Remove the temporary array for reductions written to memory.
[official-gcc/graphite-test-results.git] / gcc / cfgcleanup.c
blobbf6ca45990ed9066d720d3f1789a5d2d3e0f5fa6
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 "cselib.h"
48 #include "params.h"
49 #include "tm_p.h"
50 #include "target.h"
51 #include "cfglayout.h"
52 #include "emit-rtl.h"
53 #include "tree-pass.h"
54 #include "cfgloop.h"
55 #include "expr.h"
56 #include "df.h"
57 #include "dce.h"
58 #include "dbgcnt.h"
60 #define FORWARDER_BLOCK_P(BB) ((BB)->flags & BB_FORWARDER_BLOCK)
62 /* Set to true when we are running first pass of try_optimize_cfg loop. */
63 static bool first_pass;
65 /* Set to true if crossjumps occured in the latest run of try_optimize_cfg. */
66 static bool crossjumps_occured;
68 /* Set to true if we couldn't run an optimization due to stale liveness
69 information; we should run df_analyze to enable more opportunities. */
70 static bool block_was_dirty;
72 static bool try_crossjump_to_edge (int, edge, edge);
73 static bool try_crossjump_bb (int, basic_block);
74 static bool outgoing_edges_match (int, basic_block, basic_block);
75 static bool old_insns_match_p (int, rtx, rtx);
77 static void merge_blocks_move_predecessor_nojumps (basic_block, basic_block);
78 static void merge_blocks_move_successor_nojumps (basic_block, basic_block);
79 static bool try_optimize_cfg (int);
80 static bool try_simplify_condjump (basic_block);
81 static bool try_forward_edges (int, basic_block);
82 static edge thread_jump (edge, basic_block);
83 static bool mark_effect (rtx, bitmap);
84 static void notice_new_block (basic_block);
85 static void update_forwarder_flag (basic_block);
86 static int mentions_nonequal_regs (rtx *, void *);
87 static void merge_memattrs (rtx, rtx);
89 /* Set flags for newly created block. */
91 static void
92 notice_new_block (basic_block bb)
94 if (!bb)
95 return;
97 if (forwarder_block_p (bb))
98 bb->flags |= BB_FORWARDER_BLOCK;
101 /* Recompute forwarder flag after block has been modified. */
103 static void
104 update_forwarder_flag (basic_block bb)
106 if (forwarder_block_p (bb))
107 bb->flags |= BB_FORWARDER_BLOCK;
108 else
109 bb->flags &= ~BB_FORWARDER_BLOCK;
112 /* Simplify a conditional jump around an unconditional jump.
113 Return true if something changed. */
115 static bool
116 try_simplify_condjump (basic_block cbranch_block)
118 basic_block jump_block, jump_dest_block, cbranch_dest_block;
119 edge cbranch_jump_edge, cbranch_fallthru_edge;
120 rtx cbranch_insn;
122 /* Verify that there are exactly two successors. */
123 if (EDGE_COUNT (cbranch_block->succs) != 2)
124 return false;
126 /* Verify that we've got a normal conditional branch at the end
127 of the block. */
128 cbranch_insn = BB_END (cbranch_block);
129 if (!any_condjump_p (cbranch_insn))
130 return false;
132 cbranch_fallthru_edge = FALLTHRU_EDGE (cbranch_block);
133 cbranch_jump_edge = BRANCH_EDGE (cbranch_block);
135 /* The next block must not have multiple predecessors, must not
136 be the last block in the function, and must contain just the
137 unconditional jump. */
138 jump_block = cbranch_fallthru_edge->dest;
139 if (!single_pred_p (jump_block)
140 || jump_block->next_bb == EXIT_BLOCK_PTR
141 || !FORWARDER_BLOCK_P (jump_block))
142 return false;
143 jump_dest_block = single_succ (jump_block);
145 /* If we are partitioning hot/cold basic blocks, we don't want to
146 mess up unconditional or indirect jumps that cross between hot
147 and cold sections.
149 Basic block partitioning may result in some jumps that appear to
150 be optimizable (or blocks that appear to be mergeable), but which really
151 must be left untouched (they are required to make it safely across
152 partition boundaries). See the comments at the top of
153 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
155 if (BB_PARTITION (jump_block) != BB_PARTITION (jump_dest_block)
156 || (cbranch_jump_edge->flags & EDGE_CROSSING))
157 return false;
159 /* The conditional branch must target the block after the
160 unconditional branch. */
161 cbranch_dest_block = cbranch_jump_edge->dest;
163 if (cbranch_dest_block == EXIT_BLOCK_PTR
164 || !can_fallthru (jump_block, cbranch_dest_block))
165 return false;
167 /* Invert the conditional branch. */
168 if (!invert_jump (cbranch_insn, block_label (jump_dest_block), 0))
169 return false;
171 if (dump_file)
172 fprintf (dump_file, "Simplifying condjump %i around jump %i\n",
173 INSN_UID (cbranch_insn), INSN_UID (BB_END (jump_block)));
175 /* Success. Update the CFG to match. Note that after this point
176 the edge variable names appear backwards; the redirection is done
177 this way to preserve edge profile data. */
178 cbranch_jump_edge = redirect_edge_succ_nodup (cbranch_jump_edge,
179 cbranch_dest_block);
180 cbranch_fallthru_edge = redirect_edge_succ_nodup (cbranch_fallthru_edge,
181 jump_dest_block);
182 cbranch_jump_edge->flags |= EDGE_FALLTHRU;
183 cbranch_fallthru_edge->flags &= ~EDGE_FALLTHRU;
184 update_br_prob_note (cbranch_block);
186 /* Delete the block with the unconditional jump, and clean up the mess. */
187 delete_basic_block (jump_block);
188 tidy_fallthru_edge (cbranch_jump_edge);
189 update_forwarder_flag (cbranch_block);
191 return true;
194 /* Attempt to prove that operation is NOOP using CSElib or mark the effect
195 on register. Used by jump threading. */
197 static bool
198 mark_effect (rtx exp, regset nonequal)
200 int regno;
201 rtx dest;
202 switch (GET_CODE (exp))
204 /* In case we do clobber the register, mark it as equal, as we know the
205 value is dead so it don't have to match. */
206 case CLOBBER:
207 if (REG_P (XEXP (exp, 0)))
209 dest = XEXP (exp, 0);
210 regno = REGNO (dest);
211 CLEAR_REGNO_REG_SET (nonequal, regno);
212 if (regno < FIRST_PSEUDO_REGISTER)
214 int n = hard_regno_nregs[regno][GET_MODE (dest)];
215 while (--n > 0)
216 CLEAR_REGNO_REG_SET (nonequal, regno + n);
219 return false;
221 case SET:
222 if (rtx_equal_for_cselib_p (SET_DEST (exp), SET_SRC (exp)))
223 return false;
224 dest = SET_DEST (exp);
225 if (dest == pc_rtx)
226 return false;
227 if (!REG_P (dest))
228 return true;
229 regno = REGNO (dest);
230 SET_REGNO_REG_SET (nonequal, regno);
231 if (regno < FIRST_PSEUDO_REGISTER)
233 int n = hard_regno_nregs[regno][GET_MODE (dest)];
234 while (--n > 0)
235 SET_REGNO_REG_SET (nonequal, regno + n);
237 return false;
239 default:
240 return false;
244 /* Return nonzero if X is a register set in regset DATA.
245 Called via for_each_rtx. */
246 static int
247 mentions_nonequal_regs (rtx *x, void *data)
249 regset nonequal = (regset) data;
250 if (REG_P (*x))
252 int regno;
254 regno = REGNO (*x);
255 if (REGNO_REG_SET_P (nonequal, regno))
256 return 1;
257 if (regno < FIRST_PSEUDO_REGISTER)
259 int n = hard_regno_nregs[regno][GET_MODE (*x)];
260 while (--n > 0)
261 if (REGNO_REG_SET_P (nonequal, regno + n))
262 return 1;
265 return 0;
267 /* Attempt to prove that the basic block B will have no side effects and
268 always continues in the same edge if reached via E. Return the edge
269 if exist, NULL otherwise. */
271 static edge
272 thread_jump (edge e, basic_block b)
274 rtx set1, set2, cond1, cond2, insn;
275 enum rtx_code code1, code2, reversed_code2;
276 bool reverse1 = false;
277 unsigned i;
278 regset nonequal;
279 bool failed = false;
280 reg_set_iterator rsi;
282 if (b->flags & BB_NONTHREADABLE_BLOCK)
283 return NULL;
285 /* At the moment, we do handle only conditional jumps, but later we may
286 want to extend this code to tablejumps and others. */
287 if (EDGE_COUNT (e->src->succs) != 2)
288 return NULL;
289 if (EDGE_COUNT (b->succs) != 2)
291 b->flags |= BB_NONTHREADABLE_BLOCK;
292 return NULL;
295 /* Second branch must end with onlyjump, as we will eliminate the jump. */
296 if (!any_condjump_p (BB_END (e->src)))
297 return NULL;
299 if (!any_condjump_p (BB_END (b)) || !onlyjump_p (BB_END (b)))
301 b->flags |= BB_NONTHREADABLE_BLOCK;
302 return NULL;
305 set1 = pc_set (BB_END (e->src));
306 set2 = pc_set (BB_END (b));
307 if (((e->flags & EDGE_FALLTHRU) != 0)
308 != (XEXP (SET_SRC (set1), 1) == pc_rtx))
309 reverse1 = true;
311 cond1 = XEXP (SET_SRC (set1), 0);
312 cond2 = XEXP (SET_SRC (set2), 0);
313 if (reverse1)
314 code1 = reversed_comparison_code (cond1, BB_END (e->src));
315 else
316 code1 = GET_CODE (cond1);
318 code2 = GET_CODE (cond2);
319 reversed_code2 = reversed_comparison_code (cond2, BB_END (b));
321 if (!comparison_dominates_p (code1, code2)
322 && !comparison_dominates_p (code1, reversed_code2))
323 return NULL;
325 /* Ensure that the comparison operators are equivalent.
326 ??? This is far too pessimistic. We should allow swapped operands,
327 different CCmodes, or for example comparisons for interval, that
328 dominate even when operands are not equivalent. */
329 if (!rtx_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
330 || !rtx_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
331 return NULL;
333 /* Short circuit cases where block B contains some side effects, as we can't
334 safely bypass it. */
335 for (insn = NEXT_INSN (BB_HEAD (b)); insn != NEXT_INSN (BB_END (b));
336 insn = NEXT_INSN (insn))
337 if (INSN_P (insn) && side_effects_p (PATTERN (insn)))
339 b->flags |= BB_NONTHREADABLE_BLOCK;
340 return NULL;
343 cselib_init (0);
345 /* First process all values computed in the source basic block. */
346 for (insn = NEXT_INSN (BB_HEAD (e->src));
347 insn != NEXT_INSN (BB_END (e->src));
348 insn = NEXT_INSN (insn))
349 if (INSN_P (insn))
350 cselib_process_insn (insn);
352 nonequal = BITMAP_ALLOC (NULL);
353 CLEAR_REG_SET (nonequal);
355 /* Now assume that we've continued by the edge E to B and continue
356 processing as if it were same basic block.
