Update.
[glibc.git] / stdlib / strtod.c
blobdb0938404cd67509af834030be2347de59784b89
1 /* Read decimal floating point numbers.
2 This file is part of the GNU C Library.
3 Copyright (C) 1995,96,97,98,99,2000,2001 Free Software Foundation, Inc.
4 Contributed by Ulrich Drepper <drepper@gnu.org>, 1995.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, write to the Free
18 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
19 02111-1307 USA. */
21 /* Configuration part. These macros are defined by `strtold.c',
22 `strtof.c', `wcstod.c', `wcstold.c', and `wcstof.c' to produce the
23 `long double' and `float' versions of the reader. */
24 #ifndef FLOAT
25 # define FLOAT double
26 # define FLT DBL
27 # ifdef USE_WIDE_CHAR
28 # ifdef USE_IN_EXTENDED_LOCALE_MODEL
29 # define STRTOF __wcstod_l
30 # else
31 # define STRTOF wcstod
32 # endif
33 # else
34 # ifdef USE_IN_EXTENDED_LOCALE_MODEL
35 # define STRTOF __strtod_l
36 # else
37 # define STRTOF strtod
38 # endif
39 # endif
40 # define MPN2FLOAT __mpn_construct_double
41 # define FLOAT_HUGE_VAL HUGE_VAL
42 # define SET_MANTISSA(flt, mant) \
43 do { union ieee754_double u; \
44 u.d = (flt); \
45 if ((mant & 0xfffffffffffffULL) == 0) \
46 mant = 0x8000000000000ULL; \
47 u.ieee.mantissa0 = ((mant) >> 32) & 0xfffff; \
48 u.ieee.mantissa1 = (mant) & 0xffffffff; \
49 (flt) = u.d; \
50 } while (0)
51 #endif
52 /* End of configuration part. */
54 #include <ctype.h>
55 #include <errno.h>
56 #include <float.h>
57 #include <ieee754.h>
58 #include "../locale/localeinfo.h"
59 #include <locale.h>
60 #include <math.h>
61 #include <stdlib.h>
62 #include <string.h>
64 /* The gmp headers need some configuration frobs. */
65 #define HAVE_ALLOCA 1
67 #include <gmp.h>
68 #include <gmp-impl.h>
69 #include <gmp-mparam.h>
70 #include <longlong.h>
71 #include "fpioconst.h"
73 #define NDEBUG 1
74 #include <assert.h>
77 /* We use this code also for the extended locale handling where the
78 function gets as an additional argument the locale which has to be
79 used. To access the values we have to redefine the _NL_CURRENT
80 macro. */
81 #ifdef USE_IN_EXTENDED_LOCALE_MODEL
82 # undef _NL_CURRENT
83 # define _NL_CURRENT(category, item) \
84 (current->values[_NL_ITEM_INDEX (item)].string)
85 # define LOCALE_PARAM , loc
86 # define LOCALE_PARAM_DECL __locale_t loc;
87 #else
88 # define LOCALE_PARAM
89 # define LOCALE_PARAM_DECL
90 #endif
92 #if defined _LIBC || defined HAVE_WCHAR_H
93 # include <wchar.h>
94 #endif
96 #ifdef USE_WIDE_CHAR
97 # include <wctype.h>
98 # define STRING_TYPE wchar_t
99 # define CHAR_TYPE wint_t
100 # define L_(Ch) L##Ch
101 # ifdef USE_IN_EXTENDED_LOCALE_MODEL
102 # define ISSPACE(Ch) __iswspace_l ((Ch), loc)
103 # define ISDIGIT(Ch) __iswdigit_l ((Ch), loc)
104 # define ISXDIGIT(Ch) __iswxdigit_l ((Ch), loc)
105 # define TOLOWER(Ch) __towlower_l ((Ch), loc)
106 # define STRNCASECMP(S1, S2, N) __wcsncasecmp_l ((S1), (S2), (N), loc)
107 # define STRTOULL(S, E, B) ____wcstoull_l_internal ((S), (E), (B), 0, loc)
108 # else
109 # define ISSPACE(Ch) iswspace (Ch)
110 # define ISDIGIT(Ch) iswdigit (Ch)
111 # define ISXDIGIT(Ch) iswxdigit (Ch)
112 # define TOLOWER(Ch) towlower (Ch)
113 # define STRNCASECMP(S1, S2, N) __wcsncasecmp ((S1), (S2), (N))
114 # define STRTOULL(S, E, B) __wcstoull_internal ((S), (E), (B), 0)
115 # endif
116 #else
117 # define STRING_TYPE char
118 # define CHAR_TYPE char
119 # define L_(Ch) Ch
120 # ifdef USE_IN_EXTENDED_LOCALE_MODEL
121 # define ISSPACE(Ch) __isspace_l ((Ch), loc)
122 # define ISDIGIT(Ch) __isdigit_l ((Ch), loc)
123 # define ISXDIGIT(Ch) __isxdigit_l ((Ch), loc)
124 # define TOLOWER(Ch) __tolower_l ((Ch), loc)
125 # define STRNCASECMP(S1, S2, N) __strncasecmp_l ((S1), (S2), (N), loc)
126 # define STRTOULL(S, E, B) ____strtoull_l_internal ((S), (E), (B), 0, loc)
127 # else
128 # define ISSPACE(Ch) isspace (Ch)
129 # define ISDIGIT(Ch) isdigit (Ch)
130 # define ISXDIGIT(Ch) isxdigit (Ch)
131 # define TOLOWER(Ch) tolower (Ch)
132 # define STRNCASECMP(S1, S2, N) __strncasecmp ((S1), (S2), (N))
133 # define STRTOULL(S, E, B) __strtoull_internal ((S), (E), 0, (B))
134 # endif
135 #endif
138 /* Constants we need from float.h; select the set for the FLOAT precision. */
139 #define MANT_DIG PASTE(FLT,_MANT_DIG)
140 #define DIG PASTE(FLT,_DIG)
141 #define MAX_EXP PASTE(FLT,_MAX_EXP)
142 #define MIN_EXP PASTE(FLT,_MIN_EXP)
143 #define MAX_10_EXP PASTE(FLT,_MAX_10_EXP)
144 #define MIN_10_EXP PASTE(FLT,_MIN_10_EXP)
146 /* Extra macros required to get FLT expanded before the pasting. */
147 #define PASTE(a,b) PASTE1(a,b)
148 #define PASTE1(a,b) a##b
150 /* Function to construct a floating point number from an MP integer
151 containing the fraction bits, a base 2 exponent, and a sign flag. */
152 extern FLOAT MPN2FLOAT (mp_srcptr mpn, int exponent, int negative);
154 /* Definitions according to limb size used. */
155 #if BITS_PER_MP_LIMB == 32
156 # define MAX_DIG_PER_LIMB 9
157 # define MAX_FAC_PER_LIMB 1000000000UL
158 #elif BITS_PER_MP_LIMB == 64
159 # define MAX_DIG_PER_LIMB 19
160 # define MAX_FAC_PER_LIMB 10000000000000000000UL
161 #else
162 # error "mp_limb_t size " BITS_PER_MP_LIMB "not accounted for"
163 #endif
166 /* Local data structure. */
167 static const mp_limb_t _tens_in_limb[MAX_DIG_PER_LIMB + 1] =
168 { 0, 10, 100,
169 1000, 10000, 100000,
170 1000000, 10000000, 100000000,
171 1000000000
172 #if BITS_PER_MP_LIMB > 32
173 , 10000000000U, 100000000000U,
