Remove assert in get_def_bb_for_const
[official-gcc.git] / gcc / tree-scalar-evolution.c
blobd6f2a2fba1403b80cf5defda8f705f3f1b1d7a1a
1 /* Scalar evolution detector.
2 Copyright (C) 2003-2016 Free Software Foundation, Inc.
3 Contributed by Sebastian Pop <s.pop@laposte.net>
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 3, or (at your option) any later
10 version.
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 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/>. */
22 Description:
24 This pass analyzes the evolution of scalar variables in loop
25 structures. The algorithm is based on the SSA representation,
26 and on the loop hierarchy tree. This algorithm is not based on
27 the notion of versions of a variable, as it was the case for the
28 previous implementations of the scalar evolution algorithm, but
29 it assumes that each defined name is unique.
31 The notation used in this file is called "chains of recurrences",
32 and has been proposed by Eugene Zima, Robert Van Engelen, and
33 others for describing induction variables in programs. For example
34 "b -> {0, +, 2}_1" means that the scalar variable "b" is equal to 0
35 when entering in the loop_1 and has a step 2 in this loop, in other
36 words "for (b = 0; b < N; b+=2);". Note that the coefficients of
37 this chain of recurrence (or chrec [shrek]) can contain the name of
38 other variables, in which case they are called parametric chrecs.
39 For example, "b -> {a, +, 2}_1" means that the initial value of "b"
40 is the value of "a". In most of the cases these parametric chrecs
41 are fully instantiated before their use because symbolic names can
42 hide some difficult cases such as self-references described later
43 (see the Fibonacci example).
45 A short sketch of the algorithm is:
47 Given a scalar variable to be analyzed, follow the SSA edge to
48 its definition:
50 - When the definition is a GIMPLE_ASSIGN: if the right hand side
51 (RHS) of the definition cannot be statically analyzed, the answer
52 of the analyzer is: "don't know".
53 Otherwise, for all the variables that are not yet analyzed in the
54 RHS, try to determine their evolution, and finally try to
55 evaluate the operation of the RHS that gives the evolution
56 function of the analyzed variable.
58 - When the definition is a condition-phi-node: determine the
59 evolution function for all the branches of the phi node, and
60 finally merge these evolutions (see chrec_merge).
62 - When the definition is a loop-phi-node: determine its initial
63 condition, that is the SSA edge defined in an outer loop, and
64 keep it symbolic. Then determine the SSA edges that are defined
65 in the body of the loop. Follow the inner edges until ending on
66 another loop-phi-node of the same analyzed loop. If the reached
67 loop-phi-node is not the starting loop-phi-node, then we keep
68 this definition under a symbolic form. If the reached
69 loop-phi-node is the same as the starting one, then we compute a
70 symbolic stride on the return path. The result is then the
71 symbolic chrec {initial_condition, +, symbolic_stride}_loop.
73 Examples:
75 Example 1: Illustration of the basic algorithm.
77 | a = 3
78 | loop_1
79 | b = phi (a, c)
80 | c = b + 1
81 | if (c > 10) exit_loop
82 | endloop
84 Suppose that we want to know the number of iterations of the
85 loop_1. The exit_loop is controlled by a COND_EXPR (c > 10). We
86 ask the scalar evolution analyzer two questions: what's the
87 scalar evolution (scev) of "c", and what's the scev of "10". For
88 "10" the answer is "10" since it is a scalar constant. For the
89 scalar variable "c", it follows the SSA edge to its definition,
90 "c = b + 1", and then asks again what's the scev of "b".
91 Following the SSA edge, we end on a loop-phi-node "b = phi (a,
92 c)", where the initial condition is "a", and the inner loop edge
93 is "c". The initial condition is kept under a symbolic form (it
94 may be the case that the copy constant propagation has done its
95 work and we end with the constant "3" as one of the edges of the
96 loop-phi-node). The update edge is followed to the end of the
97 loop, and until reaching again the starting loop-phi-node: b -> c
98 -> b. At this point we have drawn a path from "b" to "b" from
99 which we compute the stride in the loop: in this example it is
100 "+1". The resulting scev for "b" is "b -> {a, +, 1}_1". Now
101 that the scev for "b" is known, it is possible to compute the
102 scev for "c", that is "c -> {a + 1, +, 1}_1". In order to
103 determine the number of iterations in the loop_1, we have to
104 instantiate_parameters (loop_1, {a + 1, +, 1}_1), that gives after some
105 more analysis the scev {4, +, 1}_1, or in other words, this is
106 the function "f (x) = x + 4", where x is the iteration count of
107 the loop_1. Now we have to solve the inequality "x + 4 > 10",
108 and take the smallest iteration number for which the loop is
109 exited: x = 7. This loop runs from x = 0 to x = 7, and in total
110 there are 8 iterations. In terms of loop normalization, we have
111 created a variable that is implicitly defined, "x" or just "_1",
112 and all the other analyzed scalars of the loop are defined in
113 function of this variable:
115 a -> 3
116 b -> {3, +, 1}_1
117 c -> {4, +, 1}_1
119 or in terms of a C program:
121 | a = 3
122 | for (x = 0; x <= 7; x++)
124 | b = x + 3
125 | c = x + 4
128 Example 2a: Illustration of the algorithm on nested loops.
130 | loop_1
131 | a = phi (1, b)
132 | c = a + 2
133 | loop_2 10 times
134 | b = phi (c, d)
135 | d = b + 3
136 | endloop
137 | endloop
139 For analyzing the scalar evolution of "a", the algorithm follows
140 the SSA edge into the loop's body: "a -> b". "b" is an inner
141 loop-phi-node, and its analysis as in Example 1, gives:
143 b -> {c, +, 3}_2
144 d -> {c + 3, +, 3}_2
146 Following the SSA edge for the initial condition, we end on "c = a
147 + 2", and then on the starting loop-phi-node "a". From this point,
148 the loop stride is computed: back on "c = a + 2" we get a "+2" in
149 the loop_1, then on the loop-phi-node "b" we compute the overall
150 effect of the inner loop that is "b = c + 30", and we get a "+30"
151 in the loop_1. That means that the overall stride in loop_1 is
152 equal to "+32", and the result is:
154 a -> {1, +, 32}_1
155 c -> {3, +, 32}_1
157 Example 2b: Multivariate chains of recurrences.
159 | loop_1
160 | k = phi (0, k + 1)
161 | loop_2 4 times
162 | j = phi (0, j + 1)
163 | loop_3 4 times
164 | i = phi (0, i + 1)
165 | A[j + k] = ...
166 | endloop
167 | endloop
168 | endloop
170 Analyzing the access function of array A with
171 instantiate_parameters (loop_1, "j + k"), we obtain the
172 instantiation and the analysis of the scalar variables "j" and "k"
173 in loop_1. This leads to the scalar evolution {4, +, 1}_1: the end
174 value of loop_2 for "j" is 4, and the evolution of "k" in loop_1 is
175 {0, +, 1}_1. To obtain the evolution function in loop_3 and
176 instantiate the scalar variables up to loop_1, one has to use:
177 instantiate_scev (block_before_loop (loop_1), loop_3, "j + k").
178 The result of this call is {{0, +, 1}_1, +, 1}_2.
180 Example 3: Higher degree polynomials.
182 | loop_1
183 | a = phi (2, b)
184 | c = phi (5, d)
185 | b = a + 1
186 | d = c + a
187 | endloop
189 a -> {2, +, 1}_1
190 b -> {3, +, 1}_1
191 c -> {5, +, a}_1
192 d -> {5 + a, +, a}_1
194 instantiate_parameters (loop_1, {5, +, a}_1) -> {5, +, 2, +, 1}_1
195 instantiate_parameters (loop_1, {5 + a, +, a}_1) -> {7, +, 3, +, 1}_1
197 Example 4: Lucas, Fibonacci, or mixers in general.
199 | loop_1
200 | a = phi (1, b)
201 | c = phi (3, d)
202 | b = c
203 | d = c + a
204 | endloop
206 a -> (1, c)_1
207 c -> {3, +, a}_1
209 The syntax "(1, c)_1" stands for a PEELED_CHREC that has the
210 following semantics: during the first iteration of the loop_1, the
211 variable contains the value 1, and then it contains the value "c".
212 Note that this syntax is close to the syntax of the loop-phi-node:
213 "a -> (1, c)_1" vs. "a = phi (1, c)".
215 The symbolic chrec representation contains all the semantics of the
216 original code. What is more difficult is to use this information.
218 Example 5: Flip-flops, or exchangers.
220 | loop_1
221 | a = phi (1, b)
222 | c = phi (3, d)
223 | b = c
224 | d = a
225 | endloop
227 a -> (1, c)_1
228 c -> (3, a)_1
230 Based on these symbolic chrecs, it is possible to refine this
231 information into the more precise PERIODIC_CHRECs:
233 a -> |1, 3|_1
234 c -> |3, 1|_1
236 This transformation is not yet implemented.
238 Further readings:
240 You can find a more detailed description of the algorithm in:
241 http://icps.u-strasbg.fr/~pop/DEA_03_Pop.pdf
242 http://icps.u-strasbg.fr/~pop/DEA_03_Pop.ps.gz. But note that
243 this is a preliminary report and some of the details of the
244 algorithm have changed. I'm working on a research report that
245 updates the description of the algorithms to reflect the design
246 choices used in this implementation.
248 A set of slides show a high level overview of the algorithm and run
249 an example through the scalar evolution analyzer:
250 http://cri.ensmp.fr/~pop/gcc/mar04/slides.pdf
252 The slides that I have presented at the GCC Summit'04 are available
253 at: http://cri.ensmp.fr/~pop/gcc/20040604/gccsummit-lno-spop.pdf
256 #include "config.h"
257 #include "system.h"
258 #include "coretypes.h"
259 #include "backend.h"
260 #include "rtl.h"
261 #include "tree.h"
262 #include "gimple.h"
263 #include "ssa.h"
264 #include "gimple-pretty-print.h"
265 #include "fold-const.h"
266 #include "gimplify.h"
267 #include "gimple-iterator.h"
268 #include "gimplify-me.h"
269 #include "tree-cfg.h"
270 #include "tree-ssa-loop-ivopts.h"
271 #include "tree-ssa-loop-manip.h"
272 #include "tree-ssa-loop-niter.h"
273 #include "tree-ssa-loop.h"
274 #include "tree-ssa.h"
275 #include "cfgloop.h"
276 #include "tree-chrec.h"
277 #include "tree-affine.h"
278 #include "tree-scalar-evolution.h"
279 #include "dumpfile.h"
280 #include "params.h"
281 #include "tree-ssa-propagate.h"
282 #include "gimple-fold.h"
284 static tree analyze_scalar_evolution_1 (struct loop *, tree, tree);
285 static tree analyze_scalar_evolution_for_address_of (struct loop *loop,
286 tree var);
288 /* The cached information about an SSA name with version NAME_VERSION,
289 claiming that below basic block with index INSTANTIATED_BELOW, the
290 value of the SSA name can be expressed as CHREC. */
292 struct GTY((for_user)) scev_info_str {
293 unsigned int name_version;
294 int instantiated_below;
295 tree chrec;
298 /* Counters for the scev database. */
299 static unsigned nb_set_scev = 0;
300 static unsigned nb_get_scev = 0;
302 /* The following trees are unique elements. Thus the comparison of
303 another element to these elements should be done on the pointer to
304 these trees, and not on their value. */
306 /* The SSA_NAMEs that are not yet analyzed are qualified with NULL_TREE. */
307 tree chrec_not_analyzed_yet;
309 /* Reserved to the cases where the analyzer has detected an
310 undecidable property at compile time. */
311 tree chrec_dont_know;
313 /* When the analyzer has detected that a property will never
314 happen, then it qualifies it with chrec_known. */
315 tree chrec_known;
317 struct scev_info_hasher : ggc_ptr_hash<scev_info_str>
319 static hashval_t hash (scev_info_str *i);
320 static bool equal (const scev_info_str *a, const scev_info_str *b);
323 static GTY (()) hash_table<scev_info_hasher> *scalar_evolution_info;
326 /* Constructs a new SCEV_INFO_STR structure for VAR and INSTANTIATED_BELOW. */
328 static inline struct scev_info_str *
329 new_scev_info_str (basic_block instantiated_below, tree var)
331 struct scev_info_str *res;
333 res = ggc_alloc<scev_info_str> ();
334 res->name_version = SSA_NAME_VERSION (var);
335 res->chrec = chrec_not_analyzed_yet;
336 res->instantiated_below = instantiated_below->index;
338 return res;
341 /* Computes a hash function for database element ELT. */
343 hashval_t
344 scev_info_hasher::hash (scev_info_str *elt)
346 return elt->name_version ^ elt->instantiated_below;
349 /* Compares database elements E1 and E2. */
351 bool
352 scev_info_hasher::equal (const scev_info_str *elt1, const scev_info_str *elt2)
354 return (elt1->name_version == elt2->name_version
355 && elt1->instantiated_below == elt2->instantiated_below);
358 /* Get the scalar evolution of VAR for INSTANTIATED_BELOW basic block.
359 A first query on VAR returns chrec_not_analyzed_yet. */
361 static tree *
362 find_var_scev_info (basic_block instantiated_below, tree var)
364 struct scev_info_str *res;
365 struct scev_info_str tmp;
367 tmp.name_version = SSA_NAME_VERSION (var);
368 tmp.instantiated_below = instantiated_below->index;
369 scev_info_str **slot = scalar_evolution_info->find_slot (&tmp, INSERT);
371 if (!*slot)
372 *slot = new_scev_info_str (instantiated_below, var);
373 res = *slot;
375 return &res->chrec;
378 /* Return true when CHREC contains symbolic names defined in
379 LOOP_NB. */
381 bool
382 chrec_contains_symbols_defined_in_loop (const_tree chrec, unsigned loop_nb)
384 int i, n;
386 if (chrec == NULL_TREE)
387 return false;
389 if (is_gimple_min_invariant (chrec))
390 return false;
392 if (TREE_CODE (chrec) == SSA_NAME)
394 gimple *def;
395 loop_p def_loop, loop;
397 if (SSA_NAME_IS_DEFAULT_DEF (chrec))
398 return false;
400 def = SSA_NAME_DEF_STMT (chrec);
401 def_loop = loop_containing_stmt (def);
402 loop = get_loop (cfun, loop_nb);
404 if (def_loop == NULL)
405 return false;
407 if (loop == def_loop || flow_loop_nested_p (loop, def_loop))
408 return true;
410 return false;
413 n = TREE_OPERAND_LENGTH (chrec);
414 for (i = 0; i < n; i++)
415 if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec, i),
416 loop_nb))
417 return true;
418 return false;
421 /* Return true when PHI is a loop-phi-node. */
423 static bool
424 loop_phi_node_p (gimple *phi)
426 /* The implementation of this function is based on the following
427 property: "all the loop-phi-nodes of a loop are contained in the
428 loop's header basic block". */
430 return loop_containing_stmt (phi)->header == gimple_bb (phi);
433 /* Compute the scalar evolution for EVOLUTION_FN after crossing LOOP.
434 In general, in the case of multivariate evolutions we want to get
435 the evolution in different loops. LOOP specifies the level for
436 which to get the evolution.
438 Example:
440 | for (j = 0; j < 100; j++)
442 | for (k = 0; k < 100; k++)
444 | i = k + j; - Here the value of i is a function of j, k.
446 | ... = i - Here the value of i is a function of j.
448 | ... = i - Here the value of i is a scalar.
450 Example:
452 | i_0 = ...