357 Our goal is to prove that whole block is an NOOP. */
359 for (insn = NEXT_INSN (BB_HEAD (b));
360 insn != NEXT_INSN (BB_END (b)) && !failed;
361 insn = NEXT_INSN (insn))
363 if (INSN_P (insn))
365 rtx pat = PATTERN (insn);
367 if (GET_CODE (pat) == PARALLEL)
369 for (i = 0; i < (unsigned)XVECLEN (pat, 0); i++)
370 failed |= mark_effect (XVECEXP (pat, 0, i), nonequal);
372 else
373 failed |= mark_effect (pat, nonequal);
376 cselib_process_insn (insn);
379 /* Later we should clear nonequal of dead registers. So far we don't
380 have life information in cfg_cleanup. */
381 if (failed)
383 b->flags |= BB_NONTHREADABLE_BLOCK;
384 goto failed_exit;
387 /* cond2 must not mention any register that is not equal to the
388 former block. */
389 if (for_each_rtx (&cond2, mentions_nonequal_regs, nonequal))
390 goto failed_exit;
392 EXECUTE_IF_SET_IN_REG_SET (nonequal, 0, i, rsi)
393 goto failed_exit;
395 BITMAP_FREE (nonequal);
396 cselib_finish ();
397 if ((comparison_dominates_p (code1, code2) != 0)
398 != (XEXP (SET_SRC (set2), 1) == pc_rtx))
399 return BRANCH_EDGE (b);
400 else
401 return FALLTHRU_EDGE (b);
403 failed_exit:
404 BITMAP_FREE (nonequal);
405 cselib_finish ();
406 return NULL;
409 /* Attempt to forward edges leaving basic block B.
410 Return true if successful. */
412 static bool
413 try_forward_edges (int mode, basic_block b)
415 bool changed = false;
416 edge_iterator ei;
417 edge e, *threaded_edges = NULL;
419 /* If we are partitioning hot/cold basic blocks, we don't want to
420 mess up unconditional or indirect jumps that cross between hot
421 and cold sections.
423 Basic block partitioning may result in some jumps that appear to
424 be optimizable (or blocks that appear to be mergeable), but which really
425 must be left untouched (they are required to make it safely across
426 partition boundaries). See the comments at the top of
427 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
429 if (find_reg_note (BB_END (b), REG_CROSSING_JUMP, NULL_RTX))
430 return false;
432 for (ei = ei_start (b->succs); (e = ei_safe_edge (ei)); )
434 basic_block target, first;
435 int counter, goto_locus;
436 bool threaded = false;
437 int nthreaded_edges = 0;
438 bool may_thread = first_pass || (b->flags & BB_MODIFIED) != 0;
440 /* Skip complex edges because we don't know how to update them.
442 Still handle fallthru edges, as we can succeed to forward fallthru
443 edge to the same place as the branch edge of conditional branch
444 and turn conditional branch to an unconditional branch. */
445 if (e->flags & EDGE_COMPLEX)
447 ei_next (&ei);
448 continue;
451 target = first = e->dest;
452 counter = NUM_FIXED_BLOCKS;
453 goto_locus = e->goto_locus;
455 /* If we are partitioning hot/cold basic_blocks, we don't want to mess
456 up jumps that cross between hot/cold sections.
458 Basic block partitioning may result in some jumps that appear
459 to be optimizable (or blocks that appear to be mergeable), but which
460 really must be left untouched (they are required to make it safely
461 across partition boundaries). See the comments at the top of
462 bb-reorder.c:partition_hot_cold_basic_blocks for complete
463 details. */
465 if (first != EXIT_BLOCK_PTR
466 && find_reg_note (BB_END (first), REG_CROSSING_JUMP, NULL_RTX))
467 return false;
469 while (counter < n_basic_blocks)
471 basic_block new_target = NULL;
472 bool new_target_threaded = false;
473 may_thread |= (target->flags & BB_MODIFIED) != 0;
475 if (FORWARDER_BLOCK_P (target)
476 && !(single_succ_edge (target)->flags & EDGE_CROSSING)
477 && single_succ (target) != EXIT_BLOCK_PTR)
479 /* Bypass trivial infinite loops. */
480 new_target = single_succ (target);
481 if (target == new_target)
482 counter = n_basic_blocks;
483 else if (!optimize)
485 /* When not optimizing, ensure that edges or forwarder
486 blocks with different locus are not optimized out. */
487 int new_locus = single_succ_edge (target)->goto_locus;
488 int locus = goto_locus;
490 if (new_locus && locus && !locator_eq (new_locus, locus))
491 new_target = NULL;
492 else
494 rtx last;
496 if (new_locus)
497 locus = new_locus;
499 last = BB_END (target);
500 if (DEBUG_INSN_P (last))
501 last = prev_nondebug_insn (last);
503 new_locus = last && INSN_P (last)
504 ? INSN_LOCATOR (last) : 0;
506 if (new_locus && locus && !locator_eq (new_locus, locus))
507 new_target = NULL;
508 else
510 if (new_locus)
511 locus = new_locus;
513 goto_locus = locus;
519 /* Allow to thread only over one edge at time to simplify updating
520 of probabilities. */
521 else if ((mode & CLEANUP_THREADING) && may_thread)
523 edge t = thread_jump (e, target);
524 if (t)
526 if (!threaded_edges)
527 threaded_edges = XNEWVEC (edge, n_basic_blocks);
528 else
530 int i;
532 /* Detect an infinite loop across blocks not
533 including the start block. */
534 for (i = 0; i < nthreaded_edges; ++i)
535 if (threaded_edges[i] == t)
536 break;
537 if (i < nthreaded_edges)
539 counter = n_basic_blocks;
540 break;
544 /* Detect an infinite loop across the start block. */
545 if (t->dest == b)
546 break;
548 gcc_assert (nthreaded_edges < n_basic_blocks - NUM_FIXED_BLOCKS);
549 threaded_edges[nthreaded_edges++] = t;
551 new_target = t->dest;
552 new_target_threaded = true;
556 if (!new_target)
557 break;
559 counter++;
560 target = new_target;
561 threaded |= new_target_threaded;
564 if (counter >= n_basic_blocks)
566 if (dump_file)
567 fprintf (dump_file, "Infinite loop in BB %i.\n",
568 target->index);
570 else if (target == first)
571 ; /* We didn't do anything. */
572 else
574 /* Save the values now, as the edge may get removed. */
575 gcov_type edge_count = e->count;
576 int edge_probability = e->probability;
577 int edge_frequency;
578 int n = 0;
580 e->goto_locus = goto_locus;
582 /* Don't force if target is exit block. */
583 if (threaded && target != EXIT_BLOCK_PTR)
585 notice_new_block (redirect_edge_and_branch_force (e, target));
586 if (dump_file)
587 fprintf (dump_file, "Conditionals threaded.\n");
589 else if (!redirect_edge_and_branch (e, target))
591 if (dump_file)
592 fprintf (dump_file,
593 "Forwarding edge %i->%i to %i failed.\n",
594 b->index, e->dest->index, target->index);
595 ei_next (&ei);
596 continue;
599 /* We successfully forwarded the edge. Now update profile
600 data: for each edge we traversed in the chain, remove
601 the original edge's execution count. */
602 edge_frequency = ((edge_probability * b->frequency
603 + REG_BR_PROB_BASE / 2)
604 / REG_BR_PROB_BASE);
606 if (!FORWARDER_BLOCK_P (b) && forwarder_block_p (b))
607 b->flags |= BB_FORWARDER_BLOCK;
611 edge t;
613 if (!single_succ_p (first))
615 gcc_assert (n < nthreaded_edges);
616 t = threaded_edges [n++];
617 gcc_assert (t->src == first);
618 update_bb_profile_for_threading (first, edge_frequency,
619 edge_count, t);
620 update_br_prob_note (first);
622 else
624 first->count -= edge_count;
625 if (first->count < 0)
626 first->count = 0;
627 first->frequency -= edge_frequency;
628 if (first->frequency < 0)
629 first->frequency = 0;
630 /* It is possible that as the result of
631 threading we've removed edge as it is
632 threaded to the fallthru edge. Avoid
633 getting out of sync. */
634 if (n < nthreaded_edges
635 && first == threaded_edges [n]->src)
636 n++;
637 t = single_succ_edge (first);
640 t->count -= edge_count;
641 if (t->count < 0)
642 t->count = 0;
643 first = t->dest;
645 while (first != target);
647 changed = true;
648 continue;
650 ei_next (&ei);
653 if (threaded_edges)
654 free (threaded_edges);
655 return changed;
659 /* Blocks A and B are to be merged into a single block. A has no incoming
660 fallthru edge, so it can be moved before B without adding or modifying
661 any jumps (aside from the jump from A to B). */
663 static void
664 merge_blocks_move_predecessor_nojumps (basic_block a, basic_block b)
666 rtx barrier;
668 /* If we are partitioning hot/cold basic blocks, we don't want to
669 mess up unconditional or indirect jumps that cross between hot
670 and cold sections.
672 Basic block partitioning may result in some jumps that appear to
673 be optimizable (or blocks that appear to be mergeable), but which really
674 must be left untouched (they are required to make it safely across
675 partition boundaries). See the comments at the top of
676 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
678 if (BB_PARTITION (a) != BB_PARTITION (b))
679 return;
681 barrier = next_nonnote_insn (BB_END (a));
682 gcc_assert (BARRIER_P (barrier));
683 delete_insn (barrier);
685 /* Scramble the insn chain. */
686 if (BB_END (a) != PREV_INSN (BB_HEAD (b)))
687 reorder_insns_nobb (BB_HEAD (a), BB_END (a), PREV_INSN (BB_HEAD (b)));
688 df_set_bb_dirty (a);
690 if (dump_file)
691 fprintf (dump_file, "Moved block %d before %d and merged.\n",
692 a->index, b->index);
694 /* Swap the records for the two blocks around. */
696 unlink_block (a);
697 link_block (a, b->prev_bb);
699 /* Now blocks A and B are contiguous. Merge them. */
700 merge_blocks (a, b);
703 /* Blocks A and B are to be merged into a single block. B has no outgoing
704 fallthru edge, so it can be moved after A without adding or modifying
705 any jumps (aside from the jump from A to B). */
707 static void
708 merge_blocks_move_successor_nojumps (basic_block a, basic_block b)
710 rtx barrier, real_b_end;
711 rtx label, table;
713 /* If we are partitioning hot/cold basic blocks, we don't want to
714 mess up unconditional or indirect jumps that cross between hot
715 and cold sections.
717 Basic block partitioning may result in some jumps that appear to
718 be optimizable (or blocks that appear to be mergeable), but which really
719 must be left untouched (they are required to make it safely across
720 partition boundaries). See the comments at the top of
721 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
723 if (BB_PARTITION (a) != BB_PARTITION (b))
724 return;
726 real_b_end = BB_END (b);
728 /* If there is a jump table following block B temporarily add the jump table
729 to block B so that it will also be moved to the correct location. */
730 if (tablejump_p (BB_END (b), &label, &table)
731 && prev_active_insn (label) == BB_END (b))
733 BB_END (b) = table;
736 /* There had better have been a barrier there. Delete it. */
737 barrier = NEXT_INSN (BB_END (b));
738 if (barrier && BARRIER_P (barrier))
739 delete_insn (barrier);
742 /* Scramble the insn chain. */
743 reorder_insns_nobb (BB_HEAD (b), BB_END (b), BB_END (a));
745 /* Restore the real end of b. */
746 BB_END (b) = real_b_end;
748 if (dump_file)
749 fprintf (dump_file, "Moved block %d after %d and merged.\n",
750 b->index, a->index);
752 /* Now blocks A and B are contiguous. Merge them. */
753 merge_blocks (a, b);
756 /* Attempt to merge basic blocks that are potentially non-adjacent.
757 Return NULL iff the attempt failed, otherwise return basic block
758 where cleanup_cfg should continue. Because the merging commonly
759 moves basic block away or introduces another optimization
760 possibility, return basic block just before B so cleanup_cfg don't
761 need to iterate.
