most programs, this amount is sufficient. If your program requires more
complex or flexible locking facilities, use the ussetlock family of subroutines
directly.

mp_set_slave_stacksize
The mp_set_slave_stacksize routine sets the stack size (in bytes) to be used by
the slave processes when they are created (using sprocsp). The default size is 16
MB. Slave processes only allocate their local data onto their stack, shared data (even if
allocated on the master’s stack) is not counted.

007–0701–150 117
11: Multiprocessing Advanced Features

Run-time Environment Variables

The SGI multiprocessing C and C++ compiler provides the following environment
variables that you can use to customize your program.

MP_SET_NUMTHREADS, MP_BLOCKTIME, MP_SETUP

The MP_SET_NUMTHREADS, MP_BLOCKTIME, and MP_SETUP environment variables
act as an implicit call to the corresponding routine(s) of the same name at program
start-up time.
For example, the following csh command causes the program to create two threads
regardless of the number of CPUs actually on the machine, as does the source
statement below it:
csh command:
% setenv MP_SET_NUMTHREADS 2

Source statement:
mp_set_numthreads (2)

Similarly, the following sh commands prevent the slave threads from autoblocking, as
does the source statement:
sh commands:
% set MP_BLOCKTIME 0
% export MP_BLOCKTIME

Source statement:
mp_blocktime (0);

For compatibility with older releases, the environment variable NUM_THREADS is

supported as a synonym for MP_SET_NUMTHREADS.
To help support networks with several multiprocessors and several CPUs, the
environment variable MP_SET_NUMTHREADS also accepts an expression involving
integers +, –, min, max, and the special symbol “all,” which stands for the number of
CPUs on the current machine. For example, the following command selects the
number of threads to be two fewer than the total number of CPUs (but always at
least one):

118 007–0701–150
 C Language Reference Manual

% setenv MP_SET_NUMTHREADS max(1,all-2)

MP_SUGNUMTHD, MP_SUGNUMTHD_MIN, MP_SUGNUMTHD_MAX, MP_SUGNUMTHD_VERBOSE

In an environment with long running jobs and varying workloads, it may be
preferable to vary the number of threads during execution of some jobs.
Setting MP_SUGNUMTHD causes the run-time library to create an additional,
asynchronous process that periodically wakes up and monitors the system load.
When idle processors exist, this process increases the number of threads, up to a
maximum of MP_SET_NUMTHREADS. When the system load increases, it decreases the
number of threads, possibly to as few as 1. When MP_SUGNUMTHD has no value, this
feature is disabled and multithreading works as before.
The environment variables MP_SUGNUMTHD_MIN and MP_SUGNUMTHD_MAX are used
to limit this feature as desired. When MP_SUGNUMTHD_MIN is set to an integer value
between 1 and MP_SET_NUMTHREADS, the process will not decrease the number of
threads below that value.
When MP_SUGNUMTHD_MAX is set to an integer value between the minimum number
of threads and MP_SET_NUMTHREADS, the process will not increase the number of
threads above that value.
If you set any value in the environment variable MP_SUGNUMTHD_VERBOSE,
informational messages are written to stderr whenever the process changes the
number of threads in use.
Calls to mp_numthreads and mp_set_numthreads are taken as a sign that the
application depends on the number of threads in use. The number in use is frozen
upon either of these calls; and if MP_SUGNUMTHD_VERBOSE is set, a message to that
effect is written to stderr.

MP_SCHEDTYPE, CHUNK
These environment variables specify the type of scheduling to use on for loops that
have their scheduling type set to RUNTIME. For example, the following csh
commands cause loops with the RUNTIME scheduling type to be executed as
interleaved loops with a chunk size of 4:

% setenv MP_SCHEDTYPE INTERLEAVE

% setenv CHUNK 4

007–0701–150 119
11: Multiprocessing Advanced Features

The defaults are the same as on the #pragma pfor directive; if neither variable is
set, SIMPLE scheduling is assumed. If MP_SCHEDTYPE is set, but CHUNK is not set, a
CHUNK of 1 is assumed. If CHUNK is set, but MP_SCHEDTYPE is not, DYNAMIC
scheduling is assumed.

MP_SLAVE_STACKSIZE
The stack size of slave processes can be controlled through the environment variable
MP_SLAVE_STACKSIZE, which may be set to the desired stacksize in bytes. The
default value is 16 MB (4 MB for more than 64 threads).

MPC_GANG
MPC_GANG specifies gang scheduling. Set MPC_GANG to ON to enable gang scheduling.
To disable gang scheduling, set MPC_GANG to OFF.

Communicating Between Threads Through Thread Local Data

The routines described in this section allow you to perform explicit communication
between threads within their multiprocessing C program. These communication
mechanisms are similar to message-passing, one-sided-communication, or shmem, and
may be desirable for reasons of performance and/or style.
The operations allow a thread to fetch from (get) or send to (put) data belonging to
other threads. Therefore, these operations can be performed only on data that has
been declared to be -Xlocal (that is, each thread has its own private copy of that
data; see the ld(1) man page for details on Xlocal). A get operation requires that
the source parameter point to Xlocal data, while a put operation requires that the
target parameter point to Xlocal data.
The following routines are available as part of the Message Passing Toolkit (MPT) and
are similar to the original shmem routines (see the shmem reference page), but are
prefixed by mp_:

120 007–0701–150
 C Language Reference Manual

void mp_shmem_get32 (int *target,

int *source,
int length,
int source_thread)

void mp_shmem_put32 (int *target,

int *source,
int length,
int target_thread)

void mp_shmem_iget32 (int *target,

int *source,
int target_inc,
int source_inc,
int length,
int source_thread)

void mp_shmem_iput32 (int *target,

int *source,
int target_inc,
int source_inc,
int length,
int target_thread)

void mp_shmem_get64(long long *target,

long long *source,
int length,
int source_thread)

void mp_shmem_put64 (long long *target,

long long *source,
int length,
int target_thread)

void mp_shmem_iget64 (long long *target,

long long *source,
int target_inc,
int source_inc,
int length,
int source_thread)

007–0701–150 121
11: Multiprocessing Advanced Features

void mp_shmem_iput64 (long long *target,

long long *source,
int target_inc,
int source_inc,
int length,
int target_thread)

The following rules apply to the preceding listed routines:

• Both source and target are pointers to 32-bit quantities for the 32-bit versions, and
to 64-bit quantities for the 64-bit versions of the calls. The actual type of the data
is not important, because the routines perform a bit-wise copy.
• For a put operation, the target must be Xlocal. For a get operation, the source
must be Xlocal.
• length specifies the number of elements to be copied, in units of 32 or 64-bit
elements, as appropriate.
• source_thread and target_thread specify the thread-number of the remote processing
element (PE).
• A get operation copies from the remote PE. A put operation copies to the remote
PE.
• target_inc and source_inc are specified for the strided iget and iput operations.
They specify the increment (in units of 32–bit or 64–bit elements) for source and
target when performing the data transfer. The number of elements copied during
a strided put or get operation is still determined by length.

Note: Call these routines only after the threads have been created (typically, the first
pfor/parallel region). Performing these operations while the program is still serial
leads to a run-time error because each thread’s copy has not yet been created.

In the example below, compiling with -Wl,-Xlocal, myvars ensures that each
thread has a private copy of x and y.
struct {
int x;
double y[100];
} myvars;

122 007–0701–150
 C Language Reference Manual

The following example copies the value of x on thread 3 into the private copy of x for
the current thread.
mp_shmem_get32 (&x, &x, 1, 3)

The next example copies the value of localvar into the thread 5 copy of x.
mp_shmem_put32 (&x, &localvar, 1, 5)

The example below fetches values from the thread 7 copy of array y into
localarray.
mp_shmem_get64 (&localarray, &y, 100, 7)

The next example copies the value of every other element of localarray into the
thread 9 copy of y.
mp_shmem_iput64 (&y, &localarray, 2, 2, 50, 9)

Synchronization Intrinsics
The intrinsics described in this section provide a variety of primitive synchronization
operations. Besides performing the particular synchronization operation, each of these
intrinsics has two key properties:
• The function performed is guaranteed to be atomic (typically achieved by
implementing the operation using a sequence of load-linked and/or
store-conditional instructions in a loop).
• Associated with each instrinsic are certain memory barrier properties that restrict
the movement of memory references to visible data across the intrinsic operation
(by either the compiler or the processor).
A visible memory reference is a reference to a data object potentially accessible by
another thread executing in the same shared address space. A visible data object can
be one of the following:
• C/C++ global data
• Data declared extern
• Volatile data
• Static data (either file-scope or function-scope)

007–0701–150 123
11: Multiprocessing Advanced Features

• Data accessible via function parameters

• Automatic data (local-scope) that has had its address taken and assigned to some
visible object (recursively)
The memory barrier semantics of an intrinsic can be one of the following three types:
• acquire barrier: disallows the movement of memory references to visible data
from after the intrinsic (in program order) to before the intrinsic. (This behavior is
desirable at lock-acquire operations.)
• release barrier: disallows the movement of memory references to visible data from
before the intrinsic (in program order) to after the intrinsic. (This behavior is
desirable at lock-release operations.)
• full barrier: disallows the movement of memory references to visible data past the
intrinsic (in either direction), and is thus both an acquire and a release barrier. A
barrier restricts only the movement of memory references to visible data across the
intrinsic operation: between synchronization operations (or in their absence),
memory references to visible data may be freely reordered subject to the usual
data-dependence constraints.
By default, it is assumed that a memory barrier applies to all visible data. If you
know the precise set of data objects that must be restricted by the memory barrier,
you can specify the set of data objects as additional arguments to the intrinsic. In this
case, the memory barrier restricts the movement of memory references to the
specified list of data objects only, possibly resulting in better performance. The
specified data objects must be simple variables and cannot be expressions (for
example, &p and *p are disallowed).

Caution: Conditional execution of a synchronization intrinsic (such as within an if

! or a while statement) does not prevent the movement of memory references to
visible data past the overall if or while construct.

Atomic fetch-and-op Operations

The fetch-and-op operations are as follows:
<type> __fetch_and_add (<type>* ptr, <type> value, ...)
<type> __fetch_and_sub (<type>* ptr, <type> value, ...)
<type> __fetch_and_or (<type>* ptr, <type> value, ...)
<type> __fetch_and_and (<type>* ptr, <type> value, ...)

124 007–0701–150
 C Language Reference Manual

<type> __fetch_and_xor (<type>* ptr, <type> value, ...)

<type> __fetch_and_nand(<type>* ptr, <type> value, ...)
<type> __fetch_and_mpy (<type>* ptr, <type> value, ...)
<type> __fetch_and_min (<type>* ptr, <type> value, ...)
<type> __fetch_and_max (<type>* ptr, <type> value, ...)

<type> can be any of the following:

int
long
long long
unsigned int
unsigned long
unsigned long long

The ellipses (...) refer to an optional list of variables protected by the memory
barrier.
Each of these operations behaves as follows:
• Atomically performs the specified operation with the given value on *ptr, and
returns the old value of *ptr.
{tmp = *ptr; *ptr <op>= value; return tmp;}

• Full barrier

Atomic op-and-fetch Operations

The op-and-fetch operations are as follows:
<type> __add_and_fetch (<type>* ptr, <type> value, ...)
<type> __sub_and_fetch (<type>* ptr, <type> value, ...)
<type> __or_and_fetch (<type>* ptr, <type> value, ...)
<type> __and_and_fetch (<type>* ptr, <type> value, ...)
<type> __xor_and_fetch (<type>* ptr, <type> value, ...)
<type> __nand_and_fetch(<type>* ptr, <type> value, ...)
<type> __mpy_and_fetch (<type>* ptr, <type> value, ...)
<type> __min_and_fetch (<type>* ptr, <type> value, ...)
<type> __max_and_fetch (<type>* ptr, <type> value, ...)

