The GNU Multiple Precision Arithmetic Library
Edition 5.0.1
6 February 2010
by Torbj¨orn Granlund and the GMP development team
This manual describes how to install and use the GNU multiple precision arithmetic library,
version 5.0.1.
Copyright 1991, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005,
2006, 2007, 2008, 2009, 2010 Free Software Foundation, Inc.
Permission is granted to copy, distribute and/or modify this document under the terms of
the GNU Free Documentation License, Version 1.3 or any later version published by the Free
Software Foundation; with no Invariant Sections, with the FrontCover Texts being “A GNU
Manual”, and with the BackCover Texts being “You have freedom to copy and modify this
GNU Manual, like GNU software”. A copy of the license is included in Appendix C [GNU Free
Documentation License], page 124.
i
Table of Contents
GNU MP Copying Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Introduction to GNU MP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1 How to use this Manual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Installing GMP. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Build Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 ABI and ISA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Notes for Package Builds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4 Notes for Particular Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Known Build Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.6 Performance optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 GMP Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1 Headers and Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Nomenclature and Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Function Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.4 Variable Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.5 Parameter Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.6 Memory Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.7 Reentrancy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.8 Useful Macros and Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.9 Compatibility with older versions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.10 Demonstration programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.11 Efciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.12 Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.13 Profling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.14 Autoconf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.15 Emacs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4 Reporting Bugs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5 Integer Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.1 Initialization Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.2 Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.3 Combined Initialization and Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.4 Conversion Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.5 Arithmetic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.6 Division Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.7 Exponentiation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.8 Root Extraction Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.9 Number Theoretic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.10 Comparison Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.11 Logical and Bit Manipulation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.12 Input and Output Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5.13 Random Number Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
ii GNU MP 5.0.1
5.14 Integer Import and Export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5.15 Miscellaneous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.16 Special Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6 Rational Number Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.1 Initialization and Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.2 Conversion Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.3 Arithmetic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
6.4 Comparison Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.5 Applying Integer Functions to Rationals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
6.6 Input and Output Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7 Floatingpoint Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
7.1 Initialization Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
7.2 Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.3 Combined Initialization and Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.4 Conversion Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
7.5 Arithmetic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
7.6 Comparison Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.7 Input and Output Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.8 Miscellaneous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
8 Lowlevel Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
8.1 Nails . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
9 Random Number Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
9.1 Random State Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
9.2 Random State Seeding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
9.3 Random State Miscellaneous . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
10 Formatted Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
10.1 Format Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
10.2 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
10.3 C++ Formatted Output. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
11 Formatted Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
11.1 Formatted Input Strings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
11.2 Formatted Input Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
11.3 C++ Formatted Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
12 C++ Class Interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
12.1 C++ Interface General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
12.2 C++ Interface Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
12.3 C++ Interface Rationals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
12.4 C++ Interface Floats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
12.5 C++ Interface Random Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
12.6 C++ Interface Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
13 Berkeley MP Compatible Functions . . . . . . . . . . . . . . . . . . . . . . . 84
iii
14 Custom Allocation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
15 Language Bindings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
16 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
16.1 Multiplication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
16.1.1 Basecase Multiplication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
16.1.2 Karatsuba Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
16.1.3 Toom 3Way Multiplication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
16.1.4 Toom 4Way Multiplication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
16.1.5 FFT Multiplication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
16.1.6 Other Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
16.1.7 Unbalanced Multiplication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
16.2 Division Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
16.2.1 Single Limb Division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
16.2.2 Basecase Division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
16.2.3 Divide and Conquer Division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
16.2.4 BlockWise Barrett Division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
16.2.5 Exact Division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
16.2.6 Exact Remainder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
16.2.7 Small Quotient Division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
16.3 Greatest Common Divisor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
16.3.1 Binary GCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
16.3.2 Lehmer’s algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
16.3.3 Subquadratic GCD. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
16.3.4 Extended GCD. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
16.3.5 Jacobi Symbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
16.4 Powering Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
16.4.1 Normal Powering. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
16.4.2 Modular Powering. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
16.5 Root Extraction Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
16.5.1 Square Root . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
16.5.2 Nth Root . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
16.5.3 Perfect Square . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
16.5.4 Perfect Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
16.6 Radix Conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
16.6.1 Binary to Radix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
16.6.2 Radix to Binary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
16.7 Other Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
16.7.1 Prime Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
16.7.2 Factorial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
16.7.3 Binomial Coefcients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
16.7.4 Fibonacci Numbers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
16.7.5 Lucas Numbers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
16.7.6 Random Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
16.8 Assembly Coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
16.8.1 Code Organisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
16.8.2 Assembly Basics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
16.8.3 Carry Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
16.8.4 Cache Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
16.8.5 Functional Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
16.8.6 Floating Point. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
16.8.7 SIMD Instructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
iv GNU MP 5.0.1
16.8.8 Software Pipelining. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
16.8.9 Loop Unrolling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
16.8.10 Writing Guide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
17 Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
17.1 Integer Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
17.2 Rational Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
17.3 Float Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
17.4 Raw Output Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
17.5 C++ Interface Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Appendix A Contributors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Appendix B References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
B.1 Books . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
B.2 Papers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Appendix C GNU Free Documentation License . . . . . . . . . . . 124
Concept Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
Function and Type Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
GNU MP Copying Conditions 1
GNU MP Copying Conditions
This library is free; this means that everyone is free to use it and free to redistribute it on a free
basis. The library is not in the public domain; it is copyrighted and there are restrictions on its
distribution, but these restrictions are designed to permit everything that a good cooperating
citizen would want to do. What is not allowed is to try to prevent others from further sharing
any version of this library that they might get from you.
Specifcally, we want to make sure that you have the right to give away copies of the library,
that you receive source code or else can get it if you want it, that you can change this library
or use pieces of it in new free programs, and that you know you can do these things.
To make sure that everyone has such rights, we have to forbid you to deprive anyone else of
these rights. For example, if you distribute copies of the GNU MP library, you must give the
recipients all the rights that you have. You must make sure that they, too, receive or can get
the source code. And you must tell them their rights.
Also, for our own protection, we must make certain that everyone fnds out that there is no
warranty for the GNU MP library. If it is modifed by someone else and passed on, we want
their recipients to know that what they have is not what we distributed, so that any problems
introduced by others will not refect on our reputation.
The precise conditions of the license for the GNU MP library are found in the Lesser General
Public License version 3 that accompanies the source code, see ‘COPYING.LIB’. Certain demon
stration programs are provided under the terms of the plain General Public License version 3,
see ‘COPYING’.
2 GNU MP 5.0.1
1 Introduction to GNU MP
GNU MP is a portable library written in C for arbitrary precision arithmetic on integers, rational
numbers, and foatingpoint numbers. It aims to provide the fastest possible arithmetic for all
applications that need higher precision than is directly supported by the basic C types.
Many applications use just a few hundred bits of precision; but some applications may need
thousands or even millions of bits. GMP is designed to give good performance for both, by
choosing algorithms based on the sizes of the operands, and by carefully keeping the overhead
at a minimum.
The speed of GMP is achieved by using fullwords as the basic arithmetic type, by using sophis
ticated algorithms, by including carefully optimized assembly code for the most common inner
loops for many diferent CPUs, and by a general emphasis on speed (as opposed to simplicity
or elegance).
There is assembly code for these CPUs: ARM, DEC Alpha 21064, 21164, and 21264, AMD
29000, AMD K6, K62, Athlon, and Athlon64, Hitachi SuperH and SH2, HPPA 1.0, 1.1 and
2.0, Intel Pentium, Pentium Pro/II/III, Pentium 4, generic x86, Intel IA64, i960, Motorola
MC68000, MC68020, MC88100, and MC88110, Motorola/IBM PowerPC 32 and 64, National
NS32000, IBM POWER, MIPS R3000, R4000, SPARCv7, SuperSPARC, generic SPARCv8,
UltraSPARC, DEC VAX, and Zilog Z8000. Some optimizations also for Cray vector systems,
Clipper, IBM ROMP (RT), and Pyramid AP/XP.
For uptodate information on GMP, please see the GMP web pages at
http://gmplib.org/
The latest version of the library is available at
ftp://ftp.gnu.org/gnu/gmp/
Many sites around the world mirror ‘ftp.gnu.org’, please use a mirror near you, see
http://www.gnu.org/order/ftp.html for a full list.
There are three public mailing lists of interest. One for release announcements, one for general
questions and discussions about usage of the GMP library and one for bug reports. For more
information, see
http://gmplib.org/mailman/listinfo/.
The proper place for bug reports is gmpbugs@gmplib.org. See Chapter 4 [Reporting Bugs],
page 28 for information about reporting bugs.
1.1 How to use this Manual
Everyone should read Chapter 3 [GMP Basics], page 16. If you need to install the library
yourself, then read Chapter 2 [Installing GMP], page 3. If you have a system with multiple
ABIs, then read Section 2.2 [ABI and ISA], page 8, for the compiler options that must be used
on applications.
The rest of the manual can be used for later reference, although it is probably a good idea to
glance through it.
Chapter 2: Installing GMP 3
2 Installing GMP
GMP has an autoconf/automake/libtool based confguration system. On a Unixlike system a
basic build can be done with
./configure
make
Some selftests can be run with
make check
And you can install (under ‘/usr/local’ by default) with
make install
If you experience problems, please report them to gmpbugs@gmplib.org. See Chapter 4 [Re
porting Bugs], page 28, for information on what to include in useful bug reports.
2.1 Build Options
All the usual autoconf confgure options are available, run ‘./configure help’ for a summary.
The fle ‘INSTALL.autoconf’ has some generic installation information too.
Tools ‘configure’ requires various Unixlike tools. See Section 2.4 [Notes for Particular
Systems], page 12, for some options on nonUnix systems.
It might be possible to build without the help of ‘configure’, certainly all the code
is there, but unfortunately you’ll be on your own.
Build Directory
To compile in a separate build directory, cd to that directory, and prefx the confgure
command with the path to the GMP source directory. For example
cd /my/build/dir
/my/sources/gmp5.0.1/configure
Not all ‘make’ programs have the necessary features (VPATH) to support this. In
particular, SunOS and Slowaris make have bugs that make them unable to build in
a separate directory. Use GNU make instead.
‘prefix’ and ‘execprefix’
The ‘prefix’ option can be used in the normal way to direct GMP to install
under a particular tree. The default is ‘/usr/local’.
‘execprefix’ can be used to direct architecturedependent fles like ‘libgmp.a’
to a diferent location. This can be used to share architectureindependent
parts like the documentation, but separate the dependent parts. Note however
that ‘gmp.h’ and ‘mp.h’ are architecturedependent since they encode certain as
pects of ‘libgmp’, so it will be necessary to ensure both ‘$prefix/include’ and
‘$exec_prefix/include’ are available to the compiler.
‘disableshared’, ‘disablestatic’
By default both shared and static libraries are built (where possible), but one or
other can be disabled. Shared libraries result in smaller executables and permit code
sharing between separate running processes, but on some CPUs are slightly slower,
having a small cost on each function call.
Native Compilation, ‘build=CPUVENDOROS’
For normal native compilation, the system can be specifed with ‘build’. By
default ‘./configure’ uses the output from running ‘./config.guess’. On some
4 GNU MP 5.0.1
systems ‘./config.guess’ can determine the exact CPU type, on others it will be
necessary to give it explicitly. For example,
./configure build=ultrasparcsunsolaris2.7
In all cases the ‘OS’ part is important, since it controls how libtool generates shared
libraries. Running ‘./config.guess’ is the simplest way to see what it should be,
if you don’t know already.
Cross Compilation, ‘host=CPUVENDOROS’
When crosscompiling, the system used for compiling is given by ‘build’ and the
system where the library will run is given by ‘host’. For example when using a
FreeBSD Athlon system to build GNU/Linux m68k binaries,
./configure build=athlonpcfreebsd3.5 host=m68kmaclinuxgnu
Compiler tools are sought frst with the host system type as a prefx. For example
m68kmaclinuxgnuranlib is tried, then plain ranlib. This makes it possible
for a set of crosscompiling tools to coexist with native tools. The prefx is the
argument to ‘host’, and this can be an alias, such as ‘m68klinux’. But note
that tools don’t have to be setup this way, it’s enough to just have a PATH with a
suitable crosscompiling cc etc.
Compiling for a diferent CPU in the same family as the build system is a form of
crosscompilation, though very possibly this would merely be special options on a
native compiler. In any case ‘./configure’ avoids depending on being able to run
code on the build system, which is important when creating binaries for a newer
CPU since they very possibly won’t run on the build system.
In all cases the compiler must be able to produce an executable (of whatever format)
from a standard C main. Although only object fles will go to make up ‘libgmp’,
‘./configure’ uses linking tests for various purposes, such as determining what
functions are available on the host system.
Currently a warning is given unless an explicit ‘build’ is used when cross
compiling, because it may not be possible to correctly guess the build system type
if the PATH has only a crosscompiling cc.
Note that the ‘target’ option is not appropriate for GMP. It’s for use when
building compiler tools, with ‘host’ being where they will run, and ‘target’
what they’ll produce code for. Ordinary programs or libraries like GMP are only
interested in the ‘host’ part, being where they’ll run. (Some past versions of
GMP used ‘target’ incorrectly.)
CPU types
In general, if you want a library that runs as fast as possible, you should confgure
GMP for the exact CPU type your system uses. However, this may mean the binaries
won’t run on older members of the family, and might run slower on other members,
older or newer. The best idea is always to build GMP for the exact machine type
you intend to run it on.
The following CPUs have specifc support. See ‘configure.in’ for details of what
code and compiler options they select.
• Alpha: ‘alpha’, ‘alphaev5’, ‘alphaev56’, ‘alphapca56’, ‘alphapca57’,
‘alphaev6’, ‘alphaev67’, ‘alphaev68’ ‘alphaev7’
• Cray: ‘c90’, ‘j90’, ‘t90’, ‘sv1’
• HPPA: ‘hppa1.0’, ‘hppa1.1’, ‘hppa2.0’, ‘hppa2.0n’, ‘hppa2.0w’, ‘hppa64’
• IA64: ‘ia64’, ‘itanium’, ‘itanium2’
• MIPS: ‘mips’, ‘mips3’, ‘mips64’
Chapter 2: Installing GMP 5
• Motorola: ‘m68k’, ‘m68000’, ‘m68010’, ‘m68020’, ‘m68030’, ‘m68040’, ‘m68060’,
‘m68302’, ‘m68360’, ‘m88k’, ‘m88110’
• POWER: ‘power’, ‘power1’, ‘power2’, ‘power2sc’
• PowerPC: ‘powerpc’, ‘powerpc64’, ‘powerpc401’, ‘powerpc403’, ‘powerpc405’,
‘powerpc505’, ‘powerpc601’, ‘powerpc602’, ‘powerpc603’, ‘powerpc603e’,
‘powerpc604’, ‘powerpc604e’, ‘powerpc620’, ‘powerpc630’, ‘powerpc740’,
‘powerpc7400’, ‘powerpc7450’, ‘powerpc750’, ‘powerpc801’, ‘powerpc821’,
‘powerpc823’, ‘powerpc860’, ‘powerpc970’
• SPARC: ‘sparc’, ‘sparcv8’, ‘microsparc’, ‘supersparc’, ‘sparcv9’,
‘ultrasparc’, ‘ultrasparc2’, ‘ultrasparc2i’, ‘ultrasparc3’, ‘sparc64’
• x86 family: ‘i386’, ‘i486’, ‘i586’, ‘pentium’, ‘pentiummmx’, ‘pentiumpro’,
‘pentium2’, ‘pentium3’, ‘pentium4’, ‘k6’, ‘k62’, ‘k63’, ‘athlon’, ‘amd64’,
‘viac3’, ‘viac32’
• Other: ‘a29k’, ‘arm’, ‘clipper’, ‘i960’, ‘ns32k’, ‘pyramid’, ‘sh’, ‘sh2’, ‘vax’,
‘z8k’
CPUs not listed will use generic C code.
Generic C Build
If some of the assembly code causes problems, or if otherwise desired, the generic C
code can be selected with CPU ‘none’. For example,
./configure host=noneunknownfreebsd3.5
Note that this will run quite slowly, but it should be portable and should at least
make it possible to get something running if all else fails.
Fat binary, ‘enablefat’
Using ‘enablefat’ selects a “fat binary” build on x86, where optimized low
level subroutines are chosen at runtime according to the CPU detected. This means
more code, but gives good performance on all x86 chips. (This option might become
available for more architectures in the future.)
‘ABI’ On some systems GMP supports multiple ABIs (application binary interfaces),
meaning data type sizes and calling conventions. By default GMP chooses the
best ABI available, but a particular ABI can be selected. For example
./configure host=mips64sgiirix6 ABI=n32
See Section 2.2 [ABI and ISA], page 8, for the available choices on relevant CPUs,
and what applications need to do.
‘CC’, ‘CFLAGS’
By default the C compiler used is chosen from among some likely candidates, with
gcc normally preferred if it’s present. The usual ‘CC=whatever’ can be passed to
‘./configure’ to choose something diferent.
For various systems, default compiler fags are set based on the CPU and compiler.
The usual ‘CFLAGS="whatever"’ can be passed to ‘./configure’ to use something
diferent or to set good fags for systems GMP doesn’t otherwise know.
The ‘CC’ and ‘CFLAGS’ used are printed during ‘./configure’, and can be found
in each generated ‘Makefile’. This is the easiest way to check the defaults when
considering changing or adding something.
Note that when ‘CC’ and ‘CFLAGS’ are specifed on a system supporting multiple
ABIs it’s important to give an explicit ‘ABI=whatever’, since GMP can’t determine
the ABI just from the fags and won’t be able to select the correct assembly code.
If just ‘CC’ is selected then normal default ‘CFLAGS’ for that compiler will be used
(if GMP recognises it). For example ‘CC=gcc’ can be used to force the use of GCC,
with default fags (and default ABI).
6 GNU MP 5.0.1
‘CPPFLAGS’
Any fags like ‘D’ defnes or ‘I’ includes required by the preprocessor should be
set in ‘CPPFLAGS’ rather than ‘CFLAGS’. Compiling is done with both ‘CPPFLAGS’
and ‘CFLAGS’, but preprocessing uses just ‘CPPFLAGS’. This distinction is because
most preprocessors won’t accept all the fags the compiler does. Preprocessing is
done separately in some confgure tests, and in the ‘ansi2knr’ support for K&R
compilers.
‘CC_FOR_BUILD’
Some buildtime programs are compiled and run to generate hostspecifc data ta
bles. ‘CC_FOR_BUILD’ is the compiler used for this. It doesn’t need to be in any
particular ABI or mode, it merely needs to generate executables that can run. The
default is to try the selected ‘CC’ and some likely candidates such as ‘cc’ and ‘gcc’,
looking for something that works.
No fags are used with ‘CC_FOR_BUILD’ because a simple invocation like ‘cc foo.c’
should be enough. If some particular options are required they can be included as
for instance ‘CC_FOR_BUILD="cc whatever"’.
C++ Support, ‘enablecxx’
C++ support in GMP can be enabled with ‘enablecxx’, in which case a C++
compiler will be required. As a convenience ‘enablecxx=detect’ can be used
to enable C++ support only if a compiler can be found. The C++ support consists
of a library ‘libgmpxx.la’ and header fle ‘gmpxx.h’ (see Section 3.1 [Headers and
Libraries], page 16).
A separate ‘libgmpxx.la’ has been adopted rather than having C++ objects within
‘libgmp.la’ in order to ensure dynamic linked C programs aren’t bloated by a
dependency on the C++ standard library, and to avoid any chance that the C++
compiler could be required when linking plain C programs.
‘libgmpxx.la’ will use certain internals from ‘libgmp.la’ and can only be expected
to work with ‘libgmp.la’ from the same GMP version. Future changes to the rele
vant internals will be accompanied by renaming, so a mismatch will cause unresolved
symbols rather than perhaps mysterious misbehaviour.
In general ‘libgmpxx.la’ will be usable only with the C++ compiler that built it,
since name mangling and runtime support are usually incompatible between diferent
compilers.
‘CXX’, ‘CXXFLAGS’
When C++ support is enabled, the C++ compiler and its fags can be set with vari
ables ‘CXX’ and ‘CXXFLAGS’ in the usual way. The default for ‘CXX’ is the frst compiler
that works from a list of likely candidates, with g++ normally preferred when avail
able. The default for ‘CXXFLAGS’ is to try ‘CFLAGS’, ‘CFLAGS’ without ‘g’, then for
g++ either ‘g O2’ or ‘O2’, or for other compilers ‘g’ or nothing. Trying ‘CFLAGS’
this way is convenient when using ‘gcc’ and ‘g++’ together, since the fags for ‘gcc’
will usually suit ‘g++’.
It’s important that the C and C++ compilers match, meaning their startup and
runtime support routines are compatible and that they generate code in the same
ABI (if there’s a choice of ABIs on the system). ‘./configure’ isn’t currently able to
check these things very well itself, so for that reason ‘disablecxx’ is the default,
to avoid a build failure due to a compiler mismatch. Perhaps this will change in the
future.
Incidentally, it’s normally not good enough to set ‘CXX’ to the same as ‘CC’. Although
gcc for instance recognises ‘foo.cc’ as C++ code, only g++ will invoke the linker the
right way when building an executable or shared library from C++ object fles.
Chapter 2: Installing GMP 7
Temporary Memory, ‘enablealloca=<choice>’
GMP allocates temporary workspace using one of the following three methods, which
can be selected with for instance ‘enablealloca=mallocreentrant’.
• ‘alloca’  C library or compiler builtin.
• ‘mallocreentrant’  the heap, in a reentrant fashion.
• ‘mallocnotreentrant’  the heap, with global variables.
For convenience, the following choices are also available. ‘disablealloca’ is the
same as ‘no’.
• ‘yes’  a synonym for ‘alloca’.
• ‘no’  a synonym for ‘mallocreentrant’.
• ‘reentrant’  alloca if available, otherwise ‘mallocreentrant’. This is the
default.
• ‘notreentrant’  alloca if available, otherwise ‘mallocnotreentrant’.
alloca is reentrant and fast, and is recommended. It actually allocates just small
blocks on the stack; larger ones use mallocreentrant.
‘mallocreentrant’ is, as the name suggests, reentrant and thread safe, but
‘mallocnotreentrant’ is faster and should be used if reentrancy is not required.
The two malloc methods in fact use the memory allocation functions selected by mp_
set_memory_functions, these being malloc and friends by default. See Chapter 14
[Custom Allocation], page 86.
An additional choice ‘enablealloca=debug’ is available, to help when debugging
memory related problems (see Section 3.12 [Debugging], page 23).
FFT Multiplication, ‘disablefft’
By default multiplications are done using Karatsuba, 3way Toom, and Fermat FFT.
The FFT is only used on large to very large operands and can be disabled to save
code size if desired.
Berkeley MP, ‘enablempbsd’
The Berkeley MP compatibility library (‘libmp’) and header fle (‘mp.h’) are built
and installed only if ‘enablempbsd’ is used. See Chapter 13 [BSD Compatible
Functions], page 84.
Assertion Checking, ‘enableassert’
This option enables some consistency checking within the library. This can be of
use while debugging, see Section 3.12 [Debugging], page 23.
Execution Profling, ‘enableprofiling=prof/gprof/instrument’
Enable profling support, in one of various styles, see Section 3.13 [Profling], page 25.
‘MPN_PATH’
Various assembly versions of each mpn subroutines are provided. For a given CPU,
a search is made though a path to choose a version of each. For example ‘sparcv8’
has
MPN_PATH="sparc32/v8 sparc32 generic"
which means look frst for v8 code, then plain sparc32 (which is v7), and fnally
fall back on generic C. Knowledgeable users with special requirements can specify
a diferent path. Normally this is completely unnecessary.
Documentation
The source for the document you’re now reading is ‘doc/gmp.texi’, in Texinfo
format, see Texinfo.
8 GNU MP 5.0.1
Info format ‘doc/gmp.info’ is included in the distribution. The usual automake
targets are available to make PostScript, DVI, PDF and HTML (these will require
various T
E
X and Texinfo tools).
DocBook and XML can be generated by the Texinfo makeinfo program too, see
Section “Options for makeinfo” in Texinfo.
Some supplementary notes can also be found in the ‘doc’ subdirectory.
2.2 ABI and ISA
ABI (Application Binary Interface) refers to the calling conventions between functions, meaning
what registers are used and what sizes the various C data types are. ISA (Instruction Set
Architecture) refers to the instructions and registers a CPU has available.
Some 64bit ISA CPUs have both a 64bit ABI and a 32bit ABI defned, the latter for com
patibility with older CPUs in the family. GMP supports some CPUs like this in both ABIs. In
fact within GMP ‘ABI’ means a combination of chip ABI, plus how GMP chooses to use it. For
example in some 32bit ABIs, GMP may support a limb as either a 32bit long or a 64bit long
long.
By default GMP chooses the best ABI available for a given system, and this generally gives
signifcantly greater speed. But an ABI can be chosen explicitly to make GMP compatible with
other libraries, or particular application requirements. For example,
./configure ABI=32
In all cases it’s vital that all object code used in a given program is compiled for the same ABI.
Usually a limb is implemented as a long. When a long long limb is used this is encoded in
the generated ‘gmp.h’. This is convenient for applications, but it does mean that ‘gmp.h’ will
vary, and can’t be just copied around. ‘gmp.h’ remains compiler independent though, since all
compilers for a particular ABI will be expected to use the same limb type.
Currently no attempt is made to follow whatever conventions a system has for installing library
or header fles built for a particular ABI. This will probably only matter when installing multiple
builds of GMP, and it might be as simple as confguring with a special ‘libdir’, or it might
require more than that. Note that builds for diferent ABIs need to done separately, with a fresh
./configure and make each.
AMD64 (‘x86_64’)
On AMD64 systems supporting both 32bit and 64bit modes for applications, the
following ABI choices are available.
‘ABI=64’ The 64bit ABI uses 64bit limbs and pointers and makes full use of
the chip architecture. This is the default. Applications will usually not
need special compiler fags, but for reference the option is
gcc m64
‘ABI=32’ The 32bit ABI is the usual i386 conventions. This will be slower, and
is not recommended except for interoperating with other code not yet
64bit capable. Applications must be compiled with
gcc m32
(In GCC 2.95 and earlier there’s no ‘m32’ option, it’s the only mode.)
Chapter 2: Installing GMP 9
HPPA 2.0 (‘hppa2.0*’, ‘hppa64’)
‘ABI=2.0w’
The 2.0w ABI uses 64bit limbs and pointers and is available on HPUX
11 or up. Applications must be compiled with
gcc [built for 2.0w]
cc +DD64
‘ABI=2.0n’
The 2.0n ABI means the 32bit HPPA 1.0 ABI and all its normal calling
conventions, but with 64bit instructions permitted within functions.
GMP uses a 64bit long long for a limb. This ABI is available on
hppa64 GNU/Linux and on HPUX 10 or higher. Applications must be
compiled with
gcc [built for 2.0n]
cc +DA2.0 +e
Note that current versions of GCC (eg. 3.2) don’t generate 64bit in
structions for long long operations and so may be slower than for 2.0w.
(The GMP assembly code is the same though.)
‘ABI=1.0’ HPPA 2.0 CPUs can run all HPPA 1.0 and 1.1 code in the 32bit HPPA
1.0 ABI. No special compiler options are needed for applications.
All three ABIs are available for CPU types ‘hppa2.0w’, ‘hppa2.0’ and ‘hppa64’, but
for CPU type ‘hppa2.0n’ only 2.0n or 1.0 are considered.
Note that GCC on HPUX has no options to choose between 2.0n and 2.0w modes,
unlike HP cc. Instead it must be built for one or the other ABI. GMP will detect
how it was built, and skip to the corresponding ‘ABI’.
IA64 under HPUX (‘ia64**hpux*’, ‘itanium**hpux*’)
HPUX supports two ABIs for IA64. GMP performance is the same in both.
‘ABI=32’ In the 32bit ABI, pointers, ints and longs are 32 bits and GMP uses
a 64 bit long long for a limb. Applications can be compiled without
any special fags since this ABI is the default in both HP C and GCC,
but for reference the fags are
gcc milp32
cc +DD32
‘ABI=64’ In the 64bit ABI, longs and pointers are 64 bits and GMP uses a long
for a limb. Applications must be compiled with
gcc mlp64
cc +DD64
On other IA64 systems, GNU/Linux for instance, ‘ABI=64’ is the only choice.
MIPS under IRIX 6 (‘mips**irix[6789]’)
IRIX 6 always has a 64bit MIPS 3 or better CPU, and supports ABIs o32, n32,
and 64. n32 or 64 are recommended, and GMP performance will be the same in
each. The default is n32.
‘ABI=o32’ The o32 ABI is 32bit pointers and integers, and no 64bit operations.
GMP will be slower than in n32 or 64, this option only exists to support
old compilers, eg. GCC 2.7.2. Applications can be compiled with no
special fags on an old compiler, or on a newer compiler with
10 GNU MP 5.0.1
gcc mabi=32
cc 32
‘ABI=n32’ The n32 ABI is 32bit pointers and integers, but with a 64bit limb
using a long long. Applications must be compiled with
gcc mabi=n32
cc n32
‘ABI=64’ The 64bit ABI is 64bit pointers and integers. Applications must be
compiled with
gcc mabi=64
cc 64
Note that MIPS GNU/Linux, as of kernel version 2.2, doesn’t have the necessary
support for n32 or 64 and so only gets a 32bit limb and the MIPS 2 code.
PowerPC 64 (‘powerpc64’, ‘powerpc620’, ‘powerpc630’, ‘powerpc970’, ‘power4’, ‘power5’)
‘ABI=aix64’
The AIX 64 ABI uses 64bit limbs and pointers and is the default on
PowerPC 64 ‘**aix*’ systems. Applications must be compiled with
gcc maix64
xlc q64
‘ABI=mode64’
The ‘mode64’ ABI uses 64bit limbs and pointers, and is the default on
64bit GNU/Linux, BSD, and Mac OS X/Darwin systems. Applications
must be compiled with
gcc m64
‘ABI=mode32’
The ‘mode32’ ABI uses a 64bit long long limb but with the chip still
in 32bit mode and using 32bit calling conventions. This is the default
on for systems where the true 64bit ABIs are unavailable. No special
compiler options are needed for applications.
‘ABI=32’ This is the basic 32bit PowerPC ABI, with a 32bit limb. No special
compiler options are needed for applications.
GMP speed is greatest in ‘aix64’ and ‘mode32’. In ‘ABI=32’ only the 32bit ISA is
used and this doesn’t make full use of a 64bit chip. On a suitable system we could
perhaps use more of the ISA, but there are no plans to do so.
Sparc V9 (‘sparc64’, ‘sparcv9’, ‘ultrasparc*’)
‘ABI=64’ The 64bit V9 ABI is available on the various BSD sparc64 ports, recent
versions of Sparc64 GNU/Linux, and Solaris 2.7 and up (when the kernel
is in 64bit mode). GCC 3.2 or higher, or Sun cc is required. On
GNU/Linux, depending on the default gcc mode, applications must be
compiled with
gcc m64
On Solaris applications must be compiled with
gcc m64 mptr64 Wa,xarch=v9 mcpu=v9
cc xarch=v9
On the BSD sparc64 systems no special options are required, since 64
bits is the only ABI available.
Chapter 2: Installing GMP 11
‘ABI=32’ For the basic 32bit ABI, GMP still uses as much of the V9 ISA as it
can. In the Sun documentation this combination is known as “v8plus”.
On GNU/Linux, depending on the default gcc mode, applications may
need to be compiled with
gcc m32
On Solaris, no special compiler options are required for applications,
though using something like the following is recommended. (gcc 2.8
and earlier only support ‘mv8’ though.)
gcc mv8plus
cc xarch=v8plus
GMP speed is greatest in ‘ABI=64’, so it’s the default where available. The speed
is partly because there are extra registers available and partly because 64bits is
considered the more important case and has therefore had better code written for
it.
Don’t be confused by the names of the ‘m’ and ‘x’ compiler options, they’re called
‘arch’ but efectively control both ABI and ISA.
On Solaris 2.6 and earlier, only ‘ABI=32’ is available since the kernel doesn’t save
all registers.
On Solaris 2.7 with the kernel in 32bit mode, a normal native build will reject
‘ABI=64’ because the resulting executables won’t run. ‘ABI=64’ can still be built if
desired by making it look like a crosscompile, for example
./configure build=none host=sparcv9sunsolaris2.7 ABI=64
2.3 Notes for Package Builds
GMP should present no great difculties for packaging in a binary distribution.
Libtool is used to build the library and ‘versioninfo’ is set appropriately, having started
from ‘3:0:0’ in GMP 3.0 (see Section “Library interface versions” in GNU Libtool).
The GMP 4 series will be upwardly binary compatible in each release and will be upwardly
binary compatible with all of the GMP 3 series. Additional function interfaces may be added
in each release, so on systems where libtool versioning is not fully checked by the loader an
auxiliary mechanism may be needed to express that a dynamic linked application depends on a
new enough GMP.
An auxiliary mechanism may also be needed to express that ‘libgmpxx.la’ (from
‘enablecxx’, see Section 2.1 [Build Options], page 3) requires ‘libgmp.la’ from the same
GMP version, since this is not done by the libtool versioning, nor otherwise. A mismatch will
result in unresolved symbols from the linker, or perhaps the loader.
When building a package for a CPU family, care should be taken to use ‘host’ (or ‘build’)
to choose the least common denominator among the CPUs which might use the package. For
example this might mean plain ‘sparc’ (meaning V7) for SPARCs.
For x86s, ‘enablefat’ sets things up for a fat binary build, making a runtime selection of
optimized low level routines. This is a good choice for packaging to run on a range of x86 chips.
Users who care about speed will want GMP built for their exact CPU type, to make best use
of the available optimizations. Providing a way to suitably rebuild a package may be useful.
This could be as simple as making it possible for a user to omit ‘build’ (and ‘host’) so
‘./config.guess’ will detect the CPU. But a way to manually specify a ‘build’ will be
wanted for systems where ‘./config.guess’ is inexact.
12 GNU MP 5.0.1
On systems with multiple ABIs, a packaged build will need to decide which among the choices
is to be provided, see Section 2.2 [ABI and ISA], page 8. A given run of ‘./configure’ etc will
only build one ABI. If a second ABI is also required then a second run of ‘./configure’ etc
must be made, starting from a clean directory tree (‘make distclean’).
As noted under “ABI and ISA”, currently no attempt is made to follow system conventions for
install locations that vary with ABI, such as ‘/usr/lib/sparcv9’ for ‘ABI=64’ as opposed to
‘/usr/lib’ for ‘ABI=32’. A package build can override ‘libdir’ and other standard variables as
necessary.
Note that ‘gmp.h’ is a generated fle, and will be architecture and ABI dependent. When
attempting to install two ABIs simultaneously it will be important that an application compile
gets the correct ‘gmp.h’ for its desired ABI. If compiler include paths don’t vary with ABI
options then it might be necessary to create a ‘/usr/include/gmp.h’ which tests preprocessor
symbols and chooses the correct actual ‘gmp.h’.
2.4 Notes for Particular Systems
AIX 3 and 4
On systems ‘**aix[34]*’ shared libraries are disabled by default, since some
versions of the native ar fail on the convenience libraries used. A shared build can
be attempted with
./configure enableshared disablestatic
Note that the ‘disablestatic’ is necessary because in a shared build libtool
makes ‘libgmp.a’ a symlink to ‘libgmp.so’, apparently for the beneft of old ver
sions of ld which only recognise ‘.a’, but unfortunately this is done even if a fully
functional ld is available.
ARM On systems ‘arm***’, versions of GCC up to and including 2.95.3 have a bug in
unsigned division, giving wrong results for some operands. GMP ‘./configure’ will
demand GCC 2.95.4 or later.
Compaq C++
Compaq C++ on OSF 5.1 has two favours of iostream, a standard one and an old
prestandard one (see ‘man iostream_intro’). GMP can only use the standard one,
which unfortunately is not the default but must be selected by defning __USE_STD_
IOSTREAM. Confgure with for instance
./configure enablecxx CPPFLAGS=D__USE_STD_IOSTREAM
Floating Point Mode
On some systems, the hardware foating point has a control mode which can set
all operations to be done in a particular precision, for instance single, double or
extended on x86 systems (x87 foating point). The GMP functions involving a
double cannot be expected to operate to their full precision when the hardware is
in single precision mode. Of course this afects all code, including application code,
not just GMP.
MSDOS and MS Windows
On an MSDOS system DJGPP can be used to build GMP, and on an MS Windows
system Cygwin, DJGPP and MINGW can be used. All three are excellent ports of
GCC and the various GNU tools.
http://www.cygwin.com/
http://www.delorie.com/djgpp/
http://www.mingw.org/
Microsoft also publishes an Interix “Services for Unix” which can be used to build
GMP on Windows (with a normal ‘./configure’), but it’s not free software.
Chapter 2: Installing GMP 13
MS Windows DLLs
On systems ‘**cygwin*’, ‘**mingw*’ and ‘**pw32*’ by default GMP builds
only a static library, but a DLL can be built instead using
./configure disablestatic enableshared
Static and DLL libraries can’t both be built, since certain export directives in ‘gmp.h’
must be diferent.
A MINGW DLL build of GMP can be used with Microsoft C. Libtool doesn’t install
a ‘.lib’ format import library, but it can be created with MS lib as follows, and
copied to the install directory. Similarly for ‘libmp’ and ‘libgmpxx’.
cd .libs
lib /def:libgmp3.dll.def /out:libgmp3.lib
MINGW uses the C runtime library ‘msvcrt.dll’ for I/O, so applications wanting
to use the GMP I/O routines must be compiled with ‘cl /MD’ to do the same. If
one of the other C runtime library choices provided by MS C is desired then the
suggestion is to use the GMP string functions and confne I/O to the application.
Motorola 68k CPU Types
‘m68k’ is taken to mean 68000. ‘m68020’ or higher will give a performance boost on
applicable CPUs. ‘m68360’ can be used for CPU32 series chips. ‘m68302’ can be
used for “Dragonball” series chips, though this is merely a synonym for ‘m68000’.
OpenBSD 2.6
m4 in this release of OpenBSD has a bug in eval that makes it unsuitable for ‘.asm’
fle processing. ‘./configure’ will detect the problem and either abort or choose
another m4 in the PATH. The bug is fxed in OpenBSD 2.7, so either upgrade or use
GNU m4.
Power CPU Types
In GMP, CPU types ‘power*’ and ‘powerpc*’ will each use instructions not available
on the other, so it’s important to choose the right one for the CPU that will be
used. Currently GMP has no assembly code support for using just the common
instruction subset. To get executables that run on both, the current suggestion is
to use the generic C code (CPU ‘none’), possibly with appropriate compiler options
(like ‘mcpu=common’ for gcc). CPU ‘rs6000’ (which is not a CPU but a family of
workstations) is accepted by ‘config.sub’, but is currently equivalent to ‘none’.
Sparc CPU Types
‘sparcv8’ or ‘supersparc’ on relevant systems will give a signifcant performance
increase over the V7 code selected by plain ‘sparc’.
Sparc App Regs
The GMP assembly code for both 32bit and 64bit Sparc clobbers the “application
registers” g2, g3 and g4, the same way that the GCC default ‘mappregs’ does
(see Section “SPARC Options” in Using the GNU Compiler Collection (GCC)).
This makes that code unsuitable for use with the special V9 ‘mcmodel=embmedany’
(which uses g4 as a data segment pointer), and for applications wanting to use those
registers for special purposes. In these cases the only suggestion currently is to build
GMP with CPU ‘none’ to avoid the assembly code.
SunOS 4 /usr/bin/m4 lacks various features needed to process ‘.asm’ fles, and instead
‘./configure’ will automatically use /usr/5bin/m4, which we believe is always
available (if not then use GNU m4).
14 GNU MP 5.0.1
x86 CPU Types
‘i586’, ‘pentium’ or ‘pentiummmx’ code is good for its intended P5 Pentium chips,
but quite slow when run on Intel P6 class chips (PPro, PII, PIII). ‘i386’ is a
better choice when making binaries that must run on both.
x86 MMX and SSE2 Code
If the CPU selected has MMX code but the assembler doesn’t support it, a warning
is given and nonMMX code is used instead. This will be an inferior build, since the
MMX code that’s present is there because it’s faster than the corresponding plain
integer code. The same applies to SSE2.
Old versions of ‘gas’ don’t support MMX instructions, in particular version 1.92.3
that comes with FreeBSD 2.2.8 or the more recent OpenBSD 3.1 doesn’t.
Solaris 2.6 and 2.7 as generate incorrect object code for register to register movq
instructions, and so can’t be used for MMX code. Install a recent gas if MMX code
is wanted on these systems.
2.5 Known Build Problems
You might fnd more uptodate information at http://gmplib.org/.
Compiler link options
The version of libtool currently in use rather aggressively strips compiler options
when linking a shared library. This will hopefully be relaxed in the future, but for
now if this is a problem the suggestion is to create a little script to hide them, and
for instance confgure with
./configure CC=gccwithmyoptions
DJGPP (‘**msdosdjgpp*’)
The DJGPP port of bash 2.03 is unable to run the ‘configure’ script, it exits
silently, having died writing a preamble to ‘config.log’. Use bash 2.04 or higher.
‘make all’ was found to run out of memory during the fnal ‘libgmp.la’ link on one
system tested, despite having 64Mb available. Running ‘make libgmp.la’ directly
helped, perhaps recursing into the various subdirectories uses up memory.
GNU binutils strip prior to 2.12
strip from GNU binutils 2.11 and earlier should not be used on the static libraries
‘libgmp.a’ and ‘libmp.a’ since it will discard all but the last of multiple archive
members with the same name, like the three versions of ‘init.o’ in ‘libgmp.a’.
Binutils 2.12 or higher can be used successfully.
The shared libraries ‘libgmp.so’ and ‘libmp.so’ are not afected by this and any
version of strip can be used on them.
make syntax error
On certain versions of SCO OpenServer 5 and IRIX 6.5 the native make is unable
to handle the long dependencies list for ‘libgmp.la’. The symptom is a “syntax
error” on the following line of the toplevel ‘Makefile’.
libgmp.la: $(libgmp_la_OBJECTS) $(libgmp_la_DEPENDENCIES)
Either use GNU Make, or as a workaround remove $(libgmp_la_DEPENDENCIES)
from that line (which will make the initial build work, but if any recompiling is done
‘libgmp.la’ might not be rebuilt).
MacOS X (‘**darwin*’)
Libtool currently only knows how to create shared libraries on MacOS X using the
native cc (which is a modifed GCC), not a plain GCC. A staticonly build should
work though (‘disableshared’).
Chapter 2: Installing GMP 15
NeXT prior to 3.3
The system compiler on old versions of NeXT was a massacred and old GCC, even
if it called itself ‘cc’. This compiler cannot be used to build GMP, you need to
get a real GCC, and install that. (NeXT may have fxed this in release 3.3 of their
system.)
POWER and PowerPC
Bugs in GCC 2.7.2 (and 2.6.3) mean it can’t be used to compile GMP on POWER
or PowerPC. If you want to use GCC for these machines, get GCC 2.7.2.1 (or later).
Sequent Symmetry
Use the GNU assembler instead of the system assembler, since the latter has serious
bugs.
Solaris 2.6 The system sed prints an error “Output line too long” when libtool builds
‘libgmp.la’. This doesn’t seem to cause any obvious ill efects, but GNU sed
is recommended, to avoid any doubt.
Sparc Solaris 2.7 with gcc 2.95.2 in ‘ABI=32’
A shared library build of GMP seems to fail in this combination, it builds but
then fails the tests, apparently due to some incorrect data relocations within gmp_
randinit_lc_2exp_size. The exact cause is unknown, ‘disableshared’ is rec
ommended.
2.6 Performance optimization
For optimal performance, build GMP for the exact CPU type of the target computer, see
Section 2.1 [Build Options], page 3.
Unlike what is the case for most other programs, the compiler typically doesn’t matter much,
since GMP uses assembly language for the most critical operation.
In particular for longrunning GMP applications, and applications demanding extremely large
numbers, building and running the tuneup program in the ‘tune’ subdirectory, can be important.
For example,
cd tune
make tuneup
./tuneup
will generate better contents for the ‘gmpmparam.h’ parameter fle.
To use the results, put the output in the fle fle indicated in the ‘Parameters for ...’ header.
Then recompile from scratch.
The tuneup program takes one useful parameter, ‘f NNN’, which instructs the program how long
to check FFT multiply parameters. If you’re going to use GMP for extremely large numbers,
you may want to run tuneup with a large NNN value.
16 GNU MP 5.0.1
3 GMP Basics
Using functions, macros, data types, etc. not documented in this manual is strongly discouraged.
If you do so your application is guaranteed to be incompatible with future versions of GMP.
3.1 Headers and Libraries
All declarations needed to use GMP are collected in the include fle ‘gmp.h’. It is designed to
work with both C and C++ compilers.
#include <gmp.h>
Note however that prototypes for GMP functions with FILE * parameters are only provided if
<stdio.h> is included too.
#include <stdio.h>
#include <gmp.h>
Likewise <stdarg.h> (or <varargs.h>) is required for prototypes with va_list parameters,
such as gmp_vprintf. And <obstack.h> for prototypes with struct obstack parameters, such
as gmp_obstack_printf, when available.
All programs using GMP must link against the ‘libgmp’ library. On a typical Unixlike system
this can be done with ‘lgmp’, for example
gcc myprogram.c lgmp
GMP C++ functions are in a separate ‘libgmpxx’ library. This is built and installed if C++
support has been enabled (see Section 2.1 [Build Options], page 3). For example,
g++ mycxxprog.cc lgmpxx lgmp
GMP is built using Libtool and an application can use that to link if desired, see GNU Libtool .
If GMP has been installed to a nonstandard location then it may be necessary to use ‘I’ and
‘L’ compiler options to point to the right directories, and some sort of runtime path for a
shared library.
3.2 Nomenclature and Types
In this manual, integer usually means a multiple precision integer, as defned by the GMP
library. The C data type for such integers is mpz_t. Here are some examples of how to declare
such integers:
mpz_t sum;
struct foo { mpz_t x, y; };
mpz_t vec[20];
Rational number means a multiple precision fraction. The C data type for these fractions is
mpq_t. For example:
mpq_t quotient;
Floating point number or Float for short, is an arbitrary precision mantissa with a limited
precision exponent. The C data type for such objects is mpf_t. For example:
mpf_t fp;
Chapter 3: GMP Basics 17
The foating point functions accept and return exponents in the C type mp_exp_t. Currently
this is usually a long, but on some systems it’s an int for efciency.
A limb means the part of a multiprecision number that fts in a single machine word. (We chose
this word because a limb of the human body is analogous to a digit, only larger, and containing
several digits.) Normally a limb is 32 or 64 bits. The C data type for a limb is mp_limb_t.
Counts of limbs of a multiprecision number represented in the C type mp_size_t. Currently
this is normally a long, but on some systems it’s an int for efciency, and on some systems it
will be long long in the future.
Counts of bits of a multiprecision number are represented in the C type mp_bitcnt_t. Currently
this is always an unsigned long, but on some systems it will be an unsigned long long in the
future .
Random state means an algorithm selection and current state data. The C data type for such
objects is gmp_randstate_t. For example:
gmp_randstate_t rstate;
Also, in general mp_bitcnt_t is used for bit counts and ranges, and size_t is used for byte or
character counts.
3.3 Function Classes
There are six classes of functions in the GMP library:
1. Functions for signed integer arithmetic, with names beginning with mpz_. The associated
type is mpz_t. There are about 150 functions in this class. (see Chapter 5 [Integer Func
tions], page 29)
2. Functions for rational number arithmetic, with names beginning with mpq_. The associated
type is mpq_t. There are about 40 functions in this class, but the integer functions can be
used for arithmetic on the numerator and denominator separately. (see Chapter 6 [Rational
Number Functions], page 44)
3. Functions for foatingpoint arithmetic, with names beginning with mpf_. The associated
type is mpf_t. There are about 60 functions is this class. (see Chapter 7 [Floatingpoint
Functions], page 48)
4. Functions compatible with Berkeley MP, such as itom, madd, and mult. The associated
type is MINT. (see Chapter 13 [BSD Compatible Functions], page 84)
5. Fast lowlevel functions that operate on natural numbers. These are used by the functions
in the preceding groups, and you can also call them directly from very timecritical user
programs. These functions’ names begin with mpn_. The associated type is array of mp_
limb_t. There are about 30 (hardtouse) functions in this class. (see Chapter 8 [Lowlevel
Functions], page 56)
6. Miscellaneous functions. Functions for setting up custom allocation and functions for gen
erating random numbers. (see Chapter 14 [Custom Allocation], page 86, and see Chapter 9
[Random Number Functions], page 65)
3.4 Variable Conventions
GMP functions generally have output arguments before input arguments. This notation is by
analogy with the assignment operator. The BSD MP compatibility functions are exceptions,
having the output arguments last.
18 GNU MP 5.0.1
GMP lets you use the same variable for both input and output in one call. For example, the
main function for integer multiplication, mpz_mul, can be used to square x and put the result
back in x with
mpz_mul (x, x, x);
Before you can assign to a GMP variable, you need to initialize it by calling one of the special
initialization functions. When you’re done with a variable, you need to clear it out, using one
of the functions for that purpose. Which function to use depends on the type of variable. See
the chapters on integer functions, rational number functions, and foatingpoint functions for
details.
A variable should only be initialized once, or at least cleared between each initialization. After
a variable has been initialized, it may be assigned to any number of times.
For efciency reasons, avoid excessive initializing and clearing. In general, initialize near the
start of a function and clear near the end. For example,
void
foo (void)
{
mpz_t n;
int i;
mpz_init (n);
for (i = 1; i < 100; i++)
{
mpz_mul (n, ...);
mpz_fdiv_q (n, ...);
...
}
mpz_clear (n);
}
3.5 Parameter Conventions
When a GMP variable is used as a function parameter, it’s efectively a callbyreference, meaning
if the function stores a value there it will change the original in the caller. Parameters which
are inputonly can be designated const to provoke a compiler error or warning on attempting
to modify them.
When a function is going to return a GMP result, it should designate a parameter that it sets,
like the library functions do. More than one value can be returned by having more than one
output parameter, again like the library functions. A return of an mpz_t etc doesn’t return the
object, only a pointer, and this is almost certainly not what’s wanted.
Here’s an example accepting an mpz_t parameter, doing a calculation, and storing the result to
the indicated parameter.
void
foo (mpz_t result, const mpz_t param, unsigned long n)
{
unsigned long i;
mpz_mul_ui (result, param, n);
for (i = 1; i < n; i++)
mpz_add_ui (result, result, i*7);
}
Chapter 3: GMP Basics 19
int
main (void)
{
mpz_t r, n;
mpz_init (r);
mpz_init_set_str (n, "123456", 0);
foo (r, n, 20L);
gmp_printf ("%Zd\n", r);
return 0;
}
foo works even if the mainline passes the same variable for param and result, just like the
library functions. But sometimes it’s tricky to make that work, and an application might not
want to bother supporting that sort of thing.
For interest, the GMP types mpz_t etc are implemented as oneelement arrays of certain struc
tures. This is why declaring a variable creates an object with the felds GMP needs, but then
using it as a parameter passes a pointer to the object. Note that the actual felds in each mpz_t
etc are for internal use only and should not be accessed directly by code that expects to be
compatible with future GMP releases.
3.6 Memory Management
The GMP types like mpz_t are small, containing only a couple of sizes, and pointers to allocated
data. Once a variable is initialized, GMP takes care of all space allocation. Additional space is
allocated whenever a variable doesn’t have enough.
mpz_t and mpq_t variables never reduce their allocated space. Normally this is the best policy,
since it avoids frequent reallocation. Applications that need to return memory to the heap at
some particular point can use mpz_realloc2, or clear variables no longer needed.
mpf_t variables, in the current implementation, use a fxed amount of space, determined by the
chosen precision and allocated at initialization, so their size doesn’t change.
All memory is allocated using malloc and friends by default, but this can be changed, see
Chapter 14 [Custom Allocation], page 86. Temporary memory on the stack is also used (via
alloca), but this can be changed at buildtime if desired, see Section 2.1 [Build Options], page 3.
3.7 Reentrancy
GMP is reentrant and threadsafe, with some exceptions:
• If confgured with ‘enablealloca=mallocnotreentrant’ (or with
‘enablealloca=notreentrant’ when alloca is not available), then naturally
GMP is not reentrant.
• mpf_set_default_prec and mpf_init use a global variable for the selected precision. mpf_
init2 can be used instead, and in the C++ interface an explicit precision to the mpf_class
constructor.
• mpz_random and the other old random number functions use a global random state and are
hence not reentrant. The newer random number functions that accept a gmp_randstate_t
parameter can be used instead.
• gmp_randinit (obsolete) returns an error indication through a global variable, which is not
thread safe. Applications are advised to use gmp_randinit_default or gmp_randinit_lc_
2exp instead.
20 GNU MP 5.0.1
• mp_set_memory_functions uses global variables to store the selected memory allocation
functions.
• If the memory allocation functions set by a call to mp_set_memory_functions (or malloc
and friends by default) are not reentrant, then GMP will not be reentrant either.
• If the standard I/O functions such as fwrite are not reentrant then the GMP I/O functions
using them will not be reentrant either.
• It’s safe for two threads to read from the same GMP variable simultaneously, but it’s
not safe for one to read while the another might be writing, nor for two threads to write
simultaneously. It’s not safe for two threads to generate a random number from the same
gmp_randstate_t simultaneously, since this involves an update of that variable.
3.8 Useful Macros and Constants
[Global Constant] const int mp_bits_per_limb
The number of bits per limb.
[Macro] __GNU_MP_VERSION
[Macro] __GNU_MP_VERSION_MINOR
[Macro] __GNU_MP_VERSION_PATCHLEVEL
The major and minor GMP version, and patch level, respectively, as integers. For GMP i.j,
these numbers will be i, j, and 0, respectively. For GMP i.j.k, these numbers will be i, j, and
k, respectively.
[Global Constant] const char * const gmp_version
The GMP version number, as a nullterminated string, in the form “i.j.k”. This release is
"5.0.1". Note that the format “i.j” was used when k was zero was used before version 4.3.0.
[Macro] __GMP_CC
[Macro] __GMP_CFLAGS
The compiler and compiler fags, respectively, used when compiling GMP, as strings.
3.9 Compatibility with older versions
This version of GMP is upwardly binary compatible with all 4.x and 3.x versions, and upwardly
compatible at the source level with all 2.x versions, with the following exceptions.
• mpn_gcd had its source arguments swapped as of GMP 3.0, for consistency with other mpn
functions.
• mpf_get_prec counted precision slightly diferently in GMP 3.0 and 3.0.1, but in 3.1 re
verted to the 2.x style.
There are a number of compatibility issues between GMP 1 and GMP 2 that of course also
apply when porting applications from GMP 1 to GMP 4. Please see the GMP 2 manual for
details.
The Berkeley MP compatibility library (see Chapter 13 [BSD Compatible Functions], page 84)
is source and binary compatible with the standard ‘libmp’.
3.10 Demonstration programs
The ‘demos’ subdirectory has some sample programs using GMP. These aren’t built or installed,
but there’s a ‘Makefile’ with rules for them. For instance,
Chapter 3: GMP Basics 21
make pexpr
./pexpr 68^975+10
The following programs are provided
• ‘pexpr’ is an expression evaluator, the program used on the GMP web page.
• The ‘calc’ subdirectory has a similar but simpler evaluator using lex and yacc.
• The ‘expr’ subdirectory is yet another expression evaluator, a library designed for ease of
use within a C program. See ‘demos/expr/README’ for more information.
• ‘factorize’ is a PollardRho factorization program.
• ‘isprime’ is a commandline interface to the mpz_probab_prime_p function.
• ‘primes’ counts or lists primes in an interval, using a sieve.
• ‘qcn’ is an example use of mpz_kronecker_ui to estimate quadratic class numbers.
• The ‘perl’ subdirectory is a comprehensive perl interface to GMP. See
‘demos/perl/INSTALL’ for more information. Documentation is in POD format
in ‘demos/perl/GMP.pm’.
As an aside, consideration has been given at various times to some sort of expression evaluation
within the main GMP library. Going beyond something minimal quickly leads to matters like
userdefned functions, looping, fxnums for control variables, etc, which are considered outside
the scope of GMP (much closer to language interpreters or compilers, See Chapter 15 [Language
Bindings], page 88.) Something simple for program input convenience may yet be a possibility,
a combination of the ‘expr’ demo and the ‘pexpr’ tree backend perhaps. But for now the above
evaluators are ofered as illustrations.
3.11 Efciency
Small Operands
On small operands, the time for function call overheads and memory allocation can
be signifcant in comparison to actual calculation. This is unavoidable in a general
purpose variable precision library, although GMP attempts to be as efcient as it
can on both large and small operands.
Static Linking
On some CPUs, in particular the x86s, the static ‘libgmp.a’ should be used for
maximum speed, since the PIC code in the shared ‘libgmp.so’ will have a small
overhead on each function call and global data address. For many programs this
will be insignifcant, but for long calculations there’s a gain to be had.
Initializing and Clearing
Avoid excessive initializing and clearing of variables, since this can be quite time
consuming, especially in comparison to otherwise fast operations like addition.
A language interpreter might want to keep a free list or stack of initialized variables
ready for use. It should be possible to integrate something like that with a garbage
collector too.
Reallocations
An mpz_t or mpq_t variable used to hold successively increasing values will have
its memory repeatedly realloced, which could be quite slow or could fragment
memory, depending on the C library. If an application can estimate the fnal size
then mpz_init2 or mpz_realloc2 can be called to allocate the necessary space from
the beginning (see Section 5.1 [Initializing Integers], page 29).
22 GNU MP 5.0.1
It doesn’t matter if a size set with mpz_init2 or mpz_realloc2 is too small, since all
functions will do a further reallocation if necessary. Badly overestimating memory
required will waste space though.
2exp Functions
It’s up to an application to call functions like mpz_mul_2exp when appropriate.
General purpose functions like mpz_mul make no attempt to identify powers of two
or other special forms, because such inputs will usually be very rare and testing
every time would be wasteful.
ui and si Functions
The ui functions and the small number of si functions exist for convenience and
should be used where applicable. But if for example an mpz_t contains a value that
fts in an unsigned long there’s no need extract it and call a ui function, just use
the regular mpz function.
InPlace Operations
mpz_abs, mpq_abs, mpf_abs, mpz_neg, mpq_neg and mpf_neg are fast when used for
inplace operations like mpz_abs(x,x), since in the current implementation only a
single feld of x needs changing. On suitable compilers (GCC for instance) this is
inlined too.
mpz_add_ui, mpz_sub_ui, mpf_add_ui and mpf_sub_ui beneft from an inplace
operation like mpz_add_ui(x,x,y), since usually only one or two limbs of x will
need to be changed. The same applies to the full precision mpz_add etc if y is small.
If y is big then cache locality may be helped, but that’s all.
mpz_mul is currently the opposite, a separate destination is slightly better. A call
like mpz_mul(x,x,y) will, unless y is only one limb, make a temporary copy of x
before forming the result. Normally that copying will only be a tiny fraction of the
time for the multiply, so this is not a particularly important consideration.
mpz_set, mpq_set, mpq_set_num, mpf_set, etc, make no attempt to recognise a
copy of something to itself, so a call like mpz_set(x,x) will be wasteful. Naturally
that would never be written deliberately, but if it might arise from two pointers to
the same object then a test to avoid it might be desirable.
if (x != y)
mpz_set (x, y);
Note that it’s never worth introducing extra mpz_set calls just to get inplace op
erations. If a result should go to a particular variable then just direct it there and
let GMP take care of data movement.
Divisibility Testing (Small Integers)
mpz_divisible_ui_p and mpz_congruent_ui_p are the best functions for testing
whether an mpz_t is divisible by an individual small integer. They use an algorithm
which is faster than mpz_tdiv_ui, but which gives no useful information about the
actual remainder, only whether it’s zero (or a particular value).
However when testing divisibility by several small integers, it’s best to take a re
mainder modulo their product, to save multiprecision operations. For instance to
test whether a number is divisible by any of 23, 29 or 31 take a remainder modulo
23 29 31 = 20677 and then test that.
The division functions like mpz_tdiv_q_ui which give a quotient as well as a re
mainder are generally a little slower than the remainderonly functions like mpz_
tdiv_ui. If the quotient is only rarely wanted then it’s probably best to just take
a remainder and then go back and calculate the quotient if and when it’s wanted
(mpz_divexact_ui can be used if the remainder is zero).
Chapter 3: GMP Basics 23
Rational Arithmetic
The mpq functions operate on mpq_t values with no common factors in the numerator
and denominator. Common factors are checkedfor and cast out as necessary. In
general, cancelling factors every time is the best approach since it minimizes the
sizes for subsequent operations.
However, applications that know something about the factorization of the values
they’re working with might be able to avoid some of the GCDs used for canonical
ization, or swap them for divisions. For example when multiplying by a prime it’s
enough to check for factors of it in the denominator instead of doing a full GCD.
Or when forming a big product it might be known that very little cancellation will
be possible, and so canonicalization can be left to the end.
The mpq_numref and mpq_denref macros give access to the numerator and denom
inator to do things outside the scope of the supplied mpq functions. See Section 6.5
[Applying Integer Functions], page 46.
The canonical form for rationals allows mixedtype mpq_t and integer additions or
subtractions to be done directly with multiples of the denominator. This will be
somewhat faster than mpq_add. For example,
/* mpq increment */
mpz_add (mpq_numref(q), mpq_numref(q), mpq_denref(q));
/* mpq += unsigned long */
mpz_addmul_ui (mpq_numref(q), mpq_denref(q), 123UL);
/* mpq = mpz */
mpz_submul (mpq_numref(q), mpq_denref(q), z);
Number Sequences
Functions like mpz_fac_ui, mpz_fib_ui and mpz_bin_uiui are designed for calcu
lating isolated values. If a range of values is wanted it’s probably best to call to get
a starting point and iterate from there.
Text Input/Output
Hexadecimal or octal are suggested for input or output in text form. Powerof
2 bases like these can be converted much more efciently than other bases, like
decimal. For big numbers there’s usually nothing of particular interest to be seen
in the digits, so the base doesn’t matter much.
Maybe we can hope octal will one day become the normal base for everyday use, as
proposed by King Charles XII of Sweden and later reformers.
3.12 Debugging
Stack Overfow
Depending on the system, a segmentation violation or bus error might be the only
indication of stack overfow. See ‘enablealloca’ choices in Section 2.1 [Build
Options], page 3, for how to address this.
In new enough versions of GCC, ‘fstackcheck’ may be able to ensure
an overfow is recognised by the system before too much damage is done, or
‘fstacklimitsymbol’ or ‘fstacklimitregister’ may be able to add
checking if the system itself doesn’t do any (see Section “Options for Code
Generation” in Using the GNU Compiler Collection (GCC)). These options must
be added to the ‘CFLAGS’ used in the GMP build (see Section 2.1 [Build Options],
page 3), adding them just to an application will have no efect. Note also they’re a
slowdown, adding overhead to each function call and each stack allocation.
24 GNU MP 5.0.1
Heap Problems
The most likely cause of application problems with GMP is heap corruption. Fail
ing to init GMP variables will have unpredictable efects, and corruption arising
elsewhere in a program may well afect GMP. Initializing GMP variables more than
once or failing to clear them will cause memory leaks.
In all such cases a malloc debugger is recommended. On a GNU or BSD system
the standard C library malloc has some diagnostic facilities, see Section “Allocation
Debugging” in The GNU C Library Reference Manual, or ‘man 3 malloc’. Other
possibilities, in no particular order, include
http://www.inf.ethz.ch/personal/biere/projects/ccmalloc/
http://dmalloc.com/
http://www.perens.com/FreeSoftware/ (electric fence)
http://packages.debian.org/stable/devel/fda
http://www.gnupdate.org/components/leakbug/
http://people.redhat.com/~otaylor/memprof/
http://www.cbmamiga.demon.co.uk/mpatrol/
The GMP default allocation routines in ‘memory.c’ also have a simple sentinel
scheme which can be enabled with #define DEBUG in that fle. This is mainly de
signed for detecting bufer overruns during GMP development, but might fnd other
uses.
Stack Backtraces
On some systems the compiler options GMP uses by default can interfere with
debugging. In particular on x86 and 68k systems ‘fomitframepointer’ is used
and this generally inhibits stack backtracing. Recompiling without such options
may help while debugging, though the usual caveats about it potentially moving a
memory problem or hiding a compiler bug will apply.
GDB, the GNU Debugger
A sample ‘.gdbinit’ is included in the distribution, showing how to call some un
documented dump functions to print GMP variables from within GDB. Note that
these functions shouldn’t be used in fnal application code since they’re undocu
mented and may be subject to incompatible changes in future versions of GMP.
Source File Paths
GMP has multiple source fles with the same name, in diferent directories. For
example ‘mpz’, ‘mpq’ and ‘mpf’ each have an ‘init.c’. If the debugger can’t already
determine the right one it may help to build with absolute paths on each C fle.
One way to do that is to use a separate object directory with an absolute path to
the source directory.
cd /my/build/dir
/my/source/dir/gmp5.0.1/configure
This works via VPATH, and might require GNU make. Alternately it might be possible
to change the .c.lo rules appropriately.
Assertion Checking
The build option ‘enableassert’ is available to add some consistency checks
to the library (see Section 2.1 [Build Options], page 3). These are likely to be of
limited value to most applications. Assertion failures are just as likely to indicate
memory corruption as a library or compiler bug.
Applications using the lowlevel mpn functions, however, will beneft from
‘enableassert’ since it adds checks on the parameters of most such functions,
many of which have subtle restrictions on their usage. Note however that only the
Chapter 3: GMP Basics 25
generic C code has checks, not the assembly code, so CPU ‘none’ should be used
for maximum checking.
Temporary Memory Checking
The build option ‘enablealloca=debug’ arranges that each block of temporary
memory in GMP is allocated with a separate call to malloc (or the allocation
function set with mp_set_memory_functions).
This can help a malloc debugger detect accesses outside the intended bounds, or
detect memory not released. In a normal build, on the other hand, temporary
memory is allocated in blocks which GMP divides up for its own use, or may be
allocated with a compiler builtin alloca which will go nowhere near any malloc
debugger hooks.
Maximum Debuggability
To summarize the above, a GMP build for maximum debuggability would be
./configure disableshared enableassert \
enablealloca=debug host=none CFLAGS=g
For C++, add ‘enablecxx CXXFLAGS=g’.
Checker The GCC checker (http://savannah.nongnu.org/projects/checker/) can be
used with GMP. It contains a stub library which means GMP applications compiled
with checker can use a normal GMP build.
A build of GMP with checking within GMP itself can be made. This will run very
very slowly. On GNU/Linux for example,
./configure host=nonepclinuxgnu CC=checkergcc
‘host=none’ must be used, since the GMP assembly code doesn’t support the
checking scheme. The GMP C++ features cannot be used, since current versions of
checker (0.9.9.1) don’t yet support the standard C++ library.
Valgrind The valgrind program (http://valgrind.org/) is a memory checker for x86s. It
translates and emulates machine instructions to do strong checks for uninitialized
data (at the level of individual bits), memory accesses through bad pointers, and
memory leaks.
Recent versions of Valgrind are getting support for MMX and SSE/SSE2 instruc
tions, for past versions GMP will need to be confgured not to use those, ie. for an
x86 without them (for instance plain ‘i486’).
Other Problems
Any suspected bug in GMP itself should be isolated to make sure it’s not an appli
cation problem, see Chapter 4 [Reporting Bugs], page 28.
3.13 Profling
Running a program under a profler is a good way to fnd where it’s spending most time and
where improvements can be best sought. The profling choices for a GMP build are as follows.
‘disableprofiling’
The default is to add nothing special for profling.
It should be possible to just compile the mainline of a program with p and use prof
to get a profle consisting of timerbased sampling of the program counter. Most of
the GMP assembly code has the necessary symbol information.
This approach has the advantage of minimizing interference with normal program
operation, but on most systems the resolution of the sampling is quite low (10
milliseconds for instance), requiring long runs to get accurate information.
26 GNU MP 5.0.1
‘enableprofiling=prof’
Build with support for the system prof, which means ‘p’ added to the ‘CFLAGS’.
This provides call counting in addition to program counter sampling, which allows
the most frequently called routines to be identifed, and an average time spent in
each routine to be determined.
The x86 assembly code has support for this option, but on other processors the
assembly routines will be as if compiled without ‘p’ and therefore won’t appear in
the call counts.
On some systems, such as GNU/Linux, ‘p’ in fact means ‘pg’ and in this case
‘enableprofiling=gprof’ described below should be used instead.
‘enableprofiling=gprof’
Build with support for gprof, which means ‘pg’ added to the ‘CFLAGS’.
This provides call graph construction in addition to call counting and program
counter sampling, which makes it possible to count calls coming from diferent loca
tions. For example the number of calls to mpn_mul from mpz_mul versus the number
from mpf_mul. The program counter sampling is still fat though, so only a total
time in mpn_mul would be accumulated, not a separate amount for each call site.
The x86 assembly code has support for this option, but on other processors the
assembly routines will be as if compiled without ‘pg’ and therefore not be included
in the call counts.
On x86 and m68k systems ‘pg’ and ‘fomitframepointer’ are incompatible, so
the latter is omitted from the default fags in that case, which might result in poorer
code generation.
Incidentally, it should be possible to use the gprof program with a plain
‘enableprofiling=prof’ build. But in that case only the ‘gprof p’ fat profle
and call counts can be expected to be valid, not the ‘gprof q’ call graph.
‘enableprofiling=instrument’
Build with the GCC option ‘finstrumentfunctions’ added to the ‘CFLAGS’ (see
Section “Options for Code Generation” in Using the GNU Compiler Collection
(GCC)).
This inserts special instrumenting calls at the start and end of each function, allowing
exact timing and full call graph construction.
This instrumenting is not normally a standard system feature and will require sup
port from an external library, such as
http://sourceforge.net/projects/fnccheck/
This should be included in ‘LIBS’ during the GMP confgure so that test programs
will link. For example,
./configure enableprofiling=instrument LIBS=lfc
On a GNU system the C library provides dummy instrumenting functions, so pro
grams compiled with this option will link. In this case it’s only necessary to ensure
the correct library is added when linking an application.
The x86 assembly code supports this option, but on other processors the assembly
routines will be as if compiled without ‘finstrumentfunctions’ meaning time
spent in them will efectively be attributed to their caller.
3.14 Autoconf
Autoconf based applications can easily check whether GMP is installed. The only thing to be
noted is that GMP library symbols from version 3 onwards have prefxes like __gmpz. The
following therefore would be a simple test,
Chapter 3: GMP Basics 27
AC_CHECK_LIB(gmp, __gmpz_init)
This just uses the default AC_CHECK_LIB actions for found or not found, but an application that
must have GMP would want to generate an error if not found. For example,
AC_CHECK_LIB(gmp, __gmpz_init, ,
[AC_MSG_ERROR([GNU MP not found, see http://gmplib.org/])])
If functions added in some particular version of GMP are required, then one of those can be
used when checking. For example mpz_mul_si was added in GMP 3.1,
AC_CHECK_LIB(gmp, __gmpz_mul_si, ,
[AC_MSG_ERROR(
[GNU MP not found, or not 3.1 or up, see http://gmplib.org/])])
An alternative would be to test the version number in ‘gmp.h’ using say AC_EGREP_CPP. That
would make it possible to test the exact version, if some particular subminor release is known
to be necessary.
In general it’s recommended that applications should simply demand a new enough GMP rather
than trying to provide supplements for features not available in past versions.
Occasionally an application will need or want to know the size of a type at confguration or
preprocessing time, not just with sizeof in the code. This can be done in the normal way
with mp_limb_t etc, but GMP 4.0 or up is best for this, since prior versions needed certain ‘D’
defnes on systems using a long long limb. The following would suit Autoconf 2.50 or up,
AC_CHECK_SIZEOF(mp_limb_t, , [#include <gmp.h>])
3.15 Emacs
CH CI (infolookupsymbol) is a good way to fnd documentation on C functions while
editing (see Section “Info Documentation Lookup” in The Emacs Editor).
The GMP manual can be included in such lookups by putting the following in your ‘.emacs’,
(evalafterload "infolook"
’(let ((modevalue (assoc ’cmode (assoc ’symbol infolookupalist))))
(setcar (nthcdr 3 modevalue)
(cons ’("(gmp)Function Index" nil "^ .* " "\\>")
(nth 3 modevalue)))))
28 GNU MP 5.0.1
4 Reporting Bugs
If you think you have found a bug in the GMP library, please investigate it and report it. We
have made this library available to you, and it is not too much to ask you to report the bugs
you fnd.
Before you report a bug, check it’s not already addressed in Section 2.5 [Known Build Problems],
page 14, or perhaps Section 2.4 [Notes for Particular Systems], page 12. You may also want to
check http://gmplib.org/ for patches for this release.
Please include the following in any report,
• The GMP version number, and if prepackaged or patched then say so.
• A test program that makes it possible for us to reproduce the bug. Include instructions on
how to run the program.
• A description of what is wrong. If the results are incorrect, in what way. If you get a crash,
say so.
• If you get a crash, include a stack backtrace from the debugger if it’s informative (‘where’
in gdb, or ‘$C’ in adb).
• Please do not send core dumps, executables or straces.
• The confguration options you used when building GMP, if any.
• The name of the compiler and its version. For gcc, get the version with ‘gcc v’, otherwise
perhaps ‘what ‘which cc‘’, or similar.
• The output from running ‘uname a’.
• The output from running ‘./config.guess’, and from running ‘./configfsf.guess’
(might be the same).
• If the bug is related to ‘configure’, then the compressed contents of ‘config.log’.
• If the bug is related to an ‘asm’ fle not assembling, then the contents of ‘config.m4’ and
the ofending line or lines from the temporary ‘mpn/tmp<file>.s’.
Please make an efort to produce a selfcontained report, with something defnite that can be
tested or debugged. Vague queries or piecemeal messages are difcult to act on and don’t help
the development efort.
It is not uncommon that an observed problem is actually due to a bug in the compiler; the GMP
code tends to explore interesting corners in compilers.
If your bug report is good, we will do our best to help you get a corrected version of the library;
if the bug report is poor, we won’t do anything about it (except maybe ask you to send a better
report).
Send your report to: gmpbugs@gmplib.org.
If you think something in this manual is unclear, or downright incorrect, or if the language needs
to be improved, please send a note to the same address.
Chapter 5: Integer Functions 29
5 Integer Functions
This chapter describes the GMP functions for performing integer arithmetic. These functions
start with the prefx mpz_.
GMP integers are stored in objects of type mpz_t.
5.1 Initialization Functions
The functions for integer arithmetic assume that all integer objects are initialized. You do that
by calling the function mpz_init. For example,
{
mpz_t integ;
mpz_init (integ);
...
mpz_add (integ, ...);
...
mpz_sub (integ, ...);
/* Unless the program is about to exit, do ... */
mpz_clear (integ);
}
As you can see, you can store new values any number of times, once an object is initialized.
[Function] void mpz_init (mpz t x)
Initialize x, and set its value to 0.
[Function] void mpz_inits (mpz t x, ...)
Initialize a NULLterminated list of mpz_t variables, and set their values to 0.
[Function] void mpz_init2 (mpz t x, mp bitcnt t n)
Initialize x, with space for nbit numbers, and set its value to 0. Calling this function instead
of mpz_init or mpz_inits is never necessary; reallocation is handled automatically by GMP
when needed.
n is only the initial space, x will grow automatically in the normal way, if necessary, for sub
sequent values stored. mpz_init2 makes it possible to avoid such reallocations if a maximum
size is known in advance.
[Function] void mpz_clear (mpz t x)
Free the space occupied by x. Call this function for all mpz_t variables when you are done
with them.
[Function] void mpz_clears (mpz t x, ...)
Free the space occupied by a NULLterminated list of mpz_t variables.
[Function] void mpz_realloc2 (mpz t x, mp bitcnt t n)
Change the space allocated for x to n bits. The value in x is preserved if it fts, or is set to
0 if not.
Calling this function is never necessary; reallocation is handled automatically by GMP when
needed. But this function can be used to increase the space for a variable in order to avoid
repeated automatic reallocations, or to decrease it to give memory back to the heap.
30 GNU MP 5.0.1
5.2 Assignment Functions
These functions assign new values to already initialized integers (see Section 5.1 [Initializing
Integers], page 29).
[Function] void mpz_set (mpz t rop, mpz t op)
[Function] void mpz_set_ui (mpz t rop, unsigned long int op)
[Function] void mpz_set_si (mpz t rop, signed long int op)
[Function] void mpz_set_d (mpz t rop, double op)
[Function] void mpz_set_q (mpz t rop, mpq t op)
[Function] void mpz_set_f (mpz t rop, mpf t op)
Set the value of rop from op.
mpz_set_d, mpz_set_q and mpz_set_f truncate op to make it an integer.
[Function] int mpz_set_str (mpz t rop, char *str, int base)
Set the value of rop from str, a nullterminated C string in base base. White space is allowed
in the string, and is simply ignored.
The base may vary from 2 to 62, or if base is 0, then the leading characters are used: 0x and
0X for hexadecimal, 0b and 0B for binary, 0 for octal, or decimal otherwise.
For bases up to 36, case is ignored; uppercase and lowercase letters have the same value. For
bases 37 to 62, uppercase letter represent the usual 10..35 while lowercase letter represent
36..61.
This function returns 0 if the entire string is a valid number in base base. Otherwise it returns
−1.
[Function] void mpz_swap (mpz t rop1, mpz t rop2)
Swap the values rop1 and rop2 efciently.
5.3 Combined Initialization and Assignment Functions
For convenience, GMP provides a parallel series of initializeandset functions which initialize the
output and then store the value there. These functions’ names have the form mpz_init_set...
Here is an example of using one:
{
mpz_t pie;
mpz_init_set_str (pie, "3141592653589793238462643383279502884", 10);
...
mpz_sub (pie, ...);
...
mpz_clear (pie);
}
Once the integer has been initialized by any of the mpz_init_set... functions, it can be used
as the source or destination operand for the ordinary integer functions. Don’t use an initialize
andset function on a variable already initialized!
[Function] void mpz_init_set (mpz t rop, mpz t op)
[Function] void mpz_init_set_ui (mpz t rop, unsigned long int op)
[Function] void mpz_init_set_si (mpz t rop, signed long int op)
Chapter 5: Integer Functions 31
[Function] void mpz_init_set_d (mpz t rop, double op)
Initialize rop with limb space and set the initial numeric value from op.
[Function] int mpz_init_set_str (mpz t rop, char *str, int base)
Initialize rop and set its value like mpz_set_str (see its documentation above for details).
If the string is a correct base base number, the function returns 0; if an error occurs it returns
−1. rop is initialized even if an error occurs. (I.e., you have to call mpz_clear for it.)
5.4 Conversion Functions
This section describes functions for converting GMP integers to standard C types. Functions
for converting to GMP integers are described in Section 5.2 [Assigning Integers], page 30 and
Section 5.12 [I/O of Integers], page 39.
[Function] unsigned long int mpz_get_ui (mpz t op)
Return the value of op as an unsigned long.
If op is too big to ft an unsigned long then just the least signifcant bits that do ft are
returned. The sign of op is ignored, only the absolute value is used.
[Function] signed long int mpz_get_si (mpz t op)
If op fts into a signed long int return the value of op. Otherwise return the least signifcant
part of op, with the same sign as op.
If op is too big to ft in a signed long int, the returned result is probably not very useful.
To fnd out if the value will ft, use the function mpz_fits_slong_p.
[Function] double mpz_get_d (mpz t op)
Convert op to a double, truncating if necessary (ie. rounding towards zero).
If the exponent from the conversion is too big, the result is system dependent. An infnity is
returned where available. A hardware overfow trap may or may not occur.
[Function] double mpz_get_d_2exp (signed long int *exp, mpz t op)
Convert op to a double, truncating if necessary (ie. rounding towards zero), and returning
the exponent separately.
The return value is in the range 0.5 ≤ [d[ < 1 and the exponent is stored to *exp. d ∗ 2
exp
is
the (truncated) op value. If op is zero, the return is 0.0 and 0 is stored to *exp.
This is similar to the standard C frexp function (see Section “Normalization Functions” in
The GNU C Library Reference Manual).
[Function] char * mpz_get_str (char *str, int base, mpz t op)
Convert op to a string of digits in base base. The base argument may vary from 2 to 62 or
from −2 to −36.
For base in the range 2..36, digits and lowercase letters are used; for −2..−36, digits and
uppercase letters are used; for 37..62, digits, uppercase letters, and lowercase letters (in
that signifcance order) are used.
If str is NULL, the result string is allocated using the current allocation function (see
Chapter 14 [Custom Allocation], page 86). The block will be strlen(str)+1 bytes, that
being exactly enough for the string and nullterminator.
32 GNU MP 5.0.1
If str is not NULL, it should point to a block of storage large enough for the result, that being
mpz_sizeinbase (op, base) + 2. The two extra bytes are for a possible minus sign, and the
nullterminator.
A pointer to the result string is returned, being either the allocated block, or the given str.
5.5 Arithmetic Functions
[Function] void mpz_add (mpz t rop, mpz t op1, mpz t op2)
[Function] void mpz_add_ui (mpz t rop, mpz t op1, unsigned long int op2)
Set rop to op1 + op2.
[Function] void mpz_sub (mpz t rop, mpz t op1, mpz t op2)
[Function] void mpz_sub_ui (mpz t rop, mpz t op1, unsigned long int op2)
[Function] void mpz_ui_sub (mpz t rop, unsigned long int op1, mpz t op2)
Set rop to op1 − op2.
[Function] void mpz_mul (mpz t rop, mpz t op1, mpz t op2)
[Function] void mpz_mul_si (mpz t rop, mpz t op1, long int op2)
[Function] void mpz_mul_ui (mpz t rop, mpz t op1, unsigned long int op2)
Set rop to op1 op2.
[Function] void mpz_addmul (mpz t rop, mpz t op1, mpz t op2)
[Function] void mpz_addmul_ui (mpz t rop, mpz t op1, unsigned long int op2)
Set rop to rop + op1 op2.
[Function] void mpz_submul (mpz t rop, mpz t op1, mpz t op2)
[Function] void mpz_submul_ui (mpz t rop, mpz t op1, unsigned long int op2)
Set rop to rop −op1 op2.
[Function] void mpz_mul_2exp (mpz t rop, mpz t op1, mp bitcnt t op2)
Set rop to op1 2
op2
. This operation can also be defned as a left shift by op2 bits.
[Function] void mpz_neg (mpz t rop, mpz t op)
Set rop to −op.
[Function] void mpz_abs (mpz t rop, mpz t op)
Set rop to the absolute value of op.
5.6 Division Functions
Division is undefned if the divisor is zero. Passing a zero divisor to the division or modulo
functions (including the modular powering functions mpz_powm and mpz_powm_ui), will cause an
intentional division by zero. This lets a program handle arithmetic exceptions in these functions
the same way as for normal C int arithmetic.
[Function] void mpz_cdiv_q (mpz t q, mpz t n, mpz t d)
[Function] void mpz_cdiv_r (mpz t r, mpz t n, mpz t d)
[Function] void mpz_cdiv_qr (mpz t q, mpz t r, mpz t n, mpz t d)
[Function] unsigned long int mpz_cdiv_q_ui (mpz t q, mpz t n,
unsigned long int d)
[Function] unsigned long int mpz_cdiv_r_ui (mpz t r, mpz t n,
unsigned long int d)
Chapter 5: Integer Functions 33
[Function] unsigned long int mpz_cdiv_qr_ui (mpz t q, mpz t r, mpz t n,
unsigned long int d)
[Function] unsigned long int mpz_cdiv_ui (mpz t n, unsigned long int d)
[Function] void mpz_cdiv_q_2exp (mpz t q, mpz t n, mp bitcnt t b)
[Function] void mpz_cdiv_r_2exp (mpz t r, mpz t n, mp bitcnt t b)
[Function] void mpz_fdiv_q (mpz t q, mpz t n, mpz t d)
[Function] void mpz_fdiv_r (mpz t r, mpz t n, mpz t d)
[Function] void mpz_fdiv_qr (mpz t q, mpz t r, mpz t n, mpz t d)
[Function] unsigned long int mpz_fdiv_q_ui (mpz t q, mpz t n,
unsigned long int d)
[Function] unsigned long int mpz_fdiv_r_ui (mpz t r, mpz t n,
unsigned long int d)
[Function] unsigned long int mpz_fdiv_qr_ui (mpz t q, mpz t r, mpz t n,
unsigned long int d)
[Function] unsigned long int mpz_fdiv_ui (mpz t n, unsigned long int d)
[Function] void mpz_fdiv_q_2exp (mpz t q, mpz t n, mp bitcnt t b)
[Function] void mpz_fdiv_r_2exp (mpz t r, mpz t n, mp bitcnt t b)
[Function] void mpz_tdiv_q (mpz t q, mpz t n, mpz t d)
[Function] void mpz_tdiv_r (mpz t r, mpz t n, mpz t d)
[Function] void mpz_tdiv_qr (mpz t q, mpz t r, mpz t n, mpz t d)
[Function] unsigned long int mpz_tdiv_q_ui (mpz t q, mpz t n,
unsigned long int d)
[Function] unsigned long int mpz_tdiv_r_ui (mpz t r, mpz t n,
unsigned long int d)
[Function] unsigned long int mpz_tdiv_qr_ui (mpz t q, mpz t r, mpz t n,
unsigned long int d)
[Function] unsigned long int mpz_tdiv_ui (mpz t n, unsigned long int d)
[Function] void mpz_tdiv_q_2exp (mpz t q, mpz t n, mp bitcnt t b)
[Function] void mpz_tdiv_r_2exp (mpz t r, mpz t n, mp bitcnt t b)
Divide n by d, forming a quotient q and/or remainder r. For the 2exp functions, d = 2
b
. The
rounding is in three styles, each suiting diferent applications.
• cdiv rounds q up towards +∞, and r will have the opposite sign to d. The c stands for
“ceil”.
• fdiv rounds q down towards −∞, and r will have the same sign as d. The f stands for
“foor”.
• tdiv rounds q towards zero, and r will have the same sign as n. The t stands for
“truncate”.
In all cases q and r will satisfy n = qd + r, and r will satisfy 0 ≤ [r[ < [d[.