453 | loop_1 10 times
454 | i_1 = phi (i_0, i_2)
455 | i_2 = i_1 + 2
456 | endloop
458 This loop has the same effect as:
459 LOOP_1 has the same effect as:
461 | i_1 = i_0 + 20
463 The overall effect of the loop, "i_0 + 20" in the previous example,
464 is obtained by passing in the parameters: LOOP = 1,
465 EVOLUTION_FN = {i_0, +, 2}_1.
468 tree
469 compute_overall_effect_of_inner_loop (struct loop *loop, tree evolution_fn)
471 bool val = false;
473 if (evolution_fn == chrec_dont_know)
474 return chrec_dont_know;
476 else if (TREE_CODE (evolution_fn) == POLYNOMIAL_CHREC)
478 struct loop *inner_loop = get_chrec_loop (evolution_fn);
480 if (inner_loop == loop
481 || flow_loop_nested_p (loop, inner_loop))
483 tree nb_iter = number_of_latch_executions (inner_loop);
485 if (nb_iter == chrec_dont_know)
486 return chrec_dont_know;
487 else
489 tree res;
491 /* evolution_fn is the evolution function in LOOP. Get
492 its value in the nb_iter-th iteration. */
493 res = chrec_apply (inner_loop->num, evolution_fn, nb_iter);
495 if (chrec_contains_symbols_defined_in_loop (res, loop->num))
496 res = instantiate_parameters (loop, res);
498 /* Continue the computation until ending on a parent of LOOP. */
499 return compute_overall_effect_of_inner_loop (loop, res);
502 else
503 return evolution_fn;
506 /* If the evolution function is an invariant, there is nothing to do. */
507 else if (no_evolution_in_loop_p (evolution_fn, loop->num, &val) && val)
508 return evolution_fn;
510 else
511 return chrec_dont_know;
514 /* Associate CHREC to SCALAR. */
516 static void
517 set_scalar_evolution (basic_block instantiated_below, tree scalar, tree chrec)
519 tree *scalar_info;
521 if (TREE_CODE (scalar) != SSA_NAME)
522 return;
524 scalar_info = find_var_scev_info (instantiated_below, scalar);
526 if (dump_file)
528 if (dump_flags & TDF_SCEV)
530 fprintf (dump_file, "(set_scalar_evolution \n");
531 fprintf (dump_file, " instantiated_below = %d \n",
532 instantiated_below->index);
533 fprintf (dump_file, " (scalar = ");
534 print_generic_expr (dump_file, scalar, 0);
535 fprintf (dump_file, ")\n (scalar_evolution = ");
536 print_generic_expr (dump_file, chrec, 0);
537 fprintf (dump_file, "))\n");
539 if (dump_flags & TDF_STATS)
540 nb_set_scev++;
543 *scalar_info = chrec;
546 /* Retrieve the chrec associated to SCALAR instantiated below
547 INSTANTIATED_BELOW block. */
549 static tree
550 get_scalar_evolution (basic_block instantiated_below, tree scalar)
552 tree res;
554 if (dump_file)
556 if (dump_flags & TDF_SCEV)
558 fprintf (dump_file, "(get_scalar_evolution \n");
559 fprintf (dump_file, " (scalar = ");
560 print_generic_expr (dump_file, scalar, 0);
561 fprintf (dump_file, ")\n");
563 if (dump_flags & TDF_STATS)
564 nb_get_scev++;
567 switch (TREE_CODE (scalar))
569 case SSA_NAME:
570 res = *find_var_scev_info (instantiated_below, scalar);
571 break;
573 case REAL_CST:
574 case FIXED_CST:
575 case INTEGER_CST:
576 res = scalar;
577 break;
579 default:
580 res = chrec_not_analyzed_yet;
581 break;
584 if (dump_file && (dump_flags & TDF_SCEV))
586 fprintf (dump_file, " (scalar_evolution = ");
587 print_generic_expr (dump_file, res, 0);
588 fprintf (dump_file, "))\n");
591 return res;
594 /* Helper function for add_to_evolution. Returns the evolution
595 function for an assignment of the form "a = b + c", where "a" and
596 "b" are on the strongly connected component. CHREC_BEFORE is the
597 information that we already have collected up to this point.
598 TO_ADD is the evolution of "c".
600 When CHREC_BEFORE has an evolution part in LOOP_NB, add to this
601 evolution the expression TO_ADD, otherwise construct an evolution
602 part for this loop. */
604 static tree
605 add_to_evolution_1 (unsigned loop_nb, tree chrec_before, tree to_add,
606 gimple *at_stmt)
608 tree type, left, right;
609 struct loop *loop = get_loop (cfun, loop_nb), *chloop;
611 switch (TREE_CODE (chrec_before))
613 case POLYNOMIAL_CHREC:
614 chloop = get_chrec_loop (chrec_before);
615 if (chloop == loop
616 || flow_loop_nested_p (chloop, loop))
618 unsigned var;
620 type = chrec_type (chrec_before);
622 /* When there is no evolution part in this loop, build it. */
623 if (chloop != loop)
625 var = loop_nb;
626 left = chrec_before;
627 right = SCALAR_FLOAT_TYPE_P (type)
628 ? build_real (type, dconst0)
629 : build_int_cst (type, 0);
631 else
633 var = CHREC_VARIABLE (chrec_before);
634 left = CHREC_LEFT (chrec_before);
635 right = CHREC_RIGHT (chrec_before);
638 to_add = chrec_convert (type, to_add, at_stmt);
639 right = chrec_convert_rhs (type, right, at_stmt);
640 right = chrec_fold_plus (chrec_type (right), right, to_add);
641 return build_polynomial_chrec (var, left, right);
643 else
645 gcc_assert (flow_loop_nested_p (loop, chloop));
647 /* Search the evolution in LOOP_NB. */
648 left = add_to_evolution_1 (loop_nb, CHREC_LEFT (chrec_before),
649 to_add, at_stmt);
650 right = CHREC_RIGHT (chrec_before);
651 right = chrec_convert_rhs (chrec_type (left), right, at_stmt);
652 return build_polynomial_chrec (CHREC_VARIABLE (chrec_before),
653 left, right);
656 default:
657 /* These nodes do not depend on a loop. */
658 if (chrec_before == chrec_dont_know)
659 return chrec_dont_know;
661 left = chrec_before;
662 right = chrec_convert_rhs (chrec_type (left), to_add, at_stmt);
663 return build_polynomial_chrec (loop_nb, left, right);
667 /* Add TO_ADD to the evolution part of CHREC_BEFORE in the dimension
668 of LOOP_NB.
670 Description (provided for completeness, for those who read code in
671 a plane, and for my poor 62 bytes brain that would have forgotten
672 all this in the next two or three months):
674 The algorithm of translation of programs from the SSA representation
675 into the chrecs syntax is based on a pattern matching. After having
676 reconstructed the overall tree expression for a loop, there are only
677 two cases that can arise:
679 1. a = loop-phi (init, a + expr)
680 2. a = loop-phi (init, expr)
682 where EXPR is either a scalar constant with respect to the analyzed
683 loop (this is a degree 0 polynomial), or an expression containing
684 other loop-phi definitions (these are higher degree polynomials).
686 Examples:
689 | init = ...
690 | loop_1
691 | a = phi (init, a + 5)
692 | endloop
695 | inita = ...
696 | initb = ...
697 | loop_1
698 | a = phi (inita, 2 * b + 3)
699 | b = phi (initb, b + 1)
700 | endloop
702 For the first case, the semantics of the SSA representation is:
704 | a (x) = init + \sum_{j = 0}^{x - 1} expr (j)
706 that is, there is a loop index "x" that determines the scalar value
707 of the variable during the loop execution. During the first
708 iteration, the value is that of the initial condition INIT, while
709 during the subsequent iterations, it is the sum of the initial
710 condition with the sum of all the values of EXPR from the initial
711 iteration to the before last considered iteration.
713 For the second case, the semantics of the SSA program is:
715 | a (x) = init, if x = 0;
716 | expr (x - 1), otherwise.
718 The second case corresponds to the PEELED_CHREC, whose syntax is
719 close to the syntax of a loop-phi-node:
721 | phi (init, expr) vs. (init, expr)_x
723 The proof of the translation algorithm for the first case is a
724 proof by structural induction based on the degree of EXPR.
726 Degree 0:
727 When EXPR is a constant with respect to the analyzed loop, or in
728 other words when EXPR is a polynomial of degree 0, the evolution of
729 the variable A in the loop is an affine function with an initial
730 condition INIT, and a step EXPR. In order to show this, we start
731 from the semantics of the SSA representation:
733 f (x) = init + \sum_{j = 0}^{x - 1} expr (j)
735 and since "expr (j)" is a constant with respect to "j",
737 f (x) = init + x * expr
739 Finally, based on the semantics of the pure sum chrecs, by
740 identification we get the corresponding chrecs syntax:
742 f (x) = init * \binom{x}{0} + expr * \binom{x}{1}
743 f (x) -> {init, +, expr}_x
745 Higher degree:
746 Suppose that EXPR is a polynomial of degree N with respect to the
747 analyzed loop_x for which we have already determined that it is
748 written under the chrecs syntax:
750 | expr (x) -> {b_0, +, b_1, +, ..., +, b_{n-1}} (x)
752 We start from the semantics of the SSA program:
754 | f (x) = init + \sum_{j = 0}^{x - 1} expr (j)
756 | f (x) = init + \sum_{j = 0}^{x - 1}
757 | (b_0 * \binom{j}{0} + ... + b_{n-1} * \binom{j}{n-1})
759 | f (x) = init + \sum_{j = 0}^{x - 1}
760 | \sum_{k = 0}^{n - 1} (b_k * \binom{j}{k})
762 | f (x) = init + \sum_{k = 0}^{n - 1}
763 | (b_k * \sum_{j = 0}^{x - 1} \binom{j}{k})
765 | f (x) = init + \sum_{k = 0}^{n - 1}
766 | (b_k * \binom{x}{k + 1})
768 | f (x) = init + b_0 * \binom{x}{1} + ...
769 | + b_{n-1} * \binom{x}{n}
771 | f (x) = init * \binom{x}{0} + b_0 * \binom{x}{1} + ...
772 | + b_{n-1} * \binom{x}{n}
775 And finally from the definition of the chrecs syntax, we identify:
776 | f (x) -> {init, +, b_0, +, ..., +, b_{n-1}}_x
778 This shows the mechanism that stands behind the add_to_evolution
779 function. An important point is that the use of symbolic
780 parameters avoids the need of an analysis schedule.
782 Example:
784 | inita = ...
785 | initb = ...
786 | loop_1
787 | a = phi (inita, a + 2 + b)
788 | b = phi (initb, b + 1)
789 | endloop
791 When analyzing "a", the algorithm keeps "b" symbolically:
793 | a -> {inita, +, 2 + b}_1
795 Then, after instantiation, the analyzer ends on the evolution:
797 | a -> {inita, +, 2 + initb, +, 1}_1
801 static tree
802 add_to_evolution (unsigned loop_nb, tree chrec_before, enum tree_code code,
803 tree to_add, gimple *at_stmt)
805 tree type = chrec_type (to_add);
806 tree res = NULL_TREE;
808 if (to_add == NULL_TREE)
809 return chrec_before;
811 /* TO_ADD is either a scalar, or a parameter. TO_ADD is not
812 instantiated at this point. */
813 if (TREE_CODE (to_add) == POLYNOMIAL_CHREC)
814 /* This should not happen. */
815 return chrec_dont_know;
817 if (dump_file && (dump_flags & TDF_SCEV))
819 fprintf (dump_file, "(add_to_evolution \n");
820 fprintf (dump_file, " (loop_nb = %d)\n", loop_nb);
821 fprintf (dump_file, " (chrec_before = ");
822 print_generic_expr (dump_file, chrec_before, 0);
823 fprintf (dump_file, ")\n (to_add = ");
824 print_generic_expr (dump_file, to_add, 0);
825 fprintf (dump_file, ")\n");
828 if (code == MINUS_EXPR)
829 to_add = chrec_fold_multiply (type, to_add, SCALAR_FLOAT_TYPE_P (type)
830 ? build_real (type, dconstm1)
831 : build_int_cst_type (type, -1));
833 res = add_to_evolution_1 (loop_nb, chrec_before, to_add, at_stmt);
835 if (dump_file && (dump_flags & TDF_SCEV))
837 fprintf (dump_file, " (res = ");
838 print_generic_expr (dump_file, res, 0);
839 fprintf (dump_file, "))\n");
842 return res;
847 /* This section selects the loops that will be good candidates for the
848 scalar evolution analysis. For the moment, greedily select all the
849 loop nests we could analyze. */
851 /* For a loop with a single exit edge, return the COND_EXPR that
852 guards the exit edge. If the expression is too difficult to
853 analyze, then give up. */
855 gcond *
856 get_loop_exit_condition (const struct loop *loop)
858 gcond *res = NULL;
859 edge exit_edge = single_exit (loop);
861 if (dump_file && (dump_flags & TDF_SCEV))
862 fprintf (dump_file, "(get_loop_exit_condition \n ");
864 if (exit_edge)
866 gimple *stmt;
868 stmt = last_stmt (exit_edge->src);
869 if (gcond *cond_stmt = dyn_cast <gcond *> (stmt))
870 res = cond_stmt;
873 if (dump_file && (dump_flags & TDF_SCEV))
875 print_gimple_stmt (dump_file, res, 0, 0);
876 fprintf (dump_file, ")\n");
879 return res;
883 /* Depth first search algorithm. */
885 enum t_bool {
886 t_false,
887 t_true,
888 t_dont_know
892 static t_bool follow_ssa_edge (struct loop *loop, gimple *, gphi *,
893 tree *, int);
895 /* Follow the ssa edge into the binary expression RHS0 CODE RHS1.
896 Return true if the strongly connected component has been found. */
898 static t_bool
899 follow_ssa_edge_binary (struct loop *loop, gimple *at_stmt,
900 tree type, tree rhs0, enum tree_code code, tree rhs1,
901 gphi *halting_phi, tree *evolution_of_loop,
902 int limit)
904 t_bool res = t_false;
905 tree evol;
907 switch (code)
909 case POINTER_PLUS_EXPR:
910 case PLUS_EXPR:
911 if (TREE_CODE (rhs0) == SSA_NAME)
913 if (TREE_CODE (rhs1) == SSA_NAME)
915 /* Match an assignment under the form:
916 "a = b + c". */
918 /* We want only assignments of form "name + name" contribute to
919 LIMIT, as the other cases do not necessarily contribute to
920 the complexity of the expression. */
921 limit++;
923 evol = *evolution_of_loop;
924 evol = add_to_evolution
925 (loop->num,
926 chrec_convert (type, evol, at_stmt),
927 code, rhs1, at_stmt);
928 res = follow_ssa_edge
929 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi, &evol, limit);
930 if (res == t_true)
931 *evolution_of_loop = evol;
932 else if (res == t_false)
934 *evolution_of_loop = add_to_evolution
935 (loop->num,
936 chrec_convert (type, *evolution_of_loop, at_stmt),
937 code, rhs0, at_stmt);
938 res = follow_ssa_edge
939 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
940 evolution_of_loop, limit);
941 if (res == t_true)
943 else if (res == t_dont_know)
944 *evolution_of_loop = chrec_dont_know;
947 else if (res == t_dont_know)
948 *evolution_of_loop = chrec_dont_know;
951 else
953 /* Match an assignment under the form:
954 "a = b + ...". */
955 *evolution_of_loop = add_to_evolution
956 (loop->num, chrec_convert (type, *evolution_of_loop,
957 at_stmt),
958 code, rhs1, at_stmt);
959 res = follow_ssa_edge
960 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
961 evolution_of_loop, limit);
962 if (res == t_true)
964 else if (res == t_dont_know)
965 *evolution_of_loop = chrec_dont_know;
969 else if (TREE_CODE (rhs1) == SSA_NAME)
971 /* Match an assignment under the form:
972 "a = ... + c". */
973 *evolution_of_loop = add_to_evolution
974 (loop->num, chrec_convert (type, *evolution_of_loop,
975 at_stmt),
976 code, rhs0, at_stmt);
977 res = follow_ssa_edge
978 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
979 evolution_of_loop, limit);
980 if (res == t_true)
982 else if (res == t_dont_know)
983 *evolution_of_loop = chrec_dont_know;
986 else
987 /* Otherwise, match an assignment under the form:
988 "a = ... + ...". */
989 /* And there is nothing to do. */
990 res = t_false;
991 break;
993 case MINUS_EXPR:
994 /* This case is under the form "opnd0 = rhs0 - rhs1". */
995 if (TREE_CODE (rhs0) == SSA_NAME)
997 /* Match an assignment under the form:
998 "a = b - ...". */
1000 /* We want only assignments of form "name - name" contribute to
1001 LIMIT, as the other cases do not necessarily contribute to
1002 the complexity of the expression. */
1003 if (TREE_CODE (rhs1) == SSA_NAME)
1004 limit++;
1006 *evolution_of_loop = add_to_evolution
1007 (loop->num, chrec_convert (type, *evolution_of_loop, at_stmt),
1008 MINUS_EXPR, rhs1, at_stmt);
1009 res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1010 evolution_of_loop, limit);
1011 if (res == t_true)
1013 else if (res == t_dont_know)
1014 *evolution_of_loop = chrec_dont_know;
1016 else
1017 /* Otherwise, match an assignment under the form:
1018 "a = ... - ...". */
1019 /* And there is nothing to do. */
1020 res = t_false;
1021 break;
1023 default:
1024 res = t_false;
1027 return res;
1030 /* Follow the ssa edge into the expression EXPR.
