• 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.
76  * #define Check_FLT_ROUNDS if FLT_ROUNDS can assume the values 2 or 3
77  *	and Honor_FLT_ROUNDS is not #defined.
78  * #define RND_PRODQUOT to use rnd_prod and rnd_quot (assembly routines
79  *	that use extended-precision instructions to compute rounded
80  *	products and quotients) with IBM.
81  * #define ROUND_BIASED for IEEE-format with biased rounding.
82  * #define Inaccurate_Divide for IEEE-format with correctly rounded
83  *	products but inaccurate quotients, e.g., for Intel i860.
84  * #define NO_LONG_LONG on machines that do not have a "long long"
85  *	integer type (of >= 64 bits).  On such machines, you can
86  *	#define Just_16 to store 16 bits per 32-bit Long when doing
87  *	high-precision integer arithmetic.  Whether this speeds things
88  *	up or slows things down depends on the machine and the number
89  *	being converted.  If long long is available and the name is
90  *	something other than "long long", #define Llong to be the name,
91  *	and if "unsigned Llong" does not work as an unsigned version of
92  *	Llong, #define #ULLong to be the corresponding unsigned type.
93  * #define KR_headers for old-style C function headers.
94  * #define Bad_float_h if your system lacks a float.h or if it does not
95  *	define some or all of DBL_DIG, DBL_MAX_10_EXP, DBL_MAX_EXP,
96  *	FLT_RADIX, FLT_ROUNDS, and DBL_MAX.
97  * #define MALLOC your_malloc, where your_malloc(n) acts like malloc(n)
98  *	if memory is available and otherwise does something you deem
99  *	appropriate.  If MALLOC is undefined, malloc will be invoked
100  *	directly -- and assumed always to succeed.
101  * #define Omit_Private_Memory to omit logic (added Jan. 1998) for making
102  *	memory allocations from a private pool of memory when possible.
103  *	When used, the private pool is PRIVATE_MEM bytes long:  2304 bytes,
104  *	unless #defined to be a different length.  This default length
105  *	suffices to get rid of MALLOC calls except for unusual cases,
106  *	such as decimal-to-binary conversion of a very long string of
107  *	digits.  The longest string dtoa can return is about 751 bytes
108  *	long.  For conversions by strtod of strings of 800 digits and
109  *	all dtoa conversions in single-threaded executions with 8-byte
110  *	pointers, PRIVATE_MEM >= 7400 appears to suffice; with 4-byte
111  *	pointers, PRIVATE_MEM >= 7112 appears adequate.
112  * #define INFNAN_CHECK on IEEE systems to cause strtod to check for
113  *	Infinity and NaN (case insensitively).  On some systems (e.g.,
114  *	some HP systems), it may be necessary to #define NAN_WORD0
115  *	appropriately -- to the most significant word of a quiet NaN.
116  *	(On HP Series 700/800 machines, -DNAN_WORD0=0x7ff40000 works.)
117  *	When INFNAN_CHECK is #defined and No_Hex_NaN is not #defined,
118  *	strtod also accepts (case insensitively) strings of the form
119  *	NaN(x), where x is a string of hexadecimal digits and spaces;
120  *	if there is only one string of hexadecimal digits, it is taken
121  *	for the 52 fraction bits of the resulting NaN; if there are two
122  *	or more strings of hex digits, the first is for the high 20 bits,
123  *	the second and subsequent for the low 32 bits, with intervening
124  *	white space ignored; but if this results in none of the 52
125  *	fraction bits being on (an IEEE Infinity symbol), then NAN_WORD0
126  *	and NAN_WORD1 are used instead.
127  * #define MULTIPLE_THREADS if the system offers preemptively scheduled
128  *	multiple threads.  In this case, you must provide (or suitably
129  *	#define) two locks, acquired by ACQUIRE_DTOA_LOCK(n) and freed
130  *	by FREE_DTOA_LOCK(n) for n = 0 or 1.  (The second lock, accessed
131  *	in pow5mult, ensures lazy evaluation of only one copy of high
132  *	powers of 5; omitting this lock would introduce a small
133  *	probability of wasting memory, but would otherwise be harmless.)
134  *	You must also invoke freedtoa(s) to free the value s returned by
135  *	dtoa.  You may do so whether or not MULTIPLE_THREADS is #defined.
136  * #define NO_IEEE_Scale to disable new (Feb. 1997) logic in strtod that
137  *	avoids underflows on inputs whose result does not underflow.
138  *	If you #define NO_IEEE_Scale on a machine that uses IEEE-format
139  *	floating-point numbers and flushes underflows to zero rather
140  *	than implementing gradual underflow, then you must also #define
141  *	Sudden_Underflow.
142  * #define YES_ALIAS to permit aliasing certain double values with
143  *	arrays of ULongs.  This leads to slightly better code with
144  *	some compilers and was always used prior to 19990916, but it
145  *	is not strictly legal and can cause trouble with aggressively
146  *	optimizing compilers (e.g., gcc 2.95.1 under -O2).
147  * #define USE_LOCALE to use the current locale's decimal_point value.
148  * #define SET_INEXACT if IEEE arithmetic is being used and extra
149  *	computation should be done to set the inexact flag when the
150  *	result is inexact and avoid setting inexact when the result
151  *	is exact.  In this case, dtoa.c must be compiled in
152  *	an environment, perhaps provided by #include "dtoa.c" in a
153  *	suitable wrapper, that defines two functions,
154  *		int get_inexact(void);
155  *		void clear_inexact(void);
156  *	such that get_inexact() returns a nonzero value if the
157  *	inexact bit is already set, and clear_inexact() sets the
158  *	inexact bit to 0.  When SET_INEXACT is #defined, strtod
159  *	also does extra computations to set the underflow and overflow
160  *	flags when appropriate (i.e., when the result is tiny and
161  *	inexact or when it is a numeric value rounded to +-infinity).
162  * #define NO_ERRNO if strtod should not assign errno = ERANGE when
163  *	the result overflows to +-Infinity or underflows to 0.
164  */
165 
166 #ifndef Long
167 #define Long long
168 #endif
169 #ifndef ULong
170 typedef unsigned Long ULong;
171 #endif
172 
173 #ifdef DEBUG
174 #include "stdio.h"
175 #define Bug(x) {fprintf(stderr, "%s\n", x); exit(1);}
176 #endif
177 
178 #include "stdlib.h"
179 #include "string.h"
180 
181 #ifdef USE_LOCALE
182 #include "locale.h"
183 #endif
184 
185 #ifdef MALLOC
186 #ifdef KR_headers
187 extern char *MALLOC();
188 #else
189 extern void *MALLOC(size_t);
190 #endif
191 #else
192 #define MALLOC malloc
193 #endif
194 
195 #ifndef Omit_Private_Memory
196 #ifndef PRIVATE_MEM
197 #define PRIVATE_MEM 2304
198 #endif
199 #define PRIVATE_mem ((PRIVATE_MEM+sizeof(double)-1)/sizeof(double))
200 static double private_mem[PRIVATE_mem], *pmem_next = private_mem;
201 #endif
202 
203 #undef IEEE_Arith
204 #undef Avoid_Underflow
205 #ifdef IEEE_MC68k
206 #define IEEE_Arith
207 #endif
208 #ifdef IEEE_8087
209 #define IEEE_Arith
210 #endif
211 
212 #include "errno.h"
213 
214 #ifdef Bad_float_h
215 
216 #ifdef IEEE_Arith
217 #define DBL_DIG 15
218 #define DBL_MAX_10_EXP 308
219 #define DBL_MAX_EXP 1024
220 #define FLT_RADIX 2
221 #endif /*IEEE_Arith*/
222 
223 #ifdef IBM
224 #define DBL_DIG 16
225 #define DBL_MAX_10_EXP 75
226 #define DBL_MAX_EXP 63
227 #define FLT_RADIX 16
228 #define DBL_MAX 7.2370055773322621e+75
229 #endif
230 
231 #ifdef VAX
232 #define DBL_DIG 16
233 #define DBL_MAX_10_EXP 38
234 #define DBL_MAX_EXP 127
235 #define FLT_RADIX 2
236 #define DBL_MAX 1.7014118346046923e+38
237 #endif
238 
239 #ifndef LONG_MAX
240 #define LONG_MAX 2147483647
241 #endif
242 
243 #else /* ifndef Bad_float_h */
244 #include "float.h"
245 #endif /* Bad_float_h */
246 
247 #ifndef __MATH_H__
248 #include "math.h"
249 #endif
250 
251 #ifdef __cplusplus
252 extern "C" {
253 #endif
254 
255 #ifndef CONST
256 #ifdef KR_headers
257 #define CONST /* blank */
258 #else
259 #define CONST const
260 #endif
261 #endif
262 
263 #if defined(IEEE_8087) + defined(IEEE_MC68k) + defined(VAX) + defined(IBM) != 1
264 Exactly one of IEEE_8087, IEEE_MC68k, VAX, or IBM should be defined.
265 #endif
266 
267 typedef union { double d; ULong L[2]; } U;
268 
269 #ifdef YES_ALIAS
270 #define dval(x) x
271 #ifdef IEEE_8087
272 #define word0(x) ((ULong *)&x)[1]
273 #define word1(x) ((ULong *)&x)[0]
274 #else
275 #define word0(x) ((ULong *)&x)[0]
276 #define word1(x) ((ULong *)&x)[1]
277 #endif
278 #else
279 #ifdef IEEE_8087
280 #define word0(x) ((U*)&x)->L[1]
281 #define word1(x) ((U*)&x)->L[0]
282 #else
283 #define word0(x) ((U*)&x)->L[0]
284 #define word1(x) ((U*)&x)->L[1]
285 #endif
286 #define dval(x) ((U*)&x)->d
287 #endif
288 
289 /* The following definition of Storeinc is appropriate for MIPS processors.
290  * An alternative that might be better on some machines is
291  * #define Storeinc(a,b,c) (*a++ = b << 16 | c & 0xffff)
292  */
293 #if defined(IEEE_8087) + defined(VAX)
294 #define Storeinc(a,b,c) (((unsigned short *)a)[1] = (unsigned short)b, \
295 ((unsigned short *)a)[0] = (unsigned short)c, a++)
296 #else
297 #define Storeinc(a,b,c) (((unsigned short *)a)[0] = (unsigned short)b, \
298 ((unsigned short *)a)[1] = (unsigned short)c, a++)
299 #endif
300 
301 /* #define P DBL_MANT_DIG */
302 /* Ten_pmax = floor(P*log(2)/log(5)) */
303 /* Bletch = (highest power of 2 < DBL_MAX_10_EXP) / 16 */
304 /* Quick_max = floor((P-1)*log(FLT_RADIX)/log(10) - 1) */
305 /* Int_max = floor(P*log(FLT_RADIX)/log(10) - 1) */
306 
307 #ifdef IEEE_Arith
308 #define Exp_shift  20
309 #define Exp_shift1 20
310 #define Exp_msk1    0x100000
311 #define Exp_msk11   0x100000
312 #define Exp_mask  0x7ff00000
313 #define P 53
314 #define Bias 1023
315 #define Emin (-1022)
316 #define Exp_1  0x3ff00000
317 #define Exp_11 0x3ff00000
318 #define Ebits 11
319 #define Frac_mask  0xfffff
320 #define Frac_mask1 0xfffff
321 #define Ten_pmax 22
322 #define Bletch 0x10
323 #define Bndry_mask  0xfffff
324 #define Bndry_mask1 0xfffff
325 #define LSB 1
326 #define Sign_bit 0x80000000
327 #define Log2P 1
328 #define Tiny0 0
329 #define Tiny1 1
330 #define Quick_max 14
331 #define Int_max 14
332 #ifndef NO_IEEE_Scale
333 #define Avoid_Underflow
334 #ifdef Flush_Denorm	/* debugging option */
335 #undef Sudden_Underflow
336 #endif
337 #endif
338 
339 #ifndef Flt_Rounds
340 #ifdef FLT_ROUNDS
341 #define Flt_Rounds FLT_ROUNDS
342 #else
343 #define Flt_Rounds 1
344 #endif
345 #endif /*Flt_Rounds*/
346 
347 #ifdef Honor_FLT_ROUNDS
348 #define Rounding rounding
349 #undef Check_FLT_ROUNDS
350 #define Check_FLT_ROUNDS
351 #else
352 #define Rounding Flt_Rounds
353 #endif
354 
355 #else /* ifndef IEEE_Arith */
356 #undef Check_FLT_ROUNDS
357 #undef Honor_FLT_ROUNDS
358 #undef SET_INEXACT
359 #undef  Sudden_Underflow
360 #define Sudden_Underflow
361 #ifdef IBM
362 #undef Flt_Rounds
363 #define Flt_Rounds 0
364 #define Exp_shift  24
365 #define Exp_shift1 24
366 #define Exp_msk1   0x1000000
367 #define Exp_msk11  0x1000000
368 #define Exp_mask  0x7f000000
369 #define P 14
370 #define Bias 65
371 #define Exp_1  0x41000000
372 #define Exp_11 0x41000000
373 #define Ebits 8	/* exponent has 7 bits, but 8 is the right value in b2d */
374 #define Frac_mask  0xffffff
375 #define Frac_mask1 0xffffff
376 #define Bletch 4
377 #define Ten_pmax 22
378 #define Bndry_mask  0xefffff
379 #define Bndry_mask1 0xffffff
380 #define LSB 1
381 #define Sign_bit 0x80000000
382 #define Log2P 4
383 #define Tiny0 0x100000
384 #define Tiny1 0
385 #define Quick_max 14
386 #define Int_max 15
387 #else /* VAX */
388 #undef Flt_Rounds
389 #define Flt_Rounds 1
390 #define Exp_shift  23
391 #define Exp_shift1 7
392 #define Exp_msk1    0x80
393 #define Exp_msk11   0x800000
394 #define Exp_mask  0x7f80
395 #define P 56
396 #define Bias 129
397 #define Exp_1  0x40800000
398 #define Exp_11 0x4080
399 #define Ebits 8
400 #define Frac_mask  0x7fffff
401 #define Frac_mask1 0xffff007f
402 #define Ten_pmax 24
403 #define Bletch 2
404 #define Bndry_mask  0xffff007f
405 #define Bndry_mask1 0xffff007f
406 #define LSB 0x10000
407 #define Sign_bit 0x8000
408 #define Log2P 1
409 #define Tiny0 0x80
410 #define Tiny1 0
411 #define Quick_max 15
412 #define Int_max 15
413 #endif /* IBM, VAX */
414 #endif /* IEEE_Arith */
415 
416 #ifndef IEEE_Arith
417 #define ROUND_BIASED
418 #endif
419 
420 #ifdef RND_PRODQUOT
421 #define rounded_product(a,b) a = rnd_prod(a, b)
422 #define rounded_quotient(a,b) a = rnd_quot(a, b)
423 #ifdef KR_headers
424 extern double rnd_prod(), rnd_quot();
425 #else
426 extern double rnd_prod(double, double), rnd_quot(double, double);
427 #endif
428 #else
429 #define rounded_product(a,b) a *= b
430 #define rounded_quotient(a,b) a /= b
431 #endif
432 
433 #define Big0 (Frac_mask1 | Exp_msk1*(DBL_MAX_EXP+Bias-1))
434 #define Big1 0xffffffff
435 
436 #ifndef Pack_32
437 #define Pack_32
438 #endif
439 
440 #ifdef KR_headers
441 #define FFFFFFFF ((((unsigned long)0xffff)<<16)|(unsigned long)0xffff)
442 #else
443 #define FFFFFFFF 0xffffffffUL
444 #endif
445 
446 #ifdef NO_LONG_LONG
447 #undef ULLong
448 #ifdef Just_16
449 #undef Pack_32
450 /* When Pack_32 is not defined, we store 16 bits per 32-bit Long.
