2010-11-23 Tobias Burnus <burnus@net-b.de>
[official-gcc.git] / gcc / cfgcleanup.c
blobdc41f45bd6d0428af4aeb8086442b56396a51176
1 /* Control flow optimization code for GNU compiler.
2 Copyright (C) 1987, 1988, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
3 1999, 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008, 2010
4 Free Software Foundation, Inc.
6 This file is part of GCC.
8 GCC is free software; you can redistribute it and/or modify it under
9 the terms of the GNU General Public License as published by the Free
10 Software Foundation; either version 3, or (at your option) any later
11 version.
13 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 for more details.
18 You should have received a copy of the GNU General Public License
19 along with GCC; see the file COPYING3. If not see
20 <http://www.gnu.org/licenses/>. */
22 /* This file contains optimizer of the control flow. The main entry point is
23 cleanup_cfg. Following optimizations are performed:
25 - Unreachable blocks removal
26 - Edge forwarding (edge to the forwarder block is forwarded to its
27 successor. Simplification of the branch instruction is performed by
28 underlying infrastructure so branch can be converted to simplejump or
29 eliminated).
30 - Cross jumping (tail merging)
31 - Conditional jump-around-simplejump simplification
32 - Basic block merging. */
34 #include "config.h"
35 #include "system.h"
36 #include "coretypes.h"
37 #include "tm.h"
38 #include "rtl.h"
39 #include "hard-reg-set.h"
40 #include "regs.h"
41 #include "timevar.h"
42 #include "output.h"
43 #include "insn-config.h"
44 #include "flags.h"
45 #include "recog.h"
46 #include "diagnostic-core.h"
47 #include "toplev.h"
48 #include "cselib.h"
49 #include "params.h"
50 #include "tm_p.h"
51 #include "target.h"
52 #include "cfglayout.h"
53 #include "emit-rtl.h"
54 #include "tree-pass.h"
55 #include "cfgloop.h"
56 #include "expr.h"
57 #include "df.h"
58 #include "dce.h"
59 #include "dbgcnt.h"
61 #define FORWARDER_BLOCK_P(BB) ((BB)->flags & BB_FORWARDER_BLOCK)
63 /* Set to true when we are running first pass of try_optimize_cfg loop. */
64 static bool first_pass;
66 /* Set to true if crossjumps occured in the latest run of try_optimize_cfg. */
67 static bool crossjumps_occured;
69 static bool try_crossjump_to_edge (int, edge, edge);
70 static bool try_crossjump_bb (int, basic_block);
71 static bool outgoing_edges_match (int, basic_block, basic_block);
72 static bool old_insns_match_p (int, rtx, rtx);
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 int mentions_nonequal_regs (rtx *, void *);
84 static void merge_memattrs (rtx, rtx);
86 /* Set flags for newly created block. */
88 static void
89 notice_new_block (basic_block bb)
91 if (!bb)
92 return;
94 if (forwarder_block_p (bb))
95 bb->flags |= BB_FORWARDER_BLOCK;
98 /* Recompute forwarder flag after block has been modified. */
100 static void
101 update_forwarder_flag (basic_block bb)
103 if (forwarder_block_p (bb))
104 bb->flags |= BB_FORWARDER_BLOCK;
105 else
106 bb->flags &= ~BB_FORWARDER_BLOCK;
109 /* Simplify a conditional jump around an unconditional jump.
110 Return true if something changed. */
112 static bool
113 try_simplify_condjump (basic_block cbranch_block)
115 basic_block jump_block, jump_dest_block, cbranch_dest_block;
116 edge cbranch_jump_edge, cbranch_fallthru_edge;
117 rtx cbranch_insn;
119 /* Verify that there are exactly two successors. */
120 if (EDGE_COUNT (cbranch_block->succs) != 2)
121 return false;
123 /* Verify that we've got a normal conditional branch at the end
124 of the block. */
125 cbranch_insn = BB_END (cbranch_block);
126 if (!any_condjump_p (cbranch_insn))
127 return false;
129 cbranch_fallthru_edge = FALLTHRU_EDGE (cbranch_block);
130 cbranch_jump_edge = BRANCH_EDGE (cbranch_block);
132 /* The next block must not have multiple predecessors, must not
133 be the last block in the function, and must contain just the
134 unconditional jump. */
135 jump_block = cbranch_fallthru_edge->dest;
136 if (!single_pred_p (jump_block)
137 || jump_block->next_bb == EXIT_BLOCK_PTR
138 || !FORWARDER_BLOCK_P (jump_block))
139 return false;
140 jump_dest_block = single_succ (jump_block);
142 /* If we are partitioning hot/cold basic blocks, we don't want to
143 mess up unconditional or indirect jumps that cross between hot
144 and cold sections.
146 Basic block partitioning may result in some jumps that appear to
147 be optimizable (or blocks that appear to be mergeable), but which really
148 must be left untouched (they are required to make it safely across
149 partition boundaries). See the comments at the top of
150 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
152 if (BB_PARTITION (jump_block) != BB_PARTITION (jump_dest_block)
153 || (cbranch_jump_edge->flags & EDGE_CROSSING))
154 return false;
156 /* The conditional branch must target the block after the
157 unconditional branch. */
158 cbranch_dest_block = cbranch_jump_edge->dest;
160 if (cbranch_dest_block == EXIT_BLOCK_PTR
161 || !can_fallthru (jump_block, cbranch_dest_block))
162 return false;
164 /* Invert the conditional branch. */
165 if (!invert_jump (cbranch_insn, block_label (jump_dest_block), 0))
166 return false;
168 if (dump_file)
169 fprintf (dump_file, "Simplifying condjump %i around jump %i\n",
170 INSN_UID (cbranch_insn), INSN_UID (BB_END (jump_block)));
172 /* Success. Update the CFG to match. Note that after this point
173 the edge variable names appear backwards; the redirection is done
174 this way to preserve edge profile data. */
175 cbranch_jump_edge = redirect_edge_succ_nodup (cbranch_jump_edge,
176 cbranch_dest_block);
177 cbranch_fallthru_edge = redirect_edge_succ_nodup (cbranch_fallthru_edge,
178 jump_dest_block);
179 cbranch_jump_edge->flags |= EDGE_FALLTHRU;
180 cbranch_fallthru_edge->flags &= ~EDGE_FALLTHRU;
181 update_br_prob_note (cbranch_block);
183 /* Delete the block with the unconditional jump, and clean up the mess. */
184 delete_basic_block (jump_block);
185 tidy_fallthru_edge (cbranch_jump_edge);
186 update_forwarder_flag (cbranch_block);
188 return true;
191 /* Attempt to prove that operation is NOOP using CSElib or mark the effect
192 on register. Used by jump threading. */
194 static bool
195 mark_effect (rtx exp, regset nonequal)
197 int regno;
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 if (REG_P (XEXP (exp, 0)))
206 dest = XEXP (exp, 0);
207 regno = REGNO (dest);
208 CLEAR_REGNO_REG_SET (nonequal, regno);
209 if (regno < FIRST_PSEUDO_REGISTER)
211 int n = hard_regno_nregs[regno][GET_MODE (dest)];
212 while (--n > 0)
213 CLEAR_REGNO_REG_SET (nonequal, regno + n);
216 return false;
218 case SET:
219 if (rtx_equal_for_cselib_p (SET_DEST (exp), SET_SRC (exp)))
220 return false;
221 dest = SET_DEST (exp);
222 if (dest == pc_rtx)
223 return false;
224 if (!REG_P (dest))
225 return true;
226 regno = REGNO (dest);
227 SET_REGNO_REG_SET (nonequal, regno);
228 if (regno < FIRST_PSEUDO_REGISTER)
230 int n = hard_regno_nregs[regno][GET_MODE (dest)];
231 while (--n > 0)
232 SET_REGNO_REG_SET (nonequal, regno + n);
234 return false;
236 default:
237 return false;
241 /* Return nonzero if X is a register set in regset DATA.
242 Called via for_each_rtx. */
243 static int
244 mentions_nonequal_regs (rtx *x, void *data)
246 regset nonequal = (regset) data;
247 if (REG_P (*x))
249 int regno;
251 regno = REGNO (*x);
252 if (REGNO_REG_SET_P (nonequal, regno))
253 return 1;
254 if (regno < FIRST_PSEUDO_REGISTER)
256 int n = hard_regno_nregs[regno][GET_MODE (*x)];
257 while (--n > 0)
258 if (REGNO_REG_SET_P (nonequal, regno + n))
259 return 1;
262 return 0;
264 /* Attempt to prove that the basic block B will have no side effects and
265 always continues in the same edge if reached via E. Return the edge
266 if exist, NULL otherwise. */
268 static edge
269 thread_jump (edge e, basic_block b)
271 rtx set1, set2, cond1, cond2, insn;
272 enum rtx_code code1, code2, reversed_code2;
273 bool reverse1 = false;
274 unsigned i;
275 regset nonequal;
276 bool failed = false;
277 reg_set_iterator rsi;
279 if (b->flags & BB_NONTHREADABLE_BLOCK)
280 return NULL;
282 /* At the moment, we do handle only conditional jumps, but later we may
283 want to extend this code to tablejumps and others. */
284 if (EDGE_COUNT (e->src->succs) != 2)
285 return NULL;
286 if (EDGE_COUNT (b->succs) != 2)
288 b->flags |= BB_NONTHREADABLE_BLOCK;
289 return NULL;
292 /* Second branch must end with onlyjump, as we will eliminate the jump. */
293 if (!any_condjump_p (BB_END (e->src)))
294 return NULL;
296 if (!any_condjump_p (BB_END (b)) || !onlyjump_p (BB_END (b)))
298 b->flags |= BB_NONTHREADABLE_BLOCK;
299 return NULL;
302 set1 = pc_set (BB_END (e->src));
303 set2 = pc_set (BB_END (b));
304 if (((e->flags & EDGE_FALLTHRU) != 0)
305 != (XEXP (SET_SRC (set1), 1) == pc_rtx))
306 reverse1 = true;
308 cond1 = XEXP (SET_SRC (set1), 0);
309 cond2 = XEXP (SET_SRC (set2), 0);
310 if (reverse1)
311 code1 = reversed_comparison_code (cond1, BB_END (e->src));
312 else
313 code1 = GET_CODE (cond1);
315 code2 = GET_CODE (cond2);
316 reversed_code2 = reversed_comparison_code (cond2, BB_END (b));
318 if (!comparison_dominates_p (code1, code2)
319 && !comparison_dominates_p (code1, reversed_code2))
320 return NULL;
322 /* Ensure that the comparison operators are equivalent.
