* jvspec.c (jvgenmain_spec): Don't handle -fnew-verifier.
[gcc.git] / gcc / double-int.c
1 /* Operations with long integers.
2 Copyright (C) 2006, 2007, 2009, 2010 Free Software Foundation, Inc.
3
4 This file is part of GCC.
5
6 GCC is free software; you can redistribute it and/or modify it
7 under the terms of the GNU General Public License as published by the
8 Free Software Foundation; either version 3, or (at your option) any
9 later version.
10
11 GCC is distributed in the hope that it will be useful, but WITHOUT
12 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 for more details.
15
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
19
20 #include "config.h"
21 #include "system.h"
22 #include "coretypes.h"
23 #include "tm.h"
24 #include "tree.h"
25 #include "toplev.h"
26
27 /* We know that A1 + B1 = SUM1, using 2's complement arithmetic and ignoring
28 overflow. Suppose A, B and SUM have the same respective signs as A1, B1,
29 and SUM1. Then this yields nonzero if overflow occurred during the
30 addition.
31
32 Overflow occurs if A and B have the same sign, but A and SUM differ in
33 sign. Use `^' to test whether signs differ, and `< 0' to isolate the
34 sign. */
35 #define OVERFLOW_SUM_SIGN(a, b, sum) ((~((a) ^ (b)) & ((a) ^ (sum))) < 0)
36
37 /* To do constant folding on INTEGER_CST nodes requires two-word arithmetic.
38 We do that by representing the two-word integer in 4 words, with only
39 HOST_BITS_PER_WIDE_INT / 2 bits stored in each word, as a positive
40 number. The value of the word is LOWPART + HIGHPART * BASE. */
41
42 #define LOWPART(x) \
43 ((x) & (((unsigned HOST_WIDE_INT) 1 << (HOST_BITS_PER_WIDE_INT / 2)) - 1))
44 #define HIGHPART(x) \
45 ((unsigned HOST_WIDE_INT) (x) >> HOST_BITS_PER_WIDE_INT / 2)
46 #define BASE ((unsigned HOST_WIDE_INT) 1 << HOST_BITS_PER_WIDE_INT / 2)
47
48 /* Unpack a two-word integer into 4 words.
49 LOW and HI are the integer, as two `HOST_WIDE_INT' pieces.
50 WORDS points to the array of HOST_WIDE_INTs. */
51
52 static void
53 encode (HOST_WIDE_INT *words, unsigned HOST_WIDE_INT low, HOST_WIDE_INT hi)
54 {
55 words[0] = LOWPART (low);
56 words[1] = HIGHPART (low);
57 words[2] = LOWPART (hi);
58 words[3] = HIGHPART (hi);
59 }
60
61 /* Pack an array of 4 words into a two-word integer.
62 WORDS points to the array of words.
63 The integer is stored into *LOW and *HI as two `HOST_WIDE_INT' pieces. */
64
65 static void
66 decode (HOST_WIDE_INT *words, unsigned HOST_WIDE_INT *low,
67 HOST_WIDE_INT *hi)
68 {
69 *low = words[0] + words[1] * BASE;
70 *hi = words[2] + words[3] * BASE;
71 }
72
73 /* Add two doubleword integers with doubleword result.
74 Return nonzero if the operation overflows according to UNSIGNED_P.
75 Each argument is given as two `HOST_WIDE_INT' pieces.
76 One argument is L1 and H1; the other, L2 and H2.
77 The value is stored as two `HOST_WIDE_INT' pieces in *LV and *HV. */
78
79 int
80 add_double_with_sign (unsigned HOST_WIDE_INT l1, HOST_WIDE_INT h1,
81 unsigned HOST_WIDE_INT l2, HOST_WIDE_INT h2,
82 unsigned HOST_WIDE_INT *lv, HOST_WIDE_INT *hv,
83 bool unsigned_p)
84 {
85 unsigned HOST_WIDE_INT l;
86 HOST_WIDE_INT h;
87
88 l = l1 + l2;
89 h = (HOST_WIDE_INT) ((unsigned HOST_WIDE_INT) h1
90 + (unsigned HOST_WIDE_INT) h2
91 + (l < l1));
92
93 *lv = l;
94 *hv = h;
95
96 if (unsigned_p)
97 return ((unsigned HOST_WIDE_INT) h < (unsigned HOST_WIDE_INT) h1
98 || (h == h1
99 && l < l1));
100 else
101 return OVERFLOW_SUM_SIGN (h1, h2, h);
102 }
103
104 /* Negate a doubleword integer with doubleword result.
105 Return nonzero if the operation overflows, assuming it's signed.
106 The argument is given as two `HOST_WIDE_INT' pieces in L1 and H1.
107 The value is stored as two `HOST_WIDE_INT' pieces in *LV and *HV. */
108
109 int
110 neg_double (unsigned HOST_WIDE_INT l1, HOST_WIDE_INT h1,
111 unsigned HOST_WIDE_INT *lv, HOST_WIDE_INT *hv)
112 {
113 if (l1 == 0)
114 {
115 *lv = 0;
116 *hv = - h1;
117 return (*hv & h1) < 0;
118 }
119 else
120 {
121 *lv = -l1;
122 *hv = ~h1;
123 return 0;
124 }
125 }
126
127 /* Multiply two doubleword integers with doubleword result.
