1 | /* Copyright (c) 2007, 2012, Oracle and/or its affiliates. All rights reserved. |
2 | 2016,2018 MariaDB Corporation AB |
3 | |
4 | This library is free software; you can redistribute it and/or |
5 | modify it under the terms of the GNU Library General Public |
6 | License as published by the Free Software Foundation; version 2 |
7 | of the License. |
8 | |
9 | This program is distributed in the hope that it will be useful, |
10 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
11 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
12 | GNU General Public License for more details. |
13 | |
14 | You should have received a copy of the GNU General Public License |
15 | along with this program; if not, write to the Free Software |
16 | Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */ |
17 | |
18 | /**************************************************************** |
19 | |
20 | This file incorporates work covered by the following copyright and |
21 | permission notice: |
22 | |
23 | The author of this software is David M. Gay. |
24 | |
25 | Copyright (c) 1991, 2000, 2001 by Lucent Technologies. |
26 | |
27 | Permission to use, copy, modify, and distribute this software for any |
28 | purpose without fee is hereby granted, provided that this entire notice |
29 | is included in all copies of any software which is or includes a copy |
30 | or modification of this software and in all copies of the supporting |
31 | documentation for such software. |
32 | |
33 | THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR IMPLIED |
34 | WARRANTY. IN PARTICULAR, NEITHER THE AUTHOR NOR LUCENT MAKES ANY |
35 | REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE MERCHANTABILITY |
36 | OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR PURPOSE. |
37 | |
38 | ***************************************************************/ |
39 | |
40 | //#include "strings_def.h" |
41 | //#include <my_base.h> /* for EOVERFLOW on Windows */ |
42 | #include <ma_global.h> |
43 | #include <memory.h> |
44 | #include "ma_string.h" |
45 | |
46 | /** |
47 | Appears to suffice to not call malloc() in most cases. |
48 | @todo |
49 | see if it is possible to get rid of malloc(). |
50 | this constant is sufficient to avoid malloc() on all inputs I have tried. |
51 | */ |
52 | #define DTOA_BUFF_SIZE (460 * sizeof(void *)) |
53 | |
54 | /* Magic value returned by dtoa() to indicate overflow */ |
55 | #define DTOA_OVERFLOW 9999 |
56 | |
57 | static char *dtoa(double, int, int, int *, int *, char **, char *, size_t); |
58 | static void dtoa_free(char *, char *, size_t); |
59 | |
60 | /** |
61 | @brief |
62 | Converts a given floating point number to a zero-terminated string |
63 | representation using the 'f' format. |
64 | |
65 | @details |
66 | This function is a wrapper around dtoa() to do the same as |
67 | sprintf(to, "%-.*f", precision, x), though the conversion is usually more |
68 | precise. The only difference is in handling [-,+]infinity and nan values, |
69 | in which case we print '0\0' to the output string and indicate an overflow. |
70 | |
71 | @param x the input floating point number. |
72 | @param precision the number of digits after the decimal point. |
73 | All properties of sprintf() apply: |
74 | - if the number of significant digits after the decimal |
75 | point is less than precision, the resulting string is |
76 | right-padded with zeros |
77 | - if the precision is 0, no decimal point appears |
78 | - if a decimal point appears, at least one digit appears |
79 | before it |
80 | @param to pointer to the output buffer. The longest string which |
81 | my_fcvt() can return is FLOATING_POINT_BUFFER bytes |
82 | (including the terminating '\0'). |
83 | @param error if not NULL, points to a location where the status of |
84 | conversion is stored upon return. |
85 | FALSE successful conversion |
86 | TRUE the input number is [-,+]infinity or nan. |
87 | The output string in this case is always '0'. |
88 | @return number of written characters (excluding terminating '\0') |
89 | */ |
90 | |
91 | size_t ma_fcvt(double x, int precision, char *to, my_bool *error) |
92 | { |
93 | int decpt, sign, len, i; |
94 | char *res, *src, *end, *dst= to; |
95 | char buf[DTOA_BUFF_SIZE]; |
96 | DBUG_ASSERT(precision >= 0 && precision < NOT_FIXED_DEC && to != NULL); |
97 | |
98 | res= dtoa(x, 5, precision, &decpt, &sign, &end, buf, sizeof(buf)); |
99 | |
100 | if (decpt == DTOA_OVERFLOW) |
101 | { |
102 | dtoa_free(res, buf, sizeof(buf)); |
103 | *to++= '0'; |
104 | *to= '\0'; |
105 | if (error != NULL) |
106 | *error= TRUE; |
107 | return 1; |
108 | } |
109 | |
110 | src= res; |
111 | len= (int)(end - src); |
112 | |
113 | if (sign) |
114 | *dst++= '-'; |
115 | |
116 | if (decpt <= 0) |
117 | { |
118 | *dst++= '0'; |
119 | *dst++= '.'; |
120 | for (i= decpt; i < 0; i++) |
121 | *dst++= '0'; |
122 | } |
123 | |
124 | for (i= 1; i <= len; i++) |
125 | { |
126 | *dst++= *src++; |
127 | if (i == decpt && i < len) |
128 | *dst++= '.'; |
129 | } |
130 | while (i++ <= decpt) |
131 | *dst++= '0'; |
132 | |
133 | if (precision > 0) |
134 | { |
135 | if (len <= decpt) |
136 | *dst++= '.'; |
137 | |
138 | for (i= precision - MAX(0, (len - decpt)); i > 0; i--) |
139 | *dst++= '0'; |
140 | } |
141 | |
142 | *dst= '\0'; |
143 | if (error != NULL) |
144 | *error= FALSE; |
145 | |
146 | dtoa_free(res, buf, sizeof(buf)); |
147 | |
148 | return dst - to; |
149 | } |
150 | |
151 | /** |
152 | @brief |
153 | Converts a given floating point number to a zero-terminated string |
154 | representation with a given field width using the 'e' format |
155 | (aka scientific notation) or the 'f' one. |
156 | |
157 | @details |
158 | The format is chosen automatically to provide the most number of significant |
159 | digits (and thus, precision) with a given field width. In many cases, the |
160 | result is similar to that of sprintf(to, "%g", x) with a few notable |
161 | differences: |
162 | - the conversion is usually more precise than C library functions. |
163 | - there is no 'precision' argument. instead, we specify the number of |
164 | characters available for conversion (i.e. a field width). |
165 | - the result never exceeds the specified field width. If the field is too |
166 | short to contain even a rounded decimal representation, ma_gcvt() |
167 | indicates overflow and truncates the output string to the specified width. |
168 | - float-type arguments are handled differently than double ones. For a |
169 | float input number (i.e. when the 'type' argument is MY_GCVT_ARG_FLOAT) |
170 | we deliberately limit the precision of conversion by FLT_DIG digits to |
171 | avoid garbage past the significant digits. |
172 | - unlike sprintf(), in cases where the 'e' format is preferred, we don't |
173 | zero-pad the exponent to save space for significant digits. The '+' sign |