451  * This makes some inner loops simpler and sometimes saves work
452  * during multiplications, but it often seems to make things slightly
453  * slower.  Hence the default is now to store 32 bits per Long.
454  */
455 #endif
456 #else	/* long long available */
457 #ifndef Llong
458 #define Llong long long
459 #endif
460 #ifndef ULLong
461 #define ULLong unsigned Llong
462 #endif
463 #endif /* NO_LONG_LONG */
464 
465 #ifndef MULTIPLE_THREADS
466 #define ACQUIRE_DTOA_LOCK(n)	/*nothing*/
467 #define FREE_DTOA_LOCK(n)	/*nothing*/
468 #endif
469 
470 #define Kmax 15
471 
472 #ifdef __cplusplus
473 extern "C" double strtod(const char *s00, char **se);
474 extern "C" char *dtoa(double d, int mode, int ndigits,
475 			int *decpt, int *sign, char **rve);
476 #endif
477 
478  struct
479 Bigint {
480 	struct Bigint *next;
481 	int k, maxwds, sign, wds;
482 	ULong x[1];
483 	};
484 
485  typedef struct Bigint Bigint;
486 
487  static Bigint *freelist[Kmax+1];
488 
489  static Bigint *
Balloc(k)490 Balloc
491 #ifdef KR_headers
492 	(k) int k;
493 #else
494 	(int k)
495 #endif
496 {
497 	int x;
498 	Bigint *rv;
499 #ifndef Omit_Private_Memory
500 	unsigned int len;
501 #endif
502 
503 	ACQUIRE_DTOA_LOCK(0);
504         /* The k > Kmax case does not need ACQUIRE_DTOA_LOCK(0). */
505         /* but this case seems very unlikely. */
506 	if (k <= Kmax && (rv = freelist[k])) {
507 		freelist[k] = rv->next;
508 		}
509 	else {
510 		x = 1 << k;
511 #ifdef Omit_Private_Memory
512 		rv = (Bigint *)MALLOC(sizeof(Bigint) + (x-1)*sizeof(ULong));
513 #else
514 		len = (sizeof(Bigint) + (x-1)*sizeof(ULong) + sizeof(double) - 1)
515 			/sizeof(double);
516 		if (k <= Kmax && pmem_next - private_mem + len <= PRIVATE_mem) {
517 			rv = (Bigint*)pmem_next;
518 			pmem_next += len;
519 			}
520 		else
521 			rv = (Bigint*)MALLOC(len*sizeof(double));
522 #endif
523 		rv->k = k;
524 		rv->maxwds = x;
525 		}
526 	FREE_DTOA_LOCK(0);
527 	rv->sign = rv->wds = 0;
528 	return rv;
529 	}
530 
531  static void
Bfree(v)532 Bfree
533 #ifdef KR_headers
534 	(v) Bigint *v;
535 #else
536 	(Bigint *v)
537 #endif
538 {
539 	if (v) {
540                 if (v->k > Kmax)
541                         free((void*)v);
542                 else {
543          		ACQUIRE_DTOA_LOCK(0);
544          		v->next = freelist[v->k];
545         		freelist[v->k] = v;
546         		FREE_DTOA_LOCK(0);
547                         }
548 		}
549 	}
550 
551 #define Bcopy(x,y) memcpy((char *)&x->sign, (char *)&y->sign, \
552 y->wds*sizeof(Long) + 2*sizeof(int))
553 
554  static Bigint *
multadd(b,m,a)555 multadd
556 #ifdef KR_headers
557 	(b, m, a) Bigint *b; int m, a;
558 #else
559 	(Bigint *b, int m, int a)	/* multiply by m and add a */
560 #endif
561 {
562 	int i, wds;
563 #ifdef ULLong
564 	ULong *x;
565 	ULLong carry, y;
566 #else
567 	ULong carry, *x, y;
568 #ifdef Pack_32
569 	ULong xi, z;
570 #endif
571 #endif
572 	Bigint *b1;
573 
574 	wds = b->wds;
575 	x = b->x;
576 	i = 0;
577 	carry = a;
578 	do {
579 #ifdef ULLong
580 		y = *x * (ULLong)m + carry;
581 		carry = y >> 32;
582 		*x++ = y & FFFFFFFF;
583 #else
584 #ifdef Pack_32
585 		xi = *x;
586 		y = (xi & 0xffff) * m + carry;
587 		z = (xi >> 16) * m + (y >> 16);
588 		carry = z >> 16;
589 		*x++ = (z << 16) + (y & 0xffff);
590 #else
591 		y = *x * m + carry;
592 		carry = y >> 16;
593 		*x++ = y & 0xffff;
594 #endif
595 #endif
596 		}
597 		while(++i < wds);
598 	if (carry) {
599 		if (wds >= b->maxwds) {
600 			b1 = Balloc(b->k+1);
601 			Bcopy(b1, b);
602 			Bfree(b);
603 			b = b1;
604 			}
605 		b->x[wds++] = carry;
606 		b->wds = wds;
607 		}
608 	return b;
609 	}
610 
611  static Bigint *
s2b(s,nd0,nd,y9)612 s2b
613 #ifdef KR_headers
614 	(s, nd0, nd, y9) CONST char *s; int nd0, nd; ULong y9;
615 #else
616 	(CONST char *s, int nd0, int nd, ULong y9)
617 #endif
618 {
619 	Bigint *b;
620 	int i, k;
621 	Long x, y;
622 
623 	x = (nd + 8) / 9;
624 	for(k = 0, y = 1; x > y; y <<= 1, k++) ;
625 #ifdef Pack_32
626 	b = Balloc(k);
627 	b->x[0] = y9;
628 	b->wds = 1;
629 #else
630 	b = Balloc(k+1);
631 	b->x[0] = y9 & 0xffff;
632 	b->wds = (b->x[1] = y9 >> 16) ? 2 : 1;
633 #endif
634 
635 	i = 9;
636 	if (9 < nd0) {
637 		s += 9;
638 		do b = multadd(b, 10, *s++ - '0');
639 			while(++i < nd0);
640 		s++;
641 		}
642 	else
643 		s += 10;
644 	for(; i < nd; i++)
645 		b = multadd(b, 10, *s++ - '0');
646 	return b;
647 	}
648 
649  static int
hi0bits(x)650 hi0bits
651 #ifdef KR_headers
652 	(x) register ULong x;
653 #else
654 	(register ULong x)
655 #endif
656 {
657 	register int k = 0;
658 
659 	if (!(x & 0xffff0000)) {
660 		k = 16;
661 		x <<= 16;
662 		}
663 	if (!(x & 0xff000000)) {
664 		k += 8;
665 		x <<= 8;
666 		}
667 	if (!(x & 0xf0000000)) {
668 		k += 4;
669 		x <<= 4;
670 		}
671 	if (!(x & 0xc0000000)) {
672 		k += 2;
673 		x <<= 2;
674 		}
675 	if (!(x & 0x80000000)) {
676 		k++;
677 		if (!(x & 0x40000000))
678 			return 32;
679 		}
680 	return k;
681 	}
682 
683  static int
lo0bits(y)684 lo0bits
685 #ifdef KR_headers
686 	(y) ULong *y;
687 #else
688 	(ULong *y)
689 #endif
690 {
691 	register int k;
692 	register ULong x = *y;
693 
694 	if (x & 7) {
695 		if (x & 1)
696 			return 0;
697 		if (x & 2) {
698 			*y = x >> 1;
699 			return 1;
700 			}
701 		*y = x >> 2;
702 		return 2;
703 		}
704 	k = 0;
705 	if (!(x & 0xffff)) {
706 		k = 16;
707 		x >>= 16;
708 		}
709 	if (!(x & 0xff)) {
710 		k += 8;
711 		x >>= 8;
712 		}
713 	if (!(x & 0xf)) {
714 		k += 4;
715 		x >>= 4;
716 		}
717 	if (!(x & 0x3)) {
718 		k += 2;
719 		x >>= 2;
720 		}
721 	if (!(x & 1)) {
722 		k++;
723 		x >>= 1;
724 		if (!x)
725 			return 32;
726 		}
727 	*y = x;
728 	return k;
729 	}
730 
731  static Bigint *
i2b(i)732 i2b
733 #ifdef KR_headers
734 	(i) int i;
735 #else
736 	(int i)
737 #endif
738 {
739 	Bigint *b;
740 
741 	b = Balloc(1);
742 	b->x[0] = i;
743 	b->wds = 1;
744 	return b;
745 	}
746 
747  static Bigint *
mult(a,b)748 mult
749 #ifdef KR_headers
750 	(a, b) Bigint *a, *b;
751 #else
752 	(Bigint *a, Bigint *b)
753 #endif
754 {
755 	Bigint *c;
756 	int k, wa, wb, wc;
757 	ULong *x, *xa, *xae, *xb, *xbe, *xc, *xc0;
758 	ULong y;
759 #ifdef ULLong
760 	ULLong carry, z;
761 #else
762 	ULong carry, z;
763 #ifdef Pack_32
764 	ULong z2;
765 #endif
766 #endif
767 
768 	if (a->wds < b->wds) {
769 		c = a;
770 		a = b;
771 		b = c;
772 		}
773 	k = a->k;
774 	wa = a->wds;
775 	wb = b->wds;
776 	wc = wa + wb;
777 	if (wc > a->maxwds)
778 		k++;
779 	c = Balloc(k);
780 	for(x = c->x, xa = x + wc; x < xa; x++)
781 		*x = 0;
782 	xa = a->x;
783 	xae = xa + wa;
784 	xb = b->x;
785 	xbe = xb + wb;
786 	xc0 = c->x;
787 #ifdef ULLong
788 	for(; xb < xbe; xc0++) {
789 		if ((y = *xb++)) {
790 			x = xa;
791 			xc = xc0;
792 			carry = 0;
793 			do {
794 				z = *x++ * (ULLong)y + *xc + carry;
795 				carry = z >> 32;
796 				*xc++ = z & FFFFFFFF;
797 				}
798 				while(x < xae);
799 			*xc = carry;
800 			}
801 		}
802 #else
803 #ifdef Pack_32
804 	for(; xb < xbe; xb++, xc0++) {
805 		if (y = *xb & 0xffff) {
806 			x = xa;
807 			xc = xc0;
808 			carry = 0;
809 			do {
810 				z = (*x & 0xffff) * y + (*xc & 0xffff) + carry;
811 				carry = z >> 16;
812 				z2 = (*x++ >> 16) * y + (*xc >> 16) + carry;
813 				carry = z2 >> 16;
814 				Storeinc(xc, z2, z);
815 				}
816 				while(x < xae);
817 			*xc = carry;
818 			}
819 		if (y = *xb >> 16) {
820 			x = xa;
821 			xc = xc0;
822 			carry = 0;
823 			z2 = *xc;
824 			do {
825 				z = (*x & 0xffff) * y + (*xc >> 16) + carry;
826 				carry = z >> 16;
827 				Storeinc(xc, z, z2);
828 				z2 = (*x++ >> 16) * y + (*xc & 0xffff) + carry;
829 				carry = z2 >> 16;
830 				}
831 				while(x < xae);
832 			*xc = z2;
833 			}
834 		}
835 #else
836 	for(; xb < xbe; xc0++) {
837 		if (y = *xb++) {
838 			x = xa;
839 			xc = xc0;
840 			carry = 0;
841 			do {
842 				z = *x++ * y + *xc + carry;
843 				carry = z >> 16;
844 				*xc++ = z & 0xffff;
845 				}
846 				while(x < xae);
847 			*xc = carry;
848 			}
849 		}
850 #endif
851 #endif
852 	for(xc0 = c->x, xc = xc0 + wc; wc > 0 && !*--xc; --wc) ;
853 	c->wds = wc;
854 	return c;
855 	}
856 
857  static Bigint *p5s;
858 
859  static Bigint *
pow5mult(b,k)860 pow5mult
861 #ifdef KR_headers
862 	(b, k) Bigint *b; int k;
863 #else
864 	(Bigint *b, int k)
865 #endif
866 {
867 	Bigint *b1, *p5, *p51;
868 	int i;
869 	static int p05[3] = { 5, 25, 125 };
870 
871 	if ((i = k & 3))
872 		b = multadd(b, p05[i-1], 0);