The q functions calculate only the quotient, the r functions only the remainder, and the qr
functions calculate both. Note that for qr the same variable cannot be passed for both q and
r, or results will be unpredictable.
For the ui variants the return value is the remainder, and in fact returning the remainder is
all the div_ui functions do. For tdiv and cdiv the remainder can be negative, so for those
the return value is the absolute value of the remainder.
For the 2exp variants the divisor is 2
b
. These functions are implemented as right shifts and
bit masks, but of course they round the same as the other functions.
34 GNU MP 5.0.1
For positive n both mpz_fdiv_q_2exp and mpz_tdiv_q_2exp are simple bitwise right shifts.
For negative n, mpz_fdiv_q_2exp is efectively an arithmetic right shift treating n as twos
complement the same as the bitwise logical functions do, whereas mpz_tdiv_q_2exp efec
tively treats n as sign and magnitude.
[Function] void mpz_mod (mpz t r, mpz t n, mpz t d)
[Function] unsigned long int mpz_mod_ui (mpz t r, mpz t n, unsigned long int d)
Set r to n mod d. The sign of the divisor is ignored; the result is always nonnegative.
mpz_mod_ui is identical to mpz_fdiv_r_ui above, returning the remainder as well as setting
r. See mpz_fdiv_ui above if only the return value is wanted.
[Function] void mpz_divexact (mpz t q, mpz t n, mpz t d)
[Function] void mpz_divexact_ui (mpz t q, mpz t n, unsigned long d)
Set q to n/d. These functions produce correct results only when it is known in advance that
d divides n.
These routines are much faster than the other division functions, and are the best choice
when exact division is known to occur, for example reducing a rational to lowest terms.
[Function] int mpz_divisible_p (mpz t n, mpz t d)
[Function] int mpz_divisible_ui_p (mpz t n, unsigned long int d)
[Function] int mpz_divisible_2exp_p (mpz t n, mp bitcnt t b)
Return nonzero if n is exactly divisible by d, or in the case of mpz_divisible_2exp_p by 2
b
.
n is divisible by d if there exists an integer q satisfying n = qd. Unlike the other division
functions, d = 0 is accepted and following the rule it can be seen that only 0 is considered
divisible by 0.
[Function] int mpz_congruent_p (mpz t n, mpz t c, mpz t d)
[Function] int mpz_congruent_ui_p (mpz t n, unsigned long int c, unsigned long int d)
[Function] int mpz_congruent_2exp_p (mpz t n, mpz t c, mp bitcnt t b)
Return nonzero if n is congruent to c modulo d, or in the case of mpz_congruent_2exp_p
modulo 2
b
.
n is congruent to c mod d if there exists an integer q satisfying n = c +qd. Unlike the other
division functions, d = 0 is accepted and following the rule it can be seen that n and c are
considered congruent mod 0 only when exactly equal.
5.7 Exponentiation Functions
[Function] void mpz_powm (mpz t rop, mpz t base, mpz t exp, mpz t mod)
[Function] void mpz_powm_ui (mpz t rop, mpz t base, unsigned long int exp, mpz t
mod)
Set rop to base
exp
mod mod.
Negative exp is supported if an inverse base
−1
mod mod exists (see mpz_invert in Section 5.9
[Number Theoretic Functions], page 35). If an inverse doesn’t exist then a divide by zero is
raised.
[Function] void mpz_powm_sec (mpz t rop, mpz t base, mpz t exp, mpz t mod)
Set rop to base
exp
mod mod.
It is required that exp > 0 and that mod is odd.
Chapter 5: Integer Functions 35
This function is designed to take the same time and have the same cache access patterns
for any two samesize arguments, assuming that function arguments are placed at the same
position and that the machine state is identical upon function entry. This function is intended
for cryptographic purposes, where resilience to sidechannel attacks is desired.
[Function] void mpz_pow_ui (mpz t rop, mpz t base, unsigned long int exp)
[Function] void mpz_ui_pow_ui (mpz t rop, unsigned long int base, unsigned long int
exp)
Set rop to base
exp
. The case 0
0
yields 1.
5.8 Root Extraction Functions
[Function] int mpz_root (mpz t rop, mpz t op, unsigned long int n)
Set rop to
n
√
op, the truncated integer part of the nth root of op. Return nonzero if the
computation was exact, i.e., if op is rop to the nth power.
[Function] void mpz_rootrem (mpz t root, mpz t rem, mpz t u, unsigned long int n)
Set root to
n
√
u, the truncated integer part of the nth root of u. Set rem to the remainder,
(u −root
n
).
[Function] void mpz_sqrt (mpz t rop, mpz t op)
Set rop to
√
op, the truncated integer part of the square root of op.
[Function] void mpz_sqrtrem (mpz t rop1, mpz t rop2, mpz t op)
Set rop1 to
√
op, like mpz_sqrt. Set rop2 to the remainder (op − rop1
2
), which will be
zero if op is a perfect square.
If rop1 and rop2 are the same variable, the results are undefned.
[Function] int mpz_perfect_power_p (mpz t op)
Return nonzero if op is a perfect power, i.e., if there exist integers a and b, with b > 1, such
that op = a
b
.
Under this defnition both 0 and 1 are considered to be perfect powers. Negative values of
op are accepted, but of course can only be odd perfect powers.
[Function] int mpz_perfect_square_p (mpz t op)
Return nonzero if op is a perfect square, i.e., if the square root of op is an integer. Under
this defnition both 0 and 1 are considered to be perfect squares.
5.9 Number Theoretic Functions
[Function] int mpz_probab_prime_p (mpz t n, int reps)
Determine whether n is prime. Return 2 if n is defnitely prime, return 1 if n is probably
prime (without being certain), or return 0 if n is defnitely composite.
This function does some trial divisions, then some MillerRabin probabilistic primality tests.
reps controls how many such tests are done, 5 to 10 is a reasonable number, more will reduce
the chances of a composite being returned as “probably prime”.
MillerRabin and similar tests can be more properly called compositeness tests. Numbers
which fail are known to be composite but those which pass might be prime or might be
composite. Only a few composites pass, hence those which pass are considered probably
prime.
36 GNU MP 5.0.1
[Function] void mpz_nextprime (mpz t rop, mpz t op)
Set rop to the next prime greater than op.
This function uses a probabilistic algorithm to identify primes. For practical purposes it’s
adequate, the chance of a composite passing will be extremely small.
[Function] void mpz_gcd (mpz t rop, mpz t op1, mpz t op2)
Set rop to the greatest common divisor of op1 and op2. The result is always positive even if
one or both input operands are negative.
[Function] unsigned long int mpz_gcd_ui (mpz t rop, mpz t op1, unsigned long int
op2)
Compute the greatest common divisor of op1 and op2. If rop is not NULL, store the result
there.
If the result is small enough to ft in an unsigned long int, it is returned. If the result does
not ft, 0 is returned, and the result is equal to the argument op1. Note that the result will
always ft if op2 is nonzero.
[Function] void mpz_gcdext (mpz t g, mpz t s, mpz t t, mpz t a, mpz t b)
Set g to the greatest common divisor of a and b, and in addition set s and t to coefcients
satisfying as + bt = g. The value in g is always positive, even if one or both of a and b are
negative. The values in s and t are chosen such that [s[ ≤ [b[ and [t[ ≤ [a[.
If t is NULL then that value is not computed.
[Function] void mpz_lcm (mpz t rop, mpz t op1, mpz t op2)
[Function] void mpz_lcm_ui (mpz t rop, mpz t op1, unsigned long op2)
Set rop to the least common multiple of op1 and op2. rop is always positive, irrespective of
the signs of op1 and op2. rop will be zero if either op1 or op2 is zero.
[Function] int mpz_invert (mpz t rop, mpz t op1, mpz t op2)
Compute the inverse of op1 modulo op2 and put the result in rop. If the inverse exists, the
return value is nonzero and rop will satisfy 0 ≤ rop < op2. If an inverse doesn’t exist the
return value is zero and rop is undefned.
[Function] int mpz_jacobi (mpz t a, mpz t b)
Calculate the Jacobi symbol
a
b
. This is defned only for b odd.
[Function] int mpz_legendre (mpz t a, mpz t p)
Calculate the Legendre symbol
a
p
. This is defned only for p an odd positive prime, and
for such p it’s identical to the Jacobi symbol.
[Function] int mpz_kronecker (mpz t a, mpz t b)
[Function] int mpz_kronecker_si (mpz t a, long b)
[Function] int mpz_kronecker_ui (mpz t a, unsigned long b)
[Function] int mpz_si_kronecker (long a, mpz t b)
[Function] int mpz_ui_kronecker (unsigned long a, mpz t b)
Calculate the Jacobi symbol
a
b
with the Kronecker extension
a
2
=
2
a
when a odd, or
a
2
= 0 when a even.
When b is odd the Jacobi symbol and Kronecker symbol are identical, so mpz_kronecker_ui
etc can be used for mixed precision Jacobi symbols too.
Chapter 5: Integer Functions 37
For more information see Henri Cohen section 1.4.2 (see Appendix B [References], page 122),
or any number theory textbook. See also the example program ‘demos/qcn.c’ which uses
mpz_kronecker_ui.
[Function] mp_bitcnt_t mpz_remove (mpz t rop, mpz t op, mpz t f)
Remove all occurrences of the factor f from op and store the result in rop. The return value
is how many such occurrences were removed.
[Function] void mpz_fac_ui (mpz t rop, unsigned long int op)
Set rop to op!, the factorial of op.
[Function] void mpz_bin_ui (mpz t rop, mpz t n, unsigned long int k)
[Function] void mpz_bin_uiui (mpz t rop, unsigned long int n, unsigned long int k)
Compute the binomial coefcient
n
k
and store the result in rop. Negative values of n are
supported by mpz_bin_ui, using the identity
−n
k
= (−1)
k
n+k−1
k
, see Knuth volume 1
section 1.2.6 part G.
[Function] void mpz_fib_ui (mpz t fn, unsigned long int n)
[Function] void mpz_fib2_ui (mpz t fn, mpz t fnsub1, unsigned long int n)
mpz_fib_ui sets fn to to F
n
, the n’th Fibonacci number. mpz_fib2_ui sets fn to F
n
, and
fnsub1 to F
n−1
.
These functions are designed for calculating isolated Fibonacci numbers. When a sequence of
values is wanted it’s best to start with mpz_fib2_ui and iterate the defning F
n+1
= F
n
+F
n−1
or similar.
[Function] void mpz_lucnum_ui (mpz t ln, unsigned long int n)
[Function] void mpz_lucnum2_ui (mpz t ln, mpz t lnsub1, unsigned long int n)
mpz_lucnum_ui sets ln to to L
n
, the n’th Lucas number. mpz_lucnum2_ui sets ln to L
n
, and
lnsub1 to L
n−1
.
These functions are designed for calculating isolated Lucas numbers. When a sequence of
values is wanted it’s best to start with mpz_lucnum2_ui and iterate the defning L
n+1
=
L
n
+L
n−1
or similar.
The Fibonacci numbers and Lucas numbers are related sequences, so it’s never necessary
to call both mpz_fib2_ui and mpz_lucnum2_ui. The formulas for going from Fibonacci to
Lucas can be found in Section 16.7.5 [Lucas Numbers Algorithm], page 108, the reverse is
straightforward too.
5.10 Comparison Functions
[Function] int mpz_cmp (mpz t op1, mpz t op2)
[Function] int mpz_cmp_d (mpz t op1, double op2)
[Macro] int mpz_cmp_si (mpz t op1, signed long int op2)
[Macro] int mpz_cmp_ui (mpz t op1, unsigned long int op2)
Compare op1 and op2. Return a positive value if op1 > op2, zero if op1 = op2, or a negative
value if op1 < op2.
mpz_cmp_ui and mpz_cmp_si are macros and will evaluate their arguments more than once.
mpz_cmp_d can be called with an infnity, but results are undefned for a NaN.
[Function] int mpz_cmpabs (mpz t op1, mpz t op2)
[Function] int mpz_cmpabs_d (mpz t op1, double op2)
38 GNU MP 5.0.1
[Function] int mpz_cmpabs_ui (mpz t op1, unsigned long int op2)
Compare the absolute values of op1 and op2. Return a positive value if [op1[ > [op2[, zero
if [op1[ = [op2[, or a negative value if [op1[ < [op2[.
mpz_cmpabs_d can be called with an infnity, but results are undefned for a NaN.
[Macro] int mpz_sgn (mpz t op)
Return +1 if op > 0, 0 if op = 0, and −1 if op < 0.
This function is actually implemented as a macro. It evaluates its argument multiple times.
5.11 Logical and Bit Manipulation Functions
These functions behave as if twos complement arithmetic were used (although signmagnitude
is the actual implementation). The least signifcant bit is number 0.
[Function] void mpz_and (mpz t rop, mpz t op1, mpz t op2)
Set rop to op1 bitwiseand op2.
[Function] void mpz_ior (mpz t rop, mpz t op1, mpz t op2)
Set rop to op1 bitwise inclusiveor op2.
[Function] void mpz_xor (mpz t rop, mpz t op1, mpz t op2)
Set rop to op1 bitwise exclusiveor op2.
[Function] void mpz_com (mpz t rop, mpz t op)
Set rop to the one’s complement of op.
[Function] mp_bitcnt_t mpz_popcount (mpz t op)
If op ≥ 0, return the population count of op, which is the number of 1 bits in the binary
representation. If op < 0, the number of 1s is infnite, and the return value is the largest
possible mp_bitcnt_t.
[Function] mp_bitcnt_t mpz_hamdist (mpz t op1, mpz t op2)
If op1 and op2 are both ≥ 0 or both < 0, return the hamming distance between the two
operands, which is the number of bit positions where op1 and op2 have diferent bit values.
If one operand is ≥ 0 and the other < 0 then the number of bits diferent is infnite, and the
return value is the largest possible mp_bitcnt_t.
[Function] mp_bitcnt_t mpz_scan0 (mpz t op, mp bitcnt t starting_bit)
[Function] mp_bitcnt_t mpz_scan1 (mpz t op, mp bitcnt t starting_bit)
Scan op, starting from bit starting bit, towards more signifcant bits, until the frst 0 or 1 bit
(respectively) is found. Return the index of the found bit.
If the bit at starting bit is already what’s sought, then starting bit is returned.
If there’s no bit found, then the largest possible mp_bitcnt_t is returned. This will happen
in mpz_scan0 past the end of a negative number, or mpz_scan1 past the end of a nonnegative
number.
[Function] void mpz_setbit (mpz t rop, mp bitcnt t bit_index)
Set bit bit index in rop.
[Function] void mpz_clrbit (mpz t rop, mp bitcnt t bit_index)
Clear bit bit index in rop.
Chapter 5: Integer Functions 39
[Function] void mpz_combit (mpz t rop, mp bitcnt t bit_index)
Complement bit bit index in rop.
[Function] int mpz_tstbit (mpz t op, mp bitcnt t bit_index)
Test bit bit index in op and return 0 or 1 accordingly.
5.12 Input and Output Functions
Functions that perform input from a stdio stream, and functions that output to a stdio stream.
Passing a NULL pointer for a stream argument to any of these functions will make them read
from stdin and write to stdout, respectively.
When using any of these functions, it is a good idea to include ‘stdio.h’ before ‘gmp.h’, since
that will allow ‘gmp.h’ to defne prototypes for these functions.
[Function] size_t mpz_out_str (FILE *stream, int base, mpz t op)
Output op on stdio stream stream, as a string of digits in base base. The base argument may
vary from 2 to 62 or from −2 to −36.
For base in the range 2..36, digits and lowercase letters are used; for −2..−36, digits and
uppercase letters are used; for 37..62, digits, uppercase letters, and lowercase letters (in
that signifcance order) are used.
Return the number of bytes written, or if an error occurred, return 0.
[Function] size_t mpz_inp_str (mpz t rop, FILE *stream, int base)
Input a possibly whitespace preceded string in base base from stdio stream stream, and put
the read integer in rop.
The base may vary from 2 to 62, or if base is 0, then the leading characters are used: 0x and
0X for hexadecimal, 0b and 0B for binary, 0 for octal, or decimal otherwise.
For bases up to 36, case is ignored; uppercase and lowercase letters have the same value. For
bases 37 to 62, uppercase letter represent the usual 10..35 while lowercase letter represent
36..61.
Return the number of bytes read, or if an error occurred, return 0.
[Function] size_t mpz_out_raw (FILE *stream, mpz t op)
Output op on stdio stream stream, in raw binary format. The integer is written in a portable
format, with 4 bytes of size information, and that many bytes of limbs. Both the size and
the limbs are written in decreasing signifcance order (i.e., in bigendian).
The output can be read with mpz_inp_raw.
Return the number of bytes written, or if an error occurred, return 0.
The output of this can not be read by mpz_inp_raw from GMP 1, because of changes necessary
for compatibility between 32bit and 64bit machines.
[Function] size_t mpz_inp_raw (mpz t rop, FILE *stream)
Input from stdio stream stream in the format written by mpz_out_raw, and put the result in
rop. Return the number of bytes read, or if an error occurred, return 0.
This routine can read the output from mpz_out_raw also from GMP 1, in spite of changes
necessary for compatibility between 32bit and 64bit machines.
40 GNU MP 5.0.1
5.13 Random Number Functions
The random number functions of GMP come in two groups; older function that rely on a global
state, and newer functions that accept a state parameter that is read and modifed. Please see
the Chapter 9 [Random Number Functions], page 65 for more information on how to use and
not to use random number functions.
[Function] void mpz_urandomb (mpz t rop, gmp randstate t state, mp bitcnt t n)
Generate a uniformly distributed random integer in the range 0 to 2
n
−1, inclusive.
The variable state must be initialized by calling one of the gmp_randinit functions
(Section 9.1 [Random State Initialization], page 65) before invoking this function.
[Function] void mpz_urandomm (mpz t rop, gmp randstate t state, mpz t n)
Generate a uniform random integer in the range 0 to n −1, inclusive.
The variable state must be initialized by calling one of the gmp_randinit functions
(Section 9.1 [Random State Initialization], page 65) before invoking this function.
[Function] void mpz_rrandomb (mpz t rop, gmp randstate t state, mp bitcnt t n)
Generate a random integer with long strings of zeros and ones in the binary representation.
Useful for testing functions and algorithms, since this kind of random numbers have proven
to be more likely to trigger cornercase bugs. The random number will be in the range 0 to
2
n
−1, inclusive.
The variable state must be initialized by calling one of the gmp_randinit functions
(Section 9.1 [Random State Initialization], page 65) before invoking this function.
[Function] void mpz_random (mpz t rop, mp size t max_size)
Generate a random integer of at most max size limbs. The generated random number doesn’t
satisfy any particular requirements of randomness. Negative random numbers are generated
when max size is negative.
This function is obsolete. Use mpz_urandomb or mpz_urandomm instead.
[Function] void mpz_random2 (mpz t rop, mp size t max_size)
Generate a random integer of at most max size limbs, with long strings of zeros and ones
in the binary representation. Useful for testing functions and algorithms, since this kind of
random numbers have proven to be more likely to trigger cornercase bugs. Negative random
numbers are generated when max size is negative.
This function is obsolete. Use mpz_rrandomb instead.
5.14 Integer Import and Export
mpz_t variables can be converted to and from arbitrary words of binary data with the following
functions.
[Function] void mpz_import (mpz t rop, size t count, int order, size t size, int
endian, size t nails, const void *op)
Set rop from an array of word data at op.
The parameters specify the format of the data. count many words are read, each size bytes.
order can be 1 for most signifcant word frst or 1 for least signifcant frst. Within each
Chapter 5: Integer Functions 41
word endian can be 1 for most signifcant byte frst, 1 for least signifcant frst, or 0 for the
native endianness of the host CPU. The most signifcant nails bits of each word are skipped,
this can be 0 to use the full words.
There is no sign taken from the data, rop will simply be a positive integer. An application
can handle any sign itself, and apply it for instance with mpz_neg.
There are no data alignment restrictions on op, any address is allowed.
Here’s an example converting an array of unsigned long data, most signifcant element frst,
and host byte order within each value.
unsigned long a[20];
/* Initialize z and a */
mpz_import (z, 20, 1, sizeof(a[0]), 0, 0, a);
This example assumes the full sizeof bytes are used for data in the given type, which is
usually true, and certainly true for unsigned long everywhere we know of. However on Cray
vector systems it may be noted that short and int are always stored in 8 bytes (and with
sizeof indicating that) but use only 32 or 46 bits. The nails feature can account for this,
by passing for instance 8*sizeof(int)INT_BIT.
[Function] void * mpz_export (void *rop, size t *countp, int order, size t size, int
endian, size t nails, mpz t op)
Fill rop with word data from op.
The parameters specify the format of the data produced. Each word will be size bytes and
order can be 1 for most signifcant word frst or 1 for least signifcant frst. Within each
word endian can be 1 for most signifcant byte frst, 1 for least signifcant frst, or 0 for the
native endianness of the host CPU. The most signifcant nails bits of each word are unused
and set to zero, this can be 0 to produce full words.
The number of words produced is written to *countp, or countp can be NULL to discard the
count. rop must have enough space for the data, or if rop is NULL then a result array of
the necessary size is allocated using the current GMP allocation function (see Chapter 14
[Custom Allocation], page 86). In either case the return value is the destination used, either
rop or the allocated block.
If op is nonzero then the most signifcant word produced will be nonzero. If op is zero then
the count returned will be zero and nothing written to rop. If rop is NULL in this case, no
block is allocated, just NULL is returned.
The sign of op is ignored, just the absolute value is exported. An application can use mpz_sgn
to get the sign and handle it as desired. (see Section 5.10 [Integer Comparisons], page 37)
There are no data alignment restrictions on rop, any address is allowed.
When an application is allocating space itself the required size can be determined with a
calculation like the following. Since mpz_sizeinbase always returns at least 1, count here
will be at least one, which avoids any portability problems with malloc(0), though if z is
zero no space at all is actually needed (or written).
numb = 8*size  nail;
count = (mpz_sizeinbase (z, 2) + numb1) / numb;
p = malloc (count * size);
42 GNU MP 5.0.1
5.15 Miscellaneous Functions
[Function] int mpz_fits_ulong_p (mpz t op)
[Function] int mpz_fits_slong_p (mpz t op)
[Function] int mpz_fits_uint_p (mpz t op)
[Function] int mpz_fits_sint_p (mpz t op)
[Function] int mpz_fits_ushort_p (mpz t op)
[Function] int mpz_fits_sshort_p (mpz t op)
Return nonzero if the value of op fts in an unsigned long int, signed long int, unsigned
int, signed int, unsigned short int, or signed short int, respectively. Otherwise, re
turn zero.
[Macro] int mpz_odd_p (mpz t op)
[Macro] int mpz_even_p (mpz t op)
Determine whether op is odd or even, respectively. Return nonzero if yes, zero if no. These
macros evaluate their argument more than once.
[Function] size_t mpz_sizeinbase (mpz t op, int base)
Return the size of op measured in number of digits in the given base. base can vary from 2
to 62. The sign of op is ignored, just the absolute value is used. The result will be either
exact or 1 too big. If base is a power of 2, the result is always exact. If op is zero the return
value is always 1.
This function can be used to determine the space required when converting op to a string. The
right amount of allocation is normally two more than the value returned by mpz_sizeinbase,
one extra for a minus sign and one for the nullterminator.
It will be noted that mpz_sizeinbase(op,2) can be used to locate the most signifcant 1 bit
in op, counting from 1. (Unlike the bitwise functions which start from 0, See Section 5.11
[Logical and Bit Manipulation Functions], page 38.)
5.16 Special Functions
The functions in this section are for various special purposes. Most applications will not need
them.
[Function] void mpz_array_init (mpz t integer_array, mp size t array_size,
mp size t fixed_num_bits)
This is a special type of initialization. Fixed space of fxed num bits is allocated to each of
the array size integers in integer array. There is no way to free the storage allocated by this
function. Don’t call mpz_clear!
The integer array parameter is the frst mpz_t in the array. For example,
mpz_t arr[20000];
mpz_array_init (arr[0], 20000, 512);
This function is only intended for programs that create a large number of integers and need
to reduce memory usage by avoiding the overheads of allocating and reallocating lots of small
blocks. In normal programs this function is not recommended.
The space allocated to each integer by this function will not be automatically increased, unlike
the normal mpz_init, so an application must ensure it is sufcient for any value stored. The
following space requirements apply to various routines,
Chapter 5: Integer Functions 43
• mpz_abs, mpz_neg, mpz_set, mpz_set_si and mpz_set_ui need room for the value they
store.
• mpz_add, mpz_add_ui, mpz_sub and mpz_sub_ui need room for the larger of the two
operands, plus an extra mp_bits_per_limb.
• mpz_mul, mpz_mul_ui and mpz_mul_ui need room for the sum of the number of bits in
their operands, but each rounded up to a multiple of mp_bits_per_limb.
• mpz_swap can be used between two array variables, but not between an array and a
normal variable.
For other functions, or if in doubt, the suggestion is to calculate in a regular mpz_init variable
and copy the result to an array variable with mpz_set.
[Function] void * _mpz_realloc (mpz t integer, mp size t new_alloc)
Change the space for integer to new alloc limbs. The value in integer is preserved if it fts,
or is set to 0 if not. The return value is not useful to applications and should be ignored.
mpz_realloc2 is the preferred way to accomplish allocation changes like this. mpz_realloc2
and _mpz_realloc are the same except that _mpz_realloc takes its size in limbs.
[Function] mp_limb_t mpz_getlimbn (mpz t op, mp size t n)
Return limb number n from op. The sign of op is ignored, just the absolute value is used.
The least signifcant limb is number 0.
mpz_size can be used to fnd how many limbs make up op. mpz_getlimbn returns zero if n
is outside the range 0 to mpz_size(op)1.
[Function] size_t mpz_size (mpz t op)
Return the size of op measured in number of limbs. If op is zero, the returned value will be
zero.
44 GNU MP 5.0.1
6 Rational Number Functions
This chapter describes the GMP functions for performing arithmetic on rational numbers. These
functions start with the prefx mpq_.
Rational numbers are stored in objects of type mpq_t.
All rational arithmetic functions assume operands have a canonical form, and canonicalize their
result. The canonical from means that the denominator and the numerator have no common
factors, and that the denominator is positive. Zero has the unique representation 0/1.
Pure assignment functions do not canonicalize the assigned variable. It is the responsibility of
the user to canonicalize the assigned variable before any arithmetic operations are performed on
that variable.
[Function] void mpq_canonicalize (mpq t op)
Remove any factors that are common to the numerator and denominator of op, and make
the denominator positive.
6.1 Initialization and Assignment Functions
[Function] void mpq_init (mpq t x)
Initialize x and set it to 0/1. Each variable should normally only be initialized once, or at
least cleared out (using the function mpq_clear) between each initialization.
[Function] void mpq_inits (mpq t x, ...)
Initialize a NULLterminated list of mpq_t variables, and set their values to 0/1.
[Function] void mpq_clear (mpq t x)
Free the space occupied by x. Make sure to call this function for all mpq_t variables when
you are done with them.
[Function] void mpq_clears (mpq t x, ...)
Free the space occupied by a NULLterminated list of mpq_t variables.
[Function] void mpq_set (mpq t rop, mpq t op)
[Function] void mpq_set_z (mpq t rop, mpz t op)
Assign rop from op.
[Function] void mpq_set_ui (mpq t rop, unsigned long int op1, unsigned long int op2)
[Function] void mpq_set_si (mpq t rop, signed long int op1, unsigned long int op2)
Set the value of rop to op1/op2. Note that if op1 and op2 have common factors, rop has to
be passed to mpq_canonicalize before any operations are performed on rop.
[Function] int mpq_set_str (mpq t rop, char *str, int base)
Set rop from a nullterminated string str in the given base.
The string can be an integer like “41” or a fraction like “41/152”. The fraction must be
in canonical form (see Chapter 6 [Rational Number Functions], page 44), or if not then
mpq_canonicalize must be called.
The numerator and optional denominator are parsed the same as in mpz_set_str (see
Section 5.2 [Assigning Integers], page 30). White space is allowed in the string, and is simply
ignored. The base can vary from 2 to 62, or if base is 0 then the leading characters are used:
Chapter 6: Rational Number Functions 45
0x or 0X for hex, 0b or 0B for binary, 0 for octal, or decimal otherwise. Note that this is done
separately for the numerator and denominator, so for instance 0xEF/100 is 239/100, whereas
0xEF/0x100 is 239/256.
The return value is 0 if the entire string is a valid number, or −1 if not.
[Function] void mpq_swap (mpq t rop1, mpq t rop2)
Swap the values rop1 and rop2 efciently.
6.2 Conversion Functions
[Function] double mpq_get_d (mpq t op)
Convert op to a double, truncating if necessary (ie. rounding towards zero).
If the exponent from the conversion is too big or too small to ft a double then the result is
system dependent. For too big an infnity is returned when available. For too small 0.0 is
normally returned. Hardware overfow, underfow and denorm traps may or may not occur.
[Function] void mpq_set_d (mpq t rop, double op)
[Function] void mpq_set_f (mpq t rop, mpf t op)
Set rop to the value of op. There is no rounding, this conversion is exact.
[Function] char * mpq_get_str (char *str, int base, mpq t op)
Convert op to a string of digits in base base. The base may vary from 2 to 36. The string
will be of the form ‘num/den’, or if the denominator is 1 then just ‘num’.
If str is NULL, the result string is allocated using the current allocation function (see
Chapter 14 [Custom Allocation], page 86). The block will be strlen(str)+1 bytes, that
being exactly enough for the string and nullterminator.
If str is not NULL, it should point to a block of storage large enough for the result, that being
mpz_sizeinbase (mpq_numref(op), base)
+ mpz_sizeinbase (mpq_denref(op), base) + 3
The three extra bytes are for a possible minus sign, possible slash, and the nullterminator.
A pointer to the result string is returned, being either the allocated block, or the given str.
6.3 Arithmetic Functions
[Function] void mpq_add (mpq t sum, mpq t addend1, mpq t addend2)
Set sum to addend1 + addend2.
[Function] void mpq_sub (mpq t difference, mpq t minuend, mpq t subtrahend)
Set diference to minuend − subtrahend.
[Function] void mpq_mul (mpq t product, mpq t multiplier, mpq t multiplicand)
Set product to multiplier multiplicand.
[Function] void mpq_mul_2exp (mpq t rop, mpq t op1, mp bitcnt t op2)
Set rop to op1 2
op2
.
[Function] void mpq_div (mpq t quotient, mpq t dividend, mpq t divisor)
Set quotient to dividend/divisor.
46 GNU MP 5.0.1
[Function] void mpq_div_2exp (mpq t rop, mpq t op1, mp bitcnt t op2)
Set rop to op1/2
op2
.
[Function] void mpq_neg (mpq t negated_operand, mpq t operand)
Set negated operand to −operand.
[Function] void mpq_abs (mpq t rop, mpq t op)
Set rop to the absolute value of op.
[Function] void mpq_inv (mpq t inverted_number, mpq t number)
Set inverted number to 1/number. If the new denominator is zero, this routine will divide
by zero.
6.4 Comparison Functions
[Function] int mpq_cmp (mpq t op1, mpq t op2)
Compare op1 and op2. Return a positive value if op1 > op2, zero if op1 = op2, and a
negative value if op1 < op2.
To determine if two rationals are equal, mpq_equal is faster than mpq_cmp.
[Macro] int mpq_cmp_ui (mpq t op1, unsigned long int num2, unsigned long int den2)
[Macro] int mpq_cmp_si (mpq t op1, long int num2, unsigned long int den2)
Compare op1 and num2/den2. Return a positive value if op1 > num2/den2, zero if op1 =
num2/den2, and a negative value if op1 < num2/den2.
num2 and den2 are allowed to have common factors.
These functions are implemented as a macros and evaluate their arguments multiple times.
[Macro] int mpq_sgn (mpq t op)
Return +1 if op > 0, 0 if op = 0, and −1 if op < 0.
This function is actually implemented as a macro. It evaluates its arguments multiple times.
[Function] int mpq_equal (mpq t op1, mpq t op2)
Return nonzero if op1 and op2 are equal, zero if they are nonequal. Although mpq_cmp can
be used for the same purpose, this function is much faster.
6.5 Applying Integer Functions to Rationals
The set of mpq functions is quite small. In particular, there are few functions for either input
or output. The following functions give direct access to the numerator and denominator of an
mpq_t.
Note that if an assignment to the numerator and/or denominator could take an mpq_t out
of the canonical form described at the start of this chapter (see Chapter 6 [Rational Number
Functions], page 44) then mpq_canonicalize must be called before any other mpq functions are
applied to that mpq_t.
[Macro] mpz_t mpq_numref (mpq t op)
[Macro] mpz_t mpq_denref (mpq t op)
Return a reference to the numerator and denominator of op, respectively. The mpz functions
can be used on the result of these macros.
Chapter 6: Rational Number Functions 47
[Function] void mpq_get_num (mpz t numerator, mpq t rational)
[Function] void mpq_get_den (mpz t denominator, mpq t rational)
[Function] void mpq_set_num (mpq t rational, mpz t numerator)
[Function] void mpq_set_den (mpq t rational, mpz t denominator)
Get or set the numerator or denominator of a rational. These functions are equivalent to
calling mpz_set with an appropriate mpq_numref or mpq_denref. Direct use of mpq_numref
or mpq_denref is recommended instead of these functions.
6.6 Input and Output Functions
When using any of these functions, it’s a good idea to include ‘stdio.h’ before ‘gmp.h’, since
that will allow ‘gmp.h’ to defne prototypes for these functions.
Passing a NULL pointer for a stream argument to any of these functions will make them read
from stdin and write to stdout, respectively.
[Function] size_t mpq_out_str (FILE *stream, int base, mpq t op)
Output op on stdio stream stream, as a string of digits in base base. The base may vary from
2 to 36. Output is in the form ‘num/den’ or if the denominator is 1 then just ‘num’.
Return the number of bytes written, or if an error occurred, return 0.
[Function] size_t mpq_inp_str (mpq t rop, FILE *stream, int base)
Read a string of digits from stream and convert them to a rational in rop. Any initial white
space characters are read and discarded. Return the number of characters read (including
white space), or 0 if a rational could not be read.
The input can be a fraction like ‘17/63’ or just an integer like ‘123’. Reading stops at the
frst character not in this form, and white space is not permitted within the string. If the
input might not be in canonical form, then mpq_canonicalize must be called (see Chapter 6
[Rational Number Functions], page 44).
The base can be between 2 and 36, or can be 0 in which case the leading characters of the
string determine the base, ‘0x’ or ‘0X’ for hexadecimal, ‘0’ for octal, or decimal otherwise.
The leading characters are examined separately for the numerator and denominator of a
fraction, so for instance ‘0x10/11’ is 16/11, whereas ‘0x10/0x11’ is 16/17.
48 GNU MP 5.0.1
7 Floatingpoint Functions
GMP foating point numbers are stored in objects of type mpf_t and functions operating on
them have an mpf_ prefx.
The mantissa of each foat has a userselectable precision, limited only by available memory.
Each variable has its own precision, and that can be increased or decreased at any time.
The exponent of each foat is a fxed precision, one machine word on most systems. In the
current implementation the exponent is a count of limbs, so for example on a 32bit system this
means a range of roughly 2
−68719476768
to 2
68719476736
, or on a 64bit system this will be greater.
Note however mpf_get_str can only return an exponent which fts an mp_exp_t and currently
mpf_set_str doesn’t accept exponents bigger than a long.
Each variable keeps a size for the mantissa data actually in use. This means that if a foat is
exactly represented in only a few bits then only those bits will be used in a calculation, even if
the selected precision is high.
All calculations are performed to the precision of the destination variable. Each function is
defned to calculate with “infnite precision” followed by a truncation to the destination precision,
but of course the work done is only what’s needed to determine a result under that defnition.
The precision selected for a variable is a minimum value, GMP may increase it a little to facilitate
efcient calculation. Currently this means rounding up to a whole limb, and then sometimes
having a further partial limb, depending on the high limb of the mantissa. But applications
shouldn’t be concerned by such details.
The mantissa in stored in binary, as might be imagined from the fact precisions are expressed
in bits. One consequence of this is that decimal fractions like 0.1 cannot be represented exactly.
The same is true of plain IEEE double foats. This makes both highly unsuitable for calculations
involving money or other values that should be exact decimal fractions. (Suitably scaled integers,
or perhaps rationals, are better choices.)
mpf functions and variables have no special notion of infnity or notanumber, and applications
must take care not to overfow the exponent or results will be unpredictable. This might change
in a future release.
Note that the mpf functions are not intended as a smooth extension to IEEE P754 arithmetic.
In particular results obtained on one computer often difer from the results on a computer with
a diferent word size.
7.1 Initialization Functions
[Function] void mpf_set_default_prec (mp bitcnt t prec)
Set the default precision to be at least prec bits. All subsequent calls to mpf_init will use
this precision, but previously initialized variables are unafected.
[Function] mp_bitcnt_t mpf_get_default_prec (void)
Return the default precision actually used.
An mpf_t object must be initialized before storing the frst value in it. The functions mpf_init
and mpf_init2 are used for that purpose.
Chapter 7: Floatingpoint Functions 49
[Function] void mpf_init (mpf t x)
Initialize x to 0. Normally, a variable should be initialized once only or at least be cleared,
using mpf_clear, between initializations. The precision of x is undefned unless a default
precision has already been established by a call to mpf_set_default_prec.
[Function] void mpf_init2 (mpf t x, mp bitcnt t prec)
Initialize x to 0 and set its precision to be at least prec bits. Normally, a variable should be
initialized once only or at least be cleared, using mpf_clear, between initializations.
[Function] void mpf_inits (mpf t x, ...)
Initialize a NULLterminated list of mpf_t variables, and set their values to 0. The precision
of the initialized variables is undefned unless a default precision has already been established
by a call to mpf_set_default_prec.
[Function] void mpf_clear (mpf t x)
Free the space occupied by x. Make sure to call this function for all mpf_t variables when
you are done with them.
[Function] void mpf_clears (mpf t x, ...)
Free the space occupied by a NULLterminated list of mpf_t variables.
Here is an example on how to initialize foatingpoint variables:
{
mpf_t x, y;
mpf_init (x); /* use default precision */
mpf_init2 (y, 256); /* precision at least 256 bits */
...
/* Unless the program is about to exit, do ... */
mpf_clear (x);
mpf_clear (y);
}
The following three functions are useful for changing the precision during a calculation. A typical
use would be for adjusting the precision gradually in iterative algorithms like NewtonRaphson,
making the computation precision closely match the actual accurate part of the numbers.
[Function] mp_bitcnt_t mpf_get_prec (mpf t op)
Return the current precision of op, in bits.
[Function] void mpf_set_prec (mpf t rop, mp bitcnt t prec)
Set the precision of rop to be at least prec bits. The value in rop will be truncated to the
new precision.
This function requires a call to realloc, and so should not be used in a tight loop.
[Function] void mpf_set_prec_raw (mpf t rop, mp bitcnt t prec)
Set the precision of rop to be at least prec bits, without changing the memory allocated.
prec must be no more than the allocated precision for rop, that being the precision when rop
was initialized, or in the most recent mpf_set_prec.
The value in rop is unchanged, and in particular if it had a higher precision than prec it will
retain that higher precision. New values written to rop will use the new prec.
50 GNU MP 5.0.1
Before calling mpf_clear or the full mpf_set_prec, another mpf_set_prec_raw call must be
made to restore rop to its original allocated precision. Failing to do so will have unpredictable
results.
mpf_get_prec can be used before mpf_set_prec_raw to get the original allocated precision.
After mpf_set_prec_raw it refects the prec value set.
mpf_set_prec_raw is an efcient way to use an mpf_t variable at diferent precisions during
a calculation, perhaps to gradually increase precision in an iteration, or just to use various
diferent precisions for diferent purposes during a calculation.
7.2 Assignment Functions
These functions assign new values to already initialized foats (see Section 7.1 [Initializing Floats],
page 48).
[Function] void mpf_set (mpf t rop, mpf t op)
[Function] void mpf_set_ui (mpf t rop, unsigned long int op)
[Function] void mpf_set_si (mpf t rop, signed long int op)
[Function] void mpf_set_d (mpf t rop, double op)
[Function] void mpf_set_z (mpf t rop, mpz t op)
[Function] void mpf_set_q (mpf t rop, mpq t op)
Set the value of rop from op.
[Function] int mpf_set_str (mpf t rop, char *str, int base)
Set the value of rop from the string in str. The string is of the form ‘M@N’ or, if the base is 10
or less, alternatively ‘MeN’. ‘M’ is the mantissa and ‘N’ is the exponent. The mantissa is always
in the specifed base. The exponent is either in the specifed base or, if base is negative, in
decimal. The decimal point expected is taken from the current locale, on systems providing
localeconv.
The argument base may be in the ranges 2 to 62, or −62 to −2. Negative values are used to
specify that the exponent is in decimal.
For bases up to 36, case is ignored; uppercase and lowercase letters have the same value; for
bases 37 to 62, uppercase letter represent the usual 10..35 while lowercase letter represent
36..61.
Unlike the corresponding mpz function, the base will not be determined from the leading
characters of the string if base is 0. This is so that numbers like ‘0.23’ are not interpreted
as octal.
White space is allowed in the string, and is simply ignored. [This is not really true; white
space is ignored in the beginning of the string and within the mantissa, but not in other
places, such as after a minus sign or in the exponent. We are considering changing the
defnition of this function, making it fail when there is any whitespace in the input, since
that makes a lot of sense. Please tell us your opinion about this change. Do you really want
it to accept "3 14" as meaning 314 as it does now?]
This function returns 0 if the entire string is a valid number in base base. Otherwise it returns
−1.
[Function] void mpf_swap (mpf t rop1, mpf t rop2)
Swap rop1 and rop2 efciently. Both the values and the precisions of the two variables are
swapped.
Chapter 7: Floatingpoint Functions 51
7.3 Combined Initialization and Assignment Functions
For convenience, GMP provides a parallel series of initializeandset functions which initialize the
output and then store the value there. These functions’ names have the form mpf_init_set...
Once the foat has been initialized by any of the mpf_init_set... functions, it can be used as
the source or destination operand for the ordinary foat functions. Don’t use an initializeandset
function on a variable already initialized!
[Function] void mpf_init_set (mpf t rop, mpf t op)
[Function] void mpf_init_set_ui (mpf t rop, unsigned long int op)
[Function] void mpf_init_set_si (mpf t rop, signed long int op)
[Function] void mpf_init_set_d (mpf t rop, double op)
Initialize rop and set its value from op.
The precision of rop will be taken from the active default precision, as set by mpf_set_
default_prec.
[Function] int mpf_init_set_str (mpf t rop, char *str, int base)
Initialize rop and set its value from the string in str. See mpf_set_str above for details on
the assignment operation.
Note that rop is initialized even if an error occurs. (I.e., you have to call mpf_clear for it.)
The precision of rop will be taken from the active default precision, as set by mpf_set_
default_prec.
7.4 Conversion Functions
[Function] double mpf_get_d (mpf t op)
Convert op to a double, truncating if necessary (ie. rounding towards zero).
If the exponent in op is too big or too small to ft a double then the result is system dependent.
For too big an infnity is returned when available. For too small 0.0 is normally returned.
Hardware overfow, underfow and denorm traps may or may not occur.
[Function] double mpf_get_d_2exp (signed long int *exp, mpf t op)
Convert op to a double, truncating if necessary (ie. rounding towards zero), and with an
exponent returned separately.
The return value is in the range 0.5 ≤ [d[ < 1 and the exponent is stored to *exp. d ∗ 2
exp
is
the (truncated) op value. If op is zero, the return is 0.0 and 0 is stored to *exp.
This is similar to the standard C frexp function (see Section “Normalization Functions” in
The GNU C Library Reference Manual).
[Function] long mpf_get_si (mpf t op)
[Function] unsigned long mpf_get_ui (mpf t op)
Convert op to a long or unsigned long, truncating any fraction part. If op is too big for
the return type, the result is undefned.
See also mpf_fits_slong_p and mpf_fits_ulong_p (see Section 7.8 [Miscellaneous Float
Functions], page 54).
52 GNU MP 5.0.1
[Function] char * mpf_get_str (char *str, mp exp t *expptr, int base, size t
n_digits, mpf t op)
Convert op to a string of digits in base base. The base argument may vary from 2 to 62 or
from −2 to −36. Up to n digits digits will be generated. Trailing zeros are not returned.
No more digits than can be accurately represented by op are ever generated. If n digits is 0
then that accurate maximum number of digits are generated.
For base in the range 2..36, digits and lowercase letters are used; for −2..−36, digits and
uppercase letters are used; for 37..62, digits, uppercase letters, and lowercase letters (in
that signifcance order) are used.
If str is NULL, the result string is allocated using the current allocation function (see
Chapter 14 [Custom Allocation], page 86). The block will be strlen(str)+1 bytes, that
being exactly enough for the string and nullterminator.
If str is not NULL, it should point to a block of n digits + 2 bytes, that being enough for
the mantissa, a possible minus sign, and a nullterminator. When n digits is 0 to get all
signifcant digits, an application won’t be able to know the space required, and str should be
NULL in that case.
The generated string is a fraction, with an implicit radix point immediately to the left of the
frst digit. The applicable exponent is written through the expptr pointer. For example, the
number 3.1416 would be returned as string "31416" and exponent 1.
When op is zero, an empty string is produced and the exponent returned is 0.
A pointer to the result string is returned, being either the allocated block or the given str.
7.5 Arithmetic Functions
[Function] void mpf_add (mpf t rop, mpf t op1, mpf t op2)
[Function] void mpf_add_ui (mpf t rop, mpf t op1, unsigned long int op2)
Set rop to op1 + op2.
[Function] void mpf_sub (mpf t rop, mpf t op1, mpf t op2)
[Function] void mpf_ui_sub (mpf t rop, unsigned long int op1, mpf t op2)
[Function] void mpf_sub_ui (mpf t rop, mpf t op1, unsigned long int op2)
Set rop to op1 − op2.
[Function] void mpf_mul (mpf t rop, mpf t op1, mpf t op2)
[Function] void mpf_mul_ui (mpf t rop, mpf t op1, unsigned long int op2)
Set rop to op1 op2.
Division is undefned if the divisor is zero, and passing a zero divisor to the divide functions
will make these functions intentionally divide by zero. This lets the user handle arithmetic
exceptions in these functions in the same manner as other arithmetic exceptions.
[Function] void mpf_div (mpf t rop, mpf t op1, mpf t op2)
[Function] void mpf_ui_div (mpf t rop, unsigned long int op1, mpf t op2)
[Function] void mpf_div_ui (mpf t rop, mpf t op1, unsigned long int op2)
Set rop to op1/op2.
[Function] void mpf_sqrt (mpf t rop, mpf t op)
[Function] void mpf_sqrt_ui (mpf t rop, unsigned long int op)
Set rop to
√
op.
Chapter 7: Floatingpoint Functions 53
[Function] void mpf_pow_ui (mpf t rop, mpf t op1, unsigned long int op2)
Set rop to op1
op2
.
[Function] void mpf_neg (mpf t rop, mpf t op)
Set rop to −op.
[Function] void mpf_abs (mpf t rop, mpf t op)
Set rop to the absolute value of op.
[Function] void mpf_mul_2exp (mpf t rop, mpf t op1, mp bitcnt t op2)
Set rop to op1 2
op2
.
[Function] void mpf_div_2exp (mpf t rop, mpf t op1, mp bitcnt t op2)
Set rop to op1/2
op2
.
7.6 Comparison Functions
[Function] int mpf_cmp (mpf t op1, mpf t op2)
[Function] int mpf_cmp_d (mpf t op1, double op2)
[Function] int mpf_cmp_ui (mpf t op1, unsigned long int op2)
[Function] int mpf_cmp_si (mpf t op1, signed long int op2)
Compare op1 and op2. Return a positive value if op1 > op2, zero if op1 = op2, and a
negative value if op1 < op2.
mpf_cmp_d can be called with an infnity, but results are undefned for a NaN.
[Function] int mpf_eq (mpf t op1, mpf t op2, mp bitcnt t op3)
Return nonzero if the frst op3 bits of op1 and op2 are equal, zero otherwise. I.e., test if
op1 and op2 are approximately equal.
Caution 1: All version of GMP up to version 4.2.4 compared just whole limbs, meaning
sometimes more than op3 bits, sometimes fewer.
Caution 2: This function will consider XXX11...111 and XX100...000 diferent, even if ... is
replaced by a semiinfnite number of bits. Such numbers are really just one ulp of, and
should be considered equal.
[Function] void mpf_reldiff (mpf t rop, mpf t op1, mpf t op2)
Compute the relative diference between op1 and op2 and store the result in rop. This is
[op1 −op2[/op1.
[Macro] int mpf_sgn (mpf t op)
Return +1 if op > 0, 0 if op = 0, and −1 if op < 0.
This function is actually implemented as a macro. It evaluates its arguments multiple times.
7.7 Input and Output Functions
Functions that perform input from a stdio stream, and functions that output to a stdio stream.
Passing a NULL pointer for a stream argument to any of these functions will make them read
from stdin and write to stdout, respectively.
When using any of these functions, it is a good idea to include ‘stdio.h’ before ‘gmp.h’, since
that will allow ‘gmp.h’ to defne prototypes for these functions.
54 GNU MP 5.0.1
[Function] size_t mpf_out_str (FILE *stream, int base, size t n_digits, mpf t op)
Print op to stream, as a string of digits. Return the number of bytes written, or if an error
occurred, return 0.
The mantissa is prefxed with an ‘0.’ and is in the given base, which may vary from 2 to 62
or from −2 to −36. An exponent is then printed, separated by an ‘e’, or if the base is greater
than 10 then by an ‘@’. The exponent is always in decimal. The decimal point follows the
current locale, on systems providing localeconv.
For base in the range 2..36, digits and lowercase letters are used; for −2..−36, digits and
uppercase letters are used; for 37..62, digits, uppercase letters, and lowercase letters (in
that signifcance order) are used.
Up to n digits will be printed from the mantissa, except that no more digits than are accu
rately representable by op will be printed. n digits can be 0 to select that accurate maximum.
[Function] size_t mpf_inp_str (mpf t rop, FILE *stream, int base)
Read a string in base base from stream, and put the read foat in rop. The string is of
the form ‘M@N’ or, if the base is 10 or less, alternatively ‘MeN’. ‘M’ is the mantissa and ‘N’ is
the exponent. The mantissa is always in the specifed base. The exponent is either in the
specifed base or, if base is negative, in decimal. The decimal point expected is taken from
the current locale, on systems providing localeconv.
The argument base may be in the ranges 2 to 36, or −36 to −2. Negative values are used to
specify that the exponent is in decimal.
Unlike the corresponding mpz function, the base will not be determined from the leading
characters of the string if base is 0. This is so that numbers like ‘0.23’ are not interpreted
as octal.
Return the number of bytes read, or if an error occurred, return 0.
7.8 Miscellaneous Functions
[Function] void mpf_ceil (mpf t rop, mpf t op)
[Function] void mpf_floor (mpf t rop, mpf t op)
[Function] void mpf_trunc (mpf t rop, mpf t op)
Set rop to op rounded to an integer. mpf_ceil rounds to the next higher integer, mpf_floor
to the next lower, and mpf_trunc to the integer towards zero.
[Function] int mpf_integer_p (mpf t op)
Return nonzero if op is an integer.
[Function] int mpf_fits_ulong_p (mpf t op)
[Function] int mpf_fits_slong_p (mpf t op)
[Function] int mpf_fits_uint_p (mpf t op)
[Function] int mpf_fits_sint_p (mpf t op)
[Function] int mpf_fits_ushort_p (mpf t op)
[Function] int mpf_fits_sshort_p (mpf t op)
Return nonzero if op would ft in the respective C data type, when truncated to an integer.
[Function] void mpf_urandomb (mpf t rop, gmp randstate t state, mp bitcnt t
nbits)
Generate a uniformly distributed random foat in rop, such that 0 ≤ rop < 1, with nbits
signifcant bits in the mantissa.
Chapter 7: Floatingpoint Functions 55
The variable state must be initialized by calling one of the gmp_randinit functions
(Section 9.1 [Random State Initialization], page 65) before invoking this function.
[Function] void mpf_random2 (mpf t rop, mp size t max_size, mp exp t exp)
Generate a random foat of at most max size limbs, with long strings of zeros and ones
in the binary representation. The exponent of the number is in the interval −exp to exp
(in limbs). This function is useful for testing functions and algorithms, since these kind of
random numbers have proven to be more likely to trigger cornercase bugs. Negative random
numbers are generated when max size is negative.
56 GNU MP 5.0.1
8 Lowlevel Functions
This chapter describes lowlevel GMP functions, used to implement the highlevel GMP func
tions, but also intended for timecritical user code.
These functions start with the prefx mpn_.
The mpn functions are designed to be as fast as possible, not to provide a coherent calling
interface. The diferent functions have somewhat similar interfaces, but there are variations that
make them hard to use. These functions do as little as possible apart from the real multiple
precision computation, so that no time is spent on things that not all callers need.
A source operand is specifed by a pointer to the least signifcant limb and a limb count. A
destination operand is specifed by just a pointer. It is the responsibility of the caller to ensure
that the destination has enough space for storing the result.
With this way of specifying operands, it is possible to perform computations on subranges of an
argument, and store the result into a subrange of a destination.
A common requirement for all functions is that each source area needs at least one limb. No size
argument may be zero. Unless otherwise stated, inplace operations are allowed where source
and destination are the same, but not where they only partly overlap.
The mpn functions are the base for the implementation of the mpz_, mpf_, and mpq_ functions.
This example adds the number beginning at s1p and the number beginning at s2p and writes
the sum at destp. All areas have n limbs.
cy = mpn_add_n (destp, s1p, s2p, n)
It should be noted that the mpn functions make no attempt to identify high or low zero limbs
on their operands, or other special forms. On random data such cases will be unlikely and it’d
be wasteful for every function to check every time. An application knowing something about its
data can take steps to trim or perhaps split its calculations.
In the notation used below, a source operand is identifed by the pointer to the least signifcant
limb, and the limb count in braces. For example, {s1p, s1n}.
[Function] mp_limb_t mpn_add_n (mp limb t *rp, const mp limb t *s1p, const
mp limb t *s2p, mp size t n)
Add {s1p, n} and {s2p, n}, and write the n least signifcant limbs of the result to rp. Return
carry, either 0 or 1.
This is the lowestlevel function for addition. It is the preferred function for addition, since
it is written in assembly for most CPUs. For addition of a variable to itself (i.e., s1p equals
s2p) use mpn_lshift with a count of 1 for optimal speed.
[Function] mp_limb_t mpn_add_1 (mp limb t *rp, const mp limb t *s1p, mp size t n,
mp limb t s2limb)
Add {s1p, n} and s2limb, and write the n least signifcant limbs of the result to rp. Return
carry, either 0 or 1.
[Function] mp_limb_t mpn_add (mp limb t *rp, const mp limb t *s1p, mp size t s1n,
const mp limb t *s2p, mp size t s2n)
Add {s1p, s1n} and {s2p, s2n}, and write the s1n least signifcant limbs of the result to rp.
Return carry, either 0 or 1.
Chapter 8: Lowlevel Functions 57
This function requires that s1n is greater than or equal to s2n.
[Function] mp_limb_t mpn_sub_n (mp limb t *rp, const mp limb t *s1p, const
mp limb t *s2p, mp size t n)
Subtract {s2p, n} from {s1p, n}, and write the n least signifcant limbs of the result to rp.
Return borrow, either 0 or 1.
This is the lowestlevel function for subtraction. It is the preferred function for subtraction,
since it is written in assembly for most CPUs.
[Function] mp_limb_t mpn_sub_1 (mp limb t *rp, const mp limb t *s1p, mp size t n,
mp limb t s2limb)
Subtract s2limb from {s1p, n}, and write the n least signifcant limbs of the result to rp.
Return borrow, either 0 or 1.
[Function] mp_limb_t mpn_sub (mp limb t *rp, const mp limb t *s1p, mp size t s1n,
const mp limb t *s2p, mp size t s2n)
Subtract {s2p, s2n} from {s1p, s1n}, and write the s1n least signifcant limbs of the result
to rp. Return borrow, either 0 or 1.
This function requires that s1n is greater than or equal to s2n.
[Function] void mpn_neg (mp limb t *rp, const mp limb t *sp, mp size t n)
Perform the negation of {sp, n}, and write the result to {rp, n}. Return carryout.
[Function] void mpn_mul_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Multiply {s1p, n} and {s2p, n}, and write the 2*nlimb result to rp.
The destination has to have space for 2*n limbs, even if the product’s most signifcant limb
is zero. No overlap is permitted between the destination and either source.
If the two input operands are the same, use mpn_sqr.
[Function] mp_limb_t mpn_mul (mp limb t *rp, const mp limb t *s1p, mp size t s1n,
const mp limb t *s2p, mp size t s2n)
Multiply {s1p, s1n} and {s2p, s2n}, and write the (s1n+s2n)limb result to rp. Return the
most signifcant limb of the result.
The destination has to have space for s1n + s2n limbs, even if the product’s most signifcant
limb is zero. No overlap is permitted between the destination and either source.
This function requires that s1n is greater than or equal to s2n.
[Function] void mpn_sqr (mp limb t *rp, const mp limb t *s1p, mp size t n)
Compute the square of {s1p, n} and write the 2*nlimb result to rp.
The destination has to have space for 2*n limbs, even if the result’s most signifcant limb is
zero. No overlap is permitted between the destination and the source.
[Function] mp_limb_t mpn_mul_1 (mp limb t *rp, const mp limb t *s1p, mp size t n,
mp limb t s2limb)
Multiply {s1p, n} by s2limb, and write the n least signifcant limbs of the product to rp.
Return the most signifcant limb of the product. {s1p, n} and {rp, n} are allowed to overlap
provided rp ≤ s1p.
58 GNU MP 5.0.1
This is a lowlevel function that is a building block for general multiplication as well as other
operations in GMP. It is written in assembly for most CPUs.
Don’t call this function if s2limb is a power of 2; use mpn_lshift with a count equal to the
logarithm of s2limb instead, for optimal speed.
[Function] mp_limb_t mpn_addmul_1 (mp limb t *rp, const mp limb t *s1p, mp size t
n, mp limb t s2limb)
Multiply {s1p, n} and s2limb, and add the n least signifcant limbs of the product to {rp, n}
and write the result to rp. Return the most signifcant limb of the product, plus carryout
from the addition.
This is a lowlevel function that is a building block for general multiplication as well as other
operations in GMP. It is written in assembly for most CPUs.
[Function] mp_limb_t mpn_submul_1 (mp limb t *rp, const mp limb t *s1p, mp size t
n, mp limb t s2limb)
Multiply {s1p, n} and s2limb, and subtract the n least signifcant limbs of the product from
{rp, n} and write the result to rp. Return the most signifcant limb of the product, plus
borrowout from the subtraction.
This is a lowlevel function that is a building block for general multiplication and division as
well as other operations in GMP. It is written in assembly for most CPUs.
[Function] void mpn_tdiv_qr (mp limb t *qp, mp limb t *rp, mp size t qxn, const
mp limb t *np, mp size t nn, const mp limb t *dp, mp size t dn)
Divide {np, nn} by {dp, dn} and put the quotient at {qp, nn−dn+1} and the remainder at
{rp, dn}. The quotient is rounded towards 0.
No overlap is permitted between arguments, except that np might equal rp. The dividend
size nn must be greater than or equal to divisor size dn. The most signifcant limb of the
divisor must be nonzero. The qxn operand must be zero.
[Function] mp_limb_t mpn_divrem (mp limb t *r1p, mp size t qxn, mp limb t *rs2p,
mp size t rs2n, const mp limb t *s3p, mp size t s3n)
[This function is obsolete. Please call mpn_tdiv_qr instead for best performance.]
Divide {rs2p, rs2n} by {s3p, s3n}, and write the quotient at r1p, with the exception of the
most signifcant limb, which is returned. The remainder replaces the dividend at rs2p; it will
be s3n limbs long (i.e., as many limbs as the divisor).
In addition to an integer quotient, qxn fraction limbs are developed, and stored after the
integral limbs. For most usages, qxn will be zero.
It is required that rs2n is greater than or equal to s3n. It is required that the most signifcant
bit of the divisor is set.
If the quotient is not needed, pass rs2p + s3n as r1p. Aside from that special case, no overlap
between arguments is permitted.
Return the most signifcant limb of the quotient, either 0 or 1.
The area at r1p needs to be rs2n − s3n + qxn limbs large.
Chapter 8: Lowlevel Functions 59
[Function] mp_limb_t mpn_divrem_1 (mp limb t *r1p, mp size t qxn,
mp limb t *s2p, mp size t s2n, mp limb t s3limb)
[Macro] mp_limb_t mpn_divmod_1 (mp limb t *r1p, mp limb t *s2p, mp size t s2n,
mp limb t s3limb)
Divide {s2p, s2n} by s3limb, and write the quotient at r1p. Return the remainder.
The integer quotient is written to {r1p+qxn, s2n} and in addition qxn fraction limbs are
developed and written to {r1p, qxn}. Either or both s2n and qxn can be zero. For most
usages, qxn will be zero.
mpn_divmod_1 exists for upward source compatibility and is simply a macro calling mpn_
divrem_1 with a qxn of 0.
The areas at r1p and s2p have to be identical or completely separate, not partially overlap
ping.
[Function] mp_limb_t mpn_divmod (mp limb t *r1p, mp limb t *rs2p, mp size t rs2n,
const mp limb t *s3p, mp size t s3n)
[This function is obsolete. Please call mpn_tdiv_qr instead for best performance.]
[Macro] mp_limb_t mpn_divexact_by3 (mp limb t *rp, mp limb t *sp, mp size t n)
[Function] mp_limb_t mpn_divexact_by3c (mp limb t *rp, mp limb t *sp,
mp size t n, mp limb t carry)
Divide {sp, n} by 3, expecting it to divide exactly, and writing the result to {rp, n}. If 3
divides exactly, the return value is zero and the result is the quotient. If not, the return value
is nonzero and the result won’t be anything useful.
mpn_divexact_by3c takes an initial carry parameter, which can be the return value from
a previous call, so a large calculation can be done piece by piece from low to high. mpn_
divexact_by3 is simply a macro calling mpn_divexact_by3c with a 0 carry parameter.
These routines use a multiplybyinverse and will be faster than mpn_divrem_1 on CPUs with
fast multiplication but slow division.
The source a, result q, size n, initial carry i, and return value c satisfy cb
n
+a−i = 3q, where
b = 2
GMP NUMB BITS
. The return c is always 0, 1 or 2, and the initial carry i must also be 0,
1 or 2 (these are both borrows really). When c = 0 clearly q = (a − i)/3. When c = 0, the
remainder (a −i) mod 3 is given by 3 −c, because b ≡ 1 mod 3 (when mp_bits_per_limb is
even, which is always so currently).
[Function] mp_limb_t mpn_mod_1 (mp limb t *s1p, mp size t s1n, mp limb t s2limb)
Divide {s1p, s1n} by s2limb, and return the remainder. s1n can be zero.
[Function] mp_limb_t mpn_lshift (mp limb t *rp, const mp limb t *sp, mp size t n,
unsigned int count)
Shift {sp, n} left by count bits, and write the result to {rp, n}. The bits shifted out at the
left are returned in the least signifcant count bits of the return value (the rest of the return
value is zero).
count must be in the range 1 to mp_bits_per_limb−1. The regions {sp, n} and {rp, n} may
overlap, provided rp ≥ sp.
This function is written in assembly for most CPUs.
60 GNU MP 5.0.1
[Function] mp_limb_t mpn_rshift (mp limb t *rp, const mp limb t *sp, mp size t n,
unsigned int count)
Shift {sp, n} right by count bits, and write the result to {rp, n}. The bits shifted out at
the right are returned in the most signifcant count bits of the return value (the rest of the
return value is zero).
count must be in the range 1 to mp_bits_per_limb−1. The regions {sp, n} and {rp, n} may
overlap, provided rp ≤ sp.
This function is written in assembly for most CPUs.
[Function] int mpn_cmp (const mp limb t *s1p, const mp limb t *s2p, mp size t n)
Compare {s1p, n} and {s2p, n} and return a positive value if s1 > s2, 0 if they are equal, or
a negative value if s1 < s2.
[Function] mp_size_t mpn_gcd (mp limb t *rp, mp limb t *xp, mp size t xn,
mp limb t *yp, mp size t yn)
Set {rp, retval} to the greatest common divisor of {xp, xn} and {yp, yn}. The result can be
up to yn limbs, the return value is the actual number produced. Both source operands are
destroyed.
{xp, xn} must have at least as many bits as {yp, yn}. {yp, yn} must be odd. Both operands
must have nonzero most signifcant limbs. No overlap is permitted between {xp, xn} and
{yp, yn}.
[Function] mp_limb_t mpn_gcd_1 (const mp limb t *xp, mp size t xn, mp limb t
ylimb)
Return the greatest common divisor of {xp, xn} and ylimb. Both operands must be nonzero.
[Function] mp_size_t mpn_gcdext (mp limb t *gp, mp limb t *sp, mp size t *sn,
mp limb t *xp, mp size t xn, mp limb t *yp, mp size t yn)
Let U be defned by {xp, xn} and let V be defned by {yp, yn}.
Compute the greatest common divisor G of U and V . Compute a cofactor S such that
G = US + V T. The second cofactor T is not computed but can easily be obtained from
(G−US)/V (the division will be exact). It is required that U ≥ V > 0.
S satisfes S = 1 or [S[ < V/(2G). S = 0 if and only if V divides U (i.e., G = V ).
Store G at gp and let the return value defne its limb count. Store S at sp and let *sn
defne its limb count. S can be negative; when this happens *sn will be negative. The areas
at gp and sp should each have room for xn + 1 limbs.
The areas {xp, xn +1} and {yp, yn +1} are destroyed (i.e. the input operands plus an extra
limb past the end of each).
Compatibility note: GMP 4.3.0 and 4.3.1 defned S less strictly. Earlier as well as later GMP
releases defne S as described here.
[Function] mp_size_t mpn_sqrtrem (mp limb t *r1p, mp limb t *r2p, const
mp limb t *sp, mp size t n)
Compute the square root of {sp, n} and put the result at {r1p, n/2} and the remainder
at {r2p, retval}. r2p needs space for n limbs, but the return value indicates how many are
produced.
Chapter 8: Lowlevel Functions 61
The most signifcant limb of {sp, n} must be nonzero. The areas {r1p, n/2} and {sp, n}
must be completely separate. The areas {r2p, n} and {sp, n} must be either identical or
completely separate.
If the remainder is not wanted then r2p can be NULL, and in this case the return value is zero
or nonzero according to whether the remainder would have been zero or nonzero.
A return value of zero indicates a perfect square. See also mpz_perfect_square_p.
[Function] mp_size_t mpn_get_str (unsigned char *str, int base, mp limb t *s1p,
mp size t s1n)
Convert {s1p, s1n} to a raw unsigned char array at str in base base, and return the number
of characters produced. There may be leading zeros in the string. The string is not in ASCII;
to convert it to printable format, add the ASCII codes for ‘0’ or ‘A’, depending on the base
and range. base can vary from 2 to 256.
The most signifcant limb of the input {s1p, s1n} must be nonzero. The input {s1p, s1n} is
clobbered, except when base is a power of 2, in which case it’s unchanged.
The area at str has to have space for the largest possible number represented by a s1n long
limb array, plus one extra character.
[Function] mp_size_t mpn_set_str (mp limb t *rp, const unsigned char *str, size t
strsize, int base)
Convert bytes {str,strsize} in the given base to limbs at rp.
str[0] is the most signifcant byte and str[strsize −1] is the least signifcant. Each byte should
be a value in the range 0 to base −1, not an ASCII character. base can vary from 2 to 256.
The return value is the number of limbs written to rp. If the most signifcant input byte is
nonzero then the high limb at rp will be nonzero, and only that exact number of limbs will
be required there.
If the most signifcant input byte is zero then there may be high zero limbs written to rp and
included in the return value.
strsize must be at least 1, and no overlap is permitted between {str,strsize} and the result
at rp.
[Function] mp_bitcnt_t mpn_scan0 (const mp limb t *s1p, mp bitcnt t bit)
Scan s1p from bit position bit for the next clear bit.
It is required that there be a clear bit within the area at s1p at or beyond bit position bit,
so that the function has something to return.
[Function] mp_bitcnt_t mpn_scan1 (const mp limb t *s1p, mp bitcnt t bit)
Scan s1p from bit position bit for the next set bit.
It is required that there be a set bit within the area at s1p at or beyond bit position bit, so
that the function has something to return.
[Function] void mpn_random (mp limb t *r1p, mp size t r1n)
[Function] void mpn_random2 (mp limb t *r1p, mp size t r1n)
Generate a random number of length r1n and store it at r1p. The most signifcant limb
is always nonzero. mpn_random generates uniformly distributed limb data, mpn_random2
generates long strings of zeros and ones in the binary representation.
62 GNU MP 5.0.1
mpn_random2 is intended for testing the correctness of the mpn routines.
[Function] mp_bitcnt_t mpn_popcount (const mp limb t *s1p, mp size t n)
Count the number of set bits in {s1p, n}.
[Function] mp_bitcnt_t mpn_hamdist (const mp limb t *s1p, const mp limb t *s2p,
mp size t n)
Compute the hamming distance between {s1p, n} and {s2p, n}, which is the number of bit
positions where the two operands have diferent bit values.
[Function] int mpn_perfect_square_p (const mp limb t *s1p, mp size t n)
Return nonzero if {s1p, n} is a perfect square.
[Function] void mpn_and_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Perform the bitwise logical and of {s1p, n} and {s2p, n}, and write the result to {rp, n}.
[Function] void mpn_ior_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Perform the bitwise logical inclusive or of {s1p, n} and {s2p, n}, and write the result to {rp,
n}.
[Function] void mpn_xor_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Perform the bitwise logical exclusive or of {s1p, n} and {s2p, n}, and write the result to {rp,
n}.
[Function] void mpn_andn_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Perform the bitwise logical and of {s1p, n} and the bitwise complement of {s2p, n}, and
write the result to {rp, n}.
[Function] void mpn_iorn_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Perform the bitwise logical inclusive or of {s1p, n} and the bitwise complement of {s2p, n},
and write the result to {rp, n}.
[Function] void mpn_nand_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Perform the bitwise logical and of {s1p, n} and {s2p, n}, and write the bitwise complement
of the result to {rp, n}.
[Function] void mpn_nior_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Perform the bitwise logical inclusive or of {s1p, n} and {s2p, n}, and write the bitwise
complement of the result to {rp, n}.
[Function] void mpn_xnor_n (mp limb t *rp, const mp limb t *s1p, const mp limb t
*s2p, mp size t n)
Perform the bitwise logical exclusive or of {s1p, n} and {s2p, n}, and write the bitwise
complement of the result to {rp, n}.
Chapter 8: Lowlevel Functions 63
[Function] void mpn_com (mp limb t *rp, const mp limb t *sp, mp size t n)
Perform the bitwise complement of {sp, n}, and write the result to {rp, n}.
[Function] void mpn_copyi (mp limb t *rp, const mp limb t *s1p, mp size t n)
Copy from {s1p, n} to {rp, n}, increasingly.
[Function] void mpn_copyd (mp limb t *rp, const mp limb t *s1p, mp size t n)
Copy from {s1p, n} to {rp, n}, decreasingly.
[Function] void mpn_zero (mp limb t *rp, mp size t n)
Zero {rp, n}.
8.1 Nails
Everything in this section is highly experimental and may disappear or be subject to incompat
ible changes in a future version of GMP.
Nails are an experimental feature whereby a few bits are left unused at the top of each mp_limb_
t. This can signifcantly improve carry handling on some processors.
All the mpn functions accepting limb data will expect the nail bits to be zero on entry, and will
return data with the nails similarly all zero. This applies both to limb vectors and to single limb
arguments.
Nails can be enabled by confguring with ‘enablenails’. By default the number of bits will
be chosen according to what suits the host processor, but a particular number can be selected
with ‘enablenails=N’.
At the mpn level, a nail build is neither source nor binary compatible with a nonnail build,
strictly speaking. But programs acting on limbs only through the mpn functions are likely to
work equally well with either build, and judicious use of the defnitions below should make any
program compatible with either build, at the source level.
For the higher level routines, meaning mpz etc, a nail build should be fully source and binary
compatible with a nonnail build.
[Macro] GMP_NAIL_BITS
[Macro] GMP_NUMB_BITS
[Macro] GMP_LIMB_BITS
GMP_NAIL_BITS is the number of nail bits, or 0 when nails are not in use. GMP_NUMB_BITS
is the number of data bits in a limb. GMP_LIMB_BITS is the total number of bits in an
mp_limb_t. In all cases
GMP_LIMB_BITS == GMP_NAIL_BITS + GMP_NUMB_BITS
[Macro] GMP_NAIL_MASK
[Macro] GMP_NUMB_MASK
Bit masks for the nail and number parts of a limb. GMP_NAIL_MASK is 0 when nails are not
in use.
GMP_NAIL_MASK is not often needed, since the nail part can be obtained with x >> GMP_NUMB_
BITS, and that means one less large constant, which can help various RISC chips.
64 GNU MP 5.0.1
[Macro] GMP_NUMB_MAX
The maximum value that can be stored in the number part of a limb. This is the same as
GMP_NUMB_MASK, but can be used for clarity when doing comparisons rather than bitwise
operations.
The term “nails” comes from fnger or toe nails, which are at the ends of a limb (arm or leg).
“numb” is short for number, but is also how the developers felt after trying for a long time to
come up with sensible names for these things.
In the future (the distant future most likely) a nonzero nail might be permitted, giving non
unique representations for numbers in a limb vector. This would help vector processors since
carries would only ever need to propagate one or two limbs.
Chapter 9: Random Number Functions 65
9 Random Number Functions
Sequences of pseudorandom numbers in GMP are generated using a variable of type gmp_
randstate_t, which holds an algorithm selection and a current state. Such a variable must be
initialized by a call to one of the gmp_randinit functions, and can be seeded with one of the
gmp_randseed functions.
The functions actually generating random numbers are described in Section 5.13 [Integer Ran
dom Numbers], page 40, and Section 7.8 [Miscellaneous Float Functions], page 54.
The older style random number functions don’t accept a gmp_randstate_t parameter but in
stead share a global variable of that type. They use a default algorithm and are currently
not seeded (though perhaps that will change in the future). The new functions accepting a
gmp_randstate_t are recommended for applications that care about randomness.
9.1 Random State Initialization
[Function] void gmp_randinit_default (gmp randstate t state)
Initialize state with a default algorithm. This will be a compromise between speed and
randomness, and is recommended for applications with no special requirements. Currently
this is gmp_randinit_mt.
[Function] void gmp_randinit_mt (gmp randstate t state)
Initialize state for a Mersenne Twister algorithm. This algorithm is fast and has good ran
domness properties.
[Function] void gmp_randinit_lc_2exp (gmp randstate t state, mpz t a,
unsigned long c, mp bitcnt t m2exp)
Initialize state with a linear congruential algorithm X = (aX + c) mod 2
m2exp
.
The low bits of X in this algorithm are not very random. The least signifcant bit will have
a period no more than 2, and the second bit no more than 4, etc. For this reason only the
high half of each X is actually used.
When a random number of more than m2exp/2 bits is to be generated, multiple iterations
of the recurrence are used and the results concatenated.
[Function] int gmp_randinit_lc_2exp_size (gmp randstate t state, mp bitcnt t
size)
Initialize state for a linear congruential algorithm as per gmp_randinit_lc_2exp. a, c and
m2exp are selected from a table, chosen so that size bits (or more) of each X will be used,
ie. m2exp/2 ≥ size.
If successful the return value is nonzero. If size is bigger than the table data provides then
the return value is zero. The maximum size currently supported is 128.
[Function] void gmp_randinit_set (gmp randstate t rop, gmp randstate t op)
Initialize rop with a copy of the algorithm and state from op.
[Function] void gmp_randinit (gmp randstate t state, gmp randalg t alg, . . . )
This function is obsolete.
Initialize state with an algorithm selected by alg. The only choice is GMP_RAND_ALG_LC, which
is gmp_randinit_lc_2exp_size described above. A third parameter of type unsigned long
66 GNU MP 5.0.1
is required, this is the size for that function. GMP_RAND_ALG_DEFAULT or 0 are the same as
GMP_RAND_ALG_LC.
gmp_randinit sets bits in the global variable gmp_errno to indicate an error. GMP_ERROR_
UNSUPPORTED_ARGUMENT if alg is unsupported, or GMP_ERROR_INVALID_ARGUMENT if the size
parameter is too big. It may be noted this error reporting is not thread safe (a good reason
to use gmp_randinit_lc_2exp_size instead).
[Function] void gmp_randclear (gmp randstate t state)
Free all memory occupied by state.
9.2 Random State Seeding
[Function] void gmp_randseed (gmp randstate t state, mpz t seed)
[Function] void gmp_randseed_ui (gmp randstate t state, unsigned long int seed)
Set an initial seed value into state.
The size of a seed determines how many diferent sequences of random numbers that it’s
possible to generate. The “quality” of the seed is the randomness of a given seed compared
to the previous seed used, and this afects the randomness of separate number sequences. The
method for choosing a seed is critical if the generated numbers are to be used for important
applications, such as generating cryptographic keys.
Traditionally the system time has been used to seed, but care needs to be taken with this.
If an application seeds often and the resolution of the system clock is low, then the same
sequence of numbers might be repeated. Also, the system time is quite easy to guess, so if
unpredictability is required then it should defnitely not be the only source for the seed value.
On some systems there’s a special device ‘/dev/random’ which provides random data better
suited for use as a seed.
9.3 Random State Miscellaneous
[Function] unsigned long gmp_urandomb_ui (gmp randstate t state, unsigned long
n)
Return a uniformly distributed random number of n bits, ie. in the range 0 to 2
n
−1 inclusive.
n must be less than or equal to the number of bits in an unsigned long.
[Function] unsigned long gmp_urandomm_ui (gmp randstate t state, unsigned long
n)
Return a uniformly distributed random number in the range 0 to n −1, inclusive.
Chapter 10: Formatted Output 67
10 Formatted Output
10.1 Format Strings
gmp_printf and friends accept format strings similar to the standard C printf (see Section
“Formatted Output” in The GNU C Library Reference Manual). A format specifcation is of
the form
% [flags] [width] [.[precision]] [type] conv
GMP adds types ‘Z’, ‘Q’ and ‘F’ for mpz_t, mpq_t and mpf_t respectively, ‘M’ for mp_limb_t,
and ‘N’ for an mp_limb_t array. ‘Z’, ‘Q’, ‘M’ and ‘N’ behave like integers. ‘Q’ will print a ‘/’ and
a denominator, if needed. ‘F’ behaves like a foat. For example,
mpz_t z;
gmp_printf ("%s is an mpz %Zd\n", "here", z);
mpq_t q;
gmp_printf ("a hex rational: %#40Qx\n", q);
mpf_t f;
int n;
gmp_printf ("fixed point mpf %.*Ff with %d digits\n", n, f, n);
mp_limb_t l;
gmp_printf ("limb %Mu\n", l);
const mp_limb_t *ptr;
mp_size_t size;
gmp_printf ("limb array %Nx\n", ptr, size);
For ‘N’ the limbs are expected least signifcant frst, as per the mpn functions (see Chapter 8
[Lowlevel Functions], page 56). A negative size can be given to print the value as a negative.
All the standard C printf types behave the same as the C library printf, and can be freely
intermixed with the GMP extensions. In the current implementation the standard parts of the
format string are simply handed to printf and only the GMP extensions handled directly.
The fags accepted are as follows. GLIBC style ‘’’ is only for the standard C types (not the
GMP types), and only if the C library supports it.
0 pad with zeros (rather than spaces)
# show the base with ‘0x’, ‘0X’ or ‘0’
+ always show a sign
(space) show a space or a ‘’ sign
’ group digits, GLIBC style (not GMP types)
The optional width and precision can be given as a number within the format string, or as a ‘*’
to take an extra parameter of type int, the same as the standard printf.
The standard types accepted are as follows. ‘h’ and ‘l’ are portable, the rest will depend on the
compiler (or include fles) for the type and the C library for the output.
h short
hh char
68 GNU MP 5.0.1
j intmax_t or uintmax_t
l long or wchar_t
ll long long
L long double
q quad_t or u_quad_t
t ptrdiff_t
z size_t
The GMP types are
F mpf_t, foat conversions
Q mpq_t, integer conversions
M mp_limb_t, integer conversions
N mp_limb_t array, integer conversions
Z mpz_t, integer conversions
The conversions accepted are as follows. ‘a’ and ‘A’ are always supported for mpf_t but depend
on the C library for standard C foat types. ‘m’ and ‘p’ depend on the C library.
a A hex foats, C99 style
c character
d decimal integer
e E scientifc format foat
f fxed point foat
i same as d
g G fxed or scientifc foat
m strerror string, GLIBC style
n store characters written so far
o octal integer
p pointer
s string
u unsigned integer
x X hex integer
‘o’, ‘x’ and ‘X’ are unsigned for the standard C types, but for types ‘Z’, ‘Q’ and ‘N’ they are
signed. ‘u’ is not meaningful for ‘Z’, ‘Q’ and ‘N’.
‘M’ is a proxy for the C library ‘l’ or ‘L’, according to the size of mp_limb_t. Unsigned conver
sions will be usual, but a signed conversion can be used and will interpret the value as a twos
complement negative.
‘n’ can be used with any type, even the GMP types.
Other types or conversions that might be accepted by the C library printf cannot be used
through gmp_printf, this includes for instance extensions registered with GLIBC register_
printf_function. Also currently there’s no support for POSIX ‘$’ style numbered arguments
(perhaps this will be added in the future).
The precision feld has it’s usual meaning for integer ‘Z’ and foat ‘F’ types, but is currently
undefned for ‘Q’ and should not be used with that.
mpf_t conversions only ever generate as many digits as can be accurately represented by the
operand, the same as mpf_get_str does. Zeros will be used if necessary to pad to the requested
precision. This happens even for an ‘f’ conversion of an mpf_t which is an integer, for instance
Chapter 10: Formatted Output 69
2
1024
in an mpf_t of 128 bits precision will only produce about 40 digits, then pad with zeros
to the decimal point. An empty precision feld like ‘%.Fe’ or ‘%.Ff’ can be used to specifcally
request just the signifcant digits.
The decimal point character (or string) is taken from the current locale settings on systems which
provide localeconv (see Section “Locales and Internationalization” in The GNU C Library
Reference Manual). The C library will normally do the same for standard foat output.
The format string is only interpreted as plain chars, multibyte characters are not recognised.
Perhaps this will change in the future.
10.2 Functions
Each of the following functions is similar to the corresponding C library function. The basic
printf forms take a variable argument list. The vprintf forms take an argument pointer, see
Section “Variadic Functions” in The GNU C Library Reference Manual, or ‘man 3 va_start’.
It should be emphasised that if a format string is invalid, or the arguments don’t match what
the format specifes, then the behaviour of any of these functions will be unpredictable. GCC
format string checking is not available, since it doesn’t recognise the GMP extensions.
The fle based functions gmp_printf and gmp_fprintf will return −1 to indicate a write error.
Output is not “atomic”, so partial output may be produced if a write error occurs. All the
functions can return −1 if the C library printf variant in use returns −1, but this shouldn’t
normally occur.
[Function] int gmp_printf (const char *fmt, . . . )
[Function] int gmp_vprintf (const char *fmt, va list ap)
Print to the standard output stdout. Return the number of characters written, or −1 if an
error occurred.
[Function] int gmp_fprintf (FILE *fp, const char *fmt, . . . )
[Function] int gmp_vfprintf (FILE *fp, const char *fmt, va list ap)
Print to the stream fp. Return the number of characters written, or −1 if an error occurred.
[Function] int gmp_sprintf (char *buf, const char *fmt, . . . )
[Function] int gmp_vsprintf (char *buf, const char *fmt, va list ap)
Form a nullterminated string in buf. Return the number of characters written, excluding
the terminating null.
No overlap is permitted between the space at buf and the string fmt.
These functions are not recommended, since there’s no protection against exceeding the space
available at buf.
[Function] int gmp_snprintf (char *buf, size t size, const char *fmt, . . . )
[Function] int gmp_vsnprintf (char *buf, size t size, const char *fmt, va list ap)
Form a nullterminated string in buf. No more than size bytes will be written. To get the
full output, size must be enough for the string and nullterminator.
The return value is the total number of characters which ought to have been produced,
excluding the terminating null. If retval ≥ size then the actual output has been truncated to
the frst size −1 characters, and a null appended.
No overlap is permitted between the region {buf,size} and the fmt string.
70 GNU MP 5.0.1
Notice the return value is in ISO C99 snprintf style. This is so even if the C library
vsnprintf is the older GLIBC 2.0.x style.
[Function] int gmp_asprintf (char **pp, const char *fmt, . . . )
[Function] int gmp_vasprintf (char **pp, const char *fmt, va list ap)
Form a nullterminated string in a block of memory obtained from the current memory
allocation function (see Chapter 14 [Custom Allocation], page 86). The block will be the size
of the string and nullterminator. The address of the block in stored to *pp. The return
value is the number of characters produced, excluding the nullterminator.
Unlike the C library asprintf, gmp_asprintf doesn’t return −1 if there’s no more memory
available, it lets the current allocation function handle that.
[Function] int gmp_obstack_printf (struct obstack *ob, const char *fmt, . . . )
[Function] int gmp_obstack_vprintf (struct obstack *ob, const char *fmt, va list ap)
Append to the current object in ob. The return value is the number of characters written.
A nullterminator is not written.
fmt cannot be within the current object in ob, since that object might move as it grows.
These functions are available only when the C library provides the obstack feature, which
probably means only on GNU systems, see Section “Obstacks” in The GNU C Library Ref
erence Manual.
10.3 C++ Formatted Output
The following functions are provided in ‘libgmpxx’ (see Section 3.1 [Headers and Libraries],
page 16), which is built if C++ support is enabled (see Section 2.1 [Build Options], page 3).
Prototypes are available from <gmp.h>.
[Function] ostream& operator<< (ostream& stream, mpz t op)
Print op to stream, using its ios formatting settings. ios::width is reset to 0 after output,
the same as the standard ostream operator<< routines do.
