* arm.c (FL_WBUF): Define.
[official-gcc.git] / gcc / tree-scalar-evolution.c
blobffead4903b778513c918fad890ec09d0154871bd
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 && TREE_INT_CST_LOW (res) == 0)
926 || TREE_OVERFLOW (res))
927 res = chrec_dont_know;
929 if (dump_file && (dump_flags & TDF_DETAILS))
931 fprintf (dump_file, " (set_nb_iterations_in_loop = ");
932 print_generic_expr (dump_file, res, 0);
933 fprintf (dump_file, "))\n");
936 loop->nb_iterations = res;
937 return res;
942 /* This section selects the loops that will be good candidates for the
943 scalar evolution analysis. For the moment, greedily select all the
944 loop nests we could analyze. */
946 /* Return true when it is possible to analyze the condition expression
947 EXPR. */
949 static bool
950 analyzable_condition (tree expr)
952 tree condition;
954 if (TREE_CODE (expr) != COND_EXPR)
955 return false;
957 condition = TREE_OPERAND (expr, 0);
959 switch (TREE_CODE (condition))
961 case SSA_NAME:
962 return true;
964 case LT_EXPR:
965 case LE_EXPR:
966 case GT_EXPR:
967 case GE_EXPR:
968 case EQ_EXPR:
969 case NE_EXPR:
970 return true;
972 default:
973 return false;
976 return false;
979 /* For a loop with a single exit edge, return the COND_EXPR that
980 guards the exit edge. If the expression is too difficult to
981 analyze, then give up. */
983 tree
984 get_loop_exit_condition (struct loop *loop)
986 tree res = NULL_TREE;
987 edge exit_edge = loop->single_exit;
990 if (dump_file && (dump_flags & TDF_DETAILS))
991 fprintf (dump_file, "(get_loop_exit_condition \n ");
993 if (exit_edge)
995 tree expr;
997 expr = last_stmt (exit_edge->src);
998 if (analyzable_condition (expr))
999 res = expr;
1002 if (dump_file && (dump_flags & TDF_DETAILS))
1004 print_generic_expr (dump_file, res, 0);
1005 fprintf (dump_file, ")\n");
1008 return res;
1011 /* Recursively determine and enqueue the exit conditions for a loop. */
1013 static void
1014 get_exit_conditions_rec (struct loop *loop,
1015 varray_type *exit_conditions)
1017 if (!loop)
1018 return;
1020 /* Recurse on the inner loops, then on the next (sibling) loops. */
1021 get_exit_conditions_rec (loop->inner, exit_conditions);
1022 get_exit_conditions_rec (loop->next, exit_conditions);
1024 if (loop->single_exit)
1026 tree loop_condition = get_loop_exit_condition (loop);
1028 if (loop_condition)
1029 VARRAY_PUSH_TREE (*exit_conditions, loop_condition);
1033 /* Select the candidate loop nests for the analysis. This function
1034 initializes the EXIT_CONDITIONS array. */
1036 static void
1037 select_loops_exit_conditions (struct loops *loops,
1038 varray_type *exit_conditions)
1040 struct loop *function_body = loops->parray[0];
1042 get_exit_conditions_rec (function_body->inner, exit_conditions);
1046 /* Depth first search algorithm. */
1048 static bool follow_ssa_edge (struct loop *loop, tree, tree, tree *);
1050 /* Follow the ssa edge into the right hand side RHS of an assignment.
1051 Return true if the strongly connected component has been found. */
1053 static bool
1054 follow_ssa_edge_in_rhs (struct loop *loop,
1055 tree rhs,
1056 tree halting_phi,
1057 tree *evolution_of_loop)
1059 bool res = false;
1060 tree rhs0, rhs1;
1061 tree type_rhs = TREE_TYPE (rhs);
1063 /* The RHS is one of the following cases:
1064 - an SSA_NAME,
1065 - an INTEGER_CST,
1066 - a PLUS_EXPR,
1067 - a MINUS_EXPR,
1068 - an ASSERT_EXPR,
1069 - other cases are not yet handled. */
1070 switch (TREE_CODE (rhs))
1072 case NOP_EXPR:
1073 /* This assignment is under the form "a_1 = (cast) rhs. */
1074 res = follow_ssa_edge_in_rhs (loop, TREE_OPERAND (rhs, 0), halting_phi,
1075 evolution_of_loop);
1076 *evolution_of_loop = chrec_convert (TREE_TYPE (rhs), *evolution_of_loop);
1077 break;
1079 case INTEGER_CST:
1080 /* This assignment is under the form "a_1 = 7". */
1081 res = false;
1082 break;
1084 case SSA_NAME:
1085 /* This assignment is under the form: "a_1 = b_2". */
1086 res = follow_ssa_edge
1087 (loop, SSA_NAME_DEF_STMT (rhs), halting_phi, evolution_of_loop);
1088 break;
1090 case PLUS_EXPR:
1091 /* This case is under the form "rhs0 + rhs1". */
1092 rhs0 = TREE_OPERAND (rhs, 0);
1093 rhs1 = TREE_OPERAND (rhs, 1);
1094 STRIP_TYPE_NOPS (rhs0);
1095 STRIP_TYPE_NOPS (rhs1);
1097 if (TREE_CODE (rhs0) == SSA_NAME)
1099 if (TREE_CODE (rhs1) == SSA_NAME)
1101 /* Match an assignment under the form:
1102 "a = b + c". */
1103 res = follow_ssa_edge
1104 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1105 evolution_of_loop);
1107 if (res)
1108 *evolution_of_loop = add_to_evolution
1109 (loop->num,
1110 chrec_convert (type_rhs, *evolution_of_loop),
1111 PLUS_EXPR, rhs1);
1113 else
1115 res = follow_ssa_edge
1116 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1117 evolution_of_loop);
1119 if (res)
1120 *evolution_of_loop = add_to_evolution
1121 (loop->num,
1122 chrec_convert (type_rhs, *evolution_of_loop),
1123 PLUS_EXPR, rhs0);
1127 else
1129 /* Match an assignment under the form:
1130 "a = b + ...". */
1131 res = follow_ssa_edge
1132 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1133 evolution_of_loop);
1134 if (res)
1135 *evolution_of_loop = add_to_evolution
1136 (loop->num, chrec_convert (type_rhs, *evolution_of_loop),
1137 PLUS_EXPR, rhs1);
1141 else if (TREE_CODE (rhs1) == SSA_NAME)
1143 /* Match an assignment under the form:
1144 "a = ... + c". */
1145 res = follow_ssa_edge
1146 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1147 evolution_of_loop);
