blob: fd7ff3d91e6a920ff084beca09d10b5b9abba981 [file] [log] [blame]
David Howellsf0d1b0b2006-12-08 02:37:49 -08001/* Integer base 2 logarithm calculation
2 *
3 * Copyright (C) 2006 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#ifndef _LINUX_LOG2_H
13#define _LINUX_LOG2_H
14
15#include <linux/types.h>
16#include <linux/bitops.h>
17
18/*
19 * deal with unrepresentable constant logarithms
20 */
21extern __attribute__((const, noreturn))
22int ____ilog2_NaN(void);
23
24/*
25 * non-constant log of base 2 calculators
26 * - the arch may override these in asm/bitops.h if they can be implemented
27 * more efficiently than using fls() and fls64()
28 * - the arch is not required to handle n==0 if implementing the fallback
29 */
30#ifndef CONFIG_ARCH_HAS_ILOG2_U32
31static inline __attribute__((const))
32int __ilog2_u32(u32 n)
33{
34 return fls(n) - 1;
35}
36#endif
37
38#ifndef CONFIG_ARCH_HAS_ILOG2_U64
39static inline __attribute__((const))
40int __ilog2_u64(u64 n)
41{
42 return fls64(n) - 1;
43}
44#endif
45
David Howells312a0c12006-12-08 02:37:51 -080046/*
Robert P. J. Day63c2f782007-01-30 06:06:00 -050047 * Determine whether some value is a power of two, where zero is
48 * *not* considered a power of two.
49 */
50
51static inline __attribute__((const))
52bool is_power_of_2(unsigned long n)
53{
54 return (n != 0 && ((n & (n - 1)) == 0));
55}
56
57/*
David Howells312a0c12006-12-08 02:37:51 -080058 * round up to nearest power of two
59 */
60static inline __attribute__((const))
61unsigned long __roundup_pow_of_two(unsigned long n)
62{
63 return 1UL << fls_long(n - 1);
64}
65
Robert P. J. Dayb311e922007-10-16 23:29:32 -070066/*
67 * round down to nearest power of two
68 */
69static inline __attribute__((const))
70unsigned long __rounddown_pow_of_two(unsigned long n)
71{
72 return 1UL << (fls_long(n) - 1);
73}
74
David Howellsf0d1b0b2006-12-08 02:37:49 -080075/**
76 * ilog2 - log of base 2 of 32-bit or a 64-bit unsigned value
77 * @n - parameter
78 *
79 * constant-capable log of base 2 calculation
80 * - this can be used to initialise global variables from constant data, hence
81 * the massive ternary operator construction
82 *
83 * selects the appropriately-sized optimised version depending on sizeof(n)
84 */
85#define ilog2(n) \
86( \
87 __builtin_constant_p(n) ? ( \
88 (n) < 1 ? ____ilog2_NaN() : \
89 (n) & (1ULL << 63) ? 63 : \
90 (n) & (1ULL << 62) ? 62 : \
91 (n) & (1ULL << 61) ? 61 : \
92 (n) & (1ULL << 60) ? 60 : \
93 (n) & (1ULL << 59) ? 59 : \
94 (n) & (1ULL << 58) ? 58 : \
95 (n) & (1ULL << 57) ? 57 : \
96 (n) & (1ULL << 56) ? 56 : \
97 (n) & (1ULL << 55) ? 55 : \
98 (n) & (1ULL << 54) ? 54 : \
99 (n) & (1ULL << 53) ? 53 : \
100 (n) & (1ULL << 52) ? 52 : \
101 (n) & (1ULL << 51) ? 51 : \
102 (n) & (1ULL << 50) ? 50 : \
103 (n) & (1ULL << 49) ? 49 : \
104 (n) & (1ULL << 48) ? 48 : \
105 (n) & (1ULL << 47) ? 47 : \
106 (n) & (1ULL << 46) ? 46 : \
107 (n) & (1ULL << 45) ? 45 : \
108 (n) & (1ULL << 44) ? 44 : \
109 (n) & (1ULL << 43) ? 43 : \
110 (n) & (1ULL << 42) ? 42 : \
111 (n) & (1ULL << 41) ? 41 : \
112 (n) & (1ULL << 40) ? 40 : \