In hex or octal, op is printed as a signed number, the same as for decimal. This is unlike the
standard operator<< routines on int etc, which instead give twos complement.
[Function] ostream& operator<< (ostream& stream, mpq t op)
Print op to stream, using its ios formatting settings. ios::width is reset to 0 after output,
the same as the standard ostream operator<< routines do.
Output will be a fraction like ‘5/9’, or if the denominator is 1 then just a plain integer like
‘123’.
In hex or octal, op is printed as a signed value, the same as for decimal. If ios::showbase is
set then a base indicator is shown on both the numerator and denominator (if the denominator
is required).
[Function] ostream& operator<< (ostream& stream, mpf t op)
Print op to stream, using its ios formatting settings. ios::width is reset to 0 after output,
the same as the standard ostream operator<< routines do.
The decimal point follows the standard library foat operator<<, which on recent systems
means the std::locale imbued on stream.
Chapter 10: Formatted Output 71
Hex and octal are supported, unlike the standard operator<< on double. The mantissa will
be in hex or octal, the exponent will be in decimal. For hex the exponent delimiter is an ‘@’.
This is as per mpf_out_str.
ios::showbase is supported, and will put a base on the mantissa, for example hex ‘0x1.8’ or
‘0x0.8’, or octal ‘01.4’ or ‘00.4’. This last form is slightly strange, but at least diferentiates
itself from decimal.
These operators mean that GMP types can be printed in the usual C++ way, for example,
mpz_t z;
int n;
...
cout << "iteration " << n << " value " << z << "\n";
But note that ostream output (and istream input, see Section 11.3 [C++ Formatted Input],
page 74) is the only overloading available for the GMP types and that for instance using + with
an mpz_t will have unpredictable results. For classes with overloading, see Chapter 12 [C++
Class Interface], page 76.
72 GNU MP 5.0.1
11 Formatted Input
11.1 Formatted Input Strings
gmp_scanf and friends accept format strings similar to the standard C scanf (see Section
“Formatted Input” in The GNU C Library Reference Manual). A format specifcation is of the
form
% [flags] [width] [type] conv
GMP adds types ‘Z’, ‘Q’ and ‘F’ for mpz_t, mpq_t and mpf_t respectively. ‘Z’ and ‘Q’ behave like
integers. ‘Q’ will read a ‘/’ and a denominator, if present. ‘F’ behaves like a foat.
GMP variables don’t require an & when passed to gmp_scanf, since they’re already “callby
reference”. For example,
/* to read say "a(5) = 1234" */
int n;
mpz_t z;
gmp_scanf ("a(%d) = %Zd\n", &n, z);
mpq_t q1, q2;
gmp_sscanf ("0377 + 0x10/0x11", "%Qi + %Qi", q1, q2);
/* to read say "topleft (1.55,2.66)" */
mpf_t x, y;
char buf[32];
gmp_scanf ("%31s (%Ff,%Ff)", buf, x, y);
All the standard C scanf types behave the same as in the C library scanf, and can be freely
intermixed with the GMP extensions. In the current implementation the standard parts of the
format string are simply handed to scanf and only the GMP extensions handled directly.
The fags accepted are as follows. ‘a’ and ‘’’ will depend on support from the C library, and ‘’’
cannot be used with GMP types.
* read but don’t store
a allocate a bufer (string conversions)
’ grouped digits, GLIBC style (not GMP types)
The standard types accepted are as follows. ‘h’ and ‘l’ are portable, the rest will depend on the
compiler (or include fles) for the type and the C library for the input.
h short
hh char
j intmax_t or uintmax_t
l long int, double or wchar_t
ll long long
L long double
q quad_t or u_quad_t
t ptrdiff_t
z size_t
The GMP types are
Chapter 11: Formatted Input 73
F mpf_t, foat conversions
Q mpq_t, integer conversions
Z mpz_t, integer conversions
The conversions accepted are as follows. ‘p’ and ‘[’ will depend on support from the C library,
the rest are standard.
c character or characters
d decimal integer
e E f g
G
foat
i integer with base indicator
n characters read so far
o octal integer
p pointer
s string of nonwhitespace characters
u decimal integer
x X hex integer
[ string of characters in a set
‘e’, ‘E’, ‘f’, ‘g’ and ‘G’ are identical, they all read either fxed point or scientifc format, and
either upper or lower case ‘e’ for the exponent in scientifc format.
C99 style hex foat format (printf %a, see Section 10.1 [Formatted Output Strings], page 67) is
always accepted for mpf_t, but for the standard foat types it will depend on the C library.
‘x’ and ‘X’ are identical, both accept both upper and lower case hexadecimal.
‘o’, ‘u’, ‘x’ and ‘X’ all read positive or negative values. For the standard C types these are
described as “unsigned” conversions, but that merely afects certain overfow handling, negatives
are still allowed (per strtoul, see Section “Parsing of Integers” in The GNU C Library Reference
Manual). For GMP types there are no overfows, so ‘d’ and ‘u’ are identical.
‘Q’ type reads the numerator and (optional) denominator as given. If the value might not be in
canonical form then mpq_canonicalize must be called before using it in any calculations (see
Chapter 6 [Rational Number Functions], page 44).
‘Qi’ will read a base specifcation separately for the numerator and denominator. For example
‘0x10/11’ would be 16/11, whereas ‘0x10/0x11’ would be 16/17.
‘n’ can be used with any of the types above, even the GMP types. ‘*’ to suppress assignment is
allowed, though in that case it would do nothing at all.
Other conversions or types that might be accepted by the C library scanf cannot be used
through gmp_scanf.
Whitespace is read and discarded before a feld, except for ‘c’ and ‘[’ conversions.
For foat conversions, the decimal point character (or string) expected is taken from the current
locale settings on systems which provide localeconv (see Section “Locales and International
ization” in The GNU C Library Reference Manual). The C library will normally do the same
for standard foat input.
The format string is only interpreted as plain chars, multibyte characters are not recognised.
Perhaps this will change in the future.
74 GNU MP 5.0.1
11.2 Formatted Input Functions
Each of the following functions is similar to the corresponding C library function. The plain
scanf forms take a variable argument list. The vscanf forms take an argument pointer, see
Section “Variadic Functions” in The GNU C Library Reference Manual, or ‘man 3 va_start’.
It should be emphasised that if a format string is invalid, or the arguments don’t match what
the format specifes, then the behaviour of any of these functions will be unpredictable. GCC
format string checking is not available, since it doesn’t recognise the GMP extensions.
No overlap is permitted between the fmt string and any of the results produced.
[Function] int gmp_scanf (const char *fmt, . . . )
[Function] int gmp_vscanf (const char *fmt, va list ap)
Read from the standard input stdin.
[Function] int gmp_fscanf (FILE *fp, const char *fmt, . . . )
[Function] int gmp_vfscanf (FILE *fp, const char *fmt, va list ap)
Read from the stream fp.
[Function] int gmp_sscanf (const char *s, const char *fmt, . . . )
[Function] int gmp_vsscanf (const char *s, const char *fmt, va list ap)
Read from a nullterminated string s.
The return value from each of these functions is the same as the standard C99 scanf, namely
the number of felds successfully parsed and stored. ‘%n’ felds and felds read but suppressed by
‘*’ don’t count towards the return value.
If end of input (or a fle error) is reached before a character for a feld or a literal, and if
no previous nonsuppressed felds have matched, then the return value is EOF instead of 0. A
whitespace character in the format string is only an optional match and doesn’t induce an EOF
in this fashion. Leading whitespace read and discarded for a feld don’t count as characters for
that feld.
For the GMP types, input parsing follows C99 rules, namely one character of lookahead is used
and characters are read while they continue to meet the format requirements. If this doesn’t
provide a complete number then the function terminates, with that feld not stored nor counted
towards the return value. For instance with mpf_t an input ‘1.23eXYZ’ would be read up to
the ‘X’ and that character pushed back since it’s not a digit. The string ‘1.23e’ would then be
considered invalid since an ‘e’ must be followed by at least one digit.
For the standard C types, in the current implementation GMP calls the C library scanf func
tions, which might have looser rules about what constitutes a valid input.
Note that gmp_sscanf is the same as gmp_fscanf and only does one character of lookahead
when parsing. Although clearly it could look at its entire input, it is deliberately made identical
to gmp_fscanf, the same way C99 sscanf is the same as fscanf.
11.3 C++ Formatted Input
The following functions are provided in ‘libgmpxx’ (see Section 3.1 [Headers and Libraries],
page 16), which is built only if C++ support is enabled (see Section 2.1 [Build Options], page 3).
Prototypes are available from <gmp.h>.
[Function] istream& operator>> (istream& stream, mpz t rop)
Read rop from stream, using its ios formatting settings.
Chapter 11: Formatted Input 75
[Function] istream& operator>> (istream& stream, mpq t rop)
An integer like ‘123’ will be read, or a fraction like ‘5/9’. No whitespace is allowed around
the ‘/’. If the fraction is not in canonical form then mpq_canonicalize must be called (see
Chapter 6 [Rational Number Functions], page 44) before operating on it.
As per integer input, an ‘0’ or ‘0x’ base indicator is read when none of ios::dec, ios::oct
or ios::hex are set. This is done separately for numerator and denominator, so that for
instance ‘0x10/11’ is 16/11 and ‘0x10/0x11’ is 16/17.
[Function] istream& operator>> (istream& stream, mpf t rop)
Read rop from stream, using its ios formatting settings.
Hex or octal foats are not supported, but might be in the future, or perhaps it’s best to
accept only what the standard foat operator>> does.
Note that digit grouping specifed by the istream locale is currently not accepted. Perhaps this
will change in the future.
These operators mean that GMP types can be read in the usual C++ way, for example,
mpz_t z;
...
cin >> z;
But note that istream input (and ostream output, see Section 10.3 [C++ Formatted Output],
page 70) is the only overloading available for the GMP types and that for instance using + with
an mpz_t will have unpredictable results. For classes with overloading, see Chapter 12 [C++
Class Interface], page 76.
76 GNU MP 5.0.1
12 C++ Class Interface
This chapter describes the C++ class based interface to GMP.
All GMP C language types and functions can be used in C++ programs, since ‘gmp.h’ has extern
"C" qualifers, but the class interface ofers overloaded functions and operators which may be
more convenient.
Due to the implementation of this interface, a reasonably recent C++ compiler is required, one
supporting namespaces, partial specialization of templates and member templates. For GCC
this means version 2.91 or later.
Everything described in this chapter is to be considered preliminary and might be subject to
incompatible changes if some unforeseen difculty reveals itself.
12.1 C++ Interface General
All the C++ classes and functions are available with
#include <gmpxx.h>
Programs should be linked with the ‘libgmpxx’ and ‘libgmp’ libraries. For example,
g++ mycxxprog.cc lgmpxx lgmp
The classes defned are
[Class] mpz_class
[Class] mpq_class
[Class] mpf_class
The standard operators and various standard functions are overloaded to allow arithmetic with
these classes. For example,
int
main (void)
{
mpz_class a, b, c;
a = 1234;
b = "5678";
c = a+b;
cout << "sum is " << c << "\n";
cout << "absolute value is " << abs(c) << "\n";
return 0;
}
An important feature of the implementation is that an expression like a=b+c results in a single
call to the corresponding mpz_add, without using a temporary for the b+c part. Expressions
which by their nature imply intermediate values, like a=b*c+d*e, still use temporaries though.
The classes can be freely intermixed in expressions, as can the classes and the standard types
long, unsigned long and double. Smaller types like int or float can also be intermixed, since
C++ will promote them.
Note that bool is not accepted directly, but must be explicitly cast to an int frst. This is
because C++ will automatically convert any pointer to a bool, so if GMP accepted bool it
Chapter 12: C++ Class Interface 77
would make all sorts of invalid class and pointer combinations compile but almost certainly not
do anything sensible.
Conversions back from the classes to standard C++ types aren’t done automatically, instead
member functions like get_si are provided (see the following sections for details).
Also there are no automatic conversions from the classes to the corresponding GMP C types,
instead a reference to the underlying C object can be obtained with the following functions,
[Function] mpz_t mpz_class::get_mpz_t ()
[Function] mpq_t mpq_class::get_mpq_t ()
[Function] mpf_t mpf_class::get_mpf_t ()
These can be used to call a C function which doesn’t have a C++ class interface. For example
to set a to the GCD of b and c,
mpz_class a, b, c;
...
mpz_gcd (a.get_mpz_t(), b.get_mpz_t(), c.get_mpz_t());
In the other direction, a class can be initialized from the corresponding GMP C type, or assigned
to if an explicit constructor is used. In both cases this makes a copy of the value, it doesn’t
create any sort of association. For example,
mpz_t z;
// ... init and calculate z ...
mpz_class x(z);
mpz_class y;
y = mpz_class (z);
There are no namespace setups in ‘gmpxx.h’, all types and functions are simply put into the
global namespace. This is what ‘gmp.h’ has done in the past, and continues to do for compat
ibility. The extras provided by ‘gmpxx.h’ follow GMP naming conventions and are unlikely to
clash with anything.
12.2 C++ Interface Integers
[Function] void mpz_class::mpz_class (type n)
Construct an mpz_class. All the standard C++ types may be used, except long long and
long double, and all the GMP C++ classes can be used. Any necessary conversion follows the
corresponding C function, for example double follows mpz_set_d (see Section 5.2 [Assigning
Integers], page 30).
[Function] void mpz_class::mpz_class (mpz t z)
Construct an mpz_class from an mpz_t. The value in z is copied into the new mpz_class,
there won’t be any permanent association between it and z.
[Function] void mpz_class::mpz_class (const char *s)
[Function] void mpz_class::mpz_class (const char *s, int base = 0)
[Function] void mpz_class::mpz_class (const string& s)
[Function] void mpz_class::mpz_class (const string& s, int base = 0)
Construct an mpz_class converted from a string using mpz_set_str (see Section 5.2 [As
signing Integers], page 30).
If the string is not a valid integer, an std::invalid_argument exception is thrown. The
same applies to operator=.
78 GNU MP 5.0.1
[Function] mpz_class operator/ (mpz class a, mpz class d)
[Function] mpz_class operator% (mpz class a, mpz class d)
Divisions involving mpz_class round towards zero, as per the mpz_tdiv_q and mpz_tdiv_r
functions (see Section 5.6 [Integer Division], page 32). This is the same as the C99 / and %
operators.
The mpz_fdiv... or mpz_cdiv... functions can always be called directly if desired. For
example,
mpz_class q, a, d;
...
mpz_fdiv_q (q.get_mpz_t(), a.get_mpz_t(), d.get_mpz_t());
[Function] mpz_class abs (mpz class op1)
[Function] int cmp (mpz class op1, type op2)
[Function] int cmp (type op1, mpz class op2)
[Function] bool mpz_class::fits_sint_p (void)
[Function] bool mpz_class::fits_slong_p (void)
[Function] bool mpz_class::fits_sshort_p (void)
[Function] bool mpz_class::fits_uint_p (void)
[Function] bool mpz_class::fits_ulong_p (void)
[Function] bool mpz_class::fits_ushort_p (void)
[Function] double mpz_class::get_d (void)
[Function] long mpz_class::get_si (void)
[Function] string mpz_class::get_str (int base = 10)
[Function] unsigned long mpz_class::get_ui (void)
[Function] int mpz_class::set_str (const char *str, int base)
[Function] int mpz_class::set_str (const string& str, int base)
[Function] int sgn (mpz class op)
[Function] mpz_class sqrt (mpz class op)
These functions provide a C++ class interface to the corresponding GMP C routines.
cmp can be used with any of the classes or the standard C++ types, except long long and
long double.
Overloaded operators for combinations of mpz_class and double are provided for completeness,
but it should be noted that if the given double is not an integer then the way any rounding is
done is currently unspecifed. The rounding might take place at the start, in the middle, or at
the end of the operation, and it might change in the future.
Conversions between mpz_class and double, however, are defned to follow the corresponding
C functions mpz_get_d and mpz_set_d. And comparisons are always made exactly, as per
mpz_cmp_d.
12.3 C++ Interface Rationals
In all the following constructors, if a fraction is given then it should be in canonical form, or if
not then mpq_class::canonicalize called.
[Function] void mpq_class::mpq_class (type op)
[Function] void mpq_class::mpq_class (integer num, integer den)
Construct an mpq_class. The initial value can be a single value of any type, or a pair of
integers (mpz_class or standard C++ integer types) representing a fraction, except that long
long and long double are not supported. For example,
Chapter 12: C++ Class Interface 79
mpq_class q (99);
mpq_class q (1.75);
mpq_class q (1, 3);
[Function] void mpq_class::mpq_class (mpq t q)
Construct an mpq_class from an mpq_t. The value in q is copied into the new mpq_class,
there won’t be any permanent association between it and q.
[Function] void mpq_class::mpq_class (const char *s)
[Function] void mpq_class::mpq_class (const char *s, int base = 0)
[Function] void mpq_class::mpq_class (const string& s)
[Function] void mpq_class::mpq_class (const string& s, int base = 0)
Construct an mpq_class converted from a string using mpq_set_str (see Section 6.1 [Initial
izing Rationals], page 44).
If the string is not a valid rational, an std::invalid_argument exception is thrown. The
same applies to operator=.
[Function] void mpq_class::canonicalize ()
Put an mpq_class into canonical form, as per Chapter 6 [Rational Number Functions],
page 44. All arithmetic operators require their operands in canonical form, and will return
results in canonical form.
[Function] mpq_class abs (mpq class op)
[Function] int cmp (mpq class op1, type op2)
[Function] int cmp (type op1, mpq class op2)
[Function] double mpq_class::get_d (void)
[Function] string mpq_class::get_str (int base = 10)
[Function] int mpq_class::set_str (const char *str, int base)
[Function] int mpq_class::set_str (const string& str, int base)
[Function] int sgn (mpq class op)
These functions provide a C++ class interface to the corresponding GMP C routines.
cmp can be used with any of the classes or the standard C++ types, except long long and
long double.
[Function] mpz_class& mpq_class::get_num ()
[Function] mpz_class& mpq_class::get_den ()
Get a reference to an mpz_class which is the numerator or denominator of an mpq_class.
This can be used both for read and write access. If the object returned is modifed, it modifes
the original mpq_class.
If direct manipulation might produce a noncanonical value, then mpq_class::canonicalize
must be called before further operations.
[Function] mpz_t mpq_class::get_num_mpz_t ()
[Function] mpz_t mpq_class::get_den_mpz_t ()
Get a reference to the underlying mpz_t numerator or denominator of an mpq_class. This
can be passed to C functions expecting an mpz_t. Any modifcations made to the mpz_t will
modify the original mpq_class.
If direct manipulation might produce a noncanonical value, then mpq_class::canonicalize
must be called before further operations.
80 GNU MP 5.0.1
[Function] istream& operator>> (istream& stream, mpq class& rop);
Read rop from stream, using its ios formatting settings, the same as mpq_t operator>> (see
Section 11.3 [C++ Formatted Input], page 74).
If the rop read might not be in canonical form then mpq_class::canonicalize must be
called.
12.4 C++ Interface Floats
When an expression requires the use of temporary intermediate mpf_class values, like
f=g*h+x*y, those temporaries will have the same precision as the destination f. Explicit con
structors can be used if this doesn’t suit.
[Function] mpf_class::mpf_class (type op)
[Function] mpf_class::mpf_class (type op, unsigned long prec)
Construct an mpf_class. Any standard C++ type can be used, except long long and long
double, and any of the GMP C++ classes can be used.
If prec is given, the initial precision is that value, in bits. If prec is not given, then the
initial precision is determined by the type of op given. An mpz_class, mpq_class, or C++
builtin type will give the default mpf precision (see Section 7.1 [Initializing Floats], page 48).
An mpf_class or expression will give the precision of that value. The precision of a binary
expression is the higher of the two operands.
mpf_class f(1.5); // default precision
mpf_class f(1.5, 500); // 500 bits (at least)
mpf_class f(x); // precision of x
mpf_class f(abs(x)); // precision of x
mpf_class f(g, 1000); // 1000 bits (at least)
mpf_class f(x+y); // greater of precisions of x and y
[Function] void mpf_class::mpf_class (const char *s)
[Function] void mpf_class::mpf_class (const char *s, unsigned long prec, int base
= 0)
[Function] void mpf_class::mpf_class (const string& s)
[Function] void mpf_class::mpf_class (const string& s, unsigned long prec, int
base = 0)
Construct an mpf_class converted from a string using mpf_set_str (see Section 7.2 [As
signing Floats], page 50). If prec is given, the initial precision is that value, in bits. If not,
the default mpf precision (see Section 7.1 [Initializing Floats], page 48) is used.
If the string is not a valid foat, an std::invalid_argument exception is thrown. The same
applies to operator=.
[Function] mpf_class& mpf_class::operator= (type op)
Convert and store the given op value to an mpf_class object. The same types are accepted
as for the constructors above.
Note that operator= only stores a new value, it doesn’t copy or change the precision of the
destination, instead the value is truncated if necessary. This is the same as mpf_set etc.
Note in particular this means for mpf_class a copy constructor is not the same as a default
constructor plus assignment.
mpf_class x (y); // x created with precision of y
Chapter 12: C++ Class Interface 81
mpf_class x; // x created with default precision
x = y; // value truncated to that precision
Applications using templated code may need to be careful about the assumptions the code
makes in this area, when working with mpf_class values of various diferent or nondefault
precisions. For instance implementations of the standard complex template have been seen
in both styles above, though of course complex is normally only actually specifed for use
with the builtin foat types.
[Function] mpf_class abs (mpf class op)
[Function] mpf_class ceil (mpf class op)
[Function] int cmp (mpf class op1, type op2)
[Function] int cmp (type op1, mpf class op2)
[Function] bool mpf_class::fits_sint_p (void)
[Function] bool mpf_class::fits_slong_p (void)
[Function] bool mpf_class::fits_sshort_p (void)
[Function] bool mpf_class::fits_uint_p (void)
[Function] bool mpf_class::fits_ulong_p (void)
[Function] bool mpf_class::fits_ushort_p (void)
[Function] mpf_class floor (mpf class op)
[Function] mpf_class hypot (mpf class op1, mpf class op2)
[Function] double mpf_class::get_d (void)
[Function] long mpf_class::get_si (void)
[Function] string mpf_class::get_str (mp exp t& exp, int base = 10, size t
digits = 0)
[Function] unsigned long mpf_class::get_ui (void)
[Function] int mpf_class::set_str (const char *str, int base)
[Function] int mpf_class::set_str (const string& str, int base)
[Function] int sgn (mpf class op)
[Function] mpf_class sqrt (mpf class op)
[Function] mpf_class trunc (mpf class op)
These functions provide a C++ class interface to the corresponding GMP C routines.
cmp can be used with any of the classes or the standard C++ types, except long long and
long double.
The accuracy provided by hypot is not currently guaranteed.
[Function] mp_bitcnt_t mpf_class::get_prec ()
[Function] void mpf_class::set_prec (mp bitcnt t prec)
[Function] void mpf_class::set_prec_raw (mp bitcnt t prec)
Get or set the current precision of an mpf_class.
The restrictions described for mpf_set_prec_raw (see Section 7.1 [Initializing Floats],
page 48) apply to mpf_class::set_prec_raw. Note in particular that the mpf_class must
be restored to it’s allocated precision before being destroyed. This must be done by applica
tion code, there’s no automatic mechanism for it.
12.5 C++ Interface Random Numbers
[Class] gmp_randclass
The C++ class interface to the GMP random number functions uses gmp_randclass to hold
an algorithm selection and current state, as per gmp_randstate_t.
82 GNU MP 5.0.1
[Function] gmp_randclass::gmp_randclass (void (*randinit) (gmp randstate t,
. . . ), . . . )
Construct a gmp_randclass, using a call to the given randinit function (see Section 9.1
[Random State Initialization], page 65). The arguments expected are the same as randinit,
but with mpz_class instead of mpz_t. For example,
gmp_randclass r1 (gmp_randinit_default);
gmp_randclass r2 (gmp_randinit_lc_2exp_size, 32);
gmp_randclass r3 (gmp_randinit_lc_2exp, a, c, m2exp);
gmp_randclass r4 (gmp_randinit_mt);
gmp_randinit_lc_2exp_size will fail if the size requested is too big, an std::length_error
exception is thrown in that case.
[Function] gmp_randclass::gmp_randclass (gmp randalg t alg, . . . )
Construct a gmp_randclass using the same parameters as gmp_randinit (see Section 9.1
[Random State Initialization], page 65). This function is obsolete and the above randinit
style should be preferred.
[Function] void gmp_randclass::seed (unsigned long int s)
[Function] void gmp_randclass::seed (mpz class s)
Seed a random number generator. See see Chapter 9 [Random Number Functions], page 65,
for how to choose a good seed.
[Function] mpz_class gmp_randclass::get_z_bits (unsigned long bits)
[Function] mpz_class gmp_randclass::get_z_bits (mpz class bits)
Generate a random integer with a specifed number of bits.
[Function] mpz_class gmp_randclass::get_z_range (mpz class n)
Generate a random integer in the range 0 to n −1 inclusive.
[Function] mpf_class gmp_randclass::get_f ()
[Function] mpf_class gmp_randclass::get_f (unsigned long prec)
Generate a random foat f in the range 0 <= f < 1. f will be to prec bits precision, or if
prec is not given then to the precision of the destination. For example,
gmp_randclass r;
...
mpf_class f (0, 512); // 512 bits precision
f = r.get_f(); // random number, 512 bits
12.6 C++ Interface Limitations
mpq_class and Templated Reading
A generic piece of template code probably won’t know that mpq_class requires a
canonicalize call if inputs read with operator>> might be noncanonical. This
can lead to incorrect results.
operator>> behaves as it does for reasons of efciency. A canonicalize can be quite
time consuming on large operands, and is best avoided if it’s not necessary.
But this potential difculty reduces the usefulness of mpq_class. Perhaps a mech
anism to tell operator>> what to do will be adopted in the future, maybe a pre
processor defne, a global fag, or an ios fag pressed into service. Or maybe, at
the risk of inconsistency, the mpq_class operator>> could canonicalize and leave
Chapter 12: C++ Class Interface 83
mpq_t operator>> not doing so, for use on those occasions when that’s acceptable.
Send feedback or alternate ideas to gmpbugs@gmplib.org.
Subclassing
Subclassing the GMP C++ classes works, but is not currently recommended.
Expressions involving subclasses resolve correctly (or seem to), but in normal C++
fashion the subclass doesn’t inherit constructors and assignments. There’s many of
those in the GMP classes, and a good way to reestablish them in a subclass is not
yet provided.
Templated Expressions
A subtle difculty exists when using expressions together with applicationdefned
template functions. Consider the following, with T intended to be some numeric
type,
template <class T>
T fun (const T &, const T &);
When used with, say, plain mpz_class variables, it works fne: T is resolved as
mpz_class.
mpz_class f(1), g(2);
fun (f, g); // Good
But when one of the arguments is an expression, it doesn’t work.
mpz_class f(1), g(2), h(3);
fun (f, g+h); // Bad
This is because g+h ends up being a certain expression template type internal to
gmpxx.h, which the C++ template resolution rules are unable to automatically con
vert to mpz_class. The workaround is simply to add an explicit cast.
mpz_class f(1), g(2), h(3);
fun (f, mpz_class(g+h)); // Good
Similarly, within fun it may be necessary to cast an expression to type T when
calling a templated fun2.
template <class T>
void fun (T f, T g)
{
fun2 (f, f+g); // Bad
}
template <class T>
void fun (T f, T g)
{
fun2 (f, T(f+g)); // Good
}
84 GNU MP 5.0.1
13 Berkeley MP Compatible Functions
These functions are intended to be fully compatible with the Berkeley MP library which is
available on many BSD derived U*ix systems. The ‘enablempbsd’ option must be used
when building GNU MP to make these available (see Chapter 2 [Installing GMP], page 3).
The original Berkeley MP library has a usage restriction: you cannot use the same variable as
both source and destination in a single function call. The compatible functions in GNU MP do
not share this restriction—inputs and outputs may overlap.
It is not recommended that new programs are written using these functions. Apart from the
incomplete set of functions, the interface for initializing MINT objects is more error prone, and
the pow function collides with pow in ‘libm.a’.
Include the header ‘mp.h’ to get the defnition of the necessary types and functions. If you are
on a BSD derived system, make sure to include GNU ‘mp.h’ if you are going to link the GNU
‘libmp.a’ to your program. This means that you probably need to give the ‘I<dir>’ option to
the compiler, where ‘<dir>’ is the directory where you have GNU ‘mp.h’.
[Function] MINT * itom (signed short int initial_value)
Allocate an integer consisting of a MINT object and dynamic limb space. Initialize the integer
to initial value. Return a pointer to the MINT object.
[Function] MINT * xtom (char *initial_value)
Allocate an integer consisting of a MINT object and dynamic limb space. Initialize the integer
from initial value, a hexadecimal, nullterminated C string. Return a pointer to the MINT
object.
[Function] void move (MINT *src, MINT *dest)
Set dest to src by copying. Both variables must be previously initialized.
[Function] void madd (MINT *src_1, MINT *src_2, MINT *destination)
Add src 1 and src 2 and put the sum in destination.
[Function] void msub (MINT *src_1, MINT *src_2, MINT *destination)
Subtract src 2 from src 1 and put the diference in destination.
[Function] void mult (MINT *src_1, MINT *src_2, MINT *destination)
Multiply src 1 and src 2 and put the product in destination.
[Function] void mdiv (MINT *dividend, MINT *divisor, MINT *quotient, MINT
*remainder)
[Function] void sdiv (MINT *dividend, signed short int divisor, MINT *quotient,
signed short int *remainder)
Set quotient to dividend/divisor, and remainder to dividend mod divisor. The quotient is
rounded towards zero; the remainder has the same sign as the dividend unless it is zero.
Some implementations of these functions work diferently—or not at all—for negative argu
ments.
[Function] void msqrt (MINT *op, MINT *root, MINT *remainder)
Set root to
√
op, like mpz_sqrt. Set remainder to (op −root
2
), i.e. zero if op is a perfect
square.
If root and remainder are the same variable, the results are undefned.
Chapter 13: Berkeley MP Compatible Functions 85
[Function] void pow (MINT *base, MINT *exp, MINT *mod, MINT *dest)
Set dest to (base raised to exp) modulo mod.
Note that the name pow clashes with pow from the standard C math library (see Section
“Exponentiation and Logarithms” in The GNU C Library Reference Manual). An application
will only be able to use one or the other.
[Function] void rpow (MINT *base, signed short int exp, MINT *dest)
Set dest to base raised to exp.
[Function] void gcd (MINT *op1, MINT *op2, MINT *res)
Set res to the greatest common divisor of op1 and op2.
[Function] int mcmp (MINT *op1, MINT *op2)
Compare op1 and op2. Return a positive value if op1 > op2, zero if op1 = op2, and a negative
value if op1 < op2.
[Function] void min (MINT *dest)
Input a decimal string from stdin, and put the read integer in dest. SPC and TAB are
allowed in the number string, and are ignored.
[Function] void mout (MINT *src)
Output src to stdout, as a decimal string. Also output a newline.
[Function] char * mtox (MINT *op)
Convert op to a hexadecimal string, and return a pointer to the string. The returned string
is allocated using the default memory allocation function, malloc by default. It will be
strlen(str)+1 bytes, that being exactly enough for the string and nullterminator.
[Function] void mfree (MINT *op)
Deallocate, the space used by op. This function should only be passed a value returned by
itom or xtom.
86 GNU MP 5.0.1
14 Custom Allocation
By default GMP uses malloc, realloc and free for memory allocation, and if they fail GMP
prints a message to the standard error output and terminates the program.
Alternate functions can be specifed, to allocate memory in a diferent way or to have a diferent
error action on running out of memory.
This feature is available in the Berkeley compatibility library (see Chapter 13 [BSD Compatible
Functions], page 84) as well as the main GMP library.
[Function] void mp_set_memory_functions (
void *(*alloc_func_ptr) (size t),
void *(*realloc_func_ptr) (void *, size t, size t),
void (*free_func_ptr) (void *, size t))
Replace the current allocation functions from the arguments. If an argument is NULL, the
corresponding default function is used.
These functions will be used for all memory allocation done by GMP, apart from temporary
space from alloca if that function is available and GMP is confgured to use it (see Section 2.1
[Build Options], page 3).
Be sure to call mp_set_memory_functions only when there are no active GMP objects
allocated using the previous memory functions! Usually that means calling it before any
other GMP function.
The functions supplied should ft the following declarations:
[Function] void * allocate_function (size t alloc_size)
Return a pointer to newly allocated space with at least alloc size bytes.
[Function] void * reallocate_function (void *ptr, size t old_size, size t
new_size)
Resize a previously allocated block ptr of old size bytes to be new size bytes.
The block may be moved if necessary or if desired, and in that case the smaller of old size
and new size bytes must be copied to the new location. The return value is a pointer to the
resized block, that being the new location if moved or just ptr if not.
ptr is never NULL, it’s always a previously allocated block. new size may be bigger or smaller
than old size.
[Function] void free_function (void *ptr, size t size)
Deallocate the space pointed to by ptr.
ptr is never NULL, it’s always a previously allocated block of size bytes.
A byte here means the unit used by the sizeof operator.
The old size parameters to reallocate function and free function are passed for convenience,
but of course can be ignored if not needed. The default functions using malloc and friends for
instance don’t use them.
No error return is allowed from any of these functions, if they return then they must have per
formed the specifed operation. In particular note that allocate function or reallocate function
mustn’t return NULL.
Chapter 14: Custom Allocation 87
Getting a diferent fatal error action is a good use for custom allocation functions, for example
giving a graphical dialog rather than the default print to stderr. How much is possible when
genuinely out of memory is another question though.
There’s currently no defned way for the allocation functions to recover from an error such as out
of memory, they must terminate program execution. A longjmp or throwing a C++ exception
will have undefned results. This may change in the future.
GMP may use allocated blocks to hold pointers to other allocated blocks. This will limit the
assumptions a conservative garbage collection scheme can make.
Since the default GMP allocation uses malloc and friends, those functions will be linked in even
if the frst thing a program does is an mp_set_memory_functions. It’s necessary to change the
GMP sources if this is a problem.
[Function] void mp_get_memory_functions (
void *(**alloc_func_ptr) (size t),
void *(**realloc_func_ptr) (void *, size t, size t),
void (**free_func_ptr) (void *, size t))
Get the current allocation functions, storing function pointers to the locations given by the
arguments. If an argument is NULL, that function pointer is not stored.
For example, to get just the current free function,
void (*freefunc) (void *, size_t);
mp_get_memory_functions (NULL, NULL, &freefunc);
88 GNU MP 5.0.1
15 Language Bindings
The following packages and projects ofer access to GMP from languages other than C, though
perhaps with varying levels of functionality and efciency.
C++
• GMP C++ class interface, see Chapter 12 [C++ Class Interface], page 76
Straightforward interface, expression templates to eliminate temporaries.
• ALP http://wwwsop.inria.fr/saga/logiciels/ALP/
Linear algebra and polynomials using templates.
• Arithmos http://www.win.ua.ac.be/~cant/arithmos/
Rationals with infnities and square roots.
• CLN http://www.ginac.de/CLN/
High level classes for arithmetic.
• LiDIA http://www.cdc.informatik.tudarmstadt.de/TI/LiDIA/
A C++ library for computational number theory.
• Linbox http://www.linalg.org/
Sparse vectors and matrices.
• NTL http://www.shoup.net/ntl/
A C++ number theory library.
Fortran
• Omni F77 http://phase.hpcc.jp/Omni/home.html
Arbitrary precision foats.
Haskell
• Glasgow Haskell Compiler http://www.haskell.org/ghc/
Java
• Kafe http://www.kaffe.org/
• Kissme http://kissme.sourceforge.net/
Lisp
• GNU Common Lisp http://www.gnu.org/software/gcl/gcl.html
• Librep http://librep.sourceforge.net/
• XEmacs (21.5.18 beta and up) http://www.xemacs.org
Optional big integers, rationals and foats using GMP.
M4
• GNU m4 betas http://www.seindal.dk/rene/gnu/
Optionally provides an arbitrary precision mpeval.
ML
• MLton compiler http://mlton.org/
Objective Caml
• MLGMP http://www.di.ens.fr/~monniaux/programmes.html.en
• Numerix http://pauillac.inria.fr/~quercia/
Optionally using GMP.
Oz
• Mozart http://www.mozartoz.org/
Chapter 15: Language Bindings 89
Pascal
• GNU Pascal Compiler http://www.gnupascal.de/
GMP unit.
• Numerix http://pauillac.inria.fr/~quercia/
For Free Pascal, optionally using GMP.
Perl
• GMP module, see ‘demos/perl’ in the GMP sources (see Section 3.10 [Demon
stration Programs], page 20).
• Math::GMP http://www.cpan.org/
Compatible with Math::BigInt, but not as many functions as the GMP module
above.
• Math::BigInt::GMP http://www.cpan.org/
Plug Math::GMP into normal Math::BigInt operations.
Pike
• mpz module in the standard distribution, http://pike.ida.liu.se/
Prolog
• SWI Prolog http://www.swiprolog.org/
Arbitrary precision foats.
Python
• mpz module in the standard distribution, http://www.python.org/
• GMPY http://gmpy.sourceforge.net/
Scheme
• GNU Guile (upcoming 1.8) http://www.gnu.org/software/guile/guile.html
• RScheme http://www.rscheme.org/
• STklos http://www.stklos.org/
Smalltalk
• GNU Smalltalk http://www.smalltalk.org/versions/GNUSmalltalk.html
Other
• Axiom http://savannah.nongnu.org/projects/axiom
Computer algebra using GCL.
• DrGenius http://drgenius.seul.org/
Geometry system and mathematical programming language.
• GiNaC http://www.ginac.de/
C++ computer algebra using CLN.
• GOO http://www.googoogaga.org/
Dynamic object oriented language.
• Maxima http://www.ma.utexas.edu/users/wfs/maxima.html
Macsyma computer algebra using GCL.
• Q http://qlang.sourceforge.net/
Equational programming system.
• Regina http://regina.sourceforge.net/
Topological calculator.
• Yacas http://www.xs4all.nl/~apinkus/yacas.html
Yet another computer algebra system.
90 GNU MP 5.0.1
16 Algorithms
This chapter is an introduction to some of the algorithms used for various GMP operations.
The code is likely to be hard to understand without knowing something about the algorithms.
Some GMP internals are mentioned, but applications that expect to be compatible with future
GMP releases should take care to use only the documented functions.
16.1 Multiplication
NN limb multiplications and squares are done using one of fve algorithms, as the size N
increases.
Algorithm Threshold
Basecase (none)
Karatsuba MUL_TOOM22_THRESHOLD
Toom3 MUL_TOOM33_THRESHOLD
Toom4 MUL_TOOM44_THRESHOLD
FFT MUL_FFT_THRESHOLD
Similarly for squaring, with the SQR thresholds.
NM multiplications of operands with diferent sizes above MUL_TOOM22_THRESHOLD are cur
rently done by special Toominspired algorithms or directly with FFT, depending on operand
size (see Section 16.1.7 [Unbalanced Multiplication], page 96).
16.1.1 Basecase Multiplication
Basecase NM multiplication is a straightforward rectangular set of crossproducts, the same
as long multiplication done by hand and for that reason sometimes known as the schoolbook or
grammar school method. This is an O(NM) algorithm. See Knuth section 4.3.1 algorithm M
(see Appendix B [References], page 122), and the ‘mpn/generic/mul_basecase.c’ code.
Assembly implementations of mpn_mul_basecase are essentially the same as the generic C code,
but have all the usual assembly tricks and obscurities introduced for speed.
A square can be done in roughly half the time of a multiply, by using the fact that the cross
products above and below the diagonal are the same. A triangle of products below the diagonal is
formed, doubled (left shift by one bit), and then the products on the diagonal added. This can be
seen in ‘mpn/generic/sqr_basecase.c’. Again the assembly implementations take essentially
the same approach.
u0
u1
u2
u3
u4
u0 u1 u2 u3 u4
d
d
d
d
d
In practice squaring isn’t a full 2 faster than multiplying, it’s usually around 1.5. Less than
1.5 probably indicates mpn_sqr_basecase wants improving on that CPU.
Chapter 16: Algorithms 91
On some CPUs mpn_mul_basecase can be faster than the generic C mpn_sqr_basecase on some
small sizes. SQR_BASECASE_THRESHOLD is the size at which to use mpn_sqr_basecase, this will
be zero if that routine should be used always.
16.1.2 Karatsuba Multiplication
The Karatsuba multiplication algorithm is described in Knuth section 4.3.3 part A, and various
other textbooks. A brief description is given here.
The inputs x and y are treated as each split into two parts of equal length (or the most signifcant
part one limb shorter if N is odd).
high low
x
1
x
0
y
1
y
0
Let b be the power of 2 where the split occurs, ie. if x
0
is k limbs (y
0
the same) then b =
2
k∗mp bits per limb
. With that x = x
1
b +x
0
and y = y
1
b +y
0
, and the following holds,
xy = (b
2
+b)x
1
y
1
−b(x
1
−x
0
)(y
1
−y
0
) + (b + 1)x
0
y
0
This formula means doing only three multiplies of (N/2)(N/2) limbs, whereas a basecase
multiply of NN limbs is equivalent to four multiplies of (N/2)(N/2). The factors (b
2
+b) etc
represent the positions where the three products must be added.
high low
x
1
y
1
x
0
y
0
+ x
1
y
1
+ x
0
y
0
− (x
1
−x
0
)(y
1
−y
0
)
The term (x
1
−x
0
)(y
1
−y
0
) is best calculated as an absolute value, and the sign used to choose
to add or subtract. Notice the sum high(x
0
y
0
) +low(x
1
y
1
) occurs twice, so it’s possible to do 5k
limb additions, rather than 6k, but in GMP extra function call overheads outweigh the saving.
Squaring is similar to multiplying, but with x = y the formula reduces to an equivalent with
three squares,
x
2
= (b
2
+b)x
2
1
−b(x
1
−x
0
)
2
+ (b + 1)x
2
0
The fnal result is accumulated from those three squares the same way as for the three multiplies
above. The middle term (x
1
−x
0
)
2
is now always positive.
A similar formula for both multiplying and squaring can be constructed with a middle term
(x
1
+ x
0
)(y
1
+ y
0
). But those sums can exceed k limbs, leading to more carry handling and
additions than the form above.
Karatsuba multiplication is asymptotically an O(N
1.585
) algorithm, the exponent being
log 3/ log 2, representing 3 multiplies each 1/2 the size of the inputs. This is a big improvement
over the basecase multiply at O(N
2
) and the advantage soon overcomes the extra additions
Karatsuba performs. MUL_TOOM22_THRESHOLD can be as little as 10 limbs. The SQR threshold is
usually about twice the MUL.
The basecase algorithm will take a time of the form M(N) = aN
2
+bN +c and the Karatsuba
algorithm K(N) = 3M(N/2)+dN+e, which expands to K(N) =
3
4
aN
2
+
3
2
bN+3c+dN+e. The
92 GNU MP 5.0.1
factor
3
4
for a means percrossproduct speedups in the basecase code will increase the threshold
since they beneft M(N) more than K(N). And conversely the
3
2
for b means linear style
speedups of b will increase the threshold since they beneft K(N) more than M(N). The latter
can be seen for instance when adding an optimized mpn_sqr_diagonal to mpn_sqr_basecase.
Of course all speedups reduce total time, and in that sense the algorithm thresholds are merely
of academic interest.
16.1.3 Toom 3Way Multiplication
The Karatsuba formula is the simplest case of a general approach to splitting inputs that leads
to both Toom and FFT algorithms. A description of Toom can be found in Knuth section 4.3.3,
with an example 3way calculation after Theorem A. The 3way form used in GMP is described
here.
The operands are each considered split into 3 pieces of equal length (or the most signifcant part
1 or 2 limbs shorter than the other two).
high low
x
2
x
1
x
0
y
2
y
1
y
0
These parts are treated as the coefcients of two polynomials
X(t) = x
2
t
2
+x
1
t +x
0
Y (t) = y
2
t
2
+y
1
t +y
0
Let b equal the power of 2 which is the size of the x
0
, x
1
, y
0
and y
1
pieces, ie. if they’re k limbs
each then b = 2
k∗mp bits per limb
. With this x = X(b) and y = Y (b).
Let a polynomial W(t) = X(t)Y (t) and suppose its coefcients are
W(t) = w
4
t
4
+w
3
t
3
+w
2
t
2
+w
1
t +w
0
The w
i
are going to be determined, and when they are they’ll give the fnal result using w = W(b),
since xy = X(b)Y (b). The coefcients will be roughly b
2
each, and the fnal W(b) will be an
addition like,
high low
w
4
w
3
w
2
w
1
w
0
The w
i
coefcients could be formed by a simple set of cross products, like w
4
= x
2
y
2
, w
3
=
x
2
y
1
+x
1
y
2
, w
2
= x
2
y
0
+x
1
y
1
+x
0
y
2
etc, but this would need all nine x
i
y
j
for i, j = 0, 1, 2, and
would be equivalent merely to a basecase multiply. Instead the following approach is used.
X(t) and Y (t) are evaluated and multiplied at 5 points, giving values of W(t) at those points.
In GMP the following points are used,
Point Value
t = 0 x
0
y
0
, which gives w
0
immediately
t = 1 (x
2
+x
1
+x
0
)(y
2
+y
1
+y
0
)
Chapter 16: Algorithms 93
t = −1 (x
2
−x
1
+x
0
)(y
2
−y
1
+y
0
)
t = 2 (4x
2
+ 2x
1
+x
0
)(4y
2
+ 2y
1
+y
0
)
t = ∞ x
2
y
2
, which gives w
4
immediately
At t = −1 the values can be negative and that’s handled using the absolute values and tracking
the sign separately. At t = ∞ the value is actually lim
t→∞
X(t)Y (t)
t
4
, but it’s much easier to think
of as simply x
2
y
2
giving w
4
immediately (much like x
0
y
0
at t = 0 gives w
0
immediately).
Each of the points substituted into W(t) = w
4
t
4
+ +w
0
gives a linear combination of the w
i
coefcients, and the value of those combinations has just been calculated.
W(0) = w
0
W(1) = w
4
+ w
3
+ w
2
+ w
1
+ w
0
W(−1) = w
4
− w
3
+ w
2
− w
1
+ w
0
W(2) = 16w
4
+ 8w
3
+ 4w
2
+ 2w
1
+ w
0
W(∞) = w
4
This is a set of fve equations in fve unknowns, and some elementary linear algebra quickly
isolates each w
i
. This involves adding or subtracting one W(t) value from another, and a couple
of divisions by powers of 2 and one division by 3, the latter using the special mpn_divexact_by3
(see Section 16.2.5 [Exact Division], page 98).
The conversion of W(t) values to the coefcients is interpolation. A polynomial of degree 4 like
W(t) is uniquely determined by values known at 5 diferent points. The points are arbitrary and
can be chosen to make the linear equations come out with a convenient set of steps for quickly
isolating the w
i
.
Squaring follows the same procedure as multiplication, but there’s only one X(t) and it’s evalu
ated at the 5 points, and those values squared to give values of W(t). The interpolation is then
identical, and in fact the same toom3_interpolate subroutine is used for both squaring and
multiplying.
Toom3 is asymptotically O(N
1.465
), the exponent being log 5/ log 3, representing 5 recursive
multiplies of 1/3 the original size each. This is an improvement over Karatsuba at O(N
1.585
),
though Toom does more work in the evaluation and interpolation and so it only realizes its
advantage above a certain size.
Near the crossover between Toom3 and Karatsuba there’s generally a range of sizes where the
diference between the two is small. MUL_TOOM33_THRESHOLD is a somewhat arbitrary point in
that range and successive runs of the tune program can give diferent values due to small varia
tions in measuring. A graph of time versus size for the two shows the efect, see ‘tune/README’.
At the fairly small sizes where the Toom3 thresholds occur it’s worth remembering that the
asymptotic behaviour for Karatsuba and Toom3 can’t be expected to make accurate predictions,
due of course to the big infuence of all sorts of overheads, and the fact that only a few recursions
of each are being performed. Even at large sizes there’s a good chance machine dependent efects
like cache architecture will mean actual performance deviates from what might be predicted.
The formula given for the Karatsuba algorithm (see Section 16.1.2 [Karatsuba Multiplication],
page 91) has an equivalent for Toom3 involving only fve multiplies, but this would be compli
cated and unenlightening.
An alternate view of Toom3 can be found in Zuras (see Appendix B [References], page 122),
using a vector to represent the x and y splits and a matrix multiplication for the evaluation
and interpolation stages. The matrix inverses are not meant to be actually used, and they have
elements with values much greater than in fact arise in the interpolation steps. The diagram
94 GNU MP 5.0.1
shown for the 3way is attractive, but again doesn’t have to be implemented that way and for
example with a bit of rearrangement just one division by 6 can be done.
16.1.4 Toom 4Way Multiplication
Karatsuba and Toom3 split the operands into 2 and 3 coefcients, respectively. Toom4 anal
ogously splits the operands into 4 coefcients. Using the notation from the section on Toom3
multiplication, we form two polynomials:
X(t) = x
3
t
3
+x
2
t
2
+x
1
t +x
0
Y (t) = y
3
t
3
+y
2
t
2
+y
1
t +y
0
X(t) and Y (t) are evaluated and multiplied at 7 points, giving values of W(t) at those points.
In GMP the following points are used,
Point Value
t = 0 x
0
y
0
, which gives w
0
immediately
t = 1/2 (x
3
+ 2x
2
+ 4x
1
+ 8x
0
)(y
3
+ 2y
2
+ 4y
1
+ 8y
0
)
t = −1/2 (−x
3
+ 2x
2
−4x
1
+ 8x
0
)(−y
3
+ 2y
2
−4y
1
+ 8y
0
)
t = 1 (x
3
+x
2
+x
1
+x
0
)(y
3
+y
2
+y
1
+y
0
)
t = −1 (−x
3
+x
2
−x
1
+x
0
)(−y
3
+y
2
−y
1
+y
0
)
t = 2 (8x
3
+ 4x
2
+ 2x
1
+x
0
)(8y
3
+ 4y
2
+ 2y
1
+y
0
)
t = ∞ x
3
y
3
, which gives w
6
immediately
The number of additions and subtractions for Toom4 is much larger than for Toom3. But
several subexpressions occur multiple times, for example x
2
+ x
0
, occurs for both t = 1 and
t = −1.
Toom4 is asymptotically O(N
1.404
), the exponent being log 7/ log 4, representing 7 recursive
multiplies of 1/4 the original size each.
16.1.5 FFT Multiplication
At large to very large sizes a Fermat style FFT multiplication is used, following Sch¨onhage and
Strassen (see Appendix B [References], page 122). Descriptions of FFTs in various forms can
be found in many textbooks, for instance Knuth section 4.3.3 part C or Lipson chapter IX. A
brief description of the form used in GMP is given here.
The multiplication done is xy mod 2
N
+ 1, for a given N. A full product xy is obtained by
choosing N ≥ bits(x) +bits(y) and padding x and y with high zero limbs. The modular product
is the native form for the algorithm, so padding to get a full product is unavoidable.
The algorithm follows a split, evaluate, pointwise multiply, interpolate and combine similar to
that described above for Karatsuba and Toom3. A k parameter controls the split, with an FFT
k splitting into 2
k
pieces of M = N/2
k
bits each. N must be a multiple of 2
k
mp bits per limb
so the split falls on limb boundaries, avoiding bit shifts in the split and combine stages.
The evaluations, pointwise multiplications, and interpolation, are all done modulo 2
N
+1 where
N
is 2M + k + 3 rounded up to a multiple of 2
k
and of mp_bits_per_limb. The results of
interpolation will be the following negacyclic convolution of the input pieces, and the choice of
N
ensures these sums aren’t truncated.
w
n
=
i+j=b2
k
+n
b=0,1
(−1)
b
x
i
y
j
The points used for the evaluation are g
i
for i = 0 to 2
k
−1 where g = 2
2N
/2
k
. g is a 2
k
th root
of unity mod 2
N
+ 1, which produces necessary cancellations at the interpolation stage, and
Chapter 16: Algorithms 95
it’s also a power of 2 so the fast Fourier transforms used for the evaluation and interpolation do
only shifts, adds and negations.
The pointwise multiplications are done modulo 2
N
+ 1 and either recurse into a further FFT
or use a plain multiplication (Toom3, Karatsuba or basecase), whichever is optimal at the size
N
. The interpolation is an inverse fast Fourier transform. The resulting set of sums of x
i
y
j
are
added at appropriate ofsets to give the fnal result.
Squaring is the same, but x is the only input so it’s one transform at the evaluate stage and the
pointwise multiplies are squares. The interpolation is the same.
For a mod 2
N
+ 1 product, an FFTk is an O(N
k/(k−1)
) algorithm, the exponent representing
2
k
recursed modular multiplies each 1/2
k−1
the size of the original. Each successive k is an
asymptotic improvement, but overheads mean each is only faster at bigger and bigger sizes. In
the code, MUL_FFT_TABLE and SQR_FFT_TABLE are the thresholds where each k is used. Each
new k efectively swaps some multiplying for some shifts, adds and overheads.
A mod 2
N
+1 product can be formed with a normal NN →2N bit multiply plus a subtraction,
so an FFT and Toom3 etc can be compared directly. A k = 4 FFT at O(N
1.333
) can be expected
to be the frst faster than Toom3 at O(N
1.465
). In practice this is what’s found, with MUL_FFT_
MODF_THRESHOLD and SQR_FFT_MODF_THRESHOLD being between 300 and 1000 limbs, depending
on the CPU. So far it’s been found that only very large FFTs recurse into pointwise multiplies
above these sizes.
When an FFT is to give a full product, the change of N to 2N doesn’t alter the theoretical
complexity for a given k, but for the purposes of considering where an FFT might be frst used
it can be assumed that the FFT is recursing into a normal multiply and that on that basis it’s
doing 2
k
recursed multiplies each 1/2
k−2
the size of the inputs, making it O(N
k/(k−2)
). This
would mean k = 7 at O(N
1.4
) would be the frst FFT faster than Toom3. In practice MUL_
FFT_THRESHOLD and SQR_FFT_THRESHOLD have been found to be in the k = 8 range, somewhere
between 3000 and 10000 limbs.
The way N is split into 2
k
pieces and then 2M + k + 3 is rounded up to a multiple of 2
k
and
mp_bits_per_limb means that when 2
k
≥ mp bits per limb the efective N is a multiple of
2
2k−1
bits. The +k + 3 means some values of N just under such a multiple will be rounded
to the next. The complexity calculations above assume that a favourable size is used, meaning
one which isn’t padded through rounding, and it’s also assumed that the extra +k + 3 bits are
negligible at typical FFT sizes.
The practical efect of the 2
2k−1
constraint is to introduce a stepefect into measured speeds.
For example k = 8 will round N up to a multiple of 32768 bits, so for a 32bit limb there’ll be
512 limb groups of sizes for which mpn_mul_n runs at the same speed. Or for k = 9 groups of
2048 limbs, k = 10 groups of 8192 limbs, etc. In practice it’s been found each k is used at quite
small multiples of its size constraint and so the step efect is quite noticeable in a time versus
size graph.
The threshold determinations currently measure at the midpoints of size steps, but this is sub
optimal since at the start of a new step it can happen that it’s better to go back to the previous
k for a while. Something more sophisticated for MUL_FFT_TABLE and SQR_FFT_TABLE will be
needed.
16.1.6 Other Multiplication
The Toom algorithms described above (see Section 16.1.3 [Toom 3Way Multiplication], page 92,
see Section 16.1.4 [Toom 4Way Multiplication], page 94) generalizes to split into an arbitrary
96 GNU MP 5.0.1
number of pieces, as per Knuth section 4.3.3 algorithm C. This is not currently used. The notes
here are merely for interest.
In general a split into r + 1 pieces is made, and evaluations and pointwise multiplications done
at 2r +1 points. A 4way split does 7 pointwise multiplies, 5way does 9, etc. Asymptotically an
(r +1)way algorithm is O(N
log(2r+1)/log(r+1)
. Only the pointwise multiplications count towards
bigO complexity, but the time spent in the evaluate and interpolate stages grows with r and has
a signifcant practical impact, with the asymptotic advantage of each r realized only at bigger
and bigger sizes. The overheads grow as O(Nr), whereas in an r = 2
k
FFT they grow only as
O(N log r).
Knuth algorithm C evaluates at points 0,1,2,. . . ,2r, but exercise 4 uses −r,. . . ,0,. . . ,r and the
latter saves some small multiplies in the evaluate stage (or rather trades them for additions),
and has a further saving of nearly half the interpolate steps. The idea is to separate odd and
even fnal coefcients and then perform algorithm C steps C7 and C8 on them separately. The
divisors at step C7 become j
2
and the multipliers at C8 become 2tj −j
2
.
Splitting odd and even parts through positive and negative points can be thought of as using −1
as a square root of unity. If a 4th root of unity was available then a further split and speedup
would be possible, but no such root exists for plain integers. Going to complex integers with
i =
√
−1 doesn’t help, essentially because in Cartesian form it takes three real multiplies to do
a complex multiply. The existence of 2
k
th roots of unity in a suitable ring or feld lets the fast
Fourier transform keep splitting and get to O(N log r).
Floating point FFTs use complex numbers approximating Nth roots of unity. Some processors
have special support for such FFTs. But these are not used in GMP since it’s very difcult to
guarantee an exact result (to some number of bits). An occasional diference of 1 in the last bit
might not matter to a typical signal processing algorithm, but is of course of vital importance
to GMP.
16.1.7 Unbalanced Multiplication
Multiplication of operands with diferent sizes, both below MUL_TOOM22_THRESHOLD are done
with plain schoolbook multiplication (see Section 16.1.1 [Basecase Multiplication], page 90).
For really large operands, we invoke FFT directly.
For operands between these sizes, we use Toom inspired algorithms suggested by Alberto Zanoni
and Marco Bodrato. The idea is to split the operands into polynomials of diferent degree. GMP
currently splits the smaller operand onto 2 coefcients, i.e., a polynomial of degree 1, but the
larger operand can be split into 2, 3, or 4 coefcients, i.e., a polynomial of degree 1 to 3.
16.2 Division Algorithms
16.2.1 Single Limb Division
N1 division is implemented using repeated 21 divisions from high to low, either with a
hardware divide instruction or a multiplication by inverse, whichever is best on a given CPU.
The multiply by inverse follows “Improved division by invariant integers” by M¨oller and
Granlund (see Appendix B [References], page 122) and is implemented as udiv_qrnnd_preinv
in ‘gmpimpl.h’. The idea is to have a fxedpoint approximation to 1/d (see invert_limb)
and then multiply by the high limb (plus one bit) of the dividend to get a quotient q. With d
normalized (high bit set), q is no more than 1 too small. Subtracting qd from the dividend gives
a remainder, and reveals whether q or q −1 is correct.
Chapter 16: Algorithms 97
The result is a division done with two multiplications and four or fve arithmetic operations. On
CPUs with low latency multipliers this can be much faster than a hardware divide, though the
cost of calculating the inverse at the start may mean it’s only better on inputs bigger than say
4 or 5 limbs.
When a divisor must be normalized, either for the generic C __udiv_qrnnd_c or the multiply
by inverse, the division performed is actually a2
k
by d2
k
where a is the dividend and k is the
power necessary to have the high bit of d2
k
set. The bit shifts for the dividend are usually
accomplished “on the fy” meaning by extracting the appropriate bits at each step. Done this
way the quotient limbs come out aligned ready to store. When only the remainder is wanted,
an alternative is to take the dividend limbs unshifted and calculate r = a mod d2
k
followed by
an extra fnal step r2
k
mod d2
k
. This can help on CPUs with poor bit shifts or few registers.
The multiply by inverse can be done two limbs at a time. The calculation is basically the same,
but the inverse is two limbs and the divisor treated as if padded with a low zero limb. This
means more work, since the inverse will need a 22 multiply, but the four 11s to do that
are independent and can therefore be done partly or wholly in parallel. Likewise for a 21
calculating qd. The net efect is to process two limbs with roughly the same two multiplies
worth of latency that one limb at a time gives. This extends to 3 or 4 limbs at a time, though
the extra work to apply the inverse will almost certainly soon reach the limits of multiplier
throughput.
A similar approach in reverse can be taken to process just half a limb at a time if the divisor is
only a half limb. In this case the 11 multiply for the inverse efectively becomes two
1
2
1 for
each limb, which can be a saving on CPUs with a fast half limb multiply, or in fact if the only
multiply is a half limb, and especially if it’s not pipelined.
16.2.2 Basecase Division
Basecase NM division is like long division done by hand, but in base 2
mp bits per limb
. See
Knuth section 4.3.1 algorithm D, and ‘mpn/generic/sb_divrem_mn.c’.
Briefy stated, while the dividend remains larger than the divisor, a high quotient limb is formed
and the N1 product qd subtracted at the top end of the dividend. With a normalized divisor
(most signifcant bit set), each quotient limb can be formed with a 21 division and a 11
multiplication plus some subtractions. The 21 division is by the high limb of the divisor and
is done either with a hardware divide or a multiply by inverse (the same as in Section 16.2.1
[Single Limb Division], page 96) whichever is faster. Such a quotient is sometimes one too big,
requiring an addback of the divisor, but that happens rarely.
With Q=N−M being the number of quotient limbs, this is an O(QM) algorithm and will run
at a speed similar to a basecase QM multiplication, difering in fact only in the extra multiply
and divide for each of the Q quotient limbs.
16.2.3 Divide and Conquer Division
For divisors larger than DC_DIV_QR_THRESHOLD, division is done by dividing. Or to be precise
by a recursive divide and conquer algorithm based on work by Moenck and Borodin, Jebelean,
and Burnikel and Ziegler (see Appendix B [References], page 122).
The algorithm consists essentially of recognising that a 2NN division can be done with the
basecase division algorithm (see Section 16.2.2 [Basecase Division], page 97), but using N/2
limbs as a base, not just a single limb. This way the multiplications that arise are (N/2)(N/2)
and can take advantage of Karatsuba and higher multiplication algorithms (see Section 16.1
[Multiplication Algorithms], page 90). The two “digits” of the quotient are formed by recursive
N(N/2) divisions.
98 GNU MP 5.0.1
If the (N/2)(N/2) multiplies are done with a basecase multiplication then the work is about the
same as a basecase division, but with more function call overheads and with some subtractions
separated from the multiplies. These overheads mean that it’s only when N/2 is above MUL_
TOOM22_THRESHOLD that divide and conquer is of use.
DC_DIV_QR_THRESHOLD is based on the divisor size N, so it will be somewhere above twice
MUL_TOOM22_THRESHOLD, but how much above depends on the CPU. An optimized mpn_mul_
basecase can lower DC_DIV_QR_THRESHOLD a little by ofering a readymade advantage over
repeated mpn_submul_1 calls.
Divide and conquer is asymptotically O(M(N) log N) where M(N) is the time for an NN
multiplication done with FFTs. The actual time is a sum over multiplications of the recursed
sizes, as can be seen near the end of section 2.2 of Burnikel and Ziegler. For example, within
the Toom3 range, divide and conquer is 2.63M(N). With higher algorithms the M(N) term
improves and the multiplier tends to log N. In practice, at moderate to large sizes, a 2NN
division is about 2 to 4 times slower than an NN multiplication.
16.2.4 BlockWise Barrett Division
For the largest divisions, a blockwise Barrett division algorithm is used. Here, the divisor is
inverted to a precision determined by the relative size of the dividend and divisor. Blocks of
quotient limbs are then generated by multiplying blocks from the dividend by the inverse.
Our blockwise algorithm computes a smaller inverse than in the plain Barrett algorithm. For
a 2n/n division, the inverse will be just n/2 limbs.
16.2.5 Exact Division
A socalled exact division is when the dividend is known to be an exact multiple of the divisor.
Jebelean’s exact division algorithm uses this knowledge to make some signifcant optimizations
(see Appendix B [References], page 122).
The idea can be illustrated in decimal for example with 368154 divided by 543. Because the
low digit of the dividend is 4, the low digit of the quotient must be 8. This is arrived at from
47 mod 10, using the fact 7 is the modular inverse of 3 (the low digit of the divisor), since
37 ≡1 mod 10. So 8543 = 4344 can be subtracted from the dividend leaving 363810. Notice
the low digit has become zero.
The procedure is repeated at the second digit, with the next quotient digit 7 (17 mod 10),
subtracting 7543 = 3801, leaving 325800. And fnally at the third digit with quotient digit 6
(87 mod 10), subtracting 6543 = 3258 leaving 0. So the quotient is 678.
Notice however that the multiplies and subtractions don’t need to extend past the low three
digits of the dividend, since that’s enough to determine the three quotient digits. For the last
quotient digit no subtraction is needed at all. On a 2NN division like this one, only about half
the work of a normal basecase division is necessary.
For an NM exact division producing Q=N−M quotient limbs, the saving over a normal basecase
division is in two parts. Firstly, each of the Q quotient limbs needs only one multiply, not a 21
divide and multiply. Secondly, the crossproducts are reduced when Q > M to QM−M(M+1)/2,
or when Q ≤ M to Q(Q− 1)/2. Notice the savings are complementary. If Q is big then many
divisions are saved, or if Q is small then the crossproducts reduce to a small number.
The modular inverse used is calculated efciently by binvert_limb in ‘gmpimpl.h’. This does
four multiplies for a 32bit limb, or six for a 64bit limb. ‘tune/modlinv.c’ has some alternate
implementations that might suit processors better at bit twiddling than multiplying.
Chapter 16: Algorithms 99
The subquadratic exact division described by Jebelean in “Exact Division with Karatsuba
Complexity” is not currently implemented. It uses a rearrangement similar to the divide and
conquer for normal division (see Section 16.2.3 [Divide and Conquer Division], page 97), but
operating from low to high. A further possibility not currently implemented is “Bidirectional
Exact Integer Division” by Krandick and Jebelean which forms quotient limbs from both the
high and low ends of the dividend, and can halve once more the number of crossproducts needed
in a 2NN division.
A special case exact division by 3 exists in mpn_divexact_by3, supporting Toom3 multiplication
and mpq canonicalizations. It forms quotient digits with a multiply by the modular inverse of 3
(which is 0xAA..AAB) and uses two comparisons to determine a borrow for the next limb. The
multiplications don’t need to be on the dependent chain, as long as the efect of the borrows is
applied, which can help chips with pipelined multipliers.
16.2.6 Exact Remainder
If the exact division algorithm is done with a full subtraction at each stage and the dividend
isn’t a multiple of the divisor, then low zero limbs are produced but with a remainder in the
high limbs. For dividend a, divisor d, quotient q, and b = 2
mp bits per limb
, this remainder
r is of the form
a = qd +rb
n
n represents the number of zero limbs produced by the subtractions, that being the number of
limbs produced for q. r will be in the range 0 ≤ r < d and can be viewed as a remainder, but
one shifted up by a factor of b
n
.
Carrying out full subtractions at each stage means the same number of cross products must be
done as a normal division, but there’s still some single limb divisions saved. When d is a single
limb some simplifcations arise, providing good speedups on a number of processors.
mpn_divexact_by3, mpn_modexact_1_odd and the mpn_redc_X functions difer subtly in how
they return r, leading to some negations in the above formula, but all are essentially the same.
Clearly r is zero when a is a multiple of d, and this leads to divisibility or congruence tests which
are potentially more efcient than a normal division.
The factor of b
n
on r can be ignored in a GCD when d is odd, hence the use of mpn_modexact_
1_odd by mpn_gcd_1 and mpz_kronecker_ui etc (see Section 16.3 [Greatest Common Divisor
Algorithms], page 100).
Montgomery’s REDC method for modular multiplications uses operands of the form of xb
−n
and yb
−n
and on calculating (xb
−n
)(yb
−n
) uses the factor of b
n
in the exact remainder to reach a
product in the same form (xy)b
−n
(see Section 16.4.2 [Modular Powering Algorithm], page 103).
Notice that r generally gives no useful information about the ordinary remainder a mod d since
b
n
mod d could be anything. If however b
n
≡ 1 mod d, then r is the negative of the ordinary
remainder. This occurs whenever d is a factor of b
n
−1, as for example with 3 in mpn_divexact_
by3. For a 32 or 64 bit limb other such factors include 5, 17 and 257, but no particular use has
been found for this.
16.2.7 Small Quotient Division
An NM division where the number of quotient limbs Q=N−M is small can be optimized
somewhat.
An ordinary basecase division normalizes the divisor by shifting it to make the high bit set,
shifting the dividend accordingly, and shifting the remainder back down at the end of the
calculation. This is wasteful if only a few quotient limbs are to be formed. Instead a division
100 GNU MP 5.0.1
of just the top 2Q limbs of the dividend by the top Q limbs of the divisor can be used to form
a trial quotient. This requires only those limbs normalized, not the whole of the divisor and
dividend.
A multiply and subtract then applies the trial quotient to the M−Q unused limbs of the divisor
and N−Q dividend limbs (which includes Q limbs remaining from the trial quotient division).
The starting trial quotient can be 1 or 2 too big, but all cases of 2 too big and most cases of
1 too big are detected by frst comparing the most signifcant limbs that will arise from the
subtraction. An addback is done if the quotient still turns out to be 1 too big.
This whole procedure is essentially the same as one step of the basecase algorithm done in a Q
limb base, though with the trial quotient test done only with the high limbs, not an entire Q
limb “digit” product. The correctness of this weaker test can be established by following the
argument of Knuth section 4.3.1 exercise 20 but with the v
2
^q > b^r +u
2
condition appropriately
relaxed.
16.3 Greatest Common Divisor
16.3.1 Binary GCD
At small sizes GMP uses an O(N
2
) binary style GCD. This is described in many textbooks,
for example Knuth section 4.5.2 algorithm B. It simply consists of successively reducing odd
operands a and b using
a, b = abs (a −b), min (a, b)
strip factors of 2 from a
The Euclidean GCD algorithm, as per Knuth algorithms E and A, repeatedly computes the
quotient q = a/b and replaces a, b by v, u − qv. The binary algorithm has so far been found
to be faster than the Euclidean algorithm everywhere. One reason the binary method does well
is that the implied quotient at each step is usually small, so often only one or two subtractions
are needed to get the same efect as a division. Quotients 1, 2 and 3 for example occur 67.7%
of the time, see Knuth section 4.5.3 Theorem E.
When the implied quotient is large, meaning b is much smaller than a, then a division is worth
while. This is the basis for the initial a mod b reductions in mpn_gcd and mpn_gcd_1 (the latter
for both N1 and 11 cases). But after that initial reduction, big quotients occur too rarely to
make it worth checking for them.
The fnal 1 1 GCD in mpn_gcd_1 is done in the generic C code as described above. For two
Nbit operands, the algorithm takes about 0.68 iterations per bit. For optimum performance
some attention needs to be paid to the way the factors of 2 are stripped from a.
Firstly it may be noted that in twos complement the number of low zero bits on a − b is the
same as b − a, so counting or testing can begin on a − b without waiting for abs (a − b) to be
determined.
A loop stripping low zero bits tends not to branch predict well, since the condition is data
dependent. But on average there’s only a few low zeros, so an option is to strip one or two bits
arithmetically then loop for more (as done for AMD K6). Or use a lookup table to get a count
for several bits then loop for more (as done for AMD K7). An alternative approach is to keep
just one of a or b odd and iterate
a, b = abs (a −b), min (a, b)
a = a/2 if even
b = b/2 if even
Chapter 16: Algorithms 101
This requires about 1.25 iterations per bit, but stripping of a single bit at each step avoids
any branching. Repeating the bit strip reduces to about 0.9 iterations per bit, which may be a
worthwhile tradeof.
Generally with the above approaches a speed of perhaps 6 cycles per bit can be achieved, which
is still not terribly fast with for instance a 64bit GCD taking nearly 400 cycles. It’s this sort
of time which means it’s not usually advantageous to combine a set of divisibility tests into a
GCD.
Currently, the binary algorithm is used for GCD only when N < 3.
16.3.2 Lehmer’s algorithm
Lehmer’s improvement of the Euclidean algorithms is based on the observation that the initial
part of the quotient sequence depends only on the most signifcant parts of the inputs. The
variant of Lehmer’s algorithm used in GMP splits of the most signifcant two limbs, as suggested,
e.g., in “A DoubleDigit LehmerEuclid Algorithm” by Jebelean (see Appendix B [References],
page 122). The quotients of two doublelimb inputs are collected as a 2 by 2 matrix with single
limb elements. This is done by the function mpn_hgcd2. The resulting matrix is applied to the
inputs using mpn_mul_1 and mpn_submul_1. Each iteration usually reduces the inputs by almost
one limb. In the rare case of a large quotient, no progress can be made by examining just the
most signifcant two limbs, and the quotient is computing using plain division.
The resulting algorithm is asymptotically O(N
2
), just as the Euclidean algorithm and the binary
algorithm. The quadratic part of the work are the calls to mpn_mul_1 and mpn_submul_1. For
small sizes, the linear work is also signifcant. There are roughly N calls to the mpn_hgcd2
function. This function uses a couple of important optimizations:
• It uses the same relaxed notion of correctness as mpn_hgcd (see next section). This means
that when called with the most signifcant two limbs of two large numbers, the returned
matrix does not always correspond exactly to the initial quotient sequence for the two large
numbers; the fnal quotient may sometimes be one of.
• It takes advantage of the fact the quotients are usually small. The division operator is
not used, since the corresponding assembler instruction is very slow on most architectures.
(This code could probably be improved further, it uses many branches that are unfriendly
to prediction).
• It switches from doublelimb calculations to singlelimb calculations halfway through, when
the input numbers have been reduced in size from two limbs to one and a half.
16.3.3 Subquadratic GCD
For inputs larger than GCD_DC_THRESHOLD, GCD is computed via the HGCD (Half GCD) func
tion, as a generalization to Lehmer’s algorithm.
Let the inputs a, b be of size N limbs each. Put S = N/2 + 1. Then HGCD(a,b) returns a
transformation matrix T with nonnegative elements, and reduced numbers (c; d) = T
−1
(a; b).
The reduced numbers c, d must be larger than S limbs, while their diference abs(c −d) must ft
in S limbs. The matrix elements will also be of size roughly N/2.
The HGCD base case uses Lehmer’s algorithm, but with the above stop condition that returns
reduced numbers and the corresponding transformation matrix halfway through. For inputs
larger than HGCD_THRESHOLD, HGCD is computed recursively, using the divide and conquer
algorithm in “On Sch¨onhage’s algorithm and subquadratic integer GCD computation” by M¨oller
(see Appendix B [References], page 122). The recursive algorithm consists of these main steps.
• Call HGCD recursively, on the most signifcant N/2 limbs. Apply the resulting matrix T
1
to the full numbers, reducing them to a size just above 3N/2.
102 GNU MP 5.0.1
• Perform a small number of division or subtraction steps to reduce the numbers to size below
3N/2. This is essential mainly for the unlikely case of large quotients.
• Call HGCD recursively, on the most signifcant N/2 limbs of the reduced numbers. Apply
the resulting matrix T
2
to the full numbers, reducing them to a size just above N/2.
• Compute T = T
1
T
2
.
• Perform a small number of division and subtraction steps to satisfy the requirements, and
return.
GCD is then implemented as a loop around HGCD, similarly to Lehmer’s algorithm. Where
Lehmer repeatedly chops of the top two limbs, calls mpn_hgcd2, and applies the resulting matrix
to the full numbers, the subquadratic GCD chops of the most signifcant third of the limbs (the
proportion is a tuning parameter, and 1/3 seems to be more efcient than, e.g, 1/2), calls
mpn_hgcd, and applies the resulting matrix. Once the input numbers are reduced to size below
GCD_DC_THRESHOLD, Lehmer’s algorithm is used for the rest of the work.
The asymptotic running time of both HGCD and GCD is O(M(N) log N), where M(N) is the
time for multiplying two Nlimb numbers.
16.3.4 Extended GCD
The extended GCD function, or GCDEXT, calculates gcd (a, b) and also cofactors x and y
satisfying ax + by = gcd(a,b). All the algorithms used for plain GCD are extended to handle
this case. The binary algorithm is used only for singlelimb GCDEXT. Lehmer’s algorithm is
used for sizes up to GCDEXT_DC_THRESHOLD. Above this threshold, GCDEXT is implemented as
a loop around HGCD, but with more bookkeeping to keep track of the cofactors. This gives
the same asymptotic running time as for GCD and HGCD, O(M(N) log N)
One diference to plain GCD is that while the inputs a and b are reduced as the algorithm
proceeds, the cofactors x and y grow in size. This makes the tuning of the choppingpoint more
difcult. The current code chops of the most signifcant half of the inputs for the call to HGCD
in the frst iteration, and the most signifcant two thirds for the remaining calls. This strategy
could surely be improved. Also the stop condition for the loop, where Lehmer’s algorithm is
invoked once the inputs are reduced below GCDEXT_DC_THRESHOLD, could maybe be improved
by taking into account the current size of the cofactors.
16.3.5 Jacobi Symbol
mpz_jacobi and mpz_kronecker are currently implemented with a simple binary algorithm
similar to that described for the GCDs (see Section 16.3.1 [Binary GCD], page 100). They’re
not very fast when both inputs are large. Lehmer’s multistep improvement or a binary based
multistep algorithm is likely to be better.
When one operand fts a single limb, and that includes mpz_kronecker_ui and friends, an
initial reduction is done with either mpn_mod_1 or mpn_modexact_1_odd, followed by the binary
algorithm on a single limb. The binary algorithm is well suited to a single limb, and the whole
calculation in this case is quite efcient.
In all the routines sign changes for the result are accumulated using some bit twiddling, avoiding
table lookups or conditional jumps.
16.4 Powering Algorithms
16.4.1 Normal Powering
Normal mpz or mpf powering uses a simple binary algorithm, successively squaring and then
multiplying by the base when a 1 bit is seen in the exponent, as per Knuth section 4.6.3. The
Chapter 16: Algorithms 103
“left to right” variant described there is used rather than algorithm A, since it’s just as easy
and can be done with somewhat less temporary memory.
16.4.2 Modular Powering
Modular powering is implemented using a 2
k
ary sliding window algorithm, as per “Handbook
of Applied Cryptography” algorithm 14.85 (see Appendix B [References], page 122). k is chosen
according to the size of the exponent. Larger exponents use larger values of k, the choice being
made to minimize the average number of multiplications that must supplement the squaring.
The modular multiplies and squares use either a simple division or the REDC method by Mont
gomery (see Appendix B [References], page 122). REDC is a little faster, essentially saving
N single limb divisions in a fashion similar to an exact remainder (see Section 16.2.6 [Exact
Remainder], page 99).
16.5 Root Extraction Algorithms
16.5.1 Square Root
Square roots are taken using the “Karatsuba Square Root” algorithm by Paul Zimmermann (see
Appendix B [References], page 122).
An input n is split into four parts of k bits each, so with b = 2
k
we have n = a
3
b
3
+a
2
b
2
+a
1
b+a
0
.
Part a
3
must be “normalized” so that either the high or second highest bit is set. In GMP, k is
kept on a limb boundary and the input is left shifted (by an even number of bits) to normalize.
The square root of the high two parts is taken, by recursive application of the algorithm (bot
toming out in a onelimb Newton’s method),
s
, r
= sqrtrem (a
3
b +a
2
)
This is an approximation to the desired root and is extended by a division to give s,r,
q, u = divrem (r
b +a
1
, 2s
)
s = s
b +q
r = ub +a
0
−q
2
The normalization requirement on a
3
means at this point s is either correct or 1 too big. r is
negative in the latter case, so
if r < 0 then
r ←r + 2s −1
s ←s −1
The algorithm is expressed in a divide and conquer form, but as noted in the paper it can also
be viewed as a discrete variant of Newton’s method, or as a variation on the schoolboy method
(no longer taught) for square roots two digits at a time.
If the remainder r is not required then usually only a few high limbs of r and u need to be cal
culated to determine whether an adjustment to s is required. This optimization is not currently
implemented.
In the Karatsuba multiplication range this algorithm is O(
3
2
M(N/2)), where M(n) is the time
to multiply two numbers of n limbs. In the FFT multiplication range this grows to a bound of
O(6M(N/2)). In practice a factor of about 1.5 to 1.8 is found in the Karatsuba and Toom3
ranges, growing to 2 or 3 in the FFT range.
The algorithm does all its calculations in integers and the resulting mpn_sqrtrem is used for
both mpz_sqrt and mpf_sqrt. The extended precision given by mpf_sqrt_ui is obtained by
padding with zero limbs.
104 GNU MP 5.0.1
16.5.2 Nth Root
Integer Nth roots are taken using Newton’s method with the following iteration, where A is the
input and n is the root to be taken.
a
i+1
=
1
n
A
a
n−1
i
+ (n −1)a
i
The initial approximation a
1
is generated bitwise by successively powering a trial root with or
without new 1 bits, aiming to be just above the true root. The iteration converges quadratically
when started from a good approximation. When n is large more initial bits are needed to get
good convergence. The current implementation is not particularly well optimized.
16.5.3 Perfect Square
A signifcant fraction of nonsquares can be quickly identifed by checking whether the input is
a quadratic residue modulo small integers.
mpz_perfect_square_p frst tests the input mod 256, which means just examining the low
byte. Only 44 diferent values occur for squares mod 256, so 82.8% of inputs can be immediately
identifed as nonsquares.
On a 32bit system similar tests are done mod 9, 5, 7, 13 and 17, for a total 99.25% of inputs
identifed as nonsquares. On a 64bit system 97 is tested too, for a total 99.62%.
These moduli are chosen because they’re factors of 2
24
− 1 (or 2
48
− 1 for 64bits), and such a
remainder can be quickly taken just using additions (see mpn_mod_34lsub1).
When nails are in use moduli are instead selected by the ‘genpsqr.c’ program and applied
with an mpn_mod_1. The same 2
24
−1 or 2
48
−1 could be done with nails using some extra bit
shifts, but this is not currently implemented.
In any case each modulus is applied to the mpn_mod_34lsub1 or mpn_mod_1 remainder and
a table lookup identifes nonsquares. By using a “modexact” style calculation, and suitably
permuted tables, just one multiply each is required, see the code for details. Moduli are also
combined to save operations, so long as the lookup tables don’t become too big. ‘genpsqr.c’
does all the precalculations.
A square root must still be taken for any value that passes these tests, to verify it’s really a
square and not one of the small fraction of nonsquares that get through (ie. a pseudosquare to
all the tested bases).
Clearly more residue tests could be done, mpz_perfect_square_p only uses a compact and
efcient set. Big inputs would probably beneft from more residue testing, small inputs might
be better of with less. The assumed distribution of squares versus nonsquares in the input
would afect such considerations.
16.5.4 Perfect Power
Detecting perfect powers is required by some factorization algorithms. Currently mpz_perfect_
power_p is implemented using repeated Nth root extractions, though naturally only prime roots
need to be considered. (See Section 16.5.2 [Nth Root Algorithm], page 104.)
If a prime divisor p with multiplicity e can be found, then only roots which are divisors of e
need to be considered, much reducing the work necessary. To this end divisibility by a set of
small primes is checked.
Chapter 16: Algorithms 105
16.6 Radix Conversion
Radix conversions are less important than other algorithms. A program dominated by conver
sions should probably use a diferent data representation.
16.6.1 Binary to Radix
Conversions from binary to a powerof2 radix use a simple and fast O(N) bit extraction algo
rithm.
Conversions from binary to other radices use one of two algorithms. Sizes below GET_STR_
PRECOMPUTE_THRESHOLD use a basic O(N
2
) method. Repeated divisions by b
n
are made, where
b is the radix and n is the biggest power that fts in a limb. But instead of simply using the
remainder r from such divisions, an extra divide step is done to give a fractional limb representing
r/b
n
. The digits of r can then be extracted using multiplications by b rather than divisions.
Special case code is provided for decimal, allowing multiplications by 10 to optimize to shifts
and adds.
Above GET_STR_PRECOMPUTE_THRESHOLD a subquadratic algorithm is used. For an input t,
powers b
n2
i
of the radix are calculated, until a power between t and
√
t is reached. t is then
divided by that largest power, giving a quotient which is the digits above that power, and a
remainder which is those below. These two parts are in turn divided by the second highest power,
and so on recursively. When a piece has been divided down to less than GET_STR_DC_THRESHOLD
limbs, the basecase algorithm described above is used.
The advantage of this algorithm is that big divisions can make use of the subquadratic divide and
conquer division (see Section 16.2.3 [Divide and Conquer Division], page 97), and big divisions
tend to have less overheads than lots of separate single limb divisions anyway. But in any case
the cost of calculating the powers b
n2
i
must frst be overcome.
GET_STR_PRECOMPUTE_THRESHOLD and GET_STR_DC_THRESHOLD represent the same basic thing,
the point where it becomes worth doing a big division to cut the input in half. GET_STR_
PRECOMPUTE_THRESHOLD includes the cost of calculating the radix power required, whereas GET_
STR_DC_THRESHOLD assumes that’s already available, which is the case when recursing.
Since the base case produces digits from least to most signifcant but they want to be stored
from most to least, it’s necessary to calculate in advance how many digits there will be, or at
least be sure not to underestimate that. For GMP the number of input bits is multiplied by
chars_per_bit_exactly from mp_bases, rounding up. The result is either correct or one too
big.
Examining some of the high bits of the input could increase the chance of getting the exact
number of digits, but an exact result every time would not be practical, since in general the
diference between numbers 100. . . and 99. . . is only in the last few bits and the work to identify
99. . . might well be almost as much as a full conversion.
mpf_get_str doesn’t currently use the algorithm described here, it multiplies or divides by a
power of b to move the radix point to the just above the highest nonzero digit (or at worst one
above that location), then multiplies by b
n
to bring out digits. This is O(N
2
) and is certainly
not optimal.
The r/b
n
scheme described above for using multiplications to bring out digits might be useful
for more than a single limb. Some brief experiments with it on the base case when recursing
didn’t give a noticeable improvement, but perhaps that was only due to the implementation.
Something similar would work for the subquadratic divisions too, though there would be the
cost of calculating a bigger radix power.
106 GNU MP 5.0.1
Another possible improvement for the subquadratic part would be to arrange for radix powers
that balanced the sizes of quotient and remainder produced, ie. the highest power would be an
b
nk
approximately equal to
√
t, not restricted to a 2
i
factor. That ought to smooth out a graph
of times against sizes, but may or may not be a net speedup.