1148 if (res)
1149 *evolution_of_loop = add_to_evolution
1150 (loop->num, chrec_convert (type_rhs, *evolution_of_loop),
1151 PLUS_EXPR, rhs0);
1154 else
1155 /* Otherwise, match an assignment under the form:
1156 "a = ... + ...". */
1157 /* And there is nothing to do. */
1158 res = false;
1160 break;
1162 case MINUS_EXPR:
1163 /* This case is under the form "opnd0 = rhs0 - rhs1". */
1164 rhs0 = TREE_OPERAND (rhs, 0);
1165 rhs1 = TREE_OPERAND (rhs, 1);
1166 STRIP_TYPE_NOPS (rhs0);
1167 STRIP_TYPE_NOPS (rhs1);
1169 if (TREE_CODE (rhs0) == SSA_NAME)
1171 /* Match an assignment under the form:
1172 "a = b - ...". */
1173 res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1174 evolution_of_loop);
1175 if (res)
1176 *evolution_of_loop = add_to_evolution
1177 (loop->num, chrec_convert (type_rhs, *evolution_of_loop),
1178 MINUS_EXPR, rhs1);
1180 else
1181 /* Otherwise, match an assignment under the form:
1182 "a = ... - ...". */
1183 /* And there is nothing to do. */
1184 res = false;
1186 break;
1188 case MULT_EXPR:
1189 /* This case is under the form "opnd0 = rhs0 * rhs1". */
1190 rhs0 = TREE_OPERAND (rhs, 0);
1191 rhs1 = TREE_OPERAND (rhs, 1);
1192 STRIP_TYPE_NOPS (rhs0);
1193 STRIP_TYPE_NOPS (rhs1);
1195 if (TREE_CODE (rhs0) == SSA_NAME)
1197 if (TREE_CODE (rhs1) == SSA_NAME)
1199 /* Match an assignment under the form:
1200 "a = b * c". */
1201 res = follow_ssa_edge
1202 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1203 evolution_of_loop);
1205 if (res)
1206 *evolution_of_loop = chrec_dont_know;
1208 else
1210 res = follow_ssa_edge
1211 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1212 evolution_of_loop);
1214 if (res)
1215 *evolution_of_loop = chrec_dont_know;
1219 else
1221 /* Match an assignment under the form:
1222 "a = b * ...". */
1223 res = follow_ssa_edge
1224 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1225 evolution_of_loop);
1226 if (res)
1227 *evolution_of_loop = chrec_dont_know;
1231 else if (TREE_CODE (rhs1) == SSA_NAME)
1233 /* Match an assignment under the form:
1234 "a = ... * c". */
1235 res = follow_ssa_edge
1236 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1237 evolution_of_loop);
1238 if (res)
1239 *evolution_of_loop = chrec_dont_know;
1242 else
1243 /* Otherwise, match an assignment under the form:
1244 "a = ... * ...". */
1245 /* And there is nothing to do. */
1246 res = false;
1248 break;
1250 case ASSERT_EXPR:
1252 /* This assignment is of the form: "a_1 = ASSERT_EXPR <a_2, ...>"
1253 It must be handled as a copy assignment of the form a_1 = a_2. */
1254 tree op0 = ASSERT_EXPR_VAR (rhs);
1255 if (TREE_CODE (op0) == SSA_NAME)
1256 res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (op0),
1257 halting_phi, evolution_of_loop);
1258 else
1259 res = false;
1260 break;
1264 default:
1265 res = false;
1266 break;
1269 return res;
1272 /* Checks whether the I-th argument of a PHI comes from a backedge. */
1274 static bool
1275 backedge_phi_arg_p (tree phi, int i)
1277 edge e = PHI_ARG_EDGE (phi, i);
1279 /* We would in fact like to test EDGE_DFS_BACK here, but we do not care
1280 about updating it anywhere, and this should work as well most of the
1281 time. */
1282 if (e->flags & EDGE_IRREDUCIBLE_LOOP)
1283 return true;
1285 return false;
1288 /* Helper function for one branch of the condition-phi-node. Return
1289 true if the strongly connected component has been found following
1290 this path. */
1292 static inline bool
1293 follow_ssa_edge_in_condition_phi_branch (int i,
1294 struct loop *loop,
1295 tree condition_phi,
1296 tree halting_phi,
1297 tree *evolution_of_branch,
1298 tree init_cond)
1300 tree branch = PHI_ARG_DEF (condition_phi, i);
1301 *evolution_of_branch = chrec_dont_know;
1303 /* Do not follow back edges (they must belong to an irreducible loop, which
1304 we really do not want to worry about). */
1305 if (backedge_phi_arg_p (condition_phi, i))
1306 return false;
1308 if (TREE_CODE (branch) == SSA_NAME)
1310 *evolution_of_branch = init_cond;
1311 return follow_ssa_edge (loop, SSA_NAME_DEF_STMT (branch), halting_phi,
1312 evolution_of_branch);
1315 /* This case occurs when one of the condition branches sets
1316 the variable to a constant: i.e. a phi-node like
1317 "a_2 = PHI <a_7(5), 2(6)>;".
1319 FIXME: This case have to be refined correctly:
1320 in some cases it is possible to say something better than
1321 chrec_dont_know, for example using a wrap-around notation. */
1322 return false;
1325 /* This function merges the branches of a condition-phi-node in a
1326 loop. */
1328 static bool
1329 follow_ssa_edge_in_condition_phi (struct loop *loop,
1330 tree condition_phi,
1331 tree halting_phi,
1332 tree *evolution_of_loop)
1334 int i;
1335 tree init = *evolution_of_loop;
1336 tree evolution_of_branch;
1338 if (!follow_ssa_edge_in_condition_phi_branch (0, loop, condition_phi,
1339 halting_phi,
1340 &evolution_of_branch,
1341 init))
1342 return false;
1343 *evolution_of_loop = evolution_of_branch;
1345 for (i = 1; i < PHI_NUM_ARGS (condition_phi); i++)
1347 /* Quickly give up when the evolution of one of the branches is
1348 not known. */
1349 if (*evolution_of_loop == chrec_dont_know)
1350 return true;
1352 if (!follow_ssa_edge_in_condition_phi_branch (i, loop, condition_phi,
1353 halting_phi,
1354 &evolution_of_branch,
1355 init))
1356 return false;
1358 *evolution_of_loop = chrec_merge (*evolution_of_loop,
1359 evolution_of_branch);
1362 return true;
