1 /* Scalar evolution detector.
2 Copyright (C) 2003, 2004 Free Software Foundation, Inc.
3 Contributed by Sebastian Pop <s.pop@laposte.net>
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify it under
8 the terms of the GNU General Public License as published by the Free
9 Software Foundation; either version 2, or (at your option) any later
12 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
13 WARRANTY; without even the implied warranty of MERCHANTABILITY or
14 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING. If not, write to the Free
19 Software Foundation, 59 Temple Place - Suite 330, Boston, MA
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
51 - When the definition is a MODIFY_EXPR: if the right hand side
52 (RHS) of the definition cannot be statically analyzed, the answer
53 of the analyzer is: "don't know".
54 Otherwise, for all the variables that are not yet analyzed in the
55 RHS, try to determine their evolution, and finally try to
56 evaluate the operation of the RHS that gives the evolution
57 function of the analyzed variable.
59 - When the definition is a condition-phi-node: determine the
60 evolution function for all the branches of the phi node, and
61 finally merge these evolutions (see chrec_merge).
63 - When the definition is a loop-phi-node: determine its initial
64 condition, that is the SSA edge defined in an outer loop, and
65 keep it symbolic. Then determine the SSA edges that are defined
66 in the body of the loop. Follow the inner edges until ending on
67 another loop-phi-node of the same analyzed loop. If the reached
68 loop-phi-node is not the starting loop-phi-node, then we keep
69 this definition under a symbolic form. If the reached
70 loop-phi-node is the same as the starting one, then we compute a
71 symbolic stride on the return path. The result is then the
72 symbolic chrec {initial_condition, +, symbolic_stride}_loop.
76 Example 1: Illustration of the basic algorithm.
82 | if (c > 10) exit_loop
85 Suppose that we want to know the number of iterations of the
86 loop_1. The exit_loop is controlled by a COND_EXPR (c > 10). We
87 ask the scalar evolution analyzer two questions: what's the
88 scalar evolution (scev) of "c", and what's the scev of "10". For
89 "10" the answer is "10" since it is a scalar constant. For the
90 scalar variable "c", it follows the SSA edge to its definition,
91 "c = b + 1", and then asks again what's the scev of "b".
92 Following the SSA edge, we end on a loop-phi-node "b = phi (a,
93 c)", where the initial condition is "a", and the inner loop edge
94 is "c". The initial condition is kept under a symbolic form (it
95 may be the case that the copy constant propagation has done its
96 work and we end with the constant "3" as one of the edges of the
97 loop-phi-node). The update edge is followed to the end of the
98 loop, and until reaching again the starting loop-phi-node: b -> c
99 -> b. At this point we have drawn a path from "b" to "b" from
100 which we compute the stride in the loop: in this example it is
101 "+1". The resulting scev for "b" is "b -> {a, +, 1}_1". Now
102 that the scev for "b" is known, it is possible to compute the
103 scev for "c", that is "c -> {a + 1, +, 1}_1". In order to
104 determine the number of iterations in the loop_1, we have to
105 instantiate_parameters ({a + 1, +, 1}_1), that gives after some
106 more analysis the scev {4, +, 1}_1, or in other words, this is
107 the function "f (x) = x + 4", where x is the iteration count of
108 the loop_1. Now we have to solve the inequality "x + 4 > 10",
109 and take the smallest iteration number for which the loop is
110 exited: x = 7. This loop runs from x = 0 to x = 7, and in total
111 there are 8 iterations. In terms of loop normalization, we have
112 created a variable that is implicitly defined, "x" or just "_1",
113 and all the other analyzed scalars of the loop are defined in
114 function of this variable:
120 or in terms of a C program:
123 | for (x = 0; x <= 7; x++)
129 Example 2: Illustration of the algorithm on nested loops.
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:
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:
158 Example 3: Higher degree polynomials.
172 instantiate_parameters ({5, +, a}_1) -> {5, +, 2, +, 1}_1
173 instantiate_parameters ({5 + a, +, a}_1) -> {7, +, 3, +, 1}_1
175 Example 4: Lucas, Fibonacci, or mixers in general.
187 The syntax "(1, c)_1" stands for a PEELED_CHREC that has the
188 following semantics: during the first iteration of the loop_1, the
189 variable contains the value 1, and then it contains the value "c".
190 Note that this syntax is close to the syntax of the loop-phi-node:
191 "a -> (1, c)_1" vs. "a = phi (1, c)".
193 The symbolic chrec representation contains all the semantics of the
194 original code. What is more difficult is to use this information.
196 Example 5: Flip-flops, or exchangers.
208 Based on these symbolic chrecs, it is possible to refine this
209 information into the more precise PERIODIC_CHRECs:
214 This transformation is not yet implemented.
218 You can find a more detailed description of the algorithm in:
219 http://icps.u-strasbg.fr/~pop/DEA_03_Pop.pdf
220 http://icps.u-strasbg.fr/~pop/DEA_03_Pop.ps.gz. But note that
221 this is a preliminary report and some of the details of the
222 algorithm have changed. I'm working on a research report that
223 updates the description of the algorithms to reflect the design
224 choices used in this implementation.
226 A set of slides show a high level overview of the algorithm and run
227 an example through the scalar evolution analyzer:
228 http://cri.ensmp.fr/~pop/gcc/mar04/slides.pdf
230 The slides that I have presented at the GCC Summit'04 are available
231 at: http://cri.ensmp.fr/~pop/gcc/20040604/gccsummit-lno-spop.pdf
236 #include "coretypes.h"
242 /* These RTL headers are needed for basic-block.h. */
244 #include "basic-block.h"
245 #include "diagnostic.h"
246 #include "tree-flow.h"
247 #include "tree-dump.h"
250 #include "tree-chrec.h"
251 #include "tree-scalar-evolution.h"
252 #include "tree-pass.h"
255 static tree
analyze_scalar_evolution_1 (struct loop
*, tree
, tree
);
256 static tree
resolve_mixers (struct loop
*, tree
);
258 /* The cached information about a ssa name VAR, claiming that inside LOOP,
259 the value of VAR can be expressed as CHREC. */
267 /* Counters for the scev database. */
268 static unsigned nb_set_scev
= 0;
269 static unsigned nb_get_scev
= 0;
271 /* The following trees are unique elements. Thus the comparison of
272 another element to these elements should be done on the pointer to
273 these trees, and not on their value. */
275 /* The SSA_NAMEs that are not yet analyzed are qualified with NULL_TREE. */
276 tree chrec_not_analyzed_yet
;
278 /* Reserved to the cases where the analyzer has detected an
279 undecidable property at compile time. */
280 tree chrec_dont_know
;
282 /* When the analyzer has detected that a property will never
283 happen, then it qualifies it with chrec_known. */
286 static bitmap already_instantiated
;
288 static htab_t scalar_evolution_info
;
291 /* Constructs a new SCEV_INFO_STR structure. */
293 static inline struct scev_info_str
*
294 new_scev_info_str (tree var
)
296 struct scev_info_str
*res
;
298 res
= xmalloc (sizeof (struct scev_info_str
));
300 res
->chrec
= chrec_not_analyzed_yet
;
305 /* Computes a hash function for database element ELT. */
308 hash_scev_info (const void *elt
)
310 return SSA_NAME_VERSION (((struct scev_info_str
*) elt
)->var
);
313 /* Compares database elements E1 and E2. */
316 eq_scev_info (const void *e1
, const void *e2
)
318 const struct scev_info_str
*elt1
= e1
;
319 const struct scev_info_str
*elt2
= e2
;
321 return elt1
->var
== elt2
->var
;
324 /* Deletes database element E. */
327 del_scev_info (void *e
)
332 /* Get the index corresponding to VAR in the current LOOP. If
333 it's the first time we ask for this VAR, then we return
334 chrec_not_analyzed_yet for this VAR and return its index. */
337 find_var_scev_info (tree var
)
339 struct scev_info_str
*res
;
340 struct scev_info_str tmp
;
344 slot
= htab_find_slot (scalar_evolution_info
, &tmp
, INSERT
);
347 *slot
= new_scev_info_str (var
);
353 /* Tries to express CHREC in wider type TYPE. */
356 count_ev_in_wider_type (tree type
, tree chrec
)
361 if (!evolution_function_is_affine_p (chrec
))
362 return fold_convert (type
, chrec
);
364 base
= CHREC_LEFT (chrec
);
365 step
= CHREC_RIGHT (chrec
);
366 loop
= current_loops
->parray
[CHREC_VARIABLE (chrec
)];
368 /* TODO -- if we knew the statement at that the conversion occurs,
369 we could pass it to can_count_iv_in_wider_type and get a better
371 step
= can_count_iv_in_wider_type (loop
, type
, base
, step
, NULL_TREE
);
373 return fold_convert (type
, chrec
);
374 base
= chrec_convert (type
, base
);
376 return build_polynomial_chrec (CHREC_VARIABLE (chrec
),
380 /* Return true when CHREC contains symbolic names defined in
384 chrec_contains_symbols_defined_in_loop (tree chrec
, unsigned loop_nb
)
386 if (chrec
== NULL_TREE
)
389 if (TREE_INVARIANT (chrec
))
392 if (TREE_CODE (chrec
) == VAR_DECL
393 || TREE_CODE (chrec
) == PARM_DECL
394 || TREE_CODE (chrec
) == FUNCTION_DECL
395 || TREE_CODE (chrec
) == LABEL_DECL
396 || TREE_CODE (chrec
) == RESULT_DECL
397 || TREE_CODE (chrec
) == FIELD_DECL
)
400 if (TREE_CODE (chrec
) == SSA_NAME
)
402 tree def
= SSA_NAME_DEF_STMT (chrec
);
403 struct loop
*def_loop
= loop_containing_stmt (def
);
404 struct loop
*loop
= current_loops
->parray
[loop_nb
];
406 if (def_loop
== NULL
)
409 if (loop
== def_loop
|| flow_loop_nested_p (loop
, def_loop
))
415 switch (TREE_CODE_LENGTH (TREE_CODE (chrec
)))
418 if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec
, 2),
423 if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec
, 1),
428 if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec
, 0),
437 /* Return true when PHI is a loop-phi-node. */
440 loop_phi_node_p (tree phi
)
442 /* The implementation of this function is based on the following
443 property: "all the loop-phi-nodes of a loop are contained in the
444 loop's header basic block". */
446 return loop_containing_stmt (phi
)->header
== bb_for_stmt (phi
);
449 /* Compute the scalar evolution for EVOLUTION_FN after crossing LOOP.
