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
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
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/>. */
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 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.
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 (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:
120 or in terms of a C program:
123 | for (x = 0; x <= 7; x++)
129 Example 2a: 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 2b: Multivariate chains of recurrences.
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.
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.
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.
231 Based on these symbolic chrecs, it is possible to refine this
232 information into the more precise PERIODIC_CHRECs:
237 This transformation is not yet implemented.
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
259 #include "coretypes.h"
265 /* These RTL headers are needed for basic-block.h. */
267 #include "basic-block.h"
268 #include "diagnostic.h"
269 #include "tree-flow.h"
270 #include "tree-dump.h"
273 #include "tree-chrec.h"
274 #include "tree-scalar-evolution.h"
275 #include "tree-pass.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
285 struct GTY(()) scev_info_str
{
286 basic_block instantiated_below
;
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. */
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
);
322 res
->chrec
= chrec_not_analyzed_yet
;
323 res
->instantiated_below
= instantiated_below
;
328 /* Computes a hash function for database element ELT. */
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. */
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. */
351 del_scev_info (void *e
)
356 /* Get the scalar evolution of VAR for INSTANTIATED_BELOW basic block.
357 A first query on VAR returns chrec_not_analyzed_yet. */
360 find_var_scev_info (basic_block instantiated_below
, tree var
)
362 struct scev_info_str
*res
;
363 struct scev_info_str tmp
;
367 tmp
.instantiated_below
= instantiated_below
;
368 slot
= htab_find_slot (scalar_evolution_info
, &tmp
, INSERT
);
371 *slot
= new_scev_info_str (instantiated_below
, var
);
372 res
= (struct scev_info_str
*) *slot
;
377 /* Return true when CHREC contains symbolic names defined in
381 chrec_contains_symbols_defined_in_loop (const_tree chrec
, unsigned loop_nb
)
385 if (chrec
== NULL_TREE
)
388 if (is_gimple_min_invariant (chrec
))
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
)
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
)
408 if (loop
== def_loop
|| flow_loop_nested_p (loop
, def_loop
))
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
),
422 /* Return true when PHI is a loop-phi-node. */
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.
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.
455 | i_1 = phi (i_0, i_2)
459 This loop has the same effect as:
460 LOOP_1 has the same effect as:
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.
470 compute_overall_effect_of_inner_loop (struct loop
*loop
, tree evolution_fn
)
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
;
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
);
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
)
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
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
))
532 /* FIXME -- overflows. */
533 if (value0
== value1
)
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
))
546 nb_iter
= number_of_latch_executions (get_chrec_loop (chrec
));
547 if (chrec_contains_undetermined (nb_iter
))
551 /* TODO -- If the test is after the exit, we may decrease the number of
552 iterations by one. */
554 nb_iter
= chrec_fold_minus (type
, nb_iter
, build_int_cst (type
, 1));
557 end_value
= chrec_apply (CHREC_VARIABLE (chrec
), chrec
, nb_iter
);
559 if (!chrec_is_positive (end_value
, &value2
))
563 return value0
== value1
;
566 *value
= (tree_int_cst_sgn (chrec
) == 1);
574 /* Associate CHREC to SCALAR. */
577 set_scalar_evolution (basic_block instantiated_below
, tree scalar
, tree chrec
)
581 if (TREE_CODE (scalar
) != SSA_NAME
)
584 scalar_info
= find_var_scev_info (instantiated_below
, scalar
);
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
)
603 *scalar_info
= chrec
;
606 /* Retrieve the chrec associated to SCALAR instantiated below
607 INSTANTIATED_BELOW block. */
610 get_scalar_evolution (basic_block instantiated_below
, tree scalar
)
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
)
627 switch (TREE_CODE (scalar
))
630 res
= *find_var_scev_info (instantiated_below
, scalar
);
640 res
= chrec_not_analyzed_yet
;
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");
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. */
665 add_to_evolution_1 (unsigned loop_nb
, tree chrec_before
, tree to_add
,
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
);
676 || flow_loop_nested_p (chloop
, loop
))
680 type
= chrec_type (chrec_before
);
682 /* When there is no evolution part in this loop, build it. */
687 right
= SCALAR_FLOAT_TYPE_P (type
)
688 ? build_real (type
, dconst0
)
689 : build_int_cst (type
, 0);
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
);
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
),
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
),
717 /* These nodes do not depend on a loop. */
718 if (chrec_before
== chrec_dont_know
)
719 return chrec_dont_know
;
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
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).
751 | a = phi (init, a + 5)
758 | a = phi (inita, 2 * b + 3)
759 | b = phi (initb, b + 1)
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.
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
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.
847 | a = phi (inita, a + 2 + b)
848 | b = phi (initb, b + 1)
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
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
)
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");
905 /* Helper function. */
908 set_nb_iterations_in_loop (struct loop
*loop
,
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
;
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. */
933 get_loop_exit_condition (const struct loop
*loop
)
936 edge exit_edge
= single_exit (loop
);
938 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
939 fprintf (dump_file
, "(get_loop_exit_condition \n ");
945 stmt
= last_stmt (exit_edge
->src
);
946 if (gimple_code (stmt
) == GIMPLE_COND
)
950 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
952 print_gimple_stmt (dump_file
, res
, 0, 0);
953 fprintf (dump_file
, ")\n");
959 /* Recursively determine and enqueue the exit conditions for a loop. */
962 get_exit_conditions_rec (struct loop
*loop
,
963 VEC(gimple
,heap
) **exit_conditions
)
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
);
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. */
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
{
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. */
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
;
1017 case POINTER_PLUS_EXPR
:
1019 if (TREE_CODE (rhs0
) == SSA_NAME
)
1021 if (TREE_CODE (rhs1
) == SSA_NAME
)
1023 /* Match an assignment under the form:
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. */
1031 evol
= *evolution_of_loop
;
1032 res
= follow_ssa_edge
1033 (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
, &evol
, limit
);
1036 *evolution_of_loop
= add_to_evolution
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
);
1048 *evolution_of_loop
= add_to_evolution
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
;
1063 /* Match an assignment under the form:
1065 res
= follow_ssa_edge
1066 (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
,
1067 evolution_of_loop
, limit
);
1069 *evolution_of_loop
= add_to_evolution
1070 (loop
->num
, chrec_convert (type
, *evolution_of_loop
,
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:
1083 res
= follow_ssa_edge
1084 (loop
, SSA_NAME_DEF_STMT (rhs1
), halting_phi
,
1085 evolution_of_loop
, limit
);
1087 *evolution_of_loop
= add_to_evolution
1088 (loop
->num
, chrec_convert (type
, *evolution_of_loop
,
1090 code
, rhs0
, at_stmt
);
1092 else if (res
== t_dont_know
)
1093 *evolution_of_loop
= chrec_dont_know
;
1097 /* Otherwise, match an assignment under the form:
1099 /* And there is nothing to do. */
1104 /* This case is under the form "opnd0 = rhs0 - rhs1". */
1105 if (TREE_CODE (rhs0
) == SSA_NAME
)
1107 /* Match an assignment under the form:
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
)
1116 res
= follow_ssa_edge (loop
, SSA_NAME_DEF_STMT (rhs0
), halting_phi
,
1117 evolution_of_loop
, limit
);
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
;
1127 /* Otherwise, match an assignment under the form:
1129 /* And there is nothing to do. */
1140 /* Follow the ssa edge into the expression EXPR.
