PR c++/85952
[official-gcc.git] / gcc / cfgcleanup.c
blob4a5dc29d14fb7ab7abf70a89a7512b4c4edc1440
1 /* Control flow optimization code for GNU compiler.
2 Copyright (C) 1987-2018 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 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 /* This file contains optimizer of the control flow. The main entry point is
21 cleanup_cfg. Following optimizations are performed:
23 - Unreachable blocks removal
24 - Edge forwarding (edge to the forwarder block is forwarded to its
25 successor. Simplification of the branch instruction is performed by
26 underlying infrastructure so branch can be converted to simplejump or
27 eliminated).
28 - Cross jumping (tail merging)
29 - Conditional jump-around-simplejump simplification
30 - Basic block merging. */
32 #include "config.h"
33 #include "system.h"
34 #include "coretypes.h"
35 #include "backend.h"
36 #include "target.h"
37 #include "rtl.h"
38 #include "tree.h"
39 #include "cfghooks.h"
40 #include "df.h"
41 #include "memmodel.h"
42 #include "tm_p.h"
43 #include "insn-config.h"
44 #include "emit-rtl.h"
45 #include "cselib.h"
46 #include "params.h"
47 #include "tree-pass.h"
48 #include "cfgloop.h"
49 #include "cfgrtl.h"
50 #include "cfganal.h"
51 #include "cfgbuild.h"
52 #include "cfgcleanup.h"
53 #include "dce.h"
54 #include "dbgcnt.h"
55 #include "rtl-iter.h"
57 #define FORWARDER_BLOCK_P(BB) ((BB)->flags & BB_FORWARDER_BLOCK)
59 /* Set to true when we are running first pass of try_optimize_cfg loop. */
60 static bool first_pass;
62 /* Set to true if crossjumps occurred in the latest run of try_optimize_cfg. */
63 static bool crossjumps_occurred;
65 /* Set to true if we couldn't run an optimization due to stale liveness
66 information; we should run df_analyze to enable more opportunities. */
67 static bool block_was_dirty;
69 static bool try_crossjump_to_edge (int, edge, edge, enum replace_direction);
70 static bool try_crossjump_bb (int, basic_block);
71 static bool outgoing_edges_match (int, basic_block, basic_block);
72 static enum replace_direction old_insns_match_p (int, rtx_insn *, rtx_insn *);
74 static void merge_blocks_move_predecessor_nojumps (basic_block, basic_block);
75 static void merge_blocks_move_successor_nojumps (basic_block, basic_block);
76 static bool try_optimize_cfg (int);
77 static bool try_simplify_condjump (basic_block);
78 static bool try_forward_edges (int, basic_block);
79 static edge thread_jump (edge, basic_block);
80 static bool mark_effect (rtx, bitmap);
81 static void notice_new_block (basic_block);
82 static void update_forwarder_flag (basic_block);
83 static void merge_memattrs (rtx, rtx);
85 /* Set flags for newly created block. */
87 static void
88 notice_new_block (basic_block bb)
90 if (!bb)
91 return;
93 if (forwarder_block_p (bb))
94 bb->flags |= BB_FORWARDER_BLOCK;
97 /* Recompute forwarder flag after block has been modified. */
99 static void
100 update_forwarder_flag (basic_block bb)
102 if (forwarder_block_p (bb))
103 bb->flags |= BB_FORWARDER_BLOCK;
104 else
105 bb->flags &= ~BB_FORWARDER_BLOCK;
108 /* Simplify a conditional jump around an unconditional jump.
109 Return true if something changed. */
111 static bool
112 try_simplify_condjump (basic_block cbranch_block)
114 basic_block jump_block, jump_dest_block, cbranch_dest_block;
115 edge cbranch_jump_edge, cbranch_fallthru_edge;
116 rtx_insn *cbranch_insn;
118 /* Verify that there are exactly two successors. */
119 if (EDGE_COUNT (cbranch_block->succs) != 2)
120 return false;
122 /* Verify that we've got a normal conditional branch at the end
123 of the block. */
124 cbranch_insn = BB_END (cbranch_block);
125 if (!any_condjump_p (cbranch_insn))
126 return false;
128 cbranch_fallthru_edge = FALLTHRU_EDGE (cbranch_block);
129 cbranch_jump_edge = BRANCH_EDGE (cbranch_block);
131 /* The next block must not have multiple predecessors, must not
132 be the last block in the function, and must contain just the
133 unconditional jump. */
134 jump_block = cbranch_fallthru_edge->dest;
135 if (!single_pred_p (jump_block)
136 || jump_block->next_bb == EXIT_BLOCK_PTR_FOR_FN (cfun)
137 || !FORWARDER_BLOCK_P (jump_block))
138 return false;
139 jump_dest_block = single_succ (jump_block);
141 /* If we are partitioning hot/cold basic blocks, we don't want to
142 mess up unconditional or indirect jumps that cross between hot
143 and cold sections.
145 Basic block partitioning may result in some jumps that appear to
146 be optimizable (or blocks that appear to be mergeable), but which really
147 must be left untouched (they are required to make it safely across
148 partition boundaries). See the comments at the top of
149 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
151 if (BB_PARTITION (jump_block) != BB_PARTITION (jump_dest_block)
152 || (cbranch_jump_edge->flags & EDGE_CROSSING))
153 return false;
155 /* The conditional branch must target the block after the
156 unconditional branch. */
157 cbranch_dest_block = cbranch_jump_edge->dest;
159 if (cbranch_dest_block == EXIT_BLOCK_PTR_FOR_FN (cfun)
160 || jump_dest_block == EXIT_BLOCK_PTR_FOR_FN (cfun)
161 || !can_fallthru (jump_block, cbranch_dest_block))
162 return false;
164 /* Invert the conditional branch. */
165 if (!invert_jump (as_a <rtx_jump_insn *> (cbranch_insn),
166 block_label (jump_dest_block), 0))
167 return false;
169 if (dump_file)
170 fprintf (dump_file, "Simplifying condjump %i around jump %i\n",
171 INSN_UID (cbranch_insn), INSN_UID (BB_END (jump_block)));
173 /* Success. Update the CFG to match. Note that after this point
174 the edge variable names appear backwards; the redirection is done
175 this way to preserve edge profile data. */
176 cbranch_jump_edge = redirect_edge_succ_nodup (cbranch_jump_edge,
177 cbranch_dest_block);
178 cbranch_fallthru_edge = redirect_edge_succ_nodup (cbranch_fallthru_edge,
179 jump_dest_block);
180 cbranch_jump_edge->flags |= EDGE_FALLTHRU;
181 cbranch_fallthru_edge->flags &= ~EDGE_FALLTHRU;
182 update_br_prob_note (cbranch_block);
184 /* Delete the block with the unconditional jump, and clean up the mess. */
185 delete_basic_block (jump_block);
186 tidy_fallthru_edge (cbranch_jump_edge);
187 update_forwarder_flag (cbranch_block);
189 return true;
192 /* Attempt to prove that operation is NOOP using CSElib or mark the effect
193 on register. Used by jump threading. */
195 static bool
196 mark_effect (rtx exp, regset nonequal)
198 rtx dest;
199 switch (GET_CODE (exp))
201 /* In case we do clobber the register, mark it as equal, as we know the
202 value is dead so it don't have to match. */
203 case CLOBBER:
204 dest = XEXP (exp, 0);
205 if (REG_P (dest))
206 bitmap_clear_range (nonequal, REGNO (dest), REG_NREGS (dest));
207 return false;
209 case SET:
210 if (rtx_equal_for_cselib_p (SET_DEST (exp), SET_SRC (exp)))
211 return false;
212 dest = SET_DEST (exp);
213 if (dest == pc_rtx)
214 return false;
215 if (!REG_P (dest))
216 return true;
217 bitmap_set_range (nonequal, REGNO (dest), REG_NREGS (dest));
218 return false;
220 default:
221 return false;
225 /* Return true if X contains a register in NONEQUAL. */
226 static bool
227 mentions_nonequal_regs (const_rtx x, regset nonequal)
229 subrtx_iterator::array_type array;
230 FOR_EACH_SUBRTX (iter, array, x, NONCONST)
232 const_rtx x = *iter;
233 if (REG_P (x))
235 unsigned int end_regno = END_REGNO (x);
236 for (unsigned int regno = REGNO (x); regno < end_regno; ++regno)
237 if (REGNO_REG_SET_P (nonequal, regno))
238 return true;
241 return false;
244 /* Attempt to prove that the basic block B will have no side effects and
245 always continues in the same edge if reached via E. Return the edge
246 if exist, NULL otherwise. */
248 static edge
249 thread_jump (edge e, basic_block b)
251 rtx set1, set2, cond1, cond2;
252 rtx_insn *insn;
253 enum rtx_code code1, code2, reversed_code2;
254 bool reverse1 = false;
255 unsigned i;
256 regset nonequal;
257 bool failed = false;
258 reg_set_iterator rsi;
260 if (b->flags & BB_NONTHREADABLE_BLOCK)
261 return NULL;
263 /* At the moment, we do handle only conditional jumps, but later we may
264 want to extend this code to tablejumps and others. */
265 if (EDGE_COUNT (e->src->succs) != 2)
266 return NULL;
267 if (EDGE_COUNT (b->succs) != 2)
269 b->flags |= BB_NONTHREADABLE_BLOCK;
270 return NULL;
273 /* Second branch must end with onlyjump, as we will eliminate the jump. */
274 if (!any_condjump_p (BB_END (e->src)))
275 return NULL;
277 if (!any_condjump_p (BB_END (b)) || !onlyjump_p (BB_END (b)))
279 b->flags |= BB_NONTHREADABLE_BLOCK;
280 return NULL;
283 set1 = pc_set (BB_END (e->src));
284 set2 = pc_set (BB_END (b));
285 if (((e->flags & EDGE_FALLTHRU) != 0)
286 != (XEXP (SET_SRC (set1), 1) == pc_rtx))
287 reverse1 = true;
289 cond1 = XEXP (SET_SRC (set1), 0);
290 cond2 = XEXP (SET_SRC (set2), 0);
291 if (reverse1)
292 code1 = reversed_comparison_code (cond1, BB_END (e->src));
293 else
294 code1 = GET_CODE (cond1);
296 code2 = GET_CODE (cond2);
297 reversed_code2 = reversed_comparison_code (cond2, BB_END (b));
299 if (!comparison_dominates_p (code1, code2)
300 && !comparison_dominates_p (code1, reversed_code2))
301 return NULL;
303 /* Ensure that the comparison operators are equivalent.
304 ??? This is far too pessimistic. We should allow swapped operands,
305 different CCmodes, or for example comparisons for interval, that
306 dominate even when operands are not equivalent. */
307 if (!rtx_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
308 || !rtx_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
309 return NULL;
311 /* Short circuit cases where block B contains some side effects, as we can't
312 safely bypass it. */
313 for (insn = NEXT_INSN (BB_HEAD (b)); insn != NEXT_INSN (BB_END (b));
314 insn = NEXT_INSN (insn))
315 if (INSN_P (insn) && side_effects_p (PATTERN (insn)))
317 b->flags |= BB_NONTHREADABLE_BLOCK;
318 return NULL;
321 cselib_init (0);
323 /* First process all values computed in the source basic block. */
324 for (insn = NEXT_INSN (BB_HEAD (e->src));
325 insn != NEXT_INSN (BB_END (e->src));
326 insn = NEXT_INSN (insn))
327 if (INSN_P (insn))
328 cselib_process_insn (insn);
330 nonequal = BITMAP_ALLOC (NULL);
331 CLEAR_REG_SET (nonequal);
333 /* Now assume that we've continued by the edge E to B and continue
334 processing as if it were same basic block.
335 Our goal is to prove that whole block is an NOOP. */
337 for (insn = NEXT_INSN (BB_HEAD (b));
338 insn != NEXT_INSN (BB_END (b)) && !failed;
339 insn = NEXT_INSN (insn))
341 if (INSN_P (insn))
343 rtx pat = PATTERN (insn);
345 if (GET_CODE (pat) == PARALLEL)
347 for (i = 0; i < (unsigned)XVECLEN (pat, 0); i++)
348 failed |= mark_effect (XVECEXP (pat, 0, i), nonequal);
350 else
351 failed |= mark_effect (pat, nonequal);
354 cselib_process_insn (insn);
357 /* Later we should clear nonequal of dead registers. So far we don't
358 have life information in cfg_cleanup. */
359 if (failed)
361 b->flags |= BB_NONTHREADABLE_BLOCK;
362 goto failed_exit;
365 /* cond2 must not mention any register that is not equal to the
366 former block. */
367 if (mentions_nonequal_regs (cond2, nonequal))
368 goto failed_exit;
370 EXECUTE_IF_SET_IN_REG_SET (nonequal, 0, i, rsi)
371 goto failed_exit;
373 BITMAP_FREE (nonequal);
374 cselib_finish ();
375 if ((comparison_dominates_p (code1, code2) != 0)
376 != (XEXP (SET_SRC (set2), 1) == pc_rtx))
377 return BRANCH_EDGE (b);
378 else
379 return FALLTHRU_EDGE (b);
381 failed_exit:
382 BITMAP_FREE (nonequal);
383 cselib_finish ();
384 return NULL;
387 /* Attempt to forward edges leaving basic block B.
388 Return true if successful. */
390 static bool
391 try_forward_edges (int mode, basic_block b)
393 bool changed = false;
394 edge_iterator ei;
395 edge e, *threaded_edges = NULL;
397 for (ei = ei_start (b->succs); (e = ei_safe_edge (ei)); )
399 basic_block target, first;
400 location_t goto_locus;
401 int counter;
402 bool threaded = false;
403 int nthreaded_edges = 0;
404 bool may_thread = first_pass || (b->flags & BB_MODIFIED) != 0;
405 bool new_target_threaded = false;
407 /* Skip complex edges because we don't know how to update them.