763 It may be good idea to return basic block before C in the case
764 C has been moved after B and originally appeared earlier in the
765 insn sequence, but we have no information available about the
766 relative ordering of these two. Hopefully it is not too common. */
768 static basic_block
769 merge_blocks_move (edge e, basic_block b, basic_block c, int mode)
771 basic_block next;
773 /* If we are partitioning hot/cold basic blocks, we don't want to
774 mess up unconditional or indirect jumps that cross between hot
775 and cold sections.
777 Basic block partitioning may result in some jumps that appear to
778 be optimizable (or blocks that appear to be mergeable), but which really
779 must be left untouched (they are required to make it safely across
780 partition boundaries). See the comments at the top of
781 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
783 if (BB_PARTITION (b) != BB_PARTITION (c))
784 return NULL;
786 /* If B has a fallthru edge to C, no need to move anything. */
787 if (e->flags & EDGE_FALLTHRU)
789 int b_index = b->index, c_index = c->index;
790 merge_blocks (b, c);
791 update_forwarder_flag (b);
793 if (dump_file)
794 fprintf (dump_file, "Merged %d and %d without moving.\n",
795 b_index, c_index);
797 return b->prev_bb == ENTRY_BLOCK_PTR ? b : b->prev_bb;
800 /* Otherwise we will need to move code around. Do that only if expensive
801 transformations are allowed. */
802 else if (mode & CLEANUP_EXPENSIVE)
804 edge tmp_edge, b_fallthru_edge;
805 bool c_has_outgoing_fallthru;
806 bool b_has_incoming_fallthru;
808 /* Avoid overactive code motion, as the forwarder blocks should be
809 eliminated by edge redirection instead. One exception might have
810 been if B is a forwarder block and C has no fallthru edge, but
811 that should be cleaned up by bb-reorder instead. */
812 if (FORWARDER_BLOCK_P (b) || FORWARDER_BLOCK_P (c))
813 return NULL;
815 /* We must make sure to not munge nesting of lexical blocks,
816 and loop notes. This is done by squeezing out all the notes
817 and leaving them there to lie. Not ideal, but functional. */
819 tmp_edge = find_fallthru_edge (c->succs);
820 c_has_outgoing_fallthru = (tmp_edge != NULL);
822 tmp_edge = find_fallthru_edge (b->preds);
823 b_has_incoming_fallthru = (tmp_edge != NULL);
824 b_fallthru_edge = tmp_edge;
825 next = b->prev_bb;
826 if (next == c)
827 next = next->prev_bb;
829 /* Otherwise, we're going to try to move C after B. If C does
830 not have an outgoing fallthru, then it can be moved
831 immediately after B without introducing or modifying jumps. */
832 if (! c_has_outgoing_fallthru)
834 merge_blocks_move_successor_nojumps (b, c);
835 return next == ENTRY_BLOCK_PTR ? next->next_bb : next;
838 /* If B does not have an incoming fallthru, then it can be moved
839 immediately before C without introducing or modifying jumps.
840 C cannot be the first block, so we do not have to worry about
841 accessing a non-existent block. */
843 if (b_has_incoming_fallthru)
845 basic_block bb;
847 if (b_fallthru_edge->src == ENTRY_BLOCK_PTR)
848 return NULL;
849 bb = force_nonfallthru (b_fallthru_edge);
850 if (bb)
851 notice_new_block (bb);
854 merge_blocks_move_predecessor_nojumps (b, c);
855 return next == ENTRY_BLOCK_PTR ? next->next_bb : next;
858 return NULL;
862 /* Removes the memory attributes of MEM expression
863 if they are not equal. */
865 void
866 merge_memattrs (rtx x, rtx y)
868 int i;
869 int j;
870 enum rtx_code code;
871 const char *fmt;
873 if (x == y)
874 return;
875 if (x == 0 || y == 0)
876 return;
878 code = GET_CODE (x);
880 if (code != GET_CODE (y))
881 return;
883 if (GET_MODE (x) != GET_MODE (y))
884 return;
886 if (code == MEM && MEM_ATTRS (x) != MEM_ATTRS (y))
888 if (! MEM_ATTRS (x))
889 MEM_ATTRS (y) = 0;
890 else if (! MEM_ATTRS (y))
891 MEM_ATTRS (x) = 0;
892 else
894 rtx mem_size;
896 if (MEM_ALIAS_SET (x) != MEM_ALIAS_SET (y))
898 set_mem_alias_set (x, 0);
899 set_mem_alias_set (y, 0);
902 if (! mem_expr_equal_p (MEM_EXPR (x), MEM_EXPR (y)))
904 set_mem_expr (x, 0);
905 set_mem_expr (y, 0);
906 set_mem_offset (x, 0);
907 set_mem_offset (y, 0);
909 else if (MEM_OFFSET (x) != MEM_OFFSET (y))
911 set_mem_offset (x, 0);
912 set_mem_offset (y, 0);
915 if (!MEM_SIZE (x))
916 mem_size = NULL_RTX;
917 else if (!MEM_SIZE (y))
918 mem_size = NULL_RTX;
919 else
920 mem_size = GEN_INT (MAX (INTVAL (MEM_SIZE (x)),
921 INTVAL (MEM_SIZE (y))));
922 set_mem_size (x, mem_size);
923 set_mem_size (y, mem_size);
925 set_mem_align (x, MIN (MEM_ALIGN (x), MEM_ALIGN (y)));
926 set_mem_align (y, MEM_ALIGN (x));
930 fmt = GET_RTX_FORMAT (code);
931 for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
933 switch (fmt[i])
935 case 'E':
936 /* Two vectors must have the same length. */
937 if (XVECLEN (x, i) != XVECLEN (y, i))
938 return;
940 for (j = 0; j < XVECLEN (x, i); j++)
941 merge_memattrs (XVECEXP (x, i, j), XVECEXP (y, i, j));
943 break;
945 case 'e':
946 merge_memattrs (XEXP (x, i), XEXP (y, i));
949 return;
953 /* Return true if I1 and I2 are equivalent and thus can be crossjumped. */
955 static bool
956 old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx i1, rtx i2)
958 rtx p1, p2;
960 /* Verify that I1 and I2 are equivalent. */
961 if (GET_CODE (i1) != GET_CODE (i2))
962 return false;
964 /* __builtin_unreachable() may lead to empty blocks (ending with
965 NOTE_INSN_BASIC_BLOCK). They may be crossjumped. */
966 if (NOTE_INSN_BASIC_BLOCK_P (i1) && NOTE_INSN_BASIC_BLOCK_P (i2))
967 return true;
969 p1 = PATTERN (i1);
970 p2 = PATTERN (i2);
972 if (GET_CODE (p1) != GET_CODE (p2))
973 return false;
975 /* If this is a CALL_INSN, compare register usage information.
976 If we don't check this on stack register machines, the two
977 CALL_INSNs might be merged leaving reg-stack.c with mismatching
978 numbers of stack registers in the same basic block.
979 If we don't check this on machines with delay slots, a delay slot may
980 be filled that clobbers a parameter expected by the subroutine.
982 ??? We take the simple route for now and assume that if they're
983 equal, they were constructed identically.
985 Also check for identical exception regions. */
987 if (CALL_P (i1))
989 /* Ensure the same EH region. */
990 rtx n1 = find_reg_note (i1, REG_EH_REGION, 0);
991 rtx n2 = find_reg_note (i2, REG_EH_REGION, 0);
993 if (!n1 && n2)
994 return false;
996 if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0)))
997 return false;
999 if (!rtx_equal_p (CALL_INSN_FUNCTION_USAGE (i1),
1000 CALL_INSN_FUNCTION_USAGE (i2))
1001 || SIBLING_CALL_P (i1) != SIBLING_CALL_P (i2))
1002 return false;
1005 #ifdef STACK_REGS
1006 /* If cross_jump_death_matters is not 0, the insn's mode
1007 indicates whether or not the insn contains any stack-like
1008 regs. */
1010 if ((mode & CLEANUP_POST_REGSTACK) && stack_regs_mentioned (i1))
1012 /* If register stack conversion has already been done, then
1013 death notes must also be compared before it is certain that
1014 the two instruction streams match. */
1016 rtx note;
1017 HARD_REG_SET i1_regset, i2_regset;
1019 CLEAR_HARD_REG_SET (i1_regset);
1020 CLEAR_HARD_REG_SET (i2_regset);
1022 for (note = REG_NOTES (i1); note; note = XEXP (note, 1))
1023 if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
1024 SET_HARD_REG_BIT (i1_regset, REGNO (XEXP (note, 0)));
1026 for (note = REG_NOTES (i2); note; note = XEXP (note, 1))
1027 if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
1028 SET_HARD_REG_BIT (i2_regset, REGNO (XEXP (note, 0)));
1030 if (!hard_reg_set_equal_p (i1_regset, i2_regset))
1031 return false;
1033 #endif
1035 if (reload_completed
1036 ? rtx_renumbered_equal_p (p1, p2) : rtx_equal_p (p1, p2))
1037 return true;
1039 return false;
1042 /* When comparing insns I1 and I2 in flow_find_cross_jump or
1043 flow_find_head_matching_sequence, ensure the notes match. */
1045 static void
1046 merge_notes (rtx i1, rtx i2)
1048 /* If the merged insns have different REG_EQUAL notes, then
1049 remove them. */
1050 rtx equiv1 = find_reg_equal_equiv_note (i1);
1051 rtx equiv2 = find_reg_equal_equiv_note (i2);
1053 if (equiv1 && !equiv2)
1054 remove_note (i1, equiv1);
1055 else if (!equiv1 && equiv2)
1056 remove_note (i2, equiv2);
1057 else if (equiv1 && equiv2
1058 && !rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0)))
1060 remove_note (i1, equiv1);
1061 remove_note (i2, equiv2);
1065 /* Look through the insns at the end of BB1 and BB2 and find the longest
1066 sequence that are equivalent. Store the first insns for that sequence
1067 in *F1 and *F2 and return the sequence length.
1069 To simplify callers of this function, if the blocks match exactly,
1070 store the head of the blocks in *F1 and *F2. */
1073 flow_find_cross_jump (basic_block bb1, basic_block bb2, rtx *f1, rtx *f2)
1075 rtx i1, i2, last1, last2, afterlast1, afterlast2;
1076 int ninsns = 0;
1078 /* Skip simple jumps at the end of the blocks. Complex jumps still
1079 need to be compared for equivalence, which we'll do below. */
1081 i1 = BB_END (bb1);
1082 last1 = afterlast1 = last2 = afterlast2 = NULL_RTX;
1083 if (onlyjump_p (i1)
1084 || (returnjump_p (i1) && !side_effects_p (PATTERN (i1))))
1086 last1 = i1;
1087 i1 = PREV_INSN (i1);
1090 i2 = BB_END (bb2);
1091 if (onlyjump_p (i2)
1092 || (returnjump_p (i2) && !side_effects_p (PATTERN (i2))))
1094 last2 = i2;
1095 /* Count everything except for unconditional jump as insn. */
1096 if (!simplejump_p (i2) && !returnjump_p (i2) && last1)
1097 ninsns++;
1098 i2 = PREV_INSN (i2);
1101 while (true)
1103 /* Ignore notes. */
1104 while (!NONDEBUG_INSN_P (i1) && i1 != BB_HEAD (bb1))
1105 i1 = PREV_INSN (i1);
1107 while (!NONDEBUG_INSN_P (i2) && i2 != BB_HEAD (bb2))
1108 i2 = PREV_INSN (i2);
1110 if (i1 == BB_HEAD (bb1) || i2 == BB_HEAD (bb2))
1111 break;
1113 if (!old_insns_match_p (0, i1, i2))
1114 break;
1116 merge_memattrs (i1, i2);
1118 /* Don't begin a cross-jump with a NOTE insn. */
1119 if (INSN_P (i1))
1121 merge_notes (i1, i2);
1123 afterlast1 = last1, afterlast2 = last2;
1124 last1 = i1, last2 = i2;
1125 ninsns++;
1128 i1 = PREV_INSN (i1);
1129 i2 = PREV_INSN (i2);
1132 #ifdef HAVE_cc0
1133 /* Don't allow the insn after a compare to be shared by
1134 cross-jumping unless the compare is also shared. */
1135 if (ninsns && reg_mentioned_p (cc0_rtx, last1) && ! sets_cc0_p (last1))
1136 last1 = afterlast1, last2 = afterlast2, ninsns--;
1137 #endif
1139 /* Include preceding notes and labels in the cross-jump. One,
1140 this may bring us to the head of the blocks as requested above.
