/src/mpdecimal-4.0.0/libmpdec/mpdecimal.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Copyright (c) 2008-2024 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 | | * 2. Redistributions in binary form must reproduce the above copyright |
11 | | * notice, this list of conditions and the following disclaimer in the |
12 | | * documentation and/or other materials provided with the distribution. |
13 | | * |
14 | | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND |
15 | | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
16 | | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
17 | | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
18 | | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
19 | | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
20 | | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
21 | | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
22 | | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
23 | | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
24 | | * SUCH DAMAGE. |
25 | | */ |
26 | | |
27 | | |
28 | | #include <assert.h> |
29 | | #include <limits.h> |
30 | | #include <math.h> |
31 | | #include <stdio.h> |
32 | | #include <stdlib.h> |
33 | | #include <string.h> |
34 | | |
35 | | #include "basearith.h" |
36 | | #include "bits.h" |
37 | | #include "constants.h" |
38 | | #include "convolute.h" |
39 | | #include "crt.h" |
40 | | #include "mpalloc.h" |
41 | | #include "mpdecimal.h" |
42 | | #include "typearith.h" |
43 | | |
44 | | |
45 | | #ifdef PPRO |
46 | | #if defined(_MSC_VER) |
47 | | #include <float.h> |
48 | | #pragma float_control(precise, on) |
49 | | #pragma fenv_access(on) |
50 | | #elif !defined(__OpenBSD__) && !defined(__NetBSD__) |
51 | | /* The C99 standard requires the pragma */ |
52 | | #include <fenv.h> |
53 | | #pragma STDC FENV_ACCESS ON |
54 | | #endif |
55 | | #endif |
56 | | |
57 | | |
58 | | /* Disable warning that is part of -Wextra since gcc 7.0 */ |
59 | | #if defined(__GNUC__) && !defined(__INTEL_COMPILER) && __GNUC__ >= 7 |
60 | | #pragma GCC diagnostic ignored "-Wimplicit-fallthrough" |
61 | | #endif |
62 | | |
63 | | |
64 | | #if defined(_MSC_VER) |
65 | | #define ALWAYS_INLINE __forceinline |
66 | | #elif defined (__IBMC__) || defined(__COMPCERT__) || defined(LEGACY_COMPILER) |
67 | | #define ALWAYS_INLINE |
68 | | #undef inline |
69 | | #define inline |
70 | | #else |
71 | | #ifdef TEST_COVERAGE |
72 | | #define ALWAYS_INLINE |
73 | | #else |
74 | | #define ALWAYS_INLINE inline |
75 | | #endif |
76 | | #endif |
77 | | |
78 | | |
79 | 2.65M | #define MPD_NEWTONDIV_CUTOFF 1024L |
80 | | |
81 | | #define MPD_NEW_STATIC(name, flags, exp, digits, len) \ |
82 | 9.25M | mpd_uint_t name##_data[MPD_MINALLOC_MAX]; \ |
83 | 9.25M | mpd_t name = {flags|MPD_STATIC|MPD_STATIC_DATA, exp, digits, \ |
84 | 9.25M | len, MPD_MINALLOC_MAX, name##_data} |
85 | | |
86 | | #define MPD_NEW_CONST(name, flags, exp, digits, len, alloc, initval) \ |
87 | 4.10k | mpd_uint_t name##_data[alloc] = {initval}; \ |
88 | 4.10k | mpd_t name = {flags|MPD_STATIC|MPD_CONST_DATA, exp, digits, \ |
89 | 4.10k | len, alloc, name##_data} |
90 | | |
91 | | #define MPD_NEW_SHARED(name, a) \ |
92 | 0 | mpd_t name = {(a->flags&~MPD_DATAFLAGS)|MPD_STATIC|MPD_SHARED_DATA, \ |
93 | 0 | a->exp, a->digits, a->len, a->alloc, a->data} |
94 | | |
95 | | |
96 | | static mpd_uint_t data_one[1] = {1}; |
97 | | static mpd_uint_t data_zero[1] = {0}; |
98 | | static const mpd_t one = {MPD_STATIC|MPD_CONST_DATA, 0, 1, 1, 1, data_one}; |
99 | | static const mpd_t minus_one = {MPD_NEG|MPD_STATIC|MPD_CONST_DATA, 0, 1, 1, 1, |
100 | | data_one}; |
101 | | static const mpd_t zero = {MPD_STATIC|MPD_CONST_DATA, 0, 1, 1, 1, data_zero}; |
102 | | |
103 | | static inline void _mpd_check_exp(mpd_t *dec, const mpd_context_t *ctx, |
104 | | uint32_t *status); |
105 | | static void _settriple(mpd_t *result, uint8_t sign, mpd_uint_t a, |
106 | | mpd_ssize_t exp); |
107 | | static inline mpd_ssize_t _mpd_real_size(mpd_uint_t *data, mpd_ssize_t size); |
108 | | |
109 | | static int _mpd_cmp_abs(const mpd_t *a, const mpd_t *b); |
110 | | |
111 | | static void _mpd_qadd(mpd_t *result, const mpd_t *a, const mpd_t *b, |
112 | | const mpd_context_t *ctx, uint32_t *status); |
113 | | static inline void _mpd_qmul(mpd_t *result, const mpd_t *a, const mpd_t *b, |
114 | | const mpd_context_t *ctx, uint32_t *status); |
115 | | static void _mpd_base_ndivmod(mpd_t *q, mpd_t *r, const mpd_t *a, |
116 | | const mpd_t *b, uint32_t *status); |
117 | | static inline void _mpd_qpow_uint(mpd_t *result, const mpd_t *base, |
118 | | mpd_uint_t exp, uint8_t resultsign, |
119 | | const mpd_context_t *ctx, uint32_t *status); |
120 | | |
121 | | static mpd_uint_t mpd_qsshiftr(mpd_t *result, const mpd_t *a, mpd_ssize_t n); |
122 | | |
123 | | |
124 | | /******************************************************************************/ |
125 | | /* Version */ |
126 | | /******************************************************************************/ |
127 | | |
128 | | const char * |
129 | | mpd_version(void) |
130 | 0 | { |
131 | 0 | return MPD_VERSION; |
132 | 0 | } |
133 | | |
134 | | |
135 | | /******************************************************************************/ |
136 | | /* Performance critical inline functions */ |
137 | | /******************************************************************************/ |
138 | | |
139 | | #ifdef CONFIG_64 |
140 | | /* Digits in a word, primarily useful for the most significant word. */ |
141 | | ALWAYS_INLINE int |
142 | | mpd_word_digits(mpd_uint_t word) |
143 | 12.0M | { |
144 | 12.0M | if (word < mpd_pow10[9]) { |
145 | 6.58M | if (word < mpd_pow10[4]) { |
146 | 4.11M | if (word < mpd_pow10[2]) { |
147 | 3.47M | return (word < mpd_pow10[1]) ? 1 : 2; |
148 | 3.47M | } |
149 | 633k | return (word < mpd_pow10[3]) ? 3 : 4; |
150 | 4.11M | } |
151 | 2.47M | if (word < mpd_pow10[6]) { |
152 | 767k | return (word < mpd_pow10[5]) ? 5 : 6; |
153 | 767k | } |
154 | 1.70M | if (word < mpd_pow10[8]) { |
155 | 963k | return (word < mpd_pow10[7]) ? 7 : 8; |
156 | 963k | } |
157 | 740k | return 9; |
158 | 1.70M | } |
159 | 5.43M | if (word < mpd_pow10[14]) { |
160 | 2.51M | if (word < mpd_pow10[11]) { |
161 | 648k | return (word < mpd_pow10[10]) ? 10 : 11; |
162 | 648k | } |
163 | 1.86M | if (word < mpd_pow10[13]) { |
164 | 1.48M | return (word < mpd_pow10[12]) ? 12 : 13; |
165 | 1.48M | } |
166 | 378k | return 14; |
167 | 1.86M | } |
168 | 2.92M | if (word < mpd_pow10[18]) { |
169 | 2.04M | if (word < mpd_pow10[16]) { |
170 | 711k | return (word < mpd_pow10[15]) ? 15 : 16; |
171 | 711k | } |
172 | 1.33M | return (word < mpd_pow10[17]) ? 17 : 18; |
173 | 2.04M | } |
174 | | |
175 | 874k | return (word < mpd_pow10[19]) ? 19 : 20; |
176 | 2.92M | } |
177 | | #else |
178 | | ALWAYS_INLINE int |
179 | | mpd_word_digits(mpd_uint_t word) |
180 | | { |
181 | | if (word < mpd_pow10[4]) { |
182 | | if (word < mpd_pow10[2]) { |
183 | | return (word < mpd_pow10[1]) ? 1 : 2; |
184 | | } |
185 | | return (word < mpd_pow10[3]) ? 3 : 4; |
186 | | } |
187 | | if (word < mpd_pow10[6]) { |
188 | | return (word < mpd_pow10[5]) ? 5 : 6; |
189 | | } |
190 | | if (word < mpd_pow10[8]) { |
191 | | return (word < mpd_pow10[7]) ? 7 : 8; |
192 | | } |
193 | | |
194 | | return (word < mpd_pow10[9]) ? 9 : 10; |
195 | | } |
196 | | #endif |
197 | | |
198 | | |
199 | | /* Adjusted exponent */ |
200 | | ALWAYS_INLINE mpd_ssize_t |
201 | | mpd_adjexp(const mpd_t *dec) |
202 | 16.6M | { |
203 | 16.6M | return (dec->exp + dec->digits) - 1; |
204 | 16.6M | } |
205 | | |
206 | | /* Etiny */ |
207 | | ALWAYS_INLINE mpd_ssize_t |
208 | | mpd_etiny(const mpd_context_t *ctx) |
209 | 0 | { |
210 | 0 | return ctx->emin - (ctx->prec - 1); |
211 | 0 | } |
212 | | |
213 | | /* Etop: used for folding down in IEEE clamping */ |
214 | | ALWAYS_INLINE mpd_ssize_t |
215 | | mpd_etop(const mpd_context_t *ctx) |
216 | 0 | { |
217 | 0 | return ctx->emax - (ctx->prec - 1); |
218 | 0 | } |
219 | | |
220 | | /* Most significant word */ |
221 | | ALWAYS_INLINE mpd_uint_t |
222 | | mpd_msword(const mpd_t *dec) |
223 | 25.0M | { |
224 | 25.0M | assert(dec->len > 0); |
225 | 25.0M | return dec->data[dec->len-1]; |
226 | 25.0M | } |
227 | | |
228 | | /* Most significant digit of a word */ |
229 | | inline mpd_uint_t |
230 | | mpd_msd(mpd_uint_t word) |
231 | 0 | { |
232 | 0 | int n; |
233 | |
|
234 | 0 | n = mpd_word_digits(word); |
235 | 0 | return word / mpd_pow10[n-1]; |
236 | 0 | } |
237 | | |
238 | | /* Least significant digit of a word */ |
239 | | ALWAYS_INLINE mpd_uint_t |
240 | | mpd_lsd(mpd_uint_t word) |
241 | 153 | { |
242 | 153 | return word % 10; |
243 | 153 | } |
244 | | |
245 | | /* Coefficient size needed to store 'digits' */ |
246 | | mpd_ssize_t |
247 | | mpd_digits_to_size(mpd_ssize_t digits) |
248 | 388 | { |
249 | 388 | mpd_ssize_t q, r; |
250 | | |
251 | 388 | _mpd_idiv_word(&q, &r, digits, MPD_RDIGITS); |
252 | 388 | return (r == 0) ? q : q+1; |
253 | 388 | } |
254 | | |
255 | | /* Number of digits in the exponent. Not defined for MPD_SSIZE_MIN. */ |
256 | | inline int |
257 | | mpd_exp_digits(mpd_ssize_t exp) |
258 | 0 | { |
259 | 0 | exp = (exp < 0) ? -exp : exp; |
260 | 0 | return mpd_word_digits(exp); |
261 | 0 | } |
262 | | |
263 | | /* Canonical */ |
264 | | ALWAYS_INLINE int |
265 | | mpd_iscanonical(const mpd_t *dec) |
266 | 0 | { |
267 | 0 | (void)dec; |
268 | 0 | return 1; |
269 | 0 | } |
270 | | |
271 | | /* Finite */ |
272 | | ALWAYS_INLINE int |
273 | | mpd_isfinite(const mpd_t *dec) |
274 | 1.85M | { |
275 | 1.85M | return !(dec->flags & MPD_SPECIAL); |
276 | 1.85M | } |
277 | | |
278 | | /* Infinite */ |
279 | | ALWAYS_INLINE int |
280 | | mpd_isinfinite(const mpd_t *dec) |
281 | 414 | { |
282 | 414 | return dec->flags & MPD_INF; |
283 | 414 | } |
284 | | |
285 | | /* NaN */ |
286 | | ALWAYS_INLINE int |
287 | | mpd_isnan(const mpd_t *dec) |
288 | 0 | { |
289 | 0 | return dec->flags & (MPD_NAN|MPD_SNAN); |
290 | 0 | } |
291 | | |
292 | | /* Negative */ |
293 | | ALWAYS_INLINE int |
294 | | mpd_isnegative(const mpd_t *dec) |
295 | 12.1k | { |
296 | 12.1k | return dec->flags & MPD_NEG; |
297 | 12.1k | } |
298 | | |
299 | | /* Positive */ |
300 | | ALWAYS_INLINE int |
301 | | mpd_ispositive(const mpd_t *dec) |
302 | 0 | { |
303 | 0 | return !(dec->flags & MPD_NEG); |
304 | 0 | } |
305 | | |
306 | | /* qNaN */ |
307 | | ALWAYS_INLINE int |
308 | | mpd_isqnan(const mpd_t *dec) |
309 | 80 | { |
310 | 80 | return dec->flags & MPD_NAN; |
311 | 80 | } |
312 | | |
313 | | /* Signed */ |
314 | | ALWAYS_INLINE int |
315 | | mpd_issigned(const mpd_t *dec) |
316 | 0 | { |
317 | 0 | return dec->flags & MPD_NEG; |
318 | 0 | } |
319 | | |
320 | | /* sNaN */ |
321 | | ALWAYS_INLINE int |
322 | | mpd_issnan(const mpd_t *dec) |
323 | 0 | { |
324 | 0 | return dec->flags & MPD_SNAN; |
325 | 0 | } |
326 | | |
327 | | /* Special */ |
328 | | ALWAYS_INLINE int |
329 | | mpd_isspecial(const mpd_t *dec) |
330 | 31.6M | { |
331 | 31.6M | return dec->flags & MPD_SPECIAL; |
332 | 31.6M | } |
333 | | |
334 | | /* Zero */ |
335 | | ALWAYS_INLINE int |
336 | | mpd_iszero(const mpd_t *dec) |
337 | 1.85M | { |
338 | 1.85M | return !mpd_isspecial(dec) && mpd_msword(dec) == 0; |
339 | 1.85M | } |
340 | | |
341 | | /* Test for zero when specials have been ruled out already */ |
342 | | ALWAYS_INLINE int |
343 | | mpd_iszerocoeff(const mpd_t *dec) |
344 | 11.1M | { |
345 | 11.1M | return mpd_msword(dec) == 0; |
346 | 11.1M | } |
347 | | |
348 | | /* Normal */ |
349 | | inline int |
350 | | mpd_isnormal(const mpd_t *dec, const mpd_context_t *ctx) |
351 | 0 | { |
352 | 0 | if (mpd_isspecial(dec)) return 0; |
353 | 0 | if (mpd_iszerocoeff(dec)) return 0; |
354 | | |
355 | 0 | return mpd_adjexp(dec) >= ctx->emin; |
356 | 0 | } |
357 | | |
358 | | /* Subnormal */ |
359 | | inline int |
360 | | mpd_issubnormal(const mpd_t *dec, const mpd_context_t *ctx) |
361 | 3.50k | { |
362 | 3.50k | if (mpd_isspecial(dec)) return 0; |
363 | 3.50k | if (mpd_iszerocoeff(dec)) return 0; |
364 | | |
365 | 3.50k | return mpd_adjexp(dec) < ctx->emin; |
366 | 3.50k | } |
367 | | |
368 | | /* Odd word */ |
369 | | ALWAYS_INLINE int |
370 | | mpd_isoddword(mpd_uint_t word) |
371 | 1.84M | { |
372 | 1.84M | return word & 1; |
373 | 1.84M | } |
374 | | |
375 | | /* Odd coefficient */ |
376 | | ALWAYS_INLINE int |
377 | | mpd_isoddcoeff(const mpd_t *dec) |
378 | 0 | { |
379 | 0 | return mpd_isoddword(dec->data[0]); |
380 | 0 | } |
381 | | |
382 | | /* 0 if dec is positive, 1 if dec is negative */ |
383 | | ALWAYS_INLINE uint8_t |
384 | | mpd_sign(const mpd_t *dec) |
385 | 23.0M | { |
386 | 23.0M | return dec->flags & MPD_NEG; |
387 | 23.0M | } |
388 | | |
389 | | /* 1 if dec is positive, -1 if dec is negative */ |
390 | | ALWAYS_INLINE int |
391 | | mpd_arith_sign(const mpd_t *dec) |
392 | 116 | { |
393 | 116 | return 1 - 2 * mpd_isnegative(dec); |
394 | 116 | } |
395 | | |
396 | | /* Radix */ |
397 | | ALWAYS_INLINE long |
398 | | mpd_radix(void) |
399 | 0 | { |
400 | 0 | return 10; |
401 | 0 | } |
402 | | |
403 | | /* Dynamic decimal */ |
404 | | ALWAYS_INLINE int |
405 | | mpd_isdynamic(const mpd_t *dec) |
406 | 9.37M | { |
407 | 9.37M | return !(dec->flags & MPD_STATIC); |
408 | 9.37M | } |
409 | | |
410 | | /* Static decimal */ |
411 | | ALWAYS_INLINE int |
412 | | mpd_isstatic(const mpd_t *dec) |
413 | 0 | { |
414 | 0 | return dec->flags & MPD_STATIC; |
415 | 0 | } |
416 | | |
417 | | /* Data of decimal is dynamic */ |
418 | | ALWAYS_INLINE int |
419 | | mpd_isdynamic_data(const mpd_t *dec) |
420 | 10.9M | { |
421 | 10.9M | return !(dec->flags & MPD_DATAFLAGS); |
422 | 10.9M | } |
423 | | |
424 | | /* Data of decimal is static */ |
425 | | ALWAYS_INLINE int |
426 | | mpd_isstatic_data(const mpd_t *dec) |
427 | 2.17M | { |
428 | 2.17M | return dec->flags & MPD_STATIC_DATA; |
429 | 2.17M | } |
430 | | |
431 | | /* Data of decimal is shared */ |
432 | | ALWAYS_INLINE int |
433 | | mpd_isshared_data(const mpd_t *dec) |
434 | 0 | { |
435 | 0 | return dec->flags & MPD_SHARED_DATA; |
436 | 0 | } |
437 | | |
438 | | /* Data of decimal is const */ |
439 | | ALWAYS_INLINE int |
440 | | mpd_isconst_data(const mpd_t *dec) |
441 | 0 | { |
442 | 0 | return dec->flags & MPD_CONST_DATA; |
443 | 0 | } |
444 | | |
445 | | |
446 | | /******************************************************************************/ |
447 | | /* Inline memory handling */ |
448 | | /******************************************************************************/ |
449 | | |
450 | | /* Fill destination with zeros */ |
451 | | ALWAYS_INLINE void |
452 | | mpd_uint_zero(mpd_uint_t *dest, mpd_size_t len) |
453 | 519k | { |
454 | 519k | mpd_size_t i; |
455 | | |
456 | 13.7M | for (i = 0; i < len; i++) { |
457 | 13.2M | dest[i] = 0; |
458 | 13.2M | } |
459 | 519k | } |
460 | | |
461 | | /* Free a decimal */ |
462 | | ALWAYS_INLINE void |
463 | | mpd_del(mpd_t *dec) |
464 | 9.37M | { |
465 | 9.37M | if (mpd_isdynamic_data(dec)) { |
466 | 169k | mpd_free(dec->data); |
467 | 169k | } |
468 | 9.37M | if (mpd_isdynamic(dec)) { |
469 | 167k | mpd_free(dec); |
470 | 167k | } |
471 | 9.37M | } |
472 | | |
473 | | /* |
474 | | * Resize the coefficient. Existing data up to 'nwords' is left untouched. |
475 | | * Return 1 on success, 0 otherwise. |
476 | | * |
477 | | * Input invariant: MPD_MINALLOC <= result->alloc. |
478 | | * |
479 | | * Case nwords == result->alloc: |
480 | | * 'result' is unchanged. Return 1. |
481 | | * |
482 | | * Case nwords > result->alloc: |
483 | | * Case realloc success: |
484 | | * The value of 'result' does not change. Return 1. |
485 | | * Case realloc failure: |
486 | | * 'result' is NaN, status is updated with MPD_Malloc_error. Return 0. |
487 | | * |
488 | | * Case nwords < result->alloc: |
489 | | * Case is_static_data or realloc failure [1]: |
490 | | * 'result' is unchanged. Return 1. |
491 | | * Case realloc success: |
492 | | * The value of result is undefined (expected). Return 1. |
493 | | * |
494 | | * |
495 | | * [1] In that case the old (now oversized) area is still valid. |
496 | | */ |
497 | | ALWAYS_INLINE int |
498 | | mpd_qresize(mpd_t *result, mpd_ssize_t nwords, uint32_t *status) |
499 | 19.6M | { |
500 | 19.6M | assert(!mpd_isconst_data(result)); /* illegal operation for a const */ |
501 | 19.6M | assert(!mpd_isshared_data(result)); /* illegal operation for a shared */ |
502 | 19.6M | assert(MPD_MINALLOC <= result->alloc); |
503 | | |
504 | 19.6M | nwords = (nwords <= MPD_MINALLOC) ? MPD_MINALLOC : nwords; |
505 | 19.6M | if (nwords == result->alloc) { |
506 | 17.5M | return 1; |
507 | 17.5M | } |
508 | 2.12M | if (mpd_isstatic_data(result)) { |
509 | 194 | if (nwords > result->alloc) { |
510 | 194 | return mpd_switch_to_dyn(result, nwords, status); |
511 | 194 | } |
512 | 0 | return 1; |
513 | 194 | } |
514 | | |
515 | 2.12M | return mpd_realloc_dyn(result, nwords, status); |
516 | 2.12M | } |
517 | | |
518 | | /* Same as mpd_qresize, but do not set the result to NaN on failure. */ |
519 | | static ALWAYS_INLINE int |
520 | | mpd_qresize_cxx(mpd_t *result, mpd_ssize_t nwords) |
521 | 0 | { |
522 | 0 | assert(!mpd_isconst_data(result)); /* illegal operation for a const */ |
523 | 0 | assert(!mpd_isshared_data(result)); /* illegal operation for a shared */ |
524 | 0 | assert(MPD_MINALLOC <= result->alloc); |
525 | |
|
526 | 0 | nwords = (nwords <= MPD_MINALLOC) ? MPD_MINALLOC : nwords; |
527 | 0 | if (nwords == result->alloc) { |
528 | 0 | return 1; |
529 | 0 | } |
530 | 0 | if (mpd_isstatic_data(result)) { |
531 | 0 | if (nwords > result->alloc) { |
532 | 0 | return mpd_switch_to_dyn_cxx(result, nwords); |
533 | 0 | } |
534 | 0 | return 1; |
535 | 0 | } |
536 | | |
537 | 0 | return mpd_realloc_dyn_cxx(result, nwords); |
538 | 0 | } |
539 | | |
540 | | /* Same as mpd_qresize, but the complete coefficient (including the old |
541 | | * memory area!) is initialized to zero. */ |
542 | | ALWAYS_INLINE int |
543 | | mpd_qresize_zero(mpd_t *result, mpd_ssize_t nwords, uint32_t *status) |
544 | 0 | { |
545 | 0 | assert(!mpd_isconst_data(result)); /* illegal operation for a const */ |
546 | 0 | assert(!mpd_isshared_data(result)); /* illegal operation for a shared */ |
547 | 0 | assert(MPD_MINALLOC <= result->alloc); |
548 | |
|
549 | 0 | nwords = (nwords <= MPD_MINALLOC) ? MPD_MINALLOC : nwords; |
550 | 0 | if (nwords != result->alloc) { |
551 | 0 | if (mpd_isstatic_data(result)) { |
552 | 0 | if (nwords > result->alloc) { |
553 | 0 | return mpd_switch_to_dyn_zero(result, nwords, status); |
554 | 0 | } |
555 | 0 | } |
556 | 0 | else if (!mpd_realloc_dyn(result, nwords, status)) { |
557 | 0 | return 0; |
558 | 0 | } |
559 | 0 | } |
560 | | |
561 | 0 | mpd_uint_zero(result->data, nwords); |
562 | 0 | return 1; |
563 | 0 | } |
564 | | |
565 | | /* |
566 | | * Reduce memory size for the coefficient to MPD_MINALLOC. In theory, |
567 | | * realloc may fail even when reducing the memory size. But in that case |
568 | | * the old memory area is always big enough, so checking for MPD_Malloc_error |
569 | | * is not imperative. |
570 | | */ |
571 | | ALWAYS_INLINE void |
572 | | mpd_minalloc(mpd_t *result) |
573 | 58.5k | { |
574 | 58.5k | assert(!mpd_isconst_data(result)); /* illegal operation for a const */ |
575 | 58.5k | assert(!mpd_isshared_data(result)); /* illegal operation for a shared */ |
576 | | |
577 | 58.5k | if (!mpd_isstatic_data(result) && result->alloc > MPD_MINALLOC) { |
578 | 32 | uint8_t err = 0; |
579 | 32 | result->data = mpd_realloc(result->data, MPD_MINALLOC, |
580 | 32 | sizeof *result->data, &err); |
581 | 32 | if (!err) { |
582 | 32 | result->alloc = MPD_MINALLOC; |
583 | 32 | } |
584 | 32 | } |
585 | 58.5k | } |
586 | | |
587 | | int |
588 | | mpd_resize(mpd_t *result, mpd_ssize_t nwords, mpd_context_t *ctx) |
589 | 0 | { |
590 | 0 | uint32_t status = 0; |
591 | 0 | if (!mpd_qresize(result, nwords, &status)) { |
592 | 0 | mpd_addstatus_raise(ctx, status); |
593 | 0 | return 0; |
594 | 0 | } |
595 | 0 | return 1; |
596 | 0 | } |
597 | | |
598 | | int |
599 | | mpd_resize_zero(mpd_t *result, mpd_ssize_t nwords, mpd_context_t *ctx) |
600 | 0 | { |
601 | 0 | uint32_t status = 0; |
602 | 0 | if (!mpd_qresize_zero(result, nwords, &status)) { |
603 | 0 | mpd_addstatus_raise(ctx, status); |
604 | 0 | return 0; |
605 | 0 | } |
606 | 0 | return 1; |
607 | 0 | } |
608 | | |
609 | | |
610 | | /******************************************************************************/ |
611 | | /* Set attributes of a decimal */ |
612 | | /******************************************************************************/ |
613 | | |
614 | | /* Set digits. Assumption: result->len is initialized and > 0. */ |
615 | | inline void |
616 | | mpd_setdigits(mpd_t *result) |
617 | 12.0M | { |
618 | 12.0M | mpd_ssize_t wdigits = mpd_word_digits(mpd_msword(result)); |
619 | 12.0M | result->digits = wdigits + (result->len-1) * MPD_RDIGITS; |
620 | 12.0M | } |
621 | | |
622 | | /* Set sign */ |
623 | | ALWAYS_INLINE void |
624 | | mpd_set_sign(mpd_t *result, uint8_t sign) |
625 | 3.50k | { |
626 | 3.50k | result->flags &= ~MPD_NEG; |
627 | 3.50k | result->flags |= sign; |
628 | 3.50k | } |
629 | | |
630 | | /* Copy sign from another decimal */ |
631 | | ALWAYS_INLINE void |
632 | | mpd_signcpy(mpd_t *result, const mpd_t *a) |
633 | 0 | { |
634 | 0 | uint8_t sign = a->flags&MPD_NEG; |
635 | |
|
636 | 0 | result->flags &= ~MPD_NEG; |
637 | 0 | result->flags |= sign; |
638 | 0 | } |
639 | | |
640 | | /* Set infinity */ |
641 | | ALWAYS_INLINE void |
642 | | mpd_set_infinity(mpd_t *result) |
643 | 0 | { |
644 | 0 | result->flags &= ~MPD_SPECIAL; |
645 | 0 | result->flags |= MPD_INF; |
646 | 0 | } |
647 | | |
648 | | /* Set qNaN */ |
649 | | ALWAYS_INLINE void |
650 | | mpd_set_qnan(mpd_t *result) |
651 | 332 | { |
652 | 332 | result->flags &= ~MPD_SPECIAL; |
653 | 332 | result->flags |= MPD_NAN; |
654 | 332 | } |
655 | | |
656 | | /* Set sNaN */ |
657 | | ALWAYS_INLINE void |
658 | | mpd_set_snan(mpd_t *result) |
659 | 0 | { |
660 | 0 | result->flags &= ~MPD_SPECIAL; |
661 | 0 | result->flags |= MPD_SNAN; |
662 | 0 | } |
663 | | |
664 | | /* Set to negative */ |
665 | | ALWAYS_INLINE void |
666 | | mpd_set_negative(mpd_t *result) |
667 | 0 | { |
668 | 0 | result->flags |= MPD_NEG; |
669 | 0 | } |
670 | | |
671 | | /* Set to positive */ |
672 | | ALWAYS_INLINE void |
673 | | mpd_set_positive(mpd_t *result) |
674 | 7.36k | { |
675 | 7.36k | result->flags &= ~MPD_NEG; |
676 | 7.36k | } |
677 | | |
678 | | /* Set to dynamic */ |
679 | | ALWAYS_INLINE void |
680 | | mpd_set_dynamic(mpd_t *result) |
681 | 0 | { |
682 | 0 | result->flags &= ~MPD_STATIC; |
683 | 0 | } |
684 | | |
685 | | /* Set to static */ |
686 | | ALWAYS_INLINE void |
687 | | mpd_set_static(mpd_t *result) |
688 | 0 | { |
689 | 0 | result->flags |= MPD_STATIC; |
690 | 0 | } |
691 | | |
692 | | /* Set data to dynamic */ |
693 | | ALWAYS_INLINE void |
694 | | mpd_set_dynamic_data(mpd_t *result) |
695 | 1.61M | { |
696 | 1.61M | result->flags &= ~MPD_DATAFLAGS; |
697 | 1.61M | } |
698 | | |
699 | | /* Set data to static */ |
700 | | ALWAYS_INLINE void |
701 | | mpd_set_static_data(mpd_t *result) |
702 | 0 | { |
703 | 0 | result->flags &= ~MPD_DATAFLAGS; |
704 | 0 | result->flags |= MPD_STATIC_DATA; |
705 | 0 | } |
706 | | |
707 | | /* Set data to shared */ |
708 | | ALWAYS_INLINE void |
709 | | mpd_set_shared_data(mpd_t *result) |
710 | 0 | { |
711 | 0 | result->flags &= ~MPD_DATAFLAGS; |
712 | 0 | result->flags |= MPD_SHARED_DATA; |
713 | 0 | } |
714 | | |
715 | | /* Set data to const */ |
716 | | ALWAYS_INLINE void |
717 | | mpd_set_const_data(mpd_t *result) |
718 | 0 | { |
719 | 0 | result->flags &= ~MPD_DATAFLAGS; |
720 | 0 | result->flags |= MPD_CONST_DATA; |
721 | 0 | } |
722 | | |
723 | | /* Clear flags, preserving memory attributes. */ |
724 | | ALWAYS_INLINE void |
725 | | mpd_clear_flags(mpd_t *result) |
726 | 4 | { |
727 | 4 | result->flags &= (MPD_STATIC|MPD_DATAFLAGS); |
728 | 4 | } |
729 | | |
730 | | /* Set flags, preserving memory attributes. */ |
731 | | ALWAYS_INLINE void |
732 | | mpd_set_flags(mpd_t *result, uint8_t flags) |
733 | 12.0M | { |
734 | 12.0M | result->flags &= (MPD_STATIC|MPD_DATAFLAGS); |
735 | 12.0M | result->flags |= flags; |
736 | 12.0M | } |
737 | | |
738 | | /* Copy flags, preserving memory attributes of result. */ |
739 | | ALWAYS_INLINE void |
740 | | mpd_copy_flags(mpd_t *result, const mpd_t *a) |
741 | 121k | { |
742 | 121k | uint8_t aflags = a->flags; |
743 | 121k | result->flags &= (MPD_STATIC|MPD_DATAFLAGS); |
744 | 121k | result->flags |= (aflags & ~(MPD_STATIC|MPD_DATAFLAGS)); |
745 | 121k | } |
746 | | |
747 | | /* Initialize a workcontext from ctx. Set traps, flags and newtrap to 0. */ |
748 | | static inline void |
749 | | mpd_workcontext(mpd_context_t *workctx, const mpd_context_t *ctx) |
750 | 0 | { |
751 | 0 | workctx->prec = ctx->prec; |
752 | 0 | workctx->emax = ctx->emax; |
753 | 0 | workctx->emin = ctx->emin; |
754 | 0 | workctx->round = ctx->round; |
755 | 0 | workctx->traps = 0; |
756 | 0 | workctx->status = 0; |
757 | 0 | workctx->newtrap = 0; |
758 | 0 | workctx->clamp = ctx->clamp; |
759 | 0 | workctx->allcr = ctx->allcr; |
760 | 0 | } |
761 | | |
762 | | |
763 | | /******************************************************************************/ |
764 | | /* Getting and setting parts of decimals */ |
765 | | /******************************************************************************/ |
766 | | |
767 | | /* Flip the sign of a decimal */ |
768 | | static inline void |
769 | | _mpd_negate(mpd_t *dec) |
770 | 0 | { |
771 | 0 | dec->flags ^= MPD_NEG; |
772 | 0 | } |
773 | | |
774 | | /* Set coefficient to zero */ |
775 | | void |
776 | | mpd_zerocoeff(mpd_t *result) |
777 | 0 | { |
778 | 0 | mpd_minalloc(result); |
779 | 0 | result->digits = 1; |
780 | 0 | result->len = 1; |
781 | 0 | result->data[0] = 0; |
782 | 0 | } |
783 | | |
784 | | /* Set the coefficient to all nines. */ |
785 | | void |
786 | | mpd_qmaxcoeff(mpd_t *result, const mpd_context_t *ctx, uint32_t *status) |
787 | 0 | { |
788 | 0 | mpd_ssize_t len, r; |
789 | |
|
790 | 0 | _mpd_idiv_word(&len, &r, ctx->prec, MPD_RDIGITS); |
791 | 0 | len = (r == 0) ? len : len+1; |
792 | |
|
793 | 0 | if (!mpd_qresize(result, len, status)) { |
794 | 0 | return; |
795 | 0 | } |
796 | | |
797 | 0 | result->len = len; |
798 | 0 | result->digits = ctx->prec; |
799 | |
|
800 | 0 | --len; |
801 | 0 | if (r > 0) { |
802 | 0 | result->data[len--] = mpd_pow10[r]-1; |
803 | 0 | } |
804 | 0 | for (; len >= 0; --len) { |
805 | 0 | result->data[len] = MPD_RADIX-1; |
806 | 0 | } |
807 | 0 | } |
808 | | |
809 | | /* |
810 | | * Cut off the most significant digits so that the rest fits in ctx->prec. |
811 | | * Cannot fail. |
812 | | */ |
813 | | static void |
814 | | _mpd_cap(mpd_t *result, const mpd_context_t *ctx) |
815 | 4 | { |
816 | 4 | uint32_t dummy; |
817 | 4 | mpd_ssize_t len, r; |
818 | | |
819 | 4 | if (result->len > 0 && result->digits > ctx->prec) { |
820 | 0 | _mpd_idiv_word(&len, &r, ctx->prec, MPD_RDIGITS); |
821 | 0 | len = (r == 0) ? len : len+1; |
822 | |
|
823 | 0 | if (r != 0) { |
824 | 0 | result->data[len-1] %= mpd_pow10[r]; |
825 | 0 | } |
826 | |
|
827 | 0 | len = _mpd_real_size(result->data, len); |
828 | | /* resize to fewer words cannot fail */ |
829 | 0 | mpd_qresize(result, len, &dummy); |
830 | 0 | result->len = len; |
831 | 0 | mpd_setdigits(result); |
832 | 0 | } |
833 | 4 | if (mpd_iszero(result)) { |
834 | 4 | _settriple(result, mpd_sign(result), 0, result->exp); |
835 | 4 | } |
836 | 4 | } |
837 | | |
838 | | /* |
839 | | * Cut off the most significant digits of a NaN payload so that the rest |
840 | | * fits in ctx->prec - ctx->clamp. Cannot fail. |
841 | | */ |
842 | | static void |
843 | | _mpd_fix_nan(mpd_t *result, const mpd_context_t *ctx) |
844 | 0 | { |
845 | 0 | uint32_t dummy; |
846 | 0 | mpd_ssize_t prec; |
847 | 0 | mpd_ssize_t len, r; |
848 | |
|
849 | 0 | prec = ctx->prec - ctx->clamp; |
850 | 0 | if (result->len > 0 && result->digits > prec) { |
851 | 0 | if (prec == 0) { |
852 | 0 | mpd_minalloc(result); |
853 | 0 | result->len = result->digits = 0; |
854 | 0 | } |
855 | 0 | else { |
856 | 0 | _mpd_idiv_word(&len, &r, prec, MPD_RDIGITS); |
857 | 0 | len = (r == 0) ? len : len+1; |
858 | |
|
859 | 0 | if (r != 0) { |
860 | 0 | result->data[len-1] %= mpd_pow10[r]; |
861 | 0 | } |
862 | |
|
863 | 0 | len = _mpd_real_size(result->data, len); |
864 | | /* resize to fewer words cannot fail */ |
865 | 0 | mpd_qresize(result, len, &dummy); |
866 | 0 | result->len = len; |
867 | 0 | mpd_setdigits(result); |
868 | 0 | if (mpd_iszerocoeff(result)) { |
869 | | /* NaN0 is not a valid representation */ |
870 | 0 | result->len = result->digits = 0; |
871 | 0 | } |
872 | 0 | } |
873 | 0 | } |
874 | 0 | } |
875 | | |
876 | | /* |
877 | | * Get n most significant digits from a decimal, where 0 < n <= MPD_UINT_DIGITS. |
878 | | * Assumes MPD_UINT_DIGITS == MPD_RDIGITS+1, which is true for 32 and 64 bit |
879 | | * machines. |
880 | | * |
881 | | * The result of the operation will be in lo. If the operation is impossible, |
882 | | * hi will be nonzero. This is used to indicate an error. |
883 | | */ |
884 | | static inline void |
885 | | _mpd_get_msdigits(mpd_uint_t *hi, mpd_uint_t *lo, const mpd_t *dec, |
886 | | unsigned int n) |
887 | 0 | { |
888 | 0 | mpd_uint_t r, tmp; |
889 | |
|
890 | 0 | assert(0 < n && n <= MPD_RDIGITS+1); |
891 | |
|
892 | 0 | _mpd_div_word(&tmp, &r, dec->digits, MPD_RDIGITS); |
893 | 0 | r = (r == 0) ? MPD_RDIGITS : r; /* digits in the most significant word */ |
894 | |
|
895 | 0 | *hi = 0; |
896 | 0 | *lo = dec->data[dec->len-1]; |
897 | 0 | if (n <= r) { |
898 | 0 | *lo /= mpd_pow10[r-n]; |
899 | 0 | } |
900 | 0 | else if (dec->len > 1) { |
901 | | /* at this point 1 <= r < n <= MPD_RDIGITS+1 */ |
902 | 0 | _mpd_mul_words(hi, lo, *lo, mpd_pow10[n-r]); |
903 | 0 | tmp = dec->data[dec->len-2] / mpd_pow10[MPD_RDIGITS-(n-r)]; |
904 | 0 | *lo = *lo + tmp; |
905 | 0 | if (*lo < tmp) (*hi)++; |
906 | 0 | } |
907 | 0 | } |
908 | | |
909 | | |
910 | | /******************************************************************************/ |
911 | | /* Gathering information about a decimal */ |
912 | | /******************************************************************************/ |
913 | | |
914 | | /* The real size of the coefficient without leading zero words. */ |
915 | | static inline mpd_ssize_t |
916 | | _mpd_real_size(mpd_uint_t *data, mpd_ssize_t size) |
917 | 11.8M | { |
918 | 14.8M | while (size > 1 && data[size-1] == 0) { |
919 | 2.94M | size--; |
920 | 2.94M | } |
921 | | |
922 | 11.8M | return size; |
923 | 11.8M | } |
924 | | |
925 | | /* Return number of trailing zeros. No errors are possible. */ |
926 | | mpd_ssize_t |
927 | | mpd_trail_zeros(const mpd_t *dec) |
928 | 10.8k | { |
929 | 10.8k | mpd_uint_t word; |
930 | 10.8k | mpd_ssize_t i, tz = 0; |
931 | | |
932 | 226k | for (i=0; i < dec->len; ++i) { |
933 | 226k | if (dec->data[i] != 0) { |
934 | 10.8k | word = dec->data[i]; |
935 | 10.8k | tz = i * MPD_RDIGITS; |
936 | 26.9k | while (word % 10 == 0) { |
937 | 16.1k | word /= 10; |
938 | 16.1k | tz++; |
939 | 16.1k | } |
940 | 10.8k | break; |
941 | 10.8k | } |
942 | 226k | } |
943 | | |
944 | 10.8k | return tz; |
945 | 10.8k | } |
946 | | |
947 | | /* Integer: Undefined for specials */ |
948 | | static int |
949 | | _mpd_isint(const mpd_t *dec) |
950 | 11.0k | { |
951 | 11.0k | mpd_ssize_t tz; |
952 | | |
953 | 11.0k | if (mpd_iszerocoeff(dec)) { |
954 | 218 | return 1; |
955 | 218 | } |
956 | | |
957 | 10.8k | tz = mpd_trail_zeros(dec); |
958 | 10.8k | return (dec->exp + tz >= 0); |
959 | 11.0k | } |
960 | | |
961 | | /* Integer */ |
962 | | int |
963 | | mpd_isinteger(const mpd_t *dec) |
964 | 0 | { |
965 | 0 | if (mpd_isspecial(dec)) { |
966 | 0 | return 0; |
967 | 0 | } |
968 | 0 | return _mpd_isint(dec); |
969 | 0 | } |
970 | | |
971 | | /* Word is a power of 10 */ |
972 | | static int |
973 | | mpd_word_ispow10(mpd_uint_t word) |
974 | 0 | { |
975 | 0 | int n; |
976 | |
|
977 | 0 | n = mpd_word_digits(word); |
978 | 0 | if (word == mpd_pow10[n-1]) { |
979 | 0 | return 1; |
980 | 0 | } |
981 | | |
982 | 0 | return 0; |
983 | 0 | } |
984 | | |
985 | | /* Coefficient is a power of 10 */ |
986 | | static int |
987 | | mpd_coeff_ispow10(const mpd_t *dec) |
988 | 0 | { |
989 | 0 | if (mpd_word_ispow10(mpd_msword(dec))) { |
990 | 0 | if (_mpd_isallzero(dec->data, dec->len-1)) { |
991 | 0 | return 1; |
992 | 0 | } |
993 | 0 | } |
994 | | |
995 | 0 | return 0; |
996 | 0 | } |
997 | | |
998 | | /* All digits of a word are nines */ |
999 | | static int |
1000 | | mpd_word_isallnine(mpd_uint_t word) |
1001 | 0 | { |
1002 | 0 | int n; |
1003 | |
|
1004 | 0 | n = mpd_word_digits(word); |
1005 | 0 | if (word == mpd_pow10[n]-1) { |
1006 | 0 | return 1; |
1007 | 0 | } |
1008 | | |
1009 | 0 | return 0; |
1010 | 0 | } |
1011 | | |
1012 | | /* All digits of the coefficient are nines */ |
1013 | | static int |
1014 | | mpd_coeff_isallnine(const mpd_t *dec) |
1015 | 0 | { |
1016 | 0 | if (mpd_word_isallnine(mpd_msword(dec))) { |
1017 | 0 | if (_mpd_isallnine(dec->data, dec->len-1)) { |
1018 | 0 | return 1; |
1019 | 0 | } |
1020 | 0 | } |
1021 | | |
1022 | 0 | return 0; |
1023 | 0 | } |
1024 | | |
1025 | | /* Odd decimal: Undefined for non-integers! */ |
1026 | | int |
1027 | | mpd_isodd(const mpd_t *dec) |
1028 | 1.84M | { |
1029 | 1.84M | mpd_uint_t q, r; |
1030 | 1.84M | assert(mpd_isinteger(dec)); |
1031 | 1.84M | if (mpd_iszerocoeff(dec)) return 0; |
1032 | 1.84M | if (dec->exp < 0) { |
1033 | 0 | _mpd_div_word(&q, &r, -dec->exp, MPD_RDIGITS); |
1034 | 0 | q = dec->data[q] / mpd_pow10[r]; |
1035 | 0 | return mpd_isoddword(q); |
1036 | 0 | } |
1037 | 1.84M | return dec->exp == 0 && mpd_isoddword(dec->data[0]); |
1038 | 1.84M | } |
1039 | | |
1040 | | /* Even: Undefined for non-integers! */ |
1041 | | int |
1042 | | mpd_iseven(const mpd_t *dec) |
1043 | 0 | { |
1044 | 0 | return !mpd_isodd(dec); |
1045 | 0 | } |
1046 | | |
1047 | | /******************************************************************************/ |
1048 | | /* Getting and setting decimals */ |
1049 | | /******************************************************************************/ |
1050 | | |
1051 | | /* Internal function: Set a static decimal from a triple, no error checking. */ |
1052 | | static void |
1053 | | _ssettriple(mpd_t *result, uint8_t sign, mpd_uint_t a, mpd_ssize_t exp) |
1054 | 0 | { |
1055 | 0 | mpd_set_flags(result, sign); |
1056 | 0 | result->exp = exp; |
1057 | 0 | _mpd_div_word(&result->data[1], &result->data[0], a, MPD_RADIX); |
1058 | 0 | result->len = (result->data[1] == 0) ? 1 : 2; |
1059 | 0 | mpd_setdigits(result); |
1060 | 0 | } |
1061 | | |
1062 | | /* Internal function: Set a decimal from a triple, no error checking. */ |
1063 | | static void |
1064 | | _settriple(mpd_t *result, uint8_t sign, mpd_uint_t a, mpd_ssize_t exp) |
1065 | 58.1k | { |
1066 | 58.1k | mpd_minalloc(result); |
1067 | 58.1k | mpd_set_flags(result, sign); |
1068 | 58.1k | result->exp = exp; |
1069 | 58.1k | _mpd_div_word(&result->data[1], &result->data[0], a, MPD_RADIX); |
1070 | 58.1k | result->len = (result->data[1] == 0) ? 1 : 2; |
1071 | 58.1k | mpd_setdigits(result); |
1072 | 58.1k | } |
1073 | | |
1074 | | /* Set a special number from a triple */ |
1075 | | void |
1076 | | mpd_setspecial(mpd_t *result, uint8_t sign, uint8_t type) |
1077 | 12 | { |
1078 | 12 | mpd_minalloc(result); |
1079 | 12 | result->flags &= ~(MPD_NEG|MPD_SPECIAL); |
1080 | 12 | result->flags |= (sign|type); |
1081 | 12 | result->exp = result->digits = result->len = 0; |
1082 | 12 | } |
1083 | | |
1084 | | /* Set result of NaN with an error status */ |
1085 | | void |
1086 | | mpd_seterror(mpd_t *result, uint32_t flags, uint32_t *status) |
1087 | 332 | { |
1088 | 332 | mpd_minalloc(result); |
1089 | 332 | mpd_set_qnan(result); |
1090 | 332 | mpd_set_positive(result); |
1091 | 332 | result->exp = result->digits = result->len = 0; |
1092 | 332 | *status |= flags; |
1093 | 332 | } |
1094 | | |
1095 | | /* quietly set a static decimal from an mpd_ssize_t */ |
1096 | | void |
1097 | | mpd_qsset_ssize(mpd_t *result, mpd_ssize_t a, const mpd_context_t *ctx, |
1098 | | uint32_t *status) |
1099 | 0 | { |
1100 | 0 | mpd_uint_t u; |
1101 | 0 | uint8_t sign = MPD_POS; |
1102 | |
|
1103 | 0 | if (a < 0) { |
1104 | 0 | if (a == MPD_SSIZE_MIN) { |
1105 | 0 | u = (mpd_uint_t)MPD_SSIZE_MAX + |
1106 | 0 | (-(MPD_SSIZE_MIN+MPD_SSIZE_MAX)); |
1107 | 0 | } |
1108 | 0 | else { |
1109 | 0 | u = -a; |
1110 | 0 | } |
1111 | 0 | sign = MPD_NEG; |
1112 | 0 | } |
1113 | 0 | else { |
1114 | 0 | u = a; |
1115 | 0 | } |
1116 | 0 | _ssettriple(result, sign, u, 0); |
1117 | 0 | mpd_qfinalize(result, ctx, status); |
1118 | 0 | } |
1119 | | |
1120 | | /* quietly set a static decimal from an mpd_uint_t */ |
1121 | | void |
1122 | | mpd_qsset_uint(mpd_t *result, mpd_uint_t a, const mpd_context_t *ctx, |
1123 | | uint32_t *status) |
1124 | 0 | { |
1125 | 0 | _ssettriple(result, MPD_POS, a, 0); |
1126 | 0 | mpd_qfinalize(result, ctx, status); |
1127 | 0 | } |
1128 | | |
1129 | | /* quietly set a static decimal from an int32_t */ |
1130 | | void |
1131 | | mpd_qsset_i32(mpd_t *result, int32_t a, const mpd_context_t *ctx, |
1132 | | uint32_t *status) |
1133 | 0 | { |
1134 | 0 | mpd_qsset_ssize(result, a, ctx, status); |
1135 | 0 | } |
1136 | | |
1137 | | /* quietly set a static decimal from a uint32_t */ |
1138 | | void |
1139 | | mpd_qsset_u32(mpd_t *result, uint32_t a, const mpd_context_t *ctx, |
1140 | | uint32_t *status) |
1141 | 0 | { |
1142 | 0 | mpd_qsset_uint(result, a, ctx, status); |
1143 | 0 | } |
1144 | | |
1145 | | #ifdef CONFIG_64 |
1146 | | /* quietly set a static decimal from an int64_t */ |
1147 | | void |
1148 | | mpd_qsset_i64(mpd_t *result, int64_t a, const mpd_context_t *ctx, |
1149 | | uint32_t *status) |
1150 | 0 | { |
1151 | 0 | mpd_qsset_ssize(result, a, ctx, status); |
1152 | 0 | } |
1153 | | |
1154 | | /* quietly set a static decimal from a uint64_t */ |
1155 | | void |
1156 | | mpd_qsset_u64(mpd_t *result, uint64_t a, const mpd_context_t *ctx, |
1157 | | uint32_t *status) |
1158 | 0 | { |
1159 | 0 | mpd_qsset_uint(result, a, ctx, status); |
1160 | 0 | } |
1161 | | #endif |
1162 | | |
1163 | | /* quietly set a decimal from an mpd_ssize_t */ |
1164 | | void |
1165 | | mpd_qset_ssize(mpd_t *result, mpd_ssize_t a, const mpd_context_t *ctx, |
1166 | | uint32_t *status) |
1167 | 0 | { |
1168 | 0 | mpd_minalloc(result); |
1169 | 0 | mpd_qsset_ssize(result, a, ctx, status); |
1170 | 0 | } |
1171 | | |
1172 | | /* quietly set a decimal from an mpd_uint_t */ |
1173 | | void |
1174 | | mpd_qset_uint(mpd_t *result, mpd_uint_t a, const mpd_context_t *ctx, |
1175 | | uint32_t *status) |
1176 | 0 | { |
1177 | 0 | _settriple(result, MPD_POS, a, 0); |
1178 | 0 | mpd_qfinalize(result, ctx, status); |
1179 | 0 | } |
1180 | | |
1181 | | /* quietly set a decimal from an int32_t */ |
1182 | | void |
1183 | | mpd_qset_i32(mpd_t *result, int32_t a, const mpd_context_t *ctx, |
1184 | | uint32_t *status) |
1185 | 0 | { |
1186 | 0 | mpd_qset_ssize(result, a, ctx, status); |
1187 | 0 | } |
1188 | | |
1189 | | /* quietly set a decimal from a uint32_t */ |
1190 | | void |
1191 | | mpd_qset_u32(mpd_t *result, uint32_t a, const mpd_context_t *ctx, |
1192 | | uint32_t *status) |
1193 | 0 | { |
1194 | 0 | mpd_qset_uint(result, a, ctx, status); |
1195 | 0 | } |
1196 | | |
1197 | | #if defined(CONFIG_32) && !defined(LEGACY_COMPILER) |
1198 | | /* set a decimal from a uint64_t */ |
1199 | | static void |
1200 | | _c32setu64(mpd_t *result, uint64_t u, uint8_t sign, uint32_t *status) |
1201 | | { |
1202 | | mpd_uint_t w[3]; |
1203 | | uint64_t q; |