<type> can be any of the following:

007–0701–150 125
11: Multiprocessing Advanced Features

int
long
long long
unsigned int
unsigned long
unsigned long long

Each of these operations behaves as follows:

• Atomically performs the specified operation with the given value on *ptr, and
returns the new value of *ptr.
{*ptr <op>= value; return *ptr;}

• Full barrier

Atomic compare-and-swap Operation

The compare-and-swap operation is as follows:
int __compare_and_swap (<type>* ptr, <type> oldvalue, <type> newvalue, ...)

<type> can be one of the following:

int
long
long long
unsigned int
unsigned long
unsigned long long

This operation behaves as follows:

• Atomically compares *ptr to oldvalue. If equal, it stores the new value and returns
1, otherwise it returns 0.
if (*ptr != oldvalue) return 0;
else {
*ptr = newvalue;
return 1;
}

• Full barrier

126 007–0701–150
 C Language Reference Manual

Atomic synchronize Operation

The synchronize operation is as follows:
__synchronize (...)

The ellipses (...) refer to an optional list of variables protected by the memory
barrier.

This operation behaves as follows:

• Issues a sync operation
• Full barrier

Atomic lock and unlock Operations

Atomic lock-test-and-set Operation

The lock-test-and-set operation is as follows:

<type> __lock_test_and_set (<type>* ptr, <type> value, ...)

<type> can be any of the following:

int
long
long long
unsigned int
unsigned long
unsigned long long

This operation behaves as follows:

• Atomically stores the supplied value in *ptr and returns the old value of *ptr
{tmp = *ptr; *ptr = value; return tmp;}

• Acquire barrier

Atomic lock-release Operation

The lock_release operation is as follows:

void __lock_release (<type>* ptr, ...)

007–0701–150 127
11: Multiprocessing Advanced Features

<type> can be one of the following:

int
long
long long
unsigned int
unsigned long
unsigned long long

This operation behaves as follows:

• Issues sync then sets *ptr to 0 and flushes it from the register
{*ptr = 0}

• Release barrier

Example of Implementing a Pure Spin-Wait Lock

The following example shows implementation of a spin-wait lock:
int lockvar = 0;
while (__lock_test_and_set (&lockvar, 1) != 0); /* acquire the lock */
... read and update shared variables ...
__lock_release (&lockvar); /* release the lock */

The memory barrier semantics of the intrinsics guarantee that no memory reference to
visible data is moved out of the above critical section, either ahead of the lock-acquire
or past the lock-release.

Note: Pure spin-wait locks can perform poorly under heavy contention.

If the data structures protected by the lock are known precisely (for example, x, y,
and z in the example below), then those data structures can be precisely identified as
follows:
int lockvar = 0;
while (__lock_test_and_set (&lockvar, 1, x, y, z) != 0);
... read/modify the variables x, y, and z ...
__lock_release (&lockvar, x, y, z);

128 007–0701–150
 Appendix A

Implementation-Defined Behavior

The sections in this appendix describe implementation-defined behavior. Each section

is keyed to the ANSI C Standard (ANSI X3.159-1989), Appendix F, and each point is
keyed to the section number of the ANSI C Standard. The bold lines, usually marked
with bullets, are items from Appendix F of the ANSI C Standard. Text following the
italic lines describes the SGI implementation.

Translation (F.3.1)
• Whether each nonempty sequence of white-space characters other than newline
is retained or replaced by one space character (2.1.1.2).
A nonempty sequence of white-space characters (other than newline) is retained.
• How a diagnostic is identified (2.1.1.3).
Successful compilations are silent. Diagnostics are, in general, emitted to standard
error. Diagnostic messages have the general pattern of
file-name,line-number:severity(number): message in -n32 and -64 modes.
Diagnostics have a slightly different pattern in -o32 mode. Also, the range of
numbers in -o32 mode is disjointed from the range in -n32 and -64 modes.
For example, typical messages from the ANSI C compiler front end in -n32 and
-64 mode look like this:
"t4.c’’, line 4: error(1020):identifier "x’’ is undefined
"t4.c’’, line 5: warning(1551):variable "y’’ is used before its
value is set

Messages can also be issued by other internal compiler passes.

• Classes of diagnostic messages, their return codes and control over them.
Three classes of messages exist: warning, error, and remark. Warning messages
include the notation “warning” (which can be capitalized), and allow the
compilation to continue (return code 0). Error messages cause the compilation to
fail (return code 1).
Remark messages appear in -n32 and -64 modes only. Typically, remarks are
issued only if the -fullwarn option appears on the command line. More control

007–0701–150 129
A: Implementation-Defined Behavior

is available with the -diag_warning, -diag_remark, and -diag_error

options. (See the cc reference page for more information.)
Warning messages from the compiler front end have a unique diagnostic number.
You can suppress these messages individually by putting the number in the
numberlist of a -woff numberlist switch to the cc command. numberlist is a
comma-separated list of warning numbers and ranges of warning numbers. For
example, to suppress the warning message in the previous example, enter
-woff 1551
To suppress warning messages numbered 1642, 1643, 1644, and 1759, enter
-woff 1642-1644,1759

Environment (F.3.2)
• Support of freestanding environments.
No support is provided for a freestanding environment.
• The semantics of the arguments to main (2.1.2.2.1).
main is defined to have the two required parameters argc and argv. A third
parameter, envp, is provided as an extension. That is, main would have the
equivalent of the following prototype:
int main(int argc, char *argv[], char *envp[])

The parameters have the following semantics:

– argc is the number of arguments on the command line.
– argv[0..argc-1] are pointers to the command-line arguments (strings).
– argv[0] is the program name, as it appeared on the command line.
– argv[argc] is a null pointer.
– envp is an array of pointers to strings of the form NAME=value, where NAME
is the name of an environment variable and value is its value. The array is
terminated by a null pointer.
• What constitutes an interactive device (2.1.2.3).

130 007–0701–150
 C Language Reference Manual

Asynchronous terminals, including windows, are interactive devices and are, by

default, line buffered. In addition, the standard error device, stderr, is
unbuffered by default.

Identifiers (F.3.3)
• The number of significant initial characters (beyond 31) in an identifier without
external linkage (3.1.2).
All characters are significant.
• The number of significant initial characters (beyond 6) in an identifier with
external linkage (3.1.2).
All characters are significant.
• Whether case distinctions are significant in an identifier with external linkage
(3.1.2).
Case distinctions are always significant.

Characters (F.3.4)
• The members of the source and execution character sets, except as explicitly
specified in the standard (2.2.1).
Only the mandated characters are present. The source character set includes all
printable ASCII characters, hexadecimal 0x20 through 0x7e, and 0x7 through 0xc
(the standard escape sequences).
• The values to which the standard escape sequences are translated (2.2.2).
The escape sequences are translated as specified for standard ASCII: \a = 0x7, \b
= 0x8, \f = 0xc, \n = 0xa, \r = 0xd, \t = 0x9, \v=0xb
• The shift states used for the encoding of multibyte characters (2.2.1.2).

The multibyte character set is identical to the source and execution character sets.
There are no shift states.
• The number of bits in a character in the execution character set (2.2.4.2.1).
There are eight bits per character.

007–0701–150 131
A: Implementation-Defined Behavior

• The mapping of members of the source character set (in character constants and
string literals) to members of the execution character set (3.1.3.4).
The mapping is the identity mapping.
• The value of an integer character constant that contains a character or escape
sequence not represented in the basic execution character set or in the extended
character set for a wide character constant (3.1.3.4).
With the exception of newline (0xa), backslash (’\’), and 0xff (end-of-file), eight-bit
values appearing in an integer character constant are placed in the resultant
integer in the same fashion as are characters that are members of the execution
character set (see below). A backslash, newline, or 0xff can be placed in a
character constant by preceding it with a backslash (that is, “escaping” it).
• The value of an integer character constant that contains more than one character
or a wide character constant that contains more than one multibyte character
(3.1.3.4).
You can assign up to four characters to an int using a character constant, as the
following example illustrates:
int t = ’a’; /* integer value 0x61 */
int t2 = ’ab’; /* integer value 0x6162 */
int t4 = ’abcd’; /* integer value 0x61626364 */
int t4 = ’abcde’; /* error: too many characters for */
/* character constant */

The encoding of multiple characters in an integer consists of the assignment of the

corresponding character values of the ncharacters in the constant to the
least-significant n bytes of the integer, filling any unused bytes with zeros. The
most significant byte assigned contains the value of the lexically first character in
the constant.
Because the multibyte character set is identical to the source and execution
character sets, the above discussion applies to the assignment of more than one
multibyte character to a wide character constant.
• The current locale used to convert multibyte characters into corresponding wide
character (codes) for a wide character constant (3.1.3.4).
The mapping is the identity mapping to the standard ASCII character set. The C
locale is used.

132 007–0701–150
 C Language Reference Manual

• Whether a “plain” char has the same range of values as signed char or
unsigned char.
Plain char is the same as unsigned char by default. Use the -signed option
to cc to switch the range to be that of signed char.

Integers (F.3.5)
• The representations and sets of values of the various types of integers (3.1.2.5).
Integers are two’s complement binary. Table A-1, page 133, lists the sizes and
ranges of the various types of integer. The use of long long results in a warning
in -ansi and -ansiposix modes.
In -o32 and -n32 mode implementations, to take full advantage of the support
for 64-bit integral values in -ansi and -ansiposix modes, you can define the
macro _LONGLONG on the cc command line when using the types __uint64_t,
__int64_t, or library routines that are prototyped in terms of these types.

Table A-1 Integer Types and Ranges

Type Range: Low High Size (bits)

signed char –128 127 8
char, unsigned char 0 255 8
short, signed short –32768 32767 16
unsigned short int 0 65535 16
int, signed int –2147483648 2147483647 32
unsigned int 0 4294967295 32
long, signed long int –2147483648 (-32 and -n32 2147483647 (-32 and -n32 32
modes) modes) 64
–9223372036854775808 (-64 9223372036854775807 (-64
mode) mode)

007–0701–150 133
A: Implementation-Defined Behavior

Type Range: Low High Size (bits)

unsigned long int 0 4294967295 (-32 and -n32 32
modes) 64
18446744073709551615 (-64
mode)
long long signed long –9223372036854775808 9223372036854775807 64
long int
unsigned long long int 0 18446744073709551615 64

• The result of converting an integer to a shorter signed integer, or the result of

converting an unsigned integer to a signed integer of equal length, if the value
cannot be represented (3.2.1.2).
The least significant n bits (n being the length of the result integer) of the source
are copied to the result.
• The results of bitwise operations on signed integers (3.3).
With the exception of right-shift of a negative signed integer (defined below),
operations on signed and unsigned integers produce the same bitwise results.
• The sign of the remainder on integer division (3.3.5).
The sign of the remainder is that of the numerator.
• The result of a right shift of a negative-valued signed integral type (3.3.7).
The sign bit is propagated, so the result value is still negative.

Floating Point (F.3.6)

• The representations and sets of values of the various types of floating-point
numbers (3.1.2.5).
The representation is IEEE:
– Single (for float values)
– Double (for double values and for long double values in -o32 mode)
– Quad precision (for long double values in -n32 and -64 mode).

134 007–0701–150
 C Language Reference Manual

See ANSI/IEEE Standard 754-1985 and IEEE Standard for Binary Floating-Point
Arithmetic. Table A-2, page 135, lists ranges of floating point types.

