PR c++/16115
[official-gcc.git] / gcc / tree-ssa-dom.c
blob740fe9bc9b2d42c36e2e4f2411c515b6997f1911
1 /* SSA Dominator optimizations for trees
2 Copyright (C) 2001, 2002, 2003, 2004 Free Software Foundation, Inc.
3 Contributed by Diego Novillo <dnovillo@redhat.com>
5 This file is part of GCC.
7 GCC is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
10 any later version.
12 GCC is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with GCC; see the file COPYING. If not, write to
19 the Free Software Foundation, 59 Temple Place - Suite 330,
20 Boston, MA 02111-1307, USA. */
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 "output.h"
33 #include "errors.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 "langhooks.h"
45 /* This file implements optimizations on the dominator tree. */
47 /* Hash table with expressions made available during the renaming process.
48 When an assignment of the form X_i = EXPR is found, the statement is
49 stored in this table. If the same expression EXPR is later found on the
50 RHS of another statement, it is replaced with X_i (thus performing
51 global redundancy elimination). Similarly as we pass through conditionals
52 we record the conditional itself as having either a true or false value
53 in this table. */
54 static htab_t avail_exprs;
56 /* Structure for entries in the expression hash table.
58 This requires more memory for the hash table entries, but allows us
59 to avoid creating silly tree nodes and annotations for conditionals,
60 eliminates 2 global hash tables and two block local varrays.
62 It also allows us to reduce the number of hash table lookups we
63 have to perform in lookup_avail_expr and finally it allows us to
64 significantly reduce the number of calls into the hashing routine
65 itself. */
66 struct expr_hash_elt
68 /* The value (lhs) of this expression. */
69 tree lhs;
71 /* The expression (rhs) we want to record. */
72 tree rhs;
74 /* The annotation if this element corresponds to a statement. */
75 stmt_ann_t ann;
77 /* The hash value for RHS/ann. */
78 hashval_t hash;
81 /* Table of constant values and copies indexed by SSA name. When the
82 renaming pass finds an assignment of a constant (X_i = C) or a copy
83 assignment from another SSA variable (X_i = Y_j), it creates a mapping
84 between X_i and the RHS in this table. This mapping is used later on,
85 when renaming uses of X_i. If an assignment to X_i is found in this
86 table, instead of using X_i, we use the RHS of the statement stored in
87 this table (thus performing very simplistic copy and constant
88 propagation). */
89 static varray_type const_and_copies;
91 /* Bitmap of SSA_NAMEs known to have a nonzero value, even if we do not
92 know their exact value. */
93 static bitmap nonzero_vars;
95 /* Track whether or not we have changed the control flow graph. */
96 static bool cfg_altered;
98 /* Bitmap of blocks that have had EH statements cleaned. We should
99 remove their dead edges eventually. */
100 static bitmap need_eh_cleanup;
102 /* Statistics for dominator optimizations. */
103 struct opt_stats_d
105 long num_stmts;
106 long num_exprs_considered;
107 long num_re;
110 /* Value range propagation record. Each time we encounter a conditional
111 of the form SSA_NAME COND CONST we create a new vrp_element to record
112 how the condition affects the possible values SSA_NAME may have.
114 Each record contains the condition tested (COND), and the the range of
115 values the variable may legitimately have if COND is true. Note the
116 range of values may be a smaller range than COND specifies if we have
117 recorded other ranges for this variable. Each record also contains the
118 block in which the range was recorded for invalidation purposes.
120 Note that the current known range is computed lazily. This allows us
121 to avoid the overhead of computing ranges which are never queried.
123 When we encounter a conditional, we look for records which constrain
124 the SSA_NAME used in the condition. In some cases those records allow
125 us to determine the condition's result at compile time. In other cases
126 they may allow us to simplify the condition.
128 We also use value ranges to do things like transform signed div/mod
129 operations into unsigned div/mod or to simplify ABS_EXPRs.
131 Simple experiments have shown these optimizations to not be all that
132 useful on switch statements (much to my surprise). So switch statement
133 optimizations are not performed.
135 Note carefully we do not propagate information through each statement
136 in the block. ie, if we know variable X has a value defined of
137 [0, 25] and we encounter Y = X + 1, we do not track a value range
138 for Y (which would be [1, 26] if we cared). Similarly we do not
139 constrain values as we encounter narrowing typecasts, etc. */
141 struct vrp_element
143 /* The highest and lowest values the variable in COND may contain when
144 COND is true. Note this may not necessarily be the same values
145 tested by COND if the same variable was used in earlier conditionals.
147 Note this is computed lazily and thus can be NULL indicating that
148 the values have not been computed yet. */
149 tree low;
150 tree high;
152 /* The actual conditional we recorded. This is needed since we compute
153 ranges lazily. */
154 tree cond;
156 /* The basic block where this record was created. We use this to determine
157 when to remove records. */
158 basic_block bb;
161 static struct opt_stats_d opt_stats;
163 /* This virtual array holds pairs of edges which describe a scheduled
164 edge redirection from jump threading.
166 The first entry in each pair is the edge we are going to redirect.
168 The second entry in each pair is the edge leading to our final
169 destination block. By providing this as an edge rather than the
170 final target block itself we can correctly handle redirections
171 when the target block had PHIs which required edge insertions/splitting
172 to remove the PHIs. */
173 static GTY(()) varray_type redirection_edges;
175 /* A virtual array holding value range records for the variable identified
176 by the index, SSA_VERSION. */
177 static varray_type vrp_data;
179 /* Datastructure for block local data used during the dominator walk.
180 We maintain a stack of these as we recursively walk down the
181 dominator tree. */
183 struct dom_walk_block_data
185 /* Array of all the expressions entered into the global expression
186 hash table by this block. During finalization we use this array to
187 know what expressions to remove from the global expression hash
188 table. */
189 varray_type avail_exprs;
191 /* Array of dest, src pairs that need to be restored during finalization
192 into the global const/copies table during finalization. */
193 varray_type const_and_copies;
195 /* Similarly for the nonzero state of variables that needs to be
196 restored during finalization. */
197 varray_type nonzero_vars;
199 /* Array of statements we need to rescan during finalization for newly
200 exposed variables. */
201 varray_type stmts_to_rescan;
203 /* Array of variables which have their values constrained by operations
204 in this basic block. We use this during finalization to know
205 which variables need their VRP data updated. */
206 varray_type vrp_variables;
208 /* Array of tree pairs used to restore the global currdefs to its
209 original state after completing optimization of a block and its
210 dominator children. */
211 varray_type block_defs;
214 struct eq_expr_value
216 tree src;
217 tree dst;
220 /* Local functions. */
221 static void optimize_stmt (struct dom_walk_data *,
222 basic_block bb,
223 block_stmt_iterator);
224 static inline tree get_value_for (tree, varray_type table);
225 static inline void set_value_for (tree, tree, varray_type table);
226 static tree lookup_avail_expr (tree, varray_type *, bool);
227 static struct eq_expr_value get_eq_expr_value (tree, int, varray_type *,
228 basic_block, varray_type *);
229 static hashval_t avail_expr_hash (const void *);
230 static hashval_t real_avail_expr_hash (const void *);
231 static int avail_expr_eq (const void *, const void *);
232 static void htab_statistics (FILE *, htab_t);
233 static void record_cond (tree, tree, varray_type *);
234 static void record_dominating_conditions (tree, varray_type *);
235 static void record_const_or_copy (tree, tree, varray_type *);
236 static void record_equality (tree, tree, varray_type *);
237 static tree update_rhs_and_lookup_avail_expr (tree, tree, varray_type *,
238 stmt_ann_t, bool);
239 static tree simplify_rhs_and_lookup_avail_expr (struct dom_walk_data *,
240 tree, stmt_ann_t, int);
241 static tree simplify_cond_and_lookup_avail_expr (tree, varray_type *,
242 stmt_ann_t, int);
243 static tree simplify_switch_and_lookup_avail_expr (tree, varray_type *,
244 stmt_ann_t, int);
245 static tree find_equivalent_equality_comparison (tree);
246 static void record_range (tree, basic_block, varray_type *);
247 static bool extract_range_from_cond (tree, tree *, tree *, int *);
248 static void record_equivalences_from_phis (struct dom_walk_data *, basic_block);
249 static void record_equivalences_from_incoming_edge (struct dom_walk_data *,
250 basic_block);
251 static bool eliminate_redundant_computations (struct dom_walk_data *,
252 tree, stmt_ann_t);
253 static void record_equivalences_from_stmt (tree, varray_type *, varray_type *,
254 int, stmt_ann_t);
255 static void thread_across_edge (struct dom_walk_data *, edge);
256 static void dom_opt_finalize_block (struct dom_walk_data *, basic_block);
257 static void dom_opt_initialize_block_local_data (struct dom_walk_data *,
258 basic_block, bool);
259 static void dom_opt_initialize_block (struct dom_walk_data *, basic_block);
260 static void cprop_into_phis (struct dom_walk_data *, basic_block);
261 static void remove_local_expressions_from_table (varray_type locals,
262 unsigned limit,
263 htab_t table);
264 static void restore_vars_to_original_value (varray_type locals,
265 unsigned limit,
266 varray_type table);
267 static void restore_currdefs_to_original_value (varray_type locals,
268 unsigned limit);
269 static void register_definitions_for_stmt (stmt_ann_t, varray_type *);
270 static void redirect_edges_and_update_ssa_graph (varray_type);
272 /* Local version of fold that doesn't introduce cruft. */
274 static tree
275 local_fold (tree t)
277 t = fold (t);
279 /* Strip away useless type conversions. Both the NON_LVALUE_EXPR that
280 may have been added by fold, and "useless" type conversions that might
281 now be apparent due to propagation. */
282 STRIP_USELESS_TYPE_CONVERSION (t);
284 return t;
287 /* Return the value associated with variable VAR in TABLE. */
289 static inline tree
290 get_value_for (tree var, varray_type table)
292 return VARRAY_TREE (table, SSA_NAME_VERSION (var));
295 /* Associate VALUE to variable VAR in TABLE. */
297 static inline void
298 set_value_for (tree var, tree value, varray_type table)
300 VARRAY_TREE (table, SSA_NAME_VERSION (var)) = value;
303 /* REDIRECTION_EDGES contains edge pairs where we want to revector the
304 destination of the first edge to the destination of the second edge.
306 These redirections may significantly change the SSA graph since we
307 allow redirection through blocks with PHI nodes and blocks with
308 real instructions in some cases.
310 This routine will perform the requested redirections and incrementally
311 update the SSA graph.
313 Note in some cases requested redirections may be ignored as they can
314 not be safely implemented. */
316 static void
317 redirect_edges_and_update_ssa_graph (varray_type redirection_edges)
319 basic_block tgt, bb;
320 tree phi;
321 unsigned int i;
322 size_t old_num_referenced_vars = num_referenced_vars;
323 bitmap virtuals_to_rename = BITMAP_XMALLOC ();
325 /* First note any variables which we are going to have to take
326 out of SSA form as well as any virtuals which need updating. */
327 for (i = 0; i < VARRAY_ACTIVE_SIZE (redirection_edges); i += 2)
329 block_stmt_iterator bsi;
330 edge e;
331 basic_block tgt;
332 tree phi;
334 e = VARRAY_EDGE (redirection_edges, i);
335 tgt = VARRAY_EDGE (redirection_edges, i + 1)->dest;
337 /* All variables referenced in PHI nodes we bypass must be
338 renamed. */
339 for (phi = phi_nodes (e->dest); phi; phi = PHI_CHAIN (phi))
341 tree result = SSA_NAME_VAR (PHI_RESULT (phi));
343 if (is_gimple_reg (PHI_RESULT (phi)))
344 bitmap_set_bit (vars_to_rename, var_ann (result)->uid);
345 else
346 bitmap_set_bit (virtuals_to_rename, var_ann (result)->uid);
349 /* Any variables set by statements at the start of the block we
350 are bypassing must also be taken our of SSA form. */
351 for (bsi = bsi_start (e->dest); ! bsi_end_p (bsi); bsi_next (&bsi))
353 unsigned int j;
354 def_optype defs;
355 v_may_def_optype v_may_defs;
356 v_must_def_optype v_must_defs;
357 tree stmt = bsi_stmt (bsi);
358 stmt_ann_t ann = stmt_ann (stmt);
360 if (TREE_CODE (stmt) == COND_EXPR)
361 break;
363 get_stmt_operands (stmt);
365 defs = DEF_OPS (ann);
366 for (j = 0; j < NUM_DEFS (defs); j++)
368 tree op = SSA_NAME_VAR (DEF_OP (defs, j));
369 bitmap_set_bit (vars_to_rename, var_ann (op)->uid);
372 v_may_defs = STMT_V_MAY_DEF_OPS (stmt);
373 for (j = 0; j < NUM_V_MAY_DEFS (v_may_defs); j++)
375 tree op = V_MAY_DEF_RESULT (v_may_defs, j);
376 bitmap_set_bit (vars_to_rename, var_ann (op)->uid);
379 v_must_defs = STMT_V_MUST_DEF_OPS (stmt);
380 for (j = 0; j < NUM_V_MUST_DEFS (v_must_defs); j++)
382 tree op = V_MUST_DEF_OP (v_must_defs, j);
383 bitmap_set_bit (vars_to_rename, var_ann (op)->uid);
387 /* Finally, any variables in PHI nodes at our final destination
388 must also be taken our of SSA form. */
389 for (phi = phi_nodes (tgt); phi; phi = PHI_CHAIN (phi))
391 tree result = SSA_NAME_VAR (PHI_RESULT (phi));
393 if (is_gimple_reg (PHI_RESULT (phi)))
394 bitmap_set_bit (vars_to_rename, var_ann (result)->uid);
395 else
396 bitmap_set_bit (virtuals_to_rename, var_ann (result)->uid);
400 /* Take those selected variables out of SSA form. This must be
401 done before we start redirecting edges. */
402 if (bitmap_first_set_bit (vars_to_rename) >= 0)
403 rewrite_vars_out_of_ssa (vars_to_rename);
405 /* The out of SSA translation above may split the edge from
406 E->src to E->dest. This could potentially cause us to lose
407 an assignment leading to invalid warnings about uninitialized
408 variables or incorrect code.
410 Luckily, we can detect this by looking at the last statement
411 in E->dest. If it is not a COND_EXPR or SWITCH_EXPR, then
412 the edge was split and instead of E, we want E->dest->succ. */
413 for (i = 0; i < VARRAY_ACTIVE_SIZE (redirection_edges); i += 2)
415 edge e = VARRAY_EDGE (redirection_edges, i);
416 tree last = last_stmt (e->dest);
418 if (last
419 && TREE_CODE (last) != COND_EXPR
420 && TREE_CODE (last) != SWITCH_EXPR)
422 e = e->dest->succ;
424 #ifdef ENABLE_CHECKING
425 /* There should only be a single successor if the
426 original edge was split. */
427 if (e->succ_next)
428 abort ();
429 #endif
430 /* Replace the edge in REDIRECTION_EDGES for the
431 loop below. */
432 VARRAY_EDGE (redirection_edges, i) = e;
436 /* If we created any new variables as part of the out-of-ssa
437 translation, then any jump threads must be invalidated if they
438 bypass a block in which we skipped instructions.
440 This is necessary as instructions which appeared to be NOPS
441 may be necessary after the out-of-ssa translation. */
442 if (num_referenced_vars != old_num_referenced_vars)
444 for (i = 0; i < VARRAY_ACTIVE_SIZE (redirection_edges); i += 2)
446 block_stmt_iterator bsi;
447 edge e;
449 e = VARRAY_EDGE (redirection_edges, i);
450 for (bsi = bsi_start (e->dest); ! bsi_end_p (bsi); bsi_next (&bsi))
452 tree stmt = bsi_stmt (bsi);
454 if (IS_EMPTY_STMT (stmt)
455 || TREE_CODE (stmt) == LABEL_EXPR)
456 continue;
458 if (TREE_CODE (stmt) == COND_EXPR)
459 break;
461 /* Invalidate the jump thread. */
462 VARRAY_EDGE (redirection_edges, i) = NULL;
463 VARRAY_EDGE (redirection_edges, i + 1) = NULL;
464 break;
469 /* Now redirect the edges. */
470 for (i = 0; i < VARRAY_ACTIVE_SIZE (redirection_edges); i += 2)
472 basic_block src;
473 edge e;
475 e = VARRAY_EDGE (redirection_edges, i);
476 if (!e)
477 continue;
479 tgt = VARRAY_EDGE (redirection_edges, i + 1)->dest;
482 if (dump_file && (dump_flags & TDF_DETAILS))
483 fprintf (dump_file, " Threaded jump %d --> %d to %d\n",
484 e->src->index, e->dest->index, tgt->index);
486 src = e->src;
488 e = redirect_edge_and_branch (e, tgt);
489 PENDING_STMT (e) = NULL_TREE;
491 /* Updating the dominance information would be nontrivial. */
492 free_dominance_info (CDI_DOMINATORS);
494 if ((dump_file && (dump_flags & TDF_DETAILS))
495 && e->src != src)
496 fprintf (dump_file, " basic block %d created\n",
497 e->src->index);
499 cfg_altered = true;
502 VARRAY_CLEAR (redirection_edges);
504 for (i = old_num_referenced_vars; i < num_referenced_vars; i++)
506 bitmap_set_bit (vars_to_rename, i);
507 var_ann (referenced_var (i))->out_of_ssa_tag = 0;
510 bitmap_a_or_b (vars_to_rename, vars_to_rename, virtuals_to_rename);
512 /* We must remove any PHIs for virtual variables that we are going to
513 re-rename. Hopefully we'll be able to simply update these incrementally
514 soon. */
515 FOR_EACH_BB (bb)
517 tree next;
519 for (phi = phi_nodes (bb); phi; phi = next)
521 tree result = PHI_RESULT (phi);
523 next = PHI_CHAIN (phi);
525 if (bitmap_bit_p (virtuals_to_rename,
526 var_ann (SSA_NAME_VAR (result))->uid))
527 remove_phi_node (phi, NULL, bb);
530 BITMAP_XFREE (virtuals_to_rename);
533 /* Jump threading, redundancy elimination and const/copy propagation.