450 In general, in the case of multivariate evolutions we want to get
451 the evolution in different loops. LOOP specifies the level for
452 which to get the evolution.
456 | for (j = 0; j < 100; j++)
458 | for (k = 0; k < 100; k++)
460 | i = k + j; - Here the value of i is a function of j, k.
462 | ... = i - Here the value of i is a function of j.
464 | ... = i - Here the value of i is a scalar.
470 | i_1 = phi (i_0, i_2)
474 This loop has the same effect as:
475 LOOP_1 has the same effect as:
479 The overall effect of the loop, "i_0 + 20" in the previous example,
480 is obtained by passing in the parameters: LOOP = 1,
481 EVOLUTION_FN = {i_0, +, 2}_1.
485 compute_overall_effect_of_inner_loop (struct loop
*loop
, tree evolution_fn
)
489 if (evolution_fn
== chrec_dont_know
)
490 return chrec_dont_know
;
492 else if (TREE_CODE (evolution_fn
) == POLYNOMIAL_CHREC
)
494 if (CHREC_VARIABLE (evolution_fn
) >= (unsigned) loop
->num
)
496 struct loop
*inner_loop
=
497 current_loops
->parray
[CHREC_VARIABLE (evolution_fn
)];
498 tree nb_iter
= number_of_iterations_in_loop (inner_loop
);
500 if (nb_iter
== chrec_dont_know
)
501 return chrec_dont_know
;
506 /* Number of iterations is off by one (the ssa name we
507 analyze must be defined before the exit). */
508 nb_iter
= chrec_fold_minus (chrec_type (nb_iter
),
510 fold_convert (chrec_type (nb_iter
),
513 /* evolution_fn is the evolution function in LOOP. Get
514 its value in the nb_iter-th iteration. */
515 res
= chrec_apply (inner_loop
->num
, evolution_fn
, nb_iter
);
517 /* Continue the computation until ending on a parent of LOOP. */
518 return compute_overall_effect_of_inner_loop (loop
, res
);
525 /* If the evolution function is an invariant, there is nothing to do. */
526 else if (no_evolution_in_loop_p (evolution_fn
, loop
->num
, &val
) && val
)
530 return chrec_dont_know
;
533 /* Determine whether the CHREC is always positive/negative. If the expression
534 cannot be statically analyzed, return false, otherwise set the answer into
538 chrec_is_positive (tree chrec
, bool *value
)
545 switch (TREE_CODE (chrec
))
547 case POLYNOMIAL_CHREC
:
548 if (!chrec_is_positive (CHREC_LEFT (chrec
), &value0
)
549 || !chrec_is_positive (CHREC_RIGHT (chrec
), &value1
))
552 /* FIXME -- overflows. */
553 if (value0
== value1
)
559 /* Otherwise the chrec is under the form: "{-197, +, 2}_1",
560 and the proof consists in showing that the sign never
561 changes during the execution of the loop, from 0 to
562 loop->nb_iterations. */
563 if (!evolution_function_is_affine_p (chrec
))
566 nb_iter
= number_of_iterations_in_loop
567 (current_loops
->parray
[CHREC_VARIABLE (chrec
)]);
569 if (chrec_contains_undetermined (nb_iter
))
572 nb_iter
= chrec_fold_minus
573 (chrec_type (nb_iter
), nb_iter
,
574 build_int_cst (chrec_type (nb_iter
), 1));
577 /* TODO -- If the test is after the exit, we may decrease the number of
578 iterations by one. */
580 nb_iter
= chrec_fold_minus
581 (chrec_type (nb_iter
), nb_iter
,
582 build_int_cst (chrec_type (nb_iter
), 1));
585 end_value
= chrec_apply (CHREC_VARIABLE (chrec
), chrec
, nb_iter
);
587 if (!chrec_is_positive (end_value
, &value2
))
591 return value0
== value1
;
594 *value
= (tree_int_cst_sgn (chrec
) == 1);
602 /* Associate CHREC to SCALAR. */
605 set_scalar_evolution (tree scalar
, tree chrec
)
609 if (TREE_CODE (scalar
) != SSA_NAME
)
612 scalar_info
= find_var_scev_info (scalar
);
616 if (dump_flags
& TDF_DETAILS
)
618 fprintf (dump_file
, "(set_scalar_evolution \n");
619 fprintf (dump_file
, " (scalar = ");
620 print_generic_expr (dump_file
, scalar
, 0);
621 fprintf (dump_file
, ")\n (scalar_evolution = ");
622 print_generic_expr (dump_file
, chrec
, 0);
623 fprintf (dump_file
, "))\n");
625 if (dump_flags
& TDF_STATS
)
629 *scalar_info
= chrec
;
632 /* Retrieve the chrec associated to SCALAR in the LOOP. */
635 get_scalar_evolution (tree scalar
)
641 if (dump_flags
& TDF_DETAILS
)
643 fprintf (dump_file
, "(get_scalar_evolution \n");
644 fprintf (dump_file
, " (scalar = ");
645 print_generic_expr (dump_file
, scalar
, 0);
646 fprintf (dump_file
, ")\n");
648 if (dump_flags
& TDF_STATS
)
652 switch (TREE_CODE (scalar
))
655 res
= *find_var_scev_info (scalar
);
664 res
= chrec_not_analyzed_yet
;
668 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
670 fprintf (dump_file
, " (scalar_evolution = ");
671 print_generic_expr (dump_file
, res
, 0);
672 fprintf (dump_file
, "))\n");
678 /* Helper function for add_to_evolution. Returns the evolution
679 function for an assignment of the form "a = b + c", where "a" and
680 "b" are on the strongly connected component. CHREC_BEFORE is the
681 information that we already have collected up to this point.
682 TO_ADD is the evolution of "c".
684 When CHREC_BEFORE has an evolution part in LOOP_NB, add to this
685 evolution the expression TO_ADD, otherwise construct an evolution
686 part for this loop. */
689 add_to_evolution_1 (unsigned loop_nb
,
693 switch (TREE_CODE (chrec_before
))
695 case POLYNOMIAL_CHREC
:
696 if (CHREC_VARIABLE (chrec_before
) <= loop_nb
)
700 tree type
= chrec_type (chrec_before
);
702 /* When there is no evolution part in this loop, build it. */
703 if (CHREC_VARIABLE (chrec_before
) < loop_nb
)
707 right
= build_int_cst (type
, 0);
711 var
= CHREC_VARIABLE (chrec_before
);
712 left
= CHREC_LEFT (chrec_before
);
713 right
= CHREC_RIGHT (chrec_before
);
716 return build_polynomial_chrec
717 (var
, left
, chrec_fold_plus (type
, right
, to_add
));
720 /* Search the evolution in LOOP_NB. */
721 return build_polynomial_chrec
722 (CHREC_VARIABLE (chrec_before
),
723 add_to_evolution_1 (loop_nb
, CHREC_LEFT (chrec_before
), to_add
),
724 CHREC_RIGHT (chrec_before
));
727 /* These nodes do not depend on a loop. */
728 if (chrec_before
== chrec_dont_know
)
729 return chrec_dont_know
;
730 return build_polynomial_chrec (loop_nb
, chrec_before
, to_add
);
734 /* Add TO_ADD to the evolution part of CHREC_BEFORE in the dimension
737 Description (provided for completeness, for those who read code in
738 a plane, and for my poor 62 bytes brain that would have forgotten
739 all this in the next two or three months):
741 The algorithm of translation of programs from the SSA representation
742 into the chrecs syntax is based on a pattern matching. After having
743 reconstructed the overall tree expression for a loop, there are only
744 two cases that can arise:
746 1. a = loop-phi (init, a + expr)
747 2. a = loop-phi (init, expr)
749 where EXPR is either a scalar constant with respect to the analyzed
750 loop (this is a degree 0 polynomial), or an expression containing
751 other loop-phi definitions (these are higher degree polynomials).