174 1000000000000U, 10000000000000U, 100000000000000U,
175 1000000000000000U, 10000000000000000U, 100000000000000000U,
176 1000000000000000000U, 10000000000000000000U
177 #endif
178 #if BITS_PER_MP_LIMB > 64
179 #error "Need to expand tens_in_limb table to" MAX_DIG_PER_LIMB
180 #endif
183 #ifndef howmany
184 #define howmany(x,y) (((x)+((y)-1))/(y))
185 #endif
186 #define SWAP(x, y) ({ typeof(x) _tmp = x; x = y; y = _tmp; })
188 #define NDIG (MAX_10_EXP - MIN_10_EXP + 2 * MANT_DIG)
189 #define HEXNDIG ((MAX_EXP - MIN_EXP + 7) / 8 + 2 * MANT_DIG)
190 #define RETURN_LIMB_SIZE howmany (MANT_DIG, BITS_PER_MP_LIMB)
192 #define RETURN(val,end) \
193 do { if (endptr != NULL) *endptr = (STRING_TYPE *) (end); \
194 return val; } while (0)
196 /* Maximum size necessary for mpn integers to hold floating point numbers. */
197 #define MPNSIZE (howmany (MAX_EXP + 2 * MANT_DIG, BITS_PER_MP_LIMB) \
198 + 2)
199 /* Declare an mpn integer variable that big. */
200 #define MPN_VAR(name) mp_limb_t name[MPNSIZE]; mp_size_t name##size
201 /* Copy an mpn integer value. */
202 #define MPN_ASSIGN(dst, src) \
203 memcpy (dst, src, (dst##size = src##size) * sizeof (mp_limb_t))
206 /* Return a floating point number of the needed type according to the given
207 multi-precision number after possible rounding. */
208 static inline FLOAT
209 round_and_return (mp_limb_t *retval, int exponent, int negative,
210 mp_limb_t round_limb, mp_size_t round_bit, int more_bits)
212 if (exponent < MIN_EXP - 1)
214 mp_size_t shift = MIN_EXP - 1 - exponent;
216 if (shift > MANT_DIG)
218 __set_errno (EDOM);
219 return 0.0;
222 more_bits |= (round_limb & ((((mp_limb_t) 1) << round_bit) - 1)) != 0;
223 if (shift == MANT_DIG)
224 /* This is a special case to handle the very seldom case where
225 the mantissa will be empty after the shift. */
227 int i;
229 round_limb = retval[RETURN_LIMB_SIZE - 1];
230 round_bit = (MANT_DIG - 1) % BITS_PER_MP_LIMB;
231 for (i = 0; i < RETURN_LIMB_SIZE; ++i)
232 more_bits |= retval[i] != 0;
233 MPN_ZERO (retval, RETURN_LIMB_SIZE);
235 else if (shift >= BITS_PER_MP_LIMB)
237 int i;
239 round_limb = retval[(shift - 1) / BITS_PER_MP_LIMB];
240 round_bit = (shift - 1) % BITS_PER_MP_LIMB;
241 for (i = 0; i < (shift - 1) / BITS_PER_MP_LIMB; ++i)
242 more_bits |= retval[i] != 0;
243 more_bits |= ((round_limb & ((((mp_limb_t) 1) << round_bit) - 1))
244 != 0);
246 (void) __mpn_rshift (retval, &retval[shift / BITS_PER_MP_LIMB],
247 RETURN_LIMB_SIZE - (shift / BITS_PER_MP_LIMB),
248 shift % BITS_PER_MP_LIMB);
249 MPN_ZERO (&retval[RETURN_LIMB_SIZE - (shift / BITS_PER_MP_LIMB)],
250 shift / BITS_PER_MP_LIMB);
252 else if (shift > 0)
254 round_limb = retval[0];
255 round_bit = shift - 1;
256 (void) __mpn_rshift (retval, retval, RETURN_LIMB_SIZE, shift);
258 /* This is a hook for the m68k long double format, where the
259 exponent bias is the same for normalized and denormalized
260 numbers. */
261 #ifndef DENORM_EXP
262 # define DENORM_EXP (MIN_EXP - 2)
263 #endif
264 exponent = DENORM_EXP;
267 if ((round_limb & (((mp_limb_t) 1) << round_bit)) != 0
268 && (more_bits || (retval[0] & 1) != 0
269 || (round_limb & ((((mp_limb_t) 1) << round_bit) - 1)) != 0))
271 mp_limb_t cy = __mpn_add_1 (retval, retval, RETURN_LIMB_SIZE, 1);
273 if (((MANT_DIG % BITS_PER_MP_LIMB) == 0 && cy) ||
274 ((MANT_DIG % BITS_PER_MP_LIMB) != 0 &&
275 (retval[RETURN_LIMB_SIZE - 1]
276 & (((mp_limb_t) 1) << (MANT_DIG % BITS_PER_MP_LIMB))) != 0))
278 ++exponent;
279 (void) __mpn_rshift (retval, retval, RETURN_LIMB_SIZE, 1);
280 retval[RETURN_LIMB_SIZE - 1]
281 |= ((mp_limb_t) 1) << ((MANT_DIG - 1) % BITS_PER_MP_LIMB);
283 else if (exponent == DENORM_EXP
284 && (retval[RETURN_LIMB_SIZE - 1]
285 & (((mp_limb_t) 1) << ((MANT_DIG - 1) % BITS_PER_MP_LIMB)))
286 != 0)
287 /* The number was denormalized but now normalized. */
288 exponent = MIN_EXP - 1;
291 if (exponent > MAX_EXP)
292 return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
294 return MPN2FLOAT (retval, exponent, negative);
298 /* Read a multi-precision integer starting at STR with exactly DIGCNT digits
299 into N. Return the size of the number limbs in NSIZE at the first
300 character od the string that is not part of the integer as the function
301 value. If the EXPONENT is small enough to be taken as an additional
302 factor for the resulting number (see code) multiply by it. */
303 static inline const STRING_TYPE *
304 str_to_mpn (const STRING_TYPE *str, int digcnt, mp_limb_t *n, mp_size_t *nsize,
305 int *exponent
306 #ifndef USE_WIDE_CHAR
307 , const char *decimal, size_t decimal_len, const char *thousands
308 #endif
312 /* Number of digits for actual limb. */
313 int cnt = 0;
314 mp_limb_t low = 0;
315 mp_limb_t start;
317 *nsize = 0;
318 assert (digcnt > 0);
321 if (cnt == MAX_DIG_PER_LIMB)
323 if (*nsize == 0)
325 n[0] = low;
326 *nsize = 1;
328 else
330 mp_limb_t cy;
331 cy = __mpn_mul_1 (n, n, *nsize, MAX_FAC_PER_LIMB);
332 cy += __mpn_add_1 (n, n, *nsize, low);
333 if (cy != 0)
335 n[*nsize] = cy;
336 ++(*nsize);
339 cnt = 0;
340 low = 0;
343 /* There might be thousands separators or radix characters in
344 the string. But these all can be ignored because we know the
345 format of the number is correct and we have an exact number
346 of characters to read. */
347 #ifdef USE_WIDE_CHAR
348 if (*str < L'0' || *str > L'9')
349 ++str;
350 #else
351 if (*str < '0' || *str > '9')
353 int inner = 0;
354 if (thousands != NULL && *str == *thousands
