* sv.po: Update.
[official-gcc.git] / gcc / tree-scalar-evolution.c
blob4ff50e618904a99b65c58ef2532931dc3c3855d3
1 /* Scalar evolution detector.
2 Copyright (C) 2003, 2004, 2005 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 2, 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 COPYING. If not, write to the Free
19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
20 02111-1307, USA. */
22 /*
23 Description:
25 This pass analyzes the evolution of scalar variables in loop
26 structures. The algorithm is based on the SSA representation,
27 and on the loop hierarchy tree. This algorithm is not based on
28 the notion of versions of a variable, as it was the case for the
29 previous implementations of the scalar evolution algorithm, but
30 it assumes that each defined name is unique.
32 The notation used in this file is called "chains of recurrences",
33 and has been proposed by Eugene Zima, Robert Van Engelen, and
34 others for describing induction variables in programs. For example
35 "b -> {0, +, 2}_1" means that the scalar variable "b" is equal to 0
36 when entering in the loop_1 and has a step 2 in this loop, in other
37 words "for (b = 0; b < N; b+=2);". Note that the coefficients of
38 this chain of recurrence (or chrec [shrek]) can contain the name of
39 other variables, in which case they are called parametric chrecs.
40 For example, "b -> {a, +, 2}_1" means that the initial value of "b"
41 is the value of "a". In most of the cases these parametric chrecs
42 are fully instantiated before their use because symbolic names can
43 hide some difficult cases such as self-references described later
44 (see the Fibonacci example).
46 A short sketch of the algorithm is:
48 Given a scalar variable to be analyzed, follow the SSA edge to
49 its definition:
51 - When the definition is a MODIFY_EXPR: if the right hand side
52 (RHS) of the definition cannot be statically analyzed, the answer
53 of the analyzer is: "don't know".
54 Otherwise, for all the variables that are not yet analyzed in the
55 RHS, try to determine their evolution, and finally try to
56 evaluate the operation of the RHS that gives the evolution
57 function of the analyzed variable.
59 - When the definition is a condition-phi-node: determine the
60 evolution function for all the branches of the phi node, and
61 finally merge these evolutions (see chrec_merge).
63 - When the definition is a loop-phi-node: determine its initial
64 condition, that is the SSA edge defined in an outer loop, and
65 keep it symbolic. Then determine the SSA edges that are defined
66 in the body of the loop. Follow the inner edges until ending on
67 another loop-phi-node of the same analyzed loop. If the reached
68 loop-phi-node is not the starting loop-phi-node, then we keep
69 this definition under a symbolic form. If the reached
70 loop-phi-node is the same as the starting one, then we compute a
71 symbolic stride on the return path. The result is then the
72 symbolic chrec {initial_condition, +, symbolic_stride}_loop.
74 Examples:
76 Example 1: Illustration of the basic algorithm.
78 | a = 3
79 | loop_1
80 | b = phi (a, c)
81 | c = b + 1
82 | if (c > 10) exit_loop
83 | endloop
85 Suppose that we want to know the number of iterations of the
86 loop_1. The exit_loop is controlled by a COND_EXPR (c > 10). We
87 ask the scalar evolution analyzer two questions: what's the
88 scalar evolution (scev) of "c", and what's the scev of "10". For
89 "10" the answer is "10" since it is a scalar constant. For the
90 scalar variable "c", it follows the SSA edge to its definition,
91 "c = b + 1", and then asks again what's the scev of "b".
92 Following the SSA edge, we end on a loop-phi-node "b = phi (a,
93 c)", where the initial condition is "a", and the inner loop edge
94 is "c". The initial condition is kept under a symbolic form (it
95 may be the case that the copy constant propagation has done its
96 work and we end with the constant "3" as one of the edges of the
97 loop-phi-node). The update edge is followed to the end of the
98 loop, and until reaching again the starting loop-phi-node: b -> c
99 -> b. At this point we have drawn a path from "b" to "b" from
100 which we compute the stride in the loop: in this example it is
101 "+1". The resulting scev for "b" is "b -> {a, +, 1}_1". Now
102 that the scev for "b" is known, it is possible to compute the
103 scev for "c", that is "c -> {a + 1, +, 1}_1". In order to
104 determine the number of iterations in the loop_1, we have to
105 instantiate_parameters ({a + 1, +, 1}_1), that gives after some
106 more analysis the scev {4, +, 1}_1, or in other words, this is
107 the function "f (x) = x + 4", where x is the iteration count of
108 the loop_1. Now we have to solve the inequality "x + 4 > 10",
109 and take the smallest iteration number for which the loop is
110 exited: x = 7. This loop runs from x = 0 to x = 7, and in total
111 there are 8 iterations. In terms of loop normalization, we have
112 created a variable that is implicitly defined, "x" or just "_1",
113 and all the other analyzed scalars of the loop are defined in
114 function of this variable:
116 a -> 3
117 b -> {3, +, 1}_1
118 c -> {4, +, 1}_1
120 or in terms of a C program:
122 | a = 3
123 | for (x = 0; x <= 7; x++)
125 | b = x + 3
126 | c = x + 4
129 Example 2: Illustration of the algorithm on nested loops.
131 | loop_1
132 | a = phi (1, b)
133 | c = a + 2
134 | loop_2 10 times
135 | b = phi (c, d)
136 | d = b + 3
137 | endloop
138 | endloop
140 For analyzing the scalar evolution of "a", the algorithm follows
141 the SSA edge into the loop's body: "a -> b". "b" is an inner
142 loop-phi-node, and its analysis as in Example 1, gives:
144 b -> {c, +, 3}_2
145 d -> {c + 3, +, 3}_2
147 Following the SSA edge for the initial condition, we end on "c = a
148 + 2", and then on the starting loop-phi-node "a". From this point,
149 the loop stride is computed: back on "c = a + 2" we get a "+2" in
150 the loop_1, then on the loop-phi-node "b" we compute the overall
151 effect of the inner loop that is "b = c + 30", and we get a "+30"
152 in the loop_1. That means that the overall stride in loop_1 is
153 equal to "+32", and the result is:
155 a -> {1, +, 32}_1
156 c -> {3, +, 32}_1
158 Example 3: Higher degree polynomials.
160 | loop_1
161 | a = phi (2, b)
162 | c = phi (5, d)
163 | b = a + 1
164 | d = c + a
165 | endloop
167 a -> {2, +, 1}_1
168 b -> {3, +, 1}_1
169 c -> {5, +, a}_1
170 d -> {5 + a, +, a}_1
172 instantiate_parameters ({5, +, a}_1) -> {5, +, 2, +, 1}_1
173 instantiate_parameters ({5 + a, +, a}_1) -> {7, +, 3, +, 1}_1
175 Example 4: Lucas, Fibonacci, or mixers in general.
177 | loop_1
178 | a = phi (1, b)
179 | c = phi (3, d)
180 | b = c
181 | d = c + a
182 | endloop
184 a -> (1, c)_1
185 c -> {3, +, a}_1
187 The syntax "(1, c)_1" stands for a PEELED_CHREC that has the
188 following semantics: during the first iteration of the loop_1, the
189 variable contains the value 1, and then it contains the value "c".
190 Note that this syntax is close to the syntax of the loop-phi-node:
191 "a -> (1, c)_1" vs. "a = phi (1, c)".
193 The symbolic chrec representation contains all the semantics of the
194 original code. What is more difficult is to use this information.
196 Example 5: Flip-flops, or exchangers.
198 | loop_1
199 | a = phi (1, b)
200 | c = phi (3, d)
201 | b = c
202 | d = a
203 | endloop
205 a -> (1, c)_1
206 c -> (3, a)_1
208 Based on these symbolic chrecs, it is possible to refine this
209 information into the more precise PERIODIC_CHRECs:
211 a -> |1, 3|_1
212 c -> |3, 1|_1
214 This transformation is not yet implemented.
216 Further readings:
218 You can find a more detailed description of the algorithm in:
219 http://icps.u-strasbg.fr/~pop/DEA_03_Pop.pdf
220 http://icps.u-strasbg.fr/~pop/DEA_03_Pop.ps.gz. But note that
221 this is a preliminary report and some of the details of the
222 algorithm have changed. I'm working on a research report that
223 updates the description of the algorithms to reflect the design
224 choices used in this implementation.