323 ??? This is far too pessimistic. We should allow swapped operands,
324 different CCmodes, or for example comparisons for interval, that
325 dominate even when operands are not equivalent. */
326 if (!rtx_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
327 || !rtx_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
328 return NULL;
330 /* Short circuit cases where block B contains some side effects, as we can't
331 safely bypass it. */
332 for (insn = NEXT_INSN (BB_HEAD (b)); insn != NEXT_INSN (BB_END (b));
333 insn = NEXT_INSN (insn))
334 if (INSN_P (insn) && side_effects_p (PATTERN (insn)))
336 b->flags |= BB_NONTHREADABLE_BLOCK;
337 return NULL;
340 cselib_init (0);
342 /* First process all values computed in the source basic block. */
343 for (insn = NEXT_INSN (BB_HEAD (e->src));
344 insn != NEXT_INSN (BB_END (e->src));
345 insn = NEXT_INSN (insn))
346 if (INSN_P (insn))
347 cselib_process_insn (insn);
349 nonequal = BITMAP_ALLOC (NULL);
350 CLEAR_REG_SET (nonequal);
352 /* Now assume that we've continued by the edge E to B and continue
353 processing as if it were same basic block.
354 Our goal is to prove that whole block is an NOOP. */
356 for (insn = NEXT_INSN (BB_HEAD (b));
357 insn != NEXT_INSN (BB_END (b)) && !failed;
358 insn = NEXT_INSN (insn))
360 if (INSN_P (insn))
362 rtx pat = PATTERN (insn);
364 if (GET_CODE (pat) == PARALLEL)
366 for (i = 0; i < (unsigned)XVECLEN (pat, 0); i++)
367 failed |= mark_effect (XVECEXP (pat, 0, i), nonequal);
369 else
370 failed |= mark_effect (pat, nonequal);
373 cselib_process_insn (insn);
376 /* Later we should clear nonequal of dead registers. So far we don't
377 have life information in cfg_cleanup. */
378 if (failed)
380 b->flags |= BB_NONTHREADABLE_BLOCK;
381 goto failed_exit;
384 /* cond2 must not mention any register that is not equal to the
385 former block. */
386 if (for_each_rtx (&cond2, mentions_nonequal_regs, nonequal))
387 goto failed_exit;
389 EXECUTE_IF_SET_IN_REG_SET (nonequal, 0, i, rsi)
390 goto failed_exit;
392 BITMAP_FREE (nonequal);
393 cselib_finish ();
394 if ((comparison_dominates_p (code1, code2) != 0)
395 != (XEXP (SET_SRC (set2), 1) == pc_rtx))
396 return BRANCH_EDGE (b);
397 else
398 return FALLTHRU_EDGE (b);
400 failed_exit:
401 BITMAP_FREE (nonequal);
402 cselib_finish ();
403 return NULL;
406 /* Attempt to forward edges leaving basic block B.
407 Return true if successful. */
409 static bool
410 try_forward_edges (int mode, basic_block b)
412 bool changed = false;
413 edge_iterator ei;
414 edge e, *threaded_edges = NULL;
416 /* If we are partitioning hot/cold basic blocks, we don't want to
417 mess up unconditional or indirect jumps that cross between hot
418 and cold sections.
420 Basic block partitioning may result in some jumps that appear to
421 be optimizable (or blocks that appear to be mergeable), but which really
422 must be left untouched (they are required to make it safely across
423 partition boundaries). See the comments at the top of
424 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
426 if (find_reg_note (BB_END (b), REG_CROSSING_JUMP, NULL_RTX))
427 return false;
429 for (ei = ei_start (b->succs); (e = ei_safe_edge (ei)); )
431 basic_block target, first;
432 int counter, goto_locus;
433 bool threaded = false;
434 int nthreaded_edges = 0;
435 bool may_thread = first_pass | df_get_bb_dirty (b);
437 /* Skip complex edges because we don't know how to update them.
439 Still handle fallthru edges, as we can succeed to forward fallthru
440 edge to the same place as the branch edge of conditional branch
441 and turn conditional branch to an unconditional branch. */
442 if (e->flags & EDGE_COMPLEX)
444 ei_next (&ei);
445 continue;
448 target = first = e->dest;
449 counter = NUM_FIXED_BLOCKS;
450 goto_locus = e->goto_locus;
452 /* If we are partitioning hot/cold basic_blocks, we don't want to mess
453 up jumps that cross between hot/cold sections.
455 Basic block partitioning may result in some jumps that appear
456 to be optimizable (or blocks that appear to be mergeable), but which
457 really must be left untouched (they are required to make it safely
458 across partition boundaries). See the comments at the top of
459 bb-reorder.c:partition_hot_cold_basic_blocks for complete
460 details. */
462 if (first != EXIT_BLOCK_PTR
463 && find_reg_note (BB_END (first), REG_CROSSING_JUMP, NULL_RTX))
464 return false;
466 while (counter < n_basic_blocks)
468 basic_block new_target = NULL;
469 bool new_target_threaded = false;
470 may_thread |= df_get_bb_dirty (target);
472 if (FORWARDER_BLOCK_P (target)
473 && !(single_succ_edge (target)->flags & EDGE_CROSSING)
474 && single_succ (target) != EXIT_BLOCK_PTR)
476 /* Bypass trivial infinite loops. */
477 new_target = single_succ (target);
478 if (target == new_target)
479 counter = n_basic_blocks;
480 else if (!optimize)
482 /* When not optimizing, ensure that edges or forwarder
483 blocks with different locus are not optimized out. */
484 int new_locus = single_succ_edge (target)->goto_locus;
485 int locus = goto_locus;
487 if (new_locus && locus && !locator_eq (new_locus, locus))
488 new_target = NULL;
489 else
491 if (new_locus)
492 locus = new_locus;
494 new_locus = INSN_P (BB_END (target))
495 ? INSN_LOCATOR (BB_END (target)) : 0;
497 if (new_locus && locus && !locator_eq (new_locus, locus))
498 new_target = NULL;
499 else
501 if (new_locus)
502 locus = new_locus;
504 goto_locus = locus;
510 /* Allow to thread only over one edge at time to simplify updating
511 of probabilities. */
512 else if ((mode & CLEANUP_THREADING) && may_thread)
514 edge t = thread_jump (e, target);
515 if (t)
517 if (!threaded_edges)
518 threaded_edges = XNEWVEC (edge, n_basic_blocks);
519 else
521 int i;
523 /* Detect an infinite loop across blocks not
524 including the start block. */
525 for (i = 0; i < nthreaded_edges; ++i)
526 if (threaded_edges[i] == t)
527 break;
528 if (i < nthreaded_edges)
530 counter = n_basic_blocks;
531 break;
535 /* Detect an infinite loop across the start block. */
536 if (t->dest == b)
537 break;
539 gcc_assert (nthreaded_edges < n_basic_blocks - NUM_FIXED_BLOCKS);
540 threaded_edges[nthreaded_edges++] = t;
542 new_target = t->dest;
543 new_target_threaded = true;
547 if (!new_target)
548 break;
550 counter++;
551 target = new_target;
552 threaded |= new_target_threaded;
555 if (counter >= n_basic_blocks)
557 if (dump_file)
558 fprintf (dump_file, "Infinite loop in BB %i.\n",
559 target->index);
561 else if (target == first)
562 ; /* We didn't do anything. */
563 else
565 /* Save the values now, as the edge may get removed. */
566 gcov_type edge_count = e->count;
567 int edge_probability = e->probability;
568 int edge_frequency;
569 int n = 0;
571 e->goto_locus = goto_locus;
573 /* Don't force if target is exit block. */
574 if (threaded && target != EXIT_BLOCK_PTR)
576 notice_new_block (redirect_edge_and_branch_force (e, target));
577 if (dump_file)
578 fprintf (dump_file, "Conditionals threaded.\n");
580 else if (!redirect_edge_and_branch (e, target))
582 if (dump_file)
583 fprintf (dump_file,
584 "Forwarding edge %i->%i to %i failed.\n",
585 b->index, e->dest->index, target->index);
586 ei_next (&ei);
587 continue;
590 /* We successfully forwarded the edge. Now update profile
591 data: for each edge we traversed in the chain, remove
592 the original edge's execution count. */
593 edge_frequency = ((edge_probability * b->frequency
594 + REG_BR_PROB_BASE / 2)
595 / REG_BR_PROB_BASE);
597 if (!FORWARDER_BLOCK_P (b) && forwarder_block_p (b))
598 b->flags |= BB_FORWARDER_BLOCK;
602 edge t;
604 if (!single_succ_p (first))
606 gcc_assert (n < nthreaded_edges);
607 t = threaded_edges [n++];
608 gcc_assert (t->src == first);
609 update_bb_profile_for_threading (first, edge_frequency,
610 edge_count, t);
611 update_br_prob_note (first);
613 else
615 first->count -= edge_count;
616 if (first->count < 0)
617 first->count = 0;
618 first->frequency -= edge_frequency;
619 if (first->frequency < 0)
620 first->frequency = 0;
621 /* It is possible that as the result of
622 threading we've removed edge as it is
623 threaded to the fallthru edge. Avoid
624 getting out of sync. */
625 if (n < nthreaded_edges
626 && first == threaded_edges [n]->src)
627 n++;
628 t = single_succ_edge (first);
631 t->count -= edge_count;
632 if (t->count < 0)
633 t->count = 0;
634 first = t->dest;
636 while (first != target);
638 changed = true;
639 continue;
641 ei_next (&ei);
644 if (threaded_edges)
645 free (threaded_edges);
646 return changed;
650 /* Blocks A and B are to be merged into a single block. A has no incoming
651 fallthru edge, so it can be moved before B without adding or modifying
652 any jumps (aside from the jump from A to B). */
654 static void
655 merge_blocks_move_predecessor_nojumps (basic_block a, basic_block b)
657 rtx barrier;
659 /* If we are partitioning hot/cold basic blocks, we don't want to
660 mess up unconditional or indirect jumps that cross between hot
661 and cold sections.
