1 | /* Operations with very long integers. |
2 | Copyright (C) 2012-2024 Free Software Foundation, Inc. |
3 | Contributed by Kenneth Zadeck <zadeck@naturalbridge.com> |
4 | |
5 | This file is part of GCC. |
6 | |
7 | GCC is free software; you can redistribute it and/or modify it |
8 | under the terms of the GNU General Public License as published by the |
9 | Free Software Foundation; either version 3, or (at your option) any |
10 | later version. |
11 | |
12 | GCC is distributed in the hope that it will be useful, but WITHOUT |
13 | ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
14 | FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
15 | for more details. |
16 | |
17 | You should have received a copy of the GNU General Public License |
18 | along with GCC; see the file COPYING3. If not see |
19 | <http://www.gnu.org/licenses/>. */ |
20 | |
21 | #include "config.h" |
22 | #include "system.h" |
23 | #include "coretypes.h" |
24 | #include "tm.h" |
25 | #include "tree.h" |
26 | #include "selftest.h" |
27 | |
28 | |
29 | #define HOST_BITS_PER_HALF_WIDE_INT 32 |
30 | #if HOST_BITS_PER_HALF_WIDE_INT == HOST_BITS_PER_LONG |
31 | # define HOST_HALF_WIDE_INT long |
32 | #elif HOST_BITS_PER_HALF_WIDE_INT == HOST_BITS_PER_INT |
33 | # define HOST_HALF_WIDE_INT int |
34 | #else |
35 | #error Please add support for HOST_HALF_WIDE_INT |
36 | #endif |
37 | |
38 | #define W_TYPE_SIZE HOST_BITS_PER_WIDE_INT |
39 | /* Do not include longlong.h when compiler is clang-based. See PR61146. */ |
40 | #if GCC_VERSION >= 3000 && (W_TYPE_SIZE == 32 || defined (__SIZEOF_INT128__)) && !defined(__clang__) |
41 | typedef unsigned HOST_HALF_WIDE_INT UHWtype; |
42 | typedef unsigned HOST_WIDE_INT UWtype; |
43 | typedef unsigned int UQItype __attribute__ ((mode (QI))); |
44 | typedef unsigned int USItype __attribute__ ((mode (SI))); |
45 | typedef unsigned int UDItype __attribute__ ((mode (DI))); |
46 | #if W_TYPE_SIZE == 32 |
47 | typedef unsigned int UDWtype __attribute__ ((mode (DI))); |
48 | #else |
49 | typedef unsigned int UDWtype __attribute__ ((mode (TI))); |
50 | #endif |
51 | #include "longlong.h" |
52 | #endif |
53 | |
54 | static const HOST_WIDE_INT zeros[1] = {}; |
55 | |
56 | /* |
57 | * Internal utilities. |
58 | */ |
59 | |
60 | /* Quantities to deal with values that hold half of a wide int. Used |
61 | in multiply and divide. */ |
62 | #define HALF_INT_MASK ((HOST_WIDE_INT_1 << HOST_BITS_PER_HALF_WIDE_INT) - 1) |
63 | |
64 | #define BLOCK_OF(TARGET) ((TARGET) / HOST_BITS_PER_WIDE_INT) |
65 | #define BLOCKS_NEEDED(PREC) (PREC ? CEIL (PREC, HOST_BITS_PER_WIDE_INT) : 1) |
66 | #define SIGN_MASK(X) ((HOST_WIDE_INT) (X) < 0 ? -1 : 0) |
67 | |
68 | /* Return the value a VAL[I] if I < LEN, otherwise, return 0 or -1 |
69 | based on the top existing bit of VAL. */ |
70 | |
71 | static unsigned HOST_WIDE_INT |
72 | safe_uhwi (const HOST_WIDE_INT *val, unsigned int len, unsigned int i) |
73 | { |
74 | return i < len ? val[i] : val[len - 1] < 0 ? HOST_WIDE_INT_M1 : 0; |
75 | } |
76 | |
77 | /* Convert the integer in VAL to canonical form, returning its new length. |
78 | LEN is the number of blocks currently in VAL and PRECISION is the number |
79 | of bits in the integer it represents. |
80 | |
81 | This function only changes the representation, not the value. */ |
82 | static unsigned int |
83 | canonize (HOST_WIDE_INT *val, unsigned int len, unsigned int precision) |
84 | { |
85 | unsigned int blocks_needed = BLOCKS_NEEDED (precision); |
86 | HOST_WIDE_INT top; |
87 | int i; |
88 | |
89 | if (len > blocks_needed) |
90 | len = blocks_needed; |
91 | |
92 | if (len == 1) |
93 | return len; |
94 | |
95 | top = val[len - 1]; |
96 | if (len * HOST_BITS_PER_WIDE_INT > precision) |
97 | val[len - 1] = top = sext_hwi (src: top, prec: precision % HOST_BITS_PER_WIDE_INT); |
98 | if (top != 0 && top != HOST_WIDE_INT_M1) |
99 | return len; |
100 | |
101 | /* At this point we know that the top is either 0 or -1. Find the |
102 | first block that is not a copy of this. */ |
103 | for (i = len - 2; i >= 0; i--) |
104 | { |
105 | HOST_WIDE_INT x = val[i]; |
106 | if (x != top) |
107 | { |
108 | if (SIGN_MASK (x) == top) |
109 | return i + 1; |
110 | |
111 | /* We need an extra block because the top bit block i does |
112 | not match the extension. */ |
113 | return i + 2; |
114 | } |
115 | } |
116 | |
117 | /* The number is 0 or -1. */ |
118 | return 1; |
119 | } |
120 | |
121 | /* VAL[0] is the unsigned result of an operation. Canonize it by adding |
122 | another 0 block if needed, and return number of blocks needed. */ |
123 | |
124 | static inline unsigned int |
125 | canonize_uhwi (HOST_WIDE_INT *val, unsigned int precision) |
126 | { |
127 | if (val[0] < 0 && precision > HOST_BITS_PER_WIDE_INT) |
128 | { |
129 | val[1] = 0; |
130 | return 2; |
131 | } |
132 | return 1; |
133 | } |
134 | |
135 | /* |
136 | * Conversion routines in and out of wide_int. |
137 | */ |
138 | |
139 | /* Copy XLEN elements from XVAL to VAL. If NEED_CANON, canonize the |
140 | result for an integer with precision PRECISION. Return the length |
141 | of VAL (after any canonization). */ |
142 | unsigned int |
143 | wi::from_array (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
144 | unsigned int xlen, unsigned int precision, bool need_canon) |
145 | { |
146 | for (unsigned i = 0; i < xlen; i++) |
147 | val[i] = xval[i]; |
148 | return need_canon ? canonize (val, len: xlen, precision) : xlen; |
149 | } |
150 | |
151 | /* Construct a wide int from a buffer of length LEN. BUFFER will be |
152 | read according to byte endianness and word endianness of the target. |
153 | Only the lower BUFFER_LEN bytes of the result are set; the remaining |
154 | high bytes are cleared. */ |
155 | wide_int |
156 | wi::from_buffer (const unsigned char *buffer, unsigned int buffer_len) |
157 | { |
158 | unsigned int precision = buffer_len * BITS_PER_UNIT; |
159 | wide_int result = wide_int::create (precision); |
160 | unsigned int words = buffer_len / UNITS_PER_WORD; |
161 | |
162 | /* We have to clear all the bits ourself, as we merely or in values |
163 | below. */ |
164 | unsigned int len = BLOCKS_NEEDED (precision); |
165 | HOST_WIDE_INT *val = result.write_val (0); |
166 | for (unsigned int i = 0; i < len; ++i) |
167 | val[i] = 0; |
168 | |
169 | for (unsigned int byte = 0; byte < buffer_len; byte++) |
170 | { |
171 | unsigned int offset; |
172 | unsigned int index; |
173 | unsigned int bitpos = byte * BITS_PER_UNIT; |
174 | unsigned HOST_WIDE_INT value; |
175 | |
176 | if (buffer_len > UNITS_PER_WORD) |
177 | { |
178 | unsigned int word = byte / UNITS_PER_WORD; |
179 | |
180 | if (WORDS_BIG_ENDIAN) |
181 | word = (words - 1) - word; |
182 | |
183 | offset = word * UNITS_PER_WORD; |
184 | |
185 | if (BYTES_BIG_ENDIAN) |
186 | offset += (UNITS_PER_WORD - 1) - (byte % UNITS_PER_WORD); |
187 | else |
188 | offset += byte % UNITS_PER_WORD; |
189 | } |
190 | else |
191 | offset = BYTES_BIG_ENDIAN ? (buffer_len - 1) - byte : byte; |
192 | |
193 | value = (unsigned HOST_WIDE_INT) buffer[offset]; |
194 | |
195 | index = bitpos / HOST_BITS_PER_WIDE_INT; |
196 | val[index] |= value << (bitpos % HOST_BITS_PER_WIDE_INT); |
197 | } |
198 | |
199 | result.set_len (l: canonize (val, len, precision)); |
200 | |
201 | return result; |
202 | } |
203 | |
204 | /* Sets RESULT from X, the sign is taken according to SGN. */ |
205 | void |
206 | wi::to_mpz (const wide_int_ref &x, mpz_t result, signop sgn) |
207 | { |
208 | int len = x.get_len (); |
209 | const HOST_WIDE_INT *v = x.get_val (); |
210 | int excess = len * HOST_BITS_PER_WIDE_INT - x.get_precision (); |
211 | |
212 | if (wi::neg_p (x, sgn)) |
213 | { |
214 | /* We use ones complement to avoid -x80..0 edge case that - |
215 | won't work on. */ |
216 | HOST_WIDE_INT *t = XALLOCAVEC (HOST_WIDE_INT, len); |
217 | for (int i = 0; i < len; i++) |
218 | t[i] = ~v[i]; |
219 | if (excess > 0) |
220 | t[len - 1] = (unsigned HOST_WIDE_INT) t[len - 1] << excess >> excess; |
221 | mpz_import (result, len, -1, sizeof (HOST_WIDE_INT), 0, 0, t); |
222 | mpz_com (result, result); |
223 | } |
224 | else if (excess > 0) |
225 | { |
226 | HOST_WIDE_INT *t = XALLOCAVEC (HOST_WIDE_INT, len); |
227 | for (int i = 0; i < len - 1; i++) |
228 | t[i] = v[i]; |
229 | t[len - 1] = (unsigned HOST_WIDE_INT) v[len - 1] << excess >> excess; |
230 | mpz_import (result, len, -1, sizeof (HOST_WIDE_INT), 0, 0, t); |
231 | } |
232 | else if (excess < 0 && wi::neg_p (x)) |
233 | { |
234 | int = CEIL (-excess, HOST_BITS_PER_WIDE_INT); |
235 | HOST_WIDE_INT *t = XALLOCAVEC (HOST_WIDE_INT, len + extra); |
236 | for (int i = 0; i < len; i++) |
237 | t[i] = v[i]; |
238 | for (int i = 0; i < extra; i++) |
239 | t[len + i] = -1; |
240 | excess = (-excess) % HOST_BITS_PER_WIDE_INT; |
241 | if (excess) |
242 | t[len + extra - 1] = (HOST_WIDE_INT_1U << excess) - 1; |
243 | mpz_import (result, len + extra, -1, sizeof (HOST_WIDE_INT), 0, 0, t); |
244 | } |
245 | else |
246 | mpz_import (result, len, -1, sizeof (HOST_WIDE_INT), 0, 0, v); |
247 | } |
248 | |
249 | /* Returns X converted to TYPE. If WRAP is true, then out-of-range |
250 | values of VAL will be wrapped; otherwise, they will be set to the |
251 | appropriate minimum or maximum TYPE bound. */ |
252 | wide_int |
253 | wi::from_mpz (const_tree type, mpz_t x, bool wrap) |
254 | { |
255 | size_t count, numb; |
256 | unsigned int prec = TYPE_PRECISION (type); |
257 | wide_int res = wide_int::create (precision: prec); |
258 | |
259 | if (!wrap) |
260 | { |
261 | mpz_t min, max; |
262 | |
263 | mpz_init (min); |
264 | mpz_init (max); |
265 | get_type_static_bounds (type, min, max); |
266 | |
267 | if (mpz_cmp (x, min) < 0) |
268 | mpz_set (x, min); |
269 | else if (mpz_cmp (x, max) > 0) |
270 | mpz_set (x, max); |
271 | |
272 | mpz_clear (min); |
273 | mpz_clear (max); |
274 | } |
275 | |
276 | /* Determine the number of unsigned HOST_WIDE_INTs that are required |
277 | for representing the absolute value. The code to calculate count is |
278 | extracted from the GMP manual, section "Integer Import and Export": |
279 | http://gmplib.org/manual/Integer-Import-and-Export.html */ |
280 | numb = CHAR_BIT * sizeof (HOST_WIDE_INT); |
281 | count = CEIL (mpz_sizeinbase (x, 2), numb); |
282 | HOST_WIDE_INT *val = res.write_val (0); |
283 | /* Read the absolute value. |
284 | |
285 | Write directly to the wide_int storage if possible, otherwise leave |
286 | GMP to allocate the memory for us. It might be slightly more efficient |
287 | to use mpz_tdiv_r_2exp for the latter case, but the situation is |
288 | pathological and it seems safer to operate on the original mpz value |
289 | in all cases. */ |
290 | void *valres = mpz_export (count <= WIDE_INT_MAX_INL_ELTS ? val : 0, |
291 | &count, -1, sizeof (HOST_WIDE_INT), 0, 0, x); |
292 | if (count < 1) |
293 | { |
294 | val[0] = 0; |
295 | count = 1; |
296 | } |
297 | count = MIN (count, BLOCKS_NEEDED (prec)); |
298 | if (valres != val) |
299 | { |
300 | memcpy (dest: val, src: valres, n: count * sizeof (HOST_WIDE_INT)); |
301 | free (ptr: valres); |
302 | } |
303 | /* Zero-extend the absolute value to PREC bits. */ |
304 | if (count < BLOCKS_NEEDED (prec) && val[count - 1] < 0) |
305 | val[count++] = 0; |
306 | else |
307 | count = canonize (val, len: count, precision: prec); |
308 | res.set_len (l: count); |
309 | |
310 | if (mpz_sgn (x) < 0) |
311 | res = -res; |
312 | |
313 | return res; |
314 | } |
315 | |
316 | /* |
317 | * Largest and smallest values in a mode. |
318 | */ |
319 | |
320 | /* Return the largest SGNed number that is representable in PRECISION bits. |
321 | |
322 | TODO: There is still code from the double_int era that trys to |
323 | make up for the fact that double int's could not represent the |
324 | min and max values of all types. This code should be removed |
