Line data Source code
1 : /* Control flow optimization code for GNU compiler.
2 : Copyright (C) 1987-2026 Free Software Foundation, Inc.
3 :
4 : This file is part of GCC.
5 :
6 : GCC is free software; you can redistribute it and/or modify it under
7 : the terms of the GNU General Public License as published by the Free
8 : Software Foundation; either version 3, or (at your option) any later
9 : version.
10 :
11 : GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 : WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 : FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 : for more details.
15 :
16 : You should have received a copy of the GNU General Public License
17 : along with GCC; see the file COPYING3. If not see
18 : <http://www.gnu.org/licenses/>. */
19 :
20 : /* This file contains optimizer of the control flow. The main entry point is
21 : cleanup_cfg. Following optimizations are performed:
22 :
23 : - Unreachable blocks removal
24 : - Edge forwarding (edge to the forwarder block is forwarded to its
25 : successor. Simplification of the branch instruction is performed by
26 : underlying infrastructure so branch can be converted to simplejump or
27 : eliminated).
28 : - Cross jumping (tail merging)
29 : - Conditional jump-around-simplejump simplification
30 : - Basic block merging. */
31 :
32 : #include "config.h"
33 : #include "system.h"
34 : #include "coretypes.h"
35 : #include "backend.h"
36 : #include "target.h"
37 : #include "rtl.h"
38 : #include "tree.h"
39 : #include "cfghooks.h"
40 : #include "df.h"
41 : #include "memmodel.h"
42 : #include "tm_p.h"
43 : #include "insn-config.h"
44 : #include "emit-rtl.h"
45 : #include "cselib.h"
46 : #include "tree-pass.h"
47 : #include "cfgloop.h"
48 : #include "cfgrtl.h"
49 : #include "cfganal.h"
50 : #include "cfgbuild.h"
51 : #include "cfgcleanup.h"
52 : #include "dce.h"
53 : #include "dbgcnt.h"
54 : #include "rtl-iter.h"
55 : #include "regs.h"
56 : #include "function-abi.h"
57 :
58 : #define FORWARDER_BLOCK_P(BB) ((BB)->flags & BB_FORWARDER_BLOCK)
59 :
60 : /* Set to true when we are running first pass of try_optimize_cfg loop. */
61 : static bool first_pass;
62 :
63 : /* Set to true if crossjumps occurred in the latest run of try_optimize_cfg. */
64 : static bool crossjumps_occurred;
65 :
66 : /* Set to true if we couldn't run an optimization due to stale liveness
67 : information; we should run df_analyze to enable more opportunities. */
68 : static bool block_was_dirty;
69 :
70 : static bool try_crossjump_to_edge (int, edge, edge, enum replace_direction);
71 : static bool try_crossjump_bb (int, basic_block);
72 : static bool outgoing_edges_match (int, basic_block, basic_block);
73 : static enum replace_direction old_insns_match_p (int, rtx_insn *, rtx_insn *);
74 :
75 : static void merge_blocks_move_predecessor_nojumps (basic_block, basic_block);
76 : static void merge_blocks_move_successor_nojumps (basic_block, basic_block);
77 : static bool try_optimize_cfg (int);
78 : static bool try_simplify_condjump (basic_block);
79 : static bool try_forward_edges (int, basic_block);
80 : static edge thread_jump (edge, basic_block);
81 : static bool mark_effect (rtx, bitmap);
82 : static void notice_new_block (basic_block);
83 : static void update_forwarder_flag (basic_block);
84 : static void merge_memattrs (rtx, rtx);
85 :
86 : /* Set flags for newly created block. */
87 :
88 : static void
89 20877 : notice_new_block (basic_block bb)
90 : {
91 20877 : if (!bb)
92 : return;
93 :
94 14986 : if (forwarder_block_p (bb))
95 14986 : bb->flags |= BB_FORWARDER_BLOCK;
96 : }
97 :
98 : /* Recompute forwarder flag after block has been modified. */
99 :
100 : static void
101 321415162 : update_forwarder_flag (basic_block bb)
102 : {
103 321415162 : if (forwarder_block_p (bb))
104 16885431 : bb->flags |= BB_FORWARDER_BLOCK;
105 : else
106 304529731 : bb->flags &= ~BB_FORWARDER_BLOCK;
107 321415162 : }
108 :
109 : /* Simplify a conditional jump around an unconditional jump.
110 : Return true if something changed. */
111 :
112 : static bool
113 35378687 : try_simplify_condjump (basic_block cbranch_block)
114 : {
115 35378687 : basic_block jump_block, jump_dest_block, cbranch_dest_block;
116 35378687 : edge cbranch_jump_edge, cbranch_fallthru_edge;
117 35378687 : rtx_insn *cbranch_insn;
118 :
119 : /* Verify that there are exactly two successors. */
120 35378687 : if (EDGE_COUNT (cbranch_block->succs) != 2)
121 : return false;
122 :
123 : /* Verify that we've got a normal conditional branch at the end
124 : of the block. */
125 17773906 : cbranch_insn = BB_END (cbranch_block);
126 17773906 : if (!any_condjump_p (cbranch_insn))
127 : return false;
128 :
129 15466911 : cbranch_fallthru_edge = FALLTHRU_EDGE (cbranch_block);
130 15466911 : cbranch_jump_edge = BRANCH_EDGE (cbranch_block);
131 :
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 15466911 : jump_block = cbranch_fallthru_edge->dest;
136 15466911 : if (!single_pred_p (jump_block)
137 13737563 : || jump_block->next_bb == EXIT_BLOCK_PTR_FOR_FN (cfun)
138 29038651 : || !FORWARDER_BLOCK_P (jump_block))
139 : return false;
140 1726018 : jump_dest_block = single_succ (jump_block);
141 :
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.
145 :
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.cc:partition_hot_cold_basic_blocks for complete details. */
151 :
152 1726018 : if (BB_PARTITION (jump_block) != BB_PARTITION (jump_dest_block)
153 1686739 : || (cbranch_jump_edge->flags & EDGE_CROSSING))
154 : return false;
155 :
156 : /* The conditional branch must target the block after the
157 : unconditional branch. */
158 1440104 : cbranch_dest_block = cbranch_jump_edge->dest;
159 :
160 1440104 : if (cbranch_dest_block == EXIT_BLOCK_PTR_FOR_FN (cfun)
161 1440104 : || jump_dest_block == EXIT_BLOCK_PTR_FOR_FN (cfun)
162 2879392 : || !can_fallthru (jump_block, cbranch_dest_block))
163 : return false;
164 :
165 : /* Invert the conditional branch. */
166 7527 : if (!invert_jump (as_a <rtx_jump_insn *> (cbranch_insn),
167 7527 : block_label (jump_dest_block), 0))
168 : return false;
169 :
170 7527 : if (dump_file)
171 1 : fprintf (dump_file, "Simplifying condjump %i around jump %i\n",
172 1 : INSN_UID (cbranch_insn), INSN_UID (BB_END (jump_block)));
173 :
174 : /* Success. Update the CFG to match. Note that after this point
175 : the edge variable names appear backwards; the redirection is done
176 : this way to preserve edge profile data. */
177 7527 : cbranch_jump_edge = redirect_edge_succ_nodup (cbranch_jump_edge,
178 : cbranch_dest_block);
179 7527 : cbranch_fallthru_edge = redirect_edge_succ_nodup (cbranch_fallthru_edge,
180 : jump_dest_block);
181 7527 : cbranch_jump_edge->flags |= EDGE_FALLTHRU;
182 7527 : cbranch_fallthru_edge->flags &= ~EDGE_FALLTHRU;
183 7527 : update_br_prob_note (cbranch_block);
184 :
185 : /* Delete the block with the unconditional jump, and clean up the mess. */
186 7527 : delete_basic_block (jump_block);
187 7527 : tidy_fallthru_edge (cbranch_jump_edge);
188 7527 : update_forwarder_flag (cbranch_block);
189 :
190 7527 : return true;
191 : }
192 :
193 : /* Attempt to prove that operation is NOOP using CSElib or mark the effect
194 : on register. Used by jump threading. */
195 :
196 : static bool
197 34607905 : mark_effect (rtx exp, regset nonequal)
198 : {
199 34607905 : rtx dest;
200 34607905 : switch (GET_CODE (exp))
201 : {
202 : /* In case we do clobber the register, mark it as equal, as we know the
203 : value is dead so it don't have to match. */
204 2535260 : case CLOBBER:
205 2535260 : dest = XEXP (exp, 0);
206 2535260 : if (REG_P (dest))
207 2501414 : bitmap_clear_range (nonequal, REGNO (dest), REG_NREGS (dest));
208 : return false;
209 :
210 11346430 : case SET:
211 11346430 : if (cselib_redundant_set_p (exp))
212 : return false;
213 11154981 : dest = SET_DEST (exp);
214 11154981 : if (dest == pc_rtx)
215 : return false;
216 11147140 : if (!REG_P (dest))
217 : return true;
218 10589643 : bitmap_set_range (nonequal, REGNO (dest), REG_NREGS (dest));
219 10589643 : return false;
220 :
221 : default:
222 : return false;
223 : }
224 : }
225 :
226 : /* Return true if X contains a register in NONEQUAL. */
227 : static bool
228 3661804 : mentions_nonequal_regs (const_rtx x, regset nonequal)
229 : {
230 3661804 : subrtx_iterator::array_type array;
231 7340430 : FOR_EACH_SUBRTX (iter, array, x, NONCONST)
232 : {
233 7332589 : const_rtx x = *iter;
234 7332589 : if (REG_P (x))
235 : {
236 3662089 : unsigned int end_regno = END_REGNO (x);
237 3670215 : for (unsigned int regno = REGNO (x); regno < end_regno; ++regno)
238 3662089 : if (REGNO_REG_SET_P (nonequal, regno))
239 3653963 : return true;
240 : }
241 : }
242 7841 : return false;
243 3661804 : }
244 :
245 : /* Attempt to prove that the basic block B will have no side effects and
246 : always continues in the same edge if reached via E. Return the edge
247 : if exist, NULL otherwise. */
248 :
249 : static edge
250 28947251 : thread_jump (edge e, basic_block b)
251 : {
252 28947251 : rtx set1, set2, cond1, cond2;
253 28947251 : rtx_insn *insn;
254 28947251 : enum rtx_code code1, code2, reversed_code2;
255 28947251 : bool reverse1 = false;
256 28947251 : unsigned i;
257 28947251 : regset nonequal;
258 28947251 : bool failed = false;
259 :
260 : /* Jump threading may cause fixup_partitions to introduce new crossing edges,
261 : which is not allowed after reload. */
262 28947251 : gcc_checking_assert (!reload_completed || !crtl->has_bb_partition);
263 :
264 28947251 : if (b->flags & BB_NONTHREADABLE_BLOCK)
265 : return NULL;
266 :
267 : /* At the moment, we do handle only conditional jumps, but later we may
268 : want to extend this code to tablejumps and others. */
269 24892569 : if (EDGE_COUNT (e->src->succs) != 2)
270 : return NULL;
271 17493196 : if (EDGE_COUNT (b->succs) != 2)
272 : {
273 7712227 : b->flags |= BB_NONTHREADABLE_BLOCK;
274 7712227 : return NULL;
275 : }
276 :
277 : /* Second branch must end with onlyjump, as we will eliminate the jump. */
278 9780969 : if (!any_condjump_p (BB_END (e->src)))
279 : return NULL;
280 :
281 8873587 : if (!any_condjump_p (BB_END (b)) || !onlyjump_p (BB_END (b)))
282 : {
283 672076 : b->flags |= BB_NONTHREADABLE_BLOCK;
284 672076 : return NULL;
285 : }
286 :
287 8201511 : set1 = pc_set (BB_END (e->src));
288 8201511 : set2 = pc_set (BB_END (b));
289 8201511 : if (((e->flags & EDGE_FALLTHRU) != 0)
290 8201511 : != (XEXP (SET_SRC (set1), 1) == pc_rtx))
291 4059149 : reverse1 = true;
292 :
293 8201511 : cond1 = XEXP (SET_SRC (set1), 0);
294 8201511 : cond2 = XEXP (SET_SRC (set2), 0);
295 8201511 : if (reverse1)
296 4059149 : code1 = reversed_comparison_code (cond1, BB_END (e->src));
297 : else
298 4142362 : code1 = GET_CODE (cond1);
299 :
300 8201511 : code2 = GET_CODE (cond2);
301 8201511 : reversed_code2 = reversed_comparison_code (cond2, BB_END (b));
302 :
303 8201511 : if (!comparison_dominates_p (code1, code2)
304 8201511 : && !comparison_dominates_p (code1, reversed_code2))
305 : return NULL;
306 :
307 : /* Ensure that the comparison operators are equivalent.
308 : ??? This is far too pessimistic. We should allow swapped operands,
309 : different CCmodes, or for example comparisons for interval, that
310 : dominate even when operands are not equivalent. */
311 6597141 : if (!rtx_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
312 6597141 : || !rtx_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
313 : return NULL;
314 :
315 : /* Punt if BB_END (e->src) is doloop-like conditional jump that modifies
316 : the registers used in cond1. */
317 5662799 : if (modified_in_p (cond1, BB_END (e->src)))
318 : return NULL;
319 :
320 : /* Short circuit cases where block B contains some side effects, as we can't
321 : safely bypass it. */
322 65914811 : for (insn = NEXT_INSN (BB_HEAD (b)); insn != NEXT_INSN (BB_END (b));
323 54589213 : insn = NEXT_INSN (insn))
324 56032711 : if (INSN_P (insn) && side_effects_p (PATTERN (insn)))
325 : {
326 1443498 : b->flags |= BB_NONTHREADABLE_BLOCK;
327 1443498 : return NULL;
328 : }
329 :
330 4219301 : cselib_init (0);
331 :
332 : /* First process all values computed in the source basic block. */
333 60988372 : for (insn = NEXT_INSN (BB_HEAD (e->src));
334 60988372 : insn != NEXT_INSN (BB_END (e->src));
335 56769071 : insn = NEXT_INSN (insn))
336 56769071 : if (INSN_P (insn))
337 52996318 : cselib_process_insn (insn);
338 :
339 4219301 : nonequal = BITMAP_ALLOC (NULL);
340 4219301 : CLEAR_REG_SET (nonequal);
341 :
342 : /* Now assume that we've continued by the edge E to B and continue
343 : processing as if it were same basic block.
344 : Our goal is to prove that whole block is an NOOP. */
345 :
346 39315469 : for (insn = NEXT_INSN (BB_HEAD (b));
347 39315469 : insn != NEXT_INSN (BB_END (b)) && !failed;
348 35096168 : insn = NEXT_INSN (insn))
349 : {
350 : /* cond2 must not mention any register that is not equal to the
351 : former block. Check this before processing that instruction,
352 : as BB_END (b) could contain also clobbers. */
353 38750131 : if (insn == BB_END (b)
354 38750131 : && mentions_nonequal_regs (cond2, nonequal))
355 3653963 : goto failed_exit;
356 :
357 35096168 : if (INSN_P (insn))
358 : {
359 31972196 : rtx pat = PATTERN (insn);
360 :
361 31972196 : if (GET_CODE (pat) == PARALLEL)
362 : {
363 7725701 : for (i = 0; i < (unsigned)XVECLEN (pat, 0); i++)
364 5180705 : failed |= mark_effect (XVECEXP (pat, 0, i), nonequal);
365 : }
366 : else
367 29427200 : failed |= mark_effect (pat, nonequal);
368 : }
369 :
370 35096168 : cselib_process_insn (insn);
371 : }
372 :
373 : /* Later we should clear nonequal of dead registers. So far we don't
374 : have life information in cfg_cleanup. */
375 565338 : if (failed)
376 : {
377 557497 : b->flags |= BB_NONTHREADABLE_BLOCK;
378 557497 : goto failed_exit;
379 : }
380 :
381 7841 : if (!REG_SET_EMPTY_P (nonequal))
382 539 : goto failed_exit;
383 :
384 7302 : BITMAP_FREE (nonequal);
385 7302 : cselib_finish ();
386 7302 : if ((comparison_dominates_p (code1, code2) != 0)
387 7302 : != (XEXP (SET_SRC (set2), 1) == pc_rtx))
388 5144 : return BRANCH_EDGE (b);
389 : else
390 2158 : return FALLTHRU_EDGE (b);
391 :
392 4211999 : failed_exit:
393 4211999 : BITMAP_FREE (nonequal);
394 4211999 : cselib_finish ();
395 4211999 : return NULL;
396 : }
397 :
398 : /* Attempt to forward edges leaving basic block B.
399 : Return true if successful. */
400 :
401 : static bool
402 411192794 : try_forward_edges (int mode, basic_block b)
403 : {
404 411192794 : bool changed = false;
405 411192794 : edge_iterator ei;
406 411192794 : edge e, *threaded_edges = NULL;
407 :
408 1030858015 : for (ei = ei_start (b->succs); (e = ei_safe_edge (ei)); )
409 : {
410 619665221 : basic_block target, first;
411 619665221 : location_t goto_locus;
412 619665221 : int counter;
413 619665221 : bool threaded = false;
414 619665221 : int nthreaded_edges = 0;
415 619665221 : bool may_thread = first_pass || (b->flags & BB_MODIFIED) != 0;
416 619665221 : bool new_target_threaded = false;
417 :
418 : /* Skip complex edges because we don't know how to update them.
