Line data Source code
1 : /* Perform instruction reorganizations for delay slot filling.
2 : Copyright (C) 1992-2026 Free Software Foundation, Inc.
3 : Contributed by Richard Kenner (kenner@vlsi1.ultra.nyu.edu).
4 : Hacked by Michael Tiemann (tiemann@cygnus.com).
5 :
6 : This file is part of GCC.
7 :
8 : GCC is free software; you can redistribute it and/or modify it under
9 : the terms of the GNU General Public License as published by the Free
10 : Software Foundation; either version 3, or (at your option) any later
11 : version.
12 :
13 : GCC is distributed in the hope that it will be useful, but WITHOUT ANY
14 : WARRANTY; without even the implied warranty of MERCHANTABILITY or
15 : FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 : for more details.
17 :
18 : You should have received a copy of the GNU General Public License
19 : along with GCC; see the file COPYING3. If not see
20 : <http://www.gnu.org/licenses/>. */
21 :
22 : /* Instruction reorganization pass.
23 :
24 : This pass runs after register allocation and final jump
25 : optimization. It should be the last pass to run before peephole.
26 : It serves primarily to fill delay slots of insns, typically branch
27 : and call insns. Other insns typically involve more complicated
28 : interactions of data dependencies and resource constraints, and
29 : are better handled by scheduling before register allocation (by the
30 : function `schedule_insns').
31 :
32 : The Branch Penalty is the number of extra cycles that are needed to
33 : execute a branch insn. On an ideal machine, branches take a single
34 : cycle, and the Branch Penalty is 0. Several RISC machines approach
35 : branch delays differently:
36 :
37 : The MIPS has a single branch delay slot. Most insns
38 : (except other branches) can be used to fill this slot. When the
39 : slot is filled, two insns execute in two cycles, reducing the
40 : branch penalty to zero.
41 :
42 : The SPARC always has a branch delay slot, but its effects can be
43 : annulled when the branch is not taken. This means that failing to
44 : find other sources of insns, we can hoist an insn from the branch
45 : target that would only be safe to execute knowing that the branch
46 : is taken.
47 :
48 : The HP-PA always has a branch delay slot. For unconditional branches
49 : its effects can be annulled when the branch is taken. The effects
50 : of the delay slot in a conditional branch can be nullified for forward
51 : taken branches, or for untaken backward branches. This means
52 : we can hoist insns from the fall-through path for forward branches or
53 : steal insns from the target of backward branches.
54 :
55 : The TMS320C3x and C4x have three branch delay slots. When the three
56 : slots are filled, the branch penalty is zero. Most insns can fill the
57 : delay slots except jump insns.
58 :
59 : Three techniques for filling delay slots have been implemented so far:
60 :
61 : (1) `fill_simple_delay_slots' is the simplest, most efficient way
62 : to fill delay slots. This pass first looks for insns which come
63 : from before the branch and which are safe to execute after the
64 : branch. Then it searches after the insn requiring delay slots or,
65 : in the case of a branch, for insns that are after the point at
66 : which the branch merges into the fallthrough code, if such a point
67 : exists. When such insns are found, the branch penalty decreases
68 : and no code expansion takes place.
69 :
70 : (2) `fill_eager_delay_slots' is more complicated: it is used for
71 : scheduling conditional jumps, or for scheduling jumps which cannot
72 : be filled using (1). A machine need not have annulled jumps to use
73 : this strategy, but it helps (by keeping more options open).
74 : `fill_eager_delay_slots' tries to guess the direction the branch
75 : will go; if it guesses right 100% of the time, it can reduce the
76 : branch penalty as much as `fill_simple_delay_slots' does. If it
77 : guesses wrong 100% of the time, it might as well schedule nops. When
78 : `fill_eager_delay_slots' takes insns from the fall-through path of
79 : the jump, usually there is no code expansion; when it takes insns
80 : from the branch target, there is code expansion if it is not the
81 : only way to reach that target.
82 :
83 : (3) `relax_delay_slots' uses a set of rules to simplify code that
84 : has been reorganized by (1) and (2). It finds cases where
85 : conditional test can be eliminated, jumps can be threaded, extra
86 : insns can be eliminated, etc. It is the job of (1) and (2) to do a
87 : good job of scheduling locally; `relax_delay_slots' takes care of
88 : making the various individual schedules work well together. It is
89 : especially tuned to handle the control flow interactions of branch
90 : insns. It does nothing for insns with delay slots that do not
91 : branch. */
92 :
93 : #include "config.h"
94 : #include "system.h"
95 : #include "coretypes.h"
96 : #include "backend.h"
97 : #include "target.h"
98 : #include "rtl.h"
99 : #include "tree.h"
100 : #include "predict.h"
101 : #include "memmodel.h"
102 : #include "tm_p.h"
103 : #include "expmed.h"
104 : #include "insn-config.h"
105 : #include "emit-rtl.h"
106 : #include "recog.h"
107 : #include "insn-attr.h"
108 : #include "resource.h"
109 : #include "tree-pass.h"
110 :
111 :
112 : /* First, some functions that were used before GCC got a control flow graph.
113 : These functions are now only used here in reorg.cc, and have therefore
114 : been moved here to avoid inadvertent misuse elsewhere in the compiler. */
115 :
116 : /* Return the last label to mark the same position as LABEL. Return LABEL
117 : itself if it is null or any return rtx. */
118 :
119 : static rtx
120 0 : skip_consecutive_labels (rtx label_or_return)
121 : {
122 0 : rtx_insn *insn;
123 :
124 0 : if (label_or_return && ANY_RETURN_P (label_or_return))
125 : return label_or_return;
126 :
127 0 : rtx_insn *label = as_a <rtx_insn *> (label_or_return);
128 :
129 : /* __builtin_unreachable can create a CODE_LABEL followed by a BARRIER.