535 This pass may expose new symbols that need to be renamed into SSA. For
536 every new symbol exposed, its corresponding bit will be set in
537 VARS_TO_RENAME. */
539 static void
540 tree_ssa_dominator_optimize (void)
542 basic_block bb;
543 struct dom_walk_data walk_data;
544 unsigned int i;
546 for (i = 0; i < num_referenced_vars; i++)
547 var_ann (referenced_var (i))->current_def = NULL;
549 /* Mark loop edges so we avoid threading across loop boundaries.
550 This may result in transforming natural loop into irreducible
551 region. */
552 mark_dfs_back_edges ();
554 /* Create our hash tables. */
555 avail_exprs = htab_create (1024, real_avail_expr_hash, avail_expr_eq, free);
556 VARRAY_TREE_INIT (const_and_copies, num_ssa_names, "const_and_copies");
557 nonzero_vars = BITMAP_XMALLOC ();
558 VARRAY_EDGE_INIT (redirection_edges, 20, "redirection_edges");
559 VARRAY_GENERIC_PTR_INIT (vrp_data, num_ssa_names, "vrp_data");
560 need_eh_cleanup = BITMAP_XMALLOC ();
562 /* Setup callbacks for the generic dominator tree walker. */
563 walk_data.walk_stmts_backward = false;
564 walk_data.dom_direction = CDI_DOMINATORS;
565 walk_data.initialize_block_local_data = dom_opt_initialize_block_local_data;
566 walk_data.before_dom_children_before_stmts = dom_opt_initialize_block;
567 walk_data.before_dom_children_walk_stmts = optimize_stmt;
568 walk_data.before_dom_children_after_stmts = cprop_into_phis;
569 walk_data.after_dom_children_before_stmts = NULL;
570 walk_data.after_dom_children_walk_stmts = NULL;
571 walk_data.after_dom_children_after_stmts = dom_opt_finalize_block;
572 /* Right now we only attach a dummy COND_EXPR to the global data pointer.
573 When we attach more stuff we'll need to fill this out with a real
574 structure. */
575 walk_data.global_data = NULL;
576 walk_data.block_local_data_size = sizeof (struct dom_walk_block_data);
578 /* Now initialize the dominator walker. */
579 init_walk_dominator_tree (&walk_data);
581 /* Reset block_forwardable in each block's annotation. We use that
582 attribute when threading through COND_EXPRs. */
583 FOR_EACH_BB (bb)
584 bb_ann (bb)->forwardable = 1;
586 calculate_dominance_info (CDI_DOMINATORS);
588 /* If we prove certain blocks are unreachable, then we want to
589 repeat the dominator optimization process as PHI nodes may
590 have turned into copies which allows better propagation of
591 values. So we repeat until we do not identify any new unreachable
592 blocks. */
595 /* Optimize the dominator tree. */
596 cfg_altered = false;
598 /* Recursively walk the dominator tree optimizing statements. */
599 walk_dominator_tree (&walk_data, ENTRY_BLOCK_PTR);
601 /* Wipe the hash tables. */
603 if (VARRAY_ACTIVE_SIZE (redirection_edges) > 0)
604 redirect_edges_and_update_ssa_graph (redirection_edges);
606 if (bitmap_first_set_bit (need_eh_cleanup) >= 0)
608 cfg_altered = tree_purge_all_dead_eh_edges (need_eh_cleanup);
609 bitmap_zero (need_eh_cleanup);
612 /* We may have made some basic blocks unreachable, remove them. */
613 cfg_altered |= delete_unreachable_blocks ();
615 /* If the CFG was altered, then recompute the dominator tree. This
616 is not strictly needed if we only removed unreachable blocks, but
617 may produce better results. If we threaded jumps, then rebuilding
618 the dominator tree is strictly necessary. Likewise with EH cleanup.
619 Free the dominance info first so that cleanup_tree_cfg doesn't try
620 to verify it. */
621 if (cfg_altered)
623 free_dominance_info (CDI_DOMINATORS);
624 cleanup_tree_cfg ();
625 calculate_dominance_info (CDI_DOMINATORS);
628 /* If we are going to iterate (CFG_ALTERED is true), then we must
629 perform any queued renaming before the next iteration. */
630 if (cfg_altered
631 && bitmap_first_set_bit (vars_to_rename) >= 0)
633 rewrite_into_ssa (false);
634 bitmap_clear (vars_to_rename);
636 /* The into SSA translation may have created new SSA_NAMES whic
637 affect the size of CONST_AND_COPIES and VRP_DATA. */
638 VARRAY_GROW (const_and_copies, num_ssa_names);
639 VARRAY_GROW (vrp_data, num_ssa_names);
642 /* Reinitialize the various tables. */
643 bitmap_clear (nonzero_vars);
644 htab_empty (avail_exprs);
645 VARRAY_CLEAR (const_and_copies);
646 VARRAY_CLEAR (vrp_data);
648 for (i = 0; i < num_referenced_vars; i++)
649 var_ann (referenced_var (i))->current_def = NULL;
651 while (cfg_altered);
653 /* Remove any unreachable blocks left behind and linearize the CFG. */
654 cleanup_tree_cfg ();
656 /* Debugging dumps. */
657 if (dump_file && (dump_flags & TDF_STATS))
658 dump_dominator_optimization_stats (dump_file);
660 /* We emptied the hash table earlier, now delete it completely. */
661 htab_delete (avail_exprs);
663 /* It is not necessary to clear CURRDEFS, REDIRECTION_EDGES, VRP_DATA,
664 CONST_AND_COPIES, and NONZERO_VARS as they all get cleared at the bottom
665 of the do-while loop above. */
667 /* And finalize the dominator walker. */
668 fini_walk_dominator_tree (&walk_data);
670 /* Free nonzero_vars. */
671 BITMAP_XFREE (nonzero_vars);
672 BITMAP_XFREE (need_eh_cleanup);
675 static bool
676 gate_dominator (void)
678 return flag_tree_dom != 0;
681 struct tree_opt_pass pass_dominator =
683 "dom", /* name */
684 gate_dominator, /* gate */
685 tree_ssa_dominator_optimize, /* execute */
686 NULL, /* sub */
687 NULL, /* next */
688 0, /* static_pass_number */
689 TV_TREE_SSA_DOMINATOR_OPTS, /* tv_id */
690 PROP_cfg | PROP_ssa, /* properties_required */
691 0, /* properties_provided */
692 0, /* properties_destroyed */
693 0, /* todo_flags_start */
694 TODO_dump_func | TODO_rename_vars
695 | TODO_verify_ssa /* todo_flags_finish */
699 /* We are exiting BB, see if the target block begins with a conditional
700 jump which has a known value when reached via BB. */
702 static void
703 thread_across_edge (struct dom_walk_data *walk_data, edge e)
705 struct dom_walk_block_data *bd
706 = VARRAY_TOP_GENERIC_PTR (walk_data->block_data_stack);
707 block_stmt_iterator bsi;
708 tree stmt = NULL;
709 tree phi;
711 /* Each PHI creates a temporary equivalence, record them. */
712 for (phi = phi_nodes (e->dest); phi; phi = PHI_CHAIN (phi))
714 tree src = PHI_ARG_DEF_FROM_EDGE (phi, e);
715 tree dst = PHI_RESULT (phi);
716 record_const_or_copy (dst, src, &bd->const_and_copies);
717 register_new_def (dst, &bd->block_defs);
720 for (bsi = bsi_start (e->dest); ! bsi_end_p (bsi); bsi_next (&bsi))
722 tree lhs, cached_lhs;
724 stmt = bsi_stmt (bsi);
726 /* Ignore empty statements and labels. */
727 if (IS_EMPTY_STMT (stmt) || TREE_CODE (stmt) == LABEL_EXPR)
728 continue;
730 /* If this is not a MODIFY_EXPR which sets an SSA_NAME to a new
731 value, then stop our search here. Ideally when we stop a
732 search we stop on a COND_EXPR or SWITCH_EXPR. */
733 if (TREE_CODE (stmt) != MODIFY_EXPR
734 || TREE_CODE (TREE_OPERAND (stmt, 0)) != SSA_NAME)
735 break;
737 /* At this point we have a statement which assigns an RHS to an
738 SSA_VAR on the LHS. We want to prove that the RHS is already
739 available and that its value is held in the current definition
740 of the LHS -- meaning that this assignment is a NOP when
741 reached via edge E. */
742 if (TREE_CODE (TREE_OPERAND (stmt, 1)) == SSA_NAME)
743 cached_lhs = TREE_OPERAND (stmt, 1);
744 else
745 cached_lhs = lookup_avail_expr (stmt, NULL, false);
747 lhs = TREE_OPERAND (stmt, 0);
749 /* This can happen if we thread around to the start of a loop. */
750 if (lhs == cached_lhs)
751 break;
753 /* If we did not find RHS in the hash table, then try again after
754 temporarily const/copy propagating the operands. */
755 if (!cached_lhs)
757 /* Copy the operands. */
758 stmt_ann_t ann = stmt_ann (stmt);
759 use_optype uses = USE_OPS (ann);
760 vuse_optype vuses = VUSE_OPS (ann);
761 tree *uses_copy = xcalloc (NUM_USES (uses), sizeof (tree));
762 tree *vuses_copy = xcalloc (NUM_VUSES (vuses), sizeof (tree));
763 unsigned int i;
765 /* Make a copy of the uses into USES_COPY, then cprop into
766 the use operands. */
767 for (i = 0; i < NUM_USES (uses); i++)
769 tree tmp = NULL;
771 uses_copy[i] = USE_OP (uses, i);
772 if (TREE_CODE (USE_OP (uses, i)) == SSA_NAME)
773 tmp = get_value_for (USE_OP (uses, i), const_and_copies);
774 if (tmp)
775 SET_USE_OP (uses, i, tmp);
778 /* Similarly for virtual uses. */
779 for (i = 0; i < NUM_VUSES (vuses); i++)
781 tree tmp = NULL;
783 vuses_copy[i] = VUSE_OP (vuses, i);
784 if (TREE_CODE (VUSE_OP (vuses, i)) == SSA_NAME)
785 tmp = get_value_for (VUSE_OP (vuses, i), const_and_copies);
786 if (tmp)
787 SET_VUSE_OP (vuses, i, tmp);
790 /* Try to lookup the new expression. */
791 cached_lhs = lookup_avail_expr (stmt, NULL, false);
793 /* Restore the statement's original uses/defs. */
794 for (i = 0; i < NUM_USES (uses); i++)
795 SET_USE_OP (uses, i, uses_copy[i]);
797 for (i = 0; i < NUM_VUSES (vuses); i++)
798 SET_VUSE_OP (vuses, i, vuses_copy[i]);
800 free (uses_copy);
801 free (vuses_copy);
803 /* If we still did not find the expression in the hash table,
804 then we can not ignore this statement. */
805 if (! cached_lhs)
806 break;
809 /* If the expression in the hash table was not assigned to an
810 SSA_NAME, then we can not ignore this statement. */
811 if (TREE_CODE (cached_lhs) != SSA_NAME)
812 break;
814 /* If we have different underlying variables, then we can not
815 ignore this statement. */
816 if (SSA_NAME_VAR (cached_lhs) != SSA_NAME_VAR (lhs))
817 break;
819 /* If CACHED_LHS does not represent the current value of the undering
820 variable in CACHED_LHS/LHS, then we can not ignore this statement. */
821 if (var_ann (SSA_NAME_VAR (lhs))->current_def != cached_lhs)
822 break;
824 /* If we got here, then we can ignore this statement and continue
825 walking through the statements in the block looking for a threadable
826 COND_EXPR.
828 We want to record an equivalence lhs = cache_lhs so that if
829 the result of this statement is used later we can copy propagate
830 suitably. */
831 record_const_or_copy (lhs, cached_lhs, &bd->const_and_copies);
832 register_new_def (lhs, &bd->block_defs);
835 /* If we stopped at a COND_EXPR or SWITCH_EXPR, then see if we know which
836 arm will be taken. */
837 if (stmt
838 && (TREE_CODE (stmt) == COND_EXPR
839 || TREE_CODE (stmt) == SWITCH_EXPR))
841 tree cond, cached_lhs;
842 edge e1;
844 /* Do not forward entry edges into the loop. In the case loop
845 has multiple entry edges we may end up in constructing irreducible
846 region.
847 ??? We may consider forwarding the edges in the case all incoming
848 edges forward to the same destination block. */
849 if (!e->flags & EDGE_DFS_BACK)
851 for (e1 = e->dest->pred; e; e = e->pred_next)
852 if (e1->flags & EDGE_DFS_BACK)
853 break;
854 if (e1)
855 return;
858 /* Now temporarily cprop the operands and try to find the resulting
859 expression in the hash tables. */
860 if (TREE_CODE (stmt) == COND_EXPR)
861 cond = COND_EXPR_COND (stmt);
862 else
863 cond = SWITCH_COND (stmt);
865 if (TREE_CODE_CLASS (TREE_CODE (cond)) == '<')
867 tree dummy_cond, op0, op1;
868 enum tree_code cond_code;
870 op0 = TREE_OPERAND (cond, 0);
871 op1 = TREE_OPERAND (cond, 1);
872 cond_code = TREE_CODE (cond);
874 /* Get the current value of both operands. */
875 if (TREE_CODE (op0) == SSA_NAME)
877 tree tmp = get_value_for (op0, const_and_copies);
878 if (tmp)
879 op0 = tmp;
882 if (TREE_CODE (op1) == SSA_NAME)
884 tree tmp = get_value_for (op1, const_and_copies);
885 if (tmp)
886 op1 = tmp;
889 /* Stuff the operator and operands into our dummy conditional
890 expression, creating the dummy conditional if necessary. */
891 dummy_cond = walk_data->global_data;
892 if (! dummy_cond)
894 dummy_cond = build (cond_code, boolean_type_node, op0, op1);
895 dummy_cond = build (COND_EXPR, void_type_node,
896 dummy_cond, NULL, NULL);
897 walk_data->global_data = dummy_cond;
899 else
901 TREE_SET_CODE (TREE_OPERAND (dummy_cond, 0), cond_code);
902 TREE_OPERAND (TREE_OPERAND (dummy_cond, 0), 0) = op0;
903 TREE_OPERAND (TREE_OPERAND (dummy_cond, 0), 1) = op1;
906 /* If the conditional folds to an invariant, then we are done,
907 otherwise look it up in the hash tables. */
908 cached_lhs = local_fold (COND_EXPR_COND (dummy_cond));
909 if (! is_gimple_min_invariant (cached_lhs))
910 cached_lhs = lookup_avail_expr (dummy_cond, NULL, false);
911 if (!cached_lhs || ! is_gimple_min_invariant (cached_lhs))
913 stmt_ann_t ann = get_stmt_ann (dummy_cond);
914 cached_lhs = simplify_cond_and_lookup_avail_expr (dummy_cond,
915 NULL,
916 ann,
917 false);
920 /* We can have conditionals which just test the state of a
921 variable rather than use a relational operator. These are
922 simpler to handle. */
923 else if (TREE_CODE (cond) == SSA_NAME)
925 cached_lhs = cond;
926 cached_lhs = get_value_for (cached_lhs, const_and_copies);
927 if (cached_lhs && ! is_gimple_min_invariant (cached_lhs))
928 cached_lhs = 0;
930 else
931 cached_lhs = lookup_avail_expr (stmt, NULL, false);
933 if (cached_lhs)
935 edge taken_edge = find_taken_edge (e->dest, cached_lhs);
936 basic_block dest = (taken_edge ? taken_edge->dest : NULL);
938 if (dest == e->dest)
939 return;
941 /* If we have a known destination for the conditional, then
942 we can perform this optimization, which saves at least one
943 conditional jump each time it applies since we get to
944 bypass the conditional at our original destination.
946 Note that we can either thread through a block with PHIs
947 or to a block with PHIs, but not both. At this time the
948 bookkeeping to keep the CFG & SSA up-to-date has proven
949 difficult. */
950 if (dest)
952 int saved_forwardable = bb_ann (e->src)->forwardable;
953 edge tmp_edge;
955 bb_ann (e->src)->forwardable = 0;
956 tmp_edge = tree_block_forwards_to (dest);
957 taken_edge = (tmp_edge ? tmp_edge : taken_edge);
958 bb_ann (e->src)->forwardable = saved_forwardable;
959 VARRAY_PUSH_EDGE (redirection_edges, e);
960 VARRAY_PUSH_EDGE (redirection_edges, taken_edge);
967 /* Initialize the local stacks.