1141 Two, it keeps line number notes as matched as may be. */
1142 if (ninsns)
1144 while (last1 != BB_HEAD (bb1) && !NONDEBUG_INSN_P (PREV_INSN (last1)))
1145 last1 = PREV_INSN (last1);
1147 if (last1 != BB_HEAD (bb1) && LABEL_P (PREV_INSN (last1)))
1148 last1 = PREV_INSN (last1);
1150 while (last2 != BB_HEAD (bb2) && !NONDEBUG_INSN_P (PREV_INSN (last2)))
1151 last2 = PREV_INSN (last2);
1153 if (last2 != BB_HEAD (bb2) && LABEL_P (PREV_INSN (last2)))
1154 last2 = PREV_INSN (last2);
1156 *f1 = last1;
1157 *f2 = last2;
1160 return ninsns;
1163 /* Like flow_find_cross_jump, except start looking for a matching sequence from
1164 the head of the two blocks. Do not include jumps at the end.
1165 If STOP_AFTER is nonzero, stop after finding that many matching
1166 instructions. */
1169 flow_find_head_matching_sequence (basic_block bb1, basic_block bb2, rtx *f1,
1170 rtx *f2, int stop_after)
1172 rtx i1, i2, last1, last2, beforelast1, beforelast2;
1173 int ninsns = 0;
1174 edge e;
1175 edge_iterator ei;
1176 int nehedges1 = 0, nehedges2 = 0;
1178 FOR_EACH_EDGE (e, ei, bb1->succs)
1179 if (e->flags & EDGE_EH)
1180 nehedges1++;
1181 FOR_EACH_EDGE (e, ei, bb2->succs)
1182 if (e->flags & EDGE_EH)
1183 nehedges2++;
1185 i1 = BB_HEAD (bb1);
1186 i2 = BB_HEAD (bb2);
1187 last1 = beforelast1 = last2 = beforelast2 = NULL_RTX;
1189 while (true)
1191 /* Ignore notes, except NOTE_INSN_EPILOGUE_BEG. */
1192 while (!NONDEBUG_INSN_P (i1) && i1 != BB_END (bb1))
1194 if (NOTE_P (i1) && NOTE_KIND (i1) == NOTE_INSN_EPILOGUE_BEG)
1195 break;
1196 i1 = NEXT_INSN (i1);
1199 while (!NONDEBUG_INSN_P (i2) && i2 != BB_END (bb2))
1201 if (NOTE_P (i2) && NOTE_KIND (i2) == NOTE_INSN_EPILOGUE_BEG)
1202 break;
1203 i2 = NEXT_INSN (i2);
1206 if ((i1 == BB_END (bb1) && !NONDEBUG_INSN_P (i1))
1207 || (i2 == BB_END (bb2) && !NONDEBUG_INSN_P (i2)))
1208 break;
1210 if (NOTE_P (i1) || NOTE_P (i2)
1211 || JUMP_P (i1) || JUMP_P (i2))
1212 break;
1214 /* A sanity check to make sure we're not merging insns with different
1215 effects on EH. If only one of them ends a basic block, it shouldn't
1216 have an EH edge; if both end a basic block, there should be the same
1217 number of EH edges. */
1218 if ((i1 == BB_END (bb1) && i2 != BB_END (bb2)
1219 && nehedges1 > 0)
1220 || (i2 == BB_END (bb2) && i1 != BB_END (bb1)
1221 && nehedges2 > 0)
1222 || (i1 == BB_END (bb1) && i2 == BB_END (bb2)
1223 && nehedges1 != nehedges2))
1224 break;
1226 if (!old_insns_match_p (0, i1, i2))
1227 break;
1229 merge_memattrs (i1, i2);
1231 /* Don't begin a cross-jump with a NOTE insn. */
1232 if (INSN_P (i1))
1234 merge_notes (i1, i2);
1236 beforelast1 = last1, beforelast2 = last2;
1237 last1 = i1, last2 = i2;
1238 ninsns++;
1241 if (i1 == BB_END (bb1) || i2 == BB_END (bb2)
1242 || (stop_after > 0 && ninsns == stop_after))
1243 break;
1245 i1 = NEXT_INSN (i1);
1246 i2 = NEXT_INSN (i2);
1249 #ifdef HAVE_cc0
1250 /* Don't allow a compare to be shared by cross-jumping unless the insn
1251 after the compare is also shared. */
1252 if (ninsns && reg_mentioned_p (cc0_rtx, last1) && sets_cc0_p (last1))
1253 last1 = beforelast1, last2 = beforelast2, ninsns--;
1254 #endif
1256 if (ninsns)
1258 *f1 = last1;
1259 *f2 = last2;
1262 return ninsns;
1265 /* Return true iff outgoing edges of BB1 and BB2 match, together with
1266 the branch instruction. This means that if we commonize the control
1267 flow before end of the basic block, the semantic remains unchanged.
1269 We may assume that there exists one edge with a common destination. */
1271 static bool
1272 outgoing_edges_match (int mode, basic_block bb1, basic_block bb2)
1274 int nehedges1 = 0, nehedges2 = 0;
1275 edge fallthru1 = 0, fallthru2 = 0;
1276 edge e1, e2;
1277 edge_iterator ei;
1279 /* If BB1 has only one successor, we may be looking at either an
1280 unconditional jump, or a fake edge to exit. */
1281 if (single_succ_p (bb1)
1282 && (single_succ_edge (bb1)->flags & (EDGE_COMPLEX | EDGE_FAKE)) == 0
1283 && (!JUMP_P (BB_END (bb1)) || simplejump_p (BB_END (bb1))))
1284 return (single_succ_p (bb2)
1285 && (single_succ_edge (bb2)->flags
1286 & (EDGE_COMPLEX | EDGE_FAKE)) == 0
1287 && (!JUMP_P (BB_END (bb2)) || simplejump_p (BB_END (bb2))));
1289 /* Match conditional jumps - this may get tricky when fallthru and branch
1290 edges are crossed. */
1291 if (EDGE_COUNT (bb1->succs) == 2
1292 && any_condjump_p (BB_END (bb1))
1293 && onlyjump_p (BB_END (bb1)))
1295 edge b1, f1, b2, f2;
1296 bool reverse, match;
1297 rtx set1, set2, cond1, cond2;
1298 enum rtx_code code1, code2;
1300 if (EDGE_COUNT (bb2->succs) != 2
1301 || !any_condjump_p (BB_END (bb2))
1302 || !onlyjump_p (BB_END (bb2)))
1303 return false;
1305 b1 = BRANCH_EDGE (bb1);
1306 b2 = BRANCH_EDGE (bb2);
1307 f1 = FALLTHRU_EDGE (bb1);
1308 f2 = FALLTHRU_EDGE (bb2);
1310 /* Get around possible forwarders on fallthru edges. Other cases
1311 should be optimized out already. */
1312 if (FORWARDER_BLOCK_P (f1->dest))
1313 f1 = single_succ_edge (f1->dest);
1315 if (FORWARDER_BLOCK_P (f2->dest))
1316 f2 = single_succ_edge (f2->dest);
1318 /* To simplify use of this function, return false if there are
1319 unneeded forwarder blocks. These will get eliminated later
1320 during cleanup_cfg. */
1321 if (FORWARDER_BLOCK_P (f1->dest)
1322 || FORWARDER_BLOCK_P (f2->dest)
1323 || FORWARDER_BLOCK_P (b1->dest)
1324 || FORWARDER_BLOCK_P (b2->dest))
1325 return false;
1327 if (f1->dest == f2->dest && b1->dest == b2->dest)
1328 reverse = false;
1329 else if (f1->dest == b2->dest && b1->dest == f2->dest)
1330 reverse = true;
1331 else
1332 return false;
1334 set1 = pc_set (BB_END (bb1));
1335 set2 = pc_set (BB_END (bb2));
1336 if ((XEXP (SET_SRC (set1), 1) == pc_rtx)
1337 != (XEXP (SET_SRC (set2), 1) == pc_rtx))
1338 reverse = !reverse;
1340 cond1 = XEXP (SET_SRC (set1), 0);
1341 cond2 = XEXP (SET_SRC (set2), 0);
1342 code1 = GET_CODE (cond1);
1343 if (reverse)
1344 code2 = reversed_comparison_code (cond2, BB_END (bb2));
1345 else
1346 code2 = GET_CODE (cond2);
1348 if (code2 == UNKNOWN)
1349 return false;
1351 /* Verify codes and operands match. */
1352 match = ((code1 == code2
1353 && rtx_renumbered_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
1354 && rtx_renumbered_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
1355 || (code1 == swap_condition (code2)
1356 && rtx_renumbered_equal_p (XEXP (cond1, 1),
1357 XEXP (cond2, 0))
1358 && rtx_renumbered_equal_p (XEXP (cond1, 0),
1359 XEXP (cond2, 1))));
1361 /* If we return true, we will join the blocks. Which means that
1362 we will only have one branch prediction bit to work with. Thus
1363 we require the existing branches to have probabilities that are
1364 roughly similar. */
1365 if (match
1366 && optimize_bb_for_speed_p (bb1)
1367 && optimize_bb_for_speed_p (bb2))
1369 int prob2;
1371 if (b1->dest == b2->dest)
1372 prob2 = b2->probability;
1373 else
1374 /* Do not use f2 probability as f2 may be forwarded. */
1375 prob2 = REG_BR_PROB_BASE - b2->probability;
1377 /* Fail if the difference in probabilities is greater than 50%.
1378 This rules out two well-predicted branches with opposite
1379 outcomes. */
1380 if (abs (b1->probability - prob2) > REG_BR_PROB_BASE / 2)
1382 if (dump_file)
1383 fprintf (dump_file,
1384 "Outcomes of branch in bb %i and %i differ too much (%i %i)\n",
1385 bb1->index, bb2->index, b1->probability, prob2);
1387 return false;
1391 if (dump_file && match)
1392 fprintf (dump_file, "Conditionals in bb %i and %i match.\n",
1393 bb1->index, bb2->index);
1395 return match;
1398 /* Generic case - we are seeing a computed jump, table jump or trapping
1399 instruction. */
1401 /* Check whether there are tablejumps in the end of BB1 and BB2.