Table A-2 Ranges of floating point Types

Type Range: Min Max Size (Bits)

float 1.1755e-38 3.4028e+38 32
double 2.225e-308 1.7977e+308 64
long double 2.225e-308 1.7977e+308 128 (-n32 and -64
modes)

• The type of rounding or truncation used when representing a floating-point

constant which is within its range.
Per IEEE, the rounding is round-to-nearest (IEEE Standard 754, sections 4.1 and
5.5). If the two values are equally near, then the one with the least significant bit
zero is chosen.
• The direction of truncation when an integral number is converted to a
floating-point number that cannot exactly represent the original value (3.2.1.3).
Conversion of an integral type to a float type, if the integral value is too large to
be exactly represented, gives the next higher value.
• The direction of truncation or rounding when a floating-point number is
converted to a narrower floating-point number.
Per IEEE, the rounding is round-to-nearest (IEEE Standard 754, Section 4.1 and
5.5). If the two values are equally near, then the one with the least significant bit
zero is chosen.

Arrays and Pointers (F.3.7)

• The type of integer required to hold the maximum size of an array— that is, the
type of the sizeof operator, size_t(3.3.3.4, 4.1.1).
An unsigned long holds the maximum array size.

007–0701–150 135
A: Implementation-Defined Behavior

• The size of integer required for a pointer to be converted to an integer type

(3.3.4).
long ints are large enough to hold pointers in -n32 and -o32 mode. Both are
32 bits wide.
long ints are large enough to hold pointers in -64 mode. Both are 64 bits wide.
• The result of casting a pointer to an integer or vice versa (3.3.4).
The result is bitwise exact provided the integer type is large enough to hold a
pointer.
• The type of integer required to hold the difference between two pointers to
elements of the same array, ptrdiff_t (3.3.6, 4.1.1).
An int is large enough to hold the difference between two pointers to elements of
the same array in -o32 and -n32 modes.
A long int is large enough to hold the difference between two pointers to
elements of the same array in -n32, -o32, and -64 modes.

Registers (F.3.8)
• The extent to which objects can actually be placed in registers by use of the
register storage-class specifier (3.5.1).
The compilation system can use up to eight of the register storage-class specifiers
for nonoptimized code in -32 mode, and it ignores register specifiers for formal
parameters. Use of register specifiers is not recommended.
The register storage-class specifier is always ignored and the compilation system
makes its own decision about what should be in registers for optimized code (-O2
and above).

Structures, Unions, Enumerations, and Bitfields (F.3.9)

• What is the result if a member of a union object is accessed using a member of
a different type (3.3.2.3).
The bits of the accessed member are interpreted according to the type used to
access the member. For integral types, the N bits of the type are simply accessed.

136 007–0701–150
 C Language Reference Manual

For floating types, the access might cause a trap if the bits are not a legal floating
point value. For pointer types, the 32 bits (64 bits if in -64 mode) of the pointer
are picked up. The usability of the pointer depends on whether it points to a valid
object or function, and whether it is used appropriately. For example, a pointer
whose least-significant bit is set can point to a character, but not to an integer.
• The padding and alignment of members of structures (3.5.2.1).
This should present no problem unless binary data written by one implementation
are read by another.
Members of structures are on the same boundaries as the base data type
alignments anywhere else. A word is 32 bits and is aligned on an address, which
is a multiple of 4. unsigned and signed versions of a basic type use identical
alignment. Type alignments are given in Table A-3, page 137.

Table A-3 Alignment of Structure Members

Type Alignment
long double Double- word boundary (-32 mode)
Quad-word boundary (-n32 and -64 modes)
double Double-word boundary
float Word boundary
long long Double-word boundary
long Word boundary (-n32 and -32 modes)
double-word boundary (-64 mode)
int Word boundary
pointer Word boundary
short Half-word boundary
char Byte boundary

• Whether a “plain” int bit-field is treated as a signed int bit-field or as an

unsigned int bit-field (3.5.2.1).
A “plain” int bit-field is treated as a signed int bit-field.
• The order of allocation of bitfields within a unit (3.5.2.1).

007–0701–150 137
A: Implementation-Defined Behavior

Bits in a bitfield are allocated with the most-significant bit first within a unit.
• Whether a bitfield can straddle a storage-unit boundary (3.5.2.1).
Bitfields cannot straddle storage unit boundaries (relative to the beginning of the
struct or union), where a storage unit can be of size 8, 16, 32, or 64 bits.
• The integer type chosen to represent the values of an enumeration type (3.5.2.2).

The int type is always used.

Note: long or long long enumerations are not supported.

Qualifiers (F.3.10)
• What constitutes an access to an object that has volatile-qualified type (3.5.3).
Objects of volatile-qualified type are accessed only as specified by the abstract
semantics, and as would be expected on a RISC architecture, no complex
instructions exist (for example, read-modify-write). Volatile objects appearing on
the left side of an assignment expression are accessed once for the write. If the
assignment is not simple, an additional read access is performed. Volatile objects
appearing in other contexts are accessed once per instance. Incrementation and
decrementation require both a read and a write access.
Volatile objects that are memory-mapped are accessed only as specified. If such an
object is of size char, for example, adjacent bytes are not accessed. If the object is
a bitfield, a read may access the entire storage unit containing the field. A write of
an unaligned field necessitates a read and write of the storage unit that contains it.

Declarators (F.3.11)
• The maximum number of declarators that can modify an arithmetic, structure, or
union type (3.5.4).
There is no limit.

138 007–0701–150
 C Language Reference Manual

Statements (F.3.12)
• The maximum number of case values in a switch statement (3.6.4.2).
There is no limit.

Preprocessing Directives (F.3.13)

• Whether the value of a single-character character constant in a constant
expression that controls conditional inclusion matches the value of the same
character constant in the execution character set. Whether such a character
constant can have a negative value (3.8.1).
The preprocessing and execution phases use exactly the same meanings for
character constants.
A single-character character constant is always positive.
• The method for locating includable source files (3.8.2).
For filenames surrounded by < >, the includable source files are searched for in
/usr/include.
The default search list includes /usr/include. You can change this list with
various compiler options. See the cc(1) reference page and the -I and
-nostdinc options.
• The support of quoted names for includable source files (3.8.2).
Quoted names are supported for includable source files. For filenames surrounded
by ‘‘ ’’, the includable source files are searched for in the directory of the
current include file, then in /usr/include.
The default search list includes /usr/include. You can change this list with
various compiler options. See the cc(1) reference page and the -I and
-nostdinc options.
• The mapping of source file character sequences (3.8.2).
The mapping is the identity mapping.
• The behavior on each recognized #pragma directive.

007–0701–150 139
A: Implementation-Defined Behavior

See MIPSpro C and C++ Pragmas on the SGI Tech Pubs Library
(http://techpubs.sgi.com/library) for details on all supported #pragma
directives.
• The definitions for __DATE__ and __TIME__ when, respectively, the date and
time of translation are not available.
The date and time of translation are always available in this implementation.
• What is the maximum nesting depth of include files (3.8.2).
The maximum nesting depth of include files is 200.

Library Functions (F.3.14)

• The null pointer constant to which the macro NULL expands (4.1.5).
The NULL pointer constant expands to an int with value zero. That is,
#define NULL 0

• The diagnostic printed by and the termination behavior of the assert function
(4.2).
If an assertion given by assert(EX) fails, the following message is printed on
stderr using _write to its underlying fileno:
Assertion failed: EX, file <filename>, line <linenumber>

This is followed by a call to abort (which exits with a SIGABRT).

• The sets of characters tested for by the isalnum, isalpha, iscntrl, islower,
isprint, and isupper functions (4.3.1).
The statements in the following list are true when operating in the C locale. The C
locale is in effect at program start up for programs compiled for pure ANSI C
(that is, -ansi), or by invoking setlocale(LC_ALL,’’C’’). The C locale can
be overridden at start up for any program that does not explicitly invoke
setlocale by setting the value of the CHRCLASS environment variable. (See the
ctype(3c) reference page .)
– isalnum is nonzero for the 26 letters a–z, the 26 letters A–Z, and the digits 0–9.
– isalpha is nonzero for the 26 letters a–z and the 26 letters A–Z.

140 007–0701–150
 C Language Reference Manual

– islower is nonzero for the 26 letters a–z.

– isupper is nonzero for the 26 letters A–Z.
– isprint is nonzero for the ASCII characters space through tilde (~) (0x20
through 0x7e).
– iscntrl is nonzero for the ASCII characters NUL through US (0x0 through
0x1f).
• The values returned by the mathematics functions on domain errors (4.5.1).
The value returned by the math functions on domain errors is the default IEEE
Quiet NaN in all cases except the following:
– The functions pow and powf return -HUGE_VAL when the first argument is
zero and the second argument is negative. When both arguments are zero,
pow() and powf() return 1.0.
– The functions atan2 and atan2f return zero when both arguments are zero.
• Whether mathematics functions set the integer expression errno to the value of
the macro RANGE on underflow range errors (4.5.1).
Yes, except intrinsic functions that have been inlined. Note that fabs, fabsf,
sqrt, sqrtf, hypotf, fhypot, pow, and powf are intrinsic by default in -xansi
and -cckr modes and can be made intrinsic in -ansi mode by using the
D__INLINE_INTRINSICS compiler option.
• Whether a domain error occurs or zero is returned when the fmodfunction has a
second argument of zero (4.5.6.4).
fmod(x,0) gives a domain error and returns the default IEEE Quiet NaN.

Signals
• The set of signals for the signal function (4.7.1.1).
The signal set is listed in Table A-4, page 142, which is from the signal(2)
reference page. The set of signals conforms to the SVR4 ABI. Note that some of
the signals are not defined in -ansiposix mode. References in square brackets
beside the signal numbers are described under “Signal Notes” in the discussion of
signal semantics.

007–0701–150 141
A: Implementation-Defined Behavior

Table A-4 Signals

Signal Number[Note] Meaning

SIGHUP 01 Hangup
SIGINT 02 Interrupt
SIGQUIT 03[1] Quit
SIGILL 04[1] Illegal instruction (not reset when
caught)
SIGTRAP 05[1][5] Race trap (not reset when caught)
SIGIOT 06 IOT instruction
SIGABRT 06[1] Abort
SIGEMT 07[1][4] MT instruction
SIGFPE 08[1] Floating point exception
SIGKILL 09 Kill (cannot be caught or ignored)
SIGBUS 10[1] Bus error
SIGSEGV 11[1] Segmentation violation
SIGSYS 12[1] Bad argument to system call
SIGPIPE 13 Write on a pipe with no one to read it
SIGALRM 14 Alarm clock
SIGTERM 15 Software termination signal
SIGUSR1 16 User-defined signal 1
SIGUSR2 17 User-defined signal 2
SIGCLD 18[2] Termination of a child process
®
SIGGHLD 18 4.3 BSD and POSIX name
SIGPWR 19[2] Power fail (not reset when caught)
SIGWINCH 20[2] Window size changes
SIGURG 21[2] Urgent condition on I/O channel
SIGIO 22[2] Input/output possible
SIGPOLL 22[3] Selectable event pending

142 007–0701–150
 C Language Reference Manual

Signal Number[Note] Meaning

SIGSTOP 23[6] Stop (cannot be caught or ignored)
SIGTSTP 24[6] Stop signal generated from keyboard
SIGCONT 25[6] Continue after stop (cannot be ignored)
SIGTTIN 26[6] Background read from control terminal
SIGTTOU 27[6] Background write to control terminal
SIGVTALRM 28 Virtual time alarm
SIGPROF 29 Profiling alarm
SIGXCPU 30 CPU time limit exceeded [see
setrlimit(2)]
SIGXFSZ 31 File size limit exceeded [see
setrlimit(2)]
SIG32 32 Reserved for kernel usage

• The semantics for each signal recognized by the signal function (4.7.1.1).
In the signal invocation signal(sig, func), func can be the address of a
signal handler, handler, or one of the two constant values (defined in
<sys/signal.h>) SIG_DFL or SIG_IGN. The semantics of these values are as
follows:
SIG_DFL Terminate process upon receipt of signal sig. (This
is the default if no call to signal for signal sig
occurs.) Upon receipt of the signal sig, the
receiving process is to be terminated with all of the
consequences outlined in the exit(2) reference
page. See note 1 under "Signal Notes", page 145.
SIG_IGN Ignore signal. The signal sig is to be ignored.
handler Catch signal. func is the address of function
handler.