969 AVAIL_EXPRS stores all the expressions made available in this block.
971 CONST_AND_COPIES stores var/value pairs to restore at the end of this
972 block.
974 NONZERO_VARS stores the vars which have a nonzero value made in this
975 block.
977 STMTS_TO_RESCAN is a list of statements we will rescan for operands.
979 VRP_VARIABLES is the list of variables which have had their values
980 constrained by an operation in this block.
982 These stacks are cleared in the finalization routine run for each
983 block. */
985 static void
986 dom_opt_initialize_block_local_data (struct dom_walk_data *walk_data ATTRIBUTE_UNUSED,
987 basic_block bb ATTRIBUTE_UNUSED,
988 bool recycled ATTRIBUTE_UNUSED)
990 #ifdef ENABLE_CHECKING
991 struct dom_walk_block_data *bd
992 = (struct dom_walk_block_data *)VARRAY_TOP_GENERIC_PTR (walk_data->block_data_stack);
994 /* We get cleared memory from the allocator, so if the memory is not
995 cleared, then we are re-using a previously allocated entry. In
996 that case, we can also re-use the underlying virtual arrays. Just
997 make sure we clear them before using them! */
998 if (recycled)
1000 if (bd->avail_exprs && VARRAY_ACTIVE_SIZE (bd->avail_exprs) > 0)
1001 abort ();
1002 if (bd->const_and_copies && VARRAY_ACTIVE_SIZE (bd->const_and_copies) > 0)
1003 abort ();
1004 if (bd->nonzero_vars && VARRAY_ACTIVE_SIZE (bd->nonzero_vars) > 0)
1005 abort ();
1006 if (bd->stmts_to_rescan && VARRAY_ACTIVE_SIZE (bd->stmts_to_rescan) > 0)
1007 abort ();
1008 if (bd->vrp_variables && VARRAY_ACTIVE_SIZE (bd->vrp_variables) > 0)
1009 abort ();
1010 if (bd->block_defs && VARRAY_ACTIVE_SIZE (bd->block_defs) > 0)
1011 abort ();
1013 #endif
1016 /* Initialize local stacks for this optimizer and record equivalences
1017 upon entry to BB. Equivalences can come from the edge traversed to
1018 reach BB or they may come from PHI nodes at the start of BB. */
1020 static void
1021 dom_opt_initialize_block (struct dom_walk_data *walk_data, basic_block bb)
1023 if (dump_file && (dump_flags & TDF_DETAILS))
1024 fprintf (dump_file, "\n\nOptimizing block #%d\n\n", bb->index);
1026 record_equivalences_from_incoming_edge (walk_data, bb);
1028 /* PHI nodes can create equivalences too. */
1029 record_equivalences_from_phis (walk_data, bb);
1032 /* Given an expression EXPR (a relational expression or a statement),
1033 initialize the hash table element pointed by by ELEMENT. */
1035 static void
1036 initialize_hash_element (tree expr, tree lhs, struct expr_hash_elt *element)
1038 /* Hash table elements may be based on conditional expressions or statements.
1040 For the former case, we have no annotation and we want to hash the
1041 conditional expression. In the latter case we have an annotation and
1042 we want to record the expression the statement evaluates. */
1043 if (TREE_CODE_CLASS (TREE_CODE (expr)) == '<'
1044 || TREE_CODE (expr) == TRUTH_NOT_EXPR)
1046 element->ann = NULL;
1047 element->rhs = expr;
1049 else if (TREE_CODE (expr) == COND_EXPR)
1051 element->ann = stmt_ann (expr);
1052 element->rhs = COND_EXPR_COND (expr);
1054 else if (TREE_CODE (expr) == SWITCH_EXPR)
1056 element->ann = stmt_ann (expr);
1057 element->rhs = SWITCH_COND (expr);
1059 else if (TREE_CODE (expr) == RETURN_EXPR && TREE_OPERAND (expr, 0))
1061 element->ann = stmt_ann (expr);
1062 element->rhs = TREE_OPERAND (TREE_OPERAND (expr, 0), 1);
1064 else
1066 element->ann = stmt_ann (expr);
1067 element->rhs = TREE_OPERAND (expr, 1);
1070 element->lhs = lhs;
1071 element->hash = avail_expr_hash (element);
1074 /* Remove all the expressions in LOCALS from TABLE, stopping when there are
1075 LIMIT entries left in LOCALs. */
1077 static void
1078 remove_local_expressions_from_table (varray_type locals,
1079 unsigned limit,
1080 htab_t table)
1082 if (! locals)
1083 return;
1085 /* Remove all the expressions made available in this block. */
1086 while (VARRAY_ACTIVE_SIZE (locals) > limit)
1088 struct expr_hash_elt element;
1089 tree expr = VARRAY_TOP_TREE (locals);
1090 VARRAY_POP (locals);
1092 initialize_hash_element (expr, NULL, &element);
1093 htab_remove_elt_with_hash (table, &element, element.hash);
1097 /* Use the SSA_NAMES in LOCALS to restore TABLE to its original
1098 state, stopping when there are LIMIT entries left in LOCALs. */
1100 static void
1101 restore_nonzero_vars_to_original_value (varray_type locals,
1102 unsigned limit,
1103 bitmap table)
1105 if (!locals)
1106 return;
1108 while (VARRAY_ACTIVE_SIZE (locals) > limit)
1110 tree name = VARRAY_TOP_TREE (locals);
1111 VARRAY_POP (locals);
1112 bitmap_clear_bit (table, SSA_NAME_VERSION (name));
1116 /* Use the source/dest pairs in LOCALS to restore TABLE to its original
1117 state, stopping when there are LIMIT entries left in LOCALs. */
1119 static void
1120 restore_vars_to_original_value (varray_type locals,
1121 unsigned limit,
1122 varray_type table)
1124 if (! locals)
1125 return;
1127 while (VARRAY_ACTIVE_SIZE (locals) > limit)
1129 tree prev_value, dest;
1131 prev_value = VARRAY_TOP_TREE (locals);
1132 VARRAY_POP (locals);
1133 dest = VARRAY_TOP_TREE (locals);
1134 VARRAY_POP (locals);
1136 set_value_for (dest, prev_value, table);
1140 /* Similar to restore_vars_to_original_value, except that it restores
1141 CURRDEFS to its original value. */
1142 static void
1143 restore_currdefs_to_original_value (varray_type locals, unsigned limit)
1145 if (!locals)
1146 return;
1148 /* Restore CURRDEFS to its original state. */
1149 while (VARRAY_ACTIVE_SIZE (locals) > limit)
1151 tree tmp = VARRAY_TOP_TREE (locals);
1152 tree saved_def, var;
1154 VARRAY_POP (locals);
1156 /* If we recorded an SSA_NAME, then make the SSA_NAME the current
1157 definition of its underlying variable. If we recorded anything
1158 else, it must have been an _DECL node and its current reaching
1159 definition must have been NULL. */
1160 if (TREE_CODE (tmp) == SSA_NAME)
1162 saved_def = tmp;
1163 var = SSA_NAME_VAR (saved_def);
1165 else
1167 saved_def = NULL;
1168 var = tmp;
1171 var_ann (var)->current_def = saved_def;
1175 /* We have finished processing the dominator children of BB, perform
1176 any finalization actions in preparation for leaving this node in
1177 the dominator tree. */
1179 static void
1180 dom_opt_finalize_block (struct dom_walk_data *walk_data, basic_block bb)
1182 struct dom_walk_block_data *bd
1183 = VARRAY_TOP_GENERIC_PTR (walk_data->block_data_stack);
1184 tree last;
1186 /* If we are at a leaf node in the dominator graph, see if we can thread
1187 the edge from BB through its successor.
1189 Do this before we remove entries from our equivalence tables. */
1190 if (bb->succ
1191 && ! bb->succ->succ_next
1192 && (bb->succ->flags & EDGE_ABNORMAL) == 0
1193 && (get_immediate_dominator (CDI_DOMINATORS, bb->succ->dest) != bb
1194 || phi_nodes (bb->succ->dest)))
1197 thread_across_edge (walk_data, bb->succ);
1199 else if ((last = last_stmt (bb))
1200 && TREE_CODE (last) == COND_EXPR
1201 && (TREE_CODE_CLASS (TREE_CODE (COND_EXPR_COND (last))) == '<'
1202 || TREE_CODE (COND_EXPR_COND (last)) == SSA_NAME)
1203 && bb->succ
1204 && (bb->succ->flags & EDGE_ABNORMAL) == 0
1205 && bb->succ->succ_next
1206 && (bb->succ->succ_next->flags & EDGE_ABNORMAL) == 0
1207 && ! bb->succ->succ_next->succ_next)
1209 edge true_edge, false_edge;
1210 tree cond, inverted = NULL;
1211 enum tree_code cond_code;
1213 extract_true_false_edges_from_block (bb, &true_edge, &false_edge);
1215 cond = COND_EXPR_COND (last);
1216 cond_code = TREE_CODE (cond);
1218 if (TREE_CODE_CLASS (cond_code) == '<')
1219 inverted = invert_truthvalue (cond);
1221 /* If the THEN arm is the end of a dominator tree or has PHI nodes,
1222 then try to thread through its edge. */
1223 if (get_immediate_dominator (CDI_DOMINATORS, true_edge->dest) != bb
1224 || phi_nodes (true_edge->dest))
1226 unsigned avail_expr_limit;
1227 unsigned const_and_copies_limit;
1228 unsigned currdefs_limit;
1230 avail_expr_limit
1231 = bd->avail_exprs ? VARRAY_ACTIVE_SIZE (bd->avail_exprs) : 0;
1232 const_and_copies_limit
1233 = bd->const_and_copies ? VARRAY_ACTIVE_SIZE (bd->const_and_copies)
1234 : 0;
1235 currdefs_limit
1236 = bd->block_defs ? VARRAY_ACTIVE_SIZE (bd->block_defs) : 0;
1238 /* Record any equivalences created by following this edge. */
1239 if (TREE_CODE_CLASS (cond_code) == '<')
1241 record_cond (cond, boolean_true_node, &bd->avail_exprs);
1242 record_dominating_conditions (cond, &bd->avail_exprs);
1243 record_cond (inverted, boolean_false_node, &bd->avail_exprs);
1245 else if (cond_code == SSA_NAME)
1246 record_const_or_copy (cond, boolean_true_node,
1247 &bd->const_and_copies);
1249 /* Now thread the edge. */
1250 thread_across_edge (walk_data, true_edge);
1252 /* And restore the various tables to their state before
1253 we threaded this edge. */
1254 remove_local_expressions_from_table (bd->avail_exprs,
1255 avail_expr_limit,
1256 avail_exprs);
1257 restore_vars_to_original_value (bd->const_and_copies,
1258 const_and_copies_limit,
1259 const_and_copies);
1260 restore_currdefs_to_original_value (bd->block_defs, currdefs_limit);
1263 /* Similarly for the ELSE arm. */
1264 if (get_immediate_dominator (CDI_DOMINATORS, false_edge->dest) != bb
1265 || phi_nodes (false_edge->dest))
1267 /* Record any equivalences created by following this edge. */
1268 if (TREE_CODE_CLASS (cond_code) == '<')
1270 record_cond (cond, boolean_false_node, &bd->avail_exprs);
1271 record_cond (inverted, boolean_true_node, &bd->avail_exprs);
1272 record_dominating_conditions (inverted, &bd->avail_exprs);
1274 else if (cond_code == SSA_NAME)
1275 record_const_or_copy (cond, boolean_false_node,
1276 &bd->const_and_copies);
1278 thread_across_edge (walk_data, false_edge);
1280 /* No need to remove local expressions from our tables
1281 or restore vars to their original value as that will
1282 be done immediately below. */
1286 remove_local_expressions_from_table (bd->avail_exprs, 0, avail_exprs);
1287 restore_nonzero_vars_to_original_value (bd->nonzero_vars, 0, nonzero_vars);
1288 restore_vars_to_original_value (bd->const_and_copies, 0, const_and_copies);
1289 restore_currdefs_to_original_value (bd->block_defs, 0);
1291 /* Remove VRP records associated with this basic block. They are no
1292 longer valid.
1294 To be efficient, we note which variables have had their values
1295 constrained in this block. So walk over each variable in the
1296 VRP_VARIABLEs array. */
1297 while (bd->vrp_variables && VARRAY_ACTIVE_SIZE (bd->vrp_variables) > 0)
1299 tree var = VARRAY_TOP_TREE (bd->vrp_variables);
1301 /* Each variable has a stack of value range records. We want to
1302 invalidate those associated with our basic block. So we walk
1303 the array backwards popping off records associated with our
1304 block. Once we hit a record not associated with our block
1305 we are done. */
1306 varray_type var_vrp_records = VARRAY_GENERIC_PTR (vrp_data,
1307 SSA_NAME_VERSION (var));
1309 while (VARRAY_ACTIVE_SIZE (var_vrp_records) > 0)
1311 struct vrp_element *element
1312 = (struct vrp_element *)VARRAY_TOP_GENERIC_PTR (var_vrp_records);
1314 if (element->bb != bb)
1315 break;
1317 VARRAY_POP (var_vrp_records);
1320 VARRAY_POP (bd->vrp_variables);
1323 /* Re-scan operands in all statements that may have had new symbols
1324 exposed. */
1325 while (bd->stmts_to_rescan && VARRAY_ACTIVE_SIZE (bd->stmts_to_rescan) > 0)
1327 tree stmt = VARRAY_TOP_TREE (bd->stmts_to_rescan);
1328 VARRAY_POP (bd->stmts_to_rescan);
1329 mark_new_vars_to_rename (stmt, vars_to_rename);
1333 /* PHI nodes can create equivalences too.
1335 Ignoring any alternatives which are the same as the result, if
1336 all the alternatives are equal, then the PHI node creates an
1337 equivalence.
1339 Additionally, if all the PHI alternatives are known to have a nonzero
1340 value, then the result of this PHI is known to have a nonzero value,
1341 even if we do not know its exact value. */
1343 static void
1344 record_equivalences_from_phis (struct dom_walk_data *walk_data, basic_block bb)
1346 struct dom_walk_block_data *bd
1347 = VARRAY_TOP_GENERIC_PTR (walk_data->block_data_stack);
1348 tree phi;
1350 for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi))
1352 tree lhs = PHI_RESULT (phi);
1353 tree rhs = NULL;
1354 int i;
1356 for (i = 0; i < PHI_NUM_ARGS (phi); i++)
1358 tree t = PHI_ARG_DEF (phi, i);
1360 if (TREE_CODE (t) == SSA_NAME || is_gimple_min_invariant (t))
1362 /* Ignore alternatives which are the same as our LHS. */
1363 if (operand_equal_p (lhs, t, 0))
1364 continue;
1366 /* If we have not processed an alternative yet, then set
1367 RHS to this alternative. */
1368 if (rhs == NULL)
1369 rhs = t;
1370 /* If we have processed an alternative (stored in RHS), then
1371 see if it is equal to this one. If it isn't, then stop
1372 the search. */
1373 else if (! operand_equal_p (rhs, t, 0))
1374 break;
1376 else
1377 break;
1380 /* If we had no interesting alternatives, then all the RHS alternatives
1381 must have been the same as LHS. */
1382 if (!rhs)
1383 rhs = lhs;
1385 /* If we managed to iterate through each PHI alternative without
1386 breaking out of the loop, then we have a PHI which may create
1387 a useful equivalence. We do not need to record unwind data for
1388 this, since this is a true assignment and not an equivalence
1389 inferred from a comparison. All uses of this ssa name are dominated
1390 by this assignment, so unwinding just costs time and space. */
1391 if (i == PHI_NUM_ARGS (phi)
1392 && may_propagate_copy (lhs, rhs))
1393 set_value_for (lhs, rhs, const_and_copies);
1395 /* Now see if we know anything about the nonzero property for the
1396 result of this PHI. */
1397 for (i = 0; i < PHI_NUM_ARGS (phi); i++)
1399 if (!PHI_ARG_NONZERO (phi, i))
1400 break;
1403 if (i == PHI_NUM_ARGS (phi))
1404 bitmap_set_bit (nonzero_vars, SSA_NAME_VERSION (PHI_RESULT (phi)));
1406 register_new_def (lhs, &bd->block_defs);
1410 /* Record any equivalences created by the incoming edge to BB. If BB
1411 has more than one incoming edge, then no equivalence is created. */
1413 static void
1414 record_equivalences_from_incoming_edge (struct dom_walk_data *walk_data,
1415 basic_block bb)
1417 int edge_flags;
1418 basic_block parent;
1419 struct eq_expr_value eq_expr_value;
1420 tree parent_block_last_stmt = NULL;
1421 struct dom_walk_block_data *bd
1422 = VARRAY_TOP_GENERIC_PTR (walk_data->block_data_stack);
1424 /* If our parent block ended with a control statment, then we may be
1425 able to record some equivalences based on which outgoing edge from
1426 the parent was followed. */
1427 parent = get_immediate_dominator (CDI_DOMINATORS, bb);
1428 if (parent)
1430 parent_block_last_stmt = last_stmt (parent);
1431 if (parent_block_last_stmt && !is_ctrl_stmt (parent_block_last_stmt))
1432 parent_block_last_stmt = NULL;
1435 eq_expr_value.src = NULL;
1436 eq_expr_value.dst = NULL;
1438 /* If we have a single predecessor, then extract EDGE_FLAGS from
1439 our single incoming edge. Otherwise clear EDGE_FLAGS and
1440 PARENT_BLOCK_LAST_STMT since they're not needed. */
1441 if (bb->pred
1442 && ! bb->pred->pred_next
1443 && parent_block_last_stmt
1444 && bb_for_stmt (parent_block_last_stmt) == bb->pred->src)
1446 edge_flags = bb->pred->flags;
1448 else
1450 edge_flags = 0;
1451 parent_block_last_stmt = NULL;
1454 /* If our parent block ended in a COND_EXPR, add any equivalences
1455 created by the COND_EXPR to the hash table and initialize
1456 EQ_EXPR_VALUE appropriately.
1458 EQ_EXPR_VALUE is an assignment expression created when BB's immediate
1459 dominator ends in a COND_EXPR statement whose predicate is of the form
1460 'VAR == VALUE', where VALUE may be another variable or a constant.