226 A set of slides show a high level overview of the algorithm and run
227 an example through the scalar evolution analyzer:
228 http://cri.ensmp.fr/~pop/gcc/mar04/slides.pdf
230 The slides that I have presented at the GCC Summit'04 are available
231 at: http://cri.ensmp.fr/~pop/gcc/20040604/gccsummit-lno-spop.pdf
234 #include "config.h"
235 #include "system.h"
236 #include "coretypes.h"
237 #include "tm.h"
238 #include "errors.h"
239 #include "ggc.h"
240 #include "tree.h"
242 /* These RTL headers are needed for basic-block.h. */
243 #include "rtl.h"
244 #include "basic-block.h"
245 #include "diagnostic.h"
246 #include "tree-flow.h"
247 #include "tree-dump.h"
248 #include "timevar.h"
249 #include "cfgloop.h"
250 #include "tree-chrec.h"
251 #include "tree-scalar-evolution.h"
252 #include "tree-pass.h"
253 #include "flags.h"
255 static tree analyze_scalar_evolution_1 (struct loop *, tree, tree);
256 static tree resolve_mixers (struct loop *, tree);
258 /* The cached information about a ssa name VAR, claiming that inside LOOP,
259 the value of VAR can be expressed as CHREC. */
261 struct scev_info_str
263 tree var;
264 tree chrec;
267 /* Counters for the scev database. */
268 static unsigned nb_set_scev = 0;
269 static unsigned nb_get_scev = 0;
271 /* The following trees are unique elements. Thus the comparison of
272 another element to these elements should be done on the pointer to
273 these trees, and not on their value. */
275 /* The SSA_NAMEs that are not yet analyzed are qualified with NULL_TREE. */
276 tree chrec_not_analyzed_yet;
278 /* Reserved to the cases where the analyzer has detected an
279 undecidable property at compile time. */
280 tree chrec_dont_know;
282 /* When the analyzer has detected that a property will never
283 happen, then it qualifies it with chrec_known. */
284 tree chrec_known;
286 static bitmap already_instantiated;
288 static htab_t scalar_evolution_info;
291 /* Constructs a new SCEV_INFO_STR structure. */
293 static inline struct scev_info_str *
294 new_scev_info_str (tree var)
296 struct scev_info_str *res;
298 res = xmalloc (sizeof (struct scev_info_str));
299 res->var = var;
300 res->chrec = chrec_not_analyzed_yet;
302 return res;
305 /* Computes a hash function for database element ELT. */
307 static hashval_t
308 hash_scev_info (const void *elt)
310 return SSA_NAME_VERSION (((struct scev_info_str *) elt)->var);
313 /* Compares database elements E1 and E2. */
315 static int
316 eq_scev_info (const void *e1, const void *e2)
318 const struct scev_info_str *elt1 = e1;
319 const struct scev_info_str *elt2 = e2;
321 return elt1->var == elt2->var;
324 /* Deletes database element E. */
326 static void
327 del_scev_info (void *e)
329 free (e);
332 /* Get the index corresponding to VAR in the current LOOP. If
333 it's the first time we ask for this VAR, then we return
334 chrec_not_analyzed_yet for this VAR and return its index. */
336 static tree *
337 find_var_scev_info (tree var)
339 struct scev_info_str *res;
340 struct scev_info_str tmp;
341 PTR *slot;
343 tmp.var = var;
344 slot = htab_find_slot (scalar_evolution_info, &tmp, INSERT);
346 if (!*slot)
347 *slot = new_scev_info_str (var);
348 res = *slot;
350 return &res->chrec;
353 /* Tries to express CHREC in wider type TYPE. */
355 tree
356 count_ev_in_wider_type (tree type, tree chrec)
358 tree base, step;
359 struct loop *loop;
361 if (!evolution_function_is_affine_p (chrec))
362 return fold_convert (type, chrec);
364 base = CHREC_LEFT (chrec);
365 step = CHREC_RIGHT (chrec);
366 loop = current_loops->parray[CHREC_VARIABLE (chrec)];
368 /* TODO -- if we knew the statement at that the conversion occurs,
369 we could pass it to can_count_iv_in_wider_type and get a better
370 result. */
371 step = can_count_iv_in_wider_type (loop, type, base, step, NULL_TREE);
372 if (!step)
373 return fold_convert (type, chrec);
374 base = chrec_convert (type, base);
376 return build_polynomial_chrec (CHREC_VARIABLE (chrec),
377 base, step);
380 /* Return true when CHREC contains symbolic names defined in
381 LOOP_NB. */
383 bool
384 chrec_contains_symbols_defined_in_loop (tree chrec, unsigned loop_nb)
386 if (chrec == NULL_TREE)
387 return false;
389 if (TREE_INVARIANT (chrec))
390 return false;
392 if (TREE_CODE (chrec) == VAR_DECL
393 || TREE_CODE (chrec) == PARM_DECL
394 || TREE_CODE (chrec) == FUNCTION_DECL
395 || TREE_CODE (chrec) == LABEL_DECL
396 || TREE_CODE (chrec) == RESULT_DECL
397 || TREE_CODE (chrec) == FIELD_DECL)
398 return true;
400 if (TREE_CODE (chrec) == SSA_NAME)
402 tree def = SSA_NAME_DEF_STMT (chrec);
403 struct loop *def_loop = loop_containing_stmt (def);
404 struct loop *loop = current_loops->parray[loop_nb];
406 if (def_loop == NULL)
407 return false;
409 if (loop == def_loop || flow_loop_nested_p (loop, def_loop))
410 return true;
412 return false;
415 switch (TREE_CODE_LENGTH (TREE_CODE (chrec)))
417 case 3:
418 if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec, 2),
419 loop_nb))
420 return true;
422 case 2:
423 if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec, 1),
424 loop_nb))
425 return true;
427 case 1:
428 if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec, 0),
429 loop_nb))
430 return true;
432 default:
433 return false;
437 /* Return true when PHI is a loop-phi-node. */
439 static bool
440 loop_phi_node_p (tree phi)
442 /* The implementation of this function is based on the following
443 property: "all the loop-phi-nodes of a loop are contained in the
444 loop's header basic block". */
446 return loop_containing_stmt (phi)->header == bb_for_stmt (phi);
449 /* Compute the scalar evolution for EVOLUTION_FN after crossing LOOP.
450 In general, in the case of multivariate evolutions we want to get
451 the evolution in different loops. LOOP specifies the level for
452 which to get the evolution.
454 Example:
456 | for (j = 0; j < 100; j++)
458 | for (k = 0; k < 100; k++)
460 | i = k + j; - Here the value of i is a function of j, k.
462 | ... = i - Here the value of i is a function of j.
464 | ... = i - Here the value of i is a scalar.
466 Example:
468 | i_0 = ...
469 | loop_1 10 times
470 | i_1 = phi (i_0, i_2)
471 | i_2 = i_1 + 2
472 | endloop
474 This loop has the same effect as:
475 LOOP_1 has the same effect as:
477 | i_1 = i_0 + 20
479 The overall effect of the loop, "i_0 + 20" in the previous example,
480 is obtained by passing in the parameters: LOOP = 1,
481 EVOLUTION_FN = {i_0, +, 2}_1.
484 static tree
485 compute_overall_effect_of_inner_loop (struct loop *loop, tree evolution_fn)
487 bool val = false;
489 if (evolution_fn == chrec_dont_know)
490 return chrec_dont_know;
492 else if (TREE_CODE (evolution_fn) == POLYNOMIAL_CHREC)
494 if (CHREC_VARIABLE (evolution_fn) >= (unsigned) loop->num)
496 struct loop *inner_loop =
497 current_loops->parray[CHREC_VARIABLE (evolution_fn)];
498 tree nb_iter = number_of_iterations_in_loop (inner_loop);
500 if (nb_iter == chrec_dont_know)
501 return chrec_dont_know;
502 else
504 tree res;
506 /* Number of iterations is off by one (the ssa name we
507 analyze must be defined before the exit). */
508 nb_iter = chrec_fold_minus (chrec_type (nb_iter),
509 nb_iter,
510 build_int_cst_type (chrec_type (nb_iter), 1));
512 /* evolution_fn is the evolution function in LOOP. Get
513 its value in the nb_iter-th iteration. */
514 res = chrec_apply (inner_loop->num, evolution_fn, nb_iter);
516 /* Continue the computation until ending on a parent of LOOP. */
517 return compute_overall_effect_of_inner_loop (loop, res);
520 else
521 return evolution_fn;
524 /* If the evolution function is an invariant, there is nothing to do. */
525 else if (no_evolution_in_loop_p (evolution_fn, loop->num, &val) && val)
526 return evolution_fn;
528 else
529 return chrec_dont_know;
532 /* Determine whether the CHREC is always positive/negative. If the expression
533 cannot be statically analyzed, return false, otherwise set the answer into
534 VALUE. */
536 bool
537 chrec_is_positive (tree chrec, bool *value)
539 bool value0, value1;
540 bool value2;
541 tree end_value;
542 tree nb_iter;
544 switch (TREE_CODE (chrec))
546 case POLYNOMIAL_CHREC:
547 if (!chrec_is_positive (CHREC_LEFT (chrec), &value0)
548 || !chrec_is_positive (CHREC_RIGHT (chrec), &value1))
549 return false;
551 /* FIXME -- overflows. */
552 if (value0 == value1)
554 *value = value0;
555 return true;
558 /* Otherwise the chrec is under the form: "{-197, +, 2}_1",
559 and the proof consists in showing that the sign never
560 changes during the execution of the loop, from 0 to
561 loop->nb_iterations. */
562 if (!evolution_function_is_affine_p (chrec))
563 return false;
565 nb_iter = number_of_iterations_in_loop
566 (current_loops->parray[CHREC_VARIABLE (chrec)]);
568 if (chrec_contains_undetermined (nb_iter))
569 return false;
571 nb_iter = chrec_fold_minus
572 (chrec_type (nb_iter), nb_iter,
573 build_int_cst (chrec_type (nb_iter), 1));
575 #if 0
576 /* TODO -- If the test is after the exit, we may decrease the number of
577 iterations by one. */
578 if (after_exit)
579 nb_iter = chrec_fold_minus
580 (chrec_type (nb_iter), nb_iter,
581 build_int_cst (chrec_type (nb_iter), 1));
582 #endif
584 end_value = chrec_apply (CHREC_VARIABLE (chrec), chrec, nb_iter);
586 if (!chrec_is_positive (end_value, &value2))
587 return false;
589 *value = value0;
590 return value0 == value1;
592 case INTEGER_CST:
593 *value = (tree_int_cst_sgn (chrec) == 1);
594 return true;
596 default:
597 return false;
601 /* Associate CHREC to SCALAR. */
603 static void
604 set_scalar_evolution (tree scalar, tree chrec)
606 tree *scalar_info;
608 if (TREE_CODE (scalar) != SSA_NAME)
609 return;
611 scalar_info = find_var_scev_info (scalar);
613 if (dump_file)
615 if (dump_flags & TDF_DETAILS)
617 fprintf (dump_file, "(set_scalar_evolution \n");
618 fprintf (dump_file, " (scalar = ");
619 print_generic_expr (dump_file, scalar, 0);
620 fprintf (dump_file, ")\n (scalar_evolution = ");
621 print_generic_expr (dump_file, chrec, 0);
622 fprintf (dump_file, "))\n");
624 if (dump_flags & TDF_STATS)
625 nb_set_scev++;
628 *scalar_info = chrec;
631 /* Retrieve the chrec associated to SCALAR in the LOOP. */
633 static tree
634 get_scalar_evolution (tree scalar)
636 tree res;
638 if (dump_file)
640 if (dump_flags & TDF_DETAILS)
642 fprintf (dump_file, "(get_scalar_evolution \n");
643 fprintf (dump_file, " (scalar = ");
644 print_generic_expr (dump_file, scalar, 0);
645 fprintf (dump_file, ")\n");
647 if (dump_flags & TDF_STATS)
648 nb_get_scev++;
651 switch (TREE_CODE (scalar))
653 case SSA_NAME:
654 res = *find_var_scev_info (scalar);
655 break;
657 case REAL_CST:
658 case INTEGER_CST:
659 res = scalar;
660 break;
662 default:
663 res = chrec_not_analyzed_yet;
664 break;
667 if (dump_file && (dump_flags & TDF_DETAILS))
669 fprintf (dump_file, " (scalar_evolution = ");
670 print_generic_expr (dump_file, res, 0);
671 fprintf (dump_file, "))\n");
674 return res;
677 /* Helper function for add_to_evolution. Returns the evolution
678 function for an assignment of the form "a = b + c", where "a" and
679 "b" are on the strongly connected component. CHREC_BEFORE is the
680 information that we already have collected up to this point.