758 | a = phi (init, a + 5)
765 | a = phi (inita, 2 * b + 3)
766 | b = phi (initb, b + 1)
769 For the first case, the semantics of the SSA representation is:
771 | a (x) = init + \sum_{j = 0}^{x - 1} expr (j)
773 that is, there is a loop index "x" that determines the scalar value
774 of the variable during the loop execution. During the first
775 iteration, the value is that of the initial condition INIT, while
776 during the subsequent iterations, it is the sum of the initial
777 condition with the sum of all the values of EXPR from the initial
778 iteration to the before last considered iteration.
780 For the second case, the semantics of the SSA program is:
782 | a (x) = init, if x = 0;
783 | expr (x - 1), otherwise.
785 The second case corresponds to the PEELED_CHREC, whose syntax is
786 close to the syntax of a loop-phi-node:
788 | phi (init, expr) vs. (init, expr)_x
790 The proof of the translation algorithm for the first case is a
791 proof by structural induction based on the degree of EXPR.
794 When EXPR is a constant with respect to the analyzed loop, or in
795 other words when EXPR is a polynomial of degree 0, the evolution of
796 the variable A in the loop is an affine function with an initial
797 condition INIT, and a step EXPR. In order to show this, we start
798 from the semantics of the SSA representation:
800 f (x) = init + \sum_{j = 0}^{x - 1} expr (j)
802 and since "expr (j)" is a constant with respect to "j",
804 f (x) = init + x * expr
806 Finally, based on the semantics of the pure sum chrecs, by
807 identification we get the corresponding chrecs syntax:
809 f (x) = init * \binom{x}{0} + expr * \binom{x}{1}
810 f (x) -> {init, +, expr}_x
813 Suppose that EXPR is a polynomial of degree N with respect to the
814 analyzed loop_x for which we have already determined that it is
815 written under the chrecs syntax:
817 | expr (x) -> {b_0, +, b_1, +, ..., +, b_{n-1}} (x)
819 We start from the semantics of the SSA program:
821 | f (x) = init + \sum_{j = 0}^{x - 1} expr (j)
823 | f (x) = init + \sum_{j = 0}^{x - 1}
824 | (b_0 * \binom{j}{0} + ... + b_{n-1} * \binom{j}{n-1})
826 | f (x) = init + \sum_{j = 0}^{x - 1}
827 | \sum_{k = 0}^{n - 1} (b_k * \binom{j}{k})
829 | f (x) = init + \sum_{k = 0}^{n - 1}
830 | (b_k * \sum_{j = 0}^{x - 1} \binom{j}{k})
832 | f (x) = init + \sum_{k = 0}^{n - 1}
833 | (b_k * \binom{x}{k + 1})
835 | f (x) = init + b_0 * \binom{x}{1} + ...
836 | + b_{n-1} * \binom{x}{n}
838 | f (x) = init * \binom{x}{0} + b_0 * \binom{x}{1} + ...
839 | + b_{n-1} * \binom{x}{n}
842 And finally from the definition of the chrecs syntax, we identify:
843 | f (x) -> {init, +, b_0, +, ..., +, b_{n-1}}_x
845 This shows the mechanism that stands behind the add_to_evolution
846 function. An important point is that the use of symbolic
847 parameters avoids the need of an analysis schedule.
854 | a = phi (inita, a + 2 + b)
855 | b = phi (initb, b + 1)
858 When analyzing "a", the algorithm keeps "b" symbolically:
860 | a -> {inita, +, 2 + b}_1
862 Then, after instantiation, the analyzer ends on the evolution:
864 | a -> {inita, +, 2 + initb, +, 1}_1
869 add_to_evolution (unsigned loop_nb
,
874 tree type
= chrec_type (to_add
);
875 tree res
= NULL_TREE
;
877 if (to_add
== NULL_TREE
)
880 /* TO_ADD is either a scalar, or a parameter. TO_ADD is not
881 instantiated at this point. */
882 if (TREE_CODE (to_add
) == POLYNOMIAL_CHREC
)
883 /* This should not happen. */
884 return chrec_dont_know
;
886 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
888 fprintf (dump_file
, "(add_to_evolution \n");
889 fprintf (dump_file
, " (loop_nb = %d)\n", loop_nb
);
890 fprintf (dump_file
, " (chrec_before = ");
891 print_generic_expr (dump_file
, chrec_before
, 0);
892 fprintf (dump_file
, ")\n (to_add = ");
893 print_generic_expr (dump_file
, to_add
, 0);
894 fprintf (dump_file
, ")\n");
897 if (code
== MINUS_EXPR
)
898 to_add
= chrec_fold_multiply (type
, to_add
,
899 fold_convert (type
, integer_minus_one_node
));
901 res
= add_to_evolution_1 (loop_nb
, chrec_before
, to_add
);
903 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
905 fprintf (dump_file
, " (res = ");
906 print_generic_expr (dump_file
, res
, 0);
907 fprintf (dump_file
, "))\n");
913 /* Helper function. */
916 set_nb_iterations_in_loop (struct loop
*loop
,
919 res
= chrec_fold_plus (chrec_type (res
), res
, integer_one_node
);
920 /* FIXME HWI: However we want to store one iteration less than the
921 count of the loop in order to be compatible with the other
922 nb_iter computations in loop-iv. This also allows the
923 representation of nb_iters that are equal to MAX_INT. */
924 if ((TREE_CODE (res
) == INTEGER_CST
&& TREE_INT_CST_LOW (res
) == 0)
925 || TREE_OVERFLOW (res
))
926 res
= chrec_dont_know
;
928 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
930 fprintf (dump_file
, " (set_nb_iterations_in_loop = ");
931 print_generic_expr (dump_file
, res
, 0);
932 fprintf (dump_file
, "))\n");
935 loop
->nb_iterations
= res
;
941 /* This section selects the loops that will be good candidates for the
942 scalar evolution analysis. For the moment, greedily select all the
943 loop nests we could analyze. */
945 /* Return true when it is possible to analyze the condition expression
949 analyzable_condition (tree expr
)
953 if (TREE_CODE (expr
) != COND_EXPR
)
956 condition
= TREE_OPERAND (expr
, 0);
958 switch (TREE_CODE (condition
))
961 /* Volatile expressions are not analyzable. */
962 if (TREE_THIS_VOLATILE (SSA_NAME_VAR (condition
)))
975 opnd0
= TREE_OPERAND (condition
, 0);
976 opnd1
= TREE_OPERAND (condition
, 1);
978 if (TREE_CODE (opnd0
) == SSA_NAME
979 && TREE_THIS_VOLATILE (SSA_NAME_VAR (opnd0
)))
982 if (TREE_CODE (opnd1
) == SSA_NAME
983 && TREE_THIS_VOLATILE (SSA_NAME_VAR (opnd1
)))
996 /* For a loop with a single exit edge, return the COND_EXPR that
997 guards the exit edge. If the expression is too difficult to
998 analyze, then give up. */
1001 get_loop_exit_condition (struct loop
*loop
)
1003 tree res
= NULL_TREE
;
1004 edge exit_edge
= loop
->single_exit
;
1007 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1008 fprintf (dump_file
, "(get_loop_exit_condition \n ");
1014 expr
= last_stmt (exit_edge
->src
);
1015 if (analyzable_condition (expr
))
1019 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1021 print_generic_expr (dump_file
, res
, 0);
1022 fprintf (dump_file
, ")\n");
1028 /* Recursively determine and enqueue the exit conditions for a loop. */
1031 get_exit_conditions_rec (struct loop
*loop
,
1032 varray_type
*exit_conditions
)
1037 /* Recurse on the inner loops, then on the next (sibling) loops. */
1038 get_exit_conditions_rec (loop
->inner
, exit_conditions
);
1039 get_exit_conditions_rec (loop
->next
, exit_conditions
);
1041 if (loop
->single_exit
)
1043 tree loop_condition
= get_loop_exit_condition (loop
);
1046 VARRAY_PUSH_TREE (*exit_conditions
, loop_condition
);
1050 /* Select the candidate loop nests for the analysis. This function
1051 initializes the EXIT_CONDITIONS array. */
1054 select_loops_exit_conditions (struct loops
*loops
,
1055 varray_type
*exit_conditions
)
1057 struct loop
*function_body
= loops
->parray
[0];
1059 get_exit_conditions_rec (function_body
->inner
, exit_conditions
);
1063 /* Depth first search algorithm. */
1065 static bool follow_ssa_edge (struct loop
*loop
, tree
, tree
, tree
*);
1067 /* Follow the ssa edge into the right hand side RHS of an assignment.
