re PR target/35659 (Miscompiled code with -O2 (but not with -O2 -funroll-loops) on...
[official-gcc.git] / gcc / tree-ssa-loop-niter.c
blob13b10c9f1c41477d2ab4693131ea8f1159b03b30
1 /* Functions to determine/estimate number of iterations of a loop.
2 Copyright (C) 2004, 2005, 2006, 2007, 2008 Free Software Foundation,
3 Inc.
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it
8 under the terms of the GNU General Public License as published by the
9 Free Software Foundation; either version 3, or (at your option) any
10 later version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT
13 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
15 for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING3. If not see
19 <http://www.gnu.org/licenses/>. */
21 #include "config.h"
22 #include "system.h"
23 #include "coretypes.h"
24 #include "tm.h"
25 #include "tree.h"
26 #include "rtl.h"
27 #include "tm_p.h"
28 #include "hard-reg-set.h"
29 #include "basic-block.h"
30 #include "output.h"
31 #include "diagnostic.h"
32 #include "intl.h"
33 #include "tree-flow.h"
34 #include "tree-dump.h"
35 #include "cfgloop.h"
36 #include "tree-pass.h"
37 #include "ggc.h"
38 #include "tree-chrec.h"
39 #include "tree-scalar-evolution.h"
40 #include "tree-data-ref.h"
41 #include "params.h"
42 #include "flags.h"
43 #include "toplev.h"
44 #include "tree-inline.h"
45 #include "gmp.h"
47 #define SWAP(X, Y) do { affine_iv *tmp = (X); (X) = (Y); (Y) = tmp; } while (0)
49 /* The maximum number of dominator BBs we search for conditions
50 of loop header copies we use for simplifying a conditional
51 expression. */
52 #define MAX_DOMINATORS_TO_WALK 8
56 Analysis of number of iterations of an affine exit test.
60 /* Bounds on some value, BELOW <= X <= UP. */
62 typedef struct
64 mpz_t below, up;
65 } bounds;
68 /* Splits expression EXPR to a variable part VAR and constant OFFSET. */
70 static void
71 split_to_var_and_offset (tree expr, tree *var, mpz_t offset)
73 tree type = TREE_TYPE (expr);
74 tree op0, op1;
75 double_int off;
76 bool negate = false;
78 *var = expr;
79 mpz_set_ui (offset, 0);
81 switch (TREE_CODE (expr))
83 case MINUS_EXPR:
84 negate = true;
85 /* Fallthru. */
87 case PLUS_EXPR:
88 case POINTER_PLUS_EXPR:
89 op0 = TREE_OPERAND (expr, 0);
90 op1 = TREE_OPERAND (expr, 1);
92 if (TREE_CODE (op1) != INTEGER_CST)
93 break;
95 *var = op0;
96 /* Always sign extend the offset. */
97 off = double_int_sext (tree_to_double_int (op1),
98 TYPE_PRECISION (type));
99 mpz_set_double_int (offset, off, false);
100 break;
102 case INTEGER_CST:
103 *var = build_int_cst_type (type, 0);
104 off = tree_to_double_int (expr);
105 mpz_set_double_int (offset, off, TYPE_UNSIGNED (type));
106 break;
108 default:
109 break;
113 /* Stores estimate on the minimum/maximum value of the expression VAR + OFF
114 in TYPE to MIN and MAX. */
116 static void
117 determine_value_range (tree type, tree var, mpz_t off,
118 mpz_t min, mpz_t max)
120 /* If the expression is a constant, we know its value exactly. */
121 if (integer_zerop (var))
123 mpz_set (min, off);
124 mpz_set (max, off);
125 return;
128 /* If the computation may wrap, we know nothing about the value, except for
129 the range of the type. */
130 get_type_static_bounds (type, min, max);
131 if (!nowrap_type_p (type))
132 return;
134 /* Since the addition of OFF does not wrap, if OFF is positive, then we may
135 add it to MIN, otherwise to MAX. */
136 if (mpz_sgn (off) < 0)
137 mpz_add (max, max, off);
138 else
139 mpz_add (min, min, off);
142 /* Stores the bounds on the difference of the values of the expressions
143 (var + X) and (var + Y), computed in TYPE, to BNDS. */
145 static void
146 bound_difference_of_offsetted_base (tree type, mpz_t x, mpz_t y,
147 bounds *bnds)
149 int rel = mpz_cmp (x, y);
150 bool may_wrap = !nowrap_type_p (type);
151 mpz_t m;
153 /* If X == Y, then the expressions are always equal.
154 If X > Y, there are the following possibilities:
155 a) neither of var + X and var + Y overflow or underflow, or both of
156 them do. Then their difference is X - Y.
157 b) var + X overflows, and var + Y does not. Then the values of the
158 expressions are var + X - M and var + Y, where M is the range of
159 the type, and their difference is X - Y - M.
160 c) var + Y underflows and var + X does not. Their difference again
161 is M - X + Y.
162 Therefore, if the arithmetics in type does not overflow, then the
163 bounds are (X - Y, X - Y), otherwise they are (X - Y - M, X - Y)
164 Similarly, if X < Y, the bounds are either (X - Y, X - Y) or
165 (X - Y, X - Y + M). */
167 if (rel == 0)
169 mpz_set_ui (bnds->below, 0);
170 mpz_set_ui (bnds->up, 0);
171 return;
174 mpz_init (m);
175 mpz_set_double_int (m, double_int_mask (TYPE_PRECISION (type)), true);
176 mpz_add_ui (m, m, 1);
177 mpz_sub (bnds->up, x, y);
178 mpz_set (bnds->below, bnds->up);
180 if (may_wrap)
182 if (rel > 0)
183 mpz_sub (bnds->below, bnds->below, m);
184 else
185 mpz_add (bnds->up, bnds->up, m);
188 mpz_clear (m);
191 /* From condition C0 CMP C1 derives information regarding the
192 difference of values of VARX + OFFX and VARY + OFFY, computed in TYPE,
193 and stores it to BNDS. */
195 static void
196 refine_bounds_using_guard (tree type, tree varx, mpz_t offx,
197 tree vary, mpz_t offy,
198 tree c0, enum tree_code cmp, tree c1,
199 bounds *bnds)
201 tree varc0, varc1, tmp, ctype;
202 mpz_t offc0, offc1, loffx, loffy, bnd;
203 bool lbound = false;
204 bool no_wrap = nowrap_type_p (type);
205 bool x_ok, y_ok;
207 switch (cmp)
209 case LT_EXPR:
210 case LE_EXPR:
211 case GT_EXPR:
212 case GE_EXPR:
213 STRIP_SIGN_NOPS (c0);
214 STRIP_SIGN_NOPS (c1);
215 ctype = TREE_TYPE (c0);
216 if (!useless_type_conversion_p (ctype, type))
217 return;
219 break;
221 case EQ_EXPR:
222 /* We could derive quite precise information from EQ_EXPR, however, such
223 a guard is unlikely to appear, so we do not bother with handling
224 it. */
225 return;
227 case NE_EXPR:
228 /* NE_EXPR comparisons do not contain much of useful information, except for
229 special case of comparing with the bounds of the type. */
230 if (TREE_CODE (c1) != INTEGER_CST
231 || !INTEGRAL_TYPE_P (type))
232 return;
234 /* Ensure that the condition speaks about an expression in the same type
235 as X and Y. */
236 ctype = TREE_TYPE (c0);
237 if (TYPE_PRECISION (ctype) != TYPE_PRECISION (type))
238 return;
239 c0 = fold_convert (type, c0);
240 c1 = fold_convert (type, c1);
242 if (TYPE_MIN_VALUE (type)
243 && operand_equal_p (c1, TYPE_MIN_VALUE (type), 0))
245 cmp = GT_EXPR;
246 break;
248 if (TYPE_MAX_VALUE (type)
249 && operand_equal_p (c1, TYPE_MAX_VALUE (type), 0))
251 cmp = LT_EXPR;
252 break;
255 return;
256 default:
257 return;
260 mpz_init (offc0);
261 mpz_init (offc1);
262 split_to_var_and_offset (expand_simple_operations (c0), &varc0, offc0);
263 split_to_var_and_offset (expand_simple_operations (c1), &varc1, offc1);
265 /* We are only interested in comparisons of expressions based on VARX and
266 VARY. TODO -- we might also be able to derive some bounds from
267 expressions containing just one of the variables. */
269 if (operand_equal_p (varx, varc1, 0))
271 tmp = varc0; varc0 = varc1; varc1 = tmp;
272 mpz_swap (offc0, offc1);
273 cmp = swap_tree_comparison (cmp);
276 if (!operand_equal_p (varx, varc0, 0)
277 || !operand_equal_p (vary, varc1, 0))
278 goto end;
280 mpz_init_set (loffx, offx);
281 mpz_init_set (loffy, offy);
283 if (cmp == GT_EXPR || cmp == GE_EXPR)
285 tmp = varx; varx = vary; vary = tmp;
286 mpz_swap (offc0, offc1);
287 mpz_swap (loffx, loffy);
288 cmp = swap_tree_comparison (cmp);
289 lbound = true;
292 /* If there is no overflow, the condition implies that
294 (VARX + OFFX) cmp (VARY + OFFY) + (OFFX - OFFY + OFFC1 - OFFC0).
296 The overflows and underflows may complicate things a bit; each
297 overflow decreases the appropriate offset by M, and underflow
298 increases it by M. The above inequality would not necessarily be
299 true if
301 -- VARX + OFFX underflows and VARX + OFFC0 does not, or
302 VARX + OFFC0 overflows, but VARX + OFFX does not.
303 This may only happen if OFFX < OFFC0.
304 -- VARY + OFFY overflows and VARY + OFFC1 does not, or
305 VARY + OFFC1 underflows and VARY + OFFY does not.
306 This may only happen if OFFY > OFFC1. */
308 if (no_wrap)
310 x_ok = true;
311 y_ok = true;
313 else
315 x_ok = (integer_zerop (varx)
316 || mpz_cmp (loffx, offc0) >= 0);
317 y_ok = (integer_zerop (vary)
318 || mpz_cmp (loffy, offc1) <= 0);
321 if (x_ok && y_ok)
323 mpz_init (bnd);
324 mpz_sub (bnd, loffx, loffy);
325 mpz_add (bnd, bnd, offc1);
326 mpz_sub (bnd, bnd, offc0);
328 if (cmp == LT_EXPR)
329 mpz_sub_ui (bnd, bnd, 1);
331 if (lbound)
333 mpz_neg (bnd, bnd);
334 if (mpz_cmp (bnds->below, bnd) < 0)
335 mpz_set (bnds->below, bnd);
337 else
339 if (mpz_cmp (bnd, bnds->up) < 0)
340 mpz_set (bnds->up, bnd);
342 mpz_clear (bnd);
345 mpz_clear (loffx);
346 mpz_clear (loffy);
347 end:
348 mpz_clear (offc0);
349 mpz_clear (offc1);
352 /* Stores the bounds on the value of the expression X - Y in LOOP to BNDS.
353 The subtraction is considered to be performed in arbitrary precision,
354 without overflows.
