• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /****************************************************************
2  *
3  * The author of this software is David M. Gay.
4  *
5  * Copyright (c) 1991, 2000, 2001 by Lucent Technologies.
6  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose without fee is hereby granted, provided that this entire notice
9  * is included in all copies of any software which is or includes a copy
10  * or modification of this software and in all copies of the supporting
11  * documentation for such software.
12  *
13  * THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR IMPLIED
14  * WARRANTY.  IN PARTICULAR, NEITHER THE AUTHOR NOR LUCENT MAKES ANY
15  * REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE MERCHANTABILITY
16  * OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR PURPOSE.
17  *
18  ***************************************************************/
19 
20 /* Please send bug reports to David M. Gay (dmg at acm dot org,
21  * with " at " changed at "@" and " dot " changed to ".").	*/
22 
23 /* On a machine with IEEE extended-precision registers, it is
24  * necessary to specify double-precision (53-bit) rounding precision
25  * before invoking strtod or dtoa.  If the machine uses (the equivalent
26  * of) Intel 80x87 arithmetic, the call
27  *	_control87(PC_53, MCW_PC);
28  * does this with many compilers.  Whether this or another call is
29  * appropriate depends on the compiler; for this to work, it may be
30  * necessary to #include "float.h" or another system-dependent header
31  * file.
32  */
33 
34 /* strtod for IEEE-, VAX-, and IBM-arithmetic machines.
35  *
36  * This strtod returns a nearest machine number to the input decimal
37  * string (or sets errno to ERANGE).  With IEEE arithmetic, ties are
38  * broken by the IEEE round-even rule.  Otherwise ties are broken by
39  * biased rounding (add half and chop).
40  *
41  * Inspired loosely by William D. Clinger's paper "How to Read Floating
42  * Point Numbers Accurately" [Proc. ACM SIGPLAN '90, pp. 92-101].
43  *
44  * Modifications:
45  *
46  *	1. We only require IEEE, IBM, or VAX double-precision
47  *		arithmetic (not IEEE double-extended).
48  *	2. We get by with floating-point arithmetic in a case that
49  *		Clinger missed -- when we're computing d * 10^n
50  *		for a small integer d and the integer n is not too
51  *		much larger than 22 (the maximum integer k for which
52  *		we can represent 10^k exactly), we may be able to
53  *		compute (d*10^k) * 10^(e-k) with just one roundoff.
54  *	3. Rather than a bit-at-a-time adjustment of the binary
55  *		result in the hard case, we use floating-point
56  *		arithmetic to determine the adjustment to within
57  *		one bit; only in really hard cases do we need to
58  *		compute a second residual.
59  *	4. Because of 3., we don't need a large table of powers of 10
60  *		for ten-to-e (just some small tables, e.g. of 10^k
61  *		for 0 <= k <= 22).
62  */
63 
64 /*
65  * #define IEEE_8087 for IEEE-arithmetic machines where the least
66  *	significant byte has the lowest address.
67  * #define IEEE_MC68k for IEEE-arithmetic machines where the most
68  *	significant byte has the lowest address.
69  * #define Long int on machines with 32-bit ints and 64-bit longs.
70  * #define IBM for IBM mainframe-style floating-point arithmetic.
71  * #define VAX for VAX-style floating-point arithmetic (D_floating).
72  * #define No_leftright to omit left-right logic in fast floating-point
73  *	computation of dtoa.
74  * #define Honor_FLT_ROUNDS if FLT_ROUNDS can assume the values 2 or 3
75  *	and strtod and dtoa should round accordingly.  Unless Trust_FLT_ROUNDS
76  *	is also #defined, fegetround() will be queried for the rounding mode.
77  *	Note that both FLT_ROUNDS and fegetround() are specified by the C99
78  *	standard (and are specified to be consistent, with fesetround()
79  *	affecting the value of FLT_ROUNDS), but that some (Linux) systems
80  *	do not work correctly in this regard, so using fegetround() is more
81  *	portable than using FLT_FOUNDS directly.
82  * #define Check_FLT_ROUNDS if FLT_ROUNDS can assume the values 2 or 3
83  *	and Honor_FLT_ROUNDS is not #defined.
84  * #define RND_PRODQUOT to use rnd_prod and rnd_quot (assembly routines
85  *	that use extended-precision instructions to compute rounded
86  *	products and quotients) with IBM.
87  * #define ROUND_BIASED for IEEE-format with biased rounding.
88  * #define Inaccurate_Divide for IEEE-format with correctly rounded
89  *	products but inaccurate quotients, e.g., for Intel i860.
90  * #define NO_LONG_LONG on machines that do not have a "long long"
91  *	integer type (of >= 64 bits).  On such machines, you can
92  *	#define Just_16 to store 16 bits per 32-bit Long when doing
93  *	high-precision integer arithmetic.  Whether this speeds things
94  *	up or slows things down depends on the machine and the number
95  *	being converted.  If long long is available and the name is
96  *	something other than "long long", #define Llong to be the name,
97  *	and if "unsigned Llong" does not work as an unsigned version of
98  *	Llong, #define #ULLong to be the corresponding unsigned type.
99  * #define KR_headers for old-style C function headers.
100  * #define Bad_float_h if your system lacks a float.h or if it does not
101  *	define some or all of DBL_DIG, DBL_MAX_10_EXP, DBL_MAX_EXP,
102  *	FLT_RADIX, FLT_ROUNDS, and DBL_MAX.
103  * #define MALLOC your_malloc, where your_malloc(n) acts like malloc(n)
104  *	if memory is available and otherwise does something you deem
105  *	appropriate.  If MALLOC is undefined, malloc will be invoked
106  *	directly -- and assumed always to succeed.  Similarly, if you
107  *	want something other than the system's free() to be called to
108  *	recycle memory acquired from MALLOC, #define FREE to be the
109  *	name of the alternate routine.  (FREE or free is only called in
110  *	pathological cases, e.g., in a dtoa call after a dtoa return in
111  *	mode 3 with thousands of digits requested.)
112  * #define Omit_Private_Memory to omit logic (added Jan. 1998) for making
113  *	memory allocations from a private pool of memory when possible.
114  *	When used, the private pool is PRIVATE_MEM bytes long:  2304 bytes,
115  *	unless #defined to be a different length.  This default length
116  *	suffices to get rid of MALLOC calls except for unusual cases,
117  *	such as decimal-to-binary conversion of a very long string of
118  *	digits.  The longest string dtoa can return is about 751 bytes
119  *	long.  For conversions by strtod of strings of 800 digits and
120  *	all dtoa conversions in single-threaded executions with 8-byte
121  *	pointers, PRIVATE_MEM >= 7400 appears to suffice; with 4-byte
122  *	pointers, PRIVATE_MEM >= 7112 appears adequate.
123  * #define NO_INFNAN_CHECK if you do not wish to have INFNAN_CHECK
124  *	#defined automatically on IEEE systems.  On such systems,
125  *	when INFNAN_CHECK is #defined, strtod checks
126  *	for Infinity and NaN (case insensitively).  On some systems
127  *	(e.g., some HP systems), it may be necessary to #define NAN_WORD0
128  *	appropriately -- to the most significant word of a quiet NaN.
129  *	(On HP Series 700/800 machines, -DNAN_WORD0=0x7ff40000 works.)
130  *	When INFNAN_CHECK is #defined and No_Hex_NaN is not #defined,
131  *	strtod also accepts (case insensitively) strings of the form
132  *	NaN(x), where x is a string of hexadecimal digits and spaces;
133  *	if there is only one string of hexadecimal digits, it is taken
134  *	for the 52 fraction bits of the resulting NaN; if there are two
135  *	or more strings of hex digits, the first is for the high 20 bits,
136  *	the second and subsequent for the low 32 bits, with intervening
137  *	white space ignored; but if this results in none of the 52
138  *	fraction bits being on (an IEEE Infinity symbol), then NAN_WORD0
139  *	and NAN_WORD1 are used instead.
140  * #define MULTIPLE_THREADS if the system offers preemptively scheduled
141  *	multiple threads.  In this case, you must provide (or suitably
142  *	#define) two locks, acquired by ACQUIRE_DTOA_LOCK(n) and freed
143  *	by FREE_DTOA_LOCK(n) for n = 0 or 1.  (The second lock, accessed
144  *	in pow5mult, ensures lazy evaluation of only one copy of high
145  *	powers of 5; omitting this lock would introduce a small
146  *	probability of wasting memory, but would otherwise be harmless.)
147  *	You must also invoke freedtoa(s) to free the value s returned by
148  *	dtoa.  You may do so whether or not MULTIPLE_THREADS is #defined.
149  * #define NO_IEEE_Scale to disable new (Feb. 1997) logic in strtod that
150  *	avoids underflows on inputs whose result does not underflow.
151  *	If you #define NO_IEEE_Scale on a machine that uses IEEE-format
152  *	floating-point numbers and flushes underflows to zero rather
153  *	than implementing gradual underflow, then you must also #define
154  *	Sudden_Underflow.
155  * #define USE_LOCALE to use the current locale's decimal_point value.
156  * #define SET_INEXACT if IEEE arithmetic is being used and extra
157  *	computation should be done to set the inexact flag when the
158  *	result is inexact and avoid setting inexact when the result
159  *	is exact.  In this case, dtoa.c must be compiled in
160  *	an environment, perhaps provided by #include "dtoa.c" in a
161  *	suitable wrapper, that defines two functions,
162  *		int get_inexact(void);
163  *		void clear_inexact(void);
164  *	such that get_inexact() returns a nonzero value if the
165  *	inexact bit is already set, and clear_inexact() sets the
166  *	inexact bit to 0.  When SET_INEXACT is #defined, strtod
167  *	also does extra computations to set the underflow and overflow
168  *	flags when appropriate (i.e., when the result is tiny and
169  *	inexact or when it is a numeric value rounded to +-infinity).
170  * #define NO_ERRNO if strtod should not assign errno = ERANGE when
171  *	the result overflows to +-Infinity or underflows to 0.
172  * #define NO_HEX_FP to omit recognition of hexadecimal floating-point
173  *	values by strtod.
174  * #define NO_STRTOD_BIGCOMP (on IEEE-arithmetic systems only for now)
175  *	to disable logic for "fast" testing of very long input strings
176  *	to strtod.  This testing proceeds by initially truncating the
177  *	input string, then if necessary comparing the whole string with
178  *	a decimal expansion to decide close cases. This logic is only
179  *	used for input more than STRTOD_DIGLIM digits long (default 40).
180  */
181 
182 #define IEEE_8087
183 #define NO_HEX_FP
184 
185 #ifndef Long
186 #if __LP64__
187 #define Long int
188 #else
189 #define Long long
190 #endif
191 #endif
192 #ifndef ULong
193 typedef unsigned Long ULong;
194 #endif
195 
196 #ifdef DEBUG
197 #include "stdio.h"
198 #define Bug(x) {fprintf(stderr, "%s\n", x); exit(1);}
199 #endif
200 
201 #include "stdlib.h"
202 #include "string.h"
203 
204 #ifdef USE_LOCALE
205 #include "locale.h"
206 #endif
207 
208 #ifdef Honor_FLT_ROUNDS
209 #ifndef Trust_FLT_ROUNDS
210 #include <fenv.h>
211 #endif
212 #endif
213 
214 #ifdef MALLOC
215 #ifdef KR_headers
216 extern char *MALLOC();
217 #else
218 extern void *MALLOC(size_t);
219 #endif
220 #else
221 #define MALLOC malloc
222 #endif
223 
224 #ifndef Omit_Private_Memory
225 #ifndef PRIVATE_MEM
226 #define PRIVATE_MEM 2304
227 #endif
228 #define PRIVATE_mem ((unsigned)((PRIVATE_MEM+sizeof(double)-1)/sizeof(double)))
229 static double private_mem[PRIVATE_mem], *pmem_next = private_mem;
230 #endif
231 
232 #undef IEEE_Arith
233 #undef Avoid_Underflow
234 #ifdef IEEE_MC68k
235 #define IEEE_Arith
236 #endif
237 #ifdef IEEE_8087
238 #define IEEE_Arith
239 #endif
240 
241 #ifdef IEEE_Arith
242 #ifndef NO_INFNAN_CHECK
243 #undef INFNAN_CHECK
244 #define INFNAN_CHECK
245 #endif
246 #else
247 #undef INFNAN_CHECK
248 #define NO_STRTOD_BIGCOMP
249 #endif
250 
251 #include "errno.h"
252 
253 #ifdef Bad_float_h
254 
255 #ifdef IEEE_Arith
256 #define DBL_DIG 15
257 #define DBL_MAX_10_EXP 308
258 #define DBL_MAX_EXP 1024
259 #define FLT_RADIX 2
260 #endif /*IEEE_Arith*/
261 
262 #ifdef IBM
263 #define DBL_DIG 16
264 #define DBL_MAX_10_EXP 75
265 #define DBL_MAX_EXP 63
266 #define FLT_RADIX 16
267 #define DBL_MAX 7.2370055773322621e+75
268 #endif
269 
270 #ifdef VAX
271 #define DBL_DIG 16
272 #define DBL_MAX_10_EXP 38
273 #define DBL_MAX_EXP 127
274 #define FLT_RADIX 2
275 #define DBL_MAX 1.7014118346046923e+38
276 #endif
277 
278 #ifndef LONG_MAX
279 #define LONG_MAX 2147483647
280 #endif
281 
282 #else /* ifndef Bad_float_h */
283 #include "float.h"
284 #endif /* Bad_float_h */
285 
286 #ifndef __MATH_H__
287 #include "math.h"
288 #endif
289 
290 namespace dmg_fp {
291 
292 #ifndef CONST
293 #ifdef KR_headers
294 #define CONST /* blank */
295 #else
296 #define CONST const
297 #endif
298 #endif
299 
300 #if defined(IEEE_8087) + defined(IEEE_MC68k) + defined(VAX) + defined(IBM) != 1
301 Exactly one of IEEE_8087, IEEE_MC68k, VAX, or IBM should be defined.
302 #endif
303 
304 typedef union { double d; ULong L[2]; } U;
305 
306 #ifdef IEEE_8087
307 #define word0(x) (x)->L[1]
308 #define word1(x) (x)->L[0]
309 #else
310 #define word0(x) (x)->L[0]
311 #define word1(x) (x)->L[1]
312 #endif
313 #define dval(x) (x)->d
314 
315 #ifndef STRTOD_DIGLIM
316 #define STRTOD_DIGLIM 40
317 #endif
318 
319 #ifdef DIGLIM_DEBUG
320 extern int strtod_diglim;
321 #else
322 #define strtod_diglim STRTOD_DIGLIM
323 #endif
324 
325 /* The following definition of Storeinc is appropriate for MIPS processors.
326  * An alternative that might be better on some machines is
327  * #define Storeinc(a,b,c) (*a++ = b << 16 | c & 0xffff)
328  */
329 #if defined(IEEE_8087) + defined(VAX)
330 #define Storeinc(a,b,c) (((unsigned short *)a)[1] = (unsigned short)b, \
331 ((unsigned short *)a)[0] = (unsigned short)c, a++)
332 #else
333 #define Storeinc(a,b,c) (((unsigned short *)a)[0] = (unsigned short)b, \
334 ((unsigned short *)a)[1] = (unsigned short)c, a++)
335 #endif
336 
337 /* #define P DBL_MANT_DIG */
338 /* Ten_pmax = floor(P*log(2)/log(5)) */
339 /* Bletch = (highest power of 2 < DBL_MAX_10_EXP) / 16 */
340 /* Quick_max = floor((P-1)*log(FLT_RADIX)/log(10) - 1) */
341 /* Int_max = floor(P*log(FLT_RADIX)/log(10) - 1) */
342 
343 #ifdef IEEE_Arith
344 #define Exp_shift  20
345 #define Exp_shift1 20
346 #define Exp_msk1    0x100000
347 #define Exp_msk11   0x100000
348 #define Exp_mask  0x7ff00000
349 #define P 53
350 #define Nbits 53
351 #define Bias 1023
352 #define Emax 1023
353 #define Emin (-1022)
354 #define Exp_1  0x3ff00000
355 #define Exp_11 0x3ff00000
356 #define Ebits 11
357 #define Frac_mask  0xfffff
358 #define Frac_mask1 0xfffff
359 #define Ten_pmax 22
360 #define Bletch 0x10
361 #define Bndry_mask  0xfffff
362 #define Bndry_mask1 0xfffff
363 #define LSB 1
364 #define Sign_bit 0x80000000
365 #define Log2P 1
366 #define Tiny0 0
367 #define Tiny1 1
368 #define Quick_max 14
369 #define Int_max 14
370 #ifndef NO_IEEE_Scale
371 #define Avoid_Underflow
372 #ifdef Flush_Denorm	/* debugging option */
373 #undef Sudden_Underflow
374 #endif
375 #endif
376 
377 #ifndef Flt_Rounds
378 #ifdef FLT_ROUNDS
379 #define Flt_Rounds FLT_ROUNDS
380 #else
381 #define Flt_Rounds 1
382 #endif
383 #endif /*Flt_Rounds*/
384 
385 #ifdef Honor_FLT_ROUNDS
386 #undef Check_FLT_ROUNDS
387 #define Check_FLT_ROUNDS
388 #else
389 #define Rounding Flt_Rounds
390 #endif
391 
392 #else /* ifndef IEEE_Arith */
393 #undef Check_FLT_ROUNDS
394 #undef Honor_FLT_ROUNDS
395 #undef SET_INEXACT
396 #undef  Sudden_Underflow
397 #define Sudden_Underflow
398 #ifdef IBM
399 #undef Flt_Rounds
400 #define Flt_Rounds 0
401 #define Exp_shift  24
402 #define Exp_shift1 24
403 #define Exp_msk1   0x1000000
404 #define Exp_msk11  0x1000000
405 #define Exp_mask  0x7f000000
406 #define P 14
407 #define Nbits 56
408 #define Bias 65
409 #define Emax 248
410 #define Emin (-260)
411 #define Exp_1  0x41000000
412 #define Exp_11 0x41000000
413 #define Ebits 8	/* exponent has 7 bits, but 8 is the right value in b2d */
414 #define Frac_mask  0xffffff
415 #define Frac_mask1 0xffffff
416 #define Bletch 4
417 #define Ten_pmax 22
418 #define Bndry_mask  0xefffff
419 #define Bndry_mask1 0xffffff
420 #define LSB 1
421 #define Sign_bit 0x80000000
422 #define Log2P 4
423 #define Tiny0 0x100000
424 #define Tiny1 0
425 #define Quick_max 14
426 #define Int_max 15
427 #else /* VAX */
428 #undef Flt_Rounds
429 #define Flt_Rounds 1
430 #define Exp_shift  23
431 #define Exp_shift1 7
432 #define Exp_msk1    0x80
433 #define Exp_msk11   0x800000
434 #define Exp_mask  0x7f80
435 #define P 56
436 #define Nbits 56
437 #define Bias 129
438 #define Emax 126
439 #define Emin (-129)
440 #define Exp_1  0x40800000
441 #define Exp_11 0x4080
442 #define Ebits 8
443 #define Frac_mask  0x7fffff
444 #define Frac_mask1 0xffff007f
445 #define Ten_pmax 24
446 #define Bletch 2
447 #define Bndry_mask  0xffff007f
448 #define Bndry_mask1 0xffff007f
449 #define LSB 0x10000
450 #define Sign_bit 0x8000
451 #define Log2P 1
452 #define Tiny0 0x80
453 #define Tiny1 0
454 #define Quick_max 15
455 #define Int_max 15
456 #endif /* IBM, VAX */
457 #endif /* IEEE_Arith */
458 
459 #ifndef IEEE_Arith
460 #define ROUND_BIASED
461 #endif
462 
463 #ifdef RND_PRODQUOT
464 #define rounded_product(a,b) a = rnd_prod(a, b)
465 #define rounded_quotient(a,b) a = rnd_quot(a, b)
466 #ifdef KR_headers
467 extern double rnd_prod(), rnd_quot();
468 #else
469 extern double rnd_prod(double, double), rnd_quot(double, double);
470 #endif
471 #else
472 #define rounded_product(a,b) a *= b
473 #define rounded_quotient(a,b) a /= b
474 #endif
475 
476 #define Big0 (Frac_mask1 | Exp_msk1*(DBL_MAX_EXP+Bias-1))
477 #define Big1 0xffffffff
478 
479 #ifndef Pack_32
480 #define Pack_32
481 #endif
482 
483 typedef struct BCinfo BCinfo;
484  struct
485 BCinfo { int dp0, dp1, dplen, dsign, e0, inexact, nd, nd0, rounding, scale, uflchk; };
486 
487 #ifdef KR_headers
488 #define FFFFFFFF ((((unsigned long)0xffff)<<16)|(unsigned long)0xffff)
489 #else
490 #define FFFFFFFF 0xffffffffUL
491 #endif
492 
493 #ifdef NO_LONG_LONG
494 #undef ULLong
495 #ifdef Just_16
496 #undef Pack_32
497 /* When Pack_32 is not defined, we store 16 bits per 32-bit Long.