Note: The signals SIGKILL, SIGSTOP, and SIGCONT cannot be ignored.

If func is the address of handler, upon receipt of the signal sig, the receiving
process is to invoke handler as follows:

007–0701–150 143
A: Implementation-Defined Behavior

handler (int sig, int code, struct sigcontext *sc);

The remaining arguments are supplied as extensions and are optional. The value
of the second argument code is meaningful only in the cases shown in Table A-5,
page 144.

Table A-5 Valid Codes in a Signal-Catching Function

Condition Signal Code

User breakpoint SIGTRAP BRK_USERBP
User breakpoint SIGTRAP BRK_SSTEPBP
Integer overflow SIGTRAP BRK_OVERFLOW
Divide by zero SIGTRAP BRK_DIVZERO
Multiply overflow SIGTRAP BRK_MULOVF
Invalid virtual address SIGSEGV EFAULT
Read-only address SIGSEGV EACCESS
Read beyond mapped object SIGSEGV ENXIO

The third argument, sc, is a pointer to a struct sigcontext (defined in

<sys/signal.h>) that contains the processor context at the time of the signal.
Upon return from handler, the receiving process resumes execution at the point
where it was interrupted.
Before entering the signal-catching function, the value of func for the caught signal
is set to SIG_DFL, unless the signal is SIGILL, SIGTRAP, or SIGPWR. This means
that before exiting the handler, a call to signal is necessary to catch future signals.
Suppose a signal that is to be caught occurs during one of the following routines:
– A read, write, or open
– An ioctl system call on a slow device (like a terminal, but not a file)
– A pause (system call)
– A wait system call that does not return immediately due to the existence of a
previously stopped or zombie process

144 007–0701–150
 C Language Reference Manual

The signal catching function is executed and then the interrupted system call
returns a -1 to the calling process with errno set to EINTR.

Note: The signals SIGKILL and SIGSTOP cannot be caught.

Signal Notes

1. If SIG_DFL is assigned for SIGQUIT, SIGILL, SIGTRAP, SIGABRT, SIGEMT,

SIGFPE, SIGBUS, SIGSEGV, or SIGSYS, in addition to the process being
terminated, a “core image” is constructed in the current working directory of the
process, if the following two conditions are met:
a. The effective user ID and the real user ID of the receiving process are equal.
b. An ordinary file named core exists and is writable or can be created.
If the file must be created, it has the following properties:
• A mode of 0666 modified by the file creation mask (see the umask(2) reference
page)
• A file owner ID that is the same as the effective user ID of the receiving process
• A file group ID that is the same as the effective group ID of the receiving
process

Note: The core file can be truncated if the resultant file size would exceed either
ulimit (see the ulimit(2) reference page) or the process’s maximum core file
size (see the setrlimit(2) reference page).

2. For the signals SIGCLD, SIGWINCH, SIGPWR, SIGURG, and SIGIO, the actions
associated with each of the three possible values for func are as follows:
SIG_DFL Ignore signal. The signal is to be ignored.
SIG_IGN Ignore signal. The signal is to be ignored. Also, if
sig is SIGCLD, the calling process’s child processes
do not create zombie processes when they
terminate (see the exit(2) reference page).
handler Catch signal. If the signal is SIGPWR, SIGURG,
SIGIO, or SIGWINCH, the action to be taken is the
same as that previously described when func is the

007–0701–150 145
A: Implementation-Defined Behavior

address of a function. The same is true if the signal

is SIGCLD with one exception: while the process is
executing the signal-catching function, all
terminating child processes are queued. The wait
system call removes the first entry of the queue. If
the signal system call is used to catch SIGCLD,
the signal handler must be reattached when exiting
the handler, and at that time—if the queue is not
empty—SIGCLD is raised again before signal
returns. (See the wait(2) reference page.)
In addition, SIGCLD affects the wait and exit system calls as follows:
wait If the handler parameter of SIGCLD is set to
SIG_IGN and a wait is executed, the wait blocks
until all of the calling process’s child processes
terminate; it then returns a value of -1 with errno
set to ECHILD.
exit If, in the exiting process’s parent process, the
handler parameter of SIGCLD is set to SIG_IGN, the
exiting process does not create a zombie process.
When processing a pipeline, the shell makes the last process in the pipeline the
parent of the preceding processes. Do not set SIGCLD to be caught for a process
that can be piped into in this manner (and thus become the parent of other
processes).
3. SIGPOLL is issued when a file descriptor corresponding to a STREAMS (see
intro(2)) file has a “selectable” event pending. A process must specifically
request that this signal be sent using the I_SETSIG ioctl call. Otherwise, the
process never receives SIGPOLL.
4. SIGEMT is never generated on an IRIS 4D system.
5. SIGTRAP is generated for breakpoint instructions, overflows, divide by zeros,
range errors, and multiply overflows. The second argument code gives specific
details of the cause of the signal. Possible values are described in
<sys/signal.h>.
6. The signals SIGSTOP, SIGTSTP, SIGTTIN, SIGTTOU, and SIGCONT are used by
command interpreters like the C shell (see the csh(1) reference page) to provide
job control. The first four signals listed stop the receiving process unless the
signal is caught or ignored. SIGCONT resumes a stopped process. SIGTSTP is

146 007–0701–150
 C Language Reference Manual

sent from the terminal driver in response to the SWTCH character being entered
from the keyboard (see the termio(7) reference page. SIGTTIN is sent from the
terminal driver when a background process attempts to read from its controlling
terminal. If SIGTTIN is ignored by the process, then the read returns EIO.
SIGTTOU is sent from the terminal driver when a background process attempts to
write to its controlling terminal when the terminal is in TOSTOP mode. If
SIGTTOU is ignored by the process, then the write succeeds, regardless of the
state of the controlling terminal.
signal does not catch an invalid function argument, func, and results are undefined
when an attempt is made to execute the function at the bad address.
SIGKILL immediately terminates a process, regardless of its state.
Processes stopped via job control (typically CTRL+Z) do not act upon any delivered
signals other than SIGKILL until the job is restarted. Processes blocked via a
blockproc system call unblock if they receive a signal that is fatal (that is, a
non-job-control signal that they are not catching). These processes remained stopped,
however, if the job they are a part of is stopped. Only upon restart do they die. Any
non-fatal signals received by a blocked process do not cause the process to be
unblocked. An unblockproc or unblockprocall system call is necessary.
If an instance of signal sig is pending when signal(sig, func) is executed, the pending
signal is cancelled unless it is SIGKILL.
signal fails if sig is an illegal signal number, including SIGKILL and SIGSTOP, or if
an illegal operation is requested (such as ignoring SIGCONT, which is ignored by
default). In these cases, signal returns SIG_ERR and sets errno to EINVAL.
After a fork, the child inherits all handlers and signal masks. If any signals are
pending for the parent, they are not inherited by the child.
The exec routines reset all caught signals to the default action; ignored signals remain
ignored; the blocked signal mask is unchanged and pending signals remain pending.
The following reference pages contain other relevant information: intro(2),
blockproc(2), kill(2), pause(2), ptrace(2), sigaction(2), sigset(2), wait(2),
setjmp(3c), sigvec, and kill(1).

Diagnostics

Upon successful completion, signal returns the previous value of func for the
specified signal sig. Otherwise, a value of SIG_ERR is returned and errno is set to
indicate the error. SIG_ERR is defined in the <sys/signal.h> header file.

007–0701–150 147
A: Implementation-Defined Behavior

Caution: Signals raised by the instruction stream, SIGILL, SIGEMT, SIGBUS, and
! SIGSEGV, will cause infinite loops if their handler returns, or the action is set to
SIG_IGN. The POSIX signal routines (sigaction, sigpending, sigprocmask,
sigsuspend, sigsetjmp), and the BSD 4.3 signal routines (sigvec, signal,
sigblock, sigpause, sigsetmask) must never be used with signal or sigset.

Before entering the signal-catching function, the value of func for the caught signal is
set to SIG_DFL, unless the signal is SIGILL, SIGTRAP, or SIGPWR. This means that
before exiting the handler, a signal call is necessary to again set the disposition to
catch the signal.
Note that handlers installed by signal execute with no signals blocked, not even the
one that invoked the handler.
• The default handling and the handling at program startup for each signal
recognized by the signal function (4.7.1.1).
Each signal is set to SIG_DFL at program start up.
• If the equivalent of signal (sig, SIG_DFL); is not executed prior to the call of a
signal handler, the blocking of the signal that is performed(4.7.1.1).
The equivalent of signal(sig, SIG_DFL) is executed prior to the call of a signal
handler unless the signal is SIGILL, SIGTRAP, or SIGPWR. See the signal
reference page for information on the support for the BSD 4.3 signal facilities.
• Whether the default handling is reset if the SIGILL signal is received by a
handler specified to the signal function (4.7.1.1).
No.

Streams and Files

• Whether the last line of a text stream requires a terminating newline character
(4.9.2).
There is no requirement that the last line of a text stream have a terminating
newline: the output is flushed when the program terminates, if not earlier (as a
result of fflush call). However, subsequent processes or programs reading the
text stream or file might expect the newline to be present; it customarily is in IRIX
text files.

148 007–0701–150
 C Language Reference Manual

• Whether space characters that are written out to a text stream immediately
before a newline character appear when read in (4.9.2).
All text characters (including spaces before a newline character) written out to a
text stream appear exactly as written when read back in.
• The number of null characters that can be appended to data written to a binary
stream (4.9.2).
The library never appends nulls to data written to a binary stream. Only the
characters written by the application are written to the output stream, whether
binary or text. Text and binary streams are identical: there is no distinction.
• Whether the file position indicator of an append mode stream is initially
positioned at the beginning or end of the file (4.9.2).
The file position indicator of an append stream is initially positioned at the end of
the file.
• Whether a write on a text stream causes the associated file to be truncated
beyond that point (4.9.3).
A write on a text stream does not cause the associated file to be truncated.
• The characteristics of file buffering (4.9.3).
Files are fully buffered, as described in paragraph 3, section 4.9.3, of ANSI
X3.159-1989.
• Whether a zero-length file actually exists (4.9.3).
Zero-length files exist, but have no data, so a read on such a file returns an
immediate EOF.
• The rules for composing valid file names (4.9.3).
Filenames consist of 1 to FILENAME_MAX characters. These characters can be
selected from the set of all character values excluding \0 (null) and the ASCII code
for / (slash).
It is generally unwise to use *, ?, [, or ]as part of filenames because of the special
meaning attached to these characters by the shell (see the sh(1) reference page).
Although permitted, the use of unprintable characters should be avoided.
• Whether the same file can be opened multiple times (4.9.3).

007–0701–150 149
A: Implementation-Defined Behavior

A file can be open any number of times.

• The effect of the remove function on an open file (4.9.4.1).
For local disk files, a remove(1) removes a directory entry pointing to the file but
has no effect on the file or the program with the file open. For files remotely
mounted via NFS software, the effect is unpredictable (the file might be removed
making further I/O impossible through open streams, or it might behave like a
local disk file) and might depend on the version(s) of NFS involved.
• The effect if a file with the new name exists prior to a call to the rename
function (4.9.4.2).
If the new name exists, the file with that new name is removed (See the rm(1)
reference page) before the rename is done.
• The output for %p conversion in the fprintf function (4.9.6.1).
%p is treated the same as %x.
• The input for %p conversion in the fscanf function (4.9.6.2).
%p is treated the same as %x.
• The interpretation of a – character that is neither the first nor the last character
in the scanlist for %[ conversion in the fscanf function (4.9.6.2).
A – character that does not fit the pattern mentioned above is used as a shorthand
for ranges of characters. For example, [xabcdefgh] and [xa-h] mean that characters
a through h and the character x are in the range (called a scanset in 4.9.6.2).