174 | for a positive exponent does not appear for the same reason. |
175 | |
176 | @param x the input floating point number. |
177 | @param type is either MY_GCVT_ARG_FLOAT or MY_GCVT_ARG_DOUBLE. |
178 | Specifies the type of the input number (see notes above). |
179 | @param width field width in characters. The minimal field width to |
180 | hold any number representation (albeit rounded) is 7 |
181 | characters ("-Ne-NNN"). |
182 | @param to pointer to the output buffer. The result is always |
183 | zero-terminated, and the longest returned string is thus |
184 | 'width + 1' bytes. |
185 | @param error if not NULL, points to a location where the status of |
186 | conversion is stored upon return. |
187 | FALSE successful conversion |
188 | TRUE the input number is [-,+]infinity or nan. |
189 | The output string in this case is always '0'. |
190 | @return number of written characters (excluding terminating '\0') |
191 | |
192 | @todo |
193 | Check if it is possible and makes sense to do our own rounding on top of |
194 | dtoa() instead of calling dtoa() twice in (rare) cases when the resulting |
195 | string representation does not fit in the specified field width and we want |
196 | to re-round the input number with fewer significant digits. Examples: |
197 | |
198 | ma_gcvt(-9e-3, ..., 4, ...); |
199 | ma_gcvt(-9e-3, ..., 2, ...); |
200 | ma_gcvt(1.87e-3, ..., 4, ...); |
201 | ma_gcvt(55, ..., 1, ...); |
202 | |
203 | We do our best to minimize such cases by: |
204 | |
205 | - passing to dtoa() the field width as the number of significant digits |
206 | |
207 | - removing the sign of the number early (and decreasing the width before |
208 | passing it to dtoa()) |
209 | |
210 | - choosing the proper format to preserve the most number of significant |
211 | digits. |
212 | */ |
213 | |
214 | size_t ma_gcvt(double x, my_gcvt_arg_type type, int width, char *to, |
215 | my_bool *error) |
216 | { |
217 | int decpt, sign, len, exp_len; |
218 | char *res, *src, *end, *dst= to, *dend= dst + width; |
219 | char buf[DTOA_BUFF_SIZE]; |
220 | my_bool have_space, force_e_format; |
221 | DBUG_ASSERT(width > 0 && to != NULL); |
222 | |
223 | /* We want to remove '-' from equations early */ |
224 | if (x < 0.) |
225 | width--; |
226 | |
227 | res= dtoa(x, 4, type == MY_GCVT_ARG_DOUBLE ? width : MIN(width, FLT_DIG), |
228 | &decpt, &sign, &end, buf, sizeof(buf)); |
229 | if (decpt == DTOA_OVERFLOW) |
230 | { |
231 | dtoa_free(res, buf, sizeof(buf)); |
232 | *to++= '0'; |
233 | *to= '\0'; |
234 | if (error != NULL) |
235 | *error= TRUE; |
236 | return 1; |
237 | } |
238 | |
239 | if (error != NULL) |
240 | *error= FALSE; |
241 | |
242 | src= res; |
243 | len= (int)(end - res); |
244 | |
245 | /* |
246 | Number of digits in the exponent from the 'e' conversion. |
247 | The sign of the exponent is taken into account separetely, we don't need |
248 | to count it here. |
249 | */ |
250 | exp_len= 1 + (decpt >= 101 || decpt <= -99) + (decpt >= 11 || decpt <= -9); |
251 | |
252 | /* |
253 | Do we have enough space for all digits in the 'f' format? |
254 | Let 'len' be the number of significant digits returned by dtoa, |
255 | and F be the length of the resulting decimal representation. |
256 | Consider the following cases: |
257 | 1. decpt <= 0, i.e. we have "0.NNN" => F = len - decpt + 2 |
258 | 2. 0 < decpt < len, i.e. we have "NNN.NNN" => F = len + 1 |
259 | 3. len <= decpt, i.e. we have "NNN00" => F = decpt |
260 | */ |
261 | have_space= (decpt <= 0 ? len - decpt + 2 : |
262 | decpt > 0 && decpt < len ? len + 1 : |
263 | decpt) <= width; |
264 | /* |
265 | The following is true when no significant digits can be placed with the |
266 | specified field width using the 'f' format, and the 'e' format |
267 | will not be truncated. |
268 | */ |
269 | force_e_format= (decpt <= 0 && width <= 2 - decpt && width >= 3 + exp_len); |
270 | /* |
271 | Assume that we don't have enough space to place all significant digits in |
272 | the 'f' format. We have to choose between the 'e' format and the 'f' one |
273 | to keep as many significant digits as possible. |
274 | Let E and F be the lengths of decimal representation in the 'e' and 'f' |
275 | formats, respectively. We want to use the 'f' format if, and only if F <= E. |
276 | Consider the following cases: |
277 | 1. decpt <= 0. |
278 | F = len - decpt + 2 (see above) |
279 | E = len + (len > 1) + 1 + 1 (decpt <= -99) + (decpt <= -9) + 1 |
280 | ("N.NNe-MMM") |
281 | (F <= E) <=> (len == 1 && decpt >= -1) || (len > 1 && decpt >= -2) |
282 | We also need to ensure that if the 'f' format is chosen, |
283 | the field width allows us to place at least one significant digit |
284 | (i.e. width > 2 - decpt). If not, we prefer the 'e' format. |
285 | 2. 0 < decpt < len |
286 | F = len + 1 (see above) |
287 | E = len + 1 + 1 + ... ("N.NNeMMM") |
288 | F is always less than E. |
289 | 3. len <= decpt <= width |
290 | In this case we have enough space to represent the number in the 'f' |
291 | format, so we prefer it with some exceptions. |
292 | 4. width < decpt |
293 | The number cannot be represented in the 'f' format at all, always use |
294 | the 'e' 'one. |
295 | */ |
296 | if ((have_space || |
297 | /* |
298 | Not enough space, let's see if the 'f' format provides the most number |
299 | of significant digits. |
300 | */ |
301 | ((decpt <= width && (decpt >= -1 || (decpt == -2 && |
302 | (len > 1 || !force_e_format)))) && |
303 | !force_e_format)) && |
304 | |
305 | /* |
306 | Use the 'e' format in some cases even if we have enough space for the |
307 | 'f' one. See comment for DBL_DIG. |
308 | */ |
309 | (!have_space || (decpt >= -DBL_DIG + 1 && |
310 | (decpt <= DBL_DIG || len > decpt)))) |
311 | { |
312 | /* 'f' format */ |
313 | int i; |
314 | |
315 | width-= (decpt < len) + (decpt <= 0 ? 1 - decpt : 0); |
316 | |
317 | /* Do we have to truncate any digits? */ |
318 | if (width < len) |
319 | { |
320 | if (width < decpt) |
321 | { |
322 | if (error != NULL) |
323 | *error= TRUE; |
324 | width= decpt; |
325 | } |
326 | |
327 | /* |
328 | We want to truncate (len - width) least significant digits after the |
329 | decimal point. For this we are calling dtoa with mode=5, passing the |
330 | number of significant digits = (len-decpt) - (len-width) = width-decpt |
331 | */ |
332 | dtoa_free(res, buf, sizeof(buf)); |
333 | res= dtoa(x, 5, width - decpt, &decpt, &sign, &end, buf, sizeof(buf)); |
334 | src= res; |
335 | len= (int)(end - res); |
336 | } |
337 | |
338 | if (len == 0) |
339 | { |
340 | /* Underflow. Just print '0' and exit */ |
341 | *dst++= '0'; |
342 | goto end; |
343 | } |
344 | |
345 | /* |
346 | At this point we are sure we have enough space to put all digits |
347 | returned by dtoa |
348 | */ |
349 | if (sign && dst < dend) |
350 | *dst++= '-'; |
351 | if (decpt <= 0) |
352 | { |
353 | if (dst < dend) |
354 | *dst++= '0'; |