1204 | | int i, len; |
1205 | | |
1206 | | len = 0; |
1207 | | do { |
1208 | | q = u / MPD_RADIX; |
1209 | | w[len] = (mpd_uint_t)(u - q * MPD_RADIX); |
1210 | | u = q; len++; |
1211 | | } while (u != 0); |
1212 | | |
1213 | | if (!mpd_qresize(result, len, status)) { |
1214 | | return; |
1215 | | } |
1216 | | for (i = 0; i < len; i++) { |
1217 | | result->data[i] = w[i]; |
1218 | | } |
1219 | | |
1220 | | mpd_set_flags(result, sign); |
1221 | | result->exp = 0; |
1222 | | result->len = len; |
1223 | | mpd_setdigits(result); |
1224 | | } |
1225 | | |
1226 | | static void |
1227 | | _c32_qset_u64(mpd_t *result, uint64_t a, const mpd_context_t *ctx, |
1228 | | uint32_t *status) |
1229 | | { |
1230 | | _c32setu64(result, a, MPD_POS, status); |
1231 | | mpd_qfinalize(result, ctx, status); |
1232 | | } |
1233 | | |
1234 | | /* set a decimal from an int64_t */ |
1235 | | static void |
1236 | | _c32_qset_i64(mpd_t *result, int64_t a, const mpd_context_t *ctx, |
1237 | | uint32_t *status) |
1238 | | { |
1239 | | uint64_t u; |
1240 | | uint8_t sign = MPD_POS; |
1241 | | |
1242 | | if (a < 0) { |
1243 | | if (a == INT64_MIN) { |
1244 | | u = (uint64_t)INT64_MAX + (-(INT64_MIN+INT64_MAX)); |
1245 | | } |
1246 | | else { |
1247 | | u = -a; |
1248 | | } |
1249 | | sign = MPD_NEG; |
1250 | | } |
1251 | | else { |
1252 | | u = a; |
1253 | | } |
1254 | | _c32setu64(result, u, sign, status); |
1255 | | mpd_qfinalize(result, ctx, status); |
1256 | | } |
1257 | | #endif /* CONFIG_32 && !LEGACY_COMPILER */ |
1258 | | |
1259 | | #ifndef LEGACY_COMPILER |
1260 | | /* quietly set a decimal from an int64_t */ |
1261 | | void |
1262 | | mpd_qset_i64(mpd_t *result, int64_t a, const mpd_context_t *ctx, |
1263 | | uint32_t *status) |
1264 | 0 | { |
1265 | 0 | #ifdef CONFIG_64 |
1266 | 0 | mpd_qset_ssize(result, a, ctx, status); |
1267 | | #else |
1268 | | _c32_qset_i64(result, a, ctx, status); |
1269 | | #endif |
1270 | 0 | } |
1271 | | |
1272 | | /* quietly set a decimal from an int64_t, use a maxcontext for conversion */ |
1273 | | void |
1274 | | mpd_qset_i64_exact(mpd_t *result, int64_t a, uint32_t *status) |
1275 | 0 | { |
1276 | 0 | mpd_context_t maxcontext; |
1277 | 0 | uint32_t workstatus = 0; |
1278 | |
|
1279 | 0 | mpd_maxcontext(&maxcontext); |
1280 | 0 | #ifdef CONFIG_64 |
1281 | 0 | mpd_qset_ssize(result, a, &maxcontext, &workstatus); |
1282 | | #else |
1283 | | _c32_qset_i64(result, a, &maxcontext, &workstatus); |
1284 | | #endif |
1285 | |
|
1286 | 0 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
1287 | | /* we want exact results */ |
1288 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); /* GCOV_NOT_REACHED */ |
1289 | 0 | } |
1290 | 0 | *status |= (workstatus&MPD_Errors); |
1291 | 0 | } |
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 | 0 | { |
1298 | 0 | #ifdef CONFIG_64 |
1299 | 0 | mpd_qset_uint(result, a, ctx, status); |
1300 | | #else |
1301 | | _c32_qset_u64(result, a, ctx, status); |
1302 | | #endif |
1303 | 0 | } |
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 | 0 | { |
1309 | 0 | mpd_context_t maxcontext; |
1310 | 0 | uint32_t workstatus = 0; |
1311 | |
|
1312 | 0 | mpd_maxcontext(&maxcontext); |
1313 | 0 | #ifdef CONFIG_64 |
1314 | 0 | mpd_qset_uint(result, a, &maxcontext, &workstatus); |
1315 | | #else |
1316 | | _c32_qset_u64(result, a, &maxcontext, &workstatus); |
1317 | | #endif |
1318 | |
|
1319 | 0 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
1320 | | /* we want exact results */ |
1321 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); /* GCOV_NOT_REACHED */ |
1322 | 0 | } |
1323 | 0 | *status |= (workstatus&MPD_Errors); |
1324 | 0 | } |
1325 | | #endif /* !LEGACY_COMPILER */ |
1326 | | |
1327 | | /* |
1328 | | * Quietly get an mpd_uint_t from a decimal. Assumes |
1329 | | * MPD_UINT_DIGITS == MPD_RDIGITS+1, which is true for |
1330 | | * 32 and 64 bit machines. |
1331 | | * |
1332 | | * If the operation is impossible, MPD_Invalid_operation is set. |
1333 | | */ |
1334 | | static mpd_uint_t |
1335 | | _mpd_qget_uint(int use_sign, const mpd_t *a, uint32_t *status) |
1336 | 0 | { |
1337 | 0 | mpd_t tmp; |
1338 | 0 | mpd_uint_t tmp_data[2]; |
1339 | 0 | mpd_uint_t lo, hi; |
1340 | |
|
1341 | 0 | if (mpd_isspecial(a)) { |
1342 | 0 | *status |= MPD_Invalid_operation; |
1343 | 0 | return MPD_UINT_MAX; |
1344 | 0 | } |
1345 | 0 | if (mpd_iszero(a)) { |
1346 | 0 | return 0; |
1347 | 0 | } |
1348 | 0 | if (use_sign && mpd_isnegative(a)) { |
1349 | 0 | *status |= MPD_Invalid_operation; |
1350 | 0 | return MPD_UINT_MAX; |
1351 | 0 | } |
1352 | | |
1353 | 0 | if (a->digits+a->exp > MPD_RDIGITS+1) { |
1354 | 0 | *status |= MPD_Invalid_operation; |
1355 | 0 | return MPD_UINT_MAX; |
1356 | 0 | } |
1357 | | |
1358 | 0 | if (a->exp < 0) { |
1359 | 0 | if (!_mpd_isint(a)) { |
1360 | 0 | *status |= MPD_Invalid_operation; |
1361 | 0 | return MPD_UINT_MAX; |
1362 | 0 | } |
1363 | | /* At this point a->digits+a->exp <= MPD_RDIGITS+1, |
1364 | | * so the shift fits. */ |
1365 | 0 | tmp.data = tmp_data; |
1366 | 0 | tmp.flags = MPD_STATIC|MPD_STATIC_DATA; |
1367 | 0 | tmp.alloc = 2; |
1368 | 0 | mpd_qsshiftr(&tmp, a, -a->exp); |
1369 | 0 | tmp.exp = 0; |
1370 | 0 | a = &tmp; |
1371 | 0 | } |
1372 | | |
1373 | 0 | _mpd_get_msdigits(&hi, &lo, a, MPD_RDIGITS+1); |
1374 | 0 | if (hi) { |
1375 | 0 | *status |= MPD_Invalid_operation; |
1376 | 0 | return MPD_UINT_MAX; |
1377 | 0 | } |
1378 | | |
1379 | 0 | if (a->exp > 0) { |
1380 | 0 | _mpd_mul_words(&hi, &lo, lo, mpd_pow10[a->exp]); |
1381 | 0 | if (hi) { |
1382 | 0 | *status |= MPD_Invalid_operation; |
1383 | 0 | return MPD_UINT_MAX; |
1384 | 0 | } |
1385 | 0 | } |
1386 | | |
1387 | 0 | return lo; |
1388 | 0 | } |
1389 | | |
1390 | | /* |
1391 | | * Sets Invalid_operation for: |
1392 | | * - specials |
1393 | | * - negative numbers (except negative zero) |
1394 | | * - non-integers |
1395 | | * - overflow |
1396 | | */ |
1397 | | mpd_uint_t |
1398 | | mpd_qget_uint(const mpd_t *a, uint32_t *status) |
1399 | 0 | { |
1400 | 0 | return _mpd_qget_uint(1, a, status); |
1401 | 0 | } |
1402 | | |
1403 | | /* Same as above, but gets the absolute value, i.e. the sign is ignored. */ |
1404 | | mpd_uint_t |
1405 | | mpd_qabs_uint(const mpd_t *a, uint32_t *status) |
1406 | 0 | { |
1407 | 0 | return _mpd_qget_uint(0, a, status); |
1408 | 0 | } |
1409 | | |
1410 | | /* quietly get an mpd_ssize_t from a decimal */ |
1411 | | mpd_ssize_t |
1412 | | mpd_qget_ssize(const mpd_t *a, uint32_t *status) |
1413 | 0 | { |
1414 | 0 | uint32_t workstatus = 0; |
1415 | 0 | mpd_uint_t u; |
1416 | 0 | int isneg; |
1417 | |
|
1418 | 0 | u = mpd_qabs_uint(a, &workstatus); |
1419 | 0 | if (workstatus&MPD_Invalid_operation) { |
1420 | 0 | *status |= workstatus; |
1421 | 0 | return MPD_SSIZE_MAX; |
1422 | 0 | } |
1423 | | |
1424 | 0 | isneg = mpd_isnegative(a); |
1425 | 0 | if (u <= MPD_SSIZE_MAX) { |
1426 | 0 | return isneg ? -((mpd_ssize_t)u) : (mpd_ssize_t)u; |
1427 | 0 | } |
1428 | 0 | else if (isneg && u+(mpd_uint_t)(MPD_SSIZE_MIN+MPD_SSIZE_MAX) == MPD_SSIZE_MAX) { |
1429 | 0 | return MPD_SSIZE_MIN; |
1430 | 0 | } |
1431 | | |
1432 | 0 | *status |= MPD_Invalid_operation; |
1433 | 0 | return MPD_SSIZE_MAX; |
1434 | 0 | } |
1435 | | |
1436 | | #if defined(CONFIG_32) && !defined(LEGACY_COMPILER) |
1437 | | /* |
1438 | | * Quietly get a uint64_t from a decimal. If the operation is impossible, |
1439 | | * MPD_Invalid_operation is set. |
1440 | | */ |
1441 | | static uint64_t |
1442 | | _c32_qget_u64(int use_sign, const mpd_t *a, uint32_t *status) |
1443 | | { |
1444 | | MPD_NEW_STATIC(tmp,0,0,20,3); |
1445 | | mpd_context_t maxcontext; |
1446 | | uint64_t ret; |
1447 | | |
1448 | | tmp_data[0] = 709551615; |
1449 | | tmp_data[1] = 446744073; |
1450 | | tmp_data[2] = 18; |
1451 | | |
1452 | | if (mpd_isspecial(a)) { |
1453 | | *status |= MPD_Invalid_operation; |
1454 | | return UINT64_MAX; |
1455 | | } |
1456 | | if (mpd_iszero(a)) { |
1457 | | return 0; |
1458 | | } |
1459 | | if (use_sign && mpd_isnegative(a)) { |
1460 | | *status |= MPD_Invalid_operation; |
1461 | | return UINT64_MAX; |
1462 | | } |
1463 | | if (!_mpd_isint(a)) { |
1464 | | *status |= MPD_Invalid_operation; |
1465 | | return UINT64_MAX; |
1466 | | } |
1467 | | |
1468 | | if (_mpd_cmp_abs(a, &tmp) > 0) { |
1469 | | *status |= MPD_Invalid_operation; |
1470 | | return UINT64_MAX; |
1471 | | } |
1472 | | |
1473 | | mpd_maxcontext(&maxcontext); |
1474 | | mpd_qrescale(&tmp, a, 0, &maxcontext, &maxcontext.status); |
1475 | | maxcontext.status &= ~MPD_Rounded; |
1476 | | if (maxcontext.status != 0) { |
1477 | | *status |= (maxcontext.status|MPD_Invalid_operation); /* GCOV_NOT_REACHED */ |
1478 | | return UINT64_MAX; /* GCOV_NOT_REACHED */ |
1479 | | } |
1480 | | |
1481 | | ret = 0; |
1482 | | switch (tmp.len) { |
1483 | | case 3: |
1484 | | ret += (uint64_t)tmp_data[2] * 1000000000000000000ULL; |
1485 | | case 2: |
1486 | | ret += (uint64_t)tmp_data[1] * 1000000000ULL; |
1487 | | case 1: |
1488 | | ret += tmp_data[0]; |
1489 | | break; |
1490 | | default: |
1491 | | abort(); /* GCOV_NOT_REACHED */ |
1492 | | } |
1493 | | |
1494 | | return ret; |
1495 | | } |
1496 | | |
1497 | | static int64_t |
1498 | | _c32_qget_i64(const mpd_t *a, uint32_t *status) |
1499 | | { |
1500 | | uint64_t u; |
1501 | | int isneg; |
1502 | | |
1503 | | u = _c32_qget_u64(0, a, status); |
1504 | | if (*status&MPD_Invalid_operation) { |
1505 | | return INT64_MAX; |
1506 | | } |
1507 | | |
1508 | | isneg = mpd_isnegative(a); |
1509 | | if (u <= INT64_MAX) { |
1510 | | return isneg ? -((int64_t)u) : (int64_t)u; |
1511 | | } |
1512 | | else if (isneg && u+(uint64_t)(INT64_MIN+INT64_MAX) == INT64_MAX) { |
1513 | | return INT64_MIN; |
1514 | | } |
1515 | | |
1516 | | *status |= MPD_Invalid_operation; |
1517 | | return INT64_MAX; |
1518 | | } |
1519 | | #endif /* CONFIG_32 && !LEGACY_COMPILER */ |
1520 | | |
1521 | | #ifdef CONFIG_64 |
1522 | | /* quietly get a uint64_t from a decimal */ |
1523 | | uint64_t |
1524 | | mpd_qget_u64(const mpd_t *a, uint32_t *status) |
1525 | 0 | { |
1526 | 0 | return mpd_qget_uint(a, status); |
1527 | 0 | } |
1528 | | |
1529 | | /* quietly get an int64_t from a decimal */ |
1530 | | int64_t |
1531 | | mpd_qget_i64(const mpd_t *a, uint32_t *status) |
1532 | 0 | { |
1533 | 0 | return mpd_qget_ssize(a, status); |
1534 | 0 | } |
1535 | | |
1536 | | /* quietly get a uint32_t from a decimal */ |
1537 | | uint32_t |
1538 | | mpd_qget_u32(const mpd_t *a, uint32_t *status) |
1539 | 0 | { |
1540 | 0 | uint32_t workstatus = 0; |
1541 | 0 | uint64_t x = mpd_qget_uint(a, &workstatus); |
1542 | |
|
1543 | 0 | if (workstatus&MPD_Invalid_operation) { |
1544 | 0 | *status |= workstatus; |
1545 | 0 | return UINT32_MAX; |
1546 | 0 | } |
1547 | 0 | if (x > UINT32_MAX) { |
1548 | 0 | *status |= MPD_Invalid_operation; |
1549 | 0 | return UINT32_MAX; |
1550 | 0 | } |
1551 | | |
1552 | 0 | return (uint32_t)x; |
1553 | 0 | } |
1554 | | |
1555 | | /* quietly get an int32_t from a decimal */ |
1556 | | int32_t |
1557 | | mpd_qget_i32(const mpd_t *a, uint32_t *status) |
1558 | 0 | { |
1559 | 0 | uint32_t workstatus = 0; |
1560 | 0 | int64_t x = mpd_qget_ssize(a, &workstatus); |
1561 | |
|
1562 | 0 | if (workstatus&MPD_Invalid_operation) { |
1563 | 0 | *status |= workstatus; |
1564 | 0 | return INT32_MAX; |
1565 | 0 | } |
1566 | 0 | if (x < INT32_MIN || x > INT32_MAX) { |
1567 | 0 | *status |= MPD_Invalid_operation; |
1568 | 0 | return INT32_MAX; |
1569 | 0 | } |
1570 | | |
1571 | 0 | return (int32_t)x; |
1572 | 0 | } |
1573 | | #else |
1574 | | #ifndef LEGACY_COMPILER |
1575 | | /* quietly get a uint64_t from a decimal */ |
1576 | | uint64_t |
1577 | | mpd_qget_u64(const mpd_t *a, uint32_t *status) |
1578 | | { |
1579 | | uint32_t workstatus = 0; |
1580 | | uint64_t x = _c32_qget_u64(1, a, &workstatus); |
1581 | | *status |= workstatus; |
1582 | | return x; |
1583 | | } |
1584 | | |
1585 | | /* quietly get an int64_t from a decimal */ |
1586 | | int64_t |
1587 | | mpd_qget_i64(const mpd_t *a, uint32_t *status) |
1588 | | { |
1589 | | uint32_t workstatus = 0; |
1590 | | int64_t x = _c32_qget_i64(a, &workstatus); |
1591 | | *status |= workstatus; |
1592 | | return x; |
1593 | | } |
1594 | | #endif |
1595 | | |
1596 | | /* quietly get a uint32_t from a decimal */ |
1597 | | uint32_t |
1598 | | mpd_qget_u32(const mpd_t *a, uint32_t *status) |
1599 | | { |
1600 | | return mpd_qget_uint(a, status); |
1601 | | } |
1602 | | |
1603 | | /* quietly get an int32_t from a decimal */ |
1604 | | int32_t |
1605 | | mpd_qget_i32(const mpd_t *a, uint32_t *status) |
1606 | | { |
1607 | | return mpd_qget_ssize(a, status); |
1608 | | } |
1609 | | #endif |
1610 | | |
1611 | | |
1612 | | /******************************************************************************/ |
1613 | | /* Filtering input of functions, finalizing output of functions */ |
1614 | | /******************************************************************************/ |
1615 | | |
1616 | | /* |
1617 | | * Check if the operand is NaN, copy to result and return 1 if this is |
1618 | | * the case. Copying can fail since NaNs are allowed to have a payload that |
1619 | | * does not fit in MPD_MINALLOC. |
1620 | | */ |
1621 | | int |
1622 | | mpd_qcheck_nan(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
1623 | | uint32_t *status) |
1624 | 0 | { |
1625 | 0 | if (mpd_isnan(a)) { |
1626 | 0 | *status |= mpd_issnan(a) ? MPD_Invalid_operation : 0; |
1627 | 0 | mpd_qcopy(result, a, status); |
1628 | 0 | mpd_set_qnan(result); |
1629 | 0 | _mpd_fix_nan(result, ctx); |
1630 | 0 | return 1; |
1631 | 0 | } |
1632 | 0 | return 0; |
1633 | 0 | } |
1634 | | |
1635 | | /* |
1636 | | * Check if either operand is NaN, copy to result and return 1 if this |
1637 | | * is the case. Copying can fail since NaNs are allowed to have a payload |
1638 | | * that does not fit in MPD_MINALLOC. |
1639 | | */ |
1640 | | int |
1641 | | mpd_qcheck_nans(mpd_t *result, const mpd_t *a, const mpd_t *b, |
1642 | | const mpd_context_t *ctx, uint32_t *status) |
1643 | 40 | { |
1644 | 40 | if ((a->flags|b->flags)&(MPD_NAN|MPD_SNAN)) { |
1645 | 0 | const mpd_t *choice = b; |
1646 | 0 | if (mpd_issnan(a)) { |
1647 | 0 | choice = a; |
1648 | 0 | *status |= MPD_Invalid_operation; |
1649 | 0 | } |
1650 | 0 | else if (mpd_issnan(b)) { |
1651 | 0 | *status |= MPD_Invalid_operation; |
1652 | 0 | } |
1653 | 0 | else if (mpd_isqnan(a)) { |
1654 | 0 | choice = a; |
1655 | 0 | } |
1656 | 0 | mpd_qcopy(result, choice, status); |
1657 | 0 | mpd_set_qnan(result); |
1658 | 0 | _mpd_fix_nan(result, ctx); |
1659 | 0 | return 1; |
1660 | 0 | } |
1661 | 40 | return 0; |
1662 | 40 | } |
1663 | | |
1664 | | /* |
1665 | | * Check if one of the operands is NaN, copy to result and return 1 if this |
1666 | | * is the case. Copying can fail since NaNs are allowed to have a payload |
1667 | | * that does not fit in MPD_MINALLOC. |
1668 | | */ |
1669 | | static int |
1670 | | mpd_qcheck_3nans(mpd_t *result, const mpd_t *a, const mpd_t *b, const mpd_t *c, |
1671 | | const mpd_context_t *ctx, uint32_t *status) |
1672 | 0 | { |
1673 | 0 | if ((a->flags|b->flags|c->flags)&(MPD_NAN|MPD_SNAN)) { |
1674 | 0 | const mpd_t *choice = c; |
1675 | 0 | if (mpd_issnan(a)) { |
1676 | 0 | choice = a; |
1677 | 0 | *status |= MPD_Invalid_operation; |
1678 | 0 | } |
1679 | 0 | else if (mpd_issnan(b)) { |
1680 | 0 | choice = b; |
1681 | 0 | *status |= MPD_Invalid_operation; |
1682 | 0 | } |
1683 | 0 | else if (mpd_issnan(c)) { |
1684 | 0 | *status |= MPD_Invalid_operation; |
1685 | 0 | } |
1686 | 0 | else if (mpd_isqnan(a)) { |
1687 | 0 | choice = a; |
1688 | 0 | } |
1689 | 0 | else if (mpd_isqnan(b)) { |
1690 | 0 | choice = b; |
1691 | 0 | } |
1692 | 0 | mpd_qcopy(result, choice, status); |
1693 | 0 | mpd_set_qnan(result); |
1694 | 0 | _mpd_fix_nan(result, ctx); |
1695 | 0 | return 1; |
1696 | 0 | } |
1697 | 0 | return 0; |
1698 | 0 | } |
1699 | | |
1700 | | /* Check if rounding digit 'rnd' leads to an increment. */ |
1701 | | static inline int |
1702 | | _mpd_rnd_incr(const mpd_t *dec, mpd_uint_t rnd, const mpd_context_t *ctx) |
1703 | 153 | { |
1704 | 153 | int ld; |
1705 | | |
1706 | 153 | switch (ctx->round) { |
1707 | 0 | case MPD_ROUND_DOWN: case MPD_ROUND_TRUNC: |
1708 | 0 | return 0; |
1709 | 153 | case MPD_ROUND_HALF_UP: |
1710 | 153 | return (rnd >= 5); |
1711 | 0 | case MPD_ROUND_HALF_EVEN: |
1712 | 0 | return (rnd > 5) || ((rnd == 5) && mpd_isoddcoeff(dec)); |
1713 | 0 | case MPD_ROUND_CEILING: |
1714 | 0 | return !(rnd == 0 || mpd_isnegative(dec)); |
1715 | 0 | case MPD_ROUND_FLOOR: |
1716 | 0 | return !(rnd == 0 || mpd_ispositive(dec)); |
1717 | 0 | case MPD_ROUND_HALF_DOWN: |
1718 | 0 | return (rnd > 5); |
1719 | 0 | case MPD_ROUND_UP: |
1720 | 0 | return !(rnd == 0); |
1721 | 0 | case MPD_ROUND_05UP: |
1722 | 0 | ld = (int)mpd_lsd(dec->data[0]); |
1723 | 0 | return (!(rnd == 0) && (ld == 0 || ld == 5)); |
1724 | 0 | default: |
1725 | | /* Without a valid context, further results will be undefined. */ |
1726 | 0 | return 0; /* GCOV_NOT_REACHED */ |
1727 | 153 | } |
1728 | 153 | } |
1729 | | |
1730 | | /* |
1731 | | * Apply rounding to a decimal that has been right-shifted into a full |
1732 | | * precision decimal. If an increment leads to an overflow of the precision, |
1733 | | * adjust the coefficient and the exponent and check the new exponent for |
1734 | | * overflow. |
1735 | | */ |
1736 | | static inline void |
1737 | | _mpd_apply_round(mpd_t *dec, mpd_uint_t rnd, const mpd_context_t *ctx, |
1738 | | uint32_t *status) |
1739 | 153 | { |
1740 | 153 | if (_mpd_rnd_incr(dec, rnd, ctx)) { |
1741 | | /* We have a number with exactly ctx->prec digits. The increment |
1742 | | * can only lead to an overflow if the decimal is all nines. In |
1743 | | * that case, the result is a power of ten with prec+1 digits. |
1744 | | * |
1745 | | * If the precision is a multiple of MPD_RDIGITS, this situation is |
1746 | | * detected by _mpd_baseincr returning a carry. |
1747 | | * If the precision is not a multiple of MPD_RDIGITS, we have to |
1748 | | * check if the result has one digit too many. |
1749 | | */ |
1750 | 71 | mpd_uint_t carry = _mpd_baseincr(dec->data, dec->len); |
1751 | 71 | if (carry) { |
1752 | 0 | dec->data[dec->len-1] = mpd_pow10[MPD_RDIGITS-1]; |
1753 | 0 | dec->exp += 1; |
1754 | 0 | _mpd_check_exp(dec, ctx, status); |
1755 | 0 | return; |
1756 | 0 | } |
1757 | 71 | mpd_setdigits(dec); |
1758 | 71 | if (dec->digits > ctx->prec) { |
1759 | 0 | mpd_qshiftr_inplace(dec, 1); |
1760 | 0 | dec->exp += 1; |
1761 | 0 | dec->digits = ctx->prec; |
1762 | 0 | _mpd_check_exp(dec, ctx, status); |
1763 | 0 | } |
1764 | 71 | } |
1765 | 153 | } |
1766 | | |
1767 | | /* |
1768 | | * Apply rounding to a decimal. Allow overflow of the precision. |
1769 | | */ |
1770 | | static inline void |
1771 | | _mpd_apply_round_excess(mpd_t *dec, mpd_uint_t rnd, const mpd_context_t *ctx, |
1772 | | uint32_t *status) |
1773 | 0 | { |
1774 | 0 | if (_mpd_rnd_incr(dec, rnd, ctx)) { |
1775 | 0 | mpd_uint_t carry = _mpd_baseincr(dec->data, dec->len); |
1776 | 0 | if (carry) { |
1777 | 0 | if (!mpd_qresize(dec, dec->len+1, status)) { |
1778 | 0 | return; |
1779 | 0 | } |
1780 | 0 | dec->data[dec->len] = 1; |
1781 | 0 | dec->len += 1; |
1782 | 0 | } |
1783 | 0 | mpd_setdigits(dec); |
1784 | 0 | } |
1785 | 0 | } |
1786 | | |
1787 | | /* |
1788 | | * Apply rounding to a decimal that has been right-shifted into a decimal |
1789 | | * with full precision or less. Return failure if an increment would |
1790 | | * overflow the precision. |
1791 | | */ |
1792 | | static inline int |
1793 | | _mpd_apply_round_fit(mpd_t *dec, mpd_uint_t rnd, const mpd_context_t *ctx, |
1794 | | uint32_t *status) |
1795 | 0 | { |
1796 | 0 | if (_mpd_rnd_incr(dec, rnd, ctx)) { |
1797 | 0 | mpd_uint_t carry = _mpd_baseincr(dec->data, dec->len); |
1798 | 0 | if (carry) { |
1799 | 0 | if (!mpd_qresize(dec, dec->len+1, status)) { |
1800 | 0 | return 0; |
1801 | 0 | } |
1802 | 0 | dec->data[dec->len] = 1; |
1803 | 0 | dec->len += 1; |
1804 | 0 | } |
1805 | 0 | mpd_setdigits(dec); |
1806 | 0 | if (dec->digits > ctx->prec) { |
1807 | 0 | mpd_seterror(dec, MPD_Invalid_operation, status); |
1808 | 0 | return 0; |
1809 | 0 | } |
1810 | 0 | } |
1811 | 0 | return 1; |
1812 | 0 | } |
1813 | | |
1814 | | /* Check a normal number for overflow, underflow, clamping. If the operand |
1815 | | is modified, it will be zero, special or (sub)normal with a coefficient |
1816 | | that fits into the current context precision. */ |
1817 | | static inline void |
1818 | | _mpd_check_exp(mpd_t *dec, const mpd_context_t *ctx, uint32_t *status) |
1819 | 7.45M | { |
1820 | 7.45M | mpd_ssize_t adjexp, etiny, shift; |
1821 | 7.45M | int rnd; |
1822 | | |
1823 | 7.45M | adjexp = mpd_adjexp(dec); |
1824 | 7.45M | if (adjexp > ctx->emax) { |
1825 | |
|
1826 | 0 | if (mpd_iszerocoeff(dec)) { |
1827 | 0 | dec->exp = ctx->emax; |
1828 | 0 | if (ctx->clamp) { |
1829 | 0 | dec->exp -= (ctx->prec-1); |
1830 | 0 | } |
1831 | 0 | mpd_zerocoeff(dec); |
1832 | 0 | *status |= MPD_Clamped; |
1833 | 0 | return; |
1834 | 0 | } |
1835 | | |
1836 | 0 | switch (ctx->round) { |
1837 | 0 | case MPD_ROUND_HALF_UP: case MPD_ROUND_HALF_EVEN: |
1838 | 0 | case MPD_ROUND_HALF_DOWN: case MPD_ROUND_UP: |
1839 | 0 | case MPD_ROUND_TRUNC: |
1840 | 0 | mpd_setspecial(dec, mpd_sign(dec), MPD_INF); |
1841 | 0 | break; |
1842 | 0 | case MPD_ROUND_DOWN: case MPD_ROUND_05UP: |
1843 | 0 | mpd_qmaxcoeff(dec, ctx, status); |
1844 | 0 | dec->exp = ctx->emax - ctx->prec + 1; |
1845 | 0 | break; |
1846 | 0 | case MPD_ROUND_CEILING: |
1847 | 0 | if (mpd_isnegative(dec)) { |
1848 | 0 | mpd_qmaxcoeff(dec, ctx, status); |
1849 | 0 | dec->exp = ctx->emax - ctx->prec + 1; |
1850 | 0 | } |
1851 | 0 | else { |
1852 | 0 | mpd_setspecial(dec, MPD_POS, MPD_INF); |
1853 | 0 | } |
1854 | 0 | break; |
1855 | 0 | case MPD_ROUND_FLOOR: |
1856 | 0 | if (mpd_ispositive(dec)) { |
1857 | 0 | mpd_qmaxcoeff(dec, ctx, status); |
1858 | 0 | dec->exp = ctx->emax - ctx->prec + 1; |
1859 | 0 | } |
1860 | 0 | else { |
1861 | 0 | mpd_setspecial(dec, MPD_NEG, MPD_INF); |
1862 | 0 | } |
1863 | 0 | break; |
1864 | 0 | default: /* debug */ |
1865 | 0 | abort(); /* GCOV_NOT_REACHED */ |
1866 | 0 | } |
1867 | | |
1868 | 0 | *status |= MPD_Overflow|MPD_Inexact|MPD_Rounded; |
1869 | |
|
1870 | 0 | } /* fold down */ |
1871 | 7.45M | else if (ctx->clamp && dec->exp > mpd_etop(ctx)) { |
1872 | | /* At this point adjexp=exp+digits-1 <= emax and exp > etop=emax-prec+1: |
1873 | | * (1) shift = exp -emax+prec-1 > 0 |
1874 | | * (2) digits+shift = exp+digits-1 - emax + prec <= prec */ |
1875 | 0 | shift = dec->exp - mpd_etop(ctx); |
1876 | 0 | if (!mpd_qshiftl(dec, dec, shift, status)) { |
1877 | 0 | return; |
1878 | 0 | } |
1879 | 0 | dec->exp -= shift; |
1880 | 0 | *status |= MPD_Clamped; |
1881 | 0 | if (!mpd_iszerocoeff(dec) && adjexp < ctx->emin) { |
1882 | | /* Underflow is impossible, since exp < etiny=emin-prec+1 |
1883 | | * and exp > etop=emax-prec+1 would imply emax < emin. */ |
1884 | 0 | *status |= MPD_Subnormal; |
1885 | 0 | } |
1886 | 0 | } |
1887 | 7.45M | else if (adjexp < ctx->emin) { |
1888 | |
|
1889 | 0 | etiny = mpd_etiny(ctx); |
1890 | |
|
1891 | 0 | if (mpd_iszerocoeff(dec)) { |
1892 | 0 | if (dec->exp < etiny) { |
1893 | 0 | dec->exp = etiny; |
1894 | 0 | mpd_zerocoeff(dec); |
1895 | 0 | *status |= MPD_Clamped; |
1896 | 0 | } |
1897 | 0 | return; |
1898 | 0 | } |
1899 | | |
1900 | 0 | *status |= MPD_Subnormal; |
1901 | 0 | if (dec->exp < etiny) { |
1902 | | /* At this point adjexp=exp+digits-1 < emin and exp < etiny=emin-prec+1: |
1903 | | * (1) shift = emin-prec+1 - exp > 0 |
1904 | | * (2) digits-shift = exp+digits-1 - emin + prec < prec */ |
1905 | 0 | shift = etiny - dec->exp; |
1906 | 0 | rnd = (int)mpd_qshiftr_inplace(dec, shift); |
1907 | 0 | dec->exp = etiny; |
1908 | | /* We always have a spare digit in case of an increment. */ |
1909 | 0 | _mpd_apply_round_excess(dec, rnd, ctx, status); |
1910 | 0 | *status |= MPD_Rounded; |
1911 | 0 | if (rnd) { |
1912 | 0 | *status |= (MPD_Inexact|MPD_Underflow); |
1913 | 0 | if (mpd_iszerocoeff(dec)) { |
1914 | 0 | mpd_zerocoeff(dec); |
1915 | 0 | *status |= MPD_Clamped; |
1916 | 0 | } |
1917 | 0 | } |
1918 | 0 | } |
1919 | | /* Case exp >= etiny=emin-prec+1: |
1920 | | * (1) adjexp=exp+digits-1 < emin |
1921 | | * (2) digits < emin-exp+1 <= prec */ |
1922 | 0 | } |
1923 | 7.45M | } |
1924 | | |
1925 | | /* Transcendental functions do not always set Underflow reliably, |
1926 | | * since they only use as much precision as is necessary for correct |
1927 | | * rounding. If a result like 1.0000000000e-101 is finalized, there |
1928 | | * is no rounding digit that would trigger Underflow. But we can |
1929 | | * assume Inexact, so a short check suffices. */ |
1930 | | static inline void |
1931 | | mpd_check_underflow(mpd_t *dec, const mpd_context_t *ctx, uint32_t *status) |
1932 | 0 | { |
1933 | 0 | if (mpd_adjexp(dec) < ctx->emin && !mpd_iszero(dec) && |
1934 | 0 | dec->exp < mpd_etiny(ctx)) { |
1935 | 0 | *status |= MPD_Underflow; |
1936 | 0 | } |
1937 | 0 | } |
1938 | | |
1939 | | /* Check if a normal number must be rounded after the exponent has been checked. */ |
1940 | | static inline void |
1941 | | _mpd_check_round(mpd_t *dec, const mpd_context_t *ctx, uint32_t *status) |
1942 | 7.45M | { |
1943 | 7.45M | mpd_uint_t rnd; |
1944 | 7.45M | mpd_ssize_t shift; |
1945 | | |
1946 | | /* must handle specials: _mpd_check_exp() can produce infinities or NaNs */ |
1947 | 7.45M | if (mpd_isspecial(dec)) { |
1948 | 0 | return; |
1949 | 0 | } |
1950 | | |
1951 | 7.45M | if (dec->digits > ctx->prec) { |
1952 | 153 | shift = dec->digits - ctx->prec; |
1953 | 153 | rnd = mpd_qshiftr_inplace(dec, shift); |
1954 | 153 | dec->exp += shift; |
1955 | 153 | _mpd_apply_round(dec, rnd, ctx, status); |
1956 | 153 | *status |= MPD_Rounded; |
1957 | 153 | if (rnd) { |
1958 | 153 | *status |= MPD_Inexact; |
1959 | 153 | } |
1960 | 153 | } |
1961 | 7.45M | } |
1962 | | |
1963 | | /* Finalize all operations. */ |
1964 | | void |
1965 | | mpd_qfinalize(mpd_t *result, const mpd_context_t *ctx, uint32_t *status) |
1966 | 7.45M | { |
1967 | 7.45M | if (mpd_isspecial(result)) { |
1968 | 0 | if (mpd_isnan(result)) { |
1969 | 0 | _mpd_fix_nan(result, ctx); |
1970 | 0 | } |
1971 | 0 | return; |
1972 | 0 | } |
1973 | | |
1974 | 7.45M | _mpd_check_exp(result, ctx, status); |
1975 | 7.45M | _mpd_check_round(result, ctx, status); |
1976 | 7.45M | } |
1977 | | |
1978 | | |
1979 | | /******************************************************************************/ |
1980 | | /* Copying */ |
1981 | | /******************************************************************************/ |
1982 | | |
1983 | | /* Internal function: Copy a decimal, share data with src: USE WITH CARE! */ |
1984 | | static inline void |
1985 | | _mpd_copy_shared(mpd_t *dest, const mpd_t *src) |
1986 | 0 | { |
1987 | 0 | dest->flags = src->flags; |
1988 | 0 | dest->exp = src->exp; |
1989 | 0 | dest->digits = src->digits; |
1990 | 0 | dest->len = src->len; |
1991 | 0 | dest->alloc = src->alloc; |
1992 | 0 | dest->data = src->data; |
1993 | |
|
1994 | 0 | mpd_set_shared_data(dest); |
1995 | 0 | } |
1996 | | |
1997 | | /* |
1998 | | * Copy a decimal. In case of an error, status is set to MPD_Malloc_error. |
1999 | | */ |
2000 | | int |
2001 | | mpd_qcopy(mpd_t *result, const mpd_t *a, uint32_t *status) |
2002 | 178k | { |
2003 | 178k | if (result == a) return 1; |
2004 | | |
2005 | 120k | if (!mpd_qresize(result, a->len, status)) { |
2006 | 0 | return 0; |
2007 | 0 | } |
2008 | | |
2009 | 120k | mpd_copy_flags(result, a); |
2010 | 120k | result->exp = a->exp; |
2011 | 120k | result->digits = a->digits; |
2012 | 120k | result->len = a->len; |
2013 | 120k | memcpy(result->data, a->data, a->len * (sizeof *result->data)); |
2014 | | |
2015 | 120k | return 1; |
2016 | 120k | } |
2017 | | |
2018 | | /* Same as mpd_qcopy, but do not set the result to NaN on failure. */ |
2019 | | int |
2020 | | mpd_qcopy_cxx(mpd_t *result, const mpd_t *a) |
2021 | 0 | { |
2022 | 0 | if (result == a) return 1; |
2023 | | |
2024 | 0 | if (!mpd_qresize_cxx(result, a->len)) { |
2025 | 0 | return 0; |
2026 | 0 | } |
2027 | | |
2028 | 0 | mpd_copy_flags(result, a); |
2029 | 0 | result->exp = a->exp; |
2030 | 0 | result->digits = a->digits; |
2031 | 0 | result->len = a->len; |
2032 | 0 | memcpy(result->data, a->data, a->len * (sizeof *result->data)); |
2033 | |
|
2034 | 0 | return 1; |
2035 | 0 | } |
2036 | | |
2037 | | /* |
2038 | | * Copy to a decimal with a static buffer. The caller has to make sure that |
2039 | | * the buffer is big enough. Cannot fail. |
2040 | | */ |
2041 | | static void |
2042 | | mpd_qcopy_static(mpd_t *result, const mpd_t *a) |
2043 | 0 | { |
2044 | 0 | if (result == a) return; |
2045 | | |
2046 | 0 | memcpy(result->data, a->data, a->len * (sizeof *result->data)); |
2047 | |
|
2048 | 0 | mpd_copy_flags(result, a); |
2049 | 0 | result->exp = a->exp; |
2050 | 0 | result->digits = a->digits; |
2051 | 0 | result->len = a->len; |
2052 | 0 | } |
2053 | | |
2054 | | /* |
2055 | | * Return a newly allocated copy of the operand. In case of an error, |
2056 | | * status is set to MPD_Malloc_error and the return value is NULL. |
2057 | | */ |
2058 | | mpd_t * |
2059 | | mpd_qncopy(const mpd_t *a) |
2060 | 0 | { |
2061 | 0 | mpd_t *result; |
2062 | |
|
2063 | 0 | if ((result = mpd_qnew_size(a->len)) == NULL) { |
2064 | 0 | return NULL; |
2065 | 0 | } |
2066 | 0 | memcpy(result->data, a->data, a->len * (sizeof *result->data)); |
2067 | 0 | mpd_copy_flags(result, a); |
2068 | 0 | result->exp = a->exp; |
2069 | 0 | result->digits = a->digits; |
2070 | 0 | result->len = a->len; |
2071 | |
|
2072 | 0 | return result; |
2073 | 0 | } |
2074 | | |
2075 | | /* |
2076 | | * Copy a decimal and set the sign to positive. In case of an error, the |
2077 | | * status is set to MPD_Malloc_error. |
2078 | | */ |
2079 | | int |
2080 | | mpd_qcopy_abs(mpd_t *result, const mpd_t *a, uint32_t *status) |
2081 | 6 | { |
2082 | 6 | if (!mpd_qcopy(result, a, status)) { |
2083 | 0 | return 0; |
2084 | 0 | } |
2085 | 6 | mpd_set_positive(result); |
2086 | 6 | return 1; |
2087 | 6 | } |
2088 | | |
2089 | | /* |
2090 | | * Copy a decimal and negate the sign. In case of an error, the |
2091 | | * status is set to MPD_Malloc_error. |
2092 | | */ |
2093 | | int |
2094 | | mpd_qcopy_negate(mpd_t *result, const mpd_t *a, uint32_t *status) |
2095 | 0 | { |
2096 | 0 | if (!mpd_qcopy(result, a, status)) { |
2097 | 0 | return 0; |
2098 | 0 | } |
2099 | 0 | _mpd_negate(result); |
2100 | 0 | return 1; |
2101 | 0 | } |
2102 | | |
2103 | | /* |
2104 | | * Copy a decimal, setting the sign of the first operand to the sign of the |
2105 | | * second operand. In case of an error, the status is set to MPD_Malloc_error. |
2106 | | */ |
2107 | | int |
2108 | | mpd_qcopy_sign(mpd_t *result, const mpd_t *a, const mpd_t *b, uint32_t *status) |
2109 | 0 | { |
2110 | 0 | uint8_t sign_b = mpd_sign(b); /* result may equal b! */ |
2111 | |
|
2112 | 0 | if (!mpd_qcopy(result, a, status)) { |
2113 | 0 | return 0; |
2114 | 0 | } |
2115 | 0 | mpd_set_sign(result, sign_b); |
2116 | 0 | return 1; |
2117 | 0 | } |
2118 | | |
2119 | | |
2120 | | /******************************************************************************/ |
2121 | | /* Comparisons */ |
2122 | | /******************************************************************************/ |
2123 | | |
2124 | | /* |
2125 | | * For all functions that compare two operands and return an int the usual |
2126 | | * convention applies to the return value: |
2127 | | * |
2128 | | * -1 if op1 < op2 |
2129 | | * 0 if op1 == op2 |
2130 | | * 1 if op1 > op2 |
2131 | | * |
2132 | | * INT_MAX for error |
2133 | | */ |
2134 | | |
2135 | | |
2136 | | /* Convenience macro. If a and b are not equal, return from the calling |
2137 | | * function with the correct comparison value. */ |
2138 | | #define CMP_EQUAL_OR_RETURN(a, b) \ |
2139 | 74 | if (a != b) { \ |
2140 | 20 | if (a < b) { \ |
2141 | 6 | return -1; \ |
2142 | 6 | } \ |
2143 | 20 | return 1; \ |
2144 | 20 | } |
2145 | | |
2146 | | /* |
2147 | | * Compare the data of big and small. This function does the equivalent |
2148 | | * of first shifting small to the left and then comparing the data of |
2149 | | * big and small, except that no allocation for the left shift is needed. |
2150 | | */ |
2151 | | static int |
2152 | | _mpd_basecmp(mpd_uint_t *big, mpd_uint_t *small, mpd_size_t n, mpd_size_t m, |
2153 | | mpd_size_t shift) |
2154 | 0 | { |
2155 | | #if defined(__GNUC__) && !defined(__INTEL_COMPILER) && !defined(__clang__) |
2156 | | /* spurious uninitialized warnings */ |
2157 | | mpd_uint_t l=l, lprev=lprev, h=h; |
2158 | | #else |
2159 | 0 | mpd_uint_t l, lprev, h; |
2160 | 0 | #endif |
2161 | 0 | mpd_uint_t q, r; |
2162 | 0 | mpd_uint_t ph, x; |
2163 | |
|
2164 | 0 | assert(m > 0 && n >= m && shift > 0); |
2165 | |
|
2166 | 0 | _mpd_div_word(&q, &r, (mpd_uint_t)shift, MPD_RDIGITS); |
2167 | |
|
2168 | 0 | if (r != 0) { |
2169 | |
|
2170 | 0 | ph = mpd_pow10[r]; |
2171 | |
|
2172 | 0 | --m; --n; |
2173 | 0 | _mpd_divmod_pow10(&h, &lprev, small[m--], MPD_RDIGITS-r); |
2174 | 0 | if (h != 0) { |
2175 | 0 | CMP_EQUAL_OR_RETURN(big[n], h) |
2176 | 0 | --n; |
2177 | 0 | } |
2178 | 0 | for (; m != MPD_SIZE_MAX; m--,n--) { |
2179 | 0 | _mpd_divmod_pow10(&h, &l, small[m], MPD_RDIGITS-r); |
2180 | 0 | x = ph * lprev + h; |
2181 | 0 | CMP_EQUAL_OR_RETURN(big[n], x) |
2182 | 0 | lprev = l; |
2183 | 0 | } |
2184 | 0 | x = ph * lprev; |
2185 | 0 | CMP_EQUAL_OR_RETURN(big[q], x) |
2186 | 0 | } |
2187 | 0 | else { |
2188 | 0 | while (--m != MPD_SIZE_MAX) { |
2189 | 0 | CMP_EQUAL_OR_RETURN(big[m+q], small[m]) |
2190 | 0 | } |
2191 | 0 | } |
2192 | | |
2193 | 0 | return !_mpd_isallzero(big, q); |
2194 | 0 | } |
2195 | | |
2196 | | /* Compare two decimals with the same adjusted exponent. */ |
2197 | | static int |
2198 | | _mpd_cmp_same_adjexp(const mpd_t *a, const mpd_t *b) |
2199 | 37 | { |
2200 | 37 | mpd_ssize_t shift, i; |
2201 | | |
2202 | 37 | if (a->exp != b->exp) { |
2203 | | /* Cannot wrap: a->exp + a->digits = b->exp + b->digits, so |
2204 | | * a->exp - b->exp = b->digits - a->digits. */ |
2205 | 0 | shift = a->exp - b->exp; |
2206 | 0 | if (shift > 0) { |
2207 | 0 | return -1 * _mpd_basecmp(b->data, a->data, b->len, a->len, shift); |
2208 | 0 | } |
2209 | 0 | else { |
2210 | 0 | return _mpd_basecmp(a->data, b->data, a->len, b->len, -shift); |
2211 | 0 | } |
2212 | 0 | } |
2213 | | |
2214 | | /* |
2215 | | * At this point adjexp(a) == adjexp(b) and a->exp == b->exp, |
2216 | | * so a->digits == b->digits, therefore a->len == b->len. |
2217 | | */ |
2218 | 91 | for (i = a->len-1; i >= 0; --i) { |
2219 | 74 | CMP_EQUAL_OR_RETURN(a->data[i], b->data[i]) |
2220 | 74 | } |
2221 | | |
2222 | 17 | return 0; |
2223 | 37 | } |
2224 | | |
2225 | | /* Compare two numerical values. */ |
2226 | | static int |
2227 | | _mpd_cmp(const mpd_t *a, const mpd_t *b) |
2228 | 131 | { |
2229 | 131 | mpd_ssize_t adjexp_a, adjexp_b; |
2230 | | |
2231 | | /* equal pointers */ |
2232 | 131 | if (a == b) { |
2233 | 0 | return 0; |
2234 | 0 | } |
2235 | | |
2236 | | /* infinities */ |
2237 | 131 | if (mpd_isinfinite(a)) { |
2238 | 0 | if (mpd_isinfinite(b)) { |
2239 | 0 | return mpd_isnegative(b) - mpd_isnegative(a); |
2240 | 0 | } |
2241 | 0 | return mpd_arith_sign(a); |
2242 | 0 | } |
2243 | 131 | if (mpd_isinfinite(b)) { |
2244 | 0 | return -mpd_arith_sign(b); |
2245 | 0 | } |
2246 | | |
2247 | | /* zeros */ |
2248 | 131 | if (mpd_iszerocoeff(a)) { |
2249 | 23 | if (mpd_iszerocoeff(b)) { |
2250 | 18 | return 0; |
2251 | 18 | } |
2252 | 5 | return -mpd_arith_sign(b); |
2253 | 23 | } |
2254 | 108 | if (mpd_iszerocoeff(b)) { |
2255 | 15 | return mpd_arith_sign(a); |
2256 | 15 | } |
2257 | | |
2258 | | /* different signs */ |
2259 | 93 | if (mpd_sign(a) != mpd_sign(b)) { |
2260 | 0 | return mpd_sign(b) - mpd_sign(a); |
2261 | 0 | } |
2262 | | |
2263 | | /* different adjusted exponents */ |
2264 | 93 | adjexp_a = mpd_adjexp(a); |
2265 | 93 | adjexp_b = mpd_adjexp(b); |
2266 | 93 | if (adjexp_a != adjexp_b) { |
2267 | 69 | if (adjexp_a < adjexp_b) { |
2268 | 31 | return -1 * mpd_arith_sign(a); |
2269 | 31 | } |
2270 | 38 | return mpd_arith_sign(a); |
2271 | 69 | } |
2272 | | |
2273 | | /* same adjusted exponents */ |
2274 | 24 | return _mpd_cmp_same_adjexp(a, b) * mpd_arith_sign(a); |
2275 | 93 | } |
2276 | | |
2277 | | /* Compare the absolutes of two numerical values. */ |
2278 | | static int |
2279 | | _mpd_cmp_abs(const mpd_t *a, const mpd_t *b) |
2280 | 76 | { |
2281 | 76 | mpd_ssize_t adjexp_a, adjexp_b; |
2282 | | |
2283 | | /* equal pointers */ |
2284 | 76 | if (a == b) { |
2285 | 0 | return 0; |
2286 | 0 | } |
2287 | | |
2288 | | /* infinities */ |
2289 | 76 | if (mpd_isinfinite(a)) { |
2290 | 0 | if (mpd_isinfinite(b)) { |
2291 | 0 | return 0; |
2292 | 0 | } |
2293 | 0 | return 1; |
2294 | 0 | } |
2295 | 76 | if (mpd_isinfinite(b)) { |
2296 | 0 | return -1; |
2297 | 0 | } |
2298 | | |
2299 | | /* zeros */ |
2300 | 76 | if (mpd_iszerocoeff(a)) { |
2301 | 0 | if (mpd_iszerocoeff(b)) { |
2302 | 0 | return 0; |
2303 | 0 | } |
2304 | 0 | return -1; |
2305 | 0 | } |
2306 | 76 | if (mpd_iszerocoeff(b)) { |
2307 | 0 | return 1; |
2308 | 0 | } |
2309 | | |
2310 | | /* different adjusted exponents */ |
2311 | 76 | adjexp_a = mpd_adjexp(a); |
2312 | 76 | adjexp_b = mpd_adjexp(b); |
2313 | 76 | if (adjexp_a != adjexp_b) { |
2314 | 63 | if (adjexp_a < adjexp_b) { |
2315 | 0 | return -1; |
2316 | 0 | } |
2317 | 63 | return 1; |
2318 | 63 | } |
2319 | | |
2320 | | /* same adjusted exponents */ |
2321 | 13 | return _mpd_cmp_same_adjexp(a, b); |
2322 | 76 | } |
2323 | | |
2324 | | /* Compare two values and return an integer result. */ |
2325 | | int |
2326 | | mpd_qcmp(const mpd_t *a, const mpd_t *b, uint32_t *status) |
2327 | 0 | { |
2328 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
2329 | 0 | if (mpd_isnan(a) || mpd_isnan(b)) { |
2330 | 0 | *status |= MPD_Invalid_operation; |
2331 | 0 | return INT_MAX; |
2332 | 0 | } |
2333 | 0 | } |
2334 | | |
2335 | 0 | return _mpd_cmp(a, b); |
2336 | 0 | } |
2337 | | |
2338 | | /* |
2339 | | * Compare a and b, convert the usual integer result to a decimal and |
2340 | | * store it in 'result'. For convenience, the integer result of the comparison |
2341 | | * is returned. Comparisons involving NaNs return NaN/INT_MAX. |
2342 | | */ |
2343 | | int |
2344 | | mpd_qcompare(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2345 | | const mpd_context_t *ctx, uint32_t *status) |
2346 | 91 | { |
2347 | 91 | int c; |
2348 | | |
2349 | 91 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
2350 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
2351 | 0 | return INT_MAX; |
2352 | 0 | } |
2353 | 0 | } |
2354 | | |
2355 | 91 | c = _mpd_cmp(a, b); |
2356 | 91 | _settriple(result, (c < 0), (c != 0), 0); |
2357 | 91 | return c; |
2358 | 91 | } |
2359 | | |
2360 | | /* Same as mpd_compare(), but signal for all NaNs, i.e. also for quiet NaNs. */ |
2361 | | int |
2362 | | mpd_qcompare_signal(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2363 | | const mpd_context_t *ctx, uint32_t *status) |
2364 | 0 | { |
2365 | 0 | int c; |
2366 | |
|
2367 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
2368 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
2369 | 0 | *status |= MPD_Invalid_operation; |
2370 | 0 | return INT_MAX; |
2371 | 0 | } |
2372 | 0 | } |
2373 | | |
2374 | 0 | c = _mpd_cmp(a, b); |
2375 | 0 | _settriple(result, (c < 0), (c != 0), 0); |
2376 | 0 | return c; |
2377 | 0 | } |
2378 | | |
2379 | | /* Compare the operands using a total order. */ |
2380 | | int |
2381 | | mpd_cmp_total(const mpd_t *a, const mpd_t *b) |
2382 | 0 | { |
2383 | 0 | mpd_t aa, bb; |
