gcc/
[official-gcc.git] / gcc / tree-ssa-dom.c
blob64ff4efca852cb43424c4d3daa1c160bdff6b1a2
1 /* SSA Dominator optimizations for trees
2 Copyright (C) 2001, 2002, 2003, 2004, 2005, 2006, 2007
3 Free Software Foundation, Inc.
4 Contributed by Diego Novillo <dnovillo@redhat.com>
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3, or (at your option)
11 any later version.
13 GCC is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
22 #include "config.h"
23 #include "system.h"
24 #include "coretypes.h"
25 #include "tm.h"
26 #include "tree.h"
27 #include "flags.h"
28 #include "rtl.h"
29 #include "tm_p.h"
30 #include "ggc.h"
31 #include "basic-block.h"
32 #include "cfgloop.h"
33 #include "output.h"
34 #include "expr.h"
35 #include "function.h"
36 #include "diagnostic.h"
37 #include "timevar.h"
38 #include "tree-dump.h"
39 #include "tree-flow.h"
40 #include "domwalk.h"
41 #include "real.h"
42 #include "tree-pass.h"
43 #include "tree-ssa-propagate.h"
44 #include "langhooks.h"
45 #include "params.h"
47 /* This file implements optimizations on the dominator tree. */
50 /* Structure for recording edge equivalences as well as any pending
51 edge redirections during the dominator optimizer.
53 Computing and storing the edge equivalences instead of creating
54 them on-demand can save significant amounts of time, particularly
55 for pathological cases involving switch statements.
57 These structures live for a single iteration of the dominator
58 optimizer in the edge's AUX field. At the end of an iteration we
59 free each of these structures and update the AUX field to point
60 to any requested redirection target (the code for updating the
61 CFG and SSA graph for edge redirection expects redirection edge
62 targets to be in the AUX field for each edge. */
64 struct edge_info
66 /* If this edge creates a simple equivalence, the LHS and RHS of
67 the equivalence will be stored here. */
68 tree lhs;
69 tree rhs;
71 /* Traversing an edge may also indicate one or more particular conditions
72 are true or false. The number of recorded conditions can vary, but
73 can be determined by the condition's code. So we have an array
74 and its maximum index rather than use a varray. */
75 tree *cond_equivalences;
76 unsigned int max_cond_equivalences;
80 /* Hash table with expressions made available during the renaming process.
81 When an assignment of the form X_i = EXPR is found, the statement is
82 stored in this table. If the same expression EXPR is later found on the
83 RHS of another statement, it is replaced with X_i (thus performing
84 global redundancy elimination). Similarly as we pass through conditionals
85 we record the conditional itself as having either a true or false value
86 in this table. */
87 static htab_t avail_exprs;
89 /* Stack of available expressions in AVAIL_EXPRs. Each block pushes any
90 expressions it enters into the hash table along with a marker entry
91 (null). When we finish processing the block, we pop off entries and
92 remove the expressions from the global hash table until we hit the
93 marker. */
94 static VEC(tree,heap) *avail_exprs_stack;
96 /* Stack of statements we need to rescan during finalization for newly
97 exposed variables.
99 Statement rescanning must occur after the current block's available
100 expressions are removed from AVAIL_EXPRS. Else we may change the
101 hash code for an expression and be unable to find/remove it from
102 AVAIL_EXPRS. */
103 typedef tree *tree_p;
104 DEF_VEC_P(tree_p);
105 DEF_VEC_ALLOC_P(tree_p,heap);
107 static VEC(tree_p,heap) *stmts_to_rescan;
109 /* Structure for entries in the expression hash table.
111 This requires more memory for the hash table entries, but allows us
112 to avoid creating silly tree nodes and annotations for conditionals,
113 eliminates 2 global hash tables and two block local varrays.
115 It also allows us to reduce the number of hash table lookups we
116 have to perform in lookup_avail_expr and finally it allows us to
117 significantly reduce the number of calls into the hashing routine
118 itself. */
120 struct expr_hash_elt
122 /* The value (lhs) of this expression. */
123 tree lhs;
125 /* The expression (rhs) we want to record. */
126 tree rhs;
128 /* The stmt pointer if this element corresponds to a statement. */
129 tree stmt;
131 /* The hash value for RHS/ann. */
132 hashval_t hash;
135 /* Stack of dest,src pairs that need to be restored during finalization.
137 A NULL entry is used to mark the end of pairs which need to be
138 restored during finalization of this block. */
139 static VEC(tree,heap) *const_and_copies_stack;
141 /* Track whether or not we have changed the control flow graph. */
142 static bool cfg_altered;
144 /* Bitmap of blocks that have had EH statements cleaned. We should
145 remove their dead edges eventually. */
146 static bitmap need_eh_cleanup;
148 /* Statistics for dominator optimizations. */
149 struct opt_stats_d
151 long num_stmts;
152 long num_exprs_considered;
153 long num_re;
154 long num_const_prop;
155 long num_copy_prop;
158 static struct opt_stats_d opt_stats;
160 struct eq_expr_value
162 tree src;
163 tree dst;
166 /* Local functions. */
167 static void optimize_stmt (struct dom_walk_data *,
168 basic_block bb,
169 block_stmt_iterator);
170 static tree lookup_avail_expr (tree, bool);
171 static hashval_t avail_expr_hash (const void *);
172 static hashval_t real_avail_expr_hash (const void *);
173 static int avail_expr_eq (const void *, const void *);
174 static void htab_statistics (FILE *, htab_t);
175 static void record_cond (tree, tree);
176 static void record_const_or_copy (tree, tree);
177 static void record_equality (tree, tree);
178 static void record_equivalences_from_phis (basic_block);
179 static void record_equivalences_from_incoming_edge (basic_block);
180 static bool eliminate_redundant_computations (tree);
181 static void record_equivalences_from_stmt (tree, int, stmt_ann_t);
182 static void dom_thread_across_edge (struct dom_walk_data *, edge);
183 static void dom_opt_finalize_block (struct dom_walk_data *, basic_block);
184 static void dom_opt_initialize_block (struct dom_walk_data *, basic_block);
185 static void propagate_to_outgoing_edges (struct dom_walk_data *, basic_block);
186 static void remove_local_expressions_from_table (void);
187 static void restore_vars_to_original_value (void);
188 static edge single_incoming_edge_ignoring_loop_edges (basic_block);
191 /* Allocate an EDGE_INFO for edge E and attach it to E.
192 Return the new EDGE_INFO structure. */
194 static struct edge_info *
195 allocate_edge_info (edge e)
197 struct edge_info *edge_info;
199 edge_info = XCNEW (struct edge_info);
201 e->aux = edge_info;
202 return edge_info;
205 /* Free all EDGE_INFO structures associated with edges in the CFG.
206 If a particular edge can be threaded, copy the redirection
207 target from the EDGE_INFO structure into the edge's AUX field
208 as required by code to update the CFG and SSA graph for
209 jump threading. */
211 static void
212 free_all_edge_infos (void)
214 basic_block bb;
215 edge_iterator ei;
216 edge e;
218 FOR_EACH_BB (bb)
220 FOR_EACH_EDGE (e, ei, bb->preds)
222 struct edge_info *edge_info = (struct edge_info *) e->aux;
224 if (edge_info)
226 if (edge_info->cond_equivalences)
227 free (edge_info->cond_equivalences);
228 free (edge_info);
229 e->aux = NULL;
235 /* Jump threading, redundancy elimination and const/copy propagation.
237 This pass may expose new symbols that need to be renamed into SSA. For
238 every new symbol exposed, its corresponding bit will be set in
239 VARS_TO_RENAME. */
241 static unsigned int
242 tree_ssa_dominator_optimize (void)
244 struct dom_walk_data walk_data;
245 unsigned int i;
247 memset (&opt_stats, 0, sizeof (opt_stats));
249 /* Create our hash tables. */
250 avail_exprs = htab_create (1024, real_avail_expr_hash, avail_expr_eq, free);
251 avail_exprs_stack = VEC_alloc (tree, heap, 20);
252 const_and_copies_stack = VEC_alloc (tree, heap, 20);
253 stmts_to_rescan = VEC_alloc (tree_p, heap, 20);
254 need_eh_cleanup = BITMAP_ALLOC (NULL);
256 /* Setup callbacks for the generic dominator tree walker. */
257 walk_data.walk_stmts_backward = false;
258 walk_data.dom_direction = CDI_DOMINATORS;
259 walk_data.initialize_block_local_data = NULL;
260 walk_data.before_dom_children_before_stmts = dom_opt_initialize_block;
261 walk_data.before_dom_children_walk_stmts = optimize_stmt;
262 walk_data.before_dom_children_after_stmts = propagate_to_outgoing_edges;
263 walk_data.after_dom_children_before_stmts = NULL;
264 walk_data.after_dom_children_walk_stmts = NULL;
265 walk_data.after_dom_children_after_stmts = dom_opt_finalize_block;
266 /* Right now we only attach a dummy COND_EXPR to the global data pointer.
267 When we attach more stuff we'll need to fill this out with a real
268 structure. */
269 walk_data.global_data = NULL;
270 walk_data.block_local_data_size = 0;
271 walk_data.interesting_blocks = NULL;
273 /* Now initialize the dominator walker. */
274 init_walk_dominator_tree (&walk_data);
276 calculate_dominance_info (CDI_DOMINATORS);
277 cfg_altered = false;
279 /* We need to know loop structures in order to avoid destroying them
280 in jump threading. Note that we still can e.g. thread through loop
281 headers to an exit edge, or through loop header to the loop body, assuming
282 that we update the loop info. */
283 loop_optimizer_init (LOOPS_HAVE_SIMPLE_LATCHES);
285 /* We need accurate information regarding back edges in the CFG
286 for jump threading; this may include back edes that are not part of
287 a single loop. */
288 mark_dfs_back_edges ();
290 /* Recursively walk the dominator tree optimizing statements. */
291 walk_dominator_tree (&walk_data, ENTRY_BLOCK_PTR);
294 block_stmt_iterator bsi;
295 basic_block bb;
296 FOR_EACH_BB (bb)
298 for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
299 update_stmt_if_modified (bsi_stmt (bsi));
303 /* If we exposed any new variables, go ahead and put them into
304 SSA form now, before we handle jump threading. This simplifies
305 interactions between rewriting of _DECL nodes into SSA form
306 and rewriting SSA_NAME nodes into SSA form after block
307 duplication and CFG manipulation. */
308 update_ssa (TODO_update_ssa);
310 free_all_edge_infos ();
312 /* Thread jumps, creating duplicate blocks as needed. */
313 cfg_altered |= thread_through_all_blocks (first_pass_instance);
315 if (cfg_altered)
316 free_dominance_info (CDI_DOMINATORS);
318 /* Removal of statements may make some EH edges dead. Purge
319 such edges from the CFG as needed. */
320 if (!bitmap_empty_p (need_eh_cleanup))
322 tree_purge_all_dead_eh_edges (need_eh_cleanup);
323 bitmap_zero (need_eh_cleanup);
326 /* Finally, remove everything except invariants in SSA_NAME_VALUE.