355 && ({ for (inner = 1; thousands[inner] != '\0'; ++inner)
356 if (thousands[inner] != str[inner])
357 break;
358 thousands[inner] == '\0'; }))
359 str += inner;
360 else
361 str += decimal_len;
363 #endif
364 low = low * 10 + *str++ - L_('0');
365 ++cnt;
367 while (--digcnt > 0);
369 if (*exponent > 0 && cnt + *exponent <= MAX_DIG_PER_LIMB)
371 low *= _tens_in_limb[*exponent];
372 start = _tens_in_limb[cnt + *exponent];
373 *exponent = 0;
375 else
376 start = _tens_in_limb[cnt];
378 if (*nsize == 0)
380 n[0] = low;
381 *nsize = 1;
383 else
385 mp_limb_t cy;
386 cy = __mpn_mul_1 (n, n, *nsize, start);
387 cy += __mpn_add_1 (n, n, *nsize, low);
388 if (cy != 0)
389 n[(*nsize)++] = cy;
392 return str;
396 /* Shift {PTR, SIZE} COUNT bits to the left, and fill the vacated bits
397 with the COUNT most significant bits of LIMB.
399 Tege doesn't like this function so I have to write it here myself. :)
400 --drepper */
401 static inline void
402 __mpn_lshift_1 (mp_limb_t *ptr, mp_size_t size, unsigned int count,
403 mp_limb_t limb)
405 if (count == BITS_PER_MP_LIMB)
407 /* Optimize the case of shifting by exactly a word:
408 just copy words, with no actual bit-shifting. */
409 mp_size_t i;
410 for (i = size - 1; i > 0; --i)
411 ptr[i] = ptr[i - 1];
412 ptr[0] = limb;
414 else
416 (void) __mpn_lshift (ptr, ptr, size, count);
417 ptr[0] |= limb >> (BITS_PER_MP_LIMB - count);
422 #define INTERNAL(x) INTERNAL1(x)
423 #define INTERNAL1(x) __##x##_internal
425 /* This file defines a function to check for correct grouping. */
426 #include "grouping.h"
429 /* Return a floating point number with the value of the given string NPTR.
430 Set *ENDPTR to the character after the last used one. If the number is
431 smaller than the smallest representable number, set `errno' to ERANGE and
432 return 0.0. If the number is too big to be represented, set `errno' to
433 ERANGE and return HUGE_VAL with the appropriate sign. */
434 FLOAT
435 INTERNAL (STRTOF) (nptr, endptr, group LOCALE_PARAM)
436 const STRING_TYPE *nptr;
437 STRING_TYPE **endptr;
438 int group;
439 LOCALE_PARAM_DECL
441 int negative; /* The sign of the number. */
442 MPN_VAR (num); /* MP representation of the number. */
443 int exponent; /* Exponent of the number. */
445 /* Numbers starting `0X' or `0x' have to be processed with base 16. */
446 int base = 10;
448 /* When we have to compute fractional digits we form a fraction with a
449 second multi-precision number (and we sometimes need a second for
450 temporary results). */
451 MPN_VAR (den);
453 /* Representation for the return value. */
454 mp_limb_t retval[RETURN_LIMB_SIZE];
455 /* Number of bits currently in result value. */
456 int bits;
458 /* Running pointer after the last character processed in the string. */
459 const STRING_TYPE *cp, *tp;
460 /* Start of significant part of the number. */
461 const STRING_TYPE *startp, *start_of_digits;
462 /* Points at the character following the integer and fractional digits. */
463 const STRING_TYPE *expp;
464 /* Total number of digit and number of digits in integer part. */
465 int dig_no, int_no, lead_zero;
466 /* Contains the last character read. */
467 CHAR_TYPE c;
469 /* We should get wint_t from <stddef.h>, but not all GCC versions define it
470 there. So define it ourselves if it remains undefined. */
471 #ifndef _WINT_T
472 typedef unsigned int wint_t;
473 #endif
474 /* The radix character of the current locale. */
475 #ifdef USE_WIDE_CHAR
476 wchar_t decimal;
477 #else
478 const char *decimal;
479 size_t decimal_len;
480 #endif
481 /* The thousands character of the current locale. */
482 #ifdef USE_WIDE_CHAR
483 wchar_t thousands = L'\0';
484 #else
485 const char *thousands = NULL;
486 #endif
487 /* The numeric grouping specification of the current locale,
488 in the format described in <locale.h>. */
489 const char *grouping;
490 /* Used in several places. */
491 int cnt;
493 #ifdef USE_IN_EXTENDED_LOCALE_MODEL
494 struct locale_data *current = loc->__locales[LC_NUMERIC];
495 #endif
497 if (group)
499 grouping = _NL_CURRENT (LC_NUMERIC, GROUPING);
500 if (*grouping <= 0 || *grouping == CHAR_MAX)
501 grouping = NULL;
502 else
504 /* Figure out the thousands separator character. */
505 #ifdef USE_WIDE_CHAR
506 thousands = _NL_CURRENT_WORD (LC_NUMERIC,
507 _NL_NUMERIC_THOUSANDS_SEP_WC);
508 if (thousands == L'\0')
509 grouping = NULL;
510 #else
511 thousands = _NL_CURRENT (LC_NUMERIC, THOUSANDS_SEP);
512 if (*thousands == '\0')
514 thousands = NULL;
515 grouping = NULL;
517 #endif
520 else
521 grouping = NULL;
523 /* Find the locale's decimal point character. */
524 #ifdef USE_WIDE_CHAR
525 decimal = _NL_CURRENT_WORD (LC_NUMERIC, _NL_NUMERIC_DECIMAL_POINT_WC);
526 assert (decimal != L'\0');
527 # define decimal_len 1
528 #else
529 decimal = _NL_CURRENT (LC_NUMERIC, DECIMAL_POINT);
530 decimal_len = strlen (decimal);
531 assert (decimal_len > 0);
532 #endif
534 /* Prepare number representation. */
535 exponent = 0;
536 negative = 0;
537 bits = 0;
539 /* Parse string to get maximal legal prefix. We need the number of
540 characters of the integer part, the fractional part and the exponent. */
541 cp = nptr - 1;
542 /* Ignore leading white space. */
544 c = *++cp;
545 while (ISSPACE (c));
547 /* Get sign of the result. */
548 if (c == L_('-'))
550 negative = 1;
551 c = *++cp;
553 else if (c == L_('+'))
554 c = *++cp;
556 /* Return 0.0 if no legal string is found.