1068 Return true if the strongly connected component has been found. */
1071 follow_ssa_edge_in_rhs (struct loop
*loop
,
1074 tree
*evolution_of_loop
)
1078 tree type_rhs
= TREE_TYPE (rhs
);
1080 /* The RHS is one of the following cases:
1085 - other cases are not yet handled.
1087 switch (TREE_CODE (rhs
))
1090 /* This assignment is under the form "a_1 = (cast) rhs. */
1091 res
= follow_ssa_edge_in_rhs (loop
, TREE_OPERAND (rhs
, 0), halting_phi
,
1093 *evolution_of_loop
= chrec_convert (TREE_TYPE (rhs
), *evolution_of_loop
);
1097 /* This assignment is under the form "a_1 = 7". */
1102 /* This assignment is under the form: "a_1 = b_2". */
1103 res
= follow_ssa_edge
1104 (loop
, SSA_NAME_DEF_STMT (rhs
), halting_phi
, evolution_of_loop
);
1108 /* This case is under the form "rhs0 + rhs1". */
1109 rhs0
= TREE_OPERAND (rhs
, 0);
1110 rhs1
= TREE_OPERAND (rhs
, 1);
1111 STRIP_TYPE_NOPS (rhs0
);
1112 STRIP_TYPE_NOPS (rhs1
);
1114 if (TREE_CODE (rhs0
) == SSA_NAME
)
1116 if (TREE_CODE (rhs1
) == SSA_NAME
)
1118 /* Match an assignment under the form:
1120 res
= follow_ssa_edge
1121 (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
,
1125 *evolution_of_loop
= add_to_evolution
1127 chrec_convert (type_rhs
, *evolution_of_loop
),
1132 res
= follow_ssa_edge
1133 (loop
, SSA_NAME_DEF_STMT (rhs1
), halting_phi
,
1137 *evolution_of_loop
= add_to_evolution
1139 chrec_convert (type_rhs
, *evolution_of_loop
),
1146 /* Match an assignment under the form:
1148 res
= follow_ssa_edge
1149 (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
,
1152 *evolution_of_loop
= add_to_evolution
1153 (loop
->num
, chrec_convert (type_rhs
, *evolution_of_loop
),
1158 else if (TREE_CODE (rhs1
) == SSA_NAME
)
1160 /* Match an assignment under the form:
1162 res
= follow_ssa_edge
1163 (loop
, SSA_NAME_DEF_STMT (rhs1
), halting_phi
,
1166 *evolution_of_loop
= add_to_evolution
1167 (loop
->num
, chrec_convert (type_rhs
, *evolution_of_loop
),
1172 /* Otherwise, match an assignment under the form:
1174 /* And there is nothing to do. */
1180 /* This case is under the form "opnd0 = rhs0 - rhs1". */
1181 rhs0
= TREE_OPERAND (rhs
, 0);
1182 rhs1
= TREE_OPERAND (rhs
, 1);
1183 STRIP_TYPE_NOPS (rhs0
);
1184 STRIP_TYPE_NOPS (rhs1
);
1186 if (TREE_CODE (rhs0
) == SSA_NAME
)
1188 if (TREE_CODE (rhs1
) == SSA_NAME
)
1190 /* Match an assignment under the form:
1192 res
= follow_ssa_edge
1193 (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
,
1197 *evolution_of_loop
= add_to_evolution
1198 (loop
->num
, chrec_convert (type_rhs
, *evolution_of_loop
),
1203 res
= follow_ssa_edge
1204 (loop
, SSA_NAME_DEF_STMT (rhs1
), halting_phi
,
1208 *evolution_of_loop
= add_to_evolution
1210 chrec_fold_multiply (type_rhs
,
1212 fold_convert (type_rhs
,
1213 integer_minus_one_node
)),
1220 /* Match an assignment under the form:
1222 res
= follow_ssa_edge
1223 (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
,
1226 *evolution_of_loop
= add_to_evolution
1227 (loop
->num
, chrec_convert (type_rhs
, *evolution_of_loop
),
1232 else if (TREE_CODE (rhs1
) == SSA_NAME
)
1234 /* Match an assignment under the form:
1236 res
= follow_ssa_edge
1237 (loop
, SSA_NAME_DEF_STMT (rhs1
), halting_phi
,
1240 *evolution_of_loop
= add_to_evolution
1242 chrec_fold_multiply (type_rhs
,
1244 fold_convert (type_rhs
, integer_minus_one_node
)),
1249 /* Otherwise, match an assignment under the form:
1251 /* And there is nothing to do. */
1257 /* This case is under the form "opnd0 = rhs0 * rhs1". */
1258 rhs0
= TREE_OPERAND (rhs
, 0);
1259 rhs1
= TREE_OPERAND (rhs
, 1);
1260 STRIP_TYPE_NOPS (rhs0
);
1261 STRIP_TYPE_NOPS (rhs1
);
1263 if (TREE_CODE (rhs0
) == SSA_NAME
)
1265 if (TREE_CODE (rhs1
) == SSA_NAME
)
1267 /* Match an assignment under the form:
1269 res
= follow_ssa_edge
1270 (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
,
1274 *evolution_of_loop
= chrec_dont_know
;
1278 res
= follow_ssa_edge
1279 (loop
, SSA_NAME_DEF_STMT (rhs1
), halting_phi
,
1283 *evolution_of_loop
= chrec_dont_know
;
1289 /* Match an assignment under the form:
1291 res
= follow_ssa_edge
1292 (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
,
1295 *evolution_of_loop
= chrec_dont_know
;
1299 else if (TREE_CODE (rhs1
) == SSA_NAME
)
1301 /* Match an assignment under the form:
1303 res
= follow_ssa_edge
1304 (loop
, SSA_NAME_DEF_STMT (rhs1
), halting_phi
,
1307 *evolution_of_loop
= chrec_dont_know
;
1311 /* Otherwise, match an assignment under the form:
1313 /* And there is nothing to do. */
1326 /* Checks whether the I-th argument of a PHI comes from a backedge. */
1329 backedge_phi_arg_p (tree phi
, int i
)
1331 edge e
= PHI_ARG_EDGE (phi
, i
);
1333 /* We would in fact like to test EDGE_DFS_BACK here, but we do not care
1334 about updating it anywhere, and this should work as well most of the
1336 if (e
->flags
& EDGE_IRREDUCIBLE_LOOP
)
1342 /* Helper function for one branch of the condition-phi-node. Return
1343 true if the strongly connected component has been found following
1347 follow_ssa_edge_in_condition_phi_branch (int i
,
1351 tree
*evolution_of_branch
,
1354 tree branch
= PHI_ARG_DEF (condition_phi
, i
);
1355 *evolution_of_branch
= chrec_dont_know
;
1357 /* Do not follow back edges (they must belong to an irreducible loop, which
1358 we really do not want to worry about). */
1359 if (backedge_phi_arg_p (condition_phi
, i
))
1362 if (TREE_CODE (branch
) == SSA_NAME
)
1364 *evolution_of_branch
= init_cond
;
1365 return follow_ssa_edge (loop
, SSA_NAME_DEF_STMT (branch
), halting_phi
,
1366 evolution_of_branch
);
1369 /* This case occurs when one of the condition branches sets
1370 the variable to a constant: i.e. a phi-node like
1371 "a_2 = PHI <a_7(5), 2(6)>;".