328 Long term we will be able to let everything in SSA_NAME_VALUE
329 persist. However, for now, we know this is the safe thing to do. */
330 for (i = 0; i < num_ssa_names; i++)
332 tree name = ssa_name (i);
333 tree value;
335 if (!name)
336 continue;
338 value = SSA_NAME_VALUE (name);
339 if (value && !is_gimple_min_invariant (value))
340 SSA_NAME_VALUE (name) = NULL;
343 /* Debugging dumps. */
344 if (dump_file && (dump_flags & TDF_STATS))
345 dump_dominator_optimization_stats (dump_file);
347 loop_optimizer_finalize ();
349 /* Delete our main hashtable. */
350 htab_delete (avail_exprs);
352 /* And finalize the dominator walker. */
353 fini_walk_dominator_tree (&walk_data);
355 /* Free asserted bitmaps and stacks. */
356 BITMAP_FREE (need_eh_cleanup);
358 VEC_free (tree, heap, avail_exprs_stack);
359 VEC_free (tree, heap, const_and_copies_stack);
360 VEC_free (tree_p, heap, stmts_to_rescan);
361 return 0;
364 static bool
365 gate_dominator (void)
367 return flag_tree_dom != 0;
370 struct gimple_opt_pass pass_dominator =
373 GIMPLE_PASS,
374 "dom", /* name */
375 gate_dominator, /* gate */
376 tree_ssa_dominator_optimize, /* execute */
377 NULL, /* sub */
378 NULL, /* next */
379 0, /* static_pass_number */
380 TV_TREE_SSA_DOMINATOR_OPTS, /* tv_id */
381 PROP_cfg | PROP_ssa | PROP_alias, /* properties_required */
382 0, /* properties_provided */
383 0, /* properties_destroyed */
384 0, /* todo_flags_start */
385 TODO_dump_func
386 | TODO_update_ssa
387 | TODO_cleanup_cfg
388 | TODO_verify_ssa /* todo_flags_finish */
393 /* Given a stmt CONDSTMT containing a COND_EXPR, canonicalize the
394 COND_EXPR into a canonical form. */
396 static void
397 canonicalize_comparison (tree condstmt)
399 tree cond = COND_EXPR_COND (condstmt);
400 tree op0;
401 tree op1;
402 enum tree_code code = TREE_CODE (cond);
404 if (!COMPARISON_CLASS_P (cond))
405 return;
407 op0 = TREE_OPERAND (cond, 0);
408 op1 = TREE_OPERAND (cond, 1);
410 /* If it would be profitable to swap the operands, then do so to
411 canonicalize the statement, enabling better optimization.
413 By placing canonicalization of such expressions here we
414 transparently keep statements in canonical form, even
415 when the statement is modified. */
416 if (tree_swap_operands_p (op0, op1, false))
418 /* For relationals we need to swap the operands
419 and change the code. */
420 if (code == LT_EXPR
421 || code == GT_EXPR
422 || code == LE_EXPR
423 || code == GE_EXPR)
425 TREE_SET_CODE (cond, swap_tree_comparison (code));
426 swap_tree_operands (condstmt,
427 &TREE_OPERAND (cond, 0),
428 &TREE_OPERAND (cond, 1));
429 /* If one operand was in the operand cache, but the other is
430 not, because it is a constant, this is a case that the
431 internal updating code of swap_tree_operands can't handle
432 properly. */
433 if (TREE_CODE_CLASS (TREE_CODE (op0))
434 != TREE_CODE_CLASS (TREE_CODE (op1)))
435 update_stmt (condstmt);
440 /* Initialize local stacks for this optimizer and record equivalences
441 upon entry to BB. Equivalences can come from the edge traversed to
442 reach BB or they may come from PHI nodes at the start of BB. */
444 static void
445 dom_opt_initialize_block (struct dom_walk_data *walk_data ATTRIBUTE_UNUSED,
446 basic_block bb)
448 if (dump_file && (dump_flags & TDF_DETAILS))
449 fprintf (dump_file, "\n\nOptimizing block #%d\n\n", bb->index);
451 /* Push a marker on the stacks of local information so that we know how
452 far to unwind when we finalize this block. */
453 VEC_safe_push (tree, heap, avail_exprs_stack, NULL_TREE);
454 VEC_safe_push (tree, heap, const_and_copies_stack, NULL_TREE);
456 record_equivalences_from_incoming_edge (bb);
458 /* PHI nodes can create equivalences too. */
459 record_equivalences_from_phis (bb);
462 /* Given an expression EXPR (a relational expression or a statement),
463 initialize the hash table element pointed to by ELEMENT. */
465 static void
466 initialize_hash_element (tree expr, tree lhs, struct expr_hash_elt *element)
468 /* Hash table elements may be based on conditional expressions or statements.
470 For the former case, we have no annotation and we want to hash the
471 conditional expression. In the latter case we have an annotation and
472 we want to record the expression the statement evaluates. */
473 if (COMPARISON_CLASS_P (expr) || TREE_CODE (expr) == TRUTH_NOT_EXPR)
475 element->stmt = NULL;
476 element->rhs = expr;
478 else if (TREE_CODE (expr) == COND_EXPR)
480 element->stmt = expr;
481 element->rhs = COND_EXPR_COND (expr);
483 else if (TREE_CODE (expr) == SWITCH_EXPR)
485 element->stmt = expr;
486 element->rhs = SWITCH_COND (expr);
488 else if (TREE_CODE (expr) == RETURN_EXPR && TREE_OPERAND (expr, 0))
490 element->stmt = expr;
491 element->rhs = GIMPLE_STMT_OPERAND (TREE_OPERAND (expr, 0), 1);
493 else if (TREE_CODE (expr) == GOTO_EXPR)
495 element->stmt = expr;
496 element->rhs = GOTO_DESTINATION (expr);
498 else
500 element->stmt = expr;
501 element->rhs = GENERIC_TREE_OPERAND (expr, 1);
504 element->lhs = lhs;
505 element->hash = avail_expr_hash (element);
508 /* Remove all the expressions in LOCALS from TABLE, stopping when there are
509 LIMIT entries left in LOCALs. */
511 static void
512 remove_local_expressions_from_table (void)
514 /* Remove all the expressions made available in this block. */
515 while (VEC_length (tree, avail_exprs_stack) > 0)
517 struct expr_hash_elt element;
518 tree expr = VEC_pop (tree, avail_exprs_stack);
520 if (expr == NULL_TREE)
521 break;
523 initialize_hash_element (expr, NULL, &element);
524 htab_remove_elt_with_hash (avail_exprs, &element, element.hash);
528 /* Use the source/dest pairs in CONST_AND_COPIES_STACK to restore
529 CONST_AND_COPIES to its original state, stopping when we hit a
530 NULL marker. */
532 static void
533 restore_vars_to_original_value (void)
535 while (VEC_length (tree, const_and_copies_stack) > 0)
537 tree prev_value, dest;
539 dest = VEC_pop (tree, const_and_copies_stack);
541 if (dest == NULL)
542 break;
544 prev_value = VEC_pop (tree, const_and_copies_stack);
545 SSA_NAME_VALUE (dest) = prev_value;
549 /* A trivial wrapper so that we can present the generic jump
550 threading code with a simple API for simplifying statements. */
551 static tree
552 simplify_stmt_for_jump_threading (tree stmt, tree within_stmt ATTRIBUTE_UNUSED)
554 return lookup_avail_expr (stmt, false);
557 /* Wrapper for common code to attempt to thread an edge. For example,
558 it handles lazily building the dummy condition and the bookkeeping
559 when jump threading is successful. */
561 static void
562 dom_thread_across_edge (struct dom_walk_data *walk_data, edge e)
564 /* If we don't already have a dummy condition, build it now. */
565 if (! walk_data->global_data)
567 tree dummy_cond = build2 (NE_EXPR, boolean_type_node,
568 integer_zero_node, integer_zero_node);
569 dummy_cond = build3 (COND_EXPR, void_type_node, dummy_cond, NULL, NULL);
570 walk_data->global_data = dummy_cond;
573 thread_across_edge ((tree) walk_data->global_data, e, false,
574 &const_and_copies_stack,
575 simplify_stmt_for_jump_threading);
578 /* We have finished processing the dominator children of BB, perform
579 any finalization actions in preparation for leaving this node in
580 the dominator tree. */
582 static void
583 dom_opt_finalize_block (struct dom_walk_data *walk_data, basic_block bb)
585 tree last;
588 /* If we have an outgoing edge to a block with multiple incoming and
589 outgoing edges, then we may be able to thread the edge. ie, we
590 may be able to statically determine which of the outgoing edges
591 will be traversed when the incoming edge from BB is traversed. */
592 if (single_succ_p (bb)
593 && (single_succ_edge (bb)->flags & EDGE_ABNORMAL) == 0
594 && potentially_threadable_block (single_succ (bb)))
596 dom_thread_across_edge (walk_data, single_succ_edge (bb));
598 else if ((last = last_stmt (bb))
599 && TREE_CODE (last) == COND_EXPR
600 && (COMPARISON_CLASS_P (COND_EXPR_COND (last))
601 || TREE_CODE (COND_EXPR_COND (last)) == SSA_NAME)
602 && EDGE_COUNT (bb->succs) == 2
603 && (EDGE_SUCC (bb, 0)->flags & EDGE_ABNORMAL) == 0
604 && (EDGE_SUCC (bb, 1)->flags & EDGE_ABNORMAL) == 0)
606 edge true_edge, false_edge;
608 extract_true_false_edges_from_block (bb, &true_edge, &false_edge);
610 /* Only try to thread the edge if it reaches a target block with
611 more than one predecessor and more than one successor. */
612 if (potentially_threadable_block (true_edge->dest))
614 struct edge_info *edge_info;
615 unsigned int i;
617 /* Push a marker onto the available expression stack so that we
618 unwind any expressions related to the TRUE arm before processing
619 the false arm below. */
620 VEC_safe_push (tree, heap, avail_exprs_stack, NULL_TREE);
621 VEC_safe_push (tree, heap, const_and_copies_stack, NULL_TREE);
623 edge_info = (struct edge_info *) true_edge->aux;
625 /* If we have info associated with this edge, record it into
626 our equivalency tables. */
627 if (edge_info)
629 tree *cond_equivalences = edge_info->cond_equivalences;
630 tree lhs = edge_info->lhs;
631 tree rhs = edge_info->rhs;
633 /* If we have a simple NAME = VALUE equivalency record it. */
634 if (lhs && TREE_CODE (lhs) == SSA_NAME)
635 record_const_or_copy (lhs, rhs);
637 /* If we have 0 = COND or 1 = COND equivalences, record them
638 into our expression hash tables. */
639 if (cond_equivalences)
640 for (i = 0; i < edge_info->max_cond_equivalences; i += 2)
642 tree expr = cond_equivalences[i];
643 tree value = cond_equivalences[i + 1];
645 record_cond (expr, value);
649 dom_thread_across_edge (walk_data, true_edge);
651 /* And restore the various tables to their state before
652 we threaded this edge. */
653 remove_local_expressions_from_table ();
656 /* Similarly for the ELSE arm. */
657 if (potentially_threadable_block (false_edge->dest))
659 struct edge_info *edge_info;
660 unsigned int i;
662 VEC_safe_push (tree, heap, const_and_copies_stack, NULL_TREE);
663 edge_info = (struct edge_info *) false_edge->aux;
665 /* If we have info associated with this edge, record it into
666 our equivalency tables. */
667 if (edge_info)
669 tree *cond_equivalences = edge_info->cond_equivalences;
670 tree lhs = edge_info->lhs;
671 tree rhs = edge_info->rhs;
673 /* If we have a simple NAME = VALUE equivalency record it. */
674 if (lhs && TREE_CODE (lhs) == SSA_NAME)
675 record_const_or_copy (lhs, rhs);
677 /* If we have 0 = COND or 1 = COND equivalences, record them
678 into our expression hash tables. */
679 if (cond_equivalences)
680 for (i = 0; i < edge_info->max_cond_equivalences; i += 2)
682 tree expr = cond_equivalences[i];
683 tree value = cond_equivalences[i + 1];
685 record_cond (expr, value);
689 /* Now thread the edge. */
690 dom_thread_across_edge (walk_data, false_edge);
692 /* No need to remove local expressions from our tables
693 or restore vars to their original value as that will
694 be done immediately below. */
698 remove_local_expressions_from_table ();
699 restore_vars_to_original_value ();
701 /* If we queued any statements to rescan in this block, then
702 go ahead and rescan them now. */
703 while (VEC_length (tree_p, stmts_to_rescan) > 0)
705 tree *stmt_p = VEC_last (tree_p, stmts_to_rescan);
706 tree stmt = *stmt_p;
707 basic_block stmt_bb = bb_for_stmt (stmt);
709 if (stmt_bb != bb)
710 break;
712 VEC_pop (tree_p, stmts_to_rescan);
713 pop_stmt_changes (stmt_p);
717 /* PHI nodes can create equivalences too.
