1 | /* |
2 | * Copyright (c) 2008-2020 Stefan Krah. All rights reserved. |
3 | * |
4 | * Redistribution and use in source and binary forms, with or without |
5 | * modification, are permitted provided that the following conditions |
6 | * are met: |
7 | * |
8 | * 1. Redistributions of source code must retain the above copyright |
9 | * notice, this list of conditions and the following disclaimer. |
10 | * |
11 | * 2. Redistributions in binary form must reproduce the above copyright |
12 | * notice, this list of conditions and the following disclaimer in the |
13 | * documentation and/or other materials provided with the distribution. |
14 | * |
15 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS "AS IS" AND |
16 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
17 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
18 | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
19 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
20 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
21 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
22 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
23 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
24 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
25 | * SUCH DAMAGE. |
26 | */ |
27 | |
28 | |
29 | #include "mpdecimal.h" |
30 | |
31 | #include <assert.h> |
32 | #include <limits.h> |
33 | #include <math.h> |
34 | #include <stdio.h> |
35 | #include <stdlib.h> |
36 | #include <string.h> |
37 | |
38 | #include "basearith.h" |
39 | #include "bits.h" |
40 | #include "constants.h" |
41 | #include "convolute.h" |
42 | #include "crt.h" |
43 | #include "mpalloc.h" |
44 | #include "typearith.h" |
45 | |
46 | #ifdef PPRO |
47 | #if defined(_MSC_VER) |
48 | #include <float.h> |
49 | #pragma float_control(precise, on) |
50 | #pragma fenv_access(on) |
51 | #elif !defined(__OpenBSD__) && !defined(__NetBSD__) |
52 | /* C99 */ |
53 | #include <fenv.h> |
54 | #pragma STDC FENV_ACCESS ON |
55 | #endif |
56 | #endif |
57 | |
58 | |
59 | /* Disable warning that is part of -Wextra since gcc 7.0. */ |
60 | #if defined(__GNUC__) && !defined(__INTEL_COMPILER) && __GNUC__ >= 7 |
61 | #pragma GCC diagnostic ignored "-Wimplicit-fallthrough" |
62 | #endif |
63 | |
64 | |
65 | #if defined(_MSC_VER) |
66 | #define ALWAYS_INLINE __forceinline |
67 | #elif defined (__IBMC__) || defined(LEGACY_COMPILER) |
68 | #define ALWAYS_INLINE |
69 | #undef inline |
70 | #define inline |
71 | #else |
72 | #ifdef TEST_COVERAGE |
73 | #define ALWAYS_INLINE |
74 | #else |
75 | #define ALWAYS_INLINE inline __attribute__ ((always_inline)) |
76 | #endif |
77 | #endif |
78 | |
79 | |
80 | #define MPD_NEWTONDIV_CUTOFF 1024L |
81 | |
82 | #define MPD_NEW_STATIC(name, flags, exp, digits, len) \ |
83 | mpd_uint_t name##_data[MPD_MINALLOC_MAX]; \ |
84 | mpd_t name = {flags|MPD_STATIC|MPD_STATIC_DATA, exp, digits, \ |
85 | len, MPD_MINALLOC_MAX, name##_data} |
86 | |
87 | #define MPD_NEW_CONST(name, flags, exp, digits, len, alloc, initval) \ |
88 | mpd_uint_t name##_data[alloc] = {initval}; \ |
89 | mpd_t name = {flags|MPD_STATIC|MPD_CONST_DATA, exp, digits, \ |
90 | len, alloc, name##_data} |
91 | |
92 | #define MPD_NEW_SHARED(name, a) \ |
93 | mpd_t name = {(a->flags&~MPD_DATAFLAGS)|MPD_STATIC|MPD_SHARED_DATA, \ |
94 | a->exp, a->digits, a->len, a->alloc, a->data} |
95 | |
96 | |
97 | static mpd_uint_t data_one[1] = {1}; |
98 | static mpd_uint_t data_zero[1] = {0}; |
99 | static const mpd_t one = {MPD_STATIC|MPD_CONST_DATA, 0, 1, 1, 1, data_one}; |
100 | static const mpd_t minus_one = {MPD_NEG|MPD_STATIC|MPD_CONST_DATA, 0, 1, 1, 1, |
101 | data_one}; |
102 | static const mpd_t zero = {MPD_STATIC|MPD_CONST_DATA, 0, 1, 1, 1, data_zero}; |
103 | |
104 | static inline void _mpd_check_exp(mpd_t *dec, const mpd_context_t *ctx, |
105 | uint32_t *status); |
106 | static void _settriple(mpd_t *result, uint8_t sign, mpd_uint_t a, |
107 | mpd_ssize_t exp); |
108 | static inline mpd_ssize_t _mpd_real_size(mpd_uint_t *data, mpd_ssize_t size); |
109 | |
110 | static int _mpd_cmp_abs(const mpd_t *a, const mpd_t *b); |
111 | |
112 | static void _mpd_qadd(mpd_t *result, const mpd_t *a, const mpd_t *b, |
113 | const mpd_context_t *ctx, uint32_t *status); |
114 | static inline void _mpd_qmul(mpd_t *result, const mpd_t *a, const mpd_t *b, |
115 | const mpd_context_t *ctx, uint32_t *status); |
116 | static void _mpd_base_ndivmod(mpd_t *q, mpd_t *r, const mpd_t *a, |
117 | const mpd_t *b, uint32_t *status); |
118 | static inline void _mpd_qpow_uint(mpd_t *result, const mpd_t *base, |
119 | mpd_uint_t exp, uint8_t resultsign, |
120 | const mpd_context_t *ctx, uint32_t *status); |
121 | |
122 | static mpd_uint_t mpd_qsshiftr(mpd_t *result, const mpd_t *a, mpd_ssize_t n); |
123 | |
124 | |
125 | /******************************************************************************/ |
126 | /* Version */ |
127 | /******************************************************************************/ |
128 | |
129 | const char * |
130 | mpd_version(void) |
131 | { |
132 | return MPD_VERSION; |
133 | } |
134 | |
135 | |
136 | /******************************************************************************/ |
137 | /* Performance critical inline functions */ |
138 | /******************************************************************************/ |
139 | |
140 | #ifdef CONFIG_64 |
141 | /* Digits in a word, primarily useful for the most significant word. */ |
142 | ALWAYS_INLINE int |
143 | mpd_word_digits(mpd_uint_t word) |
144 | { |
145 | if (word < mpd_pow10[9]) { |
146 | if (word < mpd_pow10[4]) { |
147 | if (word < mpd_pow10[2]) { |
148 | return (word < mpd_pow10[1]) ? 1 : 2; |
149 | } |
150 | return (word < mpd_pow10[3]) ? 3 : 4; |
151 | } |
152 | if (word < mpd_pow10[6]) { |
153 | return (word < mpd_pow10[5]) ? 5 : 6; |
154 | } |
155 | if (word < mpd_pow10[8]) { |
156 | return (word < mpd_pow10[7]) ? 7 : 8; |
157 | } |
158 | return 9; |
159 | } |
160 | if (word < mpd_pow10[14]) { |
161 | if (word < mpd_pow10[11]) { |
162 | return (word < mpd_pow10[10]) ? 10 : 11; |
163 | } |
164 | if (word < mpd_pow10[13]) { |
165 | return (word < mpd_pow10[12]) ? 12 : 13; |
166 | } |
167 | return 14; |
168 | } |
169 | if (word < mpd_pow10[18]) { |
170 | if (word < mpd_pow10[16]) { |
171 | return (word < mpd_pow10[15]) ? 15 : 16; |
172 | } |
173 | return (word < mpd_pow10[17]) ? 17 : 18; |
174 | } |
175 | |
176 | return (word < mpd_pow10[19]) ? 19 : 20; |
177 | } |
178 | #else |
179 | ALWAYS_INLINE int |
180 | mpd_word_digits(mpd_uint_t word) |
181 | { |
182 | if (word < mpd_pow10[4]) { |
183 | if (word < mpd_pow10[2]) { |
184 | return (word < mpd_pow10[1]) ? 1 : 2; |
185 | } |
186 | return (word < mpd_pow10[3]) ? 3 : 4; |
187 | } |
188 | if (word < mpd_pow10[6]) { |
189 | return (word < mpd_pow10[5]) ? 5 : 6; |
190 | } |
191 | if (word < mpd_pow10[8]) { |
192 | return (word < mpd_pow10[7]) ? 7 : 8; |
193 | } |
194 | |
195 | return (word < mpd_pow10[9]) ? 9 : 10; |
196 | } |
197 | #endif |
198 | |
199 | |
200 | /* Adjusted exponent */ |
201 | ALWAYS_INLINE mpd_ssize_t |
202 | mpd_adjexp(const mpd_t *dec) |
203 | { |
204 | return (dec->exp + dec->digits) - 1; |
205 | } |
206 | |
207 | /* Etiny */ |
208 | ALWAYS_INLINE mpd_ssize_t |
209 | mpd_etiny(const mpd_context_t *ctx) |
210 | { |
211 | return ctx->emin - (ctx->prec - 1); |
212 | } |
213 | |
214 | /* Etop: used for folding down in IEEE clamping */ |
215 | ALWAYS_INLINE mpd_ssize_t |
216 | mpd_etop(const mpd_context_t *ctx) |
217 | { |
218 | return ctx->emax - (ctx->prec - 1); |
219 | } |
220 | |
221 | /* Most significant word */ |
222 | ALWAYS_INLINE mpd_uint_t |
223 | mpd_msword(const mpd_t *dec) |
224 | { |
225 | assert(dec->len > 0); |
226 | return dec->data[dec->len-1]; |
227 | } |
228 | |
229 | /* Most significant digit of a word */ |
230 | inline mpd_uint_t |
231 | mpd_msd(mpd_uint_t word) |
232 | { |
233 | int n; |
234 | |
235 | n = mpd_word_digits(word); |
236 | return word / mpd_pow10[n-1]; |
237 | } |
238 | |
239 | /* Least significant digit of a word */ |
240 | ALWAYS_INLINE mpd_uint_t |
241 | mpd_lsd(mpd_uint_t word) |
242 | { |
243 | return word % 10; |
244 | } |
245 | |
246 | /* Coefficient size needed to store 'digits' */ |
247 | mpd_ssize_t |
248 | mpd_digits_to_size(mpd_ssize_t digits) |
249 | { |
250 | mpd_ssize_t q, r; |
251 | |
252 | _mpd_idiv_word(&q, &r, digits, MPD_RDIGITS); |
253 | return (r == 0) ? q : q+1; |
254 | } |
255 | |
256 | /* Number of digits in the exponent. Not defined for MPD_SSIZE_MIN. */ |
257 | inline int |
258 | mpd_exp_digits(mpd_ssize_t exp) |
259 | { |
260 | exp = (exp < 0) ? -exp : exp; |
261 | return mpd_word_digits(exp); |
262 | } |
263 | |
264 | /* Canonical */ |
265 | ALWAYS_INLINE int |
266 | mpd_iscanonical(const mpd_t *dec) |
267 | { |
268 | (void)dec; |
269 | return 1; |
270 | } |
271 | |
272 | /* Finite */ |
273 | ALWAYS_INLINE int |
274 | mpd_isfinite(const mpd_t *dec) |
275 | { |
276 | return !(dec->flags & MPD_SPECIAL); |
277 | } |
278 | |
279 | /* Infinite */ |
280 | ALWAYS_INLINE int |
281 | mpd_isinfinite(const mpd_t *dec) |
282 | { |
283 | return dec->flags & MPD_INF; |
284 | } |
285 | |
286 | /* NaN */ |
287 | ALWAYS_INLINE int |
288 | mpd_isnan(const mpd_t *dec) |
289 | { |
290 | return dec->flags & (MPD_NAN|MPD_SNAN); |
291 | } |
292 | |
293 | /* Negative */ |
294 | ALWAYS_INLINE int |
295 | mpd_isnegative(const mpd_t *dec) |
296 | { |
297 | return dec->flags & MPD_NEG; |
298 | } |
299 | |
300 | /* Positive */ |
301 | ALWAYS_INLINE int |
302 | mpd_ispositive(const mpd_t *dec) |
303 | { |
304 | return !(dec->flags & MPD_NEG); |
305 | } |
306 | |
307 | /* qNaN */ |
308 | ALWAYS_INLINE int |
309 | mpd_isqnan(const mpd_t *dec) |
310 | { |
311 | return dec->flags & MPD_NAN; |
312 | } |
313 | |
314 | /* Signed */ |
315 | ALWAYS_INLINE int |
316 | mpd_issigned(const mpd_t *dec) |
317 | { |
318 | return dec->flags & MPD_NEG; |
319 | } |
320 | |
321 | /* sNaN */ |
322 | ALWAYS_INLINE int |
323 | mpd_issnan(const mpd_t *dec) |
324 | { |
325 | return dec->flags & MPD_SNAN; |
326 | } |
327 | |
328 | /* Special */ |
329 | ALWAYS_INLINE int |
330 | mpd_isspecial(const mpd_t *dec) |
331 | { |
332 | return dec->flags & MPD_SPECIAL; |
333 | } |
334 | |
335 | /* Zero */ |
336 | ALWAYS_INLINE int |
337 | mpd_iszero(const mpd_t *dec) |
338 | { |
339 | return !mpd_isspecial(dec) && mpd_msword(dec) == 0; |
340 | } |
341 | |
342 | /* Test for zero when specials have been ruled out already */ |
343 | ALWAYS_INLINE int |
344 | mpd_iszerocoeff(const mpd_t *dec) |
345 | { |
346 | return mpd_msword(dec) == 0; |
347 | } |
348 | |
349 | /* Normal */ |
350 | inline int |
351 | mpd_isnormal(const mpd_t *dec, const mpd_context_t *ctx) |
352 | { |
353 | if (mpd_isspecial(dec)) return 0; |
354 | if (mpd_iszerocoeff(dec)) return 0; |
355 | |
356 | return mpd_adjexp(dec) >= ctx->emin; |
357 | } |
358 | |
359 | /* Subnormal */ |
360 | inline int |
361 | mpd_issubnormal(const mpd_t *dec, const mpd_context_t *ctx) |
362 | { |
363 | if (mpd_isspecial(dec)) return 0; |
364 | if (mpd_iszerocoeff(dec)) return 0; |
365 | |
366 | return mpd_adjexp(dec) < ctx->emin; |
367 | } |
368 | |
369 | /* Odd word */ |
370 | ALWAYS_INLINE int |
371 | mpd_isoddword(mpd_uint_t word) |
372 | { |
373 | return word & 1; |
374 | } |
375 | |
376 | /* Odd coefficient */ |
377 | ALWAYS_INLINE int |
378 | mpd_isoddcoeff(const mpd_t *dec) |
379 | { |
380 | return mpd_isoddword(dec->data[0]); |
381 | } |
382 | |
383 | /* 0 if dec is positive, 1 if dec is negative */ |
384 | ALWAYS_INLINE uint8_t |
385 | mpd_sign(const mpd_t *dec) |
386 | { |
387 | return dec->flags & MPD_NEG; |
388 | } |
389 | |
390 | /* 1 if dec is positive, -1 if dec is negative */ |
391 | ALWAYS_INLINE int |
392 | mpd_arith_sign(const mpd_t *dec) |
393 | { |
394 | return 1 - 2 * mpd_isnegative(dec); |
395 | } |
396 | |
397 | /* Radix */ |
398 | ALWAYS_INLINE long |
399 | mpd_radix(void) |
400 | { |
401 | return 10; |
402 | } |
403 | |
404 | /* Dynamic decimal */ |
405 | ALWAYS_INLINE int |
406 | mpd_isdynamic(const mpd_t *dec) |
407 | { |
408 | return !(dec->flags & MPD_STATIC); |
409 | } |
410 | |
411 | /* Static decimal */ |
412 | ALWAYS_INLINE int |
413 | mpd_isstatic(const mpd_t *dec) |
414 | { |
415 | return dec->flags & MPD_STATIC; |
416 | } |
417 | |
418 | /* Data of decimal is dynamic */ |
419 | ALWAYS_INLINE int |
420 | mpd_isdynamic_data(const mpd_t *dec) |
421 | { |
422 | return !(dec->flags & MPD_DATAFLAGS); |
423 | } |
424 | |
425 | /* Data of decimal is static */ |
426 | ALWAYS_INLINE int |
427 | mpd_isstatic_data(const mpd_t *dec) |
428 | { |
429 | return dec->flags & MPD_STATIC_DATA; |
430 | } |
431 | |
432 | /* Data of decimal is shared */ |
433 | ALWAYS_INLINE int |
434 | mpd_isshared_data(const mpd_t *dec) |
435 | { |
436 | return dec->flags & MPD_SHARED_DATA; |
437 | } |
438 | |
439 | /* Data of decimal is const */ |
440 | ALWAYS_INLINE int |
441 | mpd_isconst_data(const mpd_t *dec) |
442 | { |
443 | return dec->flags & MPD_CONST_DATA; |
444 | } |
445 | |
446 | |
447 | /******************************************************************************/ |
448 | /* Inline memory handling */ |
449 | /******************************************************************************/ |
450 | |
451 | /* Fill destination with zeros */ |
452 | ALWAYS_INLINE void |
453 | mpd_uint_zero(mpd_uint_t *dest, mpd_size_t len) |
454 | { |
455 | mpd_size_t i; |
456 | |
457 | for (i = 0; i < len; i++) { |
458 | dest[i] = 0; |
459 | } |
460 | } |
461 | |
462 | /* Free a decimal */ |
463 | ALWAYS_INLINE void |
464 | mpd_del(mpd_t *dec) |
465 | { |
466 | if (mpd_isdynamic_data(dec)) { |
467 | mpd_free(dec->data); |
468 | } |
469 | if (mpd_isdynamic(dec)) { |
470 | mpd_free(dec); |
471 | } |
472 | } |
473 | |
474 | /* |
475 | * Resize the coefficient. Existing data up to 'nwords' is left untouched. |
476 | * Return 1 on success, 0 otherwise. |
477 | * |
478 | * Input invariant: MPD_MINALLOC <= result->alloc. |
479 | * |
480 | * Case nwords == result->alloc: |
481 | * 'result' is unchanged. Return 1. |
482 | * |
483 | * Case nwords > result->alloc: |
484 | * Case realloc success: |
485 | * The value of 'result' does not change. Return 1. |
486 | * Case realloc failure: |
487 | * 'result' is NaN, status is updated with MPD_Malloc_error. Return 0. |
488 | * |
489 | * Case nwords < result->alloc: |
490 | * Case is_static_data or realloc failure [1]: |
491 | * 'result' is unchanged. Return 1. |
492 | * Case realloc success: |
493 | * The value of result is undefined (expected). Return 1. |
494 | * |
495 | * |
496 | * [1] In that case the old (now oversized) area is still valid. |
497 | */ |
498 | ALWAYS_INLINE int |
499 | mpd_qresize(mpd_t *result, mpd_ssize_t nwords, uint32_t *status) |
500 | { |
501 | assert(!mpd_isconst_data(result)); /* illegal operation for a const */ |
502 | assert(!mpd_isshared_data(result)); /* illegal operation for a shared */ |
503 | assert(MPD_MINALLOC <= result->alloc); |
504 | |
505 | nwords = (nwords <= MPD_MINALLOC) ? MPD_MINALLOC : nwords; |
506 | if (nwords == result->alloc) { |
507 | return 1; |
508 | } |
509 | if (mpd_isstatic_data(result)) { |
510 | if (nwords > result->alloc) { |
511 | return mpd_switch_to_dyn(result, nwords, status); |
512 | } |
513 | return 1; |
514 | } |
515 | |
516 | return mpd_realloc_dyn(result, nwords, status); |
517 | } |
518 | |
519 | /* Same as mpd_qresize, but do not set the result no NaN on failure. */ |
520 | static ALWAYS_INLINE int |
521 | mpd_qresize_cxx(mpd_t *result, mpd_ssize_t nwords) |
522 | { |
523 | assert(!mpd_isconst_data(result)); /* illegal operation for a const */ |
524 | assert(!mpd_isshared_data(result)); /* illegal operation for a shared */ |
525 | assert(MPD_MINALLOC <= result->alloc); |
526 | |
527 | nwords = (nwords <= MPD_MINALLOC) ? MPD_MINALLOC : nwords; |
528 | if (nwords == result->alloc) { |
529 | return 1; |
530 | } |
531 | if (mpd_isstatic_data(result)) { |
532 | if (nwords > result->alloc) { |
533 | return mpd_switch_to_dyn_cxx(result, nwords); |
534 | } |
535 | return 1; |
536 | } |
537 | |
538 | return mpd_realloc_dyn_cxx(result, nwords); |
539 | } |
540 | |
541 | /* Same as mpd_qresize, but the complete coefficient (including the old |
542 | * memory area!) is initialized to zero. */ |
543 | ALWAYS_INLINE int |
544 | mpd_qresize_zero(mpd_t *result, mpd_ssize_t nwords, uint32_t *status) |
545 | { |
546 | assert(!mpd_isconst_data(result)); /* illegal operation for a const */ |
547 | assert(!mpd_isshared_data(result)); /* illegal operation for a shared */ |
548 | assert(MPD_MINALLOC <= result->alloc); |
549 | |
550 | nwords = (nwords <= MPD_MINALLOC) ? MPD_MINALLOC : nwords; |
551 | if (nwords != result->alloc) { |
552 | if (mpd_isstatic_data(result)) { |
553 | if (nwords > result->alloc) { |
554 | return mpd_switch_to_dyn_zero(result, nwords, status); |
555 | } |
556 | } |
557 | else if (!mpd_realloc_dyn(result, nwords, status)) { |
558 | return 0; |
559 | } |
560 | } |
561 | |
562 | mpd_uint_zero(result->data, nwords); |
563 | return 1; |
564 | } |
565 | |
566 | /* |
567 | * Reduce memory size for the coefficient to MPD_MINALLOC. In theory, |
568 | * realloc may fail even when reducing the memory size. But in that case |
569 | * the old memory area is always big enough, so checking for MPD_Malloc_error |
570 | * is not imperative. |
571 | */ |
572 | ALWAYS_INLINE void |
573 | mpd_minalloc(mpd_t *result) |
574 | { |
575 | assert(!mpd_isconst_data(result)); /* illegal operation for a const */ |
576 | assert(!mpd_isshared_data(result)); /* illegal operation for a shared */ |
577 | |
578 | if (!mpd_isstatic_data(result) && result->alloc > MPD_MINALLOC) { |
579 | uint8_t err = 0; |
580 | result->data = mpd_realloc(result->data, MPD_MINALLOC, |
581 | sizeof *result->data, &err); |
582 | if (!err) { |
583 | result->alloc = MPD_MINALLOC; |
584 | } |
585 | } |
586 | } |
587 | |
588 | int |
589 | mpd_resize(mpd_t *result, mpd_ssize_t nwords, mpd_context_t *ctx) |
590 | { |
591 | uint32_t status = 0; |
592 | if (!mpd_qresize(result, nwords, &status)) { |
593 | mpd_addstatus_raise(ctx, status); |
594 | return 0; |
595 | } |
596 | return 1; |
597 | } |
598 | |
599 | int |
600 | mpd_resize_zero(mpd_t *result, mpd_ssize_t nwords, mpd_context_t *ctx) |
601 | { |
602 | uint32_t status = 0; |
603 | if (!mpd_qresize_zero(result, nwords, &status)) { |
604 | mpd_addstatus_raise(ctx, status); |
605 | return 0; |
606 | } |
607 | return 1; |
608 | } |
609 | |
610 | |
611 | /******************************************************************************/ |
612 | /* Set attributes of a decimal */ |
613 | /******************************************************************************/ |
614 | |
615 | /* Set digits. Assumption: result->len is initialized and > 0. */ |
616 | inline void |
617 | mpd_setdigits(mpd_t *result) |
618 | { |
619 | mpd_ssize_t wdigits = mpd_word_digits(mpd_msword(result)); |
620 | result->digits = wdigits + (result->len-1) * MPD_RDIGITS; |
621 | } |
622 | |
623 | /* Set sign */ |
624 | ALWAYS_INLINE void |
625 | mpd_set_sign(mpd_t *result, uint8_t sign) |
626 | { |
627 | result->flags &= ~MPD_NEG; |
628 | result->flags |= sign; |
629 | } |
630 | |
631 | /* Copy sign from another decimal */ |
632 | ALWAYS_INLINE void |
633 | mpd_signcpy(mpd_t *result, const mpd_t *a) |
634 | { |
635 | uint8_t sign = a->flags&MPD_NEG; |
636 | |
637 | result->flags &= ~MPD_NEG; |
638 | result->flags |= sign; |
639 | } |
640 | |
641 | /* Set infinity */ |
642 | ALWAYS_INLINE void |
643 | mpd_set_infinity(mpd_t *result) |
644 | { |
645 | result->flags &= ~MPD_SPECIAL; |
646 | result->flags |= MPD_INF; |
647 | } |
648 | |
649 | /* Set qNaN */ |
650 | ALWAYS_INLINE void |
651 | mpd_set_qnan(mpd_t *result) |
652 | { |
653 | result->flags &= ~MPD_SPECIAL; |
654 | result->flags |= MPD_NAN; |
655 | } |
656 | |
657 | /* Set sNaN */ |
658 | ALWAYS_INLINE void |
659 | mpd_set_snan(mpd_t *result) |
660 | { |
661 | result->flags &= ~MPD_SPECIAL; |
662 | result->flags |= MPD_SNAN; |
663 | } |
664 | |
665 | /* Set to negative */ |
666 | ALWAYS_INLINE void |
667 | mpd_set_negative(mpd_t *result) |
668 | { |
669 | result->flags |= MPD_NEG; |
670 | } |
671 | |
672 | /* Set to positive */ |
673 | ALWAYS_INLINE void |
674 | mpd_set_positive(mpd_t *result) |
675 | { |
676 | result->flags &= ~MPD_NEG; |
677 | } |
678 | |
679 | /* Set to dynamic */ |
680 | ALWAYS_INLINE void |
681 | mpd_set_dynamic(mpd_t *result) |
682 | { |
683 | result->flags &= ~MPD_STATIC; |
684 | } |
685 | |
686 | /* Set to static */ |
687 | ALWAYS_INLINE void |
688 | mpd_set_static(mpd_t *result) |
689 | { |
690 | result->flags |= MPD_STATIC; |
691 | } |
692 | |
693 | /* Set data to dynamic */ |
694 | ALWAYS_INLINE void |
695 | mpd_set_dynamic_data(mpd_t *result) |
696 | { |
697 | result->flags &= ~MPD_DATAFLAGS; |
698 | } |
699 | |
700 | /* Set data to static */ |
701 | ALWAYS_INLINE void |
702 | mpd_set_static_data(mpd_t *result) |
703 | { |
704 | result->flags &= ~MPD_DATAFLAGS; |
705 | result->flags |= MPD_STATIC_DATA; |
706 | } |
707 | |
708 | /* Set data to shared */ |
709 | ALWAYS_INLINE void |
710 | mpd_set_shared_data(mpd_t *result) |
711 | { |
712 | result->flags &= ~MPD_DATAFLAGS; |
713 | result->flags |= MPD_SHARED_DATA; |
714 | } |
715 | |
716 | /* Set data to const */ |
717 | ALWAYS_INLINE void |
718 | mpd_set_const_data(mpd_t *result) |
719 | { |
720 | result->flags &= ~MPD_DATAFLAGS; |
721 | result->flags |= MPD_CONST_DATA; |
722 | } |
723 | |
724 | /* Clear flags, preserving memory attributes. */ |
725 | ALWAYS_INLINE void |
726 | mpd_clear_flags(mpd_t *result) |
727 | { |
728 | result->flags &= (MPD_STATIC|MPD_DATAFLAGS); |
729 | } |
730 | |
731 | /* Set flags, preserving memory attributes. */ |
732 | ALWAYS_INLINE void |
733 | mpd_set_flags(mpd_t *result, uint8_t flags) |
734 | { |
735 | result->flags &= (MPD_STATIC|MPD_DATAFLAGS); |
736 | result->flags |= flags; |
737 | } |
738 | |
739 | /* Copy flags, preserving memory attributes of result. */ |
740 | ALWAYS_INLINE void |
741 | mpd_copy_flags(mpd_t *result, const mpd_t *a) |
742 | { |
743 | uint8_t aflags = a->flags; |
744 | result->flags &= (MPD_STATIC|MPD_DATAFLAGS); |
745 | result->flags |= (aflags & ~(MPD_STATIC|MPD_DATAFLAGS)); |
746 | } |
747 | |
748 | /* Initialize a workcontext from ctx. Set traps, flags and newtrap to 0. */ |
749 | static inline void |
750 | mpd_workcontext(mpd_context_t *workctx, const mpd_context_t *ctx) |
751 | { |
752 | workctx->prec = ctx->prec; |
753 | workctx->emax = ctx->emax; |
754 | workctx->emin = ctx->emin; |
755 | workctx->round = ctx->round; |
756 | workctx->traps = 0; |
757 | workctx->status = 0; |
758 | workctx->newtrap = 0; |
759 | workctx->clamp = ctx->clamp; |
760 | workctx->allcr = ctx->allcr; |
761 | } |
762 | |
763 | |
764 | /******************************************************************************/ |
765 | /* Getting and setting parts of decimals */ |
766 | /******************************************************************************/ |
767 | |
768 | /* Flip the sign of a decimal */ |
769 | static inline void |
770 | _mpd_negate(mpd_t *dec) |
771 | { |
772 | dec->flags ^= MPD_NEG; |
773 | } |
774 | |
775 | /* Set coefficient to zero */ |
776 | void |
777 | mpd_zerocoeff(mpd_t *result) |
778 | { |
779 | mpd_minalloc(result); |
780 | result->digits = 1; |
781 | result->len = 1; |
782 | result->data[0] = 0; |
783 | } |
784 | |
785 | /* Set the coefficient to all nines. */ |
786 | void |
787 | mpd_qmaxcoeff(mpd_t *result, const mpd_context_t *ctx, uint32_t *status) |
788 | { |
789 | mpd_ssize_t len, r; |
790 | |
791 | _mpd_idiv_word(&len, &r, ctx->prec, MPD_RDIGITS); |
792 | len = (r == 0) ? len : len+1; |
793 | |
794 | if (!mpd_qresize(result, len, status)) { |
795 | return; |
796 | } |
797 | |
798 | result->len = len; |
799 | result->digits = ctx->prec; |
800 | |
801 | --len; |
802 | if (r > 0) { |
803 | result->data[len--] = mpd_pow10[r]-1; |
804 | } |
805 | for (; len >= 0; --len) { |
806 | result->data[len] = MPD_RADIX-1; |
807 | } |
808 | } |
809 | |
810 | /* |
811 | * Cut off the most significant digits so that the rest fits in ctx->prec. |
812 | * Cannot fail. |
813 | */ |
814 | static void |
815 | _mpd_cap(mpd_t *result, const mpd_context_t *ctx) |
816 | { |
817 | uint32_t dummy; |
818 | mpd_ssize_t len, r; |
819 | |
820 | if (result->len > 0 && result->digits > ctx->prec) { |
821 | _mpd_idiv_word(&len, &r, ctx->prec, MPD_RDIGITS); |
822 | len = (r == 0) ? len : len+1; |
823 | |
824 | if (r != 0) { |
825 | result->data[len-1] %= mpd_pow10[r]; |
826 | } |
827 | |
828 | len = _mpd_real_size(result->data, len); |
829 | /* resize to fewer words cannot fail */ |
830 | mpd_qresize(result, len, &dummy); |
831 | result->len = len; |
832 | mpd_setdigits(result); |
833 | } |
834 | if (mpd_iszero(result)) { |
835 | _settriple(result, mpd_sign(result), 0, result->exp); |
836 | } |
837 | } |
838 | |
839 | /* |
840 | * Cut off the most significant digits of a NaN payload so that the rest |
841 | * fits in ctx->prec - ctx->clamp. Cannot fail. |
842 | */ |
843 | static void |
844 | _mpd_fix_nan(mpd_t *result, const mpd_context_t *ctx) |
845 | { |
846 | uint32_t dummy; |
847 | mpd_ssize_t prec; |
848 | mpd_ssize_t len, r; |
849 | |
850 | prec = ctx->prec - ctx->clamp; |
851 | if (result->len > 0 && result->digits > prec) { |
852 | if (prec == 0) { |
853 | mpd_minalloc(result); |
854 | result->len = result->digits = 0; |
855 | } |
856 | else { |
857 | _mpd_idiv_word(&len, &r, prec, MPD_RDIGITS); |
858 | len = (r == 0) ? len : len+1; |
859 | |
860 | if (r != 0) { |
861 | result->data[len-1] %= mpd_pow10[r]; |
862 | } |
863 | |
864 | len = _mpd_real_size(result->data, len); |
865 | /* resize to fewer words cannot fail */ |
866 | mpd_qresize(result, len, &dummy); |
867 | result->len = len; |
868 | mpd_setdigits(result); |
869 | if (mpd_iszerocoeff(result)) { |
870 | /* NaN0 is not a valid representation */ |
871 | result->len = result->digits = 0; |
872 | } |
873 | } |
874 | } |
875 | } |
876 | |
877 | /* |
878 | * Get n most significant digits from a decimal, where 0 < n <= MPD_UINT_DIGITS. |
879 | * Assumes MPD_UINT_DIGITS == MPD_RDIGITS+1, which is true for 32 and 64 bit |
880 | * machines. |
881 | * |
882 | * The result of the operation will be in lo. If the operation is impossible, |
883 | * hi will be nonzero. This is used to indicate an error. |
884 | */ |
885 | static inline void |
886 | _mpd_get_msdigits(mpd_uint_t *hi, mpd_uint_t *lo, const mpd_t *dec, |
887 | unsigned int n) |
888 | { |
889 | mpd_uint_t r, tmp; |
890 | |
891 | assert(0 < n && n <= MPD_RDIGITS+1); |
892 | |
893 | _mpd_div_word(&tmp, &r, dec->digits, MPD_RDIGITS); |
894 | r = (r == 0) ? MPD_RDIGITS : r; /* digits in the most significant word */ |
895 | |
896 | *hi = 0; |
897 | *lo = dec->data[dec->len-1]; |
898 | if (n <= r) { |
899 | *lo /= mpd_pow10[r-n]; |
900 | } |
901 | else if (dec->len > 1) { |
902 | /* at this point 1 <= r < n <= MPD_RDIGITS+1 */ |
903 | _mpd_mul_words(hi, lo, *lo, mpd_pow10[n-r]); |
904 | tmp = dec->data[dec->len-2] / mpd_pow10[MPD_RDIGITS-(n-r)]; |
905 | *lo = *lo + tmp; |
906 | if (*lo < tmp) (*hi)++; |
907 | } |
908 | } |
909 | |
910 | |
911 | /******************************************************************************/ |
912 | /* Gathering information about a decimal */ |
913 | /******************************************************************************/ |
914 | |
915 | /* The real size of the coefficient without leading zero words. */ |
916 | static inline mpd_ssize_t |
917 | _mpd_real_size(mpd_uint_t *data, mpd_ssize_t size) |
918 | { |
919 | while (size > 1 && data[size-1] == 0) { |
920 | size--; |
921 | } |
922 | |
923 | return size; |
924 | } |
925 | |
926 | /* Return number of trailing zeros. No errors are possible. */ |
927 | mpd_ssize_t |
928 | mpd_trail_zeros(const mpd_t *dec) |
929 | { |
930 | mpd_uint_t word; |
931 | mpd_ssize_t i, tz = 0; |
932 | |
933 | for (i=0; i < dec->len; ++i) { |
934 | if (dec->data[i] != 0) { |
935 | word = dec->data[i]; |
936 | tz = i * MPD_RDIGITS; |
937 | while (word % 10 == 0) { |
938 | word /= 10; |
939 | tz++; |
940 | } |
941 | break; |
942 | } |
943 | } |
944 | |
945 | return tz; |
946 | } |
947 | |
948 | /* Integer: Undefined for specials */ |
949 | static int |
950 | _mpd_isint(const mpd_t *dec) |
951 | { |
952 | mpd_ssize_t tz; |
953 | |
954 | if (mpd_iszerocoeff(dec)) { |
955 | return 1; |
956 | } |
957 | |
958 | tz = mpd_trail_zeros(dec); |
959 | return (dec->exp + tz >= 0); |
960 | } |
961 | |
962 | /* Integer */ |
963 | int |
964 | mpd_isinteger(const mpd_t *dec) |
965 | { |
966 | if (mpd_isspecial(dec)) { |
967 | return 0; |
968 | } |
969 | return _mpd_isint(dec); |
970 | } |
971 | |
972 | /* Word is a power of 10 */ |
973 | static int |
974 | mpd_word_ispow10(mpd_uint_t word) |
975 | { |
976 | int n; |
977 | |
978 | n = mpd_word_digits(word); |
979 | if (word == mpd_pow10[n-1]) { |
980 | return 1; |
981 | } |
982 | |
983 | return 0; |
984 | } |
985 | |
986 | /* Coefficient is a power of 10 */ |
987 | static int |
988 | mpd_coeff_ispow10(const mpd_t *dec) |
989 | { |
990 | if (mpd_word_ispow10(mpd_msword(dec))) { |
991 | if (_mpd_isallzero(dec->data, dec->len-1)) { |
992 | return 1; |
993 | } |
994 | } |
995 | |
996 | return 0; |
997 | } |
998 | |
999 | /* All digits of a word are nines */ |
1000 | static int |
1001 | mpd_word_isallnine(mpd_uint_t word) |
1002 | { |
1003 | int n; |
1004 | |
1005 | n = mpd_word_digits(word); |
1006 | if (word == mpd_pow10[n]-1) { |
1007 | return 1; |
1008 | } |
1009 | |
1010 | return 0; |
1011 | } |
1012 | |
1013 | /* All digits of the coefficient are nines */ |
1014 | static int |
1015 | mpd_coeff_isallnine(const mpd_t *dec) |
1016 | { |
1017 | if (mpd_word_isallnine(mpd_msword(dec))) { |
1018 | if (_mpd_isallnine(dec->data, dec->len-1)) { |
1019 | return 1; |
1020 | } |
1021 | } |
1022 | |
1023 | return 0; |
1024 | } |
1025 | |
1026 | /* Odd decimal: Undefined for non-integers! */ |
1027 | int |
1028 | mpd_isodd(const mpd_t *dec) |
1029 | { |
1030 | mpd_uint_t q, r; |
1031 | assert(mpd_isinteger(dec)); |
1032 | if (mpd_iszerocoeff(dec)) return 0; |
1033 | if (dec->exp < 0) { |
1034 | _mpd_div_word(&q, &r, -dec->exp, MPD_RDIGITS); |
1035 | q = dec->data[q] / mpd_pow10[r]; |
1036 | return mpd_isoddword(q); |
1037 | } |
1038 | return dec->exp == 0 && mpd_isoddword(dec->data[0]); |
1039 | } |
1040 | |
1041 | /* Even: Undefined for non-integers! */ |
1042 | int |
1043 | mpd_iseven(const mpd_t *dec) |
1044 | { |
1045 | return !mpd_isodd(dec); |
1046 | } |
1047 | |
1048 | /******************************************************************************/ |
1049 | /* Getting and setting decimals */ |
1050 | /******************************************************************************/ |
1051 | |
1052 | /* Internal function: Set a static decimal from a triple, no error checking. */ |
1053 | static void |
1054 | _ssettriple(mpd_t *result, uint8_t sign, mpd_uint_t a, mpd_ssize_t exp) |
1055 | { |
1056 | mpd_set_flags(result, sign); |
1057 | result->exp = exp; |
1058 | _mpd_div_word(&result->data[1], &result->data[0], a, MPD_RADIX); |
1059 | result->len = (result->data[1] == 0) ? 1 : 2; |
1060 | mpd_setdigits(result); |
1061 | } |
1062 | |
1063 | /* Internal function: Set a decimal from a triple, no error checking. */ |
1064 | static void |
1065 | _settriple(mpd_t *result, uint8_t sign, mpd_uint_t a, mpd_ssize_t exp) |
1066 | { |
1067 | mpd_minalloc(result); |
1068 | mpd_set_flags(result, sign); |
1069 | result->exp = exp; |
1070 | _mpd_div_word(&result->data[1], &result->data[0], a, MPD_RADIX); |
1071 | result->len = (result->data[1] == 0) ? 1 : 2; |
1072 | mpd_setdigits(result); |
1073 | } |
1074 | |
1075 | /* Set a special number from a triple */ |
1076 | void |
1077 | mpd_setspecial(mpd_t *result, uint8_t sign, uint8_t type) |
1078 | { |
1079 | mpd_minalloc(result); |
1080 | result->flags &= ~(MPD_NEG|MPD_SPECIAL); |
1081 | result->flags |= (sign|type); |
1082 | result->exp = result->digits = result->len = 0; |
1083 | } |
1084 | |
1085 | /* Set result of NaN with an error status */ |
1086 | void |
1087 | mpd_seterror(mpd_t *result, uint32_t flags, uint32_t *status) |
1088 | { |
1089 | mpd_minalloc(result); |
1090 | mpd_set_qnan(result); |
1091 | mpd_set_positive(result); |
1092 | result->exp = result->digits = result->len = 0; |
1093 | *status |= flags; |
1094 | } |
1095 | |
1096 | /* quietly set a static decimal from an mpd_ssize_t */ |
1097 | void |
1098 | mpd_qsset_ssize(mpd_t *result, mpd_ssize_t a, const mpd_context_t *ctx, |
1099 | uint32_t *status) |
1100 | { |
1101 | mpd_uint_t u; |
1102 | uint8_t sign = MPD_POS; |
1103 | |
1104 | if (a < 0) { |
1105 | if (a == MPD_SSIZE_MIN) { |
1106 | u = (mpd_uint_t)MPD_SSIZE_MAX + |
1107 | (-(MPD_SSIZE_MIN+MPD_SSIZE_MAX)); |
1108 | } |
1109 | else { |
1110 | u = -a; |
1111 | } |
1112 | sign = MPD_NEG; |
1113 | } |
1114 | else { |
1115 | u = a; |
1116 | } |
1117 | _ssettriple(result, sign, u, 0); |
1118 | mpd_qfinalize(result, ctx, status); |
1119 | } |
1120 | |
1121 | /* quietly set a static decimal from an mpd_uint_t */ |
1122 | void |
1123 | mpd_qsset_uint(mpd_t *result, mpd_uint_t a, const mpd_context_t *ctx, |
1124 | uint32_t *status) |
1125 | { |
1126 | _ssettriple(result, MPD_POS, a, 0); |
1127 | mpd_qfinalize(result, ctx, status); |
1128 | } |
1129 | |
1130 | /* quietly set a static decimal from an int32_t */ |
1131 | void |
1132 | mpd_qsset_i32(mpd_t *result, int32_t a, const mpd_context_t *ctx, |
1133 | uint32_t *status) |
1134 | { |
1135 | mpd_qsset_ssize(result, a, ctx, status); |
1136 | } |
1137 | |
1138 | /* quietly set a static decimal from a uint32_t */ |
1139 | void |
1140 | mpd_qsset_u32(mpd_t *result, uint32_t a, const mpd_context_t *ctx, |
1141 | uint32_t *status) |
1142 | { |
1143 | mpd_qsset_uint(result, a, ctx, status); |
1144 | } |
1145 | |
1146 | #ifdef CONFIG_64 |
1147 | /* quietly set a static decimal from an int64_t */ |
1148 | void |
1149 | mpd_qsset_i64(mpd_t *result, int64_t a, const mpd_context_t *ctx, |
1150 | uint32_t *status) |
1151 | { |
1152 | mpd_qsset_ssize(result, a, ctx, status); |
1153 | } |
1154 | |
1155 | /* quietly set a static decimal from a uint64_t */ |
1156 | void |
1157 | mpd_qsset_u64(mpd_t *result, uint64_t a, const mpd_context_t *ctx, |
1158 | uint32_t *status) |
1159 | { |
1160 | mpd_qsset_uint(result, a, ctx, status); |
1161 | } |
1162 | #endif |
1163 | |
1164 | /* quietly set a decimal from an mpd_ssize_t */ |
1165 | void |
1166 | mpd_qset_ssize(mpd_t *result, mpd_ssize_t a, const mpd_context_t *ctx, |
1167 | uint32_t *status) |
1168 | { |
1169 | mpd_minalloc(result); |
1170 | mpd_qsset_ssize(result, a, ctx, status); |
1171 | } |
1172 | |
1173 | /* quietly set a decimal from an mpd_uint_t */ |
1174 | void |
1175 | mpd_qset_uint(mpd_t *result, mpd_uint_t a, const mpd_context_t *ctx, |
1176 | uint32_t *status) |
1177 | { |
1178 | _settriple(result, MPD_POS, a, 0); |
1179 | mpd_qfinalize(result, ctx, status); |
1180 | } |
1181 | |
1182 | /* quietly set a decimal from an int32_t */ |
1183 | void |
1184 | mpd_qset_i32(mpd_t *result, int32_t a, const mpd_context_t *ctx, |
1185 | uint32_t *status) |
1186 | { |
1187 | mpd_qset_ssize(result, a, ctx, status); |
1188 | } |
1189 | |
1190 | /* quietly set a decimal from a uint32_t */ |
1191 | void |
1192 | mpd_qset_u32(mpd_t *result, uint32_t a, const mpd_context_t *ctx, |
1193 | uint32_t *status) |