663 Basic block partitioning may result in some jumps that appear to
664 be optimizable (or blocks that appear to be mergeable), but which really
665 must be left untouched (they are required to make it safely across
666 partition boundaries). See the comments at the top of
667 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
669 if (BB_PARTITION (a) != BB_PARTITION (b))
670 return;
672 barrier = next_nonnote_insn (BB_END (a));
673 gcc_assert (BARRIER_P (barrier));
674 delete_insn (barrier);
676 /* Scramble the insn chain. */
677 if (BB_END (a) != PREV_INSN (BB_HEAD (b)))
678 reorder_insns_nobb (BB_HEAD (a), BB_END (a), PREV_INSN (BB_HEAD (b)));
679 df_set_bb_dirty (a);
681 if (dump_file)
682 fprintf (dump_file, "Moved block %d before %d and merged.\n",
683 a->index, b->index);
685 /* Swap the records for the two blocks around. */
687 unlink_block (a);
688 link_block (a, b->prev_bb);
690 /* Now blocks A and B are contiguous. Merge them. */
691 merge_blocks (a, b);
694 /* Blocks A and B are to be merged into a single block. B has no outgoing
695 fallthru edge, so it can be moved after A without adding or modifying
696 any jumps (aside from the jump from A to B). */
698 static void
699 merge_blocks_move_successor_nojumps (basic_block a, basic_block b)
701 rtx barrier, real_b_end;
702 rtx label, table;
704 /* If we are partitioning hot/cold basic blocks, we don't want to
705 mess up unconditional or indirect jumps that cross between hot
706 and cold sections.
708 Basic block partitioning may result in some jumps that appear to
709 be optimizable (or blocks that appear to be mergeable), but which really
710 must be left untouched (they are required to make it safely across
711 partition boundaries). See the comments at the top of
712 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
714 if (BB_PARTITION (a) != BB_PARTITION (b))
715 return;
717 real_b_end = BB_END (b);
719 /* If there is a jump table following block B temporarily add the jump table
720 to block B so that it will also be moved to the correct location. */
721 if (tablejump_p (BB_END (b), &label, &table)
722 && prev_active_insn (label) == BB_END (b))
724 BB_END (b) = table;
727 /* There had better have been a barrier there. Delete it. */
728 barrier = NEXT_INSN (BB_END (b));
729 if (barrier && BARRIER_P (barrier))
730 delete_insn (barrier);
733 /* Scramble the insn chain. */
734 reorder_insns_nobb (BB_HEAD (b), BB_END (b), BB_END (a));
736 /* Restore the real end of b. */
737 BB_END (b) = real_b_end;
739 if (dump_file)
740 fprintf (dump_file, "Moved block %d after %d and merged.\n",
741 b->index, a->index);
743 /* Now blocks A and B are contiguous. Merge them. */
744 merge_blocks (a, b);
747 /* Attempt to merge basic blocks that are potentially non-adjacent.
748 Return NULL iff the attempt failed, otherwise return basic block
749 where cleanup_cfg should continue. Because the merging commonly
750 moves basic block away or introduces another optimization
751 possibility, return basic block just before B so cleanup_cfg don't
752 need to iterate.
754 It may be good idea to return basic block before C in the case
755 C has been moved after B and originally appeared earlier in the
756 insn sequence, but we have no information available about the
757 relative ordering of these two. Hopefully it is not too common. */
759 static basic_block
760 merge_blocks_move (edge e, basic_block b, basic_block c, int mode)
762 basic_block next;
764 /* If we are partitioning hot/cold basic blocks, we don't want to
765 mess up unconditional or indirect jumps that cross between hot
766 and cold sections.
768 Basic block partitioning may result in some jumps that appear to
769 be optimizable (or blocks that appear to be mergeable), but which really
770 must be left untouched (they are required to make it safely across
771 partition boundaries). See the comments at the top of
772 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
774 if (BB_PARTITION (b) != BB_PARTITION (c))
775 return NULL;
777 /* If B has a fallthru edge to C, no need to move anything. */
778 if (e->flags & EDGE_FALLTHRU)
780 int b_index = b->index, c_index = c->index;
781 merge_blocks (b, c);
782 update_forwarder_flag (b);
784 if (dump_file)
785 fprintf (dump_file, "Merged %d and %d without moving.\n",
786 b_index, c_index);
788 return b->prev_bb == ENTRY_BLOCK_PTR ? b : b->prev_bb;
791 /* Otherwise we will need to move code around. Do that only if expensive
792 transformations are allowed. */
793 else if (mode & CLEANUP_EXPENSIVE)
795 edge tmp_edge, b_fallthru_edge;
796 bool c_has_outgoing_fallthru;
797 bool b_has_incoming_fallthru;
799 /* Avoid overactive code motion, as the forwarder blocks should be
800 eliminated by edge redirection instead. One exception might have
801 been if B is a forwarder block and C has no fallthru edge, but
802 that should be cleaned up by bb-reorder instead. */
803 if (FORWARDER_BLOCK_P (b) || FORWARDER_BLOCK_P (c))
804 return NULL;
806 /* We must make sure to not munge nesting of lexical blocks,
807 and loop notes. This is done by squeezing out all the notes
808 and leaving them there to lie. Not ideal, but functional. */
810 tmp_edge = find_fallthru_edge (c->succs);
811 c_has_outgoing_fallthru = (tmp_edge != NULL);
813 tmp_edge = find_fallthru_edge (b->preds);
814 b_has_incoming_fallthru = (tmp_edge != NULL);
815 b_fallthru_edge = tmp_edge;
816 next = b->prev_bb;
817 if (next == c)
818 next = next->prev_bb;
820 /* Otherwise, we're going to try to move C after B. If C does
821 not have an outgoing fallthru, then it can be moved
822 immediately after B without introducing or modifying jumps. */
823 if (! c_has_outgoing_fallthru)
825 merge_blocks_move_successor_nojumps (b, c);
826 return next == ENTRY_BLOCK_PTR ? next->next_bb : next;
829 /* If B does not have an incoming fallthru, then it can be moved
830 immediately before C without introducing or modifying jumps.
831 C cannot be the first block, so we do not have to worry about
832 accessing a non-existent block. */
834 if (b_has_incoming_fallthru)
836 basic_block bb;
838 if (b_fallthru_edge->src == ENTRY_BLOCK_PTR)
839 return NULL;
840 bb = force_nonfallthru (b_fallthru_edge);
841 if (bb)
842 notice_new_block (bb);
845 merge_blocks_move_predecessor_nojumps (b, c);
846 return next == ENTRY_BLOCK_PTR ? next->next_bb : next;
849 return NULL;
853 /* Removes the memory attributes of MEM expression
854 if they are not equal. */
856 void
857 merge_memattrs (rtx x, rtx y)
859 int i;
860 int j;
861 enum rtx_code code;
862 const char *fmt;
864 if (x == y)
865 return;
866 if (x == 0 || y == 0)
867 return;
869 code = GET_CODE (x);
871 if (code != GET_CODE (y))
872 return;
874 if (GET_MODE (x) != GET_MODE (y))
875 return;
877 if (code == MEM && MEM_ATTRS (x) != MEM_ATTRS (y))
879 if (! MEM_ATTRS (x))
880 MEM_ATTRS (y) = 0;
881 else if (! MEM_ATTRS (y))
882 MEM_ATTRS (x) = 0;
883 else
885 rtx mem_size;
887 if (MEM_ALIAS_SET (x) != MEM_ALIAS_SET (y))
889 set_mem_alias_set (x, 0);
890 set_mem_alias_set (y, 0);
893 if (! mem_expr_equal_p (MEM_EXPR (x), MEM_EXPR (y)))
895 set_mem_expr (x, 0);
896 set_mem_expr (y, 0);
897 set_mem_offset (x, 0);
898 set_mem_offset (y, 0);
900 else if (MEM_OFFSET (x) != MEM_OFFSET (y))
902 set_mem_offset (x, 0);
903 set_mem_offset (y, 0);
906 if (!MEM_SIZE (x))
907 mem_size = NULL_RTX;
908 else if (!MEM_SIZE (y))
909 mem_size = NULL_RTX;
910 else
911 mem_size = GEN_INT (MAX (INTVAL (MEM_SIZE (x)),
912 INTVAL (MEM_SIZE (y))));
913 set_mem_size (x, mem_size);
914 set_mem_size (y, mem_size);
916 set_mem_align (x, MIN (MEM_ALIGN (x), MEM_ALIGN (y)));
917 set_mem_align (y, MEM_ALIGN (x));
921 fmt = GET_RTX_FORMAT (code);
922 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
924 switch (fmt[i])
926 case 'E':
927 /* Two vectors must have the same length. */
928 if (XVECLEN (x, i) != XVECLEN (y, i))
929 return;
931 for (j = 0; j < XVECLEN (x, i); j++)
932 merge_memattrs (XVECEXP (x, i, j), XVECEXP (y, i, j));
934 break;
936 case 'e':
937 merge_memattrs (XEXP (x, i), XEXP (y, i));
940 return;
944 /* Return true if I1 and I2 are equivalent and thus can be crossjumped. */
946 static bool
947 old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx i1, rtx i2)
949 rtx p1, p2;
951 /* Verify that I1 and I2 are equivalent. */
952 if (GET_CODE (i1) != GET_CODE (i2))
953 return false;
955 /* __builtin_unreachable() may lead to empty blocks (ending with
956 NOTE_INSN_BASIC_BLOCK). They may be crossjumped. */
957 if (NOTE_INSN_BASIC_BLOCK_P (i1) && NOTE_INSN_BASIC_BLOCK_P (i2))
958 return true;
960 p1 = PATTERN (i1);
961 p2 = PATTERN (i2);
963 if (GET_CODE (p1) != GET_CODE (p2))
964 return false;
966 /* If this is a CALL_INSN, compare register usage information.
967 If we don't check this on stack register machines, the two
968 CALL_INSNs might be merged leaving reg-stack.c with mismatching
969 numbers of stack registers in the same basic block.
970 If we don't check this on machines with delay slots, a delay slot may
971 be filled that clobbers a parameter expected by the subroutine.
973 ??? We take the simple route for now and assume that if they're
974 equal, they were constructed identically.