2384 | 0 | int nan_a, nan_b; |
2385 | 0 | int c; |
2386 | |
|
2387 | 0 | if (mpd_sign(a) != mpd_sign(b)) { |
2388 | 0 | return mpd_sign(b) - mpd_sign(a); |
2389 | 0 | } |
2390 | | |
2391 | | |
2392 | 0 | if (mpd_isnan(a)) { |
2393 | 0 | c = 1; |
2394 | 0 | if (mpd_isnan(b)) { |
2395 | 0 | nan_a = (mpd_isqnan(a)) ? 1 : 0; |
2396 | 0 | nan_b = (mpd_isqnan(b)) ? 1 : 0; |
2397 | 0 | if (nan_b == nan_a) { |
2398 | 0 | if (a->len > 0 && b->len > 0) { |
2399 | 0 | _mpd_copy_shared(&aa, a); |
2400 | 0 | _mpd_copy_shared(&bb, b); |
2401 | 0 | aa.exp = bb.exp = 0; |
2402 | | /* compare payload */ |
2403 | 0 | c = _mpd_cmp_abs(&aa, &bb); |
2404 | 0 | } |
2405 | 0 | else { |
2406 | 0 | c = (a->len > 0) - (b->len > 0); |
2407 | 0 | } |
2408 | 0 | } |
2409 | 0 | else { |
2410 | 0 | c = nan_a - nan_b; |
2411 | 0 | } |
2412 | 0 | } |
2413 | 0 | } |
2414 | 0 | else if (mpd_isnan(b)) { |
2415 | 0 | c = -1; |
2416 | 0 | } |
2417 | 0 | else { |
2418 | 0 | c = _mpd_cmp_abs(a, b); |
2419 | 0 | if (c == 0 && a->exp != b->exp) { |
2420 | 0 | c = (a->exp < b->exp) ? -1 : 1; |
2421 | 0 | } |
2422 | 0 | } |
2423 | |
|
2424 | 0 | return c * mpd_arith_sign(a); |
2425 | 0 | } |
2426 | | |
2427 | | /* |
2428 | | * Compare a and b according to a total order, convert the usual integer result |
2429 | | * to a decimal and store it in 'result'. For convenience, the integer result |
2430 | | * of the comparison is returned. |
2431 | | */ |
2432 | | int |
2433 | | mpd_compare_total(mpd_t *result, const mpd_t *a, const mpd_t *b) |
2434 | 0 | { |
2435 | 0 | int c; |
2436 | |
|
2437 | 0 | c = mpd_cmp_total(a, b); |
2438 | 0 | _settriple(result, (c < 0), (c != 0), 0); |
2439 | 0 | return c; |
2440 | 0 | } |
2441 | | |
2442 | | /* Compare the magnitude of the operands using a total order. */ |
2443 | | int |
2444 | | mpd_cmp_total_mag(const mpd_t *a, const mpd_t *b) |
2445 | 0 | { |
2446 | 0 | mpd_t aa, bb; |
2447 | |
|
2448 | 0 | _mpd_copy_shared(&aa, a); |
2449 | 0 | _mpd_copy_shared(&bb, b); |
2450 | |
|
2451 | 0 | mpd_set_positive(&aa); |
2452 | 0 | mpd_set_positive(&bb); |
2453 | |
|
2454 | 0 | return mpd_cmp_total(&aa, &bb); |
2455 | 0 | } |
2456 | | |
2457 | | /* |
2458 | | * Compare the magnitude of a and b according to a total order, convert the |
2459 | | * the usual integer result to a decimal and store it in 'result'. |
2460 | | * For convenience, the integer result of the comparison is returned. |
2461 | | */ |
2462 | | int |
2463 | | mpd_compare_total_mag(mpd_t *result, const mpd_t *a, const mpd_t *b) |
2464 | 0 | { |
2465 | 0 | int c; |
2466 | |
|
2467 | 0 | c = mpd_cmp_total_mag(a, b); |
2468 | 0 | _settriple(result, (c < 0), (c != 0), 0); |
2469 | 0 | return c; |
2470 | 0 | } |
2471 | | |
2472 | | /* Determine an ordering for operands that are numerically equal. */ |
2473 | | static inline int |
2474 | | _mpd_cmp_numequal(const mpd_t *a, const mpd_t *b) |
2475 | 3 | { |
2476 | 3 | int sign_a, sign_b; |
2477 | 3 | int c; |
2478 | | |
2479 | 3 | sign_a = mpd_sign(a); |
2480 | 3 | sign_b = mpd_sign(b); |
2481 | 3 | if (sign_a != sign_b) { |
2482 | 0 | c = sign_b - sign_a; |
2483 | 0 | } |
2484 | 3 | else { |
2485 | 3 | c = (a->exp < b->exp) ? -1 : 1; |
2486 | 3 | c *= mpd_arith_sign(a); |
2487 | 3 | } |
2488 | | |
2489 | 3 | return c; |
2490 | 3 | } |
2491 | | |
2492 | | |
2493 | | /******************************************************************************/ |
2494 | | /* Shifting the coefficient */ |
2495 | | /******************************************************************************/ |
2496 | | |
2497 | | /* |
2498 | | * Shift the coefficient of the operand to the left, no check for specials. |
2499 | | * Both operands may be the same pointer. If the result length has to be |
2500 | | * increased, mpd_qresize() might fail with MPD_Malloc_error. |
2501 | | */ |
2502 | | int |
2503 | | mpd_qshiftl(mpd_t *result, const mpd_t *a, mpd_ssize_t n, uint32_t *status) |
2504 | 7.20k | { |
2505 | 7.20k | mpd_ssize_t size; |
2506 | | |
2507 | 7.20k | assert(!mpd_isspecial(a)); |
2508 | 7.20k | assert(n >= 0); |
2509 | | |
2510 | 7.20k | if (mpd_iszerocoeff(a) || n == 0) { |
2511 | 7.01k | return mpd_qcopy(result, a, status); |
2512 | 7.01k | } |
2513 | | |
2514 | 194 | size = mpd_digits_to_size(a->digits+n); |
2515 | 194 | if (!mpd_qresize(result, size, status)) { |
2516 | 0 | return 0; /* result is NaN */ |
2517 | 0 | } |
2518 | | |
2519 | 194 | _mpd_baseshiftl(result->data, a->data, size, a->len, n); |
2520 | | |
2521 | 194 | mpd_copy_flags(result, a); |
2522 | 194 | result->exp = a->exp; |
2523 | 194 | result->digits = a->digits+n; |
2524 | 194 | result->len = size; |
2525 | | |
2526 | 194 | return 1; |
2527 | 194 | } |
2528 | | |
2529 | | /* Determine the rounding indicator if all digits of the coefficient are shifted |
2530 | | * out of the picture. */ |
2531 | | static mpd_uint_t |
2532 | | _mpd_get_rnd(const mpd_uint_t *data, mpd_ssize_t len, int use_msd) |
2533 | 0 | { |
2534 | 0 | mpd_uint_t rnd = 0, rest = 0, word; |
2535 | |
|
2536 | 0 | word = data[len-1]; |
2537 | | /* special treatment for the most significant digit if shift == digits */ |
2538 | 0 | if (use_msd) { |
2539 | 0 | _mpd_divmod_pow10(&rnd, &rest, word, mpd_word_digits(word)-1); |
2540 | 0 | if (len > 1 && rest == 0) { |
2541 | 0 | rest = !_mpd_isallzero(data, len-1); |
2542 | 0 | } |
2543 | 0 | } |
2544 | 0 | else { |
2545 | 0 | rest = !_mpd_isallzero(data, len); |
2546 | 0 | } |
2547 | |
|
2548 | 0 | return (rnd == 0 || rnd == 5) ? rnd + !!rest : rnd; |
2549 | 0 | } |
2550 | | |
2551 | | /* |
2552 | | * Same as mpd_qshiftr(), but 'result' is an mpd_t with a static coefficient. |
2553 | | * It is the caller's responsibility to ensure that the coefficient is big |
2554 | | * enough. The function cannot fail. |
2555 | | */ |
2556 | | static mpd_uint_t |
2557 | | mpd_qsshiftr(mpd_t *result, const mpd_t *a, mpd_ssize_t n) |
2558 | 0 | { |
2559 | 0 | mpd_uint_t rnd; |
2560 | 0 | mpd_ssize_t size; |
2561 | |
|
2562 | 0 | assert(!mpd_isspecial(a)); |
2563 | 0 | assert(n >= 0); |
2564 | |
|
2565 | 0 | if (mpd_iszerocoeff(a) || n == 0) { |
2566 | 0 | mpd_qcopy_static(result, a); |
2567 | 0 | return 0; |
2568 | 0 | } |
2569 | | |
2570 | 0 | if (n >= a->digits) { |
2571 | 0 | rnd = _mpd_get_rnd(a->data, a->len, (n==a->digits)); |
2572 | 0 | mpd_zerocoeff(result); |
2573 | 0 | } |
2574 | 0 | else { |
2575 | 0 | result->digits = a->digits-n; |
2576 | 0 | size = mpd_digits_to_size(result->digits); |
2577 | 0 | rnd = _mpd_baseshiftr(result->data, a->data, a->len, n); |
2578 | 0 | result->len = size; |
2579 | 0 | } |
2580 | |
|
2581 | 0 | mpd_copy_flags(result, a); |
2582 | 0 | result->exp = a->exp; |
2583 | |
|
2584 | 0 | return rnd; |
2585 | 0 | } |
2586 | | |
2587 | | /* |
2588 | | * Inplace shift of the coefficient to the right, no check for specials. |
2589 | | * Returns the rounding indicator for mpd_rnd_incr(). |
2590 | | * The function cannot fail. |
2591 | | */ |
2592 | | mpd_uint_t |
2593 | | mpd_qshiftr_inplace(mpd_t *result, mpd_ssize_t n) |
2594 | 194 | { |
2595 | 194 | uint32_t dummy; |
2596 | 194 | mpd_uint_t rnd; |
2597 | 194 | mpd_ssize_t size; |
2598 | | |
2599 | 194 | assert(!mpd_isspecial(result)); |
2600 | 194 | assert(n >= 0); |
2601 | | |
2602 | 194 | if (mpd_iszerocoeff(result) || n == 0) { |
2603 | 0 | return 0; |
2604 | 0 | } |
2605 | | |
2606 | 194 | if (n >= result->digits) { |
2607 | 0 | rnd = _mpd_get_rnd(result->data, result->len, (n==result->digits)); |
2608 | 0 | mpd_zerocoeff(result); |
2609 | 0 | } |
2610 | 194 | else { |
2611 | 194 | rnd = _mpd_baseshiftr(result->data, result->data, result->len, n); |
2612 | 194 | result->digits -= n; |
2613 | 194 | size = mpd_digits_to_size(result->digits); |
2614 | | /* reducing the size cannot fail */ |
2615 | 194 | mpd_qresize(result, size, &dummy); |
2616 | 194 | result->len = size; |
2617 | 194 | } |
2618 | | |
2619 | 194 | return rnd; |
2620 | 194 | } |
2621 | | |
2622 | | /* |
2623 | | * Shift the coefficient of the operand to the right, no check for specials. |
2624 | | * Both operands may be the same pointer. Returns the rounding indicator to |
2625 | | * be used by mpd_rnd_incr(). If the result length has to be increased, |
2626 | | * mpd_qcopy() or mpd_qresize() might fail with MPD_Malloc_error. In those |
2627 | | * cases, MPD_UINT_MAX is returned. |
2628 | | */ |
2629 | | mpd_uint_t |
2630 | | mpd_qshiftr(mpd_t *result, const mpd_t *a, mpd_ssize_t n, uint32_t *status) |
2631 | 0 | { |
2632 | 0 | mpd_uint_t rnd; |
2633 | 0 | mpd_ssize_t size; |
2634 | |
|
2635 | 0 | assert(!mpd_isspecial(a)); |
2636 | 0 | assert(n >= 0); |
2637 | |
|
2638 | 0 | if (mpd_iszerocoeff(a) || n == 0) { |
2639 | 0 | if (!mpd_qcopy(result, a, status)) { |
2640 | 0 | return MPD_UINT_MAX; |
2641 | 0 | } |
2642 | 0 | return 0; |
2643 | 0 | } |
2644 | | |
2645 | 0 | if (n >= a->digits) { |
2646 | 0 | rnd = _mpd_get_rnd(a->data, a->len, (n==a->digits)); |
2647 | 0 | mpd_zerocoeff(result); |
2648 | 0 | } |
2649 | 0 | else { |
2650 | 0 | result->digits = a->digits-n; |
2651 | 0 | size = mpd_digits_to_size(result->digits); |
2652 | 0 | if (result == a) { |
2653 | 0 | rnd = _mpd_baseshiftr(result->data, a->data, a->len, n); |
2654 | | /* reducing the size cannot fail */ |
2655 | 0 | mpd_qresize(result, size, status); |
2656 | 0 | } |
2657 | 0 | else { |
2658 | 0 | if (!mpd_qresize(result, size, status)) { |
2659 | 0 | return MPD_UINT_MAX; |
2660 | 0 | } |
2661 | 0 | rnd = _mpd_baseshiftr(result->data, a->data, a->len, n); |
2662 | 0 | } |
2663 | 0 | result->len = size; |
2664 | 0 | } |
2665 | | |
2666 | 0 | mpd_copy_flags(result, a); |
2667 | 0 | result->exp = a->exp; |
2668 | |
|
2669 | 0 | return rnd; |
2670 | 0 | } |
2671 | | |
2672 | | |
2673 | | /******************************************************************************/ |
2674 | | /* Miscellaneous operations */ |
2675 | | /******************************************************************************/ |
2676 | | |
2677 | | /* Logical And */ |
2678 | | void |
2679 | | mpd_qand(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2680 | | const mpd_context_t *ctx, uint32_t *status) |
2681 | 79 | { |
2682 | 79 | const mpd_t *big = a, *small = b; |
2683 | 79 | mpd_uint_t x, y, z, xbit, ybit; |
2684 | 79 | int k, mswdigits; |
2685 | 79 | mpd_ssize_t i; |
2686 | | |
2687 | 79 | if (mpd_isspecial(a) || mpd_isspecial(b) || |
2688 | 79 | mpd_isnegative(a) || mpd_isnegative(b) || |
2689 | 79 | a->exp != 0 || b->exp != 0) { |
2690 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
2691 | 0 | return; |
2692 | 0 | } |
2693 | 79 | if (b->digits > a->digits) { |
2694 | 33 | big = b; |
2695 | 33 | small = a; |
2696 | 33 | } |
2697 | 79 | if (!mpd_qresize(result, big->len, status)) { |
2698 | 0 | return; |
2699 | 0 | } |
2700 | | |
2701 | | |
2702 | | /* full words */ |
2703 | 85 | for (i = 0; i < small->len-1; i++) { |
2704 | 58 | x = small->data[i]; |
2705 | 58 | y = big->data[i]; |
2706 | 58 | z = 0; |
2707 | 268 | for (k = 0; k < MPD_RDIGITS; k++) { |
2708 | 262 | xbit = x % 10; |
2709 | 262 | x /= 10; |
2710 | 262 | ybit = y % 10; |
2711 | 262 | y /= 10; |
2712 | 262 | if (xbit > 1 || ybit > 1) { |
2713 | 52 | goto invalid_operation; |
2714 | 52 | } |
2715 | 210 | z += (xbit&ybit) ? mpd_pow10[k] : 0; |
2716 | 210 | } |
2717 | 6 | result->data[i] = z; |
2718 | 6 | } |
2719 | | /* most significant word of small */ |
2720 | 27 | x = small->data[i]; |
2721 | 27 | y = big->data[i]; |
2722 | 27 | z = 0; |
2723 | 27 | mswdigits = mpd_word_digits(x); |
2724 | 67 | for (k = 0; k < mswdigits; k++) { |
2725 | 55 | xbit = x % 10; |
2726 | 55 | x /= 10; |
2727 | 55 | ybit = y % 10; |
2728 | 55 | y /= 10; |
2729 | 55 | if (xbit > 1 || ybit > 1) { |
2730 | 15 | goto invalid_operation; |
2731 | 15 | } |
2732 | 40 | z += (xbit&ybit) ? mpd_pow10[k] : 0; |
2733 | 40 | } |
2734 | 12 | result->data[i++] = z; |
2735 | | |
2736 | | /* scan the rest of y for digits > 1 */ |
2737 | 166 | for (; k < MPD_RDIGITS; k++) { |
2738 | 160 | ybit = y % 10; |
2739 | 160 | y /= 10; |
2740 | 160 | if (ybit > 1) { |
2741 | 6 | goto invalid_operation; |
2742 | 6 | } |
2743 | 160 | } |
2744 | | /* scan the rest of big for digits > 1 */ |
2745 | 63 | for (; i < big->len; i++) { |
2746 | 61 | y = big->data[i]; |
2747 | 1.16k | for (k = 0; k < MPD_RDIGITS; k++) { |
2748 | 1.10k | ybit = y % 10; |
2749 | 1.10k | y /= 10; |
2750 | 1.10k | if (ybit > 1) { |
2751 | 4 | goto invalid_operation; |
2752 | 4 | } |
2753 | 1.10k | } |
2754 | 61 | } |
2755 | | |
2756 | 2 | mpd_clear_flags(result); |
2757 | 2 | result->exp = 0; |
2758 | 2 | result->len = _mpd_real_size(result->data, small->len); |
2759 | 2 | mpd_qresize(result, result->len, status); |
2760 | 2 | mpd_setdigits(result); |
2761 | 2 | _mpd_cap(result, ctx); |
2762 | 2 | return; |
2763 | | |
2764 | 77 | invalid_operation: |
2765 | 77 | mpd_seterror(result, MPD_Invalid_operation, status); |
2766 | 77 | } |
2767 | | |
2768 | | /* Class of an operand. Returns a pointer to the constant name. */ |
2769 | | const char * |
2770 | | mpd_class(const mpd_t *a, const mpd_context_t *ctx) |
2771 | 0 | { |
2772 | 0 | if (mpd_isnan(a)) { |
2773 | 0 | if (mpd_isqnan(a)) |
2774 | 0 | return "NaN"; |
2775 | 0 | else |
2776 | 0 | return "sNaN"; |
2777 | 0 | } |
2778 | 0 | else if (mpd_ispositive(a)) { |
2779 | 0 | if (mpd_isinfinite(a)) |
2780 | 0 | return "+Infinity"; |
2781 | 0 | else if (mpd_iszero(a)) |
2782 | 0 | return "+Zero"; |
2783 | 0 | else if (mpd_isnormal(a, ctx)) |
2784 | 0 | return "+Normal"; |
2785 | 0 | else |
2786 | 0 | return "+Subnormal"; |
2787 | 0 | } |
2788 | 0 | else { |
2789 | 0 | if (mpd_isinfinite(a)) |
2790 | 0 | return "-Infinity"; |
2791 | 0 | else if (mpd_iszero(a)) |
2792 | 0 | return "-Zero"; |
2793 | 0 | else if (mpd_isnormal(a, ctx)) |
2794 | 0 | return "-Normal"; |
2795 | 0 | else |
2796 | 0 | return "-Subnormal"; |
2797 | 0 | } |
2798 | 0 | } |
2799 | | |
2800 | | /* Logical Xor */ |
2801 | | void |
2802 | | mpd_qinvert(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
2803 | | uint32_t *status) |
2804 | 0 | { |
2805 | 0 | mpd_uint_t x, z, xbit; |
2806 | 0 | mpd_ssize_t i, digits, len; |
2807 | 0 | mpd_ssize_t q, r; |
2808 | 0 | int k; |
2809 | |
|
2810 | 0 | if (mpd_isspecial(a) || mpd_isnegative(a) || a->exp != 0) { |
2811 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
2812 | 0 | return; |
2813 | 0 | } |
2814 | | |
2815 | 0 | digits = (a->digits < ctx->prec) ? ctx->prec : a->digits; |
2816 | 0 | _mpd_idiv_word(&q, &r, digits, MPD_RDIGITS); |
2817 | 0 | len = (r == 0) ? q : q+1; |
2818 | 0 | if (!mpd_qresize(result, len, status)) { |
2819 | 0 | return; |
2820 | 0 | } |
2821 | | |
2822 | 0 | for (i = 0; i < len; i++) { |
2823 | 0 | x = (i < a->len) ? a->data[i] : 0; |
2824 | 0 | z = 0; |
2825 | 0 | for (k = 0; k < MPD_RDIGITS; k++) { |
2826 | 0 | xbit = x % 10; |
2827 | 0 | x /= 10; |
2828 | 0 | if (xbit > 1) { |
2829 | 0 | goto invalid_operation; |
2830 | 0 | } |
2831 | 0 | z += !xbit ? mpd_pow10[k] : 0; |
2832 | 0 | } |
2833 | 0 | result->data[i] = z; |
2834 | 0 | } |
2835 | | |
2836 | 0 | mpd_clear_flags(result); |
2837 | 0 | result->exp = 0; |
2838 | 0 | result->len = _mpd_real_size(result->data, len); |
2839 | 0 | mpd_qresize(result, result->len, status); |
2840 | 0 | mpd_setdigits(result); |
2841 | 0 | _mpd_cap(result, ctx); |
2842 | 0 | return; |
2843 | | |
2844 | 0 | invalid_operation: |
2845 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
2846 | 0 | } |
2847 | | |
2848 | | /* Exponent of the magnitude of the most significant digit of the operand. */ |
2849 | | void |
2850 | | mpd_qlogb(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
2851 | | uint32_t *status) |
2852 | 0 | { |
2853 | 0 | if (mpd_isspecial(a)) { |
2854 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
2855 | 0 | return; |
2856 | 0 | } |
2857 | 0 | mpd_setspecial(result, MPD_POS, MPD_INF); |
2858 | 0 | } |
2859 | 0 | else if (mpd_iszerocoeff(a)) { |
2860 | 0 | mpd_setspecial(result, MPD_NEG, MPD_INF); |
2861 | 0 | *status |= MPD_Division_by_zero; |
2862 | 0 | } |
2863 | 0 | else { |
2864 | 0 | mpd_qset_ssize(result, mpd_adjexp(a), ctx, status); |
2865 | 0 | } |
2866 | 0 | } |
2867 | | |
2868 | | /* Logical Or */ |
2869 | | void |
2870 | | mpd_qor(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2871 | | const mpd_context_t *ctx, uint32_t *status) |
2872 | 80 | { |
2873 | 80 | const mpd_t *big = a, *small = b; |
2874 | 80 | mpd_uint_t x, y, z, xbit, ybit; |
2875 | 80 | int k, mswdigits; |
2876 | 80 | mpd_ssize_t i; |
2877 | | |
2878 | 80 | if (mpd_isspecial(a) || mpd_isspecial(b) || |
2879 | 80 | mpd_isnegative(a) || mpd_isnegative(b) || |
2880 | 80 | a->exp != 0 || b->exp != 0) { |
2881 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
2882 | 0 | return; |
2883 | 0 | } |
2884 | 80 | if (b->digits > a->digits) { |
2885 | 38 | big = b; |
2886 | 38 | small = a; |
2887 | 38 | } |
2888 | 80 | if (!mpd_qresize(result, big->len, status)) { |
2889 | 0 | return; |
2890 | 0 | } |
2891 | | |
2892 | | |
2893 | | /* full words */ |
2894 | 92 | for (i = 0; i < small->len-1; i++) { |
2895 | 68 | x = small->data[i]; |
2896 | 68 | y = big->data[i]; |
2897 | 68 | z = 0; |
2898 | 346 | for (k = 0; k < MPD_RDIGITS; k++) { |
2899 | 334 | xbit = x % 10; |
2900 | 334 | x /= 10; |
2901 | 334 | ybit = y % 10; |
2902 | 334 | y /= 10; |
2903 | 334 | if (xbit > 1 || ybit > 1) { |
2904 | 56 | goto invalid_operation; |
2905 | 56 | } |
2906 | 278 | z += (xbit|ybit) ? mpd_pow10[k] : 0; |
2907 | 278 | } |
2908 | 12 | result->data[i] = z; |
2909 | 12 | } |
2910 | | /* most significant word of small */ |
2911 | 24 | x = small->data[i]; |
2912 | 24 | y = big->data[i]; |
2913 | 24 | z = 0; |
2914 | 24 | mswdigits = mpd_word_digits(x); |
2915 | 37 | for (k = 0; k < mswdigits; k++) { |
2916 | 24 | xbit = x % 10; |
2917 | 24 | x /= 10; |
2918 | 24 | ybit = y % 10; |
2919 | 24 | y /= 10; |
2920 | 24 | if (xbit > 1 || ybit > 1) { |
2921 | 11 | goto invalid_operation; |
2922 | 11 | } |
2923 | 13 | z += (xbit|ybit) ? mpd_pow10[k] : 0; |
2924 | 13 | } |
2925 | | |
2926 | | /* scan for digits > 1 and copy the rest of y */ |
2927 | 159 | for (; k < MPD_RDIGITS; k++) { |
2928 | 152 | ybit = y % 10; |
2929 | 152 | y /= 10; |
2930 | 152 | if (ybit > 1) { |
2931 | 6 | goto invalid_operation; |
2932 | 6 | } |
2933 | 146 | z += ybit*mpd_pow10[k]; |
2934 | 146 | } |
2935 | 7 | result->data[i++] = z; |
2936 | | /* scan for digits > 1 and copy the rest of big */ |
2937 | 137 | for (; i < big->len; i++) { |
2938 | 137 | y = big->data[i]; |
2939 | 2.67k | for (k = 0; k < MPD_RDIGITS; k++) { |
2940 | 2.54k | ybit = y % 10; |
2941 | 2.54k | y /= 10; |
2942 | 2.54k | if (ybit > 1) { |
2943 | 7 | goto invalid_operation; |
2944 | 7 | } |
2945 | 2.54k | } |
2946 | 130 | result->data[i] = big->data[i]; |
2947 | 130 | } |
2948 | | |
2949 | 0 | mpd_clear_flags(result); |
2950 | 0 | result->exp = 0; |
2951 | 0 | result->len = _mpd_real_size(result->data, big->len); |
2952 | 0 | mpd_qresize(result, result->len, status); |
2953 | 0 | mpd_setdigits(result); |
2954 | 0 | _mpd_cap(result, ctx); |
2955 | 0 | return; |
2956 | | |
2957 | 80 | invalid_operation: |
2958 | 80 | mpd_seterror(result, MPD_Invalid_operation, status); |
2959 | 80 | } |
2960 | | |
2961 | | /* |
2962 | | * Rotate the coefficient of 'a' by 'b' digits. 'b' must be an integer with |
2963 | | * exponent 0. |
2964 | | */ |
2965 | | void |
2966 | | mpd_qrotate(mpd_t *result, const mpd_t *a, const mpd_t *b, |
2967 | | const mpd_context_t *ctx, uint32_t *status) |
2968 | 0 | { |
2969 | 0 | uint32_t workstatus = 0; |
2970 | 0 | MPD_NEW_STATIC(tmp,0,0,0,0); |
2971 | 0 | MPD_NEW_STATIC(big,0,0,0,0); |
2972 | 0 | MPD_NEW_STATIC(small,0,0,0,0); |
2973 | 0 | mpd_ssize_t n, lshift, rshift; |
2974 | |
|
2975 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
2976 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
2977 | 0 | return; |
2978 | 0 | } |
2979 | 0 | } |
2980 | 0 | if (b->exp != 0 || mpd_isinfinite(b)) { |
2981 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
2982 | 0 | return; |
2983 | 0 | } |
2984 | | |
2985 | 0 | n = mpd_qget_ssize(b, &workstatus); |
2986 | 0 | if (workstatus&MPD_Invalid_operation) { |
2987 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
2988 | 0 | return; |
2989 | 0 | } |
2990 | 0 | if (n > ctx->prec || n < -ctx->prec) { |
2991 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
2992 | 0 | return; |
2993 | 0 | } |
2994 | 0 | if (mpd_isinfinite(a)) { |
2995 | 0 | mpd_qcopy(result, a, status); |
2996 | 0 | return; |
2997 | 0 | } |
2998 | | |
2999 | 0 | if (n >= 0) { |
3000 | 0 | lshift = n; |
3001 | 0 | rshift = ctx->prec-n; |
3002 | 0 | } |
3003 | 0 | else { |
3004 | 0 | lshift = ctx->prec+n; |
3005 | 0 | rshift = -n; |
3006 | 0 | } |
3007 | |
|
3008 | 0 | if (a->digits > ctx->prec) { |
3009 | 0 | if (!mpd_qcopy(&tmp, a, status)) { |
3010 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
3011 | 0 | goto finish; |
3012 | 0 | } |
3013 | 0 | _mpd_cap(&tmp, ctx); |
3014 | 0 | a = &tmp; |
3015 | 0 | } |
3016 | | |
3017 | 0 | if (!mpd_qshiftl(&big, a, lshift, status)) { |
3018 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
3019 | 0 | goto finish; |
3020 | 0 | } |
3021 | 0 | _mpd_cap(&big, ctx); |
3022 | |
|
3023 | 0 | if (mpd_qshiftr(&small, a, rshift, status) == MPD_UINT_MAX) { |
3024 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
3025 | 0 | goto finish; |
3026 | 0 | } |
3027 | 0 | _mpd_qadd(result, &big, &small, ctx, status); |
3028 | | |
3029 | |
|
3030 | 0 | finish: |
3031 | 0 | mpd_del(&tmp); |
3032 | 0 | mpd_del(&big); |
3033 | 0 | mpd_del(&small); |
3034 | 0 | } |
3035 | | |
3036 | | /* |
3037 | | * b must be an integer with exponent 0 and in the range +-2*(emax + prec). |
3038 | | * XXX: In my opinion +-(2*emax + prec) would be more sensible. |
3039 | | * The result is a with the value of b added to its exponent. |
3040 | | */ |
3041 | | void |
3042 | | mpd_qscaleb(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3043 | | const mpd_context_t *ctx, uint32_t *status) |
3044 | 0 | { |
3045 | 0 | uint32_t workstatus = 0; |
3046 | 0 | mpd_uint_t n, maxjump; |
3047 | 0 | #ifndef LEGACY_COMPILER |
3048 | 0 | int64_t exp; |
3049 | | #else |
3050 | | mpd_uint_t x; |
3051 | | int x_sign, n_sign; |
3052 | | mpd_ssize_t exp; |
3053 | | #endif |
3054 | |
|
3055 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3056 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
3057 | 0 | return; |
3058 | 0 | } |
3059 | 0 | } |
3060 | 0 | if (b->exp != 0 || mpd_isinfinite(b)) { |
3061 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3062 | 0 | return; |
3063 | 0 | } |
3064 | | |
3065 | 0 | n = mpd_qabs_uint(b, &workstatus); |
3066 | | /* the spec demands this */ |
3067 | 0 | maxjump = 2 * (mpd_uint_t)(ctx->emax + ctx->prec); |
3068 | |
|
3069 | 0 | if (n > maxjump || workstatus&MPD_Invalid_operation) { |
3070 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3071 | 0 | return; |
3072 | 0 | } |
3073 | 0 | if (mpd_isinfinite(a)) { |
3074 | 0 | mpd_qcopy(result, a, status); |
3075 | 0 | return; |
3076 | 0 | } |
3077 | | |
3078 | 0 | #ifndef LEGACY_COMPILER |
3079 | 0 | exp = a->exp + (int64_t)n * mpd_arith_sign(b); |
3080 | 0 | exp = (exp > MPD_EXP_INF) ? MPD_EXP_INF : exp; |
3081 | 0 | exp = (exp < MPD_EXP_CLAMP) ? MPD_EXP_CLAMP : exp; |
3082 | | #else |
3083 | | x = (a->exp < 0) ? -a->exp : a->exp; |
3084 | | x_sign = (a->exp < 0) ? 1 : 0; |
3085 | | n_sign = mpd_isnegative(b) ? 1 : 0; |
3086 | | |
3087 | | if (x_sign == n_sign) { |
3088 | | x = x + n; |
3089 | | if (x < n) x = MPD_UINT_MAX; |
3090 | | } |
3091 | | else { |
3092 | | x_sign = (x >= n) ? x_sign : n_sign; |
3093 | | x = (x >= n) ? x - n : n - x; |
3094 | | } |
3095 | | if (!x_sign && x > MPD_EXP_INF) x = MPD_EXP_INF; |
3096 | | if (x_sign && x > -MPD_EXP_CLAMP) x = -MPD_EXP_CLAMP; |
3097 | | exp = x_sign ? -((mpd_ssize_t)x) : (mpd_ssize_t)x; |
3098 | | #endif |
3099 | |
|
3100 | 0 | mpd_qcopy(result, a, status); |
3101 | 0 | result->exp = (mpd_ssize_t)exp; |
3102 | |
|
3103 | 0 | mpd_qfinalize(result, ctx, status); |
3104 | 0 | } |
3105 | | |
3106 | | /* |
3107 | | * Shift the coefficient by n digits, positive n is a left shift. In the case |
3108 | | * of a left shift, the result is decapitated to fit the context precision. If |
3109 | | * you don't want that, use mpd_shiftl(). |
3110 | | */ |
3111 | | void |
3112 | | mpd_qshiftn(mpd_t *result, const mpd_t *a, mpd_ssize_t n, const mpd_context_t *ctx, |
3113 | | uint32_t *status) |
3114 | 0 | { |
3115 | 0 | if (mpd_isspecial(a)) { |
3116 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
3117 | 0 | return; |
3118 | 0 | } |
3119 | 0 | mpd_qcopy(result, a, status); |
3120 | 0 | return; |
3121 | 0 | } |
3122 | | |
3123 | 0 | if (n >= 0 && n <= ctx->prec) { |
3124 | 0 | mpd_qshiftl(result, a, n, status); |
3125 | 0 | _mpd_cap(result, ctx); |
3126 | 0 | } |
3127 | 0 | else if (n < 0 && n >= -ctx->prec) { |
3128 | 0 | if (!mpd_qcopy(result, a, status)) { |
3129 | 0 | return; |
3130 | 0 | } |
3131 | 0 | _mpd_cap(result, ctx); |
3132 | 0 | mpd_qshiftr_inplace(result, -n); |
3133 | 0 | } |
3134 | 0 | else { |
3135 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3136 | 0 | } |
3137 | 0 | } |
3138 | | |
3139 | | /* |
3140 | | * Same as mpd_shiftn(), but the shift is specified by the decimal b, which |
3141 | | * must be an integer with a zero exponent. Infinities remain infinities. |
3142 | | */ |
3143 | | void |
3144 | | mpd_qshift(mpd_t *result, const mpd_t *a, const mpd_t *b, const mpd_context_t *ctx, |
3145 | | uint32_t *status) |
3146 | 0 | { |
3147 | 0 | uint32_t workstatus = 0; |
3148 | 0 | mpd_ssize_t n; |
3149 | |
|
3150 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3151 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
3152 | 0 | return; |
3153 | 0 | } |
3154 | 0 | } |
3155 | 0 | if (b->exp != 0 || mpd_isinfinite(b)) { |
3156 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3157 | 0 | return; |
3158 | 0 | } |
3159 | | |
3160 | 0 | n = mpd_qget_ssize(b, &workstatus); |
3161 | 0 | if (workstatus&MPD_Invalid_operation) { |
3162 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3163 | 0 | return; |
3164 | 0 | } |
3165 | 0 | if (n > ctx->prec || n < -ctx->prec) { |
3166 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3167 | 0 | return; |
3168 | 0 | } |
3169 | 0 | if (mpd_isinfinite(a)) { |
3170 | 0 | mpd_qcopy(result, a, status); |
3171 | 0 | return; |
3172 | 0 | } |
3173 | | |
3174 | 0 | if (n >= 0) { |
3175 | 0 | mpd_qshiftl(result, a, n, status); |
3176 | 0 | _mpd_cap(result, ctx); |
3177 | 0 | } |
3178 | 0 | else { |
3179 | 0 | if (!mpd_qcopy(result, a, status)) { |
3180 | 0 | return; |
3181 | 0 | } |
3182 | 0 | _mpd_cap(result, ctx); |
3183 | 0 | mpd_qshiftr_inplace(result, -n); |
3184 | 0 | } |
3185 | 0 | } |
3186 | | |
3187 | | /* Logical Xor */ |
3188 | | void |
3189 | | mpd_qxor(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3190 | | const mpd_context_t *ctx, uint32_t *status) |
3191 | 93 | { |
3192 | 93 | const mpd_t *big = a, *small = b; |
3193 | 93 | mpd_uint_t x, y, z, xbit, ybit; |
3194 | 93 | int k, mswdigits; |
3195 | 93 | mpd_ssize_t i; |
3196 | | |
3197 | 93 | if (mpd_isspecial(a) || mpd_isspecial(b) || |
3198 | 93 | mpd_isnegative(a) || mpd_isnegative(b) || |
3199 | 93 | a->exp != 0 || b->exp != 0) { |
3200 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3201 | 0 | return; |
3202 | 0 | } |
3203 | 93 | if (b->digits > a->digits) { |
3204 | 46 | big = b; |
3205 | 46 | small = a; |
3206 | 46 | } |
3207 | 93 | if (!mpd_qresize(result, big->len, status)) { |
3208 | 0 | return; |
3209 | 0 | } |
3210 | | |
3211 | | |
3212 | | /* full words */ |
3213 | 125 | for (i = 0; i < small->len-1; i++) { |
3214 | 92 | x = small->data[i]; |
3215 | 92 | y = big->data[i]; |
3216 | 92 | z = 0; |
3217 | 762 | for (k = 0; k < MPD_RDIGITS; k++) { |
3218 | 730 | xbit = x % 10; |
3219 | 730 | x /= 10; |
3220 | 730 | ybit = y % 10; |
3221 | 730 | y /= 10; |
3222 | 730 | if (xbit > 1 || ybit > 1) { |
3223 | 60 | goto invalid_operation; |
3224 | 60 | } |
3225 | 670 | z += (xbit^ybit) ? mpd_pow10[k] : 0; |
3226 | 670 | } |
3227 | 32 | result->data[i] = z; |
3228 | 32 | } |
3229 | | /* most significant word of small */ |
3230 | 33 | x = small->data[i]; |
3231 | 33 | y = big->data[i]; |
3232 | 33 | z = 0; |
3233 | 33 | mswdigits = mpd_word_digits(x); |
3234 | 60 | for (k = 0; k < mswdigits; k++) { |
3235 | 34 | xbit = x % 10; |
3236 | 34 | x /= 10; |
3237 | 34 | ybit = y % 10; |
3238 | 34 | y /= 10; |
3239 | 34 | if (xbit > 1 || ybit > 1) { |
3240 | 7 | goto invalid_operation; |
3241 | 7 | } |
3242 | 27 | z += (xbit^ybit) ? mpd_pow10[k] : 0; |
3243 | 27 | } |
3244 | | |
3245 | | /* scan for digits > 1 and copy the rest of y */ |
3246 | 321 | for (; k < MPD_RDIGITS; k++) { |
3247 | 308 | ybit = y % 10; |
3248 | 308 | y /= 10; |
3249 | 308 | if (ybit > 1) { |
3250 | 13 | goto invalid_operation; |
3251 | 13 | } |
3252 | 295 | z += ybit*mpd_pow10[k]; |
3253 | 295 | } |
3254 | 13 | result->data[i++] = z; |
3255 | | /* scan for digits > 1 and copy the rest of big */ |
3256 | 43 | for (; i < big->len; i++) { |
3257 | 41 | y = big->data[i]; |
3258 | 656 | for (k = 0; k < MPD_RDIGITS; k++) { |
3259 | 626 | ybit = y % 10; |
3260 | 626 | y /= 10; |
3261 | 626 | if (ybit > 1) { |
3262 | 11 | goto invalid_operation; |
3263 | 11 | } |
3264 | 626 | } |
3265 | 30 | result->data[i] = big->data[i]; |
3266 | 30 | } |
3267 | | |
3268 | 2 | mpd_clear_flags(result); |
3269 | 2 | result->exp = 0; |
3270 | 2 | result->len = _mpd_real_size(result->data, big->len); |
3271 | 2 | mpd_qresize(result, result->len, status); |
3272 | 2 | mpd_setdigits(result); |
3273 | 2 | _mpd_cap(result, ctx); |
3274 | 2 | return; |
3275 | | |
3276 | 91 | invalid_operation: |
3277 | 91 | mpd_seterror(result, MPD_Invalid_operation, status); |
3278 | 91 | } |
3279 | | |
3280 | | |
3281 | | /******************************************************************************/ |
3282 | | /* Arithmetic operations */ |
3283 | | /******************************************************************************/ |
3284 | | |
3285 | | /* |
3286 | | * The absolute value of a. If a is negative, the result is the same |
3287 | | * as the result of the minus operation. Otherwise, the result is the |
3288 | | * result of the plus operation. |
3289 | | */ |
3290 | | void |
3291 | | mpd_qabs(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
3292 | | uint32_t *status) |
3293 | 42 | { |
3294 | 42 | if (mpd_isspecial(a)) { |
3295 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
3296 | 0 | return; |
3297 | 0 | } |
3298 | 0 | } |
3299 | | |
3300 | 42 | if (mpd_isnegative(a)) { |
3301 | 0 | mpd_qminus(result, a, ctx, status); |
3302 | 0 | } |
3303 | 42 | else { |
3304 | 42 | mpd_qplus(result, a, ctx, status); |
3305 | 42 | } |
3306 | 42 | } |
3307 | | |
3308 | | static inline void |
3309 | | _mpd_ptrswap(const mpd_t **a, const mpd_t **b) |
3310 | 45.0k | { |
3311 | 45.0k | const mpd_t *t = *a; |
3312 | 45.0k | *a = *b; |
3313 | 45.0k | *b = t; |
3314 | 45.0k | } |
3315 | | |
3316 | | /* Add or subtract infinities. */ |
3317 | | static void |
3318 | | _mpd_qaddsub_inf(mpd_t *result, const mpd_t *a, const mpd_t *b, uint8_t sign_b, |
3319 | | uint32_t *status) |
3320 | 0 | { |
3321 | 0 | if (mpd_isinfinite(a)) { |
3322 | 0 | if (mpd_sign(a) != sign_b && mpd_isinfinite(b)) { |
3323 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3324 | 0 | } |
3325 | 0 | else { |
3326 | 0 | mpd_setspecial(result, mpd_sign(a), MPD_INF); |
3327 | 0 | } |
3328 | 0 | return; |
3329 | 0 | } |
3330 | 0 | assert(mpd_isinfinite(b)); |
3331 | 0 | mpd_setspecial(result, sign_b, MPD_INF); |
3332 | 0 | } |
3333 | | |
3334 | | /* Add or subtract non-special numbers. */ |
3335 | | static void |
3336 | | _mpd_qaddsub(mpd_t *result, const mpd_t *a, const mpd_t *b, uint8_t sign_b, |
3337 | | const mpd_context_t *ctx, uint32_t *status) |
3338 | 427 | { |
3339 | 427 | const mpd_t *big, *small; |
3340 | 427 | MPD_NEW_STATIC(big_aligned,0,0,0,0); |
3341 | 427 | MPD_NEW_CONST(tiny,0,0,1,1,1,1); |
3342 | 427 | mpd_uint_t carry; |
3343 | 427 | mpd_ssize_t newsize, shift; |
3344 | 427 | mpd_ssize_t exp, i; |
3345 | 427 | int swap = 0; |
3346 | | |
3347 | | |
3348 | | /* compare exponents */ |
3349 | 427 | big = a; small = b; |
3350 | 427 | if (big->exp != small->exp) { |
3351 | 0 | if (small->exp > big->exp) { |
3352 | 0 | _mpd_ptrswap(&big, &small); |
3353 | 0 | swap++; |
3354 | 0 | } |
3355 | | /* align the coefficients */ |
3356 | 0 | if (!mpd_iszerocoeff(big)) { |
3357 | 0 | exp = big->exp - 1; |
3358 | 0 | exp += (big->digits > ctx->prec) ? 0 : big->digits-ctx->prec-1; |
3359 | 0 | if (mpd_adjexp(small) < exp) { |
3360 | | /* |
3361 | | * Avoid huge shifts by substituting a value for small that is |
3362 | | * guaranteed to produce the same results. |
3363 | | * |
3364 | | * adjexp(small) < exp if and only if: |
3365 | | * |
3366 | | * bdigits <= prec AND |
3367 | | * bdigits+shift >= prec+2+sdigits AND |
3368 | | * exp = bexp+bdigits-prec-2 |
3369 | | * |
3370 | | * 1234567000000000 -> bdigits + shift |
3371 | | * ----------XX1234 -> sdigits |
3372 | | * ----------X1 -> tiny-digits |
3373 | | * |- prec -| |
3374 | | * |
3375 | | * OR |
3376 | | * |
3377 | | * bdigits > prec AND |
3378 | | * shift > sdigits AND |
3379 | | * exp = bexp-1 |
3380 | | * |
3381 | | * 1234567892100000 -> bdigits + shift |
3382 | | * ----------XX1234 -> sdigits |
3383 | | * ----------X1 -> tiny-digits |
3384 | | * |- prec -| |
3385 | | * |
3386 | | * If tiny is zero, adding or subtracting is a no-op. |
3387 | | * Otherwise, adding tiny generates a non-zero digit either |
3388 | | * below the rounding digit or the least significant digit |
3389 | | * of big. When subtracting, tiny is in the same position as |
3390 | | * the carry that would be generated by subtracting sdigits. |
3391 | | */ |
3392 | 0 | mpd_copy_flags(&tiny, small); |
3393 | 0 | tiny.exp = exp; |
3394 | 0 | tiny.digits = 1; |
3395 | 0 | tiny.len = 1; |
3396 | 0 | tiny.data[0] = mpd_iszerocoeff(small) ? 0 : 1; |
3397 | 0 | small = &tiny; |
3398 | 0 | } |
3399 | | /* This cannot wrap: the difference is positive and <= maxprec */ |
3400 | 0 | shift = big->exp - small->exp; |
3401 | 0 | if (!mpd_qshiftl(&big_aligned, big, shift, status)) { |
3402 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
3403 | 0 | goto finish; |
3404 | 0 | } |
3405 | 0 | big = &big_aligned; |
3406 | 0 | } |
3407 | 0 | } |
3408 | 427 | result->exp = small->exp; |
3409 | | |
3410 | | |
3411 | | /* compare length of coefficients */ |
3412 | 427 | if (big->len < small->len) { |
3413 | 159 | _mpd_ptrswap(&big, &small); |
3414 | 159 | swap++; |
3415 | 159 | } |
3416 | | |
3417 | 427 | newsize = big->len; |
3418 | 427 | if (!mpd_qresize(result, newsize, status)) { |
3419 | 0 | goto finish; |
3420 | 0 | } |
3421 | | |
3422 | 427 | if (mpd_sign(a) == sign_b) { |
3423 | | |
3424 | 196 | carry = _mpd_baseadd(result->data, big->data, small->data, |
3425 | 196 | big->len, small->len); |
3426 | | |
3427 | 196 | if (carry) { |
3428 | 0 | newsize = big->len + 1; |
3429 | 0 | if (!mpd_qresize(result, newsize, status)) { |
3430 | 0 | goto finish; |
3431 | 0 | } |
3432 | 0 | result->data[newsize-1] = carry; |
3433 | 0 | } |
3434 | | |
3435 | 196 | result->len = newsize; |
3436 | 196 | mpd_set_flags(result, sign_b); |
3437 | 196 | } |
3438 | 231 | else { |
3439 | 231 | if (big->len == small->len) { |
3440 | 76 | for (i=big->len-1; i >= 0; --i) { |
3441 | 68 | if (big->data[i] != small->data[i]) { |
3442 | 48 | if (big->data[i] < small->data[i]) { |
3443 | 36 | _mpd_ptrswap(&big, &small); |
3444 | 36 | swap++; |
3445 | 36 | } |
3446 | 48 | break; |
3447 | 48 | } |
3448 | 68 | } |
3449 | 56 | } |
3450 | | |
3451 | 231 | _mpd_basesub(result->data, big->data, small->data, |
3452 | 231 | big->len, small->len); |
3453 | 231 | newsize = _mpd_real_size(result->data, big->len); |
3454 | | /* resize to smaller cannot fail */ |
3455 | 231 | (void)mpd_qresize(result, newsize, status); |
3456 | | |
3457 | 231 | result->len = newsize; |
3458 | 231 | sign_b = (swap & 1) ? sign_b : mpd_sign(a); |
3459 | 231 | mpd_set_flags(result, sign_b); |
3460 | | |
3461 | 231 | if (mpd_iszerocoeff(result)) { |
3462 | 8 | mpd_set_positive(result); |
3463 | 8 | if (ctx->round == MPD_ROUND_FLOOR) { |
3464 | 0 | mpd_set_negative(result); |
3465 | 0 | } |
3466 | 8 | } |
3467 | 231 | } |
3468 | | |
3469 | 427 | mpd_setdigits(result); |
3470 | | |
3471 | 427 | finish: |
3472 | 427 | mpd_del(&big_aligned); |
3473 | 427 | } |
3474 | | |
3475 | | /* Add a and b. No specials, no finalizing. */ |
3476 | | static void |
3477 | | _mpd_qadd(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3478 | | const mpd_context_t *ctx, uint32_t *status) |
3479 | 0 | { |
3480 | 0 | _mpd_qaddsub(result, a, b, mpd_sign(b), ctx, status); |
3481 | 0 | } |
3482 | | |
3483 | | /* Subtract b from a. No specials, no finalizing. */ |
3484 | | static void |
3485 | | _mpd_qsub(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3486 | | const mpd_context_t *ctx, uint32_t *status) |
3487 | 0 | { |
3488 | 0 | _mpd_qaddsub(result, a, b, !mpd_sign(b), ctx, status); |
3489 | 0 | } |
3490 | | |
3491 | | /* Add a and b. */ |
3492 | | void |
3493 | | mpd_qadd(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3494 | | const mpd_context_t *ctx, uint32_t *status) |
3495 | 196 | { |
3496 | 196 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3497 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
3498 | 0 | return; |
3499 | 0 | } |
3500 | 0 | _mpd_qaddsub_inf(result, a, b, mpd_sign(b), status); |
3501 | 0 | return; |
3502 | 0 | } |
3503 | | |
3504 | 196 | _mpd_qaddsub(result, a, b, mpd_sign(b), ctx, status); |
3505 | 196 | mpd_qfinalize(result, ctx, status); |
3506 | 196 | } |
3507 | | |
3508 | | /* Add a and b. Set NaN/Invalid_operation if the result is inexact. */ |
3509 | | static void |
3510 | | _mpd_qadd_exact(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3511 | | const mpd_context_t *ctx, uint32_t *status) |
3512 | 0 | { |
3513 | 0 | uint32_t workstatus = 0; |
3514 | |
|
3515 | 0 | mpd_qadd(result, a, b, ctx, &workstatus); |
3516 | 0 | *status |= workstatus; |
3517 | 0 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
3518 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3519 | 0 | } |
3520 | 0 | } |
3521 | | |
3522 | | /* Subtract b from a. */ |
3523 | | void |
3524 | | mpd_qsub(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3525 | | const mpd_context_t *ctx, uint32_t *status) |
3526 | 231 | { |
3527 | 231 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3528 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
3529 | 0 | return; |
3530 | 0 | } |
3531 | 0 | _mpd_qaddsub_inf(result, a, b, !mpd_sign(b), status); |
3532 | 0 | return; |
3533 | 0 | } |
3534 | | |
3535 | 231 | _mpd_qaddsub(result, a, b, !mpd_sign(b), ctx, status); |
3536 | 231 | mpd_qfinalize(result, ctx, status); |
3537 | 231 | } |
3538 | | |
3539 | | /* Subtract b from a. Set NaN/Invalid_operation if the result is inexact. */ |
3540 | | static void |
3541 | | _mpd_qsub_exact(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3542 | | const mpd_context_t *ctx, uint32_t *status) |
3543 | 0 | { |
3544 | 0 | uint32_t workstatus = 0; |
3545 | |
|
3546 | 0 | mpd_qsub(result, a, b, ctx, &workstatus); |
3547 | 0 | *status |= workstatus; |
3548 | 0 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
3549 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3550 | 0 | } |
3551 | 0 | } |
3552 | | |
3553 | | /* Add decimal and mpd_ssize_t. */ |
3554 | | void |
3555 | | mpd_qadd_ssize(mpd_t *result, const mpd_t *a, mpd_ssize_t b, |
3556 | | const mpd_context_t *ctx, uint32_t *status) |
3557 | 0 | { |
3558 | 0 | mpd_context_t maxcontext; |
3559 | 0 | MPD_NEW_STATIC(bb,0,0,0,0); |
3560 | |
