Line data Source code
1 : /* Functions to determine/estimate number of iterations of a loop.
2 : Copyright (C) 2004-2026 Free Software Foundation, Inc.
3 :
4 : This file is part of GCC.
5 :
6 : GCC is free software; you can redistribute it and/or modify it
7 : under the terms of the GNU General Public License as published by the
8 : Free Software Foundation; either version 3, or (at your option) any
9 : later version.
10 :
11 : GCC is distributed in the hope that it will be useful, but WITHOUT
12 : ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 : FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 : for more details.
15 :
16 : You should have received a copy of the GNU General Public License
17 : along with GCC; see the file COPYING3. If not see
18 : <http://www.gnu.org/licenses/>. */
19 :
20 : #include "config.h"
21 : #include "system.h"
22 : #include "coretypes.h"
23 : #include "backend.h"
24 : #include "rtl.h"
25 : #include "tree.h"
26 : #include "gimple.h"
27 : #include "tree-pass.h"
28 : #include "ssa.h"
29 : #include "gimple-pretty-print.h"
30 : #include "diagnostic-core.h"
31 : #include "stor-layout.h"
32 : #include "fold-const.h"
33 : #include "calls.h"
34 : #include "intl.h"
35 : #include "gimple-iterator.h"
36 : #include "tree-cfg.h"
37 : #include "tree-ssa-loop-ivopts.h"
38 : #include "tree-ssa-loop-niter.h"
39 : #include "tree-ssa-loop.h"
40 : #include "cfgloop.h"
41 : #include "tree-chrec.h"
42 : #include "tree-scalar-evolution.h"
43 : #include "tree-data-ref.h"
44 : #include "tree-dfa.h"
45 : #include "internal-fn.h"
46 : #include "gimple-range.h"
47 : #include "sreal.h"
48 :
49 :
50 : /*
51 :
52 : Analysis of number of iterations of an affine exit test.
53 :
54 : */
55 :
56 : /* Bounds on some value, BELOW <= X <= UP. */
57 :
58 : struct bounds
59 : {
60 : mpz_t below, up;
61 : };
62 :
63 : /* Splits expression EXPR to a variable part VAR and constant OFFSET. */
64 :
65 : static void
66 57872840 : split_to_var_and_offset (tree expr, tree *var, mpz_t offset)
67 : {
68 57872840 : tree type = TREE_TYPE (expr);
69 57872840 : tree op0, op1;
70 57872840 : bool negate = false;
71 :
72 57872840 : *var = expr;
73 57872840 : mpz_set_ui (offset, 0);
74 :
75 57872840 : switch (TREE_CODE (expr))
76 : {
77 8183 : case MINUS_EXPR:
78 8183 : negate = true;
79 : /* Fallthru. */
80 :
81 5869337 : case PLUS_EXPR:
82 5869337 : case POINTER_PLUS_EXPR:
83 5869337 : op0 = TREE_OPERAND (expr, 0);
84 5869337 : op1 = TREE_OPERAND (expr, 1);
85 :
86 5869337 : if (TREE_CODE (op1) != INTEGER_CST)
87 : break;
88 :
89 5857803 : *var = op0;
90 : /* Always sign extend the offset. */
91 5857803 : wi::to_mpz (wi::to_wide (op1), offset, SIGNED);
92 5857803 : if (negate)
93 34 : mpz_neg (offset, offset);
94 : break;
95 :
96 25690389 : case INTEGER_CST:
97 25690389 : *var = build_int_cst_type (type, 0);
98 25690389 : wi::to_mpz (wi::to_wide (expr), offset, TYPE_SIGN (type));
99 25690389 : break;
100 :
101 : default:
102 : break;
103 : }
104 57872840 : }
105 :
106 : /* From condition C0 CMP C1 derives information regarding the value range
107 : of VAR, which is of TYPE. Results are stored in to BELOW and UP. */
108 :
109 : static void
110 20792921 : refine_value_range_using_guard (tree type, tree var,
111 : tree c0, enum tree_code cmp, tree c1,
112 : mpz_t below, mpz_t up)
113 : {
114 20792921 : tree varc0, varc1, ctype;
115 20792921 : mpz_t offc0, offc1;
116 20792921 : mpz_t mint, maxt, minc1, maxc1;
117 20792921 : bool no_wrap = nowrap_type_p (type);
118 20792921 : bool c0_ok, c1_ok;
119 20792921 : signop sgn = TYPE_SIGN (type);
120 :
121 20792921 : switch (cmp)
122 : {
123 9743506 : case LT_EXPR:
124 9743506 : case LE_EXPR:
125 9743506 : case GT_EXPR:
126 9743506 : case GE_EXPR:
127 9743506 : STRIP_SIGN_NOPS (c0);
128 9743506 : STRIP_SIGN_NOPS (c1);
129 9743506 : ctype = TREE_TYPE (c0);
130 9743506 : if (!useless_type_conversion_p (ctype, type))
131 17844329 : return;
132 :
133 : break;
134 :
135 : case EQ_EXPR:
136 : /* We could derive quite precise information from EQ_EXPR, however,
137 : such a guard is unlikely to appear, so we do not bother with
138 : handling it. */
139 : return;
140 :
141 5145761 : case NE_EXPR:
142 : /* NE_EXPR comparisons do not contain much of useful information,
143 : except for cases of comparing with bounds. */
144 5145761 : if (TREE_CODE (c1) != INTEGER_CST
145 4823498 : || !INTEGRAL_TYPE_P (type))
146 : return;
147 :
148 : /* Ensure that the condition speaks about an expression in the same
149 : type as X and Y. */
150 4823498 : ctype = TREE_TYPE (c0);
151 4823498 : if (TYPE_PRECISION (ctype) != TYPE_PRECISION (type))
152 : return;
153 2906488 : c0 = fold_convert (type, c0);
154 2906488 : c1 = fold_convert (type, c1);
155 :
156 2906488 : if (operand_equal_p (var, c0, 0))
157 : {
158 : /* Case of comparing VAR with its below/up bounds. */
159 309367 : auto_mpz valc1;
160 309367 : wi::to_mpz (wi::to_wide (c1), valc1, TYPE_SIGN (type));
161 309367 : if (mpz_cmp (valc1, below) == 0)
162 102542 : cmp = GT_EXPR;
163 309367 : if (mpz_cmp (valc1, up) == 0)
164 49197 : cmp = LT_EXPR;
165 309367 : }
166 : else
167 : {
168 : /* Case of comparing with the bounds of the type. */
169 2597121 : wide_int min = wi::min_value (type);
170 2597121 : wide_int max = wi::max_value (type);
171 :
172 2597121 : if (wi::to_wide (c1) == min)
173 745065 : cmp = GT_EXPR;
174 2597121 : if (wi::to_wide (c1) == max)
175 32430 : cmp = LT_EXPR;
176 2597121 : }
177 :
178 : /* Quick return if no useful information. */
179 2906488 : if (cmp == NE_EXPR)
180 : return;
181 :
182 : break;
183 :
184 : default:
185 : return;
186 : }
187 :
188 9016997 : mpz_init (offc0);
189 9016997 : mpz_init (offc1);
190 9016997 : split_to_var_and_offset (expand_simple_operations (c0), &varc0, offc0);
191 9016997 : split_to_var_and_offset (expand_simple_operations (c1), &varc1, offc1);
192 :
193 : /* We are only interested in comparisons of expressions based on VAR. */
194 9016997 : if (operand_equal_p (var, varc1, 0))
195 : {
196 853655 : std::swap (varc0, varc1);
197 853655 : mpz_swap (offc0, offc1);
198 853655 : cmp = swap_tree_comparison (cmp);
199 : }
200 8163342 : else if (!operand_equal_p (var, varc0, 0))
201 : {
202 6068405 : mpz_clear (offc0);
203 6068405 : mpz_clear (offc1);
204 6068405 : return;
205 : }
206 :
207 2948592 : mpz_init (mint);
208 2948592 : mpz_init (maxt);
209 2948592 : get_type_static_bounds (type, mint, maxt);
210 2948592 : mpz_init (minc1);
211 2948592 : mpz_init (maxc1);
212 5897184 : int_range_max r (TREE_TYPE (varc1));
213 : /* Setup range information for varc1. */
214 2948592 : if (integer_zerop (varc1))
215 : {
216 1438899 : wi::to_mpz (0, minc1, TYPE_SIGN (type));
217 1438899 : wi::to_mpz (0, maxc1, TYPE_SIGN (type));
218 : }
219 1509693 : else if (TREE_CODE (varc1) == SSA_NAME
220 1418600 : && INTEGRAL_TYPE_P (type)
221 2837200 : && get_range_query (cfun)->range_of_expr (r, varc1)
222 1418600 : && !r.undefined_p ()
223 2925612 : && !r.varying_p ())
224 : {
225 1500604 : gcc_assert (wi::le_p (r.lower_bound (), r.upper_bound (), sgn));
226 750302 : wi::to_mpz (r.lower_bound (), minc1, sgn);
227 750302 : wi::to_mpz (r.upper_bound (), maxc1, sgn);
228 : }
229 : else
230 : {
231 759391 : mpz_set (minc1, mint);
232 759391 : mpz_set (maxc1, maxt);
233 : }
234 :
235 : /* Compute valid range information for varc1 + offc1. Note nothing
236 : useful can be derived if it overflows or underflows. Overflow or
237 : underflow could happen when:
238 :
239 : offc1 > 0 && varc1 + offc1 > MAX_VAL (type)
240 : offc1 < 0 && varc1 + offc1 < MIN_VAL (type). */
241 2948592 : mpz_add (minc1, minc1, offc1);
242 2948592 : mpz_add (maxc1, maxc1, offc1);
243 5897184 : c1_ok = (no_wrap
244 837774 : || mpz_sgn (offc1) == 0
245 133881 : || (mpz_sgn (offc1) < 0 && mpz_cmp (minc1, mint) >= 0)
246 3081048 : || (mpz_sgn (offc1) > 0 && mpz_cmp (maxc1, maxt) <= 0));
247 29880 : if (!c1_ok)
248 29880 : goto end;
249 :
250 2918712 : if (mpz_cmp (minc1, mint) < 0)
251 26381 : mpz_set (minc1, mint);
252 2918712 : if (mpz_cmp (maxc1, maxt) > 0)
253 17122 : mpz_set (maxc1, maxt);
254 :
255 2918712 : if (cmp == LT_EXPR)
256 : {
257 412776 : cmp = LE_EXPR;
258 412776 : mpz_sub_ui (maxc1, maxc1, 1);
259 : }
260 2918712 : if (cmp == GT_EXPR)
261 : {
262 1484682 : cmp = GE_EXPR;
263 1484682 : mpz_add_ui (minc1, minc1, 1);
264 : }
265 :
266 : /* Compute range information for varc0. If there is no overflow,
267 : the condition implied that
268 :
269 : (varc0) cmp (varc1 + offc1 - offc0)
270 :
271 : We can possibly improve the upper bound of varc0 if cmp is LE_EXPR,
272 : or the below bound if cmp is GE_EXPR.
273 :
274 : To prove there is no overflow/underflow, we need to check below
275 : four cases:
276 : 1) cmp == LE_EXPR && offc0 > 0
277 :
278 : (varc0 + offc0) doesn't overflow
279 : && (varc1 + offc1 - offc0) doesn't underflow
280 :
281 : 2) cmp == LE_EXPR && offc0 < 0
282 :
283 : (varc0 + offc0) doesn't underflow
284 : && (varc1 + offc1 - offc0) doesn't overfloe
285 :
286 : In this case, (varc0 + offc0) will never underflow if we can
287 : prove (varc1 + offc1 - offc0) doesn't overflow.
288 :
289 : 3) cmp == GE_EXPR && offc0 < 0
290 :
291 : (varc0 + offc0) doesn't underflow
292 : && (varc1 + offc1 - offc0) doesn't overflow
293 :
294 : 4) cmp == GE_EXPR && offc0 > 0
295 :
296 : (varc0 + offc0) doesn't overflow
297 : && (varc1 + offc1 - offc0) doesn't underflow
298 :
299 : In this case, (varc0 + offc0) will never overflow if we can
300 : prove (varc1 + offc1 - offc0) doesn't underflow.
301 :
302 : Note we only handle case 2 and 4 in below code. */
303 :
304 2918712 : mpz_sub (minc1, minc1, offc0);
305 2918712 : mpz_sub (maxc1, maxc1, offc0);
306 5837424 : c0_ok = (no_wrap
307 807894 : || mpz_sgn (offc0) == 0
308 51733 : || (cmp == LE_EXPR
309 33739 : && mpz_sgn (offc0) < 0 && mpz_cmp (maxc1, maxt) <= 0)
310 2967701 : || (cmp == GE_EXPR
311 17994 : && mpz_sgn (offc0) > 0 && mpz_cmp (minc1, mint) >= 0));
312 46641 : if (!c0_ok)
313 46641 : goto end;
314 :
315 2872071 : if (cmp == LE_EXPR)
316 : {
317 1103202 : if (mpz_cmp (up, maxc1) > 0)
318 350975 : mpz_set (up, maxc1);
319 : }
320 : else
321 : {
322 1768869 : if (mpz_cmp (below, minc1) < 0)
323 1172944 : mpz_set (below, minc1);
324 : }
325 :
326 595925 : end:
327 2948592 : mpz_clear (mint);
328 2948592 : mpz_clear (maxt);
329 2948592 : mpz_clear (minc1);
330 2948592 : mpz_clear (maxc1);
331 2948592 : mpz_clear (offc0);
332 2948592 : mpz_clear (offc1);
333 : }
334 :
335 : /* Stores estimates of the minimum and maximum values of the expression
336 : VAR + OFF in TYPE to MIN and MAX. The estimates are valid on entry
337 : to LOOP, i.e. on the loop preheader edge. */
338 :
339 : static void
340 25378450 : determine_value_range (class loop *loop, tree type, tree var, mpz_t off,
341 : mpz_t min, mpz_t max)
342 : {
343 25378450 : int cnt = 0;
344 25378450 : mpz_t minm, maxm;
345 25378450 : basic_block bb;
346 25378450 : wide_int minv, maxv;
347 25378450 : enum value_range_kind rtype = VR_VARYING;
348 :
349 : /* If the expression is a constant, we know its value exactly. */
350 25378450 : if (integer_zerop (var))
351 : {
352 17931228 : mpz_set (min, off);
353 17931228 : mpz_set (max, off);
354 17931228 : return;
355 : }
356 :
357 7447222 : get_type_static_bounds (type, min, max);
358 :
359 : /* See if we have some range info from VRP. */
360 7447222 : if (INTEGRAL_TYPE_P (type))
361 : {
362 5922414 : edge e = loop_preheader_edge (loop);
363 5922414 : signop sgn = TYPE_SIGN (type);
364 5922414 : gphi_iterator gsi;
365 :
366 : /* Either for VAR itself... */
367 5922414 : int_range_max var_range (TREE_TYPE (var));
368 11844828 : get_range_query (cfun)->range_on_edge (var_range, e, var);
369 5922414 : if (var_range.varying_p () || var_range.undefined_p ())
370 : rtype = VR_VARYING;
371 : else
372 : rtype = VR_RANGE;
373 5922414 : if (!var_range.undefined_p ())
374 : {
375 5905910 : minv = var_range.lower_bound ();
376 5905978 : maxv = var_range.upper_bound ();
377 : }
378 :
379 : /* Or for PHI results in loop->header where VAR is used as
380 : PHI argument from the loop preheader edge. */
381 5922414 : int_range_max phi_range (TREE_TYPE (var));
382 21815660 : for (gsi = gsi_start_phis (loop->header); !gsi_end_p (gsi); gsi_next (&gsi))
383 : {
384 15893247 : gphi *phi = gsi.phi ();
385 15893247 : if (PHI_ARG_DEF_FROM_EDGE (phi, e) == var
386 2623114 : && get_range_query (cfun)->range_on_edge (phi_range,
387 : e, gimple_phi_result (phi))
388 1311557 : && !phi_range.varying_p ()
389 16638311 : && !phi_range.undefined_p ())
390 : {
391 745060 : if (rtype != VR_RANGE)
392 : {
393 253615 : rtype = VR_RANGE;
394 253615 : minv = phi_range.lower_bound ();
395 253630 : maxv = phi_range.upper_bound ();
396 : }
397 : else
398 : {
399 491447 : minv = wi::max (minv, phi_range.lower_bound (), sgn);
400 491447 : maxv = wi::min (maxv, phi_range.upper_bound (), sgn);
401 : /* If the PHI result range are inconsistent with
402 : the VAR range, give up on looking at the PHI
403 : results. This can happen if VR_UNDEFINED is
404 : involved. */
405 491445 : if (wi::gt_p (minv, maxv, sgn))
406 : {
407 1 : int_range_max vr (TREE_TYPE (var));
408 2 : get_range_query (cfun)->range_on_edge (vr, e, var);
409 1 : if (vr.varying_p () || vr.undefined_p ())
410 : rtype = VR_VARYING;
411 : else
412 : rtype = VR_RANGE;
413 1 : if (!vr.undefined_p ())
414 : {
415 1 : minv = vr.lower_bound ();
416 1 : maxv = vr.upper_bound ();
417 : }
418 1 : break;
419 1 : }
420 : }
421 : }
422 : }
423 5922414 : mpz_init (minm);
424 5922414 : mpz_init (maxm);
425 5922414 : if (rtype != VR_RANGE)
426 : {
427 3017283 : mpz_set (minm, min);
428 3017283 : mpz_set (maxm, max);
429 : }
430 : else
431 : {
432 2905131 : gcc_assert (wi::le_p (minv, maxv, sgn));
433 2905131 : wi::to_mpz (minv, minm, sgn);
434 2905131 : wi::to_mpz (maxv, maxm, sgn);
435 : }
436 : /* Now walk the dominators of the loop header and use the entry
437 : guards to refine the estimates. */
438 5922414 : for (bb = loop->header;
439 60138113 : bb != ENTRY_BLOCK_PTR_FOR_FN (cfun)
440 60138113 : && cnt < param_max_niter_dominators_walk;
441 54215699 : bb = get_immediate_dominator (CDI_DOMINATORS, bb))
442 : {
443 54215699 : edge e;
444 54215699 : tree c0, c1;
445 54215699 : enum tree_code cmp;
446 :
447 54215699 : if (!single_pred_p (bb))
448 26986438 : continue;
449 27229261 : e = single_pred_edge (bb);
450 :
451 27229261 : if (!(e->flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
452 6436340 : continue;
453 :
454 41585842 : gcond *cond = as_a <gcond *> (*gsi_last_bb (e->src));
455 20792921 : c0 = gimple_cond_lhs (cond);
456 20792921 : cmp = gimple_cond_code (cond);
457 20792921 : c1 = gimple_cond_rhs (cond);
458 :
459 20792921 : if (e->flags & EDGE_FALSE_VALUE)
460 11176779 : cmp = invert_tree_comparison (cmp, false);
461 :
462 20792921 : refine_value_range_using_guard (type, var, c0, cmp, c1, minm, maxm);
463 20792921 : ++cnt;
464 : }
465 :
466 5922414 : mpz_add (minm, minm, off);
467 5922414 : mpz_add (maxm, maxm, off);
468 : /* If the computation may not wrap or off is zero, then this
469 : is always fine. If off is negative and minv + off isn't
470 : smaller than type's minimum, or off is positive and
471 : maxv + off isn't bigger than type's maximum, use the more
472 : precise range too. */
473 5922414 : if (nowrap_type_p (type)
474 2323761 : || mpz_sgn (off) == 0
475 585018 : || (mpz_sgn (off) < 0 && mpz_cmp (minm, min) >= 0)
476 6422023 : || (mpz_sgn (off) > 0 && mpz_cmp (maxm, max) <= 0))
477 : {
478 5716069 : mpz_set (min, minm);
479 5716069 : mpz_set (max, maxm);
480 5716069 : mpz_clear (minm);
481 5716069 : mpz_clear (maxm);
482 5716069 : return;
483 : }
484 206345 : mpz_clear (minm);
485 206345 : mpz_clear (maxm);
486 5922414 : }
487 :
488 : /* If the computation may wrap, we know nothing about the value, except for
489 : the range of the type. */
490 1731153 : if (!nowrap_type_p (type))
491 : return;
492 :
493 : /* Since the addition of OFF does not wrap, if OFF is positive, then we may
494 : add it to MIN, otherwise to MAX. */
495 1524696 : if (mpz_sgn (off) < 0)
496 30602 : mpz_add (max, max, off);
497 : else
498 1494094 : mpz_add (min, min, off);
499 25378518 : }
500 :
501 : /* Stores the bounds on the difference of the values of the expressions
502 : (var + X) and (var + Y), computed in TYPE, to BNDS. */
503 :
504 : static void
505 172424 : bound_difference_of_offsetted_base (tree type, mpz_t x, mpz_t y,
506 : bounds *bnds)
507 : {
508 172424 : int rel = mpz_cmp (x, y);
509 172424 : bool may_wrap = !nowrap_type_p (type);
510 :
511 : /* If X == Y, then the expressions are always equal.
512 : If X > Y, there are the following possibilities:
513 : a) neither of var + X and var + Y overflow or underflow, or both of
514 : them do. Then their difference is X - Y.
515 : b) var + X overflows, and var + Y does not. Then the values of the
516 : expressions are var + X - M and var + Y, where M is the range of
517 : the type, and their difference is X - Y - M.
518 : c) var + Y underflows and var + X does not. Their difference again
519 : is M - X + Y.
520 : Therefore, if the arithmetics in type does not overflow, then the
521 : bounds are (X - Y, X - Y), otherwise they are (X - Y - M, X - Y)
522 : Similarly, if X < Y, the bounds are either (X - Y, X - Y) or
523 : (X - Y, X - Y + M). */
524 :
525 172424 : if (rel == 0)
526 : {
527 770 : mpz_set_ui (bnds->below, 0);
528 770 : mpz_set_ui (bnds->up, 0);
529 770 : return;
530 : }
531 :
532 171654 : auto_mpz m;
533 171654 : wi::to_mpz (wi::minus_one (TYPE_PRECISION (type)), m, UNSIGNED);
534 171654 : mpz_add_ui (m, m, 1);
535 171654 : mpz_sub (bnds->up, x, y);
536 171654 : mpz_set (bnds->below, bnds->up);
537 :
538 171654 : if (may_wrap)
539 : {
540 147114 : if (rel > 0)
541 145704 : mpz_sub (bnds->below, bnds->below, m);
542 : else
543 1410 : mpz_add (bnds->up, bnds->up, m);
544 : }
545 171654 : }
546 :
547 : /* From condition C0 CMP C1 derives information regarding the
548 : difference of values of VARX + OFFX and VARY + OFFY, computed in TYPE,
549 : and stores it to BNDS. */
550 :
551 : static void
552 19814659 : refine_bounds_using_guard (tree type, tree varx, mpz_t offx,
553 : tree vary, mpz_t offy,
554 : tree c0, enum tree_code cmp, tree c1,
555 : bounds *bnds)
556 : {
557 19814659 : tree varc0, varc1, ctype;
558 19814659 : mpz_t offc0, offc1, loffx, loffy, bnd;
559 19814659 : bool lbound = false;
560 19814659 : bool no_wrap = nowrap_type_p (type);
561 19814659 : bool x_ok, y_ok;
562 :
563 19814659 : switch (cmp)
564 : {
565 8394522 : case LT_EXPR:
566 8394522 : case LE_EXPR:
567 8394522 : case GT_EXPR:
568 8394522 : case GE_EXPR:
569 8394522 : STRIP_SIGN_NOPS (c0);
570 8394522 : STRIP_SIGN_NOPS (c1);
571 8394522 : ctype = TREE_TYPE (c0);
572 8394522 : if (!useless_type_conversion_p (ctype, type))
573 12756885 : return;
574 :
575 : break;
576 :
577 : case EQ_EXPR:
578 : /* We could derive quite precise information from EQ_EXPR, however, such
579 : a guard is unlikely to appear, so we do not bother with handling
580 : it. */
581 : return;
582 :
583 5542670 : case NE_EXPR:
584 : /* NE_EXPR comparisons do not contain much of useful information, except for
585 : special case of comparing with the bounds of the type. */
586 5542670 : if (TREE_CODE (c1) != INTEGER_CST
587 4760661 : || !INTEGRAL_TYPE_P (type))
588 : return;
589 :
590 : /* Ensure that the condition speaks about an expression in the same type
591 : as X and Y. */
592 4184746 : ctype = TREE_TYPE (c0);
593 4184746 : if (TYPE_PRECISION (ctype) != TYPE_PRECISION (type))
594 : return;
595 2387026 : c0 = fold_convert (type, c0);
596 2387026 : c1 = fold_convert (type, c1);
597 :
598 2387026 : if (TYPE_MIN_VALUE (type)
599 2387026 : && operand_equal_p (c1, TYPE_MIN_VALUE (type), 0))
600 : {
601 : cmp = GT_EXPR;
602 : break;
603 : }
604 1692020 : if (TYPE_MAX_VALUE (type)
605 1692020 : && operand_equal_p (c1, TYPE_MAX_VALUE (type), 0))
606 : {
607 : cmp = LT_EXPR;
608 : break;
609 : }
610 :
611 : return;
612 : default:
613 : return;
614 : }
615 :
616 7057774 : mpz_init (offc0);
617 7057774 : mpz_init (offc1);
618 7057774 : split_to_var_and_offset (expand_simple_operations (c0), &varc0, offc0);
619 7057774 : split_to_var_and_offset (expand_simple_operations (c1), &varc1, offc1);
620 :
621 : /* We are only interested in comparisons of expressions based on VARX and
622 : VARY. TODO -- we might also be able to derive some bounds from
623 : expressions containing just one of the variables. */
624 :
625 7057774 : if (operand_equal_p (varx, varc1, 0))
626 : {
627 953012 : std::swap (varc0, varc1);
628 953012 : mpz_swap (offc0, offc1);
629 953012 : cmp = swap_tree_comparison (cmp);
630 : }
631 :
632 7057774 : if (!operand_equal_p (varx, varc0, 0)
633 7057774 : || !operand_equal_p (vary, varc1, 0))
634 5245942 : goto end;
635 :
636 1811832 : mpz_init_set (loffx, offx);
637 1811832 : mpz_init_set (loffy, offy);
638 :
639 1811832 : if (cmp == GT_EXPR || cmp == GE_EXPR)
640 : {
641 1725888 : std::swap (varx, vary);
642 1725888 : mpz_swap (offc0, offc1);
643 1725888 : mpz_swap (loffx, loffy);
644 1725888 : cmp = swap_tree_comparison (cmp);
645 1725888 : lbound = true;
646 : }
647 :
648 : /* If there is no overflow, the condition implies that
649 :
650 : (VARX + OFFX) cmp (VARY + OFFY) + (OFFX - OFFY + OFFC1 - OFFC0).
651 :
652 : The overflows and underflows may complicate things a bit; each
653 : overflow decreases the appropriate offset by M, and underflow
654 : increases it by M. The above inequality would not necessarily be
655 : true if
656 :
657 : -- VARX + OFFX underflows and VARX + OFFC0 does not, or
658 : VARX + OFFC0 overflows, but VARX + OFFX does not.
659 : This may only happen if OFFX < OFFC0.
660 : -- VARY + OFFY overflows and VARY + OFFC1 does not, or
661 : VARY + OFFC1 underflows and VARY + OFFY does not.
662 : This may only happen if OFFY > OFFC1. */
663 :
664 1811832 : if (no_wrap)
665 : {
666 : x_ok = true;
667 : y_ok = true;
668 : }
669 : else
670 : {
671 611968 : x_ok = (integer_zerop (varx)
672 611968 : || mpz_cmp (loffx, offc0) >= 0);
673 611968 : y_ok = (integer_zerop (vary)
674 611968 : || mpz_cmp (loffy, offc1) <= 0);
675 : }
676 :
677 609335 : if (x_ok && y_ok)
678 : {
679 1805239 : mpz_init (bnd);
680 1805239 : mpz_sub (bnd, loffx, loffy);
681 1805239 : mpz_add (bnd, bnd, offc1);
682 1805239 : mpz_sub (bnd, bnd, offc0);
683 :
684 1805239 : if (cmp == LT_EXPR)
685 1455435 : mpz_sub_ui (bnd, bnd, 1);
686 :
687 1805239 : if (lbound)
688 : {
689 1720561 : mpz_neg (bnd, bnd);
690 1720561 : if (mpz_cmp (bnds->below, bnd) < 0)
691 463717 : mpz_set (bnds->below, bnd);
692 : }
693 : else
694 : {
695 84678 : if (mpz_cmp (bnd, bnds->up) < 0)
696 4124 : mpz_set (bnds->up, bnd);
697 : }
698 1805239 : mpz_clear (bnd);
699 : }
700 :
701 1811832 : mpz_clear (loffx);
702 1811832 : mpz_clear (loffy);
703 7057774 : end:
704 7057774 : mpz_clear (offc0);
705 7057774 : mpz_clear (offc1);
706 : }
707 :
708 : /* Stores the bounds on the value of the expression X - Y in LOOP to BNDS.
709 : The subtraction is considered to be performed in arbitrary precision,
710 : without overflows.