1141 Return true if the strongly connected component has been found. */
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
;
1151 /* The EXPR is one of the following cases:
1155 - a POINTER_PLUS_EXPR,
1158 - other cases are not yet handled. */
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
);
1170 /* This assignment is under the form "a_1 = 7". */
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
);
1180 case POINTER_PLUS_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
);
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
);
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. */
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
;
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
);
1232 case POINTER_PLUS_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
);
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
);
1254 /* Checks whether the I-th argument of a PHI comes from a backedge. */
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
1264 if (e
->flags
& EDGE_IRREDUCIBLE_LOOP
)
1270 /* Helper function for one branch of the condition-phi-node. Return
1271 true if the strongly connected component has been found following
1274 static inline t_bool
1275 follow_ssa_edge_in_condition_phi_branch (int i
,
1277 gimple condition_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
))
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. */
1307 /* This function merges the branches of a condition-phi-node in a
1311 follow_ssa_edge_in_condition_phi (struct loop
*loop
,
1312 gimple condition_phi
,
1314 tree
*evolution_of_loop
, int limit
)
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
,
1321 &evolution_of_branch
,
1323 if (res
== t_false
|| res
== t_dont_know
)
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
1333 if (*evolution_of_loop
== chrec_dont_know
)
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
,
1340 &evolution_of_branch
,
1342 if (res
== t_false
|| res
== t_dont_know
)
1345 *evolution_of_loop
= chrec_merge (*evolution_of_loop
,
1346 evolution_of_branch
);
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. */
1358 follow_ssa_edge_inner_loop_phi (struct loop
*outer_loop
,
1359 gimple loop_phi_node
,
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
);
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
,
1383 evolution_of_loop
, limit
);
1388 /* If the path crosses this loop-phi, give up. */
1390 *evolution_of_loop
= chrec_dont_know
;
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. */
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
))
1413 /* Give up if the path is longer than the MAX that we allow. */
1414 if (limit
> PARAM_VALUE (PARAM_SCEV_MAX_EXPR_SIZE
))
1417 def_loop
= loop_containing_stmt (def
);
1419 switch (gimple_code (def
))
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
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
)
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
)
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);
1451 return follow_ssa_edge_in_rhs (loop
, def
, halting_phi
,
1452 evolution_of_loop
, limit
);
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. */
1464 /* Given a LOOP_PHI_NODE, this function determines the evolution
1465 function from LOOP_PHI_NODE to LOOP_PHI_NODE in the loop. */
1468 analyze_evolution_in_loop (gimple loop_phi_node
,
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
);
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
);
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
))
1496 if (TREE_CODE (arg
) == SSA_NAME
)
1498 ssa_chain
= SSA_NAME_DEF_STMT (arg
);
1500 /* Pass in the initial condition to the follow edge function. */
1502 res
= follow_ssa_edge (loop
, ssa_chain
, loop_phi_node
, &ev_fn
, 0);
1507 /* When it is impossible to go back on the same
1508 loop_phi_node by following the ssa edges, the
1509 evolution is represented by a peeled chrec, i.e. the
1510 first iteration, EV_FN has the value INIT_COND, then
1511 all the other iterations it has the value of ARG.
1512 For the moment, PEELED_CHREC nodes are not built. */
1514 ev_fn
= chrec_dont_know
;
1516 /* When there are multiple back edges of the loop (which in fact never
1517 happens currently, but nevertheless), merge their evolutions. */
1518 evolution_function
= chrec_merge (evolution_function
, ev_fn
);
1521 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1523 fprintf (dump_file
, " (evolution_function = ");
1524 print_generic_expr (dump_file
, evolution_function
, 0);
1525 fprintf (dump_file
, "))\n");
1528 return evolution_function
;
1531 /* Given a loop-phi-node, return the initial conditions of the
1532 variable on entry of the loop. When the CCP has propagated
1533 constants into the loop-phi-node, the initial condition is
1534 instantiated, otherwise the initial condition is kept symbolic.
1535 This analyzer does not analyze the evolution outside the current
1536 loop, and leaves this task to the on-demand tree reconstructor. */
1539 analyze_initial_condition (gimple loop_phi_node
)
1542 tree init_cond
= chrec_not_analyzed_yet
;
1543 struct loop
*loop
= loop_containing_stmt (loop_phi_node
);
1545 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1547 fprintf (dump_file
, "(analyze_initial_condition \n");
1548 fprintf (dump_file
, " (loop_phi_node = \n");
1549 print_gimple_stmt (dump_file
, loop_phi_node
, 0, 0);
1550 fprintf (dump_file
, ")\n");
1553 n
= gimple_phi_num_args (loop_phi_node
);
1554 for (i
= 0; i
< n
; i
++)
1556 tree branch
= PHI_ARG_DEF (loop_phi_node
, i
);
1557 basic_block bb
= gimple_phi_arg_edge (loop_phi_node
, i
)->src
;
1559 /* When the branch is oriented to the loop's body, it does
1560 not contribute to the initial condition. */
1561 if (flow_bb_inside_loop_p (loop
, bb
))
1564 if (init_cond
== chrec_not_analyzed_yet
)
1570 if (TREE_CODE (branch
) == SSA_NAME
)
1572 init_cond
= chrec_dont_know
;
1576 init_cond
= chrec_merge (init_cond
, branch
);
1579 /* Ooops -- a loop without an entry??? */
1580 if (init_cond
== chrec_not_analyzed_yet
)
1581 init_cond
= chrec_dont_know
;
1583 /* During early loop unrolling we do not have fully constant propagated IL.