128 Return nonzero if the operation overflows according to UNSIGNED_P.
129 Each argument is given as two `HOST_WIDE_INT' pieces.
130 One argument is L1 and H1; the other, L2 and H2.
131 The value is stored as two `HOST_WIDE_INT' pieces in *LV and *HV. */
132
133 int
134 mul_double_with_sign (unsigned HOST_WIDE_INT l1, HOST_WIDE_INT h1,
135 unsigned HOST_WIDE_INT l2, HOST_WIDE_INT h2,
136 unsigned HOST_WIDE_INT *lv, HOST_WIDE_INT *hv,
137 bool unsigned_p)
138 {
139 HOST_WIDE_INT arg1[4];
140 HOST_WIDE_INT arg2[4];
141 HOST_WIDE_INT prod[4 * 2];
142 unsigned HOST_WIDE_INT carry;
143 int i, j, k;
144 unsigned HOST_WIDE_INT toplow, neglow;
145 HOST_WIDE_INT tophigh, neghigh;
146
147 encode (arg1, l1, h1);
148 encode (arg2, l2, h2);
149
150 memset (prod, 0, sizeof prod);
151
152 for (i = 0; i < 4; i++)
153 {
154 carry = 0;
155 for (j = 0; j < 4; j++)
156 {
157 k = i + j;
158 /* This product is <= 0xFFFE0001, the sum <= 0xFFFF0000. */
159 carry += arg1[i] * arg2[j];
160 /* Since prod[p] < 0xFFFF, this sum <= 0xFFFFFFFF. */
161 carry += prod[k];
162 prod[k] = LOWPART (carry);
163 carry = HIGHPART (carry);
164 }
165 prod[i + 4] = carry;
166 }
167
168 decode (prod, lv, hv);
169 decode (prod + 4, &toplow, &tophigh);
170
171 /* Unsigned overflow is immediate. */
172 if (unsigned_p)
173 return (toplow | tophigh) != 0;
174
175 /* Check for signed overflow by calculating the signed representation of the
176 top half of the result; it should agree with the low half's sign bit. */
177 if (h1 < 0)
178 {
179 neg_double (l2, h2, &neglow, &neghigh);
180 add_double (neglow, neghigh, toplow, tophigh, &toplow, &tophigh);
181 }
182 if (h2 < 0)
183 {
184 neg_double (l1, h1, &neglow, &neghigh);
185 add_double (neglow, neghigh, toplow, tophigh, &toplow, &tophigh);
186 }
187 return (*hv < 0 ? ~(toplow & tophigh) : toplow | tophigh) != 0;
188 }
189
190 /* Shift the doubleword integer in L1, H1 left by COUNT places
191 keeping only PREC bits of result.
192 Shift right if COUNT is negative.
193 ARITH nonzero specifies arithmetic shifting; otherwise use logical shift.
194 Store the value as two `HOST_WIDE_INT' pieces in *LV and *HV. */
195
196 void
197 lshift_double (unsigned HOST_WIDE_INT l1, HOST_WIDE_INT h1,
198 HOST_WIDE_INT count, unsigned int prec,
199 unsigned HOST_WIDE_INT *lv, HOST_WIDE_INT *hv, bool arith)
200 {
201 unsigned HOST_WIDE_INT signmask;
202
203 if (count < 0)
204 {
205 rshift_double (l1, h1, -count, prec, lv, hv, arith);
206 return;
207 }
208
209 if (SHIFT_COUNT_TRUNCATED)
210 count %= prec;
211
212 if (count >= 2 * HOST_BITS_PER_WIDE_INT)
213 {
214 /* Shifting by the host word size is undefined according to the
215 ANSI standard, so we must handle this as a special case. */
216 *hv = 0;
217 *lv = 0;
218 }
219 else if (count >= HOST_BITS_PER_WIDE_INT)
220 {
221 *hv = l1 << (count - HOST_BITS_PER_WIDE_INT);
222 *lv = 0;
223 }
224 else
225 {
226 *hv = (((unsigned HOST_WIDE_INT) h1 << count)
227 | (l1 >> (HOST_BITS_PER_WIDE_INT - count - 1) >> 1));
228 *lv = l1 << count;
229 }
230
231 /* Sign extend all bits that are beyond the precision. */
232
233 signmask = -((prec > HOST_BITS_PER_WIDE_INT
234 ? ((unsigned HOST_WIDE_INT) *hv
235 >> (prec - HOST_BITS_PER_WIDE_INT - 1))
236 : (*lv >> (prec - 1))) & 1);
237
238 if (prec >= 2 * HOST_BITS_PER_WIDE_INT)
239 ;
240 else if (prec >= HOST_BITS_PER_WIDE_INT)
241 {
242 *hv &= ~((HOST_WIDE_INT) (-1) << (prec - HOST_BITS_PER_WIDE_INT));
243 *hv |= signmask << (prec - HOST_BITS_PER_WIDE_INT);
244 }
245 else
246 {
247 *hv = signmask;
248 *lv &= ~((unsigned HOST_WIDE_INT) (-1) << prec);
249 *lv |= signmask << prec;
250 }
251 }
252
253 /* Shift the doubleword integer in L1, H1 right by COUNT places
254 keeping only PREC bits of result. Shift left if COUNT is negative.