356 We do not attempt to be too clever regarding the value ranges of X and
357 Y; most of the time, they are just integers or ssa names offsetted by
358 integer. However, we try to use the information contained in the
359 comparisons before the loop (usually created by loop header copying). */
361 static void
362 bound_difference (struct loop *loop, tree x, tree y, bounds *bnds)
364 tree type = TREE_TYPE (x);
365 tree varx, vary;
366 mpz_t offx, offy;
367 mpz_t minx, maxx, miny, maxy;
368 int cnt = 0;
369 edge e;
370 basic_block bb;
371 tree c0, c1;
372 gimple cond;
373 enum tree_code cmp;
375 /* Get rid of unnecessary casts, but preserve the value of
376 the expressions. */
377 STRIP_SIGN_NOPS (x);
378 STRIP_SIGN_NOPS (y);
380 mpz_init (bnds->below);
381 mpz_init (bnds->up);
382 mpz_init (offx);
383 mpz_init (offy);
384 split_to_var_and_offset (x, &varx, offx);
385 split_to_var_and_offset (y, &vary, offy);
387 if (!integer_zerop (varx)
388 && operand_equal_p (varx, vary, 0))
390 /* Special case VARX == VARY -- we just need to compare the
391 offsets. The matters are a bit more complicated in the
392 case addition of offsets may wrap. */
393 bound_difference_of_offsetted_base (type, offx, offy, bnds);
395 else
397 /* Otherwise, use the value ranges to determine the initial
398 estimates on below and up. */
399 mpz_init (minx);
400 mpz_init (maxx);
401 mpz_init (miny);
402 mpz_init (maxy);
403 determine_value_range (type, varx, offx, minx, maxx);
404 determine_value_range (type, vary, offy, miny, maxy);
406 mpz_sub (bnds->below, minx, maxy);
407 mpz_sub (bnds->up, maxx, miny);
408 mpz_clear (minx);
409 mpz_clear (maxx);
410 mpz_clear (miny);
411 mpz_clear (maxy);
414 /* If both X and Y are constants, we cannot get any more precise. */
415 if (integer_zerop (varx) && integer_zerop (vary))
416 goto end;
418 /* Now walk the dominators of the loop header and use the entry
419 guards to refine the estimates. */
420 for (bb = loop->header;
421 bb != ENTRY_BLOCK_PTR && cnt < MAX_DOMINATORS_TO_WALK;
422 bb = get_immediate_dominator (CDI_DOMINATORS, bb))
424 if (!single_pred_p (bb))
425 continue;
426 e = single_pred_edge (bb);
428 if (!(e->flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
429 continue;
431 cond = last_stmt (e->src);
432 c0 = gimple_cond_lhs (cond);
433 cmp = gimple_cond_code (cond);
434 c1 = gimple_cond_rhs (cond);
436 if (e->flags & EDGE_FALSE_VALUE)
437 cmp = invert_tree_comparison (cmp, false);
439 refine_bounds_using_guard (type, varx, offx, vary, offy,
440 c0, cmp, c1, bnds);
441 ++cnt;
444 end:
445 mpz_clear (offx);
446 mpz_clear (offy);
449 /* Update the bounds in BNDS that restrict the value of X to the bounds
450 that restrict the value of X + DELTA. X can be obtained as a
451 difference of two values in TYPE. */
453 static void
454 bounds_add (bounds *bnds, double_int delta, tree type)
456 mpz_t mdelta, max;
458 mpz_init (mdelta);
459 mpz_set_double_int (mdelta, delta, false);
461 mpz_init (max);
462 mpz_set_double_int (max, double_int_mask (TYPE_PRECISION (type)), true);
464 mpz_add (bnds->up, bnds->up, mdelta);
465 mpz_add (bnds->below, bnds->below, mdelta);
467 if (mpz_cmp (bnds->up, max) > 0)
468 mpz_set (bnds->up, max);
470 mpz_neg (max, max);
471 if (mpz_cmp (bnds->below, max) < 0)
472 mpz_set (bnds->below, max);
474 mpz_clear (mdelta);
475 mpz_clear (max);
478 /* Update the bounds in BNDS that restrict the value of X to the bounds
479 that restrict the value of -X. */
481 static void
482 bounds_negate (bounds *bnds)
484 mpz_t tmp;
486 mpz_init_set (tmp, bnds->up);
487 mpz_neg (bnds->up, bnds->below);
488 mpz_neg (bnds->below, tmp);
489 mpz_clear (tmp);
492 /* Returns inverse of X modulo 2^s, where MASK = 2^s-1. */
494 static tree
495 inverse (tree x, tree mask)
497 tree type = TREE_TYPE (x);
498 tree rslt;
499 unsigned ctr = tree_floor_log2 (mask);
501 if (TYPE_PRECISION (type) <= HOST_BITS_PER_WIDE_INT)
503 unsigned HOST_WIDE_INT ix;
504 unsigned HOST_WIDE_INT imask;
505 unsigned HOST_WIDE_INT irslt = 1;
507 gcc_assert (cst_and_fits_in_hwi (x));
508 gcc_assert (cst_and_fits_in_hwi (mask));
510 ix = int_cst_value (x);
511 imask = int_cst_value (mask);
513 for (; ctr; ctr--)
515 irslt *= ix;
516 ix *= ix;
518 irslt &= imask;
520 rslt = build_int_cst_type (type, irslt);
522 else
524 rslt = build_int_cst (type, 1);
525 for (; ctr; ctr--)
527 rslt = int_const_binop (MULT_EXPR, rslt, x, 0);
528 x = int_const_binop (MULT_EXPR, x, x, 0);
530 rslt = int_const_binop (BIT_AND_EXPR, rslt, mask, 0);
533 return rslt;
536 /* Derives the upper bound BND on the number of executions of loop with exit
537 condition S * i <> C, assuming that the loop is not infinite. If
538 NO_OVERFLOW is true, then the control variable of the loop does not
539 overflow. If NO_OVERFLOW is true or BNDS.below >= 0, then BNDS.up
540 contains the upper bound on the value of C. */
542 static void
543 number_of_iterations_ne_max (mpz_t bnd, bool no_overflow, tree c, tree s,
544 bounds *bnds)
546 double_int max;
547 mpz_t d;
549 /* If the control variable does not overflow, the number of iterations is
550 at most c / s. Otherwise it is at most the period of the control
551 variable. */
552 if (!no_overflow && !multiple_of_p (TREE_TYPE (c), c, s))
554 max = double_int_mask (TYPE_PRECISION (TREE_TYPE (c))
555 - tree_low_cst (num_ending_zeros (s), 1));
556 mpz_set_double_int (bnd, max, true);
557 return;
560 /* Determine the upper bound on C. */
561 if (no_overflow || mpz_sgn (bnds->below) >= 0)
562 mpz_set (bnd, bnds->up);
563 else if (TREE_CODE (c) == INTEGER_CST)
564 mpz_set_double_int (bnd, tree_to_double_int (c), true);
565 else
566 mpz_set_double_int (bnd, double_int_mask (TYPE_PRECISION (TREE_TYPE (c))),
567 true);
569 mpz_init (d);
570 mpz_set_double_int (d, tree_to_double_int (s), true);
571 mpz_fdiv_q (bnd, bnd, d);
572 mpz_clear (d);
575 /* Determines number of iterations of loop whose ending condition
576 is IV <> FINAL. TYPE is the type of the iv. The number of
577 iterations is stored to NITER. NEVER_INFINITE is true if
578 we know that the exit must be taken eventually, i.e., that the IV
579 ever reaches the value FINAL (we derived this earlier, and possibly set
580 NITER->assumptions to make sure this is the case). BNDS contains the
581 bounds on the difference FINAL - IV->base. */
583 static bool
584 number_of_iterations_ne (tree type, affine_iv *iv, tree final,
585 struct tree_niter_desc *niter, bool never_infinite,
586 bounds *bnds)
588 tree niter_type = unsigned_type_for (type);
589 tree s, c, d, bits, assumption, tmp, bound;
590 mpz_t max;
592 niter->control = *iv;
593 niter->bound = final;
594 niter->cmp = NE_EXPR;
596 /* Rearrange the terms so that we get inequality S * i <> C, with S
597 positive. Also cast everything to the unsigned type. If IV does
598 not overflow, BNDS bounds the value of C. Also, this is the
599 case if the computation |FINAL - IV->base| does not overflow, i.e.,
600 if BNDS->below in the result is nonnegative. */
601 if (tree_int_cst_sign_bit (iv->step))
603 s = fold_convert (niter_type,
604 fold_build1 (NEGATE_EXPR, type, iv->step));
605 c = fold_build2 (MINUS_EXPR, niter_type,
606 fold_convert (niter_type, iv->base),
607 fold_convert (niter_type, final));
608 bounds_negate (bnds);
610 else
612 s = fold_convert (niter_type, iv->step);
613 c = fold_build2 (MINUS_EXPR, niter_type,
614 fold_convert (niter_type, final),
615 fold_convert (niter_type, iv->base));
618 mpz_init (max);
619 number_of_iterations_ne_max (max, iv->no_overflow, c, s, bnds);
620 niter->max = mpz_get_double_int (niter_type, max, false);
621 mpz_clear (max);
623 /* First the trivial cases -- when the step is 1. */
624 if (integer_onep (s))
626 niter->niter = c;
627 return true;
630 /* Let nsd (step, size of mode) = d. If d does not divide c, the loop
631 is infinite. Otherwise, the number of iterations is
632 (inverse(s/d) * (c/d)) mod (size of mode/d). */
633 bits = num_ending_zeros (s);
634 bound = build_low_bits_mask (niter_type,
635 (TYPE_PRECISION (niter_type)
636 - tree_low_cst (bits, 1)));
638 d = fold_binary_to_constant (LSHIFT_EXPR, niter_type,
639 build_int_cst (niter_type, 1), bits);
640 s = fold_binary_to_constant (RSHIFT_EXPR, niter_type, s, bits);
642 if (!never_infinite)
644 /* If we cannot assume that the loop is not infinite, record the
645 assumptions for divisibility of c. */
646 assumption = fold_build2 (FLOOR_MOD_EXPR, niter_type, c, d);
647 assumption = fold_build2 (EQ_EXPR, boolean_type_node,
648 assumption, build_int_cst (niter_type, 0));
649 if (!integer_nonzerop (assumption))
650 niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
651 niter->assumptions, assumption);
654 c = fold_build2 (EXACT_DIV_EXPR, niter_type, c, d);
655 tmp = fold_build2 (MULT_EXPR, niter_type, c, inverse (s, bound));
656 niter->niter = fold_build2 (BIT_AND_EXPR, niter_type, tmp, bound);
657 return true;
660 /* Checks whether we can determine the final value of the control variable
661 of the loop with ending condition IV0 < IV1 (computed in TYPE).
662 DELTA is the difference IV1->base - IV0->base, STEP is the absolute value
663 of the step. The assumptions necessary to ensure that the computation
664 of the final value does not overflow are recorded in NITER. If we
665 find the final value, we adjust DELTA and return TRUE. Otherwise
666 we return false. BNDS bounds the value of IV1->base - IV0->base,
667 and will be updated by the same amount as DELTA. */
669 static bool
670 number_of_iterations_lt_to_ne (tree type, affine_iv *iv0, affine_iv *iv1,
671 struct tree_niter_desc *niter,
672 tree *delta, tree step,
673 bounds *bnds)
675 tree niter_type = TREE_TYPE (step);
676 tree mod = fold_build2 (FLOOR_MOD_EXPR, niter_type, *delta, step);
677 tree tmod;
678 mpz_t mmod;
679 tree assumption = boolean_true_node, bound, noloop;
680 bool ret = false;
681 tree type1 = type;
682 if (POINTER_TYPE_P (type))
683 type1 = sizetype;
685 if (TREE_CODE (mod) != INTEGER_CST)
686 return false;
687 if (integer_nonzerop (mod))
688 mod = fold_build2 (MINUS_EXPR, niter_type, step, mod);
689 tmod = fold_convert (type1, mod);
691 mpz_init (mmod);
692 mpz_set_double_int (mmod, tree_to_double_int (mod), true);
693 mpz_neg (mmod, mmod);
695 if (integer_nonzerop (iv0->step))
697 /* The final value of the iv is iv1->base + MOD, assuming that this
698 computation does not overflow, and that
699 iv0->base <= iv1->base + MOD. */
700 if (!iv0->no_overflow && !integer_zerop (mod))
702 bound = fold_build2 (MINUS_EXPR, type,
703 TYPE_MAX_VALUE (type1), tmod);
704 assumption = fold_build2 (LE_EXPR, boolean_type_node,
705 iv1->base, bound);
706 if (integer_zerop (assumption))
707 goto end;
709 if (mpz_cmp (mmod, bnds->below) < 0)
710 noloop = boolean_false_node;
711 else
712 noloop = fold_build2 (GT_EXPR, boolean_type_node,
713 iv0->base,
714 fold_build2 (PLUS_EXPR, type1,
715 iv1->base, tmod));
717 else
719 /* The final value of the iv is iv0->base - MOD, assuming that this
720 computation does not overflow, and that
721 iv0->base - MOD <= iv1->base. */
722 if (!iv1->no_overflow && !integer_zerop (mod))
724 bound = fold_build2 (PLUS_EXPR, type1,
725 TYPE_MIN_VALUE (type1), tmod);
726 assumption = fold_build2 (GE_EXPR, boolean_type_node,
727 iv0->base, bound);
728 if (integer_zerop (assumption))
729 goto end;
731 if (mpz_cmp (mmod, bnds->below) < 0)
732 noloop = boolean_false_node;
733 else
734 noloop = fold_build2 (GT_EXPR, boolean_type_node,
735 fold_build2 (MINUS_EXPR, type1,
736 iv0->base, tmod),
737 iv1->base);
740 if (!integer_nonzerop (assumption))
741 niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
742 niter->assumptions,
743 assumption);
744 if (!integer_zerop (noloop))
745 niter->may_be_zero = fold_build2 (TRUTH_OR_EXPR, boolean_type_node,
746 niter->may_be_zero,
747 noloop);
748 bounds_add (bnds, tree_to_double_int (mod), type);
749 *delta = fold_build2 (PLUS_EXPR, niter_type, *delta, mod);
751 ret = true;
752 end:
753 mpz_clear (mmod);
754 return ret;
757 /* Add assertions to NITER that ensure that the control variable of the loop
758 with ending condition IV0 < IV1 does not overflow. Types of IV0 and IV1
759 are TYPE. Returns false if we can prove that there is an overflow, true
760 otherwise. STEP is the absolute value of the step. */
762 static bool
763 assert_no_overflow_lt (tree type, affine_iv *iv0, affine_iv *iv1,
764 struct tree_niter_desc *niter, tree step)
766 tree bound, d, assumption, diff;
767 tree niter_type = TREE_TYPE (step);
769 if (integer_nonzerop (iv0->step))
771 /* for (i = iv0->base; i < iv1->base; i += iv0->step) */
772 if (iv0->no_overflow)
773 return true;
775 /* If iv0->base is a constant, we can determine the last value before
776 overflow precisely; otherwise we conservatively assume
777 MAX - STEP + 1. */
779 if (TREE_CODE (iv0->base) == INTEGER_CST)
781 d = fold_build2 (MINUS_EXPR, niter_type,
782 fold_convert (niter_type, TYPE_MAX_VALUE (type)),
783 fold_convert (niter_type, iv0->base));
784 diff = fold_build2 (FLOOR_MOD_EXPR, niter_type, d, step);
786 else
787 diff = fold_build2 (MINUS_EXPR, niter_type, step,
788 build_int_cst (niter_type, 1));
789 bound = fold_build2 (MINUS_EXPR, type,
790 TYPE_MAX_VALUE (type), fold_convert (type, diff));
791 assumption = fold_build2 (LE_EXPR, boolean_type_node,
792 iv1->base, bound);
794 else
796 /* for (i = iv1->base; i > iv0->base; i += iv1->step) */
797 if (iv1->no_overflow)
798 return true;
800 if (TREE_CODE (iv1->base) == INTEGER_CST)
802 d = fold_build2 (MINUS_EXPR, niter_type,
803 fold_convert (niter_type, iv1->base),
804 fold_convert (niter_type, TYPE_MIN_VALUE (type)));
805 diff = fold_build2 (FLOOR_MOD_EXPR, niter_type, d, step);
807 else
808 diff = fold_build2 (MINUS_EXPR, niter_type, step,
809 build_int_cst (niter_type, 1));
810 bound = fold_build2 (PLUS_EXPR, type,
811 TYPE_MIN_VALUE (type), fold_convert (type, diff));
812 assumption = fold_build2 (GE_EXPR, boolean_type_node,
813 iv0->base, bound);
816 if (integer_zerop (assumption))
817 return false;
818 if (!integer_nonzerop (assumption))
819 niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
820 niter->assumptions, assumption);
822 iv0->no_overflow = true;
823 iv1->no_overflow = true;
824 return true;
827 /* Add an assumption to NITER that a loop whose ending condition
828 is IV0 < IV1 rolls. TYPE is the type of the control iv. BNDS
829 bounds the value of IV1->base - IV0->base. */
831 static void
832 assert_loop_rolls_lt (tree type, affine_iv *iv0, affine_iv *iv1,
833 struct tree_niter_desc *niter, bounds *bnds)
835 tree assumption = boolean_true_node, bound, diff;
836 tree mbz, mbzl, mbzr, type1;
837 bool rolls_p, no_overflow_p;
838 double_int dstep;
839 mpz_t mstep, max;
841 /* We are going to compute the number of iterations as
842 (iv1->base - iv0->base + step - 1) / step, computed in the unsigned
843 variant of TYPE. This formula only works if
845 -step + 1 <= (iv1->base - iv0->base) <= MAX - step + 1
847 (where MAX is the maximum value of the unsigned variant of TYPE, and
848 the computations in this formula are performed in full precision
849 (without overflows).