498  * This makes some inner loops simpler and sometimes saves work
499  * during multiplications, but it often seems to make things slightly
500  * slower.  Hence the default is now to store 32 bits per Long.
501  */
502 #endif
503 #else	/* long long available */
504 #ifndef Llong
505 #define Llong long long
506 #endif
507 #ifndef ULLong
508 #define ULLong unsigned Llong
509 #endif
510 #endif /* NO_LONG_LONG */
511 
512 #ifndef MULTIPLE_THREADS
513 #define ACQUIRE_DTOA_LOCK(n)	/*nothing*/
514 #define FREE_DTOA_LOCK(n)	/*nothing*/
515 #endif
516 
517 #define Kmax 7
518 
519 double strtod(const char *s00, char **se);
520 char *dtoa(double d, int mode, int ndigits,
521 			int *decpt, int *sign, char **rve);
522 
523  struct
524 Bigint {
525 	struct Bigint *next;
526 	int k, maxwds, sign, wds;
527 	ULong x[1];
528 	};
529 
530  typedef struct Bigint Bigint;
531 
532  static Bigint *freelist[Kmax+1];
533 
534  static Bigint *
535 Balloc
536 #ifdef KR_headers
537 	(k) int k;
538 #else
539 	(int k)
540 #endif
541 {
542 	int x;
543 	Bigint *rv;
544 #ifndef Omit_Private_Memory
545 	unsigned int len;
546 #endif
547 
548 	ACQUIRE_DTOA_LOCK(0);
549 	/* The k > Kmax case does not need ACQUIRE_DTOA_LOCK(0), */
550 	/* but this case seems very unlikely. */
551 	if (k <= Kmax && freelist[k]) {
552 		rv = freelist[k];
553 		freelist[k] = rv->next;
554 		}
555 	else {
556 		x = 1 << k;
557 #ifdef Omit_Private_Memory
558 		rv = (Bigint *)MALLOC(sizeof(Bigint) + (x-1)*sizeof(ULong));
559 #else
560 		len = (sizeof(Bigint) + (x-1)*sizeof(ULong) + sizeof(double) - 1)
561 			/sizeof(double);
562 		if (k <= Kmax && pmem_next - private_mem + len <= PRIVATE_mem) {
563 			rv = (Bigint*)pmem_next;
564 			pmem_next += len;
565 			}
566 		else
567 			rv = (Bigint*)MALLOC(len*sizeof(double));
568 #endif
569 		rv->k = k;
570 		rv->maxwds = x;
571 		}
572 	FREE_DTOA_LOCK(0);
573 	rv->sign = rv->wds = 0;
574 	return rv;
575 	}
576 
577  static void
578 Bfree
579 #ifdef KR_headers
580 	(v) Bigint *v;
581 #else
582 	(Bigint *v)
583 #endif
584 {
585 	if (v) {
586 		if (v->k > Kmax)
587 #ifdef FREE
588 			FREE((void*)v);
589 #else
590 			free((void*)v);
591 #endif
592 		else {
593 			ACQUIRE_DTOA_LOCK(0);
594 			v->next = freelist[v->k];
595 			freelist[v->k] = v;
596 			FREE_DTOA_LOCK(0);
597 			}
598 		}
599 	}
600 
601 #define Bcopy(x,y) memcpy((char *)&x->sign, (char *)&y->sign, \
602 y->wds*sizeof(Long) + 2*sizeof(int))
603 
604  static Bigint *
605 multadd
606 #ifdef KR_headers
607 	(b, m, a) Bigint *b; int m, a;
608 #else
609 	(Bigint *b, int m, int a)	/* multiply by m and add a */
610 #endif
611 {
612 	int i, wds;
613 #ifdef ULLong
614 	ULong *x;
615 	ULLong carry, y;
616 #else
617 	ULong carry, *x, y;
618 #ifdef Pack_32
619 	ULong xi, z;
620 #endif
621 #endif
622 	Bigint *b1;
623 
624 	wds = b->wds;
625 	x = b->x;
626 	i = 0;
627 	carry = a;
628 	do {
629 #ifdef ULLong
630 		y = *x * (ULLong)m + carry;
631 		carry = y >> 32;
632 		*x++ = y & FFFFFFFF;
633 #else
634 #ifdef Pack_32
635 		xi = *x;
636 		y = (xi & 0xffff) * m + carry;
637 		z = (xi >> 16) * m + (y >> 16);
638 		carry = z >> 16;
639 		*x++ = (z << 16) + (y & 0xffff);
640 #else
641 		y = *x * m + carry;
642 		carry = y >> 16;
643 		*x++ = y & 0xffff;
644 #endif
645 #endif
646 		}
647 		while(++i < wds);
648 	if (carry) {
649 		if (wds >= b->maxwds) {
650 			b1 = Balloc(b->k+1);
651 			Bcopy(b1, b);
652 			Bfree(b);
653 			b = b1;
654 			}
655 		b->x[wds++] = (ULong)carry;
656 		b->wds = wds;
657 		}
658 	return b;
659 	}
660 
661  static Bigint *
662 s2b
663 #ifdef KR_headers
664 	(s, nd0, nd, y9, dplen) CONST char *s; int nd0, nd, dplen; ULong y9;
665 #else
666 	(CONST char *s, int nd0, int nd, ULong y9, int dplen)
667 #endif
668 {
669 	Bigint *b;
670 	int i, k;
671 	Long x, y;
672 
673 	x = (nd + 8) / 9;
674 	for(k = 0, y = 1; x > y; y <<= 1, k++) ;
675 #ifdef Pack_32
676 	b = Balloc(k);
677 	b->x[0] = y9;
678 	b->wds = 1;
679 #else
680 	b = Balloc(k+1);
681 	b->x[0] = y9 & 0xffff;
682 	b->wds = (b->x[1] = y9 >> 16) ? 2 : 1;
683 #endif
684 
685 	i = 9;
686 	if (9 < nd0) {
687 		s += 9;
688 		do b = multadd(b, 10, *s++ - '0');
689 			while(++i < nd0);
690 		s += dplen;
691 		}
692 	else
693 		s += dplen + 9;
694 	for(; i < nd; i++)
695 		b = multadd(b, 10, *s++ - '0');
696 	return b;
697 	}
698 
699  static int
700 hi0bits
701 #ifdef KR_headers
702 	(x) ULong x;
703 #else
704 	(ULong x)
705 #endif
706 {
707 	int k = 0;
708 
709 	if (!(x & 0xffff0000)) {
710 		k = 16;
711 		x <<= 16;
712 		}
713 	if (!(x & 0xff000000)) {
714 		k += 8;
715 		x <<= 8;
716 		}
717 	if (!(x & 0xf0000000)) {
718 		k += 4;
719 		x <<= 4;
720 		}
721 	if (!(x & 0xc0000000)) {
722 		k += 2;
723 		x <<= 2;
724 		}
725 	if (!(x & 0x80000000)) {
726 		k++;
727 		if (!(x & 0x40000000))
728 			return 32;
729 		}
730 	return k;
731 	}
732 
733  static int
734 lo0bits
735 #ifdef KR_headers
736 	(y) ULong *y;
737 #else
738 	(ULong *y)
739 #endif
740 {
741 	int k;
742 	ULong x = *y;
743 
744 	if (x & 7) {
745 		if (x & 1)
746 			return 0;
747 		if (x & 2) {
748 			*y = x >> 1;
749 			return 1;
750 			}
751 		*y = x >> 2;
752 		return 2;
753 		}
754 	k = 0;
755 	if (!(x & 0xffff)) {
756 		k = 16;
757 		x >>= 16;
758 		}
759 	if (!(x & 0xff)) {
760 		k += 8;
761 		x >>= 8;
762 		}
763 	if (!(x & 0xf)) {
764 		k += 4;
765 		x >>= 4;
766 		}
767 	if (!(x & 0x3)) {
768 		k += 2;
769 		x >>= 2;
770 		}
771 	if (!(x & 1)) {
772 		k++;
773 		x >>= 1;
774 		if (!x)
775 			return 32;
776 		}
777 	*y = x;
778 	return k;
779 	}
780 
781  static Bigint *
782 i2b
783 #ifdef KR_headers
784 	(i) int i;
785 #else
786 	(int i)
787 #endif
788 {
789 	Bigint *b;
790 
791 	b = Balloc(1);
792 	b->x[0] = i;
793 	b->wds = 1;
794 	return b;
795 	}
796 
797  static Bigint *
798 mult
799 #ifdef KR_headers
800 	(a, b) Bigint *a, *b;
801 #else
802 	(Bigint *a, Bigint *b)
803 #endif
804 {
805 	Bigint *c;
806 	int k, wa, wb, wc;
807 	ULong *x, *xa, *xae, *xb, *xbe, *xc, *xc0;
808 	ULong y;
809 #ifdef ULLong
810 	ULLong carry, z;
811 #else
812 	ULong carry, z;
813 #ifdef Pack_32
814 	ULong z2;
815 #endif
816 #endif
817 
818 	if (a->wds < b->wds) {
819 		c = a;
820 		a = b;
821 		b = c;
822 		}
823 	k = a->k;
824 	wa = a->wds;
825 	wb = b->wds;
826 	wc = wa + wb;
827 	if (wc > a->maxwds)
828 		k++;
829 	c = Balloc(k);
830 	for(x = c->x, xa = x + wc; x < xa; x++)
831 		*x = 0;
832 	xa = a->x;
833 	xae = xa + wa;
834 	xb = b->x;
835 	xbe = xb + wb;
836 	xc0 = c->x;
837 #ifdef ULLong
838 	for(; xb < xbe; xc0++) {
839 		y = *xb++;
840 		if (y) {
841 			x = xa;
842 			xc = xc0;
843 			carry = 0;
844 			do {
845 				z = *x++ * (ULLong)y + *xc + carry;
846 				carry = z >> 32;
847 				*xc++ = z & FFFFFFFF;
848 				}
849 				while(x < xae);
850 			*xc = (ULong)carry;
851 			}
852 		}
853 #else
854 #ifdef Pack_32
855 	for(; xb < xbe; xb++, xc0++) {
856 		if (y = *xb & 0xffff) {
857 			x = xa;
858 			xc = xc0;
859 			carry = 0;
860 			do {
861 				z = (*x & 0xffff) * y + (*xc & 0xffff) + carry;
862 				carry = z >> 16;
863 				z2 = (*x++ >> 16) * y + (*xc >> 16) + carry;
864 				carry = z2 >> 16;
865 				Storeinc(xc, z2, z);
866 				}
867 				while(x < xae);
868 			*xc = carry;
869 			}
870 		if (y = *xb >> 16) {
871 			x = xa;
872 			xc = xc0;
873 			carry = 0;
874 			z2 = *xc;
875 			do {
876 				z = (*x & 0xffff) * y + (*xc >> 16) + carry;
877 				carry = z >> 16;
878 				Storeinc(xc, z, z2);
879 				z2 = (*x++ >> 16) * y + (*xc & 0xffff) + carry;
880 				carry = z2 >> 16;
881 				}
882 				while(x < xae);
883 			*xc = z2;
884 			}
885 		}
886 #else
887 	for(; xb < xbe; xc0++) {
888 		if (y = *xb++) {
889 			x = xa;
890 			xc = xc0;
891 			carry = 0;
892 			do {
893 				z = *x++ * y + *xc + carry;
894 				carry = z >> 16;
895 				*xc++ = z & 0xffff;
896 				}
897 				while(x < xae);
898 			*xc = carry;
899 			}
900 		}
901 #endif
902 #endif
903 	for(xc0 = c->x, xc = xc0 + wc; wc > 0 && !*--xc; --wc) ;
904 	c->wds = wc;
905 	return c;
906 	}
907 
908  static Bigint *p5s;
909 
910  static Bigint *
911 pow5mult
912 #ifdef KR_headers
913 	(b, k) Bigint *b; int k;
914 #else
915 	(Bigint *b, int k)
916 #endif
917 {
918 	Bigint *b1, *p5, *p51;
919 	int i;
920 	static int p05[3] = { 5, 25, 125 };
921 
922 	i = k & 3;
923 	if (i)
924 		b = multadd(b, p05[i-1], 0);
925 
926 	if (!(k >>= 2))
927 		return b;
928 	p5 = p5s;
929 	if (!p5) {
930 		/* first time */
931 #ifdef MULTIPLE_THREADS
932 		ACQUIRE_DTOA_LOCK(1);
933 		p5 = p5s;
934 		if (!p5) {
935 			p5 = p5s = i2b(625);
936 			p5->next = 0;
937 			}
938 		FREE_DTOA_LOCK(1);
939 #else
940 		p5 = p5s = i2b(625);
941 		p5->next = 0;
942 #endif
943 		}
944 	for(;;) {
945 		if (k & 1) {
946 			b1 = mult(b, p5);
947 			Bfree(b);
948 			b = b1;
949 			}
950 		if (!(k >>= 1))
951 			break;
952 		p51 = p5->next;
953 		if (!p51) {
954 #ifdef MULTIPLE_THREADS
955 			ACQUIRE_DTOA_LOCK(1);
956 			p51 = p5->next;
957 			if (!p51) {
958 				p51 = p5->next = mult(p5,p5);
959 				p51->next = 0;
960 				}
961 			FREE_DTOA_LOCK(1);
962 #else
963 			p51 = p5->next = mult(p5,p5);
964 			p51->next = 0;
965 #endif
966 			}
967 		p5 = p51;
968 		}
969 	return b;
970 	}
971 
972  static Bigint *
973 lshift
974 #ifdef KR_headers
975 	(b, k) Bigint *b; int k;
976 #else
977 	(Bigint *b, int k)
978 #endif
979 {
980 	int i, k1, n, n1;
981 	Bigint *b1;
982 	ULong *x, *x1, *xe, z;
983 
984 #ifdef Pack_32
985 	n = k >> 5;
986 #else
987 	n = k >> 4;
988 #endif
989 	k1 = b->k;
990 	n1 = n + b->wds + 1;
991 	for(i = b->maxwds; n1 > i; i <<= 1)
992 		k1++;
993 	b1 = Balloc(k1);
994 	x1 = b1->x;
995 	for(i = 0; i < n; i++)
996 		*x1++ = 0;
997 	x = b->x;
998 	xe = x + b->wds;
999 #ifdef Pack_32
1000 	if (k &= 0x1f) {
1001 		k1 = 32 - k;
1002 		z = 0;
1003 		do {
1004 			*x1++ = *x << k | z;
1005 			z = *x++ >> k1;
1006 			}
1007 			while(x < xe);
1008 		*x1 = z;
1009 		if (*x1)
1010 			++n1;
1011 		}
1012 #else
1013 	if (k &= 0xf) {
1014 		k1 = 16 - k;
1015 		z = 0;
1016 		do {
1017 			*x1++ = *x << k  & 0xffff | z;
1018 			z = *x++ >> k1;
1019 			}
1020 			while(x < xe);
1021 		if (*x1 = z)
1022 			++n1;
1023 		}
1024 #endif
1025 	else do
1026 		*x1++ = *x++;
1027 		while(x < xe);
1028 	b1->wds = n1 - 1;
1029 	Bfree(b);
1030 	return b1;
1031 	}
1032 
1033  static int
1034 cmp
1035 #ifdef KR_headers
1036 	(a, b) Bigint *a, *b;
1037 #else
1038 	(Bigint *a, Bigint *b)
1039 #endif
1040 {
1041 	ULong *xa, *xa0, *xb, *xb0;
1042 	int i, j;
1043 
1044 	i = a->wds;
1045 	j = b->wds;
1046 #ifdef DEBUG
1047 	if (i > 1 && !a->x[i-1])
1048 		Bug("cmp called with a->x[a->wds-1] == 0");
1049 	if (j > 1 && !b->x[j-1])
1050 		Bug("cmp called with b->x[b->wds-1] == 0");
1051 #endif
1052 	if (i -= j)
1053 		return i;
1054 	xa0 = a->x;
1055 	xa = xa0 + j;
1056 	xb0 = b->x;
1057 	xb = xb0 + j;
1058 	for(;;) {
1059 		if (*--xa != *--xb)
1060 			return *xa < *xb ? -1 : 1;
1061 		if (xa <= xa0)
1062 			break;
1063 		}
1064 	return 0;
1065 	}
1066 
1067  static Bigint *
1068 diff
1069 #ifdef KR_headers
1070 	(a, b) Bigint *a, *b;
1071 #else
1072 	(Bigint *a, Bigint *b)
1073 #endif
1074 {
1075 	Bigint *c;
1076 	int i, wa, wb;
1077 	ULong *xa, *xae, *xb, *xbe, *xc;
1078 #ifdef ULLong
1079 	ULLong borrow, y;
1080 #else
1081 	ULong borrow, y;
1082 #ifdef Pack_32
1083 	ULong z;
1084 #endif
1085 #endif
1086 
1087 	i = cmp(a,b);
1088 	if (!i) {
1089 		c = Balloc(0);
1090 		c->wds = 1;
1091 		c->x[0] = 0;
1092 		return c;
1093 		}
1094 	if (i < 0) {
1095 		c = a;
1096 		a = b;
1097 		b = c;
1098 		i = 1;
1099 		}
1100 	else
1101 		i = 0;
1102 	c = Balloc(a->k);
1103 	c->sign = i;
1104 	wa = a->wds;
1105 	xa = a->x;
1106 	xae = xa + wa;
1107 	wb = b->wds;
1108 	xb = b->x;
1109 	xbe = xb + wb;
1110 	xc = c->x;
1111 	borrow = 0;
1112 #ifdef ULLong
1113 	do {
1114 		y = (ULLong)*xa++ - *xb++ - borrow;
1115 		borrow = y >> 32 & (ULong)1;
1116 		*xc++ = y & FFFFFFFF;
1117 		}
1118 		while(xb < xbe);
1119 	while(xa < xae) {
1120 		y = *xa++ - borrow;
1121 		borrow = y >> 32 & (ULong)1;
1122 		*xc++ = y & FFFFFFFF;
1123 		}
1124 #else
1125 #ifdef Pack_32
1126 	do {
1127 		y = (*xa & 0xffff) - (*xb & 0xffff) - borrow;
1128 		borrow = (y & 0x10000) >> 16;
1129 		z = (*xa++ >> 16) - (*xb++ >> 16) - borrow;
1130 		borrow = (z & 0x10000) >> 16;
1131 		Storeinc(xc, z, y);
1132 		}
1133 		while(xb < xbe);
1134 	while(xa < xae) {
1135 		y = (*xa & 0xffff) - borrow;
1136 		borrow = (y & 0x10000) >> 16;
1137 		z = (*xa++ >> 16) - borrow;
1138 		borrow = (z & 0x10000) >> 16;
1139 		Storeinc(xc, z, y);
1140 		}
1141 #else
1142 	do {
1143 		y = *xa++ - *xb++ - borrow;
1144 		borrow = (y & 0x10000) >> 16;
1145 		*xc++ = y & 0xffff;
1146 		}
1147 		while(xb < xbe);
1148 	while(xa < xae) {
1149 		y = *xa++ - borrow;
1150 		borrow = (y & 0x10000) >> 16;
1151 		*xc++ = y & 0xffff;
1152 		}
1153 #endif
1154 #endif
1155 	while(!*--xc)
1156 		wa--;
1157 	c->wds = wa;
1158 	return c;
1159 	}
1160 
1161  static double