711 :
712 : We do not attempt to be too clever regarding the value ranges of X and
713 : Y; most of the time, they are just integers or ssa names offsetted by
714 : integer. However, we try to use the information contained in the
715 : comparisons before the loop (usually created by loop header copying). */
716 :
717 : static void
718 12861649 : bound_difference (class loop *loop, tree x, tree y, bounds *bnds)
719 : {
720 12861649 : tree type = TREE_TYPE (x);
721 12861649 : tree varx, vary;
722 12861649 : mpz_t offx, offy;
723 12861649 : int cnt = 0;
724 12861649 : edge e;
725 12861649 : basic_block bb;
726 12861649 : tree c0, c1;
727 12861649 : enum tree_code cmp;
728 :
729 : /* Get rid of unnecessary casts, but preserve the value of
730 : the expressions. */
731 12861649 : STRIP_SIGN_NOPS (x);
732 12861649 : STRIP_SIGN_NOPS (y);
733 :
734 12861649 : mpz_init (bnds->below);
735 12861649 : mpz_init (bnds->up);
736 12861649 : mpz_init (offx);
737 12861649 : mpz_init (offy);
738 12861649 : split_to_var_and_offset (x, &varx, offx);
739 12861649 : split_to_var_and_offset (y, &vary, offy);
740 :
741 12861649 : if (!integer_zerop (varx)
742 12861649 : && operand_equal_p (varx, vary, 0))
743 : {
744 : /* Special case VARX == VARY -- we just need to compare the
745 : offsets. The matters are a bit more complicated in the
746 : case addition of offsets may wrap. */
747 172424 : bound_difference_of_offsetted_base (type, offx, offy, bnds);
748 : }
749 : else
750 : {
751 : /* Otherwise, use the value ranges to determine the initial
752 : estimates on below and up. */
753 12689225 : auto_mpz minx, maxx, miny, maxy;
754 12689225 : determine_value_range (loop, type, varx, offx, minx, maxx);
755 12689225 : determine_value_range (loop, type, vary, offy, miny, maxy);
756 :
757 12689225 : mpz_sub (bnds->below, minx, maxy);
758 12689225 : mpz_sub (bnds->up, maxx, miny);
759 12689225 : }
760 :
761 : /* If both X and Y are constants, we cannot get any more precise. */
762 12861649 : if (integer_zerop (varx) && integer_zerop (vary))
763 7124111 : goto end;
764 :
765 : /* Now walk the dominators of the loop header and use the entry
766 : guards to refine the estimates. */
767 5737538 : for (bb = loop->header;
768 57108415 : bb != ENTRY_BLOCK_PTR_FOR_FN (cfun)
769 57108415 : && cnt < param_max_niter_dominators_walk;
770 51370877 : bb = get_immediate_dominator (CDI_DOMINATORS, bb))
771 : {
772 51370877 : if (!single_pred_p (bb))
773 24604034 : continue;
774 26766843 : e = single_pred_edge (bb);
775 :
776 26766843 : if (!(e->flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
777 6952184 : continue;
778 :
779 39629318 : gcond *cond = as_a <gcond *> (*gsi_last_bb (e->src));
780 19814659 : c0 = gimple_cond_lhs (cond);
781 19814659 : cmp = gimple_cond_code (cond);
782 19814659 : c1 = gimple_cond_rhs (cond);
783 :
784 19814659 : if (e->flags & EDGE_FALSE_VALUE)
785 10984650 : cmp = invert_tree_comparison (cmp, false);
786 :
787 19814659 : refine_bounds_using_guard (type, varx, offx, vary, offy,
788 : c0, cmp, c1, bnds);
789 19814659 : ++cnt;
790 : }
791 :
792 5737538 : end:
793 12861649 : mpz_clear (offx);
794 12861649 : mpz_clear (offy);
795 12861649 : }
796 :
797 : /* Update the bounds in BNDS that restrict the value of X to the bounds
798 : that restrict the value of X + DELTA. X can be obtained as a
799 : difference of two values in TYPE. */
800 :
801 : static void
802 2083635 : bounds_add (bounds *bnds, const widest_int &delta, tree type)
803 : {
804 2083635 : mpz_t mdelta, max;
805 :
806 2083635 : mpz_init (mdelta);
807 2083635 : wi::to_mpz (delta, mdelta, SIGNED);
808 :
809 2083635 : mpz_init (max);
810 2083635 : wi::to_mpz (wi::minus_one (TYPE_PRECISION (type)), max, UNSIGNED);
811 :
812 2083635 : mpz_add (bnds->up, bnds->up, mdelta);
813 2083635 : mpz_add (bnds->below, bnds->below, mdelta);
814 :
815 2083635 : if (mpz_cmp (bnds->up, max) > 0)
816 121059 : mpz_set (bnds->up, max);
817 :
818 2083635 : mpz_neg (max, max);
819 2083635 : if (mpz_cmp (bnds->below, max) < 0)
820 9982 : mpz_set (bnds->below, max);
821 :
822 2083635 : mpz_clear (mdelta);
823 2083635 : mpz_clear (max);
824 2083635 : }
825 :
826 : /* Update the bounds in BNDS that restrict the value of X to the bounds
827 : that restrict the value of -X. */
828 :
829 : static void
830 2545904 : bounds_negate (bounds *bnds)
831 : {
832 2545904 : mpz_t tmp;
833 :
834 2545904 : mpz_init_set (tmp, bnds->up);
835 2545904 : mpz_neg (bnds->up, bnds->below);
836 2545904 : mpz_neg (bnds->below, tmp);
837 2545904 : mpz_clear (tmp);
838 2545904 : }
839 :
840 : /* Returns inverse of X modulo 2^s, where MASK = 2^s-1. */
841 :
842 : static tree
843 183291 : inverse (tree x, tree mask)
844 : {
845 183291 : tree type = TREE_TYPE (x);
846 183291 : tree rslt;
847 183291 : unsigned ctr = tree_floor_log2 (mask);
848 :
849 183291 : if (TYPE_PRECISION (type) <= HOST_BITS_PER_WIDE_INT)
850 : {
851 183247 : unsigned HOST_WIDE_INT ix;
852 183247 : unsigned HOST_WIDE_INT imask;
853 183247 : unsigned HOST_WIDE_INT irslt = 1;
854 :
855 183247 : gcc_assert (cst_and_fits_in_hwi (x));
856 183247 : gcc_assert (cst_and_fits_in_hwi (mask));
857 :
858 183247 : ix = int_cst_value (x);
859 183247 : imask = int_cst_value (mask);
860 :
861 10470269 : for (; ctr; ctr--)
862 : {
863 10103775 : irslt *= ix;
864 10103775 : ix *= ix;
865 : }
866 183247 : irslt &= imask;
867 :
868 183247 : rslt = build_int_cst_type (type, irslt);
869 : }
870 : else
871 : {
872 44 : rslt = build_int_cst (type, 1);
873 5632 : for (; ctr; ctr--)
874 : {
875 5588 : rslt = int_const_binop (MULT_EXPR, rslt, x);
876 5588 : x = int_const_binop (MULT_EXPR, x, x);
877 : }
878 44 : rslt = int_const_binop (BIT_AND_EXPR, rslt, mask);
879 : }
880 :
881 183291 : return rslt;
882 : }
883 :
884 : /* Derives the upper bound BND on the number of executions of loop with exit
885 : condition S * i <> C. If NO_OVERFLOW is true, then the control variable of
886 : the loop does not overflow. EXIT_MUST_BE_TAKEN is true if we are guaranteed
887 : that the loop ends through this exit, i.e., the induction variable ever
888 : reaches the value of C.
889 :
890 : The value C is equal to final - base, where final and base are the final and
891 : initial value of the actual induction variable in the analysed loop. BNDS
892 : bounds the value of this difference when computed in signed type with
893 : unbounded range, while the computation of C is performed in an unsigned
894 : type with the range matching the range of the type of the induction variable.
895 : In particular, BNDS.up contains an upper bound on C in the following cases:
896 : -- if the iv must reach its final value without overflow, i.e., if
897 : NO_OVERFLOW && EXIT_MUST_BE_TAKEN is true, or
898 : -- if final >= base, which we know to hold when BNDS.below >= 0. */
899 :
900 : static void
901 7520399 : number_of_iterations_ne_max (mpz_t bnd, bool no_overflow, tree c, tree s,
902 : bounds *bnds, bool exit_must_be_taken)
903 : {
904 7520399 : widest_int max;
905 7520399 : mpz_t d;
906 7520399 : tree type = TREE_TYPE (c);
907 15040798 : bool bnds_u_valid = ((no_overflow && exit_must_be_taken)
908 7520399 : || mpz_sgn (bnds->below) >= 0);
909 :
910 7520399 : if (integer_onep (s)
911 882220 : || (TREE_CODE (c) == INTEGER_CST
912 339580 : && TREE_CODE (s) == INTEGER_CST
913 1222469 : && wi::mod_trunc (wi::to_wide (c), wi::to_wide (s),
914 8199559 : TYPE_SIGN (type)) == 0)
915 8063708 : || (TYPE_OVERFLOW_UNDEFINED (type)
916 0 : && multiple_of_p (type, c, s)))
917 : {
918 : /* If C is an exact multiple of S, then its value will be reached before
919 : the induction variable overflows (unless the loop is exited in some
920 : other way before). Note that the actual induction variable in the
921 : loop (which ranges from base to final instead of from 0 to C) may
922 : overflow, in which case BNDS.up will not be giving a correct upper
923 : bound on C; thus, BNDS_U_VALID had to be computed in advance. */
924 : no_overflow = true;
925 : exit_must_be_taken = true;
926 : }
927 :
928 : /* If the induction variable can overflow, the number of iterations is at
929 : most the period of the control variable (or infinite, but in that case
930 : the whole # of iterations analysis will fail). */
931 543309 : if (!no_overflow)
932 : {
933 122916 : max = wi::mask <widest_int> (TYPE_PRECISION (type)
934 122916 : - wi::ctz (wi::to_wide (s)), false);
935 61458 : wi::to_mpz (max, bnd, UNSIGNED);
936 61458 : return;
937 : }
938 :
939 : /* Now we know that the induction variable does not overflow, so the loop
940 : iterates at most (range of type / S) times. */
941 7458941 : wi::to_mpz (wi::minus_one (TYPE_PRECISION (type)), bnd, UNSIGNED);
942 :
943 : /* If the induction variable is guaranteed to reach the value of C before
944 : overflow, ... */
945 7458941 : if (exit_must_be_taken)
946 : {
947 : /* ... then we can strengthen this to C / S, and possibly we can use
948 : the upper bound on C given by BNDS. */
949 7194924 : if (TREE_CODE (c) == INTEGER_CST)
950 6315203 : wi::to_mpz (wi::to_wide (c), bnd, UNSIGNED);
951 879721 : else if (bnds_u_valid)
952 744296 : mpz_set (bnd, bnds->up);
953 : }
954 :
955 7458941 : mpz_init (d);
956 7458941 : wi::to_mpz (wi::to_wide (s), d, UNSIGNED);
957 7458941 : mpz_fdiv_q (bnd, bnd, d);
958 7458941 : mpz_clear (d);
959 7520399 : }
960 :
961 : /* Determines number of iterations of loop whose ending condition
962 : is IV <> FINAL. TYPE is the type of the iv. The number of
963 : iterations is stored to NITER. EXIT_MUST_BE_TAKEN is true if
964 : we know that the exit must be taken eventually, i.e., that the IV
965 : ever reaches the value FINAL (we derived this earlier, and possibly set
966 : NITER->assumptions to make sure this is the case). BNDS contains the
967 : bounds on the difference FINAL - IV->base. */
968 :
969 : static bool
970 7520399 : number_of_iterations_ne (class loop *loop, tree type, affine_iv *iv,
971 : tree final, class tree_niter_desc *niter,
972 : bool exit_must_be_taken, bounds *bnds)
973 : {
974 7520399 : tree niter_type = unsigned_type_for (type);
975 7520399 : tree s, c, d, bits, assumption, tmp, bound;
976 :
977 7520399 : niter->control = *iv;
978 7520399 : niter->bound = final;
979 7520399 : niter->cmp = NE_EXPR;
980 :
981 : /* Rearrange the terms so that we get inequality S * i <> C, with S
982 : positive. Also cast everything to the unsigned type. If IV does
983 : not overflow, BNDS bounds the value of C. Also, this is the
984 : case if the computation |FINAL - IV->base| does not overflow, i.e.,
985 : if BNDS->below in the result is nonnegative. */
986 7520399 : if (tree_int_cst_sign_bit (iv->step))
987 : {
988 2545904 : s = fold_build1 (NEGATE_EXPR, niter_type,
989 : fold_convert (niter_type, iv->step));
990 2545904 : c = fold_build2 (MINUS_EXPR, niter_type,
991 : fold_convert (niter_type, iv->base),
992 : fold_convert (niter_type, final));
993 2545904 : bounds_negate (bnds);
994 : }
995 : else
996 : {
997 4974495 : s = fold_convert (niter_type, iv->step);
998 4974495 : c = fold_build2 (MINUS_EXPR, niter_type,
999 : fold_convert (niter_type, final),
1000 : fold_convert (niter_type, iv->base));
1001 : }
1002 :
1003 7520399 : auto_mpz max;
1004 7520399 : number_of_iterations_ne_max (max, iv->no_overflow, c, s, bnds,
1005 : exit_must_be_taken);
1006 15040798 : niter->max = widest_int::from (wi::from_mpz (niter_type, max, false),
1007 15040798 : TYPE_SIGN (niter_type));
1008 :
1009 : /* Compute no-overflow information for the control iv. This can be
1010 : proven when below two conditions are satisfied:
1011 :
1012 : 1) IV evaluates toward FINAL at beginning, i.e:
1013 : base <= FINAL ; step > 0
1014 : base >= FINAL ; step < 0
1015 :
1016 : 2) |FINAL - base| is an exact multiple of step.
1017 :
1018 : Unfortunately, it's hard to prove above conditions after pass loop-ch
1019 : because loop with exit condition (IV != FINAL) usually will be guarded
1020 : by initial-condition (IV.base - IV.step != FINAL). In this case, we
1021 : can alternatively try to prove below conditions:
1022 :
1023 : 1') IV evaluates toward FINAL at beginning, i.e:
1024 : new_base = base - step < FINAL ; step > 0
1025 : && base - step doesn't underflow
1026 : new_base = base - step > FINAL ; step < 0
1027 : && base - step doesn't overflow
1028 :
1029 : Please refer to PR34114 as an example of loop-ch's impact.
1030 :
1031 : Note, for NE_EXPR, base equals to FINAL is a special case, in
1032 : which the loop exits immediately, and the iv does not overflow.
1033 :
1034 : Also note, we prove condition 2) by checking base and final separately
1035 : along with condition 1) or 1'). Since we ensure the difference
1036 : computation of c does not wrap with cond below and the adjusted s
1037 : will fit a signed type as well as an unsigned we can safely do
1038 : this using the type of the IV if it is not pointer typed. */
1039 7520399 : tree mtype = type;
1040 7520399 : if (POINTER_TYPE_P (type))
1041 628581 : mtype = niter_type;
1042 7520399 : if (!niter->control.no_overflow
1043 7520399 : && (integer_onep (s)
1044 205159 : || (multiple_of_p (mtype, fold_convert (mtype, iv->base),
1045 205159 : fold_convert (mtype, s), false)
1046 32059 : && multiple_of_p (mtype, fold_convert (mtype, final),
1047 32059 : fold_convert (mtype, s), false))))
1048 : {
1049 2401131 : tree t, cond, relaxed_cond = boolean_false_node;
1050 :
1051 2401131 : if (tree_int_cst_sign_bit (iv->step))
1052 : {
1053 2300656 : cond = fold_build2 (GE_EXPR, boolean_type_node, iv->base, final);
1054 2300656 : if (INTEGRAL_NB_TYPE_P (type))
1055 : {
1056 : /* Only when base - step doesn't overflow. */
1057 2300656 : t = TYPE_MAX_VALUE (type);
1058 2300656 : t = fold_build2 (PLUS_EXPR, type, t, iv->step);
1059 2300656 : t = fold_build2 (GE_EXPR, boolean_type_node, t, iv->base);
1060 2300656 : if (integer_nonzerop (t))
1061 : {
1062 2139663 : t = fold_build2 (MINUS_EXPR, type, iv->base, iv->step);
1063 2139663 : relaxed_cond = fold_build2 (GT_EXPR, boolean_type_node, t,
1064 : final);
1065 : }
1066 : }
1067 : }
1068 : else
1069 : {
1070 100475 : cond = fold_build2 (LE_EXPR, boolean_type_node, iv->base, final);
1071 100475 : if (INTEGRAL_NB_TYPE_P (type))
1072 : {
1073 : /* Only when base - step doesn't underflow. */
1074 100475 : t = TYPE_MIN_VALUE (type);
1075 100475 : t = fold_build2 (PLUS_EXPR, type, t, iv->step);
1076 100475 : t = fold_build2 (LE_EXPR, boolean_type_node, t, iv->base);
1077 100475 : if (integer_nonzerop (t))
1078 : {
1079 62860 : t = fold_build2 (MINUS_EXPR, type, iv->base, iv->step);
1080 62860 : relaxed_cond = fold_build2 (LT_EXPR, boolean_type_node, t,
1081 : final);
1082 : }
1083 : }
1084 : }
1085 :
1086 2401131 : t = simplify_using_initial_conditions (loop, cond);
1087 2401131 : if (!t || !integer_onep (t))
1088 62918 : t = simplify_using_initial_conditions (loop, relaxed_cond);
1089 :
1090 2401131 : if (t && integer_onep (t))
1091 : {
1092 2338566 : niter->control.no_overflow = true;
1093 2338566 : niter->niter = fold_build2 (EXACT_DIV_EXPR, niter_type, c, s);
1094 2338566 : return true;
1095 : }
1096 : }
1097 :
1098 : /* Let nsd (step, size of mode) = d. If d does not divide c, the loop
1099 : is infinite. Otherwise, the number of iterations is
1100 : (inverse(s/d) * (c/d)) mod (size of mode/d). */
1101 5181833 : bits = num_ending_zeros (s);
1102 15545499 : bound = build_low_bits_mask (niter_type,
1103 5181833 : (TYPE_PRECISION (niter_type)
1104 5181833 : - tree_to_uhwi (bits)));
1105 :
1106 5181833 : d = fold_binary_to_constant (LSHIFT_EXPR, niter_type,
1107 : build_int_cst (niter_type, 1), bits);
1108 5181833 : s = fold_binary_to_constant (RSHIFT_EXPR, niter_type, s, bits);
1109 :
1110 5181833 : if (!exit_must_be_taken)
1111 : {
1112 : /* If we cannot assume that the exit is taken eventually, record the
1113 : assumptions for divisibility of c. */
1114 1882557 : assumption = fold_build2 (FLOOR_MOD_EXPR, niter_type, c, d);
1115 1882557 : assumption = fold_build2 (EQ_EXPR, boolean_type_node,
1116 : assumption, build_int_cst (niter_type, 0));
1117 1882557 : if (!integer_nonzerop (assumption))
1118 309747 : niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
1119 : niter->assumptions, assumption);
1120 : }
1121 :
1122 5181833 : c = fold_build2 (EXACT_DIV_EXPR, niter_type, c, d);
1123 5181833 : if (integer_onep (s))
1124 : {
1125 4998542 : niter->niter = c;
1126 : }
1127 : else
1128 : {
1129 183291 : tmp = fold_build2 (MULT_EXPR, niter_type, c, inverse (s, bound));
1130 183291 : niter->niter = fold_build2 (BIT_AND_EXPR, niter_type, tmp, bound);
1131 : }
1132 : return true;
1133 7520399 : }
1134 :
1135 : /* Checks whether we can determine the final value of the control variable
1136 : of the loop with ending condition IV0 < IV1 (computed in TYPE).
1137 : DELTA is the difference IV1->base - IV0->base, STEP is the absolute value
1138 : of the step. The assumptions necessary to ensure that the computation
1139 : of the final value does not overflow are recorded in NITER. If we
1140 : find the final value, we adjust DELTA and return TRUE. Otherwise
1141 : we return false. BNDS bounds the value of IV1->base - IV0->base,
1142 : and will be updated by the same amount as DELTA. EXIT_MUST_BE_TAKEN is
1143 : true if we know that the exit must be taken eventually. */
1144 :
1145 : static bool
1146 385882 : number_of_iterations_lt_to_ne (tree type, affine_iv *iv0, affine_iv *iv1,
1147 : class tree_niter_desc *niter,
1148 : tree *delta, tree step,
1149 : bool exit_must_be_taken, bounds *bnds)
1150 : {
1151 385882 : tree niter_type = TREE_TYPE (step);
1152 385882 : tree mod = fold_build2 (FLOOR_MOD_EXPR, niter_type, *delta, step);
1153 385882 : tree tmod;
1154 385882 : tree assumption = boolean_true_node, bound, noloop;
1155 385882 : bool fv_comp_no_overflow;
1156 385882 : tree type1 = type;
1157 385882 : if (POINTER_TYPE_P (type))
1158 128394 : type1 = sizetype;
1159 :
1160 385882 : if (TREE_CODE (mod) != INTEGER_CST)
1161 : return false;
1162 36278 : if (integer_nonzerop (mod))
1163 16458 : mod = fold_build2 (MINUS_EXPR, niter_type, step, mod);
1164 36278 : tmod = fold_convert (type1, mod);
1165 :
1166 36278 : auto_mpz mmod;
1167 36278 : wi::to_mpz (wi::to_wide (mod), mmod, UNSIGNED);
1168 36278 : mpz_neg (mmod, mmod);
1169 :
1170 : /* If the induction variable does not overflow and the exit is taken,
1171 : then the computation of the final value does not overflow. This is
1172 : also obviously the case if the new final value is equal to the
1173 : current one. Finally, we postulate this for pointer type variables,
1174 : as the code cannot rely on the object to that the pointer points being
1175 : placed at the end of the address space (and more pragmatically,
1176 : TYPE_{MIN,MAX}_VALUE is not defined for pointers). */
1177 36278 : if (integer_zerop (mod) || POINTER_TYPE_P (type))
1178 : fv_comp_no_overflow = true;
1179 15759 : else if (!exit_must_be_taken)
1180 : fv_comp_no_overflow = false;
1181 : else
1182 7846 : fv_comp_no_overflow =
1183 7846 : (iv0->no_overflow && integer_nonzerop (iv0->step))
1184 8774 : || (iv1->no_overflow && integer_nonzerop (iv1->step));
1185 :
1186 36278 : if (integer_nonzerop (iv0->step))
1187 : {
1188 : /* The final value of the iv is iv1->base + MOD, assuming that this
1189 : computation does not overflow, and that
1190 : iv0->base <= iv1->base + MOD. */
1191 31567 : if (!fv_comp_no_overflow)
1192 : {
1193 5969 : bound = fold_build2 (MINUS_EXPR, type1,
1194 : TYPE_MAX_VALUE (type1), tmod);
1195 5969 : assumption = fold_build2 (LE_EXPR, boolean_type_node,
1196 : iv1->base, bound);
1197 5969 : if (integer_zerop (assumption))
1198 : return false;
1199 : }
1200 : }
1201 : else
1202 : {
1203 : /* The final value of the iv is iv0->base - MOD, assuming that this
1204 : computation does not overflow, and that
1205 : iv0->base - MOD <= iv1->base. */
1206 4711 : if (!fv_comp_no_overflow)
1207 : {
1208 1944 : bound = fold_build2 (PLUS_EXPR, type1,
1209 : TYPE_MIN_VALUE (type1), tmod);
1210 1944 : assumption = fold_build2 (GE_EXPR, boolean_type_node,
1211 : iv0->base, bound);
1212 1944 : if (integer_zerop (assumption))
1213 : return false;
1214 : }
1215 : }
1216 :
1217 : /* IV0 < IV1 does not loop if IV0->base >= IV1->base. */
1218 36222 : if (fv_comp_no_overflow && mpz_cmp (mmod, bnds->below) < 0)
1219 17201 : noloop = boolean_false_node;
1220 : else
1221 19021 : noloop = fold_build2 (GE_EXPR, boolean_type_node,
1222 : iv0->base, iv1->base);
1223 :
1224 36222 : if (!integer_nonzerop (assumption))
1225 401 : niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
1226 : niter->assumptions,
1227 : assumption);
1228 36222 : if (!integer_zerop (noloop))
1229 3681 : niter->may_be_zero = fold_build2 (TRUTH_OR_EXPR, boolean_type_node,
1230 : niter->may_be_zero,
1231 : noloop);
1232 36222 : bounds_add (bnds, wi::to_widest (mod), type);
1233 36222 : *delta = fold_build2 (PLUS_EXPR, niter_type, *delta, mod);
1234 :
1235 36222 : return true;
1236 36278 : }
1237 :
1238 : /* Add assertions to NITER that ensure that the control variable of the loop
1239 : with ending condition IV0 < IV1 does not overflow. Types of IV0 and IV1
1240 : are TYPE. Returns false if we can prove that there is an overflow, true
1241 : otherwise. STEP is the absolute value of the step. */
1242 :
1243 : static bool
1244 349660 : assert_no_overflow_lt (tree type, affine_iv *iv0, affine_iv *iv1,
1245 : class tree_niter_desc *niter, tree step)
1246 : {
1247 349660 : tree bound, d, assumption, diff;
1248 349660 : tree niter_type = TREE_TYPE (step);
1249 :
1250 349660 : if (integer_nonzerop (iv0->step))
1251 : {
1252 : /* for (i = iv0->base; i < iv1->base; i += iv0->step) */
1253 286910 : if (iv0->no_overflow)
1254 : return true;
1255 :
1256 : /* If iv0->base is a constant, we can determine the last value before
1257 : overflow precisely; otherwise we conservatively assume
1258 : MAX - STEP + 1. */
1259 :
1260 46746 : if (TREE_CODE (iv0->base) == INTEGER_CST)
1261 : {
1262 15852 : d = fold_build2 (MINUS_EXPR, niter_type,
1263 : fold_convert (niter_type, TYPE_MAX_VALUE (type)),
1264 : fold_convert (niter_type, iv0->base));
1265 15852 : diff = fold_build2 (FLOOR_MOD_EXPR, niter_type, d, step);
1266 : }
1267 : else
1268 30894 : diff = fold_build2 (MINUS_EXPR, niter_type, step,
1269 : build_int_cst (niter_type, 1));
1270 46746 : bound = fold_build2 (MINUS_EXPR, type,
1271 : TYPE_MAX_VALUE (type), fold_convert (type, diff));
1272 46746 : assumption = fold_build2 (LE_EXPR, boolean_type_node,
1273 : iv1->base, bound);
1274 : }
1275 : else
1276 : {
1277 : /* for (i = iv1->base; i > iv0->base; i += iv1->step) */
1278 62750 : if (iv1->no_overflow)
1279 : return true;
1280 :
1281 31477 : if (TREE_CODE (iv1->base) == INTEGER_CST)
1282 : {
1283 56 : d = fold_build2 (MINUS_EXPR, niter_type,
1284 : fold_convert (niter_type, iv1->base),
1285 : fold_convert (niter_type, TYPE_MIN_VALUE (type)));
1286 56 : diff = fold_build2 (FLOOR_MOD_EXPR, niter_type, d, step);
1287 : }
1288 : else
1289 31421 : diff = fold_build2 (MINUS_EXPR, niter_type, step,
1290 : build_int_cst (niter_type, 1));
1291 31477 : bound = fold_build2 (PLUS_EXPR, type,
1292 : TYPE_MIN_VALUE (type), fold_convert (type, diff));
1293 31477 : assumption = fold_build2 (GE_EXPR, boolean_type_node,
1294 : iv0->base, bound);
1295 : }
1296 :
1297 78223 : if (integer_zerop (assumption))
1298 : return false;
1299 78157 : if (!integer_nonzerop (assumption))
1300 55805 : niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
1301 : niter->assumptions, assumption);
1302 :
1303 78157 : iv0->no_overflow = true;
1304 78157 : iv1->no_overflow = true;
1305 78157 : return true;
1306 : }
1307 :
1308 : /* Add an assumption to NITER that a loop whose ending condition
1309 : is IV0 < IV1 rolls. TYPE is the type of the control iv. BNDS
1310 : bounds the value of IV1->base - IV0->base. */
1311 :
1312 : static void
1313 349594 : assert_loop_rolls_lt (tree type, affine_iv *iv0, affine_iv *iv1,
1314 : class tree_niter_desc *niter, bounds *bnds)
1315 : {
1316 349594 : tree assumption = boolean_true_node, bound, diff;
1317 349594 : tree mbz, mbzl, mbzr, type1;
1318 349594 : bool rolls_p, no_overflow_p;
1319 349594 : widest_int dstep;
1320 349594 : mpz_t mstep, max;
1321 :
1322 : /* We are going to compute the number of iterations as
1323 : (iv1->base - iv0->base + step - 1) / step, computed in the unsigned
1324 : variant of TYPE. This formula only works if
1325 :
1326 : -step + 1 <= (iv1->base - iv0->base) <= MAX - step + 1
1327 :
1328 : (where MAX is the maximum value of the unsigned variant of TYPE, and
1329 : the computations in this formula are performed in full precision,
1330 : i.e., without overflows).
1331 :
1332 : Usually, for loops with exit condition iv0->base + step * i < iv1->base,
1333 : we have a condition of the form iv0->base - step < iv1->base before the loop,
1334 : and for loops iv0->base < iv1->base - step * i the condition
1335 : iv0->base < iv1->base + step, due to loop header copying, which enable us
1336 : to prove the lower bound.
1337 :
1338 : The upper bound is more complicated. Unless the expressions for initial
1339 : and final value themselves contain enough information, we usually cannot
1340 : derive it from the context. */
1341 :
1342 : /* First check whether the answer does not follow from the bounds we gathered
1343 : before. */
1344 349594 : if (integer_nonzerop (iv0->step))
1345 286908 : dstep = wi::to_widest (iv0->step);
1346 : else
1347 : {
1348 62686 : dstep = wi::sext (wi::to_widest (iv1->step), TYPE_PRECISION (type));
1349 62686 : dstep = -dstep;
1350 : }
1351 :
1352 349594 : mpz_init (mstep);
1353 349594 : wi::to_mpz (dstep, mstep, UNSIGNED);
1354 349594 : mpz_neg (mstep, mstep);
1355 349594 : mpz_add_ui (mstep, mstep, 1);
1356 :
1357 349594 : rolls_p = mpz_cmp (mstep, bnds->below) <= 0;
1358 :
1359 349594 : mpz_init (max);
1360 349594 : wi::to_mpz (wi::minus_one (TYPE_PRECISION (type)), max, UNSIGNED);
1361 349594 : mpz_add (max, max, mstep);
1362 699188 : no_overflow_p = (mpz_cmp (bnds->up, max) <= 0
1363 : /* For pointers, only values lying inside a single object
1364 : can be compared or manipulated by pointer arithmetics.
1365 : Gcc in general does not allow or handle objects larger
1366 : than half of the address space, hence the upper bound
1367 : is satisfied for pointers. */
1368 349594 : || POINTER_TYPE_P (type));
1369 349594 : mpz_clear (mstep);
1370 349594 : mpz_clear (max);
1371 :
1372 349594 : if (rolls_p && no_overflow_p)
1373 129958 : return;
1374 :
1375 219636 : type1 = type;
1376 219636 : if (POINTER_TYPE_P (type))
1377 67368 : type1 = sizetype;
1378 :
1379 : /* Now the hard part; we must formulate the assumption(s) as expressions, and
1380 : we must be careful not to introduce overflow. */
1381 :
1382 219636 : if (integer_nonzerop (iv0->step))
1383 : {
1384 178492 : diff = fold_build2 (MINUS_EXPR, type1,
1385 : iv0->step, build_int_cst (type1, 1));
1386 :
1387 : /* We need to know that iv0->base >= MIN + iv0->step - 1. Since
1388 : 0 address never belongs to any object, we can assume this for
1389 : pointers. */
1390 178492 : if (!POINTER_TYPE_P (type))
1391 : {
1392 114998 : bound = fold_build2 (PLUS_EXPR, type1,
1393 : TYPE_MIN_VALUE (type), diff);
1394 114998 : assumption = fold_build2 (GE_EXPR, boolean_type_node,
1395 : iv0->base, bound);
1396 : }
1397 :
1398 : /* And then we can compute iv0->base - diff, and compare it with
1399 : iv1->base. */
1400 178492 : mbzl = fold_build2 (MINUS_EXPR, type1,
1401 : fold_convert (type1, iv0->base), diff);
1402 178492 : mbzr = fold_convert (type1, iv1->base);
1403 : }
1404 : else
1405 : {
1406 41144 : diff = fold_build2 (PLUS_EXPR, type1,
1407 : iv1->step, build_int_cst (type1, 1));
1408 :
1409 41144 : if (!POINTER_TYPE_P (type))
1410 : {
1411 37270 : bound = fold_build2 (PLUS_EXPR, type1,
1412 : TYPE_MAX_VALUE (type), diff);
1413 37270 : assumption = fold_build2 (LE_EXPR, boolean_type_node,
1414 : iv1->base, bound);
1415 : }
1416 :
1417 41144 : mbzl = fold_convert (type1, iv0->base);
1418 41144 : mbzr = fold_build2 (MINUS_EXPR, type1,
1419 : fold_convert (type1, iv1->base), diff);
1420 : }
1421 :
1422 219636 : if (!integer_nonzerop (assumption))
1423 93442 : niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
1424 : niter->assumptions, assumption);
1425 219636 : if (!rolls_p)
1426 : {
1427 203396 : mbz = fold_build2 (GT_EXPR, boolean_type_node, mbzl, mbzr);
1428 203396 : niter->may_be_zero = fold_build2 (TRUTH_OR_EXPR, boolean_type_node,
1429 : niter->may_be_zero, mbz);
1430 : }
1431 349594 : }
1432 :
1433 : /* Determines number of iterations of loop whose ending condition
1434 : is IV0 < IV1 which likes: {base, -C} < n, or n < {base, C}.