409 Still handle fallthru edges, as we can succeed to forward fallthru
410 edge to the same place as the branch edge of conditional branch
411 and turn conditional branch to an unconditional branch. */
412 if (e->flags & EDGE_COMPLEX)
414 ei_next (&ei);
415 continue;
418 target = first = e->dest;
419 counter = NUM_FIXED_BLOCKS;
420 goto_locus = e->goto_locus;
422 while (counter < n_basic_blocks_for_fn (cfun))
424 basic_block new_target = NULL;
425 may_thread |= (target->flags & BB_MODIFIED) != 0;
427 if (FORWARDER_BLOCK_P (target)
428 && single_succ (target) != EXIT_BLOCK_PTR_FOR_FN (cfun))
430 /* Bypass trivial infinite loops. */
431 new_target = single_succ (target);
432 if (target == new_target)
433 counter = n_basic_blocks_for_fn (cfun);
434 else if (!optimize)
436 /* When not optimizing, ensure that edges or forwarder
437 blocks with different locus are not optimized out. */
438 location_t new_locus = single_succ_edge (target)->goto_locus;
439 location_t locus = goto_locus;
441 if (LOCATION_LOCUS (new_locus) != UNKNOWN_LOCATION
442 && LOCATION_LOCUS (locus) != UNKNOWN_LOCATION
443 && new_locus != locus)
444 new_target = NULL;
445 else
447 if (LOCATION_LOCUS (new_locus) != UNKNOWN_LOCATION)
448 locus = new_locus;
450 rtx_insn *last = BB_END (target);
451 if (DEBUG_INSN_P (last))
452 last = prev_nondebug_insn (last);
453 if (last && INSN_P (last))
454 new_locus = INSN_LOCATION (last);
455 else
456 new_locus = UNKNOWN_LOCATION;
458 if (LOCATION_LOCUS (new_locus) != UNKNOWN_LOCATION
459 && LOCATION_LOCUS (locus) != UNKNOWN_LOCATION
460 && new_locus != locus)
461 new_target = NULL;
462 else
464 if (LOCATION_LOCUS (new_locus) != UNKNOWN_LOCATION)
465 locus = new_locus;
467 goto_locus = locus;
473 /* Allow to thread only over one edge at time to simplify updating
474 of probabilities. */
475 else if ((mode & CLEANUP_THREADING) && may_thread)
477 edge t = thread_jump (e, target);
478 if (t)
480 if (!threaded_edges)
481 threaded_edges = XNEWVEC (edge,
482 n_basic_blocks_for_fn (cfun));
483 else
485 int i;
487 /* Detect an infinite loop across blocks not
488 including the start block. */
489 for (i = 0; i < nthreaded_edges; ++i)
490 if (threaded_edges[i] == t)
491 break;
492 if (i < nthreaded_edges)
494 counter = n_basic_blocks_for_fn (cfun);
495 break;
499 /* Detect an infinite loop across the start block. */
500 if (t->dest == b)
501 break;
503 gcc_assert (nthreaded_edges
504 < (n_basic_blocks_for_fn (cfun)
505 - NUM_FIXED_BLOCKS));
506 threaded_edges[nthreaded_edges++] = t;
508 new_target = t->dest;
509 new_target_threaded = true;
513 if (!new_target)
514 break;
516 counter++;
517 /* Do not turn non-crossing jump to crossing. Depending on target
518 it may require different instruction pattern. */
519 if ((e->flags & EDGE_CROSSING)
520 || BB_PARTITION (first) == BB_PARTITION (new_target))
522 target = new_target;
523 threaded |= new_target_threaded;
527 if (counter >= n_basic_blocks_for_fn (cfun))
529 if (dump_file)
530 fprintf (dump_file, "Infinite loop in BB %i.\n",
531 target->index);
533 else if (target == first)
534 ; /* We didn't do anything. */
535 else
537 /* Save the values now, as the edge may get removed. */
538 profile_count edge_count = e->count ();
539 int n = 0;
541 e->goto_locus = goto_locus;
543 /* Don't force if target is exit block. */
544 if (threaded && target != EXIT_BLOCK_PTR_FOR_FN (cfun))
546 notice_new_block (redirect_edge_and_branch_force (e, target));
547 if (dump_file)
548 fprintf (dump_file, "Conditionals threaded.\n");
550 else if (!redirect_edge_and_branch (e, target))
552 if (dump_file)
553 fprintf (dump_file,
554 "Forwarding edge %i->%i to %i failed.\n",
555 b->index, e->dest->index, target->index);
556 ei_next (&ei);
557 continue;
560 /* We successfully forwarded the edge. Now update profile
561 data: for each edge we traversed in the chain, remove
562 the original edge's execution count. */
565 edge t;
567 if (!single_succ_p (first))
569 gcc_assert (n < nthreaded_edges);
570 t = threaded_edges [n++];
571 gcc_assert (t->src == first);
572 update_bb_profile_for_threading (first, edge_count, t);
573 update_br_prob_note (first);
575 else
577 first->count -= edge_count;
578 /* It is possible that as the result of
579 threading we've removed edge as it is
580 threaded to the fallthru edge. Avoid
581 getting out of sync. */
582 if (n < nthreaded_edges
583 && first == threaded_edges [n]->src)
584 n++;
585 t = single_succ_edge (first);
588 first = t->dest;
590 while (first != target);
592 changed = true;
593 continue;
595 ei_next (&ei);
598 free (threaded_edges);
599 return changed;
603 /* Blocks A and B are to be merged into a single block. A has no incoming
604 fallthru edge, so it can be moved before B without adding or modifying
605 any jumps (aside from the jump from A to B). */
607 static void
608 merge_blocks_move_predecessor_nojumps (basic_block a, basic_block b)
610 rtx_insn *barrier;
612 /* If we are partitioning hot/cold basic blocks, we don't want to
613 mess up unconditional or indirect jumps that cross between hot
614 and cold sections.
616 Basic block partitioning may result in some jumps that appear to
617 be optimizable (or blocks that appear to be mergeable), but which really
618 must be left untouched (they are required to make it safely across
619 partition boundaries). See the comments at the top of
620 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
622 if (BB_PARTITION (a) != BB_PARTITION (b))
623 return;
625 barrier = next_nonnote_insn (BB_END (a));
626 gcc_assert (BARRIER_P (barrier));
627 delete_insn (barrier);
629 /* Scramble the insn chain. */
630 if (BB_END (a) != PREV_INSN (BB_HEAD (b)))
631 reorder_insns_nobb (BB_HEAD (a), BB_END (a), PREV_INSN (BB_HEAD (b)));
632 df_set_bb_dirty (a);
634 if (dump_file)
635 fprintf (dump_file, "Moved block %d before %d and merged.\n",
636 a->index, b->index);
638 /* Swap the records for the two blocks around. */
640 unlink_block (a);
641 link_block (a, b->prev_bb);
643 /* Now blocks A and B are contiguous. Merge them. */
644 merge_blocks (a, b);
647 /* Blocks A and B are to be merged into a single block. B has no outgoing
648 fallthru edge, so it can be moved after A without adding or modifying
649 any jumps (aside from the jump from A to B). */
651 static void
652 merge_blocks_move_successor_nojumps (basic_block a, basic_block b)
654 rtx_insn *barrier, *real_b_end;
655 rtx_insn *label;
656 rtx_jump_table_data *table;
658 /* If we are partitioning hot/cold basic blocks, we don't want to
659 mess up unconditional or indirect jumps that cross between hot
660 and cold sections.
662 Basic block partitioning may result in some jumps that appear to
663 be optimizable (or blocks that appear to be mergeable), but which really
664 must be left untouched (they are required to make it safely across
665 partition boundaries). See the comments at the top of
666 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
668 if (BB_PARTITION (a) != BB_PARTITION (b))
669 return;
671 real_b_end = BB_END (b);
673 /* If there is a jump table following block B temporarily add the jump table
674 to block B so that it will also be moved to the correct location. */
675 if (tablejump_p (BB_END (b), &label, &table)
676 && prev_active_insn (label) == BB_END (b))
678 BB_END (b) = table;
681 /* There had better have been a barrier there. Delete it. */
682 barrier = NEXT_INSN (BB_END (b));
683 if (barrier && BARRIER_P (barrier))
684 delete_insn (barrier);
687 /* Scramble the insn chain. */
688 reorder_insns_nobb (BB_HEAD (b), BB_END (b), BB_END (a));
690 /* Restore the real end of b. */
691 BB_END (b) = real_b_end;
693 if (dump_file)
694 fprintf (dump_file, "Moved block %d after %d and merged.\n",
695 b->index, a->index);
697 /* Now blocks A and B are contiguous. Merge them. */
698 merge_blocks (a, b);
701 /* Attempt to merge basic blocks that are potentially non-adjacent.
702 Return NULL iff the attempt failed, otherwise return basic block
703 where cleanup_cfg should continue. Because the merging commonly
704 moves basic block away or introduces another optimization
705 possibility, return basic block just before B so cleanup_cfg don't
706 need to iterate.
708 It may be good idea to return basic block before C in the case
709 C has been moved after B and originally appeared earlier in the
710 insn sequence, but we have no information available about the
711 relative ordering of these two. Hopefully it is not too common. */
713 static basic_block
714 merge_blocks_move (edge e, basic_block b, basic_block c, int mode)
716 basic_block next;
718 /* If we are partitioning hot/cold basic blocks, we don't want to
719 mess up unconditional or indirect jumps that cross between hot
720 and cold sections.
722 Basic block partitioning may result in some jumps that appear to
723 be optimizable (or blocks that appear to be mergeable), but which really
724 must be left untouched (they are required to make it safely across
725 partition boundaries). See the comments at the top of
726 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
728 if (BB_PARTITION (b) != BB_PARTITION (c))
729 return NULL;
731 /* If B has a fallthru edge to C, no need to move anything. */
732 if (e->flags & EDGE_FALLTHRU)
734 int b_index = b->index, c_index = c->index;
736 /* Protect the loop latches. */
737 if (current_loops && c->loop_father->latch == c)
738 return NULL;
740 merge_blocks (b, c);
741 update_forwarder_flag (b);
743 if (dump_file)
744 fprintf (dump_file, "Merged %d and %d without moving.\n",
745 b_index, c_index);
747 return b->prev_bb == ENTRY_BLOCK_PTR_FOR_FN (cfun) ? b : b->prev_bb;
750 /* Otherwise we will need to move code around. Do that only if expensive
751 transformations are allowed. */
752 else if (mode & CLEANUP_EXPENSIVE)
754 edge tmp_edge, b_fallthru_edge;
755 bool c_has_outgoing_fallthru;
756 bool b_has_incoming_fallthru;
758 /* Avoid overactive code motion, as the forwarder blocks should be
759 eliminated by edge redirection instead. One exception might have
760 been if B is a forwarder block and C has no fallthru edge, but
761 that should be cleaned up by bb-reorder instead. */
762 if (FORWARDER_BLOCK_P (b) || FORWARDER_BLOCK_P (c))
763 return NULL;
765 /* We must make sure to not munge nesting of lexical blocks,
766 and loop notes. This is done by squeezing out all the notes
767 and leaving them there to lie. Not ideal, but functional. */
769 tmp_edge = find_fallthru_edge (c->succs);
770 c_has_outgoing_fallthru = (tmp_edge != NULL);
772 tmp_edge = find_fallthru_edge (b->preds);
773 b_has_incoming_fallthru = (tmp_edge != NULL);
774 b_fallthru_edge = tmp_edge;
775 next = b->prev_bb;
776 if (next == c)
777 next = next->prev_bb;
779 /* Otherwise, we're going to try to move C after B. If C does
780 not have an outgoing fallthru, then it can be moved
781 immediately after B without introducing or modifying jumps. */
782 if (! c_has_outgoing_fallthru)
784 merge_blocks_move_successor_nojumps (b, c);
785 return next == ENTRY_BLOCK_PTR_FOR_FN (cfun) ? next->next_bb : next;
788 /* If B does not have an incoming fallthru, then it can be moved
789 immediately before C without introducing or modifying jumps.
790 C cannot be the first block, so we do not have to worry about
791 accessing a non-existent block. */
793 if (b_has_incoming_fallthru)
795 basic_block bb;
797 if (b_fallthru_edge->src == ENTRY_BLOCK_PTR_FOR_FN (cfun))
798 return NULL;
799 bb = force_nonfallthru (b_fallthru_edge);
800 if (bb)
801 notice_new_block (bb);
804 merge_blocks_move_predecessor_nojumps (b, c);
805 return next == ENTRY_BLOCK_PTR_FOR_FN (cfun) ? next->next_bb : next;
808 return NULL;
812 /* Removes the memory attributes of MEM expression
813 if they are not equal. */
815 static void
816 merge_memattrs (rtx x, rtx y)
818 int i;
819 int j;
820 enum rtx_code code;
821 const char *fmt;
823 if (x == y)
824 return;
825 if (x == 0 || y == 0)
826 return;
828 code = GET_CODE (x);
830 if (code != GET_CODE (y))
831 return;
833 if (GET_MODE (x) != GET_MODE (y))
834 return;
836 if (code == MEM && !mem_attrs_eq_p (MEM_ATTRS (x), MEM_ATTRS (y)))
838 if (! MEM_ATTRS (x))
839 MEM_ATTRS (y) = 0;
840 else if (! MEM_ATTRS (y))
841 MEM_ATTRS (x) = 0;
842 else
844 if (MEM_ALIAS_SET (x) != MEM_ALIAS_SET (y))
846 set_mem_alias_set (x, 0);
847 set_mem_alias_set (y, 0);
850 if (! mem_expr_equal_p (MEM_EXPR (x), MEM_EXPR (y)))
852 set_mem_expr (x, 0);
853 set_mem_expr (y, 0);
854 clear_mem_offset (x);
855 clear_mem_offset (y);
857 else if (MEM_OFFSET_KNOWN_P (x) != MEM_OFFSET_KNOWN_P (y)
858 || (MEM_OFFSET_KNOWN_P (x)
859 && maybe_ne (MEM_OFFSET (x), MEM_OFFSET (y))))
861 clear_mem_offset (x);
862 clear_mem_offset (y);
865 if (!MEM_SIZE_KNOWN_P (x))
866 clear_mem_size (y);
867 else if (!MEM_SIZE_KNOWN_P (y))
868 clear_mem_size (x);
869 else if (known_le (MEM_SIZE (x), MEM_SIZE (y)))
870 set_mem_size (x, MEM_SIZE (y));
871 else if (known_le (MEM_SIZE (y), MEM_SIZE (x)))
872 set_mem_size (y, MEM_SIZE (x));
873 else
875 /* The sizes aren't ordered, so we can't merge them. */
876 clear_mem_size (x);
877 clear_mem_size (y);
880 set_mem_align (x, MIN (MEM_ALIGN (x), MEM_ALIGN (y)));
881 set_mem_align (y, MEM_ALIGN (x));
884 if (code == MEM)
886 if (MEM_READONLY_P (x) != MEM_READONLY_P (y))
888 MEM_READONLY_P (x) = 0;
889 MEM_READONLY_P (y) = 0;
891 if (MEM_NOTRAP_P (x) != MEM_NOTRAP_P (y))
893 MEM_NOTRAP_P (x) = 0;
894 MEM_NOTRAP_P (y) = 0;
896 if (MEM_VOLATILE_P (x) != MEM_VOLATILE_P (y))
898 MEM_VOLATILE_P (x) = 1;
899 MEM_VOLATILE_P (y) = 1;
903 fmt = GET_RTX_FORMAT (code);
904 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
906 switch (fmt[i])
908 case 'E':
909 /* Two vectors must have the same length. */
910 if (XVECLEN (x, i) != XVECLEN (y, i))
911 return;
913 for (j = 0; j < XVECLEN (x, i); j++)
914 merge_memattrs (XVECEXP (x, i, j), XVECEXP (y, i, j));
916 break;
918 case 'e':
919 merge_memattrs (XEXP (x, i), XEXP (y, i));
922 return;
926 /* Checks if patterns P1 and P2 are equivalent, apart from the possibly
927 different single sets S1 and S2. */
929 static bool
930 equal_different_set_p (rtx p1, rtx s1, rtx p2, rtx s2)
932 int i;
933 rtx e1, e2;
935 if (p1 == s1 && p2 == s2)
936 return true;
938 if (GET_CODE (p1) != PARALLEL || GET_CODE (p2) != PARALLEL)
939 return false;
941 if (XVECLEN (p1, 0) != XVECLEN (p2, 0))
942 return false;
944 for (i = 0; i < XVECLEN (p1, 0); i++)
946 e1 = XVECEXP (p1, 0, i);
947 e2 = XVECEXP (p2, 0, i);
948 if (e1 == s1 && e2 == s2)
949 continue;
950 if (reload_completed
951 ? rtx_renumbered_equal_p (e1, e2) : rtx_equal_p (e1, e2))
952 continue;
954 return false;
957 return true;
961 /* NOTE1 is the REG_EQUAL note, if any, attached to an insn
962 that is a single_set with a SET_SRC of SRC1. Similarly
963 for NOTE2/SRC2.
