]> git.ipfire.org Git - thirdparty/glibc.git/blame - stdlib/strtod_l.c
Update.
[thirdparty/glibc.git] / stdlib / strtod_l.c
CommitLineData
0501d603 1/* Convert string representing a number to float value, using given locale.
ccadf7b5 2 Copyright (C) 1997,98,2002, 2004 Free Software Foundation, Inc.
0501d603
UD
3 This file is part of the GNU C Library.
4 Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
5
6 The GNU C Library is free software; you can redistribute it and/or
41bdb6e2
AJ
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.
0501d603
UD
10
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
41bdb6e2 14 Lesser General Public License for more details.
0501d603 15
41bdb6e2
AJ
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. */
0501d603 20
ebbad4cc
UD
21#include <xlocale.h>
22
0501d603 23extern double ____strtod_l_internal (const char *, char **, int, __locale_t);
c3d6c951
UD
24extern unsigned long long int ____strtoull_l_internal (const char *, char **,
25 int, int, __locale_t);
0501d603 26
ccadf7b5
UD
27/* Configuration part. These macros are defined by `strtold.c',
28 `strtof.c', `wcstod.c', `wcstold.c', and `wcstof.c' to produce the
29 `long double' and `float' versions of the reader. */
30#ifndef FLOAT
31# define FLOAT double
32# define FLT DBL
33# ifdef USE_WIDE_CHAR
34# define STRTOF wcstod_l
35# define __STRTOF __wcstod_l
36# else
37# define STRTOF strtod_l
38# define __STRTOF __strtod_l
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. */
53\f
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>
63
64/* The gmp headers need some configuration frobs. */
65#define HAVE_ALLOCA 1
66
67/* Include gmp-mparam.h first, such that definitions of _SHORT_LIMB
68 and _LONG_LONG_LIMB in it can take effect into gmp.h. */
69#include <gmp-mparam.h>
70#include <gmp.h>
71#include <gmp-impl.h>
72#include <longlong.h>
73#include "fpioconst.h"
74
75#define NDEBUG 1
76#include <assert.h>
77
78
79/* We use this code for the extended locale handling where the
80 function gets as an additional argument the locale which has to be
81 used. To access the values we have to redefine the _NL_CURRENT and
82 _NL_CURRENT_WORD macros. */
83#undef _NL_CURRENT
84#define _NL_CURRENT(category, item) \
85 (current->values[_NL_ITEM_INDEX (item)].string)
86#undef _NL_CURRENT_WORD
87#define _NL_CURRENT_WORD(category, item) \
88 ((uint32_t) current->values[_NL_ITEM_INDEX (item)].word)
89
90#if defined _LIBC || defined HAVE_WCHAR_H
91# include <wchar.h>
92#endif
93
94#ifdef USE_WIDE_CHAR
95# include <wctype.h>
96# define STRING_TYPE wchar_t
97# define CHAR_TYPE wint_t
98# define L_(Ch) L##Ch
99# define ISSPACE(Ch) __iswspace_l ((Ch), loc)
100# define ISDIGIT(Ch) __iswdigit_l ((Ch), loc)
101# define ISXDIGIT(Ch) __iswxdigit_l ((Ch), loc)
102# define TOLOWER(Ch) __towlower_l ((Ch), loc)
103# define STRNCASECMP(S1, S2, N) __wcsncasecmp_l ((S1), (S2), (N), loc)
104# define STRTOULL(S, E, B) ____wcstoull_l_internal ((S), (E), (B), 0, loc)
105#else
106# define STRING_TYPE char
107# define CHAR_TYPE char
108# define L_(Ch) Ch
109# define ISSPACE(Ch) __isspace_l ((Ch), loc)
110# define ISDIGIT(Ch) __isdigit_l ((Ch), loc)
111# define ISXDIGIT(Ch) __isxdigit_l ((Ch), loc)
112# define TOLOWER(Ch) __tolower_l ((Ch), loc)
113# define STRNCASECMP(S1, S2, N) __strncasecmp_l ((S1), (S2), (N), loc)
114# define STRTOULL(S, E, B) ____strtoull_l_internal ((S), (E), (B), 0, loc)
115#endif
116
117
118/* Constants we need from float.h; select the set for the FLOAT precision. */
119#define MANT_DIG PASTE(FLT,_MANT_DIG)
120#define DIG PASTE(FLT,_DIG)
121#define MAX_EXP PASTE(FLT,_MAX_EXP)
122#define MIN_EXP PASTE(FLT,_MIN_EXP)
123#define MAX_10_EXP PASTE(FLT,_MAX_10_EXP)
124#define MIN_10_EXP PASTE(FLT,_MIN_10_EXP)
125
126/* Extra macros required to get FLT expanded before the pasting. */
127#define PASTE(a,b) PASTE1(a,b)
128#define PASTE1(a,b) a##b
129
130/* Function to construct a floating point number from an MP integer
131 containing the fraction bits, a base 2 exponent, and a sign flag. */
132extern FLOAT MPN2FLOAT (mp_srcptr mpn, int exponent, int negative);
133\f
134/* Definitions according to limb size used. */
135#if BITS_PER_MP_LIMB == 32
136# define MAX_DIG_PER_LIMB 9
137# define MAX_FAC_PER_LIMB 1000000000UL
138#elif BITS_PER_MP_LIMB == 64
139# define MAX_DIG_PER_LIMB 19
140# define MAX_FAC_PER_LIMB 10000000000000000000ULL
141#else
142# error "mp_limb_t size " BITS_PER_MP_LIMB "not accounted for"
143#endif
144
145
146/* Local data structure. */
147static const mp_limb_t _tens_in_limb[MAX_DIG_PER_LIMB + 1] =
148{ 0, 10, 100,
149 1000, 10000, 100000L,
150 1000000L, 10000000L, 100000000L,
151 1000000000L
152#if BITS_PER_MP_LIMB > 32
153 , 10000000000ULL, 100000000000ULL,
154 1000000000000ULL, 10000000000000ULL, 100000000000000ULL,
155 1000000000000000ULL, 10000000000000000ULL, 100000000000000000ULL,
156 1000000000000000000ULL, 10000000000000000000ULL
157#endif
158#if BITS_PER_MP_LIMB > 64
159 #error "Need to expand tens_in_limb table to" MAX_DIG_PER_LIMB
160#endif
161};
162\f
163#ifndef howmany
164#define howmany(x,y) (((x)+((y)-1))/(y))
165#endif
166#define SWAP(x, y) ({ typeof(x) _tmp = x; x = y; y = _tmp; })
167
168#define NDIG (MAX_10_EXP - MIN_10_EXP + 2 * MANT_DIG)
169#define HEXNDIG ((MAX_EXP - MIN_EXP + 7) / 8 + 2 * MANT_DIG)
170#define RETURN_LIMB_SIZE howmany (MANT_DIG, BITS_PER_MP_LIMB)
171
172#define RETURN(val,end) \
173 do { if (endptr != NULL) *endptr = (STRING_TYPE *) (end); \
174 return val; } while (0)
175
176/* Maximum size necessary for mpn integers to hold floating point numbers. */
177#define MPNSIZE (howmany (MAX_EXP + 2 * MANT_DIG, BITS_PER_MP_LIMB) \
178 + 2)