1402 Return true if they are identical. */
1404 rtx label1, label2;
1405 rtx table1, table2;
1407 if (tablejump_p (BB_END (bb1), &label1, &table1)
1408 && tablejump_p (BB_END (bb2), &label2, &table2)
1409 && GET_CODE (PATTERN (table1)) == GET_CODE (PATTERN (table2)))
1411 /* The labels should never be the same rtx. If they really are same
1412 the jump tables are same too. So disable crossjumping of blocks BB1
1413 and BB2 because when deleting the common insns in the end of BB1
1414 by delete_basic_block () the jump table would be deleted too. */
1415 /* If LABEL2 is referenced in BB1->END do not do anything
1416 because we would loose information when replacing
1417 LABEL1 by LABEL2 and then LABEL2 by LABEL1 in BB1->END. */
1418 if (label1 != label2 && !rtx_referenced_p (label2, BB_END (bb1)))
1420 /* Set IDENTICAL to true when the tables are identical. */
1421 bool identical = false;
1422 rtx p1, p2;
1424 p1 = PATTERN (table1);
1425 p2 = PATTERN (table2);
1426 if (GET_CODE (p1) == ADDR_VEC && rtx_equal_p (p1, p2))
1428 identical = true;
1430 else if (GET_CODE (p1) == ADDR_DIFF_VEC
1431 && (XVECLEN (p1, 1) == XVECLEN (p2, 1))
1432 && rtx_equal_p (XEXP (p1, 2), XEXP (p2, 2))
1433 && rtx_equal_p (XEXP (p1, 3), XEXP (p2, 3)))
1435 int i;
1437 identical = true;
1438 for (i = XVECLEN (p1, 1) - 1; i >= 0 && identical; i--)
1439 if (!rtx_equal_p (XVECEXP (p1, 1, i), XVECEXP (p2, 1, i)))
1440 identical = false;
1443 if (identical)
1445 replace_label_data rr;
1446 bool match;
1448 /* Temporarily replace references to LABEL1 with LABEL2
1449 in BB1->END so that we could compare the instructions. */
1450 rr.r1 = label1;
1451 rr.r2 = label2;
1452 rr.update_label_nuses = false;
1453 for_each_rtx (&BB_END (bb1), replace_label, &rr);
1455 match = old_insns_match_p (mode, BB_END (bb1), BB_END (bb2));
1456 if (dump_file && match)
1457 fprintf (dump_file,
1458 "Tablejumps in bb %i and %i match.\n",
1459 bb1->index, bb2->index);
1461 /* Set the original label in BB1->END because when deleting
1462 a block whose end is a tablejump, the tablejump referenced
1463 from the instruction is deleted too. */
1464 rr.r1 = label2;
1465 rr.r2 = label1;
1466 for_each_rtx (&BB_END (bb1), replace_label, &rr);
1468 return match;
1471 return false;
1475 /* First ensure that the instructions match. There may be many outgoing
1476 edges so this test is generally cheaper. */
1477 if (!old_insns_match_p (mode, BB_END (bb1), BB_END (bb2)))
1478 return false;
1480 /* Search the outgoing edges, ensure that the counts do match, find possible
1481 fallthru and exception handling edges since these needs more
1482 validation. */
1483 if (EDGE_COUNT (bb1->succs) != EDGE_COUNT (bb2->succs))
1484 return false;
1486 FOR_EACH_EDGE (e1, ei, bb1->succs)
1488 e2 = EDGE_SUCC (bb2, ei.index);
1490 if (e1->flags & EDGE_EH)
1491 nehedges1++;
1493 if (e2->flags & EDGE_EH)
1494 nehedges2++;
1496 if (e1->flags & EDGE_FALLTHRU)
1497 fallthru1 = e1;
1498 if (e2->flags & EDGE_FALLTHRU)
1499 fallthru2 = e2;
1502 /* If number of edges of various types does not match, fail. */
1503 if (nehedges1 != nehedges2
1504 || (fallthru1 != 0) != (fallthru2 != 0))
1505 return false;
1507 /* fallthru edges must be forwarded to the same destination. */
1508 if (fallthru1)
1510 basic_block d1 = (forwarder_block_p (fallthru1->dest)
1511 ? single_succ (fallthru1->dest): fallthru1->dest);
1512 basic_block d2 = (forwarder_block_p (fallthru2->dest)
1513 ? single_succ (fallthru2->dest): fallthru2->dest);
1515 if (d1 != d2)
1516 return false;
1519 /* Ensure the same EH region. */
1521 rtx n1 = find_reg_note (BB_END (bb1), REG_EH_REGION, 0);
1522 rtx n2 = find_reg_note (BB_END (bb2), REG_EH_REGION, 0);
1524 if (!n1 && n2)
1525 return false;
1527 if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0)))
1528 return false;
1531 /* The same checks as in try_crossjump_to_edge. It is required for RTL
1532 version of sequence abstraction. */
1533 FOR_EACH_EDGE (e1, ei, bb2->succs)
1535 edge e2;
1536 edge_iterator ei;
1537 basic_block d1 = e1->dest;
1539 if (FORWARDER_BLOCK_P (d1))
1540 d1 = EDGE_SUCC (d1, 0)->dest;
1542 FOR_EACH_EDGE (e2, ei, bb1->succs)
1544 basic_block d2 = e2->dest;
1545 if (FORWARDER_BLOCK_P (d2))
1546 d2 = EDGE_SUCC (d2, 0)->dest;
1547 if (d1 == d2)
1548 break;
1551 if (!e2)
1552 return false;
1555 return true;
1558 /* Returns true if BB basic block has a preserve label. */
1560 static bool
1561 block_has_preserve_label (basic_block bb)
1563 return (bb
1564 && block_label (bb)
1565 && LABEL_PRESERVE_P (block_label (bb)));
1568 /* E1 and E2 are edges with the same destination block. Search their
1569 predecessors for common code. If found, redirect control flow from
1570 (maybe the middle of) E1->SRC to (maybe the middle of) E2->SRC. */
1572 static bool
1573 try_crossjump_to_edge (int mode, edge e1, edge e2)
1575 int nmatch;
1576 basic_block src1 = e1->src, src2 = e2->src;
1577 basic_block redirect_to, redirect_from, to_remove;
1578 rtx newpos1, newpos2;
1579 edge s;
1580 edge_iterator ei;
1582 newpos1 = newpos2 = NULL_RTX;
1584 /* If we have partitioned hot/cold basic blocks, it is a bad idea
1585 to try this optimization.
1587 Basic block partitioning may result in some jumps that appear to
1588 be optimizable (or blocks that appear to be mergeable), but which really
1589 must be left untouched (they are required to make it safely across
1590 partition boundaries). See the comments at the top of
1591 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
1593 if (flag_reorder_blocks_and_partition && reload_completed)
1594 return false;
1596 /* Search backward through forwarder blocks. We don't need to worry
1597 about multiple entry or chained forwarders, as they will be optimized
1598 away. We do this to look past the unconditional jump following a
1599 conditional jump that is required due to the current CFG shape. */
1600 if (single_pred_p (src1)
1601 && FORWARDER_BLOCK_P (src1))
1602 e1 = single_pred_edge (src1), src1 = e1->src;
1604 if (single_pred_p (src2)
1605 && FORWARDER_BLOCK_P (src2))
1606 e2 = single_pred_edge (src2), src2 = e2->src;
1608 /* Nothing to do if we reach ENTRY, or a common source block. */
1609 if (src1 == ENTRY_BLOCK_PTR || src2 == ENTRY_BLOCK_PTR)
1610 return false;
1611 if (src1 == src2)
1612 return false;
1614 /* Seeing more than 1 forwarder blocks would confuse us later... */
1615 if (FORWARDER_BLOCK_P (e1->dest)
1616 && FORWARDER_BLOCK_P (single_succ (e1->dest)))
1617 return false;
1619 if (FORWARDER_BLOCK_P (e2->dest)
1620 && FORWARDER_BLOCK_P (single_succ (e2->dest)))
1621 return false;
1623 /* Likewise with dead code (possibly newly created by the other optimizations
1624 of cfg_cleanup). */
1625 if (EDGE_COUNT (src1->preds) == 0 || EDGE_COUNT (src2->preds) == 0)
1626 return false;
1628 /* Look for the common insn sequence, part the first ... */
1629 if (!outgoing_edges_match (mode, src1, src2))
1630 return false;
1632 /* ... and part the second. */
1633 nmatch = flow_find_cross_jump (src1, src2, &newpos1, &newpos2);
1635 /* Don't proceed with the crossjump unless we found a sufficient number
1636 of matching instructions or the 'from' block was totally matched
1637 (such that its predecessors will hopefully be redirected and the
1638 block removed). */
1639 if ((nmatch < PARAM_VALUE (PARAM_MIN_CROSSJUMP_INSNS))
1640 && (newpos1 != BB_HEAD (src1)))
1641 return false;
1643 /* Avoid deleting preserve label when redirecting ABNORMAL edges. */
1644 if (block_has_preserve_label (e1->dest)
1645 && (e1->flags & EDGE_ABNORMAL))
1646 return false;
1648 /* Here we know that the insns in the end of SRC1 which are common with SRC2
1649 will be deleted.