1373 FIXME: This case have to be refined correctly:
1374 in some cases it is possible to say something better than
1375 chrec_dont_know, for example using a wrap-around notation. */
1379 /* This function merges the branches of a condition-phi-node in a
1383 follow_ssa_edge_in_condition_phi (struct loop
*loop
,
1386 tree
*evolution_of_loop
)
1389 tree init
= *evolution_of_loop
;
1390 tree evolution_of_branch
;
1392 if (!follow_ssa_edge_in_condition_phi_branch (0, loop
, condition_phi
,
1394 &evolution_of_branch
,
1397 *evolution_of_loop
= evolution_of_branch
;
1399 for (i
= 1; i
< PHI_NUM_ARGS (condition_phi
); i
++)
1401 if (!follow_ssa_edge_in_condition_phi_branch (i
, loop
, condition_phi
,
1403 &evolution_of_branch
,
1407 *evolution_of_loop
= chrec_merge (*evolution_of_loop
,
1408 evolution_of_branch
);
1414 /* Follow an SSA edge in an inner loop. It computes the overall
1415 effect of the loop, and following the symbolic initial conditions,
1416 it follows the edges in the parent loop. The inner loop is
1417 considered as a single statement. */
1420 follow_ssa_edge_inner_loop_phi (struct loop
*outer_loop
,
1423 tree
*evolution_of_loop
)
1425 struct loop
*loop
= loop_containing_stmt (loop_phi_node
);
1426 tree ev
= analyze_scalar_evolution (loop
, PHI_RESULT (loop_phi_node
));
1428 /* Sometimes, the inner loop is too difficult to analyze, and the
1429 result of the analysis is a symbolic parameter. */
1430 if (ev
== PHI_RESULT (loop_phi_node
))
1435 for (i
= 0; i
< PHI_NUM_ARGS (loop_phi_node
); i
++)
1437 tree arg
= PHI_ARG_DEF (loop_phi_node
, i
);
1440 /* Follow the edges that exit the inner loop. */
1441 bb
= PHI_ARG_EDGE (loop_phi_node
, i
)->src
;
1442 if (!flow_bb_inside_loop_p (loop
, bb
))
1443 res
= res
|| follow_ssa_edge_in_rhs (outer_loop
, arg
, halting_phi
,
1447 /* If the path crosses this loop-phi, give up. */
1449 *evolution_of_loop
= chrec_dont_know
;
1454 /* Otherwise, compute the overall effect of the inner loop. */
1455 ev
= compute_overall_effect_of_inner_loop (loop
, ev
);
1456 return follow_ssa_edge_in_rhs (outer_loop
, ev
, halting_phi
,
1460 /* Follow an SSA edge from a loop-phi-node to itself, constructing a
1461 path that is analyzed on the return walk. */
1464 follow_ssa_edge (struct loop
*loop
,
1467 tree
*evolution_of_loop
)
1469 struct loop
*def_loop
;
1471 if (TREE_CODE (def
) == NOP_EXPR
)
1474 def_loop
= loop_containing_stmt (def
);
1476 switch (TREE_CODE (def
))
1479 if (!loop_phi_node_p (def
))
1480 /* DEF is a condition-phi-node. Follow the branches, and
1481 record their evolutions. Finally, merge the collected
1482 information and set the approximation to the main
1484 return follow_ssa_edge_in_condition_phi
1485 (loop
, def
, halting_phi
, evolution_of_loop
);
1487 /* When the analyzed phi is the halting_phi, the
1488 depth-first search is over: we have found a path from
1489 the halting_phi to itself in the loop. */
1490 if (def
== halting_phi
)
1493 /* Otherwise, the evolution of the HALTING_PHI depends
1494 on the evolution of another loop-phi-node, i.e. the
1495 evolution function is a higher degree polynomial. */
1496 if (def_loop
== loop
)
1500 if (flow_loop_nested_p (loop
, def_loop
))
1501 return follow_ssa_edge_inner_loop_phi
1502 (loop
, def
, halting_phi
, evolution_of_loop
);
1508 return follow_ssa_edge_in_rhs (loop
,
1509 TREE_OPERAND (def
, 1),
1514 /* At this level of abstraction, the program is just a set
1515 of MODIFY_EXPRs and PHI_NODEs. In principle there is no
1516 other node to be handled. */
1523 /* Given a LOOP_PHI_NODE, this function determines the evolution
1524 function from LOOP_PHI_NODE to LOOP_PHI_NODE in the loop. */
1527 analyze_evolution_in_loop (tree loop_phi_node
,
1531 tree evolution_function
= chrec_not_analyzed_yet
;
1532 struct loop
*loop
= loop_containing_stmt (loop_phi_node
);
1535 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1537 fprintf (dump_file
, "(analyze_evolution_in_loop \n");
1538 fprintf (dump_file
, " (loop_phi_node = ");
1539 print_generic_expr (dump_file
, loop_phi_node
, 0);
1540 fprintf (dump_file
, ")\n");
1543 for (i
= 0; i
< PHI_NUM_ARGS (loop_phi_node
); i
++)
1545 tree arg
= PHI_ARG_DEF (loop_phi_node
, i
);
1546 tree ssa_chain
, ev_fn
;
1549 /* Select the edges that enter the loop body. */
1550 bb
= PHI_ARG_EDGE (loop_phi_node
, i
)->src
;
1551 if (!flow_bb_inside_loop_p (loop
, bb
))
1554 if (TREE_CODE (arg
) == SSA_NAME
)
1556 ssa_chain
= SSA_NAME_DEF_STMT (arg
);
1558 /* Pass in the initial condition to the follow edge function. */
1560 res
= follow_ssa_edge (loop
, ssa_chain
, loop_phi_node
, &ev_fn
);
1565 /* When it is impossible to go back on the same
1566 loop_phi_node by following the ssa edges, the
1567 evolution is represented by a peeled chrec, i.e. the
1568 first iteration, EV_FN has the value INIT_COND, then
1569 all the other iterations it has the value of ARG.
1570 For the moment, PEELED_CHREC nodes are not built. */
1572 ev_fn
= chrec_dont_know
;
1574 /* When there are multiple back edges of the loop (which in fact never
1575 happens currently, but nevertheless), merge their evolutions. */
1576 evolution_function
= chrec_merge (evolution_function
, ev_fn
);
1579 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1581 fprintf (dump_file
, " (evolution_function = ");
1582 print_generic_expr (dump_file
, evolution_function
, 0);
1583 fprintf (dump_file
, "))\n");
1586 return evolution_function
;
1589 /* Given a loop-phi-node, return the initial conditions of the
1590 variable on entry of the loop. When the CCP has propagated
1591 constants into the loop-phi-node, the initial condition is
1592 instantiated, otherwise the initial condition is kept symbolic.
1593 This analyzer does not analyze the evolution outside the current
1594 loop, and leaves this task to the on-demand tree reconstructor. */
1597 analyze_initial_condition (tree loop_phi_node
)
1600 tree init_cond
= chrec_not_analyzed_yet
;
1601 struct loop
*loop
= bb_for_stmt (loop_phi_node
)->loop_father
;
1603 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1605 fprintf (dump_file
, "(analyze_initial_condition \n");
1606 fprintf (dump_file
, " (loop_phi_node = \n");
1607 print_generic_expr (dump_file
, loop_phi_node
, 0);
1608 fprintf (dump_file
, ")\n");
1611 for (i
= 0; i
< PHI_NUM_ARGS (loop_phi_node
); i
++)
1613 tree branch
= PHI_ARG_DEF (loop_phi_node
, i
);
1614 basic_block bb
= PHI_ARG_EDGE (loop_phi_node
, i
)->src
;
1616 /* When the branch is oriented to the loop's body, it does
1617 not contribute to the initial condition. */
1618 if (flow_bb_inside_loop_p (loop
, bb
))
1621 if (init_cond
== chrec_not_analyzed_yet
)
1627 if (TREE_CODE (branch
) == SSA_NAME
)
1629 init_cond
= chrec_dont_know
;
1633 init_cond
= chrec_merge (init_cond
, branch
);
1636 /* Ooops -- a loop without an entry??? */
1637 if (init_cond
== chrec_not_analyzed_yet
)
1638 init_cond
= chrec_dont_know
;
1640 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1642 fprintf (dump_file
, " (init_cond = ");
1643 print_generic_expr (dump_file
, init_cond
, 0);
1644 fprintf (dump_file
, "))\n");
1650 /* Analyze the scalar evolution for LOOP_PHI_NODE. */