1461 This is used to propagate VALUE on the THEN_CLAUSE of that
1462 conditional. This assignment is inserted in CONST_AND_COPIES so that
1463 the copy and constant propagator can find more propagation
1464 opportunities. */
1465 if (parent_block_last_stmt
1466 && bb->pred->pred_next == NULL
1467 && TREE_CODE (parent_block_last_stmt) == COND_EXPR
1468 && (edge_flags & (EDGE_TRUE_VALUE | EDGE_FALSE_VALUE)))
1469 eq_expr_value = get_eq_expr_value (parent_block_last_stmt,
1470 (edge_flags & EDGE_TRUE_VALUE) != 0,
1471 &bd->avail_exprs,
1473 &bd->vrp_variables);
1474 /* Similarly when the parent block ended in a SWITCH_EXPR.
1475 We can only know the value of the switch's condition if the dominator
1476 parent is also the only predecessor of this block. */
1477 else if (parent_block_last_stmt
1478 && bb->pred->pred_next == NULL
1479 && bb->pred->src == parent
1480 && TREE_CODE (parent_block_last_stmt) == SWITCH_EXPR)
1482 tree switch_cond = SWITCH_COND (parent_block_last_stmt);
1484 /* If the switch's condition is an SSA variable, then we may
1485 know its value at each of the case labels. */
1486 if (TREE_CODE (switch_cond) == SSA_NAME)
1488 tree switch_vec = SWITCH_LABELS (parent_block_last_stmt);
1489 size_t i, n = TREE_VEC_LENGTH (switch_vec);
1490 int case_count = 0;
1491 tree match_case = NULL_TREE;
1493 /* Search the case labels for those whose destination is
1494 the current basic block. */
1495 for (i = 0; i < n; ++i)
1497 tree elt = TREE_VEC_ELT (switch_vec, i);
1498 if (label_to_block (CASE_LABEL (elt)) == bb)
1500 if (++case_count > 1 || CASE_HIGH (elt))
1501 break;
1502 match_case = elt;
1506 /* If we encountered precisely one CASE_LABEL_EXPR and it
1507 was not the default case, or a case range, then we know
1508 the exact value of SWITCH_COND which caused us to get to
1509 this block. Record that equivalence in EQ_EXPR_VALUE. */
1510 if (case_count == 1
1511 && match_case
1512 && CASE_LOW (match_case)
1513 && !CASE_HIGH (match_case))
1515 eq_expr_value.dst = switch_cond;
1516 eq_expr_value.src = CASE_LOW (match_case);
1521 /* If EQ_EXPR_VALUE (VAR == VALUE) is given, register the VALUE as a
1522 new value for VAR, so that occurrences of VAR can be replaced with
1523 VALUE while re-writing the THEN arm of a COND_EXPR. */
1524 if (eq_expr_value.src && eq_expr_value.dst)
1525 record_equality (eq_expr_value.dst, eq_expr_value.src,
1526 &bd->const_and_copies);
1529 /* Dump SSA statistics on FILE. */
1531 void
1532 dump_dominator_optimization_stats (FILE *file)
1534 long n_exprs;
1536 fprintf (file, "Total number of statements: %6ld\n\n",
1537 opt_stats.num_stmts);
1538 fprintf (file, "Exprs considered for dominator optimizations: %6ld\n",
1539 opt_stats.num_exprs_considered);
1541 n_exprs = opt_stats.num_exprs_considered;
1542 if (n_exprs == 0)
1543 n_exprs = 1;
1545 fprintf (file, " Redundant expressions eliminated: %6ld (%.0f%%)\n",
1546 opt_stats.num_re, PERCENT (opt_stats.num_re,
1547 n_exprs));
1549 fprintf (file, "\nHash table statistics:\n");
1551 fprintf (file, " avail_exprs: ");
1552 htab_statistics (file, avail_exprs);
1556 /* Dump SSA statistics on stderr. */
1558 void
1559 debug_dominator_optimization_stats (void)
1561 dump_dominator_optimization_stats (stderr);
1565 /* Dump statistics for the hash table HTAB. */
1567 static void
1568 htab_statistics (FILE *file, htab_t htab)
1570 fprintf (file, "size %ld, %ld elements, %f collision/search ratio\n",
1571 (long) htab_size (htab),
1572 (long) htab_elements (htab),
1573 htab_collisions (htab));
1576 /* Record the fact that VAR has a nonzero value, though we may not know
1577 its exact value. Note that if VAR is already known to have a nonzero
1578 value, then we do nothing. */
1580 static void
1581 record_var_is_nonzero (tree var, varray_type *block_nonzero_vars_p)
1583 int indx = SSA_NAME_VERSION (var);
1585 if (bitmap_bit_p (nonzero_vars, indx))
1586 return;
1588 /* Mark it in the global table. */
1589 bitmap_set_bit (nonzero_vars, indx);
1591 /* Record this SSA_NAME so that we can reset the global table
1592 when we leave this block. */
1593 if (! *block_nonzero_vars_p)
1594 VARRAY_TREE_INIT (*block_nonzero_vars_p, 2, "block_nonzero_vars");
1595 VARRAY_PUSH_TREE (*block_nonzero_vars_p, var);
1598 /* Enter a statement into the true/false expression hash table indicating
1599 that the condition COND has the value VALUE. */
1601 static void
1602 record_cond (tree cond, tree value, varray_type *block_avail_exprs_p)
1604 struct expr_hash_elt *element = xmalloc (sizeof (struct expr_hash_elt));
1605 void **slot;
1607 initialize_hash_element (cond, value, element);
1609 slot = htab_find_slot_with_hash (avail_exprs, (void *)element,
1610 element->hash, true);
1611 if (*slot == NULL)
1613 *slot = (void *) element;
1614 if (! *block_avail_exprs_p)
1615 VARRAY_TREE_INIT (*block_avail_exprs_p, 20, "block_avail_exprs");
1616 VARRAY_PUSH_TREE (*block_avail_exprs_p, cond);
1618 else
1619 free (element);
1622 /* COND is a condition which is known to be true. Record variants of
1623 COND which must also be true.
1625 For example, if a < b is true, then a <= b must also be true. */
1627 static void
1628 record_dominating_conditions (tree cond, varray_type *block_avail_exprs_p)
1630 switch (TREE_CODE (cond))
1632 case LT_EXPR:
1633 record_cond (build2 (LE_EXPR, boolean_type_node,
1634 TREE_OPERAND (cond, 0),
1635 TREE_OPERAND (cond, 1)),
1636 boolean_true_node,
1637 block_avail_exprs_p);
1638 record_cond (build2 (ORDERED_EXPR, boolean_type_node,
1639 TREE_OPERAND (cond, 0),
1640 TREE_OPERAND (cond, 1)),
1641 boolean_true_node,
1642 block_avail_exprs_p);
1643 record_cond (build2 (NE_EXPR, boolean_type_node,
1644 TREE_OPERAND (cond, 0),
1645 TREE_OPERAND (cond, 1)),
1646 boolean_true_node,
1647 block_avail_exprs_p);
1648 record_cond (build2 (LTGT_EXPR, boolean_type_node,
1649 TREE_OPERAND (cond, 0),
1650 TREE_OPERAND (cond, 1)),
1651 boolean_true_node,
1652 block_avail_exprs_p);
1653 break;
1655 case GT_EXPR:
1656 record_cond (build2 (GE_EXPR, boolean_type_node,
1657 TREE_OPERAND (cond, 0),
1658 TREE_OPERAND (cond, 1)),
1659 boolean_true_node,
1660 block_avail_exprs_p);
1661 record_cond (build2 (ORDERED_EXPR, boolean_type_node,
1662 TREE_OPERAND (cond, 0),
1663 TREE_OPERAND (cond, 1)),
1664 boolean_true_node,
1665 block_avail_exprs_p);
1666 record_cond (build2 (NE_EXPR, boolean_type_node,
1667 TREE_OPERAND (cond, 0),
1668 TREE_OPERAND (cond, 1)),
1669 boolean_true_node,
1670 block_avail_exprs_p);
1671 record_cond (build2 (LTGT_EXPR, boolean_type_node,
1672 TREE_OPERAND (cond, 0),
1673 TREE_OPERAND (cond, 1)),
1674 boolean_true_node,
1675 block_avail_exprs_p);
1676 break;
1678 case GE_EXPR:
1679 case LE_EXPR:
1680 record_cond (build2 (ORDERED_EXPR, boolean_type_node,
1681 TREE_OPERAND (cond, 0),
1682 TREE_OPERAND (cond, 1)),
1683 boolean_true_node,
1684 block_avail_exprs_p);
1685 break;
1687 case EQ_EXPR:
1688 record_cond (build2 (ORDERED_EXPR, boolean_type_node,
1689 TREE_OPERAND (cond, 0),
1690 TREE_OPERAND (cond, 1)),
1691 boolean_true_node,
1692 block_avail_exprs_p);
1693 record_cond (build2 (LE_EXPR, boolean_type_node,
1694 TREE_OPERAND (cond, 0),
1695 TREE_OPERAND (cond, 1)),
1696 boolean_true_node,
1697 block_avail_exprs_p);
1698 record_cond (build2 (GE_EXPR, boolean_type_node,
1699 TREE_OPERAND (cond, 0),
1700 TREE_OPERAND (cond, 1)),
1701 boolean_true_node,
1702 block_avail_exprs_p);
1703 break;
1705 case UNORDERED_EXPR:
1706 record_cond (build2 (NE_EXPR, boolean_type_node,
1707 TREE_OPERAND (cond, 0),
1708 TREE_OPERAND (cond, 1)),
1709 boolean_true_node,
1710 block_avail_exprs_p);
1711 record_cond (build2 (UNLE_EXPR, boolean_type_node,
1712 TREE_OPERAND (cond, 0),
1713 TREE_OPERAND (cond, 1)),
1714 boolean_true_node,
1715 block_avail_exprs_p);
1716 record_cond (build2 (UNGE_EXPR, boolean_type_node,
1717 TREE_OPERAND (cond, 0),
1718 TREE_OPERAND (cond, 1)),
1719 boolean_true_node,
1720 block_avail_exprs_p);
1721 record_cond (build2 (UNEQ_EXPR, boolean_type_node,
1722 TREE_OPERAND (cond, 0),
1723 TREE_OPERAND (cond, 1)),
1724 boolean_true_node,
1725 block_avail_exprs_p);
1726 record_cond (build2 (UNLT_EXPR, boolean_type_node,
1727 TREE_OPERAND (cond, 0),
1728 TREE_OPERAND (cond, 1)),
1729 boolean_true_node,
1730 block_avail_exprs_p);
1731 record_cond (build2 (UNGT_EXPR, boolean_type_node,
1732 TREE_OPERAND (cond, 0),
1733 TREE_OPERAND (cond, 1)),
1734 boolean_true_node,
1735 block_avail_exprs_p);
1736 break;
1738 case UNLT_EXPR:
1739 record_cond (build2 (UNLE_EXPR, boolean_type_node,
1740 TREE_OPERAND (cond, 0),
1741 TREE_OPERAND (cond, 1)),
1742 boolean_true_node,
1743 block_avail_exprs_p);
1744 record_cond (build2 (NE_EXPR, boolean_type_node,
1745 TREE_OPERAND (cond, 0),
1746 TREE_OPERAND (cond, 1)),
1747 boolean_true_node,
1748 block_avail_exprs_p);
1749 break;
1751 case UNGT_EXPR:
1752 record_cond (build2 (UNGE_EXPR, boolean_type_node,
1753 TREE_OPERAND (cond, 0),
1754 TREE_OPERAND (cond, 1)),
1755 boolean_true_node,
1756 block_avail_exprs_p);
1757 record_cond (build2 (NE_EXPR, boolean_type_node,
1758 TREE_OPERAND (cond, 0),
1759 TREE_OPERAND (cond, 1)),
1760 boolean_true_node,
1761 block_avail_exprs_p);
1762 break;
1764 case UNEQ_EXPR:
1765 record_cond (build2 (UNLE_EXPR, boolean_type_node,
1766 TREE_OPERAND (cond, 0),
1767 TREE_OPERAND (cond, 1)),
1768 boolean_true_node,
1769 block_avail_exprs_p);
1770 record_cond (build2 (UNGE_EXPR, boolean_type_node,
1771 TREE_OPERAND (cond, 0),
1772 TREE_OPERAND (cond, 1)),
1773 boolean_true_node,
1774 block_avail_exprs_p);
1775 break;
1777 case LTGT_EXPR:
1778 record_cond (build2 (NE_EXPR, boolean_type_node,
1779 TREE_OPERAND (cond, 0),
1780 TREE_OPERAND (cond, 1)),
1781 boolean_true_node,
1782 block_avail_exprs_p);
1783 record_cond (build2 (ORDERED_EXPR, boolean_type_node,
1784 TREE_OPERAND (cond, 0),
1785 TREE_OPERAND (cond, 1)),
1786 boolean_true_node,
1787 block_avail_exprs_p);
1789 default:
1790 break;
1794 /* A helper function for record_const_or_copy and record_equality.
1795 Do the work of recording the value and undo info. */
1797 static void
1798 record_const_or_copy_1 (tree x, tree y, tree prev_x,
1799 varray_type *block_const_and_copies_p)
1801 set_value_for (x, y, const_and_copies);
1803 if (!*block_const_and_copies_p)
1804 VARRAY_TREE_INIT (*block_const_and_copies_p, 2, "block_const_and_copies");
1805 VARRAY_PUSH_TREE (*block_const_and_copies_p, x);
1806 VARRAY_PUSH_TREE (*block_const_and_copies_p, prev_x);
1809 /* Record that X is equal to Y in const_and_copies. Record undo
1810 information in the block-local varray. */
1812 static void
1813 record_const_or_copy (tree x, tree y, varray_type *block_const_and_copies_p)
1815 tree prev_x = get_value_for (x, const_and_copies);
1817 if (TREE_CODE (y) == SSA_NAME)
1819 tree tmp = get_value_for (y, const_and_copies);
1820 if (tmp)
1821 y = tmp;
1824 record_const_or_copy_1 (x, y, prev_x, block_const_and_copies_p);
1827 /* Similarly, but assume that X and Y are the two operands of an EQ_EXPR.
1828 This constrains the cases in which we may treat this as assignment. */
1830 static void
1831 record_equality (tree x, tree y, varray_type *block_const_and_copies_p)
1833 tree prev_x = NULL, prev_y = NULL;
1835 if (TREE_CODE (x) == SSA_NAME)
1836 prev_x = get_value_for (x, const_and_copies);
1837 if (TREE_CODE (y) == SSA_NAME)
1838 prev_y = get_value_for (y, const_and_copies);
1840 /* If one of the previous values is invariant, then use that.
1841 Otherwise it doesn't matter which value we choose, just so
1842 long as we canonicalize on one value. */
1843 if (TREE_INVARIANT (y))
1845 else if (TREE_INVARIANT (x))
1846 prev_x = x, x = y, y = prev_x, prev_x = prev_y;
1847 else if (prev_x && TREE_INVARIANT (prev_x))
1848 x = y, y = prev_x, prev_x = prev_y;
1849 else if (prev_y)
1850 y = prev_y;
1852 /* After the swapping, we must have one SSA_NAME. */
1853 if (TREE_CODE (x) != SSA_NAME)
1854 return;
1856 /* For IEEE, -0.0 == 0.0, so we don't necessarily know the sign of a
1857 variable compared against zero. If we're honoring signed zeros,
1858 then we cannot record this value unless we know that the value is
1859 nonzero. */
1860 if (HONOR_SIGNED_ZEROS (TYPE_MODE (TREE_TYPE (x)))
1861 && (TREE_CODE (y) != REAL_CST
1862 || REAL_VALUES_EQUAL (dconst0, TREE_REAL_CST (y))))
1863 return;
1865 record_const_or_copy_1 (x, y, prev_x, block_const_and_copies_p);
1868 /* STMT is a MODIFY_EXPR for which we were unable to find RHS in the
1869 hash tables. Try to simplify the RHS using whatever equivalences
1870 we may have recorded.