1162 ulp
1163 #ifdef KR_headers
1164 	(x) U *x;
1165 #else
1166 	(U *x)
1167 #endif
1168 {
1169 	Long L;
1170 	U u;
1171 
1172 	L = (word0(x) & Exp_mask) - (P-1)*Exp_msk1;
1173 #ifndef Avoid_Underflow
1174 #ifndef Sudden_Underflow
1175 	if (L > 0) {
1176 #endif
1177 #endif
1178 #ifdef IBM
1179 		L |= Exp_msk1 >> 4;
1180 #endif
1181 		word0(&u) = L;
1182 		word1(&u) = 0;
1183 #ifndef Avoid_Underflow
1184 #ifndef Sudden_Underflow
1185 		}
1186 	else {
1187 		L = -L >> Exp_shift;
1188 		if (L < Exp_shift) {
1189 			word0(&u) = 0x80000 >> L;
1190 			word1(&u) = 0;
1191 			}
1192 		else {
1193 			word0(&u) = 0;
1194 			L -= Exp_shift;
1195 			word1(&u) = L >= 31 ? 1 : 1 << 31 - L;
1196 			}
1197 		}
1198 #endif
1199 #endif
1200 	return dval(&u);
1201 	}
1202 
1203  static double
1204 b2d
1205 #ifdef KR_headers
1206 	(a, e) Bigint *a; int *e;
1207 #else
1208 	(Bigint *a, int *e)
1209 #endif
1210 {
1211 	ULong *xa, *xa0, w, y, z;
1212 	int k;
1213 	U d;
1214 #ifdef VAX
1215 	ULong d0, d1;
1216 #else
1217 #define d0 word0(&d)
1218 #define d1 word1(&d)
1219 #endif
1220 
1221 	xa0 = a->x;
1222 	xa = xa0 + a->wds;
1223 	y = *--xa;
1224 #ifdef DEBUG
1225 	if (!y) Bug("zero y in b2d");
1226 #endif
1227 	k = hi0bits(y);
1228 	*e = 32 - k;
1229 #ifdef Pack_32
1230 	if (k < Ebits) {
1231 		d0 = Exp_1 | y >> (Ebits - k);
1232 		w = xa > xa0 ? *--xa : 0;
1233 		d1 = y << ((32-Ebits) + k) | w >> (Ebits - k);
1234 		goto ret_d;
1235 		}
1236 	z = xa > xa0 ? *--xa : 0;
1237 	if (k -= Ebits) {
1238 		d0 = Exp_1 | y << k | z >> (32 - k);
1239 		y = xa > xa0 ? *--xa : 0;
1240 		d1 = z << k | y >> (32 - k);
1241 		}
1242 	else {
1243 		d0 = Exp_1 | y;
1244 		d1 = z;
1245 		}
1246 #else
1247 	if (k < Ebits + 16) {
1248 		z = xa > xa0 ? *--xa : 0;
1249 		d0 = Exp_1 | y << k - Ebits | z >> Ebits + 16 - k;
1250 		w = xa > xa0 ? *--xa : 0;
1251 		y = xa > xa0 ? *--xa : 0;
1252 		d1 = z << k + 16 - Ebits | w << k - Ebits | y >> 16 + Ebits - k;
1253 		goto ret_d;
1254 		}
1255 	z = xa > xa0 ? *--xa : 0;
1256 	w = xa > xa0 ? *--xa : 0;
1257 	k -= Ebits + 16;
1258 	d0 = Exp_1 | y << k + 16 | z << k | w >> 16 - k;
1259 	y = xa > xa0 ? *--xa : 0;
1260 	d1 = w << k + 16 | y << k;
1261 #endif
1262  ret_d:
1263 #ifdef VAX
1264 	word0(&d) = d0 >> 16 | d0 << 16;
1265 	word1(&d) = d1 >> 16 | d1 << 16;
1266 #else
1267 #undef d0
1268 #undef d1
1269 #endif
1270 	return dval(&d);
1271 	}
1272 
1273  static Bigint *
1274 d2b
1275 #ifdef KR_headers
1276 	(d, e, bits) U *d; int *e, *bits;
1277 #else
1278 	(U *d, int *e, int *bits)
1279 #endif
1280 {
1281 	Bigint *b;
1282 	int de, k;
1283 	ULong *x, y, z;
1284 #ifndef Sudden_Underflow
1285 	int i;
1286 #endif
1287 #ifdef VAX
1288 	ULong d0, d1;
1289 	d0 = word0(d) >> 16 | word0(d) << 16;
1290 	d1 = word1(d) >> 16 | word1(d) << 16;
1291 #else
1292 #define d0 word0(d)
1293 #define d1 word1(d)
1294 #endif
1295 
1296 #ifdef Pack_32
1297 	b = Balloc(1);
1298 #else
1299 	b = Balloc(2);
1300 #endif
1301 	x = b->x;
1302 
1303 	z = d0 & Frac_mask;
1304 	d0 &= 0x7fffffff;	/* clear sign bit, which we ignore */
1305 #ifdef Sudden_Underflow
1306 	de = (int)(d0 >> Exp_shift);
1307 #ifndef IBM
1308 	z |= Exp_msk11;
1309 #endif
1310 #else
1311 	de = (int)(d0 >> Exp_shift);
1312 	if (de)
1313 		z |= Exp_msk1;
1314 #endif
1315 #ifdef Pack_32
1316 	y = d1;
1317 	if (y) {
1318 		k = lo0bits(&y);
1319 		if (k) {
1320 			x[0] = y | z << (32 - k);
1321 			z >>= k;
1322 			}
1323 		else
1324 			x[0] = y;
1325 		x[1] = z;
1326 		b->wds = x[1] ? 2 : 1;
1327 #ifndef Sudden_Underflow
1328 		i = b->wds;
1329 #endif
1330 		}
1331 	else {
1332 		k = lo0bits(&z);
1333 		x[0] = z;
1334 #ifndef Sudden_Underflow
1335 		i =
1336 #endif
1337 		    b->wds = 1;
1338 		k += 32;
1339 		}
1340 #else
1341 	if (y = d1) {
1342 		if (k = lo0bits(&y))
1343 			if (k >= 16) {
1344 				x[0] = y | z << 32 - k & 0xffff;
1345 				x[1] = z >> k - 16 & 0xffff;
1346 				x[2] = z >> k;
1347 				i = 2;
1348 				}
1349 			else {
1350 				x[0] = y & 0xffff;
1351 				x[1] = y >> 16 | z << 16 - k & 0xffff;
1352 				x[2] = z >> k & 0xffff;
1353 				x[3] = z >> k+16;
1354 				i = 3;
1355 				}
1356 		else {
1357 			x[0] = y & 0xffff;
1358 			x[1] = y >> 16;
1359 			x[2] = z & 0xffff;
1360 			x[3] = z >> 16;
1361 			i = 3;
1362 			}
1363 		}
1364 	else {
1365 #ifdef DEBUG
1366 		if (!z)
1367 			Bug("Zero passed to d2b");
1368 #endif
1369 		k = lo0bits(&z);
1370 		if (k >= 16) {
1371 			x[0] = z;
1372 			i = 0;
1373 			}
1374 		else {
1375 			x[0] = z & 0xffff;
1376 			x[1] = z >> 16;
1377 			i = 1;
1378 			}
1379 		k += 32;
1380 		}
1381 	while(!x[i])
1382 		--i;
1383 	b->wds = i + 1;
1384 #endif
1385 #ifndef Sudden_Underflow
1386 	if (de) {
1387 #endif
1388 #ifdef IBM
1389 		*e = (de - Bias - (P-1) << 2) + k;
1390 		*bits = 4*P + 8 - k - hi0bits(word0(d) & Frac_mask);
1391 #else
1392 		*e = de - Bias - (P-1) + k;
1393 		*bits = P - k;
1394 #endif
1395 #ifndef Sudden_Underflow
1396 		}
1397 	else {
1398 		*e = de - Bias - (P-1) + 1 + k;
1399 #ifdef Pack_32
1400 		*bits = 32*i - hi0bits(x[i-1]);
1401 #else
1402 		*bits = (i+2)*16 - hi0bits(x[i]);
1403 #endif
1404 		}
1405 #endif
1406 	return b;
1407 	}
1408 #undef d0
1409 #undef d1
1410 
1411  static double
1412 ratio
1413 #ifdef KR_headers
1414 	(a, b) Bigint *a, *b;
1415 #else
1416 	(Bigint *a, Bigint *b)
1417 #endif
1418 {
1419 	U da, db;
1420 	int k, ka, kb;
1421 
1422 	dval(&da) = b2d(a, &ka);
1423 	dval(&db) = b2d(b, &kb);
1424 #ifdef Pack_32
1425 	k = ka - kb + 32*(a->wds - b->wds);
1426 #else
1427 	k = ka - kb + 16*(a->wds - b->wds);
1428 #endif
1429 #ifdef IBM
1430 	if (k > 0) {
1431 		word0(&da) += (k >> 2)*Exp_msk1;
1432 		if (k &= 3)
1433 			dval(&da) *= 1 << k;
1434 		}
1435 	else {
1436 		k = -k;
1437 		word0(&db) += (k >> 2)*Exp_msk1;
1438 		if (k &= 3)
1439 			dval(&db) *= 1 << k;
1440 		}
1441 #else
1442 	if (k > 0)
1443 		word0(&da) += k*Exp_msk1;
1444 	else {
1445 		k = -k;
1446 		word0(&db) += k*Exp_msk1;
1447 		}
1448 #endif
1449 	return dval(&da) / dval(&db);
1450 	}
1451 
1452  static CONST double
1453 tens[] = {
1454 		1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9,
1455 		1e10, 1e11, 1e12, 1e13, 1e14, 1e15, 1e16, 1e17, 1e18, 1e19,
1456 		1e20, 1e21, 1e22
1457 #ifdef VAX
1458 		, 1e23, 1e24
1459 #endif
1460 		};
1461 
1462  static CONST double
1463 #ifdef IEEE_Arith
1464 bigtens[] = { 1e16, 1e32, 1e64, 1e128, 1e256 };
1465 static CONST double tinytens[] = { 1e-16, 1e-32, 1e-64, 1e-128,
1466 #ifdef Avoid_Underflow
1467 		9007199254740992.*9007199254740992.e-256
1468 		/* = 2^106 * 1e-256 */
1469 #else
1470 		1e-256
1471 #endif
1472 		};
1473 /* The factor of 2^53 in tinytens[4] helps us avoid setting the underflow */
1474 /* flag unnecessarily.  It leads to a song and dance at the end of strtod. */
1475 #define Scale_Bit 0x10
1476 #define n_bigtens 5
1477 #else
1478 #ifdef IBM
1479 bigtens[] = { 1e16, 1e32, 1e64 };
1480 static CONST double tinytens[] = { 1e-16, 1e-32, 1e-64 };
1481 #define n_bigtens 3
1482 #else
1483 bigtens[] = { 1e16, 1e32 };
1484 static CONST double tinytens[] = { 1e-16, 1e-32 };
1485 #define n_bigtens 2
1486 #endif
1487 #endif
1488 
1489 #undef Need_Hexdig
1490 #ifdef INFNAN_CHECK
1491 #ifndef No_Hex_NaN
1492 #define Need_Hexdig
1493 #endif
1494 #endif
1495 
1496 #ifndef Need_Hexdig
1497 #ifndef NO_HEX_FP
1498 #define Need_Hexdig
1499 #endif
1500 #endif
1501 
1502 #ifdef Need_Hexdig /*{*/
1503 static unsigned char hexdig[256];
1504 
1505  static void
1506 #ifdef KR_headers
1507 htinit(h, s, inc) unsigned char *h; unsigned char *s; int inc;
1508 #else
1509 htinit(unsigned char *h, unsigned char *s, int inc)
1510 #endif
1511 {
1512 	int i, j;
1513 	for(i = 0; (j = s[i]) !=0; i++)
1514 		h[j] = (unsigned char)(i + inc);
1515 	}
1516 
1517  static void
1518 #ifdef KR_headers
hexdig_init()1519 hexdig_init()
1520 #else
1521 hexdig_init(void)
1522 #endif
1523 {
1524 #define USC (unsigned char *)
1525 	htinit(hexdig, USC "0123456789", 0x10);
1526 	htinit(hexdig, USC "abcdef", 0x10 + 10);
1527 	htinit(hexdig, USC "ABCDEF", 0x10 + 10);
1528 	}
1529 #endif /* } Need_Hexdig */
1530 
1531 #ifdef INFNAN_CHECK
1532 
1533 #ifndef NAN_WORD0
1534 #define NAN_WORD0 0x7ff80000
1535 #endif
1536 
1537 #ifndef NAN_WORD1
1538 #define NAN_WORD1 0
1539 #endif
1540 
1541  static int
1542 match
1543 #ifdef KR_headers
1544 	(sp, t) char **sp, *t;
1545 #else
1546 	(CONST char **sp, CONST char *t)
1547 #endif
1548 {
1549 	int c, d;
1550 	CONST char *s = *sp;
1551 
1552 	for(d = *t++; d; d = *t++) {
1553 		if ((c = *++s) >= 'A' && c <= 'Z')
1554 			c += 'a' - 'A';
1555 		if (c != d)
1556 			return 0;
1557 		}
1558 	*sp = s + 1;
1559 	return 1;
1560 	}
1561 
1562 #ifndef No_Hex_NaN
1563  static void
1564 hexnan
1565 #ifdef KR_headers
1566 	(rvp, sp) U *rvp; CONST char **sp;
1567 #else
1568 	(U *rvp, CONST char **sp)
1569 #endif
1570 {
1571 	ULong c, x[2];
1572 	CONST char *s;
1573 	int c1, havedig, udx0, xshift;
1574 
1575 	if (!hexdig['0'])
1576 		hexdig_init();
1577 	x[0] = x[1] = 0;
1578 	havedig = xshift = 0;
1579 	udx0 = 1;
1580 	s = *sp;
1581 	/* allow optional initial 0x or 0X */
1582 	for(c = *(CONST unsigned char*)(s+1); c && c <= ' '; c = *(CONST unsigned char*)(s+1))
1583 		++s;
1584 	if (s[1] == '0' && (s[2] == 'x' || s[2] == 'X'))
1585 		s += 2;
1586 	for(c = *(CONST unsigned char*)++s; c; c = *(CONST unsigned char*)++s) {
1587 		c1 = hexdig[c];
1588 		if (c1)
1589 			c  = c1 & 0xf;
1590 		else if (c <= ' ') {
1591 			if (udx0 && havedig) {
1592 				udx0 = 0;
1593 				xshift = 1;
1594 				}
1595 			continue;
1596 			}
1597 #ifdef GDTOA_NON_PEDANTIC_NANCHECK
1598 		else if (/*(*/ c == ')' && havedig) {
1599 			*sp = s + 1;
1600 			break;
1601 			}
1602 		else
1603 			return;	/* invalid form: don't change *sp */
1604 #else
1605 		else {
1606 			do {
1607 				if (/*(*/ c == ')') {
1608 					*sp = s + 1;
1609 					break;
1610 					}
1611 				c = *++s;
1612 				} while(c);
1613 			break;
1614 			}
1615 #endif
1616 		havedig = 1;
1617 		if (xshift) {
1618 			xshift = 0;
1619 			x[0] = x[1];
1620 			x[1] = 0;
1621 			}
1622 		if (udx0)
1623 			x[0] = (x[0] << 4) | (x[1] >> 28);
1624 		x[1] = (x[1] << 4) | c;
1625 		}
1626 	if ((x[0] &= 0xfffff) || x[1]) {
1627 		word0(rvp) = Exp_mask | x[0];
1628 		word1(rvp) = x[1];
1629 		}
1630 	}
1631 #endif /*No_Hex_NaN*/
1632 #endif /* INFNAN_CHECK */
1633 
1634 #ifdef Pack_32
1635 #define ULbits 32
1636 #define kshift 5
1637 #define kmask 31
1638 #else
1639 #define ULbits 16
1640 #define kshift 4
1641 #define kmask 15
1642 #endif
1643 #ifndef NO_HEX_FP /*{*/
1644 
1645  static void
1646 #ifdef KR_headers
1647 rshift(b, k) Bigint *b; int k;
1648 #else
1649 rshift(Bigint *b, int k)
1650 #endif
1651 {
1652 	ULong *x, *x1, *xe, y;
1653 	int n;
1654 
1655 	x = x1 = b->x;
1656 	n = k >> kshift;
1657 	if (n < b->wds) {
1658 		xe = x + b->wds;
1659 		x += n;
1660 		if (k &= kmask) {
1661 			n = 32 - k;
1662 			y = *x++ >> k;
1663 			while(x < xe) {
1664 				*x1++ = (y | (*x << n)) & 0xffffffff;
1665 				y = *x++ >> k;
1666 				}
1667 			if ((*x1 = y) !=0)
1668 				x1++;
1669 			}
1670 		else
1671 			while(x < xe)
1672 				*x1++ = *x++;
1673 		}
1674 	if ((b->wds = x1 - b->x) == 0)
1675 		b->x[0] = 0;
1676 	}
1677 
1678  static ULong
1679 #ifdef KR_headers
1680 any_on(b, k) Bigint *b; int k;
1681 #else
1682 any_on(Bigint *b, int k)
1683 #endif
1684 {
1685 	int n, nwds;
1686 	ULong *x, *x0, x1, x2;
1687 
1688 	x = b->x;
1689 	nwds = b->wds;
1690 	n = k >> kshift;
1691 	if (n > nwds)
1692 		n = nwds;
1693 	else if (n < nwds && (k &= kmask)) {
1694 		x1 = x2 = x[n];
1695 		x1 >>= k;
1696 		x1 <<= k;
1697 		if (x1 != x2)
1698 			return 1;
1699 		}
1700 	x0 = x;
1701 	x += n;
1702 	while(x > x0)
1703 		if (*--x)
1704 			return 1;
1705 	return 0;
1706 	}
1707 
1708 enum {	/* rounding values: same as FLT_ROUNDS */
1709 	Round_zero = 0,
1710 	Round_near = 1,
1711 	Round_up = 2,
1712 	Round_down = 3
1713 	};
1714 
1715  static Bigint *
1716 #ifdef KR_headers
1717 increment(b) Bigint *b;
1718 #else
1719 increment(Bigint *b)
1720 #endif
1721 {
1722 	ULong *x, *xe;
1723 	Bigint *b1;
1724 
1725 	x = b->x;
1726 	xe = x + b->wds;
1727 	do {
1728 		if (*x < (ULong)0xffffffffL) {
1729 			++*x;
1730 			return b;
1731 			}
1732 		*x++ = 0;
1733 		} while(x < xe);
1734 	{
1735 		if (b->wds >= b->maxwds) {
1736 			b1 = Balloc(b->k+1);
1737 			Bcopy(b1,b);
1738 			Bfree(b);
1739 			b = b1;
1740 			}
1741 		b->x[b->wds++] = 1;
1742 		}
1743 	return b;
1744 	}
1745 
1746  void
1747 #ifdef KR_headers
1748 gethex(sp, rvp, rounding, sign)
1749 	CONST char **sp; U *rvp; int rounding, sign;
1750 #else
1751 gethex( CONST char **sp, U *rvp, int rounding, int sign)
1752 #endif
1753 {
1754 	Bigint *b;
1755 	CONST unsigned char *decpt, *s0, *s, *s1;
1756 	Long e, e1;
1757 	ULong L, lostbits, *x;
1758 	int big, denorm, esign, havedig, k, n, nbits, up, zret;
1759 #ifdef IBM
1760 	int j;
1761 #endif
1762 	enum {