16.6.2 Radix to Binary
This section needs to be rewritten, it currently describes the algorithms used before GMP 4.3.
Conversions from a powerof2 radix into binary use a simple and fast O(N) bitwise concatena
tion algorithm.
Conversions from other radices use one of two algorithms. Sizes below SET_STR_PRECOMPUTE_
THRESHOLD use a basic O(N
2
) method. Groups of n digits are converted to limbs, where n is
the biggest power of the base b which will ft in a limb, then those groups are accumulated into
the result by multiplying by b
n
and adding. This saves multiprecision operations, as per Knuth
section 4.4 part E (see Appendix B [References], page 122). Some special case code is provided
for decimal, giving the compiler a chance to optimize multiplications by 10.
Above SET_STR_PRECOMPUTE_THRESHOLD a subquadratic algorithm is used. First groups of n
digits are converted into limbs. Then adjacent limbs are combined into limb pairs with xb
n
+y,
where x and y are the limbs. Adjacent limb pairs are combined into quads similarly with xb
2n
+y.
This continues until a single block remains, that being the result.
The advantage of this method is that the multiplications for each x are big blocks, allowing
Karatsuba and higher algorithms to be used. But the cost of calculating the powers b
n2
i
must
be overcome. SET_STR_PRECOMPUTE_THRESHOLD usually ends up quite big, around 5000 digits,
and on some processors much bigger still.
SET_STR_PRECOMPUTE_THRESHOLD is based on the input digits (and tuned for decimal), though
it might be better based on a limb count, so as to be independent of the base. But that sort of
count isn’t used by the base case and so would need some sort of initial calculation or estimate.
The main reason SET_STR_PRECOMPUTE_THRESHOLD is so much bigger than the corresponding
GET_STR_PRECOMPUTE_THRESHOLD is that mpn_mul_1 is much faster than mpn_divrem_1 (often
by a factor of 5, or more).
16.7 Other Algorithms
16.7.1 Prime Testing
The primality testing in mpz_probab_prime_p (see Section 5.9 [Number Theoretic Functions],
page 35) frst does some trial division by small factors and then uses the MillerRabin probabilis
tic primality testing algorithm, as described in Knuth section 4.5.4 algorithm P (see Appendix B
[References], page 122).
For an odd input n, and with n = q2
k
+ 1 where q is odd, this algorithm selects a random base
x and tests whether x
q
mod n is 1 or −1, or an x
q2
j
mod n is 1, for 1 ≤ j ≤ k. If so then n is
probably prime, if not then n is defnitely composite.
Any prime n will pass the test, but some composites do too. Such composites are known as
strong pseudoprimes to base x. No n is a strong pseudoprime to more than 1/4 of all bases
(see Knuth exercise 22), hence with x chosen at random there’s no more than a 1/4 chance a
“probable prime” will in fact be composite.
In fact strong pseudoprimes are quite rare, making the test much more powerful than this
analysis would suggest, but 1/4 is all that’s proven for an arbitrary n.
Chapter 16: Algorithms 107
16.7.2 Factorial
Factorials are calculated by a combination of removal of twos, powering, and binary splitting.
The procedure can be best illustrated with an example,
23! = 1.2.3.4.5.6.7.8.9.10.11.12.13.14.15.16.17.18.19.20.21.22.23
has factors of two removed,
23! = 2
19
.1.1.3.1.5.3.7.1.9.5.11.3.13.7.15.1.17.9.19.5.21.11.23
and the resulting terms collected up according to their multiplicity,
23! = 2
19
.(3.5)
3
.(7.9.11)
2
.(13.15.17.19.21.23)
Each sequence such as 13.15.17.19.21.23 is evaluated by splitting into every second term, as
for instance (13.17.21).(15.19.23), and the same recursively on each half. This is implemented
iteratively using some bit twiddling.
Such splitting is more efcient than repeated N1 multiplies since it forms big multiplies, al
lowing Karatsuba and higher algorithms to be used. And even below the Karatsuba threshold
a big block of work can be more efcient for the basecase algorithm.
Splitting into subsequences of every second term keeps the resulting products more nearly equal
in size than would the simpler approach of say taking the frst half and second half of the
sequence. Nearly equal products are more efcient for the current multiply implementation.
16.7.3 Binomial Coefcients
Binomial coefcients
n
k
are calculated by frst arranging k ≤ n/2 using
n
k
=
n
n−k
if neces
sary, and then evaluating the following product simply from i = 2 to i = k.
n
k
= (n −k + 1)
k
i=2
n −k +i
i
It’s easy to show that each denominator i will divide the product so far, so the exact division
algorithm is used (see Section 16.2.5 [Exact Division], page 98).
The numerators n −k +i and denominators i are frst accumulated into as many ft a limb, to
save multiprecision operations, though for mpz_bin_ui this applies only to the divisors, since
n is an mpz_t and n −k +i in general won’t ft in a limb at all.
16.7.4 Fibonacci Numbers
The Fibonacci functions mpz_fib_ui and mpz_fib2_ui are designed for calculating isolated F
n
or F
n
,F
n−1
values efciently.
For small n, a table of single limb values in __gmp_fib_table is used. On a 32bit limb this
goes up to F
47
, or on a 64bit limb up to F
93
. For convenience the table starts at F
−1
.
Beyond the table, values are generated with a binary powering algorithm, calculating a pair F
n
and F
n−1
working from high to low across the bits of n. The formulas used are
F
2k+1
= 4F
2
k
−F
2
k−1
+ 2(−1)
k
F
2k−1
= F
2
k
+F
2
k−1
F
2k
= F
2k+1
−F
2k−1
At each step, k is the high b bits of n. If the next bit of n is 0 then F
2k
,F
2k−1
is used, or if it’s
a 1 then F
2k+1
,F
2k
is used, and the process repeated until all bits of n are incorporated. Notice
these formulas require just two squares per bit of n.
108 GNU MP 5.0.1
It’d be possible to handle the frst few n above the single limb table with simple additions, using
the defning Fibonacci recurrence F
k+1
= F
k
+ F
k−1
, but this is not done since it usually turns
out to be faster for only about 10 or 20 values of n, and including a block of code for just those
doesn’t seem worthwhile. If they really mattered it’d be better to extend the data table.
Using a table avoids lots of calculations on small numbers, and makes small n go fast. A bigger
table would make more small n go fast, it’s just a question of balancing size against desired
speed. For GMP the code is kept compact, with the emphasis primarily on a good powering
algorithm.
mpz_fib2_ui returns both F
n
and F
n−1
, but mpz_fib_ui is only interested in F
n
. In this case
the last step of the algorithm can become one multiply instead of two squares. One of the
following two formulas is used, according as n is odd or even.
F
2k
= F
k
(F
k
+ 2F
k−1
)
F
2k+1
= (2F
k
+F
k−1
)(2F
k
−F
k−1
) + 2(−1)
k
F
2k+1
here is the same as above, just rearranged to be a multiply. For interest, the 2(−1)
k
term
both here and above can be applied just to the low limb of the calculation, without a carry or
borrow into further limbs, which saves some code size. See comments with mpz_fib_ui and the
internal mpn_fib2_ui for how this is done.
16.7.5 Lucas Numbers
mpz_lucnum2_ui derives a pair of Lucas numbers from a pair of Fibonacci numbers with the
following simple formulas.
L
k
= F
k
+ 2F
k−1
L
k−1
= 2F
k
−F
k−1
mpz_lucnum_ui is only interested in L
n
, and some work can be saved. Trailing zero bits on n
can be handled with a single square each.
L
2k
= L
2
k
−2(−1)
k
And the lowest 1 bit can be handled with one multiply of a pair of Fibonacci numbers, similar
to what mpz_fib_ui does.
L
2k+1
= 5F
k−1
(2F
k
+F
k−1
) −4(−1)
k
16.7.6 Random Numbers
For the urandomb functions, random numbers are generated simply by concatenating bits pro
duced by the generator. As long as the generator has good randomness properties this will
produce welldistributed N bit numbers.
For the urandomm functions, random numbers in a range 0 ≤ R < N are generated by taking
values R of log
2
N bits each until one satisfes R < N. This will normally require only one
or two attempts, but the attempts are limited in case the generator is somehow degenerate and
produces only 1 bits or similar.
The Mersenne Twister generator is by Matsumoto and Nishimura (see Appendix B [References],
page 122). It has a nonrepeating period of 2
19937
−1, which is a Mersenne prime, hence the name
of the generator. The state is 624 words of 32bits each, which is iterated with one XOR and
shift for each 32bit word generated, making the algorithm very fast. Randomness properties
are also very good and this is the default algorithm used by GMP.
Linear congruential generators are described in many text books, for instance Knuth volume
2 (see Appendix B [References], page 122). With a modulus M and parameters A and C, a
Chapter 16: Algorithms 109
integer state S is iterated by the formula S ←AS +C mod M. At each step the new state is a
linear function of the previous, mod M, hence the name of the generator.
In GMP only moduli of the form 2
N
are supported, and the current implementation is not as well
optimized as it could be. Overheads are signifcant when N is small, and when N is large clearly
the multiply at each step will become slow. This is not a big concern, since the Mersenne Twister
generator is better in every respect and is therefore recommended for all normal applications.
For both generators the current state can be deduced by observing enough output and applying
some linear algebra (over GF(2) in the case of the Mersenne Twister). This generally means
raw output is unsuitable for cryptographic applications without further hashing or the like.
16.8 Assembly Coding
The assembly subroutines in GMP are the most signifcant source of speed at small to moderate
sizes. At larger sizes algorithm selection becomes more important, but of course speedups in
low level routines will still speed up everything proportionally.
Carry handling and widening multiplies that are important for GMP can’t be easily expressed
in C. GCC asm blocks help a lot and are provided in ‘longlong.h’, but hand coding low level
routines invariably ofers a speedup over generic C by a factor of anything from 2 to 10.
16.8.1 Code Organisation
The various ‘mpn’ subdirectories contain machinedependent code, written in C or assembly.
The ‘mpn/generic’ subdirectory contains default code, used when there’s no machinespecifc
version of a particular fle.
Each ‘mpn’ subdirectory is for an ISA family. Generally 32bit and 64bit variants in a family
cannot share code and have separate directories. Within a family further subdirectories may
exist for CPU variants.
In each directory a ‘nails’ subdirectory may exist, holding code with nails support for that CPU
variant. A NAILS_SUPPORT directive in each fle indicates the nails values the code handles. Nails
code only exists where it’s faster, or promises to be faster, than plain code. There’s no efort
put into nails if they’re not going to enhance a given CPU.
16.8.2 Assembly Basics
mpn_addmul_1 and mpn_submul_1 are the most important routines for overall GMP performance.
All multiplications and divisions come down to repeated calls to these. mpn_add_n, mpn_sub_n,
mpn_lshift and mpn_rshift are next most important.
On some CPUs assembly versions of the internal functions mpn_mul_basecase and mpn_sqr_
basecase give signifcant speedups, mainly through avoiding function call overheads. They can
also potentially make better use of a wide superscalar processor, as can bigger primitives like
mpn_addmul_2 or mpn_addmul_4.
The restrictions on overlaps between sources and destinations (see Chapter 8 [Lowlevel Func
tions], page 56) are designed to facilitate a variety of implementations. For example, knowing
mpn_add_n won’t have partly overlapping sources and destination means reading can be done far
ahead of writing on superscalar processors, and loops can be vectorized on a vector processor,
depending on the carry handling.
110 GNU MP 5.0.1
16.8.3 Carry Propagation
The problem that presents most challenges in GMP is propagating carries from one limb to the
next. In functions like mpn_addmul_1 and mpn_add_n, carries are the only dependencies between
limb operations.
On processors with carry fags, a straightforward CISC style adc is generally best. AMD K6
mpn_addmul_1 however is an example of an unusual set of circumstances where a branch works
out better.
On RISC processors generally an add and compare for overfow is used. This sort of thing can
be seen in ‘mpn/generic/aors_n.c’. Some carry propagation schemes require 4 instructions,
meaning at least 4 cycles per limb, but other schemes may use just 1 or 2. On wide superscalar
processors performance may be completely determined by the number of dependent instructions
between carryin and carryout for each limb.
On vector processors good use can be made of the fact that a carry bit only very rarely propagates
more than one limb. When adding a single bit to a limb, there’s only a carry out if that limb was
0xFF...FF which on random data will be only 1 in 2
mp bits per limb
. ‘mpn/cray/add_n.c’
is an example of this, it adds all limbs in parallel, adds one set of carry bits in parallel and then
only rarely needs to fall through to a loop propagating further carries.
On the x86s, GCC (as of version 2.95.2) doesn’t generate particularly good code for the RISC
style idioms that are necessary to handle carry bits in C. Often conditional jumps are generated
where adc or sbb forms would be better. And so unfortunately almost any loop involving carry
bits needs to be coded in assembly for best results.
16.8.4 Cache Handling
GMP aims to perform well both on operands that ft entirely in L1 cache and those which don’t.
Basic routines like mpn_add_n or mpn_lshift are often used on large operands, so L2 and main
memory performance is important for them. mpn_mul_1 and mpn_addmul_1 are mostly used
for multiply and square basecases, so L1 performance matters most for them, unless assembly
versions of mpn_mul_basecase and mpn_sqr_basecase exist, in which case the remaining uses
are mostly for larger operands.
For L2 or main memory operands, memory access times will almost certainly be more than
the calculation time. The aim therefore is to maximize memory throughput, by starting a load
of the next cache line while processing the contents of the previous one. Clearly this is only
possible if the chip has a lockup free cache or some sort of prefetch instruction. Most current
chips have both these features.
Prefetching sources combines well with loop unrolling, since a prefetch can be initiated once per
unrolled loop (or more than once if the loop covers more than one cache line).
On CPUs without writeallocate caches, prefetching destinations will ensure individual stores
don’t go further down the cache hierarchy, limiting bandwidth. Of course for calculations which
are slow anyway, like mpn_divrem_1, writethroughs might be fne.
The distance ahead to prefetch will be determined by memory latency versus throughput. The
aim of course is to have data arriving continuously, at peak throughput. Some CPUs have limits
on the number of fetches or prefetches in progress.
If a special prefetch instruction doesn’t exist then a plain load can be used, but in that case care
must be taken not to attempt to read past the end of an operand, since that might produce a
segmentation violation.
Chapter 16: Algorithms 111
Some CPUs or systems have hardware that detects sequential memory accesses and initiates
suitable cache movements automatically, making life easy.
16.8.5 Functional Units
When choosing an approach for an assembly loop, consideration is given to what operations
can execute simultaneously and what throughput can thereby be achieved. In some cases an
algorithm can be tweaked to accommodate available resources.
Loop control will generally require a counter and pointer updates, costing as much as 5 in
structions, plus any delays a branch introduces. CPU addressing modes might reduce pointer
updates, perhaps by allowing just one updating pointer and others expressed as ofsets from it,
or on CISC chips with all addressing done with the loop counter as a scaled index.
The fnal loop control cost can be amortised by processing several limbs in each iteration (see
Section 16.8.9 [Assembly Loop Unrolling], page 113). This at least ensures loop control isn’t a
big fraction the work done.
Memory throughput is always a limit. If perhaps only one load or one store can be done per
cycle then 3 cycles/limb will the top speed for “binary” operations like mpn_add_n, and any
code achieving that is optimal.
Integer resources can be freed up by having the loop counter in a foat register, or by pressing
the foat units into use for some multiplying, perhaps doing every second limb on the foat side
(see Section 16.8.6 [Assembly Floating Point], page 111).
Float resources can be freed up by doing carry propagation on the integer side, or even by doing
integer to foat conversions in integers using bit twiddling.
16.8.6 Floating Point
Floating point arithmetic is used in GMP for multiplications on CPUs with poor integer multi
pliers. It’s mostly useful for mpn_mul_1, mpn_addmul_1 and mpn_submul_1 on 64bit machines,
and mpn_mul_basecase on both 32bit and 64bit machines.
With IEEE 53bit double precision foats, integer multiplications producing up to 53 bits will give
exact results. Breaking a 6464 multiplication into eight 1632 → 48 bit pieces is convenient.
With some care though six 2132 →53 bit products can be used, if one of the lower two 21bit
pieces also uses the sign bit.
For the mpn_mul_1 family of functions on a 64bit machine, the invariant single limb is split
at the start, into 3 or 4 pieces. Inside the loop, the bignum operand is split into 32bit pieces.
Fast conversion of these unsigned 32bit pieces to foating point is highly machinedependent. In
some cases, reading the data into the integer unit, zeroextending to 64bits, then transferring
to the foating point unit back via memory is the only option.
Converting partial products back to 64bit limbs is usually best done as a signed conversion.
Since all values are smaller than 2
53
, signed and unsigned are the same, but most processors
lack unsigned conversions.
Here is a diagram showing 1632 bit products for an mpn_mul_1 or mpn_addmul_1 with a 64bit
limb. The single limb operand V is split into four 16bit parts. The multilimb operand U is
split in the loop into two 32bit parts.
112 GNU MP 5.0.1
v48 v32 v16 v00 V Operand
u32 u00 U Operand (one limb)
u00 v00 p00 48bit products
u00 v16 p16
u00 v32 p32
u00 v48 p48
u32 v00 r32
u32 v16 r48
u32 v32 r64
u32 v48 r80
p32 and r32 can be summed using foatingpoint addition, and likewise p48 and r48. p00 and
p16 can be summed with r64 and r80 from the previous iteration.
For each loop then, four 49bit quantities are transferred to the integer unit, aligned as follows,
64 bits 64 bits
p00 +r64
i00
p16 +r80
i16
p32 +r32 i32
p48 +r48 i48
The challenge then is to sum these efciently and add in a carry limb, generating a low 64bit
result limb and a high 33bit carry limb (i48 extends 33 bits into the high half).
16.8.7 SIMD Instructions
The singleinstruction multipledata support in current microprocessors is aimed at signal pro
cessing algorithms where each data point can be treated more or less independently. There’s
generally not much support for propagating the sort of carries that arise in GMP.
SIMD multiplications of say four 1616 bit multiplies only do as much work as one 3232 from
GMP’s point of view, and need some shifts and adds besides. But of course if say the SIMD
form is fully pipelined and uses less instruction decoding then it may still be worthwhile.
On the x86 chips, MMX has so far found a use in mpn_rshift and mpn_lshift, and is used in a
special case for 16bit multipliers in the P55 mpn_mul_1. SSE2 is used for Pentium 4 mpn_mul_1,
mpn_addmul_1, and mpn_submul_1.
16.8.8 Software Pipelining
Software pipelining consists of scheduling instructions around the branch point in a loop. For
example a loop might issue a load not for use in the present iteration but the next, thereby
allowing extra cycles for the data to arrive from memory.
Naturally this is wanted only when doing things like loads or multiplies that take several cycles
to complete, and only where a CPU has multiple functional units so that other work can be
done in the meantime.
A pipeline with several stages will have a data value in progress at each stage and each loop
iteration moves them along one stage. This is like juggling.
Chapter 16: Algorithms 113
If the latency of some instruction is greater than the loop time then it will be necessary to unroll,
so one register has a result ready to use while another (or multiple others) are still in progress.
(see Section 16.8.9 [Assembly Loop Unrolling], page 113).
16.8.9 Loop Unrolling
Loop unrolling consists of replicating code so that several limbs are processed in each loop.
At a minimum this reduces loop overheads by a corresponding factor, but it can also allow
better register usage, for example alternately using one register combination and then another.
Judicious use of m4 macros can help avoid lots of duplication in the source code.
Any amount of unrolling can be handled with a loop counter that’s decremented by N each
time, stopping when the remaining count is less than the further N the loop will process. Or by
subtracting N at the start, the termination condition becomes when the counter C is less than
0 (and the count of remaining limbs is C +N).
Alternately for a power of 2 unroll the loop count and remainder can be established with a shift
and mask. This is convenient if also making a computed jump into the middle of a large loop.
The limbs not a multiple of the unrolling can be handled in various ways, for example
• A simple loop at the end (or the start) to process the excess. Care will be wanted that it
isn’t too much slower than the unrolled part.
• A set of binary tests, for example after an 8limb unrolling, test for 4 more limbs to process,
then a further 2 more or not, and fnally 1 more or not. This will probably take more code
space than a simple loop.
• A switch statement, providing separate code for each possible excess, for example an 8limb
unrolling would have separate code for 0 remaining, 1 remaining, etc, up to 7 remaining.
This might take a lot of code, but may be the best way to optimize all cases in combination
with a deep pipelined loop.
• A computed jump into the middle of the loop, thus making the frst iteration handle the
excess. This should make times smoothly increase with size, which is attractive, but setups
for the jump and adjustments for pointers can be tricky and could become quite difcult in
combination with deep pipelining.
16.8.10 Writing Guide
This is a guide to writing software pipelined loops for processing limb vectors in assembly.
First determine the algorithm and which instructions are needed. Code it without unrolling or
scheduling, to make sure it works. On a 3operand CPU try to write each new value to a new
register, this will greatly simplify later steps.
Then note for each instruction the functional unit and/or issue port requirements. If an instruc
tion can use either of two units, like U0 or U1 then make a category “U0/U1”. Count the total
using each unit (or combined unit), and count all instructions.
Figure out from those counts the best possible loop time. The goal will be to fnd a perfect
schedule where instruction latencies are completely hidden. The total instruction count might
be the limiting factor, or perhaps a particular functional unit. It might be possible to tweak the
instructions to help the limiting factor.
Suppose the loop time is N, then make N issue buckets, with the fnal loop branch at the end of
the last. Now fll the buckets with dummy instructions using the functional units desired. Run
this to make sure the intended speed is reached.
114 GNU MP 5.0.1
Now replace the dummy instructions with the real instructions from the slow but correct loop
you started with. The frst will typically be a load instruction. Then the instruction using that
value is placed in a bucket an appropriate distance down. Run the loop again, to check it still
runs at target speed.
Keep placing instructions, frequently measuring the loop. After a few you will need to wrap
around from the last bucket back to the top of the loop. If you used the newregister for new
value strategy above then there will be no register conficts. If not then take care not to clobber
something already in use. Changing registers at this time is very error prone.
The loop will overlap two or more of the original loop iterations, and the computation of one
vector element result will be started in one iteration of the new loop, and completed one or
several iterations later.
The fnal step is to create feedin and winddown code for the loop. A good way to do this is
to make a copy (or copies) of the loop at the start and delete those instructions which don’t
have valid antecedents, and at the end replicate and delete those whose results are unwanted
(including any further loads).
The loop will have a minimum number of limbs loaded and processed, so the feedin code must
test if the request size is smaller and skip either to a suitable part of the winddown or to special
code for small sizes.
Chapter 17: Internals 115
17 Internals
This chapter is provided only for informational purposes and the various internals described
here may change in future GMP releases. Applications expecting to be compatible with future
releases should use only the documented interfaces described in previous chapters.
17.1 Integer Internals
mpz_t variables represent integers using sign and magnitude, in space dynamically allocated and
reallocated. The felds are as follows.
_mp_size The number of limbs, or the negative of that when representing a negative integer.
Zero is represented by _mp_size set to zero, in which case the _mp_d data is unused.
_mp_d A pointer to an array of limbs which is the magnitude. These are stored “little
endian” as per the mpn functions, so _mp_d[0] is the least signifcant limb and _mp_
d[ABS(_mp_size)1] is the most signifcant. Whenever _mp_size is nonzero, the
most signifcant limb is nonzero.
Currently there’s always at least one limb allocated, so for instance mpz_set_ui
never needs to reallocate, and mpz_get_ui can fetch _mp_d[0] unconditionally
(though its value is then only wanted if _mp_size is nonzero).
_mp_alloc
_mp_alloc is the number of limbs currently allocated at _mp_d, and naturally _mp_
alloc >= ABS(_mp_size). When an mpz routine is about to (or might be about to)
increase _mp_size, it checks _mp_alloc to see whether there’s enough space, and
reallocates if not. MPZ_REALLOC is generally used for this.
The various bitwise logical functions like mpz_and behave as if negative values were twos com
plement. But sign and magnitude is always used internally, and necessary adjustments are made
during the calculations. Sometimes this isn’t pretty, but sign and magnitude are best for other
routines.
Some internal temporary variables are setup with MPZ_TMP_INIT and these have _mp_d space
obtained from TMP_ALLOC rather than the memory allocation functions. Care is taken to ensure
that these are big enough that no reallocation is necessary (since it would have unpredictable
consequences).
_mp_size and _mp_alloc are int, although mp_size_t is usually a long. This is done to make
the felds just 32 bits on some 64 bits systems, thereby saving a few bytes of data space but still
providing plenty of range.
17.2 Rational Internals
mpq_t variables represent rationals using an mpz_t numerator and denominator (see Section 17.1
[Integer Internals], page 115).
The canonical form adopted is denominator positive (and nonzero), no common factors between
numerator and denominator, and zero uniquely represented as 0/1.
It’s believed that casting out common factors at each stage of a calculation is best in general. A
GCD is an O(N
2
) operation so it’s better to do a few small ones immediately than to delay and
have to do a big one later. Knowing the numerator and denominator have no common factors
can be used for example in mpq_mul to make only two cross GCDs necessary, not four.
This general approach to common factors is badly suboptimal in the presence of simple factor
izations or little prospect for cancellation, but GMP has no way to know when this will occur.
116 GNU MP 5.0.1
As per Section 3.11 [Efciency], page 21, that’s left to applications. The mpq_t framework might
still suit, with mpq_numref and mpq_denref for direct access to the numerator and denominator,
or of course mpz_t variables can be used directly.
17.3 Float Internals
Efcient calculation is the primary aim of GMP foats and the use of whole limbs and simple
rounding facilitates this.
mpf_t foats have a variable precision mantissa and a single machine word signed exponent. The
mantissa is represented using sign and magnitude.
most signifcant limb least signifcant limb
mp exp → mp d
← radix point
← mp size →
The felds are as follows.
_mp_size The number of limbs currently in use, or the negative of that when representing a
negative value. Zero is represented by _mp_size and _mp_exp both set to zero, and
in that case the _mp_d data is unused. (In the future _mp_exp might be undefned
when representing zero.)
_mp_prec The precision of the mantissa, in limbs. In any calculation the aim is to produce
_mp_prec limbs of result (the most signifcant being nonzero).
_mp_d A pointer to the array of limbs which is the absolute value of the mantissa. These are
stored “little endian” as per the mpn functions, so _mp_d[0] is the least signifcant
limb and _mp_d[ABS(_mp_size)1] the most signifcant.
The most signifcant limb is always nonzero, but there are no other restrictions on
its value, in particular the highest 1 bit can be anywhere within the limb.
_mp_prec+1 limbs are allocated to _mp_d, the extra limb being for convenience (see
below). There are no reallocations during a calculation, only in a change of precision
with mpf_set_prec.
_mp_exp The exponent, in limbs, determining the location of the implied radix point. Zero
means the radix point is just above the most signifcant limb. Positive values mean
a radix point ofset towards the lower limbs and hence a value ≥ 1, as for example
in the diagram above. Negative exponents mean a radix point further above the
highest limb.
Naturally the exponent can be any value, it doesn’t have to fall within the limbs as
the diagram shows, it can be a long way above or a long way below. Limbs other
than those included in the {_mp_d,_mp_size} data are treated as zero.
The _mp_size and _mp_prec felds are int, although the mp_size_t type is usually a long.
The _mp_exp feld is usually long. This is done to make some felds just 32 bits on some 64 bits
systems, thereby saving a few bytes of data space but still providing plenty of precision and a
very large range.
The following various points should be noted.
Low Zeros The least signifcant limbs _mp_d[0] etc can be zero, though such low zeros can
always be ignored. Routines likely to produce low zeros check and avoid them to
Chapter 17: Internals 117
save time in subsequent calculations, but for most routines they’re quite unlikely
and aren’t checked.
Mantissa Size Range
The _mp_size count of limbs in use can be less than _mp_prec if the value can be
represented in less. This means low precision values or small integers stored in a
high precision mpf_t can still be operated on efciently.
_mp_size can also be greater than _mp_prec. Firstly a value is allowed to use all
of the _mp_prec+1 limbs available at _mp_d, and secondly when mpf_set_prec_raw
lowers _mp_prec it leaves _mp_size unchanged and so the size can be arbitrarily
bigger than _mp_prec.
Rounding All rounding is done on limb boundaries. Calculating _mp_prec limbs with the high
nonzero will ensure the application requested minimum precision is obtained.
The use of simple “trunc” rounding towards zero is efcient, since there’s no need
to examine extra limbs and increment or decrement.
Bit Shifts Since the exponent is in limbs, there are no bit shifts in basic operations like mpf_
add and mpf_mul. When difering exponents are encountered all that’s needed is to
adjust pointers to line up the relevant limbs.
Of course mpf_mul_2exp and mpf_div_2exp will require bit shifts, but the choice
is between an exponent in limbs which requires shifts there, or one in bits which
requires them almost everywhere else.
Use of _mp_prec+1 Limbs
The extra limb on _mp_d (_mp_prec+1 rather than just _mp_prec) helps when an
mpf routine might get a carry from its operation. mpf_add for instance will do an
mpn_add of _mp_prec limbs. If there’s no carry then that’s the result, but if there is a
carry then it’s stored in the extra limb of space and _mp_size becomes _mp_prec+1.
Whenever _mp_prec+1 limbs are held in a variable, the low limb is not needed for
the intended precision, only the _mp_prec high limbs. But zeroing it out or moving
the rest down is unnecessary. Subsequent routines reading the value will simply take
the high limbs they need, and this will be _mp_prec if their target has that same
precision. This is no more than a pointer adjustment, and must be checked anyway
since the destination precision can be diferent from the sources.
Copy functions like mpf_set will retain a full _mp_prec+1 limbs if available. This
ensures that a variable which has _mp_size equal to _mp_prec+1 will get its full
exact value copied. Strictly speaking this is unnecessary since only _mp_prec limbs
are needed for the application’s requested precision, but it’s considered that an mpf_
set from one variable into another of the same precision ought to produce an exact
copy.
Application Precisions
__GMPF_BITS_TO_PREC converts an application requested precision to an _mp_prec.
The value in bits is rounded up to a whole limb then an extra limb is added since
the most signifcant limb of _mp_d is only nonzero and therefore might contain only
one bit.
__GMPF_PREC_TO_BITS does the reverse conversion, and removes the extra limb from
_mp_prec before converting to bits. The net efect of reading back with mpf_get_
prec is simply the precision rounded up to a multiple of mp_bits_per_limb.
Note that the extra limb added here for the high only being nonzero is in addition
to the extra limb allocated to _mp_d. For example with a 32bit limb, an application
request for 250 bits will be rounded up to 8 limbs, then an extra added for the high
being only nonzero, giving an _mp_prec of 9. _mp_d then gets 10 limbs allocated.
118 GNU MP 5.0.1
Reading back with mpf_get_prec will take _mp_prec subtract 1 limb and multiply
by 32, giving 256 bits.
Strictly speaking, the fact the high limb has at least one bit means that a foat with,
say, 3 limbs of 32bits each will be holding at least 65 bits, but for the purposes of
mpf_t it’s considered simply to be 64 bits, a nice multiple of the limb size.
17.4 Raw Output Internals
mpz_out_raw uses the following format.
size data bytes
The size is 4 bytes written most signifcant byte frst, being the number of subsequent data
bytes, or the twos complement negative of that when a negative integer is represented. The
data bytes are the absolute value of the integer, written most signifcant byte frst.
The most signifcant data byte is always nonzero, so the output is the same on all systems,
irrespective of limb size.
In GMP 1, leading zero bytes were written to pad the data bytes to a multiple of the limb size.
mpz_inp_raw will still accept this, for compatibility.
The use of “big endian” for both the size and data felds is deliberate, it makes the data easy to
read in a hex dump of a fle. Unfortunately it also means that the limb data must be reversed
when reading or writing, so neither a big endian nor little endian system can just read and write
_mp_d.
17.5 C++ Interface Internals
A system of expression templates is used to ensure something like a=b+c turns into a simple call
to mpz_add etc. For mpf_class the scheme also ensures the precision of the fnal destination
is used for any temporaries within a statement like f=w*x+y*z. These are important features
which a naive implementation cannot provide.
A simplifed description of the scheme follows. The true scheme is complicated by the fact that
expressions have diferent return types. For detailed information, refer to the source code.
To perform an operation, say, addition, we frst defne a “function object” evaluating it,
struct __gmp_binary_plus
{
static void eval(mpf_t f, mpf_t g, mpf_t h) { mpf_add(f, g, h); }
};
And an “additive expression” object,
__gmp_expr<__gmp_binary_expr<mpf_class, mpf_class, __gmp_binary_plus> >
operator+(const mpf_class &f, const mpf_class &g)
{
return __gmp_expr
<__gmp_binary_expr<mpf_class, mpf_class, __gmp_binary_plus> >(f, g);
}
The seemingly redundant __gmp_expr<__gmp_binary_expr<...>> is used to encapsulate any
possible kind of expression into a single template type. In fact even mpf_class etc are typedef
specializations of __gmp_expr.
Chapter 17: Internals 119
Next we defne assignment of __gmp_expr to mpf_class.
template <class T>
mpf_class & mpf_class::operator=(const __gmp_expr<T> &expr)
{
expr.eval(this>get_mpf_t(), this>precision());
return *this;
}
template <class Op>
void __gmp_expr<__gmp_binary_expr<mpf_class, mpf_class, Op> >::eval
(mpf_t f, mp_bitcnt_t precision)
{
Op::eval(f, expr.val1.get_mpf_t(), expr.val2.get_mpf_t());
}
where expr.val1 and expr.val2 are references to the expression’s operands (here expr is the
__gmp_binary_expr stored within the __gmp_expr).
This way, the expression is actually evaluated only at the time of assignment, when the required
precision (that of f) is known. Furthermore the target mpf_t is now available, thus we can call
mpf_add directly with f as the output argument.
Compound expressions are handled by defning operators taking subexpressions as their argu
ments, like this:
template <class T, class U>
__gmp_expr
<__gmp_binary_expr<__gmp_expr<T>, __gmp_expr<U>, __gmp_binary_plus> >
operator+(const __gmp_expr<T> &expr1, const __gmp_expr<U> &expr2)
{
return __gmp_expr
<__gmp_binary_expr<__gmp_expr<T>, __gmp_expr<U>, __gmp_binary_plus> >
(expr1, expr2);
}
And the corresponding specializations of __gmp_expr::eval:
template <class T, class U, class Op>
void __gmp_expr
<__gmp_binary_expr<__gmp_expr<T>, __gmp_expr<U>, Op> >::eval
(mpf_t f, mp_bitcnt_t precision)
{
// declare two temporaries
mpf_class temp1(expr.val1, precision), temp2(expr.val2, precision);
Op::eval(f, temp1.get_mpf_t(), temp2.get_mpf_t());
}
The expression is thus recursively evaluated to any level of complexity and all subexpressions
are evaluated to the precision of f.
120 GNU MP 5.0.1
Appendix A Contributors
Torbj¨orn Granlund wrote the original GMP library and is still the main developer. Code not
explicitly attributed to others, was contributed by Torbj¨orn. Several other individuals and
organizations have contributed GMP. Here is a list in chronological order on frst contribution:
Gunnar Sj¨odin and Hans Riesel helped with mathematical problems in early versions of the
library.
Richard Stallman helped with the interface design and revised the frst version of this manual.
Brian Beuning and Doug Lea helped with testing of early versions of the library and made
creative suggestions.
John Amanatides of York University in Canada contributed the function mpz_probab_prime_p.
Paul Zimmermann wrote the REDCbased mpz powm code, the Sch¨onhageStrassen FFT mul
tiply code, and the Karatsuba square root code. He also improved the Toom3 code for GMP
4.2. Paul sparked the development of GMP 2, with his comparisons between bignum packages.
The ECMNET project Paul is organizing was a driving force behind many of the optimizations
in GMP 3. Paul also wrote the new GMP 4.3 nth root code (with Torbj¨orn).
Ken Weber (Kent State University, Universidade Federal do Rio Grande do Sul) contributed now
defunct versions of mpz_gcd, mpz_divexact, mpn_gcd, and mpn_bdivmod, partially supported by
CNPq (Brazil) grant 3013141942.
Per Bothner of Cygnus Support helped to set up GMP to use Cygnus’ confgure. He has also
made valuable suggestions and tested numerous intermediary releases.
Joachim Hollman was involved in the design of the mpf interface, and in the mpz design revisions
for version 2.
Bennet Yee contributed the initial versions of mpz_jacobi and mpz_legendre.
Andreas Schwab contributed the fles ‘mpn/m68k/lshift.S’ and ‘mpn/m68k/rshift.S’ (now in
‘.asm’ form).
Robert Harley of Inria, France and David Seal of ARM, England, suggested clever improve
ments for population count. Robert also wrote highly optimized Karatsuba and 3way Toom
multiplication functions for GMP 3, and contributed the ARM assembly code.
Torsten Ekedahl of the Mathematical department of Stockholm University provided signifcant
inspiration during several phases of the GMP development. His mathematical expertise helped
improve several algorithms.
Linus Nordberg wrote the new confgure system based on autoconf and implemented the new
random functions.
Kevin Ryde worked on a large number of things: optimized x86 code, m4 asm macros, parameter
tuning, speed measuring, the confgure system, function inlining, divisibility tests, bit scanning,
Jacobi symbols, Fibonacci and Lucas number functions, printf and scanf functions, perl interface,
demo expression parser, the algorithms chapter in the manual, ‘gmpasmmode.el’, and various
miscellaneous improvements elsewhere.
Kent Boortz made the Mac OS 9 port.
Steve Root helped write the optimized alpha 21264 assembly code.
Gerardo Ballabio wrote the ‘gmpxx.h’ C++ class interface and the C++ istream input routines.
Appendix A: Contributors 121
Jason Moxham rewrote mpz_fac_ui.
Pedro Gimeno implemented the Mersenne Twister and made other random number improve
ments.
Niels M¨oller wrote the subquadratic GCD and extended GCD code, the quadratic Hensel divi
sion code, and (with Torbj¨orn) the new divide and conquer division code for GMP 4.3. Niels also
helped implement the new Toom multiply code for GMP 4.3 and implemented helper functions
to simplify Toom evaluations for GMP 5.0. He wrote the original version of mpn mulmod bnm1.
Alberto Zanoni and Marco Bodrato suggested the unbalanced multiply strategy, and found the
optimal strategies for evaluation and interpolation in Toom multiplication.
Marco Bodrato helped implement the new Toom multiply code for GMP 4.3 and implemented
most of the new Toom multiply and squaring code for 5.0. He is the main author of the
current mpn mulmod bnm1 and mpn mullo n. Marco also wrote the functions mpn invert and
mpn invertappr.
David Harvey suggested the internal function mpn_bdiv_dbm1, implementing division relevant
to Toom multiplication. He also worked on fast assembly sequences, in particular on a fast
AMD64 mpn_mul_basecase.
Martin Boij wrote mpn_perfect_power_p.
(This list is chronological, not ordered after signifcance. If you have contributed to GMP but
are not listed above, please tell gmpdevel@gmplib.org about the omission!)
The development of foating point functions of GNU MP 2, were supported in part by the
ESPRITBRA (Basic Research Activities) 6846 project POSSO (POlynomial System SOlving).
The development of GMP 2, 3, and 4 was supported in part by the IDA Center for Computing
Sciences.
Thanks go to Hans Thorsen for donating an SGI system for the GMP test system environment.
122 GNU MP 5.0.1
Appendix B References
B.1 Books
• Jonathan M. Borwein and Peter B. Borwein, “Pi and the AGM: A Study in Analytic
Number Theory and Computational Complexity”, Wiley, 1998.
• Richard Crandall and Carl Pomerance, “Prime Numbers: A Computational Perspective”,
2nd edition, SpringerVerlag, 2005.
http://math.dartmouth.edu/~carlp/
• Henri Cohen, “A Course in Computational Algebraic Number Theory”, Graduate Texts in
Mathematics number 138, SpringerVerlag, 1993.
http://www.math.ubordeaux.fr/~cohen/
• Donald E. Knuth, “The Art of Computer Programming”, volume 2, “Seminumerical Algo
rithms”, 3rd edition, AddisonWesley, 1998.
http://wwwcsfaculty.stanford.edu/~knuth/taocp.html
• John D. Lipson, “Elements of Algebra and Algebraic Computing”, The Benjamin Cummings
Publishing Company Inc, 1981.
• Alfred J. Menezes, Paul C. van Oorschot and Scott A. Vanstone, “Handbook of Applied
Cryptography”, http://www.cacr.math.uwaterloo.ca/hac/
• Richard M. Stallman and the GCC Developer Community, “Using the
GNU Compiler Collection”, Free Software Foundation, 2008, avail
able online http://gcc.gnu.org/onlinedocs/, and in the GCC package
ftp://ftp.gnu.org/gnu/gcc/
B.2 Papers
• Yves Bertot, Nicolas Magaud and Paul Zimmermann, “A Proof of GMP Square Root”,
Journal of Automated Reasoning, volume 29, 2002, pp. 225252. Also available online as
INRIA Research Report 4475, June 2001, http://www.inria.fr/rrrt/rr4475.html
• Christoph Burnikel and Joachim Ziegler, “Fast Recursive Division”, MaxPlanckInstitut
fuer Informatik Research Report MPII981022,
http://data.mpisb.mpg.de/internet/reports.nsf/NumberView/19981022
• Torbj¨orn Granlund and Peter L. Montgomery, “Division by Invariant Integers using Multi
plication”, in Proceedings of the SIGPLAN PLDI’94 Conference, June 1994. Also available
ftp://ftp.cwi.nl/pub/pmontgom/divcnst.psa4.gz (and .psl.gz).
• Niels M¨oller and Torbj¨orn Granlund, “Improved division by invariant integers”, to appear.
• Torbj¨orn Granlund and Niels M¨oller, “Division of integers large and small”, to appear.
• Tudor Jebelean, “An algorithm for exact division”, Journal of Symbolic Computation, vol
ume 15, 1993, pp. 169180. Research report version available
ftp://ftp.risc.unilinz.ac.at/pub/techreports/1992/9235.ps.gz
• Tudor Jebelean, “Exact Division with Karatsuba Complexity  Extended Abstract”, RISC
Linz technical report 9631,
ftp://ftp.risc.unilinz.ac.at/pub/techreports/1996/9631.ps.gz
• Tudor Jebelean, “Practical Integer Division with Karatsuba Complexity”, ISSAC 97, pp.
339341. Technical report available
ftp://ftp.risc.unilinz.ac.at/pub/techreports/1996/9629.ps.gz
• Tudor Jebelean, “A Generalization of the Binary GCD Algorithm”, ISSAC 93, pp. 111116.
Technical report version available
ftp://ftp.risc.unilinz.ac.at/pub/techreports/1993/9301.ps.gz
Appendix B: References 123
• Tudor Jebelean, “A DoubleDigit LehmerEuclid Algorithm for Finding the GCD of Long
Integers”, Journal of Symbolic Computation, volume 19, 1995, pp. 145157. Technical
report version also available
ftp://ftp.risc.unilinz.ac.at/pub/techreports/1992/9269.ps.gz
• Werner Krandick and Tudor Jebelean, “Bidirectional Exact Integer Division”, Journal of
Symbolic Computation, volume 21, 1996, pp. 441455. Early technical report version also
available ftp://ftp.risc.unilinz.ac.at/pub/techreports/1994/9450.ps.gz
• Makoto Matsumoto and Takuji Nishimura, “Mersenne Twister: A 623dimensionally
equidistributed uniform pseudorandom number generator”, ACM Transactions on Mod
elling and Computer Simulation, volume 8, January 1998, pp. 330. Available online
http://www.math.sci.hiroshimau.ac.jp/~mmat/MT/ARTICLES/mt.ps.gz (or .pdf)
• R. Moenck and A. Borodin, “Fast Modular Transforms via Division”, Proceedings of the
13th Annual IEEE Symposium on Switching and Automata Theory, October 1972, pp. 90
96. Reprinted as “Fast Modular Transforms”, Journal of Computer and System Sciences,
volume 8, number 3, June 1974, pp. 366386.
• Niels M¨oller, “On Sch¨onhage’s algorithm and subquadratic integer GCD computation”, in
Mathematics of Computation, volume 77, January 2008, pp. 589607.
• Peter L. Montgomery, “Modular Multiplication Without Trial Division”, in Mathematics
of Computation, volume 44, number 170, April 1985.
• Arnold Sch¨onhage and Volker Strassen, “Schnelle Multiplikation grosser Zahlen”, Comput
ing 7, 1971, pp. 281292.
• Kenneth Weber, “The accelerated integer GCD algorithm”, ACM Transactions on Mathe
matical Software, volume 21, number 1, March 1995, pp. 111122.
• Paul Zimmermann, “Karatsuba Square Root”, INRIA Research Report 3805, November
1999, http://www.inria.fr/rrrt/rr3805.html
• Paul Zimmermann, “A Proof of GMP Fast Division and Square Root Implementations”,
http://www.loria.fr/~zimmerma/papers/proofdivsqrt.ps.gz
• Dan Zuras, “On Squaring and Multiplying Large Integers”, ARITH11: IEEE Symposium
on Computer Arithmetic, 1993, pp. 260 to 271. Reprinted as “More on Multiplying and
Squaring Large Integers”, IEEE Transactions on Computers, volume 43, number 8, August
1994, pp. 899908.
124 GNU MP 5.0.1
Appendix C GNU Free Documentation License
Version 1.3, 3 November 2008
Copyright c ( 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc.
http://fsf.org/
Everyone is permitted to copy and distribute verbatim copies
of this license document, but changing it is not allowed.
0. PREAMBLE
The purpose of this License is to make a manual, textbook, or other functional and useful
document free in the sense of freedom: to assure everyone the efective freedom to copy
and redistribute it, with or without modifying it, either commercially or noncommercially.
Secondarily, this License preserves for the author and publisher a way to get credit for their
work, while not being considered responsible for modifcations made by others.
This License is a kind of “copyleft”, which means that derivative works of the document
must themselves be free in the same sense. It complements the GNU General Public License,
which is a copyleft license designed for free software.
We have designed this License in order to use it for manuals for free software, because free
software needs free documentation: a free program should come with manuals providing the
same freedoms that the software does. But this License is not limited to software manuals;
it can be used for any textual work, regardless of subject matter or whether it is published
as a printed book. We recommend this License principally for works whose purpose is
instruction or reference.
1. APPLICABILITY AND DEFINITIONS
This License applies to any manual or other work, in any medium, that contains a notice
placed by the copyright holder saying it can be distributed under the terms of this License.
Such a notice grants a worldwide, royaltyfree license, unlimited in duration, to use that
work under the conditions stated herein. The “Document”, below, refers to any such manual
or work. Any member of the public is a licensee, and is addressed as “you”. You accept
the license if you copy, modify or distribute the work in a way requiring permission under
copyright law.
A “Modifed Version” of the Document means any work containing the Document or a
portion of it, either copied verbatim, or with modifcations and/or translated into another
language.
A “Secondary Section” is a named appendix or a frontmatter section of the Document
that deals exclusively with the relationship of the publishers or authors of the Document
to the Document’s overall subject (or to related matters) and contains nothing that could
fall directly within that overall subject. (Thus, if the Document is in part a textbook of
mathematics, a Secondary Section may not explain any mathematics.) The relationship
could be a matter of historical connection with the subject or with related matters, or of
legal, commercial, philosophical, ethical or political position regarding them.
The “Invariant Sections” are certain Secondary Sections whose titles are designated, as
being those of Invariant Sections, in the notice that says that the Document is released
under this License. If a section does not ft the above defnition of Secondary then it is not
allowed to be designated as Invariant. The Document may contain zero Invariant Sections.
If the Document does not identify any Invariant Sections then there are none.
The “Cover Texts” are certain short passages of text that are listed, as FrontCover Texts or
BackCover Texts, in the notice that says that the Document is released under this License.
A FrontCover Text may be at most 5 words, and a BackCover Text may be at most 25
words.
Appendix C: GNU Free Documentation License 125
A “Transparent” copy of the Document means a machinereadable copy, represented in a
format whose specifcation is available to the general public, that is suitable for revising
the document straightforwardly with generic text editors or (for images composed of pixels)
generic paint programs or (for drawings) some widely available drawing editor, and that is
suitable for input to text formatters or for automatic translation to a variety of formats
suitable for input to text formatters. A copy made in an otherwise Transparent fle format
whose markup, or absence of markup, has been arranged to thwart or discourage subsequent
modifcation by readers is not Transparent. An image format is not Transparent if used for
any substantial amount of text. A copy that is not “Transparent” is called “Opaque”.
Examples of suitable formats for Transparent copies include plain ascii without markup,
Texinfo input format, LaT
E
X input format, SGML or XML using a publicly available DTD,
and standardconforming simple HTML, PostScript or PDF designed for human modifca
tion. Examples of transparent image formats include PNG, XCF and JPG. Opaque formats
include proprietary formats that can be read and edited only by proprietary word proces
sors, SGML or XML for which the DTD and/or processing tools are not generally available,
and the machinegenerated HTML, PostScript or PDF produced by some word processors
for output purposes only.
The “Title Page” means, for a printed book, the title page itself, plus such following pages
as are needed to hold, legibly, the material this License requires to appear in the title page.
For works in formats which do not have any title page as such, “Title Page” means the
text near the most prominent appearance of the work’s title, preceding the beginning of the
body of the text.
The “publisher” means any person or entity that distributes copies of the Document to the
public.
A section “Entitled XYZ” means a named subunit of the Document whose title either
is precisely XYZ or contains XYZ in parentheses following text that translates XYZ in
another language. (Here XYZ stands for a specifc section name mentioned below, such
as “Acknowledgements”, “Dedications”, “Endorsements”, or “History”.) To “Preserve the
Title” of such a section when you modify the Document means that it remains a section
“Entitled XYZ” according to this defnition.
The Document may include Warranty Disclaimers next to the notice which states that
this License applies to the Document. These Warranty Disclaimers are considered to be
included by reference in this License, but only as regards disclaiming warranties: any other
implication that these Warranty Disclaimers may have is void and has no efect on the
meaning of this License.
2. VERBATIM COPYING
You may copy and distribute the Document in any medium, either commercially or noncom
mercially, provided that this License, the copyright notices, and the license notice saying
this License applies to the Document are reproduced in all copies, and that you add no
other conditions whatsoever to those of this License. You may not use technical measures
to obstruct or control the reading or further copying of the copies you make or distribute.
However, you may accept compensation in exchange for copies. If you distribute a large
enough number of copies you must also follow the conditions in section 3.
You may also lend copies, under the same conditions stated above, and you may publicly
display copies.
3. COPYING IN QUANTITY
If you publish printed copies (or copies in media that commonly have printed covers) of the
Document, numbering more than 100, and the Document’s license notice requires Cover
Texts, you must enclose the copies in covers that carry, clearly and legibly, all these Cover
Texts: FrontCover Texts on the front cover, and BackCover Texts on the back cover. Both
126 GNU MP 5.0.1
covers must also clearly and legibly identify you as the publisher of these copies. The front
cover must present the full title with all words of the title equally prominent and visible.
You may add other material on the covers in addition. Copying with changes limited to
the covers, as long as they preserve the title of the Document and satisfy these conditions,
can be treated as verbatim copying in other respects.
If the required texts for either cover are too voluminous to ft legibly, you should put the
frst ones listed (as many as ft reasonably) on the actual cover, and continue the rest onto
adjacent pages.
If you publish or distribute Opaque copies of the Document numbering more than 100, you
must either include a machinereadable Transparent copy along with each Opaque copy,
or state in or with each Opaque copy a computernetwork location from which the general
networkusing public has access to download using publicstandard network protocols a
complete Transparent copy of the Document, free of added material. If you use the latter
option, you must take reasonably prudent steps, when you begin distribution of Opaque
copies in quantity, to ensure that this Transparent copy will remain thus accessible at the
stated location until at least one year after the last time you distribute an Opaque copy
(directly or through your agents or retailers) of that edition to the public.
It is requested, but not required, that you contact the authors of the Document well before
redistributing any large number of copies, to give them a chance to provide you with an
updated version of the Document.
4. MODIFICATIONS
You may copy and distribute a Modifed Version of the Document under the conditions
of sections 2 and 3 above, provided that you release the Modifed Version under precisely
this License, with the Modifed Version flling the role of the Document, thus licensing
distribution and modifcation of the Modifed Version to whoever possesses a copy of it. In
addition, you must do these things in the Modifed Version:
A. Use in the Title Page (and on the covers, if any) a title distinct from that of the
Document, and from those of previous versions (which should, if there were any, be
listed in the History section of the Document). You may use the same title as a previous
version if the original publisher of that version gives permission.
B. List on the Title Page, as authors, one or more persons or entities responsible for
authorship of the modifcations in the Modifed Version, together with at least fve of
the principal authors of the Document (all of its principal authors, if it has fewer than
fve), unless they release you from this requirement.
C. State on the Title page the name of the publisher of the Modifed Version, as the
publisher.
D. Preserve all the copyright notices of the Document.
E. Add an appropriate copyright notice for your modifcations adjacent to the other copy
right notices.
F. Include, immediately after the copyright notices, a license notice giving the public
permission to use the Modifed Version under the terms of this License, in the form
shown in the Addendum below.
G. Preserve in that license notice the full lists of Invariant Sections and required Cover
Texts given in the Document’s license notice.
H. Include an unaltered copy of this License.
I. Preserve the section Entitled “History”, Preserve its Title, and add to it an item stating
at least the title, year, new authors, and publisher of the Modifed Version as given
on the Title Page. If there is no section Entitled “History” in the Document, create
one stating the title, year, authors, and publisher of the Document as given on its
Appendix C: GNU Free Documentation License 127
Title Page, then add an item describing the Modifed Version as stated in the previous
sentence.
J. Preserve the network location, if any, given in the Document for public access to a
Transparent copy of the Document, and likewise the network locations given in the
Document for previous versions it was based on. These may be placed in the “History”
section. You may omit a network location for a work that was published at least four
years before the Document itself, or if the original publisher of the version it refers to
gives permission.
K. For any section Entitled “Acknowledgements” or “Dedications”, Preserve the Title
of the section, and preserve in the section all the substance and tone of each of the
contributor acknowledgements and/or dedications given therein.
L. Preserve all the Invariant Sections of the Document, unaltered in their text and in their
titles. Section numbers or the equivalent are not considered part of the section titles.
M. Delete any section Entitled “Endorsements”. Such a section may not be included in
the Modifed Version.
N. Do not retitle any existing section to be Entitled “Endorsements” or to confict in title
with any Invariant Section.
O. Preserve any Warranty Disclaimers.
If the Modifed Version includes new frontmatter sections or appendices that qualify as
Secondary Sections and contain no material copied from the Document, you may at your
option designate some or all of these sections as invariant. To do this, add their titles to
the list of Invariant Sections in the Modifed Version’s license notice. These titles must be
distinct from any other section titles.
You may add a section Entitled “Endorsements”, provided it contains nothing but endorse
ments of your Modifed Version by various parties—for example, statements of peer review
or that the text has been approved by an organization as the authoritative defnition of a
standard.
You may add a passage of up to fve words as a FrontCover Text, and a passage of up
to 25 words as a BackCover Text, to the end of the list of Cover Texts in the Modifed
Version. Only one passage of FrontCover Text and one of BackCover Text may be added
by (or through arrangements made by) any one entity. If the Document already includes
a cover text for the same cover, previously added by you or by arrangement made by the
same entity you are acting on behalf of, you may not add another; but you may replace the
old one, on explicit permission from the previous publisher that added the old one.
The author(s) and publisher(s) of the Document do not by this License give permission to
use their names for publicity for or to assert or imply endorsement of any Modifed Version.
5. COMBINING DOCUMENTS
You may combine the Document with other documents released under this License, under
the terms defned in section 4 above for modifed versions, provided that you include in the
combination all of the Invariant Sections of all of the original documents, unmodifed, and
list them all as Invariant Sections of your combined work in its license notice, and that you
preserve all their Warranty Disclaimers.
The combined work need only contain one copy of this License, and multiple identical
Invariant Sections may be replaced with a single copy. If there are multiple Invariant
Sections with the same name but diferent contents, make the title of each such section
unique by adding at the end of it, in parentheses, the name of the original author or
publisher of that section if known, or else a unique number. Make the same adjustment to
the section titles in the list of Invariant Sections in the license notice of the combined work.
In the combination, you must combine any sections Entitled “History” in the various original
documents, forming one section Entitled “History”; likewise combine any sections Entitled
128 GNU MP 5.0.1
“Acknowledgements”, and any sections Entitled “Dedications”. You must delete all sections
Entitled “Endorsements.”
6. COLLECTIONS OF DOCUMENTS
You may make a collection consisting of the Document and other documents released under
this License, and replace the individual copies of this License in the various documents with
a single copy that is included in the collection, provided that you follow the rules of this
License for verbatim copying of each of the documents in all other respects.
You may extract a single document from such a collection, and distribute it individually
under this License, provided you insert a copy of this License into the extracted document,
and follow this License in all other respects regarding verbatim copying of that document.
7. AGGREGATION WITH INDEPENDENT WORKS
A compilation of the Document or its derivatives with other separate and independent
documents or works, in or on a volume of a storage or distribution medium, is called an
“aggregate” if the copyright resulting from the compilation is not used to limit the legal
rights of the compilation’s users beyond what the individual works permit. When the
Document is included in an aggregate, this License does not apply to the other works in
the aggregate which are not themselves derivative works of the Document.
If the Cover Text requirement of section 3 is applicable to these copies of the Document,
then if the Document is less than one half of the entire aggregate, the Document’s Cover
Texts may be placed on covers that bracket the Document within the aggregate, or the
electronic equivalent of covers if the Document is in electronic form. Otherwise they must
appear on printed covers that bracket the whole aggregate.
8. TRANSLATION
Translation is considered a kind of modifcation, so you may distribute translations of the
Document under the terms of section 4. Replacing Invariant Sections with translations
requires special permission from their copyright holders, but you may include translations
of some or all Invariant Sections in addition to the original versions of these Invariant
Sections. You may include a translation of this License, and all the license notices in
the Document, and any Warranty Disclaimers, provided that you also include the original
English version of this License and the original versions of those notices and disclaimers. In
case of a disagreement between the translation and the original version of this License or a
notice or disclaimer, the original version will prevail.
If a section in the Document is Entitled “Acknowledgements”, “Dedications”, or “History”,
the requirement (section 4) to Preserve its Title (section 1) will typically require changing
the actual title.
9. TERMINATION
You may not copy, modify, sublicense, or distribute the Document except as expressly pro
vided under this License. Any attempt otherwise to copy, modify, sublicense, or distribute
it is void, and will automatically terminate your rights under this License.
However, if you cease all violation of this License, then your license from a particular copy
right holder is reinstated (a) provisionally, unless and until the copyright holder explicitly
and fnally terminates your license, and (b) permanently, if the copyright holder fails to
notify you of the violation by some reasonable means prior to 60 days after the cessation.
Moreover, your license from a particular copyright holder is reinstated permanently if the
copyright holder notifes you of the violation by some reasonable means, this is the frst
time you have received notice of violation of this License (for any work) from that copyright
holder, and you cure the violation prior to 30 days after your receipt of the notice.
Termination of your rights under this section does not terminate the licenses of parties
who have received copies or rights from you under this License. If your rights have been
Appendix C: GNU Free Documentation License 129
terminated and not permanently reinstated, receipt of a copy of some or all of the same
material does not give you any rights to use it.
10. FUTURE REVISIONS OF THIS LICENSE
The Free Software Foundation may publish new, revised versions of the GNU Free Doc
umentation License from time to time. Such new versions will be similar in spirit to
the present version, but may difer in detail to address new problems or concerns. See
http://www.gnu.org/copyleft/.
Each version of the License is given a distinguishing version number. If the Document
specifes that a particular numbered version of this License “or any later version” applies
to it, you have the option of following the terms and conditions either of that specifed
version or of any later version that has been published (not as a draft) by the Free Software
Foundation. If the Document does not specify a version number of this License, you may
choose any version ever published (not as a draft) by the Free Software Foundation. If the
Document specifes that a proxy can decide which future versions of this License can be
used, that proxy’s public statement of acceptance of a version permanently authorizes you
to choose that version for the Document.
11. RELICENSING
“Massive Multiauthor Collaboration Site” (or “MMC Site”) means any World Wide Web
server that publishes copyrightable works and also provides prominent facilities for anybody
to edit those works. A public wiki that anybody can edit is an example of such a server. A
“Massive Multiauthor Collaboration” (or “MMC”) contained in the site means any set of
copyrightable works thus published on the MMC site.
“CCBYSA” means the Creative Commons AttributionShare Alike 3.0 license published
by Creative Commons Corporation, a notforproft corporation with a principal place of
business in San Francisco, California, as well as future copyleft versions of that license
published by that same organization.
“Incorporate” means to publish or republish a Document, in whole or in part, as part of
another Document.
An MMC is “eligible for relicensing” if it is licensed under this License, and if all works that
were frst published under this License somewhere other than this MMC, and subsequently
incorporated in whole or in part into the MMC, (1) had no cover texts or invariant sections,
and (2) were thus incorporated prior to November 1, 2008.
The operator of an MMC Site may republish an MMC contained in the site under CCBY
SA on the same site at any time before August 1, 2009, provided the MMC is eligible for
relicensing.
130 GNU MP 5.0.1
ADDENDUM: How to use this License for your documents
To use this License in a document you have written, include a copy of the License in the document
and put the following copyright and license notices just after the title page:
Copyright (C) year your name.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.3
or any later version published by the Free Software Foundation;
with no Invariant Sections, no FrontCover Texts, and no BackCover
Texts. A copy of the license is included in the section entitled ‘‘GNU
Free Documentation License’’.
If you have Invariant Sections, FrontCover Texts and BackCover Texts, replace the
“with. . . Texts.” line with this:
with the Invariant Sections being list their titles, with
the FrontCover Texts being list, and with the BackCover Texts
being list.
If you have Invariant Sections without Cover Texts, or some other combination of the three,
merge those two alternatives to suit the situation.
If your document contains nontrivial examples of program code, we recommend releasing these
examples in parallel under your choice of free software license, such as the GNU General Public
License, to permit their use in free software.
Concept Index 131
Concept Index
#
#include . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

build. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
disablefft. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
disableshared . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
disablestatic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
enablealloca . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
enableassert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
enablecxx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
‘enablefat’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
enablempbsd. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
enableprofiling . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 25
execprefix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
host. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
prefix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
finstrumentfunctions . . . . . . . . . . . . . . . . . . . . . . . . 26
2
2exp functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
6
68000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
8
80x86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
A
ABI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5, 8
About this manual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
AC_CHECK_LIB. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
AIX. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10, 12
Algorithms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
alloca. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Allocation of memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
AMD64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Anonymous FTP of latest version . . . . . . . . . . . . . . . . . 2
Application Binary Interface. . . . . . . . . . . . . . . . . . . . . . . 8
Arithmetic functions . . . . . . . . . . . . . . . . . . . . . . 32, 45, 52
ARM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Assembly cache handling. . . . . . . . . . . . . . . . . . . . . . . . 110
Assembly carry propagation. . . . . . . . . . . . . . . . . . . . . 110
Assembly code organisation . . . . . . . . . . . . . . . . . . . . . 109
Assembly coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Assembly foating Point . . . . . . . . . . . . . . . . . . . . . . . . . 111
Assembly loop unrolling. . . . . . . . . . . . . . . . . . . . . . . . . 113
Assembly SIMD. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Assembly software pipelining . . . . . . . . . . . . . . . . . . . . 112
Assembly writing guide . . . . . . . . . . . . . . . . . . . . . . . . . 113
Assertion checking. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 24
Assignment functions. . . . . . . . . . . . . . . . . . 30, 44, 50, 51
Autoconf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
B
Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Berkeley MP compatible functions . . . . . . . . . . . . . 7, 84
Binomial coefcient algorithm. . . . . . . . . . . . . . . . . . . 107
Binomial coefcient functions. . . . . . . . . . . . . . . . . . . . . 37
Binutils strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Bit manipulation functions . . . . . . . . . . . . . . . . . . . . . . . 38
Bit scanning functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Bit shift left . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Bit shift right . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Bits per limb. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
BSD MP compatible functions . . . . . . . . . . . . . . . . . 7, 84
Bug reporting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Build directory. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Build notes for binary packaging . . . . . . . . . . . . . . . . . 11
Build notes for particular systems . . . . . . . . . . . . . . . . 12
Build options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Build problems known . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Build system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Building GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Bus error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
C
C compiler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
C++ compiler. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
C++ interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
C++ interface internals . . . . . . . . . . . . . . . . . . . . . . . . . . 118
C++ istream input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
C++ ostream output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
C++ support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
CC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
CC_FOR_BUILD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
CFLAGS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Checker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
checkergcc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Code organisation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Compaq C++. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Comparison functions . . . . . . . . . . . . . . . . . . . . . 37, 46, 53
Compatibility with older versions . . . . . . . . . . . . . . . . . 20
Conditions for copying GNU MP . . . . . . . . . . . . . . . . . . 1
Confguring GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Congruence algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Congruence functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Conventions for parameters . . . . . . . . . . . . . . . . . . . . . . 18
Conventions for variables . . . . . . . . . . . . . . . . . . . . . . . . . 17
Conversion functions . . . . . . . . . . . . . . . . . . . . . . 31, 45, 51
Copying conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
CPPFLAGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
CPU types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2, 4
Cross compiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Custom allocation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
CXX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
CXXFLAGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Cygwin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
132 GNU MP 5.0.1
D
Darwin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Demonstration programs . . . . . . . . . . . . . . . . . . . . . . . . . 20
Digits in an integer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Divisibility algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Divisibility functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Divisibility testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Division algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Division functions. . . . . . . . . . . . . . . . . . . . . . . . . 32, 45, 52
DJGPP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12, 14
DLLs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
DocBook. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Documentation formats . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Documentation license . . . . . . . . . . . . . . . . . . . . . . . . . . 124
DVI. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
E
Efciency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Emacs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Exact division functions . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Exact remainder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Example programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Exec prefx. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Execution profling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 25
Exponentiation functions . . . . . . . . . . . . . . . . . . . . . 34, 53
Export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Expression parsing demo . . . . . . . . . . . . . . . . . . . . . . . . . 21
Extended GCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
F
Factor removal functions . . . . . . . . . . . . . . . . . . . . . . . . . 37
Factorial algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Factorial functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Factorization demo. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Fast Fourier Transform. . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Fat binary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
FFT multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 94
Fibonacci number algorithm . . . . . . . . . . . . . . . . . . . . 107
Fibonacci sequence functions . . . . . . . . . . . . . . . . . . . . . 37
Float arithmetic functions . . . . . . . . . . . . . . . . . . . . . . . . 52
Float assignment functions . . . . . . . . . . . . . . . . . . . 50, 51
Float comparison functions . . . . . . . . . . . . . . . . . . . . . . . 53
Float conversion functions . . . . . . . . . . . . . . . . . . . . . . . . 51
Float functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Float initialization functions . . . . . . . . . . . . . . . . . . 48, 51
Float input and output functions . . . . . . . . . . . . . . . . . 53
Float internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Float miscellaneous functions. . . . . . . . . . . . . . . . . . . . . 54
Float random number functions . . . . . . . . . . . . . . . . . . 54
Float rounding functions . . . . . . . . . . . . . . . . . . . . . . . . . 54
Float sign tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Floating point mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Floatingpoint functions . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Floatingpoint number . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
fnccheck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Formatted input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Formatted output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
Free Documentation License. . . . . . . . . . . . . . . . . . . . . 124
frexp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31, 51
FTP of latest version. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Function classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
FunctionCheck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
G
GCC Checker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
GCD algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
GCD extended . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
GCD functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
GDB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Generic C. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
GMP Perl module. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
GMP version number . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
‘gmp.h’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
gmpxx.h. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
GNU Debugger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
GNU Free Documentation License. . . . . . . . . . . . . . . 124
GNU strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
gprof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Greatest common divisor algorithms. . . . . . . . . . . . . 100
Greatest common divisor functions . . . . . . . . . . . . . . . 36
H
Hardware foating point mode . . . . . . . . . . . . . . . . . . . . 12
Headers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Heap problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Home page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Host system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
HPUX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
HPPA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
I
I/O functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39, 47, 53
i386. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
IA64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Inplace operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Include fles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
infolookupsymbol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Initialization functions . . . . . . . . . 29, 30, 44, 48, 51, 65
Initializing and clearing . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Input functions. . . . . . . . . . . . . . . . . . . . . . . . 39, 47, 53, 74
Install prefx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Installing GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Instruction Set Architecture . . . . . . . . . . . . . . . . . . . . . . . 8
instrumentfunctions. . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Integer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Integer arithmetic functions . . . . . . . . . . . . . . . . . . . . . . 32
Integer assignment functions . . . . . . . . . . . . . . . . . . . . . 30
Integer bit manipulation functions . . . . . . . . . . . . . . . . 38
Integer comparison functions . . . . . . . . . . . . . . . . . . . . . 37
Integer conversion functions . . . . . . . . . . . . . . . . . . . . . . 31
Integer division functions . . . . . . . . . . . . . . . . . . . . . . . . . 32
Integer exponentiation functions . . . . . . . . . . . . . . . . . . 34
Integer export. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Integer functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Integer import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Integer initialization functions . . . . . . . . . . . . . . . . 29, 30
Integer input and output functions . . . . . . . . . . . . . . . 39
Integer internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Integer logical functions . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Integer miscellaneous functions . . . . . . . . . . . . . . . . . . . 42
Integer random number functions. . . . . . . . . . . . . . . . . 40
Integer root functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Concept Index 133
Integer sign tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Integer special functions . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Interix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Inverse modulo functions . . . . . . . . . . . . . . . . . . . . . . . . . 36
IRIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9, 14
ISA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
istream input. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
J
Jacobi symbol algorithm . . . . . . . . . . . . . . . . . . . . . . . . 102
Jacobi symbol functions . . . . . . . . . . . . . . . . . . . . . . . . . . 36
K
Karatsuba multiplication. . . . . . . . . . . . . . . . . . . . . . . . . 91
Karatsuba square root algorithm. . . . . . . . . . . . . . . . 103
Kronecker symbol functions . . . . . . . . . . . . . . . . . . . . . . 36
L
Language bindings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Latest version of GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
LCM functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Least common multiple functions . . . . . . . . . . . . . . . . . 36
Legendre symbol functions . . . . . . . . . . . . . . . . . . . . . . . 36
libgmp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
libgmpxx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Libtool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Libtool versioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
License conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Limb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Limb size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Linear congruential algorithm . . . . . . . . . . . . . . . . . . . 108
Linear congruential random numbers . . . . . . . . . . . . . 65
Linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Logical functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
Lowlevel functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Lucas number algorithm . . . . . . . . . . . . . . . . . . . . . . . . 108
Lucas number functions . . . . . . . . . . . . . . . . . . . . . . . . . . 37
M
MacOS X. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Mailing lists. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Malloc debugger . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Malloc problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Memory allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
Memory management . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Mersenne twister algorithm . . . . . . . . . . . . . . . . . . . . . 108
Mersenne twister random numbers . . . . . . . . . . . . . . . 65
MINGW. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
MIPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Miscellaneous foat functions . . . . . . . . . . . . . . . . . . . . . 54
Miscellaneous integer functions . . . . . . . . . . . . . . . . . . . 42
MMX. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Modular inverse functions . . . . . . . . . . . . . . . . . . . . . . . . 36
Most signifcant bit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mp.h . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
MPN_PATH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
MS Windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12, 13
MSDOS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Multithreading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Multiplication algorithms . . . . . . . . . . . . . . . . . . . . . . . . 90
N
Nails. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
Native compilation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
NeXT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Next prime function. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
NonUnix systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Nth root algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Number sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Number theoretic functions. . . . . . . . . . . . . . . . . . . . . . . 35
Numerator and denominator . . . . . . . . . . . . . . . . . . . . . 46
O
obstack output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
OpenBSD. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Optimizing performance. . . . . . . . . . . . . . . . . . . . . . . . . . 15
ostream output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Other languages. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Output functions . . . . . . . . . . . . . . . . . . . . . . 39, 47, 53, 69
P
Packaged builds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Parameter conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Parsing expressions demo . . . . . . . . . . . . . . . . . . . . . . . . 21
Particular systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Past GMP versions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
PDF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Perfect power algorithm. . . . . . . . . . . . . . . . . . . . . . . . . 104
Perfect power functions . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Perfect square algorithm . . . . . . . . . . . . . . . . . . . . . . . . 104
Perfect square functions . . . . . . . . . . . . . . . . . . . . . . . . . . 35
perl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Perl module. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Postscript . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Power/PowerPC. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13, 15
Powering algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Powering functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 34, 53
PowerPC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Precision of foats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Precision of hardware foating point . . . . . . . . . . . . . . 12
Prefx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Prime testing algorithms . . . . . . . . . . . . . . . . . . . . . . . . 106
Prime testing functions. . . . . . . . . . . . . . . . . . . . . . . . . . . 35
printf formatted output . . . . . . . . . . . . . . . . . . . . . . . . . 67
Probable prime testing functions . . . . . . . . . . . . . . . . . 35
prof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Profling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
R
Radix conversion algorithms. . . . . . . . . . . . . . . . . . . . . 105
Random number algorithms . . . . . . . . . . . . . . . . . . . . . 108
Random number functions . . . . . . . . . . . . . . . . 40, 54, 65
Random number seeding . . . . . . . . . . . . . . . . . . . . . . . . . 66
Random number state. . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
Random state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Rational arithmetic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
134 GNU MP 5.0.1
Rational arithmetic functions. . . . . . . . . . . . . . . . . . . . . 45
Rational assignment functions . . . . . . . . . . . . . . . . . . . . 44
Rational comparison functions. . . . . . . . . . . . . . . . . . . . 46
Rational conversion functions. . . . . . . . . . . . . . . . . . . . . 45
Rational initialization functions . . . . . . . . . . . . . . . . . . 44
Rational input and output functions . . . . . . . . . . . . . . 47
Rational internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Rational number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Rational number functions . . . . . . . . . . . . . . . . . . . . . . . 44
Rational numerator and denominator . . . . . . . . . . . . . 46
Rational sign tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Raw output internals. . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Reallocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Reentrancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Remove factor functions . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Reporting bugs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Root extraction algorithm. . . . . . . . . . . . . . . . . . . . . . . 104
Root extraction algorithms . . . . . . . . . . . . . . . . . . . . . . 103
Root extraction functions . . . . . . . . . . . . . . . . . . . . . 35, 52
Root testing functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Rounding functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
S
Sample programs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Scan bit functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
scanf formatted input . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
SCO. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Seeding random numbers . . . . . . . . . . . . . . . . . . . . . . . . . 66
Segmentation violation . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Sequent Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
Services for Unix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Shared library versioning . . . . . . . . . . . . . . . . . . . . . . . . . 11
Sign tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38, 46, 53
Size in digits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Small operands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Solaris . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10, 15
Sparc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Sparc V9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Special integer functions . . . . . . . . . . . . . . . . . . . . . . . . . 42
Square root algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . 103
SSE2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Stack backtrace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Stack overfow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 23
Static linking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
stdarg.h . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
stdio.h. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Stripped libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
Sun. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
SunOS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
T
Temporary memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Texinfo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Text input/output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Thread safety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
Toom multiplication . . . . . . . . . . . . . . . . . . . . . . 92, 94, 95
Types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
U
ui and si functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
Unbalanced multiplication. . . . . . . . . . . . . . . . . . . . . . . . 96
Upward compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Useful macros and constants . . . . . . . . . . . . . . . . . . . . . 20
Userdefned precision . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
V
Valgrind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Variable conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
Version number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
W
Web page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Windows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12, 13
X
x86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
x87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
XML. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Function and Type Index 135
Function and Type Index
__GMP_CC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
__GMP_CFLAGS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
__GNU_MP_VERSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
__GNU_MP_VERSION_MINOR . . . . . . . . . . . . . . . . . . . . . . . . 20
__GNU_MP_VERSION_PATCHLEVEL. . . . . . . . . . . . . . . . . . . 20
_mpz_realloc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
A
abs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78, 79, 81
C
ceil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
cmp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78, 79, 81
F
floor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
G
gcd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
gmp_asprintf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
gmp_errno . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
GMP_ERROR_INVALID_ARGUMENT. . . . . . . . . . . . . . . . . . . . 66
GMP_ERROR_UNSUPPORTED_ARGUMENT . . . . . . . . . . . . . . . 66
gmp_fprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
gmp_fscanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
GMP_LIMB_BITS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
GMP_NAIL_BITS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
GMP_NAIL_MASK. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
GMP_NUMB_BITS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
GMP_NUMB_MASK. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
GMP_NUMB_MAX. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
gmp_obstack_printf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
gmp_obstack_vprintf. . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
gmp_printf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
GMP_RAND_ALG_DEFAULT. . . . . . . . . . . . . . . . . . . . . . . . . . . 65
GMP_RAND_ALG_LC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
gmp_randclass. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
gmp_randclass::get_f. . . . . . . . . . . . . . . . . . . . . . . . . . . 82
gmp_randclass::get_z_bits . . . . . . . . . . . . . . . . . . . . . 82
gmp_randclass::get_z_range. . . . . . . . . . . . . . . . . . . . 82
gmp_randclass::gmp_randclass . . . . . . . . . . . . . . . . . 82
gmp_randclass::seed. . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
gmp_randclear. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
gmp_randinit. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
gmp_randinit_default. . . . . . . . . . . . . . . . . . . . . . . . . . . 65
gmp_randinit_lc_2exp. . . . . . . . . . . . . . . . . . . . . . . . . . . 65
gmp_randinit_lc_2exp_size . . . . . . . . . . . . . . . . . . . . . 65
gmp_randinit_mt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
gmp_randinit_set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
gmp_randseed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
gmp_randseed_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
gmp_randstate_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
gmp_scanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
gmp_snprintf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
gmp_sprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
gmp_sscanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
gmp_urandomb_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
gmp_urandomm_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
gmp_vasprintf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
gmp_version . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
gmp_vfprintf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
gmp_vfscanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
gmp_vprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
gmp_vscanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
gmp_vsnprintf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
gmp_vsprintf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
gmp_vsscanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
H
hypot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
I
itom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
M
madd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
mcmp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
mdiv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
mfree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
min . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
MINT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
mout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
move . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
mp_bitcnt_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
mp_bits_per_limb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
mp_exp_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
mp_get_memory_functions . . . . . . . . . . . . . . . . . . . . . . . 87
mp_limb_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
mp_set_memory_functions . . . . . . . . . . . . . . . . . . . . . . . 86
mp_size_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
mpf_abs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_add. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_add_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_ceil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
mpf_class::fits_sint_p . . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::fits_slong_p . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::fits_sshort_p . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::fits_uint_p . . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::fits_ulong_p . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::fits_ushort_p . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::get_mpf_t. . . . . . . . . . . . . . . . . . . . . . . . . . . 77
mpf_class::get_prec. . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::get_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::get_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::mpf_class. . . . . . . . . . . . . . . . . . . . . . . . . . . 80
mpf_class::operator=. . . . . . . . . . . . . . . . . . . . . . . . . . . 80
mpf_class::set_prec. . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::set_prec_raw . . . . . . . . . . . . . . . . . . . . . . . 81
mpf_class::set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
136 GNU MP 5.0.1
mpf_clear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
mpf_clears . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
mpf_cmp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_cmp_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_cmp_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_cmp_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_div. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_div_2exp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_div_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_eq. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_fits_sint_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_fits_slong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_fits_sshort_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_fits_uint_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_fits_ulong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_fits_ushort_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_floor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_get_d_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_get_default_prec. . . . . . . . . . . . . . . . . . . . . . . . . . . 48
mpf_get_prec. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
mpf_get_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_get_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_init . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
mpf_init_set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_init_set_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_init_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_init_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_init_set_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
mpf_init2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
mpf_inits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
mpf_inp_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_integer_p. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_mul. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_mul_2exp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_mul_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_neg. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_out_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_pow_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_random2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
mpf_reldiff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
mpf_set_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
mpf_set_default_prec. . . . . . . . . . . . . . . . . . . . . . . . . . . 48
mpf_set_prec. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
mpf_set_prec_raw . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
mpf_set_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
mpf_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
mpf_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
mpf_set_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
mpf_set_z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
mpf_sgn. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
mpf_sqrt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_sqrt_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_sub. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_sub_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
mpf_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
mpf_trunc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpf_ui_div . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_ui_sub . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
mpf_urandomb. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
mpn_add. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
mpn_add_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
mpn_add_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
mpn_addmul_1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
mpn_and_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_andn_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_cmp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
mpn_com. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
mpn_copyd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
mpn_copyi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
mpn_divexact_by3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
mpn_divexact_by3c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
mpn_divmod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
mpn_divmod_1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
mpn_divrem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
mpn_divrem_1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
mpn_gcd. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
mpn_gcd_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
mpn_gcdext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
mpn_get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
mpn_hamdist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_ior_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_iorn_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_lshift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
mpn_mod_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
mpn_mul. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
mpn_mul_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
mpn_mul_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
mpn_nand_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_neg. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
mpn_nior_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_perfect_square_p. . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_popcount. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_random . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
mpn_random2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
mpn_rshift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
mpn_scan0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
mpn_scan1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
mpn_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
mpn_sqr. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
mpn_sqrtrem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
mpn_sub. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
mpn_sub_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
mpn_sub_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
mpn_submul_1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
mpn_tdiv_qr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
mpn_xnor_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_xor_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
mpn_zero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
mpq_abs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_add. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_canonicalize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
mpq_class::canonicalize . . . . . . . . . . . . . . . . . . . . . . . 79
mpq_class::get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
mpq_class::get_den . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
mpq_class::get_den_mpz_t . . . . . . . . . . . . . . . . . . . . . . 79
mpq_class::get_mpq_t. . . . . . . . . . . . . . . . . . . . . . . . . . . 77
mpq_class::get_num . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
mpq_class::get_num_mpz_t . . . . . . . . . . . . . . . . . . . . . . 79
mpq_class::get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
mpq_class::mpq_class . . . . . . . . . . . . . . . . . . . . . . . 78, 79
mpq_class::set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
mpq_clear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_clears . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_cmp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_cmp_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Function and Type Index 137
mpq_cmp_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_denref . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_div. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_div_2exp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_equal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_get_den . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
mpq_get_num . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
mpq_get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_init . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_inits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_inp_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
mpq_inv. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_mul. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_mul_2exp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_neg. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_numref . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_out_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
mpq_set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_set_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_set_den . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
mpq_set_f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_set_num . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
mpq_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_set_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_set_z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
mpq_sgn. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
mpq_sub. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
mpq_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
mpz_abs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_add. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_add_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_addmul . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_addmul_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_and. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_array_init . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_bin_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_bin_uiui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_cdiv_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_cdiv_q_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_cdiv_q_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_cdiv_qr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_cdiv_qr_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_cdiv_r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_cdiv_r_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_cdiv_r_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_cdiv_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
mpz_class::fits_sint_p . . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::fits_slong_p . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::fits_sshort_p . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::fits_uint_p . . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::fits_ulong_p . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::fits_ushort_p . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::get_mpz_t. . . . . . . . . . . . . . . . . . . . . . . . . . . 77
mpz_class::get_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::get_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_class::mpz_class. . . . . . . . . . . . . . . . . . . . . . . . . . . 77
mpz_class::set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
mpz_clear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
mpz_clears . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
mpz_clrbit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_cmp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_cmp_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_cmp_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_cmp_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_cmpabs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_cmpabs_d. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_cmpabs_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_com. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_combit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
mpz_congruent_2exp_p. . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_congruent_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_congruent_ui_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_divexact. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_divexact_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_divisible_2exp_p. . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_divisible_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_divisible_ui_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_even_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_export . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
mpz_fac_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_fdiv_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fdiv_q_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fdiv_q_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fdiv_qr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fdiv_qr_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fdiv_r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fdiv_r_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fdiv_r_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fdiv_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_fib_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_fib2_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_fits_sint_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_fits_slong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_fits_sshort_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_fits_uint_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_fits_ulong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_fits_ushort_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_gcd. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_gcd_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_gcdext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
mpz_get_d_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
mpz_get_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
mpz_get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
mpz_get_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
mpz_getlimbn. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
mpz_hamdist . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
mpz_init . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
mpz_init_set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_init_set_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_init_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_init_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
mpz_init_set_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_init2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
mpz_inits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
mpz_inp_raw . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
mpz_inp_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
mpz_invert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_ior. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_jacobi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_kronecker. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_kronecker_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_kronecker_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
138 GNU MP 5.0.1
mpz_lcm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_lcm_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_legendre. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_lucnum_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_lucnum2_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_mod. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_mod_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_mul. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_mul_2exp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_mul_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_mul_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_neg. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_nextprime. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_odd_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_out_raw . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
mpz_out_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
mpz_perfect_power_p. . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_perfect_square_p. . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_popcount. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_pow_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_powm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_powm_sec. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_powm_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
mpz_probab_prime_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_random . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
mpz_random2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
mpz_realloc2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
mpz_remove . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
mpz_root . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_rootrem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_rrandomb. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
mpz_scan0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_scan1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_set_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_set_f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_set_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_set_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_setbit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_sgn. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
mpz_si_kronecker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
mpz_sizeinbase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
mpz_sqrt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_sqrtrem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_sub. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_sub_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_submul . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_submul_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
mpz_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
mpz_tdiv_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tdiv_q_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tdiv_q_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tdiv_qr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tdiv_qr_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tdiv_r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tdiv_r_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tdiv_r_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tdiv_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
mpz_tstbit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
mpz_ui_kronecker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
mpz_ui_pow_ui. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
mpz_ui_sub . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
mpz_urandomb. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
mpz_urandomm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
mpz_xor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
msqrt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
msub . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
mtox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
mult . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
O
operator% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
operator/ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
operator<< . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
operator>> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74, 75, 80
P
pow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
R
rpow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
S
sdiv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
sgn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78, 79, 81
sqrt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78, 81
T
trunc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
X
xtom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
This manual describes how to install and use the GNU multiple precision arithmetic library, version 5.0.1. Copyright 1991, 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2009, 2010 Free Software Foundation, Inc. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, with the FrontCover Texts being “A GNU Manual”, and with the BackCover Texts being “You have freedom to copy and modify this GNU Manual, like GNU software”. A copy of the license is included in Appendix C [GNU Free Documentation License], page 124.
i
Table of Contents
GNU MP Copying Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1 2 Introduction to GNU MP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1 How to use this Manual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Installing GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 2.2 2.3 2.4 2.5 2.6 Build Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 ABI and ISA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Notes for Package Builds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Notes for Particular Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Known Build Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Performance optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3
GMP Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 Headers and Libraries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nomenclature and Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Function Classes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Variable Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Parameter Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Memory Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reentrancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Useful Macros and Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Compatibility with older versions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Demonstration programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Autoconf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Emacs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 16 17 17 18 19 19 20 20 20 21 23 25 26 27
4 5
Reporting Bugs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Integer Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 5.13 Initialization Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Combined Initialization and Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conversion Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Arithmetic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Division Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exponentiation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Root Extraction Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Number Theoretic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Logical and Bit Manipulation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Input and Output Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Random Number Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 30 30 31 32 32 34 35 35 37 38 39 40
. . . . . . . . . . . . . . . . . 65 9. . . . . . . . Comparison Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 10. . . . . . . . . . . . . . . . 74 12 C++ Class Interface . .ii GNU MP 5. . . . . . Conversion Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Applying Integer Functions to Rationals . .5 6. . . . . . . . . . . 48 7. . . . . . . . . . . . . . . . . . . . . . . . . .1 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 C++ Formatted Input . . . . . . . . . . . . . . . . . . . . . . . . 72 11. . . . . . . . . . . . . . 66 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Random Numbers . . . . . . . .2 Formatted Input Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 5. . . . . . . . . . .1 Formatted Input Strings . . . . . . . . . . . . . . . . . . . . . . . 67 10. . . .3 7. . . . . . . . . . . . . . . . 56 8. . . .2 Random State Seeding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 6. . . . . . . . . . . . . . . . . . . .15 Miscellaneous Functions . . . 66 10 Formatted Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 12. . . . . . .1 Random State Initialization . . . . . . . . 74 11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conversion Functions . . . . . . . . . . . 65 9. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 45 45 46 46 47 7 Floatingpoint Functions . .2 12. . . Integers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 7. . . . . .5 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Miscellaneous Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rationals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Assignment Functions . . . . . . . . . . . . . 72 11. . . . . . . . . . . . . .3 Random State Miscellaneous . . . . . . . . Input and Output Functions . 76 C++ C++ C++ C++ C++ C++ Interface Interface Interface Interface Interface Interface General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 13 Berkeley MP Compatible Functions . . . . . . . . . . . . . . .1 Nails . . . . . . . . . . . . . . . . . . . . . .0. . . . . . . . . . . .2 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 70 11 Formatted Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Input and Output Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 7. . . . . . . .2 7. . . . . . . . . . . . . . . . . . . . . . . .8 Initialization Functions . . . . . . . . . . . . . . . . . . .3 C++ Formatted Output. . . . . . . . . . . .5 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 12. . . . . . . . .2 6. . . . . . . . . . . . . . . . . 44 6. . . .1 12. . . . . . . . . . . . . . . . . . . . . . . . 63 Random Number Functions . . . . . Arithmetic Functions . . . . . . . . . . . . . . . . . . . . . . Floats . . . . 42 6 Rational Number Functions . . . . . . . . . . . . . . . . . . . .14 Integer Import and Export . . . . . . . . . . . . . . . . . . . .16 Special Functions . . . . . . . . .6 7. . . . . . . . 76 77 78 80 81 82 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . Arithmetic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 . . . . . . . . . . . . . . .1 Format Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Initialization and Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Combined Initialization and Assignment Functions . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 7. . . . . . . . . . . . . Limitations . . . . . . . . . . . . . . 48 50 51 51 52 53 53 54 8 9 Lowlevel Functions . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .6 Radix Conversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . 110 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3.5 Lucas Numbers. . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8. . . . . . . . . 98 16. . . . . . . . . . . .2 Factorial . . . . . .4 Cache Handling . . . . . . . . . . . . . . . . . . . . . . 108 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8. . . . . . . . . .3 Carry Propagation . . . 107 16. . 112 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Root Extraction Algorithms . 95 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 16. . . . . . . . . . . . . . . . . . . . . . . . .3 Perfect Square . . . . . . 107 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Other Multiplication . . . . . .2. . . 92 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 SIMD Instructions . . . . . . . . . . . . . . . . . . . . 94 16. . . . . . . . . .1 Basecase Multiplication. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Divide and Conquer Division . . . . .7 Other Algorithms . . . . . . . . . . . . . . . 109 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6. 106 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . .1. . . . . . . . . .1 Square Root . . . . . . . . . . . . . . . . . . . . . . . . 108 16. . . . . . . . . . . . . . . .1. . . . 96 16. . . . 107 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Small Quotient Division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . 103 16. . . . . . . . . . . . . . . . . . . . 104 16. . . . . . . . . . . . . . . . . . . . . . . . .4 Fibonacci Numbers . . . . . . . . . . . . . . . . . . . . .8. . . . . . . . . . . . 86 Language Bindings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 Assembly Coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Binomial Coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 16. . . . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Division Algorithms . 101 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Modular Powering . . . . . . . 105 16. . . . . .5. . . . . . . . . . . . . . . . . . . .1 Normal Powering . . .2 Basecase Division . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . .8. . . . . . . .3. . . . . . . . . . .5. . . . . 96 16. .8. . . . . . . . . . . . . . . . . . . . . . . . .8. . .2 Nth Root . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Prime Testing . . . . . . . . . . . . . . 98 16. . . . .7. . . . . . . . . . . . . . . . . . . 106 16. . . . . . . . .5 Jacobi Symbol . . . . . . . . . . . . . 90 16. . . . . . . . . . . . . . 102 16. . . . . . . . .4 Perfect Power . . . . . . . . . . . 97 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Exact Remainder . . . . . . . . 105 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Algorithms . . . . . . . . . . . .7. . . 102 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . 110 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 BlockWise Barrett Division . . . . . 111 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 16. . . . . . . . . . .1 Binary to Radix . . . . . 100 16. . . . . . . . . . . . . . .3 Toom 3Way Multiplication. . . .2. . . . . . . . .1 Code Organisation . . . . .5 FFT Multiplication . . .7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . 106 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Extended GCD. . . . . . . . . . . . . . . .2 Lehmer’s algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 16. .1 Binary GCD . . . . . . . . . . . . . . . . . . . . . . .2 Radix to Binary . . . . 91 16. . . . . . . . .3 Subquadratic GCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. .7. . . . . . . . . . . . . . . . . . . .6. . . . . . . . . . . . . . . . . .1 Multiplication . . . . . . .1. . . . . . . . . .7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Powering Algorithms . . . . . . 109 16. . . . . . . . . .7. . . . . . . . . 104 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8. . . . . . . . . . . . . . 90 16. 103 16. . . . . . . . . . . . . . . . . .iii 14 15 16 Custom Allocation . . . . . .2 Karatsuba Multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Single Limb Division . . . . . . . . . . . . . . . . .6 Random Numbers . .2 Assembly Basics. . . . . . . . . . . . . . . . 101 16. . . . . . . . . . . . . . . . . . . . . . . .4 Toom 4Way Multiplication. . . . . . .7 Unbalanced Multiplication . . . . . . . . . . 96 16. . . . . . . . . . .5 Exact Division . . . . . . . . . . . . . . . . . . . .3 Greatest Common Divisor . . . . . . . . . . . . . . . . . 94 16. . . . . . . . . . . . . . . . . .1. . . . . . .3.1. .5. . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . 99 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Floating Point.5 Functional Units . . . . . . 104 16. . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 References .2 17. . . . . . . . . . . . . . . . . .1 16. . . . . . . . . . . . . . . . . . . . . . . . 122 Appendix C GNU Free Documentation License . . . . . 122 B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Integer Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .0. . . . . . . . 131 Function and Type Index . . . . . . . . . . . . . . . . . . . . . . . . . .8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 Concept Index . . . . . . . . . . . . . . . . . . . . . . . . .1 Books . . . . . . . . . . . . . . . . . . . 122 B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 17 Internals . . . . . . . . . . . . . . . . . . . . 112 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 Software Pipelining . . . . . .5 Appendix A Appendix B Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .iv GNU MP 5. . . . . . . . . . . . . .1 17. C++ Interface Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 . . . . . . . Float Internals . . . . . Raw Output Internals . . . . . . . . . . . . . . . . . . . . . .4 17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 16. . . . . . . . . . .2 Papers. . . . . . . . . . . . . . . . . . . . . . . . . .3 17.9 Loop Unrolling . . . . . . . . . . . . . . . . . . . . . . . . . .8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10 Writing Guide . . . . . . . . . . . . . . . . 115 115 116 118 118 17. . . . . . . . . . . . Rational Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
we want their recipients to know that what they have is not what we distributed. and that you know you can do these things. The precise conditions of the license for the GNU MP library are found in the Lesser General Public License version 3 that accompanies the source code. . receive or can get the source code. that you receive source code or else can get it if you want it. The library is not in the public domain. You must make sure that they. Certain demonstration programs are provided under the terms of the plain General Public License version 3. but these restrictions are designed to permit everything that a good cooperating citizen would want to do.GNU MP Copying Conditions 1 GNU MP Copying Conditions This library is free.LIB’. for our own protection. you must give the recipients all the rights that you have. so that any problems introduced by others will not reflect on our reputation. this means that everyone is free to use it and free to redistribute it on a free basis. if you distribute copies of the GNU MP library. Specifically. see ‘COPYING’. we have to forbid you to deprive anyone else of these rights. that you can change this library or use pieces of it in new free programs. Also. For example. To make sure that everyone has such rights. If it is modified by someone else and passed on. we must make certain that everyone finds out that there is no warranty for the GNU MP library. And you must tell them their rights. it is copyrighted and there are restrictions on its distribution. we want to make sure that you have the right to give away copies of the library. What is not allowed is to try to prevent others from further sharing any version of this library that they might get from you. see ‘COPYING. too.
and 21264. by including carefully optimized assembly code for the most common inner loops for many different CPUs. MC88100. GMP is designed to give good performance for both. by using sophisticated algorithms. For uptodate information on GMP. IBM POWER. generic x86. DEC Alpha 21064.gnu. . for the compiler options that must be used on applications. Many applications use just a few hundred bits of precision. There are three public mailing lists of interest. Pentium 4. generic SPARCv8. MC68020. UltraSPARC. AMD K6. although it is probably a good idea to glance through it. one for general questions and discussions about usage of the GMP library and one for bug reports. The proper place for bug reports is gmpbugs@gmplib.org’.org/gnu/gmp/ Many sites around the world mirror ‘ftp.gnu. IBM ROMP (RT). and Pyramid AP/XP. It aims to provide the fastest possible arithmetic for all applications that need higher precision than is directly supported by the basic C types.org/order/ftp. page 3. SuperSPARC.org/ The latest version of the library is available at ftp://ftp. Some optimizations also for Cray vector systems. Motorola MC68000. page 8. MIPS R3000. Pentium Pro/II/III. 1. SPARCv7. The rest of the manual can be used for later reference. If you have a system with multiple ABIs.0.0. please use a mirror near you.org/mailman/listinfo/. The speed of GMP is achieved by using fullwords as the basic arithmetic type. then read Section 2. then read Chapter 2 [Installing GMP]. 1. and by carefully keeping the overhead at a minimum.0.org.2 [ABI and ISA]. page 16. please see the GMP web pages at http://gmplib. see http://gmplib. For more information. and by a general emphasis on speed (as opposed to simplicity or elegance).html for a full list. National NS32000.gnu. Motorola/IBM PowerPC 32 and 64. Clipper. rational numbers. K62. page 28 for information about reporting bugs. There is assembly code for these CPUs: ARM. Intel IA64. Hitachi SuperH and SH2. R4000.1 How to use this Manual Everyone should read Chapter 3 [GMP Basics]. HPPA 1. and Athlon64. 21164. DEC VAX. If you need to install the library yourself. See Chapter 4 [Reporting Bugs]. AMD 29000. Athlon. by choosing algorithms based on the sizes of the operands. i960. but some applications may need thousands or even millions of bits. see http://www. One for release announcements.2 GNU MP 5. and Zilog Z8000.1 1 Introduction to GNU MP GNU MP is a portable library written in C for arbitrary precision arithmetic on integers. and MC88110. Intel Pentium.1 and 2. and floatingpoint numbers.