113 (n) & (1ULL << 39) ? 39 : \
114 (n) & (1ULL << 38) ? 38 : \
115 (n) & (1ULL << 37) ? 37 : \
116 (n) & (1ULL << 36) ? 36 : \
117 (n) & (1ULL << 35) ? 35 : \
118 (n) & (1ULL << 34) ? 34 : \
119 (n) & (1ULL << 33) ? 33 : \
120 (n) & (1ULL << 32) ? 32 : \
121 (n) & (1ULL << 31) ? 31 : \
122 (n) & (1ULL << 30) ? 30 : \
123 (n) & (1ULL << 29) ? 29 : \
124 (n) & (1ULL << 28) ? 28 : \
125 (n) & (1ULL << 27) ? 27 : \
126 (n) & (1ULL << 26) ? 26 : \
127 (n) & (1ULL << 25) ? 25 : \
128 (n) & (1ULL << 24) ? 24 : \
129 (n) & (1ULL << 23) ? 23 : \
130 (n) & (1ULL << 22) ? 22 : \
131 (n) & (1ULL << 21) ? 21 : \
132 (n) & (1ULL << 20) ? 20 : \
133 (n) & (1ULL << 19) ? 19 : \
134 (n) & (1ULL << 18) ? 18 : \
135 (n) & (1ULL << 17) ? 17 : \
136 (n) & (1ULL << 16) ? 16 : \
137 (n) & (1ULL << 15) ? 15 : \
138 (n) & (1ULL << 14) ? 14 : \
139 (n) & (1ULL << 13) ? 13 : \
140 (n) & (1ULL << 12) ? 12 : \
141 (n) & (1ULL << 11) ? 11 : \
142 (n) & (1ULL << 10) ? 10 : \
143 (n) & (1ULL << 9) ? 9 : \
144 (n) & (1ULL << 8) ? 8 : \
145 (n) & (1ULL << 7) ? 7 : \
146 (n) & (1ULL << 6) ? 6 : \
147 (n) & (1ULL << 5) ? 5 : \
148 (n) & (1ULL << 4) ? 4 : \
149 (n) & (1ULL << 3) ? 3 : \
150 (n) & (1ULL << 2) ? 2 : \
151 (n) & (1ULL << 1) ? 1 : \
152 (n) & (1ULL << 0) ? 0 : \
153 ____ilog2_NaN() \
154 ) : \
155 (sizeof(n) <= 4) ? \
156 __ilog2_u32(n) : \
157 __ilog2_u64(n) \
158 )
159
David Howells312a0c12006-12-08 02:37:51 -0800160/**
161 * roundup_pow_of_two - round the given value up to nearest power of two
162 * @n - parameter
163 *
Robert P. J. Day6fb189c2007-02-17 19:17:37 +0100164 * round the given value up to the nearest power of two
David Howells312a0c12006-12-08 02:37:51 -0800165 * - the result is undefined when n == 0
166 * - this can be used to initialise global variables from constant data
167 */
168#define roundup_pow_of_two(n) \
169( \
170 __builtin_constant_p(n) ? ( \
Rolf Eike Beer1a06a522007-05-17 23:56:56 +0200171 (n == 1) ? 1 : \
David Howells312a0c12006-12-08 02:37:51 -0800172 (1UL << (ilog2((n) - 1) + 1)) \
173 ) : \
174 __roundup_pow_of_two(n) \
175 )
176
Robert P. J. Dayb311e922007-10-16 23:29:32 -0700177/**
178 * rounddown_pow_of_two - round the given value down to nearest power of two
179 * @n - parameter
180 *
181 * round the given value down to the nearest power of two
182 * - the result is undefined when n == 0
183 * - this can be used to initialise global variables from constant data
184 */
185#define rounddown_pow_of_two(n) \
186( \
187 __builtin_constant_p(n) ? ( \
Robert P. J. Dayb311e922007-10-16 23:29:32 -0700188 (1UL << ilog2(n))) : \
189 __rounddown_pow_of_two(n) \
190 )
191
Robert P. J. Dayde9330d2008-02-06 01:36:54 -0800192/**
193 * order_base_2 - calculate the (rounded up) base 2 order of the argument
194 * @n: parameter
195 *
196 * The first few values calculated by this routine:
197 * ob2(0) = 0
198 * ob2(1) = 0
199 * ob2(2) = 1
200 * ob2(3) = 2
201 * ob2(4) = 2
202 * ob2(5) = 3
203 * ... and so on.
204 */
205
206#define order_base_2(n) ilog2(roundup_pow_of_two(n))
207
David Howellsf0d1b0b2006-12-08 02:37:49 -0800208#endif /* _LINUX_LOG2_H */