976 Also check for identical exception regions. */
978 if (CALL_P (i1))
980 /* Ensure the same EH region. */
981 rtx n1 = find_reg_note (i1, REG_EH_REGION, 0);
982 rtx n2 = find_reg_note (i2, REG_EH_REGION, 0);
984 if (!n1 && n2)
985 return false;
987 if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0)))
988 return false;
990 if (!rtx_equal_p (CALL_INSN_FUNCTION_USAGE (i1),
991 CALL_INSN_FUNCTION_USAGE (i2))
992 || SIBLING_CALL_P (i1) != SIBLING_CALL_P (i2))
993 return false;
996 #ifdef STACK_REGS
997 /* If cross_jump_death_matters is not 0, the insn's mode
998 indicates whether or not the insn contains any stack-like
999 regs. */
1001 if ((mode & CLEANUP_POST_REGSTACK) && stack_regs_mentioned (i1))
1003 /* If register stack conversion has already been done, then
1004 death notes must also be compared before it is certain that
1005 the two instruction streams match. */
1007 rtx note;
1008 HARD_REG_SET i1_regset, i2_regset;
1010 CLEAR_HARD_REG_SET (i1_regset);
1011 CLEAR_HARD_REG_SET (i2_regset);
1013 for (note = REG_NOTES (i1); note; note = XEXP (note, 1))
1014 if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
1015 SET_HARD_REG_BIT (i1_regset, REGNO (XEXP (note, 0)));
1017 for (note = REG_NOTES (i2); note; note = XEXP (note, 1))
1018 if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
1019 SET_HARD_REG_BIT (i2_regset, REGNO (XEXP (note, 0)));
1021 if (!hard_reg_set_equal_p (i1_regset, i2_regset))
1022 return false;
1024 #endif
1026 if (reload_completed
1027 ? rtx_renumbered_equal_p (p1, p2) : rtx_equal_p (p1, p2))
1028 return true;
1030 return false;
1033 /* When comparing insns I1 and I2 in flow_find_cross_jump or
1034 flow_find_head_matching_sequence, ensure the notes match. */
1036 static void
1037 merge_notes (rtx i1, rtx i2)
1039 /* If the merged insns have different REG_EQUAL notes, then
1040 remove them. */
1041 rtx equiv1 = find_reg_equal_equiv_note (i1);
1042 rtx equiv2 = find_reg_equal_equiv_note (i2);
1044 if (equiv1 && !equiv2)
1045 remove_note (i1, equiv1);
1046 else if (!equiv1 && equiv2)
1047 remove_note (i2, equiv2);
1048 else if (equiv1 && equiv2
1049 && !rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0)))
1051 remove_note (i1, equiv1);
1052 remove_note (i2, equiv2);
1056 /* Look through the insns at the end of BB1 and BB2 and find the longest
1057 sequence that are equivalent. Store the first insns for that sequence
1058 in *F1 and *F2 and return the sequence length.
1060 To simplify callers of this function, if the blocks match exactly,
1061 store the head of the blocks in *F1 and *F2. */
1064 flow_find_cross_jump (basic_block bb1, basic_block bb2, rtx *f1, rtx *f2)
1066 rtx i1, i2, last1, last2, afterlast1, afterlast2;
1067 int ninsns = 0;
1069 /* Skip simple jumps at the end of the blocks. Complex jumps still
1070 need to be compared for equivalence, which we'll do below. */
1072 i1 = BB_END (bb1);
1073 last1 = afterlast1 = last2 = afterlast2 = NULL_RTX;
1074 if (onlyjump_p (i1)
1075 || (returnjump_p (i1) && !side_effects_p (PATTERN (i1))))
1077 last1 = i1;
1078 i1 = PREV_INSN (i1);
1081 i2 = BB_END (bb2);
1082 if (onlyjump_p (i2)
1083 || (returnjump_p (i2) && !side_effects_p (PATTERN (i2))))
1085 last2 = i2;
1086 /* Count everything except for unconditional jump as insn. */
1087 if (!simplejump_p (i2) && !returnjump_p (i2) && last1)
1088 ninsns++;
1089 i2 = PREV_INSN (i2);
1092 while (true)
1094 /* Ignore notes. */
1095 while (!NONDEBUG_INSN_P (i1) && i1 != BB_HEAD (bb1))
1096 i1 = PREV_INSN (i1);
1098 while (!NONDEBUG_INSN_P (i2) && i2 != BB_HEAD (bb2))
1099 i2 = PREV_INSN (i2);
1101 if (i1 == BB_HEAD (bb1) || i2 == BB_HEAD (bb2))
1102 break;
1104 if (!old_insns_match_p (0, i1, i2))
1105 break;
1107 merge_memattrs (i1, i2);
1109 /* Don't begin a cross-jump with a NOTE insn. */
1110 if (INSN_P (i1))
1112 merge_notes (i1, i2);
1114 afterlast1 = last1, afterlast2 = last2;
1115 last1 = i1, last2 = i2;
1116 ninsns++;
1119 i1 = PREV_INSN (i1);
1120 i2 = PREV_INSN (i2);
1123 #ifdef HAVE_cc0
1124 /* Don't allow the insn after a compare to be shared by
1125 cross-jumping unless the compare is also shared. */
1126 if (ninsns && reg_mentioned_p (cc0_rtx, last1) && ! sets_cc0_p (last1))
1127 last1 = afterlast1, last2 = afterlast2, ninsns--;
1128 #endif
1130 /* Include preceding notes and labels in the cross-jump. One,
1131 this may bring us to the head of the blocks as requested above.
1132 Two, it keeps line number notes as matched as may be. */
1133 if (ninsns)
1135 while (last1 != BB_HEAD (bb1) && !NONDEBUG_INSN_P (PREV_INSN (last1)))
1136 last1 = PREV_INSN (last1);
1138 if (last1 != BB_HEAD (bb1) && LABEL_P (PREV_INSN (last1)))
1139 last1 = PREV_INSN (last1);
1141 while (last2 != BB_HEAD (bb2) && !NONDEBUG_INSN_P (PREV_INSN (last2)))
1142 last2 = PREV_INSN (last2);
1144 if (last2 != BB_HEAD (bb2) && LABEL_P (PREV_INSN (last2)))
1145 last2 = PREV_INSN (last2);
1147 *f1 = last1;
1148 *f2 = last2;
1151 return ninsns;
1154 /* Like flow_find_cross_jump, except start looking for a matching sequence from
1155 the head of the two blocks. Do not include jumps at the end.
1156 If STOP_AFTER is nonzero, stop after finding that many matching
1157 instructions. */
1160 flow_find_head_matching_sequence (basic_block bb1, basic_block bb2, rtx *f1,
1161 rtx *f2, int stop_after)
1163 rtx i1, i2, last1, last2, beforelast1, beforelast2;
1164 int ninsns = 0;
1165 edge e;
1166 edge_iterator ei;
1167 int nehedges1 = 0, nehedges2 = 0;
1169 FOR_EACH_EDGE (e, ei, bb1->succs)
1170 if (e->flags & EDGE_EH)
1171 nehedges1++;
1172 FOR_EACH_EDGE (e, ei, bb2->succs)
1173 if (e->flags & EDGE_EH)
1174 nehedges2++;
1176 i1 = BB_HEAD (bb1);
1177 i2 = BB_HEAD (bb2);
1178 last1 = beforelast1 = last2 = beforelast2 = NULL_RTX;
1180 while (true)
1182 /* Ignore notes. */
1183 while (!NONDEBUG_INSN_P (i1) && i1 != BB_END (bb1))
1184 i1 = NEXT_INSN (i1);
1186 while (!NONDEBUG_INSN_P (i2) && i2 != BB_END (bb2))
1187 i2 = NEXT_INSN (i2);
1189 if ((i1 == BB_END (bb1) && !NONDEBUG_INSN_P (i1))
1190 || (i2 == BB_END (bb2) && !NONDEBUG_INSN_P (i2)))
1191 break;
1193 if (NOTE_P (i1) || NOTE_P (i2)
1194 || JUMP_P (i1) || JUMP_P (i2))
1195 break;
1197 /* A sanity check to make sure we're not merging insns with different
1198 effects on EH. If only one of them ends a basic block, it shouldn't
1199 have an EH edge; if both end a basic block, there should be the same
1200 number of EH edges. */
1201 if ((i1 == BB_END (bb1) && i2 != BB_END (bb2)
1202 && nehedges1 > 0)
1203 || (i2 == BB_END (bb2) && i1 != BB_END (bb1)
1204 && nehedges2 > 0)
1205 || (i1 == BB_END (bb1) && i2 == BB_END (bb2)
1206 && nehedges1 != nehedges2))
1207 break;
1209 if (!old_insns_match_p (0, i1, i2))
1210 break;
1212 merge_memattrs (i1, i2);
1214 /* Don't begin a cross-jump with a NOTE insn. */
1215 if (INSN_P (i1))
1217 merge_notes (i1, i2);
1219 beforelast1 = last1, beforelast2 = last2;
1220 last1 = i1, last2 = i2;
1221 ninsns++;
1224 if (i1 == BB_END (bb1) || i2 == BB_END (bb2)
1225 || (stop_after > 0 && ninsns == stop_after))
1226 break;
1228 i1 = NEXT_INSN (i1);
1229 i2 = NEXT_INSN (i2);
1232 #ifdef HAVE_cc0
1233 /* Don't allow a compare to be shared by cross-jumping unless the insn
1234 after the compare is also shared. */
1235 if (ninsns && reg_mentioned_p (cc0_rtx, last1) && sets_cc0_p (last1))
1236 last1 = beforelast1, last2 = beforelast2, ninsns--;
1237 #endif
1239 if (ninsns)
1241 *f1 = last1;
1242 *f2 = last2;
1245 return ninsns;
1248 /* Return true iff outgoing edges of BB1 and BB2 match, together with
1249 the branch instruction. This means that if we commonize the control
1250 flow before end of the basic block, the semantic remains unchanged.