873 
874 	if (!(k >>= 2))
875 		return b;
876 	if (!(p5 = p5s)) {
877 		/* first time */
878 #ifdef MULTIPLE_THREADS
879 		ACQUIRE_DTOA_LOCK(1);
880 		if (!(p5 = p5s)) {
881 			p5 = p5s = i2b(625);
882 			p5->next = 0;
883 			}
884 		FREE_DTOA_LOCK(1);
885 #else
886 		p5 = p5s = i2b(625);
887 		p5->next = 0;
888 #endif
889 		}
890 	for(;;) {
891 		if (k & 1) {
892 			b1 = mult(b, p5);
893 			Bfree(b);
894 			b = b1;
895 			}
896 		if (!(k >>= 1))
897 			break;
898 		if (!(p51 = p5->next)) {
899 #ifdef MULTIPLE_THREADS
900 			ACQUIRE_DTOA_LOCK(1);
901 			if (!(p51 = p5->next)) {
902 				p51 = p5->next = mult(p5,p5);
903 				p51->next = 0;
904 				}
905 			FREE_DTOA_LOCK(1);
906 #else
907 			p51 = p5->next = mult(p5,p5);
908 			p51->next = 0;
909 #endif
910 			}
911 		p5 = p51;
912 		}
913 	return b;
914 	}
915 
916  static Bigint *
lshift(b,k)917 lshift
918 #ifdef KR_headers
919 	(b, k) Bigint *b; int k;
920 #else
921 	(Bigint *b, int k)
922 #endif
923 {
924 	int i, k1, n, n1;
925 	Bigint *b1;
926 	ULong *x, *x1, *xe, z;
927 
928 #ifdef Pack_32
929 	n = k >> 5;
930 #else
931 	n = k >> 4;
932 #endif
933 	k1 = b->k;
934 	n1 = n + b->wds + 1;
935 	for(i = b->maxwds; n1 > i; i <<= 1)
936 		k1++;
937 	b1 = Balloc(k1);
938 	x1 = b1->x;
939 	for(i = 0; i < n; i++)
940 		*x1++ = 0;
941 	x = b->x;
942 	xe = x + b->wds;
943 #ifdef Pack_32
944 	if (k &= 0x1f) {
945 		k1 = 32 - k;
946 		z = 0;
947 		do {
948 			*x1++ = *x << k | z;
949 			z = *x++ >> k1;
950 			}
951 			while(x < xe);
952 		if ((*x1 = z))
953 			++n1;
954 		}
955 #else
956 	if (k &= 0xf) {
957 		k1 = 16 - k;
958 		z = 0;
959 		do {
960 			*x1++ = *x << k  & 0xffff | z;
961 			z = *x++ >> k1;
962 			}
963 			while(x < xe);
964 		if (*x1 = z)
965 			++n1;
966 		}
967 #endif
968 	else do
969 		*x1++ = *x++;
970 		while(x < xe);
971 	b1->wds = n1 - 1;
972 	Bfree(b);
973 	return b1;
974 	}
975 
976  static int
cmp(a,b)977 cmp
978 #ifdef KR_headers
979 	(a, b) Bigint *a, *b;
980 #else
981 	(Bigint *a, Bigint *b)
982 #endif
983 {
984 	ULong *xa, *xa0, *xb, *xb0;
985 	int i, j;
986 
987 	i = a->wds;
988 	j = b->wds;
989 #ifdef DEBUG
990 	if (i > 1 && !a->x[i-1])
991 		Bug("cmp called with a->x[a->wds-1] == 0");
992 	if (j > 1 && !b->x[j-1])
993 		Bug("cmp called with b->x[b->wds-1] == 0");
994 #endif
995 	if (i -= j)
996 		return i;
997 	xa0 = a->x;
998 	xa = xa0 + j;
999 	xb0 = b->x;
1000 	xb = xb0 + j;
1001 	for(;;) {
1002 		if (*--xa != *--xb)
1003 			return *xa < *xb ? -1 : 1;
1004 		if (xa <= xa0)
1005 			break;
1006 		}
1007 	return 0;
1008 	}
1009 
1010  static Bigint *
diff(a,b)1011 diff
1012 #ifdef KR_headers
1013 	(a, b) Bigint *a, *b;
1014 #else
1015 	(Bigint *a, Bigint *b)
1016 #endif
1017 {
1018 	Bigint *c;
1019 	int i, wa, wb;
1020 	ULong *xa, *xae, *xb, *xbe, *xc;
1021 #ifdef ULLong
1022 	ULLong borrow, y;
1023 #else
1024 	ULong borrow, y;
1025 #ifdef Pack_32
1026 	ULong z;
1027 #endif
1028 #endif
1029 
1030 	i = cmp(a,b);
1031 	if (!i) {
1032 		c = Balloc(0);
1033 		c->wds = 1;
1034 		c->x[0] = 0;
1035 		return c;
1036 		}
1037 	if (i < 0) {
1038 		c = a;
1039 		a = b;
1040 		b = c;
1041 		i = 1;
1042 		}
1043 	else
1044 		i = 0;
1045 	c = Balloc(a->k);
1046 	c->sign = i;
1047 	wa = a->wds;
1048 	xa = a->x;
1049 	xae = xa + wa;
1050 	wb = b->wds;
1051 	xb = b->x;
1052 	xbe = xb + wb;
1053 	xc = c->x;
1054 	borrow = 0;
1055 #ifdef ULLong
1056 	do {
1057 		y = (ULLong)*xa++ - *xb++ - borrow;
1058 		borrow = y >> 32 & (ULong)1;
1059 		*xc++ = y & FFFFFFFF;
1060 		}
1061 		while(xb < xbe);
1062 	while(xa < xae) {
1063 		y = *xa++ - borrow;
1064 		borrow = y >> 32 & (ULong)1;
1065 		*xc++ = y & FFFFFFFF;
1066 		}
1067 #else
1068 #ifdef Pack_32
1069 	do {
1070 		y = (*xa & 0xffff) - (*xb & 0xffff) - borrow;
1071 		borrow = (y & 0x10000) >> 16;
1072 		z = (*xa++ >> 16) - (*xb++ >> 16) - borrow;
1073 		borrow = (z & 0x10000) >> 16;
1074 		Storeinc(xc, z, y);
1075 		}
1076 		while(xb < xbe);
1077 	while(xa < xae) {
1078 		y = (*xa & 0xffff) - borrow;
1079 		borrow = (y & 0x10000) >> 16;
1080 		z = (*xa++ >> 16) - borrow;
1081 		borrow = (z & 0x10000) >> 16;
1082 		Storeinc(xc, z, y);
1083 		}
1084 #else
1085 	do {
1086 		y = *xa++ - *xb++ - borrow;
1087 		borrow = (y & 0x10000) >> 16;
1088 		*xc++ = y & 0xffff;
1089 		}
1090 		while(xb < xbe);
1091 	while(xa < xae) {
1092 		y = *xa++ - borrow;
1093 		borrow = (y & 0x10000) >> 16;
1094 		*xc++ = y & 0xffff;
1095 		}
1096 #endif
1097 #endif
1098 	while(!*--xc)
1099 		wa--;
1100 	c->wds = wa;
1101 	return c;
1102 	}
1103 
1104  static double
ulp(x)1105 ulp
1106 #ifdef KR_headers
1107 	(x) double x;
1108 #else
1109 	(double x)
1110 #endif
1111 {
1112 	register Long L;
1113 	double a;
1114 
1115 	L = (word0(x) & Exp_mask) - (P-1)*Exp_msk1;
1116 #ifndef Avoid_Underflow
1117 #ifndef Sudden_Underflow
1118 	if (L > 0) {
1119 #endif
1120 #endif
1121 #ifdef IBM
1122 		L |= Exp_msk1 >> 4;
1123 #endif
1124 		word0(a) = L;
1125 		word1(a) = 0;
1126 #ifndef Avoid_Underflow
1127 #ifndef Sudden_Underflow
1128 		}
1129 	else {
1130 		L = -L >> Exp_shift;
1131 		if (L < Exp_shift) {
1132 			word0(a) = 0x80000 >> L;
1133 			word1(a) = 0;
1134 			}
1135 		else {
1136 			word0(a) = 0;
1137 			L -= Exp_shift;
1138 			word1(a) = L >= 31 ? 1 : 1 << 31 - L;
1139 			}
1140 		}
1141 #endif
1142 #endif
1143 	return dval(a);
1144 	}
1145 
1146  static double
b2d(a,e)1147 b2d
1148 #ifdef KR_headers
1149 	(a, e) Bigint *a; int *e;
1150 #else
1151 	(Bigint *a, int *e)
1152 #endif
1153 {
1154 	ULong *xa, *xa0, w, y, z;
1155 	int k;
1156 	double d;
1157 #ifdef VAX
1158 	ULong d0, d1;
1159 #else
1160 #define d0 word0(d)
1161 #define d1 word1(d)
1162 #endif
1163 
1164 	xa0 = a->x;
1165 	xa = xa0 + a->wds;
1166 	y = *--xa;
1167 #ifdef DEBUG
1168 	if (!y) Bug("zero y in b2d");
1169 #endif
1170 	k = hi0bits(y);
1171 	*e = 32 - k;
1172 #ifdef Pack_32
1173 	if (k < Ebits) {
1174 		d0 = Exp_1 | (y >> (Ebits - k));
1175 		w = xa > xa0 ? *--xa : 0;
1176 		d1 = (y << ((32-Ebits) + k)) | (w >> (Ebits - k));
1177 		goto ret_d;
1178 		}
1179 	z = xa > xa0 ? *--xa : 0;
1180 	if (k -= Ebits) {
1181 		d0 = Exp_1 | (y << k) | (z >> (32 - k));
1182 		y = xa > xa0 ? *--xa : 0;
1183 		d1 = (z << k) | (y >> (32 - k));
1184 		}
1185 	else {
1186 		d0 = Exp_1 | y;
1187 		d1 = z;
1188 		}
1189 #else
1190 	if (k < Ebits + 16) {
1191 		z = xa > xa0 ? *--xa : 0;
1192 		d0 = Exp_1 | (y << (k - Ebits)) | (z >> (Ebits + 16 - k));
1193 		w = xa > xa0 ? *--xa : 0;
1194 		y = xa > xa0 ? *--xa : 0;
1195 		d1 = (z << (k + 16 - Ebits)) | (w << (k - Ebits)) | (y >> (16 + Ebits - k));
1196 		goto ret_d;
1197 		}
1198 	z = xa > xa0 ? *--xa : 0;
1199 	w = xa > xa0 ? *--xa : 0;
1200 	k -= Ebits + 16;
1201 	d0 = Exp_1 | y << k + 16 | z << k | w >> 16 - k;
1202 	y = xa > xa0 ? *--xa : 0;
1203 	d1 = w << k + 16 | y << k;
1204 #endif
1205  ret_d:
1206 #ifdef VAX
1207 	word0(d) = d0 >> 16 | d0 << 16;
1208 	word1(d) = d1 >> 16 | d1 << 16;
1209 #else
1210 #undef d0
1211 #undef d1
1212 #endif
1213 	return dval(d);
1214 	}
1215 
1216  static Bigint *
d2b(d,e,bits)1217 d2b
1218 #ifdef KR_headers
1219 	(d, e, bits) double d; int *e, *bits;
1220 #else
1221 	(double d, int *e, int *bits)
1222 #endif
1223 {
1224 	Bigint *b;
1225 	int de, k;
1226 	ULong *x, y, z;
1227 #ifndef Sudden_Underflow
1228 	int i;
1229 #endif
1230 #ifdef VAX
1231 	ULong d0, d1;
1232 	d0 = word0(d) >> 16 | word0(d) << 16;
1233 	d1 = word1(d) >> 16 | word1(d) << 16;
1234 #else
1235 #define d0 word0(d)
1236 #define d1 word1(d)
1237 #endif
1238 
1239 #ifdef Pack_32
1240 	b = Balloc(1);
1241 #else
1242 	b = Balloc(2);
1243 #endif
1244 	x = b->x;
1245 
1246 	z = d0 & Frac_mask;
1247 	d0 &= 0x7fffffff;	/* clear sign bit, which we ignore */
1248 #ifdef Sudden_Underflow
1249 	de = (int)(d0 >> Exp_shift);
1250 #ifndef IBM
1251 	z |= Exp_msk11;
1252 #endif
1253 #else
1254 	if ((de = (int)(d0 >> Exp_shift)))
1255 		z |= Exp_msk1;
1256 #endif
1257 #ifdef Pack_32
1258 	if ((y = d1)) {
1259 		if ((k = lo0bits(&y))) {
1260 			x[0] = y | (z << (32 - k));
1261 			z >>= k;
1262 			}
1263 		else
1264 			x[0] = y;
1265 #ifndef Sudden_Underflow
1266 		i =
1267 #endif
1268 		    b->wds = (x[1] = z) ? 2 : 1;
1269 		}
1270 	else {
1271                /* This assertion fails for "1e-500" and other very
1272                 * small numbers.  It provides the right result (0)
1273                 * though. This assert has also been removed from KJS's
1274                 * version of dtoa.c.