1584 Handle degenerate PHIs here to not miss important unrollings. */
1585 if (TREE_CODE (init_cond
) == SSA_NAME
)
1587 gimple def
= SSA_NAME_DEF_STMT (init_cond
);
1589 if (gimple_code (def
) == GIMPLE_PHI
1590 && (res
= degenerate_phi_result (def
)) != NULL_TREE
1591 /* Only allow invariants here, otherwise we may break
1592 loop-closed SSA form. */
1593 && is_gimple_min_invariant (res
))
1597 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1599 fprintf (dump_file
, " (init_cond = ");
1600 print_generic_expr (dump_file
, init_cond
, 0);
1601 fprintf (dump_file
, "))\n");
1607 /* Analyze the scalar evolution for LOOP_PHI_NODE. */
1610 interpret_loop_phi (struct loop
*loop
, gimple loop_phi_node
)
1613 struct loop
*phi_loop
= loop_containing_stmt (loop_phi_node
);
1616 if (phi_loop
!= loop
)
1618 struct loop
*subloop
;
1619 tree evolution_fn
= analyze_scalar_evolution
1620 (phi_loop
, PHI_RESULT (loop_phi_node
));
1622 /* Dive one level deeper. */
1623 subloop
= superloop_at_depth (phi_loop
, loop_depth (loop
) + 1);
1625 /* Interpret the subloop. */
1626 res
= compute_overall_effect_of_inner_loop (subloop
, evolution_fn
);
1630 /* Otherwise really interpret the loop phi. */
1631 init_cond
= analyze_initial_condition (loop_phi_node
);
1632 res
= analyze_evolution_in_loop (loop_phi_node
, init_cond
);
1637 /* This function merges the branches of a condition-phi-node,
1638 contained in the outermost loop, and whose arguments are already
1642 interpret_condition_phi (struct loop
*loop
, gimple condition_phi
)
1644 int i
, n
= gimple_phi_num_args (condition_phi
);
1645 tree res
= chrec_not_analyzed_yet
;
1647 for (i
= 0; i
< n
; i
++)
1651 if (backedge_phi_arg_p (condition_phi
, i
))
1653 res
= chrec_dont_know
;
1657 branch_chrec
= analyze_scalar_evolution
1658 (loop
, PHI_ARG_DEF (condition_phi
, i
));
1660 res
= chrec_merge (res
, branch_chrec
);
1666 /* Interpret the operation RHS1 OP RHS2. If we didn't
1667 analyze this node before, follow the definitions until ending
1668 either on an analyzed GIMPLE_ASSIGN, or on a loop-phi-node. On the
1669 return path, this function propagates evolutions (ala constant copy
1670 propagation). OPND1 is not a GIMPLE expression because we could
1671 analyze the effect of an inner loop: see interpret_loop_phi. */
1674 interpret_rhs_expr (struct loop
*loop
, gimple at_stmt
,
1675 tree type
, tree rhs1
, enum tree_code code
, tree rhs2
)
1677 tree res
, chrec1
, chrec2
;
1679 if (get_gimple_rhs_class (code
) == GIMPLE_SINGLE_RHS
)
1681 if (is_gimple_min_invariant (rhs1
))
1682 return chrec_convert (type
, rhs1
, at_stmt
);
1684 if (code
== SSA_NAME
)
1685 return chrec_convert (type
, analyze_scalar_evolution (loop
, rhs1
),
1688 if (code
== ASSERT_EXPR
)
1690 rhs1
= ASSERT_EXPR_VAR (rhs1
);
1691 return chrec_convert (type
, analyze_scalar_evolution (loop
, rhs1
),
1695 return chrec_dont_know
;
1700 case POINTER_PLUS_EXPR
:
1701 chrec1
= analyze_scalar_evolution (loop
, rhs1
);
1702 chrec2
= analyze_scalar_evolution (loop
, rhs2
);
1703 chrec1
= chrec_convert (type
, chrec1
, at_stmt
);
1704 chrec2
= chrec_convert (sizetype
, chrec2
, at_stmt
);
1705 res
= chrec_fold_plus (type
, chrec1
, chrec2
);
1709 chrec1
= analyze_scalar_evolution (loop
, rhs1
);
1710 chrec2
= analyze_scalar_evolution (loop
, rhs2
);
1711 chrec1
= chrec_convert (type
, chrec1
, at_stmt
);
1712 chrec2
= chrec_convert (type
, chrec2
, at_stmt
);
1713 res
= chrec_fold_plus (type
, chrec1
, chrec2
);
1717 chrec1
= analyze_scalar_evolution (loop
, rhs1
);
1718 chrec2
= analyze_scalar_evolution (loop
, rhs2
);
1719 chrec1
= chrec_convert (type
, chrec1
, at_stmt
);
1720 chrec2
= chrec_convert (type
, chrec2
, at_stmt
);
1721 res
= chrec_fold_minus (type
, chrec1
, chrec2
);
1725 chrec1
= analyze_scalar_evolution (loop
, rhs1
);
1726 chrec1
= chrec_convert (type
, chrec1
, at_stmt
);
1727 /* TYPE may be integer, real or complex, so use fold_convert. */
1728 res
= chrec_fold_multiply (type
, chrec1
,
1729 fold_convert (type
, integer_minus_one_node
));
1733 /* Handle ~X as -1 - X. */
1734 chrec1
= analyze_scalar_evolution (loop
, rhs1
);
1735 chrec1
= chrec_convert (type
, chrec1
, at_stmt
);
1736 res
= chrec_fold_minus (type
,
1737 fold_convert (type
, integer_minus_one_node
),
1742 chrec1
= analyze_scalar_evolution (loop
, rhs1
);
1743 chrec2
= analyze_scalar_evolution (loop
, rhs2
);
1744 chrec1
= chrec_convert (type
, chrec1
, at_stmt
);
1745 chrec2
= chrec_convert (type
, chrec2
, at_stmt
);
1746 res
= chrec_fold_multiply (type
, chrec1
, chrec2
);
1750 chrec1
= analyze_scalar_evolution (loop
, rhs1
);
1751 res
= chrec_convert (type
, chrec1
, at_stmt
);
1755 res
= chrec_dont_know
;
1762 /* Interpret the expression EXPR. */
1765 interpret_expr (struct loop
*loop
, gimple at_stmt
, tree expr
)
1767 enum tree_code code
;
1768 tree type
= TREE_TYPE (expr
), op0
, op1
;
1770 if (automatically_generated_chrec_p (expr
))
1773 if (TREE_CODE (expr
) == POLYNOMIAL_CHREC
)
1774 return chrec_dont_know
;
1776 extract_ops_from_tree (expr
, &code
, &op0
, &op1
);
1778 return interpret_rhs_expr (loop
, at_stmt
, type
,
1782 /* Interpret the rhs of the assignment STMT. */
1785 interpret_gimple_assign (struct loop
*loop
, gimple stmt
)
1787 tree type
= TREE_TYPE (gimple_assign_lhs (stmt
));
1788 enum tree_code code
= gimple_assign_rhs_code (stmt
);
1790 return interpret_rhs_expr (loop
, stmt
, type
,
1791 gimple_assign_rhs1 (stmt
), code
,
1792 gimple_assign_rhs2 (stmt
));
1797 /* This section contains all the entry points:
1798 - number_of_iterations_in_loop,
1799 - analyze_scalar_evolution,
1800 - instantiate_parameters.
1803 /* Compute and return the evolution function in WRTO_LOOP, the nearest
1804 common ancestor of DEF_LOOP and USE_LOOP. */
1807 compute_scalar_evolution_in_loop (struct loop
*wrto_loop
,
1808 struct loop
*def_loop
,
1812 if (def_loop
== wrto_loop
)
1815 def_loop
= superloop_at_depth (def_loop
, loop_depth (wrto_loop
) + 1);
1816 res
= compute_overall_effect_of_inner_loop (def_loop
, ev
);
1818 return analyze_scalar_evolution_1 (wrto_loop
, res
, chrec_not_analyzed_yet
);
1821 /* Helper recursive function. */
1824 analyze_scalar_evolution_1 (struct loop
*loop
, tree var
, tree res
)
1826 tree type
= TREE_TYPE (var
);
1829 struct loop
*def_loop
;
1831 if (loop
== NULL
|| TREE_CODE (type
) == VECTOR_TYPE
)
1832 return chrec_dont_know
;
1834 if (TREE_CODE (var
) != SSA_NAME
)
1835 return interpret_expr (loop
, NULL
, var
);
1837 def
= SSA_NAME_DEF_STMT (var
);
1838 bb
= gimple_bb (def
);
1839 def_loop
= bb
? bb
->loop_father
: NULL
;
1842 || !flow_bb_inside_loop_p (loop
, bb
))
1844 /* Keep the symbolic form. */
1849 if (res
!= chrec_not_analyzed_yet
)
1851 if (loop
!= bb
->loop_father
)
1852 res
= compute_scalar_evolution_in_loop
1853 (find_common_loop (loop
, bb
->loop_father
), bb
->loop_father
, res
);
1858 if (loop
!= def_loop
)
1860 res
= analyze_scalar_evolution_1 (def_loop
, var
, chrec_not_analyzed_yet
);
1861 res
= compute_scalar_evolution_in_loop (loop
, def_loop
, res
);
1866 switch (gimple_code (def
))
1869 res
= interpret_gimple_assign (loop
, def
);
1873 if (loop_phi_node_p (def
))
1874 res
= interpret_loop_phi (loop
, def
);
1876 res
= interpret_condition_phi (loop
, def
);
1880 res
= chrec_dont_know
;
1886 /* Keep the symbolic form. */
1887 if (res
== chrec_dont_know
)
1890 if (loop
== def_loop
)
1891 set_scalar_evolution (block_before_loop (loop
), var
, res
);