965 So effectively NOTE1/NOTE2 are an alternate form of
966 SRC1/SRC2 respectively.
968 Return nonzero if SRC1 or NOTE1 has the same constant
969 integer value as SRC2 or NOTE2. Else return zero. */
970 static int
971 values_equal_p (rtx note1, rtx note2, rtx src1, rtx src2)
973 if (note1
974 && note2
975 && CONST_INT_P (XEXP (note1, 0))
976 && rtx_equal_p (XEXP (note1, 0), XEXP (note2, 0)))
977 return 1;
979 if (!note1
980 && !note2
981 && CONST_INT_P (src1)
982 && CONST_INT_P (src2)
983 && rtx_equal_p (src1, src2))
984 return 1;
986 if (note1
987 && CONST_INT_P (src2)
988 && rtx_equal_p (XEXP (note1, 0), src2))
989 return 1;
991 if (note2
992 && CONST_INT_P (src1)
993 && rtx_equal_p (XEXP (note2, 0), src1))
994 return 1;
996 return 0;
999 /* Examine register notes on I1 and I2 and return:
1000 - dir_forward if I1 can be replaced by I2, or
1001 - dir_backward if I2 can be replaced by I1, or
1002 - dir_both if both are the case. */
1004 static enum replace_direction
1005 can_replace_by (rtx_insn *i1, rtx_insn *i2)
1007 rtx s1, s2, d1, d2, src1, src2, note1, note2;
1008 bool c1, c2;
1010 /* Check for 2 sets. */
1011 s1 = single_set (i1);
1012 s2 = single_set (i2);
1013 if (s1 == NULL_RTX || s2 == NULL_RTX)
1014 return dir_none;
1016 /* Check that the 2 sets set the same dest. */
1017 d1 = SET_DEST (s1);
1018 d2 = SET_DEST (s2);
1019 if (!(reload_completed
1020 ? rtx_renumbered_equal_p (d1, d2) : rtx_equal_p (d1, d2)))
1021 return dir_none;
1023 /* Find identical req_equiv or reg_equal note, which implies that the 2 sets
1024 set dest to the same value. */
1025 note1 = find_reg_equal_equiv_note (i1);
1026 note2 = find_reg_equal_equiv_note (i2);
1028 src1 = SET_SRC (s1);
1029 src2 = SET_SRC (s2);
1031 if (!values_equal_p (note1, note2, src1, src2))
1032 return dir_none;
1034 if (!equal_different_set_p (PATTERN (i1), s1, PATTERN (i2), s2))
1035 return dir_none;
1037 /* Although the 2 sets set dest to the same value, we cannot replace
1038 (set (dest) (const_int))
1040 (set (dest) (reg))
1041 because we don't know if the reg is live and has the same value at the
1042 location of replacement. */
1043 c1 = CONST_INT_P (src1);
1044 c2 = CONST_INT_P (src2);
1045 if (c1 && c2)
1046 return dir_both;
1047 else if (c2)
1048 return dir_forward;
1049 else if (c1)
1050 return dir_backward;
1052 return dir_none;
1055 /* Merges directions A and B. */
1057 static enum replace_direction
1058 merge_dir (enum replace_direction a, enum replace_direction b)
1060 /* Implements the following table:
1061 |bo fw bw no
1062 ---+-----------
1063 bo |bo fw bw no
1064 fw |-- fw no no
1065 bw |-- -- bw no
1066 no |-- -- -- no. */
1068 if (a == b)
1069 return a;
1071 if (a == dir_both)
1072 return b;
1073 if (b == dir_both)
1074 return a;
1076 return dir_none;
1079 /* Array of flags indexed by reg note kind, true if the given
1080 reg note is CFA related. */
1081 static const bool reg_note_cfa_p[] = {
1082 #undef REG_CFA_NOTE
1083 #define DEF_REG_NOTE(NAME) false,
1084 #define REG_CFA_NOTE(NAME) true,
1085 #include "reg-notes.def"
1086 #undef REG_CFA_NOTE
1087 #undef DEF_REG_NOTE
1088 false
1091 /* Return true if I1 and I2 have identical CFA notes (the same order
1092 and equivalent content). */
1094 static bool
1095 insns_have_identical_cfa_notes (rtx_insn *i1, rtx_insn *i2)
1097 rtx n1, n2;
1098 for (n1 = REG_NOTES (i1), n2 = REG_NOTES (i2); ;
1099 n1 = XEXP (n1, 1), n2 = XEXP (n2, 1))
1101 /* Skip over reg notes not related to CFI information. */
1102 while (n1 && !reg_note_cfa_p[REG_NOTE_KIND (n1)])
1103 n1 = XEXP (n1, 1);
1104 while (n2 && !reg_note_cfa_p[REG_NOTE_KIND (n2)])
1105 n2 = XEXP (n2, 1);
1106 if (n1 == NULL_RTX && n2 == NULL_RTX)
1107 return true;
1108 if (n1 == NULL_RTX || n2 == NULL_RTX)
1109 return false;
1110 if (XEXP (n1, 0) == XEXP (n2, 0))
1112 else if (XEXP (n1, 0) == NULL_RTX || XEXP (n2, 0) == NULL_RTX)
1113 return false;
1114 else if (!(reload_completed
1115 ? rtx_renumbered_equal_p (XEXP (n1, 0), XEXP (n2, 0))
1116 : rtx_equal_p (XEXP (n1, 0), XEXP (n2, 0))))
1117 return false;
1121 /* Examine I1 and I2 and return:
1122 - dir_forward if I1 can be replaced by I2, or
1123 - dir_backward if I2 can be replaced by I1, or
1124 - dir_both if both are the case. */
1126 static enum replace_direction
1127 old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx_insn *i1, rtx_insn *i2)
1129 rtx p1, p2;
1131 /* Verify that I1 and I2 are equivalent. */
1132 if (GET_CODE (i1) != GET_CODE (i2))
1133 return dir_none;
1135 /* __builtin_unreachable() may lead to empty blocks (ending with
1136 NOTE_INSN_BASIC_BLOCK). They may be crossjumped. */
1137 if (NOTE_INSN_BASIC_BLOCK_P (i1) && NOTE_INSN_BASIC_BLOCK_P (i2))
1138 return dir_both;
1140 /* ??? Do not allow cross-jumping between different stack levels. */
1141 p1 = find_reg_note (i1, REG_ARGS_SIZE, NULL);
1142 p2 = find_reg_note (i2, REG_ARGS_SIZE, NULL);
1143 if (p1 && p2)
1145 p1 = XEXP (p1, 0);
1146 p2 = XEXP (p2, 0);
1147 if (!rtx_equal_p (p1, p2))
1148 return dir_none;
1150 /* ??? Worse, this adjustment had better be constant lest we
1151 have differing incoming stack levels. */
1152 if (!frame_pointer_needed
1153 && known_eq (find_args_size_adjust (i1), HOST_WIDE_INT_MIN))
1154 return dir_none;
1156 else if (p1 || p2)
1157 return dir_none;
1159 /* Do not allow cross-jumping between frame related insns and other
1160 insns. */
1161 if (RTX_FRAME_RELATED_P (i1) != RTX_FRAME_RELATED_P (i2))
1162 return dir_none;
1164 p1 = PATTERN (i1);
1165 p2 = PATTERN (i2);
1167 if (GET_CODE (p1) != GET_CODE (p2))
1168 return dir_none;
1170 /* If this is a CALL_INSN, compare register usage information.
1171 If we don't check this on stack register machines, the two
1172 CALL_INSNs might be merged leaving reg-stack.c with mismatching
1173 numbers of stack registers in the same basic block.
1174 If we don't check this on machines with delay slots, a delay slot may
1175 be filled that clobbers a parameter expected by the subroutine.
1177 ??? We take the simple route for now and assume that if they're
1178 equal, they were constructed identically.
1180 Also check for identical exception regions. */
1182 if (CALL_P (i1))
1184 /* Ensure the same EH region. */
1185 rtx n1 = find_reg_note (i1, REG_EH_REGION, 0);
1186 rtx n2 = find_reg_note (i2, REG_EH_REGION, 0);
1188 if (!n1 && n2)
1189 return dir_none;
1191 if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0)))
1192 return dir_none;
1194 if (!rtx_equal_p (CALL_INSN_FUNCTION_USAGE (i1),
1195 CALL_INSN_FUNCTION_USAGE (i2))
1196 || SIBLING_CALL_P (i1) != SIBLING_CALL_P (i2))
1197 return dir_none;
1199 /* For address sanitizer, never crossjump __asan_report_* builtins,
1200 otherwise errors might be reported on incorrect lines. */
1201 if (flag_sanitize & SANITIZE_ADDRESS)
1203 rtx call = get_call_rtx_from (i1);
1204 if (call && GET_CODE (XEXP (XEXP (call, 0), 0)) == SYMBOL_REF)
1206 rtx symbol = XEXP (XEXP (call, 0), 0);
1207 if (SYMBOL_REF_DECL (symbol)
1208 && TREE_CODE (SYMBOL_REF_DECL (symbol)) == FUNCTION_DECL)
1210 if ((DECL_BUILT_IN_CLASS (SYMBOL_REF_DECL (symbol))
1211 == BUILT_IN_NORMAL)
1212 && DECL_FUNCTION_CODE (SYMBOL_REF_DECL (symbol))
1213 >= BUILT_IN_ASAN_REPORT_LOAD1
1214 && DECL_FUNCTION_CODE (SYMBOL_REF_DECL (symbol))
1215 <= BUILT_IN_ASAN_STOREN)
1216 return dir_none;
1222 /* If both i1 and i2 are frame related, verify all the CFA notes
1223 in the same order and with the same content. */
1224 if (RTX_FRAME_RELATED_P (i1) && !insns_have_identical_cfa_notes (i1, i2))
1225 return dir_none;
1227 #ifdef STACK_REGS
1228 /* If cross_jump_death_matters is not 0, the insn's mode
1229 indicates whether or not the insn contains any stack-like
1230 regs. */
1232 if ((mode & CLEANUP_POST_REGSTACK) && stack_regs_mentioned (i1))
1234 /* If register stack conversion has already been done, then
1235 death notes must also be compared before it is certain that
1236 the two instruction streams match. */
1238 rtx note;
1239 HARD_REG_SET i1_regset, i2_regset;
1241 CLEAR_HARD_REG_SET (i1_regset);
1242 CLEAR_HARD_REG_SET (i2_regset);
1244 for (note = REG_NOTES (i1); note; note = XEXP (note, 1))
1245 if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
1246 SET_HARD_REG_BIT (i1_regset, REGNO (XEXP (note, 0)));
1248 for (note = REG_NOTES (i2); note; note = XEXP (note, 1))
1249 if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
1250 SET_HARD_REG_BIT (i2_regset, REGNO (XEXP (note, 0)));
1252 if (!hard_reg_set_equal_p (i1_regset, i2_regset))
1253 return dir_none;
1255 #endif
1257 if (reload_completed
1258 ? rtx_renumbered_equal_p (p1, p2) : rtx_equal_p (p1, p2))
1259 return dir_both;
1261 return can_replace_by (i1, i2);
1264 /* When comparing insns I1 and I2 in flow_find_cross_jump or
1265 flow_find_head_matching_sequence, ensure the notes match. */
1267 static void
1268 merge_notes (rtx_insn *i1, rtx_insn *i2)
1270 /* If the merged insns have different REG_EQUAL notes, then
1271 remove them. */
1272 rtx equiv1 = find_reg_equal_equiv_note (i1);
1273 rtx equiv2 = find_reg_equal_equiv_note (i2);
1275 if (equiv1 && !equiv2)
1276 remove_note (i1, equiv1);
1277 else if (!equiv1 && equiv2)
1278 remove_note (i2, equiv2);
1279 else if (equiv1 && equiv2
1280 && !rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0)))
1282 remove_note (i1, equiv1);
1283 remove_note (i2, equiv2);
1287 /* Walks from I1 in BB1 backward till the next non-debug insn, and returns the
1288 resulting insn in I1, and the corresponding bb in BB1. At the head of a
1289 bb, if there is a predecessor bb that reaches this bb via fallthru, and
1290 FOLLOW_FALLTHRU, walks further in the predecessor bb and registers this in
1291 DID_FALLTHRU. Otherwise, stops at the head of the bb. */
1293 static void
1294 walk_to_nondebug_insn (rtx_insn **i1, basic_block *bb1, bool follow_fallthru,
1295 bool *did_fallthru)
1297 edge fallthru;
1299 *did_fallthru = false;
1301 /* Ignore notes. */
1302 while (!NONDEBUG_INSN_P (*i1))
1304 if (*i1 != BB_HEAD (*bb1))
1306 *i1 = PREV_INSN (*i1);
1307 continue;
1310 if (!follow_fallthru)
1311 return;
1313 fallthru = find_fallthru_edge ((*bb1)->preds);
1314 if (!fallthru || fallthru->src == ENTRY_BLOCK_PTR_FOR_FN (cfun)
1315 || !single_succ_p (fallthru->src))
1316 return;
1318 *bb1 = fallthru->src;
1319 *i1 = BB_END (*bb1);
1320 *did_fallthru = true;
1324 /* Look through the insns at the end of BB1 and BB2 and find the longest
1325 sequence that are either equivalent, or allow forward or backward
1326 replacement. Store the first insns for that sequence in *F1 and *F2 and
1327 return the sequence length.
