Bjorn Reese | 4502960 | 2001-08-21 09:23:53 +0000 | [diff] [blame] | 1 | /************************************************************************* |
| 2 | * |
| 3 | * $Id$ |
| 4 | * |
| 5 | * Copyright (C) 2001 Bjorn Reese <breese@users.sourceforge.net> |
| 6 | * |
| 7 | * Permission to use, copy, modify, and distribute this software for any |
| 8 | * purpose with or without fee is hereby granted, provided that the above |
| 9 | * copyright notice and this permission notice appear in all copies. |
| 10 | * |
| 11 | * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED |
| 12 | * WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF |
| 13 | * MERCHANTIBILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE AUTHORS AND |
| 14 | * CONTRIBUTORS ACCEPT NO RESPONSIBILITY IN ANY CONCEIVABLE MANNER. |
| 15 | * |
| 16 | ************************************************************************ |
| 17 | * |
| 18 | * Functions to handle special quantities in floating-point numbers |
| 19 | * (that is, NaNs and infinity). They provide the capability to detect |
| 20 | * and fabricate special quantities. |
| 21 | * |
| 22 | * Although written to be as portable as possible, it can never be |
| 23 | * guaranteed to work on all platforms, as not all hardware supports |
| 24 | * special quantities. |
| 25 | * |
| 26 | * The approach used here (approximately) is to: |
| 27 | * |
| 28 | * 1. Use C99 functionality when available. |
| 29 | * 2. Use IEEE 754 bit-patterns if possible. |
| 30 | * 3. Use platform-specific techniques. |
| 31 | * |
| 32 | * This program has been tested on the following platforms (in |
| 33 | * alphabetic order) |
| 34 | * |
| 35 | * OS CPU Compiler |
| 36 | * ------------------------------------------------- |
| 37 | * AIX 4.1.4 PowerPC gcc |
| 38 | * Darwin 1.3.7 PowerPC gcc |
| 39 | * FreeBSD 2.2 x86 gcc |
| 40 | * FreeBSD 3.3 x86 gcc |
| 41 | * FreeBSD 4.3 x86 gcc |
| 42 | * FreeBSD 4.3 Alpha gcc |
| 43 | * HP-UX 10.20 PA-RISC gcc |
| 44 | * HP-UX 10.20 PA-RISC HP C++ |
| 45 | * IRIX 6.5 MIPS MIPSpro C |
| 46 | * Linux 2.2 x86 gcc |
| 47 | * Linux 2.2 Alpha gcc |
| 48 | * Linux 2.4 IA64 gcc |
| 49 | * Linux 2.4 StrongARM gcc |
| 50 | * NetBSD 1.4 x86 gcc |
| 51 | * NetBSD 1.4 StrongARM gcc |
| 52 | * NetBSD 1.5 Alpha gcc |
| 53 | * RISC OS 4 StrongARM Norcroft C |
| 54 | * Solaris 2.5.1 x86 gcc |
| 55 | * Solaris 2.5.1 Sparc gcc |
| 56 | * Solaris 2.6 Sparc WorkShop 4.2 |
| 57 | * Solaris 8 Sparc Forte C 6 |
| 58 | * Tru64 4.0D Alpha gcc |
| 59 | * Tru64 5.1 Alpha gcc |
| 60 | * WinNT x86 MSVC 5.0 & 6.0 |
| 61 | * |
| 62 | ************************************************************************/ |
| 63 | |
| 64 | static const char rcsid[] = "@(#)$Id$"; |
| 65 | |
| 66 | |
| 67 | /************************************************************************* |
| 68 | * Include files |
| 69 | */ |
| 70 | #include "triodef.h" |
| 71 | #include "trionan.h" |
| 72 | |
| 73 | #include <math.h> |
| 74 | #include <string.h> |
| 75 | #include <limits.h> |
| 76 | #include <float.h> |
| 77 | #if defined(TRIO_PLATFORM_UNIX) |
| 78 | # include <signal.h> |
| 79 | #endif |
| 80 | #include <assert.h> |
| 81 | |
| 82 | #ifndef __STDC__ |
| 83 | # define volatile |
| 84 | # define const |
| 85 | #endif |