557 No character is used even if a sign was found. */
558 #ifdef USE_WIDE_CHAR
559 if (c == decimal && cp[1] >= L'0' && cp[1] <= L'9')
561 /* We accept it. This funny construct is here only to indent
562 the code directly. */
564 #else
565 for (cnt = 0; decimal[cnt] != '\0'; ++cnt)
566 if (cp[cnt] != decimal[cnt])
567 break;
568 if (decimal[cnt] == '\0' && cp[1] >= '0' && cp[1] <= '9')
570 /* We accept it. This funny construct is here only to indent
571 the code directly. */
573 #endif
574 else if (c < L_('0') || c > L_('9'))
576 /* Check for `INF' or `INFINITY'. */
577 if (TOLOWER (c) == L_('i') && STRNCASECMP (cp, L_("inf"), 3) == 0)
579 /* Return +/- infinity. */
580 if (endptr != NULL)
581 *endptr = (STRING_TYPE *)
582 (cp + (STRNCASECMP (cp + 3, L_("inity"), 5) == 0
583 ? 8 : 3));
585 return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
588 if (TOLOWER (c) == L_('n') && STRNCASECMP (cp, L_("nan"), 3) == 0)
590 /* Return NaN. */
591 FLOAT retval = NAN;
593 cp += 3;
595 /* Match `(n-char-sequence-digit)'. */
596 if (*cp == L_('('))
598 const STRING_TYPE *startp = cp;
600 ++cp;
601 while ((*cp >= L_('0') && *cp <= L_('9'))
602 || (TOLOWER (*cp) >= L_('a') && TOLOWER (*cp) <= L_('z'))
603 || *cp == L_('_'));
605 if (*cp != L_(')'))
606 /* The closing brace is missing. Only match the NAN
607 part. */
608 cp = startp;
609 else
611 /* This is a system-dependent way to specify the
612 bitmask used for the NaN. We expect it to be
613 a number which is put in the mantissa of the
614 number. */
615 STRING_TYPE *endp;
616 unsigned long long int mant;
618 mant = STRTOULL (startp + 1, &endp, 0);
619 if (endp == cp)
620 SET_MANTISSA (retval, mant);
624 if (endptr != NULL)
625 *endptr = (STRING_TYPE *) cp;
627 return retval;
630 /* It is really a text we do not recognize. */
631 RETURN (0.0, nptr);
634 /* First look whether we are faced with a hexadecimal number. */
635 if (c == L_('0') && TOLOWER (cp[1]) == L_('x'))
637 /* Okay, it is a hexa-decimal number. Remember this and skip
638 the characters. BTW: hexadecimal numbers must not be
639 grouped. */
640 base = 16;
641 cp += 2;
642 c = *cp;
643 grouping = NULL;
646 /* Record the start of the digits, in case we will check their grouping. */
647 start_of_digits = startp = cp;
649 /* Ignore leading zeroes. This helps us to avoid useless computations. */
650 #ifdef USE_WIDE_CHAR
651 while (c == L'0' || (thousands != L'\0' && c == thousands))
652 c = *++cp;
653 #else
654 if (thousands == NULL)
655 while (c == '0')
656 c = *++cp;
657 else
659 /* We also have the multibyte thousands string. */
660 while (1)
662 if (c != '0')
664 for (cnt = 0; thousands[cnt] != '\0'; ++cnt)
665 if (c != thousands[cnt])
666 break;
667 if (thousands[cnt] != '\0')
668 break;
670 c = *++cp;
673 #endif
675 /* If no other digit but a '0' is found the result is 0.0.