130 :
131 : Since reaching the CODE_LABEL is undefined behavior, we can return
132 : any code label and we're OK at run time.
133 :
134 : However, if we return a CODE_LABEL which leads to a shrink-wrapped
135 : epilogue, but the path does not have a prologue, then we will trip
136 : a sanity check in the dwarf2 cfi code which wants to verify that
137 : the CFIs are all the same on the traces leading to the epilogue.
138 :
139 : So we explicitly disallow looking through BARRIERS here. */
140 0 : for (insn = label;
141 0 : insn != 0 && !INSN_P (insn) && !BARRIER_P (insn);
142 0 : insn = NEXT_INSN (insn))
143 0 : if (LABEL_P (insn))
144 0 : label = insn;
145 :
146 : return label;
147 : }
148 :
149 : /* Insns which have delay slots that have not yet been filled. */
150 :
151 : static struct obstack unfilled_slots_obstack;
152 : static rtx *unfilled_firstobj;
153 :
154 : /* Define macros to refer to the first and last slot containing unfilled
155 : insns. These are used because the list may move and its address
156 : should be recomputed at each use. */
157 :
158 : #define unfilled_slots_base \
159 : ((rtx_insn **) obstack_base (&unfilled_slots_obstack))
160 :
161 : #define unfilled_slots_next \
162 : ((rtx_insn **) obstack_next_free (&unfilled_slots_obstack))
163 :
164 : /* Points to the label before the end of the function, or before a
165 : return insn. */
166 : static rtx_code_label *function_return_label;
167 : /* Likewise for a simple_return. */
168 : static rtx_code_label *function_simple_return_label;
169 :
170 : /* Mapping between INSN_UID's and position in the code since INSN_UID's do
171 : not always monotonically increase. */
172 : static int *uid_to_ruid;
173 :
174 : /* Highest valid index in `uid_to_ruid'. */
175 : static int max_uid;
176 :
177 : static bool stop_search_p (rtx_insn *, bool);
178 : static bool resource_conflicts_p (struct resources *, struct resources *);
179 : static bool insn_references_resource_p (rtx, struct resources *, bool);
180 : static bool insn_sets_resource_p (rtx, struct resources *, bool);
181 : static rtx_code_label *find_end_label (rtx);
182 : static rtx_insn *emit_delay_sequence (rtx_insn *, const vec<rtx_insn *> &,
183 : int);
184 : static void add_to_delay_list (rtx_insn *, vec<rtx_insn *> *);
185 : static rtx_insn *delete_from_delay_slot (rtx_insn *);
186 : static void delete_scheduled_jump (rtx_insn *);
187 : static void note_delay_statistics (int, int);
188 : static int get_jump_flags (const rtx_insn *, rtx);
189 : static int mostly_true_jump (rtx);
190 : static rtx get_branch_condition (const rtx_insn *, rtx);
191 : static bool condition_dominates_p (rtx, const rtx_insn *);
192 : static bool redirect_with_delay_slots_safe_p (rtx_insn *, rtx, rtx);
193 : static bool redirect_with_delay_list_safe_p (rtx_insn *, rtx,
194 : const vec<rtx_insn *> &);
195 : static bool check_annul_list_true_false (bool, const vec<rtx_insn *> &);
196 : static void steal_delay_list_from_target (rtx_insn *, rtx, rtx_sequence *,
197 : vec<rtx_insn *> *,
198 : struct resources *,
199 : struct resources *,
200 : struct resources *,
201 : int, int *, bool *,
202 : rtx *);
203 : static void steal_delay_list_from_fallthrough (rtx_insn *, rtx, rtx_sequence *,
204 : vec<rtx_insn *> *,
205 : struct resources *,
206 : struct resources *,
207 : struct resources *,
208 : int, int *, bool *);
209 : static void try_merge_delay_insns (rtx_insn *, rtx_insn *);
210 : static rtx_insn *redundant_insn (rtx, rtx_insn *, const vec<rtx_insn *> &);
211 : static bool own_thread_p (rtx, rtx, bool);
212 : static void update_block (rtx_insn *, rtx_insn *);
213 : static bool reorg_redirect_jump (rtx_jump_insn *, rtx);
214 : static void update_reg_dead_notes (rtx_insn *, rtx_insn *);
215 : static void fix_reg_dead_note (rtx_insn *, rtx);
216 : static void update_reg_unused_notes (rtx_insn *, rtx);
217 : static void fill_simple_delay_slots (bool);
218 : static void fill_slots_from_thread (rtx_jump_insn *, rtx, rtx, rtx,
219 : bool, bool, bool, int,
220 : int *, vec<rtx_insn *> *);
221 : static void fill_eager_delay_slots (void);
222 : static void relax_delay_slots (rtx_insn *);
223 : static void make_return_insns (rtx_insn *);
224 :
225 : /* A wrapper around next_active_insn which takes care to return ret_rtx
226 : unchanged. */
227 :
228 : static rtx
229 0 : first_active_target_insn (rtx insn)
230 : {
231 0 : if (ANY_RETURN_P (insn))
232 : return insn;
233 0 : return next_active_insn (as_a <rtx_insn *> (insn));
234 : }
235 :
236 : /* Return true iff INSN is a simplejump, or any kind of return insn. */
237 :
238 : static bool
239 0 : simplejump_or_return_p (rtx insn)
240 : {
241 0 : return (JUMP_P (insn)
242 0 : && (simplejump_p (as_a <rtx_insn *> (insn))
243 0 : || ANY_RETURN_P (PATTERN (insn))));
244 : }
245 :
246 : /* Return TRUE if this insn should stop the search for insn to fill delay
247 : slots. LABELS_P indicates that labels should terminate the search.