In particular. page 28. ‘execprefix’ can be used to direct architecturedependent files like ‘libgmp. ‘build=CPUVENDOROS’ For normal native compilation. This can be used to share architectureindependent parts like the documentation.guess’. so it will be necessary to ensure both ‘$prefix/include’ and ‘$exec_prefix/include’ are available to the compiler. See Chapter 4 [Reporting Bugs].1 Build Options All the usual autoconf configure options are available.0. page 12. but on some CPUs are slightly slower. certainly all the code is there./configure’ uses the output from running ‘. Use GNU make instead. SunOS and Slowaris make have bugs that make them unable to build in a separate directory.4 [Notes for Particular Systems]. For example cd /my/build/dir /my/sources/gmp5. By default ‘. Tools ‘configure’ requires various Unixlike tools./config. It might be possible to build without the help of ‘configure’. The file ‘INSTALL. On some ./configure make Some selftests can be run with make check And you can install (under ‘/usr/local’ by default) with make install If you experience problems. On a Unixlike system a basic build can be done with . for information on what to include in useful bug reports. Build Directory To compile in a separate build directory. please report them to gmpbugs@gmplib. cd to that directory. but one or other can be disabled. for some options on nonUnix systems. 2. but unfortunately you’ll be on your own.h’ are architecturedependent since they encode certain aspects of ‘libgmp’.org. ‘disablestatic’ By default both shared and static libraries are built (where possible). Shared libraries result in smaller executables and permit code sharing between separate running processes. ‘disableshared’.a’ to a different location. Native Compilation. run ‘.1/configure Not all ‘make’ programs have the necessary features (VPATH) to support this. having a small cost on each function call.Chapter 2: Installing GMP 3 2 Installing GMP GMP has an autoconf/automake/libtool based configuration system./configure help’ for a summary. The default is ‘/usr/local’. but separate the dependent parts.autoconf’ has some generic installation information too. See Section 2. the system can be specified with ‘build’. and prefix the configure command with the path to the GMP source directory.h’ and ‘mp. Note however that ‘gmp. ‘prefix’ and ‘execprefix’ The ‘prefix’ option can be used in the normal way to direct GMP to install under a particular tree.
In all cases the compiler must be able to produce an executable (of whatever format) from a standard C main. ‘alphaev56’. ‘hppa1. Currently a warning is given unless an explicit ‘build’ is used when crosscompiling. such as ‘m68klinux’. But note that tools don’t have to be setup this way. ‘j90’. ‘host=CPUVENDOROS’ When crosscompiling. (Some past versions of GMP used ‘target’ incorrectly.0’. which is important when creating binaries for a newer CPU since they very possibly won’t run on the build system. See ‘configure.0w’./config. ‘hppa64’ • IA64: ‘ia64’. ‘mips64’ . ‘itanium’. though very possibly this would merely be special options on a native compiler. the system used for compiling is given by ‘build’ and the system where the library will run is given by ‘host’. and this can be an alias. since it controls how libtool generates shared libraries. ‘t90’.guess’ can determine the exact CPU type. such as determining what functions are available on the host system. ‘alphaev68’ ‘alphaev7’ • Cray: ‘c90’. and might run slower on other members. ‘hppa2. For example when using a FreeBSD Athlon system to build GNU/Linux m68k binaries./configure build=ultrasparcsunsolaris2. ‘alphaev6’./configure build=athlonpcfreebsd3. older or newer. However. ‘alphapca57’./configure’ avoids depending on being able to run code on the build system. • Alpha: ‘alpha’.1 systems ‘. Ordinary programs or libraries like GMP are only interested in the ‘host’ part.) CPU types In general.in’ for details of what code and compiler options they select. it’s enough to just have a PATH with a suitable crosscompiling cc etc.guess’ is the simplest way to see what it should be. because it may not be possible to correctly guess the build system type if the PATH has only a crosscompiling cc. if you don’t know already. ‘hppa2. Cross Compilation.4 GNU MP 5. ‘.7 In all cases the ‘OS’ part is important. The prefix is the argument to ‘host’. Running ‘. ‘hppa2. . and ‘target’ what they’ll produce code for./configure’ uses linking tests for various purposes. ‘mips3’. For example m68kmaclinuxgnuranlib is tried. Compiling for a different CPU in the same family as the build system is a form of crosscompilation.5 host=m68kmaclinuxgnu Compiler tools are sought first with the host system type as a prefix./config. ‘alphapca56’. you should configure GMP for the exact CPU type your system uses. It’s for use when building compiler tools. ‘alphaev67’. being where they’ll run. Although only object files will go to make up ‘libgmp’.0n’. Note that the ‘target’ option is not appropriate for GMP. ‘sv1’ • HPPA: ‘hppa1. The following CPUs have specific support. This makes it possible for a set of crosscompiling tools to coexist with native tools. The best idea is always to build GMP for the exact machine type you intend to run it on. then plain ranlib. In any case ‘.0. For example. this may mean the binaries won’t run on older members of the family. with ‘host’ being where they will run. ‘alphaev5’. ‘itanium2’ • MIPS: ‘mips’. on others it will be necessary to give it explicitly. if you want a library that runs as fast as possible. .1’.0’.
‘k62’. For example . with gcc normally preferred if it’s present. ‘ns32k’. ‘powerpc620’. ‘i586’. ‘ultrasparc2’. The usual ‘CFLAGS="whatever"’ can be passed to ‘. ‘powerpc970’ • SPARC: ‘sparc’./configure host=noneunknownfreebsd3. ‘pentium2’. ‘enablefat’ Using ‘enablefat’ selects a “fat binary” build on x86. ‘ultrasparc2i’. . the generic C code can be selected with CPU ‘none’. but it should be portable and should at least make it possible to get something running if all else fails. ‘powerpc603e’. ‘powerpc823’. ‘powerpc7400’. ‘m88k’. ‘arm’. ‘viac32’ • Other: ‘a29k’. ‘powerpc750’. ‘m68010’. ‘viac3’. ‘vax’. For various systems. ‘ultrasparc’. ‘ultrasparc3’. with default flags (and default ABI). ‘k6’. default compiler flags are set based on the CPU and compiler. meaning data type sizes and calling conventions./configure’ to choose something different.2 [ABI and ISA]. ‘powerpc401’.Chapter 2: Installing GMP 5 • Motorola: ‘m68k’. page 8. ‘powerpc604e’. ‘powerpc630’. ‘microsparc’. This is the easiest way to check the defaults when considering changing or adding something. ‘m88110’ • POWER: ‘power’. Generic C Build If some of the assembly code causes problems. ‘pentiumpro’. The ‘CC’ and ‘CFLAGS’ used are printed during ‘. for the available choices on relevant CPUs. ‘powerpc603’. ‘powerpc801’. ‘m68000’./configure’ to use something different or to set good flags for systems GMP doesn’t otherwise know.) ‘ABI’ On some systems GMP supports multiple ABIs (application binary interfaces). but gives good performance on all x86 chips. ‘powerpc505’. but a particular ABI can be selected. For example. The usual ‘CC=whatever’ can be passed to ‘. ‘sh’. ‘sparcv8’. ‘power2sc’ • PowerPC: ‘powerpc’. ‘sparcv9’. where optimized low level subroutines are chosen at runtime according to the CPU detected. ‘CFLAGS’ By default the C compiler used is chosen from among some likely candidates. ‘powerpc860’. ‘pyramid’. ‘m68020’. ‘m68060’./configure’./configure host=mips64sgiirix6 ABI=n32 See Section 2. ‘m68360’. ‘m68302’. ‘power1’. (This option might become available for more architectures in the future. ‘sparc64’ • x86 family: ‘i386’. ‘k63’. ‘i486’. and what applications need to do. ‘powerpc604’. . ‘powerpc64’. ‘CC’. and can be found in each generated ‘Makefile’. ‘athlon’. ‘amd64’. By default GMP chooses the best ABI available. or if otherwise desired. ‘supersparc’. ‘powerpc403’. ‘powerpc740’. For example ‘CC=gcc’ can be used to force the use of GCC. ‘powerpc7450’. since GMP can’t determine the ABI just from the flags and won’t be able to select the correct assembly code. ‘powerpc602’. ‘sh2’. ‘z8k’ CPUs not listed will use generic C code. If just ‘CC’ is selected then normal default ‘CFLAGS’ for that compiler will be used (if GMP recognises it). ‘pentium4’. ‘i960’. ‘m68040’. ‘pentium3’. ‘pentium’. Fat binary. This means more code. ‘pentiummmx’. ‘powerpc821’. ‘m68030’.5 Note that this will run quite slowly. ‘power2’. Note that when ‘CC’ and ‘CFLAGS’ are specified on a system supporting multiple ABIs it’s important to give an explicit ‘ABI=whatever’. ‘powerpc405’. ‘powerpc601’. ‘clipper’.
la’ and header file ‘gmpxx. It’s important that the C and C++ compilers match. ‘CFLAGS’ without ‘g’. so for that reason ‘disablecxx’ is the default. The default for ‘CXXFLAGS’ is to try ‘CFLAGS’.la’ from the same GMP version. meaning their startup and runtime support routines are compatible and that they generate code in the same ABI (if there’s a choice of ABIs on the system). The default is to try the selected ‘CC’ and some likely candidates such as ‘cc’ and ‘gcc’. and in the ‘ansi2knr’ support for K&R compilers. to avoid a build failure due to a compiler mismatch. so a mismatch will cause unresolved symbols rather than perhaps mysterious misbehaviour. Incidentally.0. ‘CXXFLAGS’ When C++ support is enabled. The C++ support consists of a library ‘libgmpxx. . the C++ compiler and its flags can be set with variables ‘CXX’ and ‘CXXFLAGS’ in the usual way. since name mangling and runtime support are usually incompatible between different compilers. and to avoid any chance that the C++ compiler could be required when linking plain C programs.1 ‘CPPFLAGS’ Any flags like ‘D’ defines or ‘I’ includes required by the preprocessor should be set in ‘CPPFLAGS’ rather than ‘CFLAGS’. Future changes to the relevant internals will be accompanied by renaming. ‘CC_FOR_BUILD’ Some buildtime programs are compiled and run to generate hostspecific data tables.la’ in order to ensure dynamic linked C programs aren’t bloated by a dependency on the C++ standard library. it’s normally not good enough to set ‘CXX’ to the same as ‘CC’. If some particular options are required they can be included as for instance ‘CC_FOR_BUILD="cc whatever"’.1 [Headers and Libraries]. ‘.la’ and can only be expected to work with ‘libgmp. Compiling is done with both ‘CPPFLAGS’ and ‘CFLAGS’. then for g++ either ‘g O2’ or ‘O2’.la’ has been adopted rather than having C++ objects within ‘libgmp.c’ should be enough. C++ Support. only g++ will invoke the linker the right way when building an executable or shared library from C++ object files. or for other compilers ‘g’ or nothing. looking for something that works. As a convenience ‘enablecxx=detect’ can be used to enable C++ support only if a compiler can be found. page 16). in which case a C++ compiler will be required. ‘enablecxx’ C++ support in GMP can be enabled with ‘enablecxx’. Although gcc for instance recognises ‘foo.la’ will use certain internals from ‘libgmp. Trying ‘CFLAGS’ this way is convenient when using ‘gcc’ and ‘g++’ together.6 GNU MP 5. In general ‘libgmpxx. it merely needs to generate executables that can run.la’ will be usable only with the C++ compiler that built it. ‘CC_FOR_BUILD’ is the compiler used for this. The default for ‘CXX’ is the first compiler that works from a list of likely candidates. ‘CXX’. but preprocessing uses just ‘CPPFLAGS’. with g++ normally preferred when available. ‘libgmpxx./configure’ isn’t currently able to check these things very well itself. A separate ‘libgmpxx. since the flags for ‘gcc’ will usually suit ‘g++’.h’ (see Section 3. This distinction is because most preprocessors won’t accept all the flags the compiler does. Preprocessing is done separately in some configure tests. No flags are used with ‘CC_FOR_BUILD’ because a simple invocation like ‘cc foo. It doesn’t need to be in any particular ABI or mode. Perhaps this will change in the future.cc’ as C++ code.
See Chapter 13 [BSD Compatible Functions]. This can be of use while debugging.h’) are built and installed only if ‘enablempbsd’ is used. Knowledgeable users with special requirements can specify a different path. these being malloc and friends by default. see Texinfo. For example ‘sparcv8’ has MPN_PATH="sparc32/v8 sparc32 generic" which means look first for v8 code. page 84. Assertion Checking. a search is made though a path to choose a version of each. in Texinfo format. ‘MPN_PATH’ Various assembly versions of each mpn subroutines are provided. ‘enablealloca=<choice>’ GMP allocates temporary workspace using one of the following three methods. • ‘alloca’ . then plain sparc32 (which is v7). Berkeley MP.the heap. larger ones use mallocreentrant. FFT Multiplication. ‘enableprofiling=prof/gprof/instrument’ Enable profiling support. An additional choice ‘enablealloca=debug’ is available. • ‘notreentrant’ .alloca if available.a synonym for ‘alloca’.12 [Debugging]. ‘enableassert’ This option enables some consistency checking within the library. The FFT is only used on large to very large operands and can be disabled to save code size if desired.a synonym for ‘mallocreentrant’.13 [Profiling]. • ‘mallocnotreentrant’ . • ‘yes’ . page 23.alloca if available. and finally fall back on generic C. otherwise ‘mallocreentrant’. • ‘reentrant’ .C library or compiler builtin. ‘disablefft’ By default multiplications are done using Karatsuba. see Section 3. reentrant and thread safe. see Section 3. the following choices are also available.Chapter 2: Installing GMP 7 Temporary Memory. otherwise ‘mallocnotreentrant’. For convenience. • ‘no’ . and is recommended. page 25. but ‘mallocnotreentrant’ is faster and should be used if reentrancy is not required. as the name suggests.the heap. For a given CPU. ‘mallocreentrant’ is. Normally this is completely unnecessary. This is the default. • ‘mallocreentrant’ . ‘enablempbsd’ The Berkeley MP compatibility library (‘libmp’) and header file (‘mp. with global variables. ‘disablealloca’ is the same as ‘no’. alloca is reentrant and fast.texi’. It actually allocates just small blocks on the stack. Documentation The source for the document you’re now reading is ‘doc/gmp. See Chapter 14 [Custom Allocation]. The two malloc methods in fact use the memory allocation functions selected by mp_ set_memory_functions. and Fermat FFT. which can be selected with for instance ‘enablealloca=mallocreentrant’. to help when debugging memory related problems (see Section 3.12 [Debugging]. in one of various styles. page 23). 3way Toom. page 86. Execution Profiling. in a reentrant fashion. .
Applications must be compiled with gcc m32 (In GCC 2. and is not recommended except for interoperating with other code not yet 64bit capable. and this generally gives significantly greater speed. the following ABI choices are available.h’. This is convenient for applications.0. ISA (Instruction Set Architecture) refers to the instructions and registers a CPU has available. This will probably only matter when installing multiple builds of GMP.info’ is included in the distribution. DocBook and XML can be generated by the Texinfo makeinfo program too. This is the default. PDF and HTML (these will require various TEX and Texinfo tools). 2. The usual automake targets are available to make PostScript. since all compilers for a particular ABI will be expected to use the same limb type. When a long long limb is used this is encoded in the generated ‘gmp. the latter for compatibility with older CPUs in the family.2 ABI and ISA ABI (Application Binary Interface) refers to the calling conventions between functions./configure ABI=32 In all cases it’s vital that all object code used in a given program is compiled for the same ABI.1 Info format ‘doc/gmp. AMD64 (‘x86_64’) On AMD64 systems supporting both 32bit and 64bit modes for applications. ‘gmp. or it might require more than that. ‘ABI=64’ The 64bit ABI uses 64bit limbs and pointers and makes full use of the chip architecture./configure and make each. see Section “Options for makeinfo” in Texinfo. GMP may support a limb as either a 32bit long or a 64bit long long. But an ABI can be chosen explicitly to make GMP compatible with other libraries. but for reference the option is gcc ‘ABI=32’ m64 The 32bit ABI is the usual i386 conventions. it’s the only mode. For example in some 32bit ABIs. and it might be as simple as configuring with a special ‘libdir’. or particular application requirements. This will be slower. Some supplementary notes can also be found in the ‘doc’ subdirectory.) . plus how GMP chooses to use it.h’ will vary. and can’t be just copied around. Currently no attempt is made to follow whatever conventions a system has for installing library or header files built for a particular ABI. For example. Usually a limb is implemented as a long. DVI. GMP supports some CPUs like this in both ABIs.95 and earlier there’s no ‘m32’ option. Note that builds for different ABIs need to done separately. meaning what registers are used and what sizes the various C data types are. .8 GNU MP 5. By default GMP chooses the best ABI available for a given system. with a fresh . Applications will usually not need special compiler flags.h’ remains compiler independent though. but it does mean that ‘gmp. In fact within GMP ‘ABI’ means a combination of chip ABI. Some 64bit ISA CPUs have both a 64bit ABI and a 32bit ABI defined.
pointers.0n’ The 2.0w’ The 2.0 +e Note that current versions of GCC (eg. No special compiler options are needed for applications.0’ HPPA 2.0n and 2. MIPS under IRIX 6 (‘mips**irix[6789]’) IRIX 6 always has a 64bit MIPS 3 or better CPU. and no 64bit operations. Applications must be compiled with gcc [built for 2.0 (‘hppa2. Instead it must be built for one or the other ABI. and GMP performance will be the same in each. This ABI is available on hppa64 GNU/Linux and on HPUX 10 or higher. ‘hppa64’) ‘ABI=2. Applications must be compiled with gcc mlp64 cc +DD64 ‘ABI=64’ On other IA64 systems.0w.0n ABI means the 32bit HPPA 1.0’ and ‘hppa64’. but for CPU type ‘hppa2.Chapter 2: Installing GMP 9 HPPA 2. n32 or 64 are recommended. ‘ABI=o32’ The o32 ABI is 32bit pointers and integers. but for reference the flags are gcc milp32 cc +DD32 In the 64bit ABI. and skip to the corresponding ‘ABI’. ints and longs are 32 bits and GMP uses a 64 bit long long for a limb. eg. Note that GCC on HPUX has no options to choose between 2.0n’ only 2. but with 64bit instructions permitted within functions. Applications can be compiled with no special flags on an old compiler.0n] cc +DA2.0 CPUs can run all HPPA 1.0w] cc +DD64 ‘ABI=2.2) don’t generate 64bit instructions for long long operations and so may be slower than for 2.0 ABI and all its normal calling conventions. ‘hppa2. (The GMP assembly code is the same though. GMP uses a 64bit long long for a limb.) ‘ABI=1.0w’.0 and 1. and 64. Applications can be compiled without any special flags since this ABI is the default in both HP C and GCC. and supports ABIs o32.0n or 1. unlike HP cc.2. 3. GNU/Linux for instance.7. ‘itanium**hpux*’) HPUX supports two ABIs for IA64. All three ABIs are available for CPU types ‘hppa2.0w modes. IA64 under HPUX (‘ia64**hpux*’.0 are considered. or on a newer compiler with .0w ABI uses 64bit limbs and pointers and is available on HPUX 11 or up. GMP performance is the same in both. GMP will detect how it was built. this option only exists to support old compilers. GMP will be slower than in n32 or 64. GCC 2.1 code in the 32bit HPPA 1. ‘ABI=64’ is the only choice.0*’. The default is n32. ‘ABI=32’ In the 32bit ABI.0 ABI. n32. Applications must be compiled with gcc [built for 2. longs and pointers are 64 bits and GMP uses a long for a limb.
‘powerpc630’. depending on the default gcc mode. ‘ultrasparc*’) ‘ABI=64’ The 64bit V9 ABI is available on the various BSD sparc64 ports.0. Applications must be compiled with gcc m64 ‘ABI=mode32’ The ‘mode32’ ABI uses a 64bit long long limb but with the chip still in 32bit mode and using 32bit calling conventions. On a suitable system we could perhaps use more of the ISA.7 and up (when the kernel is in 64bit mode).2. This is the default on for systems where the true 64bit ABIs are unavailable. but there are no plans to do so. No special compiler options are needed for applications. ‘powerpc970’. ‘power5’) ‘ABI=aix64’ The AIX 64 ABI uses 64bit limbs and pointers and is the default on PowerPC 64 ‘**aix*’ systems. ‘sparcv9’. recent versions of Sparc64 GNU/Linux.2 or higher. ‘powerpc620’. Applications must be compiled with gcc mabi=n32 cc n32 The 64bit ABI is 64bit pointers and integers.1 gcc cc ‘ABI=n32’ mabi=32 32 The n32 ABI is 32bit pointers and integers. doesn’t have the necessary support for n32 or 64 and so only gets a 32bit limb and the MIPS 2 code. ‘power4’. and Mac OS X/Darwin systems. GCC 3. or Sun cc is required. but with a 64bit limb using a long long. PowerPC 64 (‘powerpc64’. with a 32bit limb. ‘ABI=32’ This is the basic 32bit PowerPC ABI. . On GNU/Linux. as of kernel version 2. since 64bits is the only ABI available. Sparc V9 (‘sparc64’. Applications must be compiled with gcc mabi=64 cc 64 ‘ABI=64’ Note that MIPS GNU/Linux. Applications must be compiled with gcc maix64 xlc q64 ‘ABI=mode64’ The ‘mode64’ ABI uses 64bit limbs and pointers. GMP speed is greatest in ‘aix64’ and ‘mode32’.xarch=v9 mcpu=v9 cc xarch=v9 On the BSD sparc64 systems no special options are required. In ‘ABI=32’ only the 32bit ISA is used and this doesn’t make full use of a 64bit chip. and Solaris 2. applications must be compiled with gcc m64 On Solaris applications must be compiled with gcc m64 mptr64 Wa. No special compiler options are needed for applications.10 GNU MP 5. and is the default on 64bit GNU/Linux. BSD.
see Section 2.guess’ will detect the CPU. applications may need to be compiled with gcc m32 On Solaris.) gcc mv8plus cc xarch=v8plus GMP speed is greatest in ‘ABI=64’. though using something like the following is recommended. Libtool is used to build the library and ‘versioninfo’ is set appropriately.6 and earlier. This is a good choice for packaging to run on a range of x86 chips.0 (see Section “Library interface versions” in GNU Libtool). depending on the default gcc mode. For example this might mean plain ‘sparc’ (meaning V7) for SPARCs./config.guess’ is inexact.8 and earlier only support ‘mv8’ though. or perhaps the loader.Chapter 2: Installing GMP 11 ‘ABI=32’ For the basic 32bit ABI. since this is not done by the libtool versioning.la’ (from ‘enablecxx’.3 Notes for Package Builds GMP should present no great difficulties for packaging in a binary distribution. having started from ‘3:0:0’ in GMP 3. On Solaris 2. no special compiler options are required for applications. In the Sun documentation this combination is known as “v8plus”.1 [Build Options]. A mismatch will result in unresolved symbols from the linker./config. ‘ABI=64’ can still be built if desired by making it look like a crosscompile. Providing a way to suitably rebuild a package may be useful. Additional function interfaces may be added in each release. The GMP 4 series will be upwardly binary compatible in each release and will be upwardly binary compatible with all of the GMP 3 series. For x86s. The speed is partly because there are extra registers available and partly because 64bits is considered the more important case and has therefore had better code written for it. nor otherwise. But a way to manually specify a ‘build’ will be wanted for systems where ‘. they’re called ‘arch’ but effectively control both ABI and ISA. page 3) requires ‘libgmp. Users who care about speed will want GMP built for their exact CPU type. GMP still uses as much of the V9 ISA as it can. On GNU/Linux. This could be as simple as making it possible for a user to omit ‘build’ (and ‘host’) so ‘. so it’s the default where available. care should be taken to use ‘host’ (or ‘build’) to choose the least common denominator among the CPUs which might use the package. to make best use of the available optimizations. making a runtime selection of optimized low level routines. so on systems where libtool versioning is not fully checked by the loader an auxiliary mechanism may be needed to express that a dynamic linked application depends on a new enough GMP. Don’t be confused by the names of the ‘m’ and ‘x’ compiler options. for example .7 ABI=64 2. An auxiliary mechanism may also be needed to express that ‘libgmpxx./configure build=none host=sparcv9sunsolaris2. . ‘enablefat’ sets things up for a fat binary build.la’ from the same GMP version. (gcc 2. a normal native build will reject ‘ABI=64’ because the resulting executables won’t run. only ‘ABI=32’ is available since the kernel doesn’t save all registers. When building a package for a CPU family. On Solaris 2.7 with the kernel in 32bit mode.
versions of GCC up to and including 2. but unfortunately this is done even if a fully functional ld is available./configure’).cygwin.4 or later. Compaq C++ Compaq C++ on OSF 5.com/ http://www. A given run of ‘./configure enablecxx CPPFLAGS=D__USE_STD_IOSTREAM Floating Point Mode On some systems. MSDOS and MS Windows On an MSDOS system DJGPP can be used to build GMP. including application code. currently no attempt is made to follow system conventions for install locations that vary with ABI.com/djgpp/ http://www. starting from a clean directory tree (‘make distclean’). a packaged build will need to decide which among the choices is to be provided.so’. If compiler include paths don’t vary with ABI options then it might be necessary to create a ‘/usr/include/gmp.0. A shared build can be attempted with . page 8. not just GMP. Of course this affects all code. apparently for the benefit of old versions of ld which only recognise ‘. The GMP functions involving a double cannot be expected to operate to their full precision when the hardware is in single precision mode.1 has two flavours of iostream.95.3 have a bug in unsigned division.delorie.4 Notes for Particular Systems AIX 3 and 4 On systems ‘**aix[34]*’ shared libraries are disabled by default. GMP ‘. When attempting to install two ABIs simultaneously it will be important that an application compile gets the correct ‘gmp. see Section 2. If a second ABI is also required then a second run of ‘.2 [ABI and ISA]. such as ‘/usr/lib/sparcv9’ for ‘ABI=64’ as opposed to ‘/usr/lib’ for ‘ABI=32’. Note that ‘gmp. All three are excellent ports of GCC and the various GNU tools. since some versions of the native ar fail on the convenience libraries used. double or extended on x86 systems (x87 floating point).h’. and on an MS Windows system Cygwin. .org/ Microsoft also publishes an Interix “Services for Unix” which can be used to build GMP on Windows (with a normal ‘. a standard one and an old prestandard one (see ‘man iostream_intro’). the hardware floating point has a control mode which can set all operations to be done in a particular precision.12 GNU MP 5. A package build can override ‘libdir’ and other standard variables as necessary. DJGPP and MINGW can be used.h’ is a generated file. but it’s not free software.a’ a symlink to ‘libgmp.h’ which tests preprocessor symbols and chooses the correct actual ‘gmp. for instance single. http://www. Configure with for instance . 2. As noted under “ABI and ISA”./configure’ etc will only build one ABI.mingw.1 On systems with multiple ABIs. and will be architecture and ABI dependent./configure’ will demand GCC 2./configure enableshared disablestatic Note that the ‘disablestatic’ is necessary because in a shared build libtool makes ‘libgmp.95. giving wrong results for some operands. which unfortunately is not the default but must be selected by defining __USE_STD_ IOSTREAM.h’ for its desired ABI. GMP can only use the standard one./configure’ etc must be made.a’. ARM On systems ‘arm***’.
Chapter 2: Installing GMP 13 MS Windows DLLs On systems ‘**cygwin*’. OpenBSD 2.lib MINGW uses the C runtime library ‘msvcrt. A MINGW DLL build of GMP can be used with Microsoft C. but a DLL can be built instead using . which we believe is always available (if not then use GNU m4). Power CPU Types In GMP. ‘m68020’ or higher will give a performance boost on applicable CPUs. though this is merely a synonym for ‘m68000’. since certain export directives in ‘gmp.6 m4 in this release of OpenBSD has a bug in eval that makes it unsuitable for ‘. SunOS 4 /usr/bin/m4 lacks various features needed to process ‘. In these cases the only suggestion currently is to build GMP with CPU ‘none’ to avoid the assembly code.dll’ for I/O. Currently GMP has no assembly code support for using just the common instruction subset. and copied to the install directory./configure’ will detect the problem and either abort or choose another m4 in the PATH. but it can be created with MS lib as follows. To get executables that run on both.lib’ format import library. Sparc App Regs The GMP assembly code for both 32bit and 64bit Sparc clobbers the “application registers” g2.sub’. the current suggestion is to use the generic C code (CPU ‘none’).h’ must be different. and instead ‘. If one of the other C runtime library choices provided by MS C is desired then the suggestion is to use the GMP string functions and confine I/O to the application.libs lib /def:libgmp3. ‘**mingw*’ and ‘**pw32*’ by default GMP builds only a static library. the same way that the GCC default ‘mappregs’ does (see Section “SPARC Options” in Using the GNU Compiler Collection (GCC)). so it’s important to choose the right one for the CPU that will be used./configure’ will automatically use /usr/5bin/m4.asm’ files. ‘m68360’ can be used for CPU32 series chips. Libtool doesn’t install a ‘.asm’ file processing. ‘m68302’ can be used for “Dragonball” series chips. but is currently equivalent to ‘none’. CPU ‘rs6000’ (which is not a CPU but a family of workstations) is accepted by ‘config. This makes that code unsuitable for use with the special V9 ‘mcmodel=embmedany’ (which uses g4 as a data segment pointer). g3 and g4. The bug is fixed in OpenBSD 2. and for applications wanting to use those registers for special purposes. Similarly for ‘libmp’ and ‘libgmpxx’./configure disablestatic enableshared Static and DLL libraries can’t both be built. Motorola 68k CPU Types ‘m68k’ is taken to mean 68000.dll. ‘. so either upgrade or use GNU m4. so applications wanting to use the GMP I/O routines must be compiled with ‘cl /MD’ to do the same. cd .def /out:libgmp3.7. Sparc CPU Types ‘sparcv8’ or ‘supersparc’ on relevant systems will give a significant performance increase over the V7 code selected by plain ‘sparc’. . possibly with appropriate compiler options (like ‘mcpu=common’ for gcc). CPU types ‘power*’ and ‘powerpc*’ will each use instructions not available on the other.
2.la: $(libgmp_la_OBJECTS) $(libgmp_la_DEPENDENCIES) Either use GNU Make. and for instance configure with . since the MMX code that’s present is there because it’s faster than the corresponding plain integer code. but if any recompiling is done ‘libgmp.12 strip from GNU binutils 2.5 the native make is unable to handle the long dependencies list for ‘libgmp.12 or higher can be used successfully. 2. or as a workaround remove $(libgmp_la_DEPENDENCIES) from that line (which will make the initial build work. like the three versions of ‘init.14 GNU MP 5.0. GNU binutils strip prior to 2.3 that comes with FreeBSD 2.o’ in ‘libgmp. in particular version 1.so’ and ‘libmp. A staticonly build should work though (‘disableshared’). perhaps recursing into the various subdirectories uses up memory. The symptom is a “syntax error” on the following line of the toplevel ‘Makefile’. Old versions of ‘gas’ don’t support MMX instructions. it exits silently. ‘pentium’ or ‘pentiummmx’ code is good for its intended P5 Pentium chips.a’ since it will discard all but the last of multiple archive members with the same name.92. PIII). The shared libraries ‘libgmp. but quite slow when run on Intel P6 class chips (PPro.5 Known Build Problems You might find more uptodate information at http://gmplib. ‘make all’ was found to run out of memory during the final ‘libgmp. a warning is given and nonMMX code is used instead. This will be an inferior build.04 or higher. Running ‘make libgmp. ‘i386’ is a better choice when making binaries that must run on both. libgmp. not a plain GCC. Compiler link options The version of libtool currently in use rather aggressively strips compiler options when linking a shared library.a’ and ‘libmp.la’ might not be rebuilt).la’.7 as generate incorrect object code for register to register movq instructions.6 and 2. make syntax error On certain versions of SCO OpenServer 5 and IRIX 6. The same applies to SSE2.so’ are not affected by this and any version of strip can be used on them.1 x86 CPU Types ‘i586’. and so can’t be used for MMX code. x86 MMX and SSE2 Code If the CPU selected has MMX code but the assembler doesn’t support it.11 and earlier should not be used on the static libraries ‘libgmp. Binutils 2./configure CC=gccwithmyoptions DJGPP (‘**msdosdjgpp*’) The DJGPP port of bash 2. Use bash 2. but for now if this is a problem the suggestion is to create a little script to hide them. This will hopefully be relaxed in the future.a’. despite having 64Mb available.la’ directly helped.8 or the more recent OpenBSD 3.log’. having died writing a preamble to ‘config.03 is unable to run the ‘configure’ script. .1 doesn’t. Install a recent gas if MMX code is wanted on these systems. Solaris 2. PII.org/.la’ link on one system tested. MacOS X (‘**darwin*’) Libtool currently only knows how to create shared libraries on MacOS X using the native cc (which is a modified GCC).
which instructs the program how long to check FFT multiply parameters. and install that. ‘disableshared’ is recommended. building and running the tuneup program in the ‘tune’ subdirectory.2 (and 2. This compiler cannot be used to build GMP. and applications demanding extremely large numbers.2 in ‘ABI=32’ A shared library build of GMP seems to fail in this combination. If you’re going to use GMP for extremely large numbers. 2.3 The system compiler on old versions of NeXT was a massacred and old GCC.3 of their system. ‘f NNN’.6 The system sed prints an error “Output line too long” when libtool builds ‘libgmp.6.. since GMP uses assembly language for the most critical operation. even if it called itself ‘cc’.h’ parameter file.1 (or later).7. Solaris 2. see Section 2. Sparc Solaris 2. you need to get a real GCC.) POWER and PowerPC Bugs in GCC 2. (NeXT may have fixed this in release 3. but GNU sed is recommended.la’. since the latter has serious bugs. you may want to run tuneup with a large NNN value. Then recompile from scratch. Unlike what is the case for most other programs. The exact cause is unknown. cd tune make tuneup .7.3) mean it can’t be used to compile GMP on POWER or PowerPC.Chapter 2: Installing GMP 15 NeXT prior to 3.6 Performance optimization For optimal performance./tuneup will generate better contents for the ‘gmpmparam. the compiler typically doesn’t matter much.95. it builds but then fails the tests. put the output in the file file indicated in the ‘Parameters for . If you want to use GCC for these machines. Sequent Symmetry Use the GNU assembler instead of the system assembler. To use the results. to avoid any doubt.2. can be important. This doesn’t seem to cause any obvious ill effects.7 with gcc 2..’ header. build GMP for the exact CPU type of the target computer.1 [Build Options]. page 3. For example. In particular for longrunning GMP applications. apparently due to some incorrect data relocations within gmp_ randinit_lc_2exp_size. get GCC 2. . The tuneup program takes one useful parameter.
see GNU Libtool .h> for prototypes with struct obstack parameters. macros. struct foo { mpz_t x. not documented in this manual is strongly discouraged. If you do so your application is guaranteed to be incompatible with future versions of GMP. For example: mpq_t quotient.1 Headers and Libraries All declarations needed to use GMP are collected in the include file ‘gmp. g++ mycxxprog.h> Likewise <stdarg. when available. On a typical Unixlike system this can be done with ‘lgmp’.16 GNU MP 5. 3. y.h> is included too. For example. #include <gmp. For example: mpf_t fp.h> Note however that prototypes for GMP functions with FILE * parameters are only provided if <stdio. And <obstack.1 3 GMP Basics Using functions. This is built and installed if C++ support has been enabled (see Section 2. .cc lgmpxx lgmp GMP is built using Libtool and an application can use that to link if desired. All programs using GMP must link against the ‘libgmp’ library. Floating point number or Float for short. page 3). integer usually means a multiple precision integer. The C data type for such integers is mpz_t. for example gcc myprogram. such as gmp_vprintf. data types. 3. If GMP has been installed to a nonstandard location then it may be necessary to use ‘I’ and ‘L’ compiler options to point to the right directories. Rational number means a multiple precision fraction. It is designed to work with both C and C++ compilers. }.2 Nomenclature and Types In this manual. The C data type for such objects is mpf_t.h> (or <varargs. mpz_t vec[20]. as defined by the GMP library.h’. Here are some examples of how to declare such integers: mpz_t sum. etc. The C data type for these fractions is mpq_t. and some sort of runtime path for a shared library.c lgmp GMP C++ functions are in a separate ‘libgmpxx’ library.1 [Build Options]. such as gmp_obstack_printf. #include <stdio.0. is an arbitrary precision mantissa with a limited precision exponent.h>) is required for prototypes with va_list parameters.h> #include <gmp.
page 44) 3. Functions for rational number arithmetic. Fast lowlevel functions that operate on natural numbers. The associated type is mpq_t. but on some systems it’s an int for efficiency. The associated type is mpf_t. and on some systems it will be long long in the future. and size_t is used for byte or character counts. Currently this is always an unsigned long. The C data type for a limb is mp_limb_t. (see Chapter 5 [Integer Functions]. with names beginning with mpz_. Miscellaneous functions. These are used by the functions in the preceding groups.4 Variable Conventions GMP functions generally have output arguments before input arguments. page 56) 6.3 Function Classes There are six classes of functions in the GMP library: 1. The associated type is MINT. page 65) 3. There are about 150 functions in this class. A limb means the part of a multiprecision number that fits in a single machine word. in general mp_bitcnt_t is used for bit counts and ranges.) Normally a limb is 32 or 64 bits. There are about 40 functions in this class. and containing several digits. and mult. (see Chapter 7 [Floatingpoint Functions]. Currently this is usually a long. There are about 60 functions is this class. Random state means an algorithm selection and current state data. (see Chapter 8 [Lowlevel Functions]. Counts of limbs of a multiprecision number represented in the C type mp_size_t. (We chose this word because a limb of the human body is analogous to a digit. page 48) 4. with names beginning with mpf_. There are about 30 (hardtouse) functions in this class. page 29) 2. but on some systems it’s an int for efficiency. (see Chapter 14 [Custom Allocation]. This notation is by analogy with the assignment operator. The associated type is mpz_t. (see Chapter 6 [Rational Number Functions]. and you can also call them directly from very timecritical user programs. Currently this is normally a long. page 84) 5. Also. Functions compatible with Berkeley MP. Functions for signed integer arithmetic. page 86. only larger. These functions’ names begin with mpn_. but on some systems it will be an unsigned long long in the future . Functions for floatingpoint arithmetic. Counts of bits of a multiprecision number are represented in the C type mp_bitcnt_t. Functions for setting up custom allocation and functions for generating random numbers. but the integer functions can be used for arithmetic on the numerator and denominator separately. and see Chapter 9 [Random Number Functions]. The BSD MP compatibility functions are exceptions. For example: gmp_randstate_t rstate. with names beginning with mpq_. such as itom. The associated type is array of mp_ limb_t. (see Chapter 13 [BSD Compatible Functions]. .Chapter 3: GMP Basics 17 The floating point functions accept and return exponents in the C type mp_exp_t. 3. madd. The C data type for such objects is gmp_randstate_t. having the output arguments last.
More than one value can be returned by having more than one output parameter.. for (i = 1. mpz_init (n). } mpz_clear (n). result. . meaning if the function stores a value there it will change the original in the caller. using one of the functions for that purpose.18 GNU MP 5.. After a variable has been initialized. Parameters which are inputonly can be designated const to provoke a compiler error or warning on attempting to modify them. or at least cleared between each initialization. param. void foo (void) { mpz_t n. x. Here’s an example accepting an mpz_t parameter. i++) mpz_add_ui (result. } . mpz_mul_ui (result. See the chapters on integer functions. When a function is going to return a GMP result. int i. const mpz_t param. unsigned long n) { unsigned long i. For example. . i++) { mpz_mul (n. For example. the main function for integer multiplication. For efficiency reasons. Before you can assign to a GMP variable. only a pointer. initialize near the start of a function and clear near the end. avoid excessive initializing and clearing. for (i = 1. A variable should only be initialized once. n). In general.). Which function to use depends on the type of variable. you need to clear it out. like the library functions do.. x).).. it should designate a parameter that it sets. A return of an mpz_t etc doesn’t return the object. again like the library functions.. When you’re done with a variable.. and floatingpoint functions for details. it’s effectively a callbyreference. . mpz_fdiv_q (n. i*7). it may be assigned to any number of times. i < 100. rational number functions. and this is almost certainly not what’s wanted. you need to initialize it by calling one of the special initialization functions.5 Parameter Conventions When a GMP variable is used as a function parameter. doing a calculation. can be used to square x and put the result back in x with mpz_mul (x.1 GMP lets you use the same variable for both input and output in one call. void foo (mpz_t result. i < n. mpz_mul. } 3. and storing the result to the indicated parameter.0.
n. • mpz_random and the other old random number functions use a global random state and are hence not reentrant. Applications are advised to use gmp_randinit_default or gmp_randinit_lc_ 2exp instead. page 3. page 86. But sometimes it’s tricky to make that work. • mpf_set_default_prec and mpf_init use a global variable for the selected precision.1 [Build Options]. mpz_t and mpq_t variables never reduce their allocated space. 3. n. r). . Normally this is the best policy. Note that the actual fields in each mpz_t etc are for internal use only and should not be accessed directly by code that expects to be compatible with future GMP releases. For interest.7 Reentrancy GMP is reentrant and threadsafe. since it avoids frequent reallocation. containing only a couple of sizes. which is not thread safe. then naturally GMP is not reentrant. see Section 2. gmp_printf ("%Zd\n". in the current implementation. Applications that need to return memory to the heap at some particular point can use mpz_realloc2. but this can be changed. determined by the chosen precision and allocated at initialization. and in the C++ interface an explicit precision to the mpf_class constructor. "123456". Once a variable is initialized. mpf_ init2 can be used instead. the GMP types mpz_t etc are implemented as oneelement arrays of certain structures. and pointers to allocated data. with some exceptions: • If configured with ‘enablealloca=mallocnotreentrant’ (or with ‘enablealloca=notreentrant’ when alloca is not available). mpz_init (r). see Chapter 14 [Custom Allocation].Chapter 3: GMP Basics 19 int main (void) { mpz_t r. 3. but then using it as a parameter passes a pointer to the object. just like the library functions. 20L). This is why declaring a variable creates an object with the fields GMP needs. The newer random number functions that accept a gmp_randstate_t parameter can be used instead.6 Memory Management The GMP types like mpz_t are small. foo (r. and an application might not want to bother supporting that sort of thing. All memory is allocated using malloc and friends by default. Temporary memory on the stack is also used (via alloca). or clear variables no longer needed. } foo works even if the mainline passes the same variable for param and result. • gmp_randinit (obsolete) returns an error indication through a global variable. GMP takes care of all space allocation. mpf_t variables. mpz_init_set_str (n. 0). use a fixed amount of space. Additional space is allocated whenever a variable doesn’t have enough. return 0. so their size doesn’t change. but this can be changed at buildtime if desired.
k. [Global Constant] The GMP version number. as strings.x versions.1 • mp_set_memory_functions uses global variables to store the selected memory allocation functions. For GMP i.10 Demonstration programs The ‘demos’ subdirectory has some sample programs using GMP. since this involves an update of that variable. For GMP i. as integers. These aren’t built or installed.j. 3. nor for two threads to write simultaneously. For instance. • If the standard I/O functions such as fwrite are not reentrant then the GMP I/O functions using them will not be reentrant either. Please see the GMP 2 manual for details.8 Useful Macros and Constants const int mp_bits_per_limb The number of bits per limb. then GMP will not be reentrant either.0.x versions. 3. and patch level.j. but it’s not safe for one to read while the another might be writing.x style. used when compiling GMP.9 Compatibility with older versions This version of GMP is upwardly binary compatible with all 4.1. and k.0.x and 3. as a nullterminated string. • If the memory allocation functions set by a call to mp_set_memory_functions (or malloc and friends by default) are not reentrant. and upwardly compatible at the source level with all 2. • mpn_gcd had its source arguments swapped as of GMP 3.1 reverted to the 2. respectively. for consistency with other mpn functions. Note that the format “i. j. respectively. . [Macro] [Macro] The compiler and compiler flags. in the form “i.j” was used when k was zero was used before version 4.0. j. respectively. [Global Constant] __GNU_MP_VERSION __GNU_MP_VERSION_MINOR __GNU_MP_VERSION_PATCHLEVEL [Macro] [Macro] [Macro] The major and minor GMP version. This release is "5. The Berkeley MP compatibility library (see Chapter 13 [BSD Compatible Functions].0. these numbers will be i. these numbers will be i.20 GNU MP 5. with the following exceptions. but there’s a ‘Makefile’ with rules for them. and 0. It’s not safe for two threads to generate a random number from the same gmp_randstate_t simultaneously. const char * const gmp_version __GMP_CC __GMP_CFLAGS 3.k”. page 84) is source and binary compatible with the standard ‘libmp’.0.j. • It’s safe for two threads to read from the same GMP variable simultaneously. but in 3. There are a number of compatibility issues between GMP 1 and GMP 2 that of course also apply when porting applications from GMP 1 to GMP 4.3. • mpf_get_prec counted precision slightly differently in GMP 3. respectively.1".0 and 3.
For many programs this will be insignificant. the static ‘libgmp. page 88. 3. looping. especially in comparison to otherwise fast operations like addition.Chapter 3: GMP Basics 21 make pexpr . See Chapter 15 [Language Bindings].so’ will have a small overhead on each function call and global data address. Documentation is in POD format in ‘demos/perl/GMP. See ‘demos/perl/INSTALL’ for more information. This is unavoidable in a general purpose variable precision library. which are considered outside the scope of GMP (much closer to language interpreters or compilers. fixnums for control variables.1 [Initializing Integers]. As an aside. although GMP attempts to be as efficient as it can on both large and small operands. but for long calculations there’s a gain to be had. • The ‘expr’ subdirectory is yet another expression evaluator. etc. which could be quite slow or could fragment memory. • ‘factorize’ is a PollardRho factorization program. Reallocations An mpz_t or mpq_t variable used to hold successively increasing values will have its memory repeatedly realloced. a combination of the ‘expr’ demo and the ‘pexpr’ tree backend perhaps. • The ‘calc’ subdirectory has a similar but simpler evaluator using lex and yacc. Going beyond something minimal quickly leads to matters like userdefined functions. • The ‘perl’ subdirectory is a comprehensive perl interface to GMP. . consideration has been given at various times to some sort of expression evaluation within the main GMP library. a library designed for ease of use within a C program. depending on the C library. It should be possible to integrate something like that with a garbage collector too.11 Efficiency Small Operands On small operands. But for now the above evaluators are offered as illustrations. the time for function call overheads and memory allocation can be significant in comparison to actual calculation. A language interpreter might want to keep a free list or stack of initialized variables ready for use. since this can be quite time consuming. See ‘demos/expr/README’ for more information. • ‘primes’ counts or lists primes in an interval.) Something simple for program input convenience may yet be a possibility./pexpr 68^975+10 The following programs are provided • ‘pexpr’ is an expression evaluator. Static Linking On some CPUs.pm’. the program used on the GMP web page.a’ should be used for maximum speed. since the PIC code in the shared ‘libgmp. • ‘isprime’ is a commandline interface to the mpz_probab_prime_p function. If an application can estimate the final size then mpz_init2 or mpz_realloc2 can be called to allocate the necessary space from the beginning (see Section 5. in particular the x86s. page 29). Initializing and Clearing Avoid excessive initializing and clearing of variables. using a sieve. • ‘qcn’ is an example use of mpz_kronecker_ui to estimate quadratic class numbers.
y) will. but that’s all. They use an algorithm which is faster than mpz_tdiv_ui. because such inputs will usually be very rare and testing every time would be wasteful. If y is big then cache locality may be helped. a separate destination is slightly better. 29 or 31 take a remainder modulo 23 × 29 × 31 = 20677 and then test that. unless y is only one limb. If the quotient is only rarely wanted then it’s probably best to just take a remainder and then go back and calculate the quotient if and when it’s wanted (mpz_divexact_ui can be used if the remainder is zero). but if it might arise from two pointers to the same object then a test to avoid it might be desirable. only whether it’s zero (or a particular value). but which gives no useful information about the actual remainder. 2exp Functions It’s up to an application to call functions like mpz_mul_2exp when appropriate. Note that it’s never worth introducing extra mpz_set calls just to get inplace operations.x). Normally that copying will only be a tiny fraction of the time for the multiply. so a call like mpz_set(x. mpz_add_ui.x. The division functions like mpz_tdiv_q_ui which give a quotient as well as a remainder are generally a little slower than the remainderonly functions like mpz_ tdiv_ui. make a temporary copy of x before forming the result.22 GNU MP 5. For instance to test whether a number is divisible by any of 23. mpf_set. But if for example an mpz_t contains a value that fits in an unsigned long there’s no need extract it and call a ui function.1 It doesn’t matter if a size set with mpz_init2 or mpz_realloc2 is too small. Badly overestimating memory required will waste space though. Naturally that would never be written deliberately. etc. mpq_set_num. mpq_neg and mpf_neg are fast when used for inplace operations like mpz_abs(x. since all functions will do a further reallocation if necessary. since usually only one or two limbs of x will need to be changed. mpf_add_ui and mpf_sub_ui benefit from an inplace operation like mpz_add_ui(x.0. so this is not a particularly important consideration. A call like mpz_mul(x. mpz_sub_ui. mpq_set. mpf_abs. General purpose functions like mpz_mul make no attempt to identify powers of two or other special forms. InPlace Operations mpz_abs. However when testing divisibility by several small integers. mpq_abs. y). The same applies to the full precision mpz_add etc if y is small. Divisibility Testing (Small Integers) mpz_divisible_ui_p and mpz_congruent_ui_p are the best functions for testing whether an mpz_t is divisible by an individual small integer. since in the current implementation only a single field of x needs changing. ui and si Functions The ui functions and the small number of si functions exist for convenience and should be used where applicable. it’s best to take a remainder modulo their product.y). just use the regular mpz function.x) will be wasteful. make no attempt to recognise a copy of something to itself. if (x != y) mpz_set (x. mpz_set. mpz_mul is currently the opposite. If a result should go to a particular variable then just direct it there and let GMP take care of data movement.x. to save multiprecision operations. mpz_neg. On suitable compilers (GCC for instance) this is inlined too. .
z). 3. In new enough versions of GCC. mpq_denref(q)). Or when forming a big product it might be known that very little cancellation will be possible. adding overhead to each function call and each stack allocation. page 3.5 [Applying Integer Functions]. For example when multiplying by a prime it’s enough to check for factors of it in the denominator instead of doing a full GCD. /* mpq = mpz */ mpz_submul (mpq_numref(q). The mpq_numref and mpq_denref macros give access to the numerator and denominator to do things outside the scope of the supplied mpq functions. The canonical form for rationals allows mixedtype mpq_t and integer additions or subtractions to be done directly with multiples of the denominator. cancelling factors every time is the best approach since it minimizes the sizes for subsequent operations. ‘fstackcheck’ may be able to ensure an overflow is recognised by the system before too much damage is done.12 Debugging Stack Overflow Depending on the system. as proposed by King Charles XII of Sweden and later reformers. These options must be added to the ‘CFLAGS’ used in the GMP build (see Section 2.Chapter 3: GMP Basics 23 Rational Arithmetic The mpq functions operate on mpq_t values with no common factors in the numerator and denominator. 123UL).1 [Build Options]. This will be somewhat faster than mpq_add. . Common factors are checkedfor and cast out as necessary. If a range of values is wanted it’s probably best to call to get a starting point and iterate from there. /* mpq increment */ mpz_add (mpq_numref(q). or swap them for divisions. See Section 6. so the base doesn’t matter much. for how to address this. mpq_denref(q).1 [Build Options]. Maybe we can hope octal will one day become the normal base for everyday use. Powerof2 bases like these can be converted much more efficiently than other bases. adding them just to an application will have no effect. In general. like decimal. Note also they’re a slowdown. For big numbers there’s usually nothing of particular interest to be seen in the digits. page 46. See ‘enablealloca’ choices in Section 2. Text Input/Output Hexadecimal or octal are suggested for input or output in text form. Number Sequences Functions like mpz_fac_ui. However. For example. mpz_fib_ui and mpz_bin_uiui are designed for calculating isolated values. mpq_numref(q). a segmentation violation or bus error might be the only indication of stack overflow. page 3). or ‘fstacklimitsymbol’ or ‘fstacklimitregister’ may be able to add checking if the system itself doesn’t do any (see Section “Options for Code Generation” in Using the GNU Compiler Collection (GCC)). applications that know something about the factorization of the values they’re working with might be able to avoid some of the GCDs used for canonicalization. and so canonicalization can be left to the end. /* mpq += unsigned long */ mpz_addmul_ui (mpq_numref(q). mpq_denref(q).
Assertion Checking The build option ‘enableassert’ is available to add some consistency checks to the library (see Section 2. For example ‘mpz’.c’.lo rules appropriately. In particular on x86 and 68k systems ‘fomitframepointer’ is used and this generally inhibits stack backtracing. and might require GNU make. see Section “Allocation Debugging” in The GNU C Library Reference Manual.perens. Other possibilities. Source File Paths GMP has multiple source files with the same name.gdbinit’ is included in the distribution.cbmamiga. include http://www. Note that these functions shouldn’t be used in final application code since they’re undocumented and may be subject to incompatible changes in future versions of GMP. Note however that only the .com/~otaylor/memprof/ http://www.ethz. GDB. One way to do that is to use a separate object directory with an absolute path to the source directory. showing how to call some undocumented dump functions to print GMP variables from within GDB.gnupdate.24 GNU MP 5.inf.org/stable/devel/fda http://www.debian.uk/mpatrol/ The GMP default allocation routines in ‘memory. Failing to init GMP variables will have unpredictable effects.0. If the debugger can’t already determine the right one it may help to build with absolute paths on each C file. will benefit from ‘enableassert’ since it adds checks on the parameters of most such functions.ch/personal/biere/projects/ccmalloc/ http://dmalloc. many of which have subtle restrictions on their usage. These are likely to be of limited value to most applications.com/FreeSoftware/ (electric fence) http://packages. however.c.demon. but might find other uses.1 Heap Problems The most likely cause of application problems with GMP is heap corruption. Alternately it might be possible to change the . Stack Backtraces On some systems the compiler options GMP uses by default can interfere with debugging.1 [Build Options].org/components/leakbug/ http://people. Assertion failures are just as likely to indicate memory corruption as a library or compiler bug.c’ also have a simple sentinel scheme which can be enabled with #define DEBUG in that file. In all such cases a malloc debugger is recommended. and corruption arising elsewhere in a program may well affect GMP. or ‘man 3 malloc’. Initializing GMP variables more than once or failing to clear them will cause memory leaks.redhat. cd /my/build/dir /my/source/dir/gmp5. On a GNU or BSD system the standard C library malloc has some diagnostic facilities. though the usual caveats about it potentially moving a memory problem or hiding a compiler bug will apply. This is mainly designed for detecting buffer overruns during GMP development.co.com/ http://www.1/configure This works via VPATH. ‘mpq’ and ‘mpf’ each have an ‘init. Applications using the lowlevel mpn functions. page 3). in no particular order.0. in different directories. the GNU Debugger A sample ‘. Recompiling without such options may help while debugging.
so CPU ‘none’ should be used for maximum checking. and memory leaks. on the other hand. add ‘enablecxx CXXFLAGS=g’. . or may be allocated with a compiler builtin alloca which will go nowhere near any malloc debugger hooks. memory accesses through bad pointers. The GMP C++ features cannot be used. for an x86 without them (for instance plain ‘i486’). On GNU/Linux for example. page 28.9. This will run very very slowly. Recent versions of Valgrind are getting support for MMX and SSE/SSE2 instructions. In a normal build. but on most systems the resolution of the sampling is quite low (10 milliseconds for instance).Chapter 3: GMP Basics 25 generic C code has checks. ie. Checker The GCC checker (http://savannah. Most of the GMP assembly code has the necessary symbol information. Valgrind Other Problems Any suspected bug in GMP itself should be isolated to make sure it’s not an application problem. A build of GMP with checking within GMP itself can be made. for past versions GMP will need to be configured not to use those. This can help a malloc debugger detect accesses outside the intended bounds. a GMP build for maximum debuggability would be .org/) is a memory checker for x86s. Maximum Debuggability To summarize the above./configure host=nonepclinuxgnu CC=checkergcc ‘host=none’ must be used.9. not the assembly code. 3. temporary memory is allocated in blocks which GMP divides up for its own use. It translates and emulates machine instructions to do strong checks for uninitialized data (at the level of individual bits). ‘disableprofiling’ The default is to add nothing special for profiling. since the GMP assembly code doesn’t support the checking scheme. or detect memory not released.13 Profiling Running a program under a profiler is a good way to find where it’s spending most time and where improvements can be best sought.nongnu. The valgrind program (http://valgrind. requiring long runs to get accurate information. ./configure disableshared enableassert \ enablealloca=debug host=none CFLAGS=g For C++. The profiling choices for a GMP build are as follows. This approach has the advantage of minimizing interference with normal program operation. Temporary Memory Checking The build option ‘enablealloca=debug’ arranges that each block of temporary memory in GMP is allocated with a separate call to malloc (or the allocation function set with mp_set_memory_functions). It should be possible to just compile the mainline of a program with p and use prof to get a profile consisting of timerbased sampling of the program counter.1) don’t yet support the standard C++ library. see Chapter 4 [Reporting Bugs].org/projects/checker/) can be used with GMP. since current versions of checker (0. It contains a stub library which means GMP applications compiled with checker can use a normal GMP build.
On some systems. This provides call counting in addition to program counter sampling. but on other processors the assembly routines will be as if compiled without ‘p’ and therefore won’t appear in the call counts. which makes it possible to count calls coming from different locations.net/projects/fnccheck/ This should be included in ‘LIBS’ during the GMP configure so that test programs will link. 3. so programs compiled with this option will link. such as GNU/Linux. But in that case only the ‘gprof p’ flat profile and call counts can be expected to be valid. which allows the most frequently called routines to be identified. The x86 assembly code supports this option. This inserts special instrumenting calls at the start and end of each function. which means ‘pg’ added to the ‘CFLAGS’.1 ‘enableprofiling=prof’ Build with support for the system prof. ‘enableprofiling=gprof’ Build with support for gprof. This instrumenting is not normally a standard system feature and will require support from an external library.26 GNU MP 5. which might result in poorer code generation. . so only a total time in mpn_mul would be accumulated. On x86 and m68k systems ‘pg’ and ‘fomitframepointer’ are incompatible. For example. but on other processors the assembly routines will be as if compiled without ‘finstrumentfunctions’ meaning time spent in them will effectively be attributed to their caller.0./configure enableprofiling=instrument LIBS=lfc On a GNU system the C library provides dummy instrumenting functions. This provides call graph construction in addition to call counting and program counter sampling. allowing exact timing and full call graph construction. The program counter sampling is still flat though. ‘p’ in fact means ‘pg’ and in this case ‘enableprofiling=gprof’ described below should be used instead. it should be possible to use the gprof program with a plain ‘enableprofiling=prof’ build. . not the ‘gprof q’ call graph. ‘enableprofiling=instrument’ Build with the GCC option ‘finstrumentfunctions’ added to the ‘CFLAGS’ (see Section “Options for Code Generation” in Using the GNU Compiler Collection (GCC)). For example the number of calls to mpn_mul from mpz_mul versus the number from mpf_mul. In this case it’s only necessary to ensure the correct library is added when linking an application. The x86 assembly code has support for this option. and an average time spent in each routine to be determined. not a separate amount for each call site. which means ‘p’ added to the ‘CFLAGS’. such as http://sourceforge.14 Autoconf Autoconf based applications can easily check whether GMP is installed. but on other processors the assembly routines will be as if compiled without ‘pg’ and therefore not be included in the call counts. The only thing to be noted is that GMP library symbols from version 3 onwards have prefixes like __gmpz. The following therefore would be a simple test. so the latter is omitted from the default flags in that case. The x86 assembly code has support for this option. Incidentally.
The following would suit Autoconf 2. see http://gmplib. That would make it possible to test the exact version.emacs’. see http://gmplib. This can be done in the normal way with mp_limb_t etc.Chapter 3: GMP Basics 27 AC_CHECK_LIB(gmp.0 or up is best for this. or not 3. but an application that must have GMP would want to generate an error if not found.50 or up. . AC_CHECK_LIB(gmp.* " "\\>") (nth 3 modevalue))))) . For example mpz_mul_si was added in GMP 3.h>]) 3.15 Emacs CH CI (infolookupsymbol) is a good way to find documentation on C functions while editing (see Section “Info Documentation Lookup” in The Emacs Editor). In general it’s recommended that applications should simply demand a new enough GMP rather than trying to provide supplements for features not available in past versions. AC_CHECK_LIB(gmp. .1. if some particular subminor release is known to be necessary.h’ using say AC_EGREP_CPP. AC_CHECK_SIZEOF(mp_limb_t. but GMP 4.1 or up. [#include <gmp. __gmpz_init) This just uses the default AC_CHECK_LIB actions for found or not found. __gmpz_init. (evalafterload "infolook" ’(let ((modevalue (assoc ’cmode (assoc ’symbol infolookupalist)))) (setcar (nthcdr 3 modevalue) (cons ’("(gmp)Function Index" nil "^ . Occasionally an application will need or want to know the size of a type at configuration or preprocessing time. [AC_MSG_ERROR([GNU MP not found. not just with sizeof in the code. For example.org/])]) An alternative would be to test the version number in ‘gmp.org/])]) If functions added in some particular version of GMP are required. [AC_MSG_ERROR( [GNU MP not found. then one of those can be used when checking. . since prior versions needed certain ‘D’ defines on systems using a long long limb. The GMP manual can be included in such lookups by putting the following in your ‘. __gmpz_mul_si.
please investigate it and report it. We have made this library available to you.guess’ (might be the same). if the bug report is poor. Before you report a bug. • The GMP version number.guess’. please send a note to the same address. and it is not too much to ask you to report the bugs you find. If you get a crash. or downright incorrect. check it’s not already addressed in Section 2. executables or straces. For gcc. get the version with ‘gcc v’.org.org/ for patches for this release. • The output from running ‘. • The name of the compiler and its version. If the results are incorrect. Vague queries or piecemeal messages are difficult to act on and don’t help the development effort. • A description of what is wrong. It is not uncommon that an observed problem is actually due to a bug in the compiler. • The output from running ‘uname a’. include a stack backtrace from the debugger if it’s informative (‘where’ in gdb. page 12. then the contents of ‘config. or ‘$C’ in adb). otherwise perhaps ‘what ‘which cc‘’. in what way. Send your report to: gmpbugs@gmplib. we will do our best to help you get a corrected version of the library. say so. Please include the following in any report. • Please do not send core dumps./configfsf. You may also want to check http://gmplib. • The configuration options you used when building GMP. if any. If your bug report is good. we won’t do anything about it (except maybe ask you to send a better report). • If the bug is related to ‘configure’. • A test program that makes it possible for us to reproduce the bug. with something definite that can be tested or debugged. . or if the language needs to be improved. If you think something in this manual is unclear. then the compressed contents of ‘config. page 14. or perhaps Section 2.log’. and if prepackaged or patched then say so. or similar. Include instructions on how to run the program.0.5 [Known Build Problems]. • If the bug is related to an ‘asm’ file not assembling.28 GNU MP 5. and from running ‘.4 [Notes for Particular Systems].1 4 Reporting Bugs If you think you have found a bug in the GMP library. Please make an effort to produce a selfcontained report.s’.m4’ and the offending line or lines from the temporary ‘mpn/tmp<file>. • If you get a crash./config. the GMP code tends to explore interesting corners in compilers.
[Function] void mpz_clears (mpz t x. The value in x is preserved if it fits. Calling this function instead of mpz_init or mpz_inits is never necessary.. mp bitcnt t n ) [Function] Initialize x. { mpz_t integ. /* Unless the program is about to exit. x will grow automatically in the normal way. mpz_sub (integ... mp bitcnt t n ) [Function] Change the space allocated for x to n bits. These functions start with the prefix mpz_. or to decrease it to give memory back to the heap. But this function can be used to increase the space for a variable in order to avoid repeated automatic reallocations.. void mpz_realloc2 (mpz t x. . . do . and set its value to 0.1 Initialization Functions The functions for integer arithmetic assume that all integer objects are initialized.. .. GMP integers are stored in objects of type mpz_t. for subsequent values stored. void mpz_clear (mpz t x ) [Function] Free the space occupied by x. ...). or is set to 0 if not.). You do that by calling the function mpz_init... Call this function for all mpz_t variables when you are done with them. [Function] void mpz_init2 (mpz t x. once an object is initialized.) Free the space occupied by a NULLterminated list of mpz_t variables... you can store new values any number of times. and set their values to 0. Calling this function is never necessary. n is only the initial space. and set its value to 0. mpz_init2 makes it possible to avoid such reallocations if a maximum size is known in advance.. reallocation is handled automatically by GMP when needed. void mpz_init (mpz t x ) Initialize x. with space for nbit numbers. if necessary. . */ mpz_clear (integ). For example..) Initialize a NULLterminated list of mpz_t variables. mpz_add (integ. [Function] void mpz_inits (mpz t x. 5. . .Chapter 5: Integer Functions 29 5 Integer Functions This chapter describes the GMP functions for performing integer arithmetic. reallocation is handled automatically by GMP when needed. } As you can see. mpz_init (integ).
or decimal otherwise. and is simply ignored. .. Here is an example of using one: { mpz_t pie. void void void void void void mpz_set (mpz t rop. The base may vary from 2 to 62. unsigned long int op ) mpz_set_si (mpz t rop.. int base ) [Function] Set the value of rop from str.. page 29). double op ) mpz_set_q (mpz t rop. mpz_sub (pie. [Function] 5.. } Once the integer has been initialized by any of the mpz_init_set.1 [Initializing Integers]. . "3141592653589793238462643383279502884".. These functions’ names have the form mpz_init_set. For bases 37 to 62. then the leading characters are used: 0x and 0X for hexadecimal. For bases up to 36... case is ignored.61. . unsigned long int op ) void mpz_init_set_si (mpz t rop...30 GNU MP 5. or if base is 0. mpz_set_d.1 5..). mpz_set_q and mpz_set_f truncate op to make it an integer.3 Combined Initialization and Assignment Functions For convenience. signed long int op ) [Function] [Function] [Function] .2 Assignment Functions These functions assign new values to already initialized integers (see Section 5. mpz_init_set_str (pie. mpz t rop2 ) Swap the values rop1 and rop2 efficiently. This function returns 0 if the entire string is a valid number in base base. mpz t op ) mpz_set_ui (mpz t rop. White space is allowed in the string. signed long int op ) mpz_set_d (mpz t rop. uppercase letter represent the usual 10. Don’t use an initializeandset function on a variable already initialized! void mpz_init_set (mpz t rop. 10). int mpz_set_str (mpz t rop. 0b and 0B for binary. GMP provides a parallel series of initializeandset functions which initialize the output and then store the value there. uppercase and lowercase letters have the same value.. Otherwise it returns −1.. char *str. mpq t op ) mpz_set_f (mpz t rop.0. mpf t op ) [Function] [Function] [Function] [Function] [Function] [Function] Set the value of rop from op. 0 for octal. mpz t op ) void mpz_init_set_ui (mpz t rop. it can be used as the source or destination operand for the ordinary integer functions. a nullterminated C string in base base. functions. void mpz_swap (mpz t rop1.35 while lowercase letter represent 36. mpz_clear (pie).
and lowercase letters (in that significance order) are used. Otherwise return the least significant part of op.) 5. If op is too big to fit in a signed long int. digits.. double mpz_get_d_2exp (signed long int *exp. only the absolute value is used. If the string is a correct base base number.−36. (I. A hardware overflow trap may or may not occur.12 [I/O of Integers]. the returned result is probably not very useful. The base argument may vary from 2 to 62 or from −2 to −36. the return is 0.. rounding towards zero). The block will be strlen(str)+1 bytes. you have to call mpz_clear for it. mpz t op ) [Function] Convert op to a string of digits in base base. digits and lowercase letters are used. char * mpz_get_str (char *str. Functions for converting to GMP integers are described in Section 5..Chapter 5: Integer Functions 31 void mpz_init_set_d (mpz t rop. signed long int mpz_get_si (mpz t op ) [Function] If op fits into a signed long int return the value of op. To find out if the value will fit. truncating if necessary (ie. unsigned long int mpz_get_ui (mpz t op ) Return the value of op as an unsigned long. char *str. digits and uppercase letters are used.36. int base ) [Function] Initialize rop and set its value like mpz_set_str (see its documentation above for details). double mpz_get_d (mpz t op ) Convert op to a double. rounding towards zero). the result string is allocated using the current allocation function (see Chapter 14 [Custom Allocation]. if an error occurs it returns −1. If str is NULL. page 39. int base.2 [Assigning Integers].0 and 0 is stored to *exp . the result is system dependent. d ∗ 2exp is the (truncated) op value.e. page 30 and Section 5. for −2. and returning the exponent separately. [Function] int mpz_init_set_str (mpz t rop. . rop is initialized even if an error occurs. For base in the range 2. page 86). use the function mpz_fits_slong_p. The sign of op is ignored.5 ≤ d < 1 and the exponent is stored to *exp .62. mpz t op ) [Function] Convert op to a double. [Function] If the exponent from the conversion is too big. with the same sign as op. An infinity is returned where available. This is similar to the standard C frexp function (see Section “Normalization Functions” in The GNU C Library Reference Manual). that being exactly enough for the string and nullterminator. uppercase letters. If op is zero.. The return value is in the range 0. the function returns 0.4 Conversion Functions This section describes functions for converting GMP integers to standard C types. [Function] If op is too big to fit an unsigned long then just the least significant bits that do fit are returned. truncating if necessary (ie. double op ) Initialize rop with limb space and set the initial numeric value from op. for 37.
[Function] void mpz_abs (mpz t rop. mpz t op1. unsigned long int op2 ) Set rop to op1 × op2. mpz t op1. mpz t op2 ) void mpz_addmul_ui (mpz t rop. mpz t r. mpz t n.1 If str is not NULL. mpz t op2 ) Set rop to op1 − op2. mpz t d ) unsigned long int mpz_cdiv_q_ui (mpz t q. long int op2 ) void mpz_mul_ui (mpz t rop.0. it should point to a block of storage large enough for the result. mpz t op1. mpz t op1. mpz t op1. mpz t d ) void mpz_cdiv_r (mpz t r. mpz t op ) Set rop to the absolute value of op. mpz t n. This lets a program handle arithmetic exceptions in these functions the same way as for normal C int arithmetic. base ) + 2. being either the allocated block. mpz t op ) Set rop to −op.32 GNU MP 5. mp bitcnt t op2 ) void mpz_neg (mpz t rop. The two extra bytes are for a possible minus sign. 5. mpz t op1. mpz t op1. mpz t op2 ) void mpz_add_ui (mpz t rop.6 Division Functions Division is undefined if the divisor is zero. [Function] 5. unsigned long int op2 ) Set rop to op1 + op2. mpz t op1. unsigned long int op2 ) Set rop to rop − op1 × op2. mpz t op2 ) void mpz_sub_ui (mpz t rop. mpz t op2 ) void mpz_mul_si (mpz t rop. mpz t op1. and the nullterminator. [Function] [Function] void mpz_mul_2exp (mpz t rop. mpz t n. mpz t d ) void mpz_cdiv_qr (mpz t q. [Function] [Function] void mpz_submul (mpz t rop. A pointer to the result string is returned. mpz t op1. will cause an intentional division by zero.5 Arithmetic Functions void mpz_add (mpz t rop. unsigned long int op1. unsigned long int d ) [Function] [Function] [Function] [Function] [Function] . void mpz_cdiv_q (mpz t q. mpz t n. [Function] [Function] void mpz_sub (mpz t rop. [Function] [Function] [Function] void mpz_mul (mpz t rop. [Function] Set rop to op1 × 2op2 . mpz t op1. mpz t n. that being mpz_sizeinbase (op. This operation can also be defined as a left shift by op2 bits. or the given str. unsigned long int op2 ) Set rop to rop + op1 × op2. [Function] [Function] [Function] void mpz_addmul (mpz t rop. Passing a zero divisor to the division or modulo functions (including the modular powering functions mpz_powm and mpz_powm_ui). mpz t op2 ) void mpz_submul_ui (mpz t rop. unsigned long int d ) unsigned long int mpz_cdiv_r_ui (mpz t r. unsigned long int op2 ) void mpz_ui_sub (mpz t rop. mpz t op1.