| 86 | |
| 87 | /************************************************************************* |
| 88 | * Definitions |
| 89 | */ |
| 90 | |
| 91 | /* We must enable IEEE floating-point on Alpha */ |
| 92 | #if defined(__alpha) && !defined(_IEEE_FP) |
| 93 | # if defined(TRIO_COMPILER_DECC) |
| 94 | # error "Must be compiled with option -ieee" |
| 95 | # elif defined(TRIO_COMPILER_GCC) && (defined(__osf__) || defined(__linux__)) |
| 96 | # error "Must be compiled with option -mieee" |
| 97 | # endif |
| 98 | #endif /* __alpha && ! _IEEE_FP */ |
| 99 | |
| 100 | /* |
| 101 | * In ANSI/IEEE 754-1985 64-bits double format numbers have the |
| 102 | * following properties (amoungst others) |
| 103 | * |
| 104 | * o FLT_RADIX == 2: binary encoding |
| 105 | * o DBL_MAX_EXP == 1024: 11 bits exponent, where one bit is used |
| 106 | * to indicate special numbers (e.g. NaN and Infinity), so the |
| 107 | * maximum exponent is 10 bits wide (2^10 == 1024). |
| 108 | * o DBL_MANT_DIG == 53: The mantissa is 52 bits wide, but because |
| 109 | * numbers are normalized the initial binary 1 is represented |
| 110 | * implictly (the so-called "hidden bit"), which leaves us with |
| 111 | * the ability to represent 53 bits wide mantissa. |
| 112 | */ |
| 113 | #if (FLT_RADIX == 2) && (DBL_MAX_EXP == 1024) && (DBL_MANT_DIG == 53) |
| 114 | # define USE_IEEE_754 |
| 115 | #endif |
| 116 | |
| 117 | |
| 118 | /************************************************************************* |
| 119 | * Data |
| 120 | */ |
| 121 | |
| 122 | #if defined(USE_IEEE_754) |
| 123 | |
| 124 | /* |
| 125 | * Endian-agnostic indexing macro. |
| 126 | * |
| 127 | * The value of internalEndianMagic, when converted into a 64-bit |
| 128 | * integer, becomes 0x0001020304050607 (we could have used a 64-bit |
| 129 | * integer value instead of a double, but not all platforms supports |
| 130 | * that type). The value is automatically encoded with the correct |
| 131 | * endianess by the compiler, which means that we can support any |
| 132 | * kind of endianess. The individual bytes are then used as an index |
| 133 | * for the IEEE 754 bit-patterns and masks. |
| 134 | */ |
| 135 | #define TRIO_DOUBLE_INDEX(x) (((unsigned char *)&internalEndianMagic)[(x)]) |
| 136 | |
| 137 | static const double internalEndianMagic = 1.4015997730788920e-309; |
| 138 | |
| 139 | /* Mask for the exponent */ |
| 140 | static const unsigned char ieee_754_exponent_mask[] = { |
| 141 | 0x7F, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 |
| 142 | }; |
| 143 | |
| 144 | /* Mask for the mantissa */ |
| 145 | static const unsigned char ieee_754_mantissa_mask[] = { |
| 146 | 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF |
| 147 | }; |
| 148 | |
| 149 | /* Bit-pattern for infinity */ |
| 150 | static const unsigned char ieee_754_infinity_array[] = { |
| 151 | 0x7F, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 |
| 152 | }; |
| 153 | |
| 154 | /* Bit-pattern for quiet NaN */ |
| 155 | static const unsigned char ieee_754_qnan_array[] = { |
| 156 | 0x7F, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 |
| 157 | }; |
| 158 | |
| 159 | |
| 160 | /************************************************************************* |
| 161 | * trio_make_double |