325 | because the min and max values can always be represented in |
326 | wide_ints and int-csts. */ |
327 | wide_int |
328 | wi::max_value (unsigned int precision, signop sgn) |
329 | { |
330 | gcc_checking_assert (precision != 0); |
331 | if (sgn == UNSIGNED) |
332 | /* The unsigned max is just all ones. */ |
333 | return shwi (val: -1, precision); |
334 | else |
335 | /* The signed max is all ones except the top bit. This must be |
336 | explicitly represented. */ |
337 | return mask (width: precision - 1, negate_p: false, precision); |
338 | } |
339 | |
340 | /* Return the largest SGNed number that is representable in PRECISION bits. */ |
341 | wide_int |
342 | wi::min_value (unsigned int precision, signop sgn) |
343 | { |
344 | gcc_checking_assert (precision != 0); |
345 | if (sgn == UNSIGNED) |
346 | return uhwi (val: 0, precision); |
347 | else |
348 | /* The signed min is all zeros except the top bit. This must be |
349 | explicitly represented. */ |
350 | return wi::set_bit_in_zero (bit: precision - 1, precision); |
351 | } |
352 | |
353 | /* |
354 | * Public utilities. |
355 | */ |
356 | |
357 | /* Convert the number represented by XVAL, XLEN and XPRECISION, which has |
358 | signedness SGN, to an integer that has PRECISION bits. Store the blocks |
359 | in VAL and return the number of blocks used. |
360 | |
361 | This function can handle both extension (PRECISION > XPRECISION) |
362 | and truncation (PRECISION < XPRECISION). */ |
363 | unsigned int |
364 | wi::force_to_size (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
365 | unsigned int xlen, unsigned int xprecision, |
366 | unsigned int precision, signop sgn) |
367 | { |
368 | unsigned int blocks_needed = BLOCKS_NEEDED (precision); |
369 | unsigned int len = blocks_needed < xlen ? blocks_needed : xlen; |
370 | for (unsigned i = 0; i < len; i++) |
371 | val[i] = xval[i]; |
372 | |
373 | if (precision > xprecision) |
374 | { |
375 | unsigned int small_xprecision = xprecision % HOST_BITS_PER_WIDE_INT; |
376 | |
377 | /* Expanding. */ |
378 | if (sgn == UNSIGNED) |
379 | { |
380 | if (small_xprecision && len == BLOCKS_NEEDED (xprecision)) |
381 | val[len - 1] = zext_hwi (src: val[len - 1], prec: small_xprecision); |
382 | else if (val[len - 1] < 0) |
383 | { |
384 | while (len < BLOCKS_NEEDED (xprecision)) |
385 | val[len++] = -1; |
386 | if (small_xprecision) |
387 | val[len - 1] = zext_hwi (src: val[len - 1], prec: small_xprecision); |
388 | else |
389 | val[len++] = 0; |
390 | } |
391 | } |
392 | else |
393 | { |
394 | if (small_xprecision && len == BLOCKS_NEEDED (xprecision)) |
395 | val[len - 1] = sext_hwi (src: val[len - 1], prec: small_xprecision); |
396 | } |
397 | } |
398 | len = canonize (val, len, precision); |
399 | |
400 | return len; |
401 | } |
402 | |
403 | /* This function hides the fact that we cannot rely on the bits beyond |
404 | the precision. This issue comes up in the relational comparisions |
405 | where we do allow comparisons of values of different precisions. */ |
406 | static inline HOST_WIDE_INT |
407 | selt (const HOST_WIDE_INT *a, unsigned int len, |
408 | unsigned int blocks_needed, unsigned int small_prec, |
409 | unsigned int index, signop sgn) |
410 | { |
411 | HOST_WIDE_INT val; |
412 | if (index < len) |
413 | val = a[index]; |
414 | else if (index < blocks_needed || sgn == SIGNED) |
415 | /* Signed or within the precision. */ |
416 | val = SIGN_MASK (a[len - 1]); |
417 | else |
418 | /* Unsigned extension beyond the precision. */ |
419 | val = 0; |
420 | |
421 | if (small_prec && index == blocks_needed - 1) |
422 | return (sgn == SIGNED |
423 | ? sext_hwi (src: val, prec: small_prec) |
424 | : zext_hwi (src: val, prec: small_prec)); |
425 | else |
426 | return val; |
427 | } |
428 | |
429 | /* Find the highest bit represented in a wide int. This will in |
430 | general have the same value as the sign bit. */ |
431 | static inline HOST_WIDE_INT |
432 | top_bit_of (const HOST_WIDE_INT *a, unsigned int len, unsigned int prec) |
433 | { |
434 | int excess = len * HOST_BITS_PER_WIDE_INT - prec; |
435 | unsigned HOST_WIDE_INT val = a[len - 1]; |
436 | if (excess > 0) |
437 | val <<= excess; |
438 | return val >> (HOST_BITS_PER_WIDE_INT - 1); |
439 | } |
440 | |
441 | /* |
442 | * Comparisons, note that only equality is an operator. The other |
443 | * comparisons cannot be operators since they are inherently signed or |
444 | * unsigned and C++ has no such operators. |
445 | */ |
446 | |
447 | /* Return true if OP0 == OP1. */ |
448 | bool |
449 | wi::eq_p_large (const HOST_WIDE_INT *op0, unsigned int op0len, |
450 | const HOST_WIDE_INT *op1, unsigned int op1len, |
451 | unsigned int prec) |
452 | { |
453 | int l0 = op0len - 1; |
454 | unsigned int small_prec = prec & (HOST_BITS_PER_WIDE_INT - 1); |
455 | |
456 | if (op0len != op1len) |
457 | return false; |
458 | |
459 | if (op0len == BLOCKS_NEEDED (prec) && small_prec) |
460 | { |
461 | /* It does not matter if we zext or sext here, we just have to |
462 | do both the same way. */ |
463 | if (zext_hwi (src: op0 [l0], prec: small_prec) != zext_hwi (src: op1 [l0], prec: small_prec)) |
464 | return false; |
465 | l0--; |
466 | } |
467 | |
468 | while (l0 >= 0) |
469 | if (op0[l0] != op1[l0]) |
470 | return false; |
471 | else |
472 | l0--; |
473 | |
474 | return true; |
475 | } |
476 | |
477 | /* Return true if OP0 < OP1 using signed comparisons. */ |
478 | bool |
479 | wi::lts_p_large (const HOST_WIDE_INT *op0, unsigned int op0len, |
480 | unsigned int precision, |
481 | const HOST_WIDE_INT *op1, unsigned int op1len) |
482 | { |
483 | HOST_WIDE_INT s0, s1; |
484 | unsigned HOST_WIDE_INT u0, u1; |
485 | unsigned int blocks_needed = BLOCKS_NEEDED (precision); |
486 | unsigned int small_prec = precision & (HOST_BITS_PER_WIDE_INT - 1); |
487 | int l = MAX (op0len - 1, op1len - 1); |
488 | |
489 | /* Only the top block is compared as signed. The rest are unsigned |
490 | comparisons. */ |
491 | s0 = selt (a: op0, len: op0len, blocks_needed, small_prec, index: l, sgn: SIGNED); |
492 | s1 = selt (a: op1, len: op1len, blocks_needed, small_prec, index: l, sgn: SIGNED); |
493 | if (s0 < s1) |
494 | return true; |
495 | if (s0 > s1) |
496 | return false; |
497 | |
498 | l--; |
499 | while (l >= 0) |
500 | { |
501 | u0 = selt (a: op0, len: op0len, blocks_needed, small_prec, index: l, sgn: SIGNED); |
502 | u1 = selt (a: op1, len: op1len, blocks_needed, small_prec, index: l, sgn: SIGNED); |
503 | |
504 | if (u0 < u1) |
505 | return true; |
506 | if (u0 > u1) |
507 | return false; |
508 | l--; |
509 | } |
510 | |
511 | return false; |
512 | } |
513 | |
514 | /* Returns -1 if OP0 < OP1, 0 if OP0 == OP1 and 1 if OP0 > OP1 using |
515 | signed compares. */ |
516 | int |
517 | wi::cmps_large (const HOST_WIDE_INT *op0, unsigned int op0len, |
518 | unsigned int precision, |
519 | const HOST_WIDE_INT *op1, unsigned int op1len) |
520 | { |
521 | HOST_WIDE_INT s0, s1; |
522 | unsigned HOST_WIDE_INT u0, u1; |
523 | unsigned int blocks_needed = BLOCKS_NEEDED (precision); |
524 | unsigned int small_prec = precision & (HOST_BITS_PER_WIDE_INT - 1); |
525 | int l = MAX (op0len - 1, op1len - 1); |
526 | |
527 | /* Only the top block is compared as signed. The rest are unsigned |
528 | comparisons. */ |
529 | s0 = selt (a: op0, len: op0len, blocks_needed, small_prec, index: l, sgn: SIGNED); |
530 | s1 = selt (a: op1, len: op1len, blocks_needed, small_prec, index: l, sgn: SIGNED); |
531 | if (s0 < s1) |
532 | return -1; |
533 | if (s0 > s1) |
534 | return 1; |
535 | |
536 | l--; |
537 | while (l >= 0) |
538 | { |
539 | u0 = selt (a: op0, len: op0len, blocks_needed, small_prec, index: l, sgn: SIGNED); |
540 | u1 = selt (a: op1, len: op1len, blocks_needed, small_prec, index: l, sgn: SIGNED); |
541 | |
542 | if (u0 < u1) |
543 | return -1; |
544 | if (u0 > u1) |
545 | return 1; |
546 | l--; |
547 | } |
548 | |
549 | return 0; |
550 | } |
551 | |
552 | /* Return true if OP0 < OP1 using unsigned comparisons. */ |
553 | bool |
554 | wi::ltu_p_large (const HOST_WIDE_INT *op0, unsigned int op0len, |
555 | unsigned int precision, |
556 | const HOST_WIDE_INT *op1, unsigned int op1len) |
557 | { |
558 | unsigned HOST_WIDE_INT x0; |
559 | unsigned HOST_WIDE_INT x1; |
560 | unsigned int blocks_needed = BLOCKS_NEEDED (precision); |
561 | unsigned int small_prec = precision & (HOST_BITS_PER_WIDE_INT - 1); |
562 | int l = MAX (op0len - 1, op1len - 1); |
563 | |
564 | while (l >= 0) |
565 | { |
566 | x0 = selt (a: op0, len: op0len, blocks_needed, small_prec, index: l, sgn: UNSIGNED); |
567 | x1 = selt (a: op1, len: op1len, blocks_needed, small_prec, index: l, sgn: UNSIGNED); |
568 | if (x0 < x1) |
569 | return true; |
570 | if (x0 > x1) |
571 | return false; |
572 | l--; |
573 | } |
574 | |
575 | return false; |
576 | } |
577 | |
578 | /* Returns -1 if OP0 < OP1, 0 if OP0 == OP1 and 1 if OP0 > OP1 using |
579 | unsigned compares. */ |
580 | int |
581 | wi::cmpu_large (const HOST_WIDE_INT *op0, unsigned int op0len, |
582 | unsigned int precision, |
583 | const HOST_WIDE_INT *op1, unsigned int op1len) |
584 | { |
585 | unsigned HOST_WIDE_INT x0; |
586 | unsigned HOST_WIDE_INT x1; |
587 | unsigned int blocks_needed = BLOCKS_NEEDED (precision); |
588 | unsigned int small_prec = precision & (HOST_BITS_PER_WIDE_INT - 1); |
589 | int l = MAX (op0len - 1, op1len - 1); |
590 | |
591 | while (l >= 0) |
592 | { |
593 | x0 = selt (a: op0, len: op0len, blocks_needed, small_prec, index: l, sgn: UNSIGNED); |
594 | x1 = selt (a: op1, len: op1len, blocks_needed, small_prec, index: l, sgn: UNSIGNED); |
595 | if (x0 < x1) |
596 | return -1; |
597 | if (x0 > x1) |
598 | return 1; |
599 | l--; |
600 | } |
601 | |
602 | return 0; |
603 | } |
604 | |
605 | /* |
606 | * Extension. |
607 | */ |
608 | |
609 | /* Sign-extend the number represented by XVAL and XLEN into VAL, |
610 | starting at OFFSET. Return the number of blocks in VAL. Both XVAL |
611 | and VAL have PRECISION bits. */ |
612 | unsigned int |
613 | wi::sext_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
614 | unsigned int xlen, unsigned int precision, unsigned int offset) |
615 | { |
616 | unsigned int len = offset / HOST_BITS_PER_WIDE_INT; |
617 | /* Extending beyond the precision is a no-op. If we have only stored |
618 | OFFSET bits or fewer, the rest are already signs. */ |
619 | if (offset >= precision || len >= xlen) |
620 | { |
621 | for (unsigned i = 0; i < xlen; ++i) |
622 | val[i] = xval[i]; |
623 | return xlen; |
624 | } |
625 | unsigned int suboffset = offset % HOST_BITS_PER_WIDE_INT; |
626 | for (unsigned int i = 0; i < len; i++) |
627 | val[i] = xval[i]; |
628 | if (suboffset > 0) |
629 | { |
630 | val[len] = sext_hwi (src: xval[len], prec: suboffset); |
631 | len += 1; |
632 | } |
633 | return canonize (val, len, precision); |
634 | } |
635 | |
636 | /* Zero-extend the number represented by XVAL and XLEN into VAL, |
637 | starting at OFFSET. Return the number of blocks in VAL. Both XVAL |
638 | and VAL have PRECISION bits. */ |
639 | unsigned int |
640 | wi::zext_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
641 | unsigned int xlen, unsigned int precision, unsigned int offset) |
642 | { |
643 | unsigned int len = offset / HOST_BITS_PER_WIDE_INT; |
644 | /* Extending beyond the precision is a no-op. If we have only stored |
645 | OFFSET bits or fewer, and the upper stored bit is zero, then there |
646 | is nothing to do. */ |
647 | if (offset >= precision || (len >= xlen && xval[xlen - 1] >= 0)) |
648 | { |
649 | for (unsigned i = 0; i < xlen; ++i) |
650 | val[i] = xval[i]; |
651 | return xlen; |
652 | } |
653 | unsigned int suboffset = offset % HOST_BITS_PER_WIDE_INT; |
654 | for (unsigned int i = 0; i < len; i++) |
655 | val[i] = i < xlen ? xval[i] : -1; |
656 | if (suboffset > 0) |
657 | val[len] = zext_hwi (src: len < xlen ? xval[len] : -1, prec: suboffset); |