1896 /* Analyzes and returns the scalar evolution of the ssa_name VAR in
1897 LOOP. LOOP is the loop in which the variable is used.
1899 Example of use: having a pointer VAR to a SSA_NAME node, STMT a
1900 pointer to the statement that uses this variable, in order to
1901 determine the evolution function of the variable, use the following
1904 loop_p loop = loop_containing_stmt (stmt);
1905 tree chrec_with_symbols = analyze_scalar_evolution (loop, var);
1906 tree chrec_instantiated = instantiate_parameters (loop, chrec_with_symbols);
1910 analyze_scalar_evolution (struct loop
*loop
, tree var
)
1914 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1916 fprintf (dump_file
, "(analyze_scalar_evolution \n");
1917 fprintf (dump_file
, " (loop_nb = %d)\n", loop
->num
);
1918 fprintf (dump_file
, " (scalar = ");
1919 print_generic_expr (dump_file
, var
, 0);
1920 fprintf (dump_file
, ")\n");
1923 res
= get_scalar_evolution (block_before_loop (loop
), var
);
1924 res
= analyze_scalar_evolution_1 (loop
, var
, res
);
1926 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
1927 fprintf (dump_file
, ")\n");
1932 /* Analyze scalar evolution of use of VERSION in USE_LOOP with respect to
1933 WRTO_LOOP (which should be a superloop of USE_LOOP)
1935 FOLDED_CASTS is set to true if resolve_mixers used
1936 chrec_convert_aggressive (TODO -- not really, we are way too conservative
1937 at the moment in order to keep things simple).
1939 To illustrate the meaning of USE_LOOP and WRTO_LOOP, consider the following
1942 for (i = 0; i < 100; i++) -- loop 1
1944 for (j = 0; j < 100; j++) -- loop 2
1951 for (t = 0; t < 100; t++) -- loop 3
1958 Both k1 and k2 are invariants in loop3, thus
1959 analyze_scalar_evolution_in_loop (loop3, loop3, k1) = k1
1960 analyze_scalar_evolution_in_loop (loop3, loop3, k2) = k2
1962 As they are invariant, it does not matter whether we consider their
1963 usage in loop 3 or loop 2, hence
1964 analyze_scalar_evolution_in_loop (loop2, loop3, k1) =
1965 analyze_scalar_evolution_in_loop (loop2, loop2, k1) = i
1966 analyze_scalar_evolution_in_loop (loop2, loop3, k2) =
1967 analyze_scalar_evolution_in_loop (loop2, loop2, k2) = [0,+,1]_2
1969 Similarly for their evolutions with respect to loop 1. The values of K2
1970 in the use in loop 2 vary independently on loop 1, thus we cannot express
1971 the evolution with respect to loop 1:
1972 analyze_scalar_evolution_in_loop (loop1, loop3, k1) =
1973 analyze_scalar_evolution_in_loop (loop1, loop2, k1) = [0,+,1]_1
1974 analyze_scalar_evolution_in_loop (loop1, loop3, k2) =
1975 analyze_scalar_evolution_in_loop (loop1, loop2, k2) = dont_know
1977 The value of k2 in the use in loop 1 is known, though:
1978 analyze_scalar_evolution_in_loop (loop1, loop1, k1) = [0,+,1]_1
1979 analyze_scalar_evolution_in_loop (loop1, loop1, k2) = 100
1983 analyze_scalar_evolution_in_loop (struct loop
*wrto_loop
, struct loop
*use_loop
,
1984 tree version
, bool *folded_casts
)
1987 tree ev
= version
, tmp
;
1989 /* We cannot just do
1991 tmp = analyze_scalar_evolution (use_loop, version);
1992 ev = resolve_mixers (wrto_loop, tmp);
1994 as resolve_mixers would query the scalar evolution with respect to
1995 wrto_loop. For example, in the situation described in the function
1996 comment, suppose that wrto_loop = loop1, use_loop = loop3 and
1999 analyze_scalar_evolution (use_loop, version) = k2
2001 and resolve_mixers (loop1, k2) finds that the value of k2 in loop 1
2002 is 100, which is a wrong result, since we are interested in the
2005 Instead, we need to proceed from use_loop to wrto_loop loop by loop,
2006 each time checking that there is no evolution in the inner loop. */
2009 *folded_casts
= false;
2012 tmp
= analyze_scalar_evolution (use_loop
, ev
);
2013 ev
= resolve_mixers (use_loop
, tmp
);
2015 if (folded_casts
&& tmp
!= ev
)
2016 *folded_casts
= true;
2018 if (use_loop
== wrto_loop
)
2021 /* If the value of the use changes in the inner loop, we cannot express
2022 its value in the outer loop (we might try to return interval chrec,
2023 but we do not have a user for it anyway) */
2024 if (!no_evolution_in_loop_p (ev
, use_loop
->num
, &val
)
2026 return chrec_dont_know
;
2028 use_loop
= loop_outer (use_loop
);
2032 /* Returns from CACHE the value for VERSION instantiated below
2033 INSTANTIATED_BELOW block. */
2036 get_instantiated_value (htab_t cache
, basic_block instantiated_below
,
2039 struct scev_info_str
*info
, pattern
;
2041 pattern
.var
= version
;
2042 pattern
.instantiated_below
= instantiated_below
;
2043 info
= (struct scev_info_str
*) htab_find (cache
, &pattern
);
2051 /* Sets in CACHE the value of VERSION instantiated below basic block
2052 INSTANTIATED_BELOW to VAL. */
2055 set_instantiated_value (htab_t cache
, basic_block instantiated_below
,
2056 tree version
, tree val
)
2058 struct scev_info_str
*info
, pattern
;
2061 pattern
.var
= version
;
2062 pattern
.instantiated_below
= instantiated_below
;
2063 slot
= htab_find_slot (cache
, &pattern
, INSERT
);
2066 *slot
= new_scev_info_str (instantiated_below
, version
);
2067 info
= (struct scev_info_str
*) *slot
;
2071 /* Return the closed_loop_phi node for VAR. If there is none, return
2075 loop_closed_phi_def (tree var
)
2080 gimple_stmt_iterator psi
;
2082 if (var
== NULL_TREE
2083 || TREE_CODE (var
) != SSA_NAME
)
2086 loop
= loop_containing_stmt (SSA_NAME_DEF_STMT (var
));
2087 exit
= single_exit (loop
);
2091 for (psi
= gsi_start_phis (exit
->dest
); !gsi_end_p (psi
); gsi_next (&psi
))
2093 phi
= gsi_stmt (psi
);
2094 if (PHI_ARG_DEF_FROM_EDGE (phi
, exit
) == var
)
2095 return PHI_RESULT (phi
);
2101 /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW
2102 and EVOLUTION_LOOP, that were left under a symbolic form.
2104 CHREC is the scalar evolution to instantiate.
2106 CACHE is the cache of already instantiated values.
2108 FOLD_CONVERSIONS should be set to true when the conversions that
2109 may wrap in signed/pointer type are folded, as long as the value of
2110 the chrec is preserved.