719 Ignoring any alternatives which are the same as the result, if
720 all the alternatives are equal, then the PHI node creates an
721 equivalence. */
723 static void
724 record_equivalences_from_phis (basic_block bb)
726 tree phi;
728 for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
730 tree lhs = PHI_RESULT (phi);
731 tree rhs = NULL;
732 int i;
734 for (i = 0; i < PHI_NUM_ARGS (phi); i++)
736 tree t = PHI_ARG_DEF (phi, i);
738 /* Ignore alternatives which are the same as our LHS. Since
739 LHS is a PHI_RESULT, it is known to be a SSA_NAME, so we
740 can simply compare pointers. */
741 if (lhs == t)
742 continue;
744 /* If we have not processed an alternative yet, then set
745 RHS to this alternative. */
746 if (rhs == NULL)
747 rhs = t;
748 /* If we have processed an alternative (stored in RHS), then
749 see if it is equal to this one. If it isn't, then stop
750 the search. */
751 else if (! operand_equal_for_phi_arg_p (rhs, t))
752 break;
755 /* If we had no interesting alternatives, then all the RHS alternatives
756 must have been the same as LHS. */
757 if (!rhs)
758 rhs = lhs;
760 /* If we managed to iterate through each PHI alternative without
761 breaking out of the loop, then we have a PHI which may create
762 a useful equivalence. We do not need to record unwind data for
763 this, since this is a true assignment and not an equivalence
764 inferred from a comparison. All uses of this ssa name are dominated
765 by this assignment, so unwinding just costs time and space. */
766 if (i == PHI_NUM_ARGS (phi)
767 && may_propagate_copy (lhs, rhs))
768 SSA_NAME_VALUE (lhs) = rhs;
772 /* Ignoring loop backedges, if BB has precisely one incoming edge then
773 return that edge. Otherwise return NULL. */
774 static edge
775 single_incoming_edge_ignoring_loop_edges (basic_block bb)
777 edge retval = NULL;
778 edge e;
779 edge_iterator ei;
781 FOR_EACH_EDGE (e, ei, bb->preds)
783 /* A loop back edge can be identified by the destination of
784 the edge dominating the source of the edge. */
785 if (dominated_by_p (CDI_DOMINATORS, e->src, e->dest))
786 continue;
788 /* If we have already seen a non-loop edge, then we must have
789 multiple incoming non-loop edges and thus we return NULL. */
790 if (retval)
791 return NULL;
793 /* This is the first non-loop incoming edge we have found. Record
794 it. */
795 retval = e;
798 return retval;
801 /* Record any equivalences created by the incoming edge to BB. If BB
802 has more than one incoming edge, then no equivalence is created. */
804 static void
805 record_equivalences_from_incoming_edge (basic_block bb)
807 edge e;
808 basic_block parent;
809 struct edge_info *edge_info;
811 /* If our parent block ended with a control statement, then we may be
812 able to record some equivalences based on which outgoing edge from
813 the parent was followed. */
814 parent = get_immediate_dominator (CDI_DOMINATORS, bb);
816 e = single_incoming_edge_ignoring_loop_edges (bb);
818 /* If we had a single incoming edge from our parent block, then enter
819 any data associated with the edge into our tables. */
820 if (e && e->src == parent)
822 unsigned int i;
824 edge_info = (struct edge_info *) e->aux;
826 if (edge_info)
828 tree lhs = edge_info->lhs;
829 tree rhs = edge_info->rhs;
830 tree *cond_equivalences = edge_info->cond_equivalences;
832 if (lhs)
833 record_equality (lhs, rhs);
835 if (cond_equivalences)
837 for (i = 0; i < edge_info->max_cond_equivalences; i += 2)
839 tree expr = cond_equivalences[i];
840 tree value = cond_equivalences[i + 1];
842 record_cond (expr, value);
849 /* Dump SSA statistics on FILE. */
851 void
852 dump_dominator_optimization_stats (FILE *file)
854 long n_exprs;
856 fprintf (file, "Total number of statements: %6ld\n\n",
857 opt_stats.num_stmts);
858 fprintf (file, "Exprs considered for dominator optimizations: %6ld\n",
859 opt_stats.num_exprs_considered);
861 n_exprs = opt_stats.num_exprs_considered;
862 if (n_exprs == 0)
863 n_exprs = 1;
865 fprintf (file, " Redundant expressions eliminated: %6ld (%.0f%%)\n",
866 opt_stats.num_re, PERCENT (opt_stats.num_re,
867 n_exprs));
868 fprintf (file, " Constants propagated: %6ld\n",
869 opt_stats.num_const_prop);
870 fprintf (file, " Copies propagated: %6ld\n",
871 opt_stats.num_copy_prop);
873 fprintf (file, "\nHash table statistics:\n");
875 fprintf (file, " avail_exprs: ");
876 htab_statistics (file, avail_exprs);
880 /* Dump SSA statistics on stderr. */
882 void
883 debug_dominator_optimization_stats (void)
885 dump_dominator_optimization_stats (stderr);
889 /* Dump statistics for the hash table HTAB. */
891 static void
892 htab_statistics (FILE *file, htab_t htab)
894 fprintf (file, "size %ld, %ld elements, %f collision/search ratio\n",
895 (long) htab_size (htab),
896 (long) htab_elements (htab),
897 htab_collisions (htab));
900 /* Enter a statement into the true/false expression hash table indicating
901 that the condition COND has the value VALUE. */
903 static void
904 record_cond (tree cond, tree value)
906 struct expr_hash_elt *element = XCNEW (struct expr_hash_elt);
907 void **slot;
909 initialize_hash_element (cond, value, element);
911 slot = htab_find_slot_with_hash (avail_exprs, (void *)element,
912 element->hash, INSERT);
913 if (*slot == NULL)
915 *slot = (void *) element;
916 VEC_safe_push (tree, heap, avail_exprs_stack, cond);
918 else
919 free (element);
922 /* Build a new conditional using NEW_CODE, OP0 and OP1 and store
923 the new conditional into *p, then store a boolean_true_node
924 into *(p + 1). */
926 static void
927 build_and_record_new_cond (enum tree_code new_code, tree op0, tree op1, tree *p)
929 *p = build2 (new_code, boolean_type_node, op0, op1);
930 p++;
931 *p = boolean_true_node;
934 /* Record that COND is true and INVERTED is false into the edge information
935 structure. Also record that any conditions dominated by COND are true
936 as well.
938 For example, if a < b is true, then a <= b must also be true. */
940 static void
941 record_conditions (struct edge_info *edge_info, tree cond, tree inverted)
943 tree op0, op1;
945 if (!COMPARISON_CLASS_P (cond))
946 return;
948 op0 = TREE_OPERAND (cond, 0);
949 op1 = TREE_OPERAND (cond, 1);
951 switch (TREE_CODE (cond))
953 case LT_EXPR:
954 case GT_EXPR:
955 if (FLOAT_TYPE_P (TREE_TYPE (op0)))
957 edge_info->max_cond_equivalences = 12;
958 edge_info->cond_equivalences = XNEWVEC (tree, 12);
959 build_and_record_new_cond (ORDERED_EXPR, op0, op1,
960 &edge_info->cond_equivalences[8]);
961 build_and_record_new_cond (LTGT_EXPR, op0, op1,
962 &edge_info->cond_equivalences[10]);
964 else
966 edge_info->max_cond_equivalences = 8;
967 edge_info->cond_equivalences = XNEWVEC (tree, 8);
970 build_and_record_new_cond ((TREE_CODE (cond) == LT_EXPR
971 ? LE_EXPR : GE_EXPR),
972 op0, op1, &edge_info->cond_equivalences[4]);
973 build_and_record_new_cond (NE_EXPR, op0, op1,
974 &edge_info->cond_equivalences[6]);
975 break;
977 case GE_EXPR:
978 case LE_EXPR:
979 if (FLOAT_TYPE_P (TREE_TYPE (op0)))
981 edge_info->max_cond_equivalences = 6;
982 edge_info->cond_equivalences = XNEWVEC (tree, 6);
983 build_and_record_new_cond (ORDERED_EXPR, op0, op1,
984 &edge_info->cond_equivalences[4]);
986 else
988 edge_info->max_cond_equivalences = 4;
989 edge_info->cond_equivalences = XNEWVEC (tree, 4);
991 break;
993 case EQ_EXPR:
994 if (FLOAT_TYPE_P (TREE_TYPE (op0)))
996 edge_info->max_cond_equivalences = 10;
997 edge_info->cond_equivalences = XNEWVEC (tree, 10);
998 build_and_record_new_cond (ORDERED_EXPR, op0, op1,
999 &edge_info->cond_equivalences[8]);
1001 else
1003 edge_info->max_cond_equivalences = 8;
1004 edge_info->cond_equivalences = XNEWVEC (tree, 8);
1006 build_and_record_new_cond (LE_EXPR, op0, op1,
1007 &edge_info->cond_equivalences[4]);
1008 build_and_record_new_cond (GE_EXPR, op0, op1,
1009 &edge_info->cond_equivalences[6]);
1010 break;
1012 case UNORDERED_EXPR:
1013 edge_info->max_cond_equivalences = 16;
1014 edge_info->cond_equivalences = XNEWVEC (tree, 16);
1015 build_and_record_new_cond (NE_EXPR, op0, op1,
1016 &edge_info->cond_equivalences[4]);
1017 build_and_record_new_cond (UNLE_EXPR, op0, op1,
1018 &edge_info->cond_equivalences[6]);
1019 build_and_record_new_cond (UNGE_EXPR, op0, op1,
1020 &edge_info->cond_equivalences[8]);
1021 build_and_record_new_cond (UNEQ_EXPR, op0, op1,
1022 &edge_info->cond_equivalences[10]);
1023 build_and_record_new_cond (UNLT_EXPR, op0, op1,
1024 &edge_info->cond_equivalences[12]);
1025 build_and_record_new_cond (UNGT_EXPR, op0, op1,
1026 &edge_info->cond_equivalences[14]);
1027 break;
1029 case UNLT_EXPR:
1030 case UNGT_EXPR:
1031 edge_info->max_cond_equivalences = 8;
1032 edge_info->cond_equivalences = XNEWVEC (tree, 8);
1033 build_and_record_new_cond ((TREE_CODE (cond) == UNLT_EXPR
1034 ? UNLE_EXPR : UNGE_EXPR),
1035 op0, op1, &edge_info->cond_equivalences[4]);
1036 build_and_record_new_cond (NE_EXPR, op0, op1,
1037 &edge_info->cond_equivalences[6]);
1038 break;
1040 case UNEQ_EXPR:
1041 edge_info->max_cond_equivalences = 8;
1042 edge_info->cond_equivalences = XNEWVEC (tree, 8);
1043 build_and_record_new_cond (UNLE_EXPR, op0, op1,
1044 &edge_info->cond_equivalences[4]);
1045 build_and_record_new_cond (UNGE_EXPR, op0, op1,
1046 &edge_info->cond_equivalences[6]);
1047 break;
1049 case LTGT_EXPR:
1050 edge_info->max_cond_equivalences = 8;
1051 edge_info->cond_equivalences = XNEWVEC (tree, 8);
1052 build_and_record_new_cond (NE_EXPR, op0, op1,
1053 &edge_info->cond_equivalences[4]);
1054 build_and_record_new_cond (ORDERED_EXPR, op0, op1,
1055 &edge_info->cond_equivalences[6]);
1056 break;
1058 default:
1059 edge_info->max_cond_equivalences = 4;
1060 edge_info->cond_equivalences = XNEWVEC (tree, 4);
1061 break;
1064 /* Now store the original true and false conditions into the first
1065 two slots. */
1066 edge_info->cond_equivalences[0] = cond;
1067 edge_info->cond_equivalences[1] = boolean_true_node;
1068 edge_info->cond_equivalences[2] = inverted;
1069 edge_info->cond_equivalences[3] = boolean_false_node;
1072 /* A helper function for record_const_or_copy and record_equality.
