| 1 |
/* Copyright (C) 2009-2015 Free Software Foundation, Inc. |
| 2 |
|
| 3 |
This file is part of GCC. |
| 4 |
|
| 5 |
GCC is free software; you can redistribute it and/or modify |
| 6 |
it under the terms of the GNU General Public License as published by |
| 7 |
the Free Software Foundation; either version 3, or (at your option) |
| 8 |
any later version. |
| 9 |
|
| 10 |
GCC is distributed in the hope that it will be useful, |
| 11 |
but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 13 |
GNU General Public License for more details. |
| 14 |
|
| 15 |
Under Section 7 of GPL version 3, you are granted additional |
| 16 |
permissions described in the GCC Runtime Library Exception, version |
| 17 |
3.1, as published by the Free Software Foundation. |
| 18 |
|
| 19 |
You should have received a copy of the GNU General Public License and |
| 20 |
a copy of the GCC Runtime Library Exception along with this program; |
| 21 |
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see |
| 22 |
<http://www.gnu.org/licenses/>. */ |
| 23 |
|
| 24 |
#ifndef _X86INTRIN_H_INCLUDED |
| 25 |
# error "Never use <ia32intrin.h> directly; include <x86intrin.h> instead." |
| 26 |
#endif |
| 27 |
|
| 28 |
/* 32bit bsf */ |
| 29 |
extern __inline int |
| 30 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 31 |
__bsfd (int __X) |
| 32 |
{ |
| 33 |
return __builtin_ctz (__X); |
| 34 |
} |
| 35 |
|
| 36 |
/* 32bit bsr */ |
| 37 |
extern __inline int |
| 38 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 39 |
__bsrd (int __X) |
| 40 |
{ |
| 41 |
return __builtin_ia32_bsrsi (__X); |
| 42 |
} |
| 43 |
|
| 44 |
/* 32bit bswap */ |
| 45 |
extern __inline int |
| 46 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 47 |
__bswapd (int __X) |
| 48 |
{ |
| 49 |
return __builtin_bswap32 (__X); |
| 50 |
} |
| 51 |
|
| 52 |
#ifndef __SSE4_2__ |
| 53 |
#pragma GCC push_options |
| 54 |
#pragma GCC target("sse4.2") |
| 55 |
#define __DISABLE_SSE4_2__ |
| 56 |
#endif /* __SSE4_2__ */ |
| 57 |
|
| 58 |
/* 32bit accumulate CRC32 (polynomial 0x11EDC6F41) value. */ |
| 59 |
extern __inline unsigned int |
| 60 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 61 |
__crc32b (unsigned int __C, unsigned char __V) |
| 62 |
{ |
| 63 |
return __builtin_ia32_crc32qi (__C, __V); |
| 64 |
} |
| 65 |
|
| 66 |
extern __inline unsigned int |
| 67 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 68 |
__crc32w (unsigned int __C, unsigned short __V) |
| 69 |
{ |
| 70 |
return __builtin_ia32_crc32hi (__C, __V); |
| 71 |
} |
| 72 |
|
| 73 |
extern __inline unsigned int |
| 74 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 75 |
__crc32d (unsigned int __C, unsigned int __V) |
| 76 |
{ |
| 77 |
return __builtin_ia32_crc32si (__C, __V); |
| 78 |
} |
| 79 |
|
| 80 |
#ifdef __DISABLE_SSE4_2__ |
| 81 |
#undef __DISABLE_SSE4_2__ |
| 82 |
#pragma GCC pop_options |
| 83 |
#endif /* __DISABLE_SSE4_2__ */ |
| 84 |
|
| 85 |
/* 32bit popcnt */ |
| 86 |
extern __inline int |
| 87 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 88 |
__popcntd (unsigned int __X) |
| 89 |
{ |
| 90 |
return __builtin_popcount (__X); |
| 91 |
} |
| 92 |
|
| 93 |
/* rdpmc */ |
| 94 |
extern __inline unsigned long long |