658 | else |
659 | val[len] = 0; |
660 | return canonize (val, len: len + 1, precision); |
661 | } |
662 | |
663 | /* |
664 | * Masking, inserting, shifting, rotating. |
665 | */ |
666 | |
667 | /* Insert WIDTH bits from Y into X starting at START. */ |
668 | wide_int |
669 | wi::insert (const wide_int &x, const wide_int &y, unsigned int start, |
670 | unsigned int width) |
671 | { |
672 | wide_int result; |
673 | wide_int mask; |
674 | wide_int tmp; |
675 | |
676 | unsigned int precision = x.get_precision (); |
677 | if (start >= precision) |
678 | return x; |
679 | |
680 | gcc_checking_assert (precision >= width); |
681 | |
682 | if (start + width >= precision) |
683 | width = precision - start; |
684 | |
685 | mask = wi::shifted_mask (start, width, negate_p: false, precision); |
686 | tmp = wi::lshift (x: wide_int::from (x: y, precision, sgn: UNSIGNED), y: start); |
687 | result = tmp & mask; |
688 | |
689 | tmp = wi::bit_and_not (x, y: mask); |
690 | result = result | tmp; |
691 | |
692 | return result; |
693 | } |
694 | |
695 | /* Copy the number represented by XVAL and XLEN into VAL, setting bit BIT. |
696 | Return the number of blocks in VAL. Both XVAL and VAL have PRECISION |
697 | bits. */ |
698 | unsigned int |
699 | wi::set_bit_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
700 | unsigned int xlen, unsigned int precision, unsigned int bit) |
701 | { |
702 | unsigned int block = bit / HOST_BITS_PER_WIDE_INT; |
703 | unsigned int subbit = bit % HOST_BITS_PER_WIDE_INT; |
704 | |
705 | if (block + 1 >= xlen) |
706 | { |
707 | /* The operation either affects the last current block or needs |
708 | a new block. */ |
709 | unsigned int len = block + 1; |
710 | for (unsigned int i = 0; i < len; i++) |
711 | val[i] = safe_uhwi (val: xval, len: xlen, i); |
712 | val[block] |= HOST_WIDE_INT_1U << subbit; |
713 | |
714 | /* If the bit we just set is at the msb of the block, make sure |
715 | that any higher bits are zeros. */ |
716 | if (bit + 1 < precision && subbit == HOST_BITS_PER_WIDE_INT - 1) |
717 | { |
718 | val[len++] = 0; |
719 | return len; |
720 | } |
721 | return canonize (val, len, precision); |
722 | } |
723 | else |
724 | { |
725 | for (unsigned int i = 0; i < xlen; i++) |
726 | val[i] = xval[i]; |
727 | val[block] |= HOST_WIDE_INT_1U << subbit; |
728 | return canonize (val, len: xlen, precision); |
729 | } |
730 | } |
731 | |
732 | /* Byte swap the integer represented by XVAL and XLEN into VAL. Return |
733 | the number of blocks in VAL. Both XVAL and VAL have PRECISION bits. */ |
734 | unsigned int |
735 | wi::bswap_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
736 | unsigned int xlen, unsigned int precision) |
737 | { |
738 | unsigned int s, len = BLOCKS_NEEDED (precision); |
739 | |
740 | /* This is not a well defined operation if the precision is not a |
741 | multiple of 8. */ |
742 | gcc_assert ((precision & 0x7) == 0); |
743 | |
744 | memset (s: val, c: 0, n: sizeof (unsigned HOST_WIDE_INT) * len); |
745 | |
746 | /* Only swap the bytes that are not the padding. */ |
747 | for (s = 0; s < precision; s += 8) |
748 | { |
749 | unsigned int d = precision - s - 8; |
750 | unsigned HOST_WIDE_INT byte; |
751 | |
752 | unsigned int block = s / HOST_BITS_PER_WIDE_INT; |
753 | unsigned int offset = s & (HOST_BITS_PER_WIDE_INT - 1); |
754 | |
755 | byte = (safe_uhwi (val: xval, len: xlen, i: block) >> offset) & 0xff; |
756 | |
757 | block = d / HOST_BITS_PER_WIDE_INT; |
758 | offset = d & (HOST_BITS_PER_WIDE_INT - 1); |
759 | |
760 | val[block] |= byte << offset; |
761 | } |
762 | |
763 | return canonize (val, len, precision); |
764 | } |
765 | |
766 | /* Bitreverse the integer represented by XVAL and LEN into VAL. Return |
767 | the number of blocks in VAL. Both XVAL and VAL have PRECISION bits. */ |
768 | unsigned int |
769 | wi::bitreverse_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
770 | unsigned int len, unsigned int precision) |
771 | { |
772 | unsigned int i, s; |
773 | |
774 | for (i = 0; i < len; i++) |
775 | val[i] = 0; |
776 | |
777 | for (s = 0; s < precision; s++) |
778 | { |
779 | unsigned int block = s / HOST_BITS_PER_WIDE_INT; |
780 | unsigned int offset = s & (HOST_BITS_PER_WIDE_INT - 1); |
781 | if (((safe_uhwi (val: xval, len, i: block) >> offset) & 1) != 0) |
782 | { |
783 | unsigned int d = (precision - 1) - s; |
784 | block = d / HOST_BITS_PER_WIDE_INT; |
785 | offset = d & (HOST_BITS_PER_WIDE_INT - 1); |
786 | val[block] |= HOST_WIDE_INT_1U << offset; |
787 | } |
788 | } |
789 | |
790 | return canonize (val, len, precision); |
791 | } |
792 | |
793 | /* Fill VAL with a mask where the lower WIDTH bits are ones and the bits |
794 | above that up to PREC are zeros. The result is inverted if NEGATE |
795 | is true. Return the number of blocks in VAL. */ |
796 | unsigned int |
797 | wi::mask (HOST_WIDE_INT *val, unsigned int width, bool negate, |
798 | unsigned int prec) |
799 | { |
800 | if (width >= prec) |
801 | { |
802 | val[0] = negate ? 0 : -1; |
803 | return 1; |
804 | } |
805 | else if (width == 0) |
806 | { |
807 | val[0] = negate ? -1 : 0; |
808 | return 1; |
809 | } |
810 | |
811 | unsigned int i = 0; |
812 | while (i < width / HOST_BITS_PER_WIDE_INT) |
813 | val[i++] = negate ? 0 : -1; |
814 | |
815 | unsigned int shift = width & (HOST_BITS_PER_WIDE_INT - 1); |
816 | if (shift != 0) |
817 | { |
818 | HOST_WIDE_INT last = (HOST_WIDE_INT_1U << shift) - 1; |
819 | val[i++] = negate ? ~last : last; |
820 | } |
821 | else |
822 | val[i++] = negate ? -1 : 0; |
823 | |
824 | return i; |
825 | } |
826 | |
827 | /* Fill VAL with a mask where the lower START bits are zeros, the next WIDTH |
828 | bits are ones, and the bits above that up to PREC are zeros. The result |
829 | is inverted if NEGATE is true. Return the number of blocks in VAL. */ |
830 | unsigned int |
831 | wi::shifted_mask (HOST_WIDE_INT *val, unsigned int start, unsigned int width, |
832 | bool negate, unsigned int prec) |
833 | { |
834 | if (start >= prec || width == 0) |
835 | { |
836 | val[0] = negate ? -1 : 0; |
837 | return 1; |
838 | } |
839 | |
840 | if (width > prec - start) |
841 | width = prec - start; |
842 | unsigned int end = start + width; |
843 | |
844 | unsigned int i = 0; |
845 | while (i < start / HOST_BITS_PER_WIDE_INT) |
846 | val[i++] = negate ? -1 : 0; |
847 | |
848 | unsigned int shift = start & (HOST_BITS_PER_WIDE_INT - 1); |
849 | if (shift) |
850 | { |
851 | HOST_WIDE_INT block = (HOST_WIDE_INT_1U << shift) - 1; |
852 | shift += width; |
853 | if (shift < HOST_BITS_PER_WIDE_INT) |
854 | { |
855 | /* case 000111000 */ |
856 | block = (HOST_WIDE_INT_1U << shift) - block - 1; |
857 | val[i++] = negate ? ~block : block; |
858 | return i; |
859 | } |
860 | else |
861 | /* ...111000 */ |
862 | val[i++] = negate ? block : ~block; |
863 | } |
864 | |
865 | if (end >= prec) |
866 | { |
867 | if (!shift) |
868 | val[i++] = negate ? 0 : -1; |
869 | return i; |
870 | } |
871 | |
872 | while (i < end / HOST_BITS_PER_WIDE_INT) |
873 | /* 1111111 */ |
874 | val[i++] = negate ? 0 : -1; |
875 | |
876 | shift = end & (HOST_BITS_PER_WIDE_INT - 1); |
877 | if (shift != 0) |
878 | { |
879 | /* 000011111 */ |
880 | HOST_WIDE_INT block = (HOST_WIDE_INT_1U << shift) - 1; |
881 | val[i++] = negate ? ~block : block; |
882 | } |
883 | else |
884 | val[i++] = negate ? -1 : 0; |
885 | |
886 | return i; |
887 | } |
888 | |
889 | /* |
890 | * logical operations. |
891 | */ |
892 | |
893 | /* Set VAL to OP0 & OP1. Return the number of blocks used. */ |
894 | unsigned int |
895 | wi::and_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *op0, |
896 | unsigned int op0len, const HOST_WIDE_INT *op1, |
897 | unsigned int op1len, unsigned int prec) |
898 | { |
899 | int l0 = op0len - 1; |
900 | int l1 = op1len - 1; |
901 | bool need_canon = true; |
902 | |
903 | unsigned int len = MAX (op0len, op1len); |
904 | if (l0 > l1) |
905 | { |
906 | HOST_WIDE_INT op1mask = -top_bit_of (a: op1, len: op1len, prec); |
907 | if (op1mask == 0) |
908 | { |
909 | l0 = l1; |
910 | len = l1 + 1; |
911 | } |
912 | else |
913 | { |
914 | need_canon = false; |
915 | while (l0 > l1) |
916 | { |
917 | val[l0] = op0[l0]; |
918 | l0--; |
919 | } |
920 | } |
921 | } |
922 | else if (l1 > l0) |
923 | { |
924 | HOST_WIDE_INT op0mask = -top_bit_of (a: op0, len: op0len, prec); |
925 | if (op0mask == 0) |
926 | len = l0 + 1; |
927 | else |
928 | { |
929 | need_canon = false; |
930 | while (l1 > l0) |
931 | { |
932 | val[l1] = op1[l1]; |
933 | l1--; |
934 | } |
935 | } |
936 | } |
937 | |
938 | while (l0 >= 0) |
939 | { |
940 | val[l0] = op0[l0] & op1[l0]; |
941 | l0--; |
942 | } |
943 | |
944 | if (need_canon) |
945 | len = canonize (val, len, precision: prec); |
946 | |
947 | return len; |
948 | } |
949 | |
950 | /* Set VAL to OP0 & ~OP1. Return the number of blocks used. */ |
951 | unsigned int |
952 | wi::and_not_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *op0, |
953 | unsigned int op0len, const HOST_WIDE_INT *op1, |
954 | unsigned int op1len, unsigned int prec) |
955 | { |
956 | wide_int result; |
957 | int l0 = op0len - 1; |
958 | int l1 = op1len - 1; |
959 | bool need_canon = true; |
960 | |
961 | unsigned int len = MAX (op0len, op1len); |
962 | if (l0 > l1) |
963 | { |
964 | HOST_WIDE_INT op1mask = -top_bit_of (a: op1, len: op1len, prec); |
965 | if (op1mask != 0) |
966 | { |
967 | l0 = l1; |
968 | len = l1 + 1; |
969 | } |
970 | else |
971 | { |
972 | need_canon = false; |
973 | while (l0 > l1) |
974 | { |
975 | val[l0] = op0[l0]; |
976 | l0--; |
977 | } |
978 | } |
979 | } |
980 | else if (l1 > l0) |
981 | { |
982 | HOST_WIDE_INT op0mask = -top_bit_of (a: op0, len: op0len, prec); |
983 | if (op0mask == 0) |
984 | len = l0 + 1; |
985 | else |
986 | { |
987 | need_canon = false; |
988 | while (l1 > l0) |
989 | { |
990 | val[l1] = ~op1[l1]; |
991 | l1--; |
992 | } |
993 | } |
994 | } |
995 | |
996 | while (l0 >= 0) |
997 | { |
998 | val[l0] = op0[l0] & ~op1[l0]; |
999 | l0--; |
1000 | } |
1001 | |
1002 | if (need_canon) |
1003 | len = canonize (val, len, precision: prec); |
1004 | |
1005 | return len; |
1006 | } |
1007 | |
1008 | /* Set VAL to OP0 | OP1. Return the number of blocks used. */ |
1009 | unsigned int |
1010 | wi::or_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *op0, |
1011 | unsigned int op0len, const HOST_WIDE_INT *op1, |
1012 | unsigned int op1len, unsigned int prec) |
1013 | { |
1014 | wide_int result; |
1015 | int l0 = op0len - 1; |
1016 | int l1 = op1len - 1; |
1017 | bool need_canon = true; |
1018 | |
1019 | unsigned int len = MAX (op0len, op1len); |
1020 | if (l0 > l1) |
1021 | { |
1022 | HOST_WIDE_INT op1mask = -top_bit_of (a: op1, len: op1len, prec); |
1023 | if (op1mask != 0) |
1024 | { |
1025 | l0 = l1; |
1026 | len = l1 + 1; |
1027 | } |
1028 | else |
1029 | { |
1030 | need_canon = false; |
1031 | while (l0 > l1) |
1032 | { |
1033 | val[l0] = op0[l0]; |
1034 | l0--; |
1035 | } |
1036 | } |
1037 | } |
1038 | else if (l1 > l0) |
1039 | { |
1040 | HOST_WIDE_INT op0mask = -top_bit_of (a: op0, len: op0len, prec); |
1041 | if (op0mask != 0) |
1042 | len = l0 + 1; |
1043 | else |
1044 | { |
1045 | need_canon = false; |
1046 | while (l1 > l0) |
1047 | { |
1048 | val[l1] = op1[l1]; |
1049 | l1--; |
1050 | } |
1051 | } |
1052 | } |
1053 | |
1054 | while (l0 >= 0) |
1055 | { |
1056 | val[l0] = op0[l0] | op1[l0]; |
1057 | l0--; |
1058 | } |
1059 | |
1060 | if (need_canon) |
1061 | len = canonize (val, len, precision: prec); |
1062 | |
1063 | return len; |
1064 | } |
1065 | |
1066 | /* Set VAL to OP0 | ~OP1. Return the number of blocks used. */ |
1067 | unsigned int |
1068 | wi::or_not_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *op0, |
1069 | unsigned int op0len, const HOST_WIDE_INT *op1, |
1070 | unsigned int op1len, unsigned int prec) |
1071 | { |
1072 | wide_int result; |