Temporary Files
• Whether a temporary file is removed if a program terminates abnormally
(4.9.4.3).
Temporary files are removed if a program terminates abnormally.

errno and perror

• The value to which the macro errno is set by the fgetpos or ftell function on
failure (4.9.9.1, 4.9.9.4).

150 007–0701–150
 C Language Reference Manual

errno is set to EBADF (9) by the fgetpos or ftell function on failure.

• The messages generated by the perror function (4.9.10.4).
The message generated is simply a string. The content of the message given for
each legal value of errno is given in the list below, which is of the format
errno_value:message.
1: No permission match (-o32 mode) 1: Not privileged (-n32 and -64 modes)
2: No such file or directory
3: No such process
4: Interrupted system call
5: I/O error
6: No such device or address
7: Arg list too long
8: Exec format error
9: Bad file number

10: No child processes

11: Resource temporarily unavailable
12: Not enough space
13: Permission denied
14: Bad address
15: Block device required
16: Device or resource busy (-o32 mode) 16: Device busy (-n32 and -64 modes)
17: File exists
18: Cross-device link
19: No such device
20: Not a directory
21: Is a directory

007–0701–150 151
A: Implementation-Defined Behavior

22: Invalid argument

23: Too many open files in system (-o32 mode) 23: File table overflow (-n32 and
-64 modes)
24: Too many open files in a process (-o32 mode) 24: Too many open files (-n32
and -64 modes)
25: Inappropriate IOCTL operation (-o32 mode) 25: Not a typewriter (-n32 and
-64 modes)
26: Text file busy
27: File too large
28: No space left on device
29: Illegal seek
30: Read-only filesystem
31: Too many links
32: Broken pipe
33: Argument out of domain
34: Result too large
35: No message of desired type
36: Identifier removed
37: Channel number out of range
38: Level 2 not synchronized
39: Level 3 halted
40: Level 3 reset
41: Link number out of range
42: Protocol driver not attached

43: No CSI structure available

44: Level 2 halted

152 007–0701–150
 C Language Reference Manual

45: Deadlock situation detected/avoided

46: No record locks available
47: Error 47
48: Error 48
49: Error 49
50: Bad exchange descriptor
51: Bad request descriptor
52: Message tables full
53: Anode table overflow
54: Bad request code
55: Invalid slot
56: File locking deadlock
57: Bad font file format
58: Error 58
59: Error 59
60: Not a stream device
61: No data available
62: Timer expired
63: Out of stream resources
64: Machine is not on the network
65: Package not installed
66: Object is remote
67: Link has been severed
68: Advertise error
69: Srmount error

007–0701–150 153
A: Implementation-Defined Behavior

70: Communication error on send

71: Protocol error
72: Error 72
73: Error 73
74: Multihop attempted
75: Error 75
76: Error 76
77: Not a data message
78: Error 78 (-o32 mode) 78: Filename too long (-n32 and -64 modes)
79: Error 79 (-o32 mode) 79: Value too large for defined data type (-n32 and -64
modes)
80: Name not unique on network
81: File descriptor in bad state
82: Remote address changed
83: Cannot access a needed shared library
84: Accessing a corrupted shared library
85: .lib section in a.out corrupted
86: Attempting to link in more shared libraries than system limit
87: Cannot exec a shared library directly
88: Invalid System Call (-o32 mode) 88: Illegal byte sequence (-n32 and -64
modes)
89: Error 89 (-o32 mode) 89: Operation not applicable (-n32 and -64 modes)
90: Error 90 (-o32 mode) 90: Too many symbolic links in pathname traversal
(-n32 and -64 modes)

91: Error 91 (-o32 mode) 91: Restartable system call (-n32 and -64 modes)

154 007–0701–150
 C Language Reference Manual

92: Error 92 (-o32 mode) 92: If pipe/FIFO, don’t sleep in stream head (-n32 and
-64 modes)
93: Error 93 (-o32 mode) 93: Directory not empty (-n32 and -64 modes)
94: Error 94 (-o32 mode) 94: Too many users (-n32 and -64 modes)
95: Error 95 (-o32 mode) 95: Socket operation on non-socket (-n32 and -64
modes)
96: Error 96 (-o32 mode) 96: Destination address required (-n32 and -64 modes)
97: Error 97 (-o32 mode) 97: Message too long (-n32 and -64 modes)
98: Error 98 (-o32 mode) 98: Protocol wrong type for socket (-n32 and -64
modes)
99: Error 99 (-o32 mode) 99: Option not supported by protocol (-n32 and -64
modes)
100: Error 100
101: Operation would block (-o32 mode) 101: Error 101 (-n32 and -64 modes)
102: Operation now in progress (-o32 mode) 102: Error 102 (-n32 and -64
modes)
103: Operation already in progress (-o32 mode) 103: Error 103 (-n32 and -64
modes)
104: Socket operation on non-socket (-o32 mode) 104: Error 104 (-n32 and -64
modes)
105: Destination address required (-o32 mode) 105: Error 105 (-n32 and -64
modes)
106: Message too long (-o32 mode) 106: Error 106 (-n32 and -64 modes)
107: Protocol wrong type for socket (-o32 mode) 107: Error 107 (-n32 and -64
modes)
108: Option not supported by protocol (-o32 mode) 108: Error 108 (-n32 and
-64 modes)
109: Protocol not supported (-o32 mode) 109: Error 109 (-n32 and -64 modes)
110: Socket type not supported (-o32 mode) 110: Error 110 (-n32 and -64 modes)

007–0701–150 155
A: Implementation-Defined Behavior

111: Operation not supported on socket (-o32 mode) 111: Error 111 (-n32 and
-64 modes)
112: Protocol family not supported (-o32 mode) 112: Error 112 (-n32 and -64
modes)
113: Address family not supported by protocol family (-o32 mode) 113: Error 113
(-n32 and -64 modes)
114: Address already in use (-o32 mode) 114: Error 114 (-n32 and -64 modes)
115: Can’t assign requested address (-o32 mode) 115: Error 115 (-n32 and -64
modes)
116: Network is down (-o32 mode) 116: Error 116 (-n32 and -64 modes)
117: Network is unreachable (-o32 mode) 117: Error 117 (-n32 and -64 modes)
118: Network dropped connection on reset (-o32 mode) 118: Error 118 (-n32 and
-64 modes)
119: Software caused connection abort (-o32 mode) 119: Error 119 (-n32 and -64
modes)
120: Connection reset by peer (-o32 mode) 120: Protocol not supported (-n32
and -64 modes)
121: No buffer space available (-o32 mode) 121: Socket type not supported (-n32
and -64 modes)
122: Socket is already connected (-o32 mode) 122: Operation not supported on
transport endpoint (-n32 and -64 modes)
123: Socket is not connected (-o32 mode) 123: Protocol family not supported
(-n32 and -64 modes)
124: Can’t send after socket shutdown (-o32 mode) 124: Address family not
supported by protocol family (-n32 and -64 modes)
125: Too many references: can’t splice (-o32 mode) 125: Address already in use
(-n32 and -64 modes)
126: Connection timed out (-o32 mode) 126: Cannot assign requested address
(-n32 and -64 modes)

156 007–0701–150
 C Language Reference Manual

127: Connection refused (-o32 mode) 127: Network is down (-n32 and -64
modes)
128: Host is down (-o32 mode) 128: Network is unreachable (-n32 and -64
modes)
129: Host is unreachable (-o32 mode) 129: Network dropped connection because
of reset (-n32 and -64 modes)
130: Too many levels of symbolic links (-o32 mode) 130: Software caused
connection abort (-n32 and -64 modes)
131: Filename too long (-o32 mode) 131: Connection reset by peer (-n32 and -64
modes)
132: Directory not empty (-o32 mode) 132: No buffer space available (-n32 and
-64 modes)
133: Disk quota exceeded (-o32 mode) 133: Transport endpoint is already
connected (-n32 and -64 modes)
134: Stale NFS 14 file handle (-o32 mode) 134: Transport endpoint is not
®

connected (-n32 and -64 modes)

135: Structure needs cleaning (-n32 and -64 modes)
136: Error 136 (-n32 and -64 modes)
137: Not a name file (-n32 and -64 modes)
138: Not available (-n32 and -64 modes)
139: Is a name file (-n32 and -64 modes)
140: Remote I/O error (-n32 and -64 modes)
141: Reserved for future use (-n32 and -64 modes)
142: Error 142 (-n32 and -64 modes)
143: Cannot send after socket shutdown (-n32 and -64 modes)
144: Too many references: cannot splice (-n32 and -64 modes)

145: Connection timed out (-n32 and -64 modes)

146: Connection refused (-n32 and -64 modes)

007–0701–150 157
A: Implementation-Defined Behavior

147: Host is down (-n32 and -64 modes)

148: No route to host (-n32 and -64 modes)
149: Operation already in progress (-n32 and -64 modes)
150: Operation now in progress (-n32 and -64 modes)
151: Stale NFS file handle (-n32 and -64 modes)
See the perror(3c) reference page for further information.

Memory Allocation
• The behavior of the calloc, malloc, or realloc function if the size requested
is zero (4.10.3).
The malloc in libc.a returns a pointer to a zero-length space if a size of zero is
requested. Successive calls to malloc return different zero-length pointers. If the
library libmalloc.a is used, malloc returns 0 (the NULL pointer).

abort Function
• The behavior of the abort function with regard to open and temporary files
(4.10.4.1).
Open files are not flushed, but are closed. Temporary files are removed.

exit Function
• The status returned by the exit function if the value of the argument is other
than zero, EXIT_SUCCESS or EXIT_FAILURE (4.10.4.3).
The status returned to the environment is the least significant eight bits of the
value passed to exit.

getenv Function
• The set of environment names and the method for altering the environment list
used by the getenv function (4.10.4.4).

158 007–0701–150
 C Language Reference Manual

Any string can be used as the name of an environment variable, and any string
can be used for its value. The function putenv alters the environment list of the
application. For example,
putenv(‘‘MYNAME=foo’’)

This sets the value of the environment variable MYNAME to “foo.” If the
environment variable MYNAME already existed, its value is changed. If it did not
exist, it is added. The string passed to putenv actually becomes part of the
environment, and changing it later alters the environment. Further, the string
should not be space that was automatically allocated (for example, an auto array);
rather, it should be space that is either global or malloced. For more information,
see the putenv(3c) reference page.
It is not wise to alter the value of well-known environment variables. For the
current list, see the environ(5) reference page.

system Function
• The contents and mode of execution of the string passed to the system function
(4.10.4.5).
The contents of the string should be a command string, as if typed to a normal
IRIX shell, such as sh(1). A shell (sh) is forked, and the string is passed to it.
The current process waits until the shell has completed and returns the exit status
of the shell as the return value.

strerror Function
• The contents of the error message strings returned by the strerror function
(4.11.6.2).
The string is exactly the same as the string output by perror, which is
documented in "errno and perror", page 150.

Time Zones and the clock Function

• The local time zone and daylight saving time (4.12.1).
Local time and daylight saving time are determined by the value of the TZ
environment variable. TZ is set by init to the default value indicated in the file

007–0701–150 159
A: Implementation-Defined Behavior

/etc/TIMEZONE, and this value is inherited in the environment of all processes.

If TZ is unset, the local time zone defaults to GMT (Greenwich mean time, or
coordinated universal time), and daylight saving time is not in effect. See the
reference pages ctime(3c), time(2), timezone(4), environ(5), getenv(3), and
other related reference pages for the format of TZ.
• The era for the clock function (4.12.2.1).
clock counts seconds from 00:00:00: GMT, January 1, 1970. What was once
known as Greenwich mean time (GMT) is now known as coordinated universal
time, though the reference pages do not reflect this change yet. See the ctime(3c)
reference page for further information.