681 TO_ADD is the evolution of "c".
683 When CHREC_BEFORE has an evolution part in LOOP_NB, add to this
684 evolution the expression TO_ADD, otherwise construct an evolution
685 part for this loop. */
687 static tree
688 add_to_evolution_1 (unsigned loop_nb,
689 tree chrec_before,
690 tree to_add)
692 switch (TREE_CODE (chrec_before))
694 case POLYNOMIAL_CHREC:
695 if (CHREC_VARIABLE (chrec_before) <= loop_nb)
697 unsigned var;
698 tree left, right;
699 tree type = chrec_type (chrec_before);
701 /* When there is no evolution part in this loop, build it. */
702 if (CHREC_VARIABLE (chrec_before) < loop_nb)
704 var = loop_nb;
705 left = chrec_before;
706 right = build_int_cst (type, 0);
708 else
710 var = CHREC_VARIABLE (chrec_before);
711 left = CHREC_LEFT (chrec_before);
712 right = CHREC_RIGHT (chrec_before);
715 return build_polynomial_chrec
716 (var, left, chrec_fold_plus (type, right, to_add));
718 else
719 /* Search the evolution in LOOP_NB. */
720 return build_polynomial_chrec
721 (CHREC_VARIABLE (chrec_before),
722 add_to_evolution_1 (loop_nb, CHREC_LEFT (chrec_before), to_add),
723 CHREC_RIGHT (chrec_before));
725 default:
726 /* These nodes do not depend on a loop. */
727 if (chrec_before == chrec_dont_know)
728 return chrec_dont_know;
729 return build_polynomial_chrec (loop_nb, chrec_before, to_add);
733 /* Add TO_ADD to the evolution part of CHREC_BEFORE in the dimension
734 of LOOP_NB.
736 Description (provided for completeness, for those who read code in
737 a plane, and for my poor 62 bytes brain that would have forgotten
738 all this in the next two or three months):
740 The algorithm of translation of programs from the SSA representation
741 into the chrecs syntax is based on a pattern matching. After having
742 reconstructed the overall tree expression for a loop, there are only
743 two cases that can arise:
745 1. a = loop-phi (init, a + expr)
746 2. a = loop-phi (init, expr)
748 where EXPR is either a scalar constant with respect to the analyzed
749 loop (this is a degree 0 polynomial), or an expression containing
750 other loop-phi definitions (these are higher degree polynomials).
752 Examples:
755 | init = ...
756 | loop_1
757 | a = phi (init, a + 5)
758 | endloop
761 | inita = ...
762 | initb = ...
763 | loop_1
764 | a = phi (inita, 2 * b + 3)
765 | b = phi (initb, b + 1)
766 | endloop
768 For the first case, the semantics of the SSA representation is:
770 | a (x) = init + \sum_{j = 0}^{x - 1} expr (j)
772 that is, there is a loop index "x" that determines the scalar value
773 of the variable during the loop execution. During the first
774 iteration, the value is that of the initial condition INIT, while
775 during the subsequent iterations, it is the sum of the initial
776 condition with the sum of all the values of EXPR from the initial
777 iteration to the before last considered iteration.
779 For the second case, the semantics of the SSA program is:
781 | a (x) = init, if x = 0;
782 | expr (x - 1), otherwise.
784 The second case corresponds to the PEELED_CHREC, whose syntax is
785 close to the syntax of a loop-phi-node:
787 | phi (init, expr) vs. (init, expr)_x
789 The proof of the translation algorithm for the first case is a
790 proof by structural induction based on the degree of EXPR.
792 Degree 0:
793 When EXPR is a constant with respect to the analyzed loop, or in
794 other words when EXPR is a polynomial of degree 0, the evolution of
795 the variable A in the loop is an affine function with an initial
796 condition INIT, and a step EXPR. In order to show this, we start
797 from the semantics of the SSA representation:
799 f (x) = init + \sum_{j = 0}^{x - 1} expr (j)
801 and since "expr (j)" is a constant with respect to "j",
803 f (x) = init + x * expr
805 Finally, based on the semantics of the pure sum chrecs, by
806 identification we get the corresponding chrecs syntax:
808 f (x) = init * \binom{x}{0} + expr * \binom{x}{1}
809 f (x) -> {init, +, expr}_x
811 Higher degree:
812 Suppose that EXPR is a polynomial of degree N with respect to the
813 analyzed loop_x for which we have already determined that it is
814 written under the chrecs syntax:
816 | expr (x) -> {b_0, +, b_1, +, ..., +, b_{n-1}} (x)
818 We start from the semantics of the SSA program:
820 | f (x) = init + \sum_{j = 0}^{x - 1} expr (j)
822 | f (x) = init + \sum_{j = 0}^{x - 1}
823 | (b_0 * \binom{j}{0} + ... + b_{n-1} * \binom{j}{n-1})
825 | f (x) = init + \sum_{j = 0}^{x - 1}
826 | \sum_{k = 0}^{n - 1} (b_k * \binom{j}{k})
828 | f (x) = init + \sum_{k = 0}^{n - 1}
829 | (b_k * \sum_{j = 0}^{x - 1} \binom{j}{k})
831 | f (x) = init + \sum_{k = 0}^{n - 1}
832 | (b_k * \binom{x}{k + 1})
834 | f (x) = init + b_0 * \binom{x}{1} + ...
835 | + b_{n-1} * \binom{x}{n}
837 | f (x) = init * \binom{x}{0} + b_0 * \binom{x}{1} + ...
838 | + b_{n-1} * \binom{x}{n}
841 And finally from the definition of the chrecs syntax, we identify:
842 | f (x) -> {init, +, b_0, +, ..., +, b_{n-1}}_x
844 This shows the mechanism that stands behind the add_to_evolution
845 function. An important point is that the use of symbolic
846 parameters avoids the need of an analysis schedule.
848 Example:
850 | inita = ...
851 | initb = ...
852 | loop_1
853 | a = phi (inita, a + 2 + b)
854 | b = phi (initb, b + 1)
855 | endloop
857 When analyzing "a", the algorithm keeps "b" symbolically:
859 | a -> {inita, +, 2 + b}_1
861 Then, after instantiation, the analyzer ends on the evolution:
863 | a -> {inita, +, 2 + initb, +, 1}_1
867 static tree
868 add_to_evolution (unsigned loop_nb,
869 tree chrec_before,
870 enum tree_code code,
871 tree to_add)
873 tree type = chrec_type (to_add);
874 tree res = NULL_TREE;
876 if (to_add == NULL_TREE)
877 return chrec_before;
879 /* TO_ADD is either a scalar, or a parameter. TO_ADD is not
880 instantiated at this point. */
881 if (TREE_CODE (to_add) == POLYNOMIAL_CHREC)
882 /* This should not happen. */
883 return chrec_dont_know;
885 if (dump_file && (dump_flags & TDF_DETAILS))
887 fprintf (dump_file, "(add_to_evolution \n");
888 fprintf (dump_file, " (loop_nb = %d)\n", loop_nb);
889 fprintf (dump_file, " (chrec_before = ");
890 print_generic_expr (dump_file, chrec_before, 0);
891 fprintf (dump_file, ")\n (to_add = ");
892 print_generic_expr (dump_file, to_add, 0);
893 fprintf (dump_file, ")\n");
896 if (code == MINUS_EXPR)
897 to_add = chrec_fold_multiply (type, to_add,
898 build_int_cst_type (type, -1));
900 res = add_to_evolution_1 (loop_nb, chrec_before, to_add);
902 if (dump_file && (dump_flags & TDF_DETAILS))
904 fprintf (dump_file, " (res = ");
905 print_generic_expr (dump_file, res, 0);
906 fprintf (dump_file, "))\n");
909 return res;
912 /* Helper function. */
914 static inline tree
915 set_nb_iterations_in_loop (struct loop *loop,
916 tree res)
918 res = chrec_fold_plus (chrec_type (res), res,
919 build_int_cst_type (chrec_type (res), 1));
921 /* FIXME HWI: However we want to store one iteration less than the
922 count of the loop in order to be compatible with the other
923 nb_iter computations in loop-iv. This also allows the
924 representation of nb_iters that are equal to MAX_INT. */
925 if (TREE_CODE (res) == INTEGER_CST
926 && (TREE_INT_CST_LOW (res) == 0
927 || TREE_OVERFLOW (res)))
928 res = chrec_dont_know;
930 if (dump_file && (dump_flags & TDF_DETAILS))
932 fprintf (dump_file, " (set_nb_iterations_in_loop = ");
933 print_generic_expr (dump_file, res, 0);
934 fprintf (dump_file, "))\n");
937 loop->nb_iterations = res;
938 return res;
943 /* This section selects the loops that will be good candidates for the
944 scalar evolution analysis. For the moment, greedily select all the
945 loop nests we could analyze. */
947 /* Return true when it is possible to analyze the condition expression
948 EXPR. */
950 static bool
951 analyzable_condition (tree expr)
953 tree condition;
955 if (TREE_CODE (expr) != COND_EXPR)
956 return false;
958 condition = TREE_OPERAND (expr, 0);
960 switch (TREE_CODE (condition))
962 case SSA_NAME:
963 return true;
965 case LT_EXPR:
966 case LE_EXPR:
967 case GT_EXPR:
968 case GE_EXPR:
969 case EQ_EXPR:
970 case NE_EXPR:
971 return true;
973 default:
974 return false;
977 return false;
980 /* For a loop with a single exit edge, return the COND_EXPR that
981 guards the exit edge. If the expression is too difficult to
982 analyze, then give up. */
984 tree
985 get_loop_exit_condition (struct loop *loop)
987 tree res = NULL_TREE;
988 edge exit_edge = loop->single_exit;
991 if (dump_file && (dump_flags & TDF_DETAILS))
992 fprintf (dump_file, "(get_loop_exit_condition \n ");
994 if (exit_edge)
996 tree expr;
998 expr = last_stmt (exit_edge->src);
999 if (analyzable_condition (expr))
1000 res = expr;
1003 if (dump_file && (dump_flags & TDF_DETAILS))
1005 print_generic_expr (dump_file, res, 0);
1006 fprintf (dump_file, ")\n");
1009 return res;
1012 /* Recursively determine and enqueue the exit conditions for a loop. */
1014 static void
1015 get_exit_conditions_rec (struct loop *loop,
1016 VEC(tree,heap) **exit_conditions)
1018 if (!loop)
1019 return;
1021 /* Recurse on the inner loops, then on the next (sibling) loops. */
1022 get_exit_conditions_rec (loop->inner, exit_conditions);
1023 get_exit_conditions_rec (loop->next, exit_conditions);
1025 if (loop->single_exit)
1027 tree loop_condition = get_loop_exit_condition (loop);
1029 if (loop_condition)
1030 VEC_safe_push (tree, heap, *exit_conditions, loop_condition);
1034 /* Select the candidate loop nests for the analysis. This function
1035 initializes the EXIT_CONDITIONS array. */
1037 static void
1038 select_loops_exit_conditions (struct loops *loops,
1039 VEC(tree,heap) **exit_conditions)
1041 struct loop *function_body = loops->parray[0];
1043 get_exit_conditions_rec (function_body->inner, exit_conditions);
1047 /* Depth first search algorithm. */
1049 static bool follow_ssa_edge (struct loop *loop, tree, tree, tree *);
1051 /* Follow the ssa edge into the right hand side RHS of an assignment.