1194 | { |
1195 | mpd_qset_uint(result, a, ctx, status); |
1196 | } |
1197 | |
1198 | #if defined(CONFIG_32) && !defined(LEGACY_COMPILER) |
1199 | /* set a decimal from a uint64_t */ |
1200 | static void |
1201 | _c32setu64(mpd_t *result, uint64_t u, uint8_t sign, uint32_t *status) |
1202 | { |
1203 | mpd_uint_t w[3]; |
1204 | uint64_t q; |
1205 | int i, len; |
1206 | |
1207 | len = 0; |
1208 | do { |
1209 | q = u / MPD_RADIX; |
1210 | w[len] = (mpd_uint_t)(u - q * MPD_RADIX); |
1211 | u = q; len++; |
1212 | } while (u != 0); |
1213 | |
1214 | if (!mpd_qresize(result, len, status)) { |
1215 | return; |
1216 | } |
1217 | for (i = 0; i < len; i++) { |
1218 | result->data[i] = w[i]; |
1219 | } |
1220 | |
1221 | mpd_set_flags(result, sign); |
1222 | result->exp = 0; |
1223 | result->len = len; |
1224 | mpd_setdigits(result); |
1225 | } |
1226 | |
1227 | static void |
1228 | _c32_qset_u64(mpd_t *result, uint64_t a, const mpd_context_t *ctx, |
1229 | uint32_t *status) |
1230 | { |
1231 | _c32setu64(result, a, MPD_POS, status); |
1232 | mpd_qfinalize(result, ctx, status); |
1233 | } |
1234 | |
1235 | /* set a decimal from an int64_t */ |
1236 | static void |
1237 | _c32_qset_i64(mpd_t *result, int64_t a, const mpd_context_t *ctx, |
1238 | uint32_t *status) |
1239 | { |
1240 | uint64_t u; |
1241 | uint8_t sign = MPD_POS; |
1242 | |
1243 | if (a < 0) { |
1244 | if (a == INT64_MIN) { |
1245 | u = (uint64_t)INT64_MAX + (-(INT64_MIN+INT64_MAX)); |
1246 | } |
1247 | else { |
1248 | u = -a; |
1249 | } |
1250 | sign = MPD_NEG; |
1251 | } |
1252 | else { |
1253 | u = a; |
1254 | } |
1255 | _c32setu64(result, u, sign, status); |
1256 | mpd_qfinalize(result, ctx, status); |
1257 | } |
1258 | #endif /* CONFIG_32 && !LEGACY_COMPILER */ |
1259 | |
1260 | #ifndef LEGACY_COMPILER |
1261 | /* quietly set a decimal from an int64_t */ |
1262 | void |
1263 | mpd_qset_i64(mpd_t *result, int64_t a, const mpd_context_t *ctx, |
1264 | uint32_t *status) |
1265 | { |
1266 | #ifdef CONFIG_64 |
1267 | mpd_qset_ssize(result, a, ctx, status); |
1268 | #else |
1269 | _c32_qset_i64(result, a, ctx, status); |
1270 | #endif |
1271 | } |
1272 | |
1273 | /* quietly set a decimal from an int64_t, use a maxcontext for conversion */ |
1274 | void |
1275 | mpd_qset_i64_exact(mpd_t *result, int64_t a, uint32_t *status) |
1276 | { |
1277 | mpd_context_t maxcontext; |
1278 | |
1279 | mpd_maxcontext(&maxcontext); |
1280 | #ifdef CONFIG_64 |
1281 | mpd_qset_ssize(result, a, &maxcontext, status); |
1282 | #else |
1283 | _c32_qset_i64(result, a, &maxcontext, status); |
1284 | #endif |
1285 | |
1286 | if (*status & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
1287 | /* we want exact results */ |
1288 | mpd_seterror(result, MPD_Invalid_operation, status); |
1289 | } |
1290 | *status &= MPD_Errors; |
1291 | } |
1292 | |
1293 | /* quietly set a decimal from a uint64_t */ |
1294 | void |
1295 | mpd_qset_u64(mpd_t *result, uint64_t a, const mpd_context_t *ctx, |
1296 | uint32_t *status) |
1297 | { |
1298 | #ifdef CONFIG_64 |
1299 | mpd_qset_uint(result, a, ctx, status); |
1300 | #else |
1301 | _c32_qset_u64(result, a, ctx, status); |
1302 | #endif |
1303 | } |
1304 | |
1305 | /* quietly set a decimal from a uint64_t, use a maxcontext for conversion */ |
1306 | void |
1307 | mpd_qset_u64_exact(mpd_t *result, uint64_t a, uint32_t *status) |
1308 | { |
1309 | mpd_context_t maxcontext; |
1310 | |
1311 | mpd_maxcontext(&maxcontext); |
1312 | #ifdef CONFIG_64 |
1313 | mpd_qset_uint(result, a, &maxcontext, status); |
1314 | #else |
1315 | _c32_qset_u64(result, a, &maxcontext, status); |
1316 | #endif |
1317 | |
1318 | if (*status & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
1319 | /* we want exact results */ |
1320 | mpd_seterror(result, MPD_Invalid_operation, status); |
1321 | } |
1322 | *status &= MPD_Errors; |
1323 | } |
1324 | #endif /* !LEGACY_COMPILER */ |
1325 | |
1326 | /* |
1327 | * Quietly get an mpd_uint_t from a decimal. Assumes |
1328 | * MPD_UINT_DIGITS == MPD_RDIGITS+1, which is true for |
1329 | * 32 and 64 bit machines. |
1330 | * |
1331 | * If the operation is impossible, MPD_Invalid_operation is set. |
1332 | */ |
1333 | static mpd_uint_t |
1334 | _mpd_qget_uint(int use_sign, const mpd_t *a, uint32_t *status) |
1335 | { |
1336 | mpd_t tmp; |
1337 | mpd_uint_t tmp_data[2]; |
1338 | mpd_uint_t lo, hi; |
1339 | |
1340 | if (mpd_isspecial(a)) { |
1341 | *status |= MPD_Invalid_operation; |
1342 | return MPD_UINT_MAX; |
1343 | } |
1344 | if (mpd_iszero(a)) { |
1345 | return 0; |
1346 | } |
1347 | if (use_sign && mpd_isnegative(a)) { |
1348 | *status |= MPD_Invalid_operation; |
1349 | return MPD_UINT_MAX; |
1350 | } |
1351 | |
1352 | if (a->digits+a->exp > MPD_RDIGITS+1) { |
1353 | *status |= MPD_Invalid_operation; |
1354 | return MPD_UINT_MAX; |
1355 | } |
1356 | |
1357 | if (a->exp < 0) { |
1358 | if (!_mpd_isint(a)) { |
1359 | *status |= MPD_Invalid_operation; |
1360 | return MPD_UINT_MAX; |
1361 | } |
1362 | /* At this point a->digits+a->exp <= MPD_RDIGITS+1, |
1363 | * so the shift fits. */ |
1364 | tmp.data = tmp_data; |
1365 | tmp.flags = MPD_STATIC|MPD_STATIC_DATA; |
1366 | tmp.alloc = 2; |
1367 | mpd_qsshiftr(&tmp, a, -a->exp); |
1368 | tmp.exp = 0; |
1369 | a = &tmp; |
1370 | } |
1371 | |
1372 | _mpd_get_msdigits(&hi, &lo, a, MPD_RDIGITS+1); |
1373 | if (hi) { |
1374 | *status |= MPD_Invalid_operation; |
1375 | return MPD_UINT_MAX; |
1376 | } |
1377 | |
1378 | if (a->exp > 0) { |
1379 | _mpd_mul_words(&hi, &lo, lo, mpd_pow10[a->exp]); |
1380 | if (hi) { |
1381 | *status |= MPD_Invalid_operation; |
1382 | return MPD_UINT_MAX; |
1383 | } |
1384 | } |
1385 | |
1386 | return lo; |
1387 | } |
1388 | |
1389 | /* |
1390 | * Sets Invalid_operation for: |
1391 | * - specials |
1392 | * - negative numbers (except negative zero) |
1393 | * - non-integers |
1394 | * - overflow |
1395 | */ |
1396 | mpd_uint_t |
1397 | mpd_qget_uint(const mpd_t *a, uint32_t *status) |
1398 | { |
1399 | return _mpd_qget_uint(1, a, status); |
1400 | } |
1401 | |
1402 | /* Same as above, but gets the absolute value, i.e. the sign is ignored. */ |
1403 | mpd_uint_t |
1404 | mpd_qabs_uint(const mpd_t *a, uint32_t *status) |
1405 | { |
1406 | return _mpd_qget_uint(0, a, status); |
1407 | } |
1408 | |
1409 | /* quietly get an mpd_ssize_t from a decimal */ |
1410 | mpd_ssize_t |
1411 | mpd_qget_ssize(const mpd_t *a, uint32_t *status) |
1412 | { |
1413 | uint32_t workstatus = 0; |
1414 | mpd_uint_t u; |
1415 | int isneg; |
1416 | |
1417 | u = mpd_qabs_uint(a, &workstatus); |
1418 | if (workstatus&MPD_Invalid_operation) { |
1419 | *status |= workstatus; |
1420 | return MPD_SSIZE_MAX; |
1421 | } |
1422 | |
1423 | isneg = mpd_isnegative(a); |
1424 | if (u <= MPD_SSIZE_MAX) { |
1425 | return isneg ? -((mpd_ssize_t)u) : (mpd_ssize_t)u; |
1426 | } |
1427 | else if (isneg && u+(MPD_SSIZE_MIN+MPD_SSIZE_MAX) == MPD_SSIZE_MAX) { |
1428 | return MPD_SSIZE_MIN; |
1429 | } |
1430 | |
1431 | *status |= MPD_Invalid_operation; |
1432 | return MPD_SSIZE_MAX; |
1433 | } |
1434 | |
1435 | #if defined(CONFIG_32) && !defined(LEGACY_COMPILER) |
1436 | /* |
1437 | * Quietly get a uint64_t from a decimal. If the operation is impossible, |
1438 | * MPD_Invalid_operation is set. |
1439 | */ |
1440 | static uint64_t |
1441 | _c32_qget_u64(int use_sign, const mpd_t *a, uint32_t *status) |
1442 | { |
1443 | MPD_NEW_STATIC(tmp,0,0,20,3); |
1444 | mpd_context_t maxcontext; |
1445 | uint64_t ret; |
1446 | |
1447 | tmp_data[0] = 709551615; |
1448 | tmp_data[1] = 446744073; |
1449 | tmp_data[2] = 18; |
1450 | |
1451 | if (mpd_isspecial(a)) { |
1452 | *status |= MPD_Invalid_operation; |
1453 | return UINT64_MAX; |
1454 | } |
1455 | if (mpd_iszero(a)) { |
1456 | return 0; |
1457 | } |
1458 | if (use_sign && mpd_isnegative(a)) { |
1459 | *status |= MPD_Invalid_operation; |
1460 | return UINT64_MAX; |
1461 | } |
1462 | if (!_mpd_isint(a)) { |
1463 | *status |= MPD_Invalid_operation; |
1464 | return UINT64_MAX; |
1465 | } |
1466 | |
1467 | if (_mpd_cmp_abs(a, &tmp) > 0) { |
1468 | *status |= MPD_Invalid_operation; |
1469 | return UINT64_MAX; |
1470 | } |
1471 | |
1472 | mpd_maxcontext(&maxcontext); |
1473 | mpd_qrescale(&tmp, a, 0, &maxcontext, &maxcontext.status); |
1474 | maxcontext.status &= ~MPD_Rounded; |
1475 | if (maxcontext.status != 0) { |
1476 | *status |= (maxcontext.status|MPD_Invalid_operation); /* GCOV_NOT_REACHED */ |
1477 | return UINT64_MAX; /* GCOV_NOT_REACHED */ |
1478 | } |
1479 | |
1480 | ret = 0; |
1481 | switch (tmp.len) { |
1482 | case 3: |
1483 | ret += (uint64_t)tmp_data[2] * 1000000000000000000ULL; |
1484 | case 2: |
1485 | ret += (uint64_t)tmp_data[1] * 1000000000ULL; |
1486 | case 1: |
1487 | ret += tmp_data[0]; |
1488 | break; |
1489 | default: |
1490 | abort(); /* GCOV_NOT_REACHED */ |
1491 | } |
1492 | |
1493 | return ret; |
1494 | } |
1495 | |
1496 | static int64_t |
1497 | _c32_qget_i64(const mpd_t *a, uint32_t *status) |
1498 | { |
1499 | uint64_t u; |
1500 | int isneg; |
1501 | |
1502 | u = _c32_qget_u64(0, a, status); |
1503 | if (*status&MPD_Invalid_operation) { |
1504 | return INT64_MAX; |
1505 | } |
1506 | |
1507 | isneg = mpd_isnegative(a); |
1508 | if (u <= INT64_MAX) { |
1509 | return isneg ? -((int64_t)u) : (int64_t)u; |
1510 | } |
1511 | else if (isneg && u+(INT64_MIN+INT64_MAX) == INT64_MAX) { |
1512 | return INT64_MIN; |
1513 | } |
1514 | |
1515 | *status |= MPD_Invalid_operation; |
1516 | return INT64_MAX; |
1517 | } |
1518 | #endif /* CONFIG_32 && !LEGACY_COMPILER */ |
1519 | |
1520 | #ifdef CONFIG_64 |
1521 | /* quietly get a uint64_t from a decimal */ |
1522 | uint64_t |
1523 | mpd_qget_u64(const mpd_t *a, uint32_t *status) |
1524 | { |
1525 | return mpd_qget_uint(a, status); |
1526 | } |
1527 | |
1528 | /* quietly get an int64_t from a decimal */ |
1529 | int64_t |
1530 | mpd_qget_i64(const mpd_t *a, uint32_t *status) |
1531 | { |
1532 | return mpd_qget_ssize(a, status); |
1533 | } |
1534 | |
1535 | /* quietly get a uint32_t from a decimal */ |
1536 | uint32_t |
1537 | mpd_qget_u32(const mpd_t *a, uint32_t *status) |
1538 | { |
1539 | uint32_t workstatus = 0; |
1540 | uint64_t x = mpd_qget_uint(a, &workstatus); |
1541 | |
1542 | if (workstatus&MPD_Invalid_operation) { |
1543 | *status |= workstatus; |
1544 | return UINT32_MAX; |
1545 | } |
1546 | if (x > UINT32_MAX) { |
1547 | *status |= MPD_Invalid_operation; |
1548 | return UINT32_MAX; |
1549 | } |
1550 | |
1551 | return (uint32_t)x; |
1552 | } |
1553 | |
1554 | /* quietly get an int32_t from a decimal */ |
1555 | int32_t |
1556 | mpd_qget_i32(const mpd_t *a, uint32_t *status) |
1557 | { |
1558 | uint32_t workstatus = 0; |
1559 | int64_t x = mpd_qget_ssize(a, &workstatus); |
1560 | |
1561 | if (workstatus&MPD_Invalid_operation) { |
1562 | *status |= workstatus; |
1563 | return INT32_MAX; |
1564 | } |
1565 | if (x < INT32_MIN || x > INT32_MAX) { |
1566 | *status |= MPD_Invalid_operation; |
1567 | return INT32_MAX; |
1568 | } |
1569 | |
1570 | return (int32_t)x; |
1571 | } |
1572 | #else |
1573 | #ifndef LEGACY_COMPILER |
1574 | /* quietly get a uint64_t from a decimal */ |
1575 | uint64_t |
1576 | mpd_qget_u64(const mpd_t *a, uint32_t *status) |
1577 | { |
1578 | uint32_t workstatus = 0; |
1579 | uint64_t x = _c32_qget_u64(1, a, &workstatus); |
1580 | *status |= workstatus; |
1581 | return x; |
1582 | } |
1583 | |
1584 | /* quietly get an int64_t from a decimal */ |
1585 | int64_t |
1586 | mpd_qget_i64(const mpd_t *a, uint32_t *status) |
1587 | { |
1588 | uint32_t workstatus = 0; |
1589 | int64_t x = _c32_qget_i64(a, &workstatus); |
1590 | *status |= workstatus; |
1591 | return x; |
1592 | } |
1593 | #endif |
1594 | |
1595 | /* quietly get a uint32_t from a decimal */ |
1596 | uint32_t |
1597 | mpd_qget_u32(const mpd_t *a, uint32_t *status) |
1598 | { |
1599 | return mpd_qget_uint(a, status); |
1600 | } |
1601 | |
1602 | /* quietly get an int32_t from a decimal */ |
1603 | int32_t |
1604 | mpd_qget_i32(const mpd_t *a, uint32_t *status) |
1605 | { |
1606 | return mpd_qget_ssize(a, status); |
1607 | } |
1608 | #endif |
1609 | |
1610 | |
1611 | /******************************************************************************/ |
1612 | /* Filtering input of functions, finalizing output of functions */ |
1613 | /******************************************************************************/ |
1614 | |
1615 | /* |
1616 | * Check if the operand is NaN, copy to result and return 1 if this is |
1617 | * the case. Copying can fail since NaNs are allowed to have a payload that |
1618 | * does not fit in MPD_MINALLOC. |
1619 | */ |
1620 | int |
1621 | mpd_qcheck_nan(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
1622 | uint32_t *status) |
1623 | { |
1624 | if (mpd_isnan(a)) { |
1625 | *status |= mpd_issnan(a) ? MPD_Invalid_operation : 0; |
1626 | mpd_qcopy(result, a, status); |
1627 | mpd_set_qnan(result); |
1628 | _mpd_fix_nan(result, ctx); |
1629 | return 1; |
1630 | } |
1631 | return 0; |
1632 | } |
1633 | |
1634 | /* |
1635 | * Check if either operand is NaN, copy to result and return 1 if this |
1636 | * is the case. Copying can fail since NaNs are allowed to have a payload |
1637 | * that does not fit in MPD_MINALLOC. |
1638 | */ |
1639 | int |
1640 | mpd_qcheck_nans(mpd_t *result, const mpd_t *a, const mpd_t *b, |
1641 | const mpd_context_t *ctx, uint32_t *status) |
1642 | { |
1643 | if ((a->flags|b->flags)&(MPD_NAN|MPD_SNAN)) { |
1644 | const mpd_t *choice = b; |
1645 | if (mpd_issnan(a)) { |
1646 | choice = a; |
1647 | *status |= MPD_Invalid_operation; |
1648 | } |
1649 | else if (mpd_issnan(b)) { |
1650 | *status |= MPD_Invalid_operation; |
1651 | } |
1652 | else if (mpd_isqnan(a)) { |
1653 | choice = a; |
1654 | } |
1655 | mpd_qcopy(result, choice, status); |
1656 | mpd_set_qnan(result); |
1657 | _mpd_fix_nan(result, ctx); |
1658 | return 1; |
1659 | } |
1660 | return 0; |
1661 | } |
1662 | |
1663 | /* |
1664 | * Check if one of the operands is NaN, copy to result and return 1 if this |
1665 | * is the case. Copying can fail since NaNs are allowed to have a payload |
1666 | * that does not fit in MPD_MINALLOC. |
1667 | */ |
1668 | static int |
1669 | mpd_qcheck_3nans(mpd_t *result, const mpd_t *a, const mpd_t *b, const mpd_t *c, |
1670 | const mpd_context_t *ctx, uint32_t *status) |
1671 | { |
1672 | if ((a->flags|b->flags|c->flags)&(MPD_NAN|MPD_SNAN)) { |
1673 | const mpd_t *choice = c; |
1674 | if (mpd_issnan(a)) { |
1675 | choice = a; |
1676 | *status |= MPD_Invalid_operation; |
1677 | } |
1678 | else if (mpd_issnan(b)) { |
1679 | choice = b; |
1680 | *status |= MPD_Invalid_operation; |
1681 | } |
1682 | else if (mpd_issnan(c)) { |
1683 | *status |= MPD_Invalid_operation; |
1684 | } |
1685 | else if (mpd_isqnan(a)) { |
1686 | choice = a; |
1687 | } |
1688 | else if (mpd_isqnan(b)) { |
1689 | choice = b; |
1690 | } |
1691 | mpd_qcopy(result, choice, status); |
1692 | mpd_set_qnan(result); |
1693 | _mpd_fix_nan(result, ctx); |
1694 | return 1; |
1695 | } |
1696 | return 0; |
1697 | } |
1698 | |
1699 | /* Check if rounding digit 'rnd' leads to an increment. */ |
1700 | static inline int |
1701 | _mpd_rnd_incr(const mpd_t *dec, mpd_uint_t rnd, const mpd_context_t *ctx) |
1702 | { |
1703 | int ld; |
1704 | |
1705 | switch (ctx->round) { |
1706 | case MPD_ROUND_DOWN: case MPD_ROUND_TRUNC: |
1707 | return 0; |
1708 | case MPD_ROUND_HALF_UP: |
1709 | return (rnd >= 5); |
1710 | case MPD_ROUND_HALF_EVEN: |
1711 | return (rnd > 5) || ((rnd == 5) && mpd_isoddcoeff(dec)); |
1712 | case MPD_ROUND_CEILING: |
1713 | return !(rnd == 0 || mpd_isnegative(dec)); |
1714 | case MPD_ROUND_FLOOR: |
1715 | return !(rnd == 0 || mpd_ispositive(dec)); |
1716 | case MPD_ROUND_HALF_DOWN: |
1717 | return (rnd > 5); |
1718 | case MPD_ROUND_UP: |
1719 | return !(rnd == 0); |
1720 | case MPD_ROUND_05UP: |
1721 | ld = (int)mpd_lsd(dec->data[0]); |
1722 | return (!(rnd == 0) && (ld == 0 || ld == 5)); |
1723 | default: |
1724 | /* Without a valid context, further results will be undefined. */ |
1725 | return 0; /* GCOV_NOT_REACHED */ |
1726 | } |
1727 | } |
1728 | |
1729 | /* |
1730 | * Apply rounding to a decimal that has been right-shifted into a full |
1731 | * precision decimal. If an increment leads to an overflow of the precision, |
1732 | * adjust the coefficient and the exponent and check the new exponent for |
1733 | * overflow. |
1734 | */ |
1735 | static inline void |
1736 | _mpd_apply_round(mpd_t *dec, mpd_uint_t rnd, const mpd_context_t *ctx, |
1737 | uint32_t *status) |
1738 | { |
1739 | if (_mpd_rnd_incr(dec, rnd, ctx)) { |
1740 | /* We have a number with exactly ctx->prec digits. The increment |
1741 | * can only lead to an overflow if the decimal is all nines. In |
1742 | * that case, the result is a power of ten with prec+1 digits. |
1743 | * |
1744 | * If the precision is a multiple of MPD_RDIGITS, this situation is |
1745 | * detected by _mpd_baseincr returning a carry. |
1746 | * If the precision is not a multiple of MPD_RDIGITS, we have to |
1747 | * check if the result has one digit too many. |
1748 | */ |
1749 | mpd_uint_t carry = _mpd_baseincr(dec->data, dec->len); |
1750 | if (carry) { |
1751 | dec->data[dec->len-1] = mpd_pow10[MPD_RDIGITS-1]; |
1752 | dec->exp += 1; |
1753 | _mpd_check_exp(dec, ctx, status); |
1754 | return; |
1755 | } |
1756 | mpd_setdigits(dec); |
1757 | if (dec->digits > ctx->prec) { |
1758 | mpd_qshiftr_inplace(dec, 1); |
1759 | dec->exp += 1; |
1760 | dec->digits = ctx->prec; |
1761 | _mpd_check_exp(dec, ctx, status); |
1762 | } |
1763 | } |
1764 | } |
1765 | |
1766 | /* |
1767 | * Apply rounding to a decimal. Allow overflow of the precision. |
1768 | */ |
1769 | static inline void |
1770 | _mpd_apply_round_excess(mpd_t *dec, mpd_uint_t rnd, const mpd_context_t *ctx, |
1771 | uint32_t *status) |
1772 | { |
1773 | if (_mpd_rnd_incr(dec, rnd, ctx)) { |
1774 | mpd_uint_t carry = _mpd_baseincr(dec->data, dec->len); |
1775 | if (carry) { |
1776 | if (!mpd_qresize(dec, dec->len+1, status)) { |
1777 | return; |
1778 | } |
1779 | dec->data[dec->len] = 1; |
1780 | dec->len += 1; |
1781 | } |
1782 | mpd_setdigits(dec); |
1783 | } |
1784 | } |
1785 | |
1786 | /* |
1787 | * Apply rounding to a decimal that has been right-shifted into a decimal |
1788 | * with full precision or less. Return failure if an increment would |
1789 | * overflow the precision. |
1790 | */ |
1791 | static inline int |
1792 | _mpd_apply_round_fit(mpd_t *dec, mpd_uint_t rnd, const mpd_context_t *ctx, |
1793 | uint32_t *status) |
1794 | { |
1795 | if (_mpd_rnd_incr(dec, rnd, ctx)) { |
1796 | mpd_uint_t carry = _mpd_baseincr(dec->data, dec->len); |
1797 | if (carry) { |
1798 | if (!mpd_qresize(dec, dec->len+1, status)) { |
1799 | return 0; |
1800 | } |
1801 | dec->data[dec->len] = 1; |
1802 | dec->len += 1; |
1803 | } |
1804 | mpd_setdigits(dec); |
1805 | if (dec->digits > ctx->prec) { |
1806 | mpd_seterror(dec, MPD_Invalid_operation, status); |
1807 | return 0; |
1808 | } |
1809 | } |
1810 | return 1; |
1811 | } |
1812 | |
1813 | /* Check a normal number for overflow, underflow, clamping. If the operand |
1814 | is modified, it will be zero, special or (sub)normal with a coefficient |
1815 | that fits into the current context precision. */ |
1816 | static inline void |
1817 | _mpd_check_exp(mpd_t *dec, const mpd_context_t *ctx, uint32_t *status) |
1818 | { |
1819 | mpd_ssize_t adjexp, etiny, shift; |
1820 | int rnd; |
1821 | |
1822 | adjexp = mpd_adjexp(dec); |
1823 | if (adjexp > ctx->emax) { |
1824 | |
1825 | if (mpd_iszerocoeff(dec)) { |
1826 | dec->exp = ctx->emax; |
1827 | if (ctx->clamp) { |
1828 | dec->exp -= (ctx->prec-1); |
1829 | } |
1830 | mpd_zerocoeff(dec); |
1831 | *status |= MPD_Clamped; |
1832 | return; |
1833 | } |
1834 | |
1835 | switch (ctx->round) { |
1836 | case MPD_ROUND_HALF_UP: case MPD_ROUND_HALF_EVEN: |
1837 | case MPD_ROUND_HALF_DOWN: case MPD_ROUND_UP: |
1838 | case MPD_ROUND_TRUNC: |
1839 | mpd_setspecial(dec, mpd_sign(dec), MPD_INF); |
1840 | break; |
1841 | case MPD_ROUND_DOWN: case MPD_ROUND_05UP: |
1842 | mpd_qmaxcoeff(dec, ctx, status); |
1843 | dec->exp = ctx->emax - ctx->prec + 1; |
1844 | break; |
1845 | case MPD_ROUND_CEILING: |
1846 | if (mpd_isnegative(dec)) { |
1847 | mpd_qmaxcoeff(dec, ctx, status); |
1848 | dec->exp = ctx->emax - ctx->prec + 1; |
1849 | } |
1850 | else { |
1851 | mpd_setspecial(dec, MPD_POS, MPD_INF); |
1852 | } |
1853 | break; |
1854 | case MPD_ROUND_FLOOR: |
1855 | if (mpd_ispositive(dec)) { |
1856 | mpd_qmaxcoeff(dec, ctx, status); |
1857 | dec->exp = ctx->emax - ctx->prec + 1; |
1858 | } |
1859 | else { |
1860 | mpd_setspecial(dec, MPD_NEG, MPD_INF); |
1861 | } |
1862 | break; |
1863 | default: /* debug */ |
1864 | abort(); /* GCOV_NOT_REACHED */ |
1865 | } |
1866 | |
1867 | *status |= MPD_Overflow|MPD_Inexact|MPD_Rounded; |
1868 | |
1869 | } /* fold down */ |
1870 | else if (ctx->clamp && dec->exp > mpd_etop(ctx)) { |
1871 | /* At this point adjexp=exp+digits-1 <= emax and exp > etop=emax-prec+1: |
1872 | * (1) shift = exp -emax+prec-1 > 0 |
1873 | * (2) digits+shift = exp+digits-1 - emax + prec <= prec */ |
1874 | shift = dec->exp - mpd_etop(ctx); |
1875 | if (!mpd_qshiftl(dec, dec, shift, status)) { |
1876 | return; |
1877 | } |
1878 | dec->exp -= shift; |
1879 | *status |= MPD_Clamped; |
1880 | if (!mpd_iszerocoeff(dec) && adjexp < ctx->emin) { |
1881 | /* Underflow is impossible, since exp < etiny=emin-prec+1 |
1882 | * and exp > etop=emax-prec+1 would imply emax < emin. */ |
1883 | *status |= MPD_Subnormal; |
1884 | } |
1885 | } |
1886 | else if (adjexp < ctx->emin) { |
1887 | |
1888 | etiny = mpd_etiny(ctx); |
1889 | |
1890 | if (mpd_iszerocoeff(dec)) { |
1891 | if (dec->exp < etiny) { |
1892 | dec->exp = etiny; |
1893 | mpd_zerocoeff(dec); |
1894 | *status |= MPD_Clamped; |
1895 | } |
1896 | return; |
1897 | } |
1898 | |
1899 | *status |= MPD_Subnormal; |
1900 | if (dec->exp < etiny) { |
1901 | /* At this point adjexp=exp+digits-1 < emin and exp < etiny=emin-prec+1: |
1902 | * (1) shift = emin-prec+1 - exp > 0 |
1903 | * (2) digits-shift = exp+digits-1 - emin + prec < prec */ |
1904 | shift = etiny - dec->exp; |
1905 | rnd = (int)mpd_qshiftr_inplace(dec, shift); |
1906 | dec->exp = etiny; |
1907 | /* We always have a spare digit in case of an increment. */ |
1908 | _mpd_apply_round_excess(dec, rnd, ctx, status); |
1909 | *status |= MPD_Rounded; |
1910 | if (rnd) { |
1911 | *status |= (MPD_Inexact|MPD_Underflow); |
1912 | if (mpd_iszerocoeff(dec)) { |
1913 | mpd_zerocoeff(dec); |
1914 | *status |= MPD_Clamped; |
1915 | } |
1916 | } |
1917 | } |
1918 | /* Case exp >= etiny=emin-prec+1: |
1919 | * (1) adjexp=exp+digits-1 < emin |
1920 | * (2) digits < emin-exp+1 <= prec */ |
1921 | } |
1922 | } |
1923 | |
1924 | /* Transcendental functions do not always set Underflow reliably, |
1925 | * since they only use as much precision as is necessary for correct |
1926 | * rounding. If a result like 1.0000000000e-101 is finalized, there |
1927 | * is no rounding digit that would trigger Underflow. But we can |
1928 | * assume Inexact, so a short check suffices. */ |
1929 | static inline void |
1930 | mpd_check_underflow(mpd_t *dec, const mpd_context_t *ctx, uint32_t *status) |
1931 | { |
1932 | if (mpd_adjexp(dec) < ctx->emin && !mpd_iszero(dec) && |
1933 | dec->exp < mpd_etiny(ctx)) { |
1934 | *status |= MPD_Underflow; |
1935 | } |
1936 | } |
1937 | |
1938 | /* Check if a normal number must be rounded after the exponent has been checked. */ |
1939 | static inline void |
1940 | _mpd_check_round(mpd_t *dec, const mpd_context_t *ctx, uint32_t *status) |
1941 | { |
1942 | mpd_uint_t rnd; |
1943 | mpd_ssize_t shift; |
1944 | |
1945 | /* must handle specials: _mpd_check_exp() can produce infinities or NaNs */ |
1946 | if (mpd_isspecial(dec)) { |
1947 | return; |
1948 | } |
1949 | |
1950 | if (dec->digits > ctx->prec) { |
1951 | shift = dec->digits - ctx->prec; |
1952 | rnd = mpd_qshiftr_inplace(dec, shift); |
1953 | dec->exp += shift; |
1954 | _mpd_apply_round(dec, rnd, ctx, status); |
1955 | *status |= MPD_Rounded; |
1956 | if (rnd) { |
1957 | *status |= MPD_Inexact; |
1958 | } |
1959 | } |
1960 | } |
1961 | |
1962 | /* Finalize all operations. */ |
1963 | void |
1964 | mpd_qfinalize(mpd_t *result, const mpd_context_t *ctx, uint32_t *status) |
1965 | { |
1966 | if (mpd_isspecial(result)) { |
1967 | if (mpd_isnan(result)) { |
1968 | _mpd_fix_nan(result, ctx); |
1969 | } |
1970 | return; |
1971 | } |
1972 | |
1973 | _mpd_check_exp(result, ctx, status); |
1974 | _mpd_check_round(result, ctx, status); |
1975 | } |
1976 | |
1977 | |
1978 | /******************************************************************************/ |
1979 | /* Copying */ |
1980 | /******************************************************************************/ |
1981 | |
1982 | /* Internal function: Copy a decimal, share data with src: USE WITH CARE! */ |
1983 | static inline void |
1984 | _mpd_copy_shared(mpd_t *dest, const mpd_t *src) |
1985 | { |
1986 | dest->flags = src->flags; |
1987 | dest->exp = src->exp; |
1988 | dest->digits = src->digits; |
1989 | dest->len = src->len; |
1990 | dest->alloc = src->alloc; |
1991 | dest->data = src->data; |
1992 | |
1993 | mpd_set_shared_data(dest); |
1994 | } |
1995 | |
1996 | /* |
1997 | * Copy a decimal. In case of an error, status is set to MPD_Malloc_error. |
1998 | */ |
1999 | int |
2000 | mpd_qcopy(mpd_t *result, const mpd_t *a, uint32_t *status) |
2001 | { |
2002 | if (result == a) return 1; |
2003 | |
2004 | if (!mpd_qresize(result, a->len, status)) { |
2005 | return 0; |
2006 | } |
2007 | |
2008 | mpd_copy_flags(result, a); |
2009 | result->exp = a->exp; |
2010 | result->digits = a->digits; |
2011 | result->len = a->len; |
2012 | memcpy(result->data, a->data, a->len * (sizeof *result->data)); |
2013 | |
2014 | return 1; |
2015 | } |
2016 | |
2017 | /* Same as mpd_qcopy, but do not set the result to NaN on failure. */ |
2018 | int |
2019 | mpd_qcopy_cxx(mpd_t *result, const mpd_t *a) |
2020 | { |
2021 | if (result == a) return 1; |
2022 | |
2023 | if (!mpd_qresize_cxx(result, a->len)) { |
2024 | return 0; |
2025 | } |
2026 | |
2027 | mpd_copy_flags(result, a); |
2028 | result->exp = a->exp; |
2029 | result->digits = a->digits; |
2030 | result->len = a->len; |
2031 | memcpy(result->data, a->data, a->len * (sizeof *result->data)); |
2032 | |
2033 | return 1; |
2034 | } |
2035 | |
2036 | /* |
2037 | * Copy to a decimal with a static buffer. The caller has to make sure that |
2038 | * the buffer is big enough. Cannot fail. |
2039 | */ |
2040 | static void |
2041 | mpd_qcopy_static(mpd_t *result, const mpd_t *a) |
2042 | { |
2043 | if (result == a) return; |
2044 | |
2045 | memcpy(result->data, a->data, a->len * (sizeof *result->data)); |
2046 | |
2047 | mpd_copy_flags(result, a); |
2048 | result->exp = a->exp; |
2049 | result->digits = a->digits; |
2050 | result->len = a->len; |
2051 | } |
2052 | |
2053 | /* |
2054 | * Return a newly allocated copy of the operand. In case of an error, |
2055 | * status is set to MPD_Malloc_error and the return value is NULL. |
2056 | */ |
2057 | mpd_t * |
2058 | mpd_qncopy(const mpd_t *a) |
2059 | { |
2060 | mpd_t *result; |
2061 | |
2062 | if ((result = mpd_qnew_size(a->len)) == NULL) { |
2063 | return NULL; |
2064 | } |
2065 | memcpy(result->data, a->data, a->len * (sizeof *result->data)); |
2066 | mpd_copy_flags(result, a); |
2067 | result->exp = a->exp; |
2068 | result->digits = a->digits; |
2069 | result->len = a->len; |
2070 | |
2071 | return result; |
2072 | } |
2073 | |
2074 | /* |
2075 | * Copy a decimal and set the sign to positive. In case of an error, the |
2076 | * status is set to MPD_Malloc_error. |
2077 | */ |
2078 | int |
2079 | mpd_qcopy_abs(mpd_t *result, const mpd_t *a, uint32_t *status) |
2080 | { |
2081 | if (!mpd_qcopy(result, a, status)) { |
2082 | return 0; |
2083 | } |
2084 | mpd_set_positive(result); |
2085 | return 1; |
2086 | } |
2087 | |
2088 | /* |
2089 | * Copy a decimal and negate the sign. In case of an error, the |
2090 | * status is set to MPD_Malloc_error. |
2091 | */ |
2092 | int |
2093 | mpd_qcopy_negate(mpd_t *result, const mpd_t *a, uint32_t *status) |
2094 | { |
2095 | if (!mpd_qcopy(result, a, status)) { |
2096 | return 0; |
2097 | } |
2098 | _mpd_negate(result); |
2099 | return 1; |
2100 | } |
2101 | |
2102 | /* |
2103 | * Copy a decimal, setting the sign of the first operand to the sign of the |
2104 | * second operand. In case of an error, the status is set to MPD_Malloc_error. |
2105 | */ |
2106 | int |
2107 | mpd_qcopy_sign(mpd_t *result, const mpd_t *a, const mpd_t *b, uint32_t *status) |
2108 | { |
2109 | uint8_t sign_b = mpd_sign(b); /* result may equal b! */ |
2110 | |
2111 | if (!mpd_qcopy(result, a, status)) { |
2112 | return 0; |
2113 | } |
2114 | mpd_set_sign(result, sign_b); |
2115 | return 1; |
2116 | } |
2117 | |
2118 | |
2119 | /******************************************************************************/ |
2120 | /* Comparisons */ |
2121 | /******************************************************************************/ |
2122 | |
2123 | /* |
2124 | * For all functions that compare two operands and return an int the usual |
2125 | * convention applies to the return value: |
2126 | * |
2127 | * -1 if op1 < op2 |
2128 | * 0 if op1 == op2 |
2129 | * 1 if op1 > op2 |
2130 | * |
2131 | * INT_MAX for error |
2132 | */ |
2133 | |
2134 | |
2135 | /* Convenience macro. If a and b are not equal, return from the calling |
2136 | * function with the correct comparison value. */ |
2137 | #define CMP_EQUAL_OR_RETURN(a, b) \ |
2138 | if (a != b) { \ |
2139 | if (a < b) { \ |
2140 | return -1; \ |
2141 | } \ |
2142 | return 1; \ |
2143 | } |
2144 | |
2145 | /* |
2146 | * Compare the data of big and small. This function does the equivalent |
2147 | * of first shifting small to the left and then comparing the data of |
2148 | * big and small, except that no allocation for the left shift is needed. |
2149 | */ |
2150 | static int |
2151 | _mpd_basecmp(mpd_uint_t *big, mpd_uint_t *small, mpd_size_t n, mpd_size_t m, |
2152 | mpd_size_t shift) |
2153 | { |
2154 | #if defined(__GNUC__) && !defined(__INTEL_COMPILER) && !defined(__clang__) |
2155 | /* spurious uninitialized warnings */ |
2156 | mpd_uint_t l=l, lprev=lprev, h=h; |
2157 | #else |
2158 | mpd_uint_t l, lprev, h; |
2159 | #endif |
2160 | mpd_uint_t q, r; |
2161 | mpd_uint_t ph, x; |
2162 | |
2163 | assert(m > 0 && n >= m && shift > 0); |
2164 | |
2165 | _mpd_div_word(&q, &r, (mpd_uint_t)shift, MPD_RDIGITS); |
2166 | |
2167 | if (r != 0) { |
2168 | |
2169 | ph = mpd_pow10[r]; |
2170 | |
2171 | --m; --n; |
2172 | _mpd_divmod_pow10(&h, &lprev, small[m--], MPD_RDIGITS-r); |
2173 | if (h != 0) { |
2174 | CMP_EQUAL_OR_RETURN(big[n], h) |
2175 | --n; |
2176 | } |
2177 | for (; m != MPD_SIZE_MAX; m--,n--) { |
2178 | _mpd_divmod_pow10(&h, &l, small[m], MPD_RDIGITS-r); |
2179 | x = ph * lprev + h; |
2180 | CMP_EQUAL_OR_RETURN(big[n], x) |
2181 | lprev = l; |
2182 | } |
2183 | x = ph * lprev; |
2184 | CMP_EQUAL_OR_RETURN(big[q], x) |
2185 | } |
2186 | else { |
2187 | while (--m != MPD_SIZE_MAX) { |
2188 | CMP_EQUAL_OR_RETURN(big[m+q], small[m]) |
2189 | } |
2190 | } |
2191 | |
2192 | return !_mpd_isallzero(big, q); |
2193 | } |
2194 | |
2195 | /* Compare two decimals with the same adjusted exponent. */ |
2196 | static int |
2197 | _mpd_cmp_same_adjexp(const mpd_t *a, const mpd_t *b) |
2198 | { |
2199 | mpd_ssize_t shift, i; |
2200 | |
2201 | if (a->exp != b->exp) { |
2202 | /* Cannot wrap: a->exp + a->digits = b->exp + b->digits, so |
2203 | * a->exp - b->exp = b->digits - a->digits. */ |
2204 | shift = a->exp - b->exp; |
2205 | if (shift > 0) { |
2206 | return -1 * _mpd_basecmp(b->data, a->data, b->len, a->len, shift); |
2207 | } |
2208 | else { |
2209 | return _mpd_basecmp(a->data, b->data, a->len, b->len, -shift); |
2210 | } |
2211 | } |
2212 | |
2213 | /* |
2214 | * At this point adjexp(a) == adjexp(b) and a->exp == b->exp, |
2215 | * so a->digits == b->digits, therefore a->len == b->len. |
2216 | */ |
2217 | for (i = a->len-1; i >= 0; --i) { |
2218 | CMP_EQUAL_OR_RETURN(a->data[i], b->data[i]) |
2219 | } |
2220 | |
2221 | return 0; |
2222 | } |
2223 | |
2224 | /* Compare two numerical values. */ |
2225 | static int |
2226 | _mpd_cmp(const mpd_t *a, const mpd_t *b) |
2227 | { |
2228 | mpd_ssize_t adjexp_a, adjexp_b; |
2229 | |
2230 | /* equal pointers */ |
2231 | if (a == b) { |
2232 | return 0; |
2233 | } |
2234 | |
2235 | /* infinities */ |
2236 | if (mpd_isinfinite(a)) { |
2237 | if (mpd_isinfinite(b)) { |
2238 | return mpd_isnegative(b) - mpd_isnegative(a); |
2239 | } |
2240 | return mpd_arith_sign(a); |
2241 | } |
2242 | if (mpd_isinfinite(b)) { |
2243 | return -mpd_arith_sign(b); |
2244 | } |
2245 | |
2246 | /* zeros */ |
2247 | if (mpd_iszerocoeff(a)) { |
2248 | if (mpd_iszerocoeff(b)) { |
2249 | return 0; |
2250 | } |
2251 | return -mpd_arith_sign(b); |
2252 | } |
2253 | if (mpd_iszerocoeff(b)) { |
2254 | return mpd_arith_sign(a); |
2255 | } |
2256 | |
2257 | /* different signs */ |
2258 | if (mpd_sign(a) != mpd_sign(b)) { |
2259 | return mpd_sign(b) - mpd_sign(a); |
2260 | } |
2261 | |
2262 | /* different adjusted exponents */ |
2263 | adjexp_a = mpd_adjexp(a); |
2264 | adjexp_b = mpd_adjexp(b); |
2265 | if (adjexp_a != adjexp_b) { |
2266 | if (adjexp_a < adjexp_b) { |
2267 | return -1 * mpd_arith_sign(a); |
2268 | } |
2269 | return mpd_arith_sign(a); |
2270 | } |
2271 | |
2272 | /* same adjusted exponents */ |
2273 | return _mpd_cmp_same_adjexp(a, b) * mpd_arith_sign(a); |
2274 | } |
2275 | |
2276 | /* Compare the absolutes of two numerical values. */ |
2277 | static int |
2278 | _mpd_cmp_abs(const mpd_t *a, const mpd_t *b) |
2279 | { |
2280 | mpd_ssize_t adjexp_a, adjexp_b; |
2281 | |
2282 | /* equal pointers */ |
2283 | if (a == b) { |
2284 | return 0; |
2285 | } |
2286 | |
2287 | /* infinities */ |
2288 | if (mpd_isinfinite(a)) { |
2289 | if (mpd_isinfinite(b)) { |
2290 | return 0; |
2291 | } |
2292 | return 1; |
2293 | } |
2294 | if (mpd_isinfinite(b)) { |
2295 | return -1; |
2296 | } |
2297 | |
2298 | /* zeros */ |
2299 | if (mpd_iszerocoeff(a)) { |
2300 | if (mpd_iszerocoeff(b)) { |
2301 | return 0; |
2302 | } |
2303 | return -1; |
2304 | } |
2305 | if (mpd_iszerocoeff(b)) { |
2306 | return 1; |
2307 | } |
2308 | |
2309 | /* different adjusted exponents */ |
2310 | adjexp_a = mpd_adjexp(a); |
2311 | adjexp_b = mpd_adjexp(b); |
2312 | if (adjexp_a != adjexp_b) { |
2313 | if (adjexp_a < adjexp_b) { |
2314 | return -1; |
2315 | } |
2316 | return 1; |
2317 | } |
2318 | |
2319 | /* same adjusted exponents */ |
2320 | return _mpd_cmp_same_adjexp(a, b); |
2321 | } |
2322 | |
2323 | /* Compare two values and return an integer result. */ |
2324 | int |
2325 | mpd_qcmp(const mpd_t *a, const mpd_t *b, uint32_t *status) |
2326 | { |
2327 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
2328 | if (mpd_isnan(a) || mpd_isnan(b)) { |
2329 | *status |= MPD_Invalid_operation; |
2330 | return INT_MAX; |
2331 | } |
2332 | } |
2333 | |
2334 | return _mpd_cmp(a, b); |
2335 | } |
2336 | |
2337 | /* |
2338 | * Compare a and b, convert the usual integer result to a decimal and |
2339 | * store it in 'result'. For convenience, the integer result of the comparison |
2340 | * is returned. Comparisons involving NaNs return NaN/INT_MAX. |
2341 | */ |
2342 | int |
2343 | mpd_qcompare(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2344 | const mpd_context_t *ctx, uint32_t *status) |