1031 Return true if the strongly connected component has been found. */
1033 static t_bool
1034 follow_ssa_edge_expr (struct loop *loop, gimple *at_stmt, tree expr,
1035 gphi *halting_phi, tree *evolution_of_loop,
1036 int limit)
1038 enum tree_code code = TREE_CODE (expr);
1039 tree type = TREE_TYPE (expr), rhs0, rhs1;
1040 t_bool res;
1042 /* The EXPR is one of the following cases:
1043 - an SSA_NAME,
1044 - an INTEGER_CST,
1045 - a PLUS_EXPR,
1046 - a POINTER_PLUS_EXPR,
1047 - a MINUS_EXPR,
1048 - an ASSERT_EXPR,
1049 - other cases are not yet handled. */
1051 switch (code)
1053 CASE_CONVERT:
1054 /* This assignment is under the form "a_1 = (cast) rhs. */
1055 res = follow_ssa_edge_expr (loop, at_stmt, TREE_OPERAND (expr, 0),
1056 halting_phi, evolution_of_loop, limit);
1057 *evolution_of_loop = chrec_convert (type, *evolution_of_loop, at_stmt);
1058 break;
1060 case INTEGER_CST:
1061 /* This assignment is under the form "a_1 = 7". */
1062 res = t_false;
1063 break;
1065 case SSA_NAME:
1066 /* This assignment is under the form: "a_1 = b_2". */
1067 res = follow_ssa_edge
1068 (loop, SSA_NAME_DEF_STMT (expr), halting_phi, evolution_of_loop, limit);
1069 break;
1071 case POINTER_PLUS_EXPR:
1072 case PLUS_EXPR:
1073 case MINUS_EXPR:
1074 /* This case is under the form "rhs0 +- rhs1". */
1075 rhs0 = TREE_OPERAND (expr, 0);
1076 rhs1 = TREE_OPERAND (expr, 1);
1077 type = TREE_TYPE (rhs0);
1078 STRIP_USELESS_TYPE_CONVERSION (rhs0);
1079 STRIP_USELESS_TYPE_CONVERSION (rhs1);
1080 res = follow_ssa_edge_binary (loop, at_stmt, type, rhs0, code, rhs1,
1081 halting_phi, evolution_of_loop, limit);
1082 break;
1084 case ADDR_EXPR:
1085 /* Handle &MEM[ptr + CST] which is equivalent to POINTER_PLUS_EXPR. */
1086 if (TREE_CODE (TREE_OPERAND (expr, 0)) == MEM_REF)
1088 expr = TREE_OPERAND (expr, 0);
1089 rhs0 = TREE_OPERAND (expr, 0);
1090 rhs1 = TREE_OPERAND (expr, 1);
1091 type = TREE_TYPE (rhs0);
1092 STRIP_USELESS_TYPE_CONVERSION (rhs0);
1093 STRIP_USELESS_TYPE_CONVERSION (rhs1);
1094 res = follow_ssa_edge_binary (loop, at_stmt, type,
1095 rhs0, POINTER_PLUS_EXPR, rhs1,
1096 halting_phi, evolution_of_loop, limit);
1098 else
1099 res = t_false;
1100 break;
1102 case ASSERT_EXPR:
1103 /* This assignment is of the form: "a_1 = ASSERT_EXPR <a_2, ...>"
1104 It must be handled as a copy assignment of the form a_1 = a_2. */
1105 rhs0 = ASSERT_EXPR_VAR (expr);
1106 if (TREE_CODE (rhs0) == SSA_NAME)
1107 res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (rhs0),
1108 halting_phi, evolution_of_loop, limit);
1109 else
1110 res = t_false;
1111 break;
1113 default:
1114 res = t_false;
1115 break;
1118 return res;
1121 /* Follow the ssa edge into the right hand side of an assignment STMT.
1122 Return true if the strongly connected component has been found. */
1124 static t_bool
1125 follow_ssa_edge_in_rhs (struct loop *loop, gimple *stmt,
1126 gphi *halting_phi, tree *evolution_of_loop,
1127 int limit)
1129 enum tree_code code = gimple_assign_rhs_code (stmt);
1130 tree type = gimple_expr_type (stmt), rhs1, rhs2;
1131 t_bool res;
1133 switch (code)
1135 CASE_CONVERT:
1136 /* This assignment is under the form "a_1 = (cast) rhs. */
1137 res = follow_ssa_edge_expr (loop, stmt, gimple_assign_rhs1 (stmt),
1138 halting_phi, evolution_of_loop, limit);
1139 *evolution_of_loop = chrec_convert (type, *evolution_of_loop, stmt);
1140 break;
1142 case POINTER_PLUS_EXPR:
1143 case PLUS_EXPR:
1144 case MINUS_EXPR:
1145 rhs1 = gimple_assign_rhs1 (stmt);
1146 rhs2 = gimple_assign_rhs2 (stmt);
1147 type = TREE_TYPE (rhs1);
1148 res = follow_ssa_edge_binary (loop, stmt, type, rhs1, code, rhs2,
1149 halting_phi, evolution_of_loop, limit);
1150 break;
1152 default:
1153 if (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS)
1154 res = follow_ssa_edge_expr (loop, stmt, gimple_assign_rhs1 (stmt),
1155 halting_phi, evolution_of_loop, limit);
1156 else
1157 res = t_false;
1158 break;
1161 return res;
1164 /* Checks whether the I-th argument of a PHI comes from a backedge. */
1166 static bool
1167 backedge_phi_arg_p (gphi *phi, int i)
1169 const_edge e = gimple_phi_arg_edge (phi, i);
1171 /* We would in fact like to test EDGE_DFS_BACK here, but we do not care
1172 about updating it anywhere, and this should work as well most of the
1173 time. */
1174 if (e->flags & EDGE_IRREDUCIBLE_LOOP)
1175 return true;
1177 return false;
1180 /* Helper function for one branch of the condition-phi-node. Return
1181 true if the strongly connected component has been found following
1182 this path. */
1184 static inline t_bool
1185 follow_ssa_edge_in_condition_phi_branch (int i,
1186 struct loop *loop,
1187 gphi *condition_phi,
1188 gphi *halting_phi,
1189 tree *evolution_of_branch,
1190 tree init_cond, int limit)
1192 tree branch = PHI_ARG_DEF (condition_phi, i);
1193 *evolution_of_branch = chrec_dont_know;
1195 /* Do not follow back edges (they must belong to an irreducible loop, which
1196 we really do not want to worry about). */
1197 if (backedge_phi_arg_p (condition_phi, i))
1198 return t_false;
1200 if (TREE_CODE (branch) == SSA_NAME)
1202 *evolution_of_branch = init_cond;
1203 return follow_ssa_edge (loop, SSA_NAME_DEF_STMT (branch), halting_phi,
1204 evolution_of_branch, limit);
1207 /* This case occurs when one of the condition branches sets
1208 the variable to a constant: i.e. a phi-node like
1209 "a_2 = PHI <a_7(5), 2(6)>;".
1211 FIXME: This case have to be refined correctly:
1212 in some cases it is possible to say something better than
1213 chrec_dont_know, for example using a wrap-around notation. */
1214 return t_false;
1217 /* This function merges the branches of a condition-phi-node in a
1218 loop. */
1220 static t_bool
1221 follow_ssa_edge_in_condition_phi (struct loop *loop,
1222 gphi *condition_phi,
1223 gphi *halting_phi,
1224 tree *evolution_of_loop, int limit)
1226 int i, n;
1227 tree init = *evolution_of_loop;
1228 tree evolution_of_branch;
1229 t_bool res = follow_ssa_edge_in_condition_phi_branch (0, loop, condition_phi,
1230 halting_phi,
1231 &evolution_of_branch,
1232 init, limit);
1233 if (res == t_false || res == t_dont_know)
1234 return res;
1236 *evolution_of_loop = evolution_of_branch;
1238 n = gimple_phi_num_args (condition_phi);
1239 for (i = 1; i < n; i++)
1241 /* Quickly give up when the evolution of one of the branches is
1242 not known. */
1243 if (*evolution_of_loop == chrec_dont_know)
1244 return t_true;
1246 /* Increase the limit by the PHI argument number to avoid exponential
1247 time and memory complexity. */
1248 res = follow_ssa_edge_in_condition_phi_branch (i, loop, condition_phi,
1249 halting_phi,
1250 &evolution_of_branch,
1251 init, limit + i);
1252 if (res == t_false || res == t_dont_know)
1253 return res;
1255 *evolution_of_loop = chrec_merge (*evolution_of_loop,
1256 evolution_of_branch);
1259 return t_true;
1262 /* Follow an SSA edge in an inner loop. It computes the overall
1263 effect of the loop, and following the symbolic initial conditions,
1264 it follows the edges in the parent loop. The inner loop is
1265 considered as a single statement. */
1267 static t_bool
1268 follow_ssa_edge_inner_loop_phi (struct loop *outer_loop,
1269 gphi *loop_phi_node,
1270 gphi *halting_phi,
1271 tree *evolution_of_loop, int limit)
1273 struct loop *loop = loop_containing_stmt (loop_phi_node);
1274 tree ev = analyze_scalar_evolution (loop, PHI_RESULT (loop_phi_node));
1276 /* Sometimes, the inner loop is too difficult to analyze, and the
1277 result of the analysis is a symbolic parameter. */
1278 if (ev == PHI_RESULT (loop_phi_node))
1280 t_bool res = t_false;
1281 int i, n = gimple_phi_num_args (loop_phi_node);
1283 for (i = 0; i < n; i++)
1285 tree arg = PHI_ARG_DEF (loop_phi_node, i);
1286 basic_block bb;
1288 /* Follow the edges that exit the inner loop. */
1289 bb = gimple_phi_arg_edge (loop_phi_node, i)->src;
1290 if (!flow_bb_inside_loop_p (loop, bb))
1291 res = follow_ssa_edge_expr (outer_loop, loop_phi_node,
1292 arg, halting_phi,
1293 evolution_of_loop, limit);
1294 if (res == t_true)
1295 break;
1298 /* If the path crosses this loop-phi, give up. */
1299 if (res == t_true)
1300 *evolution_of_loop = chrec_dont_know;
1302 return res;
1305 /* Otherwise, compute the overall effect of the inner loop. */
1306 ev = compute_overall_effect_of_inner_loop (loop, ev);
1307 return follow_ssa_edge_expr (outer_loop, loop_phi_node, ev, halting_phi,
1308 evolution_of_loop, limit);
1311 /* Follow an SSA edge from a loop-phi-node to itself, constructing a
1312 path that is analyzed on the return walk. */
1314 static t_bool
1315 follow_ssa_edge (struct loop *loop, gimple *def, gphi *halting_phi,
1316 tree *evolution_of_loop, int limit)
1318 struct loop *def_loop;
1320 if (gimple_nop_p (def))
1321 return t_false;
1323 /* Give up if the path is longer than the MAX that we allow. */
1324 if (limit > PARAM_VALUE (PARAM_SCEV_MAX_EXPR_COMPLEXITY))
1325 return t_dont_know;
1327 def_loop = loop_containing_stmt (def);
1329 switch (gimple_code (def))
1331 case GIMPLE_PHI:
1332 if (!loop_phi_node_p (def))
1333 /* DEF is a condition-phi-node. Follow the branches, and
1334 record their evolutions. Finally, merge the collected
1335 information and set the approximation to the main
1336 variable. */
1337 return follow_ssa_edge_in_condition_phi
1338 (loop, as_a <gphi *> (def), halting_phi, evolution_of_loop,
1339 limit);
1341 /* When the analyzed phi is the halting_phi, the
1342 depth-first search is over: we have found a path from
1343 the halting_phi to itself in the loop. */
1344 if (def == halting_phi)
1345 return t_true;
1347 /* Otherwise, the evolution of the HALTING_PHI depends
1348 on the evolution of another loop-phi-node, i.e. the
1349 evolution function is a higher degree polynomial. */
1350 if (def_loop == loop)
1351 return t_false;
1353 /* Inner loop. */
1354 if (flow_loop_nested_p (loop, def_loop))
1355 return follow_ssa_edge_inner_loop_phi
1356 (loop, as_a <gphi *> (def), halting_phi, evolution_of_loop,
1357 limit + 1);
1359 /* Outer loop. */
1360 return t_false;
1362 case GIMPLE_ASSIGN:
1363 return follow_ssa_edge_in_rhs (loop, def, halting_phi,
1364 evolution_of_loop, limit);
1366 default:
1367 /* At this level of abstraction, the program is just a set
1368 of GIMPLE_ASSIGNs and PHI_NODEs. In principle there is no
1369 other node to be handled. */
1370 return t_false;
1375 /* Simplify PEELED_CHREC represented by (init_cond, arg) in LOOP.
1376 Handle below case and return the corresponding POLYNOMIAL_CHREC:
1378 # i_17 = PHI <i_13(5), 0(3)>
1379 # _20 = PHI <_5(5), start_4(D)(3)>
1381 i_13 = i_17 + 1;
1382 _5 = start_4(D) + i_13;
1384 Though variable _20 appears as a PEELED_CHREC in the form of
1385 (start_4, _5)_LOOP, it's a POLYNOMIAL_CHREC like {start_4, 1}_LOOP.