1052 Return true if the strongly connected component has been found. */
1054 static bool
1055 follow_ssa_edge_in_rhs (struct loop *loop,
1056 tree rhs,
1057 tree halting_phi,
1058 tree *evolution_of_loop)
1060 bool res = false;
1061 tree rhs0, rhs1;
1062 tree type_rhs = TREE_TYPE (rhs);
1064 /* The RHS is one of the following cases:
1065 - an SSA_NAME,
1066 - an INTEGER_CST,
1067 - a PLUS_EXPR,
1068 - a MINUS_EXPR,
1069 - an ASSERT_EXPR,
1070 - other cases are not yet handled. */
1071 switch (TREE_CODE (rhs))
1073 case NOP_EXPR:
1074 /* This assignment is under the form "a_1 = (cast) rhs. */
1075 res = follow_ssa_edge_in_rhs (loop, TREE_OPERAND (rhs, 0), halting_phi,
1076 evolution_of_loop);
1077 *evolution_of_loop = chrec_convert (TREE_TYPE (rhs), *evolution_of_loop);
1078 break;
1080 case INTEGER_CST:
1081 /* This assignment is under the form "a_1 = 7". */
1082 res = false;
1083 break;
1085 case SSA_NAME:
1086 /* This assignment is under the form: "a_1 = b_2". */
1087 res = follow_ssa_edge
1088 (loop, SSA_NAME_DEF_STMT (rhs), halting_phi, evolution_of_loop);
1089 break;
1091 case PLUS_EXPR:
1092 /* This case is under the form "rhs0 + rhs1". */
1093 rhs0 = TREE_OPERAND (rhs, 0);
1094 rhs1 = TREE_OPERAND (rhs, 1);
1095 STRIP_TYPE_NOPS (rhs0);
1096 STRIP_TYPE_NOPS (rhs1);
1098 if (TREE_CODE (rhs0) == SSA_NAME)
1100 if (TREE_CODE (rhs1) == SSA_NAME)
1102 /* Match an assignment under the form:
1103 "a = b + c". */
1104 res = follow_ssa_edge
1105 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1106 evolution_of_loop);
1108 if (res)
1109 *evolution_of_loop = add_to_evolution
1110 (loop->num,
1111 chrec_convert (type_rhs, *evolution_of_loop),
1112 PLUS_EXPR, rhs1);
1114 else
1116 res = follow_ssa_edge
1117 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1118 evolution_of_loop);
1120 if (res)
1121 *evolution_of_loop = add_to_evolution
1122 (loop->num,
1123 chrec_convert (type_rhs, *evolution_of_loop),
1124 PLUS_EXPR, rhs0);
1128 else
1130 /* Match an assignment under the form:
1131 "a = b + ...". */
1132 res = follow_ssa_edge
1133 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1134 evolution_of_loop);
1135 if (res)
1136 *evolution_of_loop = add_to_evolution
1137 (loop->num, chrec_convert (type_rhs, *evolution_of_loop),
1138 PLUS_EXPR, rhs1);
1142 else if (TREE_CODE (rhs1) == SSA_NAME)
1144 /* Match an assignment under the form:
1145 "a = ... + c". */
1146 res = follow_ssa_edge
1147 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1148 evolution_of_loop);
1149 if (res)
1150 *evolution_of_loop = add_to_evolution
1151 (loop->num, chrec_convert (type_rhs, *evolution_of_loop),
1152 PLUS_EXPR, rhs0);
1155 else
1156 /* Otherwise, match an assignment under the form:
1157 "a = ... + ...". */
1158 /* And there is nothing to do. */
1159 res = false;
1161 break;
1163 case MINUS_EXPR:
1164 /* This case is under the form "opnd0 = rhs0 - rhs1". */
1165 rhs0 = TREE_OPERAND (rhs, 0);
1166 rhs1 = TREE_OPERAND (rhs, 1);
1167 STRIP_TYPE_NOPS (rhs0);
1168 STRIP_TYPE_NOPS (rhs1);
1170 if (TREE_CODE (rhs0) == SSA_NAME)
1172 /* Match an assignment under the form:
1173 "a = b - ...". */
1174 res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1175 evolution_of_loop);
1176 if (res)
1177 *evolution_of_loop = add_to_evolution
1178 (loop->num, chrec_convert (type_rhs, *evolution_of_loop),
1179 MINUS_EXPR, rhs1);
1181 else
1182 /* Otherwise, match an assignment under the form:
1183 "a = ... - ...". */
1184 /* And there is nothing to do. */
1185 res = false;
1187 break;
1189 case MULT_EXPR:
1190 /* This case is under the form "opnd0 = rhs0 * rhs1". */
1191 rhs0 = TREE_OPERAND (rhs, 0);
1192 rhs1 = TREE_OPERAND (rhs, 1);
1193 STRIP_TYPE_NOPS (rhs0);
1194 STRIP_TYPE_NOPS (rhs1);
1196 if (TREE_CODE (rhs0) == SSA_NAME)
1198 if (TREE_CODE (rhs1) == SSA_NAME)
1200 /* Match an assignment under the form:
1201 "a = b * c". */
1202 res = follow_ssa_edge
1203 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1204 evolution_of_loop);
1206 if (res)
1207 *evolution_of_loop = chrec_dont_know;
1209 else
1211 res = follow_ssa_edge
1212 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1213 evolution_of_loop);
1215 if (res)
1216 *evolution_of_loop = chrec_dont_know;
1220 else
1222 /* Match an assignment under the form:
1223 "a = b * ...". */
1224 res = follow_ssa_edge
1225 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1226 evolution_of_loop);
1227 if (res)
1228 *evolution_of_loop = chrec_dont_know;
1232 else if (TREE_CODE (rhs1) == SSA_NAME)
1234 /* Match an assignment under the form:
1235 "a = ... * c". */
1236 res = follow_ssa_edge
1237 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1238 evolution_of_loop);
1239 if (res)
1240 *evolution_of_loop = chrec_dont_know;
1243 else
1244 /* Otherwise, match an assignment under the form:
1245 "a = ... * ...". */
1246 /* And there is nothing to do. */
1247 res = false;
1249 break;
1251 case ASSERT_EXPR:
1253 /* This assignment is of the form: "a_1 = ASSERT_EXPR <a_2, ...>"
1254 It must be handled as a copy assignment of the form a_1 = a_2. */
1255 tree op0 = ASSERT_EXPR_VAR (rhs);
1256 if (TREE_CODE (op0) == SSA_NAME)
1257 res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (op0),
1258 halting_phi, evolution_of_loop);
1259 else
1260 res = false;
1261 break;
1265 default:
1266 res = false;
1267 break;
1270 return res;
1273 /* Checks whether the I-th argument of a PHI comes from a backedge. */
1275 static bool
1276 backedge_phi_arg_p (tree phi, int i)
1278 edge e = PHI_ARG_EDGE (phi, i);
1280 /* We would in fact like to test EDGE_DFS_BACK here, but we do not care
1281 about updating it anywhere, and this should work as well most of the
1282 time. */
1283 if (e->flags & EDGE_IRREDUCIBLE_LOOP)
1284 return true;
1286 return false;
1289 /* Helper function for one branch of the condition-phi-node. Return
1290 true if the strongly connected component has been found following
1291 this path. */
1293 static inline bool
1294 follow_ssa_edge_in_condition_phi_branch (int i,
1295 struct loop *loop,
1296 tree condition_phi,
1297 tree halting_phi,
1298 tree *evolution_of_branch,
1299 tree init_cond)
1301 tree branch = PHI_ARG_DEF (condition_phi, i);
1302 *evolution_of_branch = chrec_dont_know;
1304 /* Do not follow back edges (they must belong to an irreducible loop, which
1305 we really do not want to worry about). */
1306 if (backedge_phi_arg_p (condition_phi, i))
1307 return false;
1309 if (TREE_CODE (branch) == SSA_NAME)
1311 *evolution_of_branch = init_cond;
1312 return follow_ssa_edge (loop, SSA_NAME_DEF_STMT (branch), halting_phi,
1313 evolution_of_branch);
1316 /* This case occurs when one of the condition branches sets
1317 the variable to a constant: i.e. a phi-node like
1318 "a_2 = PHI <a_7(5), 2(6)>;".