1365 /* Follow an SSA edge in an inner loop. It computes the overall
1366 effect of the loop, and following the symbolic initial conditions,
1367 it follows the edges in the parent loop. The inner loop is
1368 considered as a single statement. */
1370 static bool
1371 follow_ssa_edge_inner_loop_phi (struct loop *outer_loop,
1372 tree loop_phi_node,
1373 tree halting_phi,
1374 tree *evolution_of_loop)
1376 struct loop *loop = loop_containing_stmt (loop_phi_node);
1377 tree ev = analyze_scalar_evolution (loop, PHI_RESULT (loop_phi_node));
1379 /* Sometimes, the inner loop is too difficult to analyze, and the
1380 result of the analysis is a symbolic parameter. */
1381 if (ev == PHI_RESULT (loop_phi_node))
1383 bool res = false;
1384 int i;
1386 for (i = 0; i < PHI_NUM_ARGS (loop_phi_node); i++)
1388 tree arg = PHI_ARG_DEF (loop_phi_node, i);
1389 basic_block bb;
1391 /* Follow the edges that exit the inner loop. */
1392 bb = PHI_ARG_EDGE (loop_phi_node, i)->src;
1393 if (!flow_bb_inside_loop_p (loop, bb))
1394 res = res || follow_ssa_edge_in_rhs (outer_loop, arg, halting_phi,
1395 evolution_of_loop);
1398 /* If the path crosses this loop-phi, give up. */
1399 if (res == true)
1400 *evolution_of_loop = chrec_dont_know;
1402 return res;
1405 /* Otherwise, compute the overall effect of the inner loop. */
1406 ev = compute_overall_effect_of_inner_loop (loop, ev);
1407 return follow_ssa_edge_in_rhs (outer_loop, ev, halting_phi,
1408 evolution_of_loop);
1411 /* Follow an SSA edge from a loop-phi-node to itself, constructing a
1412 path that is analyzed on the return walk. */
1414 static bool
1415 follow_ssa_edge (struct loop *loop,
1416 tree def,
1417 tree halting_phi,
1418 tree *evolution_of_loop)
1420 struct loop *def_loop;
1422 if (TREE_CODE (def) == NOP_EXPR)
1423 return false;
1425 def_loop = loop_containing_stmt (def);
1427 switch (TREE_CODE (def))
1429 case PHI_NODE:
1430 if (!loop_phi_node_p (def))
1431 /* DEF is a condition-phi-node. Follow the branches, and
1432 record their evolutions. Finally, merge the collected
1433 information and set the approximation to the main
1434 variable. */
1435 return follow_ssa_edge_in_condition_phi
1436 (loop, def, halting_phi, evolution_of_loop);
1438 /* When the analyzed phi is the halting_phi, the
1439 depth-first search is over: we have found a path from
1440 the halting_phi to itself in the loop. */
1441 if (def == halting_phi)
1442 return true;
1444 /* Otherwise, the evolution of the HALTING_PHI depends
1445 on the evolution of another loop-phi-node, i.e. the
1446 evolution function is a higher degree polynomial. */
1447 if (def_loop == loop)
1448 return false;
1450 /* Inner loop. */
1451 if (flow_loop_nested_p (loop, def_loop))
1452 return follow_ssa_edge_inner_loop_phi
1453 (loop, def, halting_phi, evolution_of_loop);
1455 /* Outer loop. */
1456 return false;
1458 case MODIFY_EXPR:
1459 return follow_ssa_edge_in_rhs (loop,
1460 TREE_OPERAND (def, 1),
1461 halting_phi,
1462 evolution_of_loop);
1464 default:
1465 /* At this level of abstraction, the program is just a set
1466 of MODIFY_EXPRs and PHI_NODEs. In principle there is no
1467 other node to be handled. */
1468 return false;
1474 /* Given a LOOP_PHI_NODE, this function determines the evolution
1475 function from LOOP_PHI_NODE to LOOP_PHI_NODE in the loop. */
1477 static tree
1478 analyze_evolution_in_loop (tree loop_phi_node,
1479 tree init_cond)
1481 int i;
1482 tree evolution_function = chrec_not_analyzed_yet;
1483 struct loop *loop = loop_containing_stmt (loop_phi_node);
1484 basic_block bb;
1486 if (dump_file && (dump_flags & TDF_DETAILS))
1488 fprintf (dump_file, "(analyze_evolution_in_loop \n");
1489 fprintf (dump_file, " (loop_phi_node = ");
1490 print_generic_expr (dump_file, loop_phi_node, 0);
1491 fprintf (dump_file, ")\n");
1494 for (i = 0; i < PHI_NUM_ARGS (loop_phi_node); i++)
1496 tree arg = PHI_ARG_DEF (loop_phi_node, i);
1497 tree ssa_chain, ev_fn;
1498 bool res;
1500 /* Select the edges that enter the loop body. */
1501 bb = PHI_ARG_EDGE (loop_phi_node, i)->src;
1502 if (!flow_bb_inside_loop_p (loop, bb))
1503 continue;
1505 if (TREE_CODE (arg) == SSA_NAME)
1507 ssa_chain = SSA_NAME_DEF_STMT (arg);
1509 /* Pass in the initial condition to the follow edge function. */
1510 ev_fn = init_cond;
1511 res = follow_ssa_edge (loop, ssa_chain, loop_phi_node, &ev_fn);
1513 else
1514 res = false;
1516 /* When it is impossible to go back on the same
1517 loop_phi_node by following the ssa edges, the
1518 evolution is represented by a peeled chrec, i.e. the
1519 first iteration, EV_FN has the value INIT_COND, then
1520 all the other iterations it has the value of ARG.
1521 For the moment, PEELED_CHREC nodes are not built. */
1522 if (!res)
1523 ev_fn = chrec_dont_know;
1525 /* When there are multiple back edges of the loop (which in fact never
1526 happens currently, but nevertheless), merge their evolutions. */
1527 evolution_function = chrec_merge (evolution_function, ev_fn);
1530 if (dump_file && (dump_flags & TDF_DETAILS))
1532 fprintf (dump_file, " (evolution_function = ");
1533 print_generic_expr (dump_file, evolution_function, 0);
1534 fprintf (dump_file, "))\n");
1537 return evolution_function;
1540 /* Given a loop-phi-node, return the initial conditions of the
1541 variable on entry of the loop. When the CCP has propagated
1542 constants into the loop-phi-node, the initial condition is
1543 instantiated, otherwise the initial condition is kept symbolic.
