Fix DealII type problems.
[official-gcc/Ramakrishna.git] / gcc / tree-scalar-evolution.c
blob8b442e549945144be6f813c475a11977a5fd73d8
1 /* Scalar evolution detector.
2 Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009
3 Free Software Foundation, Inc.
4 Contributed by Sebastian Pop <s.pop@laposte.net>
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify it under
9 the terms of the GNU General Public License as published by the Free
10 Software Foundation; either version 3, or (at your option) any later
11 version.
13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 for more details.
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
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 GIMPLE_ASSIGN: 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 (loop_1, {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 2a: 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 2b: Multivariate chains of recurrences.
160 | loop_1
161 | k = phi (0, k + 1)
162 | loop_2 4 times
163 | j = phi (0, j + 1)
164 | loop_3 4 times
165 | i = phi (0, i + 1)
166 | A[j + k] = ...
167 | endloop
168 | endloop
169 | endloop
171 Analyzing the access function of array A with
172 instantiate_parameters (loop_1, "j + k"), we obtain the
173 instantiation and the analysis of the scalar variables "j" and "k"
174 in loop_1. This leads to the scalar evolution {4, +, 1}_1: the end
175 value of loop_2 for "j" is 4, and the evolution of "k" in loop_1 is
176 {0, +, 1}_1. To obtain the evolution function in loop_3 and
177 instantiate the scalar variables up to loop_1, one has to use:
178 instantiate_scev (block_before_loop (loop_1), loop_3, "j + k").
179 The result of this call is {{0, +, 1}_1, +, 1}_2.
181 Example 3: Higher degree polynomials.
183 | loop_1
184 | a = phi (2, b)
185 | c = phi (5, d)
186 | b = a + 1
187 | d = c + a
188 | endloop
190 a -> {2, +, 1}_1
191 b -> {3, +, 1}_1
192 c -> {5, +, a}_1
193 d -> {5 + a, +, a}_1
195 instantiate_parameters (loop_1, {5, +, a}_1) -> {5, +, 2, +, 1}_1
196 instantiate_parameters (loop_1, {5 + a, +, a}_1) -> {7, +, 3, +, 1}_1
198 Example 4: Lucas, Fibonacci, or mixers in general.
200 | loop_1
201 | a = phi (1, b)
202 | c = phi (3, d)
203 | b = c
204 | d = c + a
205 | endloop
207 a -> (1, c)_1
208 c -> {3, +, a}_1
210 The syntax "(1, c)_1" stands for a PEELED_CHREC that has the
211 following semantics: during the first iteration of the loop_1, the
212 variable contains the value 1, and then it contains the value "c".
213 Note that this syntax is close to the syntax of the loop-phi-node:
214 "a -> (1, c)_1" vs. "a = phi (1, c)".
216 The symbolic chrec representation contains all the semantics of the
217 original code. What is more difficult is to use this information.
219 Example 5: Flip-flops, or exchangers.
221 | loop_1
222 | a = phi (1, b)
223 | c = phi (3, d)
224 | b = c
225 | d = a
226 | endloop
228 a -> (1, c)_1
229 c -> (3, a)_1
231 Based on these symbolic chrecs, it is possible to refine this
232 information into the more precise PERIODIC_CHRECs:
234 a -> |1, 3|_1
235 c -> |3, 1|_1
237 This transformation is not yet implemented.
239 Further readings:
241 You can find a more detailed description of the algorithm in:
242 http://icps.u-strasbg.fr/~pop/DEA_03_Pop.pdf
243 http://icps.u-strasbg.fr/~pop/DEA_03_Pop.ps.gz. But note that
244 this is a preliminary report and some of the details of the
245 algorithm have changed. I'm working on a research report that
246 updates the description of the algorithms to reflect the design
247 choices used in this implementation.
249 A set of slides show a high level overview of the algorithm and run
250 an example through the scalar evolution analyzer:
251 http://cri.ensmp.fr/~pop/gcc/mar04/slides.pdf
253 The slides that I have presented at the GCC Summit'04 are available
254 at: http://cri.ensmp.fr/~pop/gcc/20040604/gccsummit-lno-spop.pdf
257 #include "config.h"
258 #include "system.h"
259 #include "coretypes.h"
260 #include "tm.h"
261 #include "ggc.h"
262 #include "tree.h"
263 #include "real.h"
265 /* These RTL headers are needed for basic-block.h. */
266 #include "rtl.h"
267 #include "basic-block.h"
268 #include "diagnostic.h"
269 #include "tree-flow.h"
270 #include "tree-dump.h"
271 #include "timevar.h"
272 #include "cfgloop.h"
273 #include "tree-chrec.h"
274 #include "tree-scalar-evolution.h"
275 #include "tree-pass.h"
276 #include "flags.h"
277 #include "params.h"
279 static tree analyze_scalar_evolution_1 (struct loop *, tree, tree);
281 /* The cached information about an SSA name VAR, claiming that below
282 basic block INSTANTIATED_BELOW, the value of VAR can be expressed
283 as CHREC. */
285 struct GTY(()) scev_info_str {
286 basic_block instantiated_below;
287 tree var;
288 tree chrec;
291 /* Counters for the scev database. */
292 static unsigned nb_set_scev = 0;
293 static unsigned nb_get_scev = 0;
295 /* The following trees are unique elements. Thus the comparison of
296 another element to these elements should be done on the pointer to
297 these trees, and not on their value. */
299 /* The SSA_NAMEs that are not yet analyzed are qualified with NULL_TREE. */
300 tree chrec_not_analyzed_yet;
302 /* Reserved to the cases where the analyzer has detected an
303 undecidable property at compile time. */
304 tree chrec_dont_know;
306 /* When the analyzer has detected that a property will never
307 happen, then it qualifies it with chrec_known. */
308 tree chrec_known;
310 static GTY ((param_is (struct scev_info_str))) htab_t scalar_evolution_info;
313 /* Constructs a new SCEV_INFO_STR structure for VAR and INSTANTIATED_BELOW. */
315 static inline struct scev_info_str *
316 new_scev_info_str (basic_block instantiated_below, tree var)
318 struct scev_info_str *res;
320 res = GGC_NEW (struct scev_info_str);
321 res->var = var;
322 res->chrec = chrec_not_analyzed_yet;
323 res->instantiated_below = instantiated_below;
325 return res;
328 /* Computes a hash function for database element ELT. */
330 static hashval_t
331 hash_scev_info (const void *elt)
333 return SSA_NAME_VERSION (((const struct scev_info_str *) elt)->var);
336 /* Compares database elements E1 and E2. */
338 static int
339 eq_scev_info (const void *e1, const void *e2)
341 const struct scev_info_str *elt1 = (const struct scev_info_str *) e1;
342 const struct scev_info_str *elt2 = (const struct scev_info_str *) e2;
344 return (elt1->var == elt2->var
345 && elt1->instantiated_below == elt2->instantiated_below);
348 /* Deletes database element E. */
350 static void
351 del_scev_info (void *e)
353 ggc_free (e);
356 /* Get the scalar evolution of VAR for INSTANTIATED_BELOW basic block.
357 A first query on VAR returns chrec_not_analyzed_yet. */
359 static tree *
360 find_var_scev_info (basic_block instantiated_below, tree var)
362 struct scev_info_str *res;
363 struct scev_info_str tmp;
364 PTR *slot;
366 tmp.var = var;
367 tmp.instantiated_below = instantiated_below;
368 slot = htab_find_slot (scalar_evolution_info, &tmp, INSERT);
370 if (!*slot)
371 *slot = new_scev_info_str (instantiated_below, var);
372 res = (struct scev_info_str *) *slot;
374 return &res->chrec;
377 /* Return true when CHREC contains symbolic names defined in
378 LOOP_NB. */
380 bool
381 chrec_contains_symbols_defined_in_loop (const_tree chrec, unsigned loop_nb)
383 int i, n;
385 if (chrec == NULL_TREE)
386 return false;
388 if (is_gimple_min_invariant (chrec))
389 return false;
391 if (TREE_CODE (chrec) == VAR_DECL
392 || TREE_CODE (chrec) == PARM_DECL
393 || TREE_CODE (chrec) == FUNCTION_DECL
394 || TREE_CODE (chrec) == LABEL_DECL
395 || TREE_CODE (chrec) == RESULT_DECL
396 || TREE_CODE (chrec) == FIELD_DECL)
397 return true;
399 if (TREE_CODE (chrec) == SSA_NAME)
401 gimple def = SSA_NAME_DEF_STMT (chrec);
402 struct loop *def_loop = loop_containing_stmt (def);
403 struct loop *loop = get_loop (loop_nb);
405 if (def_loop == NULL)
406 return false;
408 if (loop == def_loop || flow_loop_nested_p (loop, def_loop))
409 return true;
411 return false;
414 n = TREE_OPERAND_LENGTH (chrec);
415 for (i = 0; i < n; i++)
416 if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec, i),
417 loop_nb))
418 return true;
419 return false;
422 /* Return true when PHI is a loop-phi-node. */
424 static bool
425 loop_phi_node_p (gimple phi)
427 /* The implementation of this function is based on the following
428 property: "all the loop-phi-nodes of a loop are contained in the
429 loop's header basic block". */
431 return loop_containing_stmt (phi)->header == gimple_bb (phi);
434 /* Compute the scalar evolution for EVOLUTION_FN after crossing LOOP.
435 In general, in the case of multivariate evolutions we want to get
436 the evolution in different loops. LOOP specifies the level for
437 which to get the evolution.
439 Example:
441 | for (j = 0; j < 100; j++)
443 | for (k = 0; k < 100; k++)
445 | i = k + j; - Here the value of i is a function of j, k.
447 | ... = i - Here the value of i is a function of j.
449 | ... = i - Here the value of i is a scalar.
451 Example:
453 | i_0 = ...
454 | loop_1 10 times
455 | i_1 = phi (i_0, i_2)
456 | i_2 = i_1 + 2
457 | endloop
459 This loop has the same effect as:
460 LOOP_1 has the same effect as:
462 | i_1 = i_0 + 20
464 The overall effect of the loop, "i_0 + 20" in the previous example,
465 is obtained by passing in the parameters: LOOP = 1,
466 EVOLUTION_FN = {i_0, +, 2}_1.
469 tree
470 compute_overall_effect_of_inner_loop (struct loop *loop, tree evolution_fn)
472 bool val = false;
474 if (evolution_fn == chrec_dont_know)
475 return chrec_dont_know;
477 else if (TREE_CODE (evolution_fn) == POLYNOMIAL_CHREC)
479 struct loop *inner_loop = get_chrec_loop (evolution_fn);
481 if (inner_loop == loop
482 || flow_loop_nested_p (loop, inner_loop))
484 tree nb_iter = number_of_latch_executions (inner_loop);
486 if (nb_iter == chrec_dont_know)
487 return chrec_dont_know;
488 else
490 tree res;
492 /* evolution_fn is the evolution function in LOOP. Get
493 its value in the nb_iter-th iteration. */
494 res = chrec_apply (inner_loop->num, evolution_fn, nb_iter);
496 if (chrec_contains_symbols_defined_in_loop (res, loop->num))
497 res = instantiate_parameters (loop, res);
499 /* Continue the computation until ending on a parent of LOOP. */
500 return compute_overall_effect_of_inner_loop (loop, res);
503 else
504 return evolution_fn;
507 /* If the evolution function is an invariant, there is nothing to do. */
508 else if (no_evolution_in_loop_p (evolution_fn, loop->num, &val) && val)
509 return evolution_fn;
511 else
512 return chrec_dont_know;
515 /* Determine whether the CHREC is always positive/negative. If the expression
516 cannot be statically analyzed, return false, otherwise set the answer into
517 VALUE. */
519 bool
520 chrec_is_positive (tree chrec, bool *value)
522 bool value0, value1, value2;
523 tree end_value, nb_iter;
525 switch (TREE_CODE (chrec))
527 case POLYNOMIAL_CHREC:
528 if (!chrec_is_positive (CHREC_LEFT (chrec), &value0)
529 || !chrec_is_positive (CHREC_RIGHT (chrec), &value1))
530 return false;
532 /* FIXME -- overflows. */
533 if (value0 == value1)
535 *value = value0;
536 return true;
539 /* Otherwise the chrec is under the form: "{-197, +, 2}_1",
540 and the proof consists in showing that the sign never
541 changes during the execution of the loop, from 0 to
542 loop->nb_iterations. */
543 if (!evolution_function_is_affine_p (chrec))
544 return false;
546 nb_iter = number_of_latch_executions (get_chrec_loop (chrec));
547 if (chrec_contains_undetermined (nb_iter))
548 return false;
550 #if 0
551 /* TODO -- If the test is after the exit, we may decrease the number of
552 iterations by one. */
553 if (after_exit)
554 nb_iter = chrec_fold_minus (type, nb_iter, build_int_cst (type, 1));
555 #endif
557 end_value = chrec_apply (CHREC_VARIABLE (chrec), chrec, nb_iter);
559 if (!chrec_is_positive (end_value, &value2))
560 return false;
562 *value = value0;
563 return value0 == value1;
565 case INTEGER_CST:
566 *value = (tree_int_cst_sgn (chrec) == 1);
567 return true;
569 default:
570 return false;
574 /* Associate CHREC to SCALAR. */
576 static void
577 set_scalar_evolution (basic_block instantiated_below, tree scalar, tree chrec)
579 tree *scalar_info;
581 if (TREE_CODE (scalar) != SSA_NAME)
582 return;
584 scalar_info = find_var_scev_info (instantiated_below, scalar);
586 if (dump_file)
588 if (dump_flags & TDF_DETAILS)
590 fprintf (dump_file, "(set_scalar_evolution \n");
591 fprintf (dump_file, " instantiated_below = %d \n",
592 instantiated_below->index);
593 fprintf (dump_file, " (scalar = ");
594 print_generic_expr (dump_file, scalar, 0);
595 fprintf (dump_file, ")\n (scalar_evolution = ");
596 print_generic_expr (dump_file, chrec, 0);
597 fprintf (dump_file, "))\n");
599 if (dump_flags & TDF_STATS)
600 nb_set_scev++;
603 *scalar_info = chrec;
606 /* Retrieve the chrec associated to SCALAR instantiated below
607 INSTANTIATED_BELOW block. */
609 static tree
610 get_scalar_evolution (basic_block instantiated_below, tree scalar)
612 tree res;
614 if (dump_file)
616 if (dump_flags & TDF_DETAILS)
618 fprintf (dump_file, "(get_scalar_evolution \n");
619 fprintf (dump_file, " (scalar = ");
620 print_generic_expr (dump_file, scalar, 0);
621 fprintf (dump_file, ")\n");
623 if (dump_flags & TDF_STATS)
624 nb_get_scev++;
627 switch (TREE_CODE (scalar))
629 case SSA_NAME:
630 res = *find_var_scev_info (instantiated_below, scalar);
631 break;
633 case REAL_CST:
634 case FIXED_CST:
635 case INTEGER_CST:
636 res = scalar;
637 break;
639 default:
640 res = chrec_not_analyzed_yet;
641 break;
644 if (dump_file && (dump_flags & TDF_DETAILS))
646 fprintf (dump_file, " (scalar_evolution = ");
647 print_generic_expr (dump_file, res, 0);
648 fprintf (dump_file, "))\n");
651 return res;
654 /* Helper function for add_to_evolution. Returns the evolution
655 function for an assignment of the form "a = b + c", where "a" and
656 "b" are on the strongly connected component. CHREC_BEFORE is the
657 information that we already have collected up to this point.