2345 | { |
2346 | int c; |
2347 | |
2348 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
2349 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
2350 | return INT_MAX; |
2351 | } |
2352 | } |
2353 | |
2354 | c = _mpd_cmp(a, b); |
2355 | _settriple(result, (c < 0), (c != 0), 0); |
2356 | return c; |
2357 | } |
2358 | |
2359 | /* Same as mpd_compare(), but signal for all NaNs, i.e. also for quiet NaNs. */ |
2360 | int |
2361 | mpd_qcompare_signal(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2362 | const mpd_context_t *ctx, uint32_t *status) |
2363 | { |
2364 | int c; |
2365 | |
2366 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
2367 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
2368 | *status |= MPD_Invalid_operation; |
2369 | return INT_MAX; |
2370 | } |
2371 | } |
2372 | |
2373 | c = _mpd_cmp(a, b); |
2374 | _settriple(result, (c < 0), (c != 0), 0); |
2375 | return c; |
2376 | } |
2377 | |
2378 | /* Compare the operands using a total order. */ |
2379 | int |
2380 | mpd_cmp_total(const mpd_t *a, const mpd_t *b) |
2381 | { |
2382 | mpd_t aa, bb; |
2383 | int nan_a, nan_b; |
2384 | int c; |
2385 | |
2386 | if (mpd_sign(a) != mpd_sign(b)) { |
2387 | return mpd_sign(b) - mpd_sign(a); |
2388 | } |
2389 | |
2390 | |
2391 | if (mpd_isnan(a)) { |
2392 | c = 1; |
2393 | if (mpd_isnan(b)) { |
2394 | nan_a = (mpd_isqnan(a)) ? 1 : 0; |
2395 | nan_b = (mpd_isqnan(b)) ? 1 : 0; |
2396 | if (nan_b == nan_a) { |
2397 | if (a->len > 0 && b->len > 0) { |
2398 | _mpd_copy_shared(&aa, a); |
2399 | _mpd_copy_shared(&bb, b); |
2400 | aa.exp = bb.exp = 0; |
2401 | /* compare payload */ |
2402 | c = _mpd_cmp_abs(&aa, &bb); |
2403 | } |
2404 | else { |
2405 | c = (a->len > 0) - (b->len > 0); |
2406 | } |
2407 | } |
2408 | else { |
2409 | c = nan_a - nan_b; |
2410 | } |
2411 | } |
2412 | } |
2413 | else if (mpd_isnan(b)) { |
2414 | c = -1; |
2415 | } |
2416 | else { |
2417 | c = _mpd_cmp_abs(a, b); |
2418 | if (c == 0 && a->exp != b->exp) { |
2419 | c = (a->exp < b->exp) ? -1 : 1; |
2420 | } |
2421 | } |
2422 | |
2423 | return c * mpd_arith_sign(a); |
2424 | } |
2425 | |
2426 | /* |
2427 | * Compare a and b according to a total order, convert the usual integer result |
2428 | * to a decimal and store it in 'result'. For convenience, the integer result |
2429 | * of the comparison is returned. |
2430 | */ |
2431 | int |
2432 | mpd_compare_total(mpd_t *result, const mpd_t *a, const mpd_t *b) |
2433 | { |
2434 | int c; |
2435 | |
2436 | c = mpd_cmp_total(a, b); |
2437 | _settriple(result, (c < 0), (c != 0), 0); |
2438 | return c; |
2439 | } |
2440 | |
2441 | /* Compare the magnitude of the operands using a total order. */ |
2442 | int |
2443 | mpd_cmp_total_mag(const mpd_t *a, const mpd_t *b) |
2444 | { |
2445 | mpd_t aa, bb; |
2446 | |
2447 | _mpd_copy_shared(&aa, a); |
2448 | _mpd_copy_shared(&bb, b); |
2449 | |
2450 | mpd_set_positive(&aa); |
2451 | mpd_set_positive(&bb); |
2452 | |
2453 | return mpd_cmp_total(&aa, &bb); |
2454 | } |
2455 | |
2456 | /* |
2457 | * Compare the magnitude of a and b according to a total order, convert the |
2458 | * the usual integer result to a decimal and store it in 'result'. |
2459 | * For convenience, the integer result of the comparison is returned. |
2460 | */ |
2461 | int |
2462 | mpd_compare_total_mag(mpd_t *result, const mpd_t *a, const mpd_t *b) |
2463 | { |
2464 | int c; |
2465 | |
2466 | c = mpd_cmp_total_mag(a, b); |
2467 | _settriple(result, (c < 0), (c != 0), 0); |
2468 | return c; |
2469 | } |
2470 | |
2471 | /* Determine an ordering for operands that are numerically equal. */ |
2472 | static inline int |
2473 | _mpd_cmp_numequal(const mpd_t *a, const mpd_t *b) |
2474 | { |
2475 | int sign_a, sign_b; |
2476 | int c; |
2477 | |
2478 | sign_a = mpd_sign(a); |
2479 | sign_b = mpd_sign(b); |
2480 | if (sign_a != sign_b) { |
2481 | c = sign_b - sign_a; |
2482 | } |
2483 | else { |
2484 | c = (a->exp < b->exp) ? -1 : 1; |
2485 | c *= mpd_arith_sign(a); |
2486 | } |
2487 | |
2488 | return c; |
2489 | } |
2490 | |
2491 | |
2492 | /******************************************************************************/ |
2493 | /* Shifting the coefficient */ |
2494 | /******************************************************************************/ |
2495 | |
2496 | /* |
2497 | * Shift the coefficient of the operand to the left, no check for specials. |
2498 | * Both operands may be the same pointer. If the result length has to be |
2499 | * increased, mpd_qresize() might fail with MPD_Malloc_error. |
2500 | */ |
2501 | int |
2502 | mpd_qshiftl(mpd_t *result, const mpd_t *a, mpd_ssize_t n, uint32_t *status) |
2503 | { |
2504 | mpd_ssize_t size; |
2505 | |
2506 | assert(!mpd_isspecial(a)); |
2507 | assert(n >= 0); |
2508 | |
2509 | if (mpd_iszerocoeff(a) || n == 0) { |
2510 | return mpd_qcopy(result, a, status); |
2511 | } |
2512 | |
2513 | size = mpd_digits_to_size(a->digits+n); |
2514 | if (!mpd_qresize(result, size, status)) { |
2515 | return 0; /* result is NaN */ |
2516 | } |
2517 | |
2518 | _mpd_baseshiftl(result->data, a->data, size, a->len, n); |
2519 | |
2520 | mpd_copy_flags(result, a); |
2521 | result->exp = a->exp; |
2522 | result->digits = a->digits+n; |
2523 | result->len = size; |
2524 | |
2525 | return 1; |
2526 | } |
2527 | |
2528 | /* Determine the rounding indicator if all digits of the coefficient are shifted |
2529 | * out of the picture. */ |
2530 | static mpd_uint_t |
2531 | _mpd_get_rnd(const mpd_uint_t *data, mpd_ssize_t len, int use_msd) |
2532 | { |
2533 | mpd_uint_t rnd = 0, rest = 0, word; |
2534 | |
2535 | word = data[len-1]; |
2536 | /* special treatment for the most significant digit if shift == digits */ |
2537 | if (use_msd) { |
2538 | _mpd_divmod_pow10(&rnd, &rest, word, mpd_word_digits(word)-1); |
2539 | if (len > 1 && rest == 0) { |
2540 | rest = !_mpd_isallzero(data, len-1); |
2541 | } |
2542 | } |
2543 | else { |
2544 | rest = !_mpd_isallzero(data, len); |
2545 | } |
2546 | |
2547 | return (rnd == 0 || rnd == 5) ? rnd + !!rest : rnd; |
2548 | } |
2549 | |
2550 | /* |
2551 | * Same as mpd_qshiftr(), but 'result' is an mpd_t with a static coefficient. |
2552 | * It is the caller's responsibility to ensure that the coefficient is big |
2553 | * enough. The function cannot fail. |
2554 | */ |
2555 | static mpd_uint_t |
2556 | mpd_qsshiftr(mpd_t *result, const mpd_t *a, mpd_ssize_t n) |
2557 | { |
2558 | mpd_uint_t rnd; |
2559 | mpd_ssize_t size; |
2560 | |
2561 | assert(!mpd_isspecial(a)); |
2562 | assert(n >= 0); |
2563 | |
2564 | if (mpd_iszerocoeff(a) || n == 0) { |
2565 | mpd_qcopy_static(result, a); |
2566 | return 0; |
2567 | } |
2568 | |
2569 | if (n >= a->digits) { |
2570 | rnd = _mpd_get_rnd(a->data, a->len, (n==a->digits)); |
2571 | mpd_zerocoeff(result); |
2572 | } |
2573 | else { |
2574 | result->digits = a->digits-n; |
2575 | size = mpd_digits_to_size(result->digits); |
2576 | rnd = _mpd_baseshiftr(result->data, a->data, a->len, n); |
2577 | result->len = size; |
2578 | } |
2579 | |
2580 | mpd_copy_flags(result, a); |
2581 | result->exp = a->exp; |
2582 | |
2583 | return rnd; |
2584 | } |
2585 | |
2586 | /* |
2587 | * Inplace shift of the coefficient to the right, no check for specials. |
2588 | * Returns the rounding indicator for mpd_rnd_incr(). |
2589 | * The function cannot fail. |
2590 | */ |
2591 | mpd_uint_t |
2592 | mpd_qshiftr_inplace(mpd_t *result, mpd_ssize_t n) |
2593 | { |
2594 | uint32_t dummy; |
2595 | mpd_uint_t rnd; |
2596 | mpd_ssize_t size; |
2597 | |
2598 | assert(!mpd_isspecial(result)); |
2599 | assert(n >= 0); |
2600 | |
2601 | if (mpd_iszerocoeff(result) || n == 0) { |
2602 | return 0; |
2603 | } |
2604 | |
2605 | if (n >= result->digits) { |
2606 | rnd = _mpd_get_rnd(result->data, result->len, (n==result->digits)); |
2607 | mpd_zerocoeff(result); |
2608 | } |
2609 | else { |
2610 | rnd = _mpd_baseshiftr(result->data, result->data, result->len, n); |
2611 | result->digits -= n; |
2612 | size = mpd_digits_to_size(result->digits); |
2613 | /* reducing the size cannot fail */ |
2614 | mpd_qresize(result, size, &dummy); |
2615 | result->len = size; |
2616 | } |
2617 | |
2618 | return rnd; |
2619 | } |
2620 | |
2621 | /* |
2622 | * Shift the coefficient of the operand to the right, no check for specials. |
2623 | * Both operands may be the same pointer. Returns the rounding indicator to |
2624 | * be used by mpd_rnd_incr(). If the result length has to be increased, |
2625 | * mpd_qcopy() or mpd_qresize() might fail with MPD_Malloc_error. In those |
2626 | * cases, MPD_UINT_MAX is returned. |
2627 | */ |
2628 | mpd_uint_t |
2629 | mpd_qshiftr(mpd_t *result, const mpd_t *a, mpd_ssize_t n, uint32_t *status) |
2630 | { |
2631 | mpd_uint_t rnd; |
2632 | mpd_ssize_t size; |
2633 | |
2634 | assert(!mpd_isspecial(a)); |
2635 | assert(n >= 0); |
2636 | |
2637 | if (mpd_iszerocoeff(a) || n == 0) { |
2638 | if (!mpd_qcopy(result, a, status)) { |
2639 | return MPD_UINT_MAX; |
2640 | } |
2641 | return 0; |
2642 | } |
2643 | |
2644 | if (n >= a->digits) { |
2645 | rnd = _mpd_get_rnd(a->data, a->len, (n==a->digits)); |
2646 | mpd_zerocoeff(result); |
2647 | } |
2648 | else { |
2649 | result->digits = a->digits-n; |
2650 | size = mpd_digits_to_size(result->digits); |
2651 | if (result == a) { |
2652 | rnd = _mpd_baseshiftr(result->data, a->data, a->len, n); |
2653 | /* reducing the size cannot fail */ |
2654 | mpd_qresize(result, size, status); |
2655 | } |
2656 | else { |
2657 | if (!mpd_qresize(result, size, status)) { |
2658 | return MPD_UINT_MAX; |
2659 | } |
2660 | rnd = _mpd_baseshiftr(result->data, a->data, a->len, n); |
2661 | } |
2662 | result->len = size; |
2663 | } |
2664 | |
2665 | mpd_copy_flags(result, a); |
2666 | result->exp = a->exp; |
2667 | |
2668 | return rnd; |
2669 | } |
2670 | |
2671 | |
2672 | /******************************************************************************/ |
2673 | /* Miscellaneous operations */ |
2674 | /******************************************************************************/ |
2675 | |
2676 | /* Logical And */ |
2677 | void |
2678 | mpd_qand(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2679 | const mpd_context_t *ctx, uint32_t *status) |
2680 | { |
2681 | const mpd_t *big = a, *small = b; |
2682 | mpd_uint_t x, y, z, xbit, ybit; |
2683 | int k, mswdigits; |
2684 | mpd_ssize_t i; |
2685 | |
2686 | if (mpd_isspecial(a) || mpd_isspecial(b) || |
2687 | mpd_isnegative(a) || mpd_isnegative(b) || |
2688 | a->exp != 0 || b->exp != 0) { |
2689 | mpd_seterror(result, MPD_Invalid_operation, status); |
2690 | return; |
2691 | } |
2692 | if (b->digits > a->digits) { |
2693 | big = b; |
2694 | small = a; |
2695 | } |
2696 | if (!mpd_qresize(result, big->len, status)) { |
2697 | return; |
2698 | } |
2699 | |
2700 | |
2701 | /* full words */ |
2702 | for (i = 0; i < small->len-1; i++) { |
2703 | x = small->data[i]; |
2704 | y = big->data[i]; |
2705 | z = 0; |
2706 | for (k = 0; k < MPD_RDIGITS; k++) { |
2707 | xbit = x % 10; |
2708 | x /= 10; |
2709 | ybit = y % 10; |
2710 | y /= 10; |
2711 | if (xbit > 1 || ybit > 1) { |
2712 | goto invalid_operation; |
2713 | } |
2714 | z += (xbit&ybit) ? mpd_pow10[k] : 0; |
2715 | } |
2716 | result->data[i] = z; |
2717 | } |
2718 | /* most significant word of small */ |
2719 | x = small->data[i]; |
2720 | y = big->data[i]; |
2721 | z = 0; |
2722 | mswdigits = mpd_word_digits(x); |
2723 | for (k = 0; k < mswdigits; k++) { |
2724 | xbit = x % 10; |
2725 | x /= 10; |
2726 | ybit = y % 10; |
2727 | y /= 10; |
2728 | if (xbit > 1 || ybit > 1) { |
2729 | goto invalid_operation; |
2730 | } |
2731 | z += (xbit&ybit) ? mpd_pow10[k] : 0; |
2732 | } |
2733 | result->data[i++] = z; |
2734 | |
2735 | /* scan the rest of y for digits > 1 */ |
2736 | for (; k < MPD_RDIGITS; k++) { |
2737 | ybit = y % 10; |
2738 | y /= 10; |
2739 | if (ybit > 1) { |
2740 | goto invalid_operation; |
2741 | } |
2742 | } |
2743 | /* scan the rest of big for digits > 1 */ |
2744 | for (; i < big->len; i++) { |
2745 | y = big->data[i]; |
2746 | for (k = 0; k < MPD_RDIGITS; k++) { |
2747 | ybit = y % 10; |
2748 | y /= 10; |
2749 | if (ybit > 1) { |
2750 | goto invalid_operation; |
2751 | } |
2752 | } |
2753 | } |
2754 | |
2755 | mpd_clear_flags(result); |
2756 | result->exp = 0; |
2757 | result->len = _mpd_real_size(result->data, small->len); |
2758 | mpd_qresize(result, result->len, status); |
2759 | mpd_setdigits(result); |
2760 | _mpd_cap(result, ctx); |
2761 | return; |
2762 | |
2763 | invalid_operation: |
2764 | mpd_seterror(result, MPD_Invalid_operation, status); |
2765 | } |
2766 | |
2767 | /* Class of an operand. Returns a pointer to the constant name. */ |
2768 | const char * |
2769 | mpd_class(const mpd_t *a, const mpd_context_t *ctx) |
2770 | { |
2771 | if (mpd_isnan(a)) { |
2772 | if (mpd_isqnan(a)) |
2773 | return "NaN" ; |
2774 | else |
2775 | return "sNaN" ; |
2776 | } |
2777 | else if (mpd_ispositive(a)) { |
2778 | if (mpd_isinfinite(a)) |
2779 | return "+Infinity" ; |
2780 | else if (mpd_iszero(a)) |
2781 | return "+Zero" ; |
2782 | else if (mpd_isnormal(a, ctx)) |
2783 | return "+Normal" ; |
2784 | else |
2785 | return "+Subnormal" ; |
2786 | } |
2787 | else { |
2788 | if (mpd_isinfinite(a)) |
2789 | return "-Infinity" ; |
2790 | else if (mpd_iszero(a)) |
2791 | return "-Zero" ; |
2792 | else if (mpd_isnormal(a, ctx)) |
2793 | return "-Normal" ; |
2794 | else |
2795 | return "-Subnormal" ; |
2796 | } |
2797 | } |
2798 | |
2799 | /* Logical Xor */ |
2800 | void |
2801 | mpd_qinvert(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
2802 | uint32_t *status) |
2803 | { |
2804 | mpd_uint_t x, z, xbit; |
2805 | mpd_ssize_t i, digits, len; |
2806 | mpd_ssize_t q, r; |
2807 | int k; |
2808 | |
2809 | if (mpd_isspecial(a) || mpd_isnegative(a) || a->exp != 0) { |
2810 | mpd_seterror(result, MPD_Invalid_operation, status); |
2811 | return; |
2812 | } |
2813 | |
2814 | digits = (a->digits < ctx->prec) ? ctx->prec : a->digits; |
2815 | _mpd_idiv_word(&q, &r, digits, MPD_RDIGITS); |
2816 | len = (r == 0) ? q : q+1; |
2817 | if (!mpd_qresize(result, len, status)) { |
2818 | return; |
2819 | } |
2820 | |
2821 | for (i = 0; i < len; i++) { |
2822 | x = (i < a->len) ? a->data[i] : 0; |
2823 | z = 0; |
2824 | for (k = 0; k < MPD_RDIGITS; k++) { |
2825 | xbit = x % 10; |
2826 | x /= 10; |
2827 | if (xbit > 1) { |
2828 | goto invalid_operation; |
2829 | } |
2830 | z += !xbit ? mpd_pow10[k] : 0; |
2831 | } |
2832 | result->data[i] = z; |
2833 | } |
2834 | |
2835 | mpd_clear_flags(result); |
2836 | result->exp = 0; |
2837 | result->len = _mpd_real_size(result->data, len); |
2838 | mpd_qresize(result, result->len, status); |
2839 | mpd_setdigits(result); |
2840 | _mpd_cap(result, ctx); |
2841 | return; |
2842 | |
2843 | invalid_operation: |
2844 | mpd_seterror(result, MPD_Invalid_operation, status); |
2845 | } |
2846 | |
2847 | /* Exponent of the magnitude of the most significant digit of the operand. */ |
2848 | void |
2849 | mpd_qlogb(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
2850 | uint32_t *status) |
2851 | { |
2852 | if (mpd_isspecial(a)) { |
2853 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
2854 | return; |
2855 | } |
2856 | mpd_setspecial(result, MPD_POS, MPD_INF); |
2857 | } |
2858 | else if (mpd_iszerocoeff(a)) { |
2859 | mpd_setspecial(result, MPD_NEG, MPD_INF); |
2860 | *status |= MPD_Division_by_zero; |
2861 | } |
2862 | else { |
2863 | mpd_qset_ssize(result, mpd_adjexp(a), ctx, status); |
2864 | } |
2865 | } |
2866 | |
2867 | /* Logical Or */ |
2868 | void |
2869 | mpd_qor(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2870 | const mpd_context_t *ctx, uint32_t *status) |
2871 | { |
2872 | const mpd_t *big = a, *small = b; |
2873 | mpd_uint_t x, y, z, xbit, ybit; |
2874 | int k, mswdigits; |
2875 | mpd_ssize_t i; |
2876 | |
2877 | if (mpd_isspecial(a) || mpd_isspecial(b) || |
2878 | mpd_isnegative(a) || mpd_isnegative(b) || |
2879 | a->exp != 0 || b->exp != 0) { |
2880 | mpd_seterror(result, MPD_Invalid_operation, status); |
2881 | return; |
2882 | } |
2883 | if (b->digits > a->digits) { |
2884 | big = b; |
2885 | small = a; |
2886 | } |
2887 | if (!mpd_qresize(result, big->len, status)) { |
2888 | return; |
2889 | } |
2890 | |
2891 | |
2892 | /* full words */ |
2893 | for (i = 0; i < small->len-1; i++) { |
2894 | x = small->data[i]; |
2895 | y = big->data[i]; |
2896 | z = 0; |
2897 | for (k = 0; k < MPD_RDIGITS; k++) { |
2898 | xbit = x % 10; |
2899 | x /= 10; |
2900 | ybit = y % 10; |
2901 | y /= 10; |
2902 | if (xbit > 1 || ybit > 1) { |
2903 | goto invalid_operation; |
2904 | } |
2905 | z += (xbit|ybit) ? mpd_pow10[k] : 0; |
2906 | } |
2907 | result->data[i] = z; |
2908 | } |
2909 | /* most significant word of small */ |
2910 | x = small->data[i]; |
2911 | y = big->data[i]; |
2912 | z = 0; |
2913 | mswdigits = mpd_word_digits(x); |
2914 | for (k = 0; k < mswdigits; k++) { |
2915 | xbit = x % 10; |
2916 | x /= 10; |
2917 | ybit = y % 10; |
2918 | y /= 10; |
2919 | if (xbit > 1 || ybit > 1) { |
2920 | goto invalid_operation; |
2921 | } |
2922 | z += (xbit|ybit) ? mpd_pow10[k] : 0; |
2923 | } |
2924 | |
2925 | /* scan for digits > 1 and copy the rest of y */ |
2926 | for (; k < MPD_RDIGITS; k++) { |
2927 | ybit = y % 10; |
2928 | y /= 10; |
2929 | if (ybit > 1) { |
2930 | goto invalid_operation; |
2931 | } |
2932 | z += ybit*mpd_pow10[k]; |
2933 | } |
2934 | result->data[i++] = z; |
2935 | /* scan for digits > 1 and copy the rest of big */ |
2936 | for (; i < big->len; i++) { |
2937 | y = big->data[i]; |
2938 | for (k = 0; k < MPD_RDIGITS; k++) { |
2939 | ybit = y % 10; |
2940 | y /= 10; |
2941 | if (ybit > 1) { |
2942 | goto invalid_operation; |
2943 | } |
2944 | } |
2945 | result->data[i] = big->data[i]; |
2946 | } |
2947 | |
2948 | mpd_clear_flags(result); |
2949 | result->exp = 0; |
2950 | result->len = _mpd_real_size(result->data, big->len); |
2951 | mpd_qresize(result, result->len, status); |
2952 | mpd_setdigits(result); |
2953 | _mpd_cap(result, ctx); |
2954 | return; |
2955 | |
2956 | invalid_operation: |
2957 | mpd_seterror(result, MPD_Invalid_operation, status); |
2958 | } |
2959 | |
2960 | /* |
2961 | * Rotate the coefficient of 'a' by 'b' digits. 'b' must be an integer with |
2962 | * exponent 0. |
2963 | */ |
2964 | void |
2965 | mpd_qrotate(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2966 | const mpd_context_t *ctx, uint32_t *status) |
2967 | { |
2968 | uint32_t workstatus = 0; |
2969 | MPD_NEW_STATIC(tmp,0,0,0,0); |
2970 | MPD_NEW_STATIC(big,0,0,0,0); |
2971 | MPD_NEW_STATIC(small,0,0,0,0); |
2972 | mpd_ssize_t n, lshift, rshift; |
2973 | |
2974 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
2975 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
2976 | return; |
2977 | } |
2978 | } |
2979 | if (b->exp != 0 || mpd_isinfinite(b)) { |
2980 | mpd_seterror(result, MPD_Invalid_operation, status); |
2981 | return; |
2982 | } |
2983 | |
2984 | n = mpd_qget_ssize(b, &workstatus); |
2985 | if (workstatus&MPD_Invalid_operation) { |
2986 | mpd_seterror(result, MPD_Invalid_operation, status); |
2987 | return; |
2988 | } |
2989 | if (n > ctx->prec || n < -ctx->prec) { |
2990 | mpd_seterror(result, MPD_Invalid_operation, status); |
2991 | return; |
2992 | } |
2993 | if (mpd_isinfinite(a)) { |
2994 | mpd_qcopy(result, a, status); |
2995 | return; |
2996 | } |
2997 | |
2998 | if (n >= 0) { |
2999 | lshift = n; |
3000 | rshift = ctx->prec-n; |
3001 | } |
3002 | else { |
3003 | lshift = ctx->prec+n; |
3004 | rshift = -n; |
3005 | } |
3006 | |
3007 | if (a->digits > ctx->prec) { |
3008 | if (!mpd_qcopy(&tmp, a, status)) { |
3009 | mpd_seterror(result, MPD_Malloc_error, status); |
3010 | goto finish; |
3011 | } |
3012 | _mpd_cap(&tmp, ctx); |
3013 | a = &tmp; |
3014 | } |
3015 | |
3016 | if (!mpd_qshiftl(&big, a, lshift, status)) { |
3017 | mpd_seterror(result, MPD_Malloc_error, status); |
3018 | goto finish; |
3019 | } |
3020 | _mpd_cap(&big, ctx); |
3021 | |
3022 | if (mpd_qshiftr(&small, a, rshift, status) == MPD_UINT_MAX) { |
3023 | mpd_seterror(result, MPD_Malloc_error, status); |
3024 | goto finish; |
3025 | } |
3026 | _mpd_qadd(result, &big, &small, ctx, status); |
3027 | |
3028 | |
3029 | finish: |
3030 | mpd_del(&tmp); |
3031 | mpd_del(&big); |
3032 | mpd_del(&small); |
3033 | } |
3034 | |
3035 | /* |
3036 | * b must be an integer with exponent 0 and in the range +-2*(emax + prec). |
3037 | * XXX: In my opinion +-(2*emax + prec) would be more sensible. |
3038 | * The result is a with the value of b added to its exponent. |
3039 | */ |
3040 | void |
3041 | mpd_qscaleb(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3042 | const mpd_context_t *ctx, uint32_t *status) |
3043 | { |
3044 | uint32_t workstatus = 0; |
3045 | mpd_uint_t n, maxjump; |
3046 | #ifndef LEGACY_COMPILER |
3047 | int64_t exp; |
3048 | #else |
3049 | mpd_uint_t x; |
3050 | int x_sign, n_sign; |
3051 | mpd_ssize_t exp; |
3052 | #endif |
3053 | |
3054 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3055 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
3056 | return; |
3057 | } |
3058 | } |
3059 | if (b->exp != 0 || mpd_isinfinite(b)) { |
3060 | mpd_seterror(result, MPD_Invalid_operation, status); |
3061 | return; |
3062 | } |
3063 | |
3064 | n = mpd_qabs_uint(b, &workstatus); |
3065 | /* the spec demands this */ |
3066 | maxjump = 2 * (mpd_uint_t)(ctx->emax + ctx->prec); |
3067 | |
3068 | if (n > maxjump || workstatus&MPD_Invalid_operation) { |
3069 | mpd_seterror(result, MPD_Invalid_operation, status); |
3070 | return; |
3071 | } |
3072 | if (mpd_isinfinite(a)) { |
3073 | mpd_qcopy(result, a, status); |
3074 | return; |
3075 | } |
3076 | |
3077 | #ifndef LEGACY_COMPILER |
3078 | exp = a->exp + (int64_t)n * mpd_arith_sign(b); |
3079 | exp = (exp > MPD_EXP_INF) ? MPD_EXP_INF : exp; |
3080 | exp = (exp < MPD_EXP_CLAMP) ? MPD_EXP_CLAMP : exp; |
3081 | #else |
3082 | x = (a->exp < 0) ? -a->exp : a->exp; |
3083 | x_sign = (a->exp < 0) ? 1 : 0; |
3084 | n_sign = mpd_isnegative(b) ? 1 : 0; |
3085 | |
3086 | if (x_sign == n_sign) { |
3087 | x = x + n; |
3088 | if (x < n) x = MPD_UINT_MAX; |
3089 | } |
3090 | else { |
3091 | x_sign = (x >= n) ? x_sign : n_sign; |
3092 | x = (x >= n) ? x - n : n - x; |
3093 | } |
3094 | if (!x_sign && x > MPD_EXP_INF) x = MPD_EXP_INF; |
3095 | if (x_sign && x > -MPD_EXP_CLAMP) x = -MPD_EXP_CLAMP; |
3096 | exp = x_sign ? -((mpd_ssize_t)x) : (mpd_ssize_t)x; |
3097 | #endif |
3098 | |
3099 | mpd_qcopy(result, a, status); |
3100 | result->exp = (mpd_ssize_t)exp; |
3101 | |
3102 | mpd_qfinalize(result, ctx, status); |
3103 | } |
3104 | |
3105 | /* |
3106 | * Shift the coefficient by n digits, positive n is a left shift. In the case |
3107 | * of a left shift, the result is decapitated to fit the context precision. If |
3108 | * you don't want that, use mpd_shiftl(). |
3109 | */ |
3110 | void |
3111 | mpd_qshiftn(mpd_t *result, const mpd_t *a, mpd_ssize_t n, const mpd_context_t *ctx, |
3112 | uint32_t *status) |
3113 | { |
3114 | if (mpd_isspecial(a)) { |
3115 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
3116 | return; |
3117 | } |
3118 | mpd_qcopy(result, a, status); |
3119 | return; |
3120 | } |
3121 | |
3122 | if (n >= 0 && n <= ctx->prec) { |
3123 | mpd_qshiftl(result, a, n, status); |
3124 | _mpd_cap(result, ctx); |
3125 | } |
3126 | else if (n < 0 && n >= -ctx->prec) { |
3127 | if (!mpd_qcopy(result, a, status)) { |
3128 | return; |
3129 | } |
3130 | _mpd_cap(result, ctx); |
3131 | mpd_qshiftr_inplace(result, -n); |
3132 | } |
3133 | else { |
3134 | mpd_seterror(result, MPD_Invalid_operation, status); |
3135 | } |
3136 | } |
3137 | |
3138 | /* |
3139 | * Same as mpd_shiftn(), but the shift is specified by the decimal b, which |
3140 | * must be an integer with a zero exponent. Infinities remain infinities. |
3141 | */ |
3142 | void |
3143 | mpd_qshift(mpd_t *result, const mpd_t *a, const mpd_t *b, const mpd_context_t *ctx, |
3144 | uint32_t *status) |
3145 | { |
3146 | uint32_t workstatus = 0; |
3147 | mpd_ssize_t n; |
3148 | |
3149 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3150 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
3151 | return; |
3152 | } |
3153 | } |
3154 | if (b->exp != 0 || mpd_isinfinite(b)) { |
3155 | mpd_seterror(result, MPD_Invalid_operation, status); |
3156 | return; |
3157 | } |
3158 | |
3159 | n = mpd_qget_ssize(b, &workstatus); |
3160 | if (workstatus&MPD_Invalid_operation) { |
3161 | mpd_seterror(result, MPD_Invalid_operation, status); |
3162 | return; |
3163 | } |
3164 | if (n > ctx->prec || n < -ctx->prec) { |
3165 | mpd_seterror(result, MPD_Invalid_operation, status); |
3166 | return; |
3167 | } |
3168 | if (mpd_isinfinite(a)) { |
3169 | mpd_qcopy(result, a, status); |
3170 | return; |
3171 | } |
3172 | |
3173 | if (n >= 0) { |
3174 | mpd_qshiftl(result, a, n, status); |
3175 | _mpd_cap(result, ctx); |
3176 | } |
3177 | else { |
3178 | if (!mpd_qcopy(result, a, status)) { |
3179 | return; |
3180 | } |
3181 | _mpd_cap(result, ctx); |
3182 | mpd_qshiftr_inplace(result, -n); |
3183 | } |
3184 | } |
3185 | |
3186 | /* Logical Xor */ |
3187 | void |
3188 | mpd_qxor(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3189 | const mpd_context_t *ctx, uint32_t *status) |
3190 | { |
3191 | const mpd_t *big = a, *small = b; |
3192 | mpd_uint_t x, y, z, xbit, ybit; |
3193 | int k, mswdigits; |
3194 | mpd_ssize_t i; |
3195 | |
3196 | if (mpd_isspecial(a) || mpd_isspecial(b) || |
3197 | mpd_isnegative(a) || mpd_isnegative(b) || |
3198 | a->exp != 0 || b->exp != 0) { |
3199 | mpd_seterror(result, MPD_Invalid_operation, status); |
3200 | return; |
3201 | } |
3202 | if (b->digits > a->digits) { |
3203 | big = b; |
3204 | small = a; |
3205 | } |
3206 | if (!mpd_qresize(result, big->len, status)) { |
3207 | return; |
3208 | } |
3209 | |
3210 | |
3211 | /* full words */ |
3212 | for (i = 0; i < small->len-1; i++) { |
3213 | x = small->data[i]; |
3214 | y = big->data[i]; |
3215 | z = 0; |
3216 | for (k = 0; k < MPD_RDIGITS; k++) { |
3217 | xbit = x % 10; |
3218 | x /= 10; |
3219 | ybit = y % 10; |
3220 | y /= 10; |
3221 | if (xbit > 1 || ybit > 1) { |
3222 | goto invalid_operation; |
3223 | } |
3224 | z += (xbit^ybit) ? mpd_pow10[k] : 0; |
3225 | } |
3226 | result->data[i] = z; |
3227 | } |
3228 | /* most significant word of small */ |
3229 | x = small->data[i]; |
3230 | y = big->data[i]; |
3231 | z = 0; |
3232 | mswdigits = mpd_word_digits(x); |
3233 | for (k = 0; k < mswdigits; k++) { |
3234 | xbit = x % 10; |
3235 | x /= 10; |
3236 | ybit = y % 10; |
3237 | y /= 10; |
3238 | if (xbit > 1 || ybit > 1) { |
3239 | goto invalid_operation; |
3240 | } |
3241 | z += (xbit^ybit) ? mpd_pow10[k] : 0; |
3242 | } |
3243 | |
3244 | /* scan for digits > 1 and copy the rest of y */ |
3245 | for (; k < MPD_RDIGITS; k++) { |
3246 | ybit = y % 10; |
3247 | y /= 10; |
3248 | if (ybit > 1) { |
3249 | goto invalid_operation; |
3250 | } |
3251 | z += ybit*mpd_pow10[k]; |
3252 | } |
3253 | result->data[i++] = z; |
3254 | /* scan for digits > 1 and copy the rest of big */ |
3255 | for (; i < big->len; i++) { |
3256 | y = big->data[i]; |
3257 | for (k = 0; k < MPD_RDIGITS; k++) { |
3258 | ybit = y % 10; |
3259 | y /= 10; |
3260 | if (ybit > 1) { |
3261 | goto invalid_operation; |
3262 | } |
3263 | } |
3264 | result->data[i] = big->data[i]; |
3265 | } |
3266 | |
3267 | mpd_clear_flags(result); |
3268 | result->exp = 0; |
3269 | result->len = _mpd_real_size(result->data, big->len); |
3270 | mpd_qresize(result, result->len, status); |
3271 | mpd_setdigits(result); |
3272 | _mpd_cap(result, ctx); |
3273 | return; |
3274 | |
3275 | invalid_operation: |
3276 | mpd_seterror(result, MPD_Invalid_operation, status); |
3277 | } |
3278 | |
3279 | |
3280 | /******************************************************************************/ |
3281 | /* Arithmetic operations */ |
3282 | /******************************************************************************/ |
3283 | |
3284 | /* |
3285 | * The absolute value of a. If a is negative, the result is the same |
3286 | * as the result of the minus operation. Otherwise, the result is the |
3287 | * result of the plus operation. |
3288 | */ |
3289 | void |
3290 | mpd_qabs(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
3291 | uint32_t *status) |
3292 | { |
3293 | if (mpd_isspecial(a)) { |
3294 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
3295 | return; |
3296 | } |
3297 | } |
3298 | |
3299 | if (mpd_isnegative(a)) { |
3300 | mpd_qminus(result, a, ctx, status); |
3301 | } |
3302 | else { |
3303 | mpd_qplus(result, a, ctx, status); |
3304 | } |
3305 | } |
3306 | |
3307 | static inline void |
3308 | _mpd_ptrswap(const mpd_t **a, const mpd_t **b) |
3309 | { |
3310 | const mpd_t *t = *a; |
3311 | *a = *b; |
3312 | *b = t; |
3313 | } |
3314 | |
3315 | /* Add or subtract infinities. */ |
3316 | static void |
3317 | _mpd_qaddsub_inf(mpd_t *result, const mpd_t *a, const mpd_t *b, uint8_t sign_b, |
3318 | uint32_t *status) |
3319 | { |
3320 | if (mpd_isinfinite(a)) { |
3321 | if (mpd_sign(a) != sign_b && mpd_isinfinite(b)) { |
3322 | mpd_seterror(result, MPD_Invalid_operation, status); |
3323 | } |
3324 | else { |
3325 | mpd_setspecial(result, mpd_sign(a), MPD_INF); |
3326 | } |
3327 | return; |
3328 | } |
3329 | assert(mpd_isinfinite(b)); |
3330 | mpd_setspecial(result, sign_b, MPD_INF); |
3331 | } |
3332 | |
3333 | /* Add or subtract non-special numbers. */ |
3334 | static void |
3335 | _mpd_qaddsub(mpd_t *result, const mpd_t *a, const mpd_t *b, uint8_t sign_b, |
3336 | const mpd_context_t *ctx, uint32_t *status) |
3337 | { |
3338 | const mpd_t *big, *small; |
3339 | MPD_NEW_STATIC(big_aligned,0,0,0,0); |
3340 | MPD_NEW_CONST(tiny,0,0,1,1,1,1); |
3341 | mpd_uint_t carry; |
3342 | mpd_ssize_t newsize, shift; |
3343 | mpd_ssize_t exp, i; |
3344 | int swap = 0; |
3345 | |
3346 | |
3347 | /* compare exponents */ |
3348 | big = a; small = b; |
3349 | if (big->exp != small->exp) { |
3350 | if (small->exp > big->exp) { |
3351 | _mpd_ptrswap(&big, &small); |
3352 | swap++; |
3353 | } |
3354 | /* align the coefficients */ |
3355 | if (!mpd_iszerocoeff(big)) { |
3356 | exp = big->exp - 1; |
3357 | exp += (big->digits > ctx->prec) ? 0 : big->digits-ctx->prec-1; |
3358 | if (mpd_adjexp(small) < exp) { |
3359 | /* |
3360 | * Avoid huge shifts by substituting a value for small that is |
3361 | * guaranteed to produce the same results. |
3362 | * |
3363 | * adjexp(small) < exp if and only if: |
3364 | * |
3365 | * bdigits <= prec AND |
3366 | * bdigits+shift >= prec+2+sdigits AND |
3367 | * exp = bexp+bdigits-prec-2 |
3368 | * |
3369 | * 1234567000000000 -> bdigits + shift |
3370 | * ----------XX1234 -> sdigits |
3371 | * ----------X1 -> tiny-digits |
3372 | * |- prec -| |
3373 | * |
3374 | * OR |
3375 | * |
3376 | * bdigits > prec AND |
3377 | * shift > sdigits AND |
3378 | * exp = bexp-1 |
3379 | * |
3380 | * 1234567892100000 -> bdigits + shift |
3381 | * ----------XX1234 -> sdigits |
3382 | * ----------X1 -> tiny-digits |
3383 | * |- prec -| |
3384 | * |
3385 | * If tiny is zero, adding or subtracting is a no-op. |
3386 | * Otherwise, adding tiny generates a non-zero digit either |
3387 | * below the rounding digit or the least significant digit |
3388 | * of big. When subtracting, tiny is in the same position as |
3389 | * the carry that would be generated by subtracting sdigits. |
3390 | */ |
3391 | mpd_copy_flags(&tiny, small); |
3392 | tiny.exp = exp; |
3393 | tiny.digits = 1; |
3394 | tiny.len = 1; |
3395 | tiny.data[0] = mpd_iszerocoeff(small) ? 0 : 1; |
3396 | small = &tiny; |
3397 | } |
3398 | /* This cannot wrap: the difference is positive and <= maxprec */ |
3399 | shift = big->exp - small->exp; |
3400 | if (!mpd_qshiftl(&big_aligned, big, shift, status)) { |
3401 | mpd_seterror(result, MPD_Malloc_error, status); |
3402 | goto finish; |
3403 | } |
3404 | big = &big_aligned; |
3405 | } |
3406 | } |
3407 | result->exp = small->exp; |
3408 | |
3409 | |
3410 | /* compare length of coefficients */ |
3411 | if (big->len < small->len) { |
3412 | _mpd_ptrswap(&big, &small); |
3413 | swap++; |
3414 | } |
3415 | |
3416 | newsize = big->len; |
3417 | if (!mpd_qresize(result, newsize, status)) { |
3418 | goto finish; |
3419 | } |
3420 | |
3421 | if (mpd_sign(a) == sign_b) { |
3422 | |
3423 | carry = _mpd_baseadd(result->data, big->data, small->data, |
3424 | big->len, small->len); |
3425 | |
3426 | if (carry) { |
3427 | newsize = big->len + 1; |
3428 | if (!mpd_qresize(result, newsize, status)) { |
3429 | goto finish; |
3430 | } |
3431 | result->data[newsize-1] = carry; |
3432 | } |
3433 | |
3434 | result->len = newsize; |
3435 | mpd_set_flags(result, sign_b); |
3436 | } |
3437 | else { |
3438 | if (big->len == small->len) { |
3439 | for (i=big->len-1; i >= 0; --i) { |
3440 | if (big->data[i] != small->data[i]) { |
3441 | if (big->data[i] < small->data[i]) { |
3442 | _mpd_ptrswap(&big, &small); |
3443 | swap++; |
3444 | } |
3445 | break; |
3446 | } |
3447 | } |
3448 | } |
3449 | |
3450 | _mpd_basesub(result->data, big->data, small->data, |
3451 | big->len, small->len); |
3452 | newsize = _mpd_real_size(result->data, big->len); |
3453 | /* resize to smaller cannot fail */ |
3454 | (void)mpd_qresize(result, newsize, status); |
3455 | |
3456 | result->len = newsize; |
3457 | sign_b = (swap & 1) ? sign_b : mpd_sign(a); |
3458 | mpd_set_flags(result, sign_b); |
3459 | |
3460 | if (mpd_iszerocoeff(result)) { |
3461 | mpd_set_positive(result); |
3462 | if (ctx->round == MPD_ROUND_FLOOR) { |
3463 | mpd_set_negative(result); |
3464 | } |
3465 | } |
3466 | } |
3467 | |
3468 | mpd_setdigits(result); |
3469 | |
3470 | finish: |
3471 | mpd_del(&big_aligned); |
3472 | } |
3473 | |
3474 | /* Add a and b. No specials, no finalizing. */ |
3475 | static void |
3476 | _mpd_qadd(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3477 | const mpd_context_t *ctx, uint32_t *status) |
3478 | { |
3479 | _mpd_qaddsub(result, a, b, mpd_sign(b), ctx, status); |
3480 | } |
3481 | |
3482 | /* Subtract b from a. No specials, no finalizing. */ |
3483 | static void |
3484 | _mpd_qsub(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3485 | const mpd_context_t *ctx, uint32_t *status) |
3486 | { |
3487 | _mpd_qaddsub(result, a, b, !mpd_sign(b), ctx, status); |
3488 | } |
3489 | |
3490 | /* Add a and b. */ |
3491 | void |
3492 | mpd_qadd(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3493 | const mpd_context_t *ctx, uint32_t *status) |
3494 | { |
3495 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3496 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
3497 | return; |
3498 | } |
3499 | _mpd_qaddsub_inf(result, a, b, mpd_sign(b), status); |
3500 | return; |
3501 | } |
3502 | |
3503 | _mpd_qaddsub(result, a, b, mpd_sign(b), ctx, status); |
3504 | mpd_qfinalize(result, ctx, status); |
3505 | } |
3506 | |
3507 | /* Add a and b. Set NaN/Invalid_operation if the result is inexact. */ |
3508 | static void |
3509 | _mpd_qadd_exact(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3510 | const mpd_context_t *ctx, uint32_t *status) |
3511 | { |
3512 | uint32_t workstatus = 0; |
3513 | |
3514 | mpd_qadd(result, a, b, ctx, &workstatus); |
3515 | *status |= workstatus; |
3516 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
3517 | mpd_seterror(result, MPD_Invalid_operation, status); |
3518 | } |
3519 | } |
3520 | |
3521 | /* Subtract b from a. */ |
3522 | void |