1387 See PR41488. */
1389 static tree
1390 simplify_peeled_chrec (struct loop *loop, tree arg, tree init_cond)
1392 aff_tree aff1, aff2;
1393 tree ev, left, right, type, step_val;
1394 hash_map<tree, name_expansion *> *peeled_chrec_map = NULL;
1396 ev = instantiate_parameters (loop, analyze_scalar_evolution (loop, arg));
1397 if (ev == NULL_TREE || TREE_CODE (ev) != POLYNOMIAL_CHREC)
1398 return chrec_dont_know;
1400 left = CHREC_LEFT (ev);
1401 right = CHREC_RIGHT (ev);
1402 type = TREE_TYPE (left);
1403 step_val = chrec_fold_plus (type, init_cond, right);
1405 /* Transform (init, {left, right}_LOOP)_LOOP to {init, right}_LOOP
1406 if "left" equals to "init + right". */
1407 if (operand_equal_p (left, step_val, 0))
1409 if (dump_file && (dump_flags & TDF_SCEV))
1410 fprintf (dump_file, "Simplify PEELED_CHREC into POLYNOMIAL_CHREC.\n");
1412 return build_polynomial_chrec (loop->num, init_cond, right);
1415 /* Try harder to check if they are equal. */
1416 tree_to_aff_combination_expand (left, type, &aff1, &peeled_chrec_map);
1417 tree_to_aff_combination_expand (step_val, type, &aff2, &peeled_chrec_map);
1418 free_affine_expand_cache (&peeled_chrec_map);
1419 aff_combination_scale (&aff2, -1);
1420 aff_combination_add (&aff1, &aff2);
1422 /* Transform (init, {left, right}_LOOP)_LOOP to {init, right}_LOOP
1423 if "left" equals to "init + right". */
1424 if (aff_combination_zero_p (&aff1))
1426 if (dump_file && (dump_flags & TDF_SCEV))
1427 fprintf (dump_file, "Simplify PEELED_CHREC into POLYNOMIAL_CHREC.\n");
1429 return build_polynomial_chrec (loop->num, init_cond, right);
1431 return chrec_dont_know;
1434 /* Given a LOOP_PHI_NODE, this function determines the evolution
1435 function from LOOP_PHI_NODE to LOOP_PHI_NODE in the loop. */
1437 static tree
1438 analyze_evolution_in_loop (gphi *loop_phi_node,
1439 tree init_cond)
1441 int i, n = gimple_phi_num_args (loop_phi_node);
1442 tree evolution_function = chrec_not_analyzed_yet;
1443 struct loop *loop = loop_containing_stmt (loop_phi_node);
1444 basic_block bb;
1445 static bool simplify_peeled_chrec_p = true;
1447 if (dump_file && (dump_flags & TDF_SCEV))
1449 fprintf (dump_file, "(analyze_evolution_in_loop \n");
1450 fprintf (dump_file, " (loop_phi_node = ");
1451 print_gimple_stmt (dump_file, loop_phi_node, 0, 0);
1452 fprintf (dump_file, ")\n");
1455 for (i = 0; i < n; i++)
1457 tree arg = PHI_ARG_DEF (loop_phi_node, i);
1458 gimple *ssa_chain;
1459 tree ev_fn;
1460 t_bool res;
1462 /* Select the edges that enter the loop body. */
1463 bb = gimple_phi_arg_edge (loop_phi_node, i)->src;
1464 if (!flow_bb_inside_loop_p (loop, bb))
1465 continue;
1467 if (TREE_CODE (arg) == SSA_NAME)
1469 bool val = false;
1471 ssa_chain = SSA_NAME_DEF_STMT (arg);
1473 /* Pass in the initial condition to the follow edge function. */
1474 ev_fn = init_cond;
1475 res = follow_ssa_edge (loop, ssa_chain, loop_phi_node, &ev_fn, 0);
1477 /* If ev_fn has no evolution in the inner loop, and the
1478 init_cond is not equal to ev_fn, then we have an
1479 ambiguity between two possible values, as we cannot know
1480 the number of iterations at this point. */
1481 if (TREE_CODE (ev_fn) != POLYNOMIAL_CHREC
1482 && no_evolution_in_loop_p (ev_fn, loop->num, &val) && val
1483 && !operand_equal_p (init_cond, ev_fn, 0))
1484 ev_fn = chrec_dont_know;
1486 else
1487 res = t_false;
1489 /* When it is impossible to go back on the same
1490 loop_phi_node by following the ssa edges, the
1491 evolution is represented by a peeled chrec, i.e. the
1492 first iteration, EV_FN has the value INIT_COND, then
1493 all the other iterations it has the value of ARG.
1494 For the moment, PEELED_CHREC nodes are not built. */
1495 if (res != t_true)
1497 ev_fn = chrec_dont_know;
1498 /* Try to recognize POLYNOMIAL_CHREC which appears in
1499 the form of PEELED_CHREC, but guard the process with
1500 a bool variable to keep the analyzer from infinite
1501 recurrence for real PEELED_RECs. */
1502 if (simplify_peeled_chrec_p && TREE_CODE (arg) == SSA_NAME)
1504 simplify_peeled_chrec_p = false;
1505 ev_fn = simplify_peeled_chrec (loop, arg, init_cond);
1506 simplify_peeled_chrec_p = true;
1510 /* When there are multiple back edges of the loop (which in fact never
1511 happens currently, but nevertheless), merge their evolutions. */
1512 evolution_function = chrec_merge (evolution_function, ev_fn);
1514 if (evolution_function == chrec_dont_know)
1515 break;
1518 if (dump_file && (dump_flags & TDF_SCEV))
1520 fprintf (dump_file, " (evolution_function = ");
1521 print_generic_expr (dump_file, evolution_function, 0);
1522 fprintf (dump_file, "))\n");
1525 return evolution_function;
1528 /* Looks to see if VAR is a copy of a constant (via straightforward assignments
1529 or degenerate phi's). If so, returns the constant; else, returns VAR. */
1531 static tree
1532 follow_copies_to_constant (tree var)
1534 tree res = var;
1535 while (TREE_CODE (res) == SSA_NAME)
1537 gimple *def = SSA_NAME_DEF_STMT (res);
1538 if (gphi *phi = dyn_cast <gphi *> (def))
1540 if (tree rhs = degenerate_phi_result (phi))
1541 res = rhs;
1542 else
1543 break;
1545 else if (gimple_assign_single_p (def))
1546 /* Will exit loop if not an SSA_NAME. */
1547 res = gimple_assign_rhs1 (def);
1548 else
1549 break;
1551 if (CONSTANT_CLASS_P (res))
1552 return res;
1553 return var;
1556 /* Given a loop-phi-node, return the initial conditions of the
1557 variable on entry of the loop. When the CCP has propagated
1558 constants into the loop-phi-node, the initial condition is
1559 instantiated, otherwise the initial condition is kept symbolic.
1560 This analyzer does not analyze the evolution outside the current
1561 loop, and leaves this task to the on-demand tree reconstructor. */
1563 static tree
1564 analyze_initial_condition (gphi *loop_phi_node)
1566 int i, n;
1567 tree init_cond = chrec_not_analyzed_yet;
1568 struct loop *loop = loop_containing_stmt (loop_phi_node);
1570 if (dump_file && (dump_flags & TDF_SCEV))
1572 fprintf (dump_file, "(analyze_initial_condition \n");
1573 fprintf (dump_file, " (loop_phi_node = \n");
1574 print_gimple_stmt (dump_file, loop_phi_node, 0, 0);
1575 fprintf (dump_file, ")\n");
1578 n = gimple_phi_num_args (loop_phi_node);
1579 for (i = 0; i < n; i++)
1581 tree branch = PHI_ARG_DEF (loop_phi_node, i);
1582 basic_block bb = gimple_phi_arg_edge (loop_phi_node, i)->src;
1584 /* When the branch is oriented to the loop's body, it does
1585 not contribute to the initial condition. */
1586 if (flow_bb_inside_loop_p (loop, bb))
1587 continue;
1589 if (init_cond == chrec_not_analyzed_yet)
1591 init_cond = branch;
1592 continue;
1595 if (TREE_CODE (branch) == SSA_NAME)
1597 init_cond = chrec_dont_know;
1598 break;
1601 init_cond = chrec_merge (init_cond, branch);
1604 /* Ooops -- a loop without an entry??? */
1605 if (init_cond == chrec_not_analyzed_yet)
1606 init_cond = chrec_dont_know;
1608 /* We may not have fully constant propagated IL. Handle degenerate PHIs here
1609 to not miss important early loop unrollings. */
1610 init_cond = follow_copies_to_constant (init_cond);
1612 if (dump_file && (dump_flags & TDF_SCEV))
1614 fprintf (dump_file, " (init_cond = ");
1615 print_generic_expr (dump_file, init_cond, 0);
1616 fprintf (dump_file, "))\n");
1619 return init_cond;
1622 /* Analyze the scalar evolution for LOOP_PHI_NODE. */
1624 static tree
1625 interpret_loop_phi (struct loop *loop, gphi *loop_phi_node)
1627 tree res;
1628 struct loop *phi_loop = loop_containing_stmt (loop_phi_node);
1629 tree init_cond;
1631 if (phi_loop != loop)
1633 struct loop *subloop;
1634 tree evolution_fn = analyze_scalar_evolution
1635 (phi_loop, PHI_RESULT (loop_phi_node));
1637 /* Dive one level deeper. */
1638 subloop = superloop_at_depth (phi_loop, loop_depth (loop) + 1);
1640 /* Interpret the subloop. */
1641 res = compute_overall_effect_of_inner_loop (subloop, evolution_fn);
1642 return res;
1645 /* Otherwise really interpret the loop phi. */
1646 init_cond = analyze_initial_condition (loop_phi_node);
1647 res = analyze_evolution_in_loop (loop_phi_node, init_cond);
1649 /* Verify we maintained the correct initial condition throughout
1650 possible conversions in the SSA chain. */
1651 if (res != chrec_dont_know)
1653 tree new_init = res;
1654 if (CONVERT_EXPR_P (res)
1655 && TREE_CODE (TREE_OPERAND (res, 0)) == POLYNOMIAL_CHREC)
1656 new_init = fold_convert (TREE_TYPE (res),
1657 CHREC_LEFT (TREE_OPERAND (res, 0)));
1658 else if (TREE_CODE (res) == POLYNOMIAL_CHREC)
1659 new_init = CHREC_LEFT (res);
1660 STRIP_USELESS_TYPE_CONVERSION (new_init);
1661 if (TREE_CODE (new_init) == POLYNOMIAL_CHREC
1662 || !operand_equal_p (init_cond, new_init, 0))
1663 return chrec_dont_know;
1666 return res;
1669 /* This function merges the branches of a condition-phi-node,
1670 contained in the outermost loop, and whose arguments are already
1671 analyzed. */
1673 static tree
1674 interpret_condition_phi (struct loop *loop, gphi *condition_phi)
1676 int i, n = gimple_phi_num_args (condition_phi);
1677 tree res = chrec_not_analyzed_yet;
1679 for (i = 0; i < n; i++)
1681 tree branch_chrec;
1683 if (backedge_phi_arg_p (condition_phi, i))
1685 res = chrec_dont_know;
1686 break;
1689 branch_chrec = analyze_scalar_evolution
1690 (loop, PHI_ARG_DEF (condition_phi, i));
1692 res = chrec_merge (res, branch_chrec);
1693 if (res == chrec_dont_know)
1694 break;
1697 return res;
1700 /* Interpret the operation RHS1 OP RHS2. If we didn't
1701 analyze this node before, follow the definitions until ending
1702 either on an analyzed GIMPLE_ASSIGN, or on a loop-phi-node. On the
1703 return path, this function propagates evolutions (ala constant copy
1704 propagation). OPND1 is not a GIMPLE expression because we could
1705 analyze the effect of an inner loop: see interpret_loop_phi. */
1707 static tree
1708 interpret_rhs_expr (struct loop *loop, gimple *at_stmt,
1709 tree type, tree rhs1, enum tree_code code, tree rhs2)
1711 tree res, chrec1, chrec2, ctype;
1712 gimple *def;
1714 if (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS)
1716 if (is_gimple_min_invariant (rhs1))
1717 return chrec_convert (type, rhs1, at_stmt);
1719 if (code == SSA_NAME)
1720 return chrec_convert (type, analyze_scalar_evolution (loop, rhs1),
1721 at_stmt);
1723 if (code == ASSERT_EXPR)
1725 rhs1 = ASSERT_EXPR_VAR (rhs1);
1726 return chrec_convert (type, analyze_scalar_evolution (loop, rhs1),
1727 at_stmt);
1731 switch (code)
1733 case ADDR_EXPR:
1734 if (TREE_CODE (TREE_OPERAND (rhs1, 0)) == MEM_REF
1735 || handled_component_p (TREE_OPERAND (rhs1, 0)))
1737 machine_mode mode;
1738 HOST_WIDE_INT bitsize, bitpos;
1739 int unsignedp, reversep;
1740 int volatilep = 0;
1741 tree base, offset;
1742 tree chrec3;
1743 tree unitpos;
1745 base = get_inner_reference (TREE_OPERAND (rhs1, 0),
1746 &bitsize, &bitpos, &offset, &mode,
1747 &unsignedp, &reversep, &volatilep,
1748 false);
1750 if (TREE_CODE (base) == MEM_REF)
1752 rhs2 = TREE_OPERAND (base, 1);
1753 rhs1 = TREE_OPERAND (base, 0);
1755 chrec1 = analyze_scalar_evolution (loop, rhs1);
1756 chrec2 = analyze_scalar_evolution (loop, rhs2);
1757 chrec1 = chrec_convert (type, chrec1, at_stmt);
1758 chrec2 = chrec_convert (TREE_TYPE (rhs2), chrec2, at_stmt);
1759 chrec1 = instantiate_parameters (loop, chrec1);
1760 chrec2 = instantiate_parameters (loop, chrec2);
1761 res = chrec_fold_plus (type, chrec1, chrec2);
1763 else
1765 chrec1 = analyze_scalar_evolution_for_address_of (loop, base);
1766 chrec1 = chrec_convert (type, chrec1, at_stmt);
1767 res = chrec1;
1770 if (offset != NULL_TREE)
1772 chrec2 = analyze_scalar_evolution (loop, offset);
1773 chrec2 = chrec_convert (TREE_TYPE (offset), chrec2, at_stmt);
1774 chrec2 = instantiate_parameters (loop, chrec2);
1775 res = chrec_fold_plus (type, res, chrec2);
1778 if (bitpos != 0)
1780 gcc_assert ((bitpos % BITS_PER_UNIT) == 0);
1782 unitpos = size_int (bitpos / BITS_PER_UNIT);
1783 chrec3 = analyze_scalar_evolution (loop, unitpos);
1784 chrec3 = chrec_convert (TREE_TYPE (unitpos), chrec3, at_stmt);
1785 chrec3 = instantiate_parameters (loop, chrec3);
1786 res = chrec_fold_plus (type, res, chrec3);
1789 else
1790 res = chrec_dont_know;
1791 break;
1793 case POINTER_PLUS_EXPR:
1794 chrec1 = analyze_scalar_evolution (loop, rhs1);
1795 chrec2 = analyze_scalar_evolution (loop, rhs2);
1796 chrec1 = chrec_convert (type, chrec1, at_stmt);
1797 chrec2 = chrec_convert (TREE_TYPE (rhs2), chrec2, at_stmt);
1798 chrec1 = instantiate_parameters (loop, chrec1);
1799 chrec2 = instantiate_parameters (loop, chrec2);
1800 res = chrec_fold_plus (type, chrec1, chrec2);
1801 break;
1803 case PLUS_EXPR:
1804 chrec1 = analyze_scalar_evolution (loop, rhs1);
1805 chrec2 = analyze_scalar_evolution (loop, rhs2);
1806 ctype = type;
1807 /* When the stmt is conditionally executed re-write the CHREC
1808 into a form that has well-defined behavior on overflow. */
1809 if (at_stmt
1810 && INTEGRAL_TYPE_P (type)
1811 && ! TYPE_OVERFLOW_WRAPS (type)
1812 && ! dominated_by_p (CDI_DOMINATORS, loop->latch,
1813 gimple_bb (at_stmt)))
1814 ctype = unsigned_type_for (type);
1815 chrec1 = chrec_convert (ctype, chrec1, at_stmt);
1816 chrec2 = chrec_convert (ctype, chrec2, at_stmt);
1817 chrec1 = instantiate_parameters (loop, chrec1);
1818 chrec2 = instantiate_parameters (loop, chrec2);
1819 res = chrec_fold_plus (ctype, chrec1, chrec2);
1820 if (type != ctype)
1821 res = chrec_convert (type, res, at_stmt);
1822 break;
1824 case MINUS_EXPR:
1825 chrec1 = analyze_scalar_evolution (loop, rhs1);
1826 chrec2 = analyze_scalar_evolution (loop, rhs2);
1827 ctype = type;
1828 /* When the stmt is conditionally executed re-write the CHREC
1829 into a form that has well-defined behavior on overflow. */
1830 if (at_stmt
1831 && INTEGRAL_TYPE_P (type)
1832 && ! TYPE_OVERFLOW_WRAPS (type)
1833 && ! dominated_by_p (CDI_DOMINATORS,