658 TO_ADD is the evolution of "c".
660 When CHREC_BEFORE has an evolution part in LOOP_NB, add to this
661 evolution the expression TO_ADD, otherwise construct an evolution
662 part for this loop. */
664 static tree
665 add_to_evolution_1 (unsigned loop_nb, tree chrec_before, tree to_add,
666 gimple at_stmt)
668 tree type, left, right;
669 struct loop *loop = get_loop (loop_nb), *chloop;
671 switch (TREE_CODE (chrec_before))
673 case POLYNOMIAL_CHREC:
674 chloop = get_chrec_loop (chrec_before);
675 if (chloop == loop
676 || flow_loop_nested_p (chloop, loop))
678 unsigned var;
680 type = chrec_type (chrec_before);
682 /* When there is no evolution part in this loop, build it. */
683 if (chloop != loop)
685 var = loop_nb;
686 left = chrec_before;
687 right = SCALAR_FLOAT_TYPE_P (type)
688 ? build_real (type, dconst0)
689 : build_int_cst (type, 0);
691 else
693 var = CHREC_VARIABLE (chrec_before);
694 left = CHREC_LEFT (chrec_before);
695 right = CHREC_RIGHT (chrec_before);
698 to_add = chrec_convert (type, to_add, at_stmt);
699 right = chrec_convert_rhs (type, right, at_stmt);
700 right = chrec_fold_plus (chrec_type (right), right, to_add);
701 return build_polynomial_chrec (var, left, right);
703 else
705 gcc_assert (flow_loop_nested_p (loop, chloop));
707 /* Search the evolution in LOOP_NB. */
708 left = add_to_evolution_1 (loop_nb, CHREC_LEFT (chrec_before),
709 to_add, at_stmt);
710 right = CHREC_RIGHT (chrec_before);
711 right = chrec_convert_rhs (chrec_type (left), right, at_stmt);
712 return build_polynomial_chrec (CHREC_VARIABLE (chrec_before),
713 left, right);
716 default:
717 /* These nodes do not depend on a loop. */
718 if (chrec_before == chrec_dont_know)
719 return chrec_dont_know;
721 left = chrec_before;
722 right = chrec_convert_rhs (chrec_type (left), to_add, at_stmt);
723 return build_polynomial_chrec (loop_nb, left, right);
727 /* Add TO_ADD to the evolution part of CHREC_BEFORE in the dimension
728 of LOOP_NB.
730 Description (provided for completeness, for those who read code in
731 a plane, and for my poor 62 bytes brain that would have forgotten
732 all this in the next two or three months):
734 The algorithm of translation of programs from the SSA representation
735 into the chrecs syntax is based on a pattern matching. After having
736 reconstructed the overall tree expression for a loop, there are only
737 two cases that can arise:
739 1. a = loop-phi (init, a + expr)
740 2. a = loop-phi (init, expr)
742 where EXPR is either a scalar constant with respect to the analyzed
743 loop (this is a degree 0 polynomial), or an expression containing
744 other loop-phi definitions (these are higher degree polynomials).
746 Examples:
749 | init = ...
750 | loop_1
751 | a = phi (init, a + 5)
752 | endloop
755 | inita = ...
756 | initb = ...
757 | loop_1
758 | a = phi (inita, 2 * b + 3)
759 | b = phi (initb, b + 1)
760 | endloop
762 For the first case, the semantics of the SSA representation is:
764 | a (x) = init + \sum_{j = 0}^{x - 1} expr (j)
766 that is, there is a loop index "x" that determines the scalar value
767 of the variable during the loop execution. During the first
768 iteration, the value is that of the initial condition INIT, while
769 during the subsequent iterations, it is the sum of the initial
770 condition with the sum of all the values of EXPR from the initial
771 iteration to the before last considered iteration.
773 For the second case, the semantics of the SSA program is:
775 | a (x) = init, if x = 0;
776 | expr (x - 1), otherwise.
778 The second case corresponds to the PEELED_CHREC, whose syntax is
779 close to the syntax of a loop-phi-node:
781 | phi (init, expr) vs. (init, expr)_x
783 The proof of the translation algorithm for the first case is a
784 proof by structural induction based on the degree of EXPR.
786 Degree 0:
787 When EXPR is a constant with respect to the analyzed loop, or in
788 other words when EXPR is a polynomial of degree 0, the evolution of
789 the variable A in the loop is an affine function with an initial
790 condition INIT, and a step EXPR. In order to show this, we start
791 from the semantics of the SSA representation:
793 f (x) = init + \sum_{j = 0}^{x - 1} expr (j)
795 and since "expr (j)" is a constant with respect to "j",
797 f (x) = init + x * expr
799 Finally, based on the semantics of the pure sum chrecs, by
800 identification we get the corresponding chrecs syntax:
802 f (x) = init * \binom{x}{0} + expr * \binom{x}{1}
803 f (x) -> {init, +, expr}_x
805 Higher degree:
806 Suppose that EXPR is a polynomial of degree N with respect to the
807 analyzed loop_x for which we have already determined that it is
808 written under the chrecs syntax:
810 | expr (x) -> {b_0, +, b_1, +, ..., +, b_{n-1}} (x)
812 We start from the semantics of the SSA program:
814 | f (x) = init + \sum_{j = 0}^{x - 1} expr (j)
816 | f (x) = init + \sum_{j = 0}^{x - 1}
817 | (b_0 * \binom{j}{0} + ... + b_{n-1} * \binom{j}{n-1})
819 | f (x) = init + \sum_{j = 0}^{x - 1}
820 | \sum_{k = 0}^{n - 1} (b_k * \binom{j}{k})
822 | f (x) = init + \sum_{k = 0}^{n - 1}
823 | (b_k * \sum_{j = 0}^{x - 1} \binom{j}{k})
825 | f (x) = init + \sum_{k = 0}^{n - 1}
826 | (b_k * \binom{x}{k + 1})
828 | f (x) = init + b_0 * \binom{x}{1} + ...
829 | + b_{n-1} * \binom{x}{n}
831 | f (x) = init * \binom{x}{0} + b_0 * \binom{x}{1} + ...
832 | + b_{n-1} * \binom{x}{n}
835 And finally from the definition of the chrecs syntax, we identify:
836 | f (x) -> {init, +, b_0, +, ..., +, b_{n-1}}_x
838 This shows the mechanism that stands behind the add_to_evolution
839 function. An important point is that the use of symbolic
840 parameters avoids the need of an analysis schedule.
842 Example:
844 | inita = ...
845 | initb = ...
846 | loop_1
847 | a = phi (inita, a + 2 + b)
848 | b = phi (initb, b + 1)
849 | endloop
851 When analyzing "a", the algorithm keeps "b" symbolically:
853 | a -> {inita, +, 2 + b}_1
855 Then, after instantiation, the analyzer ends on the evolution:
857 | a -> {inita, +, 2 + initb, +, 1}_1
861 static tree
862 add_to_evolution (unsigned loop_nb, tree chrec_before, enum tree_code code,
863 tree to_add, gimple at_stmt)
865 tree type = chrec_type (to_add);
866 tree res = NULL_TREE;
868 if (to_add == NULL_TREE)
869 return chrec_before;
871 /* TO_ADD is either a scalar, or a parameter. TO_ADD is not
872 instantiated at this point. */
873 if (TREE_CODE (to_add) == POLYNOMIAL_CHREC)
874 /* This should not happen. */
875 return chrec_dont_know;
877 if (dump_file && (dump_flags & TDF_DETAILS))
879 fprintf (dump_file, "(add_to_evolution \n");
880 fprintf (dump_file, " (loop_nb = %d)\n", loop_nb);
881 fprintf (dump_file, " (chrec_before = ");
882 print_generic_expr (dump_file, chrec_before, 0);
883 fprintf (dump_file, ")\n (to_add = ");
884 print_generic_expr (dump_file, to_add, 0);
885 fprintf (dump_file, ")\n");
888 if (code == MINUS_EXPR)
889 to_add = chrec_fold_multiply (type, to_add, SCALAR_FLOAT_TYPE_P (type)
890 ? build_real (type, dconstm1)
891 : build_int_cst_type (type, -1));
893 res = add_to_evolution_1 (loop_nb, chrec_before, to_add, at_stmt);
895 if (dump_file && (dump_flags & TDF_DETAILS))
897 fprintf (dump_file, " (res = ");
898 print_generic_expr (dump_file, res, 0);
899 fprintf (dump_file, "))\n");
902 return res;
905 /* Helper function. */
907 static inline tree
908 set_nb_iterations_in_loop (struct loop *loop,
909 tree res)
911 if (dump_file && (dump_flags & TDF_DETAILS))
913 fprintf (dump_file, " (set_nb_iterations_in_loop = ");
914 print_generic_expr (dump_file, res, 0);
915 fprintf (dump_file, "))\n");
918 loop->nb_iterations = res;
919 return res;
924 /* This section selects the loops that will be good candidates for the
925 scalar evolution analysis. For the moment, greedily select all the
926 loop nests we could analyze. */
928 /* For a loop with a single exit edge, return the COND_EXPR that
929 guards the exit edge. If the expression is too difficult to
930 analyze, then give up. */
932 gimple
933 get_loop_exit_condition (const struct loop *loop)
935 gimple res = NULL;
936 edge exit_edge = single_exit (loop);
938 if (dump_file && (dump_flags & TDF_DETAILS))
939 fprintf (dump_file, "(get_loop_exit_condition \n ");
941 if (exit_edge)
943 gimple stmt;
945 stmt = last_stmt (exit_edge->src);
946 if (gimple_code (stmt) == GIMPLE_COND)
947 res = stmt;
950 if (dump_file && (dump_flags & TDF_DETAILS))
952 print_gimple_stmt (dump_file, res, 0, 0);
953 fprintf (dump_file, ")\n");
956 return res;
959 /* Recursively determine and enqueue the exit conditions for a loop. */
961 static void
962 get_exit_conditions_rec (struct loop *loop,
963 VEC(gimple,heap) **exit_conditions)
965 if (!loop)
966 return;
968 /* Recurse on the inner loops, then on the next (sibling) loops. */
969 get_exit_conditions_rec (loop->inner, exit_conditions);
970 get_exit_conditions_rec (loop->next, exit_conditions);
972 if (single_exit (loop))
974 gimple loop_condition = get_loop_exit_condition (loop);
976 if (loop_condition)
977 VEC_safe_push (gimple, heap, *exit_conditions, loop_condition);
981 /* Select the candidate loop nests for the analysis. This function
982 initializes the EXIT_CONDITIONS array. */
984 static void
985 select_loops_exit_conditions (VEC(gimple,heap) **exit_conditions)
987 struct loop *function_body = current_loops->tree_root;
989 get_exit_conditions_rec (function_body->inner, exit_conditions);
993 /* Depth first search algorithm. */
995 typedef enum t_bool {
996 t_false,
997 t_true,
998 t_dont_know
999 } t_bool;
1002 static t_bool follow_ssa_edge (struct loop *loop, gimple, gimple, tree *, int);
1004 /* Follow the ssa edge into the binary expression RHS0 CODE RHS1.