1320 FIXME: This case have to be refined correctly:
1321 in some cases it is possible to say something better than
1322 chrec_dont_know, for example using a wrap-around notation. */
1323 return false;
1326 /* This function merges the branches of a condition-phi-node in a
1327 loop. */
1329 static bool
1330 follow_ssa_edge_in_condition_phi (struct loop *loop,
1331 tree condition_phi,
1332 tree halting_phi,
1333 tree *evolution_of_loop)
1335 int i;
1336 tree init = *evolution_of_loop;
1337 tree evolution_of_branch;
1339 if (!follow_ssa_edge_in_condition_phi_branch (0, loop, condition_phi,
1340 halting_phi,
1341 &evolution_of_branch,
1342 init))
1343 return false;
1344 *evolution_of_loop = evolution_of_branch;
1346 for (i = 1; i < PHI_NUM_ARGS (condition_phi); i++)
1348 /* Quickly give up when the evolution of one of the branches is
1349 not known. */
1350 if (*evolution_of_loop == chrec_dont_know)
1351 return true;
1353 if (!follow_ssa_edge_in_condition_phi_branch (i, loop, condition_phi,
1354 halting_phi,
1355 &evolution_of_branch,
1356 init))
1357 return false;
1359 *evolution_of_loop = chrec_merge (*evolution_of_loop,
1360 evolution_of_branch);
1363 return true;
1366 /* Follow an SSA edge in an inner loop. It computes the overall
1367 effect of the loop, and following the symbolic initial conditions,
1368 it follows the edges in the parent loop. The inner loop is
1369 considered as a single statement. */
1371 static bool
1372 follow_ssa_edge_inner_loop_phi (struct loop *outer_loop,
1373 tree loop_phi_node,
1374 tree halting_phi,
1375 tree *evolution_of_loop)
1377 struct loop *loop = loop_containing_stmt (loop_phi_node);
1378 tree ev = analyze_scalar_evolution (loop, PHI_RESULT (loop_phi_node));
1380 /* Sometimes, the inner loop is too difficult to analyze, and the
1381 result of the analysis is a symbolic parameter. */
1382 if (ev == PHI_RESULT (loop_phi_node))
1384 bool res = false;
1385 int i;
1387 for (i = 0; i < PHI_NUM_ARGS (loop_phi_node); i++)
1389 tree arg = PHI_ARG_DEF (loop_phi_node, i);
1390 basic_block bb;
1392 /* Follow the edges that exit the inner loop. */
1393 bb = PHI_ARG_EDGE (loop_phi_node, i)->src;
1394 if (!flow_bb_inside_loop_p (loop, bb))
1395 res = res || follow_ssa_edge_in_rhs (outer_loop, arg, halting_phi,
1396 evolution_of_loop);
1399 /* If the path crosses this loop-phi, give up. */
1400 if (res == true)
1401 *evolution_of_loop = chrec_dont_know;
1403 return res;
1406 /* Otherwise, compute the overall effect of the inner loop. */
1407 ev = compute_overall_effect_of_inner_loop (loop, ev);
1408 return follow_ssa_edge_in_rhs (outer_loop, ev, halting_phi,
1409 evolution_of_loop);
1412 /* Follow an SSA edge from a loop-phi-node to itself, constructing a
1413 path that is analyzed on the return walk. */
1415 static bool
1416 follow_ssa_edge (struct loop *loop,
1417 tree def,
1418 tree halting_phi,
1419 tree *evolution_of_loop)
1421 struct loop *def_loop;
1423 if (TREE_CODE (def) == NOP_EXPR)
1424 return false;
1426 def_loop = loop_containing_stmt (def);
1428 switch (TREE_CODE (def))
1430 case PHI_NODE:
1431 if (!loop_phi_node_p (def))
1432 /* DEF is a condition-phi-node. Follow the branches, and
1433 record their evolutions. Finally, merge the collected
1434 information and set the approximation to the main
1435 variable. */
1436 return follow_ssa_edge_in_condition_phi
1437 (loop, def, halting_phi, evolution_of_loop);
1439 /* When the analyzed phi is the halting_phi, the
1440 depth-first search is over: we have found a path from
1441 the halting_phi to itself in the loop. */
1442 if (def == halting_phi)
1443 return true;
1445 /* Otherwise, the evolution of the HALTING_PHI depends
1446 on the evolution of another loop-phi-node, i.e. the
1447 evolution function is a higher degree polynomial. */
1448 if (def_loop == loop)
1449 return false;
1451 /* Inner loop. */
1452 if (flow_loop_nested_p (loop, def_loop))
1453 return follow_ssa_edge_inner_loop_phi
1454 (loop, def, halting_phi, evolution_of_loop);
1456 /* Outer loop. */
1457 return false;
1459 case MODIFY_EXPR:
1460 return follow_ssa_edge_in_rhs (loop,
1461 TREE_OPERAND (def, 1),
1462 halting_phi,
1463 evolution_of_loop);
1465 default:
1466 /* At this level of abstraction, the program is just a set
1467 of MODIFY_EXPRs and PHI_NODEs. In principle there is no
1468 other node to be handled. */
1469 return false;
1475 /* Given a LOOP_PHI_NODE, this function determines the evolution
1476 function from LOOP_PHI_NODE to LOOP_PHI_NODE in the loop. */
1478 static tree
1479 analyze_evolution_in_loop (tree loop_phi_node,
1480 tree init_cond)
1482 int i;
1483 tree evolution_function = chrec_not_analyzed_yet;
1484 struct loop *loop = loop_containing_stmt (loop_phi_node);
1485 basic_block bb;
1487 if (dump_file && (dump_flags & TDF_DETAILS))
1489 fprintf (dump_file, "(analyze_evolution_in_loop \n");
1490 fprintf (dump_file, " (loop_phi_node = ");
1491 print_generic_expr (dump_file, loop_phi_node, 0);
1492 fprintf (dump_file, ")\n");
1495 for (i = 0; i < PHI_NUM_ARGS (loop_phi_node); i++)
1497 tree arg = PHI_ARG_DEF (loop_phi_node, i);
1498 tree ssa_chain, ev_fn;
1499 bool res;
1501 /* Select the edges that enter the loop body. */
1502 bb = PHI_ARG_EDGE (loop_phi_node, i)->src;
1503 if (!flow_bb_inside_loop_p (loop, bb))
1504 continue;
1506 if (TREE_CODE (arg) == SSA_NAME)
1508 ssa_chain = SSA_NAME_DEF_STMT (arg);
1510 /* Pass in the initial condition to the follow edge function. */
1511 ev_fn = init_cond;
1512 res = follow_ssa_edge (loop, ssa_chain, loop_phi_node, &ev_fn);
1514 else
1515 res = false;
1517 /* When it is impossible to go back on the same
1518 loop_phi_node by following the ssa edges, the
1519 evolution is represented by a peeled chrec, i.e. the
1520 first iteration, EV_FN has the value INIT_COND, then
1521 all the other iterations it has the value of ARG.
1522 For the moment, PEELED_CHREC nodes are not built. */
1523 if (!res)
1524 ev_fn = chrec_dont_know;
1526 /* When there are multiple back edges of the loop (which in fact never
1527 happens currently, but nevertheless), merge their evolutions. */
1528 evolution_function = chrec_merge (evolution_function, ev_fn);
1531 if (dump_file && (dump_flags & TDF_DETAILS))
1533 fprintf (dump_file, " (evolution_function = ");
1534 print_generic_expr (dump_file, evolution_function, 0);
1535 fprintf (dump_file, "))\n");
1538 return evolution_function;
1541 /* Given a loop-phi-node, return the initial conditions of the
1542 variable on entry of the loop. When the CCP has propagated
1543 constants into the loop-phi-node, the initial condition is
1544 instantiated, otherwise the initial condition is kept symbolic.
1545 This analyzer does not analyze the evolution outside the current
1546 loop, and leaves this task to the on-demand tree reconstructor. */
1548 static tree
1549 analyze_initial_condition (tree loop_phi_node)
1551 int i;
1552 tree init_cond = chrec_not_analyzed_yet;
1553 struct loop *loop = bb_for_stmt (loop_phi_node)->loop_father;
1555 if (dump_file && (dump_flags & TDF_DETAILS))
1557 fprintf (dump_file, "(analyze_initial_condition \n");
1558 fprintf (dump_file, " (loop_phi_node = \n");
1559 print_generic_expr (dump_file, loop_phi_node, 0);
1560 fprintf (dump_file, ")\n");
1563 for (i = 0; i < PHI_NUM_ARGS (loop_phi_node); i++)
1565 tree branch = PHI_ARG_DEF (loop_phi_node, i);
1566 basic_block bb = PHI_ARG_EDGE (loop_phi_node, i)->src;
1568 /* When the branch is oriented to the loop's body, it does
1569 not contribute to the initial condition. */
1570 if (flow_bb_inside_loop_p (loop, bb))
1571 continue;
1573 if (init_cond == chrec_not_analyzed_yet)
1575 init_cond = branch;
1576 continue;
1579 if (TREE_CODE (branch) == SSA_NAME)
1581 init_cond = chrec_dont_know;
1582 break;
1585 init_cond = chrec_merge (init_cond, branch);
1588 /* Ooops -- a loop without an entry??? */
1589 if (init_cond == chrec_not_analyzed_yet)
1590 init_cond = chrec_dont_know;
1592 if (dump_file && (dump_flags & TDF_DETAILS))
1594 fprintf (dump_file, " (init_cond = ");
1595 print_generic_expr (dump_file, init_cond, 0);
1596 fprintf (dump_file, "))\n");
1599 return init_cond;
1602 /* Analyze the scalar evolution for LOOP_PHI_NODE. */
1604 static tree
1605 interpret_loop_phi (struct loop *loop, tree loop_phi_node)
1607 tree res;
1608 struct loop *phi_loop = loop_containing_stmt (loop_phi_node);
1609 tree init_cond;
1611 if (phi_loop != loop)
1613 struct loop *subloop;
1614 tree evolution_fn = analyze_scalar_evolution
1615 (phi_loop, PHI_RESULT (loop_phi_node));
1617 /* Dive one level deeper. */
1618 subloop = superloop_at_depth (phi_loop, loop->depth + 1);
1620 /* Interpret the subloop. */
1621 res = compute_overall_effect_of_inner_loop (subloop, evolution_fn);
1622 return res;
1625 /* Otherwise really interpret the loop phi. */
1626 init_cond = analyze_initial_condition (loop_phi_node);
1627 res = analyze_evolution_in_loop (loop_phi_node, init_cond);
1629 return res;
1632 /* This function merges the branches of a condition-phi-node,
1633 contained in the outermost loop, and whose arguments are already
1634 analyzed. */
1636 static tree
1637 interpret_condition_phi (struct loop *loop, tree condition_phi)
1639 int i;
1640 tree res = chrec_not_analyzed_yet;
1642 for (i = 0; i < PHI_NUM_ARGS (condition_phi); i++)
1644 tree branch_chrec;
1646 if (backedge_phi_arg_p (condition_phi, i))
1648 res = chrec_dont_know;
1649 break;
1652 branch_chrec = analyze_scalar_evolution
1653 (loop, PHI_ARG_DEF (condition_phi, i));
1655 res = chrec_merge (res, branch_chrec);
1658 return res;
1661 /* Interpret the right hand side of a modify_expr OPND1. If we didn't
1662 analyze this node before, follow the definitions until ending
1663 either on an analyzed modify_expr, or on a loop-phi-node. On the
1664 return path, this function propagates evolutions (ala constant copy
1665 propagation). OPND1 is not a GIMPLE expression because we could
1666 analyze the effect of an inner loop: see interpret_loop_phi. */
1668 static tree
1669 interpret_rhs_modify_expr (struct loop *loop,
1670 tree opnd1, tree type)
1672 tree res, opnd10, opnd11, chrec10, chrec11;
1674 if (is_gimple_min_invariant (opnd1))
1675 return chrec_convert (type, opnd1);
1677 switch (TREE_CODE (opnd1))
1679 case PLUS_EXPR:
1680 opnd10 = TREE_OPERAND (opnd1, 0);
1681 opnd11 = TREE_OPERAND (opnd1, 1);
1682 chrec10 = analyze_scalar_evolution (loop, opnd10);
1683 chrec11 = analyze_scalar_evolution (loop, opnd11);
1684 chrec10 = chrec_convert (type, chrec10);
1685 chrec11 = chrec_convert (type, chrec11);
1686 res = chrec_fold_plus (type, chrec10, chrec11);
1687 break;
1689 case MINUS_EXPR:
1690 opnd10 = TREE_OPERAND (opnd1, 0);
1691 opnd11 = TREE_OPERAND (opnd1, 1);
1692 chrec10 = analyze_scalar_evolution (loop, opnd10);
1693 chrec11 = analyze_scalar_evolution (loop, opnd11);
1694 chrec10 = chrec_convert (type, chrec10);
1695 chrec11 = chrec_convert (type, chrec11);
1696 res = chrec_fold_minus (type, chrec10, chrec11);
1697 break;
1699 case NEGATE_EXPR:
1700 opnd10 = TREE_OPERAND (opnd1, 0);
1701 chrec10 = analyze_scalar_evolution (loop, opnd10);
1702 chrec10 = chrec_convert (type, chrec10);
1703 res = chrec_fold_minus (type, build_int_cst (type, 0), chrec10);
1704 break;
1706 case MULT_EXPR:
1707 opnd10 = TREE_OPERAND (opnd1, 0);
1708 opnd11 = TREE_OPERAND (opnd1, 1);
1709 chrec10 = analyze_scalar_evolution (loop, opnd10);
1710 chrec11 = analyze_scalar_evolution (loop, opnd11);
1711 chrec10 = chrec_convert (type, chrec10);
1712 chrec11 = chrec_convert (type, chrec11);
1713 res = chrec_fold_multiply (type, chrec10, chrec11);
1714 break;
1716 case SSA_NAME:
1717 res = chrec_convert (type, analyze_scalar_evolution (loop, opnd1));
1718 break;
1720 case ASSERT_EXPR:
1721 opnd10 = ASSERT_EXPR_VAR (opnd1);
1722 res = chrec_convert (type, analyze_scalar_evolution (loop, opnd10));
1723 break;
1725 case NOP_EXPR:
1726 case CONVERT_EXPR:
1727 opnd10 = TREE_OPERAND (opnd1, 0);
1728 chrec10 = analyze_scalar_evolution (loop, opnd10);
1729 res = chrec_convert (type, chrec10);
1730 break;
1732 default:
1733 res = chrec_dont_know;
1734 break;
1737 return res;
1742 /* This section contains all the entry points:
1743 - number_of_iterations_in_loop,
1744 - analyze_scalar_evolution,
1745 - instantiate_parameters.