|
3561 | 0 | mpd_maxcontext(&maxcontext); |
3562 | 0 | mpd_qsset_ssize(&bb, b, &maxcontext, status); |
3563 | 0 | mpd_qadd(result, a, &bb, ctx, status); |
3564 | 0 | mpd_del(&bb); |
3565 | 0 | } |
3566 | | |
3567 | | /* Add decimal and mpd_uint_t. */ |
3568 | | void |
3569 | | mpd_qadd_uint(mpd_t *result, const mpd_t *a, mpd_uint_t b, |
3570 | | const mpd_context_t *ctx, uint32_t *status) |
3571 | 0 | { |
3572 | 0 | mpd_context_t maxcontext; |
3573 | 0 | MPD_NEW_STATIC(bb,0,0,0,0); |
3574 | |
|
3575 | 0 | mpd_maxcontext(&maxcontext); |
3576 | 0 | mpd_qsset_uint(&bb, b, &maxcontext, status); |
3577 | 0 | mpd_qadd(result, a, &bb, ctx, status); |
3578 | 0 | mpd_del(&bb); |
3579 | 0 | } |
3580 | | |
3581 | | /* Subtract mpd_ssize_t from decimal. */ |
3582 | | void |
3583 | | mpd_qsub_ssize(mpd_t *result, const mpd_t *a, mpd_ssize_t b, |
3584 | | const mpd_context_t *ctx, uint32_t *status) |
3585 | 0 | { |
3586 | 0 | mpd_context_t maxcontext; |
3587 | 0 | MPD_NEW_STATIC(bb,0,0,0,0); |
3588 | |
|
3589 | 0 | mpd_maxcontext(&maxcontext); |
3590 | 0 | mpd_qsset_ssize(&bb, b, &maxcontext, status); |
3591 | 0 | mpd_qsub(result, a, &bb, ctx, status); |
3592 | 0 | mpd_del(&bb); |
3593 | 0 | } |
3594 | | |
3595 | | /* Subtract mpd_uint_t from decimal. */ |
3596 | | void |
3597 | | mpd_qsub_uint(mpd_t *result, const mpd_t *a, mpd_uint_t b, |
3598 | | const mpd_context_t *ctx, uint32_t *status) |
3599 | 0 | { |
3600 | 0 | mpd_context_t maxcontext; |
3601 | 0 | MPD_NEW_STATIC(bb,0,0,0,0); |
3602 | |
|
3603 | 0 | mpd_maxcontext(&maxcontext); |
3604 | 0 | mpd_qsset_uint(&bb, b, &maxcontext, status); |
3605 | 0 | mpd_qsub(result, a, &bb, ctx, status); |
3606 | 0 | mpd_del(&bb); |
3607 | 0 | } |
3608 | | |
3609 | | /* Add decimal and int32_t. */ |
3610 | | void |
3611 | | mpd_qadd_i32(mpd_t *result, const mpd_t *a, int32_t b, |
3612 | | const mpd_context_t *ctx, uint32_t *status) |
3613 | 0 | { |
3614 | 0 | mpd_qadd_ssize(result, a, b, ctx, status); |
3615 | 0 | } |
3616 | | |
3617 | | /* Add decimal and uint32_t. */ |
3618 | | void |
3619 | | mpd_qadd_u32(mpd_t *result, const mpd_t *a, uint32_t b, |
3620 | | const mpd_context_t *ctx, uint32_t *status) |
3621 | 0 | { |
3622 | 0 | mpd_qadd_uint(result, a, b, ctx, status); |
3623 | 0 | } |
3624 | | |
3625 | | #ifdef CONFIG_64 |
3626 | | /* Add decimal and int64_t. */ |
3627 | | void |
3628 | | mpd_qadd_i64(mpd_t *result, const mpd_t *a, int64_t b, |
3629 | | const mpd_context_t *ctx, uint32_t *status) |
3630 | 0 | { |
3631 | 0 | mpd_qadd_ssize(result, a, b, ctx, status); |
3632 | 0 | } |
3633 | | |
3634 | | /* Add decimal and uint64_t. */ |
3635 | | void |
3636 | | mpd_qadd_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
3637 | | const mpd_context_t *ctx, uint32_t *status) |
3638 | 0 | { |
3639 | 0 | mpd_qadd_uint(result, a, b, ctx, status); |
3640 | 0 | } |
3641 | | #elif !defined(LEGACY_COMPILER) |
3642 | | /* Add decimal and int64_t. */ |
3643 | | void |
3644 | | mpd_qadd_i64(mpd_t *result, const mpd_t *a, int64_t b, |
3645 | | const mpd_context_t *ctx, uint32_t *status) |
3646 | | { |
3647 | | mpd_context_t maxcontext; |
3648 | | MPD_NEW_STATIC(bb,0,0,0,0); |
3649 | | |
3650 | | mpd_maxcontext(&maxcontext); |
3651 | | mpd_qset_i64(&bb, b, &maxcontext, status); |
3652 | | mpd_qadd(result, a, &bb, ctx, status); |
3653 | | mpd_del(&bb); |
3654 | | } |
3655 | | |
3656 | | /* Add decimal and uint64_t. */ |
3657 | | void |
3658 | | mpd_qadd_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
3659 | | const mpd_context_t *ctx, uint32_t *status) |
3660 | | { |
3661 | | mpd_context_t maxcontext; |
3662 | | MPD_NEW_STATIC(bb,0,0,0,0); |
3663 | | |
3664 | | mpd_maxcontext(&maxcontext); |
3665 | | mpd_qset_u64(&bb, b, &maxcontext, status); |
3666 | | mpd_qadd(result, a, &bb, ctx, status); |
3667 | | mpd_del(&bb); |
3668 | | } |
3669 | | #endif |
3670 | | |
3671 | | /* Subtract int32_t from decimal. */ |
3672 | | void |
3673 | | mpd_qsub_i32(mpd_t *result, const mpd_t *a, int32_t b, |
3674 | | const mpd_context_t *ctx, uint32_t *status) |
3675 | 0 | { |
3676 | 0 | mpd_qsub_ssize(result, a, b, ctx, status); |
3677 | 0 | } |
3678 | | |
3679 | | /* Subtract uint32_t from decimal. */ |
3680 | | void |
3681 | | mpd_qsub_u32(mpd_t *result, const mpd_t *a, uint32_t b, |
3682 | | const mpd_context_t *ctx, uint32_t *status) |
3683 | 0 | { |
3684 | 0 | mpd_qsub_uint(result, a, b, ctx, status); |
3685 | 0 | } |
3686 | | |
3687 | | #ifdef CONFIG_64 |
3688 | | /* Subtract int64_t from decimal. */ |
3689 | | void |
3690 | | mpd_qsub_i64(mpd_t *result, const mpd_t *a, int64_t b, |
3691 | | const mpd_context_t *ctx, uint32_t *status) |
3692 | 0 | { |
3693 | 0 | mpd_qsub_ssize(result, a, b, ctx, status); |
3694 | 0 | } |
3695 | | |
3696 | | /* Subtract uint64_t from decimal. */ |
3697 | | void |
3698 | | mpd_qsub_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
3699 | | const mpd_context_t *ctx, uint32_t *status) |
3700 | 0 | { |
3701 | 0 | mpd_qsub_uint(result, a, b, ctx, status); |
3702 | 0 | } |
3703 | | #elif !defined(LEGACY_COMPILER) |
3704 | | /* Subtract int64_t from decimal. */ |
3705 | | void |
3706 | | mpd_qsub_i64(mpd_t *result, const mpd_t *a, int64_t b, |
3707 | | const mpd_context_t *ctx, uint32_t *status) |
3708 | | { |
3709 | | mpd_context_t maxcontext; |
3710 | | MPD_NEW_STATIC(bb,0,0,0,0); |
3711 | | |
3712 | | mpd_maxcontext(&maxcontext); |
3713 | | mpd_qset_i64(&bb, b, &maxcontext, status); |
3714 | | mpd_qsub(result, a, &bb, ctx, status); |
3715 | | mpd_del(&bb); |
3716 | | } |
3717 | | |
3718 | | /* Subtract uint64_t from decimal. */ |
3719 | | void |
3720 | | mpd_qsub_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
3721 | | const mpd_context_t *ctx, uint32_t *status) |
3722 | | { |
3723 | | mpd_context_t maxcontext; |
3724 | | MPD_NEW_STATIC(bb,0,0,0,0); |
3725 | | |
3726 | | mpd_maxcontext(&maxcontext); |
3727 | | mpd_qset_u64(&bb, b, &maxcontext, status); |
3728 | | mpd_qsub(result, a, &bb, ctx, status); |
3729 | | mpd_del(&bb); |
3730 | | } |
3731 | | #endif |
3732 | | |
3733 | | |
3734 | | /* Divide infinities. */ |
3735 | | static void |
3736 | | _mpd_qdiv_inf(mpd_t *result, const mpd_t *a, const mpd_t *b, |
3737 | | const mpd_context_t *ctx, uint32_t *status) |
3738 | 0 | { |
3739 | 0 | if (mpd_isinfinite(a)) { |
3740 | 0 | if (mpd_isinfinite(b)) { |
3741 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
3742 | 0 | return; |
3743 | 0 | } |
3744 | 0 | mpd_setspecial(result, mpd_sign(a)^mpd_sign(b), MPD_INF); |
3745 | 0 | return; |
3746 | 0 | } |
3747 | 0 | assert(mpd_isinfinite(b)); |
3748 | 0 | _settriple(result, mpd_sign(a)^mpd_sign(b), 0, mpd_etiny(ctx)); |
3749 | 0 | *status |= MPD_Clamped; |
3750 | 0 | } |
3751 | | |
3752 | | enum {NO_IDEAL_EXP, SET_IDEAL_EXP}; |
3753 | | /* Divide a by b. */ |
3754 | | static void |
3755 | | _mpd_qdiv(int action, mpd_t *q, const mpd_t *a, const mpd_t *b, |
3756 | | const mpd_context_t *ctx, uint32_t *status) |
3757 | 235 | { |
3758 | 235 | MPD_NEW_STATIC(aligned,0,0,0,0); |
3759 | 235 | mpd_uint_t ld; |
3760 | 235 | mpd_ssize_t shift, exp, tz; |
3761 | 235 | mpd_ssize_t newsize; |
3762 | 235 | mpd_ssize_t ideal_exp; |
3763 | 235 | mpd_uint_t rem; |
3764 | 235 | uint8_t sign_a = mpd_sign(a); |
3765 | 235 | uint8_t sign_b = mpd_sign(b); |
3766 | | |
3767 | | |
3768 | 235 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
3769 | 0 | if (mpd_qcheck_nans(q, a, b, ctx, status)) { |
3770 | 0 | return; |
3771 | 0 | } |
3772 | 0 | _mpd_qdiv_inf(q, a, b, ctx, status); |
3773 | 0 | return; |
3774 | 0 | } |
3775 | 235 | if (mpd_iszerocoeff(b)) { |
3776 | 21 | if (mpd_iszerocoeff(a)) { |
3777 | 9 | mpd_seterror(q, MPD_Division_undefined, status); |
3778 | 9 | } |
3779 | 12 | else { |
3780 | 12 | mpd_setspecial(q, sign_a^sign_b, MPD_INF); |
3781 | 12 | *status |= MPD_Division_by_zero; |
3782 | 12 | } |
3783 | 21 | return; |
3784 | 21 | } |
3785 | 214 | if (mpd_iszerocoeff(a)) { |
3786 | 20 | exp = a->exp - b->exp; |
3787 | 20 | _settriple(q, sign_a^sign_b, 0, exp); |
3788 | 20 | mpd_qfinalize(q, ctx, status); |
3789 | 20 | return; |
3790 | 20 | } |
3791 | | |
3792 | 194 | shift = (b->digits - a->digits) + ctx->prec + 1; |
3793 | 194 | ideal_exp = a->exp - b->exp; |
3794 | 194 | exp = ideal_exp - shift; |
3795 | 194 | if (shift > 0) { |
3796 | 194 | if (!mpd_qshiftl(&aligned, a, shift, status)) { |
3797 | 0 | mpd_seterror(q, MPD_Malloc_error, status); |
3798 | 0 | goto finish; |
3799 | 0 | } |
3800 | 194 | a = &aligned; |
3801 | 194 | } |
3802 | 0 | else if (shift < 0) { |
3803 | 0 | shift = -shift; |
3804 | 0 | if (!mpd_qshiftl(&aligned, b, shift, status)) { |
3805 | 0 | mpd_seterror(q, MPD_Malloc_error, status); |
3806 | 0 | goto finish; |
3807 | 0 | } |
3808 | 0 | b = &aligned; |
3809 | 0 | } |
3810 | | |
3811 | | |
3812 | 194 | newsize = a->len - b->len + 1; |
3813 | 194 | if ((q != b && q != a) || (q == b && newsize > b->len)) { |
3814 | 194 | if (!mpd_qresize(q, newsize, status)) { |
3815 | 0 | mpd_seterror(q, MPD_Malloc_error, status); |
3816 | 0 | goto finish; |
3817 | 0 | } |
3818 | 194 | } |
3819 | | |
3820 | | |
3821 | 194 | if (b->len == 1) { |
3822 | 74 | rem = _mpd_shortdiv(q->data, a->data, a->len, b->data[0]); |
3823 | 74 | } |
3824 | 120 | else if (b->len <= MPD_NEWTONDIV_CUTOFF) { |
3825 | 120 | int ret = _mpd_basedivmod(q->data, NULL, a->data, b->data, |
3826 | 120 | a->len, b->len); |
3827 | 120 | if (ret < 0) { |
3828 | 0 | mpd_seterror(q, MPD_Malloc_error, status); |
3829 | 0 | goto finish; |
3830 | 0 | } |
3831 | 120 | rem = ret; |
3832 | 120 | } |
3833 | 0 | else { |
3834 | 0 | MPD_NEW_STATIC(r,0,0,0,0); |
3835 | 0 | _mpd_base_ndivmod(q, &r, a, b, status); |
3836 | 0 | if (mpd_isspecial(q) || mpd_isspecial(&r)) { |
3837 | 0 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
3838 | 0 | mpd_del(&r); |
3839 | 0 | goto finish; |
3840 | 0 | } |
3841 | 0 | rem = !mpd_iszerocoeff(&r); |
3842 | 0 | mpd_del(&r); |
3843 | 0 | newsize = q->len; |
3844 | 0 | } |
3845 | | |
3846 | 194 | newsize = _mpd_real_size(q->data, newsize); |
3847 | | /* resize to smaller cannot fail */ |
3848 | 194 | mpd_qresize(q, newsize, status); |
3849 | 194 | mpd_set_flags(q, sign_a^sign_b); |
3850 | 194 | q->len = newsize; |
3851 | 194 | mpd_setdigits(q); |
3852 | | |
3853 | 194 | shift = ideal_exp - exp; |
3854 | 194 | if (rem) { |
3855 | 153 | ld = mpd_lsd(q->data[0]); |
3856 | 153 | if (ld == 0 || ld == 5) { |
3857 | 35 | q->data[0] += 1; |
3858 | 35 | } |
3859 | 153 | } |
3860 | 41 | else if (action == SET_IDEAL_EXP && shift > 0) { |
3861 | 41 | tz = mpd_trail_zeros(q); |
3862 | 41 | shift = (tz > shift) ? shift : tz; |
3863 | 41 | mpd_qshiftr_inplace(q, shift); |
3864 | 41 | exp += shift; |
3865 | 41 | } |
3866 | | |
3867 | 194 | q->exp = exp; |
3868 | | |
3869 | | |
3870 | 194 | finish: |
3871 | 194 | mpd_del(&aligned); |
3872 | 194 | mpd_qfinalize(q, ctx, status); |
3873 | 194 | } |
3874 | | |
3875 | | /* Divide a by b. */ |
3876 | | void |
3877 | | mpd_qdiv(mpd_t *q, const mpd_t *a, const mpd_t *b, |
3878 | | const mpd_context_t *ctx, uint32_t *status) |
3879 | 235 | { |
3880 | 235 | MPD_NEW_STATIC(aa,0,0,0,0); |
3881 | 235 | MPD_NEW_STATIC(bb,0,0,0,0); |
3882 | 235 | uint32_t xstatus = 0; |
3883 | | |
3884 | 235 | if (q == a) { |
3885 | 0 | if (!mpd_qcopy(&aa, a, status)) { |
3886 | 0 | mpd_seterror(q, MPD_Malloc_error, status); |
3887 | 0 | goto out; |
3888 | 0 | } |
3889 | 0 | a = &aa; |
3890 | 0 | } |
3891 | | |
3892 | 235 | if (q == b) { |
3893 | 0 | if (!mpd_qcopy(&bb, b, status)) { |
3894 | 0 | mpd_seterror(q, MPD_Malloc_error, status); |
3895 | 0 | goto out; |
3896 | 0 | } |
3897 | 0 | b = &bb; |
3898 | 0 | } |
3899 | | |
3900 | 235 | _mpd_qdiv(SET_IDEAL_EXP, q, a, b, ctx, &xstatus); |
3901 | | |
3902 | 235 | if (xstatus & (MPD_Malloc_error|MPD_Division_impossible)) { |
3903 | | /* Inexact quotients (the usual case) fill the entire context precision, |
3904 | | * which can lead to the above errors for very high precisions. Retry |
3905 | | * the operation with a lower precision in case the result is exact. |
3906 | | * |
3907 | | * We need an upper bound for the number of digits of a_coeff / b_coeff |
3908 | | * when the result is exact. If a_coeff' * 1 / b_coeff' is in lowest |
3909 | | * terms, then maxdigits(a_coeff') + maxdigits(1 / b_coeff') is a suitable |
3910 | | * bound. |
3911 | | * |
3912 | | * 1 / b_coeff' is exact iff b_coeff' exclusively has prime factors 2 or 5. |
3913 | | * The largest amount of digits is generated if b_coeff' is a power of 2 or |
3914 | | * a power of 5 and is less than or equal to log5(b_coeff') <= log2(b_coeff'). |
3915 | | * |
3916 | | * We arrive at a total upper bound: |
3917 | | * |
3918 | | * maxdigits(a_coeff') + maxdigits(1 / b_coeff') <= |
3919 | | * log10(a_coeff) + log2(b_coeff) = |
3920 | | * log10(a_coeff) + log10(b_coeff) / log10(2) <= |
3921 | | * a->digits + b->digits * 4; |
3922 | | */ |
3923 | 0 | uint32_t ystatus = 0; |
3924 | 0 | mpd_context_t workctx; |
3925 | 0 | mpd_workcontext(&workctx, ctx); |
3926 | |
|
3927 | 0 | workctx.prec = a->digits + b->digits * 4; |
3928 | 0 | if (workctx.prec >= ctx->prec) { |
3929 | 0 | *status |= (xstatus&MPD_Errors); |
3930 | 0 | goto out; /* No point in retrying, keep the original error. */ |
3931 | 0 | } |
3932 | | |
3933 | 0 | _mpd_qdiv(SET_IDEAL_EXP, q, a, b, &workctx, &ystatus); |
3934 | 0 | if (ystatus != 0) { |
3935 | 0 | ystatus = *status | ((ystatus|xstatus)&MPD_Errors); |
3936 | 0 | mpd_seterror(q, ystatus, status); |
3937 | 0 | } |
3938 | 0 | } |
3939 | 235 | else { |
3940 | 235 | *status |= xstatus; |
3941 | 235 | } |
3942 | | |
3943 | | |
3944 | 235 | out: |
3945 | 235 | mpd_del(&aa); |
3946 | 235 | mpd_del(&bb); |
3947 | 235 | } |
3948 | | |
3949 | | /* Internal function. */ |
3950 | | static void |
3951 | | _mpd_qdivmod(mpd_t *q, mpd_t *r, const mpd_t *a, const mpd_t *b, |
3952 | | const mpd_context_t *ctx, uint32_t *status) |
3953 | 4.62M | { |
3954 | 4.62M | MPD_NEW_STATIC(aligned,0,0,0,0); |
3955 | 4.62M | mpd_ssize_t qsize, rsize; |
3956 | 4.62M | mpd_ssize_t ideal_exp, expdiff, shift; |
3957 | 4.62M | uint8_t sign_a = mpd_sign(a); |
3958 | 4.62M | uint8_t sign_ab = mpd_sign(a)^mpd_sign(b); |
3959 | | |
3960 | | |
3961 | 4.62M | ideal_exp = (a->exp > b->exp) ? b->exp : a->exp; |
3962 | 4.62M | if (mpd_iszerocoeff(a)) { |
3963 | 29.1k | if (!mpd_qcopy(r, a, status)) { |
3964 | 0 | goto nanresult; /* GCOV_NOT_REACHED */ |
3965 | 0 | } |
3966 | 29.1k | r->exp = ideal_exp; |
3967 | 29.1k | _settriple(q, sign_ab, 0, 0); |
3968 | 29.1k | return; |
3969 | 29.1k | } |
3970 | | |
3971 | 4.59M | expdiff = mpd_adjexp(a) - mpd_adjexp(b); |
3972 | 4.59M | if (expdiff < 0) { |
3973 | 28.8k | if (a->exp > b->exp) { |
3974 | | /* positive and less than b->digits - a->digits */ |
3975 | 0 | shift = a->exp - b->exp; |
3976 | 0 | if (!mpd_qshiftl(r, a, shift, status)) { |
3977 | 0 | goto nanresult; |
3978 | 0 | } |
3979 | 0 | r->exp = ideal_exp; |
3980 | 0 | } |
3981 | 28.8k | else { |
3982 | 28.8k | if (!mpd_qcopy(r, a, status)) { |
3983 | 0 | goto nanresult; |
3984 | 0 | } |
3985 | 28.8k | } |
3986 | 28.8k | _settriple(q, sign_ab, 0, 0); |
3987 | 28.8k | return; |
3988 | 28.8k | } |
3989 | 4.56M | if (expdiff > ctx->prec) { |
3990 | 0 | *status |= MPD_Division_impossible; |
3991 | 0 | goto nanresult; |
3992 | 0 | } |
3993 | | |
3994 | | |
3995 | | /* |
3996 | | * At this point we have: |
3997 | | * (1) 0 <= a->exp + a->digits - b->exp - b->digits <= prec |
3998 | | * (2) a->exp - b->exp >= b->digits - a->digits |
3999 | | * (3) a->exp - b->exp <= prec + b->digits - a->digits |
4000 | | */ |
4001 | 4.56M | if (a->exp != b->exp) { |
4002 | 0 | shift = a->exp - b->exp; |
4003 | 0 | if (shift > 0) { |
4004 | | /* by (3), after the shift a->digits <= prec + b->digits */ |
4005 | 0 | if (!mpd_qshiftl(&aligned, a, shift, status)) { |
4006 | 0 | goto nanresult; |
4007 | 0 | } |
4008 | 0 | a = &aligned; |
4009 | 0 | } |
4010 | 0 | else { |
4011 | 0 | shift = -shift; |
4012 | | /* by (2), after the shift b->digits <= a->digits */ |
4013 | 0 | if (!mpd_qshiftl(&aligned, b, shift, status)) { |
4014 | 0 | goto nanresult; |
4015 | 0 | } |
4016 | 0 | b = &aligned; |
4017 | 0 | } |
4018 | 0 | } |
4019 | | |
4020 | | |
4021 | 4.56M | qsize = a->len - b->len + 1; |
4022 | 4.56M | if (!(q == a && qsize < a->len) && !(q == b && qsize < b->len)) { |
4023 | 4.56M | if (!mpd_qresize(q, qsize, status)) { |
4024 | 0 | goto nanresult; |
4025 | 0 | } |
4026 | 4.56M | } |
4027 | | |
4028 | 4.56M | rsize = b->len; |
4029 | 4.56M | if (!(r == a && rsize < a->len)) { |
4030 | 2.09M | if (!mpd_qresize(r, rsize, status)) { |
4031 | 0 | goto nanresult; |
4032 | 0 | } |
4033 | 2.09M | } |
4034 | | |
4035 | 4.56M | if (b->len == 1) { |
4036 | 1.91M | assert(b->data[0] != 0); /* annotation for scan-build */ |
4037 | 1.91M | if (a->len == 1) { |
4038 | 236k | _mpd_div_word(&q->data[0], &r->data[0], a->data[0], b->data[0]); |
4039 | 236k | } |
4040 | 1.67M | else { |
4041 | 1.67M | r->data[0] = _mpd_shortdiv(q->data, a->data, a->len, b->data[0]); |
4042 | 1.67M | } |
4043 | 1.91M | } |
4044 | 2.65M | else if (b->len <= MPD_NEWTONDIV_CUTOFF) { |
4045 | 2.65M | int ret; |
4046 | 2.65M | ret = _mpd_basedivmod(q->data, r->data, a->data, b->data, |
4047 | 2.65M | a->len, b->len); |
4048 | 2.65M | if (ret == -1) { |
4049 | 0 | *status |= MPD_Malloc_error; |
4050 | 0 | goto nanresult; |
4051 | 0 | } |
4052 | 2.65M | } |
4053 | 0 | else { |
4054 | 0 | _mpd_base_ndivmod(q, r, a, b, status); |
4055 | 0 | if (mpd_isspecial(q) || mpd_isspecial(r)) { |
4056 | 0 | goto nanresult; |
4057 | 0 | } |
4058 | 0 | qsize = q->len; |
4059 | 0 | rsize = r->len; |
4060 | 0 | } |
4061 | | |
4062 | 4.56M | qsize = _mpd_real_size(q->data, qsize); |
4063 | | /* resize to smaller cannot fail */ |
4064 | 4.56M | mpd_qresize(q, qsize, status); |
4065 | 4.56M | q->len = qsize; |
4066 | 4.56M | mpd_setdigits(q); |
4067 | 4.56M | mpd_set_flags(q, sign_ab); |
4068 | 4.56M | q->exp = 0; |
4069 | 4.56M | if (q->digits > ctx->prec) { |
4070 | 0 | *status |= MPD_Division_impossible; |
4071 | 0 | goto nanresult; |
4072 | 0 | } |
4073 | | |
4074 | 4.56M | rsize = _mpd_real_size(r->data, rsize); |
4075 | | /* resize to smaller cannot fail */ |
4076 | 4.56M | mpd_qresize(r, rsize, status); |
4077 | 4.56M | r->len = rsize; |
4078 | 4.56M | mpd_setdigits(r); |
4079 | 4.56M | mpd_set_flags(r, sign_a); |
4080 | 4.56M | r->exp = ideal_exp; |
4081 | | |
4082 | 4.56M | out: |
4083 | 4.56M | mpd_del(&aligned); |
4084 | 4.56M | return; |
4085 | | |
4086 | 0 | nanresult: |
4087 | 0 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
4088 | 0 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
4089 | 0 | goto out; |
4090 | 4.56M | } |
4091 | | |
4092 | | /* Integer division with remainder. */ |
4093 | | void |
4094 | | mpd_qdivmod(mpd_t *q, mpd_t *r, const mpd_t *a, const mpd_t *b, |
4095 | | const mpd_context_t *ctx, uint32_t *status) |
4096 | 0 | { |
4097 | 0 | uint8_t sign = mpd_sign(a)^mpd_sign(b); |
4098 | |
|
4099 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
4100 | 0 | if (mpd_qcheck_nans(q, a, b, ctx, status)) { |
4101 | 0 | mpd_qcopy(r, q, status); |
4102 | 0 | return; |
4103 | 0 | } |
4104 | 0 | if (mpd_isinfinite(a)) { |
4105 | 0 | if (mpd_isinfinite(b)) { |
4106 | 0 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
4107 | 0 | } |
4108 | 0 | else { |
4109 | 0 | mpd_setspecial(q, sign, MPD_INF); |
4110 | 0 | } |
4111 | 0 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
4112 | 0 | *status |= MPD_Invalid_operation; |
4113 | 0 | return; |
4114 | 0 | } |
4115 | 0 | if (mpd_isinfinite(b)) { |
4116 | 0 | if (!mpd_qcopy(r, a, status)) { |
4117 | 0 | mpd_seterror(q, MPD_Malloc_error, status); |
4118 | 0 | return; |
4119 | 0 | } |
4120 | 0 | mpd_qfinalize(r, ctx, status); |
4121 | 0 | _settriple(q, sign, 0, 0); |
4122 | 0 | return; |
4123 | 0 | } |
4124 | | /* debug */ |
4125 | 0 | abort(); /* GCOV_NOT_REACHED */ |
4126 | 0 | } |
4127 | 0 | if (mpd_iszerocoeff(b)) { |
4128 | 0 | if (mpd_iszerocoeff(a)) { |
4129 | 0 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
4130 | 0 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
4131 | 0 | *status |= MPD_Division_undefined; |
4132 | 0 | } |
4133 | 0 | else { |
4134 | 0 | mpd_setspecial(q, sign, MPD_INF); |
4135 | 0 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
4136 | 0 | *status |= (MPD_Division_by_zero|MPD_Invalid_operation); |
4137 | 0 | } |
4138 | 0 | return; |
4139 | 0 | } |
4140 | | |
4141 | 0 | _mpd_qdivmod(q, r, a, b, ctx, status); |
4142 | 0 | mpd_qfinalize(q, ctx, status); |
4143 | 0 | mpd_qfinalize(r, ctx, status); |
4144 | 0 | } |
4145 | | |
4146 | | void |
4147 | | mpd_qdivint(mpd_t *q, const mpd_t *a, const mpd_t *b, |
4148 | | const mpd_context_t *ctx, uint32_t *status) |
4149 | 1.84M | { |
4150 | 1.84M | MPD_NEW_STATIC(r,0,0,0,0); |
4151 | 1.84M | uint8_t sign = mpd_sign(a)^mpd_sign(b); |
4152 | | |
4153 | 1.84M | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
4154 | 0 | if (mpd_qcheck_nans(q, a, b, ctx, status)) { |
4155 | 0 | return; |
4156 | 0 | } |
4157 | 0 | if (mpd_isinfinite(a) && mpd_isinfinite(b)) { |
4158 | 0 | mpd_seterror(q, MPD_Invalid_operation, status); |
4159 | 0 | return; |
4160 | 0 | } |
4161 | 0 | if (mpd_isinfinite(a)) { |
4162 | 0 | mpd_setspecial(q, sign, MPD_INF); |
4163 | 0 | return; |
4164 | 0 | } |
4165 | 0 | if (mpd_isinfinite(b)) { |
4166 | 0 | _settriple(q, sign, 0, 0); |
4167 | 0 | return; |
4168 | 0 | } |
4169 | | /* debug */ |
4170 | 0 | abort(); /* GCOV_NOT_REACHED */ |
4171 | 0 | } |
4172 | 1.84M | if (mpd_iszerocoeff(b)) { |
4173 | 0 | if (mpd_iszerocoeff(a)) { |
4174 | 0 | mpd_seterror(q, MPD_Division_undefined, status); |
4175 | 0 | } |
4176 | 0 | else { |
4177 | 0 | mpd_setspecial(q, sign, MPD_INF); |
4178 | 0 | *status |= MPD_Division_by_zero; |
4179 | 0 | } |
4180 | 0 | return; |
4181 | 0 | } |
4182 | | |
4183 | | |
4184 | 1.84M | _mpd_qdivmod(q, &r, a, b, ctx, status); |
4185 | 1.84M | mpd_del(&r); |
4186 | 1.84M | mpd_qfinalize(q, ctx, status); |
4187 | 1.84M | } |
4188 | | |
4189 | | /* Divide decimal by mpd_ssize_t. */ |
4190 | | void |
4191 | | mpd_qdiv_ssize(mpd_t *result, const mpd_t *a, mpd_ssize_t b, |
4192 | | const mpd_context_t *ctx, uint32_t *status) |
4193 | 0 | { |
4194 | 0 | mpd_context_t maxcontext; |
4195 | 0 | MPD_NEW_STATIC(bb,0,0,0,0); |
4196 | |
|
4197 | 0 | mpd_maxcontext(&maxcontext); |
4198 | 0 | mpd_qsset_ssize(&bb, b, &maxcontext, status); |
4199 | 0 | mpd_qdiv(result, a, &bb, ctx, status); |
4200 | 0 | mpd_del(&bb); |
4201 | 0 | } |
4202 | | |
4203 | | /* Divide decimal by mpd_uint_t. */ |
4204 | | void |
4205 | | mpd_qdiv_uint(mpd_t *result, const mpd_t *a, mpd_uint_t b, |
4206 | | const mpd_context_t *ctx, uint32_t *status) |
4207 | 0 | { |
4208 | 0 | mpd_context_t maxcontext; |
4209 | 0 | MPD_NEW_STATIC(bb,0,0,0,0); |
4210 | |
|
4211 | 0 | mpd_maxcontext(&maxcontext); |
4212 | 0 | mpd_qsset_uint(&bb, b, &maxcontext, status); |
4213 | 0 | mpd_qdiv(result, a, &bb, ctx, status); |
4214 | 0 | mpd_del(&bb); |
4215 | 0 | } |
4216 | | |
4217 | | /* Divide decimal by int32_t. */ |
4218 | | void |
4219 | | mpd_qdiv_i32(mpd_t *result, const mpd_t *a, int32_t b, |
4220 | | const mpd_context_t *ctx, uint32_t *status) |
4221 | 0 | { |
4222 | 0 | mpd_qdiv_ssize(result, a, b, ctx, status); |
4223 | 0 | } |
4224 | | |
4225 | | /* Divide decimal by uint32_t. */ |
4226 | | void |
4227 | | mpd_qdiv_u32(mpd_t *result, const mpd_t *a, uint32_t b, |
4228 | | const mpd_context_t *ctx, uint32_t *status) |
4229 | 0 | { |
4230 | 0 | mpd_qdiv_uint(result, a, b, ctx, status); |
4231 | 0 | } |
4232 | | |
4233 | | #ifdef CONFIG_64 |
4234 | | /* Divide decimal by int64_t. */ |
4235 | | void |
4236 | | mpd_qdiv_i64(mpd_t *result, const mpd_t *a, int64_t b, |
4237 | | const mpd_context_t *ctx, uint32_t *status) |
4238 | 0 | { |
4239 | 0 | mpd_qdiv_ssize(result, a, b, ctx, status); |
4240 | 0 | } |
4241 | | |
4242 | | /* Divide decimal by uint64_t. */ |
4243 | | void |
4244 | | mpd_qdiv_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
4245 | | const mpd_context_t *ctx, uint32_t *status) |
4246 | 0 | { |
4247 | 0 | mpd_qdiv_uint(result, a, b, ctx, status); |
4248 | 0 | } |
4249 | | #elif !defined(LEGACY_COMPILER) |
4250 | | /* Divide decimal by int64_t. */ |
4251 | | void |
4252 | | mpd_qdiv_i64(mpd_t *result, const mpd_t *a, int64_t b, |
4253 | | const mpd_context_t *ctx, uint32_t *status) |
4254 | | { |
4255 | | mpd_context_t maxcontext; |
4256 | | MPD_NEW_STATIC(bb,0,0,0,0); |
4257 | | |
4258 | | mpd_maxcontext(&maxcontext); |
4259 | | mpd_qset_i64(&bb, b, &maxcontext, status); |
4260 | | mpd_qdiv(result, a, &bb, ctx, status); |
4261 | | mpd_del(&bb); |
4262 | | } |
4263 | | |
4264 | | /* Divide decimal by uint64_t. */ |
4265 | | void |
4266 | | mpd_qdiv_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
4267 | | const mpd_context_t *ctx, uint32_t *status) |
4268 | | { |
4269 | | mpd_context_t maxcontext; |
4270 | | MPD_NEW_STATIC(bb,0,0,0,0); |
4271 | | |
4272 | | mpd_maxcontext(&maxcontext); |
4273 | | mpd_qset_u64(&bb, b, &maxcontext, status); |
4274 | | mpd_qdiv(result, a, &bb, ctx, status); |
4275 | | mpd_del(&bb); |
4276 | | } |
4277 | | #endif |
4278 | | |
4279 | | /* Pad the result with trailing zeros if it has fewer digits than prec. */ |
4280 | | static void |
4281 | | _mpd_zeropad(mpd_t *result, const mpd_context_t *ctx, uint32_t *status) |
4282 | 0 | { |
4283 | 0 | if (!mpd_isspecial(result) && !mpd_iszero(result) && |
4284 | 0 | result->digits < ctx->prec) { |
4285 | 0 | mpd_ssize_t shift = ctx->prec - result->digits; |
4286 | 0 | mpd_qshiftl(result, result, shift, status); |
4287 | 0 | result->exp -= shift; |
4288 | 0 | } |
4289 | 0 | } |
4290 | | |
4291 | | /* Check if the result is guaranteed to be one. */ |
4292 | | static int |
4293 | | _mpd_qexp_check_one(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
4294 | | uint32_t *status) |
4295 | 0 | { |
4296 | 0 | MPD_NEW_CONST(lim,0,-(ctx->prec+1),1,1,1,9); |
4297 | 0 | MPD_NEW_SHARED(aa, a); |
4298 | |
|
4299 | 0 | mpd_set_positive(&aa); |
4300 | | |
4301 | | /* abs(a) <= 9 * 10**(-prec-1) */ |
4302 | 0 | if (_mpd_cmp(&aa, &lim) <= 0) { |
4303 | 0 | _settriple(result, 0, 1, 0); |
4304 | 0 | *status |= MPD_Rounded|MPD_Inexact; |
4305 | 0 | return 1; |
4306 | 0 | } |
4307 | | |
4308 | 0 | return 0; |
4309 | 0 | } |
4310 | | |
4311 | | /* |
4312 | | * Get the number of iterations for the Horner scheme in _mpd_qexp(). |
4313 | | */ |
4314 | | static inline mpd_ssize_t |
4315 | | _mpd_get_exp_iterations(const mpd_t *r, mpd_ssize_t p) |
4316 | 0 | { |
4317 | 0 | mpd_ssize_t log10pbyr; /* lower bound for log10(p / abs(r)) */ |
4318 | 0 | mpd_ssize_t n; |
4319 | |
|
4320 | 0 | assert(p >= 10); |
4321 | 0 | assert(!mpd_iszero(r)); |
4322 | 0 | assert(-p < mpd_adjexp(r) && mpd_adjexp(r) <= -1); |
4323 | |
|
4324 | 0 | #ifdef CONFIG_64 |
4325 | 0 | if (p > (mpd_ssize_t)(1ULL<<52)) { |
4326 | 0 | return MPD_SSIZE_MAX; |
4327 | 0 | } |
4328 | 0 | #endif |
4329 | | |
4330 | | /* |
4331 | | * Lower bound for log10(p / abs(r)): adjexp(p) - (adjexp(r) + 1) |
4332 | | * At this point (for CONFIG_64, CONFIG_32 is not problematic): |
4333 | | * 1) 10 <= p <= 2**52 |
4334 | | * 2) -p < adjexp(r) <= -1 |
4335 | | * 3) 1 <= log10pbyr <= 2**52 + 14 |
4336 | | */ |
4337 | 0 | log10pbyr = (mpd_word_digits(p)-1) - (mpd_adjexp(r)+1); |
4338 | | |
4339 | | /* |
4340 | | * The numerator in the paper is 1.435 * p - 1.182, calculated |
4341 | | * exactly. We compensate for rounding errors by using 1.43503. |
4342 | | * ACL2 proofs: |
4343 | | * 1) exp-iter-approx-lower-bound: The term below evaluated |
4344 | | * in 53-bit floating point arithmetic is greater than or |
4345 | | * equal to the exact term used in the paper. |
4346 | | * 2) exp-iter-approx-upper-bound: The term below is less than |
4347 | | * or equal to 3/2 * p <= 3/2 * 2**52. |
4348 | | */ |
4349 | 0 | n = (mpd_ssize_t)ceil((1.43503*(double)p - 1.182) / (double)log10pbyr); |
4350 | 0 | return n >= 3 ? n : 3; |
4351 | 0 | } |
4352 | | |
4353 | | /* |
4354 | | * Internal function, specials have been dealt with. Apart from Overflow |
4355 | | * and Underflow, two cases must be considered for the error of the result: |
4356 | | * |
4357 | | * 1) abs(a) <= 9 * 10**(-prec-1) ==> result == 1 |
4358 | | * |
4359 | | * Absolute error: abs(1 - e**x) < 10**(-prec) |
4360 | | * ------------------------------------------- |
4361 | | * |
4362 | | * 2) abs(a) > 9 * 10**(-prec-1) |
4363 | | * |
4364 | | * Relative error: abs(result - e**x) < 0.5 * 10**(-prec) * e**x |
4365 | | * ------------------------------------------------------------- |
4366 | | * |
4367 | | * The algorithm is from Hull&Abrham, Variable Precision Exponential Function, |
4368 | | * ACM Transactions on Mathematical Software, Vol. 12, No. 2, June 1986. |
4369 | | * |
4370 | | * Main differences: |
4371 | | * |
4372 | | * - The number of iterations for the Horner scheme is calculated using |
4373 | | * 53-bit floating point arithmetic. |
4374 | | * |
4375 | | * - In the error analysis for ER (relative error accumulated in the |
4376 | | * evaluation of the truncated series) the reduced operand r may |
4377 | | * have any number of digits. |
4378 | | * ACL2 proof: exponent-relative-error |
4379 | | * |
4380 | | * - The analysis for early abortion has been adapted for the mpd_t |
4381 | | * ranges. |
4382 | | */ |
4383 | | static int |
4384 | | _mpd_qexp(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
4385 | | uint32_t *status) |
4386 | 0 | { |
4387 | 0 | mpd_context_t workctx; |
4388 | 0 | MPD_NEW_STATIC(tmp,0,0,0,0); |
4389 | 0 | MPD_NEW_STATIC(sum,0,0,0,0); |
4390 | 0 | MPD_NEW_CONST(word,0,0,1,1,1,1); |
4391 | 0 | mpd_ssize_t j, n, t; |
4392 | |
|
4393 | 0 | assert(!mpd_isspecial(a)); |
4394 | |
|
4395 | 0 | if (mpd_iszerocoeff(a)) { |
4396 | 0 | _settriple(result, MPD_POS, 1, 0); |
4397 | 0 | return 1; |
4398 | 0 | } |
4399 | | |
4400 | | /* |
4401 | | * We are calculating e^x = e^(r*10^t) = (e^r)^(10^t), where abs(r) < 1 and t >= 0. |
4402 | | * |
4403 | | * If t > 0, we have: |
4404 | | * |
4405 | | * (1) 0.1 <= r < 1, so e^0.1 <= e^r. If t > MAX_T, overflow occurs: |
4406 | | * |
4407 | | * MAX-EMAX+1 < log10(e^(0.1*10*t)) <= log10(e^(r*10^t)) < adjexp(e^(r*10^t))+1 |
4408 | | * |
4409 | | * (2) -1 < r <= -0.1, so e^r <= e^-0.1. If t > MAX_T, underflow occurs: |
4410 | | * |
4411 | | * adjexp(e^(r*10^t)) <= log10(e^(r*10^t)) <= log10(e^(-0.1*10^t)) < MIN-ETINY |
4412 | | */ |
4413 | 0 | #if defined(CONFIG_64) |
4414 | 0 | #define MPD_EXP_MAX_T 19 |
4415 | | #elif defined(CONFIG_32) |
4416 | | #define MPD_EXP_MAX_T 10 |
4417 | | #endif |
4418 | 0 | t = a->digits + a->exp; |
4419 | 0 | t = (t > 0) ? t : 0; |
4420 | 0 | if (t > MPD_EXP_MAX_T) { |
4421 | 0 | if (mpd_ispositive(a)) { |
4422 | 0 | mpd_setspecial(result, MPD_POS, MPD_INF); |
4423 | 0 | *status |= MPD_Overflow|MPD_Inexact|MPD_Rounded; |
4424 | 0 | } |
4425 | 0 | else { |
4426 | 0 | _settriple(result, MPD_POS, 0, mpd_etiny(ctx)); |
4427 | 0 | *status |= (MPD_Inexact|MPD_Rounded|MPD_Subnormal| |
4428 | 0 | MPD_Underflow|MPD_Clamped); |
4429 | 0 | } |
4430 | 0 | return 1; |
4431 | 0 | } |
4432 | | |
4433 | | /* abs(a) <= 9 * 10**(-prec-1) */ |
4434 | 0 | if (_mpd_qexp_check_one(result, a, ctx, status)) { |
4435 | 0 | return 1; |
4436 | 0 | } |
4437 | | |
4438 | 0 | mpd_maxcontext(&workctx); |
4439 | 0 | workctx.prec = ctx->prec + t + 2; |
4440 | 0 | workctx.prec = (workctx.prec < 10) ? 10 : workctx.prec; |
4441 | 0 | workctx.round = MPD_ROUND_HALF_EVEN; |
4442 | |
|
4443 | 0 | if (!mpd_qcopy(result, a, status)) { |
4444 | 0 | return 1; |
4445 | 0 | } |
4446 | 0 | result->exp -= t; |
4447 | | |
4448 | | /* |
4449 | | * At this point: |
4450 | | * 1) 9 * 10**(-prec-1) < abs(a) |
4451 | | * 2) 9 * 10**(-prec-t-1) < abs(r) |
4452 | | * 3) log10(9) - prec - t - 1 < log10(abs(r)) < adjexp(abs(r)) + 1 |
4453 | | * 4) - prec - t - 2 < adjexp(abs(r)) <= -1 |
4454 | | */ |
4455 | 0 | n = _mpd_get_exp_iterations(result, workctx.prec); |
4456 | 0 | if (n == MPD_SSIZE_MAX) { |
4457 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); /* GCOV_UNLIKELY */ |
4458 | 0 | return 1; /* GCOV_UNLIKELY */ |
4459 | 0 | } |
4460 | | |
4461 | 0 | _settriple(&sum, MPD_POS, 1, 0); |
4462 | |
|
4463 | 0 | for (j = n-1; j >= 1; j--) { |
4464 | 0 | word.data[0] = j; |
4465 | 0 | mpd_setdigits(&word); |
4466 | 0 | mpd_qdiv(&tmp, result, &word, &workctx, &workctx.status); |
4467 | 0 | mpd_qfma(&sum, &sum, &tmp, &one, &workctx, &workctx.status); |
4468 | 0 | } |
4469 | |
|
4470 | 0 | #ifdef CONFIG_64 |
4471 | 0 | _mpd_qpow_uint(result, &sum, mpd_pow10[t], MPD_POS, &workctx, status); |
4472 | | #else |
4473 | | if (t <= MPD_MAX_POW10) { |
4474 | | _mpd_qpow_uint(result, &sum, mpd_pow10[t], MPD_POS, &workctx, status); |
4475 | | } |
4476 | | else { |
4477 | | t -= MPD_MAX_POW10; |
4478 | | _mpd_qpow_uint(&tmp, &sum, mpd_pow10[MPD_MAX_POW10], MPD_POS, |
4479 | | &workctx, status); |
4480 | | _mpd_qpow_uint(result, &tmp, mpd_pow10[t], MPD_POS, &workctx, status); |
4481 | | } |
4482 | | #endif |
4483 | |
|
4484 | 0 | mpd_del(&tmp); |
4485 | 0 | mpd_del(&sum); |
4486 | 0 | *status |= (workctx.status&MPD_Errors); |
4487 | 0 | *status |= (MPD_Inexact|MPD_Rounded); |
4488 | |
|
4489 | 0 | return 0; |
4490 | 0 | } |
4491 | | |
4492 | | /* exp(a) */ |
4493 | | void |
4494 | | mpd_qexp(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
4495 | | uint32_t *status) |
4496 | 0 | { |
4497 | 0 | mpd_context_t workctx; |
4498 | |
|
4499 | 0 | if (mpd_isspecial(a)) { |
4500 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
4501 | 0 | return; |
4502 | 0 | } |
4503 | 0 | if (mpd_isnegative(a)) { |
4504 | 0 | _settriple(result, MPD_POS, 0, 0); |
4505 | 0 | } |
4506 | 0 | else { |
4507 | 0 | mpd_setspecial(result, MPD_POS, MPD_INF); |
4508 | 0 | } |
4509 | 0 | return; |
4510 | 0 | } |
4511 | 0 | if (mpd_iszerocoeff(a)) { |
4512 | 0 | _settriple(result, MPD_POS, 1, 0); |
4513 | 0 | return; |
4514 | 0 | } |
4515 | | |
4516 | 0 | mpd_workcontext(&workctx, ctx); |
4517 | 0 | workctx.round = MPD_ROUND_HALF_EVEN; |
4518 | |
|
4519 | 0 | if (ctx->allcr) { |
4520 | 0 | MPD_NEW_STATIC(hi, 0,0,0,0); |
4521 | 0 | MPD_NEW_STATIC(lo, 0,0,0,0); |
4522 | 0 | MPD_NEW_STATIC(ulp, 0,0,0,0); |
4523 | 0 | MPD_NEW_STATIC(aa, 0,0,0,0); |
4524 | 0 | uint32_t loop_protect = 0; |
4525 | 0 | mpd_ssize_t prec; |
4526 | 0 | mpd_ssize_t ulpexp; |
4527 | |
|
4528 | 0 | if (result == a) { |
4529 | 0 | if (!mpd_qcopy(&aa, a, status)) { |
4530 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
4531 | 0 | return; |
4532 | 0 | } |
4533 | 0 | a = &aa; |
4534 | 0 | } |
4535 | | |
4536 | 0 | workctx.clamp = 0; |
4537 | 0 | prec = ctx->prec + 3; |
4538 | 0 | while (1) { |
4539 | 0 | uint32_t status_res = 0; |
4540 | 0 | uint32_t status_hi = 0; |
4541 | 0 | uint32_t status_lo = 0; |
4542 | 0 | int shortcut; |
4543 | 0 | int bounds_eq; |
4544 | 0 | int subnormal_eq; |
4545 | |
|
4546 | 0 | workctx.prec = prec; |
4547 | 0 | shortcut = _mpd_qexp(result, a, &workctx, &status_res); |
4548 | 0 | if (mpd_isnan(result)) { |
4549 | 0 | mpd_seterror(result, status_res, status); |
4550 | 0 | break; |
4551 | 0 | } |
4552 | 0 | if (shortcut || mpd_isinfinite(result) || mpd_iszero(result)) { |
4553 | 0 | *status |= status_res; |
4554 | 0 | workctx.prec = ctx->prec; |
4555 | 0 | workctx.clamp = ctx->clamp; |
4556 | 0 | _mpd_zeropad(result, &workctx, status); |
4557 | 0 | mpd_qfinalize(result, &workctx, status); |
4558 | 0 | break; |
4559 | 0 | } |
4560 | | |
4561 | 0 | ulpexp = result->exp + result->digits - workctx.prec; |
4562 | 0 | if (status_res & MPD_Underflow) { |
4563 | | /* The effective work precision is result->digits. */ |
4564 | 0 | ulpexp = result->exp; |
4565 | 0 | } |
4566 | 0 | _ssettriple(&ulp, MPD_POS, 1, ulpexp); |
4567 | | |
4568 | | /* |
4569 | | * At this point [1]: |
4570 | | * 1) abs(result - e**x) < 0.5 * 10**(-prec) * e**x |
4571 | | * 2) result - ulp < e**x < result + ulp |
4572 | | * 3) result - ulp < result < result + ulp |
4573 | | * |
4574 | | * If round(result-ulp)==round(result+ulp), then |
4575 | | * round(result)==round(e**x). Therefore the result |
4576 | | * is correctly rounded. |
4577 | | * |
4578 | | * [1] If abs(a) <= 9 * 10**(-prec-1), use the absolute |
4579 | | * error for a similar argument. |
4580 | | */ |
4581 | 0 | workctx.prec = ctx->prec; |
4582 | 0 | mpd_qadd(&hi, result, &ulp, &workctx, &status_hi); |
4583 | 0 | mpd_qsub(&lo, result, &ulp, &workctx, &status_lo); |
4584 | 0 | if (mpd_isnan(&hi) || mpd_isnan(&lo)) { |
4585 | 0 | mpd_seterror(result, status_hi|status_lo, status); |
4586 | 0 | break; |
4587 | 0 | } |
4588 | | |
4589 | 0 | subnormal_eq = (status_hi&MPD_Subnormal) == (status_lo&MPD_Subnormal); |
4590 | 0 | bounds_eq = mpd_qcmp(&hi, &lo, status) == 0; |
4591 | 0 | if (bounds_eq && ++loop_protect > 5) { |
4592 | | /* If the bounds are equal, the result is always correctly rounded. |
4593 | | |
4594 | | Resolving the subnormal status can take more iterations (around |
4595 | | three) in extremely rare cases. 'hi' and 'lo' are so close that |
4596 | | subnormal/underflow is largely cosmetic, so allow a maximum of |
4597 | | five additional iterations. */ |
4598 | 0 | subnormal_eq = 1; /* GCOV_NOT_REACHED */ |
4599 | 0 | } |
4600 | |
|
4601 | 0 | if (subnormal_eq && bounds_eq) { |
4602 | 0 | *status |= status_lo; |
4603 | 0 | workctx.clamp = ctx->clamp; |
4604 | 0 | _mpd_zeropad(result, &workctx, status); |
4605 | 0 | mpd_qfinalize(result, &workctx, status); |
4606 | 0 | break; |
4607 | 0 | } |
4608 | | |
4609 | 0 | if (subnormal_eq) { |
4610 | 0 | prec += MPD_RDIGITS; |
4611 | 0 | } |
4612 | 0 | else { |
4613 | 0 | prec *= 2; |
4614 | 0 | } |
4615 | 0 | if (prec > MPD_MAX_PREC) { |
4616 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
4617 | 0 | break; |
4618 | 0 | } |
4619 | 0 | } |
4620 | 0 | mpd_del(&hi); |
4621 | 0 | mpd_del(&lo); |
4622 | 0 | mpd_del(&ulp); |
4623 | 0 | mpd_del(&aa); |
4624 | 0 | } |
4625 | 0 | else { |
4626 | 0 | _mpd_qexp(result, a, &workctx, status); |
4627 | 0 | _mpd_zeropad(result, &workctx, status); |
4628 | 0 | mpd_check_underflow(result, &workctx, status); |
4629 | 0 | mpd_qfinalize(result, &workctx, status); |
4630 | 0 | } |
4631 | 0 | } |
4632 | | |
4633 | | /* Fused multiply-add: (a * b) + c, with a single final rounding. */ |
4634 | | void |
4635 | | mpd_qfma(mpd_t *result, const mpd_t *a, const mpd_t *b, const mpd_t *c, |
4636 | | const mpd_context_t *ctx, uint32_t *status) |
4637 | 45 | { |
4638 | 45 | uint32_t workstatus = 0; |
4639 | 45 | mpd_t *cc = NULL; |
4640 | | |
4641 | 45 | if (result == c) { |
4642 | 0 | if ((cc = mpd_qncopy(c)) == NULL) { |
4643 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
4644 | 0 | return; |
4645 | 0 | } |
4646 | 0 | c = cc; |
4647 | 0 | } |
4648 | | |
4649 | 45 | _mpd_qmul(result, a, b, ctx, &workstatus); |
4650 | 45 | if (!(workstatus&MPD_Invalid_operation)) { |
4651 | 45 | mpd_qadd(result, result, c, ctx, &workstatus); |
4652 | 45 | } |
4653 | | |
4654 | 45 | if (cc) mpd_del(cc); |
4655 | 45 | *status |= workstatus; |
4656 | 45 | } |
4657 | | |
4658 | | /* |
4659 | | * Schedule the optimal precision increase for the Newton iteration. |
4660 | | * v := input operand |
4661 | | * z_0 := initial approximation |
4662 | | * initprec := natural number such that abs(log(v) - z_0) < 10**-initprec |
4663 | | * maxprec := target precision |
4664 | | * |
4665 | | * For convenience the output klist contains the elements in reverse order: |
4666 | | * klist := [k_n-1, ..., k_0], where |
4667 | | * 1) k_0 <= initprec and |
4668 | | * 2) abs(log(v) - result) < 10**(-2*k_n-1 + 1) <= 10**-maxprec. |
4669 | | */ |
4670 | | static inline int |
4671 | | ln_schedule_prec(mpd_ssize_t klist[MPD_MAX_PREC_LOG2], mpd_ssize_t maxprec, |
4672 | | mpd_ssize_t initprec) |