2112 SIZE_EXPR is used for computing the size of the expression to be
2113 instantiated, and to stop if it exceeds some limit. */
2116 instantiate_scev_1 (basic_block instantiate_below
,
2117 struct loop
*evolution_loop
, tree chrec
,
2118 bool fold_conversions
, htab_t cache
, int size_expr
)
2120 tree res
, op0
, op1
, op2
;
2122 struct loop
*def_loop
;
2123 tree type
= chrec_type (chrec
);
2125 /* Give up if the expression is larger than the MAX that we allow. */
2126 if (size_expr
++ > PARAM_VALUE (PARAM_SCEV_MAX_EXPR_SIZE
))
2127 return chrec_dont_know
;
2129 if (automatically_generated_chrec_p (chrec
)
2130 || is_gimple_min_invariant (chrec
))
2133 switch (TREE_CODE (chrec
))
2136 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. */
2141 || loop_depth (def_bb
->loop_father
) == 0
2142 || dominated_by_p (CDI_DOMINATORS
, instantiate_below
, def_bb
))
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
);
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
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
))))
2174 res
= loop_closed_phi_def (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_1 (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
);
2192 case POLYNOMIAL_CHREC
:
2193 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2194 CHREC_LEFT (chrec
), fold_conversions
, cache
,
2196 if (op0
== chrec_dont_know
)
2197 return chrec_dont_know
;
2199 op1
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2200 CHREC_RIGHT (chrec
), fold_conversions
, cache
,
2202 if (op1
== chrec_dont_know
)
2203 return chrec_dont_know
;
2205 if (CHREC_LEFT (chrec
) != op0
2206 || CHREC_RIGHT (chrec
) != op1
)
2208 op1
= chrec_convert_rhs (chrec_type (op0
), op1
, NULL
);
2209 chrec
= build_polynomial_chrec (CHREC_VARIABLE (chrec
), op0
, op1
);
2213 case POINTER_PLUS_EXPR
:
2215 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2216 TREE_OPERAND (chrec
, 0), fold_conversions
, cache
,
2218 if (op0
== chrec_dont_know
)
2219 return chrec_dont_know
;
2221 op1
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2222 TREE_OPERAND (chrec
, 1), fold_conversions
, cache
,
2224 if (op1
== chrec_dont_know
)
2225 return chrec_dont_know
;
2227 if (TREE_OPERAND (chrec
, 0) != op0
2228 || TREE_OPERAND (chrec
, 1) != op1
)
2230 op0
= chrec_convert (type
, op0
, NULL
);
2231 op1
= chrec_convert_rhs (type
, op1
, NULL
);
2232 chrec
= chrec_fold_plus (type
, op0
, op1
);
2237 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2238 TREE_OPERAND (chrec
, 0), fold_conversions
, cache
,
2240 if (op0
== chrec_dont_know
)
2241 return chrec_dont_know
;
2243 op1
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2244 TREE_OPERAND (chrec
, 1),
2245 fold_conversions
, cache
, size_expr
);
2246 if (op1
== chrec_dont_know
)
2247 return chrec_dont_know
;
2249 if (TREE_OPERAND (chrec
, 0) != op0
2250 || TREE_OPERAND (chrec
, 1) != op1
)
2252 op0
= chrec_convert (type
, op0
, NULL
);
2253 op1
= chrec_convert (type
, op1
, NULL
);
2254 chrec
= chrec_fold_minus (type
, op0
, op1
);
2259 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2260 TREE_OPERAND (chrec
, 0),
2261 fold_conversions
, cache
, size_expr
);
2262 if (op0
== chrec_dont_know
)
2263 return chrec_dont_know
;
2265 op1
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2266 TREE_OPERAND (chrec
, 1),
2267 fold_conversions
, cache
, size_expr
);
2268 if (op1
== chrec_dont_know
)
2269 return chrec_dont_know
;
2271 if (TREE_OPERAND (chrec
, 0) != op0
2272 || TREE_OPERAND (chrec
, 1) != op1
)
2274 op0
= chrec_convert (type
, op0
, NULL
);
2275 op1
= chrec_convert (type
, op1
, NULL
);
2276 chrec
= chrec_fold_multiply (type
, op0
, op1
);
2281 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2282 TREE_OPERAND (chrec
, 0),
2283 fold_conversions
, cache
, size_expr
);
2284 if (op0
== chrec_dont_know
)
2285 return chrec_dont_know
;
2287 if (fold_conversions
)
2289 tree tmp
= chrec_convert_aggressive (TREE_TYPE (chrec
), op0
);
2294 if (op0
== TREE_OPERAND (chrec
, 0))
2297 /* If we used chrec_convert_aggressive, we can no longer assume that
2298 signed chrecs do not overflow, as chrec_convert does, so avoid
2299 calling it in that case. */
2300 if (fold_conversions
)
2301 return fold_convert (TREE_TYPE (chrec
), op0
);
2303 return chrec_convert (TREE_TYPE (chrec
), op0
, NULL
);
2306 /* Handle ~X as -1 - X. */
2307 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2308 TREE_OPERAND (chrec
, 0),
2309 fold_conversions
, cache
, size_expr
);
2310 if (op0
== chrec_dont_know
)
2311 return chrec_dont_know
;
2313 if (TREE_OPERAND (chrec
, 0) != op0
)
2315 op0
= chrec_convert (type
, op0
, NULL
);
2316 chrec
= chrec_fold_minus (type
,
2318 integer_minus_one_node
),
2323 case SCEV_NOT_KNOWN
:
2324 return chrec_dont_know
;
2333 if (VL_EXP_CLASS_P (chrec
))
2334 return chrec_dont_know
;
2336 switch (TREE_CODE_LENGTH (TREE_CODE (chrec
)))
2339 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2340 TREE_OPERAND (chrec
, 0),
2341 fold_conversions
, cache
, size_expr
);
2342 if (op0
== chrec_dont_know
)
2343 return chrec_dont_know
;
2345 op1
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2346 TREE_OPERAND (chrec
, 1),
2347 fold_conversions
, cache
, size_expr
);
2348 if (op1
== chrec_dont_know
)
2349 return chrec_dont_know
;
2351 op2
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2352 TREE_OPERAND (chrec
, 2),
2353 fold_conversions
, cache
, size_expr
);
2354 if (op2
== chrec_dont_know
)
2355 return chrec_dont_know
;
2357 if (op0
== TREE_OPERAND (chrec
, 0)
2358 && op1
== TREE_OPERAND (chrec
, 1)
2359 && op2
== TREE_OPERAND (chrec
, 2))
2362 return fold_build3 (TREE_CODE (chrec
),
2363 TREE_TYPE (chrec
), op0
, op1
, op2
);
2366 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2367 TREE_OPERAND (chrec
, 0),
2368 fold_conversions
, cache
, size_expr
);
2369 if (op0
== chrec_dont_know
)
2370 return chrec_dont_know
;
2372 op1
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2373 TREE_OPERAND (chrec
, 1),
2374 fold_conversions
, cache
, size_expr
);
2375 if (op1
== chrec_dont_know
)
2376 return chrec_dont_know
;
2378 if (op0
== TREE_OPERAND (chrec
, 0)
2379 && op1
== TREE_OPERAND (chrec
, 1))
2381 return fold_build2 (TREE_CODE (chrec
), TREE_TYPE (chrec
), op0
, op1
);
2384 op0
= instantiate_scev_1 (instantiate_below
, evolution_loop
,
2385 TREE_OPERAND (chrec
, 0),
2386 fold_conversions
, cache
, size_expr
);
2387 if (op0
== chrec_dont_know
)
2388 return chrec_dont_know
;
2389 if (op0
== TREE_OPERAND (chrec
, 0))
2391 return fold_build1 (TREE_CODE (chrec
), TREE_TYPE (chrec
), op0
);
2400 /* Too complicated to handle. */
2401 return chrec_dont_know
;
2404 /* Analyze all the parameters of the chrec that were left under a
2405 symbolic form. INSTANTIATE_BELOW is the basic block that stops the
2406 recursive instantiation of parameters: a parameter is a variable
2407 that is defined in a basic block that dominates INSTANTIATE_BELOW or
2408 a function parameter. */
2411 instantiate_scev (basic_block instantiate_below
, struct loop
*evolution_loop
,
2415 htab_t cache
= htab_create (10, hash_scev_info
, eq_scev_info
, del_scev_info
);
2417 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2419 fprintf (dump_file
, "(instantiate_scev \n");
2420 fprintf (dump_file
, " (instantiate_below = %d)\n", instantiate_below
->index
);
2421 fprintf (dump_file
, " (evolution_loop = %d)\n", evolution_loop
->num
);
2422 fprintf (dump_file
, " (chrec = ");
2423 print_generic_expr (dump_file
, chrec
, 0);
2424 fprintf (dump_file
, ")\n");
2427 res
= instantiate_scev_1 (instantiate_below
, evolution_loop
, chrec
, false,
2430 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2432 fprintf (dump_file
, " (res = ");
2433 print_generic_expr (dump_file
, res
, 0);
2434 fprintf (dump_file
, "))\n");
2437 htab_delete (cache
);
2442 /* Similar to instantiate_parameters, but does not introduce the
2443 evolutions in outer loops for LOOP invariants in CHREC, and does not
2444 care about causing overflows, as long as they do not affect value
2445 of an expression. */
2448 resolve_mixers (struct loop
*loop
, tree chrec
)
2450 htab_t cache
= htab_create (10, hash_scev_info
, eq_scev_info
, del_scev_info
);
2451 tree ret
= instantiate_scev_1 (block_before_loop (loop
), loop
, chrec
, true,
2453 htab_delete (cache
);
2457 /* Entry point for the analysis of the number of iterations pass.