1872 If we are able to simplify the RHS, then lookup the simplified form in
1873 the hash table and return the result. Otherwise return NULL. */
1875 static tree
1876 simplify_rhs_and_lookup_avail_expr (struct dom_walk_data *walk_data,
1877 tree stmt,
1878 stmt_ann_t ann,
1879 int insert)
1881 tree rhs = TREE_OPERAND (stmt, 1);
1882 enum tree_code rhs_code = TREE_CODE (rhs);
1883 tree result = NULL;
1884 struct dom_walk_block_data *bd
1885 = VARRAY_TOP_GENERIC_PTR (walk_data->block_data_stack);
1887 /* If we have lhs = ~x, look and see if we earlier had x = ~y.
1888 In which case we can change this statement to be lhs = y.
1889 Which can then be copy propagated.
1891 Similarly for negation. */
1892 if ((rhs_code == BIT_NOT_EXPR || rhs_code == NEGATE_EXPR)
1893 && TREE_CODE (TREE_OPERAND (rhs, 0)) == SSA_NAME)
1895 /* Get the definition statement for our RHS. */
1896 tree rhs_def_stmt = SSA_NAME_DEF_STMT (TREE_OPERAND (rhs, 0));
1898 /* See if the RHS_DEF_STMT has the same form as our statement. */
1899 if (TREE_CODE (rhs_def_stmt) == MODIFY_EXPR
1900 && TREE_CODE (TREE_OPERAND (rhs_def_stmt, 1)) == rhs_code)
1902 tree rhs_def_operand;
1904 rhs_def_operand = TREE_OPERAND (TREE_OPERAND (rhs_def_stmt, 1), 0);
1906 /* Verify that RHS_DEF_OPERAND is a suitable SSA variable. */
1907 if (TREE_CODE (rhs_def_operand) == SSA_NAME
1908 && ! SSA_NAME_OCCURS_IN_ABNORMAL_PHI (rhs_def_operand))
1909 result = update_rhs_and_lookup_avail_expr (stmt,
1910 rhs_def_operand,
1911 &bd->avail_exprs,
1912 ann,
1913 insert);
1917 /* If we have z = (x OP C1), see if we earlier had x = y OP C2.
1918 If OP is associative, create and fold (y OP C2) OP C1 which
1919 should result in (y OP C3), use that as the RHS for the
1920 assignment. Add minus to this, as we handle it specially below. */
1921 if ((associative_tree_code (rhs_code) || rhs_code == MINUS_EXPR)
1922 && TREE_CODE (TREE_OPERAND (rhs, 0)) == SSA_NAME
1923 && is_gimple_min_invariant (TREE_OPERAND (rhs, 1)))
1925 tree rhs_def_stmt = SSA_NAME_DEF_STMT (TREE_OPERAND (rhs, 0));
1927 /* See if the RHS_DEF_STMT has the same form as our statement. */
1928 if (TREE_CODE (rhs_def_stmt) == MODIFY_EXPR)
1930 tree rhs_def_rhs = TREE_OPERAND (rhs_def_stmt, 1);
1931 enum tree_code rhs_def_code = TREE_CODE (rhs_def_rhs);
1933 if (rhs_code == rhs_def_code
1934 || (rhs_code == PLUS_EXPR && rhs_def_code == MINUS_EXPR)
1935 || (rhs_code == MINUS_EXPR && rhs_def_code == PLUS_EXPR))
1937 tree def_stmt_op0 = TREE_OPERAND (rhs_def_rhs, 0);
1938 tree def_stmt_op1 = TREE_OPERAND (rhs_def_rhs, 1);
1940 if (TREE_CODE (def_stmt_op0) == SSA_NAME
1941 && ! SSA_NAME_OCCURS_IN_ABNORMAL_PHI (def_stmt_op0)
1942 && is_gimple_min_invariant (def_stmt_op1))
1944 tree outer_const = TREE_OPERAND (rhs, 1);
1945 tree type = TREE_TYPE (TREE_OPERAND (stmt, 0));
1946 tree t;
1948 /* If we care about correct floating point results, then
1949 don't fold x + c1 - c2. Note that we need to take both
1950 the codes and the signs to figure this out. */
1951 if (FLOAT_TYPE_P (type)
1952 && !flag_unsafe_math_optimizations
1953 && (rhs_def_code == PLUS_EXPR
1954 || rhs_def_code == MINUS_EXPR))
1956 bool neg = false;
1958 neg ^= (rhs_code == MINUS_EXPR);
1959 neg ^= (rhs_def_code == MINUS_EXPR);
1960 neg ^= real_isneg (TREE_REAL_CST_PTR (outer_const));
1961 neg ^= real_isneg (TREE_REAL_CST_PTR (def_stmt_op1));
1963 if (neg)
1964 goto dont_fold_assoc;
1967 /* Ho hum. So fold will only operate on the outermost
1968 thingy that we give it, so we have to build the new
1969 expression in two pieces. This requires that we handle
1970 combinations of plus and minus. */
1971 if (rhs_def_code != rhs_code)
1973 if (rhs_def_code == MINUS_EXPR)
1974 t = build (MINUS_EXPR, type, outer_const, def_stmt_op1);
1975 else
1976 t = build (MINUS_EXPR, type, def_stmt_op1, outer_const);
1977 rhs_code = PLUS_EXPR;
1979 else if (rhs_def_code == MINUS_EXPR)
1980 t = build (PLUS_EXPR, type, def_stmt_op1, outer_const);
1981 else
1982 t = build (rhs_def_code, type, def_stmt_op1, outer_const);
1983 t = local_fold (t);
1984 t = build (rhs_code, type, def_stmt_op0, t);
1985 t = local_fold (t);
1987 /* If the result is a suitable looking gimple expression,
1988 then use it instead of the original for STMT. */
1989 if (TREE_CODE (t) == SSA_NAME
1990 || (TREE_CODE_CLASS (TREE_CODE (t)) == '1'
1991 && TREE_CODE (TREE_OPERAND (t, 0)) == SSA_NAME)
1992 || ((TREE_CODE_CLASS (TREE_CODE (t)) == '2'
1993 || TREE_CODE_CLASS (TREE_CODE (t)) == '<')
1994 && TREE_CODE (TREE_OPERAND (t, 0)) == SSA_NAME
1995 && is_gimple_val (TREE_OPERAND (t, 1))))
1996 result = update_rhs_and_lookup_avail_expr
1997 (stmt, t, &bd->avail_exprs, ann, insert);
2001 dont_fold_assoc:;
2004 /* Transform TRUNC_DIV_EXPR and TRUNC_MOD_EXPR into RSHIFT_EXPR
2005 and BIT_AND_EXPR respectively if the first operand is greater
2006 than zero and the second operand is an exact power of two. */
2007 if ((rhs_code == TRUNC_DIV_EXPR || rhs_code == TRUNC_MOD_EXPR)
2008 && INTEGRAL_TYPE_P (TREE_TYPE (TREE_OPERAND (rhs, 0)))
2009 && integer_pow2p (TREE_OPERAND (rhs, 1)))
2011 tree val;
2012 tree op = TREE_OPERAND (rhs, 0);
2014 if (TYPE_UNSIGNED (TREE_TYPE (op)))
2016 val = integer_one_node;
2018 else
2020 tree dummy_cond = walk_data->global_data;
2022 if (! dummy_cond)
2024 dummy_cond = build (GT_EXPR, boolean_type_node,
2025 op, integer_zero_node);
2026 dummy_cond = build (COND_EXPR, void_type_node,
2027 dummy_cond, NULL, NULL);
2028 walk_data->global_data = dummy_cond;
2030 else
2032 TREE_SET_CODE (TREE_OPERAND (dummy_cond, 0), GT_EXPR);
2033 TREE_OPERAND (TREE_OPERAND (dummy_cond, 0), 0) = op;
2034 TREE_OPERAND (TREE_OPERAND (dummy_cond, 0), 1)
2035 = integer_zero_node;
2037 val = simplify_cond_and_lookup_avail_expr (dummy_cond,
2038 &bd->avail_exprs,
2039 NULL, false);
2042 if (val && integer_onep (val))
2044 tree t;
2045 tree op0 = TREE_OPERAND (rhs, 0);
2046 tree op1 = TREE_OPERAND (rhs, 1);
2048 if (rhs_code == TRUNC_DIV_EXPR)
2049 t = build (RSHIFT_EXPR, TREE_TYPE (op0), op0,
2050 build_int_2 (tree_log2 (op1), 0));
2051 else
2052 t = build (BIT_AND_EXPR, TREE_TYPE (op0), op0,
2053 local_fold (build (MINUS_EXPR, TREE_TYPE (op1),
2054 op1, integer_one_node)));
2056 result = update_rhs_and_lookup_avail_expr (stmt, t,
2057 &bd->avail_exprs,
2058 ann, insert);
2062 /* Transform ABS (X) into X or -X as appropriate. */
2063 if (rhs_code == ABS_EXPR
2064 && INTEGRAL_TYPE_P (TREE_TYPE (TREE_OPERAND (rhs, 0))))
2066 tree val;
2067 tree op = TREE_OPERAND (rhs, 0);
2068 tree type = TREE_TYPE (op);
2070 if (TYPE_UNSIGNED (type))
2072 val = integer_zero_node;
2074 else
2076 tree dummy_cond = walk_data->global_data;
2078 if (! dummy_cond)
2080 dummy_cond = build (LE_EXPR, boolean_type_node,
2081 op, integer_zero_node);
2082 dummy_cond = build (COND_EXPR, void_type_node,
2083 dummy_cond, NULL, NULL);
2084 walk_data->global_data = dummy_cond;
2086 else
2088 TREE_SET_CODE (TREE_OPERAND (dummy_cond, 0), LE_EXPR);
2089 TREE_OPERAND (TREE_OPERAND (dummy_cond, 0), 0) = op;
2090 TREE_OPERAND (TREE_OPERAND (dummy_cond, 0), 1)
2091 = fold_convert (type, integer_zero_node);
2093 val = simplify_cond_and_lookup_avail_expr (dummy_cond,
2094 &bd->avail_exprs,
2095 NULL, false);
2097 if (!val)
2099 TREE_SET_CODE (TREE_OPERAND (dummy_cond, 0), GE_EXPR);
2100 TREE_OPERAND (TREE_OPERAND (dummy_cond, 0), 0) = op;
2101 TREE_OPERAND (TREE_OPERAND (dummy_cond, 0), 1)
2102 = fold_convert (type, integer_zero_node);
2104 val = simplify_cond_and_lookup_avail_expr (dummy_cond,
2105 &bd->avail_exprs,
2106 NULL, false);
2108 if (val)
2110 if (integer_zerop (val))
2111 val = integer_one_node;
2112 else if (integer_onep (val))
2113 val = integer_zero_node;
2118 if (val
2119 && (integer_onep (val) || integer_zerop (val)))
2121 tree t;
2123 if (integer_onep (val))
2124 t = build1 (NEGATE_EXPR, TREE_TYPE (op), op);
2125 else
2126 t = op;
2128 result = update_rhs_and_lookup_avail_expr (stmt, t,
2129 &bd->avail_exprs,
2130 ann, insert);
2134 /* Optimize *"foo" into 'f'. This is done here rather than
2135 in fold to avoid problems with stuff like &*"foo". */
2136 if (TREE_CODE (rhs) == INDIRECT_REF || TREE_CODE (rhs) == ARRAY_REF)
2138 tree t = fold_read_from_constant_string (rhs);
2140 if (t)
2141 result = update_rhs_and_lookup_avail_expr (stmt, t,
2142 &bd->avail_exprs,
2143 ann, insert);
2146 return result;
2149 /* COND is a condition of the form:
2151 x == const or x != const
2153 Look back to x's defining statement and see if x is defined as
2155 x = (type) y;
2157 If const is unchanged if we convert it to type, then we can build
2158 the equivalent expression:
2161 y == const or y != const
2163 Which may allow further optimizations.
2165 Return the equivalent comparison or NULL if no such equivalent comparison
2166 was found. */
2168 static tree
2169 find_equivalent_equality_comparison (tree cond)
2171 tree op0 = TREE_OPERAND (cond, 0);
2172 tree op1 = TREE_OPERAND (cond, 1);
2173 tree def_stmt = SSA_NAME_DEF_STMT (op0);
2175 /* OP0 might have been a parameter, so first make sure it
2176 was defined by a MODIFY_EXPR. */
2177 if (def_stmt && TREE_CODE (def_stmt) == MODIFY_EXPR)
2179 tree def_rhs = TREE_OPERAND (def_stmt, 1);
2181 /* Now make sure the RHS of the MODIFY_EXPR is a typecast. */
2182 if ((TREE_CODE (def_rhs) == NOP_EXPR
2183 || TREE_CODE (def_rhs) == CONVERT_EXPR)
2184 && TREE_CODE (TREE_OPERAND (def_rhs, 0)) == SSA_NAME)
2186 tree def_rhs_inner = TREE_OPERAND (def_rhs, 0);
2187 tree def_rhs_inner_type = TREE_TYPE (def_rhs_inner);
2188 tree new;
2190 if (TYPE_PRECISION (def_rhs_inner_type)
2191 > TYPE_PRECISION (TREE_TYPE (def_rhs)))
2192 return NULL;
2194 /* What we want to prove is that if we convert OP1 to
2195 the type of the object inside the NOP_EXPR that the
2196 result is still equivalent to SRC.
2198 If that is true, the build and return new equivalent
2199 condition which uses the source of the typecast and the
2200 new constant (which has only changed its type). */
2201 new = build1 (TREE_CODE (def_rhs), def_rhs_inner_type, op1);
2202 new = local_fold (new);
2203 if (is_gimple_val (new) && tree_int_cst_equal (new, op1))
2204 return build (TREE_CODE (cond), TREE_TYPE (cond),
2205 def_rhs_inner, new);
2208 return NULL;
2211 /* STMT is a COND_EXPR for which we could not trivially determine its
2212 result. This routine attempts to find equivalent forms of the
2213 condition which we may be able to optimize better. It also
2214 uses simple value range propagation to optimize conditionals. */
2216 static tree
2217 simplify_cond_and_lookup_avail_expr (tree stmt,
2218 varray_type *block_avail_exprs_p,
2219 stmt_ann_t ann,
2220 int insert)
2222 tree cond = COND_EXPR_COND (stmt);
2224 if (TREE_CODE_CLASS (TREE_CODE (cond)) == '<')
2226 tree op0 = TREE_OPERAND (cond, 0);
2227 tree op1 = TREE_OPERAND (cond, 1);
2229 if (TREE_CODE (op0) == SSA_NAME && is_gimple_min_invariant (op1))
2231 int limit;
2232 tree low, high, cond_low, cond_high;
2233 int lowequal, highequal, swapped, no_overlap, subset, cond_inverted;
2234 varray_type vrp_records;
2235 struct vrp_element *element;
2237 /* First see if we have test of an SSA_NAME against a constant
2238 where the SSA_NAME is defined by an earlier typecast which
2239 is irrelevant when performing tests against the given
2240 constant. */
2241 if (TREE_CODE (cond) == EQ_EXPR || TREE_CODE (cond) == NE_EXPR)
2243 tree new_cond = find_equivalent_equality_comparison (cond);
2245 if (new_cond)
2247 /* Update the statement to use the new equivalent
2248 condition. */
2249 COND_EXPR_COND (stmt) = new_cond;
2250 ann->modified = 1;
2252 /* Lookup the condition and return its known value if it
2253 exists. */
2254 new_cond = lookup_avail_expr (stmt, block_avail_exprs_p,
2255 insert);
2256 if (new_cond)
2257 return new_cond;
2259 /* The operands have changed, so update op0 and op1. */
2260 op0 = TREE_OPERAND (cond, 0);
2261 op1 = TREE_OPERAND (cond, 1);
2265 /* Consult the value range records for this variable (if they exist)
2266 to see if we can eliminate or simplify this conditional.
2268 Note two tests are necessary to determine no records exist.
2269 First we have to see if the virtual array exists, if it
2270 exists, then we have to check its active size.
2272 Also note the vast majority of conditionals are not testing
2273 a variable which has had its range constrained by an earlier
2274 conditional. So this filter avoids a lot of unnecessary work. */
2275 vrp_records = VARRAY_GENERIC_PTR (vrp_data, SSA_NAME_VERSION (op0));
2276 if (vrp_records == NULL)
2277 return NULL;
2279 limit = VARRAY_ACTIVE_SIZE (vrp_records);
2281 /* If we have no value range records for this variable, or we are
2282 unable to extract a range for this condition, then there is
2283 nothing to do. */
2284 if (limit == 0
2285 || ! extract_range_from_cond (cond, &cond_high,
2286 &cond_low, &cond_inverted))
2287 return NULL;
2289 /* We really want to avoid unnecessary computations of range
2290 info. So all ranges are computed lazily; this avoids a
2291 lot of unnecessary work. ie, we record the conditional,
2292 but do not process how it constrains the variable's
2293 potential values until we know that processing the condition
2294 could be helpful.