4673 | 0 | { |
4674 | 0 | mpd_ssize_t k; |
4675 | 0 | int i; |
4676 | |
|
4677 | 0 | assert(maxprec >= 2 && initprec >= 2); |
4678 | 0 | if (maxprec <= initprec) return -1; |
4679 | | |
4680 | 0 | i = 0; k = maxprec; |
4681 | 0 | do { |
4682 | 0 | k = (k+2) / 2; |
4683 | 0 | klist[i++] = k; |
4684 | 0 | } while (k > initprec); |
4685 | |
|
4686 | 0 | return i-1; |
4687 | 0 | } |
4688 | | |
4689 | | /* The constants have been verified with both decimal.py and mpfr. */ |
4690 | | #ifdef CONFIG_64 |
4691 | | #if MPD_RDIGITS != 19 |
4692 | | #error "mpdecimal.c: MPD_RDIGITS must be 19." |
4693 | | #endif |
4694 | | static const mpd_uint_t mpd_ln10_data[MPD_MINALLOC_MAX] = { |
4695 | | 6983716328982174407ULL, 9089704281976336583ULL, 1515961135648465461ULL, |
4696 | | 4416816335727555703ULL, 2900988039194170265ULL, 2307925037472986509ULL, |
4697 | | 107598438319191292ULL, 3466624107184669231ULL, 4450099781311469159ULL, |
4698 | | 9807828059751193854ULL, 7713456862091670584ULL, 1492198849978748873ULL, |
4699 | | 6528728696511086257ULL, 2385392051446341972ULL, 8692180205189339507ULL, |
4700 | | 6518769751037497088ULL, 2375253577097505395ULL, 9095610299291824318ULL, |
4701 | | 982748238504564801ULL, 5438635917781170543ULL, 7547331541421808427ULL, |
4702 | | 752371033310119785ULL, 3171643095059950878ULL, 9785265383207606726ULL, |
4703 | | 2932258279850258550ULL, 5497347726624257094ULL, 2976979522110718264ULL, |
4704 | | 9221477656763693866ULL, 1979650047149510504ULL, 6674183485704422507ULL, |
4705 | | 9702766860595249671ULL, 9278096762712757753ULL, 9314848524948644871ULL, |
4706 | | 6826928280848118428ULL, 754403708474699401ULL, 230105703089634572ULL, |
4707 | | 1929203337658714166ULL, 7589402567763113569ULL, 4208241314695689016ULL, |
4708 | | 2922455440575892572ULL, 9356734206705811364ULL, 2684916746550586856ULL, |
4709 | | 644507064800027750ULL, 9476834636167921018ULL, 5659121373450747856ULL, |
4710 | | 2835522011480466371ULL, 6470806855677432162ULL, 7141748003688084012ULL, |
4711 | | 9619404400222105101ULL, 5504893431493939147ULL, 6674744042432743651ULL, |
4712 | | 2287698219886746543ULL, 7773262884616336622ULL, 1985283935053089653ULL, |
4713 | | 4680843799894826233ULL, 8168948290720832555ULL, 8067566662873690987ULL, |
4714 | | 6248633409525465082ULL, 9829834196778404228ULL, 3524802359972050895ULL, |
4715 | | 3327900967572609677ULL, 110148862877297603ULL, 179914546843642076ULL, |
4716 | | 2302585092994045684ULL |
4717 | | }; |
4718 | | #else |
4719 | | #if MPD_RDIGITS != 9 |
4720 | | #error "mpdecimal.c: MPD_RDIGITS must be 9." |
4721 | | #endif |
4722 | | static const mpd_uint_t mpd_ln10_data[MPD_MINALLOC_MAX] = { |
4723 | | 401682692UL, 708474699UL, 720754403UL, 30896345UL, 602301057UL, 765871416UL, |
4724 | | 192920333UL, 763113569UL, 589402567UL, 956890167UL, 82413146UL, 589257242UL, |
4725 | | 245544057UL, 811364292UL, 734206705UL, 868569356UL, 167465505UL, 775026849UL, |
4726 | | 706480002UL, 18064450UL, 636167921UL, 569476834UL, 734507478UL, 156591213UL, |
4727 | | 148046637UL, 283552201UL, 677432162UL, 470806855UL, 880840126UL, 417480036UL, |
4728 | | 210510171UL, 940440022UL, 939147961UL, 893431493UL, 436515504UL, 440424327UL, |
4729 | | 654366747UL, 821988674UL, 622228769UL, 884616336UL, 537773262UL, 350530896UL, |
4730 | | 319852839UL, 989482623UL, 468084379UL, 720832555UL, 168948290UL, 736909878UL, |
4731 | | 675666628UL, 546508280UL, 863340952UL, 404228624UL, 834196778UL, 508959829UL, |
4732 | | 23599720UL, 967735248UL, 96757260UL, 603332790UL, 862877297UL, 760110148UL, |
4733 | | 468436420UL, 401799145UL, 299404568UL, 230258509UL |
4734 | | }; |
4735 | | #endif |
4736 | | /* _mpd_ln10 is used directly for precisions smaller than MINALLOC_MAX*RDIGITS. |
4737 | | Otherwise, it serves as the initial approximation for calculating ln(10). */ |
4738 | | static const mpd_t _mpd_ln10 = { |
4739 | | MPD_STATIC|MPD_CONST_DATA, -(MPD_MINALLOC_MAX*MPD_RDIGITS-1), |
4740 | | MPD_MINALLOC_MAX*MPD_RDIGITS, MPD_MINALLOC_MAX, MPD_MINALLOC_MAX, |
4741 | | (mpd_uint_t *)mpd_ln10_data |
4742 | | }; |
4743 | | |
4744 | | /* |
4745 | | * Set 'result' to log(10). |
4746 | | * Ulp error: abs(result - log(10)) < ulp(log(10)) |
4747 | | * Relative error: abs(result - log(10)) < 5 * 10**-prec * log(10) |
4748 | | * |
4749 | | * NOTE: The relative error is not derived from the ulp error, but |
4750 | | * calculated separately using the fact that 23/10 < log(10) < 24/10. |
4751 | | */ |
4752 | | void |
4753 | | mpd_qln10(mpd_t *result, mpd_ssize_t prec, uint32_t *status) |
4754 | 0 | { |
4755 | 0 | mpd_context_t varcontext, maxcontext; |
4756 | 0 | MPD_NEW_STATIC(tmp, 0,0,0,0); |
4757 | 0 | MPD_NEW_CONST(static10, 0,0,2,1,1,10); |
4758 | 0 | mpd_ssize_t klist[MPD_MAX_PREC_LOG2]; |
4759 | 0 | mpd_uint_t rnd; |
4760 | 0 | mpd_ssize_t shift; |
4761 | 0 | int i; |
4762 | |
|
4763 | 0 | assert(prec >= 1); |
4764 | |
|
4765 | 0 | shift = MPD_MINALLOC_MAX*MPD_RDIGITS-prec; |
4766 | 0 | shift = shift < 0 ? 0 : shift; |
4767 | |
|
4768 | 0 | rnd = mpd_qshiftr(result, &_mpd_ln10, shift, status); |
4769 | 0 | if (rnd == MPD_UINT_MAX) { |
4770 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
4771 | 0 | return; |
4772 | 0 | } |
4773 | 0 | result->exp = -(result->digits-1); |
4774 | |
|
4775 | 0 | mpd_maxcontext(&maxcontext); |
4776 | 0 | if (prec < MPD_MINALLOC_MAX*MPD_RDIGITS) { |
4777 | 0 | maxcontext.prec = prec; |
4778 | 0 | _mpd_apply_round_excess(result, rnd, &maxcontext, status); |
4779 | 0 | *status |= (MPD_Inexact|MPD_Rounded); |
4780 | 0 | return; |
4781 | 0 | } |
4782 | | |
4783 | 0 | mpd_maxcontext(&varcontext); |
4784 | 0 | varcontext.round = MPD_ROUND_TRUNC; |
4785 | |
|
4786 | 0 | i = ln_schedule_prec(klist, prec+2, -result->exp); |
4787 | 0 | for (; i >= 0; i--) { |
4788 | 0 | varcontext.prec = 2*klist[i]+3; |
4789 | 0 | result->flags ^= MPD_NEG; |
4790 | 0 | _mpd_qexp(&tmp, result, &varcontext, status); |
4791 | 0 | result->flags ^= MPD_NEG; |
4792 | 0 | mpd_qmul(&tmp, &static10, &tmp, &varcontext, status); |
4793 | 0 | mpd_qsub(&tmp, &tmp, &one, &maxcontext, status); |
4794 | 0 | mpd_qadd(result, result, &tmp, &maxcontext, status); |
4795 | 0 | if (mpd_isspecial(result)) { |
4796 | 0 | break; |
4797 | 0 | } |
4798 | 0 | } |
4799 | |
|
4800 | 0 | mpd_del(&tmp); |
4801 | 0 | maxcontext.prec = prec; |
4802 | 0 | mpd_qfinalize(result, &maxcontext, status); |
4803 | 0 | } |
4804 | | |
4805 | | /* |
4806 | | * Initial approximations for the ln() iteration. The values have the |
4807 | | * following properties (established with both decimal.py and mpfr): |
4808 | | * |
4809 | | * Index 0 - 400, logarithms of x in [1.00, 5.00]: |
4810 | | * abs(lnapprox[i] * 10**-3 - log((i+100)/100)) < 10**-2 |
4811 | | * abs(lnapprox[i] * 10**-3 - log((i+1+100)/100)) < 10**-2 |
4812 | | * |
4813 | | * Index 401 - 899, logarithms of x in (0.500, 0.999]: |
4814 | | * abs(-lnapprox[i] * 10**-3 - log((i+100)/1000)) < 10**-2 |
4815 | | * abs(-lnapprox[i] * 10**-3 - log((i+1+100)/1000)) < 10**-2 |
4816 | | */ |
4817 | | static const uint16_t lnapprox[900] = { |
4818 | | /* index 0 - 400: log((i+100)/100) * 1000 */ |
4819 | | 0, 10, 20, 30, 39, 49, 58, 68, 77, 86, 95, 104, 113, 122, 131, 140, 148, 157, |
4820 | | 166, 174, 182, 191, 199, 207, 215, 223, 231, 239, 247, 255, 262, 270, 278, |
4821 | | 285, 293, 300, 308, 315, 322, 329, 336, 344, 351, 358, 365, 372, 378, 385, |
4822 | | 392, 399, 406, 412, 419, 425, 432, 438, 445, 451, 457, 464, 470, 476, 482, |
4823 | | 489, 495, 501, 507, 513, 519, 525, 531, 536, 542, 548, 554, 560, 565, 571, |
4824 | | 577, 582, 588, 593, 599, 604, 610, 615, 621, 626, 631, 637, 642, 647, 652, |
4825 | | 658, 663, 668, 673, 678, 683, 688, 693, 698, 703, 708, 713, 718, 723, 728, |
4826 | | 732, 737, 742, 747, 751, 756, 761, 766, 770, 775, 779, 784, 788, 793, 798, |
4827 | | 802, 806, 811, 815, 820, 824, 829, 833, 837, 842, 846, 850, 854, 859, 863, |
4828 | | 867, 871, 876, 880, 884, 888, 892, 896, 900, 904, 908, 912, 916, 920, 924, |
4829 | | 928, 932, 936, 940, 944, 948, 952, 956, 959, 963, 967, 971, 975, 978, 982, |
4830 | | 986, 990, 993, 997, 1001, 1004, 1008, 1012, 1015, 1019, 1022, 1026, 1030, |
4831 | | 1033, 1037, 1040, 1044, 1047, 1051, 1054, 1058, 1061, 1065, 1068, 1072, 1075, |
4832 | | 1078, 1082, 1085, 1089, 1092, 1095, 1099, 1102, 1105, 1109, 1112, 1115, 1118, |
4833 | | 1122, 1125, 1128, 1131, 1135, 1138, 1141, 1144, 1147, 1151, 1154, 1157, 1160, |
4834 | | 1163, 1166, 1169, 1172, 1176, 1179, 1182, 1185, 1188, 1191, 1194, 1197, 1200, |
4835 | | 1203, 1206, 1209, 1212, 1215, 1218, 1221, 1224, 1227, 1230, 1233, 1235, 1238, |
4836 | | 1241, 1244, 1247, 1250, 1253, 1256, 1258, 1261, 1264, 1267, 1270, 1273, 1275, |
4837 | | 1278, 1281, 1284, 1286, 1289, 1292, 1295, 1297, 1300, 1303, 1306, 1308, 1311, |
4838 | | 1314, 1316, 1319, 1322, 1324, 1327, 1330, 1332, 1335, 1338, 1340, 1343, 1345, |
4839 | | 1348, 1351, 1353, 1356, 1358, 1361, 1364, 1366, 1369, 1371, 1374, 1376, 1379, |
4840 | | 1381, 1384, 1386, 1389, 1391, 1394, 1396, 1399, 1401, 1404, 1406, 1409, 1411, |
4841 | | 1413, 1416, 1418, 1421, 1423, 1426, 1428, 1430, 1433, 1435, 1437, 1440, 1442, |
4842 | | 1445, 1447, 1449, 1452, 1454, 1456, 1459, 1461, 1463, 1466, 1468, 1470, 1472, |
4843 | | 1475, 1477, 1479, 1482, 1484, 1486, 1488, 1491, 1493, 1495, 1497, 1500, 1502, |
4844 | | 1504, 1506, 1509, 1511, 1513, 1515, 1517, 1520, 1522, 1524, 1526, 1528, 1530, |
4845 | | 1533, 1535, 1537, 1539, 1541, 1543, 1545, 1548, 1550, 1552, 1554, 1556, 1558, |
4846 | | 1560, 1562, 1564, 1567, 1569, 1571, 1573, 1575, 1577, 1579, 1581, 1583, 1585, |
4847 | | 1587, 1589, 1591, 1593, 1595, 1597, 1599, 1601, 1603, 1605, 1607, 1609, |
4848 | | /* index 401 - 899: -log((i+100)/1000) * 1000 */ |
4849 | | 691, 689, 687, 685, 683, 681, 679, 677, 675, 673, 671, 669, 668, 666, 664, |
4850 | | 662, 660, 658, 656, 654, 652, 650, 648, 646, 644, 642, 641, 639, 637, 635, |
4851 | | 633, 631, 629, 627, 626, 624, 622, 620, 618, 616, 614, 612, 611, 609, 607, |
4852 | | 605, 603, 602, 600, 598, 596, 594, 592, 591, 589, 587, 585, 583, 582, 580, |
4853 | | 578, 576, 574, 573, 571, 569, 567, 566, 564, 562, 560, 559, 557, 555, 553, |
4854 | | 552, 550, 548, 546, 545, 543, 541, 540, 538, 536, 534, 533, 531, 529, 528, |
4855 | | 526, 524, 523, 521, 519, 518, 516, 514, 512, 511, 509, 508, 506, 504, 502, |
4856 | | 501, 499, 498, 496, 494, 493, 491, 489, 488, 486, 484, 483, 481, 480, 478, |
4857 | | 476, 475, 473, 472, 470, 468, 467, 465, 464, 462, 460, 459, 457, 456, 454, |
4858 | | 453, 451, 449, 448, 446, 445, 443, 442, 440, 438, 437, 435, 434, 432, 431, |
4859 | | 429, 428, 426, 425, 423, 422, 420, 419, 417, 416, 414, 412, 411, 410, 408, |
4860 | | 406, 405, 404, 402, 400, 399, 398, 396, 394, 393, 392, 390, 389, 387, 386, |
4861 | | 384, 383, 381, 380, 378, 377, 375, 374, 372, 371, 370, 368, 367, 365, 364, |
4862 | | 362, 361, 360, 358, 357, 355, 354, 352, 351, 350, 348, 347, 345, 344, 342, |
4863 | | 341, 340, 338, 337, 336, 334, 333, 331, 330, 328, 327, 326, 324, 323, 322, |
4864 | | 320, 319, 318, 316, 315, 313, 312, 311, 309, 308, 306, 305, 304, 302, 301, |
4865 | | 300, 298, 297, 296, 294, 293, 292, 290, 289, 288, 286, 285, 284, 282, 281, |
4866 | | 280, 278, 277, 276, 274, 273, 272, 270, 269, 268, 267, 265, 264, 263, 261, |
4867 | | 260, 259, 258, 256, 255, 254, 252, 251, 250, 248, 247, 246, 245, 243, 242, |
4868 | | 241, 240, 238, 237, 236, 234, 233, 232, 231, 229, 228, 227, 226, 224, 223, |
4869 | | 222, 221, 219, 218, 217, 216, 214, 213, 212, 211, 210, 208, 207, 206, 205, |
4870 | | 203, 202, 201, 200, 198, 197, 196, 195, 194, 192, 191, 190, 189, 188, 186, |
4871 | | 185, 184, 183, 182, 180, 179, 178, 177, 176, 174, 173, 172, 171, 170, 168, |
4872 | | 167, 166, 165, 164, 162, 161, 160, 159, 158, 157, 156, 154, 153, 152, 151, |
4873 | | 150, 148, 147, 146, 145, 144, 143, 142, 140, 139, 138, 137, 136, 135, 134, |
4874 | | 132, 131, 130, 129, 128, 127, 126, 124, 123, 122, 121, 120, 119, 118, 116, |
4875 | | 115, 114, 113, 112, 111, 110, 109, 108, 106, 105, 104, 103, 102, 101, 100, |
4876 | | 99, 98, 97, 95, 94, 93, 92, 91, 90, 89, 88, 87, 86, 84, 83, 82, 81, 80, 79, |
4877 | | 78, 77, 76, 75, 74, 73, 72, 70, 69, 68, 67, 66, 65, 64, 63, 62, 61, 60, 59, |
4878 | | 58, 57, 56, 54, 53, 52, 51, 50, 49, 48, 47, 46, 45, 44, 43, 42, 41, 40, 39, |
4879 | | 38, 37, 36, 35, 34, 33, 31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, |
4880 | | 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1 |
4881 | | }; |
4882 | | |
4883 | | /* |
4884 | | * Internal ln() function that does not check for specials, zero or one. |
4885 | | * Relative error: abs(result - log(a)) < 0.1 * 10**-prec * abs(log(a)) |
4886 | | */ |
4887 | | static int |
4888 | | _mpd_qln(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
4889 | | uint32_t *status) |
4890 | 0 | { |
4891 | 0 | mpd_context_t varcontext, maxcontext; |
4892 | 0 | mpd_t *z = result; |
4893 | 0 | MPD_NEW_STATIC(v,0,0,0,0); |
4894 | 0 | MPD_NEW_STATIC(vtmp,0,0,0,0); |
4895 | 0 | MPD_NEW_STATIC(tmp,0,0,0,0); |
4896 | 0 | mpd_ssize_t klist[MPD_MAX_PREC_LOG2]; |
4897 | 0 | mpd_ssize_t maxprec, shift, t; |
4898 | 0 | mpd_ssize_t a_digits, a_exp; |
4899 | 0 | mpd_uint_t dummy, x; |
4900 | 0 | int ret = 1; |
4901 | 0 | int i; |
4902 | |
|
4903 | 0 | assert(!mpd_isspecial(a) && !mpd_iszerocoeff(a)); |
4904 | | |
4905 | | /* |
4906 | | * We are calculating ln(a) = ln(v * 10^t) = ln(v) + t*ln(10), |
4907 | | * where 0.5 < v <= 5. |
4908 | | */ |
4909 | 0 | if (!mpd_qcopy(&v, a, status)) { |
4910 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
4911 | 0 | goto finish; |
4912 | 0 | } |
4913 | | |
4914 | | /* Initial approximation: we have at least one non-zero digit */ |
4915 | 0 | _mpd_get_msdigits(&dummy, &x, &v, 3); |
4916 | 0 | if (x < 10) x *= 10; |
4917 | 0 | if (x < 100) x *= 10; |
4918 | 0 | x -= 100; |
4919 | | |
4920 | | /* a may equal z */ |
4921 | 0 | a_digits = a->digits; |
4922 | 0 | a_exp = a->exp; |
4923 | |
|
4924 | 0 | mpd_minalloc(z); |
4925 | 0 | mpd_clear_flags(z); |
4926 | 0 | z->data[0] = lnapprox[x]; |
4927 | 0 | z->len = 1; |
4928 | 0 | z->exp = -3; |
4929 | 0 | mpd_setdigits(z); |
4930 | |
|
4931 | 0 | if (x <= 400) { |
4932 | | /* Reduce the input operand to 1.00 <= v <= 5.00. Let y = x + 100, |
4933 | | * so 100 <= y <= 500. Since y contains the most significant digits |
4934 | | * of v, y/100 <= v < (y+1)/100 and abs(z - log(v)) < 10**-2. */ |
4935 | 0 | v.exp = -(a_digits - 1); |
4936 | 0 | t = a_exp + a_digits - 1; |
4937 | 0 | } |
4938 | 0 | else { |
4939 | | /* Reduce the input operand to 0.500 < v <= 0.999. Let y = x + 100, |
4940 | | * so 500 < y <= 999. Since y contains the most significant digits |
4941 | | * of v, y/1000 <= v < (y+1)/1000 and abs(z - log(v)) < 10**-2. */ |
4942 | 0 | v.exp = -a_digits; |
4943 | 0 | t = a_exp + a_digits; |
4944 | 0 | mpd_set_negative(z); |
4945 | 0 | } |
4946 | |
|
4947 | 0 | mpd_maxcontext(&maxcontext); |
4948 | 0 | mpd_maxcontext(&varcontext); |
4949 | 0 | varcontext.round = MPD_ROUND_TRUNC; |
4950 | |
|
4951 | 0 | maxprec = ctx->prec + 2; |
4952 | 0 | if (t == 0 && (x <= 15 || x >= 800)) { |
4953 | | /* 0.900 <= v <= 1.15: Estimate the magnitude of the logarithm. |
4954 | | * If ln(v) will underflow, skip the loop. Otherwise, adjust the |
4955 | | * precision upwards in order to obtain a sufficient number of |
4956 | | * significant digits. |
4957 | | * |
4958 | | * Case v > 1: |
4959 | | * abs((v-1)/10) < abs((v-1)/v) < abs(ln(v)) < abs(v-1) |
4960 | | * Case v < 1: |
4961 | | * abs(v-1) < abs(ln(v)) < abs((v-1)/v) < abs((v-1)*10) |
4962 | | */ |
4963 | 0 | int cmp = _mpd_cmp(&v, &one); |
4964 | | |
4965 | | /* Upper bound (assume v > 1): abs(v-1), unrounded */ |
4966 | 0 | _mpd_qsub(&tmp, &v, &one, &maxcontext, &maxcontext.status); |
4967 | 0 | if (maxcontext.status & MPD_Errors) { |
4968 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
4969 | 0 | goto finish; |
4970 | 0 | } |
4971 | | |
4972 | 0 | if (cmp < 0) { |
4973 | | /* v < 1: abs((v-1)*10) */ |
4974 | 0 | tmp.exp += 1; |
4975 | 0 | } |
4976 | 0 | if (mpd_adjexp(&tmp) < mpd_etiny(ctx)) { |
4977 | | /* The upper bound is less than etiny: Underflow to zero */ |
4978 | 0 | _settriple(result, (cmp<0), 1, mpd_etiny(ctx)-1); |
4979 | 0 | goto finish; |
4980 | 0 | } |
4981 | | /* Lower bound: abs((v-1)/10) or abs(v-1) */ |
4982 | 0 | tmp.exp -= 1; |
4983 | 0 | if (mpd_adjexp(&tmp) < 0) { |
4984 | | /* Absolute error of the loop: abs(z - log(v)) < 10**-p. If |
4985 | | * p = ctx->prec+2-adjexp(lower), then the relative error of |
4986 | | * the result is (using 10**adjexp(x) <= abs(x)): |
4987 | | * |
4988 | | * abs(z - log(v)) / abs(log(v)) < 10**-p / abs(log(v)) |
4989 | | * <= 10**(-ctx->prec-2) |
4990 | | */ |
4991 | 0 | maxprec = maxprec - mpd_adjexp(&tmp); |
4992 | 0 | } |
4993 | 0 | } |
4994 | | |
4995 | 0 | i = ln_schedule_prec(klist, maxprec, 2); |
4996 | 0 | for (; i >= 0; i--) { |
4997 | 0 | varcontext.prec = 2*klist[i]+3; |
4998 | 0 | z->flags ^= MPD_NEG; |
4999 | 0 | _mpd_qexp(&tmp, z, &varcontext, status); |
5000 | 0 | z->flags ^= MPD_NEG; |
5001 | |
|
5002 | 0 | if (v.digits > varcontext.prec) { |
5003 | 0 | shift = v.digits - varcontext.prec; |
5004 | 0 | mpd_qshiftr(&vtmp, &v, shift, status); |
5005 | 0 | vtmp.exp += shift; |
5006 | 0 | mpd_qmul(&tmp, &vtmp, &tmp, &varcontext, status); |
5007 | 0 | } |
5008 | 0 | else { |
5009 | 0 | mpd_qmul(&tmp, &v, &tmp, &varcontext, status); |
5010 | 0 | } |
5011 | |
|
5012 | 0 | mpd_qsub(&tmp, &tmp, &one, &maxcontext, status); |
5013 | 0 | mpd_qadd(z, z, &tmp, &maxcontext, status); |
5014 | 0 | if (mpd_isspecial(z)) { |
5015 | 0 | break; |
5016 | 0 | } |
5017 | 0 | } |
5018 | | |
5019 | | /* |
5020 | | * Case t == 0: |
5021 | | * t * log(10) == 0, the result does not change and the analysis |
5022 | | * above applies. If v < 0.900 or v > 1.15, the relative error is |
5023 | | * less than 10**(-ctx.prec-1). |
5024 | | * Case t != 0: |
5025 | | * z := approx(log(v)) |
5026 | | * y := approx(log(10)) |
5027 | | * p := maxprec = ctx->prec + 2 |
5028 | | * Absolute errors: |
5029 | | * 1) abs(z - log(v)) < 10**-p |
5030 | | * 2) abs(y - log(10)) < 10**-p |
5031 | | * The multiplication is exact, so: |
5032 | | * 3) abs(t*y - t*log(10)) < t*10**-p |
5033 | | * The sum is exact, so: |
5034 | | * 4) abs((z + t*y) - (log(v) + t*log(10))) < (abs(t) + 1) * 10**-p |
5035 | | * Bounds for log(v) and log(10): |
5036 | | * 5) -7/10 < log(v) < 17/10 |
5037 | | * 6) 23/10 < log(10) < 24/10 |
5038 | | * Using 4), 5), 6) and t != 0, the relative error is: |
5039 | | * |
5040 | | * 7) relerr < ((abs(t) + 1)*10**-p) / abs(log(v) + t*log(10)) |
5041 | | * < 0.5 * 10**(-p + 1) = 0.5 * 10**(-ctx->prec-1) |
5042 | | */ |
5043 | 0 | mpd_qln10(&v, maxprec+1, status); |
5044 | 0 | mpd_qmul_ssize(&tmp, &v, t, &maxcontext, status); |
5045 | 0 | mpd_qadd(result, &tmp, z, &maxcontext, status); |
5046 | |
|
5047 | 0 | ret = 0; |
5048 | |
|
5049 | 0 | finish: |
5050 | 0 | *status |= (MPD_Inexact|MPD_Rounded); |
5051 | 0 | mpd_del(&v); |
5052 | 0 | mpd_del(&vtmp); |
5053 | 0 | mpd_del(&tmp); |
5054 | 0 | return ret; |
5055 | 0 | } |
5056 | | |
5057 | | /* ln(a) */ |
5058 | | void |
5059 | | mpd_qln(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
5060 | | uint32_t *status) |
5061 | 0 | { |
5062 | 0 | mpd_context_t workctx; |
5063 | 0 | mpd_ssize_t adjexp, t; |
5064 | |
|
5065 | 0 | if (mpd_isspecial(a)) { |
5066 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
5067 | 0 | return; |
5068 | 0 | } |
5069 | 0 | if (mpd_isnegative(a)) { |
5070 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
5071 | 0 | return; |
5072 | 0 | } |
5073 | 0 | mpd_setspecial(result, MPD_POS, MPD_INF); |
5074 | 0 | return; |
5075 | 0 | } |
5076 | 0 | if (mpd_iszerocoeff(a)) { |
5077 | 0 | mpd_setspecial(result, MPD_NEG, MPD_INF); |
5078 | 0 | return; |
5079 | 0 | } |
5080 | 0 | if (mpd_isnegative(a)) { |
5081 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
5082 | 0 | return; |
5083 | 0 | } |
5084 | 0 | if (_mpd_cmp(a, &one) == 0) { |
5085 | 0 | _settriple(result, MPD_POS, 0, 0); |
5086 | 0 | return; |
5087 | 0 | } |
5088 | | /* |
5089 | | * Check if the result will overflow (0 < x, x != 1): |
5090 | | * 1) log10(x) < 0 iff adjexp(x) < 0 |
5091 | | * 2) 0 < x /\ x <= y ==> adjexp(x) <= adjexp(y) |
5092 | | * 3) 0 < x /\ x != 1 ==> 2 * abs(log10(x)) < abs(log(x)) |
5093 | | * 4) adjexp(x) <= log10(x) < adjexp(x) + 1 |
5094 | | * |
5095 | | * Case adjexp(x) >= 0: |
5096 | | * 5) 2 * adjexp(x) < abs(log(x)) |
5097 | | * Case adjexp(x) > 0: |
5098 | | * 6) adjexp(2 * adjexp(x)) <= adjexp(abs(log(x))) |
5099 | | * Case adjexp(x) == 0: |
5100 | | * mpd_exp_digits(t)-1 == 0 <= emax (the shortcut is not triggered) |
5101 | | * |
5102 | | * Case adjexp(x) < 0: |
5103 | | * 7) 2 * (-adjexp(x) - 1) < abs(log(x)) |
5104 | | * Case adjexp(x) < -1: |
5105 | | * 8) adjexp(2 * (-adjexp(x) - 1)) <= adjexp(abs(log(x))) |
5106 | | * Case adjexp(x) == -1: |
5107 | | * mpd_exp_digits(t)-1 == 0 <= emax (the shortcut is not triggered) |
5108 | | */ |
5109 | 0 | adjexp = mpd_adjexp(a); |
5110 | 0 | t = (adjexp < 0) ? -adjexp-1 : adjexp; |
5111 | 0 | t *= 2; |
5112 | 0 | if (mpd_exp_digits(t)-1 > ctx->emax) { |
5113 | 0 | *status |= MPD_Overflow|MPD_Inexact|MPD_Rounded; |
5114 | 0 | mpd_setspecial(result, (adjexp<0), MPD_INF); |
5115 | 0 | return; |
5116 | 0 | } |
5117 | | |
5118 | 0 | mpd_workcontext(&workctx, ctx); |
5119 | 0 | workctx.round = MPD_ROUND_HALF_EVEN; |
5120 | |
|
5121 | 0 | if (ctx->allcr) { |
5122 | 0 | MPD_NEW_STATIC(hi, 0,0,0,0); |
5123 | 0 | MPD_NEW_STATIC(lo, 0,0,0,0); |
5124 | 0 | MPD_NEW_STATIC(ulp, 0,0,0,0); |
5125 | 0 | MPD_NEW_STATIC(aa, 0,0,0,0); |
5126 | 0 | uint32_t loop_protect = 0; |
5127 | 0 | mpd_ssize_t prec; |
5128 | |
|
5129 | 0 | if (result == a) { |
5130 | 0 | if (!mpd_qcopy(&aa, a, status)) { |
5131 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
5132 | 0 | return; |
5133 | 0 | } |
5134 | 0 | a = &aa; |
5135 | 0 | } |
5136 | | |
5137 | 0 | workctx.clamp = 0; |
5138 | 0 | prec = ctx->prec + 3; |
5139 | 0 | while (1) { |
5140 | 0 | uint32_t status_res = 0; |
5141 | 0 | uint32_t status_hi = 0; |
5142 | 0 | uint32_t status_lo = 0; |
5143 | 0 | int shortcut; |
5144 | 0 | int bounds_eq; |
5145 | 0 | int subnormal_eq; |
5146 | |
|
5147 | 0 | workctx.prec = prec; |
5148 | 0 | shortcut = _mpd_qln(result, a, &workctx, &status_res); |
5149 | 0 | if (mpd_isnan(result)) { |
5150 | 0 | mpd_seterror(result, status_res, status); |
5151 | 0 | break; |
5152 | 0 | } |
5153 | 0 | if (shortcut || mpd_isinfinite(result) || mpd_iszero(result)) { |
5154 | 0 | *status |= status_res; |
5155 | 0 | workctx.prec = ctx->prec; |
5156 | 0 | workctx.clamp = ctx->clamp; |
5157 | 0 | mpd_qfinalize(result, &workctx, status); |
5158 | 0 | break; |
5159 | 0 | } |
5160 | | |
5161 | 0 | _ssettriple(&ulp, MPD_POS, 1, |
5162 | 0 | result->exp + result->digits-workctx.prec); |
5163 | |
|
5164 | 0 | workctx.prec = ctx->prec; |
5165 | 0 | mpd_qadd(&hi, result, &ulp, &workctx, &status_hi); |
5166 | 0 | mpd_qsub(&lo, result, &ulp, &workctx, &status_lo); |
5167 | 0 | if (mpd_isnan(&hi) || mpd_isnan(&lo)) { |
5168 | 0 | mpd_seterror(result, status_hi|status_lo, status); |
5169 | 0 | break; |
5170 | 0 | } |
5171 | 0 | subnormal_eq = (status_hi&MPD_Subnormal) == (status_lo&MPD_Subnormal); |
5172 | 0 | bounds_eq = mpd_qcmp(&hi, &lo, status) == 0; |
5173 | 0 | if (bounds_eq && ++loop_protect > 5) { |
5174 | | /* If the bounds are equal, the result is always correctly rounded. |
5175 | | |
5176 | | Resolving the subnormal status can take more iterations (around |
5177 | | three) in extremely rare cases. 'hi' and 'lo' are so close that |
5178 | | subnormal/underflow is largely cosmetic, so allow a maximum of |
5179 | | five additional iterations. */ |
5180 | 0 | subnormal_eq = 1; /* GCOV_NOT_REACHED */ |
5181 | 0 | } |
5182 | |
|
5183 | 0 | if (subnormal_eq && bounds_eq) { |
5184 | 0 | *status |= status_lo; |
5185 | 0 | workctx.clamp = ctx->clamp; |
5186 | 0 | mpd_qfinalize(result, &workctx, status); |
5187 | 0 | break; |
5188 | 0 | } |
5189 | | |
5190 | 0 | if (subnormal_eq) { |
5191 | 0 | prec += MPD_RDIGITS; |
5192 | 0 | } |
5193 | 0 | else { |
5194 | 0 | prec *= 2; |
5195 | 0 | } |
5196 | |
|
5197 | 0 | if (prec > MPD_MAX_PREC) { |
5198 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
5199 | 0 | break; |
5200 | 0 | } |
5201 | 0 | } |
5202 | 0 | mpd_del(&hi); |
5203 | 0 | mpd_del(&lo); |
5204 | 0 | mpd_del(&ulp); |
5205 | 0 | mpd_del(&aa); |
5206 | 0 | } |
5207 | 0 | else { |
5208 | 0 | _mpd_qln(result, a, &workctx, status); |
5209 | 0 | mpd_check_underflow(result, &workctx, status); |
5210 | 0 | mpd_qfinalize(result, &workctx, status); |
5211 | 0 | } |
5212 | 0 | } |
5213 | | |
5214 | | /* |
5215 | | * Internal log10() function that does not check for specials, zero or one. |
5216 | | * Case SKIP_FINALIZE: |
5217 | | * Relative error: abs(result - log10(a)) < 0.1 * 10**-prec * abs(log10(a)) |
5218 | | * Case DO_FINALIZE: |
5219 | | * Ulp error: abs(result - log10(a)) < ulp(log10(a)) |
5220 | | */ |
5221 | | enum {SKIP_FINALIZE, DO_FINALIZE}; |
5222 | | static int |
5223 | | _mpd_qlog10(int action, mpd_t *result, const mpd_t *a, |
5224 | | const mpd_context_t *ctx, uint32_t *status) |
5225 | 0 | { |
5226 | 0 | mpd_context_t workctx; |
5227 | 0 | MPD_NEW_STATIC(ln10,0,0,0,0); |
5228 | 0 | int ret; |
5229 | |
|
5230 | 0 | mpd_maxcontext(&workctx); |
5231 | 0 | workctx.prec = ctx->prec + 3; |
5232 | | /* relative error: 0.1 * 10**(-p-3). The specific underflow shortcut |
5233 | | * in _mpd_qln() does not change the final result. */ |
5234 | 0 | ret = _mpd_qln(result, a, &workctx, status); |
5235 | | /* relative error: 5 * 10**(-p-3) */ |
5236 | 0 | mpd_qln10(&ln10, workctx.prec, status); |
5237 | |
|
5238 | 0 | if (action == DO_FINALIZE) { |
5239 | 0 | mpd_workcontext(&workctx, ctx); |
5240 | 0 | workctx.round = MPD_ROUND_HALF_EVEN; |
5241 | 0 | } |
5242 | | /* SKIP_FINALIZE: relative error: 5 * 10**(-p-3) */ |
5243 | 0 | _mpd_qdiv(NO_IDEAL_EXP, result, result, &ln10, &workctx, status); |
5244 | |
|
5245 | 0 | mpd_del(&ln10); |
5246 | 0 | return ret; |
5247 | 0 | } |
5248 | | |
5249 | | /* log10(a) */ |
5250 | | void |
5251 | | mpd_qlog10(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
5252 | | uint32_t *status) |
5253 | 0 | { |
5254 | 0 | mpd_context_t workctx; |
5255 | 0 | mpd_ssize_t adjexp, t; |
5256 | |
|
5257 | 0 | mpd_workcontext(&workctx, ctx); |
5258 | 0 | workctx.round = MPD_ROUND_HALF_EVEN; |
5259 | |
|
5260 | 0 | if (mpd_isspecial(a)) { |
5261 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
5262 | 0 | return; |
5263 | 0 | } |
5264 | 0 | if (mpd_isnegative(a)) { |
5265 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
5266 | 0 | return; |
5267 | 0 | } |
5268 | 0 | mpd_setspecial(result, MPD_POS, MPD_INF); |
5269 | 0 | return; |
5270 | 0 | } |
5271 | 0 | if (mpd_iszerocoeff(a)) { |
5272 | 0 | mpd_setspecial(result, MPD_NEG, MPD_INF); |
5273 | 0 | return; |
5274 | 0 | } |
5275 | 0 | if (mpd_isnegative(a)) { |
5276 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
5277 | 0 | return; |
5278 | 0 | } |
5279 | 0 | if (mpd_coeff_ispow10(a)) { |
5280 | 0 | uint8_t sign = 0; |
5281 | 0 | adjexp = mpd_adjexp(a); |
5282 | 0 | if (adjexp < 0) { |
5283 | 0 | sign = 1; |
5284 | 0 | adjexp = -adjexp; |
5285 | 0 | } |
5286 | 0 | _settriple(result, sign, adjexp, 0); |
5287 | 0 | mpd_qfinalize(result, &workctx, status); |
5288 | 0 | return; |
5289 | 0 | } |
5290 | | /* |
5291 | | * Check if the result will overflow (0 < x, x != 1): |
5292 | | * 1) log10(x) < 0 iff adjexp(x) < 0 |
5293 | | * 2) 0 < x /\ x <= y ==> adjexp(x) <= adjexp(y) |
5294 | | * 3) adjexp(x) <= log10(x) < adjexp(x) + 1 |
5295 | | * |
5296 | | * Case adjexp(x) >= 0: |
5297 | | * 4) adjexp(x) <= abs(log10(x)) |
5298 | | * Case adjexp(x) > 0: |
5299 | | * 5) adjexp(adjexp(x)) <= adjexp(abs(log10(x))) |
5300 | | * Case adjexp(x) == 0: |
5301 | | * mpd_exp_digits(t)-1 == 0 <= emax (the shortcut is not triggered) |
5302 | | * |
5303 | | * Case adjexp(x) < 0: |
5304 | | * 6) -adjexp(x) - 1 < abs(log10(x)) |
5305 | | * Case adjexp(x) < -1: |
5306 | | * 7) adjexp(-adjexp(x) - 1) <= adjexp(abs(log(x))) |
5307 | | * Case adjexp(x) == -1: |
5308 | | * mpd_exp_digits(t)-1 == 0 <= emax (the shortcut is not triggered) |
5309 | | */ |
5310 | 0 | adjexp = mpd_adjexp(a); |
5311 | 0 | t = (adjexp < 0) ? -adjexp-1 : adjexp; |
5312 | 0 | if (mpd_exp_digits(t)-1 > ctx->emax) { |
5313 | 0 | *status |= MPD_Overflow|MPD_Inexact|MPD_Rounded; |
5314 | 0 | mpd_setspecial(result, (adjexp<0), MPD_INF); |
5315 | 0 | return; |
5316 | 0 | } |
5317 | | |
5318 | 0 | if (ctx->allcr) { |
5319 | 0 | MPD_NEW_STATIC(hi, 0,0,0,0); |
5320 | 0 | MPD_NEW_STATIC(lo, 0,0,0,0); |
5321 | 0 | MPD_NEW_STATIC(ulp, 0,0,0,0); |
5322 | 0 | MPD_NEW_STATIC(aa, 0,0,0,0); |
5323 | 0 | uint32_t loop_protect = 0; |
5324 | 0 | mpd_ssize_t prec; |
5325 | |
|
5326 | 0 | if (result == a) { |
5327 | 0 | if (!mpd_qcopy(&aa, a, status)) { |
5328 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
5329 | 0 | return; |
5330 | 0 | } |
5331 | 0 | a = &aa; |
5332 | 0 | } |
5333 | | |
5334 | 0 | workctx.clamp = 0; |
5335 | 0 | prec = ctx->prec + 3; |
5336 | 0 | while (1) { |
5337 | 0 | uint32_t status_res = 0; |
5338 | 0 | uint32_t status_hi = 0; |
5339 | 0 | uint32_t status_lo = 0; |
5340 | 0 | int shortcut; |
5341 | 0 | int bounds_eq; |
5342 | 0 | int subnormal_eq; |
5343 | |
|
5344 | 0 | workctx.prec = prec; |
5345 | 0 | shortcut = _mpd_qlog10(SKIP_FINALIZE, result, a, &workctx, &status_res); |
5346 | 0 | if (mpd_isnan(result)) { |
5347 | 0 | mpd_seterror(result, status_res, status); |
5348 | 0 | break; |
5349 | 0 | } |
5350 | 0 | if (shortcut || mpd_isinfinite(result) || mpd_iszero(result)) { |
5351 | 0 | *status |= status_res; |
5352 | 0 | workctx.prec = ctx->prec; |
5353 | 0 | workctx.clamp = ctx->clamp; |
5354 | 0 | mpd_qfinalize(result, &workctx, status); |
5355 | 0 | break; |
5356 | 0 | } |
5357 | | |
5358 | 0 | _ssettriple(&ulp, MPD_POS, 1, |
5359 | 0 | result->exp + result->digits-workctx.prec); |
5360 | |
|
5361 | 0 | workctx.prec = ctx->prec; |
5362 | 0 | mpd_qadd(&hi, result, &ulp, &workctx, &status_hi); |
5363 | 0 | mpd_qsub(&lo, result, &ulp, &workctx, &status_lo); |
5364 | 0 | if (mpd_isnan(&hi) || mpd_isnan(&lo)) { |
5365 | 0 | mpd_seterror(result, status_hi|status_lo, status); |
5366 | 0 | break; |
5367 | 0 | } |
5368 | 0 | subnormal_eq = (status_hi&MPD_Subnormal) == (status_lo&MPD_Subnormal); |
5369 | 0 | bounds_eq = mpd_qcmp(&hi, &lo, status) == 0; |
5370 | 0 | if (bounds_eq && ++loop_protect > 5) { |
5371 | | /* If the bounds are equal, the result is always correctly rounded. |
5372 | | |
5373 | | Resolving the subnormal status can take more iterations (around |
5374 | | three) in extremely rare cases. 'hi' and 'lo' are so close that |
5375 | | subnormal/underflow is largely cosmetic, so allow a maximum of |
5376 | | five additional iterations. */ |
5377 | 0 | subnormal_eq = 1; /* GCOV_NOT_REACHED */ |
5378 | 0 | } |
5379 | |
|
5380 | 0 | if (subnormal_eq && bounds_eq) { |
5381 | 0 | *status |= status_lo; |
5382 | 0 | workctx.clamp = ctx->clamp; |
5383 | 0 | mpd_qfinalize(result, &workctx, status); |
5384 | 0 | break; |
5385 | 0 | } |
5386 | | |
5387 | 0 | if (subnormal_eq) { |
5388 | 0 | prec += MPD_RDIGITS; |
5389 | 0 | } |
5390 | 0 | else { |
5391 | 0 | prec *= 2; |
5392 | 0 | } |
5393 | |
|
5394 | 0 | if (prec > MPD_MAX_PREC) { |
5395 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
5396 | 0 | break; |
5397 | 0 | } |
5398 | 0 | } |
5399 | 0 | mpd_del(&hi); |
5400 | 0 | mpd_del(&lo); |
5401 | 0 | mpd_del(&ulp); |
5402 | 0 | mpd_del(&aa); |
5403 | 0 | } |
5404 | 0 | else { |
5405 | 0 | _mpd_qlog10(DO_FINALIZE, result, a, &workctx, status); |
5406 | 0 | mpd_check_underflow(result, &workctx, status); |
5407 | 0 | } |
5408 | 0 | } |
5409 | | |
5410 | | /* |
5411 | | * Maximum of the two operands. Attention: If one operand is a quiet NaN and the |
5412 | | * other is numeric, the numeric operand is returned. This may not be what one |
5413 | | * expects. |
5414 | | */ |
5415 | | void |
5416 | | mpd_qmax(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5417 | | const mpd_context_t *ctx, uint32_t *status) |
5418 | 16 | { |
5419 | 16 | int c; |
5420 | | |
5421 | 16 | if (mpd_isqnan(a) && !mpd_isnan(b)) { |
5422 | 0 | mpd_qcopy(result, b, status); |
5423 | 0 | } |
5424 | 16 | else if (mpd_isqnan(b) && !mpd_isnan(a)) { |
5425 | 0 | mpd_qcopy(result, a, status); |
5426 | 0 | } |
5427 | 16 | else if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5428 | 0 | return; |
5429 | 0 | } |
5430 | 16 | else { |
5431 | 16 | c = _mpd_cmp(a, b); |
5432 | 16 | if (c == 0) { |
5433 | 0 | c = _mpd_cmp_numequal(a, b); |
5434 | 0 | } |
5435 | | |
5436 | 16 | if (c < 0) { |
5437 | 8 | mpd_qcopy(result, b, status); |
5438 | 8 | } |
5439 | 8 | else { |
5440 | 8 | mpd_qcopy(result, a, status); |
5441 | 8 | } |
5442 | 16 | } |
5443 | | |
5444 | 16 | mpd_qfinalize(result, ctx, status); |
5445 | 16 | } |
5446 | | |
5447 | | /* |
5448 | | * Maximum magnitude: Same as mpd_max(), but compares the operands with their |
5449 | | * sign ignored. |
5450 | | */ |
5451 | | void |
5452 | | mpd_qmax_mag(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5453 | | const mpd_context_t *ctx, uint32_t *status) |
5454 | 0 | { |
5455 | 0 | int c; |
5456 | |
|
5457 | 0 | if (mpd_isqnan(a) && !mpd_isnan(b)) { |
5458 | 0 | mpd_qcopy(result, b, status); |
5459 | 0 | } |
5460 | 0 | else if (mpd_isqnan(b) && !mpd_isnan(a)) { |
5461 | 0 | mpd_qcopy(result, a, status); |
5462 | 0 | } |
5463 | 0 | else if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5464 | 0 | return; |
5465 | 0 | } |
5466 | 0 | else { |
5467 | 0 | c = _mpd_cmp_abs(a, b); |
5468 | 0 | if (c == 0) { |
5469 | 0 | c = _mpd_cmp_numequal(a, b); |
5470 | 0 | } |
5471 | |
|
5472 | 0 | if (c < 0) { |
5473 | 0 | mpd_qcopy(result, b, status); |
5474 | 0 | } |
5475 | 0 | else { |
5476 | 0 | mpd_qcopy(result, a, status); |
5477 | 0 | } |
5478 | 0 | } |
5479 | | |
5480 | 0 | mpd_qfinalize(result, ctx, status); |
5481 | 0 | } |
5482 | | |
5483 | | /* |
5484 | | * Minimum of the two operands. Attention: If one operand is a quiet NaN and the |
5485 | | * other is numeric, the numeric operand is returned. This may not be what one |
5486 | | * expects. |
5487 | | */ |
5488 | | void |
5489 | | mpd_qmin(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5490 | | const mpd_context_t *ctx, uint32_t *status) |
5491 | 24 | { |
5492 | 24 | int c; |
5493 | | |
5494 | 24 | if (mpd_isqnan(a) && !mpd_isnan(b)) { |
5495 | 0 | mpd_qcopy(result, b, status); |
5496 | 0 | } |
5497 | 24 | else if (mpd_isqnan(b) && !mpd_isnan(a)) { |
5498 | 0 | mpd_qcopy(result, a, status); |
5499 | 0 | } |
5500 | 24 | else if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5501 | 0 | return; |
5502 | 0 | } |
5503 | 24 | else { |
5504 | 24 | c = _mpd_cmp(a, b); |
5505 | 24 | if (c == 0) { |
5506 | 3 | c = _mpd_cmp_numequal(a, b); |
5507 | 3 | } |
5508 | | |
5509 | 24 | if (c < 0) { |
5510 | 5 | mpd_qcopy(result, a, status); |
5511 | 5 | } |
5512 | 19 | else { |
5513 | 19 | mpd_qcopy(result, b, status); |
5514 | 19 | } |
5515 | 24 | } |
5516 | | |
5517 | 24 | mpd_qfinalize(result, ctx, status); |
5518 | 24 | } |
5519 | | |
5520 | | /* |
5521 | | * Minimum magnitude: Same as mpd_min(), but compares the operands with their |
5522 | | * sign ignored. |
5523 | | */ |
5524 | | void |
5525 | | mpd_qmin_mag(mpd_t *result, const mpd_t *a, const mpd_t *b, |
5526 | | const mpd_context_t *ctx, uint32_t *status) |
5527 | 0 | { |
5528 | 0 | int c; |
5529 | |
|
5530 | 0 | if (mpd_isqnan(a) && !mpd_isnan(b)) { |
5531 | 0 | mpd_qcopy(result, b, status); |
5532 | 0 | } |
5533 | 0 | else if (mpd_isqnan(b) && !mpd_isnan(a)) { |
5534 | 0 | mpd_qcopy(result, a, status); |
5535 | 0 | } |
5536 | 0 | else if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
5537 | 0 | return; |
5538 | 0 | } |
5539 | 0 | else { |
5540 | 0 | c = _mpd_cmp_abs(a, b); |
5541 | 0 | if (c == 0) { |
5542 | 0 | c = _mpd_cmp_numequal(a, b); |
5543 | 0 | } |
5544 | |
|
5545 | 0 | if (c < 0) { |
5546 | 0 | mpd_qcopy(result, a, status); |
5547 | 0 | } |
5548 | 0 | else { |
5549 | 0 | mpd_qcopy(result, b, status); |
5550 | 0 | } |
5551 | 0 | } |
5552 | | |
5553 | 0 | mpd_qfinalize(result, ctx, status); |
5554 | 0 | } |
5555 | | |
5556 | | /* Minimum space needed for the result array in _karatsuba_rec(). */ |
5557 | | static inline mpd_size_t |
5558 | | _kmul_resultsize(mpd_size_t la, mpd_size_t lb) |
5559 | 0 | { |
5560 | 0 | mpd_size_t n, m; |
5561 | |
|
5562 | 0 | n = add_size_t(la, lb); |
5563 | 0 | n = add_size_t(n, 1); |
5564 | |
|
5565 | 0 | m = (la+1)/2 + 1; |
5566 | 0 | m = mul_size_t(m, 3); |
5567 | |
|
5568 | 0 | return (m > n) ? m : n; |
5569 | 0 | } |
5570 | | |
5571 | | /* Work space needed in _karatsuba_rec(). lim >= 4 */ |
5572 | | static inline mpd_size_t |
5573 | | _kmul_worksize(mpd_size_t n, mpd_size_t lim) |
5574 | 0 | { |
5575 | 0 | mpd_size_t m; |
5576 | |
|
5577 | 0 | if (n <= lim) { |
5578 | 0 | return 0; |
5579 | 0 | } |
5580 | | |
5581 | 0 | m = (n+1)/2 + 1; |
5582 | |
|
5583 | 0 | return add_size_t(mul_size_t(m, 2), _kmul_worksize(m, lim)); |
5584 | 0 | } |
5585 | | |
5586 | | |
5587 | 0 | #define MPD_KARATSUBA_BASECASE 16 /* must be >= 4 */ |
5588 | | |
5589 | | /* |
5590 | | * Add the product of a and b to c. |
5591 | | * c must be _kmul_resultsize(la, lb) in size. |
5592 | | * w is used as a work array and must be _kmul_worksize(a, lim) in size. |
5593 | | * Roman E. Maeder, Storage Allocation for the Karatsuba Integer Multiplication |
5594 | | * Algorithm. In "Design and implementation of symbolic computation systems", |
5595 | | * Springer, 1993, ISBN 354057235X, 9783540572350. |
5596 | | */ |
5597 | | static void |
5598 | | _karatsuba_rec(mpd_uint_t *c, const mpd_uint_t *a, const mpd_uint_t *b, |
5599 | | mpd_uint_t *w, mpd_size_t la, mpd_size_t lb) |
5600 | 0 | { |
5601 | 0 | mpd_size_t m, lt; |
5602 | |
|
5603 | 0 | assert(la >= lb && lb > 0); |
5604 | 0 | assert(la <= MPD_KARATSUBA_BASECASE || w != NULL); |
5605 | |
|
5606 | 0 | if (la <= MPD_KARATSUBA_BASECASE) { |
5607 | 0 | _mpd_basemul(c, a, b, la, lb); |
5608 | 0 | return; |
5609 | 0 | } |
5610 | | |
5611 | 0 | m = (la+1)/2; /* ceil(la/2) */ |
5612 | | |
5613 | | /* lb <= m < la */ |
5614 | 0 | if (lb <= m) { |
5615 | | |
5616 | | /* lb can now be larger than la-m */ |
5617 | 0 | if (lb > la-m) { |
5618 | 0 | lt = lb + lb + 1; /* space needed for result array */ |
5619 | 0 | mpd_uint_zero(w, lt); /* clear result array */ |
5620 | 0 | _karatsuba_rec(w, b, a+m, w+lt, lb, la-m); /* b*ah */ |
5621 | 0 | } |
5622 | 0 | else { |
5623 | 0 | lt = (la-m) + (la-m) + 1; /* space needed for result array */ |
5624 | 0 | mpd_uint_zero(w, lt); /* clear result array */ |
5625 | 0 | _karatsuba_rec(w, a+m, b, w+lt, la-m, lb); /* ah*b */ |
5626 | 0 | } |
5627 | 0 | _mpd_baseaddto(c+m, w, (la-m)+lb); /* add ah*b*B**m */ |
5628 | |
|
5629 | 0 | lt = m + m + 1; /* space needed for the result array */ |
5630 | 0 | mpd_uint_zero(w, lt); /* clear result array */ |
5631 | 0 | _karatsuba_rec(w, a, b, w+lt, m, lb); /* al*b */ |
5632 | 0 | _mpd_baseaddto(c, w, m+lb); /* add al*b */ |
5633 | |
|
5634 | 0 | return; |
5635 | 0 | } |
5636 | | |
5637 | | /* la >= lb > m */ |
5638 | 0 | memcpy(w, a, m * sizeof *w); |
5639 | 0 | w[m] = 0; |
5640 | 0 | _mpd_baseaddto(w, a+m, la-m); |
5641 | |
|
5642 | 0 | memcpy(w+(m+1), b, m * sizeof *w); |
5643 | 0 | w[m+1+m] = 0; |
5644 | 0 | _mpd_baseaddto(w+(m+1), b+m, lb-m); |
5645 | |
|
5646 | 0 | _karatsuba_rec(c+m, w, w+(m+1), w+2*(m+1), m+1, m+1); |
5647 | |
|
5648 | 0 | lt = (la-m) + (la-m) + 1; |
5649 | 0 | mpd_uint_zero(w, lt); |
5650 | |
|
5651 | 0 | _karatsuba_rec(w, a+m, b+m, w+lt, la-m, lb-m); |
5652 | |
|