255 ARITH nonzero specifies arithmetic shifting; otherwise use logical shift.
256 Store the value as two `HOST_WIDE_INT' pieces in *LV and *HV. */
257
258 void
259 rshift_double (unsigned HOST_WIDE_INT l1, HOST_WIDE_INT h1,
260 HOST_WIDE_INT count, unsigned int prec,
261 unsigned HOST_WIDE_INT *lv, HOST_WIDE_INT *hv,
262 bool arith)
263 {
264 unsigned HOST_WIDE_INT signmask;
265
266 if (count < 0)
267 {
268 lshift_double (l1, h1, -count, prec, lv, hv, arith);
269 return;
270 }
271
272 signmask = (arith
273 ? -((unsigned HOST_WIDE_INT) h1 >> (HOST_BITS_PER_WIDE_INT - 1))
274 : 0);
275
276 if (SHIFT_COUNT_TRUNCATED)
277 count %= prec;
278
279 if (count >= 2 * HOST_BITS_PER_WIDE_INT)
280 {
281 /* Shifting by the host word size is undefined according to the
282 ANSI standard, so we must handle this as a special case. */
283 *hv = 0;
284 *lv = 0;
285 }
286 else if (count >= HOST_BITS_PER_WIDE_INT)
287 {
288 *hv = 0;
289 *lv = (unsigned HOST_WIDE_INT) h1 >> (count - HOST_BITS_PER_WIDE_INT);
290 }
291 else
292 {
293 *hv = (unsigned HOST_WIDE_INT) h1 >> count;
294 *lv = ((l1 >> count)
295 | ((unsigned HOST_WIDE_INT) h1
296 << (HOST_BITS_PER_WIDE_INT - count - 1) << 1));
297 }
298
299 /* Zero / sign extend all bits that are beyond the precision. */
300
301 if (count >= (HOST_WIDE_INT)prec)
302 {
303 *hv = signmask;
304 *lv = signmask;
305 }
306 else if ((prec - count) >= 2 * HOST_BITS_PER_WIDE_INT)
307 ;
308 else if ((prec - count) >= HOST_BITS_PER_WIDE_INT)
309 {
310 *hv &= ~((HOST_WIDE_INT) (-1) << (prec - count - HOST_BITS_PER_WIDE_INT));
311 *hv |= signmask << (prec - count - HOST_BITS_PER_WIDE_INT);
312 }
313 else
314 {
315 *hv = signmask;
316 *lv &= ~((unsigned HOST_WIDE_INT) (-1) << (prec - count));
317 *lv |= signmask << (prec - count);
318 }
319 }
320
321 /* Divide doubleword integer LNUM, HNUM by doubleword integer LDEN, HDEN
322 for a quotient (stored in *LQUO, *HQUO) and remainder (in *LREM, *HREM).
323 CODE is a tree code for a kind of division, one of
324 TRUNC_DIV_EXPR, FLOOR_DIV_EXPR, CEIL_DIV_EXPR, ROUND_DIV_EXPR
325 or EXACT_DIV_EXPR
326 It controls how the quotient is rounded to an integer.
327 Return nonzero if the operation overflows.
328 UNS nonzero says do unsigned division. */
329
330 int
331 div_and_round_double (unsigned code, int uns,
332 /* num == numerator == dividend */
333 unsigned HOST_WIDE_INT lnum_orig,
334 HOST_WIDE_INT hnum_orig,
335 /* den == denominator == divisor */
336 unsigned HOST_WIDE_INT lden_orig,
337 HOST_WIDE_INT hden_orig,
338 unsigned HOST_WIDE_INT *lquo,
339 HOST_WIDE_INT *hquo, unsigned HOST_WIDE_INT *lrem,
340 HOST_WIDE_INT *hrem)
341 {
342 int quo_neg = 0;
343 HOST_WIDE_INT num[4 + 1]; /* extra element for scaling. */
344 HOST_WIDE_INT den[4], quo[4];
345 int i, j;
346 unsigned HOST_WIDE_INT work;
347 unsigned HOST_WIDE_INT carry = 0;
348 unsigned HOST_WIDE_INT lnum = lnum_orig;
349 HOST_WIDE_INT hnum = hnum_orig;
350 unsigned HOST_WIDE_INT lden = lden_orig;
351 HOST_WIDE_INT hden = hden_orig;
352 int overflow = 0;
353
354 if (hden == 0 && lden == 0)
355 overflow = 1, lden = 1;
356
357 /* Calculate quotient sign and convert operands to unsigned. */