1435 : The number of iterations is stored to NITER. */
1436 :
1437 : static bool
1438 67517 : number_of_iterations_until_wrap (class loop *loop, tree type, affine_iv *iv0,
1439 : affine_iv *iv1, class tree_niter_desc *niter)
1440 : {
1441 67517 : tree niter_type = unsigned_type_for (type);
1442 67517 : tree step, num, assumptions, may_be_zero, span;
1443 67517 : wide_int high, low, max, min;
1444 :
1445 67517 : may_be_zero = fold_build2 (LE_EXPR, boolean_type_node, iv1->base, iv0->base);
1446 67517 : if (integer_onep (may_be_zero))
1447 : return false;
1448 :
1449 67303 : int prec = TYPE_PRECISION (type);
1450 67303 : signop sgn = TYPE_SIGN (type);
1451 67303 : min = wi::min_value (prec, sgn);
1452 67303 : max = wi::max_value (prec, sgn);
1453 :
1454 : /* n < {base, C}. */
1455 67303 : if (integer_zerop (iv0->step) && !tree_int_cst_sign_bit (iv1->step))
1456 : {
1457 : /* MIN + C - 1 <= n. */
1458 42329 : tree last = wide_int_to_tree (type, min + wi::to_wide (iv1->step) - 1);
1459 42329 : assumptions = fold_build2 (LE_EXPR, boolean_type_node, last, iv0->base);
1460 42329 : if (integer_zerop (assumptions))
1461 : return false;
1462 :
1463 42327 : step = fold_convert (niter_type, iv1->step);
1464 42327 : num = fold_build2 (MINUS_EXPR, niter_type,
1465 : wide_int_to_tree (niter_type, max),
1466 : fold_convert (niter_type, iv1->base));
1467 :
1468 : /* When base has the form iv + 1, if we know iv >= n, then iv + 1 < n
1469 : only when iv + 1 overflows, i.e. when iv == TYPE_VALUE_MAX. */
1470 42327 : if (sgn == UNSIGNED
1471 6750 : && integer_onep (step)
1472 1113 : && TREE_CODE (iv1->base) == PLUS_EXPR
1473 42994 : && integer_onep (TREE_OPERAND (iv1->base, 1)))
1474 : {
1475 484 : tree cond = fold_build2 (GE_EXPR, boolean_type_node,
1476 : TREE_OPERAND (iv1->base, 0), iv0->base);
1477 484 : cond = simplify_using_initial_conditions (loop, cond);
1478 484 : if (integer_onep (cond))
1479 18 : may_be_zero = fold_build2 (EQ_EXPR, boolean_type_node,
1480 : TREE_OPERAND (iv1->base, 0),
1481 : TYPE_MAX_VALUE (type));
1482 : }
1483 :
1484 42327 : high = max;
1485 42327 : if (TREE_CODE (iv1->base) == INTEGER_CST)
1486 28936 : low = wi::to_wide (iv1->base) - 1;
1487 13391 : else if (TREE_CODE (iv0->base) == INTEGER_CST)
1488 6347 : low = wi::to_wide (iv0->base);
1489 : else
1490 7044 : low = min;
1491 : }
1492 : /* {base, -C} < n. */
1493 24974 : else if (tree_int_cst_sign_bit (iv0->step) && integer_zerop (iv1->step))
1494 : {
1495 : /* MAX + (-C) + 1 >= n. */
1496 24974 : tree last = wide_int_to_tree (type, max + wi::to_wide (iv0->step) + 1);
1497 24974 : assumptions = fold_build2 (GE_EXPR, boolean_type_node, last, iv1->base);
1498 24974 : if (integer_zerop (assumptions))
1499 : return false;
1500 :
1501 24972 : step = fold_build1 (NEGATE_EXPR, niter_type,
1502 : fold_convert (niter_type, iv0->step));
1503 24972 : num = fold_build2 (MINUS_EXPR, niter_type,
1504 : fold_convert (niter_type, iv0->base),
1505 : wide_int_to_tree (niter_type, min));
1506 24972 : low = min;
1507 24972 : if (TREE_CODE (iv0->base) == INTEGER_CST)
1508 1332 : high = wi::to_wide (iv0->base) + 1;
1509 23640 : else if (TREE_CODE (iv1->base) == INTEGER_CST)
1510 2353 : high = wi::to_wide (iv1->base);
1511 : else
1512 21287 : high = max;
1513 : }
1514 : else
1515 : return false;
1516 :
1517 : /* (delta + step - 1) / step */
1518 67299 : num = fold_build2 (PLUS_EXPR, niter_type, num, step);
1519 67299 : niter->niter = fold_build2 (FLOOR_DIV_EXPR, niter_type, num, step);
1520 :
1521 67299 : widest_int delta, s;
1522 67299 : delta = widest_int::from (high, sgn) - widest_int::from (low, sgn);
1523 67299 : s = wi::to_widest (step);
1524 67299 : delta = delta + s - 1;
1525 67299 : niter->max = wi::udiv_floor (delta, s);
1526 :
1527 67299 : niter->may_be_zero = may_be_zero;
1528 :
1529 67299 : if (!integer_nonzerop (assumptions))
1530 9269 : niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
1531 : niter->assumptions, assumptions);
1532 :
1533 67299 : niter->control.no_overflow = false;
1534 :
1535 : /* Update bound and exit condition as:
1536 : bound = niter * STEP + (IVbase - STEP).
1537 : { IVbase - STEP, +, STEP } != bound
1538 : Here, biasing IVbase by 1 step makes 'bound' be the value before wrap.
1539 : */
1540 67299 : tree base_type = TREE_TYPE (niter->control.base);
1541 67299 : if (POINTER_TYPE_P (base_type))
1542 : {
1543 6661 : tree utype = unsigned_type_for (base_type);
1544 6661 : niter->control.base
1545 6661 : = fold_build2 (MINUS_EXPR, utype,
1546 : fold_convert (utype, niter->control.base),
1547 : fold_convert (utype, niter->control.step));
1548 6661 : niter->control.base = fold_convert (base_type, niter->control.base);
1549 : }
1550 : else
1551 60638 : niter->control.base
1552 60638 : = fold_build2 (MINUS_EXPR, base_type, niter->control.base,
1553 : niter->control.step);
1554 :
1555 67299 : span = fold_build2 (MULT_EXPR, niter_type, niter->niter,
1556 : fold_convert (niter_type, niter->control.step));
1557 67299 : niter->bound = fold_build2 (PLUS_EXPR, niter_type, span,
1558 : fold_convert (niter_type, niter->control.base));
1559 67299 : niter->bound = fold_convert (type, niter->bound);
1560 67299 : niter->cmp = NE_EXPR;
1561 :
1562 67299 : return true;
1563 67517 : }
1564 :
1565 : /* Determines number of iterations of loop whose ending condition
1566 : is IV0 < IV1. TYPE is the type of the iv. The number of
1567 : iterations is stored to NITER. BNDS bounds the difference
1568 : IV1->base - IV0->base. EXIT_MUST_BE_TAKEN is true if we know
1569 : that the exit must be taken eventually. */
1570 :
1571 : static bool
1572 5377442 : number_of_iterations_lt (class loop *loop, tree type, affine_iv *iv0,
1573 : affine_iv *iv1, class tree_niter_desc *niter,
1574 : bool exit_must_be_taken, bounds *bnds)
1575 : {
1576 5377442 : tree niter_type = unsigned_type_for (type);
1577 5377442 : tree delta, step, s;
1578 5377442 : mpz_t mstep, tmp;
1579 :
1580 5377442 : if (integer_nonzerop (iv0->step))
1581 : {
1582 5044083 : niter->control = *iv0;
1583 5044083 : niter->cmp = LT_EXPR;
1584 5044083 : niter->bound = iv1->base;
1585 : }
1586 : else
1587 : {
1588 333359 : niter->control = *iv1;
1589 333359 : niter->cmp = GT_EXPR;
1590 333359 : niter->bound = iv0->base;
1591 : }
1592 :
1593 : /* {base, -C} < n, or n < {base, C} */
1594 5377442 : if (tree_int_cst_sign_bit (iv0->step)
1595 5377442 : || (!integer_zerop (iv1->step) && !tree_int_cst_sign_bit (iv1->step)))
1596 67517 : return number_of_iterations_until_wrap (loop, type, iv0, iv1, niter);
1597 :
1598 5309925 : delta = fold_build2 (MINUS_EXPR, niter_type,
1599 : fold_convert (niter_type, iv1->base),
1600 : fold_convert (niter_type, iv0->base));
1601 :
1602 : /* First handle the special case that the step is +-1. */
1603 10010557 : if ((integer_onep (iv0->step) && integer_zerop (iv1->step))
1604 5309925 : || (integer_all_onesp (iv1->step) && integer_zerop (iv0->step)))
1605 : {
1606 : /* for (i = iv0->base; i < iv1->base; i++)
1607 :
1608 : or
1609 :
1610 : for (i = iv1->base; i > iv0->base; i--).
1611 :
1612 : In both cases # of iterations is iv1->base - iv0->base, assuming that
1613 : iv1->base >= iv0->base.
1614 :
1615 : First try to derive a lower bound on the value of
1616 : iv1->base - iv0->base, computed in full precision. If the difference
1617 : is nonnegative, we are done, otherwise we must record the
1618 : condition. */
1619 :
1620 4924043 : if (mpz_sgn (bnds->below) < 0)
1621 1843521 : niter->may_be_zero = fold_build2 (LT_EXPR, boolean_type_node,
1622 : iv1->base, iv0->base);
1623 4924043 : niter->niter = delta;
1624 9848086 : niter->max = widest_int::from (wi::from_mpz (niter_type, bnds->up, false),
1625 9848086 : TYPE_SIGN (niter_type));
1626 4924043 : niter->control.no_overflow = true;
1627 4924043 : return true;
1628 : }
1629 :
1630 385882 : if (integer_nonzerop (iv0->step))
1631 318477 : step = fold_convert (niter_type, iv0->step);
1632 : else
1633 67405 : step = fold_build1 (NEGATE_EXPR, niter_type,
1634 : fold_convert (niter_type, iv1->step));
1635 :
1636 : /* If we can determine the final value of the control iv exactly, we can
1637 : transform the condition to != comparison. In particular, this will be
1638 : the case if DELTA is constant. */
1639 385882 : if (number_of_iterations_lt_to_ne (type, iv0, iv1, niter, &delta, step,
1640 : exit_must_be_taken, bnds))
1641 : {
1642 36222 : affine_iv zps;
1643 :
1644 36222 : zps.base = build_int_cst (niter_type, 0);
1645 36222 : zps.step = step;
1646 : /* number_of_iterations_lt_to_ne will add assumptions that ensure that
1647 : zps does not overflow. */
1648 36222 : zps.no_overflow = true;
1649 :
1650 36222 : return number_of_iterations_ne (loop, type, &zps,
1651 : delta, niter, true, bnds);
1652 : }
1653 :
1654 : /* Make sure that the control iv does not overflow. */
1655 349660 : if (!assert_no_overflow_lt (type, iv0, iv1, niter, step))
1656 : return false;
1657 :
1658 : /* We determine the number of iterations as (delta + step - 1) / step. For
1659 : this to work, we must know that iv1->base >= iv0->base - step + 1,
1660 : otherwise the loop does not roll. */
1661 349594 : assert_loop_rolls_lt (type, iv0, iv1, niter, bnds);
1662 :
1663 349594 : s = fold_build2 (MINUS_EXPR, niter_type,
1664 : step, build_int_cst (niter_type, 1));
1665 349594 : delta = fold_build2 (PLUS_EXPR, niter_type, delta, s);
1666 349594 : niter->niter = fold_build2 (FLOOR_DIV_EXPR, niter_type, delta, step);
1667 :
1668 349594 : mpz_init (mstep);
1669 349594 : mpz_init (tmp);
1670 349594 : wi::to_mpz (wi::to_wide (step), mstep, UNSIGNED);
1671 349594 : mpz_add (tmp, bnds->up, mstep);
1672 349594 : mpz_sub_ui (tmp, tmp, 1);
1673 349594 : mpz_fdiv_q (tmp, tmp, mstep);
1674 699188 : niter->max = widest_int::from (wi::from_mpz (niter_type, tmp, false),
1675 699188 : TYPE_SIGN (niter_type));
1676 349594 : mpz_clear (mstep);
1677 349594 : mpz_clear (tmp);
1678 :
1679 349594 : return true;
1680 : }
1681 :
1682 : /* Determines number of iterations of loop whose ending condition
1683 : is IV0 <= IV1. TYPE is the type of the iv. The number of
1684 : iterations is stored to NITER. EXIT_MUST_BE_TAKEN is true if
1685 : we know that this condition must eventually become false (we derived this
1686 : earlier, and possibly set NITER->assumptions to make sure this
1687 : is the case). BNDS bounds the difference IV1->base - IV0->base. */
1688 :
1689 : static bool
1690 2047443 : number_of_iterations_le (class loop *loop, tree type, affine_iv *iv0,
1691 : affine_iv *iv1, class tree_niter_desc *niter,
1692 : bool exit_must_be_taken, bounds *bnds)
1693 : {
1694 2047443 : tree assumption;
1695 2047443 : tree type1 = type;
1696 2047443 : if (POINTER_TYPE_P (type))
1697 6631 : type1 = sizetype;
1698 :
1699 : /* Say that IV0 is the control variable. Then IV0 <= IV1 iff
1700 : IV0 < IV1 + 1, assuming that IV1 is not equal to the greatest
1701 : value of the type. This we must know anyway, since if it is
1702 : equal to this value, the loop rolls forever. We do not check
1703 : this condition for pointer type ivs, as the code cannot rely on
1704 : the object to that the pointer points being placed at the end of
1705 : the address space (and more pragmatically, TYPE_{MIN,MAX}_VALUE is
1706 : not defined for pointers). */
1707 :
1708 2047443 : if (!exit_must_be_taken && !POINTER_TYPE_P (type))
1709 : {
1710 643733 : if (integer_nonzerop (iv0->step))
1711 579389 : assumption = fold_build2 (NE_EXPR, boolean_type_node,
1712 : iv1->base, TYPE_MAX_VALUE (type));
1713 : else
1714 64344 : assumption = fold_build2 (NE_EXPR, boolean_type_node,
1715 : iv0->base, TYPE_MIN_VALUE (type));
1716 :
1717 643733 : if (integer_zerop (assumption))
1718 : return false;
1719 643703 : if (!integer_nonzerop (assumption))
1720 252995 : niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
1721 : niter->assumptions, assumption);
1722 : }
1723 :
1724 2047413 : if (integer_nonzerop (iv0->step))
1725 : {
1726 1949853 : if (POINTER_TYPE_P (type))
1727 1245 : iv1->base = fold_build_pointer_plus_hwi (iv1->base, 1);
1728 : else
1729 1948608 : iv1->base = fold_build2 (PLUS_EXPR, type1, iv1->base,
1730 : build_int_cst (type1, 1));
1731 : }
1732 97560 : else if (POINTER_TYPE_P (type))
1733 5386 : iv0->base = fold_build_pointer_plus_hwi (iv0->base, -1);
1734 : else
1735 92174 : iv0->base = fold_build2 (MINUS_EXPR, type1,
1736 : iv0->base, build_int_cst (type1, 1));
1737 :
1738 2047413 : bounds_add (bnds, 1, type1);
1739 :
1740 2047413 : return number_of_iterations_lt (loop, type, iv0, iv1, niter, exit_must_be_taken,
1741 2047413 : bnds);
1742 : }
1743 :
1744 : /* Dumps description of affine induction variable IV to FILE. */
1745 :
1746 : static void
1747 87034 : dump_affine_iv (FILE *file, affine_iv *iv)
1748 : {
1749 87034 : if (!integer_zerop (iv->step))
1750 43517 : fprintf (file, "[");
1751 :
1752 87034 : print_generic_expr (file, iv->base, TDF_SLIM);
1753 :
1754 87034 : if (!integer_zerop (iv->step))
1755 : {
1756 43517 : fprintf (file, ", + , ");
1757 43517 : print_generic_expr (file, iv->step, TDF_SLIM);
1758 77822 : fprintf (file, "]%s", iv->no_overflow ? "(no_overflow)" : "");
1759 : }
1760 87034 : }
1761 :
1762 : DEBUG_FUNCTION void
1763 0 : debug (affine_iv *iv)
1764 : {
1765 0 : dump_affine_iv (stderr, iv);
1766 0 : fputc ('\n', stderr);
1767 0 : }
1768 :
1769 : /* Determine the number of iterations according to condition (for staying
1770 : inside loop) which compares two induction variables using comparison
1771 : operator CODE. The induction variable on left side of the comparison
1772 : is IV0, the right-hand side is IV1. Both induction variables must have
1773 : type TYPE, which must be an integer or pointer type. The steps of the
1774 : ivs must be constants (or NULL_TREE, which is interpreted as constant zero).
1775 :
1776 : LOOP is the loop whose number of iterations we are determining.
1777 :
1778 : ONLY_EXIT is true if we are sure this is the only way the loop could be
1779 : exited (including possibly non-returning function calls, exceptions, etc.)
1780 : -- in this case we can use the information whether the control induction
1781 : variables can overflow or not in a more efficient way.
1782 :
1783 : if EVERY_ITERATION is true, we know the test is executed on every iteration.
1784 :
1785 : The results (number of iterations and assumptions as described in
1786 : comments at class tree_niter_desc in tree-ssa-loop.h) are stored to NITER.
1787 : Returns false if it fails to determine number of iterations, true if it
1788 : was determined (possibly with some assumptions). */
1789 :
1790 : static bool
1791 13085692 : number_of_iterations_cond (class loop *loop,
1792 : tree type, affine_iv *iv0, enum tree_code code,
1793 : affine_iv *iv1, class tree_niter_desc *niter,
1794 : bool only_exit, bool every_iteration)
1795 : {
1796 13085692 : bool exit_must_be_taken = false, ret;
1797 13085692 : bounds bnds;
1798 :
1799 : /* If the test is not executed every iteration, wrapping may make the test
1800 : to pass again.
1801 : TODO: the overflow case can be still used as unreliable estimate of upper
1802 : bound. But we have no API to pass it down to number of iterations code
1803 : and, at present, it will not use it anyway. */
1804 13085692 : if (!every_iteration
1805 67493 : && (!iv0->no_overflow || !iv1->no_overflow
1806 46004 : || code == NE_EXPR || code == EQ_EXPR))
1807 : return false;
1808 :
1809 : /* The meaning of these assumptions is this:
1810 : if !assumptions
1811 : then the rest of information does not have to be valid
1812 : if may_be_zero then the loop does not roll, even if
1813 : niter != 0. */
1814 13040133 : niter->assumptions = boolean_true_node;
1815 13040133 : niter->may_be_zero = boolean_false_node;
1816 13040133 : niter->niter = NULL_TREE;
1817 13040133 : niter->max = 0;
1818 13040133 : niter->bound = NULL_TREE;
1819 13040133 : niter->cmp = ERROR_MARK;
1820 :
1821 : /* Make < comparison from > ones, and for NE_EXPR comparisons, ensure that
1822 : the control variable is on lhs. */
1823 13040133 : if (code == GE_EXPR || code == GT_EXPR
1824 13040133 : || (code == NE_EXPR && integer_zerop (iv0->step)))
1825 : {
1826 3030096 : std::swap (iv0, iv1);
1827 3030096 : code = swap_tree_comparison (code);
1828 : }
1829 :
1830 13040133 : if (POINTER_TYPE_P (type))
1831 : {
1832 : /* Comparison of pointers is undefined unless both iv0 and iv1 point
1833 : to the same object. If they do, the control variable cannot wrap
1834 : (as wrap around the bounds of memory will never return a pointer
1835 : that would be guaranteed to point to the same object, even if we
1836 : avoid undefined behavior by casting to size_t and back). */
1837 815241 : iv0->no_overflow = true;
1838 815241 : iv1->no_overflow = true;
1839 : }
1840 :
1841 : /* If the control induction variable does not overflow and the only exit
1842 : from the loop is the one that we analyze, we know it must be taken
1843 : eventually. */
1844 13040133 : if (only_exit)
1845 : {
1846 8484497 : if (!integer_zerop (iv0->step) && iv0->no_overflow)
1847 : exit_must_be_taken = true;
1848 2150455 : else if (!integer_zerop (iv1->step) && iv1->no_overflow)
1849 6492094 : exit_must_be_taken = true;
1850 : }
1851 :
1852 : /* We can handle cases which neither of the sides of the comparison is
1853 : invariant:
1854 :
1855 : {iv0.base, iv0.step} cmp_code {iv1.base, iv1.step}
1856 : as if:
1857 : {iv0.base, iv0.step - iv1.step} cmp_code {iv1.base, 0}
1858 :
1859 : provided that either below condition is satisfied:
1860 :
1861 : a) the test is NE_EXPR;
1862 : b) iv0 and iv1 do not overflow and iv0.step - iv1.step is of
1863 : the same sign and of less or equal magnitude than iv0.step
1864 :
1865 : This rarely occurs in practice, but it is simple enough to manage. */
1866 13040133 : if (!integer_zerop (iv0->step) && !integer_zerop (iv1->step))
1867 : {
1868 5558 : tree step_type = POINTER_TYPE_P (type) ? sizetype : type;
1869 5558 : tree step = fold_binary_to_constant (MINUS_EXPR, step_type,
1870 : iv0->step, iv1->step);
1871 :
1872 : /* For code other than NE_EXPR we have to ensure moving the evolution
1873 : of IV1 to that of IV0 does not introduce overflow. */
1874 5558 : if (TREE_CODE (step) != INTEGER_CST
1875 5558 : || !iv0->no_overflow || !iv1->no_overflow)
1876 : {
1877 1284 : if (code != NE_EXPR)
1878 : return false;
1879 0 : iv0->no_overflow = false;
1880 : }
1881 : /* If the new step of IV0 has changed sign or is of greater
1882 : magnitude then we do not know whether IV0 does overflow
1883 : and thus the transform is not valid for code other than NE_EXPR. */
1884 4274 : else if (tree_int_cst_sign_bit (step) != tree_int_cst_sign_bit (iv0->step)
1885 7952 : || wi::gtu_p (wi::abs (wi::to_widest (step)),
1886 11630 : wi::abs (wi::to_widest (iv0->step))))
1887 : {
1888 2949 : if (POINTER_TYPE_P (type) && code != NE_EXPR)
1889 : /* For relational pointer compares we have further guarantees
1890 : that the pointers always point to the same object (or one
1891 : after it) and that objects do not cross the zero page. So
1892 : not only is the transform always valid for relational
1893 : pointer compares, we also know the resulting IV does not
1894 : overflow. */
1895 : ;
1896 813 : else if (code != NE_EXPR)
1897 : return false;
1898 : else
1899 408 : iv0->no_overflow = false;
1900 : }
1901 :
1902 3461 : iv0->step = step;
1903 3461 : iv1->step = build_int_cst (step_type, 0);
1904 3461 : iv1->no_overflow = true;
1905 : }
1906 :
1907 : /* If the result of the comparison is a constant, the loop is weird. More
1908 : precise handling would be possible, but the situation is not common enough
1909 : to waste time on it. */
1910 13038036 : if (integer_zerop (iv0->step) && integer_zerop (iv1->step))
1911 : return false;
1912 :
1913 : /* If the loop exits immediately, there is nothing to do. */
1914 12943318 : tree tem = fold_binary (code, boolean_type_node, iv0->base, iv1->base);
1915 12943318 : if (tem && integer_zerop (tem))
1916 : {
1917 81669 : if (!every_iteration)
1918 : return false;
1919 81606 : niter->niter = build_int_cst (unsigned_type_for (type), 0);
1920 81606 : niter->max = 0;
1921 81606 : return true;
1922 : }
1923 :
1924 : /* OK, now we know we have a senseful loop. Handle several cases, depending
1925 : on what comparison operator is used. */
1926 12861649 : bound_difference (loop, iv1->base, iv0->base, &bnds);
1927 :
1928 12861649 : if (dump_file && (dump_flags & TDF_DETAILS))
1929 : {
1930 43517 : fprintf (dump_file,
1931 : "Analyzing # of iterations of loop %d\n", loop->num);
1932 :
1933 43517 : fprintf (dump_file, " exit condition ");
1934 43517 : dump_affine_iv (dump_file, iv0);
1935 51967 : fprintf (dump_file, " %s ",
1936 : code == NE_EXPR ? "!="
1937 8450 : : code == LT_EXPR ? "<"
1938 : : "<=");
1939 43517 : dump_affine_iv (dump_file, iv1);
1940 43517 : fprintf (dump_file, "\n");
1941 :
1942 43517 : fprintf (dump_file, " bounds on difference of bases: ");
1943 43517 : mpz_out_str (dump_file, 10, bnds.below);
1944 43517 : fprintf (dump_file, " ... ");
1945 43517 : mpz_out_str (dump_file, 10, bnds.up);
1946 43517 : fprintf (dump_file, "\n");
1947 : }
1948 :
1949 12861649 : switch (code)
1950 : {
1951 7484177 : case NE_EXPR:
1952 7484177 : gcc_assert (integer_zerop (iv1->step));
1953 7484177 : ret = number_of_iterations_ne (loop, type, iv0, iv1->base, niter,
1954 : exit_must_be_taken, &bnds);
1955 7484177 : break;
1956 :
1957 3330029 : case LT_EXPR:
1958 3330029 : ret = number_of_iterations_lt (loop, type, iv0, iv1, niter,
1959 : exit_must_be_taken, &bnds);
1960 3330029 : break;
1961 :
1962 2047443 : case LE_EXPR:
1963 2047443 : ret = number_of_iterations_le (loop, type, iv0, iv1, niter,
1964 : exit_must_be_taken, &bnds);
1965 2047443 : break;
1966 :
1967 0 : default:
1968 0 : gcc_unreachable ();
1969 : }
1970 :
1971 12861649 : mpz_clear (bnds.up);
1972 12861649 : mpz_clear (bnds.below);
1973 :
1974 12861649 : if (dump_file && (dump_flags & TDF_DETAILS))
1975 : {
1976 43517 : if (ret)
1977 : {
1978 43517 : fprintf (dump_file, " result:\n");
1979 43517 : if (!integer_nonzerop (niter->assumptions))
1980 : {
1981 239 : fprintf (dump_file, " under assumptions ");
1982 239 : print_generic_expr (dump_file, niter->assumptions, TDF_SLIM);
1983 239 : fprintf (dump_file, "\n");
1984 : }
1985 :
1986 43517 : if (!integer_zerop (niter->may_be_zero))
1987 : {
1988 696 : fprintf (dump_file, " zero if ");
1989 696 : print_generic_expr (dump_file, niter->may_be_zero, TDF_SLIM);
1990 696 : fprintf (dump_file, "\n");
1991 : }
1992 :
1993 43517 : fprintf (dump_file, " # of iterations ");
1994 43517 : print_generic_expr (dump_file, niter->niter, TDF_SLIM);
1995 43517 : fprintf (dump_file, ", bounded by ");
1996 43517 : print_decu (niter->max, dump_file);
1997 43517 : fprintf (dump_file, "\n");
1998 : }
1999 : else
2000 0 : fprintf (dump_file, " failed\n\n");
2001 : }
2002 : return ret;
2003 : }
2004 :
2005 : /* Return an expression that computes the popcount of src. */
2006 :
2007 : static tree
2008 3587 : build_popcount_expr (tree src)
2009 : {
2010 3587 : tree fn;
2011 3587 : bool use_ifn = false;
2012 3587 : int prec = TYPE_PRECISION (TREE_TYPE (src));
2013 3587 : int i_prec = TYPE_PRECISION (integer_type_node);
2014 3587 : int li_prec = TYPE_PRECISION (long_integer_type_node);
2015 3587 : int lli_prec = TYPE_PRECISION (long_long_integer_type_node);
2016 :
2017 3587 : tree utype = unsigned_type_for (TREE_TYPE (src));
2018 3587 : src = fold_convert (utype, src);
2019 :
2020 3587 : if (direct_internal_fn_supported_p (IFN_POPCOUNT, utype, OPTIMIZE_FOR_BOTH))
2021 : use_ifn = true;
2022 3536 : else if (prec <= i_prec)
2023 3460 : fn = builtin_decl_implicit (BUILT_IN_POPCOUNT);
2024 76 : else if (prec == li_prec)
2025 59 : fn = builtin_decl_implicit (BUILT_IN_POPCOUNTL);
2026 17 : else if (prec == lli_prec || prec == 2 * lli_prec)
2027 17 : fn = builtin_decl_implicit (BUILT_IN_POPCOUNTLL);
2028 : else
2029 : return NULL_TREE;
2030 :
2031 3587 : tree call;
2032 3536 : if (use_ifn)
2033 51 : call = build_call_expr_internal_loc (UNKNOWN_LOCATION, IFN_POPCOUNT,
2034 : integer_type_node, 1, src);
2035 3536 : else if (prec == 2 * lli_prec)
2036 : {
2037 17 : tree src1 = fold_convert (long_long_unsigned_type_node,
2038 : fold_build2 (RSHIFT_EXPR, TREE_TYPE (src),
2039 : unshare_expr (src),
2040 : build_int_cst (integer_type_node,
2041 : lli_prec)));
2042 17 : tree src2 = fold_convert (long_long_unsigned_type_node, src);
2043 17 : tree call1 = build_call_expr (fn, 1, src1);
2044 17 : tree call2 = build_call_expr (fn, 1, src2);
2045 17 : call = fold_build2 (PLUS_EXPR, integer_type_node, call1, call2);
2046 : }
2047 : else
2048 : {
2049 3519 : if (prec < i_prec)
2050 293 : src = fold_convert (unsigned_type_node, src);
2051 :
2052 3519 : call = build_call_expr (fn, 1, src);
2053 : }
2054 :
2055 : return call;
2056 : }
2057 :
2058 : /* Utility function to check if OP is defined by a stmt
2059 : that is a val - 1. */
2060 :
2061 : static bool
2062 135929 : ssa_defined_by_minus_one_stmt_p (tree op, tree val)
2063 : {
2064 135929 : gimple *stmt;
2065 135929 : return (TREE_CODE (op) == SSA_NAME
2066 87894 : && (stmt = SSA_NAME_DEF_STMT (op))
2067 87894 : && is_gimple_assign (stmt)
2068 74942 : && (gimple_assign_rhs_code (stmt) == PLUS_EXPR)
2069 10294 : && val == gimple_assign_rhs1 (stmt)
2070 139766 : && integer_minus_onep (gimple_assign_rhs2 (stmt)));
2071 : }
2072 :
2073 : /* See comment below for number_of_iterations_bitcount.
2074 : For popcount, we have:
2075 :
2076 : modify:
2077 : _1 = iv_1 + -1
2078 : iv_2 = iv_1 & _1
2079 :
2080 : test:
2081 : if (iv != 0)
2082 :
2083 : modification count:
2084 : popcount (src)
2085 :
2086 : */
2087 :
2088 : static bool
2089 3897926 : number_of_iterations_popcount (loop_p loop, edge exit,
2090 : enum tree_code code,
2091 : class tree_niter_desc *niter)
2092 : {
2093 3897926 : bool modify_before_test = true;
2094 3897926 : HOST_WIDE_INT max;
2095 :
2096 : /* Check that condition for staying inside the loop is like
2097 : if (iv != 0). */
2098 7795852 : gcond *cond_stmt = safe_dyn_cast <gcond *> (*gsi_last_bb (exit->src));
2099 3897926 : if (!cond_stmt
2100 3897926 : || code != NE_EXPR
2101 1897195 : || !integer_zerop (gimple_cond_rhs (cond_stmt))
2102 5255860 : || TREE_CODE (gimple_cond_lhs (cond_stmt)) != SSA_NAME)
2103 : return false;
2104 :
2105 1357934 : tree iv_2 = gimple_cond_lhs (cond_stmt);
2106 1357934 : gimple *iv_2_stmt = SSA_NAME_DEF_STMT (iv_2);
2107 :
2108 : /* If the test comes before the iv modification, then these will actually be
2109 : iv_1 and a phi node. */
2110 1357934 : if (gimple_code (iv_2_stmt) == GIMPLE_PHI
2111 350903 : && gimple_bb (iv_2_stmt) == loop->header
2112 273107 : && gimple_phi_num_args (iv_2_stmt) == 2
2113 1631041 : && (TREE_CODE (gimple_phi_arg_def (iv_2_stmt,
2114 : loop_latch_edge (loop)->dest_idx))
2115 : == SSA_NAME))
2116 : {
2117 : /* iv_2 is actually one of the inputs to the phi. */
2118 268398 : iv_2 = gimple_phi_arg_def (iv_2_stmt, loop_latch_edge (loop)->dest_idx);
2119 268398 : iv_2_stmt = SSA_NAME_DEF_STMT (iv_2);
2120 268398 : modify_before_test = false;
2121 : }
2122 :
2123 : /* Make sure iv_2_stmt is an and stmt (iv_2 = _1 & iv_1). */
2124 1357934 : if (!is_gimple_assign (iv_2_stmt)
2125 1357934 : || gimple_assign_rhs_code (iv_2_stmt) != BIT_AND_EXPR)
2126 : return false;
2127 :
2128 68957 : tree iv_1 = gimple_assign_rhs1 (iv_2_stmt);
2129 68957 : tree _1 = gimple_assign_rhs2 (iv_2_stmt);
2130 :
2131 : /* Check that _1 is defined by (_1 = iv_1 + -1).