1073 Do the work of recording the value and undo info. */
1075 static void
1076 record_const_or_copy_1 (tree x, tree y, tree prev_x)
1078 SSA_NAME_VALUE (x) = y;
1080 VEC_reserve (tree, heap, const_and_copies_stack, 2);
1081 VEC_quick_push (tree, const_and_copies_stack, prev_x);
1082 VEC_quick_push (tree, const_and_copies_stack, x);
1086 /* Return the loop depth of the basic block of the defining statement of X.
1087 This number should not be treated as absolutely correct because the loop
1088 information may not be completely up-to-date when dom runs. However, it
1089 will be relatively correct, and as more passes are taught to keep loop info
1090 up to date, the result will become more and more accurate. */
1093 loop_depth_of_name (tree x)
1095 tree defstmt;
1096 basic_block defbb;
1098 /* If it's not an SSA_NAME, we have no clue where the definition is. */
1099 if (TREE_CODE (x) != SSA_NAME)
1100 return 0;
1102 /* Otherwise return the loop depth of the defining statement's bb.
1103 Note that there may not actually be a bb for this statement, if the
1104 ssa_name is live on entry. */
1105 defstmt = SSA_NAME_DEF_STMT (x);
1106 defbb = bb_for_stmt (defstmt);
1107 if (!defbb)
1108 return 0;
1110 return defbb->loop_depth;
1114 /* Record that X is equal to Y in const_and_copies. Record undo
1115 information in the block-local vector. */
1117 static void
1118 record_const_or_copy (tree x, tree y)
1120 tree prev_x = SSA_NAME_VALUE (x);
1122 if (TREE_CODE (y) == SSA_NAME)
1124 tree tmp = SSA_NAME_VALUE (y);
1125 if (tmp)
1126 y = tmp;
1129 record_const_or_copy_1 (x, y, prev_x);
1132 /* Similarly, but assume that X and Y are the two operands of an EQ_EXPR.
1133 This constrains the cases in which we may treat this as assignment. */
1135 static void
1136 record_equality (tree x, tree y)
1138 tree prev_x = NULL, prev_y = NULL;
1140 if (TREE_CODE (x) == SSA_NAME)
1141 prev_x = SSA_NAME_VALUE (x);
1142 if (TREE_CODE (y) == SSA_NAME)
1143 prev_y = SSA_NAME_VALUE (y);
1145 /* If one of the previous values is invariant, or invariant in more loops
1146 (by depth), then use that.
1147 Otherwise it doesn't matter which value we choose, just so
1148 long as we canonicalize on one value. */
1149 if (is_gimple_min_invariant (y))
1151 else if (is_gimple_min_invariant (x)
1152 || (loop_depth_of_name (x) <= loop_depth_of_name (y)))
1153 prev_x = x, x = y, y = prev_x, prev_x = prev_y;
1154 else if (prev_x && is_gimple_min_invariant (prev_x))
1155 x = y, y = prev_x, prev_x = prev_y;
1156 else if (prev_y && TREE_CODE (prev_y) != VALUE_HANDLE)
1157 y = prev_y;
1159 /* After the swapping, we must have one SSA_NAME. */
1160 if (TREE_CODE (x) != SSA_NAME)
1161 return;
1163 /* For IEEE, -0.0 == 0.0, so we don't necessarily know the sign of a
1164 variable compared against zero. If we're honoring signed zeros,
1165 then we cannot record this value unless we know that the value is
1166 nonzero. */
1167 if (HONOR_SIGNED_ZEROS (TYPE_MODE (TREE_TYPE (x)))
1168 && (TREE_CODE (y) != REAL_CST
1169 || REAL_VALUES_EQUAL (dconst0, TREE_REAL_CST (y))))
1170 return;
1172 record_const_or_copy_1 (x, y, prev_x);
1175 /* Returns true when STMT is a simple iv increment. It detects the
1176 following situation:
1178 i_1 = phi (..., i_2)
1179 i_2 = i_1 +/- ... */
1181 static bool
1182 simple_iv_increment_p (tree stmt)
1184 tree lhs, rhs, preinc, phi;
1185 unsigned i;
1187 if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
1188 return false;
1190 lhs = GIMPLE_STMT_OPERAND (stmt, 0);
1191 if (TREE_CODE (lhs) != SSA_NAME)
1192 return false;
1194 rhs = GIMPLE_STMT_OPERAND (stmt, 1);
1196 if (TREE_CODE (rhs) != PLUS_EXPR
1197 && TREE_CODE (rhs) != MINUS_EXPR)
1198 return false;
1200 preinc = TREE_OPERAND (rhs, 0);
1201 if (TREE_CODE (preinc) != SSA_NAME)
1202 return false;
1204 phi = SSA_NAME_DEF_STMT (preinc);
1205 if (TREE_CODE (phi) != PHI_NODE)
1206 return false;
1208 for (i = 0; i < (unsigned) PHI_NUM_ARGS (phi); i++)
1209 if (PHI_ARG_DEF (phi, i) == lhs)
1210 return true;
1212 return false;
1215 /* CONST_AND_COPIES is a table which maps an SSA_NAME to the current
1216 known value for that SSA_NAME (or NULL if no value is known).
1218 Propagate values from CONST_AND_COPIES into the PHI nodes of the
1219 successors of BB. */
1221 static void
1222 cprop_into_successor_phis (basic_block bb)
1224 edge e;
1225 edge_iterator ei;
1227 FOR_EACH_EDGE (e, ei, bb->succs)
1229 tree phi;
1230 int indx;
1232 /* If this is an abnormal edge, then we do not want to copy propagate
1233 into the PHI alternative associated with this edge. */
1234 if (e->flags & EDGE_ABNORMAL)
1235 continue;
1237 phi = phi_nodes (e->dest);
1238 if (! phi)
1239 continue;
1241 indx = e->dest_idx;
1242 for ( ; phi; phi = PHI_CHAIN (phi))
1244 tree new_val;
1245 use_operand_p orig_p;
1246 tree orig_val;
1248 /* The alternative may be associated with a constant, so verify
1249 it is an SSA_NAME before doing anything with it. */
1250 orig_p = PHI_ARG_DEF_PTR (phi, indx);
1251 orig_val = USE_FROM_PTR (orig_p);
1252 if (TREE_CODE (orig_val) != SSA_NAME)
1253 continue;
1255 /* If we have *ORIG_P in our constant/copy table, then replace
1256 ORIG_P with its value in our constant/copy table. */
1257 new_val = SSA_NAME_VALUE (orig_val);
1258 if (new_val
1259 && new_val != orig_val
1260 && (TREE_CODE (new_val) == SSA_NAME
1261 || is_gimple_min_invariant (new_val))
1262 && may_propagate_copy (orig_val, new_val))
1263 propagate_value (orig_p, new_val);
1268 /* We have finished optimizing BB, record any information implied by
1269 taking a specific outgoing edge from BB. */
1271 static void
1272 record_edge_info (basic_block bb)
1274 block_stmt_iterator bsi = bsi_last (bb);
1275 struct edge_info *edge_info;
1277 if (! bsi_end_p (bsi))
1279 tree stmt = bsi_stmt (bsi);
1281 if (stmt && TREE_CODE (stmt) == SWITCH_EXPR)
1283 tree cond = SWITCH_COND (stmt);
1285 if (TREE_CODE (cond) == SSA_NAME)
1287 tree labels = SWITCH_LABELS (stmt);
1288 int i, n_labels = TREE_VEC_LENGTH (labels);
1289 tree *info = XCNEWVEC (tree, last_basic_block);
1290 edge e;
1291 edge_iterator ei;
1293 for (i = 0; i < n_labels; i++)
1295 tree label = TREE_VEC_ELT (labels, i);
1296 basic_block target_bb = label_to_block (CASE_LABEL (label));
1298 if (CASE_HIGH (label)
1299 || !CASE_LOW (label)
1300 || info[target_bb->index])
1301 info[target_bb->index] = error_mark_node;
1302 else
1303 info[target_bb->index] = label;
1306 FOR_EACH_EDGE (e, ei, bb->succs)
1308 basic_block target_bb = e->dest;
1309 tree node = info[target_bb->index];
1311 if (node != NULL && node != error_mark_node)
1313 tree x = fold_convert (TREE_TYPE (cond), CASE_LOW (node));
1314 edge_info = allocate_edge_info (e);
1315 edge_info->lhs = cond;
1316 edge_info->rhs = x;
1319 free (info);
1323 /* A COND_EXPR may create equivalences too. */
1324 if (stmt && TREE_CODE (stmt) == COND_EXPR)
1326 tree cond = COND_EXPR_COND (stmt);
1327 edge true_edge;
1328 edge false_edge;
1330 extract_true_false_edges_from_block (bb, &true_edge, &false_edge);
1332 /* If the conditional is a single variable 'X', record 'X = 1'
1333 for the true edge and 'X = 0' on the false edge. */
1334 if (SSA_VAR_P (cond))
1336 struct edge_info *edge_info;
1338 edge_info = allocate_edge_info (true_edge);
1339 edge_info->lhs = cond;
1340 edge_info->rhs = constant_boolean_node (1, TREE_TYPE (cond));
1342 edge_info = allocate_edge_info (false_edge);
1343 edge_info->lhs = cond;
1344 edge_info->rhs = constant_boolean_node (0, TREE_TYPE (cond));
1346 /* Equality tests may create one or two equivalences. */
1347 else if (COMPARISON_CLASS_P (cond))
1349 tree op0 = TREE_OPERAND (cond, 0);
1350 tree op1 = TREE_OPERAND (cond, 1);
1352 /* Special case comparing booleans against a constant as we
1353 know the value of OP0 on both arms of the branch. i.e., we
1354 can record an equivalence for OP0 rather than COND. */
1355 if ((TREE_CODE (cond) == EQ_EXPR || TREE_CODE (cond) == NE_EXPR)
1356 && TREE_CODE (op0) == SSA_NAME
1357 && TREE_CODE (TREE_TYPE (op0)) == BOOLEAN_TYPE
1358 && is_gimple_min_invariant (op1))
1360 if (TREE_CODE (cond) == EQ_EXPR)
1362 edge_info = allocate_edge_info (true_edge);
1363 edge_info->lhs = op0;
1364 edge_info->rhs = (integer_zerop (op1)
1365 ? boolean_false_node
1366 : boolean_true_node);
1368 edge_info = allocate_edge_info (false_edge);
1369 edge_info->lhs = op0;
1370 edge_info->rhs = (integer_zerop (op1)
1371 ? boolean_true_node
1372 : boolean_false_node);
1374 else
1376 edge_info = allocate_edge_info (true_edge);
1377 edge_info->lhs = op0;
1378 edge_info->rhs = (integer_zerop (op1)
1379 ? boolean_true_node
1380 : boolean_false_node);
1382 edge_info = allocate_edge_info (false_edge);
1383 edge_info->lhs = op0;
1384 edge_info->rhs = (integer_zerop (op1)
1385 ? boolean_false_node
1386 : boolean_true_node);
1390 else if (is_gimple_min_invariant (op0)
1391 && (TREE_CODE (op1) == SSA_NAME
1392 || is_gimple_min_invariant (op1)))
1394 tree inverted = invert_truthvalue (cond);
1395 struct edge_info *edge_info;
1397 edge_info = allocate_edge_info (true_edge);
1398 record_conditions (edge_info, cond, inverted);
1400 if (TREE_CODE (cond) == EQ_EXPR)
1402 edge_info->lhs = op1;
1403 edge_info->rhs = op0;
1406 edge_info = allocate_edge_info (false_edge);
1407 record_conditions (edge_info, inverted, cond);
1409 if (TREE_CODE (cond) == NE_EXPR)
1411 edge_info->lhs = op1;
1412 edge_info->rhs = op0;
1416 else if (TREE_CODE (op0) == SSA_NAME
1417 && (is_gimple_min_invariant (op1)
1418 || TREE_CODE (op1) == SSA_NAME))
1420 tree inverted = invert_truthvalue (cond);
1421 struct edge_info *edge_info;
1423 edge_info = allocate_edge_info (true_edge);
1424 record_conditions (edge_info, cond, inverted);
1426 if (TREE_CODE (cond) == EQ_EXPR)
1428 edge_info->lhs = op0;
1429 edge_info->rhs = op1;
1432 edge_info = allocate_edge_info (false_edge);
1433 record_conditions (edge_info, inverted, cond);
1435 if (TREE_CODE (cond) == NE_EXPR)
1437 edge_info->lhs = op0;
1438 edge_info->rhs = op1;
1443 /* ??? TRUTH_NOT_EXPR can create an equivalence too. */
1448 /* Propagate information from BB to its outgoing edges.