676 Return current read pointer. */
677 if ((c < L_('0') || c > L_('9'))
678 && (base == 16 && (c < TOLOWER (L_('a')) || c > TOLOWER (L_('f'))))
679 #ifdef USE_WIDE_CHAR
680 && c != decimal
681 #else
682 && ({ for (cnt = 0; decimal[cnt] != '\0'; ++cnt)
683 if (decimal[cnt] != cp[cnt])
684 break;
685 decimal[cnt] != '\0'; })
686 #endif
687 && (base == 16 && (cp == start_of_digits || TOLOWER (c) != L_('p')))
688 && (base != 16 && TOLOWER (c) != L_('e')))
690 tp = correctly_grouped_prefix (start_of_digits, cp, thousands, grouping);
691 /* If TP is at the start of the digits, there was no correctly
692 grouped prefix of the string; so no number found. */
693 RETURN (0.0, tp == start_of_digits ? (base == 16 ? cp - 1 : nptr) : tp);
696 /* Remember first significant digit and read following characters until the
697 decimal point, exponent character or any non-FP number character. */
698 startp = cp;
699 dig_no = 0;
700 while (1)
702 if ((c >= L_('0') && c <= L_('9'))
703 || (base == 16 && TOLOWER (c) >= L_('a') && TOLOWER (c) <= L_('f')))
704 ++dig_no;
705 else
707 #ifdef USE_WIDE_CHAR
708 if (thousands == L'\0' || c != thousands)
709 /* Not a digit or separator: end of the integer part. */
710 break;
711 #else
712 if (thousands == NULL)
713 break;
714 else
716 for (cnt = 0; thousands[cnt] != '\0'; ++cnt)
717 if (thousands[cnt] != cp[cnt])
718 break;
719 if (thousands[cnt] != '\0')
720 break;
722 #endif
724 c = *++cp;
727 if (grouping && dig_no > 0)
729 /* Check the grouping of the digits. */
730 tp = correctly_grouped_prefix (start_of_digits, cp, thousands, grouping);
731 if (cp != tp)
733 /* Less than the entire string was correctly grouped. */
735 if (tp == start_of_digits)
736 /* No valid group of numbers at all: no valid number. */
737 RETURN (0.0, nptr);
739 if (tp < startp)
740 /* The number is validly grouped, but consists
741 only of zeroes. The whole value is zero. */
742 RETURN (0.0, tp);
744 /* Recompute DIG_NO so we won't read more digits than
745 are properly grouped. */
746 cp = tp;
747 dig_no = 0;
748 for (tp = startp; tp < cp; ++tp)
749 if (*tp >= L_('0') && *tp <= L_('9'))
750 ++dig_no;
752 int_no = dig_no;
753 lead_zero = 0;
755 goto number_parsed;
759 /* We have the number digits in the integer part. Whether these are all or
760 any is really a fractional digit will be decided later. */
761 int_no = dig_no;
762 lead_zero = int_no == 0 ? -1 : 0;
764 /* Read the fractional digits. A special case are the 'american style'
765 numbers like `16.' i.e. with decimal but without trailing digits. */
766 if (
767 #ifdef USE_WIDE_CHAR
768 c == decimal
769 #else
770 ({ for (cnt = 0; decimal[cnt] != '\0'; ++cnt)
771 if (decimal[cnt] != cp[cnt])
772 break;
773 decimal[cnt] == '\0'; })
774 #endif
777 cp += decimal_len;
778 c = *cp;
779 while ((c >= L_('0') && c <= L_('9')) ||
780 (base == 16 && TOLOWER (c) >= L_('a') && TOLOWER (c) <= L_('f')))
782 if (c != L_('0') && lead_zero == -1)
783 lead_zero = dig_no - int_no;
784 ++dig_no;
785 c = *++cp;
789 /* Remember start of exponent (if any). */
790 expp = cp;
792 /* Read exponent. */
793 if ((base == 16 && TOLOWER (c) == L_('p'))
794 || (base != 16 && TOLOWER (c) == L_('e')))
796 int exp_negative = 0;
798 c = *++cp;
799 if (c == L_('-'))
801 exp_negative = 1;
802 c = *++cp;
804 else if (c == L_('+'))
805 c = *++cp;
807 if (c >= L_('0') && c <= L_('9'))
809 int exp_limit;
811 /* Get the exponent limit. */
812 if (base == 16)
813 exp_limit = (exp_negative ?
814 -MIN_EXP + MANT_DIG + 4 * int_no :
815 MAX_EXP - 4 * int_no + lead_zero);
816 else
817 exp_limit = (exp_negative ?
818 -MIN_10_EXP + MANT_DIG + int_no :
819 MAX_10_EXP - int_no + lead_zero);
823 exponent *= 10;
825 if (exponent > exp_limit)
826 /* The exponent is too large/small to represent a valid
827 number. */
829 FLOAT result;
831 /* We have to take care for special situation: a joker
832 might have written "0.0e100000" which is in fact
833 zero. */
834 if (lead_zero == -1)
835 result = negative ? -0.0 : 0.0;
836 else
838 /* Overflow or underflow. */
839 __set_errno (ERANGE);
840 result = (exp_negative ? 0.0 :
841 negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL);
844 /* Accept all following digits as part of the exponent. */
846 ++cp;
847 while (*cp >= L_('0') && *cp <= L_('9'));
849 RETURN (result, cp);
850 /* NOTREACHED */
853 exponent += c - L_('0');
854 c = *++cp;
856 while (c >= L_('0') && c <= L_('9'));
858 if (exp_negative)
859 exponent = -exponent;
861 else
862 cp = expp;
865 /* We don't want to have to work with trailing zeroes after the radix. */
866 if (dig_no > int_no)
868 while (expp[-1] == L_('0'))
870 --expp;
871 --dig_no;
873 assert (dig_no >= int_no);
876 if (dig_no == int_no && dig_no > 0 && exponent < 0)
879 while (expp[-1] < L_('0') || expp[-1] > L_('9'))
880 --expp;
882 if (expp[-1] != L_('0'))
883 break;
885 --expp;
886 --dig_no;
887 --int_no;
888 ++exponent;
890 while (dig_no > 0 && exponent < 0);
892 number_parsed:
894 /* The whole string is parsed. Store the address of the next character. */
895 if (endptr)
896 *endptr = (STRING_TYPE *) cp;
898 if (dig_no == 0)
899 return negative ? -0.0 : 0.0;
901 if (lead_zero)
903 /* Find the decimal point */
904 #ifdef USE_WIDE_CHAR
905 while (*startp != decimal)
906 ++startp;
907 #else
908 while (1)
910 if (*startp == decimal[0])
912 for (cnt = 1; decimal[cnt] != '\0'; ++cnt)
913 if (decimal[cnt] != startp[cnt])
914 break;
915 if (decimal[cnt] == '\0')
916 break;
918 ++startp;
920 #endif
921 startp += lead_zero + decimal_len;
922 exponent -= base == 16 ? 4 * lead_zero : lead_zero;
923 dig_no -= lead_zero;
926 /* If the BASE is 16 we can use a simpler algorithm. */
927 if (base == 16)
929 static const int nbits[16] = { 0, 1, 2, 2, 3, 3, 3, 3,
930 4, 4, 4, 4, 4, 4, 4, 4 };
931 int idx = (MANT_DIG - 1) / BITS_PER_MP_LIMB;
932 int pos = (MANT_DIG - 1) % BITS_PER_MP_LIMB;
933 mp_limb_t val;
935 while (!ISXDIGIT (*startp))
936 ++startp;
937 while (*startp == L_('0'))
938 ++startp;
939 if (ISDIGIT (*startp))
940 val = *startp++ - L_('0');
941 else
942 val = 10 + TOLOWER (*startp++) - L_('a');
943 bits = nbits[val];
944 /* We cannot have a leading zero. */
945 assert (bits != 0);
947 if (pos + 1 >= 4 || pos + 1 >= bits)
949 /* We don't have to care for wrapping. This is the normal
950 case so we add the first clause in the `if' expression as
951 an optimization. It is a compile-time constant and so does
952 not cost anything. */
953 retval[idx] = val << (pos - bits + 1);
954 pos -= bits;
956 else
958 retval[idx--] = val >> (bits - pos - 1);
959 retval[idx] = val << (BITS_PER_MP_LIMB - (bits - pos - 1));