851 Usually, for loops with exit condition iv0->base + step * i < iv1->base,
852 we have a condition of form iv0->base - step < iv1->base before the loop,
853 and for loops iv0->base < iv1->base - step * i the condition
854 iv0->base < iv1->base + step, due to loop header copying, which enable us
855 to prove the lower bound.
857 The upper bound is more complicated. Unless the expressions for initial
858 and final value themselves contain enough information, we usually cannot
859 derive it from the context. */
861 /* First check whether the answer does not follow from the bounds we gathered
862 before. */
863 if (integer_nonzerop (iv0->step))
864 dstep = tree_to_double_int (iv0->step);
865 else
867 dstep = double_int_sext (tree_to_double_int (iv1->step),
868 TYPE_PRECISION (type));
869 dstep = double_int_neg (dstep);
872 mpz_init (mstep);
873 mpz_set_double_int (mstep, dstep, true);
874 mpz_neg (mstep, mstep);
875 mpz_add_ui (mstep, mstep, 1);
877 rolls_p = mpz_cmp (mstep, bnds->below) <= 0;
879 mpz_init (max);
880 mpz_set_double_int (max, double_int_mask (TYPE_PRECISION (type)), true);
881 mpz_add (max, max, mstep);
882 no_overflow_p = (mpz_cmp (bnds->up, max) <= 0
883 /* For pointers, only values lying inside a single object
884 can be compared or manipulated by pointer arithmetics.
885 Gcc in general does not allow or handle objects larger
886 than half of the address space, hence the upper bound
887 is satisfied for pointers. */
888 || POINTER_TYPE_P (type));
889 mpz_clear (mstep);
890 mpz_clear (max);
892 if (rolls_p && no_overflow_p)
893 return;
895 type1 = type;
896 if (POINTER_TYPE_P (type))
897 type1 = sizetype;
899 /* Now the hard part; we must formulate the assumption(s) as expressions, and
900 we must be careful not to introduce overflow. */
902 if (integer_nonzerop (iv0->step))
904 diff = fold_build2 (MINUS_EXPR, type1,
905 iv0->step, build_int_cst (type1, 1));
907 /* We need to know that iv0->base >= MIN + iv0->step - 1. Since
908 0 address never belongs to any object, we can assume this for
909 pointers. */
910 if (!POINTER_TYPE_P (type))
912 bound = fold_build2 (PLUS_EXPR, type1,
913 TYPE_MIN_VALUE (type), diff);
914 assumption = fold_build2 (GE_EXPR, boolean_type_node,
915 iv0->base, bound);
918 /* And then we can compute iv0->base - diff, and compare it with
919 iv1->base. */
920 mbzl = fold_build2 (MINUS_EXPR, type1,
921 fold_convert (type1, iv0->base), diff);
922 mbzr = fold_convert (type1, iv1->base);
924 else
926 diff = fold_build2 (PLUS_EXPR, type1,
927 iv1->step, build_int_cst (type1, 1));
929 if (!POINTER_TYPE_P (type))
931 bound = fold_build2 (PLUS_EXPR, type1,
932 TYPE_MAX_VALUE (type), diff);
933 assumption = fold_build2 (LE_EXPR, boolean_type_node,
934 iv1->base, bound);
937 mbzl = fold_convert (type1, iv0->base);
938 mbzr = fold_build2 (MINUS_EXPR, type1,
939 fold_convert (type1, iv1->base), diff);
942 if (!integer_nonzerop (assumption))
943 niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
944 niter->assumptions, assumption);
945 if (!rolls_p)
947 mbz = fold_build2 (GT_EXPR, boolean_type_node, mbzl, mbzr);
948 niter->may_be_zero = fold_build2 (TRUTH_OR_EXPR, boolean_type_node,
949 niter->may_be_zero, mbz);
953 /* Determines number of iterations of loop whose ending condition
954 is IV0 < IV1. TYPE is the type of the iv. The number of
955 iterations is stored to NITER. BNDS bounds the difference
956 IV1->base - IV0->base. */
958 static bool
959 number_of_iterations_lt (tree type, affine_iv *iv0, affine_iv *iv1,
960 struct tree_niter_desc *niter,
961 bool never_infinite ATTRIBUTE_UNUSED,
962 bounds *bnds)
964 tree niter_type = unsigned_type_for (type);
965 tree delta, step, s;
966 mpz_t mstep, tmp;
968 if (integer_nonzerop (iv0->step))
970 niter->control = *iv0;
971 niter->cmp = LT_EXPR;
972 niter->bound = iv1->base;
974 else
976 niter->control = *iv1;
977 niter->cmp = GT_EXPR;
978 niter->bound = iv0->base;
981 delta = fold_build2 (MINUS_EXPR, niter_type,
982 fold_convert (niter_type, iv1->base),
983 fold_convert (niter_type, iv0->base));
985 /* First handle the special case that the step is +-1. */
986 if ((integer_onep (iv0->step) && integer_zerop (iv1->step))
987 || (integer_all_onesp (iv1->step) && integer_zerop (iv0->step)))
989 /* for (i = iv0->base; i < iv1->base; i++)
993 for (i = iv1->base; i > iv0->base; i--).
995 In both cases # of iterations is iv1->base - iv0->base, assuming that
996 iv1->base >= iv0->base.
998 First try to derive a lower bound on the value of
999 iv1->base - iv0->base, computed in full precision. If the difference
1000 is nonnegative, we are done, otherwise we must record the
1001 condition. */
1003 if (mpz_sgn (bnds->below) < 0)
1004 niter->may_be_zero = fold_build2 (LT_EXPR, boolean_type_node,
1005 iv1->base, iv0->base);
1006 niter->niter = delta;
1007 niter->max = mpz_get_double_int (niter_type, bnds->up, false);
1008 return true;
1011 if (integer_nonzerop (iv0->step))
1012 step = fold_convert (niter_type, iv0->step);
1013 else
1014 step = fold_convert (niter_type,
1015 fold_build1 (NEGATE_EXPR, type, iv1->step));
1017 /* If we can determine the final value of the control iv exactly, we can
1018 transform the condition to != comparison. In particular, this will be
1019 the case if DELTA is constant. */
1020 if (number_of_iterations_lt_to_ne (type, iv0, iv1, niter, &delta, step,
1021 bnds))
1023 affine_iv zps;
1025 zps.base = build_int_cst (niter_type, 0);
1026 zps.step = step;
1027 /* number_of_iterations_lt_to_ne will add assumptions that ensure that
1028 zps does not overflow. */
1029 zps.no_overflow = true;
1031 return number_of_iterations_ne (type, &zps, delta, niter, true, bnds);
1034 /* Make sure that the control iv does not overflow. */
1035 if (!assert_no_overflow_lt (type, iv0, iv1, niter, step))
1036 return false;
1038 /* We determine the number of iterations as (delta + step - 1) / step. For
1039 this to work, we must know that iv1->base >= iv0->base - step + 1,
1040 otherwise the loop does not roll. */
1041 assert_loop_rolls_lt (type, iv0, iv1, niter, bnds);
1043 s = fold_build2 (MINUS_EXPR, niter_type,
1044 step, build_int_cst (niter_type, 1));
1045 delta = fold_build2 (PLUS_EXPR, niter_type, delta, s);
1046 niter->niter = fold_build2 (FLOOR_DIV_EXPR, niter_type, delta, step);
1048 mpz_init (mstep);
1049 mpz_init (tmp);
1050 mpz_set_double_int (mstep, tree_to_double_int (step), true);
1051 mpz_add (tmp, bnds->up, mstep);
1052 mpz_sub_ui (tmp, tmp, 1);
1053 mpz_fdiv_q (tmp, tmp, mstep);
1054 niter->max = mpz_get_double_int (niter_type, tmp, false);
1055 mpz_clear (mstep);
1056 mpz_clear (tmp);
1058 return true;
1061 /* Determines number of iterations of loop whose ending condition
1062 is IV0 <= IV1. TYPE is the type of the iv. The number of
1063 iterations is stored to NITER. NEVER_INFINITE is true if
1064 we know that this condition must eventually become false (we derived this
1065 earlier, and possibly set NITER->assumptions to make sure this
1066 is the case). BNDS bounds the difference IV1->base - IV0->base. */
1068 static bool
1069 number_of_iterations_le (tree type, affine_iv *iv0, affine_iv *iv1,
1070 struct tree_niter_desc *niter, bool never_infinite,
1071 bounds *bnds)
1073 tree assumption;
1074 tree type1 = type;
1075 if (POINTER_TYPE_P (type))
1076 type1 = sizetype;
1078 /* Say that IV0 is the control variable. Then IV0 <= IV1 iff
1079 IV0 < IV1 + 1, assuming that IV1 is not equal to the greatest
1080 value of the type. This we must know anyway, since if it is
1081 equal to this value, the loop rolls forever. */
1083 if (!never_infinite)
1085 if (integer_nonzerop (iv0->step))
1086 assumption = fold_build2 (NE_EXPR, boolean_type_node,
1087 iv1->base, TYPE_MAX_VALUE (type1));
1088 else
1089 assumption = fold_build2 (NE_EXPR, boolean_type_node,
1090 iv0->base, TYPE_MIN_VALUE (type1));
1092 if (integer_zerop (assumption))
1093 return false;
1094 if (!integer_nonzerop (assumption))
1095 niter->assumptions = fold_build2 (TRUTH_AND_EXPR, boolean_type_node,
1096 niter->assumptions, assumption);
1099 if (integer_nonzerop (iv0->step))
1100 iv1->base = fold_build2 (PLUS_EXPR, type1,
1101 iv1->base, build_int_cst (type1, 1));
1102 else
1103 iv0->base = fold_build2 (MINUS_EXPR, type1,
1104 iv0->base, build_int_cst (type1, 1));
1106 bounds_add (bnds, double_int_one, type1);
1108 return number_of_iterations_lt (type, iv0, iv1, niter, never_infinite, bnds);
1111 /* Dumps description of affine induction variable IV to FILE. */
1113 static void
1114 dump_affine_iv (FILE *file, affine_iv *iv)
1116 if (!integer_zerop (iv->step))
1117 fprintf (file, "[");
1119 print_generic_expr (dump_file, iv->base, TDF_SLIM);
1121 if (!integer_zerop (iv->step))
1123 fprintf (file, ", + , ");
1124 print_generic_expr (dump_file, iv->step, TDF_SLIM);
1125 fprintf (file, "]%s", iv->no_overflow ? "(no_overflow)" : "");
1129 /* Determine the number of iterations according to condition (for staying
1130 inside loop) which compares two induction variables using comparison
1131 operator CODE. The induction variable on left side of the comparison
1132 is IV0, the right-hand side is IV1. Both induction variables must have
1133 type TYPE, which must be an integer or pointer type. The steps of the
1134 ivs must be constants (or NULL_TREE, which is interpreted as constant zero).
1136 LOOP is the loop whose number of iterations we are determining.
1138 ONLY_EXIT is true if we are sure this is the only way the loop could be
1139 exited (including possibly non-returning function calls, exceptions, etc.)
1140 -- in this case we can use the information whether the control induction
1141 variables can overflow or not in a more efficient way.
1143 The results (number of iterations and assumptions as described in
1144 comments at struct tree_niter_desc in tree-flow.h) are stored to NITER.