179/* Declare an mpn integer variable that big. */
180#define MPN_VAR(name) mp_limb_t name[MPNSIZE]; mp_size_t name##size
181/* Copy an mpn integer value. */
182#define MPN_ASSIGN(dst, src) \
183 memcpy (dst, src, (dst##size = src##size) * sizeof (mp_limb_t))
184
185
186/* Return a floating point number of the needed type according to the given
187 multi-precision number after possible rounding. */
188static FLOAT
189round_and_return (mp_limb_t *retval, int exponent, int negative,
190 mp_limb_t round_limb, mp_size_t round_bit, int more_bits)
191{
192 if (exponent < MIN_EXP - 1)
193 {
194 mp_size_t shift = MIN_EXP - 1 - exponent;
195
196 if (shift > MANT_DIG)
197 {
198 __set_errno (EDOM);
199 return 0.0;
200 }
201
202 more_bits |= (round_limb & ((((mp_limb_t) 1) << round_bit) - 1)) != 0;
203 if (shift == MANT_DIG)
204 /* This is a special case to handle the very seldom case where
205 the mantissa will be empty after the shift. */
206 {
207 int i;
208
209 round_limb = retval[RETURN_LIMB_SIZE - 1];
210 round_bit = (MANT_DIG - 1) % BITS_PER_MP_LIMB;
211 for (i = 0; i < RETURN_LIMB_SIZE; ++i)
212 more_bits |= retval[i] != 0;
213 MPN_ZERO (retval, RETURN_LIMB_SIZE);
214 }
215 else if (shift >= BITS_PER_MP_LIMB)
216 {
217 int i;
218
219 round_limb = retval[(shift - 1) / BITS_PER_MP_LIMB];
220 round_bit = (shift - 1) % BITS_PER_MP_LIMB;
221 for (i = 0; i < (shift - 1) / BITS_PER_MP_LIMB; ++i)
222 more_bits |= retval[i] != 0;
223 more_bits |= ((round_limb & ((((mp_limb_t) 1) << round_bit) - 1))
224 != 0);
225
226 (void) __mpn_rshift (retval, &retval[shift / BITS_PER_MP_LIMB],
227 RETURN_LIMB_SIZE - (shift / BITS_PER_MP_LIMB),
228 shift % BITS_PER_MP_LIMB);
229 MPN_ZERO (&retval[RETURN_LIMB_SIZE - (shift / BITS_PER_MP_LIMB)],
230 shift / BITS_PER_MP_LIMB);
231 }
232 else if (shift > 0)
233 {
234 round_limb = retval[0];
235 round_bit = shift - 1;
236 (void) __mpn_rshift (retval, retval, RETURN_LIMB_SIZE, shift);
237 }
238 /* This is a hook for the m68k long double format, where the
239 exponent bias is the same for normalized and denormalized
240 numbers. */
241#ifndef DENORM_EXP
242# define DENORM_EXP (MIN_EXP - 2)
243#endif
244 exponent = DENORM_EXP;
245 }
246
247 if ((round_limb & (((mp_limb_t) 1) << round_bit)) != 0
248 && (more_bits || (retval[0] & 1) != 0
249 || (round_limb & ((((mp_limb_t) 1) << round_bit) - 1)) != 0))
250 {
251 mp_limb_t cy = __mpn_add_1 (retval, retval, RETURN_LIMB_SIZE, 1);
252
253 if (((MANT_DIG % BITS_PER_MP_LIMB) == 0 && cy) ||
254 ((MANT_DIG % BITS_PER_MP_LIMB) != 0 &&
255 (retval[RETURN_LIMB_SIZE - 1]
256 & (((mp_limb_t) 1) << (MANT_DIG % BITS_PER_MP_LIMB))) != 0))
257 {
258 ++exponent;
259 (void) __mpn_rshift (retval, retval, RETURN_LIMB_SIZE, 1);
260 retval[RETURN_LIMB_SIZE - 1]
261 |= ((mp_limb_t) 1) << ((MANT_DIG - 1) % BITS_PER_MP_LIMB);
262 }
263 else if (exponent == DENORM_EXP
264 && (retval[RETURN_LIMB_SIZE - 1]
265 & (((mp_limb_t) 1) << ((MANT_DIG - 1) % BITS_PER_MP_LIMB)))
266 != 0)
267 /* The number was denormalized but now normalized. */
268 exponent = MIN_EXP - 1;
269 }
270
271 if (exponent > MAX_EXP)
272 return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
273
274 return MPN2FLOAT (retval, exponent, negative);
275}
276
277
278/* Read a multi-precision integer starting at STR with exactly DIGCNT digits
279 into N. Return the size of the number limbs in NSIZE at the first
280 character od the string that is not part of the integer as the function
281 value. If the EXPONENT is small enough to be taken as an additional
282 factor for the resulting number (see code) multiply by it. */
283static const STRING_TYPE *
284str_to_mpn (const STRING_TYPE *str, int digcnt, mp_limb_t *n, mp_size_t *nsize,
285 int *exponent
286#ifndef USE_WIDE_CHAR
287 , const char *decimal, size_t decimal_len, const char *thousands
288#endif
289
290 )
291{
292 /* Number of digits for actual limb. */
293 int cnt = 0;
294 mp_limb_t low = 0;
295 mp_limb_t start;
296
297 *nsize = 0;
298 assert (digcnt > 0);
299 do
300 {
301 if (cnt == MAX_DIG_PER_LIMB)
302 {
303 if (*nsize == 0)
304 {
305 n[0] = low;
306 *nsize = 1;
307 }
308 else
309 {
310 mp_limb_t cy;
311 cy = __mpn_mul_1 (n, n, *nsize, MAX_FAC_PER_LIMB);
312 cy += __mpn_add_1 (n, n, *nsize, low);
313 if (cy != 0)
314 {
315 n[*nsize] = cy;
316 ++(*nsize);
317 }
318 }
319 cnt = 0;
320 low = 0;
321 }
322
323 /* There might be thousands separators or radix characters in
324 the string. But these all can be ignored because we know the
325 format of the number is correct and we have an exact number
326 of characters to read. */
327#ifdef USE_WIDE_CHAR
328 if (*str < L'0' || *str > L'9')
329 ++str;
330#else
331 if (*str < '0' || *str > '9')
332 {
333 int inner = 0;
334 if (thousands != NULL && *str == *thousands
335 && ({ for (inner = 1; thousands[inner] != '\0'; ++inner)
336 if (thousands[inner] != str[inner])
337 break;
338 thousands[inner] == '\0'; }))
339 str += inner;
340 else
341 str += decimal_len;
342 }
343#endif
344 low = low * 10 + *str++ - L_('0');
345 ++cnt;
346 }
347 while (--digcnt > 0);
348
349 if (*exponent > 0 && cnt + *exponent <= MAX_DIG_PER_LIMB)
350 {
351 low *= _tens_in_limb[*exponent];
352 start = _tens_in_limb[cnt + *exponent];
353 *exponent = 0;
354 }
355 else
356 start = _tens_in_limb[cnt];
357
358 if (*nsize == 0)
359 {
360 n[0] = low;
361 *nsize = 1;
362 }
363 else
364 {
365 mp_limb_t cy;
366 cy = __mpn_mul_1 (n, n, *nsize, start);
367 cy += __mpn_add_1 (n, n, *nsize, low);
368 if (cy != 0)
369 n[(*nsize)++] = cy;
370 }
371
372 return str;
373}
374
375
376/* Shift {PTR, SIZE} COUNT bits to the left, and fill the vacated bits
377 with the COUNT most significant bits of LIMB.