2458 This function tries to safely approximate the number of iterations
2459 the loop will run. When this property is not decidable at compile
2460 time, the result is chrec_dont_know. Otherwise the result is
2461 a scalar or a symbolic parameter.
2463 Example of analysis: suppose that the loop has an exit condition:
2465 "if (b > 49) goto end_loop;"
2467 and that in a previous analysis we have determined that the
2468 variable 'b' has an evolution function:
2470 "EF = {23, +, 5}_2".
2472 When we evaluate the function at the point 5, i.e. the value of the
2473 variable 'b' after 5 iterations in the loop, we have EF (5) = 48,
2474 and EF (6) = 53. In this case the value of 'b' on exit is '53' and
2475 the loop body has been executed 6 times. */
2478 number_of_latch_executions (struct loop
*loop
)
2482 struct tree_niter_desc niter_desc
;
2484 /* Determine whether the number_of_iterations_in_loop has already
2486 res
= loop
->nb_iterations
;
2489 res
= chrec_dont_know
;
2491 if (dump_file
&& (dump_flags
& TDF_DETAILS
))
2492 fprintf (dump_file
, "(number_of_iterations_in_loop\n");
2494 exit
= single_exit (loop
);
2498 if (!number_of_iterations_exit (loop
, exit
, &niter_desc
, false))
2501 type
= TREE_TYPE (niter_desc
.niter
);
2502 if (integer_nonzerop (niter_desc
.may_be_zero
))
2503 res
= build_int_cst (type
, 0);
2504 else if (integer_zerop (niter_desc
.may_be_zero
))
2505 res
= niter_desc
.niter
;
2507 res
= chrec_dont_know
;
2510 return set_nb_iterations_in_loop (loop
, res
);
2513 /* Returns the number of executions of the exit condition of LOOP,
2514 i.e., the number by one higher than number_of_latch_executions.
2515 Note that unlike number_of_latch_executions, this number does
2516 not necessarily fit in the unsigned variant of the type of
2517 the control variable -- if the number of iterations is a constant,
2518 we return chrec_dont_know if adding one to number_of_latch_executions
2519 overflows; however, in case the number of iterations is symbolic
2520 expression, the caller is responsible for dealing with this
2521 the possible overflow. */
2524 number_of_exit_cond_executions (struct loop
*loop
)
2526 tree ret
= number_of_latch_executions (loop
);
2527 tree type
= chrec_type (ret
);
2529 if (chrec_contains_undetermined (ret
))
2532 ret
= chrec_fold_plus (type
, ret
, build_int_cst (type
, 1));
2533 if (TREE_CODE (ret
) == INTEGER_CST
2534 && TREE_OVERFLOW (ret
))
2535 return chrec_dont_know
;
2540 /* One of the drivers for testing the scalar evolutions analysis.
2541 This function computes the number of iterations for all the loops
2542 from the EXIT_CONDITIONS array. */
2545 number_of_iterations_for_all_loops (VEC(gimple
,heap
) **exit_conditions
)
2548 unsigned nb_chrec_dont_know_loops
= 0;
2549 unsigned nb_static_loops
= 0;
2552 for (i
= 0; VEC_iterate (gimple
, *exit_conditions
, i
, cond
); i
++)
2554 tree res
= number_of_latch_executions (loop_containing_stmt (cond
));
2555 if (chrec_contains_undetermined (res
))
2556 nb_chrec_dont_know_loops
++;
2563 fprintf (dump_file
, "\n(\n");
2564 fprintf (dump_file
, "-----------------------------------------\n");
2565 fprintf (dump_file
, "%d\tnb_chrec_dont_know_loops\n", nb_chrec_dont_know_loops
);
2566 fprintf (dump_file
, "%d\tnb_static_loops\n", nb_static_loops
);
2567 fprintf (dump_file
, "%d\tnb_total_loops\n", number_of_loops ());
2568 fprintf (dump_file
, "-----------------------------------------\n");
2569 fprintf (dump_file
, ")\n\n");
2571 print_loops (dump_file
, 3);
2577 /* Counters for the stats. */
2583 unsigned nb_affine_multivar
;
2584 unsigned nb_higher_poly
;
2585 unsigned nb_chrec_dont_know
;
2586 unsigned nb_undetermined
;
2589 /* Reset the counters. */
2592 reset_chrecs_counters (struct chrec_stats
*stats
)
2594 stats
->nb_chrecs
= 0;
2595 stats
->nb_affine
= 0;
2596 stats
->nb_affine_multivar
= 0;
2597 stats
->nb_higher_poly
= 0;
2598 stats
->nb_chrec_dont_know
= 0;
2599 stats
->nb_undetermined
= 0;
2602 /* Dump the contents of a CHREC_STATS structure. */
2605 dump_chrecs_stats (FILE *file
, struct chrec_stats
*stats
)
2607 fprintf (file
, "\n(\n");
2608 fprintf (file
, "-----------------------------------------\n");
2609 fprintf (file
, "%d\taffine univariate chrecs\n", stats
->nb_affine
);
2610 fprintf (file
, "%d\taffine multivariate chrecs\n", stats
->nb_affine_multivar
);
2611 fprintf (file
, "%d\tdegree greater than 2 polynomials\n",
2612 stats
->nb_higher_poly
);
2613 fprintf (file
, "%d\tchrec_dont_know chrecs\n", stats
->nb_chrec_dont_know
);
2614 fprintf (file
, "-----------------------------------------\n");
2615 fprintf (file
, "%d\ttotal chrecs\n", stats
->nb_chrecs
);
2616 fprintf (file
, "%d\twith undetermined coefficients\n",
2617 stats
->nb_undetermined
);
2618 fprintf (file
, "-----------------------------------------\n");
2619 fprintf (file
, "%d\tchrecs in the scev database\n",
2620 (int) htab_elements (scalar_evolution_info
));
2621 fprintf (file
, "%d\tsets in the scev database\n", nb_set_scev
);
2622 fprintf (file
, "%d\tgets in the scev database\n", nb_get_scev
);
2623 fprintf (file
, "-----------------------------------------\n");
2624 fprintf (file
, ")\n\n");
2627 /* Gather statistics about CHREC. */
2630 gather_chrec_stats (tree chrec
, struct chrec_stats
*stats
)
2632 if (dump_file
&& (dump_flags
& TDF_STATS
))
2634 fprintf (dump_file
, "(classify_chrec ");
2635 print_generic_expr (dump_file
, chrec
, 0);
2636 fprintf (dump_file
, "\n");
2641 if (chrec
== NULL_TREE
)
2643 stats
->nb_undetermined
++;
2647 switch (TREE_CODE (chrec
))
2649 case POLYNOMIAL_CHREC
:
2650 if (evolution_function_is_affine_p (chrec
))
2652 if (dump_file
&& (dump_flags
& TDF_STATS
))
2653 fprintf (dump_file
, " affine_univariate\n");
2656 else if (evolution_function_is_affine_multivariate_p (chrec
, 0))
2658 if (dump_file
&& (dump_flags
& TDF_STATS
))
2659 fprintf (dump_file
, " affine_multivariate\n");
2660 stats
->nb_affine_multivar
++;
2664 if (dump_file
&& (dump_flags
& TDF_STATS
))
2665 fprintf (dump_file
, " higher_degree_polynomial\n");
2666 stats
->nb_higher_poly
++;
2675 if (chrec_contains_undetermined (chrec
))
2677 if (dump_file
&& (dump_flags
& TDF_STATS
))
2678 fprintf (dump_file
, " undetermined\n");
2679 stats
->nb_undetermined
++;
2682 if (dump_file
&& (dump_flags
& TDF_STATS
))
2683 fprintf (dump_file
, ")\n");
2686 /* One of the drivers for testing the scalar evolutions analysis.