355 | if (len > 0 && dst < dend) |
356 | *dst++= '.'; |
357 | for (; decpt < 0 && dst < dend; decpt++) |
358 | *dst++= '0'; |
359 | } |
360 | |
361 | for (i= 1; i <= len && dst < dend; i++) |
362 | { |
363 | *dst++= *src++; |
364 | if (i == decpt && i < len && dst < dend) |
365 | *dst++= '.'; |
366 | } |
367 | while (i++ <= decpt && dst < dend) |
368 | *dst++= '0'; |
369 | } |
370 | else |
371 | { |
372 | /* 'e' format */ |
373 | int decpt_sign= 0; |
374 | |
375 | if (--decpt < 0) |
376 | { |
377 | decpt= -decpt; |
378 | width--; |
379 | decpt_sign= 1; |
380 | } |
381 | width-= 1 + exp_len; /* eNNN */ |
382 | |
383 | if (len > 1) |
384 | width--; |
385 | |
386 | if (width <= 0) |
387 | { |
388 | /* Overflow */ |
389 | if (error != NULL) |
390 | *error= TRUE; |
391 | width= 0; |
392 | } |
393 | |
394 | /* Do we have to truncate any digits? */ |
395 | if (width < len) |
396 | { |
397 | /* Yes, re-convert with a smaller width */ |
398 | dtoa_free(res, buf, sizeof(buf)); |
399 | res= dtoa(x, 4, width, &decpt, &sign, &end, buf, sizeof(buf)); |
400 | src= res; |
401 | len= (int)(end - res); |
402 | if (--decpt < 0) |
403 | decpt= -decpt; |
404 | } |
405 | /* |
406 | At this point we are sure we have enough space to put all digits |
407 | returned by dtoa |
408 | */ |
409 | if (sign && dst < dend) |
410 | *dst++= '-'; |
411 | if (dst < dend) |
412 | *dst++= *src++; |
413 | if (len > 1 && dst < dend) |
414 | { |
415 | *dst++= '.'; |
416 | while (src < end && dst < dend) |
417 | *dst++= *src++; |
418 | } |
419 | if (dst < dend) |
420 | *dst++= 'e'; |
421 | if (decpt_sign && dst < dend) |
422 | *dst++= '-'; |
423 | |
424 | if (decpt >= 100 && dst < dend) |
425 | { |
426 | *dst++= decpt / 100 + '0'; |
427 | decpt%= 100; |
428 | if (dst < dend) |
429 | *dst++= decpt / 10 + '0'; |
430 | } |
431 | else if (decpt >= 10 && dst < dend) |
432 | *dst++= decpt / 10 + '0'; |
433 | if (dst < dend) |
434 | *dst++= decpt % 10 + '0'; |
435 | |
436 | } |
437 | |
438 | end: |
439 | dtoa_free(res, buf, sizeof(buf)); |
440 | *dst= '\0'; |
441 | |
442 | return dst - to; |
443 | } |
444 | |
445 | /**************************************************************** |
446 | * |
447 | * The author of this software is David M. Gay. |
448 | * |
449 | * Copyright (c) 1991, 2000, 2001 by Lucent Technologies. |
450 | * |
451 | * Permission to use, copy, modify, and distribute this software for any |
452 | * purpose without fee is hereby granted, provided that this entire notice |
453 | * is included in all copies of any software which is or includes a copy |
454 | * or modification of this software and in all copies of the supporting |
455 | * documentation for such software. |
456 | * |
457 | * THIS SOFTWARE IS BEING PROVIDED "AS IS", WITHOUT ANY EXPRESS OR IMPLIED |
458 | * WARRANTY. IN PARTICULAR, NEITHER THE AUTHOR NOR LUCENT MAKES ANY |
459 | * REPRESENTATION OR WARRANTY OF ANY KIND CONCERNING THE MERCHANTABILITY |
460 | * OF THIS SOFTWARE OR ITS FITNESS FOR ANY PARTICULAR PURPOSE. |
461 | * |
462 | ***************************************************************/ |
463 | /* Please send bug reports to David M. Gay (dmg at acm dot org, |
464 | * with " at " changed at "@" and " dot " changed to "."). */ |
465 | |
466 | /* |
467 | Original copy of the software is located at http://www.netlib.org/fp/dtoa.c |
468 | It was adjusted to serve MySQL server needs: |
469 | * strtod() was modified to not expect a zero-terminated string. |
470 | It now honors 'se' (end of string) argument as the input parameter, |
471 | not just as the output one. |
472 | * in dtoa(), in case of overflow/underflow/NaN result string now contains "0"; |
473 | decpt is set to DTOA_OVERFLOW to indicate overflow. |
474 | * support for VAX, IBM mainframe and 16-bit hardware removed |
475 | * we always assume that 64-bit integer type is available |
476 | * support for Kernigan-Ritchie style headers (pre-ANSI compilers) |
477 | removed |
478 | * all gcc warnings ironed out |
479 | * we always assume multithreaded environment, so we had to change |
480 | memory allocation procedures to use stack in most cases; |
481 | malloc is used as the last resort. |
482 | * pow5mult rewritten to use pre-calculated pow5 list instead of |
483 | the one generated on the fly. |
484 | */ |
485 | |
486 | |
487 | /* |
488 | On a machine with IEEE extended-precision registers, it is |
489 | necessary to specify double-precision (53-bit) rounding precision |
490 | before invoking strtod or dtoa. If the machine uses (the equivalent |
491 | of) Intel 80x87 arithmetic, the call |
492 | _control87(PC_53, MCW_PC); |
493 | does this with many compilers. Whether this or another call is |
494 | appropriate depends on the compiler; for this to work, it may be |
495 | necessary to #include "float.h" or another system-dependent header |
496 | file. |
497 | */ |
498 | |
499 | /* |
500 | #define Honor_FLT_ROUNDS if FLT_ROUNDS can assume the values 2 or 3 |
501 | and dtoa should round accordingly. |
502 | #define Check_FLT_ROUNDS if FLT_ROUNDS can assume the values 2 or 3 |
503 | and Honor_FLT_ROUNDS is not #defined. |
504 | |
505 | TODO: check if we can get rid of the above two |
506 | */ |
507 | |
508 | typedef int32 Long; |
509 | typedef uint32 ULong; |
510 | typedef int64 LLong; |
511 | typedef uint64 ULLong; |
512 | |
513 | typedef union { double d; ULong L[2]; } U; |
514 | |
515 | #if defined(HAVE_BIGENDIAN) || defined(WORDS_BIGENDIAN) || \ |
516 | (defined(__FLOAT_WORD_ORDER) && (__FLOAT_WORD_ORDER == __BIG_ENDIAN)) |
517 | #define word0(x) (x)->L[0] |
518 | #define word1(x) (x)->L[1] |
519 | #else |
520 | #define word0(x) (x)->L[1] |
521 | #define word1(x) (x)->L[0] |
522 | #endif |
523 | |
524 | #define dval(x) (x)->d |
525 | |
526 | /* #define P DBL_MANT_DIG */ |
527 | /* Ten_pmax= floor(P*log(2)/log(5)) */ |
528 | /* Bletch= (highest power of 2 < DBL_MAX_10_EXP) / 16 */ |
529 | /* Quick_max= floor((P-1)*log(FLT_RADIX)/log(10) - 1) */ |
530 | /* Int_max= floor(P*log(FLT_RADIX)/log(10) - 1) */ |
531 | |
532 | #define Exp_shift 20 |
533 | #define Exp_shift1 20 |
534 | #define Exp_msk1 0x100000 |
535 | #define Exp_mask 0x7ff00000 |
536 | #define P 53 |
537 | #define Bias 1023 |
538 | #define Emin (-1022) |
539 | #define Exp_1 0x3ff00000 |
540 | #define Exp_11 0x3ff00000 |
541 | #define Ebits 11 |
542 | #define Frac_mask 0xfffff |
543 | #define Frac_mask1 0xfffff |
544 | #define Ten_pmax 22 |
545 | #define Bletch 0x10 |
546 | #define Bndry_mask 0xfffff |
547 | #define Bndry_mask1 0xfffff |
548 | #define LSB 1 |
549 | #define Sign_bit 0x80000000 |
550 | #define Log2P 1 |
551 | #define Tiny1 1 |
552 | #define Quick_max 14 |
553 | #define Int_max 14 |
554 | |
555 | #ifndef Flt_Rounds |
556 | #ifdef FLT_ROUNDS |
557 | #define Flt_Rounds FLT_ROUNDS |
558 | #else |
559 | #define Flt_Rounds 1 |
560 | #endif |
561 | #endif /*Flt_Rounds*/ |
562 | |
563 | #ifdef Honor_FLT_ROUNDS |
