* gimple-ssa-store-merging.c (struct store_immediate_info): Add
[official-gcc.git] / gcc / tree-ssa-phiopt.c
blob6e0e18689dab15eb932e8d7d7c28564271ea2b53
1 /* Optimization of PHI nodes by converting them into straightline code.
2 Copyright (C) 2004-2017 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it
7 under the terms of the GNU General Public License as published by the
8 Free Software Foundation; either version 3, or (at your option) any
9 later version.
11 GCC is distributed in the hope that it will be useful, but WITHOUT
12 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 #include "config.h"
21 #include "system.h"
22 #include "coretypes.h"
23 #include "backend.h"
24 #include "insn-codes.h"
25 #include "rtl.h"
26 #include "tree.h"
27 #include "gimple.h"
28 #include "cfghooks.h"
29 #include "tree-pass.h"
30 #include "ssa.h"
31 #include "optabs-tree.h"
32 #include "insn-config.h"
33 #include "gimple-pretty-print.h"
34 #include "fold-const.h"
35 #include "stor-layout.h"
36 #include "cfganal.h"
37 #include "gimplify.h"
38 #include "gimple-iterator.h"
39 #include "gimplify-me.h"
40 #include "tree-cfg.h"
41 #include "tree-dfa.h"
42 #include "domwalk.h"
43 #include "cfgloop.h"
44 #include "tree-data-ref.h"
45 #include "tree-scalar-evolution.h"
46 #include "tree-inline.h"
47 #include "params.h"
49 static unsigned int tree_ssa_phiopt_worker (bool, bool);
50 static bool conditional_replacement (basic_block, basic_block,
51 edge, edge, gphi *, tree, tree);
52 static gphi *factor_out_conditional_conversion (edge, edge, gphi *, tree, tree,
53 gimple *);
54 static int value_replacement (basic_block, basic_block,
55 edge, edge, gimple *, tree, tree);
56 static bool minmax_replacement (basic_block, basic_block,
57 edge, edge, gimple *, tree, tree);
58 static bool abs_replacement (basic_block, basic_block,
59 edge, edge, gimple *, tree, tree);
60 static bool cond_store_replacement (basic_block, basic_block, edge, edge,
61 hash_set<tree> *);
62 static bool cond_if_else_store_replacement (basic_block, basic_block, basic_block);
63 static hash_set<tree> * get_non_trapping ();
64 static void replace_phi_edge_with_variable (basic_block, edge, gimple *, tree);
65 static void hoist_adjacent_loads (basic_block, basic_block,
66 basic_block, basic_block);
67 static bool gate_hoist_loads (void);
69 /* This pass tries to transform conditional stores into unconditional
70 ones, enabling further simplifications with the simpler then and else
71 blocks. In particular it replaces this:
73 bb0:
74 if (cond) goto bb2; else goto bb1;
75 bb1:
76 *p = RHS;
77 bb2:
79 with
81 bb0:
82 if (cond) goto bb1; else goto bb2;
83 bb1:
84 condtmp' = *p;
85 bb2:
86 condtmp = PHI <RHS, condtmp'>
87 *p = condtmp;
89 This transformation can only be done under several constraints,
90 documented below. It also replaces:
92 bb0:
93 if (cond) goto bb2; else goto bb1;
94 bb1:
95 *p = RHS1;
96 goto bb3;
97 bb2:
98 *p = RHS2;
99 bb3:
101 with
103 bb0:
104 if (cond) goto bb3; else goto bb1;
105 bb1:
106 bb3:
107 condtmp = PHI <RHS1, RHS2>
108 *p = condtmp; */
110 static unsigned int
111 tree_ssa_cs_elim (void)
113 unsigned todo;
114 /* ??? We are not interested in loop related info, but the following
115 will create it, ICEing as we didn't init loops with pre-headers.
116 An interfacing issue of find_data_references_in_bb. */
117 loop_optimizer_init (LOOPS_NORMAL);
118 scev_initialize ();
119 todo = tree_ssa_phiopt_worker (true, false);
120 scev_finalize ();
121 loop_optimizer_finalize ();
122 return todo;
125 /* Return the singleton PHI in the SEQ of PHIs for edges E0 and E1. */
127 static gphi *
128 single_non_singleton_phi_for_edges (gimple_seq seq, edge e0, edge e1)
130 gimple_stmt_iterator i;
131 gphi *phi = NULL;
132 if (gimple_seq_singleton_p (seq))
133 return as_a <gphi *> (gsi_stmt (gsi_start (seq)));
134 for (i = gsi_start (seq); !gsi_end_p (i); gsi_next (&i))
136 gphi *p = as_a <gphi *> (gsi_stmt (i));
137 /* If the PHI arguments are equal then we can skip this PHI. */
138 if (operand_equal_for_phi_arg_p (gimple_phi_arg_def (p, e0->dest_idx),
139 gimple_phi_arg_def (p, e1->dest_idx)))
140 continue;
142 /* If we already have a PHI that has the two edge arguments are
143 different, then return it is not a singleton for these PHIs. */
144 if (phi)
145 return NULL;
147 phi = p;
149 return phi;
152 /* The core routine of conditional store replacement and normal
153 phi optimizations. Both share much of the infrastructure in how
154 to match applicable basic block patterns. DO_STORE_ELIM is true
155 when we want to do conditional store replacement, false otherwise.
156 DO_HOIST_LOADS is true when we want to hoist adjacent loads out
157 of diamond control flow patterns, false otherwise. */
158 static unsigned int
159 tree_ssa_phiopt_worker (bool do_store_elim, bool do_hoist_loads)
161 basic_block bb;
162 basic_block *bb_order;
163 unsigned n, i;
164 bool cfgchanged = false;
165 hash_set<tree> *nontrap = 0;
167 if (do_store_elim)
168 /* Calculate the set of non-trapping memory accesses. */
169 nontrap = get_non_trapping ();
171 /* Search every basic block for COND_EXPR we may be able to optimize.
173 We walk the blocks in order that guarantees that a block with
174 a single predecessor is processed before the predecessor.
175 This ensures that we collapse inner ifs before visiting the
176 outer ones, and also that we do not try to visit a removed
177 block. */
178 bb_order = single_pred_before_succ_order ();
179 n = n_basic_blocks_for_fn (cfun) - NUM_FIXED_BLOCKS;
181 for (i = 0; i < n; i++)
183 gimple *cond_stmt;
184 gphi *phi;
185 basic_block bb1, bb2;
186 edge e1, e2;
187 tree arg0, arg1;
189 bb = bb_order[i];
191 cond_stmt = last_stmt (bb);
192 /* Check to see if the last statement is a GIMPLE_COND. */
193 if (!cond_stmt
194 || gimple_code (cond_stmt) != GIMPLE_COND)
195 continue;
197 e1 = EDGE_SUCC (bb, 0);
198 bb1 = e1->dest;
199 e2 = EDGE_SUCC (bb, 1);
200 bb2 = e2->dest;
202 /* We cannot do the optimization on abnormal edges. */
203 if ((e1->flags & EDGE_ABNORMAL) != 0
204 || (e2->flags & EDGE_ABNORMAL) != 0)
205 continue;
207 /* If either bb1's succ or bb2 or bb2's succ is non NULL. */
208 if (EDGE_COUNT (bb1->succs) == 0
209 || bb2 == NULL
210 || EDGE_COUNT (bb2->succs) == 0)
211 continue;
213 /* Find the bb which is the fall through to the other. */
214 if (EDGE_SUCC (bb1, 0)->dest == bb2)
216 else if (EDGE_SUCC (bb2, 0)->dest == bb1)
218 std::swap (bb1, bb2);
219 std::swap (e1, e2);
221 else if (do_store_elim
222 && EDGE_SUCC (bb1, 0)->dest == EDGE_SUCC (bb2, 0)->dest)
224 basic_block bb3 = EDGE_SUCC (bb1, 0)->dest;
226 if (!single_succ_p (bb1)
227 || (EDGE_SUCC (bb1, 0)->flags & EDGE_FALLTHRU) == 0
228 || !single_succ_p (bb2)
229 || (EDGE_SUCC (bb2, 0)->flags & EDGE_FALLTHRU) == 0
230 || EDGE_COUNT (bb3->preds) != 2)
231 continue;
232 if (cond_if_else_store_replacement (bb1, bb2, bb3))
233 cfgchanged = true;
234 continue;
236 else if (do_hoist_loads
237 && EDGE_SUCC (bb1, 0)->dest == EDGE_SUCC (bb2, 0)->dest)
239 basic_block bb3 = EDGE_SUCC (bb1, 0)->dest;
241 if (!FLOAT_TYPE_P (TREE_TYPE (gimple_cond_lhs (cond_stmt)))
242 && single_succ_p (bb1)
243 && single_succ_p (bb2)
244 && single_pred_p (bb1)
245 && single_pred_p (bb2)
246 && EDGE_COUNT (bb->succs) == 2
247 && EDGE_COUNT (bb3->preds) == 2
248 /* If one edge or the other is dominant, a conditional move
249 is likely to perform worse than the well-predicted branch. */
250 && !predictable_edge_p (EDGE_SUCC (bb, 0))
251 && !predictable_edge_p (EDGE_SUCC (bb, 1)))
252 hoist_adjacent_loads (bb, bb1, bb2, bb3);
253 continue;
255 else
256 continue;
258 e1 = EDGE_SUCC (bb1, 0);
260 /* Make sure that bb1 is just a fall through. */
261 if (!single_succ_p (bb1)
262 || (e1->flags & EDGE_FALLTHRU) == 0)
263 continue;
265 /* Also make sure that bb1 only have one predecessor and that it
266 is bb. */
267 if (!single_pred_p (bb1)
268 || single_pred (bb1) != bb)
269 continue;
271 if (do_store_elim)
273 /* bb1 is the middle block, bb2 the join block, bb the split block,
274 e1 the fallthrough edge from bb1 to bb2. We can't do the
275 optimization if the join block has more than two predecessors. */
276 if (EDGE_COUNT (bb2->preds) > 2)
277 continue;
278 if (cond_store_replacement (bb1, bb2, e1, e2, nontrap))
279 cfgchanged = true;
281 else
283 gimple_seq phis = phi_nodes (bb2);
284 gimple_stmt_iterator gsi;
285 bool candorest = true;
287 /* Value replacement can work with more than one PHI
288 so try that first. */
289 for (gsi = gsi_start (phis); !gsi_end_p (gsi); gsi_next (&gsi))
291 phi = as_a <gphi *> (gsi_stmt (gsi));
292 arg0 = gimple_phi_arg_def (phi, e1->dest_idx);
293 arg1 = gimple_phi_arg_def (phi, e2->dest_idx);
294 if (value_replacement (bb, bb1, e1, e2, phi, arg0, arg1) == 2)
296 candorest = false;
297 cfgchanged = true;
298 break;
302 if (!candorest)
303 continue;
305 phi = single_non_singleton_phi_for_edges (phis, e1, e2);
306 if (!phi)
307 continue;
309 arg0 = gimple_phi_arg_def (phi, e1->dest_idx);
310 arg1 = gimple_phi_arg_def (phi, e2->dest_idx);
312 /* Something is wrong if we cannot find the arguments in the PHI
313 node. */
314 gcc_assert (arg0 != NULL_TREE && arg1 != NULL_TREE);
316 gphi *newphi = factor_out_conditional_conversion (e1, e2, phi,
317 arg0, arg1,
318 cond_stmt);
319 if (newphi != NULL)
321 phi = newphi;
322 /* factor_out_conditional_conversion may create a new PHI in
323 BB2 and eliminate an existing PHI in BB2. Recompute values
324 that may be affected by that change. */
325 arg0 = gimple_phi_arg_def (phi, e1->dest_idx);
326 arg1 = gimple_phi_arg_def (phi, e2->dest_idx);
327 gcc_assert (arg0 != NULL_TREE && arg1 != NULL_TREE);
330 /* Do the replacement of conditional if it can be done. */
331 if (conditional_replacement (bb, bb1, e1, e2, phi, arg0, arg1))
332 cfgchanged = true;
333 else if (abs_replacement (bb, bb1, e1, e2, phi, arg0, arg1))
334 cfgchanged = true;
335 else if (minmax_replacement (bb, bb1, e1, e2, phi, arg0, arg1))
336 cfgchanged = true;
340 free (bb_order);
342 if (do_store_elim)
343 delete nontrap;
344 /* If the CFG has changed, we should cleanup the CFG. */
345 if (cfgchanged && do_store_elim)
347 /* In cond-store replacement we have added some loads on edges
348 and new VOPS (as we moved the store, and created a load). */
349 gsi_commit_edge_inserts ();
350 return TODO_cleanup_cfg | TODO_update_ssa_only_virtuals;
352 else if (cfgchanged)
353 return TODO_cleanup_cfg;
354 return 0;
357 /* Replace PHI node element whose edge is E in block BB with variable NEW.