1748 /* Compute and return the evolution function in WRTO_LOOP, the nearest
1749 common ancestor of DEF_LOOP and USE_LOOP. */
1751 static tree
1752 compute_scalar_evolution_in_loop (struct loop *wrto_loop,
1753 struct loop *def_loop,
1754 tree ev)
1756 tree res;
1757 if (def_loop == wrto_loop)
1758 return ev;
1760 def_loop = superloop_at_depth (def_loop, wrto_loop->depth + 1);
1761 res = compute_overall_effect_of_inner_loop (def_loop, ev);
1763 return analyze_scalar_evolution_1 (wrto_loop, res, chrec_not_analyzed_yet);
1766 /* Helper recursive function. */
1768 static tree
1769 analyze_scalar_evolution_1 (struct loop *loop, tree var, tree res)
1771 tree def, type = TREE_TYPE (var);
1772 basic_block bb;
1773 struct loop *def_loop;
1775 if (loop == NULL)
1776 return chrec_dont_know;
1778 if (TREE_CODE (var) != SSA_NAME)
1779 return interpret_rhs_modify_expr (loop, var, type);
1781 def = SSA_NAME_DEF_STMT (var);
1782 bb = bb_for_stmt (def);
1783 def_loop = bb ? bb->loop_father : NULL;
1785 if (bb == NULL
1786 || !flow_bb_inside_loop_p (loop, bb))
1788 /* Keep the symbolic form. */
1789 res = var;
1790 goto set_and_end;
1793 if (res != chrec_not_analyzed_yet)
1795 if (loop != bb->loop_father)
1796 res = compute_scalar_evolution_in_loop
1797 (find_common_loop (loop, bb->loop_father), bb->loop_father, res);
1799 goto set_and_end;
1802 if (loop != def_loop)
1804 res = analyze_scalar_evolution_1 (def_loop, var, chrec_not_analyzed_yet);
1805 res = compute_scalar_evolution_in_loop (loop, def_loop, res);
1807 goto set_and_end;
1810 switch (TREE_CODE (def))
1812 case MODIFY_EXPR:
1813 res = interpret_rhs_modify_expr (loop, TREE_OPERAND (def, 1), type);
1814 break;
1816 case PHI_NODE:
1817 if (loop_phi_node_p (def))
1818 res = interpret_loop_phi (loop, def);
1819 else
1820 res = interpret_condition_phi (loop, def);
1821 break;
1823 default:
1824 res = chrec_dont_know;
1825 break;
1828 set_and_end:
1830 /* Keep the symbolic form. */
1831 if (res == chrec_dont_know)
1832 res = var;
1834 if (loop == def_loop)
1835 set_scalar_evolution (var, res);
1837 return res;
1840 /* Entry point for the scalar evolution analyzer.
1841 Analyzes and returns the scalar evolution of the ssa_name VAR.
1842 LOOP_NB is the identifier number of the loop in which the variable
1843 is used.
1845 Example of use: having a pointer VAR to a SSA_NAME node, STMT a
1846 pointer to the statement that uses this variable, in order to
1847 determine the evolution function of the variable, use the following
1848 calls:
1850 unsigned loop_nb = loop_containing_stmt (stmt)->num;
1851 tree chrec_with_symbols = analyze_scalar_evolution (loop_nb, var);
1852 tree chrec_instantiated = instantiate_parameters
1853 (loop_nb, chrec_with_symbols);
1856 tree
1857 analyze_scalar_evolution (struct loop *loop, tree var)
1859 tree res;
1861 if (dump_file && (dump_flags & TDF_DETAILS))
1863 fprintf (dump_file, "(analyze_scalar_evolution \n");
1864 fprintf (dump_file, " (loop_nb = %d)\n", loop->num);
1865 fprintf (dump_file, " (scalar = ");
1866 print_generic_expr (dump_file, var, 0);
1867 fprintf (dump_file, ")\n");
1870 res = analyze_scalar_evolution_1 (loop, var, get_scalar_evolution (var));
1872 if (TREE_CODE (var) == SSA_NAME && res == chrec_dont_know)
1873 res = var;
1875 if (dump_file && (dump_flags & TDF_DETAILS))
1876 fprintf (dump_file, ")\n");
1878 return res;
1881 /* Analyze scalar evolution of use of VERSION in USE_LOOP with respect to
1882 WRTO_LOOP (which should be a superloop of both USE_LOOP and definition
1883 of VERSION). */
1885 static tree
1886 analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop,
1887 tree version)
1889 bool val = false;
1890 tree ev = version;
1892 while (1)
1894 ev = analyze_scalar_evolution (use_loop, ev);
1895 ev = resolve_mixers (use_loop, ev);
1897 if (use_loop == wrto_loop)
1898 return ev;
1900 /* If the value of the use changes in the inner loop, we cannot express
1901 its value in the outer loop (we might try to return interval chrec,
1902 but we do not have a user for it anyway) */
1903 if (!no_evolution_in_loop_p (ev, use_loop->num, &val)
1904 || !val)
1905 return chrec_dont_know;
1907 use_loop = use_loop->outer;
1911 /* Returns instantiated value for VERSION in CACHE. */
1913 static tree
1914 get_instantiated_value (htab_t cache, tree version)
1916 struct scev_info_str *info, pattern;
1918 pattern.var = version;
1919 info = htab_find (cache, &pattern);
1921 if (info)
1922 return info->chrec;
1923 else
1924 return NULL_TREE;
1927 /* Sets instantiated value for VERSION to VAL in CACHE. */
1929 static void
1930 set_instantiated_value (htab_t cache, tree version, tree val)
1932 struct scev_info_str *info, pattern;
1933 PTR *slot;
1935 pattern.var = version;
1936 slot = htab_find_slot (cache, &pattern, INSERT);
1938 if (*slot)
1939 info = *slot;
1940 else
1941 info = *slot = new_scev_info_str (version);
1942 info->chrec = val;
1945 /* Analyze all the parameters of the chrec that were left under a symbolic form,
1946 with respect to LOOP. CHREC is the chrec to instantiate. If
1947 ALLOW_SUPERLOOP_CHRECS is true, replacing loop invariants with
1948 outer loop chrecs is done. CACHE is the cache of already instantiated
1949 values. */
1951 static tree
1952 instantiate_parameters_1 (struct loop *loop, tree chrec,
1953 bool allow_superloop_chrecs,
1954 htab_t cache)
1956 tree res, op0, op1, op2;
1957 basic_block def_bb;
1958 struct loop *def_loop;
1960 if (chrec == NULL_TREE
1961 || automatically_generated_chrec_p (chrec))
1962 return chrec;
1964 if (is_gimple_min_invariant (chrec))
1965 return chrec;
1967 switch (TREE_CODE (chrec))
1969 case SSA_NAME:
1970 def_bb = bb_for_stmt (SSA_NAME_DEF_STMT (chrec));
1972 /* A parameter (or loop invariant and we do not want to include
1973 evolutions in outer loops), nothing to do. */
1974 if (!def_bb
1975 || (!allow_superloop_chrecs
1976 && !flow_bb_inside_loop_p (loop, def_bb)))
1977 return chrec;
1979 /* We cache the value of instantiated variable to avoid exponential
1980 time complexity due to reevaluations. We also store the convenient
1981 value in the cache in order to prevent infinite recursion -- we do
1982 not want to instantiate the SSA_NAME if it is in a mixer
1983 structure. This is used for avoiding the instantiation of
1984 recursively defined functions, such as:
1986 | a_2 -> {0, +, 1, +, a_2}_1 */
1988 res = get_instantiated_value (cache, chrec);
1989 if (res)
1990 return res;
1992 /* Store the convenient value for chrec in the structure. If it
1993 is defined outside of the loop, we may just leave it in symbolic
1994 form, otherwise we need to admit that we do not know its behavior
1995 inside the loop. */
1996 res = !flow_bb_inside_loop_p (loop, def_bb) ? chrec : chrec_dont_know;
1997 set_instantiated_value (cache, chrec, res);
1999 /* To make things even more complicated, instantiate_parameters_1
2000 calls analyze_scalar_evolution that may call # of iterations
2001 analysis that may in turn call instantiate_parameters_1 again.