1275                 *
1276                 * #ifdef DEBUG
1277                 *     if (!z) Bug("zero z in b2d");
1278                 * #endif
1279                 */
1280 		k = lo0bits(&z);
1281 		x[0] = z;
1282 #ifndef Sudden_Underflow
1283 		i =
1284 #endif
1285 		    b->wds = 1;
1286 		k += 32;
1287 		}
1288 #else
1289 	if (y = d1) {
1290 		if (k = lo0bits(&y))
1291 			if (k >= 16) {
1292 				x[0] = y | z << 32 - k & 0xffff;
1293 				x[1] = z >> k - 16 & 0xffff;
1294 				x[2] = z >> k;
1295 				i = 2;
1296 				}
1297 			else {
1298 				x[0] = y & 0xffff;
1299 				x[1] = y >> 16 | z << 16 - k & 0xffff;
1300 				x[2] = z >> k & 0xffff;
1301 				x[3] = z >> k+16;
1302 				i = 3;
1303 				}
1304 		else {
1305 			x[0] = y & 0xffff;
1306 			x[1] = y >> 16;
1307 			x[2] = z & 0xffff;
1308 			x[3] = z >> 16;
1309 			i = 3;
1310 			}
1311 		}
1312 	else {
1313 #ifdef DEBUG
1314 		if (!z)
1315 			Bug("Zero passed to d2b");
1316 #endif
1317 		k = lo0bits(&z);
1318 		if (k >= 16) {
1319 			x[0] = z;
1320 			i = 0;
1321 			}
1322 		else {
1323 			x[0] = z & 0xffff;
1324 			x[1] = z >> 16;
1325 			i = 1;
1326 			}
1327 		k += 32;
1328 		}
1329 	while(!x[i])
1330 		--i;
1331 	b->wds = i + 1;
1332 #endif
1333 #ifndef Sudden_Underflow
1334 	if (de) {
1335 #endif
1336 #ifdef IBM
1337 		*e = (de - Bias - (P-1) << 2) + k;
1338 		*bits = 4*P + 8 - k - hi0bits(word0(d) & Frac_mask);
1339 #else
1340 		*e = de - Bias - (P-1) + k;
1341 		*bits = P - k;
1342 #endif
1343 #ifndef Sudden_Underflow
1344 		}
1345 	else {
1346 		*e = de - Bias - (P-1) + 1 + k;
1347 #ifdef Pack_32
1348 		*bits = 32*i - hi0bits(x[i-1]);
1349 #else
1350 		*bits = (i+2)*16 - hi0bits(x[i]);
1351 #endif
1352 		}
1353 #endif
1354 	return b;
1355 	}
1356 #undef d0
1357 #undef d1
1358 
1359  static double
ratio(a,b)1360 ratio
1361 #ifdef KR_headers
1362 	(a, b) Bigint *a, *b;
1363 #else
1364 	(Bigint *a, Bigint *b)
1365 #endif
1366 {
1367 	double da, db;
1368 	int k, ka, kb;
1369 
1370 	dval(da) = b2d(a, &ka);
1371 	dval(db) = b2d(b, &kb);
1372 #ifdef Pack_32
1373 	k = ka - kb + 32*(a->wds - b->wds);
1374 #else
1375 	k = ka - kb + 16*(a->wds - b->wds);
1376 #endif
1377 #ifdef IBM
1378 	if (k > 0) {
1379 		word0(da) += (k >> 2)*Exp_msk1;
1380 		if (k &= 3)
1381 			dval(da) *= 1 << k;
1382 		}
1383 	else {
1384 		k = -k;
1385 		word0(db) += (k >> 2)*Exp_msk1;
1386 		if (k &= 3)
1387 			dval(db) *= 1 << k;
1388 		}
1389 #else
1390 	if (k > 0)
1391 		word0(da) += k*Exp_msk1;
1392 	else {
1393 		k = -k;
1394 		word0(db) += k*Exp_msk1;
1395 		}
1396 #endif
1397 	return dval(da) / dval(db);
1398 	}
1399 
1400  static CONST double
1401 tens[] = {
1402 		1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9,
1403 		1e10, 1e11, 1e12, 1e13, 1e14, 1e15, 1e16, 1e17, 1e18, 1e19,
1404 		1e20, 1e21, 1e22
1405 #ifdef VAX
1406 		, 1e23, 1e24
1407 #endif
1408 		};
1409 
1410  static CONST double
1411 #ifdef IEEE_Arith
1412 bigtens[] = { 1e16, 1e32, 1e64, 1e128, 1e256 };
1413 static CONST double tinytens[] = { 1e-16, 1e-32, 1e-64, 1e-128,
1414 #ifdef Avoid_Underflow
1415 		9007199254740992.*9007199254740992.e-256
1416 		/* = 2^106 * 1e-53 */
1417 #else
1418 		1e-256
1419 #endif
1420 		};
1421 /* The factor of 2^53 in tinytens[4] helps us avoid setting the underflow */
1422 /* flag unnecessarily.  It leads to a song and dance at the end of strtod. */
1423 #define Scale_Bit 0x10
1424 #define n_bigtens 5
1425 #else
1426 #ifdef IBM
1427 bigtens[] = { 1e16, 1e32, 1e64 };
1428 static CONST double tinytens[] = { 1e-16, 1e-32, 1e-64 };
1429 #define n_bigtens 3
1430 #else
1431 bigtens[] = { 1e16, 1e32 };
1432 static CONST double tinytens[] = { 1e-16, 1e-32 };
1433 #define n_bigtens 2
1434 #endif
1435 #endif
1436 
1437 #ifndef IEEE_Arith
1438 #undef INFNAN_CHECK
1439 #endif
1440 
1441 #ifdef INFNAN_CHECK
1442 
1443 #ifndef NAN_WORD0
1444 #define NAN_WORD0 0x7ff80000
1445 #endif
1446 
1447 #ifndef NAN_WORD1
1448 #define NAN_WORD1 0
1449 #endif
1450 
1451  static int
match(sp,t)1452 match
1453 #ifdef KR_headers
1454 	(sp, t) char **sp, *t;
1455 #else
1456 	(CONST char **sp, char *t)
1457 #endif
1458 {
1459 	int c, d;
1460 	CONST char *s = *sp;
1461 
1462 	while(d = *t++) {
1463 		if ((c = *++s) >= 'A' && c <= 'Z')
1464 			c += 'a' - 'A';
1465 		if (c != d)
1466 			return 0;
1467 		}
1468 	*sp = s + 1;
1469 	return 1;
1470 	}
1471 
1472 #ifndef No_Hex_NaN
1473  static void
hexnan(rvp,sp)1474 hexnan
1475 #ifdef KR_headers
1476 	(rvp, sp) double *rvp; CONST char **sp;
1477 #else
1478 	(double *rvp, CONST char **sp)
1479 #endif
1480 {
1481 	ULong c, x[2];
1482 	CONST char *s;
1483 	int havedig, udx0, xshift;
1484 
1485 	x[0] = x[1] = 0;
1486 	havedig = xshift = 0;
1487 	udx0 = 1;
1488 	s = *sp;
1489 	while(c = *(CONST unsigned char*)++s) {
1490 		if (c >= '0' && c <= '9')
1491 			c -= '0';
1492 		else if (c >= 'a' && c <= 'f')
1493 			c += 10 - 'a';
1494 		else if (c >= 'A' && c <= 'F')
1495 			c += 10 - 'A';
1496 		else if (c <= ' ') {
1497 			if (udx0 && havedig) {
1498 				udx0 = 0;
1499 				xshift = 1;
1500 				}
1501 			continue;
1502 			}
1503 		else if (/*(*/ c == ')' && havedig) {
1504 			*sp = s + 1;
1505 			break;
1506 			}
1507 		else
1508 			return;	/* invalid form: don't change *sp */
1509 		havedig = 1;
1510 		if (xshift) {
1511 			xshift = 0;
1512 			x[0] = x[1];
1513 			x[1] = 0;
1514 			}
1515 		if (udx0)
1516 			x[0] = (x[0] << 4) | (x[1] >> 28);
1517 		x[1] = (x[1] << 4) | c;
1518 		}
1519 	if ((x[0] &= 0xfffff) || x[1]) {
1520 		word0(*rvp) = Exp_mask | x[0];
1521 		word1(*rvp) = x[1];
1522 		}
1523 	}
1524 #endif /*No_Hex_NaN*/
1525 #endif /* INFNAN_CHECK */
1526 
1527  double
strtod(s00,se)1528 strtod
1529 #ifdef KR_headers
1530 	(s00, se) CONST char *s00; char **se;
1531 #else
1532 	(CONST char *s00, char **se)
1533 #endif
1534 {
1535 #ifdef Avoid_Underflow
1536 	int scale;
1537 #endif
1538 	int bb2, bb5, bbe, bd2, bd5, bbbits, bs2, c, dsign,
1539 		 e, e1, esign, i, j, k, nd, nd0, nf, nz, nz0, sign;
1540 	CONST char *s, *s0, *s1;
1541 	double aadj, aadj1, adj, rv, rv0;
1542 	Long L;
1543 	ULong y, z;
1544 	Bigint *bb = NULL, *bb1, *bd = NULL, *bd0, *bs = NULL, *delta = NULL;
1545 #ifdef SET_INEXACT
1546 	int inexact, oldinexact;
1547 #endif
1548 #ifdef Honor_FLT_ROUNDS
1549 	int rounding;
1550 #endif
1551 #ifdef USE_LOCALE
1552 	CONST char *s2;
1553 #endif
1554 
1555 	sign = nz0 = nz = 0;
1556 	dval(rv) = 0.;
1557 	for(s = s00;;s++) switch(*s) {
1558 		case '-':
1559 			sign = 1;
1560 			/* no break */
1561 		case '+':
1562 			if (*++s)
1563 				goto break2;
1564 			/* no break */
1565 		case 0:
1566 			goto ret0;
1567 		case '\t':
1568 		case '\n':
1569 		case '\v':
1570 		case '\f':
1571 		case '\r':
1572 		case ' ':
1573 			continue;
1574 		default:
1575 			goto break2;
1576 		}
1577  break2:
1578 	if (*s == '0') {
1579 		nz0 = 1;
1580 		while(*++s == '0') ;
1581 		if (!*s)
1582 			goto ret;
1583 		}
1584 	s0 = s;
1585 	y = z = 0;
1586 	for(nd = nf = 0; (c = *s) >= '0' && c <= '9'; nd++, s++)
1587 		if (nd < 9)
1588 			y = 10*y + c - '0';
1589 		else if (nd < 16)
1590 			z = 10*z + c - '0';
1591 	nd0 = nd;
1592 #ifdef USE_LOCALE
1593 	s1 = localeconv()->decimal_point;
1594 	if (c == *s1) {
1595 		c = '.';
1596 		if (*++s1) {
1597 			s2 = s;
1598 			for(;;) {
1599 				if (*++s2 != *s1) {
1600 					c = 0;
1601 					break;
1602 					}
1603 				if (!*++s1) {
1604 					s = s2;
1605 					break;
1606 					}
1607 				}
1608 			}
1609 		}
1610 #endif
1611 	if (c == '.') {
1612 		c = *++s;
1613 		if (!nd) {
1614 			for(; c == '0'; c = *++s)
1615 				nz++;
1616 			if (c > '0' && c <= '9') {
1617 				s0 = s;
1618 				nf += nz;
1619 				nz = 0;
1620 				goto have_dig;
1621 				}
1622 			goto dig_done;
1623 			}
1624 		for(; c >= '0' && c <= '9'; c = *++s) {
1625  have_dig:
1626 			nz++;
1627 			if (c -= '0') {
1628 				nf += nz;
1629 				for(i = 1; i < nz; i++)
1630 					if (nd++ < 9)
1631 						y *= 10;
1632 					else if (nd <= DBL_DIG + 1)
1633 						z *= 10;
1634 				if (nd++ < 9)
1635 					y = 10*y + c;
1636 				else if (nd <= DBL_DIG + 1)
1637 					z = 10*z + c;
1638 				nz = 0;
1639 				}
1640 			}
1641 		}
1642  dig_done:
1643 	e = 0;
1644 	if (c == 'e' || c == 'E') {
1645 		if (!nd && !nz && !nz0) {
1646 			goto ret0;
1647 			}
1648 		s00 = s;
1649 		esign = 0;
1650 		switch(c = *++s) {
1651 			case '-':
1652 				esign = 1;
1653 			case '+':
1654 				c = *++s;
1655 			}
1656 		if (c >= '0' && c <= '9') {
1657 			while(c == '0')
1658 				c = *++s;
1659 			if (c > '0' && c <= '9') {
1660 				L = c - '0';
1661 				s1 = s;
1662 				while((c = *++s) >= '0' && c <= '9')
1663 					L = 10*L + c - '0';
1664 				if (s - s1 > 8 || L > 19999)
1665 					/* Avoid confusion from exponents
1666 					 * so large that e might overflow.
1667 					 */
1668 					e = 19999; /* safe for 16 bit ints */
1669 				else
1670 					e = (int)L;
1671 				if (esign)
1672 					e = -e;
1673 				}
1674 			else
1675 				e = 0;
1676 			}
1677 		else
1678 			s = s00;
1679 		}
1680 	if (!nd) {
1681 		if (!nz && !nz0) {
1682 #ifdef INFNAN_CHECK
1683 			/* Check for Nan and Infinity */
1684 			switch(c) {
1685 			  case 'i':
1686 			  case 'I':
1687 				if (match(&s,"nf")) {
1688 					--s;
1689 					if (!match(&s,"inity"))
1690 						++s;
1691 					word0(rv) = 0x7ff00000;
1692 					word1(rv) = 0;
1693 					goto ret;
1694 					}
1695 				break;
1696 			  case 'n':
1697 			  case 'N':
1698 				if (match(&s, "an")) {
1699 					word0(rv) = NAN_WORD0;
1700 					word1(rv) = NAN_WORD1;
1701 #ifndef No_Hex_NaN
1702 					if (*s == '(') /*)*/
1703 						hexnan(&rv, &s);
1704 #endif
1705 					goto ret;
1706 					}
1707 			  }
1708 #endif /* INFNAN_CHECK */
1709  ret0:
1710 			s = s00;
1711 			sign = 0;
1712 			}
1713 		goto ret;
1714 		}
1715 	e1 = e -= nf;
1716 
1717 	/* Now we have nd0 digits, starting at s0, followed by a
1718 	 * decimal point, followed by nd-nd0 digits.  The number we're
1719 	 * after is the integer represented by those digits times
1720 	 * 10**e */
1721 
1722 	if (!nd0)
1723 		nd0 = nd;
1724 	k = nd < DBL_DIG + 1 ? nd : DBL_DIG + 1;
1725 	dval(rv) = y;
1726 	if (k > 9) {
1727 #ifdef SET_INEXACT
1728 		if (k > DBL_DIG)
1729 			oldinexact = get_inexact();
1730 #endif
1731 		dval(rv) = tens[k - 9] * dval(rv) + z;
1732 		}
1733 	bd0 = 0;
1734 	if (nd <= DBL_DIG
1735 #ifndef RND_PRODQUOT
1736 #ifndef Honor_FLT_ROUNDS
1737 		&& Flt_Rounds == 1
1738 #endif
1739 #endif
1740 			) {
1741 		if (!e)
1742 			goto ret;
1743 		if (e > 0) {
1744 			if (e <= Ten_pmax) {
1745 #ifdef VAX
1746 				goto vax_ovfl_check;
1747 #else
1748 #ifdef Honor_FLT_ROUNDS
1749 				/* round correctly FLT_ROUNDS = 2 or 3 */
1750 				if (sign) {
1751 					rv = -rv;
1752 					sign = 0;
1753 					}
1754 #endif
1755 				/* rv = */ rounded_product(dval(rv), tens[e]);
1756 				goto ret;
1757 #endif
1758 				}
1759 			i = DBL_DIG - nd;
1760 			if (e <= Ten_pmax + i) {
1761 				/* A fancier test would sometimes let us do
1762 				 * this for larger i values.
1763 				 */
1764 #ifdef Honor_FLT_ROUNDS
1765 				/* round correctly FLT_ROUNDS = 2 or 3 */
1766 				if (sign) {
1767 					rv = -rv;
1768 					sign = 0;
1769 					}
1770 #endif
1771 				e -= i;
1772 				dval(rv) *= tens[i];
1773 #ifdef VAX
1774 				/* VAX exponent range is so narrow we must
1775 				 * worry about overflow here...