248 : In all cases, jumps terminate the search. */
249 :
250 : static bool
251 0 : stop_search_p (rtx_insn *insn, bool labels_p)
252 : {
253 0 : if (insn == 0)
254 : return true;
255 :
256 : /* If the insn can throw an exception that is caught within the function,
257 : it may effectively perform a jump from the viewpoint of the function.
258 : Therefore act like for a jump. */
259 0 : if (can_throw_internal (insn))
260 : return true;
261 :
262 0 : switch (GET_CODE (insn))
263 : {
264 : case NOTE:
265 : case CALL_INSN:
266 : case DEBUG_INSN:
267 : return false;
268 :
269 : case CODE_LABEL:
270 : return labels_p;
271 :
272 : case JUMP_INSN:
273 : case BARRIER:
274 : return true;
275 :
276 0 : case INSN:
277 : /* OK unless it contains a delay slot or is an `asm' insn of some type.
278 : We don't know anything about these. */
279 0 : return (GET_CODE (PATTERN (insn)) == SEQUENCE
280 0 : || GET_CODE (PATTERN (insn)) == ASM_INPUT
281 0 : || asm_noperands (PATTERN (insn)) >= 0);
282 :
283 0 : default:
284 0 : gcc_unreachable ();
285 : }
286 : }
287 :
288 : /* Return TRUE if any resources are marked in both RES1 and RES2 or if either
289 : resource set contains a volatile memory reference. Otherwise, return FALSE. */
290 :
291 : static bool
292 0 : resource_conflicts_p (struct resources *res1, struct resources *res2)
293 : {
294 0 : if ((res1->cc && res2->cc) || (res1->memory && res2->memory)
295 0 : || res1->volatil || res2->volatil)
296 : return true;
297 :
298 0 : return hard_reg_set_intersect_p (res1->regs, res2->regs);
299 : }
300 :
301 : /* Return TRUE if any resource marked in RES, a `struct resources', is
302 : referenced by INSN. If INCLUDE_DELAYED_EFFECTS is set, return if the called
303 : routine is using those resources.
304 :
305 : We compute this by computing all the resources referenced by INSN and
306 : seeing if this conflicts with RES. It might be faster to directly check
307 : ourselves, and this is the way it used to work, but it means duplicating
308 : a large block of complex code. */
309 :
310 : static bool
311 0 : insn_references_resource_p (rtx insn, struct resources *res,
312 : bool include_delayed_effects)
313 : {
314 0 : struct resources insn_res;
315 :
316 0 : CLEAR_RESOURCE (&insn_res);
317 0 : mark_referenced_resources (insn, &insn_res, include_delayed_effects);
318 0 : return resource_conflicts_p (&insn_res, res);
319 : }
320 :
321 : /* Return TRUE if INSN modifies resources that are marked in RES.
322 : INCLUDE_DELAYED_EFFECTS is set if the actions of that routine should be
323 : included. */
324 :
325 : static bool
326 0 : insn_sets_resource_p (rtx insn, struct resources *res,
327 : bool include_delayed_effects)
328 : {
329 0 : struct resources insn_sets;
330 :
331 0 : CLEAR_RESOURCE (&insn_sets);
332 0 : mark_set_resources (insn, &insn_sets, 0,
333 : (include_delayed_effects
334 : ? MARK_SRC_DEST_CALL
335 : : MARK_SRC_DEST));
336 0 : return resource_conflicts_p (&insn_sets, res);
337 : }
338 :
339 : /* Find a label before a RETURN. If there is none, try to make one; if this
340 : fails, return 0. KIND is either ret_rtx or simple_return_rtx, indicating
341 : which type of RETURN we're looking for.
342 :
343 : The property of the label is that it is placed just before a bare RETURN
344 : insn, so that another bare RETURN can be turned into a jump to the label
345 : unconditionally. In particular, the label cannot be placed before a
346 : RETURN insn with a filled delay slot.
347 :
348 : ??? There may be a problem with the current implementation. Suppose
349 : we start with a bare RETURN insn and call find_end_label. It may set
350 : function_return_label just before the RETURN. Suppose the machinery
351 : is able to fill the delay slot of the RETURN insn afterwards. Then
352 : function_return_label is no longer valid according to the property
353 : described above and find_end_label will still return it unmodified.