419 :
420 : Still handle fallthru edges, as we can succeed to forward fallthru
421 : edge to the same place as the branch edge of conditional branch
422 : and turn conditional branch to an unconditional branch. */
423 619665221 : if (e->flags & EDGE_COMPLEX)
424 : {
425 36219621 : ei_next (&ei);
426 36219621 : continue;
427 : }
428 :
429 583445600 : target = first = e->dest;
430 583445600 : counter = NUM_FIXED_BLOCKS;
431 583445600 : goto_locus = e->goto_locus;
432 :
433 613495692 : while (counter < n_basic_blocks_for_fn (cfun))
434 : {
435 613094235 : basic_block new_target = NULL;
436 613094235 : may_thread |= (target->flags & BB_MODIFIED) != 0;
437 :
438 613094235 : if (FORWARDER_BLOCK_P (target)
439 652712260 : && single_succ (target) != EXIT_BLOCK_PTR_FOR_FN (cfun))
440 : {
441 : /* Bypass trivial infinite loops. */
442 30232800 : new_target = single_succ (target);
443 30232800 : if (target == new_target)
444 : counter = n_basic_blocks_for_fn (cfun);
445 29970749 : else if (!optimize)
446 : {
447 : /* When not optimizing, ensure that edges or forwarder
448 : blocks with different locus are not optimized out. */
449 1189972 : location_t new_locus = single_succ_edge (target)->goto_locus;
450 1189972 : location_t locus = goto_locus;
451 :
452 1189972 : if (LOCATION_LOCUS (new_locus) != UNKNOWN_LOCATION
453 473766 : && LOCATION_LOCUS (locus) != UNKNOWN_LOCATION
454 1500954 : && new_locus != locus)
455 : new_target = NULL;
456 : else
457 : {
458 1009776 : if (LOCATION_LOCUS (new_locus) != UNKNOWN_LOCATION)
459 293570 : locus = new_locus;
460 :
461 1009776 : rtx_insn *last = BB_END (target);
462 1009776 : if (DEBUG_INSN_P (last))
463 6 : last = prev_nondebug_insn (last);
464 1009776 : if (last && INSN_P (last))
465 501609 : new_locus = INSN_LOCATION (last);
466 : else
467 : new_locus = UNKNOWN_LOCATION;
468 :
469 501609 : if (LOCATION_LOCUS (new_locus) != UNKNOWN_LOCATION
470 484456 : && LOCATION_LOCUS (locus) != UNKNOWN_LOCATION
471 1380093 : && new_locus != locus)
472 : new_target = NULL;
473 : else
474 : {
475 1001301 : if (LOCATION_LOCUS (new_locus) != UNKNOWN_LOCATION)
476 475981 : locus = new_locus;
477 :
478 : goto_locus = locus;
479 : }
480 : }
481 : }
482 : }
483 :
484 : /* Allow to thread only over one edge at time to simplify updating
485 : of probabilities. */
486 582861435 : else if ((mode & CLEANUP_THREADING) && may_thread)
487 : {
488 28947251 : edge t = thread_jump (e, target);
489 28947251 : if (t)
490 : {
491 7302 : if (!threaded_edges)
492 7228 : threaded_edges = XNEWVEC (edge,
493 : n_basic_blocks_for_fn (cfun));
494 : else
495 : {
496 : int i;
497 :
498 : /* Detect an infinite loop across blocks not
499 : including the start block. */
500 221 : for (i = 0; i < nthreaded_edges; ++i)
501 166 : if (threaded_edges[i] == t)
502 : break;
503 74 : if (i < nthreaded_edges)
504 : {
505 19 : counter = n_basic_blocks_for_fn (cfun);
506 19 : break;
507 : }
508 : }
509 :
510 : /* Detect an infinite loop across the start block. */
511 7283 : if (t->dest == b)
512 : break;
513 :
514 5963 : gcc_assert (nthreaded_edges
515 : < (n_basic_blocks_for_fn (cfun)
516 : - NUM_FIXED_BLOCKS));
517 5963 : threaded_edges[nthreaded_edges++] = t;
518 :
519 5963 : new_target = t->dest;
520 5963 : new_target_threaded = true;
521 : }
522 : }
523 :
524 30050092 : if (!new_target)
525 : break;
526 :
527 30050092 : counter++;
528 : /* Do not turn non-crossing jump to crossing. Depending on target
529 : it may require different instruction pattern. */
530 30050092 : if ((e->flags & EDGE_CROSSING)
531 30016091 : || BB_PARTITION (first) == BB_PARTITION (new_target))
532 : {
533 15458002 : target = new_target;
534 15458002 : threaded |= new_target_threaded;
535 : }
536 : }
537 :
538 583445600 : if (counter >= n_basic_blocks_for_fn (cfun))
539 : {
540 401476 : if (dump_file)
541 4 : fprintf (dump_file, "Infinite loop in BB %i.\n",
542 : target->index);
543 : }
544 583044124 : else if (target == first)
545 : ; /* We didn't do anything. */
546 : else
547 : {
548 : /* Save the values now, as the edge may get removed. */
549 14322796 : profile_count edge_count = e->count ();
550 14322796 : int n = 0;
551 :
552 14322796 : e->goto_locus = goto_locus;
553 :
554 : /* Don't force if target is exit block. */
555 14322796 : if (threaded && target != EXIT_BLOCK_PTR_FOR_FN (cfun))
556 : {
557 5891 : notice_new_block (redirect_edge_and_branch_force (e, target));
558 5891 : if (dump_file)
559 0 : fprintf (dump_file, "Conditionals threaded.\n");
560 : }
561 14316905 : else if (!redirect_edge_and_branch (e, target))
562 : {
563 7589812 : if (dump_file)
564 183 : fprintf (dump_file,
565 : "Forwarding edge %i->%i to %i failed.\n",
566 183 : b->index, e->dest->index, target->index);
567 7589812 : ei_next (&ei);
568 7589812 : continue;
569 : }
570 :
571 : /* We successfully forwarded the edge. Now update profile
572 : data: for each edge we traversed in the chain, remove
573 : the original edge's execution count. */
574 6860928 : do
575 : {
576 6860928 : edge t;
577 :
578 6860928 : if (!single_succ_p (first))
579 : {
580 5944 : gcc_assert (n < nthreaded_edges);
581 5944 : t = threaded_edges [n++];
582 5944 : gcc_assert (t->src == first);
583 5944 : update_bb_profile_for_threading (first, edge_count, t);
584 5944 : update_br_prob_note (first);
585 : }
586 : else
587 : {
588 6854984 : first->count -= edge_count;
589 : /* It is possible that as the result of
590 : threading we've removed edge as it is
591 : threaded to the fallthru edge. Avoid
592 : getting out of sync. */
593 6854984 : if (n < nthreaded_edges
594 0 : && first == threaded_edges [n]->src)
595 0 : n++;
596 6854984 : t = single_succ_edge (first);
597 : }
598 :
599 6860928 : first = t->dest;
600 : }
601 6860928 : while (first != target);
602 :
603 6732984 : changed = true;
604 6732984 : continue;
605 6732984 : }
606 569122804 : ei_next (&ei);
607 : }
608 :
609 411192794 : free (threaded_edges);
610 411192794 : return changed;
611 : }
612 :
613 :
614 : /* Blocks A and B are to be merged into a single block. A has no incoming
615 : fallthru edge, so it can be moved before B without adding or modifying
616 : any jumps (aside from the jump from A to B). */
617 :
618 : static void
619 21562 : merge_blocks_move_predecessor_nojumps (basic_block a, basic_block b)
620 : {
621 21562 : rtx_insn *barrier;
622 :
623 : /* If we are partitioning hot/cold basic blocks, we don't want to
624 : mess up unconditional or indirect jumps that cross between hot
625 : and cold sections.
626 :
627 : Basic block partitioning may result in some jumps that appear to
628 : be optimizable (or blocks that appear to be mergeable), but which really
629 : must be left untouched (they are required to make it safely across
630 : partition boundaries). See the comments at the top of
631 : bb-reorder.cc:partition_hot_cold_basic_blocks for complete details. */
632 :
633 21562 : if (BB_PARTITION (a) != BB_PARTITION (b))
634 : return;
635 :
636 21562 : barrier = next_nonnote_insn (BB_END (a));
637 21562 : gcc_assert (BARRIER_P (barrier));
638 21562 : delete_insn (barrier);
639 :
640 : /* Scramble the insn chain. */
641 21562 : if (BB_END (a) != PREV_INSN (BB_HEAD (b)))
642 21529 : reorder_insns_nobb (BB_HEAD (a), BB_END (a), PREV_INSN (BB_HEAD (b)));
643 21562 : df_set_bb_dirty (a);
644 :
645 21562 : if (dump_file)
646 0 : fprintf (dump_file, "Moved block %d before %d and merged.\n",
647 : a->index, b->index);
648 :
649 : /* Swap the records for the two blocks around. */
650 :
651 21562 : unlink_block (a);
652 21562 : link_block (a, b->prev_bb);
653 :
654 : /* Now blocks A and B are contiguous. Merge them. */
655 21562 : merge_blocks (a, b);
656 : }
657 :
658 : /* Blocks A and B are to be merged into a single block. B has no outgoing
659 : fallthru edge, so it can be moved after A without adding or modifying
660 : any jumps (aside from the jump from A to B). */
661 :
662 : static void
663 30867 : merge_blocks_move_successor_nojumps (basic_block a, basic_block b)
664 : {
665 30867 : rtx_insn *barrier, *real_b_end;
666 30867 : rtx_insn *label;
667 30867 : rtx_jump_table_data *table;
668 :
669 : /* If we are partitioning hot/cold basic blocks, we don't want to
670 : mess up unconditional or indirect jumps that cross between hot
671 : and cold sections.
672 :
673 : Basic block partitioning may result in some jumps that appear to
674 : be optimizable (or blocks that appear to be mergeable), but which really
675 : must be left untouched (they are required to make it safely across
676 : partition boundaries). See the comments at the top of
677 : bb-reorder.cc:partition_hot_cold_basic_blocks for complete details. */
678 :
679 30867 : if (BB_PARTITION (a) != BB_PARTITION (b))
680 0 : return;
681 :
682 30867 : real_b_end = BB_END (b);
683 :
684 : /* If there is a jump table following block B temporarily add the jump table
685 : to block B so that it will also be moved to the correct location. */
686 30867 : if (tablejump_p (BB_END (b), &label, &table)
687 30867 : && prev_active_insn (label) == BB_END (b))
688 : {
689 0 : BB_END (b) = table;
690 : }
691 :
692 : /* There had better have been a barrier there. Delete it. */
693 30867 : barrier = NEXT_INSN (BB_END (b));
694 30867 : if (barrier && BARRIER_P (barrier))
695 30867 : delete_insn (barrier);
696 :
697 :
698 : /* Scramble the insn chain. */
699 30867 : reorder_insns_nobb (BB_HEAD (b), BB_END (b), BB_END (a));
700 :
701 : /* Restore the real end of b. */
702 30867 : BB_END (b) = real_b_end;
703 :
704 30867 : if (dump_file)
705 22 : fprintf (dump_file, "Moved block %d after %d and merged.\n",
706 : b->index, a->index);
707 :
708 : /* Now blocks A and B are contiguous. Merge them. */
709 30867 : merge_blocks (a, b);
710 : }
711 :
712 : /* Attempt to merge basic blocks that are potentially non-adjacent.
713 : Return NULL iff the attempt failed, otherwise return basic block
714 : where cleanup_cfg should continue. Because the merging commonly
715 : moves basic block away or introduces another optimization
716 : possibility, return basic block just before B so cleanup_cfg don't
717 : need to iterate.
718 :
719 : It may be good idea to return basic block before C in the case
720 : C has been moved after B and originally appeared earlier in the
721 : insn sequence, but we have no information available about the
722 : relative ordering of these two. Hopefully it is not too common. */
723 :
724 : static basic_block
725 6986668 : merge_blocks_move (edge e, basic_block b, basic_block c, int mode)
726 : {
727 6986668 : basic_block next;
728 :
729 : /* If we are partitioning hot/cold basic blocks, we don't want to
730 : mess up unconditional or indirect jumps that cross between hot
731 : and cold sections.
732 :
733 : Basic block partitioning may result in some jumps that appear to
734 : be optimizable (or blocks that appear to be mergeable), but which really
735 : must be left untouched (they are required to make it safely across
736 : partition boundaries). See the comments at the top of
737 : bb-reorder.cc:partition_hot_cold_basic_blocks for complete details. */
738 :
739 6986668 : if (BB_PARTITION (b) != BB_PARTITION (c))
740 : return NULL;
741 :
742 : /* If B has a fallthru edge to C, no need to move anything. */
743 6593037 : if (e->flags & EDGE_FALLTHRU)
744 : {
745 3846862 : int b_index = b->index, c_index = c->index;
746 :
747 : /* Protect the loop latches. */
748 3846862 : if (current_loops && c->loop_father->latch == c)
749 : return NULL;
750 :
751 3802096 : merge_blocks (b, c);
752 3802096 : update_forwarder_flag (b);
753 :
754 3802096 : if (dump_file)
755 910 : fprintf (dump_file, "Merged %d and %d without moving.\n",
756 : b_index, c_index);
757 :
758 3802096 : return b->prev_bb == ENTRY_BLOCK_PTR_FOR_FN (cfun) ? b : b->prev_bb;
759 : }
760 :
761 : /* Otherwise we will need to move code around. Do that only if expensive
762 : transformations are allowed. */
763 2746175 : else if (mode & CLEANUP_EXPENSIVE)
764 : {
765 918587 : edge tmp_edge, b_fallthru_edge;
766 918587 : bool c_has_outgoing_fallthru;
767 918587 : bool b_has_incoming_fallthru;
768 :
769 : /* Avoid overactive code motion, as the forwarder blocks should be
770 : eliminated by edge redirection instead. One exception might have
771 : been if B is a forwarder block and C has no fallthru edge, but
772 : that should be cleaned up by bb-reorder instead. */
773 918587 : if (FORWARDER_BLOCK_P (b) || FORWARDER_BLOCK_P (c))
774 : return NULL;
775 :
776 : /* We must make sure to not munge nesting of lexical blocks,
777 : and loop notes. This is done by squeezing out all the notes
778 : and leaving them there to lie. Not ideal, but functional. */
779 :
780 54575 : tmp_edge = find_fallthru_edge (c->succs);
781 54575 : c_has_outgoing_fallthru = (tmp_edge != NULL);
782 :
783 54575 : tmp_edge = find_fallthru_edge (b->preds);
784 54575 : b_has_incoming_fallthru = (tmp_edge != NULL);
785 54575 : b_fallthru_edge = tmp_edge;
786 54575 : next = b->prev_bb;
787 54575 : if (next == c)
788 63 : next = next->prev_bb;
789 :
790 : /* Otherwise, we're going to try to move C after B. If C does
791 : not have an outgoing fallthru, then it can be moved
792 : immediately after B without introducing or modifying jumps. */
793 54575 : if (! c_has_outgoing_fallthru)
794 : {
795 30867 : merge_blocks_move_successor_nojumps (b, c);
796 30867 : return next == ENTRY_BLOCK_PTR_FOR_FN (cfun) ? next->next_bb : next;
797 : }
798 :
799 : /* If B does not have an incoming fallthru, then it can be moved
800 : immediately before C without introducing or modifying jumps.
801 : C cannot be the first block, so we do not have to worry about
802 : accessing a non-existent block. */
803 :
804 23708 : if (b_has_incoming_fallthru)
805 : {
806 21621 : basic_block bb;
807 :
808 21621 : if (b_fallthru_edge->src == ENTRY_BLOCK_PTR_FOR_FN (cfun))
809 : return NULL;
810 19475 : bb = force_nonfallthru (b_fallthru_edge);
811 19475 : if (bb)
812 14986 : notice_new_block (bb);
813 : }
814 :
815 21562 : merge_blocks_move_predecessor_nojumps (b, c);
816 21562 : return next == ENTRY_BLOCK_PTR_FOR_FN (cfun) ? next->next_bb : next;
817 : }
818 :
819 : return NULL;
820 : }
821 :
822 :
823 : /* Removes the memory attributes of MEM expression
824 : if they are not equal. */
825 :
826 : static void
827 149552785 : merge_memattrs (rtx x, rtx y)
828 : {
829 149552785 : int i;
830 149552785 : int j;
831 149552785 : enum rtx_code code;
832 149552785 : const char *fmt;
833 :
834 149552785 : if (x == y)
835 : return;
836 111571284 : if (x == 0 || y == 0)
837 : return;
838 :
839 111532817 : code = GET_CODE (x);
840 :
841 111532817 : if (code != GET_CODE (y))
842 : return;
843 :
844 111532805 : if (GET_MODE (x) != GET_MODE (y))
845 : return;
846 :
847 111529613 : if (code == MEM && !mem_attrs_eq_p (MEM_ATTRS (x), MEM_ATTRS (y)))
848 : {
849 47944 : if (! MEM_ATTRS (x))
850 803 : MEM_ATTRS (y) = 0;
851 47141 : else if (! MEM_ATTRS (y))
852 1358 : MEM_ATTRS (x) = 0;
853 : else
854 : {
855 45783 : if (MEM_ALIAS_SET (x) != MEM_ALIAS_SET (y))
856 : {
857 2679 : set_mem_alias_set (x, 0);
858 2679 : set_mem_alias_set (y, 0);
859 : }
860 :
861 45783 : if (! mem_expr_equal_p (MEM_EXPR (x), MEM_EXPR (y)))
862 : {
863 45178 : set_mem_expr (x, 0);
864 45178 : set_mem_expr (y, 0);
865 45178 : clear_mem_offset (x);
866 45178 : clear_mem_offset (y);
867 : }
868 605 : else if (MEM_OFFSET_KNOWN_P (x) != MEM_OFFSET_KNOWN_P (y)
869 605 : || (MEM_OFFSET_KNOWN_P (x)
870 0 : && maybe_ne (MEM_OFFSET (x), MEM_OFFSET (y))))
871 : {
872 0 : clear_mem_offset (x);
873 0 : clear_mem_offset (y);
874 : }
875 :
876 52672 : if (!MEM_SIZE_KNOWN_P (x))
877 221 : clear_mem_size (y);
878 52269 : else if (!MEM_SIZE_KNOWN_P (y))
879 8 : clear_mem_size (x);
880 45554 : else if (known_le (MEM_SIZE (x), MEM_SIZE (y)))
881 45510 : set_mem_size (x, MEM_SIZE (y));
882 44 : else if (known_le (MEM_SIZE (y), MEM_SIZE (x)))
883 44 : set_mem_size (y, MEM_SIZE (x));
884 : else
885 : {
886 : /* The sizes aren't ordered, so we can't merge them. */
887 : clear_mem_size (x);
888 : clear_mem_size (y);
889 : }
890 :
891 59602 : set_mem_align (x, MIN (MEM_ALIGN (x), MEM_ALIGN (y)));
892 52713 : set_mem_align (y, MEM_ALIGN (x));
893 : }
894 : }
895 111529613 : if (code == MEM)
896 : {
897 6123685 : if (MEM_READONLY_P (x) != MEM_READONLY_P (y))
898 : {
899 0 : MEM_READONLY_P (x) = 0;
900 0 : MEM_READONLY_P (y) = 0;
901 : }
902 6123685 : if (MEM_NOTRAP_P (x) != MEM_NOTRAP_P (y))
903 : {
904 214 : MEM_NOTRAP_P (x) = 0;
905 214 : MEM_NOTRAP_P (y) = 0;
906 : }
907 6123685 : if (MEM_VOLATILE_P (x) != MEM_VOLATILE_P (y))
908 : {
909 150 : MEM_VOLATILE_P (x) = 1;
910 150 : MEM_VOLATILE_P (y) = 1;
911 : }
912 : }
913 :
914 111529613 : fmt = GET_RTX_FORMAT (code);
915 338872107 : for (i = GET_RTX_LENGTH (code) - 1; i >= 0; i--)
916 : {
917 227342494 : switch (fmt[i])
918 : {
919 396008 : case 'E':
920 : /* Two vectors must have the same length. */
921 396008 : if (XVECLEN (x, i) != XVECLEN (y, i))
922 : return;
923 :
924 1170982 : for (j = 0; j < XVECLEN (x, i); j++)
925 774974 : merge_memattrs (XVECEXP (x, i, j), XVECEXP (y, i, j));
926 :
927 : break;
928 :
929 139979602 : case 'e':
930 139979602 : merge_memattrs (XEXP (x, i), XEXP (y, i));
931 : }
932 : }
933 : return;
934 : }
935 :
936 :
937 : /* Checks if patterns P1 and P2 are equivalent, apart from the possibly
938 : different single sets S1 and S2. */
939 :
940 : static bool
941 110 : equal_different_set_p (rtx p1, rtx s1, rtx p2, rtx s2)
942 : {
943 110 : int i;
944 110 : rtx e1, e2;
945 :
946 110 : if (p1 == s1 && p2 == s2)
947 : return true;
948 :
949 0 : if (GET_CODE (p1) != PARALLEL || GET_CODE (p2) != PARALLEL)
950 : return false;
951 :
952 0 : if (XVECLEN (p1, 0) != XVECLEN (p2, 0))
953 : return false;
954 :
955 0 : for (i = 0; i < XVECLEN (p1, 0); i++)
956 : {
957 0 : e1 = XVECEXP (p1, 0, i);
958 0 : e2 = XVECEXP (p2, 0, i);
959 0 : if (e1 == s1 && e2 == s2)
960 0 : continue;
961 0 : if (reload_completed
962 0 : ? rtx_renumbered_equal_p (e1, e2) : rtx_equal_p (e1, e2))
963 0 : continue;
964 :
965 : return false;
966 : }
967 :
968 : return true;
969 : }
970 :
971 :
972 : /* NOTE1 is the REG_EQUAL note, if any, attached to an insn
973 : that is a single_set with a SET_SRC of SRC1. Similarly
974 : for NOTE2/SRC2.