1544 This analyzer does not analyze the evolution outside the current
1545 loop, and leaves this task to the on-demand tree reconstructor. */
1547 static tree
1548 analyze_initial_condition (tree loop_phi_node)
1550 int i;
1551 tree init_cond = chrec_not_analyzed_yet;
1552 struct loop *loop = bb_for_stmt (loop_phi_node)->loop_father;
1554 if (dump_file && (dump_flags & TDF_DETAILS))
1556 fprintf (dump_file, "(analyze_initial_condition \n");
1557 fprintf (dump_file, " (loop_phi_node = \n");
1558 print_generic_expr (dump_file, loop_phi_node, 0);
1559 fprintf (dump_file, ")\n");
1562 for (i = 0; i < PHI_NUM_ARGS (loop_phi_node); i++)
1564 tree branch = PHI_ARG_DEF (loop_phi_node, i);
1565 basic_block bb = PHI_ARG_EDGE (loop_phi_node, i)->src;
1567 /* When the branch is oriented to the loop's body, it does
1568 not contribute to the initial condition. */
1569 if (flow_bb_inside_loop_p (loop, bb))
1570 continue;
1572 if (init_cond == chrec_not_analyzed_yet)
1574 init_cond = branch;
1575 continue;
1578 if (TREE_CODE (branch) == SSA_NAME)
1580 init_cond = chrec_dont_know;
1581 break;
1584 init_cond = chrec_merge (init_cond, branch);
1587 /* Ooops -- a loop without an entry??? */
1588 if (init_cond == chrec_not_analyzed_yet)
1589 init_cond = chrec_dont_know;
1591 if (dump_file && (dump_flags & TDF_DETAILS))
1593 fprintf (dump_file, " (init_cond = ");
1594 print_generic_expr (dump_file, init_cond, 0);
1595 fprintf (dump_file, "))\n");
1598 return init_cond;
1601 /* Analyze the scalar evolution for LOOP_PHI_NODE. */
1603 static tree
1604 interpret_loop_phi (struct loop *loop, tree loop_phi_node)
1606 tree res;
1607 struct loop *phi_loop = loop_containing_stmt (loop_phi_node);
1608 tree init_cond;
1610 if (phi_loop != loop)
1612 struct loop *subloop;
1613 tree evolution_fn = analyze_scalar_evolution
1614 (phi_loop, PHI_RESULT (loop_phi_node));
1616 /* Dive one level deeper. */
1617 subloop = superloop_at_depth (phi_loop, loop->depth + 1);
1619 /* Interpret the subloop. */
1620 res = compute_overall_effect_of_inner_loop (subloop, evolution_fn);
1621 return res;
1624 /* Otherwise really interpret the loop phi. */
1625 init_cond = analyze_initial_condition (loop_phi_node);
1626 res = analyze_evolution_in_loop (loop_phi_node, init_cond);
1628 return res;
1631 /* This function merges the branches of a condition-phi-node,
1632 contained in the outermost loop, and whose arguments are already
1633 analyzed. */
1635 static tree
1636 interpret_condition_phi (struct loop *loop, tree condition_phi)
1638 int i;
1639 tree res = chrec_not_analyzed_yet;
1641 for (i = 0; i < PHI_NUM_ARGS (condition_phi); i++)
1643 tree branch_chrec;
1645 if (backedge_phi_arg_p (condition_phi, i))
1647 res = chrec_dont_know;
1648 break;
1651 branch_chrec = analyze_scalar_evolution
1652 (loop, PHI_ARG_DEF (condition_phi, i));
1654 res = chrec_merge (res, branch_chrec);
1657 return res;
1660 /* Interpret the right hand side of a modify_expr OPND1. If we didn't
1661 analyzed this node before, follow the definitions until ending
1662 either on an analyzed modify_expr, or on a loop-phi-node. On the
1663 return path, this function propagates evolutions (ala constant copy
1664 propagation). OPND1 is not a GIMPLE expression because we could
1665 analyze the effect of an inner loop: see interpret_loop_phi. */
1667 static tree
1668 interpret_rhs_modify_expr (struct loop *loop,
1669 tree opnd1, tree type)
1671 tree res, opnd10, opnd11, chrec10, chrec11;
1673 if (is_gimple_min_invariant (opnd1))
1674 return chrec_convert (type, opnd1);
1676 switch (TREE_CODE (opnd1))
1678 case PLUS_EXPR:
1679 opnd10 = TREE_OPERAND (opnd1, 0);
1680 opnd11 = TREE_OPERAND (opnd1, 1);
1681 chrec10 = analyze_scalar_evolution (loop, opnd10);
1682 chrec11 = analyze_scalar_evolution (loop, opnd11);
1683 chrec10 = chrec_convert (type, chrec10);
1684 chrec11 = chrec_convert (type, chrec11);
1685 res = chrec_fold_plus (type, chrec10, chrec11);
1686 break;
1688 case MINUS_EXPR:
1689 opnd10 = TREE_OPERAND (opnd1, 0);
1690 opnd11 = TREE_OPERAND (opnd1, 1);
1691 chrec10 = analyze_scalar_evolution (loop, opnd10);
1692 chrec11 = analyze_scalar_evolution (loop, opnd11);
1693 chrec10 = chrec_convert (type, chrec10);
1694 chrec11 = chrec_convert (type, chrec11);
1695 res = chrec_fold_minus (type, chrec10, chrec11);
1696 break;
1698 case NEGATE_EXPR:
1699 opnd10 = TREE_OPERAND (opnd1, 0);
1700 chrec10 = analyze_scalar_evolution (loop, opnd10);
1701 chrec10 = chrec_convert (type, chrec10);
1702 res = chrec_fold_minus (type, build_int_cst (type, 0), chrec10);
1703 break;
1705 case MULT_EXPR:
1706 opnd10 = TREE_OPERAND (opnd1, 0);
1707 opnd11 = TREE_OPERAND (opnd1, 1);
1708 chrec10 = analyze_scalar_evolution (loop, opnd10);
1709 chrec11 = analyze_scalar_evolution (loop, opnd11);
1710 chrec10 = chrec_convert (type, chrec10);
1711 chrec11 = chrec_convert (type, chrec11);
1712 res = chrec_fold_multiply (type, chrec10, chrec11);
1713 break;
1715 case SSA_NAME:
1716 res = chrec_convert (type, analyze_scalar_evolution (loop, opnd1));
1717 break;
1719 case ASSERT_EXPR:
1720 opnd10 = ASSERT_EXPR_VAR (opnd1);
1721 res = chrec_convert (type, analyze_scalar_evolution (loop, opnd10));
1722 break;
1724 case NOP_EXPR:
1725 case CONVERT_EXPR:
1726 opnd10 = TREE_OPERAND (opnd1, 0);
1727 chrec10 = analyze_scalar_evolution (loop, opnd10);
1728 res = chrec_convert (type, chrec10);
1729 break;
1731 default:
1732 res = chrec_dont_know;
1733 break;
1736 return res;
1741 /* This section contains all the entry points:
1742 - number_of_iterations_in_loop,
1743 - analyze_scalar_evolution,
1744 - instantiate_parameters.