358 Remove the edge from COND_BLOCK which does not lead to BB (COND_BLOCK
359 is known to have two edges, one of which must reach BB). */
361 static void
362 replace_phi_edge_with_variable (basic_block cond_block,
363 edge e, gimple *phi, tree new_tree)
365 basic_block bb = gimple_bb (phi);
366 basic_block block_to_remove;
367 gimple_stmt_iterator gsi;
369 /* Change the PHI argument to new. */
370 SET_USE (PHI_ARG_DEF_PTR (phi, e->dest_idx), new_tree);
372 /* Remove the empty basic block. */
373 if (EDGE_SUCC (cond_block, 0)->dest == bb)
375 EDGE_SUCC (cond_block, 0)->flags |= EDGE_FALLTHRU;
376 EDGE_SUCC (cond_block, 0)->flags &= ~(EDGE_TRUE_VALUE | EDGE_FALSE_VALUE);
377 EDGE_SUCC (cond_block, 0)->probability = profile_probability::always ();
379 block_to_remove = EDGE_SUCC (cond_block, 1)->dest;
381 else
383 EDGE_SUCC (cond_block, 1)->flags |= EDGE_FALLTHRU;
384 EDGE_SUCC (cond_block, 1)->flags
385 &= ~(EDGE_TRUE_VALUE | EDGE_FALSE_VALUE);
386 EDGE_SUCC (cond_block, 1)->probability = profile_probability::always ();
388 block_to_remove = EDGE_SUCC (cond_block, 0)->dest;
390 delete_basic_block (block_to_remove);
392 /* Eliminate the COND_EXPR at the end of COND_BLOCK. */
393 gsi = gsi_last_bb (cond_block);
394 gsi_remove (&gsi, true);
396 if (dump_file && (dump_flags & TDF_DETAILS))
397 fprintf (dump_file,
398 "COND_EXPR in block %d and PHI in block %d converted to straightline code.\n",
399 cond_block->index,
400 bb->index);
403 /* PR66726: Factor conversion out of COND_EXPR. If the arguments of the PHI
404 stmt are CONVERT_STMT, factor out the conversion and perform the conversion
405 to the result of PHI stmt. COND_STMT is the controlling predicate.
406 Return the newly-created PHI, if any. */
408 static gphi *
409 factor_out_conditional_conversion (edge e0, edge e1, gphi *phi,
410 tree arg0, tree arg1, gimple *cond_stmt)
412 gimple *arg0_def_stmt = NULL, *arg1_def_stmt = NULL, *new_stmt;
413 tree new_arg0 = NULL_TREE, new_arg1 = NULL_TREE;
414 tree temp, result;
415 gphi *newphi;
416 gimple_stmt_iterator gsi, gsi_for_def;
417 source_location locus = gimple_location (phi);
418 enum tree_code convert_code;
420 /* Handle only PHI statements with two arguments. TODO: If all
421 other arguments to PHI are INTEGER_CST or if their defining
422 statement have the same unary operation, we can handle more
423 than two arguments too. */
424 if (gimple_phi_num_args (phi) != 2)
425 return NULL;
427 /* First canonicalize to simplify tests. */
428 if (TREE_CODE (arg0) != SSA_NAME)
430 std::swap (arg0, arg1);
431 std::swap (e0, e1);
434 if (TREE_CODE (arg0) != SSA_NAME
435 || (TREE_CODE (arg1) != SSA_NAME
436 && TREE_CODE (arg1) != INTEGER_CST))
437 return NULL;
439 /* Check if arg0 is an SSA_NAME and the stmt which defines arg0 is
440 a conversion. */
441 arg0_def_stmt = SSA_NAME_DEF_STMT (arg0);
442 if (!gimple_assign_cast_p (arg0_def_stmt))
443 return NULL;
445 /* Use the RHS as new_arg0. */
446 convert_code = gimple_assign_rhs_code (arg0_def_stmt);
447 new_arg0 = gimple_assign_rhs1 (arg0_def_stmt);
448 if (convert_code == VIEW_CONVERT_EXPR)
450 new_arg0 = TREE_OPERAND (new_arg0, 0);
451 if (!is_gimple_reg_type (TREE_TYPE (new_arg0)))
452 return NULL;
455 if (TREE_CODE (arg1) == SSA_NAME)
457 /* Check if arg1 is an SSA_NAME and the stmt which defines arg1
458 is a conversion. */
459 arg1_def_stmt = SSA_NAME_DEF_STMT (arg1);
460 if (!is_gimple_assign (arg1_def_stmt)
461 || gimple_assign_rhs_code (arg1_def_stmt) != convert_code)
462 return NULL;
464 /* Use the RHS as new_arg1. */
465 new_arg1 = gimple_assign_rhs1 (arg1_def_stmt);
466 if (convert_code == VIEW_CONVERT_EXPR)
467 new_arg1 = TREE_OPERAND (new_arg1, 0);
469 else
471 /* If arg1 is an INTEGER_CST, fold it to new type. */
472 if (INTEGRAL_TYPE_P (TREE_TYPE (new_arg0))
473 && int_fits_type_p (arg1, TREE_TYPE (new_arg0)))
475 if (gimple_assign_cast_p (arg0_def_stmt))
477 /* For the INTEGER_CST case, we are just moving the
478 conversion from one place to another, which can often
479 hurt as the conversion moves further away from the
480 statement that computes the value. So, perform this
481 only if new_arg0 is an operand of COND_STMT, or
482 if arg0_def_stmt is the only non-debug stmt in
483 its basic block, because then it is possible this
484 could enable further optimizations (minmax replacement
485 etc.). See PR71016. */
486 if (new_arg0 != gimple_cond_lhs (cond_stmt)
487 && new_arg0 != gimple_cond_rhs (cond_stmt)
488 && gimple_bb (arg0_def_stmt) == e0->src)
490 gsi = gsi_for_stmt (arg0_def_stmt);
491 gsi_prev_nondebug (&gsi);
492 if (!gsi_end_p (gsi))
493 return NULL;
494 gsi = gsi_for_stmt (arg0_def_stmt);
495 gsi_next_nondebug (&gsi);
496 if (!gsi_end_p (gsi))
497 return NULL;
499 new_arg1 = fold_convert (TREE_TYPE (new_arg0), arg1);
501 else
502 return NULL;
504 else
505 return NULL;
508 /* If arg0/arg1 have > 1 use, then this transformation actually increases
509 the number of expressions evaluated at runtime. */
510 if (!has_single_use (arg0)
511 || (arg1_def_stmt && !has_single_use (arg1)))
512 return NULL;
514 /* If types of new_arg0 and new_arg1 are different bailout. */
515 if (!types_compatible_p (TREE_TYPE (new_arg0), TREE_TYPE (new_arg1)))
516 return NULL;
518 /* Create a new PHI stmt. */
519 result = PHI_RESULT (phi);
520 temp = make_ssa_name (TREE_TYPE (new_arg0), NULL);
521 newphi = create_phi_node (temp, gimple_bb (phi));
523 if (dump_file && (dump_flags & TDF_DETAILS))
525 fprintf (dump_file, "PHI ");
526 print_generic_expr (dump_file, gimple_phi_result (phi));
527 fprintf (dump_file,
528 " changed to factor conversion out from COND_EXPR.\n");
529 fprintf (dump_file, "New stmt with CAST that defines ");
530 print_generic_expr (dump_file, result);
531 fprintf (dump_file, ".\n");
534 /* Remove the old cast(s) that has single use. */
535 gsi_for_def = gsi_for_stmt (arg0_def_stmt);
536 gsi_remove (&gsi_for_def, true);
537 release_defs (arg0_def_stmt);
539 if (arg1_def_stmt)
541 gsi_for_def = gsi_for_stmt (arg1_def_stmt);
542 gsi_remove (&gsi_for_def, true);
543 release_defs (arg1_def_stmt);
546 add_phi_arg (newphi, new_arg0, e0, locus);
547 add_phi_arg (newphi, new_arg1, e1, locus);
549 /* Create the conversion stmt and insert it. */
550 if (convert_code == VIEW_CONVERT_EXPR)
551 temp = fold_build1 (VIEW_CONVERT_EXPR, TREE_TYPE (result), temp);
552 new_stmt = gimple_build_assign (result, convert_code, temp);
553 gsi = gsi_after_labels (gimple_bb (phi));
554 gsi_insert_before (&gsi, new_stmt, GSI_SAME_STMT);
556 /* Remove the original PHI stmt. */
557 gsi = gsi_for_stmt (phi);
558 gsi_remove (&gsi, true);
559 return newphi;
562 /* The function conditional_replacement does the main work of doing the
563 conditional replacement. Return true if the replacement is done.
564 Otherwise return false.
565 BB is the basic block where the replacement is going to be done on. ARG0
566 is argument 0 from PHI. Likewise for ARG1. */
568 static bool
569 conditional_replacement (basic_block cond_bb, basic_block middle_bb,
570 edge e0, edge e1, gphi *phi,
571 tree arg0, tree arg1)
573 tree result;
574 gimple *stmt;
575 gassign *new_stmt;
576 tree cond;
577 gimple_stmt_iterator gsi;
578 edge true_edge, false_edge;
579 tree new_var, new_var2;
580 bool neg;
582 /* FIXME: Gimplification of complex type is too hard for now. */
583 /* We aren't prepared to handle vectors either (and it is a question
584 if it would be worthwhile anyway). */
585 if (!(INTEGRAL_TYPE_P (TREE_TYPE (arg0))
586 || POINTER_TYPE_P (TREE_TYPE (arg0)))
587 || !(INTEGRAL_TYPE_P (TREE_TYPE (arg1))
588 || POINTER_TYPE_P (TREE_TYPE (arg1))))
589 return false;
591 /* The PHI arguments have the constants 0 and 1, or 0 and -1, then
592 convert it to the conditional. */
593 if ((integer_zerop (arg0) && integer_onep (arg1))
594 || (integer_zerop (arg1) && integer_onep (arg0)))
595 neg = false;
596 else if ((integer_zerop (arg0) && integer_all_onesp (arg1))
597 || (integer_zerop (arg1) && integer_all_onesp (arg0)))
598 neg = true;
599 else
600 return false;
602 if (!empty_block_p (middle_bb))
603 return false;
605 /* At this point we know we have a GIMPLE_COND with two successors.
606 One successor is BB, the other successor is an empty block which
607 falls through into BB.
609 There is a single PHI node at the join point (BB) and its arguments
610 are constants (0, 1) or (0, -1).