975 :
976 : So effectively NOTE1/NOTE2 are an alternate form of
977 : SRC1/SRC2 respectively.
978 :
979 : Return nonzero if SRC1 or NOTE1 has the same constant
980 : integer value as SRC2 or NOTE2. Else return zero. */
981 : static int
982 7958658 : values_equal_p (rtx note1, rtx note2, rtx src1, rtx src2)
983 : {
984 7958658 : if (note1
985 7958658 : && note2
986 1285770 : && CONST_INT_P (XEXP (note1, 0))
987 8016989 : && rtx_equal_p (XEXP (note1, 0), XEXP (note2, 0)))
988 : return 1;
989 :
990 7958550 : if (!note1
991 7958550 : && !note2
992 6199589 : && CONST_INT_P (src1)
993 2665480 : && CONST_INT_P (src2)
994 10521438 : && rtx_equal_p (src1, src2))
995 : return 1;
996 :
997 7958550 : if (note1
998 1562556 : && CONST_INT_P (src2)
999 8107159 : && rtx_equal_p (XEXP (note1, 0), src2))
1000 : return 1;
1001 :
1002 7958550 : if (note2
1003 1482067 : && CONST_INT_P (src1)
1004 8029661 : && rtx_equal_p (XEXP (note2, 0), src1))
1005 2 : return 1;
1006 :
1007 : return 0;
1008 : }
1009 :
1010 : /* Examine register notes on I1 and I2 and return:
1011 : - dir_forward if I1 can be replaced by I2, or
1012 : - dir_backward if I2 can be replaced by I1, or
1013 : - dir_both if both are the case. */
1014 :
1015 : static enum replace_direction
1016 14165656 : can_replace_by (rtx_insn *i1, rtx_insn *i2)
1017 : {
1018 14165656 : rtx s1, s2, d1, d2, src1, src2, note1, note2;
1019 14165656 : bool c1, c2;
1020 :
1021 : /* Check for 2 sets. */
1022 14165656 : s1 = single_set (i1);
1023 14165656 : s2 = single_set (i2);
1024 14165656 : if (s1 == NULL_RTX || s2 == NULL_RTX)
1025 : return dir_none;
1026 :
1027 : /* Check that the 2 sets set the same dest. */
1028 13391427 : d1 = SET_DEST (s1);
1029 13391427 : d2 = SET_DEST (s2);
1030 26782854 : if (!(reload_completed
1031 13391427 : ? rtx_renumbered_equal_p (d1, d2) : rtx_equal_p (d1, d2)))
1032 : return dir_none;
1033 :
1034 : /* Find identical req_equiv or reg_equal note, which implies that the 2 sets
1035 : set dest to the same value. */
1036 7958658 : note1 = find_reg_equal_equiv_note (i1);
1037 7958658 : note2 = find_reg_equal_equiv_note (i2);
1038 :
1039 7958658 : src1 = SET_SRC (s1);
1040 7958658 : src2 = SET_SRC (s2);
1041 :
1042 7958658 : if (!values_equal_p (note1, note2, src1, src2))
1043 : return dir_none;
1044 :
1045 110 : if (!equal_different_set_p (PATTERN (i1), s1, PATTERN (i2), s2))
1046 : return dir_none;
1047 :
1048 : /* Although the 2 sets set dest to the same value, we cannot replace
1049 : (set (dest) (const_int))
1050 : by
1051 : (set (dest) (reg))
1052 : because we don't know if the reg is live and has the same value at the
1053 : location of replacement. */
1054 110 : c1 = CONST_INT_P (src1);
1055 110 : c2 = CONST_INT_P (src2);
1056 110 : if (c1 && c2)
1057 : return dir_both;
1058 110 : else if (c2)
1059 : return dir_forward;
1060 110 : else if (c1)
1061 2 : return dir_backward;
1062 :
1063 : return dir_none;
1064 : }
1065 :
1066 : /* Merges directions A and B. */
1067 :
1068 : static enum replace_direction
1069 23798082 : merge_dir (enum replace_direction a, enum replace_direction b)
1070 : {
1071 : /* Implements the following table:
1072 : |bo fw bw no
1073 : ---+-----------
1074 : bo |bo fw bw no
1075 : fw |-- fw no no
1076 : bw |-- -- bw no
1077 : no |-- -- -- no. */
1078 :
1079 0 : if (a == b)
1080 : return a;
1081 :
1082 16347498 : if (a == dir_both)
1083 : return b;
1084 3912667 : if (b == dir_both)
1085 0 : return a;
1086 :
1087 : return dir_none;
1088 : }
1089 :
1090 : /* Array of flags indexed by reg note kind, true if the given
1091 : reg note is CFA related. */
1092 : static const bool reg_note_cfa_p[] = {
1093 : #undef REG_CFA_NOTE
1094 : #define DEF_REG_NOTE(NAME) false,
1095 : #define REG_CFA_NOTE(NAME) true,
1096 : #include "reg-notes.def"
1097 : #undef REG_CFA_NOTE
1098 : #undef DEF_REG_NOTE
1099 : false
1100 : };
1101 :
1102 : /* Return true if I1 and I2 have identical CFA notes (the same order
1103 : and equivalent content). */
1104 :
1105 : static bool
1106 617547 : insns_have_identical_cfa_notes (rtx_insn *i1, rtx_insn *i2)
1107 : {
1108 617547 : rtx n1, n2;
1109 617547 : for (n1 = REG_NOTES (i1), n2 = REG_NOTES (i2); ;
1110 490090 : n1 = XEXP (n1, 1), n2 = XEXP (n2, 1))
1111 : {
1112 : /* Skip over reg notes not related to CFI information. */
1113 1622898 : while (n1 && !reg_note_cfa_p[REG_NOTE_KIND (n1)])
1114 25171 : n1 = XEXP (n1, 1);
1115 1131294 : while (n2 && !reg_note_cfa_p[REG_NOTE_KIND (n2)])
1116 23657 : n2 = XEXP (n2, 1);
1117 1107637 : if (n1 == NULL_RTX && n2 == NULL_RTX)
1118 : return true;
1119 601300 : if (n1 == NULL_RTX || n2 == NULL_RTX)
1120 : return false;
1121 596606 : if (XEXP (n1, 0) == XEXP (n2, 0))
1122 : ;
1123 596600 : else if (XEXP (n1, 0) == NULL_RTX || XEXP (n2, 0) == NULL_RTX)
1124 : return false;
1125 1193200 : else if (!(reload_completed
1126 596600 : ? rtx_renumbered_equal_p (XEXP (n1, 0), XEXP (n2, 0))
1127 0 : : rtx_equal_p (XEXP (n1, 0), XEXP (n2, 0))))
1128 : return false;
1129 : }
1130 : }
1131 :
1132 : /* Examine I1 and I2 and return:
1133 : - dir_forward if I1 can be replaced by I2, or
1134 : - dir_backward if I2 can be replaced by I1, or
1135 : - dir_both if both are the case. */
1136 :
1137 : static enum replace_direction
1138 41217958 : old_insns_match_p (int mode ATTRIBUTE_UNUSED, rtx_insn *i1, rtx_insn *i2)
1139 : {
1140 41217958 : rtx p1, p2;
1141 :
1142 : /* Verify that I1 and I2 are equivalent. */
1143 41217958 : if (GET_CODE (i1) != GET_CODE (i2))
1144 : return dir_none;
1145 :
1146 : /* __builtin_unreachable() may lead to empty blocks (ending with
1147 : NOTE_INSN_BASIC_BLOCK). They may be crossjumped. */
1148 35612978 : if (NOTE_INSN_BASIC_BLOCK_P (i1) && NOTE_INSN_BASIC_BLOCK_P (i2))
1149 : return dir_both;
1150 :
1151 : /* ??? Do not allow cross-jumping between different stack levels. */
1152 35525072 : p1 = find_reg_note (i1, REG_ARGS_SIZE, NULL);
1153 35525072 : p2 = find_reg_note (i2, REG_ARGS_SIZE, NULL);
1154 35525072 : if (p1 && p2)
1155 : {
1156 3994757 : p1 = XEXP (p1, 0);
1157 3994757 : p2 = XEXP (p2, 0);
1158 3994757 : if (!rtx_equal_p (p1, p2))
1159 : return dir_none;
1160 :
1161 : /* ??? Worse, this adjustment had better be constant lest we
1162 : have differing incoming stack levels. */
1163 3893702 : if (!frame_pointer_needed
1164 3893702 : && known_eq (find_args_size_adjust (i1), HOST_WIDE_INT_MIN))
1165 27 : return dir_none;
1166 : }
1167 31530315 : else if (p1 || p2)
1168 : return dir_none;
1169 :
1170 : /* Do not allow cross-jumping between frame related insns and other
1171 : insns. */
1172 33818670 : if (RTX_FRAME_RELATED_P (i1) != RTX_FRAME_RELATED_P (i2))
1173 : return dir_none;
1174 :
1175 33377275 : p1 = PATTERN (i1);
1176 33377275 : p2 = PATTERN (i2);
1177 :
1178 33377275 : if (GET_CODE (p1) != GET_CODE (p2))
1179 : return dir_none;
1180 :
1181 : /* If this is a CALL_INSN, compare register usage information.
1182 : If we don't check this on stack register machines, the two
1183 : CALL_INSNs might be merged leaving reg-stack.cc with mismatching
1184 : numbers of stack registers in the same basic block.
1185 : If we don't check this on machines with delay slots, a delay slot may
1186 : be filled that clobbers a parameter expected by the subroutine.
1187 :
1188 : ??? We take the simple route for now and assume that if they're
1189 : equal, they were constructed identically.
1190 :
1191 : Also check for identical exception regions. */
1192 :
1193 30786541 : if (CALL_P (i1))
1194 : {
1195 : /* Ensure the same EH region. */
1196 12391994 : rtx n1 = find_reg_note (i1, REG_EH_REGION, 0);
1197 12391994 : rtx n2 = find_reg_note (i2, REG_EH_REGION, 0);
1198 :
1199 12391994 : if (!n1 && n2)
1200 : return dir_none;
1201 :
1202 12300139 : if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0)))
1203 : return dir_none;
1204 :
1205 11405911 : if (!rtx_equal_p (CALL_INSN_FUNCTION_USAGE (i1),
1206 11405911 : CALL_INSN_FUNCTION_USAGE (i2))
1207 7535125 : || CALL_INSN_ABI_ID (i1) != CALL_INSN_ABI_ID (i2)
1208 18941036 : || SIBLING_CALL_P (i1) != SIBLING_CALL_P (i2))
1209 : return dir_none;
1210 :
1211 : /* For address sanitizer, never crossjump __asan_report_* builtins,
1212 : otherwise errors might be reported on incorrect lines. */
1213 7535125 : if (flag_sanitize & SANITIZE_ADDRESS)
1214 : {
1215 368884 : rtx call = get_call_rtx_from (i1);
1216 368884 : if (call && GET_CODE (XEXP (XEXP (call, 0), 0)) == SYMBOL_REF)
1217 : {
1218 368884 : rtx symbol = XEXP (XEXP (call, 0), 0);
1219 368884 : if (SYMBOL_REF_DECL (symbol)
1220 368884 : && TREE_CODE (SYMBOL_REF_DECL (symbol)) == FUNCTION_DECL)
1221 : {
1222 368884 : if ((DECL_BUILT_IN_CLASS (SYMBOL_REF_DECL (symbol))
1223 368884 : == BUILT_IN_NORMAL)
1224 366879 : && DECL_FUNCTION_CODE (SYMBOL_REF_DECL (symbol))
1225 : >= BUILT_IN_ASAN_REPORT_LOAD1
1226 734255 : && DECL_FUNCTION_CODE (SYMBOL_REF_DECL (symbol))
1227 : <= BUILT_IN_ASAN_STOREN)
1228 : return dir_none;
1229 : }
1230 : }
1231 : }
1232 :
1233 7170704 : if (insn_callee_abi (i1) != insn_callee_abi (i2))
1234 : return dir_none;
1235 : }
1236 :
1237 : /* If both i1 and i2 are frame related, verify all the CFA notes
1238 : in the same order and with the same content. */
1239 25504701 : if (RTX_FRAME_RELATED_P (i1) && !insns_have_identical_cfa_notes (i1, i2))
1240 : return dir_none;
1241 :
1242 : #ifdef STACK_REGS
1243 : /* If cross_jump_death_matters is not 0, the insn's mode
1244 : indicates whether or not the insn contains any stack-like
1245 : regs. */
1246 :
1247 25393491 : if ((mode & CLEANUP_POST_REGSTACK) && stack_regs_mentioned (i1))
1248 : {
1249 : /* If register stack conversion has already been done, then
1250 : death notes must also be compared before it is certain that
1251 : the two instruction streams match. */
1252 :
1253 : rtx note;
1254 : HARD_REG_SET i1_regset, i2_regset;
1255 :
1256 0 : CLEAR_HARD_REG_SET (i1_regset);
1257 0 : CLEAR_HARD_REG_SET (i2_regset);
1258 :
1259 0 : for (note = REG_NOTES (i1); note; note = XEXP (note, 1))
1260 0 : if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
1261 0 : SET_HARD_REG_BIT (i1_regset, REGNO (XEXP (note, 0)));
1262 :
1263 0 : for (note = REG_NOTES (i2); note; note = XEXP (note, 1))
1264 0 : if (REG_NOTE_KIND (note) == REG_DEAD && STACK_REG_P (XEXP (note, 0)))
1265 0 : SET_HARD_REG_BIT (i2_regset, REGNO (XEXP (note, 0)));
1266 :
1267 0 : if (i1_regset != i2_regset)
1268 0 : return dir_none;
1269 : }
1270 : #endif
1271 :
1272 25393491 : if (reload_completed
1273 25393491 : ? rtx_renumbered_equal_p (p1, p2) : rtx_equal_p (p1, p2))
1274 : return dir_both;
1275 :
1276 14165656 : return can_replace_by (i1, i2);
1277 : }
1278 :
1279 : /* When comparing insns I1 and I2 in flow_find_cross_jump or
1280 : flow_find_head_matching_sequence, ensure the notes match. */
1281 :
1282 : static void
1283 8798209 : merge_notes (rtx_insn *i1, rtx_insn *i2)
1284 : {
1285 : /* If the merged insns have different REG_EQUAL notes, then
1286 : remove them. */
1287 8798209 : rtx equiv1 = find_reg_equal_equiv_note (i1);
1288 8798209 : rtx equiv2 = find_reg_equal_equiv_note (i2);
1289 :
1290 8798209 : if (equiv1 && !equiv2)
1291 24185 : remove_note (i1, equiv1);
1292 8774024 : else if (!equiv1 && equiv2)
1293 5786 : remove_note (i2, equiv2);
1294 8768238 : else if (equiv1 && equiv2
1295 8768238 : && !rtx_equal_p (XEXP (equiv1, 0), XEXP (equiv2, 0)))
1296 : {
1297 348 : remove_note (i1, equiv1);
1298 348 : remove_note (i2, equiv2);
1299 : }
1300 8798209 : }
1301 :
1302 : /* Walks from I1 in BB1 backward till the next non-debug insn, and returns the
1303 : resulting insn in I1, and the corresponding bb in BB1. At the head of a
1304 : bb, if there is a predecessor bb that reaches this bb via fallthru, and
1305 : FOLLOW_FALLTHRU, walks further in the predecessor bb and registers this in
1306 : DID_FALLTHRU. Otherwise, stops at the head of the bb. */
1307 :
1308 : static void
1309 49256844 : walk_to_nondebug_insn (rtx_insn **i1, basic_block *bb1, bool follow_fallthru,
1310 : bool *did_fallthru)
1311 : {
1312 49256844 : edge fallthru;
1313 :
1314 49256844 : *did_fallthru = false;
1315 :
1316 : /* Ignore notes. */
1317 68669420 : while (!NONDEBUG_INSN_P (*i1))
1318 : {
1319 20646576 : if (*i1 != BB_HEAD (*bb1))
1320 : {
1321 19226495 : *i1 = PREV_INSN (*i1);
1322 19226495 : continue;
1323 : }
1324 :
1325 1420081 : if (!follow_fallthru)
1326 : return;
1327 :
1328 1234919 : fallthru = find_fallthru_edge ((*bb1)->preds);
1329 477718 : if (!fallthru || fallthru->src == ENTRY_BLOCK_PTR_FOR_FN (cfun)
1330 1712561 : || !single_succ_p (fallthru->src))
1331 : return;
1332 :
1333 186081 : *bb1 = fallthru->src;
1334 186081 : *i1 = BB_END (*bb1);
1335 186081 : *did_fallthru = true;
1336 : }
1337 : }
1338 :
1339 : /* Look through the insns at the end of BB1 and BB2 and find the longest
1340 : sequence that are either equivalent, or allow forward or backward
1341 : replacement. Store the first insns for that sequence in *F1 and *F2 and
1342 : return the sequence length.