5653 | 0 | _mpd_baseaddto(c+2*m, w, (la-m) + (lb-m)); |
5654 | 0 | _mpd_basesubfrom(c+m, w, (la-m) + (lb-m)); |
5655 | |
|
5656 | 0 | lt = m + m + 1; |
5657 | 0 | mpd_uint_zero(w, lt); |
5658 | |
|
5659 | 0 | _karatsuba_rec(w, a, b, w+lt, m, m); |
5660 | 0 | _mpd_baseaddto(c, w, m+m); |
5661 | 0 | _mpd_basesubfrom(c+m, w, m+m); |
5662 | |
|
5663 | 0 | return; |
5664 | 0 | } |
5665 | | |
5666 | | /* |
5667 | | * Multiply u and v, using Karatsuba multiplication. Returns a pointer |
5668 | | * to the result or NULL in case of failure (malloc error). |
5669 | | * Conditions: ulen >= vlen, ulen >= 4 |
5670 | | */ |
5671 | | static mpd_uint_t * |
5672 | | _mpd_kmul(const mpd_uint_t *u, const mpd_uint_t *v, |
5673 | | mpd_size_t ulen, mpd_size_t vlen, |
5674 | | mpd_size_t *rsize) |
5675 | 0 | { |
5676 | 0 | mpd_uint_t *result = NULL, *w = NULL; |
5677 | 0 | mpd_size_t m; |
5678 | |
|
5679 | 0 | assert(ulen >= 4); |
5680 | 0 | assert(ulen >= vlen); |
5681 | |
|
5682 | 0 | *rsize = _kmul_resultsize(ulen, vlen); |
5683 | 0 | if ((result = mpd_calloc(*rsize, sizeof *result)) == NULL) { |
5684 | 0 | return NULL; |
5685 | 0 | } |
5686 | | |
5687 | 0 | m = _kmul_worksize(ulen, MPD_KARATSUBA_BASECASE); |
5688 | 0 | if (m && ((w = mpd_calloc(m, sizeof *w)) == NULL)) { |
5689 | 0 | mpd_free(result); |
5690 | 0 | return NULL; |
5691 | 0 | } |
5692 | | |
5693 | 0 | _karatsuba_rec(result, u, v, w, ulen, vlen); |
5694 | | |
5695 | |
|
5696 | 0 | if (w) mpd_free(w); |
5697 | 0 | return result; |
5698 | 0 | } |
5699 | | |
5700 | | |
5701 | | /* |
5702 | | * Determine the minimum length for the number theoretic transform. Valid |
5703 | | * transform lengths are 2**n or 3*2**n, where 2**n <= MPD_MAXTRANSFORM_2N. |
5704 | | * The function finds the shortest length m such that rsize <= m. |
5705 | | */ |
5706 | | static inline mpd_size_t |
5707 | | _mpd_get_transform_len(mpd_size_t rsize) |
5708 | 0 | { |
5709 | 0 | mpd_size_t log2rsize; |
5710 | 0 | mpd_size_t x, step; |
5711 | |
|
5712 | 0 | assert(rsize >= 4); |
5713 | 0 | log2rsize = mpd_bsr(rsize); |
5714 | |
|
5715 | 0 | if (rsize <= 1024) { |
5716 | | /* 2**n is faster in this range. */ |
5717 | 0 | x = ((mpd_size_t)1)<<log2rsize; |
5718 | 0 | return (rsize == x) ? x : x<<1; |
5719 | 0 | } |
5720 | 0 | else if (rsize <= MPD_MAXTRANSFORM_2N) { |
5721 | 0 | x = ((mpd_size_t)1)<<log2rsize; |
5722 | 0 | if (rsize == x) return x; |
5723 | 0 | step = x>>1; |
5724 | 0 | x += step; |
5725 | 0 | return (rsize <= x) ? x : x + step; |
5726 | 0 | } |
5727 | 0 | else if (rsize <= MPD_MAXTRANSFORM_2N+MPD_MAXTRANSFORM_2N/2) { |
5728 | 0 | return MPD_MAXTRANSFORM_2N+MPD_MAXTRANSFORM_2N/2; |
5729 | 0 | } |
5730 | 0 | else if (rsize <= 3*MPD_MAXTRANSFORM_2N) { |
5731 | 0 | return 3*MPD_MAXTRANSFORM_2N; |
5732 | 0 | } |
5733 | 0 | else { |
5734 | 0 | return MPD_SIZE_MAX; |
5735 | 0 | } |
5736 | 0 | } |
5737 | | |
5738 | | #ifdef PPRO |
5739 | | #ifndef _MSC_VER |
5740 | | static inline unsigned short |
5741 | | _mpd_get_control87(void) |
5742 | | { |
5743 | | unsigned short cw; |
5744 | | |
5745 | | __asm__ __volatile__ ("fnstcw %0" : "=m" (cw)); |
5746 | | return cw; |
5747 | | } |
5748 | | |
5749 | | static inline void |
5750 | | _mpd_set_control87(unsigned short cw) |
5751 | | { |
5752 | | __asm__ __volatile__ ("fldcw %0" : : "m" (cw)); |
5753 | | } |
5754 | | #endif |
5755 | | |
5756 | | static unsigned int |
5757 | | mpd_set_fenv(void) |
5758 | | { |
5759 | | unsigned int cw; |
5760 | | #ifdef _MSC_VER |
5761 | | unsigned int flags = |
5762 | | _EM_INVALID|_EM_DENORMAL|_EM_ZERODIVIDE|_EM_OVERFLOW| |
5763 | | _EM_UNDERFLOW|_EM_INEXACT|_RC_CHOP|_PC_64; |
5764 | | unsigned int mask = _MCW_EM|_MCW_RC|_MCW_PC; |
5765 | | unsigned int dummy; |
5766 | | |
5767 | | __control87_2(0, 0, &cw, NULL); |
5768 | | __control87_2(flags, mask, &dummy, NULL); |
5769 | | #else |
5770 | | cw = _mpd_get_control87(); |
5771 | | _mpd_set_control87(cw|0xF3F); |
5772 | | #endif |
5773 | | return cw; |
5774 | | } |
5775 | | |
5776 | | static void |
5777 | | mpd_restore_fenv(unsigned int cw) |
5778 | | { |
5779 | | #ifdef _MSC_VER |
5780 | | unsigned int mask = _MCW_EM|_MCW_RC|_MCW_PC; |
5781 | | unsigned int dummy; |
5782 | | |
5783 | | __control87_2(cw, mask, &dummy, NULL); |
5784 | | #else |
5785 | | _mpd_set_control87((unsigned short)cw); |
5786 | | #endif |
5787 | | } |
5788 | | #endif /* PPRO */ |
5789 | | |
5790 | | /* |
5791 | | * Multiply u and v, using the fast number theoretic transform. Returns |
5792 | | * a pointer to the result or NULL in case of failure (malloc error). |
5793 | | */ |
5794 | | static mpd_uint_t * |
5795 | | _mpd_fntmul(const mpd_uint_t *u, const mpd_uint_t *v, |
5796 | | mpd_size_t ulen, mpd_size_t vlen, |
5797 | | mpd_size_t *rsize) |
5798 | 0 | { |
5799 | 0 | mpd_uint_t *c1 = NULL, *c2 = NULL, *c3 = NULL, *vtmp = NULL; |
5800 | 0 | mpd_size_t n; |
5801 | |
|
5802 | | #ifdef PPRO |
5803 | | unsigned int cw; |
5804 | | cw = mpd_set_fenv(); |
5805 | | #endif |
5806 | |
|
5807 | 0 | *rsize = add_size_t(ulen, vlen); |
5808 | 0 | if ((n = _mpd_get_transform_len(*rsize)) == MPD_SIZE_MAX) { |
5809 | 0 | goto malloc_error; |
5810 | 0 | } |
5811 | | |
5812 | 0 | if ((c1 = mpd_calloc(n, sizeof *c1)) == NULL) { |
5813 | 0 | goto malloc_error; |
5814 | 0 | } |
5815 | 0 | if ((c2 = mpd_calloc(n, sizeof *c2)) == NULL) { |
5816 | 0 | goto malloc_error; |
5817 | 0 | } |
5818 | 0 | if ((c3 = mpd_calloc(n, sizeof *c3)) == NULL) { |
5819 | 0 | goto malloc_error; |
5820 | 0 | } |
5821 | | |
5822 | 0 | memcpy(c1, u, ulen * (sizeof *c1)); |
5823 | 0 | memcpy(c2, u, ulen * (sizeof *c2)); |
5824 | 0 | memcpy(c3, u, ulen * (sizeof *c3)); |
5825 | |
|
5826 | 0 | if (u == v) { |
5827 | 0 | if (!fnt_autoconvolute(c1, n, P1) || |
5828 | 0 | !fnt_autoconvolute(c2, n, P2) || |
5829 | 0 | !fnt_autoconvolute(c3, n, P3)) { |
5830 | 0 | goto malloc_error; |
5831 | 0 | } |
5832 | 0 | } |
5833 | 0 | else { |
5834 | 0 | if ((vtmp = mpd_calloc(n, sizeof *vtmp)) == NULL) { |
5835 | 0 | goto malloc_error; |
5836 | 0 | } |
5837 | | |
5838 | 0 | memcpy(vtmp, v, vlen * (sizeof *vtmp)); |
5839 | 0 | if (!fnt_convolute(c1, vtmp, n, P1)) { |
5840 | 0 | mpd_free(vtmp); |
5841 | 0 | goto malloc_error; |
5842 | 0 | } |
5843 | | |
5844 | 0 | memcpy(vtmp, v, vlen * (sizeof *vtmp)); |
5845 | 0 | mpd_uint_zero(vtmp+vlen, n-vlen); |
5846 | 0 | if (!fnt_convolute(c2, vtmp, n, P2)) { |
5847 | 0 | mpd_free(vtmp); |
5848 | 0 | goto malloc_error; |
5849 | 0 | } |
5850 | | |
5851 | 0 | memcpy(vtmp, v, vlen * (sizeof *vtmp)); |
5852 | 0 | mpd_uint_zero(vtmp+vlen, n-vlen); |
5853 | 0 | if (!fnt_convolute(c3, vtmp, n, P3)) { |
5854 | 0 | mpd_free(vtmp); |
5855 | 0 | goto malloc_error; |
5856 | 0 | } |
5857 | | |
5858 | 0 | mpd_free(vtmp); |
5859 | 0 | } |
5860 | | |
5861 | 0 | crt3(c1, c2, c3, *rsize); |
5862 | |
|
5863 | 0 | out: |
5864 | | #ifdef PPRO |
5865 | | mpd_restore_fenv(cw); |
5866 | | #endif |
5867 | 0 | if (c2) mpd_free(c2); |
5868 | 0 | if (c3) mpd_free(c3); |
5869 | 0 | return c1; |
5870 | | |
5871 | 0 | malloc_error: |
5872 | 0 | if (c1) mpd_free(c1); |
5873 | 0 | c1 = NULL; |
5874 | 0 | goto out; |
5875 | 0 | } |
5876 | | |
5877 | | |
5878 | | /* |
5879 | | * Karatsuba multiplication with FNT/basemul as the base case. |
5880 | | */ |
5881 | | static int |
5882 | | _karatsuba_rec_fnt(mpd_uint_t *c, const mpd_uint_t *a, const mpd_uint_t *b, |
5883 | | mpd_uint_t *w, mpd_size_t la, mpd_size_t lb) |
5884 | 0 | { |
5885 | 0 | mpd_size_t m, lt; |
5886 | |
|
5887 | 0 | assert(la >= lb && lb > 0); |
5888 | 0 | assert(la <= 3*(MPD_MAXTRANSFORM_2N/2) || w != NULL); |
5889 | |
|
5890 | 0 | if (la <= 3*(MPD_MAXTRANSFORM_2N/2)) { |
5891 | |
|
5892 | 0 | if (lb <= 192) { |
5893 | 0 | _mpd_basemul(c, b, a, lb, la); |
5894 | 0 | } |
5895 | 0 | else { |
5896 | 0 | mpd_uint_t *result; |
5897 | 0 | mpd_size_t dummy; |
5898 | |
|
5899 | 0 | if ((result = _mpd_fntmul(a, b, la, lb, &dummy)) == NULL) { |
5900 | 0 | return 0; |
5901 | 0 | } |
5902 | 0 | memcpy(c, result, (la+lb) * (sizeof *result)); |
5903 | 0 | mpd_free(result); |
5904 | 0 | } |
5905 | 0 | return 1; |
5906 | 0 | } |
5907 | | |
5908 | 0 | m = (la+1)/2; /* ceil(la/2) */ |
5909 | | |
5910 | | /* lb <= m < la */ |
5911 | 0 | if (lb <= m) { |
5912 | | |
5913 | | /* lb can now be larger than la-m */ |
5914 | 0 | if (lb > la-m) { |
5915 | 0 | lt = lb + lb + 1; /* space needed for result array */ |
5916 | 0 | mpd_uint_zero(w, lt); /* clear result array */ |
5917 | 0 | if (!_karatsuba_rec_fnt(w, b, a+m, w+lt, lb, la-m)) { /* b*ah */ |
5918 | 0 | return 0; /* GCOV_UNLIKELY */ |
5919 | 0 | } |
5920 | 0 | } |
5921 | 0 | else { |
5922 | 0 | lt = (la-m) + (la-m) + 1; /* space needed for result array */ |
5923 | 0 | mpd_uint_zero(w, lt); /* clear result array */ |
5924 | 0 | if (!_karatsuba_rec_fnt(w, a+m, b, w+lt, la-m, lb)) { /* ah*b */ |
5925 | 0 | return 0; /* GCOV_UNLIKELY */ |
5926 | 0 | } |
5927 | 0 | } |
5928 | 0 | _mpd_baseaddto(c+m, w, (la-m)+lb); /* add ah*b*B**m */ |
5929 | |
|
5930 | 0 | lt = m + m + 1; /* space needed for the result array */ |
5931 | 0 | mpd_uint_zero(w, lt); /* clear result array */ |
5932 | 0 | if (!_karatsuba_rec_fnt(w, a, b, w+lt, m, lb)) { /* al*b */ |
5933 | 0 | return 0; /* GCOV_UNLIKELY */ |
5934 | 0 | } |
5935 | 0 | _mpd_baseaddto(c, w, m+lb); /* add al*b */ |
5936 | |
|
5937 | 0 | return 1; |
5938 | 0 | } |
5939 | | |
5940 | | /* la >= lb > m */ |
5941 | 0 | memcpy(w, a, m * sizeof *w); |
5942 | 0 | w[m] = 0; |
5943 | 0 | _mpd_baseaddto(w, a+m, la-m); |
5944 | |
|
5945 | 0 | memcpy(w+(m+1), b, m * sizeof *w); |
5946 | 0 | w[m+1+m] = 0; |
5947 | 0 | _mpd_baseaddto(w+(m+1), b+m, lb-m); |
5948 | |
|
5949 | 0 | if (!_karatsuba_rec_fnt(c+m, w, w+(m+1), w+2*(m+1), m+1, m+1)) { |
5950 | 0 | return 0; /* GCOV_UNLIKELY */ |
5951 | 0 | } |
5952 | | |
5953 | 0 | lt = (la-m) + (la-m) + 1; |
5954 | 0 | mpd_uint_zero(w, lt); |
5955 | |
|
5956 | 0 | if (!_karatsuba_rec_fnt(w, a+m, b+m, w+lt, la-m, lb-m)) { |
5957 | 0 | return 0; /* GCOV_UNLIKELY */ |
5958 | 0 | } |
5959 | | |
5960 | 0 | _mpd_baseaddto(c+2*m, w, (la-m) + (lb-m)); |
5961 | 0 | _mpd_basesubfrom(c+m, w, (la-m) + (lb-m)); |
5962 | |
|
5963 | 0 | lt = m + m + 1; |
5964 | 0 | mpd_uint_zero(w, lt); |
5965 | |
|
5966 | 0 | if (!_karatsuba_rec_fnt(w, a, b, w+lt, m, m)) { |
5967 | 0 | return 0; /* GCOV_UNLIKELY */ |
5968 | 0 | } |
5969 | 0 | _mpd_baseaddto(c, w, m+m); |
5970 | 0 | _mpd_basesubfrom(c+m, w, m+m); |
5971 | |
|
5972 | 0 | return 1; |
5973 | 0 | } |
5974 | | |
5975 | | /* |
5976 | | * Multiply u and v, using Karatsuba multiplication with the FNT as the |
5977 | | * base case. Returns a pointer to the result or NULL in case of failure |
5978 | | * (malloc error). Conditions: ulen >= vlen, ulen >= 4. |
5979 | | */ |
5980 | | static mpd_uint_t * |
5981 | | _mpd_kmul_fnt(const mpd_uint_t *u, const mpd_uint_t *v, |
5982 | | mpd_size_t ulen, mpd_size_t vlen, |
5983 | | mpd_size_t *rsize) |
5984 | 0 | { |
5985 | 0 | mpd_uint_t *result = NULL, *w = NULL; |
5986 | 0 | mpd_size_t m; |
5987 | |
|
5988 | 0 | assert(ulen >= 4); |
5989 | 0 | assert(ulen >= vlen); |
5990 | |
|
5991 | 0 | *rsize = _kmul_resultsize(ulen, vlen); |
5992 | 0 | if ((result = mpd_calloc(*rsize, sizeof *result)) == NULL) { |
5993 | 0 | return NULL; |
5994 | 0 | } |
5995 | | |
5996 | 0 | m = _kmul_worksize(ulen, 3*(MPD_MAXTRANSFORM_2N/2)); |
5997 | 0 | if (m && ((w = mpd_calloc(m, sizeof *w)) == NULL)) { |
5998 | 0 | mpd_free(result); /* GCOV_UNLIKELY */ |
5999 | 0 | return NULL; /* GCOV_UNLIKELY */ |
6000 | 0 | } |
6001 | | |
6002 | 0 | if (!_karatsuba_rec_fnt(result, u, v, w, ulen, vlen)) { |
6003 | 0 | mpd_free(result); |
6004 | 0 | result = NULL; |
6005 | 0 | } |
6006 | | |
6007 | |
|
6008 | 0 | if (w) mpd_free(w); |
6009 | 0 | return result; |
6010 | 0 | } |
6011 | | |
6012 | | |
6013 | | /* Deal with the special cases of multiplying infinities. */ |
6014 | | static void |
6015 | | _mpd_qmul_inf(mpd_t *result, const mpd_t *a, const mpd_t *b, uint32_t *status) |
6016 | 0 | { |
6017 | 0 | if (mpd_isinfinite(a)) { |
6018 | 0 | if (mpd_iszero(b)) { |
6019 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
6020 | 0 | } |
6021 | 0 | else { |
6022 | 0 | mpd_setspecial(result, mpd_sign(a)^mpd_sign(b), MPD_INF); |
6023 | 0 | } |
6024 | 0 | return; |
6025 | 0 | } |
6026 | 0 | assert(mpd_isinfinite(b)); |
6027 | 0 | if (mpd_iszero(a)) { |
6028 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
6029 | 0 | } |
6030 | 0 | else { |
6031 | 0 | mpd_setspecial(result, mpd_sign(a)^mpd_sign(b), MPD_INF); |
6032 | 0 | } |
6033 | 0 | } |
6034 | | |
6035 | | /* |
6036 | | * Internal function: Multiply a and b. _mpd_qmul deals with specials but |
6037 | | * does NOT finalize the result. This is for use in mpd_fma(). |
6038 | | */ |
6039 | | static inline void |
6040 | | _mpd_qmul(mpd_t *result, const mpd_t *a, const mpd_t *b, |
6041 | | const mpd_context_t *ctx, uint32_t *status) |
6042 | 2.76M | { |
6043 | 2.76M | const mpd_t *big = a, *small = b; |
6044 | 2.76M | mpd_uint_t *rdata = NULL; |
6045 | 2.76M | mpd_uint_t rbuf[MPD_MINALLOC_MAX]; |
6046 | 2.76M | mpd_size_t rsize, i; |
6047 | | |
6048 | | |
6049 | 2.76M | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
6050 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
6051 | 0 | return; |
6052 | 0 | } |
6053 | 0 | _mpd_qmul_inf(result, a, b, status); |
6054 | 0 | return; |
6055 | 0 | } |
6056 | | |
6057 | 2.76M | if (small->len > big->len) { |
6058 | 44.8k | _mpd_ptrswap(&big, &small); |
6059 | 44.8k | } |
6060 | | |
6061 | 2.76M | rsize = big->len + small->len; |
6062 | | |
6063 | 2.76M | if (big->len == 1) { |
6064 | 257k | _mpd_singlemul(result->data, big->data[0], small->data[0]); |
6065 | 257k | goto finish; |
6066 | 257k | } |
6067 | 2.50M | if (rsize <= (mpd_size_t)MPD_MINALLOC_MAX) { |
6068 | 898k | if (big->len == 2) { |
6069 | 378k | _mpd_mul_2_le2(rbuf, big->data, small->data, small->len); |
6070 | 378k | } |
6071 | 519k | else { |
6072 | 519k | mpd_uint_zero(rbuf, rsize); |
6073 | 519k | if (small->len == 1) { |
6074 | 3.07k | _mpd_shortmul(rbuf, big->data, big->len, small->data[0]); |
6075 | 3.07k | } |
6076 | 516k | else { |
6077 | 516k | _mpd_basemul(rbuf, small->data, big->data, small->len, big->len); |
6078 | 516k | } |
6079 | 519k | } |
6080 | 898k | if (!mpd_qresize(result, rsize, status)) { |
6081 | 0 | return; |
6082 | 0 | } |
6083 | 14.5M | for(i = 0; i < rsize; i++) { |
6084 | 13.6M | result->data[i] = rbuf[i]; |
6085 | 13.6M | } |
6086 | 898k | goto finish; |
6087 | 898k | } |
6088 | | |
6089 | | |
6090 | 1.61M | if (small->len <= 256) { |
6091 | 1.61M | rdata = mpd_calloc(rsize, sizeof *rdata); |
6092 | 1.61M | if (rdata != NULL) { |
6093 | 1.61M | if (small->len == 1) { |
6094 | 3 | _mpd_shortmul(rdata, big->data, big->len, small->data[0]); |
6095 | 3 | } |
6096 | 1.61M | else { |
6097 | 1.61M | _mpd_basemul(rdata, small->data, big->data, small->len, big->len); |
6098 | 1.61M | } |
6099 | 1.61M | } |
6100 | 1.61M | } |
6101 | 0 | else if (rsize <= 1024) { |
6102 | 0 | rdata = _mpd_kmul(big->data, small->data, big->len, small->len, &rsize); |
6103 | 0 | } |
6104 | 0 | else if (rsize <= 3*MPD_MAXTRANSFORM_2N) { |
6105 | 0 | rdata = _mpd_fntmul(big->data, small->data, big->len, small->len, &rsize); |
6106 | 0 | } |
6107 | 0 | else { |
6108 | 0 | rdata = _mpd_kmul_fnt(big->data, small->data, big->len, small->len, &rsize); |
6109 | 0 | } |
6110 | | |
6111 | 1.61M | if (rdata == NULL) { |
6112 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
6113 | 0 | return; |
6114 | 0 | } |
6115 | | |
6116 | 1.61M | if (mpd_isdynamic_data(result)) { |
6117 | 1.60M | mpd_free(result->data); |
6118 | 1.60M | } |
6119 | 1.61M | result->data = rdata; |
6120 | 1.61M | result->alloc = rsize; |
6121 | 1.61M | mpd_set_dynamic_data(result); |
6122 | | |
6123 | | |
6124 | 2.76M | finish: |
6125 | 2.76M | mpd_set_flags(result, mpd_sign(a)^mpd_sign(b)); |
6126 | 2.76M | result->exp = big->exp + small->exp; |
6127 | 2.76M | result->len = _mpd_real_size(result->data, rsize); |
6128 | | /* resize to smaller cannot fail */ |
6129 | 2.76M | mpd_qresize(result, result->len, status); |
6130 | 2.76M | mpd_setdigits(result); |
6131 | 2.76M | } |
6132 | | |
6133 | | /* Multiply a and b. */ |
6134 | | void |
6135 | | mpd_qmul(mpd_t *result, const mpd_t *a, const mpd_t *b, |
6136 | | const mpd_context_t *ctx, uint32_t *status) |
6137 | 2.76M | { |
6138 | 2.76M | _mpd_qmul(result, a, b, ctx, status); |
6139 | 2.76M | mpd_qfinalize(result, ctx, status); |
6140 | 2.76M | } |
6141 | | |
6142 | | /* Multiply a and b. Set NaN/Invalid_operation if the result is inexact. */ |
6143 | | static void |
6144 | | _mpd_qmul_exact(mpd_t *result, const mpd_t *a, const mpd_t *b, |
6145 | | const mpd_context_t *ctx, uint32_t *status) |
6146 | 2.76M | { |
6147 | 2.76M | uint32_t workstatus = 0; |
6148 | | |
6149 | 2.76M | mpd_qmul(result, a, b, ctx, &workstatus); |
6150 | 2.76M | *status |= workstatus; |
6151 | 2.76M | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
6152 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
6153 | 0 | } |
6154 | 2.76M | } |
6155 | | |
6156 | | /* Multiply decimal and mpd_ssize_t. */ |
6157 | | void |
6158 | | mpd_qmul_ssize(mpd_t *result, const mpd_t *a, mpd_ssize_t b, |
6159 | | const mpd_context_t *ctx, uint32_t *status) |
6160 | 0 | { |
6161 | 0 | mpd_context_t maxcontext; |
6162 | 0 | MPD_NEW_STATIC(bb,0,0,0,0); |
6163 | |
|
6164 | 0 | mpd_maxcontext(&maxcontext); |
6165 | 0 | mpd_qsset_ssize(&bb, b, &maxcontext, status); |
6166 | 0 | mpd_qmul(result, a, &bb, ctx, status); |
6167 | 0 | mpd_del(&bb); |
6168 | 0 | } |
6169 | | |
6170 | | /* Multiply decimal and mpd_uint_t. */ |
6171 | | void |
6172 | | mpd_qmul_uint(mpd_t *result, const mpd_t *a, mpd_uint_t b, |
6173 | | const mpd_context_t *ctx, uint32_t *status) |
6174 | 0 | { |
6175 | 0 | mpd_context_t maxcontext; |
6176 | 0 | MPD_NEW_STATIC(bb,0,0,0,0); |
6177 | |
|
6178 | 0 | mpd_maxcontext(&maxcontext); |
6179 | 0 | mpd_qsset_uint(&bb, b, &maxcontext, status); |
6180 | 0 | mpd_qmul(result, a, &bb, ctx, status); |
6181 | 0 | mpd_del(&bb); |
6182 | 0 | } |
6183 | | |
6184 | | void |
6185 | | mpd_qmul_i32(mpd_t *result, const mpd_t *a, int32_t b, |
6186 | | const mpd_context_t *ctx, uint32_t *status) |
6187 | 0 | { |
6188 | 0 | mpd_qmul_ssize(result, a, b, ctx, status); |
6189 | 0 | } |
6190 | | |
6191 | | void |
6192 | | mpd_qmul_u32(mpd_t *result, const mpd_t *a, uint32_t b, |
6193 | | const mpd_context_t *ctx, uint32_t *status) |
6194 | 0 | { |
6195 | 0 | mpd_qmul_uint(result, a, b, ctx, status); |
6196 | 0 | } |
6197 | | |
6198 | | #ifdef CONFIG_64 |
6199 | | void |
6200 | | mpd_qmul_i64(mpd_t *result, const mpd_t *a, int64_t b, |
6201 | | const mpd_context_t *ctx, uint32_t *status) |
6202 | 0 | { |
6203 | 0 | mpd_qmul_ssize(result, a, b, ctx, status); |
6204 | 0 | } |
6205 | | |
6206 | | void |
6207 | | mpd_qmul_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
6208 | | const mpd_context_t *ctx, uint32_t *status) |
6209 | 0 | { |
6210 | 0 | mpd_qmul_uint(result, a, b, ctx, status); |
6211 | 0 | } |
6212 | | #elif !defined(LEGACY_COMPILER) |
6213 | | /* Multiply decimal and int64_t. */ |
6214 | | void |
6215 | | mpd_qmul_i64(mpd_t *result, const mpd_t *a, int64_t b, |
6216 | | const mpd_context_t *ctx, uint32_t *status) |
6217 | | { |
6218 | | mpd_context_t maxcontext; |
6219 | | MPD_NEW_STATIC(bb,0,0,0,0); |
6220 | | |
6221 | | mpd_maxcontext(&maxcontext); |
6222 | | mpd_qset_i64(&bb, b, &maxcontext, status); |
6223 | | mpd_qmul(result, a, &bb, ctx, status); |
6224 | | mpd_del(&bb); |
6225 | | } |
6226 | | |
6227 | | /* Multiply decimal and uint64_t. */ |
6228 | | void |
6229 | | mpd_qmul_u64(mpd_t *result, const mpd_t *a, uint64_t b, |
6230 | | const mpd_context_t *ctx, uint32_t *status) |
6231 | | { |
6232 | | mpd_context_t maxcontext; |
6233 | | MPD_NEW_STATIC(bb,0,0,0,0); |
6234 | | |
6235 | | mpd_maxcontext(&maxcontext); |
6236 | | mpd_qset_u64(&bb, b, &maxcontext, status); |
6237 | | mpd_qmul(result, a, &bb, ctx, status); |
6238 | | mpd_del(&bb); |
6239 | | } |
6240 | | #endif |
6241 | | |
6242 | | /* Like the minus operator. */ |
6243 | | void |
6244 | | mpd_qminus(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
6245 | | uint32_t *status) |
6246 | 0 | { |
6247 | 0 | if (mpd_isspecial(a)) { |
6248 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
6249 | 0 | return; |
6250 | 0 | } |
6251 | 0 | } |
6252 | | |
6253 | 0 | if (mpd_iszero(a) && ctx->round != MPD_ROUND_FLOOR) { |
6254 | 0 | mpd_qcopy_abs(result, a, status); |
6255 | 0 | } |
6256 | 0 | else { |
6257 | 0 | mpd_qcopy_negate(result, a, status); |
6258 | 0 | } |
6259 | |
|
6260 | 0 | mpd_qfinalize(result, ctx, status); |
6261 | 0 | } |
6262 | | |
6263 | | /* Like the plus operator. */ |
6264 | | void |
6265 | | mpd_qplus(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
6266 | | uint32_t *status) |
6267 | 42 | { |
6268 | 42 | if (mpd_isspecial(a)) { |
6269 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
6270 | 0 | return; |
6271 | 0 | } |
6272 | 0 | } |
6273 | | |
6274 | 42 | if (mpd_iszero(a) && ctx->round != MPD_ROUND_FLOOR) { |
6275 | 6 | mpd_qcopy_abs(result, a, status); |
6276 | 6 | } |
6277 | 36 | else { |
6278 | 36 | mpd_qcopy(result, a, status); |
6279 | 36 | } |
6280 | | |
6281 | 42 | mpd_qfinalize(result, ctx, status); |
6282 | 42 | } |
6283 | | |
6284 | | /* The largest representable number that is smaller than the operand. */ |
6285 | | void |
6286 | | mpd_qnext_minus(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
6287 | | uint32_t *status) |
6288 | 0 | { |
6289 | 0 | mpd_context_t workctx; |
6290 | 0 | MPD_NEW_CONST(tiny,MPD_POS,mpd_etiny(ctx)-1,1,1,1,1); |
6291 | |
|
6292 | 0 | if (mpd_isspecial(a)) { |
6293 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
6294 | 0 | return; |
6295 | 0 | } |
6296 | | |
6297 | 0 | assert(mpd_isinfinite(a)); |
6298 | 0 | if (mpd_isnegative(a)) { |
6299 | 0 | mpd_qcopy(result, a, status); |
6300 | 0 | return; |
6301 | 0 | } |
6302 | 0 | else { |
6303 | 0 | mpd_clear_flags(result); |
6304 | 0 | mpd_qmaxcoeff(result, ctx, status); |
6305 | 0 | if (mpd_isnan(result)) { |
6306 | 0 | return; |
6307 | 0 | } |
6308 | 0 | result->exp = mpd_etop(ctx); |
6309 | 0 | return; |
6310 | 0 | } |
6311 | 0 | } |
6312 | | |
6313 | 0 | mpd_workcontext(&workctx, ctx); |
6314 | 0 | workctx.round = MPD_ROUND_FLOOR; |
6315 | |
|
6316 | 0 | if (!mpd_qcopy(result, a, status)) { |
6317 | 0 | return; |
6318 | 0 | } |
6319 | | |
6320 | 0 | mpd_qfinalize(result, &workctx, &workctx.status); |
6321 | 0 | if (workctx.status&(MPD_Inexact|MPD_Errors)) { |
6322 | 0 | *status |= (workctx.status&MPD_Errors); |
6323 | 0 | return; |
6324 | 0 | } |
6325 | | |
6326 | 0 | workctx.status = 0; |
6327 | 0 | mpd_qsub(result, a, &tiny, &workctx, &workctx.status); |
6328 | 0 | *status |= (workctx.status&MPD_Errors); |
6329 | 0 | } |
6330 | | |
6331 | | /* The smallest representable number that is larger than the operand. */ |
6332 | | void |
6333 | | mpd_qnext_plus(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
6334 | | uint32_t *status) |
6335 | 0 | { |
6336 | 0 | mpd_context_t workctx; |
6337 | 0 | MPD_NEW_CONST(tiny,MPD_POS,mpd_etiny(ctx)-1,1,1,1,1); |
6338 | |
|
6339 | 0 | if (mpd_isspecial(a)) { |
6340 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
6341 | 0 | return; |
6342 | 0 | } |
6343 | | |
6344 | 0 | assert(mpd_isinfinite(a)); |
6345 | 0 | if (mpd_ispositive(a)) { |
6346 | 0 | mpd_qcopy(result, a, status); |
6347 | 0 | } |
6348 | 0 | else { |
6349 | 0 | mpd_clear_flags(result); |
6350 | 0 | mpd_qmaxcoeff(result, ctx, status); |
6351 | 0 | if (mpd_isnan(result)) { |
6352 | 0 | return; |
6353 | 0 | } |
6354 | 0 | mpd_set_flags(result, MPD_NEG); |
6355 | 0 | result->exp = mpd_etop(ctx); |
6356 | 0 | } |
6357 | 0 | return; |
6358 | 0 | } |
6359 | | |
6360 | 0 | mpd_workcontext(&workctx, ctx); |
6361 | 0 | workctx.round = MPD_ROUND_CEILING; |
6362 | |
|
6363 | 0 | if (!mpd_qcopy(result, a, status)) { |
6364 | 0 | return; |
6365 | 0 | } |
6366 | | |
6367 | 0 | mpd_qfinalize(result, &workctx, &workctx.status); |
6368 | 0 | if (workctx.status & (MPD_Inexact|MPD_Errors)) { |
6369 | 0 | *status |= (workctx.status&MPD_Errors); |
6370 | 0 | return; |
6371 | 0 | } |
6372 | | |
6373 | 0 | workctx.status = 0; |
6374 | 0 | mpd_qadd(result, a, &tiny, &workctx, &workctx.status); |
6375 | 0 | *status |= (workctx.status&MPD_Errors); |
6376 | 0 | } |
6377 | | |
6378 | | /* |
6379 | | * The number closest to the first operand that is in the direction towards |
6380 | | * the second operand. |
6381 | | */ |
6382 | | void |
6383 | | mpd_qnext_toward(mpd_t *result, const mpd_t *a, const mpd_t *b, |
6384 | | const mpd_context_t *ctx, uint32_t *status) |
6385 | 0 | { |
6386 | 0 | int c; |
6387 | |
|
6388 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
6389 | 0 | return; |
6390 | 0 | } |
6391 | | |
6392 | 0 | c = _mpd_cmp(a, b); |
6393 | 0 | if (c == 0) { |
6394 | 0 | mpd_qcopy_sign(result, a, b, status); |
6395 | 0 | return; |
6396 | 0 | } |
6397 | | |
6398 | 0 | if (c < 0) { |
6399 | 0 | mpd_qnext_plus(result, a, ctx, status); |
6400 | 0 | } |
6401 | 0 | else { |
6402 | 0 | mpd_qnext_minus(result, a, ctx, status); |
6403 | 0 | } |
6404 | |
|
6405 | 0 | if (mpd_isinfinite(result)) { |
6406 | 0 | *status |= (MPD_Overflow|MPD_Rounded|MPD_Inexact); |
6407 | 0 | } |
6408 | 0 | else if (mpd_adjexp(result) < ctx->emin) { |
6409 | 0 | *status |= (MPD_Underflow|MPD_Subnormal|MPD_Rounded|MPD_Inexact); |
6410 | 0 | if (mpd_iszero(result)) { |
6411 | 0 | *status |= MPD_Clamped; |
6412 | 0 | } |
6413 | 0 | } |
6414 | 0 | } |
6415 | | |
6416 | | /* |
6417 | | * Internal function: Integer power with mpd_uint_t exponent. The function |
6418 | | * can fail with MPD_Malloc_error. |
6419 | | * |
6420 | | * The error is equal to the error incurred in k-1 multiplications. Assuming |
6421 | | * the upper bound for the relative error in each operation: |
6422 | | * |
6423 | | * abs(err) = 5 * 10**-prec |
6424 | | * result = x**k * (1 + err)**(k-1) |
6425 | | */ |
6426 | | static inline void |
6427 | | _mpd_qpow_uint(mpd_t *result, const mpd_t *base, mpd_uint_t exp, |
6428 | | uint8_t resultsign, const mpd_context_t *ctx, uint32_t *status) |
6429 | 0 | { |
6430 | 0 | uint32_t workstatus = 0; |
6431 | 0 | mpd_uint_t n; |
6432 | |
|
6433 | 0 | if (exp == 0) { |
6434 | 0 | _settriple(result, resultsign, 1, 0); /* GCOV_NOT_REACHED */ |
6435 | 0 | return; /* GCOV_NOT_REACHED */ |
6436 | 0 | } |
6437 | | |
6438 | 0 | if (!mpd_qcopy(result, base, status)) { |
6439 | 0 | return; |
6440 | 0 | } |
6441 | | |
6442 | 0 | n = mpd_bits[mpd_bsr(exp)]; |
6443 | 0 | while (n >>= 1) { |
6444 | 0 | mpd_qmul(result, result, result, ctx, &workstatus); |
6445 | 0 | if (exp & n) { |
6446 | 0 | mpd_qmul(result, result, base, ctx, &workstatus); |
6447 | 0 | } |
6448 | 0 | if (mpd_isspecial(result) || |
6449 | 0 | (mpd_iszerocoeff(result) && (workstatus & MPD_Clamped))) { |
6450 | 0 | break; |
6451 | 0 | } |
6452 | 0 | } |
6453 | |
|
6454 | 0 | *status |= workstatus; |
6455 | 0 | mpd_set_sign(result, resultsign); |
6456 | 0 | } |
6457 | | |
6458 | | /* |
6459 | | * Internal function: Integer power with mpd_t exponent, tbase and texp |
6460 | | * are modified!! Function can fail with MPD_Malloc_error. |
6461 | | * |
6462 | | * The error is equal to the error incurred in k multiplications. Assuming |
6463 | | * the upper bound for the relative error in each operation: |
6464 | | * |
6465 | | * abs(err) = 5 * 10**-prec |
6466 | | * result = x**k * (1 + err)**k |
6467 | | */ |
6468 | | static inline void |
6469 | | _mpd_qpow_mpd(mpd_t *result, mpd_t *tbase, mpd_t *texp, uint8_t resultsign, |
6470 | | const mpd_context_t *ctx, uint32_t *status) |
6471 | 0 | { |
6472 | 0 | uint32_t workstatus = 0; |
6473 | 0 | mpd_context_t maxctx; |
6474 | 0 | MPD_NEW_CONST(two,0,0,1,1,1,2); |
6475 | | |
6476 | |
|
6477 | 0 | mpd_maxcontext(&maxctx); |
6478 | | |
6479 | | /* resize to smaller cannot fail */ |
6480 | 0 | mpd_qcopy(result, &one, status); |
6481 | |
|
6482 | 0 | while (!mpd_iszero(texp)) { |
6483 | 0 | if (mpd_isodd(texp)) { |
6484 | 0 | mpd_qmul(result, result, tbase, ctx, &workstatus); |
6485 | 0 | *status |= workstatus; |
6486 | 0 | if (mpd_isspecial(result) || |
6487 | 0 | (mpd_iszerocoeff(result) && (workstatus & MPD_Clamped))) { |
6488 | 0 | break; |
6489 | 0 | } |
6490 | 0 | } |
6491 | 0 | mpd_qmul(tbase, tbase, tbase, ctx, &workstatus); |
6492 | 0 | mpd_qdivint(texp, texp, &two, &maxctx, &workstatus); |
6493 | 0 | if (mpd_isnan(tbase) || mpd_isnan(texp)) { |
6494 | 0 | mpd_seterror(result, workstatus&MPD_Errors, status); |
6495 | 0 | return; |
6496 | 0 | } |
6497 | 0 | } |
6498 | 0 | mpd_set_sign(result, resultsign); |
6499 | 0 | } |
6500 | | |
6501 | | /* |
6502 | | * The power function for integer exponents. Relative error _before_ the |
6503 | | * final rounding to prec: |
6504 | | * abs(result - base**exp) < 0.1 * 10**-prec * abs(base**exp) |
6505 | | */ |
6506 | | static void |
6507 | | _mpd_qpow_int(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6508 | | uint8_t resultsign, |
6509 | | const mpd_context_t *ctx, uint32_t *status) |
6510 | 0 | { |
6511 | 0 | mpd_context_t workctx; |
6512 | 0 | MPD_NEW_STATIC(tbase,0,0,0,0); |
6513 | 0 | MPD_NEW_STATIC(texp,0,0,0,0); |
6514 | 0 | mpd_uint_t n; |
6515 | | |
6516 | |
|
6517 | 0 | mpd_workcontext(&workctx, ctx); |
6518 | 0 | workctx.prec += (exp->digits + exp->exp + 2); |
6519 | 0 | workctx.round = MPD_ROUND_HALF_EVEN; |
6520 | 0 | workctx.clamp = 0; |
6521 | 0 | if (mpd_isnegative(exp)) { |
6522 | 0 | uint32_t workstatus = 0; |
6523 | 0 | workctx.prec += 1; |
6524 | 0 | mpd_qdiv(&tbase, &one, base, &workctx, &workstatus); |
6525 | 0 | *status |= workstatus; |
6526 | 0 | if (workstatus&MPD_Errors) { |
6527 | 0 | mpd_setspecial(result, MPD_POS, MPD_NAN); |
6528 | 0 | goto finish; |
6529 | 0 | } |
6530 | 0 | } |
6531 | 0 | else { |
6532 | 0 | if (!mpd_qcopy(&tbase, base, status)) { |
6533 | 0 | mpd_setspecial(result, MPD_POS, MPD_NAN); |
6534 | 0 | goto finish; |
6535 | 0 | } |
6536 | 0 | } |
6537 | | |
6538 | 0 | n = mpd_qabs_uint(exp, &workctx.status); |
6539 | 0 | if (workctx.status&MPD_Invalid_operation) { |
6540 | 0 | if (!mpd_qcopy(&texp, exp, status)) { |
6541 | 0 | mpd_setspecial(result, MPD_POS, MPD_NAN); /* GCOV_UNLIKELY */ |
6542 | 0 | goto finish; /* GCOV_UNLIKELY */ |
6543 | 0 | } |
6544 | 0 | _mpd_qpow_mpd(result, &tbase, &texp, resultsign, &workctx, status); |
6545 | 0 | } |
6546 | 0 | else { |
6547 | 0 | _mpd_qpow_uint(result, &tbase, n, resultsign, &workctx, status); |
6548 | 0 | } |
6549 | | |
6550 | 0 | if (mpd_isinfinite(result)) { |
6551 | | /* for ROUND_DOWN, ROUND_FLOOR, etc. */ |
6552 | 0 | _settriple(result, resultsign, 1, MPD_EXP_INF); |
6553 | 0 | } |
6554 | |
|
6555 | 0 | finish: |
6556 | 0 | mpd_del(&tbase); |
6557 | 0 | mpd_del(&texp); |
6558 | 0 | mpd_qfinalize(result, ctx, status); |
6559 | 0 | } |
6560 | | |
6561 | | /* |
6562 | | * If the exponent is infinite and base equals one, the result is one |
6563 | | * with a coefficient of length prec. Otherwise, result is undefined. |
6564 | | * Return the value of the comparison against one. |
6565 | | */ |
6566 | | static int |
6567 | | _qcheck_pow_one_inf(mpd_t *result, const mpd_t *base, uint8_t resultsign, |
6568 | | const mpd_context_t *ctx, uint32_t *status) |
6569 | 0 | { |
6570 | 0 | mpd_ssize_t shift; |
6571 | 0 | int cmp; |
6572 | |
|
6573 | 0 | if ((cmp = _mpd_cmp(base, &one)) == 0) { |
6574 | 0 | shift = ctx->prec-1; |
6575 | 0 | mpd_qshiftl(result, &one, shift, status); |
6576 | 0 | result->exp = -shift; |
6577 | 0 | mpd_set_flags(result, resultsign); |
6578 | 0 | *status |= (MPD_Inexact|MPD_Rounded); |
6579 | 0 | } |
6580 | |
|
6581 | 0 | return cmp; |
6582 | 0 | } |
6583 | | |
6584 | | /* |
6585 | | * If abs(base) equals one, calculate the correct power of one result. |
6586 | | * Otherwise, result is undefined. Return the value of the comparison |
6587 | | * against 1. |
6588 | | * |
6589 | | * This is an internal function that does not check for specials. |
6590 | | */ |
6591 | | static int |
6592 | | _qcheck_pow_one(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6593 | | uint8_t resultsign, |
6594 | | const mpd_context_t *ctx, uint32_t *status) |
6595 | 0 | { |
6596 | 0 | uint32_t workstatus = 0; |
6597 | 0 | mpd_ssize_t shift; |
6598 | 0 | int cmp; |
6599 | |
|
6600 | 0 | if ((cmp = _mpd_cmp_abs(base, &one)) == 0) { |
6601 | 0 | if (_mpd_isint(exp)) { |
6602 | 0 | if (mpd_isnegative(exp)) { |
6603 | 0 | _settriple(result, resultsign, 1, 0); |
6604 | 0 | return 0; |
6605 | 0 | } |
6606 | | /* 1.000**3 = 1.000000000 */ |
6607 | 0 | mpd_qmul_ssize(result, exp, -base->exp, ctx, &workstatus); |
6608 | 0 | if (workstatus&MPD_Errors) { |
6609 | 0 | *status |= (workstatus&MPD_Errors); |
6610 | 0 | return 0; |
6611 | 0 | } |
6612 | | /* digits-1 after exponentiation */ |
6613 | 0 | shift = mpd_qget_ssize(result, &workstatus); |
6614 | | /* shift is MPD_SSIZE_MAX if result is too large */ |
6615 | 0 | if (shift > ctx->prec-1) { |
6616 | 0 | shift = ctx->prec-1; |
6617 | 0 | *status |= MPD_Rounded; |
6618 | 0 | } |
6619 | 0 | } |
6620 | 0 | else if (mpd_ispositive(base)) { |
6621 | 0 | shift = ctx->prec-1; |
6622 | 0 | *status |= (MPD_Inexact|MPD_Rounded); |
6623 | 0 | } |
6624 | 0 | else { |
6625 | 0 | return -2; /* GCOV_NOT_REACHED */ |
6626 | 0 | } |
6627 | 0 | if (!mpd_qshiftl(result, &one, shift, status)) { |
6628 | 0 | return 0; |
6629 | 0 | } |
6630 | 0 | result->exp = -shift; |
6631 | 0 | mpd_set_flags(result, resultsign); |
6632 | 0 | } |
6633 | | |
6634 | 0 | return cmp; |
6635 | 0 | } |
6636 | | |
6637 | | /* |
6638 | | * Detect certain over/underflow of x**y. |
6639 | | * ACL2 proof: pow-bounds.lisp. |
6640 | | * |
6641 | | * Symbols: |
6642 | | * |
6643 | | * e: EXP_INF or EXP_CLAMP |
6644 | | * x: base |
6645 | | * y: exponent |
6646 | | * |
6647 | | * omega(e) = log10(abs(e)) |
6648 | | * zeta(x) = log10(abs(log10(x))) |
6649 | | * theta(y) = log10(abs(y)) |
6650 | | * |
6651 | | * Upper and lower bounds: |
6652 | | * |
6653 | | * ub_omega(e) = ceil(log10(abs(e))) |
6654 | | * lb_theta(y) = floor(log10(abs(y))) |
6655 | | * |
6656 | | * | floor(log10(floor(abs(log10(x))))) if x < 1/10 or x >= 10 |
6657 | | * lb_zeta(x) = | floor(log10(abs(x-1)/10)) if 1/10 <= x < 1 |
6658 | | * | floor(log10(abs((x-1)/100))) if 1 < x < 10 |
6659 | | * |
6660 | | * ub_omega(e) and lb_theta(y) are obviously upper and lower bounds |
6661 | | * for omega(e) and theta(y). |
6662 | | * |
6663 | | * lb_zeta is a lower bound for zeta(x): |
6664 | | * |
6665 | | * x < 1/10 or x >= 10: |
6666 | | * |
6667 | | * abs(log10(x)) >= 1, so the outer log10 is well defined. Since log10 |
6668 | | * is strictly increasing, the end result is a lower bound. |
6669 | | * |
6670 | | * 1/10 <= x < 1: |
6671 | | * |
6672 | | * We use: log10(x) <= (x-1)/log(10) |
6673 | | * abs(log10(x)) >= abs(x-1)/log(10) |
6674 | | * abs(log10(x)) >= abs(x-1)/10 |
6675 | | * |
6676 | | * 1 < x < 10: |
6677 | | * |
6678 | | * We use: (x-1)/(x*log(10)) < log10(x) |
6679 | | * abs((x-1)/100) < abs(log10(x)) |
6680 | | * |
6681 | | * XXX: abs((x-1)/10) would work, need ACL2 proof. |
6682 | | * |
6683 | | * |
6684 | | * Let (0 < x < 1 and y < 0) or (x > 1 and y > 0). (H1) |
6685 | | * Let ub_omega(exp_inf) < lb_zeta(x) + lb_theta(y) (H2) |
6686 | | * |
6687 | | * Then: |
6688 | | * log10(abs(exp_inf)) < log10(abs(log10(x))) + log10(abs(y)). (1) |
6689 | | * exp_inf < log10(x) * y (2) |
6690 | | * 10**exp_inf < x**y (3) |
6691 | | * |
6692 | | * Let (0 < x < 1 and y > 0) or (x > 1 and y < 0). (H3) |
6693 | | * Let ub_omega(exp_clamp) < lb_zeta(x) + lb_theta(y) (H4) |
6694 | | * |
6695 | | * Then: |
6696 | | * log10(abs(exp_clamp)) < log10(abs(log10(x))) + log10(abs(y)). (4) |
6697 | | * log10(x) * y < exp_clamp (5) |
6698 | | * x**y < 10**exp_clamp (6) |
6699 | | * |
6700 | | */ |
6701 | | static mpd_ssize_t |
6702 | | _lower_bound_zeta(const mpd_t *x, uint32_t *status) |
6703 | 0 | { |
6704 | 0 | mpd_context_t maxctx; |
6705 | 0 | MPD_NEW_STATIC(scratch,0,0,0,0); |
6706 | 0 | mpd_ssize_t t, u; |
6707 | |
|
6708 | 0 | t = mpd_adjexp(x); |
6709 | 0 | if (t > 0) { |
6710 | | /* x >= 10 -> floor(log10(floor(abs(log10(x))))) */ |
6711 | 0 | return mpd_exp_digits(t) - 1; |
6712 | 0 | } |
6713 | 0 | else if (t < -1) { |
6714 | | /* x < 1/10 -> floor(log10(floor(abs(log10(x))))) */ |
6715 | 0 | return mpd_exp_digits(t+1) - 1; |
6716 | 0 | } |
6717 | 0 | else { |
6718 | 0 | mpd_maxcontext(&maxctx); |
6719 | 0 | mpd_qsub(&scratch, x, &one, &maxctx, status); |
6720 | 0 | if (mpd_isspecial(&scratch)) { |
6721 | 0 | mpd_del(&scratch); |
6722 | 0 | return MPD_SSIZE_MAX; |
6723 | 0 | } |
6724 | 0 | u = mpd_adjexp(&scratch); |
6725 | 0 | mpd_del(&scratch); |
6726 | | |
6727 | | /* t == -1, 1/10 <= x < 1 -> floor(log10(abs(x-1)/10)) |
6728 | | * t == 0, 1 < x < 10 -> floor(log10(abs(x-1)/100)) */ |
6729 | 0 | return (t == 0) ? u-2 : u-1; |
6730 | 0 | } |
6731 | 0 | } |
6732 | | |
6733 | | /* |
6734 | | * Detect cases of certain overflow/underflow in the power function. |
6735 | | * Assumptions: x != 1, y != 0. The proof above is for positive x. |
6736 | | * If x is negative and y is an odd integer, x**y == -(abs(x)**y), |
6737 | | * so the analysis does not change. |
6738 | | */ |
6739 | | static int |
6740 | | _qcheck_pow_bounds(mpd_t *result, const mpd_t *x, const mpd_t *y, |
6741 | | uint8_t resultsign, |
6742 | | const mpd_context_t *ctx, uint32_t *status) |
6743 | 0 | { |
6744 | 0 | MPD_NEW_SHARED(abs_x, x); |
6745 | 0 | mpd_ssize_t ub_omega, lb_zeta, lb_theta; |
6746 | 0 | uint8_t sign; |
6747 | |
|
6748 | 0 | mpd_set_positive(&abs_x); |
6749 | |
|
6750 | 0 | lb_theta = mpd_adjexp(y); |
6751 | 0 | lb_zeta = _lower_bound_zeta(&abs_x, status); |
6752 | 0 | if (lb_zeta == MPD_SSIZE_MAX) { |
6753 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
6754 | 0 | return 1; |
6755 | 0 | } |
6756 | | |
6757 | 0 | sign = (mpd_adjexp(&abs_x) < 0) ^ mpd_sign(y); |
6758 | 0 | if (sign == 0) { |
6759 | | /* (0 < |x| < 1 and y < 0) or (|x| > 1 and y > 0) */ |
6760 | 0 | ub_omega = mpd_exp_digits(ctx->emax); |
6761 | 0 | if (ub_omega < lb_zeta + lb_theta) { |
6762 | 0 | _settriple(result, resultsign, 1, MPD_EXP_INF); |
6763 | 0 | mpd_qfinalize(result, ctx, status); |
6764 | 0 | return 1; |
6765 | 0 | } |
6766 | 0 | } |
6767 | 0 | else { |
6768 | | /* (0 < |x| < 1 and y > 0) or (|x| > 1 and y < 0). */ |
6769 | 0 | ub_omega = mpd_exp_digits(mpd_etiny(ctx)); |
6770 | 0 | if (ub_omega < lb_zeta + lb_theta) { |
6771 | 0 | _settriple(result, resultsign, 1, mpd_etiny(ctx)-1); |
6772 | 0 | mpd_qfinalize(result, ctx, status); |
6773 | 0 | return 1; |
6774 | 0 | } |
6775 | 0 | } |
6776 | | |
6777 | 0 | return 0; |
6778 | 0 | } |
6779 | | |
6780 | | /* |
6781 | | * TODO: Implement algorithm for computing exact powers from decimal.py. |
6782 | | * In order to prevent infinite loops, this has to be called before |
6783 | | * using Ziv's strategy for correct rounding. |
6784 | | */ |
6785 | | /* |
6786 | | static int |
6787 | | _mpd_qpow_exact(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6788 | | const mpd_context_t *ctx, uint32_t *status) |
6789 | | { |
6790 | | return 0; |
6791 | | } |
6792 | | */ |
6793 | | |
6794 | | /* |
6795 | | * The power function for real exponents. |
6796 | | * Relative error: abs(result - e**y) < e**y * 1/5 * 10**(-prec - 1) |
6797 | | */ |
6798 | | static void |
6799 | | _mpd_qpow_real(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6800 | | const mpd_context_t *ctx, uint32_t *status) |
6801 | 0 | { |
6802 | 0 | mpd_context_t workctx; |
6803 | 0 | MPD_NEW_STATIC(texp,0,0,0,0); |
6804 | |
|
6805 | 0 | if (!mpd_qcopy(&texp, exp, status)) { |
6806 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
6807 | 0 | return; |
6808 | 0 | } |
6809 | | |