1145 Returns false if it fails to determine number of iterations, true if it
1146 was determined (possibly with some assumptions). */
1148 static bool
1149 number_of_iterations_cond (struct loop *loop,
1150 tree type, affine_iv *iv0, enum tree_code code,
1151 affine_iv *iv1, struct tree_niter_desc *niter,
1152 bool only_exit)
1154 bool never_infinite, ret;
1155 bounds bnds;
1157 /* The meaning of these assumptions is this:
1158 if !assumptions
1159 then the rest of information does not have to be valid
1160 if may_be_zero then the loop does not roll, even if
1161 niter != 0. */
1162 niter->assumptions = boolean_true_node;
1163 niter->may_be_zero = boolean_false_node;
1164 niter->niter = NULL_TREE;
1165 niter->max = double_int_zero;
1167 niter->bound = NULL_TREE;
1168 niter->cmp = ERROR_MARK;
1170 /* Make < comparison from > ones, and for NE_EXPR comparisons, ensure that
1171 the control variable is on lhs. */
1172 if (code == GE_EXPR || code == GT_EXPR
1173 || (code == NE_EXPR && integer_zerop (iv0->step)))
1175 SWAP (iv0, iv1);
1176 code = swap_tree_comparison (code);
1179 if (!only_exit)
1181 /* If this is not the only possible exit from the loop, the information
1182 that the induction variables cannot overflow as derived from
1183 signedness analysis cannot be relied upon. We use them e.g. in the
1184 following way: given loop for (i = 0; i <= n; i++), if i is
1185 signed, it cannot overflow, thus this loop is equivalent to
1186 for (i = 0; i < n + 1; i++); however, if n == MAX, but the loop
1187 is exited in some other way before i overflows, this transformation
1188 is incorrect (the new loop exits immediately). */
1189 iv0->no_overflow = false;
1190 iv1->no_overflow = false;
1193 if (POINTER_TYPE_P (type))
1195 /* Comparison of pointers is undefined unless both iv0 and iv1 point
1196 to the same object. If they do, the control variable cannot wrap
1197 (as wrap around the bounds of memory will never return a pointer
1198 that would be guaranteed to point to the same object, even if we
1199 avoid undefined behavior by casting to size_t and back). The
1200 restrictions on pointer arithmetics and comparisons of pointers
1201 ensure that using the no-overflow assumptions is correct in this
1202 case even if ONLY_EXIT is false. */
1203 iv0->no_overflow = true;
1204 iv1->no_overflow = true;
1207 /* If the control induction variable does not overflow, the loop obviously
1208 cannot be infinite. */
1209 if (!integer_zerop (iv0->step) && iv0->no_overflow)
1210 never_infinite = true;
1211 else if (!integer_zerop (iv1->step) && iv1->no_overflow)
1212 never_infinite = true;
1213 else
1214 never_infinite = false;
1216 /* We can handle the case when neither of the sides of the comparison is
1217 invariant, provided that the test is NE_EXPR. This rarely occurs in
1218 practice, but it is simple enough to manage. */
1219 if (!integer_zerop (iv0->step) && !integer_zerop (iv1->step))
1221 if (code != NE_EXPR)
1222 return false;
1224 iv0->step = fold_binary_to_constant (MINUS_EXPR, type,
1225 iv0->step, iv1->step);
1226 iv0->no_overflow = false;
1227 iv1->step = build_int_cst (type, 0);
1228 iv1->no_overflow = true;
1231 /* If the result of the comparison is a constant, the loop is weird. More
1232 precise handling would be possible, but the situation is not common enough
1233 to waste time on it. */
1234 if (integer_zerop (iv0->step) && integer_zerop (iv1->step))
1235 return false;
1237 /* Ignore loops of while (i-- < 10) type. */
1238 if (code != NE_EXPR)
1240 if (iv0->step && tree_int_cst_sign_bit (iv0->step))
1241 return false;
1243 if (!integer_zerop (iv1->step) && !tree_int_cst_sign_bit (iv1->step))
1244 return false;
1247 /* If the loop exits immediately, there is nothing to do. */
1248 if (integer_zerop (fold_build2 (code, boolean_type_node, iv0->base, iv1->base)))
1250 niter->niter = build_int_cst (unsigned_type_for (type), 0);
1251 niter->max = double_int_zero;
1252 return true;
1255 /* OK, now we know we have a senseful loop. Handle several cases, depending
1256 on what comparison operator is used. */
1257 bound_difference (loop, iv1->base, iv0->base, &bnds);
1259 if (dump_file && (dump_flags & TDF_DETAILS))
1261 fprintf (dump_file,
1262 "Analyzing # of iterations of loop %d\n", loop->num);
1264 fprintf (dump_file, " exit condition ");
1265 dump_affine_iv (dump_file, iv0);
1266 fprintf (dump_file, " %s ",
1267 code == NE_EXPR ? "!="
1268 : code == LT_EXPR ? "<"
1269 : "<=");
1270 dump_affine_iv (dump_file, iv1);
1271 fprintf (dump_file, "\n");
1273 fprintf (dump_file, " bounds on difference of bases: ");
1274 mpz_out_str (dump_file, 10, bnds.below);
1275 fprintf (dump_file, " ... ");
1276 mpz_out_str (dump_file, 10, bnds.up);
1277 fprintf (dump_file, "\n");
1280 switch (code)
1282 case NE_EXPR:
1283 gcc_assert (integer_zerop (iv1->step));
1284 ret = number_of_iterations_ne (type, iv0, iv1->base, niter,
1285 never_infinite, &bnds);
1286 break;
1288 case LT_EXPR:
1289 ret = number_of_iterations_lt (type, iv0, iv1, niter, never_infinite,
1290 &bnds);
1291 break;
1293 case LE_EXPR:
1294 ret = number_of_iterations_le (type, iv0, iv1, niter, never_infinite,
1295 &bnds);
1296 break;
1298 default:
1299 gcc_unreachable ();
1302 mpz_clear (bnds.up);
1303 mpz_clear (bnds.below);
1305 if (dump_file && (dump_flags & TDF_DETAILS))
1307 if (ret)
1309 fprintf (dump_file, " result:\n");
1310 if (!integer_nonzerop (niter->assumptions))
1312 fprintf (dump_file, " under assumptions ");
1313 print_generic_expr (dump_file, niter->assumptions, TDF_SLIM);
1314 fprintf (dump_file, "\n");
1317 if (!integer_zerop (niter->may_be_zero))
1319 fprintf (dump_file, " zero if ");
1320 print_generic_expr (dump_file, niter->may_be_zero, TDF_SLIM);
1321 fprintf (dump_file, "\n");
1324 fprintf (dump_file, " # of iterations ");
1325 print_generic_expr (dump_file, niter->niter, TDF_SLIM);
1326 fprintf (dump_file, ", bounded by ");
1327 dump_double_int (dump_file, niter->max, true);
1328 fprintf (dump_file, "\n");
1330 else
1331 fprintf (dump_file, " failed\n\n");
1333 return ret;
1336 /* Substitute NEW for OLD in EXPR and fold the result. */
1338 static tree
1339 simplify_replace_tree (tree expr, tree old, tree new_tree)
1341 unsigned i, n;
1342 tree ret = NULL_TREE, e, se;
1344 if (!expr)
1345 return NULL_TREE;
1347 if (expr == old
1348 || operand_equal_p (expr, old, 0))
1349 return unshare_expr (new_tree);
1351 if (!EXPR_P (expr))
1352 return expr;
1354 n = TREE_OPERAND_LENGTH (expr);
1355 for (i = 0; i < n; i++)
1357 e = TREE_OPERAND (expr, i);
1358 se = simplify_replace_tree (e, old, new_tree);
1359 if (e == se)
1360 continue;
1362 if (!ret)
1363 ret = copy_node (expr);
1365 TREE_OPERAND (ret, i) = se;
1368 return (ret ? fold (ret) : expr);
1371 /* Expand definitions of ssa names in EXPR as long as they are simple
1372 enough, and return the new expression. */
1374 tree
1375 expand_simple_operations (tree expr)
1377 unsigned i, n;
1378 tree ret = NULL_TREE, e, ee, e1;
1379 enum tree_code code;
1380 gimple stmt;
1382 if (expr == NULL_TREE)
1383 return expr;
1385 if (is_gimple_min_invariant (expr))
1386 return expr;
1388 code = TREE_CODE (expr);
1389 if (IS_EXPR_CODE_CLASS (TREE_CODE_CLASS (code)))
1391 n = TREE_OPERAND_LENGTH (expr);
1392 for (i = 0; i < n; i++)
1394 e = TREE_OPERAND (expr, i);
1395 ee = expand_simple_operations (e);
1396 if (e == ee)
1397 continue;
1399 if (!ret)
1400 ret = copy_node (expr);
1402 TREE_OPERAND (ret, i) = ee;
1405 if (!ret)
1406 return expr;
1408 fold_defer_overflow_warnings ();
1409 ret = fold (ret);
1410 fold_undefer_and_ignore_overflow_warnings ();
1411 return ret;
1414 if (TREE_CODE (expr) != SSA_NAME)
1415 return expr;
1417 stmt = SSA_NAME_DEF_STMT (expr);
1418 if (gimple_code (stmt) == GIMPLE_PHI)
1420 basic_block src, dest;
1422 if (gimple_phi_num_args (stmt) != 1)
1423 return expr;
1424 e = PHI_ARG_DEF (stmt, 0);
1426 /* Avoid propagating through loop exit phi nodes, which
1427 could break loop-closed SSA form restrictions. */
1428 dest = gimple_bb (stmt);
1429 src = single_pred (dest);
1430 if (TREE_CODE (e) == SSA_NAME
1431 && src->loop_father != dest->loop_father)
1432 return expr;
1434 return expand_simple_operations (e);
1436 if (gimple_code (stmt) != GIMPLE_ASSIGN)
1437 return expr;
1439 e = gimple_assign_rhs1 (stmt);
1440 code = gimple_assign_rhs_code (stmt);
1441 if (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS)
1443 if (is_gimple_min_invariant (e))
1444 return e;
1446 if (code == SSA_NAME)
1447 return expand_simple_operations (e);
1449 return expr;
1452 switch (code)
1454 CASE_CONVERT:
1455 /* Casts are simple. */
1456 ee = expand_simple_operations (e);
1457 return fold_build1 (code, TREE_TYPE (expr), ee);
1459 case PLUS_EXPR:
1460 case MINUS_EXPR:
1461 case POINTER_PLUS_EXPR:
1462 /* And increments and decrements by a constant are simple. */
1463 e1 = gimple_assign_rhs2 (stmt);
1464 if (!is_gimple_min_invariant (e1))
1465 return expr;
1467 ee = expand_simple_operations (e);
1468 return fold_build2 (code, TREE_TYPE (expr), ee, e1);
1470 default:
1471 return expr;
1475 /* Tries to simplify EXPR using the condition COND. Returns the simplified
1476 expression (or EXPR unchanged, if no simplification was possible). */
1478 static tree
1479 tree_simplify_using_condition_1 (tree cond, tree expr)
1481 bool changed;
1482 tree e, te, e0, e1, e2, notcond;
1483 enum tree_code code = TREE_CODE (expr);
1485 if (code == INTEGER_CST)
1486 return expr;
1488 if (code == TRUTH_OR_EXPR
1489 || code == TRUTH_AND_EXPR
1490 || code == COND_EXPR)
1492 changed = false;
1494 e0 = tree_simplify_using_condition_1 (cond, TREE_OPERAND (expr, 0));
1495 if (TREE_OPERAND (expr, 0) != e0)
1496 changed = true;
1498 e1 = tree_simplify_using_condition_1 (cond, TREE_OPERAND (expr, 1));
1499 if (TREE_OPERAND (expr, 1) != e1)
1500 changed = true;
1502 if (code == COND_EXPR)
1504 e2 = tree_simplify_using_condition_1 (cond, TREE_OPERAND (expr, 2));
1505 if (TREE_OPERAND (expr, 2) != e2)
1506 changed = true;
1508 else
1509 e2 = NULL_TREE;
1511 if (changed)
1513 if (code == COND_EXPR)
1514 expr = fold_build3 (code, boolean_type_node, e0, e1, e2);
1515 else
1516 expr = fold_build2 (code, boolean_type_node, e0, e1);
1519 return expr;
1522 /* In case COND is equality, we may be able to simplify EXPR by copy/constant
1523 propagation, and vice versa. Fold does not handle this, since it is
1524 considered too expensive. */
1525 if (TREE_CODE (cond) == EQ_EXPR)
1527 e0 = TREE_OPERAND (cond, 0);
1528 e1 = TREE_OPERAND (cond, 1);
1530 /* We know that e0 == e1. Check whether we cannot simplify expr
1531 using this fact. */
1532 e = simplify_replace_tree (expr, e0, e1);
1533 if (integer_zerop (e) || integer_nonzerop (e))
1534 return e;
1536 e = simplify_replace_tree (expr, e1, e0);
1537 if (integer_zerop (e) || integer_nonzerop (e))
1538 return e;
1540 if (TREE_CODE (expr) == EQ_EXPR)
1542 e0 = TREE_OPERAND (expr, 0);
1543 e1 = TREE_OPERAND (expr, 1);
1545 /* If e0 == e1 (EXPR) implies !COND, then EXPR cannot be true. */
1546 e = simplify_replace_tree (cond, e0, e1);
1547 if (integer_zerop (e))
1548 return e;
1549 e = simplify_replace_tree (cond, e1, e0);
1550 if (integer_zerop (e))
1551 return e;
1553 if (TREE_CODE (expr) == NE_EXPR)
1555 e0 = TREE_OPERAND (expr, 0);
1556 e1 = TREE_OPERAND (expr, 1);
1558 /* If e0 == e1 (!EXPR) implies !COND, then EXPR must be true. */
1559 e = simplify_replace_tree (cond, e0, e1);
1560 if (integer_zerop (e))
1561 return boolean_true_node;
1562 e = simplify_replace_tree (cond, e1, e0);
1563 if (integer_zerop (e))
1564 return boolean_true_node;
1567 te = expand_simple_operations (expr);
1569 /* Check whether COND ==> EXPR. */
1570 notcond = invert_truthvalue (cond);
1571 e = fold_binary (TRUTH_OR_EXPR, boolean_type_node, notcond, te);
1572 if (e && integer_nonzerop (e))
1573 return e;
1575 /* Check whether COND ==> not EXPR. */
1576 e = fold_binary (TRUTH_AND_EXPR, boolean_type_node, cond, te);
1577 if (e && integer_zerop (e))
1578 return e;
1580 return expr;
1583 /* Tries to simplify EXPR using the condition COND. Returns the simplified
1584 expression (or EXPR unchanged, if no simplification was possible).