1343 :
1344 : DIR_P indicates the allowed replacement direction on function entry, and
1345 : the actual replacement direction on function exit. If NULL, only equivalent
1346 : sequences are allowed.
1347 :
1348 : To simplify callers of this function, if the blocks match exactly,
1349 : store the head of the blocks in *F1 and *F2. */
1350 :
1351 : int
1352 16049269 : flow_find_cross_jump (basic_block bb1, basic_block bb2, rtx_insn **f1,
1353 : rtx_insn **f2, enum replace_direction *dir_p)
1354 : {
1355 16049269 : rtx_insn *i1, *i2, *last1, *last2, *afterlast1, *afterlast2;
1356 16049269 : int ninsns = 0;
1357 16049269 : enum replace_direction dir, last_dir, afterlast_dir;
1358 16049269 : bool follow_fallthru, did_fallthru;
1359 :
1360 16049269 : if (dir_p)
1361 16049269 : dir = *dir_p;
1362 : else
1363 : dir = dir_both;
1364 16049269 : afterlast_dir = dir;
1365 16049269 : last_dir = afterlast_dir;
1366 :
1367 : /* Skip simple jumps at the end of the blocks. Complex jumps still
1368 : need to be compared for equivalence, which we'll do below. */
1369 :
1370 16049269 : i1 = BB_END (bb1);
1371 16049269 : last1 = afterlast1 = last2 = afterlast2 = NULL;
1372 16049269 : if (onlyjump_p (i1)
1373 16049269 : || (returnjump_p (i1) && !side_effects_p (PATTERN (i1))))
1374 : {
1375 14381895 : last1 = i1;
1376 14381895 : i1 = PREV_INSN (i1);
1377 : }
1378 :
1379 16049269 : i2 = BB_END (bb2);
1380 16049269 : if (onlyjump_p (i2)
1381 16049269 : || (returnjump_p (i2) && !side_effects_p (PATTERN (i2))))
1382 : {
1383 11647576 : last2 = i2;
1384 : /* Count everything except for unconditional jump as insn.
1385 : Don't count any jumps if dir_p is NULL. */
1386 11647576 : if (!simplejump_p (i2) && !returnjump_p (i2) && last1 && dir_p)
1387 : ninsns++;
1388 11647576 : i2 = PREV_INSN (i2);
1389 : }
1390 :
1391 33207575 : while (true)
1392 : {
1393 : /* In the following example, we can replace all jumps to C by jumps to A.
1394 :
1395 : This removes 4 duplicate insns.
1396 : [bb A] insn1 [bb C] insn1
1397 : insn2 insn2
1398 : [bb B] insn3 insn3
1399 : insn4 insn4
1400 : jump_insn jump_insn
1401 :
1402 : We could also replace all jumps to A by jumps to C, but that leaves B
1403 : alive, and removes only 2 duplicate insns. In a subsequent crossjump
1404 : step, all jumps to B would be replaced with jumps to the middle of C,
1405 : achieving the same result with more effort.
1406 : So we allow only the first possibility, which means that we don't allow
1407 : fallthru in the block that's being replaced. */
1408 :
1409 24628422 : follow_fallthru = dir_p && dir != dir_forward;
1410 24628422 : walk_to_nondebug_insn (&i1, &bb1, follow_fallthru, &did_fallthru);
1411 24628422 : if (did_fallthru)
1412 28481 : dir = dir_backward;
1413 :
1414 24628422 : follow_fallthru = dir_p && dir != dir_backward;
1415 24628422 : walk_to_nondebug_insn (&i2, &bb2, follow_fallthru, &did_fallthru);
1416 24628422 : if (did_fallthru)
1417 157590 : dir = dir_forward;
1418 :
1419 24628422 : if (i1 == BB_HEAD (bb1) || i2 == BB_HEAD (bb2))
1420 : break;
1421 :
1422 : /* Do not turn corssing edge to non-crossing or vice versa after
1423 : reload. */
1424 23798082 : if (BB_PARTITION (BLOCK_FOR_INSN (i1))
1425 23798082 : != BB_PARTITION (BLOCK_FOR_INSN (i2))
1426 23798082 : && reload_completed)
1427 : break;
1428 :
1429 23798082 : dir = merge_dir (dir, old_insns_match_p (0, i1, i2));
1430 21013984 : if (dir == dir_none || (!dir_p && dir != dir_both))
1431 : break;
1432 :
1433 8579153 : merge_memattrs (i1, i2);
1434 :
1435 : /* Don't begin a cross-jump with a NOTE insn. */
1436 8579153 : if (INSN_P (i1))
1437 : {
1438 8579153 : merge_notes (i1, i2);
1439 :
1440 8579153 : afterlast1 = last1, afterlast2 = last2;
1441 8579153 : last1 = i1, last2 = i2;
1442 8579153 : afterlast_dir = last_dir;
1443 8579153 : last_dir = dir;
1444 8579153 : if (active_insn_p (i1))
1445 8475560 : ninsns++;
1446 : }
1447 :
1448 8579153 : i1 = PREV_INSN (i1);
1449 8579153 : i2 = PREV_INSN (i2);
1450 : }
1451 :
1452 : /* Include preceding notes and labels in the cross-jump. One,
1453 : this may bring us to the head of the blocks as requested above.
1454 : Two, it keeps line number notes as matched as may be. */
1455 16049269 : if (ninsns)
1456 : {
1457 5177897 : bb1 = BLOCK_FOR_INSN (last1);
1458 13392817 : while (last1 != BB_HEAD (bb1) && !NONDEBUG_INSN_P (PREV_INSN (last1)))
1459 : last1 = PREV_INSN (last1);
1460 :
1461 10060417 : if (last1 != BB_HEAD (bb1) && LABEL_P (PREV_INSN (last1)))
1462 : last1 = PREV_INSN (last1);
1463 :
1464 5177897 : bb2 = BLOCK_FOR_INSN (last2);
1465 13442112 : while (last2 != BB_HEAD (bb2) && !NONDEBUG_INSN_P (PREV_INSN (last2)))
1466 : last2 = PREV_INSN (last2);
1467 :
1468 9769262 : if (last2 != BB_HEAD (bb2) && LABEL_P (PREV_INSN (last2)))
1469 : last2 = PREV_INSN (last2);
1470 :
1471 5177897 : *f1 = last1;
1472 5177897 : *f2 = last2;
1473 : }
1474 :
1475 16049269 : if (dir_p)
1476 16049269 : *dir_p = last_dir;
1477 16049269 : return ninsns;
1478 : }
1479 :
1480 : /* Like flow_find_cross_jump, except start looking for a matching sequence from
1481 : the head of the two blocks. Do not include jumps at the end.
1482 : If STOP_AFTER is nonzero, stop after finding that many matching
1483 : instructions. If STOP_AFTER is zero, count all INSN_P insns, if it is
1484 : non-zero, only count active insns. */
1485 :
1486 : int
1487 4696632 : flow_find_head_matching_sequence (basic_block bb1, basic_block bb2, rtx_insn **f1,
1488 : rtx_insn **f2, int stop_after)
1489 : {
1490 4696632 : rtx_insn *i1, *i2, *last1, *last2, *beforelast1, *beforelast2;
1491 4696632 : int ninsns = 0;
1492 4696632 : edge e;
1493 4696632 : edge_iterator ei;
1494 4696632 : int nehedges1 = 0, nehedges2 = 0;
1495 :
1496 11265394 : FOR_EACH_EDGE (e, ei, bb1->succs)
1497 6568762 : if (e->flags & EDGE_EH)
1498 353432 : nehedges1++;
1499 11624080 : FOR_EACH_EDGE (e, ei, bb2->succs)
1500 6927448 : if (e->flags & EDGE_EH)
1501 384916 : nehedges2++;
1502 :
1503 4696632 : i1 = BB_HEAD (bb1);
1504 4696632 : i2 = BB_HEAD (bb2);
1505 4696632 : last1 = beforelast1 = last2 = beforelast2 = NULL;
1506 :
1507 205311 : while (true)
1508 : {
1509 : /* Ignore notes, except NOTE_INSN_EPILOGUE_BEG. */
1510 23212613 : while (!NONDEBUG_INSN_P (i1) && i1 != BB_END (bb1))
1511 : {
1512 18129409 : if (NOTE_P (i1) && NOTE_KIND (i1) == NOTE_INSN_EPILOGUE_BEG)
1513 : break;
1514 18105359 : i1 = NEXT_INSN (i1);
1515 : }
1516 :
1517 22909316 : while (!NONDEBUG_INSN_P (i2) && i2 != BB_END (bb2))
1518 : {
1519 18054629 : if (NOTE_P (i2) && NOTE_KIND (i2) == NOTE_INSN_EPILOGUE_BEG)
1520 : break;
1521 18007373 : i2 = NEXT_INSN (i2);
1522 : }
1523 :
1524 4901943 : if ((i1 == BB_END (bb1) && !NONDEBUG_INSN_P (i1))
1525 4889706 : || (i2 == BB_END (bb2) && !NONDEBUG_INSN_P (i2)))
1526 : break;
1527 :
1528 4884802 : if (NOTE_P (i1) || NOTE_P (i2)
1529 4819388 : || JUMP_P (i1) || JUMP_P (i2))
1530 : break;
1531 :
1532 : /* A sanity check to make sure we're not merging insns with different
1533 : effects on EH. If only one of them ends a basic block, it shouldn't
1534 : have an EH edge; if both end a basic block, there should be the same
1535 : number of EH edges. */
1536 3952855 : if ((i1 == BB_END (bb1) && i2 != BB_END (bb2)
1537 217592 : && nehedges1 > 0)
1538 3915341 : || (i2 == BB_END (bb2) && i1 != BB_END (bb1)
1539 250855 : && nehedges2 > 0)
1540 3886690 : || (i1 == BB_END (bb1) && i2 == BB_END (bb2)
1541 25315 : && nehedges1 != nehedges2))
1542 : break;
1543 :
1544 3883937 : if (old_insns_match_p (0, i1, i2) != dir_both)
1545 : break;
1546 :
1547 219056 : merge_memattrs (i1, i2);
1548 :
1549 : /* Don't begin a cross-jump with a NOTE insn. */
1550 219056 : if (INSN_P (i1))
1551 : {
1552 219056 : merge_notes (i1, i2);
1553 :
1554 219056 : beforelast1 = last1, beforelast2 = last2;
1555 219056 : last1 = i1, last2 = i2;
1556 219056 : if (!stop_after || active_insn_p (i1))
1557 219056 : ninsns++;
1558 : }
1559 :
1560 219056 : if (i1 == BB_END (bb1) || i2 == BB_END (bb2)
1561 205311 : || (stop_after > 0 && ninsns == stop_after))
1562 : break;
1563 :
1564 205311 : i1 = NEXT_INSN (i1);
1565 205311 : i2 = NEXT_INSN (i2);
1566 : }
1567 :
1568 4696632 : if (ninsns)
1569 : {
1570 125051 : *f1 = last1;
1571 125051 : *f2 = last2;
1572 : }
1573 :
1574 4696632 : return ninsns;
1575 : }
1576 :
1577 : /* Return true iff outgoing edges of BB1 and BB2 match, together with
1578 : the branch instruction. This means that if we commonize the control
1579 : flow before end of the basic block, the semantic remains unchanged.
1580 :
1581 : We may assume that there exists one edge with a common destination. */
1582 :
1583 : static bool
1584 47740121 : outgoing_edges_match (int mode, basic_block bb1, basic_block bb2)
1585 : {
1586 47740121 : int nehedges1 = 0, nehedges2 = 0;
1587 47740121 : edge fallthru1 = 0, fallthru2 = 0;
1588 47740121 : edge e1, e2;
1589 47740121 : edge_iterator ei;
1590 :
1591 : /* If we performed shrink-wrapping, edges to the exit block can
1592 : only be distinguished for JUMP_INSNs. The two paths may differ in
1593 : whether they went through the prologue. Sibcalls are fine, we know
1594 : that we either didn't need or inserted an epilogue before them. */
1595 47740121 : if (crtl->shrink_wrapped
1596 1234674 : && single_succ_p (bb1)
1597 471591 : && single_succ (bb1) == EXIT_BLOCK_PTR_FOR_FN (cfun)
1598 151681 : && (!JUMP_P (BB_END (bb1))
1599 : /* Punt if the only successor is a fake edge to exit, the jump
1600 : must be some weird one. */
1601 64113 : || (single_succ_edge (bb1)->flags & EDGE_FAKE) != 0)
1602 47827769 : && !(CALL_P (BB_END (bb1)) && SIBLING_CALL_P (BB_END (bb1))))
1603 : return false;
1604 :
1605 : /* If BB1 has only one successor, we may be looking at either an
1606 : unconditional jump, or a fake edge to exit. */
1607 47690240 : if (single_succ_p (bb1)
1608 19067445 : && (single_succ_edge (bb1)->flags & (EDGE_COMPLEX | EDGE_FAKE)) == 0
1609 63161270 : && (!JUMP_P (BB_END (bb1)) || simplejump_p (BB_END (bb1))))
1610 15209541 : return (single_succ_p (bb2)
1611 13989749 : && (single_succ_edge (bb2)->flags
1612 13989749 : & (EDGE_COMPLEX | EDGE_FAKE)) == 0
1613 29118431 : && (!JUMP_P (BB_END (bb2)) || simplejump_p (BB_END (bb2))));
1614 :
1615 : /* Match conditional jumps - this may get tricky when fallthru and branch
1616 : edges are crossed. */
1617 32480699 : if (EDGE_COUNT (bb1->succs) == 2
1618 28524610 : && any_condjump_p (BB_END (bb1))
1619 51423528 : && onlyjump_p (BB_END (bb1)))
1620 : {
1621 18942829 : edge b1, f1, b2, f2;
1622 18942829 : bool reverse, match;
1623 18942829 : rtx set1, set2, cond1, cond2;
1624 18942829 : enum rtx_code code1, code2;
1625 :
1626 18942829 : if (EDGE_COUNT (bb2->succs) != 2
1627 8530262 : || !any_condjump_p (BB_END (bb2))
1628 27433020 : || !onlyjump_p (BB_END (bb2)))
1629 : return false;
1630 :
1631 8490191 : b1 = BRANCH_EDGE (bb1);
1632 8490191 : b2 = BRANCH_EDGE (bb2);
1633 8490191 : f1 = FALLTHRU_EDGE (bb1);
1634 8490191 : f2 = FALLTHRU_EDGE (bb2);
1635 :
1636 : /* Get around possible forwarders on fallthru edges. Other cases
1637 : should be optimized out already. */
1638 8490191 : if (FORWARDER_BLOCK_P (f1->dest))
1639 3300326 : f1 = single_succ_edge (f1->dest);
1640 :
1641 8490191 : if (FORWARDER_BLOCK_P (f2->dest))
1642 2438971 : f2 = single_succ_edge (f2->dest);
1643 :
1644 : /* To simplify use of this function, return false if there are
1645 : unneeded forwarder blocks. These will get eliminated later
1646 : during cleanup_cfg. */
1647 8490191 : if (FORWARDER_BLOCK_P (f1->dest)
1648 8481132 : || FORWARDER_BLOCK_P (f2->dest)
1649 8476396 : || FORWARDER_BLOCK_P (b1->dest)
1650 8452844 : || FORWARDER_BLOCK_P (b2->dest))
1651 : return false;
1652 :
1653 8424623 : if (f1->dest == f2->dest && b1->dest == b2->dest)
1654 : reverse = false;
1655 8213185 : else if (f1->dest == b2->dest && b1->dest == f2->dest)
1656 : reverse = true;
1657 : else
1658 : return false;
1659 :
1660 355770 : set1 = pc_set (BB_END (bb1));
1661 355770 : set2 = pc_set (BB_END (bb2));
1662 355770 : if ((XEXP (SET_SRC (set1), 1) == pc_rtx)
1663 355770 : != (XEXP (SET_SRC (set2), 1) == pc_rtx))
1664 0 : reverse = !reverse;
1665 :
1666 355770 : cond1 = XEXP (SET_SRC (set1), 0);
1667 355770 : cond2 = XEXP (SET_SRC (set2), 0);
1668 355770 : code1 = GET_CODE (cond1);
1669 355770 : if (reverse)
1670 144332 : code2 = reversed_comparison_code (cond2, BB_END (bb2));
1671 : else
1672 211438 : code2 = GET_CODE (cond2);
1673 :
1674 355770 : if (code2 == UNKNOWN)
1675 : return false;
1676 :
1677 : /* Verify codes and operands match. */
1678 355770 : match = ((code1 == code2
1679 304102 : && rtx_renumbered_equal_p (XEXP (cond1, 0), XEXP (cond2, 0))
1680 302032 : && rtx_renumbered_equal_p (XEXP (cond1, 1), XEXP (cond2, 1)))
1681 357840 : || (code1 == swap_condition (code2)
1682 4783 : && rtx_renumbered_equal_p (XEXP (cond1, 1),
1683 4783 : XEXP (cond2, 0))
1684 0 : && rtx_renumbered_equal_p (XEXP (cond1, 0),
1685 0 : XEXP (cond2, 1))));
1686 :
1687 : /* If we return true, we will join the blocks. Which means that
1688 : we will only have one branch prediction bit to work with. Thus
1689 : we require the existing branches to have probabilities that are
1690 : roughly similar. */
1691 : if (match
1692 302032 : && optimize_bb_for_speed_p (bb1)
1693 273978 : && optimize_bb_for_speed_p (bb2))
1694 : {
1695 263896 : profile_probability prob2;
1696 :
1697 263896 : if (b1->dest == b2->dest)
1698 159362 : prob2 = b2->probability;
1699 : else
1700 : /* Do not use f2 probability as f2 may be forwarded. */
1701 104534 : prob2 = b2->probability.invert ();
1702 :
1703 : /* Fail if the difference in probabilities is greater than 50%.
