1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2004-2005 David Schultz <das@FreeBSD.ORG> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #ifndef _FENV_H_ 30 #define _FENV_H_ 31 32 #include <sys/cdefs.h> 33 #include <sys/_types.h> 34 #include <ieeefp.h> 35 36 #ifndef __fenv_static 37 #define __fenv_static static 38 #endif 39 40 typedef __uint16_t fexcept_t; 41 42 /* Exception flags */ 43 #define FE_INVALID 0x01 44 #define FE_DENORMAL 0x02 45 #define FE_DIVBYZERO 0x04 46 #define FE_OVERFLOW 0x08 47 #define FE_UNDERFLOW 0x10 48 #define FE_INEXACT 0x20 49 #define FE_ALL_EXCEPT (FE_DIVBYZERO | FE_DENORMAL | FE_INEXACT | \ 50 FE_INVALID | FE_OVERFLOW | FE_UNDERFLOW) 51 52 /* Rounding modes */ 53 #define FE_TONEAREST 0x0000 54 #define FE_DOWNWARD 0x0400 55 #define FE_UPWARD 0x0800 56 #define FE_TOWARDZERO 0x0c00 57 #define _ROUND_MASK (FE_TONEAREST | FE_DOWNWARD | \ 58 FE_UPWARD | FE_TOWARDZERO) 59 60 /* 61 * As compared to the x87 control word, the SSE unit's control word 62 * has the rounding control bits offset by 3 and the exception mask 63 * bits offset by 7. 64 */ 65 #define _SSE_ROUND_SHIFT 3 66 #define _SSE_EMASK_SHIFT 7 67 68 #ifdef __i386__ 69 /* 70 * To preserve binary compatibility with FreeBSD 5.3, we pack the 71 * mxcsr into some reserved fields, rather than changing sizeof(fenv_t). 72 */ 73 typedef struct { 74 __uint16_t __control; 75 __uint16_t __mxcsr_hi; 76 __uint16_t __status; 77 __uint16_t __mxcsr_lo; 78 __uint32_t __tag; 79 char __other[16]; 80 } fenv_t; 81 #else /* __amd64__ */ 82 typedef struct { 83 struct { 84 __uint32_t __control; 85 __uint32_t __status; 86 __uint32_t __tag; 87 char __other[16]; 88 } __x87; 89 __uint32_t __mxcsr; 90 } fenv_t; 91 #endif /* __i386__ */ 92 93 /* 94 * Control modes: x87 control word plus MXCSR bits other than the 95 * exception flags. fesetmode() restores those bits without changing 96 * the raised exception flags. 97 */ 98 typedef struct { 99 __uint16_t __control; 100 __uint16_t __reserved; 101 __uint32_t __mxcsr; 102 } femode_t; 103 104 __BEGIN_DECLS 105 106 /* Default floating-point environment */ 107 extern const fenv_t __fe_dfl_env; 108 #define FE_DFL_ENV (&__fe_dfl_env) 109 110 /* Default floating-point control modes */ 111 extern const femode_t __fe_dfl_mode; 112 #define FE_DFL_MODE (&__fe_dfl_mode) 113 114 #define __fldenvx(__env) __asm __volatile("fldenv %0" : : "m" (__env) \ 115 : "st", "st(1)", "st(2)", "st(3)", "st(4)", \ 116 "st(5)", "st(6)", "st(7)") 117 #define __fwait() __asm __volatile("fwait") 118 119 int fegetenv(fenv_t *__envp); 120 int feholdexcept(fenv_t *__envp); 121 int fesetexceptflag(const fexcept_t *__flagp, int __excepts); 122 int feraiseexcept(int __excepts); 123 int feupdateenv(const fenv_t *__envp); 124 125 __fenv_static inline int 126 __fegetround_int(void) 127 { 128 __uint16_t __control; 129 130 /* 131 * We assume that the x87 and the SSE unit agree on the 132 * rounding mode. Reading the control word on the x87 turns 133 * out to be about 5 times faster than reading it on the SSE 134 * unit on an Opteron 244. 135 */ 136 __fnstcw(&__control); 137 return (__control & _ROUND_MASK); 138 } 139 140 #if __BSD_VISIBLE 141 142 int feenableexcept(int __mask); 143 int fedisableexcept(int __mask); 144 145 /* We currently provide no external definition of fegetexcept(). */ 146 static inline int 147 fegetexcept(void) 148 { 149 __uint16_t __control; 150 151 /* 152 * We assume that the masks for the x87 and the SSE unit are 153 * the same. 154 */ 155 __fnstcw(&__control); 156 return (~__control & FE_ALL_EXCEPT); 157 } 158 159 #endif /* __BSD_VISIBLE */ 160 161 int feclearexcept(int); 