2132 : Also make sure that _1 is the same in and_stmt and _1 defining stmt.
2133 : Also canonicalize if _1 and _b11 are reversed. */
2134 68957 : if (ssa_defined_by_minus_one_stmt_p (iv_1, _1))
2135 : std::swap (iv_1, _1);
2136 66972 : else if (ssa_defined_by_minus_one_stmt_p (_1, iv_1))
2137 : ;
2138 : else
2139 : return false;
2140 :
2141 : /* Check the recurrence. */
2142 3606 : gimple *phi = SSA_NAME_DEF_STMT (iv_1);
2143 3606 : if (gimple_code (phi) != GIMPLE_PHI
2144 3606 : || (gimple_bb (phi) != loop_latch_edge (loop)->dest)
2145 7193 : || (iv_2 != gimple_phi_arg_def (phi, loop_latch_edge (loop)->dest_idx)))
2146 : return false;
2147 :
2148 : /* We found a match. */
2149 3587 : tree src = gimple_phi_arg_def (phi, loop_preheader_edge (loop)->dest_idx);
2150 3587 : int src_precision = TYPE_PRECISION (TREE_TYPE (src));
2151 :
2152 : /* Get the corresponding popcount builtin. */
2153 3587 : tree expr = build_popcount_expr (src);
2154 :
2155 3587 : if (!expr)
2156 : return false;
2157 :
2158 3587 : max = src_precision;
2159 :
2160 3587 : tree may_be_zero = boolean_false_node;
2161 :
2162 3587 : if (modify_before_test)
2163 : {
2164 1735 : expr = fold_build2 (MINUS_EXPR, integer_type_node, expr,
2165 : integer_one_node);
2166 1735 : max = max - 1;
2167 1735 : may_be_zero = fold_build2 (EQ_EXPR, boolean_type_node, src,
2168 : build_zero_cst (TREE_TYPE (src)));
2169 : }
2170 :
2171 3587 : expr = fold_convert (unsigned_type_node, expr);
2172 :
2173 3587 : niter->assumptions = boolean_true_node;
2174 3587 : niter->may_be_zero = simplify_using_initial_conditions (loop, may_be_zero);
2175 3587 : niter->niter = simplify_using_initial_conditions(loop, expr);
2176 :
2177 3587 : if (TREE_CODE (niter->niter) == INTEGER_CST)
2178 411 : niter->max = tree_to_uhwi (niter->niter);
2179 : else
2180 3176 : niter->max = max;
2181 :
2182 3587 : niter->bound = NULL_TREE;
2183 3587 : niter->cmp = ERROR_MARK;
2184 3587 : return true;
2185 : }
2186 :
2187 : /* Return an expression that counts the leading/trailing zeroes of src.
2188 :
2189 : If define_at_zero is true, then the built expression will be defined to
2190 : return the precision of src when src == 0 (using either a conditional
2191 : expression or a suitable internal function).
2192 : Otherwise, we can elide the conditional expression and let src = 0 invoke
2193 : undefined behaviour. */
2194 :
2195 : static tree
2196 9268 : build_cltz_expr (tree src, bool leading, bool define_at_zero)
2197 : {
2198 9268 : tree fn;
2199 9268 : internal_fn ifn = leading ? IFN_CLZ : IFN_CTZ;
2200 9268 : bool use_ifn = false;
2201 9268 : int prec = TYPE_PRECISION (TREE_TYPE (src));
2202 9268 : int i_prec = TYPE_PRECISION (integer_type_node);
2203 9268 : int li_prec = TYPE_PRECISION (long_integer_type_node);
2204 9268 : int lli_prec = TYPE_PRECISION (long_long_integer_type_node);
2205 :
2206 9268 : tree utype = unsigned_type_for (TREE_TYPE (src));
2207 9268 : src = fold_convert (utype, src);
2208 :
2209 9268 : if (direct_internal_fn_supported_p (ifn, utype, OPTIMIZE_FOR_BOTH))
2210 : use_ifn = true;
2211 3522 : else if (prec <= i_prec)
2212 1728 : fn = leading ? builtin_decl_implicit (BUILT_IN_CLZ)
2213 12790 : : builtin_decl_implicit (BUILT_IN_CTZ);
2214 1794 : else if (prec == li_prec)
2215 0 : fn = leading ? builtin_decl_implicit (BUILT_IN_CLZL)
2216 12790 : : builtin_decl_implicit (BUILT_IN_CTZL);
2217 1794 : else if (prec == lli_prec || prec == 2 * lli_prec)
2218 1794 : fn = leading ? builtin_decl_implicit (BUILT_IN_CLZLL)
2219 12790 : : builtin_decl_implicit (BUILT_IN_CTZLL);
2220 : else
2221 : return NULL_TREE;
2222 :
2223 9268 : tree call;
2224 3522 : if (use_ifn)
2225 : {
2226 5746 : int val;
2227 5746 : int optab_defined_at_zero
2228 : = (leading
2229 5746 : ? CLZ_DEFINED_VALUE_AT_ZERO (SCALAR_INT_TYPE_MODE (utype), val)
2230 6781 : : CTZ_DEFINED_VALUE_AT_ZERO (SCALAR_INT_TYPE_MODE (utype), val));
2231 5746 : tree arg2 = NULL_TREE;
2232 5746 : if (define_at_zero && optab_defined_at_zero == 2 && val == prec)
2233 0 : arg2 = build_int_cst (integer_type_node, val);
2234 5746 : call = build_call_expr_internal_loc (UNKNOWN_LOCATION, ifn,
2235 : integer_type_node, arg2 ? 2 : 1,
2236 : src, arg2);
2237 5746 : if (define_at_zero && arg2 == NULL_TREE)
2238 : {
2239 4831 : tree is_zero = fold_build2 (NE_EXPR, boolean_type_node, src,
2240 : build_zero_cst (TREE_TYPE (src)));
2241 4831 : call = fold_build3 (COND_EXPR, integer_type_node, is_zero, call,
2242 : build_int_cst (integer_type_node, prec));
2243 : }
2244 : }
2245 3522 : else if (fn == NULL_TREE)
2246 : return NULL_TREE;
2247 3522 : else if (prec == 2 * lli_prec)
2248 : {
2249 844 : tree src1 = fold_convert (long_long_unsigned_type_node,
2250 : fold_build2 (RSHIFT_EXPR, TREE_TYPE (src),
2251 : unshare_expr (src),
2252 : build_int_cst (integer_type_node,
2253 : lli_prec)));
2254 844 : tree src2 = fold_convert (long_long_unsigned_type_node, src);
2255 : /* We count the zeroes in src1, and add the number in src2 when src1
2256 : is 0. */
2257 844 : if (!leading)
2258 0 : std::swap (src1, src2);
2259 844 : tree call1 = build_call_expr (fn, 1, src1);
2260 844 : tree call2 = build_call_expr (fn, 1, src2);
2261 844 : if (define_at_zero)
2262 : {
2263 728 : tree is_zero2 = fold_build2 (NE_EXPR, boolean_type_node, src2,
2264 : build_zero_cst (TREE_TYPE (src2)));
2265 728 : call2 = fold_build3 (COND_EXPR, integer_type_node, is_zero2, call2,
2266 : build_int_cst (integer_type_node, lli_prec));
2267 : }
2268 844 : tree is_zero1 = fold_build2 (NE_EXPR, boolean_type_node, src1,
2269 : build_zero_cst (TREE_TYPE (src1)));
2270 844 : call = fold_build3 (COND_EXPR, integer_type_node, is_zero1, call1,
2271 : fold_build2 (PLUS_EXPR, integer_type_node, call2,
2272 : build_int_cst (integer_type_node,
2273 : lli_prec)));
2274 : }
2275 : else
2276 : {
2277 2678 : if (prec < i_prec)
2278 1728 : src = fold_convert (unsigned_type_node, src);
2279 :
2280 2678 : call = build_call_expr (fn, 1, src);
2281 2678 : if (leading && prec < i_prec)
2282 1547 : call = fold_build2 (MINUS_EXPR, integer_type_node, call,
2283 : build_int_cst (integer_type_node, i_prec - prec));
2284 2678 : if (define_at_zero)
2285 : {
2286 2388 : tree is_zero = fold_build2 (NE_EXPR, boolean_type_node, src,
2287 : build_zero_cst (TREE_TYPE (src)));
2288 2388 : call = fold_build3 (COND_EXPR, integer_type_node, is_zero, call,
2289 : build_int_cst (integer_type_node, prec));
2290 : }
2291 : }
2292 :
2293 : return call;
2294 : }
2295 :
2296 : /* Returns true if STMT is equivalent to x << 1. */
2297 :
2298 : static bool
2299 1172896 : is_lshift_by_1 (gassign *stmt)
2300 : {
2301 1172896 : if (gimple_assign_rhs_code (stmt) == LSHIFT_EXPR
2302 1174230 : && integer_onep (gimple_assign_rhs2 (stmt)))
2303 : return true;
2304 1172128 : if (gimple_assign_rhs_code (stmt) == MULT_EXPR
2305 1337 : && tree_fits_shwi_p (gimple_assign_rhs2 (stmt))
2306 1173146 : && tree_to_shwi (gimple_assign_rhs2 (stmt)) == 2)
2307 227 : return true;
2308 : return false;
2309 : }
2310 :
2311 : /* Returns true if STMT is equivalent to x >> 1. */
2312 :
2313 : static bool
2314 1171901 : is_rshift_by_1 (gassign *stmt)
2315 : {
2316 1171901 : if (gimple_assign_rhs_code (stmt) == RSHIFT_EXPR
2317 1194956 : && integer_onep (gimple_assign_rhs2 (stmt)))
2318 : return true;
2319 2101713 : if (trunc_or_exact_div_p (gimple_assign_rhs_code (stmt))
2320 1467 : && tree_fits_shwi_p (gimple_assign_rhs2 (stmt))
2321 1161410 : && tree_to_shwi (gimple_assign_rhs2 (stmt)) == 2)
2322 99 : return true;
2323 : return false;
2324 : }
2325 :
2326 : /* Helper for number_of_iterations_cltz that uses ranger to determine
2327 : if SRC's range, shifted left (when LEFT_SHIFT is true) or right
2328 : by NUM_IGNORED_BITS, is guaranteed to be != 0 on LOOP's preheader
2329 : edge.
2330 : Return true if so or false otherwise. */
2331 :
2332 : static bool
2333 1321 : shifted_range_nonzero_p (loop_p loop, tree src,
2334 : bool left_shift, int num_ignored_bits)
2335 : {
2336 1321 : int_range_max r (TREE_TYPE (src));
2337 1321 : gcc_assert (num_ignored_bits >= 0);
2338 :
2339 1321 : if (get_range_query (cfun)->range_on_edge
2340 1321 : (r, loop_preheader_edge (loop), src)
2341 1321 : && !r.varying_p ()
2342 2070 : && !r.undefined_p ())
2343 : {
2344 745 : if (num_ignored_bits)
2345 : {
2346 689 : range_op_handler op (left_shift ? LSHIFT_EXPR : RSHIFT_EXPR);
2347 443 : int_range_max shifted_range (TREE_TYPE (src));
2348 443 : wide_int shift_count = wi::shwi (num_ignored_bits,
2349 443 : TYPE_PRECISION (TREE_TYPE
2350 443 : (src)));
2351 443 : int_range_max shift_amount
2352 443 : (TREE_TYPE (src), shift_count, shift_count);
2353 :
2354 443 : if (op.fold_range (shifted_range, TREE_TYPE (src), r,
2355 : shift_amount))
2356 443 : r = shifted_range;
2357 443 : }
2358 :
2359 : /* If the range does not contain zero we are good. */
2360 745 : if (!range_includes_zero_p (r))
2361 498 : return true;
2362 : }
2363 :
2364 : return false;
2365 1321 : }
2366 :
2367 :
2368 : /* See comment below for number_of_iterations_bitcount.
2369 : For c[lt]z, we have:
2370 :
2371 : modify:
2372 : iv_2 = iv_1 << 1 OR iv_1 >> 1
2373 :
2374 : test:
2375 : if (iv & 1 << (prec-1)) OR (iv & 1)
2376 :
2377 : modification count:
2378 : src precision - c[lt]z (src)
2379 :
2380 : */
2381 :
2382 : static bool
2383 7108915 : number_of_iterations_cltz (loop_p loop, edge exit,
2384 : enum tree_code code,
2385 : class tree_niter_desc *niter)
2386 : {
2387 7108915 : bool modify_before_test = true;
2388 7108915 : HOST_WIDE_INT max;
2389 7108915 : int checked_bit;
2390 7108915 : tree iv_2;
2391 :
2392 : /* Check that condition for staying inside the loop is like
2393 : if (iv == 0). */
2394 14217830 : gcond *cond_stmt = safe_dyn_cast <gcond *> (*gsi_last_bb (exit->src));
2395 7108915 : if (!cond_stmt
2396 7108915 : || (code != EQ_EXPR && code != GE_EXPR)
2397 3531093 : || !integer_zerop (gimple_cond_rhs (cond_stmt))
2398 1354464 : || TREE_CODE (gimple_cond_lhs (cond_stmt)) != SSA_NAME)
2399 : return false;
2400 :
2401 1354422 : if (code == EQ_EXPR)
2402 : {
2403 : /* Make sure we check a bitwise and with a suitable constant */
2404 1215515 : gimple *and_stmt = SSA_NAME_DEF_STMT (gimple_cond_lhs (cond_stmt));
2405 1215515 : if (!is_gimple_assign (and_stmt)
2406 748173 : || gimple_assign_rhs_code (and_stmt) != BIT_AND_EXPR
2407 139610 : || !integer_pow2p (gimple_assign_rhs2 (and_stmt))
2408 1234182 : || TREE_CODE (gimple_assign_rhs1 (and_stmt)) != SSA_NAME)
2409 : return false;
2410 :
2411 18667 : checked_bit = tree_log2 (gimple_assign_rhs2 (and_stmt));
2412 :
2413 18667 : iv_2 = gimple_assign_rhs1 (and_stmt);
2414 : }
2415 : else
2416 : {
2417 : /* We have a GE_EXPR - a signed comparison with zero is equivalent to
2418 : testing the leading bit, so check for this pattern too. */
2419 :
2420 138907 : iv_2 = gimple_cond_lhs (cond_stmt);
2421 138907 : tree test_value_type = TREE_TYPE (iv_2);
2422 :
2423 138907 : if (TYPE_UNSIGNED (test_value_type))
2424 : return false;
2425 :
2426 138907 : gimple *test_value_stmt = SSA_NAME_DEF_STMT (iv_2);
2427 :
2428 138907 : if (is_gimple_assign (test_value_stmt)
2429 138907 : && gimple_assign_rhs_code (test_value_stmt) == NOP_EXPR)
2430 : {
2431 : /* If the test value comes from a NOP_EXPR, then we need to unwrap
2432 : this. We conservatively require that both types have the same
2433 : precision. */
2434 5452 : iv_2 = gimple_assign_rhs1 (test_value_stmt);
2435 5452 : tree rhs_type = TREE_TYPE (iv_2);
2436 5452 : if (TREE_CODE (iv_2) != SSA_NAME
2437 5452 : || !INTEGRAL_NB_TYPE_P (rhs_type)
2438 10893 : || (TYPE_PRECISION (rhs_type)
2439 5441 : != TYPE_PRECISION (test_value_type)))
2440 : return false;
2441 : }
2442 :
2443 138501 : checked_bit = TYPE_PRECISION (test_value_type) - 1;
2444 : }
2445 :
2446 157168 : gimple *iv_2_stmt = SSA_NAME_DEF_STMT (iv_2);
2447 :
2448 : /* If the test comes before the iv modification, then these will actually be
2449 : iv_1 and a phi node. */
2450 157168 : if (gimple_code (iv_2_stmt) == GIMPLE_PHI
2451 25632 : && gimple_bb (iv_2_stmt) == loop->header
2452 14703 : && gimple_phi_num_args (iv_2_stmt) == 2
2453 171871 : && (TREE_CODE (gimple_phi_arg_def (iv_2_stmt,
2454 : loop_latch_edge (loop)->dest_idx))
2455 : == SSA_NAME))
2456 : {
2457 : /* iv_2 is actually one of the inputs to the phi. */
2458 14604 : iv_2 = gimple_phi_arg_def (iv_2_stmt, loop_latch_edge (loop)->dest_idx);
2459 14604 : iv_2_stmt = SSA_NAME_DEF_STMT (iv_2);
2460 14604 : modify_before_test = false;
2461 : }
2462 :
2463 : /* Make sure iv_2_stmt is a logical shift by one stmt:
2464 : iv_2 = iv_1 {<<|>>} 1 */
2465 157168 : if (!is_gimple_assign (iv_2_stmt))
2466 : return false;
2467 127791 : bool left_shift = false;
2468 255073 : if (!((left_shift = is_lshift_by_1 (as_a <gassign *> (iv_2_stmt)))
2469 127282 : || is_rshift_by_1 (as_a <gassign *> (iv_2_stmt))))
2470 : return false;
2471 :
2472 2036 : tree iv_1 = gimple_assign_rhs1 (iv_2_stmt);
2473 :
2474 : /* Check the recurrence. */
2475 2036 : gimple *phi = SSA_NAME_DEF_STMT (iv_1);
2476 2036 : if (gimple_code (phi) != GIMPLE_PHI
2477 1975 : || (gimple_bb (phi) != loop_latch_edge (loop)->dest)
2478 4005 : || (iv_2 != gimple_phi_arg_def (phi, loop_latch_edge (loop)->dest_idx)))
2479 : return false;
2480 :
2481 : /* We found a match. */
2482 1969 : tree src = gimple_phi_arg_def (phi, loop_preheader_edge (loop)->dest_idx);
2483 :
2484 : /* If the type is signed, verify via Ranger on the preheader edge
2485 : that the initial value entering the loop is non-negative. */
2486 1969 : if (!TYPE_UNSIGNED (TREE_TYPE (src)))
2487 : {
2488 1037 : int_range_max r (TREE_TYPE (src));
2489 1037 : edge e = loop_preheader_edge (loop);
2490 2074 : if (!get_range_query (cfun)->range_on_edge (r, e, src)
2491 1037 : || r.undefined_p ()
2492 2074 : || !r.nonnegative_p ())
2493 648 : return false;
2494 1037 : }
2495 :
2496 1321 : int src_precision = TYPE_PRECISION (TREE_TYPE (src));
2497 :
2498 : /* Save the original SSA name before preprocessing for ranger queries. */
2499 1321 : tree unshifted_src = src;
2500 :
2501 : /* Apply any needed preprocessing to src. */
2502 1321 : int num_ignored_bits;
2503 1321 : if (left_shift)
2504 382 : num_ignored_bits = src_precision - checked_bit - 1;
2505 : else
2506 : num_ignored_bits = checked_bit;
2507 :
2508 1321 : if (modify_before_test)
2509 575 : num_ignored_bits++;
2510 :
2511 1321 : if (num_ignored_bits != 0)
2512 1245 : src = fold_build2 (left_shift ? LSHIFT_EXPR : RSHIFT_EXPR,
2513 : TREE_TYPE (src), src,
2514 : build_int_cst (integer_type_node, num_ignored_bits));
2515 :
2516 : /* Get the corresponding c[lt]z builtin. */
2517 1321 : tree expr = build_cltz_expr (src, left_shift, false);
2518 :
2519 1321 : if (!expr)
2520 : return false;
2521 :
2522 1321 : max = src_precision - num_ignored_bits - 1;
2523 :
2524 1321 : expr = fold_convert (unsigned_type_node, expr);
2525 :
2526 : /* If the copy-header (ch) pass peeled one iteration we're shifting
2527 : SRC by preprocessing it above.
2528 :
2529 : A loop like
2530 : if (bits)
2531 : {
2532 : while (!(bits & 1))
2533 : {
2534 : bits >>= 1;
2535 : cnt += 1;
2536 : }
2537 : return cnt;
2538 : }
2539 : ch (roughly) transforms into:
2540 : if (bits)
2541 : {
2542 : if (!(bits & 1)
2543 : {
2544 : do
2545 : {
2546 : bits >>= 1;
2547 : cnt += 1;
2548 : } while (!(bits & 1));
2549 : }
2550 : else
2551 : cnt = 1;
2552 : return cnt;
2553 : }
2554 :
2555 : Then, our preprocessed SRC (that is used for c[tl]z computation)
2556 : will be bits >> 1, and the assumption is bits >> 1 != 0. */
2557 :
2558 1321 : tree assumptions;
2559 1321 : if (shifted_range_nonzero_p (loop, unshifted_src,
2560 : left_shift, num_ignored_bits))
2561 498 : assumptions = boolean_true_node;
2562 : else
2563 : {
2564 : /* If ranger couldn't prove the assumption, try
2565 : simplify_using_initial_conditions. */
2566 823 : assumptions = fold_build2 (NE_EXPR, boolean_type_node, src,
2567 : build_zero_cst (TREE_TYPE (src)));
2568 823 : assumptions = simplify_using_initial_conditions (loop, assumptions);
2569 : }
2570 :
2571 1321 : niter->assumptions = assumptions;
2572 1321 : niter->may_be_zero = boolean_false_node;
2573 1321 : niter->niter = simplify_using_initial_conditions (loop, expr);
2574 :
2575 1321 : if (TREE_CODE (niter->niter) == INTEGER_CST)
2576 100 : niter->max = tree_to_uhwi (niter->niter);
2577 : else
2578 1221 : niter->max = max;
2579 :
2580 1321 : niter->bound = NULL_TREE;
2581 1321 : niter->cmp = ERROR_MARK;
2582 :
2583 1321 : return true;
2584 : }
2585 :
2586 : /* See comment below for number_of_iterations_bitcount.
2587 : For c[lt]z complement, we have:
2588 :
2589 : modify:
2590 : iv_2 = iv_1 >> 1 OR iv_1 << 1
2591 :
2592 : test:
2593 : if (iv != 0)
2594 :
2595 : modification count:
2596 : src precision - c[lt]z (src)
2597 :
2598 : */
2599 :
2600 : static bool
2601 3894260 : number_of_iterations_cltz_complement (loop_p loop, edge exit,
2602 : enum tree_code code,
2603 : class tree_niter_desc *niter)
2604 : {
2605 3894260 : bool modify_before_test = true;
2606 3894260 : HOST_WIDE_INT max;
2607 :
2608 : /* Check that condition for staying inside the loop is like
2609 : if (iv != 0). */
2610 7788520 : gcond *cond_stmt = safe_dyn_cast <gcond *> (*gsi_last_bb (exit->src));
2611 3894260 : if (!cond_stmt
2612 3894260 : || code != NE_EXPR
2613 1893608 : || !integer_zerop (gimple_cond_rhs (cond_stmt))
2614 5248607 : || TREE_CODE (gimple_cond_lhs (cond_stmt)) != SSA_NAME)
2615 : return false;
2616 :
2617 1354347 : tree iv_2 = gimple_cond_lhs (cond_stmt);
2618 1354347 : gimple *iv_2_stmt = SSA_NAME_DEF_STMT (iv_2);
2619 :
2620 : /* If the test comes before the iv modification, then these will actually be
2621 : iv_1 and a phi node. */
2622 1354347 : if (gimple_code (iv_2_stmt) == GIMPLE_PHI
2623 349051 : && gimple_bb (iv_2_stmt) == loop->header
2624 271255 : && gimple_phi_num_args (iv_2_stmt) == 2
2625 1625602 : && (TREE_CODE (gimple_phi_arg_def (iv_2_stmt,
2626 : loop_latch_edge (loop)->dest_idx))
2627 : == SSA_NAME))
2628 : {
2629 : /* iv_2 is actually one of the inputs to the phi. */
2630 266546 : iv_2 = gimple_phi_arg_def (iv_2_stmt, loop_latch_edge (loop)->dest_idx);
2631 266546 : iv_2_stmt = SSA_NAME_DEF_STMT (iv_2);
2632 266546 : modify_before_test = false;
2633 : }
2634 :
2635 : /* Make sure iv_2_stmt is a logical shift by one stmt:
2636 : iv_2 = iv_1 {>>|<<} 1 */
2637 1354347 : if (!is_gimple_assign (iv_2_stmt))
2638 : return false;
2639 1045105 : bool left_shift = false;
2640 2089724 : if (!((left_shift = is_lshift_by_1 (as_a <gassign *> (iv_2_stmt)))
2641 1044619 : || is_rshift_by_1 (as_a <gassign *> (iv_2_stmt))))
2642 : return false;
2643 :
2644 10544 : tree iv_1 = gimple_assign_rhs1 (iv_2_stmt);
2645 :
2646 : /* Check the recurrence. */
2647 10544 : gimple *phi = SSA_NAME_DEF_STMT (iv_1);
2648 10544 : if (gimple_code (phi) != GIMPLE_PHI
2649 9809 : || (gimple_bb (phi) != loop_latch_edge (loop)->dest)
2650 19180 : || (iv_2 != gimple_phi_arg_def (phi, loop_latch_edge (loop)->dest_idx)))
2651 : return false;
2652 :
2653 : /* We found a match. */
2654 8383 : tree src = gimple_phi_arg_def (phi, loop_preheader_edge (loop)->dest_idx);
2655 :
2656 : /* If the type is signed, verify via Ranger on the preheader edge
2657 : that the initial value entering the loop is non-negative. */
2658 8383 : if (!TYPE_UNSIGNED (TREE_TYPE (src)))
2659 : {
2660 888 : int_range_max r (TREE_TYPE (src));
2661 888 : edge e = loop_preheader_edge (loop);
2662 1776 : if (!get_range_query (cfun)->range_on_edge (r, e, src)
2663 888 : || r.undefined_p ()
2664 1776 : || !r.nonnegative_p ())
2665 436 : return false;
2666 888 : }
2667 :
2668 7947 : int src_precision = TYPE_PRECISION (TREE_TYPE (src));
2669 :
2670 : /* Get the corresponding c[lt]z builtin. */
2671 7947 : tree expr = build_cltz_expr (src, !left_shift, true);
2672 :
2673 7947 : if (!expr)
2674 : return false;
2675 :
2676 7947 : expr = fold_build2 (MINUS_EXPR, integer_type_node,
2677 : build_int_cst (integer_type_node, src_precision),
2678 : expr);
2679 :
2680 7947 : max = src_precision;
2681 :
2682 7947 : tree may_be_zero = boolean_false_node;
2683 :
2684 7947 : if (modify_before_test)
2685 : {
2686 6454 : expr = fold_build2 (MINUS_EXPR, integer_type_node, expr,
2687 : integer_one_node);
2688 6454 : max = max - 1;
2689 6454 : may_be_zero = fold_build2 (EQ_EXPR, boolean_type_node, src,
2690 : build_zero_cst (TREE_TYPE (src)));
2691 : }
2692 :
2693 7947 : expr = fold_convert (unsigned_type_node, expr);
2694 :
2695 7947 : niter->assumptions = boolean_true_node;
2696 7947 : niter->may_be_zero = simplify_using_initial_conditions (loop, may_be_zero);
2697 7947 : niter->niter = simplify_using_initial_conditions (loop, expr);
2698 :
2699 7947 : if (TREE_CODE (niter->niter) == INTEGER_CST)
2700 69 : niter->max = tree_to_uhwi (niter->niter);
2701 : else
2702 7878 : niter->max = max;
2703 :
2704 7947 : niter->bound = NULL_TREE;
2705 7947 : niter->cmp = ERROR_MARK;
2706 7947 : return true;
2707 : }
2708 :
2709 : /* See if LOOP contains a bit counting idiom. The idiom consists of two parts:
2710 : 1. A modification to the induction variabler;.
2711 : 2. A test to determine whether or not to exit the loop.
2712 :
2713 : These can come in either order - i.e.:
2714 :
2715 : <bb 3>
2716 : iv_1 = PHI <src(2), iv_2(4)>
2717 : if (test (iv_1))
2718 : goto <bb 4>
2719 : else
2720 : goto <bb 5>
2721 :
2722 : <bb 4>
2723 : iv_2 = modify (iv_1)
2724 : goto <bb 3>
2725 :
2726 : OR
2727 :
2728 : <bb 3>
2729 : iv_1 = PHI <src(2), iv_2(4)>
2730 : iv_2 = modify (iv_1)
2731 :
2732 : <bb 4>
2733 : if (test (iv_2))
2734 : goto <bb 3>
2735 : else
2736 : goto <bb 5>
2737 :
2738 : The second form can be generated by copying the loop header out of the loop.
2739 :
2740 : In the first case, the number of latch executions will be equal to the
2741 : number of induction variable modifications required before the test fails.
2742 :
2743 : In the second case (modify_before_test), if we assume that the number of
2744 : modifications required before the test fails is nonzero, then the number of
2745 : latch executions will be one less than this number.
2746 :
2747 : If we recognise the pattern, then we update niter accordingly, and return
2748 : true. */
2749 :
2750 : static bool
2751 3897926 : number_of_iterations_bitcount (loop_p loop, edge exit,
2752 : enum tree_code code,
2753 : class tree_niter_desc *niter)
2754 : {
2755 3897926 : return (number_of_iterations_popcount (loop, exit, code, niter)
2756 3894339 : || number_of_iterations_cltz (loop, exit, code, niter)
2757 7792186 : || number_of_iterations_cltz_complement (loop, exit, code, niter));
2758 : }
2759 :
2760 : /* Substitute NEW_TREE for OLD in EXPR and fold the result.
2761 : If VALUEIZE is non-NULL then OLD and NEW_TREE are ignored and instead
2762 : all SSA names are replaced with the result of calling the VALUEIZE
2763 : function with the SSA name as argument. */
2764 :
2765 : tree
2766 160471027 : simplify_replace_tree (tree expr, tree old, tree new_tree,
2767 : tree (*valueize) (tree, void*), void *context,
2768 : bool do_fold)
2769 : {
2770 160471027 : unsigned i, n;
2771 160471027 : tree ret = NULL_TREE, e, se;
2772 :
2773 160471027 : if (!expr)
2774 : return NULL_TREE;
2775 :
2776 : /* Do not bother to replace constants. */
2777 32399918 : if (CONSTANT_CLASS_P (expr))
2778 : return expr;
2779 :
2780 24477779 : if (valueize)
2781 : {
2782 264892 : if (TREE_CODE (expr) == SSA_NAME)
2783 : {
2784 82284 : new_tree = valueize (expr, context);
2785 82284 : if (new_tree != expr)
2786 : return new_tree;
2787 : }
2788 : }
2789 24212887 : else if (expr == old
2790 24212887 : || operand_equal_p (expr, old, 0))
2791 221845 : return unshare_expr (new_tree);
2792 :
2793 24255328 : if (!EXPR_P (expr))
2794 : return expr;
2795 :
2796 13016274 : n = TREE_OPERAND_LENGTH (expr);
2797 36485507 : for (i = 0; i < n; i++)
2798 : {
2799 23469233 : e = TREE_OPERAND (expr, i);
2800 23469233 : se = simplify_replace_tree (e, old, new_tree, valueize, context, do_fold);
2801 23469233 : if (e == se)
2802 23180585 : continue;
2803 :
2804 288648 : if (!ret)
2805 288257 : ret = copy_node (expr);
2806 :
2807 288648 : TREE_OPERAND (ret, i) = se;
2808 : }
2809 :
2810 13016274 : return (ret ? (do_fold ? fold (ret) : ret) : expr);
2811 : }
2812 :
2813 : /* Expand definitions of ssa names in EXPR as long as they are simple
2814 : enough, and return the new expression. If STOP is specified, stop
2815 : expanding if EXPR equals to it. */
2816 :
2817 : static tree
2818 123874757 : expand_simple_operations (tree expr, tree stop, hash_map<tree, tree> &cache)
2819 : {
2820 124186884 : unsigned i, n;
2821 124186884 : tree ret = NULL_TREE, e, ee, e1;
2822 124186884 : enum tree_code code;
2823 124186884 : gimple *stmt;
2824 :
2825 124186884 : if (expr == NULL_TREE)
2826 : return expr;
2827 :
2828 124186884 : if (is_gimple_min_invariant (expr))
2829 : return expr;
2830 :
2831 84941909 : code = TREE_CODE (expr);
2832 84941909 : if (IS_EXPR_CODE_CLASS (TREE_CODE_CLASS (code)))
2833 : {
2834 20619799 : n = TREE_OPERAND_LENGTH (expr);
2835 56997762 : for (i = 0; i < n; i++)
2836 : {
2837 36377963 : e = TREE_OPERAND (expr, i);
2838 36377963 : if (!e)
2839 32971721 : continue;
2840 : /* SCEV analysis feeds us with a proper expression
2841 : graph matching the SSA graph. Avoid turning it
2842 : into a tree here, thus handle tree sharing
2843 : properly.
