ia32intrin.h revision 1.5 1 1.5 mrg /* Copyright (C) 2009-2015 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 1.5 mrg #ifndef __SSE4_2__
53 1.5 mrg #pragma GCC push_options
54 1.5 mrg #pragma GCC target("sse4.2")
55 1.5 mrg #define __DISABLE_SSE4_2__
56 1.5 mrg #endif /* __SSE4_2__ */
57 1.5 mrg
58 1.1 mrg /* 32bit accumulate CRC32 (polynomial 0x11EDC6F41) value. */
59 1.1 mrg extern __inline unsigned int
60 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
61 1.1 mrg __crc32b (unsigned int __C, unsigned char __V)
62 1.1 mrg {
63 1.1 mrg return __builtin_ia32_crc32qi (__C, __V);
64 1.1 mrg }
65 1.1 mrg
66 1.1 mrg extern __inline unsigned int
67 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
68 1.1 mrg __crc32w (unsigned int __C, unsigned short __V)
69 1.1 mrg {
70 1.1 mrg return __builtin_ia32_crc32hi (__C, __V);
71 1.1 mrg }
72 1.1 mrg
73 1.1 mrg extern __inline unsigned int
74 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
75 1.1 mrg __crc32d (unsigned int __C, unsigned int __V)
76 1.1 mrg {
77 1.1 mrg return __builtin_ia32_crc32si (__C, __V);
78 1.1 mrg }
79 1.5 mrg
80 1.5 mrg #ifdef __DISABLE_SSE4_2__
81 1.5 mrg #undef __DISABLE_SSE4_2__
82 1.5 mrg #pragma GCC pop_options
83 1.5 mrg #endif /* __DISABLE_SSE4_2__ */
84 1.1 mrg
85 1.1 mrg /* 32bit popcnt */
86 1.1 mrg extern __inline int
87 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
88 1.1 mrg __popcntd (unsigned int __X)
89 1.1 mrg {
90 1.1 mrg return __builtin_popcount (__X);
91 1.1 mrg }
92 1.1 mrg
93 1.1 mrg /* rdpmc */
94 1.1 mrg extern __inline unsigned long long
95 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
96 1.1 mrg __rdpmc (int __S)
97 1.1 mrg {
98 1.1 mrg return __builtin_ia32_rdpmc (__S);
99 1.1 mrg }
100 1.1 mrg
101 1.1 mrg /* rdtsc */
102 1.1 mrg extern __inline unsigned long long
103 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
104 1.1 mrg __rdtsc (void)
105 1.1 mrg {
106 1.1 mrg return __builtin_ia32_rdtsc ();
107 1.1 mrg }
108 1.1 mrg
109 1.1 mrg /* rdtscp */
110 1.1 mrg extern __inline unsigned long long
111 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
112 1.1 mrg __rdtscp (unsigned int *__A)
113 1.1 mrg {
114 1.1 mrg return __builtin_ia32_rdtscp (__A);
115 1.1 mrg }
116 1.1 mrg
117 1.1 mrg /* 8bit rol */
118 1.1 mrg extern __inline unsigned char
119 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
120 1.1 mrg __rolb (unsigned char __X, int __C)
121 1.1 mrg {
122 1.1 mrg return __builtin_ia32_rolqi (__X, __C);
123 1.1 mrg }
124 1.1 mrg
125 1.1 mrg /* 16bit rol */
126 1.1 mrg extern __inline unsigned short
127 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
128 1.1 mrg __rolw (unsigned short __X, int __C)
129 1.1 mrg {
130 1.1 mrg return __builtin_ia32_rolhi (__X, __C);
131 1.1 mrg }
132 1.1 mrg
133 1.1 mrg /* 32bit rol */
134 1.1 mrg extern __inline unsigned int
135 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
136 1.1 mrg __rold (unsigned int __X, int __C)
137 1.1 mrg {
138 1.1 mrg return (__X << __C) | (__X >> (32 - __C));
139 1.1 mrg }
140 1.1 mrg
141 1.1 mrg /* 8bit ror */
142 1.1 mrg extern __inline unsigned char
143 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
144 1.1 mrg __rorb (unsigned char __X, int __C)
145 1.1 mrg {
146 1.1 mrg return __builtin_ia32_rorqi (__X, __C);
147 1.1 mrg }
148 1.1 mrg
149 1.1 mrg /* 16bit ror */
150 1.1 mrg extern __inline unsigned short
151 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
152 1.1 mrg __rorw (unsigned short __X, int __C)
153 1.1 mrg {
154 1.1 mrg return __builtin_ia32_rorhi (__X, __C);
155 1.1 mrg }
156 1.1 mrg
157 1.1 mrg /* 32bit ror */
158 1.1 mrg extern __inline unsigned int
159 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
160 1.1 mrg __rord (unsigned int __X, int __C)
161 1.1 mrg {
162 1.1 mrg return (__X >> __C) | (__X << (32 - __C));
163 1.1 mrg }
164 1.1 mrg
165 1.3 mrg /* Pause */
166 1.3 mrg extern __inline void
167 1.3 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
168 1.3 mrg __pause (void)
169 1.3 mrg {
170 1.3 mrg __builtin_ia32_pause ();
171 1.3 mrg }
172 1.3 mrg
173 1.1 mrg #ifdef __x86_64__
174 1.1 mrg /* 64bit bsf */
175 1.1 mrg extern __inline int
176 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