1704 : This rules out two well-predicted branches with opposite
1705 : outcomes. */
1706 263896 : if (b1->probability.differs_lot_from_p (prob2))
1707 : {
1708 2528 : if (dump_file)
1709 : {
1710 0 : fprintf (dump_file,
1711 : "Outcomes of branch in bb %i and %i differ too"
1712 : " much (", bb1->index, bb2->index);
1713 0 : b1->probability.dump (dump_file);
1714 0 : prob2.dump (dump_file);
1715 0 : fprintf (dump_file, ")\n");
1716 : }
1717 2528 : return false;
1718 : }
1719 : }
1720 :
1721 353242 : if (dump_file && match)
1722 16 : fprintf (dump_file, "Conditionals in bb %i and %i match.\n",
1723 : bb1->index, bb2->index);
1724 :
1725 : return match;
1726 : }
1727 :
1728 : /* Generic case - we are seeing a computed jump, table jump or trapping
1729 : instruction. */
1730 :
1731 : /* Check whether there are tablejumps in the end of BB1 and BB2.
1732 : Return true if they are identical. */
1733 13537870 : {
1734 13537870 : rtx_insn *label1, *label2;
1735 13537870 : rtx_jump_table_data *table1, *table2;
1736 :
1737 13537870 : if (tablejump_p (BB_END (bb1), &label1, &table1)
1738 95602 : && tablejump_p (BB_END (bb2), &label2, &table2)
1739 13540038 : && GET_CODE (PATTERN (table1)) == GET_CODE (PATTERN (table2)))
1740 : {
1741 : /* The labels should never be the same rtx. If they really are same
1742 : the jump tables are same too. So disable crossjumping of blocks BB1
1743 : and BB2 because when deleting the common insns in the end of BB1
1744 : by delete_basic_block () the jump table would be deleted too. */
1745 : /* If LABEL2 is referenced in BB1->END do not do anything
1746 : because we would loose information when replacing
1747 : LABEL1 by LABEL2 and then LABEL2 by LABEL1 in BB1->END. */
1748 2168 : if (label1 != label2 && !rtx_referenced_p (label2, BB_END (bb1)))
1749 : {
1750 : /* Set IDENTICAL to true when the tables are identical. */
1751 2168 : bool identical = false;
1752 2168 : rtx p1, p2;
1753 :
1754 2168 : p1 = PATTERN (table1);
1755 2168 : p2 = PATTERN (table2);
1756 2168 : if (GET_CODE (p1) == ADDR_VEC && rtx_equal_p (p1, p2))
1757 : {
1758 : identical = true;
1759 : }
1760 1979 : else if (GET_CODE (p1) == ADDR_DIFF_VEC
1761 389 : && (XVECLEN (p1, 1) == XVECLEN (p2, 1))
1762 267 : && rtx_equal_p (XEXP (p1, 2), XEXP (p2, 2))
1763 2246 : && rtx_equal_p (XEXP (p1, 3), XEXP (p2, 3)))
1764 : {
1765 267 : int i;
1766 :
1767 267 : identical = true;
1768 1170 : for (i = XVECLEN (p1, 1) - 1; i >= 0 && identical; i--)
1769 903 : if (!rtx_equal_p (XVECEXP (p1, 1, i), XVECEXP (p2, 1, i)))
1770 : identical = false;
1771 : }
1772 :
1773 267 : if (identical)
1774 : {
1775 237 : bool match;
1776 :
1777 : /* Temporarily replace references to LABEL1 with LABEL2
1778 : in BB1->END so that we could compare the instructions. */
1779 237 : replace_label_in_insn (BB_END (bb1), label1, label2, false);
1780 :
1781 237 : match = (old_insns_match_p (mode, BB_END (bb1), BB_END (bb2))
1782 : == dir_both);
1783 237 : if (dump_file && match)
1784 0 : fprintf (dump_file,
1785 : "Tablejumps in bb %i and %i match.\n",
1786 : bb1->index, bb2->index);
1787 :
1788 : /* Set the original label in BB1->END because when deleting
1789 : a block whose end is a tablejump, the tablejump referenced
1790 : from the instruction is deleted too. */
1791 237 : replace_label_in_insn (BB_END (bb1), label2, label1, false);
1792 :
1793 2168 : return match;
1794 : }
1795 : }
1796 : return false;
1797 : }
1798 : }
1799 :
1800 : /* Find the last non-debug non-note instruction in each bb, except
1801 : stop when we see the NOTE_INSN_BASIC_BLOCK, as old_insns_match_p
1802 : handles that case specially. old_insns_match_p does not handle
1803 : other types of instruction notes. */
1804 13535702 : rtx_insn *last1 = BB_END (bb1);
1805 13535702 : rtx_insn *last2 = BB_END (bb2);
1806 13535793 : while (!NOTE_INSN_BASIC_BLOCK_P (last1) &&
1807 13405251 : (DEBUG_INSN_P (last1) || NOTE_P (last1)))
1808 91 : last1 = PREV_INSN (last1);
1809 13554176 : while (!NOTE_INSN_BASIC_BLOCK_P (last2) &&
1810 13441312 : (DEBUG_INSN_P (last2) || NOTE_P (last2)))
1811 18474 : last2 = PREV_INSN (last2);
1812 13535702 : gcc_assert (last1 && last2);
1813 :
1814 : /* First ensure that the instructions match. There may be many outgoing
1815 : edges so this test is generally cheaper. */
1816 13535702 : if (old_insns_match_p (mode, last1, last2) != dir_both)
1817 : return false;
1818 :
1819 : /* Search the outgoing edges, ensure that the counts do match, find possible
1820 : fallthru and exception handling edges since these needs more
1821 : validation. */
1822 7551918 : if (EDGE_COUNT (bb1->succs) != EDGE_COUNT (bb2->succs))
1823 : return false;
1824 :
1825 2517293 : bool nonfakeedges = false;
1826 6212165 : FOR_EACH_EDGE (e1, ei, bb1->succs)
1827 : {
1828 3694872 : e2 = EDGE_SUCC (bb2, ei.index);
1829 :
1830 3694872 : if ((e1->flags & EDGE_FAKE) == 0)
1831 2708404 : nonfakeedges = true;
1832 :
1833 3694872 : if (e1->flags & EDGE_EH)
1834 1223378 : nehedges1++;
1835 :
1836 3694872 : if (e2->flags & EDGE_EH)
1837 1223378 : nehedges2++;
1838 :
1839 3694872 : if (e1->flags & EDGE_FALLTHRU)
1840 1273600 : fallthru1 = e1;
1841 3694872 : if (e2->flags & EDGE_FALLTHRU)
1842 1273600 : fallthru2 = e2;
1843 : }
1844 :
1845 : /* If number of edges of various types does not match, fail. */
1846 2517293 : if (nehedges1 != nehedges2
1847 2517293 : || (fallthru1 != 0) != (fallthru2 != 0))
1848 : return false;
1849 :
1850 : /* If !ACCUMULATE_OUTGOING_ARGS, bb1 (and bb2) have no successors
1851 : and the last real insn doesn't have REG_ARGS_SIZE note, don't
1852 : attempt to optimize, as the two basic blocks might have different
1853 : REG_ARGS_SIZE depths. For noreturn calls and unconditional
1854 : traps there should be REG_ARG_SIZE notes, they could be missing
1855 : for __builtin_unreachable () uses though. */
1856 2517293 : if (!nonfakeedges
1857 986468 : && !ACCUMULATE_OUTGOING_ARGS
1858 3503754 : && (!INSN_P (last1)
1859 986461 : || !find_reg_note (last1, REG_ARGS_SIZE, NULL)))
1860 : return false;
1861 :
1862 : /* fallthru edges must be forwarded to the same destination. */
1863 2517282 : if (fallthru1)
1864 : {
1865 1273600 : basic_block d1 = (FORWARDER_BLOCK_P (fallthru1->dest)
1866 1273600 : ? single_succ (fallthru1->dest) : fallthru1->dest);
1867 1273600 : basic_block d2 = (FORWARDER_BLOCK_P (fallthru2->dest)
1868 1273600 : ? single_succ (fallthru2->dest) : fallthru2->dest);
1869 :
1870 1273600 : if (d1 != d2)
1871 : return false;
1872 : }
1873 :
1874 : /* Ensure the same EH region. */
1875 1840680 : {
1876 1840680 : rtx n1 = find_reg_note (last1, REG_EH_REGION, 0);
1877 1840680 : rtx n2 = find_reg_note (last2, REG_EH_REGION, 0);
1878 :
1879 1840680 : if (!n1 && n2)
1880 : return false;
1881 :
1882 1840680 : if (n1 && (!n2 || XEXP (n1, 0) != XEXP (n2, 0)))
1883 : return false;
1884 : }
1885 :
1886 : /* The same checks as in try_crossjump_to_edge. It is required for RTL
1887 : version of sequence abstraction. */
1888 4181695 : FOR_EACH_EDGE (e1, ei, bb2->succs)
1889 : {
1890 2341045 : edge e2;
1891 2341045 : edge_iterator ei;
1892 2341045 : basic_block d1 = e1->dest;
1893 :
1894 2341045 : if (FORWARDER_BLOCK_P (d1))
1895 615209 : d1 = EDGE_SUCC (d1, 0)->dest;
1896 :
1897 2842880 : FOR_EACH_EDGE (e2, ei, bb1->succs)
1898 : {
1899 2842880 : basic_block d2 = e2->dest;
1900 2842880 : if (FORWARDER_BLOCK_P (d2))
1901 801159 : d2 = EDGE_SUCC (d2, 0)->dest;
1902 2842880 : if (d1 == d2)
1903 : break;
1904 : }
1905 :
1906 2341045 : if (!e2)
1907 0 : return false;
1908 : }
1909 :
1910 : return true;
1911 : }
1912 :
1913 : /* Returns true if BB basic block has a preserve label. */
1914 :
1915 : static bool
1916 385009 : block_has_preserve_label (basic_block bb)
1917 : {
1918 385009 : return (bb
1919 385009 : && block_label (bb)
1920 717372 : && LABEL_PRESERVE_P (block_label (bb)));
1921 : }
1922 :
1923 : /* E1 and E2 are edges with the same destination block. Search their
1924 : predecessors for common code. If found, redirect control flow from
1925 : (maybe the middle of) E1->SRC to (maybe the middle of) E2->SRC (dir_forward),
1926 : or the other way around (dir_backward). DIR specifies the allowed
1927 : replacement direction. */
1928 :
1929 : static bool
1930 52636356 : try_crossjump_to_edge (int mode, edge e1, edge e2,
1931 : enum replace_direction dir)
1932 : {
1933 52636356 : int nmatch;
1934 52636356 : basic_block src1 = e1->src, src2 = e2->src;
1935 52636356 : basic_block redirect_to, redirect_from, to_remove;
1936 52636356 : basic_block osrc1, osrc2, redirect_edges_to, tmp;
1937 52636356 : rtx_insn *newpos1, *newpos2;
1938 52636356 : edge s;
1939 52636356 : edge_iterator ei;
1940 :
1941 52636356 : newpos1 = newpos2 = NULL;
1942 :
1943 : /* Search backward through forwarder blocks. We don't need to worry
1944 : about multiple entry or chained forwarders, as they will be optimized
1945 : away. We do this to look past the unconditional jump following a
1946 : conditional jump that is required due to the current CFG shape. */
1947 52636356 : if (single_pred_p (src1)
1948 52636356 : && FORWARDER_BLOCK_P (src1))
1949 11836720 : e1 = single_pred_edge (src1), src1 = e1->src;
1950 :
1951 52636356 : if (single_pred_p (src2)
1952 52636144 : && FORWARDER_BLOCK_P (src2))
1953 5562252 : e2 = single_pred_edge (src2), src2 = e2->src;
1954 :
1955 : /* Nothing to do if we reach ENTRY, or a common source block. */
1956 52636356 : if (src1 == ENTRY_BLOCK_PTR_FOR_FN (cfun) || src2
1957 : == ENTRY_BLOCK_PTR_FOR_FN (cfun))
1958 : return false;
1959 52635940 : if (src1 == src2)
1960 : return false;
1961 :
1962 : /* Seeing more than 1 forwarder blocks would confuse us later... */
1963 52635811 : if (FORWARDER_BLOCK_P (e1->dest)
1964 52635811 : && FORWARDER_BLOCK_P (single_succ (e1->dest)))
1965 : return false;
1966 :
1967 52624206 : if (FORWARDER_BLOCK_P (e2->dest)
1968 52624206 : && FORWARDER_BLOCK_P (single_succ (e2->dest)))
1969 : return false;
1970 :
1971 : /* Likewise with dead code (possibly newly created by the other optimizations
1972 : of cfg_cleanup). */
1973 52623656 : if (EDGE_COUNT (src1->preds) == 0 || EDGE_COUNT (src2->preds) == 0)
1974 : return false;
1975 :
1976 : /* Do not turn corssing edge to non-crossing or vice versa after reload. */
1977 47948371 : if (BB_PARTITION (src1) != BB_PARTITION (src2)
1978 208250 : && reload_completed)
1979 : return false;
1980 :
1981 : /* Look for the common insn sequence, part the first ... */
1982 47740121 : if (!outgoing_edges_match (mode, src1, src2))
1983 : return false;
1984 :
1985 : /* ... and part the second. */
1986 16049269 : nmatch = flow_find_cross_jump (src1, src2, &newpos1, &newpos2, &dir);
1987 :
1988 16049269 : osrc1 = src1;
1989 16049269 : osrc2 = src2;
1990 16049269 : if (newpos1 != NULL_RTX)
1991 5177897 : src1 = BLOCK_FOR_INSN (newpos1);
1992 16049269 : if (newpos2 != NULL_RTX)
1993 5177897 : src2 = BLOCK_FOR_INSN (newpos2);
1994 :
1995 : /* Check that SRC1 and SRC2 have preds again. They may have changed
1996 : above due to the call to flow_find_cross_jump. */
1997 16049269 : if (EDGE_COUNT (src1->preds) == 0 || EDGE_COUNT (src2->preds) == 0)
1998 : return false;
1999 :
2000 16049269 : if (dir == dir_backward)
2001 : {
2002 1748 : std::swap (osrc1, osrc2);
2003 1748 : std::swap (src1, src2);
2004 1748 : std::swap (e1, e2);
2005 1748 : std::swap (newpos1, newpos2);
2006 : }
2007 :
2008 : /* Don't proceed with the crossjump unless we found a sufficient number
2009 : of matching instructions or the 'from' block was totally matched
2010 : (such that its predecessors will hopefully be redirected and the
2011 : block removed). */
2012 16049269 : if ((nmatch < param_min_crossjump_insns)
2013 15918714 : && (newpos1 != BB_HEAD (src1)))
2014 : return false;
2015 :
2016 : /* Avoid deleting preserve label when redirecting ABNORMAL edges. */
2017 385009 : if (block_has_preserve_label (e1->dest)
2018 385009 : && (e1->flags & EDGE_ABNORMAL))
2019 : return false;
2020 :
2021 : /* Here we know that the insns in the end of SRC1 which are common with SRC2
2022 : will be deleted.