1450 This can include equivalency information implied by control statements
1451 at the end of BB and const/copy propagation into PHIs in BB's
1452 successor blocks. */
1454 static void
1455 propagate_to_outgoing_edges (struct dom_walk_data *walk_data ATTRIBUTE_UNUSED,
1456 basic_block bb)
1458 record_edge_info (bb);
1459 cprop_into_successor_phis (bb);
1462 /* Search for redundant computations in STMT. If any are found, then
1463 replace them with the variable holding the result of the computation.
1465 If safe, record this expression into the available expression hash
1466 table. */
1468 static bool
1469 eliminate_redundant_computations (tree stmt)
1471 tree *expr_p, def = NULL_TREE;
1472 bool insert = true;
1473 tree cached_lhs;
1474 bool retval = false;
1475 bool modify_expr_p = false;
1477 if (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT)
1478 def = GIMPLE_STMT_OPERAND (stmt, 0);
1480 /* Certain expressions on the RHS can be optimized away, but can not
1481 themselves be entered into the hash tables. */
1482 if (! def
1483 || TREE_CODE (def) != SSA_NAME
1484 || SSA_NAME_OCCURS_IN_ABNORMAL_PHI (def)
1485 || !ZERO_SSA_OPERANDS (stmt, SSA_OP_VDEF)
1486 /* Do not record equivalences for increments of ivs. This would create
1487 overlapping live ranges for a very questionable gain. */
1488 || simple_iv_increment_p (stmt))
1489 insert = false;
1491 /* Check if the expression has been computed before. */
1492 cached_lhs = lookup_avail_expr (stmt, insert);
1494 opt_stats.num_exprs_considered++;
1496 /* Get a pointer to the expression we are trying to optimize. */
1497 if (TREE_CODE (stmt) == COND_EXPR)
1498 expr_p = &COND_EXPR_COND (stmt);
1499 else if (TREE_CODE (stmt) == SWITCH_EXPR)
1500 expr_p = &SWITCH_COND (stmt);
1501 else if (TREE_CODE (stmt) == RETURN_EXPR && TREE_OPERAND (stmt, 0))
1503 expr_p = &GIMPLE_STMT_OPERAND (TREE_OPERAND (stmt, 0), 1);
1504 modify_expr_p = true;
1506 else
1508 expr_p = &GENERIC_TREE_OPERAND (stmt, 1);
1509 modify_expr_p = true;
1512 /* It is safe to ignore types here since we have already done
1513 type checking in the hashing and equality routines. In fact
1514 type checking here merely gets in the way of constant
1515 propagation. Also, make sure that it is safe to propagate
1516 CACHED_LHS into *EXPR_P. */
1517 if (cached_lhs
1518 && ((TREE_CODE (cached_lhs) != SSA_NAME
1519 && (modify_expr_p
1520 || useless_type_conversion_p (TREE_TYPE (*expr_p),
1521 TREE_TYPE (cached_lhs))))
1522 || may_propagate_copy (*expr_p, cached_lhs)))
1524 if (dump_file && (dump_flags & TDF_DETAILS))
1526 fprintf (dump_file, " Replaced redundant expr '");
1527 print_generic_expr (dump_file, *expr_p, dump_flags);
1528 fprintf (dump_file, "' with '");
1529 print_generic_expr (dump_file, cached_lhs, dump_flags);
1530 fprintf (dump_file, "'\n");
1533 opt_stats.num_re++;
1535 #if defined ENABLE_CHECKING
1536 gcc_assert (TREE_CODE (cached_lhs) == SSA_NAME
1537 || is_gimple_min_invariant (cached_lhs));
1538 #endif
1540 if (TREE_CODE (cached_lhs) == ADDR_EXPR
1541 || (POINTER_TYPE_P (TREE_TYPE (*expr_p))
1542 && is_gimple_min_invariant (cached_lhs)))
1543 retval = true;
1545 if (modify_expr_p
1546 && !useless_type_conversion_p (TREE_TYPE (*expr_p),
1547 TREE_TYPE (cached_lhs)))
1548 cached_lhs = fold_convert (TREE_TYPE (*expr_p), cached_lhs);
1550 propagate_tree_value (expr_p, cached_lhs);
1551 mark_stmt_modified (stmt);
1553 return retval;
1556 /* STMT, a GIMPLE_MODIFY_STMT, may create certain equivalences, in either
1557 the available expressions table or the const_and_copies table.
1558 Detect and record those equivalences. */
1560 static void
1561 record_equivalences_from_stmt (tree stmt, int may_optimize_p, stmt_ann_t ann)
1563 tree lhs = GIMPLE_STMT_OPERAND (stmt, 0);
1564 enum tree_code lhs_code = TREE_CODE (lhs);
1566 if (lhs_code == SSA_NAME)
1568 tree rhs = GIMPLE_STMT_OPERAND (stmt, 1);
1570 /* Strip away any useless type conversions. */
1571 STRIP_USELESS_TYPE_CONVERSION (rhs);
1573 /* If the RHS of the assignment is a constant or another variable that
1574 may be propagated, register it in the CONST_AND_COPIES table. We
1575 do not need to record unwind data for this, since this is a true
1576 assignment and not an equivalence inferred from a comparison. All
1577 uses of this ssa name are dominated by this assignment, so unwinding
1578 just costs time and space. */
1579 if (may_optimize_p
1580 && (TREE_CODE (rhs) == SSA_NAME
1581 || is_gimple_min_invariant (rhs)))
1582 SSA_NAME_VALUE (lhs) = rhs;
1585 /* A memory store, even an aliased store, creates a useful
1586 equivalence. By exchanging the LHS and RHS, creating suitable
1587 vops and recording the result in the available expression table,
1588 we may be able to expose more redundant loads. */
1589 if (!ann->has_volatile_ops
1590 && stmt_references_memory_p (stmt)
1591 && (TREE_CODE (GIMPLE_STMT_OPERAND (stmt, 1)) == SSA_NAME
1592 || is_gimple_min_invariant (GIMPLE_STMT_OPERAND (stmt, 1)))
1593 && !is_gimple_reg (lhs))
1595 tree rhs = GIMPLE_STMT_OPERAND (stmt, 1);
1596 tree new_stmt;
1598 /* FIXME: If the LHS of the assignment is a bitfield and the RHS
1599 is a constant, we need to adjust the constant to fit into the
1600 type of the LHS. If the LHS is a bitfield and the RHS is not
1601 a constant, then we can not record any equivalences for this
1602 statement since we would need to represent the widening or
1603 narrowing of RHS. This fixes gcc.c-torture/execute/921016-1.c
1604 and should not be necessary if GCC represented bitfields
1605 properly. */
1606 if (lhs_code == COMPONENT_REF
1607 && DECL_BIT_FIELD (TREE_OPERAND (lhs, 1)))
1609 if (TREE_CONSTANT (rhs))
1610 rhs = widen_bitfield (rhs, TREE_OPERAND (lhs, 1), lhs);
1611 else
1612 rhs = NULL;
1614 /* If the value overflowed, then we can not use this equivalence. */
1615 if (rhs && ! is_gimple_min_invariant (rhs))
1616 rhs = NULL;
1619 if (rhs)
1621 /* Build a new statement with the RHS and LHS exchanged. */
1622 new_stmt = build_gimple_modify_stmt (rhs, lhs);
1624 create_ssa_artificial_load_stmt (new_stmt, stmt, true);
1626 /* Finally enter the statement into the available expression
1627 table. */
1628 lookup_avail_expr (new_stmt, true);
1633 /* Replace *OP_P in STMT with any known equivalent value for *OP_P from
1634 CONST_AND_COPIES. */
1636 static bool
1637 cprop_operand (tree stmt, use_operand_p op_p)
1639 bool may_have_exposed_new_symbols = false;
1640 tree val;
1641 tree op = USE_FROM_PTR (op_p);
1643 /* If the operand has a known constant value or it is known to be a
1644 copy of some other variable, use the value or copy stored in
1645 CONST_AND_COPIES. */
1646 val = SSA_NAME_VALUE (op);
1647 if (val && val != op && TREE_CODE (val) != VALUE_HANDLE)
1649 tree op_type, val_type;
1651 /* Do not change the base variable in the virtual operand
1652 tables. That would make it impossible to reconstruct
1653 the renamed virtual operand if we later modify this
1654 statement. Also only allow the new value to be an SSA_NAME
1655 for propagation into virtual operands. */
1656 if (!is_gimple_reg (op)
1657 && (TREE_CODE (val) != SSA_NAME
1658 || is_gimple_reg (val)
1659 || get_virtual_var (val) != get_virtual_var (op)))
1660 return false;
1662 /* Do not replace hard register operands in asm statements. */
1663 if (TREE_CODE (stmt) == ASM_EXPR
1664 && !may_propagate_copy_into_asm (op))
1665 return false;
1667 /* Get the toplevel type of each operand. */
1668 op_type = TREE_TYPE (op);
1669 val_type = TREE_TYPE (val);
1671 /* While both types are pointers, get the type of the object
1672 pointed to. */
1673 while (POINTER_TYPE_P (op_type) && POINTER_TYPE_P (val_type))
1675 op_type = TREE_TYPE (op_type);
1676 val_type = TREE_TYPE (val_type);
1679 /* Make sure underlying types match before propagating a constant by
1680 converting the constant to the proper type. Note that convert may
1681 return a non-gimple expression, in which case we ignore this
1682 propagation opportunity. */
1683 if (TREE_CODE (val) != SSA_NAME)
1685 if (!useless_type_conversion_p (op_type, val_type))
1687 val = fold_convert (TREE_TYPE (op), val);
1688 if (!is_gimple_min_invariant (val))
1689 return false;
1693 /* Certain operands are not allowed to be copy propagated due
1694 to their interaction with exception handling and some GCC
1695 extensions. */
1696 else if (!may_propagate_copy (op, val))
1697 return false;
1699 /* Do not propagate copies if the propagated value is at a deeper loop
1700 depth than the propagatee. Otherwise, this may move loop variant
1701 variables outside of their loops and prevent coalescing
1702 opportunities. If the value was loop invariant, it will be hoisted
1703 by LICM and exposed for copy propagation. */
1704 if (loop_depth_of_name (val) > loop_depth_of_name (op))
1705 return false;
1707 /* Dump details. */
1708 if (dump_file && (dump_flags & TDF_DETAILS))
1710 fprintf (dump_file, " Replaced '");
1711 print_generic_expr (dump_file, op, dump_flags);
1712 fprintf (dump_file, "' with %s '",
1713 (TREE_CODE (val) != SSA_NAME ? "constant" : "variable"));
1714 print_generic_expr (dump_file, val, dump_flags);
1715 fprintf (dump_file, "'\n");
1718 /* If VAL is an ADDR_EXPR or a constant of pointer type, note
1719 that we may have exposed a new symbol for SSA renaming. */
1720 if (TREE_CODE (val) == ADDR_EXPR
1721 || (POINTER_TYPE_P (TREE_TYPE (op))
1722 && is_gimple_min_invariant (val)))
1723 may_have_exposed_new_symbols = true;
1725 if (TREE_CODE (val) != SSA_NAME)
1726 opt_stats.num_const_prop++;
1727 else
1728 opt_stats.num_copy_prop++;
1730 propagate_value (op_p, val);
1732 /* And note that we modified this statement. This is now
1733 safe, even if we changed virtual operands since we will
1734 rescan the statement and rewrite its operands again. */
1735 mark_stmt_modified (stmt);
1737 return may_have_exposed_new_symbols;
1740 /* CONST_AND_COPIES is a table which maps an SSA_NAME to the current
1741 known value for that SSA_NAME (or NULL if no value is known).