1763 #ifdef IEEE_Arith /*{{*/
1764 		emax = 0x7fe - Bias - P + 1,
1765 		emin = Emin - P + 1
1766 #else /*}{*/
1767 		emin = Emin - P,
1768 #ifdef VAX
1769 		emax = 0x7ff - Bias - P + 1
1770 #endif
1771 #ifdef IBM
1772 		emax = 0x7f - Bias - P
1773 #endif
1774 #endif /*}}*/
1775 		};
1776 #ifdef USE_LOCALE
1777 	int i;
1778 #ifdef NO_LOCALE_CACHE
1779 	const unsigned char *decimalpoint = (unsigned char*)
1780 		localeconv()->decimal_point;
1781 #else
1782 	const unsigned char *decimalpoint;
1783 	static unsigned char *decimalpoint_cache;
1784 	if (!(s0 = decimalpoint_cache)) {
1785 		s0 = (unsigned char*)localeconv()->decimal_point;
1786 		if ((decimalpoint_cache = (unsigned char*)
1787 				MALLOC(strlen((CONST char*)s0) + 1))) {
1788 			strcpy((char*)decimalpoint_cache, (CONST char*)s0);
1789 			s0 = decimalpoint_cache;
1790 			}
1791 		}
1792 	decimalpoint = s0;
1793 #endif
1794 #endif
1795 
1796 	if (!hexdig['0'])
1797 		hexdig_init();
1798 	havedig = 0;
1799 	s0 = *(CONST unsigned char **)sp + 2;
1800 	while(s0[havedig] == '0')
1801 		havedig++;
1802 	s0 += havedig;
1803 	s = s0;
1804 	decpt = 0;
1805 	zret = 0;
1806 	e = 0;
1807 	if (hexdig[*s])
1808 		havedig++;
1809 	else {
1810 		zret = 1;
1811 #ifdef USE_LOCALE
1812 		for(i = 0; decimalpoint[i]; ++i) {
1813 			if (s[i] != decimalpoint[i])
1814 				goto pcheck;
1815 			}
1816 		decpt = s += i;
1817 #else
1818 		if (*s != '.')
1819 			goto pcheck;
1820 		decpt = ++s;
1821 #endif
1822 		if (!hexdig[*s])
1823 			goto pcheck;
1824 		while(*s == '0')
1825 			s++;
1826 		if (hexdig[*s])
1827 			zret = 0;
1828 		havedig = 1;
1829 		s0 = s;
1830 		}
1831 	while(hexdig[*s])
1832 		s++;
1833 #ifdef USE_LOCALE
1834 	if (*s == *decimalpoint && !decpt) {
1835 		for(i = 1; decimalpoint[i]; ++i) {
1836 			if (s[i] != decimalpoint[i])
1837 				goto pcheck;
1838 			}
1839 		decpt = s += i;
1840 #else
1841 	if (*s == '.' && !decpt) {
1842 		decpt = ++s;
1843 #endif
1844 		while(hexdig[*s])
1845 			s++;
1846 		}/*}*/
1847 	if (decpt)
1848 		e = -(((Long)(s-decpt)) << 2);
1849  pcheck:
1850 	s1 = s;
1851 	big = esign = 0;
1852 	switch(*s) {
1853 	  case 'p':
1854 	  case 'P':
1855 		switch(*++s) {
1856 		  case '-':
1857 			esign = 1;
1858 			/* no break */
1859 		  case '+':
1860 			s++;
1861 		  }
1862 		if ((n = hexdig[*s]) == 0 || n > 0x19) {
1863 			s = s1;
1864 			break;
1865 			}
1866 		e1 = n - 0x10;
1867 		while((n = hexdig[*++s]) !=0 && n <= 0x19) {
1868 			if (e1 & 0xf8000000)
1869 				big = 1;
1870 			e1 = 10*e1 + n - 0x10;
1871 			}
1872 		if (esign)
1873 			e1 = -e1;
1874 		e += e1;
1875 	  }
1876 	*sp = (char*)s;
1877 	if (!havedig)
1878 		*sp = (char*)s0 - 1;
1879 	if (zret)
1880 		goto retz1;
1881 	if (big) {
1882 		if (esign) {
1883 #ifdef IEEE_Arith
1884 			switch(rounding) {
1885 			  case Round_up:
1886 				if (sign)
1887 					break;
1888 				goto ret_tiny;
1889 			  case Round_down:
1890 				if (!sign)
1891 					break;
1892 				goto ret_tiny;
1893 			  }
1894 #endif
1895 			goto retz;
1896 #ifdef IEEE_Arith
1897  ret_tiny:
1898 #ifndef NO_ERRNO
1899 			errno = ERANGE;
1900 #endif
1901 			word0(rvp) = 0;
1902 			word1(rvp) = 1;
1903 			return;
1904 #endif /* IEEE_Arith */
1905 			}
1906 		switch(rounding) {
1907 		  case Round_near:
1908 			goto ovfl1;
1909 		  case Round_up:
1910 			if (!sign)
1911 				goto ovfl1;
1912 			goto ret_big;
1913 		  case Round_down:
1914 			if (sign)
1915 				goto ovfl1;
1916 			goto ret_big;
1917 		  }
1918  ret_big:
1919 		word0(rvp) = Big0;
1920 		word1(rvp) = Big1;
1921 		return;
1922 		}
1923 	n = s1 - s0 - 1;
1924 	for(k = 0; n > (1 << (kshift-2)) - 1; n >>= 1)
1925 		k++;
1926 	b = Balloc(k);
1927 	x = b->x;
1928 	n = 0;
1929 	L = 0;
1930 #ifdef USE_LOCALE
1931 	for(i = 0; decimalpoint[i+1]; ++i);
1932 #endif
1933 	while(s1 > s0) {
1934 #ifdef USE_LOCALE
1935 		if (*--s1 == decimalpoint[i]) {
1936 			s1 -= i;
1937 			continue;
1938 			}
1939 #else
1940 		if (*--s1 == '.')
1941 			continue;
1942 #endif
1943 		if (n == ULbits) {
1944 			*x++ = L;
1945 			L = 0;
1946 			n = 0;
1947 			}
1948 		L |= (hexdig[*s1] & 0x0f) << n;
1949 		n += 4;
1950 		}
1951 	*x++ = L;
1952 	b->wds = n = x - b->x;
1953 	n = ULbits*n - hi0bits(L);
1954 	nbits = Nbits;
1955 	lostbits = 0;
1956 	x = b->x;
1957 	if (n > nbits) {
1958 		n -= nbits;
1959 		if (any_on(b,n)) {
1960 			lostbits = 1;
1961 			k = n - 1;
1962 			if (x[k>>kshift] & 1 << (k & kmask)) {
1963 				lostbits = 2;
1964 				if (k > 0 && any_on(b,k))
1965 					lostbits = 3;
1966 				}
1967 			}
1968 		rshift(b, n);
1969 		e += n;
1970 		}
1971 	else if (n < nbits) {
1972 		n = nbits - n;
1973 		b = lshift(b, n);
1974 		e -= n;
1975 		x = b->x;
1976 		}
1977 	if (e > Emax) {
1978  ovfl:
1979 		Bfree(b);
1980  ovfl1:
1981 #ifndef NO_ERRNO
1982 		errno = ERANGE;
1983 #endif
1984 		word0(rvp) = Exp_mask;
1985 		word1(rvp) = 0;
1986 		return;
1987 		}
1988 	denorm = 0;
1989 	if (e < emin) {
1990 		denorm = 1;
1991 		n = emin - e;
1992 		if (n >= nbits) {
1993 #ifdef IEEE_Arith /*{*/
1994 			switch (rounding) {
1995 			  case Round_near:
1996 				if (n == nbits && (n < 2 || any_on(b,n-1)))
1997 					goto ret_tiny;
1998 				break;
1999 			  case Round_up:
2000 				if (!sign)
2001 					goto ret_tiny;
2002 				break;
2003 			  case Round_down:
2004 				if (sign)
2005 					goto ret_tiny;
2006 			  }
2007 #endif /* } IEEE_Arith */
2008 			Bfree(b);
2009  retz:
2010 #ifndef NO_ERRNO
2011 			errno = ERANGE;
2012 #endif
2013  retz1:
2014 			rvp->d = 0.;
2015 			return;
2016 			}
2017 		k = n - 1;
2018 		if (lostbits)
2019 			lostbits = 1;
2020 		else if (k > 0)
2021 			lostbits = any_on(b,k);
2022 		if (x[k>>kshift] & 1 << (k & kmask))
2023 			lostbits |= 2;
2024 		nbits -= n;
2025 		rshift(b,n);
2026 		e = emin;
2027 		}
2028 	if (lostbits) {
2029 		up = 0;
2030 		switch(rounding) {
2031 		  case Round_zero:
2032 			break;
2033 		  case Round_near:
2034 			if (lostbits & 2
2035 			 && (lostbits & 1) | (x[0] & 1))
2036 				up = 1;
2037 			break;
2038 		  case Round_up:
2039 			up = 1 - sign;
2040 			break;
2041 		  case Round_down:
2042 			up = sign;
2043 		  }
2044 		if (up) {
2045 			k = b->wds;
2046 			b = increment(b);
2047 			x = b->x;
2048 			if (denorm) {
2049 #if 0
2050 				if (nbits == Nbits - 1
2051 				 && x[nbits >> kshift] & 1 << (nbits & kmask))
2052 					denorm = 0; /* not currently used */
2053 #endif
2054 				}
2055 			else if (b->wds > k
2056 			 || ((n = nbits & kmask) !=0
2057 			     && hi0bits(x[k-1]) < 32-n)) {
2058 				rshift(b,1);
2059 				if (++e > Emax)
2060 					goto ovfl;
2061 				}
2062 			}
2063 		}
2064 #ifdef IEEE_Arith
2065 	if (denorm)
2066 		word0(rvp) = b->wds > 1 ? b->x[1] & ~0x100000 : 0;
2067 	else
2068 		word0(rvp) = (b->x[1] & ~0x100000) | ((e + 0x3ff + 52) << 20);
2069 	word1(rvp) = b->x[0];
2070 #endif
2071 #ifdef IBM
2072 	if ((j = e & 3)) {
2073 		k = b->x[0] & ((1 << j) - 1);
2074 		rshift(b,j);
2075 		if (k) {
2076 			switch(rounding) {
2077 			  case Round_up:
2078 				if (!sign)
2079 					increment(b);
2080 				break;
2081 			  case Round_down:
2082 				if (sign)
2083 					increment(b);
2084 				break;
2085 			  case Round_near:
2086 				j = 1 << (j-1);
2087 				if (k & j && ((k & (j-1)) | lostbits))
2088 					increment(b);
2089 			  }
2090 			}
2091 		}
2092 	e >>= 2;
2093 	word0(rvp) = b->x[1] | ((e + 65 + 13) << 24);
2094 	word1(rvp) = b->x[0];
2095 #endif
2096 #ifdef VAX
2097 	/* The next two lines ignore swap of low- and high-order 2 bytes. */
2098 	/* word0(rvp) = (b->x[1] & ~0x800000) | ((e + 129 + 55) << 23); */
2099 	/* word1(rvp) = b->x[0]; */
2100 	word0(rvp) = ((b->x[1] & ~0x800000) >> 16) | ((e + 129 + 55) << 7) | (b->x[1] << 16);
2101 	word1(rvp) = (b->x[0] >> 16) | (b->x[0] << 16);
2102 #endif
2103 	Bfree(b);
2104 	}
2105 #endif /*}!NO_HEX_FP*/
2106 
2107  static int
2108 #ifdef KR_headers
2109 dshift(b, p2) Bigint *b; int p2;
2110 #else
2111 dshift(Bigint *b, int p2)
2112 #endif
2113 {
2114 	int rv = hi0bits(b->x[b->wds-1]) - 4;
2115 	if (p2 > 0)
2116 		rv -= p2;
2117 	return rv & kmask;
2118 	}
2119 
2120  static int
2121 quorem
2122 #ifdef KR_headers
2123 	(b, S) Bigint *b, *S;
2124 #else
2125 	(Bigint *b, Bigint *S)
2126 #endif
2127 {
2128 	int n;
2129 	ULong *bx, *bxe, q, *sx, *sxe;
2130 #ifdef ULLong
2131 	ULLong borrow, carry, y, ys;
2132 #else
2133 	ULong borrow, carry, y, ys;
2134 #ifdef Pack_32
2135 	ULong si, z, zs;
2136 #endif
2137 #endif
2138 
2139 	n = S->wds;
2140 #ifdef DEBUG
2141 	/*debug*/ if (b->wds > n)
2142 	/*debug*/	Bug("oversize b in quorem");
2143 #endif
2144 	if (b->wds < n)
2145 		return 0;
2146 	sx = S->x;
2147 	sxe = sx + --n;
2148 	bx = b->x;
2149 	bxe = bx + n;
2150 	q = *bxe / (*sxe + 1);	/* ensure q <= true quotient */
2151 #ifdef DEBUG
2152 	/*debug*/ if (q > 9)
2153 	/*debug*/	Bug("oversized quotient in quorem");
2154 #endif
2155 	if (q) {
2156 		borrow = 0;
2157 		carry = 0;
2158 		do {
2159 #ifdef ULLong
2160 			ys = *sx++ * (ULLong)q + carry;
2161 			carry = ys >> 32;
2162 			y = *bx - (ys & FFFFFFFF) - borrow;
2163 			borrow = y >> 32 & (ULong)1;
2164 			*bx++ = y & FFFFFFFF;
2165 #else
2166 #ifdef Pack_32
2167 			si = *sx++;
2168 			ys = (si & 0xffff) * q + carry;
2169 			zs = (si >> 16) * q + (ys >> 16);
2170 			carry = zs >> 16;
2171 			y = (*bx & 0xffff) - (ys & 0xffff) - borrow;
2172 			borrow = (y & 0x10000) >> 16;
2173 			z = (*bx >> 16) - (zs & 0xffff) - borrow;
2174 			borrow = (z & 0x10000) >> 16;
2175 			Storeinc(bx, z, y);
2176 #else
2177 			ys = *sx++ * q + carry;
2178 			carry = ys >> 16;
2179 			y = *bx - (ys & 0xffff) - borrow;
2180 			borrow = (y & 0x10000) >> 16;
2181 			*bx++ = y & 0xffff;
2182 #endif
2183 #endif
2184 			}
2185 			while(sx <= sxe);
2186 		if (!*bxe) {
2187 			bx = b->x;
2188 			while(--bxe > bx && !*bxe)
2189 				--n;
2190 			b->wds = n;
2191 			}
2192 		}
2193 	if (cmp(b, S) >= 0) {
2194 		q++;
2195 		borrow = 0;
2196 		carry = 0;
2197 		bx = b->x;
2198 		sx = S->x;
2199 		do {
2200 #ifdef ULLong
2201 			ys = *sx++ + carry;
2202 			carry = ys >> 32;
2203 			y = *bx - (ys & FFFFFFFF) - borrow;
2204 			borrow = y >> 32 & (ULong)1;
2205 			*bx++ = y & FFFFFFFF;
2206 #else
2207 #ifdef Pack_32
2208 			si = *sx++;
2209 			ys = (si & 0xffff) + carry;
2210 			zs = (si >> 16) + (ys >> 16);
2211 			carry = zs >> 16;
2212 			y = (*bx & 0xffff) - (ys & 0xffff) - borrow;
2213 			borrow = (y & 0x10000) >> 16;
2214 			z = (*bx >> 16) - (zs & 0xffff) - borrow;
2215 			borrow = (z & 0x10000) >> 16;
2216 			Storeinc(bx, z, y);
2217 #else
2218 			ys = *sx++ + carry;
2219 			carry = ys >> 16;
2220 			y = *bx - (ys & 0xffff) - borrow;
2221 			borrow = (y & 0x10000) >> 16;
2222 			*bx++ = y & 0xffff;
2223 #endif
2224 #endif
2225 			}
2226 			while(sx <= sxe);
2227 		bx = b->x;
2228 		bxe = bx + n;
2229 		if (!*bxe) {
2230 			while(--bxe > bx && !*bxe)
2231 				--n;
2232 			b->wds = n;
2233 			}
2234 		}
2235 	return q;
2236 	}
2237 
2238 #ifndef NO_STRTOD_BIGCOMP
2239 
2240  static void
2241 bigcomp
2242 #ifdef KR_headers
2243 	(rv, s0, bc)
2244 	U *rv; CONST char *s0; BCinfo *bc;
2245 #else
2246 	(U *rv, CONST char *s0, BCinfo *bc)
2247 #endif
2248 {
2249 	Bigint *b, *d;
2250 	int b2, bbits, d2, dd, dig, dsign, i, j, nd, nd0, p2, p5, speccase;
2251 
2252 	dsign = bc->dsign;
2253 	nd = bc->nd;
2254 	nd0 = bc->nd0;
2255 	p5 = nd + bc->e0 - 1;
2256 	dd = speccase = 0;
2257 #ifndef Sudden_Underflow
2258 	if (rv->d == 0.) {	/* special case: value near underflow-to-zero */
2259 				/* threshold was rounded to zero */
2260 		b = i2b(1);
2261 		p2 = Emin - P + 1;
2262 		bbits = 1;
2263 #ifdef Avoid_Underflow
2264 		word0(rv) = (P+2) << Exp_shift;
2265 #else
2266 		word1(rv) = 1;
2267 #endif
2268 		i = 0;
2269 #ifdef Honor_FLT_ROUNDS
2270 		if (bc->rounding == 1)
2271 #endif
2272 			{
2273 			speccase = 1;
2274 			--p2;
2275 			dsign = 0;
2276 			goto have_i;
2277 			}
2278 		}
2279 	else
2280 #endif
2281 		b = d2b(rv, &p2, &bbits);
2282 #ifdef Avoid_Underflow
2283 	p2 -= bc->scale;
2284 #endif
2285 	/* floor(log2(rv)) == bbits - 1 + p2 */
2286 	/* Check for denormal case. */
2287 	i = P - bbits;
2288 	if (i > (j = P - Emin - 1 + p2)) {
2289 #ifdef Sudden_Underflow
2290 		Bfree(b);
2291 		b = i2b(1);
2292 		p2 = Emin;
2293 		i = P - 1;
2294 #ifdef Avoid_Underflow
2295 		word0(rv) = (1 + bc->scale) << Exp_shift;
2296 #else
2297 		word0(rv) = Exp_msk1;
2298 #endif
2299 		word1(rv) = 0;
2300 #else
2301 		i = j;
2302 #endif
2303 		}
2304 #ifdef Honor_FLT_ROUNDS
2305 	if (bc->rounding != 1) {
2306 		if (i > 0)
2307 			b = lshift(b, i);
2308 		if (dsign)
2309 			b = increment(b);
2310 		}
2311 	else
2312 #endif
2313 		{
2314 		b = lshift(b, ++i);
2315 		b->x[0] |= 1;
2316 		}
2317 #ifndef Sudden_Underflow
2318  have_i:
2319 #endif
2320 	p2 -= p5 + i;
2321 	d = i2b(1);
2322 	/* Arrange for convenient computation of quotients:
2323 	 * shift left if necessary so divisor has 4 leading 0 bits.