378
379 Tege doesn't like this function so I have to write it here myself. :)
380 --drepper */
381static inline void
382__attribute ((always_inline))
383__mpn_lshift_1 (mp_limb_t *ptr, mp_size_t size, unsigned int count,
384 mp_limb_t limb)
385{
386 if (__builtin_constant_p (count) && count == BITS_PER_MP_LIMB)
387 {
388 /* Optimize the case of shifting by exactly a word:
389 just copy words, with no actual bit-shifting. */
390 mp_size_t i;
391 for (i = size - 1; i > 0; --i)
392 ptr[i] = ptr[i - 1];
393 ptr[0] = limb;
394 }
395 else
396 {
397 (void) __mpn_lshift (ptr, ptr, size, count);
398 ptr[0] |= limb >> (BITS_PER_MP_LIMB - count);
399 }
400}
401
402
403#define INTERNAL(x) INTERNAL1(x)
404#define INTERNAL1(x) __##x##_internal
405
406/* This file defines a function to check for correct grouping. */
407#include "grouping.h"
408
409
410/* Return a floating point number with the value of the given string NPTR.
411 Set *ENDPTR to the character after the last used one. If the number is
412 smaller than the smallest representable number, set `errno' to ERANGE and
413 return 0.0. If the number is too big to be represented, set `errno' to
414 ERANGE and return HUGE_VAL with the appropriate sign. */
415FLOAT
416INTERNAL (__STRTOF) (nptr, endptr, group, loc)
417 const STRING_TYPE *nptr;
418 STRING_TYPE **endptr;
419 int group;
420 __locale_t loc;
421{
422 int negative; /* The sign of the number. */
423 MPN_VAR (num); /* MP representation of the number. */
424 int exponent; /* Exponent of the number. */
425
426 /* Numbers starting `0X' or `0x' have to be processed with base 16. */
427 int base = 10;
428
429 /* When we have to compute fractional digits we form a fraction with a
430 second multi-precision number (and we sometimes need a second for
431 temporary results). */
432 MPN_VAR (den);
433
434 /* Representation for the return value. */
435 mp_limb_t retval[RETURN_LIMB_SIZE];
436 /* Number of bits currently in result value. */
437 int bits;
438
439 /* Running pointer after the last character processed in the string. */
440 const STRING_TYPE *cp, *tp;
441 /* Start of significant part of the number. */
442 const STRING_TYPE *startp, *start_of_digits;
443 /* Points at the character following the integer and fractional digits. */
444 const STRING_TYPE *expp;
445 /* Total number of digit and number of digits in integer part. */
446 int dig_no, int_no, lead_zero;
447 /* Contains the last character read. */
448 CHAR_TYPE c;
449
450/* We should get wint_t from <stddef.h>, but not all GCC versions define it
451 there. So define it ourselves if it remains undefined. */
452#ifndef _WINT_T
453 typedef unsigned int wint_t;
454#endif
455 /* The radix character of the current locale. */
456#ifdef USE_WIDE_CHAR
457 wchar_t decimal;
458#else
459 const char *decimal;
460 size_t decimal_len;
461#endif
462 /* The thousands character of the current locale. */
463#ifdef USE_WIDE_CHAR
464 wchar_t thousands = L'\0';
465#else
466 const char *thousands = NULL;
467#endif
468 /* The numeric grouping specification of the current locale,
469 in the format described in <locale.h>. */
470 const char *grouping;
471 /* Used in several places. */
472 int cnt;
473
474 struct locale_data *current = loc->__locales[LC_NUMERIC];
475
476 if (group)
477 {
478 grouping = _NL_CURRENT (LC_NUMERIC, GROUPING);
479 if (*grouping <= 0 || *grouping == CHAR_MAX)
480 grouping = NULL;
481 else
482 {
483 /* Figure out the thousands separator character. */
484#ifdef USE_WIDE_CHAR
485 thousands = _NL_CURRENT_WORD (LC_NUMERIC,
486 _NL_NUMERIC_THOUSANDS_SEP_WC);
487 if (thousands == L'\0')
488 grouping = NULL;
489#else
490 thousands = _NL_CURRENT (LC_NUMERIC, THOUSANDS_SEP);
491 if (*thousands == '\0')
492 {
493 thousands = NULL;
494 grouping = NULL;
495 }
496#endif
497 }
498 }
499 else
500 grouping = NULL;
501
502 /* Find the locale's decimal point character. */
503#ifdef USE_WIDE_CHAR
504 decimal = _NL_CURRENT_WORD (LC_NUMERIC, _NL_NUMERIC_DECIMAL_POINT_WC);
505 assert (decimal != L'\0');
506# define decimal_len 1
507#else
508 decimal = _NL_CURRENT (LC_NUMERIC, DECIMAL_POINT);
509 decimal_len = strlen (decimal);
510 assert (decimal_len > 0);
511#endif
512
513 /* Prepare number representation. */
514 exponent = 0;
515 negative = 0;
516 bits = 0;
517
518 /* Parse string to get maximal legal prefix. We need the number of
519 characters of the integer part, the fractional part and the exponent. */
520 cp = nptr - 1;
521 /* Ignore leading white space. */
522 do
523 c = *++cp;
524 while (ISSPACE (c));
525
526 /* Get sign of the result. */
527 if (c == L_('-'))
528 {
529 negative = 1;
530 c = *++cp;
531 }
532 else if (c == L_('+'))
533 c = *++cp;
534
535 /* Return 0.0 if no legal string is found.