1073 | int l0 = op0len - 1; |
1074 | int l1 = op1len - 1; |
1075 | bool need_canon = true; |
1076 | |
1077 | unsigned int len = MAX (op0len, op1len); |
1078 | if (l0 > l1) |
1079 | { |
1080 | HOST_WIDE_INT op1mask = -top_bit_of (a: op1, len: op1len, prec); |
1081 | if (op1mask == 0) |
1082 | { |
1083 | l0 = l1; |
1084 | len = l1 + 1; |
1085 | } |
1086 | else |
1087 | { |
1088 | need_canon = false; |
1089 | while (l0 > l1) |
1090 | { |
1091 | val[l0] = op0[l0]; |
1092 | l0--; |
1093 | } |
1094 | } |
1095 | } |
1096 | else if (l1 > l0) |
1097 | { |
1098 | HOST_WIDE_INT op0mask = -top_bit_of (a: op0, len: op0len, prec); |
1099 | if (op0mask != 0) |
1100 | len = l0 + 1; |
1101 | else |
1102 | { |
1103 | need_canon = false; |
1104 | while (l1 > l0) |
1105 | { |
1106 | val[l1] = ~op1[l1]; |
1107 | l1--; |
1108 | } |
1109 | } |
1110 | } |
1111 | |
1112 | while (l0 >= 0) |
1113 | { |
1114 | val[l0] = op0[l0] | ~op1[l0]; |
1115 | l0--; |
1116 | } |
1117 | |
1118 | if (need_canon) |
1119 | len = canonize (val, len, precision: prec); |
1120 | |
1121 | return len; |
1122 | } |
1123 | |
1124 | /* Set VAL to OP0 ^ OP1. Return the number of blocks used. */ |
1125 | unsigned int |
1126 | wi::xor_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *op0, |
1127 | unsigned int op0len, const HOST_WIDE_INT *op1, |
1128 | unsigned int op1len, unsigned int prec) |
1129 | { |
1130 | wide_int result; |
1131 | int l0 = op0len - 1; |
1132 | int l1 = op1len - 1; |
1133 | |
1134 | unsigned int len = MAX (op0len, op1len); |
1135 | if (l0 > l1) |
1136 | { |
1137 | HOST_WIDE_INT op1mask = -top_bit_of (a: op1, len: op1len, prec); |
1138 | while (l0 > l1) |
1139 | { |
1140 | val[l0] = op0[l0] ^ op1mask; |
1141 | l0--; |
1142 | } |
1143 | } |
1144 | |
1145 | if (l1 > l0) |
1146 | { |
1147 | HOST_WIDE_INT op0mask = -top_bit_of (a: op0, len: op0len, prec); |
1148 | while (l1 > l0) |
1149 | { |
1150 | val[l1] = op0mask ^ op1[l1]; |
1151 | l1--; |
1152 | } |
1153 | } |
1154 | |
1155 | while (l0 >= 0) |
1156 | { |
1157 | val[l0] = op0[l0] ^ op1[l0]; |
1158 | l0--; |
1159 | } |
1160 | |
1161 | return canonize (val, len, precision: prec); |
1162 | } |
1163 | |
1164 | /* |
1165 | * math |
1166 | */ |
1167 | |
1168 | /* Set VAL to OP0 + OP1. If OVERFLOW is nonnull, record in *OVERFLOW |
1169 | whether the result overflows when OP0 and OP1 are treated as having |
1170 | signedness SGN. Return the number of blocks in VAL. */ |
1171 | unsigned int |
1172 | wi::add_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *op0, |
1173 | unsigned int op0len, const HOST_WIDE_INT *op1, |
1174 | unsigned int op1len, unsigned int prec, |
1175 | signop sgn, wi::overflow_type *overflow) |
1176 | { |
1177 | unsigned HOST_WIDE_INT o0 = 0; |
1178 | unsigned HOST_WIDE_INT o1 = 0; |
1179 | unsigned HOST_WIDE_INT x = 0; |
1180 | unsigned HOST_WIDE_INT carry = 0; |
1181 | unsigned HOST_WIDE_INT old_carry = 0; |
1182 | unsigned HOST_WIDE_INT mask0, mask1; |
1183 | unsigned int i; |
1184 | |
1185 | unsigned int len = MAX (op0len, op1len); |
1186 | mask0 = -top_bit_of (a: op0, len: op0len, prec); |
1187 | mask1 = -top_bit_of (a: op1, len: op1len, prec); |
1188 | /* Add all of the explicitly defined elements. */ |
1189 | |
1190 | for (i = 0; i < len; i++) |
1191 | { |
1192 | o0 = i < op0len ? (unsigned HOST_WIDE_INT) op0[i] : mask0; |
1193 | o1 = i < op1len ? (unsigned HOST_WIDE_INT) op1[i] : mask1; |
1194 | x = o0 + o1 + carry; |
1195 | val[i] = x; |
1196 | old_carry = carry; |
1197 | carry = carry == 0 ? x < o0 : x <= o0; |
1198 | } |
1199 | |
1200 | if (len * HOST_BITS_PER_WIDE_INT < prec) |
1201 | { |
1202 | val[len] = mask0 + mask1 + carry; |
1203 | len++; |
1204 | if (overflow) |
1205 | *overflow |
1206 | = (sgn == UNSIGNED && carry) ? wi::OVF_OVERFLOW : wi::OVF_NONE; |
1207 | } |
1208 | else if (overflow) |
1209 | { |
1210 | unsigned int shift = -prec % HOST_BITS_PER_WIDE_INT; |
1211 | if (sgn == SIGNED) |
1212 | { |
1213 | unsigned HOST_WIDE_INT x = (val[len - 1] ^ o0) & (val[len - 1] ^ o1); |
1214 | if ((HOST_WIDE_INT) (x << shift) < 0) |
1215 | { |
1216 | if (o0 > (unsigned HOST_WIDE_INT) val[len - 1]) |
1217 | *overflow = wi::OVF_UNDERFLOW; |
1218 | else if (o0 < (unsigned HOST_WIDE_INT) val[len - 1]) |
1219 | *overflow = wi::OVF_OVERFLOW; |
1220 | else |
1221 | *overflow = wi::OVF_NONE; |
1222 | } |
1223 | else |
1224 | *overflow = wi::OVF_NONE; |
1225 | } |
1226 | else |
1227 | { |
1228 | /* Put the MSB of X and O0 and in the top of the HWI. */ |
1229 | x <<= shift; |
1230 | o0 <<= shift; |
1231 | if (old_carry) |
1232 | *overflow = (x <= o0) ? wi::OVF_OVERFLOW : wi::OVF_NONE; |
1233 | else |
1234 | *overflow = (x < o0) ? wi::OVF_OVERFLOW : wi::OVF_NONE; |
1235 | } |
1236 | } |
1237 | |
1238 | return canonize (val, len, precision: prec); |
1239 | } |
1240 | |
1241 | /* Subroutines of the multiplication and division operations. Unpack |
1242 | the first IN_LEN HOST_WIDE_INTs in INPUT into 2 * IN_LEN |
1243 | HOST_HALF_WIDE_INTs of RESULT. The rest of RESULT is filled by |
1244 | uncompressing the top bit of INPUT[IN_LEN - 1]. */ |
1245 | static void |
1246 | wi_unpack (unsigned HOST_HALF_WIDE_INT *result, const HOST_WIDE_INT *input, |
1247 | unsigned int in_len, unsigned int out_len, |
1248 | unsigned int prec, signop sgn) |
1249 | { |
1250 | unsigned int i; |
1251 | unsigned int j = 0; |
1252 | unsigned int small_prec = prec & (HOST_BITS_PER_WIDE_INT - 1); |
1253 | unsigned int blocks_needed = BLOCKS_NEEDED (prec); |
1254 | HOST_WIDE_INT mask; |
1255 | |
1256 | if (sgn == SIGNED) |
1257 | { |
1258 | mask = -top_bit_of (a: (const HOST_WIDE_INT *) input, len: in_len, prec); |
1259 | mask &= HALF_INT_MASK; |
1260 | } |
1261 | else |
1262 | mask = 0; |
1263 | |
1264 | for (i = 0; i < blocks_needed - 1; i++) |
1265 | { |
1266 | HOST_WIDE_INT x = safe_uhwi (val: input, len: in_len, i); |
1267 | result[j++] = x; |
1268 | result[j++] = x >> HOST_BITS_PER_HALF_WIDE_INT; |
1269 | } |
1270 | |
1271 | HOST_WIDE_INT x = safe_uhwi (val: input, len: in_len, i); |
1272 | if (small_prec) |
1273 | { |
1274 | if (sgn == SIGNED) |
1275 | x = sext_hwi (src: x, prec: small_prec); |
1276 | else |
1277 | x = zext_hwi (src: x, prec: small_prec); |
1278 | } |
1279 | result[j++] = x; |
1280 | result[j++] = x >> HOST_BITS_PER_HALF_WIDE_INT; |
1281 | |
1282 | /* Smear the sign bit. */ |
1283 | while (j < out_len) |
1284 | result[j++] = mask; |
1285 | } |
1286 | |
1287 | /* The inverse of wi_unpack. IN_LEN is the number of input |
1288 | blocks and PRECISION is the precision of the result. Return the |
1289 | number of blocks in the canonicalized result. */ |
1290 | static unsigned int |
1291 | wi_pack (HOST_WIDE_INT *result, |
1292 | const unsigned HOST_HALF_WIDE_INT *input, |
1293 | unsigned int in_len, unsigned int precision) |
1294 | { |
1295 | unsigned int i = 0; |
1296 | unsigned int j = 0; |
1297 | unsigned int blocks_needed = BLOCKS_NEEDED (precision); |
1298 | |
1299 | while (i + 1 < in_len) |
1300 | { |
1301 | result[j++] = ((unsigned HOST_WIDE_INT) input[i] |
1302 | | ((unsigned HOST_WIDE_INT) input[i + 1] |
1303 | << HOST_BITS_PER_HALF_WIDE_INT)); |
1304 | i += 2; |
1305 | } |
1306 | |
1307 | /* Handle the case where in_len is odd. For this we zero extend. */ |
1308 | if (in_len & 1) |
1309 | result[j++] = (unsigned HOST_WIDE_INT) input[i]; |
1310 | else if (j < blocks_needed) |
1311 | result[j++] = 0; |
1312 | return canonize (val: result, len: j, precision); |
1313 | } |
1314 | |
1315 | /* Multiply Op1 by Op2. If HIGH is set, only the upper half of the |
1316 | result is returned. |
1317 | |
1318 | If HIGH is not set, throw away the upper half after the check is |
1319 | made to see if it overflows. Unfortunately there is no better way |
1320 | to check for overflow than to do this. If OVERFLOW is nonnull, |
1321 | record in *OVERFLOW whether the result overflowed. SGN controls |
1322 | the signedness and is used to check overflow or if HIGH is set. |
1323 | |
1324 | NOTE: Overflow type for signed overflow is not yet implemented. */ |
1325 | unsigned int |
1326 | wi::mul_internal (HOST_WIDE_INT *val, const HOST_WIDE_INT *op1val, |
1327 | unsigned int op1len, const HOST_WIDE_INT *op2val, |
1328 | unsigned int op2len, unsigned int prec, signop sgn, |
1329 | wi::overflow_type *overflow, bool high) |
1330 | { |
1331 | unsigned HOST_WIDE_INT o0, o1, k, t; |
1332 | unsigned int i; |
1333 | unsigned int j; |
1334 | |
1335 | /* If the top level routine did not really pass in an overflow, then |
1336 | just make sure that we never attempt to set it. */ |
1337 | bool needs_overflow = (overflow != 0); |
1338 | if (needs_overflow) |
1339 | *overflow = wi::OVF_NONE; |
1340 | |
1341 | wide_int_ref op1 = wi::storage_ref (op1val, op1len, prec); |
1342 | wide_int_ref op2 = wi::storage_ref (op2val, op2len, prec); |
1343 | |
1344 | /* This is a surprisingly common case, so do it first. */ |
1345 | if (op1 == 0 || op2 == 0) |
1346 | { |
1347 | val[0] = 0; |
1348 | return 1; |
1349 | } |
1350 | |
1351 | #ifdef umul_ppmm |
1352 | if (sgn == UNSIGNED) |
1353 | { |
1354 | /* If the inputs are single HWIs and the output has room for at |
1355 | least two HWIs, we can use umul_ppmm directly. */ |
1356 | if (prec >= HOST_BITS_PER_WIDE_INT * 2 |
1357 | && wi::fits_uhwi_p (op1) |
1358 | && wi::fits_uhwi_p (op2)) |
1359 | { |
1360 | /* This case never overflows. */ |
1361 | if (high) |
1362 | { |
1363 | val[0] = 0; |
1364 | return 1; |
1365 | } |
1366 | umul_ppmm (val[1], val[0], op1.ulow (), op2.ulow ()); |
1367 | if (val[1] < 0 && prec > HOST_BITS_PER_WIDE_INT * 2) |
1368 | { |
1369 | val[2] = 0; |
1370 | return 3; |
1371 | } |
1372 | return 1 + (val[1] != 0 || val[0] < 0); |
1373 | } |
1374 | /* Likewise if the output is a full single HWI, except that the |
1375 | upper HWI of the result is only used for determining overflow. |
1376 | (We handle this case inline when overflow isn't needed.) */ |
1377 | else if (prec == HOST_BITS_PER_WIDE_INT) |
1378 | { |
1379 | unsigned HOST_WIDE_INT upper; |
1380 | umul_ppmm (upper, val[0], op1.ulow (), op2.ulow ()); |
1381 | if (needs_overflow) |
1382 | /* Unsigned overflow can only be +OVERFLOW. */ |
1383 | *overflow = (upper != 0) ? wi::OVF_OVERFLOW : wi::OVF_NONE; |
1384 | if (high) |
1385 | val[0] = upper; |
1386 | return 1; |
1387 | } |
1388 | } |
1389 | #endif |
1390 | |
1391 | /* Handle multiplications by 1. */ |
1392 | if (op1 == 1) |
1393 | { |
1394 | if (high) |
1395 | { |
1396 | val[0] = wi::neg_p (x: op2, sgn) ? -1 : 0; |
1397 | return 1; |
1398 | } |
1399 | for (i = 0; i < op2len; i++) |
1400 | val[i] = op2val[i]; |
1401 | return op2len; |
1402 | } |
1403 | if (op2 == 1) |
1404 | { |
1405 | if (high) |
1406 | { |
1407 | val[0] = wi::neg_p (x: op1, sgn) ? -1 : 0; |
1408 | return 1; |
1409 | } |
1410 | for (i = 0; i < op1len; i++) |
1411 | val[i] = op1val[i]; |
1412 | return op1len; |
1413 | } |
1414 | |
1415 | /* If we need to check for overflow, we can only do half wide |
1416 | multiplies quickly because we need to look at the top bits to |
1417 | check for the overflow. */ |
1418 | if ((high || needs_overflow) |
1419 | && (prec <= HOST_BITS_PER_HALF_WIDE_INT)) |
1420 | { |
1421 | unsigned HOST_WIDE_INT r; |
1422 | |
1423 | if (sgn == SIGNED) |
1424 | { |
1425 | o0 = op1.to_shwi (); |
1426 | o1 = op2.to_shwi (); |
1427 | } |
1428 | else |
1429 | { |
1430 | o0 = op1.to_uhwi (); |
1431 | o1 = op2.to_uhwi (); |
1432 | } |
1433 | |
1434 | r = o0 * o1; |
1435 | if (needs_overflow) |
1436 | { |
1437 | if (sgn == SIGNED) |
1438 | { |
1439 | if ((HOST_WIDE_INT) r != sext_hwi (src: r, prec)) |
1440 | /* FIXME: Signed overflow type is not implemented yet. */ |
1441 | *overflow = OVF_UNKNOWN; |
1442 | } |
1443 | else |
1444 | { |
1445 | if ((r >> prec) != 0) |
1446 | /* Unsigned overflow can only be +OVERFLOW. */ |
1447 | *overflow = OVF_OVERFLOW; |
1448 | } |
