1 /* $OpenBSD: fenv.c,v 1.2 2011/04/28 17:34:23 martynas Exp $ */ 2 /* $NetBSD: fenv.c,v 1.1 2010/07/31 21:47:53 joerg Exp $ */ 3 4 /*- 5 * Copyright (c) 2004-2005 David Schultz <das (at) FreeBSD.ORG> 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 #include <machine/fpu.h> 32 33 #include <fenv.h> 34 35 /* 36 * The following constant represents the default floating-point environment 37 * (that is, the one installed at program startup) and has type pointer to 38 * const-qualified fenv_t. 39 * 40 * It can be used as an argument to the functions within the <fenv.h> header 41 * that manage the floating-point environment, namely fesetenv() and 42 * feupdateenv(). 43 * 44 * x87 fpu registers are 16bit wide. The upper bits, 31-16, are marked as 45 * RESERVED. 46 */ 47 fenv_t __fe_dfl_env = { 48 { 49 0xffff0000 | __INITIAL_NPXCW__, /* Control word register */ 50 0xffff0000, /* Status word register */ 51 0xffffffff, /* Tag word register */ 52 { 53 0x00000000, 54 0x00000000, 55 0x00000000, 56 0xffff0000 57 } 58 }, 59 __INITIAL_MXCSR__ /* MXCSR register */ 60 }; 61 62 63 /* 64 * The feclearexcept() function clears the supported floating-point exceptions 65 * represented by `excepts'. 66 */ 67 int 68 feclearexcept(int excepts) 69 { 70 fenv_t fenv; 71 unsigned int mxcsr; 72 73 excepts &= FE_ALL_EXCEPT; 74 75 /* Store the current x87 floating-point environment */ 76 __asm__ __volatile__ ("fnstenv %0" : "=m" (fenv)); 77 78 /* Clear the requested floating-point exceptions */ 79 fenv.__x87.__status &= ~excepts; 80 81 /* Load the x87 floating-point environent */ 82 __asm__ __volatile__ ("fldenv %0" : : "m" (fenv)); 83 84 /* Same for SSE environment */ 85 __asm__ __volatile__ ("stmxcsr %0" : "=m" (mxcsr)); 86 mxcsr &= ~excepts; 87 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (mxcsr)); 88 89 return (0); 90 } 91 92 /* 93 * The fegetexceptflag() function stores an implementation-defined 94 * representation of the states of the floating-point status flags indicated by 95 * the argument excepts in the object pointed to by the argument flagp. 96 */ 97 int 98 fegetexceptflag(fexcept_t *flagp, int excepts) 99 { 100 unsigned short status; 101 unsigned int mxcsr; 102 103 excepts &= FE_ALL_EXCEPT; 104 105 /* Store the current x87 status register */ 106 __asm__ __volatile__ ("fnstsw %0" : "=am" (status)); 107 108 /* Store the MXCSR register */ 109 __asm__ __volatile__ ("stmxcsr %0" : "=m" (mxcsr)); 110 111 /* Store the results in flagp */ 112 *flagp = (status | mxcsr) & excepts; 113 114 return (0); 115 } 116 117 /* 118 * The feraiseexcept() function raises the supported floating-point exceptions 119 * represented by the argument `excepts'. 120 * 121 * The standard explicitly allows us to execute an instruction that has the 122 * exception as a side effect, but we choose to manipulate the status register 123 * directly. 124 * 125 * The validation of input is being deferred to fesetexceptflag(). 126 */ 127 int 128 feraiseexcept(int excepts) 129 { 130 excepts &= FE_ALL_EXCEPT; 131 132 fesetexceptflag((fexcept_t *)&excepts, excepts); 133 __asm__ __volatile__ ("fwait"); 134 135 return (0); 136 } 137 138 /* 139 * This function sets the floating-point status flags indicated by the argument 140 * `excepts' to the states stored in the object pointed to by `flagp'. It does 141 * NOT raise any floating-point exceptions, but only sets the state of the flags. 142 */ 143 int 144 fesetexceptflag(const fexcept_t *flagp, int excepts) 145 { 146 fenv_t fenv; 147 unsigned int mxcsr; 148 149 excepts &= FE_ALL_EXCEPT; 150 151 /* Store the current x87 floating-point environment */ 152 __asm__ __volatile__ ("fnstenv %0" : "=m" (fenv)); 153 154 /* Set the requested status flags */ 155 fenv.__x87.__status &= ~excepts; 156 fenv.__x87.