ia32intrin.h revision 1.3 1 1.3 mrg /* Copyright (C) 2009-2013 Free Software Foundation, Inc.
2 1.1 mrg
3 1.1 mrg This file is part of GCC.
4 1.1 mrg
5 1.1 mrg GCC is free software; you can redistribute it and/or modify
6 1.1 mrg it under the terms of the GNU General Public License as published by
7 1.1 mrg the Free Software Foundation; either version 3, or (at your option)
8 1.1 mrg any later version.
9 1.1 mrg
10 1.1 mrg GCC is distributed in the hope that it will be useful,
11 1.1 mrg but WITHOUT ANY WARRANTY; without even the implied warranty of
12 1.1 mrg MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 1.1 mrg GNU General Public License for more details.
14 1.1 mrg
15 1.1 mrg Under Section 7 of GPL version 3, you are granted additional
16 1.1 mrg permissions described in the GCC Runtime Library Exception, version
17 1.1 mrg 3.1, as published by the Free Software Foundation.
18 1.1 mrg
19 1.1 mrg You should have received a copy of the GNU General Public License and
20 1.1 mrg a copy of the GCC Runtime Library Exception along with this program;
21 1.1 mrg see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
22 1.1 mrg <http://www.gnu.org/licenses/>. */
23 1.1 mrg
24 1.1 mrg #ifndef _X86INTRIN_H_INCLUDED
25 1.1 mrg # error "Never use <ia32intrin.h> directly; include <x86intrin.h> instead."
26 1.1 mrg #endif
27 1.1 mrg
28 1.1 mrg /* 32bit bsf */
29 1.1 mrg extern __inline int
30 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
31 1.1 mrg __bsfd (int __X)
32 1.1 mrg {
33 1.1 mrg return __builtin_ctz (__X);
34 1.1 mrg }
35 1.1 mrg
36 1.1 mrg /* 32bit bsr */
37 1.1 mrg extern __inline int
38 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
39 1.1 mrg __bsrd (int __X)
40 1.1 mrg {
41 1.1 mrg return __builtin_ia32_bsrsi (__X);
42 1.1 mrg }
43 1.1 mrg
44 1.1 mrg /* 32bit bswap */
45 1.1 mrg extern __inline int
46 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
47 1.1 mrg __bswapd (int __X)
48 1.1 mrg {
49 1.1 mrg return __builtin_bswap32 (__X);
50 1.1 mrg }
51 1.1 mrg
52 1.1 mrg #ifdef __SSE4_2__
53 1.1 mrg /* 32bit accumulate CRC32 (polynomial 0x11EDC6F41) value. */
54 1.1 mrg extern __inline unsigned int
55 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
56 1.1 mrg __crc32b (unsigned int __C, unsigned char __V)
57 1.1 mrg {
58 1.1 mrg return __builtin_ia32_crc32qi (__C, __V);
59 1.1 mrg }
60 1.1 mrg
61 1.1 mrg extern __inline unsigned int
62 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
63 1.1 mrg __crc32w (unsigned int __C, unsigned short __V)
64 1.1 mrg {
65 1.1 mrg return __builtin_ia32_crc32hi (__C, __V);
66 1.1 mrg }
67 1.1 mrg
68 1.1 mrg extern __inline unsigned int
69 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
70 1.1 mrg __crc32d (unsigned int __C, unsigned int __V)
71 1.1 mrg {
72 1.1 mrg return __builtin_ia32_crc32si (__C, __V);
73 1.1 mrg }
74 1.1 mrg #endif /* SSE4.2 */
75 1.1 mrg
76 1.1 mrg /* 32bit popcnt */
77 1.1 mrg extern __inline int
78 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
79 1.1 mrg __popcntd (unsigned int __X)
80 1.1 mrg {
81 1.1 mrg return __builtin_popcount (__X);
82 1.1 mrg }
83 1.1 mrg
84 1.1 mrg /* rdpmc */
85 1.1 mrg extern __inline unsigned long long
86 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
87 1.1 mrg __rdpmc (int __S)
88 1.1 mrg {
89 1.1 mrg return __builtin_ia32_rdpmc (__S);
90 1.1 mrg }
91 1.1 mrg
92 1.1 mrg /* rdtsc */
93 1.1 mrg extern __inline unsigned long long
94 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
95 1.1 mrg __rdtsc (void)
96 1.1 mrg {
97 1.1 mrg return __builtin_ia32_rdtsc ();
98 1.1 mrg }
99 1.1 mrg
100 1.1 mrg /* rdtscp */
101 1.1 mrg extern __inline unsigned long long
102 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
103 1.1 mrg __rdtscp (unsigned int *__A)
104 1.1 mrg {
105 1.1 mrg return __builtin_ia32_rdtscp (__A);
106 1.1 mrg }
107 1.1 mrg
108 1.1 mrg /* 8bit rol */
109 1.1 mrg extern __inline unsigned char
110 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
111 1.1 mrg __rolb (unsigned char __X, int __C)
112 1.1 mrg {
113 1.1 mrg return __builtin_ia32_rolqi (__X, __C);
114 1.1 mrg }
115 1.1 mrg
116 1.1 mrg /* 16bit rol */
117 1.1 mrg extern __inline unsigned short
118 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