2687 This function analyzes the scalar evolution of all the scalars
2688 defined as loop phi nodes in one of the loops from the
2689 EXIT_CONDITIONS array.
2691 TODO Optimization: A loop is in canonical form if it contains only
2692 a single scalar loop phi node. All the other scalars that have an
2693 evolution in the loop are rewritten in function of this single
2694 index. This allows the parallelization of the loop. */
2697 analyze_scalar_evolution_for_all_loop_phi_nodes (VEC(gimple
,heap
) **exit_conditions
)
2700 struct chrec_stats stats
;
2702 gimple_stmt_iterator psi
;
2704 reset_chrecs_counters (&stats
);
2706 for (i
= 0; VEC_iterate (gimple
, *exit_conditions
, i
, cond
); i
++)
2712 loop
= loop_containing_stmt (cond
);
2715 for (psi
= gsi_start_phis (bb
); !gsi_end_p (psi
); gsi_next (&psi
))
2717 phi
= gsi_stmt (psi
);
2718 if (is_gimple_reg (PHI_RESULT (phi
)))
2720 chrec
= instantiate_parameters
2722 analyze_scalar_evolution (loop
, PHI_RESULT (phi
)));
2724 if (dump_file
&& (dump_flags
& TDF_STATS
))
2725 gather_chrec_stats (chrec
, &stats
);
2730 if (dump_file
&& (dump_flags
& TDF_STATS
))
2731 dump_chrecs_stats (dump_file
, &stats
);
2734 /* Callback for htab_traverse, gathers information on chrecs in the
2738 gather_stats_on_scev_database_1 (void **slot
, void *stats
)
2740 struct scev_info_str
*entry
= (struct scev_info_str
*) *slot
;
2742 gather_chrec_stats (entry
->chrec
, (struct chrec_stats
*) stats
);
2747 /* Classify the chrecs of the whole database. */
2750 gather_stats_on_scev_database (void)
2752 struct chrec_stats stats
;
2757 reset_chrecs_counters (&stats
);
2759 htab_traverse (scalar_evolution_info
, gather_stats_on_scev_database_1
,
2762 dump_chrecs_stats (dump_file
, &stats
);
2770 initialize_scalar_evolutions_analyzer (void)
2772 /* The elements below are unique. */
2773 if (chrec_dont_know
== NULL_TREE
)
2775 chrec_not_analyzed_yet
= NULL_TREE
;
2776 chrec_dont_know
= make_node (SCEV_NOT_KNOWN
);
2777 chrec_known
= make_node (SCEV_KNOWN
);
2778 TREE_TYPE (chrec_dont_know
) = void_type_node
;
2779 TREE_TYPE (chrec_known
) = void_type_node
;
2783 /* Initialize the analysis of scalar evolutions for LOOPS. */
2786 scev_initialize (void)
2791 scalar_evolution_info
= htab_create_alloc (100,
2798 initialize_scalar_evolutions_analyzer ();
2800 FOR_EACH_LOOP (li
, loop
, 0)
2802 loop
->nb_iterations
= NULL_TREE
;
2806 /* Cleans up the information cached by the scalar evolutions analysis. */
2814 if (!scalar_evolution_info
|| !current_loops
)
2817 htab_empty (scalar_evolution_info
);
2818 FOR_EACH_LOOP (li
, loop
, 0)
2820 loop
->nb_iterations
= NULL_TREE
;
2824 /* Checks whether use of OP in USE_LOOP behaves as a simple affine iv with
2825 respect to WRTO_LOOP and returns its base and step in IV if possible
2826 (see analyze_scalar_evolution_in_loop for more details on USE_LOOP
2827 and WRTO_LOOP). If ALLOW_NONCONSTANT_STEP is true, we want step to be
2828 invariant in LOOP. Otherwise we require it to be an integer constant.