162 int fegetexceptflag(fexcept_t *, int); 163 int fetestexcept(int); 164 int fesetround(int); 165 int fegetround(void); 166 int fegetmode(femode_t *); 167 int fesetmode(const femode_t *); 168 int fesetenv(const fenv_t *); 169 170 /* 171 * C permits a standard library function to also be exposed as a function-like 172 * macro (C23 7.1.4), and msun uses that here to inline the fast path. C++ 173 * forbids it: <cfenv> imports these names into namespace std (using 174 * ::feclearexcept; etc.), so std::feclearexcept() and friends must denote the 175 * actual functions. Expose the inlining macros to C only; C++ uses the real 176 * extern functions (defined in the matching lib/msun/<arch>/fenv.c). 177 */ 178 #ifndef __cplusplus 179 #define feclearexcept(a) __feclearexcept_int(a) 180 #define fegetexceptflag(e, a) __fegetexceptflag_int(e, a) 181 #define fetestexcept(a) __fetestexcept_int(a) 182 #define fesetround(a) __fesetround_int(a) 183 #define fegetround() __fegetround_int() 184 #define fegetmode(m) __fegetmode_int(m) 185 #define fesetmode(m) __fesetmode_int(m) 186 #define fesetenv(a) __fesetenv_int(a) 187 #endif /* !__cplusplus */ 188 189 #ifdef __i386__ 190 191 /* After testing for SSE support once, we cache the result in __has_sse. */ 192 enum __sse_support { __SSE_YES, __SSE_NO, __SSE_UNK }; 193 extern enum __sse_support __has_sse; 194 int __test_sse(void); 195 #ifdef __SSE__ 196 #define __HAS_SSE() 1 197 #else 198 #define __HAS_SSE() (__has_sse == __SSE_YES || \ 199 (__has_sse == __SSE_UNK && __test_sse())) 200 #endif 201 202 #define __get_mxcsr(env) (((env).__mxcsr_hi << 16) | \ 203 ((env).__mxcsr_lo)) 204 #define __set_mxcsr(env, x) do { \ 205 (env).__mxcsr_hi = (__uint32_t)(x) >> 16; \ 206 (env).__mxcsr_lo = (__uint16_t)(x); \ 207 } while (0) 208 209 __fenv_static inline int 210 __feclearexcept_int(int __excepts) 211 { 212 fenv_t __env; 213 __uint32_t __mxcsr; 214 215 if (__excepts == FE_ALL_EXCEPT) { 216 __fnclex(); 217 } else { 218 __fnstenv(&__env); 219 __env.__status &= ~__excepts; 220 __fldenv(&__env); 221 } 222 if (__HAS_SSE()) { 223 __stmxcsr(&__mxcsr); 224 __mxcsr &= ~__excepts; 225 __ldmxcsr(&__mxcsr); 226 } 227 return (0); 228 } 229 230 __fenv_static inline int 231 __fegetexceptflag_int(fexcept_t *__flagp, int __excepts) 232 { 233 __uint32_t __mxcsr; 234 __uint16_t __status; 235 236 __fnstsw(&__status); 237 if (__HAS_SSE()) 238 __stmxcsr(&__mxcsr); 239 else 240 __mxcsr = 0; 241 *__flagp = (__mxcsr | __status) & __excepts; 242 return (0); 243 } 244 245 __fenv_static inline int 246 __fetestexcept_int(int __excepts) 247 { 248 __uint32_t __mxcsr; 249 __uint16_t __status; 250 251 __fnstsw(&__status); 252 if (__HAS_SSE()) 253 __stmxcsr(&__mxcsr); 254 else 255 __mxcsr = 0; 256 return ((__status | __mxcsr) & __excepts); 257 } 258 259 __fenv_static inline int 260 __fesetround_int(int __round) 261 { 262 __uint32_t __mxcsr; 263 __uint16_t __control; 264 265 if (__round & ~_ROUND_MASK) 266 return (-1); 267 268 __fnstcw(&__control); 269 __control &= ~_ROUND_MASK; 270 __control |= __round; 271 __fldcw(&__control); 272 273 if (__HAS_SSE()) { 274 __stmxcsr(&__mxcsr); 275 __mxcsr &= ~(_ROUND_MASK << _SSE_ROUND_SHIFT); 276 __mxcsr |= __round << _SSE_ROUND_SHIFT; 277 __ldmxcsr(&__mxcsr); 278 } 279 280 return (0); 281 } 282 283 __fenv_static inline int 284 __fegetmode_int(femode_t *__modep) 285 { 286 287 __fnstcw(&__modep->__control); 288 __modep->__reserved = 0; 289 if (__HAS_SSE()) { 290 __stmxcsr(&__modep->__mxcsr); 291 __modep->__mxcsr &= ~FE_ALL_EXCEPT; 292 } else 293 __modep->__mxcsr = 0; 294 295 return (0); 296 } 297 298 __fenv_static inline int 299 __fesetmode_int(const femode_t *__modep) 300 { 301 __uint32_t __mxcsr; 302 303 __fldcw(&__modep->__control); 304 305 if (__HAS_SSE()) { 306 __stmxcsr(&__mxcsr); 307 __mxcsr &= FE_ALL_EXCEPT; 308 __mxcsr |= __modep->__mxcsr & ~FE_ALL_EXCEPT; 309 __ldmxcsr(&__mxcsr); 310 } 311 312 return (0); 313 } 314 315 __fenv_static inline int 316 __fesetenv_int(const fenv_t *__envp) 317 { 318 fenv_t __env = *__envp; 319 __uint32_t __mxcsr; 320 321 __mxcsr = __get_mxcsr(__env); 322 __set_mxcsr(__env, 0xffffffff); 323 /* 324 * XXX Using fldenvx() instead of fldenv() tells the compiler that this 325 * instruction clobbers the i387 register stack. This happens because 326 * we restore the tag word from the saved environment. Normally, this 327 * would happen anyway and we wouldn't care, because the ABI allows 328 * function calls to clobber the i387 regs. However, fesetenv() is 329 * inlined, so we need to be more careful. 330 */ 331 __fldenvx(__env); 332 if (__HAS_SSE()) 333 __ldmxcsr(&__mxcsr); 334 return (0); 335 } 336 337 #else /* __amd64__ */ 338 339 __fenv_static inline int 340 __feclearexcept_int(int __excepts) 341 { 342 fenv_t __env; 343 344 if (__excepts == FE_ALL_EXCEPT) { 345 __fnclex(); 346 } else { 347 __fnstenv(&__env.__x87); 348 __env.__x87.__status &= ~__excepts; 349 __fldenv(&__env.__x87); 350 } 351 __stmxcsr(&__env.__mxcsr); 352 __env.__mxcsr &= ~__excepts; 353 __ldmxcsr(&__env.__mxcsr); 354 return (0); 355 } 356 357 __fenv_static inline int 358 __fegetexceptflag_int(fexcept_t *__flagp, int __excepts) 359 { 360 __uint32_t __mxcsr; 361 __uint16_t __status; 362 363 __stmxcsr(&__mxcsr); 364 __fnstsw(&__status); 365 *__flagp = (__mxcsr | __status) & __excepts; 366 return (0); 367 } 368 369 __fenv_static inline int 370 __fetestexcept_int(int __excepts) 371 { 372 __uint32_t __mxcsr; 373 __uint16_t __status; 374 375 __stmxcsr(&__mxcsr); 376 __fnstsw(&__status); 377 return ((__status | __mxcsr) & __excepts); 378 } 379 380 __fenv_static inline int 381 __fesetround_int(int __round) 382 { 383 __uint32_t __mxcsr; 384 __uint16_t __control; 385 386 if (__round & ~_ROUND_MASK) 387 return (-1); 388 389 __fnstcw(&__control); 390 __control &= ~_ROUND_MASK; 391 __control |= __round; 392 __fldcw(&__control); 393 394 __stmxcsr(&__mxcsr); 395 __mxcsr &= ~(_ROUND_MASK << _SSE_ROUND_SHIFT); 396 __mxcsr |= __round << _SSE_ROUND_SHIFT; 397 __ldmxcsr(&__mxcsr); 398 399 return (0); 400 } 401 402 __fenv_static inline int 403 __fegetmode_int(femode_t *__modep) 404 { 405 406 __fnstcw(&__modep->__control); 407 __modep->__reserved = 0; 408 __stmxcsr(&__modep->__mxcsr); 409 __modep->__mxcsr &= ~FE_ALL_EXCEPT; 410 411 return (0); 412 } 413 414 __fenv_static inline int 415 __fesetmode_int(const femode_t *__modep) 416 { 417 __uint32_t __mxcsr; 418 419 __fldcw(&__modep->__control); 420 421 __stmxcsr(&__mxcsr); 422 __mxcsr &= FE_ALL_EXCEPT; 423 __mxcsr |= __modep->__mxcsr & ~FE_ALL_EXCEPT; 424 __ldmxcsr(&__mxcsr); 425 426 return (0); 427 } 428 429 __fenv_static inline int 430 __fesetenv_int(const fenv_t *__envp) 431 { 432 433 /* 434 * XXX Using fldenvx() instead of fldenv() tells the compiler that this 435 * instruction clobbers the i387 register stack. This happens because 436 * we restore the tag word from the saved environment. Normally, this 437 * would happen anyway and we wouldn't care, because the ABI allows 438 * function calls to clobber the i387 regs. However, fesetenv() is 439 * inlined, so we need to be more careful. 440 */ 441 __fldenvx(__envp->__x87); 442 __ldmxcsr(&__envp->__mxcsr); 443 return (0); 444 } 445 446 #endif /* __i386__ */ 447 448 __END_DECLS 449 450 #endif /* !_FENV_H_ */ 451