2324 	 */
2325 	if (p5 > 0)
2326 		d = pow5mult(d, p5);
2327 	else if (p5 < 0)
2328 		b = pow5mult(b, -p5);
2329 	if (p2 > 0) {
2330 		b2 = p2;
2331 		d2 = 0;
2332 		}
2333 	else {
2334 		b2 = 0;
2335 		d2 = -p2;
2336 		}
2337 	i = dshift(d, d2);
2338 	if ((b2 += i) > 0)
2339 		b = lshift(b, b2);
2340 	if ((d2 += i) > 0)
2341 		d = lshift(d, d2);
2342 
2343 	/* Now b/d = exactly half-way between the two floating-point values */
2344 	/* on either side of the input string.  Compute first digit of b/d. */
2345 
2346 	dig = quorem(b,d);
2347 	if (!dig) {
2348 		b = multadd(b, 10, 0);	/* very unlikely */
2349 		dig = quorem(b,d);
2350 		}
2351 
2352 	/* Compare b/d with s0 */
2353 
2354 	for(i = 0; i < nd0; ) {
2355 		dd = s0[i++] - '0' - dig;
2356 		if (dd)
2357 			goto ret;
2358 		if (!b->x[0] && b->wds == 1) {
2359 			if (i < nd)
2360 				dd = 1;
2361 			goto ret;
2362 			}
2363 		b = multadd(b, 10, 0);
2364 		dig = quorem(b,d);
2365 		}
2366 	for(j = bc->dp1; i++ < nd;) {
2367 		dd = s0[j++] - '0' - dig;
2368 		if (dd)
2369 			goto ret;
2370 		if (!b->x[0] && b->wds == 1) {
2371 			if (i < nd)
2372 				dd = 1;
2373 			goto ret;
2374 			}
2375 		b = multadd(b, 10, 0);
2376 		dig = quorem(b,d);
2377 		}
2378 	if (b->x[0] || b->wds > 1)
2379 		dd = -1;
2380  ret:
2381 	Bfree(b);
2382 	Bfree(d);
2383 #ifdef Honor_FLT_ROUNDS
2384 	if (bc->rounding != 1) {
2385 		if (dd < 0) {
2386 			if (bc->rounding == 0) {
2387 				if (!dsign)
2388 					goto retlow1;
2389 				}
2390 			else if (dsign)
2391 				goto rethi1;
2392 			}
2393 		else if (dd > 0) {
2394 			if (bc->rounding == 0) {
2395 				if (dsign)
2396 					goto rethi1;
2397 				goto ret1;
2398 				}
2399 			if (!dsign)
2400 				goto rethi1;
2401 			dval(rv) += 2.*ulp(rv);
2402 			}
2403 		else {
2404 			bc->inexact = 0;
2405 			if (dsign)
2406 				goto rethi1;
2407 			}
2408 		}
2409 	else
2410 #endif
2411 	if (speccase) {
2412 		if (dd <= 0)
2413 			rv->d = 0.;
2414 		}
2415 	else if (dd < 0) {
2416 		if (!dsign)	/* does not happen for round-near */
2417 retlow1:
2418 			dval(rv) -= ulp(rv);
2419 		}
2420 	else if (dd > 0) {
2421 		if (dsign) {
2422  rethi1:
2423 			dval(rv) += ulp(rv);
2424 			}
2425 		}
2426 	else {
2427 		/* Exact half-way case:  apply round-even rule. */
2428 		if (word1(rv) & 1) {
2429 			if (dsign)
2430 				goto rethi1;
2431 			goto retlow1;
2432 			}
2433 		}
2434 
2435 #ifdef Honor_FLT_ROUNDS
2436  ret1:
2437 #endif
2438 	return;
2439 	}
2440 #endif /* NO_STRTOD_BIGCOMP */
2441 
2442  double
2443 strtod
2444 #ifdef KR_headers
2445 	(s00, se) CONST char *s00; char **se;
2446 #else
2447 	(CONST char *s00, char **se)
2448 #endif
2449 {
2450 	int bb2, bb5, bbe, bd2, bd5, bbbits, bs2, c, e, e1;
2451 	int esign, i, j, k, nd, nd0, nf, nz, nz0, sign;
2452 	CONST char *s, *s0, *s1;
2453 	double aadj, aadj1;
2454 	Long L;
2455 	U aadj2, adj, rv, rv0;
2456 	ULong y, z;
2457 	BCinfo bc;
2458 	Bigint *bb, *bb1, *bd, *bd0, *bs, *delta;
2459 #ifdef SET_INEXACT
2460 	int oldinexact;
2461 #endif
2462 #ifdef Honor_FLT_ROUNDS /*{*/
2463 #ifdef Trust_FLT_ROUNDS /*{{ only define this if FLT_ROUNDS really works! */
2464 	bc.rounding = Flt_Rounds;
2465 #else /*}{*/
2466 	bc.rounding = 1;
2467 	switch(fegetround()) {
2468 	  case FE_TOWARDZERO:	bc.rounding = 0; break;
2469 	  case FE_UPWARD:	bc.rounding = 2; break;
2470 	  case FE_DOWNWARD:	bc.rounding = 3;
2471 	  }
2472 #endif /*}}*/
2473 #endif /*}*/
2474 #ifdef USE_LOCALE
2475 	CONST char *s2;
2476 #endif
2477 
2478 	sign = nz0 = nz = bc.dplen = bc.uflchk = 0;
2479 	dval(&rv) = 0.;
2480 	for(s = s00;;s++) switch(*s) {
2481 		case '-':
2482 			sign = 1;
2483 			/* no break */
2484 		case '+':
2485 			if (*++s)
2486 				goto break2;
2487 			/* no break */
2488 		case 0:
2489 			goto ret0;
2490 		case '\t':
2491 		case '\n':
2492 		case '\v':
2493 		case '\f':
2494 		case '\r':
2495 		case ' ':
2496 			continue;
2497 		default:
2498 			goto break2;
2499 		}
2500  break2:
2501 	if (*s == '0') {
2502 #ifndef NO_HEX_FP /*{*/
2503 		switch(s[1]) {
2504 		  case 'x':
2505 		  case 'X':
2506 #ifdef Honor_FLT_ROUNDS
2507 			gethex(&s, &rv, bc.rounding, sign);
2508 #else
2509 			gethex(&s, &rv, 1, sign);
2510 #endif
2511 			goto ret;
2512 		  }
2513 #endif /*}*/
2514 		nz0 = 1;
2515 		while(*++s == '0') ;
2516 		if (!*s)
2517 			goto ret;
2518 		}
2519 	s0 = s;
2520 	y = z = 0;
2521 	for(nd = nf = 0; (c = *s) >= '0' && c <= '9'; nd++, s++)
2522 		if (nd < 9)
2523 			y = 10*y + c - '0';
2524 		else if (nd < 16)
2525 			z = 10*z + c - '0';
2526 	nd0 = nd;
2527 	bc.dp0 = bc.dp1 = s - s0;
2528 #ifdef USE_LOCALE
2529 	s1 = localeconv()->decimal_point;
2530 	if (c == *s1) {
2531 		c = '.';
2532 		if (*++s1) {
2533 			s2 = s;
2534 			for(;;) {
2535 				if (*++s2 != *s1) {
2536 					c = 0;
2537 					break;
2538 					}
2539 				if (!*++s1) {
2540 					s = s2;
2541 					break;
2542 					}
2543 				}
2544 			}
2545 		}
2546 #endif
2547 	if (c == '.') {
2548 		c = *++s;
2549 		bc.dp1 = s - s0;
2550 		bc.dplen = bc.dp1 - bc.dp0;
2551 		if (!nd) {
2552 			for(; c == '0'; c = *++s)
2553 				nz++;
2554 			if (c > '0' && c <= '9') {
2555 				s0 = s;
2556 				nf += nz;
2557 				nz = 0;
2558 				goto have_dig;
2559 				}
2560 			goto dig_done;
2561 			}
2562 		for(; c >= '0' && c <= '9'; c = *++s) {
2563  have_dig:
2564 			nz++;
2565 			if (c -= '0') {
2566 				nf += nz;
2567 				for(i = 1; i < nz; i++)
2568 					if (nd++ < 9)
2569 						y *= 10;
2570 					else if (nd <= DBL_DIG + 1)
2571 						z *= 10;
2572 				if (nd++ < 9)
2573 					y = 10*y + c;
2574 				else if (nd <= DBL_DIG + 1)
2575 					z = 10*z + c;
2576 				nz = 0;
2577 				}
2578 			}
2579 		}
2580  dig_done:
2581 	e = 0;
2582 	if (c == 'e' || c == 'E') {
2583 		if (!nd && !nz && !nz0) {
2584 			goto ret0;
2585 			}
2586 		s00 = s;
2587 		esign = 0;
2588 		switch(c = *++s) {
2589 			case '-':
2590 				esign = 1;
2591 			case '+':
2592 				c = *++s;
2593 			}
2594 		if (c >= '0' && c <= '9') {
2595 			while(c == '0')
2596 				c = *++s;
2597 			if (c > '0' && c <= '9') {
2598 				L = c - '0';
2599 				s1 = s;
2600 				while((c = *++s) >= '0' && c <= '9')
2601 					L = 10*L + c - '0';
2602 				if (s - s1 > 8 || L > 19999)
2603 					/* Avoid confusion from exponents
2604 					 * so large that e might overflow.
2605 					 */
2606 					e = 19999; /* safe for 16 bit ints */
2607 				else
2608 					e = (int)L;
2609 				if (esign)
2610 					e = -e;
2611 				}
2612 			else
2613 				e = 0;
2614 			}
2615 		else
2616 			s = s00;
2617 		}
2618 	if (!nd) {
2619 		if (!nz && !nz0) {
2620 #ifdef INFNAN_CHECK
2621 			/* Check for Nan and Infinity */
2622 			if (!bc.dplen)
2623 			 switch(c) {
2624 			  case 'i':
2625 			  case 'I':
2626 				if (match(&s,"nf")) {
2627 					--s;
2628 					if (!match(&s,"inity"))
2629 						++s;
2630 					word0(&rv) = 0x7ff00000;
2631 					word1(&rv) = 0;
2632 					goto ret;
2633 					}
2634 				break;
2635 			  case 'n':
2636 			  case 'N':
2637 				if (match(&s, "an")) {
2638 					word0(&rv) = NAN_WORD0;
2639 					word1(&rv) = NAN_WORD1;
2640 #ifndef No_Hex_NaN
2641 					if (*s == '(') /*)*/
2642 						hexnan(&rv, &s);
2643 #endif
2644 					goto ret;
2645 					}
2646 			  }
2647 #endif /* INFNAN_CHECK */
2648  ret0:
2649 			s = s00;
2650 			sign = 0;
2651 			}
2652 		goto ret;
2653 		}
2654 	bc.e0 = e1 = e -= nf;
2655 
2656 	/* Now we have nd0 digits, starting at s0, followed by a
2657 	 * decimal point, followed by nd-nd0 digits.  The number we're
2658 	 * after is the integer represented by those digits times
2659 	 * 10**e */
2660 
2661 	if (!nd0)
2662 		nd0 = nd;
2663 	k = nd < DBL_DIG + 1 ? nd : DBL_DIG + 1;
2664 	dval(&rv) = y;
2665 	if (k > 9) {
2666 #ifdef SET_INEXACT
2667 		if (k > DBL_DIG)
2668 			oldinexact = get_inexact();
2669 #endif
2670 		dval(&rv) = tens[k - 9] * dval(&rv) + z;
2671 		}
2672 	bd0 = 0;
2673 	if (nd <= DBL_DIG
2674 #ifndef RND_PRODQUOT
2675 #ifndef Honor_FLT_ROUNDS
2676 		&& Flt_Rounds == 1
2677 #endif
2678 #endif
2679 			) {
2680 		if (!e)
2681 			goto ret;
2682 		if (e > 0) {
2683 			if (e <= Ten_pmax) {
2684 #ifdef VAX
2685 				goto vax_ovfl_check;
2686 #else
2687 #ifdef Honor_FLT_ROUNDS
2688 				/* round correctly FLT_ROUNDS = 2 or 3 */
2689 				if (sign) {
2690 					rv.d = -rv.d;
2691 					sign = 0;
2692 					}
2693 #endif
2694 				/* rv = */ rounded_product(dval(&rv), tens[e]);
2695 				goto ret;
2696 #endif
2697 				}
2698 			i = DBL_DIG - nd;
2699 			if (e <= Ten_pmax + i) {
2700 				/* A fancier test would sometimes let us do
2701 				 * this for larger i values.
2702 				 */
2703 #ifdef Honor_FLT_ROUNDS
2704 				/* round correctly FLT_ROUNDS = 2 or 3 */
2705 				if (sign) {
2706 					rv.d = -rv.d;
2707 					sign = 0;
2708 					}
2709 #endif
2710 				e -= i;
2711 				dval(&rv) *= tens[i];
2712 #ifdef VAX
2713 				/* VAX exponent range is so narrow we must
2714 				 * worry about overflow here...