| 95 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 96 |
__rdpmc (int __S) |
| 97 |
{ |
| 98 |
return __builtin_ia32_rdpmc (__S); |
| 99 |
} |
| 100 |
|
| 101 |
/* rdtsc */ |
| 102 |
extern __inline unsigned long long |
| 103 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 104 |
__rdtsc (void) |
| 105 |
{ |
| 106 |
return __builtin_ia32_rdtsc (); |
| 107 |
} |
| 108 |
|
| 109 |
/* rdtscp */ |
| 110 |
extern __inline unsigned long long |
| 111 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 112 |
__rdtscp (unsigned int *__A) |
| 113 |
{ |
| 114 |
return __builtin_ia32_rdtscp (__A); |
| 115 |
} |
| 116 |
|
| 117 |
/* 8bit rol */ |
| 118 |
extern __inline unsigned char |
| 119 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 120 |
__rolb (unsigned char __X, int __C) |
| 121 |
{ |
| 122 |
return __builtin_ia32_rolqi (__X, __C); |
| 123 |
} |
| 124 |
|
| 125 |
/* 16bit rol */ |
| 126 |
extern __inline unsigned short |
| 127 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 128 |
__rolw (unsigned short __X, int __C) |
| 129 |
{ |
| 130 |
return __builtin_ia32_rolhi (__X, __C); |
| 131 |
} |
| 132 |
|
| 133 |
/* 32bit rol */ |
| 134 |
extern __inline unsigned int |
| 135 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 136 |
__rold (unsigned int __X, int __C) |
| 137 |
{ |
| 138 |
return (__X << __C) | (__X >> (32 - __C)); |
| 139 |
} |
| 140 |
|
| 141 |
/* 8bit ror */ |
| 142 |
extern __inline unsigned char |
| 143 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 144 |
__rorb (unsigned char __X, int __C) |
| 145 |
{ |
| 146 |
return __builtin_ia32_rorqi (__X, __C); |
| 147 |
} |
| 148 |
|
| 149 |
/* 16bit ror */ |
| 150 |
extern __inline unsigned short |
| 151 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 152 |
__rorw (unsigned short __X, int __C) |
| 153 |
{ |
| 154 |
return __builtin_ia32_rorhi (__X, __C); |
| 155 |
} |
| 156 |
|
| 157 |
/* 32bit ror */ |
| 158 |
extern __inline unsigned int |
| 159 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 160 |
__rord (unsigned int __X, int __C) |
| 161 |
{ |
| 162 |
return (__X >> __C) | (__X << (32 - __C)); |
| 163 |
} |
| 164 |
|
| 165 |
/* Pause */ |
| 166 |
extern __inline void |
| 167 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 168 |
__pause (void) |
| 169 |
{ |
| 170 |
__builtin_ia32_pause (); |
| 171 |
} |
| 172 |
|
| 173 |
#ifdef __x86_64__ |
| 174 |
/* 64bit bsf */ |
| 175 |
extern __inline int |
| 176 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 177 |
__bsfq (long long __X) |
| 178 |
{ |
| 179 |
return __builtin_ctzll (__X); |
| 180 |
} |
| 181 |
|
| 182 |
/* 64bit bsr */ |
| 183 |
extern __inline int |
| 184 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 185 |
__bsrq (long long __X) |
| 186 |
{ |
| 187 |
return __builtin_ia32_bsrdi (__X); |
| 188 |
} |
| 189 |
|
| 190 |
/* 64bit bswap */ |
| 191 |
extern __inline long long |
| 192 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 193 |
__bswapq (long long __X) |
| 194 |
{ |
| 195 |
return __builtin_bswap64 (__X); |
| 196 |
} |
| 197 |
|
| 198 |
#ifndef __SSE4_2__ |
| 199 |
#pragma GCC push_options |
| 200 |
#pragma GCC target("sse4.2") |
| 201 |
#define __DISABLE_SSE4_2__ |
| 202 |
#endif /* __SSE4_2__ */ |
| 203 |
|
| 204 |