The t stands for “truncate”. mpz t d ) unsigned long int mpz_tdiv_q_ui (mpz t q. unsigned long int d ) void mpz_fdiv_q_2exp (mpz t q. mpz t n. mpz t n. These functions are implemented as right shifts and bit masks. mpz t n. mpz t r. mpz t n. mpz t n. mpz t n. unsigned long int d ) void mpz_cdiv_q_2exp (mpz t q. or results will be unpredictable. • tdiv rounds q towards zero. mpz t n. mp bitcnt t b ) void mpz_fdiv_r_2exp (mpz t r. forming a quotient q and/or remainder r. mpz t n. mpz t d ) void mpz_fdiv_r (mpz t r. d = 2b . unsigned long int d ) unsigned long int mpz_cdiv_ui (mpz t n. and r will satisfy 0 ≤ r < d. mpz t d ) void mpz_tdiv_qr (mpz t q. For the ui variants the return value is the remainder. mpz t n . The c stands for “ceil”. The q functions calculate only the quotient. The rounding is in three styles. • cdiv rounds q up towards +∞. mpz t n. mpz t r. mpz t n. mpz t d ) void mpz_fdiv_qr (mpz t q. mpz t n. mpz t n. unsigned long int d ) unsigned long int mpz_tdiv_ui (mpz t n. mp bitcnt t b ) [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] Divide n by d. mpz t n. and r will have the same sign as d. mpz t n. unsigned long int d ) unsigned long int mpz_fdiv_qr_ui (mpz t q.Chapter 5: Integer Functions 33 unsigned long int mpz_cdiv_qr_ui (mpz t q. and r will have the same sign as n. mpz t n . mpz t r. Note that for qr the same variable cannot be passed for both q and r. unsigned long int d ) void mpz_tdiv_q_2exp (mpz t q. • fdiv rounds q down towards −∞. and the qr functions calculate both. mp bitcnt t b ) void mpz_fdiv_q (mpz t q. In all cases q and r will satisfy n = qd + r. mp bitcnt t b ) void mpz_cdiv_r_2exp (mpz t r. unsigned long int d ) unsigned long int mpz_tdiv_qr_ui (mpz t q. For the 2exp functions. mpz t n. so for those the return value is the absolute value of the remainder. mpz t d ) unsigned long int mpz_fdiv_q_ui (mpz t q. . mpz t r. but of course they round the same as the other functions. and r will have the opposite sign to d. mpz t n. unsigned long int d ) unsigned long int mpz_fdiv_r_ui (mpz t r. mpz t n . mp bitcnt t b ) void mpz_tdiv_q (mpz t q. mp bitcnt t b ) void mpz_tdiv_r_2exp (mpz t r. and in fact returning the remainder is all the div_ui functions do. mpz t r. The f stands for “floor”. unsigned long int d ) unsigned long int mpz_tdiv_r_ui (mpz t r. For the 2exp variants the divisor is 2b . For tdiv and cdiv the remainder can be negative. the r functions only the remainder. unsigned long int d ) unsigned long int mpz_fdiv_ui (mpz t n. each suiting different applications. mpz t d ) void mpz_tdiv_r (mpz t r.
void mpz_mod (mpz t r.9 [Number Theoretic Functions].34 GNU MP 5. unsigned long int c. or in the case of mpz_divisible_2exp_p by 2b . returning the remainder as well as setting r.0. mpz t base. or in the case of mpz_congruent_2exp_p modulo 2b .1 For positive n both mpz_fdiv_q_2exp and mpz_tdiv_q_2exp are simple bitwise right shifts. mpz t exp. mpz_fdiv_q_2exp is effectively an arithmetic right shift treating n as twos complement the same as the bitwise logical functions do. mpz t d ) void mpz_divexact_ui (mpz t q. unsigned long int d ) int mpz_congruent_2exp_p (mpz t n. and are the best choice when exact division is known to occur. for example reducing a rational to lowest terms. mpz t d ) int mpz_divisible_ui_p (mpz t n. For negative n. [Function] [Function] void mpz_powm_sec (mpz t rop. void mpz_divexact (mpz t q. mpz t d ) int mpz_congruent_ui_p (mpz t n. unsigned long int d ) int mpz_divisible_2exp_p (mpz t n.7 Exponentiation Functions void mpz_powm (mpz t rop. If an inverse doesn’t exist then a divide by zero is raised. See mpz_fdiv_ui above if only the return value is wanted. mpz t mod ) void mpz_powm_ui (mpz t rop. Unlike the other division functions. mpz t d ) unsigned long int mpz_mod_ui (mpz t r. [Function] . These routines are much faster than the other division functions. mpz t n. Negative exp is supported if an inverse base −1 mod mod exists (see mpz_invert in Section 5. The sign of the divisor is ignored. unsigned long int exp. mpz t mod ) Set rop to baseexp mod mod. mp bitcnt t b ) [Function] [Function] [Function] Return nonzero if n is congruent to c modulo d. mpz t base. d = 0 is accepted and following the rule it can be seen that n and c are considered congruent mod 0 only when exactly equal. mp bitcnt t b ) [Function] [Function] [Function] Return nonzero if n is exactly divisible by d. mpz_mod_ui is identical to mpz_fdiv_r_ui above. d = 0 is accepted and following the rule it can be seen that only 0 is considered divisible by 0. mpz t exp. mpz t n. 5. int mpz_congruent_p (mpz t n. These functions produce correct results only when it is known in advance that d divides n. the result is always nonnegative. mpz t n. Unlike the other division functions. mpz t base. It is required that exp > 0 and that mod is odd. unsigned long int d ) [Function] [Function] Set r to n mod d. whereas mpz_tdiv_q_2exp effectively treats n as sign and magnitude. mpz t c. mpz t n. n is divisible by d if there exists an integer q satisfying n = qd. unsigned long d ) [Function] [Function] Set q to n/d. page 35). n is congruent to c mod d if there exists an integer q satisfying n = c + qd. mpz t c. mpz t mod ) Set rop to baseexp mod mod. int mpz_divisible_p (mpz t n.
where resilience to sidechannel attacks is desired. Set rop2 to the remainder (op − rop1 ). assuming that function arguments are placed at the same position and that the machine state is identical upon function entry. like mpz_sqrt.. n void mpz_sqrt (mpz t rop. Under this definition both 0 and 1 are considered to be perfect squares. int mpz_perfect_power_p (mpz t op ) [Function] Return nonzero if op is a perfect power. return 1 if n is probably prime (without being certain). [Function] void mpz_sqrtrem (mpz t rop1. more will reduce the chances of a composite being returned as “probably prime”. with b > 1. if the square root of op is an integer. the truncated integer part of the nth root of op. int mpz_perfect_square_p (mpz t op ) [Function] Return nonzero if op is a perfect square. This function does some trial divisions. i. mpz t base. [Function] u . mpz t op. unsigned long int base. MillerRabin and similar tests can be more properly called compositeness tests. 5.8 Root Extraction Functions int mpz_root (mpz t rop. hence those which pass are considered probably prime. reps controls how many such tests are done. [Function] [Function] 5.e. mpz t rop2. then some MillerRabin probabilistic primality tests.e. which will be zero if op is a perfect square. the truncated integer part of the square root of op. int reps ) [Function] Determine whether n is prime. void mpz_rootrem (mpz t root. if op is rop to the nth power. void mpz_pow_ui (mpz t rop. mpz t op ) √ [Function] Set rop1 to op . Under this definition both 0 and 1 are considered to be perfect powers. Return nonzero if the computation was exact. i.Chapter 5: Integer Functions 35 This function is designed to take the same time and have the same cache access patterns for any two samesize arguments. i. or return 0 if n is definitely composite. This function is intended for cryptographic purposes. mpz t rem. Negative values of op are accepted. such that op = ab . the results are undefined. . but of course can only be odd perfect powers. unsigned long int exp ) Set rop to baseexp . 2 If rop1 and rop2 are the same variable.. unsigned long int n ) [Function] √ Set rop to n op . mpz t u.e.9 Number Theoretic Functions int mpz_probab_prime_p (mpz t n. Set rem to the remainder. 5 to 10 is a reasonable number. The case 00 yields 1. mpz t op ) Set rop to √ op . the truncated integer part of the nth root of u.. Return 2 if n is definitely prime. Only a few composites pass. if there exist integers a and b. unsigned long int n ) √ Set root to (u − rootn ). unsigned long int exp ) void mpz_ui_pow_ui (mpz t rop. Numbers which fail are known to be composite but those which pass might be prime or might be composite.
[Function] This function uses a probabilistic algorithm to identify primes. and the result is equal to the argument op1. [Function] int mpz_invert (mpz t rop. If the result does not fit. and in addition set s and t to coefficients satisfying as + bt = g.0. unsigned long op2 ) [Function] [Function] Set rop to the least common multiple of op1 and op2. mpz t b ) mpz_kronecker_si (mpz t a.36 GNU MP 5. mpz t s. If t is NULL then that value is not computed. mpz t a. . mpz t op ) Set rop to the next prime greater than op. or When b is odd the Jacobi symbol and Kronecker symbol are identical. and for such p it’s identical to the Jacobi symbol. [Function] unsigned long int mpz_gcd_ui (mpz t rop. mpz t op2 ) void mpz_lcm_ui (mpz t rop. mpz t op2 ) [Function] Set rop to the greatest common divisor of op1 and op2. If the result is small enough to fit in an unsigned long int. mpz t op1. void mpz_gcdext (mpz t g. mpz t b ) a b Calculate the Jacobi symbol a = 0 when a even. int int int int int mpz_kronecker (mpz t a. unsigned long int op2 ) Compute the greatest common divisor of op1 and op2. 0 is returned. void mpz_gcd (mpz t rop. mpz t b ) [Function] Set g to the greatest common divisor of a and b. the return value is nonzero and rop will satisfy 0 ≤ rop < op2. it is returned. mpz t b ) Calculate the Jacobi symbol a b . mpz t op1. mpz t op1. [Function] Compute the inverse of op1 modulo op2 and put the result in rop. This is defined only for b odd. the chance of a composite passing will be extremely small. mpz t t. The values in s and t are chosen such that s ≤ b and t ≤ a. void mpz_lcm (mpz t rop. mpz t op1. mpz t b ) mpz_ui_kronecker (unsigned long a. If rop is not NULL. irrespective of the signs of op1 and op2. rop will be zero if either op1 or op2 is zero. mpz t op2 ) int mpz_jacobi (mpz t a. If an inverse doesn’t exist the return value is zero and rop is undefined. mpz t p ) a p [Function] Calculate the Legendre symbol .1 void mpz_nextprime (mpz t rop. int mpz_legendre (mpz t a. 2 with the Kronecker extension a 2 = 2 a [Function] [Function] [Function] [Function] [Function] when a odd. rop is always positive. store the result there. This is defined only for p an odd positive prime. so mpz_kronecker_ui etc can be used for mixed precision Jacobi symbols too. The value in g is always positive. If the inverse exists. long b ) mpz_kronecker_ui (mpz t a. For practical purposes it’s adequate. Note that the result will always fit if op2 is nonzero. mpz t op1. unsigned long b ) mpz_si_kronecker (long a. even if one or both of a and b are negative. The result is always positive even if one or both input operands are negative.
2. unsigned long int n ) [Function] [Function] mpz_fib_ui sets fn to to Fn . [Function] void mpz_fac_ui (mpz t rop. mpz t op2 ) mpz_cmp_d (mpz t op1. mpz t op.4. Negative values of n are −n k supported by mpz_bin_ui. When a sequence of values is wanted it’s best to start with mpz_fib2_ui and iterate the defining Fn+1 = Fn +Fn−1 or similar. and fnsub1 to Fn−1 . zero if op1 = op2. double op2 ) mpz_cmp_si (mpz t op1.2 (see Appendix B [References]. mpz t fnsub1. unsigned long int k ) void mpz_bin_uiui (mpz t rop. 5. unsigned long int k ) Compute the binomial coefficient n k [Function] [Function] and store the result in rop. double op2 ) [Function] [Function] . see Knuth volume 1 void mpz_fib_ui (mpz t fn. mpz t lnsub1. mpz_cmp_d can be called with an infinity. unsigned long int n ) void mpz_lucnum2_ui (mpz t ln.c’ which uses mpz_kronecker_ui. or a negative value if op1 < op2. mpz t f ) [Function] Remove all occurrences of the factor f from op and store the result in rop. using the identity section 1. These functions are designed for calculating isolated Fibonacci numbers. unsigned long int op ) Set rop to op!. unsigned long int n ) void mpz_fib2_ui (mpz t fn.6 part G. = (−1)k n+k−1 k . page 108. and lnsub1 to Ln−1 . unsigned long int n ) [Function] [Function] mpz_lucnum_ui sets ln to to Ln . but results are undefined for a NaN. These functions are designed for calculating isolated Lucas numbers.7. mpz t op2 ) int mpz_cmpabs_d (mpz t op1. When a sequence of values is wanted it’s best to start with mpz_lucnum2_ui and iterate the defining Ln+1 = Ln + Ln−1 or similar. mpz_lucnum2_ui sets ln to Ln .10 Comparison Functions int int int int [Function] [Function] [Macro] [Macro] Compare op1 and op2. The formulas for going from Fibonacci to Lucas can be found in Section 16. or any number theory textbook. signed long int op2 ) mpz_cmp_ui (mpz t op1. mpz t n. The return value is how many such occurrences were removed. so it’s never necessary to call both mpz_fib2_ui and mpz_lucnum2_ui. unsigned long int op2 ) int mpz_cmpabs (mpz t op1. mpz_cmp_ui and mpz_cmp_si are macros and will evaluate their arguments more than once. page 122). the reverse is straightforward too. The Fibonacci numbers and Lucas numbers are related sequences.5 [Lucas Numbers Algorithm]. mpz_fib2_ui sets fn to Fn .Chapter 5: Integer Functions 37 For more information see Henri Cohen section 1. mp_bitcnt_t mpz_remove (mpz t rop. mpz_cmp (mpz t op1. unsigned long int n. the n’th Fibonacci number. void mpz_lucnum_ui (mpz t ln. See also the example program ‘demos/qcn. Return a positive value if op1 > op2. the n’th Lucas number. the factorial of op. void mpz_bin_ui (mpz t rop.
which is the number of 1 bits in the binary representation. [Function] void mpz_xor (mpz t rop. mpz_cmpabs_d can be called with an infinity. mpz t op2 ) Set rop to op1 bitwiseand op2. mpz t op1. towards more significant bits. but results are undefined for a NaN. and −1 if op < 0.11 Logical and Bit Manipulation Functions These functions behave as if twos complement arithmetic were used (although signmagnitude is the actual implementation). [Function] mp_bitcnt_t mpz_popcount (mpz t op ) [Function] If op ≥ 0. mp bitcnt t bit_index ) Set bit bit index in rop. mp bitcnt t bit_index ) Clear bit bit index in rop. mpz t op1. [Macro] This function is actually implemented as a macro. return the population count of op. then the largest possible mp_bitcnt_t is returned. [Function] If op1 and op2 are both ≥ 0 or both < 0. or mpz_scan1 past the end of a nonnegative number. If there’s no bit found. Return a positive value if op1 > op2. until the first 0 or 1 bit (respectively) is found. the number of 1s is infinite. 0 if op = 0. zero if op1 = op2. mpz t op ) Set rop to the one’s complement of op. int mpz_sgn (mpz t op ) Return +1 if op > 0. void mpz_setbit (mpz t rop. If one operand is ≥ 0 and the other < 0 then the number of bits different is infinite. [Function] void mpz_com (mpz t rop. and the return value is the largest possible mp_bitcnt_t. void mpz_and (mpz t rop. then starting bit is returned. return the hamming distance between the two operands. If the bit at starting bit is already what’s sought. mpz t op1. mpz t op2 ) Set rop to op1 bitwise exclusiveor op2.0.1 int mpz_cmpabs_ui (mpz t op1. and the return value is the largest possible mp_bitcnt_t. mp bitcnt t starting_bit ) mp_bitcnt_t mpz_scan1 (mpz t op. [Function] void mpz_ior (mpz t rop. which is the number of bit positions where op1 and op2 have different bit values. [Function] . [Function] void mpz_clrbit (mpz t rop. 5. It evaluates its argument multiple times. If op < 0. starting from bit starting bit. This will happen in mpz_scan0 past the end of a negative number. The least significant bit is number 0. mpz t op2 ) Set rop to op1 bitwise inclusiveor op2. unsigned long int op2 ) [Function] Compare the absolute values of op1 and op2. or a negative value if op1 < op2.38 GNU MP 5. mp bitcnt t starting_bit ) [Function] [Function] Scan op. Return the index of the found bit. mpz t op2 ) mp_bitcnt_t mpz_scan0 (mpz t op. mp_bitcnt_t mpz_hamdist (mpz t op1.
return 0. size_t mpz_out_raw (FILE *stream. digits and uppercase letters are used. then the leading characters are used: 0x and 0X for hexadecimal. case is ignored..e. with 4 bytes of size information. mp bitcnt t bit_index ) Complement bit bit index in rop. return 0. The base may vary from 2 to 62. For bases 37 to 62. int base ) [Function] Input a possibly whitespace preceded string in base base from stdio stream stream. or if an error occurred. uppercase and lowercase letters have the same value.12 Input and Output Functions Functions that perform input from a stdio stream.35 while lowercase letter represent 36. and that many bytes of limbs. int base. FILE *stream. Return the number of bytes read. The output of this can not be read by mpz_inp_raw from GMP 1. for −2.36. Both the size and the limbs are written in decreasing significance order (i.Chapter 5: Integer Functions 39 void mpz_combit (mpz t rop. The base argument may vary from 2 to 62 or from −2 to −36. and functions that output to a stdio stream. digits and lowercase letters are used. size_t mpz_inp_str (mpz t rop. size_t mpz_out_str (FILE *stream. or if base is 0. . uppercase letters.. respectively. Passing a NULL pointer for a stream argument to any of these functions will make them read from stdin and write to stdout.h’ to define prototypes for these functions. and put the result in rop. 0b and 0B for binary. in spite of changes necessary for compatibility between 32bit and 64bit machines.. return 0.. or decimal otherwise. return 0. The output can be read with mpz_inp_raw. size_t mpz_inp_raw (mpz t rop. This routine can read the output from mpz_out_raw also from GMP 1. Return the number of bytes written.. 0 for octal. The integer is written in a portable format. since that will allow ‘gmp. for 37.−36. in raw binary format. in bigendian). FILE *stream ) [Function] Input from stdio stream stream in the format written by mpz_out_raw. and put the read integer in rop. as a string of digits in base base. and lowercase letters (in that significance order) are used. mpz t op ) [Function] Output op on stdio stream stream.. [Function] int mpz_tstbit (mpz t op. Return the number of bytes read. uppercase letter represent the usual 10. For base in the range 2. Return the number of bytes written. or if an error occurred. digits.h’ before ‘gmp. or if an error occurred. because of changes necessary for compatibility between 32bit and 64bit machines.62. [Function] 5. it is a good idea to include ‘stdio.61. When using any of these functions. or if an error occurred. For bases up to 36. mpz t op ) [Function] Output op on stdio stream stream. mp bitcnt t bit_index ) Test bit bit index in op and return 0 or 1 accordingly.h’.
size t size. void mpz_random (mpz t rop. [Function] The variable state must be initialized by calling one of the gmp_randinit functions (Section 9. inclusive. Please see the Chapter 9 [Random Number Functions]. page 65 for more information on how to use and not to use random number functions. void mpz_rrandomb (mpz t rop.1 [Random State Initialization]. mp bitcnt t n ) [Function] Generate a uniformly distributed random integer in the range 0 to 2n − 1. since this kind of random numbers have proven to be more likely to trigger cornercase bugs. page 65) before invoking this function. inclusive. with long strings of zeros and ones in the binary representation. gmp randstate t state. mp size t max_size ) [Function] Generate a random integer of at most max size limbs. void mpz_urandomm (mpz t rop.1 5. since this kind of random numbers have proven to be more likely to trigger cornercase bugs. order can be 1 for most significant word first or 1 for least significant first. and newer functions that accept a state parameter that is read and modified. Negative random numbers are generated when max size is negative. gmp randstate t state. mpz t n ) Generate a uniform random integer in the range 0 to n − 1. int endian. count many words are read. The variable state must be initialized by calling one of the gmp_randinit functions (Section 9. void mpz_urandomb (mpz t rop.14 Integer Import and Export mpz_t variables can be converted to and from arbitrary words of binary data with the following functions. page 65) before invoking this function. Use mpz_urandomb or mpz_urandomm instead. mp bitcnt t n ) [Function] Generate a random integer with long strings of zeros and ones in the binary representation.13 Random Number Functions The random number functions of GMP come in two groups. void mpz_random2 (mpz t rop. Use mpz_rrandomb instead. The variable state must be initialized by calling one of the gmp_randinit functions (Section 9. page 65) before invoking this function. This function is obsolete. inclusive. The generated random number doesn’t satisfy any particular requirements of randomness. size t nails. const void *op ) Set rop from an array of word data at op. gmp randstate t state. Within each . Useful for testing functions and algorithms. This function is obsolete. each size bytes. Useful for testing functions and algorithms. The random number will be in the range 0 to 2n − 1.40 GNU MP 5.1 [Random State Initialization]. void mpz_import (mpz t rop.0. int order. older function that rely on a global state. 5. size t count. Negative random numbers are generated when max size is negative.1 [Random State Initialization]. [Function] The parameters specify the format of the data. mp size t max_size ) [Function] Generate a random integer of at most max size limbs.
The nails feature can account for this. any address is allowed. sizeof(a[0]).Chapter 5: Integer Functions 41 word endian can be 1 for most significant byte first. which avoids any portability problems with malloc(0). most significant element first. rop will simply be a positive integer. When an application is allocating space itself the required size can be determined with a calculation like the following. Within each word endian can be 1 for most significant byte first. which is usually true. count = (mpz_sizeinbase (z. this can be 0 to use the full words. int order. mpz t op ) Fill rop with word data from op. 2) + numb1) / numb. or 0 for the native endianness of the host CPU. page 86). In either case the return value is the destination used. int endian.nail. this can be 0 to produce full words. 0. page 37) There are no data alignment restrictions on rop. void * mpz_export (void *rop. just the absolute value is exported. unsigned long a[20]. by passing for instance 8*sizeof(int)INT_BIT. 20. or 0 for the native endianness of the host CPU. numb = 8*size . /* Initialize z and a */ mpz_import (z. The most significant nails bits of each word are skipped. and certainly true for unsigned long everywhere we know of. Here’s an example converting an array of unsigned long data. just NULL is returned. size t size. 1 for least significant first. though if z is zero no space at all is actually needed (or written). 0. If op is zero then the count returned will be zero and nothing written to rop. (see Section 5. either rop or the allocated block. any address is allowed. Since mpz_sizeinbase always returns at least 1. and apply it for instance with mpz_neg. 1 for least significant first. However on Cray vector systems it may be noted that short and int are always stored in 8 bytes (and with sizeof indicating that) but use only 32 or 46 bits. or countp can be NULL to discard the count. An application can handle any sign itself. The number of words produced is written to *countp . Each word will be size bytes and order can be 1 for most significant word first or 1 for least significant first.10 [Integer Comparisons]. size t *countp. or if rop is NULL then a result array of the necessary size is allocated using the current GMP allocation function (see Chapter 14 [Custom Allocation]. and host byte order within each value. The sign of op is ignored. An application can use mpz_sgn to get the sign and handle it as desired. . There are no data alignment restrictions on op. The most significant nails bits of each word are unused and set to zero. This example assumes the full sizeof bytes are used for data in the given type. 1. size t nails. count here will be at least one. There is no sign taken from the data. If rop is NULL in this case. rop must have enough space for the data. p = malloc (count * size). If op is nonzero then the most significant word produced will be nonzero. a). [Function] The parameters specify the format of the data produced. no block is allocated.
42 GNU MP 5. signed int. unlike the normal mpz_init. In normal programs this function is not recommended.15 Miscellaneous Functions int int int int int int [Function] [Function] [Function] [Function] [Function] [Function] Return nonzero iff the value of op fits in an unsigned long int. Most applications will not need them. The result will be either exact or 1 too big. signed long int. The right amount of allocation is normally two more than the value returned by mpz_sizeinbase. Fixed space of fixed num bits is allocated to each of the array size integers in integer array.2) can be used to locate the most significant 1 bit in op. unsigned short int.) 5. 20000. See Section 5. return zero. The following space requirements apply to various routines. one extra for a minus sign and one for the nullterminator. mp size t fixed_num_bits ) [Function] This is a special type of initialization. The sign of op is ignored. counting from 1. Don’t call mpz_clear! The integer array parameter is the first mpz_t in the array. mpz_array_init (arr[0]. unsigned int. These macros evaluate their argument more than once. respectively. If op is zero the return value is always 1. so an application must ensure it is sufficient for any value stored. respectively. This function is only intended for programs that create a large number of integers and need to reduce memory usage by avoiding the overheads of allocating and reallocating lots of small blocks. It will be noted that mpz_sizeinbase(op. the result is always exact.1 5. int base ) [Function] Return the size of op measured in number of digits in the given base. [Macro] [Macro] Determine whether op is odd or even.0. or signed short int.11 [Logical and Bit Manipulation Functions]. mpz_t arr[20000]. There is no way to free the storage allocated by this function. mp size t array_size.16 Special Functions The functions in this section are for various special purposes. Return nonzero if yes. page 38. zero if no. base can vary from 2 to 62. . 512). The space allocated to each integer by this function will not be automatically increased. mpz_fits_ulong_p (mpz t op ) mpz_fits_slong_p (mpz t op ) mpz_fits_uint_p (mpz t op ) mpz_fits_sint_p (mpz t op ) mpz_fits_ushort_p (mpz t op ) mpz_fits_sshort_p (mpz t op ) int mpz_odd_p (mpz t op ) int mpz_even_p (mpz t op ) size_t mpz_sizeinbase (mpz t op. (Unlike the bitwise functions which start from 0. This function can be used to determine the space required when converting op to a string. Otherwise. just the absolute value is used. If base is a power of 2. For example. void mpz_array_init (mpz t integer_array.
Chapter 5: Integer Functions 43 • mpz_abs. the returned value will be zero. but each rounded up to a multiple of mp_bits_per_limb. • mpz_add. If op is zero. mpz_realloc2 is the preferred way to accomplish allocation changes like this. • mpz_mul. For other functions. mp_limb_t mpz_getlimbn (mpz t op. mpz_getlimbn returns zero if n is outside the range 0 to mpz_size(op )1. mp size t n ) [Function] Return limb number n from op. mpz_realloc2 and _mpz_realloc are the same except that _mpz_realloc takes its size in limbs. void * _mpz_realloc (mpz t integer. mpz_set. mpz_size can be used to find how many limbs make up op. mpz_sub and mpz_sub_ui need room for the larger of the two operands. The value in integer is preserved if it fits. The sign of op is ignored. The return value is not useful to applications and should be ignored. mpz_set_si and mpz_set_ui need room for the value they store. or is set to 0 if not. mpz_mul_ui and mpz_mul_ui need room for the sum of the number of bits in their operands. mp size t new_alloc ) [Function] Change the space for integer to new alloc limbs. the suggestion is to calculate in a regular mpz_init variable and copy the result to an array variable with mpz_set. The least significant limb is number 0. mpz_add_ui. . just the absolute value is used. or if in doubt. mpz_neg. but not between an array and a normal variable. size_t mpz_size (mpz t op ) [Function] Return the size of op measured in number of limbs. • mpz_swap can be used between two array variables. plus an extra mp_bits_per_limb.
Note that if op1 and op2 have common factors. and canonicalize their result. Make sure to call this function for all mpq_t variables when you are done with them.. rop has to be passed to mpq_canonicalize before any operations are performed on rop. page 30). and that the denominator is positive. and is simply ignored. [Function] void mpq_inits (mpq t x.0.1 6 Rational Number Functions This chapter describes the GMP functions for performing arithmetic on rational numbers. The string can be an integer like “41” or a fraction like “41/152”.) Initialize a NULLterminated list of mpq_t variables. . [Function] [Function] void mpq_set_ui (mpq t rop. or at least cleared out (using the function mpq_clear) between each initialization. Pure assignment functions do not canonicalize the assigned variable.44 GNU MP 5.2 [Assigning Integers]. mpq t op ) void mpq_set_z (mpq t rop.. The base can vary from 2 to 62. [Function] int mpq_set_str (mpq t rop. It is the responsibility of the user to canonicalize the assigned variable before any arithmetic operations are performed on that variable. and set their values to 0/1.) Free the space occupied by a NULLterminated list of mpq_t variables. Rational numbers are stored in objects of type mpq_t. . void mpq_set (mpq t rop. or if not then mpq_canonicalize must be called. 6. [Function] void mpq_clears (mpq t x. and make the denominator positive. Each variable should normally only be initialized once. signed long int op1. unsigned long int op1. unsigned long int op2 ) [Function] [Function] Set the value of rop to op1/op2. The canonical from means that the denominator and the numerator have no common factors.. The numerator and optional denominator are parsed the same as in mpz_set_str (see Section 5. unsigned long int op2 ) void mpq_set_si (mpq t rop.1 Initialization and Assignment Functions void mpq_init (mpq t x ) [Function] Initialize x and set it to 0/1. The fraction must be in canonical form (see Chapter 6 [Rational Number Functions]. or if base is 0 then the leading characters are used: . void mpq_clear (mpq t x ) [Function] Free the space occupied by x. mpz t op ) Assign rop from op. All rational arithmetic functions assume operands have a canonical form. char *str. White space is allowed in the string. These functions start with the prefix mpq_. void mpq_canonicalize (mpq t op ) [Function] Remove any factors that are common to the numerator and denominator of op. int base ) Set rop from a nullterminated string str in the given base. page 44). Zero has the unique representation 0/1..
mpq t minuend. that being exactly enough for the string and nullterminator. mpq t op ) [Function] Convert op to a string of digits in base base. page 86). If str is NULL. [Function] void mpq_div (mpq t quotient. truncating if necessary (ie. Hardware overflow. double op ) void mpq_set_f (mpq t rop. whereas 0xEF/0x100 is 239/256. mpq t addend2 ) Set sum to addend1 + addend2. or decimal otherwise. For too big an infinity is returned when available. it should point to a block of storage large enough for the result. 0 for octal. mpq t rop2 ) Swap the values rop1 and rop2 efficiently. int base. so for instance 0xEF/100 is 239/100. possible slash.3 Arithmetic Functions void mpq_add (mpq t sum. or if the denominator is 1 then just ‘num’. There is no rounding. mpq t subtrahend ) Set difference to minuend − subtrahend. [Function] void mpq_set_d (mpq t rop. If the exponent from the conversion is too big or too small to fit a double then the result is system dependent. rounding towards zero). mp bitcnt t op2 ) Set rop to op1 × 2op2 . void mpq_swap (mpq t rop1. Note that this is done separately for the numerator and denominator. The string will be of the form ‘num/den’. The base may vary from 2 to 36. [Function] void mpq_mul (mpq t product. 0b or 0B for binary.Chapter 6: Rational Number Functions 45 0x or 0X for hex. The return value is 0 if the entire string is a valid number. mpq t dividend. mpq t multiplier. For too small 0. A pointer to the result string is returned. this conversion is exact. [Function] . and the nullterminator. mpf t op ) Set rop to the value of op. that being mpz_sizeinbase (mpq_numref(op ). [Function] 6. the result string is allocated using the current allocation function (see Chapter 14 [Custom Allocation]. base ) + 3 The three extra bytes are for a possible minus sign. The block will be strlen(str)+1 bytes. If str is not NULL. [Function] void mpq_mul_2exp (mpq t rop. 6. mpq t divisor ) Set quotient to dividend/divisor. base ) + mpz_sizeinbase (mpq_denref(op ). or −1 if not. mpq t op1. mpq t addend1. or the given str. being either the allocated block. [Function] [Function] char * mpq_get_str (char *str. [Function] void mpq_sub (mpq t difference.2 Conversion Functions double mpq_get_d (mpq t op ) Convert op to a double. mpq t multiplicand ) Set product to multiplier × multiplicand.0 is normally returned. underflow and denorm traps may or may not occur.
unsigned long int den2 ) [Macro] [Macro] Compare op1 and num2/den2. Return a positive value if op1 > op2. int mpq_equal (mpq t op1. [Function] void mpq_neg (mpq t negated_operand. void mpq_abs (mpq t rop. there are few functions for either input or output.0. Note that if an assignment to the numerator and/or denominator could take an mpq_t out of the canonical form described at the start of this chapter (see Chapter 6 [Rational Number Functions]. zero if op1 = num2/den2. mpz_t mpq_numref (mpq t op ) mpz_t mpq_denref (mpq t op ) [Macro] [Macro] Return a reference to the numerator and denominator of op. [Function] void mpq_inv (mpq t inverted_number. mpq t op2 ) [Function] Return nonzero if op1 and op2 are equal. 6. this function is much faster. To determine if two rationals are equal. mp bitcnt t op2 ) Set rop to op1/2 op2 [Function] . and a negative value if op1 < op2. [Macro] This function is actually implemented as a macro. zero if op1 = op2. and a negative value if op1 < num2/den2. The following functions give direct access to the numerator and denominator of an mpq_t. 6. and −1 if op < 0.46 GNU MP 5. int mpq_sgn (mpq t op ) Return +1 if op > 0.4 Comparison Functions int mpq_cmp (mpq t op1. num2 and den2 are allowed to have common factors. this routine will divide by zero. In particular. long int num2. 0 if op = 0. mpq t op1. The mpz functions can be used on the result of these macros. mpq_equal is faster than mpq_cmp.1 void mpq_div_2exp (mpq t rop. unsigned long int num2. Although mpq_cmp can be used for the same purpose. unsigned long int den2 ) int mpq_cmp_si (mpq t op1. mpq t op2 ) [Function] Compare op1 and op2. zero if they are nonequal. mpq t number ) [Function] Set inverted number to 1/number. .5 Applying Integer Functions to Rationals The set of mpq functions is quite small. These functions are implemented as a macros and evaluate their arguments multiple times. page 44) then mpq_canonicalize must be called before any other mpq functions are applied to that mpq_t. If the new denominator is zero. Return a positive value if op1 > num2/den2. It evaluates its arguments multiple times. mpq t operand ) Set negated operand to −operand. respectively. int mpq_cmp_ui (mpq t op1. mpq t op ) Set rop to the absolute value of op.
size_t mpq_out_str (FILE *stream. Passing a NULL pointer for a stream argument to any of these functions will make them read from stdin and write to stdout. The leading characters are examined separately for the numerator and denominator of a fraction.h’ to define prototypes for these functions. If the input might not be in canonical form. Output is in the form ‘num/den’ or if the denominator is 1 then just ‘num’. Any initial whitespace characters are read and discarded. mpq_get_num mpq_get_den mpq_set_num mpq_set_den 6. or can be 0 in which case the leading characters of the string determine the base.6 Input and Output Functions When using any of these functions. mpq t rational ) [Function] (mpq t rational. int base. as a string of digits in base base. FILE *stream.Chapter 6: Rational Number Functions 47 void void void void (mpz t numerator. then mpq_canonicalize must be called (see Chapter 6 [Rational Number Functions]. The input can be a fraction like ‘17/63’ or just an integer like ‘123’. Direct use of mpq_numref or mpq_denref is recommended instead of these functions. ‘0x’ or ‘0X’ for hexadecimal. so for instance ‘0x10/11’ is 16/11. page 44). ‘0’ for octal. mpz t denominator ) [Function] Get or set the numerator or denominator of a rational.h’ before ‘gmp. mpz t numerator ) [Function] (mpq t rational. or decimal otherwise. . Reading stops at the first character not in this form. and white space is not permitted within the string. size_t mpq_inp_str (mpq t rop. The base may vary from 2 to 36. mpq t op ) [Function] Output op on stdio stream stream. or if an error occurred. Return the number of characters read (including white space). int base ) [Function] Read a string of digits from stream and convert them to a rational in rop. The base can be between 2 and 36. mpq t rational ) [Function] (mpz t denominator. or 0 if a rational could not be read. Return the number of bytes written. since that will allow ‘gmp. respectively. whereas ‘0x10/0x11’ is 16/17. These functions are equivalent to calling mpz_set with an appropriate mpq_numref or mpq_denref. return 0.h’. it’s a good idea to include ‘stdio.
. Currently this means rounding up to a whole limb. limited only by available memory. The exponent of each float is a fixed precision. or perhaps rationals. An mpf_t object must be initialized before storing the first value in it.48 GNU MP 5. In particular results obtained on one computer often differ from the results on a computer with a different word size. as might be imagined from the fact precisions are expressed in bits. Note that the mpf functions are not intended as a smooth extension to IEEE P754 arithmetic. In the current implementation the exponent is a count of limbs. This means that if a float is exactly represented in only a few bits then only those bits will be used in a calculation. [Function] mp_bitcnt_t mpf_get_default_prec (void) Return the default precision actually used. The precision selected for a variable is a minimum value. But applications shouldn’t be concerned by such details. GMP may increase it a little to facilitate efficient calculation.0.1 cannot be represented exactly. Each variable has its own precision. Each function is defined to calculate with “infinite precision” followed by a truncation to the destination precision. and applications must take care not to overflow the exponent or results will be unpredictable. and then sometimes having a further partial limb. Note however mpf_get_str can only return an exponent which fits an mp_exp_t and currently mpf_set_str doesn’t accept exponents bigger than a long. All subsequent calls to mpf_init will use this precision. 7.1 7 Floatingpoint Functions GMP floating point numbers are stored in objects of type mpf_t and functions operating on them have an mpf_ prefix. Each variable keeps a size for the mantissa data actually in use. This might change in a future release. even if the selected precision is high. The mantissa of each float has a userselectable precision. but of course the work done is only what’s needed to determine a result under that definition. One consequence of this is that decimal fractions like 0. so for example on a 32bit system this means a range of roughly 2−68719476768 to 268719476736 . The same is true of plain IEEE double floats.) mpf functions and variables have no special notion of infinity or notanumber. (Suitably scaled integers. All calculations are performed to the precision of the destination variable. The mantissa in stored in binary. but previously initialized variables are unaffected. are better choices. one machine word on most systems. The functions mpf_init and mpf_init2 are used for that purpose. or on a 64bit system this will be greater. This makes both highly unsuitable for calculations involving money or other values that should be exact decimal fractions. depending on the high limb of the mantissa.1 Initialization Functions void mpf_set_default_prec (mp bitcnt t prec ) [Function] Set the default precision to be at least prec bits. and that can be increased or decreased at any time.
. and in particular if it had a higher precision than prec it will retain that higher precision. /* Unless the program is about to exit. .. Normally.. A typical use would be for adjusting the precision gradually in iterative algorithms like NewtonRaphson. mpf_init (x). prec must be no more than the allocated precision for rop. [Function] Free the space occupied by x. */ mpf_clear (x). making the computation precision closely match the actual accurate part of the numbers. New values written to rop will use the new prec. a variable should be initialized once only or at least be cleared. or in the most recent mpf_set_prec.. between initializations. do . mpf_clear (y).Chapter 7: Floatingpoint Functions 49 void mpf_init (mpf t x ) [Function] Initialize x to 0. mp bitcnt t prec ) void mpf_inits (mpf t x. Make sure to call this function for all mpf_t variables when you are done with them. without changing the memory allocated. using mpf_clear. using mpf_clear. [Function] void mpf_clear (mpf t x ) void mpf_clears (mpf t x. Here is an example on how to initialize floatingpoint variables: { mpf_t x. between initializations. and set their values to 0. and so should not be used in a tight loop. mp bitcnt t prec ) [Function] Set the precision of rop to be at least prec bits. The value in rop is unchanged.) [Function] Initialize a NULLterminated list of mpf_t variables. void mpf_init2 (mpf t x. that being the precision when rop was initialized. a variable should be initialized once only or at least be cleared. mp bitcnt t prec ) [Function] Set the precision of rop to be at least prec bits. in bits. } The following three functions are useful for changing the precision during a calculation. y.. mp_bitcnt_t mpf_get_prec (mpf t op ) Return the current precision of op.. [Function] Initialize x to 0 and set its precision to be at least prec bits. . void mpf_set_prec_raw (mpf t rop.) Free the space occupied by a NULLterminated list of mpf_t variables. The precision of x is undefined unless a default precision has already been established by a call to mpf_set_default_prec. [Function] void mpf_set_prec (mpf t rop.. Normally. /* precision at least 256 bits */ .. The precision of the initialized variables is undefined unless a default precision has already been established by a call to mpf_set_default_prec. /* use default precision */ mpf_init2 (y. 256). The value in rop will be truncated to the new precision. . This function requires a call to realloc.
1 Before calling mpf_clear or the full mpf_set_prec. For bases up to 36. mpf t rop2 ) [Function] Swap rop1 and rop2 efficiently.35 while lowercase letter represent 36. alternatively ‘MeN’. ‘M’ is the mantissa and ‘N’ is the exponent.2 Assignment Functions These functions assign new values to already initialized floats (see Section 7. Negative values are used to specify that the exponent is in decimal. such as after a minus sign or in the exponent. char *str. mpq t op ) [Function] [Function] [Function] [Function] [Function] [Function] Set the value of rop from op. Both the values and the precisions of the two variables are swapped. in decimal. case is ignored. unsigned long int op ) mpf_set_si (mpf t rop. since that makes a lot of sense. making it fail when there is any whitespace in the input.50 GNU MP 5. int base ) [Function] Set the value of rop from the string in str. . mpf_get_prec can be used before mpf_set_prec_raw to get the original allocated precision. int mpf_set_str (mpf t rop. Do you really want it to accept "3 14" as meaning 314 as it does now?] This function returns 0 if the entire string is a valid number in base base. page 48). mpf t op ) mpf_set_ui (mpf t rop. or just to use various different precisions for different purposes during a calculation. After mpf_set_prec_raw it reflects the prec value set. void void void void void void mpf_set (mpf t rop. the base will not be determined from the leading characters of the string if base is 0. The exponent is either in the specified base or. This is so that numbers like ‘0. The string is of the form ‘M@N’ or. White space is allowed in the string. Otherwise it returns −1. The mantissa is always in the specified base. but not in other places. [This is not really true. We are considering changing the definition of this function. for bases 37 to 62. and is simply ignored. 7. on systems providing localeconv. another mpf_set_prec_raw call must be made to restore rop to its original allocated precision. uppercase and lowercase letters have the same value. or −62 to −2. uppercase letter represent the usual 10.. The decimal point expected is taken from the current locale. mpf_set_prec_raw is an efficient way to use an mpf_t variable at different precisions during a calculation.1 [Initializing Floats].. Unlike the corresponding mpz function. Failing to do so will have unpredictable results. if the base is 10 or less. The argument base may be in the ranges 2 to 62. perhaps to gradually increase precision in an iteration. double op ) mpf_set_z (mpf t rop. Please tell us your opinion about this change. whitespace is ignored in the beginning of the string and within the mantissa. void mpf_swap (mpf t rop1.0.61. signed long int op ) mpf_set_d (mpf t rop.23’ are not interpreted as octal. if base is negative. mpz t op ) mpf_set_q (mpf t rop.
If op is zero.. underflow and denorm traps may or may not occur. long mpf_get_si (mpf t op ) unsigned long mpf_get_ui (mpf t op ) [Function] [Function] Convert op to a long or unsigned long. . This is similar to the standard C frexp function (see Section “Normalization Functions” in The GNU C Library Reference Manual).4 Conversion Functions double mpf_get_d (mpf t op ) Convert op to a double. (I. truncating if necessary (ie.. d ∗ 2exp is the (truncated) op value.5 ≤ d < 1 and the exponent is stored to *exp . signed long int op ) mpf_init_set_d (mpf t rop. it can be used as the source or destination operand for the ordinary float functions. GMP provides a parallel series of initializeandset functions which initialize the output and then store the value there.0 is normally returned..Chapter 7: Floatingpoint Functions 51 7. page 54). The return value is in the range 0. For too small 0. truncating any fraction part.. These functions’ names have the form mpf_init_set. If op is too big for the return type. you have to call mpf_clear for it. Don’t use an initializeandset function on a variable already initialized! void void void void mpf_init_set (mpf t rop. and with an exponent returned separately. Note that rop is initialized even if an error occurs.. the result is undefined. int mpf_init_set_str (mpf t rop. the return is 0.3 Combined Initialization and Assignment Functions For convenience. Hardware overflow. See mpf_set_str above for details on the assignment operation. If the exponent in op is too big or too small to fit a double then the result is system dependent. 7. char *str. The precision of rop will be taken from the active default precision.0 and 0 is stored to *exp . double op ) [Function] [Function] [Function] [Function] Initialize rop and set its value from op. rounding towards zero). mpf t op ) [Function] Convert op to a double. int base ) [Function] Initialize rop and set its value from the string in str. See also mpf_fits_slong_p and mpf_fits_ulong_p (see Section 7. unsigned long int op ) mpf_init_set_si (mpf t rop. rounding towards zero). functions. as set by mpf_set_ default_prec. Once the float has been initialized by any of the mpf_init_set. [Function] double mpf_get_d_2exp (signed long int *exp. mpf t op ) mpf_init_set_ui (mpf t rop.8 [Miscellaneous Float Functions].) The precision of rop will be taken from the active default precision. as set by mpf_set_ default_prec. truncating if necessary (ie. For too big an infinity is returned when available.e.
and lowercase letters (in that significance order) are used. [Function] [Function] Division is undefined if the divisor is zero. void mpf_div (mpf t rop. mpf t op2 ) void mpf_sub_ui (mpf t rop. No more digits than can be accurately represented by op are ever generated. mpf t op1. The generated string is a fraction. it should point to a block of n digits + 2 bytes. mpf t op ) [Function] Convert op to a string of digits in base base. When op is zero. unsigned long int op2 ) Set rop to op1 × op2. Up to n digits digits will be generated. digits and lowercase letters are used. uppercase letters. a possible minus sign. mpf t op2 ) void mpf_div_ui (mpf t rop. int base.. For base in the range 2. unsigned long int op1. unsigned long int op2 ) Set rop to op1/op2. When n digits is 0 to get all significant digits. the result string is allocated using the current allocation function (see Chapter 14 [Custom Allocation].−36. and passing a zero divisor to the divide functions will make these functions intentionally divide by zero. If str is NULL. mpf t op2 ) void mpf_ui_sub (mpf t rop. Trailing zeros are not returned. being either the allocated block or the given str. The base argument may vary from 2 to 62 or from −2 to −36.62. mpf t op2 ) void mpf_mul_ui (mpf t rop. that being exactly enough for the string and nullterminator. the number 3. mpf t op2 ) void mpf_add_ui (mpf t rop. For example. If n digits is 0 then that accurate maximum number of digits are generated.0. mpf t op1. mpf t op1. for 37. mpf t op1. unsigned long int op2 ) Set rop to op1 + op2. with an implicit radix point immediately to the left of the first digit. mpf t op1. 7. The block will be strlen(str)+1 bytes. digits and uppercase letters are used. and str should be NULL in that case. [Function] [Function] [Function] void mpf_sqrt (mpf t rop. an application won’t be able to know the space required. mpf t op ) void mpf_sqrt_ui (mpf t rop. page 86). The applicable exponent is written through the expptr pointer. unsigned long int op2 ) Set rop to op1 − op2. that being enough for the mantissa. mpf t op1. mpf t op1. mpf t op2 ) void mpf_ui_div (mpf t rop. an empty string is produced and the exponent returned is 0. digits.36. unsigned long int op1..1 char * mpf_get_str (char *str. size t n_digits. and a nullterminator. [Function] [Function] void mpf_sub (mpf t rop. unsigned long int op ) Set rop to √ op. mp exp t *expptr.52 GNU MP 5. [Function] [Function] [Function] void mpf_mul (mpf t rop. [Function] [Function] . A pointer to the result string is returned.1416 would be returned as string "31416" and exponent 1. mpf t op1. This lets the user handle arithmetic exceptions in these functions in the same manner as other arithmetic exceptions. for −2. If str is not NULL..5 Arithmetic Functions void mpf_add (mpf t rop.
[Function] void mpf_div_2exp (mpf t rop. test if op1 and op2 are approximately equal.111 and XX100. 0 if op = 0. mpf t op2. mpf t op1. mpf_cmp (mpf t op1. zero if op1 = op2. meaning sometimes more than op3 bits. is replaced by a semiinfinite number of bits. double op2 ) mpf_cmp_ui (mpf t op1. even if . and should be considered equal. mpf t op ) Set rop to −op. [Function] void mpf_neg (mpf t rop.. This function is actually implemented as a macro. Caution 2: This function will consider XXX11. [Function] void mpf_mul_2exp (mpf t rop. When using any of these functions. it is a good idea to include ‘stdio. mpf t op2 ) [Function] Compute the relative difference between op1 and op2 and store the result in rop. mp bitcnt t op2 ) Set rop to op1/2op2 . since that will allow ‘gmp. and functions that output to a stdio stream. mpf t op ) Set rop to the absolute value of op. This is op1 − op2/op1. unsigned long int op2 ) mpf_cmp_si (mpf t op1. It evaluates its arguments multiple times. mp bitcnt t op3) [Function] Return nonzero if the first op3 bits of op1 and op2 are equal. [Function] 7.h’. I.Chapter 7: Floatingpoint Functions 53 void mpf_pow_ui (mpf t rop. sometimes fewer. respectively. Return a positive value if op1 > op2. Caution 1: All version of GMP up to version 4. zero otherwise... void mpf_abs (mpf t rop.000 different. 7. unsigned long int op2 ) Set rop to op1 op2 [Function] . mpf t op1..h’ to define prototypes for these functions. signed long int op2 ) mpf_cmp_d can be called with an infinity.7 Input and Output Functions Functions that perform input from a stdio stream. int mpf_eq (mpf t op1.2.. mpf t op1. Passing a NULL pointer for a stream argument to any of these functions will make them read from stdin and write to stdout.4 compared just whole limbs. and −1 if op < 0.. . mpf t op2 ) mpf_cmp_d (mpf t op1. void mpf_reldiff (mpf t rop.h’ before ‘gmp.. but results are undefined for a NaN. Such numbers are really just one ulp off. and a negative value if op1 < op2.6 Comparison Functions int int int int [Function] [Function] [Function] [Function] Compare op1 and op2. [Macro] int mpf_sgn (mpf t op ) Return +1 if op > 0.e. mp bitcnt t op2 ) Set rop to op1 × 2op2 . mpf t op1.
‘M’ is the mantissa and ‘N’ is the exponent. alternatively ‘MeN’. size_t mpf_inp_str (mpf t rop. on systems providing localeconv. The decimal point follows the current locale. with nbits significant bits in the mantissa. for 37. FILE *stream.1 size_t mpf_out_str (FILE *stream. Unlike the corresponding mpz function. and put the read float in rop. if base is negative. or if an error occurred. digits. which may vary from 2 to 62 or from −2 to −36. int base.. size t n_digits. This is so that numbers like ‘0. mpf t op ) void mpf_floor (mpf t rop. in decimal. uppercase letters. return 0. . gmp randstate t state. 7. int base ) [Function] Read a string in base base from stream. [Function] int mpf_integer_p (mpf t op ) Return nonzero if op is an integer. [Function] mpf_fits_ulong_p (mpf t op ) mpf_fits_slong_p (mpf t op ) mpf_fits_uint_p (mpf t op ) mpf_fits_sint_p (mpf t op ) mpf_fits_ushort_p (mpf t op ) mpf_fits_sshort_p (mpf t op ) void mpf_urandomb (mpf t rop.−36.62. mpf t op ) [Function] Print op to stream. The exponent is always in decimal. digits and lowercase letters are used. such that 0 ≤ rop < 1. mpf_ceil rounds to the next higher integer. when truncated to an integer. if the base is 10 or less. Up to n digits will be printed from the mantissa. and lowercase letters (in that significance order) are used.23’ are not interpreted as octal. and mpf_trunc to the integer towards zero. or if the base is greater than 10 then by an ‘@’. as a string of digits. The argument base may be in the ranges 2 to 36. Return the number of bytes read. The mantissa is prefixed with an ‘0. The decimal point expected is taken from the current locale. or if an error occurred. except that no more digits than are accurately representable by op will be printed. mpf t op ) void mpf_trunc (mpf t rop.. Negative values are used to specify that the exponent is in decimal. Return the number of bytes written. The mantissa is always in the specified base. mp bitcnt t nbits ) Generate a uniformly distributed random float in rop.54 GNU MP 5. The exponent is either in the specified base or.8 Miscellaneous Functions void mpf_ceil (mpf t rop.’ and is in the given base. mpf_floor to the next lower. or −36 to −2. mpf t op ) [Function] [Function] [Function] Set rop to op rounded to an integer.36.. the base will not be determined from the leading characters of the string if base is 0. For base in the range 2. for −2. n digits can be 0 to select that accurate maximum. int int int int int int [Function] [Function] [Function] [Function] [Function] [Function] Return nonzero if op would fit in the respective C data type. An exponent is then printed. on systems providing localeconv. return 0. separated by an ‘e’. digits and uppercase letters are used.0. The string is of the form ‘M@N’ or.
Negative random numbers are generated when max size is negative. mp exp t exp ) [Function] Generate a random float of at most max size limbs. since these kind of random numbers have proven to be more likely to trigger cornercase bugs. . The exponent of the number is in the interval −exp to exp (in limbs).1 [Random State Initialization]. void mpf_random2 (mpf t rop.Chapter 7: Floatingpoint Functions 55 The variable state must be initialized by calling one of the gmp_randinit functions (Section 9. with long strings of zeros and ones in the binary representation. mp size t max_size. page 65) before invoking this function. This function is useful for testing functions and algorithms.
This example adds the number beginning at s1p and the number beginning at s2p and writes the sum at destp. mp size t n ) [Function] Add {s1p. {s1p. but there are variations that make them hard to use. This is the lowestlevel function for addition. since it is written in assembly for most CPUs. mp_limb_t mpn_add (mp limb t *rp. mp size t n. s2n}.56 GNU MP 5. inplace operations are allowed where source and destination are the same. All areas have n limbs. not to provide a coherent calling interface. a source operand is identified by the pointer to the least significant limb. either 0 or 1. These functions start with the prefix mpn_. and write the n least significant limbs of the result to rp. and the limb count in braces. but also intended for timecritical user code. The different functions have somewhat similar interfaces. and write the n least significant limbs of the result to rp. s1n} and {s2p. On random data such cases will be unlikely and it’d be wasteful for every function to check every time. mp size t s1n.. s1n}. In the notation used below. Return carry. const mp limb t *s1p. A common requirement for all functions is that each source area needs at least one limb. s2p. mpf_. It is the responsibility of the caller to ensure that the destination has enough space for storing the result. A destination operand is specified by just a pointer. No size argument may be zero. mp_limb_t mpn_add_n (mp limb t *rp. For addition of a variable to itself (i. so that no time is spent on things that not all callers need. const mp limb t *s2p. const mp limb t *s2p. s1p. cy = mpn_add_n (destp. either 0 or 1. The mpn functions are designed to be as fast as possible.0. either 0 or 1. and mpq_ functions. const mp limb t *s1p. used to implement the highlevel GMP functions. and store the result into a subrange of a destination. n}.1 8 Lowlevel Functions This chapter describes lowlevel GMP functions. The mpn functions are the base for the implementation of the mpz_. or other special forms. it is possible to perform computations on subranges of an argument.e. Return carry. n) It should be noted that the mpn functions make no attempt to identify high or low zero limbs on their operands. s1p equals s2p) use mpn_lshift with a count of 1 for optimal speed. These functions do as little as possible apart from the real multiple precision computation. const mp limb t *s1p. For example. n} and {s2p. n} and s2limb. Return carry. mp_limb_t mpn_add_1 (mp limb t *rp. A source operand is specified by a pointer to the least significant limb and a limb count. but not where they only partly overlap. An application knowing something about its data can take steps to trim or perhaps split its calculations. It is the preferred function for addition. With this way of specifying operands. mp size t s2n ) [Function] Add {s1p. . and write the s1n least significant limbs of the result to rp. mp limb t s2limb ) [Function] Add {s1p. Unless otherwise stated.
mp_limb_t mpn_mul (mp limb t *rp. Return borrow. n}. const mp limb t *s2p. even if the product’s most significant limb is zero. If the two input operands are the same. either 0 or 1. const mp limb t *s1p. n} and {rp. and write the 2*nlimb result to rp. mp_limb_t mpn_sub_n (mp limb t *rp. Return carryout. No overlap is permitted between the destination and the source. and write the result to {rp. mp limb t s2limb ) [Function] Subtract s2limb from {s1p. Return the most significant limb of the result. s1n} and {s2p. n} from {s1p. [Function] The destination has to have space for 2*n limbs. either 0 or 1. and write the n least significant limbs of the product to rp. The destination has to have space for s1n + s2n limbs. mp size t s2n ) [Function] Multiply {s1p. const mp limb t *s1p. [Function] void mpn_mul_n (mp limb t *rp. const mp limb t *s1p. {s1p. mp size t n ) Compute the square of {s1p. even if the result’s most significant limb is zero. either 0 or 1.Chapter 8: Lowlevel Functions 57 This function requires that s1n is greater than or equal to s2n. This function requires that s1n is greater than or equal to s2n. This is the lowestlevel function for subtraction. mp size t s1n. This function requires that s1n is greater than or equal to s2n. mp size t s1n. const mp limb t *s2p. const mp limb t *s1p. const mp limb t *s1p. mp size t n ) Multiply {s1p. mp size t n. const mp limb t *s2p. n} are allowed to overlap provided rp ≤ s1p. mp size t n ) [Function] Perform the negation of {sp. mp_limb_t mpn_mul_1 (mp limb t *rp. Return the most significant limb of the product. n} by s2limb. mp_limb_t mpn_sub_1 (mp limb t *rp. n}. n}. n} and write the 2*nlimb result to rp. . void mpn_neg (mp limb t *rp. since it is written in assembly for most CPUs. and write the s1n least significant limbs of the result to rp. const mp limb t *sp. mp size t n. s2n} from {s1p. void mpn_sqr (mp limb t *rp. mp limb t s2limb ) [Function] Multiply {s1p. No overlap is permitted between the destination and either source. even if the product’s most significant limb is zero. n}. The destination has to have space for 2*n limbs. s1n}. const mp limb t *s1p. and write the n least significant limbs of the result to rp. Return borrow. No overlap is permitted between the destination and either source. mp size t s2n ) [Function] Subtract {s2p. const mp limb t *s1p. s2n}. const mp limb t *s2p. n} and {s2p. mp_limb_t mpn_sub (mp limb t *rp. and write the n least significant limbs of the result to rp. use mpn_sqr. and write the (s1n+s2n)limb result to rp. Return borrow. mp size t n ) [Function] Subtract {s2p. It is the preferred function for subtraction. n}.
The remainder replaces the dividend at rs2p. It is written in assembly for most CPUs. nn} by {dp. mp limb t *rp. mp size t rs2n. n} and s2limb. It is required that the most significant bit of the divisor is set. mp limb t *rs2p. n} and write the result to rp. This is a lowlevel function that is a building block for general multiplication as well as other operations in GMP. mp_limb_t mpn_divrem (mp limb t *r1p. and stored after the integral limbs. If the quotient is not needed. n} and write the result to rp. mp size t qxn. dn} and put the quotient at {qp. except that np might equal rp. which is returned. It is required that rs2n is greater than or equal to s3n. The dividend size nn must be greater than or equal to divisor size dn. The area at r1p needs to be rs2n − s3n + qxn limbs large. qxn will be zero. const mp limb t *dp.e. const mp limb t *s3p. plus borrowout from the subtraction.1 This is a lowlevel function that is a building block for general multiplication as well as other operations in GMP. Please call mpn_tdiv_qr instead for best performance. Don’t call this function if s2limb is a power of 2.. mp limb t s2limb ) [Function] Multiply {s1p. Return the most significant limb of the quotient.0. use mpn_lshift with a count equal to the logarithm of s2limb instead. The qxn operand must be zero. Aside from that special case. mp size t n. n} and s2limb. Return the most significant limb of the product. const mp limb t *s1p. mp size t n. either 0 or 1. for optimal speed. Return the most significant limb of the product. and write the quotient at r1p. This is a lowlevel function that is a building block for general multiplication and division as well as other operations in GMP. The quotient is rounded towards 0. mp size t nn. pass rs2p + s3n as r1p. mp_limb_t mpn_submul_1 (mp limb t *rp. For most usages. The most significant limb of the divisor must be nonzero. it will be s3n limbs long (i. void mpn_tdiv_qr (mp limb t *qp. .] [Function] Divide {rs2p. and subtract the n least significant limbs of the product from {rp. qxn fraction limbs are developed. nn−dn+1} and the remainder at {rp. mp size t qxn. and add the n least significant limbs of the product to {rp. const mp limb t *np. s3n}. It is written in assembly for most CPUs. mp size t s3n ) [This function is obsolete. It is written in assembly for most CPUs. No overlap is permitted between arguments. mp limb t s2limb ) [Function] Multiply {s1p. as many limbs as the divisor). plus carryout from the addition. const mp limb t *s1p. no overlap between arguments is permitted. rs2n} by {s3p. mp size t dn ) [Function] Divide {np. dn}. In addition to an integer quotient.58 GNU MP 5. with the exception of the most significant limb. mp_limb_t mpn_addmul_1 (mp limb t *rp.
expecting it to divide exactly. mp size t n. the return value is nonzero and the result won’t be anything useful. mp size t s3n ) [This function is obsolete. mp limb t carry ) [Macro] [Function] Divide {sp. If 3 divides exactly. mpn_divexact_by3c takes an initial carry parameter. These routines use a multiplybyinverse and will be faster than mpn_divrem_1 on CPUs with fast multiplication but slow division. mpn_ divexact_by3 is simply a macro calling mpn_divexact_by3c with a 0 carry parameter. mp limb t s3limb ) [Function] [Macro] Divide {s2p. mp_limb_t mpn_divmod (mp limb t *r1p. the return value is zero and the result is the quotient. mp_limb_t mpn_mod_1 (mp limb t *s1p. [Function] mp_limb_t mpn_lshift (mp limb t *rp. mp size t n . Return the remainder. n} by 3. which can be the return value from a previous call. mp limb t *s2p . which is always so currently). The bits shifted out at the left are returned in the least significant count bits of the return value (the rest of the return value is zero). The integer quotient is written to {r1p+qxn. Please call mpn_tdiv_qr instead for best performance. the remainder (a − i) mod 3 is given by 3 − c. mp size t n ) mp_limb_t mpn_divexact_by3c (mp limb t *rp. s1n can be zero. The source a. because b ≡ 1 mod 3 (when mp_bits_per_limb is even. and write the result to {rp. mp limb t *sp. and write the quotient at r1p. s2n} and in addition qxn fraction limbs are developed and written to {r1p. . initial carry i. This function is written in assembly for most CPUs. so a large calculation can be done piece by piece from low to high. The regions {sp. mp size t qxn. and return the remainder. 1 or 2. mp limb t s2limb ) Divide {s1p. qxn}. not partially overlapping. mp limb t *rs2p. For most usages. count must be in the range 1 to mp_bits_per_limb−1. and the initial carry i must also be 0. qxn will be zero. where b = 2 GMP NUMB BITS.Chapter 8: Lowlevel Functions 59 mp_limb_t mpn_divrem_1 (mp limb t *r1p. n}. and return value c satisfy cbn + a − i = 3q. If not.] [Function] mp_limb_t mpn_divexact_by3 (mp limb t *rp. mp limb t *sp. s2n} by s3limb. const mp limb t *s3p. const mp limb t *sp. s1n} by s2limb. mp size t s2n . The areas at r1p and s2p have to be identical or completely separate. mp limb t *s2p. mp limb t s3limb ) mp_limb_t mpn_divmod_1 (mp limb t *r1p. When c = 0 clearly q = (a − i)/3. provided rp ≥ sp. mp size t s2n. and writing the result to {rp. unsigned int count ) [Function] Shift {sp. 1 or 2 (these are both borrows really). size n. mp size t rs2n. n}. n} and {rp. n} left by count bits. mpn_divmod_1 exists for upward source compatibility and is simply a macro calling mpn_ divrem_1 with a qxn of 0. n} may overlap. When c = 0. mp size t s1n. Either or both s2n and qxn can be zero. result q. The return c is always 0.
provided rp ≤ sp. retval}. mp size t yn ) Set {rp. yn} must be odd.e. The areas {xp. [Function] Compute the greatest common divisor G of U and V . Both operands must have nonzero most significant limbs.0. mp limb t *xp. Compatibility note: GMP 4. The second cofactor T is not computed but can easily be obtained from (G − U S)/V (the division will be exact). yn}. n}. count must be in the range 1 to mp_bits_per_limb−1. Both operands must be nonzero. n} right by count bits. unsigned int count ) [Function] Shift {sp. mp size t *sn. const mp limb t *sp. xn} must have at least as many bits as {yp. mp limb t *xp. n/2 } and the remainder at {r2p. yn}.3. n} and put the result at {r1p. mp size t xn. The areas at gp and sp should each have room for xn + 1 limbs.3. yn + 1} are destroyed (i. S = 0 if and only if V divides U (i. mp_size_t mpn_gcdext (mp limb t *gp. The result can be up to yn limbs. The bits shifted out at the right are returned in the most significant count bits of the return value (the rest of the return value is zero). {xp. or a negative value if s1 < s2. It is required that U ≥ V > 0. G = V ). when this happens *sn will be negative. xn} and let V be defined by {yp. Store G at gp and let the return value define its limb count. mp limb t *yp. [Function] mp_size_t mpn_gcd (mp limb t *rp. S satisfies S = 1 or S < V /(2G). The regions {sp. 0 if they are equal. the input operands plus an extra limb past the end of each). n} and {rp. and write the result to {rp. const mp limb t *sp. retval} to the greatest common divisor of {xp. S can be negative. xn} and {yp. mp size t xn. Store S at sp and let *sn define its limb count. mp_size_t mpn_sqrtrem (mp limb t *r1p.1 defined S less strictly. n} may overlap. mp limb t *sp. Both source operands are destroyed. mp limb t *yp. but the return value indicates how many are produced. yn}. r2p needs space for n limbs. mp size t yn ) Let U be defined by {xp. the return value is the actual number produced. mp size t n ) [Function] Compare {s1p. n} and return a positive value if s1 > s2.e. .1 mp_limb_t mpn_rshift (mp limb t *rp. Compute a cofactor S such that G = U S + V T . xn} and ylimb. const mp limb t *s2p. mp_limb_t mpn_gcd_1 (const mp limb t *xp. int mpn_cmp (const mp limb t *s1p. mp limb t *r2p. mp size t xn. {yp. mp limb t ylimb ) [Function] Return the greatest common divisor of {xp.0 and 4. n} and {s2p. xn + 1} and {yp. yn}. xn} and {yp. mp size t n. mp size t n ) [Function] Compute the square root of {sp..60 GNU MP 5. This function is written in assembly for most CPUs. No overlap is permitted between {xp. Earlier as well as later GMP releases define S as described here.
not an ASCII character. s1n} must be nonzero. mp_size_t mpn_set_str (mp limb t *rp. except when base is a power of 2. mp bitcnt t bit ) Scan s1p from bit position bit for the next set bit. and in this case the return value is zero or nonzero according to whether the remainder would have been zero or nonzero. strsize must be at least 1. See also mpz_perfect_square_p. mp size t s1n ) [Function] Convert {s1p. The areas {r1p. A return value of zero indicates a perfect square. The area at str has to have space for the largest possible number represented by a s1n long limb array. s1n} to a raw unsigned char array at str in base base. n/2 } and {sp. n} and {sp.strsize} and the result at rp. mp limb t *s1p. int base ) Convert bytes {str. depending on the base and range. so that the function has something to return. If the most significant input byte is nonzero then the high limb at rp will be nonzero. const unsigned char *str. plus one extra character. base can vary from 2 to 256. and return the number of characters produced.Chapter 8: Lowlevel Functions 61 The most significant limb of {sp.strsize} in the given base to limbs at rp. so that the function has something to return. mp size t r1n ) void mpn_random2 (mp limb t *r1p. The input {s1p. mp_bitcnt_t mpn_scan1 (const mp limb t *s1p. n} must be completely separate. . add the ASCII codes for ‘0’ or ‘A’. mp_size_t mpn_get_str (unsigned char *str. If the most significant input byte is zero then there may be high zero limbs written to rp and included in the return value. The most significant limb of the input {s1p. [Function] It is required that there be a set bit within the area at s1p at or beyond bit position bit. size t strsize. void mpn_random (mp limb t *r1p. in which case it’s unchanged. n} must be nonzero. and no overlap is permitted between {str. s1n} is clobbered. base can vary from 2 to 256. [Function] str[0] is the most significant byte and str[strsize − 1] is the least significant. The return value is the number of limbs written to rp. mp bitcnt t bit ) Scan s1p from bit position bit for the next clear bit. mpn_random2 generates long strings of zeros and ones in the binary representation. n} must be either identical or completely separate. mp size t r1n ) [Function] [Function] Generate a random number of length r1n and store it at r1p. and only that exact number of limbs will be required there. mpn_random generates uniformly distributed limb data. int base. Each byte should be a value in the range 0 to base − 1. If the remainder is not wanted then r2p can be NULL. [Function] It is required that there be a clear bit within the area at s1p at or beyond bit position bit. The most significant limb is always nonzero. to convert it to printable format. There may be leading zeros in the string. The areas {r2p. The string is not in ASCII. mp_bitcnt_t mpn_scan0 (const mp limb t *s1p.
n}.62 GNU MP 5. which is the number of bit positions where the two operands have different bit values. mp size t n ) Count the number of set bits in {s1p. void mpn_xnor_n (mp limb t *rp. and write the bitwise complement of the result to {rp. n}. n} and the bitwise complement of {s2p. const mp limb t *s1p. const mp limb t *s2p. void mpn_andn_n (mp limb t *rp. and write the result to {rp. n}. [Function] void mpn_and_n (mp limb t *rp. mp size t n ) [Function] Perform the bitwise logical and of {s1p. const mp limb t *s1p. n}. n} and {s2p. const mp limb t *s2p. mp size t n ) [Function] Perform the bitwise logical inclusive or of {s1p. const mp limb t *s2p. int mpn_perfect_square_p (const mp limb t *s1p. n}. n} is a perfect square. . n} and {s2p. and write the bitwise complement of the result to {rp. n} and {s2p. n}. mp size t n ) [Function] Perform the bitwise logical inclusive or of {s1p. mp size t n ) [Function] Perform the bitwise logical and of {s1p. n} and {s2p. n}. n}. const mp limb t *s2p. void mpn_nior_n (mp limb t *rp. const mp limb t *s2p. [Function] mp_bitcnt_t mpn_hamdist (const mp limb t *s1p. n}. const mp limb t *s2p. mp size t n ) [Function] Perform the bitwise logical and of {s1p.1 mpn_random2 is intended for testing the correctness of the mpn routines. n}. void mpn_nand_n (mp limb t *rp. mp size t n ) [Function] Perform the bitwise logical exclusive or of {s1p.0. n}. const mp limb t *s2p. void mpn_xor_n (mp limb t *rp. n}. const mp limb t *s1p. mp size t n ) [Function] Compute the hamming distance between {s1p. n} and {s2p. n}. const mp limb t *s2p. const mp limb t *s1p. mp size t n ) [Function] Perform the bitwise logical exclusive or of {s1p. const mp limb t *s1p. n} and {s2p. void mpn_iorn_n (mp limb t *rp. n}. n}. const mp limb t *s1p. void mpn_ior_n (mp limb t *rp. mp_bitcnt_t mpn_popcount (const mp limb t *s1p. mp size t n ) Return nonzero iff {s1p. const mp limb t *s1p. n} and {s2p. and write the result to {rp. and write the result to {rp. n}. const mp limb t *s2p. and write the bitwise complement of the result to {rp. and write the result to {rp. n}. mp size t n ) [Function] Perform the bitwise logical inclusive or of {s1p. and write the result to {rp. const mp limb t *s1p. n}. n} and the bitwise complement of {s2p.
[Function] void mpn_copyd (mp limb t *rp. [Function] void mpn_zero (mp limb t *rp. . But programs acting on limbs only through the mpn functions are likely to work equally well with either build. GMP_LIMB_BITS is the total number of bits in an mp_limb_t. GMP_NAIL_MASK is 0 when nails are not in use. and write the result to {rp. const mp limb t *sp. const mp limb t *s1p. but a particular number can be selected with ‘enablenails=N’. n}. and that means one less large constant. mp size t n ) Zero {rp. increasingly. In all cases GMP_LIMB_BITS == GMP_NAIL_BITS + GMP_NUMB_BITS GMP_NAIL_MASK GMP_NUMB_MASK [Macro] [Macro] Bit masks for the nail and number parts of a limb.1 Nails Everything in this section is highly experimental and may disappear or be subject to incompatible changes in a future version of GMP. mp size t n ) Perform the bitwise complement of {sp. meaning mpz etc. which can help various RISC chips. n} to {rp. At the mpn level. By default the number of bits will be chosen according to what suits the host processor. This can significantly improve carry handling on some processors. and judicious use of the definitions below should make any program compatible with either build. mp size t n ) Copy from {s1p. GMP_NAIL_MASK is not often needed.Chapter 8: Lowlevel Functions 63 void mpn_com (mp limb t *rp. All the mpn functions accepting limb data will expect the nail bits to be zero on entry. [Function] 8. mp size t n ) Copy from {s1p. at the source level. GMP_NUMB_BITS is the number of data bits in a limb. n}. n}. const mp limb t *s1p. Nails can be enabled by configuring with ‘enablenails’. a nail build should be fully source and binary compatible with a nonnail build. a nail build is neither source nor binary compatible with a nonnail build. decreasingly. n}. [Function] void mpn_copyi (mp limb t *rp. or 0 when nails are not in use. GMP_NAIL_BITS GMP_NUMB_BITS GMP_LIMB_BITS [Macro] [Macro] [Macro] GMP_NAIL_BITS is the number of nail bits. This applies both to limb vectors and to single limb arguments. strictly speaking. For the higher level routines. Nails are an experimental feature whereby a few bits are left unused at the top of each mp_limb_ t. since the nail part can be obtained with x >> GMP_NUMB_ BITS. n}. n} to {rp. and will return data with the nails similarly all zero.
This would help vector processors since carries would only ever need to propagate one or two limbs. In the future (the distant future most likely) a nonzero nail might be permitted.1 GMP_NUMB_MAX [Macro] The maximum value that can be stored in the number part of a limb. . giving nonunique representations for numbers in a limb vector.64 GNU MP 5. The term “nails” comes from finger or toe nails. This is the same as GMP_NUMB_MASK. but can be used for clarity when doing comparisons rather than bitwise operations. “numb” is short for number. but is also how the developers felt after trying for a long time to come up with sensible names for these things.0. which are at the ends of a limb (arm or leg).
. [Function] Initialize state with an algorithm selected by alg. 9.8 [Miscellaneous Float Functions]. and can be seeded with one of the gmp_randseed functions. If size is bigger than the table data provides then the return value is zero. mp bitcnt t size ) [Function] Initialize state for a linear congruential algorithm as per gmp_randinit_lc_2exp. [Function] void gmp_randinit_lc_2exp (gmp randstate t state. Such a variable must be initialized by a call to one of the gmp_randinit functions. gmp randalg t alg . They use a default algorithm and are currently not seeded (though perhaps that will change in the future). The older style random number functions don’t accept a gmp_randstate_t parameter but instead share a global variable of that type. etc. page 40. The only choice is GMP_RAND_ALG_LC. which is gmp_randinit_lc_2exp_size described above. mp bitcnt t m2exp ) Initialize state with a linear congruential algorithm X = (aX + c) mod 2m2exp . and Section 7. The maximum size currently supported is 128. If successful the return value is nonzero. which holds an algorithm selection and a current state. multiple iterations of the recurrence are used and the results concatenated. ) This function is obsolete. The functions actually generating random numbers are described in Section 5. mpz t a. ie. unsigned long c .13 [Integer Random Numbers]. c and m2exp are selected from a table. When a random number of more than m2exp/2 bits is to be generated. void gmp_randinit_set (gmp randstate t rop. and the second bit no more than 4. The low bits of X in this algorithm are not very random.Chapter 9: Random Number Functions 65 9 Random Number Functions Sequences of pseudorandom numbers in GMP are generated using a variable of type gmp_ randstate_t. void gmp_randinit_mt (gmp randstate t state ) [Function] Initialize state for a Mersenne Twister algorithm. The least significant bit will have a period no more than 2. This will be a compromise between speed and randomness. This algorithm is fast and has good randomness properties. . m2exp/2 ≥ size. gmp randstate t op ) Initialize rop with a copy of the algorithm and state from op. . A third parameter of type unsigned long . page 54. Currently this is gmp_randinit_mt. For this reason only the high half of each X is actually used. int gmp_randinit_lc_2exp_size (gmp randstate t state. chosen so that size bits (or more) of each X will be used.1 Random State Initialization void gmp_randinit_default (gmp randstate t state ) [Function] Initialize state with a default algorithm. [Function] void gmp_randinit (gmp randstate t state. and is recommended for applications with no special requirements. The new functions accepting a gmp_randstate_t are recommended for applications that care about randomness. a.
[Function] 9. ie.0. The size of a seed determines how many different sequences of random numbers that it’s possible to generate. unsigned long gmp_urandomm_ui (gmp randstate t state. or GMP_ERROR_INVALID_ARGUMENT if the size parameter is too big. n must be less than or equal to the number of bits in an unsigned long.66 GNU MP 5. so if unpredictability is required then it should definitely not be the only source for the seed value. in the range 0 to 2n −1 inclusive. such as generating cryptographic keys.2 Random State Seeding void gmp_randseed (gmp randstate t state. and this affects the randomness of separate number sequences. GMP_ERROR_ UNSUPPORTED_ARGUMENT if alg is unsupported. gmp_randinit sets bits in the global variable gmp_errno to indicate an error. but care needs to be taken with this. unsigned long int seed ) Set an initial seed value into state. mpz t seed ) void gmp_randseed_ui (gmp randstate t state. It may be noted this error reporting is not thread safe (a good reason to use gmp_randinit_lc_2exp_size instead). unsigned long n) Return a uniformly distributed random number in the range 0 to n − 1. Traditionally the system time has been used to seed. The “quality” of the seed is the randomness of a given seed compared to the previous seed used. [Function] [Function] 9.3 Random State Miscellaneous unsigned long gmp_urandomb_ui (gmp randstate t state. unsigned long n) [Function] Return a uniformly distributed random number of n bits. Also. [Function] . GMP_RAND_ALG_DEFAULT or 0 are the same as GMP_RAND_ALG_LC. If an application seeds often and the resolution of the system clock is low. void gmp_randclear (gmp randstate t state ) Free all memory occupied by state. The method for choosing a seed is critical if the generated numbers are to be used for important applications. On some systems there’s a special device ‘/dev/random’ which provides random data better suited for use as a seed. then the same sequence of numbers might be repeated. this is the size for that function. the system time is quite easy to guess. inclusive.1 is required.
if needed. For example. h hh short char . mpq_t q. gmp_printf ("limb %Mu\n".Chapter 10: Formatted Output 67 10 Formatted Output 10. the same as the standard printf. ‘M’ for mp_limb_t. ‘0X’ or ‘0’ always show a sign show a space or a ‘’ sign group digits. mp_limb_t l. and ‘N’ for an mp_limb_t array. A negative size can be given to print the value as a negative. ‘Z’. n). or as a ‘*’ to take an extra parameter of type int. ptr. mpq_t and mpf_t respectively.1 Format Strings gmp_printf and friends accept format strings similar to the standard C printf (see Section “Formatted Output” in The GNU C Library Reference Manual). z). gmp_printf ("%s is an mpz %Zd\n". size). "here". and can be freely intermixed with the GMP extensions. The flags accepted are as follows. l). In the current implementation the standard parts of the format string are simply handed to printf and only the GMP extensions handled directly. gmp_printf ("limb array %Nx\n". 0 # + (space) ’ pad with zeros (rather than spaces) show the base with ‘0x’. n. ‘M’ and ‘N’ behave like integers.[precision]] [type] conv GMP adds types ‘Z’. ‘Q’. A format specification is of the form % [flags] [width] [. All the standard C printf types behave the same as the C library printf. int n. mpf_t f. ‘Q’ will print a ‘/’ and a denominator. GLIBC style ‘’’ is only for the standard C types (not the GMP types). The standard types accepted are as follows. and only if the C library supports it. q). GLIBC style (not GMP types) The optional width and precision can be given as a number within the format string. For ‘N’ the limbs are expected least significant first. ‘F’ behaves like a float. const mp_limb_t *ptr. page 56). ‘Q’ and ‘F’ for mpz_t. the rest will depend on the compiler (or include files) for the type and the C library for the output.*Ff with %d digits\n". mpz_t z. as per the mpn functions (see Chapter 8 [Lowlevel Functions]. mp_size_t size. gmp_printf ("a hex rational: %#40Qx\n". gmp_printf ("fixed point mpf %. f. ‘h’ and ‘l’ are portable.
‘x’ and ‘X’ are unsigned for the standard C types. mpf_t conversions only ever generate as many digits as can be accurately represented by the operand. Other types or conversions that might be accepted by the C library printf cannot be used through gmp_printf. C99 style character decimal integer scientific format float fixed point float same as d fixed or scientific float strerror string. This happens even for an ‘f’ conversion of an mpf_t which is an integer. ‘n’ can be used with any type. according to the size of mp_limb_t.1 j l ll L q t z intmax_t or uintmax_t long or wchar_t long long long double quad_t or u_quad_t ptrdiff_t size_t The GMP types are F Q M N Z mpf_t. The precision field has it’s usual meaning for integer ‘Z’ and float ‘F’ types. a c d e f i g m n o p s u x A hex floats. but is currently undefined for ‘Q’ and should not be used with that. ‘u’ is not meaningful for ‘Z’. ‘a’ and ‘A’ are always supported for mpf_t but depend on the C library for standard C float types. ‘M’ is a proxy for the C library ‘l’ or ‘L’. integer conversions mpz_t. this includes for instance extensions registered with GLIBC register_ printf_function. for instance .0. but for types ‘Z’. ‘Q’ and ‘N’. float conversions mpq_t. even the GMP types. integer conversions mp_limb_t. integer conversions mp_limb_t array.68 GNU MP 5. ‘Q’ and ‘N’ they are signed. but a signed conversion can be used and will interpret the value as a twos complement negative. Also currently there’s no support for POSIX ‘$’ style numbered arguments (perhaps this will be added in the future). Zeros will be used if necessary to pad to the requested precision. Unsigned conversions will be usual. ‘m’ and ‘p’ depend on the C library. the same as mpf_get_str does. integer conversions The conversions accepted are as follows. GLIBC style store characters written so far octal integer pointer string unsigned integer hex integer E G X ‘o’.
since there’s no protection against exceeding the space available at buf. To get the full output. multibyte characters are not recognised. . GCC format string checking is not available. . . size must be enough for the string and nullterminator. [Function] [Function] Print to the stream fp. The return value is the total number of characters which ought to have been produced. All the functions can return −1 if the C library printf variant in use returns −1. It should be emphasised that if a format string is invalid. The format string is only interpreted as plain chars. Perhaps this will change in the future. The decimal point character (or string) is taken from the current locale settings on systems which provide localeconv (see Section “Locales and Internationalization” in The GNU C Library Reference Manual). const char *fmt. va list ap ) [Function] [Function] Print to the standard output stdout. excluding the terminating null. ) int gmp_vsprintf (char *buf. excluding the terminating null. const char *fmt.Fe’ or ‘%.Chapter 10: Formatted Output 69 21024 in an mpf_t of 128 bits precision will only produce about 40 digits. . The C library will normally do the same for standard float output. or ‘man 3 va_start’. int gmp_fprintf (FILE *fp. see Section “Variadic Functions” in The GNU C Library Reference Manual. int gmp_snprintf (char *buf. The file based functions gmp_printf and gmp_fprintf will return −1 to indicate a write error. . then pad with zeros to the decimal point. size t size. If retval ≥ size then the actual output has been truncated to the first size − 1 characters.Ff’ can be used to specifically request just the significant digits. const char *fmt. Return the number of characters written. These functions are not recommended. then the behaviour of any of these functions will be unpredictable. ) int gmp_vprintf (const char *fmt. . .2 Functions Each of the following functions is similar to the corresponding C library function. . ) int gmp_vfprintf (FILE *fp. const char *fmt.size} and the fmt string. va list ap ) [Function] [Function] Form a nullterminated string in buf. . or −1 if an error occurred. va list ap ) [Function] [Function] Form a nullterminated string in buf. . No overlap is permitted between the region {buf. but this shouldn’t normally occur. 10. const char *fmt. . or −1 if an error occurred. Output is not “atomic”. The basic printf forms take a variable argument list. va list ap ) int gmp_sprintf (char *buf. Return the number of characters written. Return the number of characters written. . No more than size bytes will be written. since it doesn’t recognise the GMP extensions. The vprintf forms take an argument pointer. int gmp_printf (const char *fmt. No overlap is permitted between the space at buf and the string fmt. ) int gmp_vsnprintf (char *buf. const char *fmt. so partial output may be produced if a write error occurs. and a null appended. . An empty precision field like ‘%. or the arguments don’t match what the format specifies. size t size.
which probably means only on GNU systems. ostream& operator<< (ostream& stream. using its ios formatting settings. using its ios formatting settings. it lets the current allocation function handle that. In hex or octal. const char *fmt. the same as for decimal. using its ios formatting settings. The return value is the number of characters produced. mpq t op ) [Function] Print op to stream. Prototypes are available from <gmp. ) int gmp_vasprintf (char **pp. page 86). int gmp_obstack_printf (struct obstack *ob. ios::width is reset to 0 after output. which instead give twos complement. the same as the standard ostream operator<< routines do. const char *fmt. . the same as for decimal. ios::width is reset to 0 after output. This is so even if the C library vsnprintf is the older GLIBC 2.0. mpf t op ) [Function] Print op to stream. If ios::showbase is set then a base indicator is shown on both the numerator and denominator (if the denominator is required). const char *fmt. excluding the nullterminator. page 3). gmp_asprintf doesn’t return −1 if there’s no more memory available. since that object might move as it grows. the same as the standard ostream operator<< routines do. The address of the block in stored to *pp. A nullterminator is not written. In hex or octal. op is printed as a signed value.1 [Headers and Libraries]. const char *fmt. The decimal point follows the standard library float operator<<. int gmp_asprintf (char **pp. fmt cannot be within the current object in ob. . The return value is the number of characters written.x style. Unlike the C library asprintf. . ios::width is reset to 0 after output. ostream& operator<< (ostream& stream. which is built if C++ support is enabled (see Section 2. which on recent systems means the std::locale imbued on stream.h>.1 [Build Options].0. va list ap ) [Function] [Function] Append to the current object in ob. op is printed as a signed number. the same as the standard ostream operator<< routines do. 10. These functions are available only when the C library provides the obstack feature.3 C++ Formatted Output The following functions are provided in ‘libgmpxx’ (see Section 3. . ) int gmp_obstack_vprintf (struct obstack *ob. . ostream& operator<< (ostream& stream. or if the denominator is 1 then just a plain integer like ‘123’. mpz t op ) [Function] Print op to stream. The block will be the size of the string and nullterminator. This is unlike the standard operator<< routines on int etc. see Section “Obstacks” in The GNU C Library Reference Manual. va list ap ) [Function] [Function] Form a nullterminated string in a block of memory obtained from the current memory allocation function (see Chapter 14 [Custom Allocation].1 Notice the return value is in ISO C99 snprintf style.70 GNU MP 5. page 16). . . Output will be a fraction like ‘5/9’.
page 74) is the only overloading available for the GMP types and that for instance using + with an mpz_t will have unpredictable results. This is as per mpf_out_str.Chapter 10: Formatted Output 71 Hex and octal are supported. The mantissa will be in hex or octal.. mpz_t z.3 [C++ Formatted Input]. and will put a base on the mantissa. . . These operators mean that GMP types can be printed in the usual C++ way. ios::showbase is supported.4’ or ‘00. for example hex ‘0x1. page 76. the exponent will be in decimal. unlike the standard operator<< on double. cout << "iteration " << n << " value " << z << "\n".8’ or ‘0x0. int n. see Chapter 12 [C++ Class Interface]. For hex the exponent delimiter is an ‘@’. but at least differentiates itself from decimal.. This last form is slightly strange. or octal ‘01. for example. But note that ostream output (and istream input.4’. For classes with overloading.8’. see Section 11.