358 if (!uns)
359 {
360 if (hnum < 0)
361 {
362 quo_neg = ~ quo_neg;
363 /* (minimum integer) / (-1) is the only overflow case. */
364 if (neg_double (lnum, hnum, &lnum, &hnum)
365 && ((HOST_WIDE_INT) lden & hden) == -1)
366 overflow = 1;
367 }
368 if (hden < 0)
369 {
370 quo_neg = ~ quo_neg;
371 neg_double (lden, hden, &lden, &hden);
372 }
373 }
374
375 if (hnum == 0 && hden == 0)
376 { /* single precision */
377 *hquo = *hrem = 0;
378 /* This unsigned division rounds toward zero. */
379 *lquo = lnum / lden;
380 goto finish_up;
381 }
382
383 if (hnum == 0)
384 { /* trivial case: dividend < divisor */
385 /* hden != 0 already checked. */
386 *hquo = *lquo = 0;
387 *hrem = hnum;
388 *lrem = lnum;
389 goto finish_up;
390 }
391
392 memset (quo, 0, sizeof quo);
393
394 memset (num, 0, sizeof num); /* to zero 9th element */
395 memset (den, 0, sizeof den);
396
397 encode (num, lnum, hnum);
398 encode (den, lden, hden);
399
400 /* Special code for when the divisor < BASE. */
401 if (hden == 0 && lden < (unsigned HOST_WIDE_INT) BASE)
402 {
403 /* hnum != 0 already checked. */
404 for (i = 4 - 1; i >= 0; i--)
405 {
406 work = num[i] + carry * BASE;
407 quo[i] = work / lden;
408 carry = work % lden;
409 }
410 }
411 else
412 {
413 /* Full double precision division,
414 with thanks to Don Knuth's "Seminumerical Algorithms". */
415 int num_hi_sig, den_hi_sig;
416 unsigned HOST_WIDE_INT quo_est, scale;
417
418 /* Find the highest nonzero divisor digit. */
419 for (i = 4 - 1;; i--)
420 if (den[i] != 0)
421 {
422 den_hi_sig = i;
423 break;
424 }
425
426 /* Insure that the first digit of the divisor is at least BASE/2.
427 This is required by the quotient digit estimation algorithm. */
428
429 scale = BASE / (den[den_hi_sig] + 1);
430 if (scale > 1)
431 { /* scale divisor and dividend */
432 carry = 0;
433 for (i = 0; i <= 4 - 1; i++)
434 {
435 work = (num[i] * scale) + carry;
436 num[i] = LOWPART (work);
437 carry = HIGHPART (work);
438 }
439
440 num[4] = carry;
441 carry = 0;
442 for (i = 0; i <= 4 - 1; i++)
443 {
444 work = (den[i] * scale) + carry;
445 den[i] = LOWPART (work);
446 carry = HIGHPART (work);
447 if (den[i] != 0) den_hi_sig = i;
448 }
449 }
450
451 num_hi_sig = 4;
452
453 /* Main loop */
454 for (i = num_hi_sig - den_hi_sig - 1; i >= 0; i--)
455 {
456 /* Guess the next quotient digit, quo_est, by dividing the first
457 two remaining dividend digits by the high order quotient digit.
458 quo_est is never low and is at most 2 high. */
459 unsigned HOST_WIDE_INT tmp;
460
461 num_hi_sig = i + den_hi_sig + 1;
462 work = num[num_hi_sig] * BASE + num[num_hi_sig - 1];
463 if (num[num_hi_sig] != den[den_hi_sig])
464 quo_est = work / den[den_hi_sig];
465 else
466 quo_est = BASE - 1;
467
468 /* Refine quo_est so it's usually correct, and at most one high. */
469 tmp = work - quo_est * den[den_hi_sig];
470 if (tmp < BASE
471 && (den[den_hi_sig - 1] * quo_est
472 > (tmp * BASE + num[num_hi_sig - 2])))
473 quo_est--;
474
475 /* Try QUO_EST as the quotient digit, by multiplying the
476 divisor by QUO_EST and subtracting from the remaining dividend.