3523 | mpd_qsub(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3524 | const mpd_context_t *ctx, uint32_t *status) |
3525 | { |
3526 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3527 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
3528 | return; |
3529 | } |
3530 | _mpd_qaddsub_inf(result, a, b, !mpd_sign(b), status); |
3531 | return; |
3532 | } |
3533 | |
3534 | _mpd_qaddsub(result, a, b, !mpd_sign(b), ctx, status); |
3535 | mpd_qfinalize(result, ctx, status); |
3536 | } |
3537 | |
3538 | /* Subtract b from a. Set NaN/Invalid_operation if the result is inexact. */ |
3539 | static void |
3540 | _mpd_qsub_exact(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3541 | const mpd_context_t *ctx, uint32_t *status) |
3542 | { |
3543 | uint32_t workstatus = 0; |
3544 | |
3545 | mpd_qsub(result, a, b, ctx, &workstatus); |
3546 | *status |= workstatus; |
3547 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
3548 | mpd_seterror(result, MPD_Invalid_operation, status); |
3549 | } |
3550 | } |
3551 | |
3552 | /* Add decimal and mpd_ssize_t. */ |
3553 | void |
3554 | mpd_qadd_ssize(mpd_t *result, const mpd_t *a, mpd_ssize_t b, |
3555 | const mpd_context_t *ctx, uint32_t *status) |
3556 | { |
3557 | mpd_context_t maxcontext; |
3558 | MPD_NEW_STATIC(bb,0,0,0,0); |
3559 | |
3560 | mpd_maxcontext(&maxcontext); |
3561 | mpd_qsset_ssize(&bb, b, &maxcontext, status); |
3562 | mpd_qadd(result, a, &bb, ctx, status); |
3563 | mpd_del(&bb); |
3564 | } |
3565 | |
3566 | /* Add decimal and mpd_uint_t. */ |
3567 | void |
3568 | mpd_qadd_uint(mpd_t *result, const mpd_t *a, mpd_uint_t b, |
3569 | const mpd_context_t *ctx, uint32_t *status) |
3570 | { |
3571 | mpd_context_t maxcontext; |
3572 | MPD_NEW_STATIC(bb,0,0,0,0); |
3573 | |
3574 | mpd_maxcontext(&maxcontext); |
3575 | mpd_qsset_uint(&bb, b, &maxcontext, status); |
3576 | mpd_qadd(result, a, &bb, ctx, status); |
3577 | mpd_del(&bb); |
3578 | } |
3579 | |
3580 | /* Subtract mpd_ssize_t from decimal. */ |
3581 | void |
3582 | mpd_qsub_ssize(mpd_t *result, const mpd_t *a, mpd_ssize_t b, |
3583 | const mpd_context_t *ctx, uint32_t *status) |
3584 | { |
3585 | mpd_context_t maxcontext; |
3586 | MPD_NEW_STATIC(bb,0,0,0,0); |
3587 | |
3588 | mpd_maxcontext(&maxcontext); |
3589 | mpd_qsset_ssize(&bb, b, &maxcontext, status); |
3590 | mpd_qsub(result, a, &bb, ctx, status); |
3591 | mpd_del(&bb); |
3592 | } |
3593 | |
3594 | /* Subtract mpd_uint_t from decimal. */ |
3595 | void |
3596 | mpd_qsub_uint(mpd_t *result, const mpd_t *a, mpd_uint_t b, |
3597 | const mpd_context_t *ctx, uint32_t *status) |
3598 | { |
3599 | mpd_context_t maxcontext; |
3600 | MPD_NEW_STATIC(bb,0,0,0,0); |
3601 | |
3602 | mpd_maxcontext(&maxcontext); |
3603 | mpd_qsset_uint(&bb, b, &maxcontext, status); |
3604 | mpd_qsub(result, a, &bb, ctx, status); |
3605 | mpd_del(&bb); |
3606 | } |
3607 | |
3608 | /* Add decimal and int32_t. */ |
3609 | void |
3610 | mpd_qadd_i32(mpd_t *result, const mpd_t *a, int32_t b, |
3611 | const mpd_context_t *ctx, uint32_t *status) |
3612 | { |
3613 | mpd_qadd_ssize(result, a, b, ctx, status); |
3614 | } |
3615 | |
3616 | /* Add decimal and uint32_t. */ |
3617 | void |
3618 | mpd_qadd_u32(mpd_t *result, const mpd_t *a, uint32_t b, |
3619 | const mpd_context_t *ctx, uint32_t *status) |
3620 | { |
3621 | mpd_qadd_uint(result, a, b, ctx, status); |
3622 | } |
3623 | |
3624 | #ifdef CONFIG_64 |
3625 | /* Add decimal and int64_t. */ |
3626 | void |
3627 | mpd_qadd_i64(mpd_t *result, const mpd_t *a, int64_t b, |
3628 | const mpd_context_t *ctx, uint32_t *status) |
3629 | { |
3630 | mpd_qadd_ssize(result, a, b, ctx, status); |
3631 | } |
3632 | |
3633 | /* Add decimal and uint64_t. */ |
3634 | void |
3635 | mpd_qadd_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
3636 | const mpd_context_t *ctx, uint32_t *status) |
3637 | { |
3638 | mpd_qadd_uint(result, a, b, ctx, status); |
3639 | } |
3640 | #elif !defined(LEGACY_COMPILER) |
3641 | /* Add decimal and int64_t. */ |
3642 | void |
3643 | mpd_qadd_i64(mpd_t *result, const mpd_t *a, int64_t b, |
3644 | const mpd_context_t *ctx, uint32_t *status) |
3645 | { |
3646 | mpd_context_t maxcontext; |
3647 | MPD_NEW_STATIC(bb,0,0,0,0); |
3648 | |
3649 | mpd_maxcontext(&maxcontext); |
3650 | mpd_qset_i64(&bb, b, &maxcontext, status); |
3651 | mpd_qadd(result, a, &bb, ctx, status); |
3652 | mpd_del(&bb); |
3653 | } |
3654 | |
3655 | /* Add decimal and uint64_t. */ |
3656 | void |
3657 | mpd_qadd_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
3658 | const mpd_context_t *ctx, uint32_t *status) |
3659 | { |
3660 | mpd_context_t maxcontext; |
3661 | MPD_NEW_STATIC(bb,0,0,0,0); |
3662 | |
3663 | mpd_maxcontext(&maxcontext); |
3664 | mpd_qset_u64(&bb, b, &maxcontext, status); |
3665 | mpd_qadd(result, a, &bb, ctx, status); |
3666 | mpd_del(&bb); |
3667 | } |
3668 | #endif |
3669 | |
3670 | /* Subtract int32_t from decimal. */ |
3671 | void |
3672 | mpd_qsub_i32(mpd_t *result, const mpd_t *a, int32_t b, |
3673 | const mpd_context_t *ctx, uint32_t *status) |
3674 | { |
3675 | mpd_qsub_ssize(result, a, b, ctx, status); |
3676 | } |
3677 | |
3678 | /* Subtract uint32_t from decimal. */ |
3679 | void |
3680 | mpd_qsub_u32(mpd_t *result, const mpd_t *a, uint32_t b, |
3681 | const mpd_context_t *ctx, uint32_t *status) |
3682 | { |
3683 | mpd_qsub_uint(result, a, b, ctx, status); |
3684 | } |
3685 | |
3686 | #ifdef CONFIG_64 |
3687 | /* Subtract int64_t from decimal. */ |
3688 | void |
3689 | mpd_qsub_i64(mpd_t *result, const mpd_t *a, int64_t b, |
3690 | const mpd_context_t *ctx, uint32_t *status) |
3691 | { |
3692 | mpd_qsub_ssize(result, a, b, ctx, status); |
3693 | } |
3694 | |
3695 | /* Subtract uint64_t from decimal. */ |
3696 | void |
3697 | mpd_qsub_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
3698 | const mpd_context_t *ctx, uint32_t *status) |
3699 | { |
3700 | mpd_qsub_uint(result, a, b, ctx, status); |
3701 | } |
3702 | #elif !defined(LEGACY_COMPILER) |
3703 | /* Subtract int64_t from decimal. */ |
3704 | void |
3705 | mpd_qsub_i64(mpd_t *result, const mpd_t *a, int64_t b, |
3706 | const mpd_context_t *ctx, uint32_t *status) |
3707 | { |
3708 | mpd_context_t maxcontext; |
3709 | MPD_NEW_STATIC(bb,0,0,0,0); |
3710 | |
3711 | mpd_maxcontext(&maxcontext); |
3712 | mpd_qset_i64(&bb, b, &maxcontext, status); |
3713 | mpd_qsub(result, a, &bb, ctx, status); |
3714 | mpd_del(&bb); |
3715 | } |
3716 | |
3717 | /* Subtract uint64_t from decimal. */ |
3718 | void |
3719 | mpd_qsub_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
3720 | const mpd_context_t *ctx, uint32_t *status) |
3721 | { |
3722 | mpd_context_t maxcontext; |
3723 | MPD_NEW_STATIC(bb,0,0,0,0); |
3724 | |
3725 | mpd_maxcontext(&maxcontext); |
3726 | mpd_qset_u64(&bb, b, &maxcontext, status); |
3727 | mpd_qsub(result, a, &bb, ctx, status); |
3728 | mpd_del(&bb); |
3729 | } |
3730 | #endif |
3731 | |
3732 | |
3733 | /* Divide infinities. */ |
3734 | static void |
3735 | _mpd_qdiv_inf(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3736 | const mpd_context_t *ctx, uint32_t *status) |
3737 | { |
3738 | if (mpd_isinfinite(a)) { |
3739 | if (mpd_isinfinite(b)) { |
3740 | mpd_seterror(result, MPD_Invalid_operation, status); |
3741 | return; |
3742 | } |
3743 | mpd_setspecial(result, mpd_sign(a)^mpd_sign(b), MPD_INF); |
3744 | return; |
3745 | } |
3746 | assert(mpd_isinfinite(b)); |
3747 | _settriple(result, mpd_sign(a)^mpd_sign(b), 0, mpd_etiny(ctx)); |
3748 | *status |= MPD_Clamped; |
3749 | } |
3750 | |
3751 | enum {NO_IDEAL_EXP, SET_IDEAL_EXP}; |
3752 | /* Divide a by b. */ |
3753 | static void |
3754 | _mpd_qdiv(int action, mpd_t *q, const mpd_t *a, const mpd_t *b, |
3755 | const mpd_context_t *ctx, uint32_t *status) |
3756 | { |
3757 | MPD_NEW_STATIC(aligned,0,0,0,0); |
3758 | mpd_uint_t ld; |
3759 | mpd_ssize_t shift, exp, tz; |
3760 | mpd_ssize_t newsize; |
3761 | mpd_ssize_t ideal_exp; |
3762 | mpd_uint_t rem; |
3763 | uint8_t sign_a = mpd_sign(a); |
3764 | uint8_t sign_b = mpd_sign(b); |
3765 | |
3766 | |
3767 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3768 | if (mpd_qcheck_nans(q, a, b, ctx, status)) { |
3769 | return; |
3770 | } |
3771 | _mpd_qdiv_inf(q, a, b, ctx, status); |
3772 | return; |
3773 | } |
3774 | if (mpd_iszerocoeff(b)) { |
3775 | if (mpd_iszerocoeff(a)) { |
3776 | mpd_seterror(q, MPD_Division_undefined, status); |
3777 | } |
3778 | else { |
3779 | mpd_setspecial(q, sign_a^sign_b, MPD_INF); |
3780 | *status |= MPD_Division_by_zero; |
3781 | } |
3782 | return; |
3783 | } |
3784 | if (mpd_iszerocoeff(a)) { |
3785 | exp = a->exp - b->exp; |
3786 | _settriple(q, sign_a^sign_b, 0, exp); |
3787 | mpd_qfinalize(q, ctx, status); |
3788 | return; |
3789 | } |
3790 | |
3791 | shift = (b->digits - a->digits) + ctx->prec + 1; |
3792 | ideal_exp = a->exp - b->exp; |
3793 | exp = ideal_exp - shift; |
3794 | if (shift > 0) { |
3795 | if (!mpd_qshiftl(&aligned, a, shift, status)) { |
3796 | mpd_seterror(q, MPD_Malloc_error, status); |
3797 | goto finish; |
3798 | } |
3799 | a = &aligned; |
3800 | } |
3801 | else if (shift < 0) { |
3802 | shift = -shift; |
3803 | if (!mpd_qshiftl(&aligned, b, shift, status)) { |
3804 | mpd_seterror(q, MPD_Malloc_error, status); |
3805 | goto finish; |
3806 | } |
3807 | b = &aligned; |
3808 | } |
3809 | |
3810 | |
3811 | newsize = a->len - b->len + 1; |
3812 | if ((q != b && q != a) || (q == b && newsize > b->len)) { |
3813 | if (!mpd_qresize(q, newsize, status)) { |
3814 | mpd_seterror(q, MPD_Malloc_error, status); |
3815 | goto finish; |
3816 | } |
3817 | } |
3818 | |
3819 | |
3820 | if (b->len == 1) { |
3821 | rem = _mpd_shortdiv(q->data, a->data, a->len, b->data[0]); |
3822 | } |
3823 | else if (b->len <= MPD_NEWTONDIV_CUTOFF) { |
3824 | int ret = _mpd_basedivmod(q->data, NULL, a->data, b->data, |
3825 | a->len, b->len); |
3826 | if (ret < 0) { |
3827 | mpd_seterror(q, MPD_Malloc_error, status); |
3828 | goto finish; |
3829 | } |
3830 | rem = ret; |
3831 | } |
3832 | else { |
3833 | MPD_NEW_STATIC(r,0,0,0,0); |
3834 | _mpd_base_ndivmod(q, &r, a, b, status); |
3835 | if (mpd_isspecial(q) || mpd_isspecial(&r)) { |
3836 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
3837 | mpd_del(&r); |
3838 | goto finish; |
3839 | } |
3840 | rem = !mpd_iszerocoeff(&r); |
3841 | mpd_del(&r); |
3842 | newsize = q->len; |
3843 | } |
3844 | |
3845 | newsize = _mpd_real_size(q->data, newsize); |
3846 | /* resize to smaller cannot fail */ |
3847 | mpd_qresize(q, newsize, status); |
3848 | mpd_set_flags(q, sign_a^sign_b); |
3849 | q->len = newsize; |
3850 | mpd_setdigits(q); |
3851 | |
3852 | shift = ideal_exp - exp; |
3853 | if (rem) { |
3854 | ld = mpd_lsd(q->data[0]); |
3855 | if (ld == 0 || ld == 5) { |
3856 | q->data[0] += 1; |
3857 | } |
3858 | } |
3859 | else if (action == SET_IDEAL_EXP && shift > 0) { |
3860 | tz = mpd_trail_zeros(q); |
3861 | shift = (tz > shift) ? shift : tz; |
3862 | mpd_qshiftr_inplace(q, shift); |
3863 | exp += shift; |
3864 | } |
3865 | |
3866 | q->exp = exp; |
3867 | |
3868 | |
3869 | finish: |
3870 | mpd_del(&aligned); |
3871 | mpd_qfinalize(q, ctx, status); |
3872 | } |
3873 | |
3874 | /* Divide a by b. */ |
3875 | void |
3876 | mpd_qdiv(mpd_t *q, const mpd_t *a, const mpd_t *b, |
3877 | const mpd_context_t *ctx, uint32_t *status) |
3878 | { |
3879 | MPD_NEW_STATIC(aa,0,0,0,0); |
3880 | MPD_NEW_STATIC(bb,0,0,0,0); |
3881 | uint32_t xstatus = 0; |
3882 | |
3883 | if (q == a) { |
3884 | if (!mpd_qcopy(&aa, a, status)) { |
3885 | mpd_seterror(q, MPD_Malloc_error, status); |
3886 | goto out; |
3887 | } |
3888 | a = &aa; |
3889 | } |
3890 | |
3891 | if (q == b) { |
3892 | if (!mpd_qcopy(&bb, b, status)) { |
3893 | mpd_seterror(q, MPD_Malloc_error, status); |
3894 | goto out; |
3895 | } |
3896 | b = &bb; |
3897 | } |
3898 | |
3899 | _mpd_qdiv(SET_IDEAL_EXP, q, a, b, ctx, &xstatus); |
3900 | |
3901 | if (xstatus & (MPD_Malloc_error|MPD_Division_impossible)) { |
3902 | /* Inexact quotients (the usual case) fill the entire context precision, |
3903 | * which can lead to the above errors for very high precisions. Retry |
3904 | * the operation with a lower precision in case the result is exact. |
3905 | * |
3906 | * We need an upper bound for the number of digits of a_coeff / b_coeff |
3907 | * when the result is exact. If a_coeff' * 1 / b_coeff' is in lowest |
3908 | * terms, then maxdigits(a_coeff') + maxdigits(1 / b_coeff') is a suitable |
3909 | * bound. |
3910 | * |
3911 | * 1 / b_coeff' is exact iff b_coeff' exclusively has prime factors 2 or 5. |
3912 | * The largest amount of digits is generated if b_coeff' is a power of 2 or |
3913 | * a power of 5 and is less than or equal to log5(b_coeff') <= log2(b_coeff'). |
3914 | * |
3915 | * We arrive at a total upper bound: |
3916 | * |
3917 | * maxdigits(a_coeff') + maxdigits(1 / b_coeff') <= |
3918 | * log10(a_coeff) + log2(b_coeff) = |
3919 | * log10(a_coeff) + log10(b_coeff) / log10(2) <= |
3920 | * a->digits + b->digits * 4; |
3921 | */ |
3922 | mpd_context_t workctx = *ctx; |
3923 | uint32_t ystatus = 0; |
3924 | |
3925 | workctx.prec = a->digits + b->digits * 4; |
3926 | if (workctx.prec >= ctx->prec) { |
3927 | *status |= (xstatus&MPD_Errors); |
3928 | goto out; /* No point in retrying, keep the original error. */ |
3929 | } |
3930 | |
3931 | _mpd_qdiv(SET_IDEAL_EXP, q, a, b, &workctx, &ystatus); |
3932 | if (ystatus != 0) { |
3933 | ystatus = *status | ((ystatus|xstatus)&MPD_Errors); |
3934 | mpd_seterror(q, ystatus, status); |
3935 | } |
3936 | } |
3937 | else { |
3938 | *status |= xstatus; |
3939 | } |
3940 | |
3941 | |
3942 | out: |
3943 | mpd_del(&aa); |
3944 | mpd_del(&bb); |
3945 | } |
3946 | |
3947 | /* Internal function. */ |
3948 | static void |
3949 | _mpd_qdivmod(mpd_t *q, mpd_t *r, const mpd_t *a, const mpd_t *b, |
3950 | const mpd_context_t *ctx, uint32_t *status) |
3951 | { |
3952 | MPD_NEW_STATIC(aligned,0,0,0,0); |
3953 | mpd_ssize_t qsize, rsize; |
3954 | mpd_ssize_t ideal_exp, expdiff, shift; |
3955 | uint8_t sign_a = mpd_sign(a); |
3956 | uint8_t sign_ab = mpd_sign(a)^mpd_sign(b); |
3957 | |
3958 | |
3959 | ideal_exp = (a->exp > b->exp) ? b->exp : a->exp; |
3960 | if (mpd_iszerocoeff(a)) { |
3961 | if (!mpd_qcopy(r, a, status)) { |
3962 | goto nanresult; /* GCOV_NOT_REACHED */ |
3963 | } |
3964 | r->exp = ideal_exp; |
3965 | _settriple(q, sign_ab, 0, 0); |
3966 | return; |
3967 | } |
3968 | |
3969 | expdiff = mpd_adjexp(a) - mpd_adjexp(b); |
3970 | if (expdiff < 0) { |
3971 | if (a->exp > b->exp) { |
3972 | /* positive and less than b->digits - a->digits */ |
3973 | shift = a->exp - b->exp; |
3974 | if (!mpd_qshiftl(r, a, shift, status)) { |
3975 | goto nanresult; |
3976 | } |
3977 | r->exp = ideal_exp; |
3978 | } |
3979 | else { |
3980 | if (!mpd_qcopy(r, a, status)) { |
3981 | goto nanresult; |
3982 | } |
3983 | } |
3984 | _settriple(q, sign_ab, 0, 0); |
3985 | return; |
3986 | } |
3987 | if (expdiff > ctx->prec) { |
3988 | *status |= MPD_Division_impossible; |
3989 | goto nanresult; |
3990 | } |
3991 | |
3992 | |
3993 | /* |
3994 | * At this point we have: |
3995 | * (1) 0 <= a->exp + a->digits - b->exp - b->digits <= prec |
3996 | * (2) a->exp - b->exp >= b->digits - a->digits |
3997 | * (3) a->exp - b->exp <= prec + b->digits - a->digits |
3998 | */ |
3999 | if (a->exp != b->exp) { |
4000 | shift = a->exp - b->exp; |
4001 | if (shift > 0) { |
4002 | /* by (3), after the shift a->digits <= prec + b->digits */ |
4003 | if (!mpd_qshiftl(&aligned, a, shift, status)) { |
4004 | goto nanresult; |
4005 | } |
4006 | a = &aligned; |
4007 | } |
4008 | else { |
4009 | shift = -shift; |
4010 | /* by (2), after the shift b->digits <= a->digits */ |
4011 | if (!mpd_qshiftl(&aligned, b, shift, status)) { |
4012 | goto nanresult; |
4013 | } |
4014 | b = &aligned; |
4015 | } |
4016 | } |
4017 | |
4018 | |
4019 | qsize = a->len - b->len + 1; |
4020 | if (!(q == a && qsize < a->len) && !(q == b && qsize < b->len)) { |
4021 | if (!mpd_qresize(q, qsize, status)) { |
4022 | goto nanresult; |
4023 | } |
4024 | } |
4025 | |
4026 | rsize = b->len; |
4027 | if (!(r == a && rsize < a->len)) { |
4028 | if (!mpd_qresize(r, rsize, status)) { |
4029 | goto nanresult; |
4030 | } |
4031 | } |
4032 | |
4033 | if (b->len == 1) { |
4034 | assert(b->data[0] != 0); /* annotation for scan-build */ |
4035 | if (a->len == 1) { |
4036 | _mpd_div_word(&q->data[0], &r->data[0], a->data[0], b->data[0]); |
4037 | } |
4038 | else { |
4039 | r->data[0] = _mpd_shortdiv(q->data, a->data, a->len, b->data[0]); |
4040 | } |
4041 | } |
4042 | else if (b->len <= MPD_NEWTONDIV_CUTOFF) { |
4043 | int ret; |
4044 | ret = _mpd_basedivmod(q->data, r->data, a->data, b->data, |
4045 | a->len, b->len); |
4046 | if (ret == -1) { |
4047 | *status |= MPD_Malloc_error; |
4048 | goto nanresult; |
4049 | } |
4050 | } |
4051 | else { |
4052 | _mpd_base_ndivmod(q, r, a, b, status); |
4053 | if (mpd_isspecial(q) || mpd_isspecial(r)) { |
4054 | goto nanresult; |
4055 | } |
4056 | qsize = q->len; |
4057 | rsize = r->len; |
4058 | } |
4059 | |
4060 | qsize = _mpd_real_size(q->data, qsize); |
4061 | /* resize to smaller cannot fail */ |
4062 | mpd_qresize(q, qsize, status); |
4063 | q->len = qsize; |
4064 | mpd_setdigits(q); |
4065 | mpd_set_flags(q, sign_ab); |
4066 | q->exp = 0; |
4067 | if (q->digits > ctx->prec) { |
4068 | *status |= MPD_Division_impossible; |
4069 | goto nanresult; |
4070 | } |
4071 | |
4072 | rsize = _mpd_real_size(r->data, rsize); |
4073 | /* resize to smaller cannot fail */ |
4074 | mpd_qresize(r, rsize, status); |
4075 | r->len = rsize; |
4076 | mpd_setdigits(r); |
4077 | mpd_set_flags(r, sign_a); |
4078 | r->exp = ideal_exp; |
4079 | |
4080 | out: |
4081 | mpd_del(&aligned); |
4082 | return; |
4083 | |
4084 | nanresult: |
4085 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
4086 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
4087 | goto out; |
4088 | } |
4089 | |
4090 | /* Integer division with remainder. */ |
4091 | void |
4092 | mpd_qdivmod(mpd_t *q, mpd_t *r, const mpd_t *a, const mpd_t *b, |
4093 | const mpd_context_t *ctx, uint32_t *status) |
4094 | { |
4095 | uint8_t sign = mpd_sign(a)^mpd_sign(b); |
4096 | |
4097 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
4098 | if (mpd_qcheck_nans(q, a, b, ctx, status)) { |
4099 | mpd_qcopy(r, q, status); |
4100 | return; |
4101 | } |
4102 | if (mpd_isinfinite(a)) { |
4103 | if (mpd_isinfinite(b)) { |
4104 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
4105 | } |
4106 | else { |
4107 | mpd_setspecial(q, sign, MPD_INF); |
4108 | } |
4109 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
4110 | *status |= MPD_Invalid_operation; |
4111 | return; |
4112 | } |
4113 | if (mpd_isinfinite(b)) { |
4114 | if (!mpd_qcopy(r, a, status)) { |
4115 | mpd_seterror(q, MPD_Malloc_error, status); |
4116 | return; |
4117 | } |
4118 | mpd_qfinalize(r, ctx, status); |
4119 | _settriple(q, sign, 0, 0); |
4120 | return; |
4121 | } |
4122 | /* debug */ |
4123 | abort(); /* GCOV_NOT_REACHED */ |
4124 | } |
4125 | if (mpd_iszerocoeff(b)) { |
4126 | if (mpd_iszerocoeff(a)) { |
4127 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
4128 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
4129 | *status |= MPD_Division_undefined; |
4130 | } |
4131 | else { |
4132 | mpd_setspecial(q, sign, MPD_INF); |
4133 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
4134 | *status |= (MPD_Division_by_zero|MPD_Invalid_operation); |
4135 | } |
4136 | return; |
4137 | } |
4138 | |
4139 | _mpd_qdivmod(q, r, a, b, ctx, status); |
4140 | mpd_qfinalize(q, ctx, status); |
4141 | mpd_qfinalize(r, ctx, status); |
4142 | } |
4143 | |
4144 | void |
4145 | mpd_qdivint(mpd_t *q, const mpd_t *a, const mpd_t *b, |
4146 | const mpd_context_t *ctx, uint32_t *status) |
4147 | { |
4148 | MPD_NEW_STATIC(r,0,0,0,0); |
4149 | uint8_t sign = mpd_sign(a)^mpd_sign(b); |
4150 | |
4151 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
4152 | if (mpd_qcheck_nans(q, a, b, ctx, status)) { |
4153 | return; |
4154 | } |
4155 | if (mpd_isinfinite(a) && mpd_isinfinite(b)) { |
4156 | mpd_seterror(q, MPD_Invalid_operation, status); |
4157 | return; |
4158 | } |
4159 | if (mpd_isinfinite(a)) { |
4160 | mpd_setspecial(q, sign, MPD_INF); |
4161 | return; |
4162 | } |
4163 | if (mpd_isinfinite(b)) { |
4164 | _settriple(q, sign, 0, 0); |
4165 | return; |
4166 | } |
4167 | /* debug */ |
4168 | abort(); /* GCOV_NOT_REACHED */ |
4169 | } |
4170 | if (mpd_iszerocoeff(b)) { |
4171 | if (mpd_iszerocoeff(a)) { |
4172 | mpd_seterror(q, MPD_Division_undefined, status); |
4173 | } |
4174 | else { |
4175 | mpd_setspecial(q, sign, MPD_INF); |
4176 | *status |= MPD_Division_by_zero; |
4177 | } |
4178 | return; |
4179 | } |
4180 | |
4181 | |
4182 | _mpd_qdivmod(q, &r, a, b, ctx, status); |
4183 | mpd_del(&r); |
4184 | mpd_qfinalize(q, ctx, status); |
4185 | } |
4186 | |
4187 | /* Divide decimal by mpd_ssize_t. */ |
4188 | void |
4189 | mpd_qdiv_ssize(mpd_t *result, const mpd_t *a, mpd_ssize_t b, |
4190 | const mpd_context_t *ctx, uint32_t *status) |
4191 | { |
4192 | mpd_context_t maxcontext; |
4193 | MPD_NEW_STATIC(bb,0,0,0,0); |
4194 | |
4195 | mpd_maxcontext(&maxcontext); |
4196 | mpd_qsset_ssize(&bb, b, &maxcontext, status); |
4197 | mpd_qdiv(result, a, &bb, ctx, status); |
4198 | mpd_del(&bb); |
4199 | } |
4200 | |
4201 | /* Divide decimal by mpd_uint_t. */ |
4202 | void |
4203 | mpd_qdiv_uint(mpd_t *result, const mpd_t *a, mpd_uint_t b, |
4204 | const mpd_context_t *ctx, uint32_t *status) |
4205 | { |
4206 | mpd_context_t maxcontext; |
4207 | MPD_NEW_STATIC(bb,0,0,0,0); |
4208 | |
4209 | mpd_maxcontext(&maxcontext); |
4210 | mpd_qsset_uint(&bb, b, &maxcontext, status); |
4211 | mpd_qdiv(result, a, &bb, ctx, status); |
4212 | mpd_del(&bb); |
4213 | } |
4214 | |
4215 | /* Divide decimal by int32_t. */ |
4216 | void |
4217 | mpd_qdiv_i32(mpd_t *result, const mpd_t *a, int32_t b, |
4218 | const mpd_context_t *ctx, uint32_t *status) |
4219 | { |
4220 | mpd_qdiv_ssize(result, a, b, ctx, status); |
4221 | } |
4222 | |
4223 | /* Divide decimal by uint32_t. */ |
4224 | void |
4225 | mpd_qdiv_u32(mpd_t *result, const mpd_t *a, uint32_t b, |
4226 | const mpd_context_t *ctx, uint32_t *status) |
4227 | { |
4228 | mpd_qdiv_uint(result, a, b, ctx, status); |
4229 | } |
4230 | |
4231 | #ifdef CONFIG_64 |
4232 | /* Divide decimal by int64_t. */ |
4233 | void |
4234 | mpd_qdiv_i64(mpd_t *result, const mpd_t *a, int64_t b, |
4235 | const mpd_context_t *ctx, uint32_t *status) |
4236 | { |
4237 | mpd_qdiv_ssize(result, a, b, ctx, status); |
4238 | } |
4239 | |
4240 | /* Divide decimal by uint64_t. */ |
4241 | void |
4242 | mpd_qdiv_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
4243 | const mpd_context_t *ctx, uint32_t *status) |
4244 | { |
4245 | mpd_qdiv_uint(result, a, b, ctx, status); |
4246 | } |
4247 | #elif !defined(LEGACY_COMPILER) |
4248 | /* Divide decimal by int64_t. */ |
4249 | void |
4250 | mpd_qdiv_i64(mpd_t *result, const mpd_t *a, int64_t b, |
4251 | const mpd_context_t *ctx, uint32_t *status) |
4252 | { |
4253 | mpd_context_t maxcontext; |
4254 | MPD_NEW_STATIC(bb,0,0,0,0); |
4255 | |
4256 | mpd_maxcontext(&maxcontext); |
4257 | mpd_qset_i64(&bb, b, &maxcontext, status); |
4258 | mpd_qdiv(result, a, &bb, ctx, status); |
4259 | mpd_del(&bb); |
4260 | } |
4261 | |
4262 | /* Divide decimal by uint64_t. */ |
4263 | void |
4264 | mpd_qdiv_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
4265 | const mpd_context_t *ctx, uint32_t *status) |
4266 | { |
4267 | mpd_context_t maxcontext; |
4268 | MPD_NEW_STATIC(bb,0,0,0,0); |
4269 | |
4270 | mpd_maxcontext(&maxcontext); |
4271 | mpd_qset_u64(&bb, b, &maxcontext, status); |
4272 | mpd_qdiv(result, a, &bb, ctx, status); |
4273 | mpd_del(&bb); |
4274 | } |
4275 | #endif |
4276 | |
4277 | /* Pad the result with trailing zeros if it has fewer digits than prec. */ |
4278 | static void |
4279 | _mpd_zeropad(mpd_t *result, const mpd_context_t *ctx, uint32_t *status) |
4280 | { |
4281 | if (!mpd_isspecial(result) && !mpd_iszero(result) && |
4282 | result->digits < ctx->prec) { |
4283 | mpd_ssize_t shift = ctx->prec - result->digits; |
4284 | mpd_qshiftl(result, result, shift, status); |
4285 | result->exp -= shift; |
4286 | } |
4287 | } |
4288 | |
4289 | /* Check if the result is guaranteed to be one. */ |
4290 | static int |
4291 | _mpd_qexp_check_one(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
4292 | uint32_t *status) |
4293 | { |
4294 | MPD_NEW_CONST(lim,0,-(ctx->prec+1),1,1,1,9); |
4295 | MPD_NEW_SHARED(aa, a); |
4296 | |
4297 | mpd_set_positive(&aa); |
4298 | |
4299 | /* abs(a) <= 9 * 10**(-prec-1) */ |
4300 | if (_mpd_cmp(&aa, &lim) <= 0) { |
4301 | _settriple(result, 0, 1, 0); |
4302 | *status |= MPD_Rounded|MPD_Inexact; |
4303 | return 1; |
4304 | } |
4305 | |
4306 | return 0; |
4307 | } |
4308 | |
4309 | /* |
4310 | * Get the number of iterations for the Horner scheme in _mpd_qexp(). |
4311 | */ |
4312 | static inline mpd_ssize_t |
4313 | _mpd_get_exp_iterations(const mpd_t *r, mpd_ssize_t p) |
4314 | { |
4315 | mpd_ssize_t log10pbyr; /* lower bound for log10(p / abs(r)) */ |
4316 | mpd_ssize_t n; |
4317 | |
4318 | assert(p >= 10); |
4319 | assert(!mpd_iszero(r)); |
4320 | assert(-p < mpd_adjexp(r) && mpd_adjexp(r) <= -1); |
4321 | |
4322 | #ifdef CONFIG_64 |
4323 | if (p > (mpd_ssize_t)(1ULL<<52)) { |
4324 | return MPD_SSIZE_MAX; |
4325 | } |
4326 | #endif |
4327 | |
4328 | /* |
4329 | * Lower bound for log10(p / abs(r)): adjexp(p) - (adjexp(r) + 1) |
4330 | * At this point (for CONFIG_64, CONFIG_32 is not problematic): |
4331 | * 1) 10 <= p <= 2**52 |
4332 | * 2) -p < adjexp(r) <= -1 |
4333 | * 3) 1 <= log10pbyr <= 2**52 + 14 |
4334 | */ |
4335 | log10pbyr = (mpd_word_digits(p)-1) - (mpd_adjexp(r)+1); |
4336 | |
4337 | /* |
4338 | * The numerator in the paper is 1.435 * p - 1.182, calculated |
4339 | * exactly. We compensate for rounding errors by using 1.43503. |
4340 | * ACL2 proofs: |
4341 | * 1) exp-iter-approx-lower-bound: The term below evaluated |
4342 | * in 53-bit floating point arithmetic is greater than or |
4343 | * equal to the exact term used in the paper. |
4344 | * 2) exp-iter-approx-upper-bound: The term below is less than |
4345 | * or equal to 3/2 * p <= 3/2 * 2**52. |
4346 | */ |
4347 | n = (mpd_ssize_t)ceil((1.43503*(double)p - 1.182) / (double)log10pbyr); |
4348 | return n >= 3 ? n : 3; |
4349 | } |
4350 | |
4351 | /* |
4352 | * Internal function, specials have been dealt with. Apart from Overflow |
4353 | * and Underflow, two cases must be considered for the error of the result: |
4354 | * |
4355 | * 1) abs(a) <= 9 * 10**(-prec-1) ==> result == 1 |
4356 | * |
4357 | * Absolute error: abs(1 - e**x) < 10**(-prec) |
4358 | * ------------------------------------------- |
4359 | * |
4360 | * 2) abs(a) > 9 * 10**(-prec-1) |
4361 | * |
4362 | * Relative error: abs(result - e**x) < 0.5 * 10**(-prec) * e**x |
4363 | * ------------------------------------------------------------- |
4364 | * |
4365 | * The algorithm is from Hull&Abrham, Variable Precision Exponential Function, |
4366 | * ACM Transactions on Mathematical Software, Vol. 12, No. 2, June 1986. |
4367 | * |
4368 | * Main differences: |
4369 | * |
4370 | * - The number of iterations for the Horner scheme is calculated using |
4371 | * 53-bit floating point arithmetic. |
4372 | * |
4373 | * - In the error analysis for ER (relative error accumulated in the |
4374 | * evaluation of the truncated series) the reduced operand r may |
4375 | * have any number of digits. |
4376 | * ACL2 proof: exponent-relative-error |
4377 | * |
4378 | * - The analysis for early abortion has been adapted for the mpd_t |
4379 | * ranges. |
4380 | */ |
4381 | static void |
4382 | _mpd_qexp(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
4383 | uint32_t *status) |
4384 | { |
4385 | mpd_context_t workctx; |
4386 | MPD_NEW_STATIC(tmp,0,0,0,0); |
4387 | MPD_NEW_STATIC(sum,0,0,0,0); |
4388 | MPD_NEW_CONST(word,0,0,1,1,1,1); |
4389 | mpd_ssize_t j, n, t; |
4390 | |
4391 | assert(!mpd_isspecial(a)); |
4392 | |
4393 | if (mpd_iszerocoeff(a)) { |
4394 | _settriple(result, MPD_POS, 1, 0); |
4395 | return; |
4396 | } |
4397 | |
4398 | /* |
4399 | * We are calculating e^x = e^(r*10^t) = (e^r)^(10^t), where abs(r) < 1 and t >= 0. |
4400 | * |
4401 | * If t > 0, we have: |
4402 | * |
4403 | * (1) 0.1 <= r < 1, so e^0.1 <= e^r. If t > MAX_T, overflow occurs: |
4404 | * |
4405 | * MAX-EMAX+1 < log10(e^(0.1*10*t)) <= log10(e^(r*10^t)) < adjexp(e^(r*10^t))+1 |
4406 | * |
4407 | * (2) -1 < r <= -0.1, so e^r <= e^-0.1. If t > MAX_T, underflow occurs: |
4408 | * |
4409 | * adjexp(e^(r*10^t)) <= log10(e^(r*10^t)) <= log10(e^(-0.1*10^t)) < MIN-ETINY |
4410 | */ |
4411 | #if defined(CONFIG_64) |
4412 | #define MPD_EXP_MAX_T 19 |
4413 | #elif defined(CONFIG_32) |
4414 | #define MPD_EXP_MAX_T 10 |
4415 | #endif |
4416 | t = a->digits + a->exp; |
4417 | t = (t > 0) ? t : 0; |
4418 | if (t > MPD_EXP_MAX_T) { |
4419 | if (mpd_ispositive(a)) { |
4420 | mpd_setspecial(result, MPD_POS, MPD_INF); |
4421 | *status |= MPD_Overflow|MPD_Inexact|MPD_Rounded; |
4422 | } |
4423 | else { |
4424 | _settriple(result, MPD_POS, 0, mpd_etiny(ctx)); |
4425 | *status |= (MPD_Inexact|MPD_Rounded|MPD_Subnormal| |
4426 | MPD_Underflow|MPD_Clamped); |
4427 | } |
4428 | return; |
4429 | } |
4430 | |
4431 | /* abs(a) <= 9 * 10**(-prec-1) */ |
4432 | if (_mpd_qexp_check_one(result, a, ctx, status)) { |
4433 | return; |
4434 | } |
4435 | |
4436 | mpd_maxcontext(&workctx); |
4437 | workctx.prec = ctx->prec + t + 2; |
4438 | workctx.prec = (workctx.prec < 10) ? 10 : workctx.prec; |
4439 | workctx.round = MPD_ROUND_HALF_EVEN; |
4440 | |
4441 | if (!mpd_qcopy(result, a, status)) { |
4442 | return; |
4443 | } |
4444 | result->exp -= t; |
4445 | |
4446 | /* |
4447 | * At this point: |
4448 | * 1) 9 * 10**(-prec-1) < abs(a) |
4449 | * 2) 9 * 10**(-prec-t-1) < abs(r) |
4450 | * 3) log10(9) - prec - t - 1 < log10(abs(r)) < adjexp(abs(r)) + 1 |
4451 | * 4) - prec - t - 2 < adjexp(abs(r)) <= -1 |
4452 | */ |
4453 | n = _mpd_get_exp_iterations(result, workctx.prec); |
4454 | if (n == MPD_SSIZE_MAX) { |
4455 | mpd_seterror(result, MPD_Invalid_operation, status); /* GCOV_UNLIKELY */ |
4456 | return; /* GCOV_UNLIKELY */ |
4457 | } |
4458 | |
4459 | _settriple(&sum, MPD_POS, 1, 0); |
4460 | |
4461 | for (j = n-1; j >= 1; j--) { |
4462 | word.data[0] = j; |
4463 | mpd_setdigits(&word); |
4464 | mpd_qdiv(&tmp, result, &word, &workctx, &workctx.status); |
4465 | mpd_qfma(&sum, &sum, &tmp, &one, &workctx, &workctx.status); |
4466 | } |
4467 | |
4468 | #ifdef CONFIG_64 |
4469 | _mpd_qpow_uint(result, &sum, mpd_pow10[t], MPD_POS, &workctx, status); |
4470 | #else |
4471 | if (t <= MPD_MAX_POW10) { |
4472 | _mpd_qpow_uint(result, &sum, mpd_pow10[t], MPD_POS, &workctx, status); |
4473 | } |
4474 | else { |
4475 | t -= MPD_MAX_POW10; |
4476 | _mpd_qpow_uint(&tmp, &sum, mpd_pow10[MPD_MAX_POW10], MPD_POS, |
4477 | &workctx, status); |
4478 | _mpd_qpow_uint(result, &tmp, mpd_pow10[t], MPD_POS, &workctx, status); |
4479 | } |
4480 | #endif |
4481 | |
4482 | mpd_del(&tmp); |
4483 | mpd_del(&sum); |
4484 | *status |= (workctx.status&MPD_Errors); |
4485 | *status |= (MPD_Inexact|MPD_Rounded); |
4486 | } |
4487 | |
4488 | /* exp(a) */ |
4489 | void |
4490 | mpd_qexp(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
4491 | uint32_t *status) |
4492 | { |
4493 | mpd_context_t workctx; |
4494 | |
4495 | if (mpd_isspecial(a)) { |
4496 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
4497 | return; |
4498 | } |
4499 | if (mpd_isnegative(a)) { |
4500 | _settriple(result, MPD_POS, 0, 0); |
4501 | } |
4502 | else { |
4503 | mpd_setspecial(result, MPD_POS, MPD_INF); |
4504 | } |
4505 | return; |
4506 | } |
4507 | if (mpd_iszerocoeff(a)) { |
4508 | _settriple(result, MPD_POS, 1, 0); |
4509 | return; |
4510 | } |
4511 | |
4512 | workctx = *ctx; |
4513 | workctx.round = MPD_ROUND_HALF_EVEN; |
4514 | |
4515 | if (ctx->allcr) { |
4516 | MPD_NEW_STATIC(t1, 0,0,0,0); |
4517 | MPD_NEW_STATIC(t2, 0,0,0,0); |
4518 | MPD_NEW_STATIC(ulp, 0,0,0,0); |
4519 | MPD_NEW_STATIC(aa, 0,0,0,0); |
4520 | mpd_ssize_t prec; |
4521 | mpd_ssize_t ulpexp; |
4522 | uint32_t workstatus; |
4523 | |
4524 | if (result == a) { |
4525 | if (!mpd_qcopy(&aa, a, status)) { |
4526 | mpd_seterror(result, MPD_Malloc_error, status); |
4527 | return; |
4528 | } |
4529 | a = &aa; |
4530 | } |
4531 | |
4532 | workctx.clamp = 0; |
4533 | prec = ctx->prec + 3; |
4534 | while (1) { |
4535 | workctx.prec = prec; |
4536 | workstatus = 0; |
4537 | |
4538 | _mpd_qexp(result, a, &workctx, &workstatus); |
4539 | *status |= workstatus; |
4540 | |
4541 | ulpexp = result->exp + result->digits - workctx.prec; |
4542 | if (workstatus & MPD_Underflow) { |
4543 | /* The effective work precision is result->digits. */ |
4544 | ulpexp = result->exp; |
4545 | } |
4546 | _ssettriple(&ulp, MPD_POS, 1, ulpexp); |
4547 | |
4548 | /* |
4549 | * At this point [1]: |
4550 | * 1) abs(result - e**x) < 0.5 * 10**(-prec) * e**x |
4551 | * 2) result - ulp < e**x < result + ulp |
4552 | * 3) result - ulp < result < result + ulp |
4553 | * |
4554 | * If round(result-ulp)==round(result+ulp), then |
4555 | * round(result)==round(e**x). Therefore the result |
4556 | * is correctly rounded. |
4557 | * |
4558 | * [1] If abs(a) <= 9 * 10**(-prec-1), use the absolute |
4559 | * error for a similar argument. |
4560 | */ |
4561 | workctx.prec = ctx->prec; |
4562 | mpd_qadd(&t1, result, &ulp, &workctx, &workctx.status); |
4563 | mpd_qsub(&t2, result, &ulp, &workctx, &workctx.status); |
4564 | if (mpd_isspecial(result) || mpd_iszerocoeff(result) || |
4565 | mpd_qcmp(&t1, &t2, status) == 0) { |
4566 | workctx.clamp = ctx->clamp; |
4567 | _mpd_zeropad(result, &workctx, status); |
4568 | mpd_check_underflow(result, &workctx, status); |
4569 | mpd_qfinalize(result, &workctx, status); |
4570 | break; |
4571 | } |
4572 | prec += MPD_RDIGITS; |
4573 | } |
4574 | mpd_del(&t1); |
4575 | mpd_del(&t2); |
4576 | mpd_del(&ulp); |
4577 | mpd_del(&aa); |
4578 | } |
4579 | else { |
4580 | _mpd_qexp(result, a, &workctx, status); |
4581 | _mpd_zeropad(result, &workctx, status); |
4582 | mpd_check_underflow(result, &workctx, status); |
4583 | mpd_qfinalize(result, &workctx, status); |
4584 | } |
4585 | } |
4586 | |
4587 | /* Fused multiply-add: (a * b) + c, with a single final rounding. */ |
4588 | void |
4589 | mpd_qfma(mpd_t *result, const mpd_t *a, const mpd_t *b, const mpd_t *c, |
4590 | const mpd_context_t *ctx, uint32_t *status) |
4591 | { |
4592 | uint32_t workstatus = 0; |
4593 | mpd_t *cc = NULL; |
4594 | |
4595 | if (result == c) { |
4596 | if ((cc = mpd_qncopy(c)) == NULL) { |
4597 | mpd_seterror(result, MPD_Malloc_error, status); |
4598 | return; |
4599 | } |
4600 | c = cc; |
4601 | } |
4602 | |
4603 | _mpd_qmul(result, a, b, ctx, &workstatus); |
4604 | if (!(workstatus&MPD_Invalid_operation)) { |
4605 | mpd_qadd(result, result, c, ctx, &workstatus); |
4606 | } |
4607 | |
4608 | if (cc) mpd_del(cc); |
4609 | *status |= workstatus; |
4610 | } |
4611 | |
4612 | /* |
4613 | * Schedule the optimal precision increase for the Newton iteration. |
4614 | * v := input operand |
4615 | * z_0 := initial approximation |
4616 | * initprec := natural number such that abs(log(v) - z_0) < 10**-initprec |
4617 | * maxprec := target precision |
4618 | * |
4619 | * For convenience the output klist contains the elements in reverse order: |
4620 | * klist := [k_n-1, ..., k_0], where |