1834 loop->latch, gimple_bb (at_stmt)))
1835 ctype = unsigned_type_for (type);
1836 chrec1 = chrec_convert (ctype, chrec1, at_stmt);
1837 chrec2 = chrec_convert (ctype, chrec2, at_stmt);
1838 chrec1 = instantiate_parameters (loop, chrec1);
1839 chrec2 = instantiate_parameters (loop, chrec2);
1840 res = chrec_fold_minus (ctype, chrec1, chrec2);
1841 if (type != ctype)
1842 res = chrec_convert (type, res, at_stmt);
1843 break;
1845 case NEGATE_EXPR:
1846 chrec1 = analyze_scalar_evolution (loop, rhs1);
1847 ctype = type;
1848 /* When the stmt is conditionally executed re-write the CHREC
1849 into a form that has well-defined behavior on overflow. */
1850 if (at_stmt
1851 && INTEGRAL_TYPE_P (type)
1852 && ! TYPE_OVERFLOW_WRAPS (type)
1853 && ! dominated_by_p (CDI_DOMINATORS,
1854 loop->latch, gimple_bb (at_stmt)))
1855 ctype = unsigned_type_for (type);
1856 chrec1 = chrec_convert (ctype, chrec1, at_stmt);
1857 /* TYPE may be integer, real or complex, so use fold_convert. */
1858 chrec1 = instantiate_parameters (loop, chrec1);
1859 res = chrec_fold_multiply (ctype, chrec1,
1860 fold_convert (ctype, integer_minus_one_node));
1861 if (type != ctype)
1862 res = chrec_convert (type, res, at_stmt);
1863 break;
1865 case BIT_NOT_EXPR:
1866 /* Handle ~X as -1 - X. */
1867 chrec1 = analyze_scalar_evolution (loop, rhs1);
1868 chrec1 = chrec_convert (type, chrec1, at_stmt);
1869 chrec1 = instantiate_parameters (loop, chrec1);
1870 res = chrec_fold_minus (type,
1871 fold_convert (type, integer_minus_one_node),
1872 chrec1);
1873 break;
1875 case MULT_EXPR:
1876 chrec1 = analyze_scalar_evolution (loop, rhs1);
1877 chrec2 = analyze_scalar_evolution (loop, rhs2);
1878 ctype = type;
1879 /* When the stmt is conditionally executed re-write the CHREC
1880 into a form that has well-defined behavior on overflow. */
1881 if (at_stmt
1882 && INTEGRAL_TYPE_P (type)
1883 && ! TYPE_OVERFLOW_WRAPS (type)
1884 && ! dominated_by_p (CDI_DOMINATORS,
1885 loop->latch, gimple_bb (at_stmt)))
1886 ctype = unsigned_type_for (type);
1887 chrec1 = chrec_convert (ctype, chrec1, at_stmt);
1888 chrec2 = chrec_convert (ctype, chrec2, at_stmt);
1889 chrec1 = instantiate_parameters (loop, chrec1);
1890 chrec2 = instantiate_parameters (loop, chrec2);
1891 res = chrec_fold_multiply (ctype, chrec1, chrec2);
1892 if (type != ctype)
1893 res = chrec_convert (type, res, at_stmt);
1894 break;
1896 case LSHIFT_EXPR:
1898 /* Handle A<<B as A * (1<<B). */
1899 tree uns = unsigned_type_for (type);
1900 chrec1 = analyze_scalar_evolution (loop, rhs1);
1901 chrec2 = analyze_scalar_evolution (loop, rhs2);
1902 chrec1 = chrec_convert (uns, chrec1, at_stmt);
1903 chrec1 = instantiate_parameters (loop, chrec1);
1904 chrec2 = instantiate_parameters (loop, chrec2);
1906 tree one = build_int_cst (uns, 1);
1907 chrec2 = fold_build2 (LSHIFT_EXPR, uns, one, chrec2);
1908 res = chrec_fold_multiply (uns, chrec1, chrec2);
1909 res = chrec_convert (type, res, at_stmt);
1911 break;
1913 CASE_CONVERT:
1914 /* In case we have a truncation of a widened operation that in
1915 the truncated type has undefined overflow behavior analyze
1916 the operation done in an unsigned type of the same precision
1917 as the final truncation. We cannot derive a scalar evolution
1918 for the widened operation but for the truncated result. */
1919 if (TREE_CODE (type) == INTEGER_TYPE
1920 && TREE_CODE (TREE_TYPE (rhs1)) == INTEGER_TYPE
1921 && TYPE_PRECISION (type) < TYPE_PRECISION (TREE_TYPE (rhs1))
1922 && TYPE_OVERFLOW_UNDEFINED (type)
1923 && TREE_CODE (rhs1) == SSA_NAME
1924 && (def = SSA_NAME_DEF_STMT (rhs1))
1925 && is_gimple_assign (def)
1926 && TREE_CODE_CLASS (gimple_assign_rhs_code (def)) == tcc_binary
1927 && TREE_CODE (gimple_assign_rhs2 (def)) == INTEGER_CST)
1929 tree utype = unsigned_type_for (type);
1930 chrec1 = interpret_rhs_expr (loop, at_stmt, utype,
1931 gimple_assign_rhs1 (def),
1932 gimple_assign_rhs_code (def),
1933 gimple_assign_rhs2 (def));
1935 else
1936 chrec1 = analyze_scalar_evolution (loop, rhs1);
1937 res = chrec_convert (type, chrec1, at_stmt);
1938 break;
1940 case BIT_AND_EXPR:
1941 /* Given int variable A, handle A&0xffff as (int)(unsigned short)A.
1942 If A is SCEV and its value is in the range of representable set
1943 of type unsigned short, the result expression is a (no-overflow)
1944 SCEV. */
1945 res = chrec_dont_know;
1946 if (tree_fits_uhwi_p (rhs2))
1948 int precision;
1949 unsigned HOST_WIDE_INT val = tree_to_uhwi (rhs2);
1951 val ++;
1952 /* Skip if value of rhs2 wraps in unsigned HOST_WIDE_INT or
1953 it's not the maximum value of a smaller type than rhs1. */
1954 if (val != 0
1955 && (precision = exact_log2 (val)) > 0
1956 && (unsigned) precision < TYPE_PRECISION (TREE_TYPE (rhs1)))
1958 tree utype = build_nonstandard_integer_type (precision, 1);
1960 if (TYPE_PRECISION (utype) < TYPE_PRECISION (TREE_TYPE (rhs1)))
1962 chrec1 = analyze_scalar_evolution (loop, rhs1);
1963 chrec1 = chrec_convert (utype, chrec1, at_stmt);
1964 res = chrec_convert (TREE_TYPE (rhs1), chrec1, at_stmt);
1968 break;
1970 default:
1971 res = chrec_dont_know;
1972 break;
1975 return res;
1978 /* Interpret the expression EXPR. */
1980 static tree
1981 interpret_expr (struct loop *loop, gimple *at_stmt, tree expr)
1983 enum tree_code code;
1984 tree type = TREE_TYPE (expr), op0, op1;
1986 if (automatically_generated_chrec_p (expr))
1987 return expr;
1989 if (TREE_CODE (expr) == POLYNOMIAL_CHREC
1990 || get_gimple_rhs_class (TREE_CODE (expr)) == GIMPLE_TERNARY_RHS)
1991 return chrec_dont_know;
1993 extract_ops_from_tree (expr, &code, &op0, &op1);
1995 return interpret_rhs_expr (loop, at_stmt, type,
1996 op0, code, op1);
1999 /* Interpret the rhs of the assignment STMT. */
2001 static tree
2002 interpret_gimple_assign (struct loop *loop, gimple *stmt)
2004 tree type = TREE_TYPE (gimple_assign_lhs (stmt));
2005 enum tree_code code = gimple_assign_rhs_code (stmt);
2007 return interpret_rhs_expr (loop, stmt, type,
2008 gimple_assign_rhs1 (stmt), code,
2009 gimple_assign_rhs2 (stmt));
2014 /* This section contains all the entry points:
2015 - number_of_iterations_in_loop,
2016 - analyze_scalar_evolution,
2017 - instantiate_parameters.
2020 /* Compute and return the evolution function in WRTO_LOOP, the nearest
2021 common ancestor of DEF_LOOP and USE_LOOP. */
2023 static tree
2024 compute_scalar_evolution_in_loop (struct loop *wrto_loop,
2025 struct loop *def_loop,
2026 tree ev)
2028 bool val;
2029 tree res;
2031 if (def_loop == wrto_loop)
2032 return ev;
2034 def_loop = superloop_at_depth (def_loop, loop_depth (wrto_loop) + 1);
2035 res = compute_overall_effect_of_inner_loop (def_loop, ev);
2037 if (no_evolution_in_loop_p (res, wrto_loop->num, &val) && val)
2038 return res;
2040 return analyze_scalar_evolution_1 (wrto_loop, res, chrec_not_analyzed_yet);
2043 /* Helper recursive function. */
2045 static tree
2046 analyze_scalar_evolution_1 (struct loop *loop, tree var, tree res)
2048 tree type = TREE_TYPE (var);
2049 gimple *def;
2050 basic_block bb;
2051 struct loop *def_loop;
2053 if (loop == NULL || TREE_CODE (type) == VECTOR_TYPE)
2054 return chrec_dont_know;
2056 if (TREE_CODE (var) != SSA_NAME)
2057 return interpret_expr (loop, NULL, var);
2059 def = SSA_NAME_DEF_STMT (var);
2060 bb = gimple_bb (def);
2061 def_loop = bb ? bb->loop_father : NULL;
2063 if (bb == NULL
2064 || !flow_bb_inside_loop_p (loop, bb))
2066 /* Keep symbolic form, but look through obvious copies for constants. */
2067 res = follow_copies_to_constant (var);
2068 goto set_and_end;
2071 if (res != chrec_not_analyzed_yet)
2073 if (loop != bb->loop_father)
2074 res = compute_scalar_evolution_in_loop
2075 (find_common_loop (loop, bb->loop_father), bb->loop_father, res);
2077 goto set_and_end;
2080 if (loop != def_loop)
2082 res = analyze_scalar_evolution_1 (def_loop, var, chrec_not_analyzed_yet);
2083 res = compute_scalar_evolution_in_loop (loop, def_loop, res);
2085 goto set_and_end;
2088 switch (gimple_code (def))
2090 case GIMPLE_ASSIGN:
2091 res = interpret_gimple_assign (loop, def);
2092 break;
2094 case GIMPLE_PHI:
2095 if (loop_phi_node_p (def))
2096 res = interpret_loop_phi (loop, as_a <gphi *> (def));
2097 else
2098 res = interpret_condition_phi (loop, as_a <gphi *> (def));
2099 break;
2101 default:
2102 res = chrec_dont_know;
2103 break;
2106 set_and_end:
2108 /* Keep the symbolic form. */
2109 if (res == chrec_dont_know)
2110 res = var;
2112 if (loop == def_loop)
2113 set_scalar_evolution (block_before_loop (loop), var, res);
2115 return res;
2118 /* Analyzes and returns the scalar evolution of the ssa_name VAR in
2119 LOOP. LOOP is the loop in which the variable is used.
2121 Example of use: having a pointer VAR to a SSA_NAME node, STMT a
2122 pointer to the statement that uses this variable, in order to
2123 determine the evolution function of the variable, use the following
2124 calls:
2126 loop_p loop = loop_containing_stmt (stmt);
2127 tree chrec_with_symbols = analyze_scalar_evolution (loop, var);
2128 tree chrec_instantiated = instantiate_parameters (loop, chrec_with_symbols);
2131 tree
2132 analyze_scalar_evolution (struct loop *loop, tree var)
2134 tree res;
2136 if (dump_file && (dump_flags & TDF_SCEV))
2138 fprintf (dump_file, "(analyze_scalar_evolution \n");
2139 fprintf (dump_file, " (loop_nb = %d)\n", loop->num);
2140 fprintf (dump_file, " (scalar = ");
2141 print_generic_expr (dump_file, var, 0);
2142 fprintf (dump_file, ")\n");
2145 res = get_scalar_evolution (block_before_loop (loop), var);
2146 res = analyze_scalar_evolution_1 (loop, var, res);
2148 if (dump_file && (dump_flags & TDF_SCEV))
2149 fprintf (dump_file, ")\n");
2151 return res;
2154 /* Analyzes and returns the scalar evolution of VAR address in LOOP. */
2156 static tree
2157 analyze_scalar_evolution_for_address_of (struct loop *loop, tree var)
2159 return analyze_scalar_evolution (loop, build_fold_addr_expr (var));
2162 /* Analyze scalar evolution of use of VERSION in USE_LOOP with respect to
2163 WRTO_LOOP (which should be a superloop of USE_LOOP)
2165 FOLDED_CASTS is set to true if resolve_mixers used
2166 chrec_convert_aggressive (TODO -- not really, we are way too conservative
2167 at the moment in order to keep things simple).
2169 To illustrate the meaning of USE_LOOP and WRTO_LOOP, consider the following
2170 example:
2172 for (i = 0; i < 100; i++) -- loop 1
2174 for (j = 0; j < 100; j++) -- loop 2
2176 k1 = i;
2177 k2 = j;
2179 use2 (k1, k2);
2181 for (t = 0; t < 100; t++) -- loop 3
2182 use3 (k1, k2);
2185 use1 (k1, k2);
2188 Both k1 and k2 are invariants in loop3, thus
2189 analyze_scalar_evolution_in_loop (loop3, loop3, k1) = k1
2190 analyze_scalar_evolution_in_loop (loop3, loop3, k2) = k2
2192 As they are invariant, it does not matter whether we consider their
2193 usage in loop 3 or loop 2, hence
2194 analyze_scalar_evolution_in_loop (loop2, loop3, k1) =
2195 analyze_scalar_evolution_in_loop (loop2, loop2, k1) = i
2196 analyze_scalar_evolution_in_loop (loop2, loop3, k2) =
2197 analyze_scalar_evolution_in_loop (loop2, loop2, k2) = [0,+,1]_2
2199 Similarly for their evolutions with respect to loop 1. The values of K2
2200 in the use in loop 2 vary independently on loop 1, thus we cannot express
2201 the evolution with respect to loop 1:
2202 analyze_scalar_evolution_in_loop (loop1, loop3, k1) =
2203 analyze_scalar_evolution_in_loop (loop1, loop2, k1) = [0,+,1]_1
2204 analyze_scalar_evolution_in_loop (loop1, loop3, k2) =
2205 analyze_scalar_evolution_in_loop (loop1, loop2, k2) = dont_know
2207 The value of k2 in the use in loop 1 is known, though:
2208 analyze_scalar_evolution_in_loop (loop1, loop1, k1) = [0,+,1]_1
2209 analyze_scalar_evolution_in_loop (loop1, loop1, k2) = 100
2212 static tree
2213 analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop,
2214 tree version, bool *folded_casts)
2216 bool val = false;
2217 tree ev = version, tmp;
2219 /* We cannot just do
2221 tmp = analyze_scalar_evolution (use_loop, version);
2222 ev = resolve_mixers (wrto_loop, tmp, folded_casts);
2224 as resolve_mixers would query the scalar evolution with respect to
2225 wrto_loop. For example, in the situation described in the function
2226 comment, suppose that wrto_loop = loop1, use_loop = loop3 and
2227 version = k2. Then
2229 analyze_scalar_evolution (use_loop, version) = k2
2231 and resolve_mixers (loop1, k2, folded_casts) finds that the value of
2232 k2 in loop 1 is 100, which is a wrong result, since we are interested
2233 in the value in loop 3.
2235 Instead, we need to proceed from use_loop to wrto_loop loop by loop,
2236 each time checking that there is no evolution in the inner loop. */
2238 if (folded_casts)
2239 *folded_casts = false;
2240 while (1)
2242 tmp = analyze_scalar_evolution (use_loop, ev);
2243 ev = resolve_mixers (use_loop, tmp, folded_casts);
2245 if (use_loop == wrto_loop)
2246 return ev;
2248 /* If the value of the use changes in the inner loop, we cannot express
2249 its value in the outer loop (we might try to return interval chrec,
2250 but we do not have a user for it anyway) */
2251 if (!no_evolution_in_loop_p (ev, use_loop->num, &val)
2252 || !val)
2253 return chrec_dont_know;
2255 use_loop = loop_outer (use_loop);
2260 /* Hashtable helpers for a temporary hash-table used when
2261 instantiating a CHREC or resolving mixers. For this use
2262 instantiated_below is always the same. */
2264 struct instantiate_cache_type
2266 htab_t map;
2267 vec<scev_info_str> entries;
2269 instantiate_cache_type () : map (NULL), entries (vNULL) {}
2270 ~instantiate_cache_type ();
2271 tree get (unsigned slot) { return entries[slot].chrec; }
2272 void set (unsigned slot, tree chrec) { entries[slot].chrec = chrec; }
2275 instantiate_cache_type::~instantiate_cache_type ()
2277 if (map != NULL)
2279 htab_delete (map);
2280 entries.release ();
2284 /* Cache to avoid infinite recursion when instantiating an SSA name.