2844 : ??? The SSA walk below still turns the SSA graph
2845 : into a tree but until we find a testcase do not
2846 : introduce additional tree sharing here. */
2847 36359030 : bool existed_p;
2848 36359030 : tree &cee = cache.get_or_insert (e, &existed_p);
2849 36359030 : if (existed_p)
2850 147936 : ee = cee;
2851 : else
2852 : {
2853 36211094 : cee = e;
2854 36211094 : ee = expand_simple_operations (e, stop, cache);
2855 36211094 : if (ee != e)
2856 3404391 : *cache.get (e) = ee;
2857 : }
2858 36359030 : if (e == ee)
2859 32952788 : continue;
2860 :
2861 3406242 : if (!ret)
2862 3233682 : ret = copy_node (expr);
2863 :
2864 3406242 : TREE_OPERAND (ret, i) = ee;
2865 : }
2866 :
2867 20619799 : if (!ret)
2868 : return expr;
2869 :
2870 3233682 : ret = fold (ret);
2871 3233682 : return ret;
2872 : }
2873 :
2874 : /* Stop if it's not ssa name or the one we don't want to expand. */
2875 64322110 : if (TREE_CODE (expr) != SSA_NAME || expr == stop)
2876 : return expr;
2877 :
2878 64125710 : stmt = SSA_NAME_DEF_STMT (expr);
2879 64125710 : if (gimple_code (stmt) == GIMPLE_PHI)
2880 : {
2881 17226579 : basic_block src, dest;
2882 :
2883 17226579 : if (gimple_phi_num_args (stmt) != 1)
2884 : return expr;
2885 465501 : e = PHI_ARG_DEF (stmt, 0);
2886 :
2887 : /* Avoid propagating through loop exit phi nodes, which
2888 : could break loop-closed SSA form restrictions. */
2889 465501 : dest = gimple_bb (stmt);
2890 465501 : src = single_pred (dest);
2891 465501 : if (TREE_CODE (e) == SSA_NAME
2892 462066 : && src->loop_father != dest->loop_father)
2893 : return expr;
2894 :
2895 : return expand_simple_operations (e, stop, cache);
2896 : }
2897 46899131 : if (gimple_code (stmt) != GIMPLE_ASSIGN)
2898 : return expr;
2899 :
2900 : /* Avoid expanding to expressions that contain SSA names that need
2901 : to take part in abnormal coalescing. */
2902 41392846 : ssa_op_iter iter;
2903 86060018 : FOR_EACH_SSA_TREE_OPERAND (e, stmt, iter, SSA_OP_USE)
2904 44667766 : if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (e))
2905 : return expr;
2906 :
2907 41392252 : e = gimple_assign_rhs1 (stmt);
2908 41392252 : code = gimple_assign_rhs_code (stmt);
2909 41392252 : if (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS)
2910 : {
2911 16362339 : if (is_gimple_min_invariant (e))
2912 : return e;
2913 :
2914 16357447 : if (code == SSA_NAME)
2915 : return expand_simple_operations (e, stop, cache);
2916 16136607 : else if (code == ADDR_EXPR)
2917 : {
2918 17079 : poly_int64 offset;
2919 17079 : tree base = get_addr_base_and_unit_offset (TREE_OPERAND (e, 0),
2920 : &offset);
2921 17079 : if (base
2922 16663 : && TREE_CODE (base) == MEM_REF)
2923 : {
2924 16663 : ee = expand_simple_operations (TREE_OPERAND (base, 0), stop,
2925 : cache);
2926 16663 : return fold_build2 (POINTER_PLUS_EXPR, TREE_TYPE (expr), ee,
2927 : wide_int_to_tree (sizetype,
2928 : mem_ref_offset (base)
2929 : + offset));
2930 : }
2931 : }
2932 :
2933 : return expr;
2934 : }
2935 :
2936 25029913 : switch (code)
2937 : {
2938 6256289 : CASE_CONVERT:
2939 : /* Casts are simple. */
2940 6256289 : ee = expand_simple_operations (e, stop, cache);
2941 6256289 : return fold_build1 (code, TREE_TYPE (expr), ee);
2942 :
2943 13629553 : case PLUS_EXPR:
2944 13629553 : case MINUS_EXPR:
2945 13629553 : case MULT_EXPR:
2946 27259106 : if (ANY_INTEGRAL_TYPE_P (TREE_TYPE (expr))
2947 27256371 : && TYPE_OVERFLOW_TRAPS (TREE_TYPE (expr)))
2948 : return expr;
2949 : /* Fallthru. */
2950 14357888 : case POINTER_PLUS_EXPR:
2951 : /* And increments and decrements by a constant are simple. */
2952 14357888 : e1 = gimple_assign_rhs2 (stmt);
2953 14357888 : if (!is_gimple_min_invariant (e1))
2954 : return expr;
2955 :
2956 7372377 : ee = expand_simple_operations (e, stop, cache);
2957 7372377 : return fold_build2 (code, TREE_TYPE (expr), ee, e1);
2958 :
2959 : default:
2960 : return expr;
2961 : }
2962 : }
2963 :
2964 : tree
2965 74018334 : expand_simple_operations (tree expr, tree stop)
2966 : {
2967 74018334 : hash_map<tree, tree> cache;
2968 74018334 : return expand_simple_operations (expr, stop, cache);
2969 74018334 : }
2970 :
2971 : /* Tries to simplify EXPR using the condition COND. Returns the simplified
2972 : expression (or EXPR unchanged, if no simplification was possible). */
2973 :
2974 : static tree
2975 9086863 : tree_simplify_using_condition_1 (tree cond, tree expr)
2976 : {
2977 9086863 : bool changed;
2978 9086863 : tree e, e0, e1, e2, notcond;
2979 9086863 : enum tree_code code = TREE_CODE (expr);
2980 :
2981 9086863 : if (code == INTEGER_CST)
2982 : return expr;
2983 :
2984 9076497 : if (code == TRUTH_OR_EXPR
2985 9076497 : || code == TRUTH_AND_EXPR
2986 9076497 : || code == COND_EXPR)
2987 : {
2988 111690 : changed = false;
2989 :
2990 111690 : e0 = tree_simplify_using_condition_1 (cond, TREE_OPERAND (expr, 0));
2991 111690 : if (TREE_OPERAND (expr, 0) != e0)
2992 2122 : changed = true;
2993 :
2994 111690 : e1 = tree_simplify_using_condition_1 (cond, TREE_OPERAND (expr, 1));
2995 111690 : if (TREE_OPERAND (expr, 1) != e1)
2996 4 : changed = true;
2997 :
2998 111690 : if (code == COND_EXPR)
2999 : {
3000 12248 : e2 = tree_simplify_using_condition_1 (cond, TREE_OPERAND (expr, 2));
3001 12248 : if (TREE_OPERAND (expr, 2) != e2)
3002 : changed = true;
3003 : }
3004 : else
3005 : e2 = NULL_TREE;
3006 :
3007 111690 : if (changed)
3008 : {
3009 2126 : if (code == COND_EXPR)
3010 1985 : expr = fold_build3 (code, boolean_type_node, e0, e1, e2);
3011 : else
3012 141 : expr = fold_build2 (code, boolean_type_node, e0, e1);
3013 : }
3014 :
3015 : return expr;
3016 : }
3017 :
3018 : /* In case COND is equality, we may be able to simplify EXPR by copy/constant
3019 : propagation, and vice versa. Fold does not handle this, since it is
3020 : considered too expensive. */
3021 8964807 : if (TREE_CODE (cond) == EQ_EXPR)
3022 : {
3023 2131805 : e0 = TREE_OPERAND (cond, 0);
3024 2131805 : e1 = TREE_OPERAND (cond, 1);
3025 :
3026 : /* We know that e0 == e1. Check whether we cannot simplify expr
3027 : using this fact. */
3028 2131805 : e = simplify_replace_tree (expr, e0, e1);
3029 2131805 : if (integer_zerop (e) || integer_nonzerop (e))
3030 : return e;
3031 :
3032 2129868 : e = simplify_replace_tree (expr, e1, e0);
3033 2129868 : if (integer_zerop (e) || integer_nonzerop (e))
3034 : return e;
3035 : }
3036 8962517 : if (TREE_CODE (expr) == EQ_EXPR)
3037 : {
3038 1143732 : e0 = TREE_OPERAND (expr, 0);
3039 1143732 : e1 = TREE_OPERAND (expr, 1);
3040 :
3041 : /* If e0 == e1 (EXPR) implies !COND, then EXPR cannot be true. */
3042 1143732 : e = simplify_replace_tree (cond, e0, e1);
3043 1143732 : if (integer_zerop (e))
3044 : return e;
3045 1129692 : e = simplify_replace_tree (cond, e1, e0);
3046 1129692 : if (integer_zerop (e))
3047 : return e;
3048 : }
3049 8948477 : if (TREE_CODE (expr) == NE_EXPR)
3050 : {
3051 1078214 : e0 = TREE_OPERAND (expr, 0);
3052 1078214 : e1 = TREE_OPERAND (expr, 1);
3053 :
3054 : /* If e0 == e1 (!EXPR) implies !COND, then EXPR must be true. */
3055 1078214 : e = simplify_replace_tree (cond, e0, e1);
3056 1078214 : if (integer_zerop (e))
3057 13276 : return boolean_true_node;
3058 1064938 : e = simplify_replace_tree (cond, e1, e0);
3059 1064938 : if (integer_zerop (e))
3060 0 : return boolean_true_node;
3061 : }
3062 :
3063 : /* Check whether COND ==> EXPR. */
3064 8935201 : notcond = invert_truthvalue (cond);
3065 8935201 : e = fold_binary (TRUTH_OR_EXPR, boolean_type_node, notcond, expr);
3066 8935201 : if (e && integer_nonzerop (e))
3067 : return e;
3068 :
3069 : /* Check whether COND ==> not EXPR. */
3070 8932831 : e = fold_binary (TRUTH_AND_EXPR, boolean_type_node, cond, expr);
3071 8932831 : if (e && integer_zerop (e))
3072 40735 : return e;
3073 :
3074 : return expr;
3075 : }
3076 :
3077 : /* Tries to simplify EXPR using the condition COND. Returns the simplified
3078 : expression (or EXPR unchanged, if no simplification was possible).
3079 : Wrapper around tree_simplify_using_condition_1 that ensures that chains
3080 : of simple operations in definitions of ssa names in COND are expanded,
3081 : so that things like casts or incrementing the value of the bound before
3082 : the loop do not cause us to fail. */
3083 :
3084 : static tree
3085 8851235 : tree_simplify_using_condition (tree cond, tree expr)
3086 : {
3087 8851235 : cond = expand_simple_operations (cond);
3088 :
3089 8851235 : return tree_simplify_using_condition_1 (cond, expr);
3090 : }
3091 :
3092 : /* Tries to simplify EXPR using the conditions on entry to LOOP.
3093 : Returns the simplified expression (or EXPR unchanged, if no
3094 : simplification was possible). */
3095 :
3096 : tree
3097 28449387 : simplify_using_initial_conditions (class loop *loop, tree expr)
3098 : {
3099 28449387 : edge e;
3100 28449387 : basic_block bb;
3101 28449387 : tree cond, expanded, backup;
3102 28449387 : int cnt = 0;
3103 :
3104 28449387 : if (TREE_CODE (expr) == INTEGER_CST)
3105 : return expr;
3106 :
3107 2927709 : value_range expr_range (TREE_TYPE (expr));
3108 2927709 : tree val;
3109 2927709 : if (TREE_TYPE (expr) == boolean_type_node
3110 5830868 : && get_range_query (cfun)->range_on_edge (expr_range,
3111 : loop_preheader_edge (loop),
3112 : expr)
3113 5843143 : && expr_range.singleton_p (&val))
3114 80356 : return val;
3115 :
3116 2847353 : backup = expanded = expand_simple_operations (expr);
3117 :
3118 : /* Limit walking the dominators to avoid quadraticness in
3119 : the number of BBs times the number of loops in degenerate
3120 : cases. */
3121 2847353 : for (bb = loop->header;
3122 27703253 : bb != ENTRY_BLOCK_PTR_FOR_FN (cfun)
3123 27703253 : && cnt < param_max_niter_dominators_walk;
3124 24855900 : bb = get_immediate_dominator (CDI_DOMINATORS, bb))
3125 : {
3126 24926619 : if (!single_pred_p (bb))
3127 12473099 : continue;
3128 12453520 : e = single_pred_edge (bb);
3129 :
3130 12453520 : if (!(e->flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
3131 3602285 : continue;
3132 :
3133 17702470 : gcond *stmt = as_a <gcond *> (*gsi_last_bb (e->src));
3134 8851235 : cond = fold_build2 (gimple_cond_code (stmt),
3135 : boolean_type_node,
3136 : gimple_cond_lhs (stmt),
3137 : gimple_cond_rhs (stmt));
3138 8851235 : if (e->flags & EDGE_FALSE_VALUE)
3139 4981859 : cond = invert_truthvalue (cond);
3140 8851235 : expanded = tree_simplify_using_condition (cond, expanded);
3141 : /* Break if EXPR is simplified to const values. */
3142 8851235 : if (expanded
3143 8851235 : && (integer_zerop (expanded) || integer_nonzerop (expanded)))
3144 : return expanded;
3145 :
3146 8780516 : ++cnt;
3147 : }
3148 :
3149 : /* Return the original expression if no simplification is done. */
3150 2776634 : return operand_equal_p (backup, expanded, 0) ? expr : expanded;
3151 2927709 : }
3152 :
3153 : /* Tries to simplify EXPR using the evolutions of the loop invariants
3154 : in the superloops of LOOP. Returns the simplified expression
3155 : (or EXPR unchanged, if no simplification was possible). */
3156 :
3157 : static tree
3158 6631927 : simplify_using_outer_evolutions (class loop *loop, tree expr)
3159 : {
3160 6631927 : enum tree_code code = TREE_CODE (expr);
3161 6631927 : bool changed;
3162 6631927 : tree e, e0, e1, e2;
3163 :
3164 6631927 : if (is_gimple_min_invariant (expr))
3165 : return expr;
3166 :
3167 1510728 : if (code == TRUTH_OR_EXPR
3168 1510728 : || code == TRUTH_AND_EXPR
3169 1510728 : || code == COND_EXPR)
3170 : {
3171 9146 : changed = false;
3172 :
3173 9146 : e0 = simplify_using_outer_evolutions (loop, TREE_OPERAND (expr, 0));
3174 9146 : if (TREE_OPERAND (expr, 0) != e0)
3175 0 : changed = true;
3176 :
3177 9146 : e1 = simplify_using_outer_evolutions (loop, TREE_OPERAND (expr, 1));
3178 9146 : if (TREE_OPERAND (expr, 1) != e1)
3179 0 : changed = true;
3180 :
3181 9146 : if (code == COND_EXPR)
3182 : {
3183 0 : e2 = simplify_using_outer_evolutions (loop, TREE_OPERAND (expr, 2));
3184 0 : if (TREE_OPERAND (expr, 2) != e2)
3185 : changed = true;
3186 : }
3187 : else
3188 : e2 = NULL_TREE;
3189 :
3190 9146 : if (changed)
3191 : {
3192 0 : if (code == COND_EXPR)
3193 0 : expr = fold_build3 (code, boolean_type_node, e0, e1, e2);
3194 : else
3195 0 : expr = fold_build2 (code, boolean_type_node, e0, e1);
3196 : }
3197 :
3198 : return expr;
3199 : }
3200 :
3201 1501582 : e = instantiate_parameters (loop, expr);
3202 1501582 : if (is_gimple_min_invariant (e))
3203 0 : return e;
3204 :
3205 : return expr;
3206 : }
3207 :
3208 : /* Returns true if EXIT is the only possible exit from LOOP. */
3209 :
3210 : bool
3211 13498056 : loop_only_exit_p (const class loop *loop, basic_block *body, const_edge exit)
3212 : {
3213 13498056 : gimple_stmt_iterator bsi;
3214 13498056 : unsigned i;
3215 :
3216 13498056 : if (exit != single_exit (loop))
3217 : return false;
3218 :
3219 44762358 : for (i = 0; i < loop->num_nodes; i++)
3220 302317881 : for (bsi = gsi_start_bb (body[i]); !gsi_end_p (bsi); gsi_next (&bsi))
3221 231902822 : if (stmt_can_terminate_bb_p (gsi_stmt (bsi)))
3222 : return false;
3223 :
3224 : return true;
3225 : }
3226 :
3227 : /* Stores description of number of iterations of LOOP derived from
3228 : EXIT (an exit edge of the LOOP) in NITER. Returns true if some useful
3229 : information could be derived (and fields of NITER have meaning described
3230 : in comments at class tree_niter_desc declaration), false otherwise.
3231 : When EVERY_ITERATION is true, only tests that are known to be executed
3232 : every iteration are considered (i.e. only test that alone bounds the loop).
3233 : If AT_STMT is not NULL, this function stores LOOP's condition statement in
3234 : it when returning true. */
3235 :
3236 : bool
3237 24819633 : number_of_iterations_exit_assumptions (class loop *loop, edge exit,
3238 : class tree_niter_desc *niter,
3239 : gcond **at_stmt, bool every_iteration,
3240 : basic_block *body)
3241 : {
3242 24819633 : tree type;
3243 24819633 : tree op0, op1;
3244 24819633 : enum tree_code code;
3245 24819633 : affine_iv iv0, iv1;
3246 24819633 : bool safe;
3247 :
3248 : /* The condition at a fake exit (if it exists) does not control its
3249 : execution. */
3250 24819633 : if (exit->flags & EDGE_FAKE)
3251 : return false;
3252 :
3253 : /* Nothing to analyze if the loop is known to be infinite. */
3254 24819188 : if (loop_constraint_set_p (loop, LOOP_C_INFINITE))
3255 : return false;
3256 :
3257 24819188 : safe = dominated_by_p (CDI_DOMINATORS, loop->latch, exit->src);
3258 :
3259 24819188 : if (every_iteration && !safe)
3260 : return false;
3261 :
3262 23674460 : niter->assumptions = boolean_false_node;
3263 23674460 : niter->control.base = NULL_TREE;
3264 23674460 : niter->control.step = NULL_TREE;
3265 23674460 : niter->control.no_overflow = false;
3266 47348920 : gcond *stmt = safe_dyn_cast <gcond *> (*gsi_last_bb (exit->src));
3267 21647336 : if (!stmt)
3268 : return false;
3269 :
3270 21647336 : if (at_stmt)
3271 20995247 : *at_stmt = stmt;
3272 :
3273 : /* We want the condition for staying inside loop. */
3274 21647336 : code = gimple_cond_code (stmt);
3275 21647336 : if (exit->flags & EDGE_TRUE_VALUE)
3276 8159695 : code = invert_tree_comparison (code, false);
3277 :
3278 21647336 : switch (code)
3279 : {
3280 18432403 : case GT_EXPR:
3281 18432403 : case GE_EXPR:
3282 18432403 : case LT_EXPR:
3283 18432403 : case LE_EXPR:
3284 18432403 : case NE_EXPR:
3285 18432403 : break;
3286 :
3287 3214576 : case EQ_EXPR:
3288 3214576 : return number_of_iterations_cltz (loop, exit, code, niter);
3289 :
3290 : default:
3291 : return false;
3292 : }
3293 :
3294 18432403 : op0 = gimple_cond_lhs (stmt);
3295 18432403 : op1 = gimple_cond_rhs (stmt);
3296 18432403 : type = TREE_TYPE (op0);
3297 :
3298 18432403 : if (!INTEGRAL_NB_TYPE_P (type)
3299 : && !POINTER_TYPE_P (type))
3300 : return false;
3301 :
3302 17666810 : tree iv0_niters = NULL_TREE;
3303 35333620 : if (!simple_iv_with_niters (loop, loop_containing_stmt (stmt),
3304 : op0, &iv0, safe ? &iv0_niters : NULL, false))
3305 3897926 : return number_of_iterations_bitcount (loop, exit, code, niter);
3306 13768884 : tree iv1_niters = NULL_TREE;
3307 27537768 : if (!simple_iv_with_niters (loop, loop_containing_stmt (stmt),
3308 : op1, &iv1, safe ? &iv1_niters : NULL, false))
3309 : return false;
3310 : /* Give up on complicated case. */
3311 13085692 : if (iv0_niters && iv1_niters)
3312 : return false;
3313 :
3314 13085692 : iv0.base = expand_simple_operations (iv0.base);
3315 13085692 : iv1.base = expand_simple_operations (iv1.base);
3316 13085692 : bool body_from_caller = true;
3317 13085692 : if (!body)
3318 : {
3319 7869604 : body = get_loop_body (loop);
3320 7869604 : body_from_caller = false;
3321 : }
3322 13085692 : bool only_exit_p = loop_only_exit_p (loop, body, exit);
3323 13085692 : if (!body_from_caller)
3324 7869604 : free (body);
3325 13085692 : if (!number_of_iterations_cond (loop, type, &iv0, code, &iv1, niter,
3326 : only_exit_p, safe))
3327 : {
3328 : return false;
3329 : }
3330 :
3331 : /* Incorporate additional assumption implied by control iv. */
3332 12942941 : tree iv_niters = iv0_niters ? iv0_niters : iv1_niters;
3333 12942941 : if (iv_niters)
3334 : {
3335 25902 : tree assumption = fold_build2 (LE_EXPR, boolean_type_node, niter->niter,
3336 : fold_convert (TREE_TYPE (niter->niter),
3337 : iv_niters));
3338 :
3339 25902 : if (!integer_nonzerop (assumption))
3340 25691 : niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
3341 : niter->assumptions, assumption);
3342 :
3343 : /* Refine upper bound if possible. */
3344 25902 : if (TREE_CODE (iv_niters) == INTEGER_CST
3345 51804 : && niter->max > wi::to_widest (iv_niters))
3346 22663 : niter->max = wi::to_widest (iv_niters);
3347 : }
3348 :
3349 : /* There is no assumptions if the loop is known to be finite. */
3350 12942941 : if (!integer_zerop (niter->assumptions)
3351 12942941 : && loop_constraint_set_p (loop, LOOP_C_FINITE))
3352 12134 : niter->assumptions = boolean_true_node;
3353 :
3354 12942941 : if (optimize >= 3)
3355 : {
3356 2204545 : niter->assumptions = simplify_using_outer_evolutions (loop,
3357 : niter->assumptions);
3358 2204545 : niter->may_be_zero = simplify_using_outer_evolutions (loop,
3359 : niter->may_be_zero);
3360 2204545 : niter->niter = simplify_using_outer_evolutions (loop, niter->niter);
3361 : }
3362 :
3363 12942941 : niter->assumptions
3364 12942941 : = simplify_using_initial_conditions (loop,
3365 : niter->assumptions);
3366 12942941 : niter->may_be_zero
3367 12942941 : = simplify_using_initial_conditions (loop,
3368 : niter->may_be_zero);
3369 :
3370 : /* If NITER has simplified into a constant, update MAX. */
3371 12942941 : if (TREE_CODE (niter->niter) == INTEGER_CST)
3372 7433051 : niter->max = wi::to_widest (niter->niter);
3373 :
3374 12942941 : return (!integer_zerop (niter->assumptions));
3375 : }
3376 :
3377 : /* Like number_of_iterations_exit_assumptions, but return TRUE only if
3378 : the niter information holds unconditionally. */
3379 :
3380 : bool
3381 24066910 : number_of_iterations_exit (class loop *loop, edge exit,
3382 : class tree_niter_desc *niter,
3383 : bool warn, bool every_iteration,
3384 : basic_block *body)
3385 : {
3386 24066910 : gcond *stmt;
3387 24066910 : if (!number_of_iterations_exit_assumptions (loop, exit, niter,
3388 : &stmt, every_iteration, body))
3389 : return false;
3390 :
3391 12586682 : if (integer_nonzerop (niter->assumptions))
3392 : return true;
3393 :
3394 602703 : if (warn && dump_enabled_p ())
3395 5 : dump_printf_loc (MSG_MISSED_OPTIMIZATION, stmt,
3396 : "missed loop optimization: niters analysis ends up "
3397 : "with assumptions.\n");
3398 :
3399 : return false;
3400 : }
3401 :
3402 : /* Try to determine the number of iterations of LOOP. If we succeed,
3403 : expression giving number of iterations is returned and *EXIT is
3404 : set to the edge from that the information is obtained. Otherwise
3405 : chrec_dont_know is returned. */
3406 :
3407 : tree
3408 778284 : find_loop_niter (class loop *loop, edge *exit)
3409 : {
3410 778284 : unsigned i;
3411 778284 : auto_vec<edge> exits = get_loop_exit_edges (loop);
3412 778284 : edge ex;
3413 778284 : tree niter = NULL_TREE, aniter;
3414 778284 : class tree_niter_desc desc;
3415 :
3416 778284 : *exit = NULL;
3417 3407172 : FOR_EACH_VEC_ELT (exits, i, ex)
3418 : {
3419 2628888 : if (!number_of_iterations_exit (loop, ex, &desc, false))
3420 2108735 : continue;
3421 :
3422 520153 : if (integer_nonzerop (desc.may_be_zero))
3423 : {
3424 : /* We exit in the first iteration through this exit.
3425 : We won't find anything better. */
3426 0 : niter = build_int_cst (unsigned_type_node, 0);
3427 0 : *exit = ex;
3428 0 : break;
3429 : }
3430 :
3431 520153 : if (!integer_zerop (desc.may_be_zero))
3432 56475 : continue;
3433 :
3434 463678 : aniter = desc.niter;
3435 :
3436 463678 : if (!niter)
3437 : {
3438 : /* Nothing recorded yet. */
3439 454265 : niter = aniter;
3440 454265 : *exit = ex;
3441 454265 : continue;
3442 : }
3443 :
3444 : /* Prefer constants, the lower the better. */
3445 9413 : if (TREE_CODE (aniter) != INTEGER_CST)
3446 5218 : continue;
3447 :
3448 4195 : if (TREE_CODE (niter) != INTEGER_CST)
3449 : {
3450 1016 : niter = aniter;
3451 1016 : *exit = ex;
3452 1016 : continue;
3453 : }
3454 :
3455 3179 : if (tree_int_cst_lt (aniter, niter))
3456 : {
3457 256 : niter = aniter;
3458 256 : *exit = ex;
3459 256 : continue;
3460 : }
3461 : }
3462 :
3463 778284 : return niter ? niter : chrec_dont_know;
3464 778284 : }
3465 :
3466 : /* Return true if loop is known to have bounded number of iterations. */
3467 :
3468 : bool
3469 3015300 : finite_loop_p (class loop *loop)
3470 : {
3471 3015300 : widest_int nit;
3472 3015300 : int flags;
3473 :
3474 3015300 : if (loop->finite_p)
3475 : {
3476 2236298 : unsigned i;
3477 2236298 : auto_vec<edge> exits = get_loop_exit_edges (loop);
3478 2236298 : edge ex;
3479 :
3480 : /* If the loop has a normal exit, we can assume it will terminate. */
3481 4485578 : FOR_EACH_VEC_ELT (exits, i, ex)
3482 2244488 : if (!(ex->flags & (EDGE_EH | EDGE_ABNORMAL | EDGE_FAKE)))
3483 : {
3484 2231834 : if (dump_file)
3485 177 : fprintf (dump_file, "Assume loop %i to be finite: it has an exit "
3486 : "and -ffinite-loops is on or loop was "
3487 : "previously finite.\n",
3488 : loop->num);
3489 2231834 : return true;
3490 : }
3491 2236298 : }
3492 :
3493 783466 : flags = flags_from_decl_or_type (current_function_decl);
3494 783466 : if ((flags & (ECF_CONST|ECF_PURE)) && !(flags & ECF_LOOPING_CONST_OR_PURE))
3495 : {
3496 2516 : if (dump_file && (dump_flags & TDF_DETAILS))
3497 0 : fprintf (dump_file,
3498 : "Found loop %i to be finite: it is within "
3499 : "pure or const function.\n",
3500 : loop->num);
3501 2516 : loop->finite_p = true;
3502 2516 : return true;
3503 : }
3504 :
3505 780950 : if (loop->any_upper_bound
3506 : /* Loop with no normal exit will not pass max_loop_iterations. */
3507 780950 : || (!loop->finite_p && max_loop_iterations (loop, &nit)))
3508 : {
3509 414530 : if (dump_file && (dump_flags & TDF_DETAILS))
3510 31 : fprintf (dump_file, "Found loop %i to be finite: upper bound found.\n",
3511 : loop->num);
3512 414530 : loop->finite_p = true;
3513 414530 : return true;
3514 : }
3515 :
3516 : return false;
3517 3015300 : }
3518 :
3519 : /*
3520 :
3521 : Analysis of a number of iterations of a loop by a brute-force evaluation.
3522 :
3523 : */
3524 :
3525 : /* Bound on the number of iterations we try to evaluate. */
3526 :
3527 : #define MAX_ITERATIONS_TO_TRACK \
3528 : ((unsigned) param_max_iterations_to_track)
3529 :
3530 : /* Returns the loop phi node of LOOP such that ssa name X is derived from its
3531 : result by a chain of operations such that all but exactly one of their
3532 : operands are constants. */
3533 :
3534 : static gphi *
3535 1980732 : chain_of_csts_start (class loop *loop, tree x)
3536 : {
3537 2377746 : gimple *stmt = SSA_NAME_DEF_STMT (x);
3538 2377746 : tree use;
3539 2377746 : basic_block bb = gimple_bb (stmt);
3540 2377746 : enum tree_code code;
3541 :
3542 2377746 : if (!bb
3543 2377746 : || !flow_bb_inside_loop_p (loop, bb))
3544 : return NULL;
3545 :
3546 1834492 : if (gimple_code (stmt) == GIMPLE_PHI)
3547 : {
3548 720681 : if (bb == loop->header)
3549 660938 : return as_a <gphi *> (stmt);
3550 :
3551 : return NULL;
3552 : }
3553 :
3554 1113811 : if (gimple_code (stmt) != GIMPLE_ASSIGN
3555 2103032 : || gimple_assign_rhs_class (stmt) == GIMPLE_TERNARY_RHS)
3556 : return NULL;
3557 :
3558 988949 : code = gimple_assign_rhs_code (stmt);
3559 988949 : if (gimple_references_memory_p (stmt)
3560 537917 : || TREE_CODE_CLASS (code) == tcc_reference
3561 512604 : || (code == ADDR_EXPR
3562 503 : && !is_gimple_min_invariant (gimple_assign_rhs1 (stmt))))
3563 : return NULL;
3564 :
3565 512101 : use = SINGLE_SSA_TREE_OPERAND (stmt, SSA_OP_USE);
3566 512101 : if (use == NULL_TREE)
3567 : return NULL;
3568 :
3569 : return chain_of_csts_start (loop, use);
3570 : }
3571 :
3572 : /* Determines whether the expression X is derived from a result of a phi node
3573 : in header of LOOP such that
3574 :
3575 : * the derivation of X consists only from operations with constants
3576 : * the initial value of the phi node is constant
3577 : * the value of the phi node in the next iteration can be derived from the
3578 : value in the current iteration by a chain of operations with constants,
3579 : or is also a constant
3580 :
3581 : If such phi node exists, it is returned, otherwise NULL is returned. */
3582 :
3583 : static gphi *
3584 1791581 : get_base_for (class loop *loop, tree x)
3585 : {
3586 1791581 : gphi *phi;
3587 1791581 : tree init, next;
3588 :
3589 1791581 : if (is_gimple_min_invariant (x))
3590 : return NULL;
3591 :
3592 1791581 : phi = chain_of_csts_start (loop, x);
3593 1791581 : if (!phi)
3594 : return NULL;
3595 :
3596 487199 : init = PHI_ARG_DEF_FROM_EDGE (phi, loop_preheader_edge (loop));
3597 487199 : next = PHI_ARG_DEF_FROM_EDGE (phi, loop_latch_edge (loop));
3598 :
3599 487199 : if (!is_gimple_min_invariant (init))
3600 : return NULL;
3601 :
3602 214166 : if (TREE_CODE (next) == SSA_NAME
3603 214166 : && chain_of_csts_start (loop, next) != phi)
3604 15447 : return NULL;
3605 :
3606 : return phi;
3607 : }
3608 :
3609 : /* Given an expression X, then
3610 :
3611 : * if X is NULL_TREE, we return the constant BASE.