354 : Note that this is probably mitigated by the following observation:
355 : once function_return_label is made, it is very likely the target of
356 : a jump, so filling the delay slot of the RETURN will be much more
357 : difficult. */
358 :
359 : static rtx_code_label *
360 0 : find_end_label (rtx kind)
361 : {
362 0 : rtx_insn *insn;
363 0 : rtx_code_label **plabel;
364 :
365 0 : if (kind == ret_rtx)
366 0 : plabel = &function_return_label;
367 : else
368 : {
369 0 : gcc_assert (kind == simple_return_rtx);
370 0 : plabel = &function_simple_return_label;
371 : }
372 :
373 : /* If we found one previously, return it. */
374 0 : if (*plabel)
375 : return *plabel;
376 :
377 : /* Otherwise, scan the insns backward from the end of the function. */
378 0 : insn = get_last_insn ();
379 0 : while (NOTE_P (insn)
380 0 : || (NONJUMP_INSN_P (insn)
381 0 : && (GET_CODE (PATTERN (insn)) == USE
382 0 : || GET_CODE (PATTERN (insn)) == CLOBBER)))
383 0 : insn = PREV_INSN (insn);
384 :
385 : /* First, see if there is a RETURN at the end of the function. If so,
386 : put the label before it. */
387 0 : if (BARRIER_P (insn)
388 0 : && JUMP_P (PREV_INSN (insn))
389 0 : && PATTERN (PREV_INSN (insn)) == kind)
390 : {
391 0 : rtx_insn *temp = PREV_INSN (PREV_INSN (insn));
392 0 : rtx_code_label *label = gen_label_rtx ();
393 0 : LABEL_NUSES (label) = 0;
394 :
395 : /* Put the label before any USE insns that may precede the
396 : RETURN insn. */
397 0 : while (GET_CODE (temp) == USE)
398 0 : temp = PREV_INSN (temp);
399 :
400 0 : emit_label_after (label, temp);
401 0 : *plabel = label;
402 : }
403 :
404 : /* If the basic block reordering pass has moved the return insn to some
405 : other place, try to locate it again and put the label there. */
406 : else
407 : {
408 0 : rtx_code_label *label = gen_label_rtx ();
409 0 : LABEL_NUSES (label) = 0;
410 0 : while (insn && ! (JUMP_P (insn) && (PATTERN (insn) == kind)))
411 0 : insn = PREV_INSN (insn);
412 0 : if (insn)
413 : {
414 0 : insn = PREV_INSN (insn);
415 :
416 : /* Put the label before any USE insns that may precede the
417 : RETURN insn. */
418 0 : while (GET_CODE (insn) == USE)
419 0 : insn = PREV_INSN (insn);
420 :
421 0 : emit_label_after (label, insn);
422 : }
423 : else
424 : {
425 0 : if (targetm.have_epilogue () && ! targetm.have_return ())
426 : /* The RETURN insn has its delay slot filled so we cannot
427 : emit the label just before it. Since we already have
428 : an epilogue and cannot emit a new RETURN, we cannot
429 : emit the label at all. */
430 : return NULL;
431 :
432 : /* Otherwise, make a new label and emit a RETURN and BARRIER,
433 : if needed. */
434 0 : emit_label (label);
435 0 : if (targetm.have_return ())
436 : {
437 : /* The return we make may have delay slots too. */
438 0 : rtx_insn *pat = targetm.gen_return ();
439 0 : rtx_insn *insn = emit_jump_insn (pat);
440 0 : set_return_jump_label (insn);
441 0 : emit_barrier ();
442 0 : if (num_delay_slots (insn) > 0)
443 0 : obstack_ptr_grow (&unfilled_slots_obstack, insn);
444 : }
445 : }
446 0 : *plabel = label;
447 : }
448 :
449 : /* Show one additional use for this label so it won't go away until
450 : we are done. */
451 0 : ++LABEL_NUSES (*plabel);
452 :
453 0 : return *plabel;
454 : }
455 :
456 : /* Put INSN and LIST together in a SEQUENCE rtx of LENGTH, and replace
457 : the pattern of INSN with the SEQUENCE.
458 :
459 : Returns the insn containing the SEQUENCE that replaces INSN. */
460 :
461 : static rtx_insn *
462 0 : emit_delay_sequence (rtx_insn *insn, const vec<rtx_insn *> &list, int length)
463 : {
464 : /* Allocate the rtvec to hold the insns and the SEQUENCE. */
465 0 : rtvec seqv = rtvec_alloc (length + 1);
466 0 : rtx seq = gen_rtx_SEQUENCE (VOIDmode, seqv);
467 0 : rtx_insn *seq_insn = make_insn_raw (seq);
468 :
469 : /* If DELAY_INSN has a location, use it for SEQ_INSN. If DELAY_INSN does
470 : not have a location, but one of the delayed insns does, we pick up a
471 : location from there later. */
472 0 : INSN_LOCATION (seq_insn) = INSN_LOCATION (insn);
473 :
474 : /* Unlink INSN from the insn chain, so that we can put it into
475 : the SEQUENCE. Remember where we want to emit SEQUENCE in AFTER. */
476 0 : rtx_insn *after = PREV_INSN (insn);
477 0 : remove_insn (insn);
478 0 : SET_NEXT_INSN (insn) = SET_PREV_INSN (insn) = NULL;
479 :
480 : /* Build our SEQUENCE and rebuild the insn chain. */
481 0 : start_sequence ();
482 0 : XVECEXP (seq, 0, 0) = emit_insn (insn);
483 :
484 0 : unsigned int delay_insns = list.length ();
485 0 : gcc_assert (delay_insns == (unsigned int) length);
486 0 : for (unsigned int i = 0; i < delay_insns; i++)
487 : {
488 0 : rtx_insn *tem = list[i];
489 0 : rtx note, next;
490 :
491 : /* Show that this copy of the insn isn't deleted. */
492 0 : tem->set_undeleted ();
493 :
494 : /* Unlink insn from its original place, and re-emit it into