1776 				 */
1777  vax_ovfl_check:
1778 				word0(rv) -= P*Exp_msk1;
1779 				/* rv = */ rounded_product(dval(rv), tens[e]);
1780 				if ((word0(rv) & Exp_mask)
1781 				 > Exp_msk1*(DBL_MAX_EXP+Bias-1-P))
1782 					goto ovfl;
1783 				word0(rv) += P*Exp_msk1;
1784 #else
1785 				/* rv = */ rounded_product(dval(rv), tens[e]);
1786 #endif
1787 				goto ret;
1788 				}
1789 			}
1790 #ifndef Inaccurate_Divide
1791 		else if (e >= -Ten_pmax) {
1792 #ifdef Honor_FLT_ROUNDS
1793 			/* round correctly FLT_ROUNDS = 2 or 3 */
1794 			if (sign) {
1795 				rv = -rv;
1796 				sign = 0;
1797 				}
1798 #endif
1799 			/* rv = */ rounded_quotient(dval(rv), tens[-e]);
1800 			goto ret;
1801 			}
1802 #endif
1803 		}
1804 	e1 += nd - k;
1805 
1806 #ifdef IEEE_Arith
1807 #ifdef SET_INEXACT
1808 	inexact = 1;
1809 	if (k <= DBL_DIG)
1810 		oldinexact = get_inexact();
1811 #endif
1812 #ifdef Avoid_Underflow
1813 	scale = 0;
1814 #endif
1815 #ifdef Honor_FLT_ROUNDS
1816 	if ((rounding = Flt_Rounds) >= 2) {
1817 		if (sign)
1818 			rounding = rounding == 2 ? 0 : 2;
1819 		else
1820 			if (rounding != 2)
1821 				rounding = 0;
1822 		}
1823 #endif
1824 #endif /*IEEE_Arith*/
1825 
1826 	/* Get starting approximation = rv * 10**e1 */
1827 
1828 	if (e1 > 0) {
1829 		if ((i = e1 & 15))
1830 			dval(rv) *= tens[i];
1831 		if (e1 &= ~15) {
1832 			if (e1 > DBL_MAX_10_EXP) {
1833  ovfl:
1834 #ifndef NO_ERRNO
1835 				errno = ERANGE;
1836 #endif
1837 				/* Can't trust HUGE_VAL */
1838 #ifdef IEEE_Arith
1839 #ifdef Honor_FLT_ROUNDS
1840 				switch(rounding) {
1841 				  case 0: /* toward 0 */
1842 				  case 3: /* toward -infinity */
1843 					word0(rv) = Big0;
1844 					word1(rv) = Big1;
1845 					break;
1846 				  default:
1847 					word0(rv) = Exp_mask;
1848 					word1(rv) = 0;
1849 				  }
1850 #else /*Honor_FLT_ROUNDS*/
1851 				word0(rv) = Exp_mask;
1852 				word1(rv) = 0;
1853 #endif /*Honor_FLT_ROUNDS*/
1854 #ifdef SET_INEXACT
1855 				/* set overflow bit */
1856 				dval(rv0) = 1e300;
1857 				dval(rv0) *= dval(rv0);
1858 #endif
1859 #else /*IEEE_Arith*/
1860 				word0(rv) = Big0;
1861 				word1(rv) = Big1;
1862 #endif /*IEEE_Arith*/
1863 				if (bd0)
1864 					goto retfree;
1865 				goto ret;
1866 				}
1867 			e1 >>= 4;
1868 			for(j = 0; e1 > 1; j++, e1 >>= 1)
1869 				if (e1 & 1)
1870 					dval(rv) *= bigtens[j];
1871 		/* The last multiplication could overflow. */
1872 			word0(rv) -= P*Exp_msk1;
1873 			dval(rv) *= bigtens[j];
1874 			if ((z = word0(rv) & Exp_mask)
1875 			 > Exp_msk1*(DBL_MAX_EXP+Bias-P))
1876 				goto ovfl;
1877 			if (z > Exp_msk1*(DBL_MAX_EXP+Bias-1-P)) {
1878 				/* set to largest number */
1879 				/* (Can't trust DBL_MAX) */
1880 				word0(rv) = Big0;
1881 				word1(rv) = Big1;
1882 				}
1883 			else
1884 				word0(rv) += P*Exp_msk1;
1885 			}
1886 		}
1887 	else if (e1 < 0) {
1888 		e1 = -e1;
1889 		if ((i = e1 & 15))
1890 			dval(rv) /= tens[i];
1891 		if (e1 >>= 4) {
1892 			if (e1 >= 1 << n_bigtens)
1893 				goto undfl;
1894 #ifdef Avoid_Underflow
1895 			if (e1 & Scale_Bit)
1896 				scale = 2*P;
1897 			for(j = 0; e1 > 0; j++, e1 >>= 1)
1898 				if (e1 & 1)
1899 					dval(rv) *= tinytens[j];
1900 			if (scale && (j = 2*P + 1 - ((word0(rv) & Exp_mask)
1901 						>> Exp_shift)) > 0) {
1902 				/* scaled rv is denormal; zap j low bits */
1903 				if (j >= 32) {
1904 					word1(rv) = 0;
1905 					if (j >= 53)
1906 					 word0(rv) = (P+2)*Exp_msk1;
1907 					else
1908 					 word0(rv) &= 0xffffffff << (j-32);
1909 					}
1910 				else
1911 					word1(rv) &= 0xffffffff << j;
1912 				}
1913 #else
1914 			for(j = 0; e1 > 1; j++, e1 >>= 1)
1915 				if (e1 & 1)
1916 					dval(rv) *= tinytens[j];
1917 			/* The last multiplication could underflow. */
1918 			dval(rv0) = dval(rv);
1919 			dval(rv) *= tinytens[j];
1920 			if (!dval(rv)) {
1921 				dval(rv) = 2.*dval(rv0);
1922 				dval(rv) *= tinytens[j];
1923 #endif
1924 				if (!dval(rv)) {
1925  undfl:
1926 					dval(rv) = 0.;
1927 #ifndef NO_ERRNO
1928 					errno = ERANGE;
1929 #endif
1930 					if (bd0)
1931 						goto retfree;
1932 					goto ret;
1933 					}
1934 #ifndef Avoid_Underflow
1935 				word0(rv) = Tiny0;
1936 				word1(rv) = Tiny1;
1937 				/* The refinement below will clean
1938 				 * this approximation up.
1939 				 */
1940 				}
1941 #endif
1942 			}
1943 		}
1944 
1945 	/* Now the hard part -- adjusting rv to the correct value.*/
1946 
1947 	/* Put digits into bd: true value = bd * 10^e */
1948 
1949 	bd0 = s2b(s0, nd0, nd, y);
1950 
1951 	for(;;) {
1952 		bd = Balloc(bd0->k);
1953 		Bcopy(bd, bd0);
1954 		bb = d2b(dval(rv), &bbe, &bbbits);	/* rv = bb * 2^bbe */
1955 		bs = i2b(1);
1956 
1957 		if (e >= 0) {
1958 			bb2 = bb5 = 0;
1959 			bd2 = bd5 = e;
1960 			}
1961 		else {
1962 			bb2 = bb5 = -e;
1963 			bd2 = bd5 = 0;
1964 			}
1965 		if (bbe >= 0)
1966 			bb2 += bbe;
1967 		else
1968 			bd2 -= bbe;
1969 		bs2 = bb2;
1970 #ifdef Honor_FLT_ROUNDS
1971 		if (rounding != 1)
1972 			bs2++;
1973 #endif
1974 #ifdef Avoid_Underflow
1975 		j = bbe - scale;
1976 		i = j + bbbits - 1;	/* logb(rv) */
1977 		if (i < Emin)	/* denormal */
1978 			j += P - Emin;
1979 		else
1980 			j = P + 1 - bbbits;
1981 #else /*Avoid_Underflow*/
1982 #ifdef Sudden_Underflow
1983 #ifdef IBM
1984 		j = 1 + 4*P - 3 - bbbits + ((bbe + bbbits - 1) & 3);
1985 #else
1986 		j = P + 1 - bbbits;
1987 #endif
1988 #else /*Sudden_Underflow*/
1989 		j = bbe;
1990 		i = j + bbbits - 1;	/* logb(rv) */
1991 		if (i < Emin)	/* denormal */
1992 			j += P - Emin;
1993 		else
1994 			j = P + 1 - bbbits;
1995 #endif /*Sudden_Underflow*/
1996 #endif /*Avoid_Underflow*/
1997 		bb2 += j;
1998 		bd2 += j;
1999 #ifdef Avoid_Underflow
2000 		bd2 += scale;
2001 #endif
2002 		i = bb2 < bd2 ? bb2 : bd2;
2003 		if (i > bs2)
2004 			i = bs2;
2005 		if (i > 0) {
2006 			bb2 -= i;
2007 			bd2 -= i;
2008 			bs2 -= i;
2009 			}
2010 		if (bb5 > 0) {
2011 			bs = pow5mult(bs, bb5);
2012 			bb1 = mult(bs, bb);
2013 			Bfree(bb);
2014 			bb = bb1;
2015 			}
2016 		if (bb2 > 0)
2017 			bb = lshift(bb, bb2);
2018 		if (bd5 > 0)
2019 			bd = pow5mult(bd, bd5);
2020 		if (bd2 > 0)
2021 			bd = lshift(bd, bd2);
2022 		if (bs2 > 0)
2023 			bs = lshift(bs, bs2);
2024 		delta = diff(bb, bd);
2025 		dsign = delta->sign;
2026 		delta->sign = 0;
2027 		i = cmp(delta, bs);
2028 #ifdef Honor_FLT_ROUNDS
2029 		if (rounding != 1) {
2030 			if (i < 0) {
2031 				/* Error is less than an ulp */
2032 				if (!delta->x[0] && delta->wds <= 1) {
2033 					/* exact */
2034 #ifdef SET_INEXACT
2035 					inexact = 0;
2036 #endif
2037 					break;
2038 					}
2039 				if (rounding) {
2040 					if (dsign) {
2041 						adj = 1.;
2042 						goto apply_adj;
2043 						}
2044 					}
2045 				else if (!dsign) {
2046 					adj = -1.;
2047 					if (!word1(rv)
2048 					 && !(word0(rv) & Frac_mask)) {
2049 						y = word0(rv) & Exp_mask;
2050 #ifdef Avoid_Underflow
2051 						if (!scale || y > 2*P*Exp_msk1)
2052 #else
2053 						if (y)
2054 #endif
2055 						  {
2056 						  delta = lshift(delta,Log2P);
2057 						  if (cmp(delta, bs) <= 0)
2058 							adj = -0.5;
2059 						  }
2060 						}
2061  apply_adj:
2062 #ifdef Avoid_Underflow
2063 					if (scale && (y = word0(rv) & Exp_mask)
2064 						<= 2*P*Exp_msk1)
2065 					  word0(adj) += (2*P+1)*Exp_msk1 - y;
2066 #else
2067 #ifdef Sudden_Underflow
2068 					if ((word0(rv) & Exp_mask) <=
2069 							P*Exp_msk1) {
2070 						word0(rv) += P*Exp_msk1;
2071 						dval(rv) += adj*ulp(dval(rv));
2072 						word0(rv) -= P*Exp_msk1;
2073 						}
2074 					else
2075 #endif /*Sudden_Underflow*/
2076 #endif /*Avoid_Underflow*/
2077 					dval(rv) += adj*ulp(dval(rv));
2078 					}
2079 				break;
2080 				}
2081 			adj = ratio(delta, bs);
2082 			if (adj < 1.)
2083 				adj = 1.;
2084 			if (adj <= 0x7ffffffe) {
2085 				/* adj = rounding ? ceil(adj) : floor(adj); */
2086 				y = adj;
2087 				if (y != adj) {
2088 					if (!((rounding>>1) ^ dsign))
2089 						y++;
2090 					adj = y;
2091 					}
2092 				}
2093 #ifdef Avoid_Underflow
2094 			if (scale && (y = word0(rv) & Exp_mask) <= 2*P*Exp_msk1)
2095 				word0(adj) += (2*P+1)*Exp_msk1 - y;
2096 #else
2097 #ifdef Sudden_Underflow
2098 			if ((word0(rv) & Exp_mask) <= P*Exp_msk1) {
2099 				word0(rv) += P*Exp_msk1;
2100 				adj *= ulp(dval(rv));
2101 				if (dsign)
2102 					dval(rv) += adj;
2103 				else
2104 					dval(rv) -= adj;
2105 				word0(rv) -= P*Exp_msk1;
2106 				goto cont;
2107 				}
2108 #endif /*Sudden_Underflow*/
2109 #endif /*Avoid_Underflow*/
2110 			adj *= ulp(dval(rv));
2111 			if (dsign)
2112 				dval(rv) += adj;
2113 			else
2114 				dval(rv) -= adj;
2115 			goto cont;
2116 			}
2117 #endif /*Honor_FLT_ROUNDS*/
2118 
2119 		if (i < 0) {
2120 			/* Error is less than half an ulp -- check for
2121 			 * special case of mantissa a power of two.