960 pos = BITS_PER_MP_LIMB - 1 - (bits - pos - 1);
963 /* Adjust the exponent for the bits we are shifting in. */
964 exponent += bits - 1 + (int_no - 1) * 4;
966 while (--dig_no > 0 && idx >= 0)
968 if (!ISXDIGIT (*startp))
969 startp += decimal_len;
970 if (ISDIGIT (*startp))
971 val = *startp++ - L_('0');
972 else
973 val = 10 + TOLOWER (*startp++) - L_('a');
975 if (pos + 1 >= 4)
977 retval[idx] |= val << (pos - 4 + 1);
978 pos -= 4;
980 else
982 retval[idx--] |= val >> (4 - pos - 1);
983 val <<= BITS_PER_MP_LIMB - (4 - pos - 1);
984 if (idx < 0)
985 return round_and_return (retval, exponent, negative, val,
986 BITS_PER_MP_LIMB - 1, dig_no > 0);
988 retval[idx] = val;
989 pos = BITS_PER_MP_LIMB - 1 - (4 - pos - 1);
993 /* We ran out of digits. */
994 MPN_ZERO (retval, idx);
996 return round_and_return (retval, exponent, negative, 0, 0, 0);
999 /* Now we have the number of digits in total and the integer digits as well
1000 as the exponent and its sign. We can decide whether the read digits are
1001 really integer digits or belong to the fractional part; i.e. we normalize
1002 123e-2 to 1.23. */
1004 register int incr = (exponent < 0 ? MAX (-int_no, exponent)
1005 : MIN (dig_no - int_no, exponent));
1006 int_no += incr;
1007 exponent -= incr;
1010 if (int_no + exponent > MAX_10_EXP + 1)
1012 __set_errno (ERANGE);
1013 return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
1016 if (exponent < MIN_10_EXP - (DIG + 1))
1018 __set_errno (ERANGE);
1019 return 0.0;
1022 if (int_no > 0)
1024 /* Read the integer part as a multi-precision number to NUM. */
1025 startp = str_to_mpn (startp, int_no, num, &numsize, &exponent
1026 #ifndef USE_WIDE_CHAR
1027 , decimal, decimal_len, thousands
1028 #endif
1031 if (exponent > 0)
1033 /* We now multiply the gained number by the given power of ten. */
1034 mp_limb_t *psrc = num;
1035 mp_limb_t *pdest = den;
1036 int expbit = 1;
1037 const struct mp_power *ttab = &_fpioconst_pow10[0];
1041 if ((exponent & expbit) != 0)
1043 size_t size = ttab->arraysize - _FPIO_CONST_OFFSET;
1044 mp_limb_t cy;
1045 exponent ^= expbit;
1047 /* FIXME: not the whole multiplication has to be
1048 done. If we have the needed number of bits we
1049 only need the information whether more non-zero
1050 bits follow. */
1051 if (numsize >= ttab->arraysize - _FPIO_CONST_OFFSET)
1052 cy = __mpn_mul (pdest, psrc, numsize,
1053 &__tens[ttab->arrayoff
1054 + _FPIO_CONST_OFFSET],
1055 size);
1056 else
1057 cy = __mpn_mul (pdest, &__tens[ttab->arrayoff
1058 + _FPIO_CONST_OFFSET],
1059 size, psrc, numsize);
1060 numsize += size;
1061 if (cy == 0)
1062 --numsize;
1063 (void) SWAP (psrc, pdest);
1065 expbit <<= 1;
1066 ++ttab;
1068 while (exponent != 0);
1070 if (psrc == den)
1071 memcpy (num, den, numsize * sizeof (mp_limb_t));
1074 /* Determine how many bits of the result we already have. */
1075 count_leading_zeros (bits, num[numsize - 1]);
1076 bits = numsize * BITS_PER_MP_LIMB - bits;
1078 /* Now we know the exponent of the number in base two.
1079 Check it against the maximum possible exponent. */
1080 if (bits > MAX_EXP)
1082 __set_errno (ERANGE);
1083 return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
1086 /* We have already the first BITS bits of the result. Together with
1087 the information whether more non-zero bits follow this is enough
1088 to determine the result. */
1089 if (bits > MANT_DIG)
1091 int i;
1092 const mp_size_t least_idx = (bits - MANT_DIG) / BITS_PER_MP_LIMB;
1093 const mp_size_t least_bit = (bits - MANT_DIG) % BITS_PER_MP_LIMB;
1094 const mp_size_t round_idx = least_bit == 0 ? least_idx - 1
1095 : least_idx;
1096 const mp_size_t round_bit = least_bit == 0 ? BITS_PER_MP_LIMB - 1
1097 : least_bit - 1;
1099 if (least_bit == 0)
1100 memcpy (retval, &num[least_idx],
1101 RETURN_LIMB_SIZE * sizeof (mp_limb_t));
1102 else
1104 for (i = least_idx; i < numsize - 1; ++i)
1105 retval[i - least_idx] = (num[i] >> least_bit)
1106 | (num[i + 1]
1107 << (BITS_PER_MP_LIMB - least_bit));
1108 if (i - least_idx < RETURN_LIMB_SIZE)
1109 retval[RETURN_LIMB_SIZE - 1] = num[i] >> least_bit;
1112 /* Check whether any limb beside the ones in RETVAL are non-zero. */
1113 for (i = 0; num[i] == 0; ++i)
1116 return round_and_return (retval, bits - 1, negative,
1117 num[round_idx], round_bit,
1118 int_no < dig_no || i < round_idx);
1119 /* NOTREACHED */
1121 else if (dig_no == int_no)
1123 const mp_size_t target_bit = (MANT_DIG - 1) % BITS_PER_MP_LIMB;
1124 const mp_size_t is_bit = (bits - 1) % BITS_PER_MP_LIMB;
1126 if (target_bit == is_bit)
1128 memcpy (&retval[RETURN_LIMB_SIZE - numsize], num,
1129 numsize * sizeof (mp_limb_t));
1130 /* FIXME: the following loop can be avoided if we assume a
1131 maximal MANT_DIG value. */
1132 MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize);
1134 else if (target_bit > is_bit)
1136 (void) __mpn_lshift (&retval[RETURN_LIMB_SIZE - numsize],
1137 num, numsize, target_bit - is_bit);
1138 /* FIXME: the following loop can be avoided if we assume a
1139 maximal MANT_DIG value. */
1140 MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize);
1142 else
1144 mp_limb_t cy;
1145 assert (numsize < RETURN_LIMB_SIZE);
1147 cy = __mpn_rshift (&retval[RETURN_LIMB_SIZE - numsize],
1148 num, numsize, is_bit - target_bit);
1149 retval[RETURN_LIMB_SIZE - numsize - 1] = cy;
1150 /* FIXME: the following loop can be avoided if we assume a
1151 maximal MANT_DIG value. */
1152 MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize - 1);
1155 return round_and_return (retval, bits - 1, negative, 0, 0, 0);
1156 /* NOTREACHED */
1159 /* Store the bits we already have. */
1160 memcpy (retval, num, numsize * sizeof (mp_limb_t));
1161 #if RETURN_LIMB_SIZE > 1
1162 if (numsize < RETURN_LIMB_SIZE)
1163 retval[numsize] = 0;
1164 #endif
1167 /* We have to compute at least some of the fractional digits. */
1169 /* We construct a fraction and the result of the division gives us
1170 the needed digits. The denominator is 1.0 multiplied by the
1171 exponent of the lowest digit; i.e. 0.123 gives 123 / 1000 and
1172 123e-6 gives 123 / 1000000. */
1174 int expbit;
1175 int neg_exp;
1176 int more_bits;
1177 mp_limb_t cy;
1178 mp_limb_t *psrc = den;
1179 mp_limb_t *pdest = num;
1180 const struct mp_power *ttab = &_fpioconst_pow10[0];
1182 assert (dig_no > int_no && exponent <= 0);
1185 /* For the fractional part we need not process too many digits. One
1186 decimal digits gives us log_2(10) ~ 3.32 bits. If we now compute
1187 ceil(BITS / 3) =: N
1188 digits we should have enough bits for the result. The remaining
1189 decimal digits give us the information that more bits are following.