1005 Return true if the strongly connected component has been found. */
1007 static t_bool
1008 follow_ssa_edge_binary (struct loop *loop, gimple at_stmt,
1009 tree type, tree rhs0, enum tree_code code, tree rhs1,
1010 gimple halting_phi, tree *evolution_of_loop, int limit)
1012 t_bool res = t_false;
1013 tree evol;
1015 switch (code)
1017 case POINTER_PLUS_EXPR:
1018 case PLUS_EXPR:
1019 if (TREE_CODE (rhs0) == SSA_NAME)
1021 if (TREE_CODE (rhs1) == SSA_NAME)
1023 /* Match an assignment under the form:
1024 "a = b + c". */
1026 /* We want only assignments of form "name + name" contribute to
1027 LIMIT, as the other cases do not necessarily contribute to
1028 the complexity of the expression. */
1029 limit++;
1031 evol = *evolution_of_loop;
1032 res = follow_ssa_edge
1033 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi, &evol, limit);
1035 if (res == t_true)
1036 *evolution_of_loop = add_to_evolution
1037 (loop->num,
1038 chrec_convert (type, evol, at_stmt),
1039 code, rhs1, at_stmt);
1041 else if (res == t_false)
1043 res = follow_ssa_edge
1044 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1045 evolution_of_loop, limit);
1047 if (res == t_true)
1048 *evolution_of_loop = add_to_evolution
1049 (loop->num,
1050 chrec_convert (type, *evolution_of_loop, at_stmt),
1051 code, rhs0, at_stmt);
1053 else if (res == t_dont_know)
1054 *evolution_of_loop = chrec_dont_know;
1057 else if (res == t_dont_know)
1058 *evolution_of_loop = chrec_dont_know;
1061 else
1063 /* Match an assignment under the form:
1064 "a = b + ...". */
1065 res = follow_ssa_edge
1066 (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1067 evolution_of_loop, limit);
1068 if (res == t_true)
1069 *evolution_of_loop = add_to_evolution
1070 (loop->num, chrec_convert (type, *evolution_of_loop,
1071 at_stmt),
1072 code, rhs1, at_stmt);
1074 else if (res == t_dont_know)
1075 *evolution_of_loop = chrec_dont_know;
1079 else if (TREE_CODE (rhs1) == SSA_NAME)
1081 /* Match an assignment under the form:
1082 "a = ... + c". */
1083 res = follow_ssa_edge
1084 (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi,
1085 evolution_of_loop, limit);
1086 if (res == t_true)
1087 *evolution_of_loop = add_to_evolution
1088 (loop->num, chrec_convert (type, *evolution_of_loop,
1089 at_stmt),
1090 code, rhs0, at_stmt);
1092 else if (res == t_dont_know)
1093 *evolution_of_loop = chrec_dont_know;
1096 else
1097 /* Otherwise, match an assignment under the form:
1098 "a = ... + ...". */
1099 /* And there is nothing to do. */
1100 res = t_false;
1101 break;
1103 case MINUS_EXPR:
1104 /* This case is under the form "opnd0 = rhs0 - rhs1". */
1105 if (TREE_CODE (rhs0) == SSA_NAME)
1107 /* Match an assignment under the form:
1108 "a = b - ...". */
1110 /* We want only assignments of form "name - name" contribute to
1111 LIMIT, as the other cases do not necessarily contribute to
1112 the complexity of the expression. */
1113 if (TREE_CODE (rhs1) == SSA_NAME)
1114 limit++;
1116 res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi,
1117 evolution_of_loop, limit);
1118 if (res == t_true)
1119 *evolution_of_loop = add_to_evolution
1120 (loop->num, chrec_convert (type, *evolution_of_loop, at_stmt),
1121 MINUS_EXPR, rhs1, at_stmt);
1123 else if (res == t_dont_know)
1124 *evolution_of_loop = chrec_dont_know;
1126 else
1127 /* Otherwise, match an assignment under the form:
1128 "a = ... - ...". */
1129 /* And there is nothing to do. */
1130 res = t_false;
1131 break;
1133 default:
1134 res = t_false;
1137 return res;
1140 /* Follow the ssa edge into the expression EXPR.
1141 Return true if the strongly connected component has been found. */
1143 static t_bool
1144 follow_ssa_edge_expr (struct loop *loop, gimple at_stmt, tree expr,
1145 gimple halting_phi, tree *evolution_of_loop, int limit)
1147 enum tree_code code = TREE_CODE (expr);
1148 tree type = TREE_TYPE (expr), rhs0, rhs1;
1149 t_bool res;
1151 /* The EXPR is one of the following cases:
1152 - an SSA_NAME,
1153 - an INTEGER_CST,
1154 - a PLUS_EXPR,
1155 - a POINTER_PLUS_EXPR,
1156 - a MINUS_EXPR,
1157 - an ASSERT_EXPR,
1158 - other cases are not yet handled. */
1160 switch (code)
1162 CASE_CONVERT:
1163 /* This assignment is under the form "a_1 = (cast) rhs. */
1164 res = follow_ssa_edge_expr (loop, at_stmt, TREE_OPERAND (expr, 0),
1165 halting_phi, evolution_of_loop, limit);
1166 *evolution_of_loop = chrec_convert (type, *evolution_of_loop, at_stmt);
1167 break;
1169 case INTEGER_CST:
1170 /* This assignment is under the form "a_1 = 7". */
1171 res = t_false;
1172 break;
1174 case SSA_NAME:
1175 /* This assignment is under the form: "a_1 = b_2". */
1176 res = follow_ssa_edge
1177 (loop, SSA_NAME_DEF_STMT (expr), halting_phi, evolution_of_loop, limit);
1178 break;
1180 case POINTER_PLUS_EXPR:
1181 case PLUS_EXPR:
1182 case MINUS_EXPR:
1183 /* This case is under the form "rhs0 +- rhs1". */
1184 rhs0 = TREE_OPERAND (expr, 0);
1185 rhs1 = TREE_OPERAND (expr, 1);
1186 type = TREE_TYPE (rhs0);
1187 STRIP_USELESS_TYPE_CONVERSION (rhs0);
1188 STRIP_USELESS_TYPE_CONVERSION (rhs1);
1189 res = follow_ssa_edge_binary (loop, at_stmt, type, rhs0, code, rhs1,
1190 halting_phi, evolution_of_loop, limit);
1191 break;
1193 case ASSERT_EXPR:
1194 /* This assignment is of the form: "a_1 = ASSERT_EXPR <a_2, ...>"
1195 It must be handled as a copy assignment of the form a_1 = a_2. */
1196 rhs0 = ASSERT_EXPR_VAR (expr);
1197 if (TREE_CODE (rhs0) == SSA_NAME)
1198 res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (rhs0),
1199 halting_phi, evolution_of_loop, limit);
1200 else
1201 res = t_false;
1202 break;
1204 default:
1205 res = t_false;
1206 break;
1209 return res;
1212 /* Follow the ssa edge into the right hand side of an assignment STMT.
1213 Return true if the strongly connected component has been found. */
1215 static t_bool
1216 follow_ssa_edge_in_rhs (struct loop *loop, gimple stmt,
1217 gimple halting_phi, tree *evolution_of_loop, int limit)
1219 enum tree_code code = gimple_assign_rhs_code (stmt);
1220 tree type = gimple_expr_type (stmt), rhs1, rhs2;
1221 t_bool res;
1223 switch (code)
1225 CASE_CONVERT:
1226 /* This assignment is under the form "a_1 = (cast) rhs. */
1227 res = follow_ssa_edge_expr (loop, stmt, gimple_assign_rhs1 (stmt),
1228 halting_phi, evolution_of_loop, limit);
1229 *evolution_of_loop = chrec_convert (type, *evolution_of_loop, stmt);
1230 break;
1232 case POINTER_PLUS_EXPR:
1233 case PLUS_EXPR:
1234 case MINUS_EXPR:
1235 rhs1 = gimple_assign_rhs1 (stmt);
1236 rhs2 = gimple_assign_rhs2 (stmt);
1237 type = TREE_TYPE (rhs1);
1238 res = follow_ssa_edge_binary (loop, stmt, type, rhs1, code, rhs2,
1239 halting_phi, evolution_of_loop, limit);
1240 break;
1242 default:
1243 if (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS)
1244 res = follow_ssa_edge_expr (loop, stmt, gimple_assign_rhs1 (stmt),
1245 halting_phi, evolution_of_loop, limit);
1246 else
1247 res = t_false;
1248 break;
1251 return res;
1254 /* Checks whether the I-th argument of a PHI comes from a backedge. */
1256 static bool
1257 backedge_phi_arg_p (gimple phi, int i)
1259 const_edge e = gimple_phi_arg_edge (phi, i);
1261 /* We would in fact like to test EDGE_DFS_BACK here, but we do not care
1262 about updating it anywhere, and this should work as well most of the
1263 time. */
1264 if (e->flags & EDGE_IRREDUCIBLE_LOOP)
1265 return true;
1267 return false;
1270 /* Helper function for one branch of the condition-phi-node. Return
1271 true if the strongly connected component has been found following
1272 this path. */
1274 static inline t_bool
1275 follow_ssa_edge_in_condition_phi_branch (int i,
1276 struct loop *loop,
1277 gimple condition_phi,
1278 gimple halting_phi,
1279 tree *evolution_of_branch,
1280 tree init_cond, int limit)
1282 tree branch = PHI_ARG_DEF (condition_phi, i);
1283 *evolution_of_branch = chrec_dont_know;
1285 /* Do not follow back edges (they must belong to an irreducible loop, which
1286 we really do not want to worry about). */
1287 if (backedge_phi_arg_p (condition_phi, i))
1288 return t_false;
1290 if (TREE_CODE (branch) == SSA_NAME)
1292 *evolution_of_branch = init_cond;
1293 return follow_ssa_edge (loop, SSA_NAME_DEF_STMT (branch), halting_phi,
1294 evolution_of_branch, limit);
1297 /* This case occurs when one of the condition branches sets
1298 the variable to a constant: i.e. a phi-node like
1299 "a_2 = PHI <a_7(5), 2(6)>;".