1585 Wrapper around tree_simplify_using_condition_1 that ensures that chains
1586 of simple operations in definitions of ssa names in COND are expanded,
1587 so that things like casts or incrementing the value of the bound before
1588 the loop do not cause us to fail. */
1590 static tree
1591 tree_simplify_using_condition (tree cond, tree expr)
1593 cond = expand_simple_operations (cond);
1595 return tree_simplify_using_condition_1 (cond, expr);
1598 /* Tries to simplify EXPR using the conditions on entry to LOOP.
1599 Returns the simplified expression (or EXPR unchanged, if no
1600 simplification was possible).*/
1602 static tree
1603 simplify_using_initial_conditions (struct loop *loop, tree expr)
1605 edge e;
1606 basic_block bb;
1607 gimple stmt;
1608 tree cond;
1609 int cnt = 0;
1611 if (TREE_CODE (expr) == INTEGER_CST)
1612 return expr;
1614 /* Limit walking the dominators to avoid quadraticness in
1615 the number of BBs times the number of loops in degenerate
1616 cases. */
1617 for (bb = loop->header;
1618 bb != ENTRY_BLOCK_PTR && cnt < MAX_DOMINATORS_TO_WALK;
1619 bb = get_immediate_dominator (CDI_DOMINATORS, bb))
1621 if (!single_pred_p (bb))
1622 continue;
1623 e = single_pred_edge (bb);
1625 if (!(e->flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
1626 continue;
1628 stmt = last_stmt (e->src);
1629 cond = fold_build2 (gimple_cond_code (stmt),
1630 boolean_type_node,
1631 gimple_cond_lhs (stmt),
1632 gimple_cond_rhs (stmt));
1633 if (e->flags & EDGE_FALSE_VALUE)
1634 cond = invert_truthvalue (cond);
1635 expr = tree_simplify_using_condition (cond, expr);
1636 ++cnt;
1639 return expr;
1642 /* Tries to simplify EXPR using the evolutions of the loop invariants
1643 in the superloops of LOOP. Returns the simplified expression
1644 (or EXPR unchanged, if no simplification was possible). */
1646 static tree
1647 simplify_using_outer_evolutions (struct loop *loop, tree expr)
1649 enum tree_code code = TREE_CODE (expr);
1650 bool changed;
1651 tree e, e0, e1, e2;
1653 if (is_gimple_min_invariant (expr))
1654 return expr;
1656 if (code == TRUTH_OR_EXPR
1657 || code == TRUTH_AND_EXPR
1658 || code == COND_EXPR)
1660 changed = false;
1662 e0 = simplify_using_outer_evolutions (loop, TREE_OPERAND (expr, 0));
1663 if (TREE_OPERAND (expr, 0) != e0)
1664 changed = true;
1666 e1 = simplify_using_outer_evolutions (loop, TREE_OPERAND (expr, 1));
1667 if (TREE_OPERAND (expr, 1) != e1)
1668 changed = true;
1670 if (code == COND_EXPR)
1672 e2 = simplify_using_outer_evolutions (loop, TREE_OPERAND (expr, 2));
1673 if (TREE_OPERAND (expr, 2) != e2)
1674 changed = true;
1676 else
1677 e2 = NULL_TREE;
1679 if (changed)
1681 if (code == COND_EXPR)
1682 expr = fold_build3 (code, boolean_type_node, e0, e1, e2);
1683 else
1684 expr = fold_build2 (code, boolean_type_node, e0, e1);
1687 return expr;
1690 e = instantiate_parameters (loop, expr);
1691 if (is_gimple_min_invariant (e))
1692 return e;
1694 return expr;
1697 /* Returns true if EXIT is the only possible exit from LOOP. */
1699 bool
1700 loop_only_exit_p (const struct loop *loop, const_edge exit)
1702 basic_block *body;
1703 gimple_stmt_iterator bsi;
1704 unsigned i;
1705 gimple call;
1707 if (exit != single_exit (loop))
1708 return false;
1710 body = get_loop_body (loop);
1711 for (i = 0; i < loop->num_nodes; i++)
1713 for (bsi = gsi_start_bb (body[i]); !gsi_end_p (bsi); gsi_next (&bsi))
1715 call = gsi_stmt (bsi);
1716 if (gimple_code (call) != GIMPLE_CALL)
1717 continue;
1719 if (gimple_has_side_effects (call))
1721 free (body);
1722 return false;
1727 free (body);
1728 return true;
1731 /* Stores description of number of iterations of LOOP derived from
1732 EXIT (an exit edge of the LOOP) in NITER. Returns true if some
1733 useful information could be derived (and fields of NITER has
1734 meaning described in comments at struct tree_niter_desc
1735 declaration), false otherwise. If WARN is true and
1736 -Wunsafe-loop-optimizations was given, warn if the optimizer is going to use
1737 potentially unsafe assumptions. */
1739 bool
1740 number_of_iterations_exit (struct loop *loop, edge exit,
1741 struct tree_niter_desc *niter,
1742 bool warn)
1744 gimple stmt;
1745 tree type;
1746 tree op0, op1;
1747 enum tree_code code;
1748 affine_iv iv0, iv1;
1750 if (!dominated_by_p (CDI_DOMINATORS, loop->latch, exit->src))
1751 return false;
1753 niter->assumptions = boolean_false_node;
1754 stmt = last_stmt (exit->src);
1755 if (!stmt || gimple_code (stmt) != GIMPLE_COND)
1756 return false;
1758 /* We want the condition for staying inside loop. */
1759 code = gimple_cond_code (stmt);
1760 if (exit->flags & EDGE_TRUE_VALUE)
1761 code = invert_tree_comparison (code, false);
1763 switch (code)
1765 case GT_EXPR:
1766 case GE_EXPR:
1767 case NE_EXPR:
1768 case LT_EXPR:
1769 case LE_EXPR:
1770 break;
1772 default:
1773 return false;
1776 op0 = gimple_cond_lhs (stmt);
1777 op1 = gimple_cond_rhs (stmt);
1778 type = TREE_TYPE (op0);
1780 if (TREE_CODE (type) != INTEGER_TYPE
1781 && !POINTER_TYPE_P (type))
1782 return false;
1784 if (!simple_iv (loop, stmt, op0, &iv0, false))
1785 return false;
1786 if (!simple_iv (loop, stmt, op1, &iv1, false))
1787 return false;
1789 /* We don't want to see undefined signed overflow warnings while
1790 computing the number of iterations. */
1791 fold_defer_overflow_warnings ();
1793 iv0.base = expand_simple_operations (iv0.base);
1794 iv1.base = expand_simple_operations (iv1.base);
1795 if (!number_of_iterations_cond (loop, type, &iv0, code, &iv1, niter,
1796 loop_only_exit_p (loop, exit)))
1798 fold_undefer_and_ignore_overflow_warnings ();
1799 return false;
1802 if (optimize >= 3)
1804 niter->assumptions = simplify_using_outer_evolutions (loop,
1805 niter->assumptions);
1806 niter->may_be_zero = simplify_using_outer_evolutions (loop,
1807 niter->may_be_zero);
1808 niter->niter = simplify_using_outer_evolutions (loop, niter->niter);
1811 niter->assumptions
1812 = simplify_using_initial_conditions (loop,
1813 niter->assumptions);
1814 niter->may_be_zero
1815 = simplify_using_initial_conditions (loop,
1816 niter->may_be_zero);
1818 fold_undefer_and_ignore_overflow_warnings ();
1820 if (integer_onep (niter->assumptions))
1821 return true;
1823 /* With -funsafe-loop-optimizations we assume that nothing bad can happen.
1824 But if we can prove that there is overflow or some other source of weird
1825 behavior, ignore the loop even with -funsafe-loop-optimizations. */
1826 if (integer_zerop (niter->assumptions))
1827 return false;
1829 if (flag_unsafe_loop_optimizations)
1830 niter->assumptions = boolean_true_node;
1832 if (warn)
1834 const char *wording;
1835 location_t loc = gimple_location (stmt);
1837 /* We can provide a more specific warning if one of the operator is
1838 constant and the other advances by +1 or -1. */
1839 if (!integer_zerop (iv1.step)
1840 ? (integer_zerop (iv0.step)
1841 && (integer_onep (iv1.step) || integer_all_onesp (iv1.step)))
1842 : (integer_onep (iv0.step) || integer_all_onesp (iv0.step)))
1843 wording =
1844 flag_unsafe_loop_optimizations
1845 ? N_("assuming that the loop is not infinite")
1846 : N_("cannot optimize possibly infinite loops");
1847 else
1848 wording =
1849 flag_unsafe_loop_optimizations
1850 ? N_("assuming that the loop counter does not overflow")
1851 : N_("cannot optimize loop, the loop counter may overflow");
1853 if (LOCATION_LINE (loc) > 0)
1854 warning (OPT_Wunsafe_loop_optimizations, "%H%s", &loc, gettext (wording));
1855 else
1856 warning (OPT_Wunsafe_loop_optimizations, "%s", gettext (wording));
1859 return flag_unsafe_loop_optimizations;
1862 /* Try to determine the number of iterations of LOOP. If we succeed,
1863 expression giving number of iterations is returned and *EXIT is
1864 set to the edge from that the information is obtained. Otherwise
1865 chrec_dont_know is returned. */
1867 tree
1868 find_loop_niter (struct loop *loop, edge *exit)
1870 unsigned i;
1871 VEC (edge, heap) *exits = get_loop_exit_edges (loop);
1872 edge ex;
1873 tree niter = NULL_TREE, aniter;
1874 struct tree_niter_desc desc;
1876 *exit = NULL;
1877 for (i = 0; VEC_iterate (edge, exits, i, ex); i++)
1879 if (!just_once_each_iteration_p (loop, ex->src))
1880 continue;
1882 if (!number_of_iterations_exit (loop, ex, &desc, false))
1883 continue;
1885 if (integer_nonzerop (desc.may_be_zero))
1887 /* We exit in the first iteration through this exit.
1888 We won't find anything better. */
1889 niter = build_int_cst (unsigned_type_node, 0);
1890 *exit = ex;
1891 break;
1894 if (!integer_zerop (desc.may_be_zero))
1895 continue;
1897 aniter = desc.niter;
1899 if (!niter)
1901 /* Nothing recorded yet. */
1902 niter = aniter;
1903 *exit = ex;
1904 continue;
1907 /* Prefer constants, the lower the better. */
1908 if (TREE_CODE (aniter) != INTEGER_CST)
1909 continue;
1911 if (TREE_CODE (niter) != INTEGER_CST)
1913 niter = aniter;
1914 *exit = ex;
1915 continue;
1918 if (tree_int_cst_lt (aniter, niter))
1920 niter = aniter;
1921 *exit = ex;
1922 continue;
1925 VEC_free (edge, heap, exits);
1927 return niter ? niter : chrec_dont_know;
1932 Analysis of a number of iterations of a loop by a brute-force evaluation.
1936 /* Bound on the number of iterations we try to evaluate. */
1938 #define MAX_ITERATIONS_TO_TRACK \
1939 ((unsigned) PARAM_VALUE (PARAM_MAX_ITERATIONS_TO_TRACK))
1941 /* Returns the loop phi node of LOOP such that ssa name X is derived from its
1942 result by a chain of operations such that all but exactly one of their
1943 operands are constants. */
1945 static gimple
1946 chain_of_csts_start (struct loop *loop, tree x)
1948 gimple stmt = SSA_NAME_DEF_STMT (x);
1949 tree use;
1950 basic_block bb = gimple_bb (stmt);
1951 enum tree_code code;
1953 if (!bb
1954 || !flow_bb_inside_loop_p (loop, bb))
1955 return NULL;
1957 if (gimple_code (stmt) == GIMPLE_PHI)
1959 if (bb == loop->header)
1960 return stmt;
1962 return NULL;
1965 if (gimple_code (stmt) != GIMPLE_ASSIGN)
1966 return NULL;
1968 code = gimple_assign_rhs_code (stmt);
1969 if (gimple_references_memory_p (stmt)
1970 /* Before alias information is computed, operand scanning marks
1971 statements that write memory volatile. However, the statements
1972 that only read memory are not marked, thus gimple_references_memory_p
1973 returns false for them. */
1974 || TREE_CODE_CLASS (code) == tcc_reference
1975 || TREE_CODE_CLASS (code) == tcc_declaration
1976 || SINGLE_SSA_DEF_OPERAND (stmt, SSA_OP_DEF) == NULL_DEF_OPERAND_P)
1977 return NULL;
1979 use = SINGLE_SSA_TREE_OPERAND (stmt, SSA_OP_USE);
1980 if (use == NULL_USE_OPERAND_P)
1981 return NULL;
1983 return chain_of_csts_start (loop, use);
1986 /* Determines whether the expression X is derived from a result of a phi node
1987 in header of LOOP such that
1989 * the derivation of X consists only from operations with constants
1990 * the initial value of the phi node is constant
1991 * the value of the phi node in the next iteration can be derived from the
1992 value in the current iteration by a chain of operations with constants.