1329 DIR_P indicates the allowed replacement direction on function entry, and
1330 the actual replacement direction on function exit. If NULL, only equivalent
1331 sequences are allowed.
1333 To simplify callers of this function, if the blocks match exactly,
1334 store the head of the blocks in *F1 and *F2. */
1337 flow_find_cross_jump (basic_block bb1, basic_block bb2, rtx_insn **f1,
1338 rtx_insn **f2, enum replace_direction *dir_p)
1340 rtx_insn *i1, *i2, *last1, *last2, *afterlast1, *afterlast2;
1341 int ninsns = 0;
1342 enum replace_direction dir, last_dir, afterlast_dir;
1343 bool follow_fallthru, did_fallthru;
1345 if (dir_p)
1346 dir = *dir_p;
1347 else
1348 dir = dir_both;
1349 afterlast_dir = dir;
1350 last_dir = afterlast_dir;
1352 /* Skip simple jumps at the end of the blocks. Complex jumps still
1353 need to be compared for equivalence, which we'll do below. */
1355 i1 = BB_END (bb1);
1356 last1 = afterlast1 = last2 = afterlast2 = NULL;
1357 if (onlyjump_p (i1)
1358 || (returnjump_p (i1) && !side_effects_p (PATTERN (i1))))
1360 last1 = i1;
1361 i1 = PREV_INSN (i1);
1364 i2 = BB_END (bb2);
1365 if (onlyjump_p (i2)
1366 || (returnjump_p (i2) && !side_effects_p (PATTERN (i2))))
1368 last2 = i2;
1369 /* Count everything except for unconditional jump as insn.
1370 Don't count any jumps if dir_p is NULL. */
1371 if (!simplejump_p (i2) && !returnjump_p (i2) && last1 && dir_p)
1372 ninsns++;
1373 i2 = PREV_INSN (i2);
1376 while (true)
1378 /* In the following example, we can replace all jumps to C by jumps to A.
1380 This removes 4 duplicate insns.
1381 [bb A] insn1 [bb C] insn1
1382 insn2 insn2
1383 [bb B] insn3 insn3
1384 insn4 insn4
1385 jump_insn jump_insn
1387 We could also replace all jumps to A by jumps to C, but that leaves B
1388 alive, and removes only 2 duplicate insns. In a subsequent crossjump
1389 step, all jumps to B would be replaced with jumps to the middle of C,
1390 achieving the same result with more effort.
1391 So we allow only the first possibility, which means that we don't allow
1392 fallthru in the block that's being replaced. */
1394 follow_fallthru = dir_p && dir != dir_forward;
1395 walk_to_nondebug_insn (&i1, &bb1, follow_fallthru, &did_fallthru);
1396 if (did_fallthru)
1397 dir = dir_backward;
1399 follow_fallthru = dir_p && dir != dir_backward;
1400 walk_to_nondebug_insn (&i2, &bb2, follow_fallthru, &did_fallthru);
1401 if (did_fallthru)
1402 dir = dir_forward;
1404 if (i1 == BB_HEAD (bb1) || i2 == BB_HEAD (bb2))
1405 break;
1407 /* Do not turn corssing edge to non-crossing or vice versa after
1408 reload. */
1409 if (BB_PARTITION (BLOCK_FOR_INSN (i1))
1410 != BB_PARTITION (BLOCK_FOR_INSN (i2))
1411 && reload_completed)
1412 break;
1414 dir = merge_dir (dir, old_insns_match_p (0, i1, i2));
1415 if (dir == dir_none || (!dir_p && dir != dir_both))
1416 break;
1418 merge_memattrs (i1, i2);
1420 /* Don't begin a cross-jump with a NOTE insn. */
1421 if (INSN_P (i1))
1423 merge_notes (i1, i2);
1425 afterlast1 = last1, afterlast2 = last2;
1426 last1 = i1, last2 = i2;
1427 afterlast_dir = last_dir;
1428 last_dir = dir;
1429 if (active_insn_p (i1))
1430 ninsns++;
1433 i1 = PREV_INSN (i1);
1434 i2 = PREV_INSN (i2);
1437 /* Don't allow the insn after a compare to be shared by
1438 cross-jumping unless the compare is also shared. */
1439 if (HAVE_cc0 && ninsns && reg_mentioned_p (cc0_rtx, last1)
1440 && ! sets_cc0_p (last1))
1441 last1 = afterlast1, last2 = afterlast2, last_dir = afterlast_dir, ninsns--;
1443 /* Include preceding notes and labels in the cross-jump. One,
1444 this may bring us to the head of the blocks as requested above.
1445 Two, it keeps line number notes as matched as may be. */
1446 if (ninsns)
1448 bb1 = BLOCK_FOR_INSN (last1);
1449 while (last1 != BB_HEAD (bb1) && !NONDEBUG_INSN_P (PREV_INSN (last1)))
1450 last1 = PREV_INSN (last1);
1452 if (last1 != BB_HEAD (bb1) && LABEL_P (PREV_INSN (last1)))
1453 last1 = PREV_INSN (last1);
1455 bb2 = BLOCK_FOR_INSN (last2);
1456 while (last2 != BB_HEAD (bb2) && !NONDEBUG_INSN_P (PREV_INSN (last2)))
1457 last2 = PREV_INSN (last2);
1459 if (last2 != BB_HEAD (bb2) && LABEL_P (PREV_INSN (last2)))
1460 last2 = PREV_INSN (last2);
1462 *f1 = last1;
1463 *f2 = last2;
1466 if (dir_p)
1467 *dir_p = last_dir;
1468 return ninsns;
1471 /* Like flow_find_cross_jump, except start looking for a matching sequence from
1472 the head of the two blocks. Do not include jumps at the end.
1473 If STOP_AFTER is nonzero, stop after finding that many matching
1474 instructions. If STOP_AFTER is zero, count all INSN_P insns, if it is
1475 non-zero, only count active insns. */
1478 flow_find_head_matching_sequence (basic_block bb1, basic_block bb2, rtx_insn **f1,
1479 rtx_insn **f2, int stop_after)
1481 rtx_insn *i1, *i2, *last1, *last2, *beforelast1, *beforelast2;
1482 int ninsns = 0;
1483 edge e;
1484 edge_iterator ei;
1485 int nehedges1 = 0, nehedges2 = 0;
1487 FOR_EACH_EDGE (e, ei, bb1->succs)
1488 if (e->flags & EDGE_EH)
1489 nehedges1++;
1490 FOR_EACH_EDGE (e, ei, bb2->succs)
1491 if (e->flags & EDGE_EH)
1492 nehedges2++;
1494 i1 = BB_HEAD (bb1);
1495 i2 = BB_HEAD (bb2);
1496 last1 = beforelast1 = last2 = beforelast2 = NULL;
1498 while (true)
1500 /* Ignore notes, except NOTE_INSN_EPILOGUE_BEG. */
1501 while (!NONDEBUG_INSN_P (i1) && i1 != BB_END (bb1))
1503 if (NOTE_P (i1) && NOTE_KIND (i1) == NOTE_INSN_EPILOGUE_BEG)
1504 break;
1505 i1 = NEXT_INSN (i1);
1508 while (!NONDEBUG_INSN_P (i2) && i2 != BB_END (bb2))
1510 if (NOTE_P (i2) && NOTE_KIND (i2) == NOTE_INSN_EPILOGUE_BEG)
1511 break;
1512 i2 = NEXT_INSN (i2);
1515 if ((i1 == BB_END (bb1) && !NONDEBUG_INSN_P (i1))
1516 || (i2 == BB_END (bb2) && !NONDEBUG_INSN_P (i2)))
1517 break;
1519 if (NOTE_P (i1) || NOTE_P (i2)
1520 || JUMP_P (i1) || JUMP_P (i2))
1521 break;
1523 /* A sanity check to make sure we're not merging insns with different
1524 effects on EH. If only one of them ends a basic block, it shouldn't
1525 have an EH edge; if both end a basic block, there should be the same
1526 number of EH edges. */
1527 if ((i1 == BB_END (bb1) && i2 != BB_END (bb2)
1528 && nehedges1 > 0)
1529 || (i2 == BB_END (bb2) && i1 != BB_END (bb1)
1530 && nehedges2 > 0)
1531 || (i1 == BB_END (bb1) && i2 == BB_END (bb2)
1532 && nehedges1 != nehedges2))
1533 break;
1535 if (old_insns_match_p (0, i1, i2) != dir_both)
1536 break;
1538 merge_memattrs (i1, i2);
1540 /* Don't begin a cross-jump with a NOTE insn. */
1541 if (INSN_P (i1))
1543 merge_notes (i1, i2);
1545 beforelast1 = last1, beforelast2 = last2;
1546 last1 = i1, last2 = i2;
1547 if (!stop_after || active_insn_p (i1))
1548 ninsns++;
1551 if (i1 == BB_END (bb1) || i2 == BB_END (bb2)
1552 || (stop_after > 0 && ninsns == stop_after))
1553 break;
1555 i1 = NEXT_INSN (i1);
1556 i2 = NEXT_INSN (i2);
1559 /* Don't allow a compare to be shared by cross-jumping unless the insn
1560 after the compare is also shared. */
1561 if (HAVE_cc0 && ninsns && reg_mentioned_p (cc0_rtx, last1)
1562 && sets_cc0_p (last1))
1563 last1 = beforelast1, last2 = beforelast2, ninsns--;
1565 if (ninsns)
1567 *f1 = last1;
1568 *f2 = last2;
1571 return ninsns;
1574 /* Return true iff outgoing edges of BB1 and BB2 match, together with
1575 the branch instruction. This means that if we commonize the control
1576 flow before end of the basic block, the semantic remains unchanged.
1578 We may assume that there exists one edge with a common destination. */
1580 static bool
1581 outgoing_edges_match (int mode, basic_block bb1, basic_block bb2)
1583 int nehedges1 = 0, nehedges2 = 0;
1584 edge fallthru1 = 0, fallthru2 = 0;
1585 edge e1, e2;
1586 edge_iterator ei;
1588 /* If we performed shrink-wrapping, edges to the exit block can
1589 only be distinguished for JUMP_INSNs. The two paths may differ in
1590 whether they went through the prologue. Sibcalls are fine, we know
1591 that we either didn't need or inserted an epilogue before them. */
1592 if (crtl->shrink_wrapped
1593 && single_succ_p (bb1)
1594 && single_succ (bb1) == EXIT_BLOCK_PTR_FOR_FN (cfun)
1595 && !JUMP_P (BB_END (bb1))
1596 && !(CALL_P (BB_END (bb1)) && SIBLING_CALL_P (BB_END (bb1))))
1597 return false;
1599 /* If BB1 has only one successor, we may be looking at either an
1600 unconditional jump, or a fake edge to exit. */
1601 if (single_succ_p (bb1)
1602 && (single_succ_edge (bb1)->flags & (EDGE_COMPLEX | EDGE_FAKE)) == 0
1603 && (!JUMP_P (BB_END (bb1)) || simplejump_p (BB_END (bb1))))
1604 return (single_succ_p (bb2)
1605 && (single_succ_edge (bb2)->flags
1606 & (EDGE_COMPLEX | EDGE_FAKE)) == 0
1607 && (!JUMP_P (BB_END (bb2)) || simplejump_p (BB_END (bb2))));
1609 /* Match conditional jumps - this may get tricky when fallthru and branch
1610 edges are crossed. */
1611 if (EDGE_COUNT (bb1->succs) == 2
1612 && any_condjump_p (BB_END (bb1))
1613 && onlyjump_p (BB_END (bb1)))
1615 edge b1, f1, b2, f2;
1616 bool reverse, match;
1617 rtx set1, set2, cond1, cond2;
1618 enum rtx_code code1, code2;
1620 if (EDGE_COUNT (bb2->succs) != 2
1621 || !any_condjump_p (BB_END (bb2))
1622 || !onlyjump_p (BB_END (bb2)))
1623 return false;
1625 b1 = BRANCH_EDGE (bb1);
1626 b2 = BRANCH_EDGE (bb2);
1627 f1 = FALLTHRU_EDGE (bb1);
1628 f2 = FALLTHRU_EDGE (bb2);
1630 /* Get around possible forwarders on fallthru edges. Other cases
1631 should be optimized out already. */
1632 if (FORWARDER_BLOCK_P (f1->dest))
1633 f1 = single_succ_edge (f1->dest);
1635 if (FORWARDER_BLOCK_P (f2->dest))
1636 f2 = single_succ_edge (f2->dest);
1638 /* To simplify use of this function, return false if there are
1639 unneeded forwarder blocks. These will get eliminated later
1640 during cleanup_cfg. */
1641 if (FORWARDER_BLOCK_P (f1->dest)
1642 || FORWARDER_BLOCK_P (f2->dest)
1643 || FORWARDER_BLOCK_P (b1->dest)
1644 || FORWARDER_BLOCK_P (b2->dest))
1645 return false;
1647 if (f1->dest == f2->dest && b1->dest == b2->dest)
1648 reverse = false;
1649 else if (f1->dest == b2->dest && b1->dest == f2->dest)
1650 reverse = true;
1651 else
1652 return false;
1654 set1 = pc_set (BB_END (bb1));
1655 set2 = pc_set (BB_END (bb2));
1656 if ((XEXP (SET_SRC (set1), 1) == pc_rtx)
1657 != (XEXP (SET_SRC (set2), 1) == pc_rtx))
1658 reverse = !reverse;
1660 cond1 = XEXP (SET_SRC (set1), 0);
1661 cond2 = XEXP (SET_SRC (set2), 0);
1662 code1 = GET_CODE (cond1);
1663 if (reverse)
1664 code2 = reversed_comparison_code (cond2, BB_END (bb2));
1665 else
1666 code2 = GET_CODE (cond2);
1668 if (code2 == UNKNOWN)
1669 return false;
1671 /* Verify codes and operands match. */
1672 match = ((code1 == code2
1673 && rtx_renumbered_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
1674 && rtx_renumbered_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
1675 || (code1 == swap_condition (code2)
1676 && rtx_renumbered_equal_p (XEXP (cond1, 1),
1677 XEXP (cond2, 0))
1678 && rtx_renumbered_equal_p (XEXP (cond1, 0),
1679 XEXP (cond2, 1))));
1681 /* If we return true, we will join the blocks. Which means that
1682 we will only have one branch prediction bit to work with. Thus
1683 we require the existing branches to have probabilities that are
1684 roughly similar. */
1685 if (match
1686 && optimize_bb_for_speed_p (bb1)
1687 && optimize_bb_for_speed_p (bb2))
1689 profile_probability prob2;
1691 if (b1->dest == b2->dest)
1692 prob2 = b2->probability;
1693 else
1694 /* Do not use f2 probability as f2 may be forwarded. */
1695 prob2 = b2->probability.invert ();
1697 /* Fail if the difference in probabilities is greater than 50%.