536 No character is used even if a sign was found. */
537#ifdef USE_WIDE_CHAR
538 if (c == (wint_t) decimal
539 && (wint_t) cp[1] >= L'0' && (wint_t) cp[1] <= L'9')
540 {
541 /* We accept it. This funny construct is here only to indent
542 the code directly. */
543 }
544#else
545 for (cnt = 0; decimal[cnt] != '\0'; ++cnt)
546 if (cp[cnt] != decimal[cnt])
547 break;
548 if (decimal[cnt] == '\0' && cp[cnt] >= '0' && cp[cnt] <= '9')
549 {
550 /* We accept it. This funny construct is here only to indent
551 the code directly. */
552 }
553#endif
554 else if (c < L_('0') || c > L_('9'))
555 {
556 /* Check for `INF' or `INFINITY'. */
557 if (TOLOWER (c) == L_('i') && STRNCASECMP (cp, L_("inf"), 3) == 0)
558 {
559 /* Return +/- infinity. */
560 if (endptr != NULL)
561 *endptr = (STRING_TYPE *)
562 (cp + (STRNCASECMP (cp + 3, L_("inity"), 5) == 0
563 ? 8 : 3));
564
565 return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
566 }
567
568 if (TOLOWER (c) == L_('n') && STRNCASECMP (cp, L_("nan"), 3) == 0)
569 {
570 /* Return NaN. */
571 FLOAT retval = NAN;
572
573 cp += 3;
574
575 /* Match `(n-char-sequence-digit)'. */
576 if (*cp == L_('('))
577 {
578 const STRING_TYPE *startp = cp;
579 do
580 ++cp;
581 while ((*cp >= L_('0') && *cp <= L_('9'))
582 || (TOLOWER (*cp) >= L_('a') && TOLOWER (*cp) <= L_('z'))
583 || *cp == L_('_'));
584
585 if (*cp != L_(')'))
586 /* The closing brace is missing. Only match the NAN
587 part. */
588 cp = startp;
589 else
590 {
591 /* This is a system-dependent way to specify the
592 bitmask used for the NaN. We expect it to be
593 a number which is put in the mantissa of the
594 number. */
595 STRING_TYPE *endp;
596 unsigned long long int mant;
597
598 mant = STRTOULL (startp + 1, &endp, 0);
599 if (endp == cp)
600 SET_MANTISSA (retval, mant);
601 }
602 }
603
604 if (endptr != NULL)
605 *endptr = (STRING_TYPE *) cp;
606
607 return retval;
608 }
609
610 /* It is really a text we do not recognize. */
611 RETURN (0.0, nptr);
612 }
613
614 /* First look whether we are faced with a hexadecimal number. */
615 if (c == L_('0') && TOLOWER (cp[1]) == L_('x'))
616 {
617 /* Okay, it is a hexa-decimal number. Remember this and skip
618 the characters. BTW: hexadecimal numbers must not be
619 grouped. */
620 base = 16;
621 cp += 2;
622 c = *cp;
623 grouping = NULL;
624 }
625
626 /* Record the start of the digits, in case we will check their grouping. */
627 start_of_digits = startp = cp;
628
629 /* Ignore leading zeroes. This helps us to avoid useless computations. */
630#ifdef USE_WIDE_CHAR
631 while (c == L'0' || ((wint_t) thousands != L'\0' && c == (wint_t) thousands))
632 c = *++cp;
633#else
634 if (thousands == NULL)
635 while (c == '0')
636 c = *++cp;
637 else
638 {
639 /* We also have the multibyte thousands string. */
640 while (1)
641 {
642 if (c != '0')
643 {
644 for (cnt = 0; thousands[cnt] != '\0'; ++cnt)
645 if (c != thousands[cnt])
646 break;
647 if (thousands[cnt] != '\0')
648 break;
649 }
650 c = *++cp;
651 }
652 }
653#endif
654
655 /* If no other digit but a '0' is found the result is 0.0.
656 Return current read pointer. */
657 if ((c < L_('0') || c > L_('9'))
658 && (base == 16 && (c < (CHAR_TYPE) TOLOWER (L_('a'))
659 || c > (CHAR_TYPE) TOLOWER (L_('f'))))
660#ifdef USE_WIDE_CHAR
661 && c != (wint_t) decimal
662#else
663 && ({ for (cnt = 0; decimal[cnt] != '\0'; ++cnt)
664 if (decimal[cnt] != cp[cnt])
665 break;
666 decimal[cnt] != '\0'; })
667#endif
668 && (base == 16 && (cp == start_of_digits
669 || (CHAR_TYPE) TOLOWER (c) != L_('p')))
670 && (base != 16 && (CHAR_TYPE) TOLOWER (c) != L_('e')))
671 {
672#ifdef USE_WIDE_CHAR
673 tp = __correctly_grouped_prefixwc (start_of_digits, cp, thousands,
674 grouping);
675#else
676 tp = __correctly_grouped_prefixmb (start_of_digits, cp, thousands,
677 grouping);
678#endif
679 /* If TP is at the start of the digits, there was no correctly
680 grouped prefix of the string; so no number found. */
681 RETURN (0.0, tp == start_of_digits ? (base == 16 ? cp - 1 : nptr) : tp);
682 }
683
684 /* Remember first significant digit and read following characters until the
685 decimal point, exponent character or any non-FP number character. */
686 startp = cp;
687 dig_no = 0;
688 while (1)
689 {
690 if ((c >= L_('0') && c <= L_('9'))
691 || (base == 16 && (wint_t) TOLOWER (c) >= L_('a')
692 && (wint_t) TOLOWER (c) <= L_('f')))
693 ++dig_no;
694 else
695 {
696#ifdef USE_WIDE_CHAR
697 if ((wint_t) thousands == L'\0' || c != (wint_t) thousands)
698 /* Not a digit or separator: end of the integer part. */
699 break;
700#else
701 if (thousands == NULL)
702 break;
703 else
704 {
705 for (cnt = 0; thousands[cnt] != '\0'; ++cnt)
706 if (thousands[cnt] != cp[cnt])
707 break;
708 if (thousands[cnt] != '\0')
709 break;
710 }
711#endif
712 }
713 c = *++cp;
714 }
715
716 if (grouping && dig_no > 0)
717 {
718 /* Check the grouping of the digits. */
719#ifdef USE_WIDE_CHAR
720 tp = __correctly_grouped_prefixwc (start_of_digits, cp, thousands,
721 grouping);
722#else
723 tp = __correctly_grouped_prefixmb (start_of_digits, cp, thousands,
724 grouping);
725#endif
726 if (cp != tp)
727 {
728 /* Less than the entire string was correctly grouped. */
729
730 if (tp == start_of_digits)
731 /* No valid group of numbers at all: no valid number. */
732 RETURN (0.0, nptr);
733
734 if (tp < startp)
735 /* The number is validly grouped, but consists
736 only of zeroes. The whole value is zero. */
737 RETURN (0.0, tp);
738
739 /* Recompute DIG_NO so we won't read more digits than
740 are properly grouped. */
741 cp = tp;
742 dig_no = 0;
743 for (tp = startp; tp < cp; ++tp)
744 if (*tp >= L_('0') && *tp <= L_('9'))
745 ++dig_no;
746
747 int_no = dig_no;
748 lead_zero = 0;
749
750 goto number_parsed;
751 }
752 }
753
754 /* We have the number digits in the integer part. Whether these are all or
755 any is really a fractional digit will be decided later. */
756 int_no = dig_no;
757 lead_zero = int_no == 0 ? -1 : 0;
758
759 /* Read the fractional digits. A special case are the 'american style'
760 numbers like `16.' i.e. with decimal but without trailing digits. */
761 if (
762#ifdef USE_WIDE_CHAR
763 c == (wint_t) decimal
764#else
765 ({ for (cnt = 0; decimal[cnt] != '\0'; ++cnt)
766 if (decimal[cnt] != cp[cnt])
767 break;
768 decimal[cnt] == '\0'; })
769#endif
770 )
771 {
772 cp += decimal_len;
773 c = *cp;
774 while ((c >= L_('0') && c <= L_('9')) ||
775 (base == 16 && TOLOWER (c) >= L_('a') && TOLOWER (c) <= L_('f')))
776 {
777 if (c != L_('0') && lead_zero == -1)
778 lead_zero = dig_no - int_no;
779 ++dig_no;
780 c = *++cp;
781 }
782 }
783
784 /* Remember start of exponent (if any). */
785 expp = cp;
786
787 /* Read exponent. */
788 if ((base == 16 && TOLOWER (c) == L_('p'))
789 || (base != 16 && TOLOWER (c) == L_('e')))
790 {
791 int exp_negative = 0;
792
793 c = *++cp;
794 if (c == L_('-'))
795 {
796 exp_negative = 1;
797 c = *++cp;
798 }
799 else if (c == L_('+'))
800 c = *++cp;
801
802 if (c >= L_('0') && c <= L_('9'))
803 {
804 int exp_limit;
805
806 /* Get the exponent limit. */
807 if (base == 16)
808 exp_limit = (exp_negative ?