1449 | } |
1450 | val[0] = high ? r >> prec : r; |
1451 | return 1; |
1452 | } |
1453 | |
1454 | /* The sizes here are scaled to support a 2x WIDE_INT_MAX_INL_PRECISION by 2x |
1455 | WIDE_INT_MAX_INL_PRECISION yielding a 4x WIDE_INT_MAX_INL_PRECISION |
1456 | result. */ |
1457 | |
1458 | unsigned HOST_HALF_WIDE_INT |
1459 | ubuf[4 * WIDE_INT_MAX_INL_PRECISION / HOST_BITS_PER_HALF_WIDE_INT]; |
1460 | unsigned HOST_HALF_WIDE_INT |
1461 | vbuf[4 * WIDE_INT_MAX_INL_PRECISION / HOST_BITS_PER_HALF_WIDE_INT]; |
1462 | /* The '2' in 'R' is because we are internally doing a full |
1463 | multiply. */ |
1464 | unsigned HOST_HALF_WIDE_INT |
1465 | rbuf[2 * 4 * WIDE_INT_MAX_INL_PRECISION / HOST_BITS_PER_HALF_WIDE_INT]; |
1466 | const HOST_WIDE_INT mask |
1467 | = (HOST_WIDE_INT_1 << HOST_BITS_PER_HALF_WIDE_INT) - 1; |
1468 | unsigned HOST_HALF_WIDE_INT *u = ubuf; |
1469 | unsigned HOST_HALF_WIDE_INT *v = vbuf; |
1470 | unsigned HOST_HALF_WIDE_INT *r = rbuf; |
1471 | |
1472 | if (!high) |
1473 | prec = MIN ((op1len + op2len + 1) * HOST_BITS_PER_WIDE_INT, prec); |
1474 | unsigned int blocks_needed = BLOCKS_NEEDED (prec); |
1475 | unsigned int half_blocks_needed = blocks_needed * 2; |
1476 | if (UNLIKELY (prec > WIDE_INT_MAX_INL_PRECISION)) |
1477 | { |
1478 | unsigned HOST_HALF_WIDE_INT *buf |
1479 | = XALLOCAVEC (unsigned HOST_HALF_WIDE_INT, 4 * half_blocks_needed); |
1480 | u = buf; |
1481 | v = u + half_blocks_needed; |
1482 | r = v + half_blocks_needed; |
1483 | } |
1484 | |
1485 | /* We do unsigned mul and then correct it. */ |
1486 | wi_unpack (result: u, input: op1val, in_len: op1len, out_len: half_blocks_needed, prec, sgn: UNSIGNED); |
1487 | wi_unpack (result: v, input: op2val, in_len: op2len, out_len: half_blocks_needed, prec, sgn: UNSIGNED); |
1488 | |
1489 | /* The 2 is for a full mult. */ |
1490 | memset (s: r, c: 0, n: half_blocks_needed * 2 |
1491 | * HOST_BITS_PER_HALF_WIDE_INT / CHAR_BIT); |
1492 | |
1493 | for (j = 0; j < half_blocks_needed; j++) |
1494 | { |
1495 | k = 0; |
1496 | for (i = 0; i < half_blocks_needed; i++) |
1497 | { |
1498 | t = ((unsigned HOST_WIDE_INT)u[i] * (unsigned HOST_WIDE_INT)v[j] |
1499 | + r[i + j] + k); |
1500 | r[i + j] = t & HALF_INT_MASK; |
1501 | k = t >> HOST_BITS_PER_HALF_WIDE_INT; |
1502 | } |
1503 | r[j + half_blocks_needed] = k; |
1504 | } |
1505 | |
1506 | unsigned int shift; |
1507 | if ((high || needs_overflow) && (shift = prec % HOST_BITS_PER_WIDE_INT) != 0) |
1508 | { |
1509 | /* The high or needs_overflow code assumes that the high bits |
1510 | only appear from r[half_blocks_needed] up to |
1511 | r[half_blocks_needed * 2 - 1]. If prec is not a multiple |
1512 | of HOST_BITS_PER_WIDE_INT, shift the bits above prec up |
1513 | to make that code simple. */ |
1514 | if (shift == HOST_BITS_PER_HALF_WIDE_INT) |
1515 | memmove (dest: &r[half_blocks_needed], src: &r[half_blocks_needed - 1], |
1516 | n: sizeof (r[0]) * half_blocks_needed); |
1517 | else |
1518 | { |
1519 | unsigned int skip = shift < HOST_BITS_PER_HALF_WIDE_INT; |
1520 | if (!skip) |
1521 | shift -= HOST_BITS_PER_HALF_WIDE_INT; |
1522 | for (i = 2 * half_blocks_needed - 1; i >= half_blocks_needed; i--) |
1523 | r[i] = ((r[i - skip] << (-shift % HOST_BITS_PER_HALF_WIDE_INT)) |
1524 | | (r[i - skip - 1] >> shift)); |
1525 | } |
1526 | } |
1527 | |
1528 | /* We did unsigned math above. For signed we must adjust the |
1529 | product (assuming we need to see that). */ |
1530 | if (sgn == SIGNED && (high || needs_overflow)) |
1531 | { |
1532 | unsigned HOST_WIDE_INT b; |
1533 | if (wi::neg_p (x: op1)) |
1534 | { |
1535 | b = 0; |
1536 | for (i = 0; i < half_blocks_needed; i++) |
1537 | { |
1538 | t = (unsigned HOST_WIDE_INT)r[i + half_blocks_needed] |
1539 | - (unsigned HOST_WIDE_INT)v[i] - b; |
1540 | r[i + half_blocks_needed] = t & HALF_INT_MASK; |
1541 | b = t >> (HOST_BITS_PER_WIDE_INT - 1); |
1542 | } |
1543 | } |
1544 | if (wi::neg_p (x: op2)) |
1545 | { |
1546 | b = 0; |
1547 | for (i = 0; i < half_blocks_needed; i++) |
1548 | { |
1549 | t = (unsigned HOST_WIDE_INT)r[i + half_blocks_needed] |
1550 | - (unsigned HOST_WIDE_INT)u[i] - b; |
1551 | r[i + half_blocks_needed] = t & HALF_INT_MASK; |
1552 | b = t >> (HOST_BITS_PER_WIDE_INT - 1); |
1553 | } |
1554 | } |
1555 | } |
1556 | |
1557 | if (needs_overflow) |
1558 | { |
1559 | HOST_WIDE_INT top; |
1560 | |
1561 | /* For unsigned, overflow is true if any of the top bits are set. |
1562 | For signed, overflow is true if any of the top bits are not equal |
1563 | to the sign bit. */ |
1564 | if (sgn == UNSIGNED) |
1565 | top = 0; |
1566 | else |
1567 | { |
1568 | top = r[half_blocks_needed - 1 |
1569 | - ((-prec % HOST_BITS_PER_WIDE_INT) |
1570 | >= HOST_BITS_PER_HALF_WIDE_INT)]; |
1571 | top = SIGN_MASK (((unsigned HOST_WIDE_INT) top) |
1572 | << (HOST_BITS_PER_WIDE_INT / 2 |
1573 | + (-prec % HOST_BITS_PER_HALF_WIDE_INT))); |
1574 | top &= mask; |
1575 | } |
1576 | |
1577 | for (i = half_blocks_needed; i < half_blocks_needed * 2; i++) |
1578 | if (((HOST_WIDE_INT)(r[i] & mask)) != top) |
1579 | /* FIXME: Signed overflow type is not implemented yet. */ |
1580 | *overflow = (sgn == UNSIGNED) ? wi::OVF_OVERFLOW : wi::OVF_UNKNOWN; |
1581 | } |
1582 | |
1583 | int r_offset = high ? half_blocks_needed : 0; |
1584 | return wi_pack (result: val, input: &r[r_offset], in_len: half_blocks_needed, precision: prec); |
1585 | } |
1586 | |
1587 | /* Compute the population count of X. */ |
1588 | int |
1589 | wi::popcount (const wide_int_ref &x) |
1590 | { |
1591 | unsigned int i; |
1592 | int count; |
1593 | |
1594 | /* The high order block is special if it is the last block and the |
1595 | precision is not an even multiple of HOST_BITS_PER_WIDE_INT. We |
1596 | have to clear out any ones above the precision before doing |
1597 | popcount on this block. */ |
1598 | count = x.precision - x.len * HOST_BITS_PER_WIDE_INT; |
1599 | unsigned int stop = x.len; |
1600 | if (count < 0) |
1601 | { |
1602 | count = popcount_hwi (x: x.uhigh () << -count); |
1603 | stop -= 1; |
1604 | } |
1605 | else |
1606 | { |
1607 | if (x.sign_mask () >= 0) |
1608 | count = 0; |
1609 | } |
1610 | |
1611 | for (i = 0; i < stop; ++i) |
1612 | count += popcount_hwi (x: x.val[i]); |
1613 | |
1614 | return count; |
1615 | } |
1616 | |
1617 | /* Set VAL to OP0 - OP1. If OVERFLOW is nonnull, record in *OVERFLOW |
1618 | whether the result overflows when OP0 and OP1 are treated as having |
1619 | signedness SGN. Return the number of blocks in VAL. */ |
1620 | unsigned int |
1621 | wi::sub_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *op0, |
1622 | unsigned int op0len, const HOST_WIDE_INT *op1, |
1623 | unsigned int op1len, unsigned int prec, |
1624 | signop sgn, wi::overflow_type *overflow) |
1625 | { |
1626 | unsigned HOST_WIDE_INT o0 = 0; |
1627 | unsigned HOST_WIDE_INT o1 = 0; |
1628 | unsigned HOST_WIDE_INT x = 0; |
1629 | /* We implement subtraction as an in place negate and add. Negation |
1630 | is just inversion and add 1, so we can do the add of 1 by just |
1631 | starting the borrow in of the first element at 1. */ |
1632 | unsigned HOST_WIDE_INT borrow = 0; |
1633 | unsigned HOST_WIDE_INT old_borrow = 0; |
1634 | |
1635 | unsigned HOST_WIDE_INT mask0, mask1; |
1636 | unsigned int i; |
1637 | |
1638 | unsigned int len = MAX (op0len, op1len); |
1639 | mask0 = -top_bit_of (a: op0, len: op0len, prec); |
1640 | mask1 = -top_bit_of (a: op1, len: op1len, prec); |
1641 | |
1642 | /* Subtract all of the explicitly defined elements. */ |
1643 | for (i = 0; i < len; i++) |
1644 | { |
1645 | o0 = i < op0len ? (unsigned HOST_WIDE_INT)op0[i] : mask0; |
1646 | o1 = i < op1len ? (unsigned HOST_WIDE_INT)op1[i] : mask1; |
1647 | x = o0 - o1 - borrow; |
1648 | val[i] = x; |
1649 | old_borrow = borrow; |
1650 | borrow = borrow == 0 ? o0 < o1 : o0 <= o1; |
1651 | } |
1652 | |
1653 | if (len * HOST_BITS_PER_WIDE_INT < prec) |
1654 | { |
1655 | val[len] = mask0 - mask1 - borrow; |
1656 | len++; |
1657 | if (overflow) |
1658 | *overflow = (sgn == UNSIGNED && borrow) ? OVF_UNDERFLOW : OVF_NONE; |
1659 | } |
1660 | else if (overflow) |
1661 | { |
1662 | unsigned int shift = -prec % HOST_BITS_PER_WIDE_INT; |
1663 | if (sgn == SIGNED) |
1664 | { |
1665 | unsigned HOST_WIDE_INT x = (o0 ^ o1) & (val[len - 1] ^ o0); |
1666 | if ((HOST_WIDE_INT) (x << shift) < 0) |
1667 | { |
1668 | if (o0 > o1) |
1669 | *overflow = OVF_UNDERFLOW; |
1670 | else if (o0 < o1) |
1671 | *overflow = OVF_OVERFLOW; |
1672 | else |
1673 | *overflow = OVF_NONE; |
1674 | } |
1675 | else |
1676 | *overflow = OVF_NONE; |
1677 | } |
1678 | else |
1679 | { |
1680 | /* Put the MSB of X and O0 and in the top of the HWI. */ |
1681 | x <<= shift; |
1682 | o0 <<= shift; |
1683 | if (old_borrow) |
1684 | *overflow = (x >= o0) ? OVF_UNDERFLOW : OVF_NONE; |
1685 | else |
1686 | *overflow = (x > o0) ? OVF_UNDERFLOW : OVF_NONE; |
1687 | } |
1688 | } |
1689 | |
1690 | return canonize (val, len, precision: prec); |
1691 | } |
1692 | |
1693 | |
1694 | /* |
1695 | * Division and Mod |
1696 | */ |
1697 | |
1698 | /* Compute B_QUOTIENT and B_REMAINDER from B_DIVIDEND/B_DIVISOR. The |
1699 | algorithm is a small modification of the algorithm in Hacker's |
1700 | Delight by Warren, which itself is a small modification of Knuth's |
1701 | algorithm. M is the number of significant elements of U however |
1702 | there needs to be at least one extra element of B_DIVIDEND |
1703 | allocated, N is the number of elements of B_DIVISOR. |
1704 | Return new value for N. */ |
1705 | static int |
1706 | divmod_internal_2 (unsigned HOST_HALF_WIDE_INT *b_quotient, |
1707 | unsigned HOST_HALF_WIDE_INT *b_remainder, |
1708 | unsigned HOST_HALF_WIDE_INT *b_dividend, |
1709 | unsigned HOST_HALF_WIDE_INT *b_divisor, |
1710 | int m, int n) |
1711 | { |
1712 | /* The "digits" are a HOST_HALF_WIDE_INT which the size of half of a |
1713 | HOST_WIDE_INT and stored in the lower bits of each word. This |
1714 | algorithm should work properly on both 32 and 64 bit |
1715 | machines. */ |
1716 | unsigned HOST_WIDE_INT b = HOST_WIDE_INT_1U << HOST_BITS_PER_HALF_WIDE_INT; |
1717 | unsigned HOST_WIDE_INT qhat; /* Estimate of quotient digit. */ |
1718 | unsigned HOST_WIDE_INT rhat; /* A remainder. */ |
1719 | unsigned HOST_WIDE_INT p; /* Product of two digits. */ |
1720 | HOST_WIDE_INT t, k; |
1721 | int i, j, s; |
1722 | |
1723 | /* Single digit divisor. */ |
1724 | if (n == 1) |
1725 | { |
1726 | k = 0; |
1727 | for (j = m - 1; j >= 0; j--) |
1728 | { |
1729 | b_quotient[j] = (k * b + b_dividend[j])/b_divisor[0]; |
1730 | k = ((k * b + b_dividend[j]) |
1731 | - ((unsigned HOST_WIDE_INT)b_quotient[j] |
1732 | * (unsigned HOST_WIDE_INT)b_divisor[0])); |
1733 | } |
1734 | b_remainder[0] = k; |
1735 | return 1; |
1736 | } |
1737 | |
1738 | s = clz_hwi (x: b_divisor[n-1]) - HOST_BITS_PER_HALF_WIDE_INT; /* CHECK clz */ |
1739 | |
1740 | if (s) |
1741 | { |
1742 | /* Normalize B_DIVIDEND and B_DIVISOR. Unlike the published |
1743 | algorithm, we can overwrite b_dividend and b_divisor, so we do |
1744 | that. */ |
1745 | for (i = n - 1; i > 0; i--) |
1746 | b_divisor[i] = (b_divisor[i] << s) |
1747 | | (b_divisor[i-1] >> (HOST_BITS_PER_HALF_WIDE_INT - s)); |
1748 | b_divisor[0] = b_divisor[0] << s; |
1749 | |
1750 | b_dividend[m] = b_dividend[m-1] >> (HOST_BITS_PER_HALF_WIDE_INT - s); |
1751 | for (i = m - 1; i > 0; i--) |
1752 | b_dividend[i] = (b_dividend[i] << s) |
1753 | | (b_dividend[i-1] >> (HOST_BITS_PER_HALF_WIDE_INT - s)); |
1754 | b_dividend[0] = b_dividend[0] << s; |
1755 | } |
1756 | |
1757 | /* Main loop. */ |
1758 | for (j = m - n; j >= 0; j--) |
1759 | { |
1760 | qhat = (b_dividend[j+n] * b + b_dividend[j+n-1]) / b_divisor[n-1]; |
1761 | rhat = (b_dividend[j+n] * b + b_dividend[j+n-1]) - qhat * b_divisor[n-1]; |
1762 | again: |
1763 | if (qhat >= b || qhat * b_divisor[n-2] > b * rhat + b_dividend[j+n-2]) |