1698 This rules out two well-predicted branches with opposite
1699 outcomes. */
1700 if (b1->probability.differs_lot_from_p (prob2))
1702 if (dump_file)
1704 fprintf (dump_file,
1705 "Outcomes of branch in bb %i and %i differ too"
1706 " much (", bb1->index, bb2->index);
1707 b1->probability.dump (dump_file);
1708 prob2.dump (dump_file);
1709 fprintf (dump_file, ")\n");
1711 return false;
1715 if (dump_file && match)
1716 fprintf (dump_file, "Conditionals in bb %i and %i match.\n",
1717 bb1->index, bb2->index);
1719 return match;
1722 /* Generic case - we are seeing a computed jump, table jump or trapping
1723 instruction. */
1725 /* Check whether there are tablejumps in the end of BB1 and BB2.
1726 Return true if they are identical. */
1728 rtx_insn *label1, *label2;
1729 rtx_jump_table_data *table1, *table2;
1731 if (tablejump_p (BB_END (bb1), &label1, &table1)
1732 && tablejump_p (BB_END (bb2), &label2, &table2)
1733 && GET_CODE (PATTERN (table1)) == GET_CODE (PATTERN (table2)))
1735 /* The labels should never be the same rtx. If they really are same
1736 the jump tables are same too. So disable crossjumping of blocks BB1
1737 and BB2 because when deleting the common insns in the end of BB1
1738 by delete_basic_block () the jump table would be deleted too. */
1739 /* If LABEL2 is referenced in BB1->END do not do anything
1740 because we would loose information when replacing
1741 LABEL1 by LABEL2 and then LABEL2 by LABEL1 in BB1->END. */
1742 if (label1 != label2 && !rtx_referenced_p (label2, BB_END (bb1)))
1744 /* Set IDENTICAL to true when the tables are identical. */
1745 bool identical = false;
1746 rtx p1, p2;
1748 p1 = PATTERN (table1);
1749 p2 = PATTERN (table2);
1750 if (GET_CODE (p1) == ADDR_VEC && rtx_equal_p (p1, p2))
1752 identical = true;
1754 else if (GET_CODE (p1) == ADDR_DIFF_VEC
1755 && (XVECLEN (p1, 1) == XVECLEN (p2, 1))
1756 && rtx_equal_p (XEXP (p1, 2), XEXP (p2, 2))
1757 && rtx_equal_p (XEXP (p1, 3), XEXP (p2, 3)))
1759 int i;
1761 identical = true;
1762 for (i = XVECLEN (p1, 1) - 1; i >= 0 && identical; i--)
1763 if (!rtx_equal_p (XVECEXP (p1, 1, i), XVECEXP (p2, 1, i)))
1764 identical = false;
1767 if (identical)
1769 bool match;
1771 /* Temporarily replace references to LABEL1 with LABEL2
1772 in BB1->END so that we could compare the instructions. */
1773 replace_label_in_insn (BB_END (bb1), label1, label2, false);
1775 match = (old_insns_match_p (mode, BB_END (bb1), BB_END (bb2))
1776 == dir_both);
1777 if (dump_file && match)
1778 fprintf (dump_file,
1779 "Tablejumps in bb %i and %i match.\n",
1780 bb1->index, bb2->index);
1782 /* Set the original label in BB1->END because when deleting
1783 a block whose end is a tablejump, the tablejump referenced
1784 from the instruction is deleted too. */
1785 replace_label_in_insn (BB_END (bb1), label2, label1, false);
1787 return match;
1790 return false;
1794 /* Find the last non-debug non-note instruction in each bb, except
1795 stop when we see the NOTE_INSN_BASIC_BLOCK, as old_insns_match_p
1796 handles that case specially. old_insns_match_p does not handle
1797 other types of instruction notes. */
1798 rtx_insn *last1 = BB_END (bb1);
1799 rtx_insn *last2 = BB_END (bb2);
1800 while (!NOTE_INSN_BASIC_BLOCK_P (last1) &&
1801 (DEBUG_INSN_P (last1) || NOTE_P (last1)))
1802 last1 = PREV_INSN (last1);
1803 while (!NOTE_INSN_BASIC_BLOCK_P (last2) &&
1804 (DEBUG_INSN_P (last2) || NOTE_P (last2)))
1805 last2 = PREV_INSN (last2);
1806 gcc_assert (last1 && last2);
1808 /* First ensure that the instructions match. There may be many outgoing
1809 edges so this test is generally cheaper. */
1810 if (old_insns_match_p (mode, last1, last2) != dir_both)
1811 return false;
1813 /* Search the outgoing edges, ensure that the counts do match, find possible
1814 fallthru and exception handling edges since these needs more
1815 validation. */
1816 if (EDGE_COUNT (bb1->succs) != EDGE_COUNT (bb2->succs))
1817 return false;
1819 bool nonfakeedges = false;
1820 FOR_EACH_EDGE (e1, ei, bb1->succs)
1822 e2 = EDGE_SUCC (bb2, ei.index);
1824 if ((e1->flags & EDGE_FAKE) == 0)
1825 nonfakeedges = true;
1827 if (e1->flags & EDGE_EH)
1828 nehedges1++;
1830 if (e2->flags & EDGE_EH)
1831 nehedges2++;
1833 if (e1->flags & EDGE_FALLTHRU)
1834 fallthru1 = e1;
1835 if (e2->flags & EDGE_FALLTHRU)
1836 fallthru2 = e2;
1839 /* If number of edges of various types does not match, fail. */
1840 if (nehedges1 != nehedges2
1841 || (fallthru1 != 0) != (fallthru2 != 0))
1842 return false;
1844 /* If !ACCUMULATE_OUTGOING_ARGS, bb1 (and bb2) have no successors
1845 and the last real insn doesn't have REG_ARGS_SIZE note, don't
1846 attempt to optimize, as the two basic blocks might have different
1847 REG_ARGS_SIZE depths. For noreturn calls and unconditional
1848 traps there should be REG_ARG_SIZE notes, they could be missing
1849 for __builtin_unreachable () uses though. */
1850 if (!nonfakeedges
1851 && !ACCUMULATE_OUTGOING_ARGS
1852 && (!INSN_P (last1)
1853 || !find_reg_note (last1, REG_ARGS_SIZE, NULL)))
1854 return false;
1856 /* fallthru edges must be forwarded to the same destination. */
1857 if (fallthru1)
1859 basic_block d1 = (forwarder_block_p (fallthru1->dest)
1860 ? single_succ (fallthru1->dest): fallthru1->dest);
1861 basic_block d2 = (forwarder_block_p (fallthru2->dest)
1862 ? single_succ (fallthru2->dest): fallthru2->dest);
1864 if (d1 != d2)
1865 return false;
1868 /* Ensure the same EH region. */
1870 rtx n1 = find_reg_note (BB_END (bb1), REG_EH_REGION, 0);
1871 rtx n2 = find_reg_note (BB_END (bb2), REG_EH_REGION, 0);
1873 if (!n1 && n2)
1874 return false;
1876 if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0)))
1877 return false;
1880 /* The same checks as in try_crossjump_to_edge. It is required for RTL
1881 version of sequence abstraction. */
1882 FOR_EACH_EDGE (e1, ei, bb2->succs)
1884 edge e2;
1885 edge_iterator ei;
1886 basic_block d1 = e1->dest;
1888 if (FORWARDER_BLOCK_P (d1))
1889 d1 = EDGE_SUCC (d1, 0)->dest;
1891 FOR_EACH_EDGE (e2, ei, bb1->succs)
1893 basic_block d2 = e2->dest;
1894 if (FORWARDER_BLOCK_P (d2))
1895 d2 = EDGE_SUCC (d2, 0)->dest;
1896 if (d1 == d2)
1897 break;
1900 if (!e2)
1901 return false;
1904 return true;
1907 /* Returns true if BB basic block has a preserve label. */
1909 static bool
1910 block_has_preserve_label (basic_block bb)
1912 return (bb
1913 && block_label (bb)
1914 && LABEL_PRESERVE_P (block_label (bb)));
1917 /* E1 and E2 are edges with the same destination block. Search their
1918 predecessors for common code. If found, redirect control flow from
1919 (maybe the middle of) E1->SRC to (maybe the middle of) E2->SRC (dir_forward),
1920 or the other way around (dir_backward). DIR specifies the allowed
1921 replacement direction. */
1923 static bool
1924 try_crossjump_to_edge (int mode, edge e1, edge e2,
1925 enum replace_direction dir)
1927 int nmatch;
1928 basic_block src1 = e1->src, src2 = e2->src;
1929 basic_block redirect_to, redirect_from, to_remove;
1930 basic_block osrc1, osrc2, redirect_edges_to, tmp;
1931 rtx_insn *newpos1, *newpos2;
1932 edge s;
1933 edge_iterator ei;
1935 newpos1 = newpos2 = NULL;
1937 /* Search backward through forwarder blocks. We don't need to worry
1938 about multiple entry or chained forwarders, as they will be optimized
1939 away. We do this to look past the unconditional jump following a
1940 conditional jump that is required due to the current CFG shape. */
1941 if (single_pred_p (src1)
1942 && FORWARDER_BLOCK_P (src1))
1943 e1 = single_pred_edge (src1), src1 = e1->src;
1945 if (single_pred_p (src2)
1946 && FORWARDER_BLOCK_P (src2))
1947 e2 = single_pred_edge (src2), src2 = e2->src;
1949 /* Nothing to do if we reach ENTRY, or a common source block. */
1950 if (src1 == ENTRY_BLOCK_PTR_FOR_FN (cfun) || src2
1951 == ENTRY_BLOCK_PTR_FOR_FN (cfun))
1952 return false;
1953 if (src1 == src2)
1954 return false;
1956 /* Seeing more than 1 forwarder blocks would confuse us later... */
1957 if (FORWARDER_BLOCK_P (e1->dest)
1958 && FORWARDER_BLOCK_P (single_succ (e1->dest)))
1959 return false;
1961 if (FORWARDER_BLOCK_P (e2->dest)
1962 && FORWARDER_BLOCK_P (single_succ (e2->dest)))
1963 return false;
1965 /* Likewise with dead code (possibly newly created by the other optimizations
1966 of cfg_cleanup). */
1967 if (EDGE_COUNT (src1->preds) == 0 || EDGE_COUNT (src2->preds) == 0)
1968 return false;
1970 /* Do not turn corssing edge to non-crossing or vice versa after reload. */
1971 if (BB_PARTITION (src1) != BB_PARTITION (src2)
1972 && reload_completed)
1973 return false;
1975 /* Look for the common insn sequence, part the first ... */
1976 if (!outgoing_edges_match (mode, src1, src2))
1977 return false;
1979 /* ... and part the second. */
1980 nmatch = flow_find_cross_jump (src1, src2, &newpos1, &newpos2, &dir);
1982 osrc1 = src1;
1983 osrc2 = src2;
1984 if (newpos1 != NULL_RTX)
1985 src1 = BLOCK_FOR_INSN (newpos1);
1986 if (newpos2 != NULL_RTX)
1987 src2 = BLOCK_FOR_INSN (newpos2);
1989 /* Check that SRC1 and SRC2 have preds again. They may have changed
1990 above due to the call to flow_find_cross_jump. */
1991 if (EDGE_COUNT (src1->preds) == 0 || EDGE_COUNT (src2->preds) == 0)
1992 return false;
1994 if (dir == dir_backward)
1996 std::swap (osrc1, osrc2);
1997 std::swap (src1, src2);
1998 std::swap (e1, e2);
1999 std::swap (newpos1, newpos2);
2002 /* Don't proceed with the crossjump unless we found a sufficient number
2003 of matching instructions or the 'from' block was totally matched
2004 (such that its predecessors will hopefully be redirected and the
2005 block removed). */
2006 if ((nmatch < PARAM_VALUE (PARAM_MIN_CROSSJUMP_INSNS))
2007 && (newpos1 != BB_HEAD (src1)))
2008 return false;
2010 /* Avoid deleting preserve label when redirecting ABNORMAL edges. */
2011 if (block_has_preserve_label (e1->dest)
2012 && (e1->flags & EDGE_ABNORMAL))
2013 return false;
2015 /* Here we know that the insns in the end of SRC1 which are common with SRC2
2016 will be deleted.