564 | #define Rounding rounding |
565 | #undef Check_FLT_ROUNDS |
566 | #define Check_FLT_ROUNDS |
567 | #else |
568 | #define Rounding Flt_Rounds |
569 | #endif |
570 | |
571 | #define rounded_product(a,b) a*= b |
572 | #define rounded_quotient(a,b) a/= b |
573 | |
574 | #define Big0 (Frac_mask1 | Exp_msk1*(DBL_MAX_EXP+Bias-1)) |
575 | #define Big1 0xffffffff |
576 | #define FFFFFFFF 0xffffffffUL |
577 | |
578 | /* This is tested to be enough for dtoa */ |
579 | |
580 | #define Kmax 15 |
581 | |
582 | #define Bcopy(x,y) memcpy((char *)&x->sign, (char *)&y->sign, \ |
583 | 2*sizeof(int) + y->wds*sizeof(ULong)) |
584 | |
585 | /* Arbitrary-length integer */ |
586 | |
587 | typedef struct Bigint |
588 | { |
589 | union { |
590 | ULong *x; /* points right after this Bigint object */ |
591 | struct Bigint *next; /* to maintain free lists */ |
592 | } p; |
593 | int k; /* 2^k = maxwds */ |
594 | int maxwds; /* maximum length in 32-bit words */ |
595 | int sign; /* not zero if number is negative */ |
596 | int wds; /* current length in 32-bit words */ |
597 | } Bigint; |
598 | |
599 | |
600 | /* A simple stack-memory based allocator for Bigints */ |
601 | |
602 | typedef struct Stack_alloc |
603 | { |
604 | char *begin; |
605 | char *free; |
606 | char *end; |
607 | /* |
608 | Having list of free blocks lets us reduce maximum required amount |
609 | of memory from ~4000 bytes to < 1680 (tested on x86). |
610 | */ |
611 | Bigint *freelist[Kmax+1]; |
612 | } Stack_alloc; |
613 | |
614 | |
615 | /* |
616 | Try to allocate object on stack, and resort to malloc if all |
617 | stack memory is used. Ensure allocated objects to be aligned by the pointer |
618 | size in order to not break the alignment rules when storing a pointer to a |
619 | Bigint. |
620 | */ |
621 | |
622 | static Bigint *Balloc(int k, Stack_alloc *alloc) |
623 | { |
624 | Bigint *rv; |
625 | DBUG_ASSERT(k <= Kmax); |
626 | if (k <= Kmax && alloc->freelist[k]) |
627 | { |
628 | rv= alloc->freelist[k]; |
629 | alloc->freelist[k]= rv->p.next; |
630 | } |
631 | else |
632 | { |
633 | int x, len; |
634 | |
635 | x= 1 << k; |
636 | len= MY_ALIGN(sizeof(Bigint) + x * sizeof(ULong), SIZEOF_CHARP); |
637 | |
638 | if (alloc->free + len <= alloc->end) |
639 | { |
640 | rv= (Bigint*) alloc->free; |
641 | alloc->free+= len; |
642 | } |
643 | else |
644 | rv= (Bigint*) malloc(len); |
645 | |
646 | rv->k= k; |
647 | rv->maxwds= x; |
648 | } |
649 | rv->sign= rv->wds= 0; |
650 | rv->p.x= (ULong*) (rv + 1); |
651 | return rv; |
652 | } |
653 | |
654 | |
655 | /* |
656 | If object was allocated on stack, try putting it to the free |
657 | list. Otherwise call free(). |
658 | */ |
659 | |
660 | static void Bfree(Bigint *v, Stack_alloc *alloc) |
661 | { |
662 | char *gptr= (char*) v; /* generic pointer */ |
663 | if (gptr < alloc->begin || gptr >= alloc->end) |
664 | free(gptr); |
665 | else if (v->k <= Kmax) |
666 | { |
667 | /* |
668 | Maintain free lists only for stack objects: this way we don't |
669 | have to bother with freeing lists in the end of dtoa; |
670 | heap should not be used normally anyway. |
671 | */ |
672 | v->p.next= alloc->freelist[v->k]; |
673 | alloc->freelist[v->k]= v; |
674 | } |
675 | } |
676 | |
677 | |
678 | /* |
679 | This is to place return value of dtoa in: tries to use stack |
680 | as well, but passes by free lists management and just aligns len by |
681 | the pointer size in order to not break the alignment rules when storing a |
682 | pointer to a Bigint. |
683 | */ |
684 | |
685 | static char *dtoa_alloc(int i, Stack_alloc *alloc) |
686 | { |
687 | char *rv; |
688 | int aligned_size= MY_ALIGN(i, SIZEOF_CHARP); |
689 | if (alloc->free + aligned_size <= alloc->end) |
690 | { |
691 | rv= alloc->free; |
692 | alloc->free+= aligned_size; |
693 | } |
694 | else |
695 | rv= malloc(i); |
696 | return rv; |
697 | } |
698 | |
699 | |
700 | /* |
701 | dtoa_free() must be used to free values s returned by dtoa() |
702 | This is the counterpart of dtoa_alloc() |
703 | */ |
704 | |
705 | static void dtoa_free(char *gptr, char *buf, size_t buf_size) |
706 | { |
707 | if (gptr < buf || gptr >= buf + buf_size) |
708 | free(gptr); |
709 | } |
710 | |
711 | |
712 | /* Bigint arithmetic functions */ |
713 | |
714 | /* Multiply by m and add a */ |
715 | |
716 | static Bigint *multadd(Bigint *b, int m, int a, Stack_alloc *alloc) |
717 | { |
718 | int i, wds; |
719 | ULong *x; |
720 | ULLong carry, y; |
721 | Bigint *b1; |
722 | |
723 | wds= b->wds; |
724 | x= b->p.x; |
725 | i= 0; |
726 | carry= a; |
727 | do |
728 | { |
729 | y= *x * (ULLong)m + carry; |
730 | carry= y >> 32; |
731 | *x++= (ULong)(y & FFFFFFFF); |
732 | } |
733 | while (++i < wds); |
734 | if (carry) |
735 | { |
736 | if (wds >= b->maxwds) |
737 | { |
738 | b1= Balloc(b->k+1, alloc); |
739 | Bcopy(b1, b); |
740 | Bfree(b, alloc); |
741 | b= b1; |
742 | } |
743 | b->p.x[wds++]= (ULong) carry; |
744 | b->wds= wds; |
745 | } |
746 | return b; |
747 | } |
748 | |
749 | |
750 | static int hi0bits(register ULong x) |
751 | { |
752 | register int k= 0; |
753 | |
754 | if (!(x & 0xffff0000)) |
755 | { |
756 | k= 16; |
757 | x<<= 16; |
758 | } |
759 | if (!(x & 0xff000000)) |
760 | { |
761 | k+= 8; |
762 | x<<= 8; |
763 | } |
764 | if (!(x & 0xf0000000)) |
765 | { |
766 | k+= 4; |
767 | x<<= 4; |
768 | } |
769 | if (!(x & 0xc0000000)) |
770 | { |
771 | k+= 2; |
772 | x<<= 2; |
773 | } |
774 | if (!(x & 0x80000000)) |
775 | { |
776 | k++; |
777 | if (!(x & 0x40000000)) |
778 | return 32; |
779 | } |
780 | return k; |
781 | } |
782 | |
783 | |
784 | static int lo0bits(ULong *y) |
785 | { |
786 | register int k; |
787 | register ULong x= *y; |
788 | |
789 | if (x & 7) |
790 | { |
791 | if (x & 1) |
792 | return 0; |
793 | if (x & 2) |
794 | { |
795 | *y= x >> 1; |
796 | return 1; |
797 | } |
798 | *y= x >> 2; |
799 | return 2; |
800 | } |
801 | k= 0; |
802 | if (!(x & 0xffff)) |
803 | { |
804 | k= 16; |
805 | x>>= 16; |
806 | } |
807 | if (!(x & 0xff)) |
808 | { |
809 | k+= 8; |
810 | x>>= 8; |
811 | } |
812 | if (!(x & 0xf)) |
813 | { |
814 | k+= 4; |
815 | x>>= 4; |
816 | } |
817 | if (!(x & 0x3)) |
818 | { |
819 | k+= 2; |
820 | x>>= 2; |
821 | } |
822 | if (!(x & 1)) |
823 | { |
824 | k++; |
825 | x>>= 1; |
826 | if (!x) |
827 | return 32; |
828 | } |
829 | *y= x; |
830 | return k; |
831 | } |
832 | |
833 | |
834 | /* Convert integer to Bigint number */ |
835 | |
836 | static Bigint *i2b(int i, Stack_alloc *alloc) |
837 | { |
838 | Bigint *b; |
839 | |
840 | b= Balloc(1, alloc); |
841 | b->p.x[0]= i; |
842 | b->wds= 1; |
843 | return b; |
844 | } |
845 | |
846 | |
847 | /* Multiply two Bigint numbers */ |
848 | |
849 | static Bigint *mult(Bigint *a, Bigint *b, Stack_alloc *alloc) |
850 | { |
851 | Bigint *c; |
852 | int k, wa, wb, wc; |
853 | ULong *x, *xa, *xae, *xb, *xbe, *xc, *xc0; |
854 | ULong y; |
855 | ULLong carry, z; |