612 So, given the condition COND, and the two PHI arguments, we can
613 rewrite this PHI into non-branching code:
615 dest = (COND) or dest = COND'
617 We use the condition as-is if the argument associated with the
618 true edge has the value one or the argument associated with the
619 false edge as the value zero. Note that those conditions are not
620 the same since only one of the outgoing edges from the GIMPLE_COND
621 will directly reach BB and thus be associated with an argument. */
623 stmt = last_stmt (cond_bb);
624 result = PHI_RESULT (phi);
626 /* To handle special cases like floating point comparison, it is easier and
627 less error-prone to build a tree and gimplify it on the fly though it is
628 less efficient. */
629 cond = fold_build2_loc (gimple_location (stmt),
630 gimple_cond_code (stmt), boolean_type_node,
631 gimple_cond_lhs (stmt), gimple_cond_rhs (stmt));
633 /* We need to know which is the true edge and which is the false
634 edge so that we know when to invert the condition below. */
635 extract_true_false_edges_from_block (cond_bb, &true_edge, &false_edge);
636 if ((e0 == true_edge && integer_zerop (arg0))
637 || (e0 == false_edge && !integer_zerop (arg0))
638 || (e1 == true_edge && integer_zerop (arg1))
639 || (e1 == false_edge && !integer_zerop (arg1)))
640 cond = fold_build1_loc (gimple_location (stmt),
641 TRUTH_NOT_EXPR, TREE_TYPE (cond), cond);
643 if (neg)
645 cond = fold_convert_loc (gimple_location (stmt),
646 TREE_TYPE (result), cond);
647 cond = fold_build1_loc (gimple_location (stmt),
648 NEGATE_EXPR, TREE_TYPE (cond), cond);
651 /* Insert our new statements at the end of conditional block before the
652 COND_STMT. */
653 gsi = gsi_for_stmt (stmt);
654 new_var = force_gimple_operand_gsi (&gsi, cond, true, NULL, true,
655 GSI_SAME_STMT);
657 if (!useless_type_conversion_p (TREE_TYPE (result), TREE_TYPE (new_var)))
659 source_location locus_0, locus_1;
661 new_var2 = make_ssa_name (TREE_TYPE (result));
662 new_stmt = gimple_build_assign (new_var2, CONVERT_EXPR, new_var);
663 gsi_insert_before (&gsi, new_stmt, GSI_SAME_STMT);
664 new_var = new_var2;
666 /* Set the locus to the first argument, unless is doesn't have one. */
667 locus_0 = gimple_phi_arg_location (phi, 0);
668 locus_1 = gimple_phi_arg_location (phi, 1);
669 if (locus_0 == UNKNOWN_LOCATION)
670 locus_0 = locus_1;
671 gimple_set_location (new_stmt, locus_0);
674 replace_phi_edge_with_variable (cond_bb, e1, phi, new_var);
675 reset_flow_sensitive_info_in_bb (cond_bb);
677 /* Note that we optimized this PHI. */
678 return true;
681 /* Update *ARG which is defined in STMT so that it contains the
682 computed value if that seems profitable. Return true if the
683 statement is made dead by that rewriting. */
685 static bool
686 jump_function_from_stmt (tree *arg, gimple *stmt)
688 enum tree_code code = gimple_assign_rhs_code (stmt);
689 if (code == ADDR_EXPR)
691 /* For arg = &p->i transform it to p, if possible. */
692 tree rhs1 = gimple_assign_rhs1 (stmt);
693 HOST_WIDE_INT offset;
694 tree tem = get_addr_base_and_unit_offset (TREE_OPERAND (rhs1, 0),
695 &offset);
696 if (tem
697 && TREE_CODE (tem) == MEM_REF
698 && (mem_ref_offset (tem) + offset) == 0)
700 *arg = TREE_OPERAND (tem, 0);
701 return true;
704 /* TODO: Much like IPA-CP jump-functions we want to handle constant
705 additions symbolically here, and we'd need to update the comparison
706 code that compares the arg + cst tuples in our caller. For now the
707 code above exactly handles the VEC_BASE pattern from vec.h. */
708 return false;
711 /* RHS is a source argument in a BIT_AND_EXPR which feeds a conditional
712 of the form SSA_NAME NE 0.
714 If RHS is fed by a simple EQ_EXPR comparison of two values, see if
715 the two input values of the EQ_EXPR match arg0 and arg1.
717 If so update *code and return TRUE. Otherwise return FALSE. */
719 static bool
720 rhs_is_fed_for_value_replacement (const_tree arg0, const_tree arg1,
721 enum tree_code *code, const_tree rhs)
723 /* Obviously if RHS is not an SSA_NAME, we can't look at the defining
724 statement. */
725 if (TREE_CODE (rhs) == SSA_NAME)
727 gimple *def1 = SSA_NAME_DEF_STMT (rhs);
729 /* Verify the defining statement has an EQ_EXPR on the RHS. */
730 if (is_gimple_assign (def1) && gimple_assign_rhs_code (def1) == EQ_EXPR)
732 /* Finally verify the source operands of the EQ_EXPR are equal
733 to arg0 and arg1. */
734 tree op0 = gimple_assign_rhs1 (def1);
735 tree op1 = gimple_assign_rhs2 (def1);
736 if ((operand_equal_for_phi_arg_p (arg0, op0)
737 && operand_equal_for_phi_arg_p (arg1, op1))
738 || (operand_equal_for_phi_arg_p (arg0, op1)
739 && operand_equal_for_phi_arg_p (arg1, op0)))
741 /* We will perform the optimization. */
742 *code = gimple_assign_rhs_code (def1);
743 return true;
747 return false;
750 /* Return TRUE if arg0/arg1 are equal to the rhs/lhs or lhs/rhs of COND.
752 Also return TRUE if arg0/arg1 are equal to the source arguments of a
753 an EQ comparison feeding a BIT_AND_EXPR which feeds COND.
755 Return FALSE otherwise. */
757 static bool
758 operand_equal_for_value_replacement (const_tree arg0, const_tree arg1,
759 enum tree_code *code, gimple *cond)
761 gimple *def;
762 tree lhs = gimple_cond_lhs (cond);
763 tree rhs = gimple_cond_rhs (cond);
765 if ((operand_equal_for_phi_arg_p (arg0, lhs)
766 && operand_equal_for_phi_arg_p (arg1, rhs))
767 || (operand_equal_for_phi_arg_p (arg1, lhs)
768 && operand_equal_for_phi_arg_p (arg0, rhs)))
769 return true;
771 /* Now handle more complex case where we have an EQ comparison
772 which feeds a BIT_AND_EXPR which feeds COND.
774 First verify that COND is of the form SSA_NAME NE 0. */
775 if (*code != NE_EXPR || !integer_zerop (rhs)
776 || TREE_CODE (lhs) != SSA_NAME)
777 return false;
779 /* Now ensure that SSA_NAME is set by a BIT_AND_EXPR. */
780 def = SSA_NAME_DEF_STMT (lhs);
781 if (!is_gimple_assign (def) || gimple_assign_rhs_code (def) != BIT_AND_EXPR)
782 return false;
784 /* Now verify arg0/arg1 correspond to the source arguments of an
785 EQ comparison feeding the BIT_AND_EXPR. */
787 tree tmp = gimple_assign_rhs1 (def);
788 if (rhs_is_fed_for_value_replacement (arg0, arg1, code, tmp))
789 return true;
791 tmp = gimple_assign_rhs2 (def);
792 if (rhs_is_fed_for_value_replacement (arg0, arg1, code, tmp))
793 return true;
795 return false;
798 /* Returns true if ARG is a neutral element for operation CODE
799 on the RIGHT side. */
801 static bool
802 neutral_element_p (tree_code code, tree arg, bool right)
804 switch (code)
806 case PLUS_EXPR:
807 case BIT_IOR_EXPR:
808 case BIT_XOR_EXPR:
809 return integer_zerop (arg);
811 case LROTATE_EXPR:
812 case RROTATE_EXPR:
813 case LSHIFT_EXPR:
814 case RSHIFT_EXPR:
815 case MINUS_EXPR:
816 case POINTER_PLUS_EXPR:
817 return right && integer_zerop (arg);
819 case MULT_EXPR:
820 return integer_onep (arg);
822 case TRUNC_DIV_EXPR:
823 case CEIL_DIV_EXPR:
824 case FLOOR_DIV_EXPR:
825 case ROUND_DIV_EXPR:
826 case EXACT_DIV_EXPR:
827 return right && integer_onep (arg);
829 case BIT_AND_EXPR:
830 return integer_all_onesp (arg);
832 default:
833 return false;
837 /* Returns true if ARG is an absorbing element for operation CODE. */
839 static bool
840 absorbing_element_p (tree_code code, tree arg, bool right, tree rval)
842 switch (code)
844 case BIT_IOR_EXPR:
845 return integer_all_onesp (arg);
847 case MULT_EXPR:
848 case BIT_AND_EXPR:
849 return integer_zerop (arg);
851 case LSHIFT_EXPR:
852 case RSHIFT_EXPR:
853 case LROTATE_EXPR:
854 case RROTATE_EXPR:
855 return !right && integer_zerop (arg);
857 case TRUNC_DIV_EXPR:
858 case CEIL_DIV_EXPR:
859 case FLOOR_DIV_EXPR:
860 case ROUND_DIV_EXPR:
861 case EXACT_DIV_EXPR:
862 case TRUNC_MOD_EXPR:
863 case CEIL_MOD_EXPR:
864 case FLOOR_MOD_EXPR:
865 case ROUND_MOD_EXPR:
866 return (!right
867 && integer_zerop (arg)
868 && tree_single_nonzero_warnv_p (rval, NULL));
870 default:
871 return false;
875 /* The function value_replacement does the main work of doing the value
876 replacement. Return non-zero if the replacement is done. Otherwise return
877 0. If we remove the middle basic block, return 2.
878 BB is the basic block where the replacement is going to be done on. ARG0
879 is argument 0 from the PHI. Likewise for ARG1. */
881 static int
882 value_replacement (basic_block cond_bb, basic_block middle_bb,
883 edge e0, edge e1, gimple *phi,
884 tree arg0, tree arg1)
886 gimple_stmt_iterator gsi;
887 gimple *cond;
888 edge true_edge, false_edge;
889 enum tree_code code;
890 bool emtpy_or_with_defined_p = true;
892 /* If the type says honor signed zeros we cannot do this
893 optimization. */
894 if (HONOR_SIGNED_ZEROS (arg1))
895 return 0;
897 /* If there is a statement in MIDDLE_BB that defines one of the PHI
898 arguments, then adjust arg0 or arg1. */
899 gsi = gsi_start_nondebug_after_labels_bb (middle_bb);
900 while (!gsi_end_p (gsi))
902 gimple *stmt = gsi_stmt (gsi);
903 tree lhs;
904 gsi_next_nondebug (&gsi);
905 if (!is_gimple_assign (stmt))
907 emtpy_or_with_defined_p = false;
908 continue;
910 /* Now try to adjust arg0 or arg1 according to the computation
911 in the statement. */
912 lhs = gimple_assign_lhs (stmt);
913 if (!(lhs == arg0
914 && jump_function_from_stmt (&arg0, stmt))
915 || (lhs == arg1
916 && jump_function_from_stmt (&arg1, stmt)))
917 emtpy_or_with_defined_p = false;
920 cond = last_stmt (cond_bb);
921 code = gimple_cond_code (cond);
923 /* This transformation is only valid for equality comparisons. */
924 if (code != NE_EXPR && code != EQ_EXPR)
925 return 0;
927 /* We need to know which is the true edge and which is the false
928 edge so that we know if have abs or negative abs. */
929 extract_true_false_edges_from_block (cond_bb, &true_edge, &false_edge);
931 /* At this point we know we have a COND_EXPR with two successors.
932 One successor is BB, the other successor is an empty block which
933 falls through into BB.