1650 If we have tablejumps in the end of SRC1 and SRC2
1651 they have been already compared for equivalence in outgoing_edges_match ()
1652 so replace the references to TABLE1 by references to TABLE2. */
1654 rtx label1, label2;
1655 rtx table1, table2;
1657 if (tablejump_p (BB_END (src1), &label1, &table1)
1658 && tablejump_p (BB_END (src2), &label2, &table2)
1659 && label1 != label2)
1661 replace_label_data rr;
1662 rtx insn;
1664 /* Replace references to LABEL1 with LABEL2. */
1665 rr.r1 = label1;
1666 rr.r2 = label2;
1667 rr.update_label_nuses = true;
1668 for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
1670 /* Do not replace the label in SRC1->END because when deleting
1671 a block whose end is a tablejump, the tablejump referenced
1672 from the instruction is deleted too. */
1673 if (insn != BB_END (src1))
1674 for_each_rtx (&insn, replace_label, &rr);
1679 /* Avoid splitting if possible. We must always split when SRC2 has
1680 EH predecessor edges, or we may end up with basic blocks with both
1681 normal and EH predecessor edges. */
1682 if (newpos2 == BB_HEAD (src2)
1683 && !(EDGE_PRED (src2, 0)->flags & EDGE_EH))
1684 redirect_to = src2;
1685 else
1687 if (newpos2 == BB_HEAD (src2))
1689 /* Skip possible basic block header. */
1690 if (LABEL_P (newpos2))
1691 newpos2 = NEXT_INSN (newpos2);
1692 while (DEBUG_INSN_P (newpos2))
1693 newpos2 = NEXT_INSN (newpos2);
1694 if (NOTE_P (newpos2))
1695 newpos2 = NEXT_INSN (newpos2);
1696 while (DEBUG_INSN_P (newpos2))
1697 newpos2 = NEXT_INSN (newpos2);
1700 if (dump_file)
1701 fprintf (dump_file, "Splitting bb %i before %i insns\n",
1702 src2->index, nmatch);
1703 redirect_to = split_block (src2, PREV_INSN (newpos2))->dest;
1706 if (dump_file)
1707 fprintf (dump_file,
1708 "Cross jumping from bb %i to bb %i; %i common insns\n",
1709 src1->index, src2->index, nmatch);
1711 /* We may have some registers visible through the block. */
1712 df_set_bb_dirty (redirect_to);
1714 /* Recompute the frequencies and counts of outgoing edges. */
1715 FOR_EACH_EDGE (s, ei, redirect_to->succs)
1717 edge s2;
1718 edge_iterator ei;
1719 basic_block d = s->dest;
1721 if (FORWARDER_BLOCK_P (d))
1722 d = single_succ (d);
1724 FOR_EACH_EDGE (s2, ei, src1->succs)
1726 basic_block d2 = s2->dest;
1727 if (FORWARDER_BLOCK_P (d2))
1728 d2 = single_succ (d2);
1729 if (d == d2)
1730 break;
1733 s->count += s2->count;
1735 /* Take care to update possible forwarder blocks. We verified
1736 that there is no more than one in the chain, so we can't run
1737 into infinite loop. */
1738 if (FORWARDER_BLOCK_P (s->dest))
1740 single_succ_edge (s->dest)->count += s2->count;
1741 s->dest->count += s2->count;
1742 s->dest->frequency += EDGE_FREQUENCY (s);
1745 if (FORWARDER_BLOCK_P (s2->dest))
1747 single_succ_edge (s2->dest)->count -= s2->count;
1748 if (single_succ_edge (s2->dest)->count < 0)
1749 single_succ_edge (s2->dest)->count = 0;
1750 s2->dest->count -= s2->count;
1751 s2->dest->frequency -= EDGE_FREQUENCY (s);
1752 if (s2->dest->frequency < 0)
1753 s2->dest->frequency = 0;
1754 if (s2->dest->count < 0)
1755 s2->dest->count = 0;
1758 if (!redirect_to->frequency && !src1->frequency)
1759 s->probability = (s->probability + s2->probability) / 2;
1760 else
1761 s->probability
1762 = ((s->probability * redirect_to->frequency +
1763 s2->probability * src1->frequency)
1764 / (redirect_to->frequency + src1->frequency));
1767 /* Adjust count and frequency for the block. An earlier jump
1768 threading pass may have left the profile in an inconsistent
1769 state (see update_bb_profile_for_threading) so we must be
1770 prepared for overflows. */
1771 redirect_to->count += src1->count;
1772 redirect_to->frequency += src1->frequency;
1773 if (redirect_to->frequency > BB_FREQ_MAX)
1774 redirect_to->frequency = BB_FREQ_MAX;
1775 update_br_prob_note (redirect_to);
1777 /* Edit SRC1 to go to REDIRECT_TO at NEWPOS1. */
1779 /* Skip possible basic block header. */
1780 if (LABEL_P (newpos1))
1781 newpos1 = NEXT_INSN (newpos1);
1783 while (DEBUG_INSN_P (newpos1))
1784 newpos1 = NEXT_INSN (newpos1);
1786 if (NOTE_INSN_BASIC_BLOCK_P (newpos1))
1787 newpos1 = NEXT_INSN (newpos1);
1789 while (DEBUG_INSN_P (newpos1))
1790 newpos1 = NEXT_INSN (newpos1);
1792 redirect_from = split_block (src1, PREV_INSN (newpos1))->src;
1793 to_remove = single_succ (redirect_from);
1795 redirect_edge_and_branch_force (single_succ_edge (redirect_from), redirect_to);
1796 delete_basic_block (to_remove);
1798 update_forwarder_flag (redirect_from);
1799 if (redirect_to != src2)
1800 update_forwarder_flag (src2);
1802 return true;
1805 /* Search the predecessors of BB for common insn sequences. When found,
1806 share code between them by redirecting control flow. Return true if
1807 any changes made. */
1809 static bool
1810 try_crossjump_bb (int mode, basic_block bb)
1812 edge e, e2, fallthru;
1813 bool changed;
1814 unsigned max, ix, ix2;
1815 basic_block ev, ev2;
1817 /* Nothing to do if there is not at least two incoming edges. */
1818 if (EDGE_COUNT (bb->preds) < 2)
1819 return false;
1821 /* Don't crossjump if this block ends in a computed jump,
1822 unless we are optimizing for size. */
1823 if (optimize_bb_for_size_p (bb)
1824 && bb != EXIT_BLOCK_PTR
1825 && computed_jump_p (BB_END (bb)))
1826 return false;
1828 /* If we are partitioning hot/cold basic blocks, we don't want to
1829 mess up unconditional or indirect jumps that cross between hot
1830 and cold sections.
1832 Basic block partitioning may result in some jumps that appear to
1833 be optimizable (or blocks that appear to be mergeable), but which really
1834 must be left untouched (they are required to make it safely across
1835 partition boundaries). See the comments at the top of
1836 bb-reorder.c:partition_hot_cold_basic_blocks for complete details. */
1838 if (BB_PARTITION (EDGE_PRED (bb, 0)->src) !=
1839 BB_PARTITION (EDGE_PRED (bb, 1)->src)
1840 || (EDGE_PRED (bb, 0)->flags & EDGE_CROSSING))
1841 return false;
1843 /* It is always cheapest to redirect a block that ends in a branch to
1844 a block that falls through into BB, as that adds no branches to the
1845 program. We'll try that combination first. */
1846 fallthru = NULL;
1847 max = PARAM_VALUE (PARAM_MAX_CROSSJUMP_EDGES);
1849 if (EDGE_COUNT (bb->preds) > max)
1850 return false;
1852 fallthru = find_fallthru_edge (bb->preds);
1854 changed = false;
1855 for (ix = 0, ev = bb; ix < EDGE_COUNT (ev->preds); )
1857 e = EDGE_PRED (ev, ix);
1858 ix++;
1860 /* As noted above, first try with the fallthru predecessor (or, a
1861 fallthru predecessor if we are in cfglayout mode). */
1862 if (fallthru)
1864 /* Don't combine the fallthru edge into anything else.
1865 If there is a match, we'll do it the other way around. */
1866 if (e == fallthru)
1867 continue;
1868 /* If nothing changed since the last attempt, there is nothing
1869 we can do. */
1870 if (!first_pass
1871 && !((e->src->flags & BB_MODIFIED)
1872 || (fallthru->src->flags & BB_MODIFIED)))
1873 continue;
1875 if (try_crossjump_to_edge (mode, e, fallthru))
1877 changed = true;
1878 ix = 0;
1879 ev = bb;
1880 continue;
1884 /* Non-obvious work limiting check: Recognize that we're going
1885 to call try_crossjump_bb on every basic block. So if we have
1886 two blocks with lots of outgoing edges (a switch) and they
1887 share lots of common destinations, then we would do the
1888 cross-jump check once for each common destination.
1890 Now, if the blocks actually are cross-jump candidates, then
1891 all of their destinations will be shared. Which means that
1892 we only need check them for cross-jump candidacy once. We
1893 can eliminate redundant checks of crossjump(A,B) by arbitrarily
1894 choosing to do the check from the block for which the edge
1895 in question is the first successor of A. */
1896 if (EDGE_SUCC (e->src, 0) != e)
1897 continue;
1899 for (ix2 = 0, ev2 = bb; ix2 < EDGE_COUNT (ev2->preds); )
1901 e2 = EDGE_PRED (ev2, ix2);
1902 ix2++;
1904 if (e2 == e)
1905 continue;
1907 /* We've already checked the fallthru edge above. */
1908 if (e2 == fallthru)
1909 continue;
1911 /* The "first successor" check above only prevents multiple
1912 checks of crossjump(A,B). In order to prevent redundant
1913 checks of crossjump(B,A), require that A be the block
1914 with the lowest index. */
1915 if (e->src->index > e2->src->index)
1916 continue;
1918 /* If nothing changed since the last attempt, there is nothing
1919 we can do. */
1920 if (!first_pass
1921 && !((e->src->flags & BB_MODIFIED)
1922 || (e2->src->flags & BB_MODIFIED)))
1923 continue;
1925 if (try_crossjump_to_edge (mode, e, e2))
1927 changed = true;
1928 ev2 = bb;
1929 ix = 0;
1930 break;
1935 if (changed)
1936 crossjumps_occured = true;
1938 return changed;
1941 /* Search the successors of BB for common insn sequences. When found,
1942 share code between them by moving it across the basic block
1943 boundary. Return true if any changes made. */
1945 static bool
1946 try_head_merge_bb (basic_block bb)
1948 basic_block final_dest_bb = NULL;
1949 int max_match = INT_MAX;
1950 edge e0;
1951 rtx *headptr, *currptr, *nextptr;
1952 bool changed, moveall;
1953 unsigned ix;
1954 rtx e0_last_head, cond, move_before;
1955 unsigned nedges = EDGE_COUNT (bb->succs);
1956 rtx jump = BB_END (bb);
1957 regset live, live_union;
1959 /* Nothing to do if there is not at least two outgoing edges. */
1960 if (nedges < 2)
1961 return false;
1963 /* Don't crossjump if this block ends in a computed jump,
1964 unless we are optimizing for size. */
1965 if (optimize_bb_for_size_p (bb)
1966 && bb != EXIT_BLOCK_PTR
1967 && computed_jump_p (BB_END (bb)))
1968 return false;
1970 cond = get_condition (jump, &move_before, true, false);
1971 if (cond == NULL_RTX)
1972 move_before = jump;
1974 for (ix = 0; ix < nedges; ix++)
1975 if (EDGE_SUCC (bb, ix)->dest == EXIT_BLOCK_PTR)
1976 return false;
1978 for (ix = 0; ix < nedges; ix++)
1980 edge e = EDGE_SUCC (bb, ix);
1981 basic_block other_bb = e->dest;
1983 if (df_get_bb_dirty (other_bb))
1985 block_was_dirty = true;
1986 return false;
1989 if (e->flags & EDGE_ABNORMAL)
1990 return false;
1992 /* Normally, all destination blocks must only be reachable from this
1993 block, i.e. they must have one incoming edge.
1995 There is one special case we can handle, that of multiple consecutive
1996 jumps where the first jumps to one of the targets of the second jump.
1997 This happens frequently in switch statements for default labels.
1998 The structure is as follows:
1999 FINAL_DEST_BB
2000 ....
2001 if (cond) jump A;
2002 fall through
2004 jump with targets A, B, C, D...