1252 We may assume that there exists one edge with a common destination. */
1254 static bool
1255 outgoing_edges_match (int mode, basic_block bb1, basic_block bb2)
1257 int nehedges1 = 0, nehedges2 = 0;
1258 edge fallthru1 = 0, fallthru2 = 0;
1259 edge e1, e2;
1260 edge_iterator ei;
1262 /* If BB1 has only one successor, we may be looking at either an
1263 unconditional jump, or a fake edge to exit. */
1264 if (single_succ_p (bb1)
1265 && (single_succ_edge (bb1)->flags & (EDGE_COMPLEX | EDGE_FAKE)) == 0
1266 && (!JUMP_P (BB_END (bb1)) || simplejump_p (BB_END (bb1))))
1267 return (single_succ_p (bb2)
1268 && (single_succ_edge (bb2)->flags
1269 & (EDGE_COMPLEX | EDGE_FAKE)) == 0
1270 && (!JUMP_P (BB_END (bb2)) || simplejump_p (BB_END (bb2))));
1272 /* Match conditional jumps - this may get tricky when fallthru and branch
1273 edges are crossed. */
1274 if (EDGE_COUNT (bb1->succs) == 2
1275 && any_condjump_p (BB_END (bb1))
1276 && onlyjump_p (BB_END (bb1)))
1278 edge b1, f1, b2, f2;
1279 bool reverse, match;
1280 rtx set1, set2, cond1, cond2;
1281 enum rtx_code code1, code2;
1283 if (EDGE_COUNT (bb2->succs) != 2
1284 || !any_condjump_p (BB_END (bb2))
1285 || !onlyjump_p (BB_END (bb2)))
1286 return false;
1288 b1 = BRANCH_EDGE (bb1);
1289 b2 = BRANCH_EDGE (bb2);
1290 f1 = FALLTHRU_EDGE (bb1);
1291 f2 = FALLTHRU_EDGE (bb2);
1293 /* Get around possible forwarders on fallthru edges. Other cases
1294 should be optimized out already. */
1295 if (FORWARDER_BLOCK_P (f1->dest))
1296 f1 = single_succ_edge (f1->dest);
1298 if (FORWARDER_BLOCK_P (f2->dest))
1299 f2 = single_succ_edge (f2->dest);
1301 /* To simplify use of this function, return false if there are
1302 unneeded forwarder blocks. These will get eliminated later
1303 during cleanup_cfg. */
1304 if (FORWARDER_BLOCK_P (f1->dest)
1305 || FORWARDER_BLOCK_P (f2->dest)
1306 || FORWARDER_BLOCK_P (b1->dest)
1307 || FORWARDER_BLOCK_P (b2->dest))
1308 return false;
1310 if (f1->dest == f2->dest && b1->dest == b2->dest)
1311 reverse = false;
1312 else if (f1->dest == b2->dest && b1->dest == f2->dest)
1313 reverse = true;
1314 else
1315 return false;
1317 set1 = pc_set (BB_END (bb1));
1318 set2 = pc_set (BB_END (bb2));
1319 if ((XEXP (SET_SRC (set1), 1) == pc_rtx)
1320 != (XEXP (SET_SRC (set2), 1) == pc_rtx))
1321 reverse = !reverse;
1323 cond1 = XEXP (SET_SRC (set1), 0);
1324 cond2 = XEXP (SET_SRC (set2), 0);
1325 code1 = GET_CODE (cond1);
1326 if (reverse)
1327 code2 = reversed_comparison_code (cond2, BB_END (bb2));
1328 else
1329 code2 = GET_CODE (cond2);
1331 if (code2 == UNKNOWN)
1332 return false;
1334 /* Verify codes and operands match. */
1335 match = ((code1 == code2
1336 && rtx_renumbered_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
1337 && rtx_renumbered_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
1338 || (code1 == swap_condition (code2)
1339 && rtx_renumbered_equal_p (XEXP (cond1, 1),
1340 XEXP (cond2, 0))
1341 && rtx_renumbered_equal_p (XEXP (cond1, 0),
1342 XEXP (cond2, 1))));
1344 /* If we return true, we will join the blocks. Which means that
1345 we will only have one branch prediction bit to work with. Thus
1346 we require the existing branches to have probabilities that are
1347 roughly similar. */
1348 if (match
1349 && optimize_bb_for_speed_p (bb1)
1350 && optimize_bb_for_speed_p (bb2))
1352 int prob2;
1354 if (b1->dest == b2->dest)
1355 prob2 = b2->probability;
1356 else
1357 /* Do not use f2 probability as f2 may be forwarded. */
1358 prob2 = REG_BR_PROB_BASE - b2->probability;
1360 /* Fail if the difference in probabilities is greater than 50%.
1361 This rules out two well-predicted branches with opposite
1362 outcomes. */
1363 if (abs (b1->probability - prob2) > REG_BR_PROB_BASE / 2)
1365 if (dump_file)
1366 fprintf (dump_file,
1367 "Outcomes of branch in bb %i and %i differ too much (%i %i)\n",
1368 bb1->index, bb2->index, b1->probability, prob2);
1370 return false;
1374 if (dump_file && match)
1375 fprintf (dump_file, "Conditionals in bb %i and %i match.\n",
1376 bb1->index, bb2->index);
1378 return match;
1381 /* Generic case - we are seeing a computed jump, table jump or trapping
1382 instruction. */
1384 /* Check whether there are tablejumps in the end of BB1 and BB2.
1385 Return true if they are identical. */
1387 rtx label1, label2;
1388 rtx table1, table2;
1390 if (tablejump_p (BB_END (bb1), &label1, &table1)
1391 && tablejump_p (BB_END (bb2), &label2, &table2)
1392 && GET_CODE (PATTERN (table1)) == GET_CODE (PATTERN (table2)))
1394 /* The labels should never be the same rtx. If they really are same
1395 the jump tables are same too. So disable crossjumping of blocks BB1
1396 and BB2 because when deleting the common insns in the end of BB1
1397 by delete_basic_block () the jump table would be deleted too. */
1398 /* If LABEL2 is referenced in BB1->END do not do anything
1399 because we would loose information when replacing
1400 LABEL1 by LABEL2 and then LABEL2 by LABEL1 in BB1->END. */
1401 if (label1 != label2 && !rtx_referenced_p (label2, BB_END (bb1)))
1403 /* Set IDENTICAL to true when the tables are identical. */
1404 bool identical = false;
1405 rtx p1, p2;
1407 p1 = PATTERN (table1);
1408 p2 = PATTERN (table2);
1409 if (GET_CODE (p1) == ADDR_VEC && rtx_equal_p (p1, p2))
1411 identical = true;
1413 else if (GET_CODE (p1) == ADDR_DIFF_VEC
1414 && (XVECLEN (p1, 1) == XVECLEN (p2, 1))
1415 && rtx_equal_p (XEXP (p1, 2), XEXP (p2, 2))
1416 && rtx_equal_p (XEXP (p1, 3), XEXP (p2, 3)))
1418 int i;
1420 identical = true;
1421 for (i = XVECLEN (p1, 1) - 1; i >= 0 && identical; i--)
1422 if (!rtx_equal_p (XVECEXP (p1, 1, i), XVECEXP (p2, 1, i)))
1423 identical = false;
1426 if (identical)
1428 replace_label_data rr;
1429 bool match;
1431 /* Temporarily replace references to LABEL1 with LABEL2
1432 in BB1->END so that we could compare the instructions. */
1433 rr.r1 = label1;
1434 rr.r2 = label2;
1435 rr.update_label_nuses = false;
1436 for_each_rtx (&BB_END (bb1), replace_label, &rr);
1438 match = old_insns_match_p (mode, BB_END (bb1), BB_END (bb2));
1439 if (dump_file && match)
1440 fprintf (dump_file,
1441 "Tablejumps in bb %i and %i match.\n",
1442 bb1->index, bb2->index);
1444 /* Set the original label in BB1->END because when deleting
1445 a block whose end is a tablejump, the tablejump referenced
1446 from the instruction is deleted too. */
1447 rr.r1 = label2;
1448 rr.r2 = label1;
1449 for_each_rtx (&BB_END (bb1), replace_label, &rr);
1451 return match;
1454 return false;
1458 /* First ensure that the instructions match. There may be many outgoing
1459 edges so this test is generally cheaper. */
1460 if (!old_insns_match_p (mode, BB_END (bb1), BB_END (bb2)))
1461 return false;
1463 /* Search the outgoing edges, ensure that the counts do match, find possible
1464 fallthru and exception handling edges since these needs more
1465 validation. */
1466 if (EDGE_COUNT (bb1->succs) != EDGE_COUNT (bb2->succs))
1467 return false;
1469 FOR_EACH_EDGE (e1, ei, bb1->succs)
1471 e2 = EDGE_SUCC (bb2, ei.index);
1473 if (e1->flags & EDGE_EH)
1474 nehedges1++;
1476 if (e2->flags & EDGE_EH)
1477 nehedges2++;
1479 if (e1->flags & EDGE_FALLTHRU)
1480 fallthru1 = e1;
1481 if (e2->flags & EDGE_FALLTHRU)
1482 fallthru2 = e2;
1485 /* If number of edges of various types does not match, fail. */
1486 if (nehedges1 != nehedges2
1487 || (fallthru1 != 0) != (fallthru2 != 0))
1488 return false;
1490 /* fallthru edges must be forwarded to the same destination. */
1491 if (fallthru1)
1493 basic_block d1 = (forwarder_block_p (fallthru1->dest)
1494 ? single_succ (fallthru1->dest): fallthru1->dest);
1495 basic_block d2 = (forwarder_block_p (fallthru2->dest)
1496 ? single_succ (fallthru2->dest): fallthru2->dest);
1498 if (d1 != d2)
1499 return false;
1502 /* Ensure the same EH region. */
1504 rtx n1 = find_reg_note (BB_END (bb1), REG_EH_REGION, 0);
1505 rtx n2 = find_reg_note (BB_END (bb2), REG_EH_REGION, 0);
1507 if (!n1 && n2)
1508 return false;
1510 if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0)))
1511 return false;
1514 /* The same checks as in try_crossjump_to_edge. It is required for RTL
1515 version of sequence abstraction. */
1516 FOR_EACH_EDGE (e1, ei, bb2->succs)
1518 edge e2;
1519 edge_iterator ei;
1520 basic_block d1 = e1->dest;
1522 if (FORWARDER_BLOCK_P (d1))
1523 d1 = EDGE_SUCC (d1, 0)->dest;
1525 FOR_EACH_EDGE (e2, ei, bb1->succs)
1527 basic_block d2 = e2->dest;
1528 if (FORWARDER_BLOCK_P (d2))
1529 d2 = EDGE_SUCC (d2, 0)->dest;
1530 if (d1 == d2)
1531 break;
1534 if (!e2)
1535 return false;
1538 return true;
1541 /* Returns true if BB basic block has a preserve label. */
1543 static bool
1544 block_has_preserve_label (basic_block bb)
1546 return (bb
1547 && block_label (bb)
1548 && LABEL_PRESERVE_P (block_label (bb)));
1551 /* E1 and E2 are edges with the same destination block. Search their
1552 predecessors for common code. If found, redirect control flow from
1553 (maybe the middle of) E1->SRC to (maybe the middle of) E2->SRC. */
1555 static bool
1556 try_crossjump_to_edge (int mode, edge e1, edge e2)
1558 int nmatch;
1559 basic_block src1 = e1->src, src2 = e2->src;
1560 basic_block redirect_to, redirect_from, to_remove;
1561 rtx newpos1, newpos2;
1562 edge s;
1563 edge_iterator ei;
1565 newpos1 = newpos2 = NULL_RTX;
1567 /* If we have partitioned hot/cold basic blocks, it is a bad idea
1568 to try this optimization.