2122 			 */
2123 			if (dsign || word1(rv) || word0(rv) & Bndry_mask
2124 #ifdef IEEE_Arith
2125 #ifdef Avoid_Underflow
2126 			 || (word0(rv) & Exp_mask) <= (2*P+1)*Exp_msk1
2127 #else
2128 			 || (word0(rv) & Exp_mask) <= Exp_msk1
2129 #endif
2130 #endif
2131 				) {
2132 #ifdef SET_INEXACT
2133 				if (!delta->x[0] && delta->wds <= 1)
2134 					inexact = 0;
2135 #endif
2136 				break;
2137 				}
2138 			if (!delta->x[0] && delta->wds <= 1) {
2139 				/* exact result */
2140 #ifdef SET_INEXACT
2141 				inexact = 0;
2142 #endif
2143 				break;
2144 				}
2145 			delta = lshift(delta,Log2P);
2146 			if (cmp(delta, bs) > 0)
2147 				goto drop_down;
2148 			break;
2149 			}
2150 		if (i == 0) {
2151 			/* exactly half-way between */
2152 			if (dsign) {
2153 				if ((word0(rv) & Bndry_mask1) == Bndry_mask1
2154 				 &&  word1(rv) == (
2155 #ifdef Avoid_Underflow
2156 			(scale && (y = word0(rv) & Exp_mask) <= 2*P*Exp_msk1)
2157 		? (0xffffffff & (0xffffffff << (2*P+1-(y>>Exp_shift)))) :
2158 #endif
2159 						   0xffffffff)) {
2160 					/*boundary case -- increment exponent*/
2161 					word0(rv) = (word0(rv) & Exp_mask)
2162 						+ Exp_msk1
2163 #ifdef IBM
2164 						| Exp_msk1 >> 4
2165 #endif
2166 						;
2167 					word1(rv) = 0;
2168 #ifdef Avoid_Underflow
2169 					dsign = 0;
2170 #endif
2171 					break;
2172 					}
2173 				}
2174 			else if (!(word0(rv) & Bndry_mask) && !word1(rv)) {
2175  drop_down:
2176 				/* boundary case -- decrement exponent */
2177 #ifdef Sudden_Underflow /*{{*/
2178 				L = word0(rv) & Exp_mask;
2179 #ifdef IBM
2180 				if (L <  Exp_msk1)
2181 #else
2182 #ifdef Avoid_Underflow
2183 				if (L <= (scale ? (2*P+1)*Exp_msk1 : Exp_msk1))
2184 #else
2185 				if (L <= Exp_msk1)
2186 #endif /*Avoid_Underflow*/
2187 #endif /*IBM*/
2188 					goto undfl;
2189 				L -= Exp_msk1;
2190 #else /*Sudden_Underflow}{*/
2191 #ifdef Avoid_Underflow
2192 				if (scale) {
2193 					L = word0(rv) & Exp_mask;
2194 					if (L <= (2*P+1)*Exp_msk1) {
2195 						if (L > (P+2)*Exp_msk1)
2196 							/* round even ==> */
2197 							/* accept rv */
2198 							break;
2199 						/* rv = smallest denormal */
2200 						goto undfl;
2201 						}
2202 					}
2203 #endif /*Avoid_Underflow*/
2204 				L = (word0(rv) & Exp_mask) - Exp_msk1;
2205 #endif /*Sudden_Underflow}}*/
2206 				word0(rv) = L | Bndry_mask1;
2207 				word1(rv) = 0xffffffff;
2208 #ifdef IBM
2209 				goto cont;
2210 #else
2211 				break;
2212 #endif
2213 				}
2214 #ifndef ROUND_BIASED
2215 			if (!(word1(rv) & LSB))
2216 				break;
2217 #endif
2218 			if (dsign)
2219 				dval(rv) += ulp(dval(rv));
2220 #ifndef ROUND_BIASED
2221 			else {
2222 				dval(rv) -= ulp(dval(rv));
2223 #ifndef Sudden_Underflow
2224 				if (!dval(rv))
2225 					goto undfl;
2226 #endif
2227 				}
2228 #ifdef Avoid_Underflow
2229 			dsign = 1 - dsign;
2230 #endif
2231 #endif
2232 			break;
2233 			}
2234 		if ((aadj = ratio(delta, bs)) <= 2.) {
2235 			if (dsign)
2236 				aadj = aadj1 = 1.;
2237 			else if (word1(rv) || word0(rv) & Bndry_mask) {
2238 #ifndef Sudden_Underflow
2239 				if (word1(rv) == Tiny1 && !word0(rv))
2240 					goto undfl;
2241 #endif
2242 				aadj = 1.;
2243 				aadj1 = -1.;
2244 				}
2245 			else {
2246 				/* special case -- power of FLT_RADIX to be */
2247 				/* rounded down... */
2248 
2249 				if (aadj < 2./FLT_RADIX)
2250 					aadj = 1./FLT_RADIX;
2251 				else
2252 					aadj *= 0.5;
2253 				aadj1 = -aadj;
2254 				}
2255 			}
2256 		else {
2257 			aadj *= 0.5;
2258 			aadj1 = dsign ? aadj : -aadj;
2259 #ifdef Check_FLT_ROUNDS
2260 			switch(Rounding) {
2261 				case 2: /* towards +infinity */
2262 					aadj1 -= 0.5;
2263 					break;
2264 				case 0: /* towards 0 */
2265 				case 3: /* towards -infinity */
2266 					aadj1 += 0.5;
2267 				}
2268 #else
2269 			if (Flt_Rounds == 0)
2270 				aadj1 += 0.5;
2271 #endif /*Check_FLT_ROUNDS*/
2272 			}
2273 		y = word0(rv) & Exp_mask;
2274 
2275 		/* Check for overflow */
2276 
2277 		if (y == Exp_msk1*(DBL_MAX_EXP+Bias-1)) {
2278 			dval(rv0) = dval(rv);
2279 			word0(rv) -= P*Exp_msk1;
2280 			adj = aadj1 * ulp(dval(rv));
2281 			dval(rv) += adj;
2282 			if ((word0(rv) & Exp_mask) >=
2283 					Exp_msk1*(DBL_MAX_EXP+Bias-P)) {
2284 				if (word0(rv0) == Big0 && word1(rv0) == Big1)
2285 					goto ovfl;
2286 				word0(rv) = Big0;
2287 				word1(rv) = Big1;
2288 				goto cont;
2289 				}
2290 			else
2291 				word0(rv) += P*Exp_msk1;
2292 			}
2293 		else {
2294 #ifdef Avoid_Underflow
2295 			if (scale && y <= 2*P*Exp_msk1) {
2296 				if (aadj <= 0x7fffffff) {
2297 					if ((z = aadj) <= 0)
2298 						z = 1;
2299 					aadj = z;
2300 					aadj1 = dsign ? aadj : -aadj;
2301 					}
2302 				word0(aadj1) += (2*P+1)*Exp_msk1 - y;
2303 				}
2304 			adj = aadj1 * ulp(dval(rv));
2305 			dval(rv) += adj;
2306 #else
2307 #ifdef Sudden_Underflow
2308 			if ((word0(rv) & Exp_mask) <= P*Exp_msk1) {
2309 				dval(rv0) = dval(rv);
2310 				word0(rv) += P*Exp_msk1;
2311 				adj = aadj1 * ulp(dval(rv));
2312 				dval(rv) += adj;
2313 #ifdef IBM
2314 				if ((word0(rv) & Exp_mask) <  P*Exp_msk1)
2315 #else
2316 				if ((word0(rv) & Exp_mask) <= P*Exp_msk1)
2317 #endif
2318 					{
2319 					if (word0(rv0) == Tiny0
2320 					 && word1(rv0) == Tiny1)
2321 						goto undfl;
2322 					word0(rv) = Tiny0;
2323 					word1(rv) = Tiny1;
2324 					goto cont;
2325 					}
2326 				else
2327 					word0(rv) -= P*Exp_msk1;
2328 				}
2329 			else {
2330 				adj = aadj1 * ulp(dval(rv));
2331 				dval(rv) += adj;
2332 				}
2333 #else /*Sudden_Underflow*/
2334 			/* Compute adj so that the IEEE rounding rules will
2335 			 * correctly round rv + adj in some half-way cases.
2336 			 * If rv * ulp(rv) is denormalized (i.e.,
2337 			 * y <= (P-1)*Exp_msk1), we must adjust aadj to avoid
2338 			 * trouble from bits lost to denormalization;
2339 			 * example: 1.2e-307 .
2340 			 */
2341 			if (y <= (P-1)*Exp_msk1 && aadj > 1.) {
2342 				aadj1 = (double)(int)(aadj + 0.5);
2343 				if (!dsign)
2344 					aadj1 = -aadj1;
2345 				}
2346 			adj = aadj1 * ulp(dval(rv));
2347 			dval(rv) += adj;
2348 #endif /*Sudden_Underflow*/
2349 #endif /*Avoid_Underflow*/
2350 			}
2351 		z = word0(rv) & Exp_mask;
2352 #ifndef SET_INEXACT
2353 #ifdef Avoid_Underflow
2354 		if (!scale)
2355 #endif
2356 		if (y == z) {
2357 			/* Can we stop now? */
2358 			L = (Long)aadj;
2359 			aadj -= L;
2360 			/* The tolerances below are conservative. */
2361 			if (dsign || word1(rv) || word0(rv) & Bndry_mask) {
2362 				if (aadj < .4999999 || aadj > .5000001)
2363 					break;
2364 				}
2365 			else if (aadj < .4999999/FLT_RADIX)
2366 				break;
2367 			}
2368 #endif
2369  cont:
2370 		Bfree(bb);
2371 		Bfree(bd);
2372 		Bfree(bs);
2373 		Bfree(delta);
2374 		}
2375 #ifdef SET_INEXACT
2376 	if (inexact) {
2377 		if (!oldinexact) {
2378 			word0(rv0) = Exp_1 + (70 << Exp_shift);
2379 			word1(rv0) = 0;
2380 			dval(rv0) += 1.;
2381 			}
2382 		}
2383 	else if (!oldinexact)
2384 		clear_inexact();
2385 #endif
2386 #ifdef Avoid_Underflow
2387 	if (scale) {
2388 		word0(rv0) = Exp_1 - 2*P*Exp_msk1;
2389 		word1(rv0) = 0;
2390 		dval(rv) *= dval(rv0);
2391 #ifndef NO_ERRNO
2392 		/* try to avoid the bug of testing an 8087 register value */
2393 		if (word0(rv) == 0 && word1(rv) == 0)
2394 			errno = ERANGE;
2395 #endif
2396 		}
2397 #endif /* Avoid_Underflow */
2398 #ifdef SET_INEXACT
2399 	if (inexact && !(word0(rv) & Exp_mask)) {
2400 		/* set underflow bit */
2401 		dval(rv0) = 1e-300;
2402 		dval(rv0) *= dval(rv0);
2403 		}
2404 #endif
2405  retfree:
2406 	Bfree(bb);
2407 	Bfree(bd);
2408 	Bfree(bs);
2409 	Bfree(bd0);
2410 	Bfree(delta);
2411  ret:
2412 	if (se)
2413 		*se = (char *)s;
2414 	return sign ? -dval(rv) : dval(rv);
2415 	}
2416 
2417  static int
quorem(b,S)2418 quorem
2419 #ifdef KR_headers
2420 	(b, S) Bigint *b, *S;
2421 #else
2422 	(Bigint *b, Bigint *S)
2423 #endif
2424 {
2425 	int n;
2426 	ULong *bx, *bxe, q, *sx, *sxe;
2427 #ifdef ULLong
2428 	ULLong borrow, carry, y, ys;
2429 #else
2430 	ULong borrow, carry, y, ys;
2431 #ifdef Pack_32
2432 	ULong si, z, zs;
2433 #endif
2434 #endif
2435 
2436 	n = S->wds;
2437 #ifdef DEBUG
2438 	/*debug*/ if (b->wds > n)
2439 	/*debug*/	Bug("oversize b in quorem");
2440 #endif
2441 	if (b->wds < n)
2442 		return 0;
2443 	sx = S->x;
2444 	sxe = sx + --n;
2445 	bx = b->x;
2446 	bxe = bx + n;
2447 	q = *bxe / (*sxe + 1);	/* ensure q <= true quotient */
2448 #ifdef DEBUG
2449 	/*debug*/ if (q > 9)
2450 	/*debug*/	Bug("oversized quotient in quorem");
2451 #endif
2452 	if (q) {
2453 		borrow = 0;
2454 		carry = 0;
2455 		do {
2456 #ifdef ULLong
2457 			ys = *sx++ * (ULLong)q + carry;
2458 			carry = ys >> 32;
2459 			y = *bx - (ys & FFFFFFFF) - borrow;
2460 			borrow = y >> 32 & (ULong)1;
2461 			*bx++ = y & FFFFFFFF;
2462 #else
2463 #ifdef Pack_32
2464 			si = *sx++;
2465 			ys = (si & 0xffff) * q + carry;
2466 			zs = (si >> 16) * q + (ys >> 16);
2467 			carry = zs >> 16;
2468 			y = (*bx & 0xffff) - (ys & 0xffff) - borrow;
2469 			borrow = (y & 0x10000) >> 16;
2470 			z = (*bx >> 16) - (zs & 0xffff) - borrow;
2471 			borrow = (z & 0x10000) >> 16;
2472 			Storeinc(bx, z, y);
2473 #else
2474 			ys = *sx++ * q + carry;
2475 			carry = ys >> 16;
2476 			y = *bx - (ys & 0xffff) - borrow;
2477 			borrow = (y & 0x10000) >> 16;
2478 			*bx++ = y & 0xffff;
2479 #endif
2480 #endif
2481 			}
2482 			while(sx <= sxe);
2483 		if (!*bxe) {
2484 			bx = b->x;
2485 			while(--bxe > bx && !*bxe)
2486 				--n;
2487 			b->wds = n;
2488 			}
2489 		}
2490 	if (cmp(b, S) >= 0) {
2491 		q++;
2492 		borrow = 0;
2493 		carry = 0;
2494 		bx = b->x;
2495 		sx = S->x;
2496 		do {
2497 #ifdef ULLong
2498 			ys = *sx++ + carry;
2499 			carry = ys >> 32;
2500 			y = *bx - (ys & FFFFFFFF) - borrow;
2501 			borrow = y >> 32 & (ULong)1;
2502 			*bx++ = y & FFFFFFFF;
2503 #else
2504 #ifdef Pack_32
2505 			si = *sx++;
2506 			ys = (si & 0xffff) + carry;
2507 			zs = (si >> 16) + (ys >> 16);
2508 			carry = zs >> 16;
2509 			y = (*bx & 0xffff) - (ys & 0xffff) - borrow;
2510 			borrow = (y & 0x10000) >> 16;
2511 			z = (*bx >> 16) - (zs & 0xffff) - borrow;
2512 			borrow = (z & 0x10000) >> 16;
2513 			Storeinc(bx, z, y);
2514 #else
2515 			ys = *sx++ + carry;
2516 			carry = ys >> 16;
2517 			y = *bx - (ys & 0xffff) - borrow;
2518 			borrow = (y & 0x10000) >> 16;
2519 			*bx++ = y & 0xffff;
2520 #endif
2521 #endif
2522 			}
2523 			while(sx <= sxe);
2524 		bx = b->x;
2525 		bxe = bx + n;
2526 		if (!*bxe) {
2527 			while(--bxe > bx && !*bxe)
2528 				--n;
2529 			b->wds = n;
2530 			}
2531 		}
2532 	return q;
2533 	}
2534 
2535 #ifndef MULTIPLE_THREADS
2536  static char *dtoa_result;
2537 #endif
2538 
2539  static char *
2540 #ifdef KR_headers
rv_alloc(i)2541 rv_alloc(i) int i;
2542 #else
2543 rv_alloc(int i)
2544 #endif
2545 {
2546 	int j, k, *r;
2547 
2548 	j = sizeof(ULong);
2549 	for(k = 0;
2550 		sizeof(Bigint) - sizeof(ULong) - sizeof(int) + j <= i;
2551 		j <<= 1)
2552 			k++;
2553 	r = (int*)Balloc(k);
2554 	*r = k;
2555 	return
2556 #ifndef MULTIPLE_THREADS
2557 	dtoa_result =
2558 #endif
2559 		(char *)(r+1);
2560 	}
2561 
2562  static char *
2563 #ifdef KR_headers
nrv_alloc(s,rve,n)2564 nrv_alloc(s, rve, n) char *s, **rve; int n;
2565 #else
2566 nrv_alloc(const char *s, char **rve, int n)
2567 #endif
2568 {
2569 	char *rv, *t;
2570 
2571 	t = rv = rv_alloc(n);
2572 	while ((*t = *s++)) t++;
2573 	if (rve)
2574 		*rve = t;
2575 	return rv;
2576 	}
2577 
2578 /* freedtoa(s) must be used to free values s returned by dtoa
2579  * when MULTIPLE_THREADS is #defined.  It should be used in all cases,
2580  * but for consistency with earlier versions of dtoa, it is optional
2581  * when MULTIPLE_THREADS is not defined.