856 | |
857 | if (a->wds < b->wds) |
858 | { |
859 | c= a; |
860 | a= b; |
861 | b= c; |
862 | } |
863 | k= a->k; |
864 | wa= a->wds; |
865 | wb= b->wds; |
866 | wc= wa + wb; |
867 | if (wc > a->maxwds) |
868 | k++; |
869 | c= Balloc(k, alloc); |
870 | for (x= c->p.x, xa= x + wc; x < xa; x++) |
871 | *x= 0; |
872 | xa= a->p.x; |
873 | xae= xa + wa; |
874 | xb= b->p.x; |
875 | xbe= xb + wb; |
876 | xc0= c->p.x; |
877 | for (; xb < xbe; xc0++) |
878 | { |
879 | if ((y= *xb++)) |
880 | { |
881 | x= xa; |
882 | xc= xc0; |
883 | carry= 0; |
884 | do |
885 | { |
886 | z= *x++ * (ULLong)y + *xc + carry; |
887 | carry= z >> 32; |
888 | *xc++= (ULong) (z & FFFFFFFF); |
889 | } |
890 | while (x < xae); |
891 | *xc= (ULong) carry; |
892 | } |
893 | } |
894 | for (xc0= c->p.x, xc= xc0 + wc; wc > 0 && !*--xc; --wc) ; |
895 | c->wds= wc; |
896 | return c; |
897 | } |
898 | |
899 | |
900 | /* |
901 | Precalculated array of powers of 5: tested to be enough for |
902 | vasting majority of dtoa_r cases. |
903 | */ |
904 | |
905 | static ULong powers5[]= |
906 | { |
907 | 625UL, |
908 | |
909 | 390625UL, |
910 | |
911 | 2264035265UL, 35UL, |
912 | |
913 | 2242703233UL, 762134875UL, 1262UL, |
914 | |
915 | 3211403009UL, 1849224548UL, 3668416493UL, 3913284084UL, 1593091UL, |
916 | |
917 | 781532673UL, 64985353UL, 253049085UL, 594863151UL, 3553621484UL, |
918 | 3288652808UL, 3167596762UL, 2788392729UL, 3911132675UL, 590UL, |
919 | |
920 | 2553183233UL, 3201533787UL, 3638140786UL, 303378311UL, 1809731782UL, |
921 | 3477761648UL, 3583367183UL, 649228654UL, 2915460784UL, 487929380UL, |
922 | 1011012442UL, 1677677582UL, 3428152256UL, 1710878487UL, 1438394610UL, |
923 | 2161952759UL, 4100910556UL, 1608314830UL, 349175UL |
924 | }; |
925 | |
926 | |
927 | static Bigint p5_a[]= |
928 | { |
929 | /* { x } - k - maxwds - sign - wds */ |
930 | { { powers5 }, 1, 1, 0, 1 }, |
931 | { { powers5 + 1 }, 1, 1, 0, 1 }, |
932 | { { powers5 + 2 }, 1, 2, 0, 2 }, |
933 | { { powers5 + 4 }, 2, 3, 0, 3 }, |
934 | { { powers5 + 7 }, 3, 5, 0, 5 }, |
935 | { { powers5 + 12 }, 4, 10, 0, 10 }, |
936 | { { powers5 + 22 }, 5, 19, 0, 19 } |
937 | }; |
938 | |
939 | #define P5A_MAX (sizeof(p5_a)/sizeof(*p5_a) - 1) |
940 | |
941 | static Bigint *pow5mult(Bigint *b, int k, Stack_alloc *alloc) |
942 | { |
943 | Bigint *b1, *p5, *p51=NULL; |
944 | int i; |
945 | static int p05[3]= { 5, 25, 125 }; |
946 | my_bool overflow= FALSE; |
947 | |
948 | if ((i= k & 3)) |
949 | b= multadd(b, p05[i-1], 0, alloc); |
950 | |
951 | if (!(k>>= 2)) |
952 | return b; |
953 | p5= p5_a; |
954 | for (;;) |
955 | { |
956 | if (k & 1) |
957 | { |
958 | b1= mult(b, p5, alloc); |
959 | Bfree(b, alloc); |
960 | b= b1; |
961 | } |
962 | if (!(k>>= 1)) |
963 | break; |
964 | /* Calculate next power of 5 */ |
965 | if (overflow) |
966 | { |
967 | p51= mult(p5, p5, alloc); |
968 | Bfree(p5, alloc); |
969 | p5= p51; |
970 | } |
971 | else if (p5 < p5_a + P5A_MAX) |
972 | ++p5; |
973 | else if (p5 == p5_a + P5A_MAX) |
974 | { |
975 | p5= mult(p5, p5, alloc); |
976 | overflow= TRUE; |
977 | } |
978 | } |
979 | if (p51) |
980 | Bfree(p51, alloc); |
981 | return b; |
982 | } |
983 | |
984 | |
985 | static Bigint *lshift(Bigint *b, int k, Stack_alloc *alloc) |
986 | { |
987 | int i, k1, n, n1; |
988 | Bigint *b1; |
989 | ULong *x, *x1, *xe, z; |
990 | |
991 | n= k >> 5; |
992 | k1= b->k; |
993 | n1= n + b->wds + 1; |
994 | for (i= b->maxwds; n1 > i; i<<= 1) |
995 | k1++; |
996 | b1= Balloc(k1, alloc); |
997 | x1= b1->p.x; |
998 | for (i= 0; i < n; i++) |
999 | *x1++= 0; |
1000 | x= b->p.x; |
1001 | xe= x + b->wds; |
1002 | if (k&= 0x1f) |
1003 | { |
1004 | k1= 32 - k; |
1005 | z= 0; |
1006 | do |
1007 | { |
1008 | *x1++= *x << k | z; |
1009 | z= *x++ >> k1; |
1010 | } |
1011 | while (x < xe); |
1012 | if ((*x1= z)) |
1013 | ++n1; |
1014 | } |
1015 | else |
1016 | do |
1017 | *x1++= *x++; |
1018 | while (x < xe); |
1019 | b1->wds= n1 - 1; |
1020 | Bfree(b, alloc); |
1021 | return b1; |
1022 | } |
1023 | |
1024 | |
1025 | static int cmp(Bigint *a, Bigint *b) |
1026 | { |
1027 | ULong *xa, *xa0, *xb, *xb0; |
1028 | int i, j; |
1029 | |
1030 | i= a->wds; |
1031 | j= b->wds; |
1032 | if (i-= j) |
1033 | return i; |
1034 | xa0= a->p.x; |
1035 | xa= xa0 + j; |
1036 | xb0= b->p.x; |
1037 | xb= xb0 + j; |
1038 | for (;;) |
1039 | { |
1040 | if (*--xa != *--xb) |
1041 | return *xa < *xb ? -1 : 1; |
1042 | if (xa <= xa0) |
1043 | break; |
1044 | } |
1045 | return 0; |
1046 | } |
1047 | |
1048 | |
1049 | static Bigint *diff(Bigint *a, Bigint *b, Stack_alloc *alloc) |
1050 | { |
1051 | Bigint *c; |
1052 | int i, wa, wb; |
1053 | ULong *xa, *xae, *xb, *xbe, *xc; |
1054 | ULLong borrow, y; |
1055 | |
1056 | i= cmp(a,b); |
1057 | if (!i) |
1058 | { |
1059 | c= Balloc(0, alloc); |
1060 | c->wds= 1; |
1061 | c->p.x[0]= 0; |
1062 | return c; |
1063 | } |
1064 | if (i < 0) |
1065 | { |
1066 | c= a; |
1067 | a= b; |
1068 | b= c; |
1069 | i= 1; |
1070 | } |
1071 | else |
1072 | i= 0; |
1073 | c= Balloc(a->k, alloc); |
1074 | c->sign= i; |
1075 | wa= a->wds; |
1076 | xa= a->p.x; |
1077 | xae= xa + wa; |
1078 | wb= b->wds; |
1079 | xb= b->p.x; |
1080 | xbe= xb + wb; |
1081 | xc= c->p.x; |
1082 | borrow= 0; |
1083 | do |
1084 | { |
1085 | y= (ULLong)*xa++ - *xb++ - borrow; |
1086 | borrow= y >> 32 & (ULong)1; |
1087 | *xc++= (ULong) (y & FFFFFFFF); |
1088 | } |
1089 | while (xb < xbe); |
1090 | while (xa < xae) |
1091 | { |
1092 | y= *xa++ - borrow; |
1093 | borrow= y >> 32 & (ULong)1; |
1094 | *xc++= (ULong) (y & FFFFFFFF); |
1095 | } |
1096 | while (!*--xc) |
1097 | wa--; |
1098 | c->wds= wa; |
1099 | return c; |
1100 | } |
1101 | |
1102 | |
1103 | static Bigint *d2b(U *d, int *e, int *bits, Stack_alloc *alloc) |
1104 | { |
1105 | Bigint *b; |
1106 | int de, k; |
1107 | ULong *x, y, z; |
1108 | int i; |
1109 | #define d0 word0(d) |
1110 | #define d1 word1(d) |
1111 | |
1112 | b= Balloc(1, alloc); |
1113 | x= b->p.x; |
1114 | |
1115 | z= d0 & Frac_mask; |
1116 | d0 &= 0x7fffffff; /* clear sign bit, which we ignore */ |
1117 | if ((de= (int)(d0 >> Exp_shift))) |
1118 | z|= Exp_msk1; |
1119 | if ((y= d1)) |
1120 | { |
1121 | if ((k= lo0bits(&y))) |
1122 | { |
1123 | x[0]= y | z << (32 - k); |
1124 | z>>= k; |
1125 | } |
1126 | else |
1127 | x[0]= y; |
1128 | i= b->wds= (x[1]= z) ? 2 : 1; |
1129 | } |
1130 | else |
1131 | { |
1132 | k= lo0bits(&z); |
1133 | x[0]= z; |
1134 | i= b->wds= 1; |
1135 | k+= 32; |
1136 | } |
1137 | if (de) |
1138 | { |
1139 | *e= de - Bias - (P-1) + k; |
1140 | *bits= P - k; |
1141 | } |
1142 | else |
1143 | { |
1144 | *e= de - Bias - (P-1) + 1 + k; |
1145 | *bits= 32*i - hi0bits(x[i-1]); |
1146 | } |
1147 | return b; |
1148 | #undef d0 |
1149 | #undef d1 |
1150 | } |
1151 | |
1152 | |
1153 | static const double tens[] = |
1154 | { |
1155 | 1e0, 1e1, 1e2, 1e3, 1e4, 1e5, 1e6, 1e7, 1e8, 1e9, |
1156 | 1e10, 1e11, 1e12, 1e13, 1e14, 1e15, 1e16, 1e17, 1e18, 1e19, |