All the standard C scanf types behave the same as in the C library scanf. since they’re already “callbyreference”. gmp_sscanf ("0377 + 0x10/0x11".%Ff)". q2).66)" */ mpf_t x. ‘h’ and ‘l’ are portable. and ‘’’ cannot be used with GMP types. q1. the rest will depend on the compiler (or include files) for the type and the C library for the input. mpz_t z. * a ’ read but don’t store allocate a buffer (string conversions) grouped digits. ‘a’ and ‘’’ will depend on support from the C library. mpq_t and mpf_t respectively. x. ‘Q’ will read a ‘/’ and a denominator. gmp_scanf ("a(%d) = %Zd\n". and can be freely intermixed with the GMP extensions.55. double or wchar_t long long long double quad_t or u_quad_t ptrdiff_t size_t The GMP types are .1 Formatted Input Strings gmp_scanf and friends accept format strings similar to the standard C scanf (see Section “Formatted Input” in The GNU C Library Reference Manual). buf.2. /* to read say "a(5) = 1234" */ int n. A format specification is of the form % [flags] [width] [type] conv GMP adds types ‘Z’.72 GNU MP 5. h hh j l ll L q t z short char intmax_t or uintmax_t long int. GLIBC style (not GMP types) The standard types accepted are as follows. mpq_t q1. ‘Z’ and ‘Q’ behave like integers. GMP variables don’t require an & when passed to gmp_scanf. q2. ‘F’ behaves like a float. if present. char buf[32]. &n. y.1 11 Formatted Input 11. The flags accepted are as follows.0. ‘Q’ and ‘F’ for mpz_t. For example. gmp_scanf ("%31s (%Ff. In the current implementation the standard parts of the format string are simply handed to scanf and only the GMP extensions handled directly. "%Qi + %Qi". /* to read say "topleft (1. z). y).
float conversions mpq_t. ‘n’ can be used with any of the types above. negatives are still allowed (per strtoul. page 67) is always accepted for mpf_t. see Section 10. Perhaps this will change in the future. For the standard C types these are described as “unsigned” conversions. ‘g’ and ‘G’ are identical. For float conversions. the rest are standard. C99 style hex float format (printf %a. multibyte characters are not recognised. but for the standard float types it will depend on the C library. and either upper or lower case ‘e’ for the exponent in scientific format. integer conversions The conversions accepted are as follows. ‘o’. page 44). whereas ‘0x10/0x11’ would be 16/17.1 [Formatted Output Strings]. the decimal point character (or string) expected is taken from the current locale settings on systems which provide localeconv (see Section “Locales and Internationalization” in The GNU C Library Reference Manual). The format string is only interpreted as plain chars. both accept both upper and lower case hexadecimal. The C library will normally do the same for standard float input. ‘f’. but that merely affects certain overflow handling. For GMP types there are no overflows. . ‘p’ and ‘[’ will depend on support from the C library. For example ‘0x10/11’ would be 16/11. Whitespace is read and discarded before a field. If the value might not be in canonical form then mpq_canonicalize must be called before using it in any calculations (see Chapter 6 [Rational Number Functions]. ‘*’ to suppress assignment is allowed. ‘x’ and ‘X’ are identical. see Section “Parsing of Integers” in The GNU C Library Reference Manual).Chapter 11: Formatted Input 73 F Q Z mpf_t. ‘E’. except for ‘c’ and ‘[’ conversions. ‘u’. though in that case it would do nothing at all. c d e E f g G i n o p s u xX [ character or characters decimal integer float integer with base indicator characters read so far octal integer pointer string of nonwhitespace characters decimal integer hex integer string of characters in a set ‘e’. so ‘d’ and ‘u’ are identical. even the GMP types. ‘Q’ type reads the numerator and (optional) denominator as given. integer conversions mpz_t. ‘x’ and ‘X’ all read positive or negative values. they all read either fixed point or scientific format. Other conversions or types that might be accepted by the C library scanf cannot be used through gmp_scanf. ‘Qi’ will read a base specification separately for the numerator and denominator.
23eXYZ’ would be read up to the ‘X’ and that character pushed back since it’s not a digit. . ) int gmp_vsscanf (const char *s. . .1 [Headers and Libraries]. input parsing follows C99 rules. If end of input (or a file error) is reached before a character for a field or a literal. int gmp_scanf (const char *fmt. in the current implementation GMP calls the C library scanf functions. [Function] . const char *fmt. const char *fmt. then the return value is EOF instead of 0. The plain scanf forms take a variable argument list. .1 11. If this doesn’t provide a complete number then the function terminates. A whitespace character in the format string is only an optional match and doesn’t induce an EOF in this fashion. va list ap ) Read from the standard input stdin. No overlap is permitted between the fmt string and any of the results produced. [Function] [Function] The return value from each of these functions is the same as the standard C99 scanf. const char *fmt. . ) int gmp_vscanf (const char *fmt. .74 GNU MP 5. then the behaviour of any of these functions will be unpredictable. [Function] [Function] int gmp_sscanf (const char *s. namely the number of fields successfully parsed and stored. and if no previous nonsuppressed fields have matched. ) int gmp_vfscanf (FILE *fp.23e’ would then be considered invalid since an ‘e’ must be followed by at least one digit. . or the arguments don’t match what the format specifies. Prototypes are available from <gmp. the same way C99 sscanf is the same as fscanf.2 Formatted Input Functions Each of the following functions is similar to the corresponding C library function. istream& operator>> (istream& stream. it is deliberately made identical to gmp_fscanf.3 C++ Formatted Input The following functions are provided in ‘libgmpxx’ (see Section 3. which is built only if C++ support is enabled (see Section 2. It should be emphasised that if a format string is invalid. namely one character of lookahead is used and characters are read while they continue to meet the format requirements. page 16). For instance with mpf_t an input ‘1. or ‘man 3 va_start’. mpz t rop ) Read rop from stream. For the standard C types. Note that gmp_sscanf is the same as gmp_fscanf and only does one character of lookahead when parsing. . since it doesn’t recognise the GMP extensions. . with that field not stored nor counted towards the return value. const char *fmt.0. GCC format string checking is not available. For the GMP types. [Function] [Function] int gmp_fscanf (FILE *fp. see Section “Variadic Functions” in The GNU C Library Reference Manual. va list ap ) Read from a nullterminated string s. va list ap ) Read from the stream fp.h>. page 3). The vscanf forms take an argument pointer. ‘%n’ fields and fields read but suppressed by ‘*’ don’t count towards the return value. 11. Although clearly it could look at its entire input. Leading whitespace read and discarded for a field don’t count as characters for that field. The string ‘1.1 [Build Options]. which might have looser rules about what constitutes a valid input. using its ios formatting settings.
[Function] Hex or octal floats are not supported.Chapter 11: Formatted Input 75 istream& operator>> (istream& stream. Note that digit grouping specified by the istream locale is currently not accepted. mpf t rop ) Read rop from stream. If the fraction is not in canonical form then mpq_canonicalize must be called (see Chapter 6 [Rational Number Functions]. or perhaps it’s best to accept only what the standard float operator>> does.. page 76. But note that istream input (and ostream output. istream& operator>> (istream& stream. but might be in the future. see Section 10. for example. see Chapter 12 [C++ Class Interface]. Perhaps this will change in the future.. For classes with overloading. so that for instance ‘0x10/11’ is 16/11 and ‘0x10/0x11’ is 16/17. or a fraction like ‘5/9’. This is done separately for numerator and denominator. page 70) is the only overloading available for the GMP types and that for instance using + with an mpz_t will have unpredictable results. using its ios formatting settings. . As per integer input.3 [C++ Formatted Output]. mpq t rop ) [Function] An integer like ‘123’ will be read. page 44) before operating on it. No whitespace is allowed around the ‘/’. an ‘0’ or ‘0x’ base indicator is read when none of ios::dec. cin >> z. ios::oct or ios::hex are set. mpz_t z. These operators mean that GMP types can be read in the usual C++ way. .
For example. still use temporaries though. without using a temporary for the b+c part. a = 1234.76 GNU MP 5. This is because C++ will automatically convert any pointer to a bool. cout << "absolute value is " << abs(c) << "\n". Everything described in this chapter is to be considered preliminary and might be subject to incompatible changes if some unforeseen difficulty reveals itself. like a=b*c+d*e. Due to the implementation of this interface. since ‘gmp. so if GMP accepted bool it . c. Smaller types like int or float can also be intermixed.1 C++ Interface General All the C++ classes and functions are available with #include <gmpxx. Note that bool is not accepted directly. For example.0. All GMP C language types and functions can be used in C++ programs. For GCC this means version 2. b. 12. one supporting namespaces. cout << "sum is " << c << "\n". but must be explicitly cast to an int first. return 0. unsigned long and double.h> Programs should be linked with the ‘libgmpxx’ and ‘libgmp’ libraries.91 or later. The classes can be freely intermixed in expressions. partial specialization of templates and member templates. as can the classes and the standard types long. since C++ will promote them. but the class interface offers overloaded functions and operators which may be more convenient. b = "5678". g++ mycxxprog.1 12 C++ Class Interface This chapter describes the C++ class based interface to GMP. Expressions which by their nature imply intermediate values.cc lgmpxx lgmp The classes defined are [Class] [Class] [Class] The standard operators and various standard functions are overloaded to allow arithmetic with these classes. } An important feature of the implementation is that an expression like a=b+c results in a single call to the corresponding mpz_add. a reasonably recent C++ compiler is required.h’ has extern "C" qualifiers. c = a+b. mpz_class mpq_class mpf_class int main (void) { mpz_class a.
get_mpz_t().. except long long and long double.. instead member functions like get_si are provided (see the following sections for details).. In the other direction. // . y = mpz_class (z). The same applies to operator=. In both cases this makes a copy of the value. There are no namespace setups in ‘gmpxx. The value in z is copied into the new mpz_class. 12. it doesn’t create any sort of association.. Any necessary conversion follows the corresponding C function. an std::invalid_argument exception is thrown. mpz_class x(z). void mpz_class::mpz_class (mpz t z ) void void void void mpz_class::mpz_class mpz_class::mpz_class mpz_class::mpz_class mpz_class::mpz_class If the string is not a valid integer.h’ follow GMP naming conventions and are unlikely to clash with anything. b. init and calculate z . All the standard C++ types may be used.Chapter 12: C++ Class Interface 77 would make all sorts of invalid class and pointer combinations compile but almost certainly not do anything sensible. all types and functions are simply put into the global namespace. For example to set a to the GCD of b and c. Conversions back from the classes to standard C++ types aren’t done automatically. and continues to do for compatibility. For example. int base = 0) [Function] (const string& s ) [Function] (const string& s.. mpz_class a. . page 30). The extras provided by ‘gmpxx.2 [Assigning Integers]. mpz_t z. (const char *s ) [Function] (const char *s. or assigned to if an explicit constructor is used. [Function] Construct an mpz_class from an mpz_t. c. int base = 0) [Function] Construct an mpz_class converted from a string using mpz_set_str (see Section 5.get_mpz_t(). b. This is what ‘gmp. . mpz_gcd (a.h’ has done in the past. there won’t be any permanent association between it and z.2 C++ Interface Integers void mpz_class::mpz_class (type n ) [Function] Construct an mpz_class. Also there are no automatic conversions from the classes to the corresponding GMP C types. instead a reference to the underlying C object can be obtained with the following functions. mpz_t mpz_class::get_mpz_t () mpq_t mpq_class::get_mpq_t () mpf_t mpf_class::get_mpf_t () [Function] [Function] [Function] These can be used to call a C function which doesn’t have a C++ class interface. a class can be initialized from the corresponding GMP C type. for example double follows mpz_set_d (see Section 5. mpz_class y. c. and all the GMP C++ classes can be used.h’.2 [Assigning Integers]. page 30)..get_mpz_t()).
d. as per mpz_cmp_d. Conversions between mpz_class and double.get_mpz_t()). void mpq_class::mpq_class (type op ) void mpq_class::mpq_class (integer num. integer den ) [Function] [Function] Construct an mpq_class. in the middle. and it might change in the future..78 GNU MP 5. mpz_class q. mpz class op2 ) bool mpz_class::fits_sint_p (void) bool mpz_class::fits_slong_p (void) bool mpz_class::fits_sshort_p (void) bool mpz_class::fits_uint_p (void) bool mpz_class::fits_ulong_p (void) bool mpz_class::fits_ushort_p (void) double mpz_class::get_d (void) long mpz_class::get_si (void) string mpz_class::get_str (int base = 10) unsigned long mpz_class::get_ui (void) int mpz_class::set_str (const char *str.. are defined to follow the corresponding C functions mpz_get_d and mpz_set_d. but it should be noted that if the given double is not an integer then the way any rounding is done is currently unspecified. The rounding might take place at the start. int base ) int mpz_class::set_str (const string& str.3 C++ Interface Rationals In all the following constructors.. a. type op2 ) int cmp (type op1. mpz_class abs (mpz class op1 ) int cmp (mpz class op1. or a pair of integers (mpz_class or standard C++ integer types) representing a fraction. or if not then mpq_class::canonicalize called. mpz class d ) [Function] [Function] Divisions involving mpz_class round towards zero. . functions can always be called directly if desired. And comparisons are always made exactly.. For example.1 mpz_class operator/ (mpz class a. int base ) int sgn (mpz class op ) mpz_class sqrt (mpz class op ) [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] These functions provide a C++ class interface to the corresponding GMP C routines. mpz class d ) mpz_class operator% (mpz class a.. page 32). except long long and long double. a. d. This is the same as the C99 / and % operators. however. mpz_fdiv_q (q. Overloaded operators for combinations of mpz_class and double are provided for completeness. or mpz_cdiv. The initial value can be a single value of any type. .6 [Integer Division].get_mpz_t().get_mpz_t(). The mpz_fdiv. or at the end of the operation. except that long long and long double are not supported.0. cmp can be used with any of the classes or the standard C++ types.. as per the mpz_tdiv_q and mpz_tdiv_r functions (see Section 5. if a fraction is given then it should be in canonical form. For example. 12.
then mpq_class::canonicalize must be called before further operations. int base = 0) [Function] (const string& s ) [Function] (const string& s. All arithmetic operators require their operands in canonical form. mpq_class abs (mpq class op ) int cmp (mpq class op1. int base ) int mpq_class::set_str (const string& str. except long long and long double. (const char *s ) [Function] (const char *s. If direct manipulation might produce a noncanonical value. void mpq_class::mpq_class (mpq t q ) [Function] Construct an mpq_class from an mpq_t. This can be passed to C functions expecting an mpz_t. 3). then mpq_class::canonicalize must be called before further operations. mpq_class q (1. int base ) int sgn (mpq class op ) [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] These functions provide a C++ class interface to the corresponding GMP C routines. as per Chapter 6 [Rational Number Functions]. If the object returned is modified. The same applies to operator=. mpq_class q (1. type op2 ) int cmp (type op1. Any modifications made to the mpz_t will modify the original mpq_class. This can be used both for read and write access. page 44. void void void void mpq_class::mpq_class mpq_class::mpq_class mpq_class::mpq_class mpq_class::mpq_class If the string is not a valid rational. mpz_class& mpq_class::get_num () mpz_class& mpq_class::get_den () [Function] [Function] Get a reference to an mpz_class which is the numerator or denominator of an mpq_class. . there won’t be any permanent association between it and q. The value in q is copied into the new mpq_class. an std::invalid_argument exception is thrown. int base = 0) [Function] Construct an mpq_class converted from a string using mpq_set_str (see Section 6. cmp can be used with any of the classes or the standard C++ types.75).Chapter 12: C++ Class Interface 79 mpq_class q (99). mpz_t mpq_class::get_num_mpz_t () mpz_t mpq_class::get_den_mpz_t () [Function] [Function] Get a reference to the underlying mpz_t numerator or denominator of an mpq_class. it modifies the original mpq_class. If direct manipulation might produce a noncanonical value.1 [Initializing Rationals]. page 44). void mpq_class::canonicalize () [Function] Put an mpq_class into canonical form. mpq class op2 ) double mpq_class::get_d (void) string mpq_class::get_str (int base = 10) int mpq_class::set_str (const char *str. and will return results in canonical form.
f(1. mpq_class. An mpf_class or expression will give the precision of that value. unsigned long prec ) [Function] [Function] Construct an mpf_class. The precision of a binary expression is the higher of the two operands. page 50). the initial precision is that value. an std::invalid_argument exception is thrown. An mpz_class. If prec is given. If not. 12.3 [C++ Formatted Input]. int base = 0) void mpf_class::mpf_class (const string& s ) void mpf_class::mpf_class (const string& s. The same applies to operator=. int base = 0) Construct an mpf_class converted from a string using mpf_set_str (see Section 7. in bits. f(x). page 74). f(g. those temporaries will have the same precision as the destination f. mpf_class::mpf_class (type op ) mpf_class::mpf_class (type op.80 GNU MP 5.4 C++ Interface Floats When an expression requires the use of temporary intermediate mpf_class values. the initial precision is that value. If the string is not a valid float. except long long and long double. like f=g*h+x*y. 500). mpq class& rop ).1 [Initializing Floats]. unsigned long prec. Note that operator= only stores a new value. Explicit constructors can be used if this doesn’t suit. 1000).2 [Assigning Floats]. f(abs(x)). then the initial precision is determined by the type of op given. the default mpf precision (see Section 7. in bits. mpf_class& mpf_class::operator= (type op ) [Function] Convert and store the given op value to an mpf_class object. using its ios formatting settings.1 istream& operator>> (istream& stream.5). This is the same as mpf_set etc. If prec is given.5.1 [Initializing Floats]. unsigned long prec. page 48) is used. If prec is not given.0. If the rop read might not be in canonical form then mpq_class::canonicalize must be called. and any of the GMP C++ classes can be used. mpf_class x (y). mpf_class mpf_class mpf_class mpf_class mpf_class mpf_class f(1. f(x+y). or C++ builtin type will give the default mpf precision (see Section 7. it doesn’t copy or change the precision of the destination. // x created with precision of y . the same as mpq_t operator>> (see Section 11. The same types are accepted as for the constructors above. Note in particular this means for mpf_class a copy constructor is not the same as a default constructor plus assignment. // // // // // // default precision 500 bits (at least) precision of x precision of x 1000 bits (at least) greater of precisions of x and y [Function] [Function] [Function] [Function] void mpf_class::mpf_class (const char *s ) void mpf_class::mpf_class (const char *s. [Function] Read rop from stream. instead the value is truncated if necessary. Any standard C++ type can be used. page 48).
mp_bitcnt_t mpf_class::get_prec () void mpf_class::set_prec (mp bitcnt t prec ) void mpf_class::set_prec_raw (mp bitcnt t prec ) Get or set the current precision of an mpf_class. The accuracy provided by hypot is not currently guaranteed. // x created with default precision // value truncated to that precision Applications using templated code may need to be careful about the assumptions the code makes in this area. mpf class op2 ) bool mpf_class::fits_sint_p (void) bool mpf_class::fits_slong_p (void) bool mpf_class::fits_sshort_p (void) bool mpf_class::fits_uint_p (void) bool mpf_class::fits_ulong_p (void) bool mpf_class::fits_ushort_p (void) mpf_class floor (mpf class op ) mpf_class hypot (mpf class op1. type op2 ) int cmp (type op1. Note in particular that the mpf_class must be restored to it’s allocated precision before being destroyed.Chapter 12: C++ Class Interface 81 mpf_class x.5 C++ Interface Random Numbers gmp_randclass [Class] The C++ class interface to the GMP random number functions uses gmp_randclass to hold an algorithm selection and current state. when working with mpf_class values of various different or nondefault precisions. int base = 10. page 48) apply to mpf_class::set_prec_raw. cmp can be used with any of the classes or the standard C++ types. int base ) int sgn (mpf class op ) mpf_class sqrt (mpf class op ) mpf_class trunc (mpf class op ) [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] [Function] These functions provide a C++ class interface to the corresponding GMP C routines. x = y. mpf class op2 ) double mpf_class::get_d (void) long mpf_class::get_si (void) string mpf_class::get_str (mp exp t& exp. 12. [Function] [Function] [Function] The restrictions described for mpf_set_prec_raw (see Section 7.1 [Initializing Floats]. as per gmp_randstate_t. size t digits = 0) unsigned long mpf_class::get_ui (void) int mpf_class::set_str (const char *str. there’s no automatic mechanism for it. mpf_class abs (mpf class op ) mpf_class ceil (mpf class op ) int cmp (mpf class op1. For instance implementations of the standard complex template have been seen in both styles above. int base ) int mpf_class::set_str (const string& str. though of course complex is normally only actually specified for use with the builtin float types. This must be done by application code. except long long and long double. .
a global flag. Perhaps a mechanism to tell operator>> what to do will be adopted in the future. This function is obsolete and the above randinit style should be preferred. . gmp_randclass::gmp_randclass (gmp randalg t alg. for how to choose a good seed.1 gmp_randclass::gmp_randclass (void (*randinit ) (gmp randstate t. mpf_class f (0. .1 [Random State Initialization]. gmp_randclass gmp_randclass gmp_randclass gmp_randclass r1 r2 r3 r4 (gmp_randinit_default). (gmp_randinit_lc_2exp.1 [Random State Initialization]. operator>> behaves as it does for reasons of efficiency. [Function] mpf_class gmp_randclass::get_f () mpf_class gmp_randclass::get_f (unsigned long prec ) [Function] [Function] Generate a random float f in the range 0 <= f < 1. a. For example. [Function] .get_f(). See see Chapter 9 [Random Number Functions]. 512 bits 12. at the risk of inconsistency. or if prec is not given then to the precision of the destination. [Function] [Function] void gmp_randclass::seed (unsigned long int s ) void gmp_randclass::seed (mpz class s ) mpz_class gmp_randclass::get_z_bits (unsigned long bits ) mpz_class gmp_randclass::get_z_bits (mpz class bits ) Generate a random integer with a specified number of bits. .82 GNU MP 5.. maybe a preprocessor define. c. using a call to the given randinit function (see Section 9. 32). (gmp_randinit_lc_2exp_size. page 65).0. The arguments expected are the same as randinit. m2exp). gmp_randinit_lc_2exp_size will fail if the size requested is too big. the mpq_class operator>> could canonicalize and leave .. . ) Construct a gmp_randclass. . . . f = r. .6 C++ Interface Limitations mpq_class and Templated Reading A generic piece of template code probably won’t know that mpq_class requires a canonicalize call if inputs read with operator>> might be noncanonical. // 512 bits precision // random number. 512). and is best avoided if it’s not necessary. page 65. For example. but with mpz_class instead of mpz_t. ). ) [Function] Construct a gmp_randclass using the same parameters as gmp_randinit (see Section 9. Or maybe. mpz_class gmp_randclass::get_z_range (mpz class n ) Generate a random integer in the range 0 to n − 1 inclusive. gmp_randclass r. an std::length_error exception is thrown in that case. f will be to prec bits precision. This can lead to incorrect results. A canonicalize can be quite time consuming on large operands. page 65). or an ios flag pressed into service. But this potential difficulty reduces the usefulness of mpq_class. (gmp_randinit_mt). [Function] [Function] Seed a random number generator. .
When used with. which the C++ template resolution rules are unable to automatically convert to mpz_class. template <class T> void fun (T f. g+h). mpz_class f(1). const T &). } // Good . Send feedback or alternate ideas to gmpbugs@gmplib. g). h(3). with T intended to be some numeric type. // Good Similarly. Expressions involving subclasses resolve correctly (or seem to). but in normal C++ fashion the subclass doesn’t inherit constructors and assignments. say. and a good way to reestablish them in a subclass is not yet provided. template <class T> T fun (const T &. T(f+g)). it works fine: T is resolved as mpz_class. // Good But when one of the arguments is an expression. T g) { fun2 (f. g(2). mpz_class f(1). fun (f.org. but is not currently recommended. fun (f. it doesn’t work.Chapter 12: C++ Class Interface 83 mpq_t operator>> not doing so. fun (f. Templated Expressions A subtle difficulty exists when using expressions together with applicationdefined template functions. Consider the following. within fun it may be necessary to cast an expression to type T when calling a templated fun2. // Bad } template <class T> void fun (T f. // Bad This is because g+h ends up being a certain expression template type internal to gmpxx. The workaround is simply to add an explicit cast. There’s many of those in the GMP classes. f+g).h. mpz_class(g+h)). T g) { fun2 (f. g(2). g(2). plain mpz_class variables. for use on those occasions when that’s acceptable. h(3). Subclassing Subclassing the GMP C++ classes works. mpz_class f(1).
e.84 GNU MP 5. MINT *remainder ) Set root to square. a hexadecimal. the remainder has the same sign as the dividend unless it is zero. where ‘<dir>’ is the directory where you have GNU ‘mp. void msqrt (MINT *op. The quotient is rounded towards zero. MINT *remainder ) void sdiv (MINT *dividend. and the pow function collides with pow in ‘libm. Initialize the integer to initial value. [Function] MINT * xtom (char *initial_value ) void move (MINT *src. MINT *dest ) Set dest to src by copying. It is not recommended that new programs are written using these functions. MINT *destination ) Add src 1 and src 2 and put the sum in destination. √ [Function] op . like mpz_sqrt. MINT *destination ) Multiply src 1 and src 2 and put the product in destination. The compatible functions in GNU MP do not share this restriction—inputs and outputs may overlap. Both variables must be previously initialized.h’.1 13 Berkeley MP Compatible Functions These functions are intended to be fully compatible with the Berkeley MP library which is available on many BSD derived U*ix systems. Some implementations of these functions work differently—or not at all—for negative arguments. MINT *src_2.h’ if you are going to link the GNU ‘libmp. zero if op is a perfect 2 If root and remainder are the same variable. Set remainder to (op − root ). page 3). the interface for initializing MINT objects is more error prone. MINT *divisor. Return a pointer to the MINT object. void madd (MINT *src_1. This means that you probably need to give the ‘I<dir>’ option to the compiler.h’ to get the definition of the necessary types and functions. MINT *src_2. MINT *quotient. nullterminated C string. make sure to include GNU ‘mp.a’. MINT *root. signed short int *remainder ) [Function] [Function] Set quotient to dividend/divisor. signed short int divisor. and remainder to dividend mod divisor. Include the header ‘mp. [Function] void msub (MINT *src_1. .0. The ‘enablempbsd’ option must be used when building GNU MP to make these available (see Chapter 2 [Installing GMP]. MINT * itom (signed short int initial_value ) [Function] Allocate an integer consisting of a MINT object and dynamic limb space. Apart from the incomplete set of functions. [Function] Allocate an integer consisting of a MINT object and dynamic limb space. the results are undefined. MINT *src_2.a’ to your program. [Function] void mdiv (MINT *dividend. [Function] void mult (MINT *src_1. MINT *quotient. The original Berkeley MP library has a usage restriction: you cannot use the same variable as both source and destination in a single function call. Return a pointer to the MINT object. i. If you are on a BSD derived system. Initialize the integer from initial value. MINT *destination ) Subtract src 2 from src 1 and put the difference in destination.
SPC and TAB are allowed in the number string. It will be strlen(str)+1 bytes. and a negative value if op1 < op2. [Function] void gcd (MINT *op1. and put the read integer in dest. [Function] int mcmp (MINT *op1.Chapter 13: Berkeley MP Compatible Functions 85 void pow (MINT *base. [Function] Note that the name pow clashes with pow from the standard C math library (see Section “Exponentiation and Logarithms” in The GNU C Library Reference Manual). Also output a newline. An application will only be able to use one or the other. malloc by default. void mfree (MINT *op ) . as a decimal string. and return a pointer to the string. that being exactly enough for the string and nullterminator. This function should only be passed a value returned by itom or xtom. MINT *mod. MINT *op2 ) [Function] Compare op1 and op2. MINT *res ) Set res to the greatest common divisor of op1 and op2. The returned string is allocated using the default memory allocation function. signed short int exp. Return a positive value if op1 > op2. [Function] Input a decimal string from stdin. and are ignored. MINT *op2. MINT *exp. [Function] void min (MINT *dest ) void mout (MINT *src ) Output src to stdout. zero if op1 = op2. [Function] Deallocate. char * mtox (MINT *op ) [Function] Convert op to a hexadecimal string. MINT *dest ) Set dest to (base raised to exp) modulo mod. void rpow (MINT *base. the space used by op. MINT *dest ) Set dest to base raised to exp.
In particular note that allocate function or reallocate function mustn’t return NULL. [Function] void * reallocate_function (void *ptr. void free_function (void *ptr. but of course can be ignored if not needed.1 [Build Options]. and if they fail GMP prints a message to the standard error output and terminates the program. . A byte here means the unit used by the sizeof operator. page 3). size t new_size ) Resize a previously allocated block ptr of old size bytes to be new size bytes. This feature is available in the Berkeley compatibility library (see Chapter 13 [BSD Compatible Functions]. [Function] The block may be moved if necessary or if desired. ptr is never NULL. the corresponding default function is used.86 GNU MP 5. void *(*realloc_func_ptr ) (void *.0. page 84) as well as the main GMP library. size t). realloc and free for memory allocation. and in that case the smaller of old size and new size bytes must be copied to the new location. void (*free_func_ptr ) (void *. apart from temporary space from alloca if that function is available and GMP is configured to use it (see Section 2. These functions will be used for all memory allocation done by GMP. if they return then they must have performed the specified operation.1 14 Custom Allocation By default GMP uses malloc. [Function] The old size parameters to reallocate function and free function are passed for convenience. to allocate memory in a different way or to have a different error action on running out of memory. size t)) [Function] Replace the current allocation functions from the arguments. No error return is allowed from any of these functions. size t. it’s always a previously allocated block of size bytes. that being the new location if moved or just ptr if not. new size may be bigger or smaller than old size. it’s always a previously allocated block. If an argument is NULL. ptr is never NULL. Be sure to call mp_set_memory_functions only when there are no active GMP objects allocated using the previous memory functions! Usually that means calling it before any other GMP function. Alternate functions can be specified. size t old_size. The return value is a pointer to the resized block. The default functions using malloc and friends for instance don’t use them. The functions supplied should fit the following declarations: void * allocate_function (size t alloc_size ) Return a pointer to newly allocated space with at least alloc size bytes. size t size ) Deallocate the space pointed to by ptr. void mp_set_memory_functions ( void *(*alloc_func_ptr ) (size t).
For example. void (*freefunc) (void *. void *(**realloc_func_ptr ) (void *. GMP may use allocated blocks to hold pointers to other allocated blocks. size t). storing function pointers to the locations given by the arguments. A longjmp or throwing a C++ exception will have undefined results. This will limit the assumptions a conservative garbage collection scheme can make. . How much is possible when genuinely out of memory is another question though. size t. mp_get_memory_functions (NULL. void mp_get_memory_functions ( void *(**alloc_func_ptr ) (size t). size_t). If an argument is NULL. for example giving a graphical dialog rather than the default print to stderr. There’s currently no defined way for the allocation functions to recover from an error such as out of memory. This may change in the future.Chapter 14: Custom Allocation 87 Getting a different fatal error action is a good use for custom allocation functions. &freefunc). they must terminate program execution. to get just the current free function. It’s necessary to change the GMP sources if this is a problem. Since the default GMP allocation uses malloc and friends. void (**free_func_ptr ) (void *. those functions will be linked in even if the first thing a program does is an mp_set_memory_functions. NULL. size t)) [Function] Get the current allocation functions. that function pointer is not stored.
win.cdc.haskell. Haskell • Glasgow Haskell Compiler http://www. though perhaps with varying levels of functionality and efficiency.mozartoz.net/ • XEmacs (21.dk/rene/gnu/ Optionally provides an arbitrary precision mpeval.org Optional big integers.informatik.be/~cant/arithmos/ Rationals with infinities and square roots.ua.org/ Objective Caml • MLGMP http://www.sourceforge.fr/~monniaux/programmes. • NTL http://www. C++ • GMP C++ class interface.org/ • Kissme http://kissme. • ALP http://wwwsop.inria. • LiDIA http://www.shoup.org/ . see Chapter 12 [C++ Class Interface].org/software/gcl/gcl. ML • MLton compiler http://mlton.net/ Lisp • GNU Common Lisp http://www. Fortran • Omni F77 http://phase.seindal.1 15 Language Bindings The following packages and projects offer access to GMP from languages other than C. • Arithmos http://www.ginac.html.html • Librep http://librep.jp/Omni/home. Oz • Mozart http://www.kaffe. page 76 Straightforward interface.di. • CLN http://www.tudarmstadt.ens.net/ntl/ A C++ number theory library.en • Numerix http://pauillac.org/ghc/ Java • Kaffe http://www. • Linbox http://www.fr/~quercia/ Optionally using GMP.fr/saga/logiciels/ALP/ Linear algebra and polynomials using templates. expression templates to eliminate temporaries.linalg.88 GNU MP 5.18 beta and up) http://www.5.hpcc. M4 • GNU m4 betas http://www.org/ Sparse vectors and matrices.ac.xemacs.html Arbitrary precision floats.de/TI/LiDIA/ A C++ library for computational number theory. rationals and floats using GMP.0.gnu.inria.sourceforge.de/CLN/ High level classes for arithmetic.
gnu. Perl • GMP module.gnupascal.html Macsyma computer algebra using GCL.org/ Arbitrary precision floats.liu. Pike • mpz module in the standard distribution. • Maxima http://www.stklos. • Math::BigInt::GMP http://www.edu/users/wfs/maxima.org/ Smalltalk • GNU Smalltalk http://www. page 20). • DrGenius http://drgenius. • GiNaC http://www.net/ Topological calculator. http://www.se/ Prolog • SWI Prolog http://www. • Yacas http://www.inria.Chapter 15: Language Bindings 89 Pascal • GNU Pascal Compiler http://www.html Yet another computer algebra system.org/ • GMPY http://gmpy. • Q http://qlang.ma.net/ Scheme • GNU Guile (upcoming 1. but not as many functions as the GMP module above.org/ • STklos http://www.fr/~quercia/ For Free Pascal.10 [Demonstration Programs].de/ C++ computer algebra using CLN.cpan.googoogaga.seul.org/ Compatible with Math::BigInt.ida.swiprolog.html • RScheme http://www.rscheme. • Regina http://regina. • GOO http://www.org/projects/axiom Computer algebra using GCL.sourceforge.cpan.sourceforge.org/software/guile/guile.smalltalk. see ‘demos/perl’ in the GMP sources (see Section 3. .xs4all.python.utexas.ginac. • Numerix http://pauillac.org/ Dynamic object oriented language.de/ GMP unit.org/ Plug Math::GMP into normal Math::BigInt operations.org/versions/GNUSmalltalk.net/ Equational programming system.sourceforge.nongnu.nl/~apinkus/yacas.html Other • Axiom http://savannah.8) http://www. Python • mpz module in the standard distribution. http://pike. optionally using GMP. • Math::GMP http://www.org/ Geometry system and mathematical programming language.
1.90 GNU MP 5. but applications that expect to be compatible with future GMP releases should take care to use only the documented functions. The code is likely to be hard to understand without knowing something about the algorithms. doubled (left shift by one bit). Less than 1. Some GMP internals are mentioned. page 122).1.5×. See Knuth section 4. as the size N increases.1 16 Algorithms This chapter is an introduction to some of the algorithms used for various GMP operations.1 Multiplication N×N limb multiplications and squares are done using one of five algorithms.5× probably indicates mpn_sqr_basecase wants improving on that CPU. depending on operand size (see Section 16. Assembly implementations of mpn_mul_basecase are essentially the same as the generic C code. and then the products on the diagonal added. by using the fact that the cross products above and below the diagonal are the same.c’. and the ‘mpn/generic/mul_basecase. with the SQR thresholds. A triangle of products below the diagonal is formed. but have all the usual assembly tricks and obscurities introduced for speed.0.1 Basecase Multiplication Basecase N×M multiplication is a straightforward rectangular set of crossproducts. This is an O(N M ) algorithm. . it’s usually around 1. Again the assembly implementations take essentially the same approach.c’ code. A square can be done in roughly half the time of a multiply.7 [Unbalanced Multiplication]. u0 u1 u2 u3 u4 d d d d d u0 u1 u2 u3 u4 In practice squaring isn’t a full 2× faster than multiplying. 16.1 algorithm M (see Appendix B [References]. 16.3. page 96). the same as long multiplication done by hand and for that reason sometimes known as the schoolbook or grammar school method. This can be seen in ‘mpn/generic/sqr_basecase. N×M multiplications of operands with different sizes above MUL_TOOM22_THRESHOLD are currently done by special Toominspired algorithms or directly with FFT. Algorithm Basecase Karatsuba Toom3 Toom4 FFT Threshold (none) MUL_TOOM22_THRESHOLD MUL_TOOM33_THRESHOLD MUL_TOOM44_THRESHOLD MUL_FFT_THRESHOLD Similarly for squaring.
With that x = x1 b + x0 and y = y1 b + y0 . The middle term (x1 − x0 )2 is now always positive. and the following holds. ie. But those sums can exceed k limbs. A brief description is given here. whereas a basecase multiply of N×N limbs is equivalent to four multiplies of (N/2)×(N/2).585 ) algorithm. if x0 is k limbs (y0 the same) then b = 2 k∗mp bits per limb. rather than 6k. this will be zero if that routine should be used always. but in GMP extra function call overheads outweigh the saving.3. The 4 2 . Karatsuba multiplication is asymptotically an O(N 1.Chapter 16: Algorithms 91 On some CPUs mpn_mul_basecase can be faster than the generic C mpn_sqr_basecase on some small sizes.3 part A. xy = (b2 + b)x1 y1 − b(x1 − x0 )(y1 − y0 ) + (b + 1)x0 y0 This formula means doing only three multiplies of (N/2)×(N/2) limbs. The SQR threshold is usually about twice the MUL. SQR_BASECASE_THRESHOLD is the size at which to use mpn_sqr_basecase. Notice the sum high(x0 y0 ) + low(x1 y1 ) occurs twice. high x1 y1 x0 y0 low Let b be the power of 2 where the split occurs. and various other textbooks. The inputs x and y are treated as each split into two parts of equal length (or the most significant part one limb shorter if N is odd). so it’s possible to do 5k limb additions. representing 3 multiplies each 1/2 the size of the inputs. A similar formula for both multiplying and squaring can be constructed with a middle term (x1 + x0 )(y1 + y0 ). x2 = (b2 + b)x2 − b(x1 − x0 )2 + (b + 1)x2 1 0 The final result is accumulated from those three squares the same way as for the three multiplies above. Squaring is similar to multiplying. and the sign used to choose to add or subtract. This is a big improvement over the basecase multiply at O(N 2 ) and the advantage soon overcomes the extra additions Karatsuba performs. The basecase algorithm will take a time of the form M (N ) = aN 2 + bN + c and the Karatsuba algorithm K(N ) = 3M (N/2)+dN +e. but with x = y the formula reduces to an equivalent with three squares. high x1 y1 + + − x1 y1 x0 y0 (x1 − x0 )(y1 − y0 ) x0 y0 low The term (x1 − x0 )(y1 − y0 ) is best calculated as an absolute value. 16. MUL_TOOM22_THRESHOLD can be as little as 10 limbs. which expands to K(N ) = 3 aN 2 + 3 bN +3c+dN +e.2 Karatsuba Multiplication The Karatsuba multiplication algorithm is described in Knuth section 4. The factors (b2 + b) etc represent the positions where the three products must be added. the exponent being log 3/ log 2.1. leading to more carry handling and additions than the form above.
ie. 2. which gives w0 immediately (x2 + x1 + x0 )(y2 + y1 + y0 ) low . 1. x1 . and when they are they’ll give the final result using w = W (b). j = 0. giving values of W (t) at those points.1 3 factor 4 for a means percrossproduct speedups in the basecase code will increase the threshold since they benefit M (N ) more than K(N ). Of course all speedups reduce total time. high x2 y2 x1 y1 x0 y0 low These parts are treated as the coefficients of two polynomials X(t) = x2 t2 + x1 t + x0 Y (t) = y2 t2 + y1 t + y0 Let b equal the power of 2 which is the size of the x0 .3. 16. The 3way form used in GMP is described here. Point t=0 t=1 Value x0 y0 . And conversely the 3 for b means linear style 2 speedups of b will increase the threshold since they benefit K(N ) more than M (N ). and the final W (b) will be an addition like. The coefficients will be roughly b2 each. and in that sense the algorithm thresholds are merely of academic interest. w3 = x2 y1 + x1 y2 . X(t) and Y (t) are evaluated and multiplied at 5 points. and would be equivalent merely to a basecase multiply. with an example 3way calculation after Theorem A. In GMP the following points are used. With this x = X(b) and y = Y (b). Instead the following approach is used. if they’re k limbs each then b = 2 k∗mp bits per limb.92 GNU MP 5.3 Toom 3Way Multiplication The Karatsuba formula is the simplest case of a general approach to splitting inputs that leads to both Toom and FFT algorithms. The operands are each considered split into 3 pieces of equal length (or the most significant part 1 or 2 limbs shorter than the other two). but this would need all nine xi yj for i.0. like w4 = x2 y2 . A description of Toom can be found in Knuth section 4. since xy = X(b)Y (b).3.1. w2 = x2 y0 + x1 y1 + x0 y2 etc. y0 and y1 pieces. high w4 w3 w2 w1 w0 The wi coefficients could be formed by a simple set of cross products. Let a polynomial W (t) = X(t)Y (t) and suppose its coefficients are W (t) = w4 t4 + w3 t3 + w2 t2 + w1 t + w0 The wi are going to be determined. The latter can be seen for instance when adding an optimized mpn_sqr_diagonal to mpn_sqr_basecase.
The formula given for the Karatsuba algorithm (see Section 16. This is an improvement over Karatsuba at O(N 1. Squaring follows the same procedure as multiplication. Near the crossover between Toom3 and Karatsuba there’s generally a range of sizes where the difference between the two is small. A graph of time versus size for the two shows the effect. though Toom does more work in the evaluation and interpolation and so it only realizes its advantage above a certain size.2 [Karatsuba Multiplication]. and the fact that only a few recursions of each are being performed. the latter using the special mpn_divexact_by3 (see Section 16.585 ). but there’s only one X(t) and it’s evaluated at the 5 points. The points are arbitrary and can be chosen to make the linear equations come out with a convenient set of steps for quickly isolating the wi . At t = ∞ the value is actually limt→∞ X(t)Y (t) . page 91) has an equivalent for Toom3 involving only five multiplies.465 ). Each of the points substituted into W (t) = w4 t4 + · · · + w0 gives a linear combination of the wi coefficients. representing 5 recursive multiplies of 1/3 the original size each. and those values squared to give values of W (t). the exponent being log 5/ log 3.5 [Exact Division]. An alternate view of Toom3 can be found in Zuras (see Appendix B [References]. The conversion of W (t) values to the coefficients is interpolation. but it’s much easier to think t4 of as simply x2 y2 giving w4 immediately (much like x0 y0 at t = 0 gives w0 immediately). which gives w4 immediately At t = −1 the values can be negative and that’s handled using the absolute values and tracking the sign separately. and some elementary linear algebra quickly isolates each wi . The diagram .Chapter 16: Algorithms 93 t = −1 t=2 t=∞ (x2 − x1 + x0 )(y2 − y1 + y0 ) (4x2 + 2x1 + x0 )(4y2 + 2y1 + y0 ) x2 y2 . page 98). and a couple of divisions by powers of 2 and one division by 3. and the value of those combinations has just been calculated. page 122). The matrix inverses are not meant to be actually used. This involves adding or subtracting one W (t) value from another. Even at large sizes there’s a good chance machine dependent effects like cache architecture will mean actual performance deviates from what might be predicted.1.2. but this would be complicated and unenlightening. MUL_TOOM33_THRESHOLD is a somewhat arbitrary point in that range and successive runs of the tune program can give different values due to small variations in measuring. W (0) = W (1) = w4 W (−1) = w4 W (2) = 16w4 W (∞) = w4 + w3 − w3 + 8w3 + + + w2 w2 4w2 + − + w1 w1 2w1 w0 + w0 + w0 + w0 This is a set of five equations in five unknowns. and in fact the same toom3_interpolate subroutine is used for both squaring and multiplying. due of course to the big influence of all sorts of overheads. Toom3 is asymptotically O(N 1. A polynomial of degree 4 like W (t) is uniquely determined by values known at 5 different points. The interpolation is then identical. At the fairly small sizes where the Toom3 thresholds occur it’s worth remembering that the asymptotic behaviour for Karatsuba and Toom3 can’t be expected to make accurate predictions. using a vector to represent the x and y splits and a matrix multiplication for the evaluation and interpolation stages. and they have elements with values much greater than in fact arise in the interpolation steps. see ‘tune/README’.
wn = i+j=b2k +n b=0. 16. Using the notation from the section on Toom3 multiplication. A brief description of the form used in GMP is given here. A k parameter controls the split. The multiplication done is xy mod 2N + 1.404 ). interpolate and combine similar to that described above for Karatsuba and Toom3. which produces necessary cancellations at the interpolation stage. for instance Knuth section 4. respectively. Descriptions of FFTs in various forms can be found in many textbooks.5 FFT Multiplication At large to very large sizes a Fermat style FFT multiplication is used. occurs for both t = 1 and t = −1. for a given N . giving values of W (t) at those points.0. In GMP the following points are used. we form two polynomials: X(t) = x3 t3 + x2 t2 + x1 t + x0 Y (t) = y3 t3 + y2 t2 + y1 t + y0 X(t) and Y (t) are evaluated and multiplied at 7 points. and interpolation. N must be a multiple of 2k ×mp bits per limb so the split falls on limb boundaries.94 GNU MP 5. evaluate. Toom4 is asymptotically O(N 1.3 part C or Lipson chapter IX. the exponent being log 7/ log 4. so padding to get a full product is unavoidable.1.1 k (−1)b xi yj The points used for the evaluation are g i for i = 0 to 2k − 1 where g = 22N /2 . which gives w6 immediately The number of additions and subtractions for Toom4 is much larger than for Toom3. pointwise multiplications. But several subexpressions occur multiple times. for example x2 + x0 . representing 7 recursive multiplies of 1/4 the original size each. g is a 2k th root of unity mod 2N + 1. but again doesn’t have to be implemented that way and for example with a bit of rearrangement just one division by 6 can be done. pointwise multiply. The evaluations. Point t=0 t = 1/2 t = −1/2 t=1 t = −1 t=2 t=∞ Value x0 y0 . Toom4 analogously splits the operands into 4 coefficients. following Sch¨nhage and o Strassen (see Appendix B [References]. with an FFTk splitting into 2k pieces of M = N/2k bits each. page 122). The modular product is the native form for the algorithm.1 shown for the 3way is attractive. and . 16. are all done modulo 2N + 1 where N is 2M + k + 3 rounded up to a multiple of 2k and of mp_bits_per_limb.4 Toom 4Way Multiplication Karatsuba and Toom3 split the operands into 2 and 3 coefficients. A full product xy is obtained by choosing N ≥ bits(x) + bits(y) and padding x and y with high zero limbs.3. which gives w0 immediately (x3 + 2x2 + 4x1 + 8x0 )(y3 + 2y2 + 4y1 + 8y0 ) (−x3 + 2x2 − 4x1 + 8x0 )(−y3 + 2y2 − 4y1 + 8y0 ) (x3 + x2 + x1 + x0 )(y3 + y2 + y1 + y0 ) (−x3 + x2 − x1 + x0 )(−y3 + y2 − y1 + y0 ) (8x3 + 4x2 + 2x1 + x0 )(8y3 + 4y2 + 2y1 + y0 ) x3 y3 .1. and the choice of N ensures these sums aren’t truncated. The results of interpolation will be the following negacyclic convolution of the input pieces. The algorithm follows a split. avoiding bit shifts in the split and combine stages.
The interpolation is an inverse fast Fourier transform. 16. In practice it’s been found each k is used at quite small multiples of its size constraint and so the step effect is quite noticeable in a time versus size graph.6 Other Multiplication The Toom algorithms described above (see Section 16. meaning one which isn’t padded through rounding. page 92. Karatsuba or basecase). page 94) generalizes to split into an arbitrary . In practice MUL_ FFT_THRESHOLD and SQR_FFT_THRESHOLD have been found to be in the k = 8 range. but for the purposes of considering where an FFT might be first used it can be assumed that the FFT is recursing into a normal multiply and that on that basis it’s doing 2k recursed multiplies each 1/2k−2 the size of the inputs. Or for k = 9 groups of 2048 limbs. In the code. Squaring is the same. but overheads mean each is only faster at bigger and bigger sizes. whichever is optimal at the size N . Each successive k is an asymptotic improvement. but this is suboptimal since at the start of a new step it can happen that it’s better to go back to the previous k for a while. When an FFT is to give a full product. Something more sophisticated for MUL_FFT_TABLE and SQR_FFT_TABLE will be needed. adds and negations.1. see Section 16. so for a 32bit limb there’ll be 512 limb groups of sizes for which mpn_mul_n runs at the same speed. So far it’s been found that only very large FFTs recurse into pointwise multiplies above these sizes. but x is the only input so it’s one transform at the evaluate stage and the pointwise multiplies are squares. MUL_FFT_TABLE and SQR_FFT_TABLE are the thresholds where each k is used. The pointwise multiplications are done modulo 2N + 1 and either recurse into a further FFT or use a plain multiplication (Toom3. adds and overheads.333 ) can be expected to be the first faster than Toom3 at O(N 1.4 [Toom 4Way Multiplication]. For a mod 2N + 1 product.4 ) would be the first FFT faster than Toom3. The +k + 3 means some values of N just under such a multiple will be rounded to the next.3 [Toom 3Way Multiplication]. k = 10 groups of 8192 limbs.Chapter 16: Algorithms 95 it’s also a power of 2 so the fast Fourier transforms used for the evaluation and interpolation do only shifts. making it O(N k/(k−2) ).1. so an FFT and Toom3 etc can be compared directly. depending on the CPU. The interpolation is the same.1. somewhere between 3000 and 10000 limbs. The practical effect of the 22k−1 constraint is to introduce a stepeffect into measured speeds. and it’s also assumed that the extra +k + 3 bits are negligible at typical FFT sizes. This would mean k = 7 at O(N 1.465 ). For example k = 8 will round N up to a multiple of 32768 bits. In practice this is what’s found. A mod 2N +1 product can be formed with a normal N ×N → 2N bit multiply plus a subtraction. A k = 4 FFT at O(N 1. The resulting set of sums of xi yj are added at appropriate offsets to give the final result. etc. the exponent representing 2k recursed modular multiplies each 1/2k−1 the size of the original. The complexity calculations above assume that a favourable size is used. with MUL_FFT_ MODF_THRESHOLD and SQR_FFT_MODF_THRESHOLD being between 300 and 1000 limbs. the change of N to 2N doesn’t alter the theoretical complexity for a given k. an FFTk is an O(N k/(k−1) ) algorithm. The threshold determinations currently measure at the midpoints of size steps. Each new k effectively swaps some multiplying for some shifts. The way N is split into 2k pieces and then 2M + k + 3 is rounded up to a multiple of 2k and mp_bits_per_limb means that when 2k ≥ mp bits per limb the effective N is a multiple of 22k−1 bits.
. Asymptotically an (r + 1)way algorithm is O(N log(2r+1)/log(r+1) . An occasional difference of 1 in the last bit might not matter to a typical signal processing algorithm. page 122) and is implemented as udiv_qrnnd_preinv in ‘gmpimpl.1 Single Limb Division N×1 division is implemented using repeated 2×1 divisions from high to low. we invoke FFT directly. and has a further saving of nearly half the interpolate steps..1. For really large operands. GMP currently splits the smaller operand onto 2 coefficients.1 [Basecase Multiplication]. but no such root exists for plain integers. . or 4 coefficients. 5way does 9. but the larger operand can be split into 2.e. The multiply by inverse follows “Improved division by invariant integers” by M¨ller and o Granlund (see Appendix B [References].2. but exercise 4 uses −r.. a polynomial of degree 1.r and the latter saves some small multiplies in the evaluate stage (or rather trades them for additions). The idea is to have a fixedpoint approximation to 1/d (see invert_limb) and then multiply by the high limb (plus one bit) of the dividend to get a quotient q. Some processors have special support for such FFTs. Going to complex integers with √ i = −1 doesn’t help. Floating point FFTs use complex numbers approximating Nth roots of unity. but the time spent in the evaluate and interpolate stages grows with r and has a significant practical impact.1. and reveals whether q or q − 1 is correct. we use Toom inspired algorithms suggested by Alberto Zanoni and Marco Bodrato. A 4way split does 7 pointwise multiplies. etc. i.1 number of pieces.2 Division Algorithms 16. The existence of 2k th roots of unity in a suitable ring or field lets the fast Fourier transform keep splitting and get to O(N log r). But these are not used in GMP since it’s very difficult to guarantee an exact result (to some number of bits).3 algorithm C.e. with the asymptotic advantage of each r realized only at bigger and bigger sizes. The idea is to split the operands into polynomials of different degree. page 90). . Splitting odd and even parts through positive and negative points can be thought of as using −1 as a square root of unity.3.2. a polynomial of degree 1 to 3.. With d normalized (high bit set). Subtracting qd from the dividend gives a remainder. . This is not currently used. both below MUL_TOOM22_THRESHOLD are done with plain schoolbook multiplication (see Section 16. 3. . but is of course of vital importance to GMP. i. The overheads grow as O(N r). Knuth algorithm C evaluates at points 0. In general a split into r + 1 pieces is made. .2r.0.. essentially because in Cartesian form it takes three real multiplies to do a complex multiply. 16. 16. whichever is best on a given CPU. The notes here are merely for interest. and evaluations and pointwise multiplications done at 2r + 1 points. The divisors at step C7 become j 2 and the multipliers at C8 become 2tj − j 2 . Only the pointwise multiplications count towards bigO complexity.96 GNU MP 5. If a 4th root of unity was available then a further split and speedup would be possible.1. either with a hardware divide instruction or a multiplication by inverse.h’. . For operands between these sizes. . as per Knuth section 4. q is no more than 1 too small.. The idea is to separate odd and even final coefficients and then perform algorithm C steps C7 and C8 on them separately. .7 Unbalanced Multiplication Multiplication of operands with different sizes. . whereas in an r = 2k FFT they grow only as O(N log r).0.
1 [Multiplication Algorithms]. or in fact if the only multiply is a half limb. but that happens rarely. Done this way the quotient limbs come out aligned ready to store. The bit shifts for the dividend are usually accomplished “on the fly” meaning by extracting the appropriate bits at each step. but using N/2 limbs as a base. though the cost of calculating the inverse at the start may mean it’s only better on inputs bigger than say 4 or 5 limbs. and Burnikel and Ziegler (see Appendix B [References]. page 96) whichever is faster.2 [Basecase Division]. this is an O(QM ) algorithm and will run at a speed similar to a basecase Q×M multiplication. 16. which can be a saving on CPUs with a fast half limb multiply. The net effect is to process two limbs with roughly the same two multiplies worth of latency that one limb at a time gives. but the four 1×1s to do that are independent and can therefore be done partly or wholly in parallel.1 algorithm D.c’. while the dividend remains larger than the divisor. See Knuth section 4. Likewise for a 2×1 calculating qd. division is done by dividing. When only the remainder is wanted.3 Divide and Conquer Division For divisors larger than DC_DIV_QR_THRESHOLD. but the inverse is two limbs and the divisor treated as if padded with a low zero limb. requiring an addback of the divisor. In this case the 1×1 multiply for the inverse effectively becomes two 2 × 1 for each limb.Chapter 16: Algorithms 97 The result is a division done with two multiplications and four or five arithmetic operations.2. This can help on CPUs with poor bit shifts or few registers. The two “digits” of the quotient are formed by recursive N×(N/2) divisions. 16. each quotient limb can be formed with a 2×1 division and a 1×1 multiplication plus some subtractions. and ‘mpn/generic/sb_divrem_mn. A similar approach in reverse can be taken to process just half a limb at a time if the divisor is 1 only a half limb. The 2×1 division is by the high limb of the divisor and is done either with a hardware divide or a multiply by inverse (the same as in Section 16. either for the generic C __udiv_qrnnd_c or the multiply by inverse. the division performed is actually a2k by d2k where a is the dividend and k is the power necessary to have the high bit of d2k set.3. page 90). but in base 2 mp bits per limb. not just a single limb. since the inverse will need a 2×2 multiply.2. The calculation is basically the same. differing in fact only in the extra multiply and divide for each of the Q quotient limbs. With a normalized divisor (most significant bit set). The algorithm consists essentially of recognising that a 2N×N division can be done with the basecase division algorithm (see Section 16. Briefly stated. This means more work. though the extra work to apply the inverse will almost certainly soon reach the limits of multiplier throughput. This way the multiplications that arise are (N/2)×(N/2) and can take advantage of Karatsuba and higher multiplication algorithms (see Section 16. Jebelean.2. page 97). The multiply by inverse can be done two limbs at a time. On CPUs with low latency multipliers this can be much faster than a hardware divide. an alternative is to take the dividend limbs unshifted and calculate r = a mod d2k followed by an extra final step r2k mod d2k . When a divisor must be normalized. and especially if it’s not pipelined.2. .2 Basecase Division Basecase N×M division is like long division done by hand. Such a quotient is sometimes one too big. This extends to 3 or 4 limbs at a time. a high quotient limb is formed and the N×1 product qd subtracted at the top end of the dividend. Or to be precise by a recursive divide and conquer algorithm based on work by Moenck and Borodin.1 [Single Limb Division]. With Q=N−M being the number of quotient limbs. page 122).
With higher algorithms the M (N ) term improves and the multiplier tends to log N . or when Q ≤ M to Q(Q − 1)/2. So the quotient is 678. since 3×7 ≡ 1 mod 10. For an N×M exact division producing Q=N−M quotient limbs.98 GNU MP 5. Because the low digit of the dividend is 4. or if Q is small then the crossproducts reduce to a small number.1 If the (N/2)×(N/2) multiplies are done with a basecase multiplication then the work is about the same as a basecase division. the crossproducts are reduced when Q > M to QM −M (M +1)/2. ‘tune/modlinv. as can be seen near the end of section 2. These overheads mean that it’s only when N/2 is above MUL_ TOOM22_THRESHOLD that divide and conquer is of use. For example. leaving 325800.5 Exact Division A socalled exact division is when the dividend is known to be an exact multiple of the divisor.2. within the Toom3 range. On a 2N×N division like this one. subtracting 6×543 = 3258 leaving 0. page 122). so it will be somewhere above twice MUL_TOOM22_THRESHOLD.0. the inverse will be just n/2 limbs. the saving over a normal basecase division is in two parts. the divisor is inverted to a precision determined by the relative size of the dividend and divisor. And finally at the third digit with quotient digit 6 (8×7 mod 10). This is arrived at from 4×7 mod 10. Our blockwise algorithm computes a smaller inverse than in the plain Barrett algorithm. not a 2×1 divide and multiply. a 2N×N division is about 2 to 4 times slower than an N×N multiplication. Notice the low digit has become zero. The actual time is a sum over multiplications of the recursed sizes. using the fact 7 is the modular inverse of 3 (the low digit of the divisor). This does four multiplies for a 32bit limb. Firstly. If Q is big then many divisions are saved. In practice. 16. So 8×543 = 4344 can be subtracted from the dividend leaving 363810. Notice however that the multiplies and subtractions don’t need to extend past the low three digits of the dividend. Notice the savings are complementary. Jebelean’s exact division algorithm uses this knowledge to make some significant optimizations (see Appendix B [References]. Divide and conquer is asymptotically O(M (N ) log N ) where M (N ) is the time for an N×N multiplication done with FFTs. Blocks of quotient limbs are then generated by multiplying blocks from the dividend by the inverse.2 of Burnikel and Ziegler. subtracting 7×543 = 3801. An optimized mpn_mul_ basecase can lower DC_DIV_QR_THRESHOLD a little by offering a readymade advantage over repeated mpn_submul_1 calls. 16. For the last quotient digit no subtraction is needed at all. but how much above depends on the CPU.h’. The procedure is repeated at the second digit. each of the Q quotient limbs needs only one multiply. The idea can be illustrated in decimal for example with 368154 divided by 543.4 BlockWise Barrett Division For the largest divisions. divide and conquer is 2. the low digit of the quotient must be 8. at moderate to large sizes. The modular inverse used is calculated efficiently by binvert_limb in ‘gmpimpl.2. or six for a 64bit limb. Secondly. Here. . a blockwise Barrett division algorithm is used. with the next quotient digit 7 (1×7 mod 10). For a 2n/n division.63M (N ). since that’s enough to determine the three quotient digits.c’ has some alternate implementations that might suit processors better at bit twiddling than multiplying. DC_DIV_QR_THRESHOLD is based on the divisor size N. only about half the work of a normal basecase division is necessary. but with more function call overheads and with some subtractions separated from the multiplies.
mpn_divexact_by3. r will be in the range 0 ≤ r < d and can be viewed as a remainder. and shifting the remainder back down at the end of the calculation.2 [Modular Powering Algorithm]. If however bn ≡ 1 mod d. Montgomery’s REDC method for modular multiplications uses operands of the form of xb−n and yb−n and on calculating (xb−n )(yb−n ) uses the factor of bn in the exact remainder to reach a product in the same form (xy)b−n (see Section 16. hence the use of mpn_modexact_ 1_odd by mpn_gcd_1 and mpz_kronecker_ui etc (see Section 16. A special case exact division by 3 exists in mpn_divexact_by3. quotient q. but operating from low to high. The multiplications don’t need to be on the dependent chain. 16. The factor of bn on r can be ignored in a GCD when d is odd.. which can help chips with pipelined multipliers. Notice that r generally gives no useful information about the ordinary remainder a mod d since bn mod d could be anything. page 97). page 100). and b = 2 mp bits per limb. and this leads to divisibility or congruence tests which are potentially more efficient than a normal division. For a 32 or 64 bit limb other such factors include 5. Carrying out full subtractions at each stage means the same number of cross products must be done as a normal division. then low zero limbs are produced but with a remainder in the high limbs. It uses a rearrangement similar to the divide and conquer for normal division (see Section 16.3 [Greatest Common Divisor Algorithms]. that being the number of limbs produced for q.6 Exact Remainder If the exact division algorithm is done with a full subtraction at each stage and the dividend isn’t a multiple of the divisor. It forms quotient digits with a multiply by the modular inverse of 3 (which is 0xAA. but there’s still some single limb divisions saved. supporting Toom3 multiplication and mpq canonicalizations.4. then r is the negative of the ordinary remainder. A further possibility not currently implemented is “Bidirectional Exact Integer Division” by Krandick and Jebelean which forms quotient limbs from both the high and low ends of the dividend.Chapter 16: Algorithms 99 The subquadratic exact division described by Jebelean in “Exact Division with Karatsuba Complexity” is not currently implemented.AAB) and uses two comparisons to determine a borrow for the next limb. and can halve once more the number of crossproducts needed in a 2N×N division. 16. leading to some negations in the above formula.3 [Divide and Conquer Division]. For dividend a. divisor d. but all are essentially the same. This occurs whenever d is a factor of bn − 1. this remainder r is of the form a = qd + rbn n represents the number of zero limbs produced by the subtractions.7 Small Quotient Division An N×M division where the number of quotient limbs Q=N−M is small can be optimized somewhat.2. mpn_modexact_1_odd and the mpn_redc_X functions differ subtly in how they return r. Clearly r is zero when a is a multiple of d.2. but no particular use has been found for this. as for example with 3 in mpn_divexact_ by3. but one shifted up by a factor of bn . This is wasteful if only a few quotient limbs are to be formed. page 103).2. as long as the effect of the borrows is applied. shifting the dividend accordingly. Instead a division . providing good speedups on a number of processors. An ordinary basecase division normalizes the divisor by shifting it to make the high bit set. 17 and 257. When d is a single limb some simplifications arise.
so counting or testing can begin on a − b without waiting for abs (a − b) to be determined.3 Theorem E. u − qv. The binary algorithm has so far been found to be faster than the Euclidean algorithm everywhere. b) strip factors of 2 from a The Euclidean GCD algorithm. so often only one or two subtractions are needed to get the same effect as a division.100 GNU MP 5. but all cases of 2 too big and most cases of 1 too big are detected by first comparing the most significant limbs that will arise from the subtraction.3. b by v.7% of the time. min (a. meaning b is much smaller than a. Quotients 1. b = abs (a − b). This whole procedure is essentially the same as one step of the basecase algorithm done in a Q limb base. b = abs (a − b). One reason the binary method does well is that the implied quotient at each step is usually small. since the condition is data dependent. min (a. so an option is to strip one or two bits arithmetically then loop for more (as done for AMD K6). A loop stripping low zero bits tends not to branch predict well.0.5. as per Knuth algorithms E and A. For two Nbit operands. though with the trial quotient test done only with the high limbs. This is the basis for the initial a mod b reductions in mpn_gcd and mpn_gcd_1 (the latter for both N×1 and 1×1 cases). Firstly it may be noted that in twos complement the number of low zero bits on a − b is the same as b − a. repeatedly computes the quotient q = a/b and replaces a. not an entire Q limb “digit” product. This is described in many textbooks.1 Binary GCD At small sizes GMP uses an O(N 2 ) binary style GCD.3.1 of just the top 2Q limbs of the dividend by the top Q limbs of the divisor can be used to form a trial quotient. But on average there’s only a few low zeros. It simply consists of successively reducing odd operands a and b using a. The final 1 × 1 GCD in mpn_gcd_1 is done in the generic C code as described above. A multiply and subtract then applies the trial quotient to the M−Q unused limbs of the divisor and N−Q dividend limbs (which includes Q limbs remaining from the trial quotient division). 16. The starting trial quotient can be 1 or 2 too big.68 iterations per bit. An addback is done if the quotient still turns out to be 1 too big. for example Knuth section 4. the algorithm takes about 0. But after that initial reduction. b) a = a/2 if even b = b/2 if even .1 exercise 20 but with the v2 ^ > b^ + u2 condition appropriately q r relaxed. For optimum performance some attention needs to be paid to the way the factors of 2 are stripped from a. see Knuth section 4. then a division is worthwhile. Or use a lookup table to get a count for several bits then loop for more (as done for AMD K7). The correctness of this weaker test can be established by following the argument of Knuth section 4. big quotients occur too rarely to make it worth checking for them.2 algorithm B. 2 and 3 for example occur 67.5. not the whole of the divisor and dividend.3 Greatest Common Divisor 16. When the implied quotient is large. This requires only those limbs normalized. An alternative approach is to keep just one of a or b odd and iterate a.
Chapter 16: Algorithms
101
This requires about 1.25 iterations per bit, but stripping of a single bit at each step avoids any branching. Repeating the bit strip reduces to about 0.9 iterations per bit, which may be a worthwhile tradeoff. Generally with the above approaches a speed of perhaps 6 cycles per bit can be achieved, which is still not terribly fast with for instance a 64bit GCD taking nearly 400 cycles. It’s this sort of time which means it’s not usually advantageous to combine a set of divisibility tests into a GCD. Currently, the binary algorithm is used for GCD only when N < 3.
16.3.2 Lehmer’s algorithm
Lehmer’s improvement of the Euclidean algorithms is based on the observation that the initial part of the quotient sequence depends only on the most significant parts of the inputs. The variant of Lehmer’s algorithm used in GMP splits off the most significant two limbs, as suggested, e.g., in “A DoubleDigit LehmerEuclid Algorithm” by Jebelean (see Appendix B [References], page 122). The quotients of two doublelimb inputs are collected as a 2 by 2 matrix with singlelimb elements. This is done by the function mpn_hgcd2. The resulting matrix is applied to the inputs using mpn_mul_1 and mpn_submul_1. Each iteration usually reduces the inputs by almost one limb. In the rare case of a large quotient, no progress can be made by examining just the most significant two limbs, and the quotient is computing using plain division. The resulting algorithm is asymptotically O(N 2 ), just as the Euclidean algorithm and the binary algorithm. The quadratic part of the work are the calls to mpn_mul_1 and mpn_submul_1. For small sizes, the linear work is also significant. There are roughly N calls to the mpn_hgcd2 function. This function uses a couple of important optimizations: • It uses the same relaxed notion of correctness as mpn_hgcd (see next section). This means that when called with the most significant two limbs of two large numbers, the returned matrix does not always correspond exactly to the initial quotient sequence for the two large numbers; the final quotient may sometimes be one off. • It takes advantage of the fact the quotients are usually small. The division operator is not used, since the corresponding assembler instruction is very slow on most architectures. (This code could probably be improved further, it uses many branches that are unfriendly to prediction). • It switches from doublelimb calculations to singlelimb calculations halfway through, when the input numbers have been reduced in size from two limbs to one and a half.
16.3.3 Subquadratic GCD
For inputs larger than GCD_DC_THRESHOLD, GCD is computed via the HGCD (Half GCD) function, as a generalization to Lehmer’s algorithm. Let the inputs a, b be of size N limbs each. Put S = N/2 + 1. Then HGCD(a,b) returns a transformation matrix T with nonnegative elements, and reduced numbers (c; d) = T −1 (a; b). The reduced numbers c, d must be larger than S limbs, while their difference abs(c − d) must fit in S limbs. The matrix elements will also be of size roughly N/2. The HGCD base case uses Lehmer’s algorithm, but with the above stop condition that returns reduced numbers and the corresponding transformation matrix halfway through. For inputs larger than HGCD_THRESHOLD, HGCD is computed recursively, using the divide and conquer algorithm in “On Sch¨nhage’s algorithm and subquadratic integer GCD computation” by M¨ller o o (see Appendix B [References], page 122). The recursive algorithm consists of these main steps. • Call HGCD recursively, on the most significant N/2 limbs. Apply the resulting matrix T1 to the full numbers, reducing them to a size just above 3N/2.
102
GNU MP 5.0.1
• Perform a small number of division or subtraction steps to reduce the numbers to size below 3N/2. This is essential mainly for the unlikely case of large quotients. • Call HGCD recursively, on the most significant N/2 limbs of the reduced numbers. Apply the resulting matrix T2 to the full numbers, reducing them to a size just above N/2. • Compute T = T1 T2 . • Perform a small number of division and subtraction steps to satisfy the requirements, and return. GCD is then implemented as a loop around HGCD, similarly to Lehmer’s algorithm. Where Lehmer repeatedly chops off the top two limbs, calls mpn_hgcd2, and applies the resulting matrix to the full numbers, the subquadratic GCD chops off the most significant third of the limbs (the proportion is a tuning parameter, and 1/3 seems to be more efficient than, e.g, 1/2), calls mpn_hgcd, and applies the resulting matrix. Once the input numbers are reduced to size below GCD_DC_THRESHOLD, Lehmer’s algorithm is used for the rest of the work. The asymptotic running time of both HGCD and GCD is O(M (N ) log N ), where M (N ) is the time for multiplying two N limb numbers.
16.3.4 Extended GCD
The extended GCD function, or GCDEXT, calculates gcd (a, b) and also cofactors x and y satisfying ax + by = gcd(a,b). All the algorithms used for plain GCD are extended to handle this case. The binary algorithm is used only for singlelimb GCDEXT. Lehmer’s algorithm is used for sizes up to GCDEXT_DC_THRESHOLD. Above this threshold, GCDEXT is implemented as a loop around HGCD, but with more bookkeeping to keep track of the cofactors. This gives the same asymptotic running time as for GCD and HGCD, O(M (N ) log N ) One difference to plain GCD is that while the inputs a and b are reduced as the algorithm proceeds, the cofactors x and y grow in size. This makes the tuning of the choppingpoint more difficult. The current code chops off the most significant half of the inputs for the call to HGCD in the first iteration, and the most significant two thirds for the remaining calls. This strategy could surely be improved. Also the stop condition for the loop, where Lehmer’s algorithm is invoked once the inputs are reduced below GCDEXT_DC_THRESHOLD, could maybe be improved by taking into account the current size of the cofactors.
16.3.5 Jacobi Symbol
mpz_jacobi and mpz_kronecker are currently implemented with a simple binary algorithm similar to that described for the GCDs (see Section 16.3.1 [Binary GCD], page 100). They’re not very fast when both inputs are large. Lehmer’s multistep improvement or a binary based multistep algorithm is likely to be better. When one operand fits a single limb, and that includes mpz_kronecker_ui and friends, an initial reduction is done with either mpn_mod_1 or mpn_modexact_1_odd, followed by the binary algorithm on a single limb. The binary algorithm is well suited to a single limb, and the whole calculation in this case is quite efficient. In all the routines sign changes for the result are accumulated using some bit twiddling, avoiding table lookups or conditional jumps.
16.4 Powering Algorithms
16.4.1 Normal Powering
Normal mpz or mpf powering uses a simple binary algorithm, successively squaring and then multiplying by the base when a 1 bit is seen in the exponent, as per Knuth section 4.6.3. The
Chapter 16: Algorithms
103
“left to right” variant described there is used rather than algorithm A, since it’s just as easy and can be done with somewhat less temporary memory.
16.4.2 Modular Powering
Modular powering is implemented using a 2k ary sliding window algorithm, as per “Handbook of Applied Cryptography” algorithm 14.85 (see Appendix B [References], page 122). k is chosen according to the size of the exponent. Larger exponents use larger values of k, the choice being made to minimize the average number of multiplications that must supplement the squaring. The modular multiplies and squares use either a simple division or the REDC method by Montgomery (see Appendix B [References], page 122). REDC is a little faster, essentially saving N single limb divisions in a fashion similar to an exact remainder (see Section 16.2.6 [Exact Remainder], page 99).
16.5 Root Extraction Algorithms
16.5.1 Square Root
Square roots are taken using the “Karatsuba Square Root” algorithm by Paul Zimmermann (see Appendix B [References], page 122). An input n is split into four parts of k bits each, so with b = 2k we have n = a3 b3 +a2 b2 +a1 b+a0 . Part a3 must be “normalized” so that either the high or second highest bit is set. In GMP, k is kept on a limb boundary and the input is left shifted (by an even number of bits) to normalize. The square root of the high two parts is taken, by recursive application of the algorithm (bottoming out in a onelimb Newton’s method), s , r = sqrtrem (a3 b + a2 ) This is an approximation to the desired root and is extended by a division to give s,r, q, u = divrem (r b + a1 , 2s ) s=sb+q r = ub + a0 − q 2 The normalization requirement on a3 means at this point s is either correct or 1 too big. r is negative in the latter case, so if r < 0 then r ← r + 2s − 1 s←s−1 The algorithm is expressed in a divide and conquer form, but as noted in the paper it can also be viewed as a discrete variant of Newton’s method, or as a variation on the schoolboy method (no longer taught) for square roots two digits at a time. If the remainder r is not required then usually only a few high limbs of r and u need to be calculated to determine whether an adjustment to s is required. This optimization is not currently implemented.
3 In the Karatsuba multiplication range this algorithm is O( 2 M (N/2)), where M (n) is the time to multiply two numbers of n limbs. In the FFT multiplication range this grows to a bound of O(6M (N/2)). In practice a factor of about 1.5 to 1.8 is found in the Karatsuba and Toom3 ranges, growing to 2 or 3 in the FFT range.
The algorithm does all its calculations in integers and the resulting mpn_sqrtrem is used for both mpz_sqrt and mpf_sqrt. The extended precision given by mpf_sqrt_ui is obtained by padding with zero limbs.
mpz_perfect_square_p first tests the input mod 256. page 104. then only roots which are divisors of e need to be considered.c’ program and applied with an mpn_mod_1. where A is the input and n is the root to be taken. so 82. A square root must still be taken for any value that passes these tests. Moduli are also combined to save operations.2 [Nth Root Algorithm].5. Big inputs would probably benefit from more residue testing. just one multiply each is required. but this is not currently implemented. In any case each modulus is applied to the mpn_mod_34lsub1 or mpn_mod_1 remainder and a table lookup identifies nonsquares. 5. small inputs might be better off with less. ‘genpsqr. 16. The assumed distribution of squares versus nonsquares in the input would affect such considerations. for a total 99. When nails are in use moduli are instead selected by the ‘genpsqr.5.8% of inputs can be immediately identified as nonsquares. and such a remainder can be quickly taken just using additions (see mpn_mod_34lsub1).c’ does all the precalculations. so long as the lookup tables don’t become too big. To this end divisibility by a set of small primes is checked.4 Perfect Power Detecting perfect powers is required by some factorization algorithms.) If a prime divisor p with multiplicity e can be found. a pseudosquare to all the tested bases). for a total 99. and suitably permuted tables. By using a “modexact” style calculation.1 16.5. The current implementation is not particularly well optimized. 7. Only 44 different values occur for squares mod 256.104 GNU MP 5. On a 64bit system 97 is tested too. Clearly more residue tests could be done.2 Nth Root Integer Nth roots are taken using Newton’s method with the following iteration.3 Perfect Square A significant fraction of nonsquares can be quickly identified by checking whether the input is a quadratic residue modulo small integers. 13 and 17.0. These moduli are chosen because they’re factors of 224 − 1 (or 248 − 1 for 64bits). much reducing the work necessary. 1 n A an−1 i ai+1 = + (n − 1)ai The initial approximation a1 is generated bitwise by successively powering a trial root with or without new 1 bits.62%. The iteration converges quadratically when started from a good approximation. though naturally only prime roots need to be considered. When n is large more initial bits are needed to get good convergence. (See Section 16. On a 32bit system similar tests are done mod 9. mpz_perfect_square_p only uses a compact and efficient set. The same 224 − 1 or 248 − 1 could be done with nails using some extra bit shifts. which means just examining the low byte. to verify it’s really a square and not one of the small fraction of nonsquares that get through (ie. . see the code for details. 16.25% of inputs identified as nonsquares. Currently mpz_perfect_ power_p is implemented using repeated Nth root extractions.5. aiming to be just above the true root.
the point where it becomes worth doing a big division to cut the input in half. or at least be sure not to underestimate that. √ i powers bn2 of the radix are calculated. and a remainder which is those below. The result is either correct or one too big. Some brief experiments with it on the base case when recursing didn’t give a noticeable improvement. Sizes below GET_STR_ PRECOMPUTE_THRESHOLD use a basic O(N 2 ) method. . but an exact result every time would not be practical. Examining some of the high bits of the input could increase the chance of getting the exact number of digits. which is the case when recursing. Something similar would work for the subquadratic divisions too. For an input t. rounding up. The advantage of this algorithm is that big divisions can make use of the subquadratic divide and conquer division (see Section 16.6 Radix Conversion Radix conversions are less important than other algorithms.6. . This is O(N 2 ) and is certainly not optimal. . Repeated divisions by bn are made. it multiplies or divides by a power of b to move the radix point to the just above the highest nonzero digit (or at worst one above that location). it’s necessary to calculate in advance how many digits there will be. allowing multiplications by 10 to optimize to shifts and adds. Above GET_STR_PRECOMPUTE_THRESHOLD a subquadratic algorithm is used. GET_STR_ PRECOMPUTE_THRESHOLD includes the cost of calculating the radix power required. an extra divide step is done to give a fractional limb representing r/bn . since in general the difference between numbers 100. mpf_get_str doesn’t currently use the algorithm described here. . but perhaps that was only due to the implementation. The r/bn scheme described above for using multiplications to bring out digits might be useful for more than a single limb. When a piece has been divided down to less than GET_STR_DC_THRESHOLD limbs. Conversions from binary to other radices use one of two algorithms. t is then divided by that largest power. A program dominated by conversions should probably use a different data representation. where b is the radix and n is the biggest power that fits in a limb. GET_STR_PRECOMPUTE_THRESHOLD and GET_STR_DC_THRESHOLD represent the same basic thing. 16. and big divisions tend to have less overheads than lots of separate single limb divisions anyway. page 97). . and 99. The digits of r can then be extracted using multiplications by b rather than divisions. . until a power between t and t is reached. But in any case i the cost of calculating the powers bn2 must first be overcome. then multiplies by bn to bring out digits. though there would be the cost of calculating a bigger radix power.1 Binary to Radix Conversions from binary to a powerof2 radix use a simple and fast O(N ) bit extraction algorithm. giving a quotient which is the digits above that power. Special case code is provided for decimal. the basecase algorithm described above is used.Chapter 16: Algorithms 105 16.3 [Divide and Conquer Division]. Since the base case produces digits from least to most significant but they want to be stored from most to least. might well be almost as much as a full conversion.2. is only in the last few bits and the work to identify 99. whereas GET_ STR_DC_THRESHOLD assumes that’s already available. These two parts are in turn divided by the second highest power. and so on recursively. . For GMP the number of input bits is multiplied by chars_per_bit_exactly from mp_bases. But instead of simply using the remainder r from such divisions.
page 35) first does some trial division by small factors and then uses the MillerRabin probabilistic primality testing algorithm. First groups of n digits are converted into limbs. where x and y are the limbs. .5. and with n = q2k + 1 where q is odd. page 122). Conversions from other radices use one of two algorithms.9 [Number Theoretic Functions]. This saves multiprecision operations. That ought to smooth out a graph of times against sizes. or more). and on some processors much bigger still. SET_STR_PRECOMPUTE_THRESHOLD usually ends up quite big.1 Prime Testing The primality testing in mpz_probab_prime_p (see Section 5. page 122).3.6.0. giving the compiler a chance to optimize multiplications by 10. that being the result. it currently describes the algorithms used before GMP 4. No n is a strong pseudoprime to more than 1/4 of all bases (see Knuth exercise 22). SET_STR_PRECOMPUTE_THRESHOLD is based on the input digits (and tuned for decimal). for 1 ≤ j ≤ k. where n is the biggest power of the base b which will fit in a limb. Any prime n will pass the test.1 Another possible improvement for the subquadratic part would be to arrange for radix powers that balanced the sizes of quotient and remainder produced. if not then n is definitely composite. Sizes below SET_STR_PRECOMPUTE_ THRESHOLD use a basic O(N 2 ) method. ie. Conversions from a powerof2 radix into binary use a simple and fast O(N ) bitwise concatenation algorithm. but 1/4 is all that’s proven for an arbitrary n. allowing i Karatsuba and higher algorithms to be used. But that sort of count isn’t used by the base case and so would need some sort of initial calculation or estimate. then those groups are accumulated into the result by multiplying by bn and adding.2 Radix to Binary This section needs to be rewritten. so as to be independent of the base. If so then n is probably prime. as described in Knuth section 4. But the cost of calculating the powers bn2 must be overcome. 16.4 algorithm P (see Appendix B [References]. Then adjacent limbs are combined into limb pairs with xbn + y. In fact strong pseudoprimes are quite rare. The main reason SET_STR_PRECOMPUTE_THRESHOLD is so much bigger than the corresponding GET_STR_PRECOMPUTE_THRESHOLD is that mpn_mul_1 is much faster than mpn_divrem_1 (often by a factor of 5. hence with x chosen at random there’s no more than a 1/4 chance a “probable prime” will in fact be composite. though it might be better based on a limb count. Groups of n digits are converted to limbs. or an xq2 mod n is 1.4 part E (see Appendix B [References].106 GNU MP 5.7 Other Algorithms 16. the highest power would be an √ bnk approximately equal to t. The advantage of this method is that the multiplications for each x are big blocks. Some special case code is provided for decimal. not restricted to a 2i factor.7. Above SET_STR_PRECOMPUTE_THRESHOLD a subquadratic algorithm is used. but some composites do too. Adjacent limb pairs are combined into quads similarly with xb2n +y. as per Knuth section 4. making the test much more powerful than this analysis would suggest. 16. this algorithm selects a random base j x and tests whether xq mod n is 1 or −1. but may or may not be a net speedup. For an odd input n. around 5000 digits. This continues until a single block remains. Such composites are known as strong pseudoprimes to base x.
11.17.21). or on a 64bit limb up to F93 .2 Factorial Factorials are calculated by a combination of removal of twos. to save multiprecision operations.19.19.7. For small n.19. as for instance (13.1.F2k−1 is used.3.16.7. The formulas used are 2 2 F2k+1 = 4Fk − Fk−1 + 2(−1)k 2 2 F2k−1 = Fk + Fk−1 F2k = F2k+1 − F2k−1 At each step.21.19. The procedure can be best illustrated with an example.13. and binary splitting.1. 23! = 1.14.9.1.5. and then evaluating the following product simply from i = 2 to i = k.21.(15. values are generated with a binary powering algorithm. a table of single limb values in __gmp_fib_table is used. For convenience the table starts at F−1 . On a 32bit limb this goes up to F47 .9.15.21.5.7. .1.11. though for mpz_bin_ui this applies only to the divisors. And even below the Karatsuba threshold a big block of work can be more efficient for the basecase algorithm.5 [Exact Division]. and the process repeated until all bits of n are incorporated. since n is an mpz_t and n − k + i in general won’t fit in a limb at all.17.17. Such splitting is more efficient than repeated N×1 multiplies since it forms big multiplies.(13.3.15. 23! = 219 .Fn−1 values efficiently.2.11)2 . allowing Karatsuba and higher algorithms to be used.9.15.20. Notice these formulas require just two squares per bit of n. page 98).(7.17. 23! = 219 .7.7. 16.3.21.5.5. 16.3 Binomial Coefficients Binomial coefficients n are calculated by first arranging k ≤ n/2 using n = k k sary.6. calculating a pair Fn and Fn−1 working from high to low across the bits of n.8.2. The numerators n − k + i and denominators i are first accumulated into as many fit a limb.23 and the resulting terms collected up according to their multiplicity.19.10. or if it’s a 1 then F2k+1 .4 Fibonacci Numbers The Fibonacci functions mpz_fib_ui and mpz_fib2_ui are designed for calculating isolated Fn or Fn .4. This is implemented iteratively using some bit twiddling.F2k is used.13.18. Nearly equal products are more efficient for the current multiply implementation.(3.23 has factors of two removed.15.5)3 .17. Beyond the table.7.11. k is the high b bits of n. Splitting into subsequences of every second term keeps the resulting products more nearly equal in size than would the simpler approach of say taking the first half and second half of the sequence. powering. so the exact division algorithm is used (see Section 16.12.23 is evaluated by splitting into every second term.9. and the same recursively on each half.23). If the next bit of n is 0 then F2k . n k = (n − k + 1) n−k+i i i=2 k n n−k if neces It’s easy to show that each denominator i will divide the product so far.1.22.23) Each sequence such as 13.Chapter 16: Algorithms 107 16.3.