935 The condition for the COND_EXPR is known to be NE_EXPR or EQ_EXPR.
937 There is a single PHI node at the join point (BB) with two arguments.
939 We now need to verify that the two arguments in the PHI node match
940 the two arguments to the equality comparison. */
942 if (operand_equal_for_value_replacement (arg0, arg1, &code, cond))
944 edge e;
945 tree arg;
947 /* For NE_EXPR, we want to build an assignment result = arg where
948 arg is the PHI argument associated with the true edge. For
949 EQ_EXPR we want the PHI argument associated with the false edge. */
950 e = (code == NE_EXPR ? true_edge : false_edge);
952 /* Unfortunately, E may not reach BB (it may instead have gone to
953 OTHER_BLOCK). If that is the case, then we want the single outgoing
954 edge from OTHER_BLOCK which reaches BB and represents the desired
955 path from COND_BLOCK. */
956 if (e->dest == middle_bb)
957 e = single_succ_edge (e->dest);
959 /* Now we know the incoming edge to BB that has the argument for the
960 RHS of our new assignment statement. */
961 if (e0 == e)
962 arg = arg0;
963 else
964 arg = arg1;
966 /* If the middle basic block was empty or is defining the
967 PHI arguments and this is a single phi where the args are different
968 for the edges e0 and e1 then we can remove the middle basic block. */
969 if (emtpy_or_with_defined_p
970 && single_non_singleton_phi_for_edges (phi_nodes (gimple_bb (phi)),
971 e0, e1) == phi)
973 replace_phi_edge_with_variable (cond_bb, e1, phi, arg);
974 /* Note that we optimized this PHI. */
975 return 2;
977 else
979 /* Replace the PHI arguments with arg. */
980 SET_PHI_ARG_DEF (phi, e0->dest_idx, arg);
981 SET_PHI_ARG_DEF (phi, e1->dest_idx, arg);
982 if (dump_file && (dump_flags & TDF_DETAILS))
984 fprintf (dump_file, "PHI ");
985 print_generic_expr (dump_file, gimple_phi_result (phi));
986 fprintf (dump_file, " reduced for COND_EXPR in block %d to ",
987 cond_bb->index);
988 print_generic_expr (dump_file, arg);
989 fprintf (dump_file, ".\n");
991 return 1;
996 /* Now optimize (x != 0) ? x + y : y to just x + y. */
997 gsi = gsi_last_nondebug_bb (middle_bb);
998 if (gsi_end_p (gsi))
999 return 0;
1001 gimple *assign = gsi_stmt (gsi);
1002 if (!is_gimple_assign (assign)
1003 || gimple_assign_rhs_class (assign) != GIMPLE_BINARY_RHS
1004 || (!INTEGRAL_TYPE_P (TREE_TYPE (arg0))
1005 && !POINTER_TYPE_P (TREE_TYPE (arg0))))
1006 return 0;
1008 /* Punt if there are (degenerate) PHIs in middle_bb, there should not be. */
1009 if (!gimple_seq_empty_p (phi_nodes (middle_bb)))
1010 return 0;
1012 /* Allow up to 2 cheap preparation statements that prepare argument
1013 for assign, e.g.:
1014 if (y_4 != 0)
1015 goto <bb 3>;
1016 else
1017 goto <bb 4>;
1018 <bb 3>:
1019 _1 = (int) y_4;
1020 iftmp.0_6 = x_5(D) r<< _1;
1021 <bb 4>:
1022 # iftmp.0_2 = PHI <iftmp.0_6(3), x_5(D)(2)>
1024 if (y_3(D) == 0)
1025 goto <bb 4>;
1026 else
1027 goto <bb 3>;
1028 <bb 3>:
1029 y_4 = y_3(D) & 31;
1030 _1 = (int) y_4;
1031 _6 = x_5(D) r<< _1;
1032 <bb 4>:
1033 # _2 = PHI <x_5(D)(2), _6(3)> */
1034 gimple *prep_stmt[2] = { NULL, NULL };
1035 int prep_cnt;
1036 for (prep_cnt = 0; ; prep_cnt++)
1038 gsi_prev_nondebug (&gsi);
1039 if (gsi_end_p (gsi))
1040 break;
1042 gimple *g = gsi_stmt (gsi);
1043 if (gimple_code (g) == GIMPLE_LABEL)
1044 break;
1046 if (prep_cnt == 2 || !is_gimple_assign (g))
1047 return 0;
1049 tree lhs = gimple_assign_lhs (g);
1050 tree rhs1 = gimple_assign_rhs1 (g);
1051 use_operand_p use_p;
1052 gimple *use_stmt;
1053 if (TREE_CODE (lhs) != SSA_NAME
1054 || TREE_CODE (rhs1) != SSA_NAME
1055 || !INTEGRAL_TYPE_P (TREE_TYPE (lhs))
1056 || !INTEGRAL_TYPE_P (TREE_TYPE (rhs1))
1057 || !single_imm_use (lhs, &use_p, &use_stmt)
1058 || use_stmt != (prep_cnt ? prep_stmt[prep_cnt - 1] : assign))
1059 return 0;
1060 switch (gimple_assign_rhs_code (g))
1062 CASE_CONVERT:
1063 break;
1064 case PLUS_EXPR:
1065 case BIT_AND_EXPR:
1066 case BIT_IOR_EXPR:
1067 case BIT_XOR_EXPR:
1068 if (TREE_CODE (gimple_assign_rhs2 (g)) != INTEGER_CST)
1069 return 0;
1070 break;
1071 default:
1072 return 0;
1074 prep_stmt[prep_cnt] = g;
1077 /* Only transform if it removes the condition. */
1078 if (!single_non_singleton_phi_for_edges (phi_nodes (gimple_bb (phi)), e0, e1))
1079 return 0;
1081 /* Size-wise, this is always profitable. */
1082 if (optimize_bb_for_speed_p (cond_bb)
1083 /* The special case is useless if it has a low probability. */
1084 && profile_status_for_fn (cfun) != PROFILE_ABSENT
1085 && EDGE_PRED (middle_bb, 0)->probability < profile_probability::even ()
1086 /* If assign is cheap, there is no point avoiding it. */
1087 && estimate_num_insns (bb_seq (middle_bb), &eni_time_weights)
1088 >= 3 * estimate_num_insns (cond, &eni_time_weights))
1089 return 0;
1091 tree lhs = gimple_assign_lhs (assign);
1092 tree rhs1 = gimple_assign_rhs1 (assign);
1093 tree rhs2 = gimple_assign_rhs2 (assign);
1094 enum tree_code code_def = gimple_assign_rhs_code (assign);
1095 tree cond_lhs = gimple_cond_lhs (cond);
1096 tree cond_rhs = gimple_cond_rhs (cond);
1098 /* Propagate the cond_rhs constant through preparation stmts,
1099 make sure UB isn't invoked while doing that. */
1100 for (int i = prep_cnt - 1; i >= 0; --i)
1102 gimple *g = prep_stmt[i];
1103 tree grhs1 = gimple_assign_rhs1 (g);
1104 if (!operand_equal_for_phi_arg_p (cond_lhs, grhs1))
1105 return 0;
1106 cond_lhs = gimple_assign_lhs (g);
1107 cond_rhs = fold_convert (TREE_TYPE (grhs1), cond_rhs);
1108 if (TREE_CODE (cond_rhs) != INTEGER_CST
1109 || TREE_OVERFLOW (cond_rhs))
1110 return 0;
1111 if (gimple_assign_rhs_class (g) == GIMPLE_BINARY_RHS)
1113 cond_rhs = int_const_binop (gimple_assign_rhs_code (g), cond_rhs,
1114 gimple_assign_rhs2 (g));
1115 if (TREE_OVERFLOW (cond_rhs))
1116 return 0;
1118 cond_rhs = fold_convert (TREE_TYPE (cond_lhs), cond_rhs);
1119 if (TREE_CODE (cond_rhs) != INTEGER_CST
1120 || TREE_OVERFLOW (cond_rhs))
1121 return 0;
1124 if (((code == NE_EXPR && e1 == false_edge)
1125 || (code == EQ_EXPR && e1 == true_edge))
1126 && arg0 == lhs
1127 && ((arg1 == rhs1
1128 && operand_equal_for_phi_arg_p (rhs2, cond_lhs)
1129 && neutral_element_p (code_def, cond_rhs, true))
1130 || (arg1 == rhs2
1131 && operand_equal_for_phi_arg_p (rhs1, cond_lhs)
1132 && neutral_element_p (code_def, cond_rhs, false))
1133 || (operand_equal_for_phi_arg_p (arg1, cond_rhs)
1134 && ((operand_equal_for_phi_arg_p (rhs2, cond_lhs)
1135 && absorbing_element_p (code_def, cond_rhs, true, rhs2))
1136 || (operand_equal_for_phi_arg_p (rhs1, cond_lhs)
1137 && absorbing_element_p (code_def,
1138 cond_rhs, false, rhs2))))))
1140 gsi = gsi_for_stmt (cond);
1141 if (INTEGRAL_TYPE_P (TREE_TYPE (lhs)))
1143 /* Moving ASSIGN might change VR of lhs, e.g. when moving u_6
1144 def-stmt in:
1145 if (n_5 != 0)
1146 goto <bb 3>;
1147 else
1148 goto <bb 4>;
1150 <bb 3>:
1151 # RANGE [0, 4294967294]
1152 u_6 = n_5 + 4294967295;
1154 <bb 4>:
1155 # u_3 = PHI <u_6(3), 4294967295(2)> */
1156 SSA_NAME_RANGE_INFO (lhs) = NULL;
1157 /* If available, we can use VR of phi result at least. */
1158 tree phires = gimple_phi_result (phi);
1159 struct range_info_def *phires_range_info
1160 = SSA_NAME_RANGE_INFO (phires);
1161 if (phires_range_info)
1162 duplicate_ssa_name_range_info (lhs, SSA_NAME_RANGE_TYPE (phires),
1163 phires_range_info);
1165 gimple_stmt_iterator gsi_from;
1166 for (int i = prep_cnt - 1; i >= 0; --i)
1168 tree plhs = gimple_assign_lhs (prep_stmt[i]);
1169 SSA_NAME_RANGE_INFO (plhs) = NULL;
1170 gsi_from = gsi_for_stmt (prep_stmt[i]);
1171 gsi_move_before (&gsi_from, &gsi);
1173 gsi_from = gsi_for_stmt (assign);
1174 gsi_move_before (&gsi_from, &gsi);
1175 replace_phi_edge_with_variable (cond_bb, e1, phi, lhs);
1176 return 2;
1179 return 0;
1182 /* The function minmax_replacement does the main work of doing the minmax
1183 replacement. Return true if the replacement is done. Otherwise return
1184 false.
1185 BB is the basic block where the replacement is going to be done on. ARG0
1186 is argument 0 from the PHI. Likewise for ARG1. */
1188 static bool
1189 minmax_replacement (basic_block cond_bb, basic_block middle_bb,
1190 edge e0, edge e1, gimple *phi,
1191 tree arg0, tree arg1)
1193 tree result, type;
1194 gcond *cond;
1195 gassign *new_stmt;
1196 edge true_edge, false_edge;
1197 enum tree_code cmp, minmax, ass_code;
1198 tree smaller, alt_smaller, larger, alt_larger, arg_true, arg_false;
1199 gimple_stmt_iterator gsi, gsi_from;
1201 type = TREE_TYPE (PHI_RESULT (phi));
1203 /* The optimization may be unsafe due to NaNs. */
1204 if (HONOR_NANS (type) || HONOR_SIGNED_ZEROS (type))
1205 return false;
1207 cond = as_a <gcond *> (last_stmt (cond_bb));
1208 cmp = gimple_cond_code (cond);
1210 /* This transformation is only valid for order comparisons. Record which
1211 operand is smaller/larger if the result of the comparison is true. */
1212 alt_smaller = NULL_TREE;
1213 alt_larger = NULL_TREE;
1214 if (cmp == LT_EXPR || cmp == LE_EXPR)
1216 smaller = gimple_cond_lhs (cond);
1217 larger = gimple_cond_rhs (cond);
1218 /* If we have smaller < CST it is equivalent to smaller <= CST-1.
1219 Likewise smaller <= CST is equivalent to smaller < CST+1. */
1220 if (TREE_CODE (larger) == INTEGER_CST)
1222 if (cmp == LT_EXPR)
1224 bool overflow;
1225 wide_int alt = wi::sub (wi::to_wide (larger), 1,
1226 TYPE_SIGN (TREE_TYPE (larger)),
1227 &overflow);
1228 if (! overflow)
1229 alt_larger = wide_int_to_tree (TREE_TYPE (larger), alt);
1231 else
1233 bool overflow;
1234 wide_int alt = wi::add (wi::to_wide (larger), 1,
1235 TYPE_SIGN (TREE_TYPE (larger)),
1236 &overflow);
1237 if (! overflow)
1238 alt_larger = wide_int_to_tree (TREE_TYPE (larger), alt);
1242 else if (cmp == GT_EXPR || cmp == GE_EXPR)
1244 smaller = gimple_cond_rhs (cond);
1245 larger = gimple_cond_lhs (cond);
1246 /* If we have larger > CST it is equivalent to larger >= CST+1.