1301 FIXME: This case have to be refined correctly:
1302 in some cases it is possible to say something better than
1303 chrec_dont_know, for example using a wrap-around notation. */
1304 return t_false;
1307 /* This function merges the branches of a condition-phi-node in a
1308 loop. */
1310 static t_bool
1311 follow_ssa_edge_in_condition_phi (struct loop *loop,
1312 gimple condition_phi,
1313 gimple halting_phi,
1314 tree *evolution_of_loop, int limit)
1316 int i, n;
1317 tree init = *evolution_of_loop;
1318 tree evolution_of_branch;
1319 t_bool res = follow_ssa_edge_in_condition_phi_branch (0, loop, condition_phi,
1320 halting_phi,
1321 &evolution_of_branch,
1322 init, limit);
1323 if (res == t_false || res == t_dont_know)
1324 return res;
1326 *evolution_of_loop = evolution_of_branch;
1328 n = gimple_phi_num_args (condition_phi);
1329 for (i = 1; i < n; i++)
1331 /* Quickly give up when the evolution of one of the branches is
1332 not known. */
1333 if (*evolution_of_loop == chrec_dont_know)
1334 return t_true;
1336 /* Increase the limit by the PHI argument number to avoid exponential
1337 time and memory complexity. */
1338 res = follow_ssa_edge_in_condition_phi_branch (i, loop, condition_phi,
1339 halting_phi,
1340 &evolution_of_branch,
1341 init, limit + i);
1342 if (res == t_false || res == t_dont_know)
1343 return res;
1345 *evolution_of_loop = chrec_merge (*evolution_of_loop,
1346 evolution_of_branch);
1349 return t_true;
1352 /* Follow an SSA edge in an inner loop. It computes the overall
1353 effect of the loop, and following the symbolic initial conditions,
1354 it follows the edges in the parent loop. The inner loop is
1355 considered as a single statement. */
1357 static t_bool
1358 follow_ssa_edge_inner_loop_phi (struct loop *outer_loop,
1359 gimple loop_phi_node,
1360 gimple halting_phi,
1361 tree *evolution_of_loop, int limit)
1363 struct loop *loop = loop_containing_stmt (loop_phi_node);
1364 tree ev = analyze_scalar_evolution (loop, PHI_RESULT (loop_phi_node));
1366 /* Sometimes, the inner loop is too difficult to analyze, and the
1367 result of the analysis is a symbolic parameter. */
1368 if (ev == PHI_RESULT (loop_phi_node))
1370 t_bool res = t_false;
1371 int i, n = gimple_phi_num_args (loop_phi_node);
1373 for (i = 0; i < n; i++)
1375 tree arg = PHI_ARG_DEF (loop_phi_node, i);
1376 basic_block bb;
1378 /* Follow the edges that exit the inner loop. */
1379 bb = gimple_phi_arg_edge (loop_phi_node, i)->src;
1380 if (!flow_bb_inside_loop_p (loop, bb))
1381 res = follow_ssa_edge_expr (outer_loop, loop_phi_node,
1382 arg, halting_phi,
1383 evolution_of_loop, limit);
1384 if (res == t_true)
1385 break;
1388 /* If the path crosses this loop-phi, give up. */
1389 if (res == t_true)
1390 *evolution_of_loop = chrec_dont_know;
1392 return res;
1395 /* Otherwise, compute the overall effect of the inner loop. */
1396 ev = compute_overall_effect_of_inner_loop (loop, ev);
1397 return follow_ssa_edge_expr (outer_loop, loop_phi_node, ev, halting_phi,
1398 evolution_of_loop, limit);
1401 /* Follow an SSA edge from a loop-phi-node to itself, constructing a
1402 path that is analyzed on the return walk. */
1404 static t_bool
1405 follow_ssa_edge (struct loop *loop, gimple def, gimple halting_phi,
1406 tree *evolution_of_loop, int limit)
1408 struct loop *def_loop;
1410 if (gimple_nop_p (def))
1411 return t_false;
1413 /* Give up if the path is longer than the MAX that we allow. */
1414 if (limit > PARAM_VALUE (PARAM_SCEV_MAX_EXPR_SIZE))
1415 return t_dont_know;
1417 def_loop = loop_containing_stmt (def);
1419 switch (gimple_code (def))
1421 case GIMPLE_PHI:
1422 if (!loop_phi_node_p (def))
1423 /* DEF is a condition-phi-node. Follow the branches, and
1424 record their evolutions. Finally, merge the collected
1425 information and set the approximation to the main
1426 variable. */
1427 return follow_ssa_edge_in_condition_phi
1428 (loop, def, halting_phi, evolution_of_loop, limit);
1430 /* When the analyzed phi is the halting_phi, the
1431 depth-first search is over: we have found a path from
1432 the halting_phi to itself in the loop. */
1433 if (def == halting_phi)
1434 return t_true;
1436 /* Otherwise, the evolution of the HALTING_PHI depends
1437 on the evolution of another loop-phi-node, i.e. the
1438 evolution function is a higher degree polynomial. */
1439 if (def_loop == loop)
1440 return t_false;
1442 /* Inner loop. */
1443 if (flow_loop_nested_p (loop, def_loop))
1444 return follow_ssa_edge_inner_loop_phi
1445 (loop, def, halting_phi, evolution_of_loop, limit + 1);
1447 /* Outer loop. */
1448 return t_false;
1450 case GIMPLE_ASSIGN:
1451 return follow_ssa_edge_in_rhs (loop, def, halting_phi,
1452 evolution_of_loop, limit);
1454 default:
1455 /* At this level of abstraction, the program is just a set
1456 of GIMPLE_ASSIGNs and PHI_NODEs. In principle there is no
1457 other node to be handled. */
1458 return t_false;
1464 /* Given a LOOP_PHI_NODE, this function determines the evolution
1465 function from LOOP_PHI_NODE to LOOP_PHI_NODE in the loop. */
1467 static tree
1468 analyze_evolution_in_loop (gimple loop_phi_node,
1469 tree init_cond)
1471 int i, n = gimple_phi_num_args (loop_phi_node);
1472 tree evolution_function = chrec_not_analyzed_yet;
1473 struct loop *loop = loop_containing_stmt (loop_phi_node);
1474 basic_block bb;
1476 if (dump_file && (dump_flags & TDF_DETAILS))
1478 fprintf (dump_file, "(analyze_evolution_in_loop \n");
1479 fprintf (dump_file, " (loop_phi_node = ");
1480 print_gimple_stmt (dump_file, loop_phi_node, 0, 0);
1481 fprintf (dump_file, ")\n");
1484 for (i = 0; i < n; i++)
1486 tree arg = PHI_ARG_DEF (loop_phi_node, i);
1487 gimple ssa_chain;
1488 tree ev_fn;
1489 t_bool res;
1491 /* Select the edges that enter the loop body. */
1492 bb = gimple_phi_arg_edge (loop_phi_node, i)->src;
1493 if (!flow_bb_inside_loop_p (loop, bb))
1494 continue;
1496 if (TREE_CODE (arg) == SSA_NAME)
1498 bool val = false;
1500 ssa_chain = SSA_NAME_DEF_STMT (arg);
1502 /* Pass in the initial condition to the follow edge function. */
1503 ev_fn = init_cond;
1504 res = follow_ssa_edge (loop, ssa_chain, loop_phi_node, &ev_fn, 0);
1506 /* If ev_fn has no evolution in the inner loop, and the
1507 init_cond is not equal to ev_fn, then we have an
1508 ambiguity between two possible values, as we cannot know
1509 the number of iterations at this point. */
1510 if (TREE_CODE (ev_fn) != POLYNOMIAL_CHREC
1511 && no_evolution_in_loop_p (ev_fn, loop->num, &val) && val
1512 && !operand_equal_p (init_cond, ev_fn, 0))
1513 ev_fn = chrec_dont_know;
1515 else
1516 res = t_false;
1518 /* When it is impossible to go back on the same
1519 loop_phi_node by following the ssa edges, the
1520 evolution is represented by a peeled chrec, i.e. the
1521 first iteration, EV_FN has the value INIT_COND, then
1522 all the other iterations it has the value of ARG.
1523 For the moment, PEELED_CHREC nodes are not built. */
1524 if (res != t_true)
1525 ev_fn = chrec_dont_know;
1527 /* When there are multiple back edges of the loop (which in fact never
1528 happens currently, but nevertheless), merge their evolutions. */
1529 evolution_function = chrec_merge (evolution_function, ev_fn);
1532 if (dump_file && (dump_flags & TDF_DETAILS))
1534 fprintf (dump_file, " (evolution_function = ");
1535 print_generic_expr (dump_file, evolution_function, 0);
1536 fprintf (dump_file, "))\n");
1539 return evolution_function;
1542 /* Given a loop-phi-node, return the initial conditions of the
1543 variable on entry of the loop. When the CCP has propagated
1544 constants into the loop-phi-node, the initial condition is
1545 instantiated, otherwise the initial condition is kept symbolic.
1546 This analyzer does not analyze the evolution outside the current
1547 loop, and leaves this task to the on-demand tree reconstructor. */
1549 static tree
1550 analyze_initial_condition (gimple loop_phi_node)
1552 int i, n;
1553 tree init_cond = chrec_not_analyzed_yet;
1554 struct loop *loop = loop_containing_stmt (loop_phi_node);
1556 if (dump_file && (dump_flags & TDF_DETAILS))
1558 fprintf (dump_file, "(analyze_initial_condition \n");
1559 fprintf (dump_file, " (loop_phi_node = \n");
1560 print_gimple_stmt (dump_file, loop_phi_node, 0, 0);
1561 fprintf (dump_file, ")\n");
1564 n = gimple_phi_num_args (loop_phi_node);
1565 for (i = 0; i < n; i++)
1567 tree branch = PHI_ARG_DEF (loop_phi_node, i);
1568 basic_block bb = gimple_phi_arg_edge (loop_phi_node, i)->src;
1570 /* When the branch is oriented to the loop's body, it does
1571 not contribute to the initial condition. */
1572 if (flow_bb_inside_loop_p (loop, bb))
1573 continue;
1575 if (init_cond == chrec_not_analyzed_yet)
1577 init_cond = branch;
1578 continue;
1581 if (TREE_CODE (branch) == SSA_NAME)
1583 init_cond = chrec_dont_know;
1584 break;
1587 init_cond = chrec_merge (init_cond, branch);
1590 /* Ooops -- a loop without an entry??? */
1591 if (init_cond == chrec_not_analyzed_yet)
1592 init_cond = chrec_dont_know;
1594 /* During early loop unrolling we do not have fully constant propagated IL.
1595 Handle degenerate PHIs here to not miss important unrollings. */
1596 if (TREE_CODE (init_cond) == SSA_NAME)
1598 gimple def = SSA_NAME_DEF_STMT (init_cond);
1599 tree res;
1600 if (gimple_code (def) == GIMPLE_PHI
1601 && (res = degenerate_phi_result (def)) != NULL_TREE
1602 /* Only allow invariants here, otherwise we may break
1603 loop-closed SSA form. */
1604 && is_gimple_min_invariant (res))
1605 init_cond = res;
1608 if (dump_file && (dump_flags & TDF_DETAILS))
1610 fprintf (dump_file, " (init_cond = ");
1611 print_generic_expr (dump_file, init_cond, 0);
1612 fprintf (dump_file, "))\n");
1615 return init_cond;
1618 /* Analyze the scalar evolution for LOOP_PHI_NODE. */
1620 static tree
1621 interpret_loop_phi (struct loop *loop, gimple loop_phi_node)
1623 tree res;
1624 struct loop *phi_loop = loop_containing_stmt (loop_phi_node);
1625 tree init_cond;
1627 if (phi_loop != loop)
1629 struct loop *subloop;
1630 tree evolution_fn = analyze_scalar_evolution
1631 (phi_loop, PHI_RESULT (loop_phi_node));
1633 /* Dive one level deeper. */
1634 subloop = superloop_at_depth (phi_loop, loop_depth (loop) + 1);
1636 /* Interpret the subloop. */
1637 res = compute_overall_effect_of_inner_loop (subloop, evolution_fn);
1638 return res;
1641 /* Otherwise really interpret the loop phi. */
1642 init_cond = analyze_initial_condition (loop_phi_node);
1643 res = analyze_evolution_in_loop (loop_phi_node, init_cond);
1645 return res;
1648 /* This function merges the branches of a condition-phi-node,
1649 contained in the outermost loop, and whose arguments are already
1650 analyzed. */
1652 static tree
1653 interpret_condition_phi (struct loop *loop, gimple condition_phi)
1655 int i, n = gimple_phi_num_args (condition_phi);
1656 tree res = chrec_not_analyzed_yet;
1658 for (i = 0; i < n; i++)
1660 tree branch_chrec;
1662 if (backedge_phi_arg_p (condition_phi, i))
1664 res = chrec_dont_know;
1665 break;
1668 branch_chrec = analyze_scalar_evolution
1669 (loop, PHI_ARG_DEF (condition_phi, i));
1671 res = chrec_merge (res, branch_chrec);
1674 return res;
1677 /* Interpret the operation RHS1 OP RHS2. If we didn't
1678 analyze this node before, follow the definitions until ending
1679 either on an analyzed GIMPLE_ASSIGN, or on a loop-phi-node. On the
1680 return path, this function propagates evolutions (ala constant copy
1681 propagation). OPND1 is not a GIMPLE expression because we could
1682 analyze the effect of an inner loop: see interpret_loop_phi. */
1684 static tree
1685 interpret_rhs_expr (struct loop *loop, gimple at_stmt,
1686 tree type, tree rhs1, enum tree_code code, tree rhs2)
1688 tree res, chrec1, chrec2;
1690 if (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS)
1692 if (is_gimple_min_invariant (rhs1))
1693 return chrec_convert (type, rhs1, at_stmt);
1695 if (code == SSA_NAME)
1696 return chrec_convert (type, analyze_scalar_evolution (loop, rhs1),
1697 at_stmt);
1699 if (code == ASSERT_EXPR)
1701 rhs1 = ASSERT_EXPR_VAR (rhs1);
1702 return chrec_convert (type, analyze_scalar_evolution (loop, rhs1),
1703 at_stmt);
1706 return chrec_dont_know;
1709 switch (code)
1711 case POINTER_PLUS_EXPR:
1712 chrec1 = analyze_scalar_evolution (loop, rhs1);
1713 chrec2 = analyze_scalar_evolution (loop, rhs2);
1714 chrec1 = chrec_convert (type, chrec1, at_stmt);
1715 chrec2 = chrec_convert (sizetype, chrec2, at_stmt);
1716 res = chrec_fold_plus (type, chrec1, chrec2);
1717 break;
1719 case PLUS_EXPR:
1720 chrec1 = analyze_scalar_evolution (loop, rhs1);
1721 chrec2 = analyze_scalar_evolution (loop, rhs2);
1722 chrec1 = chrec_convert (type, chrec1, at_stmt);
1723 chrec2 = chrec_convert (type, chrec2, at_stmt);
1724 res = chrec_fold_plus (type, chrec1, chrec2);
1725 break;
1727 case MINUS_EXPR:
1728 chrec1 = analyze_scalar_evolution (loop, rhs1);
1729 chrec2 = analyze_scalar_evolution (loop, rhs2);
1730 chrec1 = chrec_convert (type, chrec1, at_stmt);
1731 chrec2 = chrec_convert (type, chrec2, at_stmt);
1732 res = chrec_fold_minus (type, chrec1, chrec2);
1733 break;
1735 case NEGATE_EXPR:
1736 chrec1 = analyze_scalar_evolution (loop, rhs1);
1737 chrec1 = chrec_convert (type, chrec1, at_stmt);
1738 /* TYPE may be integer, real or complex, so use fold_convert. */
1739 res = chrec_fold_multiply (type, chrec1,
1740 fold_convert (type, integer_minus_one_node));
1741 break;
1743 case BIT_NOT_EXPR:
1744 /* Handle ~X as -1 - X. */
1745 chrec1 = analyze_scalar_evolution (loop, rhs1);
1746 chrec1 = chrec_convert (type, chrec1, at_stmt);
1747 res = chrec_fold_minus (type,
1748 fold_convert (type, integer_minus_one_node),
1749 chrec1);
1750 break;
1752 case MULT_EXPR:
1753 chrec1 = analyze_scalar_evolution (loop, rhs1);
1754 chrec2 = analyze_scalar_evolution (loop, rhs2);
1755 chrec1 = chrec_convert (type, chrec1, at_stmt);
1756 chrec2 = chrec_convert (type, chrec2, at_stmt);
1757 res = chrec_fold_multiply (type, chrec1, chrec2);
1758 break;
1760 CASE_CONVERT:
1761 chrec1 = analyze_scalar_evolution (loop, rhs1);
1762 res = chrec_convert (type, chrec1, at_stmt);
1763 break;
1765 default:
1766 res = chrec_dont_know;
1767 break;
1770 return res;
1773 /* Interpret the expression EXPR. */
1775 static tree
1776 interpret_expr (struct loop *loop, gimple at_stmt, tree expr)
1778 enum tree_code code;
1779 tree type = TREE_TYPE (expr), op0, op1;
1781 if (automatically_generated_chrec_p (expr))
1782 return expr;
1784 if (TREE_CODE (expr) == POLYNOMIAL_CHREC)
1785 return chrec_dont_know;
1787 extract_ops_from_tree (expr, &code, &op0, &op1);
1789 return interpret_rhs_expr (loop, at_stmt, type,
1790 op0, code, op1);
1793 /* Interpret the rhs of the assignment STMT. */
1795 static tree
1796 interpret_gimple_assign (struct loop *loop, gimple stmt)
1798 tree type = TREE_TYPE (gimple_assign_lhs (stmt));
1799 enum tree_code code = gimple_assign_rhs_code (stmt);
1801 return interpret_rhs_expr (loop, stmt, type,
1802 gimple_assign_rhs1 (stmt), code,
1803 gimple_assign_rhs2 (stmt));
1808 /* This section contains all the entry points:
1809 - number_of_iterations_in_loop,
1810 - analyze_scalar_evolution,
1811 - instantiate_parameters.