1994 If such phi node exists, it is returned, otherwise NULL is returned. */
1996 static gimple
1997 get_base_for (struct loop *loop, tree x)
1999 gimple phi;
2000 tree init, next;
2002 if (is_gimple_min_invariant (x))
2003 return NULL;
2005 phi = chain_of_csts_start (loop, x);
2006 if (!phi)
2007 return NULL;
2009 init = PHI_ARG_DEF_FROM_EDGE (phi, loop_preheader_edge (loop));
2010 next = PHI_ARG_DEF_FROM_EDGE (phi, loop_latch_edge (loop));
2012 if (TREE_CODE (next) != SSA_NAME)
2013 return NULL;
2015 if (!is_gimple_min_invariant (init))
2016 return NULL;
2018 if (chain_of_csts_start (loop, next) != phi)
2019 return NULL;
2021 return phi;
2024 /* Given an expression X, then
2026 * if X is NULL_TREE, we return the constant BASE.
2027 * otherwise X is a SSA name, whose value in the considered loop is derived
2028 by a chain of operations with constant from a result of a phi node in
2029 the header of the loop. Then we return value of X when the value of the
2030 result of this phi node is given by the constant BASE. */
2032 static tree
2033 get_val_for (tree x, tree base)
2035 gimple stmt;
2037 gcc_assert (is_gimple_min_invariant (base));
2039 if (!x)
2040 return base;
2042 stmt = SSA_NAME_DEF_STMT (x);
2043 if (gimple_code (stmt) == GIMPLE_PHI)
2044 return base;
2046 gcc_assert (is_gimple_assign (stmt));
2048 /* STMT must be either an assignment of a single SSA name or an
2049 expression involving an SSA name and a constant. Try to fold that
2050 expression using the value for the SSA name. */
2051 if (gimple_assign_ssa_name_copy_p (stmt))
2052 return get_val_for (gimple_assign_rhs1 (stmt), base);
2053 else if (gimple_assign_rhs_class (stmt) == GIMPLE_UNARY_RHS
2054 && TREE_CODE (gimple_assign_rhs1 (stmt)) == SSA_NAME)
2056 return fold_build1 (gimple_assign_rhs_code (stmt),
2057 gimple_expr_type (stmt),
2058 get_val_for (gimple_assign_rhs1 (stmt), base));
2060 else if (gimple_assign_rhs_class (stmt) == GIMPLE_BINARY_RHS)
2062 tree rhs1 = gimple_assign_rhs1 (stmt);
2063 tree rhs2 = gimple_assign_rhs2 (stmt);
2064 if (TREE_CODE (rhs1) == SSA_NAME)
2065 rhs1 = get_val_for (rhs1, base);
2066 else if (TREE_CODE (rhs2) == SSA_NAME)
2067 rhs2 = get_val_for (rhs2, base);
2068 else
2069 gcc_unreachable ();
2070 return fold_build2 (gimple_assign_rhs_code (stmt),
2071 gimple_expr_type (stmt), rhs1, rhs2);
2073 else
2074 gcc_unreachable ();
2078 /* Tries to count the number of iterations of LOOP till it exits by EXIT
2079 by brute force -- i.e. by determining the value of the operands of the
2080 condition at EXIT in first few iterations of the loop (assuming that
2081 these values are constant) and determining the first one in that the
2082 condition is not satisfied. Returns the constant giving the number
2083 of the iterations of LOOP if successful, chrec_dont_know otherwise. */
2085 tree
2086 loop_niter_by_eval (struct loop *loop, edge exit)
2088 tree acnd;
2089 tree op[2], val[2], next[2], aval[2];
2090 gimple phi, cond;
2091 unsigned i, j;
2092 enum tree_code cmp;
2094 cond = last_stmt (exit->src);
2095 if (!cond || gimple_code (cond) != GIMPLE_COND)
2096 return chrec_dont_know;
2098 cmp = gimple_cond_code (cond);
2099 if (exit->flags & EDGE_TRUE_VALUE)
2100 cmp = invert_tree_comparison (cmp, false);
2102 switch (cmp)
2104 case EQ_EXPR:
2105 case NE_EXPR:
2106 case GT_EXPR:
2107 case GE_EXPR:
2108 case LT_EXPR:
2109 case LE_EXPR:
2110 op[0] = gimple_cond_lhs (cond);
2111 op[1] = gimple_cond_rhs (cond);
2112 break;
2114 default:
2115 return chrec_dont_know;
2118 for (j = 0; j < 2; j++)
2120 if (is_gimple_min_invariant (op[j]))
2122 val[j] = op[j];
2123 next[j] = NULL_TREE;
2124 op[j] = NULL_TREE;
2126 else
2128 phi = get_base_for (loop, op[j]);
2129 if (!phi)
2130 return chrec_dont_know;
2131 val[j] = PHI_ARG_DEF_FROM_EDGE (phi, loop_preheader_edge (loop));
2132 next[j] = PHI_ARG_DEF_FROM_EDGE (phi, loop_latch_edge (loop));
2136 /* Don't issue signed overflow warnings. */
2137 fold_defer_overflow_warnings ();
2139 for (i = 0; i < MAX_ITERATIONS_TO_TRACK; i++)
2141 for (j = 0; j < 2; j++)
2142 aval[j] = get_val_for (op[j], val[j]);
2144 acnd = fold_binary (cmp, boolean_type_node, aval[0], aval[1]);
2145 if (acnd && integer_zerop (acnd))
2147 fold_undefer_and_ignore_overflow_warnings ();
2148 if (dump_file && (dump_flags & TDF_DETAILS))
2149 fprintf (dump_file,
2150 "Proved that loop %d iterates %d times using brute force.\n",
2151 loop->num, i);
2152 return build_int_cst (unsigned_type_node, i);
2155 for (j = 0; j < 2; j++)
2157 val[j] = get_val_for (next[j], val[j]);
2158 if (!is_gimple_min_invariant (val[j]))
2160 fold_undefer_and_ignore_overflow_warnings ();
2161 return chrec_dont_know;
2166 fold_undefer_and_ignore_overflow_warnings ();
2168 return chrec_dont_know;
2171 /* Finds the exit of the LOOP by that the loop exits after a constant
2172 number of iterations and stores the exit edge to *EXIT. The constant
2173 giving the number of iterations of LOOP is returned. The number of
2174 iterations is determined using loop_niter_by_eval (i.e. by brute force
2175 evaluation). If we are unable to find the exit for that loop_niter_by_eval
2176 determines the number of iterations, chrec_dont_know is returned. */
2178 tree
2179 find_loop_niter_by_eval (struct loop *loop, edge *exit)
2181 unsigned i;
2182 VEC (edge, heap) *exits = get_loop_exit_edges (loop);
2183 edge ex;
2184 tree niter = NULL_TREE, aniter;
2186 *exit = NULL;
2187 for (i = 0; VEC_iterate (edge, exits, i, ex); i++)
2189 if (!just_once_each_iteration_p (loop, ex->src))
2190 continue;
2192 aniter = loop_niter_by_eval (loop, ex);
2193 if (chrec_contains_undetermined (aniter))
2194 continue;
2196 if (niter
2197 && !tree_int_cst_lt (aniter, niter))
2198 continue;
2200 niter = aniter;
2201 *exit = ex;
2203 VEC_free (edge, heap, exits);
2205 return niter ? niter : chrec_dont_know;
2210 Analysis of upper bounds on number of iterations of a loop.
2214 static double_int derive_constant_upper_bound_ops (tree, tree,
2215 enum tree_code, tree);
2217 /* Returns a constant upper bound on the value of the right-hand side of
2218 an assignment statement STMT. */
2220 static double_int
2221 derive_constant_upper_bound_assign (gimple stmt)
2223 enum tree_code code = gimple_assign_rhs_code (stmt);
2224 tree op0 = gimple_assign_rhs1 (stmt);
2225 tree op1 = gimple_assign_rhs2 (stmt);
2227 return derive_constant_upper_bound_ops (TREE_TYPE (gimple_assign_lhs (stmt)),
2228 op0, code, op1);
2231 /* Returns a constant upper bound on the value of expression VAL. VAL
2232 is considered to be unsigned. If its type is signed, its value must
2233 be nonnegative. */
2235 static double_int
2236 derive_constant_upper_bound (tree val)
2238 enum tree_code code;
2239 tree op0, op1;
2241 extract_ops_from_tree (val, &code, &op0, &op1);
2242 return derive_constant_upper_bound_ops (TREE_TYPE (val), op0, code, op1);
2245 /* Returns a constant upper bound on the value of expression OP0 CODE OP1,
2246 whose type is TYPE. The expression is considered to be unsigned. If
2247 its type is signed, its value must be nonnegative. */
2249 static double_int
2250 derive_constant_upper_bound_ops (tree type, tree op0,
2251 enum tree_code code, tree op1)
2253 tree subtype, maxt;
2254 double_int bnd, max, mmax, cst;
2255 gimple stmt;
2257 if (INTEGRAL_TYPE_P (type))
2258 maxt = TYPE_MAX_VALUE (type);
2259 else
2260 maxt = upper_bound_in_type (type, type);
2262 max = tree_to_double_int (maxt);
2264 switch (code)
2266 case INTEGER_CST:
2267 return tree_to_double_int (op0);
2269 CASE_CONVERT:
2270 subtype = TREE_TYPE (op0);
2271 if (!TYPE_UNSIGNED (subtype)
2272 /* If TYPE is also signed, the fact that VAL is nonnegative implies
2273 that OP0 is nonnegative. */
2274 && TYPE_UNSIGNED (type)
2275 && !tree_expr_nonnegative_p (op0))
2277 /* If we cannot prove that the casted expression is nonnegative,
2278 we cannot establish more useful upper bound than the precision
2279 of the type gives us. */
2280 return max;
2283 /* We now know that op0 is an nonnegative value. Try deriving an upper
2284 bound for it. */
2285 bnd = derive_constant_upper_bound (op0);
2287 /* If the bound does not fit in TYPE, max. value of TYPE could be
2288 attained. */
2289 if (double_int_ucmp (max, bnd) < 0)
2290 return max;
2292 return bnd;
2294 case PLUS_EXPR:
2295 case POINTER_PLUS_EXPR:
2296 case MINUS_EXPR:
2297 if (TREE_CODE (op1) != INTEGER_CST
2298 || !tree_expr_nonnegative_p (op0))
2299 return max;
2301 /* Canonicalize to OP0 - CST. Consider CST to be signed, in order to
2302 choose the most logical way how to treat this constant regardless
2303 of the signedness of the type. */
2304 cst = tree_to_double_int (op1);
2305 cst = double_int_sext (cst, TYPE_PRECISION (type));
2306 if (code != MINUS_EXPR)
2307 cst = double_int_neg (cst);
2309 bnd = derive_constant_upper_bound (op0);
2311 if (double_int_negative_p (cst))
2313 cst = double_int_neg (cst);
2314 /* Avoid CST == 0x80000... */
2315 if (double_int_negative_p (cst))
2316 return max;;
2318 /* OP0 + CST. We need to check that
2319 BND <= MAX (type) - CST. */
2321 mmax = double_int_add (max, double_int_neg (cst));
2322 if (double_int_ucmp (bnd, mmax) > 0)
2323 return max;
2325 return double_int_add (bnd, cst);
2327 else
2329 /* OP0 - CST, where CST >= 0.
2331 If TYPE is signed, we have already verified that OP0 >= 0, and we
2332 know that the result is nonnegative. This implies that
2333 VAL <= BND - CST.
2335 If TYPE is unsigned, we must additionally know that OP0 >= CST,
2336 otherwise the operation underflows.
2339 /* This should only happen if the type is unsigned; however, for
2340 buggy programs that use overflowing signed arithmetics even with
2341 -fno-wrapv, this condition may also be true for signed values. */
2342 if (double_int_ucmp (bnd, cst) < 0)
2343 return max;
2345 if (TYPE_UNSIGNED (type))
2347 tree tem = fold_binary (GE_EXPR, boolean_type_node, op0,
2348 double_int_to_tree (type, cst));
2349 if (!tem || integer_nonzerop (tem))
2350 return max;
2353 bnd = double_int_add (bnd, double_int_neg (cst));
2356 return bnd;
2358 case FLOOR_DIV_EXPR:
2359 case EXACT_DIV_EXPR:
2360 if (TREE_CODE (op1) != INTEGER_CST
2361 || tree_int_cst_sign_bit (op1))
2362 return max;
2364 bnd = derive_constant_upper_bound (op0);
2365 return double_int_udiv (bnd, tree_to_double_int (op1), FLOOR_DIV_EXPR);
2367 case BIT_AND_EXPR:
2368 if (TREE_CODE (op1) != INTEGER_CST
2369 || tree_int_cst_sign_bit (op1))
2370 return max;
2371 return tree_to_double_int (op1);
2373 case SSA_NAME:
2374 stmt = SSA_NAME_DEF_STMT (op0);
2375 if (gimple_code (stmt) != GIMPLE_ASSIGN
2376 || gimple_assign_lhs (stmt) != op0)
2377 return max;
2378 return derive_constant_upper_bound_assign (stmt);
2380 default:
2381 return max;
2385 /* Records that every statement in LOOP is executed I_BOUND times.