2715 				 */
2716  vax_ovfl_check:
2717 				word0(&rv) -= P*Exp_msk1;
2718 				/* rv = */ rounded_product(dval(&rv), tens[e]);
2719 				if ((word0(&rv) & Exp_mask)
2720 				 > Exp_msk1*(DBL_MAX_EXP+Bias-1-P))
2721 					goto ovfl;
2722 				word0(&rv) += P*Exp_msk1;
2723 #else
2724 				/* rv = */ rounded_product(dval(&rv), tens[e]);
2725 #endif
2726 				goto ret;
2727 				}
2728 			}
2729 #ifndef Inaccurate_Divide
2730 		else if (e >= -Ten_pmax) {
2731 #ifdef Honor_FLT_ROUNDS
2732 			/* round correctly FLT_ROUNDS = 2 or 3 */
2733 			if (sign) {
2734 				rv.d = -rv.d;
2735 				sign = 0;
2736 				}
2737 #endif
2738 			/* rv = */ rounded_quotient(dval(&rv), tens[-e]);
2739 			goto ret;
2740 			}
2741 #endif
2742 		}
2743 	e1 += nd - k;
2744 
2745 #ifdef IEEE_Arith
2746 #ifdef SET_INEXACT
2747 	bc.inexact = 1;
2748 	if (k <= DBL_DIG)
2749 		oldinexact = get_inexact();
2750 #endif
2751 #ifdef Avoid_Underflow
2752 	bc.scale = 0;
2753 #endif
2754 #ifdef Honor_FLT_ROUNDS
2755 	if (bc.rounding >= 2) {
2756 		if (sign)
2757 			bc.rounding = bc.rounding == 2 ? 0 : 2;
2758 		else
2759 			if (bc.rounding != 2)
2760 				bc.rounding = 0;
2761 		}
2762 #endif
2763 #endif /*IEEE_Arith*/
2764 
2765 	/* Get starting approximation = rv * 10**e1 */
2766 
2767 	if (e1 > 0) {
2768 		i = e1 & 15;
2769 		if (i)
2770 			dval(&rv) *= tens[i];
2771 		if (e1 &= ~15) {
2772 			if (e1 > DBL_MAX_10_EXP) {
2773  ovfl:
2774 #ifndef NO_ERRNO
2775 				errno = ERANGE;
2776 #endif
2777 				/* Can't trust HUGE_VAL */
2778 #ifdef IEEE_Arith
2779 #ifdef Honor_FLT_ROUNDS
2780 				switch(bc.rounding) {
2781 				  case 0: /* toward 0 */
2782 				  case 3: /* toward -infinity */
2783 					word0(&rv) = Big0;
2784 					word1(&rv) = Big1;
2785 					break;
2786 				  default:
2787 					word0(&rv) = Exp_mask;
2788 					word1(&rv) = 0;
2789 				  }
2790 #else /*Honor_FLT_ROUNDS*/
2791 				word0(&rv) = Exp_mask;
2792 				word1(&rv) = 0;
2793 #endif /*Honor_FLT_ROUNDS*/
2794 #ifdef SET_INEXACT
2795 				/* set overflow bit */
2796 				dval(&rv0) = 1e300;
2797 				dval(&rv0) *= dval(&rv0);
2798 #endif
2799 #else /*IEEE_Arith*/
2800 				word0(&rv) = Big0;
2801 				word1(&rv) = Big1;
2802 #endif /*IEEE_Arith*/
2803 				goto ret;
2804 				}
2805 			e1 >>= 4;
2806 			for(j = 0; e1 > 1; j++, e1 >>= 1)
2807 				if (e1 & 1)
2808 					dval(&rv) *= bigtens[j];
2809 		/* The last multiplication could overflow. */
2810 			word0(&rv) -= P*Exp_msk1;
2811 			dval(&rv) *= bigtens[j];
2812 			if ((z = word0(&rv) & Exp_mask)
2813 			 > Exp_msk1*(DBL_MAX_EXP+Bias-P))
2814 				goto ovfl;
2815 			if (z > Exp_msk1*(DBL_MAX_EXP+Bias-1-P)) {
2816 				/* set to largest number */
2817 				/* (Can't trust DBL_MAX) */
2818 				word0(&rv) = Big0;
2819 				word1(&rv) = Big1;
2820 				}
2821 			else
2822 				word0(&rv) += P*Exp_msk1;
2823 			}
2824 		}
2825 	else if (e1 < 0) {
2826 		e1 = -e1;
2827 		i = e1 & 15;
2828 		if (i)
2829 			dval(&rv) /= tens[i];
2830 		if (e1 >>= 4) {
2831 			if (e1 >= 1 << n_bigtens)
2832 				goto undfl;
2833 #ifdef Avoid_Underflow
2834 			if (e1 & Scale_Bit)
2835 				bc.scale = 2*P;
2836 			for(j = 0; e1 > 0; j++, e1 >>= 1)
2837 				if (e1 & 1)
2838 					dval(&rv) *= tinytens[j];
2839 			if (bc.scale && (j = 2*P + 1 - ((word0(&rv) & Exp_mask)
2840 						>> Exp_shift)) > 0) {
2841 				/* scaled rv is denormal; clear j low bits */
2842 				if (j >= 32) {
2843 					word1(&rv) = 0;
2844 					if (j >= 53)
2845 					 word0(&rv) = (P+2)*Exp_msk1;
2846 					else
2847 					 word0(&rv) &= 0xffffffff << (j-32);
2848 					}
2849 				else
2850 					word1(&rv) &= 0xffffffff << j;
2851 				}
2852 #else
2853 			for(j = 0; e1 > 1; j++, e1 >>= 1)
2854 				if (e1 & 1)
2855 					dval(&rv) *= tinytens[j];
2856 			/* The last multiplication could underflow. */
2857 			dval(&rv0) = dval(&rv);
2858 			dval(&rv) *= tinytens[j];
2859 			if (!dval(&rv)) {
2860 				dval(&rv) = 2.*dval(&rv0);
2861 				dval(&rv) *= tinytens[j];
2862 #endif
2863 				if (!dval(&rv)) {
2864  undfl:
2865 					dval(&rv) = 0.;
2866 #ifndef NO_ERRNO
2867 					errno = ERANGE;
2868 #endif
2869 					goto ret;
2870 					}
2871 #ifndef Avoid_Underflow
2872 				word0(&rv) = Tiny0;
2873 				word1(&rv) = Tiny1;
2874 				/* The refinement below will clean
2875 				 * this approximation up.
2876 				 */
2877 				}
2878 #endif
2879 			}
2880 		}
2881 
2882 	/* Now the hard part -- adjusting rv to the correct value.*/
2883 
2884 	/* Put digits into bd: true value = bd * 10^e */
2885 
2886 	bc.nd = nd;
2887 #ifndef NO_STRTOD_BIGCOMP
2888 	bc.nd0 = nd0;	/* Only needed if nd > strtod_diglim, but done here */
2889 			/* to silence an erroneous warning about bc.nd0 */
2890 			/* possibly not being initialized. */
2891 	if (nd > strtod_diglim) {
2892 		/* ASSERT(strtod_diglim >= 18); 18 == one more than the */
2893 		/* minimum number of decimal digits to distinguish double values */
2894 		/* in IEEE arithmetic. */
2895 		i = j = 18;
2896 		if (i > nd0)
2897 			j += bc.dplen;
2898 		for(;;) {
2899 			if (--j <= bc.dp1 && j >= bc.dp0)
2900 				j = bc.dp0 - 1;
2901 			if (s0[j] != '0')
2902 				break;
2903 			--i;
2904 			}
2905 		e += nd - i;
2906 		nd = i;
2907 		if (nd0 > nd)
2908 			nd0 = nd;
2909 		if (nd < 9) { /* must recompute y */
2910 			y = 0;
2911 			for(i = 0; i < nd0; ++i)
2912 				y = 10*y + s0[i] - '0';
2913 			for(j = bc.dp1; i < nd; ++i)
2914 				y = 10*y + s0[j++] - '0';
2915 			}
2916 		}
2917 #endif
2918 	bd0 = s2b(s0, nd0, nd, y, bc.dplen);
2919 
2920 	for(;;) {
2921 		bd = Balloc(bd0->k);
2922 		Bcopy(bd, bd0);
2923 		bb = d2b(&rv, &bbe, &bbbits);	/* rv = bb * 2^bbe */
2924 		bs = i2b(1);
2925 
2926 		if (e >= 0) {
2927 			bb2 = bb5 = 0;
2928 			bd2 = bd5 = e;
2929 			}
2930 		else {
2931 			bb2 = bb5 = -e;
2932 			bd2 = bd5 = 0;
2933 			}
2934 		if (bbe >= 0)
2935 			bb2 += bbe;
2936 		else
2937 			bd2 -= bbe;
2938 		bs2 = bb2;
2939 #ifdef Honor_FLT_ROUNDS
2940 		if (bc.rounding != 1)
2941 			bs2++;
2942 #endif
2943 #ifdef Avoid_Underflow
2944 		j = bbe - bc.scale;
2945 		i = j + bbbits - 1;	/* logb(rv) */
2946 		if (i < Emin)	/* denormal */
2947 			j += P - Emin;
2948 		else
2949 			j = P + 1 - bbbits;
2950 #else /*Avoid_Underflow*/
2951 #ifdef Sudden_Underflow
2952 #ifdef IBM
2953 		j = 1 + 4*P - 3 - bbbits + ((bbe + bbbits - 1) & 3);
2954 #else
2955 		j = P + 1 - bbbits;
2956 #endif
2957 #else /*Sudden_Underflow*/
2958 		j = bbe;
2959 		i = j + bbbits - 1;	/* logb(rv) */
2960 		if (i < Emin)	/* denormal */
2961 			j += P - Emin;
2962 		else
2963 			j = P + 1 - bbbits;
2964 #endif /*Sudden_Underflow*/
2965 #endif /*Avoid_Underflow*/
2966 		bb2 += j;
2967 		bd2 += j;
2968 #ifdef Avoid_Underflow
2969 		bd2 += bc.scale;
2970 #endif
2971 		i = bb2 < bd2 ? bb2 : bd2;
2972 		if (i > bs2)
2973 			i = bs2;
2974 		if (i > 0) {
2975 			bb2 -= i;
2976 			bd2 -= i;
2977 			bs2 -= i;
2978 			}
2979 		if (bb5 > 0) {
2980 			bs = pow5mult(bs, bb5);
2981 			bb1 = mult(bs, bb);
2982 			Bfree(bb);
2983 			bb = bb1;
2984 			}
2985 		if (bb2 > 0)
2986 			bb = lshift(bb, bb2);
2987 		if (bd5 > 0)
2988 			bd = pow5mult(bd, bd5);
2989 		if (bd2 > 0)
2990 			bd = lshift(bd, bd2);
2991 		if (bs2 > 0)
2992 			bs = lshift(bs, bs2);
2993 		delta = diff(bb, bd);
2994 		bc.dsign = delta->sign;
2995 		delta->sign = 0;
2996 		i = cmp(delta, bs);
2997 #ifndef NO_STRTOD_BIGCOMP
2998 		if (bc.nd > nd && i <= 0) {
2999 			if (bc.dsign)
3000 				break;	/* Must use bigcomp(). */
3001 #ifdef Honor_FLT_ROUNDS
3002 			if (bc.rounding != 1) {
3003 				if (i < 0)
3004 					break;
3005 				}
3006 			else
3007 #endif
3008 				{
3009 				bc.nd = nd;
3010 				i = -1;	/* Discarded digits make delta smaller. */
3011 				}
3012 			}
3013 #endif
3014 #ifdef Honor_FLT_ROUNDS
3015 		if (bc.rounding != 1) {
3016 			if (i < 0) {
3017 				/* Error is less than an ulp */
3018 				if (!delta->x[0] && delta->wds <= 1) {
3019 					/* exact */
3020 #ifdef SET_INEXACT
3021 					bc.inexact = 0;
3022 #endif
3023 					break;
3024 					}
3025 				if (bc.rounding) {
3026 					if (bc.dsign) {
3027 						adj.d = 1.;
3028 						goto apply_adj;
3029 						}
3030 					}
3031 				else if (!bc.dsign) {
3032 					adj.d = -1.;
3033 					if (!word1(&rv)
3034 					 && !(word0(&rv) & Frac_mask)) {
3035 						y = word0(&rv) & Exp_mask;
3036 #ifdef Avoid_Underflow
3037 						if (!bc.scale || y > 2*P*Exp_msk1)
3038 #else
3039 						if (y)
3040 #endif
3041 						  {
3042 						  delta = lshift(delta,Log2P);
3043 						  if (cmp(delta, bs) <= 0)
3044 							adj.d = -0.5;
3045 						  }
3046 						}
3047  apply_adj:
3048 #ifdef Avoid_Underflow
3049 					if (bc.scale && (y = word0(&rv) & Exp_mask)
3050 						<= 2*P*Exp_msk1)
3051 					  word0(&adj) += (2*P+1)*Exp_msk1 - y;
3052 #else
3053 #ifdef Sudden_Underflow
3054 					if ((word0(&rv) & Exp_mask) <=
3055 							P*Exp_msk1) {
3056 						word0(&rv) += P*Exp_msk1;
3057 						dval(&rv) += adj.d*ulp(dval(&rv));
3058 						word0(&rv) -= P*Exp_msk1;
3059 						}
3060 					else
3061 #endif /*Sudden_Underflow*/
3062 #endif /*Avoid_Underflow*/
3063 					dval(&rv) += adj.d*ulp(&rv);
3064 					}
3065 				break;
3066 				}
3067 			adj.d = ratio(delta, bs);
3068 			if (adj.d < 1.)
3069 				adj.d = 1.;
3070 			if (adj.d <= 0x7ffffffe) {
3071 				/* adj = rounding ? ceil(adj) : floor(adj); */
3072 				y = adj.d;
3073 				if (y != adj.d) {
3074 					if (!((bc.rounding>>1) ^ bc.dsign))
3075 						y++;
3076 					adj.d = y;
3077 					}
3078 				}
3079 #ifdef Avoid_Underflow
3080 			if (bc.scale && (y = word0(&rv) & Exp_mask) <= 2*P*Exp_msk1)
3081 				word0(&adj) += (2*P+1)*Exp_msk1 - y;
3082 #else
3083 #ifdef Sudden_Underflow
3084 			if ((word0(&rv) & Exp_mask) <= P*Exp_msk1) {
3085 				word0(&rv) += P*Exp_msk1;
3086 				adj.d *= ulp(dval(&rv));
3087 				if (bc.dsign)
3088 					dval(&rv) += adj.d;
3089 				else
3090 					dval(&rv) -= adj.d;
3091 				word0(&rv) -= P*Exp_msk1;
3092 				goto cont;
3093 				}
3094 #endif /*Sudden_Underflow*/
3095 #endif /*Avoid_Underflow*/
3096 			adj.d *= ulp(&rv);
3097 			if (bc.dsign) {
3098 				if (word0(&rv) == Big0 && word1(&rv) == Big1)
3099 					goto ovfl;
3100 				dval(&rv) += adj.d;
3101 				}
3102 			else
3103 				dval(&rv) -= adj.d;
3104 			goto cont;
3105 			}
3106 #endif /*Honor_FLT_ROUNDS*/
3107 
3108 		if (i < 0) {
3109 			/* Error is less than half an ulp -- check for
3110 			 * special case of mantissa a power of two.