1570 Basic block partitioning may result in some jumps that appear to
1571 be optimizable (or blocks that appear to be mergeable), but which really
1572 must be left untouched (they are required to make it safely across
1573 partition boundaries). See the comments at the top of
1574 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
1576 if (flag_reorder_blocks_and_partition && reload_completed)
1577 return false;
1579 /* Search backward through forwarder blocks. We don't need to worry
1580 about multiple entry or chained forwarders, as they will be optimized
1581 away. We do this to look past the unconditional jump following a
1582 conditional jump that is required due to the current CFG shape. */
1583 if (single_pred_p (src1)
1584 && FORWARDER_BLOCK_P (src1))
1585 e1 = single_pred_edge (src1), src1 = e1->src;
1587 if (single_pred_p (src2)
1588 && FORWARDER_BLOCK_P (src2))
1589 e2 = single_pred_edge (src2), src2 = e2->src;
1591 /* Nothing to do if we reach ENTRY, or a common source block. */
1592 if (src1 == ENTRY_BLOCK_PTR || src2 == ENTRY_BLOCK_PTR)
1593 return false;
1594 if (src1 == src2)
1595 return false;
1597 /* Seeing more than 1 forwarder blocks would confuse us later... */
1598 if (FORWARDER_BLOCK_P (e1->dest)
1599 && FORWARDER_BLOCK_P (single_succ (e1->dest)))
1600 return false;
1602 if (FORWARDER_BLOCK_P (e2->dest)
1603 && FORWARDER_BLOCK_P (single_succ (e2->dest)))
1604 return false;
1606 /* Likewise with dead code (possibly newly created by the other optimizations
1607 of cfg_cleanup). */
1608 if (EDGE_COUNT (src1->preds) == 0 || EDGE_COUNT (src2->preds) == 0)
1609 return false;
1611 /* Look for the common insn sequence, part the first ... */
1612 if (!outgoing_edges_match (mode, src1, src2))
1613 return false;
1615 /* ... and part the second. */
1616 nmatch = flow_find_cross_jump (src1, src2, &newpos1, &newpos2);
1618 /* Don't proceed with the crossjump unless we found a sufficient number
1619 of matching instructions or the 'from' block was totally matched
1620 (such that its predecessors will hopefully be redirected and the
1621 block removed). */
1622 if ((nmatch < PARAM_VALUE (PARAM_MIN_CROSSJUMP_INSNS))
1623 && (newpos1 != BB_HEAD (src1)))
1624 return false;
1626 /* Avoid deleting preserve label when redirecting ABNORMAL edges. */
1627 if (block_has_preserve_label (e1->dest)
1628 && (e1->flags & EDGE_ABNORMAL))
1629 return false;
1631 /* Here we know that the insns in the end of SRC1 which are common with SRC2
1632 will be deleted.
1633 If we have tablejumps in the end of SRC1 and SRC2
1634 they have been already compared for equivalence in outgoing_edges_match ()
1635 so replace the references to TABLE1 by references to TABLE2. */
1637 rtx label1, label2;
1638 rtx table1, table2;
1640 if (tablejump_p (BB_END (src1), &label1, &table1)
1641 && tablejump_p (BB_END (src2), &label2, &table2)
1642 && label1 != label2)
1644 replace_label_data rr;
1645 rtx insn;
1647 /* Replace references to LABEL1 with LABEL2. */
1648 rr.r1 = label1;
1649 rr.r2 = label2;
1650 rr.update_label_nuses = true;
1651 for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
1653 /* Do not replace the label in SRC1->END because when deleting
1654 a block whose end is a tablejump, the tablejump referenced
1655 from the instruction is deleted too. */
1656 if (insn != BB_END (src1))
1657 for_each_rtx (&insn, replace_label, &rr);
1662 /* Avoid splitting if possible. We must always split when SRC2 has
1663 EH predecessor edges, or we may end up with basic blocks with both
1664 normal and EH predecessor edges. */
1665 if (newpos2 == BB_HEAD (src2)
1666 && !(EDGE_PRED (src2, 0)->flags & EDGE_EH))
1667 redirect_to = src2;
1668 else
1670 if (newpos2 == BB_HEAD (src2))
1672 /* Skip possible basic block header. */
1673 if (LABEL_P (newpos2))
1674 newpos2 = NEXT_INSN (newpos2);
1675 while (DEBUG_INSN_P (newpos2))
1676 newpos2 = NEXT_INSN (newpos2);
1677 if (NOTE_P (newpos2))
1678 newpos2 = NEXT_INSN (newpos2);
1679 while (DEBUG_INSN_P (newpos2))
1680 newpos2 = NEXT_INSN (newpos2);
1683 if (dump_file)
1684 fprintf (dump_file, "Splitting bb %i before %i insns\n",
1685 src2->index, nmatch);
1686 redirect_to = split_block (src2, PREV_INSN (newpos2))->dest;
1689 if (dump_file)
1690 fprintf (dump_file,
1691 "Cross jumping from bb %i to bb %i; %i common insns\n",
1692 src1->index, src2->index, nmatch);
1694 /* We may have some registers visible through the block. */
1695 df_set_bb_dirty (redirect_to);
1697 /* Recompute the frequencies and counts of outgoing edges. */
1698 FOR_EACH_EDGE (s, ei, redirect_to->succs)
1700 edge s2;
1701 edge_iterator ei;
1702 basic_block d = s->dest;
1704 if (FORWARDER_BLOCK_P (d))
1705 d = single_succ (d);
1707 FOR_EACH_EDGE (s2, ei, src1->succs)
1709 basic_block d2 = s2->dest;
1710 if (FORWARDER_BLOCK_P (d2))
1711 d2 = single_succ (d2);
1712 if (d == d2)
1713 break;
1716 s->count += s2->count;
1718 /* Take care to update possible forwarder blocks. We verified
1719 that there is no more than one in the chain, so we can't run
1720 into infinite loop. */
1721 if (FORWARDER_BLOCK_P (s->dest))
1723 single_succ_edge (s->dest)->count += s2->count;
1724 s->dest->count += s2->count;
1725 s->dest->frequency += EDGE_FREQUENCY (s);
1728 if (FORWARDER_BLOCK_P (s2->dest))
1730 single_succ_edge (s2->dest)->count -= s2->count;
1731 if (single_succ_edge (s2->dest)->count < 0)
1732 single_succ_edge (s2->dest)->count = 0;
1733 s2->dest->count -= s2->count;
1734 s2->dest->frequency -= EDGE_FREQUENCY (s);
1735 if (s2->dest->frequency < 0)
1736 s2->dest->frequency = 0;
1737 if (s2->dest->count < 0)
1738 s2->dest->count = 0;
1741 if (!redirect_to->frequency && !src1->frequency)
1742 s->probability = (s->probability + s2->probability) / 2;
1743 else
1744 s->probability
1745 = ((s->probability * redirect_to->frequency +
1746 s2->probability * src1->frequency)
1747 / (redirect_to->frequency + src1->frequency));
1750 /* Adjust count and frequency for the block. An earlier jump
1751 threading pass may have left the profile in an inconsistent
1752 state (see update_bb_profile_for_threading) so we must be
1753 prepared for overflows. */
1754 redirect_to->count += src1->count;
1755 redirect_to->frequency += src1->frequency;
1756 if (redirect_to->frequency > BB_FREQ_MAX)
1757 redirect_to->frequency = BB_FREQ_MAX;
1758 update_br_prob_note (redirect_to);
1760 /* Edit SRC1 to go to REDIRECT_TO at NEWPOS1. */
1762 /* Skip possible basic block header. */
1763 if (LABEL_P (newpos1))
1764 newpos1 = NEXT_INSN (newpos1);
1766 while (DEBUG_INSN_P (newpos1))
1767 newpos1 = NEXT_INSN (newpos1);
1769 if (NOTE_INSN_BASIC_BLOCK_P (newpos1))
1770 newpos1 = NEXT_INSN (newpos1);
1772 while (DEBUG_INSN_P (newpos1))
1773 newpos1 = NEXT_INSN (newpos1);
1775 redirect_from = split_block (src1, PREV_INSN (newpos1))->src;
1776 to_remove = single_succ (redirect_from);
1778 redirect_edge_and_branch_force (single_succ_edge (redirect_from), redirect_to);
1779 delete_basic_block (to_remove);
1781 update_forwarder_flag (redirect_from);
1782 if (redirect_to != src2)
1783 update_forwarder_flag (src2);
1785 return true;
1788 /* Search the predecessors of BB for common insn sequences. When found,
1789 share code between them by redirecting control flow. Return true if
1790 any changes made. */
1792 static bool
1793 try_crossjump_bb (int mode, basic_block bb)
1795 edge e, e2, fallthru;
1796 bool changed;
1797 unsigned max, ix, ix2;
1798 basic_block ev, ev2;
1800 /* Nothing to do if there is not at least two incoming edges. */
1801 if (EDGE_COUNT (bb->preds) < 2)
1802 return false;
1804 /* Don't crossjump if this block ends in a computed jump,
1805 unless we are optimizing for size. */
1806 if (optimize_bb_for_size_p (bb)
1807 && bb != EXIT_BLOCK_PTR
1808 && computed_jump_p (BB_END (bb)))
1809 return false;
1811 /* If we are partitioning hot/cold basic blocks, we don't want to
1812 mess up unconditional or indirect jumps that cross between hot
1813 and cold sections.