495 : the sequence. */
496 0 : SET_NEXT_INSN (tem) = SET_PREV_INSN (tem) = NULL;
497 0 : XVECEXP (seq, 0, i + 1) = emit_insn (tem);
498 :
499 : /* SPARC assembler, for instance, emit warning when debug info is output
500 : into the delay slot. */
501 0 : if (INSN_LOCATION (tem) && !INSN_LOCATION (seq_insn))
502 0 : INSN_LOCATION (seq_insn) = INSN_LOCATION (tem);
503 0 : INSN_LOCATION (tem) = 0;
504 :
505 0 : for (note = REG_NOTES (tem); note; note = next)
506 : {
507 0 : next = XEXP (note, 1);
508 0 : switch (REG_NOTE_KIND (note))
509 : {
510 0 : case REG_DEAD:
511 : /* Remove any REG_DEAD notes because we can't rely on them now
512 : that the insn has been moved. */
513 0 : remove_note (tem, note);
514 0 : break;
515 :
516 0 : case REG_LABEL_OPERAND:
517 0 : case REG_LABEL_TARGET:
518 : /* Keep the label reference count up to date. */
519 0 : if (LABEL_P (XEXP (note, 0)))
520 0 : LABEL_NUSES (XEXP (note, 0)) ++;
521 : break;
522 :
523 : default:
524 : break;
525 : }
526 : }
527 : }
528 0 : end_sequence ();
529 :
530 : /* Splice our SEQUENCE into the insn stream where INSN used to be. */
531 0 : add_insn_after (seq_insn, after, NULL);
532 :
533 0 : return seq_insn;
534 : }
535 :
536 : /* Add INSN to DELAY_LIST and return the head of the new list. The list must
537 : be in the order in which the insns are to be executed. */
538 :
539 : static void
540 0 : add_to_delay_list (rtx_insn *insn, vec<rtx_insn *> *delay_list)
541 : {
542 : /* If INSN has its block number recorded, clear it since we may
543 : be moving the insn to a new block. */
544 0 : clear_hashed_info_for_insn (insn);
545 :
546 0 : delay_list->safe_push (insn);
547 0 : }
548 :
549 : /* Delete INSN from the delay slot of the insn that it is in, which may
550 : produce an insn with no delay slots. Return the new insn. */
551 :
552 : static rtx_insn *
553 0 : delete_from_delay_slot (rtx_insn *insn)
554 : {
555 0 : rtx_insn *trial, *seq_insn, *prev;
556 0 : rtx_sequence *seq;
557 0 : bool had_barrier = false;
558 0 : int i;
559 :
560 : /* We first must find the insn containing the SEQUENCE with INSN in its
561 : delay slot. Do this by finding an insn, TRIAL, where
562 : PREV_INSN (NEXT_INSN (TRIAL)) != TRIAL. */
563 :
564 0 : for (trial = insn;
565 0 : PREV_INSN (NEXT_INSN (trial)) == trial;
566 : trial = NEXT_INSN (trial))
567 : ;
568 :
569 0 : seq_insn = PREV_INSN (NEXT_INSN (trial));
570 0 : seq = as_a <rtx_sequence *> (PATTERN (seq_insn));
571 :
572 0 : if (NEXT_INSN (seq_insn) && BARRIER_P (NEXT_INSN (seq_insn)))
573 : had_barrier = true;
574 :
575 : /* Create a delay list consisting of all the insns other than the one
576 : we are deleting (unless we were the only one). */
577 0 : auto_vec<rtx_insn *, 5> delay_list;
578 0 : if (seq->len () > 2)
579 0 : for (i = 1; i < seq->len (); i++)
580 0 : if (seq->insn (i) != insn)
581 0 : add_to_delay_list (seq->insn (i), &delay_list);
582 :
583 : /* Delete the old SEQUENCE, re-emit the insn that used to have the delay
584 : list, and rebuild the delay list if non-empty. */
585 0 : prev = PREV_INSN (seq_insn);
586 0 : trial = seq->insn (0);
587 0 : delete_related_insns (seq_insn);
588 0 : add_insn_after (trial, prev, NULL);
589 :
590 : /* If there was a barrier after the old SEQUENCE, remit it. */
591 0 : if (had_barrier)
592 0 : emit_barrier_after (trial);
593 :
594 : /* If there are any delay insns, remit them. Otherwise clear the
595 : annul flag. */
596 0 : if (!delay_list.is_empty ())
597 0 : trial = emit_delay_sequence (trial, delay_list, XVECLEN (seq, 0) - 2);
598 0 : else if (JUMP_P (trial))
599 0 : INSN_ANNULLED_BRANCH_P (trial) = 0;
600 :
601 0 : INSN_FROM_TARGET_P (insn) = 0;
602 :
603 : /* Show we need to fill this insn again. */
604 0 : obstack_ptr_grow (&unfilled_slots_obstack, trial);
605 :
606 0 : return trial;
607 0 : }
608 :
609 : /* Delete INSN, a JUMP_INSN. */
610 :
611 : static void
612 0 : delete_scheduled_jump (rtx_insn *insn)
613 : {
614 0 : delete_related_insns (insn);
615 0 : }
616 :
617 : /* Counters for delay-slot filling. */
618 :
619 : #define NUM_REORG_FUNCTIONS 2
620 : #define MAX_DELAY_HISTOGRAM 3
621 : #define MAX_REORG_PASSES 2
622 :
623 : static int num_insns_needing_delays[NUM_REORG_FUNCTIONS][MAX_REORG_PASSES];
624 :
625 : static int num_filled_delays[NUM_REORG_FUNCTIONS][MAX_DELAY_HISTOGRAM+1][MAX_REORG_PASSES];
626 :
627 : static int reorg_pass_number;
628 :
629 : static void
630 0 : note_delay_statistics (int slots_filled, int index)
631 : {
632 0 : num_insns_needing_delays[index][reorg_pass_number]++;
633 0 : if (slots_filled > MAX_DELAY_HISTOGRAM)
634 : slots_filled = MAX_DELAY_HISTOGRAM;
635 0 : num_filled_delays[index][slots_filled][reorg_pass_number]++;
636 0 : }
637 :
638 : /* Optimize the following cases:
639 :
640 : 1. When a conditional branch skips over only one instruction,
641 : use an annulling branch and put that insn in the delay slot.