2017 If we have tablejumps in the end of SRC1 and SRC2
2018 they have been already compared for equivalence in outgoing_edges_match ()
2019 so replace the references to TABLE1 by references to TABLE2. */
2021 rtx_insn *label1, *label2;
2022 rtx_jump_table_data *table1, *table2;
2024 if (tablejump_p (BB_END (osrc1), &label1, &table1)
2025 && tablejump_p (BB_END (osrc2), &label2, &table2)
2026 && label1 != label2)
2028 rtx_insn *insn;
2030 /* Replace references to LABEL1 with LABEL2. */
2031 for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
2033 /* Do not replace the label in SRC1->END because when deleting
2034 a block whose end is a tablejump, the tablejump referenced
2035 from the instruction is deleted too. */
2036 if (insn != BB_END (osrc1))
2037 replace_label_in_insn (insn, label1, label2, true);
2042 /* Avoid splitting if possible. We must always split when SRC2 has
2043 EH predecessor edges, or we may end up with basic blocks with both
2044 normal and EH predecessor edges. */
2045 if (newpos2 == BB_HEAD (src2)
2046 && !(EDGE_PRED (src2, 0)->flags & EDGE_EH))
2047 redirect_to = src2;
2048 else
2050 if (newpos2 == BB_HEAD (src2))
2052 /* Skip possible basic block header. */
2053 if (LABEL_P (newpos2))
2054 newpos2 = NEXT_INSN (newpos2);
2055 while (DEBUG_INSN_P (newpos2))
2056 newpos2 = NEXT_INSN (newpos2);
2057 if (NOTE_P (newpos2))
2058 newpos2 = NEXT_INSN (newpos2);
2059 while (DEBUG_INSN_P (newpos2))
2060 newpos2 = NEXT_INSN (newpos2);
2063 if (dump_file)
2064 fprintf (dump_file, "Splitting bb %i before %i insns\n",
2065 src2->index, nmatch);
2066 redirect_to = split_block (src2, PREV_INSN (newpos2))->dest;
2069 if (dump_file)
2070 fprintf (dump_file,
2071 "Cross jumping from bb %i to bb %i; %i common insns\n",
2072 src1->index, src2->index, nmatch);
2074 /* We may have some registers visible through the block. */
2075 df_set_bb_dirty (redirect_to);
2077 if (osrc2 == src2)
2078 redirect_edges_to = redirect_to;
2079 else
2080 redirect_edges_to = osrc2;
2082 /* Recompute the counts of destinations of outgoing edges. */
2083 FOR_EACH_EDGE (s, ei, redirect_edges_to->succs)
2085 edge s2;
2086 edge_iterator ei;
2087 basic_block d = s->dest;
2089 if (FORWARDER_BLOCK_P (d))
2090 d = single_succ (d);
2092 FOR_EACH_EDGE (s2, ei, src1->succs)
2094 basic_block d2 = s2->dest;
2095 if (FORWARDER_BLOCK_P (d2))
2096 d2 = single_succ (d2);
2097 if (d == d2)
2098 break;
2101 /* Take care to update possible forwarder blocks. We verified
2102 that there is no more than one in the chain, so we can't run
2103 into infinite loop. */
2104 if (FORWARDER_BLOCK_P (s->dest))
2105 s->dest->count += s->count ();
2107 if (FORWARDER_BLOCK_P (s2->dest))
2108 s2->dest->count -= s->count ();
2110 s->probability = s->probability.combine_with_count
2111 (redirect_edges_to->count,
2112 s2->probability, src1->count);
2115 /* Adjust count for the block. An earlier jump
2116 threading pass may have left the profile in an inconsistent
2117 state (see update_bb_profile_for_threading) so we must be
2118 prepared for overflows. */
2119 tmp = redirect_to;
2122 tmp->count += src1->count;
2123 if (tmp == redirect_edges_to)
2124 break;
2125 tmp = find_fallthru_edge (tmp->succs)->dest;
2127 while (true);
2128 update_br_prob_note (redirect_edges_to);
2130 /* Edit SRC1 to go to REDIRECT_TO at NEWPOS1. */
2132 /* Skip possible basic block header. */
2133 if (LABEL_P (newpos1))
2134 newpos1 = NEXT_INSN (newpos1);
2136 while (DEBUG_INSN_P (newpos1))
2137 newpos1 = NEXT_INSN (newpos1);
2139 if (NOTE_INSN_BASIC_BLOCK_P (newpos1))
2140 newpos1 = NEXT_INSN (newpos1);
2142 while (DEBUG_INSN_P (newpos1))
2143 newpos1 = NEXT_INSN (newpos1);
2145 redirect_from = split_block (src1, PREV_INSN (newpos1))->src;
2146 to_remove = single_succ (redirect_from);
2148 redirect_edge_and_branch_force (single_succ_edge (redirect_from), redirect_to);
2149 delete_basic_block (to_remove);
2151 update_forwarder_flag (redirect_from);
2152 if (redirect_to != src2)
2153 update_forwarder_flag (src2);
2155 return true;
2158 /* Search the predecessors of BB for common insn sequences. When found,
2159 share code between them by redirecting control flow. Return true if
2160 any changes made. */
2162 static bool
2163 try_crossjump_bb (int mode, basic_block bb)
2165 edge e, e2, fallthru;
2166 bool changed;
2167 unsigned max, ix, ix2;
2169 /* Nothing to do if there is not at least two incoming edges. */
2170 if (EDGE_COUNT (bb->preds) < 2)
2171 return false;
2173 /* Don't crossjump if this block ends in a computed jump,
2174 unless we are optimizing for size. */
2175 if (optimize_bb_for_size_p (bb)
2176 && bb != EXIT_BLOCK_PTR_FOR_FN (cfun)
2177 && computed_jump_p (BB_END (bb)))
2178 return false;
2180 /* If we are partitioning hot/cold basic blocks, we don't want to
2181 mess up unconditional or indirect jumps that cross between hot
2182 and cold sections.
2184 Basic block partitioning may result in some jumps that appear to
2185 be optimizable (or blocks that appear to be mergeable), but which really
2186 must be left untouched (they are required to make it safely across
2187 partition boundaries). See the comments at the top of
2188 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
2190 if (BB_PARTITION (EDGE_PRED (bb, 0)->src) !=
2191 BB_PARTITION (EDGE_PRED (bb, 1)->src)
2192 || (EDGE_PRED (bb, 0)->flags & EDGE_CROSSING))
2193 return false;
2195 /* It is always cheapest to redirect a block that ends in a branch to
2196 a block that falls through into BB, as that adds no branches to the
2197 program. We'll try that combination first. */
2198 fallthru = NULL;
2199 max = PARAM_VALUE (PARAM_MAX_CROSSJUMP_EDGES);
2201 if (EDGE_COUNT (bb->preds) > max)
2202 return false;
2204 fallthru = find_fallthru_edge (bb->preds);
2206 changed = false;
2207 for (ix = 0; ix < EDGE_COUNT (bb->preds);)
2209 e = EDGE_PRED (bb, ix);
2210 ix++;
2212 /* As noted above, first try with the fallthru predecessor (or, a
2213 fallthru predecessor if we are in cfglayout mode). */
2214 if (fallthru)
2216 /* Don't combine the fallthru edge into anything else.
2217 If there is a match, we'll do it the other way around. */
2218 if (e == fallthru)
2219 continue;
2220 /* If nothing changed since the last attempt, there is nothing
2221 we can do. */
2222 if (!first_pass
2223 && !((e->src->flags & BB_MODIFIED)
2224 || (fallthru->src->flags & BB_MODIFIED)))
2225 continue;
2227 if (try_crossjump_to_edge (mode, e, fallthru, dir_forward))
2229 changed = true;
2230 ix = 0;
2231 continue;
2235 /* Non-obvious work limiting check: Recognize that we're going
2236 to call try_crossjump_bb on every basic block. So if we have
2237 two blocks with lots of outgoing edges (a switch) and they
2238 share lots of common destinations, then we would do the
2239 cross-jump check once for each common destination.
2241 Now, if the blocks actually are cross-jump candidates, then
2242 all of their destinations will be shared. Which means that
2243 we only need check them for cross-jump candidacy once. We
2244 can eliminate redundant checks of crossjump(A,B) by arbitrarily
2245 choosing to do the check from the block for which the edge
2246 in question is the first successor of A. */
2247 if (EDGE_SUCC (e->src, 0) != e)
2248 continue;
2250 for (ix2 = 0; ix2 < EDGE_COUNT (bb->preds); ix2++)
2252 e2 = EDGE_PRED (bb, ix2);
2254 if (e2 == e)
2255 continue;
2257 /* We've already checked the fallthru edge above. */
2258 if (e2 == fallthru)
2259 continue;
2261 /* The "first successor" check above only prevents multiple
2262 checks of crossjump(A,B). In order to prevent redundant
2263 checks of crossjump(B,A), require that A be the block
2264 with the lowest index. */
2265 if (e->src->index > e2->src->index)
2266 continue;
2268 /* If nothing changed since the last attempt, there is nothing
2269 we can do. */
2270 if (!first_pass
2271 && !((e->src->flags & BB_MODIFIED)
2272 || (e2->src->flags & BB_MODIFIED)))
2273 continue;
2275 /* Both e and e2 are not fallthru edges, so we can crossjump in either
2276 direction. */
2277 if (try_crossjump_to_edge (mode, e, e2, dir_both))
2279 changed = true;
2280 ix = 0;
2281 break;
2286 if (changed)
2287 crossjumps_occurred = true;
2289 return changed;
2292 /* Search the successors of BB for common insn sequences. When found,
2293 share code between them by moving it across the basic block
2294 boundary. Return true if any changes made. */
2296 static bool
2297 try_head_merge_bb (basic_block bb)
2299 basic_block final_dest_bb = NULL;
2300 int max_match = INT_MAX;
2301 edge e0;
2302 rtx_insn **headptr, **currptr, **nextptr;
2303 bool changed, moveall;
2304 unsigned ix;
2305 rtx_insn *e0_last_head;
2306 rtx cond;
2307 rtx_insn *move_before;
2308 unsigned nedges = EDGE_COUNT (bb->succs);
2309 rtx_insn *jump = BB_END (bb);
2310 regset live, live_union;
2312 /* Nothing to do if there is not at least two outgoing edges. */
2313 if (nedges < 2)
2314 return false;
2316 /* Don't crossjump if this block ends in a computed jump,
2317 unless we are optimizing for size. */
2318 if (optimize_bb_for_size_p (bb)
2319 && bb != EXIT_BLOCK_PTR_FOR_FN (cfun)
2320 && computed_jump_p (BB_END (bb)))
2321 return false;
2323 cond = get_condition (jump, &move_before, true, false);
2324 if (cond == NULL_RTX)
2326 if (HAVE_cc0 && reg_mentioned_p (cc0_rtx, jump))
2327 move_before = prev_nonnote_nondebug_insn (jump);
2328 else
2329 move_before = jump;
2332 for (ix = 0; ix < nedges; ix++)
2333 if (EDGE_SUCC (bb, ix)->dest == EXIT_BLOCK_PTR_FOR_FN (cfun))
2334 return false;
2336 for (ix = 0; ix < nedges; ix++)
2338 edge e = EDGE_SUCC (bb, ix);
2339 basic_block other_bb = e->dest;
2341 if (df_get_bb_dirty (other_bb))
2343 block_was_dirty = true;
2344 return false;
2347 if (e->flags & EDGE_ABNORMAL)
2348 return false;
2350 /* Normally, all destination blocks must only be reachable from this
2351 block, i.e. they must have one incoming edge.
2353 There is one special case we can handle, that of multiple consecutive
2354 jumps where the first jumps to one of the targets of the second jump.
2355 This happens frequently in switch statements for default labels.
2356 The structure is as follows:
2357 FINAL_DEST_BB
2358 ....
2359 if (cond) jump A;
2360 fall through
2362 jump with targets A, B, C, D...