3612 : * if X is a constant, we return the constant X.
3613 : * otherwise X is a SSA name, whose value in the considered loop is derived
3614 : by a chain of operations with constant from a result of a phi node in
3615 : the header of the loop. Then we return value of X when the value of the
3616 : result of this phi node is given by the constant BASE. */
3617 :
3618 : static tree
3619 7512879 : get_val_for (tree x, tree base)
3620 : {
3621 7520276 : gimple *stmt;
3622 :
3623 7520276 : gcc_checking_assert (is_gimple_min_invariant (base));
3624 :
3625 7520276 : if (!x)
3626 : return base;
3627 5266768 : else if (is_gimple_min_invariant (x))
3628 : return x;
3629 :
3630 5242148 : stmt = SSA_NAME_DEF_STMT (x);
3631 5242148 : if (gimple_code (stmt) == GIMPLE_PHI)
3632 : return base;
3633 :
3634 2784791 : gcc_checking_assert (is_gimple_assign (stmt));
3635 :
3636 : /* STMT must be either an assignment of a single SSA name or an
3637 : expression involving an SSA name and a constant. Try to fold that
3638 : expression using the value for the SSA name. */
3639 2784791 : if (gimple_assign_ssa_name_copy_p (stmt))
3640 7397 : return get_val_for (gimple_assign_rhs1 (stmt), base);
3641 2777394 : else if (gimple_assign_rhs_class (stmt) == GIMPLE_UNARY_RHS
3642 2777394 : && TREE_CODE (gimple_assign_rhs1 (stmt)) == SSA_NAME)
3643 286932 : return fold_build1 (gimple_assign_rhs_code (stmt),
3644 : TREE_TYPE (gimple_assign_lhs (stmt)),
3645 : get_val_for (gimple_assign_rhs1 (stmt), base));
3646 2490462 : else if (gimple_assign_rhs_class (stmt) == GIMPLE_BINARY_RHS)
3647 : {
3648 2490462 : tree rhs1 = gimple_assign_rhs1 (stmt);
3649 2490462 : tree rhs2 = gimple_assign_rhs2 (stmt);
3650 2490462 : if (TREE_CODE (rhs1) == SSA_NAME)
3651 2427852 : rhs1 = get_val_for (rhs1, base);
3652 62610 : else if (TREE_CODE (rhs2) == SSA_NAME)
3653 62610 : rhs2 = get_val_for (rhs2, base);
3654 : else
3655 0 : gcc_unreachable ();
3656 2490462 : return fold_build2 (gimple_assign_rhs_code (stmt),
3657 : TREE_TYPE (gimple_assign_lhs (stmt)), rhs1, rhs2);
3658 : }
3659 : else
3660 0 : gcc_unreachable ();
3661 : }
3662 :
3663 :
3664 : /* Tries to count the number of iterations of LOOP till it exits by EXIT
3665 : by brute force -- i.e. by determining the value of the operands of the
3666 : condition at EXIT in first few iterations of the loop (assuming that
3667 : these values are constant) and determining the first one in that the
3668 : condition is not satisfied. Returns the constant giving the number
3669 : of the iterations of LOOP if successful, chrec_dont_know otherwise. */
3670 :
3671 : tree
3672 1749825 : loop_niter_by_eval (class loop *loop, edge exit)
3673 : {
3674 1749825 : tree acnd;
3675 1749825 : tree op[2], val[2], next[2], aval[2];
3676 1749825 : gphi *phi;
3677 1749825 : unsigned i, j;
3678 1749825 : enum tree_code cmp;
3679 :
3680 3499650 : gcond *cond = safe_dyn_cast <gcond *> (*gsi_last_bb (exit->src));
3681 1749825 : if (!cond)
3682 128210 : return chrec_dont_know;
3683 :
3684 1621615 : cmp = gimple_cond_code (cond);
3685 1621615 : if (exit->flags & EDGE_TRUE_VALUE)
3686 587387 : cmp = invert_tree_comparison (cmp, false);
3687 :
3688 1621615 : switch (cmp)
3689 : {
3690 1621586 : case EQ_EXPR:
3691 1621586 : case NE_EXPR:
3692 1621586 : case GT_EXPR:
3693 1621586 : case GE_EXPR:
3694 1621586 : case LT_EXPR:
3695 1621586 : case LE_EXPR:
3696 1621586 : op[0] = gimple_cond_lhs (cond);
3697 1621586 : op[1] = gimple_cond_rhs (cond);
3698 1621586 : break;
3699 :
3700 29 : default:
3701 29 : return chrec_dont_know;
3702 : }
3703 :
3704 1848788 : for (j = 0; j < 2; j++)
3705 : {
3706 1820064 : if (is_gimple_min_invariant (op[j]))
3707 : {
3708 28483 : val[j] = op[j];
3709 28483 : next[j] = NULL_TREE;
3710 28483 : op[j] = NULL_TREE;
3711 : }
3712 : else
3713 : {
3714 1791581 : phi = get_base_for (loop, op[j]);
3715 1791581 : if (!phi)
3716 : {
3717 1592862 : gassign *def;
3718 1592862 : if (j == 0
3719 1423108 : && (cmp == NE_EXPR || cmp == EQ_EXPR)
3720 823178 : && TREE_CODE (op[0]) == SSA_NAME
3721 823178 : && TREE_CODE (op[1]) == INTEGER_CST
3722 526046 : && (def = dyn_cast <gassign *> (SSA_NAME_DEF_STMT (op[0])))
3723 1882238 : && gimple_assign_rhs_code (def) == MEM_REF)
3724 : {
3725 60853 : tree mem = gimple_assign_rhs1 (def);
3726 60853 : affine_iv iv;
3727 60853 : if (TYPE_MODE (TREE_TYPE (mem)) == TYPE_MODE (char_type_node)
3728 23583 : && simple_iv (loop, loop,
3729 23583 : TREE_OPERAND (mem, 0), &iv, false)
3730 17176 : && tree_fits_uhwi_p (TREE_OPERAND (mem, 1))
3731 78029 : && tree_fits_uhwi_p (iv.step))
3732 : {
3733 17176 : tree str, off;
3734 : /* iv.base can be &"Foo" but also (char *)&"Foo" + 1. */
3735 17176 : split_constant_offset (iv.base, &str, &off);
3736 17176 : STRIP_NOPS (str);
3737 17176 : if (TREE_CODE (str) == ADDR_EXPR
3738 3248 : && TREE_CODE (TREE_OPERAND (str, 0)) == STRING_CST
3739 18934 : && tree_fits_uhwi_p (off))
3740 : {
3741 1758 : str = TREE_OPERAND (str, 0);
3742 1758 : unsigned i = 0;
3743 1758 : for (unsigned HOST_WIDE_INT idx
3744 1758 : = (tree_to_uhwi (TREE_OPERAND (mem, 1))
3745 1758 : + tree_to_uhwi (off));
3746 26034 : idx < (unsigned)TREE_STRING_LENGTH (str)
3747 13017 : && i < MAX_ITERATIONS_TO_TRACK;
3748 11259 : idx += tree_to_uhwi (iv.step), ++i)
3749 : {
3750 12191 : int res = compare_tree_int
3751 12191 : (op[1], TREE_STRING_POINTER (str)[idx]);
3752 12191 : if ((cmp == NE_EXPR && res == 0)
3753 11278 : || (cmp == EQ_EXPR && res != 0))
3754 932 : return build_int_cst (unsigned_type_node, i);
3755 : }
3756 : }
3757 : }
3758 : }
3759 1591930 : return chrec_dont_know;
3760 : }
3761 198719 : val[j] = PHI_ARG_DEF_FROM_EDGE (phi, loop_preheader_edge (loop));
3762 198719 : next[j] = PHI_ARG_DEF_FROM_EDGE (phi, loop_latch_edge (loop));
3763 : }
3764 : }
3765 :
3766 1198503 : for (i = 0; i < MAX_ITERATIONS_TO_TRACK; i++)
3767 : {
3768 3592761 : for (j = 0; j < 2; j++)
3769 2395174 : aval[j] = get_val_for (op[j], val[j]);
3770 :
3771 1197587 : acnd = fold_binary (cmp, boolean_type_node, aval[0], aval[1]);
3772 1197587 : if (acnd && integer_zerop (acnd))
3773 : {
3774 27080 : if (dump_file && (dump_flags & TDF_DETAILS))
3775 3 : fprintf (dump_file,
3776 : "Proved that loop %d iterates %d times using brute force.\n",
3777 : loop->num, i);
3778 27080 : return build_int_cst (unsigned_type_node, i);
3779 : }
3780 :
3781 3510115 : for (j = 0; j < 2; j++)
3782 : {
3783 2340311 : aval[j] = val[j];
3784 2340311 : val[j] = get_val_for (next[j], val[j]);
3785 2340311 : if (!is_gimple_min_invariant (val[j]))
3786 703 : return chrec_dont_know;
3787 : }
3788 :
3789 : /* If the next iteration would use the same base values
3790 : as the current one, there is no point looping further,
3791 : all following iterations will be the same as this one. */
3792 1169804 : if (val[0] == aval[0] && val[1] == aval[1])
3793 : break;
3794 : }
3795 :
3796 941 : return chrec_dont_know;
3797 : }
3798 :
3799 : /* Finds the exit of the LOOP by that the loop exits after a constant
3800 : number of iterations and stores the exit edge to *EXIT. The constant
3801 : giving the number of iterations of LOOP is returned. The number of
3802 : iterations is determined using loop_niter_by_eval (i.e. by brute force
3803 : evaluation). If we are unable to find the exit for that loop_niter_by_eval
3804 : determines the number of iterations, chrec_dont_know is returned. */
3805 :
3806 : tree
3807 796677 : find_loop_niter_by_eval (class loop *loop, edge *exit)
3808 : {
3809 796677 : unsigned i;
3810 796677 : auto_vec<edge> exits = get_loop_exit_edges (loop);
3811 796677 : edge ex;
3812 796677 : tree niter = NULL_TREE, aniter;
3813 :
3814 796677 : *exit = NULL;
3815 :
3816 : /* Loops with multiple exits are expensive to handle and less important. */
3817 796677 : if (!flag_expensive_optimizations
3818 817324 : && exits.length () > 1)
3819 4357 : return chrec_dont_know;
3820 :
3821 2415941 : FOR_EACH_VEC_ELT (exits, i, ex)
3822 : {
3823 1623621 : if (!just_once_each_iteration_p (loop, ex->src))
3824 446820 : continue;
3825 :
3826 1176801 : aniter = loop_niter_by_eval (loop, ex);
3827 1176801 : if (chrec_contains_undetermined (aniter))
3828 1162059 : continue;
3829 :
3830 14791 : if (niter
3831 14742 : && !tree_int_cst_lt (aniter, niter))
3832 49 : continue;
3833 :
3834 14693 : niter = aniter;
3835 14693 : *exit = ex;
3836 : }
3837 :
3838 792320 : return niter ? niter : chrec_dont_know;
3839 796677 : }
3840 :
3841 : /*
3842 :
3843 : Analysis of upper bounds on number of iterations of a loop.
3844 :
3845 : */
3846 :
3847 : static widest_int derive_constant_upper_bound_ops (tree, tree,
3848 : enum tree_code, tree);
3849 :
3850 : /* Returns a constant upper bound on the value of the right-hand side of
3851 : an assignment statement STMT. */
3852 :
3853 : static widest_int
3854 332 : derive_constant_upper_bound_assign (gimple *stmt)
3855 : {
3856 332 : enum tree_code code = gimple_assign_rhs_code (stmt);
3857 332 : tree op0 = gimple_assign_rhs1 (stmt);
3858 332 : tree op1 = gimple_assign_rhs2 (stmt);
3859 :
3860 332 : return derive_constant_upper_bound_ops (TREE_TYPE (gimple_assign_lhs (stmt)),
3861 332 : op0, code, op1);
3862 : }
3863 :
3864 : /* Returns a constant upper bound on the value of expression VAL. VAL
3865 : is considered to be unsigned. If its type is signed, its value must
3866 : be nonnegative. */
3867 :
3868 : static widest_int
3869 11307704 : derive_constant_upper_bound (tree val)
3870 : {
3871 11307704 : enum tree_code code;
3872 11307704 : tree op0, op1, op2;
3873 :
3874 11307704 : extract_ops_from_tree (val, &code, &op0, &op1, &op2);
3875 11307704 : return derive_constant_upper_bound_ops (TREE_TYPE (val), op0, code, op1);
3876 : }
3877 :
3878 : /* Returns a constant upper bound on the value of expression OP0 CODE OP1,
3879 : whose type is TYPE. The expression is considered to be unsigned. If
3880 : its type is signed, its value must be nonnegative. */
3881 :
3882 : static widest_int
3883 11308036 : derive_constant_upper_bound_ops (tree type, tree op0,
3884 : enum tree_code code, tree op1)
3885 : {
3886 11308036 : tree subtype, maxt;
3887 11308036 : widest_int bnd, max, cst;
3888 11308036 : gimple *stmt;
3889 :
3890 11308036 : if (INTEGRAL_TYPE_P (type))
3891 11308036 : maxt = TYPE_MAX_VALUE (type);
3892 : else
3893 0 : maxt = upper_bound_in_type (type, type);
3894 :
3895 11308036 : max = wi::to_widest (maxt);
3896 :
3897 11308036 : switch (code)
3898 : {
3899 11242644 : case INTEGER_CST:
3900 11242644 : return wi::to_widest (op0);
3901 :
3902 1041 : CASE_CONVERT:
3903 1041 : subtype = TREE_TYPE (op0);
3904 1041 : if (!TYPE_UNSIGNED (subtype)
3905 : /* If TYPE is also signed, the fact that VAL is nonnegative implies
3906 : that OP0 is nonnegative. */
3907 762 : && TYPE_UNSIGNED (type)
3908 1803 : && !tree_expr_nonnegative_p (op0))
3909 : {
3910 : /* If we cannot prove that the casted expression is nonnegative,
3911 : we cannot establish more useful upper bound than the precision
3912 : of the type gives us. */
3913 674 : return max;
3914 : }
3915 :
3916 : /* We now know that op0 is an nonnegative value. Try deriving an upper
3917 : bound for it. */
3918 367 : bnd = derive_constant_upper_bound (op0);
3919 :
3920 : /* If the bound does not fit in TYPE, max. value of TYPE could be
3921 : attained. */
3922 367 : if (wi::ltu_p (max, bnd))
3923 0 : return max;
3924 :
3925 367 : return bnd;
3926 :
3927 34750 : case PLUS_EXPR:
3928 34750 : case POINTER_PLUS_EXPR:
3929 34750 : case MINUS_EXPR:
3930 34750 : if (TREE_CODE (op1) != INTEGER_CST
3931 34750 : || !tree_expr_nonnegative_p (op0))
3932 34315 : return max;
3933 :
3934 : /* Canonicalize to OP0 - CST. Consider CST to be signed, in order to
3935 : choose the most logical way how to treat this constant regardless
3936 : of the signedness of the type. */
3937 435 : cst = wi::sext (wi::to_widest (op1), TYPE_PRECISION (type));
3938 435 : if (code != MINUS_EXPR)
3939 435 : cst = -cst;
3940 :
3941 435 : bnd = derive_constant_upper_bound (op0);
3942 :
3943 435 : if (wi::neg_p (cst))
3944 : {
3945 28 : cst = -cst;
3946 : /* Avoid CST == 0x80000... */
3947 28 : if (wi::neg_p (cst))
3948 0 : return max;
3949 :
3950 : /* OP0 + CST. We need to check that
3951 : BND <= MAX (type) - CST. */
3952 :
3953 28 : widest_int mmax = max - cst;
3954 28 : if (wi::leu_p (bnd, mmax))
3955 0 : return max;
3956 :
3957 28 : return bnd + cst;
3958 28 : }
3959 : else
3960 : {
3961 : /* OP0 - CST, where CST >= 0.
3962 :
3963 : If TYPE is signed, we have already verified that OP0 >= 0, and we
3964 : know that the result is nonnegative. This implies that
3965 : VAL <= BND - CST.
3966 :
3967 : If TYPE is unsigned, we must additionally know that OP0 >= CST,
3968 : otherwise the operation underflows.
3969 : */
3970 :
3971 : /* This should only happen if the type is unsigned; however, for
3972 : buggy programs that use overflowing signed arithmetics even with
3973 : -fno-wrapv, this condition may also be true for signed values. */
3974 407 : if (wi::ltu_p (bnd, cst))
3975 0 : return max;
3976 :
3977 407 : if (TYPE_UNSIGNED (type))
3978 : {
3979 402 : tree tem = fold_binary (GE_EXPR, boolean_type_node, op0,
3980 : wide_int_to_tree (type, cst));
3981 402 : if (!tem || integer_nonzerop (tem))
3982 50 : return max;
3983 : }
3984 :
3985 357 : bnd -= cst;
3986 : }
3987 :
3988 357 : return bnd;
3989 :
3990 0 : case FLOOR_DIV_EXPR:
3991 0 : case EXACT_DIV_EXPR:
3992 0 : if (TREE_CODE (op1) != INTEGER_CST
3993 0 : || tree_int_cst_sign_bit (op1))
3994 0 : return max;
3995 :
3996 0 : bnd = derive_constant_upper_bound (op0);
3997 0 : return wi::udiv_floor (bnd, wi::to_widest (op1));
3998 :
3999 7 : case BIT_AND_EXPR:
4000 7 : if (TREE_CODE (op1) != INTEGER_CST
4001 7 : || tree_int_cst_sign_bit (op1))
4002 0 : return max;
4003 7 : return wi::to_widest (op1);
4004 :
4005 543 : case SSA_NAME:
4006 543 : stmt = SSA_NAME_DEF_STMT (op0);
4007 543 : if (gimple_code (stmt) != GIMPLE_ASSIGN
4008 543 : || gimple_assign_lhs (stmt) != op0)
4009 211 : return max;
4010 332 : return derive_constant_upper_bound_assign (stmt);
4011 :
4012 29051 : default:
4013 29051 : return max;
4014 : }
4015 11308049 : }
4016 :
4017 : /* Emit a -Waggressive-loop-optimizations warning if needed. */
4018 :
4019 : static void
4020 10371234 : do_warn_aggressive_loop_optimizations (class loop *loop,
4021 : widest_int i_bound, gimple *stmt)
4022 : {
4023 : /* Don't warn if the loop doesn't have known constant bound. */
4024 10371234 : if (!loop->nb_iterations
4025 10371234 : || TREE_CODE (loop->nb_iterations) != INTEGER_CST
4026 4860694 : || !warn_aggressive_loop_optimizations
4027 : /* To avoid warning multiple times for the same loop,
4028 : only start warning when we preserve loops. */
4029 4860526 : || (cfun->curr_properties & PROP_loops) == 0
4030 : /* Only warn once per loop. */
4031 4860526 : || loop->warned_aggressive_loop_optimizations
4032 : /* Only warn if undefined behavior gives us lower estimate than the
4033 : known constant bound. */
4034 10371234 : || wi::cmpu (i_bound, wi::to_widest (loop->nb_iterations)) >= 0
4035 : /* And undefined behavior happens unconditionally. */
4036 10371289 : || !dominated_by_p (CDI_DOMINATORS, loop->latch, gimple_bb (stmt)))
4037 10371179 : return;
4038 :
4039 55 : edge e = single_exit (loop);
4040 55 : if (e == NULL)
4041 : return;
4042 :
4043 55 : gimple *estmt = last_nondebug_stmt (e->src);
4044 55 : char buf[WIDE_INT_PRINT_BUFFER_SIZE], *p;
4045 55 : unsigned len;
4046 55 : if (print_dec_buf_size (i_bound, TYPE_SIGN (TREE_TYPE (loop->nb_iterations)),
4047 : &len))
4048 0 : p = XALLOCAVEC (char, len);
4049 : else
4050 : p = buf;
4051 55 : print_dec (i_bound, p, TYPE_SIGN (TREE_TYPE (loop->nb_iterations)));
4052 55 : auto_diagnostic_group d;
4053 55 : if (warning_at (gimple_location (stmt), OPT_Waggressive_loop_optimizations,
4054 : "iteration %s invokes undefined behavior", p))
4055 9 : inform (gimple_location (estmt), "within this loop");
4056 55 : loop->warned_aggressive_loop_optimizations = true;
4057 55 : }
4058 :
4059 : /* Records that AT_STMT is executed at most BOUND + 1 times in LOOP. IS_EXIT
4060 : is true if the loop is exited immediately after STMT, and this exit
4061 : is taken at last when the STMT is executed BOUND + 1 times.
4062 : REALISTIC is true if BOUND is expected to be close to the real number
4063 : of iterations. UPPER is true if we are sure the loop iterates at most
4064 : BOUND times. I_BOUND is a widest_int upper estimate on BOUND. */
4065 :
4066 : static void
4067 16116523 : record_estimate (class loop *loop, tree bound, const widest_int &i_bound,
4068 : gimple *at_stmt, bool is_exit, bool realistic, bool upper)
4069 : {
4070 16116523 : widest_int delta;
4071 :
4072 16116523 : if (dump_file && (dump_flags & TDF_DETAILS))
4073 : {
4074 126356 : fprintf (dump_file, "Statement %s", is_exit ? "(exit)" : "");
4075 70082 : print_gimple_stmt (dump_file, at_stmt, 0, TDF_SLIM);
4076 70771 : fprintf (dump_file, " is %sexecuted at most ",
4077 : upper ? "" : "probably ");
4078 70082 : print_generic_expr (dump_file, bound, TDF_SLIM);
4079 70082 : fprintf (dump_file, " (bounded by ");
4080 70082 : print_decu (i_bound, dump_file);
4081 70082 : fprintf (dump_file, ") + 1 times in loop %d.\n", loop->num);
4082 : }
4083 :
4084 : /* If the I_BOUND is just an estimate of BOUND, it rarely is close to the
4085 : real number of iterations. */
4086 16116523 : if (TREE_CODE (bound) != INTEGER_CST)
4087 : realistic = false;
4088 : else
4089 14239443 : gcc_checking_assert (i_bound == wi::to_widest (bound));
4090 :
4091 16116523 : if (wi::min_precision (i_bound, SIGNED) > bound_wide_int ().get_precision ())
4092 : return;
4093 :
4094 : /* If we have a guaranteed upper bound, record it in the appropriate
4095 : list, unless this is an !is_exit bound (i.e. undefined behavior in
4096 : at_stmt) in a loop with known constant number of iterations. */
4097 16116496 : if (upper
4098 15575201 : && (is_exit
4099 10765594 : || loop->nb_iterations == NULL_TREE
4100 10765594 : || TREE_CODE (loop->nb_iterations) != INTEGER_CST))
4101 : {
4102 10517506 : class nb_iter_bound *elt = ggc_alloc<nb_iter_bound> ();
4103 :
4104 10517506 : elt->bound = bound_wide_int::from (i_bound, SIGNED);
4105 10517506 : elt->stmt = at_stmt;
4106 10517506 : elt->is_exit = is_exit;
4107 10517506 : elt->next = loop->bounds;
4108 10517506 : loop->bounds = elt;
4109 : }
4110 :
4111 : /* If statement is executed on every path to the loop latch, we can directly
4112 : infer the upper bound on the # of iterations of the loop. */
4113 16116496 : if (!dominated_by_p (CDI_DOMINATORS, loop->latch, gimple_bb (at_stmt)))
4114 466033 : upper = false;
4115 :
4116 : /* Update the number of iteration estimates according to the bound.
4117 : If at_stmt is an exit then the loop latch is executed at most BOUND times,
4118 : otherwise it can be executed BOUND + 1 times. We will lower the estimate
4119 : later if such statement must be executed on last iteration */
4120 16116496 : if (is_exit)
4121 4809613 : delta = 0;
4122 : else
4123 11306883 : delta = 1;
4124 16116496 : widest_int new_i_bound = i_bound + delta;
4125 :
4126 : /* If an overflow occurred, ignore the result. */
4127 16116496 : if (wi::ltu_p (new_i_bound, delta))
4128 0 : return;
4129 :
4130 16116496 : if (upper && !is_exit)
4131 10371234 : do_warn_aggressive_loop_optimizations (loop, new_i_bound, at_stmt);
4132 16116496 : record_niter_bound (loop, new_i_bound, realistic, upper);
4133 16116523 : }
4134 :
4135 : /* Records the control iv analyzed in NITER for LOOP if the iv is valid
4136 : and doesn't overflow. */
4137 :
4138 : static void
4139 4809621 : record_control_iv (class loop *loop, class tree_niter_desc *niter)
4140 : {
4141 4809621 : struct control_iv *iv;
4142 :
4143 4809621 : if (!niter->control.base || !niter->control.step)
4144 : return;
4145 :
4146 4771812 : if (!integer_onep (niter->assumptions) || !niter->control.no_overflow)
4147 : return;
4148 :
4149 4621890 : iv = ggc_alloc<control_iv> ();
4150 4621890 : iv->base = niter->control.base;
4151 4621890 : iv->step = niter->control.step;
4152 4621890 : iv->next = loop->control_ivs;
4153 4621890 : loop->control_ivs = iv;
4154 :
4155 4621890 : return;
4156 : }
4157 :
4158 : /* This function returns TRUE if below conditions are satisfied:
4159 : 1) VAR is SSA variable.
4160 : 2) VAR is an IV:{base, step} in its defining loop.
4161 : 3) IV doesn't overflow.
4162 : 4) Both base and step are integer constants.
4163 : 5) Base is the MIN/MAX value depends on IS_MIN.
4164 : Store value of base to INIT correspondingly. */
4165 :
4166 : static bool
4167 121009 : get_cst_init_from_scev (tree var, wide_int *init, bool is_min)
4168 : {
4169 121009 : if (TREE_CODE (var) != SSA_NAME)
4170 : return false;
4171 :
4172 121009 : gimple *def_stmt = SSA_NAME_DEF_STMT (var);
4173 121009 : class loop *loop = loop_containing_stmt (def_stmt);
4174 :
4175 108678 : if (loop == NULL)
4176 : return false;
4177 :
4178 108678 : affine_iv iv;
4179 108678 : if (!simple_iv (loop, loop, var, &iv, false))
4180 : return false;
4181 :
4182 5937 : if (!iv.no_overflow)
4183 : return false;
4184 :
4185 5882 : if (TREE_CODE (iv.base) != INTEGER_CST || TREE_CODE (iv.step) != INTEGER_CST)
4186 : return false;
4187 :
4188 4914 : if (is_min == tree_int_cst_sign_bit (iv.step))
4189 : return false;
4190 :
4191 4687 : *init = wi::to_wide (iv.base);
4192 4687 : return true;
4193 : }
4194 :
4195 : /* Record the estimate on number of iterations of LOOP based on the fact that
4196 : the induction variable BASE + STEP * i evaluated in STMT does not wrap and
4197 : its values belong to the range <LOW, HIGH>. REALISTIC is true if the
4198 : estimated number of iterations is expected to be close to the real one.