1157 | 1e20, 1e21, 1e22 |
1158 | }; |
1159 | |
1160 | static const double bigtens[]= { 1e16, 1e32, 1e64, 1e128, 1e256 }; |
1161 | /* |
1162 | The factor of 2^53 in tinytens[4] helps us avoid setting the underflow |
1163 | flag unnecessarily. It leads to a song and dance at the end of strtod. |
1164 | */ |
1165 | #define Scale_Bit 0x10 |
1166 | #define n_bigtens 5 |
1167 | |
1168 | |
1169 | static int quorem(Bigint *b, Bigint *S) |
1170 | { |
1171 | int n; |
1172 | ULong *bx, *bxe, q, *sx, *sxe; |
1173 | ULLong borrow, carry, y, ys; |
1174 | |
1175 | n= S->wds; |
1176 | if (b->wds < n) |
1177 | return 0; |
1178 | sx= S->p.x; |
1179 | sxe= sx + --n; |
1180 | bx= b->p.x; |
1181 | bxe= bx + n; |
1182 | q= *bxe / (*sxe + 1); /* ensure q <= true quotient */ |
1183 | if (q) |
1184 | { |
1185 | borrow= 0; |
1186 | carry= 0; |
1187 | do |
1188 | { |
1189 | ys= *sx++ * (ULLong)q + carry; |
1190 | carry= ys >> 32; |
1191 | y= *bx - (ys & FFFFFFFF) - borrow; |
1192 | borrow= y >> 32 & (ULong)1; |
1193 | *bx++= (ULong) (y & FFFFFFFF); |
1194 | } |
1195 | while (sx <= sxe); |
1196 | if (!*bxe) |
1197 | { |
1198 | bx= b->p.x; |
1199 | while (--bxe > bx && !*bxe) |
1200 | --n; |
1201 | b->wds= n; |
1202 | } |
1203 | } |
1204 | if (cmp(b, S) >= 0) |
1205 | { |
1206 | q++; |
1207 | borrow= 0; |
1208 | carry= 0; |
1209 | bx= b->p.x; |
1210 | sx= S->p.x; |
1211 | do |
1212 | { |
1213 | ys= *sx++ + carry; |
1214 | carry= ys >> 32; |
1215 | y= *bx - (ys & FFFFFFFF) - borrow; |
1216 | borrow= y >> 32 & (ULong)1; |
1217 | *bx++= (ULong) (y & FFFFFFFF); |
1218 | } |
1219 | while (sx <= sxe); |
1220 | bx= b->p.x; |
1221 | bxe= bx + n; |
1222 | if (!*bxe) |
1223 | { |
1224 | while (--bxe > bx && !*bxe) |
1225 | --n; |
1226 | b->wds= n; |
1227 | } |
1228 | } |
1229 | return q; |
1230 | } |
1231 | |
1232 | |
1233 | /* |
1234 | dtoa for IEEE arithmetic (dmg): convert double to ASCII string. |
1235 | |
1236 | Inspired by "How to Print Floating-Point Numbers Accurately" by |
1237 | Guy L. Steele, Jr. and Jon L. White [Proc. ACM SIGPLAN '90, pp. 112-126]. |
1238 | |
1239 | Modifications: |
1240 | 1. Rather than iterating, we use a simple numeric overestimate |
1241 | to determine k= floor(log10(d)). We scale relevant |
1242 | quantities using O(log2(k)) rather than O(k) multiplications. |
1243 | 2. For some modes > 2 (corresponding to ecvt and fcvt), we don't |
1244 | try to generate digits strictly left to right. Instead, we |
1245 | compute with fewer bits and propagate the carry if necessary |
1246 | when rounding the final digit up. This is often faster. |
1247 | 3. Under the assumption that input will be rounded nearest, |
1248 | mode 0 renders 1e23 as 1e23 rather than 9.999999999999999e22. |
1249 | That is, we allow equality in stopping tests when the |
1250 | round-nearest rule will give the same floating-point value |
1251 | as would satisfaction of the stopping test with strict |
1252 | inequality. |
1253 | 4. We remove common factors of powers of 2 from relevant |
1254 | quantities. |
1255 | 5. When converting floating-point integers less than 1e16, |
1256 | we use floating-point arithmetic rather than resorting |
1257 | to multiple-precision integers. |
1258 | 6. When asked to produce fewer than 15 digits, we first try |
1259 | to get by with floating-point arithmetic; we resort to |
1260 | multiple-precision integer arithmetic only if we cannot |
1261 | guarantee that the floating-point calculation has given |
1262 | the correctly rounded result. For k requested digits and |
1263 | "uniformly" distributed input, the probability is |
1264 | something like 10^(k-15) that we must resort to the Long |
1265 | calculation. |
1266 | */ |
1267 | |
1268 | static char *dtoa(double dd, int mode, int ndigits, int *decpt, int *sign, |
1269 | char **rve, char *buf, size_t buf_size) |
1270 | { |
1271 | /* |
1272 | Arguments ndigits, decpt, sign are similar to those |
1273 | of ecvt and fcvt; trailing zeros are suppressed from |
1274 | the returned string. If not null, *rve is set to point |
1275 | to the end of the return value. If d is +-Infinity or NaN, |
1276 | then *decpt is set to DTOA_OVERFLOW. |
1277 | |
1278 | mode: |
1279 | 0 ==> shortest string that yields d when read in |
1280 | and rounded to nearest. |
1281 | 1 ==> like 0, but with Steele & White stopping rule; |
1282 | e.g. with IEEE P754 arithmetic , mode 0 gives |
1283 | 1e23 whereas mode 1 gives 9.999999999999999e22. |
1284 | 2 ==> MAX(1,ndigits) significant digits. This gives a |
1285 | return value similar to that of ecvt, except |
1286 | that trailing zeros are suppressed. |
1287 | 3 ==> through ndigits past the decimal point. This |
1288 | gives a return value similar to that from fcvt, |
1289 | except that trailing zeros are suppressed, and |
1290 | ndigits can be negative. |
1291 | 4,5 ==> similar to 2 and 3, respectively, but (in |
1292 | round-nearest mode) with the tests of mode 0 to |
1293 | possibly return a shorter string that rounds to d. |
1294 | With IEEE arithmetic and compilation with |
1295 | -DHonor_FLT_ROUNDS, modes 4 and 5 behave the same |
1296 | as modes 2 and 3 when FLT_ROUNDS != 1. |
1297 | 6-9 ==> Debugging modes similar to mode - 4: don't try |
1298 | fast floating-point estimate (if applicable). |
1299 | |
1300 | Values of mode other than 0-9 are treated as mode 0. |
1301 | |
1302 | Sufficient space is allocated to the return value |
1303 | to hold the suppressed trailing zeros. |
1304 | */ |
1305 | |
1306 | int bbits, b2, b5, be, dig, i, ieps, UNINIT_VAR(ilim), ilim0, |
1307 | UNINIT_VAR(ilim1), j, j1, k, k0, k_check, leftright, m2, m5, s2, s5, |
1308 | spec_case, try_quick; |
1309 | Long L; |
1310 | int denorm; |
1311 | ULong x; |
1312 | Bigint *b, *b1, *delta, *mlo, *mhi, *S; |
1313 | U d2, eps, u; |
1314 | double ds; |
1315 | char *s, *s0; |
1316 | #ifdef Honor_FLT_ROUNDS |
1317 | int rounding; |
1318 | #endif |
1319 | Stack_alloc alloc; |
1320 | |
1321 | alloc.begin= alloc.free= buf; |
1322 | alloc.end= buf + buf_size; |
1323 | memset(alloc.freelist, 0, sizeof(alloc.freelist)); |
1324 | |
1325 | u.d= dd; |
1326 | if (word0(&u) & Sign_bit) |
1327 | { |
1328 | /* set sign for everything, including 0's and NaNs */ |
1329 | *sign= 1; |
1330 | word0(&u) &= ~Sign_bit; /* clear sign bit */ |
1331 | } |
1332 | else |
1333 | *sign= 0; |
1334 | |
1335 | /* If infinity, set decpt to DTOA_OVERFLOW, if 0 set it to 1 */ |
1336 | /* coverity[assign_where_compare_meant] */ |
1337 | if (((word0(&u) & Exp_mask) == Exp_mask && (*decpt= DTOA_OVERFLOW)) || |
1338 | /* coverity[assign_where_compare_meant] */ |
1339 | (!dval(&u) && (*decpt= 1))) |
1340 | { |
1341 | /* Infinity, NaN, 0 */ |
1342 | char *res= (char*) dtoa_alloc(2, &alloc); |
1343 | res[0]= '0'; |
1344 | res[1]= '\0'; |
1345 | if (rve) |
1346 | *rve= res + 1; |
1347 | return res; |
1348 | } |
1349 | |