2364 has two incoming edges, from FINAL_DEST_BB and BB
2366 In this case, we can try to move the insns through BB and into
2367 FINAL_DEST_BB. */
2368 if (EDGE_COUNT (other_bb->preds) != 1)
2370 edge incoming_edge, incoming_bb_other_edge;
2371 edge_iterator ei;
2373 if (final_dest_bb != NULL
2374 || EDGE_COUNT (other_bb->preds) != 2)
2375 return false;
2377 /* We must be able to move the insns across the whole block. */
2378 move_before = BB_HEAD (bb);
2379 while (!NONDEBUG_INSN_P (move_before))
2380 move_before = NEXT_INSN (move_before);
2382 if (EDGE_COUNT (bb->preds) != 1)
2383 return false;
2384 incoming_edge = EDGE_PRED (bb, 0);
2385 final_dest_bb = incoming_edge->src;
2386 if (EDGE_COUNT (final_dest_bb->succs) != 2)
2387 return false;
2388 FOR_EACH_EDGE (incoming_bb_other_edge, ei, final_dest_bb->succs)
2389 if (incoming_bb_other_edge != incoming_edge)
2390 break;
2391 if (incoming_bb_other_edge->dest != other_bb)
2392 return false;
2396 e0 = EDGE_SUCC (bb, 0);
2397 e0_last_head = NULL;
2398 changed = false;
2400 for (ix = 1; ix < nedges; ix++)
2402 edge e = EDGE_SUCC (bb, ix);
2403 rtx_insn *e0_last, *e_last;
2404 int nmatch;
2406 nmatch = flow_find_head_matching_sequence (e0->dest, e->dest,
2407 &e0_last, &e_last, 0);
2408 if (nmatch == 0)
2409 return false;
2411 if (nmatch < max_match)
2413 max_match = nmatch;
2414 e0_last_head = e0_last;
2418 /* If we matched an entire block, we probably have to avoid moving the
2419 last insn. */
2420 if (max_match > 0
2421 && e0_last_head == BB_END (e0->dest)
2422 && (find_reg_note (e0_last_head, REG_EH_REGION, 0)
2423 || control_flow_insn_p (e0_last_head)))
2425 max_match--;
2426 if (max_match == 0)
2427 return false;
2428 e0_last_head = prev_real_nondebug_insn (e0_last_head);
2431 if (max_match == 0)
2432 return false;
2434 /* We must find a union of the live registers at each of the end points. */
2435 live = BITMAP_ALLOC (NULL);
2436 live_union = BITMAP_ALLOC (NULL);
2438 currptr = XNEWVEC (rtx_insn *, nedges);
2439 headptr = XNEWVEC (rtx_insn *, nedges);
2440 nextptr = XNEWVEC (rtx_insn *, nedges);
2442 for (ix = 0; ix < nedges; ix++)
2444 int j;
2445 basic_block merge_bb = EDGE_SUCC (bb, ix)->dest;
2446 rtx_insn *head = BB_HEAD (merge_bb);
2448 while (!NONDEBUG_INSN_P (head))
2449 head = NEXT_INSN (head);
2450 headptr[ix] = head;
2451 currptr[ix] = head;
2453 /* Compute the end point and live information */
2454 for (j = 1; j < max_match; j++)
2456 head = NEXT_INSN (head);
2457 while (!NONDEBUG_INSN_P (head));
2458 simulate_backwards_to_point (merge_bb, live, head);
2459 IOR_REG_SET (live_union, live);
2462 /* If we're moving across two blocks, verify the validity of the
2463 first move, then adjust the target and let the loop below deal
2464 with the final move. */
2465 if (final_dest_bb != NULL)
2467 rtx_insn *move_upto;
2469 moveall = can_move_insns_across (currptr[0], e0_last_head, move_before,
2470 jump, e0->dest, live_union,
2471 NULL, &move_upto);
2472 if (!moveall)
2474 if (move_upto == NULL_RTX)
2475 goto out;
2477 while (e0_last_head != move_upto)
2479 df_simulate_one_insn_backwards (e0->dest, e0_last_head,
2480 live_union);
2481 e0_last_head = PREV_INSN (e0_last_head);
2484 if (e0_last_head == NULL_RTX)
2485 goto out;
2487 jump = BB_END (final_dest_bb);
2488 cond = get_condition (jump, &move_before, true, false);
2489 if (cond == NULL_RTX)
2491 if (HAVE_cc0 && reg_mentioned_p (cc0_rtx, jump))
2492 move_before = prev_nonnote_nondebug_insn (jump);
2493 else
2494 move_before = jump;
2500 rtx_insn *move_upto;
2501 moveall = can_move_insns_across (currptr[0], e0_last_head,
2502 move_before, jump, e0->dest, live_union,
2503 NULL, &move_upto);
2504 if (!moveall && move_upto == NULL_RTX)
2506 if (jump == move_before)
2507 break;
2509 /* Try again, using a different insertion point. */
2510 move_before = jump;
2512 /* Don't try moving before a cc0 user, as that may invalidate
2513 the cc0. */
2514 if (HAVE_cc0 && reg_mentioned_p (cc0_rtx, jump))
2515 break;
2517 continue;
2520 if (final_dest_bb && !moveall)
2521 /* We haven't checked whether a partial move would be OK for the first
2522 move, so we have to fail this case. */
2523 break;
2525 changed = true;
2526 for (;;)
2528 if (currptr[0] == move_upto)
2529 break;
2530 for (ix = 0; ix < nedges; ix++)
2532 rtx_insn *curr = currptr[ix];
2534 curr = NEXT_INSN (curr);
2535 while (!NONDEBUG_INSN_P (curr));
2536 currptr[ix] = curr;
2540 /* If we can't currently move all of the identical insns, remember
2541 each insn after the range that we'll merge. */
2542 if (!moveall)
2543 for (ix = 0; ix < nedges; ix++)
2545 rtx_insn *curr = currptr[ix];
2547 curr = NEXT_INSN (curr);
2548 while (!NONDEBUG_INSN_P (curr));
2549 nextptr[ix] = curr;
2552 reorder_insns (headptr[0], currptr[0], PREV_INSN (move_before));
2553 df_set_bb_dirty (EDGE_SUCC (bb, 0)->dest);
2554 if (final_dest_bb != NULL)
2555 df_set_bb_dirty (final_dest_bb);
2556 df_set_bb_dirty (bb);
2557 for (ix = 1; ix < nedges; ix++)
2559 df_set_bb_dirty (EDGE_SUCC (bb, ix)->dest);
2560 delete_insn_chain (headptr[ix], currptr[ix], false);
2562 if (!moveall)
2564 if (jump == move_before)
2565 break;
2567 /* For the unmerged insns, try a different insertion point. */
2568 move_before = jump;
2570 /* Don't try moving before a cc0 user, as that may invalidate
2571 the cc0. */
2572 if (HAVE_cc0 && reg_mentioned_p (cc0_rtx, jump))
2573 break;
2575 for (ix = 0; ix < nedges; ix++)
2576 currptr[ix] = headptr[ix] = nextptr[ix];
2579 while (!moveall);
2581 out:
2582 free (currptr);
2583 free (headptr);
2584 free (nextptr);
2586 crossjumps_occurred |= changed;
2588 return changed;
2591 /* Return true if BB contains just bb note, or bb note followed
2592 by only DEBUG_INSNs. */
2594 static bool
2595 trivially_empty_bb_p (basic_block bb)
2597 rtx_insn *insn = BB_END (bb);
2599 while (1)
2601 if (insn == BB_HEAD (bb))
2602 return true;
2603 if (!DEBUG_INSN_P (insn))
2604 return false;
2605 insn = PREV_INSN (insn);
2609 /* Return true if BB contains just a return and possibly a USE of the
2610 return value. Fill in *RET and *USE with the return and use insns
2611 if any found, otherwise NULL. All CLOBBERs are ignored. */
2613 static bool
2614 bb_is_just_return (basic_block bb, rtx_insn **ret, rtx_insn **use)
2616 *ret = *use = NULL;
2617 rtx_insn *insn;
2619 if (bb == EXIT_BLOCK_PTR_FOR_FN (cfun))
2620 return false;
2622 FOR_BB_INSNS (bb, insn)
2623 if (NONDEBUG_INSN_P (insn))
2625 rtx pat = PATTERN (insn);
2627 if (!*ret && ANY_RETURN_P (pat))
2628 *ret = insn;
2629 else if (!*ret && !*use && GET_CODE (pat) == USE
2630 && REG_P (XEXP (pat, 0))
2631 && REG_FUNCTION_VALUE_P (XEXP (pat, 0)))
2632 *use = insn;
2633 else if (GET_CODE (pat) != CLOBBER)
2634 return false;
2637 return !!*ret;
2640 /* Do simple CFG optimizations - basic block merging, simplifying of jump
2641 instructions etc. Return nonzero if changes were made. */
2643 static bool
2644 try_optimize_cfg (int mode)
2646 bool changed_overall = false;
2647 bool changed;
2648 int iterations = 0;
2649 basic_block bb, b, next;
2651 if (mode & (CLEANUP_CROSSJUMP | CLEANUP_THREADING))
2652 clear_bb_flags ();
2654 crossjumps_occurred = false;
2656 FOR_EACH_BB_FN (bb, cfun)
2657 update_forwarder_flag (bb);
2659 if (! targetm.cannot_modify_jumps_p ())
2661 first_pass = true;
2662 /* Attempt to merge blocks as made possible by edge removal. If
2663 a block has only one successor, and the successor has only
2664 one predecessor, they may be combined. */
2667 block_was_dirty = false;
2668 changed = false;
2669 iterations++;
2671 if (dump_file)
2672 fprintf (dump_file,
2673 "\n\ntry_optimize_cfg iteration %i\n\n",
2674 iterations);
2676 for (b = ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb; b
2677 != EXIT_BLOCK_PTR_FOR_FN (cfun);)
2679 basic_block c;
2680 edge s;
2681 bool changed_here = false;
2683 /* Delete trivially dead basic blocks. This is either
2684 blocks with no predecessors, or empty blocks with no
2685 successors. However if the empty block with no
2686 successors is the successor of the ENTRY_BLOCK, it is
2687 kept. This ensures that the ENTRY_BLOCK will have a
2688 successor which is a precondition for many RTL
2689 passes. Empty blocks may result from expanding
2690 __builtin_unreachable (). */
2691 if (EDGE_COUNT (b->preds) == 0
2692 || (EDGE_COUNT (b->succs) == 0
2693 && trivially_empty_bb_p (b)
2694 && single_succ_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun))->dest
2695 != b))
2697 c = b->prev_bb;
2698 if (EDGE_COUNT (b->preds) > 0)
2700 edge e;
2701 edge_iterator ei;
2703 if (current_ir_type () == IR_RTL_CFGLAYOUT)
2705 if (BB_FOOTER (b)
2706 && BARRIER_P (BB_FOOTER (b)))
2707 FOR_EACH_EDGE (e, ei, b->preds)
2708 if ((e->flags & EDGE_FALLTHRU)
2709 && BB_FOOTER (e->src) == NULL)
2711 if (BB_FOOTER (b))
2713 BB_FOOTER (e->src) = BB_FOOTER (b);
2714 BB_FOOTER (b) = NULL;
2716 else
2718 start_sequence ();
2719 BB_FOOTER (e->src) = emit_barrier ();
2720 end_sequence ();
2724 else
2726 rtx_insn *last = get_last_bb_insn (b);
2727 if (last && BARRIER_P (last))
2728 FOR_EACH_EDGE (e, ei, b->preds)
2729 if ((e->flags & EDGE_FALLTHRU))
2730 emit_barrier_after (BB_END (e->src));
2733 delete_basic_block (b);
2734 changed = true;
2735 /* Avoid trying to remove the exit block. */
2736 b = (c == ENTRY_BLOCK_PTR_FOR_FN (cfun) ? c->next_bb : c);
2737 continue;
2740 /* Remove code labels no longer used. */
2741 if (single_pred_p (b)
2742 && (single_pred_edge (b)->flags & EDGE_FALLTHRU)
2743 && !(single_pred_edge (b)->flags & EDGE_COMPLEX)
2744 && LABEL_P (BB_HEAD (b))
2745 && !LABEL_PRESERVE_P (BB_HEAD (b))
2746 /* If the previous block ends with a branch to this
2747 block, we can't delete the label. Normally this
2748 is a condjump that is yet to be simplified, but
2749 if CASE_DROPS_THRU, this can be a tablejump with
2750 some element going to the same place as the
2751 default (fallthru). */
2752 && (single_pred (b) == ENTRY_BLOCK_PTR_FOR_FN (cfun)
2753 || !JUMP_P (BB_END (single_pred (b)))
2754 || ! label_is_jump_target_p (BB_HEAD (b),
2755 BB_END (single_pred (b)))))
2757 delete_insn (BB_HEAD (b));
2758 if (dump_file)
2759 fprintf (dump_file, "Deleted label in block %i.\n",
2760 b->index);
2763 /* If we fall through an empty block, we can remove it. */
2764 if (!(mode & (CLEANUP_CFGLAYOUT | CLEANUP_NO_INSN_DEL))
2765 && single_pred_p (b)
2766 && (single_pred_edge (b)->flags & EDGE_FALLTHRU)
2767 && !LABEL_P (BB_HEAD (b))
2768 && FORWARDER_BLOCK_P (b)
2769 /* Note that forwarder_block_p true ensures that
2770 there is a successor for this block. */
2771 && (single_succ_edge (b)->flags & EDGE_FALLTHRU)
2772 && n_basic_blocks_for_fn (cfun) > NUM_FIXED_BLOCKS + 1)
2774 if (dump_file)
2775 fprintf (dump_file,
2776 "Deleting fallthru block %i.\n",
2777 b->index);
2779 c = ((b->prev_bb == ENTRY_BLOCK_PTR_FOR_FN (cfun))
2780 ? b->next_bb : b->prev_bb);
2781 redirect_edge_succ_nodup (single_pred_edge (b),
2782 single_succ (b));
2783 delete_basic_block (b);
2784 changed = true;
2785 b = c;
2786 continue;
2789 /* Merge B with its single successor, if any. */
2790 if (single_succ_p (b)
2791 && (s = single_succ_edge (b))
2792 && !(s->flags & EDGE_COMPLEX)
2793 && (c = s->dest) != EXIT_BLOCK_PTR_FOR_FN (cfun)
2794 && single_pred_p (c)
2795 && b != c)
2797 /* When not in cfg_layout mode use code aware of reordering
2798 INSN. This code possibly creates new basic blocks so it
2799 does not fit merge_blocks interface and is kept here in
2800 hope that it will become useless once more of compiler
2801 is transformed to use cfg_layout mode. */
2803 if ((mode & CLEANUP_CFGLAYOUT)
2804 && can_merge_blocks_p (b, c))
2806 merge_blocks (b, c);
2807 update_forwarder_flag (b);
2808 changed_here = true;
2810 else if (!(mode & CLEANUP_CFGLAYOUT)
2811 /* If the jump insn has side effects,
2812 we can't kill the edge. */
2813 && (!JUMP_P (BB_END (b))
2814 || (reload_completed
2815 ? simplejump_p (BB_END (b))
2816 : (onlyjump_p (BB_END (b))
2817 && !tablejump_p (BB_END (b),
2818 NULL, NULL))))
2819 && (next = merge_blocks_move (s, b, c, mode)))
2821 b = next;
2822 changed_here = true;
2826 /* Try to change a branch to a return to just that return. */
2827 rtx_insn *ret, *use;
2828 if (single_succ_p (b)
2829 && onlyjump_p (BB_END (b))
2830 && bb_is_just_return (single_succ (b), &ret, &use))
2832 if (redirect_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2833 PATTERN (ret), 0))
2835 if (use)
2836 emit_insn_before (copy_insn (PATTERN (use)),
2837 BB_END (b));
2838 if (dump_file)