4621 | * 1) k_0 <= initprec and |
4622 | * 2) abs(log(v) - result) < 10**(-2*k_n-1 + 1) <= 10**-maxprec. |
4623 | */ |
4624 | static inline int |
4625 | ln_schedule_prec(mpd_ssize_t klist[MPD_MAX_PREC_LOG2], mpd_ssize_t maxprec, |
4626 | mpd_ssize_t initprec) |
4627 | { |
4628 | mpd_ssize_t k; |
4629 | int i; |
4630 | |
4631 | assert(maxprec >= 2 && initprec >= 2); |
4632 | if (maxprec <= initprec) return -1; |
4633 | |
4634 | i = 0; k = maxprec; |
4635 | do { |
4636 | k = (k+2) / 2; |
4637 | klist[i++] = k; |
4638 | } while (k > initprec); |
4639 | |
4640 | return i-1; |
4641 | } |
4642 | |
4643 | /* The constants have been verified with both decimal.py and mpfr. */ |
4644 | #ifdef CONFIG_64 |
4645 | #if MPD_RDIGITS != 19 |
4646 | #error "mpdecimal.c: MPD_RDIGITS must be 19." |
4647 | #endif |
4648 | static const mpd_uint_t mpd_ln10_data[MPD_MINALLOC_MAX] = { |
4649 | 6983716328982174407ULL, 9089704281976336583ULL, 1515961135648465461ULL, |
4650 | 4416816335727555703ULL, 2900988039194170265ULL, 2307925037472986509ULL, |
4651 | 107598438319191292ULL, 3466624107184669231ULL, 4450099781311469159ULL, |
4652 | 9807828059751193854ULL, 7713456862091670584ULL, 1492198849978748873ULL, |
4653 | 6528728696511086257ULL, 2385392051446341972ULL, 8692180205189339507ULL, |
4654 | 6518769751037497088ULL, 2375253577097505395ULL, 9095610299291824318ULL, |
4655 | 982748238504564801ULL, 5438635917781170543ULL, 7547331541421808427ULL, |
4656 | 752371033310119785ULL, 3171643095059950878ULL, 9785265383207606726ULL, |
4657 | 2932258279850258550ULL, 5497347726624257094ULL, 2976979522110718264ULL, |
4658 | 9221477656763693866ULL, 1979650047149510504ULL, 6674183485704422507ULL, |
4659 | 9702766860595249671ULL, 9278096762712757753ULL, 9314848524948644871ULL, |
4660 | 6826928280848118428ULL, 754403708474699401ULL, 230105703089634572ULL, |
4661 | 1929203337658714166ULL, 7589402567763113569ULL, 4208241314695689016ULL, |
4662 | 2922455440575892572ULL, 9356734206705811364ULL, 2684916746550586856ULL, |
4663 | 644507064800027750ULL, 9476834636167921018ULL, 5659121373450747856ULL, |
4664 | 2835522011480466371ULL, 6470806855677432162ULL, 7141748003688084012ULL, |
4665 | 9619404400222105101ULL, 5504893431493939147ULL, 6674744042432743651ULL, |
4666 | 2287698219886746543ULL, 7773262884616336622ULL, 1985283935053089653ULL, |
4667 | 4680843799894826233ULL, 8168948290720832555ULL, 8067566662873690987ULL, |
4668 | 6248633409525465082ULL, 9829834196778404228ULL, 3524802359972050895ULL, |
4669 | 3327900967572609677ULL, 110148862877297603ULL, 179914546843642076ULL, |
4670 | 2302585092994045684ULL |
4671 | }; |
4672 | #else |
4673 | #if MPD_RDIGITS != 9 |
4674 | #error "mpdecimal.c: MPD_RDIGITS must be 9." |
4675 | #endif |
4676 | static const mpd_uint_t mpd_ln10_data[MPD_MINALLOC_MAX] = { |
4677 | 401682692UL, 708474699UL, 720754403UL, 30896345UL, 602301057UL, 765871416UL, |
4678 | 192920333UL, 763113569UL, 589402567UL, 956890167UL, 82413146UL, 589257242UL, |
4679 | 245544057UL, 811364292UL, 734206705UL, 868569356UL, 167465505UL, 775026849UL, |
4680 | 706480002UL, 18064450UL, 636167921UL, 569476834UL, 734507478UL, 156591213UL, |
4681 | 148046637UL, 283552201UL, 677432162UL, 470806855UL, 880840126UL, 417480036UL, |
4682 | 210510171UL, 940440022UL, 939147961UL, 893431493UL, 436515504UL, 440424327UL, |
4683 | 654366747UL, 821988674UL, 622228769UL, 884616336UL, 537773262UL, 350530896UL, |
4684 | 319852839UL, 989482623UL, 468084379UL, 720832555UL, 168948290UL, 736909878UL, |
4685 | 675666628UL, 546508280UL, 863340952UL, 404228624UL, 834196778UL, 508959829UL, |
4686 | 23599720UL, 967735248UL, 96757260UL, 603332790UL, 862877297UL, 760110148UL, |
4687 | 468436420UL, 401799145UL, 299404568UL, 230258509UL |
4688 | }; |
4689 | #endif |
4690 | /* _mpd_ln10 is used directly for precisions smaller than MINALLOC_MAX*RDIGITS. |
4691 | Otherwise, it serves as the initial approximation for calculating ln(10). */ |
4692 | static const mpd_t _mpd_ln10 = { |
4693 | MPD_STATIC|MPD_CONST_DATA, -(MPD_MINALLOC_MAX*MPD_RDIGITS-1), |
4694 | MPD_MINALLOC_MAX*MPD_RDIGITS, MPD_MINALLOC_MAX, MPD_MINALLOC_MAX, |
4695 | (mpd_uint_t *)mpd_ln10_data |
4696 | }; |
4697 | |
4698 | /* |
4699 | * Set 'result' to log(10). |
4700 | * Ulp error: abs(result - log(10)) < ulp(log(10)) |
4701 | * Relative error: abs(result - log(10)) < 5 * 10**-prec * log(10) |
4702 | * |
4703 | * NOTE: The relative error is not derived from the ulp error, but |
4704 | * calculated separately using the fact that 23/10 < log(10) < 24/10. |
4705 | */ |
4706 | void |
4707 | mpd_qln10(mpd_t *result, mpd_ssize_t prec, uint32_t *status) |
4708 | { |
4709 | mpd_context_t varcontext, maxcontext; |
4710 | MPD_NEW_STATIC(tmp, 0,0,0,0); |
4711 | MPD_NEW_CONST(static10, 0,0,2,1,1,10); |
4712 | mpd_ssize_t klist[MPD_MAX_PREC_LOG2]; |
4713 | mpd_uint_t rnd; |
4714 | mpd_ssize_t shift; |
4715 | int i; |
4716 | |
4717 | assert(prec >= 1); |
4718 | |
4719 | shift = MPD_MINALLOC_MAX*MPD_RDIGITS-prec; |
4720 | shift = shift < 0 ? 0 : shift; |
4721 | |
4722 | rnd = mpd_qshiftr(result, &_mpd_ln10, shift, status); |
4723 | if (rnd == MPD_UINT_MAX) { |
4724 | mpd_seterror(result, MPD_Malloc_error, status); |
4725 | return; |
4726 | } |
4727 | result->exp = -(result->digits-1); |
4728 | |
4729 | mpd_maxcontext(&maxcontext); |
4730 | if (prec < MPD_MINALLOC_MAX*MPD_RDIGITS) { |
4731 | maxcontext.prec = prec; |
4732 | _mpd_apply_round_excess(result, rnd, &maxcontext, status); |
4733 | *status |= (MPD_Inexact|MPD_Rounded); |
4734 | return; |
4735 | } |
4736 | |
4737 | mpd_maxcontext(&varcontext); |
4738 | varcontext.round = MPD_ROUND_TRUNC; |
4739 | |
4740 | i = ln_schedule_prec(klist, prec+2, -result->exp); |
4741 | for (; i >= 0; i--) { |
4742 | varcontext.prec = 2*klist[i]+3; |
4743 | result->flags ^= MPD_NEG; |
4744 | _mpd_qexp(&tmp, result, &varcontext, status); |
4745 | result->flags ^= MPD_NEG; |
4746 | mpd_qmul(&tmp, &static10, &tmp, &varcontext, status); |
4747 | mpd_qsub(&tmp, &tmp, &one, &maxcontext, status); |
4748 | mpd_qadd(result, result, &tmp, &maxcontext, status); |
4749 | if (mpd_isspecial(result)) { |
4750 | break; |
4751 | } |
4752 | } |
4753 | |
4754 | mpd_del(&tmp); |
4755 | maxcontext.prec = prec; |
4756 | mpd_qfinalize(result, &maxcontext, status); |
4757 | } |
4758 | |
4759 | /* |
4760 | * Initial approximations for the ln() iteration. The values have the |
4761 | * following properties (established with both decimal.py and mpfr): |
4762 | * |
4763 | * Index 0 - 400, logarithms of x in [1.00, 5.00]: |
4764 | * abs(lnapprox[i] * 10**-3 - log((i+100)/100)) < 10**-2 |
4765 | * abs(lnapprox[i] * 10**-3 - log((i+1+100)/100)) < 10**-2 |
4766 | * |
4767 | * Index 401 - 899, logarithms of x in (0.500, 0.999]: |
4768 | * abs(-lnapprox[i] * 10**-3 - log((i+100)/1000)) < 10**-2 |
4769 | * abs(-lnapprox[i] * 10**-3 - log((i+1+100)/1000)) < 10**-2 |
4770 | */ |
4771 | static const uint16_t lnapprox[900] = { |
4772 | /* index 0 - 400: log((i+100)/100) * 1000 */ |
4773 | 0, 10, 20, 30, 39, 49, 58, 68, 77, 86, 95, 104, 113, 122, 131, 140, 148, 157, |
4774 | 166, 174, 182, 191, 199, 207, 215, 223, 231, 239, 247, 255, 262, 270, 278, |
4775 | 285, 293, 300, 308, 315, 322, 329, 336, 344, 351, 358, 365, 372, 378, 385, |
4776 | 392, 399, 406, 412, 419, 425, 432, 438, 445, 451, 457, 464, 470, 476, 482, |
4777 | 489, 495, 501, 507, 513, 519, 525, 531, 536, 542, 548, 554, 560, 565, 571, |
4778 | 577, 582, 588, 593, 599, 604, 610, 615, 621, 626, 631, 637, 642, 647, 652, |
4779 | 658, 663, 668, 673, 678, 683, 688, 693, 698, 703, 708, 713, 718, 723, 728, |
4780 | 732, 737, 742, 747, 751, 756, 761, 766, 770, 775, 779, 784, 788, 793, 798, |
4781 | 802, 806, 811, 815, 820, 824, 829, 833, 837, 842, 846, 850, 854, 859, 863, |
4782 | 867, 871, 876, 880, 884, 888, 892, 896, 900, 904, 908, 912, 916, 920, 924, |
4783 | 928, 932, 936, 940, 944, 948, 952, 956, 959, 963, 967, 971, 975, 978, 982, |
4784 | 986, 990, 993, 997, 1001, 1004, 1008, 1012, 1015, 1019, 1022, 1026, 1030, |
4785 | 1033, 1037, 1040, 1044, 1047, 1051, 1054, 1058, 1061, 1065, 1068, 1072, 1075, |
4786 | 1078, 1082, 1085, 1089, 1092, 1095, 1099, 1102, 1105, 1109, 1112, 1115, 1118, |
4787 | 1122, 1125, 1128, 1131, 1135, 1138, 1141, 1144, 1147, 1151, 1154, 1157, 1160, |
4788 | 1163, 1166, 1169, 1172, 1176, 1179, 1182, 1185, 1188, 1191, 1194, 1197, 1200, |
4789 | 1203, 1206, 1209, 1212, 1215, 1218, 1221, 1224, 1227, 1230, 1233, 1235, 1238, |
4790 | 1241, 1244, 1247, 1250, 1253, 1256, 1258, 1261, 1264, 1267, 1270, 1273, 1275, |
4791 | 1278, 1281, 1284, 1286, 1289, 1292, 1295, 1297, 1300, 1303, 1306, 1308, 1311, |
4792 | 1314, 1316, 1319, 1322, 1324, 1327, 1330, 1332, 1335, 1338, 1340, 1343, 1345, |
4793 | 1348, 1351, 1353, 1356, 1358, 1361, 1364, 1366, 1369, 1371, 1374, 1376, 1379, |
4794 | 1381, 1384, 1386, 1389, 1391, 1394, 1396, 1399, 1401, 1404, 1406, 1409, 1411, |
4795 | 1413, 1416, 1418, 1421, 1423, 1426, 1428, 1430, 1433, 1435, 1437, 1440, 1442, |
4796 | 1445, 1447, 1449, 1452, 1454, 1456, 1459, 1461, 1463, 1466, 1468, 1470, 1472, |
4797 | 1475, 1477, 1479, 1482, 1484, 1486, 1488, 1491, 1493, 1495, 1497, 1500, 1502, |
4798 | 1504, 1506, 1509, 1511, 1513, 1515, 1517, 1520, 1522, 1524, 1526, 1528, 1530, |
4799 | 1533, 1535, 1537, 1539, 1541, 1543, 1545, 1548, 1550, 1552, 1554, 1556, 1558, |
4800 | 1560, 1562, 1564, 1567, 1569, 1571, 1573, 1575, 1577, 1579, 1581, 1583, 1585, |
4801 | 1587, 1589, 1591, 1593, 1595, 1597, 1599, 1601, 1603, 1605, 1607, 1609, |
4802 | /* index 401 - 899: -log((i+100)/1000) * 1000 */ |
4803 | 691, 689, 687, 685, 683, 681, 679, 677, 675, 673, 671, 669, 668, 666, 664, |
4804 | 662, 660, 658, 656, 654, 652, 650, 648, 646, 644, 642, 641, 639, 637, 635, |
4805 | 633, 631, 629, 627, 626, 624, 622, 620, 618, 616, 614, 612, 611, 609, 607, |
4806 | 605, 603, 602, 600, 598, 596, 594, 592, 591, 589, 587, 585, 583, 582, 580, |
4807 | 578, 576, 574, 573, 571, 569, 567, 566, 564, 562, 560, 559, 557, 555, 553, |
4808 | 552, 550, 548, 546, 545, 543, 541, 540, 538, 536, 534, 533, 531, 529, 528, |
4809 | 526, 524, 523, 521, 519, 518, 516, 514, 512, 511, 509, 508, 506, 504, 502, |
4810 | 501, 499, 498, 496, 494, 493, 491, 489, 488, 486, 484, 483, 481, 480, 478, |
4811 | 476, 475, 473, 472, 470, 468, 467, 465, 464, 462, 460, 459, 457, 456, 454, |
4812 | 453, 451, 449, 448, 446, 445, 443, 442, 440, 438, 437, 435, 434, 432, 431, |
4813 | 429, 428, 426, 425, 423, 422, 420, 419, 417, 416, 414, 412, 411, 410, 408, |
4814 | 406, 405, 404, 402, 400, 399, 398, 396, 394, 393, 392, 390, 389, 387, 386, |
4815 | 384, 383, 381, 380, 378, 377, 375, 374, 372, 371, 370, 368, 367, 365, 364, |
4816 | 362, 361, 360, 358, 357, 355, 354, 352, 351, 350, 348, 347, 345, 344, 342, |
4817 | 341, 340, 338, 337, 336, 334, 333, 331, 330, 328, 327, 326, 324, 323, 322, |
4818 | 320, 319, 318, 316, 315, 313, 312, 311, 309, 308, 306, 305, 304, 302, 301, |
4819 | 300, 298, 297, 296, 294, 293, 292, 290, 289, 288, 286, 285, 284, 282, 281, |
4820 | 280, 278, 277, 276, 274, 273, 272, 270, 269, 268, 267, 265, 264, 263, 261, |
4821 | 260, 259, 258, 256, 255, 254, 252, 251, 250, 248, 247, 246, 245, 243, 242, |
4822 | 241, 240, 238, 237, 236, 234, 233, 232, 231, 229, 228, 227, 226, 224, 223, |
4823 | 222, 221, 219, 218, 217, 216, 214, 213, 212, 211, 210, 208, 207, 206, 205, |
4824 | 203, 202, 201, 200, 198, 197, 196, 195, 194, 192, 191, 190, 189, 188, 186, |
4825 | 185, 184, 183, 182, 180, 179, 178, 177, 176, 174, 173, 172, 171, 170, 168, |
4826 | 167, 166, 165, 164, 162, 161, 160, 159, 158, 157, 156, 154, 153, 152, 151, |
4827 | 150, 148, 147, 146, 145, 144, 143, 142, 140, 139, 138, 137, 136, 135, 134, |
4828 | 132, 131, 130, 129, 128, 127, 126, 124, 123, 122, 121, 120, 119, 118, 116, |
4829 | 115, 114, 113, 112, 111, 110, 109, 108, 106, 105, 104, 103, 102, 101, 100, |
4830 | 99, 98, 97, 95, 94, 93, 92, 91, 90, 89, 88, 87, 86, 84, 83, 82, 81, 80, 79, |
4831 | 78, 77, 76, 75, 74, 73, 72, 70, 69, 68, 67, 66, 65, 64, 63, 62, 61, 60, 59, |
4832 | 58, 57, 56, 54, 53, 52, 51, 50, 49, 48, 47, 46, 45, 44, 43, 42, 41, 40, 39, |
4833 | 38, 37, 36, 35, 34, 33, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, |
4834 | 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1 |
4835 | }; |
4836 | |
4837 | /* |
4838 | * Internal ln() function that does not check for specials, zero or one. |
4839 | * Relative error: abs(result - log(a)) < 0.1 * 10**-prec * abs(log(a)) |
4840 | */ |
4841 | static void |
4842 | _mpd_qln(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
4843 | uint32_t *status) |
4844 | { |
4845 | mpd_context_t varcontext, maxcontext; |
4846 | mpd_t *z = result; |
4847 | MPD_NEW_STATIC(v,0,0,0,0); |
4848 | MPD_NEW_STATIC(vtmp,0,0,0,0); |
4849 | MPD_NEW_STATIC(tmp,0,0,0,0); |
4850 | mpd_ssize_t klist[MPD_MAX_PREC_LOG2]; |
4851 | mpd_ssize_t maxprec, shift, t; |
4852 | mpd_ssize_t a_digits, a_exp; |
4853 | mpd_uint_t dummy, x; |
4854 | int i; |
4855 | |
4856 | assert(!mpd_isspecial(a) && !mpd_iszerocoeff(a)); |
4857 | |
4858 | /* |
4859 | * We are calculating ln(a) = ln(v * 10^t) = ln(v) + t*ln(10), |
4860 | * where 0.5 < v <= 5. |
4861 | */ |
4862 | if (!mpd_qcopy(&v, a, status)) { |
4863 | mpd_seterror(result, MPD_Malloc_error, status); |
4864 | goto finish; |
4865 | } |
4866 | |
4867 | /* Initial approximation: we have at least one non-zero digit */ |
4868 | _mpd_get_msdigits(&dummy, &x, &v, 3); |
4869 | if (x < 10) x *= 10; |
4870 | if (x < 100) x *= 10; |
4871 | x -= 100; |
4872 | |
4873 | /* a may equal z */ |
4874 | a_digits = a->digits; |
4875 | a_exp = a->exp; |
4876 | |
4877 | mpd_minalloc(z); |
4878 | mpd_clear_flags(z); |
4879 | z->data[0] = lnapprox[x]; |
4880 | z->len = 1; |
4881 | z->exp = -3; |
4882 | mpd_setdigits(z); |
4883 | |
4884 | if (x <= 400) { |
4885 | /* Reduce the input operand to 1.00 <= v <= 5.00. Let y = x + 100, |
4886 | * so 100 <= y <= 500. Since y contains the most significant digits |
4887 | * of v, y/100 <= v < (y+1)/100 and abs(z - log(v)) < 10**-2. */ |
4888 | v.exp = -(a_digits - 1); |
4889 | t = a_exp + a_digits - 1; |
4890 | } |
4891 | else { |
4892 | /* Reduce the input operand to 0.500 < v <= 0.999. Let y = x + 100, |
4893 | * so 500 < y <= 999. Since y contains the most significant digits |
4894 | * of v, y/1000 <= v < (y+1)/1000 and abs(z - log(v)) < 10**-2. */ |
4895 | v.exp = -a_digits; |
4896 | t = a_exp + a_digits; |
4897 | mpd_set_negative(z); |
4898 | } |
4899 | |
4900 | mpd_maxcontext(&maxcontext); |
4901 | mpd_maxcontext(&varcontext); |
4902 | varcontext.round = MPD_ROUND_TRUNC; |
4903 | |
4904 | maxprec = ctx->prec + 2; |
4905 | if (t == 0 && (x <= 15 || x >= 800)) { |
4906 | /* 0.900 <= v <= 1.15: Estimate the magnitude of the logarithm. |
4907 | * If ln(v) will underflow, skip the loop. Otherwise, adjust the |
4908 | * precision upwards in order to obtain a sufficient number of |
4909 | * significant digits. |
4910 | * |
4911 | * Case v > 1: |
4912 | * abs((v-1)/10) < abs((v-1)/v) < abs(ln(v)) < abs(v-1) |
4913 | * Case v < 1: |
4914 | * abs(v-1) < abs(ln(v)) < abs((v-1)/v) < abs((v-1)*10) |
4915 | */ |
4916 | int cmp = _mpd_cmp(&v, &one); |
4917 | |
4918 | /* Upper bound (assume v > 1): abs(v-1), unrounded */ |
4919 | _mpd_qsub(&tmp, &v, &one, &maxcontext, &maxcontext.status); |
4920 | if (maxcontext.status & MPD_Errors) { |
4921 | mpd_seterror(result, MPD_Malloc_error, status); |
4922 | goto finish; |
4923 | } |
4924 | |
4925 | if (cmp < 0) { |
4926 | /* v < 1: abs((v-1)*10) */ |
4927 | tmp.exp += 1; |
4928 | } |
4929 | if (mpd_adjexp(&tmp) < mpd_etiny(ctx)) { |
4930 | /* The upper bound is less than etiny: Underflow to zero */ |
4931 | _settriple(result, (cmp<0), 1, mpd_etiny(ctx)-1); |
4932 | goto finish; |
4933 | } |
4934 | /* Lower bound: abs((v-1)/10) or abs(v-1) */ |
4935 | tmp.exp -= 1; |
4936 | if (mpd_adjexp(&tmp) < 0) { |
4937 | /* Absolute error of the loop: abs(z - log(v)) < 10**-p. If |
4938 | * p = ctx->prec+2-adjexp(lower), then the relative error of |
4939 | * the result is (using 10**adjexp(x) <= abs(x)): |
4940 | * |
4941 | * abs(z - log(v)) / abs(log(v)) < 10**-p / abs(log(v)) |
4942 | * <= 10**(-ctx->prec-2) |
4943 | */ |
4944 | maxprec = maxprec - mpd_adjexp(&tmp); |
4945 | } |
4946 | } |
4947 | |
4948 | i = ln_schedule_prec(klist, maxprec, 2); |
4949 | for (; i >= 0; i--) { |
4950 | varcontext.prec = 2*klist[i]+3; |
4951 | z->flags ^= MPD_NEG; |
4952 | _mpd_qexp(&tmp, z, &varcontext, status); |
4953 | z->flags ^= MPD_NEG; |
4954 | |
4955 | if (v.digits > varcontext.prec) { |
4956 | shift = v.digits - varcontext.prec; |
4957 | mpd_qshiftr(&vtmp, &v, shift, status); |
4958 | vtmp.exp += shift; |
4959 | mpd_qmul(&tmp, &vtmp, &tmp, &varcontext, status); |
4960 | } |
4961 | else { |
4962 | mpd_qmul(&tmp, &v, &tmp, &varcontext, status); |
4963 | } |
4964 | |
4965 | mpd_qsub(&tmp, &tmp, &one, &maxcontext, status); |
4966 | mpd_qadd(z, z, &tmp, &maxcontext, status); |
4967 | if (mpd_isspecial(z)) { |
4968 | break; |
4969 | } |
4970 | } |
4971 | |
4972 | /* |
4973 | * Case t == 0: |
4974 | * t * log(10) == 0, the result does not change and the analysis |
4975 | * above applies. If v < 0.900 or v > 1.15, the relative error is |
4976 | * less than 10**(-ctx.prec-1). |
4977 | * Case t != 0: |
4978 | * z := approx(log(v)) |
4979 | * y := approx(log(10)) |
4980 | * p := maxprec = ctx->prec + 2 |
4981 | * Absolute errors: |
4982 | * 1) abs(z - log(v)) < 10**-p |
4983 | * 2) abs(y - log(10)) < 10**-p |
4984 | * The multiplication is exact, so: |
4985 | * 3) abs(t*y - t*log(10)) < t*10**-p |
4986 | * The sum is exact, so: |
4987 | * 4) abs((z + t*y) - (log(v) + t*log(10))) < (abs(t) + 1) * 10**-p |
4988 | * Bounds for log(v) and log(10): |
4989 | * 5) -7/10 < log(v) < 17/10 |
4990 | * 6) 23/10 < log(10) < 24/10 |
4991 | * Using 4), 5), 6) and t != 0, the relative error is: |
4992 | * |
4993 | * 7) relerr < ((abs(t) + 1)*10**-p) / abs(log(v) + t*log(10)) |
4994 | * < 0.5 * 10**(-p + 1) = 0.5 * 10**(-ctx->prec-1) |
4995 | */ |
4996 | mpd_qln10(&v, maxprec+1, status); |
4997 | mpd_qmul_ssize(&tmp, &v, t, &maxcontext, status); |
4998 | mpd_qadd(result, &tmp, z, &maxcontext, status); |
4999 | |
5000 | |
5001 | finish: |
5002 | *status |= (MPD_Inexact|MPD_Rounded); |
5003 | mpd_del(&v); |
5004 | mpd_del(&vtmp); |
5005 | mpd_del(&tmp); |
5006 | } |
5007 | |
5008 | /* ln(a) */ |
5009 | void |
5010 | mpd_qln(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
5011 | uint32_t *status) |
5012 | { |
5013 | mpd_context_t workctx; |
5014 | mpd_ssize_t adjexp, t; |
5015 | |
5016 | if (mpd_isspecial(a)) { |
5017 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
5018 | return; |
5019 | } |
5020 | if (mpd_isnegative(a)) { |
5021 | mpd_seterror(result, MPD_Invalid_operation, status); |
5022 | return; |
5023 | } |
5024 | mpd_setspecial(result, MPD_POS, MPD_INF); |
5025 | return; |
5026 | } |
5027 | if (mpd_iszerocoeff(a)) { |
5028 | mpd_setspecial(result, MPD_NEG, MPD_INF); |
5029 | return; |
5030 | } |
5031 | if (mpd_isnegative(a)) { |
5032 | mpd_seterror(result, MPD_Invalid_operation, status); |
5033 | return; |
5034 | } |
5035 | if (_mpd_cmp(a, &one) == 0) { |
5036 | _settriple(result, MPD_POS, 0, 0); |
5037 | return; |
5038 | } |
5039 | /* |
5040 | * Check if the result will overflow (0 < x, x != 1): |
5041 | * 1) log10(x) < 0 iff adjexp(x) < 0 |
5042 | * 2) 0 < x /\ x <= y ==> adjexp(x) <= adjexp(y) |
5043 | * 3) 0 < x /\ x != 1 ==> 2 * abs(log10(x)) < abs(log(x)) |
5044 | * 4) adjexp(x) <= log10(x) < adjexp(x) + 1 |
5045 | * |
5046 | * Case adjexp(x) >= 0: |
5047 | * 5) 2 * adjexp(x) < abs(log(x)) |
5048 | * Case adjexp(x) > 0: |
5049 | * 6) adjexp(2 * adjexp(x)) <= adjexp(abs(log(x))) |
5050 | * Case adjexp(x) == 0: |
5051 | * mpd_exp_digits(t)-1 == 0 <= emax (the shortcut is not triggered) |
5052 | * |
5053 | * Case adjexp(x) < 0: |
5054 | * 7) 2 * (-adjexp(x) - 1) < abs(log(x)) |
5055 | * Case adjexp(x) < -1: |
5056 | * 8) adjexp(2 * (-adjexp(x) - 1)) <= adjexp(abs(log(x))) |
5057 | * Case adjexp(x) == -1: |
5058 | * mpd_exp_digits(t)-1 == 0 <= emax (the shortcut is not triggered) |
5059 | */ |
5060 | adjexp = mpd_adjexp(a); |
5061 | t = (adjexp < 0) ? -adjexp-1 : adjexp; |
5062 | t *= 2; |
5063 | if (mpd_exp_digits(t)-1 > ctx->emax) { |
5064 | *status |= MPD_Overflow|MPD_Inexact|MPD_Rounded; |
5065 | mpd_setspecial(result, (adjexp<0), MPD_INF); |
5066 | return; |
5067 | } |
5068 | |
5069 | workctx = *ctx; |
5070 | workctx.round = MPD_ROUND_HALF_EVEN; |
5071 | |
5072 | if (ctx->allcr) { |
5073 | MPD_NEW_STATIC(t1, 0,0,0,0); |
5074 | MPD_NEW_STATIC(t2, 0,0,0,0); |
5075 | MPD_NEW_STATIC(ulp, 0,0,0,0); |
5076 | MPD_NEW_STATIC(aa, 0,0,0,0); |
5077 | mpd_ssize_t prec; |
5078 | |
5079 | if (result == a) { |
5080 | if (!mpd_qcopy(&aa, a, status)) { |
5081 | mpd_seterror(result, MPD_Malloc_error, status); |
5082 | return; |
5083 | } |
5084 | a = &aa; |
5085 | } |
5086 | |
5087 | workctx.clamp = 0; |
5088 | prec = ctx->prec + 3; |
5089 | while (1) { |
5090 | workctx.prec = prec; |
5091 | _mpd_qln(result, a, &workctx, status); |
5092 | _ssettriple(&ulp, MPD_POS, 1, |
5093 | result->exp + result->digits-workctx.prec); |
5094 | |
5095 | workctx.prec = ctx->prec; |
5096 | mpd_qadd(&t1, result, &ulp, &workctx, &workctx.status); |
5097 | mpd_qsub(&t2, result, &ulp, &workctx, &workctx.status); |
5098 | if (mpd_isspecial(result) || mpd_iszerocoeff(result) || |
5099 | mpd_qcmp(&t1, &t2, status) == 0) { |
5100 | workctx.clamp = ctx->clamp; |
5101 | mpd_check_underflow(result, &workctx, status); |
5102 | mpd_qfinalize(result, &workctx, status); |
5103 | break; |
5104 | } |
5105 | prec += MPD_RDIGITS; |
5106 | } |
5107 | mpd_del(&t1); |
5108 | mpd_del(&t2); |
5109 | mpd_del(&ulp); |
5110 | mpd_del(&aa); |
5111 | } |
5112 | else { |
5113 | _mpd_qln(result, a, &workctx, status); |
5114 | mpd_check_underflow(result, &workctx, status); |
5115 | mpd_qfinalize(result, &workctx, status); |
5116 | } |
5117 | } |
5118 | |
5119 | /* |
5120 | * Internal log10() function that does not check for specials, zero or one. |
5121 | * Case SKIP_FINALIZE: |
5122 | * Relative error: abs(result - log10(a)) < 0.1 * 10**-prec * abs(log10(a)) |
5123 | * Case DO_FINALIZE: |
5124 | * Ulp error: abs(result - log10(a)) < ulp(log10(a)) |
5125 | */ |
5126 | enum {SKIP_FINALIZE, DO_FINALIZE}; |
5127 | static void |
5128 | _mpd_qlog10(int action, mpd_t *result, const mpd_t *a, |
5129 | const mpd_context_t *ctx, uint32_t *status) |
5130 | { |
5131 | mpd_context_t workctx; |
5132 | MPD_NEW_STATIC(ln10,0,0,0,0); |
5133 | |
5134 | mpd_maxcontext(&workctx); |
5135 | workctx.prec = ctx->prec + 3; |
5136 | /* relative error: 0.1 * 10**(-p-3). The specific underflow shortcut |
5137 | * in _mpd_qln() does not change the final result. */ |
5138 | _mpd_qln(result, a, &workctx, status); |
5139 | /* relative error: 5 * 10**(-p-3) */ |
5140 | mpd_qln10(&ln10, workctx.prec, status); |
5141 | |
5142 | if (action == DO_FINALIZE) { |
5143 | workctx = *ctx; |
5144 | workctx.round = MPD_ROUND_HALF_EVEN; |
5145 | } |
5146 | /* SKIP_FINALIZE: relative error: 5 * 10**(-p-3) */ |
5147 | _mpd_qdiv(NO_IDEAL_EXP, result, result, &ln10, &workctx, status); |
5148 | |
5149 | mpd_del(&ln10); |
5150 | } |
5151 | |
5152 | /* log10(a) */ |
5153 | void |
5154 | mpd_qlog10(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
5155 | uint32_t *status) |
5156 | { |
5157 | mpd_context_t workctx; |
5158 | mpd_ssize_t adjexp, t; |
5159 | |
5160 | workctx = *ctx; |
5161 | workctx.round = MPD_ROUND_HALF_EVEN; |
5162 | |
5163 | if (mpd_isspecial(a)) { |
5164 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
5165 | return; |
5166 | } |
5167 | if (mpd_isnegative(a)) { |
5168 | mpd_seterror(result, MPD_Invalid_operation, status); |
5169 | return; |
5170 | } |
5171 | mpd_setspecial(result, MPD_POS, MPD_INF); |
5172 | return; |
5173 | } |
5174 | if (mpd_iszerocoeff(a)) { |
5175 | mpd_setspecial(result, MPD_NEG, MPD_INF); |
5176 | return; |
5177 | } |
5178 | if (mpd_isnegative(a)) { |
5179 | mpd_seterror(result, MPD_Invalid_operation, status); |
5180 | return; |
5181 | } |
5182 | if (mpd_coeff_ispow10(a)) { |
5183 | uint8_t sign = 0; |
5184 | adjexp = mpd_adjexp(a); |
5185 | if (adjexp < 0) { |
5186 | sign = 1; |
5187 | adjexp = -adjexp; |
5188 | } |
5189 | _settriple(result, sign, adjexp, 0); |
5190 | mpd_qfinalize(result, &workctx, status); |
5191 | return; |
5192 | } |
5193 | /* |
5194 | * Check if the result will overflow (0 < x, x != 1): |
5195 | * 1) log10(x) < 0 iff adjexp(x) < 0 |
5196 | * 2) 0 < x /\ x <= y ==> adjexp(x) <= adjexp(y) |
5197 | * 3) adjexp(x) <= log10(x) < adjexp(x) + 1 |
5198 | * |
5199 | * Case adjexp(x) >= 0: |
5200 | * 4) adjexp(x) <= abs(log10(x)) |
5201 | * Case adjexp(x) > 0: |
5202 | * 5) adjexp(adjexp(x)) <= adjexp(abs(log10(x))) |
5203 | * Case adjexp(x) == 0: |
5204 | * mpd_exp_digits(t)-1 == 0 <= emax (the shortcut is not triggered) |
5205 | * |
5206 | * Case adjexp(x) < 0: |
5207 | * 6) -adjexp(x) - 1 < abs(log10(x)) |
5208 | * Case adjexp(x) < -1: |
5209 | * 7) adjexp(-adjexp(x) - 1) <= adjexp(abs(log(x))) |
5210 | * Case adjexp(x) == -1: |
5211 | * mpd_exp_digits(t)-1 == 0 <= emax (the shortcut is not triggered) |
5212 | */ |
5213 | adjexp = mpd_adjexp(a); |
5214 | t = (adjexp < 0) ? -adjexp-1 : adjexp; |
5215 | if (mpd_exp_digits(t)-1 > ctx->emax) { |
5216 | *status |= MPD_Overflow|MPD_Inexact|MPD_Rounded; |
5217 | mpd_setspecial(result, (adjexp<0), MPD_INF); |
5218 | return; |
5219 | } |
5220 | |
5221 | if (ctx->allcr) { |
5222 | MPD_NEW_STATIC(t1, 0,0,0,0); |
5223 | MPD_NEW_STATIC(t2, 0,0,0,0); |
5224 | MPD_NEW_STATIC(ulp, 0,0,0,0); |
5225 | MPD_NEW_STATIC(aa, 0,0,0,0); |
5226 | mpd_ssize_t prec; |
5227 | |
5228 | if (result == a) { |
5229 | if (!mpd_qcopy(&aa, a, status)) { |
5230 | mpd_seterror(result, MPD_Malloc_error, status); |
5231 | return; |
5232 | } |
5233 | a = &aa; |
5234 | } |
5235 | |
5236 | workctx.clamp = 0; |
5237 | prec = ctx->prec + 3; |
5238 | while (1) { |
5239 | workctx.prec = prec; |
5240 | _mpd_qlog10(SKIP_FINALIZE, result, a, &workctx, status); |
5241 | _ssettriple(&ulp, MPD_POS, 1, |
5242 | result->exp + result->digits-workctx.prec); |
5243 | |
5244 | workctx.prec = ctx->prec; |
5245 | mpd_qadd(&t1, result, &ulp, &workctx, &workctx.status); |
5246 | mpd_qsub(&t2, result, &ulp, &workctx, &workctx.status); |
5247 | if (mpd_isspecial(result) || mpd_iszerocoeff(result) || |
5248 | mpd_qcmp(&t1, &t2, status) == 0) { |
5249 | workctx.clamp = ctx->clamp; |
5250 | mpd_check_underflow(result, &workctx, status); |
5251 | mpd_qfinalize(result, &workctx, status); |
5252 | break; |
5253 | } |
5254 | prec += MPD_RDIGITS; |
5255 | } |
5256 | mpd_del(&t1); |
5257 | mpd_del(&t2); |
5258 | mpd_del(&ulp); |
5259 | mpd_del(&aa); |
5260 | } |
5261 | else { |
5262 | _mpd_qlog10(DO_FINALIZE, result, a, &workctx, status); |
5263 | mpd_check_underflow(result, &workctx, status); |
5264 | } |
5265 | } |
5266 | |
5267 | /* |
5268 | * Maximum of the two operands. Attention: If one operand is a quiet NaN and the |
5269 | * other is numeric, the numeric operand is returned. This may not be what one |
5270 | * expects. |
5271 | */ |
5272 | void |
5273 | mpd_qmax(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5274 | const mpd_context_t *ctx, uint32_t *status) |
5275 | { |
5276 | int c; |
5277 | |
5278 | if (mpd_isqnan(a) && !mpd_isnan(b)) { |
5279 | mpd_qcopy(result, b, status); |
5280 | } |
5281 | else if (mpd_isqnan(b) && !mpd_isnan(a)) { |
5282 | mpd_qcopy(result, a, status); |
5283 | } |
5284 | else if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5285 | return; |
5286 | } |
5287 | else { |
5288 | c = _mpd_cmp(a, b); |
5289 | if (c == 0) { |
5290 | c = _mpd_cmp_numequal(a, b); |
5291 | } |
5292 | |
5293 | if (c < 0) { |
5294 | mpd_qcopy(result, b, status); |
5295 | } |
5296 | else { |
5297 | mpd_qcopy(result, a, status); |
5298 | } |
5299 | } |
5300 | |
5301 | mpd_qfinalize(result, ctx, status); |
5302 | } |
5303 | |
5304 | /* |
5305 | * Maximum magnitude: Same as mpd_max(), but compares the operands with their |
5306 | * sign ignored. |
5307 | */ |
5308 | void |
5309 | mpd_qmax_mag(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5310 | const mpd_context_t *ctx, uint32_t *status) |
5311 | { |
5312 | int c; |
5313 | |
5314 | if (mpd_isqnan(a) && !mpd_isnan(b)) { |
5315 | mpd_qcopy(result, b, status); |
5316 | } |
5317 | else if (mpd_isqnan(b) && !mpd_isnan(a)) { |
5318 | mpd_qcopy(result, a, status); |
5319 | } |
5320 | else if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5321 | return; |
5322 | } |
5323 | else { |
5324 | c = _mpd_cmp_abs(a, b); |
5325 | if (c == 0) { |
5326 | c = _mpd_cmp_numequal(a, b); |
5327 | } |
5328 | |
5329 | if (c < 0) { |
5330 | mpd_qcopy(result, b, status); |
5331 | } |
5332 | else { |
5333 | mpd_qcopy(result, a, status); |
5334 | } |
5335 | } |
5336 | |
5337 | mpd_qfinalize(result, ctx, status); |
5338 | } |
5339 | |
5340 | /* |
5341 | * Minimum of the two operands. Attention: If one operand is a quiet NaN and the |
5342 | * other is numeric, the numeric operand is returned. This may not be what one |
5343 | * expects. |
5344 | */ |
5345 | void |
5346 | mpd_qmin(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5347 | const mpd_context_t *ctx, uint32_t *status) |
5348 | { |
5349 | int c; |
5350 | |
5351 | if (mpd_isqnan(a) && !mpd_isnan(b)) { |
5352 | mpd_qcopy(result, b, status); |
5353 | } |
5354 | else if (mpd_isqnan(b) && !mpd_isnan(a)) { |
5355 | mpd_qcopy(result, a, status); |
5356 | } |
5357 | else if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5358 | return; |
5359 | } |
5360 | else { |
5361 | c = _mpd_cmp(a, b); |
5362 | if (c == 0) { |
5363 | c = _mpd_cmp_numequal(a, b); |
5364 | } |
5365 | |
5366 | if (c < 0) { |
5367 | mpd_qcopy(result, a, status); |
5368 | } |
5369 | else { |
5370 | mpd_qcopy(result, b, status); |
5371 | } |
5372 | } |
5373 | |
5374 | mpd_qfinalize(result, ctx, status); |
5375 | } |
5376 | |
5377 | /* |
5378 | * Minimum magnitude: Same as mpd_min(), but compares the operands with their |
5379 | * sign ignored. |
5380 | */ |
5381 | void |
5382 | mpd_qmin_mag(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5383 | const mpd_context_t *ctx, uint32_t *status) |
5384 | { |
5385 | int c; |
5386 | |
5387 | if (mpd_isqnan(a) && !mpd_isnan(b)) { |
5388 | mpd_qcopy(result, b, status); |
5389 | } |
5390 | else if (mpd_isqnan(b) && !mpd_isnan(a)) { |
5391 | mpd_qcopy(result, a, status); |
5392 | } |
5393 | else if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5394 | return; |
5395 | } |
5396 | else { |
5397 | c = _mpd_cmp_abs(a, b); |
5398 | if (c == 0) { |
5399 | c = _mpd_cmp_numequal(a, b); |
5400 | } |
5401 | |
5402 | if (c < 0) { |
5403 | mpd_qcopy(result, a, status); |
5404 | } |
5405 | else { |
5406 | mpd_qcopy(result, b, status); |
5407 | } |
5408 | } |
5409 | |
5410 | mpd_qfinalize(result, ctx, status); |
5411 | } |
5412 | |
5413 | /* Minimum space needed for the result array in _karatsuba_rec(). */ |
5414 | static inline mpd_size_t |
5415 | _kmul_resultsize(mpd_size_t la, mpd_size_t lb) |
5416 | { |
5417 | mpd_size_t n, m; |
5418 | |
5419 | n = add_size_t(la, lb); |
5420 | n = add_size_t(n, 1); |
5421 | |
5422 | m = (la+1)/2 + 1; |
5423 | m = mul_size_t(m, 3); |
5424 | |
5425 | return (m > n) ? m : n; |
5426 | } |
5427 | |
5428 | /* Work space needed in _karatsuba_rec(). lim >= 4 */ |
5429 | static inline mpd_size_t |
5430 | _kmul_worksize(mpd_size_t n, mpd_size_t lim) |
5431 | { |
5432 | mpd_size_t m; |
5433 | |
5434 | if (n <= lim) { |
5435 | return 0; |
5436 | } |
5437 | |
5438 | m = (n+1)/2 + 1; |
5439 | |
5440 | return add_size_t(mul_size_t(m, 2), _kmul_worksize(m, lim)); |
5441 | } |
5442 | |
5443 | |
5444 | #define MPD_KARATSUBA_BASECASE 16 /* must be >= 4 */ |
5445 | |
5446 | /* |
5447 | * Add the product of a and b to c. |
5448 | * c must be _kmul_resultsize(la, lb) in size. |
5449 | * w is used as a work array and must be _kmul_worksize(a, lim) in size. |
5450 | * Roman E. Maeder, Storage Allocation for the Karatsuba Integer Multiplication |
5451 | * Algorithm. In "Design and implementation of symbolic computation systems", |
5452 | * Springer, 1993, ISBN 354057235X, 9783540572350. |
5453 | */ |
5454 | static void |
5455 | _karatsuba_rec(mpd_uint_t *c, const mpd_uint_t *a, const mpd_uint_t *b, |
5456 | mpd_uint_t *w, mpd_size_t la, mpd_size_t lb) |
5457 | { |
5458 | mpd_size_t m, lt; |
5459 | |
5460 | assert(la >= lb && lb > 0); |
5461 | assert(la <= MPD_KARATSUBA_BASECASE || w != NULL); |
5462 | |
5463 | if (la <= MPD_KARATSUBA_BASECASE) { |
5464 | _mpd_basemul(c, a, b, la, lb); |
5465 | return; |
5466 | } |
5467 | |
5468 | m = (la+1)/2; /* ceil(la/2) */ |
5469 | |
5470 | /* lb <= m < la */ |
5471 | if (lb <= m) { |
5472 | |
5473 | /* lb can now be larger than la-m */ |
5474 | if (lb > la-m) { |
5475 | lt = lb + lb + 1; /* space needed for result array */ |
5476 | mpd_uint_zero(w, lt); /* clear result array */ |
5477 | _karatsuba_rec(w, b, a+m, w+lt, lb, la-m); /* b*ah */ |
5478 | } |
5479 | else { |
5480 | lt = (la-m) + (la-m) + 1; /* space needed for result array */ |
5481 | mpd_uint_zero(w, lt); /* clear result array */ |
5482 | _karatsuba_rec(w, a+m, b, w+lt, la-m, lb); /* ah*b */ |
5483 | } |
5484 | _mpd_baseaddto(c+m, w, (la-m)+lb); /* add ah*b*B**m */ |
5485 | |
5486 | lt = m + m + 1; /* space needed for the result array */ |
5487 | mpd_uint_zero(w, lt); /* clear result array */ |
5488 | _karatsuba_rec(w, a, b, w+lt, m, lb); /* al*b */ |
5489 | _mpd_baseaddto(c, w, m+lb); /* add al*b */ |
5490 | |
5491 | return; |
5492 | } |
5493 | |
5494 | /* la >= lb > m */ |
5495 | memcpy(w, a, m * sizeof *w); |
5496 | w[m] = 0; |
5497 | _mpd_baseaddto(w, a+m, la-m); |
5498 | |
5499 | memcpy(w+(m+1), b, m * sizeof *w); |
5500 | w[m+1+m] = 0; |
5501 | _mpd_baseaddto(w+(m+1), b+m, lb-m); |
5502 | |
5503 | _karatsuba_rec(c+m, w, w+(m+1), w+2*(m+1), m+1, m+1); |
5504 | |
5505 | lt = (la-m) + (la-m) + 1; |
5506 | mpd_uint_zero(w, lt); |
5507 | |
5508 | _karatsuba_rec(w, a+m, b+m, w+lt, la-m, lb-m); |
5509 | |
5510 | _mpd_baseaddto(c+2*m, w, (la-m) + (lb-m)); |
5511 | _mpd_basesubfrom(c+m, w, (la-m) + (lb-m)); |
5512 | |
5513 | lt = m + m + 1; |
5514 | mpd_uint_zero(w, lt); |
5515 | |
5516 | _karatsuba_rec(w, a, b, w+lt, m, m); |
5517 | _mpd_baseaddto(c, w, m+m); |
5518 | _mpd_basesubfrom(c+m, w, m+m); |
5519 | |
5520 | return; |
5521 | } |
5522 | |
5523 | /* |
5524 | * Multiply u and v, using Karatsuba multiplication. Returns a pointer |
5525 | * to the result or NULL in case of failure (malloc error). |
5526 | * Conditions: ulen >= vlen, ulen >= 4 |
5527 | */ |
5528 | static mpd_uint_t * |
5529 | _mpd_kmul(const mpd_uint_t *u, const mpd_uint_t *v, |
5530 | mpd_size_t ulen, mpd_size_t vlen, |
5531 | mpd_size_t *rsize) |
5532 | { |
5533 | mpd_uint_t *result = NULL, *w = NULL; |
5534 | mpd_size_t m; |
5535 | |
5536 | assert(ulen >= 4); |
5537 | assert(ulen >= vlen); |
5538 | |
5539 | *rsize = _kmul_resultsize(ulen, vlen); |
5540 | if ((result = mpd_calloc(*rsize, sizeof *result)) == NULL) { |
5541 | return NULL; |
5542 | } |
5543 | |
5544 | m = _kmul_worksize(ulen, MPD_KARATSUBA_BASECASE); |
5545 | if (m && ((w = mpd_calloc(m, sizeof *w)) == NULL)) { |
5546 | mpd_free(result); |
5547 | return NULL; |
5548 | } |
5549 | |
5550 | _karatsuba_rec(result, u, v, w, ulen, vlen); |
5551 | |
5552 | |
5553 | if (w) mpd_free(w); |
5554 | return result; |
5555 | } |
5556 | |
5557 | |
5558 | /* |
5559 | * Determine the minimum length for the number theoretic transform. Valid |