1190 This can be used while rounding. (One added as a safety margin.) */
1191 if (dig_no - int_no > (MANT_DIG - bits + 2) / 3 + 1)
1193 dig_no = int_no + (MANT_DIG - bits + 2) / 3 + 1;
1194 more_bits = 1;
1196 else
1197 more_bits = 0;
1199 neg_exp = dig_no - int_no - exponent;
1201 /* Construct the denominator. */
1202 densize = 0;
1203 expbit = 1;
1206 if ((neg_exp & expbit) != 0)
1208 mp_limb_t cy;
1209 neg_exp ^= expbit;
1211 if (densize == 0)
1213 densize = ttab->arraysize - _FPIO_CONST_OFFSET;
1214 memcpy (psrc, &__tens[ttab->arrayoff + _FPIO_CONST_OFFSET],
1215 densize * sizeof (mp_limb_t));
1217 else
1219 cy = __mpn_mul (pdest, &__tens[ttab->arrayoff
1220 + _FPIO_CONST_OFFSET],
1221 ttab->arraysize - _FPIO_CONST_OFFSET,
1222 psrc, densize);
1223 densize += ttab->arraysize - _FPIO_CONST_OFFSET;
1224 if (cy == 0)
1225 --densize;
1226 (void) SWAP (psrc, pdest);
1229 expbit <<= 1;
1230 ++ttab;
1232 while (neg_exp != 0);
1234 if (psrc == num)
1235 memcpy (den, num, densize * sizeof (mp_limb_t));
1237 /* Read the fractional digits from the string. */
1238 (void) str_to_mpn (startp, dig_no - int_no, num, &numsize, &exponent
1239 #ifndef USE_WIDE_CHAR
1240 , decimal, decimal_len, thousands
1241 #endif
1244 /* We now have to shift both numbers so that the highest bit in the
1245 denominator is set. In the same process we copy the numerator to
1246 a high place in the array so that the division constructs the wanted
1247 digits. This is done by a "quasi fix point" number representation.
1249 num: ddddddddddd . 0000000000000000000000
1250 |--- m ---|
1251 den: ddddddddddd n >= m
1252 |--- n ---|
1255 count_leading_zeros (cnt, den[densize - 1]);
1257 if (cnt > 0)
1259 /* Don't call `mpn_shift' with a count of zero since the specification
1260 does not allow this. */
1261 (void) __mpn_lshift (den, den, densize, cnt);
1262 cy = __mpn_lshift (num, num, numsize, cnt);
1263 if (cy != 0)
1264 num[numsize++] = cy;
1267 /* Now we are ready for the division. But it is not necessary to
1268 do a full multi-precision division because we only need a small
1269 number of bits for the result. So we do not use __mpn_divmod
1270 here but instead do the division here by hand and stop whenever
1271 the needed number of bits is reached. The code itself comes
1272 from the GNU MP Library by Torbj\"orn Granlund. */
1274 exponent = bits;
1276 switch (densize)
1278 case 1:
1280 mp_limb_t d, n, quot;
1281 int used = 0;
1283 n = num[0];
1284 d = den[0];
1285 assert (numsize == 1 && n < d);
1289 udiv_qrnnd (quot, n, n, 0, d);
1291 #define got_limb \
1292 if (bits == 0) \
1294 register int cnt; \
1295 if (quot == 0) \
1296 cnt = BITS_PER_MP_LIMB; \
1297 else \
1298 count_leading_zeros (cnt, quot); \
1299 exponent -= cnt; \
1300 if (BITS_PER_MP_LIMB - cnt > MANT_DIG) \
1302 used = MANT_DIG + cnt; \
1303 retval[0] = quot >> (BITS_PER_MP_LIMB - used); \
1304 bits = MANT_DIG + 1; \
1306 else \
1308 /* Note that we only clear the second element. */ \
1309 /* The conditional is determined at compile time. */ \
1310 if (RETURN_LIMB_SIZE > 1) \
1311 retval[1] = 0; \
1312 retval[0] = quot; \
1313 bits = -cnt; \
1316 else if (bits + BITS_PER_MP_LIMB <= MANT_DIG) \
1317 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE, BITS_PER_MP_LIMB, \
1318 quot); \
1319 else \
1321 used = MANT_DIG - bits; \
1322 if (used > 0) \
1323 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, quot); \
1325 bits += BITS_PER_MP_LIMB
1327 got_limb;
1329 while (bits <= MANT_DIG);
1331 return round_and_return (retval, exponent - 1, negative,
1332 quot, BITS_PER_MP_LIMB - 1 - used,
1333 more_bits || n != 0);
1335 case 2:
1337 mp_limb_t d0, d1, n0, n1;
1338 mp_limb_t quot = 0;
1339 int used = 0;
1341 d0 = den[0];
1342 d1 = den[1];
1344 if (numsize < densize)
1346 if (num[0] >= d1)
1348 /* The numerator of the number occupies fewer bits than
1349 the denominator but the one limb is bigger than the
1350 high limb of the numerator. */
1351 n1 = 0;
1352 n0 = num[0];
1354 else
1356 if (bits <= 0)
1357 exponent -= BITS_PER_MP_LIMB;
1358 else
1360 if (bits + BITS_PER_MP_LIMB <= MANT_DIG)
1361 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE,
1362 BITS_PER_MP_LIMB, 0);
1363 else
1365 used = MANT_DIG - bits;
1366 if (used > 0)
1367 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, 0);
1369 bits += BITS_PER_MP_LIMB;
1371 n1 = num[0];
1372 n0 = 0;