2582  */
2583 
2584  void
2585 #ifdef KR_headers
freedtoa(s)2586 freedtoa(s) char *s;
2587 #else
2588 freedtoa(char *s)
2589 #endif
2590 {
2591 	Bigint *b = (Bigint *)((int *)s - 1);
2592 	b->maxwds = 1 << (b->k = *(int*)b);
2593 	Bfree(b);
2594 #ifndef MULTIPLE_THREADS
2595 	if (s == dtoa_result)
2596 		dtoa_result = 0;
2597 #endif
2598 	}
2599 
2600 /* dtoa for IEEE arithmetic (dmg): convert double to ASCII string.
2601  *
2602  * Inspired by "How to Print Floating-Point Numbers Accurately" by
2603  * Guy L. Steele, Jr. and Jon L. White [Proc. ACM SIGPLAN '90, pp. 112-126].
2604  *
2605  * Modifications:
2606  *	1. Rather than iterating, we use a simple numeric overestimate
2607  *	   to determine k = floor(log10(d)).  We scale relevant
2608  *	   quantities using O(log2(k)) rather than O(k) multiplications.
2609  *	2. For some modes > 2 (corresponding to ecvt and fcvt), we don't
2610  *	   try to generate digits strictly left to right.  Instead, we
2611  *	   compute with fewer bits and propagate the carry if necessary
2612  *	   when rounding the final digit up.  This is often faster.
2613  *	3. Under the assumption that input will be rounded nearest,
2614  *	   mode 0 renders 1e23 as 1e23 rather than 9.999999999999999e22.
2615  *	   That is, we allow equality in stopping tests when the
2616  *	   round-nearest rule will give the same floating-point value
2617  *	   as would satisfaction of the stopping test with strict
2618  *	   inequality.
2619  *	4. We remove common factors of powers of 2 from relevant
2620  *	   quantities.
2621  *	5. When converting floating-point integers less than 1e16,
2622  *	   we use floating-point arithmetic rather than resorting
2623  *	   to multiple-precision integers.
2624  *	6. When asked to produce fewer than 15 digits, we first try
2625  *	   to get by with floating-point arithmetic; we resort to
2626  *	   multiple-precision integer arithmetic only if we cannot
2627  *	   guarantee that the floating-point calculation has given
2628  *	   the correctly rounded result.  For k requested digits and
2629  *	   "uniformly" distributed input, the probability is
2630  *	   something like 10^(k-15) that we must resort to the Long
2631  *	   calculation.
2632  */
2633 
2634  char *
dtoa(d,mode,ndigits,decpt,sign,rve)2635 dtoa
2636 #ifdef KR_headers
2637 	(d, mode, ndigits, decpt, sign, rve)
2638 	double d; int mode, ndigits, *decpt, *sign; char **rve;
2639 #else
2640 	(double d, int mode, int ndigits, int *decpt, int *sign, char **rve)
2641 #endif
2642 {
2643  /*	Arguments ndigits, decpt, sign are similar to those
2644 	of ecvt and fcvt; trailing zeros are suppressed from
2645 	the returned string.  If not null, *rve is set to point
2646 	to the end of the return value.  If d is +-Infinity or NaN,
2647 	then *decpt is set to 9999.
2648 
2649 	mode:
2650 		0 ==> shortest string that yields d when read in
2651 			and rounded to nearest.
2652 		1 ==> like 0, but with Steele & White stopping rule;
2653 			e.g. with IEEE P754 arithmetic , mode 0 gives
2654 			1e23 whereas mode 1 gives 9.999999999999999e22.
2655 		2 ==> max(1,ndigits) significant digits.  This gives a
2656 			return value similar to that of ecvt, except
2657 			that trailing zeros are suppressed.
2658 		3 ==> through ndigits past the decimal point.  This
2659 			gives a return value similar to that from fcvt,
2660 			except that trailing zeros are suppressed, and
2661 			ndigits can be negative.
2662 		4,5 ==> similar to 2 and 3, respectively, but (in
2663 			round-nearest mode) with the tests of mode 0 to
2664 			possibly return a shorter string that rounds to d.
2665 			With IEEE arithmetic and compilation with
2666 			-DHonor_FLT_ROUNDS, modes 4 and 5 behave the same
2667 			as modes 2 and 3 when FLT_ROUNDS != 1.
2668 		6-9 ==> Debugging modes similar to mode - 4:  don't try
2669 			fast floating-point estimate (if applicable).
2670 
2671 		Values of mode other than 0-9 are treated as mode 0.
2672 
2673 		Sufficient space is allocated to the return value
2674 		to hold the suppressed trailing zeros.
2675 	*/
2676 
2677 	int bbits, b2, b5, be, dig, i, ieps, ilim, ilim0, ilim1,
2678 		j, j1, k, k0, k_check, leftright, m2, m5, s2, s5,
2679 		spec_case, try_quick, bias_round_up;
2680 	Long L;
2681 #ifndef Sudden_Underflow
2682 	int denorm;
2683 	ULong x;
2684 #endif
2685 	Bigint *b, *b1, *delta, *mlo, *mhi, *S;
2686 	double d2, ds, eps;
2687 	char *s, *s0;
2688 #ifdef Honor_FLT_ROUNDS
2689 	int rounding;
2690 #endif
2691 #ifdef SET_INEXACT
2692 	int inexact, oldinexact;
2693 #endif
2694 
2695         /* In mode 2 and 3 we bias rounding up when there are ties. */
2696         bias_round_up = mode == 2 || mode == 3;
2697 
2698         ilim = ilim1 = 0; /* to avoid Google3 compiler warnings */
2699 
2700 #ifndef MULTIPLE_THREADS
2701 	if (dtoa_result) {
2702 		freedtoa(dtoa_result);
2703 		dtoa_result = 0;
2704 		}
2705 #endif
2706 
2707 	if (word0(d) & Sign_bit) {
2708 		/* set sign for everything, including 0's and NaNs */
2709 		*sign = 1;
2710 		word0(d) &= ~Sign_bit;	/* clear sign bit */
2711 		}
2712 	else
2713 		*sign = 0;
2714 
2715 #if defined(IEEE_Arith) + defined(VAX)
2716 #ifdef IEEE_Arith
2717 	if ((word0(d) & Exp_mask) == Exp_mask)
2718 #else
2719 	if (word0(d)  == 0x8000)
2720 #endif
2721 		{
2722 		/* Infinity or NaN */
2723 		*decpt = 9999;
2724 #ifdef IEEE_Arith
2725 		if (!word1(d) && !(word0(d) & 0xfffff))
2726 			return nrv_alloc("Infinity", rve, 8);
2727 #endif
2728 		return nrv_alloc("NaN", rve, 3);
2729 		}
2730 #endif
2731 #ifdef IBM
2732 	dval(d) += 0; /* normalize */
2733 #endif
2734 	if (!dval(d)) {
2735 		*decpt = 1;
2736 		return nrv_alloc("0", rve, 1);
2737 		}
2738 
2739 #ifdef SET_INEXACT
2740 	try_quick = oldinexact = get_inexact();
2741 	inexact = 1;
2742 #endif
2743 #ifdef Honor_FLT_ROUNDS
2744 	if ((rounding = Flt_Rounds) >= 2) {
2745 		if (*sign)
2746 			rounding = rounding == 2 ? 0 : 2;
2747 		else
2748 			if (rounding != 2)
2749 				rounding = 0;
2750 		}
2751 #endif
2752 
2753 	b = d2b(dval(d), &be, &bbits);
2754 #ifdef Sudden_Underflow
2755 	i = (int)(word0(d) >> Exp_shift1 & (Exp_mask>>Exp_shift1));
2756 #else
2757 	if ((i = (int)(word0(d) >> Exp_shift1 & (Exp_mask>>Exp_shift1)))) {
2758 #endif
2759 		dval(d2) = dval(d);
2760 		word0(d2) &= Frac_mask1;
2761 		word0(d2) |= Exp_11;
2762 #ifdef IBM
2763 		if (j = 11 - hi0bits(word0(d2) & Frac_mask))
2764 			dval(d2) /= 1 << j;
2765 #endif
2766 
2767 		/* log(x)	~=~ log(1.5) + (x-1.5)/1.5
2768 		 * log10(x)	 =  log(x) / log(10)
2769 		 *		~=~ log(1.5)/log(10) + (x-1.5)/(1.5*log(10))
2770 		 * log10(d) = (i-Bias)*log(2)/log(10) + log10(d2)
2771 		 *
2772 		 * This suggests computing an approximation k to log10(d) by
2773 		 *
2774 		 * k = (i - Bias)*0.301029995663981
2775 		 *	+ ( (d2-1.5)*0.289529654602168 + 0.176091259055681 );
2776 		 *
2777 		 * We want k to be too large rather than too small.
2778 		 * The error in the first-order Taylor series approximation
2779 		 * is in our favor, so we just round up the constant enough
2780 		 * to compensate for any error in the multiplication of
2781 		 * (i - Bias) by 0.301029995663981; since |i - Bias| <= 1077,
2782 		 * and 1077 * 0.30103 * 2^-52 ~=~ 7.2e-14,
2783 		 * adding 1e-13 to the constant term more than suffices.
2784 		 * Hence we adjust the constant term to 0.1760912590558.
2785 		 * (We could get a more accurate k by invoking log10,
2786 		 *  but this is probably not worthwhile.)
2787 		 */
2788 
2789 		i -= Bias;
2790 #ifdef IBM
2791 		i <<= 2;
2792 		i += j;
2793 #endif
2794 #ifndef Sudden_Underflow
2795 		denorm = 0;
2796 		}
2797 	else {
2798 		/* d is denormalized */
2799 
2800 		i = bbits + be + (Bias + (P-1) - 1);
2801 		x = i > 32  ? (word0(d) << (64 - i)) | (word1(d) >> (i - 32))
2802 			    : word1(d) << (32 - i);
2803 		dval(d2) = x;
2804 		word0(d2) -= 31*Exp_msk1; /* adjust exponent */
2805 		i -= (Bias + (P-1) - 1) + 1;
2806 		denorm = 1;
2807 		}
2808 #endif
2809 	ds = (dval(d2)-1.5)*0.289529654602168 + 0.1760912590558 + i*0.301029995663981;
2810 	k = (int)ds;
2811 	if (ds < 0. && ds != k)
2812 		k--;	/* want k = floor(ds) */
2813 	k_check = 1;
2814 	if (k >= 0 && k <= Ten_pmax) {
2815 		if (dval(d) < tens[k])
2816 			k--;
2817 		k_check = 0;
2818 		}
2819 	j = bbits - i - 1;
2820 	if (j >= 0) {
2821 		b2 = 0;
2822 		s2 = j;
2823 		}
2824 	else {
2825 		b2 = -j;
2826 		s2 = 0;
2827 		}
2828 	if (k >= 0) {
2829 		b5 = 0;
2830 		s5 = k;
2831 		s2 += k;
2832 		}
2833 	else {
2834 		b2 -= k;
2835 		b5 = -k;
2836 		s5 = 0;
2837 		}
2838 	if (mode < 0 || mode > 9)
2839 		mode = 0;
2840 
2841 #ifndef SET_INEXACT
2842 #ifdef Check_FLT_ROUNDS
2843 	try_quick = Rounding == 1;
2844 #else
2845 	try_quick = 1;
2846 #endif
2847 #endif /*SET_INEXACT*/
2848 
2849 	if (mode > 5) {
2850 		mode -= 4;
2851 		try_quick = 0;
2852 		}
2853 	leftright = 1;
2854 	switch(mode) {
2855 		case 0:
2856 		case 1:
2857 			ilim = ilim1 = -1;
2858 			i = 18;
2859 			ndigits = 0;
2860 			break;
2861 		case 2:
2862 			leftright = 0;
2863 			/* no break */
2864 		case 4:
2865 			if (ndigits <= 0)
2866 				ndigits = 1;
2867 			ilim = ilim1 = i = ndigits;
2868 			break;
2869 		case 3:
2870 			leftright = 0;
2871 			/* no break */
2872 		case 5:
2873 			i = ndigits + k + 1;
2874 			ilim = i;
2875 			ilim1 = i - 1;
2876 			if (i <= 0)
2877 				i = 1;
2878 		}
2879 	s = s0 = rv_alloc(i);
2880 
2881 #ifdef Honor_FLT_ROUNDS
2882 	if (mode > 1 && rounding != 1)
2883 		leftright = 0;
2884 #endif
2885 
2886 	if (ilim >= 0 && ilim <= Quick_max && try_quick) {
2887 
2888 		/* Try to get by with floating-point arithmetic. */
2889 
2890 		i = 0;
2891 		dval(d2) = dval(d);
2892 		k0 = k;
2893 		ilim0 = ilim;
2894 		ieps = 2; /* conservative */
2895 		if (k > 0) {
2896 			ds = tens[k&0xf];
2897 			j = k >> 4;
2898 			if (j & Bletch) {
2899 				/* prevent overflows */
2900 				j &= Bletch - 1;
2901 				dval(d) /= bigtens[n_bigtens-1];
2902 				ieps++;
2903 				}
2904 			for(; j; j >>= 1, i++)
2905 				if (j & 1) {
2906 					ieps++;
2907 					ds *= bigtens[i];
2908 					}
2909 			dval(d) /= ds;
2910 			}
2911 		else if ((j1 = -k)) {
2912 			dval(d) *= tens[j1 & 0xf];
2913 			for(j = j1 >> 4; j; j >>= 1, i++)
2914 				if (j & 1) {
2915 					ieps++;
2916 					dval(d) *= bigtens[i];
2917 					}
2918 			}
2919 		if (k_check && dval(d) < 1. && ilim > 0) {
2920 			if (ilim1 <= 0)
2921 				goto fast_failed;
2922 			ilim = ilim1;
2923 			k--;
2924 			dval(d) *= 10.;
2925 			ieps++;
2926 			}
2927 		dval(eps) = ieps*dval(d) + 7.;
2928 		word0(eps) -= (P-1)*Exp_msk1;
2929 		if (ilim == 0) {
2930 			S = mhi = 0;
2931 			dval(d) -= 5.;
2932 			if (dval(d) > dval(eps))
2933 				goto one_digit;
2934 			if (dval(d) < -dval(eps))
2935 				goto no_digits;
2936 			goto fast_failed;
2937 			}
2938 #ifndef No_leftright
2939 		if (leftright) {
2940 			/* Use Steele & White method of only
2941 			 * generating digits needed.