1247 Likewise larger >= CST is equivalent to larger > CST-1. */
1248 if (TREE_CODE (smaller) == INTEGER_CST)
1250 if (cmp == GT_EXPR)
1252 bool overflow;
1253 wide_int alt = wi::add (wi::to_wide (smaller), 1,
1254 TYPE_SIGN (TREE_TYPE (smaller)),
1255 &overflow);
1256 if (! overflow)
1257 alt_smaller = wide_int_to_tree (TREE_TYPE (smaller), alt);
1259 else
1261 bool overflow;
1262 wide_int alt = wi::sub (wi::to_wide (smaller), 1,
1263 TYPE_SIGN (TREE_TYPE (smaller)),
1264 &overflow);
1265 if (! overflow)
1266 alt_smaller = wide_int_to_tree (TREE_TYPE (smaller), alt);
1270 else
1271 return false;
1273 /* We need to know which is the true edge and which is the false
1274 edge so that we know if have abs or negative abs. */
1275 extract_true_false_edges_from_block (cond_bb, &true_edge, &false_edge);
1277 /* Forward the edges over the middle basic block. */
1278 if (true_edge->dest == middle_bb)
1279 true_edge = EDGE_SUCC (true_edge->dest, 0);
1280 if (false_edge->dest == middle_bb)
1281 false_edge = EDGE_SUCC (false_edge->dest, 0);
1283 if (true_edge == e0)
1285 gcc_assert (false_edge == e1);
1286 arg_true = arg0;
1287 arg_false = arg1;
1289 else
1291 gcc_assert (false_edge == e0);
1292 gcc_assert (true_edge == e1);
1293 arg_true = arg1;
1294 arg_false = arg0;
1297 if (empty_block_p (middle_bb))
1299 if ((operand_equal_for_phi_arg_p (arg_true, smaller)
1300 || (alt_smaller
1301 && operand_equal_for_phi_arg_p (arg_true, alt_smaller)))
1302 && (operand_equal_for_phi_arg_p (arg_false, larger)
1303 || (alt_larger
1304 && operand_equal_for_phi_arg_p (arg_true, alt_larger))))
1306 /* Case
1308 if (smaller < larger)
1309 rslt = smaller;
1310 else
1311 rslt = larger; */
1312 minmax = MIN_EXPR;
1314 else if ((operand_equal_for_phi_arg_p (arg_false, smaller)
1315 || (alt_smaller
1316 && operand_equal_for_phi_arg_p (arg_false, alt_smaller)))
1317 && (operand_equal_for_phi_arg_p (arg_true, larger)
1318 || (alt_larger
1319 && operand_equal_for_phi_arg_p (arg_true, alt_larger))))
1320 minmax = MAX_EXPR;
1321 else
1322 return false;
1324 else
1326 /* Recognize the following case, assuming d <= u:
1328 if (a <= u)
1329 b = MAX (a, d);
1330 x = PHI <b, u>
1332 This is equivalent to
1334 b = MAX (a, d);
1335 x = MIN (b, u); */
1337 gimple *assign = last_and_only_stmt (middle_bb);
1338 tree lhs, op0, op1, bound;
1340 if (!assign
1341 || gimple_code (assign) != GIMPLE_ASSIGN)
1342 return false;
1344 lhs = gimple_assign_lhs (assign);
1345 ass_code = gimple_assign_rhs_code (assign);
1346 if (ass_code != MAX_EXPR && ass_code != MIN_EXPR)
1347 return false;
1348 op0 = gimple_assign_rhs1 (assign);
1349 op1 = gimple_assign_rhs2 (assign);
1351 if (true_edge->src == middle_bb)
1353 /* We got here if the condition is true, i.e., SMALLER < LARGER. */
1354 if (!operand_equal_for_phi_arg_p (lhs, arg_true))
1355 return false;
1357 if (operand_equal_for_phi_arg_p (arg_false, larger)
1358 || (alt_larger
1359 && operand_equal_for_phi_arg_p (arg_false, alt_larger)))
1361 /* Case
1363 if (smaller < larger)
1365 r' = MAX_EXPR (smaller, bound)
1367 r = PHI <r', larger> --> to be turned to MIN_EXPR. */
1368 if (ass_code != MAX_EXPR)
1369 return false;
1371 minmax = MIN_EXPR;
1372 if (operand_equal_for_phi_arg_p (op0, smaller)
1373 || (alt_smaller
1374 && operand_equal_for_phi_arg_p (op0, alt_smaller)))
1375 bound = op1;
1376 else if (operand_equal_for_phi_arg_p (op1, smaller)
1377 || (alt_smaller
1378 && operand_equal_for_phi_arg_p (op1, alt_smaller)))
1379 bound = op0;
1380 else
1381 return false;
1383 /* We need BOUND <= LARGER. */
1384 if (!integer_nonzerop (fold_build2 (LE_EXPR, boolean_type_node,
1385 bound, larger)))
1386 return false;
1388 else if (operand_equal_for_phi_arg_p (arg_false, smaller)
1389 || (alt_smaller
1390 && operand_equal_for_phi_arg_p (arg_false, alt_smaller)))
1392 /* Case
1394 if (smaller < larger)
1396 r' = MIN_EXPR (larger, bound)
1398 r = PHI <r', smaller> --> to be turned to MAX_EXPR. */
1399 if (ass_code != MIN_EXPR)
1400 return false;
1402 minmax = MAX_EXPR;
1403 if (operand_equal_for_phi_arg_p (op0, larger)
1404 || (alt_larger
1405 && operand_equal_for_phi_arg_p (op0, alt_larger)))
1406 bound = op1;
1407 else if (operand_equal_for_phi_arg_p (op1, larger)
1408 || (alt_larger
1409 && operand_equal_for_phi_arg_p (op1, alt_larger)))
1410 bound = op0;
1411 else
1412 return false;
1414 /* We need BOUND >= SMALLER. */
1415 if (!integer_nonzerop (fold_build2 (GE_EXPR, boolean_type_node,
1416 bound, smaller)))
1417 return false;
1419 else
1420 return false;
1422 else
1424 /* We got here if the condition is false, i.e., SMALLER > LARGER. */
1425 if (!operand_equal_for_phi_arg_p (lhs, arg_false))
1426 return false;
1428 if (operand_equal_for_phi_arg_p (arg_true, larger)
1429 || (alt_larger
1430 && operand_equal_for_phi_arg_p (arg_true, alt_larger)))
1432 /* Case
1434 if (smaller > larger)
1436 r' = MIN_EXPR (smaller, bound)
1438 r = PHI <r', larger> --> to be turned to MAX_EXPR. */
1439 if (ass_code != MIN_EXPR)
1440 return false;
1442 minmax = MAX_EXPR;
1443 if (operand_equal_for_phi_arg_p (op0, smaller)
1444 || (alt_smaller
1445 && operand_equal_for_phi_arg_p (op0, alt_smaller)))
1446 bound = op1;
1447 else if (operand_equal_for_phi_arg_p (op1, smaller)
1448 || (alt_smaller
1449 && operand_equal_for_phi_arg_p (op1, alt_smaller)))
1450 bound = op0;
1451 else
1452 return false;
1454 /* We need BOUND >= LARGER. */
1455 if (!integer_nonzerop (fold_build2 (GE_EXPR, boolean_type_node,
1456 bound, larger)))
1457 return false;
1459 else if (operand_equal_for_phi_arg_p (arg_true, smaller)
1460 || (alt_smaller
1461 && operand_equal_for_phi_arg_p (arg_true, alt_smaller)))
1463 /* Case
1465 if (smaller > larger)
1467 r' = MAX_EXPR (larger, bound)
1469 r = PHI <r', smaller> --> to be turned to MIN_EXPR. */
1470 if (ass_code != MAX_EXPR)
1471 return false;
1473 minmax = MIN_EXPR;
1474 if (operand_equal_for_phi_arg_p (op0, larger))
1475 bound = op1;
1476 else if (operand_equal_for_phi_arg_p (op1, larger))
1477 bound = op0;
1478 else
1479 return false;
1481 /* We need BOUND <= SMALLER. */
1482 if (!integer_nonzerop (fold_build2 (LE_EXPR, boolean_type_node,
1483 bound, smaller)))
1484 return false;
1486 else
1487 return false;
1490 /* Move the statement from the middle block. */
1491 gsi = gsi_last_bb (cond_bb);
1492 gsi_from = gsi_last_nondebug_bb (middle_bb);
1493 gsi_move_before (&gsi_from, &gsi);
1496 /* Create an SSA var to hold the min/max result. If we're the only
1497 things setting the target PHI, then we can clone the PHI
1498 variable. Otherwise we must create a new one. */
1499 result = PHI_RESULT (phi);
1500 if (EDGE_COUNT (gimple_bb (phi)->preds) == 2)
1501 result = duplicate_ssa_name (result, NULL);
1502 else
1503 result = make_ssa_name (TREE_TYPE (result));
1505 /* Emit the statement to compute min/max. */
1506 new_stmt = gimple_build_assign (result, minmax, arg0, arg1);
1507 gsi = gsi_last_bb (cond_bb);
1508 gsi_insert_before (&gsi, new_stmt, GSI_NEW_STMT);
1510 replace_phi_edge_with_variable (cond_bb, e1, phi, result);
1511 reset_flow_sensitive_info_in_bb (cond_bb);
1513 return true;
1516 /* The function absolute_replacement does the main work of doing the absolute
1517 replacement. Return true if the replacement is done. Otherwise return
1518 false.
1519 bb is the basic block where the replacement is going to be done on. arg0
1520 is argument 0 from the phi. Likewise for arg1. */
1522 static bool
1523 abs_replacement (basic_block cond_bb, basic_block middle_bb,
1524 edge e0 ATTRIBUTE_UNUSED, edge e1,
1525 gimple *phi, tree arg0, tree arg1)
1527 tree result;
1528 gassign *new_stmt;
1529 gimple *cond;
1530 gimple_stmt_iterator gsi;
1531 edge true_edge, false_edge;
1532 gimple *assign;
1533 edge e;
1534 tree rhs, lhs;
1535 bool negate;
1536 enum tree_code cond_code;
1538 /* If the type says honor signed zeros we cannot do this
1539 optimization. */
1540 if (HONOR_SIGNED_ZEROS (arg1))
1541 return false;
1543 /* OTHER_BLOCK must have only one executable statement which must have the
1544 form arg0 = -arg1 or arg1 = -arg0. */
1546 assign = last_and_only_stmt (middle_bb);
1547 /* If we did not find the proper negation assignment, then we can not
1548 optimize. */
1549 if (assign == NULL)
1550 return false;
1552 /* If we got here, then we have found the only executable statement
1553 in OTHER_BLOCK. If it is anything other than arg = -arg1 or
1554 arg1 = -arg0, then we can not optimize. */
1555 if (gimple_code (assign) != GIMPLE_ASSIGN)
1556 return false;
1558 lhs = gimple_assign_lhs (assign);
1560 if (gimple_assign_rhs_code (assign) != NEGATE_EXPR)
1561 return false;
1563 rhs = gimple_assign_rhs1 (assign);
1565 /* The assignment has to be arg0 = -arg1 or arg1 = -arg0. */
1566 if (!(lhs == arg0 && rhs == arg1)
1567 && !(lhs == arg1 && rhs == arg0))
1568 return false;
1570 cond = last_stmt (cond_bb);
1571 result = PHI_RESULT (phi);
1573 /* Only relationals comparing arg[01] against zero are interesting. */
1574 cond_code = gimple_cond_code (cond);
1575 if (cond_code != GT_EXPR && cond_code != GE_EXPR
1576 && cond_code != LT_EXPR && cond_code != LE_EXPR)
1577 return false;
1579 /* Make sure the conditional is arg[01] OP y. */
1580 if (gimple_cond_lhs (cond) != rhs)
1581 return false;
1583 if (FLOAT_TYPE_P (TREE_TYPE (gimple_cond_rhs (cond)))
1584 ? real_zerop (gimple_cond_rhs (cond))
1585 : integer_zerop (gimple_cond_rhs (cond)))
1587 else
1588 return false;
1590 /* We need to know which is the true edge and which is the false
1591 edge so that we know if have abs or negative abs. */
1592 extract_true_false_edges_from_block (cond_bb, &true_edge, &false_edge);
1594 /* For GT_EXPR/GE_EXPR, if the true edge goes to OTHER_BLOCK, then we
1595 will need to negate the result. Similarly for LT_EXPR/LE_EXPR if
1596 the false edge goes to OTHER_BLOCK. */
1597 if (cond_code == GT_EXPR || cond_code == GE_EXPR)
1598 e = true_edge;
1599 else
1600 e = false_edge;
1602 if (e->dest == middle_bb)
1603 negate = true;
1604 else
1605 negate = false;
1607 /* If the code negates only iff positive then make sure to not
1608 introduce undefined behavior when negating or computing the absolute.