2285 Live during the outermost instantiate_scev or resolve_mixers call. */
2286 static instantiate_cache_type *global_cache;
2288 /* Computes a hash function for database element ELT. */
2290 static inline hashval_t
2291 hash_idx_scev_info (const void *elt_)
2293 unsigned idx = ((size_t) elt_) - 2;
2294 return scev_info_hasher::hash (&global_cache->entries[idx]);
2297 /* Compares database elements E1 and E2. */
2299 static inline int
2300 eq_idx_scev_info (const void *e1, const void *e2)
2302 unsigned idx1 = ((size_t) e1) - 2;
2303 return scev_info_hasher::equal (&global_cache->entries[idx1],
2304 (const scev_info_str *) e2);
2307 /* Returns from CACHE the slot number of the cached chrec for NAME. */
2309 static unsigned
2310 get_instantiated_value_entry (instantiate_cache_type &cache,
2311 tree name, basic_block instantiate_below)
2313 if (!cache.map)
2315 cache.map = htab_create (10, hash_idx_scev_info, eq_idx_scev_info, NULL);
2316 cache.entries.create (10);
2319 scev_info_str e;
2320 e.name_version = SSA_NAME_VERSION (name);
2321 e.instantiated_below = instantiate_below->index;
2322 void **slot = htab_find_slot_with_hash (cache.map, &e,
2323 scev_info_hasher::hash (&e), INSERT);
2324 if (!*slot)
2326 e.chrec = chrec_not_analyzed_yet;
2327 *slot = (void *)(size_t)(cache.entries.length () + 2);
2328 cache.entries.safe_push (e);
2331 return ((size_t)*slot) - 2;
2335 /* Return the closed_loop_phi node for VAR. If there is none, return
2336 NULL_TREE. */
2338 static tree
2339 loop_closed_phi_def (tree var)
2341 struct loop *loop;
2342 edge exit;
2343 gphi *phi;
2344 gphi_iterator psi;
2346 if (var == NULL_TREE
2347 || TREE_CODE (var) != SSA_NAME)
2348 return NULL_TREE;
2350 loop = loop_containing_stmt (SSA_NAME_DEF_STMT (var));
2351 exit = single_exit (loop);
2352 if (!exit)
2353 return NULL_TREE;
2355 for (psi = gsi_start_phis (exit->dest); !gsi_end_p (psi); gsi_next (&psi))
2357 phi = psi.phi ();
2358 if (PHI_ARG_DEF_FROM_EDGE (phi, exit) == var)
2359 return PHI_RESULT (phi);
2362 return NULL_TREE;
2365 static tree instantiate_scev_r (basic_block, struct loop *, struct loop *,
2366 tree, bool *, int);
2368 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2369 and EVOLUTION_LOOP, that were left under a symbolic form.
2371 CHREC is an SSA_NAME to be instantiated.
2373 CACHE is the cache of already instantiated values.
2375 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2376 conversions that may wrap in signed/pointer type are folded, as long
2377 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2378 then we don't do such fold.
2380 SIZE_EXPR is used for computing the size of the expression to be
2381 instantiated, and to stop if it exceeds some limit. */
2383 static tree
2384 instantiate_scev_name (basic_block instantiate_below,
2385 struct loop *evolution_loop, struct loop *inner_loop,
2386 tree chrec,
2387 bool *fold_conversions,
2388 int size_expr)
2390 tree res;
2391 struct loop *def_loop;
2392 basic_block def_bb = gimple_bb (SSA_NAME_DEF_STMT (chrec));
2394 /* A parameter (or loop invariant and we do not want to include
2395 evolutions in outer loops), nothing to do. */
2396 if (!def_bb
2397 || loop_depth (def_bb->loop_father) == 0
2398 || dominated_by_p (CDI_DOMINATORS, instantiate_below, def_bb))
2399 return chrec;
2401 /* We cache the value of instantiated variable to avoid exponential
2402 time complexity due to reevaluations. We also store the convenient
2403 value in the cache in order to prevent infinite recursion -- we do
2404 not want to instantiate the SSA_NAME if it is in a mixer
2405 structure. This is used for avoiding the instantiation of
2406 recursively defined functions, such as:
2408 | a_2 -> {0, +, 1, +, a_2}_1 */
2410 unsigned si = get_instantiated_value_entry (*global_cache,
2411 chrec, instantiate_below);
2412 if (global_cache->get (si) != chrec_not_analyzed_yet)
2413 return global_cache->get (si);
2415 /* On recursion return chrec_dont_know. */
2416 global_cache->set (si, chrec_dont_know);
2418 def_loop = find_common_loop (evolution_loop, def_bb->loop_father);
2420 /* If the analysis yields a parametric chrec, instantiate the
2421 result again. */
2422 res = analyze_scalar_evolution (def_loop, chrec);
2424 /* Don't instantiate default definitions. */
2425 if (TREE_CODE (res) == SSA_NAME
2426 && SSA_NAME_IS_DEFAULT_DEF (res))
2429 /* Don't instantiate loop-closed-ssa phi nodes. */
2430 else if (TREE_CODE (res) == SSA_NAME
2431 && loop_depth (loop_containing_stmt (SSA_NAME_DEF_STMT (res)))
2432 > loop_depth (def_loop))
2434 if (res == chrec)
2435 res = loop_closed_phi_def (chrec);
2436 else
2437 res = chrec;
2439 /* When there is no loop_closed_phi_def, it means that the
2440 variable is not used after the loop: try to still compute the
2441 value of the variable when exiting the loop. */
2442 if (res == NULL_TREE)
2444 loop_p loop = loop_containing_stmt (SSA_NAME_DEF_STMT (chrec));
2445 res = analyze_scalar_evolution (loop, chrec);
2446 res = compute_overall_effect_of_inner_loop (loop, res);
2447 res = instantiate_scev_r (instantiate_below, evolution_loop,
2448 inner_loop, res,
2449 fold_conversions, size_expr);
2451 else if (!dominated_by_p (CDI_DOMINATORS, instantiate_below,
2452 gimple_bb (SSA_NAME_DEF_STMT (res))))
2453 res = chrec_dont_know;
2456 else if (res != chrec_dont_know)
2458 if (inner_loop
2459 && def_bb->loop_father != inner_loop
2460 && !flow_loop_nested_p (def_bb->loop_father, inner_loop))
2461 /* ??? We could try to compute the overall effect of the loop here. */
2462 res = chrec_dont_know;
2463 else
2464 res = instantiate_scev_r (instantiate_below, evolution_loop,
2465 inner_loop, res,
2466 fold_conversions, size_expr);
2469 /* Store the correct value to the cache. */
2470 global_cache->set (si, res);
2471 return res;
2474 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2475 and EVOLUTION_LOOP, that were left under a symbolic form.
2477 CHREC is a polynomial chain of recurrence to be instantiated.
2479 CACHE is the cache of already instantiated values.
2481 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2482 conversions that may wrap in signed/pointer type are folded, as long
2483 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2484 then we don't do such fold.
2486 SIZE_EXPR is used for computing the size of the expression to be
2487 instantiated, and to stop if it exceeds some limit. */
2489 static tree
2490 instantiate_scev_poly (basic_block instantiate_below,
2491 struct loop *evolution_loop, struct loop *,
2492 tree chrec, bool *fold_conversions, int size_expr)
2494 tree op1;
2495 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2496 get_chrec_loop (chrec),
2497 CHREC_LEFT (chrec), fold_conversions,
2498 size_expr);
2499 if (op0 == chrec_dont_know)
2500 return chrec_dont_know;
2502 op1 = instantiate_scev_r (instantiate_below, evolution_loop,
2503 get_chrec_loop (chrec),
2504 CHREC_RIGHT (chrec), fold_conversions,
2505 size_expr);
2506 if (op1 == chrec_dont_know)
2507 return chrec_dont_know;
2509 if (CHREC_LEFT (chrec) != op0
2510 || CHREC_RIGHT (chrec) != op1)
2512 op1 = chrec_convert_rhs (chrec_type (op0), op1, NULL);
2513 chrec = build_polynomial_chrec (CHREC_VARIABLE (chrec), op0, op1);
2516 return chrec;
2519 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2520 and EVOLUTION_LOOP, that were left under a symbolic form.
2522 "C0 CODE C1" is a binary expression of type TYPE to be instantiated.
2524 CACHE is the cache of already instantiated values.
2526 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2527 conversions that may wrap in signed/pointer type are folded, as long
2528 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2529 then we don't do such fold.
2531 SIZE_EXPR is used for computing the size of the expression to be
2532 instantiated, and to stop if it exceeds some limit. */
2534 static tree
2535 instantiate_scev_binary (basic_block instantiate_below,
2536 struct loop *evolution_loop, struct loop *inner_loop,
2537 tree chrec, enum tree_code code,
2538 tree type, tree c0, tree c1,
2539 bool *fold_conversions, int size_expr)
2541 tree op1;
2542 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, inner_loop,
2543 c0, fold_conversions, size_expr);
2544 if (op0 == chrec_dont_know)
2545 return chrec_dont_know;
2547 op1 = instantiate_scev_r (instantiate_below, evolution_loop, inner_loop,
2548 c1, fold_conversions, size_expr);
2549 if (op1 == chrec_dont_know)
2550 return chrec_dont_know;
2552 if (c0 != op0
2553 || c1 != op1)
2555 op0 = chrec_convert (type, op0, NULL);
2556 op1 = chrec_convert_rhs (type, op1, NULL);
2558 switch (code)
2560 case POINTER_PLUS_EXPR:
2561 case PLUS_EXPR:
2562 return chrec_fold_plus (type, op0, op1);
2564 case MINUS_EXPR:
2565 return chrec_fold_minus (type, op0, op1);
2567 case MULT_EXPR:
2568 return chrec_fold_multiply (type, op0, op1);
2570 default:
2571 gcc_unreachable ();
2575 return chrec ? chrec : fold_build2 (code, type, c0, c1);
2578 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2579 and EVOLUTION_LOOP, that were left under a symbolic form.
2581 "CHREC" is an array reference to be instantiated.
2583 CACHE is the cache of already instantiated values.
2585 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2586 conversions that may wrap in signed/pointer type are folded, as long
2587 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2588 then we don't do such fold.
2590 SIZE_EXPR is used for computing the size of the expression to be
2591 instantiated, and to stop if it exceeds some limit. */
2593 static tree
2594 instantiate_array_ref (basic_block instantiate_below,
2595 struct loop *evolution_loop, struct loop *inner_loop,
2596 tree chrec, bool *fold_conversions, int size_expr)
2598 tree res;
2599 tree index = TREE_OPERAND (chrec, 1);
2600 tree op1 = instantiate_scev_r (instantiate_below, evolution_loop,
2601 inner_loop, index,
2602 fold_conversions, size_expr);
2604 if (op1 == chrec_dont_know)
2605 return chrec_dont_know;
2607 if (chrec && op1 == index)
2608 return chrec;
2610 res = unshare_expr (chrec);
2611 TREE_OPERAND (res, 1) = op1;
2612 return res;
2615 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2616 and EVOLUTION_LOOP, that were left under a symbolic form.
2618 "CHREC" that stands for a convert expression "(TYPE) OP" is to be
2619 instantiated.
2621 CACHE is the cache of already instantiated values.
2623 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2624 conversions that may wrap in signed/pointer type are folded, as long
2625 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2626 then we don't do such fold.
2628 SIZE_EXPR is used for computing the size of the expression to be
2629 instantiated, and to stop if it exceeds some limit. */
2631 static tree
2632 instantiate_scev_convert (basic_block instantiate_below,
2633 struct loop *evolution_loop, struct loop *inner_loop,
2634 tree chrec, tree type, tree op,
2635 bool *fold_conversions, int size_expr)
2637 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2638 inner_loop, op,
2639 fold_conversions, size_expr);
2641 if (op0 == chrec_dont_know)
2642 return chrec_dont_know;
2644 if (fold_conversions)
2646 tree tmp = chrec_convert_aggressive (type, op0, fold_conversions);
2647 if (tmp)
2648 return tmp;
2650 /* If we used chrec_convert_aggressive, we can no longer assume that
2651 signed chrecs do not overflow, as chrec_convert does, so avoid
2652 calling it in that case. */
2653 if (*fold_conversions)
2655 if (chrec && op0 == op)
2656 return chrec;
2658 return fold_convert (type, op0);
2662 return chrec_convert (type, op0, NULL);
2665 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2666 and EVOLUTION_LOOP, that were left under a symbolic form.
2668 CHREC is a BIT_NOT_EXPR or a NEGATE_EXPR expression to be instantiated.
2669 Handle ~X as -1 - X.
2670 Handle -X as -1 * X.
2672 CACHE is the cache of already instantiated values.
2674 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2675 conversions that may wrap in signed/pointer type are folded, as long
2676 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2677 then we don't do such fold.
2679 SIZE_EXPR is used for computing the size of the expression to be
2680 instantiated, and to stop if it exceeds some limit. */
2682 static tree
2683 instantiate_scev_not (basic_block instantiate_below,
2684 struct loop *evolution_loop, struct loop *inner_loop,
2685 tree chrec,
2686 enum tree_code code, tree type, tree op,
2687 bool *fold_conversions, int size_expr)
2689 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2690 inner_loop, op,
2691 fold_conversions, size_expr);
2693 if (op0 == chrec_dont_know)
2694 return chrec_dont_know;
2696 if (op != op0)
2698 op0 = chrec_convert (type, op0, NULL);
2700 switch (code)
2702 case BIT_NOT_EXPR:
2703 return chrec_fold_minus
2704 (type, fold_convert (type, integer_minus_one_node), op0);
2706 case NEGATE_EXPR:
2707 return chrec_fold_multiply
2708 (type, fold_convert (type, integer_minus_one_node), op0);
2710 default:
2711 gcc_unreachable ();
2715 return chrec ? chrec : fold_build1 (code, type, op0);
2718 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2719 and EVOLUTION_LOOP, that were left under a symbolic form.
2721 CHREC is an expression with 3 operands to be instantiated.
2723 CACHE is the cache of already instantiated values.
2725 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2726 conversions that may wrap in signed/pointer type are folded, as long
2727 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2728 then we don't do such fold.
2730 SIZE_EXPR is used for computing the size of the expression to be
2731 instantiated, and to stop if it exceeds some limit. */
2733 static tree
2734 instantiate_scev_3 (basic_block instantiate_below,
2735 struct loop *evolution_loop, struct loop *inner_loop,
2736 tree chrec,
2737 bool *fold_conversions, int size_expr)
2739 tree op1, op2;
2740 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2741 inner_loop, TREE_OPERAND (chrec, 0),
2742 fold_conversions, size_expr);
2743 if (op0 == chrec_dont_know)
2744 return chrec_dont_know;
2746 op1 = instantiate_scev_r (instantiate_below, evolution_loop,
2747 inner_loop, TREE_OPERAND (chrec, 1),
2748 fold_conversions, size_expr);
2749 if (op1 == chrec_dont_know)
2750 return chrec_dont_know;
2752 op2 = instantiate_scev_r (instantiate_below, evolution_loop,
2753 inner_loop, TREE_OPERAND (chrec, 2),
2754 fold_conversions, size_expr);
2755 if (op2 == chrec_dont_know)
2756 return chrec_dont_know;
2758 if (op0 == TREE_OPERAND (chrec, 0)
2759 && op1 == TREE_OPERAND (chrec, 1)
2760 && op2 == TREE_OPERAND (chrec, 2))
2761 return chrec;
2763 return fold_build3 (TREE_CODE (chrec),
2764 TREE_TYPE (chrec), op0, op1, op2);
2767 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2768 and EVOLUTION_LOOP, that were left under a symbolic form.