5560 | * transform lengths are 2**n or 3*2**n, where 2**n <= MPD_MAXTRANSFORM_2N. |
5561 | * The function finds the shortest length m such that rsize <= m. |
5562 | */ |
5563 | static inline mpd_size_t |
5564 | _mpd_get_transform_len(mpd_size_t rsize) |
5565 | { |
5566 | mpd_size_t log2rsize; |
5567 | mpd_size_t x, step; |
5568 | |
5569 | assert(rsize >= 4); |
5570 | log2rsize = mpd_bsr(rsize); |
5571 | |
5572 | if (rsize <= 1024) { |
5573 | /* 2**n is faster in this range. */ |
5574 | x = ((mpd_size_t)1)<<log2rsize; |
5575 | return (rsize == x) ? x : x<<1; |
5576 | } |
5577 | else if (rsize <= MPD_MAXTRANSFORM_2N) { |
5578 | x = ((mpd_size_t)1)<<log2rsize; |
5579 | if (rsize == x) return x; |
5580 | step = x>>1; |
5581 | x += step; |
5582 | return (rsize <= x) ? x : x + step; |
5583 | } |
5584 | else if (rsize <= MPD_MAXTRANSFORM_2N+MPD_MAXTRANSFORM_2N/2) { |
5585 | return MPD_MAXTRANSFORM_2N+MPD_MAXTRANSFORM_2N/2; |
5586 | } |
5587 | else if (rsize <= 3*MPD_MAXTRANSFORM_2N) { |
5588 | return 3*MPD_MAXTRANSFORM_2N; |
5589 | } |
5590 | else { |
5591 | return MPD_SIZE_MAX; |
5592 | } |
5593 | } |
5594 | |
5595 | #ifdef PPRO |
5596 | #ifndef _MSC_VER |
5597 | static inline unsigned short |
5598 | _mpd_get_control87(void) |
5599 | { |
5600 | unsigned short cw; |
5601 | |
5602 | __asm__ __volatile__ ("fnstcw %0" : "=m" (cw)); |
5603 | return cw; |
5604 | } |
5605 | |
5606 | static inline void |
5607 | _mpd_set_control87(unsigned short cw) |
5608 | { |
5609 | __asm__ __volatile__ ("fldcw %0" : : "m" (cw)); |
5610 | } |
5611 | #endif |
5612 | |
5613 | static unsigned int |
5614 | mpd_set_fenv(void) |
5615 | { |
5616 | unsigned int cw; |
5617 | #ifdef _MSC_VER |
5618 | unsigned int flags = |
5619 | _EM_INVALID|_EM_DENORMAL|_EM_ZERODIVIDE|_EM_OVERFLOW| |
5620 | _EM_UNDERFLOW|_EM_INEXACT|_RC_CHOP|_PC_64; |
5621 | unsigned int mask = _MCW_EM|_MCW_RC|_MCW_PC; |
5622 | unsigned int dummy; |
5623 | |
5624 | __control87_2(0, 0, &cw, NULL); |
5625 | __control87_2(flags, mask, &dummy, NULL); |
5626 | #else |
5627 | cw = _mpd_get_control87(); |
5628 | _mpd_set_control87(cw|0xF3F); |
5629 | #endif |
5630 | return cw; |
5631 | } |
5632 | |
5633 | static void |
5634 | mpd_restore_fenv(unsigned int cw) |
5635 | { |
5636 | #ifdef _MSC_VER |
5637 | unsigned int mask = _MCW_EM|_MCW_RC|_MCW_PC; |
5638 | unsigned int dummy; |
5639 | |
5640 | __control87_2(cw, mask, &dummy, NULL); |
5641 | #else |
5642 | _mpd_set_control87((unsigned short)cw); |
5643 | #endif |
5644 | } |
5645 | #endif /* PPRO */ |
5646 | |
5647 | /* |
5648 | * Multiply u and v, using the fast number theoretic transform. Returns |
5649 | * a pointer to the result or NULL in case of failure (malloc error). |
5650 | */ |
5651 | static mpd_uint_t * |
5652 | _mpd_fntmul(const mpd_uint_t *u, const mpd_uint_t *v, |
5653 | mpd_size_t ulen, mpd_size_t vlen, |
5654 | mpd_size_t *rsize) |
5655 | { |
5656 | mpd_uint_t *c1 = NULL, *c2 = NULL, *c3 = NULL, *vtmp = NULL; |
5657 | mpd_size_t n; |
5658 | |
5659 | #ifdef PPRO |
5660 | unsigned int cw; |
5661 | cw = mpd_set_fenv(); |
5662 | #endif |
5663 | |
5664 | *rsize = add_size_t(ulen, vlen); |
5665 | if ((n = _mpd_get_transform_len(*rsize)) == MPD_SIZE_MAX) { |
5666 | goto malloc_error; |
5667 | } |
5668 | |
5669 | if ((c1 = mpd_calloc(n, sizeof *c1)) == NULL) { |
5670 | goto malloc_error; |
5671 | } |
5672 | if ((c2 = mpd_calloc(n, sizeof *c2)) == NULL) { |
5673 | goto malloc_error; |
5674 | } |
5675 | if ((c3 = mpd_calloc(n, sizeof *c3)) == NULL) { |
5676 | goto malloc_error; |
5677 | } |
5678 | |
5679 | memcpy(c1, u, ulen * (sizeof *c1)); |
5680 | memcpy(c2, u, ulen * (sizeof *c2)); |
5681 | memcpy(c3, u, ulen * (sizeof *c3)); |
5682 | |
5683 | if (u == v) { |
5684 | if (!fnt_autoconvolute(c1, n, P1) || |
5685 | !fnt_autoconvolute(c2, n, P2) || |
5686 | !fnt_autoconvolute(c3, n, P3)) { |
5687 | goto malloc_error; |
5688 | } |
5689 | } |
5690 | else { |
5691 | if ((vtmp = mpd_calloc(n, sizeof *vtmp)) == NULL) { |
5692 | goto malloc_error; |
5693 | } |
5694 | |
5695 | memcpy(vtmp, v, vlen * (sizeof *vtmp)); |
5696 | if (!fnt_convolute(c1, vtmp, n, P1)) { |
5697 | mpd_free(vtmp); |
5698 | goto malloc_error; |
5699 | } |
5700 | |
5701 | memcpy(vtmp, v, vlen * (sizeof *vtmp)); |
5702 | mpd_uint_zero(vtmp+vlen, n-vlen); |
5703 | if (!fnt_convolute(c2, vtmp, n, P2)) { |
5704 | mpd_free(vtmp); |
5705 | goto malloc_error; |
5706 | } |
5707 | |
5708 | memcpy(vtmp, v, vlen * (sizeof *vtmp)); |
5709 | mpd_uint_zero(vtmp+vlen, n-vlen); |
5710 | if (!fnt_convolute(c3, vtmp, n, P3)) { |
5711 | mpd_free(vtmp); |
5712 | goto malloc_error; |
5713 | } |
5714 | |
5715 | mpd_free(vtmp); |
5716 | } |
5717 | |
5718 | crt3(c1, c2, c3, *rsize); |
5719 | |
5720 | out: |
5721 | #ifdef PPRO |
5722 | mpd_restore_fenv(cw); |
5723 | #endif |
5724 | if (c2) mpd_free(c2); |
5725 | if (c3) mpd_free(c3); |
5726 | return c1; |
5727 | |
5728 | malloc_error: |
5729 | if (c1) mpd_free(c1); |
5730 | c1 = NULL; |
5731 | goto out; |
5732 | } |
5733 | |
5734 | |
5735 | /* |
5736 | * Karatsuba multiplication with FNT/basemul as the base case. |
5737 | */ |
5738 | static int |
5739 | _karatsuba_rec_fnt(mpd_uint_t *c, const mpd_uint_t *a, const mpd_uint_t *b, |
5740 | mpd_uint_t *w, mpd_size_t la, mpd_size_t lb) |
5741 | { |
5742 | mpd_size_t m, lt; |
5743 | |
5744 | assert(la >= lb && lb > 0); |
5745 | assert(la <= 3*(MPD_MAXTRANSFORM_2N/2) || w != NULL); |
5746 | |
5747 | if (la <= 3*(MPD_MAXTRANSFORM_2N/2)) { |
5748 | |
5749 | if (lb <= 192) { |
5750 | _mpd_basemul(c, b, a, lb, la); |
5751 | } |
5752 | else { |
5753 | mpd_uint_t *result; |
5754 | mpd_size_t dummy; |
5755 | |
5756 | if ((result = _mpd_fntmul(a, b, la, lb, &dummy)) == NULL) { |
5757 | return 0; |
5758 | } |
5759 | memcpy(c, result, (la+lb) * (sizeof *result)); |
5760 | mpd_free(result); |
5761 | } |
5762 | return 1; |
5763 | } |
5764 | |
5765 | m = (la+1)/2; /* ceil(la/2) */ |
5766 | |
5767 | /* lb <= m < la */ |
5768 | if (lb <= m) { |
5769 | |
5770 | /* lb can now be larger than la-m */ |
5771 | if (lb > la-m) { |
5772 | lt = lb + lb + 1; /* space needed for result array */ |
5773 | mpd_uint_zero(w, lt); /* clear result array */ |
5774 | if (!_karatsuba_rec_fnt(w, b, a+m, w+lt, lb, la-m)) { /* b*ah */ |
5775 | return 0; /* GCOV_UNLIKELY */ |
5776 | } |
5777 | } |
5778 | else { |
5779 | lt = (la-m) + (la-m) + 1; /* space needed for result array */ |
5780 | mpd_uint_zero(w, lt); /* clear result array */ |
5781 | if (!_karatsuba_rec_fnt(w, a+m, b, w+lt, la-m, lb)) { /* ah*b */ |
5782 | return 0; /* GCOV_UNLIKELY */ |
5783 | } |
5784 | } |
5785 | _mpd_baseaddto(c+m, w, (la-m)+lb); /* add ah*b*B**m */ |
5786 | |
5787 | lt = m + m + 1; /* space needed for the result array */ |
5788 | mpd_uint_zero(w, lt); /* clear result array */ |
5789 | if (!_karatsuba_rec_fnt(w, a, b, w+lt, m, lb)) { /* al*b */ |
5790 | return 0; /* GCOV_UNLIKELY */ |
5791 | } |
5792 | _mpd_baseaddto(c, w, m+lb); /* add al*b */ |
5793 | |
5794 | return 1; |
5795 | } |
5796 | |
5797 | /* la >= lb > m */ |
5798 | memcpy(w, a, m * sizeof *w); |
5799 | w[m] = 0; |
5800 | _mpd_baseaddto(w, a+m, la-m); |
5801 | |
5802 | memcpy(w+(m+1), b, m * sizeof *w); |
5803 | w[m+1+m] = 0; |
5804 | _mpd_baseaddto(w+(m+1), b+m, lb-m); |
5805 | |
5806 | if (!_karatsuba_rec_fnt(c+m, w, w+(m+1), w+2*(m+1), m+1, m+1)) { |
5807 | return 0; /* GCOV_UNLIKELY */ |
5808 | } |
5809 | |
5810 | lt = (la-m) + (la-m) + 1; |
5811 | mpd_uint_zero(w, lt); |
5812 | |
5813 | if (!_karatsuba_rec_fnt(w, a+m, b+m, w+lt, la-m, lb-m)) { |
5814 | return 0; /* GCOV_UNLIKELY */ |
5815 | } |
5816 | |
5817 | _mpd_baseaddto(c+2*m, w, (la-m) + (lb-m)); |
5818 | _mpd_basesubfrom(c+m, w, (la-m) + (lb-m)); |
5819 | |
5820 | lt = m + m + 1; |
5821 | mpd_uint_zero(w, lt); |
5822 | |
5823 | if (!_karatsuba_rec_fnt(w, a, b, w+lt, m, m)) { |
5824 | return 0; /* GCOV_UNLIKELY */ |
5825 | } |
5826 | _mpd_baseaddto(c, w, m+m); |
5827 | _mpd_basesubfrom(c+m, w, m+m); |
5828 | |
5829 | return 1; |
5830 | } |
5831 | |
5832 | /* |
5833 | * Multiply u and v, using Karatsuba multiplication with the FNT as the |
5834 | * base case. Returns a pointer to the result or NULL in case of failure |
5835 | * (malloc error). Conditions: ulen >= vlen, ulen >= 4. |
5836 | */ |
5837 | static mpd_uint_t * |
5838 | _mpd_kmul_fnt(const mpd_uint_t *u, const mpd_uint_t *v, |
5839 | mpd_size_t ulen, mpd_size_t vlen, |
5840 | mpd_size_t *rsize) |
5841 | { |
5842 | mpd_uint_t *result = NULL, *w = NULL; |
5843 | mpd_size_t m; |
5844 | |
5845 | assert(ulen >= 4); |
5846 | assert(ulen >= vlen); |
5847 | |
5848 | *rsize = _kmul_resultsize(ulen, vlen); |
5849 | if ((result = mpd_calloc(*rsize, sizeof *result)) == NULL) { |
5850 | return NULL; |
5851 | } |
5852 | |
5853 | m = _kmul_worksize(ulen, 3*(MPD_MAXTRANSFORM_2N/2)); |
5854 | if (m && ((w = mpd_calloc(m, sizeof *w)) == NULL)) { |
5855 | mpd_free(result); /* GCOV_UNLIKELY */ |
5856 | return NULL; /* GCOV_UNLIKELY */ |
5857 | } |
5858 | |
5859 | if (!_karatsuba_rec_fnt(result, u, v, w, ulen, vlen)) { |
5860 | mpd_free(result); |
5861 | result = NULL; |
5862 | } |
5863 | |
5864 | |
5865 | if (w) mpd_free(w); |
5866 | return result; |
5867 | } |
5868 | |
5869 | |
5870 | /* Deal with the special cases of multiplying infinities. */ |
5871 | static void |
5872 | _mpd_qmul_inf(mpd_t *result, const mpd_t *a, const mpd_t *b, uint32_t *status) |
5873 | { |
5874 | if (mpd_isinfinite(a)) { |
5875 | if (mpd_iszero(b)) { |
5876 | mpd_seterror(result, MPD_Invalid_operation, status); |
5877 | } |
5878 | else { |
5879 | mpd_setspecial(result, mpd_sign(a)^mpd_sign(b), MPD_INF); |
5880 | } |
5881 | return; |
5882 | } |
5883 | assert(mpd_isinfinite(b)); |
5884 | if (mpd_iszero(a)) { |
5885 | mpd_seterror(result, MPD_Invalid_operation, status); |
5886 | } |
5887 | else { |
5888 | mpd_setspecial(result, mpd_sign(a)^mpd_sign(b), MPD_INF); |
5889 | } |
5890 | } |
5891 | |
5892 | /* |
5893 | * Internal function: Multiply a and b. _mpd_qmul deals with specials but |
5894 | * does NOT finalize the result. This is for use in mpd_fma(). |
5895 | */ |
5896 | static inline void |
5897 | _mpd_qmul(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5898 | const mpd_context_t *ctx, uint32_t *status) |
5899 | { |
5900 | const mpd_t *big = a, *small = b; |
5901 | mpd_uint_t *rdata = NULL; |
5902 | mpd_uint_t rbuf[MPD_MINALLOC_MAX]; |
5903 | mpd_size_t rsize, i; |
5904 | |
5905 | |
5906 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
5907 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5908 | return; |
5909 | } |
5910 | _mpd_qmul_inf(result, a, b, status); |
5911 | return; |
5912 | } |
5913 | |
5914 | if (small->len > big->len) { |
5915 | _mpd_ptrswap(&big, &small); |
5916 | } |
5917 | |
5918 | rsize = big->len + small->len; |
5919 | |
5920 | if (big->len == 1) { |
5921 | _mpd_singlemul(result->data, big->data[0], small->data[0]); |
5922 | goto finish; |
5923 | } |
5924 | if (rsize <= (mpd_size_t)MPD_MINALLOC_MAX) { |
5925 | if (big->len == 2) { |
5926 | _mpd_mul_2_le2(rbuf, big->data, small->data, small->len); |
5927 | } |
5928 | else { |
5929 | mpd_uint_zero(rbuf, rsize); |
5930 | if (small->len == 1) { |
5931 | _mpd_shortmul(rbuf, big->data, big->len, small->data[0]); |
5932 | } |
5933 | else { |
5934 | _mpd_basemul(rbuf, small->data, big->data, small->len, big->len); |
5935 | } |
5936 | } |
5937 | if (!mpd_qresize(result, rsize, status)) { |
5938 | return; |
5939 | } |
5940 | for(i = 0; i < rsize; i++) { |
5941 | result->data[i] = rbuf[i]; |
5942 | } |
5943 | goto finish; |
5944 | } |
5945 | |
5946 | |
5947 | if (small->len <= 256) { |
5948 | rdata = mpd_calloc(rsize, sizeof *rdata); |
5949 | if (rdata != NULL) { |
5950 | if (small->len == 1) { |
5951 | _mpd_shortmul(rdata, big->data, big->len, small->data[0]); |
5952 | } |
5953 | else { |
5954 | _mpd_basemul(rdata, small->data, big->data, small->len, big->len); |
5955 | } |
5956 | } |
5957 | } |
5958 | else if (rsize <= 1024) { |
5959 | rdata = _mpd_kmul(big->data, small->data, big->len, small->len, &rsize); |
5960 | } |
5961 | else if (rsize <= 3*MPD_MAXTRANSFORM_2N) { |
5962 | rdata = _mpd_fntmul(big->data, small->data, big->len, small->len, &rsize); |
5963 | } |
5964 | else { |
5965 | rdata = _mpd_kmul_fnt(big->data, small->data, big->len, small->len, &rsize); |
5966 | } |
5967 | |
5968 | if (rdata == NULL) { |
5969 | mpd_seterror(result, MPD_Malloc_error, status); |
5970 | return; |
5971 | } |
5972 | |
5973 | if (mpd_isdynamic_data(result)) { |
5974 | mpd_free(result->data); |
5975 | } |
5976 | result->data = rdata; |
5977 | result->alloc = rsize; |
5978 | mpd_set_dynamic_data(result); |
5979 | |
5980 | |
5981 | finish: |
5982 | mpd_set_flags(result, mpd_sign(a)^mpd_sign(b)); |
5983 | result->exp = big->exp + small->exp; |
5984 | result->len = _mpd_real_size(result->data, rsize); |
5985 | /* resize to smaller cannot fail */ |
5986 | mpd_qresize(result, result->len, status); |
5987 | mpd_setdigits(result); |
5988 | } |
5989 | |
5990 | /* Multiply a and b. */ |
5991 | void |
5992 | mpd_qmul(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5993 | const mpd_context_t *ctx, uint32_t *status) |
5994 | { |
5995 | _mpd_qmul(result, a, b, ctx, status); |
5996 | mpd_qfinalize(result, ctx, status); |
5997 | } |
5998 | |
5999 | /* Multiply a and b. Set NaN/Invalid_operation if the result is inexact. */ |
6000 | static void |
6001 | _mpd_qmul_exact(mpd_t *result, const mpd_t *a, const mpd_t *b, |
6002 | const mpd_context_t *ctx, uint32_t *status) |
6003 | { |
6004 | uint32_t workstatus = 0; |
6005 | |
6006 | mpd_qmul(result, a, b, ctx, &workstatus); |
6007 | *status |= workstatus; |
6008 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
6009 | mpd_seterror(result, MPD_Invalid_operation, status); |
6010 | } |
6011 | } |
6012 | |
6013 | /* Multiply decimal and mpd_ssize_t. */ |
6014 | void |
6015 | mpd_qmul_ssize(mpd_t *result, const mpd_t *a, mpd_ssize_t b, |
6016 | const mpd_context_t *ctx, uint32_t *status) |
6017 | { |
6018 | mpd_context_t maxcontext; |
6019 | MPD_NEW_STATIC(bb,0,0,0,0); |
6020 | |
6021 | mpd_maxcontext(&maxcontext); |
6022 | mpd_qsset_ssize(&bb, b, &maxcontext, status); |
6023 | mpd_qmul(result, a, &bb, ctx, status); |
6024 | mpd_del(&bb); |
6025 | } |
6026 | |
6027 | /* Multiply decimal and mpd_uint_t. */ |
6028 | void |
6029 | mpd_qmul_uint(mpd_t *result, const mpd_t *a, mpd_uint_t b, |
6030 | const mpd_context_t *ctx, uint32_t *status) |
6031 | { |
6032 | mpd_context_t maxcontext; |
6033 | MPD_NEW_STATIC(bb,0,0,0,0); |
6034 | |
6035 | mpd_maxcontext(&maxcontext); |
6036 | mpd_qsset_uint(&bb, b, &maxcontext, status); |
6037 | mpd_qmul(result, a, &bb, ctx, status); |
6038 | mpd_del(&bb); |
6039 | } |
6040 | |
6041 | void |
6042 | mpd_qmul_i32(mpd_t *result, const mpd_t *a, int32_t b, |
6043 | const mpd_context_t *ctx, uint32_t *status) |
6044 | { |
6045 | mpd_qmul_ssize(result, a, b, ctx, status); |
6046 | } |
6047 | |
6048 | void |
6049 | mpd_qmul_u32(mpd_t *result, const mpd_t *a, uint32_t b, |
6050 | const mpd_context_t *ctx, uint32_t *status) |
6051 | { |
6052 | mpd_qmul_uint(result, a, b, ctx, status); |
6053 | } |
6054 | |
6055 | #ifdef CONFIG_64 |
6056 | void |
6057 | mpd_qmul_i64(mpd_t *result, const mpd_t *a, int64_t b, |
6058 | const mpd_context_t *ctx, uint32_t *status) |
6059 | { |
6060 | mpd_qmul_ssize(result, a, b, ctx, status); |
6061 | } |
6062 | |
6063 | void |
6064 | mpd_qmul_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
6065 | const mpd_context_t *ctx, uint32_t *status) |
6066 | { |
6067 | mpd_qmul_uint(result, a, b, ctx, status); |
6068 | } |
6069 | #elif !defined(LEGACY_COMPILER) |
6070 | /* Multiply decimal and int64_t. */ |
6071 | void |
6072 | mpd_qmul_i64(mpd_t *result, const mpd_t *a, int64_t b, |
6073 | const mpd_context_t *ctx, uint32_t *status) |
6074 | { |
6075 | mpd_context_t maxcontext; |
6076 | MPD_NEW_STATIC(bb,0,0,0,0); |
6077 | |
6078 | mpd_maxcontext(&maxcontext); |
6079 | mpd_qset_i64(&bb, b, &maxcontext, status); |
6080 | mpd_qmul(result, a, &bb, ctx, status); |
6081 | mpd_del(&bb); |
6082 | } |
6083 | |
6084 | /* Multiply decimal and uint64_t. */ |
6085 | void |
6086 | mpd_qmul_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
6087 | const mpd_context_t *ctx, uint32_t *status) |
6088 | { |
6089 | mpd_context_t maxcontext; |
6090 | MPD_NEW_STATIC(bb,0,0,0,0); |
6091 | |
6092 | mpd_maxcontext(&maxcontext); |
6093 | mpd_qset_u64(&bb, b, &maxcontext, status); |
6094 | mpd_qmul(result, a, &bb, ctx, status); |
6095 | mpd_del(&bb); |
6096 | } |
6097 | #endif |
6098 | |
6099 | /* Like the minus operator. */ |
6100 | void |
6101 | mpd_qminus(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
6102 | uint32_t *status) |
6103 | { |
6104 | if (mpd_isspecial(a)) { |
6105 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
6106 | return; |
6107 | } |
6108 | } |
6109 | |
6110 | if (mpd_iszero(a) && ctx->round != MPD_ROUND_FLOOR) { |
6111 | mpd_qcopy_abs(result, a, status); |
6112 | } |
6113 | else { |
6114 | mpd_qcopy_negate(result, a, status); |
6115 | } |
6116 | |
6117 | mpd_qfinalize(result, ctx, status); |
6118 | } |
6119 | |
6120 | /* Like the plus operator. */ |
6121 | void |
6122 | mpd_qplus(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
6123 | uint32_t *status) |
6124 | { |
6125 | if (mpd_isspecial(a)) { |
6126 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
6127 | return; |
6128 | } |
6129 | } |
6130 | |
6131 | if (mpd_iszero(a) && ctx->round != MPD_ROUND_FLOOR) { |
6132 | mpd_qcopy_abs(result, a, status); |
6133 | } |
6134 | else { |
6135 | mpd_qcopy(result, a, status); |
6136 | } |
6137 | |
6138 | mpd_qfinalize(result, ctx, status); |
6139 | } |
6140 | |
6141 | /* The largest representable number that is smaller than the operand. */ |
6142 | void |
6143 | mpd_qnext_minus(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
6144 | uint32_t *status) |
6145 | { |
6146 | mpd_context_t workctx; |
6147 | MPD_NEW_CONST(tiny,MPD_POS,mpd_etiny(ctx)-1,1,1,1,1); |
6148 | |
6149 | if (mpd_isspecial(a)) { |
6150 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
6151 | return; |
6152 | } |
6153 | |
6154 | assert(mpd_isinfinite(a)); |
6155 | if (mpd_isnegative(a)) { |
6156 | mpd_qcopy(result, a, status); |
6157 | return; |
6158 | } |
6159 | else { |
6160 | mpd_clear_flags(result); |
6161 | mpd_qmaxcoeff(result, ctx, status); |
6162 | if (mpd_isnan(result)) { |
6163 | return; |
6164 | } |
6165 | result->exp = mpd_etop(ctx); |
6166 | return; |
6167 | } |
6168 | } |
6169 | |
6170 | mpd_workcontext(&workctx, ctx); |
6171 | workctx.round = MPD_ROUND_FLOOR; |
6172 | |
6173 | if (!mpd_qcopy(result, a, status)) { |
6174 | return; |
6175 | } |
6176 | |
6177 | mpd_qfinalize(result, &workctx, &workctx.status); |
6178 | if (workctx.status&(MPD_Inexact|MPD_Errors)) { |
6179 | *status |= (workctx.status&MPD_Errors); |
6180 | return; |
6181 | } |
6182 | |
6183 | workctx.status = 0; |
6184 | mpd_qsub(result, a, &tiny, &workctx, &workctx.status); |
6185 | *status |= (workctx.status&MPD_Errors); |
6186 | } |
6187 | |
6188 | /* The smallest representable number that is larger than the operand. */ |
6189 | void |
6190 | mpd_qnext_plus(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
6191 | uint32_t *status) |
6192 | { |
6193 | mpd_context_t workctx; |
6194 | MPD_NEW_CONST(tiny,MPD_POS,mpd_etiny(ctx)-1,1,1,1,1); |
6195 | |
6196 | if (mpd_isspecial(a)) { |
6197 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
6198 | return; |
6199 | } |
6200 | |
6201 | assert(mpd_isinfinite(a)); |
6202 | if (mpd_ispositive(a)) { |
6203 | mpd_qcopy(result, a, status); |
6204 | } |
6205 | else { |
6206 | mpd_clear_flags(result); |
6207 | mpd_qmaxcoeff(result, ctx, status); |
6208 | if (mpd_isnan(result)) { |
6209 | return; |
6210 | } |
6211 | mpd_set_flags(result, MPD_NEG); |
6212 | result->exp = mpd_etop(ctx); |
6213 | } |
6214 | return; |
6215 | } |
6216 | |
6217 | mpd_workcontext(&workctx, ctx); |
6218 | workctx.round = MPD_ROUND_CEILING; |
6219 | |
6220 | if (!mpd_qcopy(result, a, status)) { |
6221 | return; |
6222 | } |
6223 | |
6224 | mpd_qfinalize(result, &workctx, &workctx.status); |
6225 | if (workctx.status & (MPD_Inexact|MPD_Errors)) { |
6226 | *status |= (workctx.status&MPD_Errors); |
6227 | return; |
6228 | } |
6229 | |
6230 | workctx.status = 0; |
6231 | mpd_qadd(result, a, &tiny, &workctx, &workctx.status); |
6232 | *status |= (workctx.status&MPD_Errors); |
6233 | } |
6234 | |
6235 | /* |
6236 | * The number closest to the first operand that is in the direction towards |
6237 | * the second operand. |
6238 | */ |
6239 | void |
6240 | mpd_qnext_toward(mpd_t *result, const mpd_t *a, const mpd_t *b, |
6241 | const mpd_context_t *ctx, uint32_t *status) |
6242 | { |
6243 | int c; |
6244 | |
6245 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
6246 | return; |
6247 | } |
6248 | |
6249 | c = _mpd_cmp(a, b); |
6250 | if (c == 0) { |
6251 | mpd_qcopy_sign(result, a, b, status); |
6252 | return; |
6253 | } |
6254 | |
6255 | if (c < 0) { |
6256 | mpd_qnext_plus(result, a, ctx, status); |
6257 | } |
6258 | else { |
6259 | mpd_qnext_minus(result, a, ctx, status); |
6260 | } |
6261 | |
6262 | if (mpd_isinfinite(result)) { |
6263 | *status |= (MPD_Overflow|MPD_Rounded|MPD_Inexact); |
6264 | } |
6265 | else if (mpd_adjexp(result) < ctx->emin) { |
6266 | *status |= (MPD_Underflow|MPD_Subnormal|MPD_Rounded|MPD_Inexact); |
6267 | if (mpd_iszero(result)) { |
6268 | *status |= MPD_Clamped; |
6269 | } |
6270 | } |
6271 | } |
6272 | |
6273 | /* |
6274 | * Internal function: Integer power with mpd_uint_t exponent. The function |
6275 | * can fail with MPD_Malloc_error. |
6276 | * |
6277 | * The error is equal to the error incurred in k-1 multiplications. Assuming |
6278 | * the upper bound for the relative error in each operation: |
6279 | * |
6280 | * abs(err) = 5 * 10**-prec |
6281 | * result = x**k * (1 + err)**(k-1) |
6282 | */ |
6283 | static inline void |
6284 | _mpd_qpow_uint(mpd_t *result, const mpd_t *base, mpd_uint_t exp, |
6285 | uint8_t resultsign, const mpd_context_t *ctx, uint32_t *status) |
6286 | { |
6287 | uint32_t workstatus = 0; |
6288 | mpd_uint_t n; |
6289 | |
6290 | if (exp == 0) { |
6291 | _settriple(result, resultsign, 1, 0); /* GCOV_NOT_REACHED */ |
6292 | return; /* GCOV_NOT_REACHED */ |
6293 | } |
6294 | |
6295 | if (!mpd_qcopy(result, base, status)) { |
6296 | return; |
6297 | } |
6298 | |
6299 | n = mpd_bits[mpd_bsr(exp)]; |
6300 | while (n >>= 1) { |
6301 | mpd_qmul(result, result, result, ctx, &workstatus); |
6302 | if (exp & n) { |
6303 | mpd_qmul(result, result, base, ctx, &workstatus); |
6304 | } |
6305 | if (mpd_isspecial(result) || |
6306 | (mpd_iszerocoeff(result) && (workstatus & MPD_Clamped))) { |
6307 | break; |
6308 | } |
6309 | } |
6310 | |
6311 | *status |= workstatus; |
6312 | mpd_set_sign(result, resultsign); |
6313 | } |
6314 | |
6315 | /* |
6316 | * Internal function: Integer power with mpd_t exponent, tbase and texp |
6317 | * are modified!! Function can fail with MPD_Malloc_error. |
6318 | * |
6319 | * The error is equal to the error incurred in k multiplications. Assuming |
6320 | * the upper bound for the relative error in each operation: |
6321 | * |
6322 | * abs(err) = 5 * 10**-prec |
6323 | * result = x**k * (1 + err)**k |
6324 | */ |
6325 | static inline void |
6326 | _mpd_qpow_mpd(mpd_t *result, mpd_t *tbase, mpd_t *texp, uint8_t resultsign, |
6327 | const mpd_context_t *ctx, uint32_t *status) |
6328 | { |
6329 | uint32_t workstatus = 0; |
6330 | mpd_context_t maxctx; |
6331 | MPD_NEW_CONST(two,0,0,1,1,1,2); |
6332 | |
6333 | |
6334 | mpd_maxcontext(&maxctx); |
6335 | |
6336 | /* resize to smaller cannot fail */ |
6337 | mpd_qcopy(result, &one, status); |
6338 | |
6339 | while (!mpd_iszero(texp)) { |
6340 | if (mpd_isodd(texp)) { |
6341 | mpd_qmul(result, result, tbase, ctx, &workstatus); |
6342 | *status |= workstatus; |
6343 | if (mpd_isspecial(result) || |
6344 | (mpd_iszerocoeff(result) && (workstatus & MPD_Clamped))) { |
6345 | break; |
6346 | } |
6347 | } |
6348 | mpd_qmul(tbase, tbase, tbase, ctx, &workstatus); |
6349 | mpd_qdivint(texp, texp, &two, &maxctx, &workstatus); |
6350 | if (mpd_isnan(tbase) || mpd_isnan(texp)) { |
6351 | mpd_seterror(result, workstatus&MPD_Errors, status); |
6352 | return; |
6353 | } |
6354 | } |
6355 | mpd_set_sign(result, resultsign); |
6356 | } |
6357 | |
6358 | /* |
6359 | * The power function for integer exponents. Relative error _before_ the |
6360 | * final rounding to prec: |
6361 | * abs(result - base**exp) < 0.1 * 10**-prec * abs(base**exp) |
6362 | */ |
6363 | static void |
6364 | _mpd_qpow_int(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6365 | uint8_t resultsign, |
6366 | const mpd_context_t *ctx, uint32_t *status) |
6367 | { |
6368 | mpd_context_t workctx; |
6369 | MPD_NEW_STATIC(tbase,0,0,0,0); |
6370 | MPD_NEW_STATIC(texp,0,0,0,0); |
6371 | mpd_uint_t n; |
6372 | |
6373 | |
6374 | mpd_workcontext(&workctx, ctx); |
6375 | workctx.prec += (exp->digits + exp->exp + 2); |
6376 | workctx.round = MPD_ROUND_HALF_EVEN; |
6377 | workctx.clamp = 0; |
6378 | if (mpd_isnegative(exp)) { |
6379 | uint32_t workstatus = 0; |
6380 | workctx.prec += 1; |
6381 | mpd_qdiv(&tbase, &one, base, &workctx, &workstatus); |
6382 | *status |= workstatus; |
6383 | if (workstatus&MPD_Errors) { |
6384 | mpd_setspecial(result, MPD_POS, MPD_NAN); |
6385 | goto finish; |
6386 | } |
6387 | } |
6388 | else { |
6389 | if (!mpd_qcopy(&tbase, base, status)) { |
6390 | mpd_setspecial(result, MPD_POS, MPD_NAN); |
6391 | goto finish; |
6392 | } |
6393 | } |
6394 | |
6395 | n = mpd_qabs_uint(exp, &workctx.status); |
6396 | if (workctx.status&MPD_Invalid_operation) { |
6397 | if (!mpd_qcopy(&texp, exp, status)) { |
6398 | mpd_setspecial(result, MPD_POS, MPD_NAN); /* GCOV_UNLIKELY */ |
6399 | goto finish; /* GCOV_UNLIKELY */ |
6400 | } |
6401 | _mpd_qpow_mpd(result, &tbase, &texp, resultsign, &workctx, status); |
6402 | } |
6403 | else { |
6404 | _mpd_qpow_uint(result, &tbase, n, resultsign, &workctx, status); |
6405 | } |
6406 | |
6407 | if (mpd_isinfinite(result)) { |
6408 | /* for ROUND_DOWN, ROUND_FLOOR, etc. */ |
6409 | _settriple(result, resultsign, 1, MPD_EXP_INF); |
6410 | } |
6411 | |
6412 | finish: |
6413 | mpd_del(&tbase); |
6414 | mpd_del(&texp); |
6415 | mpd_qfinalize(result, ctx, status); |
6416 | } |
6417 | |
6418 | /* |
6419 | * If the exponent is infinite and base equals one, the result is one |
6420 | * with a coefficient of length prec. Otherwise, result is undefined. |
6421 | * Return the value of the comparison against one. |
6422 | */ |
6423 | static int |
6424 | _qcheck_pow_one_inf(mpd_t *result, const mpd_t *base, uint8_t resultsign, |
6425 | const mpd_context_t *ctx, uint32_t *status) |
6426 | { |
6427 | mpd_ssize_t shift; |
6428 | int cmp; |
6429 | |
6430 | if ((cmp = _mpd_cmp(base, &one)) == 0) { |
6431 | shift = ctx->prec-1; |
6432 | mpd_qshiftl(result, &one, shift, status); |
6433 | result->exp = -shift; |
6434 | mpd_set_flags(result, resultsign); |
6435 | *status |= (MPD_Inexact|MPD_Rounded); |
6436 | } |
6437 | |
6438 | return cmp; |
6439 | } |
6440 | |
6441 | /* |
6442 | * If abs(base) equals one, calculate the correct power of one result. |
6443 | * Otherwise, result is undefined. Return the value of the comparison |
6444 | * against 1. |
6445 | * |
6446 | * This is an internal function that does not check for specials. |
6447 | */ |
6448 | static int |
6449 | _qcheck_pow_one(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6450 | uint8_t resultsign, |
6451 | const mpd_context_t *ctx, uint32_t *status) |
6452 | { |
6453 | uint32_t workstatus = 0; |
6454 | mpd_ssize_t shift; |
6455 | int cmp; |
6456 | |
6457 | if ((cmp = _mpd_cmp_abs(base, &one)) == 0) { |
6458 | if (_mpd_isint(exp)) { |
6459 | if (mpd_isnegative(exp)) { |
6460 | _settriple(result, resultsign, 1, 0); |
6461 | return 0; |
6462 | } |
6463 | /* 1.000**3 = 1.000000000 */ |
6464 | mpd_qmul_ssize(result, exp, -base->exp, ctx, &workstatus); |
6465 | if (workstatus&MPD_Errors) { |
6466 | *status |= (workstatus&MPD_Errors); |
6467 | return 0; |
6468 | } |
6469 | /* digits-1 after exponentiation */ |
6470 | shift = mpd_qget_ssize(result, &workstatus); |
6471 | /* shift is MPD_SSIZE_MAX if result is too large */ |
6472 | if (shift > ctx->prec-1) { |
6473 | shift = ctx->prec-1; |
6474 | *status |= MPD_Rounded; |
6475 | } |
6476 | } |
6477 | else if (mpd_ispositive(base)) { |
6478 | shift = ctx->prec-1; |
6479 | *status |= (MPD_Inexact|MPD_Rounded); |
6480 | } |
6481 | else { |
6482 | return -2; /* GCOV_NOT_REACHED */ |
6483 | } |
6484 | if (!mpd_qshiftl(result, &one, shift, status)) { |
6485 | return 0; |
6486 | } |
6487 | result->exp = -shift; |
6488 | mpd_set_flags(result, resultsign); |
6489 | } |
6490 | |
6491 | return cmp; |
6492 | } |
6493 | |
6494 | /* |
6495 | * Detect certain over/underflow of x**y. |
6496 | * ACL2 proof: pow-bounds.lisp. |
6497 | * |
6498 | * Symbols: |
6499 | * |
6500 | * e: EXP_INF or EXP_CLAMP |
6501 | * x: base |
6502 | * y: exponent |
6503 | * |
6504 | * omega(e) = log10(abs(e)) |
6505 | * zeta(x) = log10(abs(log10(x))) |
6506 | * theta(y) = log10(abs(y)) |
6507 | * |
6508 | * Upper and lower bounds: |
6509 | * |
6510 | * ub_omega(e) = ceil(log10(abs(e))) |
6511 | * lb_theta(y) = floor(log10(abs(y))) |
6512 | * |
6513 | * | floor(log10(floor(abs(log10(x))))) if x < 1/10 or x >= 10 |
6514 | * lb_zeta(x) = | floor(log10(abs(x-1)/10)) if 1/10 <= x < 1 |
6515 | * | floor(log10(abs((x-1)/100))) if 1 < x < 10 |
6516 | * |
6517 | * ub_omega(e) and lb_theta(y) are obviously upper and lower bounds |
6518 | * for omega(e) and theta(y). |
6519 | * |
6520 | * lb_zeta is a lower bound for zeta(x): |
6521 | * |
6522 | * x < 1/10 or x >= 10: |
6523 | * |
6524 | * abs(log10(x)) >= 1, so the outer log10 is well defined. Since log10 |
6525 | * is strictly increasing, the end result is a lower bound. |
6526 | * |
6527 | * 1/10 <= x < 1: |
6528 | * |
6529 | * We use: log10(x) <= (x-1)/log(10) |
6530 | * abs(log10(x)) >= abs(x-1)/log(10) |
6531 | * abs(log10(x)) >= abs(x-1)/10 |
6532 | * |
6533 | * 1 < x < 10: |
6534 | * |
6535 | * We use: (x-1)/(x*log(10)) < log10(x) |
6536 | * abs((x-1)/100) < abs(log10(x)) |
6537 | * |
6538 | * XXX: abs((x-1)/10) would work, need ACL2 proof. |
6539 | * |
6540 | * |
6541 | * Let (0 < x < 1 and y < 0) or (x > 1 and y > 0). (H1) |
6542 | * Let ub_omega(exp_inf) < lb_zeta(x) + lb_theta(y) (H2) |
6543 | * |
6544 | * Then: |
6545 | * log10(abs(exp_inf)) < log10(abs(log10(x))) + log10(abs(y)). (1) |
6546 | * exp_inf < log10(x) * y (2) |
6547 | * 10**exp_inf < x**y (3) |
6548 | * |
6549 | * Let (0 < x < 1 and y > 0) or (x > 1 and y < 0). (H3) |
6550 | * Let ub_omega(exp_clamp) < lb_zeta(x) + lb_theta(y) (H4) |
6551 | * |
6552 | * Then: |
6553 | * log10(abs(exp_clamp)) < log10(abs(log10(x))) + log10(abs(y)). (4) |
6554 | * log10(x) * y < exp_clamp (5) |
6555 | * x**y < 10**exp_clamp (6) |
6556 | * |
6557 | */ |
6558 | static mpd_ssize_t |
6559 | _lower_bound_zeta(const mpd_t *x, uint32_t *status) |
6560 | { |
6561 | mpd_context_t maxctx; |
6562 | MPD_NEW_STATIC(scratch,0,0,0,0); |
6563 | mpd_ssize_t t, u; |
6564 | |
6565 | t = mpd_adjexp(x); |
6566 | if (t > 0) { |
6567 | /* x >= 10 -> floor(log10(floor(abs(log10(x))))) */ |
6568 | return mpd_exp_digits(t) - 1; |
6569 | } |
6570 | else if (t < -1) { |
6571 | /* x < 1/10 -> floor(log10(floor(abs(log10(x))))) */ |
6572 | return mpd_exp_digits(t+1) - 1; |
6573 | } |
6574 | else { |
6575 | mpd_maxcontext(&maxctx); |
6576 | mpd_qsub(&scratch, x, &one, &maxctx, status); |
6577 | if (mpd_isspecial(&scratch)) { |
6578 | mpd_del(&scratch); |
6579 | return MPD_SSIZE_MAX; |
6580 | } |
6581 | u = mpd_adjexp(&scratch); |
6582 | mpd_del(&scratch); |
6583 | |
6584 | /* t == -1, 1/10 <= x < 1 -> floor(log10(abs(x-1)/10)) |
6585 | * t == 0, 1 < x < 10 -> floor(log10(abs(x-1)/100)) */ |
6586 | return (t == 0) ? u-2 : u-1; |
6587 | } |
6588 | } |
6589 | |
6590 | /* |
6591 | * Detect cases of certain overflow/underflow in the power function. |
6592 | * Assumptions: x != 1, y != 0. The proof above is for positive x. |
6593 | * If x is negative and y is an odd integer, x**y == -(abs(x)**y), |
6594 | * so the analysis does not change. |
6595 | */ |
6596 | static int |
6597 | _qcheck_pow_bounds(mpd_t *result, const mpd_t *x, const mpd_t *y, |
6598 | uint8_t resultsign, |
6599 | const mpd_context_t *ctx, uint32_t *status) |
6600 | { |
6601 | MPD_NEW_SHARED(abs_x, x); |
6602 | mpd_ssize_t ub_omega, lb_zeta, lb_theta; |
6603 | uint8_t sign; |
6604 | |
6605 | mpd_set_positive(&abs_x); |
6606 | |
6607 | lb_theta = mpd_adjexp(y); |
6608 | lb_zeta = _lower_bound_zeta(&abs_x, status); |
6609 | if (lb_zeta == MPD_SSIZE_MAX) { |
6610 | mpd_seterror(result, MPD_Malloc_error, status); |
6611 | return 1; |
6612 | } |
6613 | |
6614 | sign = (mpd_adjexp(&abs_x) < 0) ^ mpd_sign(y); |
6615 | if (sign == 0) { |
6616 | /* (0 < |x| < 1 and y < 0) or (|x| > 1 and y > 0) */ |
6617 | ub_omega = mpd_exp_digits(ctx->emax); |
6618 | if (ub_omega < lb_zeta + lb_theta) { |
6619 | _settriple(result, resultsign, 1, MPD_EXP_INF); |
6620 | mpd_qfinalize(result, ctx, status); |
6621 | return 1; |
6622 | } |
6623 | } |
6624 | else { |
6625 | /* (0 < |x| < 1 and y > 0) or (|x| > 1 and y < 0). */ |
6626 | ub_omega = mpd_exp_digits(mpd_etiny(ctx)); |
6627 | if (ub_omega < lb_zeta + lb_theta) { |
6628 | _settriple(result, resultsign, 1, mpd_etiny(ctx)-1); |
6629 | mpd_qfinalize(result, ctx, status); |
6630 | return 1; |
6631 | } |
6632 | } |
6633 | |
6634 | return 0; |
6635 | } |
6636 | |
6637 | /* |
6638 | * TODO: Implement algorithm for computing exact powers from decimal.py. |
6639 | * In order to prevent infinite loops, this has to be called before |
6640 | * using Ziv's strategy for correct rounding. |
6641 | */ |
6642 | /* |
6643 | static int |
6644 | _mpd_qpow_exact(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6645 | const mpd_context_t *ctx, uint32_t *status) |
6646 | { |
6647 | return 0; |
6648 | } |
6649 | */ |
6650 | |
6651 | /* |
6652 | * The power function for real exponents. |
6653 | * Relative error: abs(result - e**y) < e**y * 1/5 * 10**(-prec - 1) |
6654 | */ |
6655 | static void |
6656 | _mpd_qpow_real(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6657 | const mpd_context_t *ctx, uint32_t *status) |
6658 | { |
6659 | mpd_context_t workctx; |
6660 | MPD_NEW_STATIC(texp,0,0,0,0); |
6661 | |
6662 | if (!mpd_qcopy(&texp, exp, status)) { |
6663 | mpd_seterror(result, MPD_Malloc_error, status); |
6664 | return; |
6665 | } |
6666 | |
6667 | mpd_maxcontext(&workctx); |
6668 | workctx.prec = (base->digits > ctx->prec) ? base->digits : ctx->prec; |
6669 | workctx.prec += (4 + MPD_EXPDIGITS); |
6670 | workctx.round = MPD_ROUND_HALF_EVEN; |
6671 | workctx.allcr = ctx->allcr; |
6672 | |
6673 | /* |
6674 | * extra := MPD_EXPDIGITS = MPD_EXP_MAX_T |
6675 | * wp := prec + 4 + extra |
6676 | * abs(err) < 5 * 10**-wp |
6677 | * y := log(base) * exp |
6678 | * Calculate: |
6679 | * 1) e**(y * (1 + err)**2) * (1 + err) |
6680 | * = e**y * e**(y * (2*err + err**2)) * (1 + err) |
6681 | * ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
6682 | * Relative error of the underlined term: |
6683 | * 2) abs(e**(y * (2*err + err**2)) - 1) |
6684 | * Case abs(y) >= 10**extra: |
6685 | * 3) adjexp(y)+1 > log10(abs(y)) >= extra |
6686 | * This triggers the Overflow/Underflow shortcut in _mpd_qexp(), |