1764 | { |
1765 | qhat -= 1; |
1766 | rhat += b_divisor[n-1]; |
1767 | if (rhat < b) |
1768 | goto again; |
1769 | } |
1770 | |
1771 | /* Multiply and subtract. */ |
1772 | k = 0; |
1773 | for (i = 0; i < n; i++) |
1774 | { |
1775 | p = qhat * b_divisor[i]; |
1776 | t = b_dividend[i+j] - k - (p & HALF_INT_MASK); |
1777 | b_dividend[i + j] = t; |
1778 | k = ((p >> HOST_BITS_PER_HALF_WIDE_INT) |
1779 | - (t >> HOST_BITS_PER_HALF_WIDE_INT)); |
1780 | } |
1781 | t = b_dividend[j+n] - k; |
1782 | b_dividend[j+n] = t; |
1783 | |
1784 | b_quotient[j] = qhat; |
1785 | if (t < 0) |
1786 | { |
1787 | b_quotient[j] -= 1; |
1788 | k = 0; |
1789 | for (i = 0; i < n; i++) |
1790 | { |
1791 | t = (HOST_WIDE_INT)b_dividend[i+j] + b_divisor[i] + k; |
1792 | b_dividend[i+j] = t; |
1793 | k = t >> HOST_BITS_PER_HALF_WIDE_INT; |
1794 | } |
1795 | b_dividend[j+n] += k; |
1796 | } |
1797 | } |
1798 | /* If N > M, the main loop was skipped, quotient will be 0 and |
1799 | we can't copy more than M half-limbs into the remainder, as they |
1800 | aren't present in b_dividend (which has . */ |
1801 | n = MIN (n, m); |
1802 | if (s) |
1803 | for (i = 0; i < n; i++) |
1804 | b_remainder[i] = (b_dividend[i] >> s) |
1805 | | (b_dividend[i+1] << (HOST_BITS_PER_HALF_WIDE_INT - s)); |
1806 | else |
1807 | for (i = 0; i < n; i++) |
1808 | b_remainder[i] = b_dividend[i]; |
1809 | return n; |
1810 | } |
1811 | |
1812 | |
1813 | /* Divide DIVIDEND by DIVISOR, which have signedness SGN, and truncate |
1814 | the result. If QUOTIENT is nonnull, store the value of the quotient |
1815 | there and return the number of blocks in it. The return value is |
1816 | not defined otherwise. If REMAINDER is nonnull, store the value |
1817 | of the remainder there and store the number of blocks in |
1818 | *REMAINDER_LEN. If OFLOW is not null, store in *OFLOW whether |
1819 | the division overflowed. */ |
1820 | unsigned int |
1821 | wi::divmod_internal (HOST_WIDE_INT *quotient, unsigned int *remainder_len, |
1822 | HOST_WIDE_INT *remainder, |
1823 | const HOST_WIDE_INT *dividend_val, |
1824 | unsigned int dividend_len, unsigned int dividend_prec, |
1825 | const HOST_WIDE_INT *divisor_val, unsigned int divisor_len, |
1826 | unsigned int divisor_prec, signop sgn, |
1827 | wi::overflow_type *oflow) |
1828 | { |
1829 | unsigned int m, n; |
1830 | bool dividend_neg = false; |
1831 | bool divisor_neg = false; |
1832 | bool overflow = false; |
1833 | wide_int neg_dividend, neg_divisor; |
1834 | |
1835 | wide_int_ref dividend = wi::storage_ref (dividend_val, dividend_len, |
1836 | dividend_prec); |
1837 | wide_int_ref divisor = wi::storage_ref (divisor_val, divisor_len, |
1838 | divisor_prec); |
1839 | if (divisor == 0) |
1840 | overflow = true; |
1841 | |
1842 | /* The smallest signed number / -1 causes overflow. The dividend_len |
1843 | check is for speed rather than correctness. */ |
1844 | if (sgn == SIGNED |
1845 | && dividend_len == BLOCKS_NEEDED (dividend_prec) |
1846 | && divisor == -1 |
1847 | && wi::only_sign_bit_p (dividend)) |
1848 | overflow = true; |
1849 | |
1850 | /* Handle the overflow cases. Viewed as unsigned value, the quotient of |
1851 | (signed min / -1) has the same representation as the orignal dividend. |
1852 | We have traditionally made division by zero act as division by one, |
1853 | so there too we use the original dividend. */ |
1854 | if (overflow) |
1855 | { |
1856 | if (remainder) |
1857 | { |
1858 | *remainder_len = 1; |
1859 | remainder[0] = 0; |
1860 | } |
1861 | if (oflow) |
1862 | *oflow = OVF_OVERFLOW; |
1863 | if (quotient) |
1864 | for (unsigned int i = 0; i < dividend_len; ++i) |
1865 | quotient[i] = dividend_val[i]; |
1866 | return dividend_len; |
1867 | } |
1868 | |
1869 | if (oflow) |
1870 | *oflow = OVF_NONE; |
1871 | |
1872 | /* Do it on the host if you can. */ |
1873 | if (sgn == SIGNED |
1874 | && wi::fits_shwi_p (x: dividend) |
1875 | && wi::fits_shwi_p (x: divisor)) |
1876 | { |
1877 | HOST_WIDE_INT o0 = dividend.to_shwi (); |
1878 | HOST_WIDE_INT o1 = divisor.to_shwi (); |
1879 | |
1880 | if (o0 == HOST_WIDE_INT_MIN && o1 == -1) |
1881 | { |
1882 | gcc_checking_assert (dividend_prec > HOST_BITS_PER_WIDE_INT); |
1883 | if (quotient) |
1884 | { |
1885 | quotient[0] = HOST_WIDE_INT_MIN; |
1886 | quotient[1] = 0; |
1887 | } |
1888 | if (remainder) |
1889 | { |
1890 | remainder[0] = 0; |
1891 | *remainder_len = 1; |
1892 | } |
1893 | return 2; |
1894 | } |
1895 | else |
1896 | { |
1897 | if (quotient) |
1898 | quotient[0] = o0 / o1; |
1899 | if (remainder) |
1900 | { |
1901 | remainder[0] = o0 % o1; |
1902 | *remainder_len = 1; |
1903 | } |
1904 | return 1; |
1905 | } |
1906 | } |
1907 | |
1908 | if (sgn == UNSIGNED |
1909 | && wi::fits_uhwi_p (x: dividend) |
1910 | && wi::fits_uhwi_p (x: divisor)) |
1911 | { |
1912 | unsigned HOST_WIDE_INT o0 = dividend.to_uhwi (); |
1913 | unsigned HOST_WIDE_INT o1 = divisor.to_uhwi (); |
1914 | unsigned int quotient_len = 1; |
1915 | |
1916 | if (quotient) |
1917 | { |
1918 | quotient[0] = o0 / o1; |
1919 | quotient_len = canonize_uhwi (val: quotient, precision: dividend_prec); |
1920 | } |
1921 | if (remainder) |
1922 | { |
1923 | remainder[0] = o0 % o1; |
1924 | *remainder_len = canonize_uhwi (val: remainder, precision: dividend_prec); |
1925 | } |
1926 | return quotient_len; |
1927 | } |
1928 | |
1929 | /* Make the divisor and dividend positive and remember what we |
1930 | did. */ |
1931 | if (sgn == SIGNED) |
1932 | { |
1933 | if (wi::neg_p (x: dividend)) |
1934 | { |
1935 | neg_dividend = -dividend; |
1936 | dividend = neg_dividend; |
1937 | dividend_neg = true; |
1938 | } |
1939 | if (wi::neg_p (x: divisor)) |
1940 | { |
1941 | neg_divisor = -divisor; |
1942 | divisor = neg_divisor; |
1943 | divisor_neg = true; |
1944 | } |
1945 | } |
1946 | |
1947 | unsigned HOST_HALF_WIDE_INT |
1948 | b_quotient_buf[4 * WIDE_INT_MAX_INL_PRECISION |
1949 | / HOST_BITS_PER_HALF_WIDE_INT]; |
1950 | unsigned HOST_HALF_WIDE_INT |
1951 | b_remainder_buf[4 * WIDE_INT_MAX_INL_PRECISION |
1952 | / HOST_BITS_PER_HALF_WIDE_INT]; |
1953 | unsigned HOST_HALF_WIDE_INT |
1954 | b_dividend_buf[(4 * WIDE_INT_MAX_INL_PRECISION |
1955 | / HOST_BITS_PER_HALF_WIDE_INT) + 1]; |
1956 | unsigned HOST_HALF_WIDE_INT |
1957 | b_divisor_buf[4 * WIDE_INT_MAX_INL_PRECISION |
1958 | / HOST_BITS_PER_HALF_WIDE_INT]; |
1959 | unsigned HOST_HALF_WIDE_INT *b_quotient = b_quotient_buf; |
1960 | unsigned HOST_HALF_WIDE_INT *b_remainder = b_remainder_buf; |
1961 | unsigned HOST_HALF_WIDE_INT *b_dividend = b_dividend_buf; |
1962 | unsigned HOST_HALF_WIDE_INT *b_divisor = b_divisor_buf; |
1963 | |
1964 | if (sgn == SIGNED || dividend_val[dividend_len - 1] >= 0) |
1965 | dividend_prec = MIN ((dividend_len + 1) * HOST_BITS_PER_WIDE_INT, |
1966 | dividend_prec); |
1967 | if (sgn == SIGNED || divisor_val[divisor_len - 1] >= 0) |
1968 | divisor_prec = MIN (divisor_len * HOST_BITS_PER_WIDE_INT, divisor_prec); |
1969 | unsigned int dividend_blocks_needed = 2 * BLOCKS_NEEDED (dividend_prec); |
1970 | unsigned int divisor_blocks_needed = 2 * BLOCKS_NEEDED (divisor_prec); |
1971 | if (UNLIKELY (dividend_prec > WIDE_INT_MAX_INL_PRECISION) |
1972 | || UNLIKELY (divisor_prec > WIDE_INT_MAX_INL_PRECISION)) |
1973 | { |
1974 | unsigned HOST_HALF_WIDE_INT *buf |
1975 | = XALLOCAVEC (unsigned HOST_HALF_WIDE_INT, |
1976 | 3 * dividend_blocks_needed + 1 |
1977 | + divisor_blocks_needed); |
1978 | b_quotient = buf; |
1979 | b_remainder = b_quotient + dividend_blocks_needed; |
1980 | b_dividend = b_remainder + dividend_blocks_needed; |
1981 | b_divisor = b_dividend + dividend_blocks_needed + 1; |
1982 | memset (s: b_quotient, c: 0, |
1983 | n: dividend_blocks_needed * sizeof (HOST_HALF_WIDE_INT)); |
1984 | } |
1985 | wi_unpack (result: b_dividend, input: dividend.get_val (), in_len: dividend.get_len (), |
1986 | out_len: dividend_blocks_needed, prec: dividend_prec, sgn: UNSIGNED); |
1987 | wi_unpack (result: b_divisor, input: divisor.get_val (), in_len: divisor.get_len (), |
1988 | out_len: divisor_blocks_needed, prec: divisor_prec, sgn: UNSIGNED); |
1989 | |
1990 | m = dividend_blocks_needed; |
1991 | b_dividend[m] = 0; |
1992 | while (m > 1 && b_dividend[m - 1] == 0) |
1993 | m--; |
1994 | |
1995 | n = divisor_blocks_needed; |
1996 | while (n > 1 && b_divisor[n - 1] == 0) |
1997 | n--; |
1998 | |
1999 | if (b_quotient == b_quotient_buf) |
2000 | memset (s: b_quotient_buf, c: 0, n: sizeof (b_quotient_buf)); |
2001 | |
2002 | n = divmod_internal_2 (b_quotient, b_remainder, b_dividend, b_divisor, m, n); |
2003 | |
2004 | unsigned int quotient_len = 0; |
2005 | if (quotient) |
2006 | { |
2007 | quotient_len = wi_pack (result: quotient, input: b_quotient, in_len: m, precision: dividend_prec); |
2008 | /* The quotient is neg if exactly one of the divisor or dividend is |
2009 | neg. */ |
2010 | if (dividend_neg != divisor_neg) |
2011 | quotient_len = wi::sub_large (val: quotient, op0: zeros, op0len: 1, op1: quotient, |
2012 | op1len: quotient_len, prec: dividend_prec, |
2013 | sgn: UNSIGNED, overflow: 0); |
2014 | } |
2015 | |
2016 | if (remainder) |
2017 | { |
2018 | *remainder_len = wi_pack (result: remainder, input: b_remainder, in_len: n, precision: dividend_prec); |
2019 | /* The remainder is always the same sign as the dividend. */ |
2020 | if (dividend_neg) |
2021 | *remainder_len = wi::sub_large (val: remainder, op0: zeros, op0len: 1, op1: remainder, |
2022 | op1len: *remainder_len, prec: dividend_prec, |
2023 | sgn: UNSIGNED, overflow: 0); |
2024 | } |
2025 | |
2026 | return quotient_len; |
2027 | } |
2028 | |
2029 | /* |
2030 | * Shifting, rotating and extraction. |
2031 | */ |
2032 | |
2033 | /* Left shift XVAL by SHIFT and store the result in VAL. Return the |
2034 | number of blocks in VAL. Both XVAL and VAL have PRECISION bits. */ |
2035 | unsigned int |
2036 | wi::lshift_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
2037 | unsigned int xlen, unsigned int precision, |
2038 | unsigned int shift) |
2039 | { |
2040 | /* Split the shift into a whole-block shift and a subblock shift. */ |
2041 | unsigned int skip = shift / HOST_BITS_PER_WIDE_INT; |
2042 | unsigned int small_shift = shift % HOST_BITS_PER_WIDE_INT; |
2043 | |
2044 | /* The whole-block shift fills with zeros. */ |
2045 | unsigned int len = BLOCKS_NEEDED (precision); |
2046 | len = MIN (xlen + skip + 1, len); |
2047 | for (unsigned int i = 0; i < skip; ++i) |
2048 | val[i] = 0; |
2049 | |
2050 | /* It's easier to handle the simple block case specially. */ |
2051 | if (small_shift == 0) |
2052 | for (unsigned int i = skip; i < len; ++i) |
2053 | val[i] = safe_uhwi (val: xval, len: xlen, i: i - skip); |
2054 | else |
2055 | { |
2056 | /* The first unfilled output block is a left shift of the first |
2057 | block in XVAL. The other output blocks contain bits from two |
2058 | consecutive input blocks. */ |
2059 | unsigned HOST_WIDE_INT carry = 0; |
2060 | for (unsigned int i = skip; i < len; ++i) |
2061 | { |
2062 | unsigned HOST_WIDE_INT x = safe_uhwi (val: xval, len: xlen, i: i - skip); |
2063 | val[i] = (x << small_shift) | carry; |
2064 | carry = x >> (-small_shift % HOST_BITS_PER_WIDE_INT); |
2065 | } |
2066 | } |
2067 | return canonize (val, len, precision); |
2068 | } |
2069 | |
2070 | /* Right shift XVAL by SHIFT and store the result in VAL. LEN is the |
2071 | number of blocks in VAL. The input has XPRECISION bits and the |
2072 | output has XPRECISION - SHIFT bits. */ |
2073 | static void |
2074 | rshift_large_common (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
2075 | unsigned int xlen, unsigned int shift, unsigned int len) |
2076 | { |
2077 | /* Split the shift into a whole-block shift and a subblock shift. */ |
2078 | unsigned int skip = shift / HOST_BITS_PER_WIDE_INT; |
2079 | unsigned int small_shift = shift % HOST_BITS_PER_WIDE_INT; |