2386 REALISTIC is true if I_BOUND is expected to be close to the real number
2387 of iterations. UPPER is true if we are sure the loop iterates at most
2388 I_BOUND times. */
2390 static void
2391 record_niter_bound (struct loop *loop, double_int i_bound, bool realistic,
2392 bool upper)
2394 /* Update the bounds only when there is no previous estimation, or when the current
2395 estimation is smaller. */
2396 if (upper
2397 && (!loop->any_upper_bound
2398 || double_int_ucmp (i_bound, loop->nb_iterations_upper_bound) < 0))
2400 loop->any_upper_bound = true;
2401 loop->nb_iterations_upper_bound = i_bound;
2403 if (realistic
2404 && (!loop->any_estimate
2405 || double_int_ucmp (i_bound, loop->nb_iterations_estimate) < 0))
2407 loop->any_estimate = true;
2408 loop->nb_iterations_estimate = i_bound;
2412 /* Records that AT_STMT is executed at most BOUND + 1 times in LOOP. IS_EXIT
2413 is true if the loop is exited immediately after STMT, and this exit
2414 is taken at last when the STMT is executed BOUND + 1 times.
2415 REALISTIC is true if BOUND is expected to be close to the real number
2416 of iterations. UPPER is true if we are sure the loop iterates at most
2417 BOUND times. I_BOUND is an unsigned double_int upper estimate on BOUND. */
2419 static void
2420 record_estimate (struct loop *loop, tree bound, double_int i_bound,
2421 gimple at_stmt, bool is_exit, bool realistic, bool upper)
2423 double_int delta;
2424 edge exit;
2426 if (dump_file && (dump_flags & TDF_DETAILS))
2428 fprintf (dump_file, "Statement %s", is_exit ? "(exit)" : "");
2429 print_gimple_stmt (dump_file, at_stmt, 0, TDF_SLIM);
2430 fprintf (dump_file, " is %sexecuted at most ",
2431 upper ? "" : "probably ");
2432 print_generic_expr (dump_file, bound, TDF_SLIM);
2433 fprintf (dump_file, " (bounded by ");
2434 dump_double_int (dump_file, i_bound, true);
2435 fprintf (dump_file, ") + 1 times in loop %d.\n", loop->num);
2438 /* If the I_BOUND is just an estimate of BOUND, it rarely is close to the
2439 real number of iterations. */
2440 if (TREE_CODE (bound) != INTEGER_CST)
2441 realistic = false;
2442 if (!upper && !realistic)
2443 return;
2445 /* If we have a guaranteed upper bound, record it in the appropriate
2446 list. */
2447 if (upper)
2449 struct nb_iter_bound *elt = GGC_NEW (struct nb_iter_bound);
2451 elt->bound = i_bound;
2452 elt->stmt = at_stmt;
2453 elt->is_exit = is_exit;
2454 elt->next = loop->bounds;
2455 loop->bounds = elt;
2458 /* Update the number of iteration estimates according to the bound.
2459 If at_stmt is an exit, then every statement in the loop is
2460 executed at most BOUND + 1 times. If it is not an exit, then
2461 some of the statements before it could be executed BOUND + 2
2462 times, if an exit of LOOP is before stmt. */
2463 exit = single_exit (loop);
2464 if (is_exit
2465 || (exit != NULL
2466 && dominated_by_p (CDI_DOMINATORS,
2467 exit->src, gimple_bb (at_stmt))))
2468 delta = double_int_one;
2469 else
2470 delta = double_int_two;
2471 i_bound = double_int_add (i_bound, delta);
2473 /* If an overflow occurred, ignore the result. */
2474 if (double_int_ucmp (i_bound, delta) < 0)
2475 return;
2477 record_niter_bound (loop, i_bound, realistic, upper);
2480 /* Record the estimate on number of iterations of LOOP based on the fact that
2481 the induction variable BASE + STEP * i evaluated in STMT does not wrap and
2482 its values belong to the range <LOW, HIGH>. REALISTIC is true if the
2483 estimated number of iterations is expected to be close to the real one.
2484 UPPER is true if we are sure the induction variable does not wrap. */
2486 static void
2487 record_nonwrapping_iv (struct loop *loop, tree base, tree step, gimple stmt,
2488 tree low, tree high, bool realistic, bool upper)
2490 tree niter_bound, extreme, delta;
2491 tree type = TREE_TYPE (base), unsigned_type;
2492 double_int max;
2494 if (TREE_CODE (step) != INTEGER_CST || integer_zerop (step))
2495 return;
2497 if (dump_file && (dump_flags & TDF_DETAILS))
2499 fprintf (dump_file, "Induction variable (");
2500 print_generic_expr (dump_file, TREE_TYPE (base), TDF_SLIM);
2501 fprintf (dump_file, ") ");
2502 print_generic_expr (dump_file, base, TDF_SLIM);
2503 fprintf (dump_file, " + ");
2504 print_generic_expr (dump_file, step, TDF_SLIM);
2505 fprintf (dump_file, " * iteration does not wrap in statement ");
2506 print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM);
2507 fprintf (dump_file, " in loop %d.\n", loop->num);
2510 unsigned_type = unsigned_type_for (type);
2511 base = fold_convert (unsigned_type, base);
2512 step = fold_convert (unsigned_type, step);
2514 if (tree_int_cst_sign_bit (step))
2516 extreme = fold_convert (unsigned_type, low);
2517 if (TREE_CODE (base) != INTEGER_CST)
2518 base = fold_convert (unsigned_type, high);
2519 delta = fold_build2 (MINUS_EXPR, unsigned_type, base, extreme);
2520 step = fold_build1 (NEGATE_EXPR, unsigned_type, step);
2522 else
2524 extreme = fold_convert (unsigned_type, high);
2525 if (TREE_CODE (base) != INTEGER_CST)
2526 base = fold_convert (unsigned_type, low);
2527 delta = fold_build2 (MINUS_EXPR, unsigned_type, extreme, base);
2530 /* STMT is executed at most NITER_BOUND + 1 times, since otherwise the value
2531 would get out of the range. */
2532 niter_bound = fold_build2 (FLOOR_DIV_EXPR, unsigned_type, delta, step);
2533 max = derive_constant_upper_bound (niter_bound);
2534 record_estimate (loop, niter_bound, max, stmt, false, realistic, upper);
2537 /* Returns true if REF is a reference to an array at the end of a dynamically
2538 allocated structure. If this is the case, the array may be allocated larger
2539 than its upper bound implies. */
2541 static bool
2542 array_at_struct_end_p (tree ref)
2544 tree base = get_base_address (ref);
2545 tree parent, field;
2547 /* Unless the reference is through a pointer, the size of the array matches
2548 its declaration. */
2549 if (!base || !INDIRECT_REF_P (base))
2550 return false;
2552 for (;handled_component_p (ref); ref = parent)
2554 parent = TREE_OPERAND (ref, 0);
2556 if (TREE_CODE (ref) == COMPONENT_REF)
2558 /* All fields of a union are at its end. */
2559 if (TREE_CODE (TREE_TYPE (parent)) == UNION_TYPE)
2560 continue;
2562 /* Unless the field is at the end of the struct, we are done. */
2563 field = TREE_OPERAND (ref, 1);
2564 if (TREE_CHAIN (field))
2565 return false;
2568 /* The other options are ARRAY_REF, ARRAY_RANGE_REF, VIEW_CONVERT_EXPR.
2569 In all these cases, we might be accessing the last element, and
2570 although in practice this will probably never happen, it is legal for
2571 the indices of this last element to exceed the bounds of the array.
2572 Therefore, continue checking. */
2575 gcc_assert (INDIRECT_REF_P (ref));
2576 return true;
2579 /* Determine information about number of iterations a LOOP from the index
2580 IDX of a data reference accessed in STMT. RELIABLE is true if STMT is
2581 guaranteed to be executed in every iteration of LOOP. Callback for
2582 for_each_index. */
2584 struct ilb_data
2586 struct loop *loop;
2587 gimple stmt;
2588 bool reliable;
2591 static bool
2592 idx_infer_loop_bounds (tree base, tree *idx, void *dta)
2594 struct ilb_data *data = (struct ilb_data *) dta;
2595 tree ev, init, step;
2596 tree low, high, type, next;
2597 bool sign, upper = data->reliable, at_end = false;
2598 struct loop *loop = data->loop;
2600 if (TREE_CODE (base) != ARRAY_REF)
2601 return true;
2603 /* For arrays at the end of the structure, we are not guaranteed that they
2604 do not really extend over their declared size. However, for arrays of
2605 size greater than one, this is unlikely to be intended. */
2606 if (array_at_struct_end_p (base))
2608 at_end = true;
2609 upper = false;
2612 ev = instantiate_parameters (loop, analyze_scalar_evolution (loop, *idx));
2613 init = initial_condition (ev);
2614 step = evolution_part_in_loop_num (ev, loop->num);
2616 if (!init
2617 || !step
2618 || TREE_CODE (step) != INTEGER_CST
2619 || integer_zerop (step)
2620 || tree_contains_chrecs (init, NULL)
2621 || chrec_contains_symbols_defined_in_loop (init, loop->num))
2622 return true;
2624 low = array_ref_low_bound (base);
2625 high = array_ref_up_bound (base);
2627 /* The case of nonconstant bounds could be handled, but it would be
2628 complicated. */
2629 if (TREE_CODE (low) != INTEGER_CST
2630 || !high
2631 || TREE_CODE (high) != INTEGER_CST)
2632 return true;
2633 sign = tree_int_cst_sign_bit (step);
2634 type = TREE_TYPE (step);
2636 /* The array of length 1 at the end of a structure most likely extends
2637 beyond its bounds. */
2638 if (at_end
2639 && operand_equal_p (low, high, 0))
2640 return true;
2642 /* In case the relevant bound of the array does not fit in type, or
2643 it does, but bound + step (in type) still belongs into the range of the
2644 array, the index may wrap and still stay within the range of the array
2645 (consider e.g. if the array is indexed by the full range of
2646 unsigned char).
2648 To make things simpler, we require both bounds to fit into type, although
2649 there are cases where this would not be strictly necessary. */
2650 if (!int_fits_type_p (high, type)
2651 || !int_fits_type_p (low, type))
2652 return true;
2653 low = fold_convert (type, low);
2654 high = fold_convert (type, high);
2656 if (sign)
2657 next = fold_binary (PLUS_EXPR, type, low, step);
2658 else
2659 next = fold_binary (PLUS_EXPR, type, high, step);
2661 if (tree_int_cst_compare (low, next) <= 0
2662 && tree_int_cst_compare (next, high) <= 0)
2663 return true;
2665 record_nonwrapping_iv (loop, init, step, data->stmt, low, high, true, upper);
2666 return true;
2669 /* Determine information about number of iterations a LOOP from the bounds
2670 of arrays in the data reference REF accessed in STMT. RELIABLE is true if
2671 STMT is guaranteed to be executed in every iteration of LOOP.*/
2673 static void
2674 infer_loop_bounds_from_ref (struct loop *loop, gimple stmt, tree ref,
2675 bool reliable)
2677 struct ilb_data data;
2679 data.loop = loop;
2680 data.stmt = stmt;
2681 data.reliable = reliable;
2682 for_each_index (&ref, idx_infer_loop_bounds, &data);
2685 /* Determine information about number of iterations of a LOOP from the way
2686 arrays are used in STMT. RELIABLE is true if STMT is guaranteed to be
2687 executed in every iteration of LOOP. */
2689 static void
2690 infer_loop_bounds_from_array (struct loop *loop, gimple stmt, bool reliable)
2692 if (is_gimple_assign (stmt))
2694 tree op0 = gimple_assign_lhs (stmt);
2695 tree op1 = gimple_assign_rhs1 (stmt);
2697 /* For each memory access, analyze its access function
2698 and record a bound on the loop iteration domain. */
2699 if (REFERENCE_CLASS_P (op0))
2700 infer_loop_bounds_from_ref (loop, stmt, op0, reliable);
2702 if (REFERENCE_CLASS_P (op1))
2703 infer_loop_bounds_from_ref (loop, stmt, op1, reliable);
2705 else if (is_gimple_call (stmt))
2707 tree arg, lhs;
2708 unsigned i, n = gimple_call_num_args (stmt);
2710 lhs = gimple_call_lhs (stmt);
2711 if (lhs && REFERENCE_CLASS_P (lhs))
2712 infer_loop_bounds_from_ref (loop, stmt, lhs, reliable);
2714 for (i = 0; i < n; i++)
2716 arg = gimple_call_arg (stmt, i);
2717 if (REFERENCE_CLASS_P (arg))
2718 infer_loop_bounds_from_ref (loop, stmt, arg, reliable);
2723 /* Determine information about number of iterations of a LOOP from the fact
2724 that signed arithmetics in STMT does not overflow. */
2726 static void
2727 infer_loop_bounds_from_signedness (struct loop *loop, gimple stmt)
2729 tree def, base, step, scev, type, low, high;
2731 if (gimple_code (stmt) != GIMPLE_ASSIGN)
2732 return;
2734 def = gimple_assign_lhs (stmt);
2736 if (TREE_CODE (def) != SSA_NAME)
2737 return;
2739 type = TREE_TYPE (def);
2740 if (!INTEGRAL_TYPE_P (type)
2741 || !TYPE_OVERFLOW_UNDEFINED (type))
2742 return;
2744 scev = instantiate_parameters (loop, analyze_scalar_evolution (loop, def));
2745 if (chrec_contains_undetermined (scev))
2746 return;
2748 base = initial_condition_in_loop_num (scev, loop->num);
2749 step = evolution_part_in_loop_num (scev, loop->num);
2751 if (!base || !step
2752 || TREE_CODE (step) != INTEGER_CST
2753 || tree_contains_chrecs (base, NULL)
2754 || chrec_contains_symbols_defined_in_loop (base, loop->num))
2755 return;
2757 low = lower_bound_in_type (type, type);
2758 high = upper_bound_in_type (type, type);
2760 record_nonwrapping_iv (loop, base, step, stmt, low, high, false, true);
2763 /* The following analyzers are extracting informations on the bounds
2764 of LOOP from the following undefined behaviors:
2766 - data references should not access elements over the statically
2767 allocated size,
2769 - signed variables should not overflow when flag_wrapv is not set.