6810 | 0 | mpd_maxcontext(&workctx); |
6811 | 0 | workctx.prec = (base->digits > ctx->prec) ? base->digits : ctx->prec; |
6812 | 0 | workctx.prec += (4 + MPD_EXPDIGITS); |
6813 | 0 | workctx.round = MPD_ROUND_HALF_EVEN; |
6814 | 0 | workctx.allcr = ctx->allcr; |
6815 | | |
6816 | | /* |
6817 | | * extra := MPD_EXPDIGITS = MPD_EXP_MAX_T |
6818 | | * wp := prec + 4 + extra |
6819 | | * abs(err) < 5 * 10**-wp |
6820 | | * y := log(base) * exp |
6821 | | * Calculate: |
6822 | | * 1) e**(y * (1 + err)**2) * (1 + err) |
6823 | | * = e**y * e**(y * (2*err + err**2)) * (1 + err) |
6824 | | * ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
6825 | | * Relative error of the underlined term: |
6826 | | * 2) abs(e**(y * (2*err + err**2)) - 1) |
6827 | | * Case abs(y) >= 10**extra: |
6828 | | * 3) adjexp(y)+1 > log10(abs(y)) >= extra |
6829 | | * This triggers the Overflow/Underflow shortcut in _mpd_qexp(), |
6830 | | * so no further analysis is necessary. |
6831 | | * Case abs(y) < 10**extra: |
6832 | | * 4) abs(y * (2*err + err**2)) < 1/5 * 10**(-prec - 2) |
6833 | | * Use (see _mpd_qexp): |
6834 | | * 5) abs(x) <= 9/10 * 10**-p ==> abs(e**x - 1) < 10**-p |
6835 | | * With 2), 4) and 5): |
6836 | | * 6) abs(e**(y * (2*err + err**2)) - 1) < 10**(-prec - 2) |
6837 | | * The complete relative error of 1) is: |
6838 | | * 7) abs(result - e**y) < e**y * 1/5 * 10**(-prec - 1) |
6839 | | */ |
6840 | 0 | mpd_qln(result, base, &workctx, &workctx.status); |
6841 | 0 | mpd_qmul(result, result, &texp, &workctx, &workctx.status); |
6842 | 0 | mpd_qexp(result, result, &workctx, status); |
6843 | |
|
6844 | 0 | mpd_del(&texp); |
6845 | 0 | *status |= (workctx.status&MPD_Errors); |
6846 | 0 | *status |= (MPD_Inexact|MPD_Rounded); |
6847 | 0 | } |
6848 | | |
6849 | | /* The power function: base**exp */ |
6850 | | void |
6851 | | mpd_qpow(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6852 | | const mpd_context_t *ctx, uint32_t *status) |
6853 | 0 | { |
6854 | 0 | uint8_t resultsign = 0; |
6855 | 0 | int intexp = 0; |
6856 | 0 | int cmp; |
6857 | |
|
6858 | 0 | if (mpd_isspecial(base) || mpd_isspecial(exp)) { |
6859 | 0 | if (mpd_qcheck_nans(result, base, exp, ctx, status)) { |
6860 | 0 | return; |
6861 | 0 | } |
6862 | 0 | } |
6863 | 0 | if (mpd_isinteger(exp)) { |
6864 | 0 | intexp = 1; |
6865 | 0 | resultsign = mpd_isnegative(base) && mpd_isodd(exp); |
6866 | 0 | } |
6867 | |
|
6868 | 0 | if (mpd_iszero(base)) { |
6869 | 0 | if (mpd_iszero(exp)) { |
6870 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
6871 | 0 | } |
6872 | 0 | else if (mpd_isnegative(exp)) { |
6873 | 0 | mpd_setspecial(result, resultsign, MPD_INF); |
6874 | 0 | } |
6875 | 0 | else { |
6876 | 0 | _settriple(result, resultsign, 0, 0); |
6877 | 0 | } |
6878 | 0 | return; |
6879 | 0 | } |
6880 | 0 | if (mpd_isnegative(base)) { |
6881 | 0 | if (!intexp || mpd_isinfinite(exp)) { |
6882 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
6883 | 0 | return; |
6884 | 0 | } |
6885 | 0 | } |
6886 | 0 | if (mpd_isinfinite(exp)) { |
6887 | | /* power of one */ |
6888 | 0 | cmp = _qcheck_pow_one_inf(result, base, resultsign, ctx, status); |
6889 | 0 | if (cmp == 0) { |
6890 | 0 | return; |
6891 | 0 | } |
6892 | 0 | else { |
6893 | 0 | cmp *= mpd_arith_sign(exp); |
6894 | 0 | if (cmp < 0) { |
6895 | 0 | _settriple(result, resultsign, 0, 0); |
6896 | 0 | } |
6897 | 0 | else { |
6898 | 0 | mpd_setspecial(result, resultsign, MPD_INF); |
6899 | 0 | } |
6900 | 0 | } |
6901 | 0 | return; |
6902 | 0 | } |
6903 | 0 | if (mpd_isinfinite(base)) { |
6904 | 0 | if (mpd_iszero(exp)) { |
6905 | 0 | _settriple(result, resultsign, 1, 0); |
6906 | 0 | } |
6907 | 0 | else if (mpd_isnegative(exp)) { |
6908 | 0 | _settriple(result, resultsign, 0, 0); |
6909 | 0 | } |
6910 | 0 | else { |
6911 | 0 | mpd_setspecial(result, resultsign, MPD_INF); |
6912 | 0 | } |
6913 | 0 | return; |
6914 | 0 | } |
6915 | 0 | if (mpd_iszero(exp)) { |
6916 | 0 | _settriple(result, resultsign, 1, 0); |
6917 | 0 | return; |
6918 | 0 | } |
6919 | 0 | if (_qcheck_pow_one(result, base, exp, resultsign, ctx, status) == 0) { |
6920 | 0 | return; |
6921 | 0 | } |
6922 | 0 | if (_qcheck_pow_bounds(result, base, exp, resultsign, ctx, status)) { |
6923 | 0 | return; |
6924 | 0 | } |
6925 | | |
6926 | 0 | if (intexp) { |
6927 | 0 | _mpd_qpow_int(result, base, exp, resultsign, ctx, status); |
6928 | 0 | } |
6929 | 0 | else { |
6930 | 0 | _mpd_qpow_real(result, base, exp, ctx, status); |
6931 | 0 | if (!mpd_isspecial(result) && _mpd_cmp(result, &one) == 0) { |
6932 | 0 | mpd_ssize_t shift = ctx->prec-1; |
6933 | 0 | mpd_qshiftl(result, &one, shift, status); |
6934 | 0 | result->exp = -shift; |
6935 | 0 | } |
6936 | 0 | if (mpd_isinfinite(result)) { |
6937 | | /* for ROUND_DOWN, ROUND_FLOOR, etc. */ |
6938 | 0 | _settriple(result, MPD_POS, 1, MPD_EXP_INF); |
6939 | 0 | } |
6940 | 0 | mpd_qfinalize(result, ctx, status); |
6941 | 0 | } |
6942 | 0 | } |
6943 | | |
6944 | | /* |
6945 | | * Internal function: Integer powmod with mpd_uint_t exponent, base is modified! |
6946 | | * Function can fail with MPD_Malloc_error. |
6947 | | */ |
6948 | | static inline void |
6949 | | _mpd_qpowmod_uint(mpd_t *result, mpd_t *base, mpd_uint_t exp, |
6950 | | const mpd_t *mod, uint32_t *status) |
6951 | 0 | { |
6952 | 0 | mpd_context_t maxcontext; |
6953 | |
|
6954 | 0 | mpd_maxcontext(&maxcontext); |
6955 | | |
6956 | | /* resize to smaller cannot fail */ |
6957 | 0 | mpd_qcopy(result, &one, status); |
6958 | |
|
6959 | 0 | while (exp > 0) { |
6960 | 0 | if (exp & 1) { |
6961 | 0 | _mpd_qmul_exact(result, result, base, &maxcontext, status); |
6962 | 0 | mpd_qrem(result, result, mod, &maxcontext, status); |
6963 | 0 | } |
6964 | 0 | _mpd_qmul_exact(base, base, base, &maxcontext, status); |
6965 | 0 | mpd_qrem(base, base, mod, &maxcontext, status); |
6966 | 0 | exp >>= 1; |
6967 | 0 | } |
6968 | 0 | } |
6969 | | |
6970 | | /* The powmod function: (base**exp) % mod */ |
6971 | | void |
6972 | | mpd_qpowmod(mpd_t *result, const mpd_t *base, const mpd_t *exp, |
6973 | | const mpd_t *mod, |
6974 | | const mpd_context_t *ctx, uint32_t *status) |
6975 | 3.67k | { |
6976 | 3.67k | mpd_context_t maxcontext; |
6977 | 3.67k | MPD_NEW_STATIC(tbase,0,0,0,0); |
6978 | 3.67k | MPD_NEW_STATIC(texp,0,0,0,0); |
6979 | 3.67k | MPD_NEW_STATIC(tmod,0,0,0,0); |
6980 | 3.67k | MPD_NEW_STATIC(tmp,0,0,0,0); |
6981 | 3.67k | MPD_NEW_CONST(two,0,0,1,1,1,2); |
6982 | 3.67k | mpd_ssize_t tbase_exp, texp_exp; |
6983 | 3.67k | mpd_ssize_t i; |
6984 | 3.67k | mpd_t t; |
6985 | 3.67k | mpd_uint_t r; |
6986 | 3.67k | uint8_t sign; |
6987 | | |
6988 | | |
6989 | 3.67k | if (mpd_isspecial(base) || mpd_isspecial(exp) || mpd_isspecial(mod)) { |
6990 | 0 | if (mpd_qcheck_3nans(result, base, exp, mod, ctx, status)) { |
6991 | 0 | return; |
6992 | 0 | } |
6993 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
6994 | 0 | return; |
6995 | 0 | } |
6996 | | |
6997 | | |
6998 | 3.67k | if (!_mpd_isint(base) || !_mpd_isint(exp) || !_mpd_isint(mod)) { |
6999 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7000 | 0 | return; |
7001 | 0 | } |
7002 | 3.67k | if (mpd_iszerocoeff(mod)) { |
7003 | 73 | mpd_seterror(result, MPD_Invalid_operation, status); |
7004 | 73 | return; |
7005 | 73 | } |
7006 | 3.60k | if (mod->digits+mod->exp > ctx->prec) { |
7007 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7008 | 0 | return; |
7009 | 0 | } |
7010 | | |
7011 | 3.60k | sign = (mpd_isnegative(base)) && (mpd_isodd(exp)); |
7012 | 3.60k | if (mpd_iszerocoeff(exp)) { |
7013 | 78 | if (mpd_iszerocoeff(base)) { |
7014 | 2 | mpd_seterror(result, MPD_Invalid_operation, status); |
7015 | 2 | return; |
7016 | 2 | } |
7017 | 76 | r = (_mpd_cmp_abs(mod, &one)==0) ? 0 : 1; |
7018 | 76 | _settriple(result, sign, r, 0); |
7019 | 76 | return; |
7020 | 78 | } |
7021 | 3.52k | if (mpd_isnegative(exp)) { |
7022 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7023 | 0 | return; |
7024 | 0 | } |
7025 | 3.52k | if (mpd_iszerocoeff(base)) { |
7026 | 18 | _settriple(result, sign, 0, 0); |
7027 | 18 | return; |
7028 | 18 | } |
7029 | | |
7030 | 3.50k | mpd_maxcontext(&maxcontext); |
7031 | | |
7032 | 3.50k | mpd_qrescale(&tmod, mod, 0, &maxcontext, &maxcontext.status); |
7033 | 3.50k | if (maxcontext.status&MPD_Errors) { |
7034 | 0 | mpd_seterror(result, maxcontext.status&MPD_Errors, status); |
7035 | 0 | goto out; |
7036 | 0 | } |
7037 | 3.50k | maxcontext.status = 0; |
7038 | 3.50k | mpd_set_positive(&tmod); |
7039 | | |
7040 | 3.50k | mpd_qround_to_int(&tbase, base, &maxcontext, status); |
7041 | 3.50k | mpd_set_positive(&tbase); |
7042 | 3.50k | tbase_exp = tbase.exp; |
7043 | 3.50k | tbase.exp = 0; |
7044 | | |
7045 | 3.50k | mpd_qround_to_int(&texp, exp, &maxcontext, status); |
7046 | 3.50k | texp_exp = texp.exp; |
7047 | 3.50k | texp.exp = 0; |
7048 | | |
7049 | | /* base = (base.int % modulo * pow(10, base.exp, modulo)) % modulo */ |
7050 | 3.50k | mpd_qrem(&tbase, &tbase, &tmod, &maxcontext, status); |
7051 | 3.50k | mpd_qshiftl(result, &one, tbase_exp, status); |
7052 | 3.50k | mpd_qrem(result, result, &tmod, &maxcontext, status); |
7053 | 3.50k | _mpd_qmul_exact(&tbase, &tbase, result, &maxcontext, status); |
7054 | 3.50k | mpd_qrem(&tbase, &tbase, &tmod, &maxcontext, status); |
7055 | 3.50k | if (mpd_isspecial(&tbase) || |
7056 | 3.50k | mpd_isspecial(&texp) || |
7057 | 3.50k | mpd_isspecial(&tmod)) { |
7058 | 0 | goto mpd_errors; |
7059 | 0 | } |
7060 | | |
7061 | 3.50k | for (i = 0; i < texp_exp; i++) { |
7062 | 0 | _mpd_qpowmod_uint(&tmp, &tbase, 10, &tmod, status); |
7063 | 0 | t = tmp; |
7064 | 0 | tmp = tbase; |
7065 | 0 | tbase = t; |
7066 | 0 | } |
7067 | 3.50k | if (mpd_isspecial(&tbase)) { |
7068 | 0 | goto mpd_errors; /* GCOV_UNLIKELY */ |
7069 | 0 | } |
7070 | | |
7071 | | /* resize to smaller cannot fail */ |
7072 | 3.50k | mpd_qcopy(result, &one, status); |
7073 | 1.85M | while (mpd_isfinite(&texp) && !mpd_iszero(&texp)) { |
7074 | 1.84M | if (mpd_isodd(&texp)) { |
7075 | 914k | _mpd_qmul_exact(result, result, &tbase, &maxcontext, status); |
7076 | 914k | mpd_qrem(result, result, &tmod, &maxcontext, status); |
7077 | 914k | } |
7078 | 1.84M | _mpd_qmul_exact(&tbase, &tbase, &tbase, &maxcontext, status); |
7079 | 1.84M | mpd_qrem(&tbase, &tbase, &tmod, &maxcontext, status); |
7080 | 1.84M | mpd_qdivint(&texp, &texp, &two, &maxcontext, status); |
7081 | 1.84M | } |
7082 | 3.50k | if (mpd_isspecial(&texp) || mpd_isspecial(&tbase) || |
7083 | 3.50k | mpd_isspecial(&tmod) || mpd_isspecial(result)) { |
7084 | | /* MPD_Malloc_error */ |
7085 | 0 | goto mpd_errors; |
7086 | 0 | } |
7087 | 3.50k | else { |
7088 | 3.50k | mpd_set_sign(result, sign); |
7089 | 3.50k | } |
7090 | | |
7091 | 3.50k | out: |
7092 | 3.50k | mpd_del(&tbase); |
7093 | 3.50k | mpd_del(&texp); |
7094 | 3.50k | mpd_del(&tmod); |
7095 | 3.50k | mpd_del(&tmp); |
7096 | 3.50k | return; |
7097 | | |
7098 | 0 | mpd_errors: |
7099 | 0 | mpd_setspecial(result, MPD_POS, MPD_NAN); |
7100 | 0 | goto out; |
7101 | 3.50k | } |
7102 | | |
7103 | | void |
7104 | | mpd_qquantize(mpd_t *result, const mpd_t *a, const mpd_t *b, |
7105 | | const mpd_context_t *ctx, uint32_t *status) |
7106 | 0 | { |
7107 | 0 | uint32_t workstatus = 0; |
7108 | 0 | mpd_ssize_t b_exp = b->exp; |
7109 | 0 | mpd_ssize_t expdiff, shift; |
7110 | 0 | mpd_uint_t rnd; |
7111 | |
|
7112 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
7113 | 0 | if (mpd_qcheck_nans(result, a, b, ctx, status)) { |
7114 | 0 | return; |
7115 | 0 | } |
7116 | 0 | if (mpd_isinfinite(a) && mpd_isinfinite(b)) { |
7117 | 0 | mpd_qcopy(result, a, status); |
7118 | 0 | return; |
7119 | 0 | } |
7120 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7121 | 0 | return; |
7122 | 0 | } |
7123 | | |
7124 | 0 | if (b->exp > ctx->emax || b->exp < mpd_etiny(ctx)) { |
7125 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7126 | 0 | return; |
7127 | 0 | } |
7128 | | |
7129 | 0 | if (mpd_iszero(a)) { |
7130 | 0 | _settriple(result, mpd_sign(a), 0, b->exp); |
7131 | 0 | mpd_qfinalize(result, ctx, status); |
7132 | 0 | return; |
7133 | 0 | } |
7134 | | |
7135 | | |
7136 | 0 | expdiff = a->exp - b->exp; |
7137 | 0 | if (a->digits + expdiff > ctx->prec) { |
7138 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7139 | 0 | return; |
7140 | 0 | } |
7141 | | |
7142 | 0 | if (expdiff >= 0) { |
7143 | 0 | shift = expdiff; |
7144 | 0 | if (!mpd_qshiftl(result, a, shift, status)) { |
7145 | 0 | return; |
7146 | 0 | } |
7147 | 0 | result->exp = b_exp; |
7148 | 0 | } |
7149 | 0 | else { |
7150 | | /* At this point expdiff < 0 and a->digits+expdiff <= prec, |
7151 | | * so the shift before an increment will fit in prec. */ |
7152 | 0 | shift = -expdiff; |
7153 | 0 | rnd = mpd_qshiftr(result, a, shift, status); |
7154 | 0 | if (rnd == MPD_UINT_MAX) { |
7155 | 0 | return; |
7156 | 0 | } |
7157 | 0 | result->exp = b_exp; |
7158 | 0 | if (!_mpd_apply_round_fit(result, rnd, ctx, status)) { |
7159 | 0 | return; |
7160 | 0 | } |
7161 | 0 | workstatus |= MPD_Rounded; |
7162 | 0 | if (rnd) { |
7163 | 0 | workstatus |= MPD_Inexact; |
7164 | 0 | } |
7165 | 0 | } |
7166 | | |
7167 | 0 | if (mpd_adjexp(result) > ctx->emax || |
7168 | 0 | mpd_adjexp(result) < mpd_etiny(ctx)) { |
7169 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7170 | 0 | return; |
7171 | 0 | } |
7172 | | |
7173 | 0 | *status |= workstatus; |
7174 | 0 | mpd_qfinalize(result, ctx, status); |
7175 | 0 | } |
7176 | | |
7177 | | void |
7178 | | mpd_qreduce(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7179 | | uint32_t *status) |
7180 | 0 | { |
7181 | 0 | mpd_ssize_t shift, maxexp, maxshift; |
7182 | 0 | uint8_t sign_a = mpd_sign(a); |
7183 | |
|
7184 | 0 | if (mpd_isspecial(a)) { |
7185 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
7186 | 0 | return; |
7187 | 0 | } |
7188 | 0 | mpd_qcopy(result, a, status); |
7189 | 0 | return; |
7190 | 0 | } |
7191 | | |
7192 | 0 | if (!mpd_qcopy(result, a, status)) { |
7193 | 0 | return; |
7194 | 0 | } |
7195 | 0 | mpd_qfinalize(result, ctx, status); |
7196 | 0 | if (mpd_isspecial(result)) { |
7197 | 0 | return; |
7198 | 0 | } |
7199 | 0 | if (mpd_iszero(result)) { |
7200 | 0 | _settriple(result, sign_a, 0, 0); |
7201 | 0 | return; |
7202 | 0 | } |
7203 | | |
7204 | 0 | shift = mpd_trail_zeros(result); |
7205 | 0 | maxexp = (ctx->clamp) ? mpd_etop(ctx) : ctx->emax; |
7206 | | /* After the finalizing above result->exp <= maxexp. */ |
7207 | 0 | maxshift = maxexp - result->exp; |
7208 | 0 | shift = (shift > maxshift) ? maxshift : shift; |
7209 | |
|
7210 | 0 | mpd_qshiftr_inplace(result, shift); |
7211 | 0 | result->exp += shift; |
7212 | 0 | } |
7213 | | |
7214 | | void |
7215 | | mpd_qrem(mpd_t *r, const mpd_t *a, const mpd_t *b, const mpd_context_t *ctx, |
7216 | | uint32_t *status) |
7217 | 2.77M | { |
7218 | 2.77M | MPD_NEW_STATIC(q,0,0,0,0); |
7219 | | |
7220 | 2.77M | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
7221 | 0 | if (mpd_qcheck_nans(r, a, b, ctx, status)) { |
7222 | 0 | return; |
7223 | 0 | } |
7224 | 0 | if (mpd_isinfinite(a)) { |
7225 | 0 | mpd_seterror(r, MPD_Invalid_operation, status); |
7226 | 0 | return; |
7227 | 0 | } |
7228 | 0 | if (mpd_isinfinite(b)) { |
7229 | 0 | mpd_qcopy(r, a, status); |
7230 | 0 | mpd_qfinalize(r, ctx, status); |
7231 | 0 | return; |
7232 | 0 | } |
7233 | | /* debug */ |
7234 | 0 | abort(); /* GCOV_NOT_REACHED */ |
7235 | 0 | } |
7236 | 2.77M | if (mpd_iszerocoeff(b)) { |
7237 | 0 | if (mpd_iszerocoeff(a)) { |
7238 | 0 | mpd_seterror(r, MPD_Division_undefined, status); |
7239 | 0 | } |
7240 | 0 | else { |
7241 | 0 | mpd_seterror(r, MPD_Invalid_operation, status); |
7242 | 0 | } |
7243 | 0 | return; |
7244 | 0 | } |
7245 | | |
7246 | 2.77M | _mpd_qdivmod(&q, r, a, b, ctx, status); |
7247 | 2.77M | mpd_del(&q); |
7248 | 2.77M | mpd_qfinalize(r, ctx, status); |
7249 | 2.77M | } |
7250 | | |
7251 | | void |
7252 | | mpd_qrem_near(mpd_t *r, const mpd_t *a, const mpd_t *b, |
7253 | | const mpd_context_t *ctx, uint32_t *status) |
7254 | 0 | { |
7255 | 0 | mpd_context_t workctx; |
7256 | 0 | MPD_NEW_STATIC(btmp,0,0,0,0); |
7257 | 0 | MPD_NEW_STATIC(q,0,0,0,0); |
7258 | 0 | mpd_ssize_t expdiff, qdigits; |
7259 | 0 | int cmp, isodd, allnine; |
7260 | |
|
7261 | 0 | assert(r != NULL); /* annotation for scan-build */ |
7262 | |
|
7263 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
7264 | 0 | if (mpd_qcheck_nans(r, a, b, ctx, status)) { |
7265 | 0 | return; |
7266 | 0 | } |
7267 | 0 | if (mpd_isinfinite(a)) { |
7268 | 0 | mpd_seterror(r, MPD_Invalid_operation, status); |
7269 | 0 | return; |
7270 | 0 | } |
7271 | 0 | if (mpd_isinfinite(b)) { |
7272 | 0 | mpd_qcopy(r, a, status); |
7273 | 0 | mpd_qfinalize(r, ctx, status); |
7274 | 0 | return; |
7275 | 0 | } |
7276 | | /* debug */ |
7277 | 0 | abort(); /* GCOV_NOT_REACHED */ |
7278 | 0 | } |
7279 | 0 | if (mpd_iszerocoeff(b)) { |
7280 | 0 | if (mpd_iszerocoeff(a)) { |
7281 | 0 | mpd_seterror(r, MPD_Division_undefined, status); |
7282 | 0 | } |
7283 | 0 | else { |
7284 | 0 | mpd_seterror(r, MPD_Invalid_operation, status); |
7285 | 0 | } |
7286 | 0 | return; |
7287 | 0 | } |
7288 | | |
7289 | 0 | if (r == b) { |
7290 | 0 | if (!mpd_qcopy(&btmp, b, status)) { |
7291 | 0 | mpd_seterror(r, MPD_Malloc_error, status); |
7292 | 0 | return; |
7293 | 0 | } |
7294 | 0 | b = &btmp; |
7295 | 0 | } |
7296 | | |
7297 | 0 | _mpd_qdivmod(&q, r, a, b, ctx, status); |
7298 | 0 | if (mpd_isnan(&q) || mpd_isnan(r)) { |
7299 | 0 | goto finish; |
7300 | 0 | } |
7301 | 0 | if (mpd_iszerocoeff(r)) { |
7302 | 0 | goto finish; |
7303 | 0 | } |
7304 | | |
7305 | 0 | expdiff = mpd_adjexp(b) - mpd_adjexp(r); |
7306 | 0 | if (-1 <= expdiff && expdiff <= 1) { |
7307 | |
|
7308 | 0 | allnine = mpd_coeff_isallnine(&q); |
7309 | 0 | qdigits = q.digits; |
7310 | 0 | isodd = mpd_isodd(&q); |
7311 | |
|
7312 | 0 | mpd_maxcontext(&workctx); |
7313 | 0 | if (mpd_sign(a) == mpd_sign(b)) { |
7314 | | /* sign(r) == sign(b) */ |
7315 | 0 | _mpd_qsub(&q, r, b, &workctx, &workctx.status); |
7316 | 0 | } |
7317 | 0 | else { |
7318 | | /* sign(r) != sign(b) */ |
7319 | 0 | _mpd_qadd(&q, r, b, &workctx, &workctx.status); |
7320 | 0 | } |
7321 | |
|
7322 | 0 | if (workctx.status&MPD_Errors) { |
7323 | 0 | mpd_seterror(r, workctx.status&MPD_Errors, status); |
7324 | 0 | goto finish; |
7325 | 0 | } |
7326 | | |
7327 | 0 | cmp = _mpd_cmp_abs(&q, r); |
7328 | 0 | if (cmp < 0 || (cmp == 0 && isodd)) { |
7329 | | /* abs(r) > abs(b)/2 or abs(r) == abs(b)/2 and isodd(quotient) */ |
7330 | 0 | if (allnine && qdigits == ctx->prec) { |
7331 | | /* abs(quotient) + 1 == 10**prec */ |
7332 | 0 | mpd_seterror(r, MPD_Division_impossible, status); |
7333 | 0 | goto finish; |
7334 | 0 | } |
7335 | 0 | mpd_qcopy(r, &q, status); |
7336 | 0 | } |
7337 | 0 | } |
7338 | | |
7339 | | |
7340 | 0 | finish: |
7341 | 0 | mpd_del(&btmp); |
7342 | 0 | mpd_del(&q); |
7343 | 0 | mpd_qfinalize(r, ctx, status); |
7344 | 0 | } |
7345 | | |
7346 | | static void |
7347 | | _mpd_qrescale(mpd_t *result, const mpd_t *a, mpd_ssize_t exp, |
7348 | | const mpd_context_t *ctx, uint32_t *status) |
7349 | 3.50k | { |
7350 | 3.50k | mpd_ssize_t expdiff, shift; |
7351 | 3.50k | mpd_uint_t rnd; |
7352 | | |
7353 | 3.50k | if (mpd_isspecial(a)) { |
7354 | 0 | mpd_qcopy(result, a, status); |
7355 | 0 | return; |
7356 | 0 | } |
7357 | | |
7358 | 3.50k | if (mpd_iszero(a)) { |
7359 | 0 | _settriple(result, mpd_sign(a), 0, exp); |
7360 | 0 | return; |
7361 | 0 | } |
7362 | | |
7363 | 3.50k | expdiff = a->exp - exp; |
7364 | 3.50k | if (expdiff >= 0) { |
7365 | 3.50k | shift = expdiff; |
7366 | 3.50k | if (a->digits + shift > MPD_MAX_PREC+1) { |
7367 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7368 | 0 | return; |
7369 | 0 | } |
7370 | 3.50k | if (!mpd_qshiftl(result, a, shift, status)) { |
7371 | 0 | return; |
7372 | 0 | } |
7373 | 3.50k | result->exp = exp; |
7374 | 3.50k | } |
7375 | 0 | else { |
7376 | 0 | shift = -expdiff; |
7377 | 0 | rnd = mpd_qshiftr(result, a, shift, status); |
7378 | 0 | if (rnd == MPD_UINT_MAX) { |
7379 | 0 | return; |
7380 | 0 | } |
7381 | 0 | result->exp = exp; |
7382 | 0 | _mpd_apply_round_excess(result, rnd, ctx, status); |
7383 | 0 | *status |= MPD_Rounded; |
7384 | 0 | if (rnd) { |
7385 | 0 | *status |= MPD_Inexact; |
7386 | 0 | } |
7387 | 0 | } |
7388 | | |
7389 | 3.50k | if (mpd_issubnormal(result, ctx)) { |
7390 | 0 | *status |= MPD_Subnormal; |
7391 | 0 | } |
7392 | 3.50k | } |
7393 | | |
7394 | | /* |
7395 | | * Rescale a number so that it has exponent 'exp'. Does not regard context |
7396 | | * precision, emax, emin, but uses the rounding mode. Special numbers are |
7397 | | * quietly copied. Restrictions: |
7398 | | * |
7399 | | * MPD_MIN_ETINY <= exp <= MPD_MAX_EMAX+1 |
7400 | | * result->digits <= MPD_MAX_PREC+1 |
7401 | | */ |
7402 | | void |
7403 | | mpd_qrescale(mpd_t *result, const mpd_t *a, mpd_ssize_t exp, |
7404 | | const mpd_context_t *ctx, uint32_t *status) |
7405 | 3.50k | { |
7406 | 3.50k | if (exp > MPD_MAX_EMAX+1 || exp < MPD_MIN_ETINY) { |
7407 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7408 | 0 | return; |
7409 | 0 | } |
7410 | | |
7411 | 3.50k | _mpd_qrescale(result, a, exp, ctx, status); |
7412 | 3.50k | } |
7413 | | |
7414 | | /* |
7415 | | * Same as mpd_qrescale, but with relaxed restrictions. The result of this |
7416 | | * function should only be used for formatting a number and never as input |
7417 | | * for other operations. |
7418 | | * |
7419 | | * MPD_MIN_ETINY-MPD_MAX_PREC <= exp <= MPD_MAX_EMAX+1 |
7420 | | * result->digits <= MPD_MAX_PREC+1 |
7421 | | */ |
7422 | | void |
7423 | | mpd_qrescale_fmt(mpd_t *result, const mpd_t *a, mpd_ssize_t exp, |
7424 | | const mpd_context_t *ctx, uint32_t *status) |
7425 | 0 | { |
7426 | 0 | if (exp > MPD_MAX_EMAX+1 || exp < MPD_MIN_ETINY-MPD_MAX_PREC) { |
7427 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7428 | 0 | return; |
7429 | 0 | } |
7430 | | |
7431 | 0 | _mpd_qrescale(result, a, exp, ctx, status); |
7432 | 0 | } |
7433 | | |
7434 | | /* Round to an integer according to 'action' and ctx->round. */ |
7435 | | enum {TO_INT_EXACT, TO_INT_SILENT, TO_INT_TRUNC}; |
7436 | | static void |
7437 | | _mpd_qround_to_integral(int action, mpd_t *result, const mpd_t *a, |
7438 | | const mpd_context_t *ctx, uint32_t *status) |
7439 | 7.01k | { |
7440 | 7.01k | mpd_uint_t rnd; |
7441 | | |
7442 | 7.01k | if (mpd_isspecial(a)) { |
7443 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
7444 | 0 | return; |
7445 | 0 | } |
7446 | 0 | mpd_qcopy(result, a, status); |
7447 | 0 | return; |
7448 | 0 | } |
7449 | 7.01k | if (a->exp >= 0) { |
7450 | 7.01k | mpd_qcopy(result, a, status); |
7451 | 7.01k | return; |
7452 | 7.01k | } |
7453 | 0 | if (mpd_iszerocoeff(a)) { |
7454 | 0 | _settriple(result, mpd_sign(a), 0, 0); |
7455 | 0 | return; |
7456 | 0 | } |
7457 | | |
7458 | 0 | rnd = mpd_qshiftr(result, a, -a->exp, status); |
7459 | 0 | if (rnd == MPD_UINT_MAX) { |
7460 | 0 | return; |
7461 | 0 | } |
7462 | 0 | result->exp = 0; |
7463 | |
|
7464 | 0 | if (action == TO_INT_EXACT || action == TO_INT_SILENT) { |
7465 | 0 | _mpd_apply_round_excess(result, rnd, ctx, status); |
7466 | 0 | if (action == TO_INT_EXACT) { |
7467 | 0 | *status |= MPD_Rounded; |
7468 | 0 | if (rnd) { |
7469 | 0 | *status |= MPD_Inexact; |
7470 | 0 | } |
7471 | 0 | } |
7472 | 0 | } |
7473 | 0 | } |
7474 | | |
7475 | | void |
7476 | | mpd_qround_to_intx(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7477 | | uint32_t *status) |
7478 | 0 | { |
7479 | 0 | (void)_mpd_qround_to_integral(TO_INT_EXACT, result, a, ctx, status); |
7480 | 0 | } |
7481 | | |
7482 | | void |
7483 | | mpd_qround_to_int(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7484 | | uint32_t *status) |
7485 | 7.01k | { |
7486 | 7.01k | (void)_mpd_qround_to_integral(TO_INT_SILENT, result, a, ctx, status); |
7487 | 7.01k | } |
7488 | | |
7489 | | void |
7490 | | mpd_qtrunc(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7491 | | uint32_t *status) |
7492 | 0 | { |
7493 | 0 | if (mpd_isspecial(a)) { |
7494 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7495 | 0 | return; |
7496 | 0 | } |
7497 | | |
7498 | 0 | (void)_mpd_qround_to_integral(TO_INT_TRUNC, result, a, ctx, status); |
7499 | 0 | } |
7500 | | |
7501 | | void |
7502 | | mpd_qfloor(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7503 | | uint32_t *status) |
7504 | 0 | { |
7505 | 0 | mpd_context_t workctx; |
7506 | 0 | mpd_workcontext(&workctx, ctx); |
7507 | |
|
7508 | 0 | if (mpd_isspecial(a)) { |
7509 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7510 | 0 | return; |
7511 | 0 | } |
7512 | | |
7513 | 0 | workctx.round = MPD_ROUND_FLOOR; |
7514 | 0 | (void)_mpd_qround_to_integral(TO_INT_SILENT, result, a, |
7515 | 0 | &workctx, status); |
7516 | 0 | } |
7517 | | |
7518 | | void |
7519 | | mpd_qceil(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7520 | | uint32_t *status) |
7521 | 0 | { |
7522 | 0 | mpd_context_t workctx; |
7523 | 0 | mpd_workcontext(&workctx, ctx); |
7524 | |
|
7525 | 0 | if (mpd_isspecial(a)) { |
7526 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
7527 | 0 | return; |
7528 | 0 | } |
7529 | | |
7530 | 0 | workctx.round = MPD_ROUND_CEILING; |
7531 | 0 | (void)_mpd_qround_to_integral(TO_INT_SILENT, result, a, |
7532 | 0 | &workctx, status); |
7533 | 0 | } |
7534 | | |
7535 | | int |
7536 | | mpd_same_quantum(const mpd_t *a, const mpd_t *b) |
7537 | 0 | { |
7538 | 0 | if (mpd_isspecial(a) || mpd_isspecial(b)) { |
7539 | 0 | return ((mpd_isnan(a) && mpd_isnan(b)) || |
7540 | 0 | (mpd_isinfinite(a) && mpd_isinfinite(b))); |
7541 | 0 | } |
7542 | | |
7543 | 0 | return a->exp == b->exp; |
7544 | 0 | } |
7545 | | |
7546 | | /* Schedule the increase in precision for the Newton iteration. */ |
7547 | | static inline int |
7548 | | recpr_schedule_prec(mpd_ssize_t klist[MPD_MAX_PREC_LOG2], |
7549 | | mpd_ssize_t maxprec, mpd_ssize_t initprec) |
7550 | 0 | { |
7551 | 0 | mpd_ssize_t k; |
7552 | 0 | int i; |
7553 | |
|
7554 | 0 | assert(maxprec > 0 && initprec > 0); |
7555 | 0 | if (maxprec <= initprec) return -1; |
7556 | | |
7557 | 0 | i = 0; k = maxprec; |
7558 | 0 | do { |
7559 | 0 | k = (k+1) / 2; |
7560 | 0 | klist[i++] = k; |
7561 | 0 | } while (k > initprec); |
7562 | |
|
7563 | 0 | return i-1; |
7564 | 0 | } |
7565 | | |
7566 | | /* |
7567 | | * Initial approximation for the reciprocal: |
7568 | | * k_0 := MPD_RDIGITS-2 |
7569 | | * z_0 := 10**(-k_0) * floor(10**(2*k_0 + 2) / floor(v * 10**(k_0 + 2))) |
7570 | | * Absolute error: |
7571 | | * |1/v - z_0| < 10**(-k_0) |
7572 | | * ACL2 proof: maxerror-inverse-approx |
7573 | | */ |
7574 | | static void |
7575 | | _mpd_qreciprocal_approx(mpd_t *z, const mpd_t *v, uint32_t *status) |
7576 | 0 | { |
7577 | 0 | mpd_uint_t p10data[2] = {0, mpd_pow10[MPD_RDIGITS-2]}; |
7578 | 0 | mpd_uint_t dummy, word; |
7579 | 0 | int n; |
7580 | |
|
7581 | 0 | assert(v->exp == -v->digits); |
7582 | |
|
7583 | 0 | _mpd_get_msdigits(&dummy, &word, v, MPD_RDIGITS); |
7584 | 0 | n = mpd_word_digits(word); |
7585 | 0 | word *= mpd_pow10[MPD_RDIGITS-n]; |
7586 | |
|
7587 | 0 | mpd_qresize(z, 2, status); |
7588 | 0 | (void)_mpd_shortdiv(z->data, p10data, 2, word); |
7589 | |
|
7590 | 0 | mpd_clear_flags(z); |
7591 | 0 | z->exp = -(MPD_RDIGITS-2); |
7592 | 0 | z->len = (z->data[1] == 0) ? 1 : 2; |
7593 | 0 | mpd_setdigits(z); |
7594 | 0 | } |
7595 | | |
7596 | | /* |
7597 | | * Reciprocal, calculated with Newton's Method. Assumption: result != a. |
7598 | | * NOTE: The comments in the function show that certain operations are |
7599 | | * exact. The proof for the maximum error is too long to fit in here. |
7600 | | * ACL2 proof: maxerror-inverse-complete |
7601 | | */ |
7602 | | static void |
7603 | | _mpd_qreciprocal(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7604 | | uint32_t *status) |
7605 | 0 | { |
7606 | 0 | mpd_context_t varcontext, maxcontext; |
7607 | 0 | mpd_t *z = result; /* current approximation */ |
7608 | 0 | mpd_t *v; /* a, normalized to a number between 0.1 and 1 */ |
7609 | 0 | MPD_NEW_SHARED(vtmp, a); /* v shares data with a */ |
7610 | 0 | MPD_NEW_STATIC(s,0,0,0,0); /* temporary variable */ |
7611 | 0 | MPD_NEW_STATIC(t,0,0,0,0); /* temporary variable */ |
7612 | 0 | MPD_NEW_CONST(two,0,0,1,1,1,2); /* const 2 */ |
7613 | 0 | mpd_ssize_t klist[MPD_MAX_PREC_LOG2]; |
7614 | 0 | mpd_ssize_t adj, maxprec, initprec; |
7615 | 0 | uint8_t sign = mpd_sign(a); |
7616 | 0 | int i; |
7617 | |
|
7618 | 0 | assert(result != a); |
7619 | |
|
7620 | 0 | v = &vtmp; |
7621 | 0 | mpd_clear_flags(v); |
7622 | 0 | adj = v->digits + v->exp; |
7623 | 0 | v->exp = -v->digits; |
7624 | | |
7625 | | /* Initial approximation */ |
7626 | 0 | _mpd_qreciprocal_approx(z, v, status); |
7627 | |
|
7628 | 0 | mpd_maxcontext(&varcontext); |
7629 | 0 | mpd_maxcontext(&maxcontext); |
7630 | 0 | varcontext.round = maxcontext.round = MPD_ROUND_TRUNC; |
7631 | 0 | varcontext.emax = maxcontext.emax = MPD_MAX_EMAX + 100; |
7632 | 0 | varcontext.emin = maxcontext.emin = MPD_MIN_EMIN - 100; |
7633 | 0 | maxcontext.prec = MPD_MAX_PREC + 100; |
7634 | |
|
7635 | 0 | maxprec = ctx->prec; |
7636 | 0 | maxprec += 2; |
7637 | 0 | initprec = MPD_RDIGITS-3; |
7638 | |
|
7639 | 0 | i = recpr_schedule_prec(klist, maxprec, initprec); |
7640 | 0 | for (; i >= 0; i--) { |
7641 | | /* Loop invariant: z->digits <= klist[i]+7 */ |
7642 | | /* Let s := z**2, exact result */ |
7643 | 0 | _mpd_qmul_exact(&s, z, z, &maxcontext, status); |
7644 | 0 | varcontext.prec = 2*klist[i] + 5; |
7645 | 0 | if (v->digits > varcontext.prec) { |
7646 | | /* Let t := v, truncated to n >= 2*k+5 fraction digits */ |
7647 | 0 | mpd_qshiftr(&t, v, v->digits-varcontext.prec, status); |
7648 | 0 | t.exp = -varcontext.prec; |
7649 | | /* Let t := trunc(v)*s, truncated to n >= 2*k+1 fraction digits */ |
7650 | 0 | mpd_qmul(&t, &t, &s, &varcontext, status); |
7651 | 0 | } |
7652 | 0 | else { /* v->digits <= 2*k+5 */ |
7653 | | /* Let t := v*s, truncated to n >= 2*k+1 fraction digits */ |
7654 | 0 | mpd_qmul(&t, v, &s, &varcontext, status); |
7655 | 0 | } |
7656 | | /* Let s := 2*z, exact result */ |
7657 | 0 | _mpd_qmul_exact(&s, z, &two, &maxcontext, status); |
7658 | | /* s.digits < t.digits <= 2*k+5, |adjexp(s)-adjexp(t)| <= 1, |
7659 | | * so the subtraction generates at most 2*k+6 <= klist[i+1]+7 |
7660 | | * digits. The loop invariant is preserved. */ |
7661 | 0 | _mpd_qsub_exact(z, &s, &t, &maxcontext, status); |
7662 | 0 | } |
7663 | |
|
7664 | 0 | if (!mpd_isspecial(z)) { |
7665 | 0 | z->exp -= adj; |
7666 | 0 | mpd_set_flags(z, sign); |
7667 | 0 | } |
7668 | |
|
7669 | 0 | mpd_del(&s); |
7670 | 0 | mpd_del(&t); |
7671 | 0 | mpd_qfinalize(z, ctx, status); |
7672 | 0 | } |
7673 | | |
7674 | | /* |
7675 | | * Internal function for large numbers: |
7676 | | * |
7677 | | * q, r = divmod(coeff(a), coeff(b)) |
7678 | | * |
7679 | | * Strategy: Multiply the dividend by the reciprocal of the divisor. The |
7680 | | * inexact result is fixed by a small loop, using at most one iteration. |
7681 | | * |
7682 | | * ACL2 proofs: |
7683 | | * ------------ |
7684 | | * 1) q is a natural number. (ndivmod-quotient-natp) |
7685 | | * 2) r is a natural number. (ndivmod-remainder-natp) |
7686 | | * 3) a = q * b + r (ndivmod-q*b+r==a) |
7687 | | * 4) r < b (ndivmod-remainder-<-b) |
7688 | | */ |
7689 | | static void |
7690 | | _mpd_base_ndivmod(mpd_t *q, mpd_t *r, const mpd_t *a, const mpd_t *b, |
7691 | | uint32_t *status) |
7692 | 0 | { |
7693 | 0 | mpd_context_t workctx; |
7694 | 0 | mpd_t *qq = q, *rr = r; |
7695 | 0 | mpd_t aa, bb; |
7696 | 0 | int k; |
7697 | |
|
7698 | 0 | _mpd_copy_shared(&aa, a); |
7699 | 0 | _mpd_copy_shared(&bb, b); |
7700 | |
|
7701 | 0 | mpd_set_positive(&aa); |
7702 | 0 | mpd_set_positive(&bb); |
7703 | 0 | aa.exp = 0; |
7704 | 0 | bb.exp = 0; |
7705 | |
|
7706 | 0 | if (q == a || q == b) { |
7707 | 0 | if ((qq = mpd_qnew()) == NULL) { |
7708 | 0 | *status |= MPD_Malloc_error; |
7709 | 0 | goto nanresult; |
7710 | 0 | } |
7711 | 0 | } |
7712 | 0 | if (r == a || r == b) { |
7713 | 0 | if ((rr = mpd_qnew()) == NULL) { |
7714 | 0 | *status |= MPD_Malloc_error; |
7715 | 0 | goto nanresult; |
7716 | 0 | } |
7717 | 0 | } |
7718 | | |
7719 | 0 | mpd_maxcontext(&workctx); |
7720 | | |
7721 | | /* Let prec := adigits - bdigits + 4 */ |
7722 | 0 | workctx.prec = a->digits - b->digits + 1 + 3; |
7723 | 0 | if (a->digits > MPD_MAX_PREC || workctx.prec > MPD_MAX_PREC) { |
7724 | 0 | *status |= MPD_Division_impossible; |
7725 | 0 | goto nanresult; |
7726 | 0 | } |
7727 | | |
7728 | | /* Let x := _mpd_qreciprocal(b, prec) |
7729 | | * Then x is bounded by: |
7730 | | * 1) 1/b - 10**(-prec - bdigits) < x < 1/b + 10**(-prec - bdigits) |
7731 | | * 2) 1/b - 10**(-adigits - 4) < x < 1/b + 10**(-adigits - 4) |
7732 | | */ |
7733 | 0 | _mpd_qreciprocal(rr, &bb, &workctx, &workctx.status); |
7734 | | |
7735 | | /* Get an estimate for the quotient. Let q := a * x |
7736 | | * Then q is bounded by: |
7737 | | * 3) a/b - 10**-4 < q < a/b + 10**-4 |
7738 | | */ |
7739 | 0 | _mpd_qmul(qq, &aa, rr, &workctx, &workctx.status); |
7740 | | /* Truncate q to an integer: |
7741 | | * 4) a/b - 2 < trunc(q) < a/b + 1 |
7742 | | */ |
7743 | 0 | mpd_qtrunc(qq, qq, &workctx, &workctx.status); |
7744 | |
|
7745 | 0 | workctx.prec = aa.digits + 3; |
7746 | 0 | workctx.emax = MPD_MAX_EMAX + 3; |
7747 | 0 | workctx.emin = MPD_MIN_EMIN - 3; |
7748 | | /* Multiply the estimate for q by b: |
7749 | | * 5) a - 2 * b < trunc(q) * b < a + b |
7750 | | */ |
7751 | 0 | _mpd_qmul(rr, &bb, qq, &workctx, &workctx.status); |
7752 | | /* Get the estimate for r such that a = q * b + r. */ |
7753 | 0 | _mpd_qsub_exact(rr, &aa, rr, &workctx, &workctx.status); |
7754 | | |
7755 | | /* Fix the result. At this point -b < r < 2*b, so the correction loop |
7756 | | takes at most one iteration. */ |
7757 | 0 | for (k = 0;; k++) { |
7758 | 0 | if (mpd_isspecial(qq) || mpd_isspecial(rr)) { |
7759 | 0 | *status |= (workctx.status&MPD_Errors); |
7760 | 0 | goto nanresult; |
7761 | 0 | } |
7762 | 0 | if (k > 2) { /* Allow two iterations despite the proof. */ |
7763 | 0 | mpd_err_warn("libmpdec: internal error in " /* GCOV_NOT_REACHED */ |
7764 | 0 | "_mpd_base_ndivmod: please report"); /* GCOV_NOT_REACHED */ |
7765 | 0 | *status |= MPD_Invalid_operation; /* GCOV_NOT_REACHED */ |
7766 | 0 | goto nanresult; /* GCOV_NOT_REACHED */ |
7767 | 0 | } |
7768 | | /* r < 0 */ |
7769 | 0 | else if (_mpd_cmp(&zero, rr) == 1) { |
7770 | 0 | _mpd_qadd_exact(rr, rr, &bb, &workctx, &workctx.status); |
7771 | 0 | _mpd_qadd_exact(qq, qq, &minus_one, &workctx, &workctx.status); |
7772 | 0 | } |
7773 | | /* 0 <= r < b */ |
7774 | 0 | else if (_mpd_cmp(rr, &bb) == -1) { |
7775 | 0 | break; |
7776 | 0 | } |
7777 | | /* r >= b */ |
7778 | 0 | else { |
7779 | 0 | _mpd_qsub_exact(rr, rr, &bb, &workctx, &workctx.status); |
7780 | 0 | _mpd_qadd_exact(qq, qq, &one, &workctx, &workctx.status); |
7781 | 0 | } |
7782 | 0 | } |
7783 | | |
7784 | 0 | if (qq != q) { |
7785 | 0 | if (!mpd_qcopy(q, qq, status)) { |
7786 | 0 | goto nanresult; /* GCOV_UNLIKELY */ |
7787 | 0 | } |
7788 | 0 | mpd_del(qq); |
7789 | 0 | } |
7790 | 0 | if (rr != r) { |
7791 | 0 | if (!mpd_qcopy(r, rr, status)) { |
7792 | 0 | goto nanresult; /* GCOV_UNLIKELY */ |
7793 | 0 | } |
7794 | 0 | mpd_del(rr); |
7795 | 0 | } |
7796 | | |
7797 | 0 | *status |= (workctx.status&MPD_Errors); |
7798 | 0 | return; |
7799 | | |
7800 | | |
7801 | 0 | nanresult: |
7802 | 0 | if (qq && qq != q) mpd_del(qq); |
7803 | 0 | if (rr && rr != r) mpd_del(rr); |
7804 | 0 | mpd_setspecial(q, MPD_POS, MPD_NAN); |
7805 | 0 | mpd_setspecial(r, MPD_POS, MPD_NAN); |
7806 | 0 | } |
7807 | | |
7808 | | /* LIBMPDEC_ONLY */ |
7809 | | /* |
7810 | | * Schedule the optimal precision increase for the Newton iteration. |
7811 | | * v := input operand |
7812 | | * z_0 := initial approximation |
7813 | | * initprec := natural number such that abs(sqrt(v) - z_0) < 10**-initprec |
7814 | | * maxprec := target precision |
7815 | | * |
7816 | | * For convenience the output klist contains the elements in reverse order: |
7817 | | * klist := [k_n-1, ..., k_0], where |
7818 | | * 1) k_0 <= initprec and |
7819 | | * 2) abs(sqrt(v) - result) < 10**(-2*k_n-1 + 2) <= 10**-maxprec. |
7820 | | */ |
7821 | | static inline int |
7822 | | invroot_schedule_prec(mpd_ssize_t klist[MPD_MAX_PREC_LOG2], |
7823 | | mpd_ssize_t maxprec, mpd_ssize_t initprec) |
7824 | 0 | { |
7825 | 0 | mpd_ssize_t k; |
7826 | 0 | int i; |
7827 | |
|
7828 | 0 | assert(maxprec >= 3 && initprec >= 3); |
7829 | 0 | if (maxprec <= initprec) return -1; |
7830 | | |
7831 | 0 | i = 0; k = maxprec; |
7832 | 0 | do { |
7833 | 0 | k = (k+3) / 2; |
7834 | 0 | klist[i++] = k; |
7835 | 0 | } while (k > initprec); |
7836 | |
|
7837 | 0 | return i-1; |
7838 | 0 | } |
7839 | | |
7840 | | /* |
7841 | | * Initial approximation for the inverse square root function. |
7842 | | * Input: |
7843 | | * v := rational number, with 1 <= v < 100 |
7844 | | * vhat := floor(v * 10**6) |
7845 | | * Output: |
7846 | | * z := approximation to 1/sqrt(v), such that abs(z - 1/sqrt(v)) < 10**-3. |
7847 | | */ |
7848 | | static inline void |
7849 | | _invroot_init_approx(mpd_t *z, mpd_uint_t vhat) |
7850 | 0 | { |
7851 | 0 | mpd_uint_t lo = 1000; |
7852 | 0 | mpd_uint_t hi = 10000; |
7853 | 0 | mpd_uint_t a, sq; |
7854 | |
|
7855 | 0 | assert(lo*lo <= vhat && vhat < (hi+1)*(hi+1)); |
7856 | |
|
7857 | 0 | for(;;) { |
7858 | 0 | a = (lo + hi) / 2; |
7859 | 0 | sq = a * a; |
7860 | 0 | if (vhat >= sq) { |
7861 | 0 | if (vhat < sq + 2*a + 1) { |
7862 | 0 | break; |
7863 | 0 | } |
7864 | 0 | lo = a + 1; |
7865 | 0 | } |
7866 | 0 | else { |
7867 | 0 | hi = a - 1; |
7868 | 0 | } |
7869 | 0 | } |
7870 | | |
7871 | | /* |
7872 | | * After the binary search we have: |
7873 | | * 1) a**2 <= floor(v * 10**6) < (a + 1)**2 |
7874 | | * This implies: |
7875 | | * 2) a**2 <= v * 10**6 < (a + 1)**2 |
7876 | | * 3) a <= sqrt(v) * 10**3 < a + 1 |
7877 | | * Since 10**3 <= a: |
7878 | | * 4) 0 <= 10**prec/a - 1/sqrt(v) < 10**-prec |
7879 | | * We have: |
7880 | | * 5) 10**3/a - 10**-3 < floor(10**9/a) * 10**-6 <= 10**3/a |
7881 | | * Merging 4) and 5): |
7882 | | * 6) abs(floor(10**9/a) * 10**-6 - 1/sqrt(v)) < 10**-3 |
7883 | | */ |
7884 | 0 | mpd_minalloc(z); |
7885 | 0 | mpd_clear_flags(z); |
7886 | 0 | z->data[0] = 1000000000UL / a; |
7887 | 0 | z->len = 1; |
7888 | 0 | z->exp = -6; |
7889 | 0 | mpd_setdigits(z); |
7890 | 0 | } |
7891 | | |
7892 | | /* |
7893 | | * Set 'result' to 1/sqrt(a). |
7894 | | * Relative error: abs(result - 1/sqrt(a)) < 10**-prec * 1/sqrt(a) |
7895 | | */ |
7896 | | static void |
7897 | | _mpd_qinvroot(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7898 | | uint32_t *status) |
7899 | 0 | { |
7900 | 0 | uint32_t workstatus = 0; |
7901 | 0 | mpd_context_t varcontext, maxcontext; |
7902 | 0 | mpd_t *z = result; /* current approximation */ |
7903 | 0 | mpd_t *v; /* a, normalized to a number between 1 and 100 */ |
7904 | 0 | MPD_NEW_SHARED(vtmp, a); /* by default v will share data with a */ |
7905 | 0 | MPD_NEW_STATIC(s,0,0,0,0); /* temporary variable */ |
7906 | 0 | MPD_NEW_STATIC(t,0,0,0,0); /* temporary variable */ |
7907 | 0 | MPD_NEW_CONST(one_half,0,-1,1,1,1,5); |
7908 | 0 | MPD_NEW_CONST(three,0,0,1,1,1,3); |
7909 | 0 | mpd_ssize_t klist[MPD_MAX_PREC_LOG2]; |
7910 | 0 | mpd_ssize_t ideal_exp, shift; |
7911 | 0 | mpd_ssize_t adj, tz; |
7912 | 0 | mpd_ssize_t maxprec, fracdigits; |
7913 | 0 | mpd_uint_t vhat, dummy; |
7914 | 0 | int i, n; |
7915 | | |
7916 | |
|
7917 | 0 | ideal_exp = -(a->exp - (a->exp & 1)) / 2; |
7918 | |
|
7919 | 0 | v = &vtmp; |
7920 | 0 | if (result == a) { |
7921 | 0 | if ((v = mpd_qncopy(a)) == NULL) { |
7922 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
7923 | 0 | return; |
7924 | 0 | } |