477 Keep in mind that QUO_EST is the I - 1st digit. */
478
479 carry = 0;
480 for (j = 0; j <= den_hi_sig; j++)
481 {
482 work = quo_est * den[j] + carry;
483 carry = HIGHPART (work);
484 work = num[i + j] - LOWPART (work);
485 num[i + j] = LOWPART (work);
486 carry += HIGHPART (work) != 0;
487 }
488
489 /* If quo_est was high by one, then num[i] went negative and
490 we need to correct things. */
491 if (num[num_hi_sig] < (HOST_WIDE_INT) carry)
492 {
493 quo_est--;
494 carry = 0; /* add divisor back in */
495 for (j = 0; j <= den_hi_sig; j++)
496 {
497 work = num[i + j] + den[j] + carry;
498 carry = HIGHPART (work);
499 num[i + j] = LOWPART (work);
500 }
501
502 num [num_hi_sig] += carry;
503 }
504
505 /* Store the quotient digit. */
506 quo[i] = quo_est;
507 }
508 }
509
510 decode (quo, lquo, hquo);
511
512 finish_up:
513 /* If result is negative, make it so. */
514 if (quo_neg)
515 neg_double (*lquo, *hquo, lquo, hquo);
516
517 /* Compute trial remainder: rem = num - (quo * den) */
518 mul_double (*lquo, *hquo, lden_orig, hden_orig, lrem, hrem);
519 neg_double (*lrem, *hrem, lrem, hrem);
520 add_double (lnum_orig, hnum_orig, *lrem, *hrem, lrem, hrem);
521
522 switch (code)
523 {
524 case TRUNC_DIV_EXPR:
525 case TRUNC_MOD_EXPR: /* round toward zero */
526 case EXACT_DIV_EXPR: /* for this one, it shouldn't matter */
527 return overflow;
528
529 case FLOOR_DIV_EXPR:
530 case FLOOR_MOD_EXPR: /* round toward negative infinity */
531 if (quo_neg && (*lrem != 0 || *hrem != 0)) /* ratio < 0 && rem != 0 */
532 {
533 /* quo = quo - 1; */
534 add_double (*lquo, *hquo, (HOST_WIDE_INT) -1, (HOST_WIDE_INT) -1,
535 lquo, hquo);
536 }
537 else
538 return overflow;
539 break;
540
541 case CEIL_DIV_EXPR:
542 case CEIL_MOD_EXPR: /* round toward positive infinity */
543 if (!quo_neg && (*lrem != 0 || *hrem != 0)) /* ratio > 0 && rem != 0 */
544 {
545 add_double (*lquo, *hquo, (HOST_WIDE_INT) 1, (HOST_WIDE_INT) 0,
546 lquo, hquo);
547 }
548 else
549 return overflow;
550 break;
551
552 case ROUND_DIV_EXPR:
553 case ROUND_MOD_EXPR: /* round to closest integer */
554 {
555 unsigned HOST_WIDE_INT labs_rem = *lrem;
556 HOST_WIDE_INT habs_rem = *hrem;
557 unsigned HOST_WIDE_INT labs_den = lden, ltwice;
558 HOST_WIDE_INT habs_den = hden, htwice;
559
560 /* Get absolute values. */
561 if (*hrem < 0)
562 neg_double (*lrem, *hrem, &labs_rem, &habs_rem);
563 if (hden < 0)
564 neg_double (lden, hden, &labs_den, &habs_den);
565
566 /* If (2 * abs (lrem) >= abs (lden)), adjust the quotient. */
567 mul_double ((HOST_WIDE_INT) 2, (HOST_WIDE_INT) 0,
568 labs_rem, habs_rem, &ltwice, &htwice);
569
570 if (((unsigned HOST_WIDE_INT) habs_den
571 < (unsigned HOST_WIDE_INT) htwice)
572 || (((unsigned HOST_WIDE_INT) habs_den
573 == (unsigned HOST_WIDE_INT) htwice)
574 && (labs_den <= ltwice)))
575 {
576 if (*hquo < 0)
577 /* quo = quo - 1; */
578 add_double (*lquo, *hquo,
579 (HOST_WIDE_INT) -1, (HOST_WIDE_INT) -1, lquo, hquo);
580 else
581 /* quo = quo + 1; */
582 add_double (*lquo, *hquo, (HOST_WIDE_INT) 1, (HOST_WIDE_INT) 0,
583 lquo, hquo);
584 }
585 else
586 return overflow;
587 }
588 break;
589
590 default:
591 gcc_unreachable ();
592 }
593
594 /* Compute true remainder: rem = num - (quo * den) */
595 mul_double (*lquo, *hquo, lden_orig, hden_orig, lrem, hrem);
596 neg_double (*lrem, *hrem, lrem, hrem);
597 add_double (lnum_orig, hnum_orig, *lrem, *hrem, lrem, hrem);
598 return overflow;
599 }
600
601
602 /* Returns mask for PREC bits. */
603
604 double_int
605 double_int_mask (unsigned prec)
606 {
607 unsigned HOST_WIDE_INT m;
608 double_int mask;
609
610 if (prec > HOST_BITS_PER_WIDE_INT)
611 {
612 prec -= HOST_BITS_PER_WIDE_INT;
613 m = ((unsigned HOST_WIDE_INT) 2 << (prec - 1)) - 1;
614 mask.high = (HOST_WIDE_INT) m;
615 mask.low = ALL_ONES;
616 }
617 else
618 {
619 mask.high = 0;
620 mask.low = ((unsigned HOST_WIDE_INT) 2 << (prec - 1)) - 1;
621 }
622
623 return mask;
624 }
625
626 /* Clears the bits of CST over the precision PREC. If UNS is false, the bits
627 outside of the precision are set to the sign bit (i.e., the PREC-th one),
628 otherwise they are set to zero.