1815 Basic block partitioning may result in some jumps that appear to
1816 be optimizable (or blocks that appear to be mergeable), but which really
1817 must be left untouched (they are required to make it safely across
1818 partition boundaries). See the comments at the top of
1819 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
1821 if (BB_PARTITION (EDGE_PRED (bb, 0)->src) !=
1822 BB_PARTITION (EDGE_PRED (bb, 1)->src)
1823 || (EDGE_PRED (bb, 0)->flags & EDGE_CROSSING))
1824 return false;
1826 /* It is always cheapest to redirect a block that ends in a branch to
1827 a block that falls through into BB, as that adds no branches to the
1828 program. We'll try that combination first. */
1829 fallthru = NULL;
1830 max = PARAM_VALUE (PARAM_MAX_CROSSJUMP_EDGES);
1832 if (EDGE_COUNT (bb->preds) > max)
1833 return false;
1835 fallthru = find_fallthru_edge (bb->preds);
1837 changed = false;
1838 for (ix = 0, ev = bb; ix < EDGE_COUNT (ev->preds); )
1840 e = EDGE_PRED (ev, ix);
1841 ix++;
1843 /* As noted above, first try with the fallthru predecessor (or, a
1844 fallthru predecessor if we are in cfglayout mode). */
1845 if (fallthru)
1847 /* Don't combine the fallthru edge into anything else.
1848 If there is a match, we'll do it the other way around. */
1849 if (e == fallthru)
1850 continue;
1851 /* If nothing changed since the last attempt, there is nothing
1852 we can do. */
1853 if (!first_pass
1854 && (!(df_get_bb_dirty (e->src))
1855 && !(df_get_bb_dirty (fallthru->src))))
1856 continue;
1858 if (try_crossjump_to_edge (mode, e, fallthru))
1860 changed = true;
1861 ix = 0;
1862 ev = bb;
1863 continue;
1867 /* Non-obvious work limiting check: Recognize that we're going
1868 to call try_crossjump_bb on every basic block. So if we have
1869 two blocks with lots of outgoing edges (a switch) and they
1870 share lots of common destinations, then we would do the
1871 cross-jump check once for each common destination.
1873 Now, if the blocks actually are cross-jump candidates, then
1874 all of their destinations will be shared. Which means that
1875 we only need check them for cross-jump candidacy once. We
1876 can eliminate redundant checks of crossjump(A,B) by arbitrarily
1877 choosing to do the check from the block for which the edge
1878 in question is the first successor of A. */
1879 if (EDGE_SUCC (e->src, 0) != e)
1880 continue;
1882 for (ix2 = 0, ev2 = bb; ix2 < EDGE_COUNT (ev2->preds); )
1884 e2 = EDGE_PRED (ev2, ix2);
1885 ix2++;
1887 if (e2 == e)
1888 continue;
1890 /* We've already checked the fallthru edge above. */
1891 if (e2 == fallthru)
1892 continue;
1894 /* The "first successor" check above only prevents multiple
1895 checks of crossjump(A,B). In order to prevent redundant
1896 checks of crossjump(B,A), require that A be the block
1897 with the lowest index. */
1898 if (e->src->index > e2->src->index)
1899 continue;
1901 /* If nothing changed since the last attempt, there is nothing
1902 we can do. */
1903 if (!first_pass
1904 && (!(df_get_bb_dirty (e->src))
1905 && !(df_get_bb_dirty (e2->src))))
1906 continue;
1908 if (try_crossjump_to_edge (mode, e, e2))
1910 changed = true;
1911 ev2 = bb;
1912 ix = 0;
1913 break;
1918 if (changed)
1919 crossjumps_occured = true;
1921 return changed;
1924 /* Return true if BB contains just bb note, or bb note followed
1925 by only DEBUG_INSNs. */
1927 static bool
1928 trivially_empty_bb_p (basic_block bb)
1930 rtx insn = BB_END (bb);
1932 while (1)
1934 if (insn == BB_HEAD (bb))
1935 return true;
1936 if (!DEBUG_INSN_P (insn))
1937 return false;
1938 insn = PREV_INSN (insn);
1942 /* Do simple CFG optimizations - basic block merging, simplifying of jump
1943 instructions etc. Return nonzero if changes were made. */
1945 static bool
1946 try_optimize_cfg (int mode)
1948 bool changed_overall = false;
1949 bool changed;
1950 int iterations = 0;
1951 basic_block bb, b, next;
1953 if (mode & (CLEANUP_CROSSJUMP | CLEANUP_THREADING))
1954 clear_bb_flags ();
1956 crossjumps_occured = false;
1958 FOR_EACH_BB (bb)
1959 update_forwarder_flag (bb);
1961 if (! targetm.cannot_modify_jumps_p ())
1963 first_pass = true;
1964 /* Attempt to merge blocks as made possible by edge removal. If
1965 a block has only one successor, and the successor has only
1966 one predecessor, they may be combined. */
1969 changed = false;
1970 iterations++;
1972 if (dump_file)
1973 fprintf (dump_file,
1974 "\n\ntry_optimize_cfg iteration %i\n\n",
1975 iterations);
1977 for (b = ENTRY_BLOCK_PTR->next_bb; b != EXIT_BLOCK_PTR;)
1979 basic_block c;
1980 edge s;
1981 bool changed_here = false;
1983 /* Delete trivially dead basic blocks. This is either
1984 blocks with no predecessors, or empty blocks with no
1985 successors. However if the empty block with no
1986 successors is the successor of the ENTRY_BLOCK, it is
1987 kept. This ensures that the ENTRY_BLOCK will have a
1988 successor which is a precondition for many RTL
1989 passes. Empty blocks may result from expanding
1990 __builtin_unreachable (). */
1991 if (EDGE_COUNT (b->preds) == 0
1992 || (EDGE_COUNT (b->succs) == 0
1993 && trivially_empty_bb_p (b)
1994 && single_succ_edge (ENTRY_BLOCK_PTR)->dest != b))
1996 c = b->prev_bb;
1997 if (EDGE_COUNT (b->preds) > 0)
1999 edge e;
2000 edge_iterator ei;
2002 if (current_ir_type () == IR_RTL_CFGLAYOUT)
2004 if (b->il.rtl->footer
2005 && BARRIER_P (b->il.rtl->footer))
2006 FOR_EACH_EDGE (e, ei, b->preds)
2007 if ((e->flags & EDGE_FALLTHRU)
2008 && e->src->il.rtl->footer == NULL)
2010 if (b->il.rtl->footer)
2012 e->src->il.rtl->footer = b->il.rtl->footer;
2013 b->il.rtl->footer = NULL;
2015 else
2017 start_sequence ();
2018 e->src->il.rtl->footer = emit_barrier ();
2019 end_sequence ();
2023 else
2025 rtx last = get_last_bb_insn (b);
2026 if (last && BARRIER_P (last))
2027 FOR_EACH_EDGE (e, ei, b->preds)
2028 if ((e->flags & EDGE_FALLTHRU))
2029 emit_barrier_after (BB_END (e->src));
2032 delete_basic_block (b);
2033 if (!(mode & CLEANUP_CFGLAYOUT))
2034 changed = true;
2035 /* Avoid trying to remove ENTRY_BLOCK_PTR. */
2036 b = (c == ENTRY_BLOCK_PTR ? c->next_bb : c);
2037 continue;
2040 /* Remove code labels no longer used. */
2041 if (single_pred_p (b)
2042 && (single_pred_edge (b)->flags & EDGE_FALLTHRU)
2043 && !(single_pred_edge (b)->flags & EDGE_COMPLEX)
2044 && LABEL_P (BB_HEAD (b))
2045 /* If the previous block ends with a branch to this
2046 block, we can't delete the label. Normally this
2047 is a condjump that is yet to be simplified, but
2048 if CASE_DROPS_THRU, this can be a tablejump with
2049 some element going to the same place as the
2050 default (fallthru). */
2051 && (single_pred (b) == ENTRY_BLOCK_PTR
2052 || !JUMP_P (BB_END (single_pred (b)))
2053 || ! label_is_jump_target_p (BB_HEAD (b),
2054 BB_END (single_pred (b)))))
2056 rtx label = BB_HEAD (b);
2058 delete_insn_chain (label, label, false);
2059 /* If the case label is undeletable, move it after the
2060 BASIC_BLOCK note. */
2061 if (NOTE_KIND (BB_HEAD (b)) == NOTE_INSN_DELETED_LABEL)
2063 rtx bb_note = NEXT_INSN (BB_HEAD (b));
2065 reorder_insns_nobb (label, label, bb_note);
2066 BB_HEAD (b) = bb_note;
2067 if (BB_END (b) == bb_note)
2068 BB_END (b) = label;
2070 if (dump_file)
2071 fprintf (dump_file, "Deleted label in block %i.\n",
2072 b->index);
2075 /* If we fall through an empty block, we can remove it. */
2076 if (!(mode & CLEANUP_CFGLAYOUT)
2077 && single_pred_p (b)
2078 && (single_pred_edge (b)->flags & EDGE_FALLTHRU)
2079 && !LABEL_P (BB_HEAD (b))
2080 && FORWARDER_BLOCK_P (b)
2081 /* Note that forwarder_block_p true ensures that
2082 there is a successor for this block. */
2083 && (single_succ_edge (b)->flags & EDGE_FALLTHRU)
2084 && n_basic_blocks > NUM_FIXED_BLOCKS + 1)
2086 if (dump_file)
2087 fprintf (dump_file,
2088 "Deleting fallthru block %i.\n",
2089 b->index);
2091 c = b->prev_bb == ENTRY_BLOCK_PTR ? b->next_bb : b->prev_bb;
2092 redirect_edge_succ_nodup (single_pred_edge (b),
2093 single_succ (b));
2094 delete_basic_block (b);
2095 changed = true;
2096 b = c;
2097 continue;
2100 /* Merge B with its single successor, if any. */
2101 if (single_succ_p (b)
2102 && (s = single_succ_edge (b))
2103 && !(s->flags & EDGE_COMPLEX)
2104 && (c = s->dest) != EXIT_BLOCK_PTR
2105 && single_pred_p (c)
2106 && b != c)
2108 /* When not in cfg_layout mode use code aware of reordering
2109 INSN. This code possibly creates new basic blocks so it
2110 does not fit merge_blocks interface and is kept here in
2111 hope that it will become useless once more of compiler
2112 is transformed to use cfg_layout mode. */
2114 if ((mode & CLEANUP_CFGLAYOUT)