1814 /* Compute and return the evolution function in WRTO_LOOP, the nearest
1815 common ancestor of DEF_LOOP and USE_LOOP. */
1817 static tree
1818 compute_scalar_evolution_in_loop (struct loop *wrto_loop,
1819 struct loop *def_loop,
1820 tree ev)
1822 tree res;
1823 if (def_loop == wrto_loop)
1824 return ev;
1826 def_loop = superloop_at_depth (def_loop, loop_depth (wrto_loop) + 1);
1827 res = compute_overall_effect_of_inner_loop (def_loop, ev);
1829 return analyze_scalar_evolution_1 (wrto_loop, res, chrec_not_analyzed_yet);
1832 /* Helper recursive function. */
1834 static tree
1835 analyze_scalar_evolution_1 (struct loop *loop, tree var, tree res)
1837 tree type = TREE_TYPE (var);
1838 gimple def;
1839 basic_block bb;
1840 struct loop *def_loop;
1842 if (loop == NULL || TREE_CODE (type) == VECTOR_TYPE)
1843 return chrec_dont_know;
1845 if (TREE_CODE (var) != SSA_NAME)
1846 return interpret_expr (loop, NULL, var);
1848 def = SSA_NAME_DEF_STMT (var);
1849 bb = gimple_bb (def);
1850 def_loop = bb ? bb->loop_father : NULL;
1852 if (bb == NULL
1853 || !flow_bb_inside_loop_p (loop, bb))
1855 /* Keep the symbolic form. */
1856 res = var;
1857 goto set_and_end;
1860 if (res != chrec_not_analyzed_yet)
1862 if (loop != bb->loop_father)
1863 res = compute_scalar_evolution_in_loop
1864 (find_common_loop (loop, bb->loop_father), bb->loop_father, res);
1866 goto set_and_end;
1869 if (loop != def_loop)
1871 res = analyze_scalar_evolution_1 (def_loop, var, chrec_not_analyzed_yet);
1872 res = compute_scalar_evolution_in_loop (loop, def_loop, res);
1874 goto set_and_end;
1877 switch (gimple_code (def))
1879 case GIMPLE_ASSIGN:
1880 res = interpret_gimple_assign (loop, def);
1881 break;
1883 case GIMPLE_PHI:
1884 if (loop_phi_node_p (def))
1885 res = interpret_loop_phi (loop, def);
1886 else
1887 res = interpret_condition_phi (loop, def);
1888 break;
1890 default:
1891 res = chrec_dont_know;
1892 break;
1895 set_and_end:
1897 /* Keep the symbolic form. */
1898 if (res == chrec_dont_know)
1899 res = var;
1901 if (loop == def_loop)
1902 set_scalar_evolution (block_before_loop (loop), var, res);
1904 return res;
1907 /* Analyzes and returns the scalar evolution of the ssa_name VAR in
1908 LOOP. LOOP is the loop in which the variable is used.
1910 Example of use: having a pointer VAR to a SSA_NAME node, STMT a
1911 pointer to the statement that uses this variable, in order to
1912 determine the evolution function of the variable, use the following
1913 calls:
1915 loop_p loop = loop_containing_stmt (stmt);
1916 tree chrec_with_symbols = analyze_scalar_evolution (loop, var);
1917 tree chrec_instantiated = instantiate_parameters (loop, chrec_with_symbols);
1920 tree
1921 analyze_scalar_evolution (struct loop *loop, tree var)
1923 tree res;
1925 if (dump_file && (dump_flags & TDF_DETAILS))
1927 fprintf (dump_file, "(analyze_scalar_evolution \n");
1928 fprintf (dump_file, " (loop_nb = %d)\n", loop->num);
1929 fprintf (dump_file, " (scalar = ");
1930 print_generic_expr (dump_file, var, 0);
1931 fprintf (dump_file, ")\n");
1934 res = get_scalar_evolution (block_before_loop (loop), var);
1935 res = analyze_scalar_evolution_1 (loop, var, res);
1937 if (dump_file && (dump_flags & TDF_DETAILS))
1938 fprintf (dump_file, ")\n");
1940 return res;
1943 /* Analyze scalar evolution of use of VERSION in USE_LOOP with respect to
1944 WRTO_LOOP (which should be a superloop of USE_LOOP)
1946 FOLDED_CASTS is set to true if resolve_mixers used
1947 chrec_convert_aggressive (TODO -- not really, we are way too conservative
1948 at the moment in order to keep things simple).
1950 To illustrate the meaning of USE_LOOP and WRTO_LOOP, consider the following
1951 example:
1953 for (i = 0; i < 100; i++) -- loop 1
1955 for (j = 0; j < 100; j++) -- loop 2
1957 k1 = i;
1958 k2 = j;
1960 use2 (k1, k2);
1962 for (t = 0; t < 100; t++) -- loop 3
1963 use3 (k1, k2);
1966 use1 (k1, k2);
1969 Both k1 and k2 are invariants in loop3, thus
1970 analyze_scalar_evolution_in_loop (loop3, loop3, k1) = k1
1971 analyze_scalar_evolution_in_loop (loop3, loop3, k2) = k2
1973 As they are invariant, it does not matter whether we consider their
1974 usage in loop 3 or loop 2, hence
1975 analyze_scalar_evolution_in_loop (loop2, loop3, k1) =
1976 analyze_scalar_evolution_in_loop (loop2, loop2, k1) = i
1977 analyze_scalar_evolution_in_loop (loop2, loop3, k2) =
1978 analyze_scalar_evolution_in_loop (loop2, loop2, k2) = [0,+,1]_2
1980 Similarly for their evolutions with respect to loop 1. The values of K2
1981 in the use in loop 2 vary independently on loop 1, thus we cannot express
1982 the evolution with respect to loop 1:
1983 analyze_scalar_evolution_in_loop (loop1, loop3, k1) =
1984 analyze_scalar_evolution_in_loop (loop1, loop2, k1) = [0,+,1]_1
1985 analyze_scalar_evolution_in_loop (loop1, loop3, k2) =
1986 analyze_scalar_evolution_in_loop (loop1, loop2, k2) = dont_know
1988 The value of k2 in the use in loop 1 is known, though:
1989 analyze_scalar_evolution_in_loop (loop1, loop1, k1) = [0,+,1]_1
1990 analyze_scalar_evolution_in_loop (loop1, loop1, k2) = 100
1993 static tree
1994 analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop,
1995 tree version, bool *folded_casts)
1997 bool val = false;
1998 tree ev = version, tmp;
2000 /* We cannot just do
2002 tmp = analyze_scalar_evolution (use_loop, version);
2003 ev = resolve_mixers (wrto_loop, tmp);
2005 as resolve_mixers would query the scalar evolution with respect to
2006 wrto_loop. For example, in the situation described in the function
2007 comment, suppose that wrto_loop = loop1, use_loop = loop3 and
2008 version = k2. Then
2010 analyze_scalar_evolution (use_loop, version) = k2
2012 and resolve_mixers (loop1, k2) finds that the value of k2 in loop 1
2013 is 100, which is a wrong result, since we are interested in the
2014 value in loop 3.
2016 Instead, we need to proceed from use_loop to wrto_loop loop by loop,
2017 each time checking that there is no evolution in the inner loop. */
2019 if (folded_casts)
2020 *folded_casts = false;
2021 while (1)
2023 tmp = analyze_scalar_evolution (use_loop, ev);
2024 ev = resolve_mixers (use_loop, tmp);
2026 if (folded_casts && tmp != ev)
2027 *folded_casts = true;
2029 if (use_loop == wrto_loop)
2030 return ev;
2032 /* If the value of the use changes in the inner loop, we cannot express
2033 its value in the outer loop (we might try to return interval chrec,
2034 but we do not have a user for it anyway) */
2035 if (!no_evolution_in_loop_p (ev, use_loop->num, &val)
2036 || !val)
2037 return chrec_dont_know;
2039 use_loop = loop_outer (use_loop);
2043 /* Returns from CACHE the value for VERSION instantiated below
2044 INSTANTIATED_BELOW block. */
2046 static tree
2047 get_instantiated_value (htab_t cache, basic_block instantiated_below,
2048 tree version)
2050 struct scev_info_str *info, pattern;
2052 pattern.var = version;
2053 pattern.instantiated_below = instantiated_below;
2054 info = (struct scev_info_str *) htab_find (cache, &pattern);
2056 if (info)
2057 return info->chrec;
2058 else
2059 return NULL_TREE;
2062 /* Sets in CACHE the value of VERSION instantiated below basic block
2063 INSTANTIATED_BELOW to VAL. */
2065 static void
2066 set_instantiated_value (htab_t cache, basic_block instantiated_below,
2067 tree version, tree val)
2069 struct scev_info_str *info, pattern;
2070 PTR *slot;
2072 pattern.var = version;
2073 pattern.instantiated_below = instantiated_below;
2074 slot = htab_find_slot (cache, &pattern, INSERT);
2076 if (!*slot)
2077 *slot = new_scev_info_str (instantiated_below, version);
2078 info = (struct scev_info_str *) *slot;
2079 info->chrec = val;
2082 /* Return the closed_loop_phi node for VAR. If there is none, return
2083 NULL_TREE. */
2085 static tree
2086 loop_closed_phi_def (tree var)
2088 struct loop *loop;
2089 edge exit;
2090 gimple phi;
2091 gimple_stmt_iterator psi;
2093 if (var == NULL_TREE
2094 || TREE_CODE (var) != SSA_NAME)
2095 return NULL_TREE;
2097 loop = loop_containing_stmt (SSA_NAME_DEF_STMT (var));
2098 exit = single_exit (loop);
2099 if (!exit)
2100 return NULL_TREE;
2102 for (psi = gsi_start_phis (exit->dest); !gsi_end_p (psi); gsi_next (&psi))
2104 phi = gsi_stmt (psi);
2105 if (PHI_ARG_DEF_FROM_EDGE (phi, exit) == var)
2106 return PHI_RESULT (phi);
2109 return NULL_TREE;
2112 static tree instantiate_scev_r (basic_block, struct loop *, tree, bool,
2113 htab_t, int);
2115 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2116 and EVOLUTION_LOOP, that were left under a symbolic form.
2118 CHREC is an SSA_NAME to be instantiated.
2120 CACHE is the cache of already instantiated values.
2122 FOLD_CONVERSIONS should be set to true when the conversions that
2123 may wrap in signed/pointer type are folded, as long as the value of
2124 the chrec is preserved.