1375 else
1377 n1 = num[1];
1378 n0 = num[0];
1381 while (bits <= MANT_DIG)
1383 mp_limb_t r;
1385 if (n1 == d1)
1387 /* QUOT should be either 111..111 or 111..110. We need
1388 special treatment of this rare case as normal division
1389 would give overflow. */
1390 quot = ~(mp_limb_t) 0;
1392 r = n0 + d1;
1393 if (r < d1) /* Carry in the addition? */
1395 add_ssaaaa (n1, n0, r - d0, 0, 0, d0);
1396 goto have_quot;
1398 n1 = d0 - (d0 != 0);
1399 n0 = -d0;
1401 else
1403 udiv_qrnnd (quot, r, n1, n0, d1);
1404 umul_ppmm (n1, n0, d0, quot);
1407 q_test:
1408 if (n1 > r || (n1 == r && n0 > 0))
1410 /* The estimated QUOT was too large. */
1411 --quot;
1413 sub_ddmmss (n1, n0, n1, n0, 0, d0);
1414 r += d1;
1415 if (r >= d1) /* If not carry, test QUOT again. */
1416 goto q_test;
1418 sub_ddmmss (n1, n0, r, 0, n1, n0);
1420 have_quot:
1421 got_limb;
1424 return round_and_return (retval, exponent - 1, negative,
1425 quot, BITS_PER_MP_LIMB - 1 - used,
1426 more_bits || n1 != 0 || n0 != 0);
1428 default:
1430 int i;
1431 mp_limb_t cy, dX, d1, n0, n1;
1432 mp_limb_t quot = 0;
1433 int used = 0;
1435 dX = den[densize - 1];
1436 d1 = den[densize - 2];
1438 /* The division does not work if the upper limb of the two-limb
1439 numerator is greater than the denominator. */
1440 if (__mpn_cmp (num, &den[densize - numsize], numsize) > 0)
1441 num[numsize++] = 0;
1443 if (numsize < densize)
1445 mp_size_t empty = densize - numsize;
1447 if (bits <= 0)
1449 register int i;
1450 for (i = numsize; i > 0; --i)
1451 num[i + empty] = num[i - 1];
1452 MPN_ZERO (num, empty + 1);
1453 exponent -= empty * BITS_PER_MP_LIMB;
1455 else
1457 if (bits + empty * BITS_PER_MP_LIMB <= MANT_DIG)
1459 /* We make a difference here because the compiler
1460 cannot optimize the `else' case that good and
1461 this reflects all currently used FLOAT types
1462 and GMP implementations. */
1463 register int i;
1464 #if RETURN_LIMB_SIZE <= 2
1465 assert (empty == 1);
1466 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE,
1467 BITS_PER_MP_LIMB, 0);
1468 #else
1469 for (i = RETURN_LIMB_SIZE; i > empty; --i)
1470 retval[i] = retval[i - empty];
1471 #endif
1472 #if RETURN_LIMB_SIZE > 1
1473 retval[1] = 0;
1474 #endif
1475 for (i = numsize; i > 0; --i)
1476 num[i + empty] = num[i - 1];
1477 MPN_ZERO (num, empty + 1);
1479 else
1481 used = MANT_DIG - bits;
1482 if (used >= BITS_PER_MP_LIMB)
1484 register int i;
1485 (void) __mpn_lshift (&retval[used
1486 / BITS_PER_MP_LIMB],
1487 retval, RETURN_LIMB_SIZE,
1488 used % BITS_PER_MP_LIMB);
1489 for (i = used / BITS_PER_MP_LIMB; i >= 0; --i)
1490 retval[i] = 0;
1492 else if (used > 0)
1493 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, 0);
1495 bits += empty * BITS_PER_MP_LIMB;
1498 else
1500 int i;
1501 assert (numsize == densize);
1502 for (i = numsize; i > 0; --i)
1503 num[i] = num[i - 1];
1506 den[densize] = 0;
1507 n0 = num[densize];
1509 while (bits <= MANT_DIG)
1511 if (n0 == dX)
1512 /* This might over-estimate QUOT, but it's probably not
1513 worth the extra code here to find out. */
1514 quot = ~(mp_limb_t) 0;
1515 else
1517 mp_limb_t r;
1519 udiv_qrnnd (quot, r, n0, num[densize - 1], dX);
1520 umul_ppmm (n1, n0, d1, quot);
1522 while (n1 > r || (n1 == r && n0 > num[densize - 2]))
1524 --quot;
1525 r += dX;
1526 if (r < dX) /* I.e. "carry in previous addition?" */
1527 break;
1528 n1 -= n0 < d1;
1529 n0 -= d1;
1533 /* Possible optimization: We already have (q * n0) and (1 * n1)
1534 after the calculation of QUOT. Taking advantage of this, we
1535 could make this loop make two iterations less. */
1537 cy = __mpn_submul_1 (num, den, densize + 1, quot);
1539 if (num[densize] != cy)
1541 cy = __mpn_add_n (num, num, den, densize);
1542 assert (cy != 0);
1543 --quot;
1545 n0 = num[densize] = num[densize - 1];
1546 for (i = densize - 1; i > 0; --i)
1547 num[i] = num[i - 1];
1549 got_limb;
1552 for (i = densize; num[i] == 0 && i >= 0; --i)
1554 return round_and_return (retval, exponent - 1, negative,
1555 quot, BITS_PER_MP_LIMB - 1 - used,
1556 more_bits || i >= 0);
1561 /* NOTREACHED */
1564 /* External user entry point. */
1566 FLOAT
1567 #ifdef weak_function
1568 weak_function
1569 #endif
1570 STRTOF (nptr, endptr LOCALE_PARAM)
1571 const STRING_TYPE *nptr;
1572 STRING_TYPE **endptr;
1573 LOCALE_PARAM_DECL
1575 return INTERNAL (STRTOF) (nptr, endptr, 0 LOCALE_PARAM);