4199 : UPPER is true if we are sure the induction variable does not wrap. */
4200 :
4201 : static void
4202 11306896 : record_nonwrapping_iv (class loop *loop, tree base, tree step, gimple *stmt,
4203 : tree low, tree high, bool realistic, bool upper)
4204 : {
4205 11306896 : tree niter_bound, extreme, delta;
4206 11306896 : tree type = TREE_TYPE (base), unsigned_type;
4207 11306896 : tree orig_base = base;
4208 :
4209 11306896 : if (TREE_CODE (step) != INTEGER_CST || integer_zerop (step))
4210 0 : return;
4211 :
4212 11306896 : if (dump_file && (dump_flags & TDF_DETAILS))
4213 : {
4214 56274 : fprintf (dump_file, "Induction variable (");
4215 56274 : print_generic_expr (dump_file, TREE_TYPE (base), TDF_SLIM);
4216 56274 : fprintf (dump_file, ") ");
4217 56274 : print_generic_expr (dump_file, base, TDF_SLIM);
4218 56274 : fprintf (dump_file, " + ");
4219 56274 : print_generic_expr (dump_file, step, TDF_SLIM);
4220 56274 : fprintf (dump_file, " * iteration does not wrap in statement ");
4221 56274 : print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
4222 56274 : fprintf (dump_file, " in loop %d.\n", loop->num);
4223 : }
4224 :
4225 11306896 : unsigned_type = unsigned_type_for (type);
4226 11306896 : base = fold_convert (unsigned_type, base);
4227 11306896 : step = fold_convert (unsigned_type, step);
4228 :
4229 11306896 : if (tree_int_cst_sign_bit (step))
4230 : {
4231 353171 : wide_int max;
4232 353171 : value_range base_range (TREE_TYPE (orig_base));
4233 706342 : if (get_range_query (cfun)->range_of_expr (base_range, orig_base)
4234 353171 : && !base_range.undefined_p ())
4235 353092 : max = wi::to_wide (base_range.ubound ());
4236 353171 : extreme = fold_convert (unsigned_type, low);
4237 353171 : if (TREE_CODE (orig_base) == SSA_NAME
4238 20017 : && TREE_CODE (high) == INTEGER_CST
4239 20017 : && INTEGRAL_TYPE_P (TREE_TYPE (orig_base))
4240 19943 : && ((!base_range.varying_p ()
4241 6772 : && !base_range.undefined_p ())
4242 13250 : || get_cst_init_from_scev (orig_base, &max, false))
4243 359864 : && wi::gts_p (wi::to_wide (high), max))
4244 1956 : base = wide_int_to_tree (unsigned_type, max);
4245 351215 : else if (TREE_CODE (base) != INTEGER_CST
4246 563361 : && dominated_by_p (CDI_DOMINATORS,
4247 212146 : loop->latch, gimple_bb (stmt)))
4248 210219 : base = fold_convert (unsigned_type, high);
4249 353171 : delta = fold_build2 (MINUS_EXPR, unsigned_type, base, extreme);
4250 353171 : step = fold_build1 (NEGATE_EXPR, unsigned_type, step);
4251 353172 : }
4252 : else
4253 : {
4254 10953725 : wide_int min;
4255 10953725 : value_range base_range (TREE_TYPE (orig_base));
4256 21907450 : if (get_range_query (cfun)->range_of_expr (base_range, orig_base)
4257 10953725 : && !base_range.undefined_p ())
4258 10951476 : min = wi::to_wide (base_range.lbound ());
4259 10953725 : extreme = fold_convert (unsigned_type, high);
4260 10953725 : if (TREE_CODE (orig_base) == SSA_NAME
4261 863857 : && TREE_CODE (low) == INTEGER_CST
4262 863857 : && INTEGRAL_TYPE_P (TREE_TYPE (orig_base))
4263 255329 : && ((!base_range.varying_p ()
4264 149566 : && !base_range.undefined_p ())
4265 107759 : || get_cst_init_from_scev (orig_base, &min, true))
4266 11105982 : && wi::gts_p (min, wi::to_wide (low)))
4267 12331 : base = wide_int_to_tree (unsigned_type, min);
4268 10941394 : else if (TREE_CODE (base) != INTEGER_CST
4269 14357824 : && dominated_by_p (CDI_DOMINATORS,
4270 3416430 : loop->latch, gimple_bb (stmt)))
4271 3354099 : base = fold_convert (unsigned_type, low);
4272 10953725 : delta = fold_build2 (MINUS_EXPR, unsigned_type, extreme, base);
4273 10953737 : }
4274 :
4275 : /* STMT is executed at most NITER_BOUND + 1 times, since otherwise the value
4276 : would get out of the range. */
4277 11306896 : niter_bound = fold_build2 (FLOOR_DIV_EXPR, unsigned_type, delta, step);
4278 11306896 : widest_int max = derive_constant_upper_bound (niter_bound);
4279 11306896 : record_estimate (loop, niter_bound, max, stmt, false, realistic, upper);
4280 11306896 : }
4281 :
4282 : /* Determine information about number of iterations a LOOP from the index
4283 : IDX of a data reference accessed in STMT. RELIABLE is true if STMT is
4284 : guaranteed to be executed in every iteration of LOOP. Callback for
4285 : for_each_index. */
4286 :
4287 : struct ilb_data
4288 : {
4289 : class loop *loop;
4290 : gimple *stmt;
4291 : };
4292 :
4293 : static bool
4294 35984774 : idx_infer_loop_bounds (tree base, tree *idx, void *dta)
4295 : {
4296 35984774 : struct ilb_data *data = (struct ilb_data *) dta;
4297 35984774 : tree ev, init, step;
4298 35984774 : tree low, high, type, next;
4299 35984774 : bool sign, upper = true, has_flexible_size = false;
4300 35984774 : class loop *loop = data->loop;
4301 :
4302 35984774 : if (TREE_CODE (base) != ARRAY_REF)
4303 : return true;
4304 :
4305 : /* For arrays that might have flexible sizes, it is not guaranteed that they
4306 : do not really extend over their declared size. */
4307 9383521 : if (array_ref_flexible_size_p (base))
4308 : {
4309 2361028 : has_flexible_size = true;
4310 2361028 : upper = false;
4311 : }
4312 :
4313 9383521 : class loop *dloop = loop_containing_stmt (data->stmt);
4314 9383521 : if (!dloop)
4315 : return true;
4316 :
4317 9383521 : ev = analyze_scalar_evolution (dloop, *idx);
4318 9383521 : ev = instantiate_parameters (loop, ev);
4319 9383521 : init = initial_condition (ev);
4320 9383521 : step = evolution_part_in_loop_num (ev, loop->num);
4321 :
4322 9383521 : if (!init
4323 9383521 : || !step
4324 5651872 : || TREE_CODE (step) != INTEGER_CST
4325 4305001 : || integer_zerop (step)
4326 4305001 : || tree_contains_chrecs (init, NULL)
4327 13688522 : || chrec_contains_symbols_defined_in_loop (init, loop->num))
4328 : return true;
4329 :
4330 4305001 : low = array_ref_low_bound (base);
4331 4305001 : high = array_ref_up_bound (base);
4332 :
4333 : /* The case of nonconstant bounds could be handled, but it would be
4334 : complicated. */
4335 4305001 : if (TREE_CODE (low) != INTEGER_CST
4336 4305001 : || !high
4337 3604284 : || TREE_CODE (high) != INTEGER_CST)
4338 : return true;
4339 3397803 : sign = tree_int_cst_sign_bit (step);
4340 3397803 : type = TREE_TYPE (step);
4341 :
4342 : /* The array that might have flexible size most likely extends
4343 : beyond its bounds. */
4344 3397803 : if (has_flexible_size
4345 3397803 : && operand_equal_p (low, high, 0))
4346 : return true;
4347 :
4348 : /* In case the relevant bound of the array does not fit in type, or
4349 : it does, but bound + step (in type) still belongs into the range of the
4350 : array, the index may wrap and still stay within the range of the array
4351 : (consider e.g. if the array is indexed by the full range of
4352 : unsigned char).
4353 :
4354 : To make things simpler, we require both bounds to fit into type, although
4355 : there are cases where this would not be strictly necessary. */
4356 3379411 : if (!int_fits_type_p (high, type)
4357 3379379 : || !int_fits_type_p (low, type))
4358 : return true;
4359 3379379 : low = fold_convert (type, low);
4360 3379379 : high = fold_convert (type, high);
4361 :
4362 3379379 : if (sign)
4363 63254 : next = fold_binary (PLUS_EXPR, type, low, step);
4364 : else
4365 3316125 : next = fold_binary (PLUS_EXPR, type, high, step);
4366 :
4367 3379379 : if (tree_int_cst_compare (low, next) <= 0
4368 3379379 : && tree_int_cst_compare (next, high) <= 0)
4369 : return true;
4370 :
4371 : /* If access is not executed on every iteration, we must ensure that overflow
4372 : may not make the access valid later. */
4373 3379379 : if (!dominated_by_p (CDI_DOMINATORS, loop->latch, gimple_bb (data->stmt)))
4374 : {
4375 457655 : if (scev_probably_wraps_p (NULL_TREE,
4376 457655 : initial_condition_in_loop_num (ev, loop->num),
4377 : step, data->stmt, loop, true))
4378 3379379 : upper = false;
4379 : }
4380 : else
4381 2921724 : record_nonwrapping_chrec (ev);
4382 :
4383 3379379 : record_nonwrapping_iv (loop, init, step, data->stmt, low, high, false, upper);
4384 3379379 : return true;
4385 : }
4386 :
4387 : /* Determine information about number of iterations a LOOP from the bounds
4388 : of arrays in the data reference REF accessed in STMT. RELIABLE is true if
4389 : STMT is guaranteed to be executed in every iteration of LOOP.*/
4390 :
4391 : static void
4392 36844884 : infer_loop_bounds_from_ref (class loop *loop, gimple *stmt, tree ref)
4393 : {
4394 36844884 : struct ilb_data data;
4395 :
4396 36844884 : data.loop = loop;
4397 36844884 : data.stmt = stmt;
4398 0 : for_each_index (&ref, idx_infer_loop_bounds, &data);
4399 0 : }
4400 :
4401 : /* Determine information about number of iterations of a LOOP from the way
4402 : arrays are used in STMT. RELIABLE is true if STMT is guaranteed to be
4403 : executed in every iteration of LOOP. */
4404 :
4405 : static void
4406 271958416 : infer_loop_bounds_from_array (class loop *loop, gimple *stmt)
4407 : {
4408 271958416 : if (is_gimple_assign (stmt))
4409 : {
4410 104942054 : tree op0 = gimple_assign_lhs (stmt);
4411 104942054 : tree op1 = gimple_assign_rhs1 (stmt);
4412 :
4413 : /* For each memory access, analyze its access function
4414 : and record a bound on the loop iteration domain. */
4415 104942054 : if (REFERENCE_CLASS_P (op0))
4416 14362327 : infer_loop_bounds_from_ref (loop, stmt, op0);
4417 :
4418 104942054 : if (REFERENCE_CLASS_P (op1))
4419 22306686 : infer_loop_bounds_from_ref (loop, stmt, op1);
4420 : }
4421 167016362 : else if (is_gimple_call (stmt))
4422 : {
4423 7459048 : tree arg, lhs;
4424 7459048 : unsigned i, n = gimple_call_num_args (stmt);
4425 :
4426 7459048 : lhs = gimple_call_lhs (stmt);
4427 7459048 : if (lhs && REFERENCE_CLASS_P (lhs))
4428 5428 : infer_loop_bounds_from_ref (loop, stmt, lhs);
4429 :
4430 24216622 : for (i = 0; i < n; i++)
4431 : {
4432 16757574 : arg = gimple_call_arg (stmt, i);
4433 16757574 : if (REFERENCE_CLASS_P (arg))
4434 170443 : infer_loop_bounds_from_ref (loop, stmt, arg);
4435 : }
4436 : }
4437 271958416 : }
4438 :
4439 : /* Determine information about number of iterations of a LOOP from the fact
4440 : that pointer arithmetics in STMT does not overflow. */
4441 :
4442 : static void
4443 134548984 : infer_loop_bounds_from_pointer_arith (class loop *loop, gimple *stmt)
4444 : {
4445 134548984 : tree def, base, step, scev, type, low, high;
4446 134548984 : tree var, ptr;
4447 :
4448 134548984 : if (!is_gimple_assign (stmt)
4449 134548984 : || gimple_assign_rhs_code (stmt) != POINTER_PLUS_EXPR)
4450 : return;
4451 :
4452 4578689 : def = gimple_assign_lhs (stmt);
4453 4578689 : if (TREE_CODE (def) != SSA_NAME)
4454 : return;
4455 :
4456 4578689 : type = TREE_TYPE (def);
4457 4578689 : if (!nowrap_type_p (type))
4458 : return;
4459 :
4460 4578352 : ptr = gimple_assign_rhs1 (stmt);
4461 4578352 : if (!expr_invariant_in_loop_p (loop, ptr))
4462 : return;
4463 :
4464 2578914 : var = gimple_assign_rhs2 (stmt);
4465 2578914 : if (TYPE_PRECISION (type) != TYPE_PRECISION (TREE_TYPE (var)))
4466 : return;
4467 :
4468 2578914 : class loop *uloop = loop_containing_stmt (stmt);
4469 2578914 : scev = instantiate_parameters (loop, analyze_scalar_evolution (uloop, def));
4470 2578914 : if (chrec_contains_undetermined (scev))
4471 : return;
4472 :
4473 2128717 : base = initial_condition_in_loop_num (scev, loop->num);
4474 2128717 : step = evolution_part_in_loop_num (scev, loop->num);
4475 :
4476 2128717 : if (!base || !step
4477 2080674 : || TREE_CODE (step) != INTEGER_CST
4478 1895075 : || tree_contains_chrecs (base, NULL)
4479 4023792 : || chrec_contains_symbols_defined_in_loop (base, loop->num))
4480 : return;
4481 :
4482 1895075 : low = lower_bound_in_type (type, type);
4483 1895075 : high = upper_bound_in_type (type, type);
4484 :
4485 : /* In C, pointer arithmetic p + 1 cannot use a NULL pointer, and p - 1 cannot
4486 : produce a NULL pointer. The contrary would mean NULL points to an object,
4487 : while NULL is supposed to compare unequal with the address of all objects.
4488 : Furthermore, p + 1 cannot produce a NULL pointer and p - 1 cannot use a
4489 : NULL pointer since that would mean wrapping, which we assume here not to
4490 : happen. So, we can exclude NULL from the valid range of pointer
4491 : arithmetic. */
4492 1895075 : if (flag_delete_null_pointer_checks && int_cst_value (low) == 0)
4493 1895063 : low = build_int_cstu (TREE_TYPE (low), TYPE_ALIGN_UNIT (TREE_TYPE (type)));
4494 :
4495 1895075 : record_nonwrapping_chrec (scev);
4496 1895075 : record_nonwrapping_iv (loop, base, step, stmt, low, high, false, true);
4497 : }
4498 :
4499 : /* Determine information about number of iterations of a LOOP from the fact
4500 : that signed arithmetics in STMT does not overflow. */
4501 :
4502 : static void
4503 134548984 : infer_loop_bounds_from_signedness (class loop *loop, gimple *stmt)
4504 : {
4505 134548984 : tree def, base, step, scev, type, low, high;
4506 :
4507 134548984 : if (gimple_code (stmt) != GIMPLE_ASSIGN)
4508 128516542 : return;
4509 :
4510 53093392 : def = gimple_assign_lhs (stmt);
4511 :
4512 53093392 : if (TREE_CODE (def) != SSA_NAME)
4513 : return;
4514 :
4515 45248993 : type = TREE_TYPE (def);
4516 45248993 : if (!INTEGRAL_TYPE_P (type)
4517 45248993 : || !TYPE_OVERFLOW_UNDEFINED (type))
4518 : return;
4519 :
4520 15220407 : scev = instantiate_parameters (loop, analyze_scalar_evolution (loop, def));
4521 15220407 : if (chrec_contains_undetermined (scev))
4522 : return;
4523 :
4524 7508623 : base = initial_condition_in_loop_num (scev, loop->num);
4525 7508623 : step = evolution_part_in_loop_num (scev, loop->num);
4526 :
4527 7508623 : if (!base || !step
4528 6687130 : || TREE_CODE (step) != INTEGER_CST
4529 6032442 : || tree_contains_chrecs (base, NULL)
4530 13541065 : || chrec_contains_symbols_defined_in_loop (base, loop->num))
4531 : return;
4532 :
4533 6032442 : low = lower_bound_in_type (type, type);
4534 6032442 : high = upper_bound_in_type (type, type);
4535 6032442 : int_range_max r (TREE_TYPE (def));
4536 12064884 : get_range_query (cfun)->range_of_expr (r, def);
4537 6032442 : if (!r.varying_p () && !r.undefined_p ())
4538 : {
4539 5521777 : low = wide_int_to_tree (type, r.lower_bound ());
4540 5521789 : high = wide_int_to_tree (type, r.upper_bound ());
4541 : }
4542 :
4543 6032442 : record_nonwrapping_chrec (scev);
4544 6032442 : record_nonwrapping_iv (loop, base, step, stmt, low, high, false, true);
4545 6032442 : }
4546 :
4547 : /* The following analyzers are extracting information on the bounds
4548 : of LOOP from the following undefined behaviors:
4549 :
4550 : - data references should not access elements over the statically
4551 : allocated size,
4552 :
4553 : - signed variables should not overflow when flag_wrapv is not set.
4554 : */
4555 :
4556 : static void
4557 6582924 : infer_loop_bounds_from_undefined (class loop *loop, basic_block *bbs)
4558 : {
4559 6582924 : unsigned i;
4560 6582924 : gimple_stmt_iterator bsi;
4561 6582924 : basic_block bb;
4562 6582924 : bool reliable;
4563 :
4564 45187619 : for (i = 0; i < loop->num_nodes; i++)
4565 : {
4566 38604695 : bb = bbs[i];
4567 :
4568 : /* If BB is not executed in each iteration of the loop, we cannot
4569 : use the operations in it to infer reliable upper bound on the
4570 : # of iterations of the loop. However, we can use it as a guess.
4571 : Reliable guesses come only from array bounds. */
4572 38604695 : reliable = dominated_by_p (CDI_DOMINATORS, loop->latch, bb);
4573 :
4574 349167806 : for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
4575 : {
4576 271958416 : gimple *stmt = gsi_stmt (bsi);
4577 :
4578 271958416 : infer_loop_bounds_from_array (loop, stmt);
4579 :
4580 271958416 : if (reliable)
4581 : {
4582 134548984 : infer_loop_bounds_from_signedness (loop, stmt);
4583 134548984 : infer_loop_bounds_from_pointer_arith (loop, stmt);
4584 : }
4585 : }
4586 :
4587 : }
4588 6582924 : }
4589 :
4590 : /* Compare wide ints, callback for qsort. */
4591 :
4592 : static int
4593 28126 : wide_int_cmp (const void *p1, const void *p2)
4594 : {
4595 28126 : const bound_wide_int *d1 = (const bound_wide_int *) p1;
4596 28126 : const bound_wide_int *d2 = (const bound_wide_int *) p2;
4597 28126 : return wi::cmpu (*d1, *d2);
4598 : }
4599 :
4600 : /* Return index of BOUND in BOUNDS array sorted in increasing order.
4601 : Lookup by binary search. */
4602 :
4603 : static int
4604 6550 : bound_index (const vec<bound_wide_int> &bounds, const bound_wide_int &bound)
4605 : {
4606 6550 : unsigned int end = bounds.length ();
4607 6550 : unsigned int begin = 0;
4608 :
4609 : /* Find a matching index by means of a binary search. */
4610 7483 : while (begin != end)
4611 : {
4612 7483 : unsigned int middle = (begin + end) / 2;
4613 7483 : bound_wide_int index = bounds[middle];
4614 :
4615 7483 : if (index == bound)
4616 6550 : return middle;
4617 933 : else if (wi::ltu_p (index, bound))
4618 235 : begin = middle + 1;
4619 : else
4620 : end = middle;
4621 : }
4622 0 : gcc_unreachable ();
4623 : }
4624 :
4625 : /* We recorded loop bounds only for statements dominating loop latch (and thus
4626 : executed each loop iteration). If there are any bounds on statements not
4627 : dominating the loop latch we can improve the estimate by walking the loop
4628 : body and seeing if every path from loop header to loop latch contains
4629 : some bounded statement. */
4630 :
4631 : static void
4632 6623991 : discover_iteration_bound_by_body_walk (class loop *loop)
4633 : {
4634 6623991 : class nb_iter_bound *elt;
4635 6623991 : auto_vec<bound_wide_int> bounds;
4636 6623991 : vec<vec<basic_block> > queues = vNULL;
4637 6623991 : vec<basic_block> queue = vNULL;
4638 6623991 : ptrdiff_t queue_index;
4639 6623991 : ptrdiff_t latch_index = 0;
4640 :
4641 : /* Discover what bounds may interest us. */
4642 17141497 : for (elt = loop->bounds; elt; elt = elt->next)
4643 : {
4644 10517506 : bound_wide_int bound = elt->bound;
4645 :
4646 : /* Exit terminates loop at given iteration, while non-exits produce undefined
4647 : effect on the next iteration. */
4648 10517506 : if (!elt->is_exit)
4649 : {
4650 5707899 : bound += 1;
4651 : /* If an overflow occurred, ignore the result. */
4652 5707899 : if (bound == 0)
4653 0 : continue;
4654 : }
4655 :
4656 10517506 : if (!loop->any_upper_bound
4657 10517506 : || wi::ltu_p (bound, loop->nb_iterations_upper_bound))
4658 6550 : bounds.safe_push (bound);
4659 : }
4660 :
4661 : /* Exit early if there is nothing to do. */
4662 6623991 : if (!bounds.exists ())
4663 6620373 : return;
4664 :
4665 3618 : if (dump_file && (dump_flags & TDF_DETAILS))
4666 16 : fprintf (dump_file, " Trying to walk loop body to reduce the bound.\n");
4667 :
4668 : /* Sort the bounds in decreasing order. */
4669 3618 : bounds.qsort (wide_int_cmp);
4670 :
4671 : /* For every basic block record the lowest bound that is guaranteed to
4672 : terminate the loop. */
4673 :
4674 3618 : hash_map<basic_block, ptrdiff_t> bb_bounds;
4675 18000 : for (elt = loop->bounds; elt; elt = elt->next)
4676 : {
4677 14382 : bound_wide_int bound = elt->bound;
4678 14382 : if (!elt->is_exit)
4679 : {
4680 10074 : bound += 1;
4681 : /* If an overflow occurred, ignore the result. */
4682 10074 : if (bound == 0)
4683 0 : continue;
4684 : }
4685 :
4686 14382 : if (!loop->any_upper_bound
4687 14382 : || wi::ltu_p (bound, loop->nb_iterations_upper_bound))
4688 : {
4689 6550 : ptrdiff_t index = bound_index (bounds, bound);
4690 6550 : ptrdiff_t *entry = bb_bounds.get (gimple_bb (elt->stmt));
4691 6550 : if (!entry)
4692 4745 : bb_bounds.put (gimple_bb (elt->stmt), index);
4693 1805 : else if ((ptrdiff_t)*entry > index)
4694 109 : *entry = index;
4695 : }
4696 : }
4697 :
4698 3618 : hash_map<basic_block, ptrdiff_t> block_priority;
4699 :
4700 : /* Perform shortest path discovery loop->header ... loop->latch.
4701 :
4702 : The "distance" is given by the smallest loop bound of basic block
4703 : present in the path and we look for path with largest smallest bound
4704 : on it.
4705 :
4706 : To avoid the need for fibonacci heap on double ints we simply compress
4707 : double ints into indexes to BOUNDS array and then represent the queue
4708 : as arrays of queues for every index.
4709 : Index of BOUNDS.length() means that the execution of given BB has
4710 : no bounds determined.
4711 :
4712 : VISITED is a pointer map translating basic block into smallest index
4713 : it was inserted into the priority queue with. */
4714 3618 : latch_index = -1;
4715 :
4716 : /* Start walk in loop header with index set to infinite bound. */
4717 3618 : queue_index = bounds.length ();
4718 3618 : queues.safe_grow_cleared (queue_index + 1, true);
4719 3618 : queue.safe_push (loop->header);
4720 3618 : queues[queue_index] = queue;
4721 3618 : block_priority.put (loop->header, queue_index);
4722 :
4723 17404 : for (; queue_index >= 0; queue_index--)
4724 : {
4725 10168 : if (latch_index < queue_index)
4726 : {
4727 41480 : while (queues[queue_index].length ())
4728 : {
4729 37542 : basic_block bb;
4730 37542 : ptrdiff_t bound_index = queue_index;
4731 37542 : edge e;
4732 37542 : edge_iterator ei;
4733 :
4734 37542 : queue = queues[queue_index];
4735 37542 : bb = queue.pop ();
4736 :
4737 : /* OK, we later inserted the BB with lower priority, skip it. */
4738 37542 : if (*block_priority.get (bb) > queue_index)
4739 0 : continue;
4740 :
4741 : /* See if we can improve the bound. */
4742 37542 : ptrdiff_t *entry = bb_bounds.get (bb);
4743 37542 : if (entry && *entry < bound_index)
4744 4262 : bound_index = *entry;
4745 :
4746 : /* Insert successors into the queue, watch for latch edge
4747 : and record greatest index we saw. */
4748 95820 : FOR_EACH_EDGE (e, ei, bb->succs)
4749 : {
4750 58278 : bool insert = false;
4751 :
4752 58278 : if (loop_exit_edge_p (loop, e))
4753 10175 : continue;
4754 :
4755 48103 : if (e == loop_latch_edge (loop)
4756 48103 : && latch_index < bound_index)
4757 : latch_index = bound_index;
4758 44485 : else if (!(entry = block_priority.get (e->dest)))
4759 : {
4760 35344 : insert = true;
4761 35344 : block_priority.put (e->dest, bound_index);
4762 : }
4763 9141 : else if (*entry < bound_index)
4764 : {
4765 158 : insert = true;
4766 158 : *entry = bound_index;
4767 : }
4768 :
4769 35502 : if (insert)
4770 35502 : queues[bound_index].safe_push (e->dest);
4771 : }
4772 : }
4773 : }
4774 10168 : queues[queue_index].release ();
4775 : }
4776 :
4777 3618 : gcc_assert (latch_index >= 0);
4778 3618 : if ((unsigned)latch_index < bounds.length ())
4779 : {
4780 241 : if (dump_file && (dump_flags & TDF_DETAILS))
4781 : {
4782 0 : fprintf (dump_file, "Found better loop bound ");
4783 0 : print_decu (bounds[latch_index], dump_file);
4784 0 : fprintf (dump_file, "\n");
4785 : }
4786 241 : record_niter_bound (loop, widest_int::from (bounds[latch_index],
4787 : SIGNED), false, true);
4788 : }
4789 :
4790 3618 : queues.release ();
4791 6623991 : }
4792 :
4793 : /* See if every path cross the loop goes through a statement that is known
4794 : to not execute at the last iteration. In that case we can decrease iteration
4795 : count by 1. */
4796 :
4797 : static void
4798 6623991 : maybe_lower_iteration_bound (class loop *loop)
4799 : {
4800 6623991 : hash_set<gimple *> *not_executed_last_iteration = NULL;
4801 6623991 : class nb_iter_bound *elt;
4802 6623991 : bool found_exit = false;
4803 6623991 : auto_vec<basic_block> queue;
4804 6623991 : bitmap visited;
4805 :
4806 : /* Collect all statements with interesting (i.e. lower than
4807 : nb_iterations_upper_bound) bound on them.
4808 :
4809 : TODO: Due to the way record_estimate choose estimates to store, the bounds
4810 : will be always nb_iterations_upper_bound-1. We can change this to record
4811 : also statements not dominating the loop latch and update the walk below
4812 : to the shortest path algorithm. */
4813 17141497 : for (elt = loop->bounds; elt; elt = elt->next)
4814 : {
4815 10517506 : if (!elt->is_exit
4816 10517506 : && wi::ltu_p (elt->bound, loop->nb_iterations_upper_bound))
4817 : {
4818 2202374 : if (!not_executed_last_iteration)
4819 1202295 : not_executed_last_iteration = new hash_set<gimple *>;
4820 2202374 : not_executed_last_iteration->add (elt->stmt);
4821 : }
4822 : }
4823 6623991 : if (!not_executed_last_iteration)
4824 5421696 : return;
4825 :
4826 : /* Start DFS walk in the loop header and see if we can reach the
4827 : loop latch or any of the exits (including statements with side
4828 : effects that may terminate the loop otherwise) without visiting
4829 : any of the statements known to have undefined effect on the last
4830 : iteration. */
4831 1202295 : queue.safe_push (loop->header);
4832 1202295 : visited = BITMAP_ALLOC (NULL);
4833 1202295 : bitmap_set_bit (visited, loop->header->index);
4834 1202295 : found_exit = false;
4835 :
4836 1295715 : do
4837 : {
4838 1295715 : basic_block bb = queue.pop ();
4839 1295715 : gimple_stmt_iterator gsi;
4840 1295715 : bool stmt_found = false;
4841 :
4842 : /* Loop for possible exits and statements bounding the execution. */
4843 6866171 : for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
4844 : {
4845 4338333 : gimple *stmt = gsi_stmt (gsi);
4846 4338333 : if (not_executed_last_iteration->contains (stmt))
4847 : {
4848 : stmt_found = true;
4849 63592 : break;
4850 : }
4851 4297389 : if (gimple_has_side_effects (stmt))
4852 : {
4853 : found_exit = true;
4854 : break;
4855 : }
4856 : }
4857 1295715 : if (found_exit)
4858 : break;
4859 :
4860 : /* If no bounding statement is found, continue the walk. */
4861 1273067 : if (!stmt_found)
4862 : {
4863 1232123 : edge e;
4864 1232123 : edge_iterator ei;
4865 :
4866 2075430 : FOR_EACH_EDGE (e, ei, bb->succs)
4867 : {
4868 1987267 : if (loop_exit_edge_p (loop, e)
4869 843939 : || e == loop_latch_edge (loop)
4870 : /* When exiting an inner loop, verify it is finite. */
4871 843904 : || (!flow_bb_inside_loop_p (bb->loop_father, e->dest)
4872 9574 : && !finite_loop_p (bb->loop_father))
4873 : /* When we enter an irreducible region and the entry
4874 : does not contain a bounding stmt assume it might be
4875 : infinite. */
4876 2830574 : || (bb->flags & BB_IRREDUCIBLE_LOOP))
4877 : {
4878 : found_exit = true;
4879 : break;
4880 : }
4881 843307 : if (bitmap_set_bit (visited, e->dest->index))
4882 818398 : queue.safe_push (e->dest);
4883 : }
4884 : }
4885 : }
4886 1273067 : while (queue.length () && !found_exit);
4887 :
4888 : /* If every path through the loop reach bounding statement before exit,
4889 : then we know the last iteration of the loop will have undefined effect
4890 : and we can decrease number of iterations. */
4891 :
4892 1202295 : if (!found_exit)
4893 : {
4894 35687 : if (dump_file && (dump_flags & TDF_DETAILS))
4895 68 : fprintf (dump_file, "Reducing loop iteration estimate by 1; "
4896 : "undefined statement must be executed at the last iteration.\n");
4897 71374 : record_niter_bound (loop, widest_int::from (loop->nb_iterations_upper_bound,
4898 107061 : SIGNED) - 1,
4899 : false, true);
4900 : }
4901 :
4902 1202295 : BITMAP_FREE (visited);
4903 1202295 : delete not_executed_last_iteration;
4904 6623991 : }
4905 :
4906 : /* Get expected upper bound for number of loop iterations for
4907 : BUILT_IN_EXPECT_WITH_PROBABILITY for a condition COND. */
4908 :
4909 : static tree
4910 5129168 : get_upper_bound_based_on_builtin_expr_with_prob (gcond *cond)
4911 : {
4912 5129168 : if (cond == NULL)
4913 : return NULL_TREE;
4914 :
4915 4972419 : tree lhs = gimple_cond_lhs (cond);
4916 4972419 : if (TREE_CODE (lhs) != SSA_NAME)
4917 : return NULL_TREE;
4918 :
4919 4972347 : gimple *stmt = SSA_NAME_DEF_STMT (gimple_cond_lhs (cond));
4920 4972347 : gcall *def = dyn_cast<gcall *> (stmt);
4921 188783 : if (def == NULL)
4922 : return NULL_TREE;
4923 :
4924 188783 : tree decl = gimple_call_fndecl (def);
4925 188783 : if (!decl
4926 160573 : || !fndecl_built_in_p (decl, BUILT_IN_EXPECT_WITH_PROBABILITY)
4927 188791 : || gimple_call_num_args (stmt) != 3)
4928 : return NULL_TREE;
4929 :
4930 8 : tree c = gimple_call_arg (def, 1);
4931 8 : tree condt = TREE_TYPE (lhs);
4932 8 : tree res = fold_build2 (gimple_cond_code (cond),
4933 : condt, c,
4934 : gimple_cond_rhs (cond));
4935 8 : if (TREE_CODE (res) != INTEGER_CST)
4936 : return NULL_TREE;
4937 :
4938 :
4939 8 : tree prob = gimple_call_arg (def, 2);
4940 8 : tree t = TREE_TYPE (prob);
4941 8 : tree one
4942 16 : = build_real_from_int_cst (t,
4943 8 : integer_one_node);
4944 8 : if (integer_zerop (res))
4945 4 : prob = fold_build2 (MINUS_EXPR, t, one, prob);
4946 8 : tree r = fold_build2 (RDIV_EXPR, t, one, prob);
4947 8 : if (TREE_CODE (r) != REAL_CST)
4948 : return NULL_TREE;
4949 :
4950 6 : HOST_WIDE_INT probi
4951 6 : = real_to_integer (TREE_REAL_CST_PTR (r));
4952 6 : return build_int_cst (condt, probi);
4953 : }
4954 :
4955 : /* Records estimates on numbers of iterations of LOOP. If USE_UNDEFINED_P
4956 : is true also use estimates derived from undefined behavior. */
4957 :
4958 : void
4959 26148545 : estimate_numbers_of_iterations (class loop *loop)
4960 : {
4961 26148545 : tree niter, type;
4962 26148545 : unsigned i;
4963 26148545 : class tree_niter_desc niter_desc;
4964 26148545 : edge ex;
4965 26148545 : widest_int bound;
4966 26148545 : edge likely_exit;
4967 :
4968 : /* Give up if we already have tried to compute an estimation. */
4969 26148545 : if (loop->estimate_state != EST_NOT_COMPUTED)
4970 19524554 : return;
4971 :
4972 6623991 : if (dump_file && (dump_flags & TDF_DETAILS))
4973 13928 : fprintf (dump_file, "Estimating # of iterations of loop %d\n", loop->num);
4974 :
4975 6623991 : loop->estimate_state = EST_AVAILABLE;
4976 :
4977 6623991 : sreal nit;
4978 6623991 : bool reliable;
4979 :
4980 : /* If we have a measured profile, use it to estimate the number of
4981 : iterations. Normally this is recorded by branch_prob right after
4982 : reading the profile. In case we however found a new loop, record the
4983 : information here.
4984 :
4985 : Explicitly check for profile status so we do not report
4986 : wrong prediction hitrates for guessed loop iterations heuristics.