119 1.1 mrg __rolw (unsigned short __X, int __C)
120 1.1 mrg {
121 1.1 mrg return __builtin_ia32_rolhi (__X, __C);
122 1.1 mrg }
123 1.1 mrg
124 1.1 mrg /* 32bit rol */
125 1.1 mrg extern __inline unsigned int
126 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
127 1.1 mrg __rold (unsigned int __X, int __C)
128 1.1 mrg {
129 1.1 mrg return (__X << __C) | (__X >> (32 - __C));
130 1.1 mrg }
131 1.1 mrg
132 1.1 mrg /* 8bit ror */
133 1.1 mrg extern __inline unsigned char
134 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
135 1.1 mrg __rorb (unsigned char __X, int __C)
136 1.1 mrg {
137 1.1 mrg return __builtin_ia32_rorqi (__X, __C);
138 1.1 mrg }
139 1.1 mrg
140 1.1 mrg /* 16bit ror */
141 1.1 mrg extern __inline unsigned short
142 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
143 1.1 mrg __rorw (unsigned short __X, int __C)
144 1.1 mrg {
145 1.1 mrg return __builtin_ia32_rorhi (__X, __C);
146 1.1 mrg }
147 1.1 mrg
148 1.1 mrg /* 32bit ror */
149 1.1 mrg extern __inline unsigned int
150 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
151 1.1 mrg __rord (unsigned int __X, int __C)
152 1.1 mrg {
153 1.1 mrg return (__X >> __C) | (__X << (32 - __C));
154 1.1 mrg }
155 1.1 mrg
156 1.3 mrg /* Pause */
157 1.3 mrg extern __inline void
158 1.3 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
159 1.3 mrg __pause (void)
160 1.3 mrg {
161 1.3 mrg __builtin_ia32_pause ();
162 1.3 mrg }
163 1.3 mrg
164 1.1 mrg #ifdef __x86_64__
165 1.1 mrg /* 64bit bsf */
166 1.1 mrg extern __inline int
167 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
168 1.1 mrg __bsfq (long long __X)
169 1.1 mrg {
170 1.1 mrg return __builtin_ctzll (__X);
171 1.1 mrg }
172 1.1 mrg
173 1.1 mrg /* 64bit bsr */
174 1.1 mrg extern __inline int
175 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
176 1.1 mrg __bsrq (long long __X)
177 1.1 mrg {
178 1.1 mrg return __builtin_ia32_bsrdi (__X);
179 1.1 mrg }
180 1.1 mrg
181 1.1 mrg /* 64bit bswap */
182 1.1 mrg extern __inline long long
183 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
184 1.1 mrg __bswapq (long long __X)
185 1.1 mrg {
186 1.1 mrg return __builtin_bswap64 (__X);
187 1.1 mrg }
188 1.1 mrg
189 1.1 mrg #ifdef __SSE4_2__
190 1.1 mrg /* 64bit accumulate CRC32 (polynomial 0x11EDC6F41) value. */
191 1.1 mrg extern __inline unsigned long long
192 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
193 1.1 mrg __crc32q (unsigned long long __C, unsigned long long __V)
194 1.1 mrg {
195 1.1 mrg return __builtin_ia32_crc32di (__C, __V);
196 1.1 mrg }
197 1.1 mrg #endif
198 1.1 mrg
199 1.1 mrg /* 64bit popcnt */
200 1.1 mrg extern __inline long long
201 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
202 1.1 mrg __popcntq (unsigned long long __X)
203 1.1 mrg {
204 1.1 mrg return __builtin_popcountll (__X);
205 1.1 mrg }
206 1.1 mrg
207 1.1 mrg /* 64bit rol */
208 1.1 mrg extern __inline unsigned long long
209 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
210 1.1 mrg __rolq (unsigned long long __X, int __C)
211 1.1 mrg {
212 1.1 mrg return (__X << __C) | (__X >> (64 - __C));
213 1.1 mrg }
214 1.1 mrg
215 1.1 mrg /* 64bit ror */
216 1.1 mrg extern __inline unsigned long long
217 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
218 1.1 mrg __rorq (unsigned long long __X, int __C)
219 1.1 mrg {
220 1.1 mrg return (__X >> __C) | (__X << (64 - __C));
221 1.1 mrg }
222 1.1 mrg
223 1.1 mrg #define _bswap64(a) __bswapq(a)
224 1.1 mrg #define _popcnt64(a) __popcntq(a)
225 1.1 mrg #define _lrotl(a,b) __rolq((a), (b))
226 1.1 mrg #define _lrotr(a,b) __rorq((a), (b))
227 1.1 mrg #else
228 1.1 mrg #define _lrotl(a,b) __rold((a), (b))
229 1.1 mrg #define _lrotr(a,b) __rord((a), (b))
230 1.1 mrg #endif
231 1.1 mrg
232 1.1 mrg #define _bit_scan_forward(a) __bsfd(a)
233 1.1 mrg #define _bit_scan_reverse(a) __bsrd(a)
234 1.1 mrg #define _bswap(a) __bswapd(a)
235 1.1 mrg #define _popcnt32(a) __popcntd(a)
236 1.1 mrg #define _rdpmc(a) __rdpmc(a)
237 1.1 mrg #define _rdtsc() __rdtsc()
238 1.1 mrg #define _rdtscp(a) __rdtscp(a)
239 1.1 mrg #define _rotwl(a,b) __rolw((a), (b))
240 1.1 mrg #define _rotwr(a,b) __rorw((a), (b))
241 1.1 mrg #define _rotl(a,b) __rold((a), (b))
242 1.1 mrg #define _rotr(a,b) __rord((a), (b))
243