809 -MIN_EXP + MANT_DIG + 4 * int_no :
810 MAX_EXP - 4 * int_no + lead_zero);
811 else
812 exp_limit = (exp_negative ?
813 -MIN_10_EXP + MANT_DIG + int_no :
814 MAX_10_EXP - int_no + lead_zero);
815
816 do
817 {
818 exponent *= 10;
819
820 if (exponent > exp_limit)
821 /* The exponent is too large/small to represent a valid
822 number. */
823 {
824 FLOAT result;
825
826 /* We have to take care for special situation: a joker
827 might have written "0.0e100000" which is in fact
828 zero. */
829 if (lead_zero == -1)
830 result = negative ? -0.0 : 0.0;
831 else
832 {
833 /* Overflow or underflow. */
834 __set_errno (ERANGE);
835 result = (exp_negative ? 0.0 :
836 negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL);
837 }
838
839 /* Accept all following digits as part of the exponent. */
840 do
841 ++cp;
842 while (*cp >= L_('0') && *cp <= L_('9'));
843
844 RETURN (result, cp);
845 /* NOTREACHED */
846 }
847
848 exponent += c - L_('0');
849 c = *++cp;
850 }
851 while (c >= L_('0') && c <= L_('9'));
852
853 if (exp_negative)
854 exponent = -exponent;
855 }
856 else
857 cp = expp;
858 }
859
860 /* We don't want to have to work with trailing zeroes after the radix. */
861 if (dig_no > int_no)
862 {
863 while (expp[-1] == L_('0'))
864 {
865 --expp;
866 --dig_no;
867 }
868 assert (dig_no >= int_no);
869 }
870
871 if (dig_no == int_no && dig_no > 0 && exponent < 0)
872 do
873 {
874 while (! (base == 16 ? ISXDIGIT (expp[-1]) : ISDIGIT (expp[-1])))
875 --expp;
876
877 if (expp[-1] != L_('0'))
878 break;
879
880 --expp;
881 --dig_no;
882 --int_no;
883 ++exponent;
884 }
885 while (dig_no > 0 && exponent < 0);
886
887 number_parsed:
888
889 /* The whole string is parsed. Store the address of the next character. */
890 if (endptr)
891 *endptr = (STRING_TYPE *) cp;
892
893 if (dig_no == 0)
894 return negative ? -0.0 : 0.0;
895
896 if (lead_zero)
897 {
898 /* Find the decimal point */
899#ifdef USE_WIDE_CHAR
900 while (*startp != decimal)
901 ++startp;
902#else
903 while (1)
904 {
905 if (*startp == decimal[0])
906 {
907 for (cnt = 1; decimal[cnt] != '\0'; ++cnt)
908 if (decimal[cnt] != startp[cnt])
909 break;
910 if (decimal[cnt] == '\0')
911 break;
912 }
913 ++startp;
914 }
915#endif
916 startp += lead_zero + decimal_len;
917 exponent -= base == 16 ? 4 * lead_zero : lead_zero;
918 dig_no -= lead_zero;
919 }
920
921 /* If the BASE is 16 we can use a simpler algorithm. */
922 if (base == 16)
923 {
924 static const int nbits[16] = { 0, 1, 2, 2, 3, 3, 3, 3,
925 4, 4, 4, 4, 4, 4, 4, 4 };
926 int idx = (MANT_DIG - 1) / BITS_PER_MP_LIMB;
927 int pos = (MANT_DIG - 1) % BITS_PER_MP_LIMB;
928 mp_limb_t val;
929
930 while (!ISXDIGIT (*startp))
931 ++startp;
932 while (*startp == L_('0'))
933 ++startp;
934 if (ISDIGIT (*startp))
935 val = *startp++ - L_('0');
936 else
937 val = 10 + TOLOWER (*startp++) - L_('a');
938 bits = nbits[val];
939 /* We cannot have a leading zero. */
940 assert (bits != 0);
941
942 if (pos + 1 >= 4 || pos + 1 >= bits)
943 {
944 /* We don't have to care for wrapping. This is the normal
945 case so we add the first clause in the `if' expression as
946 an optimization. It is a compile-time constant and so does
947 not cost anything. */
948 retval[idx] = val << (pos - bits + 1);
949 pos -= bits;
950 }
951 else
952 {
953 retval[idx--] = val >> (bits - pos - 1);
954 retval[idx] = val << (BITS_PER_MP_LIMB - (bits - pos - 1));
955 pos = BITS_PER_MP_LIMB - 1 - (bits - pos - 1);
956 }
957
958 /* Adjust the exponent for the bits we are shifting in. */
959 exponent += bits - 1 + (int_no - 1) * 4;
960
961 while (--dig_no > 0 && idx >= 0)
962 {
963 if (!ISXDIGIT (*startp))
964 startp += decimal_len;
965 if (ISDIGIT (*startp))
966 val = *startp++ - L_('0');
967 else
968 val = 10 + TOLOWER (*startp++) - L_('a');
969
970 if (pos + 1 >= 4)
971 {
972 retval[idx] |= val << (pos - 4 + 1);
973 pos -= 4;
974 }
975 else
976 {
977 retval[idx--] |= val >> (4 - pos - 1);
978 val <<= BITS_PER_MP_LIMB - (4 - pos - 1);
979 if (idx < 0)
980 return round_and_return (retval, exponent, negative, val,
981 BITS_PER_MP_LIMB - 1, dig_no > 0);
982
983 retval[idx] = val;
984 pos = BITS_PER_MP_LIMB - 1 - (4 - pos - 1);
985 }
986 }
987
988 /* We ran out of digits. */
989 MPN_ZERO (retval, idx);
990
991 return round_and_return (retval, exponent, negative, 0, 0, 0);
992 }
993
994 /* Now we have the number of digits in total and the integer digits as well
995 as the exponent and its sign. We can decide whether the read digits are
996 really integer digits or belong to the fractional part; i.e. we normalize
997 123e-2 to 1.23. */
998 {
999 register int incr = (exponent < 0 ? MAX (-int_no, exponent)
1000 : MIN (dig_no - int_no, exponent));
1001 int_no += incr;
1002 exponent -= incr;
1003 }
1004
1005 if (int_no + exponent > MAX_10_EXP + 1)
1006 {
1007 __set_errno (ERANGE);
1008 return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
1009 }
1010
1011 if (exponent < MIN_10_EXP - (DIG + 1))
1012 {
1013 __set_errno (ERANGE);
1014 return 0.0;
1015 }
1016
1017 if (int_no > 0)
1018 {
1019 /* Read the integer part as a multi-precision number to NUM. */
1020 startp = str_to_mpn (startp, int_no, num, &numsize, &exponent
1021#ifndef USE_WIDE_CHAR