Locale-Specific Behavior (F.4)

For information on locale-specific behavior, see the chapter titled “Internationalizing
Your Application” in Topics in IRIX Programming. That chapter covers some
locale-specific topics to consider when internationalizing an application. Topics
include
• Overview of Locale-Specific Behavior
• Native Language Support and the NLS Database
• Using Regular Expressions
• Cultural Data
Also, that chapter describes setting a locale, location of locale-specific data, cultural
items to consider, and GUI concerns.
For additional information on locale-specific behavior, refer to the X/Open Portability
Guide, Volume 3, “XSI Supplementary Definitions,” published by Prentice Hall,
Englewood Cliffs, New Jersey 07632, ISBN 0-13-685-850-3.

Common Extensions (F.5)

The following extensions are widely used in many systems, but are not portable to all
implementations. The inclusion of any extension that can cause a strictly conforming
program to become invalid renders an implementation nonconforming. Examples of
such extensions are new keywords, or library functions declared in standard headers

160 007–0701–150
 C Language Reference Manual

or predefined macros with names that do not begin with an underscore. The
Standard’s description of each extension is followed by a definition of any SGI
support/nonsupport of each common extension.

Environment Arguments (F.5.1)

• In a hosted environment, the main function receives a third argument, char
*envp[], that points to a null-terminated array of pointers to char. Each of these
pointers points to a string that provides information about the environment for
this execution of the process (2.1.2.1.1).
This extension is supported.

Specialized Identifiers
• Characters other than the underscore _, letters, and digits, that are not defined in
the required source character set (such as dollar sign $, or characters in national
character sets) can appear in an identifier.
If the -dollar option is given to cc, then the dollar sign ($) is allowed in
identifiers.

Lengths and Cases of Identifiers

• All characters in identifiers (with or without external linkage) are significant
and case distinctions are observed (3.1.2).
All characters are significant. Case distinctions are observed.

Scopes of Identifiers (F.5.4)

• A function identifier, or the identifier of an object (the declaration of which
contains the keyword extern) has file scope.
This is true of the compiler when invoked with cc -cckr (that is, when
requesting traditional C). When compiling in ANSI mode (by default or with one
of the ANSI options) function identifiers (and all other identifiers) have block
scope when declared at block level.

007–0701–150 161
A: Implementation-Defined Behavior

Writable String Literals (F.5.5)

• String literals are modifiable. Identical string literals shall be distinct (3.1.4).
All string literals are distinct and writable when the -use_readwrite_const
option is in effect. Otherwise, string literals may not be writable.

Other Arithmetic Types (F.5.6)

• Other arithmetic types, such as long long int and their appropriate
conversions, are defined (3.2.2.1).
Yes.

Function Pointer Casts (F.5.7)

• A pointer to an object or to void can be cast to a pointer to a function, allowing
data to be invoked as a function (3.3.4). A pointer to a function can be cast to a
pointer to an object, or to void, allowing a function to be inspected or modified
(for example, by a debugger) (3.3.4).
Function pointers can be cast to a pointer to an object, or to void, and vice versa.
Data can be invoked as a function.
Casting a pointer to a function to a pointer to an object or void does allow a
function to be inspected. Normally, functions cannot be written to, because text
space is read-only. Dynamically loaded functions are loaded (by a user program)
into data space and can be written to.

Non-int Bit-Field Types (F.5.8)

• Types other than int, unsigned int, and signed int can be declared as
bitfields, with appropriate maximum widths (3.5.2.1).
A bitfield can be any integral type in -xansi and -cckr modes. However,
bitfields of types other than int, signed int, and unsigned int result in a
warning diagnostic in -ansi mode.

162 007–0701–150
 C Language Reference Manual

fortran Keyword (F.5.9)

• The fortran declaration specifier can be used in a function declaration to
indicate that calls suitable for Fortran should be generated, or that different
representations for external names are to be generated (3.5.4.3).
The fortran keyword is not supported in this ANSI C. With cc -cckr, that
keyword is accepted but ignored.

asm Keyword (F.5.10)

• The asm keyword can be used to insert assembly language code directly into
the translator output. The most common implementation is via statement of the
form asm (character-string-literal) (3.6).
The asm keyword is not supported.

Multiple External Definitions (F.5.11)

• There can be more than one external definition for the identifier of an object,
with or without the explicit use of the keyword extern. If the definitions
disagree, or more than one is initialized, the behavior is undefined (3.7.2).
With ANSI C, only one external definition of the object is permitted. If more than
one is present, the linker (ld(1)) gives a warning message. The Strict Ref/Def
model is followed (ANSI C Rationale, 3.1.2.2, page 23).
With cc -cckr, the Relaxed Ref/Def model is followed (ANSI C Rationale,
3.1.2.2, page 23): multiple definitions of the same identifier of an object in different
files are accepted and all but one of the definitions are treated (silently) as if they
had the extern keyword.
If the definitions in different source units disagree, the mismatch is not currently
detected by the linker (ld), and the resulting program will probably not work
correctly.

Empty Macro Arguments (F.5.12)

• A macro argument can consist of no preprocessing tokens (3.8.3).
This extension is supported. For example, one could define a macro such as

007–0701–150 163
A: Implementation-Defined Behavior

#define notokargs() macrovalue

Predefined Macro Names (F.5.13)

• Macro names that do not begin with an underscore, describing the translation
and execution environments, may be defined by the implementation before
translation begins (3.8.8).
This is not true for cc -ansi, which defines ANSI C. Only macro names
beginning with two underscores or a single underscore followed by a capital letter
are predefined by the implementation before translation begins. The name space is
not polluted.
With cc -cckr (traditional C), a C preprocessor is used with a full set of the
predefined symbols. For example, sgi is predefined.
With cc -xansi (which is the default for cc), an ANSI C preprocessor and
compiler are used and a full set of predefined symbols is defined (including sgi,
for example).

Extra Arguments for Signal Handlers (F.5.14)

• Handlers for specific signals can be called with extra arguments in addition to
the signal number.
SGI supports System V, POSIX, and BSD signal handlers. Extra arguments to the
handler are available for your use. See the signal reference page.

Additional Stream Types and File-Opening Modes (F.5.15)

• Additional mappings from files to streams may be supported (4.9.2), and
additional file-opening modes may be specified by characters appended to the
mode argument of the fopen function (4.9.5.3).
There are no additional modes supported. There are no additional mappings. The
UNIX approach is used, as mentioned in the ANSI C Rationale, Section 4.9.2,
page 90.

164 007–0701–150
 C Language Reference Manual

Defined File Position Indicator (F.5.16)

• The file position indicator is decremented by each successful call to the ungetc
function for a text stream, except if its value was zero before a call (4.9.7.11).
The SGI C compiler supports only the one character of pushback guaranteed by
the standard.

007–0701–150 165
 Appendix B

lint-style Comments

The following table lists the lint-style comments available with the SGI C compiler,
along with a short description. See the lint(1) reference page for more details.
The preprocessor automatically strips out comments. This prevents the lint-style
comments from being seen by the rest of the compiler and therefore these comments
will not work from within macros.
To work around this, turn off the offending message locally by using set woff
pragmas, as in this example:
#define MY_DEFS() \
\
int X; \
int Y; \
int Z;

void
func2(void)
{

#pragma set woff 1174

MY_DEFS();
#pragma reset woff 1174

X = 2;
Z = 4;

junk (X,X,Z);
}

In this example, 1174 refers only to the message for unreferenced variables. The
other link-style suppression will need different message numbers.

007–0701–150 167
B: lint-style Comments

Table B-1 lint–style Comments

Comment Short Description

/*PRINTFLIKEn*/ Applies lint-style check to the first (n-1) arguments as
usual. The nth argument is interpreted as a printf format
string that is used to check the remaining arguments.
/*SCANFLIKEn*/ Applies lint-style check to the first (n-1) arguments as
usual. The nth argument is interpreted as a scanf format
string that is used to check the remaining arguments.
/*ARGSUSEDn*/ Applies lint-style check to only the first n arguments for
usage; a missing n is taken to be 0 (this option acts like the
-v option for the next function).
/*VARARGSn*/ Suppresses the usual checking for variable numbers of
arguments in the following function declaration. The data
types of the first n arguments are checked; a missing n is
taken to be 0. The use of the ellipsis terminator (...) in the
definition is suggested in new or updated code.
/*NOTREACHED*/ Stops comments about unreachable code when placed at
appropriate points. (This comment is typically placed just
after calls to functions like exit).
/*REFERENCED*/ Tells the compiler that the variable defined after comment is
referenced.

168 007–0701–150
 Appendix C

Built-in Functions

The following table lists the built-in functions available in the SGI C compiler, along
with a short description.

Table C-1 Built-in Functions

Intrinsic Short Description

void *__builtin_alloca(unsigned size) Returns a pointer to a
specified number of bytes of
uninitialized local stack space.
float __builtin_fsqrt(float x) Computes the non-negative
square root of a floating point
argument.
double __builtin_sqrt(double x) Computes the non-negative
square root of a double
argument.
float __builtin_fabs(float x) Computes the absolute value
of a float argument.
double __builtin_dabs(double x) Computes the absolute value
of a double argument.

int __builtin_cast_f2i(float x) Treats float as int.

float __builtin_cast_i2f(int x) Treats int as float.
long long __builtin_cast_d2ll(double x) Treats double as long long.
double __builtin_cast_ll2d(long long x) Treats long long as double.
int __builtin_copy_dhi2i(double x) Copies high part of double to
int.
double __builtin_copy_i2dhi(int x) Copies int to high part of
double.
int __builtin_copy_dlo2i(double x) Copies low part of double to
int.

007–0701–150 169
C: Built-in Functions

Intrinsic Short Description

double __builtin_copy_i2dlo(int x, Copies int to low part of
double y) double.

<type> __high_multiply (<type>, <type> ) Multiplies two parameters as

32 (or 64) bit integers and
returns the upper 32 (or 64)
bits of a 64 (or 128) bit result.
<type> can be signed or
unsigned, int, long, or
long long.