2002 To prevent the infinite recursion, keep also the bitmap of
2003 ssa names that are being instantiated globally. */
2004 if (bitmap_bit_p (already_instantiated, SSA_NAME_VERSION (chrec)))
2005 return res;
2007 def_loop = find_common_loop (loop, def_bb->loop_father);
2009 /* If the analysis yields a parametric chrec, instantiate the
2010 result again. */
2011 bitmap_set_bit (already_instantiated, SSA_NAME_VERSION (chrec));
2012 res = analyze_scalar_evolution (def_loop, chrec);
2013 if (res != chrec_dont_know)
2014 res = instantiate_parameters_1 (loop, res, allow_superloop_chrecs,
2015 cache);
2016 bitmap_clear_bit (already_instantiated, SSA_NAME_VERSION (chrec));
2018 /* Store the correct value to the cache. */
2019 set_instantiated_value (cache, chrec, res);
2020 return res;
2022 case POLYNOMIAL_CHREC:
2023 op0 = instantiate_parameters_1 (loop, CHREC_LEFT (chrec),
2024 allow_superloop_chrecs, cache);
2025 if (op0 == chrec_dont_know)
2026 return chrec_dont_know;
2028 op1 = instantiate_parameters_1 (loop, CHREC_RIGHT (chrec),
2029 allow_superloop_chrecs, cache);
2030 if (op1 == chrec_dont_know)
2031 return chrec_dont_know;
2033 if (CHREC_LEFT (chrec) != op0
2034 || CHREC_RIGHT (chrec) != op1)
2035 chrec = build_polynomial_chrec (CHREC_VARIABLE (chrec), op0, op1);
2036 return chrec;
2038 case PLUS_EXPR:
2039 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2040 allow_superloop_chrecs, cache);
2041 if (op0 == chrec_dont_know)
2042 return chrec_dont_know;
2044 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2045 allow_superloop_chrecs, cache);
2046 if (op1 == chrec_dont_know)
2047 return chrec_dont_know;
2049 if (TREE_OPERAND (chrec, 0) != op0
2050 || TREE_OPERAND (chrec, 1) != op1)
2051 chrec = chrec_fold_plus (TREE_TYPE (chrec), op0, op1);
2052 return chrec;
2054 case MINUS_EXPR:
2055 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2056 allow_superloop_chrecs, cache);
2057 if (op0 == chrec_dont_know)
2058 return chrec_dont_know;
2060 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2061 allow_superloop_chrecs, cache);
2062 if (op1 == chrec_dont_know)
2063 return chrec_dont_know;
2065 if (TREE_OPERAND (chrec, 0) != op0
2066 || TREE_OPERAND (chrec, 1) != op1)
2067 chrec = chrec_fold_minus (TREE_TYPE (chrec), op0, op1);
2068 return chrec;
2070 case MULT_EXPR:
2071 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2072 allow_superloop_chrecs, cache);
2073 if (op0 == chrec_dont_know)
2074 return chrec_dont_know;
2076 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2077 allow_superloop_chrecs, cache);
2078 if (op1 == chrec_dont_know)
2079 return chrec_dont_know;
2081 if (TREE_OPERAND (chrec, 0) != op0
2082 || TREE_OPERAND (chrec, 1) != op1)
2083 chrec = chrec_fold_multiply (TREE_TYPE (chrec), op0, op1);
2084 return chrec;
2086 case NOP_EXPR:
2087 case CONVERT_EXPR:
2088 case NON_LVALUE_EXPR:
2089 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2090 allow_superloop_chrecs, cache);
2091 if (op0 == chrec_dont_know)
2092 return chrec_dont_know;
2094 if (op0 == TREE_OPERAND (chrec, 0))
2095 return chrec;
2097 return chrec_convert (TREE_TYPE (chrec), op0);
2099 case SCEV_NOT_KNOWN:
2100 return chrec_dont_know;
2102 case SCEV_KNOWN:
2103 return chrec_known;
2105 default:
2106 break;
2109 switch (TREE_CODE_LENGTH (TREE_CODE (chrec)))
2111 case 3:
2112 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2113 allow_superloop_chrecs, cache);
2114 if (op0 == chrec_dont_know)
2115 return chrec_dont_know;
2117 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2118 allow_superloop_chrecs, cache);
2119 if (op1 == chrec_dont_know)
2120 return chrec_dont_know;
2122 op2 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 2),
2123 allow_superloop_chrecs, cache);
2124 if (op2 == chrec_dont_know)
2125 return chrec_dont_know;
2127 if (op0 == TREE_OPERAND (chrec, 0)
2128 && op1 == TREE_OPERAND (chrec, 1)
2129 && op2 == TREE_OPERAND (chrec, 2))
2130 return chrec;
2132 return fold (build (TREE_CODE (chrec),
2133 TREE_TYPE (chrec), op0, op1, op2));
2135 case 2:
2136 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2137 allow_superloop_chrecs, cache);
2138 if (op0 == chrec_dont_know)
2139 return chrec_dont_know;
2141 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2142 allow_superloop_chrecs, cache);
2143 if (op1 == chrec_dont_know)
2144 return chrec_dont_know;
2146 if (op0 == TREE_OPERAND (chrec, 0)
2147 && op1 == TREE_OPERAND (chrec, 1))
2148 return chrec;
2149 return fold (build (TREE_CODE (chrec), TREE_TYPE (chrec), op0, op1));
2151 case 1:
2152 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2153 allow_superloop_chrecs, cache);
2154 if (op0 == chrec_dont_know)
2155 return chrec_dont_know;
2156 if (op0 == TREE_OPERAND (chrec, 0))
2157 return chrec;
2158 return fold (build1 (TREE_CODE (chrec), TREE_TYPE (chrec), op0));
2160 case 0:
2161 return chrec;
2163 default:
2164 break;
2167 /* Too complicated to handle. */
2168 return chrec_dont_know;
2171 /* Analyze all the parameters of the chrec that were left under a
2172 symbolic form. LOOP is the loop in which symbolic names have to
2173 be analyzed and instantiated. */
2175 tree
2176 instantiate_parameters (struct loop *loop,
2177 tree chrec)
2179 tree res;
2180 htab_t cache = htab_create (10, hash_scev_info, eq_scev_info, del_scev_info);
2182 if (dump_file && (dump_flags & TDF_DETAILS))
2184 fprintf (dump_file, "(instantiate_parameters \n");
2185 fprintf (dump_file, " (loop_nb = %d)\n", loop->num);
2186 fprintf (dump_file, " (chrec = ");
2187 print_generic_expr (dump_file, chrec, 0);
2188 fprintf (dump_file, ")\n");
2191 res = instantiate_parameters_1 (loop, chrec, true, cache);
2193 if (dump_file && (dump_flags & TDF_DETAILS))
2195 fprintf (dump_file, " (res = ");
2196 print_generic_expr (dump_file, res, 0);
2197 fprintf (dump_file, "))\n");
2200 htab_delete (cache);
2202 return res;
2205 /* Similar to instantiate_parameters, but does not introduce the
2206 evolutions in outer loops for LOOP invariants in CHREC. */
2208 static tree
2209 resolve_mixers (struct loop *loop, tree chrec)
2211 htab_t cache = htab_create (10, hash_scev_info, eq_scev_info, del_scev_info);
2212 tree ret = instantiate_parameters_1 (loop, chrec, false, cache);
2213 htab_delete (cache);
2214 return ret;
2217 /* Entry point for the analysis of the number of iterations pass.
2218 This function tries to safely approximate the number of iterations
2219 the loop will run. When this property is not decidable at compile
2220 time, the result is chrec_dont_know. Otherwise the result is
2221 a scalar or a symbolic parameter.
2223 Example of analysis: suppose that the loop has an exit condition:
2225 "if (b > 49) goto end_loop;"
2227 and that in a previous analysis we have determined that the
2228 variable 'b' has an evolution function:
2230 "EF = {23, +, 5}_2".
2232 When we evaluate the function at the point 5, i.e. the value of the
2233 variable 'b' after 5 iterations in the loop, we have EF (5) = 48,
2234 and EF (6) = 53. In this case the value of 'b' on exit is '53' and
2235 the loop body has been executed 6 times. */
2237 tree
2238 number_of_iterations_in_loop (struct loop *loop)
2240 tree res, type;
2241 edge exit;
2242 struct tree_niter_desc niter_desc;
2244 /* Determine whether the number_of_iterations_in_loop has already
2245 been computed. */
2246 res = loop->nb_iterations;
2247 if (res)
2248 return res;
2249 res = chrec_dont_know;
2251 if (dump_file && (dump_flags & TDF_DETAILS))
2252 fprintf (dump_file, "(number_of_iterations_in_loop\n");
2254 exit = loop->single_exit;
2255 if (!exit)
2256 goto end;
2258 if (!number_of_iterations_exit (loop, exit, &niter_desc))
2259 goto end;
2261 type = TREE_TYPE (niter_desc.niter);
2262 if (integer_nonzerop (niter_desc.may_be_zero))
2263 res = build_int_cst (type, 0);
2264 else if (integer_zerop (niter_desc.may_be_zero))
2265 res = niter_desc.niter;
2266 else
2267 res = chrec_dont_know;
2269 end:
2270 return set_nb_iterations_in_loop (loop, res);
2273 /* One of the drivers for testing the scalar evolutions analysis.
2274 This function computes the number of iterations for all the loops
2275 from the EXIT_CONDITIONS array. */
2277 static void
2278 number_of_iterations_for_all_loops (VEC(tree,heap) **exit_conditions)
2280 unsigned int i;
2281 unsigned nb_chrec_dont_know_loops = 0;
2282 unsigned nb_static_loops = 0;
2283 tree cond;
2285 for (i = 0; VEC_iterate (tree, *exit_conditions, i, cond); i++)
2287 tree res = number_of_iterations_in_loop (loop_containing_stmt (cond));
2288 if (chrec_contains_undetermined (res))
2289 nb_chrec_dont_know_loops++;
2290 else
2291 nb_static_loops++;
2294 if (dump_file)
2296 fprintf (dump_file, "\n(\n");
2297 fprintf (dump_file, "-----------------------------------------\n");
2298 fprintf (dump_file, "%d\tnb_chrec_dont_know_loops\n", nb_chrec_dont_know_loops);
2299 fprintf (dump_file, "%d\tnb_static_loops\n", nb_static_loops);
2300 fprintf (dump_file, "%d\tnb_total_loops\n", current_loops->num);
2301 fprintf (dump_file, "-----------------------------------------\n");
2302 fprintf (dump_file, ")\n\n");
2304 print_loop_ir (dump_file);
2310 /* Counters for the stats. */
2312 struct chrec_stats
2314 unsigned nb_chrecs;
2315 unsigned nb_affine;
2316 unsigned nb_affine_multivar;
2317 unsigned nb_higher_poly;
2318 unsigned nb_chrec_dont_know;
2319 unsigned nb_undetermined;
2322 /* Reset the counters. */
2324 static inline void
2325 reset_chrecs_counters (struct chrec_stats *stats)
2327 stats->nb_chrecs = 0;
2328 stats->nb_affine = 0;
2329 stats->nb_affine_multivar = 0;
2330 stats->nb_higher_poly = 0;
2331 stats->nb_chrec_dont_know = 0;
2332 stats->nb_undetermined = 0;
2335 /* Dump the contents of a CHREC_STATS structure. */
2337 static void
2338 dump_chrecs_stats (FILE *file, struct chrec_stats *stats)
2340 fprintf (file, "\n(\n");
2341 fprintf (file, "-----------------------------------------\n");
2342 fprintf (file, "%d\taffine univariate chrecs\n", stats->nb_affine);
2343 fprintf (file, "%d\taffine multivariate chrecs\n", stats->nb_affine_multivar);
2344 fprintf (file, "%d\tdegree greater than 2 polynomials\n",
2345 stats->nb_higher_poly);
2346 fprintf (file, "%d\tchrec_dont_know chrecs\n", stats->nb_chrec_dont_know);
2347 fprintf (file, "-----------------------------------------\n");
2348 fprintf (file, "%d\ttotal chrecs\n", stats->nb_chrecs);
2349 fprintf (file, "%d\twith undetermined coefficients\n",
2350 stats->nb_undetermined);
2351 fprintf (file, "-----------------------------------------\n");
2352 fprintf (file, "%d\tchrecs in the scev database\n",
2353 (int) htab_elements (scalar_evolution_info));
2354 fprintf (file, "%d\tsets in the scev database\n", nb_set_scev);
2355 fprintf (file, "%d\tgets in the scev database\n", nb_get_scev);
2356 fprintf (file, "-----------------------------------------\n");
2357 fprintf (file, ")\n\n");
2360 /* Gather statistics about CHREC. */
2362 static void
2363 gather_chrec_stats (tree chrec, struct chrec_stats *stats)
2365 if (dump_file && (dump_flags & TDF_STATS))
2367 fprintf (dump_file, "(classify_chrec ");
2368 print_generic_expr (dump_file, chrec, 0);
2369 fprintf (dump_file, "\n");
2372 stats->nb_chrecs++;
2374 if (chrec == NULL_TREE)
2376 stats->nb_undetermined++;
2377 return;
2380 switch (TREE_CODE (chrec))
2382 case POLYNOMIAL_CHREC:
2383 if (evolution_function_is_affine_p (chrec))
2385 if (dump_file && (dump_flags & TDF_STATS))
2386 fprintf (dump_file, " affine_univariate\n");
2387 stats->nb_affine++;
2389 else if (evolution_function_is_affine_multivariate_p (chrec))
2391 if (dump_file && (dump_flags & TDF_STATS))
2392 fprintf (dump_file, " affine_multivariate\n");
2393 stats->nb_affine_multivar++;
2395 else
2397 if (dump_file && (dump_flags & TDF_STATS))
2398 fprintf (dump_file, " higher_degree_polynomial\n");
2399 stats->nb_higher_poly++;
2402 break;
2404 default:
2405 break;
2408 if (chrec_contains_undetermined (chrec))
2410 if (dump_file && (dump_flags & TDF_STATS))
2411 fprintf (dump_file, " undetermined\n");
2412 stats->nb_undetermined++;
2415 if (dump_file && (dump_flags & TDF_STATS))
2416 fprintf (dump_file, ")\n");
2419 /* One of the drivers for testing the scalar evolutions analysis.