2080 | |
2081 | /* It's easier to handle the simple block case specially. */ |
2082 | if (small_shift == 0) |
2083 | for (unsigned int i = 0; i < len; ++i) |
2084 | val[i] = safe_uhwi (val: xval, len: xlen, i: i + skip); |
2085 | else |
2086 | { |
2087 | /* Each output block but the last is a combination of two input blocks. |
2088 | The last block is a right shift of the last block in XVAL. */ |
2089 | unsigned HOST_WIDE_INT curr = safe_uhwi (val: xval, len: xlen, i: skip); |
2090 | for (unsigned int i = 0; i < len; ++i) |
2091 | { |
2092 | val[i] = curr >> small_shift; |
2093 | curr = safe_uhwi (val: xval, len: xlen, i: i + skip + 1); |
2094 | val[i] |= curr << (-small_shift % HOST_BITS_PER_WIDE_INT); |
2095 | } |
2096 | } |
2097 | } |
2098 | |
2099 | /* Logically right shift XVAL by SHIFT and store the result in VAL. |
2100 | Return the number of blocks in VAL. XVAL has XPRECISION bits and |
2101 | VAL has PRECISION bits. */ |
2102 | unsigned int |
2103 | wi::lrshift_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
2104 | unsigned int xlen, unsigned int xprecision, |
2105 | unsigned int precision, unsigned int shift) |
2106 | { |
2107 | /* Work out how many blocks are needed to store the significant bits |
2108 | (excluding the upper zeros or signs). */ |
2109 | unsigned int blocks_needed = BLOCKS_NEEDED (xprecision - shift); |
2110 | unsigned int len = blocks_needed; |
2111 | if (len > xlen && xval[xlen - 1] >= 0) |
2112 | len = xlen; |
2113 | |
2114 | rshift_large_common (val, xval, xlen, shift, len); |
2115 | |
2116 | /* The value we just created has precision XPRECISION - SHIFT. |
2117 | Zero-extend it to wider precisions. */ |
2118 | if (precision > xprecision - shift && len == blocks_needed) |
2119 | { |
2120 | unsigned int small_prec = (xprecision - shift) % HOST_BITS_PER_WIDE_INT; |
2121 | if (small_prec) |
2122 | val[len - 1] = zext_hwi (src: val[len - 1], prec: small_prec); |
2123 | else if (val[len - 1] < 0) |
2124 | { |
2125 | /* Add a new block with a zero. */ |
2126 | val[len++] = 0; |
2127 | return len; |
2128 | } |
2129 | } |
2130 | return canonize (val, len, precision); |
2131 | } |
2132 | |
2133 | /* Arithmetically right shift XVAL by SHIFT and store the result in VAL. |
2134 | Return the number of blocks in VAL. XVAL has XPRECISION bits and |
2135 | VAL has PRECISION bits. */ |
2136 | unsigned int |
2137 | wi::arshift_large (HOST_WIDE_INT *val, const HOST_WIDE_INT *xval, |
2138 | unsigned int xlen, unsigned int xprecision, |
2139 | unsigned int precision, unsigned int shift) |
2140 | { |
2141 | /* Work out how many blocks are needed to store the significant bits |
2142 | (excluding the upper zeros or signs). */ |
2143 | unsigned int blocks_needed = BLOCKS_NEEDED (xprecision - shift); |
2144 | unsigned int len = MIN (xlen, blocks_needed); |
2145 | |
2146 | rshift_large_common (val, xval, xlen, shift, len); |
2147 | |
2148 | /* The value we just created has precision XPRECISION - SHIFT. |
2149 | Sign-extend it to wider types. */ |
2150 | if (precision > xprecision - shift && len == blocks_needed) |
2151 | { |
2152 | unsigned int small_prec = (xprecision - shift) % HOST_BITS_PER_WIDE_INT; |
2153 | if (small_prec) |
2154 | val[len - 1] = sext_hwi (src: val[len - 1], prec: small_prec); |
2155 | } |
2156 | return canonize (val, len, precision); |
2157 | } |
2158 | |
2159 | /* Return the number of leading (upper) zeros in X. */ |
2160 | int |
2161 | wi::clz (const wide_int_ref &x) |
2162 | { |
2163 | if (x.sign_mask () < 0) |
2164 | /* The upper bit is set, so there are no leading zeros. */ |
2165 | return 0; |
2166 | |
2167 | /* Calculate how many bits there above the highest represented block. */ |
2168 | int count = x.precision - x.len * HOST_BITS_PER_WIDE_INT; |
2169 | |
2170 | unsigned HOST_WIDE_INT high = x.uhigh (); |
2171 | if (count < 0) |
2172 | /* The upper -COUNT bits of HIGH are not part of the value. |
2173 | Clear them out. */ |
2174 | high = (high << -count) >> -count; |
2175 | |
2176 | /* We don't need to look below HIGH. Either HIGH is nonzero, |
2177 | or the top bit of the block below is nonzero; clz_hwi is |
2178 | HOST_BITS_PER_WIDE_INT in the latter case. */ |
2179 | return count + clz_hwi (x: high); |
2180 | } |
2181 | |
2182 | /* Return the number of redundant sign bits in X. (That is, the number |
2183 | of bits immediately below the sign bit that have the same value as |
2184 | the sign bit.) */ |
2185 | int |
2186 | wi::clrsb (const wide_int_ref &x) |
2187 | { |
2188 | /* Calculate how many bits there above the highest represented block. */ |
2189 | int count = x.precision - x.len * HOST_BITS_PER_WIDE_INT; |
2190 | |
2191 | unsigned HOST_WIDE_INT high = x.uhigh (); |
2192 | unsigned HOST_WIDE_INT mask = -1; |
2193 | if (count < 0) |
2194 | { |
2195 | /* The upper -COUNT bits of HIGH are not part of the value. |
2196 | Clear them from both MASK and HIGH. */ |
2197 | mask >>= -count; |
2198 | high &= mask; |
2199 | } |
2200 | |
2201 | /* If the top bit is 1, count the number of leading 1s. If the top |
2202 | bit is zero, count the number of leading zeros. */ |
2203 | if (high > mask / 2) |
2204 | high ^= mask; |
2205 | |
2206 | /* There are no sign bits below the top block, so we don't need to look |
2207 | beyond HIGH. Note that clz_hwi is HOST_BITS_PER_WIDE_INT when |
2208 | HIGH is 0. */ |
2209 | return count + clz_hwi (x: high) - 1; |
2210 | } |
2211 | |
2212 | /* Return the number of trailing (lower) zeros in X. */ |
2213 | int |
2214 | wi::ctz (const wide_int_ref &x) |
2215 | { |
2216 | if (x.len == 1 && x.ulow () == 0) |
2217 | return x.precision; |
2218 | |
2219 | /* Having dealt with the zero case, there must be a block with a |
2220 | nonzero bit. We don't care about the bits above the first 1. */ |
2221 | unsigned int i = 0; |
2222 | while (x.val[i] == 0) |
2223 | ++i; |
2224 | return i * HOST_BITS_PER_WIDE_INT + ctz_hwi (x: x.val[i]); |
2225 | } |
2226 | |
2227 | /* If X is an exact power of 2, return the base-2 logarithm, otherwise |
2228 | return -1. */ |
2229 | int |
2230 | wi::exact_log2 (const wide_int_ref &x) |
2231 | { |
2232 | /* Reject cases where there are implicit -1 blocks above HIGH. */ |
2233 | if (x.len * HOST_BITS_PER_WIDE_INT < x.precision && x.sign_mask () < 0) |
2234 | return -1; |
2235 | |
2236 | /* Set CRUX to the index of the entry that should be nonzero. |
2237 | If the top block is zero then the next lowest block (if any) |
2238 | must have the high bit set. */ |
2239 | unsigned int crux = x.len - 1; |
2240 | if (crux > 0 && x.val[crux] == 0) |
2241 | crux -= 1; |
2242 | |
2243 | /* Check that all lower blocks are zero. */ |
2244 | for (unsigned int i = 0; i < crux; ++i) |
2245 | if (x.val[i] != 0) |
2246 | return -1; |
2247 | |
2248 | /* Get a zero-extended form of block CRUX. */ |
2249 | unsigned HOST_WIDE_INT hwi = x.val[crux]; |
2250 | if ((crux + 1) * HOST_BITS_PER_WIDE_INT > x.precision) |
2251 | hwi = zext_hwi (src: hwi, prec: x.precision % HOST_BITS_PER_WIDE_INT); |
2252 | |
2253 | /* Now it's down to whether HWI is a power of 2. */ |
2254 | int res = ::exact_log2 (x: hwi); |
2255 | if (res >= 0) |
2256 | res += crux * HOST_BITS_PER_WIDE_INT; |
2257 | return res; |
2258 | } |
2259 | |
2260 | /* Return the base-2 logarithm of X, rounding down. Return -1 if X is 0. */ |
2261 | int |
2262 | wi::floor_log2 (const wide_int_ref &x) |
2263 | { |
2264 | return x.precision - 1 - clz (x); |
2265 | } |
2266 | |
2267 | /* Return the index of the first (lowest) set bit in X, counting from 1. |
2268 | Return 0 if X is 0. */ |
2269 | int |
2270 | wi::ffs (const wide_int_ref &x) |
2271 | { |
2272 | return eq_p (x, y: 0) ? 0 : ctz (x) + 1; |
2273 | } |
2274 | |
2275 | /* Return true if sign-extending X to have precision PRECISION would give |
2276 | the minimum signed value at that precision. */ |
2277 | bool |
2278 | wi::only_sign_bit_p (const wide_int_ref &x, unsigned int precision) |
2279 | { |
2280 | return ctz (x) + 1 == int (precision); |
2281 | } |
2282 | |
2283 | /* Return true if X represents the minimum signed value. */ |
2284 | bool |
2285 | wi::only_sign_bit_p (const wide_int_ref &x) |
2286 | { |
2287 | return only_sign_bit_p (x, precision: x.precision); |
2288 | } |
2289 | |
2290 | /* Return VAL if VAL has no bits set outside MASK. Otherwise round VAL |
2291 | down to the previous value that has no bits set outside MASK. |
2292 | This rounding wraps for signed values if VAL is negative and |
2293 | the top bit of MASK is clear. |
2294 | |
2295 | For example, round_down_for_mask (6, 0xf1) would give 1 and |
2296 | round_down_for_mask (24, 0xf1) would give 17. */ |
2297 | |
2298 | wide_int |
2299 | wi::round_down_for_mask (const wide_int &val, const wide_int &mask) |
2300 | { |
2301 | /* Get the bits in VAL that are outside the mask. */ |
2302 | wide_int = wi::bit_and_not (x: val, y: mask); |
2303 | if (extra_bits == 0) |
2304 | return val; |
2305 | |
2306 | /* Get a mask that includes the top bit in EXTRA_BITS and is all 1s |
2307 | below that bit. */ |
2308 | unsigned int precision = val.get_precision (); |
2309 | wide_int lower_mask = wi::mask (width: precision - wi::clz (x: extra_bits), |
2310 | negate_p: false, precision); |
2311 | |
2312 | /* Clear the bits that aren't in MASK, but ensure that all bits |
2313 | in MASK below the top cleared bit are set. */ |
2314 | return (val & mask) | (mask & lower_mask); |
2315 | } |
2316 | |
2317 | /* Return VAL if VAL has no bits set outside MASK. Otherwise round VAL |
2318 | up to the next value that has no bits set outside MASK. The rounding |
2319 | wraps if there are no suitable values greater than VAL. |
2320 | |
2321 | For example, round_up_for_mask (6, 0xf1) would give 16 and |
2322 | round_up_for_mask (24, 0xf1) would give 32. */ |
2323 | |
2324 | wide_int |
2325 | wi::round_up_for_mask (const wide_int &val, const wide_int &mask) |
2326 | { |
2327 | /* Get the bits in VAL that are outside the mask. */ |
2328 | wide_int = wi::bit_and_not (x: val, y: mask); |
2329 | if (extra_bits == 0) |
2330 | return val; |
2331 | |
2332 | /* Get a mask that is all 1s above the top bit in EXTRA_BITS. */ |
2333 | unsigned int precision = val.get_precision (); |
2334 | wide_int upper_mask = wi::mask (width: precision - wi::clz (x: extra_bits), |
2335 | negate_p: true, precision); |
2336 | |
2337 | /* Get the bits of the mask that are above the top bit in EXTRA_BITS. */ |
2338 | upper_mask &= mask; |
2339 | |
2340 | /* Conceptually we need to: |
2341 | |
2342 | - clear bits of VAL outside UPPER_MASK |
2343 | - add the lowest bit in UPPER_MASK to VAL (or add 0 if UPPER_MASK is 0) |
2344 | - propagate the carry through the bits of VAL in UPPER_MASK |
2345 | |
2346 | If (~VAL & UPPER_MASK) is nonzero, the carry eventually |
2347 | reaches that bit and the process leaves all lower bits clear. |
2348 | If (~VAL & UPPER_MASK) is zero then the result is also zero. */ |
2349 | wide_int tmp = wi::bit_and_not (x: upper_mask, y: val); |
2350 | |
2351 | return (val | tmp) & -tmp; |
2352 | } |
2353 | |
2354 | /* Compute the modular multiplicative inverse of A modulo B |
2355 | using extended Euclid's algorithm. Assumes A and B are coprime, |
2356 | and that A and B have the same precision. */ |
2357 | wide_int |
2358 | wi::mod_inv (const wide_int &a, const wide_int &b) |
2359 | { |
2360 | /* Verify the assumption. */ |
2361 | gcc_checking_assert (wi::eq_p (wi::gcd (a, b), 1)); |
2362 | |
2363 | unsigned int p = a.get_precision () + 1; |
2364 | gcc_checking_assert (b.get_precision () + 1 == p); |
2365 | wide_int c = wide_int::from (x: a, precision: p, sgn: UNSIGNED); |
2366 | wide_int d = wide_int::from (x: b, precision: p, sgn: UNSIGNED); |
2367 | wide_int x0 = wide_int::from (x: 0, precision: p, sgn: UNSIGNED); |
2368 | wide_int x1 = wide_int::from (x: 1, precision: p, sgn: UNSIGNED); |
2369 | |