1609 ??? We could use range info if present to check for arg1 == INT_MIN. */
1610 if (negate
1611 && (ANY_INTEGRAL_TYPE_P (TREE_TYPE (arg1))
1612 && ! TYPE_OVERFLOW_WRAPS (TREE_TYPE (arg1))))
1613 return false;
1615 result = duplicate_ssa_name (result, NULL);
1617 if (negate)
1618 lhs = make_ssa_name (TREE_TYPE (result));
1619 else
1620 lhs = result;
1622 /* Build the modify expression with abs expression. */
1623 new_stmt = gimple_build_assign (lhs, ABS_EXPR, rhs);
1625 gsi = gsi_last_bb (cond_bb);
1626 gsi_insert_before (&gsi, new_stmt, GSI_NEW_STMT);
1628 if (negate)
1630 /* Get the right GSI. We want to insert after the recently
1631 added ABS_EXPR statement (which we know is the first statement
1632 in the block. */
1633 new_stmt = gimple_build_assign (result, NEGATE_EXPR, lhs);
1635 gsi_insert_after (&gsi, new_stmt, GSI_NEW_STMT);
1638 replace_phi_edge_with_variable (cond_bb, e1, phi, result);
1639 reset_flow_sensitive_info_in_bb (cond_bb);
1641 /* Note that we optimized this PHI. */
1642 return true;
1645 /* Auxiliary functions to determine the set of memory accesses which
1646 can't trap because they are preceded by accesses to the same memory
1647 portion. We do that for MEM_REFs, so we only need to track
1648 the SSA_NAME of the pointer indirectly referenced. The algorithm
1649 simply is a walk over all instructions in dominator order. When
1650 we see an MEM_REF we determine if we've already seen a same
1651 ref anywhere up to the root of the dominator tree. If we do the
1652 current access can't trap. If we don't see any dominating access
1653 the current access might trap, but might also make later accesses
1654 non-trapping, so we remember it. We need to be careful with loads
1655 or stores, for instance a load might not trap, while a store would,
1656 so if we see a dominating read access this doesn't mean that a later
1657 write access would not trap. Hence we also need to differentiate the
1658 type of access(es) seen.
1660 ??? We currently are very conservative and assume that a load might
1661 trap even if a store doesn't (write-only memory). This probably is
1662 overly conservative. */
1664 /* A hash-table of SSA_NAMEs, and in which basic block an MEM_REF
1665 through it was seen, which would constitute a no-trap region for
1666 same accesses. */
1667 struct name_to_bb
1669 unsigned int ssa_name_ver;
1670 unsigned int phase;
1671 bool store;
1672 HOST_WIDE_INT offset, size;
1673 basic_block bb;
1676 /* Hashtable helpers. */
1678 struct ssa_names_hasher : free_ptr_hash <name_to_bb>
1680 static inline hashval_t hash (const name_to_bb *);
1681 static inline bool equal (const name_to_bb *, const name_to_bb *);
1684 /* Used for quick clearing of the hash-table when we see calls.
1685 Hash entries with phase < nt_call_phase are invalid. */
1686 static unsigned int nt_call_phase;
1688 /* The hash function. */
1690 inline hashval_t
1691 ssa_names_hasher::hash (const name_to_bb *n)
1693 return n->ssa_name_ver ^ (((hashval_t) n->store) << 31)
1694 ^ (n->offset << 6) ^ (n->size << 3);
1697 /* The equality function of *P1 and *P2. */
1699 inline bool
1700 ssa_names_hasher::equal (const name_to_bb *n1, const name_to_bb *n2)
1702 return n1->ssa_name_ver == n2->ssa_name_ver
1703 && n1->store == n2->store
1704 && n1->offset == n2->offset
1705 && n1->size == n2->size;
1708 class nontrapping_dom_walker : public dom_walker
1710 public:
1711 nontrapping_dom_walker (cdi_direction direction, hash_set<tree> *ps)
1712 : dom_walker (direction), m_nontrapping (ps), m_seen_ssa_names (128) {}
1714 virtual edge before_dom_children (basic_block);
1715 virtual void after_dom_children (basic_block);
1717 private:
1719 /* We see the expression EXP in basic block BB. If it's an interesting
1720 expression (an MEM_REF through an SSA_NAME) possibly insert the
1721 expression into the set NONTRAP or the hash table of seen expressions.
1722 STORE is true if this expression is on the LHS, otherwise it's on
1723 the RHS. */
1724 void add_or_mark_expr (basic_block, tree, bool);
1726 hash_set<tree> *m_nontrapping;
1728 /* The hash table for remembering what we've seen. */
1729 hash_table<ssa_names_hasher> m_seen_ssa_names;
1732 /* Called by walk_dominator_tree, when entering the block BB. */
1733 edge
1734 nontrapping_dom_walker::before_dom_children (basic_block bb)
1736 edge e;
1737 edge_iterator ei;
1738 gimple_stmt_iterator gsi;
1740 /* If we haven't seen all our predecessors, clear the hash-table. */
1741 FOR_EACH_EDGE (e, ei, bb->preds)
1742 if ((((size_t)e->src->aux) & 2) == 0)
1744 nt_call_phase++;
1745 break;
1748 /* Mark this BB as being on the path to dominator root and as visited. */
1749 bb->aux = (void*)(1 | 2);
1751 /* And walk the statements in order. */
1752 for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
1754 gimple *stmt = gsi_stmt (gsi);
1756 if ((gimple_code (stmt) == GIMPLE_ASM && gimple_vdef (stmt))
1757 || (is_gimple_call (stmt)
1758 && (!nonfreeing_call_p (stmt) || !nonbarrier_call_p (stmt))))
1759 nt_call_phase++;
1760 else if (gimple_assign_single_p (stmt) && !gimple_has_volatile_ops (stmt))
1762 add_or_mark_expr (bb, gimple_assign_lhs (stmt), true);
1763 add_or_mark_expr (bb, gimple_assign_rhs1 (stmt), false);
1766 return NULL;
1769 /* Called by walk_dominator_tree, when basic block BB is exited. */
1770 void
1771 nontrapping_dom_walker::after_dom_children (basic_block bb)
1773 /* This BB isn't on the path to dominator root anymore. */
1774 bb->aux = (void*)2;
1777 /* We see the expression EXP in basic block BB. If it's an interesting
1778 expression (an MEM_REF through an SSA_NAME) possibly insert the
1779 expression into the set NONTRAP or the hash table of seen expressions.
1780 STORE is true if this expression is on the LHS, otherwise it's on
1781 the RHS. */
1782 void
1783 nontrapping_dom_walker::add_or_mark_expr (basic_block bb, tree exp, bool store)
1785 HOST_WIDE_INT size;
1787 if (TREE_CODE (exp) == MEM_REF
1788 && TREE_CODE (TREE_OPERAND (exp, 0)) == SSA_NAME
1789 && tree_fits_shwi_p (TREE_OPERAND (exp, 1))
1790 && (size = int_size_in_bytes (TREE_TYPE (exp))) > 0)
1792 tree name = TREE_OPERAND (exp, 0);
1793 struct name_to_bb map;
1794 name_to_bb **slot;
1795 struct name_to_bb *n2bb;
1796 basic_block found_bb = 0;
1798 /* Try to find the last seen MEM_REF through the same
1799 SSA_NAME, which can trap. */
1800 map.ssa_name_ver = SSA_NAME_VERSION (name);
1801 map.phase = 0;
1802 map.bb = 0;
1803 map.store = store;
1804 map.offset = tree_to_shwi (TREE_OPERAND (exp, 1));
1805 map.size = size;
1807 slot = m_seen_ssa_names.find_slot (&map, INSERT);
1808 n2bb = *slot;
1809 if (n2bb && n2bb->phase >= nt_call_phase)
1810 found_bb = n2bb->bb;
1812 /* If we've found a trapping MEM_REF, _and_ it dominates EXP
1813 (it's in a basic block on the path from us to the dominator root)
1814 then we can't trap. */
1815 if (found_bb && (((size_t)found_bb->aux) & 1) == 1)
1817 m_nontrapping->add (exp);
1819 else
1821 /* EXP might trap, so insert it into the hash table. */
1822 if (n2bb)
1824 n2bb->phase = nt_call_phase;
1825 n2bb->bb = bb;
1827 else
1829 n2bb = XNEW (struct name_to_bb);
1830 n2bb->ssa_name_ver = SSA_NAME_VERSION (name);
1831 n2bb->phase = nt_call_phase;
1832 n2bb->bb = bb;
1833 n2bb->store = store;
1834 n2bb->offset = map.offset;
1835 n2bb->size = size;
1836 *slot = n2bb;
1842 /* This is the entry point of gathering non trapping memory accesses.
1843 It will do a dominator walk over the whole function, and it will
1844 make use of the bb->aux pointers. It returns a set of trees
1845 (the MEM_REFs itself) which can't trap. */
1846 static hash_set<tree> *
1847 get_non_trapping (void)
1849 nt_call_phase = 0;
1850 hash_set<tree> *nontrap = new hash_set<tree>;
1851 /* We're going to do a dominator walk, so ensure that we have
1852 dominance information. */
1853 calculate_dominance_info (CDI_DOMINATORS);
1855 nontrapping_dom_walker (CDI_DOMINATORS, nontrap)
1856 .walk (cfun->cfg->x_entry_block_ptr);
1858 clear_aux_for_blocks ();
1859 return nontrap;
1862 /* Do the main work of conditional store replacement. We already know
1863 that the recognized pattern looks like so:
1865 split:
1866 if (cond) goto MIDDLE_BB; else goto JOIN_BB (edge E1)
1867 MIDDLE_BB:
1868 something
1869 fallthrough (edge E0)
1870 JOIN_BB:
1871 some more
1873 We check that MIDDLE_BB contains only one store, that that store
1874 doesn't trap (not via NOTRAP, but via checking if an access to the same
1875 memory location dominates us) and that the store has a "simple" RHS. */
1877 static bool
1878 cond_store_replacement (basic_block middle_bb, basic_block join_bb,
1879 edge e0, edge e1, hash_set<tree> *nontrap)
1881 gimple *assign = last_and_only_stmt (middle_bb);
1882 tree lhs, rhs, name, name2;
1883 gphi *newphi;
1884 gassign *new_stmt;
1885 gimple_stmt_iterator gsi;
1886 source_location locus;
1888 /* Check if middle_bb contains of only one store. */
1889 if (!assign
1890 || !gimple_assign_single_p (assign)
1891 || gimple_has_volatile_ops (assign))
1892 return false;
1894 locus = gimple_location (assign);
1895 lhs = gimple_assign_lhs (assign);
1896 rhs = gimple_assign_rhs1 (assign);
1897 if (TREE_CODE (lhs) != MEM_REF
1898 || TREE_CODE (TREE_OPERAND (lhs, 0)) != SSA_NAME
1899 || !is_gimple_reg_type (TREE_TYPE (lhs)))
1900 return false;
1902 /* Prove that we can move the store down. We could also check
1903 TREE_THIS_NOTRAP here, but in that case we also could move stores,
1904 whose value is not available readily, which we want to avoid. */
1905 if (!nontrap->contains (lhs))
1906 return false;
1908 /* Now we've checked the constraints, so do the transformation:
1909 1) Remove the single store. */
1910 gsi = gsi_for_stmt (assign);
1911 unlink_stmt_vdef (assign);
1912 gsi_remove (&gsi, true);
1913 release_defs (assign);
1915 /* Make both store and load use alias-set zero as we have to
1916 deal with the case of the store being a conditional change
1917 of the dynamic type. */
1918 lhs = unshare_expr (lhs);
1919 tree *basep = &lhs;
1920 while (handled_component_p (*basep))
1921 basep = &TREE_OPERAND (*basep, 0);
1922 if (TREE_CODE (*basep) == MEM_REF
1923 || TREE_CODE (*basep) == TARGET_MEM_REF)
1924 TREE_OPERAND (*basep, 1)
1925 = fold_convert (ptr_type_node, TREE_OPERAND (*basep, 1));