177 1.1 mrg __bsfq (long long __X)
178 1.1 mrg {
179 1.1 mrg return __builtin_ctzll (__X);
180 1.1 mrg }
181 1.1 mrg
182 1.1 mrg /* 64bit bsr */
183 1.1 mrg extern __inline int
184 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
185 1.1 mrg __bsrq (long long __X)
186 1.1 mrg {
187 1.1 mrg return __builtin_ia32_bsrdi (__X);
188 1.1 mrg }
189 1.1 mrg
190 1.1 mrg /* 64bit bswap */
191 1.1 mrg extern __inline long long
192 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
193 1.1 mrg __bswapq (long long __X)
194 1.1 mrg {
195 1.1 mrg return __builtin_bswap64 (__X);
196 1.1 mrg }
197 1.1 mrg
198 1.5 mrg #ifndef __SSE4_2__
199 1.5 mrg #pragma GCC push_options
200 1.5 mrg #pragma GCC target("sse4.2")
201 1.5 mrg #define __DISABLE_SSE4_2__
202 1.5 mrg #endif /* __SSE4_2__ */
203 1.5 mrg
204 1.1 mrg /* 64bit accumulate CRC32 (polynomial 0x11EDC6F41) value. */
205 1.1 mrg extern __inline unsigned long long
206 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
207 1.1 mrg __crc32q (unsigned long long __C, unsigned long long __V)
208 1.1 mrg {
209 1.1 mrg return __builtin_ia32_crc32di (__C, __V);
210 1.1 mrg }
211 1.5 mrg
212 1.5 mrg #ifdef __DISABLE_SSE4_2__
213 1.5 mrg #undef __DISABLE_SSE4_2__
214 1.5 mrg #pragma GCC pop_options
215 1.5 mrg #endif /* __DISABLE_SSE4_2__ */
216 1.1 mrg
217 1.1 mrg /* 64bit popcnt */
218 1.1 mrg extern __inline long long
219 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
220 1.1 mrg __popcntq (unsigned long long __X)
221 1.1 mrg {
222 1.1 mrg return __builtin_popcountll (__X);
223 1.1 mrg }
224 1.1 mrg
225 1.1 mrg /* 64bit rol */
226 1.1 mrg extern __inline unsigned long long
227 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
228 1.1 mrg __rolq (unsigned long long __X, int __C)
229 1.1 mrg {
230 1.1 mrg return (__X << __C) | (__X >> (64 - __C));
231 1.1 mrg }
232 1.1 mrg
233 1.1 mrg /* 64bit ror */
234 1.1 mrg extern __inline unsigned long long
235 1.1 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
236 1.1 mrg __rorq (unsigned long long __X, int __C)
237 1.1 mrg {
238 1.1 mrg return (__X >> __C) | (__X << (64 - __C));
239 1.1 mrg }
240 1.1 mrg
241 1.5 mrg /* Read flags register */
242 1.5 mrg extern __inline unsigned long long
243 1.5 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
244 1.5 mrg __readeflags (void)
245 1.5 mrg {
246 1.5 mrg return __builtin_ia32_readeflags_u64 ();
247 1.5 mrg }
248 1.5 mrg
249 1.5 mrg /* Write flags register */
250 1.5 mrg extern __inline void
251 1.5 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
252 1.5 mrg __writeeflags (unsigned long long X)
253 1.5 mrg {
254 1.5 mrg __builtin_ia32_writeeflags_u64 (X);
255 1.5 mrg }
256 1.5 mrg
257 1.1 mrg #define _bswap64(a) __bswapq(a)
258 1.1 mrg #define _popcnt64(a) __popcntq(a)
259 1.5 mrg #else
260 1.5 mrg
261 1.5 mrg /* Read flags register */
262 1.5 mrg extern __inline unsigned int
263 1.5 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
264 1.5 mrg __readeflags (void)
265 1.5 mrg {
266 1.5 mrg return __builtin_ia32_readeflags_u32 ();
267 1.5 mrg }
268 1.5 mrg
269 1.5 mrg /* Write flags register */
270 1.5 mrg extern __inline void
271 1.5 mrg __attribute__((__gnu_inline__, __always_inline__, __artificial__))
272 1.5 mrg __writeeflags (unsigned int X)
273 1.5 mrg {
274 1.5 mrg __builtin_ia32_writeeflags_u32 (X);
275 1.5 mrg }
276 1.5 mrg
277 1.5 mrg #endif
278 1.5 mrg
279 1.5 mrg /* On LP64 systems, longs are 64-bit. Use the appropriate rotate
280 1.5 mrg * function. */
281 1.5 mrg #ifdef __LP64__
282 1.1 mrg #define _lrotl(a,b) __rolq((a), (b))
283 1.1 mrg #define _lrotr(a,b) __rorq((a), (b))
284 1.1 mrg #else
285 1.1 mrg #define _lrotl(a,b) __rold((a), (b))
286 1.1 mrg #define _lrotr(a,b) __rord((a), (b))
287 1.1 mrg #endif
288 1.1 mrg
289 1.1 mrg #define _bit_scan_forward(a) __bsfd(a)
290 1.1 mrg #define _bit_scan_reverse(a) __bsrd(a)
291 1.1 mrg #define _bswap(a) __bswapd(a)
292 1.1 mrg #define _popcnt32(a) __popcntd(a)
293 1.1 mrg #define _rdpmc(a) __rdpmc(a)
294 1.1 mrg #define _rdtsc() __rdtsc()
295 1.1 mrg #define _rdtscp(a) __rdtscp(a)
296 1.1 mrg #define _rotwl(a,b) __rolw((a), (b))
297 1.1 mrg #define _rotwr(a,b) __rorw((a), (b))
298 1.1 mrg #define _rotl(a,b) __rold((a), (b))
299 1.1 mrg #define _rotr(a,b) __rord((a), (b))
300