629
630 This corresponds to returning the value represented by PREC lowermost bits
631 of CST, with the given signedness. */
632
633 double_int
634 double_int_ext (double_int cst, unsigned prec, bool uns)
635 {
636 if (uns)
637 return double_int_zext (cst, prec);
638 else
639 return double_int_sext (cst, prec);
640 }
641
642 /* The same as double_int_ext with UNS = true. */
643
644 double_int
645 double_int_zext (double_int cst, unsigned prec)
646 {
647 double_int mask = double_int_mask (prec);
648 double_int r;
649
650 r.low = cst.low & mask.low;
651 r.high = cst.high & mask.high;
652
653 return r;
654 }
655
656 /* The same as double_int_ext with UNS = false. */
657
658 double_int
659 double_int_sext (double_int cst, unsigned prec)
660 {
661 double_int mask = double_int_mask (prec);
662 double_int r;
663 unsigned HOST_WIDE_INT snum;
664
665 if (prec <= HOST_BITS_PER_WIDE_INT)
666 snum = cst.low;
667 else
668 {
669 prec -= HOST_BITS_PER_WIDE_INT;
670 snum = (unsigned HOST_WIDE_INT) cst.high;
671 }
672 if (((snum >> (prec - 1)) & 1) == 1)
673 {
674 r.low = cst.low | ~mask.low;
675 r.high = cst.high | ~mask.high;
676 }
677 else
678 {
679 r.low = cst.low & mask.low;
680 r.high = cst.high & mask.high;
681 }
682
683 return r;
684 }
685
686 /* Returns true if CST fits in signed HOST_WIDE_INT. */
687
688 bool
689 double_int_fits_in_shwi_p (double_int cst)
690 {
691 if (cst.high == 0)
692 return (HOST_WIDE_INT) cst.low >= 0;
693 else if (cst.high == -1)
694 return (HOST_WIDE_INT) cst.low < 0;
695 else
696 return false;
697 }
698
699 /* Returns true if CST fits in HOST_WIDE_INT if UNS is false, or in
700 unsigned HOST_WIDE_INT if UNS is true. */
701
702 bool
703 double_int_fits_in_hwi_p (double_int cst, bool uns)
704 {
705 if (uns)
706 return double_int_fits_in_uhwi_p (cst);
707 else
708 return double_int_fits_in_shwi_p (cst);
709 }
710
711 /* Returns A * B. */
712
713 double_int
714 double_int_mul (double_int a, double_int b)
715 {
716 double_int ret;
717 mul_double (a.low, a.high, b.low, b.high, &ret.low, &ret.high);
718 return ret;
719 }
720
721 /* Returns A + B. */
722
723 double_int
724 double_int_add (double_int a, double_int b)
725 {
726 double_int ret;
727 add_double (a.low, a.high, b.low, b.high, &ret.low, &ret.high);
728 return ret;
729 }
730
731 /* Returns A - B. */
732
733 double_int
734 double_int_sub (double_int a, double_int b)
735 {
736 double_int ret;
737 neg_double (b.low, b.high, &b.low, &b.high);
738 add_double (a.low, a.high, b.low, b.high, &ret.low, &ret.high);
739 return ret;
740 }
741
742 /* Returns -A. */
743
744 double_int
745 double_int_neg (double_int a)
746 {
747 double_int ret;
748 neg_double (a.low, a.high, &ret.low, &ret.high);
749 return ret;
750 }
751
752 /* Returns A / B (computed as unsigned depending on UNS, and rounded as
753 specified by CODE). CODE is enum tree_code in fact, but double_int.h
754 must be included before tree.h. The remainder after the division is
755 stored to MOD. */
756
757 double_int
758 double_int_divmod (double_int a, double_int b, bool uns, unsigned code,
759 double_int *mod)
760 {
761 double_int ret;
762
763 div_and_round_double (code, uns, a.low, a.high,
764 b.low, b.high, &ret.low, &ret.high,
765 &mod->low, &mod->high);
766 return ret;
767 }
768
769 /* The same as double_int_divmod with UNS = false. */
770
771 double_int
772 double_int_sdivmod (double_int a, double_int b, unsigned code, double_int *mod)
773 {
774 return double_int_divmod (a, b, false, code, mod);
775 }
776
777 /* The same as double_int_divmod with UNS = true. */
778
779 double_int
780 double_int_udivmod (double_int a, double_int b, unsigned code, double_int *mod)
781 {
782 return double_int_divmod (a, b, true, code, mod);
783 }
784
785 /* Returns A / B (computed as unsigned depending on UNS, and rounded as
786 specified by CODE). CODE is enum tree_code in fact, but double_int.h
787 must be included before tree.h. */
788
789 double_int
790 double_int_div (double_int a, double_int b, bool uns, unsigned code)
791 {
792 double_int mod;
793
794 return double_int_divmod (a, b, uns, code, &mod);
795 }
796
797 /* The same as double_int_div with UNS = false. */
798
799 double_int
800 double_int_sdiv (double_int a, double_int b, unsigned code)
801 {
802 return double_int_div (a, b, false, code);
803 }
804
805 /* The same as double_int_div with UNS = true. */
806
807 double_int
808 double_int_udiv (double_int a, double_int b, unsigned code)