/* 64bit accumulate CRC32 (polynomial 0x11EDC6F41) value. */ |
| 205 |
extern __inline unsigned long long |
| 206 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 207 |
__crc32q (unsigned long long __C, unsigned long long __V) |
| 208 |
{ |
| 209 |
return __builtin_ia32_crc32di (__C, __V); |
| 210 |
} |
| 211 |
|
| 212 |
#ifdef __DISABLE_SSE4_2__ |
| 213 |
#undef __DISABLE_SSE4_2__ |
| 214 |
#pragma GCC pop_options |
| 215 |
#endif /* __DISABLE_SSE4_2__ */ |
| 216 |
|
| 217 |
/* 64bit popcnt */ |
| 218 |
extern __inline long long |
| 219 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 220 |
__popcntq (unsigned long long __X) |
| 221 |
{ |
| 222 |
return __builtin_popcountll (__X); |
| 223 |
} |
| 224 |
|
| 225 |
/* 64bit rol */ |
| 226 |
extern __inline unsigned long long |
| 227 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 228 |
__rolq (unsigned long long __X, int __C) |
| 229 |
{ |
| 230 |
return (__X << __C) | (__X >> (64 - __C)); |
| 231 |
} |
| 232 |
|
| 233 |
/* 64bit ror */ |
| 234 |
extern __inline unsigned long long |
| 235 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 236 |
__rorq (unsigned long long __X, int __C) |
| 237 |
{ |
| 238 |
return (__X >> __C) | (__X << (64 - __C)); |
| 239 |
} |
| 240 |
|
| 241 |
/* Read flags register */ |
| 242 |
extern __inline unsigned long long |
| 243 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 244 |
__readeflags (void) |
| 245 |
{ |
| 246 |
return __builtin_ia32_readeflags_u64 (); |
| 247 |
} |
| 248 |
|
| 249 |
/* Write flags register */ |
| 250 |
extern __inline void |
| 251 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 252 |
__writeeflags (unsigned long long X) |
| 253 |
{ |
| 254 |
__builtin_ia32_writeeflags_u64 (X); |
| 255 |
} |
| 256 |
|
| 257 |
#define _bswap64(a) __bswapq(a) |
| 258 |
#define _popcnt64(a) __popcntq(a) |
| 259 |
#else |
| 260 |
|
| 261 |
/* Read flags register */ |
| 262 |
extern __inline unsigned int |
| 263 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 264 |
__readeflags (void) |
| 265 |
{ |
| 266 |
return __builtin_ia32_readeflags_u32 (); |
| 267 |
} |
| 268 |
|
| 269 |
/* Write flags register */ |
| 270 |
extern __inline void |
| 271 |
__attribute__((__gnu_inline__, __always_inline__, __artificial__)) |
| 272 |
__writeeflags (unsigned int X) |
| 273 |
{ |
| 274 |
__builtin_ia32_writeeflags_u32 (X); |
| 275 |
} |
| 276 |
|
| 277 |
#endif |
| 278 |
|
| 279 |
/* On LP64 systems, longs are 64-bit. Use the appropriate rotate |
| 280 |
* function. */ |
| 281 |
#ifdef __LP64__ |
| 282 |
#define _lrotl(a,b) __rolq((a), (b)) |
| 283 |
#define _lrotr(a,b) __rorq((a), (b)) |
| 284 |
#else |
| 285 |
#define _lrotl(a,b) __rold((a), (b)) |
| 286 |
#define _lrotr(a,b) __rord((a), (b)) |
| 287 |
#endif |
| 288 |
|
| 289 |
#define _bit_scan_forward(a) __bsfd(a) |
| 290 |
#define _bit_scan_reverse(a) __bsrd(a) |
| 291 |
#define _bswap(a) __bswapd(a) |
| 292 |
#define _popcnt32(a) __popcntd(a) |
| 293 |
#define _rdpmc(a) __rdpmc(a) |
| 294 |
#define _rdtsc() __rdtsc() |
| 295 |
#define _rdtscp(a) __rdtscp(a) |
| 296 |
#define _rotwl(a,b) __rolw((a), (b)) |
| 297 |
#define _rotwr(a,b) __rorw((a), (b)) |
| 298 |
#define _rotl(a,b) __rold((a), (b)) |
| 299 |
#define _rotr(a,b) __rord((a), (b)) |