2023 : If we have tablejumps in the end of SRC1 and SRC2
2024 : they have been already compared for equivalence in outgoing_edges_match ()
2025 : so replace the references to TABLE1 by references to TABLE2. */
2026 343563 : {
2027 343563 : rtx_insn *label1, *label2;
2028 343563 : rtx_jump_table_data *table1, *table2;
2029 :
2030 343563 : if (tablejump_p (BB_END (osrc1), &label1, &table1)
2031 189 : && tablejump_p (BB_END (osrc2), &label2, &table2)
2032 343752 : && label1 != label2)
2033 : {
2034 189 : rtx_insn *insn;
2035 :
2036 : /* Replace references to LABEL1 with LABEL2. */
2037 565644 : for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
2038 : {
2039 : /* Do not replace the label in SRC1->END because when deleting
2040 : a block whose end is a tablejump, the tablejump referenced
2041 : from the instruction is deleted too. */
2042 565455 : if (insn != BB_END (osrc1))
2043 565266 : replace_label_in_insn (insn, label1, label2, true);
2044 : }
2045 : }
2046 : }
2047 :
2048 : /* Avoid splitting if possible. We must always split when SRC2 has
2049 : EH predecessor edges, or we may end up with basic blocks with both
2050 : normal and EH predecessor edges. */
2051 343563 : if (newpos2 == BB_HEAD (src2)
2052 343563 : && !(EDGE_PRED (src2, 0)->flags & EDGE_EH))
2053 : redirect_to = src2;
2054 : else
2055 : {
2056 89662 : if (newpos2 == BB_HEAD (src2))
2057 : {
2058 : /* Skip possible basic block header. */
2059 6619 : if (LABEL_P (newpos2))
2060 6619 : newpos2 = NEXT_INSN (newpos2);
2061 6619 : while (DEBUG_INSN_P (newpos2))
2062 0 : newpos2 = NEXT_INSN (newpos2);
2063 6619 : if (NOTE_P (newpos2))
2064 6619 : newpos2 = NEXT_INSN (newpos2);
2065 6815 : while (DEBUG_INSN_P (newpos2))
2066 196 : newpos2 = NEXT_INSN (newpos2);
2067 : }
2068 :
2069 89662 : if (dump_file)
2070 2 : fprintf (dump_file, "Splitting bb %i before %i insns\n",
2071 : src2->index, nmatch);
2072 89662 : redirect_to = split_block (src2, PREV_INSN (newpos2))->dest;
2073 : }
2074 :
2075 343563 : if (dump_file)
2076 42 : fprintf (dump_file,
2077 : "Cross jumping from bb %i to bb %i; %i common insns\n",
2078 : src1->index, src2->index, nmatch);
2079 :
2080 : /* We may have some registers visible through the block. */
2081 343563 : df_set_bb_dirty (redirect_to);
2082 :
2083 343563 : if (osrc2 == src2)
2084 : redirect_edges_to = redirect_to;
2085 : else
2086 10480 : redirect_edges_to = osrc2;
2087 :
2088 : /* Recompute the counts of destinations of outgoing edges. */
2089 759621 : FOR_EACH_EDGE (s, ei, redirect_edges_to->succs)
2090 : {
2091 416058 : edge s2;
2092 416058 : edge_iterator ei;
2093 416058 : basic_block d = s->dest;
2094 :
2095 416058 : if (FORWARDER_BLOCK_P (d))
2096 30889 : d = single_succ (d);
2097 :
2098 497082 : FOR_EACH_EDGE (s2, ei, src1->succs)
2099 : {
2100 497082 : basic_block d2 = s2->dest;
2101 497082 : if (FORWARDER_BLOCK_P (d2))
2102 104836 : d2 = single_succ (d2);
2103 497082 : if (d == d2)
2104 : break;
2105 : }
2106 :
2107 : /* Take care to update possible forwarder blocks. We verified
2108 : that there is no more than one in the chain, so we can't run
2109 : into infinite loop. */
2110 416058 : if (FORWARDER_BLOCK_P (s->dest))
2111 30889 : s->dest->count += s->count ();
2112 :
2113 416058 : if (FORWARDER_BLOCK_P (s2->dest))
2114 81025 : s2->dest->count -= s->count ();
2115 :
2116 416058 : s->probability = s->probability.combine_with_count
2117 416058 : (redirect_edges_to->count,
2118 : s2->probability, src1->count);
2119 : }
2120 :
2121 : /* Adjust count for the block. An earlier jump
2122 : threading pass may have left the profile in an inconsistent
2123 : state (see update_bb_profile_for_threading) so we must be
2124 : prepared for overflows. */
2125 : tmp = redirect_to;
2126 365759 : do
2127 : {
2128 354661 : tmp->count += src1->count;
2129 354661 : if (tmp == redirect_edges_to)
2130 : break;
2131 11098 : tmp = find_fallthru_edge (tmp->succs)->dest;
2132 : }
2133 : while (true);
2134 343563 : update_br_prob_note (redirect_edges_to);
2135 :
2136 : /* Edit SRC1 to go to REDIRECT_TO at NEWPOS1. */
2137 :
2138 : /* Skip possible basic block header. */
2139 343563 : if (LABEL_P (newpos1))
2140 143707 : newpos1 = NEXT_INSN (newpos1);
2141 :
2142 358887 : while (DEBUG_INSN_P (newpos1))
2143 15324 : newpos1 = NEXT_INSN (newpos1);
2144 :
2145 343563 : if (NOTE_INSN_BASIC_BLOCK_P (newpos1))
2146 258225 : newpos1 = NEXT_INSN (newpos1);
2147 :
2148 : /* Skip also prologue and function markers. */
2149 1042191 : while (DEBUG_INSN_P (newpos1)
2150 1042191 : || (NOTE_P (newpos1)
2151 16387 : && (NOTE_KIND (newpos1) == NOTE_INSN_PROLOGUE_END
2152 16387 : || NOTE_KIND (newpos1) == NOTE_INSN_FUNCTION_BEG)))
2153 698628 : newpos1 = NEXT_INSN (newpos1);
2154 :
2155 343563 : redirect_from = split_block (src1, PREV_INSN (newpos1))->src;
2156 343563 : to_remove = single_succ (redirect_from);
2157 :
2158 343563 : redirect_edge_and_branch_force (single_succ_edge (redirect_from), redirect_to);
2159 343563 : delete_basic_block (to_remove);
2160 :
2161 343563 : update_forwarder_flag (redirect_from);
2162 343563 : if (redirect_to != src2)
2163 89662 : update_forwarder_flag (src2);
2164 :
2165 : return true;
2166 : }
2167 :
2168 : /* Search the predecessors of BB for common insn sequences. When found,
2169 : share code between them by redirecting control flow. Return true if
2170 : any changes made. */
2171 :
2172 : static bool
2173 31371776 : try_crossjump_bb (int mode, basic_block bb)
2174 : {
2175 31371776 : edge e, e2, fallthru;
2176 31371776 : bool changed;
2177 31371776 : unsigned max, ix, ix2;
2178 :
2179 : /* Nothing to do if there is not at least two incoming edges. */
2180 31371776 : if (EDGE_COUNT (bb->preds) < 2)
2181 : return false;
2182 :
2183 : /* Don't crossjump the predecessors of a computed jump when computed
2184 : gotos are duplicated (see pass_duplicate_computed_gotos::gate), as
2185 : that copies the jump back into each predecessor. */
2186 8612885 : if (!optimize_function_for_size_p (cfun)
2187 8404794 : && flag_expensive_optimizations
2188 8404617 : && bb != EXIT_BLOCK_PTR_FOR_FN (cfun)
2189 16551225 : && computed_jump_p (BB_END (bb)))
2190 : return false;
2191 :
2192 : /* If we are partitioning hot/cold basic blocks, we don't want to
2193 : mess up unconditional or indirect jumps that cross between hot
2194 : and cold sections.
2195 :
2196 : Basic block partitioning may result in some jumps that appear to
2197 : be optimizable (or blocks that appear to be mergeable), but which really
2198 : must be left untouched (they are required to make it safely across
2199 : partition boundaries). See the comments at the top of
2200 : bb-reorder.cc:partition_hot_cold_basic_blocks for complete details. */
2201 :
2202 8612771 : if (BB_PARTITION (EDGE_PRED (bb, 0)->src) !=
2203 8612771 : BB_PARTITION (EDGE_PRED (bb, 1)->src)
2204 8612771 : || (EDGE_PRED (bb, 0)->flags & EDGE_CROSSING))
2205 : return false;
2206 :
2207 : /* It is always cheapest to redirect a block that ends in a branch to
2208 : a block that falls through into BB, as that adds no branches to the
2209 : program. We'll try that combination first. */
2210 8395257 : fallthru = NULL;
2211 8395257 : max = param_max_crossjump_edges;
2212 :
2213 8395257 : if (EDGE_COUNT (bb->preds) > max)
2214 : return false;
2215 :
2216 8392318 : fallthru = find_fallthru_edge (bb->preds);
2217 :
2218 8392318 : changed = false;
2219 41244352 : for (ix = 0; ix < EDGE_COUNT (bb->preds);)
2220 : {
2221 24459716 : e = EDGE_PRED (bb, ix);
2222 24459716 : ix++;
2223 :
2224 : /* As noted above, first try with the fallthru predecessor (or, a
2225 : fallthru predecessor if we are in cfglayout mode). */
2226 24459716 : if (fallthru)
2227 : {
2228 : /* Don't combine the fallthru edge into anything else.
2229 : If there is a match, we'll do it the other way around. */
2230 17968357 : if (e == fallthru)
2231 6771601 : continue;
2232 : /* If nothing changed since the last attempt, there is nothing
2233 : we can do. */
2234 11196756 : if (!first_pass
2235 4422429 : && !((e->src->flags & BB_MODIFIED)
2236 3517014 : || (fallthru->src->flags & BB_MODIFIED)))
2237 3144258 : continue;
2238 :
2239 8052498 : if (try_crossjump_to_edge (mode, e, fallthru, dir_forward))
2240 : {
2241 152646 : changed = true;
2242 152646 : ix = 0;
2243 152646 : continue;
2244 : }
2245 : }
2246 :
2247 : /* Non-obvious work limiting check: Recognize that we're going
2248 : to call try_crossjump_bb on every basic block. So if we have
2249 : two blocks with lots of outgoing edges (a switch) and they
2250 : share lots of common destinations, then we would do the
2251 : cross-jump check once for each common destination.
2252 :
2253 : Now, if the blocks actually are cross-jump candidates, then
2254 : all of their destinations will be shared. Which means that
2255 : we only need check them for cross-jump candidacy once. We
2256 : can eliminate redundant checks of crossjump(A,B) by arbitrarily
2257 : choosing to do the check from the block for which the edge
2258 : in question is the first successor of A. */
2259 14391211 : if (EDGE_SUCC (e->src, 0) != e)
2260 2982639 : continue;
2261 :
2262 172695956 : for (ix2 = 0; ix2 < EDGE_COUNT (bb->preds); ix2++)
2263 : {
2264 128626267 : e2 = EDGE_PRED (bb, ix2);
2265 :
2266 128626267 : if (e2 == e)
2267 11316748 : continue;
2268 :
2269 : /* We've already checked the fallthru edge above. */
2270 117309519 : if (e2 == fallthru)
2271 6297868 : continue;
2272 :
2273 : /* The "first successor" check above only prevents multiple
2274 : checks of crossjump(A,B). In order to prevent redundant
2275 : checks of crossjump(B,A), require that A be the block
2276 : with the lowest index. */
2277 111011651 : if (e->src->index > e2->src->index)
2278 55383093 : continue;
2279 :
2280 : /* If nothing changed since the last attempt, there is nothing
2281 : we can do. */
2282 55628558 : if (!first_pass
2283 14093773 : && !((e->src->flags & BB_MODIFIED)
2284 11920159 : || (e2->src->flags & BB_MODIFIED)))
2285 11044700 : continue;
2286 :
2287 : /* Both e and e2 are not fallthru edges, so we can crossjump in either
2288 : direction. */
2289 44583858 : if (try_crossjump_to_edge (mode, e, e2, dir_both))
2290 : {
2291 : changed = true;
2292 : ix = 0;
2293 : break;
2294 : }
2295 : }
2296 : }
2297 :
2298 8392318 : if (changed)
2299 168336 : crossjumps_occurred = true;
2300 :
2301 : return changed;
2302 : }
2303 :
2304 : /* Search the successors of BB for common insn sequences. When found,
2305 : share code between them by moving it across the basic block
2306 : boundary. Return true if any changes made. */
2307 :
2308 : static bool
2309 30141282 : try_head_merge_bb (basic_block bb)
2310 : {
2311 30141282 : basic_block final_dest_bb = NULL;
2312 30141282 : int max_match = INT_MAX;
2313 30141282 : edge e0;
2314 30141282 : rtx_insn **headptr, **currptr, **nextptr;
2315 30141282 : bool changed, moveall;
2316 30141282 : unsigned ix;
2317 30141282 : rtx_insn *e0_last_head;
2318 30141282 : rtx cond;
2319 30141282 : rtx_insn *move_before;
2320 30141282 : unsigned nedges = EDGE_COUNT (bb->succs);
2321 30141282 : rtx_insn *jump = BB_END (bb);
2322 30141282 : regset live, live_union;
2323 :
2324 : /* Nothing to do if there is not at least two outgoing edges. */
2325 30141282 : if (nedges < 2)
2326 : return false;
2327 :
2328 : /* Don't crossjump if this block ends in a computed jump,
2329 : unless we are optimizing for size. */
2330 15469646 : if (optimize_bb_for_size_p (bb)
2331 1818827 : && bb != EXIT_BLOCK_PTR_FOR_FN (cfun)
2332 17288473 : && computed_jump_p (BB_END (bb)))
2333 : return false;
2334 :
2335 15469598 : cond = get_condition (jump, &move_before, true, false);
2336 15469598 : if (cond == NULL_RTX)
2337 2203929 : move_before = jump;
2338 :
2339 46727930 : for (ix = 0; ix < nedges; ix++)
2340 31258332 : if (EDGE_SUCC (bb, ix)->dest == EXIT_BLOCK_PTR_FOR_FN (cfun))
2341 : return false;
2342 :
2343 28792572 : for (ix = 0; ix < nedges; ix++)
2344 : {
2345 24100757 : edge e = EDGE_SUCC (bb, ix);
2346 24100757 : basic_block other_bb = e->dest;
2347 :
2348 24100757 : if (df_get_bb_dirty (other_bb))
2349 : {
2350 559839 : block_was_dirty = true;
2351 559839 : return false;
2352 : }
2353 :
2354 23540918 : if (e->flags & EDGE_ABNORMAL)
2355 : return false;
2356 :
2357 : /* Normally, all destination blocks must only be reachable from this
2358 : block, i.e. they must have one incoming edge.
2359 :
2360 : There is one special case we can handle, that of multiple consecutive
2361 : jumps where the first jumps to one of the targets of the second jump.
2362 : This happens frequently in switch statements for default labels.
2363 : The structure is as follows:
2364 : FINAL_DEST_BB
2365 : ....
2366 : if (cond) jump A;
2367 : fall through
2368 : BB
2369 : jump with targets A, B, C, D...