2296 However, we do not want to have to walk a potentially long
2297 list of ranges, nor do we want to compute a variable's
2298 range more than once for a given path.
2300 Luckily, each time we encounter a conditional that can not
2301 be otherwise optimized we will end up here and we will
2302 compute the necessary range information for the variable
2303 used in this condition.
2305 Thus you can conclude that there will never be more than one
2306 conditional associated with a variable which has not been
2307 processed. So we never need to merge more than one new
2308 conditional into the current range.
2310 These properties also help us avoid unnecessary work. */
2311 element
2312 = (struct vrp_element *)VARRAY_GENERIC_PTR (vrp_records, limit - 1);
2314 if (element->high && element->low)
2316 /* The last element has been processed, so there is no range
2317 merging to do, we can simply use the high/low values
2318 recorded in the last element. */
2319 low = element->low;
2320 high = element->high;
2322 else
2324 tree tmp_high, tmp_low;
2325 int dummy;
2327 /* The last element has not been processed. Process it now. */
2328 extract_range_from_cond (element->cond, &tmp_high,
2329 &tmp_low, &dummy);
2331 /* If this is the only element, then no merging is necessary,
2332 the high/low values from extract_range_from_cond are all
2333 we need. */
2334 if (limit == 1)
2336 low = tmp_low;
2337 high = tmp_high;
2339 else
2341 /* Get the high/low value from the previous element. */
2342 struct vrp_element *prev
2343 = (struct vrp_element *)VARRAY_GENERIC_PTR (vrp_records,
2344 limit - 2);
2345 low = prev->low;
2346 high = prev->high;
2348 /* Merge in this element's range with the range from the
2349 previous element.
2351 The low value for the merged range is the maximum of
2352 the previous low value and the low value of this record.
2354 Similarly the high value for the merged range is the
2355 minimum of the previous high value and the high value of
2356 this record. */
2357 low = (tree_int_cst_compare (low, tmp_low) == 1
2358 ? low : tmp_low);
2359 high = (tree_int_cst_compare (high, tmp_high) == -1
2360 ? high : tmp_high);
2363 /* And record the computed range. */
2364 element->low = low;
2365 element->high = high;
2369 /* After we have constrained this variable's potential values,
2370 we try to determine the result of the given conditional.
2372 To simplify later tests, first determine if the current
2373 low value is the same low value as the conditional.
2374 Similarly for the current high value and the high value
2375 for the conditional. */
2376 lowequal = tree_int_cst_equal (low, cond_low);
2377 highequal = tree_int_cst_equal (high, cond_high);
2379 if (lowequal && highequal)
2380 return (cond_inverted ? boolean_false_node : boolean_true_node);
2382 /* To simplify the overlap/subset tests below we may want
2383 to swap the two ranges so that the larger of the two
2384 ranges occurs "first". */
2385 swapped = 0;
2386 if (tree_int_cst_compare (low, cond_low) == 1
2387 || (lowequal
2388 && tree_int_cst_compare (cond_high, high) == 1))
2390 tree temp;
2392 swapped = 1;
2393 temp = low;
2394 low = cond_low;
2395 cond_low = temp;
2396 temp = high;
2397 high = cond_high;
2398 cond_high = temp;
2401 /* Now determine if there is no overlap in the ranges
2402 or if the second range is a subset of the first range. */
2403 no_overlap = tree_int_cst_lt (high, cond_low);
2404 subset = tree_int_cst_compare (cond_high, high) != 1;
2406 /* If there was no overlap in the ranges, then this conditional
2407 always has a false value (unless we had to invert this
2408 conditional, in which case it always has a true value). */
2409 if (no_overlap)
2410 return (cond_inverted ? boolean_true_node : boolean_false_node);
2412 /* If the current range is a subset of the condition's range,
2413 then this conditional always has a true value (unless we
2414 had to invert this conditional, in which case it always
2415 has a true value). */
2416 if (subset && swapped)
2417 return (cond_inverted ? boolean_false_node : boolean_true_node);
2419 /* We were unable to determine the result of the conditional.
2420 However, we may be able to simplify the conditional. First
2421 merge the ranges in the same manner as range merging above. */
2422 low = tree_int_cst_compare (low, cond_low) == 1 ? low : cond_low;
2423 high = tree_int_cst_compare (high, cond_high) == -1 ? high : cond_high;
2425 /* If the range has converged to a single point, then turn this
2426 into an equality comparison. */
2427 if (TREE_CODE (cond) != EQ_EXPR
2428 && TREE_CODE (cond) != NE_EXPR
2429 && tree_int_cst_equal (low, high))
2431 TREE_SET_CODE (cond, EQ_EXPR);
2432 TREE_OPERAND (cond, 1) = high;
2436 return 0;
2439 /* STMT is a SWITCH_EXPR for which we could not trivially determine its
2440 result. This routine attempts to find equivalent forms of the
2441 condition which we may be able to optimize better. */
2443 static tree
2444 simplify_switch_and_lookup_avail_expr (tree stmt,
2445 varray_type *block_avail_exprs_p,
2446 stmt_ann_t ann,
2447 int insert)
2449 tree cond = SWITCH_COND (stmt);
2450 tree def, to, ti;
2452 /* The optimization that we really care about is removing unnecessary
2453 casts. That will let us do much better in propagating the inferred
2454 constant at the switch target. */
2455 if (TREE_CODE (cond) == SSA_NAME)
2457 def = SSA_NAME_DEF_STMT (cond);
2458 if (TREE_CODE (def) == MODIFY_EXPR)
2460 def = TREE_OPERAND (def, 1);
2461 if (TREE_CODE (def) == NOP_EXPR)
2463 int need_precision;
2464 bool fail;
2466 def = TREE_OPERAND (def, 0);
2468 #ifdef ENABLE_CHECKING
2469 /* ??? Why was Jeff testing this? We are gimple... */
2470 if (!is_gimple_val (def))
2471 abort ();
2472 #endif
2474 to = TREE_TYPE (cond);
2475 ti = TREE_TYPE (def);
2477 /* If we have an extension that preserves value, then we
2478 can copy the source value into the switch. */
2480 need_precision = TYPE_PRECISION (ti);
2481 fail = false;
2482 if (TYPE_UNSIGNED (to) && !TYPE_UNSIGNED (ti))
2483 fail = true;
2484 else if (!TYPE_UNSIGNED (to) && TYPE_UNSIGNED (ti))
2485 need_precision += 1;
2486 if (TYPE_PRECISION (to) < need_precision)
2487 fail = true;
2489 if (!fail)
2491 SWITCH_COND (stmt) = def;
2492 ann->modified = 1;
2494 return lookup_avail_expr (stmt, block_avail_exprs_p, insert);
2500 return 0;
2504 /* CONST_AND_COPIES is a table which maps an SSA_NAME to the current
2505 known value for that SSA_NAME (or NULL if no value is known).
2507 NONZERO_VARS is the set SSA_NAMES known to have a nonzero value,
2508 even if we don't know their precise value.
2510 Propagate values from CONST_AND_COPIES and NONZERO_VARS into the PHI
2511 nodes of the successors of BB. */
2513 static void
2514 cprop_into_successor_phis (basic_block bb,
2515 varray_type const_and_copies,
2516 bitmap nonzero_vars)
2518 edge e;
2520 /* This can get rather expensive if the implementation is naive in
2521 how it finds the phi alternative associated with a particular edge. */
2522 for (e = bb->succ; e; e = e->succ_next)
2524 tree phi;
2525 int phi_num_args;
2526 int hint;
2528 /* If this is an abnormal edge, then we do not want to copy propagate
2529 into the PHI alternative associated with this edge. */
2530 if (e->flags & EDGE_ABNORMAL)
2531 continue;
2533 phi = phi_nodes (e->dest);
2534 if (! phi)
2535 continue;
2537 /* There is no guarantee that for any two PHI nodes in a block that
2538 the phi alternative associated with a particular edge will be
2539 at the same index in the phi alternative array.
2541 However, it is very likely they will be the same. So we keep
2542 track of the index of the alternative where we found the edge in
2543 the previous phi node and check that index first in the next
2544 phi node. If that hint fails, then we actually search all
2545 the entries. */
2546 phi_num_args = PHI_NUM_ARGS (phi);
2547 hint = phi_num_args;
2548 for ( ; phi; phi = PHI_CHAIN (phi))
2550 int i;
2551 tree new;
2552 use_operand_p orig_p;
2553 tree orig;
2555 /* If the hint is valid (!= phi_num_args), see if it points
2556 us to the desired phi alternative. */
2557 if (hint != phi_num_args && PHI_ARG_EDGE (phi, hint) == e)
2559 else
2561 /* The hint was either invalid or did not point to the
2562 correct phi alternative. Search all the alternatives
2563 for the correct one. Update the hint. */
2564 for (i = 0; i < phi_num_args; i++)
2565 if (PHI_ARG_EDGE (phi, i) == e)
2566 break;
2567 hint = i;
2570 #ifdef ENABLE_CHECKING
2571 /* If we did not find the proper alternative, then something is
2572 horribly wrong. */
2573 if (hint == phi_num_args)
2574 abort ();
2575 #endif
2577 /* The alternative may be associated with a constant, so verify
2578 it is an SSA_NAME before doing anything with it. */
2579 orig_p = PHI_ARG_DEF_PTR (phi, hint);
2580 orig = USE_FROM_PTR (orig_p);
2581 if (TREE_CODE (orig) != SSA_NAME)
2582 continue;
2584 /* If the alternative is known to have a nonzero value, record
2585 that fact in the PHI node itself for future use. */
2586 if (bitmap_bit_p (nonzero_vars, SSA_NAME_VERSION (orig)))
2587 PHI_ARG_NONZERO (phi, hint) = true;
2589 /* If we have *ORIG_P in our constant/copy table, then replace
2590 ORIG_P with its value in our constant/copy table. */
2591 new = VARRAY_TREE (const_and_copies, SSA_NAME_VERSION (orig));
2592 if (new
2593 && (TREE_CODE (new) == SSA_NAME
2594 || is_gimple_min_invariant (new))
2595 && may_propagate_copy (orig, new))
2597 propagate_value (orig_p, new);
2604 /* Propagate known constants/copies into PHI nodes of BB's successor
2605 blocks. */
2607 static void
2608 cprop_into_phis (struct dom_walk_data *walk_data ATTRIBUTE_UNUSED,
2609 basic_block bb)
2611 cprop_into_successor_phis (bb, const_and_copies, nonzero_vars);
2614 /* Search for redundant computations in STMT. If any are found, then
2615 replace them with the variable holding the result of the computation.
2617 If safe, record this expression into the available expression hash
2618 table. */
2620 static bool
2621 eliminate_redundant_computations (struct dom_walk_data *walk_data,
2622 tree stmt, stmt_ann_t ann)
2624 v_may_def_optype v_may_defs = V_MAY_DEF_OPS (ann);
2625 tree *expr_p, def = NULL_TREE;
2626 bool insert = true;
2627 tree cached_lhs;
2628 bool retval = false;
2629 struct dom_walk_block_data *bd
2630 = VARRAY_TOP_GENERIC_PTR (walk_data->block_data_stack);
2632 if (TREE_CODE (stmt) == MODIFY_EXPR)
2633 def = TREE_OPERAND (stmt, 0);
2635 /* Certain expressions on the RHS can be optimized away, but can not
2636 themselves be entered into the hash tables. */
2637 if (ann->makes_aliased_stores
2638 || ! def
2639 || TREE_CODE (def) != SSA_NAME
2640 || SSA_NAME_OCCURS_IN_ABNORMAL_PHI (def)
2641 || NUM_V_MAY_DEFS (v_may_defs) != 0)
2642 insert = false;
2644 /* Check if the expression has been computed before. */
2645 cached_lhs = lookup_avail_expr (stmt, &bd->avail_exprs, insert);
2647 /* If this is an assignment and the RHS was not in the hash table,
2648 then try to simplify the RHS and lookup the new RHS in the
2649 hash table. */
2650 if (! cached_lhs && TREE_CODE (stmt) == MODIFY_EXPR)
2651 cached_lhs = simplify_rhs_and_lookup_avail_expr (walk_data,
2652 stmt,
2653 ann,
2654 insert);
2655 /* Similarly if this is a COND_EXPR and we did not find its
2656 expression in the hash table, simplify the condition and
2657 try again. */
2658 else if (! cached_lhs && TREE_CODE (stmt) == COND_EXPR)
2659 cached_lhs = simplify_cond_and_lookup_avail_expr (stmt,
2660 &bd->avail_exprs,
2661 ann,
2662 insert);
2663 /* Similarly for a SWITCH_EXPR. */
2664 else if (!cached_lhs && TREE_CODE (stmt) == SWITCH_EXPR)
2665 cached_lhs = simplify_switch_and_lookup_avail_expr (stmt,
2666 &bd->avail_exprs,
2667 ann,
2668 insert);
2670 opt_stats.num_exprs_considered++;
2672 /* Get a pointer to the expression we are trying to optimize. */
2673 if (TREE_CODE (stmt) == COND_EXPR)
2674 expr_p = &COND_EXPR_COND (stmt);
2675 else if (TREE_CODE (stmt) == SWITCH_EXPR)
2676 expr_p = &SWITCH_COND (stmt);
2677 else if (TREE_CODE (stmt) == RETURN_EXPR && TREE_OPERAND (stmt, 0))
2678 expr_p = &TREE_OPERAND (TREE_OPERAND (stmt, 0), 1);
2679 else
2680 expr_p = &TREE_OPERAND (stmt, 1);
2682 /* It is safe to ignore types here since we have already done
2683 type checking in the hashing and equality routines. In fact
2684 type checking here merely gets in the way of constant
2685 propagation. Also, make sure that it is safe to propagate
2686 CACHED_LHS into *EXPR_P. */
2687 if (cached_lhs
2688 && (TREE_CODE (cached_lhs) != SSA_NAME
2689 || may_propagate_copy (*expr_p, cached_lhs)))
2691 if (dump_file && (dump_flags & TDF_DETAILS))
2693 fprintf (dump_file, " Replaced redundant expr '");
2694 print_generic_expr (dump_file, *expr_p, dump_flags);
2695 fprintf (dump_file, "' with '");
2696 print_generic_expr (dump_file, cached_lhs, dump_flags);
2697 fprintf (dump_file, "'\n");
2700 opt_stats.num_re++;
2702 #if defined ENABLE_CHECKING
2703 if (TREE_CODE (cached_lhs) != SSA_NAME
2704 && !is_gimple_min_invariant (cached_lhs))
2705 abort ();
2706 #endif
2708 if (TREE_CODE (cached_lhs) == ADDR_EXPR
2709 || (POINTER_TYPE_P (TREE_TYPE (*expr_p))
2710 && is_gimple_min_invariant (cached_lhs)))
2711 retval = true;
2713 propagate_tree_value (expr_p, cached_lhs);
2714 ann->modified = 1;
2716 return retval;
2719 /* STMT, a MODIFY_EXPR, may create certain equivalences, in either
2720 the available expressions table or the const_and_copies table.