1653 interpret_loop_phi (struct loop
*loop
, tree loop_phi_node
)
1656 struct loop
*phi_loop
= loop_containing_stmt (loop_phi_node
);
1659 if (phi_loop
!= loop
)
1661 struct loop
*subloop
;
1662 tree evolution_fn
= analyze_scalar_evolution
1663 (phi_loop
, PHI_RESULT (loop_phi_node
));
1665 /* Dive one level deeper. */
1666 subloop
= superloop_at_depth (phi_loop
, loop
->depth
+ 1);
1668 /* Interpret the subloop. */
1669 res
= compute_overall_effect_of_inner_loop (subloop
, evolution_fn
);
1673 /* Otherwise really interpret the loop phi. */
1674 init_cond
= analyze_initial_condition (loop_phi_node
);
1675 res
= analyze_evolution_in_loop (loop_phi_node
, init_cond
);
1680 /* This function merges the branches of a condition-phi-node,
1681 contained in the outermost loop, and whose arguments are already
1685 interpret_condition_phi (struct loop
*loop
, tree condition_phi
)
1688 tree res
= chrec_not_analyzed_yet
;
1690 for (i
= 0; i
< PHI_NUM_ARGS (condition_phi
); i
++)
1694 if (backedge_phi_arg_p (condition_phi
, i
))
1696 res
= chrec_dont_know
;
1700 branch_chrec
= analyze_scalar_evolution
1701 (loop
, PHI_ARG_DEF (condition_phi
, i
));
1703 res
= chrec_merge (res
, branch_chrec
);
1709 /* Interpret the right hand side of a modify_expr OPND1. If we didn't
1710 analyzed this node before, follow the definitions until ending
1711 either on an analyzed modify_expr, or on a loop-phi-node. On the
1712 return path, this function propagates evolutions (ala constant copy
1713 propagation). OPND1 is not a GIMPLE expression because we could
1714 analyze the effect of an inner loop: see interpret_loop_phi. */
1717 interpret_rhs_modify_expr (struct loop
*loop
,
1718 tree opnd1
, tree type
)
1720 tree res
, opnd10
, opnd11
, chrec10
, chrec11
;
1722 if (is_gimple_min_invariant (opnd1
))
1723 return chrec_convert (type
, opnd1
);
1725 switch (TREE_CODE (opnd1
))
1728 opnd10
= TREE_OPERAND (opnd1
, 0);
1729 opnd11
= TREE_OPERAND (opnd1
, 1);
1730 chrec10
= analyze_scalar_evolution (loop
, opnd10
);
1731 chrec11
= analyze_scalar_evolution (loop
, opnd11
);
1732 chrec10
= chrec_convert (type
, chrec10
);
1733 chrec11
= chrec_convert (type
, chrec11
);
1734 res
= chrec_fold_plus (type
, chrec10
, chrec11
);
1738 opnd10
= TREE_OPERAND (opnd1
, 0);
1739 opnd11
= TREE_OPERAND (opnd1
, 1);
1740 chrec10
= analyze_scalar_evolution (loop
, opnd10
);
1741 chrec11
= analyze_scalar_evolution (loop
, opnd11
);
1742 chrec10
= chrec_convert (type
, chrec10
);
1743 chrec11
= chrec_convert (type
, chrec11
);
1744 res
= chrec_fold_minus (type
, chrec10
, chrec11
);
1748 opnd10
= TREE_OPERAND (opnd1
, 0);
1749 chrec10
= analyze_scalar_evolution (loop
, opnd10
);
1750 chrec10
= chrec_convert (type
, chrec10
);
1751 res
= chrec_fold_minus (type
, build_int_cst (type
, 0), chrec10
);
1755 opnd10
= TREE_OPERAND (opnd1
, 0);
1756 opnd11
= TREE_OPERAND (opnd1
, 1);
1757 chrec10
= analyze_scalar_evolution (loop
, opnd10
);
1758 chrec11
= analyze_scalar_evolution (loop
, opnd11
);
1759 chrec10
= chrec_convert (type
, chrec10
);
1760 chrec11
= chrec_convert (type
, chrec11
);
1761 res
= chrec_fold_multiply (type
, chrec10
, chrec11
);
1765 res
= chrec_convert (type
, analyze_scalar_evolution (loop
, opnd1
));
1770 opnd10
= TREE_OPERAND (opnd1
, 0);
1771 chrec10
= analyze_scalar_evolution (loop
, opnd10
);
1772 res
= chrec_convert (type
, chrec10
);
1776 res
= chrec_dont_know
;
1785 /* This section contains all the entry points:
1786 - number_of_iterations_in_loop,
1787 - analyze_scalar_evolution,
1788 - instantiate_parameters.
1791 /* Compute and return the evolution function in WRTO_LOOP, the nearest
1792 common ancestor of DEF_LOOP and USE_LOOP. */
1795 compute_scalar_evolution_in_loop (struct loop
*wrto_loop
,
1796 struct loop
*def_loop
,
1800 if (def_loop
== wrto_loop
)
1803 def_loop
= superloop_at_depth (def_loop
, wrto_loop
->depth
+ 1);
1804 res
= compute_overall_effect_of_inner_loop (def_loop
, ev
);
1806 return analyze_scalar_evolution_1 (wrto_loop
, res
, chrec_not_analyzed_yet
);
1809 /* Helper recursive function. */
1812 analyze_scalar_evolution_1 (struct loop
*loop
, tree var
, tree res
)
1814 tree def
, type
= TREE_TYPE (var
);
1816 struct loop
*def_loop
;
1819 return chrec_dont_know
;
1821 if (TREE_CODE (var
) != SSA_NAME
)
1822 return interpret_rhs_modify_expr (loop
, var
, type
);
1824 def
= SSA_NAME_DEF_STMT (var
);
1825 bb
= bb_for_stmt (def
);
1826 def_loop
= bb
? bb
->loop_father
: NULL
;
1829 || !flow_bb_inside_loop_p (loop
, bb
))
1831 /* Keep the symbolic form. */
1836 if (res
!= chrec_not_analyzed_yet
)
1838 if (loop
!= bb
->loop_father
)
1839 res
= compute_scalar_evolution_in_loop
1840 (find_common_loop (loop
, bb
->loop_father
), bb
->loop_father
, res
);
1845 if (loop
!= def_loop
)
1847 res
= analyze_scalar_evolution_1 (def_loop
, var
, chrec_not_analyzed_yet
);
1848 res
= compute_scalar_evolution_in_loop (loop
, def_loop
, res
);
1853 switch (TREE_CODE (def
))
1856 res
= interpret_rhs_modify_expr (loop
, TREE_OPERAND (def
, 1), type
);
1860 if (loop_phi_node_p (def
))
1861 res
= interpret_loop_phi (loop
, def
);
1863 res
= interpret_condition_phi (loop
, def
);
1867 res
= chrec_dont_know
;
1873 /* Keep the symbolic form. */
1874 if (res
== chrec_dont_know
)
1877 if (loop
== def_loop
)
1878 set_scalar_evolution (var
, res
);
1883 /* Entry point for the scalar evolution analyzer.
1884 Analyzes and returns the scalar evolution of the ssa_name VAR.
1885 LOOP_NB is the identifier number of the loop in which the variable
1888 Example of use: having a pointer VAR to a SSA_NAME node, STMT a
1889 pointer to the statement that uses this variable, in order to
1890 determine the evolution function of the variable, use the following
1893 unsigned loop_nb = loop_containing_stmt (stmt)->num;
1894 tree chrec_with_symbols = analyze_scalar_evolution (loop_nb, var);
1895 tree chrec_instantiated = instantiate_parameters
1896 (loop_nb, chrec_with_symbols);
1900 analyze_scalar_evolution (struct loop
*loop
, tree var
)
1904 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1906 fprintf (dump_file
, "(analyze_scalar_evolution \n");
1907 fprintf (dump_file
, " (loop_nb = %d)\n", loop
->num
);
1908 fprintf (dump_file
, " (scalar = ");
1909 print_generic_expr (dump_file
, var
, 0);
1910 fprintf (dump_file
, ")\n");
1913 res
= analyze_scalar_evolution_1 (loop
, var
, get_scalar_evolution (var
));
1915 if (TREE_CODE (var
) == SSA_NAME
&& res
== chrec_dont_know
)
1918 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1919 fprintf (dump_file
, ")\n");
1924 /* Analyze scalar evolution of use of VERSION in USE_LOOP with respect to
1925 WRTO_LOOP (which should be a superloop of both USE_LOOP and definition
1929 analyze_scalar_evolution_in_loop (struct loop
*wrto_loop
, struct loop
*use_loop
,
1937 ev
= analyze_scalar_evolution (use_loop
, ev
);
1938 ev
= resolve_mixers (use_loop
, ev
);
1940 if (use_loop
== wrto_loop
)
1943 /* If the value of the use changes in the inner loop, we cannot express
1944 its value in the outer loop (we might try to return interval chrec,
1945 but we do not have a user for it anyway) */
1946 if (!no_evolution_in_loop_p (ev