2126 SIZE_EXPR is used for computing the size of the expression to be
2127 instantiated, and to stop if it exceeds some limit. */
2129 static tree
2130 instantiate_scev_name (basic_block instantiate_below,
2131 struct loop *evolution_loop, tree chrec,
2132 bool fold_conversions, htab_t cache, int size_expr)
2134 tree res;
2135 struct loop *def_loop;
2136 basic_block def_bb = gimple_bb (SSA_NAME_DEF_STMT (chrec));
2138 /* A parameter (or loop invariant and we do not want to include
2139 evolutions in outer loops), nothing to do. */
2140 if (!def_bb
2141 || loop_depth (def_bb->loop_father) == 0
2142 || dominated_by_p (CDI_DOMINATORS, instantiate_below, def_bb))
2143 return chrec;
2145 /* We cache the value of instantiated variable to avoid exponential
2146 time complexity due to reevaluations. We also store the convenient
2147 value in the cache in order to prevent infinite recursion -- we do
2148 not want to instantiate the SSA_NAME if it is in a mixer
2149 structure. This is used for avoiding the instantiation of
2150 recursively defined functions, such as:
2152 | a_2 -> {0, +, 1, +, a_2}_1 */
2154 res = get_instantiated_value (cache, instantiate_below, chrec);
2155 if (res)
2156 return res;
2158 res = chrec_dont_know;
2159 set_instantiated_value (cache, instantiate_below, chrec, res);
2161 def_loop = find_common_loop (evolution_loop, def_bb->loop_father);
2163 /* If the analysis yields a parametric chrec, instantiate the
2164 result again. */
2165 res = analyze_scalar_evolution (def_loop, chrec);
2167 /* Don't instantiate loop-closed-ssa phi nodes. */
2168 if (TREE_CODE (res) == SSA_NAME
2169 && (loop_containing_stmt (SSA_NAME_DEF_STMT (res)) == NULL
2170 || (loop_depth (loop_containing_stmt (SSA_NAME_DEF_STMT (res)))
2171 > loop_depth (def_loop))))
2173 if (res == chrec)
2174 res = loop_closed_phi_def (chrec);
2175 else
2176 res = chrec;
2178 if (res == NULL_TREE
2179 || !dominated_by_p (CDI_DOMINATORS, instantiate_below,
2180 gimple_bb (SSA_NAME_DEF_STMT (res))))
2181 res = chrec_dont_know;
2184 else if (res != chrec_dont_know)
2185 res = instantiate_scev_r (instantiate_below, evolution_loop, res,
2186 fold_conversions, cache, size_expr);
2188 /* Store the correct value to the cache. */
2189 set_instantiated_value (cache, instantiate_below, chrec, res);
2190 return res;
2194 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2195 and EVOLUTION_LOOP, that were left under a symbolic form.
2197 CHREC is a polynomial chain of recurrence to be instantiated.
2199 CACHE is the cache of already instantiated values.
2201 FOLD_CONVERSIONS should be set to true when the conversions that
2202 may wrap in signed/pointer type are folded, as long as the value of
2203 the chrec is preserved.
2205 SIZE_EXPR is used for computing the size of the expression to be
2206 instantiated, and to stop if it exceeds some limit. */
2208 static tree
2209 instantiate_scev_poly (basic_block instantiate_below,
2210 struct loop *evolution_loop, tree chrec,
2211 bool fold_conversions, htab_t cache, int size_expr)
2213 tree op1;
2214 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2215 CHREC_LEFT (chrec), fold_conversions, cache,
2216 size_expr);
2217 if (op0 == chrec_dont_know)
2218 return chrec_dont_know;
2220 op1 = instantiate_scev_r (instantiate_below, evolution_loop,
2221 CHREC_RIGHT (chrec), fold_conversions, cache,
2222 size_expr);
2223 if (op1 == chrec_dont_know)
2224 return chrec_dont_know;
2226 if (CHREC_LEFT (chrec) != op0
2227 || CHREC_RIGHT (chrec) != op1)
2229 unsigned var = CHREC_VARIABLE (chrec);
2231 /* When the instantiated stride or base has an evolution in an
2232 innermost loop, return chrec_dont_know, as this is not a
2233 valid SCEV representation. In the reduced testcase for
2234 PR40281 we would have {0, +, {1, +, 1}_2}_1 that has no
2235 meaning. */
2236 if ((tree_is_chrec (op0) && CHREC_VARIABLE (op0) > var)
2237 || (tree_is_chrec (op1) && CHREC_VARIABLE (op1) > var))
2238 return chrec_dont_know;
2240 op1 = chrec_convert_rhs (chrec_type (op0), op1, NULL);
2241 chrec = build_polynomial_chrec (var, op0, op1);
2244 return chrec;
2247 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2248 and EVOLUTION_LOOP, that were left under a symbolic form.
2250 "C0 CODE C1" is a binary expression of type TYPE to be instantiated.
2252 CACHE is the cache of already instantiated values.
2254 FOLD_CONVERSIONS should be set to true when the conversions that
2255 may wrap in signed/pointer type are folded, as long as the value of
2256 the chrec is preserved.
2258 SIZE_EXPR is used for computing the size of the expression to be
2259 instantiated, and to stop if it exceeds some limit. */
2261 static tree
2262 instantiate_scev_binary (basic_block instantiate_below,
2263 struct loop *evolution_loop, tree chrec, enum tree_code code,
2264 tree type, tree c0, tree c1,
2265 bool fold_conversions, htab_t cache, int size_expr)
2267 tree op1;
2268 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2269 c0, fold_conversions, cache,
2270 size_expr);
2271 if (op0 == chrec_dont_know)
2272 return chrec_dont_know;
2274 op1 = instantiate_scev_r (instantiate_below, evolution_loop,
2275 c1, fold_conversions, cache,
2276 size_expr);
2277 if (op1 == chrec_dont_know)
2278 return chrec_dont_know;
2280 if (c0 != op0
2281 || c1 != op1)
2283 op0 = chrec_convert (type, op0, NULL);
2284 op1 = chrec_convert_rhs (type, op1, NULL);
2286 switch (code)
2288 case POINTER_PLUS_EXPR:
2289 case PLUS_EXPR:
2290 return chrec_fold_plus (type, op0, op1);
2292 case MINUS_EXPR:
2293 return chrec_fold_minus (type, op0, op1);
2295 case MULT_EXPR:
2296 return chrec_fold_multiply (type, op0, op1);
2298 default:
2299 gcc_unreachable ();
2303 return chrec ? chrec : fold_build2 (code, type, c0, c1);
2306 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2307 and EVOLUTION_LOOP, that were left under a symbolic form.
2309 "CHREC" that stands for a convert expression "(TYPE) OP" is to be
2310 instantiated.
2312 CACHE is the cache of already instantiated values.
2314 FOLD_CONVERSIONS should be set to true when the conversions that
2315 may wrap in signed/pointer type are folded, as long as the value of
2316 the chrec is preserved.
2318 SIZE_EXPR is used for computing the size of the expression to be
2319 instantiated, and to stop if it exceeds some limit. */
2321 static tree
2322 instantiate_scev_convert (basic_block instantiate_below,
2323 struct loop *evolution_loop, tree chrec,
2324 tree type, tree op,
2325 bool fold_conversions, htab_t cache, int size_expr)
2327 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, op,
2328 fold_conversions, cache, size_expr);
2330 if (op0 == chrec_dont_know)
2331 return chrec_dont_know;
2333 if (fold_conversions)
2335 tree tmp = chrec_convert_aggressive (type, op0);
2336 if (tmp)
2337 return tmp;
2340 if (chrec && op0 == op)
2341 return chrec;
2343 /* If we used chrec_convert_aggressive, we can no longer assume that
2344 signed chrecs do not overflow, as chrec_convert does, so avoid
2345 calling it in that case. */
2346 if (fold_conversions)
2347 return fold_convert (type, op0);
2349 return chrec_convert (type, op0, NULL);
2352 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2353 and EVOLUTION_LOOP, that were left under a symbolic form.
2355 CHREC is a BIT_NOT_EXPR or a NEGATE_EXPR expression to be instantiated.
2356 Handle ~X as -1 - X.
2357 Handle -X as -1 * X.
2359 CACHE is the cache of already instantiated values.
2361 FOLD_CONVERSIONS should be set to true when the conversions that
2362 may wrap in signed/pointer type are folded, as long as the value of
2363 the chrec is preserved.
2365 SIZE_EXPR is used for computing the size of the expression to be
2366 instantiated, and to stop if it exceeds some limit. */
2368 static tree
2369 instantiate_scev_not (basic_block instantiate_below,
2370 struct loop *evolution_loop, tree chrec,
2371 enum tree_code code, tree type, tree op,
2372 bool fold_conversions, htab_t cache, int size_expr)
2374 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, op,
2375 fold_conversions, cache, size_expr);
2377 if (op0 == chrec_dont_know)
2378 return chrec_dont_know;
2380 if (op != op0)
2382 op0 = chrec_convert (type, op0, NULL);
2384 switch (code)
2386 case BIT_NOT_EXPR:
2387 return chrec_fold_minus
2388 (type, fold_convert (type, integer_minus_one_node), op0);
2390 case NEGATE_EXPR:
2391 return chrec_fold_multiply
2392 (type, fold_convert (type, integer_minus_one_node), op0);
2394 default:
2395 gcc_unreachable ();
2399 return chrec ? chrec : fold_build1 (code, type, op0);
2402 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2403 and EVOLUTION_LOOP, that were left under a symbolic form.
2405 CHREC is an expression with 3 operands to be instantiated.
2407 CACHE is the cache of already instantiated values.
2409 FOLD_CONVERSIONS should be set to true when the conversions that
2410 may wrap in signed/pointer type are folded, as long as the value of
2411 the chrec is preserved.
2413 SIZE_EXPR is used for computing the size of the expression to be
2414 instantiated, and to stop if it exceeds some limit. */
2416 static tree
2417 instantiate_scev_3 (basic_block instantiate_below,
2418 struct loop *evolution_loop, tree chrec,
2419 bool fold_conversions, htab_t cache, int size_expr)
2421 tree op1, op2;
2422 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2423 TREE_OPERAND (chrec, 0),
2424 fold_conversions, cache, size_expr);
2425 if (op0 == chrec_dont_know)
2426 return chrec_dont_know;
2428 op1 = instantiate_scev_r (instantiate_below, evolution_loop,
2429 TREE_OPERAND (chrec, 1),
2430 fold_conversions, cache, size_expr);
2431 if (op1 == chrec_dont_know)
2432 return chrec_dont_know;
2434 op2 = instantiate_scev_r (instantiate_below, evolution_loop,
2435 TREE_OPERAND (chrec, 2),
2436 fold_conversions, cache, size_expr);
2437 if (op2 == chrec_dont_know)
2438 return chrec_dont_know;
2440 if (op0 == TREE_OPERAND (chrec, 0)
2441 && op1 == TREE_OPERAND (chrec, 1)
2442 && op2 == TREE_OPERAND (chrec, 2))
2443 return chrec;
2445 return fold_build3 (TREE_CODE (chrec),
2446 TREE_TYPE (chrec), op0, op1, op2);
2449 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2450 and EVOLUTION_LOOP, that were left under a symbolic form.
2452 CHREC is an expression with 2 operands to be instantiated.
2454 CACHE is the cache of already instantiated values.
2456 FOLD_CONVERSIONS should be set to true when the conversions that
2457 may wrap in signed/pointer type are folded, as long as the value of
2458 the chrec is preserved.
2460 SIZE_EXPR is used for computing the size of the expression to be
2461 instantiated, and to stop if it exceeds some limit. */
2463 static tree
2464 instantiate_scev_2 (basic_block instantiate_below,
2465 struct loop *evolution_loop, tree chrec,
2466 bool fold_conversions, htab_t cache, int size_expr)
2468 tree op1;
2469 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2470 TREE_OPERAND (chrec, 0),
2471 fold_conversions, cache, size_expr);
2472 if (op0 == chrec_dont_know)
2473 return chrec_dont_know;
2475 op1 = instantiate_scev_r (instantiate_below, evolution_loop,
2476 TREE_OPERAND (chrec, 1),
2477 fold_conversions, cache, size_expr);
2478 if (op1 == chrec_dont_know)
2479 return chrec_dont_know;
2481 if (op0 == TREE_OPERAND (chrec, 0)
2482 && op1 == TREE_OPERAND (chrec, 1))
2483 return chrec;
2485 return fold_build2 (TREE_CODE (chrec), TREE_TYPE (chrec), op0, op1);
2488 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2489 and EVOLUTION_LOOP, that were left under a symbolic form.
2491 CHREC is an expression with 2 operands to be instantiated.
2493 CACHE is the cache of already instantiated values.
2495 FOLD_CONVERSIONS should be set to true when the conversions that
2496 may wrap in signed/pointer type are folded, as long as the value of
2497 the chrec is preserved.
2499 SIZE_EXPR is used for computing the size of the expression to be
2500 instantiated, and to stop if it exceeds some limit. */
2502 static tree
2503 instantiate_scev_1 (basic_block instantiate_below,
2504 struct loop *evolution_loop, tree chrec,
2505 bool fold_conversions, htab_t cache, int size_expr)
2507 tree op0 = instantiate_scev_r (instantiate_below, evolution_loop,
2508 TREE_OPERAND (chrec, 0),
2509 fold_conversions, cache, size_expr);
2511 if (op0 == chrec_dont_know)
2512 return chrec_dont_know;
2514 if (op0 == TREE_OPERAND (chrec, 0))
2515 return chrec;
2517 return fold_build1 (TREE_CODE (chrec), TREE_TYPE (chrec), op0);
2520 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2521 and EVOLUTION_LOOP, that were left under a symbolic form.
2523 CHREC is the scalar evolution to instantiate.
2525 CACHE is the cache of already instantiated values.
2527 FOLD_CONVERSIONS should be set to true when the conversions that
2528 may wrap in signed/pointer type are folded, as long as the value of
2529 the chrec is preserved.