1022 , decimal, decimal_len, thousands
1023#endif
1024 );
1025
1026 if (exponent > 0)
1027 {
1028 /* We now multiply the gained number by the given power of ten. */
1029 mp_limb_t *psrc = num;
1030 mp_limb_t *pdest = den;
1031 int expbit = 1;
1032 const struct mp_power *ttab = &_fpioconst_pow10[0];
1033
1034 do
1035 {
1036 if ((exponent & expbit) != 0)
1037 {
1038 size_t size = ttab->arraysize - _FPIO_CONST_OFFSET;
1039 mp_limb_t cy;
1040 exponent ^= expbit;
1041
1042 /* FIXME: not the whole multiplication has to be
1043 done. If we have the needed number of bits we
1044 only need the information whether more non-zero
1045 bits follow. */
1046 if (numsize >= ttab->arraysize - _FPIO_CONST_OFFSET)
1047 cy = __mpn_mul (pdest, psrc, numsize,
1048 &__tens[ttab->arrayoff
1049 + _FPIO_CONST_OFFSET],
1050 size);
1051 else
1052 cy = __mpn_mul (pdest, &__tens[ttab->arrayoff
1053 + _FPIO_CONST_OFFSET],
1054 size, psrc, numsize);
1055 numsize += size;
1056 if (cy == 0)
1057 --numsize;
1058 (void) SWAP (psrc, pdest);
1059 }
1060 expbit <<= 1;
1061 ++ttab;
1062 }
1063 while (exponent != 0);
1064
1065 if (psrc == den)
1066 memcpy (num, den, numsize * sizeof (mp_limb_t));
1067 }
1068
1069 /* Determine how many bits of the result we already have. */
1070 count_leading_zeros (bits, num[numsize - 1]);
1071 bits = numsize * BITS_PER_MP_LIMB - bits;
1072
1073 /* Now we know the exponent of the number in base two.
1074 Check it against the maximum possible exponent. */
1075 if (bits > MAX_EXP)
1076 {
1077 __set_errno (ERANGE);
1078 return negative ? -FLOAT_HUGE_VAL : FLOAT_HUGE_VAL;
1079 }
1080
1081 /* We have already the first BITS bits of the result. Together with
1082 the information whether more non-zero bits follow this is enough
1083 to determine the result. */
1084 if (bits > MANT_DIG)
1085 {
1086 int i;
1087 const mp_size_t least_idx = (bits - MANT_DIG) / BITS_PER_MP_LIMB;
1088 const mp_size_t least_bit = (bits - MANT_DIG) % BITS_PER_MP_LIMB;
1089 const mp_size_t round_idx = least_bit == 0 ? least_idx - 1
1090 : least_idx;
1091 const mp_size_t round_bit = least_bit == 0 ? BITS_PER_MP_LIMB - 1
1092 : least_bit - 1;
1093
1094 if (least_bit == 0)
1095 memcpy (retval, &num[least_idx],
1096 RETURN_LIMB_SIZE * sizeof (mp_limb_t));
1097 else
1098 {
1099 for (i = least_idx; i < numsize - 1; ++i)
1100 retval[i - least_idx] = (num[i] >> least_bit)
1101 | (num[i + 1]
1102 << (BITS_PER_MP_LIMB - least_bit));
1103 if (i - least_idx < RETURN_LIMB_SIZE)
1104 retval[RETURN_LIMB_SIZE - 1] = num[i] >> least_bit;
1105 }
1106
1107 /* Check whether any limb beside the ones in RETVAL are non-zero. */
1108 for (i = 0; num[i] == 0; ++i)
1109 ;
1110
1111 return round_and_return (retval, bits - 1, negative,
1112 num[round_idx], round_bit,
1113 int_no < dig_no || i < round_idx);
1114 /* NOTREACHED */
1115 }
1116 else if (dig_no == int_no)
1117 {
1118 const mp_size_t target_bit = (MANT_DIG - 1) % BITS_PER_MP_LIMB;
1119 const mp_size_t is_bit = (bits - 1) % BITS_PER_MP_LIMB;
1120
1121 if (target_bit == is_bit)
1122 {
1123 memcpy (&retval[RETURN_LIMB_SIZE - numsize], num,
1124 numsize * sizeof (mp_limb_t));
1125 /* FIXME: the following loop can be avoided if we assume a
1126 maximal MANT_DIG value. */
1127 MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize);
1128 }
1129 else if (target_bit > is_bit)
1130 {
1131 (void) __mpn_lshift (&retval[RETURN_LIMB_SIZE - numsize],
1132 num, numsize, target_bit - is_bit);
1133 /* FIXME: the following loop can be avoided if we assume a
1134 maximal MANT_DIG value. */
1135 MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize);
1136 }
1137 else
1138 {
1139 mp_limb_t cy;
1140 assert (numsize < RETURN_LIMB_SIZE);
1141
1142 cy = __mpn_rshift (&retval[RETURN_LIMB_SIZE - numsize],
1143 num, numsize, is_bit - target_bit);
1144 retval[RETURN_LIMB_SIZE - numsize - 1] = cy;
1145 /* FIXME: the following loop can be avoided if we assume a
1146 maximal MANT_DIG value. */
1147 MPN_ZERO (retval, RETURN_LIMB_SIZE - numsize - 1);
1148 }
1149
1150 return round_and_return (retval, bits - 1, negative, 0, 0, 0);
1151 /* NOTREACHED */
1152 }
1153
1154 /* Store the bits we already have. */
1155 memcpy (retval, num, numsize * sizeof (mp_limb_t));
1156#if RETURN_LIMB_SIZE > 1
1157 if (numsize < RETURN_LIMB_SIZE)
1158 retval[numsize] = 0;
1159#endif
1160 }
1161
1162 /* We have to compute at least some of the fractional digits. */
1163 {
1164 /* We construct a fraction and the result of the division gives us
1165 the needed digits. The denominator is 1.0 multiplied by the
1166 exponent of the lowest digit; i.e. 0.123 gives 123 / 1000 and
1167 123e-6 gives 123 / 1000000. */
1168
1169 int expbit;
1170 int neg_exp;
1171 int more_bits;
1172 mp_limb_t cy;
1173 mp_limb_t *psrc = den;
1174 mp_limb_t *pdest = num;
1175 const struct mp_power *ttab = &_fpioconst_pow10[0];
1176
1177 assert (dig_no > int_no && exponent <= 0);
1178
1179
1180 /* For the fractional part we need not process too many digits. One
1181 decimal digits gives us log_2(10) ~ 3.32 bits. If we now compute
1182 ceil(BITS / 3) =: N
1183 digits we should have enough bits for the result. The remaining
1184 decimal digits give us the information that more bits are following.