| 162 | */ |
| 163 | static double |
| 164 | trio_make_double(const unsigned char *values) |
| 165 | { |
| 166 | volatile double result; |
| 167 | int i; |
| 168 | |
| 169 | for (i = 0; i < (int)sizeof(double); i++) { |
| 170 | ((volatile unsigned char *)&result)[TRIO_DOUBLE_INDEX(i)] = values[i]; |
| 171 | } |
| 172 | return result; |
| 173 | } |
| 174 | |
| 175 | /************************************************************************* |
| 176 | * trio_examine_double |
| 177 | */ |
| 178 | static int |
| 179 | trio_is_special_quantity(double number, |
| 180 | int *has_mantissa) |
| 181 | { |
| 182 | unsigned int i; |
| 183 | unsigned char current; |
| 184 | int is_special_quantity = (1 == 1); |
| 185 | |
| 186 | *has_mantissa = 0; |
| 187 | |
| 188 | for (i = 0; i < (unsigned int)sizeof(double); i++) { |
| 189 | current = ((unsigned char *)&number)[TRIO_DOUBLE_INDEX(i)]; |
| 190 | is_special_quantity |
| 191 | &= ((current & ieee_754_exponent_mask[i]) == ieee_754_exponent_mask[i]); |
| 192 | *has_mantissa |= (current & ieee_754_mantissa_mask[i]); |
| 193 | } |
| 194 | return is_special_quantity; |
| 195 | } |
| 196 | |
| 197 | #endif /* USE_IEEE_754 */ |
| 198 | |
| 199 | |
| 200 | /************************************************************************* |
| 201 | * trio_pinf |
| 202 | */ |
Daniel Veillard | cda9692 | 2001-08-21 10:56:31 +0000 | [diff] [blame] | 203 | TRIO_PUBLIC double |
Bjorn Reese | 4502960 | 2001-08-21 09:23:53 +0000 | [diff] [blame] | 204 | trio_pinf(void) |
| 205 | { |
| 206 | /* Cache the result */ |
| 207 | static double result = 0.0; |
| 208 | |
| 209 | if (result == 0.0) { |
| 210 | |
| 211 | #if defined(INFINITY) && defined(__STDC_IEC_559__) |
| 212 | result = (double)INFINITY; |
| 213 | |
| 214 | #elif defined(USE_IEEE_754) |
| 215 | result = trio_make_double(ieee_754_infinity_array); |
| 216 | |
| 217 | #else |
| 218 | /* |
| 219 | * If HUGE_VAL is different from DBL_MAX, then HUGE_VAL is used |
| 220 | * as infinity. Otherwise we have to resort to an overflow |
| 221 | * operation to generate infinity. |
| 222 | */ |
| 223 | # if defined(TRIO_PLATFORM_UNIX) |
| 224 | void (*signal_handler)(int) = signal(SIGFPE, SIG_IGN); |
| 225 | # endif |
| 226 | |
| 227 | result = HUGE_VAL; |
| 228 | if (HUGE_VAL == DBL_MAX) { |
| 229 | /* Force overflow */ |
| 230 | result += HUGE_VAL; |
| 231 | } |
| 232 | |
| 233 | # if defined(TRIO_PLATFORM_UNIX) |
| 234 | signal(SIGFPE, signal_handler); |
| 235 | # endif |
| 236 | |
| 237 | #endif |
| 238 | } |
| 239 | return result; |
| 240 | } |
| 241 | |
| 242 | /************************************************************************* |
| 243 | * trio_ninf |
| 244 | */ |
Daniel Veillard | cda9692 | 2001-08-21 10:56:31 +0000 | [diff] [blame] | 245 | TRIO_PUBLIC double |
Bjorn Reese | 4502960 | 2001-08-21 09:23:53 +0000 | [diff] [blame] | 246 | trio_ninf(void) |
| 247 | { |
| 248 | static double result = 0.0; |
| 249 | |
| 250 | if (result == 0.0) { |
| 251 | /* |
| 252 | * Negative infinity is calculated by negating positive infinity, |
| 253 | * which can be done because it is legal to do calculations on |
| 254 | * infinity (for example, 1 / infinity == 0). |
| 255 | */ |
| 256 | result = -trio_pinf(); |
| 257 | } |