7925 | 0 | } |
7926 | | |
7927 | | /* normalize a to 1 <= v < 100 */ |
7928 | 0 | if ((v->digits+v->exp) & 1) { |
7929 | 0 | fracdigits = v->digits - 1; |
7930 | 0 | v->exp = -fracdigits; |
7931 | 0 | n = (v->digits > 7) ? 7 : (int)v->digits; |
7932 | | /* Let vhat := floor(v * 10**(2*initprec)) */ |
7933 | 0 | _mpd_get_msdigits(&dummy, &vhat, v, n); |
7934 | 0 | if (n < 7) { |
7935 | 0 | vhat *= mpd_pow10[7-n]; |
7936 | 0 | } |
7937 | 0 | } |
7938 | 0 | else { |
7939 | 0 | fracdigits = v->digits - 2; |
7940 | 0 | v->exp = -fracdigits; |
7941 | 0 | n = (v->digits > 8) ? 8 : (int)v->digits; |
7942 | | /* Let vhat := floor(v * 10**(2*initprec)) */ |
7943 | 0 | _mpd_get_msdigits(&dummy, &vhat, v, n); |
7944 | 0 | if (n < 8) { |
7945 | 0 | vhat *= mpd_pow10[8-n]; |
7946 | 0 | } |
7947 | 0 | } |
7948 | 0 | adj = (a->exp-v->exp) / 2; |
7949 | | |
7950 | | /* initial approximation */ |
7951 | 0 | _invroot_init_approx(z, vhat); |
7952 | |
|
7953 | 0 | mpd_maxcontext(&maxcontext); |
7954 | 0 | mpd_maxcontext(&varcontext); |
7955 | 0 | varcontext.round = MPD_ROUND_TRUNC; |
7956 | 0 | maxprec = ctx->prec + 1; |
7957 | | |
7958 | | /* initprec == 3 */ |
7959 | 0 | i = invroot_schedule_prec(klist, maxprec, 3); |
7960 | 0 | for (; i >= 0; i--) { |
7961 | 0 | varcontext.prec = 2*klist[i]+2; |
7962 | 0 | mpd_qmul(&s, z, z, &maxcontext, &workstatus); |
7963 | 0 | if (v->digits > varcontext.prec) { |
7964 | 0 | shift = v->digits - varcontext.prec; |
7965 | 0 | mpd_qshiftr(&t, v, shift, &workstatus); |
7966 | 0 | t.exp += shift; |
7967 | 0 | mpd_qmul(&t, &t, &s, &varcontext, &workstatus); |
7968 | 0 | } |
7969 | 0 | else { |
7970 | 0 | mpd_qmul(&t, v, &s, &varcontext, &workstatus); |
7971 | 0 | } |
7972 | 0 | mpd_qsub(&t, &three, &t, &maxcontext, &workstatus); |
7973 | 0 | mpd_qmul(z, z, &t, &varcontext, &workstatus); |
7974 | 0 | mpd_qmul(z, z, &one_half, &maxcontext, &workstatus); |
7975 | 0 | } |
7976 | |
|
7977 | 0 | z->exp -= adj; |
7978 | |
|
7979 | 0 | tz = mpd_trail_zeros(result); |
7980 | 0 | shift = ideal_exp - result->exp; |
7981 | 0 | shift = (tz > shift) ? shift : tz; |
7982 | 0 | if (shift > 0) { |
7983 | 0 | mpd_qshiftr_inplace(result, shift); |
7984 | 0 | result->exp += shift; |
7985 | 0 | } |
7986 | | |
7987 | |
|
7988 | 0 | mpd_del(&s); |
7989 | 0 | mpd_del(&t); |
7990 | 0 | if (v != &vtmp) mpd_del(v); |
7991 | 0 | *status |= (workstatus&MPD_Errors); |
7992 | 0 | *status |= (MPD_Rounded|MPD_Inexact); |
7993 | 0 | } |
7994 | | |
7995 | | void |
7996 | | mpd_qinvroot(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
7997 | | uint32_t *status) |
7998 | 0 | { |
7999 | 0 | mpd_context_t workctx; |
8000 | |
|
8001 | 0 | if (mpd_isspecial(a)) { |
8002 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
8003 | 0 | return; |
8004 | 0 | } |
8005 | 0 | if (mpd_isnegative(a)) { |
8006 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
8007 | 0 | return; |
8008 | 0 | } |
8009 | | /* positive infinity */ |
8010 | 0 | _settriple(result, MPD_POS, 0, mpd_etiny(ctx)); |
8011 | 0 | *status |= MPD_Clamped; |
8012 | 0 | return; |
8013 | 0 | } |
8014 | 0 | if (mpd_iszero(a)) { |
8015 | 0 | mpd_setspecial(result, mpd_sign(a), MPD_INF); |
8016 | 0 | *status |= MPD_Division_by_zero; |
8017 | 0 | return; |
8018 | 0 | } |
8019 | 0 | if (mpd_isnegative(a)) { |
8020 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
8021 | 0 | return; |
8022 | 0 | } |
8023 | | |
8024 | 0 | mpd_workcontext(&workctx, ctx); |
8025 | 0 | workctx.prec += 2; |
8026 | 0 | workctx.round = MPD_ROUND_HALF_EVEN; |
8027 | 0 | _mpd_qinvroot(result, a, &workctx, status); |
8028 | 0 | mpd_qfinalize(result, ctx, status); |
8029 | 0 | } |
8030 | | /* END LIBMPDEC_ONLY */ |
8031 | | |
8032 | | /* Algorithm from decimal.py */ |
8033 | | static void |
8034 | | _mpd_qsqrt(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
8035 | | uint32_t *status) |
8036 | 0 | { |
8037 | 0 | mpd_context_t maxcontext; |
8038 | 0 | MPD_NEW_STATIC(c,0,0,0,0); |
8039 | 0 | MPD_NEW_STATIC(q,0,0,0,0); |
8040 | 0 | MPD_NEW_STATIC(r,0,0,0,0); |
8041 | 0 | MPD_NEW_CONST(two,0,0,1,1,1,2); |
8042 | 0 | mpd_ssize_t prec, ideal_exp; |
8043 | 0 | mpd_ssize_t l, shift; |
8044 | 0 | int exact = 0; |
8045 | | |
8046 | |
|
8047 | 0 | ideal_exp = (a->exp - (a->exp & 1)) / 2; |
8048 | |
|
8049 | 0 | if (mpd_isspecial(a)) { |
8050 | 0 | if (mpd_qcheck_nan(result, a, ctx, status)) { |
8051 | 0 | return; |
8052 | 0 | } |
8053 | 0 | if (mpd_isnegative(a)) { |
8054 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
8055 | 0 | return; |
8056 | 0 | } |
8057 | 0 | mpd_setspecial(result, MPD_POS, MPD_INF); |
8058 | 0 | return; |
8059 | 0 | } |
8060 | 0 | if (mpd_iszero(a)) { |
8061 | 0 | _settriple(result, mpd_sign(a), 0, ideal_exp); |
8062 | 0 | mpd_qfinalize(result, ctx, status); |
8063 | 0 | return; |
8064 | 0 | } |
8065 | 0 | if (mpd_isnegative(a)) { |
8066 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
8067 | 0 | return; |
8068 | 0 | } |
8069 | | |
8070 | 0 | mpd_maxcontext(&maxcontext); |
8071 | 0 | prec = ctx->prec + 1; |
8072 | |
|
8073 | 0 | if (!mpd_qcopy(&c, a, status)) { |
8074 | 0 | goto malloc_error; |
8075 | 0 | } |
8076 | 0 | c.exp = 0; |
8077 | |
|
8078 | 0 | if (a->exp & 1) { |
8079 | 0 | if (!mpd_qshiftl(&c, &c, 1, status)) { |
8080 | 0 | goto malloc_error; |
8081 | 0 | } |
8082 | 0 | l = (a->digits >> 1) + 1; |
8083 | 0 | } |
8084 | 0 | else { |
8085 | 0 | l = (a->digits + 1) >> 1; |
8086 | 0 | } |
8087 | | |
8088 | 0 | shift = prec - l; |
8089 | 0 | if (shift >= 0) { |
8090 | 0 | if (!mpd_qshiftl(&c, &c, 2*shift, status)) { |
8091 | 0 | goto malloc_error; |
8092 | 0 | } |
8093 | 0 | exact = 1; |
8094 | 0 | } |
8095 | 0 | else { |
8096 | 0 | exact = !mpd_qshiftr_inplace(&c, -2*shift); |
8097 | 0 | } |
8098 | | |
8099 | 0 | ideal_exp -= shift; |
8100 | | |
8101 | | /* find result = floor(sqrt(c)) using Newton's method */ |
8102 | 0 | if (!mpd_qshiftl(result, &one, prec, status)) { |
8103 | 0 | goto malloc_error; |
8104 | 0 | } |
8105 | | |
8106 | 0 | while (1) { |
8107 | 0 | _mpd_qdivmod(&q, &r, &c, result, &maxcontext, &maxcontext.status); |
8108 | 0 | if (mpd_isspecial(result) || mpd_isspecial(&q)) { |
8109 | 0 | mpd_seterror(result, maxcontext.status&MPD_Errors, status); |
8110 | 0 | goto out; |
8111 | 0 | } |
8112 | 0 | if (_mpd_cmp(result, &q) <= 0) { |
8113 | 0 | break; |
8114 | 0 | } |
8115 | 0 | _mpd_qadd_exact(result, result, &q, &maxcontext, &maxcontext.status); |
8116 | 0 | if (mpd_isspecial(result)) { |
8117 | 0 | mpd_seterror(result, maxcontext.status&MPD_Errors, status); |
8118 | 0 | goto out; |
8119 | 0 | } |
8120 | 0 | _mpd_qdivmod(result, &r, result, &two, &maxcontext, &maxcontext.status); |
8121 | 0 | } |
8122 | | |
8123 | 0 | if (exact) { |
8124 | 0 | _mpd_qmul_exact(&r, result, result, &maxcontext, &maxcontext.status); |
8125 | 0 | if (mpd_isspecial(&r)) { |
8126 | 0 | mpd_seterror(result, maxcontext.status&MPD_Errors, status); |
8127 | 0 | goto out; |
8128 | 0 | } |
8129 | 0 | exact = (_mpd_cmp(&r, &c) == 0); |
8130 | 0 | } |
8131 | | |
8132 | 0 | if (exact) { |
8133 | 0 | if (shift >= 0) { |
8134 | 0 | mpd_qshiftr_inplace(result, shift); |
8135 | 0 | } |
8136 | 0 | else { |
8137 | 0 | if (!mpd_qshiftl(result, result, -shift, status)) { |
8138 | 0 | goto malloc_error; |
8139 | 0 | } |
8140 | 0 | } |
8141 | 0 | ideal_exp += shift; |
8142 | 0 | } |
8143 | 0 | else { |
8144 | 0 | int lsd = (int)mpd_lsd(result->data[0]); |
8145 | 0 | if (lsd == 0 || lsd == 5) { |
8146 | 0 | result->data[0] += 1; |
8147 | 0 | } |
8148 | 0 | } |
8149 | | |
8150 | 0 | result->exp = ideal_exp; |
8151 | | |
8152 | |
|
8153 | 0 | out: |
8154 | 0 | mpd_del(&c); |
8155 | 0 | mpd_del(&q); |
8156 | 0 | mpd_del(&r); |
8157 | 0 | mpd_workcontext(&maxcontext, ctx); |
8158 | 0 | maxcontext.round = MPD_ROUND_HALF_EVEN; |
8159 | 0 | mpd_qfinalize(result, &maxcontext, status); |
8160 | 0 | return; |
8161 | | |
8162 | 0 | malloc_error: |
8163 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
8164 | 0 | goto out; |
8165 | 0 | } |
8166 | | |
8167 | | void |
8168 | | mpd_qsqrt(mpd_t *result, const mpd_t *a, const mpd_context_t *ctx, |
8169 | | uint32_t *status) |
8170 | 0 | { |
8171 | 0 | MPD_NEW_STATIC(aa,0,0,0,0); |
8172 | 0 | uint32_t xstatus = 0; |
8173 | |
|
8174 | 0 | if (result == a) { |
8175 | 0 | if (!mpd_qcopy(&aa, a, status)) { |
8176 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
8177 | 0 | goto out; |
8178 | 0 | } |
8179 | 0 | a = &aa; |
8180 | 0 | } |
8181 | | |
8182 | 0 | _mpd_qsqrt(result, a, ctx, &xstatus); |
8183 | |
|
8184 | 0 | if (xstatus & (MPD_Malloc_error|MPD_Division_impossible)) { |
8185 | | /* The above conditions can occur at very high context precisions |
8186 | | * if intermediate values get too large. Retry the operation with |
8187 | | * a lower context precision in case the result is exact. |
8188 | | * |
8189 | | * If the result is exact, an upper bound for the number of digits |
8190 | | * is the number of digits in the input. |
8191 | | * |
8192 | | * NOTE: sqrt(40e9) = 2.0e+5 /\ digits(40e9) = digits(2.0e+5) = 2 |
8193 | | */ |
8194 | 0 | uint32_t ystatus = 0; |
8195 | 0 | mpd_context_t workctx; |
8196 | 0 | mpd_workcontext(&workctx, ctx); |
8197 | |
|
8198 | 0 | workctx.prec = a->digits; |
8199 | 0 | if (workctx.prec >= ctx->prec) { |
8200 | 0 | *status |= (xstatus|MPD_Errors); |
8201 | 0 | goto out; /* No point in repeating this, keep the original error. */ |
8202 | 0 | } |
8203 | | |
8204 | 0 | _mpd_qsqrt(result, a, &workctx, &ystatus); |
8205 | 0 | if (ystatus != 0) { |
8206 | 0 | ystatus = *status | ((xstatus|ystatus)&MPD_Errors); |
8207 | 0 | mpd_seterror(result, ystatus, status); |
8208 | 0 | } |
8209 | 0 | } |
8210 | 0 | else { |
8211 | 0 | *status |= xstatus; |
8212 | 0 | } |
8213 | | |
8214 | 0 | out: |
8215 | 0 | mpd_del(&aa); |
8216 | 0 | } |
8217 | | |
8218 | | |
8219 | | /******************************************************************************/ |
8220 | | /* Base conversions */ |
8221 | | /******************************************************************************/ |
8222 | | |
8223 | | /* Space needed to represent an integer mpd_t in base 'base'. */ |
8224 | | size_t |
8225 | | mpd_sizeinbase(const mpd_t *a, uint32_t base) |
8226 | 0 | { |
8227 | 0 | double x; |
8228 | 0 | size_t digits; |
8229 | 0 | double upper_bound; |
8230 | |
|
8231 | 0 | assert(mpd_isinteger(a)); |
8232 | 0 | assert(base >= 2); |
8233 | |
|
8234 | 0 | if (mpd_iszero(a)) { |
8235 | 0 | return 1; |
8236 | 0 | } |
8237 | | |
8238 | 0 | digits = a->digits+a->exp; |
8239 | |
|
8240 | 0 | #ifdef CONFIG_64 |
8241 | | /* ceil(2711437152599294 / log10(2)) + 4 == 2**53 */ |
8242 | 0 | if (digits > 2711437152599294ULL) { |
8243 | 0 | return SIZE_MAX; |
8244 | 0 | } |
8245 | | |
8246 | 0 | upper_bound = (double)((1ULL<<53)-1); |
8247 | | #else |
8248 | | upper_bound = (double)(SIZE_MAX-1); |
8249 | | #endif |
8250 | |
|
8251 | 0 | x = (double)digits / log10(base); |
8252 | 0 | return (x > upper_bound) ? SIZE_MAX : (size_t)x + 1; |
8253 | 0 | } |
8254 | | |
8255 | | /* Space needed to import a base 'base' integer of length 'srclen'. */ |
8256 | | static mpd_ssize_t |
8257 | | _mpd_importsize(size_t srclen, uint32_t base) |
8258 | 0 | { |
8259 | 0 | double x; |
8260 | 0 | double upper_bound; |
8261 | |
|
8262 | 0 | assert(srclen > 0); |
8263 | 0 | assert(base >= 2); |
8264 | |
|
8265 | 0 | #if SIZE_MAX == UINT64_MAX |
8266 | 0 | if (srclen > (1ULL<<53)) { |
8267 | 0 | return MPD_SSIZE_MAX; |
8268 | 0 | } |
8269 | | |
8270 | 0 | assert((1ULL<<53) <= MPD_MAXIMPORT); |
8271 | 0 | upper_bound = (double)((1ULL<<53)-1); |
8272 | | #else |
8273 | | upper_bound = MPD_MAXIMPORT-1; |
8274 | | #endif |
8275 | |
|
8276 | 0 | x = (double)srclen * (log10(base)/MPD_RDIGITS); |
8277 | 0 | return (x > upper_bound) ? MPD_SSIZE_MAX : (mpd_ssize_t)x + 1; |
8278 | 0 | } |
8279 | | |
8280 | | static uint8_t |
8281 | | mpd_resize_u16(uint16_t **w, size_t nmemb) |
8282 | 0 | { |
8283 | 0 | uint8_t err = 0; |
8284 | 0 | *w = mpd_realloc(*w, nmemb, sizeof **w, &err); |
8285 | 0 | return !err; |
8286 | 0 | } |
8287 | | |
8288 | | static uint8_t |
8289 | | mpd_resize_u32(uint32_t **w, size_t nmemb) |
8290 | 0 | { |
8291 | 0 | uint8_t err = 0; |
8292 | 0 | *w = mpd_realloc(*w, nmemb, sizeof **w, &err); |
8293 | 0 | return !err; |
8294 | 0 | } |
8295 | | |
8296 | | static size_t |
8297 | | _baseconv_to_u16(uint16_t **w, size_t wlen, mpd_uint_t wbase, |
8298 | | mpd_uint_t *u, mpd_ssize_t ulen) |
8299 | 0 | { |
8300 | 0 | size_t n = 0; |
8301 | |
|
8302 | 0 | assert(wlen > 0 && ulen > 0); |
8303 | 0 | assert(wbase <= (1U<<16)); |
8304 | |
|
8305 | 0 | do { |
8306 | 0 | if (n >= wlen) { |
8307 | 0 | if (!mpd_resize_u16(w, n+1)) { |
8308 | 0 | return SIZE_MAX; |
8309 | 0 | } |
8310 | 0 | wlen = n+1; |
8311 | 0 | } |
8312 | 0 | (*w)[n++] = (uint16_t)_mpd_shortdiv(u, u, ulen, wbase); |
8313 | | /* ulen is at least 1. u[ulen-1] can only be zero if ulen == 1. */ |
8314 | 0 | ulen = _mpd_real_size(u, ulen); |
8315 | |
|
8316 | 0 | } while (u[ulen-1] != 0); |
8317 | | |
8318 | 0 | return n; |
8319 | 0 | } |
8320 | | |
8321 | | static size_t |
8322 | | _coeff_from_u16(mpd_t *w, mpd_ssize_t wlen, |
8323 | | const mpd_uint_t *u, size_t ulen, uint32_t ubase, |
8324 | | uint32_t *status) |
8325 | 0 | { |
8326 | 0 | mpd_ssize_t n = 0; |
8327 | 0 | mpd_uint_t carry; |
8328 | |
|
8329 | 0 | assert(wlen > 0 && ulen > 0); |
8330 | 0 | assert(ubase <= (1U<<16)); |
8331 | |
|
8332 | 0 | w->data[n++] = u[--ulen]; |
8333 | 0 | while (--ulen != SIZE_MAX) { |
8334 | 0 | carry = _mpd_shortmul_c(w->data, w->data, n, ubase); |
8335 | 0 | if (carry) { |
8336 | 0 | if (n >= wlen) { |
8337 | 0 | if (!mpd_qresize(w, n+1, status)) { |
8338 | 0 | return SIZE_MAX; |
8339 | 0 | } |
8340 | 0 | wlen = n+1; |
8341 | 0 | } |
8342 | 0 | w->data[n++] = carry; |
8343 | 0 | } |
8344 | 0 | carry = _mpd_shortadd(w->data, n, u[ulen]); |
8345 | 0 | if (carry) { |
8346 | 0 | if (n >= wlen) { |
8347 | 0 | if (!mpd_qresize(w, n+1, status)) { |
8348 | 0 | return SIZE_MAX; |
8349 | 0 | } |
8350 | 0 | wlen = n+1; |
8351 | 0 | } |
8352 | 0 | w->data[n++] = carry; |
8353 | 0 | } |
8354 | 0 | } |
8355 | | |
8356 | 0 | return n; |
8357 | 0 | } |
8358 | | |
8359 | | /* target base wbase < source base ubase */ |
8360 | | static size_t |
8361 | | _baseconv_to_smaller(uint32_t **w, size_t wlen, uint32_t wbase, |
8362 | | mpd_uint_t *u, mpd_ssize_t ulen, mpd_uint_t ubase) |
8363 | 0 | { |
8364 | 0 | size_t n = 0; |
8365 | |
|
8366 | 0 | assert(wlen > 0 && ulen > 0); |
8367 | 0 | assert(wbase < ubase); |
8368 | |
|
8369 | 0 | do { |
8370 | 0 | if (n >= wlen) { |
8371 | 0 | if (!mpd_resize_u32(w, n+1)) { |
8372 | 0 | return SIZE_MAX; |
8373 | 0 | } |
8374 | 0 | wlen = n+1; |
8375 | 0 | } |
8376 | 0 | (*w)[n++] = (uint32_t)_mpd_shortdiv_b(u, u, ulen, wbase, ubase); |
8377 | | /* ulen is at least 1. u[ulen-1] can only be zero if ulen == 1. */ |
8378 | 0 | ulen = _mpd_real_size(u, ulen); |
8379 | |
|
8380 | 0 | } while (u[ulen-1] != 0); |
8381 | | |
8382 | 0 | return n; |
8383 | 0 | } |
8384 | | |
8385 | | #ifdef CONFIG_32 |
8386 | | /* target base 'wbase' == source base 'ubase' */ |
8387 | | static size_t |
8388 | | _copy_equal_base(uint32_t **w, size_t wlen, |
8389 | | const uint32_t *u, size_t ulen) |
8390 | | { |
8391 | | if (wlen < ulen) { |
8392 | | if (!mpd_resize_u32(w, ulen)) { |
8393 | | return SIZE_MAX; |
8394 | | } |
8395 | | } |
8396 | | |
8397 | | memcpy(*w, u, ulen * (sizeof **w)); |
8398 | | return ulen; |
8399 | | } |
8400 | | |
8401 | | /* target base 'wbase' > source base 'ubase' */ |
8402 | | static size_t |
8403 | | _baseconv_to_larger(uint32_t **w, size_t wlen, mpd_uint_t wbase, |
8404 | | const mpd_uint_t *u, size_t ulen, mpd_uint_t ubase) |
8405 | | { |
8406 | | size_t n = 0; |
8407 | | mpd_uint_t carry; |
8408 | | |
8409 | | assert(wlen > 0 && ulen > 0); |
8410 | | assert(ubase < wbase); |
8411 | | |
8412 | | (*w)[n++] = u[--ulen]; |
8413 | | while (--ulen != SIZE_MAX) { |
8414 | | carry = _mpd_shortmul_b(*w, *w, n, ubase, wbase); |
8415 | | if (carry) { |
8416 | | if (n >= wlen) { |
8417 | | if (!mpd_resize_u32(w, n+1)) { |
8418 | | return SIZE_MAX; |
8419 | | } |
8420 | | wlen = n+1; |
8421 | | } |
8422 | | (*w)[n++] = carry; |
8423 | | } |
8424 | | carry = _mpd_shortadd_b(*w, n, u[ulen], wbase); |
8425 | | if (carry) { |
8426 | | if (n >= wlen) { |
8427 | | if (!mpd_resize_u32(w, n+1)) { |
8428 | | return SIZE_MAX; |
8429 | | } |
8430 | | wlen = n+1; |
8431 | | } |
8432 | | (*w)[n++] = carry; |
8433 | | } |
8434 | | } |
8435 | | |
8436 | | return n; |
8437 | | } |
8438 | | |
8439 | | /* target base wbase < source base ubase */ |
8440 | | static size_t |
8441 | | _coeff_from_larger_base(mpd_t *w, size_t wlen, mpd_uint_t wbase, |
8442 | | mpd_uint_t *u, mpd_ssize_t ulen, mpd_uint_t ubase, |
8443 | | uint32_t *status) |
8444 | | { |
8445 | | size_t n = 0; |
8446 | | |
8447 | | assert(wlen > 0 && ulen > 0); |
8448 | | assert(wbase < ubase); |
8449 | | |
8450 | | do { |
8451 | | if (n >= wlen) { |
8452 | | if (!mpd_qresize(w, n+1, status)) { |
8453 | | return SIZE_MAX; |
8454 | | } |
8455 | | wlen = n+1; |
8456 | | } |
8457 | | w->data[n++] = (uint32_t)_mpd_shortdiv_b(u, u, ulen, wbase, ubase); |
8458 | | /* ulen is at least 1. u[ulen-1] can only be zero if ulen == 1. */ |
8459 | | ulen = _mpd_real_size(u, ulen); |
8460 | | |
8461 | | } while (u[ulen-1] != 0); |
8462 | | |
8463 | | return n; |
8464 | | } |
8465 | | #endif |
8466 | | |
8467 | | /* target base 'wbase' > source base 'ubase' */ |
8468 | | static size_t |
8469 | | _coeff_from_smaller_base(mpd_t *w, mpd_ssize_t wlen, mpd_uint_t wbase, |
8470 | | const uint32_t *u, size_t ulen, mpd_uint_t ubase, |
8471 | | uint32_t *status) |
8472 | 0 | { |
8473 | 0 | mpd_ssize_t n = 0; |
8474 | 0 | mpd_uint_t carry; |
8475 | |
|
8476 | 0 | assert(wlen > 0 && ulen > 0); |
8477 | 0 | assert(wbase > ubase); |
8478 | |
|
8479 | 0 | w->data[n++] = u[--ulen]; |
8480 | 0 | while (--ulen != SIZE_MAX) { |
8481 | 0 | carry = _mpd_shortmul_b(w->data, w->data, n, ubase, wbase); |
8482 | 0 | if (carry) { |
8483 | 0 | if (n >= wlen) { |
8484 | 0 | if (!mpd_qresize(w, n+1, status)) { |
8485 | 0 | return SIZE_MAX; |
8486 | 0 | } |
8487 | 0 | wlen = n+1; |
8488 | 0 | } |
8489 | 0 | w->data[n++] = carry; |
8490 | 0 | } |
8491 | 0 | carry = _mpd_shortadd_b(w->data, n, u[ulen], wbase); |
8492 | 0 | if (carry) { |
8493 | 0 | if (n >= wlen) { |
8494 | 0 | if (!mpd_qresize(w, n+1, status)) { |
8495 | 0 | return SIZE_MAX; |
8496 | 0 | } |
8497 | 0 | wlen = n+1; |
8498 | 0 | } |
8499 | 0 | w->data[n++] = carry; |
8500 | 0 | } |
8501 | 0 | } |
8502 | | |
8503 | 0 | return n; |
8504 | 0 | } |
8505 | | |
8506 | | /* |
8507 | | * Convert an integer mpd_t to a multiprecision integer with base <= 2**16. |
8508 | | * The least significant word of the result is (*rdata)[0]. |
8509 | | * |
8510 | | * If rdata is NULL, space is allocated by the function and rlen is irrelevant. |
8511 | | * In case of an error any allocated storage is freed and rdata is set back to |
8512 | | * NULL. |
8513 | | * |
8514 | | * If rdata is non-NULL, it MUST be allocated by one of libmpdec's allocation |
8515 | | * functions and rlen MUST be correct. If necessary, the function will resize |
8516 | | * rdata. In case of an error the caller must free rdata. |
8517 | | * |
8518 | | * Return value: In case of success, the exact length of rdata, SIZE_MAX |
8519 | | * otherwise. |
8520 | | */ |
8521 | | size_t |
8522 | | mpd_qexport_u16(uint16_t **rdata, size_t rlen, uint32_t rbase, |
8523 | | const mpd_t *src, uint32_t *status) |
8524 | 0 | { |
8525 | 0 | MPD_NEW_STATIC(tsrc,0,0,0,0); |
8526 | 0 | int alloc = 0; /* rdata == NULL */ |
8527 | 0 | size_t n; |
8528 | |
|
8529 | 0 | assert(rbase <= (1U<<16)); |
8530 | |
|
8531 | 0 | if (mpd_isspecial(src) || !_mpd_isint(src)) { |
8532 | 0 | *status |= MPD_Invalid_operation; |
8533 | 0 | return SIZE_MAX; |
8534 | 0 | } |
8535 | | |
8536 | 0 | if (*rdata == NULL) { |
8537 | 0 | rlen = mpd_sizeinbase(src, rbase); |
8538 | 0 | if (rlen == SIZE_MAX) { |
8539 | 0 | *status |= MPD_Invalid_operation; |
8540 | 0 | return SIZE_MAX; |
8541 | 0 | } |
8542 | 0 | *rdata = mpd_alloc(rlen, sizeof **rdata); |
8543 | 0 | if (*rdata == NULL) { |
8544 | 0 | goto malloc_error; |
8545 | 0 | } |
8546 | 0 | alloc = 1; |
8547 | 0 | } |
8548 | | |
8549 | 0 | if (mpd_iszero(src)) { |
8550 | 0 | **rdata = 0; |
8551 | 0 | return 1; |
8552 | 0 | } |
8553 | | |
8554 | 0 | if (src->exp >= 0) { |
8555 | 0 | if (!mpd_qshiftl(&tsrc, src, src->exp, status)) { |
8556 | 0 | goto malloc_error; |
8557 | 0 | } |
8558 | 0 | } |
8559 | 0 | else { |
8560 | 0 | if (mpd_qshiftr(&tsrc, src, -src->exp, status) == MPD_UINT_MAX) { |
8561 | 0 | goto malloc_error; |
8562 | 0 | } |
8563 | 0 | } |
8564 | | |
8565 | 0 | n = _baseconv_to_u16(rdata, rlen, rbase, tsrc.data, tsrc.len); |
8566 | 0 | if (n == SIZE_MAX) { |
8567 | 0 | goto malloc_error; |
8568 | 0 | } |
8569 | | |
8570 | | |
8571 | 0 | out: |
8572 | 0 | mpd_del(&tsrc); |
8573 | 0 | return n; |
8574 | | |
8575 | 0 | malloc_error: |
8576 | 0 | if (alloc) { |
8577 | 0 | mpd_free(*rdata); |
8578 | 0 | *rdata = NULL; |
8579 | 0 | } |
8580 | 0 | n = SIZE_MAX; |
8581 | 0 | *status |= MPD_Malloc_error; |
8582 | 0 | goto out; |
8583 | 0 | } |
8584 | | |
8585 | | /* |
8586 | | * Convert an integer mpd_t to a multiprecision integer with base<=UINT32_MAX. |
8587 | | * The least significant word of the result is (*rdata)[0]. |
8588 | | * |
8589 | | * If rdata is NULL, space is allocated by the function and rlen is irrelevant. |
8590 | | * In case of an error any allocated storage is freed and rdata is set back to |
8591 | | * NULL. |
8592 | | * |
8593 | | * If rdata is non-NULL, it MUST be allocated by one of libmpdec's allocation |
8594 | | * functions and rlen MUST be correct. If necessary, the function will resize |
8595 | | * rdata. In case of an error the caller must free rdata. |
8596 | | * |
8597 | | * Return value: In case of success, the exact length of rdata, SIZE_MAX |
8598 | | * otherwise. |
8599 | | */ |
8600 | | size_t |
8601 | | mpd_qexport_u32(uint32_t **rdata, size_t rlen, uint32_t rbase, |
8602 | | const mpd_t *src, uint32_t *status) |
8603 | 0 | { |
8604 | 0 | MPD_NEW_STATIC(tsrc,0,0,0,0); |
8605 | 0 | int alloc = 0; /* rdata == NULL */ |
8606 | 0 | size_t n; |
8607 | |
|
8608 | 0 | if (mpd_isspecial(src) || !_mpd_isint(src)) { |
8609 | 0 | *status |= MPD_Invalid_operation; |
8610 | 0 | return SIZE_MAX; |
8611 | 0 | } |
8612 | | |
8613 | 0 | if (*rdata == NULL) { |
8614 | 0 | rlen = mpd_sizeinbase(src, rbase); |
8615 | 0 | if (rlen == SIZE_MAX) { |
8616 | 0 | *status |= MPD_Invalid_operation; |
8617 | 0 | return SIZE_MAX; |
8618 | 0 | } |
8619 | 0 | *rdata = mpd_alloc(rlen, sizeof **rdata); |
8620 | 0 | if (*rdata == NULL) { |
8621 | 0 | goto malloc_error; |
8622 | 0 | } |
8623 | 0 | alloc = 1; |
8624 | 0 | } |
8625 | | |
8626 | 0 | if (mpd_iszero(src)) { |
8627 | 0 | **rdata = 0; |
8628 | 0 | return 1; |
8629 | 0 | } |
8630 | | |
8631 | 0 | if (src->exp >= 0) { |
8632 | 0 | if (!mpd_qshiftl(&tsrc, src, src->exp, status)) { |
8633 | 0 | goto malloc_error; |
8634 | 0 | } |
8635 | 0 | } |
8636 | 0 | else { |
8637 | 0 | if (mpd_qshiftr(&tsrc, src, -src->exp, status) == MPD_UINT_MAX) { |
8638 | 0 | goto malloc_error; |
8639 | 0 | } |
8640 | 0 | } |
8641 | | |
8642 | 0 | #ifdef CONFIG_64 |
8643 | 0 | n = _baseconv_to_smaller(rdata, rlen, rbase, |
8644 | 0 | tsrc.data, tsrc.len, MPD_RADIX); |
8645 | | #else |
8646 | | if (rbase == MPD_RADIX) { |
8647 | | n = _copy_equal_base(rdata, rlen, tsrc.data, tsrc.len); |
8648 | | } |
8649 | | else if (rbase < MPD_RADIX) { |
8650 | | n = _baseconv_to_smaller(rdata, rlen, rbase, |
8651 | | tsrc.data, tsrc.len, MPD_RADIX); |
8652 | | } |
8653 | | else { |
8654 | | n = _baseconv_to_larger(rdata, rlen, rbase, |
8655 | | tsrc.data, tsrc.len, MPD_RADIX); |
8656 | | } |
8657 | | #endif |
8658 | |
|
8659 | 0 | if (n == SIZE_MAX) { |
8660 | 0 | goto malloc_error; |
8661 | 0 | } |
8662 | | |
8663 | | |
8664 | 0 | out: |
8665 | 0 | mpd_del(&tsrc); |
8666 | 0 | return n; |
8667 | | |
8668 | 0 | malloc_error: |
8669 | 0 | if (alloc) { |
8670 | 0 | mpd_free(*rdata); |
8671 | 0 | *rdata = NULL; |
8672 | 0 | } |
8673 | 0 | n = SIZE_MAX; |
8674 | 0 | *status |= MPD_Malloc_error; |
8675 | 0 | goto out; |
8676 | 0 | } |
8677 | | |
8678 | | |
8679 | | /* |
8680 | | * Converts a multiprecision integer with base <= UINT16_MAX+1 to an mpd_t. |
8681 | | * The least significant word of the source is srcdata[0]. |
8682 | | */ |
8683 | | void |
8684 | | mpd_qimport_u16(mpd_t *result, |
8685 | | const uint16_t *srcdata, size_t srclen, |
8686 | | uint8_t srcsign, uint32_t srcbase, |
8687 | | const mpd_context_t *ctx, uint32_t *status) |
8688 | 0 | { |
8689 | 0 | mpd_uint_t *usrc; /* uint16_t src copied to an mpd_uint_t array */ |
8690 | 0 | mpd_ssize_t rlen; /* length of the result */ |
8691 | 0 | size_t n; |
8692 | |
|
8693 | 0 | assert(srclen > 0); |
8694 | 0 | assert(srcbase <= (1U<<16)); |
8695 | |
|
8696 | 0 | rlen = _mpd_importsize(srclen, srcbase); |
8697 | 0 | if (rlen == MPD_SSIZE_MAX) { |
8698 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
8699 | 0 | return; |
8700 | 0 | } |
8701 | | |
8702 | 0 | usrc = mpd_alloc((mpd_size_t)srclen, sizeof *usrc); |
8703 | 0 | if (usrc == NULL) { |
8704 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
8705 | 0 | return; |
8706 | 0 | } |
8707 | 0 | for (n = 0; n < srclen; n++) { |
8708 | 0 | usrc[n] = srcdata[n]; |
8709 | 0 | } |
8710 | |
|
8711 | 0 | if (!mpd_qresize(result, rlen, status)) { |
8712 | 0 | goto finish; |
8713 | 0 | } |
8714 | | |
8715 | 0 | n = _coeff_from_u16(result, rlen, usrc, srclen, srcbase, status); |
8716 | 0 | if (n == SIZE_MAX) { |
8717 | 0 | goto finish; |
8718 | 0 | } |
8719 | | |
8720 | 0 | mpd_set_flags(result, srcsign); |
8721 | 0 | result->exp = 0; |
8722 | 0 | result->len = n; |
8723 | 0 | mpd_setdigits(result); |
8724 | |
|
8725 | 0 | mpd_qresize(result, result->len, status); |
8726 | 0 | mpd_qfinalize(result, ctx, status); |
8727 | | |
8728 | |
|
8729 | 0 | finish: |
8730 | 0 | mpd_free(usrc); |
8731 | 0 | } |
8732 | | |
8733 | | /* |
8734 | | * Converts a multiprecision integer with base <= UINT32_MAX to an mpd_t. |
8735 | | * The least significant word of the source is srcdata[0]. |
8736 | | */ |
8737 | | void |
8738 | | mpd_qimport_u32(mpd_t *result, |
8739 | | const uint32_t *srcdata, size_t srclen, |
8740 | | uint8_t srcsign, uint32_t srcbase, |
8741 | | const mpd_context_t *ctx, uint32_t *status) |
8742 | 0 | { |
8743 | 0 | mpd_ssize_t rlen; /* length of the result */ |
8744 | 0 | size_t n; |
8745 | |
|
8746 | 0 | assert(srclen > 0); |
8747 | |
|
8748 | 0 | rlen = _mpd_importsize(srclen, srcbase); |
8749 | 0 | if (rlen == MPD_SSIZE_MAX) { |
8750 | 0 | mpd_seterror(result, MPD_Invalid_operation, status); |
8751 | 0 | return; |
8752 | 0 | } |
8753 | | |
8754 | 0 | if (!mpd_qresize(result, rlen, status)) { |
8755 | 0 | return; |
8756 | 0 | } |
8757 | | |
8758 | 0 | #ifdef CONFIG_64 |
8759 | 0 | n = _coeff_from_smaller_base(result, rlen, MPD_RADIX, |
8760 | 0 | srcdata, srclen, srcbase, |
8761 | 0 | status); |
8762 | | #else |
8763 | | if (srcbase == MPD_RADIX) { |
8764 | | if (!mpd_qresize(result, srclen, status)) { |
8765 | | return; |
8766 | | } |
8767 | | memcpy(result->data, srcdata, srclen * (sizeof *srcdata)); |
8768 | | n = srclen; |
8769 | | } |
8770 | | else if (srcbase < MPD_RADIX) { |
8771 | | n = _coeff_from_smaller_base(result, rlen, MPD_RADIX, |
8772 | | srcdata, srclen, srcbase, |
8773 | | status); |
8774 | | } |
8775 | | else { |
8776 | | mpd_uint_t *usrc = mpd_alloc((mpd_size_t)srclen, sizeof *usrc); |
8777 | | if (usrc == NULL) { |
8778 | | mpd_seterror(result, MPD_Malloc_error, status); |
8779 | | return; |
8780 | | } |
8781 | | for (n = 0; n < srclen; n++) { |
8782 | | usrc[n] = srcdata[n]; |
8783 | | } |
8784 | | |
8785 | | n = _coeff_from_larger_base(result, rlen, MPD_RADIX, |
8786 | | usrc, (mpd_ssize_t)srclen, srcbase, |
8787 | | status); |
8788 | | mpd_free(usrc); |
8789 | | } |
8790 | | #endif |
8791 | |
|
8792 | 0 | if (n == SIZE_MAX) { |
8793 | 0 | return; |
8794 | 0 | } |
8795 | | |
8796 | 0 | mpd_set_flags(result, srcsign); |
8797 | 0 | result->exp = 0; |
8798 | 0 | result->len = n; |
8799 | 0 | mpd_setdigits(result); |
8800 | |
|
8801 | 0 | mpd_qresize(result, result->len, status); |
8802 | 0 | mpd_qfinalize(result, ctx, status); |
8803 | 0 | } |
8804 | | |
8805 | | |
8806 | | /******************************************************************************/ |
8807 | | /* From triple */ |
8808 | | /******************************************************************************/ |
8809 | | |
8810 | | #if defined(CONFIG_64) && defined(__SIZEOF_INT128__) && !defined(_MSC_VER) |
8811 | | static mpd_ssize_t |
8812 | | _set_coeff(uint64_t data[3], uint64_t hi, uint64_t lo) |
8813 | 0 | { |
8814 | 0 | __uint128_t d = ((__uint128_t)hi << 64) + lo; |
8815 | 0 | __uint128_t q, r; |
8816 | |
|
8817 | 0 | q = d / MPD_RADIX; |
8818 | 0 | r = d % MPD_RADIX; |
8819 | 0 | data[0] = (uint64_t)r; |
8820 | 0 | d = q; |
8821 | |
|
8822 | 0 | q = d / MPD_RADIX; |
8823 | 0 | r = d % MPD_RADIX; |
8824 | 0 | data[1] = (uint64_t)r; |
8825 | 0 | d = q; |
8826 | |
|
8827 | 0 | q = d / MPD_RADIX; |
8828 | 0 | r = d % MPD_RADIX; |
8829 | 0 | data[2] = (uint64_t)r; |
8830 | |
|
8831 | 0 | if (q != 0) { |
8832 | 0 | abort(); /* GCOV_NOT_REACHED */ |
8833 | 0 | } |
8834 | | |
8835 | 0 | return data[2] != 0 ? 3 : (data[1] != 0 ? 2 : 1); |
8836 | 0 | } |
8837 | | #else |
8838 | | static size_t |
8839 | | _uint_from_u16(mpd_uint_t *w, mpd_ssize_t wlen, const uint16_t *u, size_t ulen) |
8840 | | { |
8841 | | const mpd_uint_t ubase = 1U<<16; |
8842 | | mpd_ssize_t n = 0; |
8843 | | mpd_uint_t carry; |
8844 | | |
8845 | | assert(wlen > 0 && ulen > 0); |
8846 | | |
8847 | | w[n++] = u[--ulen]; |
8848 | | while (--ulen != SIZE_MAX) { |
8849 | | carry = _mpd_shortmul_c(w, w, n, ubase); |
8850 | | if (carry) { |
8851 | | if (n >= wlen) { |
8852 | | abort(); /* GCOV_NOT_REACHED */ |
8853 | | } |
8854 | | w[n++] = carry; |
8855 | | } |
8856 | | carry = _mpd_shortadd(w, n, u[ulen]); |
8857 | | if (carry) { |
8858 | | if (n >= wlen) { |
8859 | | abort(); /* GCOV_NOT_REACHED */ |
8860 | | } |
8861 | | w[n++] = carry; |
8862 | | } |
8863 | | } |
8864 | | |
8865 | | return n; |
8866 | | } |
8867 | | |
8868 | | static mpd_ssize_t |
8869 | | _set_coeff(mpd_uint_t *data, mpd_ssize_t len, uint64_t hi, uint64_t lo) |
8870 | | { |
8871 | | uint16_t u16[8] = {0}; |
8872 | | |
8873 | | u16[7] = (uint16_t)((hi & 0xFFFF000000000000ULL) >> 48); |
8874 | | u16[6] = (uint16_t)((hi & 0x0000FFFF00000000ULL) >> 32); |
8875 | | u16[5] = (uint16_t)((hi & 0x00000000FFFF0000ULL) >> 16); |
8876 | | u16[4] = (uint16_t) (hi & 0x000000000000FFFFULL); |
8877 | | |
8878 | | u16[3] = (uint16_t)((lo & 0xFFFF000000000000ULL) >> 48); |
8879 | | u16[2] = (uint16_t)((lo & 0x0000FFFF00000000ULL) >> 32); |
8880 | | u16[1] = (uint16_t)((lo & 0x00000000FFFF0000ULL) >> 16); |
8881 | | u16[0] = (uint16_t) (lo & 0x000000000000FFFFULL); |
8882 | | |
8883 | | return (mpd_ssize_t)_uint_from_u16(data, len, u16, 8); |
8884 | | } |
8885 | | #endif |
8886 | | |
8887 | | static int |
8888 | | _set_uint128_coeff_exp(mpd_t *result, uint64_t hi, uint64_t lo, mpd_ssize_t exp) |
8889 | 0 | { |
8890 | 0 | mpd_uint_t data[5] = {0}; |
8891 | 0 | uint32_t status = 0; |
8892 | 0 | mpd_ssize_t len; |
8893 | |
|
8894 | 0 | #if defined(CONFIG_64) && defined(__SIZEOF_INT128__) && !defined(_MSC_VER) |
8895 | 0 | len = _set_coeff(data, hi, lo); |
8896 | | #else |
8897 | | len = _set_coeff(data, 5, hi, lo); |
8898 | | #endif |
8899 | |
|
8900 | 0 | if (!mpd_qresize(result, len, &status)) { |
8901 | 0 | return -1; |
8902 | 0 | } |
8903 | | |
8904 | 0 | for (mpd_ssize_t i = 0; i < len; i++) { |
8905 | 0 | result->data[i] = data[i]; |
8906 | 0 | } |
8907 | |
|
8908 | 0 | result->exp = exp; |
8909 | 0 | result->len = len; |
8910 | 0 | mpd_setdigits(result); |
8911 | |
|
8912 | 0 | return 0; |
8913 | 0 | } |
8914 | | |
8915 | | int |
8916 | | mpd_from_uint128_triple(mpd_t *result, const mpd_uint128_triple_t *triple, uint32_t *status) |
8917 | 0 | { |
8918 | 0 | static const mpd_context_t maxcontext = { |
8919 | 0 | .prec=MPD_MAX_PREC, |
8920 | 0 | .emax=MPD_MAX_EMAX, |
8921 | 0 | .emin=MPD_MIN_EMIN, |
8922 | 0 | .round=MPD_ROUND_HALF_EVEN, |
8923 | 0 | .traps=MPD_Traps, |
8924 | 0 | .status=0, |
8925 | 0 | .newtrap=0, |
8926 | 0 | .clamp=0, |
8927 | 0 | .allcr=1, |
8928 | 0 | }; |
8929 | 0 | const enum mpd_triple_class tag = triple->tag; |
8930 | 0 | const uint8_t sign = triple->sign; |
8931 | 0 | const uint64_t hi = triple->hi; |
8932 | 0 | const uint64_t lo = triple->lo; |
8933 | 0 | mpd_ssize_t exp; |
8934 | |
|
8935 | | #ifdef CONFIG_32 |
8936 | | if (triple->exp < MPD_SSIZE_MIN || triple->exp > MPD_SSIZE_MAX) { |
8937 | | goto conversion_error; |
8938 | | } |
8939 | | #endif |
8940 | 0 | exp = (mpd_ssize_t)triple->exp; |
8941 | |
|
8942 | 0 | switch (tag) { |
8943 | 0 | case MPD_TRIPLE_QNAN: case MPD_TRIPLE_SNAN: { |
8944 | 0 | if (sign > 1 || exp != 0) { |
8945 | 0 | goto conversion_error; |
8946 | 0 | } |
8947 | | |
8948 | 0 | const uint8_t flags = tag == MPD_TRIPLE_QNAN ? MPD_NAN : MPD_SNAN; |
8949 | 0 | mpd_setspecial(result, sign, flags); |
8950 | |
|
8951 | 0 | if (hi == 0 && lo == 0) { /* no payload */ |
8952 | 0 | return 0; |
8953 | 0 | } |
8954 | | |
8955 | 0 | if (_set_uint128_coeff_exp(result, hi, lo, exp) < 0) { |
8956 | 0 | goto malloc_error; |
8957 | 0 | } |
8958 | | |
8959 | 0 | return 0; |
8960 | 0 | } |
8961 | | |
8962 | 0 | case MPD_TRIPLE_INF: { |
8963 | 0 | if (sign > 1 || hi != 0 || lo != 0 || exp != 0) { |
8964 | 0 | goto conversion_error; |
8965 | 0 | } |
8966 | | |
8967 | 0 | mpd_setspecial(result, sign, MPD_INF); |
8968 | |
|
8969 | 0 | return 0; |
8970 | 0 | } |
8971 | | |
8972 | 0 | case MPD_TRIPLE_NORMAL: { |
8973 | 0 | if (sign > 1) { |
8974 | 0 | goto conversion_error; |
8975 | 0 | } |
8976 | | |
8977 | 0 | const uint8_t flags = sign ? MPD_NEG : MPD_POS; |
8978 | 0 | mpd_set_flags(result, flags); |
8979 | |
|
8980 | 0 | if (exp > MPD_EXP_INF) { |
8981 | 0 | exp = MPD_EXP_INF; |
8982 | 0 | } |
8983 | 0 | if (exp == MPD_SSIZE_MIN) { |
8984 | 0 | exp = MPD_SSIZE_MIN+1; |
8985 | 0 | } |
8986 | |
|
8987 | 0 | if (_set_uint128_coeff_exp(result, hi, lo, exp) < 0) { |
8988 | 0 | goto malloc_error; |
8989 | 0 | } |
8990 | | |
8991 | 0 | uint32_t workstatus = 0; |
8992 | 0 | mpd_qfinalize(result, &maxcontext, &workstatus); |
8993 | 0 | if (workstatus & (MPD_Inexact|MPD_Rounded|MPD_Clamped)) { |
8994 | 0 | goto conversion_error; |
8995 | 0 | } |
8996 | | |
8997 | 0 | return 0; |
8998 | 0 | } |
8999 | | |
9000 | 0 | default: |
9001 | 0 | goto conversion_error; |
9002 | 0 | } |
9003 | | |
9004 | 0 | conversion_error: |
9005 | 0 | mpd_seterror(result, MPD_Conversion_syntax, status); |
9006 | 0 | return -1; |
9007 | | |
9008 | 0 | malloc_error: |
9009 | 0 | mpd_seterror(result, MPD_Malloc_error, status); |
9010 | 0 | return -1; |
9011 | 0 | } |
9012 | | |
9013 | | |
9014 | | /******************************************************************************/ |
9015 | | /* As triple */ |
9016 | | /******************************************************************************/ |
9017 | | |
9018 | | #if defined(CONFIG_64) && defined(__SIZEOF_INT128__) && !defined(_MSC_VER) |
9019 | | static void |
9020 | | _get_coeff(uint64_t *hi, uint64_t *lo, const mpd_t *a) |
9021 | 0 | { |
9022 | 0 | __uint128_t u128 = 0; |
9023 | |
|
9024 | 0 | switch (a->len) { |
9025 | 0 | case 3: |
9026 | 0 | u128 = a->data[2]; /* fall through */ |
9027 | 0 | case 2: |
9028 | 0 | u128 = u128 * MPD_RADIX + a->data[1]; /* fall through */ |
9029 | 0 | case 1: |
9030 | 0 | u128 = u128 * MPD_RADIX + a->data[0]; |
9031 | 0 | break; |
9032 | 0 | default: |
9033 | 0 | abort(); /* GCOV_NOT_REACHED */ |
9034 | 0 | } |
9035 | | |
9036 | 0 | *hi = u128 >> 64; |
9037 | 0 | *lo = (uint64_t)u128; |
9038 | 0 | } |
9039 | | #else |
9040 | | static size_t |
9041 | | _uint_to_u16(uint16_t w[8], mpd_uint_t *u, mpd_ssize_t ulen) |
9042 | | { |
9043 | | const mpd_uint_t wbase = 1U<<16; |
9044 | | size_t n = 0; |
9045 | | |
9046 | | assert(ulen > 0); |
9047 | | |
9048 | | do { |
9049 | | if (n >= 8) { |
9050 | | abort(); /* GCOV_NOT_REACHED */ |
9051 | | } |
9052 | | w[n++] = (uint16_t)_mpd_shortdiv(u, u, ulen, wbase); |
9053 | | /* ulen is at least 1. u[ulen-1] can only be zero if ulen == 1. */ |
9054 | | ulen = _mpd_real_size(u, ulen); |
9055 | | |
9056 | | } while (u[ulen-1] != 0); |
9057 | | |
9058 | | return n; |
9059 | | } |
9060 | | |
9061 | | static void |
9062 | | _get_coeff(uint64_t *hi, uint64_t *lo, const mpd_t *a) |
9063 | | { |
9064 | | uint16_t u16[8] = {0}; |
9065 | | mpd_uint_t data[5] = {0}; |
9066 | | |
9067 | | switch (a->len) { |
9068 | | case 5: |
9069 | | data[4] = a->data[4]; /* fall through */ |
9070 | | case 4: |
9071 | | data[3] = a->data[3]; /* fall through */ |
9072 | | case 3: |
9073 | | data[2] = a->data[2]; /* fall through */ |
9074 | | case 2: |
9075 | | data[1] = a->data[1]; /* fall through */ |
9076 | | case 1: |
9077 | | data[0] = a->data[0]; |
9078 | | break; |
9079 | | default: |
9080 | | abort(); /* GCOV_NOT_REACHED */ |
9081 | | } |
9082 | | |
9083 | | _uint_to_u16(u16, data, a->len); |
9084 | | |
9085 | | *hi = (uint64_t)u16[7] << 48; |
9086 | | *hi |= (uint64_t)u16[6] << 32; |
9087 | | *hi |= (uint64_t)u16[5] << 16; |
9088 | | *hi |= (uint64_t)u16[4]; |
9089 | | |
9090 | | *lo = (uint64_t)u16[3] << 48; |
9091 | | *lo |= (uint64_t)u16[2] << 32; |
9092 | | *lo |= (uint64_t)u16[1] << 16; |
9093 | | *lo |= (uint64_t)u16[0]; |
9094 | | } |
9095 | | #endif |
9096 | | |
9097 | | static enum mpd_triple_class |
9098 | | _coeff_as_uint128(uint64_t *hi, uint64_t *lo, const mpd_t *a) |
9099 | 0 | { |
9100 | 0 | #ifdef CONFIG_64 |
9101 | 0 | static mpd_uint_t uint128_max_data[3] = { 3374607431768211455ULL, 4028236692093846346ULL, 3ULL }; |
9102 | 0 | static const mpd_t uint128_max = { MPD_STATIC|MPD_CONST_DATA, 0, 39, 3, 3, uint128_max_data }; |
9103 | | #else |
9104 | | static mpd_uint_t uint128_max_data[5] = { 768211455U, 374607431U, 938463463U, 282366920U, 340U }; |
9105 | | static const mpd_t uint128_max = { MPD_STATIC|MPD_CONST_DATA, 0, 39, 5, 5, uint128_max_data }; |
9106 | | #endif |
9107 | 0 | enum mpd_triple_class ret = MPD_TRIPLE_NORMAL; |
9108 | 0 | uint32_t status = 0; |
9109 | 0 | mpd_t coeff; |
9110 | |
|
9111 | 0 | *hi = *lo = 0ULL; |
9112 | |
|
9113 | 0 | if (mpd_isspecial(a)) { |
9114 | 0 | if (mpd_isinfinite(a)) { |
9115 | 0 | return MPD_TRIPLE_INF; |
9116 | 0 | } |
9117 | | |
9118 | 0 | ret = mpd_isqnan(a) ? MPD_TRIPLE_QNAN : MPD_TRIPLE_SNAN; |
9119 | 0 | if (a->len == 0) { /* no payload */ |
9120 | 0 | return ret; |
9121 | 0 | } |
9122 | 0 | } |
9123 | 0 | else if (mpd_iszero(a)) { |
9124 | 0 | return ret; |
9125 | 0 | } |
9126 | | |
9127 | 0 | _mpd_copy_shared(&coeff, a); |
9128 | 0 | mpd_set_flags(&coeff, 0); |
9129 | 0 | coeff.exp = 0; |
9130 | |
|
9131 | 0 | if (mpd_qcmp(&coeff, &uint128_max, &status) > 0) { |
9132 | 0 | return MPD_TRIPLE_ERROR; |
9133 | 0 | } |
9134 | | |
9135 | 0 | _get_coeff(hi, lo, &coeff); |
9136 | 0 | return ret; |
9137 | 0 | } |
9138 | | |
9139 | | mpd_uint128_triple_t |
9140 | | mpd_as_uint128_triple(const mpd_t *a) |
9141 | 0 | { |
9142 | 0 | mpd_uint128_triple_t triple = { MPD_TRIPLE_ERROR, 0, 0, 0, 0 }; |
9143 | |
|
9144 | 0 | triple.tag = _coeff_as_uint128(&triple.hi, &triple.lo, a); |
9145 | 0 | if (triple.tag == MPD_TRIPLE_ERROR) { |
9146 | 0 | return triple; |
9147 | 0 | } |
9148 | | |
9149 | 0 | triple.sign = !!mpd_isnegative(a); |
9150 | 0 | if (triple.tag == MPD_TRIPLE_NORMAL) { |
9151 | 0 | triple.exp = a->exp; |
9152 | 0 | } |
9153 | |
|
9154 | 0 | return triple; |
9155 | 0 | } |