1743 Propagate values from CONST_AND_COPIES into the uses, vuses and
1744 vdef_ops of STMT. */
1746 static bool
1747 cprop_into_stmt (tree stmt)
1749 bool may_have_exposed_new_symbols = false;
1750 use_operand_p op_p;
1751 ssa_op_iter iter;
1753 FOR_EACH_SSA_USE_OPERAND (op_p, stmt, iter, SSA_OP_ALL_USES)
1755 if (TREE_CODE (USE_FROM_PTR (op_p)) == SSA_NAME)
1756 may_have_exposed_new_symbols |= cprop_operand (stmt, op_p);
1759 return may_have_exposed_new_symbols;
1763 /* Optimize the statement pointed to by iterator SI.
1765 We try to perform some simplistic global redundancy elimination and
1766 constant propagation:
1768 1- To detect global redundancy, we keep track of expressions that have
1769 been computed in this block and its dominators. If we find that the
1770 same expression is computed more than once, we eliminate repeated
1771 computations by using the target of the first one.
1773 2- Constant values and copy assignments. This is used to do very
1774 simplistic constant and copy propagation. When a constant or copy
1775 assignment is found, we map the value on the RHS of the assignment to
1776 the variable in the LHS in the CONST_AND_COPIES table. */
1778 static void
1779 optimize_stmt (struct dom_walk_data *walk_data ATTRIBUTE_UNUSED,
1780 basic_block bb, block_stmt_iterator si)
1782 stmt_ann_t ann;
1783 tree stmt, old_stmt;
1784 bool may_optimize_p;
1785 bool may_have_exposed_new_symbols = false;
1787 old_stmt = stmt = bsi_stmt (si);
1789 if (TREE_CODE (stmt) == COND_EXPR)
1790 canonicalize_comparison (stmt);
1792 update_stmt_if_modified (stmt);
1793 ann = stmt_ann (stmt);
1794 opt_stats.num_stmts++;
1795 may_have_exposed_new_symbols = false;
1796 push_stmt_changes (bsi_stmt_ptr (si));
1798 if (dump_file && (dump_flags & TDF_DETAILS))
1800 fprintf (dump_file, "Optimizing statement ");
1801 print_generic_stmt (dump_file, stmt, TDF_SLIM);
1804 /* Const/copy propagate into USES, VUSES and the RHS of VDEFs. */
1805 may_have_exposed_new_symbols = cprop_into_stmt (stmt);
1807 /* If the statement has been modified with constant replacements,
1808 fold its RHS before checking for redundant computations. */
1809 if (ann->modified)
1811 tree rhs;
1813 /* Try to fold the statement making sure that STMT is kept
1814 up to date. */
1815 if (fold_stmt (bsi_stmt_ptr (si)))
1817 stmt = bsi_stmt (si);
1818 ann = stmt_ann (stmt);
1820 if (dump_file && (dump_flags & TDF_DETAILS))
1822 fprintf (dump_file, " Folded to: ");
1823 print_generic_stmt (dump_file, stmt, TDF_SLIM);
1827 rhs = get_rhs (stmt);
1828 if (rhs && TREE_CODE (rhs) == ADDR_EXPR)
1829 recompute_tree_invariant_for_addr_expr (rhs);
1831 /* Constant/copy propagation above may change the set of
1832 virtual operands associated with this statement. Folding
1833 may remove the need for some virtual operands.
1835 Indicate we will need to rescan and rewrite the statement. */
1836 may_have_exposed_new_symbols = true;
1839 /* Check for redundant computations. Do this optimization only
1840 for assignments that have no volatile ops and conditionals. */
1841 may_optimize_p = (!ann->has_volatile_ops
1842 && ((TREE_CODE (stmt) == RETURN_EXPR
1843 && TREE_OPERAND (stmt, 0)
1844 && TREE_CODE (TREE_OPERAND (stmt, 0))
1845 == GIMPLE_MODIFY_STMT
1846 && ! (TREE_SIDE_EFFECTS
1847 (GIMPLE_STMT_OPERAND
1848 (TREE_OPERAND (stmt, 0), 1))))
1849 || (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT
1850 && ! TREE_SIDE_EFFECTS (GIMPLE_STMT_OPERAND (stmt,
1851 1)))
1852 || TREE_CODE (stmt) == COND_EXPR
1853 || TREE_CODE (stmt) == SWITCH_EXPR));
1855 if (may_optimize_p)
1856 may_have_exposed_new_symbols |= eliminate_redundant_computations (stmt);
1858 /* Record any additional equivalences created by this statement. */
1859 if (TREE_CODE (stmt) == GIMPLE_MODIFY_STMT)
1860 record_equivalences_from_stmt (stmt, may_optimize_p, ann);
1862 /* If STMT is a COND_EXPR and it was modified, then we may know
1863 where it goes. If that is the case, then mark the CFG as altered.
1865 This will cause us to later call remove_unreachable_blocks and
1866 cleanup_tree_cfg when it is safe to do so. It is not safe to
1867 clean things up here since removal of edges and such can trigger
1868 the removal of PHI nodes, which in turn can release SSA_NAMEs to
1869 the manager.
1871 That's all fine and good, except that once SSA_NAMEs are released
1872 to the manager, we must not call create_ssa_name until all references
1873 to released SSA_NAMEs have been eliminated.
1875 All references to the deleted SSA_NAMEs can not be eliminated until
1876 we remove unreachable blocks.
1878 We can not remove unreachable blocks until after we have completed
1879 any queued jump threading.
1881 We can not complete any queued jump threads until we have taken
1882 appropriate variables out of SSA form. Taking variables out of
1883 SSA form can call create_ssa_name and thus we lose.
1885 Ultimately I suspect we're going to need to change the interface
1886 into the SSA_NAME manager. */
1887 if (ann->modified)
1889 tree val = NULL;
1891 if (TREE_CODE (stmt) == COND_EXPR)
1892 val = COND_EXPR_COND (stmt);
1893 else if (TREE_CODE (stmt) == SWITCH_EXPR)
1894 val = SWITCH_COND (stmt);
1896 if (val && TREE_CODE (val) == INTEGER_CST && find_taken_edge (bb, val))
1897 cfg_altered = true;
1899 /* If we simplified a statement in such a way as to be shown that it
1900 cannot trap, update the eh information and the cfg to match. */
1901 if (maybe_clean_or_replace_eh_stmt (old_stmt, stmt))
1903 bitmap_set_bit (need_eh_cleanup, bb->index);
1904 if (dump_file && (dump_flags & TDF_DETAILS))
1905 fprintf (dump_file, " Flagged to clear EH edges.\n");
1909 if (may_have_exposed_new_symbols)
1911 /* Queue the statement to be re-scanned after all the
1912 AVAIL_EXPRS have been processed. The change buffer stack for
1913 all the pushed statements will be processed when this queue
1914 is emptied. */
1915 VEC_safe_push (tree_p, heap, stmts_to_rescan, bsi_stmt_ptr (si));
1917 else
1919 /* Otherwise, just discard the recently pushed change buffer. If
1920 not, the STMTS_TO_RESCAN queue will get out of synch with the
1921 change buffer stack. */
1922 discard_stmt_changes (bsi_stmt_ptr (si));
1926 /* Search for an existing instance of STMT in the AVAIL_EXPRS table. If
1927 found, return its LHS. Otherwise insert STMT in the table and return
1928 NULL_TREE.
1930 Also, when an expression is first inserted in the AVAIL_EXPRS table, it
1931 is also added to the stack pointed to by BLOCK_AVAIL_EXPRS_P, so that they
1932 can be removed when we finish processing this block and its children.
1934 NOTE: This function assumes that STMT is a GIMPLE_MODIFY_STMT node that
1935 contains no CALL_EXPR on its RHS and makes no volatile nor
1936 aliased references. */
1938 static tree
1939 lookup_avail_expr (tree stmt, bool insert)
1941 void **slot;
1942 tree lhs;
1943 tree temp;
1944 struct expr_hash_elt *element = XNEW (struct expr_hash_elt);
1946 lhs = TREE_CODE (stmt) == GIMPLE_MODIFY_STMT
1947 ? GIMPLE_STMT_OPERAND (stmt, 0) : NULL;
1949 initialize_hash_element (stmt, lhs, element);
1951 /* Don't bother remembering constant assignments and copy operations.