1747 /* Compute and return the evolution function in WRTO_LOOP, the nearest
1748 common ancestor of DEF_LOOP and USE_LOOP. */
1750 static tree
1751 compute_scalar_evolution_in_loop (struct loop *wrto_loop,
1752 struct loop *def_loop,
1753 tree ev)
1755 tree res;
1756 if (def_loop == wrto_loop)
1757 return ev;
1759 def_loop = superloop_at_depth (def_loop, wrto_loop->depth + 1);
1760 res = compute_overall_effect_of_inner_loop (def_loop, ev);
1762 return analyze_scalar_evolution_1 (wrto_loop, res, chrec_not_analyzed_yet);
1765 /* Helper recursive function. */
1767 static tree
1768 analyze_scalar_evolution_1 (struct loop *loop, tree var, tree res)
1770 tree def, type = TREE_TYPE (var);
1771 basic_block bb;
1772 struct loop *def_loop;
1774 if (loop == NULL)
1775 return chrec_dont_know;
1777 if (TREE_CODE (var) != SSA_NAME)
1778 return interpret_rhs_modify_expr (loop, var, type);
1780 def = SSA_NAME_DEF_STMT (var);
1781 bb = bb_for_stmt (def);
1782 def_loop = bb ? bb->loop_father : NULL;
1784 if (bb == NULL
1785 || !flow_bb_inside_loop_p (loop, bb))
1787 /* Keep the symbolic form. */
1788 res = var;
1789 goto set_and_end;
1792 if (res != chrec_not_analyzed_yet)
1794 if (loop != bb->loop_father)
1795 res = compute_scalar_evolution_in_loop
1796 (find_common_loop (loop, bb->loop_father), bb->loop_father, res);
1798 goto set_and_end;
1801 if (loop != def_loop)
1803 res = analyze_scalar_evolution_1 (def_loop, var, chrec_not_analyzed_yet);
1804 res = compute_scalar_evolution_in_loop (loop, def_loop, res);
1806 goto set_and_end;
1809 switch (TREE_CODE (def))
1811 case MODIFY_EXPR:
1812 res = interpret_rhs_modify_expr (loop, TREE_OPERAND (def, 1), type);
1813 break;
1815 case PHI_NODE:
1816 if (loop_phi_node_p (def))
1817 res = interpret_loop_phi (loop, def);
1818 else
1819 res = interpret_condition_phi (loop, def);
1820 break;
1822 default:
1823 res = chrec_dont_know;
1824 break;
1827 set_and_end:
1829 /* Keep the symbolic form. */
1830 if (res == chrec_dont_know)
1831 res = var;
1833 if (loop == def_loop)
1834 set_scalar_evolution (var, res);
1836 return res;
1839 /* Entry point for the scalar evolution analyzer.
1840 Analyzes and returns the scalar evolution of the ssa_name VAR.
1841 LOOP_NB is the identifier number of the loop in which the variable
1842 is used.
1844 Example of use: having a pointer VAR to a SSA_NAME node, STMT a
1845 pointer to the statement that uses this variable, in order to
1846 determine the evolution function of the variable, use the following
1847 calls:
1849 unsigned loop_nb = loop_containing_stmt (stmt)->num;
1850 tree chrec_with_symbols = analyze_scalar_evolution (loop_nb, var);
1851 tree chrec_instantiated = instantiate_parameters
1852 (loop_nb, chrec_with_symbols);
1855 tree
1856 analyze_scalar_evolution (struct loop *loop, tree var)
1858 tree res;
1860 if (dump_file && (dump_flags & TDF_DETAILS))
1862 fprintf (dump_file, "(analyze_scalar_evolution \n");
1863 fprintf (dump_file, " (loop_nb = %d)\n", loop->num);
1864 fprintf (dump_file, " (scalar = ");
1865 print_generic_expr (dump_file, var, 0);
1866 fprintf (dump_file, ")\n");
1869 res = analyze_scalar_evolution_1 (loop, var, get_scalar_evolution (var));
1871 if (TREE_CODE (var) == SSA_NAME && res == chrec_dont_know)
1872 res = var;
1874 if (dump_file && (dump_flags & TDF_DETAILS))
1875 fprintf (dump_file, ")\n");
1877 return res;
1880 /* Analyze scalar evolution of use of VERSION in USE_LOOP with respect to
1881 WRTO_LOOP (which should be a superloop of both USE_LOOP and definition
1882 of VERSION). */
1884 static tree
1885 analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop,
1886 tree version)
1888 bool val = false;
1889 tree ev = version;
1891 while (1)
1893 ev = analyze_scalar_evolution (use_loop, ev);
1894 ev = resolve_mixers (use_loop, ev);
1896 if (use_loop == wrto_loop)
1897 return ev;
1899 /* If the value of the use changes in the inner loop, we cannot express
1900 its value in the outer loop (we might try to return interval chrec,
1901 but we do not have a user for it anyway) */
1902 if (!no_evolution_in_loop_p (ev, use_loop->num, &val)
1903 || !val)
1904 return chrec_dont_know;
1906 use_loop = use_loop->outer;
1910 /* Returns instantiated value for VERSION in CACHE. */
1912 static tree
1913 get_instantiated_value (htab_t cache, tree version)
1915 struct scev_info_str *info, pattern;
1917 pattern.var = version;
1918 info = htab_find (cache, &pattern);
1920 if (info)
1921 return info->chrec;
1922 else
1923 return NULL_TREE;
1926 /* Sets instantiated value for VERSION to VAL in CACHE. */
1928 static void
1929 set_instantiated_value (htab_t cache, tree version, tree val)
1931 struct scev_info_str *info, pattern;
1932 PTR *slot;
1934 pattern.var = version;
1935 slot = htab_find_slot (cache, &pattern, INSERT);
1937 if (*slot)
1938 info = *slot;
1939 else
1940 info = *slot = new_scev_info_str (version);
1941 info->chrec = val;
1944 /* Analyze all the parameters of the chrec that were left under a symbolic form,
1945 with respect to LOOP. CHREC is the chrec to instantiate. If
1946 ALLOW_SUPERLOOP_CHRECS is true, replacing loop invariants with
1947 outer loop chrecs is done. CACHE is the cache of already instantiated
1948 values. */
1950 static tree
1951 instantiate_parameters_1 (struct loop *loop, tree chrec,
1952 bool allow_superloop_chrecs,
1953 htab_t cache)
1955 tree res, op0, op1, op2;
1956 basic_block def_bb;
1957 struct loop *def_loop;
1959 if (chrec == NULL_TREE
1960 || automatically_generated_chrec_p (chrec))
1961 return chrec;
1963 if (is_gimple_min_invariant (chrec))
1964 return chrec;
1966 switch (TREE_CODE (chrec))
1968 case SSA_NAME:
1969 def_bb = bb_for_stmt (SSA_NAME_DEF_STMT (chrec));
1971 /* A parameter (or loop invariant and we do not want to include
1972 evolutions in outer loops), nothing to do. */
1973 if (!def_bb
1974 || (!allow_superloop_chrecs
1975 && !flow_bb_inside_loop_p (loop, def_bb)))
1976 return chrec;
1978 /* We cache the value of instantiated variable to avoid exponential
1979 time complexity due to reevaluations. We also store the convenient
1980 value in the cache in order to prevent infinite recursion -- we do
1981 not want to instantiate the SSA_NAME if it is in a mixer
1982 structure. This is used for avoiding the instantiation of
1983 recursively defined functions, such as:
1985 | a_2 -> {0, +, 1, +, a_2}_1 */
1987 res = get_instantiated_value (cache, chrec);
1988 if (res)
1989 return res;
1991 /* Store the convenient value for chrec in the structure. If it
1992 is defined outside of the loop, we may just leave it in symbolic
1993 form, otherwise we need to admit that we do not know its behavior
1994 inside the loop. */
1995 res = !flow_bb_inside_loop_p (loop, def_bb) ? chrec : chrec_dont_know;
1996 set_instantiated_value (cache, chrec, res);
1998 /* To make things even more complicated, instantiate_parameters_1
1999 calls analyze_scalar_evolution that may call # of iterations
2000 analysis that may in turn call instantiate_parameters_1 again.