170 007–0701–150
 Index

! operator, 56 effect on temporary files, 158

!= operator, 61 acpp
% operator, 58 changes, 7
& operator, 56, 62 Additive operators
fields and, 76 pointers and, 60
&& operator, 63 Address constant, 67
* operator, 58 Address-of operator, 56
+ operator, 56, 59 fields and, 76
++ operator, 57 AND operator
+= operator, 66 bitwise, 62
- - operator, 55 logical, 63
- character ANSI C
in fscanf function, 150 allocating storage, 13
- operator, 56, 59 conversion rules, 45, 46
-= operator, 66 disambiguating identifiers, 10
/ operator, 58 floating point, 43
< operator, 60 fucntion prototype scope, 11
<< operator, 60 function prototype error, 19
<= operator, 60 guidelines, 2
= operator, 65 identifiers, 31
== operator, 61 libraries, 2
> operator, 60 linkage discrepancies, 36
>= operator, 60 linker
>> operator, 60 warnings, 13
? operator, 64 lint, 3
^ operator, 62 macro replacement, 8
| operator, 63 name space, 12
|| operator, 63 name space discrepancies, 33
~ operator, 56 name spaces, 2, 33
32-bit mode preprocessor, 7
type differences, 39 scoping differences, 10
64-bit mode, 58 strictly conforming programs, 1
string literals, 8
switches, 2
A trigraph sequences, 27
value preserving integer promotion, 43
abort function warnings, 3

007–0701–150 171
Index

ANSI C standard header files, 22 Atomic op-and-fetch operations, 125

-ansi compiler option Atomic synchronize operation, 127
external names and, 21 auto, 70
macros, 8 Auto keyword, 71
string literals, 8 Auto storage class, 71
tokens, 9 Autoblocking, 118
Append mode stream Automatic storage duration, 37
initial file position, 149
Application Program Interface
See "API", 104 B
Argument promotions, 53
Argument type promotions barrier, 117
changes, 17 Barrier function, 117
Arguments Behavior
passing, 53 locale-specific, 160
Arithmetic constant expressions, 67 Binary streams
Arithmetic conversions, 45 null characters in, 149
Arithmetic expressions, 14 Bitfield
Arithmetic types, 40 diagnostics, 162
Arithmetic value integral type, 162
64-bit mode, 58 Bitfields, 75, 136
Array integer types, 4
type required to hold maximum size, 136 order of allocation, 138
Array declarators, 81 signedness of, 137
Arrays spanning unit boundary, 76
variable length, 81 straddling int boundaries, 138
asm keyword, 163 Bits
Assert, 140 bitfields, 76
diagnostic, 140 Bits per character, 132
Assignment operators, 65 Bitwise and operator, 62
+=, 66 Bitwise not operator, 56
-=, 66 Bitwise operations
=, 65 signed integers, 134
Associativity Bitwise OR operator
examples, 49 inclusive, 63
atan2, 141 Bitwise or operator
atan2f, 141 exclusive, 62
Atomic compare–and–swap operation, 126 Blanks, 23
Atomic fetch-and-op operations, 124 Block scope
Atomic lock and unlock operations, 127 definition, 32
Atomic lock-release operation, 127 Block statements, 93
Atomic lock-test-and-set operation, 127

172 007–0701–150
 C Language Reference Manual

Blocking Case values

automatic, 116 maximum number of, 139
Blocking slave threads, 115 Cast operators, 58
blockproc, 116 Casting
break statements, 98 pointer to a function, 162
Built-in functions, 169 -cckr compiler option, 8
__builtin_alloca, 169 external names and, 21
__builtin_cast_d2ll, 169 tokens, 9
__builtin_cast_dhi2i, 169 cConversions
__builtin_cast_dlo2i, 169 void, 47
__builtin_cast_f2i, 169 char, 38
__builtin_cast_i2dhi, 169 default sign, 133
__builtin_cast_i2dlo, 170 unsigned vs. "plain", 133
__builtin_cast_i2f, 169 Character
__builtin_cast_ll2d, 169 white space, 129
__builtin_dabs, 169 Character constant, 139
__builtin_fabs, 169 Character constants, 25
__builtin_fsqrt, 169 wide, 25
__builtin_sqrt, 169 Character set, 131
__high_multiply, 170 Character types, 38
__builtin_alloca, 169 Characters, 131
__builtin_cast_d2ll, 169 conversions to integer, 43
__builtin_cast_dhi2i, 169 integer constants, 132
__builtin_cast_dlo2i, 169 multibyte, 38, 131, 133
__builtin_cast_f2i, 169 nongraphic, 25
__builtin_cast_i2dhi, 169 number of bits, 132
__builtin_cast_i2dlo, 170 shift states, 131
__builtin_cast_i2f, 169 source set vs. execution set, 132
__builtin_cast_ll2d, 169 special, 25
__builtin_dabs, 169 wide, 133
__builtin_fabs, 169 initialization, 90
__builtin_fsqrt, 169 CHRCLASS environment variable, 140
__builtin_sqrt, 169 CHUNK, 119
clock function, 160
Coding rules
C #pragma directives, 105
Comma operator, 66
calloc, 158 Comments, 23
Case distinctions -common compiler option, 13
in identifiers, 131 Communication
case label, 95 between processors, 120
case labels, 99

007–0701–150 173
Index

compare–and–swap operation, 126 pointers, 48

Compatible types, 17 rules
Compilation, 2 ANSI C, 46
Compilation mode Traditional C, 46
effect on names, 21 cpp
Compiler restrictions, 111 changes, 7
exception handling, 112
on #pragma pfor, 111
scoping, 113 D
Compound assignment, 66
Compound statements, 93 Data area names changes, 21
scope of declarations, 94 Date
Conditional operator, 64 availability, 140
const object, 4 __DATE__, 140
const type qualifier Daylight saving time, 159
qualifiers Declarations
const, 77 as definitions, 70
Constant expression, 139 enumerations, 76
arithmetic, 67 implicit, 87
Constant expressions, 24, 66 multiple, 69
address constant, 67 structure, 73
integral, 67 union, 73
Constants, 51 Declarators
character, 25 array, 81
enumeration, 28 definition, 78
floating, 27 maximum number of, 139
integer, 24 pointer, 79
long double precision, 43 restrictions, 85
types of, 24 syntax, 78
wide character, 25 Decrement operator, 57
continue statements, 96, 97 Default argument promotions, 53
Controlling expression Default labels, 95, 99
definition, 94 Definition
Conversions, 43 declaration, 70
arithmetic, 45 Definitions
character, 43 external, 101
floating-point, 43 Denoting a bitfield, 57
function designators, 47 Derived types, 40
integer, 44 Device
promotions, 45 interactive, 131
lvalues, 47 Diagnostics
pointer, 44

174 007–0701–150
 C Language Reference Manual

classes, 129, 130 type of int used, 138

control, 129 Enumeration variables, 76
identification errors, 129 Enumerations, 136
return codes, 129 Environment
Directives altering, 159
multiprocessing, 104 names, 159
OpenMP, 104 variables, 159
#pragma Environment variables
coding rules, 105 CHRCLASS, 140
#pragma critical, 106 CHUNK, 119
#pragma enter gate, 106 gang scheduling, 120
#pragma exit gate, 106 MP_BLOCKTIME, 118
#pragma independent, 107 MP_SCHEDTYPE, 119
#pragma parallel, 107 MP_SET_NUMTHREADS, 118
#pragma pfor, 107 MP_SETUP, 118
#pragma synchronize, 106 MP_SLAVE_STACKSIZE, 120
preprocessing, 139 MP_SUGNUMTHD, 119
Disambiguating identifiers, 10 MP_SUGNUMTHD_MAX, 119
Disambiguating names, 31 MP_SUGNUMTHD_MIN, 119
Division MP_SUGNUMTHD_VERBOSE, 119
integer, 58 MPC_GANG, 120
sign of remainder, 134 Environments, 130
Division by zero, 58, 67 freestanding, 130
do statements, 96 Equality operators, 61
Domain errors ERANGE macro, 141
return values, 141 errno, 141
Double, 134 errno macro, 151
representation of, 134 Escape sequences, 131
double, 39 hexadecimal, 26
Double precision, 43 Exception handling, 67
Exception handling restrictions, 112
Exclusive or operator, 62
E exit function, 158
Expression statements, 93
else statements, 94 Expressions
–ansi switch, 2 ++, 55
–xansi switch, 2 - -, 55
enum, 74 constant, 66
changes, 12 postfix, 51
Enumeration constants, 28, 40, 76 function calls, 52
changes, 12 structure references, 54
Enumeration types

007–0701–150 175
Index

subscripts, 52 __fetch_and_xor, 124

union references, 54 fgetpos function
primary, 51 errno on failure, 151
Extensions, 2, 160 File buffering, 149
arithmetic types, 162 File position indicator
asm keyword, 163 initial position, 149
bitfield types, 162 filenames, 149
defined file position indicator, 165 Files
empty macro arguments, 163 renaming, 150
environment arguments, 161 temporary, 150, 158
external definitions, 163 zero-length, 149
fortran keyword, 163 files
function pointer casts, 162 opening multiple times, 150
identifiers, 161 remove on an open file, 150
predefined macro names, 164 valid names, 149
scopes, 161 Float
signal handlers, 164 representation of, 134
specialized identifiers, 161 float, 39
stream types, 164 float variables, 14
writable string literals, 162 -float compiler option, 43
extern, 70 effect on conversions, 43
definitions, 13 type promotions, 14
function definitions, 101 floating constants
External definitions, 101 definition, 27
External function definitions, 101 Floating point constants, 15
External linkage, 34 Floating point conversions, 17
External names, 20 Floating types, 40
compiler options and, 21 Floating-point
External object definitions, 102 conversions, 43
exception handling, 67
sizes, 39
F types, 134
fmod, 141
fetch-and-op operations for statements, 96
__fetch_and_add, 124 fortran keyword, 163
__fetch_and_and, 124 fprintf, 150
__fetch_and_max, 124 fscanf, 150
__fetch_and_min, 124 ftell function
__fetch_and_mpy, 124 errno on failure, 151
__fetch_and_nand, 124 -fullwarn compiler option
__fetch_and_or, 124 scope, 12
__fetch_and_sub, 124

176 007–0701–150
 C Language Reference Manual

Function definitions, 101 Hints, 3

Function names
changes, 20
Function pointer casts, 162 I
Function prototype scope, 11, 32
Function prototypes, 18 -I compiler option, 139
incompatible types, 19 Identifiers, 23, 131
inconsistent, 20 case distinctions, 131
Function scope, 11, 32 definition, 23
Functions disambiguating, 10
built-in linkage, 12, 34
See "Built-in functions", 169 scope, 10
calls, 52 IEEE
declarators, 82 floating point, 44
designators if statements, 94
conversions, 47 iIntegers
external sizes, 133
definition, 101 Implicit declarations, 87
mixed use, 18 Include files, 139
non-void, 4 maximum nesting depth, 140
nonprototyped, 54 quoted names, 139
prototyped, 53 Inclusive or operator, 63
prototypes, 82, 85 Incompatible types
storage-class specifiers, 101 function prototypes and, 19
type, 102 Increment operator, 57
Indirect references, 54
Indirection operator, 56
G init-declarator-list
definition, 69
Gang scheduling, 120 Initialization, 88
getenv function, 159 and storage duration, 70
goto statements, 97 examples, 90
structs, 89
unions, 89
H initialization
aggregates, 89
Header files int, 39
changes, 22 pointer conversion, 58
Headers Integer
standard, 22 conversions to character, 43
Hexadecimal escape sequences, 26 divide-by-zero, 67
__high_multiply, 170

007–0701–150 177
Index

sizes, 39 list of, 23

Integer character constants, 132
Integer constants, 24
Integer division, 58 L
sign of remainder, 134
Integers Labeled statements, 99
bitwise operations, 134 Labels
conversions, 134 case, 95
exception conditions, 67 default, 95
pointers, 44 name spaces, 99
ranges, 133 libmalloc.a, 158
representations, 133 Libraries
unsigned C, 2
conversions, 44 Math, 2
Integral constant expressions, 67 shared, 3
Integral promotions, 45, 46 Library functions, 140
Integral types, 40 prototypes, 22
Interactive device, 131 Linkage, 70
Internal linkage, 34 determination of, 34
Intrinsics discrepancies, 36
example, 128 external, 34
memory barrier, 124 identifiers, 34
synchronization, 123 internal, 34
isalnum, 140 none, 34
isalpha, 140 Linker-defined names, 20
iscntrl, 140 lint, 3
islower, 140 lint-style comments, 167
isprint, 140 /*ARGUSED*/, 168
isupper, 140 /*NOTREACHED*/, 168
Iteration statements, 95 /*PRINTFLIKE*/, 168
controlling expression, 95 /*REFERENCED*/, 168
flow of control, 95 /*SCANF like*/, 168
/*VARARGS*/, 168
Literals, 28
J Local time, 159
Locale-specific behavior, 160
Jump statements, 97 lock and unlock operations, 127
Lock example, 128
lock-release operation, 127
K lock-test-and-set operation, 127
Logical operators
Keywords