2721 Detect and record those equivalences. */
2723 static void
2724 record_equivalences_from_stmt (tree stmt,
2725 varray_type *block_avail_exprs_p,
2726 varray_type *block_nonzero_vars_p,
2727 int may_optimize_p,
2728 stmt_ann_t ann)
2730 tree lhs = TREE_OPERAND (stmt, 0);
2731 enum tree_code lhs_code = TREE_CODE (lhs);
2732 int i;
2734 if (lhs_code == SSA_NAME)
2736 tree rhs = TREE_OPERAND (stmt, 1);
2738 /* Strip away any useless type conversions. */
2739 STRIP_USELESS_TYPE_CONVERSION (rhs);
2741 /* If the RHS of the assignment is a constant or another variable that
2742 may be propagated, register it in the CONST_AND_COPIES table. We
2743 do not need to record unwind data for this, since this is a true
2744 assignment and not an equivalence inferred from a comparison. All
2745 uses of this ssa name are dominated by this assignment, so unwinding
2746 just costs time and space. */
2747 if (may_optimize_p
2748 && (TREE_CODE (rhs) == SSA_NAME
2749 || is_gimple_min_invariant (rhs)))
2750 set_value_for (lhs, rhs, const_and_copies);
2752 /* alloca never returns zero and the address of a non-weak symbol
2753 is never zero. NOP_EXPRs and CONVERT_EXPRs can be completely
2754 stripped as they do not affect this equivalence. */
2755 while (TREE_CODE (rhs) == NOP_EXPR
2756 || TREE_CODE (rhs) == CONVERT_EXPR)
2757 rhs = TREE_OPERAND (rhs, 0);
2759 if (alloca_call_p (rhs)
2760 || (TREE_CODE (rhs) == ADDR_EXPR
2761 && DECL_P (TREE_OPERAND (rhs, 0))
2762 && ! DECL_WEAK (TREE_OPERAND (rhs, 0))))
2763 record_var_is_nonzero (lhs, block_nonzero_vars_p);
2765 /* IOR of any value with a nonzero value will result in a nonzero
2766 value. Even if we do not know the exact result recording that
2767 the result is nonzero is worth the effort. */
2768 if (TREE_CODE (rhs) == BIT_IOR_EXPR
2769 && integer_nonzerop (TREE_OPERAND (rhs, 1)))
2770 record_var_is_nonzero (lhs, block_nonzero_vars_p);
2773 /* Look at both sides for pointer dereferences. If we find one, then
2774 the pointer must be nonnull and we can enter that equivalence into
2775 the hash tables. */
2776 if (flag_delete_null_pointer_checks)
2777 for (i = 0; i < 2; i++)
2779 tree t = TREE_OPERAND (stmt, i);
2781 /* Strip away any COMPONENT_REFs. */
2782 while (TREE_CODE (t) == COMPONENT_REF)
2783 t = TREE_OPERAND (t, 0);
2785 /* Now see if this is a pointer dereference. */
2786 if (TREE_CODE (t) == INDIRECT_REF)
2788 tree op = TREE_OPERAND (t, 0);
2790 /* If the pointer is a SSA variable, then enter new
2791 equivalences into the hash table. */
2792 while (TREE_CODE (op) == SSA_NAME)
2794 tree def = SSA_NAME_DEF_STMT (op);
2796 record_var_is_nonzero (op, block_nonzero_vars_p);
2798 /* And walk up the USE-DEF chains noting other SSA_NAMEs
2799 which are known to have a nonzero value. */
2800 if (def
2801 && TREE_CODE (def) == MODIFY_EXPR
2802 && TREE_CODE (TREE_OPERAND (def, 1)) == NOP_EXPR)
2803 op = TREE_OPERAND (TREE_OPERAND (def, 1), 0);
2804 else
2805 break;
2810 /* A memory store, even an aliased store, creates a useful
2811 equivalence. By exchanging the LHS and RHS, creating suitable
2812 vops and recording the result in the available expression table,
2813 we may be able to expose more redundant loads. */
2814 if (!ann->has_volatile_ops
2815 && (TREE_CODE (TREE_OPERAND (stmt, 1)) == SSA_NAME
2816 || is_gimple_min_invariant (TREE_OPERAND (stmt, 1)))
2817 && !is_gimple_reg (lhs))
2819 tree rhs = TREE_OPERAND (stmt, 1);
2820 tree new;
2821 size_t j;
2823 /* FIXME: If the LHS of the assignment is a bitfield and the RHS
2824 is a constant, we need to adjust the constant to fit into the
2825 type of the LHS. If the LHS is a bitfield and the RHS is not
2826 a constant, then we can not record any equivalences for this
2827 statement since we would need to represent the widening or
2828 narrowing of RHS. This fixes gcc.c-torture/execute/921016-1.c
2829 and should not be necessary if GCC represented bitfields
2830 properly. */
2831 if (lhs_code == COMPONENT_REF
2832 && DECL_BIT_FIELD (TREE_OPERAND (lhs, 1)))
2834 if (TREE_CONSTANT (rhs))
2835 rhs = widen_bitfield (rhs, TREE_OPERAND (lhs, 1), lhs);
2836 else
2837 rhs = NULL;
2839 /* If the value overflowed, then we can not use this equivalence. */
2840 if (rhs && ! is_gimple_min_invariant (rhs))
2841 rhs = NULL;
2844 if (rhs)
2846 v_may_def_optype v_may_defs = V_MAY_DEF_OPS (ann);
2847 v_must_def_optype v_must_defs = V_MUST_DEF_OPS (ann);
2849 /* Build a new statement with the RHS and LHS exchanged. */
2850 new = build (MODIFY_EXPR, TREE_TYPE (stmt), rhs, lhs);
2852 /* Get an annotation and set up the real operands. */
2853 get_stmt_ann (new);
2854 get_stmt_operands (new);
2856 /* Clear out the virtual operands on the new statement, we are
2857 going to set them explicitly below. */
2858 remove_vuses (new);
2859 remove_v_may_defs (new);
2860 remove_v_must_defs (new);
2862 start_ssa_stmt_operands (new);
2863 /* For each VDEF on the original statement, we want to create a
2864 VUSE of the V_MAY_DEF result or V_MUST_DEF op on the new
2865 statement. */
2866 for (j = 0; j < NUM_V_MAY_DEFS (v_may_defs); j++)
2868 tree op = V_MAY_DEF_RESULT (v_may_defs, j);
2869 add_vuse (op, new);
2872 for (j = 0; j < NUM_V_MUST_DEFS (v_must_defs); j++)
2874 tree op = V_MUST_DEF_OP (v_must_defs, j);
2875 add_vuse (op, new);
2878 finalize_ssa_stmt_operands (new);
2880 /* Finally enter the statement into the available expression
2881 table. */
2882 lookup_avail_expr (new, block_avail_exprs_p, true);
2887 /* Replace *OP_P in STMT with any known equivalent value for *OP_P from
2888 CONST_AND_COPIES. */
2890 static bool
2891 cprop_operand (stmt_ann_t ann, use_operand_p op_p, varray_type const_and_copies)
2893 bool may_have_exposed_new_symbols = false;
2894 tree val;
2895 tree op = USE_FROM_PTR (op_p);
2897 /* If the operand has a known constant value or it is known to be a
2898 copy of some other variable, use the value or copy stored in
2899 CONST_AND_COPIES. */
2900 val = VARRAY_TREE (const_and_copies, SSA_NAME_VERSION (op));
2901 if (val)
2903 tree op_type, val_type;
2905 /* Do not change the base variable in the virtual operand
2906 tables. That would make it impossible to reconstruct
2907 the renamed virtual operand if we later modify this
2908 statement. Also only allow the new value to be an SSA_NAME
2909 for propagation into virtual operands. */
2910 if (!is_gimple_reg (op)
2911 && (get_virtual_var (val) != get_virtual_var (op)
2912 || TREE_CODE (val) != SSA_NAME))
2913 return false;
2915 /* Get the toplevel type of each operand. */
2916 op_type = TREE_TYPE (op);
2917 val_type = TREE_TYPE (val);
2919 /* While both types are pointers, get the type of the object
2920 pointed to. */
2921 while (POINTER_TYPE_P (op_type) && POINTER_TYPE_P (val_type))
2923 op_type = TREE_TYPE (op_type);
2924 val_type = TREE_TYPE (val_type);
2927 /* Make sure underlying types match before propagating a constant by
2928 converting the constant to the proper type. Note that convert may
2929 return a non-gimple expression, in which case we ignore this
2930 propagation opportunity. */
2931 if (TREE_CODE (val) != SSA_NAME)
2933 if (!lang_hooks.types_compatible_p (op_type, val_type))
2935 val = fold_convert (TREE_TYPE (op), val);
2936 if (!is_gimple_min_invariant (val))
2937 return false;
2941 /* Certain operands are not allowed to be copy propagated due
2942 to their interaction with exception handling and some GCC
2943 extensions. */
2944 else if (!may_propagate_copy (op, val))
2945 return false;
2947 /* Dump details. */
2948 if (dump_file && (dump_flags & TDF_DETAILS))
2950 fprintf (dump_file, " Replaced '");
2951 print_generic_expr (dump_file, op, dump_flags);
2952 fprintf (dump_file, "' with %s '",
2953 (TREE_CODE (val) != SSA_NAME ? "constant" : "variable"));
2954 print_generic_expr (dump_file, val, dump_flags);
2955 fprintf (dump_file, "'\n");
2958 /* If VAL is an ADDR_EXPR or a constant of pointer type, note
2959 that we may have exposed a new symbol for SSA renaming. */
2960 if (TREE_CODE (val) == ADDR_EXPR
2961 || (POINTER_TYPE_P (TREE_TYPE (op))
2962 && is_gimple_min_invariant (val)))
2963 may_have_exposed_new_symbols = true;
2965 propagate_value (op_p, val);
2967 /* And note that we modified this statement. This is now
2968 safe, even if we changed virtual operands since we will
2969 rescan the statement and rewrite its operands again. */
2970 ann->modified = 1;
2972 return may_have_exposed_new_symbols;
2975 /* CONST_AND_COPIES is a table which maps an SSA_NAME to the current
2976 known value for that SSA_NAME (or NULL if no value is known).
2978 Propagate values from CONST_AND_COPIES into the uses, vuses and
2979 v_may_def_ops of STMT. */
2981 static bool
2982 cprop_into_stmt (tree stmt, varray_type const_and_copies)
2984 bool may_have_exposed_new_symbols = false;
2985 stmt_ann_t ann = stmt_ann (stmt);
2986 size_t i, num_uses, num_vuses, num_v_may_defs;
2987 vuse_optype vuses;
2988 v_may_def_optype v_may_defs;
2989 use_optype uses;
2991 uses = USE_OPS (ann);
2992 num_uses = NUM_USES (uses);
2993 for (i = 0; i < num_uses; i++)
2995 use_operand_p op_p = USE_OP_PTR (uses, i);
2996 if (TREE_CODE (USE_FROM_PTR (op_p)) == SSA_NAME)
2997 may_have_exposed_new_symbols
2998 |= cprop_operand (ann, op_p, const_and_copies);
3001 vuses = VUSE_OPS (ann);
3002 num_vuses = NUM_VUSES (vuses);
3003 for (i = 0; i < num_vuses; i++)
3005 use_operand_p op_p = VUSE_OP_PTR (vuses, i);
3006 if (TREE_CODE (USE_FROM_PTR (op_p)) == SSA_NAME)
3007 may_have_exposed_new_symbols
3008 |= cprop_operand (ann, op_p, const_and_copies);
3011 v_may_defs = V_MAY_DEF_OPS (ann);
3012 num_v_may_defs = NUM_V_MAY_DEFS (v_may_defs);
3013 for (i = 0; i < num_v_may_defs; i++)
3015 use_operand_p op_p = V_MAY_DEF_OP_PTR (v_may_defs, i);
3016 if (TREE_CODE (USE_FROM_PTR (op_p)) == SSA_NAME)
3017 may_have_exposed_new_symbols
3018 |= cprop_operand (ann, op_p, const_and_copies);
3020 return may_have_exposed_new_symbols;
3024 /* Optimize the statement pointed by iterator SI.
3026 We try to perform some simplistic global redundancy elimination and
3027 constant propagation:
3029 1- To detect global redundancy, we keep track of expressions that have
3030 been computed in this block and its dominators. If we find that the
3031 same expression is computed more than once, we eliminate repeated
3032 computations by using the target of the first one.
3034 2- Constant values and copy assignments. This is used to do very
3035 simplistic constant and copy propagation. When a constant or copy
3036 assignment is found, we map the value on the RHS of the assignment to
3037 the variable in the LHS in the CONST_AND_COPIES table. */
3039 static void
3040 optimize_stmt (struct dom_walk_data *walk_data, basic_block bb,
3041 block_stmt_iterator si)
3043 stmt_ann_t ann;
3044 tree stmt;
3045 bool may_optimize_p;
3046 bool may_have_exposed_new_symbols = false;
3047 struct dom_walk_block_data *bd
3048 = VARRAY_TOP_GENERIC_PTR (walk_data->block_data_stack);
3050 stmt = bsi_stmt (si);
3052 get_stmt_operands (stmt);
3053 ann = stmt_ann (stmt);
3054 opt_stats.num_stmts++;
3055 may_have_exposed_new_symbols = false;
3057 if (dump_file && (dump_flags & TDF_DETAILS))
3059 fprintf (dump_file, "Optimizing statement ");
3060 print_generic_stmt (dump_file, stmt, TDF_SLIM);
3063 /* Const/copy propagate into USES, VUSES and the RHS of V_MAY_DEFs. */
3064 may_have_exposed_new_symbols = cprop_into_stmt (stmt, const_and_copies);
3066 /* If the statement has been modified with constant replacements,
3067 fold its RHS before checking for redundant computations. */
3068 if (ann->modified)
3070 /* Try to fold the statement making sure that STMT is kept
3071 up to date. */
3072 if (fold_stmt (bsi_stmt_ptr (si)))
3074 stmt = bsi_stmt (si);
3075 ann = stmt_ann (stmt);
3077 if (dump_file && (dump_flags & TDF_DETAILS))
3079 fprintf (dump_file, " Folded to: ");
3080 print_generic_stmt (dump_file, stmt, TDF_SLIM);
3084 /* Constant/copy propagation above may change the set of
3085 virtual operands associated with this statement. Folding
3086 may remove the need for some virtual operands.
3088 Indicate we will need to rescan and rewrite the statement. */
3089 may_have_exposed_new_symbols = true;
3092 /* Check for redundant computations. Do this optimization only
3093 for assignments that have no volatile ops and conditionals. */
3094 may_optimize_p = (!ann->has_volatile_ops
3095 && ((TREE_CODE (stmt) == RETURN_EXPR
3096 && TREE_OPERAND (stmt, 0)
3097 && TREE_CODE (TREE_OPERAND (stmt, 0)) == MODIFY_EXPR
3098 && ! (TREE_SIDE_EFFECTS
3099 (TREE_OPERAND (TREE_OPERAND (stmt, 0), 1))))
3100 || (TREE_CODE (stmt) == MODIFY_EXPR
3101 && ! TREE_SIDE_EFFECTS (TREE_OPERAND (stmt, 1)))
3102 || TREE_CODE (stmt) == COND_EXPR
3103 || TREE_CODE (stmt) == SWITCH_EXPR));
3105 if (may_optimize_p)
3106 may_have_exposed_new_symbols
3107 |= eliminate_redundant_computations (walk_data, stmt, ann);
3109 /* Record any additional equivalences created by this statement. */
3110 if (TREE_CODE (stmt) == MODIFY_EXPR)
3111 record_equivalences_from_stmt (stmt,
3112 &bd->avail_exprs,
3113 &bd->nonzero_vars,
3114 may_optimize_p,
3115 ann);
3117 register_definitions_for_stmt (ann, &bd->block_defs);
3119 /* If STMT is a COND_EXPR and it was modified, then we may know
3120 where it goes. If that is the case, then mark the CFG as altered.
3122 This will cause us to later call remove_unreachable_blocks and
3123 cleanup_tree_cfg when it is safe to do so. It is not safe to
3124 clean things up here since removal of edges and such can trigger
3125 the removal of PHI nodes, which in turn can release SSA_NAMEs to
3126 the manager.
3128 That's all fine and good, except that once SSA_NAMEs are released
3129 to the manager, we must not call create_ssa_name until all references
3130 to released SSA_NAMEs have been eliminated.
3132 All references to the deleted SSA_NAMEs can not be eliminated until
3133 we remove unreachable blocks.
3135 We can not remove unreachable blocks until after we have completed
3136 any queued jump threading.
3138 We can not complete any queued jump threads until we have taken
3139 appropriate variables out of SSA form. Taking variables out of
3140 SSA form can call create_ssa_name and thus we lose.
3142 Ultimately I suspect we're going to need to change the interface
3143 into the SSA_NAME manager. */
3145 if (ann->modified)
3147 tree val = NULL;
3149 if (TREE_CODE (stmt) == COND_EXPR)
3150 val = COND_EXPR_COND (stmt);
3151 else if (TREE_CODE (stmt) == SWITCH_EXPR)
3152 val = SWITCH_COND (stmt);
3154 if (val && TREE_CODE (val) == INTEGER_CST && find_taken_edge (bb, val))
3155 cfg_altered = true;
3157 /* If we simplified a statement in such a way as to be shown that it
3158 cannot trap, update the eh information and the cfg to match. */
3159 if (maybe_clean_eh_stmt (stmt))
3161 bitmap_set_bit (need_eh_cleanup, bb->index);
3162 if (dump_file && (dump_flags & TDF_DETAILS))
3163 fprintf (dump_file, " Flagged to clear EH edges.\n");
3167 if (may_have_exposed_new_symbols)
3169 if (! bd->stmts_to_rescan)
3170 VARRAY_TREE_INIT (bd->stmts_to_rescan, 20, "stmts_to_rescan");
3171 VARRAY_PUSH_TREE (bd->stmts_to_rescan, bsi_stmt (si));
3175 /* Replace the RHS of STMT with NEW_RHS. If RHS can be found in the
3176 available expression hashtable, then return the LHS from the hash
3177 table.
3179 If INSERT is true, then we also update the available expression
3180 hash table to account for the changes made to STMT. */
3182 static tree
3183 update_rhs_and_lookup_avail_expr (tree stmt, tree new_rhs,
3184 varray_type *block_avail_exprs_p,
3185 stmt_ann_t ann,
3186 bool insert)
3188 tree cached_lhs = NULL;
3190 /* Remove the old entry from the hash table. */
3191 if (insert)
3193 struct expr_hash_elt element;
3195 initialize_hash_element (stmt, NULL, &element);
3196 htab_remove_elt_with_hash (avail_exprs, &element, element.hash);
3199 /* Now update the RHS of the assignment. */
3200 TREE_OPERAND (stmt, 1) = new_rhs;
3202 /* Now lookup the updated statement in the hash table. */
3203 cached_lhs = lookup_avail_expr (stmt, block_avail_exprs_p, insert);
3205 /* We have now called lookup_avail_expr twice with two different
3206 versions of this same statement, once in optimize_stmt, once here.
3208 We know the call in optimize_stmt did not find an existing entry
3209 in the hash table, so a new entry was created. At the same time
3210 this statement was pushed onto the BLOCK_AVAIL_EXPRS varray.