6687 | * so no further analysis is necessary. |
6688 | * Case abs(y) < 10**extra: |
6689 | * 4) abs(y * (2*err + err**2)) < 1/5 * 10**(-prec - 2) |
6690 | * Use (see _mpd_qexp): |
6691 | * 5) abs(x) <= 9/10 * 10**-p ==> abs(e**x - 1) < 10**-p |
6692 | * With 2), 4) and 5): |
6693 | * 6) abs(e**(y * (2*err + err**2)) - 1) < 10**(-prec - 2) |
6694 | * The complete relative error of 1) is: |
6695 | * 7) abs(result - e**y) < e**y * 1/5 * 10**(-prec - 1) |
6696 | */ |
6697 | mpd_qln(result, base, &workctx, &workctx.status); |
6698 | mpd_qmul(result, result, &texp, &workctx, &workctx.status); |
6699 | mpd_qexp(result, result, &workctx, status); |
6700 | |
6701 | mpd_del(&texp); |
6702 | *status |= (workctx.status&MPD_Errors); |
6703 | *status |= (MPD_Inexact|MPD_Rounded); |
6704 | } |
6705 | |
6706 | /* The power function: base**exp */ |
6707 | void |
6708 | mpd_qpow(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6709 | const mpd_context_t *ctx, uint32_t *status) |
6710 | { |
6711 | uint8_t resultsign = 0; |
6712 | int intexp = 0; |
6713 | int cmp; |
6714 | |
6715 | if (mpd_isspecial(base) || mpd_isspecial(exp)) { |
6716 | if (mpd_qcheck_nans(result, base, exp, ctx, status)) { |
6717 | return; |
6718 | } |
6719 | } |
6720 | if (mpd_isinteger(exp)) { |
6721 | intexp = 1; |
6722 | resultsign = mpd_isnegative(base) && mpd_isodd(exp); |
6723 | } |
6724 | |
6725 | if (mpd_iszero(base)) { |
6726 | if (mpd_iszero(exp)) { |
6727 | mpd_seterror(result, MPD_Invalid_operation, status); |
6728 | } |
6729 | else if (mpd_isnegative(exp)) { |
6730 | mpd_setspecial(result, resultsign, MPD_INF); |
6731 | } |
6732 | else { |
6733 | _settriple(result, resultsign, 0, 0); |
6734 | } |
6735 | return; |
6736 | } |
6737 | if (mpd_isnegative(base)) { |
6738 | if (!intexp || mpd_isinfinite(exp)) { |
6739 | mpd_seterror(result, MPD_Invalid_operation, status); |
6740 | return; |
6741 | } |
6742 | } |
6743 | if (mpd_isinfinite(exp)) { |
6744 | /* power of one */ |
6745 | cmp = _qcheck_pow_one_inf(result, base, resultsign, ctx, status); |
6746 | if (cmp == 0) { |
6747 | return; |
6748 | } |
6749 | else { |
6750 | cmp *= mpd_arith_sign(exp); |
6751 | if (cmp < 0) { |
6752 | _settriple(result, resultsign, 0, 0); |
6753 | } |
6754 | else { |
6755 | mpd_setspecial(result, resultsign, MPD_INF); |
6756 | } |
6757 | } |
6758 | return; |
6759 | } |
6760 | if (mpd_isinfinite(base)) { |
6761 | if (mpd_iszero(exp)) { |
6762 | _settriple(result, resultsign, 1, 0); |
6763 | } |
6764 | else if (mpd_isnegative(exp)) { |
6765 | _settriple(result, resultsign, 0, 0); |
6766 | } |
6767 | else { |
6768 | mpd_setspecial(result, resultsign, MPD_INF); |
6769 | } |
6770 | return; |
6771 | } |
6772 | if (mpd_iszero(exp)) { |
6773 | _settriple(result, resultsign, 1, 0); |
6774 | return; |
6775 | } |
6776 | if (_qcheck_pow_one(result, base, exp, resultsign, ctx, status) == 0) { |
6777 | return; |
6778 | } |
6779 | if (_qcheck_pow_bounds(result, base, exp, resultsign, ctx, status)) { |
6780 | return; |
6781 | } |
6782 | |
6783 | if (intexp) { |
6784 | _mpd_qpow_int(result, base, exp, resultsign, ctx, status); |
6785 | } |
6786 | else { |
6787 | _mpd_qpow_real(result, base, exp, ctx, status); |
6788 | if (!mpd_isspecial(result) && _mpd_cmp(result, &one) == 0) { |
6789 | mpd_ssize_t shift = ctx->prec-1; |
6790 | mpd_qshiftl(result, &one, shift, status); |
6791 | result->exp = -shift; |
6792 | } |
6793 | if (mpd_isinfinite(result)) { |
6794 | /* for ROUND_DOWN, ROUND_FLOOR, etc. */ |
6795 | _settriple(result, MPD_POS, 1, MPD_EXP_INF); |
6796 | } |
6797 | mpd_qfinalize(result, ctx, status); |
6798 | } |
6799 | } |
6800 | |
6801 | /* |
6802 | * Internal function: Integer powmod with mpd_uint_t exponent, base is modified! |
6803 | * Function can fail with MPD_Malloc_error. |
6804 | */ |
6805 | static inline void |
6806 | _mpd_qpowmod_uint(mpd_t *result, mpd_t *base, mpd_uint_t exp, |
6807 | const mpd_t *mod, uint32_t *status) |
6808 | { |
6809 | mpd_context_t maxcontext; |
6810 | |
6811 | mpd_maxcontext(&maxcontext); |
6812 | |
6813 | /* resize to smaller cannot fail */ |
6814 | mpd_qcopy(result, &one, status); |
6815 | |
6816 | while (exp > 0) { |
6817 | if (exp & 1) { |
6818 | _mpd_qmul_exact(result, result, base, &maxcontext, status); |
6819 | mpd_qrem(result, result, mod, &maxcontext, status); |
6820 | } |
6821 | _mpd_qmul_exact(base, base, base, &maxcontext, status); |
6822 | mpd_qrem(base, base, mod, &maxcontext, status); |
6823 | exp >>= 1; |
6824 | } |
6825 | } |
6826 | |
6827 | /* The powmod function: (base**exp) % mod */ |
6828 | void |
6829 | mpd_qpowmod(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6830 | const mpd_t *mod, |
6831 | const mpd_context_t *ctx, uint32_t *status) |
6832 | { |
6833 | mpd_context_t maxcontext; |
6834 | MPD_NEW_STATIC(tbase,0,0,0,0); |
6835 | MPD_NEW_STATIC(texp,0,0,0,0); |
6836 | MPD_NEW_STATIC(tmod,0,0,0,0); |
6837 | MPD_NEW_STATIC(tmp,0,0,0,0); |
6838 | MPD_NEW_CONST(two,0,0,1,1,1,2); |
6839 | mpd_ssize_t tbase_exp, texp_exp; |
6840 | mpd_ssize_t i; |
6841 | mpd_t t; |
6842 | mpd_uint_t r; |
6843 | uint8_t sign; |
6844 | |
6845 | |
6846 | if (mpd_isspecial(base) || mpd_isspecial(exp) || mpd_isspecial(mod)) { |
6847 | if (mpd_qcheck_3nans(result, base, exp, mod, ctx, status)) { |
6848 | return; |
6849 | } |
6850 | mpd_seterror(result, MPD_Invalid_operation, status); |
6851 | return; |
6852 | } |
6853 | |
6854 | |
6855 | if (!_mpd_isint(base) || !_mpd_isint(exp) || !_mpd_isint(mod)) { |
6856 | mpd_seterror(result, MPD_Invalid_operation, status); |
6857 | return; |
6858 | } |
6859 | if (mpd_iszerocoeff(mod)) { |
6860 | mpd_seterror(result, MPD_Invalid_operation, status); |
6861 | return; |
6862 | } |
6863 | if (mod->digits+mod->exp > ctx->prec) { |
6864 | mpd_seterror(result, MPD_Invalid_operation, status); |
6865 | return; |
6866 | } |
6867 | |
6868 | sign = (mpd_isnegative(base)) && (mpd_isodd(exp)); |
6869 | if (mpd_iszerocoeff(exp)) { |
6870 | if (mpd_iszerocoeff(base)) { |
6871 | mpd_seterror(result, MPD_Invalid_operation, status); |
6872 | return; |
6873 | } |
6874 | r = (_mpd_cmp_abs(mod, &one)==0) ? 0 : 1; |
6875 | _settriple(result, sign, r, 0); |
6876 | return; |
6877 | } |
6878 | if (mpd_isnegative(exp)) { |
6879 | mpd_seterror(result, MPD_Invalid_operation, status); |
6880 | return; |
6881 | } |
6882 | if (mpd_iszerocoeff(base)) { |
6883 | _settriple(result, sign, 0, 0); |
6884 | return; |
6885 | } |
6886 | |
6887 | mpd_maxcontext(&maxcontext); |
6888 | |
6889 | mpd_qrescale(&tmod, mod, 0, &maxcontext, &maxcontext.status); |
6890 | if (maxcontext.status&MPD_Errors) { |
6891 | mpd_seterror(result, maxcontext.status&MPD_Errors, status); |
6892 | goto out; |
6893 | } |
6894 | maxcontext.status = 0; |
6895 | mpd_set_positive(&tmod); |
6896 | |
6897 | mpd_qround_to_int(&tbase, base, &maxcontext, status); |
6898 | mpd_set_positive(&tbase); |
6899 | tbase_exp = tbase.exp; |
6900 | tbase.exp = 0; |
6901 | |
6902 | mpd_qround_to_int(&texp, exp, &maxcontext, status); |
6903 | texp_exp = texp.exp; |
6904 | texp.exp = 0; |
6905 | |
6906 | /* base = (base.int % modulo * pow(10, base.exp, modulo)) % modulo */ |
6907 | mpd_qrem(&tbase, &tbase, &tmod, &maxcontext, status); |
6908 | mpd_qshiftl(result, &one, tbase_exp, status); |
6909 | mpd_qrem(result, result, &tmod, &maxcontext, status); |
6910 | _mpd_qmul_exact(&tbase, &tbase, result, &maxcontext, status); |
6911 | mpd_qrem(&tbase, &tbase, &tmod, &maxcontext, status); |
6912 | if (mpd_isspecial(&tbase) || |
6913 | mpd_isspecial(&texp) || |
6914 | mpd_isspecial(&tmod)) { |
6915 | goto mpd_errors; |
6916 | } |
6917 | |
6918 | for (i = 0; i < texp_exp; i++) { |
6919 | _mpd_qpowmod_uint(&tmp, &tbase, 10, &tmod, status); |
6920 | t = tmp; |
6921 | tmp = tbase; |
6922 | tbase = t; |
6923 | } |
6924 | if (mpd_isspecial(&tbase)) { |
6925 | goto mpd_errors; /* GCOV_UNLIKELY */ |
6926 | } |
6927 | |
6928 | /* resize to smaller cannot fail */ |
6929 | mpd_qcopy(result, &one, status); |
6930 | while (mpd_isfinite(&texp) && !mpd_iszero(&texp)) { |
6931 | if (mpd_isodd(&texp)) { |
6932 | _mpd_qmul_exact(result, result, &tbase, &maxcontext, status); |
6933 | mpd_qrem(result, result, &tmod, &maxcontext, status); |
6934 | } |
6935 | _mpd_qmul_exact(&tbase, &tbase, &tbase, &maxcontext, status); |
6936 | mpd_qrem(&tbase, &tbase, &tmod, &maxcontext, status); |
6937 | mpd_qdivint(&texp, &texp, &two, &maxcontext, status); |
6938 | } |
6939 | if (mpd_isspecial(&texp) || mpd_isspecial(&tbase) || |
6940 | mpd_isspecial(&tmod) || mpd_isspecial(result)) { |
6941 | /* MPD_Malloc_error */ |
6942 | goto mpd_errors; |
6943 | } |
6944 | else { |
6945 | mpd_set_sign(result, sign); |
6946 | } |
6947 | |
6948 | out: |
6949 | mpd_del(&tbase); |
6950 | mpd_del(&texp); |
6951 | mpd_del(&tmod); |
6952 | mpd_del(&tmp); |
6953 | return; |
6954 | |
6955 | mpd_errors: |
6956 | mpd_setspecial(result, MPD_POS, MPD_NAN); |
6957 | goto out; |
6958 | } |
6959 | |
6960 | void |
6961 | mpd_qquantize(mpd_t *result, const mpd_t *a, const mpd_t *b, |
6962 | const mpd_context_t *ctx, uint32_t *status) |
6963 | { |
6964 | uint32_t workstatus = 0; |
6965 | mpd_ssize_t b_exp = b->exp; |
6966 | mpd_ssize_t expdiff, shift; |
6967 | mpd_uint_t rnd; |
6968 | |
6969 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
6970 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
6971 | return; |
6972 | } |
6973 | if (mpd_isinfinite(a) && mpd_isinfinite(b)) { |
6974 | mpd_qcopy(result, a, status); |
6975 | return; |
6976 | } |
6977 | mpd_seterror(result, MPD_Invalid_operation, status); |
6978 | return; |
6979 | } |
6980 | |
6981 | if (b->exp > ctx->emax || b->exp < mpd_etiny(ctx)) { |
6982 | mpd_seterror(result, MPD_Invalid_operation, status); |
6983 | return; |
6984 | } |
6985 | |
6986 | if (mpd_iszero(a)) { |
6987 | _settriple(result, mpd_sign(a), 0, b->exp); |
6988 | mpd_qfinalize(result, ctx, status); |
6989 | return; |
6990 | } |
6991 | |
6992 | |
6993 | expdiff = a->exp - b->exp; |
6994 | if (a->digits + expdiff > ctx->prec) { |
6995 | mpd_seterror(result, MPD_Invalid_operation, status); |
6996 | return; |
6997 | } |
6998 | |
6999 | if (expdiff >= 0) { |
7000 | shift = expdiff; |
7001 | if (!mpd_qshiftl(result, a, shift, status)) { |
7002 | return; |
7003 | } |
7004 | result->exp = b_exp; |
7005 | } |
7006 | else { |
7007 | /* At this point expdiff < 0 and a->digits+expdiff <= prec, |
7008 | * so the shift before an increment will fit in prec. */ |
7009 | shift = -expdiff; |
7010 | rnd = mpd_qshiftr(result, a, shift, status); |
7011 | if (rnd == MPD_UINT_MAX) { |
7012 | return; |
7013 | } |
7014 | result->exp = b_exp; |
7015 | if (!_mpd_apply_round_fit(result, rnd, ctx, status)) { |
7016 | return; |
7017 | } |
7018 | workstatus |= MPD_Rounded; |
7019 | if (rnd) { |
7020 | workstatus |= MPD_Inexact; |
7021 | } |
7022 | } |
7023 | |
7024 | if (mpd_adjexp(result) > ctx->emax || |
7025 | mpd_adjexp(result) < mpd_etiny(ctx)) { |
7026 | mpd_seterror(result, MPD_Invalid_operation, status); |
7027 | return; |
7028 | } |
7029 | |
7030 | *status |= workstatus; |
7031 | mpd_qfinalize(result, ctx, status); |
7032 | } |
7033 | |
7034 | void |
7035 | mpd_qreduce(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7036 | uint32_t *status) |
7037 | { |
7038 | mpd_ssize_t shift, maxexp, maxshift; |
7039 | uint8_t sign_a = mpd_sign(a); |
7040 | |
7041 | if (mpd_isspecial(a)) { |
7042 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
7043 | return; |
7044 | } |
7045 | mpd_qcopy(result, a, status); |
7046 | return; |
7047 | } |
7048 | |
7049 | if (!mpd_qcopy(result, a, status)) { |
7050 | return; |
7051 | } |
7052 | mpd_qfinalize(result, ctx, status); |
7053 | if (mpd_isspecial(result)) { |
7054 | return; |
7055 | } |
7056 | if (mpd_iszero(result)) { |
7057 | _settriple(result, sign_a, 0, 0); |
7058 | return; |
7059 | } |
7060 | |
7061 | shift = mpd_trail_zeros(result); |
7062 | maxexp = (ctx->clamp) ? mpd_etop(ctx) : ctx->emax; |
7063 | /* After the finalizing above result->exp <= maxexp. */ |
7064 | maxshift = maxexp - result->exp; |
7065 | shift = (shift > maxshift) ? maxshift : shift; |
7066 | |
7067 | mpd_qshiftr_inplace(result, shift); |
7068 | result->exp += shift; |
7069 | } |
7070 | |
7071 | void |
7072 | mpd_qrem(mpd_t *r, const mpd_t *a, const mpd_t *b, const mpd_context_t *ctx, |
7073 | uint32_t *status) |
7074 | { |
7075 | MPD_NEW_STATIC(q,0,0,0,0); |
7076 | |
7077 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
7078 | if (mpd_qcheck_nans(r, a, b, ctx, status)) { |
7079 | return; |
7080 | } |
7081 | if (mpd_isinfinite(a)) { |
7082 | mpd_seterror(r, MPD_Invalid_operation, status); |
7083 | return; |
7084 | } |
7085 | if (mpd_isinfinite(b)) { |
7086 | mpd_qcopy(r, a, status); |
7087 | mpd_qfinalize(r, ctx, status); |
7088 | return; |
7089 | } |
7090 | /* debug */ |
7091 | abort(); /* GCOV_NOT_REACHED */ |
7092 | } |
7093 | if (mpd_iszerocoeff(b)) { |
7094 | if (mpd_iszerocoeff(a)) { |
7095 | mpd_seterror(r, MPD_Division_undefined, status); |
7096 | } |
7097 | else { |
7098 | mpd_seterror(r, MPD_Invalid_operation, status); |
7099 | } |
7100 | return; |
7101 | } |
7102 | |
7103 | _mpd_qdivmod(&q, r, a, b, ctx, status); |
7104 | mpd_del(&q); |
7105 | mpd_qfinalize(r, ctx, status); |
7106 | } |
7107 | |
7108 | void |
7109 | mpd_qrem_near(mpd_t *r, const mpd_t *a, const mpd_t *b, |
7110 | const mpd_context_t *ctx, uint32_t *status) |
7111 | { |
7112 | mpd_context_t workctx; |
7113 | MPD_NEW_STATIC(btmp,0,0,0,0); |
7114 | MPD_NEW_STATIC(q,0,0,0,0); |
7115 | mpd_ssize_t expdiff, qdigits; |
7116 | int cmp, isodd, allnine; |
7117 | |
7118 | assert(r != NULL); /* annotation for scan-build */ |
7119 | |
7120 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
7121 | if (mpd_qcheck_nans(r, a, b, ctx, status)) { |
7122 | return; |
7123 | } |
7124 | if (mpd_isinfinite(a)) { |
7125 | mpd_seterror(r, MPD_Invalid_operation, status); |
7126 | return; |
7127 | } |
7128 | if (mpd_isinfinite(b)) { |
7129 | mpd_qcopy(r, a, status); |
7130 | mpd_qfinalize(r, ctx, status); |
7131 | return; |
7132 | } |
7133 | /* debug */ |
7134 | abort(); /* GCOV_NOT_REACHED */ |
7135 | } |
7136 | if (mpd_iszerocoeff(b)) { |
7137 | if (mpd_iszerocoeff(a)) { |
7138 | mpd_seterror(r, MPD_Division_undefined, status); |
7139 | } |
7140 | else { |
7141 | mpd_seterror(r, MPD_Invalid_operation, status); |
7142 | } |
7143 | return; |
7144 | } |
7145 | |
7146 | if (r == b) { |
7147 | if (!mpd_qcopy(&btmp, b, status)) { |
7148 | mpd_seterror(r, MPD_Malloc_error, status); |
7149 | return; |
7150 | } |
7151 | b = &btmp; |
7152 | } |
7153 | |
7154 | _mpd_qdivmod(&q, r, a, b, ctx, status); |
7155 | if (mpd_isnan(&q) || mpd_isnan(r)) { |
7156 | goto finish; |
7157 | } |
7158 | if (mpd_iszerocoeff(r)) { |
7159 | goto finish; |
7160 | } |
7161 | |
7162 | expdiff = mpd_adjexp(b) - mpd_adjexp(r); |
7163 | if (-1 <= expdiff && expdiff <= 1) { |
7164 | |
7165 | allnine = mpd_coeff_isallnine(&q); |
7166 | qdigits = q.digits; |
7167 | isodd = mpd_isodd(&q); |
7168 | |
7169 | mpd_maxcontext(&workctx); |
7170 | if (mpd_sign(a) == mpd_sign(b)) { |
7171 | /* sign(r) == sign(b) */ |
7172 | _mpd_qsub(&q, r, b, &workctx, &workctx.status); |
7173 | } |
7174 | else { |
7175 | /* sign(r) != sign(b) */ |
7176 | _mpd_qadd(&q, r, b, &workctx, &workctx.status); |
7177 | } |
7178 | |
7179 | if (workctx.status&MPD_Errors) { |
7180 | mpd_seterror(r, workctx.status&MPD_Errors, status); |
7181 | goto finish; |
7182 | } |
7183 | |
7184 | cmp = _mpd_cmp_abs(&q, r); |
7185 | if (cmp < 0 || (cmp == 0 && isodd)) { |
7186 | /* abs(r) > abs(b)/2 or abs(r) == abs(b)/2 and isodd(quotient) */ |
7187 | if (allnine && qdigits == ctx->prec) { |
7188 | /* abs(quotient) + 1 == 10**prec */ |
7189 | mpd_seterror(r, MPD_Division_impossible, status); |
7190 | goto finish; |
7191 | } |
7192 | mpd_qcopy(r, &q, status); |
7193 | } |
7194 | } |
7195 | |
7196 | |
7197 | finish: |
7198 | mpd_del(&btmp); |
7199 | mpd_del(&q); |
7200 | mpd_qfinalize(r, ctx, status); |
7201 | } |
7202 | |
7203 | static void |
7204 | _mpd_qrescale(mpd_t *result, const mpd_t *a, mpd_ssize_t exp, |
7205 | const mpd_context_t *ctx, uint32_t *status) |
7206 | { |
7207 | mpd_ssize_t expdiff, shift; |
7208 | mpd_uint_t rnd; |
7209 | |
7210 | if (mpd_isspecial(a)) { |
7211 | mpd_qcopy(result, a, status); |
7212 | return; |
7213 | } |
7214 | |
7215 | if (mpd_iszero(a)) { |
7216 | _settriple(result, mpd_sign(a), 0, exp); |
7217 | return; |
7218 | } |
7219 | |
7220 | expdiff = a->exp - exp; |
7221 | if (expdiff >= 0) { |
7222 | shift = expdiff; |
7223 | if (a->digits + shift > MPD_MAX_PREC+1) { |
7224 | mpd_seterror(result, MPD_Invalid_operation, status); |
7225 | return; |
7226 | } |
7227 | if (!mpd_qshiftl(result, a, shift, status)) { |
7228 | return; |
7229 | } |
7230 | result->exp = exp; |
7231 | } |
7232 | else { |
7233 | shift = -expdiff; |
7234 | rnd = mpd_qshiftr(result, a, shift, status); |
7235 | if (rnd == MPD_UINT_MAX) { |
7236 | return; |
7237 | } |
7238 | result->exp = exp; |
7239 | _mpd_apply_round_excess(result, rnd, ctx, status); |
7240 | *status |= MPD_Rounded; |
7241 | if (rnd) { |
7242 | *status |= MPD_Inexact; |
7243 | } |
7244 | } |
7245 | |
7246 | if (mpd_issubnormal(result, ctx)) { |
7247 | *status |= MPD_Subnormal; |
7248 | } |
7249 | } |
7250 | |
7251 | /* |
7252 | * Rescale a number so that it has exponent 'exp'. Does not regard context |
7253 | * precision, emax, emin, but uses the rounding mode. Special numbers are |
7254 | * quietly copied. Restrictions: |
7255 | * |
7256 | * MPD_MIN_ETINY <= exp <= MPD_MAX_EMAX+1 |
7257 | * result->digits <= MPD_MAX_PREC+1 |
7258 | */ |
7259 | void |
7260 | mpd_qrescale(mpd_t *result, const mpd_t *a, mpd_ssize_t exp, |
7261 | const mpd_context_t *ctx, uint32_t *status) |
7262 | { |
7263 | if (exp > MPD_MAX_EMAX+1 || exp < MPD_MIN_ETINY) { |
7264 | mpd_seterror(result, MPD_Invalid_operation, status); |
7265 | return; |
7266 | } |
7267 | |
7268 | _mpd_qrescale(result, a, exp, ctx, status); |
7269 | } |
7270 | |
7271 | /* |
7272 | * Same as mpd_qrescale, but with relaxed restrictions. The result of this |
7273 | * function should only be used for formatting a number and never as input |
7274 | * for other operations. |
7275 | * |
7276 | * MPD_MIN_ETINY-MPD_MAX_PREC <= exp <= MPD_MAX_EMAX+1 |
7277 | * result->digits <= MPD_MAX_PREC+1 |
7278 | */ |
7279 | void |
7280 | mpd_qrescale_fmt(mpd_t *result, const mpd_t *a, mpd_ssize_t exp, |
7281 | const mpd_context_t *ctx, uint32_t *status) |
7282 | { |
7283 | if (exp > MPD_MAX_EMAX+1 || exp < MPD_MIN_ETINY-MPD_MAX_PREC) { |
7284 | mpd_seterror(result, MPD_Invalid_operation, status); |
7285 | return; |
7286 | } |
7287 | |
7288 | _mpd_qrescale(result, a, exp, ctx, status); |
7289 | } |
7290 | |
7291 | /* Round to an integer according to 'action' and ctx->round. */ |
7292 | enum {TO_INT_EXACT, TO_INT_SILENT, TO_INT_TRUNC}; |
7293 | static void |
7294 | _mpd_qround_to_integral(int action, mpd_t *result, const mpd_t *a, |
7295 | const mpd_context_t *ctx, uint32_t *status) |
7296 | { |
7297 | mpd_uint_t rnd; |
7298 | |
7299 | if (mpd_isspecial(a)) { |
7300 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
7301 | return; |
7302 | } |
7303 | mpd_qcopy(result, a, status); |
7304 | return; |
7305 | } |
7306 | if (a->exp >= 0) { |
7307 | mpd_qcopy(result, a, status); |
7308 | return; |
7309 | } |
7310 | if (mpd_iszerocoeff(a)) { |
7311 | _settriple(result, mpd_sign(a), 0, 0); |
7312 | return; |
7313 | } |
7314 | |
7315 | rnd = mpd_qshiftr(result, a, -a->exp, status); |
7316 | if (rnd == MPD_UINT_MAX) { |
7317 | return; |
7318 | } |
7319 | result->exp = 0; |
7320 | |
7321 | if (action == TO_INT_EXACT || action == TO_INT_SILENT) { |
7322 | _mpd_apply_round_excess(result, rnd, ctx, status); |
7323 | if (action == TO_INT_EXACT) { |
7324 | *status |= MPD_Rounded; |
7325 | if (rnd) { |
7326 | *status |= MPD_Inexact; |
7327 | } |
7328 | } |
7329 | } |
7330 | } |
7331 | |
7332 | void |
7333 | mpd_qround_to_intx(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7334 | uint32_t *status) |
7335 | { |
7336 | (void)_mpd_qround_to_integral(TO_INT_EXACT, result, a, ctx, status); |
7337 | } |
7338 | |
7339 | void |
7340 | mpd_qround_to_int(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7341 | uint32_t *status) |
7342 | { |
7343 | (void)_mpd_qround_to_integral(TO_INT_SILENT, result, a, ctx, status); |
7344 | } |
7345 | |
7346 | void |
7347 | mpd_qtrunc(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7348 | uint32_t *status) |
7349 | { |
7350 | if (mpd_isspecial(a)) { |
7351 | mpd_seterror(result, MPD_Invalid_operation, status); |
7352 | return; |
7353 | } |
7354 | |
7355 | (void)_mpd_qround_to_integral(TO_INT_TRUNC, result, a, ctx, status); |
7356 | } |
7357 | |
7358 | void |
7359 | mpd_qfloor(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7360 | uint32_t *status) |
7361 | { |
7362 | mpd_context_t workctx = *ctx; |
7363 | |
7364 | if (mpd_isspecial(a)) { |
7365 | mpd_seterror(result, MPD_Invalid_operation, status); |
7366 | return; |
7367 | } |
7368 | |
7369 | workctx.round = MPD_ROUND_FLOOR; |
7370 | (void)_mpd_qround_to_integral(TO_INT_SILENT, result, a, |
7371 | &workctx, status); |
7372 | } |
7373 | |
7374 | void |
7375 | mpd_qceil(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7376 | uint32_t *status) |
7377 | { |
7378 | mpd_context_t workctx = *ctx; |
7379 | |
7380 | if (mpd_isspecial(a)) { |
7381 | mpd_seterror(result, MPD_Invalid_operation, status); |
7382 | return; |
7383 | } |
7384 | |
7385 | workctx.round = MPD_ROUND_CEILING; |
7386 | (void)_mpd_qround_to_integral(TO_INT_SILENT, result, a, |
7387 | &workctx, status); |
7388 | } |
7389 | |
7390 | int |
7391 | mpd_same_quantum(const mpd_t *a, const mpd_t *b) |
7392 | { |
7393 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
7394 | return ((mpd_isnan(a) && mpd_isnan(b)) || |
7395 | (mpd_isinfinite(a) && mpd_isinfinite(b))); |
7396 | } |
7397 | |
7398 | return a->exp == b->exp; |
7399 | } |
7400 | |
7401 | /* Schedule the increase in precision for the Newton iteration. */ |
7402 | static inline int |
7403 | recpr_schedule_prec(mpd_ssize_t klist[MPD_MAX_PREC_LOG2], |
7404 | mpd_ssize_t maxprec, mpd_ssize_t initprec) |
7405 | { |
7406 | mpd_ssize_t k; |
7407 | int i; |
7408 | |
7409 | assert(maxprec > 0 && initprec > 0); |
7410 | if (maxprec <= initprec) return -1; |
7411 | |
7412 | i = 0; k = maxprec; |
7413 | do { |
7414 | k = (k+1) / 2; |
7415 | klist[i++] = k; |
7416 | } while (k > initprec); |
7417 | |
7418 | return i-1; |
7419 | } |
7420 | |
7421 | /* |
7422 | * Initial approximation for the reciprocal: |
7423 | * k_0 := MPD_RDIGITS-2 |
7424 | * z_0 := 10**(-k_0) * floor(10**(2*k_0 + 2) / floor(v * 10**(k_0 + 2))) |
7425 | * Absolute error: |
7426 | * |1/v - z_0| < 10**(-k_0) |
7427 | * ACL2 proof: maxerror-inverse-approx |
7428 | */ |
7429 | static void |
7430 | _mpd_qreciprocal_approx(mpd_t *z, const mpd_t *v, uint32_t *status) |
7431 | { |
7432 | mpd_uint_t p10data[2] = {0, mpd_pow10[MPD_RDIGITS-2]}; |
7433 | mpd_uint_t dummy, word; |
7434 | int n; |
7435 | |
7436 | assert(v->exp == -v->digits); |
7437 | |
7438 | _mpd_get_msdigits(&dummy, &word, v, MPD_RDIGITS); |
7439 | n = mpd_word_digits(word); |
7440 | word *= mpd_pow10[MPD_RDIGITS-n]; |
7441 | |
7442 | mpd_qresize(z, 2, status); |
7443 | (void)_mpd_shortdiv(z->data, p10data, 2, word); |
7444 | |
7445 | mpd_clear_flags(z); |
7446 | z->exp = -(MPD_RDIGITS-2); |
7447 | z->len = (z->data[1] == 0) ? 1 : 2; |
7448 | mpd_setdigits(z); |
7449 | } |
7450 | |
7451 | /* |
7452 | * Reciprocal, calculated with Newton's Method. Assumption: result != a. |
7453 | * NOTE: The comments in the function show that certain operations are |
7454 | * exact. The proof for the maximum error is too long to fit in here. |
7455 | * ACL2 proof: maxerror-inverse-complete |
7456 | */ |
7457 | static void |
7458 | _mpd_qreciprocal(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7459 | uint32_t *status) |
7460 | { |
7461 | mpd_context_t varcontext, maxcontext; |
7462 | mpd_t *z = result; /* current approximation */ |
7463 | mpd_t *v; /* a, normalized to a number between 0.1 and 1 */ |
7464 | MPD_NEW_SHARED(vtmp, a); /* v shares data with a */ |
7465 | MPD_NEW_STATIC(s,0,0,0,0); /* temporary variable */ |
7466 | MPD_NEW_STATIC(t,0,0,0,0); /* temporary variable */ |
7467 | MPD_NEW_CONST(two,0,0,1,1,1,2); /* const 2 */ |
7468 | mpd_ssize_t klist[MPD_MAX_PREC_LOG2]; |
7469 | mpd_ssize_t adj, maxprec, initprec; |
7470 | uint8_t sign = mpd_sign(a); |
7471 | int i; |
7472 | |
7473 | assert(result != a); |
7474 | |
7475 | v = &vtmp; |
7476 | mpd_clear_flags(v); |
7477 | adj = v->digits + v->exp; |
7478 | v->exp = -v->digits; |
7479 | |
7480 | /* Initial approximation */ |
7481 | _mpd_qreciprocal_approx(z, v, status); |
7482 | |
7483 | mpd_maxcontext(&varcontext); |
7484 | mpd_maxcontext(&maxcontext); |
7485 | varcontext.round = maxcontext.round = MPD_ROUND_TRUNC; |
7486 | varcontext.emax = maxcontext.emax = MPD_MAX_EMAX + 100; |
7487 | varcontext.emin = maxcontext.emin = MPD_MIN_EMIN - 100; |
7488 | maxcontext.prec = MPD_MAX_PREC + 100; |
7489 | |
7490 | maxprec = ctx->prec; |
7491 | maxprec += 2; |
7492 | initprec = MPD_RDIGITS-3; |
7493 | |
7494 | i = recpr_schedule_prec(klist, maxprec, initprec); |
7495 | for (; i >= 0; i--) { |
7496 | /* Loop invariant: z->digits <= klist[i]+7 */ |
7497 | /* Let s := z**2, exact result */ |
7498 | _mpd_qmul_exact(&s, z, z, &maxcontext, status); |
7499 | varcontext.prec = 2*klist[i] + 5; |
7500 | if (v->digits > varcontext.prec) { |
7501 | /* Let t := v, truncated to n >= 2*k+5 fraction digits */ |
7502 | mpd_qshiftr(&t, v, v->digits-varcontext.prec, status); |
7503 | t.exp = -varcontext.prec; |
7504 | /* Let t := trunc(v)*s, truncated to n >= 2*k+1 fraction digits */ |
7505 | mpd_qmul(&t, &t, &s, &varcontext, status); |
7506 | } |
7507 | else { /* v->digits <= 2*k+5 */ |
7508 | /* Let t := v*s, truncated to n >= 2*k+1 fraction digits */ |
7509 | mpd_qmul(&t, v, &s, &varcontext, status); |
7510 | } |
7511 | /* Let s := 2*z, exact result */ |
7512 | _mpd_qmul_exact(&s, z, &two, &maxcontext, status); |
7513 | /* s.digits < t.digits <= 2*k+5, |adjexp(s)-adjexp(t)| <= 1, |
7514 | * so the subtraction generates at most 2*k+6 <= klist[i+1]+7 |
7515 | * digits. The loop invariant is preserved. */ |
7516 | _mpd_qsub_exact(z, &s, &t, &maxcontext, status); |
7517 | } |
7518 | |
7519 | if (!mpd_isspecial(z)) { |
7520 | z->exp -= adj; |
7521 | mpd_set_flags(z, sign); |
7522 | } |
7523 | |
7524 | mpd_del(&s); |
7525 | mpd_del(&t); |
7526 | mpd_qfinalize(z, ctx, status); |
7527 | } |
7528 | |
7529 | /* |
7530 | * Internal function for large numbers: |
7531 | * |
7532 | * q, r = divmod(coeff(a), coeff(b)) |
7533 | * |
7534 | * Strategy: Multiply the dividend by the reciprocal of the divisor. The |
7535 | * inexact result is fixed by a small loop, using at most one iteration. |
7536 | * |
7537 | * ACL2 proofs: |
7538 | * ------------ |
7539 | * 1) q is a natural number. (ndivmod-quotient-natp) |
7540 | * 2) r is a natural number. (ndivmod-remainder-natp) |
7541 | * 3) a = q * b + r (ndivmod-q*b+r==a) |
7542 | * 4) r < b (ndivmod-remainder-<-b) |
7543 | */ |
7544 | static void |
7545 | _mpd_base_ndivmod(mpd_t *q, mpd_t *r, const mpd_t *a, const mpd_t *b, |
7546 | uint32_t *status) |
7547 | { |
7548 | mpd_context_t workctx; |
7549 | mpd_t *qq = q, *rr = r; |
7550 | mpd_t aa, bb; |
7551 | int k; |
7552 | |
7553 | _mpd_copy_shared(&aa, a); |
7554 | _mpd_copy_shared(&bb, b); |
7555 | |
7556 | mpd_set_positive(&aa); |
7557 | mpd_set_positive(&bb); |
7558 | aa.exp = 0; |
7559 | bb.exp = 0; |
7560 | |
7561 | if (q == a || q == b) { |
7562 | if ((qq = mpd_qnew()) == NULL) { |
7563 | *status |= MPD_Malloc_error; |
7564 | goto nanresult; |
7565 | } |
7566 | } |
7567 | if (r == a || r == b) { |
7568 | if ((rr = mpd_qnew()) == NULL) { |
7569 | *status |= MPD_Malloc_error; |
7570 | goto nanresult; |
7571 | } |
7572 | } |
7573 | |
7574 | mpd_maxcontext(&workctx); |
7575 | |
7576 | /* Let prec := adigits - bdigits + 4 */ |
7577 | workctx.prec = a->digits - b->digits + 1 + 3; |
7578 | if (a->digits > MPD_MAX_PREC || workctx.prec > MPD_MAX_PREC) { |
7579 | *status |= MPD_Division_impossible; |
7580 | goto nanresult; |
7581 | } |
7582 | |
7583 | /* Let x := _mpd_qreciprocal(b, prec) |
7584 | * Then x is bounded by: |
7585 | * 1) 1/b - 10**(-prec - bdigits) < x < 1/b + 10**(-prec - bdigits) |
7586 | * 2) 1/b - 10**(-adigits - 4) < x < 1/b + 10**(-adigits - 4) |
7587 | */ |
7588 | _mpd_qreciprocal(rr, &bb, &workctx, &workctx.status); |
7589 | |
7590 | /* Get an estimate for the quotient. Let q := a * x |
7591 | * Then q is bounded by: |
7592 | * 3) a/b - 10**-4 < q < a/b + 10**-4 |
7593 | */ |
7594 | _mpd_qmul(qq, &aa, rr, &workctx, &workctx.status); |
7595 | /* Truncate q to an integer: |
7596 | * 4) a/b - 2 < trunc(q) < a/b + 1 |
7597 | */ |
7598 | mpd_qtrunc(qq, qq, &workctx, &workctx.status); |
7599 | |
7600 | workctx.prec = aa.digits + 3; |
7601 | workctx.emax = MPD_MAX_EMAX + 3; |
7602 | workctx.emin = MPD_MIN_EMIN - 3; |
7603 | /* Multiply the estimate for q by b: |
7604 | * 5) a - 2 * b < trunc(q) * b < a + b |
7605 | */ |
7606 | _mpd_qmul(rr, &bb, qq, &workctx, &workctx.status); |
7607 | /* Get the estimate for r such that a = q * b + r. */ |
7608 | _mpd_qsub_exact(rr, &aa, rr, &workctx, &workctx.status); |
7609 | |
7610 | /* Fix the result. At this point -b < r < 2*b, so the correction loop |
7611 | takes at most one iteration. */ |
7612 | for (k = 0;; k++) { |
7613 | if (mpd_isspecial(qq) || mpd_isspecial(rr)) { |
7614 | *status |= (workctx.status&MPD_Errors); |
7615 | goto nanresult; |
7616 | } |
7617 | if (k > 2) { /* Allow two iterations despite the proof. */ |
7618 | mpd_err_warn("libmpdec: internal error in " /* GCOV_NOT_REACHED */ |
7619 | "_mpd_base_ndivmod: please report" ); /* GCOV_NOT_REACHED */ |
7620 | *status |= MPD_Invalid_operation; /* GCOV_NOT_REACHED */ |
7621 | goto nanresult; /* GCOV_NOT_REACHED */ |
7622 | } |
7623 | /* r < 0 */ |
7624 | else if (_mpd_cmp(&zero, rr) == 1) { |
7625 | _mpd_qadd_exact(rr, rr, &bb, &workctx, &workctx.status); |
7626 | _mpd_qadd_exact(qq, qq, &minus_one, &workctx, &workctx.status); |
7627 | } |
7628 | /* 0 <= r < b */ |
7629 | else if (_mpd_cmp(rr, &bb) == -1) { |
7630 | break; |
7631 | } |
7632 | /* r >= b */ |
7633 | else { |
7634 | _mpd_qsub_exact(rr, rr, &bb, &workctx, &workctx.status); |
7635 | _mpd_qadd_exact(qq, qq, &one, &workctx, &workctx.status); |
7636 | } |
7637 | } |
7638 | |
7639 | if (qq != q) { |
7640 | if (!mpd_qcopy(q, qq, status)) { |
7641 | goto nanresult; /* GCOV_UNLIKELY */ |
7642 | } |
7643 | mpd_del(qq); |
7644 | } |
7645 | if (rr != r) { |
7646 | if (!mpd_qcopy(r, rr, status)) { |
7647 | goto nanresult; /* GCOV_UNLIKELY */ |
7648 | } |
7649 | mpd_del(rr); |
7650 | } |
7651 | |
7652 | *status |= (workctx.status&MPD_Errors); |
7653 | return; |
7654 | |
7655 | |
7656 | nanresult: |
7657 | if (qq && qq != q) mpd_del(qq); |
7658 | if (rr && rr != r) mpd_del(rr); |
7659 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
7660 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
7661 | } |
7662 | |
7663 | /* LIBMPDEC_ONLY */ |
7664 | /* |
7665 | * Schedule the optimal precision increase for the Newton iteration. |
7666 | * v := input operand |
7667 | * z_0 := initial approximation |
7668 | * initprec := natural number such that abs(sqrt(v) - z_0) < 10**-initprec |
7669 | * maxprec := target precision |
7670 | * |
7671 | * For convenience the output klist contains the elements in reverse order: |
7672 | * klist := [k_n-1, ..., k_0], where |
7673 | * 1) k_0 <= initprec and |
7674 | * 2) abs(sqrt(v) - result) < 10**(-2*k_n-1 + 2) <= 10**-maxprec. |
7675 | */ |
7676 | static inline int |
7677 | invroot_schedule_prec(mpd_ssize_t klist[MPD_MAX_PREC_LOG2], |
7678 | mpd_ssize_t maxprec, mpd_ssize_t initprec) |
7679 | { |
7680 | mpd_ssize_t k; |
7681 | int i; |
7682 | |
7683 | assert(maxprec >= 3 && initprec >= 3); |
7684 | if (maxprec <= initprec) return -1; |
7685 | |
7686 | i = 0; k = maxprec; |
7687 | do { |
7688 | k = (k+3) / 2; |
7689 | klist[i++] = k; |
7690 | } while (k > initprec); |
7691 | |
7692 | return i-1; |
7693 | } |
7694 | |
7695 | /* |
7696 | * Initial approximation for the inverse square root function. |
7697 | * Input: |
7698 | * v := rational number, with 1 <= v < 100 |
7699 | * vhat := floor(v * 10**6) |
7700 | * Output: |
7701 | * z := approximation to 1/sqrt(v), such that abs(z - 1/sqrt(v)) < 10**-3. |
7702 | */ |
7703 | static inline void |
7704 | _invroot_init_approx(mpd_t *z, mpd_uint_t vhat) |
7705 | { |
7706 | mpd_uint_t lo = 1000; |
7707 | mpd_uint_t hi = 10000; |
7708 | mpd_uint_t a, sq; |
7709 | |
7710 | assert(lo*lo <= vhat && vhat < (hi+1)*(hi+1)); |
7711 | |
7712 | for(;;) { |
7713 | a = (lo + hi) / 2; |
7714 | sq = a * a; |
7715 | if (vhat >= sq) { |
7716 | if (vhat < sq + 2*a + 1) { |
7717 | break; |
7718 | } |
7719 | lo = a + 1; |
7720 | } |
7721 | else { |
7722 | hi = a - 1; |
7723 | } |
7724 | } |
7725 | |
7726 | /* |
7727 | * After the binary search we have: |
7728 | * 1) a**2 <= floor(v * 10**6) < (a + 1)**2 |
7729 | * This implies: |
7730 | * 2) a**2 <= v * 10**6 < (a + 1)**2 |
7731 | * 3) a <= sqrt(v) * 10**3 < a + 1 |
7732 | * Since 10**3 <= a: |
7733 | * 4) 0 <= 10**prec/a - 1/sqrt(v) < 10**-prec |
7734 | * We have: |
7735 | * 5) 10**3/a - 10**-3 < floor(10**9/a) * 10**-6 <= 10**3/a |
7736 | * Merging 4) and 5): |
7737 | * 6) abs(floor(10**9/a) * 10**-6 - 1/sqrt(v)) < 10**-3 |
7738 | */ |
7739 | mpd_minalloc(z); |
7740 | mpd_clear_flags(z); |
7741 | z->data[0] = 1000000000UL / a; |
7742 | z->len = 1; |
7743 | z->exp = -6; |
7744 | mpd_setdigits(z); |
7745 | } |
7746 | |
7747 | /* |
7748 | * Set 'result' to 1/sqrt(a). |
7749 | * Relative error: abs(result - 1/sqrt(a)) < 10**-prec * 1/sqrt(a) |
7750 | */ |
7751 | static void |
7752 | _mpd_qinvroot(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7753 | uint32_t *status) |
7754 | { |
7755 | uint32_t workstatus = 0; |
7756 | mpd_context_t varcontext, maxcontext; |
7757 | mpd_t *z = result; /* current approximation */ |
7758 | mpd_t *v; /* a, normalized to a number between 1 and 100 */ |
7759 | MPD_NEW_SHARED(vtmp, a); /* by default v will share data with a */ |
7760 | MPD_NEW_STATIC(s,0,0,0,0); /* temporary variable */ |
7761 | MPD_NEW_STATIC(t,0,0,0,0); /* temporary variable */ |
7762 | MPD_NEW_CONST(one_half,0,-1,1,1,1,5); |
7763 | MPD_NEW_CONST(three,0,0,1,1,1,3); |
7764 | mpd_ssize_t klist[MPD_MAX_PREC_LOG2]; |
7765 | mpd_ssize_t ideal_exp, shift; |
7766 | mpd_ssize_t adj, tz; |
7767 | mpd_ssize_t maxprec, fracdigits; |
7768 | mpd_uint_t vhat, dummy; |
7769 | int i, n; |
7770 | |
7771 | |
7772 | ideal_exp = -(a->exp - (a->exp & 1)) / 2; |
7773 | |
7774 | v = &vtmp; |
7775 | if (result == a) { |
7776 | if ((v = mpd_qncopy(a)) == NULL) { |
7777 | mpd_seterror(result, MPD_Malloc_error, status); |
7778 | return; |
7779 | } |
7780 | } |
7781 | |
7782 | /* normalize a to 1 <= v < 100 */ |
7783 | if ((v->digits+v->exp) & 1) { |
7784 | fracdigits = v->digits - 1; |
7785 | v->exp = -fracdigits; |
7786 | n = (v->digits > 7) ? 7 : (int)v->digits; |
7787 | /* Let vhat := floor(v * 10**(2*initprec)) */ |
7788 | _mpd_get_msdigits(&dummy, &vhat, v, n); |
7789 | if (n < 7) { |
7790 | vhat *= mpd_pow10[7-n]; |
7791 | } |
7792 | } |
7793 | else { |
7794 | fracdigits = v->digits - 2; |
7795 | v->exp = -fracdigits; |