| 258 | return result; |
| 259 | } |
| 260 | |
| 261 | /************************************************************************* |
| 262 | * trio_nan |
| 263 | */ |
Daniel Veillard | cda9692 | 2001-08-21 10:56:31 +0000 | [diff] [blame] | 264 | TRIO_PUBLIC double |
Bjorn Reese | 4502960 | 2001-08-21 09:23:53 +0000 | [diff] [blame] | 265 | trio_nan(void) |
| 266 | { |
| 267 | /* Cache the result */ |
| 268 | static double result = 0.0; |
| 269 | |
| 270 | if (result == 0.0) { |
| 271 | |
| 272 | #if defined(TRIO_COMPILER_SUPPORTS_C99) |
| 273 | result = nan(NULL); |
| 274 | |
| 275 | #elif defined(NAN) && defined(__STDC_IEC_559__) |
| 276 | result = (double)NAN; |
| 277 | |
| 278 | #elif defined(USE_IEEE_754) |
| 279 | result = trio_make_double(ieee_754_qnan_array); |
| 280 | |
| 281 | #else |
| 282 | /* |
| 283 | * There are several ways to generate NaN. The one used here is |
| 284 | * to divide infinity by infinity. I would have preferred to add |
| 285 | * negative infinity to positive infinity, but that yields wrong |
| 286 | * result (infinity) on FreeBSD. |
| 287 | * |
| 288 | * This may fail if the hardware does not support NaN, or if |
| 289 | * the Invalid Operation floating-point exception is unmasked. |
| 290 | */ |
| 291 | # if defined(TRIO_PLATFORM_UNIX) |
| 292 | void (*signal_handler)(int) = signal(SIGFPE, SIG_IGN); |
| 293 | # endif |
| 294 | |
| 295 | result = trio_pinf() / trio_pinf(); |
| 296 | |
| 297 | # if defined(TRIO_PLATFORM_UNIX) |
| 298 | signal(SIGFPE, signal_handler); |
| 299 | # endif |
| 300 | |
| 301 | #endif |
| 302 | } |
| 303 | return result; |
| 304 | } |
| 305 | |
| 306 | /************************************************************************* |
| 307 | * trio_isnan |
| 308 | */ |
Daniel Veillard | cda9692 | 2001-08-21 10:56:31 +0000 | [diff] [blame] | 309 | TRIO_PUBLIC int |
Bjorn Reese | 4502960 | 2001-08-21 09:23:53 +0000 | [diff] [blame] | 310 | trio_isnan(volatile double number) |
| 311 | { |
| 312 | #if defined(isnan) || defined(TRIO_COMPILER_SUPPORTS_UNIX95) |
| 313 | /* |
| 314 | * C99 defines isnan() as a macro. UNIX95 defines isnan() as a |
| 315 | * function. This function was already present in XPG4, but this |
| 316 | * is a bit tricky to detect with compiler defines, so we choose |
| 317 | * the conservative approach and only use it for UNIX95. |
| 318 | */ |
| 319 | return isnan(number); |
| 320 | |
| 321 | #elif defined(TRIO_COMPILER_MSVC) |
| 322 | /* |
| 323 | * MSC has an _isnan() function |
| 324 | */ |
| 325 | return _isnan(number); |
| 326 | |
| 327 | #elif defined(USE_IEEE_754) |
| 328 | /* |
| 329 | * Examine IEEE 754 bit-pattern. A NaN must have a special exponent |
| 330 | * pattern, and a non-empty mantissa. |
| 331 | */ |
| 332 | int has_mantissa; |
| 333 | int is_special_quantity; |
| 334 | |
| 335 | is_special_quantity = trio_is_special_quantity(number, &has_mantissa); |
| 336 | |
| 337 | return (is_special_quantity && has_mantissa); |
| 338 | |
| 339 | #else |
| 340 | /* |
| 341 | * Fallback solution |
| 342 | */ |
| 343 | int status; |
| 344 | double integral, fraction; |
| 345 | |
| 346 | # if defined(TRIO_PLATFORM_UNIX) |
| 347 | void (*signal_handler)(int) = signal(SIGFPE, SIG_IGN); |
| 348 | # endif |
| 349 | |