3111 			 */
3112 			if (bc.dsign || word1(&rv) || word0(&rv) & Bndry_mask
3113 #ifdef IEEE_Arith
3114 #ifdef Avoid_Underflow
3115 			 || (word0(&rv) & Exp_mask) <= (2*P+1)*Exp_msk1
3116 #else
3117 			 || (word0(&rv) & Exp_mask) <= Exp_msk1
3118 #endif
3119 #endif
3120 				) {
3121 #ifdef SET_INEXACT
3122 				if (!delta->x[0] && delta->wds <= 1)
3123 					bc.inexact = 0;
3124 #endif
3125 				break;
3126 				}
3127 			if (!delta->x[0] && delta->wds <= 1) {
3128 				/* exact result */
3129 #ifdef SET_INEXACT
3130 				bc.inexact = 0;
3131 #endif
3132 				break;
3133 				}
3134 			delta = lshift(delta,Log2P);
3135 			if (cmp(delta, bs) > 0)
3136 				goto drop_down;
3137 			break;
3138 			}
3139 		if (i == 0) {
3140 			/* exactly half-way between */
3141 			if (bc.dsign) {
3142 				if ((word0(&rv) & Bndry_mask1) == Bndry_mask1
3143 				 &&  word1(&rv) == (
3144 #ifdef Avoid_Underflow
3145 			(bc.scale && (y = word0(&rv) & Exp_mask) <= 2*P*Exp_msk1)
3146 		? (0xffffffff & (0xffffffff << (2*P+1-(y>>Exp_shift)))) :
3147 #endif
3148 						   0xffffffff)) {
3149 					/*boundary case -- increment exponent*/
3150 					word0(&rv) = (word0(&rv) & Exp_mask)
3151 						+ Exp_msk1
3152 #ifdef IBM
3153 						| Exp_msk1 >> 4
3154 #endif
3155 						;
3156 					word1(&rv) = 0;
3157 #ifdef Avoid_Underflow
3158 					bc.dsign = 0;
3159 #endif
3160 					break;
3161 					}
3162 				}
3163 			else if (!(word0(&rv) & Bndry_mask) && !word1(&rv)) {
3164  drop_down:
3165 				/* boundary case -- decrement exponent */
3166 #ifdef Sudden_Underflow /*{{*/
3167 				L = word0(&rv) & Exp_mask;
3168 #ifdef IBM
3169 				if (L <  Exp_msk1)
3170 #else
3171 #ifdef Avoid_Underflow
3172 				if (L <= (bc.scale ? (2*P+1)*Exp_msk1 : Exp_msk1))
3173 #else
3174 				if (L <= Exp_msk1)
3175 #endif /*Avoid_Underflow*/
3176 #endif /*IBM*/
3177 					{
3178 					if (bc.nd >nd) {
3179 						bc.uflchk = 1;
3180 						break;
3181 						}
3182 					goto undfl;
3183 					}
3184 				L -= Exp_msk1;
3185 #else /*Sudden_Underflow}{*/
3186 #ifdef Avoid_Underflow
3187 				if (bc.scale) {
3188 					L = word0(&rv) & Exp_mask;
3189 					if (L <= (2*P+1)*Exp_msk1) {
3190 						if (L > (P+2)*Exp_msk1)
3191 							/* round even ==> */
3192 							/* accept rv */
3193 							break;
3194 						/* rv = smallest denormal */
3195 						if (bc.nd >nd) {
3196 							bc.uflchk = 1;
3197 							break;
3198 							}
3199 						goto undfl;
3200 						}
3201 					}
3202 #endif /*Avoid_Underflow*/
3203 				L = (word0(&rv) & Exp_mask) - Exp_msk1;
3204 #endif /*Sudden_Underflow}}*/
3205 				word0(&rv) = L | Bndry_mask1;
3206 				word1(&rv) = 0xffffffff;
3207 #ifdef IBM
3208 				goto cont;
3209 #else
3210 				break;
3211 #endif
3212 				}
3213 #ifndef ROUND_BIASED
3214 			if (!(word1(&rv) & LSB))
3215 				break;
3216 #endif
3217 			if (bc.dsign)
3218 				dval(&rv) += ulp(&rv);
3219 #ifndef ROUND_BIASED
3220 			else {
3221 				dval(&rv) -= ulp(&rv);
3222 #ifndef Sudden_Underflow
3223 				if (!dval(&rv)) {
3224 					if (bc.nd >nd) {
3225 						bc.uflchk = 1;
3226 						break;
3227 						}
3228 					goto undfl;
3229 					}
3230 #endif
3231 				}
3232 #ifdef Avoid_Underflow
3233 			bc.dsign = 1 - bc.dsign;
3234 #endif
3235 #endif
3236 			break;
3237 			}
3238 		if ((aadj = ratio(delta, bs)) <= 2.) {
3239 			if (bc.dsign)
3240 				aadj = aadj1 = 1.;
3241 			else if (word1(&rv) || word0(&rv) & Bndry_mask) {
3242 #ifndef Sudden_Underflow
3243 				if (word1(&rv) == Tiny1 && !word0(&rv)) {
3244 					if (bc.nd >nd) {
3245 						bc.uflchk = 1;
3246 						break;
3247 						}
3248 					goto undfl;
3249 					}
3250 #endif
3251 				aadj = 1.;
3252 				aadj1 = -1.;
3253 				}
3254 			else {
3255 				/* special case -- power of FLT_RADIX to be */
3256 				/* rounded down... */
3257 
3258 				if (aadj < 2./FLT_RADIX)
3259 					aadj = 1./FLT_RADIX;
3260 				else
3261 					aadj *= 0.5;
3262 				aadj1 = -aadj;
3263 				}
3264 			}
3265 		else {
3266 			aadj *= 0.5;
3267 			aadj1 = bc.dsign ? aadj : -aadj;
3268 #ifdef Check_FLT_ROUNDS
3269 			switch(bc.rounding) {
3270 				case 2: /* towards +infinity */
3271 					aadj1 -= 0.5;
3272 					break;
3273 				case 0: /* towards 0 */
3274 				case 3: /* towards -infinity */
3275 					aadj1 += 0.5;
3276 				}
3277 #else
3278 			if (Flt_Rounds == 0)
3279 				aadj1 += 0.5;
3280 #endif /*Check_FLT_ROUNDS*/
3281 			}
3282 		y = word0(&rv) & Exp_mask;
3283 
3284 		/* Check for overflow */
3285 
3286 		if (y == Exp_msk1*(DBL_MAX_EXP+Bias-1)) {
3287 			dval(&rv0) = dval(&rv);
3288 			word0(&rv) -= P*Exp_msk1;
3289 			adj.d = aadj1 * ulp(&rv);
3290 			dval(&rv) += adj.d;
3291 			if ((word0(&rv) & Exp_mask) >=
3292 					Exp_msk1*(DBL_MAX_EXP+Bias-P)) {
3293 				if (word0(&rv0) == Big0 && word1(&rv0) == Big1)
3294 					goto ovfl;
3295 				word0(&rv) = Big0;
3296 				word1(&rv) = Big1;
3297 				goto cont;
3298 				}
3299 			else
3300 				word0(&rv) += P*Exp_msk1;
3301 			}
3302 		else {
3303 #ifdef Avoid_Underflow
3304 			if (bc.scale && y <= 2*P*Exp_msk1) {
3305 				if (aadj <= 0x7fffffff) {
3306 					if ((z = (ULong)aadj) <= 0)
3307 						z = 1;
3308 					aadj = z;
3309 					aadj1 = bc.dsign ? aadj : -aadj;
3310 					}
3311 				dval(&aadj2) = aadj1;
3312 				word0(&aadj2) += (2*P+1)*Exp_msk1 - y;
3313 				aadj1 = dval(&aadj2);
3314 				}
3315 			adj.d = aadj1 * ulp(&rv);
3316 			dval(&rv) += adj.d;
3317 #else
3318 #ifdef Sudden_Underflow
3319 			if ((word0(&rv) & Exp_mask) <= P*Exp_msk1) {
3320 				dval(&rv0) = dval(&rv);
3321 				word0(&rv) += P*Exp_msk1;
3322 				adj.d = aadj1 * ulp(&rv);
3323 				dval(&rv) += adj.d;
3324 #ifdef IBM
3325 				if ((word0(&rv) & Exp_mask) <  P*Exp_msk1)
3326 #else
3327 				if ((word0(&rv) & Exp_mask) <= P*Exp_msk1)
3328 #endif
3329 					{
3330 					if (word0(&rv0) == Tiny0
3331 					 && word1(&rv0) == Tiny1) {
3332 						if (bc.nd >nd) {
3333 							bc.uflchk = 1;
3334 							break;
3335 							}
3336 						goto undfl;
3337 						}
3338 					word0(&rv) = Tiny0;
3339 					word1(&rv) = Tiny1;
3340 					goto cont;
3341 					}
3342 				else
3343 					word0(&rv) -= P*Exp_msk1;
3344 				}
3345 			else {
3346 				adj.d = aadj1 * ulp(&rv);
3347 				dval(&rv) += adj.d;
3348 				}
3349 #else /*Sudden_Underflow*/
3350 			/* Compute adj so that the IEEE rounding rules will
3351 			 * correctly round rv + adj in some half-way cases.
3352 			 * If rv * ulp(rv) is denormalized (i.e.,
3353 			 * y <= (P-1)*Exp_msk1), we must adjust aadj to avoid
3354 			 * trouble from bits lost to denormalization;
3355 			 * example: 1.2e-307 .
3356 			 */
3357 			if (y <= (P-1)*Exp_msk1 && aadj > 1.) {
3358 				aadj1 = (double)(int)(aadj + 0.5);
3359 				if (!bc.dsign)
3360 					aadj1 = -aadj1;
3361 				}
3362 			adj.d = aadj1 * ulp(&rv);
3363 			dval(&rv) += adj.d;
3364 #endif /*Sudden_Underflow*/
3365 #endif /*Avoid_Underflow*/
3366 			}
3367 		z = word0(&rv) & Exp_mask;
3368 #ifndef SET_INEXACT
3369 		if (bc.nd == nd) {
3370 #ifdef Avoid_Underflow
3371 		if (!bc.scale)
3372 #endif
3373 		if (y == z) {
3374 			/* Can we stop now? */
3375 			L = (Long)aadj;
3376 			aadj -= L;
3377 			/* The tolerances below are conservative. */
3378 			if (bc.dsign || word1(&rv) || word0(&rv) & Bndry_mask) {
3379 				if (aadj < .4999999 || aadj > .5000001)
3380 					break;
3381 				}
3382 			else if (aadj < .4999999/FLT_RADIX)
3383 				break;
3384 			}
3385 		}
3386 #endif
3387  cont:
3388 		Bfree(bb);
3389 		Bfree(bd);
3390 		Bfree(bs);
3391 		Bfree(delta);
3392 		}
3393 	Bfree(bb);
3394 	Bfree(bd);
3395 	Bfree(bs);
3396 	Bfree(bd0);
3397 	Bfree(delta);
3398 #ifndef NO_STRTOD_BIGCOMP
3399 	if (bc.nd > nd)
3400 		bigcomp(&rv, s0, &bc);
3401 #endif
3402 #ifdef SET_INEXACT
3403 	if (bc.inexact) {
3404 		if (!oldinexact) {
3405 			word0(&rv0) = Exp_1 + (70 << Exp_shift);
3406 			word1(&rv0) = 0;
3407 			dval(&rv0) += 1.;
3408 			}
3409 		}
3410 	else if (!oldinexact)
3411 		clear_inexact();
3412 #endif
3413 #ifdef Avoid_Underflow
3414 	if (bc.scale) {
3415 		word0(&rv0) = Exp_1 - 2*P*Exp_msk1;
3416 		word1(&rv0) = 0;
3417 		dval(&rv) *= dval(&rv0);
3418 #ifndef NO_ERRNO
3419 		/* try to avoid the bug of testing an 8087 register value */
3420 #ifdef IEEE_Arith
3421 		if (!(word0(&rv) & Exp_mask))
3422 #else
3423 		if (word0(&rv) == 0 && word1(&rv) == 0)
3424 #endif
3425 			errno = ERANGE;
3426 #endif
3427 		}
3428 #endif /* Avoid_Underflow */
3429 #ifdef SET_INEXACT
3430 	if (bc.inexact && !(word0(&rv) & Exp_mask)) {
3431 		/* set underflow bit */
3432 		dval(&rv0) = 1e-300;
3433 		dval(&rv0) *= dval(&rv0);
3434 		}
3435 #endif
3436  ret:
3437 	if (se)
3438 		*se = (char *)s;
3439 	return sign ? -dval(&rv) : dval(&rv);
3440 	}
3441 
3442 #ifndef MULTIPLE_THREADS
3443  static char *dtoa_result;
3444 #endif
3445 
3446  static char *
3447 #ifdef KR_headers
3448 rv_alloc(i) int i;
3449 #else
3450 rv_alloc(int i)
3451 #endif
3452 {
3453 	int j, k, *r;
3454 
3455 	j = sizeof(ULong);
3456 	for(k = 0;
3457 		sizeof(Bigint) - sizeof(ULong) - sizeof(int) + j <= (size_t)i;
3458 		j <<= 1)
3459 			k++;
3460 	r = (int*)Balloc(k);
3461 	*r = k;
3462 	return
3463 #ifndef MULTIPLE_THREADS
3464 	dtoa_result =
3465 #endif
3466 		(char *)(r+1);
3467 	}
3468 
3469  static char *
3470 #ifdef KR_headers
3471 nrv_alloc(s, rve, n) char *s, **rve; int n;
3472 #else
3473 nrv_alloc(CONST char *s, char **rve, int n)
3474 #endif
3475 {
3476 	char *rv, *t;
3477 
3478 	t = rv = rv_alloc(n);
3479 	for(*t = *s++; *t; *t = *s++) t++;
3480 	if (rve)
3481 		*rve = t;
3482 	return rv;
3483 	}
3484 
3485 /* freedtoa(s) must be used to free values s returned by dtoa
3486  * when MULTIPLE_THREADS is #defined.  It should be used in all cases,
3487  * but for consistency with earlier versions of dtoa, it is optional
3488  * when MULTIPLE_THREADS is not defined.
3489  */
3490 
3491  void
3492 #ifdef KR_headers
3493 freedtoa(s) char *s;
3494 #else
3495 freedtoa(char *s)
3496 #endif
3497 {
3498 	Bigint *b = (Bigint *)((int *)s - 1);
3499 	b->maxwds = 1 << (b->k = *(int*)b);
3500 	Bfree(b);
3501 #ifndef MULTIPLE_THREADS
3502 	if (s == dtoa_result)
3503 		dtoa_result = 0;
3504 #endif
3505 	}
3506 
3507 /* dtoa for IEEE arithmetic (dmg): convert double to ASCII string.
3508  *
3509  * Inspired by "How to Print Floating-Point Numbers Accurately" by
3510  * Guy L. Steele, Jr. and Jon L. White [Proc. ACM SIGPLAN '90, pp. 112-126].
3511  *
3512  * Modifications:
3513  *	1. Rather than iterating, we use a simple numeric overestimate
3514  *	   to determine k = floor(log10(d)).  We scale relevant
3515  *	   quantities using O(log2(k)) rather than O(k) multiplications.
3516  *	2. For some modes > 2 (corresponding to ecvt and fcvt), we don't
3517  *	   try to generate digits strictly left to right.  Instead, we
3518  *	   compute with fewer bits and propagate the carry if necessary
3519  *	   when rounding the final digit up.  This is often faster.
3520  *	3. Under the assumption that input will be rounded nearest,
3521  *	   mode 0 renders 1e23 as 1e23 rather than 9.999999999999999e22.
3522  *	   That is, we allow equality in stopping tests when the
3523  *	   round-nearest rule will give the same floating-point value
3524  *	   as would satisfaction of the stopping test with strict
3525  *	   inequality.
3526  *	4. We remove common factors of powers of 2 from relevant
3527  *	   quantities.
3528  *	5. When converting floating-point integers less than 1e16,
3529  *	   we use floating-point arithmetic rather than resorting
3530  *	   to multiple-precision integers.
3531  *	6. When asked to produce fewer than 15 digits, we first try
3532  *	   to get by with floating-point arithmetic; we resort to
3533  *	   multiple-precision integer arithmetic only if we cannot
3534  *	   guarantee that the floating-point calculation has given
3535  *	   the correctly rounded result.  For k requested digits and
3536  *	   "uniformly" distributed input, the probability is
3537  *	   something like 10^(k-15) that we must resort to the Long
3538  *	   calculation.
3539  */
3540 
3541  char *
3542 dtoa
3543 #ifdef KR_headers
3544 	(dd, mode, ndigits, decpt, sign, rve)
3545 	double dd; int mode, ndigits, *decpt, *sign; char **rve;
3546 #else
3547 	(double dd, int mode, int ndigits, int *decpt, int *sign, char **rve)
3548 #endif
3549 {
3550  /*	Arguments ndigits, decpt, sign are similar to those
3551 	of ecvt and fcvt; trailing zeros are suppressed from
3552 	the returned string.  If not null, *rve is set to point
3553 	to the end of the return value.  If d is +-Infinity or NaN,
3554 	then *decpt is set to 9999.
3555 
3556 	mode:
3557 		0 ==> shortest string that yields d when read in
3558 			and rounded to nearest.
3559 		1 ==> like 0, but with Steele & White stopping rule;
3560 			e.g. with IEEE P754 arithmetic , mode 0 gives
3561 			1e23 whereas mode 1 gives 9.999999999999999e22.
3562 		2 ==> max(1,ndigits) significant digits.  This gives a
3563 			return value similar to that of ecvt, except
3564 			that trailing zeros are suppressed.
3565 		3 ==> through ndigits past the decimal point.  This
3566 			gives a return value similar to that from fcvt,
3567 			except that trailing zeros are suppressed, and
3568 			ndigits can be negative.
3569 		4,5 ==> similar to 2 and 3, respectively, but (in
3570 			round-nearest mode) with the tests of mode 0 to
3571 			possibly return a shorter string that rounds to d.
3572 			With IEEE arithmetic and compilation with
3573 			-DHonor_FLT_ROUNDS, modes 4 and 5 behave the same
3574 			as modes 2 and 3 when FLT_ROUNDS != 1.
3575 		6-9 ==> Debugging modes similar to mode - 4:  don't try
3576 			fast floating-point estimate (if applicable).
3577 
3578 		Values of mode other than 0-9 are treated as mode 0.
3579 
3580 		Sufficient space is allocated to the return value
3581 		to hold the suppressed trailing zeros.
3582 	*/
3583 
3584 	int bbits, b2, b5, be, dig, i, ieps, ilim, ilim0, ilim1,
3585 		j, j1, k, k0, k_check, leftright, m2, m5, s2, s5,
3586 		spec_case, try_quick;
3587 	Long L;
3588 #ifndef Sudden_Underflow
3589 	int denorm;
3590 	ULong x;
3591 #endif
3592 	Bigint *b, *b1, *delta, *mlo = NULL, *mhi, *S;
3593 	U d2, eps, u;
3594 	double ds;
3595 	char *s, *s0;
3596 #ifdef SET_INEXACT
3597 	int inexact, oldinexact;
3598 #endif
3599 #ifdef Honor_FLT_ROUNDS /*{*/
3600 	int Rounding;
3601 #ifdef Trust_FLT_ROUNDS /*{{ only define this if FLT_ROUNDS really works! */
3602 	Rounding = Flt_Rounds;
3603 #else /*}{*/
3604 	Rounding = 1;
3605 	switch(fegetround()) {
3606 	  case FE_TOWARDZERO:	Rounding = 0; break;
3607 	  case FE_UPWARD:	Rounding = 2; break;
3608 	  case FE_DOWNWARD:	Rounding = 3;
3609 	  }
3610 #endif /*}}*/
3611 #endif /*}*/
3612 
3613 #ifndef MULTIPLE_THREADS
3614 	if (dtoa_result) {
3615 		freedtoa(dtoa_result);
3616 		dtoa_result = 0;
3617 		}
3618 #endif
3619 
3620 	u.d = dd;
3621 	if (word0(&u) & Sign_bit) {
3622 		/* set sign for everything, including 0's and NaNs */
3623 		*sign = 1;
3624 		word0(&u) &= ~Sign_bit;	/* clear sign bit */
3625 		}
3626 	else
3627 		*sign = 0;
3628 
3629 #if defined(IEEE_Arith) + defined(VAX)
3630 #ifdef IEEE_Arith
3631 	if ((word0(&u) & Exp_mask) == Exp_mask)
3632 #else
3633 	if (word0(&u)  == 0x8000)
3634 #endif
3635 		{
3636 		/* Infinity or NaN */
3637 		*decpt = 9999;
3638 #ifdef IEEE_Arith
3639 		if (!word1(&u) && !(word0(&u) & 0xfffff))
3640 			return nrv_alloc("Infinity", rve, 8);
3641 #endif
3642 		return nrv_alloc("NaN", rve, 3);
3643 		}
3644 #endif
3645 #ifdef IBM
3646 	dval(&u) += 0; /* normalize */
3647 #endif
3648 	if (!dval(&u)) {
3649 		*decpt = 1;
3650 		return nrv_alloc("0", rve, 1);
3651 		}
3652 
3653 #ifdef SET_INEXACT
3654 	try_quick = oldinexact = get_inexact();
3655 	inexact = 1;
3656 #endif
3657 #ifdef Honor_FLT_ROUNDS
3658 	if (Rounding >= 2) {
3659 		if (*sign)
3660 			Rounding = Rounding == 2 ? 0 : 2;
3661 		else
3662 			if (Rounding != 2)
3663 				Rounding = 0;
3664 		}
3665 #endif
3666 
3667 	b = d2b(&u, &be, &bbits);
3668 	i = (int)(word0(&u) >> Exp_shift1 & (Exp_mask>>Exp_shift1));
3669 #ifndef Sudden_Underflow
3670 	if (i) {
3671 #endif
3672 		dval(&d2) = dval(&u);
3673 		word0(&d2) &= Frac_mask1;
3674 		word0(&d2) |= Exp_11;
3675 #ifdef IBM
3676 		if (j = 11 - hi0bits(word0(&d2) & Frac_mask))
3677 			dval(&d2) /= 1 << j;
3678 #endif
3679 
3680 		/* log(x)	~=~ log(1.5) + (x-1.5)/1.5
3681 		 * log10(x)	 =  log(x) / log(10)
3682 		 *		~=~ log(1.5)/log(10) + (x-1.5)/(1.5*log(10))
3683 		 * log10(d) = (i-Bias)*log(2)/log(10) + log10(d2)
3684 		 *
3685 		 * This suggests computing an approximation k to log10(d) by
3686 		 *
3687 		 * k = (i - Bias)*0.301029995663981
3688 		 *	+ ( (d2-1.5)*0.289529654602168 + 0.176091259055681 );
3689 		 *
3690 		 * We want k to be too large rather than too small.