1926 else
1927 *basep = build2 (MEM_REF, TREE_TYPE (*basep),
1928 build_fold_addr_expr (*basep),
1929 build_zero_cst (ptr_type_node));
1931 /* 2) Insert a load from the memory of the store to the temporary
1932 on the edge which did not contain the store. */
1933 name = make_temp_ssa_name (TREE_TYPE (lhs), NULL, "cstore");
1934 new_stmt = gimple_build_assign (name, lhs);
1935 gimple_set_location (new_stmt, locus);
1936 gsi_insert_on_edge (e1, new_stmt);
1938 /* 3) Create a PHI node at the join block, with one argument
1939 holding the old RHS, and the other holding the temporary
1940 where we stored the old memory contents. */
1941 name2 = make_temp_ssa_name (TREE_TYPE (lhs), NULL, "cstore");
1942 newphi = create_phi_node (name2, join_bb);
1943 add_phi_arg (newphi, rhs, e0, locus);
1944 add_phi_arg (newphi, name, e1, locus);
1946 lhs = unshare_expr (lhs);
1947 new_stmt = gimple_build_assign (lhs, PHI_RESULT (newphi));
1949 /* 4) Insert that PHI node. */
1950 gsi = gsi_after_labels (join_bb);
1951 if (gsi_end_p (gsi))
1953 gsi = gsi_last_bb (join_bb);
1954 gsi_insert_after (&gsi, new_stmt, GSI_NEW_STMT);
1956 else
1957 gsi_insert_before (&gsi, new_stmt, GSI_NEW_STMT);
1959 return true;
1962 /* Do the main work of conditional store replacement. */
1964 static bool
1965 cond_if_else_store_replacement_1 (basic_block then_bb, basic_block else_bb,
1966 basic_block join_bb, gimple *then_assign,
1967 gimple *else_assign)
1969 tree lhs_base, lhs, then_rhs, else_rhs, name;
1970 source_location then_locus, else_locus;
1971 gimple_stmt_iterator gsi;
1972 gphi *newphi;
1973 gassign *new_stmt;
1975 if (then_assign == NULL
1976 || !gimple_assign_single_p (then_assign)
1977 || gimple_clobber_p (then_assign)
1978 || gimple_has_volatile_ops (then_assign)
1979 || else_assign == NULL
1980 || !gimple_assign_single_p (else_assign)
1981 || gimple_clobber_p (else_assign)
1982 || gimple_has_volatile_ops (else_assign))
1983 return false;
1985 lhs = gimple_assign_lhs (then_assign);
1986 if (!is_gimple_reg_type (TREE_TYPE (lhs))
1987 || !operand_equal_p (lhs, gimple_assign_lhs (else_assign), 0))
1988 return false;
1990 lhs_base = get_base_address (lhs);
1991 if (lhs_base == NULL_TREE
1992 || (!DECL_P (lhs_base) && TREE_CODE (lhs_base) != MEM_REF))
1993 return false;
1995 then_rhs = gimple_assign_rhs1 (then_assign);
1996 else_rhs = gimple_assign_rhs1 (else_assign);
1997 then_locus = gimple_location (then_assign);
1998 else_locus = gimple_location (else_assign);
2000 /* Now we've checked the constraints, so do the transformation:
2001 1) Remove the stores. */
2002 gsi = gsi_for_stmt (then_assign);
2003 unlink_stmt_vdef (then_assign);
2004 gsi_remove (&gsi, true);
2005 release_defs (then_assign);
2007 gsi = gsi_for_stmt (else_assign);
2008 unlink_stmt_vdef (else_assign);
2009 gsi_remove (&gsi, true);
2010 release_defs (else_assign);
2012 /* 2) Create a PHI node at the join block, with one argument
2013 holding the old RHS, and the other holding the temporary
2014 where we stored the old memory contents. */
2015 name = make_temp_ssa_name (TREE_TYPE (lhs), NULL, "cstore");
2016 newphi = create_phi_node (name, join_bb);
2017 add_phi_arg (newphi, then_rhs, EDGE_SUCC (then_bb, 0), then_locus);
2018 add_phi_arg (newphi, else_rhs, EDGE_SUCC (else_bb, 0), else_locus);
2020 new_stmt = gimple_build_assign (lhs, PHI_RESULT (newphi));
2022 /* 3) Insert that PHI node. */
2023 gsi = gsi_after_labels (join_bb);
2024 if (gsi_end_p (gsi))
2026 gsi = gsi_last_bb (join_bb);
2027 gsi_insert_after (&gsi, new_stmt, GSI_NEW_STMT);
2029 else
2030 gsi_insert_before (&gsi, new_stmt, GSI_NEW_STMT);
2032 return true;
2035 /* Conditional store replacement. We already know
2036 that the recognized pattern looks like so:
2038 split:
2039 if (cond) goto THEN_BB; else goto ELSE_BB (edge E1)
2040 THEN_BB:
2042 X = Y;
2044 goto JOIN_BB;
2045 ELSE_BB:
2047 X = Z;
2049 fallthrough (edge E0)
2050 JOIN_BB:
2051 some more
2053 We check that it is safe to sink the store to JOIN_BB by verifying that
2054 there are no read-after-write or write-after-write dependencies in
2055 THEN_BB and ELSE_BB. */
2057 static bool
2058 cond_if_else_store_replacement (basic_block then_bb, basic_block else_bb,
2059 basic_block join_bb)
2061 gimple *then_assign = last_and_only_stmt (then_bb);
2062 gimple *else_assign = last_and_only_stmt (else_bb);
2063 vec<data_reference_p> then_datarefs, else_datarefs;
2064 vec<ddr_p> then_ddrs, else_ddrs;
2065 gimple *then_store, *else_store;
2066 bool found, ok = false, res;
2067 struct data_dependence_relation *ddr;
2068 data_reference_p then_dr, else_dr;
2069 int i, j;
2070 tree then_lhs, else_lhs;
2071 basic_block blocks[3];
2073 if (MAX_STORES_TO_SINK == 0)
2074 return false;
2076 /* Handle the case with single statement in THEN_BB and ELSE_BB. */
2077 if (then_assign && else_assign)
2078 return cond_if_else_store_replacement_1 (then_bb, else_bb, join_bb,
2079 then_assign, else_assign);
2081 /* Find data references. */
2082 then_datarefs.create (1);
2083 else_datarefs.create (1);
2084 if ((find_data_references_in_bb (NULL, then_bb, &then_datarefs)
2085 == chrec_dont_know)
2086 || !then_datarefs.length ()
2087 || (find_data_references_in_bb (NULL, else_bb, &else_datarefs)
2088 == chrec_dont_know)
2089 || !else_datarefs.length ())
2091 free_data_refs (then_datarefs);
2092 free_data_refs (else_datarefs);
2093 return false;
2096 /* Find pairs of stores with equal LHS. */
2097 auto_vec<gimple *, 1> then_stores, else_stores;
2098 FOR_EACH_VEC_ELT (then_datarefs, i, then_dr)
2100 if (DR_IS_READ (then_dr))
2101 continue;
2103 then_store = DR_STMT (then_dr);
2104 then_lhs = gimple_get_lhs (then_store);
2105 if (then_lhs == NULL_TREE)
2106 continue;
2107 found = false;
2109 FOR_EACH_VEC_ELT (else_datarefs, j, else_dr)
2111 if (DR_IS_READ (else_dr))
2112 continue;
2114 else_store = DR_STMT (else_dr);
2115 else_lhs = gimple_get_lhs (else_store);
2116 if (else_lhs == NULL_TREE)
2117 continue;
2119 if (operand_equal_p (then_lhs, else_lhs, 0))
2121 found = true;
2122 break;
2126 if (!found)
2127 continue;
2129 then_stores.safe_push (then_store);
2130 else_stores.safe_push (else_store);
2133 /* No pairs of stores found. */
2134 if (!then_stores.length ()
2135 || then_stores.length () > (unsigned) MAX_STORES_TO_SINK)
2137 free_data_refs (then_datarefs);
2138 free_data_refs (else_datarefs);
2139 return false;
2142 /* Compute and check data dependencies in both basic blocks. */
2143 then_ddrs.create (1);
2144 else_ddrs.create (1);
2145 if (!compute_all_dependences (then_datarefs, &then_ddrs,
2146 vNULL, false)
2147 || !compute_all_dependences (else_datarefs, &else_ddrs,
2148 vNULL, false))
2150 free_dependence_relations (then_ddrs);
2151 free_dependence_relations (else_ddrs);
2152 free_data_refs (then_datarefs);
2153 free_data_refs (else_datarefs);
2154 return false;
2156 blocks[0] = then_bb;
2157 blocks[1] = else_bb;
2158 blocks[2] = join_bb;
2159 renumber_gimple_stmt_uids_in_blocks (blocks, 3);
2161 /* Check that there are no read-after-write or write-after-write dependencies
2162 in THEN_BB. */
2163 FOR_EACH_VEC_ELT (then_ddrs, i, ddr)
2165 struct data_reference *dra = DDR_A (ddr);
2166 struct data_reference *drb = DDR_B (ddr);
2168 if (DDR_ARE_DEPENDENT (ddr) != chrec_known
2169 && ((DR_IS_READ (dra) && DR_IS_WRITE (drb)
2170 && gimple_uid (DR_STMT (dra)) > gimple_uid (DR_STMT (drb)))
2171 || (DR_IS_READ (drb) && DR_IS_WRITE (dra)
2172 && gimple_uid (DR_STMT (drb)) > gimple_uid (DR_STMT (dra)))
2173 || (DR_IS_WRITE (dra) && DR_IS_WRITE (drb))))
2175 free_dependence_relations (then_ddrs);
2176 free_dependence_relations (else_ddrs);
2177 free_data_refs (then_datarefs);
2178 free_data_refs (else_datarefs);
2179 return false;
2183 /* Check that there are no read-after-write or write-after-write dependencies
2184 in ELSE_BB. */
2185 FOR_EACH_VEC_ELT (else_ddrs, i, ddr)
2187 struct data_reference *dra = DDR_A (ddr);
2188 struct data_reference *drb = DDR_B (ddr);
2190 if (DDR_ARE_DEPENDENT (ddr) != chrec_known
2191 && ((DR_IS_READ (dra) && DR_IS_WRITE (drb)
2192 && gimple_uid (DR_STMT (dra)) > gimple_uid (DR_STMT (drb)))
2193 || (DR_IS_READ (drb) && DR_IS_WRITE (dra)
2194 && gimple_uid (DR_STMT (drb)) > gimple_uid (DR_STMT (dra)))
2195 || (DR_IS_WRITE (dra) && DR_IS_WRITE (drb))))
2197 free_dependence_relations (then_ddrs);
2198 free_dependence_relations (else_ddrs);
2199 free_data_refs (then_datarefs);
2200 free_data_refs (else_datarefs);
2201 return false;
2205 /* Sink stores with same LHS. */
2206 FOR_EACH_VEC_ELT (then_stores, i, then_store)
2208 else_store = else_stores[i];
2209 res = cond_if_else_store_replacement_1 (then_bb, else_bb, join_bb,
2210 then_store, else_store);
2211 ok = ok || res;
2214 free_dependence_relations (then_ddrs);
2215 free_dependence_relations (else_ddrs);
2216 free_data_refs (then_datarefs);
2217 free_data_refs (else_datarefs);
2219 return ok;
2222 /* Return TRUE if STMT has a VUSE whose corresponding VDEF is in BB. */
2224 static bool
2225 local_mem_dependence (gimple *stmt, basic_block bb)
2227 tree vuse = gimple_vuse (stmt);
2228 gimple *def;
2230 if (!vuse)
2231 return false;
2233 def = SSA_NAME_DEF_STMT (vuse);
2234 return (def && gimple_bb (def) == bb);
2237 /* Given a "diamond" control-flow pattern where BB0 tests a condition,
2238 BB1 and BB2 are "then" and "else" blocks dependent on this test,
2239 and BB3 rejoins control flow following BB1 and BB2, look for
2240 opportunities to hoist loads as follows. If BB3 contains a PHI of
2241 two loads, one each occurring in BB1 and BB2, and the loads are
2242 provably of adjacent fields in the same structure, then move both
2243 loads into BB0. Of course this can only be done if there are no
2244 dependencies preventing such motion.
2246 One of the hoisted loads will always be speculative, so the
2247 transformation is currently conservative:
2249 - The fields must be strictly adjacent.
2250 - The two fields must occupy a single memory block that is
2251 guaranteed to not cross a page boundary.
2253 The last is difficult to prove, as such memory blocks should be
2254 aligned on the minimum of the stack alignment boundary and the
2255 alignment guaranteed by heap allocation interfaces. Thus we rely
2256 on a parameter for the alignment value.