642 : Use either a branch that annuls when the condition if true or
643 : invert the test with a branch that annuls when the condition is
644 : false. This saves insns, since otherwise we must copy an insn
645 : from the L1 target.
646 :
647 : (orig) (skip) (otherwise)
648 : Bcc.n L1 Bcc',a L1 Bcc,a L1'
649 : insn insn insn2
650 : L1: L1: L1:
651 : insn2 insn2 insn2
652 : insn3 insn3 L1':
653 : insn3
654 :
655 : 2. When a conditional branch skips over only one instruction,
656 : and after that, it unconditionally branches somewhere else,
657 : perform the similar optimization. This saves executing the
658 : second branch in the case where the inverted condition is true.
659 :
660 : Bcc.n L1 Bcc',a L2
661 : insn insn
662 : L1: L1:
663 : Bra L2 Bra L2
664 :
665 : INSN is a JUMP_INSN.
666 :
667 : This should be expanded to skip over N insns, where N is the number
668 : of delay slots required. */
669 :
670 : static void
671 0 : optimize_skip (rtx_jump_insn *insn, vec<rtx_insn *> *delay_list)
672 : {
673 0 : rtx_insn *trial = next_nonnote_insn (insn);
674 0 : rtx_insn *next_trial = next_active_insn (trial);
675 0 : int flags;
676 :
677 0 : flags = get_jump_flags (insn, JUMP_LABEL (insn));
678 :
679 0 : if (trial == 0
680 0 : || !NONJUMP_INSN_P (trial)
681 0 : || GET_CODE (PATTERN (trial)) == SEQUENCE
682 0 : || recog_memoized (trial) < 0
683 0 : || (! eligible_for_annul_false (insn, 0, trial, flags)
684 0 : && ! eligible_for_annul_true (insn, 0, trial, flags))
685 0 : || RTX_FRAME_RELATED_P (trial)
686 0 : || can_throw_internal (trial))
687 : return;
688 :
689 : /* There are two cases where we are just executing one insn (we assume
690 : here that a branch requires only one insn; this should be generalized
691 : at some point): Where the branch goes around a single insn or where
692 : we have one insn followed by a branch to the same label we branch to.
693 : In both of these cases, inverting the jump and annulling the delay
694 : slot give the same effect in fewer insns. */
695 0 : if (next_trial == next_active_insn (JUMP_LABEL_AS_INSN (insn))
696 0 : || (next_trial != 0
697 0 : && simplejump_or_return_p (next_trial)
698 0 : && JUMP_LABEL (insn) == JUMP_LABEL (next_trial)))
699 : {
700 0 : if (eligible_for_annul_false (insn, 0, trial, flags))
701 : {
702 0 : if (invert_jump (insn, JUMP_LABEL (insn), 1))
703 0 : INSN_FROM_TARGET_P (trial) = 1;
704 0 : else if (! eligible_for_annul_true (insn, 0, trial, flags))
705 : return;
706 : }
707 :
708 0 : add_to_delay_list (trial, delay_list);
709 0 : next_trial = next_active_insn (trial);
710 0 : update_block (trial, trial);
711 0 : delete_related_insns (trial);
712 :
713 : /* Also, if we are targeting an unconditional
714 : branch, thread our jump to the target of that branch. Don't
715 : change this into a RETURN here, because it may not accept what
716 : we have in the delay slot. We'll fix this up later. */
717 0 : if (next_trial && simplejump_or_return_p (next_trial))
718 : {
719 0 : rtx target_label = JUMP_LABEL (next_trial);
720 0 : if (ANY_RETURN_P (target_label))
721 0 : target_label = find_end_label (target_label);
722 :
723 0 : if (target_label)
724 : {
725 : /* Recompute the flags based on TARGET_LABEL since threading
726 : the jump to TARGET_LABEL may change the direction of the
727 : jump (which may change the circumstances in which the
728 : delay slot is nullified). */
729 0 : flags = get_jump_flags (insn, target_label);
730 0 : if (eligible_for_annul_true (insn, 0, trial, flags))
731 0 : reorg_redirect_jump (insn, target_label);
732 : }
733 : }
734 :
735 0 : INSN_ANNULLED_BRANCH_P (insn) = 1;
736 : }
737 : }
738 :
739 : /* Encode and return branch direction and prediction information for
740 : INSN assuming it will jump to LABEL.
741 :
742 : Non conditional branches return no direction information and
743 : are predicted as very likely taken. */
744 :
745 : static int
746 0 : get_jump_flags (const rtx_insn *insn, rtx label)
747 : {
748 0 : int flags;
749 :
750 : /* get_jump_flags can be passed any insn with delay slots, these may
751 : be INSNs, CALL_INSNs, or JUMP_INSNs. Only JUMP_INSNs have branch
752 : direction information, and only if they are conditional jumps.