3691 		 * The error in the first-order Taylor series approximation
3692 		 * is in our favor, so we just round up the constant enough
3693 		 * to compensate for any error in the multiplication of
3694 		 * (i - Bias) by 0.301029995663981; since |i - Bias| <= 1077,
3695 		 * and 1077 * 0.30103 * 2^-52 ~=~ 7.2e-14,
3696 		 * adding 1e-13 to the constant term more than suffices.
3697 		 * Hence we adjust the constant term to 0.1760912590558.
3698 		 * (We could get a more accurate k by invoking log10,
3699 		 *  but this is probably not worthwhile.)
3700 		 */
3701 
3702 		i -= Bias;
3703 #ifdef IBM
3704 		i <<= 2;
3705 		i += j;
3706 #endif
3707 #ifndef Sudden_Underflow
3708 		denorm = 0;
3709 		}
3710 	else {
3711 		/* d is denormalized */
3712 
3713 		i = bbits + be + (Bias + (P-1) - 1);
3714 		x = i > 32  ? word0(&u) << (64 - i) | word1(&u) >> (i - 32)
3715 			    : word1(&u) << (32 - i);
3716 		dval(&d2) = x;
3717 		word0(&d2) -= 31*Exp_msk1; /* adjust exponent */
3718 		i -= (Bias + (P-1) - 1) + 1;
3719 		denorm = 1;
3720 		}
3721 #endif
3722 	ds = (dval(&d2)-1.5)*0.289529654602168 + 0.1760912590558 + i*0.301029995663981;
3723 	k = (int)ds;
3724 	if (ds < 0. && ds != k)
3725 		k--;	/* want k = floor(ds) */
3726 	k_check = 1;
3727 	if (k >= 0 && k <= Ten_pmax) {
3728 		if (dval(&u) < tens[k])
3729 			k--;
3730 		k_check = 0;
3731 		}
3732 	j = bbits - i - 1;
3733 	if (j >= 0) {
3734 		b2 = 0;
3735 		s2 = j;
3736 		}
3737 	else {
3738 		b2 = -j;
3739 		s2 = 0;
3740 		}
3741 	if (k >= 0) {
3742 		b5 = 0;
3743 		s5 = k;
3744 		s2 += k;
3745 		}
3746 	else {
3747 		b2 -= k;
3748 		b5 = -k;
3749 		s5 = 0;
3750 		}
3751 	if (mode < 0 || mode > 9)
3752 		mode = 0;
3753 
3754 #ifndef SET_INEXACT
3755 #ifdef Check_FLT_ROUNDS
3756 	try_quick = Rounding == 1;
3757 #else
3758 	try_quick = 1;
3759 #endif
3760 #endif /*SET_INEXACT*/
3761 
3762 	if (mode > 5) {
3763 		mode -= 4;
3764 		try_quick = 0;
3765 		}
3766 	leftright = 1;
3767 	ilim = ilim1 = -1;	/* Values for cases 0 and 1; done here to */
3768 				/* silence erroneous "gcc -Wall" warning. */
3769 	switch(mode) {
3770 		case 0:
3771 		case 1:
3772 			i = 18;
3773 			ndigits = 0;
3774 			break;
3775 		case 2:
3776 			leftright = 0;
3777 			/* no break */
3778 		case 4:
3779 			if (ndigits <= 0)
3780 				ndigits = 1;
3781 			ilim = ilim1 = i = ndigits;
3782 			break;
3783 		case 3:
3784 			leftright = 0;
3785 			/* no break */
3786 		case 5:
3787 			i = ndigits + k + 1;
3788 			ilim = i;
3789 			ilim1 = i - 1;
3790 			if (i <= 0)
3791 				i = 1;
3792 		}
3793 	s = s0 = rv_alloc(i);
3794 
3795 #ifdef Honor_FLT_ROUNDS
3796 	if (mode > 1 && Rounding != 1)
3797 		leftright = 0;
3798 #endif
3799 
3800 	if (ilim >= 0 && ilim <= Quick_max && try_quick) {
3801 
3802 		/* Try to get by with floating-point arithmetic. */
3803 
3804 		i = 0;
3805 		dval(&d2) = dval(&u);
3806 		k0 = k;
3807 		ilim0 = ilim;
3808 		ieps = 2; /* conservative */
3809 		if (k > 0) {
3810 			ds = tens[k&0xf];
3811 			j = k >> 4;
3812 			if (j & Bletch) {
3813 				/* prevent overflows */
3814 				j &= Bletch - 1;
3815 				dval(&u) /= bigtens[n_bigtens-1];
3816 				ieps++;
3817 				}
3818 			for(; j; j >>= 1, i++)
3819 				if (j & 1) {
3820 					ieps++;
3821 					ds *= bigtens[i];
3822 					}
3823 			dval(&u) /= ds;
3824 			}
3825 		else {
3826 			j1 = -k;
3827 			if (j1) {
3828 				dval(&u) *= tens[j1 & 0xf];
3829 				for(j = j1 >> 4; j; j >>= 1, i++)
3830 					if (j & 1) {
3831 						ieps++;
3832 						dval(&u) *= bigtens[i];
3833 						}
3834 				}
3835 			}
3836 		if (k_check && dval(&u) < 1. && ilim > 0) {
3837 			if (ilim1 <= 0)
3838 				goto fast_failed;
3839 			ilim = ilim1;
3840 			k--;
3841 			dval(&u) *= 10.;
3842 			ieps++;
3843 			}
3844 		dval(&eps) = ieps*dval(&u) + 7.;
3845 		word0(&eps) -= (P-1)*Exp_msk1;
3846 		if (ilim == 0) {
3847 			S = mhi = 0;
3848 			dval(&u) -= 5.;
3849 			if (dval(&u) > dval(&eps))
3850 				goto one_digit;
3851 			if (dval(&u) < -dval(&eps))
3852 				goto no_digits;
3853 			goto fast_failed;
3854 			}
3855 #ifndef No_leftright
3856 		if (leftright) {
3857 			/* Use Steele & White method of only
3858 			 * generating digits needed.
3859 			 */
3860 			dval(&eps) = 0.5/tens[ilim-1] - dval(&eps);
3861 			for(i = 0;;) {
3862 				L = (long)dval(&u);
3863 				dval(&u) -= L;
3864 				*s++ = '0' + (char)L;
3865 				if (dval(&u) < dval(&eps))
3866 					goto ret1;
3867 				if (1. - dval(&u) < dval(&eps))
3868 					goto bump_up;
3869 				if (++i >= ilim)
3870 					break;
3871 				dval(&eps) *= 10.;
3872 				dval(&u) *= 10.;
3873 				}
3874 			}
3875 		else {
3876 #endif
3877 			/* Generate ilim digits, then fix them up. */
3878 			dval(&eps) *= tens[ilim-1];
3879 			for(i = 1;; i++, dval(&u) *= 10.) {
3880 				L = (Long)(dval(&u));
3881 				if (!(dval(&u) -= L))
3882 					ilim = i;
3883 				*s++ = '0' + (char)L;
3884 				if (i == ilim) {
3885 					if (dval(&u) > 0.5 + dval(&eps))
3886 						goto bump_up;
3887 					else if (dval(&u) < 0.5 - dval(&eps)) {
3888 						while(*--s == '0') {}
3889 						s++;
3890 						goto ret1;
3891 						}
3892 					break;
3893 					}
3894 				}
3895 #ifndef No_leftright
3896 			}
3897 #endif
3898  fast_failed:
3899 		s = s0;
3900 		dval(&u) = dval(&d2);
3901 		k = k0;
3902 		ilim = ilim0;
3903 		}
3904 
3905 	/* Do we have a "small" integer? */
3906 
3907 	if (be >= 0 && k <= Int_max) {
3908 		/* Yes. */
3909 		ds = tens[k];
3910 		if (ndigits < 0 && ilim <= 0) {
3911 			S = mhi = 0;
3912 			if (ilim < 0 || dval(&u) <= 5*ds)
3913 				goto no_digits;
3914 			goto one_digit;
3915 			}
3916 		for(i = 1; i <= k + 1; i++, dval(&u) *= 10.) {
3917 			L = (Long)(dval(&u) / ds);
3918 			dval(&u) -= L*ds;
3919 #ifdef Check_FLT_ROUNDS
3920 			/* If FLT_ROUNDS == 2, L will usually be high by 1 */
3921 			if (dval(&u) < 0) {
3922 				L--;
3923 				dval(&u) += ds;
3924 				}
3925 #endif
3926 			*s++ = '0' + (char)L;
3927 			if (!dval(&u)) {
3928 #ifdef SET_INEXACT
3929 				inexact = 0;
3930 #endif
3931 				break;
3932 				}
3933 			if (i == ilim) {
3934 #ifdef Honor_FLT_ROUNDS
3935 				if (mode > 1)
3936 				switch(Rounding) {
3937 				  case 0: goto ret1;
3938 				  case 2: goto bump_up;
3939 				  }
3940 #endif
3941 				dval(&u) += dval(&u);
3942 				if (dval(&u) > ds || (dval(&u) == ds && L & 1)) {
3943  bump_up:
3944 					while(*--s == '9')
3945 						if (s == s0) {
3946 							k++;
3947 							*s = '0';
3948 							break;
3949 							}
3950 					++*s++;
3951 					}
3952 				break;
3953 				}
3954 			}
3955 		goto ret1;
3956 		}
3957 
3958 	m2 = b2;
3959 	m5 = b5;
3960 	mhi = mlo = 0;
3961 	if (leftright) {
3962 		i =
3963 #ifndef Sudden_Underflow
3964 			denorm ? be + (Bias + (P-1) - 1 + 1) :
3965 #endif
3966 #ifdef IBM
3967 			1 + 4*P - 3 - bbits + ((bbits + be - 1) & 3);
3968 #else
3969 			1 + P - bbits;
3970 #endif
3971 		b2 += i;
3972 		s2 += i;
3973 		mhi = i2b(1);
3974 		}
3975 	if (m2 > 0 && s2 > 0) {
3976 		i = m2 < s2 ? m2 : s2;
3977 		b2 -= i;
3978 		m2 -= i;
3979 		s2 -= i;
3980 		}
3981 	if (b5 > 0) {
3982 		if (leftright) {
3983 			if (m5 > 0) {
3984 				mhi = pow5mult(mhi, m5);
3985 				b1 = mult(mhi, b);
3986 				Bfree(b);
3987 				b = b1;
3988 				}
3989 			j = b5 - m5;
3990 			if (j)
3991 				b = pow5mult(b, j);
3992 			}
3993 		else
3994 			b = pow5mult(b, b5);
3995 		}
3996 	S = i2b(1);
3997 	if (s5 > 0)
3998 		S = pow5mult(S, s5);
3999 
4000 	/* Check for special case that d is a normalized power of 2. */
4001 
4002 	spec_case = 0;
4003 	if ((mode < 2 || leftright)
4004 #ifdef Honor_FLT_ROUNDS
4005 			&& Rounding == 1
4006 #endif
4007 				) {
4008 		if (!word1(&u) && !(word0(&u) & Bndry_mask)
4009 #ifndef Sudden_Underflow
4010 		 && word0(&u) & (Exp_mask & ~Exp_msk1)
4011 #endif
4012 				) {
4013 			/* The special case */
4014 			b2 += Log2P;
4015 			s2 += Log2P;
4016 			spec_case = 1;
4017 			}
4018 		}
4019 
4020 	/* Arrange for convenient computation of quotients:
4021 	 * shift left if necessary so divisor has 4 leading 0 bits.
4022 	 *
4023 	 * Perhaps we should just compute leading 28 bits of S once
4024 	 * and for all and pass them and a shift to quorem, so it
4025 	 * can do shifts and ors to compute the numerator for q.
4026 	 */
4027 #ifdef Pack_32
4028 	i = ((s5 ? 32 - hi0bits(S->x[S->wds-1]) : 1) + s2) & 0x1f;
4029 	if (i)
4030 		i = 32 - i;
4031 #define iInc 28
4032 #else
4033 	if (i = ((s5 ? 32 - hi0bits(S->x[S->wds-1]) : 1) + s2) & 0xf)
4034 		i = 16 - i;
4035 #define iInc 12
4036 #endif
4037 	i = dshift(S, s2);
4038 	b2 += i;
4039 	m2 += i;
4040 	s2 += i;
4041 	if (b2 > 0)
4042 		b = lshift(b, b2);
4043 	if (s2 > 0)
4044 		S = lshift(S, s2);
4045 	if (k_check) {
4046 		if (cmp(b,S) < 0) {
4047 			k--;
4048 			b = multadd(b, 10, 0);	/* we botched the k estimate */
4049 			if (leftright)
4050 				mhi = multadd(mhi, 10, 0);
4051 			ilim = ilim1;
4052 			}
4053 		}
4054 	if (ilim <= 0 && (mode == 3 || mode == 5)) {
4055 		if (ilim < 0 || cmp(b,S = multadd(S,5,0)) <= 0) {
4056 			/* no digits, fcvt style */
4057  no_digits:
4058 			k = -1 - ndigits;
4059 			goto ret;
4060 			}
4061  one_digit:
4062 		*s++ = '1';
4063 		k++;
4064 		goto ret;
4065 		}
4066 	if (leftright) {
4067 		if (m2 > 0)
4068 			mhi = lshift(mhi, m2);
4069 
4070 		/* Compute mlo -- check for special case
4071 		 * that d is a normalized power of 2.
4072 		 */
4073 
4074 		mlo = mhi;
4075 		if (spec_case) {
4076 			mhi = Balloc(mhi->k);
4077 			Bcopy(mhi, mlo);
4078 			mhi = lshift(mhi, Log2P);
4079 			}
4080 
4081 		for(i = 1;;i++) {
4082 			dig = quorem(b,S) + '0';
4083 			/* Do we yet have the shortest decimal string
4084 			 * that will round to d?
4085 			 */
4086 			j = cmp(b, mlo);
4087 			delta = diff(S, mhi);
4088 			j1 = delta->sign ? 1 : cmp(b, delta);
4089 			Bfree(delta);
4090 #ifndef ROUND_BIASED
4091 			if (j1 == 0 && mode != 1 && !(word1(&u) & 1)
4092 #ifdef Honor_FLT_ROUNDS
4093 				&& Rounding >= 1
4094 #endif
4095 								   ) {
4096 				if (dig == '9')
4097 					goto round_9_up;
4098 				if (j > 0)
4099 					dig++;
4100 #ifdef SET_INEXACT
4101 				else if (!b->x[0] && b->wds <= 1)
4102 					inexact = 0;
4103 #endif
4104 				*s++ = (char)dig;
4105 				goto ret;
4106 				}
4107 #endif
4108 			if (j < 0 || (j == 0 && mode != 1
4109 #ifndef ROUND_BIASED
4110 							&& !(word1(&u) & 1)
4111 #endif
4112 					)) {
4113 				if (!b->x[0] && b->wds <= 1) {
4114 #ifdef SET_INEXACT
4115 					inexact = 0;
4116 #endif
4117 					goto accept_dig;
4118 					}
4119 #ifdef Honor_FLT_ROUNDS
4120 				if (mode > 1)
4121 				 switch(Rounding) {
4122 				  case 0: goto accept_dig;
4123 				  case 2: goto keep_dig;
4124 				  }
4125 #endif /*Honor_FLT_ROUNDS*/
4126 				if (j1 > 0) {
4127 					b = lshift(b, 1);
4128 					j1 = cmp(b, S);
4129 					if ((j1 > 0 || (j1 == 0 && dig & 1))
4130 					&& dig++ == '9')
4131 						goto round_9_up;
4132 					}
4133  accept_dig:
4134 				*s++ = (char)dig;
4135 				goto ret;
4136 				}
4137 			if (j1 > 0) {
4138 #ifdef Honor_FLT_ROUNDS
4139 				if (!Rounding)
4140 					goto accept_dig;
4141 #endif
4142 				if (dig == '9') { /* possible if i == 1 */
4143  round_9_up:
4144 					*s++ = '9';
4145 					goto roundoff;
4146 					}
4147 				*s++ = (char)dig + 1;
4148 				goto ret;
4149 				}
4150 #ifdef Honor_FLT_ROUNDS
4151  keep_dig:
4152 #endif
4153 			*s++ = (char)dig;
4154 			if (i == ilim)
4155 				break;
4156 			b = multadd(b, 10, 0);
4157 			if (mlo == mhi)
4158 				mlo = mhi = multadd(mhi, 10, 0);
4159 			else {
4160 				mlo = multadd(mlo, 10, 0);
4161 				mhi = multadd(mhi, 10, 0);
4162 				}
4163 			}
4164 		}
4165 	else
4166 		for(i = 1;; i++) {
4167 			dig = quorem(b,S) + '0';
4168 			*s++ = (char)dig;
4169 			if (!b->x[0] && b->wds <= 1) {
4170 #ifdef SET_INEXACT
4171 				inexact = 0;
4172 #endif
4173 				goto ret;
4174 				}
4175 			if (i >= ilim)
4176 				break;
4177 			b = multadd(b, 10, 0);
4178 			}
4179 
4180 	/* Round off last digit */
4181 
4182 #ifdef Honor_FLT_ROUNDS
4183 	switch(Rounding) {
4184 	  case 0: goto trimzeros;
4185 	  case 2: goto roundoff;
4186 	  }
4187 #endif
4188 	b = lshift(b, 1);
4189 	j = cmp(b, S);
4190 	if (j > 0 || (j == 0 && dig & 1)) {
4191  roundoff:
4192 		while(*--s == '9')
4193 			if (s == s0) {
4194 				k++;
4195 				*s++ = '1';
4196 				goto ret;
4197 				}
4198 		++*s++;
4199 		}
4200 	else {
4201 #ifdef Honor_FLT_ROUNDS
4202  trimzeros:
4203 #endif
4204 		while(*--s == '0') {}
4205 		s++;
4206 		}
4207  ret:
4208 	Bfree(S);
4209 	if (mhi) {
4210 		if (mlo && mlo != mhi)
4211 			Bfree(mlo);
4212 		Bfree(mhi);
4213 		}
4214  ret1:
4215 #ifdef SET_INEXACT
4216 	if (inexact) {
4217 		if (!oldinexact) {
4218 			word0(&u) = Exp_1 + (70 << Exp_shift);
4219 			word1(&u) = 0;
4220 			dval(&u) += 1.;
4221 			}
4222 		}
4223 	else if (!oldinexact)
4224 		clear_inexact();
4225 #endif
4226 	Bfree(b);
4227 	*s = 0;
4228 	*decpt = k + 1;
4229 	if (rve)
4230 		*rve = s;
4231 	return s0;
4232 	}
4233 
4234 }  // namespace dmg_fp
4235