4987 : Do not recompute already recorded bounds - we ought to be better on
4988 : updating iteration bounds than updating profile in general and thus
4989 : recomputing iteration bounds later in the compilation process will just
4990 : introduce random roundoff errors. */
4991 6623991 : if (!loop->any_estimate
4992 4527470 : && expected_loop_iterations_by_profile (loop, &nit, &reliable)
4993 10253387 : && reliable)
4994 : {
4995 19 : bound = nit.to_nearest_int ();
4996 19 : record_niter_bound (loop, bound, true, false);
4997 : }
4998 :
4999 : /* Ensure that loop->nb_iterations is computed if possible. If it turns out
5000 : to be constant, we avoid undefined behavior implied bounds and instead
5001 : diagnose those loops with -Waggressive-loop-optimizations. */
5002 6623991 : number_of_latch_executions (loop);
5003 :
5004 6623991 : basic_block *body = get_loop_body (loop);
5005 6623991 : auto_vec<edge> exits = get_loop_exit_edges (loop, body);
5006 6623991 : likely_exit = single_likely_exit (loop, exits);
5007 25276643 : FOR_EACH_VEC_ELT (exits, i, ex)
5008 : {
5009 12028661 : if (ex == likely_exit)
5010 : {
5011 5129168 : gimple *stmt = *gsi_last_bb (ex->src);
5012 5129168 : if (stmt != NULL)
5013 : {
5014 5129168 : gcond *cond = dyn_cast<gcond *> (stmt);
5015 5129168 : tree niter_bound
5016 5129168 : = get_upper_bound_based_on_builtin_expr_with_prob (cond);
5017 5129168 : if (niter_bound != NULL_TREE)
5018 : {
5019 6 : widest_int max = derive_constant_upper_bound (niter_bound);
5020 6 : record_estimate (loop, niter_bound, max, cond,
5021 : true, true, false);
5022 6 : }
5023 : }
5024 : }
5025 :
5026 12028661 : if (!number_of_iterations_exit (loop, ex, &niter_desc,
5027 : false, false, body))
5028 7219040 : continue;
5029 :
5030 4809621 : niter = niter_desc.niter;
5031 4809621 : type = TREE_TYPE (niter);
5032 4809621 : if (TREE_CODE (niter_desc.may_be_zero) != INTEGER_CST)
5033 742944 : niter = build3 (COND_EXPR, type, niter_desc.may_be_zero,
5034 : build_int_cst (type, 0),
5035 : niter);
5036 4809621 : record_estimate (loop, niter, niter_desc.max,
5037 : last_nondebug_stmt (ex->src),
5038 : true, ex == likely_exit, true);
5039 4809621 : record_control_iv (loop, &niter_desc);
5040 : }
5041 :
5042 6623991 : if (flag_aggressive_loop_optimizations)
5043 6582924 : infer_loop_bounds_from_undefined (loop, body);
5044 6623991 : free (body);
5045 :
5046 6623991 : discover_iteration_bound_by_body_walk (loop);
5047 :
5048 6623991 : maybe_lower_iteration_bound (loop);
5049 :
5050 : /* If we know the exact number of iterations of this loop, try to
5051 : not break code with undefined behavior by not recording smaller
5052 : maximum number of iterations. */
5053 6623991 : if (loop->nb_iterations
5054 6623991 : && TREE_CODE (loop->nb_iterations) == INTEGER_CST
5055 13247982 : && (wi::min_precision (wi::to_widest (loop->nb_iterations), SIGNED)
5056 2024916 : <= bound_wide_int ().get_precision ()))
5057 : {
5058 2024916 : loop->any_upper_bound = true;
5059 2024916 : loop->nb_iterations_upper_bound
5060 2024916 : = bound_wide_int::from (wi::to_widest (loop->nb_iterations), SIGNED);
5061 : }
5062 26148559 : }
5063 :
5064 : /* Sets NIT to the estimated number of executions of the latch of the
5065 : LOOP. If CONSERVATIVE is true, we must be sure that NIT is at least as
5066 : large as the number of iterations. If we have no reliable estimate,
5067 : the function returns false, otherwise returns true. */
5068 :
5069 : bool
5070 9286947 : estimated_loop_iterations (class loop *loop, widest_int *nit)
5071 : {
5072 : /* When SCEV information is available, try to update loop iterations
5073 : estimate. Otherwise just return whatever we recorded earlier. */
5074 9286947 : if (scev_initialized_p ())
5075 9286947 : estimate_numbers_of_iterations (loop);
5076 :
5077 9286947 : return (get_estimated_loop_iterations (loop, nit));
5078 : }
5079 :
5080 : /* Similar to estimated_loop_iterations, but returns the estimate only
5081 : if it fits to HOST_WIDE_INT. If this is not the case, or the estimate
5082 : on the number of iterations of LOOP could not be derived, returns -1. */
5083 :
5084 : HOST_WIDE_INT
5085 9020104 : estimated_loop_iterations_int (class loop *loop)
5086 : {
5087 9020104 : widest_int nit;
5088 9020104 : HOST_WIDE_INT hwi_nit;
5089 :
5090 9020104 : if (!estimated_loop_iterations (loop, &nit))
5091 : return -1;
5092 :
5093 3971483 : if (!wi::fits_shwi_p (nit))
5094 : return -1;
5095 3971477 : hwi_nit = nit.to_shwi ();
5096 :
5097 3971477 : return hwi_nit < 0 ? -1 : hwi_nit;
5098 9020104 : }
5099 :
5100 :
5101 : /* Sets NIT to an upper bound for the maximum number of executions of the
5102 : latch of the LOOP. If we have no reliable estimate, the function returns
5103 : false, otherwise returns true. */
5104 :
5105 : bool
5106 8653216 : max_loop_iterations (class loop *loop, widest_int *nit)
5107 : {
5108 : /* When SCEV information is available, try to update loop iterations
5109 : estimate. Otherwise just return whatever we recorded earlier. */
5110 8653216 : if (scev_initialized_p ())
5111 8645267 : estimate_numbers_of_iterations (loop);
5112 :
5113 8653216 : return get_max_loop_iterations (loop, nit);
5114 : }
5115 :
5116 : /* Similar to max_loop_iterations, but returns the estimate only
5117 : if it fits to HOST_WIDE_INT. If this is not the case, or the estimate
5118 : on the number of iterations of LOOP could not be derived, returns -1. */
5119 :
5120 : HOST_WIDE_INT
5121 2288847 : max_loop_iterations_int (class loop *loop)
5122 : {
5123 2288847 : widest_int nit;
5124 2288847 : HOST_WIDE_INT hwi_nit;
5125 :
5126 2288847 : if (!max_loop_iterations (loop, &nit))
5127 : return -1;
5128 :
5129 1731681 : if (!wi::fits_shwi_p (nit))
5130 : return -1;
5131 1658513 : hwi_nit = nit.to_shwi ();
5132 :
5133 1658513 : return hwi_nit < 0 ? -1 : hwi_nit;
5134 2288847 : }
5135 :
5136 : /* Sets NIT to an likely upper bound for the maximum number of executions of the
5137 : latch of the LOOP. If we have no reliable estimate, the function returns
5138 : false, otherwise returns true. */
5139 :
5140 : bool
5141 366545 : likely_max_loop_iterations (class loop *loop, widest_int *nit)
5142 : {
5143 : /* When SCEV information is available, try to update loop iterations
5144 : estimate. Otherwise just return whatever we recorded earlier. */
5145 366545 : if (scev_initialized_p ())
5146 366545 : estimate_numbers_of_iterations (loop);
5147 :
5148 366545 : return get_likely_max_loop_iterations (loop, nit);
5149 : }
5150 :
5151 : /* Similar to max_loop_iterations, but returns the estimate only
5152 : if it fits to HOST_WIDE_INT. If this is not the case, or the estimate
5153 : on the number of iterations of LOOP could not be derived, returns -1. */
5154 :
5155 : HOST_WIDE_INT
5156 108816 : likely_max_loop_iterations_int (class loop *loop)
5157 : {
5158 108816 : widest_int nit;
5159 108816 : HOST_WIDE_INT hwi_nit;
5160 :
5161 108816 : if (!likely_max_loop_iterations (loop, &nit))
5162 : return -1;
5163 :
5164 88344 : if (!wi::fits_shwi_p (nit))
5165 : return -1;
5166 80603 : hwi_nit = nit.to_shwi ();
5167 :
5168 80603 : return hwi_nit < 0 ? -1 : hwi_nit;
5169 108816 : }
5170 :
5171 : /* Returns an estimate for the number of executions of statements
5172 : in the LOOP. For statements before the loop exit, this exceeds
5173 : the number of execution of the latch by one. */
5174 :
5175 : HOST_WIDE_INT
5176 8846718 : estimated_stmt_executions_int (class loop *loop)
5177 : {
5178 8846718 : HOST_WIDE_INT nit = estimated_loop_iterations_int (loop);
5179 8846718 : HOST_WIDE_INT snit;
5180 :
5181 8846718 : if (nit == -1)
5182 : return -1;
5183 :
5184 3907071 : snit = (HOST_WIDE_INT) ((unsigned HOST_WIDE_INT) nit + 1);
5185 :
5186 : /* If the computation overflows, return -1. */
5187 3907071 : return snit < 0 ? -1 : snit;
5188 : }
5189 :
5190 : /* Sets NIT to the maximum number of executions of the latch of the
5191 : LOOP, plus one. If we have no reliable estimate, the function returns
5192 : false, otherwise returns true. */
5193 :
5194 : bool
5195 154 : max_stmt_executions (class loop *loop, widest_int *nit)
5196 : {
5197 154 : widest_int nit_minus_one;
5198 :
5199 154 : if (!max_loop_iterations (loop, nit))
5200 : return false;
5201 :
5202 154 : nit_minus_one = *nit;
5203 :
5204 154 : *nit += 1;
5205 :
5206 154 : return wi::gtu_p (*nit, nit_minus_one);
5207 154 : }
5208 :
5209 : /* Sets NIT to the estimated maximum number of executions of the latch of the
5210 : LOOP, plus one. If we have no likely estimate, the function returns
5211 : false, otherwise returns true. */
5212 :
5213 : bool
5214 257729 : likely_max_stmt_executions (class loop *loop, widest_int *nit)
5215 : {
5216 257729 : widest_int nit_minus_one;
5217 :
5218 257729 : if (!likely_max_loop_iterations (loop, nit))
5219 : return false;
5220 :
5221 145986 : nit_minus_one = *nit;
5222 :
5223 145986 : *nit += 1;
5224 :
5225 145986 : return wi::gtu_p (*nit, nit_minus_one);
5226 257729 : }
5227 :
5228 : /* Sets NIT to the estimated number of executions of the latch of the
5229 : LOOP, plus one. If we have no reliable estimate, the function returns
5230 : false, otherwise returns true. */
5231 :
5232 : bool
5233 266843 : estimated_stmt_executions (class loop *loop, widest_int *nit)
5234 : {
5235 266843 : widest_int nit_minus_one;
5236 :
5237 266843 : if (!estimated_loop_iterations (loop, nit))
5238 : return false;
5239 :
5240 7799 : nit_minus_one = *nit;
5241 :
5242 7799 : *nit += 1;
5243 :
5244 7799 : return wi::gtu_p (*nit, nit_minus_one);
5245 266843 : }
5246 :
5247 : /* Records estimates on numbers of iterations of loops. */
5248 :
5249 : void
5250 1245065 : estimate_numbers_of_iterations (function *fn)
5251 : {
5252 7493301 : for (auto loop : loops_list (fn, 0))
5253 3758106 : estimate_numbers_of_iterations (loop);
5254 1245065 : }
5255 :
5256 : /* Returns true if statement S1 dominates statement S2. */
5257 :
5258 : bool
5259 2129444 : stmt_dominates_stmt_p (gimple *s1, gimple *s2)
5260 : {
5261 2129444 : basic_block bb1 = gimple_bb (s1), bb2 = gimple_bb (s2);
5262 :
5263 2129444 : if (!bb1
5264 2129444 : || s1 == s2)
5265 : return true;
5266 :
5267 2119282 : if (bb1 == bb2)
5268 : {
5269 809301 : gimple_stmt_iterator bsi;
5270 :
5271 809301 : if (gimple_code (s2) == GIMPLE_PHI)
5272 : return false;
5273 :
5274 546724 : if (gimple_code (s1) == GIMPLE_PHI)
5275 : return true;
5276 :
5277 20267073 : for (bsi = gsi_start_bb (bb1); gsi_stmt (bsi) != s2; gsi_next (&bsi))
5278 19434553 : if (gsi_stmt (bsi) == s1)
5279 : return true;
5280 :
5281 : return false;
5282 : }
5283 :
5284 1309981 : return dominated_by_p (CDI_DOMINATORS, bb2, bb1);
5285 : }
5286 :
5287 : /* Returns true when we can prove that the number of executions of
5288 : STMT in the loop is at most NITER, according to the bound on
5289 : the number of executions of the statement NITER_BOUND->stmt recorded in
5290 : NITER_BOUND and fact that NITER_BOUND->stmt dominate STMT.
5291 :
5292 : ??? This code can become quite a CPU hog - we can have many bounds,
5293 : and large basic block forcing stmt_dominates_stmt_p to be queried
5294 : many times on a large basic blocks, so the whole thing is O(n^2)
5295 : for scev_probably_wraps_p invocation (that can be done n times).
5296 :
5297 : It would make more sense (and give better answers) to remember BB
5298 : bounds computed by discover_iteration_bound_by_body_walk. */
5299 :
5300 : static bool
5301 2295781 : n_of_executions_at_most (gimple *stmt,
5302 : class nb_iter_bound *niter_bound,
5303 : tree niter)
5304 : {
5305 2295781 : widest_int bound = widest_int::from (niter_bound->bound, SIGNED);
5306 2295781 : tree nit_type = TREE_TYPE (niter), e;
5307 2295781 : enum tree_code cmp;
5308 :
5309 2295781 : gcc_assert (TYPE_UNSIGNED (nit_type));
5310 :
5311 : /* If the bound does not even fit into NIT_TYPE, it cannot tell us that
5312 : the number of iterations is small. */
5313 2295781 : if (!wi::fits_to_tree_p (bound, nit_type))
5314 : return false;
5315 :
5316 : /* We know that NITER_BOUND->stmt is executed at most NITER_BOUND->bound + 1
5317 : times. This means that:
5318 :
5319 : -- if NITER_BOUND->is_exit is true, then everything after
5320 : it at most NITER_BOUND->bound times.
5321 :
5322 : -- If NITER_BOUND->is_exit is false, then if we can prove that when STMT
5323 : is executed, then NITER_BOUND->stmt is executed as well in the same
5324 : iteration then STMT is executed at most NITER_BOUND->bound + 1 times.
5325 :
5326 : If we can determine that NITER_BOUND->stmt is always executed
5327 : after STMT, then STMT is executed at most NITER_BOUND->bound + 2 times.
5328 : We conclude that if both statements belong to the same
5329 : basic block and STMT is before NITER_BOUND->stmt and there are no
5330 : statements with side effects in between. */
5331 :
5332 2129444 : if (niter_bound->is_exit)
5333 : {
5334 746348 : if (stmt == niter_bound->stmt
5335 746348 : || !stmt_dominates_stmt_p (niter_bound->stmt, stmt))
5336 : return false;
5337 : cmp = GE_EXPR;
5338 : }
5339 : else
5340 : {
5341 1383096 : if (!stmt_dominates_stmt_p (niter_bound->stmt, stmt))
5342 : {
5343 531492 : gimple_stmt_iterator bsi;
5344 531492 : if (gimple_bb (stmt) != gimple_bb (niter_bound->stmt)
5345 235550 : || gimple_code (stmt) == GIMPLE_PHI
5346 720483 : || gimple_code (niter_bound->stmt) == GIMPLE_PHI)
5347 348532 : return false;
5348 :
5349 : /* By stmt_dominates_stmt_p we already know that STMT appears
5350 : before NITER_BOUND->STMT. Still need to test that the loop
5351 : cannot be terinated by a side effect in between. */
5352 8389624 : for (bsi = gsi_for_stmt (stmt); gsi_stmt (bsi) != niter_bound->stmt;
5353 8200633 : gsi_next (&bsi))
5354 8201455 : if (gimple_has_side_effects (gsi_stmt (bsi)))
5355 : return false;
5356 188169 : bound += 1;
5357 376338 : if (bound == 0
5358 188169 : || !wi::fits_to_tree_p (bound, nit_type))
5359 : return false;
5360 : }
5361 : cmp = GT_EXPR;
5362 : }
5363 :
5364 1165302 : e = fold_binary (cmp, boolean_type_node,
5365 : niter, wide_int_to_tree (nit_type, bound));
5366 1165302 : return e && integer_nonzerop (e);
5367 2295781 : }
5368 :
5369 : /* Returns true if the arithmetics in TYPE can be assumed not to wrap. */
5370 :
5371 : bool
5372 87450258 : nowrap_type_p (tree type)
5373 : {
5374 12157117 : if ((ANY_INTEGRAL_TYPE_P (type) || POINTER_TYPE_P (type))
5375 99607371 : && TYPE_OVERFLOW_UNDEFINED (type))
5376 55817156 : return true;
5377 :
5378 : return false;
5379 : }
5380 :
5381 : /* Return true if we can prove LOOP is exited before evolution of induction
5382 : variable {BASE, STEP} overflows with respect to its type bound. */
5383 :
5384 : static bool
5385 3359503 : loop_exits_before_overflow (tree base, tree step,
5386 : gimple *at_stmt, class loop *loop)
5387 : {
5388 3359503 : widest_int niter;
5389 3359503 : struct control_iv *civ;
5390 3359503 : class nb_iter_bound *bound;
5391 3359503 : tree e, delta, step_abs, unsigned_base;
5392 3359503 : tree type = TREE_TYPE (step);
5393 3359503 : tree unsigned_type, valid_niter;
5394 :
5395 : /* Compute the number of iterations before we reach the bound of the
5396 : type, and verify that the loop is exited before this occurs. */
5397 3359503 : unsigned_type = unsigned_type_for (type);
5398 3359503 : unsigned_base = fold_convert (unsigned_type, base);
5399 :
5400 3359503 : if (tree_int_cst_sign_bit (step))
5401 : {
5402 478620 : tree extreme = fold_convert (unsigned_type,
5403 : lower_bound_in_type (type, type));
5404 478620 : delta = fold_build2 (MINUS_EXPR, unsigned_type, unsigned_base, extreme);
5405 478620 : step_abs = fold_build1 (NEGATE_EXPR, unsigned_type,
5406 : fold_convert (unsigned_type, step));
5407 : }
5408 : else
5409 : {
5410 2880883 : tree extreme = fold_convert (unsigned_type,
5411 : upper_bound_in_type (type, type));
5412 2880883 : delta = fold_build2 (MINUS_EXPR, unsigned_type, extreme, unsigned_base);
5413 2880883 : step_abs = fold_convert (unsigned_type, step);
5414 : }
5415 :
5416 3359503 : valid_niter = fold_build2 (FLOOR_DIV_EXPR, unsigned_type, delta, step_abs);
5417 :
5418 3359503 : estimate_numbers_of_iterations (loop);
5419 :
5420 3359503 : if (max_loop_iterations (loop, &niter)
5421 2860684 : && wi::fits_to_tree_p (niter, TREE_TYPE (valid_niter))
5422 2664804 : && (e = fold_binary (GT_EXPR, boolean_type_node, valid_niter,
5423 : wide_int_to_tree (TREE_TYPE (valid_niter),
5424 : niter))) != NULL
5425 5950566 : && integer_nonzerop (e))
5426 1164374 : return true;
5427 2195129 : if (at_stmt)
5428 3461649 : for (bound = loop->bounds; bound; bound = bound->next)
5429 : {
5430 2295781 : if (n_of_executions_at_most (at_stmt, bound, valid_niter))
5431 : return true;
5432 : }
5433 :
5434 : /* Try to prove loop is exited before {base, step} overflows with the
5435 : help of analyzed loop control IV. This is done only for IVs with
5436 : constant step because otherwise we don't have the information. */
5437 2185494 : if (TREE_CODE (step) == INTEGER_CST)
5438 : {
5439 3317648 : for (civ = loop->control_ivs; civ; civ = civ->next)
5440 : {
5441 1288294 : enum tree_code code;
5442 1288294 : tree civ_type = TREE_TYPE (civ->step);
5443 :
5444 : /* Have to consider type difference because operand_equal_p ignores
5445 : that for constants. */
5446 1288294 : if (TYPE_UNSIGNED (type) != TYPE_UNSIGNED (civ_type)
5447 1288294 : || element_precision (type) != element_precision (civ_type))
5448 761546 : continue;
5449 :
5450 : /* Only consider control IV with same step. */
5451 526748 : if (!operand_equal_p (step, civ->step, 0))
5452 207953 : continue;
5453 :
5454 : /* Done proving if this is a no-overflow control IV. */
5455 318795 : if (operand_equal_p (base, civ->base, 0))
5456 : return true;
5457 :
5458 : /* Control IV is recorded after expanding simple operations,
5459 : Here we expand base and compare it too. */
5460 228968 : tree expanded_base = expand_simple_operations (base);
5461 228968 : if (operand_equal_p (expanded_base, civ->base, 0))
5462 : return true;
5463 :
5464 : /* If this is a before stepping control IV, in other words, we have
5465 :
5466 : {civ_base, step} = {base + step, step}
5467 :
5468 : Because civ {base + step, step} doesn't overflow during loop
5469 : iterations, {base, step} will not overflow if we can prove the
5470 : operation "base + step" does not overflow. Specifically, we try
5471 : to prove below conditions are satisfied:
5472 :
5473 : base <= UPPER_BOUND (type) - step ;;step > 0
5474 : base >= LOWER_BOUND (type) - step ;;step < 0
5475 :
5476 : by proving the reverse conditions are false using loop's initial
5477 : condition. */
5478 199326 : if (POINTER_TYPE_P (TREE_TYPE (base)))
5479 : code = POINTER_PLUS_EXPR;
5480 : else
5481 : code = PLUS_EXPR;
5482 :
5483 199326 : tree stepped = fold_build2 (code, TREE_TYPE (base), base, step);
5484 199326 : tree expanded_stepped = fold_build2 (code, TREE_TYPE (base),
5485 : expanded_base, step);
5486 199326 : if (operand_equal_p (stepped, civ->base, 0)
5487 199326 : || operand_equal_p (expanded_stepped, civ->base, 0))
5488 : {
5489 46592 : tree extreme;
5490 :
5491 46592 : if (tree_int_cst_sign_bit (step))
5492 : {
5493 10752 : code = LT_EXPR;
5494 10752 : extreme = lower_bound_in_type (type, type);
5495 : }
5496 : else
5497 : {
5498 35840 : code = GT_EXPR;
5499 35840 : extreme = upper_bound_in_type (type, type);
5500 : }
5501 46592 : extreme = fold_build2 (MINUS_EXPR, type, extreme, step);
5502 46592 : e = fold_build2 (code, boolean_type_node, base, extreme);
5503 46592 : e = simplify_using_initial_conditions (loop, e);
5504 46592 : if (integer_zerop (e))
5505 : return true;
5506 : }
5507 : }
5508 : }
5509 :
5510 : return false;
5511 3359503 : }
5512 :
5513 : /* VAR is scev variable whose evolution part is constant STEP, this function
5514 : proves that VAR can't overflow by using value range info. If VAR's value
5515 : range is [MIN, MAX], it can be proven by:
5516 : MAX + step doesn't overflow ; if step > 0
5517 : or
5518 : MIN + step doesn't underflow ; if step < 0.
5519 :
5520 : We can only do this if var is computed in every loop iteration, i.e, var's
5521 : definition has to dominate loop latch. Consider below example:
5522 :
5523 : {
5524 : unsigned int i;
5525 :
5526 : <bb 3>:
5527 :
5528 : <bb 4>:
5529 : # RANGE [0, 4294967294] NONZERO 65535
5530 : # i_21 = PHI <0(3), i_18(9)>
5531 : if (i_21 != 0)
5532 : goto <bb 6>;
5533 : else
5534 : goto <bb 8>;
5535 :
5536 : <bb 6>:
5537 : # RANGE [0, 65533] NONZERO 65535
5538 : _6 = i_21 + 4294967295;
5539 : # RANGE [0, 65533] NONZERO 65535
5540 : _7 = (long unsigned int) _6;
5541 : # RANGE [0, 524264] NONZERO 524280
5542 : _8 = _7 * 8;
5543 : # PT = nonlocal escaped
5544 : _9 = a_14 + _8;
5545 : *_9 = 0;
5546 :
5547 : <bb 8>:
5548 : # RANGE [1, 65535] NONZERO 65535
5549 : i_18 = i_21 + 1;
5550 : if (i_18 >= 65535)
5551 : goto <bb 10>;
5552 : else
5553 : goto <bb 9>;
5554 :
5555 : <bb 9>:
5556 : goto <bb 4>;
5557 :
5558 : <bb 10>:
5559 : return;
5560 : }
5561 :
5562 : VAR _6 doesn't overflow only with pre-condition (i_21 != 0), here we
5563 : can't use _6 to prove no-overflow for _7. In fact, var _7 takes value
5564 : sequence (4294967295, 0, 1, ..., 65533) in loop life time, rather than
5565 : (4294967295, 4294967296, ...). */
5566 :
5567 : static bool
5568 329475 : scev_var_range_cant_overflow (tree var, tree step, class loop *loop)
5569 : {
5570 329475 : tree type;
5571 329475 : wide_int minv, maxv, diff, step_wi;
5572 :
5573 329475 : if (TREE_CODE (step) != INTEGER_CST || !INTEGRAL_TYPE_P (TREE_TYPE (var)))
5574 : return false;
5575 :
5576 : /* Check if VAR evaluates in every loop iteration. It's not the case
5577 : if VAR is default definition or does not dominate loop's latch. */
5578 329475 : basic_block def_bb = gimple_bb (SSA_NAME_DEF_STMT (var));
5579 329475 : if (!def_bb || !dominated_by_p (CDI_DOMINATORS, loop->latch, def_bb))
5580 : return false;
5581 :
5582 326854 : int_range_max r (TREE_TYPE (var));
5583 653708 : get_range_query (cfun)->range_of_expr (r, var);
5584 326854 : if (r.varying_p () || r.undefined_p ())
5585 : return false;
5586 :
5587 : /* VAR is a scev whose evolution part is STEP and value range info
5588 : is [MIN, MAX], we can prove its no-overflowness by conditions:
5589 :
5590 : type_MAX - MAX >= step ; if step > 0
5591 : MIN - type_MIN >= |step| ; if step < 0.
5592 :
5593 : Or VAR must take value outside of value range, which is not true. */
5594 127091 : step_wi = wi::to_wide (step);
5595 127091 : type = TREE_TYPE (var);
5596 127091 : if (tree_int_cst_sign_bit (step))
5597 : {
5598 6623 : diff = r.lower_bound () - wi::to_wide (lower_bound_in_type (type, type));
5599 6623 : step_wi = - step_wi;
5600 : }
5601 : else
5602 120468 : diff = wi::to_wide (upper_bound_in_type (type, type)) - r.upper_bound ();
5603 :
5604 127091 : return (wi::geu_p (diff, step_wi));
5605 656329 : }
5606 :
5607 : /* Return false only when the induction variable BASE + STEP * I is
5608 : known to not overflow: i.e. when the number of iterations is small
5609 : enough with respect to the step and initial condition in order to
5610 : keep the evolution confined in TYPEs bounds. Return true when the
5611 : iv is known to overflow or when the property is not computable.
5612 :
5613 : USE_OVERFLOW_SEMANTICS is true if this function should assume that
5614 : the rules for overflow of the given language apply (e.g., that signed
5615 : arithmetics in C does not overflow).
5616 :
5617 : If VAR is a ssa variable, this function also returns false if VAR can
5618 : be proven not overflow with value range info. */
5619 :
5620 : bool
5621 7839239 : scev_probably_wraps_p (tree var, tree base, tree step,
5622 : gimple *at_stmt, class loop *loop,
5623 : bool use_overflow_semantics)
5624 : {
5625 : /* FIXME: We really need something like
5626 : http://gcc.gnu.org/ml/gcc-patches/2005-06/msg02025.html.
5627 :
5628 : We used to test for the following situation that frequently appears
5629 : during address arithmetics:
5630 :
5631 : D.1621_13 = (long unsigned intD.4) D.1620_12;
5632 : D.1622_14 = D.1621_13 * 8;
5633 : D.1623_15 = (doubleD.29 *) D.1622_14;
5634 :
5635 : And derived that the sequence corresponding to D_14
5636 : can be proved to not wrap because it is used for computing a
5637 : memory access; however, this is not really the case -- for example,
5638 : if D_12 = (unsigned char) [254,+,1], then D_14 has values
5639 : 2032, 2040, 0, 8, ..., but the code is still legal. */
5640 :
5641 7839239 : if (chrec_contains_undetermined (base)
5642 7839239 : || chrec_contains_undetermined (step))
5643 : return true;
5644 :
5645 7839239 : if (integer_zerop (step))
5646 : return false;
5647 :
5648 : /* If we can use the fact that signed and pointer arithmetics does not
5649 : wrap, we are done. */
5650 7839239 : if (use_overflow_semantics && nowrap_type_p (TREE_TYPE (base)))
5651 : return false;
5652 :
5653 : /* To be able to use estimates on number of iterations of the loop,
5654 : we must have an upper bound on the absolute value of the step. */
5655 4133202 : if (TREE_CODE (step) != INTEGER_CST)
5656 : return true;
5657 :
5658 : /* Check if var can be proven not overflow with value range info. */
5659 343734 : if (var && TREE_CODE (var) == SSA_NAME
5660 3805162 : && scev_var_range_cant_overflow (var, step, loop))
5661 : return false;
5662 :
5663 3359503 : if (loop_exits_before_overflow (base, step, at_stmt, loop))
5664 : return false;
5665 :
5666 : /* Check the nonwrapping flag, which may be set by niter analysis (e.g., the
5667 : above loop exits before overflow). */
5668 2029354 : if (var && nonwrapping_chrec_p (analyze_scalar_evolution (loop, var)))
5669 3382 : return false;
5670 :
5671 : /* At this point we still don't have a proof that the iv does not
5672 : overflow: give up. */
5673 : return true;
5674 : }
5675 :
5676 : /* Frees the information on upper bounds on numbers of iterations of LOOP. */
5677 :
5678 : void
5679 58436237 : free_numbers_of_iterations_estimates (class loop *loop)
5680 : {
5681 58436237 : struct control_iv *civ;
5682 58436237 : class nb_iter_bound *bound;
5683 :
5684 58436237 : loop->nb_iterations = NULL;
5685 58436237 : loop->estimate_state = EST_NOT_COMPUTED;
5686 68650952 : for (bound = loop->bounds; bound;)
5687 : {
5688 10214715 : class nb_iter_bound *next = bound->next;
5689 10214715 : ggc_free (bound);
5690 10214715 : bound = next;
5691 : }
5692 58436237 : loop->bounds = NULL;
5693 :
5694 62938316 : for (civ = loop->control_ivs; civ;)
5695 : {
5696 4502079 : struct control_iv *next = civ->next;
5697 4502079 : ggc_free (civ);
5698 4502079 : civ = next;
5699 : }
5700 58436237 : loop->control_ivs = NULL;
5701 58436237 : }
5702 :
5703 : /* Frees the information on upper bounds on numbers of iterations of loops. */
5704 :
5705 : void
5706 74619722 : free_numbers_of_iterations_estimates (function *fn)
5707 : {
5708 281616652 : for (auto loop : loops_list (fn, 0))
5709 57757486 : free_numbers_of_iterations_estimates (loop);
5710 74619722 : }
5711 :
5712 : /* Substitute value VAL for ssa name NAME inside expressions held
5713 : at LOOP. */
5714 :
5715 : void
5716 128168961 : substitute_in_loop_info (class loop *loop, tree name, tree val)
5717 : {
5718 128168961 : loop->nb_iterations = simplify_replace_tree (loop->nb_iterations, name, val);
5719 128168961 : }
|