753 :
754 : If LABEL is a return, then there is no way to determine the branch
755 : direction. */
756 0 : if (JUMP_P (insn)
757 0 : && (condjump_p (insn) || condjump_in_parallel_p (insn))
758 0 : && !ANY_RETURN_P (label)
759 0 : && INSN_UID (insn) <= max_uid
760 0 : && INSN_UID (label) <= max_uid)
761 0 : flags
762 0 : = (uid_to_ruid[INSN_UID (label)] > uid_to_ruid[INSN_UID (insn)])
763 0 : ? ATTR_FLAG_forward : ATTR_FLAG_backward;
764 : /* No valid direction information. */
765 : else
766 : flags = 0;
767 :
768 0 : return flags;
769 : }
770 :
771 : /* Return truth value of the statement that this branch
772 : is mostly taken. If we think that the branch is extremely likely
773 : to be taken, we return 2. If the branch is slightly more likely to be
774 : taken, return 1. If the branch is slightly less likely to be taken,
775 : return 0 and if the branch is highly unlikely to be taken, return -1. */
776 :
777 : static int
778 0 : mostly_true_jump (rtx jump_insn)
779 : {
780 : /* If branch probabilities are available, then use that number since it
781 : always gives a correct answer. */
782 0 : rtx note = find_reg_note (jump_insn, REG_BR_PROB, 0);
783 0 : if (note)
784 : {
785 0 : int prob = profile_probability::from_reg_br_prob_note (XINT (note, 0))
786 0 : .to_reg_br_prob_base ();
787 :
788 0 : if (prob >= REG_BR_PROB_BASE * 9 / 10)
789 : return 2;
790 0 : else if (prob >= REG_BR_PROB_BASE / 2)
791 : return 1;
792 0 : else if (prob >= REG_BR_PROB_BASE / 10)
793 : return 0;
794 : else
795 0 : return -1;
796 : }
797 :
798 : /* If there is no note, assume branches are not taken.
799 : This should be rare. */
800 : return 0;
801 : }
802 :
803 : /* Return the condition under which INSN will branch to TARGET. If TARGET
804 : is zero, return the condition under which INSN will return. If INSN is
805 : an unconditional branch, return const_true_rtx. If INSN isn't a simple
806 : type of jump, or it doesn't go to TARGET, return 0. */
807 :
808 : static rtx
809 0 : get_branch_condition (const rtx_insn *insn, rtx target)
810 : {
811 0 : rtx pat = PATTERN (insn);
812 0 : rtx src;
813 :
814 0 : if (condjump_in_parallel_p (insn))
815 0 : pat = XVECEXP (pat, 0, 0);
816 :
817 0 : if (ANY_RETURN_P (pat) && pat == target)
818 0 : return const_true_rtx;
819 :
820 0 : if (GET_CODE (pat) != SET || SET_DEST (pat) != pc_rtx)
821 : return 0;
822 :
823 0 : src = SET_SRC (pat);
824 0 : if (GET_CODE (src) == LABEL_REF && label_ref_label (src) == target)
825 0 : return const_true_rtx;
826 :
827 0 : else if (GET_CODE (src) == IF_THEN_ELSE
828 0 : && XEXP (src, 2) == pc_rtx
829 0 : && ((GET_CODE (XEXP (src, 1)) == LABEL_REF
830 0 : && label_ref_label (XEXP (src, 1)) == target)
831 0 : || (ANY_RETURN_P (XEXP (src, 1)) && XEXP (src, 1) == target)))
832 0 : return XEXP (src, 0);
833 :
834 0 : else if (GET_CODE (src) == IF_THEN_ELSE
835 0 : && XEXP (src, 1) == pc_rtx
836 0 : && ((GET_CODE (XEXP (src, 2)) == LABEL_REF
837 0 : && label_ref_label (XEXP (src, 2)) == target)
838 0 : || (ANY_RETURN_P (XEXP (src, 2)) && XEXP (src, 2) == target)))
839 : {
840 0 : enum rtx_code rev;
841 0 : rev = reversed_comparison_code (XEXP (src, 0), insn);
842 0 : if (rev != UNKNOWN)
843 0 : return gen_rtx_fmt_ee (rev, GET_MODE (XEXP (src, 0)),
844 : XEXP (XEXP (src, 0), 0),
845 : XEXP (XEXP (src, 0), 1));
846 : }
847 :
848 : return 0;
849 : }
850 :
851 : /* Return true if CONDITION is more strict than the condition of
852 : INSN, i.e., if INSN will always branch if CONDITION is true. */
853 :
854 : static bool
855 0 : condition_dominates_p (rtx condition, const rtx_insn *insn)
856 : {
857 0 : rtx other_condition = get_branch_condition (insn, JUMP_LABEL (insn));
858 0 : enum rtx_code code = GET_CODE (condition);
859 0 : enum rtx_code other_code;
860 :
861 0 : if (rtx_equal_p (condition, other_condition)
862 0 : || other_condition == const_true_rtx)
863 : return true;
864 :
865 0 : else if (condition == const_true_rtx || other_condition == 0)
866 : return false;
867 :
868 0 : other_code = GET_CODE (other_condition);
869 0 : if (GET_RTX_LENGTH (code) != 2 || GET_RTX_LENGTH (other_code) != 2
870 0 : || ! rtx_equal_p (XEXP (condition, 0), XEXP (other_condition, 0))
871 0 : || ! rtx_equal_p (XEXP (condition, 1), XEXP (other_condition, 1)))
872 : return false;
873 :
874 0 : return comparison_dominates_p (code, other_code);
875 : }
876 :
877 : /* Return true if redirecting JUMP to NEWLABEL does not invalidate
878 : any insns already in the delay slot of JUMP. */
879 :
880 : static bool