__status |= *flagp & excepts; 157 158 /* Load the x87 floating-point environent */ 159 __asm__ __volatile__ ("fldenv %0" : : "m" (fenv)); 160 161 /* Same for SSE environment */ 162 __asm__ __volatile__ ("stmxcsr %0" : "=m" (mxcsr)); 163 mxcsr &= ~excepts; 164 mxcsr |= *flagp & excepts; 165 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (mxcsr)); 166 167 return (0); 168 } 169 170 /* 171 * The fetestexcept() function determines which of a specified subset of the 172 * floating-point exception flags are currently set. The `excepts' argument 173 * specifies the floating-point status flags to be queried. 174 */ 175 int 176 fetestexcept(int excepts) 177 { 178 unsigned short status; 179 unsigned int mxcsr; 180 181 excepts &= FE_ALL_EXCEPT; 182 183 /* Store the current x87 status register */ 184 __asm__ __volatile__ ("fnstsw %0" : "=am" (status)); 185 186 /* Store the MXCSR register state */ 187 __asm__ __volatile__ ("stmxcsr %0" : "=m" (mxcsr)); 188 189 return ((status | mxcsr) & excepts); 190 } 191 192 /* 193 * The fegetround() function gets the current rounding direction. 194 */ 195 int 196 fegetround(void) 197 { 198 unsigned short control; 199 200 /* 201 * We assume that the x87 and the SSE unit agree on the 202 * rounding mode. Reading the control word on the x87 turns 203 * out to be about 5 times faster than reading it on the SSE 204 * unit on an Opteron 244. 205 */ 206 __asm__ __volatile__ ("fnstcw %0" : "=m" (control)); 207 208 return (control & _X87_ROUND_MASK); 209 } 210 211 /* 212 * The fesetround() function establishes the rounding direction represented by 213 * its argument `round'. If the argument is not equal to the value of a rounding 214 * direction macro, the rounding direction is not changed. 215 */ 216 int 217 fesetround(int round) 218 { 219 unsigned short control; 220 unsigned int mxcsr; 221 222 /* Check whether requested rounding direction is supported */ 223 if (round & ~_X87_ROUND_MASK) 224 return (-1); 225 226 /* Store the current x87 control word register */ 227 __asm__ __volatile__ ("fnstcw %0" : "=m" (control)); 228 229 /* Set the rounding direction */ 230 control &= ~_X87_ROUND_MASK; 231 control |= round; 232 233 /* Load the x87 control word register */ 234 __asm__ __volatile__ ("fldcw %0" : : "m" (control)); 235 236 /* Same for the SSE environment */ 237 __asm__ __volatile__ ("stmxcsr %0" : "=m" (mxcsr)); 238 mxcsr &= ~(_X87_ROUND_MASK << _SSE_ROUND_SHIFT); 239 mxcsr |= round << _SSE_ROUND_SHIFT; 240 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (mxcsr)); 241 242 return (0); 243 } 244 245 /* 246 * The fegetenv() function attempts to store the current floating-point 247 * environment in the object pointed to by envp. 248 */ 249 int 250 fegetenv(fenv_t *envp) 251 { 252 /* Store the current x87 floating-point environment */ 253 __asm__ __volatile__ ("fnstenv %0" : "=m" (*envp)); 254 255 /* Store the MXCSR register state */ 256 __asm__ __volatile__ ("stmxcsr %0" : "=m" (envp->__mxcsr)); 257 258 /* 259 * When an FNSTENV instruction is executed, all pending exceptions are 260 * essentially lost (either the x87 FPU status register is cleared or 261 * all exceptions are masked). 262 * 263 * 8.6 X87 FPU EXCEPTION SYNCHRONIZATION - 264 * Intel(R) 64 and IA-32 Architectures Softare Developer's Manual - Vol1 265 */ 266 __asm__ __volatile__ ("fldcw %0" : : "m" (envp->__x87.__control)); 267 268 return (0); 269 } 270 271 /* 272 * The feholdexcept() function saves the current floating-point environment 273 * in the object pointed to by envp, clears the floating-point status flags, and 274 * then installs a non-stop (continue on floating-point exceptions) mode, if 275 * available, for all floating-point exceptions. 