2258 Provided a good value is used for the last case, the first
2259 restriction could possibly be relaxed. */
2261 static void
2262 hoist_adjacent_loads (basic_block bb0, basic_block bb1,
2263 basic_block bb2, basic_block bb3)
2265 int param_align = PARAM_VALUE (PARAM_L1_CACHE_LINE_SIZE);
2266 unsigned param_align_bits = (unsigned) (param_align * BITS_PER_UNIT);
2267 gphi_iterator gsi;
2269 /* Walk the phis in bb3 looking for an opportunity. We are looking
2270 for phis of two SSA names, one each of which is defined in bb1 and
2271 bb2. */
2272 for (gsi = gsi_start_phis (bb3); !gsi_end_p (gsi); gsi_next (&gsi))
2274 gphi *phi_stmt = gsi.phi ();
2275 gimple *def1, *def2;
2276 tree arg1, arg2, ref1, ref2, field1, field2;
2277 tree tree_offset1, tree_offset2, tree_size2, next;
2278 int offset1, offset2, size2;
2279 unsigned align1;
2280 gimple_stmt_iterator gsi2;
2281 basic_block bb_for_def1, bb_for_def2;
2283 if (gimple_phi_num_args (phi_stmt) != 2
2284 || virtual_operand_p (gimple_phi_result (phi_stmt)))
2285 continue;
2287 arg1 = gimple_phi_arg_def (phi_stmt, 0);
2288 arg2 = gimple_phi_arg_def (phi_stmt, 1);
2290 if (TREE_CODE (arg1) != SSA_NAME
2291 || TREE_CODE (arg2) != SSA_NAME
2292 || SSA_NAME_IS_DEFAULT_DEF (arg1)
2293 || SSA_NAME_IS_DEFAULT_DEF (arg2))
2294 continue;
2296 def1 = SSA_NAME_DEF_STMT (arg1);
2297 def2 = SSA_NAME_DEF_STMT (arg2);
2299 if ((gimple_bb (def1) != bb1 || gimple_bb (def2) != bb2)
2300 && (gimple_bb (def2) != bb1 || gimple_bb (def1) != bb2))
2301 continue;
2303 /* Check the mode of the arguments to be sure a conditional move
2304 can be generated for it. */
2305 if (optab_handler (movcc_optab, TYPE_MODE (TREE_TYPE (arg1)))
2306 == CODE_FOR_nothing)
2307 continue;
2309 /* Both statements must be assignments whose RHS is a COMPONENT_REF. */
2310 if (!gimple_assign_single_p (def1)
2311 || !gimple_assign_single_p (def2)
2312 || gimple_has_volatile_ops (def1)
2313 || gimple_has_volatile_ops (def2))
2314 continue;
2316 ref1 = gimple_assign_rhs1 (def1);
2317 ref2 = gimple_assign_rhs1 (def2);
2319 if (TREE_CODE (ref1) != COMPONENT_REF
2320 || TREE_CODE (ref2) != COMPONENT_REF)
2321 continue;
2323 /* The zeroth operand of the two component references must be
2324 identical. It is not sufficient to compare get_base_address of
2325 the two references, because this could allow for different
2326 elements of the same array in the two trees. It is not safe to
2327 assume that the existence of one array element implies the
2328 existence of a different one. */
2329 if (!operand_equal_p (TREE_OPERAND (ref1, 0), TREE_OPERAND (ref2, 0), 0))
2330 continue;
2332 field1 = TREE_OPERAND (ref1, 1);
2333 field2 = TREE_OPERAND (ref2, 1);
2335 /* Check for field adjacency, and ensure field1 comes first. */
2336 for (next = DECL_CHAIN (field1);
2337 next && TREE_CODE (next) != FIELD_DECL;
2338 next = DECL_CHAIN (next))
2341 if (next != field2)
2343 for (next = DECL_CHAIN (field2);
2344 next && TREE_CODE (next) != FIELD_DECL;
2345 next = DECL_CHAIN (next))
2348 if (next != field1)
2349 continue;
2351 std::swap (field1, field2);
2352 std::swap (def1, def2);
2355 bb_for_def1 = gimple_bb (def1);
2356 bb_for_def2 = gimple_bb (def2);
2358 /* Check for proper alignment of the first field. */
2359 tree_offset1 = bit_position (field1);
2360 tree_offset2 = bit_position (field2);
2361 tree_size2 = DECL_SIZE (field2);
2363 if (!tree_fits_uhwi_p (tree_offset1)
2364 || !tree_fits_uhwi_p (tree_offset2)
2365 || !tree_fits_uhwi_p (tree_size2))
2366 continue;
2368 offset1 = tree_to_uhwi (tree_offset1);
2369 offset2 = tree_to_uhwi (tree_offset2);
2370 size2 = tree_to_uhwi (tree_size2);
2371 align1 = DECL_ALIGN (field1) % param_align_bits;
2373 if (offset1 % BITS_PER_UNIT != 0)
2374 continue;
2376 /* For profitability, the two field references should fit within
2377 a single cache line. */
2378 if (align1 + offset2 - offset1 + size2 > param_align_bits)
2379 continue;
2381 /* The two expressions cannot be dependent upon vdefs defined
2382 in bb1/bb2. */
2383 if (local_mem_dependence (def1, bb_for_def1)
2384 || local_mem_dependence (def2, bb_for_def2))
2385 continue;
2387 /* The conditions are satisfied; hoist the loads from bb1 and bb2 into
2388 bb0. We hoist the first one first so that a cache miss is handled
2389 efficiently regardless of hardware cache-fill policy. */
2390 gsi2 = gsi_for_stmt (def1);
2391 gsi_move_to_bb_end (&gsi2, bb0);
2392 gsi2 = gsi_for_stmt (def2);
2393 gsi_move_to_bb_end (&gsi2, bb0);
2395 if (dump_file && (dump_flags & TDF_DETAILS))
2397 fprintf (dump_file,
2398 "\nHoisting adjacent loads from %d and %d into %d: \n",
2399 bb_for_def1->index, bb_for_def2->index, bb0->index);
2400 print_gimple_stmt (dump_file, def1, 0, TDF_VOPS|TDF_MEMSYMS);
2401 print_gimple_stmt (dump_file, def2, 0, TDF_VOPS|TDF_MEMSYMS);
2406 /* Determine whether we should attempt to hoist adjacent loads out of
2407 diamond patterns in pass_phiopt. Always hoist loads if
2408 -fhoist-adjacent-loads is specified and the target machine has
2409 both a conditional move instruction and a defined cache line size. */
2411 static bool
2412 gate_hoist_loads (void)
2414 return (flag_hoist_adjacent_loads == 1
2415 && PARAM_VALUE (PARAM_L1_CACHE_LINE_SIZE)
2416 && HAVE_conditional_move);
2419 /* This pass tries to replaces an if-then-else block with an
2420 assignment. We have four kinds of transformations. Some of these
2421 transformations are also performed by the ifcvt RTL optimizer.
2423 Conditional Replacement
2424 -----------------------
2426 This transformation, implemented in conditional_replacement,
2427 replaces
2429 bb0:
2430 if (cond) goto bb2; else goto bb1;
2431 bb1:
2432 bb2:
2433 x = PHI <0 (bb1), 1 (bb0), ...>;
2435 with
2437 bb0:
2438 x' = cond;
2439 goto bb2;
2440 bb2:
2441 x = PHI <x' (bb0), ...>;
2443 We remove bb1 as it becomes unreachable. This occurs often due to
2444 gimplification of conditionals.
2446 Value Replacement
2447 -----------------
2449 This transformation, implemented in value_replacement, replaces
2451 bb0:
2452 if (a != b) goto bb2; else goto bb1;
2453 bb1:
2454 bb2:
2455 x = PHI <a (bb1), b (bb0), ...>;
2457 with
2459 bb0:
2460 bb2:
2461 x = PHI <b (bb0), ...>;
2463 This opportunity can sometimes occur as a result of other
2464 optimizations.
2467 Another case caught by value replacement looks like this:
2469 bb0:
2470 t1 = a == CONST;
2471 t2 = b > c;
2472 t3 = t1 & t2;
2473 if (t3 != 0) goto bb1; else goto bb2;
2474 bb1:
2475 bb2:
2476 x = PHI (CONST, a)
2478 Gets replaced with:
2479 bb0:
2480 bb2:
2481 t1 = a == CONST;
2482 t2 = b > c;
2483 t3 = t1 & t2;
2484 x = a;
2486 ABS Replacement
2487 ---------------
2489 This transformation, implemented in abs_replacement, replaces
2491 bb0:
2492 if (a >= 0) goto bb2; else goto bb1;
2493 bb1:
2494 x = -a;
2495 bb2:
2496 x = PHI <x (bb1), a (bb0), ...>;
2498 with
2500 bb0:
2501 x' = ABS_EXPR< a >;
2502 bb2:
2503 x = PHI <x' (bb0), ...>;
2505 MIN/MAX Replacement
2506 -------------------
2508 This transformation, minmax_replacement replaces
2510 bb0:
2511 if (a <= b) goto bb2; else goto bb1;
2512 bb1:
2513 bb2:
2514 x = PHI <b (bb1), a (bb0), ...>;
2516 with
2518 bb0:
2519 x' = MIN_EXPR (a, b)
2520 bb2:
2521 x = PHI <x' (bb0), ...>;
2523 A similar transformation is done for MAX_EXPR.
2526 This pass also performs a fifth transformation of a slightly different
2527 flavor.
2529 Factor conversion in COND_EXPR
2530 ------------------------------
2532 This transformation factors the conversion out of COND_EXPR with
2533 factor_out_conditional_conversion.
2535 For example:
2536 if (a <= CST) goto <bb 3>; else goto <bb 4>;
2537 <bb 3>:
2538 tmp = (int) a;
2539 <bb 4>:
2540 tmp = PHI <tmp, CST>
2542 Into:
2543 if (a <= CST) goto <bb 3>; else goto <bb 4>;
2544 <bb 3>:
2545 <bb 4>:
2546 a = PHI <a, CST>
2547 tmp = (int) a;
2549 Adjacent Load Hoisting
2550 ----------------------
2552 This transformation replaces
2554 bb0:
2555 if (...) goto bb2; else goto bb1;
2556 bb1:
2557 x1 = (<expr>).field1;
2558 goto bb3;
2559 bb2:
2560 x2 = (<expr>).field2;
2561 bb3:
2562 # x = PHI <x1, x2>;
2564 with
2566 bb0:
2567 x1 = (<expr>).field1;
2568 x2 = (<expr>).field2;
2569 if (...) goto bb2; else goto bb1;
2570 bb1:
2571 goto bb3;
2572 bb2:
2573 bb3:
2574 # x = PHI <x1, x2>;
2576 The purpose of this transformation is to enable generation of conditional
2577 move instructions such as Intel CMOVE or PowerPC ISEL. Because one of
2578 the loads is speculative, the transformation is restricted to very
2579 specific cases to avoid introducing a page fault. We are looking for
2580 the common idiom:
2582 if (...)
2583 x = y->left;
2584 else
2585 x = y->right;
2587 where left and right are typically adjacent pointers in a tree structure. */
2589 namespace {
2591 const pass_data pass_data_phiopt =
2593 GIMPLE_PASS, /* type */
2594 "phiopt", /* name */
2595 OPTGROUP_NONE, /* optinfo_flags */
2596 TV_TREE_PHIOPT, /* tv_id */
2597 ( PROP_cfg | PROP_ssa ), /* properties_required */
2598 0, /* properties_provided */
2599 0, /* properties_destroyed */
2600 0, /* todo_flags_start */
2601 0, /* todo_flags_finish */
2604 class pass_phiopt : public gimple_opt_pass
2606 public:
2607 pass_phiopt (gcc::context *ctxt)
2608 : gimple_opt_pass (pass_data_phiopt, ctxt)
2611 /* opt_pass methods: */
2612 opt_pass * clone () { return new pass_phiopt (m_ctxt); }
2613 virtual bool gate (function *) { return flag_ssa_phiopt; }
2614 virtual unsigned int execute (function *)
2616 return tree_ssa_phiopt_worker (false, gate_hoist_loads ());
2619 }; // class pass_phiopt
2621 } // anon namespace
2623 gimple_opt_pass *
2624 make_pass_phiopt (gcc::context *ctxt)
2626 return new pass_phiopt (ctxt);
2629 namespace {
2631 const pass_data pass_data_cselim =
2633 GIMPLE_PASS, /* type */
2634 "cselim", /* name */
2635 OPTGROUP_NONE, /* optinfo_flags */
2636 TV_TREE_PHIOPT, /* tv_id */
2637 ( PROP_cfg | PROP_ssa ), /* properties_required */
2638 0, /* properties_provided */
2639 0, /* properties_destroyed */
2640 0, /* todo_flags_start */
2641 0, /* todo_flags_finish */
2644 class pass_cselim : public gimple_opt_pass
2646 public:
2647 pass_cselim (gcc::context *ctxt)
2648 : gimple_opt_pass (pass_data_cselim, ctxt)
2651 /* opt_pass methods: */
2652 virtual bool gate (function *) { return flag_tree_cselim; }
2653 virtual unsigned int execute (function *) { return tree_ssa_cs_elim (); }
2655 }; // class pass_cselim
2657 } // anon namespace
2659 gimple_opt_pass *
2660 make_pass_cselim (gcc::context *ctxt)
2662 return new pass_cselim (ctxt);