, use_loop
->num
, &val
)
1948 return chrec_dont_know
;
1950 use_loop
= use_loop
->outer
;
1954 /* Analyze all the parameters of the chrec that were left under a symbolic form,
1955 with respect to LOOP. CHREC is the chrec to instantiate. If
1956 ALLOW_SUPERLOOP_CHRECS is true, replacing loop invariants with
1957 outer loop chrecs is done. */
1960 instantiate_parameters_1 (struct loop
*loop
, tree chrec
,
1961 bool allow_superloop_chrecs
)
1963 tree res
, op0
, op1
, op2
;
1965 struct loop
*def_loop
;
1967 if (chrec
== NULL_TREE
1968 || automatically_generated_chrec_p (chrec
))
1971 if (is_gimple_min_invariant (chrec
))
1974 switch (TREE_CODE (chrec
))
1977 def_bb
= bb_for_stmt (SSA_NAME_DEF_STMT (chrec
));
1979 /* A parameter (or loop invariant and we do not want to include
1980 evolutions in outer loops), nothing to do. */
1982 || (!allow_superloop_chrecs
1983 && !flow_bb_inside_loop_p (loop
, def_bb
)))
1986 /* Don't instantiate the SSA_NAME if it is in a mixer
1987 structure. This is used for avoiding the instantiation of
1988 recursively defined functions, such as:
1990 | a_2 -> {0, +, 1, +, a_2}_1 */
1992 if (bitmap_bit_p (already_instantiated
, SSA_NAME_VERSION (chrec
)))
1994 if (!flow_bb_inside_loop_p (loop
, def_bb
))
1996 /* We may keep the loop invariant in symbolic form. */
2001 /* Something with unknown behavior in LOOP. */
2002 return chrec_dont_know
;
2006 def_loop
= find_common_loop (loop
, def_bb
->loop_father
);
2008 /* If the analysis yields a parametric chrec, instantiate the
2009 result again. Avoid the cyclic instantiation in mixers. */
2010 bitmap_set_bit (already_instantiated
, SSA_NAME_VERSION (chrec
));
2011 res
= analyze_scalar_evolution (def_loop
, chrec
);
2012 res
= instantiate_parameters_1 (loop
, res
, allow_superloop_chrecs
);
2013 bitmap_clear_bit (already_instantiated
, SSA_NAME_VERSION (chrec
));
2016 case POLYNOMIAL_CHREC
:
2017 op0
= instantiate_parameters_1 (loop
, CHREC_LEFT (chrec
),
2018 allow_superloop_chrecs
);
2019 op1
= instantiate_parameters_1 (loop
, CHREC_RIGHT (chrec
),
2020 allow_superloop_chrecs
);
2021 return build_polynomial_chrec (CHREC_VARIABLE (chrec
), op0
, op1
);
2024 op0
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 0),
2025 allow_superloop_chrecs
);
2026 op1
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 1),
2027 allow_superloop_chrecs
);
2028 return chrec_fold_plus (TREE_TYPE (chrec
), op0
, op1
);
2031 op0
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 0),
2032 allow_superloop_chrecs
);
2033 op1
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 1),
2034 allow_superloop_chrecs
);
2035 return chrec_fold_minus (TREE_TYPE (chrec
), op0
, op1
);
2038 op0
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 0),
2039 allow_superloop_chrecs
);
2040 op1
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 1),
2041 allow_superloop_chrecs
);
2042 return chrec_fold_multiply (TREE_TYPE (chrec
), op0
, op1
);
2046 case NON_LVALUE_EXPR
:
2047 op0
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 0),
2048 allow_superloop_chrecs
);
2049 if (op0
== chrec_dont_know
)
2050 return chrec_dont_know
;
2052 return chrec_convert (TREE_TYPE (chrec
), op0
);
2054 case SCEV_NOT_KNOWN
:
2055 return chrec_dont_know
;
2064 switch (TREE_CODE_LENGTH (TREE_CODE (chrec
)))
2067 op0
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 0),
2068 allow_superloop_chrecs
);
2069 op1
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 1),
2070 allow_superloop_chrecs
);
2071 op2
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 2),
2072 allow_superloop_chrecs
);
2073 if (op0
== chrec_dont_know
2074 || op1
== chrec_dont_know
2075 || op2
== chrec_dont_know
)
2076 return chrec_dont_know
;
2077 return fold (build (TREE_CODE (chrec
),
2078 TREE_TYPE (chrec
), op0
, op1
, op2
));
2081 op0
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 0),
2082 allow_superloop_chrecs
);
2083 op1
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 1),
2084 allow_superloop_chrecs
);
2085 if (op0
== chrec_dont_know
2086 || op1
== chrec_dont_know
)
2087 return chrec_dont_know
;
2088 return fold (build (TREE_CODE (chrec
), TREE_TYPE (chrec
), op0
, op1
));
2091 op0
= instantiate_parameters_1 (loop
, TREE_OPERAND (chrec
, 0),
2092 allow_superloop_chrecs
);
2093 if (op0
== chrec_dont_know
)
2094 return chrec_dont_know
;
2095 return fold (build1 (TREE_CODE (chrec
), TREE_TYPE (chrec
), op0
));
2104 /* Too complicated to handle. */
2105 return chrec_dont_know
;
2108 /* Analyze all the parameters of the chrec that were left under a
2109 symbolic form. LOOP is the loop in which symbolic names have to
2110 be analyzed and instantiated. */
2113 instantiate_parameters (struct loop
*loop
,
2118 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2120 fprintf (dump_file
, "(instantiate_parameters \n");
2121 fprintf (dump_file
, " (loop_nb = %d)\n", loop
->num
);
2122 fprintf (dump_file
, " (chrec = ");
2123 print_generic_expr (dump_file
, chrec
, 0);
2124 fprintf (dump_file
, ")\n");
2127 res
= instantiate_parameters_1 (loop
, chrec
, true);
2129 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2131 fprintf (dump_file
, " (res = ");
2132 print_generic_expr (dump_file
, res
, 0);
2133 fprintf (dump_file
, "))\n");
2139 /* Similar to instantiate_parameters, but does not introduce the
2140 evolutions in outer loops for LOOP invariants in CHREC. */
2143 resolve_mixers (struct loop
*loop
, tree chrec
)
2145 return instantiate_parameters_1 (loop
, chrec
, false);
2148 /* Entry point for the analysis of the number of iterations pass.
2149 This function tries to safely approximate the number of iterations
2150 the loop will run. When this property is not decidable at compile
2151 time, the result is chrec_dont_know. Otherwise the result is
2152 a scalar or a symbolic parameter.
2154 Example of analysis: suppose that the loop has an exit condition:
2156 "if (b > 49) goto end_loop;"
2158 and that in a previous analysis we have determined that the
2159 variable 'b' has an evolution function:
2161 "EF = {23, +, 5}_2".
2163 When we evaluate the function at the point 5, i.e. the value of the
2164 variable 'b' after 5 iterations in the loop, we have EF (5) = 48,
2165 and EF (6) = 53. In this case the value of 'b' on exit is '53' and
2166 the loop body has been executed 6 times. */
2169 number_of_iterations_in_loop (struct loop
*loop
)
2173 struct tree_niter_desc niter_desc
;
2175 /* Determine whether the number_of_iterations_in_loop has already
2177 res
= loop
->nb_iterations
;
2180 res
= chrec_dont_know
;
2182 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2183 fprintf (dump_file
, "(number_of_iterations_in_loop\n");
2185 exit
= loop
->single_exit
;
2189 if (!number_of_iterations_exit (loop
, exit
, &niter_desc
))
2192 type
= TREE_TYPE (niter_desc
.niter
);
2193 if (integer_nonzerop (niter_desc
.may_be_zero
))
2194 res
= build_int_cst (type
, 0);
2195 else if (integer_zerop (niter_desc
.may_be_zero
))
2196 res
= niter_desc
.niter
;
2198 res
= chrec_dont_know
;
2201 return set_nb_iterations_in_loop (loop
, res
);
2204 /* One of the drivers for testing the scalar evolutions analysis.