3212 If this call failed to find an existing entry on the hash table,
3213 then the new version of this statement was entered into the
3214 hash table. And this statement was pushed onto BLOCK_AVAIL_EXPR
3215 for the second time. So there are two copies on BLOCK_AVAIL_EXPRs
3217 If this call succeeded, we still have one copy of this statement
3218 on the BLOCK_AVAIL_EXPRs varray.
3220 For both cases, we need to pop the most recent entry off the
3221 BLOCK_AVAIL_EXPRs varray. For the case where we never found this
3222 statement in the hash tables, that will leave precisely one
3223 copy of this statement on BLOCK_AVAIL_EXPRs. For the case where
3224 we found a copy of this statement in the second hash table lookup
3225 we want _no_ copies of this statement in BLOCK_AVAIL_EXPRs. */
3226 if (insert)
3227 VARRAY_POP (*block_avail_exprs_p);
3229 /* And make sure we record the fact that we modified this
3230 statement. */
3231 ann->modified = 1;
3233 return cached_lhs;
3236 /* Search for an existing instance of STMT in the AVAIL_EXPRS table. If
3237 found, return its LHS. Otherwise insert STMT in the table and return
3238 NULL_TREE.
3240 Also, when an expression is first inserted in the AVAIL_EXPRS table, it
3241 is also added to the stack pointed by BLOCK_AVAIL_EXPRS_P, so that they
3242 can be removed when we finish processing this block and its children.
3244 NOTE: This function assumes that STMT is a MODIFY_EXPR node that
3245 contains no CALL_EXPR on its RHS and makes no volatile nor
3246 aliased references. */
3248 static tree
3249 lookup_avail_expr (tree stmt, varray_type *block_avail_exprs_p, bool insert)
3251 void **slot;
3252 tree lhs;
3253 tree temp;
3254 struct expr_hash_elt *element = xcalloc (sizeof (struct expr_hash_elt), 1);
3256 lhs = TREE_CODE (stmt) == MODIFY_EXPR ? TREE_OPERAND (stmt, 0) : NULL;
3258 initialize_hash_element (stmt, lhs, element);
3260 /* Don't bother remembering constant assignments and copy operations.
3261 Constants and copy operations are handled by the constant/copy propagator
3262 in optimize_stmt. */
3263 if (TREE_CODE (element->rhs) == SSA_NAME
3264 || is_gimple_min_invariant (element->rhs))
3266 free (element);
3267 return NULL_TREE;
3270 /* If this is an equality test against zero, see if we have recorded a
3271 nonzero value for the variable in question. */
3272 if ((TREE_CODE (element->rhs) == EQ_EXPR
3273 || TREE_CODE (element->rhs) == NE_EXPR)
3274 && TREE_CODE (TREE_OPERAND (element->rhs, 0)) == SSA_NAME
3275 && integer_zerop (TREE_OPERAND (element->rhs, 1)))
3277 int indx = SSA_NAME_VERSION (TREE_OPERAND (element->rhs, 0));
3279 if (bitmap_bit_p (nonzero_vars, indx))
3281 tree t = element->rhs;
3282 free (element);
3284 if (TREE_CODE (t) == EQ_EXPR)
3285 return boolean_false_node;
3286 else
3287 return boolean_true_node;
3291 /* Finally try to find the expression in the main expression hash table. */
3292 slot = htab_find_slot_with_hash (avail_exprs, element, element->hash,
3293 (insert ? INSERT : NO_INSERT));
3294 if (slot == NULL)
3296 free (element);
3297 return NULL_TREE;
3300 if (*slot == NULL)
3302 *slot = (void *) element;
3303 if (! *block_avail_exprs_p)
3304 VARRAY_TREE_INIT (*block_avail_exprs_p, 20, "block_avail_exprs");
3305 VARRAY_PUSH_TREE (*block_avail_exprs_p, stmt ? stmt : element->rhs);
3306 return NULL_TREE;
3309 /* Extract the LHS of the assignment so that it can be used as the current
3310 definition of another variable. */
3311 lhs = ((struct expr_hash_elt *)*slot)->lhs;
3313 /* See if the LHS appears in the CONST_AND_COPIES table. If it does, then
3314 use the value from the const_and_copies table. */
3315 if (TREE_CODE (lhs) == SSA_NAME)
3317 temp = get_value_for (lhs, const_and_copies);
3318 if (temp)
3319 lhs = temp;
3322 free (element);
3323 return lhs;
3326 /* Given a condition COND, record into HI_P, LO_P and INVERTED_P the
3327 range of values that result in the conditional having a true value.
3329 Return true if we are successful in extracting a range from COND and
3330 false if we are unsuccessful. */
3332 static bool
3333 extract_range_from_cond (tree cond, tree *hi_p, tree *lo_p, int *inverted_p)
3335 tree op1 = TREE_OPERAND (cond, 1);
3336 tree high, low, type;
3337 int inverted;
3339 /* Experiments have shown that it's rarely, if ever useful to
3340 record ranges for enumerations. Presumably this is due to
3341 the fact that they're rarely used directly. They are typically
3342 cast into an integer type and used that way. */
3343 if (TREE_CODE (TREE_TYPE (op1)) != INTEGER_TYPE)
3344 return 0;
3346 type = TREE_TYPE (op1);
3348 switch (TREE_CODE (cond))
3350 case EQ_EXPR:
3351 high = low = op1;
3352 inverted = 0;
3353 break;
3355 case NE_EXPR:
3356 high = low = op1;
3357 inverted = 1;
3358 break;
3360 case GE_EXPR:
3361 low = op1;
3362 high = TYPE_MAX_VALUE (type);
3363 inverted = 0;
3364 break;
3366 case GT_EXPR:
3367 low = int_const_binop (PLUS_EXPR, op1, integer_one_node, 1);
3368 high = TYPE_MAX_VALUE (type);
3369 inverted = 0;
3370 break;
3372 case LE_EXPR:
3373 high = op1;
3374 low = TYPE_MIN_VALUE (type);
3375 inverted = 0;
3376 break;
3378 case LT_EXPR:
3379 high = int_const_binop (MINUS_EXPR, op1, integer_one_node, 1);
3380 low = TYPE_MIN_VALUE (type);
3381 inverted = 0;
3382 break;
3384 default:
3385 return 0;
3388 *hi_p = high;
3389 *lo_p = low;
3390 *inverted_p = inverted;
3391 return 1;
3394 /* Record a range created by COND for basic block BB. */
3396 static void
3397 record_range (tree cond, basic_block bb, varray_type *vrp_variables_p)
3399 /* We explicitly ignore NE_EXPRs. They rarely allow for meaningful
3400 range optimizations and significantly complicate the implementation. */
3401 if (TREE_CODE_CLASS (TREE_CODE (cond)) == '<'
3402 && TREE_CODE (cond) != NE_EXPR
3403 && TREE_CODE (TREE_TYPE (TREE_OPERAND (cond, 1))) == INTEGER_TYPE)
3405 struct vrp_element *element = ggc_alloc (sizeof (struct vrp_element));
3406 int ssa_version = SSA_NAME_VERSION (TREE_OPERAND (cond, 0));
3408 varray_type *vrp_records_p
3409 = (varray_type *)&VARRAY_GENERIC_PTR (vrp_data, ssa_version);
3411 element->low = NULL;
3412 element->high = NULL;
3413 element->cond = cond;
3414 element->bb = bb;
3416 if (*vrp_records_p == NULL)
3418 VARRAY_GENERIC_PTR_INIT (*vrp_records_p, 2, "vrp records");
3419 VARRAY_GENERIC_PTR (vrp_data, ssa_version) = *vrp_records_p;
3422 VARRAY_PUSH_GENERIC_PTR (*vrp_records_p, element);
3423 if (! *vrp_variables_p)
3424 VARRAY_TREE_INIT (*vrp_variables_p, 2, "vrp_variables");
3425 VARRAY_PUSH_TREE (*vrp_variables_p, TREE_OPERAND (cond, 0));
3429 /* Given a conditional statement IF_STMT, return the assignment 'X = Y'
3430 known to be true depending on which arm of IF_STMT is taken.
3432 Not all conditional statements will result in a useful assignment.
3433 Return NULL_TREE in that case.
3435 Also enter into the available expression table statements of
3436 the form:
3438 TRUE ARM FALSE ARM
3439 1 = cond 1 = cond'
3440 0 = cond' 0 = cond
3442 This allows us to lookup the condition in a dominated block and
3443 get back a constant indicating if the condition is true. */
3445 static struct eq_expr_value
3446 get_eq_expr_value (tree if_stmt,
3447 int true_arm,
3448 varray_type *block_avail_exprs_p,
3449 basic_block bb,
3450 varray_type *vrp_variables_p)
3452 tree cond;
3453 struct eq_expr_value retval;
3455 cond = COND_EXPR_COND (if_stmt);
3456 retval.src = NULL;
3457 retval.dst = NULL;
3459 /* If the conditional is a single variable 'X', return 'X = 1' for
3460 the true arm and 'X = 0' on the false arm. */
3461 if (TREE_CODE (cond) == SSA_NAME)
3463 retval.dst = cond;
3464 retval.src = (true_arm ? integer_one_node : integer_zero_node);
3465 return retval;
3468 /* If we have a comparison expression, then record its result into
3469 the available expression table. */
3470 if (TREE_CODE_CLASS (TREE_CODE (cond)) == '<')
3472 tree op0 = TREE_OPERAND (cond, 0);
3473 tree op1 = TREE_OPERAND (cond, 1);
3475 /* Special case comparing booleans against a constant as we know
3476 the value of OP0 on both arms of the branch. ie, we can record
3477 an equivalence for OP0 rather than COND. */
3478 if ((TREE_CODE (cond) == EQ_EXPR || TREE_CODE (cond) == NE_EXPR)
3479 && TREE_CODE (op0) == SSA_NAME
3480 && TREE_CODE (TREE_TYPE (op0)) == BOOLEAN_TYPE
3481 && is_gimple_min_invariant (op1))
3483 if ((TREE_CODE (cond) == EQ_EXPR && true_arm)
3484 || (TREE_CODE (cond) == NE_EXPR && ! true_arm))
3486 retval.src = op1;
3488 else
3490 if (integer_zerop (op1))
3491 retval.src = boolean_true_node;
3492 else
3493 retval.src = boolean_false_node;
3495 retval.dst = op0;
3496 return retval;
3499 if (TREE_CODE (op0) == SSA_NAME
3500 && (is_gimple_min_invariant (op1) || TREE_CODE (op1) == SSA_NAME))
3502 tree inverted = invert_truthvalue (cond);
3504 /* When we find an available expression in the hash table, we replace
3505 the expression with the LHS of the statement in the hash table.
3507 So, we want to build statements such as "1 = <condition>" on the
3508 true arm and "0 = <condition>" on the false arm. That way if we
3509 find the expression in the table, we will replace it with its
3510 known constant value. Also insert inversions of the result and
3511 condition into the hash table. */
3512 if (true_arm)
3514 record_cond (cond, boolean_true_node, block_avail_exprs_p);
3515 record_dominating_conditions (cond, block_avail_exprs_p);
3516 record_cond (inverted, boolean_false_node, block_avail_exprs_p);
3518 if (TREE_CONSTANT (op1))
3519 record_range (cond, bb, vrp_variables_p);
3521 /* If the conditional is of the form 'X == Y', return 'X = Y'
3522 for the true arm. */
3523 if (TREE_CODE (cond) == EQ_EXPR)
3525 retval.dst = op0;
3526 retval.src = op1;
3527 return retval;
3530 else
3533 record_cond (inverted, boolean_true_node, block_avail_exprs_p);
3534 record_dominating_conditions (inverted, block_avail_exprs_p);
3535 record_cond (cond, boolean_false_node, block_avail_exprs_p);
3537 if (TREE_CONSTANT (op1))
3538 record_range (inverted, bb, vrp_variables_p);
3540 /* If the conditional is of the form 'X != Y', return 'X = Y'
3541 for the false arm. */
3542 if (TREE_CODE (cond) == NE_EXPR)
3544 retval.dst = op0;
3545 retval.src = op1;
3546 return retval;
3552 return retval;
3555 /* Hashing and equality functions for AVAIL_EXPRS. The table stores
3556 MODIFY_EXPR statements. We compute a value number for expressions using
3557 the code of the expression and the SSA numbers of its operands. */
3559 static hashval_t
3560 avail_expr_hash (const void *p)
3562 stmt_ann_t ann = ((struct expr_hash_elt *)p)->ann;
3563 tree rhs = ((struct expr_hash_elt *)p)->rhs;
3564 hashval_t val = 0;
3565 size_t i;
3566 vuse_optype vuses;
3568 /* iterative_hash_expr knows how to deal with any expression and
3569 deals with commutative operators as well, so just use it instead
3570 of duplicating such complexities here. */
3571 val = iterative_hash_expr (rhs, val);
3573 /* If the hash table entry is not associated with a statement, then we
3574 can just hash the expression and not worry about virtual operands
3575 and such. */
3576 if (!ann)
3577 return val;
3579 /* Add the SSA version numbers of every vuse operand. This is important
3580 because compound variables like arrays are not renamed in the
3581 operands. Rather, the rename is done on the virtual variable
3582 representing all the elements of the array. */
3583 vuses = VUSE_OPS (ann);
3584 for (i = 0; i < NUM_VUSES (vuses); i++)
3585 val = iterative_hash_expr (VUSE_OP (vuses, i), val);
3587 return val;
3590 static hashval_t
3591 real_avail_expr_hash (const void *p)
3593 return ((const struct expr_hash_elt *)p)->hash;
3596 static int
3597 avail_expr_eq (const void *p1, const void *p2)
3599 stmt_ann_t ann1 = ((struct expr_hash_elt *)p1)->ann;
3600 tree rhs1 = ((struct expr_hash_elt *)p1)->rhs;
3601 stmt_ann_t ann2 = ((struct expr_hash_elt *)p2)->ann;
3602 tree rhs2 = ((struct expr_hash_elt *)p2)->rhs;
3604 /* If they are the same physical expression, return true. */
3605 if (rhs1 == rhs2 && ann1 == ann2)
3606 return true;
3608 /* If their codes are not equal, then quit now. */
3609 if (TREE_CODE (rhs1) != TREE_CODE (rhs2))
3610 return false;
3612 /* In case of a collision, both RHS have to be identical and have the
3613 same VUSE operands. */
3614 if ((TREE_TYPE (rhs1) == TREE_TYPE (rhs2)
3615 || lang_hooks.types_compatible_p (TREE_TYPE (rhs1), TREE_TYPE (rhs2)))
3616 && operand_equal_p (rhs1, rhs2, OEP_PURE_SAME))
3618 vuse_optype ops1 = NULL;
3619 vuse_optype ops2 = NULL;
3620 size_t num_ops1 = 0;
3621 size_t num_ops2 = 0;
3622 size_t i;
3624 if (ann1)
3626 ops1 = VUSE_OPS (ann1);
3627 num_ops1 = NUM_VUSES (ops1);
3630 if (ann2)
3632 ops2 = VUSE_OPS (ann2);
3633 num_ops2 = NUM_VUSES (ops2);
3636 /* If the number of virtual uses is different, then we consider
3637 them not equal. */
3638 if (num_ops1 != num_ops2)
3639 return false;
3641 for (i = 0; i < num_ops1; i++)
3642 if (VUSE_OP (ops1, i) != VUSE_OP (ops2, i))
3643 return false;
3645 #ifdef ENABLE_CHECKING
3646 if (((struct expr_hash_elt *)p1)->hash
3647 != ((struct expr_hash_elt *)p2)->hash)
3648 abort ();
3649 #endif
3650 return true;
3653 return false;
3656 /* Given STMT and a pointer to the block local definitions BLOCK_DEFS_P,
3657 register register all objects set by this statement into BLOCK_DEFS_P
3658 and CURRDEFS. */
3660 static void
3661 register_definitions_for_stmt (stmt_ann_t ann, varray_type *block_defs_p)
3663 def_optype defs;
3664 v_may_def_optype v_may_defs;
3665 v_must_def_optype v_must_defs;
3666 unsigned int i;
3668 defs = DEF_OPS (ann);
3669 for (i = 0; i < NUM_DEFS (defs); i++)
3671 tree def = DEF_OP (defs, i);
3673 /* FIXME: We shouldn't be registering new defs if the variable
3674 doesn't need to be renamed. */
3675 register_new_def (def, block_defs_p);
3678 /* Register new virtual definitions made by the statement. */
3679 v_may_defs = V_MAY_DEF_OPS (ann);
3680 for (i = 0; i < NUM_V_MAY_DEFS (v_may_defs); i++)
3682 /* FIXME: We shouldn't be registering new defs if the variable
3683 doesn't need to be renamed. */
3684 register_new_def (V_MAY_DEF_RESULT (v_may_defs, i), block_defs_p);
3687 /* Register new virtual mustdefs made by the statement. */
3688 v_must_defs = V_MUST_DEF_OPS (ann);
3689 for (i = 0; i < NUM_V_MUST_DEFS (v_must_defs); i++)
3691 /* FIXME: We shouldn't be registering new defs if the variable
3692 doesn't need to be renamed. */
3693 register_new_def (V_MUST_DEF_OP (v_must_defs, i), block_defs_p);