1350 | #ifdef Honor_FLT_ROUNDS |
1351 | if ((rounding= Flt_Rounds) >= 2) |
1352 | { |
1353 | if (*sign) |
1354 | rounding= rounding == 2 ? 0 : 2; |
1355 | else |
1356 | if (rounding != 2) |
1357 | rounding= 0; |
1358 | } |
1359 | #endif |
1360 | |
1361 | b= d2b(&u, &be, &bbits, &alloc); |
1362 | if ((i= (int)(word0(&u) >> Exp_shift1 & (Exp_mask>>Exp_shift1)))) |
1363 | { |
1364 | dval(&d2)= dval(&u); |
1365 | word0(&d2) &= Frac_mask1; |
1366 | word0(&d2) |= Exp_11; |
1367 | |
1368 | /* |
1369 | log(x) ~=~ log(1.5) + (x-1.5)/1.5 |
1370 | log10(x) = log(x) / log(10) |
1371 | ~=~ log(1.5)/log(10) + (x-1.5)/(1.5*log(10)) |
1372 | log10(d)= (i-Bias)*log(2)/log(10) + log10(d2) |
1373 | |
1374 | This suggests computing an approximation k to log10(d) by |
1375 | |
1376 | k= (i - Bias)*0.301029995663981 |
1377 | + ( (d2-1.5)*0.289529654602168 + 0.176091259055681 ); |
1378 | |
1379 | We want k to be too large rather than too small. |
1380 | The error in the first-order Taylor series approximation |
1381 | is in our favor, so we just round up the constant enough |
1382 | to compensate for any error in the multiplication of |
1383 | (i - Bias) by 0.301029995663981; since |i - Bias| <= 1077, |
1384 | and 1077 * 0.30103 * 2^-52 ~=~ 7.2e-14, |
1385 | adding 1e-13 to the constant term more than suffices. |
1386 | Hence we adjust the constant term to 0.1760912590558. |
1387 | (We could get a more accurate k by invoking log10, |
1388 | but this is probably not worthwhile.) |
1389 | */ |
1390 | |
1391 | i-= Bias; |
1392 | denorm= 0; |
1393 | } |
1394 | else |
1395 | { |
1396 | /* d is denormalized */ |
1397 | |
1398 | i= bbits + be + (Bias + (P-1) - 1); |
1399 | x= i > 32 ? word0(&u) << (64 - i) | word1(&u) >> (i - 32) |
1400 | : word1(&u) << (32 - i); |
1401 | dval(&d2)= x; |
1402 | word0(&d2)-= 31*Exp_msk1; /* adjust exponent */ |
1403 | i-= (Bias + (P-1) - 1) + 1; |
1404 | denorm= 1; |
1405 | } |
1406 | ds= (dval(&d2)-1.5)*0.289529654602168 + 0.1760912590558 + i*0.301029995663981; |
1407 | k= (int)ds; |
1408 | if (ds < 0. && ds != k) |
1409 | k--; /* want k= floor(ds) */ |
1410 | k_check= 1; |
1411 | if (k >= 0 && k <= Ten_pmax) |
1412 | { |
1413 | if (dval(&u) < tens[k]) |
1414 | k--; |
1415 | k_check= 0; |
1416 | } |
1417 | j= bbits - i - 1; |
1418 | if (j >= 0) |
1419 | { |
1420 | b2= 0; |
1421 | s2= j; |
1422 | } |
1423 | else |
1424 | { |
1425 | b2= -j; |
1426 | s2= 0; |
1427 | } |
1428 | if (k >= 0) |
1429 | { |
1430 | b5= 0; |
1431 | s5= k; |
1432 | s2+= k; |
1433 | } |
1434 | else |
1435 | { |
1436 | b2-= k; |
1437 | b5= -k; |
1438 | s5= 0; |
1439 | } |
1440 | if (mode < 0 || mode > 9) |
1441 | mode= 0; |
1442 | |
1443 | #ifdef Check_FLT_ROUNDS |
1444 | try_quick= Rounding == 1; |
1445 | #else |
1446 | try_quick= 1; |
1447 | #endif |
1448 | |
1449 | if (mode > 5) |
1450 | { |
1451 | mode-= 4; |
1452 | try_quick= 0; |
1453 | } |
1454 | leftright= 1; |
1455 | switch (mode) { |
1456 | case 0: |
1457 | case 1: |
1458 | ilim= ilim1= -1; |
1459 | i= 18; |
1460 | ndigits= 0; |
1461 | break; |
1462 | case 2: |
1463 | leftright= 0; |
1464 | /* fall through */ |
1465 | case 4: |
1466 | if (ndigits <= 0) |
1467 | ndigits= 1; |
1468 | ilim= ilim1= i= ndigits; |
1469 | break; |
1470 | case 3: |
1471 | leftright= 0; |
1472 | /* fall through */ |
1473 | case 5: |
1474 | i= ndigits + k + 1; |
1475 | ilim= i; |
1476 | ilim1= i - 1; |
1477 | if (i <= 0) |
1478 | i= 1; |
1479 | } |
1480 | s= s0= dtoa_alloc(i, &alloc); |
1481 | |
1482 | #ifdef Honor_FLT_ROUNDS |
1483 | if (mode > 1 && rounding != 1) |
1484 | leftright= 0; |
1485 | #endif |
1486 | |
1487 | if (ilim >= 0 && ilim <= Quick_max && try_quick) |
1488 | { |
1489 | /* Try to get by with floating-point arithmetic. */ |
1490 | i= 0; |
1491 | dval(&d2)= dval(&u); |
1492 | k0= k; |
1493 | ilim0= ilim; |
1494 | ieps= 2; /* conservative */ |
1495 | if (k > 0) |
1496 | { |
1497 | ds= tens[k&0xf]; |
1498 | j= k >> 4; |
1499 | if (j & Bletch) |
1500 | { |
1501 | /* prevent overflows */ |
1502 | j&= Bletch - 1; |
1503 | dval(&u)/= bigtens[n_bigtens-1]; |
1504 | ieps++; |
1505 | } |
1506 | for (; j; j>>= 1, i++) |
1507 | { |
1508 | if (j & 1) |
1509 | { |
1510 | ieps++; |
1511 | ds*= bigtens[i]; |
1512 | } |
1513 | } |
1514 | dval(&u)/= ds; |
1515 | } |
1516 | else if ((j1= -k)) |
1517 | { |
1518 | dval(&u)*= tens[j1 & 0xf]; |
1519 | for (j= j1 >> 4; j; j>>= 1, i++) |
1520 | { |
1521 | if (j & 1) |
1522 | { |
1523 | ieps++; |
1524 | dval(&u)*= bigtens[i]; |
1525 | } |
1526 | } |
1527 | } |
1528 | if (k_check && dval(&u) < 1. && ilim > 0) |
1529 | { |
1530 | if (ilim1 <= 0) |
1531 | goto fast_failed; |
1532 | ilim= ilim1; |
1533 | k--; |
1534 | dval(&u)*= 10.; |
1535 | ieps++; |
1536 | } |
1537 | dval(&eps)= ieps*dval(&u) + 7.; |
1538 | word0(&eps)-= (P-1)*Exp_msk1; |
1539 | if (ilim == 0) |
1540 | { |
1541 | S= mhi= 0; |
1542 | dval(&u)-= 5.; |
1543 | if (dval(&u) > dval(&eps)) |
1544 | goto one_digit; |
1545 | if (dval(&u) < -dval(&eps)) |
1546 | goto no_digits; |
1547 | goto fast_failed; |
1548 | } |
1549 | if (leftright) |
1550 | { |
1551 | /* Use Steele & White method of only generating digits needed. */ |
1552 | dval(&eps)= 0.5/tens[ilim-1] - dval(&eps); |
1553 | for (i= 0;;) |
1554 | { |
1555 | L= (Long) dval(&u); |
1556 | dval(&u)-= L; |
1557 | *s++= '0' + (int)L; |
1558 | if (dval(&u) < dval(&eps)) |
1559 | goto ret1; |
1560 | if (1. - dval(&u) < dval(&eps)) |
1561 | goto bump_up; |
1562 | if (++i >= ilim) |
1563 | break; |
1564 | dval(&eps)*= 10.; |
1565 | dval(&u)*= 10.; |
1566 | } |
1567 | } |
1568 | else |
1569 | { |
1570 | /* Generate ilim digits, then fix them up. */ |
1571 | dval(&eps)*= tens[ilim-1]; |
1572 | for (i= 1;; i++, dval(&u)*= 10.) |
1573 | { |
1574 | L= (Long)(dval(&u)); |
1575 | if (!(dval(&u)-= L)) |
1576 | ilim= i; |
1577 | *s++= '0' + (int)L; |
1578 | if (i == ilim) |
1579 | { |
1580 | if (dval(&u) > 0.5 + dval(&eps)) |
1581 | goto bump_up; |
1582 | else if (dval(&u) < 0.5 - dval(&eps)) |
1583 | { |
1584 | while (*--s == '0'); |
1585 | s++; |
1586 | goto ret1; |
1587 | } |
1588 | break; |
1589 | } |
1590 | } |
1591 | } |
1592 | fast_failed: |
1593 | s= s0; |
1594 | dval(&u)= dval(&d2); |
1595 | k= k0; |
1596 | ilim= ilim0; |
1597 | } |
1598 | |
1599 | /* Do we have a "small" integer? */ |
1600 | |
1601 | if (be >= 0 && k <= Int_max) |
1602 | { |
1603 | /* Yes. */ |
1604 | ds= tens[k]; |
1605 | if (ndigits < 0 && ilim <= 0) |
1606 | { |
1607 | S= mhi= 0; |
1608 | if (ilim < 0 || dval(&u) <= 5*ds) |
1609 | goto no_digits; |
1610 | goto one_digit; |
1611 | } |
1612 | for (i= 1;; i++, dval(&u)*= 10.) |
1613 | { |
1614 | L= (Long)(dval(&u) / ds); |
1615 | dval(&u)-= L*ds; |
1616 | #ifdef Check_FLT_ROUNDS |
1617 | /* If FLT_ROUNDS == 2, L will usually be high by 1 */ |
1618 | if (dval(&u) < 0) |
1619 | { |
1620 | L--; |
1621 | dval(&u)+= ds; |
1622 | } |
1623 | #endif |
1624 | *s++= '0' + (int)L; |
1625 | if (!dval(&u)) |