1952 Constants and copy operations are handled by the constant/copy propagator
1953 in optimize_stmt. */
1954 if (TREE_CODE (element->rhs) == SSA_NAME
1955 || is_gimple_min_invariant (element->rhs))
1957 free (element);
1958 return NULL_TREE;
1961 /* Finally try to find the expression in the main expression hash table. */
1962 slot = htab_find_slot_with_hash (avail_exprs, element, element->hash,
1963 (insert ? INSERT : NO_INSERT));
1964 if (slot == NULL)
1966 free (element);
1967 return NULL_TREE;
1970 if (*slot == NULL)
1972 *slot = (void *) element;
1973 VEC_safe_push (tree, heap, avail_exprs_stack,
1974 stmt ? stmt : element->rhs);
1975 return NULL_TREE;
1978 /* Extract the LHS of the assignment so that it can be used as the current
1979 definition of another variable. */
1980 lhs = ((struct expr_hash_elt *)*slot)->lhs;
1982 /* See if the LHS appears in the CONST_AND_COPIES table. If it does, then
1983 use the value from the const_and_copies table. */
1984 if (TREE_CODE (lhs) == SSA_NAME)
1986 temp = SSA_NAME_VALUE (lhs);
1987 if (temp && TREE_CODE (temp) != VALUE_HANDLE)
1988 lhs = temp;
1991 free (element);
1992 return lhs;
1995 /* Hashing and equality functions for AVAIL_EXPRS. The table stores
1996 GIMPLE_MODIFY_STMT statements. We compute a value number for expressions
1997 using the code of the expression and the SSA numbers of its operands. */
1999 static hashval_t
2000 avail_expr_hash (const void *p)
2002 tree stmt = ((const struct expr_hash_elt *)p)->stmt;
2003 tree rhs = ((const struct expr_hash_elt *)p)->rhs;
2004 tree vuse;
2005 ssa_op_iter iter;
2006 hashval_t val = 0;
2008 /* iterative_hash_expr knows how to deal with any expression and
2009 deals with commutative operators as well, so just use it instead
2010 of duplicating such complexities here. */
2011 val = iterative_hash_expr (rhs, val);
2013 /* If the hash table entry is not associated with a statement, then we
2014 can just hash the expression and not worry about virtual operands
2015 and such. */
2016 if (!stmt || !stmt_ann (stmt))
2017 return val;
2019 /* Add the SSA version numbers of every vuse operand. This is important
2020 because compound variables like arrays are not renamed in the
2021 operands. Rather, the rename is done on the virtual variable
2022 representing all the elements of the array. */
2023 FOR_EACH_SSA_TREE_OPERAND (vuse, stmt, iter, SSA_OP_VUSE)
2024 val = iterative_hash_expr (vuse, val);
2026 return val;
2029 static hashval_t
2030 real_avail_expr_hash (const void *p)
2032 return ((const struct expr_hash_elt *)p)->hash;
2035 static int
2036 avail_expr_eq (const void *p1, const void *p2)
2038 tree stmt1 = ((const struct expr_hash_elt *)p1)->stmt;
2039 tree rhs1 = ((const struct expr_hash_elt *)p1)->rhs;
2040 tree stmt2 = ((const struct expr_hash_elt *)p2)->stmt;
2041 tree rhs2 = ((const struct expr_hash_elt *)p2)->rhs;
2043 /* If they are the same physical expression, return true. */
2044 if (rhs1 == rhs2 && stmt1 == stmt2)
2045 return true;
2047 /* If their codes are not equal, then quit now. */
2048 if (TREE_CODE (rhs1) != TREE_CODE (rhs2))
2049 return false;
2051 /* In case of a collision, both RHS have to be identical and have the
2052 same VUSE operands. */
2053 if (types_compatible_p (TREE_TYPE (rhs1), TREE_TYPE (rhs2))
2054 && operand_equal_p (rhs1, rhs2, OEP_PURE_SAME))
2056 bool ret = compare_ssa_operands_equal (stmt1, stmt2, SSA_OP_VUSE);
2057 gcc_assert (!ret || ((const struct expr_hash_elt *)p1)->hash
2058 == ((const struct expr_hash_elt *)p2)->hash);
2059 return ret;
2062 return false;
2065 /* PHI-ONLY copy and constant propagation. This pass is meant to clean
2066 up degenerate PHIs created by or exposed by jump threading. */
2068 /* Given PHI, return its RHS if the PHI is a degenerate, otherwise return
2069 NULL. */
2071 static tree
2072 degenerate_phi_result (tree phi)
2074 tree lhs = PHI_RESULT (phi);
2075 tree val = NULL;
2076 int i;
2078 /* Ignoring arguments which are the same as LHS, if all the remaining
2079 arguments are the same, then the PHI is a degenerate and has the
2080 value of that common argument. */
2081 for (i = 0; i < PHI_NUM_ARGS (phi); i++)
2083 tree arg = PHI_ARG_DEF (phi, i);
2085 if (arg == lhs)
2086 continue;
2087 else if (!val)
2088 val = arg;
2089 else if (!operand_equal_p (arg, val, 0))
2090 break;
2092 return (i == PHI_NUM_ARGS (phi) ? val : NULL);
2095 /* Given a tree node T, which is either a PHI_NODE or GIMPLE_MODIFY_STMT,
2096 remove it from the IL. */
2098 static void
2099 remove_stmt_or_phi (tree t)
2101 if (TREE_CODE (t) == PHI_NODE)
2102 remove_phi_node (t, NULL, true);
2103 else
2105 block_stmt_iterator bsi = bsi_for_stmt (t);
2106 bsi_remove (&bsi, true);
2107 release_defs (t);
2111 /* Given a tree node T, which is either a PHI_NODE or GIMPLE_MODIFY_STMT,
2112 return the "rhs" of the node, in the case of a non-degenerate
2113 PHI, NULL is returned. */
2115 static tree
2116 get_rhs_or_phi_arg (tree t)
2118 if (TREE_CODE (t) == PHI_NODE)
2119 return degenerate_phi_result (t);
2120 else if (TREE_CODE (t) == GIMPLE_MODIFY_STMT)
2121 return GIMPLE_STMT_OPERAND (t, 1);
2122 gcc_unreachable ();
2126 /* Given a tree node T, which is either a PHI_NODE or a GIMPLE_MODIFY_STMT,
2127 return the "lhs" of the node. */
2129 static tree
2130 get_lhs_or_phi_result (tree t)
2132 if (TREE_CODE (t) == PHI_NODE)
2133 return PHI_RESULT (t);
2134 else if (TREE_CODE (t) == GIMPLE_MODIFY_STMT)
2135 return GIMPLE_STMT_OPERAND (t, 0);
2136 gcc_unreachable ();
2139 /* Propagate RHS into all uses of LHS (when possible).
2141 RHS and LHS are derived from STMT, which is passed in solely so
2142 that we can remove it if propagation is successful.
2144 When propagating into a PHI node or into a statement which turns
2145 into a trivial copy or constant initialization, set the
2146 appropriate bit in INTERESTING_NAMEs so that we will visit those
2147 nodes as well in an effort to pick up secondary optimization
2148 opportunities. */
2150 static void
2151 propagate_rhs_into_lhs (tree stmt, tree lhs, tree rhs, bitmap interesting_names)
2153 /* First verify that propagation is valid and isn't going to move a
2154 loop variant variable outside its loop. */
2155 if (! SSA_NAME_OCCURS_IN_ABNORMAL_PHI (lhs)
2156 && (TREE_CODE (rhs) != SSA_NAME
2157 || ! SSA_NAME_OCCURS_IN_ABNORMAL_PHI (rhs))
2158 && may_propagate_copy (lhs, rhs)
2159 && loop_depth_of_name (lhs) >= loop_depth_of_name (rhs))
2161 use_operand_p use_p;
2162 imm_use_iterator iter;
2163 tree use_stmt;
2164 bool all = true;
2166 /* Dump details. */
2167 if (dump_file && (dump_flags & TDF_DETAILS))
2169 fprintf (dump_file, " Replacing '");
2170 print_generic_expr (dump_file, lhs, dump_flags);
2171 fprintf (dump_file, "' with %s '",
2172 (TREE_CODE (rhs) != SSA_NAME ? "constant" : "variable"));
2173 print_generic_expr (dump_file, rhs, dump_flags);
2174 fprintf (dump_file, "'\n");
2177 /* Walk over every use of LHS and try to replace the use with RHS.
2178 At this point the only reason why such a propagation would not
2179 be successful would be if the use occurs in an ASM_EXPR. */
2180 FOR_EACH_IMM_USE_STMT (use_stmt, iter, lhs)
2183 /* It's not always safe to propagate into an ASM_EXPR. */
2184 if (TREE_CODE (use_stmt) == ASM_EXPR
2185 && ! may_propagate_copy_into_asm (lhs))
2187 all = false;
2188 continue;
2191 /* Dump details. */
2192 if (dump_file && (dump_flags & TDF_DETAILS))
2194 fprintf (dump_file, " Original statement:");
2195 print_generic_expr (dump_file, use_stmt, dump_flags);
2196 fprintf (dump_file, "\n");
2199 push_stmt_changes (&use_stmt);
2201 /* Propagate the RHS into this use of the LHS. */
2202 FOR_EACH_IMM_USE_ON_STMT (use_p, iter)
2203 propagate_value (use_p, rhs);
2205 /* Special cases to avoid useless calls into the folding
2206 routines, operand scanning, etc.
2208 First, propagation into a PHI may cause the PHI to become
2209 a degenerate, so mark the PHI as interesting. No other
2210 actions are necessary.