2115 && can_merge_blocks_p (b, c))
2117 merge_blocks (b, c);
2118 update_forwarder_flag (b);
2119 changed_here = true;
2121 else if (!(mode & CLEANUP_CFGLAYOUT)
2122 /* If the jump insn has side effects,
2123 we can't kill the edge. */
2124 && (!JUMP_P (BB_END (b))
2125 || (reload_completed
2126 ? simplejump_p (BB_END (b))
2127 : (onlyjump_p (BB_END (b))
2128 && !tablejump_p (BB_END (b),
2129 NULL, NULL))))
2130 && (next = merge_blocks_move (s, b, c, mode)))
2132 b = next;
2133 changed_here = true;
2137 /* Simplify branch over branch. */
2138 if ((mode & CLEANUP_EXPENSIVE)
2139 && !(mode & CLEANUP_CFGLAYOUT)
2140 && try_simplify_condjump (b))
2141 changed_here = true;
2143 /* If B has a single outgoing edge, but uses a
2144 non-trivial jump instruction without side-effects, we
2145 can either delete the jump entirely, or replace it
2146 with a simple unconditional jump. */
2147 if (single_succ_p (b)
2148 && single_succ (b) != EXIT_BLOCK_PTR
2149 && onlyjump_p (BB_END (b))
2150 && !find_reg_note (BB_END (b), REG_CROSSING_JUMP, NULL_RTX)
2151 && try_redirect_by_replacing_jump (single_succ_edge (b),
2152 single_succ (b),
2153 (mode & CLEANUP_CFGLAYOUT) != 0))
2155 update_forwarder_flag (b);
2156 changed_here = true;
2159 /* Simplify branch to branch. */
2160 if (try_forward_edges (mode, b))
2161 changed_here = true;
2163 /* Look for shared code between blocks. */
2164 if ((mode & CLEANUP_CROSSJUMP)
2165 && try_crossjump_bb (mode, b))
2166 changed_here = true;
2168 /* Don't get confused by the index shift caused by
2169 deleting blocks. */
2170 if (!changed_here)
2171 b = b->next_bb;
2172 else
2173 changed = true;
2176 if ((mode & CLEANUP_CROSSJUMP)
2177 && try_crossjump_bb (mode, EXIT_BLOCK_PTR))
2178 changed = true;
2180 #ifdef ENABLE_CHECKING
2181 if (changed)
2182 verify_flow_info ();
2183 #endif
2185 changed_overall |= changed;
2186 first_pass = false;
2188 while (changed);
2191 FOR_ALL_BB (b)
2192 b->flags &= ~(BB_FORWARDER_BLOCK | BB_NONTHREADABLE_BLOCK);
2194 return changed_overall;
2197 /* Delete all unreachable basic blocks. */
2199 bool
2200 delete_unreachable_blocks (void)
2202 bool changed = false;
2203 basic_block b, prev_bb;
2205 find_unreachable_blocks ();
2207 /* When we're in GIMPLE mode and there may be debug insns, we should
2208 delete blocks in reverse dominator order, so as to get a chance
2209 to substitute all released DEFs into debug stmts. If we don't
2210 have dominators information, walking blocks backward gets us a
2211 better chance of retaining most debug information than
2212 otherwise. */
2213 if (MAY_HAVE_DEBUG_STMTS && current_ir_type () == IR_GIMPLE
2214 && dom_info_available_p (CDI_DOMINATORS))
2216 for (b = EXIT_BLOCK_PTR->prev_bb; b != ENTRY_BLOCK_PTR; b = prev_bb)
2218 prev_bb = b->prev_bb;
2220 if (!(b->flags & BB_REACHABLE))
2222 /* Speed up the removal of blocks that don't dominate
2223 others. Walking backwards, this should be the common
2224 case. */
2225 if (!first_dom_son (CDI_DOMINATORS, b))
2226 delete_basic_block (b);
2227 else
2229 VEC (basic_block, heap) *h
2230 = get_all_dominated_blocks (CDI_DOMINATORS, b);
2232 while (VEC_length (basic_block, h))
2234 b = VEC_pop (basic_block, h);
2236 prev_bb = b->prev_bb;
2238 gcc_assert (!(b->flags & BB_REACHABLE));
2240 delete_basic_block (b);
2243 VEC_free (basic_block, heap, h);
2246 changed = true;
2250 else
2252 for (b = EXIT_BLOCK_PTR->prev_bb; b != ENTRY_BLOCK_PTR; b = prev_bb)
2254 prev_bb = b->prev_bb;
2256 if (!(b->flags & BB_REACHABLE))
2258 delete_basic_block (b);
2259 changed = true;
2264 if (changed)
2265 tidy_fallthru_edges ();
2266 return changed;
2269 /* Delete any jump tables never referenced. We can't delete them at the
2270 time of removing tablejump insn as they are referenced by the preceding
2271 insns computing the destination, so we delay deleting and garbagecollect
2272 them once life information is computed. */
2273 void
2274 delete_dead_jumptables (void)
2276 basic_block bb;
2278 /* A dead jump table does not belong to any basic block. Scan insns
2279 between two adjacent basic blocks. */
2280 FOR_EACH_BB (bb)
2282 rtx insn, next;
2284 for (insn = NEXT_INSN (BB_END (bb));
2285 insn && !NOTE_INSN_BASIC_BLOCK_P (insn);
2286 insn = next)
2288 next = NEXT_INSN (insn);
2289 if (LABEL_P (insn)
2290 && LABEL_NUSES (insn) == LABEL_PRESERVE_P (insn)
2291 && JUMP_TABLE_DATA_P (next))
2293 rtx label = insn, jump = next;
2295 if (dump_file)
2296 fprintf (dump_file, "Dead jumptable %i removed\n",
2297 INSN_UID (insn));
2299 next = NEXT_INSN (next);
2300 delete_insn (jump);
2301 delete_insn (label);
2308 /* Tidy the CFG by deleting unreachable code and whatnot. */
2310 bool
2311 cleanup_cfg (int mode)
2313 bool changed = false;
2315 /* Set the cfglayout mode flag here. We could update all the callers
2316 but that is just inconvenient, especially given that we eventually
2317 want to have cfglayout mode as the default. */
2318 if (current_ir_type () == IR_RTL_CFGLAYOUT)
2319 mode |= CLEANUP_CFGLAYOUT;
2321 timevar_push (TV_CLEANUP_CFG);
2322 if (delete_unreachable_blocks ())
2324 changed = true;
2325 /* We've possibly created trivially dead code. Cleanup it right
2326 now to introduce more opportunities for try_optimize_cfg. */
2327 if (!(mode & (CLEANUP_NO_INSN_DEL))
2328 && !reload_completed)
2329 delete_trivially_dead_insns (get_insns (), max_reg_num ());
2332 compact_blocks ();
2334 /* To tail-merge blocks ending in the same noreturn function (e.g.
2335 a call to abort) we have to insert fake edges to exit. Do this
2336 here once. The fake edges do not interfere with any other CFG
2337 cleanups. */
2338 if (mode & CLEANUP_CROSSJUMP)
2339 add_noreturn_fake_exit_edges ();
2341 if (!dbg_cnt (cfg_cleanup))
2342 return changed;
2344 while (try_optimize_cfg (mode))
2346 delete_unreachable_blocks (), changed = true;
2347 if (!(mode & CLEANUP_NO_INSN_DEL))
2349 /* Try to remove some trivially dead insns when doing an expensive
2350 cleanup. But delete_trivially_dead_insns doesn't work after
2351 reload (it only handles pseudos) and run_fast_dce is too costly
2352 to run in every iteration.
2354 For effective cross jumping, we really want to run a fast DCE to
2355 clean up any dead conditions, or they get in the way of performing
2356 useful tail merges.
2358 Other transformations in cleanup_cfg are not so sensitive to dead
2359 code, so delete_trivially_dead_insns or even doing nothing at all
2360 is good enough. */
2361 if ((mode & CLEANUP_EXPENSIVE) && !reload_completed
2362 && !delete_trivially_dead_insns (get_insns (), max_reg_num ()))
2363 break;
2364 else if ((mode & CLEANUP_CROSSJUMP)
2365 && crossjumps_occured)
2366 run_fast_dce ();
2368 else
2369 break;
2372 if (mode & CLEANUP_CROSSJUMP)
2373 remove_fake_exit_edges ();
2375 /* Don't call delete_dead_jumptables in cfglayout mode, because
2376 that function assumes that jump tables are in the insns stream.
2377 But we also don't _have_ to delete dead jumptables in cfglayout
2378 mode because we shouldn't even be looking at things that are
2379 not in a basic block. Dead jumptables are cleaned up when
2380 going out of cfglayout mode. */
2381 if (!(mode & CLEANUP_CFGLAYOUT))
2382 delete_dead_jumptables ();
2384 timevar_pop (TV_CLEANUP_CFG);
2386 return changed;
2389 static unsigned int
2390 rest_of_handle_jump (void)
2392 if (crtl->tail_call_emit)
2393 fixup_tail_calls ();
2394 return 0;
2397 struct rtl_opt_pass pass_jump =
2400 RTL_PASS,
2401 "sibling", /* name */
2402 NULL, /* gate */
2403 rest_of_handle_jump, /* execute */
2404 NULL, /* sub */
2405 NULL, /* next */
2406 0, /* static_pass_number */
2407 TV_JUMP, /* tv_id */
2408 0, /* properties_required */
2409 0, /* properties_provided */
2410 0, /* properties_destroyed */
2411 TODO_ggc_collect, /* todo_flags_start */
2412 TODO_verify_flow, /* todo_flags_finish */
2417 static unsigned int
2418 rest_of_handle_jump2 (void)
2420 delete_trivially_dead_insns (get_insns (), max_reg_num ());
2421 if (dump_file)
2422 dump_flow_info (dump_file, dump_flags);
2423 cleanup_cfg ((optimize ? CLEANUP_EXPENSIVE : 0)
2424 | (flag_thread_jumps ? CLEANUP_THREADING : 0));
2425 return 0;
2429 struct rtl_opt_pass pass_jump2 =
2432 RTL_PASS,
2433 "jump", /* name */
2434 NULL, /* gate */
2435 rest_of_handle_jump2, /* execute */
2436 NULL, /* sub */
2437 NULL, /* next */
2438 0, /* static_pass_number */
2439 TV_JUMP, /* tv_id */
2440 0, /* properties_required */
2441 0, /* properties_provided */
2442 0, /* properties_destroyed */
2443 TODO_ggc_collect, /* todo_flags_start */
2444 TODO_dump_func | TODO_verify_rtl_sharing,/* todo_flags_finish */