2942 			 */
2943 			dval(eps) = 0.5/tens[ilim-1] - dval(eps);
2944 			for(i = 0;;) {
2945 				L = dval(d);
2946 				dval(d) -= L;
2947 				*s++ = '0' + (int)L;
2948 				if (dval(d) < dval(eps))
2949 					goto ret1;
2950 				if (1. - dval(d) < dval(eps))
2951 					goto bump_up;
2952 				if (++i >= ilim)
2953 					break;
2954 				dval(eps) *= 10.;
2955 				dval(d) *= 10.;
2956 				}
2957 			}
2958 		else {
2959 #endif
2960 			/* Generate ilim digits, then fix them up. */
2961 			dval(eps) *= tens[ilim-1];
2962 			for(i = 1;; i++, dval(d) *= 10.) {
2963 				L = (Long)(dval(d));
2964 				if (!(dval(d) -= L))
2965 					ilim = i;
2966 				*s++ = '0' + (int)L;
2967 				if (i == ilim) {
2968 					if (dval(d) > 0.5 + dval(eps))
2969 						goto bump_up;
2970 					else if (dval(d) < 0.5 - dval(eps)) {
2971 						while(*--s == '0');
2972 						s++;
2973 						goto ret1;
2974 						}
2975 					break;
2976 					}
2977 				}
2978 #ifndef No_leftright
2979 			}
2980 #endif
2981  fast_failed:
2982 		s = s0;
2983 		dval(d) = dval(d2);
2984 		k = k0;
2985 		ilim = ilim0;
2986 		}
2987 
2988 	/* Do we have a "small" integer? */
2989 
2990 	if (be >= 0 && k <= Int_max) {
2991 		/* Yes. */
2992 		ds = tens[k];
2993 		if (ndigits < 0 && ilim <= 0) {
2994 			S = mhi = 0;
2995 			if (ilim < 0 || dval(d) < 5*ds || ((dval(d) == 5*ds) && !bias_round_up))
2996 				goto no_digits;
2997 			goto one_digit;
2998 			}
2999 
3000                 /* Limit looping by the number of digits to produce.
3001                  * Firefox had a crash bug because some plugins reduce
3002                  * the precision of double arithmetic.  With reduced
3003                  * precision "dval(d) -= L*ds" might be imprecise and
3004                  * d might not become zero and the loop might not
3005                  * terminate.
3006                  *
3007                  * See https://bugzilla.mozilla.org/show_bug.cgi?id=358569
3008                  */
3009 		for(i = 1; i <= k+1; i++, dval(d) *= 10.) {
3010 			L = (Long)(dval(d) / ds);
3011 			dval(d) -= L*ds;
3012 #ifdef Check_FLT_ROUNDS
3013 			/* If FLT_ROUNDS == 2, L will usually be high by 1 */
3014 			if (dval(d) < 0) {
3015 				L--;
3016 				dval(d) += ds;
3017 				}
3018 #endif
3019 			*s++ = '0' + (int)L;
3020 			if (!dval(d)) {
3021 #ifdef SET_INEXACT
3022 				inexact = 0;
3023 #endif
3024 				break;
3025 				}
3026 			if (i == ilim) {
3027 #ifdef Honor_FLT_ROUNDS
3028 				if (mode > 1)
3029 				switch(rounding) {
3030 				  case 0: goto ret1;
3031 				  case 2: goto bump_up;
3032 				  }
3033 #endif
3034 				dval(d) += dval(d);
3035 				if (dval(d) > ds || (dval(d) == ds && ((L & 1) || bias_round_up))) {
3036  bump_up:
3037 					while(*--s == '9')
3038 						if (s == s0) {
3039 							k++;
3040 							*s = '0';
3041 							break;
3042 							}
3043 					++*s++;
3044 					}
3045 				break;
3046 				}
3047 			}
3048 		goto ret1;
3049 		}
3050 
3051 	m2 = b2;
3052 	m5 = b5;
3053 	mhi = mlo = 0;
3054 	if (leftright) {
3055 		i =
3056 #ifndef Sudden_Underflow
3057 			denorm ? be + (Bias + (P-1) - 1 + 1) :
3058 #endif
3059 #ifdef IBM
3060 			1 + 4*P - 3 - bbits + ((bbits + be - 1) & 3);
3061 #else
3062 			1 + P - bbits;
3063 #endif
3064 		b2 += i;
3065 		s2 += i;
3066 		mhi = i2b(1);
3067 		}
3068 	if (m2 > 0 && s2 > 0) {
3069 		i = m2 < s2 ? m2 : s2;
3070 		b2 -= i;
3071 		m2 -= i;
3072 		s2 -= i;
3073 		}
3074 	if (b5 > 0) {
3075 		if (leftright) {
3076 			if (m5 > 0) {
3077 				mhi = pow5mult(mhi, m5);
3078 				b1 = mult(mhi, b);
3079 				Bfree(b);
3080 				b = b1;
3081 				}
3082 			if ((j = b5 - m5))
3083 				b = pow5mult(b, j);
3084 			}
3085 		else
3086 			b = pow5mult(b, b5);
3087 		}
3088 	S = i2b(1);
3089 	if (s5 > 0)
3090 		S = pow5mult(S, s5);
3091 
3092 	/* Check for special case that d is a normalized power of 2. */
3093 
3094 	spec_case = 0;
3095 	if ((mode < 2 || leftright)
3096 #ifdef Honor_FLT_ROUNDS
3097 			&& rounding == 1
3098 #endif
3099 				) {
3100 		if (!word1(d) && !(word0(d) & Bndry_mask)
3101 #ifndef Sudden_Underflow
3102 		 && word0(d) & (Exp_mask & ~Exp_msk1)
3103 #endif
3104 				) {
3105 			/* The special case */
3106 			b2 += Log2P;
3107 			s2 += Log2P;
3108 			spec_case = 1;
3109 			}
3110 		}
3111 
3112 	/* Arrange for convenient computation of quotients:
3113 	 * shift left if necessary so divisor has 4 leading 0 bits.
3114 	 *
3115 	 * Perhaps we should just compute leading 28 bits of S once
3116 	 * and for all and pass them and a shift to quorem, so it
3117 	 * can do shifts and ors to compute the numerator for q.
3118 	 */
3119 #ifdef Pack_32
3120 	if ((i = ((s5 ? 32 - hi0bits(S->x[S->wds-1]) : 1) + s2) & 0x1f))
3121 		i = 32 - i;
3122 #else
3123 	if ((i = ((s5 ? 32 - hi0bits(S->x[S->wds-1]) : 1) + s2) & 0xf))
3124 		i = 16 - i;
3125 #endif
3126 	if (i > 4) {
3127 		i -= 4;
3128 		b2 += i;
3129 		m2 += i;
3130 		s2 += i;
3131 		}
3132 	else if (i < 4) {
3133 		i += 28;
3134 		b2 += i;
3135 		m2 += i;
3136 		s2 += i;
3137 		}
3138 	if (b2 > 0)
3139 		b = lshift(b, b2);
3140 	if (s2 > 0)
3141 		S = lshift(S, s2);
3142 	if (k_check) {
3143 		if (cmp(b,S) < 0) {
3144 			k--;
3145 			b = multadd(b, 10, 0);	/* we botched the k estimate */
3146 			if (leftright)
3147 				mhi = multadd(mhi, 10, 0);
3148 			ilim = ilim1;
3149 			}
3150 		}
3151 	if (ilim <= 0 && (mode == 3 || mode == 5)) {
3152                 S = multadd(S, 5, 0);
3153 		if (ilim < 0 || cmp(b, S) < 0 || ((cmp(b, S) == 0) && !bias_round_up)) {
3154 			/* no digits, fcvt style */
3155  no_digits:
3156 			k = -1 - ndigits;
3157 			goto ret;
3158 			}
3159  one_digit:
3160 		*s++ = '1';
3161 		k++;
3162 		goto ret;
3163 		}
3164 	if (leftright) {
3165 		if (m2 > 0)
3166 			mhi = lshift(mhi, m2);
3167 
3168 		/* Compute mlo -- check for special case
3169 		 * that d is a normalized power of 2.
3170 		 */
3171 
3172 		mlo = mhi;
3173 		if (spec_case) {
3174 			mhi = Balloc(mhi->k);
3175 			Bcopy(mhi, mlo);
3176 			mhi = lshift(mhi, Log2P);
3177 			}
3178 
3179 		for(i = 1;;i++) {
3180 			dig = quorem(b,S) + '0';
3181 			/* Do we yet have the shortest decimal string
3182 			 * that will round to d?
3183 			 */
3184 			j = cmp(b, mlo);
3185 			delta = diff(S, mhi);
3186 			j1 = delta->sign ? 1 : cmp(b, delta);
3187 			Bfree(delta);
3188 #ifndef ROUND_BIASED
3189 			if (j1 == 0 && mode != 1 && !(word1(d) & 1)
3190 #ifdef Honor_FLT_ROUNDS
3191 				&& rounding >= 1
3192 #endif
3193 								   ) {
3194 				if (dig == '9')
3195 					goto round_9_up;
3196 				if (j > 0)
3197 					dig++;
3198 #ifdef SET_INEXACT
3199 				else if (!b->x[0] && b->wds <= 1)
3200 					inexact = 0;
3201 #endif
3202 				*s++ = dig;
3203 				goto ret;
3204 				}
3205 #endif
3206 			if (j < 0 || (j == 0 && mode != 1
3207 #ifndef ROUND_BIASED
3208 							&& !(word1(d) & 1)
3209 #endif
3210 					)) {
3211 				if (!b->x[0] && b->wds <= 1) {
3212 #ifdef SET_INEXACT
3213 					inexact = 0;
3214 #endif
3215 					goto accept_dig;
3216 					}
3217 #ifdef Honor_FLT_ROUNDS
3218 				if (mode > 1)
3219 				 switch(rounding) {
3220 				  case 0: goto accept_dig;
3221 				  case 2: goto keep_dig;
3222 				  }
3223 #endif /*Honor_FLT_ROUNDS*/
3224 				if (j1 > 0) {
3225 					b = lshift(b, 1);
3226 					j1 = cmp(b, S);
3227 					if ((j1 > 0 || (j1 == 0 && ((dig & 1) || bias_round_up)))
3228                                             && dig++ == '9')
3229 						goto round_9_up;
3230 					}
3231  accept_dig:
3232 				*s++ = dig;
3233 				goto ret;
3234 				}
3235 			if (j1 > 0) {
3236 #ifdef Honor_FLT_ROUNDS
3237 				if (!rounding)
3238 					goto accept_dig;
3239 #endif
3240 				if (dig == '9') { /* possible if i == 1 */
3241  round_9_up:
3242 					*s++ = '9';
3243 					goto roundoff;
3244 					}
3245 				*s++ = dig + 1;
3246 				goto ret;
3247 				}
3248 #ifdef Honor_FLT_ROUNDS
3249  keep_dig:
3250 #endif
3251 			*s++ = dig;
3252 			if (i == ilim)
3253 				break;
3254 			b = multadd(b, 10, 0);
3255 			if (mlo == mhi)
3256 				mlo = mhi = multadd(mhi, 10, 0);
3257 			else {
3258 				mlo = multadd(mlo, 10, 0);
3259 				mhi = multadd(mhi, 10, 0);
3260 				}
3261 			}
3262 		}
3263 	else
3264 		for(i = 1;; i++) {
3265 			*s++ = dig = quorem(b,S) + '0';
3266 			if (!b->x[0] && b->wds <= 1) {
3267 #ifdef SET_INEXACT
3268 				inexact = 0;
3269 #endif
3270 				goto ret;
3271 				}
3272 			if (i >= ilim)
3273 				break;
3274 			b = multadd(b, 10, 0);
3275 			}
3276 
3277 	/* Round off last digit */
3278 
3279 #ifdef Honor_FLT_ROUNDS
3280 	switch(rounding) {
3281 	  case 0: goto trimzeros;
3282 	  case 2: goto roundoff;
3283 	  }
3284 #endif
3285 	b = lshift(b, 1);
3286 	j = cmp(b, S);
3287 	if (j > 0 || (j == 0 && ((dig & 1) || bias_round_up))) {
3288  roundoff:
3289 		while(*--s == '9')
3290 			if (s == s0) {
3291 				k++;
3292 				*s++ = '1';
3293 				goto ret;
3294 				}
3295 		++*s++;
3296 		}
3297 	else {
3298 /* trimzeros:  (never used) */
3299 		while(*--s == '0');
3300 		s++;
3301 		}
3302  ret:
3303 	Bfree(S);
3304 	if (mhi) {
3305 		if (mlo && mlo != mhi)
3306 			Bfree(mlo);
3307 		Bfree(mhi);
3308 		}
3309  ret1:
3310 #ifdef SET_INEXACT
3311 	if (inexact) {
3312 		if (!oldinexact) {
3313 			word0(d) = Exp_1 + (70 << Exp_shift);
3314 			word1(d) = 0;
3315 			dval(d) += 1.;
3316 			}
3317 		}
3318 	else if (!oldinexact)
3319 		clear_inexact();
3320 #endif
3321 	Bfree(b);
3322 	*s = 0;
3323 	*decpt = k + 1;
3324 	if (rve)
3325 		*rve = s;
3326 	return s0;
3327 	}
3328 #ifdef __cplusplus
3329 }
3330 #endif
3331