2212 Second, if we're propagating a virtual operand and the
2213 propagation does not change the underlying _DECL node for
2214 the virtual operand, then no further actions are necessary. */
2215 if (TREE_CODE (use_stmt) == PHI_NODE
2216 || (! is_gimple_reg (lhs)
2217 && TREE_CODE (rhs) == SSA_NAME
2218 && SSA_NAME_VAR (lhs) == SSA_NAME_VAR (rhs)))
2220 /* Dump details. */
2221 if (dump_file && (dump_flags & TDF_DETAILS))
2223 fprintf (dump_file, " Updated statement:");
2224 print_generic_expr (dump_file, use_stmt, dump_flags);
2225 fprintf (dump_file, "\n");
2228 /* Propagation into a PHI may expose new degenerate PHIs,
2229 so mark the result of the PHI as interesting. */
2230 if (TREE_CODE (use_stmt) == PHI_NODE)
2232 tree result = get_lhs_or_phi_result (use_stmt);
2233 bitmap_set_bit (interesting_names, SSA_NAME_VERSION (result));
2236 discard_stmt_changes (&use_stmt);
2237 continue;
2240 /* From this point onward we are propagating into a
2241 real statement. Folding may (or may not) be possible,
2242 we may expose new operands, expose dead EH edges,
2243 etc. */
2244 fold_stmt_inplace (use_stmt);
2246 /* Sometimes propagation can expose new operands to the
2247 renamer. Note this will call update_stmt at the
2248 appropriate time. */
2249 pop_stmt_changes (&use_stmt);
2251 /* Dump details. */
2252 if (dump_file && (dump_flags & TDF_DETAILS))
2254 fprintf (dump_file, " Updated statement:");
2255 print_generic_expr (dump_file, use_stmt, dump_flags);
2256 fprintf (dump_file, "\n");
2259 /* If we replaced a variable index with a constant, then
2260 we would need to update the invariant flag for ADDR_EXPRs. */
2261 if (TREE_CODE (use_stmt) == GIMPLE_MODIFY_STMT
2262 && TREE_CODE (GIMPLE_STMT_OPERAND (use_stmt, 1)) == ADDR_EXPR)
2263 recompute_tree_invariant_for_addr_expr
2264 (GIMPLE_STMT_OPERAND (use_stmt, 1));
2266 /* If we cleaned up EH information from the statement,
2267 mark its containing block as needing EH cleanups. */
2268 if (maybe_clean_or_replace_eh_stmt (use_stmt, use_stmt))
2270 bitmap_set_bit (need_eh_cleanup, bb_for_stmt (use_stmt)->index);
2271 if (dump_file && (dump_flags & TDF_DETAILS))
2272 fprintf (dump_file, " Flagged to clear EH edges.\n");
2275 /* Propagation may expose new trivial copy/constant propagation
2276 opportunities. */
2277 if (TREE_CODE (use_stmt) == GIMPLE_MODIFY_STMT
2278 && TREE_CODE (GIMPLE_STMT_OPERAND (use_stmt, 0)) == SSA_NAME
2279 && (TREE_CODE (GIMPLE_STMT_OPERAND (use_stmt, 1)) == SSA_NAME
2280 || is_gimple_min_invariant (GIMPLE_STMT_OPERAND (use_stmt,
2281 1))))
2283 tree result = get_lhs_or_phi_result (use_stmt);
2284 bitmap_set_bit (interesting_names, SSA_NAME_VERSION (result));
2287 /* Propagation into these nodes may make certain edges in
2288 the CFG unexecutable. We want to identify them as PHI nodes
2289 at the destination of those unexecutable edges may become
2290 degenerates. */
2291 else if (TREE_CODE (use_stmt) == COND_EXPR
2292 || TREE_CODE (use_stmt) == SWITCH_EXPR
2293 || TREE_CODE (use_stmt) == GOTO_EXPR)
2295 tree val;
2297 if (TREE_CODE (use_stmt) == COND_EXPR)
2298 val = COND_EXPR_COND (use_stmt);
2299 else if (TREE_CODE (use_stmt) == SWITCH_EXPR)
2300 val = SWITCH_COND (use_stmt);
2301 else
2302 val = GOTO_DESTINATION (use_stmt);
2304 if (is_gimple_min_invariant (val))
2306 basic_block bb = bb_for_stmt (use_stmt);
2307 edge te = find_taken_edge (bb, val);
2308 edge_iterator ei;
2309 edge e;
2310 block_stmt_iterator bsi;
2312 /* Remove all outgoing edges except TE. */
2313 for (ei = ei_start (bb->succs); (e = ei_safe_edge (ei));)
2315 if (e != te)
2317 tree phi;
2319 /* Mark all the PHI nodes at the destination of
2320 the unexecutable edge as interesting. */
2321 for (phi = phi_nodes (e->dest);
2322 phi;
2323 phi = PHI_CHAIN (phi))
2325 tree result = PHI_RESULT (phi);
2326 int version = SSA_NAME_VERSION (result);
2328 bitmap_set_bit (interesting_names, version);
2331 te->probability += e->probability;
2333 te->count += e->count;
2334 remove_edge (e);
2335 cfg_altered = true;
2337 else
2338 ei_next (&ei);
2341 bsi = bsi_last (bb_for_stmt (use_stmt));
2342 bsi_remove (&bsi, true);
2344 /* And fixup the flags on the single remaining edge. */
2345 te->flags &= ~(EDGE_TRUE_VALUE | EDGE_FALSE_VALUE);
2346 te->flags &= ~EDGE_ABNORMAL;
2347 te->flags |= EDGE_FALLTHRU;
2348 if (te->probability > REG_BR_PROB_BASE)
2349 te->probability = REG_BR_PROB_BASE;
2354 /* Ensure there is nothing else to do. */
2355 gcc_assert (!all || has_zero_uses (lhs));
2357 /* If we were able to propagate away all uses of LHS, then
2358 we can remove STMT. */
2359 if (all)
2360 remove_stmt_or_phi (stmt);
2364 /* T is either a PHI node (potentially a degenerate PHI node) or
2365 a statement that is a trivial copy or constant initialization.
2367 Attempt to eliminate T by propagating its RHS into all uses of
2368 its LHS. This may in turn set new bits in INTERESTING_NAMES
2369 for nodes we want to revisit later.
2371 All exit paths should clear INTERESTING_NAMES for the result
2372 of T. */
2374 static void
2375 eliminate_const_or_copy (tree t, bitmap interesting_names)
2377 tree lhs = get_lhs_or_phi_result (t);
2378 tree rhs;
2379 int version = SSA_NAME_VERSION (lhs);
2381 /* If the LHS of this statement or PHI has no uses, then we can
2382 just eliminate it. This can occur if, for example, the PHI
2383 was created by block duplication due to threading and its only
2384 use was in the conditional at the end of the block which was
2385 deleted. */
2386 if (has_zero_uses (lhs))
2388 bitmap_clear_bit (interesting_names, version);
2389 remove_stmt_or_phi (t);
2390 return;
2393 /* Get the RHS of the assignment or PHI node if the PHI is a
2394 degenerate. */
2395 rhs = get_rhs_or_phi_arg (t);
2396 if (!rhs)
2398 bitmap_clear_bit (interesting_names, version);
2399 return;
2402 propagate_rhs_into_lhs (t, lhs, rhs, interesting_names);
2404 /* Note that T may well have been deleted by now, so do
2405 not access it, instead use the saved version # to clear
2406 T's entry in the worklist. */
2407 bitmap_clear_bit (interesting_names, version);
2410 /* The first phase in degenerate PHI elimination.
2412 Eliminate the degenerate PHIs in BB, then recurse on the
2413 dominator children of BB. */
2415 static void
2416 eliminate_degenerate_phis_1 (basic_block bb, bitmap interesting_names)
2418 tree phi, next;
2419 basic_block son;
2421 for (phi = phi_nodes (bb); phi; phi = next)
2423 next = PHI_CHAIN (phi);
2424 eliminate_const_or_copy (phi, interesting_names);
2427 /* Recurse into the dominator children of BB. */
2428 for (son = first_dom_son (CDI_DOMINATORS, bb);
2429 son;
2430 son = next_dom_son (CDI_DOMINATORS, son))
2431 eliminate_degenerate_phis_1 (son, interesting_names);
2435 /* A very simple pass to eliminate degenerate PHI nodes from the
2436 IL. This is meant to be fast enough to be able to be run several
2437 times in the optimization pipeline.
2439 Certain optimizations, particularly those which duplicate blocks
2440 or remove edges from the CFG can create or expose PHIs which are
2441 trivial copies or constant initializations.
2443 While we could pick up these optimizations in DOM or with the
2444 combination of copy-prop and CCP, those solutions are far too
2445 heavy-weight for our needs.
2447 This implementation has two phases so that we can efficiently
2448 eliminate the first order degenerate PHIs and second order
2449 degenerate PHIs.
2451 The first phase performs a dominator walk to identify and eliminate
2452 the vast majority of the degenerate PHIs. When a degenerate PHI
2453 is identified and eliminated any affected statements or PHIs
2454 are put on a worklist.
2456 The second phase eliminates degenerate PHIs and trivial copies
2457 or constant initializations using the worklist. This is how we
2458 pick up the secondary optimization opportunities with minimal
2459 cost. */
2461 static unsigned int
2462 eliminate_degenerate_phis (void)
2464 bitmap interesting_names;
2465 bitmap interesting_names1;
2467 /* Bitmap of blocks which need EH information updated. We can not
2468 update it on-the-fly as doing so invalidates the dominator tree. */
2469 need_eh_cleanup = BITMAP_ALLOC (NULL);
2471 /* INTERESTING_NAMES is effectively our worklist, indexed by
2472 SSA_NAME_VERSION.
2474 A set bit indicates that the statement or PHI node which
2475 defines the SSA_NAME should be (re)examined to determine if
2476 it has become a degenerate PHI or trivial const/copy propagation
2477 opportunity.
2479 Experiments have show we generally get better compilation
2480 time behavior with bitmaps rather than sbitmaps. */
2481 interesting_names = BITMAP_ALLOC (NULL);
2482 interesting_names1 = BITMAP_ALLOC (NULL);
2484 calculate_dominance_info (CDI_DOMINATORS);
2485 cfg_altered = false;
2487 /* First phase. Eliminate degenerate PHIs via a dominator
2488 walk of the CFG.
2490 Experiments have indicated that we generally get better
2491 compile-time behavior by visiting blocks in the first
2492 phase in dominator order. Presumably this is because walking
2493 in dominator order leaves fewer PHIs for later examination
2494 by the worklist phase. */
2495 eliminate_degenerate_phis_1 (ENTRY_BLOCK_PTR, interesting_names);
2497 /* Second phase. Eliminate second order degenerate PHIs as well
2498 as trivial copies or constant initializations identified by
2499 the first phase or this phase. Basically we keep iterating
2500 until our set of INTERESTING_NAMEs is empty. */
2501 while (!bitmap_empty_p (interesting_names))
2503 unsigned int i;
2504 bitmap_iterator bi;
2506 /* EXECUTE_IF_SET_IN_BITMAP does not like its bitmap
2507 changed during the loop. Copy it to another bitmap and
2508 use that. */
2509 bitmap_copy (interesting_names1, interesting_names);
2511 EXECUTE_IF_SET_IN_BITMAP (interesting_names1, 0, i, bi)
2513 tree name = ssa_name (i);
2515 /* Ignore SSA_NAMEs that have been released because
2516 their defining statement was deleted (unreachable). */
2517 if (name)
2518 eliminate_const_or_copy (SSA_NAME_DEF_STMT (ssa_name (i)),
2519 interesting_names);
2523 if (cfg_altered)
2524 free_dominance_info (CDI_DOMINATORS);
2526 /* Propagation of const and copies may make some EH edges dead. Purge
2527 such edges from the CFG as needed. */
2528 if (!bitmap_empty_p (need_eh_cleanup))
2530 tree_purge_all_dead_eh_edges (need_eh_cleanup);
2531 BITMAP_FREE (need_eh_cleanup);
2534 BITMAP_FREE (interesting_names);
2535 BITMAP_FREE (interesting_names1);
2536 return 0;
2539 struct gimple_opt_pass pass_phi_only_cprop =
2542 GIMPLE_PASS,
2543 "phicprop", /* name */
2544 gate_dominator, /* gate */
2545 eliminate_degenerate_phis, /* execute */
2546 NULL, /* sub */
2547 NULL, /* next */
2548 0, /* static_pass_number */
2549 TV_TREE_PHI_CPROP, /* tv_id */
2550 PROP_cfg | PROP_ssa | PROP_alias, /* properties_required */
2551 0, /* properties_provided */
2552 0, /* properties_destroyed */
2553 0, /* todo_flags_start */
2554 TODO_cleanup_cfg
2555 | TODO_dump_func
2556 | TODO_ggc_collect
2557 | TODO_verify_ssa
2558 | TODO_verify_stmts
2559 | TODO_update_ssa /* todo_flags_finish */