2772 static void
2773 infer_loop_bounds_from_undefined (struct loop *loop)
2775 unsigned i;
2776 basic_block *bbs;
2777 gimple_stmt_iterator bsi;
2778 basic_block bb;
2779 bool reliable;
2781 bbs = get_loop_body (loop);
2783 for (i = 0; i < loop->num_nodes; i++)
2785 bb = bbs[i];
2787 /* If BB is not executed in each iteration of the loop, we cannot
2788 use the operations in it to infer reliable upper bound on the
2789 # of iterations of the loop. However, we can use it as a guess. */
2790 reliable = dominated_by_p (CDI_DOMINATORS, loop->latch, bb);
2792 for (bsi = gsi_start_bb (bb); !gsi_end_p (bsi); gsi_next (&bsi))
2794 gimple stmt = gsi_stmt (bsi);
2796 infer_loop_bounds_from_array (loop, stmt, reliable);
2798 if (reliable)
2799 infer_loop_bounds_from_signedness (loop, stmt);
2804 free (bbs);
2807 /* Converts VAL to double_int. */
2809 static double_int
2810 gcov_type_to_double_int (gcov_type val)
2812 double_int ret;
2814 ret.low = (unsigned HOST_WIDE_INT) val;
2815 /* If HOST_BITS_PER_WIDE_INT == HOST_BITS_PER_WIDEST_INT, avoid shifting by
2816 the size of type. */
2817 val >>= HOST_BITS_PER_WIDE_INT - 1;
2818 val >>= 1;
2819 ret.high = (unsigned HOST_WIDE_INT) val;
2821 return ret;
2824 /* Records estimates on numbers of iterations of LOOP. */
2826 void
2827 estimate_numbers_of_iterations_loop (struct loop *loop)
2829 VEC (edge, heap) *exits;
2830 tree niter, type;
2831 unsigned i;
2832 struct tree_niter_desc niter_desc;
2833 edge ex;
2834 double_int bound;
2836 /* Give up if we already have tried to compute an estimation. */
2837 if (loop->estimate_state != EST_NOT_COMPUTED)
2838 return;
2839 loop->estimate_state = EST_AVAILABLE;
2840 loop->any_upper_bound = false;
2841 loop->any_estimate = false;
2843 exits = get_loop_exit_edges (loop);
2844 for (i = 0; VEC_iterate (edge, exits, i, ex); i++)
2846 if (!number_of_iterations_exit (loop, ex, &niter_desc, false))
2847 continue;
2849 niter = niter_desc.niter;
2850 type = TREE_TYPE (niter);
2851 if (TREE_CODE (niter_desc.may_be_zero) != INTEGER_CST)
2852 niter = build3 (COND_EXPR, type, niter_desc.may_be_zero,
2853 build_int_cst (type, 0),
2854 niter);
2855 record_estimate (loop, niter, niter_desc.max,
2856 last_stmt (ex->src),
2857 true, true, true);
2859 VEC_free (edge, heap, exits);
2861 infer_loop_bounds_from_undefined (loop);
2863 /* If we have a measured profile, use it to estimate the number of
2864 iterations. */
2865 if (loop->header->count != 0)
2867 gcov_type nit = expected_loop_iterations_unbounded (loop) + 1;
2868 bound = gcov_type_to_double_int (nit);
2869 record_niter_bound (loop, bound, true, false);
2872 /* If an upper bound is smaller than the realistic estimate of the
2873 number of iterations, use the upper bound instead. */
2874 if (loop->any_upper_bound
2875 && loop->any_estimate
2876 && double_int_ucmp (loop->nb_iterations_upper_bound,
2877 loop->nb_iterations_estimate) < 0)
2878 loop->nb_iterations_estimate = loop->nb_iterations_upper_bound;
2881 /* Records estimates on numbers of iterations of loops. */
2883 void
2884 estimate_numbers_of_iterations (void)
2886 loop_iterator li;
2887 struct loop *loop;
2889 /* We don't want to issue signed overflow warnings while getting
2890 loop iteration estimates. */
2891 fold_defer_overflow_warnings ();
2893 FOR_EACH_LOOP (li, loop, 0)
2895 estimate_numbers_of_iterations_loop (loop);
2898 fold_undefer_and_ignore_overflow_warnings ();
2901 /* Returns true if statement S1 dominates statement S2. */
2903 bool
2904 stmt_dominates_stmt_p (gimple s1, gimple s2)
2906 basic_block bb1 = gimple_bb (s1), bb2 = gimple_bb (s2);
2908 if (!bb1
2909 || s1 == s2)
2910 return true;
2912 if (bb1 == bb2)
2914 gimple_stmt_iterator bsi;
2916 if (gimple_code (s2) == GIMPLE_PHI)
2917 return false;
2919 if (gimple_code (s1) == GIMPLE_PHI)
2920 return true;
2922 for (bsi = gsi_start_bb (bb1); gsi_stmt (bsi) != s2; gsi_next (&bsi))
2923 if (gsi_stmt (bsi) == s1)
2924 return true;
2926 return false;
2929 return dominated_by_p (CDI_DOMINATORS, bb2, bb1);
2932 /* Returns true when we can prove that the number of executions of
2933 STMT in the loop is at most NITER, according to the bound on
2934 the number of executions of the statement NITER_BOUND->stmt recorded in
2935 NITER_BOUND. If STMT is NULL, we must prove this bound for all
2936 statements in the loop. */
2938 static bool
2939 n_of_executions_at_most (gimple stmt,
2940 struct nb_iter_bound *niter_bound,
2941 tree niter)
2943 double_int bound = niter_bound->bound;
2944 tree nit_type = TREE_TYPE (niter), e;
2945 enum tree_code cmp;
2947 gcc_assert (TYPE_UNSIGNED (nit_type));
2949 /* If the bound does not even fit into NIT_TYPE, it cannot tell us that
2950 the number of iterations is small. */
2951 if (!double_int_fits_to_tree_p (nit_type, bound))
2952 return false;
2954 /* We know that NITER_BOUND->stmt is executed at most NITER_BOUND->bound + 1
2955 times. This means that:
2957 -- if NITER_BOUND->is_exit is true, then everything before
2958 NITER_BOUND->stmt is executed at most NITER_BOUND->bound + 1
2959 times, and everything after it at most NITER_BOUND->bound times.
2961 -- If NITER_BOUND->is_exit is false, then if we can prove that when STMT
2962 is executed, then NITER_BOUND->stmt is executed as well in the same
2963 iteration (we conclude that if both statements belong to the same
2964 basic block, or if STMT is after NITER_BOUND->stmt), then STMT
2965 is executed at most NITER_BOUND->bound + 1 times. Otherwise STMT is
2966 executed at most NITER_BOUND->bound + 2 times. */
2968 if (niter_bound->is_exit)
2970 if (stmt
2971 && stmt != niter_bound->stmt
2972 && stmt_dominates_stmt_p (niter_bound->stmt, stmt))
2973 cmp = GE_EXPR;
2974 else
2975 cmp = GT_EXPR;
2977 else
2979 if (!stmt
2980 || (gimple_bb (stmt) != gimple_bb (niter_bound->stmt)
2981 && !stmt_dominates_stmt_p (niter_bound->stmt, stmt)))
2983 bound = double_int_add (bound, double_int_one);
2984 if (double_int_zero_p (bound)
2985 || !double_int_fits_to_tree_p (nit_type, bound))
2986 return false;
2988 cmp = GT_EXPR;
2991 e = fold_binary (cmp, boolean_type_node,
2992 niter, double_int_to_tree (nit_type, bound));
2993 return e && integer_nonzerop (e);
2996 /* Returns true if the arithmetics in TYPE can be assumed not to wrap. */
2998 bool
2999 nowrap_type_p (tree type)
3001 if (INTEGRAL_TYPE_P (type)
3002 && TYPE_OVERFLOW_UNDEFINED (type))
3003 return true;
3005 if (POINTER_TYPE_P (type))
3006 return true;
3008 return false;
3011 /* Return false only when the induction variable BASE + STEP * I is
3012 known to not overflow: i.e. when the number of iterations is small
3013 enough with respect to the step and initial condition in order to
3014 keep the evolution confined in TYPEs bounds. Return true when the
3015 iv is known to overflow or when the property is not computable.
3017 USE_OVERFLOW_SEMANTICS is true if this function should assume that
3018 the rules for overflow of the given language apply (e.g., that signed
3019 arithmetics in C does not overflow). */
3021 bool
3022 scev_probably_wraps_p (tree base, tree step,
3023 gimple at_stmt, struct loop *loop,
3024 bool use_overflow_semantics)
3026 struct nb_iter_bound *bound;
3027 tree delta, step_abs;
3028 tree unsigned_type, valid_niter;
3029 tree type = TREE_TYPE (step);
3031 /* FIXME: We really need something like
3032 http://gcc.gnu.org/ml/gcc-patches/2005-06/msg02025.html.
3034 We used to test for the following situation that frequently appears
3035 during address arithmetics:
3037 D.1621_13 = (long unsigned intD.4) D.1620_12;
3038 D.1622_14 = D.1621_13 * 8;
3039 D.1623_15 = (doubleD.29 *) D.1622_14;
3041 And derived that the sequence corresponding to D_14
3042 can be proved to not wrap because it is used for computing a
3043 memory access; however, this is not really the case -- for example,
3044 if D_12 = (unsigned char) [254,+,1], then D_14 has values
3045 2032, 2040, 0, 8, ..., but the code is still legal. */
3047 if (chrec_contains_undetermined (base)
3048 || chrec_contains_undetermined (step))
3049 return true;
3051 if (integer_zerop (step))
3052 return false;
3054 /* If we can use the fact that signed and pointer arithmetics does not
3055 wrap, we are done. */
3056 if (use_overflow_semantics && nowrap_type_p (TREE_TYPE (base)))
3057 return false;
3059 /* To be able to use estimates on number of iterations of the loop,
3060 we must have an upper bound on the absolute value of the step. */
3061 if (TREE_CODE (step) != INTEGER_CST)
3062 return true;
3064 /* Don't issue signed overflow warnings. */
3065 fold_defer_overflow_warnings ();
3067 /* Otherwise, compute the number of iterations before we reach the
3068 bound of the type, and verify that the loop is exited before this
3069 occurs. */
3070 unsigned_type = unsigned_type_for (type);
3071 base = fold_convert (unsigned_type, base);
3073 if (tree_int_cst_sign_bit (step))
3075 tree extreme = fold_convert (unsigned_type,
3076 lower_bound_in_type (type, type));
3077 delta = fold_build2 (MINUS_EXPR, unsigned_type, base, extreme);
3078 step_abs = fold_build1 (NEGATE_EXPR, unsigned_type,
3079 fold_convert (unsigned_type, step));
3081 else
3083 tree extreme = fold_convert (unsigned_type,
3084 upper_bound_in_type (type, type));
3085 delta = fold_build2 (MINUS_EXPR, unsigned_type, extreme, base);
3086 step_abs = fold_convert (unsigned_type, step);
3089 valid_niter = fold_build2 (FLOOR_DIV_EXPR, unsigned_type, delta, step_abs);
3091 estimate_numbers_of_iterations_loop (loop);
3092 for (bound = loop->bounds; bound; bound = bound->next)
3094 if (n_of_executions_at_most (at_stmt, bound, valid_niter))
3096 fold_undefer_and_ignore_overflow_warnings ();
3097 return false;
3101 fold_undefer_and_ignore_overflow_warnings ();
3103 /* At this point we still don't have a proof that the iv does not
3104 overflow: give up. */
3105 return true;
3108 /* Frees the information on upper bounds on numbers of iterations of LOOP. */
3110 void
3111 free_numbers_of_iterations_estimates_loop (struct loop *loop)
3113 struct nb_iter_bound *bound, *next;
3115 loop->nb_iterations = NULL;
3116 loop->estimate_state = EST_NOT_COMPUTED;
3117 for (bound = loop->bounds; bound; bound = next)
3119 next = bound->next;
3120 ggc_free (bound);
3123 loop->bounds = NULL;
3126 /* Frees the information on upper bounds on numbers of iterations of loops. */
3128 void
3129 free_numbers_of_iterations_estimates (void)
3131 loop_iterator li;
3132 struct loop *loop;
3134 FOR_EACH_LOOP (li, loop, 0)
3136 free_numbers_of_iterations_estimates_loop (loop);
3140 /* Substitute value VAL for ssa name NAME inside expressions held
3141 at LOOP. */
3143 void
3144 substitute_in_loop_info (struct loop *loop, tree name, tree val)
3146 loop->nb_iterations = simplify_replace_tree (loop->nb_iterations, name, val);