| 350 | status = (/* |
| 351 | * NaN is the only number which does not compare to itself |
| 352 | */ |
| 353 | (number != number) || |
| 354 | /* |
| 355 | * Fallback solution if NaN compares to NaN |
| 356 | */ |
| 357 | ((number != 0.0) && |
| 358 | (fraction = modf(number, &integral), |
| 359 | integral == fraction))); |
| 360 | |
| 361 | # if defined(TRIO_PLATFORM_UNIX) |
| 362 | signal(SIGFPE, signal_handler); |
| 363 | # endif |
| 364 | |
| 365 | return status; |
| 366 | |
| 367 | #endif |
| 368 | } |
| 369 | |
| 370 | /************************************************************************* |
| 371 | * trio_isinf |
| 372 | */ |
Daniel Veillard | cda9692 | 2001-08-21 10:56:31 +0000 | [diff] [blame] | 373 | TRIO_PUBLIC int |
Bjorn Reese | 4502960 | 2001-08-21 09:23:53 +0000 | [diff] [blame] | 374 | trio_isinf(volatile double number) |
| 375 | { |
| 376 | #if defined(TRIO_COMPILER_DECC) |
| 377 | /* |
| 378 | * DECC has an isinf() macro, but it works differently than that |
| 379 | * of C99, so we use the fp_class() function instead. |
| 380 | */ |
| 381 | return ((fp_class(number) == FP_POS_INF) |
| 382 | ? 1 |
| 383 | : ((fp_class(number) == FP_NEG_INF) ? -1 : 0)); |
| 384 | |
| 385 | #elif defined(isinf) |
| 386 | /* |
| 387 | * C99 defines isinf() as a macro. |
| 388 | */ |
| 389 | return isinf(number); |
| 390 | |
| 391 | #elif defined(TRIO_COMPILER_MSVC) |
| 392 | /* |
| 393 | * MSVC has an _fpclass() function that can be used to detect infinity. |
| 394 | */ |
| 395 | return ((_fpclass(number) == _FPCLASS_PINF) |
| 396 | ? 1 |
| 397 | : ((_fpclass(number) == _FPCLASS_NINF) ? -1 : 0)); |
| 398 | |
| 399 | #elif defined(USE_IEEE_754) |
| 400 | /* |
| 401 | * Examine IEEE 754 bit-pattern. Infinity must have a special exponent |
| 402 | * pattern, and an empty mantissa. |
| 403 | */ |
| 404 | int has_mantissa; |
| 405 | int is_special_quantity; |
| 406 | |
| 407 | is_special_quantity = trio_is_special_quantity(number, &has_mantissa); |
| 408 | |
| 409 | return (is_special_quantity && !has_mantissa) |
| 410 | ? ((number < 0.0) ? -1 : 1) |
| 411 | : 0; |
| 412 | |
| 413 | #else |
| 414 | /* |
| 415 | * Fallback solution. |
| 416 | */ |
| 417 | int status; |
| 418 | |
| 419 | # if defined(TRIO_PLATFORM_UNIX) |
| 420 | void (*signal_handler)(int) = signal(SIGFPE, SIG_IGN); |
| 421 | # endif |
| 422 | |
| 423 | double infinity = trio_pinf(); |
| 424 | |
| 425 | status = ((number == infinity) |
| 426 | ? 1 |
| 427 | : ((number == -infinity) ? -1 : 0)); |
| 428 | |
| 429 | # if defined(TRIO_PLATFORM_UNIX) |
| 430 | signal(SIGFPE, signal_handler); |
| 431 | # endif |
| 432 | |
| 433 | return status; |
| 434 | |
| 435 | #endif |
| 436 | } |
| 437 | |
| 438 | /************************************************************************* |
| 439 | */ |
| 440 | #if defined(STANDALONE) |
| 441 | # include <stdio.h> |
| 442 | |
| 443 | int main(void) |
| 444 | { |
| 445 | double my_nan; |
| 446 | double my_pinf; |
| 447 | double my_ninf; |
| 448 | # if defined(TRIO_PLATFORM_UNIX) |
| 449 | void (*signal_handler)(int); |
| 450 | # endif |
| 451 | |
| 452 | my_nan = trio_nan(); |
| 453 | my_pinf = trio_pinf(); |
| 454 | my_ninf = trio_ninf(); |
| 455 | |
| 456 | printf("NaN : %4g 0x%02x%02x%02x%02x%02x%02x%02x%02x (%2d, %2d)\n", |