1185 This can be used while rounding. (Two added as a safety margin.) */
1186 if (dig_no - int_no > (MANT_DIG - bits + 2) / 3 + 2)
1187 {
1188 dig_no = int_no + (MANT_DIG - bits + 2) / 3 + 2;
1189 more_bits = 1;
1190 }
1191 else
1192 more_bits = 0;
1193
1194 neg_exp = dig_no - int_no - exponent;
1195
1196 /* Construct the denominator. */
1197 densize = 0;
1198 expbit = 1;
1199 do
1200 {
1201 if ((neg_exp & expbit) != 0)
1202 {
1203 mp_limb_t cy;
1204 neg_exp ^= expbit;
1205
1206 if (densize == 0)
1207 {
1208 densize = ttab->arraysize - _FPIO_CONST_OFFSET;
1209 memcpy (psrc, &__tens[ttab->arrayoff + _FPIO_CONST_OFFSET],
1210 densize * sizeof (mp_limb_t));
1211 }
1212 else
1213 {
1214 cy = __mpn_mul (pdest, &__tens[ttab->arrayoff
1215 + _FPIO_CONST_OFFSET],
1216 ttab->arraysize - _FPIO_CONST_OFFSET,
1217 psrc, densize);
1218 densize += ttab->arraysize - _FPIO_CONST_OFFSET;
1219 if (cy == 0)
1220 --densize;
1221 (void) SWAP (psrc, pdest);
1222 }
1223 }
1224 expbit <<= 1;
1225 ++ttab;
1226 }
1227 while (neg_exp != 0);
1228
1229 if (psrc == num)
1230 memcpy (den, num, densize * sizeof (mp_limb_t));
1231
1232 /* Read the fractional digits from the string. */
1233 (void) str_to_mpn (startp, dig_no - int_no, num, &numsize, &exponent
1234#ifndef USE_WIDE_CHAR
1235 , decimal, decimal_len, thousands
1236#endif
1237 );
1238
1239 /* We now have to shift both numbers so that the highest bit in the
1240 denominator is set. In the same process we copy the numerator to
1241 a high place in the array so that the division constructs the wanted
1242 digits. This is done by a "quasi fix point" number representation.
1243
1244 num: ddddddddddd . 0000000000000000000000
1245 |--- m ---|
1246 den: ddddddddddd n >= m
1247 |--- n ---|
1248 */
1249
1250 count_leading_zeros (cnt, den[densize - 1]);
1251
1252 if (cnt > 0)
1253 {
1254 /* Don't call `mpn_shift' with a count of zero since the specification
1255 does not allow this. */
1256 (void) __mpn_lshift (den, den, densize, cnt);
1257 cy = __mpn_lshift (num, num, numsize, cnt);
1258 if (cy != 0)
1259 num[numsize++] = cy;
1260 }
1261
1262 /* Now we are ready for the division. But it is not necessary to
1263 do a full multi-precision division because we only need a small
1264 number of bits for the result. So we do not use __mpn_divmod
1265 here but instead do the division here by hand and stop whenever
1266 the needed number of bits is reached. The code itself comes
1267 from the GNU MP Library by Torbj\"orn Granlund. */
1268
1269 exponent = bits;
1270
1271 switch (densize)
1272 {
1273 case 1:
1274 {
1275 mp_limb_t d, n, quot;
1276 int used = 0;
1277
1278 n = num[0];
1279 d = den[0];
1280 assert (numsize == 1 && n < d);
1281
1282 do
1283 {
1284 udiv_qrnnd (quot, n, n, 0, d);
1285
1286#define got_limb \
1287 if (bits == 0) \
1288 { \
1289 register int cnt; \
1290 if (quot == 0) \
1291 cnt = BITS_PER_MP_LIMB; \
1292 else \
1293 count_leading_zeros (cnt, quot); \
1294 exponent -= cnt; \
1295 if (BITS_PER_MP_LIMB - cnt > MANT_DIG) \
1296 { \
1297 used = MANT_DIG + cnt; \
1298 retval[0] = quot >> (BITS_PER_MP_LIMB - used); \
1299 bits = MANT_DIG + 1; \
1300 } \
1301 else \
1302 { \
1303 /* Note that we only clear the second element. */ \
1304 /* The conditional is determined at compile time. */ \
1305 if (RETURN_LIMB_SIZE > 1) \
1306 retval[1] = 0; \
1307 retval[0] = quot; \
1308 bits = -cnt; \
1309 } \
1310 } \
1311 else if (bits + BITS_PER_MP_LIMB <= MANT_DIG) \
1312 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE, BITS_PER_MP_LIMB, \
1313 quot); \
1314 else \
1315 { \
1316 used = MANT_DIG - bits; \
1317 if (used > 0) \
1318 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, quot); \
1319 } \
1320 bits += BITS_PER_MP_LIMB
1321
1322 got_limb;
1323 }
1324 while (bits <= MANT_DIG);
1325
1326 return round_and_return (retval, exponent - 1, negative,
1327 quot, BITS_PER_MP_LIMB - 1 - used,
1328 more_bits || n != 0);
1329 }
1330 case 2:
1331 {
1332 mp_limb_t d0, d1, n0, n1;
1333 mp_limb_t quot = 0;
1334 int used = 0;
1335
1336 d0 = den[0];
1337 d1 = den[1];
1338
1339 if (numsize < densize)
1340 {
1341 if (num[0] >= d1)
1342 {
1343 /* The numerator of the number occupies fewer bits than
1344 the denominator but the one limb is bigger than the
1345 high limb of the numerator. */
1346 n1 = 0;
1347 n0 = num[0];
1348 }
1349 else
1350 {
1351 if (bits <= 0)
1352 exponent -= BITS_PER_MP_LIMB;
1353 else
1354 {
1355 if (bits + BITS_PER_MP_LIMB <= MANT_DIG)
1356 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE,
1357 BITS_PER_MP_LIMB, 0);
1358 else
1359 {
1360 used = MANT_DIG - bits;
1361 if (used > 0)
1362 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, 0);
1363 }
1364 bits += BITS_PER_MP_LIMB;
1365 }
1366 n1 = num[0];
1367 n0 = 0;
1368 }
1369 }
1370 else
1371 {
1372 n1 = num[1];