2830 IV->no_overflow is set to true if we are sure the iv cannot overflow (e.g.
2831 because it is computed in signed arithmetics). Consequently, adding an
2834 for (i = IV->base; ; i += IV->step)
2836 is only safe if IV->no_overflow is false, or TYPE_OVERFLOW_UNDEFINED is
2837 false for the type of the induction variable, or you can prove that i does
2838 not wrap by some other argument. Otherwise, this might introduce undefined
2841 for (i = iv->base; ; i = (type) ((unsigned type) i + (unsigned type) iv->step))
2843 must be used instead. */
2846 simple_iv (struct loop
*wrto_loop
, struct loop
*use_loop
, tree op
,
2847 affine_iv
*iv
, bool allow_nonconstant_step
)
2852 iv
->base
= NULL_TREE
;
2853 iv
->step
= NULL_TREE
;
2854 iv
->no_overflow
= false;
2856 type
= TREE_TYPE (op
);
2857 if (TREE_CODE (type
) != INTEGER_TYPE
2858 && TREE_CODE (type
) != POINTER_TYPE
)
2861 ev
= analyze_scalar_evolution_in_loop (wrto_loop
, use_loop
, op
,
2863 if (chrec_contains_undetermined (ev
)
2864 || chrec_contains_symbols_defined_in_loop (ev
, wrto_loop
->num
))
2867 if (tree_does_not_contain_chrecs (ev
))
2870 iv
->step
= build_int_cst (TREE_TYPE (ev
), 0);
2871 iv
->no_overflow
= true;
2875 if (TREE_CODE (ev
) != POLYNOMIAL_CHREC
2876 || CHREC_VARIABLE (ev
) != (unsigned) wrto_loop
->num
)
2879 iv
->step
= CHREC_RIGHT (ev
);
2880 if ((!allow_nonconstant_step
&& TREE_CODE (iv
->step
) != INTEGER_CST
)
2881 || tree_contains_chrecs (iv
->step
, NULL
))
2884 iv
->base
= CHREC_LEFT (ev
);
2885 if (tree_contains_chrecs (iv
->base
, NULL
))
2888 iv
->no_overflow
= !folded_casts
&& TYPE_OVERFLOW_UNDEFINED (type
);
2893 /* Runs the analysis of scalar evolutions. */
2896 scev_analysis (void)
2898 VEC(gimple
,heap
) *exit_conditions
;
2900 exit_conditions
= VEC_alloc (gimple
, heap
, 37);
2901 select_loops_exit_conditions (&exit_conditions
);
2903 if (dump_file
&& (dump_flags
& TDF_STATS
))
2904 analyze_scalar_evolution_for_all_loop_phi_nodes (&exit_conditions
);
2906 number_of_iterations_for_all_loops (&exit_conditions
);
2907 VEC_free (gimple
, heap
, exit_conditions
);
2910 /* Finalize the scalar evolution analysis. */
2913 scev_finalize (void)
2915 if (!scalar_evolution_info
)
2917 htab_delete (scalar_evolution_info
);
2918 scalar_evolution_info
= NULL
;
2921 /* Returns true if the expression EXPR is considered to be too expensive
2922 for scev_const_prop. */
2925 expression_expensive_p (tree expr
)
2927 enum tree_code code
;
2929 if (is_gimple_val (expr
))
2932 code
= TREE_CODE (expr
);
2933 if (code
== TRUNC_DIV_EXPR
2934 || code
== CEIL_DIV_EXPR
2935 || code
== FLOOR_DIV_EXPR
2936 || code
== ROUND_DIV_EXPR
2937 || code
== TRUNC_MOD_EXPR
2938 || code
== CEIL_MOD_EXPR
2939 || code
== FLOOR_MOD_EXPR
2940 || code
== ROUND_MOD_EXPR
2941 || code
== EXACT_DIV_EXPR
)
2943 /* Division by power of two is usually cheap, so we allow it.
2944 Forbid anything else. */
2945 if (!integer_pow2p (TREE_OPERAND (expr
, 1)))
2949 switch (TREE_CODE_CLASS (code
))
2952 case tcc_comparison
:
2953 if (expression_expensive_p (TREE_OPERAND (expr
, 1)))
2958 return expression_expensive_p (TREE_OPERAND (expr
, 0));
2965 /* Replace ssa names for that scev can prove they are constant by the
2966 appropriate constants. Also perform final value replacement in loops,
2967 in case the replacement expressions are cheap.
2969 We only consider SSA names defined by phi nodes; rest is left to the
2970 ordinary constant propagation pass. */
2973 scev_const_prop (void)
2976 tree name
, type
, ev
;
2978 struct loop
*loop
, *ex_loop
;
2979 bitmap ssa_names_to_remove
= NULL
;
2982 gimple_stmt_iterator psi
;
2984 if (number_of_loops () <= 1)
2989 loop
= bb
->loop_father
;
2991 for (psi
= gsi_start_phis (bb
); !gsi_end_p (psi
); gsi_next (&psi
))
2993 phi
= gsi_stmt (psi
);
2994 name
= PHI_RESULT (phi
);
2996 if (!is_gimple_reg (name
))
2999 type
= TREE_TYPE (name
);
3001 if (!POINTER_TYPE_P (type
)
3002 && !INTEGRAL_TYPE_P (type
))
3005 ev
= resolve_mixers (loop
, analyze_scalar_evolution (loop
, name
));
3006 if (!is_gimple_min_invariant (ev
)
3007 || !may_propagate_copy (name
, ev
))
3010 /* Replace the uses of the name. */
3012 replace_uses_by (name
, ev
);
3014 if (!ssa_names_to_remove
)
3015 ssa_names_to_remove
= BITMAP_ALLOC (NULL
);
3016 bitmap_set_bit (ssa_names_to_remove
, SSA_NAME_VERSION (name
));
3020 /* Remove the ssa names that were replaced by constants. We do not
3021 remove them directly in the previous cycle, since this
3022 invalidates scev cache. */
3023 if (ssa_names_to_remove
)
3027 EXECUTE_IF_SET_IN_BITMAP (ssa_names_to_remove
, 0, i
, bi
)
3029 gimple_stmt_iterator psi
;
3030 name
= ssa_name (i
);
3031 phi
= SSA_NAME_DEF_STMT (name
);
3033 gcc_assert (gimple_code (phi
) == GIMPLE_PHI
);
3034 psi
= gsi_for_stmt (phi
);
3035 remove_phi_node (&psi
, true);
3038 BITMAP_FREE (ssa_names_to_remove
);
3042 /* Now the regular final value replacement. */
3043 FOR_EACH_LOOP (li
, loop
, LI_FROM_INNERMOST
)
3046 tree def
, rslt
, niter
;
3047 gimple_stmt_iterator bsi
;
3049 /* If we do not know exact number of iterations of the loop, we cannot
3050 replace the final value. */
3051 exit
= single_exit (loop
);
3055 niter
= number_of_latch_executions (loop
);
3056 if (niter
== chrec_dont_know
)
3059 /* Ensure that it is possible to insert new statements somewhere. */
3060 if (!single_pred_p (exit
->dest
))
3061 split_loop_exit_edge (exit
);
3062 bsi
= gsi_after_labels (exit
->dest
);
3064 ex_loop
= superloop_at_depth (loop
,
3065 loop_depth (exit
->dest
->loop_father
) + 1);
3067 for (psi
= gsi_start_phis (exit
->dest
); !gsi_end_p (psi
); )
3069 phi
= gsi_stmt (psi
);
3070 rslt
= PHI_RESULT (phi
);
3071 def
= PHI_ARG_DEF_FROM_EDGE (phi
, exit
);
3072 if (!is_gimple_reg (def
))
3078 if (!POINTER_TYPE_P (TREE_TYPE (def
))
3079 && !INTEGRAL_TYPE_P (TREE_TYPE (def
)))
3085 def
= analyze_scalar_evolution_in_loop (ex_loop
, loop
, def
, NULL
);
3086 def
= compute_overall_effect_of_inner_loop (ex_loop
, def
);
3087 if (!tree_does_not_contain_chrecs (def
)
3088 || chrec_contains_symbols_defined_in_loop (def
, ex_loop
->num
)
3089 /* Moving the computation from the loop may prolong life range
3090 of some ssa names, which may cause problems if they appear
3091 on abnormal edges. */
3092 || contains_abnormal_ssa_name_p (def
)
3093 /* Do not emit expensive expressions. The rationale is that
3094 when someone writes a code like
3096 while (n > 45) n -= 45;
3098 he probably knows that n is not large, and does not want it
3099 to be turned into n %= 45. */
3100 || expression_expensive_p (def
))
3106 /* Eliminate the PHI node and replace it by a computation outside
3108 def
= unshare_expr (def
);
3109 remove_phi_node (&psi
, false);
3111 def
= force_gimple_operand_gsi (&bsi
, def
, false, NULL_TREE
,
3112 true, GSI_SAME_STMT
);
3113 ass
= gimple_build_assign (rslt
, def
);
3114 gsi_insert_before (&bsi
, ass
, GSI_SAME_STMT
);
3120 #include "gt-tree-scalar-evolution.h"