2001 To prevent the infinite recursion, keep also the bitmap of
2002 ssa names that are being instantiated globally. */
2003 if (bitmap_bit_p (already_instantiated, SSA_NAME_VERSION (chrec)))
2004 return res;
2006 def_loop = find_common_loop (loop, def_bb->loop_father);
2008 /* If the analysis yields a parametric chrec, instantiate the
2009 result again. */
2010 bitmap_set_bit (already_instantiated, SSA_NAME_VERSION (chrec));
2011 res = analyze_scalar_evolution (def_loop, chrec);
2012 if (res != chrec_dont_know)
2013 res = instantiate_parameters_1 (loop, res, allow_superloop_chrecs,
2014 cache);
2015 bitmap_clear_bit (already_instantiated, SSA_NAME_VERSION (chrec));
2017 /* Store the correct value to the cache. */
2018 set_instantiated_value (cache, chrec, res);
2019 return res;
2021 case POLYNOMIAL_CHREC:
2022 op0 = instantiate_parameters_1 (loop, CHREC_LEFT (chrec),
2023 allow_superloop_chrecs, cache);
2024 if (op0 == chrec_dont_know)
2025 return chrec_dont_know;
2027 op1 = instantiate_parameters_1 (loop, CHREC_RIGHT (chrec),
2028 allow_superloop_chrecs, cache);
2029 if (op1 == chrec_dont_know)
2030 return chrec_dont_know;
2032 if (CHREC_LEFT (chrec) != op0
2033 || CHREC_RIGHT (chrec) != op1)
2034 chrec = build_polynomial_chrec (CHREC_VARIABLE (chrec), op0, op1);
2035 return chrec;
2037 case PLUS_EXPR:
2038 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2039 allow_superloop_chrecs, cache);
2040 if (op0 == chrec_dont_know)
2041 return chrec_dont_know;
2043 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2044 allow_superloop_chrecs, cache);
2045 if (op1 == chrec_dont_know)
2046 return chrec_dont_know;
2048 if (TREE_OPERAND (chrec, 0) != op0
2049 || TREE_OPERAND (chrec, 1) != op1)
2050 chrec = chrec_fold_plus (TREE_TYPE (chrec), op0, op1);
2051 return chrec;
2053 case MINUS_EXPR:
2054 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2055 allow_superloop_chrecs, cache);
2056 if (op0 == chrec_dont_know)
2057 return chrec_dont_know;
2059 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2060 allow_superloop_chrecs, cache);
2061 if (op1 == chrec_dont_know)
2062 return chrec_dont_know;
2064 if (TREE_OPERAND (chrec, 0) != op0
2065 || TREE_OPERAND (chrec, 1) != op1)
2066 chrec = chrec_fold_minus (TREE_TYPE (chrec), op0, op1);
2067 return chrec;
2069 case MULT_EXPR:
2070 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2071 allow_superloop_chrecs, cache);
2072 if (op0 == chrec_dont_know)
2073 return chrec_dont_know;
2075 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2076 allow_superloop_chrecs, cache);
2077 if (op1 == chrec_dont_know)
2078 return chrec_dont_know;
2080 if (TREE_OPERAND (chrec, 0) != op0
2081 || TREE_OPERAND (chrec, 1) != op1)
2082 chrec = chrec_fold_multiply (TREE_TYPE (chrec), op0, op1);
2083 return chrec;
2085 case NOP_EXPR:
2086 case CONVERT_EXPR:
2087 case NON_LVALUE_EXPR:
2088 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2089 allow_superloop_chrecs, cache);
2090 if (op0 == chrec_dont_know)
2091 return chrec_dont_know;
2093 if (op0 == TREE_OPERAND (chrec, 0))
2094 return chrec;
2096 return chrec_convert (TREE_TYPE (chrec), op0);
2098 case SCEV_NOT_KNOWN:
2099 return chrec_dont_know;
2101 case SCEV_KNOWN:
2102 return chrec_known;
2104 default:
2105 break;
2108 switch (TREE_CODE_LENGTH (TREE_CODE (chrec)))
2110 case 3:
2111 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2112 allow_superloop_chrecs, cache);
2113 if (op0 == chrec_dont_know)
2114 return chrec_dont_know;
2116 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2117 allow_superloop_chrecs, cache);
2118 if (op1 == chrec_dont_know)
2119 return chrec_dont_know;
2121 op2 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 2),
2122 allow_superloop_chrecs, cache);
2123 if (op2 == chrec_dont_know)
2124 return chrec_dont_know;
2126 if (op0 == TREE_OPERAND (chrec, 0)
2127 && op1 == TREE_OPERAND (chrec, 1)
2128 && op2 == TREE_OPERAND (chrec, 2))
2129 return chrec;
2131 return fold (build (TREE_CODE (chrec),
2132 TREE_TYPE (chrec), op0, op1, op2));
2134 case 2:
2135 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2136 allow_superloop_chrecs, cache);
2137 if (op0 == chrec_dont_know)
2138 return chrec_dont_know;
2140 op1 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 1),
2141 allow_superloop_chrecs, cache);
2142 if (op1 == chrec_dont_know)
2143 return chrec_dont_know;
2145 if (op0 == TREE_OPERAND (chrec, 0)
2146 && op1 == TREE_OPERAND (chrec, 1))
2147 return chrec;
2148 return fold (build (TREE_CODE (chrec), TREE_TYPE (chrec), op0, op1));
2150 case 1:
2151 op0 = instantiate_parameters_1 (loop, TREE_OPERAND (chrec, 0),
2152 allow_superloop_chrecs, cache);
2153 if (op0 == chrec_dont_know)
2154 return chrec_dont_know;
2155 if (op0 == TREE_OPERAND (chrec, 0))
2156 return chrec;
2157 return fold (build1 (TREE_CODE (chrec), TREE_TYPE (chrec), op0));
2159 case 0:
2160 return chrec;
2162 default:
2163 break;
2166 /* Too complicated to handle. */
2167 return chrec_dont_know;
2170 /* Analyze all the parameters of the chrec that were left under a
2171 symbolic form. LOOP is the loop in which symbolic names have to
2172 be analyzed and instantiated. */
2174 tree
2175 instantiate_parameters (struct loop *loop,
2176 tree chrec)
2178 tree res;
2179 htab_t cache = htab_create (10, hash_scev_info, eq_scev_info, del_scev_info);
2181 if (dump_file && (dump_flags & TDF_DETAILS))
2183 fprintf (dump_file, "(instantiate_parameters \n");
2184 fprintf (dump_file, " (loop_nb = %d)\n", loop->num);
2185 fprintf (dump_file, " (chrec = ");
2186 print_generic_expr (dump_file, chrec, 0);
2187 fprintf (dump_file, ")\n");
2190 res = instantiate_parameters_1 (loop, chrec, true, cache);
2192 if (dump_file && (dump_flags & TDF_DETAILS))
2194 fprintf (dump_file, " (res = ");
2195 print_generic_expr (dump_file, res, 0);
2196 fprintf (dump_file, "))\n");
2199 htab_delete (cache);
2201 return res;
2204 /* Similar to instantiate_parameters, but does not introduce the
2205 evolutions in outer loops for LOOP invariants in CHREC. */
2207 static tree
2208 resolve_mixers (struct loop *loop, tree chrec)
2210 htab_t cache = htab_create (10, hash_scev_info, eq_scev_info, del_scev_info);
2211 tree ret = instantiate_parameters_1 (loop, chrec, false, cache);
2212 htab_delete (cache);
2213 return ret;
2216 /* Entry point for the analysis of the number of iterations pass.