2770 CHREC is an expression with 2 operands to be instantiated.
2772 CACHE is the cache of already instantiated values.
2774 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2775 conversions that may wrap in signed/pointer type are folded, as long
2776 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2777 then we don't do such fold.
2779 SIZE_EXPR is used for computing the size of the expression to be
2780 instantiated, and to stop if it exceeds some limit. */
2782 static tree
2783 instantiate_scev_2 (basic_block instantiate_below,
2784 struct loop *evolution_loop, struct loop *inner_loop,
2785 tree chrec,
2786 bool *fold_conversions, int size_expr)
2788 tree op1;
2789 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2790 inner_loop, TREE_OPERAND (chrec, 0),
2791 fold_conversions, size_expr);
2792 if (op0 == chrec_dont_know)
2793 return chrec_dont_know;
2795 op1 = instantiate_scev_r (instantiate_below, evolution_loop,
2796 inner_loop, TREE_OPERAND (chrec, 1),
2797 fold_conversions, size_expr);
2798 if (op1 == chrec_dont_know)
2799 return chrec_dont_know;
2801 if (op0 == TREE_OPERAND (chrec, 0)
2802 && op1 == TREE_OPERAND (chrec, 1))
2803 return chrec;
2805 return fold_build2 (TREE_CODE (chrec), TREE_TYPE (chrec), op0, op1);
2808 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2809 and EVOLUTION_LOOP, that were left under a symbolic form.
2811 CHREC is an expression with 2 operands to be instantiated.
2813 CACHE is the cache of already instantiated values.
2815 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2816 conversions that may wrap in signed/pointer type are folded, as long
2817 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2818 then we don't do such fold.
2820 SIZE_EXPR is used for computing the size of the expression to be
2821 instantiated, and to stop if it exceeds some limit. */
2823 static tree
2824 instantiate_scev_1 (basic_block instantiate_below,
2825 struct loop *evolution_loop, struct loop *inner_loop,
2826 tree chrec,
2827 bool *fold_conversions, int size_expr)
2829 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2830 inner_loop, TREE_OPERAND (chrec, 0),
2831 fold_conversions, size_expr);
2833 if (op0 == chrec_dont_know)
2834 return chrec_dont_know;
2836 if (op0 == TREE_OPERAND (chrec, 0))
2837 return chrec;
2839 return fold_build1 (TREE_CODE (chrec), TREE_TYPE (chrec), op0);
2842 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2843 and EVOLUTION_LOOP, that were left under a symbolic form.
2845 CHREC is the scalar evolution to instantiate.
2847 CACHE is the cache of already instantiated values.
2849 Variable pointed by FOLD_CONVERSIONS is set to TRUE when the
2850 conversions that may wrap in signed/pointer type are folded, as long
2851 as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL
2852 then we don't do such fold.
2854 SIZE_EXPR is used for computing the size of the expression to be
2855 instantiated, and to stop if it exceeds some limit. */
2857 static tree
2858 instantiate_scev_r (basic_block instantiate_below,
2859 struct loop *evolution_loop, struct loop *inner_loop,
2860 tree chrec,
2861 bool *fold_conversions, int size_expr)
2863 /* Give up if the expression is larger than the MAX that we allow. */
2864 if (size_expr++ > PARAM_VALUE (PARAM_SCEV_MAX_EXPR_SIZE))
2865 return chrec_dont_know;
2867 if (chrec == NULL_TREE
2868 || automatically_generated_chrec_p (chrec)
2869 || is_gimple_min_invariant (chrec))
2870 return chrec;
2872 switch (TREE_CODE (chrec))
2874 case SSA_NAME:
2875 return instantiate_scev_name (instantiate_below, evolution_loop,
2876 inner_loop, chrec,
2877 fold_conversions, size_expr);
2879 case POLYNOMIAL_CHREC:
2880 return instantiate_scev_poly (instantiate_below, evolution_loop,
2881 inner_loop, chrec,
2882 fold_conversions, size_expr);
2884 case POINTER_PLUS_EXPR:
2885 case PLUS_EXPR:
2886 case MINUS_EXPR:
2887 case MULT_EXPR:
2888 return instantiate_scev_binary (instantiate_below, evolution_loop,
2889 inner_loop, chrec,
2890 TREE_CODE (chrec), chrec_type (chrec),
2891 TREE_OPERAND (chrec, 0),
2892 TREE_OPERAND (chrec, 1),
2893 fold_conversions, size_expr);
2895 CASE_CONVERT:
2896 return instantiate_scev_convert (instantiate_below, evolution_loop,
2897 inner_loop, chrec,
2898 TREE_TYPE (chrec), TREE_OPERAND (chrec, 0),
2899 fold_conversions, size_expr);
2901 case NEGATE_EXPR:
2902 case BIT_NOT_EXPR:
2903 return instantiate_scev_not (instantiate_below, evolution_loop,
2904 inner_loop, chrec,
2905 TREE_CODE (chrec), TREE_TYPE (chrec),
2906 TREE_OPERAND (chrec, 0),
2907 fold_conversions, size_expr);
2909 case ADDR_EXPR:
2910 case SCEV_NOT_KNOWN:
2911 return chrec_dont_know;
2913 case SCEV_KNOWN:
2914 return chrec_known;
2916 case ARRAY_REF:
2917 return instantiate_array_ref (instantiate_below, evolution_loop,
2918 inner_loop, chrec,
2919 fold_conversions, size_expr);
2921 default:
2922 break;
2925 if (VL_EXP_CLASS_P (chrec))
2926 return chrec_dont_know;
2928 switch (TREE_CODE_LENGTH (TREE_CODE (chrec)))
2930 case 3:
2931 return instantiate_scev_3 (instantiate_below, evolution_loop,
2932 inner_loop, chrec,
2933 fold_conversions, size_expr);
2935 case 2:
2936 return instantiate_scev_2 (instantiate_below, evolution_loop,
2937 inner_loop, chrec,
2938 fold_conversions, size_expr);
2940 case 1:
2941 return instantiate_scev_1 (instantiate_below, evolution_loop,
2942 inner_loop, chrec,
2943 fold_conversions, size_expr);
2945 case 0:
2946 return chrec;
2948 default:
2949 break;
2952 /* Too complicated to handle. */
2953 return chrec_dont_know;
2956 /* Analyze all the parameters of the chrec that were left under a
2957 symbolic form. INSTANTIATE_BELOW is the basic block that stops the
2958 recursive instantiation of parameters: a parameter is a variable
2959 that is defined in a basic block that dominates INSTANTIATE_BELOW or
2960 a function parameter. */
2962 tree
2963 instantiate_scev (basic_block instantiate_below, struct loop *evolution_loop,
2964 tree chrec)
2966 tree res;
2968 if (dump_file && (dump_flags & TDF_SCEV))
2970 fprintf (dump_file, "(instantiate_scev \n");
2971 fprintf (dump_file, " (instantiate_below = %d)\n", instantiate_below->index);
2972 fprintf (dump_file, " (evolution_loop = %d)\n", evolution_loop->num);
2973 fprintf (dump_file, " (chrec = ");
2974 print_generic_expr (dump_file, chrec, 0);
2975 fprintf (dump_file, ")\n");
2978 bool destr = false;
2979 if (!global_cache)
2981 global_cache = new instantiate_cache_type;
2982 destr = true;
2985 res = instantiate_scev_r (instantiate_below, evolution_loop,
2986 NULL, chrec, NULL, 0);
2988 if (destr)
2990 delete global_cache;
2991 global_cache = NULL;
2994 if (dump_file && (dump_flags & TDF_SCEV))
2996 fprintf (dump_file, " (res = ");
2997 print_generic_expr (dump_file, res, 0);
2998 fprintf (dump_file, "))\n");
3001 return res;
3004 /* Similar to instantiate_parameters, but does not introduce the
3005 evolutions in outer loops for LOOP invariants in CHREC, and does not
3006 care about causing overflows, as long as they do not affect value
3007 of an expression. */
3009 tree
3010 resolve_mixers (struct loop *loop, tree chrec, bool *folded_casts)
3012 bool destr = false;
3013 bool fold_conversions = false;
3014 if (!global_cache)
3016 global_cache = new instantiate_cache_type;
3017 destr = true;
3020 tree ret = instantiate_scev_r (block_before_loop (loop), loop, NULL,
3021 chrec, &fold_conversions, 0);
3023 if (folded_casts && !*folded_casts)
3024 *folded_casts = fold_conversions;
3026 if (destr)
3028 delete global_cache;
3029 global_cache = NULL;
3032 return ret;
3035 /* Entry point for the analysis of the number of iterations pass.
3036 This function tries to safely approximate the number of iterations
3037 the loop will run. When this property is not decidable at compile
3038 time, the result is chrec_dont_know. Otherwise the result is a
3039 scalar or a symbolic parameter. When the number of iterations may
3040 be equal to zero and the property cannot be determined at compile
3041 time, the result is a COND_EXPR that represents in a symbolic form
3042 the conditions under which the number of iterations is not zero.
3044 Example of analysis: suppose that the loop has an exit condition:
3046 "if (b > 49) goto end_loop;"
3048 and that in a previous analysis we have determined that the
3049 variable 'b' has an evolution function:
3051 "EF = {23, +, 5}_2".
3053 When we evaluate the function at the point 5, i.e. the value of the
3054 variable 'b' after 5 iterations in the loop, we have EF (5) = 48,
3055 and EF (6) = 53. In this case the value of 'b' on exit is '53' and
3056 the loop body has been executed 6 times. */
3058 tree
3059 number_of_latch_executions (struct loop *loop)
3061 edge exit;
3062 struct tree_niter_desc niter_desc;
3063 tree may_be_zero;
3064 tree res;
3066 /* Determine whether the number of iterations in loop has already
3067 been computed. */
3068 res = loop->nb_iterations;
3069 if (res)
3070 return res;
3072 may_be_zero = NULL_TREE;
3074 if (dump_file && (dump_flags & TDF_SCEV))
3075 fprintf (dump_file, "(number_of_iterations_in_loop = \n");
3077 res = chrec_dont_know;
3078 exit = single_exit (loop);
3080 if (exit && number_of_iterations_exit (loop, exit, &niter_desc, false))
3082 may_be_zero = niter_desc.may_be_zero;
3083 res = niter_desc.niter;
3086 if (res == chrec_dont_know
3087 || !may_be_zero
3088 || integer_zerop (may_be_zero))
3090 else if (integer_nonzerop (may_be_zero))
3091 res = build_int_cst (TREE_TYPE (res), 0);
3093 else if (COMPARISON_CLASS_P (may_be_zero))
3094 res = fold_build3 (COND_EXPR, TREE_TYPE (res), may_be_zero,
3095 build_int_cst (TREE_TYPE (res), 0), res);
3096 else
3097 res = chrec_dont_know;
3099 if (dump_file && (dump_flags & TDF_SCEV))
3101 fprintf (dump_file, " (set_nb_iterations_in_loop = ");
3102 print_generic_expr (dump_file, res, 0);
3103 fprintf (dump_file, "))\n");
3106 loop->nb_iterations = res;
3107 return res;
3111 /* Counters for the stats. */
3113 struct chrec_stats
3115 unsigned nb_chrecs;
3116 unsigned nb_affine;
3117 unsigned nb_affine_multivar;
3118 unsigned nb_higher_poly;
3119 unsigned nb_chrec_dont_know;
3120 unsigned nb_undetermined;
3123 /* Reset the counters. */
3125 static inline void
3126 reset_chrecs_counters (struct chrec_stats *stats)
3128 stats->nb_chrecs = 0;
3129 stats->nb_affine = 0;
3130 stats->nb_affine_multivar = 0;
3131 stats->nb_higher_poly = 0;
3132 stats->nb_chrec_dont_know = 0;
3133 stats->nb_undetermined = 0;
3136 /* Dump the contents of a CHREC_STATS structure. */
3138 static void
3139 dump_chrecs_stats (FILE *file, struct chrec_stats *stats)
3141 fprintf (file, "\n(\n");
3142 fprintf (file, "-----------------------------------------\n");
3143 fprintf (file, "%d\taffine univariate chrecs\n", stats->nb_affine);
3144 fprintf (file, "%d\taffine multivariate chrecs\n", stats->nb_affine_multivar);
3145 fprintf (file, "%d\tdegree greater than 2 polynomials\n",
3146 stats->nb_higher_poly);
3147 fprintf (file, "%d\tchrec_dont_know chrecs\n", stats->nb_chrec_dont_know);
3148 fprintf (file, "-----------------------------------------\n");
3149 fprintf (file, "%d\ttotal chrecs\n", stats->nb_chrecs);
3150 fprintf (file, "%d\twith undetermined coefficients\n",
3151 stats->nb_undetermined);
3152 fprintf (file, "-----------------------------------------\n");
3153 fprintf (file, "%d\tchrecs in the scev database\n",
3154 (int) scalar_evolution_info->elements ());
3155 fprintf (file, "%d\tsets in the scev database\n", nb_set_scev);
3156 fprintf (file, "%d\tgets in the scev database\n", nb_get_scev);
3157 fprintf (file, "-----------------------------------------\n");
3158 fprintf (file, ")\n\n");
3161 /* Gather statistics about CHREC. */
3163 static void
3164 gather_chrec_stats (tree chrec, struct chrec_stats *stats)
3166 if (dump_file && (dump_flags & TDF_STATS))
3168 fprintf (dump_file, "(classify_chrec ");
3169 print_generic_expr (dump_file, chrec, 0);
3170 fprintf (dump_file, "\n");
3173 stats->nb_chrecs++;
3175 if (chrec == NULL_TREE)
3177 stats->nb_undetermined++;
3178 return;
3181 switch (TREE_CODE (chrec))
3183 case POLYNOMIAL_CHREC:
3184 if (evolution_function_is_affine_p (chrec))
3186 if (dump_file && (dump_flags & TDF_STATS))
3187 fprintf (dump_file, " affine_univariate\n");
3188 stats->nb_affine++;
3190 else if (evolution_function_is_affine_multivariate_p (chrec, 0))
3192 if (dump_file && (dump_flags & TDF_STATS))
3193 fprintf (dump_file, " affine_multivariate\n");
3194 stats->nb_affine_multivar++;
3196 else
3198 if (dump_file && (dump_flags & TDF_STATS))
3199 fprintf (dump_file, " higher_degree_polynomial\n");
3200 stats->nb_higher_poly++;
3203 break;
3205 default:
3206 break;
3209 if (chrec_contains_undetermined (chrec))
3211 if (dump_file && (dump_flags & TDF_STATS))
3212 fprintf (dump_file, " undetermined\n");
3213 stats->nb_undetermined++;
3216 if (dump_file && (dump_flags & TDF_STATS))
3217 fprintf (dump_file, ")\n");
3220 /* Classify the chrecs of the whole database. */
3222 void
3223 gather_stats_on_scev_database (void)
3225 struct chrec_stats stats;
3227 if (!dump_file)
3228 return;
3230 reset_chrecs_counters (&stats);
3232 hash_table<scev_info_hasher>::iterator iter;
3233 scev_info_str *elt;
3234 FOR_EACH_HASH_TABLE_ELEMENT (*scalar_evolution_info, elt, scev_info_str *,
3235 iter)
3236 gather_chrec_stats (elt->chrec, &stats);
3238 dump_chrecs_stats (dump_file, &stats);
3243 /* Initializer. */
3245 static void
3246 initialize_scalar_evolutions_analyzer (void)
3248 /* The elements below are unique. */
3249 if (chrec_dont_know == NULL_TREE)
3251 chrec_not_analyzed_yet = NULL_TREE;
3252 chrec_dont_know = make_node (SCEV_NOT_KNOWN);
3253 chrec_known = make_node (SCEV_KNOWN);
3254 TREE_TYPE (chrec_dont_know) = void_type_node;
3255 TREE_TYPE (chrec_known) = void_type_node;
3259 /* Initialize the analysis of scalar evolutions for LOOPS. */
3261 void
3262 scev_initialize (void)
3264 struct loop *loop;
3266 scalar_evolution_info = hash_table<scev_info_hasher>::create_ggc (100);
3268 initialize_scalar_evolutions_analyzer ();
3270 FOR_EACH_LOOP (loop, 0)
3272 loop->nb_iterations = NULL_TREE;
3276 /* Return true if SCEV is initialized. */
3278 bool
3279 scev_initialized_p (void)
3281 return scalar_evolution_info != NULL;
3284 /* Cleans up the information cached by the scalar evolutions analysis
3285 in the hash table. */
3287 void
3288 scev_reset_htab (void)
3290 if (!scalar_evolution_info)
3291 return;
3293 scalar_evolution_info->empty ();
3296 /* Cleans up the information cached by the scalar evolutions analysis
3297 in the hash table and in the loop->nb_iterations. */
3299 void
3300 scev_reset (void)
3302 struct loop *loop;
3304 scev_reset_htab ();
3306 FOR_EACH_LOOP (loop, 0)
3308 loop->nb_iterations = NULL_TREE;
3312 /* Checks whether use of OP in USE_LOOP behaves as a simple affine iv with
3313 respect to WRTO_LOOP and returns its base and step in IV if possible
3314 (see analyze_scalar_evolution_in_loop for more details on USE_LOOP
3315 and WRTO_LOOP). If ALLOW_NONCONSTANT_STEP is true, we want step to be
3316 invariant in LOOP. Otherwise we require it to be an integer constant.