178 007–0701–150
 C Language Reference Manual

AND, 63 mp_block, 115

OR, 63 MP_BLOCKTIME, 118
long, 39 mp_blocktime, 116
long double, 27, 39, 134 mp_create, 115
long double precision, 43 mp_destroy, 115
long long, 39 mp_my_threadnum, 117
LONGLONG macro, 133 mp_numthreads, 116
lvalue MP_SCHEDTYPE, 119
conversions, 47 MP_SET_NUMTHREADS, 118
definition, 41 mp_set_numthreads, 116
lvalues, 57 mp_set_slave_stacksize, 117
mp_setlock, 117
MP_SETUP, 118
M mp_setup, 115, 117
mp_shmem, 120
Macros MP_SLAVE_STACKSIZE, 120
in -ansi mode, 8 mp_suggested_numthreads, 116
in -cckr mode, 8 MP_SUGNUMTHD, 119
LONGLONG, 133 MP_SUGNUMTHD_MAX, 119
main MP_SUGNUMTHD_MIN, 119
arguments to, 130 MP_SUGNUMTHD_VERBOSE, 119
malloc, 158 mp_unblock, 115
Mapping mp_unsetlock, 117
character sequences, 139 MPC_GANG, 120
Mathematics functions Multibyte characters, 38, 131, 133
domain errors, 141 Multiple definition messages, 13
underflow range errors, 141 Multiplicative operators, 58
Memory Multiprocessing directives, 104
acquire barrier, 124
allocation, 158
full barrier, 124 N
release barrier, 124
Memory barrier, 124 Name
Message passing, 120 definition, 23
Message Passing Tookit (MPT), 120 Name spaces, 2, 33
Messages changes, 12
diagnostic, 129 discrepancies, 33
error, 129 labels, 99
multiple definition, 13 Names
minus, 57 compilation mode effect on, 21
prefix, 57 data area, 21
mp_barrier, 117

007–0701–150 179
Index

external, 20 __nand_and_fetch, 125

functions __or_and_fetch, 125
changes, 20 __sub_and_fetch, 125
linker-defined, 20 __xor_and_fetch, 125
Negation, 56 OpemMP
Negative integers multiprocessing directives, 104
right shift on, 134 Operator
newline bitwise not, 56
in text streams, 148 operator, 57
Newlines, 23 ++operator, 55
Non-ANSI C Operators
compilation mode, 3 !, 56
Non-void function, 4 %, 58
Nongraphic characters, 25 &, 62
Nonprototyped function declarations, 53 *, 58
-nostdinc compiler option, 139 +, 59
NUL character, 26 unary, 56
Null, 26 ++
Null characters prefix, 57
in binary streams, 149 -, 59
NULL pointer, 140 unary, 56
NULL pointer constant, 48 /, 58
Null statement, 93 <<, 60
NUM_THREADS, 118 >>, 60
~, 56
additive, 59
O address-of, 56
AND, 62
Object assignment, 65
definition, 41 +=, 66
Objects -=, 66
definitions =, 65
external, 102 associativity, 49
external, 102 bitwise
types, 38 AND, 62
offsetof() macro, 3 cast, 58
op-and-fetch operations comma, 66
__add_and_fetch, 125 conditional, 64
__and_and_fetch, 125 conversions, 43
__max_and_fetch, 125 equality, 61
__min_and_fetch, 125 evaluation, 49
__mpy_and_fetch, 125

180 007–0701–150
 C Language Reference Manual

exclusive or, 62 Parameter list, 83

grouping, 49 Parenthesized expressions, 51
inclusive OR, 63 Passing arguments, 53
indirection, 56 perror function, 151
list of, 28 Pointer
logical convert to int, 58
AND, 63 truncation of value, 58
minus, 57 pointer, 39
multiplicative, 58 Pointer constant
OR NULL, 48
exclusive, 62 Pointer declarators, 79
inclusive, 63 Pointers
logical, 63 additive operators on, 60
order of evaluation, 49 command options, 80
precedence, 49 conversion to int, 136
relational, 60 conversions, 48
shift, 60 differences of, 136
sizeof, 57 integer additions and, 44
unary, 55 qualifiers, 79
-OPT restricted, 79
alias=disjoint, 80 to qualified types, 79
alias=restrict, 80 to void, 48
OR operator pointers
exclusive, 62 casting to int, 136
inclusive, 63 Postfix expressions, 51
logical, 63 ++, 55
Order of evaluation - -, 55
operators, 49 function calls, 52
Overflow handling, 67 indirect references, 54
structure references, 54
subscripts, 52
P union references, 54
pow, 141
%p conversion powf, 141
in fprintf function, 150 #pragma critical, 106
in fscanf function, 150 #pragma directives
Parallel computing forum (PCF), 105 changes from Fortran directives, 105
Parallel Fortran coding rules, 105
communication between threads, 120 #pragma enter gate, 106
Parallel reduction operations, 107 #pragma exit gate, 106
Parallel regions, 104 #pragma independent, 107
work-sharing constructs, 106

007–0701–150 181
Index

#pragma parallel, 107 R

#pragma pfor, 107
restrictions, 111 Ranges
#pragma synchronize, 106 floating points, 135
Precedence integers, 133
examples, 49 realloc, 158
Precedence of operators, 49 Recommendations
Precision, 27 coding practices, 3
Preprocessing directives, 139 things to avoid, 4
Preprocessor Reduction
changes, 7 example, 110
Primary expressions, 51 on user-defined type in C++, 110
Programming hints, 3 reduction clause, 108
Promotions Reduction operations, 107
arguments, 53 Register, 83
arithmetic expressions, 14 -32 mode, 136
floating-point, 14 optimized code, 136
integral, 15, 45 register
Prototyped function declarations, 53 function declaration lists, 101
Prototyped functions, 85 Register keyword, 71
Prototypes, 82 Register storage-class specifier, 136
function, 18 Registers, 136
incompatible types, 19 Relational operators, 60
inconsistent, 20 Remainder
ptrdiff_t, 136 sign of, 134
Punctuators remove function
definition, 29 on an open file, 150
list of, 29 rename function, 150
putenv function, 159 Reserved keywords, 23
__restrict type qualifier, 79
example, 80
Q Result type
definition, 46
Quad precision, 134 return statements, 98
Qualified objects, 4 Right shift
Qualifiers, 77 on negative integers, 134
access to volatile, 138 Rounding
volatile, 77 type used, 135
Routines, 117
barrier, 117
mp_barrier, 117

182 007–0701–150
 C Language Reference Manual

mp_block, 115 shmem routines

mp_blocktime, 116 rules, 122
mp_create, 115 short, 39
mp_destroy, 115 SIGCLD, 116
mp_my_threadnum, 117 Signal-catching functions
mp_numthreads, 116 valid codes, 145
mp_set_numthreads, 116 Signals
mp_set_slave_stacksize, 117 semantics, 143
mp_setlock, 117 set of, 141
mp_setup, 115 SIGCLD, 116
mp_suggested_numthreads, 116 signals
mp_unblock, 115 default handling, 148
mp_unsetlock, 117 Simple assignment, 65
ussetlock, 117 Single precision, 43
usunsetlock, 117 size_t, 57, 136
Run-time library routines sizeof, 57, 86, 136
See "Routines", 115 type of result, 57
Sizes
floating points, 135
S integers, 133
Slave processes
Scalar types, 40 stack size, 120
Scheduling, 119 Slave threads
Scheduling methods blocking, 115, 116
between processors, 120 &space, 78
gang, 120 special characters, 25
Scope Spin-wait lock example, 128
block, 32 sprocsp, 117
changes, 11 Stack size, 118, 120
definition, 31 Standard headers, 22
file, 33 Standards
function, 32 ANSI C, 1
function prototype, 32 Statements
Scoping block, 93
changes, 10 break, 98
Scoping restrictions, 113 compound, 93
Selection statements, 94 scope of declarations, 94
setlocale, 140 continue, 96, 97
Shift operators, 60 do, 96
Shift states, 131 else, 94
shmem expression, 93
See "mp_shmem", 120

007–0701–150 183
Index

for, 96 references, 54
goto, 97 Structure designators, 4
if, 94 Structures, 136
iteration, 95 alignment, 137
jump, 97 padding, 137
labeled, 99 structures
null, 93 initialization, 89
return, 98 Subroutines
selection, 94 See "Routines", 117
switch, 94 Subscripts
while, 96 in postfix expressions, 52
static Switch statements
function definitions, 101 maximum number of case values, 139
Static keyword, 70 switch statements, 94, 95
Static storage duration, 37, 70 labels, 99
stdarg, 4, 83 Switches, 2
stderr, 131 –ansi, 2
Storage class sizes, 39 –xansi, 2
storage class sizes, 39 Synchronization intrinsics, 109, 123
Storage class specifiers, 70 Synchronize operation
Storage duration, 70 __synchronize, 127
auto, 71 system function, 159
automatic, 38
static, 37, 70
strerror function, 159 T
String literals, 5, 28, 51, 162
wide, 28 Tabs, 23
wide characters, 90 Temporary files, 150, 158
Struct Text stream
namespace last line, 148
changes, 12 newline, 148
struct, 73 Text streams
initialization, 89 writes on, 149
members Thread
restrictions, 73 master, 117
Structs slave, 117
alignment, 137 Threads
Structure and processors, 120
declaration, 73 Time
indirect references, 54 availability, 140
members clock function, 160
restrictions, 73

184 007–0701–150
 C Language Reference Manual

daylight savings, 159 64-bit mode, 39

local, 159 arithmetic, 40
Time zone, 159 changes, 14
__TIME__, 140 character, 38
Token compatibility, 14, 17
definition, 23 derived, 40
Token concatenation, 9 differences, 39
Tokens float, 43
classes of, 23 floating, 44
in -ansi mode, 9 floating-point, 39
in -cckr mode, 9 int, 60
Traditional C integer, 39
allocating storage, 13 integral, 44
conversion rules, 45, 46 long double, 27
enumeration constants, 12 multibyte characters, 38
floating point, 43 promotion in arithmetic expressions, 14
function prototype error, 19 promotion rules, 14
identifiers, 31 promotions
incompatibilities, 84 arguments, 17
linkage discrepancies, 36 floating-point, 14
name space discrepancies, 33 integral, 15
scope, 12 sizes, 39
scoping differences, 11 unsigned char, 38
unsigned preserving integer promotion, 43 variably modified, 69
Traditional c void, 40
compiler option, 3 types
Translation, 129 integral, 40
Trigraph sequences, 27 TZ environment variable, 159
Truncation
direction of, 135
pointer value, 58 U
type used, 135
tType qualifiers Unary operators, 55
__restrict, 79 unblockproc, 116
Type, 15 Underflow handling, 67
Type names, 86 Underflow range errors
Type qualifiers, 77 math functions, 141
Type specifiers Union
list of, 71 indirect references, 54
typedef, 70, 72, 74, 77, 87 namespace
Types, 38 changes, 12
32-bit mode, 39

007–0701–150 185
Index

references, 54 pointers to, 48

union, 73 return statements, 98
declaration, 73 Volatile, 77
initialization, 89 volatile object, 4
members Volatile-qualified types
restrictions, 73 access to, 138
Unions, 136
32-bit mode, 137
64-bit mode, 137 W
accessing members, 137
unlock operation, 127 Warnings, 3
unsigned char, 38 while statements, 96
default, 133 White space, 23, 129
unsigned integers Wide characters, 133
conversions, 44 Wide string literals, 28
usconfig, 117 Words
-use_readwrite_const option, 162 alignment, 137
User name space, 2 size, 137
usinit, 117 Work-sharing constructs, 106
ussetlock, 117 write(), 20
usunsetlock, 117

X
V
-xansi compiler option
valid filenames, 149 external names and, 21
Variable length array -Xlocal, 120
as specifier for type declarator, 69
Variable length arrays, 81
Variables Z
float, 14
void, 40, 83 Zero-length files, 149
conversions, 47

186 007–0701–150