2839 fprintf (dump_file, "Changed jump %d->%d to return.\n",
2840 b->index, single_succ (b)->index);
2841 redirect_edge_succ (single_succ_edge (b),
2842 EXIT_BLOCK_PTR_FOR_FN (cfun));
2843 single_succ_edge (b)->flags &= ~EDGE_CROSSING;
2844 changed_here = true;
2848 /* Try to change a conditional branch to a return to the
2849 respective conditional return. */
2850 if (EDGE_COUNT (b->succs) == 2
2851 && any_condjump_p (BB_END (b))
2852 && bb_is_just_return (BRANCH_EDGE (b)->dest, &ret, &use))
2854 if (redirect_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2855 PATTERN (ret), 0))
2857 if (use)
2858 emit_insn_before (copy_insn (PATTERN (use)),
2859 BB_END (b));
2860 if (dump_file)
2861 fprintf (dump_file, "Changed conditional jump %d->%d "
2862 "to conditional return.\n",
2863 b->index, BRANCH_EDGE (b)->dest->index);
2864 redirect_edge_succ (BRANCH_EDGE (b),
2865 EXIT_BLOCK_PTR_FOR_FN (cfun));
2866 BRANCH_EDGE (b)->flags &= ~EDGE_CROSSING;
2867 changed_here = true;
2871 /* Try to flip a conditional branch that falls through to
2872 a return so that it becomes a conditional return and a
2873 new jump to the original branch target. */
2874 if (EDGE_COUNT (b->succs) == 2
2875 && BRANCH_EDGE (b)->dest != EXIT_BLOCK_PTR_FOR_FN (cfun)
2876 && any_condjump_p (BB_END (b))
2877 && bb_is_just_return (FALLTHRU_EDGE (b)->dest, &ret, &use))
2879 if (invert_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2880 JUMP_LABEL (BB_END (b)), 0))
2882 basic_block new_ft = BRANCH_EDGE (b)->dest;
2883 if (redirect_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2884 PATTERN (ret), 0))
2886 if (use)
2887 emit_insn_before (copy_insn (PATTERN (use)),
2888 BB_END (b));
2889 if (dump_file)
2890 fprintf (dump_file, "Changed conditional jump "
2891 "%d->%d to conditional return, adding "
2892 "fall-through jump.\n",
2893 b->index, BRANCH_EDGE (b)->dest->index);
2894 redirect_edge_succ (BRANCH_EDGE (b),
2895 EXIT_BLOCK_PTR_FOR_FN (cfun));
2896 BRANCH_EDGE (b)->flags &= ~EDGE_CROSSING;
2897 std::swap (BRANCH_EDGE (b)->probability,
2898 FALLTHRU_EDGE (b)->probability);
2899 update_br_prob_note (b);
2900 basic_block jb = force_nonfallthru (FALLTHRU_EDGE (b));
2901 notice_new_block (jb);
2902 if (!redirect_jump (as_a <rtx_jump_insn *> (BB_END (jb)),
2903 block_label (new_ft), 0))
2904 gcc_unreachable ();
2905 redirect_edge_succ (single_succ_edge (jb), new_ft);
2906 changed_here = true;
2908 else
2910 /* Invert the jump back to what it was. This should
2911 never fail. */
2912 if (!invert_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2913 JUMP_LABEL (BB_END (b)), 0))
2914 gcc_unreachable ();
2919 /* Simplify branch over branch. */
2920 if ((mode & CLEANUP_EXPENSIVE)
2921 && !(mode & CLEANUP_CFGLAYOUT)
2922 && try_simplify_condjump (b))
2923 changed_here = true;
2925 /* If B has a single outgoing edge, but uses a
2926 non-trivial jump instruction without side-effects, we
2927 can either delete the jump entirely, or replace it
2928 with a simple unconditional jump. */
2929 if (single_succ_p (b)
2930 && single_succ (b) != EXIT_BLOCK_PTR_FOR_FN (cfun)
2931 && onlyjump_p (BB_END (b))
2932 && !CROSSING_JUMP_P (BB_END (b))
2933 && try_redirect_by_replacing_jump (single_succ_edge (b),
2934 single_succ (b),
2935 (mode & CLEANUP_CFGLAYOUT) != 0))
2937 update_forwarder_flag (b);
2938 changed_here = true;
2941 /* Simplify branch to branch. */
2942 if (try_forward_edges (mode, b))
2944 update_forwarder_flag (b);
2945 changed_here = true;
2948 /* Look for shared code between blocks. */
2949 if ((mode & CLEANUP_CROSSJUMP)
2950 && try_crossjump_bb (mode, b))
2951 changed_here = true;
2953 if ((mode & CLEANUP_CROSSJUMP)
2954 /* This can lengthen register lifetimes. Do it only after
2955 reload. */
2956 && reload_completed
2957 && try_head_merge_bb (b))
2958 changed_here = true;
2960 /* Don't get confused by the index shift caused by
2961 deleting blocks. */
2962 if (!changed_here)
2963 b = b->next_bb;
2964 else
2965 changed = true;
2968 if ((mode & CLEANUP_CROSSJUMP)
2969 && try_crossjump_bb (mode, EXIT_BLOCK_PTR_FOR_FN (cfun)))
2970 changed = true;
2972 if (block_was_dirty)
2974 /* This should only be set by head-merging. */
2975 gcc_assert (mode & CLEANUP_CROSSJUMP);
2976 df_analyze ();
2979 if (changed)
2981 /* Edge forwarding in particular can cause hot blocks previously
2982 reached by both hot and cold blocks to become dominated only
2983 by cold blocks. This will cause the verification below to fail,
2984 and lead to now cold code in the hot section. This is not easy
2985 to detect and fix during edge forwarding, and in some cases
2986 is only visible after newly unreachable blocks are deleted,
2987 which will be done in fixup_partitions. */
2988 if ((mode & CLEANUP_NO_PARTITIONING) == 0)
2990 fixup_partitions ();
2991 checking_verify_flow_info ();
2995 changed_overall |= changed;
2996 first_pass = false;
2998 while (changed);
3001 FOR_ALL_BB_FN (b, cfun)
3002 b->flags &= ~(BB_FORWARDER_BLOCK | BB_NONTHREADABLE_BLOCK);
3004 return changed_overall;
3007 /* Delete all unreachable basic blocks. */
3009 bool
3010 delete_unreachable_blocks (void)
3012 bool changed = false;
3013 basic_block b, prev_bb;
3015 find_unreachable_blocks ();
3017 /* When we're in GIMPLE mode and there may be debug bind insns, we
3018 should delete blocks in reverse dominator order, so as to get a
3019 chance to substitute all released DEFs into debug bind stmts. If
3020 we don't have dominators information, walking blocks backward
3021 gets us a better chance of retaining most debug information than
3022 otherwise. */
3023 if (MAY_HAVE_DEBUG_BIND_INSNS && current_ir_type () == IR_GIMPLE
3024 && dom_info_available_p (CDI_DOMINATORS))
3026 for (b = EXIT_BLOCK_PTR_FOR_FN (cfun)->prev_bb;
3027 b != ENTRY_BLOCK_PTR_FOR_FN (cfun); b = prev_bb)
3029 prev_bb = b->prev_bb;
3031 if (!(b->flags & BB_REACHABLE))
3033 /* Speed up the removal of blocks that don't dominate
3034 others. Walking backwards, this should be the common
3035 case. */
3036 if (!first_dom_son (CDI_DOMINATORS, b))
3037 delete_basic_block (b);
3038 else
3040 vec<basic_block> h
3041 = get_all_dominated_blocks (CDI_DOMINATORS, b);
3043 while (h.length ())
3045 b = h.pop ();
3047 prev_bb = b->prev_bb;
3049 gcc_assert (!(b->flags & BB_REACHABLE));
3051 delete_basic_block (b);
3054 h.release ();
3057 changed = true;
3061 else
3063 for (b = EXIT_BLOCK_PTR_FOR_FN (cfun)->prev_bb;
3064 b != ENTRY_BLOCK_PTR_FOR_FN (cfun); b = prev_bb)
3066 prev_bb = b->prev_bb;
3068 if (!(b->flags & BB_REACHABLE))
3070 delete_basic_block (b);
3071 changed = true;
3076 if (changed)
3077 tidy_fallthru_edges ();
3078 return changed;
3081 /* Delete any jump tables never referenced. We can't delete them at the
3082 time of removing tablejump insn as they are referenced by the preceding
3083 insns computing the destination, so we delay deleting and garbagecollect
3084 them once life information is computed. */
3085 void
3086 delete_dead_jumptables (void)
3088 basic_block bb;
3090 /* A dead jump table does not belong to any basic block. Scan insns
3091 between two adjacent basic blocks. */
3092 FOR_EACH_BB_FN (bb, cfun)
3094 rtx_insn *insn, *next;
3096 for (insn = NEXT_INSN (BB_END (bb));
3097 insn && !NOTE_INSN_BASIC_BLOCK_P (insn);
3098 insn = next)
3100 next = NEXT_INSN (insn);
3101 if (LABEL_P (insn)
3102 && LABEL_NUSES (insn) == LABEL_PRESERVE_P (insn)
3103 && JUMP_TABLE_DATA_P (next))
3105 rtx_insn *label = insn, *jump = next;
3107 if (dump_file)
3108 fprintf (dump_file, "Dead jumptable %i removed\n",
3109 INSN_UID (insn));
3111 next = NEXT_INSN (next);
3112 delete_insn (jump);
3113 delete_insn (label);
3120 /* Tidy the CFG by deleting unreachable code and whatnot. */
3122 bool
3123 cleanup_cfg (int mode)
3125 bool changed = false;
3127 /* Set the cfglayout mode flag here. We could update all the callers
3128 but that is just inconvenient, especially given that we eventually
3129 want to have cfglayout mode as the default. */
3130 if (current_ir_type () == IR_RTL_CFGLAYOUT)
3131 mode |= CLEANUP_CFGLAYOUT;
3133 timevar_push (TV_CLEANUP_CFG);
3134 if (delete_unreachable_blocks ())
3136 changed = true;
3137 /* We've possibly created trivially dead code. Cleanup it right
3138 now to introduce more opportunities for try_optimize_cfg. */
3139 if (!(mode & (CLEANUP_NO_INSN_DEL))
3140 && !reload_completed)
3141 delete_trivially_dead_insns (get_insns (), max_reg_num ());
3144 compact_blocks ();
3146 /* To tail-merge blocks ending in the same noreturn function (e.g.
3147 a call to abort) we have to insert fake edges to exit. Do this
3148 here once. The fake edges do not interfere with any other CFG
3149 cleanups. */
3150 if (mode & CLEANUP_CROSSJUMP)
3151 add_noreturn_fake_exit_edges ();
3153 if (!dbg_cnt (cfg_cleanup))
3154 return changed;
3156 while (try_optimize_cfg (mode))
3158 delete_unreachable_blocks (), changed = true;
3159 if (!(mode & CLEANUP_NO_INSN_DEL))
3161 /* Try to remove some trivially dead insns when doing an expensive
3162 cleanup. But delete_trivially_dead_insns doesn't work after
3163 reload (it only handles pseudos) and run_fast_dce is too costly
3164 to run in every iteration.
3166 For effective cross jumping, we really want to run a fast DCE to
3167 clean up any dead conditions, or they get in the way of performing
3168 useful tail merges.
3170 Other transformations in cleanup_cfg are not so sensitive to dead
3171 code, so delete_trivially_dead_insns or even doing nothing at all
3172 is good enough. */
3173 if ((mode & CLEANUP_EXPENSIVE) && !reload_completed
3174 && !delete_trivially_dead_insns (get_insns (), max_reg_num ()))
3175 break;
3176 if ((mode & CLEANUP_CROSSJUMP) && crossjumps_occurred)
3177 run_fast_dce ();
3179 else
3180 break;
3183 if (mode & CLEANUP_CROSSJUMP)
3184 remove_fake_exit_edges ();
3186 /* Don't call delete_dead_jumptables in cfglayout mode, because
3187 that function assumes that jump tables are in the insns stream.
3188 But we also don't _have_ to delete dead jumptables in cfglayout
3189 mode because we shouldn't even be looking at things that are
3190 not in a basic block. Dead jumptables are cleaned up when
3191 going out of cfglayout mode. */
3192 if (!(mode & CLEANUP_CFGLAYOUT))
3193 delete_dead_jumptables ();
3195 /* ??? We probably do this way too often. */
3196 if (current_loops
3197 && (changed
3198 || (mode & CLEANUP_CFG_CHANGED)))
3200 timevar_push (TV_REPAIR_LOOPS);
3201 /* The above doesn't preserve dominance info if available. */
3202 gcc_assert (!dom_info_available_p (CDI_DOMINATORS));
3203 calculate_dominance_info (CDI_DOMINATORS);
3204 fix_loop_structure (NULL);
3205 free_dominance_info (CDI_DOMINATORS);
3206 timevar_pop (TV_REPAIR_LOOPS);
3209 timevar_pop (TV_CLEANUP_CFG);
3211 return changed;
3214 namespace {
3216 const pass_data pass_data_jump =
3218 RTL_PASS, /* type */
3219 "jump", /* name */
3220 OPTGROUP_NONE, /* optinfo_flags */
3221 TV_JUMP, /* tv_id */
3222 0, /* properties_required */
3223 0, /* properties_provided */
3224 0, /* properties_destroyed */
3225 0, /* todo_flags_start */
3226 0, /* todo_flags_finish */
3229 class pass_jump : public rtl_opt_pass
3231 public:
3232 pass_jump (gcc::context *ctxt)
3233 : rtl_opt_pass (pass_data_jump, ctxt)
3236 /* opt_pass methods: */
3237 virtual unsigned int execute (function *);
3239 }; // class pass_jump
3241 unsigned int
3242 pass_jump::execute (function *)
3244 delete_trivially_dead_insns (get_insns (), max_reg_num ());
3245 if (dump_file)
3246 dump_flow_info (dump_file, dump_flags);
3247 cleanup_cfg ((optimize ? CLEANUP_EXPENSIVE : 0)
3248 | (flag_thread_jumps ? CLEANUP_THREADING : 0));
3249 return 0;
3252 } // anon namespace
3254 rtl_opt_pass *
3255 make_pass_jump (gcc::context *ctxt)
3257 return new pass_jump (ctxt);
3260 namespace {
3262 const pass_data pass_data_jump2 =
3264 RTL_PASS, /* type */
3265 "jump2", /* name */
3266 OPTGROUP_NONE, /* optinfo_flags */
3267 TV_JUMP, /* tv_id */
3268 0, /* properties_required */
3269 0, /* properties_provided */
3270 0, /* properties_destroyed */
3271 0, /* todo_flags_start */
3272 0, /* todo_flags_finish */
3275 class pass_jump2 : public rtl_opt_pass
3277 public:
3278 pass_jump2 (gcc::context *ctxt)
3279 : rtl_opt_pass (pass_data_jump2, ctxt)
3282 /* opt_pass methods: */
3283 virtual unsigned int execute (function *)
3285 cleanup_cfg (flag_crossjumping ? CLEANUP_CROSSJUMP : 0);
3286 return 0;
3289 }; // class pass_jump2
3291 } // anon namespace
3293 rtl_opt_pass *
3294 make_pass_jump2 (gcc::context *ctxt)
3296 return new pass_jump2 (ctxt);