276 */ 277 int 278 feholdexcept(fenv_t *envp) 279 { 280 unsigned int mxcsr; 281 282 /* Store the current x87 floating-point environment */ 283 __asm__ __volatile__ ("fnstenv %0" : "=m" (*envp)); 284 285 /* Clear all exception flags in FPU */ 286 __asm__ __volatile__ ("fnclex"); 287 288 /* Store the MXCSR register state */ 289 __asm__ __volatile__ ("stmxcsr %0" : "=m" (envp->__mxcsr)); 290 291 /* Clear exception flags in MXCSR */ 292 mxcsr = envp->__mxcsr; 293 mxcsr &= ~FE_ALL_EXCEPT; 294 295 /* Mask all exceptions */ 296 mxcsr |= FE_ALL_EXCEPT << _SSE_MASK_SHIFT; 297 298 /* Store the MXCSR register */ 299 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (mxcsr)); 300 301 return (0); 302 } 303 304 /* 305 * The fesetenv() function attempts to establish the floating-point environment 306 * represented by the object pointed to by envp. The argument `envp' points 307 * to an object set by a call to fegetenv() or feholdexcept(), or equal a 308 * floating-point environment macro. The fesetenv() function does not raise 309 * floating-point exceptions, but only installs the state of the floating-point 310 * status flags represented through its argument. 311 */ 312 int 313 fesetenv(const fenv_t *envp) 314 { 315 /* Load the x87 floating-point environent */ 316 __asm__ __volatile__ ("fldenv %0" : : "m" (*envp)); 317 318 /* Store the MXCSR register */ 319 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (envp->__mxcsr)); 320 321 return (0); 322 } 323 324 /* 325 * The feupdateenv() function saves the currently raised floating-point 326 * exceptions in its automatic storage, installs the floating-point environment 327 * represented by the object pointed to by `envp', and then raises the saved 328 * floating-point exceptions. The argument `envp' shall point to an object set 329 * by a call to feholdexcept() or fegetenv(), or equal a floating-point 330 * environment macro. 331 */ 332 int 333 feupdateenv(const fenv_t *envp) 334 { 335 unsigned short status; 336 unsigned int mxcsr; 337 338 /* Store the x87 status register */ 339 __asm__ __volatile__ ("fnstsw %0" : "=am" (status)); 340 341 /* Store the MXCSR register */ 342 __asm__ __volatile__ ("stmxcsr %0" : "=m" (mxcsr)); 343 344 /* Install new floating-point environment */ 345 fesetenv(envp); 346 347 /* Raise any previously accumulated exceptions */ 348 feraiseexcept(status | mxcsr); 349 350 return (0); 351 } 352 353 /* 354 * The following functions are extentions to the standard 355 */ 356 int 357 feenableexcept(int mask) 358 { 359 unsigned int mxcsr, omask; 360 unsigned short control; 361 362 mask &= FE_ALL_EXCEPT; 363 364 __asm__ __volatile__ ("fnstcw %0" : "=m" (control)); 365 __asm__ __volatile__ ("stmxcsr %0" : "=m" (mxcsr)); 366 367 omask = ~(control | (mxcsr >> _SSE_MASK_SHIFT)) & FE_ALL_EXCEPT; 368 control &= ~mask; 369 __asm__ __volatile__ ("fldcw %0" : : "m" (control)); 370 371 mxcsr &= ~(mask << _SSE_MASK_SHIFT); 372 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (mxcsr)); 373 374 return (omask); 375 } 376 377 int 378 fedisableexcept(int mask) 379 { 380 unsigned int mxcsr, omask; 381 unsigned short control; 382 383 mask &= FE_ALL_EXCEPT; 384 385 __asm__ __volatile__ ("fnstcw %0" : "=m" (control)); 386 __asm__ __volatile__ ("stmxcsr %0" : "=m" (mxcsr)); 387 388 omask = ~(control | (mxcsr >> _SSE_MASK_SHIFT)) & FE_ALL_EXCEPT; 389 control |= mask; 390 __asm__ __volatile__ ("fldcw %0" : : "m" (control)); 391 392 mxcsr |= mask << _SSE_MASK_SHIFT; 393 __asm__ __volatile__ ("ldmxcsr %0" : : "m" (mxcsr)); 394 395 return (omask); 396 } 397 398 int 399 fegetexcept(void) 400 { 401 unsigned short control; 402 403 /* 404 * We assume that the masks for the x87 and the SSE unit are 405 * the same. 406 */ 407 __asm__ __volatile__ ("fnstcw %0" : "=m" (control)); 408 409 return (~control & FE_ALL_EXCEPT); 410 } 411