809 {
810 return double_int_div (a, b, true, code);
811 }
812
813 /* Returns A % B (computed as unsigned depending on UNS, and rounded as
814 specified by CODE). CODE is enum tree_code in fact, but double_int.h
815 must be included before tree.h. */
816
817 double_int
818 double_int_mod (double_int a, double_int b, bool uns, unsigned code)
819 {
820 double_int mod;
821
822 double_int_divmod (a, b, uns, code, &mod);
823 return mod;
824 }
825
826 /* The same as double_int_mod with UNS = false. */
827
828 double_int
829 double_int_smod (double_int a, double_int b, unsigned code)
830 {
831 return double_int_mod (a, b, false, code);
832 }
833
834 /* The same as double_int_mod with UNS = true. */
835
836 double_int
837 double_int_umod (double_int a, double_int b, unsigned code)
838 {
839 return double_int_mod (a, b, true, code);
840 }
841
842 /* Set BITPOS bit in A. */
843 double_int
844 double_int_setbit (double_int a, unsigned bitpos)
845 {
846 if (bitpos < HOST_BITS_PER_WIDE_INT)
847 a.low |= (unsigned HOST_WIDE_INT) 1 << bitpos;
848 else
849 a.high |= (HOST_WIDE_INT) 1 << (bitpos - HOST_BITS_PER_WIDE_INT);
850
851 return a;
852 }
853
854 /* Count trailing zeros in A. */
855 int
856 double_int_ctz (double_int a)
857 {
858 unsigned HOST_WIDE_INT w = a.low ? a.low : (unsigned HOST_WIDE_INT) a.high;
859 unsigned bits = a.low ? 0 : HOST_BITS_PER_WIDE_INT;
860 if (!w)
861 return HOST_BITS_PER_DOUBLE_INT;
862 bits += ctz_hwi (w);
863 return bits;
864 }
865
866 /* Shift A left by COUNT places keeping only PREC bits of result. Shift
867 right if COUNT is negative. ARITH true specifies arithmetic shifting;
868 otherwise use logical shift. */
869
870 double_int
871 double_int_lshift (double_int a, HOST_WIDE_INT count, unsigned int prec, bool arith)
872 {
873 double_int ret;
874 lshift_double (a.low, a.high, count, prec, &ret.low, &ret.high, arith);
875 return ret;
876 }
877
878 /* Shift A rigth by COUNT places keeping only PREC bits of result. Shift
879 left if COUNT is negative. ARITH true specifies arithmetic shifting;
880 otherwise use logical shift. */
881
882 double_int
883 double_int_rshift (double_int a, HOST_WIDE_INT count, unsigned int prec, bool arith)
884 {
885 double_int ret;
886 rshift_double (a.low, a.high, count, prec, &ret.low, &ret.high, arith);
887 return ret;
888 }
889
890 /* Rotate A left by COUNT places keeping only PREC bits of result.
891 Rotate right if COUNT is negative. */
892
893 double_int
894 double_int_lrotate (double_int a, HOST_WIDE_INT count, unsigned int prec)
895 {
896 double_int t1, t2;
897
898 count %= prec;
899 if (count < 0)
900 count += prec;
901
902 t1 = double_int_lshift (a, count, prec, false);
903 t2 = double_int_rshift (a, prec - count, prec, false);
904
905 return double_int_ior (t1, t2);
906 }
907
908 /* Rotate A rigth by COUNT places keeping only PREC bits of result.
909 Rotate right if COUNT is negative. */
910
911 double_int
912 double_int_rrotate (double_int a, HOST_WIDE_INT count, unsigned int prec)
913 {
914 double_int t1, t2;
915
916 count %= prec;
917 if (count < 0)
918 count += prec;
919
920 t1 = double_int_rshift (a, count, prec, false);
921 t2 = double_int_lshift (a, prec - count, prec, false);
922
923 return double_int_ior (t1, t2);
924 }
925
926 /* Returns -1 if A < B, 0 if A == B and 1 if A > B. Signedness of the
927 comparison is given by UNS. */
928
929 int
930 double_int_cmp (double_int a, double_int b, bool uns)
931 {
932 if (uns)
933 return double_int_ucmp (a, b);
934 else
935 return double_int_scmp (a, b);
936 }
937
938 /* Compares two unsigned values A and B. Returns -1 if A < B, 0 if A == B,
939 and 1 if A > B. */
940
941 int
942 double_int_ucmp (double_int a, double_int b)
943 {
944 if ((unsigned HOST_WIDE_INT) a.high < (unsigned HOST_WIDE_INT) b.high)
945 return -1;
946 if ((unsigned HOST_WIDE_INT) a.high > (unsigned HOST_WIDE_INT) b.high)
947 return 1;
948 if (a.low < b.low)
949 return -1;
950 if (a.low > b.low)
951 return 1;
952
953 return 0;
954 }
955
956 /* Compares two signed values A and B. Returns -1 if A < B, 0 if A == B,
957 and 1 if A > B. */
958
959 int
960 double_int_scmp (double_int a, double_int b)
961 {
962 if (a.high < b.high)
963 return -1;
964 if (a.high > b.high)
965 return 1;
966 if (a.low < b.low)
967 return -1;
968 if (a.low > b.low)
969 return 1;
970
971 return 0;
972 }
973
974 /* Compares two values A and B. Returns max value. Signedness of the
975 comparison is given by UNS. */