2370 : A
2371 : has two incoming edges, from FINAL_DEST_BB and BB
2372 :
2373 : In this case, we can try to move the insns through BB and into
2374 : FINAL_DEST_BB. */
2375 21419733 : if (EDGE_COUNT (other_bb->preds) != 1)
2376 : {
2377 8471109 : edge incoming_edge, incoming_bb_other_edge;
2378 8471109 : edge_iterator ei;
2379 :
2380 8471109 : if (final_dest_bb != NULL
2381 8471109 : || EDGE_COUNT (other_bb->preds) != 2)
2382 8096759 : return false;
2383 :
2384 : /* We must be able to move the insns across the whole block. */
2385 4688399 : move_before = BB_HEAD (bb);
2386 31476337 : while (!NONDEBUG_INSN_P (move_before))
2387 26787938 : move_before = NEXT_INSN (move_before);
2388 :
2389 4688399 : if (EDGE_COUNT (bb->preds) != 1)
2390 : return false;
2391 2828330 : incoming_edge = EDGE_PRED (bb, 0);
2392 2828330 : final_dest_bb = incoming_edge->src;
2393 10433524 : if (EDGE_COUNT (final_dest_bb->succs) != 2)
2394 : return false;
2395 3366098 : FOR_EACH_EDGE (incoming_bb_other_edge, ei, final_dest_bb->succs)
2396 3366098 : if (incoming_bb_other_edge != incoming_edge)
2397 : break;
2398 2336765 : if (incoming_bb_other_edge->dest != other_bb)
2399 : return false;
2400 : }
2401 : }
2402 :
2403 4691815 : e0 = EDGE_SUCC (bb, 0);
2404 4691815 : e0_last_head = NULL;
2405 4691815 : changed = false;
2406 :
2407 4816866 : for (ix = 1; ix < nedges; ix++)
2408 : {
2409 4696632 : edge e = EDGE_SUCC (bb, ix);
2410 4696632 : rtx_insn *e0_last, *e_last;
2411 4696632 : int nmatch;
2412 :
2413 4696632 : nmatch = flow_find_head_matching_sequence (e0->dest, e->dest,
2414 : &e0_last, &e_last, 0);
2415 4696632 : if (nmatch == 0)
2416 4571581 : return false;
2417 :
2418 125051 : if (nmatch < max_match)
2419 : {
2420 121207 : max_match = nmatch;
2421 121207 : e0_last_head = e0_last;
2422 : }
2423 : }
2424 :
2425 : /* If we matched an entire block, we probably have to avoid moving the
2426 : last insn. */
2427 120234 : if (max_match > 0
2428 120234 : && e0_last_head == BB_END (e0->dest)
2429 129404 : && (find_reg_note (e0_last_head, REG_EH_REGION, 0)
2430 7595 : || control_flow_insn_p (e0_last_head)))
2431 : {
2432 1588 : max_match--;
2433 1588 : if (max_match == 0)
2434 : return false;
2435 273 : e0_last_head = prev_real_nondebug_insn (e0_last_head);
2436 : }
2437 :
2438 118919 : if (max_match == 0)
2439 : return false;
2440 :
2441 : /* We must find a union of the live registers at each of the end points. */
2442 118919 : live = BITMAP_ALLOC (NULL);
2443 118919 : live_union = BITMAP_ALLOC (NULL);
2444 :
2445 118919 : currptr = XNEWVEC (rtx_insn *, nedges);
2446 118919 : headptr = XNEWVEC (rtx_insn *, nedges);
2447 118919 : nextptr = XNEWVEC (rtx_insn *, nedges);
2448 :
2449 479025 : for (ix = 0; ix < nedges; ix++)
2450 : {
2451 241187 : int j;
2452 241187 : basic_block merge_bb = EDGE_SUCC (bb, ix)->dest;
2453 241187 : rtx_insn *head = BB_HEAD (merge_bb);
2454 :
2455 1633706 : while (!NONDEBUG_INSN_P (head))
2456 1392519 : head = NEXT_INSN (head);
2457 241187 : headptr[ix] = head;
2458 241187 : currptr[ix] = head;
2459 :
2460 : /* Compute the end point and live information */
2461 383634 : for (j = 1; j < max_match; j++)
2462 205456 : do
2463 205456 : head = NEXT_INSN (head);
2464 205456 : while (!NONDEBUG_INSN_P (head));
2465 241187 : simulate_backwards_to_point (merge_bb, live, head);
2466 241187 : IOR_REG_SET (live_union, live);
2467 : }
2468 :
2469 : /* If we're moving across two blocks, verify the validity of the
2470 : first move, then adjust the target and let the loop below deal
2471 : with the final move. */
2472 118919 : if (final_dest_bb != NULL)
2473 : {
2474 7880 : rtx_insn *move_upto;
2475 :
2476 7880 : moveall = can_move_insns_across (currptr[0], e0_last_head, move_before,
2477 : jump, e0->dest, live_union,
2478 : NULL, &move_upto);
2479 7880 : if (!moveall)
2480 : {
2481 6215 : if (move_upto == NULL_RTX)
2482 5817 : goto out;
2483 :
2484 1217 : while (e0_last_head != move_upto)
2485 : {
2486 819 : df_simulate_one_insn_backwards (e0->dest, e0_last_head,
2487 : live_union);
2488 819 : e0_last_head = PREV_INSN (e0_last_head);
2489 : }
2490 : }
2491 2063 : if (e0_last_head == NULL_RTX)
2492 0 : goto out;
2493 :
2494 2063 : jump = BB_END (final_dest_bb);
2495 2063 : cond = get_condition (jump, &move_before, true, false);
2496 2063 : if (cond == NULL_RTX)
2497 0 : move_before = jump;
2498 : }
2499 :
2500 195255 : do
2501 : {
2502 195255 : rtx_insn *move_upto;
2503 195255 : moveall = can_move_insns_across (currptr[0], e0_last_head,
2504 : move_before, jump, e0->dest, live_union,
2505 : NULL, &move_upto);
2506 195255 : if (!moveall && move_upto == NULL_RTX)
2507 : {
2508 154976 : if (jump == move_before)
2509 : break;
2510 :
2511 : /* Try again, using a different insertion point. */
2512 78133 : move_before = jump;
2513 :
2514 78133 : continue;
2515 : }
2516 :
2517 40279 : if (final_dest_bb && !moveall)
2518 : /* We haven't checked whether a partial move would be OK for the first
2519 : move, so we have to fail this case. */
2520 : break;
2521 :
2522 57299 : changed = true;
2523 57299 : for (;;)
2524 : {
2525 57299 : if (currptr[0] == move_upto)
2526 : break;
2527 51245 : for (ix = 0; ix < nedges; ix++)
2528 : {
2529 34185 : rtx_insn *curr = currptr[ix];
2530 54164 : do
2531 54164 : curr = NEXT_INSN (curr);
2532 54164 : while (!NONDEBUG_INSN_P (curr));
2533 34185 : currptr[ix] = curr;
2534 : }
2535 : }
2536 :
2537 : /* If we can't currently move all of the identical insns, remember
2538 : each insn after the range that we'll merge. */
2539 40239 : if (!moveall)
2540 17445 : for (ix = 0; ix < nedges; ix++)
2541 : {
2542 11635 : rtx_insn *curr = currptr[ix];
2543 16481 : do
2544 16481 : curr = NEXT_INSN (curr);
2545 16481 : while (!NONDEBUG_INSN_P (curr));
2546 11635 : nextptr[ix] = curr;
2547 : }
2548 :
2549 40239 : reorder_insns (headptr[0], currptr[0], PREV_INSN (move_before));
2550 40239 : df_set_bb_dirty (EDGE_SUCC (bb, 0)->dest);
2551 40239 : if (final_dest_bb != NULL)
2552 2020 : df_set_bb_dirty (final_dest_bb);
2553 40239 : df_set_bb_dirty (bb);
2554 121989 : for (ix = 1; ix < nedges; ix++)
2555 : {
2556 41511 : df_set_bb_dirty (EDGE_SUCC (bb, ix)->dest);
2557 41511 : delete_insn_chain (headptr[ix], currptr[ix], false);
2558 : }
2559 40239 : if (!moveall)
2560 : {
2561 5810 : if (jump == move_before)
2562 : break;
2563 :
2564 : /* For the unmerged insns, try a different insertion point. */
2565 4020 : move_before = jump;
2566 :
2567 12060 : for (ix = 0; ix < nedges; ix++)
2568 8040 : currptr[ix] = headptr[ix] = nextptr[ix];
2569 : }
2570 : }
2571 116582 : while (!moveall);
2572 :
2573 34429 : out:
2574 118919 : free (currptr);
2575 118919 : free (headptr);
2576 118919 : free (nextptr);
2577 :
2578 118919 : crossjumps_occurred |= changed;
2579 :
2580 118919 : return changed;
2581 : }
2582 :
2583 : /* Return true if BB contains just bb note, or bb note followed
2584 : by only DEBUG_INSNs. */
2585 :
2586 : static bool
2587 16737144 : trivially_empty_bb_p (basic_block bb)
2588 : {
2589 16737144 : rtx_insn *insn = BB_END (bb);
2590 :
2591 5598 : while (1)
2592 : {
2593 16742742 : if (insn == BB_HEAD (bb))
2594 : return true;
2595 16741785 : if (!DEBUG_INSN_P (insn))
2596 : return false;
2597 5598 : insn = PREV_INSN (insn);
2598 : }
2599 : }
2600 :
2601 : /* Return true if BB contains just a return and possibly a USE of the
2602 : return value. Fill in *RET and *USE with the return and use insns
2603 : if any found, otherwise NULL. All CLOBBERs are ignored. */
2604 :
2605 : bool
2606 412836118 : bb_is_just_return (basic_block bb, rtx_insn **ret, rtx_insn **use)
2607 : {
2608 412836118 : *ret = *use = NULL;
2609 412836118 : rtx_insn *insn;
2610 :
2611 412836118 : if (bb == EXIT_BLOCK_PTR_FOR_FN (cfun))
2612 : return false;
2613 :
2614 554502515 : FOR_BB_INSNS_REVERSE (bb, insn)
2615 540710783 : if (NONDEBUG_INSN_P (insn))
2616 : {
2617 408050165 : rtx pat = PATTERN (insn);
2618 :
2619 408050165 : if (!*ret && ANY_RETURN_P (pat))
2620 7549504 : *ret = insn;
2621 7893500 : else if (*ret && !*use && GET_CODE (pat) == USE
2622 1141623 : && REG_P (XEXP (pat, 0))
2623 401642284 : && REG_FUNCTION_VALUE_P (XEXP (pat, 0)))
2624 1138337 : *use = insn;
2625 399362324 : else if (GET_CODE (pat) != CLOBBER)
2626 : return false;
2627 : }
2628 :
2629 13791732 : return !!*ret;
2630 : }
2631 :
2632 : /* Do simple CFG optimizations - basic block merging, simplifying of jump
2633 : instructions etc. Return nonzero if changes were made. */
2634 :
2635 : static bool
2636 22345486 : try_optimize_cfg (int mode)
2637 : {
2638 22345486 : bool changed_overall = false;
2639 22345486 : bool changed;
2640 22345486 : int iterations = 0;
2641 22345486 : basic_block bb, b, next;
2642 :
2643 22345486 : if (mode & (CLEANUP_CROSSJUMP | CLEANUP_THREADING))
2644 3058271 : clear_bb_flags ();
2645 :
2646 22345486 : crossjumps_occurred = false;
2647 :
2648 327809089 : FOR_EACH_BB_FN (bb, cfun)
2649 305463603 : update_forwarder_flag (bb);
2650 :
2651 22345486 : if (! targetm.cannot_modify_jumps_p ())
2652 : {
2653 22345486 : first_pass = true;
2654 : /* Attempt to merge blocks as made possible by edge removal. If
2655 : a block has only one successor, and the successor has only
2656 : one predecessor, they may be combined. */
2657 52487326 : do
2658 : {
2659 26243663 : block_was_dirty = false;
2660 26243663 : changed = false;
2661 26243663 : iterations++;
2662 :
2663 26243663 : if (dump_file)
2664 1863 : fprintf (dump_file,
2665 : "\n\ntry_optimize_cfg iteration %i\n\n",
2666 : iterations);
2667 :
2668 26243663 : for (b = ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb; b
2669 443279922 : != EXIT_BLOCK_PTR_FOR_FN (cfun);)
2670 : {
2671 417036259 : basic_block c;
2672 417036259 : edge s;
2673 417036259 : bool changed_here = false;
2674 :
2675 : /* Delete trivially dead basic blocks. This is either
2676 : blocks with no predecessors, or empty blocks with no
2677 : successors. However if the empty block with no
2678 : successors is the successor of the ENTRY_BLOCK, it is
2679 : kept. This ensures that the ENTRY_BLOCK will have a
2680 : successor which is a precondition for many RTL
2681 : passes. Empty blocks may result from expanding
2682 : __builtin_unreachable (). */
2683 417036259 : if (EDGE_COUNT (b->preds) == 0
2684 417036259 : || (EDGE_COUNT (b->succs) == 0
2685 16737144 : && trivially_empty_bb_p (b)
2686 957 : && single_succ_edge (ENTRY_BLOCK_PTR_FOR_FN (cfun))->dest
2687 : != b))
2688 : {
2689 5705759 : c = b->prev_bb;
2690 5705759 : if (EDGE_COUNT (b->preds) > 0)
2691 : {
2692 957 : edge e;
2693 957 : edge_iterator ei;
2694 :
2695 957 : if (current_ir_type () == IR_RTL_CFGLAYOUT)
2696 : {
2697 51 : rtx_insn *insn;
2698 51 : for (insn = BB_FOOTER (b);
2699 51 : insn; insn = NEXT_INSN (insn))
2700 51 : if (BARRIER_P (insn))
2701 : break;
2702 51 : if (insn)
2703 102 : FOR_EACH_EDGE (e, ei, b->preds)
2704 51 : if ((e->flags & EDGE_FALLTHRU))
2705 : {
2706 51 : if (BB_FOOTER (b)
2707 51 : && BB_FOOTER (e->src) == NULL)
2708 : {
2709 51 : BB_FOOTER (e->src) = BB_FOOTER (b);
2710 51 : BB_FOOTER (b) = NULL;
2711 : }
2712 : else
2713 0 : emit_barrier_after_bb (e->src);
2714 : }
2715 : }
2716 : else
2717 : {
2718 906 : rtx_insn *last = get_last_bb_insn (b);
2719 906 : if (last && BARRIER_P (last))
2720 1812 : FOR_EACH_EDGE (e, ei, b->preds)
2721 906 : if ((e->flags & EDGE_FALLTHRU))
2722 906 : emit_barrier_after (BB_END (e->src));
2723 : }
2724 : }
2725 5705759 : delete_basic_block (b);
2726 5705759 : changed = true;
2727 : /* Avoid trying to remove the exit block. */
2728 5705759 : b = (c == ENTRY_BLOCK_PTR_FOR_FN (cfun) ? c->next_bb : c);
2729 5843465 : continue;
2730 5705759 : }
2731 :
2732 : /* Remove code labels no longer used. */
2733 411330500 : if (single_pred_p (b)
2734 299194667 : && (single_pred_edge (b)->flags & EDGE_FALLTHRU)
2735 210622154 : && !(single_pred_edge (b)->flags & EDGE_COMPLEX)
2736 207229780 : && LABEL_P (BB_HEAD (b))
2737 853282 : && !LABEL_PRESERVE_P (BB_HEAD (b))
2738 : /* If the previous block ends with a branch to this
2739 : block, we can't delete the label. Normally this
2740 : is a condjump that is yet to be simplified, but
2741 : if CASE_DROPS_THRU, this can be a tablejump with
2742 : some element going to the same place as the
2743 : default (fallthru). */
2744 412168510 : && (single_pred (b) == ENTRY_BLOCK_PTR_FOR_FN (cfun)
2745 837958 : || !JUMP_P (BB_END (single_pred (b)))
2746 638929 : || ! label_is_jump_target_p (BB_HEAD (b),
2747 638929 : BB_END (single_pred (b)))))
2748 : {
2749 831533 : delete_insn (BB_HEAD (b));
2750 831533 : if (dump_file)
2751 37 : fprintf (dump_file, "Deleted label in block %i.\n",
2752 : b->index);
2753 : }
2754 :
2755 : /* If we fall through an empty block, we can remove it. */
2756 411468206 : if (!(mode & (CLEANUP_CFGLAYOUT | CLEANUP_NO_INSN_DEL))
2757 87689662 : && single_pred_p (b)
2758 64875293 : && (single_pred_edge (b)->flags & EDGE_FALLTHRU)
2759 45878867 : && !LABEL_P (BB_HEAD (b))
2760 45454697 : && FORWARDER_BLOCK_P (b)
2761 : /* Note that forwarder_block_p true ensures that
2762 : there is a successor for this block. */
2763 5403410 : && (single_succ_edge (b)->flags & EDGE_FALLTHRU)
2764 411555090 : && n_basic_blocks_for_fn (cfun) > NUM_FIXED_BLOCKS + 1)
2765 : {
2766 137706 : if (dump_file)
2767 11 : fprintf (dump_file,
2768 : "Deleting fallthru block %i.\n",
2769 : b->index);
2770 :
2771 2623 : c = ((b->prev_bb == ENTRY_BLOCK_PTR_FOR_FN (cfun))
2772 137706 : ? b->next_bb : b->prev_bb);
2773 137706 : redirect_edge_succ_nodup (single_pred_edge (b),
2774 : single_succ (b));
2775 137706 : delete_basic_block (b);
2776 137706 : changed = true;
2777 137706 : b = c;
2778 137706 : continue;
2779 : }
2780 :
2781 : /* Merge B with its single successor, if any. */
2782 820670139 : if (single_succ_p (b)
2783 180121607 : && (s = single_succ_edge (b))
2784 180121607 : && !(s->flags & EDGE_COMPLEX)
2785 170549959 : && (c = s->dest) != EXIT_BLOCK_PTR_FOR_FN (cfun)
2786 144265468 : && single_pred_p (c)
2787 418431328 : && b != c)
2788 : {
2789 : /* When not in cfg_layout mode use code aware of reordering
2790 : INSN. This code possibly creates new basic blocks so it
2791 : does not fit merge_blocks interface and is kept here in
2792 : hope that it will become useless once more of compiler
2793 : is transformed to use cfg_layout mode. */
2794 :
2795 8953983 : if ((mode & CLEANUP_CFGLAYOUT)
2796 8953983 : && can_merge_blocks_p (b, c))
2797 : {
2798 734739 : merge_blocks (b, c);
2799 734739 : update_forwarder_flag (b);
2800 734739 : changed_here = true;
2801 : }
2802 8219244 : else if (!(mode & CLEANUP_CFGLAYOUT)
2803 : /* If the jump insn has side effects,
2804 : we can't kill the edge. */
2805 6988650 : && (!JUMP_P (BB_END (b))
2806 2284171 : || (reload_completed
2807 3142059 : ? simplejump_p (BB_END (b))
2808 857888 : : (onlyjump_p (BB_END (b))
2809 857071 : && !tablejump_p (BB_END (b),
2810 : NULL, NULL))))
2811 18347971 : && (next = merge_blocks_move (s, b, c, mode)))
2812 : {
2813 : b = next;
2814 : changed_here = true;
2815 : }
2816 : }
2817 :
2818 : /* Try to change a branch to a return to just that return. */
2819 411192794 : rtx_insn *ret, *use;
2820 411192794 : if (single_succ_p (b)
2821 178463526 : && onlyjump_p (BB_END (b))
2822 439352496 : && bb_is_just_return (single_succ (b), &ret, &use))
2823 : {
2824 37349 : if (redirect_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2825 37349 : PATTERN (ret), 0))
2826 : {
2827 37349 : if (use)
2828 22057 : emit_insn_before (copy_insn (PATTERN (use)),
2829 22057 : BB_END (b));
2830 37349 : if (dump_file)
2831 23 : fprintf (dump_file, "Changed jump %d->%d to return.\n",
2832 23 : b->index, single_succ (b)->index);
2833 37349 : redirect_edge_succ (single_succ_edge (b),
2834 37349 : EXIT_BLOCK_PTR_FOR_FN (cfun));
2835 37349 : single_succ_edge (b)->flags &= ~EDGE_CROSSING;
2836 37349 : changed_here = true;
2837 : }
2838 : }
2839 :
2840 : /* Try to change a conditional branch to a return to the
2841 : respective conditional return. */
2842 411192794 : if (EDGE_COUNT (b->succs) == 2
2843 215522775 : && any_condjump_p (BB_END (b))
2844 598744343 : && bb_is_just_return (BRANCH_EDGE (b)->dest, &ret, &use))
2845 : {
2846 601490 : if (redirect_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2847 601490 : PATTERN (ret), 0))
2848 : {
2849 0 : if (use)
2850 0 : emit_insn_before (copy_insn (PATTERN (use)),