2217 This function tries to safely approximate the number of iterations
2218 the loop will run. When this property is not decidable at compile
2219 time, the result is chrec_dont_know. Otherwise the result is
2220 a scalar or a symbolic parameter.
2222 Example of analysis: suppose that the loop has an exit condition:
2224 "if (b > 49) goto end_loop;"
2226 and that in a previous analysis we have determined that the
2227 variable 'b' has an evolution function:
2229 "EF = {23, +, 5}_2".
2231 When we evaluate the function at the point 5, i.e. the value of the
2232 variable 'b' after 5 iterations in the loop, we have EF (5) = 48,
2233 and EF (6) = 53. In this case the value of 'b' on exit is '53' and
2234 the loop body has been executed 6 times. */
2236 tree
2237 number_of_iterations_in_loop (struct loop *loop)
2239 tree res, type;
2240 edge exit;
2241 struct tree_niter_desc niter_desc;
2243 /* Determine whether the number_of_iterations_in_loop has already
2244 been computed. */
2245 res = loop->nb_iterations;
2246 if (res)
2247 return res;
2248 res = chrec_dont_know;
2250 if (dump_file && (dump_flags & TDF_DETAILS))
2251 fprintf (dump_file, "(number_of_iterations_in_loop\n");
2253 exit = loop->single_exit;
2254 if (!exit)
2255 goto end;
2257 if (!number_of_iterations_exit (loop, exit, &niter_desc))
2258 goto end;
2260 type = TREE_TYPE (niter_desc.niter);
2261 if (integer_nonzerop (niter_desc.may_be_zero))
2262 res = build_int_cst (type, 0);
2263 else if (integer_zerop (niter_desc.may_be_zero))
2264 res = niter_desc.niter;
2265 else
2266 res = chrec_dont_know;
2268 end:
2269 return set_nb_iterations_in_loop (loop, res);
2272 /* One of the drivers for testing the scalar evolutions analysis.
2273 This function computes the number of iterations for all the loops
2274 from the EXIT_CONDITIONS array. */
2276 static void
2277 number_of_iterations_for_all_loops (varray_type exit_conditions)
2279 unsigned int i;
2280 unsigned nb_chrec_dont_know_loops = 0;
2281 unsigned nb_static_loops = 0;
2283 for (i = 0; i < VARRAY_ACTIVE_SIZE (exit_conditions); i++)
2285 tree res = number_of_iterations_in_loop
2286 (loop_containing_stmt (VARRAY_TREE (exit_conditions, i)));
2287 if (chrec_contains_undetermined (res))
2288 nb_chrec_dont_know_loops++;
2289 else
2290 nb_static_loops++;
2293 if (dump_file)
2295 fprintf (dump_file, "\n(\n");
2296 fprintf (dump_file, "-----------------------------------------\n");
2297 fprintf (dump_file, "%d\tnb_chrec_dont_know_loops\n", nb_chrec_dont_know_loops);
2298 fprintf (dump_file, "%d\tnb_static_loops\n", nb_static_loops);
2299 fprintf (dump_file, "%d\tnb_total_loops\n", current_loops->num);
2300 fprintf (dump_file, "-----------------------------------------\n");
2301 fprintf (dump_file, ")\n\n");
2303 print_loop_ir (dump_file);
2309 /* Counters for the stats. */
2311 struct chrec_stats
2313 unsigned nb_chrecs;
2314 unsigned nb_affine;
2315 unsigned nb_affine_multivar;
2316 unsigned nb_higher_poly;
2317 unsigned nb_chrec_dont_know;
2318 unsigned nb_undetermined;
2321 /* Reset the counters. */
2323 static inline void
2324 reset_chrecs_counters (struct chrec_stats *stats)
2326 stats->nb_chrecs = 0;
2327 stats->nb_affine = 0;
2328 stats->nb_affine_multivar = 0;
2329 stats->nb_higher_poly = 0;
2330 stats->nb_chrec_dont_know = 0;
2331 stats->nb_undetermined = 0;
2334 /* Dump the contents of a CHREC_STATS structure. */
2336 static void
2337 dump_chrecs_stats (FILE *file, struct chrec_stats *stats)
2339 fprintf (file, "\n(\n");
2340 fprintf (file, "-----------------------------------------\n");
2341 fprintf (file, "%d\taffine univariate chrecs\n", stats->nb_affine);
2342 fprintf (file, "%d\taffine multivariate chrecs\n", stats->nb_affine_multivar);
2343 fprintf (file, "%d\tdegree greater than 2 polynomials\n",
2344 stats->nb_higher_poly);
2345 fprintf (file, "%d\tchrec_dont_know chrecs\n", stats->nb_chrec_dont_know);
2346 fprintf (file, "-----------------------------------------\n");
2347 fprintf (file, "%d\ttotal chrecs\n", stats->nb_chrecs);
2348 fprintf (file, "%d\twith undetermined coefficients\n",
2349 stats->nb_undetermined);
2350 fprintf (file, "-----------------------------------------\n");
2351 fprintf (file, "%d\tchrecs in the scev database\n",
2352 (int) htab_elements (scalar_evolution_info));
2353 fprintf (file, "%d\tsets in the scev database\n", nb_set_scev);
2354 fprintf (file, "%d\tgets in the scev database\n", nb_get_scev);
2355 fprintf (file, "-----------------------------------------\n");
2356 fprintf (file, ")\n\n");
2359 /* Gather statistics about CHREC. */
2361 static void
2362 gather_chrec_stats (tree chrec, struct chrec_stats *stats)
2364 if (dump_file && (dump_flags & TDF_STATS))
2366 fprintf (dump_file, "(classify_chrec ");
2367 print_generic_expr (dump_file, chrec, 0);
2368 fprintf (dump_file, "\n");
2371 stats->nb_chrecs++;
2373 if (chrec == NULL_TREE)
2375 stats->nb_undetermined++;
2376 return;
2379 switch (TREE_CODE (chrec))
2381 case POLYNOMIAL_CHREC:
2382 if (evolution_function_is_affine_p (chrec))
2384 if (dump_file && (dump_flags & TDF_STATS))
2385 fprintf (dump_file, " affine_univariate\n");
2386 stats->nb_affine++;
2388 else if (evolution_function_is_affine_multivariate_p (chrec))
2390 if (dump_file && (dump_flags & TDF_STATS))
2391 fprintf (dump_file, " affine_multivariate\n");
2392 stats->nb_affine_multivar++;
2394 else
2396 if (dump_file && (dump_flags & TDF_STATS))
2397 fprintf (dump_file, " higher_degree_polynomial\n");
2398 stats->nb_higher_poly++;
2401 break;
2403 default:
2404 break;
2407 if (chrec_contains_undetermined (chrec))
2409 if (dump_file && (dump_flags & TDF_STATS))
2410 fprintf (dump_file, " undetermined\n");
2411 stats->nb_undetermined++;
2414 if (dump_file && (dump_flags & TDF_STATS))
2415 fprintf (dump_file, ")\n");
2418 /* One of the drivers for testing the scalar evolutions analysis.