2006 has two incoming edges, from FINAL_DEST_BB and BB
2008 In this case, we can try to move the insns through BB and into
2009 FINAL_DEST_BB. */
2010 if (EDGE_COUNT (other_bb->preds) != 1)
2012 edge incoming_edge, incoming_bb_other_edge;
2013 edge_iterator ei;
2015 if (final_dest_bb != NULL
2016 || EDGE_COUNT (other_bb->preds) != 2)
2017 return false;
2019 /* We must be able to move the insns across the whole block. */
2020 move_before = BB_HEAD (bb);
2021 while (!NONDEBUG_INSN_P (move_before))
2022 move_before = NEXT_INSN (move_before);
2024 if (EDGE_COUNT (bb->preds) != 1)
2025 return false;
2026 incoming_edge = EDGE_PRED (bb, 0);
2027 final_dest_bb = incoming_edge->src;
2028 if (EDGE_COUNT (final_dest_bb->succs) != 2)
2029 return false;
2030 FOR_EACH_EDGE (incoming_bb_other_edge, ei, final_dest_bb->succs)
2031 if (incoming_bb_other_edge != incoming_edge)
2032 break;
2033 if (incoming_bb_other_edge->dest != other_bb)
2034 return false;
2038 e0 = EDGE_SUCC (bb, 0);
2039 e0_last_head = NULL_RTX;
2040 changed = false;
2042 for (ix = 1; ix < nedges; ix++)
2044 edge e = EDGE_SUCC (bb, ix);
2045 rtx e0_last, e_last;
2046 int nmatch;
2048 nmatch = flow_find_head_matching_sequence (e0->dest, e->dest,
2049 &e0_last, &e_last, 0);
2050 if (nmatch == 0)
2051 return false;
2053 if (nmatch < max_match)
2055 max_match = nmatch;
2056 e0_last_head = e0_last;
2060 /* If we matched an entire block, we probably have to avoid moving the
2061 last insn. */
2062 if (max_match > 0
2063 && e0_last_head == BB_END (e0->dest)
2064 && (find_reg_note (e0_last_head, REG_EH_REGION, 0)
2065 || control_flow_insn_p (e0_last_head)))
2067 max_match--;
2068 if (max_match == 0)
2069 return false;
2071 e0_last_head = prev_real_insn (e0_last_head);
2072 while (DEBUG_INSN_P (e0_last_head));
2075 if (max_match == 0)
2076 return false;
2078 /* We must find a union of the live registers at each of the end points. */
2079 live = BITMAP_ALLOC (NULL);
2080 live_union = BITMAP_ALLOC (NULL);
2082 currptr = XNEWVEC (rtx, nedges);
2083 headptr = XNEWVEC (rtx, nedges);
2084 nextptr = XNEWVEC (rtx, nedges);
2086 for (ix = 0; ix < nedges; ix++)
2088 int j;
2089 basic_block merge_bb = EDGE_SUCC (bb, ix)->dest;
2090 rtx head = BB_HEAD (merge_bb);
2092 while (!NONDEBUG_INSN_P (head))
2093 head = NEXT_INSN (head);
2094 headptr[ix] = head;
2095 currptr[ix] = head;
2097 /* Compute the end point and live information */
2098 for (j = 1; j < max_match; j++)
2100 head = NEXT_INSN (head);
2101 while (!NONDEBUG_INSN_P (head));
2102 simulate_backwards_to_point (merge_bb, live, head);
2103 IOR_REG_SET (live_union, live);
2106 /* If we're moving across two blocks, verify the validity of the
2107 first move, then adjust the target and let the loop below deal
2108 with the final move. */
2109 if (final_dest_bb != NULL)
2111 rtx move_upto;
2113 moveall = can_move_insns_across (currptr[0], e0_last_head, move_before,
2114 jump, e0->dest, live_union,
2115 NULL, &move_upto);
2116 if (!moveall)
2118 if (move_upto == NULL_RTX)
2119 goto out;
2121 while (e0_last_head != move_upto)
2123 df_simulate_one_insn_backwards (e0->dest, e0_last_head,
2124 live_union);
2125 e0_last_head = PREV_INSN (e0_last_head);
2128 if (e0_last_head == NULL_RTX)
2129 goto out;
2131 jump = BB_END (final_dest_bb);
2132 cond = get_condition (jump, &move_before, true, false);
2133 if (cond == NULL_RTX)
2134 move_before = jump;
2139 rtx move_upto;
2140 moveall = can_move_insns_across (currptr[0], e0_last_head,
2141 move_before, jump, e0->dest, live_union,
2142 NULL, &move_upto);
2143 if (!moveall && move_upto == NULL_RTX)
2145 if (jump == move_before)
2146 break;
2148 /* Try again, using a different insertion point. */
2149 move_before = jump;
2151 #ifdef HAVE_cc0
2152 /* Don't try moving before a cc0 user, as that may invalidate
2153 the cc0. */
2154 if (reg_mentioned_p (cc0_rtx, jump))
2155 break;
2156 #endif
2158 continue;
2161 if (final_dest_bb && !moveall)
2162 /* We haven't checked whether a partial move would be OK for the first
2163 move, so we have to fail this case. */
2164 break;
2166 changed = true;
2167 for (;;)
2169 if (currptr[0] == move_upto)
2170 break;
2171 for (ix = 0; ix < nedges; ix++)
2173 rtx curr = currptr[ix];
2175 curr = NEXT_INSN (curr);
2176 while (!NONDEBUG_INSN_P (curr));
2177 currptr[ix] = curr;
2181 /* If we can't currently move all of the identical insns, remember
2182 each insn after the range that we'll merge. */
2183 if (!moveall)
2184 for (ix = 0; ix < nedges; ix++)
2186 rtx curr = currptr[ix];
2188 curr = NEXT_INSN (curr);
2189 while (!NONDEBUG_INSN_P (curr));
2190 nextptr[ix] = curr;
2193 reorder_insns (headptr[0], currptr[0], PREV_INSN (move_before));
2194 df_set_bb_dirty (EDGE_SUCC (bb, 0)->dest);
2195 if (final_dest_bb != NULL)
2196 df_set_bb_dirty (final_dest_bb);
2197 df_set_bb_dirty (bb);
2198 for (ix = 1; ix < nedges; ix++)
2200 df_set_bb_dirty (EDGE_SUCC (bb, ix)->dest);
2201 delete_insn_chain (headptr[ix], currptr[ix], false);
2203 if (!moveall)
2205 if (jump == move_before)
2206 break;
2208 /* For the unmerged insns, try a different insertion point. */
2209 move_before = jump;
2211 #ifdef HAVE_cc0
2212 /* Don't try moving before a cc0 user, as that may invalidate
2213 the cc0. */
2214 if (reg_mentioned_p (cc0_rtx, jump))
2215 break;
2216 #endif
2218 for (ix = 0; ix < nedges; ix++)
2219 currptr[ix] = headptr[ix] = nextptr[ix];
2222 while (!moveall);
2224 out:
2225 free (currptr);
2226 free (headptr);
2227 free (nextptr);
2229 crossjumps_occured |= changed;
2231 return changed;
2234 /* Return true if BB contains just bb note, or bb note followed
2235 by only DEBUG_INSNs. */
2237 static bool
2238 trivially_empty_bb_p (basic_block bb)
2240 rtx insn = BB_END (bb);
2242 while (1)
2244 if (insn == BB_HEAD (bb))
2245 return true;
2246 if (!DEBUG_INSN_P (insn))
2247 return false;
2248 insn = PREV_INSN (insn);
2252 /* Do simple CFG optimizations - basic block merging, simplifying of jump
2253 instructions etc. Return nonzero if changes were made. */
2255 static bool
2256 try_optimize_cfg (int mode)
2258 bool changed_overall = false;
2259 bool changed;
2260 int iterations = 0;
2261 basic_block bb, b, next;
2263 if (mode & (CLEANUP_CROSSJUMP | CLEANUP_THREADING))
2264 clear_bb_flags ();
2266 crossjumps_occured = false;
2268 FOR_EACH_BB (bb)
2269 update_forwarder_flag (bb);
2271 if (! targetm.cannot_modify_jumps_p ())
2273 first_pass = true;
2274 /* Attempt to merge blocks as made possible by edge removal. If
2275 a block has only one successor, and the successor has only
2276 one predecessor, they may be combined. */
2279 block_was_dirty = false;
2280 changed = false;
2281 iterations++;
2283 if (dump_file)
2284 fprintf (dump_file,
2285 "\n\ntry_optimize_cfg iteration %i\n\n",
2286 iterations);
2288 for (b = ENTRY_BLOCK_PTR->next_bb; b != EXIT_BLOCK_PTR;)
2290 basic_block c;
2291 edge s;
2292 bool changed_here = false;
2294 /* Delete trivially dead basic blocks. This is either
2295 blocks with no predecessors, or empty blocks with no
2296 successors. However if the empty block with no
2297 successors is the successor of the ENTRY_BLOCK, it is
2298 kept. This ensures that the ENTRY_BLOCK will have a
2299 successor which is a precondition for many RTL
2300 passes. Empty blocks may result from expanding
2301 __builtin_unreachable (). */
2302 if (EDGE_COUNT (b->preds) == 0
2303 || (EDGE_COUNT (b->succs) == 0
2304 && trivially_empty_bb_p (b)
2305 && single_succ_edge (ENTRY_BLOCK_PTR)->dest != b))
2307 c = b->prev_bb;
2308 if (EDGE_COUNT (b->preds) > 0)
2310 edge e;
2311 edge_iterator ei;
2313 if (current_ir_type () == IR_RTL_CFGLAYOUT)
2315 if (b->il.rtl->footer
2316 && BARRIER_P (b->il.rtl->footer))
2317 FOR_EACH_EDGE (e, ei, b->preds)
2318 if ((e->flags & EDGE_FALLTHRU)
2319 && e->src->il.rtl->footer == NULL)
2321 if (b->il.rtl->footer)
2323 e->src->il.rtl->footer = b->il.rtl->footer;
2324 b->il.rtl->footer = NULL;
2326 else
2328 start_sequence ();
2329 e->src->il.rtl->footer = emit_barrier ();
2330 end_sequence ();
2334 else
2336 rtx last = get_last_bb_insn (b);
2337 if (last && BARRIER_P (last))
2338 FOR_EACH_EDGE (e, ei, b->preds)
2339 if ((e->flags & EDGE_FALLTHRU))
2340 emit_barrier_after (BB_END (e->src));
2343 delete_basic_block (b);
2344 if (!(mode & CLEANUP_CFGLAYOUT))
2345 changed = true;
2346 /* Avoid trying to remove ENTRY_BLOCK_PTR. */
2347 b = (c == ENTRY_BLOCK_PTR ? c->next_bb : c);
2348 continue;
2351 /* Remove code labels no longer used. */
2352 if (single_pred_p (b)
2353 && (single_pred_edge (b)->flags & EDGE_FALLTHRU)
2354 && !(single_pred_edge (b)->flags & EDGE_COMPLEX)
2355 && LABEL_P (BB_HEAD (b))
2356 /* If the previous block ends with a branch to this
2357 block, we can't delete the label. Normally this
2358 is a condjump that is yet to be simplified, but
2359 if CASE_DROPS_THRU, this can be a tablejump with
2360 some element going to the same place as the
2361 default (fallthru). */
2362 && (single_pred (b) == ENTRY_BLOCK_PTR
2363 || !JUMP_P (BB_END (single_pred (b)))
2364 || ! label_is_jump_target_p (BB_HEAD (b),
2365 BB_END (single_pred (b)))))
2367 rtx label = BB_HEAD (b);
2369 delete_insn_chain (label, label, false);
2370 /* If the case label is undeletable, move it after the
2371 BASIC_BLOCK note. */
2372 if (NOTE_KIND (BB_HEAD (b)) == NOTE_INSN_DELETED_LABEL)
2374 rtx bb_note = NEXT_INSN (BB_HEAD (b));
2376 reorder_insns_nobb (label, label, bb_note);
2377 BB_HEAD (b) = bb_note;
2378 if (BB_END (b) == bb_note)
2379 BB_END (b) = label;
2381 if (dump_file)
2382 fprintf (dump_file, "Deleted label in block %i.\n",
2383 b->index);
2386 /* If we fall through an empty block, we can remove it. */
2387 if (!(mode & CLEANUP_CFGLAYOUT)
2388 && single_pred_p (b)
2389 && (single_pred_edge (b)->flags & EDGE_FALLTHRU)
2390 && !LABEL_P (BB_HEAD (b))
2391 && FORWARDER_BLOCK_P (b)
2392 /* Note that forwarder_block_p true ensures that
2393 there is a successor for this block. */