1626 | { |
1627 | break; |
1628 | } |
1629 | if (i == ilim) |
1630 | { |
1631 | #ifdef Honor_FLT_ROUNDS |
1632 | if (mode > 1) |
1633 | { |
1634 | switch (rounding) { |
1635 | case 0: goto ret1; |
1636 | case 2: goto bump_up; |
1637 | } |
1638 | } |
1639 | #endif |
1640 | dval(&u)+= dval(&u); |
1641 | if (dval(&u) > ds || (dval(&u) == ds && L & 1)) |
1642 | { |
1643 | bump_up: |
1644 | while (*--s == '9') |
1645 | if (s == s0) |
1646 | { |
1647 | k++; |
1648 | *s= '0'; |
1649 | break; |
1650 | } |
1651 | ++*s++; |
1652 | } |
1653 | break; |
1654 | } |
1655 | } |
1656 | goto ret1; |
1657 | } |
1658 | |
1659 | m2= b2; |
1660 | m5= b5; |
1661 | mhi= mlo= 0; |
1662 | if (leftright) |
1663 | { |
1664 | i = denorm ? be + (Bias + (P-1) - 1 + 1) : 1 + P - bbits; |
1665 | b2+= i; |
1666 | s2+= i; |
1667 | mhi= i2b(1, &alloc); |
1668 | } |
1669 | if (m2 > 0 && s2 > 0) |
1670 | { |
1671 | i= m2 < s2 ? m2 : s2; |
1672 | b2-= i; |
1673 | m2-= i; |
1674 | s2-= i; |
1675 | } |
1676 | if (b5 > 0) |
1677 | { |
1678 | if (leftright) |
1679 | { |
1680 | if (m5 > 0) |
1681 | { |
1682 | mhi= pow5mult(mhi, m5, &alloc); |
1683 | b1= mult(mhi, b, &alloc); |
1684 | Bfree(b, &alloc); |
1685 | b= b1; |
1686 | } |
1687 | if ((j= b5 - m5)) |
1688 | b= pow5mult(b, j, &alloc); |
1689 | } |
1690 | else |
1691 | b= pow5mult(b, b5, &alloc); |
1692 | } |
1693 | S= i2b(1, &alloc); |
1694 | if (s5 > 0) |
1695 | S= pow5mult(S, s5, &alloc); |
1696 | |
1697 | /* Check for special case that d is a normalized power of 2. */ |
1698 | |
1699 | spec_case= 0; |
1700 | if ((mode < 2 || leftright) |
1701 | #ifdef Honor_FLT_ROUNDS |
1702 | && rounding == 1 |
1703 | #endif |
1704 | ) |
1705 | { |
1706 | if (!word1(&u) && !(word0(&u) & Bndry_mask) && |
1707 | word0(&u) & (Exp_mask & ~Exp_msk1) |
1708 | ) |
1709 | { |
1710 | /* The special case */ |
1711 | b2+= Log2P; |
1712 | s2+= Log2P; |
1713 | spec_case= 1; |
1714 | } |
1715 | } |
1716 | |
1717 | /* |
1718 | Arrange for convenient computation of quotients: |
1719 | shift left if necessary so divisor has 4 leading 0 bits. |
1720 | |
1721 | Perhaps we should just compute leading 28 bits of S once |
1722 | a nd for all and pass them and a shift to quorem, so it |
1723 | can do shifts and ors to compute the numerator for q. |
1724 | */ |
1725 | if ((i= ((s5 ? 32 - hi0bits(S->p.x[S->wds-1]) : 1) + s2) & 0x1f)) |
1726 | i= 32 - i; |
1727 | if (i > 4) |
1728 | { |
1729 | i-= 4; |
1730 | b2+= i; |
1731 | m2+= i; |
1732 | s2+= i; |
1733 | } |
1734 | else if (i < 4) |
1735 | { |
1736 | i+= 28; |
1737 | b2+= i; |
1738 | m2+= i; |
1739 | s2+= i; |
1740 | } |
1741 | if (b2 > 0) |
1742 | b= lshift(b, b2, &alloc); |
1743 | if (s2 > 0) |
1744 | S= lshift(S, s2, &alloc); |
1745 | if (k_check) |
1746 | { |
1747 | if (cmp(b,S) < 0) |
1748 | { |
1749 | k--; |
1750 | /* we botched the k estimate */ |
1751 | b= multadd(b, 10, 0, &alloc); |
1752 | if (leftright) |
1753 | mhi= multadd(mhi, 10, 0, &alloc); |
1754 | ilim= ilim1; |
1755 | } |
1756 | } |
1757 | if (ilim <= 0 && (mode == 3 || mode == 5)) |
1758 | { |
1759 | if (ilim < 0 || cmp(b,S= multadd(S,5,0, &alloc)) <= 0) |
1760 | { |
1761 | /* no digits, fcvt style */ |
1762 | no_digits: |
1763 | k= -1 - ndigits; |
1764 | goto ret; |
1765 | } |
1766 | one_digit: |
1767 | *s++= '1'; |
1768 | k++; |
1769 | goto ret; |
1770 | } |
1771 | if (leftright) |
1772 | { |
1773 | if (m2 > 0) |
1774 | mhi= lshift(mhi, m2, &alloc); |
1775 | |
1776 | /* |
1777 | Compute mlo -- check for special case that d is a normalized power of 2. |
1778 | */ |
1779 | |
1780 | mlo= mhi; |
1781 | if (spec_case) |
1782 | { |
1783 | mhi= Balloc(mhi->k, &alloc); |
1784 | Bcopy(mhi, mlo); |
1785 | mhi= lshift(mhi, Log2P, &alloc); |
1786 | } |
1787 | |
1788 | for (i= 1;;i++) |
1789 | { |
1790 | dig= quorem(b,S) + '0'; |
1791 | /* Do we yet have the shortest decimal string that will round to d? */ |
1792 | j= cmp(b, mlo); |
1793 | delta= diff(S, mhi, &alloc); |
1794 | j1= delta->sign ? 1 : cmp(b, delta); |
1795 | Bfree(delta, &alloc); |
1796 | if (j1 == 0 && mode != 1 && !(word1(&u) & 1) |
1797 | #ifdef Honor_FLT_ROUNDS |
1798 | && rounding >= 1 |
1799 | #endif |
1800 | ) |
1801 | { |
1802 | if (dig == '9') |
1803 | goto round_9_up; |
1804 | if (j > 0) |
1805 | dig++; |
1806 | *s++= dig; |
1807 | goto ret; |
1808 | } |
1809 | if (j < 0 || (j == 0 && mode != 1 && !(word1(&u) & 1))) |
1810 | { |
1811 | if (!b->p.x[0] && b->wds <= 1) |
1812 | { |
1813 | goto accept_dig; |
1814 | } |
1815 | #ifdef Honor_FLT_ROUNDS |
1816 | if (mode > 1) |
1817 | switch (rounding) { |
1818 | case 0: goto accept_dig; |
1819 | case 2: goto keep_dig; |
1820 | } |
1821 | #endif /*Honor_FLT_ROUNDS*/ |
1822 | if (j1 > 0) |
1823 | { |
1824 | b= lshift(b, 1, &alloc); |
1825 | j1= cmp(b, S); |
1826 | if ((j1 > 0 || (j1 == 0 && dig & 1)) |
1827 | && dig++ == '9') |
1828 | goto round_9_up; |
1829 | } |
1830 | accept_dig: |
1831 | *s++= dig; |
1832 | goto ret; |
1833 | } |
1834 | if (j1 > 0) |
1835 | { |
1836 | #ifdef Honor_FLT_ROUNDS |
1837 | if (!rounding) |
1838 | goto accept_dig; |
1839 | #endif |
1840 | if (dig == '9') |
1841 | { /* possible if i == 1 */ |
1842 | round_9_up: |
1843 | *s++= '9'; |
1844 | goto roundoff; |
1845 | } |
1846 | *s++= dig + 1; |
1847 | goto ret; |
1848 | } |
1849 | #ifdef Honor_FLT_ROUNDS |
1850 | keep_dig: |
1851 | #endif |
1852 | *s++= dig; |
1853 | if (i == ilim) |
1854 | break; |
1855 | b= multadd(b, 10, 0, &alloc); |
1856 | if (mlo == mhi) |
1857 | mlo= mhi= multadd(mhi, 10, 0, &alloc); |
1858 | else |
1859 | { |
1860 | mlo= multadd(mlo, 10, 0, &alloc); |
1861 | mhi= multadd(mhi, 10, 0, &alloc); |
1862 | } |
1863 | } |
1864 | } |
1865 | else |
1866 | for (i= 1;; i++) |
1867 | { |
1868 | *s++= dig= quorem(b,S) + '0'; |
1869 | if (!b->p.x[0] && b->wds <= 1) |
1870 | { |
1871 | goto ret; |
1872 | } |
1873 | if (i >= ilim) |
1874 | break; |
1875 | b= multadd(b, 10, 0, &alloc); |
1876 | } |
1877 | |
1878 | /* Round off last digit */ |
1879 | |
1880 | #ifdef Honor_FLT_ROUNDS |
1881 | switch (rounding) { |
1882 | case 0: goto trimzeros; |
1883 | case 2: goto roundoff; |
1884 | } |
1885 | #endif |
1886 | b= lshift(b, 1, &alloc); |
1887 | j= cmp(b, S); |
1888 | if (j > 0 || (j == 0 && dig & 1)) |
1889 | { |
1890 | roundoff: |
1891 | while (*--s == '9') |
1892 | if (s == s0) |
1893 | { |
1894 | k++; |
1895 | *s++= '1'; |
1896 | goto ret; |
1897 | } |
1898 | ++*s++; |
1899 | } |
1900 | else |
1901 | { |
1902 | #ifdef Honor_FLT_ROUNDS |
1903 | trimzeros: |
1904 | #endif |
1905 | while (*--s == '0'); |
1906 | s++; |
1907 | } |
1908 | ret: |
1909 | Bfree(S, &alloc); |
1910 | if (mhi) |
1911 | { |
1912 | if (mlo && mlo != mhi) |
1913 | Bfree(mlo, &alloc); |
1914 | Bfree(mhi, &alloc); |
1915 | } |
1916 | ret1: |
1917 | Bfree(b, &alloc); |
1918 | *s= 0; |
1919 | *decpt= k + 1; |
1920 | if (rve) |
1921 | *rve= s; |
1922 | return s0; |
1923 | } |
1924 | |