2205 This function computes the number of iterations for all the loops
2206 from the EXIT_CONDITIONS array. */
2209 number_of_iterations_for_all_loops (varray_type exit_conditions
)
2212 unsigned nb_chrec_dont_know_loops
= 0;
2213 unsigned nb_static_loops
= 0;
2215 for (i
= 0; i
< VARRAY_ACTIVE_SIZE (exit_conditions
); i
++)
2217 tree res
= number_of_iterations_in_loop
2218 (loop_containing_stmt (VARRAY_TREE (exit_conditions
, i
)));
2219 if (chrec_contains_undetermined (res
))
2220 nb_chrec_dont_know_loops
++;
2227 fprintf (dump_file
, "\n(\n");
2228 fprintf (dump_file
, "-----------------------------------------\n");
2229 fprintf (dump_file
, "%d\tnb_chrec_dont_know_loops\n", nb_chrec_dont_know_loops
);
2230 fprintf (dump_file
, "%d\tnb_static_loops\n", nb_static_loops
);
2231 fprintf (dump_file
, "%d\tnb_total_loops\n", current_loops
->num
);
2232 fprintf (dump_file
, "-----------------------------------------\n");
2233 fprintf (dump_file
, ")\n\n");
2235 print_loop_ir (dump_file
);
2241 /* Counters for the stats. */
2247 unsigned nb_affine_multivar
;
2248 unsigned nb_higher_poly
;
2249 unsigned nb_chrec_dont_know
;
2250 unsigned nb_undetermined
;
2253 /* Reset the counters. */
2256 reset_chrecs_counters (struct chrec_stats
*stats
)
2258 stats
->nb_chrecs
= 0;
2259 stats
->nb_affine
= 0;
2260 stats
->nb_affine_multivar
= 0;
2261 stats
->nb_higher_poly
= 0;
2262 stats
->nb_chrec_dont_know
= 0;
2263 stats
->nb_undetermined
= 0;
2266 /* Dump the contents of a CHREC_STATS structure. */
2269 dump_chrecs_stats (FILE *file
, struct chrec_stats
*stats
)
2271 fprintf (file
, "\n(\n");
2272 fprintf (file
, "-----------------------------------------\n");
2273 fprintf (file
, "%d\taffine univariate chrecs\n", stats
->nb_affine
);
2274 fprintf (file
, "%d\taffine multivariate chrecs\n", stats
->nb_affine_multivar
);
2275 fprintf (file
, "%d\tdegree greater than 2 polynomials\n",
2276 stats
->nb_higher_poly
);
2277 fprintf (file
, "%d\tchrec_dont_know chrecs\n", stats
->nb_chrec_dont_know
);
2278 fprintf (file
, "-----------------------------------------\n");
2279 fprintf (file
, "%d\ttotal chrecs\n", stats
->nb_chrecs
);
2280 fprintf (file
, "%d\twith undetermined coefficients\n",
2281 stats
->nb_undetermined
);
2282 fprintf (file
, "-----------------------------------------\n");
2283 fprintf (file
, "%d\tchrecs in the scev database\n",
2284 (int) htab_elements (scalar_evolution_info
));
2285 fprintf (file
, "%d\tsets in the scev database\n", nb_set_scev
);
2286 fprintf (file
, "%d\tgets in the scev database\n", nb_get_scev
);
2287 fprintf (file
, "-----------------------------------------\n");
2288 fprintf (file
, ")\n\n");
2291 /* Gather statistics about CHREC. */
2294 gather_chrec_stats (tree chrec
, struct chrec_stats
*stats
)
2296 if (dump_file
&& (dump_flags
& TDF_STATS
))
2298 fprintf (dump_file
, "(classify_chrec ");
2299 print_generic_expr (dump_file
, chrec
, 0);
2300 fprintf (dump_file
, "\n");
2305 if (chrec
== NULL_TREE
)
2307 stats
->nb_undetermined
++;
2311 switch (TREE_CODE (chrec
))
2313 case POLYNOMIAL_CHREC
:
2314 if (evolution_function_is_affine_p (chrec
))
2316 if (dump_file
&& (dump_flags
& TDF_STATS
))
2317 fprintf (dump_file
, " affine_univariate\n");
2320 else if (evolution_function_is_affine_multivariate_p (chrec
))
2322 if (dump_file
&& (dump_flags
& TDF_STATS
))
2323 fprintf (dump_file
, " affine_multivariate\n");
2324 stats
->nb_affine_multivar
++;
2328 if (dump_file
&& (dump_flags
& TDF_STATS
))
2329 fprintf (dump_file
, " higher_degree_polynomial\n");
2330 stats
->nb_higher_poly
++;
2339 if (chrec_contains_undetermined (chrec
))
2341 if (dump_file
&& (dump_flags
& TDF_STATS
))
2342 fprintf (dump_file
, " undetermined\n");
2343 stats
->nb_undetermined
++;
2346 if (dump_file
&& (dump_flags
& TDF_STATS
))
2347 fprintf (dump_file
, ")\n");
2350 /* One of the drivers for testing the scalar evolutions analysis.
2351 This function analyzes the scalar evolution of all the scalars
2352 defined as loop phi nodes in one of the loops from the
2353 EXIT_CONDITIONS array.
2355 TODO Optimization: A loop is in canonical form if it contains only
2356 a single scalar loop phi node. All the other scalars that have an
2357 evolution in the loop are rewritten in function of this single
2358 index. This allows the parallelization of the loop. */
2361 analyze_scalar_evolution_for_all_loop_phi_nodes (varray_type exit_conditions
)
2364 struct chrec_stats stats
;
2366 reset_chrecs_counters (&stats
);
2368 for (i
= 0; i
< VARRAY_ACTIVE_SIZE (exit_conditions
); i
++)
2374 loop
= loop_containing_stmt (VARRAY_TREE (exit_conditions
, i
));
2377 for (phi
= phi_nodes (bb
); phi
; phi
= TREE_CHAIN (phi
))
2378 if (is_gimple_reg (PHI_RESULT (phi
)))
2380 chrec
= instantiate_parameters
2382 analyze_scalar_evolution (loop
, PHI_RESULT (phi
)));
2384 if (dump_file
&& (dump_flags
& TDF_STATS
))
2385 gather_chrec_stats (chrec
, &stats
);
2389 if (dump_file
&& (dump_flags
& TDF_STATS
))
2390 dump_chrecs_stats (dump_file
, &stats
);
2393 /* Callback for htab_traverse, gathers information on chrecs in the
2397 gather_stats_on_scev_database_1 (void **slot
, void *stats
)
2399 struct scev_info_str
*entry
= *slot
;
2401 gather_chrec_stats (entry
->chrec
, stats
);
2406 /* Classify the chrecs of the whole database. */
2409 gather_stats_on_scev_database (void)
2411 struct chrec_stats stats
;
2416 reset_chrecs_counters (&stats
);
2418 htab_traverse (scalar_evolution_info
, gather_stats_on_scev_database_1
,
2421 dump_chrecs_stats (dump_file
, &stats
);
2429 initialize_scalar_evolutions_analyzer (void)
2431 /* The elements below are unique. */
2432 if (chrec_dont_know
== NULL_TREE
)
2434 chrec_not_analyzed_yet
= NULL_TREE
;
2435 chrec_dont_know
= make_node (SCEV_NOT_KNOWN
);
2436 chrec_known
= make_node (SCEV_KNOWN
);
2437 TREE_TYPE (chrec_dont_know
) = NULL_TREE
;
2438 TREE_TYPE (chrec_known
) = NULL_TREE
;
2442 /* Initialize the analysis of scalar evolutions for LOOPS. */
2445 scev_initialize (struct loops
*loops
)
2448 current_loops
= loops
;
2450 scalar_evolution_info
= htab_create (100, hash_scev_info
,
2451 eq_scev_info
, del_scev_info
);
2452 already_instantiated
= BITMAP_XMALLOC ();
2454 initialize_scalar_evolutions_analyzer ();
2456 for (i
= 1; i
< loops
->num
; i
++)
2457 if (loops
->parray
[i
])
2458 loops
->parray
[i
]->nb_iterations
= NULL_TREE
;
2461 /* Cleans up the information cached by the scalar evolutions analysis. */
2469 if (!scalar_evolution_info
|| !current_loops
)
2472 htab_empty (scalar_evolution_info
);
2473 for (i
= 1; i
< current_loops
->num
; i
++)
2475 loop
= current_loops
->parray
[i
];
2477 loop
->nb_iterations
= NULL_TREE
;
2481 /* Checks whether OP behaves as a simple affine iv of LOOP in STMT and returns
2482 its BASE and STEP if possible. */
2485 simple_iv (struct loop
*loop
, tree stmt
, tree op
, tree
*base
, tree
*step
)
2487 basic_block bb
= bb_for_stmt (stmt
);
2493 type
= TREE_TYPE (op
);
2494 if (TREE_CODE (type
) != INTEGER_TYPE
2495 && TREE_CODE (type
) != POINTER_TYPE
)
2498 ev
= analyze_scalar_evolution_in_loop (loop
, bb
->loop_father
, op
);
2499 if (chrec_contains_undetermined (ev
))
2502 if (tree_does_not_contain_chrecs (ev
)
2503 && !chrec_contains_symbols_defined_in_loop (ev
, loop
->num
))
2509 if (TREE_CODE (ev
) != POLYNOMIAL_CHREC
2510 || CHREC_VARIABLE (ev
) != (unsigned) loop
->num
)
2513 *step
= CHREC_RIGHT (ev
);
2514 if (TREE_CODE (*step
) != INTEGER_CST
)
2516 *base
= CHREC_LEFT (ev
);
2517 if (tree_contains_chrecs (*base
)
2518 || chrec_contains_symbols_defined_in_loop (*base
, loop
->num
))
2524 /* Runs the analysis of scalar evolutions. */
2527 scev_analysis (void)
2529 varray_type exit_conditions
;
2531 VARRAY_GENERIC_PTR_INIT (exit_conditions
, 37, "exit_conditions");
2532 select_loops_exit_conditions (current_loops
, &exit_conditions
);
2534 if (dump_file
&& (dump_flags
& TDF_STATS
))
2535 analyze_scalar_evolution_for_all_loop_phi_nodes (exit_conditions
);
2537 number_of_iterations_for_all_loops (exit_conditions
);
2538 VARRAY_CLEAR (exit_conditions
);
2541 /* Finalize the scalar evolution analysis. */
2544 scev_finalize (void)
2546 htab_delete (scalar_evolution_info
);
2547 BITMAP_XFREE (already_instantiated
);