A bigger table would make more small n go fast. but mpz_fib_ui is only interested in Fn .7. 16. Trailing zero bits on n can be handled with a single square each. For GMP the code is kept compact. random numbers are generated simply by concatenating bits produced by the generator. without a carry or borrow into further limbs.5 Lucas Numbers mpz_lucnum2_ui derives a pair of Lucas numbers from a pair of Fibonacci numbers with the following simple formulas. using the defining Fibonacci recurrence Fk+1 = Fk + Fk−1 . it’s just a question of balancing size against desired speed. The state is 624 words of 32bits each. Lk = Fk + 2Fk−1 Lk−1 = 2Fk − Fk−1 mpz_lucnum_ui is only interested in Ln . but this is not done since it usually turns out to be faster for only about 10 or 20 values of n. random numbers in a range 0 ≤ R < N are generated by taking values R of log2 N bits each until one satisfies R < N . which is a Mersenne prime. If they really mattered it’d be better to extend the data table. L2k = L2 − 2(−1)k k And the lowest 1 bit can be handled with one multiply of a pair of Fibonacci numbers. but the attempts are limited in case the generator is somehow degenerate and produces only 1 bits or similar. hence the name of the generator. With a modulus M and parameters A and C. Using a table avoids lots of calculations on small numbers. with the emphasis primarily on a good powering algorithm. page 122). Linear congruential generators are described in many text books. The Mersenne Twister generator is by Matsumoto and Nishimura (see Appendix B [References]. making the algorithm very fast. and some work can be saved.6 Random Numbers For the urandomb functions. which saves some code size. page 122). One of the following two formulas is used. for instance Knuth volume 2 (see Appendix B [References]. See comments with mpz_fib_ui and the internal mpn_fib2_ui for how this is done. It has a nonrepeating period of 219937 −1.1 It’d be possible to handle the first few n above the single limb table with simple additions. F2k = Fk (Fk + 2Fk−1 ) F2k+1 = (2Fk + Fk−1 )(2Fk − Fk−1 ) + 2(−1)k F2k+1 here is the same as above. a . For the urandomm functions. just rearranged to be a multiply. similar to what mpz_fib_ui does. As long as the generator has good randomness properties this will produce welldistributed N bit numbers. and makes small n go fast. the 2(−1)k term both here and above can be applied just to the low limb of the calculation. Randomness properties are also very good and this is the default algorithm used by GMP. For interest.108 GNU MP 5. This will normally require only one or two attempts. In this case the last step of the algorithm can become one multiply instead of two squares. according as n is odd or even. L2k+1 = 5Fk−1 (2Fk + Fk−1 ) − 4(−1)k 16. which is iterated with one XOR and shift for each 32bit word generated. mpz_fib2_ui returns both Fn and Fn−1 . and including a block of code for just those doesn’t seem worthwhile.0.7.
Carry handling and widening multiplies that are important for GMP can’t be easily expressed in C. In GMP only moduli of the form 2N are supported. as can bigger primitives like mpn_addmul_2 or mpn_addmul_4.8. or promises to be faster. and the current implementation is not as well optimized as it could be.h’. Nails code only exists where it’s faster. knowing mpn_add_n won’t have partly overlapping sources and destination means reading can be done far ahead of writing on superscalar processors. mpn_add_n. . written in C or assembly. Generally 32bit and 64bit variants in a family cannot share code and have separate directories. Each ‘mpn’ subdirectory is for an ISA family. Overheads are significant when N is small. At each step the new state is a linear function of the previous. since the Mersenne Twister generator is better in every respect and is therefore recommended for all normal applications. At larger sizes algorithm selection becomes more important. In each directory a ‘nails’ subdirectory may exist. 16. mpn_lshift and mpn_rshift are next most important.8. A NAILS_SUPPORT directive in each file indicates the nails values the code handles. On some CPUs assembly versions of the internal functions mpn_mul_basecase and mpn_sqr_ basecase give significant speedups. and when N is large clearly the multiply at each step will become slow. and loops can be vectorized on a vector processor. They can also potentially make better use of a wide superscalar processor. mpn_sub_n. This is not a big concern. 16. Within a family further subdirectories may exist for CPU variants.Chapter 16: Algorithms 109 integer state S is iterated by the formula S ← AS + C mod M . holding code with nails support for that CPU variant. used when there’s no machinespecific version of a particular file. page 56) are designed to facilitate a variety of implementations.8 Assembly Coding The assembly subroutines in GMP are the most significant source of speed at small to moderate sizes.2 Assembly Basics mpn_addmul_1 and mpn_submul_1 are the most important routines for overall GMP performance. but hand coding low level routines invariably offers a speedup over generic C by a factor of anything from 2 to 10. depending on the carry handling. The restrictions on overlaps between sources and destinations (see Chapter 8 [Lowlevel Functions]. 16. GCC asm blocks help a lot and are provided in ‘longlong. For example. The ‘mpn/generic’ subdirectory contains default code. This generally means raw output is unsuitable for cryptographic applications without further hashing or the like. but of course speedups in low level routines will still speed up everything proportionally. For both generators the current state can be deduced by observing enough output and applying some linear algebra (over GF(2) in the case of the Mersenne Twister). All multiplications and divisions come down to repeated calls to these. There’s no effort put into nails if they’re not going to enhance a given CPU. mainly through avoiding function call overheads. than plain code. hence the name of the generator. mod M .1 Code Organisation The various ‘mpn’ subdirectories contain machinedependent code.
since a prefetch can be initiated once per unrolled loop (or more than once if the loop covers more than one cache line). by starting a load of the next cache line while processing the contents of the previous one. In functions like mpn_addmul_1 and mpn_add_n. but other schemes may use just 1 or 2. but in that case care must be taken not to attempt to read past the end of an operand. On CPUs without writeallocate caches. AMD K6 mpn_addmul_1 however is an example of an unusual set of circumstances where a branch works out better. The aim of course is to have data arriving continuously. since that might produce a segmentation violation. The aim therefore is to maximize memory throughput.8. at peak throughput. ‘mpn/cray/add_n. Some carry propagation schemes require 4 instructions. . so L2 and main memory performance is important for them. Clearly this is only possible if the chip has a lockup free cache or some sort of prefetch instruction. so L1 performance matters most for them. like mpn_divrem_1. memory access times will almost certainly be more than the calculation time. GCC (as of version 2. Prefetching sources combines well with loop unrolling.8.c’.110 GNU MP 5. Some CPUs have limits on the number of fetches or prefetches in progress. adds one set of carry bits in parallel and then only rarely needs to fall through to a loop propagating further carries. unless assembly versions of mpn_mul_basecase and mpn_sqr_basecase exist.1 16. Basic routines like mpn_add_n or mpn_lshift are often used on large operands. carries are the only dependencies between limb operations. On the x86s.2) doesn’t generate particularly good code for the RISC style idioms that are necessary to handle carry bits in C.. 16. Most current chips have both these features. it adds all limbs in parallel. limiting bandwidth. Of course for calculations which are slow anyway. And so unfortunately almost any loop involving carry bits needs to be coded in assembly for best results.. On RISC processors generally an add and compare for overflow is used. When adding a single bit to a limb. there’s only a carry out if that limb was 0xFF. mpn_mul_1 and mpn_addmul_1 are mostly used for multiply and square basecases. writethroughs might be fine. On wide superscalar processors performance may be completely determined by the number of dependent instructions between carryin and carryout for each limb. The distance ahead to prefetch will be determined by memory latency versus throughput.c’ is an example of this. On vector processors good use can be made of the fact that a carry bit only very rarely propagates more than one limb. Often conditional jumps are generated where adc or sbb forms would be better.95. a straightforward CISC style adc is generally best. in which case the remaining uses are mostly for larger operands.FF which on random data will be only 1 in 2 mp bits per limb.3 Carry Propagation The problem that presents most challenges in GMP is propagating carries from one limb to the next. If a special prefetch instruction doesn’t exist then a plain load can be used. meaning at least 4 cycles per limb. For L2 or main memory operands. On processors with carry flags.0. prefetching destinations will ensure individual stores don’t go further down the cache hierarchy.4 Cache Handling GMP aims to perform well both on operands that fit entirely in L1 cache and those which don’t. This sort of thing can be seen in ‘mpn/generic/aors_n.
Float resources can be freed up by doing carry propagation on the integer side. perhaps doing every second limb on the float side (see Section 16. Integer resources can be freed up by having the loop counter in a float register. . and any code achieving that is optimal. Since all values are smaller than 253 . or by pressing the float units into use for some multiplying. If perhaps only one load or one store can be done per cycle then 3 cycles/limb will the top speed for “binary” operations like mpn_add_n.5 Functional Units When choosing an approach for an assembly loop. In some cases an algorithm can be tweaked to accommodate available resources. signed and unsigned are the same. mpn_addmul_1 and mpn_submul_1 on 64bit machines. The multilimb operand U is split in the loop into two 32bit parts. With IEEE 53bit double precision floats. zeroextending to 64bits. the bignum operand is split into 32bit pieces. Loop control will generally require a counter and pointer updates. costing as much as 5 instructions.8. In some cases. or on CISC chips with all addressing done with the loop counter as a scaled index. if one of the lower two 21bit pieces also uses the sign bit. perhaps by allowing just one updating pointer and others expressed as offsets from it. consideration is given to what operations can execute simultaneously and what throughput can thereby be achieved. or even by doing integer to float conversions in integers using bit twiddling.6 Floating Point Floating point arithmetic is used in GMP for multiplications on CPUs with poor integer multipliers. Inside the loop. into 3 or 4 pieces. and mpn_mul_basecase on both 32bit and 64bit machines. Fast conversion of these unsigned 32bit pieces to floating point is highly machinedependent. Memory throughput is always a limit. Here is a diagram showing 16×32 bit products for an mpn_mul_1 or mpn_addmul_1 with a 64bit limb. The single limb operand V is split into four 16bit parts.Chapter 16: Algorithms 111 Some CPUs or systems have hardware that detects sequential memory accesses and initiates suitable cache movements automatically. The final loop control cost can be amortised by processing several limbs in each iteration (see Section 16. page 111).8.8. page 113).6 [Assembly Floating Point]. integer multiplications producing up to 53 bits will give exact results. With some care though six 21×32 → 53 bit products can be used. the invariant single limb is split at the start. but most processors lack unsigned conversions. It’s mostly useful for mpn_mul_1. Converting partial products back to 64bit limbs is usually best done as a signed conversion. This at least ensures loop control isn’t a big fraction the work done. plus any delays a branch introduces. Breaking a 64×64 multiplication into eight 16×32 → 48 bit pieces is convenient. making life easy. CPU addressing modes might reduce pointer updates.9 [Assembly Loop Unrolling].8. then transferring to the floating point unit back via memory is the only option. 16. 16. For the mpn_mul_1 family of functions on a 64bit machine. reading the data into the integer unit.
112 v48 × v32 v16 v00 V Operand GNU MP 5.8 Software Pipelining Software pipelining consists of scheduling instructions around the branch point in a loop.1 u32 u00 u00 × v00 U Operand (one limb) p00 p16 p32 p48 r32 r48 r64 r80 48bit products u00 × v16 u00 × v32 u00 × v48 u32 × v00 u32 × v16 u32 × v32 u32 × v48 p32 and r32 can be summed using floatingpoint addition.8. SIMD multiplications of say four 16×16 bit multiplies only do as much work as one 32×32 from GMP’s point of view. and need some shifts and adds besides. SSE2 is used for Pentium 4 mpn_mul_1. mpn_addmul_1.8. For example a loop might issue a load not for use in the present iteration but the next. four 49bit quantities are transferred to the integer unit. and is used in a special case for 16bit multipliers in the P55 mpn_mul_1. . There’s generally not much support for propagating the sort of carries that arise in GMP. p00 and p16 can be summed with r64 and r80 from the previous iteration. and likewise p48 and r48. On the x86 chips. 64 bits 64 bits p00 + r64 p16 + r80 p32 + r32 p48 + r48 i00 i16 i32 i48 The challenge then is to sum these efficiently and add in a carry limb. MMX has so far found a use in mpn_rshift and mpn_lshift.7 SIMD Instructions The singleinstruction multipledata support in current microprocessors is aimed at signal processing algorithms where each data point can be treated more or less independently. This is like juggling. thereby allowing extra cycles for the data to arrive from memory. But of course if say the SIMD form is fully pipelined and uses less instruction decoding then it may still be worthwhile.0. 16. A pipeline with several stages will have a data value in progress at each stage and each loop iteration moves them along one stage. generating a low 64bit result limb and a high 33bit carry limb (i48 extends 33 bits into the high half). and only where a CPU has multiple functional units so that other work can be done in the meantime. For each loop then. Naturally this is wanted only when doing things like loads or multiplies that take several cycles to complete. aligned as follows. 16. and mpn_submul_1.
for example an 8limb unrolling would have separate code for 0 remaining. The limbs not a multiple of the unrolling can be handled in various ways. If an instruction can use either of two units. so one register has a result ready to use while another (or multiple others) are still in progress. Or by subtracting N at the start. for example • A simple loop at the end (or the start) to process the excess.8. providing separate code for each possible excess. The total instruction count might be the limiting factor.10 Writing Guide This is a guide to writing software pipelined loops for processing limb vectors in assembly. (see Section 16. but it can also allow better register usage. On a 3operand CPU try to write each new value to a new register. . the termination condition becomes when the counter C is less than 0 (and the count of remaining limbs is C + N ). which is attractive. but may be the best way to optimize all cases in combination with a deep pipelined loop. • A computed jump into the middle of the loop. Any amount of unrolling can be handled with a loop counter that’s decremented by N each time. Suppose the loop time is N . Judicious use of m4 macros can help avoid lots of duplication in the source code. for example alternately using one register combination and then another.9 [Assembly Loop Unrolling]. This might take a lot of code. with the final loop branch at the end of the last. like U0 or U1 then make a category “U0/U1”. Code it without unrolling or scheduling. It might be possible to tweak the instructions to help the limiting factor. Figure out from those counts the best possible loop time.Chapter 16: Algorithms 113 If the latency of some instruction is greater than the loop time then it will be necessary to unroll. This should make times smoothly increase with size. then make N issue buckets. 1 remaining. This is convenient if also making a computed jump into the middle of a large loop. to make sure it works. and finally 1 more or not. or perhaps a particular functional unit. stopping when the remaining count is less than the further N the loop will process. 16. test for 4 more limbs to process. Then note for each instruction the functional unit and/or issue port requirements. this will greatly simplify later steps. page 113). then a further 2 more or not.9 Loop Unrolling Loop unrolling consists of replicating code so that several limbs are processed in each loop. Care will be wanted that it isn’t too much slower than the unrolled part. Now fill the buckets with dummy instructions using the functional units desired. At a minimum this reduces loop overheads by a corresponding factor. 16. Run this to make sure the intended speed is reached. This will probably take more code space than a simple loop. • A set of binary tests. Alternately for a power of 2 unroll the loop count and remainder can be established with a shift and mask. • A switch statement. but setups for the jump and adjustments for pointers can be tricky and could become quite difficult in combination with deep pipelining. etc. up to 7 remaining. The goal will be to find a perfect schedule where instruction latencies are completely hidden. for example after an 8limb unrolling.8.8. thus making the first iteration handle the excess. and count all instructions. Count the total using each unit (or combined unit). First determine the algorithm and which instructions are needed.
After a few you will need to wrap around from the last bucket back to the top of the loop. The loop will overlap two or more of the original loop iterations. and completed one or several iterations later. The first will typically be a load instruction. Changing registers at this time is very error prone. Run the loop again. to check it still runs at target speed. frequently measuring the loop. The loop will have a minimum number of limbs loaded and processed. The final step is to create feedin and winddown code for the loop. and the computation of one vector element result will be started in one iteration of the new loop.1 Now replace the dummy instructions with the real instructions from the slow but correct loop you started with. If not then take care not to clobber something already in use.114 GNU MP 5. Then the instruction using that value is placed in a bucket an appropriate distance down. If you used the newregister for newvalue strategy above then there will be no register conflicts. Keep placing instructions. .0. and at the end replicate and delete those whose results are unwanted (including any further loads). A good way to do this is to make a copy (or copies) of the loop at the start and delete those instructions which don’t have valid antecedents. so the feedin code must test if the request size is smaller and skip either to a suitable part of the winddown or to special code for small sizes.
This is done to make the fields just 32 bits on some 64 bits systems. The canonical form adopted is denominator positive (and nonzero). not four. _mp_alloc is the number of limbs currently allocated at _mp_d. Sometimes this isn’t pretty. and reallocates if not. but sign and magnitude are best for other routines. in which case the _mp_d data is unused. But sign and magnitude is always used internally.Chapter 17: Internals 115 17 Internals This chapter is provided only for informational purposes and the various internals described here may change in future GMP releases. no common factors between numerator and denominator. or the negative of that when representing a negative integer. Care is taken to ensure that these are big enough that no reallocation is necessary (since it would have unpredictable consequences). Zero is represented by _mp_size set to zero. although mp_size_t is usually a long. Whenever _mp_size is nonzero. _mp_size _mp_d The number of limbs.2 Rational Internals mpq_t variables represent rationals using an mpz_t numerator and denominator (see Section 17. so _mp_d[0] is the least significant limb and _mp_ d[ABS(_mp_size)1] is the most significant. A pointer to an array of limbs which is the magnitude. Some internal temporary variables are setup with MPZ_TMP_INIT and these have _mp_d space obtained from TMP_ALLOC rather than the memory allocation functions. the most significant limb is nonzero. The various bitwise logical functions like mpz_and behave as if negative values were twos complement. page 115). This general approach to common factors is badly suboptimal in the presence of simple factorizations or little prospect for cancellation. Currently there’s always at least one limb allocated. The fields are as follows. _mp_size and _mp_alloc are int. and mpz_get_ui can fetch _mp_d[0] unconditionally (though its value is then only wanted if _mp_size is nonzero). so for instance mpz_set_ui never needs to reallocate. . These are stored “little endian” as per the mpn functions. When an mpz routine is about to (or might be about to) increase _mp_size. 17. in space dynamically allocated and reallocated.1 [Integer Internals]. and zero uniquely represented as 0/1. but GMP has no way to know when this will occur. Knowing the numerator and denominator have no common factors can be used for example in mpq_mul to make only two cross GCDs necessary. it checks _mp_alloc to see whether there’s enough space. and necessary adjustments are made during the calculations. It’s believed that casting out common factors at each stage of a calculation is best in general. and naturally _mp_ alloc >= ABS(_mp_size). thereby saving a few bytes of data space but still providing plenty of range. MPZ_REALLOC is generally used for this. Applications expecting to be compatible with future releases should use only the documented interfaces described in previous chapters. A GCD is an O(N 2 ) operation so it’s better to do a few small ones immediately than to delay and have to do a big one later.1 Integer Internals mpz_t variables represent integers using sign and magnitude. _mp_alloc 17.
116 GNU MP 5. in limbs. In any calculation the aim is to produce _mp_prec limbs of result (the most significant being nonzero). Low Zeros The least significant limbs _mp_d[0] etc can be zero. and in that case the _mp_d data is unused. The following various points should be noted. This is done to make some fields just 32 bits on some 64 bits systems. Routines likely to produce low zeros check and avoid them to . (In the future _mp_exp might be undefined when representing zero. _mp_size The number of limbs currently in use. The _mp_exp field is usually long. it doesn’t have to fall within the limbs as the diagram shows. There are no reallocations during a calculation. Limbs other than those included in the {_mp_d.1 As per Section 3. the extra limb being for convenience (see below).3 Float Internals Efficient calculation is the primary aim of GMP floats and the use of whole limbs and simple rounding facilitates this. mpf_t floats have a variable precision mantissa and a single machine word signed exponent. The mantissa is represented using sign and magnitude. or the negative of that when representing a negative value. These are stored “little endian” as per the mpn functions.0. most significant limb mp exp least significant limb → mp d · ← radix point ← The fields are as follows. The exponent. The mpq_t framework might still suit. or of course mpz_t variables can be used directly. A pointer to the array of limbs which is the absolute value of the mantissa. it can be a long way above or a long way below. determining the location of the implied radix point. though such low zeros can always be ignored. Positive values mean a radix point offset towards the lower limbs and hence a value ≥ 1. _mp_prec+1 limbs are allocated to _mp_d. Zero is represented by _mp_size and _mp_exp both set to zero. thereby saving a few bytes of data space but still providing plenty of precision and a very large range. Naturally the exponent can be any value. only in a change of precision with mpf_set_prec.) The precision of the mantissa.11 [Efficiency]. The most significant limb is always nonzero. that’s left to applications. with mpq_numref and mpq_denref for direct access to the numerator and denominator. 17. as for example in the diagram above. so _mp_d[0] is the least significant limb and _mp_d[ABS(_mp_size)1] the most significant. although the mp_size_t type is usually a long. but there are no other restrictions on its value. Zero means the radix point is just above the most significant limb. page 21. in particular the highest 1 bit can be anywhere within the limb. Negative exponents mean a radix point further above the highest limb. in limbs. mp size → _mp_prec _mp_d _mp_exp The _mp_size and _mp_prec fields are int._mp_size} data are treated as zero.
giving an _mp_prec of 9. or one in bits which requires them almost everywhere else. This is no more than a pointer adjustment. Firstly a value is allowed to use all of the _mp_prec+1 limbs available at _mp_d. Since the exponent is in limbs. For example with a 32bit limb. Rounding All rounding is done on limb boundaries. since there’s no need to examine extra limbs and increment or decrement. Note that the extra limb added here for the high only being nonzero is in addition to the extra limb allocated to _mp_d. Strictly speaking this is unnecessary since only _mp_prec limbs are needed for the application’s requested precision. __GMPF_PREC_TO_BITS does the reverse conversion. _mp_size can also be greater than _mp_prec. Whenever _mp_prec+1 limbs are held in a variable. and secondly when mpf_set_prec_raw lowers _mp_prec it leaves _mp_size unchanged and so the size can be arbitrarily bigger than _mp_prec. The value in bits is rounded up to a whole limb then an extra limb is added since the most significant limb of _mp_d is only nonzero and therefore might contain only one bit. and must be checked anyway since the destination precision can be different from the sources. _mp_d then gets 10 limbs allocated. but the choice is between an exponent in limbs which requires shifts there. the low limb is not needed for the intended precision. The net effect of reading back with mpf_get_ prec is simply the precision rounded up to a multiple of mp_bits_per_limb. If there’s no carry then that’s the result. The use of simple “trunc” rounding towards zero is efficient. Mantissa Size Range The _mp_size count of limbs in use can be less than _mp_prec if the value can be represented in less. mpf_add for instance will do an mpn_add of _mp_prec limbs. and removes the extra limb from _mp_prec before converting to bits. Application Precisions __GMPF_BITS_TO_PREC converts an application requested precision to an _mp_prec. This ensures that a variable which has _mp_size equal to _mp_prec+1 will get its full exact value copied. When differing exponents are encountered all that’s needed is to adjust pointers to line up the relevant limbs. an application request for 250 bits will be rounded up to 8 limbs. . but for most routines they’re quite unlikely and aren’t checked. Of course mpf_mul_2exp and mpf_div_2exp will require bit shifts. Subsequent routines reading the value will simply take the high limbs they need. Calculating _mp_prec limbs with the high nonzero will ensure the application requested minimum precision is obtained. there are no bit shifts in basic operations like mpf_ add and mpf_mul. This means low precision values or small integers stored in a high precision mpf_t can still be operated on efficiently. but it’s considered that an mpf_ set from one variable into another of the same precision ought to produce an exact copy. and this will be _mp_prec if their target has that same precision. then an extra added for the high being only nonzero. but if there is a carry then it’s stored in the extra limb of space and _mp_size becomes _mp_prec+1. only the _mp_prec high limbs. But zeroing it out or moving the rest down is unnecessary. Bit Shifts Use of _mp_prec+1 Limbs The extra limb on _mp_d (_mp_prec+1 rather than just _mp_prec) helps when an mpf routine might get a carry from its operation. Copy functions like mpf_set will retain a full _mp_prec+1 limbs if available.Chapter 17: Internals 117 save time in subsequent calculations.
And an “additive expression” object. 17. For mpf_class the scheme also ensures the precision of the final destination is used for any temporaries within a statement like f=w*x+y*z. irrespective of limb size. mpf_class. . 17. Unfortunately it also means that the limb data must be reversed when reading or writing. refer to the source code. These are important features which a naive implementation cannot provide.. written most significant byte first. so the output is the same on all systems. it makes the data easy to read in a hex dump of a file. The most significant data byte is always nonzero. but for the purposes of mpf_t it’s considered simply to be 64 bits.. g). for compatibility. so neither a big endian nor little endian system can just read and write _mp_d. say. In fact even mpf_class etc are typedef specializations of __gmp_expr.118 GNU MP 5.1 Reading back with mpf_get_prec will take _mp_prec subtract 1 limb and multiply by 32. __gmp_binary_plus> > operator+(const mpf_class &f. The true scheme is complicated by the fact that expressions have different return types. mpf_t h) { mpf_add(f. } The seemingly redundant __gmp_expr<__gmp_binary_expr<. being the number of subsequent data bytes. struct __gmp_binary_plus { static void eval(mpf_t f. __gmp_binary_plus> >(f. size data bytes The size is 4 bytes written most significant byte first. h). The data bytes are the absolute value of the integer.5 C++ Interface Internals A system of expression templates is used to ensure something like a=b+c turns into a simple call to mpz_add etc. For detailed information. we first define a “function object” evaluating it.>> is used to encapsulate any possible kind of expression into a single template type. To perform an operation. The use of “big endian” for both the size and data fields is deliberate. leading zero bytes were written to pad the data bytes to a multiple of the limb size. } }. mpf_class. g. mpz_inp_raw will still accept this. In GMP 1. a nice multiple of the limb size. addition. giving 256 bits. or the twos complement negative of that when a negative integer is represented. 3 limbs of 32bits each will be holding at least 65 bits. say. A simplified description of the scheme follows. Strictly speaking. __gmp_expr<__gmp_binary_expr<mpf_class. the fact the high limb has at least one bit means that a float with.4 Raw Output Internals mpz_out_raw uses the following format. const mpf_class &g) { return __gmp_expr <__gmp_binary_expr<mpf_class.0. mpf_t g.
__gmp_expr<U>.get_mpf_t().val2. template <class T> mpf_class & mpf_class::operator=(const __gmp_expr<T> &expr) { expr.get_mpf_t()). __gmp_expr<U>.val1 and expr. return *this. class U> __gmp_expr <__gmp_binary_expr<__gmp_expr<T>. precision). when the required precision (that of f) is known. temp1. __gmp_binary_plus> > operator+(const __gmp_expr<T> &expr1. __gmp_expr<U>. Op> >::eval (mpf_t f. class U. } The expression is thus recursively evaluated to any level of complexity and all subexpressions are evaluated to the precision of f.eval(this>get_mpf_t().val1. mpf_class. this>precision()). mp_bitcnt_t precision) { Op::eval(f. expr.val2. temp2. } template <class Op> void __gmp_expr<__gmp_binary_expr<mpf_class. expr2). } And the corresponding specializations of __gmp_expr::eval: template <class T.get_mpf_t().get_mpf_t()). class Op> void __gmp_expr <__gmp_binary_expr<__gmp_expr<T>. Compound expressions are handled by defining operators taking subexpressions as their arguments. const __gmp_expr<U> &expr2) { return __gmp_expr <__gmp_binary_expr<__gmp_expr<T>. like this: template <class T. __gmp_binary_plus> > (expr1. expr. precision).val2 are references to the expression’s operands (here expr is the __gmp_binary_expr stored within the __gmp_expr). Furthermore the target mpf_t is now available. } where expr. thus we can call mpf_add directly with f as the output argument. . Op> >::eval (mpf_t f. the expression is actually evaluated only at the time of assignment. This way.val1.Chapter 17: Internals 119 Next we define assignment of __gmp_expr to mpf_class. temp2(expr. Op::eval(f. mp_bitcnt_t precision) { // declare two temporaries mpf_class temp1(expr.
120
GNU MP 5.0.1
Appendix A Contributors
Torbj¨rn Granlund wrote the original GMP library and is still the main developer. Code not o explicitly attributed to others, was contributed by Torbj¨rn. Several other individuals and o organizations have contributed GMP. Here is a list in chronological order on first contribution: Gunnar Sj¨din and Hans Riesel helped with mathematical problems in early versions of the o library. Richard Stallman helped with the interface design and revised the first version of this manual. Brian Beuning and Doug Lea helped with testing of early versions of the library and made creative suggestions. John Amanatides of York University in Canada contributed the function mpz_probab_prime_p. Paul Zimmermann wrote the REDCbased mpz powm code, the Sch¨nhageStrassen FFT mulo tiply code, and the Karatsuba square root code. He also improved the Toom3 code for GMP 4.2. Paul sparked the development of GMP 2, with his comparisons between bignum packages. The ECMNET project Paul is organizing was a driving force behind many of the optimizations in GMP 3. Paul also wrote the new GMP 4.3 nth root code (with Torbj¨rn). o Ken Weber (Kent State University, Universidade Federal do Rio Grande do Sul) contributed now defunct versions of mpz_gcd, mpz_divexact, mpn_gcd, and mpn_bdivmod, partially supported by CNPq (Brazil) grant 3013141942. Per Bothner of Cygnus Support helped to set up GMP to use Cygnus’ configure. He has also made valuable suggestions and tested numerous intermediary releases. Joachim Hollman was involved in the design of the mpf interface, and in the mpz design revisions for version 2. Bennet Yee contributed the initial versions of mpz_jacobi and mpz_legendre. Andreas Schwab contributed the files ‘mpn/m68k/lshift.S’ and ‘mpn/m68k/rshift.S’ (now in ‘.asm’ form). Robert Harley of Inria, France and David Seal of ARM, England, suggested clever improvements for population count. Robert also wrote highly optimized Karatsuba and 3way Toom multiplication functions for GMP 3, and contributed the ARM assembly code. Torsten Ekedahl of the Mathematical department of Stockholm University provided significant inspiration during several phases of the GMP development. His mathematical expertise helped improve several algorithms. Linus Nordberg wrote the new configure system based on autoconf and implemented the new random functions. Kevin Ryde worked on a large number of things: optimized x86 code, m4 asm macros, parameter tuning, speed measuring, the configure system, function inlining, divisibility tests, bit scanning, Jacobi symbols, Fibonacci and Lucas number functions, printf and scanf functions, perl interface, demo expression parser, the algorithms chapter in the manual, ‘gmpasmmode.el’, and various miscellaneous improvements elsewhere. Kent Boortz made the Mac OS 9 port. Steve Root helped write the optimized alpha 21264 assembly code. Gerardo Ballabio wrote the ‘gmpxx.h’ C++ class interface and the C++ istream input routines.
Appendix A: Contributors
121
Jason Moxham rewrote mpz_fac_ui. Pedro Gimeno implemented the Mersenne Twister and made other random number improvements. Niels M¨ller wrote the subquadratic GCD and extended GCD code, the quadratic Hensel divio sion code, and (with Torbj¨rn) the new divide and conquer division code for GMP 4.3. Niels also o helped implement the new Toom multiply code for GMP 4.3 and implemented helper functions to simplify Toom evaluations for GMP 5.0. He wrote the original version of mpn mulmod bnm1. Alberto Zanoni and Marco Bodrato suggested the unbalanced multiply strategy, and found the optimal strategies for evaluation and interpolation in Toom multiplication. Marco Bodrato helped implement the new Toom multiply code for GMP 4.3 and implemented most of the new Toom multiply and squaring code for 5.0. He is the main author of the current mpn mulmod bnm1 and mpn mullo n. Marco also wrote the functions mpn invert and mpn invertappr. David Harvey suggested the internal function mpn_bdiv_dbm1, implementing division relevant to Toom multiplication. He also worked on fast assembly sequences, in particular on a fast AMD64 mpn_mul_basecase. Martin Boij wrote mpn_perfect_power_p. (This list is chronological, not ordered after significance. If you have contributed to GMP but are not listed above, please tell gmpdevel@gmplib.org about the omission!) The development of floating point functions of GNU MP 2, were supported in part by the ESPRITBRA (Basic Research Activities) 6846 project POSSO (POlynomial System SOlving). The development of GMP 2, 3, and 4 was supported in part by the IDA Center for Computing Sciences. Thanks go to Hans Thorsen for donating an SGI system for the GMP test system environment.
122
GNU MP 5.0.1
Appendix B References
B.1 Books
• Jonathan M. Borwein and Peter B. Borwein, “Pi and the AGM: A Study in Analytic Number Theory and Computational Complexity”, Wiley, 1998. • Richard Crandall and Carl Pomerance, “Prime Numbers: A Computational Perspective”, 2nd edition, SpringerVerlag, 2005. http://math.dartmouth.edu/~carlp/ • Henri Cohen, “A Course in Computational Algebraic Number Theory”, Graduate Texts in Mathematics number 138, SpringerVerlag, 1993. http://www.math.ubordeaux.fr/~cohen/ • Donald E. Knuth, “The Art of Computer Programming”, volume 2, “Seminumerical Algorithms”, 3rd edition, AddisonWesley, 1998. http://wwwcsfaculty.stanford.edu/~knuth/taocp.html • John D. Lipson, “Elements of Algebra and Algebraic Computing”, The Benjamin Cummings Publishing Company Inc, 1981. • Alfred J. Menezes, Paul C. van Oorschot and Scott A. Vanstone, “Handbook of Applied Cryptography”, http://www.cacr.math.uwaterloo.ca/hac/ • Richard M. Stallman and the GCC Developer Community, “Using the GNU Compiler Collection”, Free Software Foundation, 2008, available online http://gcc.gnu.org/onlinedocs/, and in the GCC package ftp://ftp.gnu.org/gnu/gcc/
B.2 Papers
• Yves Bertot, Nicolas Magaud and Paul Zimmermann, “A Proof of GMP Square Root”, Journal of Automated Reasoning, volume 29, 2002, pp. 225252. Also available online as INRIA Research Report 4475, June 2001, http://www.inria.fr/rrrt/rr4475.html • Christoph Burnikel and Joachim Ziegler, “Fast Recursive Division”, MaxPlanckInstitut fuer Informatik Research Report MPII981022, http://data.mpisb.mpg.de/internet/reports.nsf/NumberView/19981022 • Torbj¨rn Granlund and Peter L. Montgomery, “Division by Invariant Integers using Multio plication”, in Proceedings of the SIGPLAN PLDI’94 Conference, June 1994. Also available ftp://ftp.cwi.nl/pub/pmontgom/divcnst.psa4.gz (and .psl.gz). • Niels M¨ller and Torbj¨rn Granlund, “Improved division by invariant integers”, to appear. o o • Torbj¨rn Granlund and Niels M¨ller, “Division of integers large and small”, to appear. o o • Tudor Jebelean, “An algorithm for exact division”, Journal of Symbolic Computation, volume 15, 1993, pp. 169180. Research report version available ftp://ftp.risc.unilinz.ac.at/pub/techreports/1992/9235.ps.gz • Tudor Jebelean, “Exact Division with Karatsuba Complexity  Extended Abstract”, RISCLinz technical report 9631, ftp://ftp.risc.unilinz.ac.at/pub/techreports/1996/9631.ps.gz • Tudor Jebelean, “Practical Integer Division with Karatsuba Complexity”, ISSAC 97, pp. 339341. Technical report available ftp://ftp.risc.unilinz.ac.at/pub/techreports/1996/9629.ps.gz • Tudor Jebelean, “A Generalization of the Binary GCD Algorithm”, ISSAC 93, pp. 111116. Technical report version available ftp://ftp.risc.unilinz.ac.at/pub/techreports/1993/9301.ps.gz
loria. • Kenneth Weber. 145157. pp. ARITH11: IEEE Symposium on Computer Arithmetic. Journal of Symbolic Computation. in Mathematics of Computation. January 2008. Proceedings of the 13th Annual IEEE Symposium on Switching and Automata Theory. April 1985. 111122. Journal of Symbolic Computation.sci.risc. • Paul Zimmermann. “Karatsuba Square Root”.unilinz. number 8.ac.html • Paul Zimmermann. INRIA Research Report 3805. Technical report version also available ftp://ftp. “On Sch¨nhage’s algorithm and subquadratic integer GCD computation”. 330. volume 44.ps. number 1.ps. 1993.ps. Reprinted as “More on Multiplying and Squaring Large Integers”. Computo ing 7.math. volume 21. 1971. Moenck and A.at/pub/techreports/1994/9450. pp. 260 to 271. October 1972. in o o Mathematics of Computation. 281292.jp/~mmat/MT/ARTICLES/mt. • Niels M¨ller.gz • Makoto Matsumoto and Takuji Nishimura.unilinz.gz • Dan Zuras.at/pub/techreports/1992/9269.pdf) • R.Appendix B: References 123 • Tudor Jebelean. volume 21. pp. pp.gz • Werner Krandick and Tudor Jebelean.fr/~zimmerma/papers/proofdivsqrt. “Mersenne Twister: A 623dimensionally equidistributed uniform pseudorandom number generator”. number 170. pp.fr/rrrt/rr3805. IEEE Transactions on Computers. ACM Transactions on Modelling and Computer Simulation. volume 43. volume 77. Montgomery. pp. 9096. http://www. “Fast Modular Transforms via Division”.ac. “On Squaring and Multiplying Large Integers”. “The accelerated integer GCD algorithm”. Journal of Computer and System Sciences. pp. “Modular Multiplication Without Trial Division”. 1996. November 1999. pp. August 1994. volume 19. pp. “Bidirectional Exact Integer Division”. 589607. 1995. Available online http://www. “Schnelle Multiplikation grosser Zahlen”. Reprinted as “Fast Modular Transforms”. volume 8. number 3.risc.hiroshimau. March 1995. • Arnold Sch¨nhage and Volker Strassen. volume 8. Borodin. 899908. pp.ps. January 1998. Early technical report version also available ftp://ftp. “A DoubleDigit LehmerEuclid Algorithm for Finding the GCD of Long Integers”.gz (or .ac. ACM Transactions on Mathematical Software. . • Peter L. 441455.inria. http://www. June 1974. 366386. “A Proof of GMP Fast Division and Square Root Implementations”.
but changing it is not allowed. in the notice that says that the Document is released under this License. as FrontCover Texts or BackCover Texts. or with modifications and/or translated into another language. . either copied verbatim.124 GNU MP 5. (Thus. to use that work under the conditions stated herein. It complements the GNU General Public License. and is addressed as “you”. But this License is not limited to software manuals. unlimited in duration. A FrontCover Text may be at most 5 words. We have designed this License in order to use it for manuals for free software. either commercially or noncommercially. in any medium. PREAMBLE The purpose of this License is to make a manual. You accept the license if you copy. Secondarily. ethical or political position regarding them. APPLICABILITY AND DEFINITIONS This License applies to any manual or other work. while not being considered responsible for modifications made by others. this License preserves for the author and publisher a way to get credit for their work. Such a notice grants a worldwide. and a BackCover Text may be at most 25 words. which is a copyleft license designed for free software. or other functional and useful document free in the sense of freedom: to assure everyone the effective freedom to copy and redistribute it. Any member of the public is a licensee. philosophical. The “Cover Texts” are certain short passages of text that are listed. 2002. commercial. If a section does not fit the above definition of Secondary then it is not allowed to be designated as Invariant. with or without modifying it. modify or distribute the work in a way requiring permission under copyright law. http://fsf. which means that derivative works of the document must themselves be free in the same sense. royaltyfree license. We recommend this License principally for works whose purpose is instruction or reference. refers to any such manual or work. This License is a kind of “copyleft”. A “Secondary Section” is a named appendix or a frontmatter section of the Document that deals exclusively with the relationship of the publishers or authors of the Document to the Document’s overall subject (or to related matters) and contains nothing that could fall directly within that overall subject. The Document may contain zero Invariant Sections. or of legal. that contains a notice placed by the copyright holder saying it can be distributed under the terms of this License. a Secondary Section may not explain any mathematics. The “Document”.) The relationship could be a matter of historical connection with the subject or with related matters. 3 November 2008 Copyright c 2000. 2008 Free Software Foundation. below.org/ Everyone is permitted to copy and distribute verbatim copies of this license document. regardless of subject matter or whether it is published as a printed book. in the notice that says that the Document is released under this License. 2001. If the Document does not identify any Invariant Sections then there are none. Inc. The “Invariant Sections” are certain Secondary Sections whose titles are designated. 0. because free software needs free documentation: a free program should come with manuals providing the same freedoms that the software does.0. it can be used for any textual work. textbook.3. A “Modified Version” of the Document means any work containing the Document or a portion of it. 2007.1 Appendix C GNU Free Documentation License Version 1. 1. as being those of Invariant Sections. if the Document is in part a textbook of mathematics.
and the Document’s license notice requires Cover Texts. SGML or XML for which the DTD and/or processing tools are not generally available. but only as regards disclaiming warranties: any other implication that these Warranty Disclaimers may have is void and has no effect on the meaning of this License. These Warranty Disclaimers are considered to be included by reference in this License. provided that this License. VERBATIM COPYING You may copy and distribute the Document in any medium. You may not use technical measures to obstruct or control the reading or further copying of the copies you make or distribute. The “Title Page” means. and BackCover Texts on the back cover. For works in formats which do not have any title page as such. 3. either commercially or noncommercially. If you distribute a large enough number of copies you must also follow the conditions in section 3. The Document may include Warranty Disclaimers next to the notice which states that this License applies to the Document. Examples of transparent image formats include PNG. A copy made in an otherwise Transparent file format whose markup. under the same conditions stated above. “Title Page” means the text near the most prominent appearance of the work’s title. numbering more than 100. LaTEX input format. You may also lend copies. SGML or XML using a publicly available DTD. “Dedications”. A copy that is not “Transparent” is called “Opaque”. Opaque formats include proprietary formats that can be read and edited only by proprietary word processors. you must enclose the copies in covers that carry. for a printed book. has been arranged to thwart or discourage subsequent modification by readers is not Transparent. such as “Acknowledgements”. PostScript or PDF designed for human modification. all these Cover Texts: FrontCover Texts on the front cover.Appendix C: GNU Free Documentation License 125 A “Transparent” copy of the Document means a machinereadable copy. “Endorsements”. and that you add no other conditions whatsoever to those of this License. or absence of markup. (Here XYZ stands for a specific section name mentioned below. clearly and legibly. or “History”. the copyright notices. the material this License requires to appear in the title page. you may accept compensation in exchange for copies. 2. and you may publicly display copies. and the license notice saying this License applies to the Document are reproduced in all copies. PostScript or PDF produced by some word processors for output purposes only. and standardconforming simple HTML.) To “Preserve the Title” of such a section when you modify the Document means that it remains a section “Entitled XYZ” according to this definition. Both . However. COPYING IN QUANTITY If you publish printed copies (or copies in media that commonly have printed covers) of the Document. represented in a format whose specification is available to the general public. legibly. A section “Entitled XYZ” means a named subunit of the Document whose title either is precisely XYZ or contains XYZ in parentheses following text that translates XYZ in another language. and that is suitable for input to text formatters or for automatic translation to a variety of formats suitable for input to text formatters. the title page itself. Texinfo input format. that is suitable for revising the document straightforwardly with generic text editors or (for images composed of pixels) generic paint programs or (for drawings) some widely available drawing editor. and the machinegenerated HTML. Examples of suitable formats for Transparent copies include plain ascii without markup. An image format is not Transparent if used for any substantial amount of text. plus such following pages as are needed to hold. The “publisher” means any person or entity that distributes copies of the Document to the public. preceding the beginning of the body of the text. XCF and JPG.
Preserve in that license notice the full lists of Invariant Sections and required Cover Texts given in the Document’s license notice. provided that you release the Modified Version under precisely this License. and from those of previous versions (which should. when you begin distribution of Opaque copies in quantity. immediately after the copyright notices. you must do these things in the Modified Version: A. you must either include a machinereadable Transparent copy along with each Opaque copy. 4. can be treated as verbatim copying in other respects. You may use the same title as a previous version if the original publisher of that version gives permission. Use in the Title Page (and on the covers. you should put the first ones listed (as many as fit reasonably) on the actual cover. E. with the Modified Version filling the role of the Document. free of added material. and add to it an item stating at least the title. Preserve all the copyright notices of the Document. You may add other material on the covers in addition. year. F. H. C. you must take reasonably prudent steps. new authors. that you contact the authors of the Document well before redistributing any large number of copies. to give them a chance to provide you with an updated version of the Document. List on the Title Page. If the required texts for either cover are too voluminous to fit legibly. as authors. thus licensing distribution and modification of the Modified Version to whoever possesses a copy of it. If you use the latter option. to ensure that this Transparent copy will remain thus accessible at the stated location until at least one year after the last time you distribute an Opaque copy (directly or through your agents or retailers) of that edition to the public. be listed in the History section of the Document). State on the Title page the name of the publisher of the Modified Version. as long as they preserve the title of the Document and satisfy these conditions. It is requested.1 covers must also clearly and legibly identify you as the publisher of these copies. authors. Copying with changes limited to the covers. unless they release you from this requirement. and publisher of the Document as given on its . B. MODIFICATIONS You may copy and distribute a Modified Version of the Document under the conditions of sections 2 and 3 above. Include. The front cover must present the full title with all words of the title equally prominent and visible. if there were any. Add an appropriate copyright notice for your modifications adjacent to the other copyright notices. create one stating the title. one or more persons or entities responsible for authorship of the modifications in the Modified Version. if it has fewer than five). and publisher of the Modified Version as given on the Title Page. year. as the publisher. and continue the rest onto adjacent pages. G.0. In addition. If you publish or distribute Opaque copies of the Document numbering more than 100. or state in or with each Opaque copy a computernetwork location from which the general networkusing public has access to download using publicstandard network protocols a complete Transparent copy of the Document. if any) a title distinct from that of the Document. Preserve its Title. I. but not required. in the form shown in the Addendum below. a license notice giving the public permission to use the Modified Version under the terms of this License. together with at least five of the principal authors of the Document (all of its principal authors.126 GNU MP 5. Preserve the section Entitled “History”. If there is no section Entitled “History” in the Document. D. Include an unaltered copy of this License.
and list them all as Invariant Sections of your combined work in its license notice. M. provided it contains nothing but endorsements of your Modified Version by various parties—for example. you may not add another. Preserve any Warranty Disclaimers. Section numbers or the equivalent are not considered part of the section titles. to the end of the list of Cover Texts in the Modified Version. Only one passage of FrontCover Text and one of BackCover Text may be added by (or through arrangements made by) any one entity. Preserve the network location. You may add a passage of up to five words as a FrontCover Text. If the Modified Version includes new frontmatter sections or appendices that qualify as Secondary Sections and contain no material copied from the Document. in parentheses. you may at your option designate some or all of these sections as invariant. If the Document already includes a cover text for the same cover. unmodified. make the title of each such section unique by adding at the end of it. previously added by you or by arrangement made by the same entity you are acting on behalf of. then add an item describing the Modified Version as stated in the previous sentence. or if the original publisher of the version it refers to gives permission. on explicit permission from the previous publisher that added the old one. K. 5. you must combine any sections Entitled “History” in the various original documents. likewise combine any sections Entitled . the name of the original author or publisher of that section if known. and that you preserve all their Warranty Disclaimers. but you may replace the old one. L. under the terms defined in section 4 above for modified versions. provided that you include in the combination all of the Invariant Sections of all of the original documents.Appendix C: GNU Free Documentation License 127 J. and preserve in the section all the substance and tone of each of the contributor acknowledgements and/or dedications given therein. or else a unique number. forming one section Entitled “History”. If there are multiple Invariant Sections with the same name but different contents. given in the Document for public access to a Transparent copy of the Document. and multiple identical Invariant Sections may be replaced with a single copy. You may add a section Entitled “Endorsements”. To do this. The author(s) and publisher(s) of the Document do not by this License give permission to use their names for publicity for or to assert or imply endorsement of any Modified Version. N. These may be placed in the “History” section. unaltered in their text and in their titles. Preserve the Title of the section. if any. Such a section may not be included in the Modified Version. COMBINING DOCUMENTS You may combine the Document with other documents released under this License. The combined work need only contain one copy of this License. These titles must be distinct from any other section titles. In the combination. Preserve all the Invariant Sections of the Document. Do not retitle any existing section to be Entitled “Endorsements” or to conflict in title with any Invariant Section. and likewise the network locations given in the Document for previous versions it was based on. and a passage of up to 25 words as a BackCover Text. O. statements of peer review or that the text has been approved by an organization as the authoritative definition of a standard. For any section Entitled “Acknowledgements” or “Dedications”. Title Page. add their titles to the list of Invariant Sections in the Modified Version’s license notice. Make the same adjustment to the section titles in the list of Invariant Sections in the license notice of the combined work. Delete any section Entitled “Endorsements”. You may omit a network location for a work that was published at least four years before the Document itself.
However.0.128 GNU MP 5. “Dedications”. You must delete all sections Entitled “Endorsements. then if the Document is less than one half of the entire aggregate. unless and until the copyright holder explicitly and finally terminates your license.1 “Acknowledgements”. Any attempt otherwise to copy. 7. When the Document is included in an aggregate. Termination of your rights under this section does not terminate the licenses of parties who have received copies or rights from you under this License. If a section in the Document is Entitled “Acknowledgements”. and (b) permanently. and any sections Entitled “Dedications”. your license from a particular copyright holder is reinstated permanently if the copyright holder notifies you of the violation by some reasonable means. If your rights have been . provided you insert a copy of this License into the extracted document. the requirement (section 4) to Preserve its Title (section 1) will typically require changing the actual title. the original version will prevail. this License does not apply to the other works in the aggregate which are not themselves derivative works of the Document. Otherwise they must appear on printed covers that bracket the whole aggregate. provided that you also include the original English version of this License and the original versions of those notices and disclaimers. In case of a disagreement between the translation and the original version of this License or a notice or disclaimer. but you may include translations of some or all Invariant Sections in addition to the original versions of these Invariant Sections. Replacing Invariant Sections with translations requires special permission from their copyright holders. this is the first time you have received notice of violation of this License (for any work) from that copyright holder. 9. if the copyright holder fails to notify you of the violation by some reasonable means prior to 60 days after the cessation. and follow this License in all other respects regarding verbatim copying of that document. or distribute it is void. COLLECTIONS OF DOCUMENTS You may make a collection consisting of the Document and other documents released under this License. and replace the individual copies of this License in the various documents with a single copy that is included in the collection. or “History”. so you may distribute translations of the Document under the terms of section 4. You may extract a single document from such a collection. and distribute it individually under this License. then your license from a particular copyright holder is reinstated (a) provisionally. TERMINATION You may not copy. is called an “aggregate” if the copyright resulting from the compilation is not used to limit the legal rights of the compilation’s users beyond what the individual works permit. TRANSLATION Translation is considered a kind of modification. modify. or the electronic equivalent of covers if the Document is in electronic form. provided that you follow the rules of this License for verbatim copying of each of the documents in all other respects. and any Warranty Disclaimers.” 6. Moreover. in or on a volume of a storage or distribution medium. 8. the Document’s Cover Texts may be placed on covers that bracket the Document within the aggregate. and all the license notices in the Document. sublicense. or distribute the Document except as expressly provided under this License. if you cease all violation of this License. modify. If the Cover Text requirement of section 3 is applicable to these copies of the Document. You may include a translation of this License. sublicense. and will automatically terminate your rights under this License. AGGREGATION WITH INDEPENDENT WORKS A compilation of the Document or its derivatives with other separate and independent documents or works. and you cure the violation prior to 30 days after your receipt of the notice.
Appendix C: GNU Free Documentation License
129
terminated and not permanently reinstated, receipt of a copy of some or all of the same material does not give you any rights to use it. 10. FUTURE REVISIONS OF THIS LICENSE The Free Software Foundation may publish new, revised versions of the GNU Free Documentation License from time to time. Such new versions will be similar in spirit to the present version, but may differ in detail to address new problems or concerns. See http://www.gnu.org/copyleft/. Each version of the License is given a distinguishing version number. If the Document specifies that a particular numbered version of this License “or any later version” applies to it, you have the option of following the terms and conditions either of that specified version or of any later version that has been published (not as a draft) by the Free Software Foundation. If the Document does not specify a version number of this License, you may choose any version ever published (not as a draft) by the Free Software Foundation. If the Document specifies that a proxy can decide which future versions of this License can be used, that proxy’s public statement of acceptance of a version permanently authorizes you to choose that version for the Document. 11. RELICENSING “Massive Multiauthor Collaboration Site” (or “MMC Site”) means any World Wide Web server that publishes copyrightable works and also provides prominent facilities for anybody to edit those works. A public wiki that anybody can edit is an example of such a server. A “Massive Multiauthor Collaboration” (or “MMC”) contained in the site means any set of copyrightable works thus published on the MMC site. “CCBYSA” means the Creative Commons AttributionShare Alike 3.0 license published by Creative Commons Corporation, a notforprofit corporation with a principal place of business in San Francisco, California, as well as future copyleft versions of that license published by that same organization. “Incorporate” means to publish or republish a Document, in whole or in part, as part of another Document. An MMC is “eligible for relicensing” if it is licensed under this License, and if all works that were first published under this License somewhere other than this MMC, and subsequently incorporated in whole or in part into the MMC, (1) had no cover texts or invariant sections, and (2) were thus incorporated prior to November 1, 2008. The operator of an MMC Site may republish an MMC contained in the site under CCBYSA on the same site at any time before August 1, 2009, provided the MMC is eligible for relicensing.
130
GNU MP 5.0.1
ADDENDUM: How to use this License for your documents
To use this License in a document you have written, include a copy of the License in the document and put the following copyright and license notices just after the title page:
Copyright (C) year your name. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, no FrontCover Texts, and no BackCover Texts. A copy of the license is included in the section entitled ‘‘GNU Free Documentation License’’.
If you have Invariant Sections, FrontCover Texts and BackCover Texts, replace the “with. . . Texts.” line with this:
with the Invariant Sections being list their titles, with the FrontCover Texts being list, and with the BackCover Texts being list.
If you have Invariant Sections without Cover Texts, or some other combination of the three, merge those two alternatives to suit the situation. If your document contains nontrivial examples of program code, we recommend releasing these examples in parallel under your choice of free software license, such as the GNU General Public License, to permit their use in free software.
Concept Index
131
Concept Index
#
#include . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
B
Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Berkeley MP compatible functions . . . . . . . . . . . . . 7, 84 Binomial coefficient algorithm . . . . . . . . . . . . . . . . . . . 107 Binomial coefficient functions. . . . . . . . . . . . . . . . . . . . . 37 Binutils strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Bit manipulation functions . . . . . . . . . . . . . . . . . . . . . . . 38 Bit scanning functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Bit shift left . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Bit shift right . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 Bits per limb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 BSD MP compatible functions . . . . . . . . . . . . . . . . . 7, 84 Bug reporting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Build directory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Build notes for binary packaging . . . . . . . . . . . . . . . . . 11 Build notes for particular systems . . . . . . . . . . . . . . . . 12 Build options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Build problems known . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Build system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Building GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Bus error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
build . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 disablefft . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 disableshared . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 disablestatic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 enablealloca . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 enableassert . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 enablecxx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 ‘enablefat’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 enablempbsd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 enableprofiling . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 25 execprefix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 host . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 prefix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 finstrumentfunctions . . . . . . . . . . . . . . . . . . . . . . . . 26
2
2exp functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
C
C compiler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 C++ compiler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 C++ interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 C++ interface internals . . . . . . . . . . . . . . . . . . . . . . . . . . 118 C++ istream input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 C++ ostream output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 C++ support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 CC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 CC_FOR_BUILD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 CFLAGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Checker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 checkergcc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Code organisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Compaq C++ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Comparison functions . . . . . . . . . . . . . . . . . . . . . 37, 46, 53 Compatibility with older versions . . . . . . . . . . . . . . . . . 20 Conditions for copying GNU MP . . . . . . . . . . . . . . . . . . 1 Configuring GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Congruence algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Congruence functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 Conventions for parameters . . . . . . . . . . . . . . . . . . . . . . 18 Conventions for variables . . . . . . . . . . . . . . . . . . . . . . . . . 17 Conversion functions . . . . . . . . . . . . . . . . . . . . . . 31, 45, 51 Copying conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 CPPFLAGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 CPU types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2, 4 Cross compiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Custom allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 CXX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 CXXFLAGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Cygwin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
6
68000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
8
80x86 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
A
ABI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5, 8 About this manual . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 AC_CHECK_LIB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 AIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10, 12 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 alloca . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Allocation of memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 AMD64 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Anonymous FTP of latest version . . . . . . . . . . . . . . . . . 2 Application Binary Interface. . . . . . . . . . . . . . . . . . . . . . . 8 Arithmetic functions . . . . . . . . . . . . . . . . . . . . . . 32, 45, 52 ARM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Assembly cache handling . . . . . . . . . . . . . . . . . . . . . . . . 110 Assembly carry propagation . . . . . . . . . . . . . . . . . . . . . 110 Assembly code organisation . . . . . . . . . . . . . . . . . . . . . 109 Assembly coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 Assembly floating Point . . . . . . . . . . . . . . . . . . . . . . . . . 111 Assembly loop unrolling . . . . . . . . . . . . . . . . . . . . . . . . . 113 Assembly SIMD. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Assembly software pipelining . . . . . . . . . . . . . . . . . . . . 112 Assembly writing guide . . . . . . . . . . . . . . . . . . . . . . . . . 113 Assertion checking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 24 Assignment functions . . . . . . . . . . . . . . . . . . 30, 44, 50, 51 Autoconf. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
. . . . . . . . . . . . . 21 Extended GCD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 GCD algorithms . . . . . . . . . . 12 Headers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Float assignment functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 Integer root functions . . . . . . . . . . . . . . 50. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Floating point mode . . . . . . . . . . . . . 51. . . . . . . . . . . . . . . . . . . 13 DocBook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48. . . . . . . . . . . . . . . . . . . . . . 22 Division algorithms . . . . . . . . . . . . . . . . . 36 GDB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Float functions . . . . . . . . . . . . . . . . . . 100 Greatest common divisor functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Divisibility functions . . . . . . . . . . . . . . . . . . . 16 fnccheck . . . . . . . . . . . . . . . . . . . . . . 17 I I/O functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 H Hardware floating point mode . . . . . . . . . . . . . . . . . . 107 Factorial functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 Fibonacci number algorithm . . . . . . . 51 Float input and output functions . . . . 26 Greatest common divisor algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 Integer miscellaneous functions . . . 9 F Factor removal functions . . . . . . . . . . . . . . . . . 48 Floatingpoint number . . . . . . . . . . . . . . . . . . . . . . . . . 47. . . . . . . . . . . . . . . . . 41 Expression parsing demo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 GNU strip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 E Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Integer import . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Exact remainder . . . . . . . . . . . . 26 G GCC Checker . . . . . . . . . . . . . . . . . .h’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Float arithmetic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 Division functions . . . . 74 Install prefix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31. . . . . . . . . . . 39. . . . . . . . . . . . . . . . . . . . . . . . 9 Import . . . . . . . 30 Integer bit manipulation functions . . . . . 48. . . . . . . . . . . . 12 Floatingpoint functions . 54 Float random number functions . . . . . . . . . 65 Initializing and clearing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 DLLs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30. . . 54 Float sign tests . . . . . . . . . . . . . . . 24 Home page . . . . . . . . . . . . 53 Export . . . . . . . . . . . . . . . . . . . . . . 32 Integer exponentiation functions . . . . . . . . . . . . . . . . 8 instrumentfunctions . . . . . . . 37 Factorial algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29. . . . . . . . . . . . . . . . . . . . . . . . . . 29. . . . . . 4 HPUX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Fibonacci sequence functions . . . . . . . . . . . . . 16 gmpxx. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Integer arithmetic functions . . . . . . . . . . . . . . . . . . . . . 7 FunctionCheck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Example programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 HPPA . 32 Integer assignment functions . . . . . . 37 Factorization demo . . . . . . . . . . . . . . . . . . . . . . . . 34. . . . . . . . . . . 27 Initialization functions . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Fast Fourier Transform . . . . . . . . 35 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 i386 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 Integer input and output functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 Float miscellaneous functions . . . . . . . . . . . . . . . . . . . . . .1 D Darwin . . 38 Integer comparison functions . . . . 14 Debugging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Integer division functions . . . . . 40 Inplace operations . . . . . . . . . . 14 gprof . . . . . . . . . 51 FTP of latest version . . . . . . . . . . . . . . . . . . . . . . 54 Float rounding functions . . 20 ‘gmp. . 36 GCD functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Documentation license . . . . . 39 Integer internals . . . . . . . . . . . . . . . . . . . . . . . . . 8 Documentation formats . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . . . . . . . 2 Host system . . . . . . . . . . . . . . . . . . . . . . . . . . 16 infolookupsymbol . . . 47. . . . . . . . . . . . . . . 3 Installing GMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Integer export. . . . . . . . . . 7. . . . . . . . . . 21 Input functions. . . . . . . . 3 Execution profiling . . . . . . . . . . . 48 Float initialization functions . . . . . . . . . 76 GNU Debugger . . . . . . . . . . . . 94 Fat binary . . . . . . . . . . . . . . . . . . . . . . . 37 Integer conversion functions . . . . . . . . .h. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Heap problems . . . . . . . . . . . . . . . . . . . . . 14 IA64 . . . . . . . . . . . . . . . . . . 5 GMP Perl module. . . . . . . . . . . . . . . . . . . . . . . . . . . . 45. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Float comparison functions . . . . . . . . . 26 Integer . . . . . . . 2 Function classes . . . . . . . . . . . 44. . . . . . . . . . . . . . . . . . . 23 Demonstration programs . . . . . . . . . . . . . . . . . . . . . . . . . . 32. . . . . . . 34 Divisibility testing . . . . . . . . . . . . . . . . . . . . . . . 40 Integer initialization functions . . . . . . 22 Include files . . . . . . . . . . . 24 GNU Free Documentation License . . . . . . . . . . . . . . . 115 Integer logical functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Exec prefix . . . . 39. . . . . . . . . . . . . . . . . . . . 124 frexp . . . 27 Exact division functions . . 53 Float internals . . . 24 Generic C . . . . . . . . . . . . . . . . 21 Emacs . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 Integer functions . . . . . . . . . . . 52 DJGPP . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 Free Documentation License. . . . . 42 Divisibility algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Float conversion functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 FFT multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . .132 GNU MP 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 GMP version number . . . . . . 124 DVI. 3 Instruction Set Architecture . . . 25 Exponentiation functions . . . . . . . . . . . . . . 26 Formatted input . . 100 GCD extended . . . . . . . . . 42 Integer random number functions. . 20 Digits in an integer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Formatted output . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Malloc problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Linear congruential algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . ostream output . . . . 14 ISA . . . . . . . . . . . . . . . . . . . . . 35 Perfect square algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Latest version of GMP . . . . . . . . . . . . 14 Mailing lists . . . . . . . . . . . . . 19 Multiplication algorithms . . . . . . . . . . 2 Malloc debugger . . . . . OpenBSD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 O obstack output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Linking . . . . . . . . . . . . . . . . . . . . . . . . . 13 R Radix conversion algorithms. . . . . . 37 P Packaged builds . . . . . . . . . . . . . 16 libgmpxx . . . . . . . . . . . . . 53 PowerPC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12. . . 16 Libraries . . . . . . . . . 104 Number sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Memory allocation . . . . . . . . . . . . . 7 Perfect power algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 13 15 70 88 69 L Language bindings . . . . . . . . . . . . . . . . . . . 53. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Least common multiple functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 NeXT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Random number algorithms . . . . . . . . 16 NonUnix systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 MSDOS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Optimizing performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Multithreading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Legendre symbol functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 MMX . . . . . . . . . . . . . . . . . . . . . . . . . . 67 Probable prime testing functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Concept Index 133 Integer sign tests . . . . . . . . . . . . . . . . . . . . . . 106 Prime testing functions . . 16 Libtool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Random state . . . . . . . . . . . . 86 Memory management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Prime testing algorithms . . . . . . . Output functions . . . . . . 16 Libtool versioning . . 14 Modular inverse functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Lucas number algorithm . . . 21 Particular systems . 2 LCM functions . . . . . . . . . . . . . . . . . . . . . . . 48 Precision of hardware floating point . . . 11 License conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 MINGW . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 perl . . . . . . . . 34. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 mp. . . . . . . . . 66 Random number state . . . . . . . . . . . . . . . . . . 46 J Jacobi symbol algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 printf formatted output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40. . 103 Kronecker symbol functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47. . . . . . . . . . . . . . . . . . . . . 84 MPN_PATH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Perl module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Precision of floats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Rational arithmetic . . . . . . . 2 Inverse modulo functions . . . . . 102 Jacobi symbol functions . . . . . . . . . . . . . . . 108 Lucas number functions . . . . . . . . . . . 36 K Karatsuba multiplication . . . . . . . . . . . . . . . . . . . . . . 16 Logical functions . . . . . . . . . . . . . . . . . . . . 18 Parsing expressions demo . . . . . . . . . . . . . . 12 Prefix . . . . . . . . 90 N Nails . . . . . . . . . . . 102 Powering functions . . . . . . . . . 108 Mersenne twister random numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 istream input. . 35 Numerator and denominator . . . . . . . . . . . . . . 12 Past GMP versions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. . . . . . . . . . . . . . . . 36 Nomenclature . . . . . . 15 Next prime function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Most significant bit . . 17 Limb size . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Linear congruential random numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 Native compilation . . . . . . . . . . 104 Perfect power functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Powering algorithms . . . 13. . . . . . . . . . . . 21 Postscript . . . 11 Parameter conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Nth root algorithm. . . . . . . . . . . 108 Random number functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 MS Windows . . . . . . . . . . . . . . . . 9 Miscellaneous float functions . . . . . . . . . . 7 Power/PowerPC . . . . . . . . . . . . . . 91 Karatsuba square root algorithm . . . . 12 MIPS . . . . . . . 38 Lowlevel functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 PDF . . . . . . . . . . . . . . . . . . 54 Miscellaneous integer functions . . . . . . . . . . . . . . . . 23 Number theoretic functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .h . . . . 35 prof . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Random number seeding . . . . . . . . . . . . . . . 23 . . . . . . . . . . . . 19 Mersenne twister algorithm . . 104 Perfect square functions . . . Other languages. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 IRIX . . . . 38 Integer special functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Limb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 M MacOS X . . . . . . . . . . . . . . . . . . . . . . . . 36 libgmp . . . . . . . 42 Interix . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 scanf formatted input . . . . . . . . . 38. . . . . . . . . . . . . 16 Rational number functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .h. . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Rational assignment functions . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Small operands . . . . . . . . . . . . . . . 12. . . . . . . 7 Text input/output . 25 Variable conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 XML. . . . . . 7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .h . . . . . . . . . . . . . . . . 118 Reallocations . . . . . . . . . . . . 19 Toom multiplication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 SCO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Rational initialization functions . . . . . . . . . 24 23 21 16 16 14 10 13 12 T Temporary memory . . . . . . . 44 Rational input and output functions . . . . . . . . . . . . . . . . . . . . . .0. . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 X x86 . 13 Sparc V9. . . . . . . . . . . . . . . . . . . SunOS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Square root algorithm . . 23 Sequent Symmetry . 53 Size in digits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Version number . . . . . . . . . . . . . . . . . . . . . . . Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Texinfo . 23 Thread safety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . stdarg. . . . . . . . . . . . . . . . . . . . . . . . 12 Shared library versioning . . . . . . . . . . . . . . . . . .1 Rational arithmetic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Rational comparison functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 Segmentation violation . . 20 Scan bit functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Static linking . . . . . . . . . . . . . . . . . . . . . . 37 Reporting bugs . . . . . . . . . 10 Special integer functions . . . . . . . 14 x87 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 W Web page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Raw output internals. . . . . . . . . . . . . . . . . . . . . . . . . . 54 Stack backtrace . 95 Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Solaris . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Sparc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Root extraction functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Rational conversion functions. . . . . . . . . . Stack overflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Rational internals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 Root extraction algorithms . . . . . stdio. . . . . . . . 19 References . . . . 46 Rational sign tests . . . . 14 V Valgrind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Reentrancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sun . . . . . . . . . . . . . . . . . . . . . . . . 15 Services for Unix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Userdefined precision . Unbalanced multiplication . . . . . . . . Upward compatibility . . . . . . . . 92. . . . . . . . . . . . . . . . . . . . . 22 96 20 20 48 S Sample programs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Sign tests . . . . . . . . . . . . . . . . . . . . . . . .134 GNU MP 5. . . . . . . . 10. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Root extraction algorithm . . . . . . . . . . . . . . . . . . . . 2 Windows . . . . . . . . . . 103 SSE2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Rational number . . . . . . . . . . . 52 Root testing functions . . . . . . . . . . . . . . . Stripped libraries . . . . . . . . 16 U ui and si functions . . . . Useful macros and constants . . . . . . . . . . . . . . . . . 14 Seeding random numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35. . . . . . . . . . . . . 122 Remove factor functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 Rounding functions . 44 Rational numerator and denominator . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . gmp_vsnprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78. . . . . . . . . . . . . . . . . . . . GMP_ERROR_INVALID_ARGUMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::get_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::set_prec . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::fits_slong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_obstack_printf . . . . . . . . . . . . . . . . . . . . . . . . . min . . mpf_class::get_prec . . . . . . . . . . mpf_class::get_d . . . . . . . . . . . 81 cmp . . . . . . . . . . . mpf_class::operator= . . . . . gmp_randinit_mt . . . . . . . . . . . . . . . . . . . gmp_obstack_vprintf . . . . . . . . . . mpf_add. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . GMP_NAIL_BITS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::set_prec_raw . gmp_randclass::get_f . . . . . mdiv . . . . . . . . . . . . . . . . . . . . . . . . . mpf_ceil . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_randclass . . . . . . . . . . . . . . . . . . mp_exp_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 20 20 20 20 43 A abs . . . . . . . . . . . . . . . . . . . . . . . . gmp_scanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mp_set_memory_functions . . . . . . . . . GMP_NUMB_MASK . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_randclear . . GMP_ERROR_UNSUPPORTED_ARGUMENT . . mpf_class::fits_sint_p . . . . . . . . . . . . . . . . . . . . mpf_class::fits_ushort_p . . . GMP_NUMB_BITS . . . gmp_vfscanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::fits_uint_p . . . . . . .Function and Type Index 135 Function and Type Index gmp_sscanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::mpf_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . __GMP_CFLAGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 G gcd . . . . . . . . . . . gmp_randinit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_add_ui . . . . . . mpf_class::fits_ulong_p . . . . . . . . . . . . . . mpf_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mp_size_t . . . . . . . . . . . . . . . . __GNU_MP_VERSION . . . . . . . . . mp_bitcnt_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_randinit_lc_2exp_size . . . __GNU_MP_VERSION_PATCHLEVEL. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_randseed_ui . . . . . . . . . . . . gmp_vscanf . . . . . . . . . . . . . . . . GMP_RAND_ALG_LC . . . . . . . . . . gmp_printf . . . gmp_randclass::get_z_range . . . . . . . . 81 F floor . . . . . . . . . . . gmp_fprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 85 84 85 85 84 85 84 17 20 16 87 17 86 17 53 52 52 54 76 81 81 81 81 81 81 81 77 81 81 81 81 80 80 81 81 81 . . . . . . . . . . . . . . . . . . . . . . mpf_class::get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . __GNU_MP_VERSION_MINOR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . GMP_NAIL_MASK . . . . . . . . . . 79. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 C ceil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_randclass::gmp_randclass . . . . . . . . . . . . . . . . . mp_bits_per_limb . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_vfprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . MINT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mcmp . . gmp_randinit_lc_2exp . . . . . . . . . . . . . . . . . . gmp_randinit_set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . GMP_LIMB_BITS . . . . gmp_snprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_vsprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mfree . . . . . . . . . . . . . . 81 I itom . . . mout . . . . . . . 81 H hypot . . . . . . . . . move . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_randseed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 70 66 66 66 69 74 63 63 63 63 63 64 70 70 69 65 65 81 82 82 82 82 82 66 65 65 65 65 65 65 66 66 17 74 69 69 M madd . . . . . GMP_NUMB_MAX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_urandomm_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::get_mpf_t . . . . gmp_errno . . . . . . . . . . . . . . . . . . . . . . . . gmp_asprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_class::fits_sshort_p . . . . . . . . . . . . . . . . . gmp_fscanf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_vasprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mp_limb_t . . . . . . . . . . . . . . . . . . . . . _mpz_realloc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_sprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_urandomb_ui . . . . . . . . gmp_vsscanf . . . . gmp_randinit_default . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_abs. . . . . . . . . . . . . . . . . . . . . . . . . . . . gmp_randclass::seed . . . . . . . . . gmp_vprintf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . GMP_RAND_ALG_DEFAULT . mpf_class::get_si . . . . . . . . . . . . . . . . . . . . . . . . . . mp_get_memory_functions . gmp_randstate_t . . . . gmp_randclass::get_z_bits . . . . . . 74 66 66 70 20 69 74 69 74 69 69 74 __GMP_CC . . . . . . . . . . . . . . . . . . gmp_version . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_ui_div . . . . . . . . . mpf_sub_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_trunc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_div_2exp . . . . . . . . . . . mpf_inp_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_sqr. mpf_sub. . . . mpf_fits_ulong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_set_prec . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_set_ui . . . mpq_abs. . . . . . mpf_eq . . mpf_t . . . . . . . . . . . . . . . . mpn_divexact_by3 . . . . . . . . . . . . . . . . . mpf_get_str . . mpf_div_ui . . . . . . . . . . . . . . . . . . . . . mpf_get_prec . . . . . . . . . . . . . mpf_set_z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_sqrtrem . . . . . . . . . . . . . . . . . . . . mpn_divmod . . . . . . . . . . . . . . . mpn_and_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_zero . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_gcd_1 . . . . mpf_get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_reldiff . . . . . . . . . . . . . . . . . . mpn_sub_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_clears . . . . . . . . . mpf_mul_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_mod_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_xnor_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_random2 . . . . . . . . . . . . . . . . . . . . . . .1 mpf_clear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_tdiv_qr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_fits_slong_p . . . . . . . . . . . . . . . . . . . . . . . . mpn_submul_1 . . . . . . . . . . . . . . . . . . . mpn_mul_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_divexact_by3c . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_set_q . . . . . mpn_mul. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_canonicalize . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_init2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_div. . . . . . . . mpn_mul_1 . . . . . . . . . . . . . mpn_copyi . . . . . . . . mpf_cmp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_nand_n . . . . . . . . . . . . 49 49 53 53 53 53 52 53 52 53 54 54 54 54 54 54 54 51 51 48 49 51 52 51 49 51 51 51 51 51 49 49 54 54 52 53 52 53 54 53 55 53 50 50 48 49 49 50 50 50 50 50 53 52 52 52 52 50 16 54 52 52 54 56 56 mpn_add_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_pow_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_class::get_num . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_sub_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_integer_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_neg. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_init_set . . . . . . . . . . . . . . . . mpn_perfect_square_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_sgn. . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_copyd . . . . . . . . . . . . . . . . mpf_init_set_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_set_prec_raw . . . . . . . . . . . . . . . . . . . mpf_get_default_prec . . . . . . mpf_set_si . . . . . . . . . . . . . . . . . . . . . mpq_class::get_den . mpf_ui_sub . . . . mpf_inits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_gcdext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_neg. . . . . . . . . . . . . . . mpq_class::get_mpq_t . . . . . . . . . . . . . mpf_fits_sshort_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_init_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_init_set_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_set_d . . . . . . . . . . . . . . . . . . . . . . mpq_add. . . . . . . . . . . . . . . . . . . . . . . . . . mpf_get_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_lshift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_class::get_den_mpz_t . . . . . . . . . . . . . . . . . . . . mpf_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_xor_n . . mpn_divmod_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_popcount . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_hamdist . . . . . . . . . . . . . . mpf_cmp_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_class::mpq_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_com. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_class::get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_class::get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_nior_n . . . . . . . . mpf_get_si . . . . . . . . . . mpf_floor . . . . . . . . . . . . . . . . mpf_sqrt_ui . mpn_add. mpq_class::get_num_mpz_t . mpn_add_1 . . . . mpf_init . . . . . . 78. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_cmp_ui . . . . . . . . . mpn_ior_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_cmp. . mpn_divrem . . . mpf_fits_ushort_p . . . . . . . . . . . . . . . . . . . mpn_scan1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_cmp. . . . . . . . . . . . . mpn_gcd. . . . . . . . . . . . . . . . . . . . . mpn_divrem_1 . mpn_rshift . . . . . . . . . . . . . . . . . . . . . . . mpn_random . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_scan0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 58 62 62 60 63 63 63 59 59 59 59 58 59 60 60 60 61 62 62 62 59 59 57 57 57 62 57 62 62 62 61 61 60 61 61 61 57 60 57 57 57 58 58 62 62 63 46 45 44 76 79 79 79 79 77 79 79 79 79 79 44 44 46 46 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_clear . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_out_str . . . . . . . mpf_mul_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_class::set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_random2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_iorn_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_fits_uint_p . . mpf_set_default_prec . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_sub. . . . . . mpf_clears . . . . . mpn_get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_fits_sint_p . . . . . . . . mpf_mul. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_set. . mpf_cmp_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_init_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpn_set_str . . . . . . . . . . . . . . . . . . . . . . . . mpf_urandomb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .0. . . . . . . . . . . . . . . . . . . . . . . . mpf_get_d_2exp . . . . . . . . . . . . . . . . . mpf_swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .136 GNU MP 5. . mpn_andn_n . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpf_sqrt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_cmp_si . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_class::canonicalize . . . . . . . . . . . . . . . mpn_addmul_1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . mpz_class::get_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cmp_ui . . . . . . . . . . . . . . . . . . . . . . . . . mpz_class::fits_slong_p . . mpz_class::mpz_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_divexact . . . . . . . . . . . . . . . mpz_cdiv_q_2exp . . . . . . . . . . . . . . . . . . . . . . . . mpq_denref . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cdiv_qr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_congruent_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_get_ui . . . . . . . . . . . . . . . . . . . . . . . . mpz_init_set_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fac_ui . . . . . . . . . . . . . . . . . . . . . . . mpz_cmpabs . . . . . . . . . . mpz_inp_raw . . mpz_fits_slong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_addmul_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_set_num . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_set_d . . . . . . . mpz_cdiv_ui . . . . . . . . . . . . . . . mpz_fits_sint_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_clear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_class::get_mpz_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_bin_ui . . . . . . . . . . . . . . . . . . . . . mpz_cdiv_r . . . . . . . . . . . . . . . mpz_fdiv_r_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_set_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fits_ulong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_set_f . . . . . . . . . . . . . . . . . . . . . . . . mpz_addmul . mpz_fdiv_q_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cmp_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cmpabs_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cmp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_equal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fdiv_r_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fib_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_class::get_str . . . . . . . . . . . . . . . mpq_sgn. . . . . . . . . . . . . . . . . . . mpq_set_den . . . . . . . mpz_class::get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_divisible_p . . mpq_div_2exp . . . . . . . . . . . . . . . . . . . . . . . mpz_clears . . . . . . . . . . . . . . . . . mpz_gcd. . mpz_divisible_2exp_p . . . mpz_get_d_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 46 45 46 46 45 47 47 45 44 44 47 46 45 45 46 46 47 44 45 47 45 47 44 44 44 44 46 45 45 16 32 32 32 32 32 38 42 37 37 32 33 32 32 32 32 33 32 33 76 78 78 78 78 78 78 78 77 78 78 78 77 78 29 29 mpz_clrbit . . . . . . . . . . . . . . . . . mpz_hamdist . . . . . . . . . . mpz_fits_ushort_p . . . . . . . . mpz_divisible_ui_p . . . . . . . . . . . . . . . . . mpq_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_get_den . . . . . . . . . . . . . . mpq_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_init_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cdiv_r_2exp . . . . . . . . . mpz_getlimbn . . . . . . . . . . . . mpq_mul_2exp . . . . . . . . . . . . . . . . . . mpz_even_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fdiv_qr . . . . . . . . . . . . . . . . . . . . . . . . . mpz_init_set_d . . . . mpz_class::fits_sshort_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_and. . . . . . . . mpq_out_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_inv. . . . . . . . . . . . . . . . . . . . . . . . . mpz_class::fits_uint_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_inits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_bin_uiui . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cdiv_qr_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cmpabs_ui . . . . . . . . . . . . . . . . mpq_inits . . . . . . . . . . mpq_numref . . . . . . . mpz_fib2_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_inp_str . . . . . . . . . . . . . . . . mpz_congruent_2exp_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_swap . . . . . . . . mpz_fits_uint_p . . . . . . . . . . . . . . . . . . . . . mpz_init_set_str . . . . . . . . . . . . . . . . . . . . mpz_init_set . . . . . . . . . . . . . . . . . . . . . . mpz_cdiv_r_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_divexact_ui . . mpz_array_init . . . . . . . . . . . . mpq_init . . . . mpz_congruent_ui_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cmp_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_class::fits_ushort_p . . . . . . . . . . . . . . . . . . . mpq_get_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_gcdext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_cdiv_q_ui . . . . . . . . . . . . mpz_combit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_ior. . . . . . . . . . . . . mpz_abs. . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_gcd_ui . . mpz_fdiv_r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Function and Type Index 137 mpq_cmp_ui . . . . . . . . . . . . . . . . . . . . mpz_get_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_add_ui . . . . . . . . . . . . . mpz_kronecker_ui . . . . . . . . . . . . . . . . . . . . . . . . mpz_class::fits_ulong_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_init . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_get_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_set_z . . . . . . . . . . . . . . . . . . . . . . 38 37 37 37 37 37 37 37 38 39 34 34 34 34 34 34 34 34 42 41 37 33 33 33 33 33 33 33 33 33 37 37 42 42 42 42 42 42 36 36 36 31 31 31 31 31 43 38 40 29 30 30 30 31 30 29 29 39 39 36 38 36 36 36 36 . . . . . . . . . . mpq_set. . . . . . . . . . . . . . . . . . . . . . mpz_cdiv_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fits_sshort_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fdiv_q_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fdiv_qr_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_invert . . . . . . . . . . . . . . . . . . mpz_class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_get_d . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_kronecker_si . . . . . . . . . . . . . . . . . . . . mpz_class::set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_class::fits_sint_p . . . . . . . . . . . . . . . . . . . . . mpq_neg. . . . . . . . . . . . . . . . . . . . . . mpz_com. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_fdiv_ui . . . . . . . . . . . . . . . . . mpq_mul. . . . . . . . . . . . . . . . . . . . . . . . . . mpz_import . . . . . . . . . . . . . . . mpz_add. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_export . . . . . . mpz_jacobi . . . . . . . . . mpz_fdiv_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_class::get_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_get_num . . . . . . . . . . . . . . . . . . . . . . . . mpz_init2 . . . . mpq_inp_str . . . . . . . . . . . . . . . . . . . . . . . mpq_div. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpq_sub. . . . . . . . . . . . . . . mpz_kronecker . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . mpz_urandomb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_ui_sub . . . . . . . . . . 74. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_submul_ui . . . . . . . msub . . . . 78. . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_mod. . . . . . mpz_perfect_power_p . . . . . . . . . . . . . mpz_odd_p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_scan1 . . . . . . . . . . . . . . . mpz_sqrtrem . . . . . . . . . . . mpz_realloc2 . . . . . . . . . . . . . . mpz_set_ui . . . . . . . . . . . . mpz_si_kronecker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mult . . 30 16 33 33 33 33 33 33 33 33 33 39 36 35 32 40 40 38 84 84 85 84 O operator% . . . . . . . . . . . mpz_random2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_nextprime . . . . . . . . . . . . . . . . . . . . . . . . . 75. . . . .138 GNU MP 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_set_str . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_set_f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_mod_ui . . . . . . . . . . . . . . . . . . mpz_t . . . mpz_sub_ui . . . . . . . . . . . . . . . . . . . . . mpz_tstbit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . operator/ . . . . . . . . . . . . . . . . . . . . . . . . . mpz_tdiv_qr_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 sqrt . . mpz_ui_kronecker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_popcount . . . . . . . . . . . . . . . mpz_powm_sec . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_mul_ui . . . . . . . . mpz_lucnum2_ui . . mpz_legendre . . . . . . . . . . . . . . . . . . . . . . . . 36 36 36 37 37 34 34 32 32 32 32 32 36 42 39 39 35 35 38 35 34 34 34 35 40 40 29 37 35 35 40 38 38 30 30 30 30 30 30 30 38 38 36 43 42 35 35 32 32 32 32 mpz_swap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_neg. . . . . . . . . . . . . . . . . . . . . . . . mpz_remove . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_out_raw . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 R rpow . . . . . mpz_tdiv_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79. . . . . . . . . . . . . . . . . . . . . mpz_sub. . . . . . mpz_set_q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_mul_2exp . . . . . . . . . . . . . . . . . . . . . . . 81 X xtom . . mpz_tdiv_r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_scan0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_random . . . . . . . . . . . . . . . . . . . . . . mpz_sizeinbase . . . . . . . . . . . . . . . . . . . . mpz_set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_mul. . . . . . . . . . . . mpz_tdiv_q_2exp . mpz_pow_ui . . . . . . . . . mpz_perfect_square_p . . . . . . . . . . . . . . . . . . mpz_submul . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 78 70 80 P pow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_ui_pow_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_powm . . . . . . . . operator>> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_lucnum_ui . mpz_set_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_xor. . . . . . . . . . . . . . . . . . . . . . . . mpz_probab_prime_p . . . . . . . . . . . . . mpz_tdiv_r_2exp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_setbit . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_sqrt . . . mpz_mul_si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_out_str . . . . . . mpz_size . operator<< . . . . . . . . . . . . . . . . . . . . . . . . . mpz_urandomm . . . . . . . . . . . . . mpz_root . . . . . . . .1 mpz_lcm. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_tdiv_ui . . . . . . . . mpz_sgn. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 T trunc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_powm_ui . . . . . . mpz_tdiv_q_ui . . mpz_tdiv_qr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 sgn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_rrandomb . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_rootrem . . . . . . . . . . . . mpz_set_d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_lcm_ui . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 S sdiv . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . msqrt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . mpz_tdiv_r_ui . . . . . . mtox . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 . . . . . . . . .