7796 | n = (v->digits > 8) ? 8 : (int)v->digits; |
7797 | /* Let vhat := floor(v * 10**(2*initprec)) */ |
7798 | _mpd_get_msdigits(&dummy, &vhat, v, n); |
7799 | if (n < 8) { |
7800 | vhat *= mpd_pow10[8-n]; |
7801 | } |
7802 | } |
7803 | adj = (a->exp-v->exp) / 2; |
7804 | |
7805 | /* initial approximation */ |
7806 | _invroot_init_approx(z, vhat); |
7807 | |
7808 | mpd_maxcontext(&maxcontext); |
7809 | mpd_maxcontext(&varcontext); |
7810 | varcontext.round = MPD_ROUND_TRUNC; |
7811 | maxprec = ctx->prec + 1; |
7812 | |
7813 | /* initprec == 3 */ |
7814 | i = invroot_schedule_prec(klist, maxprec, 3); |
7815 | for (; i >= 0; i--) { |
7816 | varcontext.prec = 2*klist[i]+2; |
7817 | mpd_qmul(&s, z, z, &maxcontext, &workstatus); |
7818 | if (v->digits > varcontext.prec) { |
7819 | shift = v->digits - varcontext.prec; |
7820 | mpd_qshiftr(&t, v, shift, &workstatus); |
7821 | t.exp += shift; |
7822 | mpd_qmul(&t, &t, &s, &varcontext, &workstatus); |
7823 | } |
7824 | else { |
7825 | mpd_qmul(&t, v, &s, &varcontext, &workstatus); |
7826 | } |
7827 | mpd_qsub(&t, &three, &t, &maxcontext, &workstatus); |
7828 | mpd_qmul(z, z, &t, &varcontext, &workstatus); |
7829 | mpd_qmul(z, z, &one_half, &maxcontext, &workstatus); |
7830 | } |
7831 | |
7832 | z->exp -= adj; |
7833 | |
7834 | tz = mpd_trail_zeros(result); |
7835 | shift = ideal_exp - result->exp; |
7836 | shift = (tz > shift) ? shift : tz; |
7837 | if (shift > 0) { |
7838 | mpd_qshiftr_inplace(result, shift); |
7839 | result->exp += shift; |
7840 | } |
7841 | |
7842 | |
7843 | mpd_del(&s); |
7844 | mpd_del(&t); |
7845 | if (v != &vtmp) mpd_del(v); |
7846 | *status |= (workstatus&MPD_Errors); |
7847 | *status |= (MPD_Rounded|MPD_Inexact); |
7848 | } |
7849 | |
7850 | void |
7851 | mpd_qinvroot(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7852 | uint32_t *status) |
7853 | { |
7854 | mpd_context_t workctx; |
7855 | |
7856 | if (mpd_isspecial(a)) { |
7857 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
7858 | return; |
7859 | } |
7860 | if (mpd_isnegative(a)) { |
7861 | mpd_seterror(result, MPD_Invalid_operation, status); |
7862 | return; |
7863 | } |
7864 | /* positive infinity */ |
7865 | _settriple(result, MPD_POS, 0, mpd_etiny(ctx)); |
7866 | *status |= MPD_Clamped; |
7867 | return; |
7868 | } |
7869 | if (mpd_iszero(a)) { |
7870 | mpd_setspecial(result, mpd_sign(a), MPD_INF); |
7871 | *status |= MPD_Division_by_zero; |
7872 | return; |
7873 | } |
7874 | if (mpd_isnegative(a)) { |
7875 | mpd_seterror(result, MPD_Invalid_operation, status); |
7876 | return; |
7877 | } |
7878 | |
7879 | workctx = *ctx; |
7880 | workctx.prec += 2; |
7881 | workctx.round = MPD_ROUND_HALF_EVEN; |
7882 | _mpd_qinvroot(result, a, &workctx, status); |
7883 | mpd_qfinalize(result, ctx, status); |
7884 | } |
7885 | /* END LIBMPDEC_ONLY */ |
7886 | |
7887 | /* Algorithm from decimal.py */ |
7888 | static void |
7889 | _mpd_qsqrt(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7890 | uint32_t *status) |
7891 | { |
7892 | mpd_context_t maxcontext; |
7893 | MPD_NEW_STATIC(c,0,0,0,0); |
7894 | MPD_NEW_STATIC(q,0,0,0,0); |
7895 | MPD_NEW_STATIC(r,0,0,0,0); |
7896 | MPD_NEW_CONST(two,0,0,1,1,1,2); |
7897 | mpd_ssize_t prec, ideal_exp; |
7898 | mpd_ssize_t l, shift; |
7899 | int exact = 0; |
7900 | |
7901 | |
7902 | ideal_exp = (a->exp - (a->exp & 1)) / 2; |
7903 | |
7904 | if (mpd_isspecial(a)) { |
7905 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
7906 | return; |
7907 | } |
7908 | if (mpd_isnegative(a)) { |
7909 | mpd_seterror(result, MPD_Invalid_operation, status); |
7910 | return; |
7911 | } |
7912 | mpd_setspecial(result, MPD_POS, MPD_INF); |
7913 | return; |
7914 | } |
7915 | if (mpd_iszero(a)) { |
7916 | _settriple(result, mpd_sign(a), 0, ideal_exp); |
7917 | mpd_qfinalize(result, ctx, status); |
7918 | return; |
7919 | } |
7920 | if (mpd_isnegative(a)) { |
7921 | mpd_seterror(result, MPD_Invalid_operation, status); |
7922 | return; |
7923 | } |
7924 | |
7925 | mpd_maxcontext(&maxcontext); |
7926 | prec = ctx->prec + 1; |
7927 | |
7928 | if (!mpd_qcopy(&c, a, status)) { |
7929 | goto malloc_error; |
7930 | } |
7931 | c.exp = 0; |
7932 | |
7933 | if (a->exp & 1) { |
7934 | if (!mpd_qshiftl(&c, &c, 1, status)) { |
7935 | goto malloc_error; |
7936 | } |
7937 | l = (a->digits >> 1) + 1; |
7938 | } |
7939 | else { |
7940 | l = (a->digits + 1) >> 1; |
7941 | } |
7942 | |
7943 | shift = prec - l; |
7944 | if (shift >= 0) { |
7945 | if (!mpd_qshiftl(&c, &c, 2*shift, status)) { |
7946 | goto malloc_error; |
7947 | } |
7948 | exact = 1; |
7949 | } |
7950 | else { |
7951 | exact = !mpd_qshiftr_inplace(&c, -2*shift); |
7952 | } |
7953 | |
7954 | ideal_exp -= shift; |
7955 | |
7956 | /* find result = floor(sqrt(c)) using Newton's method */ |
7957 | if (!mpd_qshiftl(result, &one, prec, status)) { |
7958 | goto malloc_error; |
7959 | } |
7960 | |
7961 | while (1) { |
7962 | _mpd_qdivmod(&q, &r, &c, result, &maxcontext, &maxcontext.status); |
7963 | if (mpd_isspecial(result) || mpd_isspecial(&q)) { |
7964 | mpd_seterror(result, maxcontext.status&MPD_Errors, status); |
7965 | goto out; |
7966 | } |
7967 | if (_mpd_cmp(result, &q) <= 0) { |
7968 | break; |
7969 | } |
7970 | _mpd_qadd_exact(result, result, &q, &maxcontext, &maxcontext.status); |
7971 | if (mpd_isspecial(result)) { |
7972 | mpd_seterror(result, maxcontext.status&MPD_Errors, status); |
7973 | goto out; |
7974 | } |
7975 | _mpd_qdivmod(result, &r, result, &two, &maxcontext, &maxcontext.status); |
7976 | } |
7977 | |
7978 | if (exact) { |
7979 | _mpd_qmul_exact(&r, result, result, &maxcontext, &maxcontext.status); |
7980 | if (mpd_isspecial(&r)) { |
7981 | mpd_seterror(result, maxcontext.status&MPD_Errors, status); |
7982 | goto out; |
7983 | } |
7984 | exact = (_mpd_cmp(&r, &c) == 0); |
7985 | } |
7986 | |
7987 | if (exact) { |
7988 | if (shift >= 0) { |
7989 | mpd_qshiftr_inplace(result, shift); |
7990 | } |
7991 | else { |
7992 | if (!mpd_qshiftl(result, result, -shift, status)) { |
7993 | goto malloc_error; |
7994 | } |
7995 | } |
7996 | ideal_exp += shift; |
7997 | } |
7998 | else { |
7999 | int lsd = (int)mpd_lsd(result->data[0]); |
8000 | if (lsd == 0 || lsd == 5) { |
8001 | result->data[0] += 1; |
8002 | } |
8003 | } |
8004 | |
8005 | result->exp = ideal_exp; |
8006 | |
8007 | |
8008 | out: |
8009 | mpd_del(&c); |
8010 | mpd_del(&q); |
8011 | mpd_del(&r); |
8012 | maxcontext = *ctx; |
8013 | maxcontext.round = MPD_ROUND_HALF_EVEN; |
8014 | mpd_qfinalize(result, &maxcontext, status); |
8015 | return; |
8016 | |
8017 | malloc_error: |
8018 | mpd_seterror(result, MPD_Malloc_error, status); |
8019 | goto out; |
8020 | } |
8021 | |
8022 | void |
8023 | mpd_qsqrt(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
8024 | uint32_t *status) |
8025 | { |
8026 | MPD_NEW_STATIC(aa,0,0,0,0); |
8027 | uint32_t xstatus = 0; |
8028 | |
8029 | if (result == a) { |
8030 | if (!mpd_qcopy(&aa, a, status)) { |
8031 | mpd_seterror(result, MPD_Malloc_error, status); |
8032 | goto out; |
8033 | } |
8034 | a = &aa; |
8035 | } |
8036 | |
8037 | _mpd_qsqrt(result, a, ctx, &xstatus); |
8038 | |
8039 | if (xstatus & (MPD_Malloc_error|MPD_Division_impossible)) { |
8040 | /* The above conditions can occur at very high context precisions |
8041 | * if intermediate values get too large. Retry the operation with |
8042 | * a lower context precision in case the result is exact. |
8043 | * |
8044 | * If the result is exact, an upper bound for the number of digits |
8045 | * is the number of digits in the input. |
8046 | * |
8047 | * NOTE: sqrt(40e9) = 2.0e+5 /\ digits(40e9) = digits(2.0e+5) = 2 |
8048 | */ |
8049 | uint32_t ystatus = 0; |
8050 | mpd_context_t workctx = *ctx; |
8051 | |
8052 | workctx.prec = a->digits; |
8053 | if (workctx.prec >= ctx->prec) { |
8054 | *status |= (xstatus|MPD_Errors); |
8055 | goto out; /* No point in repeating this, keep the original error. */ |
8056 | } |
8057 | |
8058 | _mpd_qsqrt(result, a, &workctx, &ystatus); |
8059 | if (ystatus != 0) { |
8060 | ystatus = *status | ((xstatus|ystatus)&MPD_Errors); |
8061 | mpd_seterror(result, ystatus, status); |
8062 | } |
8063 | } |
8064 | else { |
8065 | *status |= xstatus; |
8066 | } |
8067 | |
8068 | out: |
8069 | mpd_del(&aa); |
8070 | } |
8071 | |
8072 | |
8073 | /******************************************************************************/ |
8074 | /* Base conversions */ |
8075 | /******************************************************************************/ |
8076 | |
8077 | /* Space needed to represent an integer mpd_t in base 'base'. */ |
8078 | size_t |
8079 | mpd_sizeinbase(const mpd_t *a, uint32_t base) |
8080 | { |
8081 | double x; |
8082 | size_t digits; |
8083 | double upper_bound; |
8084 | |
8085 | assert(mpd_isinteger(a)); |
8086 | assert(base >= 2); |
8087 | |
8088 | if (mpd_iszero(a)) { |
8089 | return 1; |
8090 | } |
8091 | |
8092 | digits = a->digits+a->exp; |
8093 | |
8094 | #ifdef CONFIG_64 |
8095 | /* ceil(2711437152599294 / log10(2)) + 4 == 2**53 */ |
8096 | if (digits > 2711437152599294ULL) { |
8097 | return SIZE_MAX; |
8098 | } |
8099 | |
8100 | upper_bound = (double)((1ULL<<53)-1); |
8101 | #else |
8102 | upper_bound = (double)(SIZE_MAX-1); |
8103 | #endif |
8104 | |
8105 | x = (double)digits / log10(base); |
8106 | return (x > upper_bound) ? SIZE_MAX : (size_t)x + 1; |
8107 | } |
8108 | |
8109 | /* Space needed to import a base 'base' integer of length 'srclen'. */ |
8110 | static mpd_ssize_t |
8111 | _mpd_importsize(size_t srclen, uint32_t base) |
8112 | { |
8113 | double x; |
8114 | double upper_bound; |
8115 | |
8116 | assert(srclen > 0); |
8117 | assert(base >= 2); |
8118 | |
8119 | #if SIZE_MAX == UINT64_MAX |
8120 | if (srclen > (1ULL<<53)) { |
8121 | return MPD_SSIZE_MAX; |
8122 | } |
8123 | |
8124 | assert((1ULL<<53) <= MPD_MAXIMPORT); |
8125 | upper_bound = (double)((1ULL<<53)-1); |
8126 | #else |
8127 | upper_bound = MPD_MAXIMPORT-1; |
8128 | #endif |
8129 | |
8130 | x = (double)srclen * (log10(base)/MPD_RDIGITS); |
8131 | return (x > upper_bound) ? MPD_SSIZE_MAX : (mpd_ssize_t)x + 1; |
8132 | } |
8133 | |
8134 | static uint8_t |
8135 | mpd_resize_u16(uint16_t **w, size_t nmemb) |
8136 | { |
8137 | uint8_t err = 0; |
8138 | *w = mpd_realloc(*w, nmemb, sizeof **w, &err); |
8139 | return !err; |
8140 | } |
8141 | |
8142 | static uint8_t |
8143 | mpd_resize_u32(uint32_t **w, size_t nmemb) |
8144 | { |
8145 | uint8_t err = 0; |
8146 | *w = mpd_realloc(*w, nmemb, sizeof **w, &err); |
8147 | return !err; |
8148 | } |
8149 | |
8150 | static size_t |
8151 | _baseconv_to_u16(uint16_t **w, size_t wlen, mpd_uint_t wbase, |
8152 | mpd_uint_t *u, mpd_ssize_t ulen) |
8153 | { |
8154 | size_t n = 0; |
8155 | |
8156 | assert(wlen > 0 && ulen > 0); |
8157 | assert(wbase <= (1U<<16)); |
8158 | |
8159 | do { |
8160 | if (n >= wlen) { |
8161 | if (!mpd_resize_u16(w, n+1)) { |
8162 | return SIZE_MAX; |
8163 | } |
8164 | wlen = n+1; |
8165 | } |
8166 | (*w)[n++] = (uint16_t)_mpd_shortdiv(u, u, ulen, wbase); |
8167 | /* ulen is at least 1. u[ulen-1] can only be zero if ulen == 1. */ |
8168 | ulen = _mpd_real_size(u, ulen); |
8169 | |
8170 | } while (u[ulen-1] != 0); |
8171 | |
8172 | return n; |
8173 | } |
8174 | |
8175 | static size_t |
8176 | _coeff_from_u16(mpd_t *w, mpd_ssize_t wlen, |
8177 | const mpd_uint_t *u, size_t ulen, uint32_t ubase, |
8178 | uint32_t *status) |
8179 | { |
8180 | mpd_ssize_t n = 0; |
8181 | mpd_uint_t carry; |
8182 | |
8183 | assert(wlen > 0 && ulen > 0); |
8184 | assert(ubase <= (1U<<16)); |
8185 | |
8186 | w->data[n++] = u[--ulen]; |
8187 | while (--ulen != SIZE_MAX) { |
8188 | carry = _mpd_shortmul_c(w->data, w->data, n, ubase); |
8189 | if (carry) { |
8190 | if (n >= wlen) { |
8191 | if (!mpd_qresize(w, n+1, status)) { |
8192 | return SIZE_MAX; |
8193 | } |
8194 | wlen = n+1; |
8195 | } |
8196 | w->data[n++] = carry; |
8197 | } |
8198 | carry = _mpd_shortadd(w->data, n, u[ulen]); |
8199 | if (carry) { |
8200 | if (n >= wlen) { |
8201 | if (!mpd_qresize(w, n+1, status)) { |
8202 | return SIZE_MAX; |
8203 | } |
8204 | wlen = n+1; |
8205 | } |
8206 | w->data[n++] = carry; |
8207 | } |
8208 | } |
8209 | |
8210 | return n; |
8211 | } |
8212 | |
8213 | /* target base wbase < source base ubase */ |
8214 | static size_t |
8215 | _baseconv_to_smaller(uint32_t **w, size_t wlen, uint32_t wbase, |
8216 | mpd_uint_t *u, mpd_ssize_t ulen, mpd_uint_t ubase) |
8217 | { |
8218 | size_t n = 0; |
8219 | |
8220 | assert(wlen > 0 && ulen > 0); |
8221 | assert(wbase < ubase); |
8222 | |
8223 | do { |
8224 | if (n >= wlen) { |
8225 | if (!mpd_resize_u32(w, n+1)) { |
8226 | return SIZE_MAX; |
8227 | } |
8228 | wlen = n+1; |
8229 | } |
8230 | (*w)[n++] = (uint32_t)_mpd_shortdiv_b(u, u, ulen, wbase, ubase); |
8231 | /* ulen is at least 1. u[ulen-1] can only be zero if ulen == 1. */ |
8232 | ulen = _mpd_real_size(u, ulen); |
8233 | |
8234 | } while (u[ulen-1] != 0); |
8235 | |
8236 | return n; |
8237 | } |
8238 | |
8239 | #ifdef CONFIG_32 |
8240 | /* target base 'wbase' == source base 'ubase' */ |
8241 | static size_t |
8242 | _copy_equal_base(uint32_t **w, size_t wlen, |
8243 | const uint32_t *u, size_t ulen) |
8244 | { |
8245 | if (wlen < ulen) { |
8246 | if (!mpd_resize_u32(w, ulen)) { |
8247 | return SIZE_MAX; |
8248 | } |
8249 | } |
8250 | |
8251 | memcpy(*w, u, ulen * (sizeof **w)); |
8252 | return ulen; |
8253 | } |
8254 | |
8255 | /* target base 'wbase' > source base 'ubase' */ |
8256 | static size_t |
8257 | _baseconv_to_larger(uint32_t **w, size_t wlen, mpd_uint_t wbase, |
8258 | const mpd_uint_t *u, size_t ulen, mpd_uint_t ubase) |
8259 | { |
8260 | size_t n = 0; |
8261 | mpd_uint_t carry; |
8262 | |
8263 | assert(wlen > 0 && ulen > 0); |
8264 | assert(ubase < wbase); |
8265 | |
8266 | (*w)[n++] = u[--ulen]; |
8267 | while (--ulen != SIZE_MAX) { |
8268 | carry = _mpd_shortmul_b(*w, *w, n, ubase, wbase); |
8269 | if (carry) { |
8270 | if (n >= wlen) { |
8271 | if (!mpd_resize_u32(w, n+1)) { |
8272 | return SIZE_MAX; |
8273 | } |
8274 | wlen = n+1; |
8275 | } |
8276 | (*w)[n++] = carry; |
8277 | } |
8278 | carry = _mpd_shortadd_b(*w, n, u[ulen], wbase); |
8279 | if (carry) { |
8280 | if (n >= wlen) { |
8281 | if (!mpd_resize_u32(w, n+1)) { |
8282 | return SIZE_MAX; |
8283 | } |
8284 | wlen = n+1; |
8285 | } |
8286 | (*w)[n++] = carry; |
8287 | } |
8288 | } |
8289 | |
8290 | return n; |
8291 | } |
8292 | |
8293 | /* target base wbase < source base ubase */ |
8294 | static size_t |
8295 | _coeff_from_larger_base(mpd_t *w, size_t wlen, mpd_uint_t wbase, |
8296 | mpd_uint_t *u, mpd_ssize_t ulen, mpd_uint_t ubase, |
8297 | uint32_t *status) |
8298 | { |
8299 | size_t n = 0; |
8300 | |
8301 | assert(wlen > 0 && ulen > 0); |
8302 | assert(wbase < ubase); |
8303 | |
8304 | do { |
8305 | if (n >= wlen) { |
8306 | if (!mpd_qresize(w, n+1, status)) { |
8307 | return SIZE_MAX; |
8308 | } |
8309 | wlen = n+1; |
8310 | } |
8311 | w->data[n++] = (uint32_t)_mpd_shortdiv_b(u, u, ulen, wbase, ubase); |
8312 | /* ulen is at least 1. u[ulen-1] can only be zero if ulen == 1. */ |
8313 | ulen = _mpd_real_size(u, ulen); |
8314 | |
8315 | } while (u[ulen-1] != 0); |
8316 | |
8317 | return n; |
8318 | } |
8319 | #endif |
8320 | |
8321 | /* target base 'wbase' > source base 'ubase' */ |
8322 | static size_t |
8323 | _coeff_from_smaller_base(mpd_t *w, mpd_ssize_t wlen, mpd_uint_t wbase, |
8324 | const uint32_t *u, size_t ulen, mpd_uint_t ubase, |
8325 | uint32_t *status) |
8326 | { |
8327 | mpd_ssize_t n = 0; |
8328 | mpd_uint_t carry; |
8329 | |
8330 | assert(wlen > 0 && ulen > 0); |
8331 | assert(wbase > ubase); |
8332 | |
8333 | w->data[n++] = u[--ulen]; |
8334 | while (--ulen != SIZE_MAX) { |
8335 | carry = _mpd_shortmul_b(w->data, w->data, n, ubase, wbase); |
8336 | if (carry) { |
8337 | if (n >= wlen) { |
8338 | if (!mpd_qresize(w, n+1, status)) { |
8339 | return SIZE_MAX; |
8340 | } |
8341 | wlen = n+1; |
8342 | } |
8343 | w->data[n++] = carry; |
8344 | } |
8345 | carry = _mpd_shortadd_b(w->data, n, u[ulen], wbase); |
8346 | if (carry) { |
8347 | if (n >= wlen) { |
8348 | if (!mpd_qresize(w, n+1, status)) { |
8349 | return SIZE_MAX; |
8350 | } |
8351 | wlen = n+1; |
8352 | } |
8353 | w->data[n++] = carry; |
8354 | } |
8355 | } |
8356 | |
8357 | return n; |
8358 | } |
8359 | |
8360 | /* |
8361 | * Convert an integer mpd_t to a multiprecision integer with base <= 2**16. |
8362 | * The least significant word of the result is (*rdata)[0]. |
8363 | * |
8364 | * If rdata is NULL, space is allocated by the function and rlen is irrelevant. |
8365 | * In case of an error any allocated storage is freed and rdata is set back to |
8366 | * NULL. |
8367 | * |
8368 | * If rdata is non-NULL, it MUST be allocated by one of libmpdec's allocation |
8369 | * functions and rlen MUST be correct. If necessary, the function will resize |
8370 | * rdata. In case of an error the caller must free rdata. |
8371 | * |
8372 | * Return value: In case of success, the exact length of rdata, SIZE_MAX |
8373 | * otherwise. |
8374 | */ |
8375 | size_t |
8376 | mpd_qexport_u16(uint16_t **rdata, size_t rlen, uint32_t rbase, |
8377 | const mpd_t *src, uint32_t *status) |
8378 | { |
8379 | MPD_NEW_STATIC(tsrc,0,0,0,0); |
8380 | int alloc = 0; /* rdata == NULL */ |
8381 | size_t n; |
8382 | |
8383 | assert(rbase <= (1U<<16)); |
8384 | |
8385 | if (mpd_isspecial(src) || !_mpd_isint(src)) { |
8386 | *status |= MPD_Invalid_operation; |
8387 | return SIZE_MAX; |
8388 | } |
8389 | |
8390 | if (*rdata == NULL) { |
8391 | rlen = mpd_sizeinbase(src, rbase); |
8392 | if (rlen == SIZE_MAX) { |
8393 | *status |= MPD_Invalid_operation; |
8394 | return SIZE_MAX; |
8395 | } |
8396 | *rdata = mpd_alloc(rlen, sizeof **rdata); |
8397 | if (*rdata == NULL) { |
8398 | goto malloc_error; |
8399 | } |
8400 | alloc = 1; |
8401 | } |
8402 | |
8403 | if (mpd_iszero(src)) { |
8404 | **rdata = 0; |
8405 | return 1; |
8406 | } |
8407 | |
8408 | if (src->exp >= 0) { |
8409 | if (!mpd_qshiftl(&tsrc, src, src->exp, status)) { |
8410 | goto malloc_error; |
8411 | } |
8412 | } |
8413 | else { |
8414 | if (mpd_qshiftr(&tsrc, src, -src->exp, status) == MPD_UINT_MAX) { |
8415 | goto malloc_error; |
8416 | } |
8417 | } |
8418 | |
8419 | n = _baseconv_to_u16(rdata, rlen, rbase, tsrc.data, tsrc.len); |
8420 | if (n == SIZE_MAX) { |
8421 | goto malloc_error; |
8422 | } |
8423 | |
8424 | |
8425 | out: |
8426 | mpd_del(&tsrc); |
8427 | return n; |
8428 | |
8429 | malloc_error: |
8430 | if (alloc) { |
8431 | mpd_free(*rdata); |
8432 | *rdata = NULL; |
8433 | } |
8434 | n = SIZE_MAX; |
8435 | *status |= MPD_Malloc_error; |
8436 | goto out; |
8437 | } |
8438 | |
8439 | /* |
8440 | * Convert an integer mpd_t to a multiprecision integer with base<=UINT32_MAX. |
8441 | * The least significant word of the result is (*rdata)[0]. |
8442 | * |
8443 | * If rdata is NULL, space is allocated by the function and rlen is irrelevant. |
8444 | * In case of an error any allocated storage is freed and rdata is set back to |
8445 | * NULL. |
8446 | * |
8447 | * If rdata is non-NULL, it MUST be allocated by one of libmpdec's allocation |
8448 | * functions and rlen MUST be correct. If necessary, the function will resize |
8449 | * rdata. In case of an error the caller must free rdata. |
8450 | * |
8451 | * Return value: In case of success, the exact length of rdata, SIZE_MAX |
8452 | * otherwise. |
8453 | */ |
8454 | size_t |
8455 | mpd_qexport_u32(uint32_t **rdata, size_t rlen, uint32_t rbase, |
8456 | const mpd_t *src, uint32_t *status) |
8457 | { |
8458 | MPD_NEW_STATIC(tsrc,0,0,0,0); |
8459 | int alloc = 0; /* rdata == NULL */ |
8460 | size_t n; |
8461 | |
8462 | if (mpd_isspecial(src) || !_mpd_isint(src)) { |
8463 | *status |= MPD_Invalid_operation; |
8464 | return SIZE_MAX; |
8465 | } |
8466 | |
8467 | if (*rdata == NULL) { |
8468 | rlen = mpd_sizeinbase(src, rbase); |
8469 | if (rlen == SIZE_MAX) { |
8470 | *status |= MPD_Invalid_operation; |
8471 | return SIZE_MAX; |
8472 | } |
8473 | *rdata = mpd_alloc(rlen, sizeof **rdata); |
8474 | if (*rdata == NULL) { |
8475 | goto malloc_error; |
8476 | } |
8477 | alloc = 1; |
8478 | } |
8479 | |
8480 | if (mpd_iszero(src)) { |
8481 | **rdata = 0; |
8482 | return 1; |
8483 | } |
8484 | |
8485 | if (src->exp >= 0) { |
8486 | if (!mpd_qshiftl(&tsrc, src, src->exp, status)) { |
8487 | goto malloc_error; |
8488 | } |
8489 | } |
8490 | else { |
8491 | if (mpd_qshiftr(&tsrc, src, -src->exp, status) == MPD_UINT_MAX) { |
8492 | goto malloc_error; |
8493 | } |
8494 | } |
8495 | |
8496 | #ifdef CONFIG_64 |
8497 | n = _baseconv_to_smaller(rdata, rlen, rbase, |
8498 | tsrc.data, tsrc.len, MPD_RADIX); |
8499 | #else |
8500 | if (rbase == MPD_RADIX) { |
8501 | n = _copy_equal_base(rdata, rlen, tsrc.data, tsrc.len); |
8502 | } |
8503 | else if (rbase < MPD_RADIX) { |
8504 | n = _baseconv_to_smaller(rdata, rlen, rbase, |
8505 | tsrc.data, tsrc.len, MPD_RADIX); |
8506 | } |
8507 | else { |
8508 | n = _baseconv_to_larger(rdata, rlen, rbase, |
8509 | tsrc.data, tsrc.len, MPD_RADIX); |
8510 | } |
8511 | #endif |
8512 | |
8513 | if (n == SIZE_MAX) { |
8514 | goto malloc_error; |
8515 | } |
8516 | |
8517 | |
8518 | out: |
8519 | mpd_del(&tsrc); |
8520 | return n; |
8521 | |
8522 | malloc_error: |
8523 | if (alloc) { |
8524 | mpd_free(*rdata); |
8525 | *rdata = NULL; |
8526 | } |
8527 | n = SIZE_MAX; |
8528 | *status |= MPD_Malloc_error; |
8529 | goto out; |
8530 | } |
8531 | |
8532 | |
8533 | /* |
8534 | * Converts a multiprecision integer with base <= UINT16_MAX+1 to an mpd_t. |
8535 | * The least significant word of the source is srcdata[0]. |
8536 | */ |
8537 | void |
8538 | mpd_qimport_u16(mpd_t *result, |
8539 | const uint16_t *srcdata, size_t srclen, |
8540 | uint8_t srcsign, uint32_t srcbase, |
8541 | const mpd_context_t *ctx, uint32_t *status) |
8542 | { |
8543 | mpd_uint_t *usrc; /* uint16_t src copied to an mpd_uint_t array */ |
8544 | mpd_ssize_t rlen; /* length of the result */ |
8545 | size_t n; |
8546 | |
8547 | assert(srclen > 0); |
8548 | assert(srcbase <= (1U<<16)); |
8549 | |
8550 | rlen = _mpd_importsize(srclen, srcbase); |
8551 | if (rlen == MPD_SSIZE_MAX) { |
8552 | mpd_seterror(result, MPD_Invalid_operation, status); |
8553 | return; |
8554 | } |
8555 | |
8556 | usrc = mpd_alloc((mpd_size_t)srclen, sizeof *usrc); |
8557 | if (usrc == NULL) { |
8558 | mpd_seterror(result, MPD_Malloc_error, status); |
8559 | return; |
8560 | } |
8561 | for (n = 0; n < srclen; n++) { |
8562 | usrc[n] = srcdata[n]; |
8563 | } |
8564 | |
8565 | if (!mpd_qresize(result, rlen, status)) { |
8566 | goto finish; |
8567 | } |
8568 | |
8569 | n = _coeff_from_u16(result, rlen, usrc, srclen, srcbase, status); |
8570 | if (n == SIZE_MAX) { |
8571 | goto finish; |
8572 | } |
8573 | |
8574 | mpd_set_flags(result, srcsign); |
8575 | result->exp = 0; |
8576 | result->len = n; |
8577 | mpd_setdigits(result); |
8578 | |
8579 | mpd_qresize(result, result->len, status); |
8580 | mpd_qfinalize(result, ctx, status); |
8581 | |
8582 | |
8583 | finish: |
8584 | mpd_free(usrc); |
8585 | } |
8586 | |
8587 | /* |
8588 | * Converts a multiprecision integer with base <= UINT32_MAX to an mpd_t. |
8589 | * The least significant word of the source is srcdata[0]. |
8590 | */ |
8591 | void |
8592 | mpd_qimport_u32(mpd_t *result, |
8593 | const uint32_t *srcdata, size_t srclen, |
8594 | uint8_t srcsign, uint32_t srcbase, |
8595 | const mpd_context_t *ctx, uint32_t *status) |
8596 | { |
8597 | mpd_ssize_t rlen; /* length of the result */ |
8598 | size_t n; |
8599 | |
8600 | assert(srclen > 0); |
8601 | |
8602 | rlen = _mpd_importsize(srclen, srcbase); |
8603 | if (rlen == MPD_SSIZE_MAX) { |
8604 | mpd_seterror(result, MPD_Invalid_operation, status); |
8605 | return; |
8606 | } |
8607 | |
8608 | if (!mpd_qresize(result, rlen, status)) { |
8609 | return; |
8610 | } |
8611 | |
8612 | #ifdef CONFIG_64 |
8613 | n = _coeff_from_smaller_base(result, rlen, MPD_RADIX, |
8614 | srcdata, srclen, srcbase, |
8615 | status); |
8616 | #else |
8617 | if (srcbase == MPD_RADIX) { |
8618 | if (!mpd_qresize(result, srclen, status)) { |
8619 | return; |
8620 | } |
8621 | memcpy(result->data, srcdata, srclen * (sizeof *srcdata)); |
8622 | n = srclen; |
8623 | } |
8624 | else if (srcbase < MPD_RADIX) { |
8625 | n = _coeff_from_smaller_base(result, rlen, MPD_RADIX, |
8626 | srcdata, srclen, srcbase, |
8627 | status); |
8628 | } |
8629 | else { |
8630 | mpd_uint_t *usrc = mpd_alloc((mpd_size_t)srclen, sizeof *usrc); |
8631 | if (usrc == NULL) { |
8632 | mpd_seterror(result, MPD_Malloc_error, status); |
8633 | return; |
8634 | } |
8635 | for (n = 0; n < srclen; n++) { |
8636 | usrc[n] = srcdata[n]; |
8637 | } |
8638 | |
8639 | n = _coeff_from_larger_base(result, rlen, MPD_RADIX, |
8640 | usrc, (mpd_ssize_t)srclen, srcbase, |
8641 | status); |
8642 | mpd_free(usrc); |
8643 | } |
8644 | #endif |
8645 | |
8646 | if (n == SIZE_MAX) { |
8647 | return; |
8648 | } |
8649 | |
8650 | mpd_set_flags(result, srcsign); |
8651 | result->exp = 0; |
8652 | result->len = n; |
8653 | mpd_setdigits(result); |
8654 | |
8655 | mpd_qresize(result, result->len, status); |
8656 | mpd_qfinalize(result, ctx, status); |
8657 | } |
8658 | |
8659 | |
8660 | /******************************************************************************/ |
8661 | /* From triple */ |
8662 | /******************************************************************************/ |
8663 | |
8664 | #if defined(CONFIG_64) && defined(__SIZEOF_INT128__) |
8665 | static mpd_ssize_t |
8666 | _set_coeff(uint64_t data[3], uint64_t hi, uint64_t lo) |
8667 | { |
8668 | __uint128_t d = ((__uint128_t)hi << 64) + lo; |
8669 | __uint128_t q, r; |
8670 | |
8671 | q = d / MPD_RADIX; |
8672 | r = d % MPD_RADIX; |
8673 | data[0] = (uint64_t)r; |
8674 | d = q; |
8675 | |
8676 | q = d / MPD_RADIX; |
8677 | r = d % MPD_RADIX; |
8678 | data[1] = (uint64_t)r; |
8679 | d = q; |
8680 | |
8681 | q = d / MPD_RADIX; |
8682 | r = d % MPD_RADIX; |
8683 | data[2] = (uint64_t)r; |
8684 | |
8685 | if (q != 0) { |
8686 | abort(); /* GCOV_NOT_REACHED */ |
8687 | } |
8688 | |
8689 | return data[2] != 0 ? 3 : (data[1] != 0 ? 2 : 1); |
8690 | } |
8691 | #else |
8692 | static size_t |
8693 | _uint_from_u16(mpd_uint_t *w, mpd_ssize_t wlen, const uint16_t *u, size_t ulen) |
8694 | { |
8695 | const mpd_uint_t ubase = 1U<<16; |
8696 | mpd_ssize_t n = 0; |
8697 | mpd_uint_t carry; |
8698 | |
8699 | assert(wlen > 0 && ulen > 0); |
8700 | |
8701 | w[n++] = u[--ulen]; |
8702 | while (--ulen != SIZE_MAX) { |
8703 | carry = _mpd_shortmul_c(w, w, n, ubase); |
8704 | if (carry) { |
8705 | if (n >= wlen) { |
8706 | abort(); /* GCOV_NOT_REACHED */ |
8707 | } |
8708 | w[n++] = carry; |
8709 | } |
8710 | carry = _mpd_shortadd(w, n, u[ulen]); |
8711 | if (carry) { |
8712 | if (n >= wlen) { |
8713 | abort(); /* GCOV_NOT_REACHED */ |
8714 | } |
8715 | w[n++] = carry; |
8716 | } |
8717 | } |
8718 | |
8719 | return n; |
8720 | } |
8721 | |
8722 | static mpd_ssize_t |
8723 | _set_coeff(mpd_uint_t *data, mpd_ssize_t len, uint64_t hi, uint64_t lo) |
8724 | { |
8725 | uint16_t u16[8] = {0}; |
8726 | |
8727 | u16[7] = (uint16_t)((hi & 0xFFFF000000000000ULL) >> 48); |
8728 | u16[6] = (uint16_t)((hi & 0x0000FFFF00000000ULL) >> 32); |
8729 | u16[5] = (uint16_t)((hi & 0x00000000FFFF0000ULL) >> 16); |
8730 | u16[4] = (uint16_t) (hi & 0x000000000000FFFFULL); |
8731 | |
8732 | u16[3] = (uint16_t)((lo & 0xFFFF000000000000ULL) >> 48); |
8733 | u16[2] = (uint16_t)((lo & 0x0000FFFF00000000ULL) >> 32); |
8734 | u16[1] = (uint16_t)((lo & 0x00000000FFFF0000ULL) >> 16); |
8735 | u16[0] = (uint16_t) (lo & 0x000000000000FFFFULL); |
8736 | |
8737 | return (mpd_ssize_t)_uint_from_u16(data, len, u16, 8); |
8738 | } |
8739 | #endif |
8740 | |
8741 | static int |
8742 | _set_uint128_coeff_exp(mpd_t *result, uint64_t hi, uint64_t lo, mpd_ssize_t exp) |
8743 | { |
8744 | mpd_uint_t data[5] = {0}; |
8745 | uint32_t status = 0; |
8746 | mpd_ssize_t len; |
8747 | |
8748 | #if defined(CONFIG_64) && defined(__SIZEOF_INT128__) |
8749 | len = _set_coeff(data, hi, lo); |
8750 | #else |
8751 | len = _set_coeff(data, 5, hi, lo); |
8752 | #endif |
8753 | |
8754 | if (!mpd_qresize(result, len, &status)) { |
8755 | return -1; |
8756 | } |
8757 | |
8758 | for (mpd_ssize_t i = 0; i < len; i++) { |
8759 | result->data[i] = data[i]; |
8760 | } |
8761 | |
8762 | result->exp = exp; |
8763 | result->len = len; |
8764 | mpd_setdigits(result); |
8765 | |
8766 | return 0; |
8767 | } |
8768 | |
8769 | int |
8770 | mpd_from_uint128_triple(mpd_t *result, const mpd_uint128_triple_t *triple, uint32_t *status) |
8771 | { |
8772 | static const mpd_context_t maxcontext = { |
8773 | .prec=MPD_MAX_PREC, |
8774 | .emax=MPD_MAX_EMAX, |
8775 | .emin=MPD_MIN_EMIN, |
8776 | .round=MPD_ROUND_HALF_EVEN, |
8777 | .traps=MPD_Traps, |
8778 | .status=0, |
8779 | .newtrap=0, |
8780 | .clamp=0, |
8781 | .allcr=1, |
8782 | }; |
8783 | const enum mpd_triple_class tag = triple->tag; |
8784 | const uint8_t sign = triple->sign; |
8785 | const uint64_t hi = triple->hi; |
8786 | const uint64_t lo = triple->lo; |
8787 | mpd_ssize_t exp; |
8788 | |
8789 | #ifdef CONFIG_32 |
8790 | if (triple->exp < MPD_SSIZE_MIN || triple->exp > MPD_SSIZE_MAX) { |
8791 | goto conversion_error; |
8792 | } |
8793 | #endif |
8794 | exp = (mpd_ssize_t)triple->exp; |
8795 | |
8796 | switch (tag) { |
8797 | case MPD_TRIPLE_QNAN: case MPD_TRIPLE_SNAN: { |
8798 | if (sign > 1 || exp != 0) { |
8799 | goto conversion_error; |
8800 | } |
8801 | |
8802 | const uint8_t flags = tag == MPD_TRIPLE_QNAN ? MPD_NAN : MPD_SNAN; |
8803 | mpd_setspecial(result, sign, flags); |
8804 | |
8805 | if (hi == 0 && lo == 0) { /* no payload */ |
8806 | return 0; |
8807 | } |
8808 | |
8809 | if (_set_uint128_coeff_exp(result, hi, lo, exp) < 0) { |
8810 | goto malloc_error; |
8811 | } |
8812 | |
8813 | return 0; |
8814 | } |
8815 | |
8816 | case MPD_TRIPLE_INF: { |
8817 | if (sign > 1 || hi != 0 || lo != 0 || exp != 0) { |
8818 | goto conversion_error; |
8819 | } |
8820 | |
8821 | mpd_setspecial(result, sign, MPD_INF); |
8822 | |
8823 | return 0; |
8824 | } |
8825 | |
8826 | case MPD_TRIPLE_NORMAL: { |
8827 | if (sign > 1) { |
8828 | goto conversion_error; |
8829 | } |
8830 | |
8831 | const uint8_t flags = sign ? MPD_NEG : MPD_POS; |
8832 | mpd_set_flags(result, flags); |
8833 | |
8834 | if (exp > MPD_EXP_INF) { |
8835 | exp = MPD_EXP_INF; |
8836 | } |
8837 | if (exp == MPD_SSIZE_MIN) { |
8838 | exp = MPD_SSIZE_MIN+1; |
8839 | } |
8840 | |
8841 | if (_set_uint128_coeff_exp(result, hi, lo, exp) < 0) { |
8842 | goto malloc_error; |
8843 | } |
8844 | |
8845 | uint32_t workstatus = 0; |
8846 | mpd_qfinalize(result, &maxcontext, &workstatus); |
8847 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
8848 | goto conversion_error; |
8849 | } |
8850 | |
8851 | return 0; |
8852 | } |
8853 | |
8854 | default: |
8855 | goto conversion_error; |
8856 | } |
8857 | |
8858 | conversion_error: |
8859 | mpd_seterror(result, MPD_Conversion_syntax, status); |
8860 | return -1; |
8861 | |
8862 | malloc_error: |
8863 | mpd_seterror(result, MPD_Malloc_error, status); |
8864 | return -1; |
8865 | } |
8866 | |
8867 | |
8868 | /******************************************************************************/ |
8869 | /* As triple */ |
8870 | /******************************************************************************/ |
8871 | |
8872 | #if defined(CONFIG_64) && defined(__SIZEOF_INT128__) |
8873 | static void |
8874 | _get_coeff(uint64_t *hi, uint64_t *lo, const mpd_t *a) |
8875 | { |
8876 | __uint128_t u128 = 0; |
8877 | |
8878 | switch (a->len) { |
8879 | case 3: |
8880 | u128 = a->data[2]; /* fall through */ |
8881 | case 2: |
8882 | u128 = u128 * MPD_RADIX + a->data[1]; /* fall through */ |
8883 | case 1: |
8884 | u128 = u128 * MPD_RADIX + a->data[0]; |
8885 | break; |
8886 | default: |
8887 | abort(); /* GCOV_NOT_REACHED */ |
8888 | } |
8889 | |
8890 | *hi = u128 >> 64; |
8891 | *lo = (uint64_t)u128; |
8892 | } |
8893 | #else |
8894 | static size_t |
8895 | _uint_to_u16(uint16_t w[8], mpd_uint_t *u, mpd_ssize_t ulen) |
8896 | { |
8897 | const mpd_uint_t wbase = 1U<<16; |
8898 | size_t n = 0; |
8899 | |
8900 | assert(ulen > 0); |
8901 | |
8902 | do { |
8903 | if (n >= 8) { |
8904 | abort(); /* GCOV_NOT_REACHED */ |
8905 | } |
8906 | w[n++] = (uint16_t)_mpd_shortdiv(u, u, ulen, wbase); |
8907 | /* ulen is at least 1. u[ulen-1] can only be zero if ulen == 1. */ |
8908 | ulen = _mpd_real_size(u, ulen); |
8909 | |
8910 | } while (u[ulen-1] != 0); |
8911 | |
8912 | return n; |
8913 | } |
8914 | |
8915 | static void |
8916 | _get_coeff(uint64_t *hi, uint64_t *lo, const mpd_t *a) |
8917 | { |
8918 | uint16_t u16[8] = {0}; |
8919 | mpd_uint_t data[5] = {0}; |
8920 | |
8921 | switch (a->len) { |
8922 | case 5: |
8923 | data[4] = a->data[4]; /* fall through */ |
8924 | case 4: |
8925 | data[3] = a->data[3]; /* fall through */ |
8926 | case 3: |
8927 | data[2] = a->data[2]; /* fall through */ |
8928 | case 2: |
8929 | data[1] = a->data[1]; /* fall through */ |
8930 | case 1: |
8931 | data[0] = a->data[0]; |
8932 | break; |
8933 | default: |
8934 | abort(); /* GCOV_NOT_REACHED */ |
8935 | } |
8936 | |
8937 | _uint_to_u16(u16, data, a->len); |
8938 | |
8939 | *hi = (uint64_t)u16[7] << 48; |
8940 | *hi |= (uint64_t)u16[6] << 32; |
8941 | *hi |= (uint64_t)u16[5] << 16; |
8942 | *hi |= (uint64_t)u16[4]; |
8943 | |
8944 | *lo = (uint64_t)u16[3] << 48; |
8945 | *lo |= (uint64_t)u16[2] << 32; |
8946 | *lo |= (uint64_t)u16[1] << 16; |
8947 | *lo |= (uint64_t)u16[0]; |
8948 | } |
8949 | #endif |
8950 | |
8951 | static enum mpd_triple_class |
8952 | _coeff_as_uint128(uint64_t *hi, uint64_t *lo, const mpd_t *a) |
8953 | { |
8954 | #ifdef CONFIG_64 |
8955 | static mpd_uint_t uint128_max_data[3] = { 3374607431768211455ULL, 4028236692093846346ULL, 3ULL }; |
8956 | static const mpd_t uint128_max = { MPD_STATIC|MPD_CONST_DATA, 0, 39, 3, 3, uint128_max_data }; |
8957 | #else |
8958 | static mpd_uint_t uint128_max_data[5] = { 768211455U, 374607431U, 938463463U, 282366920U, 340U }; |
8959 | static const mpd_t uint128_max = { MPD_STATIC|MPD_CONST_DATA, 0, 39, 5, 5, uint128_max_data }; |
8960 | #endif |
8961 | enum mpd_triple_class ret = MPD_TRIPLE_NORMAL; |
8962 | uint32_t status = 0; |
8963 | mpd_t coeff; |
8964 | |
8965 | *hi = *lo = 0ULL; |
8966 | |
8967 | if (mpd_isspecial(a)) { |
8968 | if (mpd_isinfinite(a)) { |
8969 | return MPD_TRIPLE_INF; |
8970 | } |
8971 | |
8972 | ret = mpd_isqnan(a) ? MPD_TRIPLE_QNAN : MPD_TRIPLE_SNAN; |
8973 | if (a->len == 0) { /* no payload */ |
8974 | return ret; |
8975 | } |
8976 | } |
8977 | else if (mpd_iszero(a)) { |
8978 | return ret; |
8979 | } |
8980 | |
8981 | _mpd_copy_shared(&coeff, a); |
8982 | mpd_set_flags(&coeff, 0); |
8983 | coeff.exp = 0; |
8984 | |
8985 | if (mpd_qcmp(&coeff, &uint128_max, &status) > 0) { |
8986 | return MPD_TRIPLE_ERROR; |
8987 | } |
8988 | |
8989 | _get_coeff(hi, lo, &coeff); |
8990 | return ret; |
8991 | } |
8992 | |
8993 | mpd_uint128_triple_t |
8994 | mpd_as_uint128_triple(const mpd_t *a) |
8995 | { |
8996 | mpd_uint128_triple_t triple = { MPD_TRIPLE_ERROR, 0, 0, 0, 0 }; |
8997 | |
8998 | triple.tag = _coeff_as_uint128(&triple.hi, &triple.lo, a); |
8999 | if (triple.tag == MPD_TRIPLE_ERROR) { |
9000 | return triple; |
9001 | } |
9002 | |
9003 | triple.sign = !!mpd_isnegative(a); |
9004 | if (triple.tag == MPD_TRIPLE_NORMAL) { |
9005 | triple.exp = a->exp; |
9006 | } |
9007 | |
9008 | return triple; |
9009 | } |
9010 | |