2370 | if (wi::eq_p (x: b, y: 1)) |
2371 | return wide_int::from (x: 1, precision: p, sgn: UNSIGNED); |
2372 | |
2373 | while (wi::gt_p (x: c, y: 1, sgn: UNSIGNED)) |
2374 | { |
2375 | wide_int t = d; |
2376 | wide_int q = wi::divmod_trunc (x: c, y: d, sgn: UNSIGNED, remainder_ptr: &d); |
2377 | c = t; |
2378 | wide_int s = x0; |
2379 | x0 = wi::sub (x: x1, y: wi::mul (x: q, y: x0)); |
2380 | x1 = s; |
2381 | } |
2382 | if (wi::lt_p (x: x1, y: 0, sgn: SIGNED)) |
2383 | x1 += d; |
2384 | return x1; |
2385 | } |
2386 | |
2387 | /* |
2388 | * Private utilities. |
2389 | */ |
2390 | |
2391 | void gt_ggc_mx (widest_int *) { } |
2392 | void gt_pch_nx (widest_int *, void (*) (void *, void *), void *) { } |
2393 | void gt_pch_nx (widest_int *) { } |
2394 | |
2395 | template void wide_int::dump () const; |
2396 | template void generic_wide_int <wide_int_ref_storage <false> >::dump () const; |
2397 | template void generic_wide_int <wide_int_ref_storage <true> >::dump () const; |
2398 | template void offset_int::dump () const; |
2399 | template void widest_int::dump () const; |
2400 | |
2401 | /* We could add all the above ::dump variants here, but wide_int and |
2402 | widest_int should handle the common cases. Besides, you can always |
2403 | call the dump method directly. */ |
2404 | |
2405 | DEBUG_FUNCTION void |
2406 | debug (const wide_int &ref) |
2407 | { |
2408 | ref.dump (); |
2409 | } |
2410 | |
2411 | DEBUG_FUNCTION void |
2412 | debug (const wide_int *ptr) |
2413 | { |
2414 | if (ptr) |
2415 | debug (ref: *ptr); |
2416 | else |
2417 | fprintf (stderr, format: "<nil>\n" ); |
2418 | } |
2419 | |
2420 | DEBUG_FUNCTION void |
2421 | debug (const widest_int &ref) |
2422 | { |
2423 | ref.dump (); |
2424 | } |
2425 | |
2426 | DEBUG_FUNCTION void |
2427 | debug (const widest_int *ptr) |
2428 | { |
2429 | if (ptr) |
2430 | debug (ref: *ptr); |
2431 | else |
2432 | fprintf (stderr, format: "<nil>\n" ); |
2433 | } |
2434 | |
2435 | #if CHECKING_P |
2436 | |
2437 | namespace selftest { |
2438 | |
2439 | /* Selftests for wide ints. We run these multiple times, once per type. */ |
2440 | |
2441 | /* Helper function for building a test value. */ |
2442 | |
2443 | template <class VALUE_TYPE> |
2444 | static VALUE_TYPE |
2445 | from_int (int i); |
2446 | |
2447 | /* Specializations of the fixture for each wide-int type. */ |
2448 | |
2449 | /* Specialization for VALUE_TYPE == wide_int. */ |
2450 | |
2451 | template <> |
2452 | wide_int |
2453 | from_int (int i) |
2454 | { |
2455 | return wi::shwi (val: i, precision: 32); |
2456 | } |
2457 | |
2458 | /* Specialization for VALUE_TYPE == offset_int. */ |
2459 | |
2460 | template <> |
2461 | offset_int |
2462 | from_int (int i) |
2463 | { |
2464 | return offset_int (i); |
2465 | } |
2466 | |
2467 | /* Specialization for VALUE_TYPE == widest_int. */ |
2468 | |
2469 | template <> |
2470 | widest_int |
2471 | from_int (int i) |
2472 | { |
2473 | return widest_int (i); |
2474 | } |
2475 | |
2476 | /* Verify that print_dec (WI, ..., SGN) gives the expected string |
2477 | representation (using base 10). */ |
2478 | |
2479 | static void |
2480 | assert_deceq (const char *expected, const wide_int_ref &wi, signop sgn) |
2481 | { |
2482 | char buf[WIDE_INT_PRINT_BUFFER_SIZE], *p = buf; |
2483 | unsigned len; |
2484 | if (print_dec_buf_size (wi, sgn, len: &len)) |
2485 | p = XALLOCAVEC (char, len); |
2486 | print_dec (wi, buf: p, sgn); |
2487 | ASSERT_STREQ (expected, p); |
2488 | } |
2489 | |
2490 | /* Likewise for base 16. */ |
2491 | |
2492 | static void |
2493 | assert_hexeq (const char *expected, const wide_int_ref &wi) |
2494 | { |
2495 | char buf[WIDE_INT_PRINT_BUFFER_SIZE], *p = buf; |
2496 | unsigned len; |
2497 | if (print_hex_buf_size (wi, len: &len)) |
2498 | p = XALLOCAVEC (char, len); |
2499 | print_hex (wi, buf: p); |
2500 | ASSERT_STREQ (expected, p); |
2501 | } |
2502 | |
2503 | /* Test cases. */ |
2504 | |
2505 | /* Verify that print_dec and print_hex work for VALUE_TYPE. */ |
2506 | |
2507 | template <class VALUE_TYPE> |
2508 | static void |
2509 | test_printing () |
2510 | { |
2511 | VALUE_TYPE a = from_int<VALUE_TYPE> (42); |
2512 | assert_deceq ("42" , a, SIGNED); |
2513 | assert_hexeq ("0x2a" , a); |
2514 | assert_hexeq (expected: "0x1fffffffffffffffff" , wi: wi::shwi (val: -1, precision: 69)); |
2515 | assert_hexeq (expected: "0xffffffffffffffff" , wi: wi::mask (width: 64, negate_p: false, precision: 69)); |
2516 | assert_hexeq (expected: "0xffffffffffffffff" , wi: wi::mask <widest_int> (width: 64, negate_p: false)); |
2517 | if (WIDE_INT_MAX_INL_PRECISION > 128) |
2518 | { |
2519 | assert_hexeq (expected: "0x20000000000000000fffffffffffffffe" , |
2520 | wi: wi::lshift (x: 1, y: 129) + wi::lshift (x: 1, y: 64) - 2); |
2521 | assert_hexeq (expected: "0x200000000000004000123456789abcdef" , |
2522 | wi: wi::lshift (x: 1, y: 129) + wi::lshift (x: 1, y: 74) |
2523 | + wi::lshift (x: 0x1234567, y: 32) + 0x89abcdef); |
2524 | } |
2525 | } |
2526 | |
2527 | /* Verify that various operations work correctly for VALUE_TYPE, |
2528 | unary and binary, using both function syntax, and |
2529 | overloaded-operators. */ |
2530 | |
2531 | template <class VALUE_TYPE> |
2532 | static void |
2533 | test_ops () |
2534 | { |
2535 | VALUE_TYPE a = from_int<VALUE_TYPE> (7); |
2536 | VALUE_TYPE b = from_int<VALUE_TYPE> (3); |
2537 | |
2538 | /* Using functions. */ |
2539 | assert_deceq ("-7" , wi::neg (a), SIGNED); |
2540 | assert_deceq ("10" , wi::add (a, b), SIGNED); |
2541 | assert_deceq ("4" , wi::sub (a, b), SIGNED); |
2542 | assert_deceq ("-4" , wi::sub (b, a), SIGNED); |
2543 | assert_deceq ("21" , wi::mul (a, b), SIGNED); |
2544 | |
2545 | /* Using operators. */ |
2546 | assert_deceq ("-7" , -a, SIGNED); |
2547 | assert_deceq ("10" , a + b, SIGNED); |
2548 | assert_deceq ("4" , a - b, SIGNED); |
2549 | assert_deceq ("-4" , b - a, SIGNED); |
2550 | assert_deceq ("21" , a * b, SIGNED); |
2551 | } |
2552 | |
2553 | /* Verify that various comparisons work correctly for VALUE_TYPE. */ |
2554 | |
2555 | template <class VALUE_TYPE> |
2556 | static void |
2557 | test_comparisons () |
2558 | { |
2559 | VALUE_TYPE a = from_int<VALUE_TYPE> (7); |
2560 | VALUE_TYPE b = from_int<VALUE_TYPE> (3); |
2561 | |
2562 | /* == */ |
2563 | ASSERT_TRUE (wi::eq_p (a, a)); |
2564 | ASSERT_FALSE (wi::eq_p (a, b)); |
2565 | |
2566 | /* != */ |
2567 | ASSERT_TRUE (wi::ne_p (a, b)); |
2568 | ASSERT_FALSE (wi::ne_p (a, a)); |
2569 | |
2570 | /* < */ |
2571 | ASSERT_FALSE (wi::lts_p (a, a)); |
2572 | ASSERT_FALSE (wi::lts_p (a, b)); |
2573 | ASSERT_TRUE (wi::lts_p (b, a)); |
2574 | |
2575 | /* <= */ |
2576 | ASSERT_TRUE (wi::les_p (a, a)); |
2577 | ASSERT_FALSE (wi::les_p (a, b)); |
2578 | ASSERT_TRUE (wi::les_p (b, a)); |
2579 | |
2580 | /* > */ |
2581 | ASSERT_FALSE (wi::gts_p (a, a)); |
2582 | ASSERT_TRUE (wi::gts_p (a, b)); |
2583 | ASSERT_FALSE (wi::gts_p (b, a)); |
2584 | |
2585 | /* >= */ |
2586 | ASSERT_TRUE (wi::ges_p (a, a)); |
2587 | ASSERT_TRUE (wi::ges_p (a, b)); |
2588 | ASSERT_FALSE (wi::ges_p (b, a)); |
2589 | |
2590 | /* comparison */ |
2591 | ASSERT_EQ (-1, wi::cmps (b, a)); |
2592 | ASSERT_EQ (0, wi::cmps (a, a)); |
2593 | ASSERT_EQ (1, wi::cmps (a, b)); |
2594 | } |
2595 | |
2596 | /* Run all of the selftests, using the given VALUE_TYPE. */ |
2597 | |
2598 | template <class VALUE_TYPE> |
2599 | static void run_all_wide_int_tests () |
2600 | { |
2601 | test_printing <VALUE_TYPE> (); |
2602 | test_ops <VALUE_TYPE> (); |
2603 | test_comparisons <VALUE_TYPE> (); |
2604 | } |
2605 | |
2606 | /* Test overflow conditions. */ |
2607 | |
2608 | static void |
2609 | test_overflow () |
2610 | { |
2611 | static int precs[] = { 31, 32, 33, 63, 64, 65, 127, 128 }; |
2612 | static int offsets[] = { 16, 1, 0 }; |
2613 | for (unsigned int i = 0; i < ARRAY_SIZE (precs); ++i) |
2614 | for (unsigned int j = 0; j < ARRAY_SIZE (offsets); ++j) |
2615 | { |
2616 | int prec = precs[i]; |
2617 | int offset = offsets[j]; |
2618 | wi::overflow_type overflow; |
2619 | wide_int sum, diff; |
2620 | |
2621 | sum = wi::add (x: wi::max_value (precision: prec, sgn: UNSIGNED) - offset, y: 1, |
2622 | sgn: UNSIGNED, overflow: &overflow); |
2623 | ASSERT_EQ (sum, -offset); |
2624 | ASSERT_EQ (overflow != wi::OVF_NONE, offset == 0); |
2625 | |
2626 | sum = wi::add (x: 1, y: wi::max_value (precision: prec, sgn: UNSIGNED) - offset, |
2627 | sgn: UNSIGNED, overflow: &overflow); |
2628 | ASSERT_EQ (sum, -offset); |
2629 | ASSERT_EQ (overflow != wi::OVF_NONE, offset == 0); |
2630 | |
2631 | diff = wi::sub (x: wi::max_value (precision: prec, sgn: UNSIGNED) - offset, |
2632 | y: wi::max_value (precision: prec, sgn: UNSIGNED), |
2633 | sgn: UNSIGNED, overflow: &overflow); |
2634 | ASSERT_EQ (diff, -offset); |
2635 | ASSERT_EQ (overflow != wi::OVF_NONE, offset != 0); |
2636 | |
2637 | diff = wi::sub (x: wi::max_value (precision: prec, sgn: UNSIGNED) - offset, |
2638 | y: wi::max_value (precision: prec, sgn: UNSIGNED) - 1, |
2639 | sgn: UNSIGNED, overflow: &overflow); |
2640 | ASSERT_EQ (diff, 1 - offset); |
2641 | ASSERT_EQ (overflow != wi::OVF_NONE, offset > 1); |
2642 | } |
2643 | } |
2644 | |
2645 | /* Test the round_{down,up}_for_mask functions. */ |
2646 | |
2647 | static void |
2648 | test_round_for_mask () |
2649 | { |
2650 | unsigned int prec = 18; |
2651 | ASSERT_EQ (17, wi::round_down_for_mask (wi::shwi (17, prec), |
2652 | wi::shwi (0xf1, prec))); |
2653 | ASSERT_EQ (17, wi::round_up_for_mask (wi::shwi (17, prec), |
2654 | wi::shwi (0xf1, prec))); |
2655 | |
2656 | ASSERT_EQ (1, wi::round_down_for_mask (wi::shwi (6, prec), |
2657 | wi::shwi (0xf1, prec))); |
2658 | ASSERT_EQ (16, wi::round_up_for_mask (wi::shwi (6, prec), |
2659 | wi::shwi (0xf1, prec))); |
2660 | |
2661 | ASSERT_EQ (17, wi::round_down_for_mask (wi::shwi (24, prec), |
2662 | wi::shwi (0xf1, prec))); |
2663 | ASSERT_EQ (32, wi::round_up_for_mask (wi::shwi (24, prec), |
2664 | wi::shwi (0xf1, prec))); |
2665 | |
2666 | ASSERT_EQ (0x011, wi::round_down_for_mask (wi::shwi (0x22, prec), |
2667 | wi::shwi (0x111, prec))); |
2668 | ASSERT_EQ (0x100, wi::round_up_for_mask (wi::shwi (0x22, prec), |
2669 | wi::shwi (0x111, prec))); |
2670 | |
2671 | ASSERT_EQ (100, wi::round_down_for_mask (wi::shwi (101, prec), |
2672 | wi::shwi (0xfc, prec))); |
2673 | ASSERT_EQ (104, wi::round_up_for_mask (wi::shwi (101, prec), |
2674 | wi::shwi (0xfc, prec))); |
2675 | |
2676 | ASSERT_EQ (0x2bc, wi::round_down_for_mask (wi::shwi (0x2c2, prec), |
2677 | wi::shwi (0xabc, prec))); |
2678 | ASSERT_EQ (0x800, wi::round_up_for_mask (wi::shwi (0x2c2, prec), |
2679 | wi::shwi (0xabc, prec))); |
2680 | |
2681 | ASSERT_EQ (0xabc, wi::round_down_for_mask (wi::shwi (0xabd, prec), |
2682 | wi::shwi (0xabc, prec))); |
2683 | ASSERT_EQ (0, wi::round_up_for_mask (wi::shwi (0xabd, prec), |
2684 | wi::shwi (0xabc, prec))); |
2685 | |
2686 | ASSERT_EQ (0xabc, wi::round_down_for_mask (wi::shwi (0x1000, prec), |
2687 | wi::shwi (0xabc, prec))); |
2688 | ASSERT_EQ (0, wi::round_up_for_mask (wi::shwi (0x1000, prec), |
2689 | wi::shwi (0xabc, prec))); |
2690 | } |
2691 | |
2692 | /* Run all of the selftests within this file, for all value types. */ |
2693 | |
2694 | void |
2695 | wide_int_cc_tests () |
2696 | { |
2697 | run_all_wide_int_tests <wide_int> (); |
2698 | run_all_wide_int_tests <offset_int> (); |
2699 | run_all_wide_int_tests <widest_int> (); |
2700 | test_overflow (); |
2701 | test_round_for_mask (); |
2702 | ASSERT_EQ (wi::mask (128, false, 128), |
2703 | wi::shifted_mask (0, 128, false, 128)); |
2704 | ASSERT_EQ (wi::mask (128, true, 128), |
2705 | wi::shifted_mask (0, 128, true, 128)); |
2706 | ASSERT_EQ (wi::multiple_of_p (from_int <widest_int> (1), |
2707 | from_int <widest_int> (-128), UNSIGNED), |
2708 | false); |
2709 | } |
2710 | |
2711 | } // namespace selftest |
2712 | #endif /* CHECKING_P */ |
2713 | |