Home | History | Annotate | Download | only in dtoa
      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 *
    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
    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 *
    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 *
    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
    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
    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 *
    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 *
    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 *
    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 *
    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
    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 *
   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
   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
   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 *
   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
   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
   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
   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
   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
   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
   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
   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
   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 *
   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