| 457 | my_nan, |
| 458 | ((unsigned char *)&my_nan)[0], |
| 459 | ((unsigned char *)&my_nan)[1], |
| 460 | ((unsigned char *)&my_nan)[2], |
| 461 | ((unsigned char *)&my_nan)[3], |
| 462 | ((unsigned char *)&my_nan)[4], |
| 463 | ((unsigned char *)&my_nan)[5], |
| 464 | ((unsigned char *)&my_nan)[6], |
| 465 | ((unsigned char *)&my_nan)[7], |
| 466 | trio_isnan(my_nan), trio_isinf(my_nan)); |
| 467 | printf("PInf: %4g 0x%02x%02x%02x%02x%02x%02x%02x%02x (%2d, %2d)\n", |
| 468 | my_pinf, |
| 469 | ((unsigned char *)&my_pinf)[0], |
| 470 | ((unsigned char *)&my_pinf)[1], |
| 471 | ((unsigned char *)&my_pinf)[2], |
| 472 | ((unsigned char *)&my_pinf)[3], |
| 473 | ((unsigned char *)&my_pinf)[4], |
| 474 | ((unsigned char *)&my_pinf)[5], |
| 475 | ((unsigned char *)&my_pinf)[6], |
| 476 | ((unsigned char *)&my_pinf)[7], |
| 477 | trio_isnan(my_pinf), trio_isinf(my_pinf)); |
| 478 | printf("NInf: %4g 0x%02x%02x%02x%02x%02x%02x%02x%02x (%2d, %2d)\n", |
| 479 | my_ninf, |
| 480 | ((unsigned char *)&my_ninf)[0], |
| 481 | ((unsigned char *)&my_ninf)[1], |
| 482 | ((unsigned char *)&my_ninf)[2], |
| 483 | ((unsigned char *)&my_ninf)[3], |
| 484 | ((unsigned char *)&my_ninf)[4], |
| 485 | ((unsigned char *)&my_ninf)[5], |
| 486 | ((unsigned char *)&my_ninf)[6], |
| 487 | ((unsigned char *)&my_ninf)[7], |
| 488 | trio_isnan(my_ninf), trio_isinf(my_ninf)); |
| 489 | |
| 490 | # if defined(TRIO_PLATFORM_UNIX) |
| 491 | signal_handler = signal(SIGFPE, SIG_IGN); |
| 492 | # endif |
| 493 | |
| 494 | my_pinf = DBL_MAX + DBL_MAX; |
| 495 | my_ninf = -my_pinf; |
| 496 | my_nan = my_pinf / my_pinf; |
| 497 | |
| 498 | # if defined(TRIO_PLATFORM_UNIX) |
| 499 | signal(SIGFPE, signal_handler); |
| 500 | # endif |
| 501 | |
| 502 | printf("NaN : %4g 0x%02x%02x%02x%02x%02x%02x%02x%02x (%2d, %2d)\n", |
| 503 | my_nan, |
| 504 | ((unsigned char *)&my_nan)[0], |
| 505 | ((unsigned char *)&my_nan)[1], |
| 506 | ((unsigned char *)&my_nan)[2], |
| 507 | ((unsigned char *)&my_nan)[3], |
| 508 | ((unsigned char *)&my_nan)[4], |
| 509 | ((unsigned char *)&my_nan)[5], |
| 510 | ((unsigned char *)&my_nan)[6], |
| 511 | ((unsigned char *)&my_nan)[7], |
| 512 | trio_isnan(my_nan), trio_isinf(my_nan)); |
| 513 | printf("PInf: %4g 0x%02x%02x%02x%02x%02x%02x%02x%02x (%2d, %2d)\n", |
| 514 | my_pinf, |
| 515 | ((unsigned char *)&my_pinf)[0], |
| 516 | ((unsigned char *)&my_pinf)[1], |
| 517 | ((unsigned char *)&my_pinf)[2], |
| 518 | ((unsigned char *)&my_pinf)[3], |
| 519 | ((unsigned char *)&my_pinf)[4], |
| 520 | ((unsigned char *)&my_pinf)[5], |
| 521 | ((unsigned char *)&my_pinf)[6], |
| 522 | ((unsigned char *)&my_pinf)[7], |
| 523 | trio_isnan(my_pinf), trio_isinf(my_pinf)); |
| 524 | printf("NInf: %4g 0x%02x%02x%02x%02x%02x%02x%02x%02x (%2d, %2d)\n", |
| 525 | my_ninf, |
| 526 | ((unsigned char *)&my_ninf)[0], |
| 527 | ((unsigned char *)&my_ninf)[1], |
| 528 | ((unsigned char *)&my_ninf)[2], |
| 529 | ((unsigned char *)&my_ninf)[3], |
| 530 | ((unsigned char *)&my_ninf)[4], |
| 531 | ((unsigned char *)&my_ninf)[5], |
| 532 | ((unsigned char *)&my_ninf)[6], |
| 533 | ((unsigned char *)&my_ninf)[7], |
| 534 | trio_isnan(my_ninf), trio_isinf(my_ninf)); |
| 535 | |
| 536 | return 0; |
| 537 | } |
| 538 | #endif |