1373 n0 = num[0];
1374 }
1375
1376 while (bits <= MANT_DIG)
1377 {
1378 mp_limb_t r;
1379
1380 if (n1 == d1)
1381 {
1382 /* QUOT should be either 111..111 or 111..110. We need
1383 special treatment of this rare case as normal division
1384 would give overflow. */
1385 quot = ~(mp_limb_t) 0;
1386
1387 r = n0 + d1;
1388 if (r < d1) /* Carry in the addition? */
1389 {
1390 add_ssaaaa (n1, n0, r - d0, 0, 0, d0);
1391 goto have_quot;
1392 }
1393 n1 = d0 - (d0 != 0);
1394 n0 = -d0;
1395 }
1396 else
1397 {
1398 udiv_qrnnd (quot, r, n1, n0, d1);
1399 umul_ppmm (n1, n0, d0, quot);
1400 }
1401
1402 q_test:
1403 if (n1 > r || (n1 == r && n0 > 0))
1404 {
1405 /* The estimated QUOT was too large. */
1406 --quot;
1407
1408 sub_ddmmss (n1, n0, n1, n0, 0, d0);
1409 r += d1;
1410 if (r >= d1) /* If not carry, test QUOT again. */
1411 goto q_test;
1412 }
1413 sub_ddmmss (n1, n0, r, 0, n1, n0);
1414
1415 have_quot:
1416 got_limb;
1417 }
1418
1419 return round_and_return (retval, exponent - 1, negative,
1420 quot, BITS_PER_MP_LIMB - 1 - used,
1421 more_bits || n1 != 0 || n0 != 0);
1422 }
1423 default:
1424 {
1425 int i;
1426 mp_limb_t cy, dX, d1, n0, n1;
1427 mp_limb_t quot = 0;
1428 int used = 0;
1429
1430 dX = den[densize - 1];
1431 d1 = den[densize - 2];
1432
1433 /* The division does not work if the upper limb of the two-limb
1434 numerator is greater than the denominator. */
1435 if (__mpn_cmp (num, &den[densize - numsize], numsize) > 0)
1436 num[numsize++] = 0;
1437
1438 if (numsize < densize)
1439 {
1440 mp_size_t empty = densize - numsize;
1441
1442 if (bits <= 0)
1443 {
1444 register int i;
1445 for (i = numsize; i > 0; --i)
1446 num[i + empty] = num[i - 1];
1447 MPN_ZERO (num, empty + 1);
1448 exponent -= empty * BITS_PER_MP_LIMB;
1449 }
1450 else
1451 {
1452 if (bits + empty * BITS_PER_MP_LIMB <= MANT_DIG)
1453 {
1454 /* We make a difference here because the compiler
1455 cannot optimize the `else' case that good and
1456 this reflects all currently used FLOAT types
1457 and GMP implementations. */
1458 register int i;
1459#if RETURN_LIMB_SIZE <= 2
1460 assert (empty == 1);
1461 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE,
1462 BITS_PER_MP_LIMB, 0);
1463#else
1464 for (i = RETURN_LIMB_SIZE; i > empty; --i)
1465 retval[i] = retval[i - empty];
1466#endif
1467 for (i = numsize; i > 0; --i)
1468 num[i + empty] = num[i - 1];
1469 MPN_ZERO (num, empty + 1);
1470 }
1471 else
1472 {
1473 used = MANT_DIG - bits;
1474 if (used >= BITS_PER_MP_LIMB)
1475 {
1476 register int i;
1477 (void) __mpn_lshift (&retval[used
1478 / BITS_PER_MP_LIMB],
1479 retval, RETURN_LIMB_SIZE,
1480 used % BITS_PER_MP_LIMB);
1481 for (i = used / BITS_PER_MP_LIMB; i >= 0; --i)
1482 retval[i] = 0;
1483 }
1484 else if (used > 0)
1485 __mpn_lshift_1 (retval, RETURN_LIMB_SIZE, used, 0);
1486 }
1487 bits += empty * BITS_PER_MP_LIMB;
1488 }
1489 }
1490 else
1491 {
1492 int i;
1493 assert (numsize == densize);
1494 for (i = numsize; i > 0; --i)
1495 num[i] = num[i - 1];
1496 }
1497
1498 den[densize] = 0;
1499 n0 = num[densize];
1500
1501 while (bits <= MANT_DIG)
1502 {
1503 if (n0 == dX)
1504 /* This might over-estimate QUOT, but it's probably not
1505 worth the extra code here to find out. */
1506 quot = ~(mp_limb_t) 0;
1507 else
1508 {
1509 mp_limb_t r;
1510
1511 udiv_qrnnd (quot, r, n0, num[densize - 1], dX);
1512 umul_ppmm (n1, n0, d1, quot);
1513
1514 while (n1 > r || (n1 == r && n0 > num[densize - 2]))
1515 {
1516 --quot;
1517 r += dX;
1518 if (r < dX) /* I.e. "carry in previous addition?" */
1519 break;
1520 n1 -= n0 < d1;
1521 n0 -= d1;
1522 }
1523 }
1524
1525 /* Possible optimization: We already have (q * n0) and (1 * n1)
1526 after the calculation of QUOT. Taking advantage of this, we
1527 could make this loop make two iterations less. */
1528
1529 cy = __mpn_submul_1 (num, den, densize + 1, quot);
1530
1531 if (num[densize] != cy)
1532 {
1533 cy = __mpn_add_n (num, num, den, densize);
1534 assert (cy != 0);
1535 --quot;
1536 }
1537 n0 = num[densize] = num[densize - 1];
1538 for (i = densize - 1; i > 0; --i)
1539 num[i] = num[i - 1];
1540
1541 got_limb;
1542 }
1543
1544 for (i = densize; num[i] == 0 && i >= 0; --i)
1545 ;
1546 return round_and_return (retval, exponent - 1, negative,
1547 quot, BITS_PER_MP_LIMB - 1 - used,
1548 more_bits || i >= 0);
1549 }
1550 }
1551 }
1552
1553 /* NOTREACHED */
1554}
1555#if defined _LIBC && !defined USE_WIDE_CHAR
1556libc_hidden_def (INTERNAL (__STRTOF))
1557#endif
1558\f
1559/* External user entry point. */
1ab62b32 1560
ccadf7b5
UD
1561FLOAT
1562#ifdef weak_function
1563weak_function
1564#endif
1565__STRTOF (nptr, endptr, loc)
1566 const STRING_TYPE *nptr;
1567 STRING_TYPE **endptr;
1568 __locale_t loc;
1569{
1570 return INTERNAL (__STRTOF) (nptr, endptr, 0, loc);
1571}
1572weak_alias (__STRTOF, STRTOF)