2531 SIZE_EXPR is used for computing the size of the expression to be
2532 instantiated, and to stop if it exceeds some limit. */
2534 static tree
2535 instantiate_scev_r (basic_block instantiate_below,
2536 struct loop *evolution_loop, tree chrec,
2537 bool fold_conversions, htab_t cache, int size_expr)
2539 /* Give up if the expression is larger than the MAX that we allow. */
2540 if (size_expr++ > PARAM_VALUE (PARAM_SCEV_MAX_EXPR_SIZE))
2541 return chrec_dont_know;
2543 if (automatically_generated_chrec_p (chrec)
2544 || is_gimple_min_invariant (chrec))
2545 return chrec;
2547 switch (TREE_CODE (chrec))
2549 case SSA_NAME:
2550 return instantiate_scev_name (instantiate_below, evolution_loop, chrec,
2551 fold_conversions, cache, size_expr);
2553 case POLYNOMIAL_CHREC:
2554 return instantiate_scev_poly (instantiate_below, evolution_loop, chrec,
2555 fold_conversions, cache, size_expr);
2557 case POINTER_PLUS_EXPR:
2558 case PLUS_EXPR:
2559 case MINUS_EXPR:
2560 case MULT_EXPR:
2561 return instantiate_scev_binary (instantiate_below, evolution_loop, chrec,
2562 TREE_CODE (chrec), chrec_type (chrec),
2563 TREE_OPERAND (chrec, 0),
2564 TREE_OPERAND (chrec, 1),
2565 fold_conversions, cache, size_expr);
2567 CASE_CONVERT:
2568 return instantiate_scev_convert (instantiate_below, evolution_loop, chrec,
2569 TREE_TYPE (chrec), TREE_OPERAND (chrec, 0),
2570 fold_conversions, cache, size_expr);
2572 case NEGATE_EXPR:
2573 case BIT_NOT_EXPR:
2574 return instantiate_scev_not (instantiate_below, evolution_loop, chrec,
2575 TREE_CODE (chrec), TREE_TYPE (chrec),
2576 TREE_OPERAND (chrec, 0),
2577 fold_conversions, cache, size_expr);
2579 case SCEV_NOT_KNOWN:
2580 return chrec_dont_know;
2582 case SCEV_KNOWN:
2583 return chrec_known;
2585 default:
2586 break;
2589 if (VL_EXP_CLASS_P (chrec))
2590 return chrec_dont_know;
2592 switch (TREE_CODE_LENGTH (TREE_CODE (chrec)))
2594 case 3:
2595 return instantiate_scev_3 (instantiate_below, evolution_loop, chrec,
2596 fold_conversions, cache, size_expr);
2598 case 2:
2599 return instantiate_scev_2 (instantiate_below, evolution_loop, chrec,
2600 fold_conversions, cache, size_expr);
2602 case 1:
2603 return instantiate_scev_1 (instantiate_below, evolution_loop, chrec,
2604 fold_conversions, cache, size_expr);
2606 case 0:
2607 return chrec;
2609 default:
2610 break;
2613 /* Too complicated to handle. */
2614 return chrec_dont_know;
2617 /* Analyze all the parameters of the chrec that were left under a
2618 symbolic form. INSTANTIATE_BELOW is the basic block that stops the
2619 recursive instantiation of parameters: a parameter is a variable
2620 that is defined in a basic block that dominates INSTANTIATE_BELOW or
2621 a function parameter. */
2623 tree
2624 instantiate_scev (basic_block instantiate_below, struct loop *evolution_loop,
2625 tree chrec)
2627 tree res;
2628 htab_t cache = htab_create (10, hash_scev_info, eq_scev_info, del_scev_info);
2630 if (dump_file && (dump_flags & TDF_DETAILS))
2632 fprintf (dump_file, "(instantiate_scev \n");
2633 fprintf (dump_file, " (instantiate_below = %d)\n", instantiate_below->index);
2634 fprintf (dump_file, " (evolution_loop = %d)\n", evolution_loop->num);
2635 fprintf (dump_file, " (chrec = ");
2636 print_generic_expr (dump_file, chrec, 0);
2637 fprintf (dump_file, ")\n");
2640 res = instantiate_scev_r (instantiate_below, evolution_loop, chrec, false,
2641 cache, 0);
2643 if (dump_file && (dump_flags & TDF_DETAILS))
2645 fprintf (dump_file, " (res = ");
2646 print_generic_expr (dump_file, res, 0);
2647 fprintf (dump_file, "))\n");
2650 htab_delete (cache);
2652 return res;
2655 /* Similar to instantiate_parameters, but does not introduce the
2656 evolutions in outer loops for LOOP invariants in CHREC, and does not
2657 care about causing overflows, as long as they do not affect value
2658 of an expression. */
2660 tree
2661 resolve_mixers (struct loop *loop, tree chrec)
2663 htab_t cache = htab_create (10, hash_scev_info, eq_scev_info, del_scev_info);
2664 tree ret = instantiate_scev_r (block_before_loop (loop), loop, chrec, true,
2665 cache, 0);
2666 htab_delete (cache);
2667 return ret;
2670 /* Entry point for the analysis of the number of iterations pass.
2671 This function tries to safely approximate the number of iterations
2672 the loop will run. When this property is not decidable at compile
2673 time, the result is chrec_dont_know. Otherwise the result is
2674 a scalar or a symbolic parameter.
2676 Example of analysis: suppose that the loop has an exit condition:
2678 "if (b > 49) goto end_loop;"
2680 and that in a previous analysis we have determined that the
2681 variable 'b' has an evolution function:
2683 "EF = {23, +, 5}_2".
2685 When we evaluate the function at the point 5, i.e. the value of the
2686 variable 'b' after 5 iterations in the loop, we have EF (5) = 48,
2687 and EF (6) = 53. In this case the value of 'b' on exit is '53' and
2688 the loop body has been executed 6 times. */
2690 tree
2691 number_of_latch_executions (struct loop *loop)
2693 tree res, type;
2694 edge exit;
2695 struct tree_niter_desc niter_desc;
2697 /* Determine whether the number_of_iterations_in_loop has already
2698 been computed. */
2699 res = loop->nb_iterations;
2700 if (res)
2701 return res;
2702 res = chrec_dont_know;
2704 if (dump_file && (dump_flags & TDF_DETAILS))
2705 fprintf (dump_file, "(number_of_iterations_in_loop\n");
2707 exit = single_exit (loop);
2708 if (!exit)
2709 goto end;
2711 if (!number_of_iterations_exit (loop, exit, &niter_desc, false))
2712 goto end;
2714 type = TREE_TYPE (niter_desc.niter);
2715 if (integer_nonzerop (niter_desc.may_be_zero))
2716 res = build_int_cst (type, 0);
2717 else if (integer_zerop (niter_desc.may_be_zero))
2718 res = niter_desc.niter;
2719 else
2720 res = chrec_dont_know;
2722 end:
2723 return set_nb_iterations_in_loop (loop, res);
2726 /* Returns the number of executions of the exit condition of LOOP,
2727 i.e., the number by one higher than number_of_latch_executions.
2728 Note that unlike number_of_latch_executions, this number does
2729 not necessarily fit in the unsigned variant of the type of
2730 the control variable -- if the number of iterations is a constant,
2731 we return chrec_dont_know if adding one to number_of_latch_executions
2732 overflows; however, in case the number of iterations is symbolic
2733 expression, the caller is responsible for dealing with this
2734 the possible overflow. */
2736 tree
2737 number_of_exit_cond_executions (struct loop *loop)
2739 tree ret = number_of_latch_executions (loop);
2740 tree type = chrec_type (ret);
2742 if (chrec_contains_undetermined (ret))
2743 return ret;
2745 ret = chrec_fold_plus (type, ret, build_int_cst (type, 1));
2746 if (TREE_CODE (ret) == INTEGER_CST
2747 && TREE_OVERFLOW (ret))
2748 return chrec_dont_know;
2750 return ret;
2753 /* One of the drivers for testing the scalar evolutions analysis.
2754 This function computes the number of iterations for all the loops
2755 from the EXIT_CONDITIONS array. */
2757 static void
2758 number_of_iterations_for_all_loops (VEC(gimple,heap) **exit_conditions)
2760 unsigned int i;
2761 unsigned nb_chrec_dont_know_loops = 0;
2762 unsigned nb_static_loops = 0;
2763 gimple cond;
2765 for (i = 0; VEC_iterate (gimple, *exit_conditions, i, cond); i++)
2767 tree res = number_of_latch_executions (loop_containing_stmt (cond));
2768 if (chrec_contains_undetermined (res))
2769 nb_chrec_dont_know_loops++;
2770 else
2771 nb_static_loops++;
2774 if (dump_file)
2776 fprintf (dump_file, "\n(\n");
2777 fprintf (dump_file, "-----------------------------------------\n");
2778 fprintf (dump_file, "%d\tnb_chrec_dont_know_loops\n", nb_chrec_dont_know_loops);
2779 fprintf (dump_file, "%d\tnb_static_loops\n", nb_static_loops);
2780 fprintf (dump_file, "%d\tnb_total_loops\n", number_of_loops ());
2781 fprintf (dump_file, "-----------------------------------------\n");
2782 fprintf (dump_file, ")\n\n");
2784 print_loops (dump_file, 3);
2790 /* Counters for the stats. */
2792 struct chrec_stats
2794 unsigned nb_chrecs;
2795 unsigned nb_affine;
2796 unsigned nb_affine_multivar;
2797 unsigned nb_higher_poly;
2798 unsigned nb_chrec_dont_know;
2799 unsigned nb_undetermined;
2802 /* Reset the counters. */
2804 static inline void
2805 reset_chrecs_counters (struct chrec_stats *stats)
2807 stats->nb_chrecs = 0;
2808 stats->nb_affine = 0;
2809 stats->nb_affine_multivar = 0;
2810 stats->nb_higher_poly = 0;
2811 stats->nb_chrec_dont_know = 0;
2812 stats->nb_undetermined = 0;
2815 /* Dump the contents of a CHREC_STATS structure. */
2817 static void
2818 dump_chrecs_stats (FILE *file, struct chrec_stats *stats)
2820 fprintf (file, "\n(\n");
2821 fprintf (file, "-----------------------------------------\n");
2822 fprintf (file, "%d\taffine univariate chrecs\n", stats->nb_affine);
2823 fprintf (file, "%d\taffine multivariate chrecs\n", stats->nb_affine_multivar);
2824 fprintf (file, "%d\tdegree greater than 2 polynomials\n",
2825 stats->nb_higher_poly);
2826 fprintf (file, "%d\tchrec_dont_know chrecs\n", stats->nb_chrec_dont_know);
2827 fprintf (file, "-----------------------------------------\n");
2828 fprintf (file, "%d\ttotal chrecs\n", stats->nb_chrecs);
2829 fprintf (file, "%d\twith undetermined coefficients\n",
2830 stats->nb_undetermined);
2831 fprintf (file, "-----------------------------------------\n");
2832 fprintf (file, "%d\tchrecs in the scev database\n",
2833 (int) htab_elements (scalar_evolution_info));
2834 fprintf (file, "%d\tsets in the scev database\n", nb_set_scev);
2835 fprintf (file, "%d\tgets in the scev database\n", nb_get_scev);
2836 fprintf (file, "-----------------------------------------\n");
2837 fprintf (file, ")\n\n");
2840 /* Gather statistics about CHREC. */
2842 static void
2843 gather_chrec_stats (tree chrec, struct chrec_stats *stats)
2845 if (dump_file && (dump_flags & TDF_STATS))
2847 fprintf (dump_file, "(classify_chrec ");
2848 print_generic_expr (dump_file, chrec, 0);
2849 fprintf (dump_file, "\n");
2852 stats->nb_chrecs++;
2854 if (chrec == NULL_TREE)
2856 stats->nb_undetermined++;
2857 return;
2860 switch (TREE_CODE (chrec))
2862 case POLYNOMIAL_CHREC:
2863 if (evolution_function_is_affine_p (chrec))
2865 if (dump_file && (dump_flags & TDF_STATS))
2866 fprintf (dump_file, " affine_univariate\n");
2867 stats->nb_affine++;
2869 else if (evolution_function_is_affine_multivariate_p (chrec, 0))
2871 if (dump_file && (dump_flags & TDF_STATS))
2872 fprintf (dump_file, " affine_multivariate\n");
2873 stats->nb_affine_multivar++;
2875 else
2877 if (dump_file && (dump_flags & TDF_STATS))
2878 fprintf (dump_file, " higher_degree_polynomial\n");
2879 stats->nb_higher_poly++;
2882 break;
2884 default:
2885 break;
2888 if (chrec_contains_undetermined (chrec))
2890 if (dump_file && (dump_flags & TDF_STATS))
2891 fprintf (dump_file, " undetermined\n");
2892 stats->nb_undetermined++;
2895 if (dump_file && (dump_flags & TDF_STATS))
2896 fprintf (dump_file, ")\n");
2899 /* One of the drivers for testing the scalar evolutions analysis.