976
977 double_int
978 double_int_max (double_int a, double_int b, bool uns)
979 {
980 return (double_int_cmp (a, b, uns) == 1) ? a : b;
981 }
982
983 /* Compares two signed values A and B. Returns max value. */
984
985 double_int double_int_smax (double_int a, double_int b)
986 {
987 return (double_int_scmp (a, b) == 1) ? a : b;
988 }
989
990 /* Compares two unsigned values A and B. Returns max value. */
991
992 double_int double_int_umax (double_int a, double_int b)
993 {
994 return (double_int_ucmp (a, b) == 1) ? a : b;
995 }
996
997 /* Compares two values A and B. Returns mix value. Signedness of the
998 comparison is given by UNS. */
999
1000 double_int double_int_min (double_int a, double_int b, bool uns)
1001 {
1002 return (double_int_cmp (a, b, uns) == -1) ? a : b;
1003 }
1004
1005 /* Compares two signed values A and B. Returns min value. */
1006
1007 double_int double_int_smin (double_int a, double_int b)
1008 {
1009 return (double_int_scmp (a, b) == -1) ? a : b;
1010 }
1011
1012 /* Compares two unsigned values A and B. Returns min value. */
1013
1014 double_int double_int_umin (double_int a, double_int b)
1015 {
1016 return (double_int_ucmp (a, b) == -1) ? a : b;
1017 }
1018
1019 /* Splits last digit of *CST (taken as unsigned) in BASE and returns it. */
1020
1021 static unsigned
1022 double_int_split_digit (double_int *cst, unsigned base)
1023 {
1024 unsigned HOST_WIDE_INT resl, reml;
1025 HOST_WIDE_INT resh, remh;
1026
1027 div_and_round_double (FLOOR_DIV_EXPR, true, cst->low, cst->high, base, 0,
1028 &resl, &resh, &reml, &remh);
1029 cst->high = resh;
1030 cst->low = resl;
1031
1032 return reml;
1033 }
1034
1035 /* Dumps CST to FILE. If UNS is true, CST is considered to be unsigned,
1036 otherwise it is signed. */
1037
1038 void
1039 dump_double_int (FILE *file, double_int cst, bool uns)
1040 {
1041 unsigned digits[100], n;
1042 int i;
1043
1044 if (double_int_zero_p (cst))
1045 {
1046 fprintf (file, "0");
1047 return;
1048 }
1049
1050 if (!uns && double_int_negative_p (cst))
1051 {
1052 fprintf (file, "-");
1053 cst = double_int_neg (cst);
1054 }
1055
1056 for (n = 0; !double_int_zero_p (cst); n++)
1057 digits[n] = double_int_split_digit (&cst, 10);
1058 for (i = n - 1; i >= 0; i--)
1059 fprintf (file, "%u", digits[i]);
1060 }
1061
1062
1063 /* Sets RESULT to VAL, taken unsigned if UNS is true and as signed
1064 otherwise. */
1065
1066 void
1067 mpz_set_double_int (mpz_t result, double_int val, bool uns)
1068 {
1069 bool negate = false;
1070 unsigned HOST_WIDE_INT vp[2];
1071
1072 if (!uns && double_int_negative_p (val))
1073 {
1074 negate = true;
1075 val = double_int_neg (val);
1076 }
1077
1078 vp[0] = val.low;
1079 vp[1] = (unsigned HOST_WIDE_INT) val.high;
1080 mpz_import (result, 2, -1, sizeof (HOST_WIDE_INT), 0, 0, vp);
1081
1082 if (negate)
1083 mpz_neg (result, result);
1084 }
1085
1086 /* Returns VAL converted to TYPE. If WRAP is true, then out-of-range
1087 values of VAL will be wrapped; otherwise, they will be set to the
1088 appropriate minimum or maximum TYPE bound. */
1089
1090 double_int
1091 mpz_get_double_int (const_tree type, mpz_t val, bool wrap)
1092 {
1093 unsigned HOST_WIDE_INT *vp;
1094 size_t count, numb;
1095 double_int res;
1096
1097 if (!wrap)
1098 {
1099 mpz_t min, max;
1100
1101 mpz_init (min);
1102 mpz_init (max);
1103 get_type_static_bounds (type, min, max);
1104
1105 if (mpz_cmp (val, min) < 0)
1106 mpz_set (val, min);
1107 else if (mpz_cmp (val, max) > 0)
1108 mpz_set (val, max);
1109
1110 mpz_clear (min);
1111 mpz_clear (max);
1112 }
1113
1114 /* Determine the number of unsigned HOST_WIDE_INT that are required
1115 for representing the value. The code to calculate count is
1116 extracted from the GMP manual, section "Integer Import and Export":
1117 http://gmplib.org/manual/Integer-Import-and-Export.html */
1118 numb = 8*sizeof(HOST_WIDE_INT);
1119 count = (mpz_sizeinbase (val, 2) + numb-1) / numb;
1120 if (count < 2)
1121 count = 2;
1122 vp = (unsigned HOST_WIDE_INT *) alloca (count * sizeof(HOST_WIDE_INT));
1123
1124 vp[0] = 0;
1125 vp[1] = 0;
1126 mpz_export (vp, &count, -1, sizeof (HOST_WIDE_INT), 0, 0, val);
1127
1128 gcc_assert (wrap || count <= 2);
1129
1130 res.low = vp[0];
1131 res.high = (HOST_WIDE_INT) vp[1];
1132
1133 res = double_int_ext (res, TYPE_PRECISION (type), TYPE_UNSIGNED (type));
1134 if (mpz_sgn (val) < 0)
1135 res = double_int_neg (res);
1136
1137 return res;
1138 }