2851 0 : BB_END (b));
2852 0 : if (dump_file)
2853 0 : fprintf (dump_file, "Changed conditional jump %d->%d "
2854 : "to conditional return.\n",
2855 0 : b->index, BRANCH_EDGE (b)->dest->index);
2856 0 : redirect_edge_succ (BRANCH_EDGE (b),
2857 0 : EXIT_BLOCK_PTR_FOR_FN (cfun));
2858 0 : BRANCH_EDGE (b)->flags &= ~EDGE_CROSSING;
2859 0 : changed_here = true;
2860 : }
2861 : }
2862 :
2863 : /* Try to flip a conditional branch that falls through to
2864 : a return so that it becomes a conditional return and a
2865 : new jump to the original branch target. */
2866 411192794 : if (EDGE_COUNT (b->succs) == 2
2867 215522775 : && BRANCH_EDGE (b)->dest != EXIT_BLOCK_PTR_FOR_FN (cfun)
2868 215522775 : && any_condjump_p (BB_END (b))
2869 598744343 : && bb_is_just_return (FALLTHRU_EDGE (b)->dest, &ret, &use))
2870 : {
2871 163364 : if (invert_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2872 163364 : JUMP_LABEL (BB_END (b)), 0))
2873 : {
2874 163364 : basic_block new_ft = BRANCH_EDGE (b)->dest;
2875 163364 : if (redirect_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2876 163364 : PATTERN (ret), 0))
2877 : {
2878 0 : if (use)
2879 0 : emit_insn_before (copy_insn (PATTERN (use)),
2880 0 : BB_END (b));
2881 0 : if (dump_file)
2882 0 : fprintf (dump_file, "Changed conditional jump "
2883 : "%d->%d to conditional return, adding "
2884 : "fall-through jump.\n",
2885 0 : b->index, BRANCH_EDGE (b)->dest->index);
2886 0 : redirect_edge_succ (BRANCH_EDGE (b),
2887 0 : EXIT_BLOCK_PTR_FOR_FN (cfun));
2888 0 : BRANCH_EDGE (b)->flags &= ~EDGE_CROSSING;
2889 0 : std::swap (BRANCH_EDGE (b)->probability,
2890 0 : FALLTHRU_EDGE (b)->probability);
2891 0 : update_br_prob_note (b);
2892 0 : basic_block jb = force_nonfallthru (FALLTHRU_EDGE (b));
2893 0 : notice_new_block (jb);
2894 0 : if (!redirect_jump (as_a <rtx_jump_insn *> (BB_END (jb)),
2895 0 : block_label (new_ft), 0))
2896 0 : gcc_unreachable ();
2897 0 : redirect_edge_succ (single_succ_edge (jb), new_ft);
2898 0 : changed_here = true;
2899 : }
2900 : else
2901 : {
2902 : /* Invert the jump back to what it was. This should
2903 : never fail. */
2904 163364 : if (!invert_jump (as_a <rtx_jump_insn *> (BB_END (b)),
2905 163364 : JUMP_LABEL (BB_END (b)), 0))
2906 0 : gcc_unreachable ();
2907 : }
2908 : }
2909 : }
2910 :
2911 : /* Simplify branch over branch. */
2912 411192794 : if ((mode & CLEANUP_EXPENSIVE)
2913 108633738 : && !(mode & CLEANUP_CFGLAYOUT)
2914 446571481 : && try_simplify_condjump (b))
2915 : changed_here = true;
2916 :
2917 : /* If B has a single outgoing edge, but uses a
2918 : non-trivial jump instruction without side-effects, we
2919 : can either delete the jump entirely, or replace it
2920 : with a simple unconditional jump. */
2921 411192794 : if (single_succ_p (b)
2922 178463586 : && single_succ (b) != EXIT_BLOCK_PTR_FOR_FN (cfun)
2923 148067325 : && onlyjump_p (BB_END (b))
2924 29905156 : && !CROSSING_JUMP_P (BB_END (b))
2925 438112590 : && try_redirect_by_replacing_jump (single_succ_edge (b),
2926 : single_succ (b),
2927 28703205 : (mode & CLEANUP_CFGLAYOUT) != 0))
2928 : {
2929 4381479 : update_forwarder_flag (b);
2930 4381479 : changed_here = true;
2931 : }
2932 :
2933 : /* Simplify branch to branch. */
2934 411192794 : if (try_forward_edges (mode, b))
2935 : {
2936 6592493 : update_forwarder_flag (b);
2937 6592493 : changed_here = true;
2938 : }
2939 :
2940 : /* Look for shared code between blocks. */
2941 411192794 : if ((mode & CLEANUP_CROSSJUMP)
2942 411192794 : && try_crossjump_bb (mode, b))
2943 : changed_here = true;
2944 :
2945 411192794 : if ((mode & CLEANUP_CROSSJUMP)
2946 : /* This can lengthen register lifetimes. Do it only after
2947 : reload. */
2948 30141282 : && reload_completed
2949 441334076 : && try_head_merge_bb (b))
2950 : changed_here = true;
2951 :
2952 : /* Don't get confused by the index shift caused by
2953 : deleting blocks. */
2954 411153519 : if (!changed_here)
2955 396079175 : b = b->next_bb;
2956 : else
2957 : changed = true;
2958 : }
2959 :
2960 26243663 : if ((mode & CLEANUP_CROSSJUMP)
2961 26243663 : && try_crossjump_bb (mode, EXIT_BLOCK_PTR_FOR_FN (cfun)))
2962 : changed = true;
2963 :
2964 26243663 : if (block_was_dirty)
2965 : {
2966 : /* This should only be set by head-merging. */
2967 200471 : gcc_assert (mode & CLEANUP_CROSSJUMP);
2968 200471 : df_analyze ();
2969 : }
2970 :
2971 26243663 : if (changed)
2972 : {
2973 : /* Edge forwarding in particular can cause hot blocks previously
2974 : reached by both hot and cold blocks to become dominated only
2975 : by cold blocks. This will cause the verification below to fail,
2976 : and lead to now cold code in the hot section. This is not easy
2977 : to detect and fix during edge forwarding, and in some cases
2978 : is only visible after newly unreachable blocks are deleted,
2979 : which will be done in fixup_partitions. */
2980 3898177 : if ((mode & CLEANUP_NO_PARTITIONING) == 0)
2981 : {
2982 3759878 : fixup_partitions ();
2983 3759878 : checking_verify_flow_info ();
2984 : }
2985 : }
2986 :
2987 26243663 : changed_overall |= changed;
2988 26243663 : first_pass = false;
2989 : }
2990 : while (changed);
2991 : }
2992 :
2993 362121956 : FOR_ALL_BB_FN (b, cfun)
2994 339776470 : b->flags &= ~(BB_FORWARDER_BLOCK | BB_NONTHREADABLE_BLOCK);
2995 :
2996 22345486 : return changed_overall;
2997 : }
2998 :
2999 : /* Delete all unreachable basic blocks. */
3000 :
3001 : bool
3002 28842651 : delete_unreachable_blocks (void)
3003 : {
3004 28842651 : bool changed = false;
3005 28842651 : basic_block b, prev_bb;
3006 :
3007 28842651 : find_unreachable_blocks ();
3008 :
3009 : /* When we're in GIMPLE mode and there may be debug bind insns, we
3010 : should delete blocks in reverse dominator order, so as to get a
3011 : chance to substitute all released DEFs into debug bind stmts. If
3012 : we don't have dominators information, walking blocks backward
3013 : gets us a better chance of retaining most debug information than
3014 : otherwise. */
3015 13088817 : if (MAY_HAVE_DEBUG_BIND_INSNS && current_ir_type () == IR_GIMPLE
3016 29119820 : && dom_info_available_p (CDI_DOMINATORS))
3017 : {
3018 0 : for (b = EXIT_BLOCK_PTR_FOR_FN (cfun)->prev_bb;
3019 0 : b != ENTRY_BLOCK_PTR_FOR_FN (cfun); b = prev_bb)
3020 : {
3021 0 : prev_bb = b->prev_bb;
3022 :
3023 0 : if (!(b->flags & BB_REACHABLE))
3024 : {
3025 : /* Speed up the removal of blocks that don't dominate
3026 : others. Walking backwards, this should be the common
3027 : case. */
3028 0 : if (!first_dom_son (CDI_DOMINATORS, b))
3029 0 : delete_basic_block (b);
3030 : else
3031 : {
3032 0 : auto_vec<basic_block> h
3033 0 : = get_all_dominated_blocks (CDI_DOMINATORS, b);
3034 :
3035 0 : while (h.length ())
3036 : {
3037 0 : b = h.pop ();
3038 :
3039 0 : prev_bb = b->prev_bb;
3040 :
3041 0 : gcc_assert (!(b->flags & BB_REACHABLE));
3042 :
3043 0 : delete_basic_block (b);
3044 : }
3045 0 : }
3046 :
3047 : changed = true;
3048 : }
3049 : }
3050 : }
3051 : else
3052 : {
3053 28842651 : for (b = EXIT_BLOCK_PTR_FOR_FN (cfun)->prev_bb;
3054 430050377 : b != ENTRY_BLOCK_PTR_FOR_FN (cfun); b = prev_bb)
3055 : {
3056 401207726 : prev_bb = b->prev_bb;
3057 :
3058 401207726 : if (!(b->flags & BB_REACHABLE))
3059 : {
3060 1756330 : delete_basic_block (b);
3061 1756330 : changed = true;
3062 : }
3063 : }
3064 : }
3065 :
3066 28842651 : if (changed)
3067 469591 : tidy_fallthru_edges ();
3068 28842651 : return changed;
3069 : }
3070 :
3071 : /* Delete any jump tables never referenced. We can't delete them at the
3072 : time of removing tablejump insn as they are referenced by the preceding
3073 : insns computing the destination, so we delay deleting and garbagecollect
3074 : them once life information is computed. */
3075 : void
3076 9354817 : delete_dead_jumptables (void)
3077 : {
3078 9354817 : basic_block bb;
3079 :
3080 : /* A dead jump table does not belong to any basic block. Scan insns
3081 : between two adjacent basic blocks. */
3082 104483572 : FOR_EACH_BB_FN (bb, cfun)
3083 : {
3084 95128755 : rtx_insn *insn, *next;
3085 :
3086 95128755 : for (insn = NEXT_INSN (BB_END (bb));
3087 174084372 : insn && !NOTE_INSN_BASIC_BLOCK_P (insn);
3088 78955617 : insn = next)
3089 : {
3090 78955617 : next = NEXT_INSN (insn);
3091 78955617 : if (LABEL_P (insn)
3092 44767079 : && LABEL_NUSES (insn) == LABEL_PRESERVE_P (insn)
3093 81029641 : && JUMP_TABLE_DATA_P (next))
3094 : {
3095 0 : rtx_insn *label = insn, *jump = next;
3096 :
3097 0 : if (dump_file)
3098 0 : fprintf (dump_file, "Dead jumptable %i removed\n",
3099 0 : INSN_UID (insn));
3100 :
3101 0 : next = NEXT_INSN (next);
3102 0 : delete_insn (jump);
3103 0 : delete_insn (label);
3104 : }
3105 : }
3106 : }
3107 9354817 : }
3108 :
3109 :
3110 : /* Tidy the CFG by deleting unreachable code and whatnot. */
3111 :
3112 : bool
3113 20139711 : cleanup_cfg (int mode)
3114 : {
3115 20139711 : bool changed = false;
3116 :
3117 : /* Set the cfglayout mode flag here. We could update all the callers
3118 : but that is just inconvenient, especially given that we eventually
3119 : want to have cfglayout mode as the default. */
3120 20139711 : if (current_ir_type () == IR_RTL_CFGLAYOUT)
3121 13371317 : mode |= CLEANUP_CFGLAYOUT;
3122 :
3123 20139711 : timevar_push (TV_CLEANUP_CFG);
3124 20139711 : if (delete_unreachable_blocks ())
3125 : {
3126 158985 : changed = true;
3127 : /* We've possibly created trivially dead code. Cleanup it right
3128 : now to introduce more opportunities for try_optimize_cfg. */
3129 158985 : if (!(mode & (CLEANUP_NO_INSN_DEL))
3130 42056 : && !reload_completed)
3131 41991 : delete_trivially_dead_insns (get_insns (), max_reg_num ());
3132 : }
3133 :
3134 20139711 : compact_blocks ();
3135 :
3136 : /* To tail-merge blocks ending in the same noreturn function (e.g.
3137 : a call to abort) we have to insert fake edges to exit. Do this
3138 : here once. The fake edges do not interfere with any other CFG
3139 : cleanups. */
3140 20139711 : if (mode & CLEANUP_CROSSJUMP)
3141 988538 : add_noreturn_fake_exit_edges ();
3142 :
3143 20139711 : if (!dbg_cnt (cfg_cleanup))
3144 : return changed;
3145 :
3146 22345486 : while (try_optimize_cfg (mode))
3147 : {
3148 3805255 : delete_unreachable_blocks (), changed = true;
3149 3805255 : if (!(mode & CLEANUP_NO_INSN_DEL))
3150 : {
3151 : /* Try to remove some trivially dead insns when doing an expensive
3152 : cleanup. But delete_trivially_dead_insns doesn't work after
3153 : reload (it only handles pseudos) and run_fast_dce is too costly
3154 : to run in every iteration.
3155 :
3156 : For effective cross jumping, we really want to run a fast DCE to
3157 : clean up any dead conditions, or they get in the way of performing
3158 : useful tail merges.
3159 :
3160 : Other transformations in cleanup_cfg are not so sensitive to dead
3161 : code, so delete_trivially_dead_insns or even doing nothing at all
3162 : is good enough. */
3163 620250 : if ((mode & CLEANUP_EXPENSIVE) && !reload_completed
3164 2217095 : && !delete_trivially_dead_insns (get_insns (), max_reg_num ()))
3165 : break;
3166 2205775 : if ((mode & CLEANUP_CROSSJUMP) && crossjumps_occurred)
3167 : {
3168 83108 : run_fast_dce ();
3169 83108 : mode &= ~CLEANUP_FORCE_FAST_DCE;
3170 : }
3171 : }
3172 : else
3173 : break;
3174 : }
3175 :
3176 20139711 : if (mode & CLEANUP_CROSSJUMP)
3177 988538 : remove_fake_exit_edges ();
3178 :
3179 20139711 : if (mode & CLEANUP_FORCE_FAST_DCE)
3180 122029 : run_fast_dce ();
3181 :
3182 : /* Don't call delete_dead_jumptables in cfglayout mode, because
3183 : that function assumes that jump tables are in the insns stream.
3184 : But we also don't _have_ to delete dead jumptables in cfglayout
3185 : mode because we shouldn't even be looking at things that are
3186 : not in a basic block. Dead jumptables are cleaned up when
3187 : going out of cfglayout mode. */
3188 20139711 : if (!(mode & CLEANUP_CFGLAYOUT))
3189 6768394 : delete_dead_jumptables ();
3190 :
3191 : /* ??? We probably do this way too often. */
3192 20139711 : if (current_loops
3193 9381227 : && (changed
3194 6954319 : || (mode & CLEANUP_CFG_CHANGED)))
3195 : {
3196 2522142 : timevar_push (TV_REPAIR_LOOPS);
3197 : /* The above doesn't preserve dominance info if available. */
3198 2522142 : gcc_assert (!dom_info_available_p (CDI_DOMINATORS));
3199 2522142 : calculate_dominance_info (CDI_DOMINATORS);
3200 2522142 : fix_loop_structure (NULL);
3201 2522142 : free_dominance_info (CDI_DOMINATORS);
3202 2522142 : timevar_pop (TV_REPAIR_LOOPS);
3203 : }
3204 :
3205 20139711 : timevar_pop (TV_CLEANUP_CFG);
3206 :
3207 20139711 : return changed;
3208 : }
3209 :
3210 : namespace {
3211 :
3212 : const pass_data pass_data_jump =
3213 : {
3214 : RTL_PASS, /* type */
3215 : "jump", /* name */
3216 : OPTGROUP_NONE, /* optinfo_flags */
3217 : TV_JUMP, /* tv_id */
3218 : 0, /* properties_required */
3219 : 0, /* properties_provided */
3220 : 0, /* properties_destroyed */
3221 : 0, /* todo_flags_start */
3222 : 0, /* todo_flags_finish */
3223 : };
3224 :
3225 : class pass_jump : public rtl_opt_pass
3226 : {
3227 : public:
3228 294536 : pass_jump (gcc::context *ctxt)
3229 589072 : : rtl_opt_pass (pass_data_jump, ctxt)
3230 : {}
3231 :
3232 : /* opt_pass methods: */
3233 : unsigned int execute (function *) final override;
3234 :
3235 : }; // class pass_jump
3236 :
3237 : unsigned int
3238 1515918 : pass_jump::execute (function *)
3239 : {
3240 1515918 : delete_trivially_dead_insns (get_insns (), max_reg_num ());
3241 1515918 : if (dump_file)
3242 84 : dump_flow_info (dump_file, dump_flags);
3243 3031836 : cleanup_cfg ((optimize ? CLEANUP_EXPENSIVE : 0)
3244 1070056 : | (flag_thread_jumps && flag_expensive_optimizations
3245 1515918 : ? CLEANUP_THREADING : 0));
3246 1515918 : return 0;
3247 : }
3248 :
3249 : } // anon namespace
3250 :
3251 : rtl_opt_pass *
3252 294536 : make_pass_jump (gcc::context *ctxt)
3253 : {
3254 294536 : return new pass_jump (ctxt);
3255 : }
3256 :
3257 : namespace {
3258 :
3259 : const pass_data pass_data_jump_after_combine =
3260 : {
3261 : RTL_PASS, /* type */
3262 : "jump_after_combine", /* name */
3263 : OPTGROUP_NONE, /* optinfo_flags */
3264 : TV_JUMP, /* tv_id */
3265 : 0, /* properties_required */
3266 : 0, /* properties_provided */
3267 : 0, /* properties_destroyed */
3268 : 0, /* todo_flags_start */
3269 : 0, /* todo_flags_finish */
3270 : };
3271 :
3272 : class pass_jump_after_combine : public rtl_opt_pass
3273 : {
3274 : public:
3275 294536 : pass_jump_after_combine (gcc::context *ctxt)
3276 589072 : : rtl_opt_pass (pass_data_jump_after_combine, ctxt)
3277 : {}
3278 :
3279 : /* opt_pass methods: */
3280 1515929 : bool gate (function *) final override
3281 : {
3282 1515929 : return flag_thread_jumps && flag_expensive_optimizations;
3283 : }
3284 : unsigned int execute (function *) final override;
3285 :
3286 : }; // class pass_jump_after_combine
3287 :
3288 : unsigned int
3289 988377 : pass_jump_after_combine::execute (function *)
3290 : {
3291 : /* Jump threading does not keep dominators up-to-date. */
3292 988377 : free_dominance_info (CDI_DOMINATORS);
3293 988377 : cleanup_cfg (CLEANUP_THREADING);
3294 988377 : return 0;
3295 : }
3296 :
3297 : } // anon namespace
3298 :
3299 : rtl_opt_pass *
3300 294536 : make_pass_jump_after_combine (gcc::context *ctxt)
3301 : {
3302 294536 : return new pass_jump_after_combine (ctxt);
3303 : }
3304 :
3305 : namespace {
3306 :
3307 : const pass_data pass_data_jump2 =
3308 : {
3309 : RTL_PASS, /* type */
3310 : "jump2", /* name */
3311 : OPTGROUP_NONE, /* optinfo_flags */
3312 : TV_JUMP, /* tv_id */
3313 : 0, /* properties_required */
3314 : 0, /* properties_provided */
3315 : 0, /* properties_destroyed */
3316 : 0, /* todo_flags_start */
3317 : 0, /* todo_flags_finish */
3318 : };
3319 :
3320 : class pass_jump2 : public rtl_opt_pass
3321 : {
3322 : public:
3323 294536 : pass_jump2 (gcc::context *ctxt)
3324 589072 : : rtl_opt_pass (pass_data_jump2, ctxt)
3325 : {}
3326 :
3327 : /* opt_pass methods: */
3328 1515922 : unsigned int execute (function *) final override
3329 : {
3330 2043306 : cleanup_cfg (flag_crossjumping ? CLEANUP_CROSSJUMP : 0);
3331 1515922 : return 0;
3332 : }
3333 :
3334 : }; // class pass_jump2
3335 :
3336 : } // anon namespace
3337 :
3338 : rtl_opt_pass *
3339 294536 : make_pass_jump2 (gcc::context *ctxt)
3340 : {
3341 294536 : return new pass_jump2 (ctxt);
3342 : }
|