2420 This function analyzes the scalar evolution of all the scalars
2421 defined as loop phi nodes in one of the loops from the
2422 EXIT_CONDITIONS array.
2424 TODO Optimization: A loop is in canonical form if it contains only
2425 a single scalar loop phi node. All the other scalars that have an
2426 evolution in the loop are rewritten in function of this single
2427 index. This allows the parallelization of the loop. */
2429 static void
2430 analyze_scalar_evolution_for_all_loop_phi_nodes (VEC(tree,heap) **exit_conditions)
2432 unsigned int i;
2433 struct chrec_stats stats;
2434 tree cond;
2436 reset_chrecs_counters (&stats);
2438 for (i = 0; VEC_iterate (tree, *exit_conditions, i, cond); i++)
2440 struct loop *loop;
2441 basic_block bb;
2442 tree phi, chrec;
2444 loop = loop_containing_stmt (cond);
2445 bb = loop->header;
2447 for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
2448 if (is_gimple_reg (PHI_RESULT (phi)))
2450 chrec = instantiate_parameters
2451 (loop,
2452 analyze_scalar_evolution (loop, PHI_RESULT (phi)));
2454 if (dump_file && (dump_flags & TDF_STATS))
2455 gather_chrec_stats (chrec, &stats);
2459 if (dump_file && (dump_flags & TDF_STATS))
2460 dump_chrecs_stats (dump_file, &stats);
2463 /* Callback for htab_traverse, gathers information on chrecs in the
2464 hashtable. */
2466 static int
2467 gather_stats_on_scev_database_1 (void **slot, void *stats)
2469 struct scev_info_str *entry = *slot;
2471 gather_chrec_stats (entry->chrec, stats);
2473 return 1;
2476 /* Classify the chrecs of the whole database. */
2478 void
2479 gather_stats_on_scev_database (void)
2481 struct chrec_stats stats;
2483 if (!dump_file)
2484 return;
2486 reset_chrecs_counters (&stats);
2488 htab_traverse (scalar_evolution_info, gather_stats_on_scev_database_1,
2489 &stats);
2491 dump_chrecs_stats (dump_file, &stats);
2496 /* Initializer. */
2498 static void
2499 initialize_scalar_evolutions_analyzer (void)
2501 /* The elements below are unique. */
2502 if (chrec_dont_know == NULL_TREE)
2504 chrec_not_analyzed_yet = NULL_TREE;
2505 chrec_dont_know = make_node (SCEV_NOT_KNOWN);
2506 chrec_known = make_node (SCEV_KNOWN);
2507 TREE_TYPE (chrec_dont_know) = NULL_TREE;
2508 TREE_TYPE (chrec_known) = NULL_TREE;
2512 /* Initialize the analysis of scalar evolutions for LOOPS. */
2514 void
2515 scev_initialize (struct loops *loops)
2517 unsigned i;
2518 current_loops = loops;
2520 scalar_evolution_info = htab_create (100, hash_scev_info,
2521 eq_scev_info, del_scev_info);
2522 already_instantiated = BITMAP_ALLOC (NULL);
2524 initialize_scalar_evolutions_analyzer ();
2526 for (i = 1; i < loops->num; i++)
2527 if (loops->parray[i])
2528 loops->parray[i]->nb_iterations = NULL_TREE;
2531 /* Cleans up the information cached by the scalar evolutions analysis. */
2533 void
2534 scev_reset (void)
2536 unsigned i;
2537 struct loop *loop;
2539 if (!scalar_evolution_info || !current_loops)
2540 return;
2542 htab_empty (scalar_evolution_info);
2543 for (i = 1; i < current_loops->num; i++)
2545 loop = current_loops->parray[i];
2546 if (loop)
2547 loop->nb_iterations = NULL_TREE;
2551 /* Checks whether OP behaves as a simple affine iv of LOOP in STMT and returns
2552 its BASE and STEP if possible. If ALLOW_NONCONSTANT_STEP is true, we
2553 want STEP to be invariant in LOOP. Otherwise we require it to be an
2554 integer constant. */
2556 bool
2557 simple_iv (struct loop *loop, tree stmt, tree op, tree *base, tree *step,
2558 bool allow_nonconstant_step)
2560 basic_block bb = bb_for_stmt (stmt);
2561 tree type, ev;
2563 *base = NULL_TREE;
2564 *step = NULL_TREE;
2566 type = TREE_TYPE (op);
2567 if (TREE_CODE (type) != INTEGER_TYPE
2568 && TREE_CODE (type) != POINTER_TYPE)
2569 return false;
2571 ev = analyze_scalar_evolution_in_loop (loop, bb->loop_father, op);
2572 if (chrec_contains_undetermined (ev))
2573 return false;
2575 if (tree_does_not_contain_chrecs (ev)
2576 && !chrec_contains_symbols_defined_in_loop (ev, loop->num))
2578 *base = ev;
2579 return true;
2582 if (TREE_CODE (ev) != POLYNOMIAL_CHREC
2583 || CHREC_VARIABLE (ev) != (unsigned) loop->num)
2584 return false;
2586 *step = CHREC_RIGHT (ev);
2587 if (allow_nonconstant_step)
2589 if (tree_contains_chrecs (*step, NULL)
2590 || chrec_contains_symbols_defined_in_loop (*step, loop->num))
2591 return false;
2593 else if (TREE_CODE (*step) != INTEGER_CST)
2594 return false;
2596 *base = CHREC_LEFT (ev);
2597 if (tree_contains_chrecs (*base, NULL)
2598 || chrec_contains_symbols_defined_in_loop (*base, loop->num))
2599 return false;
2601 return true;
2604 /* Runs the analysis of scalar evolutions. */
2606 void
2607 scev_analysis (void)
2609 VEC(tree,heap) *exit_conditions;
2611 exit_conditions = VEC_alloc (tree, heap, 37);
2612 select_loops_exit_conditions (current_loops, &exit_conditions);
2614 if (dump_file && (dump_flags & TDF_STATS))
2615 analyze_scalar_evolution_for_all_loop_phi_nodes (&exit_conditions);
2617 number_of_iterations_for_all_loops (&exit_conditions);
2618 VEC_free (tree, heap, exit_conditions);
2621 /* Finalize the scalar evolution analysis. */
2623 void
2624 scev_finalize (void)
2626 htab_delete (scalar_evolution_info);
2627 BITMAP_FREE (already_instantiated);
2630 /* Replace ssa names for that scev can prove they are constant by the
2631 appropriate constants. Most importantly, this takes care of final
2632 value replacement.
2634 We only consider SSA names defined by phi nodes; rest is left to the
2635 ordinary constant propagation pass. */
2637 void
2638 scev_const_prop (void)
2640 basic_block bb;
2641 tree name, phi, type, ev;
2642 struct loop *loop;
2643 bitmap ssa_names_to_remove = NULL;
2645 if (!current_loops)
2646 return;
2648 FOR_EACH_BB (bb)
2650 loop = bb->loop_father;
2652 for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
2654 name = PHI_RESULT (phi);
2656 if (!is_gimple_reg (name))
2657 continue;
2659 type = TREE_TYPE (name);
2661 if (!POINTER_TYPE_P (type)
2662 && !INTEGRAL_TYPE_P (type))
2663 continue;
2665 ev = resolve_mixers (loop, analyze_scalar_evolution (loop, name));
2666 if (!is_gimple_min_invariant (ev)
2667 || !may_propagate_copy (name, ev))
2668 continue;
2670 /* Replace the uses of the name. */
2671 replace_uses_by (name, ev);
2673 if (!ssa_names_to_remove)
2674 ssa_names_to_remove = BITMAP_ALLOC (NULL);
2675 bitmap_set_bit (ssa_names_to_remove, SSA_NAME_VERSION (name));
2679 /* Remove the ssa names that were replaced by constants. We do not remove them
2680 directly in the previous cycle, since this invalidates scev cache. */
2681 if (ssa_names_to_remove)
2683 bitmap_iterator bi;
2684 unsigned i;
2686 EXECUTE_IF_SET_IN_BITMAP (ssa_names_to_remove, 0, i, bi)
2688 name = ssa_name (i);
2689 phi = SSA_NAME_DEF_STMT (name);
2691 gcc_assert (TREE_CODE (phi) == PHI_NODE);
2692 remove_phi_node (phi, NULL);
2695 BITMAP_FREE (ssa_names_to_remove);
2696 scev_reset ();