2900 This function analyzes the scalar evolution of all the scalars
2901 defined as loop phi nodes in one of the loops from the
2902 EXIT_CONDITIONS array.
2904 TODO Optimization: A loop is in canonical form if it contains only
2905 a single scalar loop phi node. All the other scalars that have an
2906 evolution in the loop are rewritten in function of this single
2907 index. This allows the parallelization of the loop. */
2909 static void
2910 analyze_scalar_evolution_for_all_loop_phi_nodes (VEC(gimple,heap) **exit_conditions)
2912 unsigned int i;
2913 struct chrec_stats stats;
2914 gimple cond, phi;
2915 gimple_stmt_iterator psi;
2917 reset_chrecs_counters (&stats);
2919 for (i = 0; VEC_iterate (gimple, *exit_conditions, i, cond); i++)
2921 struct loop *loop;
2922 basic_block bb;
2923 tree chrec;
2925 loop = loop_containing_stmt (cond);
2926 bb = loop->header;
2928 for (psi = gsi_start_phis (bb); !gsi_end_p (psi); gsi_next (&psi))
2930 phi = gsi_stmt (psi);
2931 if (is_gimple_reg (PHI_RESULT (phi)))
2933 chrec = instantiate_parameters
2934 (loop,
2935 analyze_scalar_evolution (loop, PHI_RESULT (phi)));
2937 if (dump_file && (dump_flags & TDF_STATS))
2938 gather_chrec_stats (chrec, &stats);
2943 if (dump_file && (dump_flags & TDF_STATS))
2944 dump_chrecs_stats (dump_file, &stats);
2947 /* Callback for htab_traverse, gathers information on chrecs in the
2948 hashtable. */
2950 static int
2951 gather_stats_on_scev_database_1 (void **slot, void *stats)
2953 struct scev_info_str *entry = (struct scev_info_str *) *slot;
2955 gather_chrec_stats (entry->chrec, (struct chrec_stats *) stats);
2957 return 1;
2960 /* Classify the chrecs of the whole database. */
2962 void
2963 gather_stats_on_scev_database (void)
2965 struct chrec_stats stats;
2967 if (!dump_file)
2968 return;
2970 reset_chrecs_counters (&stats);
2972 htab_traverse (scalar_evolution_info, gather_stats_on_scev_database_1,
2973 &stats);
2975 dump_chrecs_stats (dump_file, &stats);
2980 /* Initializer. */
2982 static void
2983 initialize_scalar_evolutions_analyzer (void)
2985 /* The elements below are unique. */
2986 if (chrec_dont_know == NULL_TREE)
2988 chrec_not_analyzed_yet = NULL_TREE;
2989 chrec_dont_know = make_node (SCEV_NOT_KNOWN);
2990 chrec_known = make_node (SCEV_KNOWN);
2991 TREE_TYPE (chrec_dont_know) = void_type_node;
2992 TREE_TYPE (chrec_known) = void_type_node;
2996 /* Initialize the analysis of scalar evolutions for LOOPS. */
2998 void
2999 scev_initialize (void)
3001 loop_iterator li;
3002 struct loop *loop;
3004 scalar_evolution_info = htab_create_alloc (100,
3005 hash_scev_info,
3006 eq_scev_info,
3007 del_scev_info,
3008 ggc_calloc,
3009 ggc_free);
3011 initialize_scalar_evolutions_analyzer ();
3013 FOR_EACH_LOOP (li, loop, 0)
3015 loop->nb_iterations = NULL_TREE;
3019 /* Cleans up the information cached by the scalar evolutions analysis
3020 in the hash table. */
3022 void
3023 scev_reset_htab (void)
3025 if (!scalar_evolution_info)
3026 return;
3028 htab_empty (scalar_evolution_info);
3031 /* Cleans up the information cached by the scalar evolutions analysis
3032 in the hash table and in the loop->nb_iterations. */
3034 void
3035 scev_reset (void)
3037 loop_iterator li;
3038 struct loop *loop;
3040 scev_reset_htab ();
3042 if (!current_loops)
3043 return;
3045 FOR_EACH_LOOP (li, loop, 0)
3047 loop->nb_iterations = NULL_TREE;
3051 /* Checks whether use of OP in USE_LOOP behaves as a simple affine iv with
3052 respect to WRTO_LOOP and returns its base and step in IV if possible
3053 (see analyze_scalar_evolution_in_loop for more details on USE_LOOP
3054 and WRTO_LOOP). If ALLOW_NONCONSTANT_STEP is true, we want step to be
3055 invariant in LOOP. Otherwise we require it to be an integer constant.
3057 IV->no_overflow is set to true if we are sure the iv cannot overflow (e.g.
3058 because it is computed in signed arithmetics). Consequently, adding an
3059 induction variable
3061 for (i = IV->base; ; i += IV->step)
3063 is only safe if IV->no_overflow is false, or TYPE_OVERFLOW_UNDEFINED is
3064 false for the type of the induction variable, or you can prove that i does
3065 not wrap by some other argument. Otherwise, this might introduce undefined
3066 behavior, and
3068 for (i = iv->base; ; i = (type) ((unsigned type) i + (unsigned type) iv->step))
3070 must be used instead. */
3072 bool
3073 simple_iv (struct loop *wrto_loop, struct loop *use_loop, tree op,
3074 affine_iv *iv, bool allow_nonconstant_step)
3076 tree type, ev;
3077 bool folded_casts;
3079 iv->base = NULL_TREE;
3080 iv->step = NULL_TREE;
3081 iv->no_overflow = false;
3083 type = TREE_TYPE (op);
3084 if (TREE_CODE (type) != INTEGER_TYPE
3085 && TREE_CODE (type) != POINTER_TYPE)
3086 return false;
3088 ev = analyze_scalar_evolution_in_loop (wrto_loop, use_loop, op,
3089 &folded_casts);
3090 if (chrec_contains_undetermined (ev)
3091 || chrec_contains_symbols_defined_in_loop (ev, wrto_loop->num))
3092 return false;
3094 if (tree_does_not_contain_chrecs (ev))
3096 iv->base = ev;
3097 iv->step = build_int_cst (TREE_TYPE (ev), 0);
3098 iv->no_overflow = true;
3099 return true;
3102 if (TREE_CODE (ev) != POLYNOMIAL_CHREC
3103 || CHREC_VARIABLE (ev) != (unsigned) wrto_loop->num)
3104 return false;
3106 iv->step = CHREC_RIGHT (ev);
3107 if ((!allow_nonconstant_step && TREE_CODE (iv->step) != INTEGER_CST)
3108 || tree_contains_chrecs (iv->step, NULL))
3109 return false;
3111 iv->base = CHREC_LEFT (ev);
3112 if (tree_contains_chrecs (iv->base, NULL))
3113 return false;
3115 iv->no_overflow = !folded_casts && TYPE_OVERFLOW_UNDEFINED (type);
3117 return true;
3120 /* Runs the analysis of scalar evolutions. */
3122 void
3123 scev_analysis (void)
3125 VEC(gimple,heap) *exit_conditions;
3127 exit_conditions = VEC_alloc (gimple, heap, 37);
3128 select_loops_exit_conditions (&exit_conditions);
3130 if (dump_file && (dump_flags & TDF_STATS))
3131 analyze_scalar_evolution_for_all_loop_phi_nodes (&exit_conditions);
3133 number_of_iterations_for_all_loops (&exit_conditions);
3134 VEC_free (gimple, heap, exit_conditions);
3137 /* Finalize the scalar evolution analysis. */
3139 void
3140 scev_finalize (void)
3142 if (!scalar_evolution_info)
3143 return;
3144 htab_delete (scalar_evolution_info);
3145 scalar_evolution_info = NULL;
3148 /* Returns true if the expression EXPR is considered to be too expensive
3149 for scev_const_prop. */
3151 bool
3152 expression_expensive_p (tree expr)
3154 enum tree_code code;
3156 if (is_gimple_val (expr))
3157 return false;
3159 code = TREE_CODE (expr);
3160 if (code == TRUNC_DIV_EXPR
3161 || code == CEIL_DIV_EXPR
3162 || code == FLOOR_DIV_EXPR
3163 || code == ROUND_DIV_EXPR
3164 || code == TRUNC_MOD_EXPR
3165 || code == CEIL_MOD_EXPR
3166 || code == FLOOR_MOD_EXPR
3167 || code == ROUND_MOD_EXPR
3168 || code == EXACT_DIV_EXPR)
3170 /* Division by power of two is usually cheap, so we allow it.
3171 Forbid anything else. */
3172 if (!integer_pow2p (TREE_OPERAND (expr, 1)))
3173 return true;
3176 switch (TREE_CODE_CLASS (code))
3178 case tcc_binary:
3179 case tcc_comparison:
3180 if (expression_expensive_p (TREE_OPERAND (expr, 1)))
3181 return true;
3183 /* Fallthru. */
3184 case tcc_unary:
3185 return expression_expensive_p (TREE_OPERAND (expr, 0));
3187 default:
3188 return true;
3192 /* Replace ssa names for that scev can prove they are constant by the
3193 appropriate constants. Also perform final value replacement in loops,
3194 in case the replacement expressions are cheap.
3196 We only consider SSA names defined by phi nodes; rest is left to the
3197 ordinary constant propagation pass. */
3199 unsigned int
3200 scev_const_prop (void)
3202 basic_block bb;
3203 tree name, type, ev;
3204 gimple phi, ass;
3205 struct loop *loop, *ex_loop;
3206 bitmap ssa_names_to_remove = NULL;
3207 unsigned i;
3208 loop_iterator li;
3209 gimple_stmt_iterator psi;
3211 if (number_of_loops () <= 1)
3212 return 0;
3214 FOR_EACH_BB (bb)
3216 loop = bb->loop_father;
3218 for (psi = gsi_start_phis (bb); !gsi_end_p (psi); gsi_next (&psi))
3220 phi = gsi_stmt (psi);
3221 name = PHI_RESULT (phi);
3223 if (!is_gimple_reg (name))
3224 continue;
3226 type = TREE_TYPE (name);
3228 if (!POINTER_TYPE_P (type)
3229 && !INTEGRAL_TYPE_P (type))
3230 continue;
3232 ev = resolve_mixers (loop, analyze_scalar_evolution (loop, name));
3233 if (!is_gimple_min_invariant (ev)
3234 || !may_propagate_copy (name, ev))
3235 continue;
3237 /* Replace the uses of the name. */
3238 if (name != ev)
3239 replace_uses_by (name, ev);
3241 if (!ssa_names_to_remove)
3242 ssa_names_to_remove = BITMAP_ALLOC (NULL);
3243 bitmap_set_bit (ssa_names_to_remove, SSA_NAME_VERSION (name));
3247 /* Remove the ssa names that were replaced by constants. We do not
3248 remove them directly in the previous cycle, since this
3249 invalidates scev cache. */
3250 if (ssa_names_to_remove)
3252 bitmap_iterator bi;
3254 EXECUTE_IF_SET_IN_BITMAP (ssa_names_to_remove, 0, i, bi)
3256 gimple_stmt_iterator psi;
3257 name = ssa_name (i);
3258 phi = SSA_NAME_DEF_STMT (name);
3260 gcc_assert (gimple_code (phi) == GIMPLE_PHI);
3261 psi = gsi_for_stmt (phi);
3262 remove_phi_node (&psi, true);
3265 BITMAP_FREE (ssa_names_to_remove);
3266 scev_reset ();
3269 /* Now the regular final value replacement. */
3270 FOR_EACH_LOOP (li, loop, LI_FROM_INNERMOST)
3272 edge exit;
3273 tree def, rslt, niter;
3274 gimple_stmt_iterator bsi;
3276 /* If we do not know exact number of iterations of the loop, we cannot
3277 replace the final value. */
3278 exit = single_exit (loop);
3279 if (!exit)
3280 continue;
3282 niter = number_of_latch_executions (loop);
3283 if (niter == chrec_dont_know)
3284 continue;
3286 /* Ensure that it is possible to insert new statements somewhere. */
3287 if (!single_pred_p (exit->dest))
3288 split_loop_exit_edge (exit);
3289 bsi = gsi_after_labels (exit->dest);
3291 ex_loop = superloop_at_depth (loop,
3292 loop_depth (exit->dest->loop_father) + 1);
3294 for (psi = gsi_start_phis (exit->dest); !gsi_end_p (psi); )
3296 phi = gsi_stmt (psi);
3297 rslt = PHI_RESULT (phi);
3298 def = PHI_ARG_DEF_FROM_EDGE (phi, exit);
3299 if (!is_gimple_reg (def))
3301 gsi_next (&psi);
3302 continue;
3305 if (!POINTER_TYPE_P (TREE_TYPE (def))
3306 && !INTEGRAL_TYPE_P (TREE_TYPE (def)))
3308 gsi_next (&psi);
3309 continue;
3312 def = analyze_scalar_evolution_in_loop (ex_loop, loop, def, NULL);
3313 def = compute_overall_effect_of_inner_loop (ex_loop, def);
3314 if (!tree_does_not_contain_chrecs (def)
3315 || chrec_contains_symbols_defined_in_loop (def, ex_loop->num)
3316 /* Moving the computation from the loop may prolong life range
3317 of some ssa names, which may cause problems if they appear
3318 on abnormal edges. */
3319 || contains_abnormal_ssa_name_p (def)
3320 /* Do not emit expensive expressions. The rationale is that
3321 when someone writes a code like
3323 while (n > 45) n -= 45;
3325 he probably knows that n is not large, and does not want it
3326 to be turned into n %= 45. */
3327 || expression_expensive_p (def))
3329 gsi_next (&psi);
3330 continue;
3333 /* Eliminate the PHI node and replace it by a computation outside
3334 the loop. */
3335 def = unshare_expr (def);
3336 remove_phi_node (&psi, false);
3338 def = force_gimple_operand_gsi (&bsi, def, false, NULL_TREE,
3339 true, GSI_SAME_STMT);
3340 ass = gimple_build_assign (rslt, def);
3341 gsi_insert_before (&bsi, ass, GSI_SAME_STMT);
3344 return 0;
3347 #include "gt-tree-scalar-evolution.h"