2419 This function analyzes the scalar evolution of all the scalars
2420 defined as loop phi nodes in one of the loops from the
2421 EXIT_CONDITIONS array.
2423 TODO Optimization: A loop is in canonical form if it contains only
2424 a single scalar loop phi node. All the other scalars that have an
2425 evolution in the loop are rewritten in function of this single
2426 index. This allows the parallelization of the loop. */
2428 static void
2429 analyze_scalar_evolution_for_all_loop_phi_nodes (varray_type exit_conditions)
2431 unsigned int i;
2432 struct chrec_stats stats;
2434 reset_chrecs_counters (&stats);
2436 for (i = 0; i < VARRAY_ACTIVE_SIZE (exit_conditions); i++)
2438 struct loop *loop;
2439 basic_block bb;
2440 tree phi, chrec;
2442 loop = loop_containing_stmt (VARRAY_TREE (exit_conditions, i));
2443 bb = loop->header;
2445 for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
2446 if (is_gimple_reg (PHI_RESULT (phi)))
2448 chrec = instantiate_parameters
2449 (loop,
2450 analyze_scalar_evolution (loop, PHI_RESULT (phi)));
2452 if (dump_file && (dump_flags & TDF_STATS))
2453 gather_chrec_stats (chrec, &stats);
2457 if (dump_file && (dump_flags & TDF_STATS))
2458 dump_chrecs_stats (dump_file, &stats);
2461 /* Callback for htab_traverse, gathers information on chrecs in the
2462 hashtable. */
2464 static int
2465 gather_stats_on_scev_database_1 (void **slot, void *stats)
2467 struct scev_info_str *entry = *slot;
2469 gather_chrec_stats (entry->chrec, stats);
2471 return 1;
2474 /* Classify the chrecs of the whole database. */
2476 void
2477 gather_stats_on_scev_database (void)
2479 struct chrec_stats stats;
2481 if (!dump_file)
2482 return;
2484 reset_chrecs_counters (&stats);
2486 htab_traverse (scalar_evolution_info, gather_stats_on_scev_database_1,
2487 &stats);
2489 dump_chrecs_stats (dump_file, &stats);
2494 /* Initializer. */
2496 static void
2497 initialize_scalar_evolutions_analyzer (void)
2499 /* The elements below are unique. */
2500 if (chrec_dont_know == NULL_TREE)
2502 chrec_not_analyzed_yet = NULL_TREE;
2503 chrec_dont_know = make_node (SCEV_NOT_KNOWN);
2504 chrec_known = make_node (SCEV_KNOWN);
2505 TREE_TYPE (chrec_dont_know) = NULL_TREE;
2506 TREE_TYPE (chrec_known) = NULL_TREE;
2510 /* Initialize the analysis of scalar evolutions for LOOPS. */
2512 void
2513 scev_initialize (struct loops *loops)
2515 unsigned i;
2516 current_loops = loops;
2518 scalar_evolution_info = htab_create (100, hash_scev_info,
2519 eq_scev_info, del_scev_info);
2520 already_instantiated = BITMAP_ALLOC (NULL);
2522 initialize_scalar_evolutions_analyzer ();
2524 for (i = 1; i < loops->num; i++)
2525 if (loops->parray[i])
2526 loops->parray[i]->nb_iterations = NULL_TREE;
2529 /* Cleans up the information cached by the scalar evolutions analysis. */
2531 void
2532 scev_reset (void)
2534 unsigned i;
2535 struct loop *loop;
2537 if (!scalar_evolution_info || !current_loops)
2538 return;
2540 htab_empty (scalar_evolution_info);
2541 for (i = 1; i < current_loops->num; i++)
2543 loop = current_loops->parray[i];
2544 if (loop)
2545 loop->nb_iterations = NULL_TREE;
2549 /* Checks whether OP behaves as a simple affine iv of LOOP in STMT and returns
2550 its BASE and STEP if possible. */
2552 bool
2553 simple_iv (struct loop *loop, tree stmt, tree op, tree *base, tree *step)
2555 basic_block bb = bb_for_stmt (stmt);
2556 tree type, ev;
2558 *base = NULL_TREE;
2559 *step = NULL_TREE;
2561 type = TREE_TYPE (op);
2562 if (TREE_CODE (type) != INTEGER_TYPE
2563 && TREE_CODE (type) != POINTER_TYPE)
2564 return false;
2566 ev = analyze_scalar_evolution_in_loop (loop, bb->loop_father, op);
2567 if (chrec_contains_undetermined (ev))
2568 return false;
2570 if (tree_does_not_contain_chrecs (ev)
2571 && !chrec_contains_symbols_defined_in_loop (ev, loop->num))
2573 *base = ev;
2574 return true;
2577 if (TREE_CODE (ev) != POLYNOMIAL_CHREC
2578 || CHREC_VARIABLE (ev) != (unsigned) loop->num)
2579 return false;
2581 *step = CHREC_RIGHT (ev);
2582 if (TREE_CODE (*step) != INTEGER_CST)
2583 return false;
2584 *base = CHREC_LEFT (ev);
2585 if (tree_contains_chrecs (*base)
2586 || chrec_contains_symbols_defined_in_loop (*base, loop->num))
2587 return false;
2589 return true;
2592 /* Runs the analysis of scalar evolutions. */
2594 void
2595 scev_analysis (void)
2597 varray_type exit_conditions;
2599 VARRAY_GENERIC_PTR_INIT (exit_conditions, 37, "exit_conditions");
2600 select_loops_exit_conditions (current_loops, &exit_conditions);
2602 if (dump_file && (dump_flags & TDF_STATS))
2603 analyze_scalar_evolution_for_all_loop_phi_nodes (exit_conditions);
2605 number_of_iterations_for_all_loops (exit_conditions);
2606 VARRAY_CLEAR (exit_conditions);
2609 /* Finalize the scalar evolution analysis. */
2611 void
2612 scev_finalize (void)
2614 htab_delete (scalar_evolution_info);
2615 BITMAP_FREE (already_instantiated);