2394 && (single_succ_edge (b)->flags & EDGE_FALLTHRU)
2395 && n_basic_blocks > NUM_FIXED_BLOCKS + 1)
2397 if (dump_file)
2398 fprintf (dump_file,
2399 "Deleting fallthru block %i.\n",
2400 b->index);
2402 c = b->prev_bb == ENTRY_BLOCK_PTR ? b->next_bb : b->prev_bb;
2403 redirect_edge_succ_nodup (single_pred_edge (b),
2404 single_succ (b));
2405 delete_basic_block (b);
2406 changed = true;
2407 b = c;
2408 continue;
2411 /* Merge B with its single successor, if any. */
2412 if (single_succ_p (b)
2413 && (s = single_succ_edge (b))
2414 && !(s->flags & EDGE_COMPLEX)
2415 && (c = s->dest) != EXIT_BLOCK_PTR
2416 && single_pred_p (c)
2417 && b != c)
2419 /* When not in cfg_layout mode use code aware of reordering
2420 INSN. This code possibly creates new basic blocks so it
2421 does not fit merge_blocks interface and is kept here in
2422 hope that it will become useless once more of compiler
2423 is transformed to use cfg_layout mode. */
2425 if ((mode & CLEANUP_CFGLAYOUT)
2426 && can_merge_blocks_p (b, c))
2428 merge_blocks (b, c);
2429 update_forwarder_flag (b);
2430 changed_here = true;
2432 else if (!(mode & CLEANUP_CFGLAYOUT)
2433 /* If the jump insn has side effects,
2434 we can't kill the edge. */
2435 && (!JUMP_P (BB_END (b))
2436 || (reload_completed
2437 ? simplejump_p (BB_END (b))
2438 : (onlyjump_p (BB_END (b))
2439 && !tablejump_p (BB_END (b),
2440 NULL, NULL))))
2441 && (next = merge_blocks_move (s, b, c, mode)))
2443 b = next;
2444 changed_here = true;
2448 /* Simplify branch over branch. */
2449 if ((mode & CLEANUP_EXPENSIVE)
2450 && !(mode & CLEANUP_CFGLAYOUT)
2451 && try_simplify_condjump (b))
2452 changed_here = true;
2454 /* If B has a single outgoing edge, but uses a
2455 non-trivial jump instruction without side-effects, we
2456 can either delete the jump entirely, or replace it
2457 with a simple unconditional jump. */
2458 if (single_succ_p (b)
2459 && single_succ (b) != EXIT_BLOCK_PTR
2460 && onlyjump_p (BB_END (b))
2461 && !find_reg_note (BB_END (b), REG_CROSSING_JUMP, NULL_RTX)
2462 && try_redirect_by_replacing_jump (single_succ_edge (b),
2463 single_succ (b),
2464 (mode & CLEANUP_CFGLAYOUT) != 0))
2466 update_forwarder_flag (b);
2467 changed_here = true;
2470 /* Simplify branch to branch. */
2471 if (try_forward_edges (mode, b))
2472 changed_here = true;
2474 /* Look for shared code between blocks. */
2475 if ((mode & CLEANUP_CROSSJUMP)
2476 && try_crossjump_bb (mode, b))
2477 changed_here = true;
2479 if ((mode & CLEANUP_CROSSJUMP)
2480 /* This can lengthen register lifetimes. Do it only after
2481 reload. */
2482 && reload_completed
2483 && try_head_merge_bb (b))
2484 changed_here = true;
2486 /* Don't get confused by the index shift caused by
2487 deleting blocks. */
2488 if (!changed_here)
2489 b = b->next_bb;
2490 else
2491 changed = true;
2494 if ((mode & CLEANUP_CROSSJUMP)
2495 && try_crossjump_bb (mode, EXIT_BLOCK_PTR))
2496 changed = true;
2498 if (block_was_dirty)
2500 /* This should only be set by head-merging. */
2501 gcc_assert (mode & CLEANUP_CROSSJUMP);
2502 df_analyze ();
2505 #ifdef ENABLE_CHECKING
2506 if (changed)
2507 verify_flow_info ();
2508 #endif
2510 changed_overall |= changed;
2511 first_pass = false;
2513 while (changed);
2516 FOR_ALL_BB (b)
2517 b->flags &= ~(BB_FORWARDER_BLOCK | BB_NONTHREADABLE_BLOCK);
2519 return changed_overall;
2522 /* Delete all unreachable basic blocks. */
2524 bool
2525 delete_unreachable_blocks (void)
2527 bool changed = false;
2528 basic_block b, prev_bb;
2530 find_unreachable_blocks ();
2532 /* When we're in GIMPLE mode and there may be debug insns, we should
2533 delete blocks in reverse dominator order, so as to get a chance
2534 to substitute all released DEFs into debug stmts. If we don't
2535 have dominators information, walking blocks backward gets us a
2536 better chance of retaining most debug information than
2537 otherwise. */
2538 if (MAY_HAVE_DEBUG_STMTS && current_ir_type () == IR_GIMPLE
2539 && dom_info_available_p (CDI_DOMINATORS))
2541 for (b = EXIT_BLOCK_PTR->prev_bb; b != ENTRY_BLOCK_PTR; b = prev_bb)
2543 prev_bb = b->prev_bb;
2545 if (!(b->flags & BB_REACHABLE))
2547 /* Speed up the removal of blocks that don't dominate
2548 others. Walking backwards, this should be the common
2549 case. */
2550 if (!first_dom_son (CDI_DOMINATORS, b))
2551 delete_basic_block (b);
2552 else
2554 VEC (basic_block, heap) *h
2555 = get_all_dominated_blocks (CDI_DOMINATORS, b);
2557 while (VEC_length (basic_block, h))
2559 b = VEC_pop (basic_block, h);
2561 prev_bb = b->prev_bb;
2563 gcc_assert (!(b->flags & BB_REACHABLE));
2565 delete_basic_block (b);
2568 VEC_free (basic_block, heap, h);
2571 changed = true;
2575 else
2577 for (b = EXIT_BLOCK_PTR->prev_bb; b != ENTRY_BLOCK_PTR; b = prev_bb)
2579 prev_bb = b->prev_bb;
2581 if (!(b->flags & BB_REACHABLE))
2583 delete_basic_block (b);
2584 changed = true;
2589 if (changed)
2590 tidy_fallthru_edges ();
2591 return changed;
2594 /* Delete any jump tables never referenced. We can't delete them at the
2595 time of removing tablejump insn as they are referenced by the preceding
2596 insns computing the destination, so we delay deleting and garbagecollect
2597 them once life information is computed. */
2598 void
2599 delete_dead_jumptables (void)
2601 basic_block bb;
2603 /* A dead jump table does not belong to any basic block. Scan insns
2604 between two adjacent basic blocks. */
2605 FOR_EACH_BB (bb)
2607 rtx insn, next;
2609 for (insn = NEXT_INSN (BB_END (bb));
2610 insn && !NOTE_INSN_BASIC_BLOCK_P (insn);
2611 insn = next)
2613 next = NEXT_INSN (insn);
2614 if (LABEL_P (insn)
2615 && LABEL_NUSES (insn) == LABEL_PRESERVE_P (insn)
2616 && JUMP_TABLE_DATA_P (next))
2618 rtx label = insn, jump = next;
2620 if (dump_file)
2621 fprintf (dump_file, "Dead jumptable %i removed\n",
2622 INSN_UID (insn));
2624 next = NEXT_INSN (next);
2625 delete_insn (jump);
2626 delete_insn (label);
2633 /* Tidy the CFG by deleting unreachable code and whatnot. */
2635 bool
2636 cleanup_cfg (int mode)
2638 bool changed = false;
2640 /* Set the cfglayout mode flag here. We could update all the callers
2641 but that is just inconvenient, especially given that we eventually
2642 want to have cfglayout mode as the default. */
2643 if (current_ir_type () == IR_RTL_CFGLAYOUT)
2644 mode |= CLEANUP_CFGLAYOUT;
2646 timevar_push (TV_CLEANUP_CFG);
2647 if (delete_unreachable_blocks ())
2649 changed = true;
2650 /* We've possibly created trivially dead code. Cleanup it right
2651 now to introduce more opportunities for try_optimize_cfg. */
2652 if (!(mode & (CLEANUP_NO_INSN_DEL))
2653 && !reload_completed)
2654 delete_trivially_dead_insns (get_insns (), max_reg_num ());
2657 compact_blocks ();
2659 /* To tail-merge blocks ending in the same noreturn function (e.g.
2660 a call to abort) we have to insert fake edges to exit. Do this
2661 here once. The fake edges do not interfere with any other CFG
2662 cleanups. */
2663 if (mode & CLEANUP_CROSSJUMP)
2664 add_noreturn_fake_exit_edges ();
2666 if (!dbg_cnt (cfg_cleanup))
2667 return changed;
2669 while (try_optimize_cfg (mode))
2671 delete_unreachable_blocks (), changed = true;
2672 if (!(mode & CLEANUP_NO_INSN_DEL))
2674 /* Try to remove some trivially dead insns when doing an expensive
2675 cleanup. But delete_trivially_dead_insns doesn't work after
2676 reload (it only handles pseudos) and run_fast_dce is too costly
2677 to run in every iteration.
2679 For effective cross jumping, we really want to run a fast DCE to
2680 clean up any dead conditions, or they get in the way of performing
2681 useful tail merges.
2683 Other transformations in cleanup_cfg are not so sensitive to dead
2684 code, so delete_trivially_dead_insns or even doing nothing at all
2685 is good enough. */
2686 if ((mode & CLEANUP_EXPENSIVE) && !reload_completed
2687 && !delete_trivially_dead_insns (get_insns (), max_reg_num ()))
2688 break;
2689 if ((mode & CLEANUP_CROSSJUMP) && crossjumps_occured)
2690 run_fast_dce ();
2692 else
2693 break;
2696 if (mode & CLEANUP_CROSSJUMP)
2697 remove_fake_exit_edges ();
2699 /* Don't call delete_dead_jumptables in cfglayout mode, because
2700 that function assumes that jump tables are in the insns stream.
2701 But we also don't _have_ to delete dead jumptables in cfglayout
2702 mode because we shouldn't even be looking at things that are
2703 not in a basic block. Dead jumptables are cleaned up when
2704 going out of cfglayout mode. */
2705 if (!(mode & CLEANUP_CFGLAYOUT))
2706 delete_dead_jumptables ();
2708 timevar_pop (TV_CLEANUP_CFG);
2710 return changed;
2713 static unsigned int
2714 rest_of_handle_jump (void)
2716 if (crtl->tail_call_emit)
2717 fixup_tail_calls ();
2718 return 0;
2721 struct rtl_opt_pass pass_jump =
2724 RTL_PASS,
2725 "sibling", /* name */
2726 NULL, /* gate */
2727 rest_of_handle_jump, /* execute */
2728 NULL, /* sub */
2729 NULL, /* next */
2730 0, /* static_pass_number */
2731 TV_JUMP, /* tv_id */
2732 0, /* properties_required */
2733 0, /* properties_provided */
2734 0, /* properties_destroyed */
2735 TODO_ggc_collect, /* todo_flags_start */
2736 TODO_verify_flow, /* todo_flags_finish */
2741 static unsigned int
2742 rest_of_handle_jump2 (void)
2744 delete_trivially_dead_insns (get_insns (), max_reg_num ());
2745 if (dump_file)
2746 dump_flow_info (dump_file, dump_flags);
2747 cleanup_cfg ((optimize ? CLEANUP_EXPENSIVE : 0)
2748 | (flag_thread_jumps ? CLEANUP_THREADING : 0));
2749 return 0;
2753 struct rtl_opt_pass pass_jump2 =
2756 RTL_PASS,
2757 "jump", /* name */
2758 NULL, /* gate */
2759 rest_of_handle_jump2, /* execute */
2760 NULL, /* sub */
2761 NULL, /* next */
2762 0, /* static_pass_number */
2763 TV_JUMP, /* tv_id */
2764 0, /* properties_required */
2765 0, /* properties_provided */
2766 0, /* properties_destroyed */
2767 TODO_ggc_collect, /* todo_flags_start */
2768 TODO_dump_func | TODO_verify_rtl_sharing,/* todo_flags_finish */