3318 IV->no_overflow is set to true if we are sure the iv cannot overflow (e.g.
3319 because it is computed in signed arithmetics). Consequently, adding an
3320 induction variable
3322 for (i = IV->base; ; i += IV->step)
3324 is only safe if IV->no_overflow is false, or TYPE_OVERFLOW_UNDEFINED is
3325 false for the type of the induction variable, or you can prove that i does
3326 not wrap by some other argument. Otherwise, this might introduce undefined
3327 behavior, and
3329 for (i = iv->base; ; i = (type) ((unsigned type) i + (unsigned type) iv->step))
3331 must be used instead. */
3333 bool
3334 simple_iv (struct loop *wrto_loop, struct loop *use_loop, tree op,
3335 affine_iv *iv, bool allow_nonconstant_step)
3337 enum tree_code code;
3338 tree type, ev, base, e, stop;
3339 wide_int extreme;
3340 bool folded_casts, overflow;
3342 iv->base = NULL_TREE;
3343 iv->step = NULL_TREE;
3344 iv->no_overflow = false;
3346 type = TREE_TYPE (op);
3347 if (!POINTER_TYPE_P (type)
3348 && !INTEGRAL_TYPE_P (type))
3349 return false;
3351 ev = analyze_scalar_evolution_in_loop (wrto_loop, use_loop, op,
3352 &folded_casts);
3353 if (chrec_contains_undetermined (ev)
3354 || chrec_contains_symbols_defined_in_loop (ev, wrto_loop->num))
3355 return false;
3357 if (tree_does_not_contain_chrecs (ev))
3359 iv->base = ev;
3360 iv->step = build_int_cst (TREE_TYPE (ev), 0);
3361 iv->no_overflow = true;
3362 return true;
3365 if (TREE_CODE (ev) != POLYNOMIAL_CHREC
3366 || CHREC_VARIABLE (ev) != (unsigned) wrto_loop->num)
3367 return false;
3369 iv->step = CHREC_RIGHT (ev);
3370 if ((!allow_nonconstant_step && TREE_CODE (iv->step) != INTEGER_CST)
3371 || tree_contains_chrecs (iv->step, NULL))
3372 return false;
3374 iv->base = CHREC_LEFT (ev);
3375 if (tree_contains_chrecs (iv->base, NULL))
3376 return false;
3378 iv->no_overflow = (!folded_casts && ANY_INTEGRAL_TYPE_P (type)
3379 && TYPE_OVERFLOW_UNDEFINED (type));
3381 /* Try to simplify iv base:
3383 (signed T) ((unsigned T)base + step) ;; TREE_TYPE (base) == signed T
3384 == (signed T)(unsigned T)base + step
3385 == base + step
3387 If we can prove operation (base + step) doesn't overflow or underflow.
3388 Specifically, we try to prove below conditions are satisfied:
3390 base <= UPPER_BOUND (type) - step ;;step > 0
3391 base >= LOWER_BOUND (type) - step ;;step < 0
3393 This is done by proving the reverse conditions are false using loop's
3394 initial conditions.
3396 The is necessary to make loop niter, or iv overflow analysis easier
3397 for below example:
3399 int foo (int *a, signed char s, signed char l)
3401 signed char i;
3402 for (i = s; i < l; i++)
3403 a[i] = 0;
3404 return 0;
3407 Note variable I is firstly converted to type unsigned char, incremented,
3408 then converted back to type signed char. */
3410 if (wrto_loop->num != use_loop->num)
3411 return true;
3413 if (!CONVERT_EXPR_P (iv->base) || TREE_CODE (iv->step) != INTEGER_CST)
3414 return true;
3416 type = TREE_TYPE (iv->base);
3417 e = TREE_OPERAND (iv->base, 0);
3418 if (TREE_CODE (e) != PLUS_EXPR
3419 || TREE_CODE (TREE_OPERAND (e, 1)) != INTEGER_CST
3420 || !tree_int_cst_equal (iv->step,
3421 fold_convert (type, TREE_OPERAND (e, 1))))
3422 return true;
3423 e = TREE_OPERAND (e, 0);
3424 if (!CONVERT_EXPR_P (e))
3425 return true;
3426 base = TREE_OPERAND (e, 0);
3427 if (!useless_type_conversion_p (type, TREE_TYPE (base)))
3428 return true;
3430 if (tree_int_cst_sign_bit (iv->step))
3432 code = LT_EXPR;
3433 extreme = wi::min_value (type);
3435 else
3437 code = GT_EXPR;
3438 extreme = wi::max_value (type);
3440 overflow = false;
3441 extreme = wi::sub (extreme, iv->step, TYPE_SIGN (type), &overflow);
3442 if (overflow)
3443 return true;
3444 e = fold_build2 (code, boolean_type_node, base,
3445 wide_int_to_tree (type, extreme));
3446 stop = (TREE_CODE (base) == SSA_NAME) ? base : NULL;
3447 e = simplify_using_initial_conditions (use_loop, e, stop);
3448 if (!integer_zerop (e))
3449 return true;
3451 if (POINTER_TYPE_P (TREE_TYPE (base)))
3452 code = POINTER_PLUS_EXPR;
3453 else
3454 code = PLUS_EXPR;
3456 iv->base = fold_build2 (code, TREE_TYPE (base), base, iv->step);
3457 return true;
3460 /* Finalize the scalar evolution analysis. */
3462 void
3463 scev_finalize (void)
3465 if (!scalar_evolution_info)
3466 return;
3467 scalar_evolution_info->empty ();
3468 scalar_evolution_info = NULL;
3471 /* Returns true if the expression EXPR is considered to be too expensive
3472 for scev_const_prop. */
3474 bool
3475 expression_expensive_p (tree expr)
3477 enum tree_code code;
3479 if (is_gimple_val (expr))
3480 return false;
3482 code = TREE_CODE (expr);
3483 if (code == TRUNC_DIV_EXPR
3484 || code == CEIL_DIV_EXPR
3485 || code == FLOOR_DIV_EXPR
3486 || code == ROUND_DIV_EXPR
3487 || code == TRUNC_MOD_EXPR
3488 || code == CEIL_MOD_EXPR
3489 || code == FLOOR_MOD_EXPR
3490 || code == ROUND_MOD_EXPR
3491 || code == EXACT_DIV_EXPR)
3493 /* Division by power of two is usually cheap, so we allow it.
3494 Forbid anything else. */
3495 if (!integer_pow2p (TREE_OPERAND (expr, 1)))
3496 return true;
3499 switch (TREE_CODE_CLASS (code))
3501 case tcc_binary:
3502 case tcc_comparison:
3503 if (expression_expensive_p (TREE_OPERAND (expr, 1)))
3504 return true;
3506 /* Fallthru. */
3507 case tcc_unary:
3508 return expression_expensive_p (TREE_OPERAND (expr, 0));
3510 default:
3511 return true;
3515 /* Do final value replacement for LOOP. */
3517 void
3518 final_value_replacement_loop (struct loop *loop)
3520 /* If we do not know exact number of iterations of the loop, we cannot
3521 replace the final value. */
3522 edge exit = single_exit (loop);
3523 if (!exit)
3524 return;
3526 tree niter = number_of_latch_executions (loop);
3527 if (niter == chrec_dont_know)
3528 return;
3530 /* Ensure that it is possible to insert new statements somewhere. */
3531 if (!single_pred_p (exit->dest))
3532 split_loop_exit_edge (exit);
3534 /* Set stmt insertion pointer. All stmts are inserted before this point. */
3535 gimple_stmt_iterator gsi = gsi_after_labels (exit->dest);
3537 struct loop *ex_loop
3538 = superloop_at_depth (loop,
3539 loop_depth (exit->dest->loop_father) + 1);
3541 gphi_iterator psi;
3542 for (psi = gsi_start_phis (exit->dest); !gsi_end_p (psi); )
3544 gphi *phi = psi.phi ();
3545 tree rslt = PHI_RESULT (phi);
3546 tree def = PHI_ARG_DEF_FROM_EDGE (phi, exit);
3547 if (virtual_operand_p (def))
3549 gsi_next (&psi);
3550 continue;
3553 if (!POINTER_TYPE_P (TREE_TYPE (def))
3554 && !INTEGRAL_TYPE_P (TREE_TYPE (def)))
3556 gsi_next (&psi);
3557 continue;
3560 bool folded_casts;
3561 def = analyze_scalar_evolution_in_loop (ex_loop, loop, def,
3562 &folded_casts);
3563 def = compute_overall_effect_of_inner_loop (ex_loop, def);
3564 if (!tree_does_not_contain_chrecs (def)
3565 || chrec_contains_symbols_defined_in_loop (def, ex_loop->num)
3566 /* Moving the computation from the loop may prolong life range
3567 of some ssa names, which may cause problems if they appear
3568 on abnormal edges. */
3569 || contains_abnormal_ssa_name_p (def)
3570 /* Do not emit expensive expressions. The rationale is that
3571 when someone writes a code like
3573 while (n > 45) n -= 45;
3575 he probably knows that n is not large, and does not want it
3576 to be turned into n %= 45. */
3577 || expression_expensive_p (def))
3579 if (dump_file && (dump_flags & TDF_DETAILS))
3581 fprintf (dump_file, "not replacing:\n ");
3582 print_gimple_stmt (dump_file, phi, 0, 0);
3583 fprintf (dump_file, "\n");
3585 gsi_next (&psi);
3586 continue;
3589 /* Eliminate the PHI node and replace it by a computation outside
3590 the loop. */
3591 if (dump_file)
3593 fprintf (dump_file, "\nfinal value replacement:\n ");
3594 print_gimple_stmt (dump_file, phi, 0, 0);
3595 fprintf (dump_file, " with\n ");
3597 def = unshare_expr (def);
3598 remove_phi_node (&psi, false);
3600 /* If def's type has undefined overflow and there were folded
3601 casts, rewrite all stmts added for def into arithmetics
3602 with defined overflow behavior. */
3603 if (folded_casts && ANY_INTEGRAL_TYPE_P (TREE_TYPE (def))
3604 && TYPE_OVERFLOW_UNDEFINED (TREE_TYPE (def)))
3606 gimple_seq stmts;
3607 gimple_stmt_iterator gsi2;
3608 def = force_gimple_operand (def, &stmts, true, NULL_TREE);
3609 gsi2 = gsi_start (stmts);
3610 while (!gsi_end_p (gsi2))
3612 gimple *stmt = gsi_stmt (gsi2);
3613 gimple_stmt_iterator gsi3 = gsi2;
3614 gsi_next (&gsi2);
3615 gsi_remove (&gsi3, false);
3616 if (is_gimple_assign (stmt)
3617 && arith_code_with_undefined_signed_overflow
3618 (gimple_assign_rhs_code (stmt)))
3619 gsi_insert_seq_before (&gsi,
3620 rewrite_to_defined_overflow (stmt),
3621 GSI_SAME_STMT);
3622 else
3623 gsi_insert_before (&gsi, stmt, GSI_SAME_STMT);
3626 else
3627 def = force_gimple_operand_gsi (&gsi, def, false, NULL_TREE,
3628 true, GSI_SAME_STMT);
3630 gassign *ass = gimple_build_assign (rslt, def);
3631 gsi_insert_before (&gsi, ass, GSI_SAME_STMT);
3632 if (dump_file)
3634 print_gimple_stmt (dump_file, ass, 0, 0);
3635 fprintf (dump_file, "\n");
3640 /* Replace ssa names for that scev can prove they are constant by the
3641 appropriate constants. Also perform final value replacement in loops,
3642 in case the replacement expressions are cheap.
3644 We only consider SSA names defined by phi nodes; rest is left to the
3645 ordinary constant propagation pass. */
3647 unsigned int
3648 scev_const_prop (void)
3650 basic_block bb;
3651 tree name, type, ev;
3652 gphi *phi;
3653 struct loop *loop;
3654 bitmap ssa_names_to_remove = NULL;
3655 unsigned i;
3656 gphi_iterator psi;
3658 if (number_of_loops (cfun) <= 1)
3659 return 0;
3661 FOR_EACH_BB_FN (bb, cfun)
3663 loop = bb->loop_father;
3665 for (psi = gsi_start_phis (bb); !gsi_end_p (psi); gsi_next (&psi))
3667 phi = psi.phi ();
3668 name = PHI_RESULT (phi);
3670 if (virtual_operand_p (name))
3671 continue;
3673 type = TREE_TYPE (name);
3675 if (!POINTER_TYPE_P (type)
3676 && !INTEGRAL_TYPE_P (type))
3677 continue;
3679 ev = resolve_mixers (loop, analyze_scalar_evolution (loop, name),
3680 NULL);
3681 if (!is_gimple_min_invariant (ev)
3682 || !may_propagate_copy (name, ev))
3683 continue;
3685 /* Replace the uses of the name. */
3686 if (name != ev)
3688 if (dump_file && (dump_flags & TDF_DETAILS))
3690 fprintf (dump_file, "Replacing uses of: ");
3691 print_generic_expr (dump_file, name, 0);
3692 fprintf (dump_file, " with: ");
3693 print_generic_expr (dump_file, ev, 0);
3694 fprintf (dump_file, "\n");
3696 replace_uses_by (name, ev);
3699 if (!ssa_names_to_remove)
3700 ssa_names_to_remove = BITMAP_ALLOC (NULL);
3701 bitmap_set_bit (ssa_names_to_remove, SSA_NAME_VERSION (name));
3705 /* Remove the ssa names that were replaced by constants. We do not
3706 remove them directly in the previous cycle, since this
3707 invalidates scev cache. */
3708 if (ssa_names_to_remove)
3710 bitmap_iterator bi;
3712 EXECUTE_IF_SET_IN_BITMAP (ssa_names_to_remove, 0, i, bi)
3714 gimple_stmt_iterator psi;
3715 name = ssa_name (i);
3716 phi = as_a <gphi *> (SSA_NAME_DEF_STMT (name));
3718 gcc_assert (gimple_code (phi) == GIMPLE_PHI);
3719 psi = gsi_for_stmt (phi);
3720 remove_phi_node (&psi, true);
3723 BITMAP_FREE (ssa_names_to_remove);
3724 scev_reset ();
3727 /* Now the regular final value replacement. */
3728 FOR_EACH_LOOP (loop, LI_FROM_INNERMOST)
3729 final_value_replacement_loop (loop);
3731 return 0;
3734 #include "gt-tree-scalar-evolution.h"