881 0 : redirect_with_delay_slots_safe_p (rtx_insn *jump, rtx newlabel, rtx seq)
882 : {
883 0 : int flags, i;
884 0 : rtx_sequence *pat = as_a <rtx_sequence *> (PATTERN (seq));
885 :
886 : /* Make sure all the delay slots of this jump would still
887 : be valid after threading the jump. If they are still
888 : valid, then return nonzero. */
889 :
890 0 : flags = get_jump_flags (jump, newlabel);
891 0 : for (i = 1; i < pat->len (); i++)
892 0 : if (! (
893 : #if ANNUL_IFFALSE_SLOTS
894 : (INSN_ANNULLED_BRANCH_P (jump)
895 : && INSN_FROM_TARGET_P (pat->insn (i)))
896 : ? eligible_for_annul_false (jump, i - 1, pat->insn (i), flags) :
897 : #endif
898 : #if ANNUL_IFTRUE_SLOTS
899 : (INSN_ANNULLED_BRANCH_P (jump)
900 : && ! INSN_FROM_TARGET_P (XVECEXP (pat, 0, i)))
901 : ? eligible_for_annul_true (jump, i - 1, pat->insn (i), flags) :
902 : #endif
903 0 : eligible_for_delay (jump, i - 1, pat->insn (i), flags)))
904 : break;
905 :
906 0 : return (i == pat->len ());
907 : }
908 :
909 : /* Return true if redirecting JUMP to NEWLABEL does not invalidate
910 : any insns we wish to place in the delay slot of JUMP. */
911 :
912 : static bool
913 0 : redirect_with_delay_list_safe_p (rtx_insn *jump, rtx newlabel,
914 : const vec<rtx_insn *> &delay_list)
915 : {
916 : /* Make sure all the insns in DELAY_LIST would still be
917 : valid after threading the jump. If they are still
918 : valid, then return true. */
919 :
920 0 : int flags = get_jump_flags (jump, newlabel);
921 0 : unsigned int delay_insns = delay_list.length ();
922 0 : unsigned int i = 0;
923 0 : for (; i < delay_insns; i++)
924 0 : if (! (
925 : #if ANNUL_IFFALSE_SLOTS
926 : (INSN_ANNULLED_BRANCH_P (jump)
927 : && INSN_FROM_TARGET_P (delay_list[i]))
928 : ? eligible_for_annul_false (jump, i, delay_list[i], flags) :
929 : #endif
930 : #if ANNUL_IFTRUE_SLOTS
931 : (INSN_ANNULLED_BRANCH_P (jump)
932 : && ! INSN_FROM_TARGET_P (delay_list[i]))
933 : ? eligible_for_annul_true (jump, i, delay_list[i], flags) :
934 : #endif
935 0 : eligible_for_delay (jump, i, delay_list[i], flags)))
936 : break;
937 :
938 0 : return i == delay_insns;
939 : }
940 :
941 : /* DELAY_LIST is a list of insns that have already been placed into delay
942 : slots. See if all of them have the same annulling status as ANNUL_TRUE_P.
943 : If not, return false; otherwise return true. */
944 :
945 : static bool
946 0 : check_annul_list_true_false (bool annul_true_p,
947 : const vec<rtx_insn *> &delay_list)
948 : {
949 0 : rtx_insn *trial;
950 0 : unsigned int i;
951 0 : FOR_EACH_VEC_ELT (delay_list, i, trial)
952 0 : if ((annul_true_p && INSN_FROM_TARGET_P (trial))
953 0 : || (!annul_true_p && !INSN_FROM_TARGET_P (trial)))
954 : return false;
955 :
956 : return true;
957 : }
958 :
959 : /* INSN branches to an insn whose pattern SEQ is a SEQUENCE. Given that
960 : the condition tested by INSN is CONDITION and the resources shown in
961 : OTHER_NEEDED are needed after INSN, see whether INSN can take all the insns
962 : from SEQ's delay list, in addition to whatever insns it may execute
963 : (in DELAY_LIST). SETS and NEEDED are denote resources already set and
964 : needed while searching for delay slot insns. Return the concatenated
965 : delay list if possible, otherwise, return 0.
966 :
967 : SLOTS_TO_FILL is the total number of slots required by INSN, and
968 : PSLOTS_FILLED points to the number filled so far (also the number of
969 : insns in DELAY_LIST). It is updated with the number that have been
970 : filled from the SEQUENCE, if any.
971 :
972 : PANNUL_P points to a nonzero value if we already know that we need
973 : to annul INSN. If this routine determines that annulling is needed,
974 : it may set that value to true.
975 :
976 : PNEW_THREAD points to a location that is to receive the place at which
977 : execution should continue. */
978 :
979 : static void
980 0 : steal_delay_list_from_target (rtx_insn *insn, rtx condition, rtx_sequence *seq,
981 : vec<rtx_insn *> *delay_list,
982 : struct resources *sets,
983 : struct resources *needed,
984 : struct resources *other_needed,
985 : int slots_to_fill, int *pslots_filled,
986 : bool *pannul_p, rtx *pnew_thread)
987 : {
988 0 : int slots_remaining = slots_to_fill - *pslots_filled;
989 0 : int total_slots_filled = *pslots_filled;
990 0 : auto_vec<rtx_insn *, 5> new_delay_list;
991 0 : bool must_annul = *pannul_p;
992 0 : bool used_annul = false;
993 0 : int i;
994 0 : struct resources cc_set;
995 0 : rtx_insn **redundant;
996 :
997 : /* We can't do anything if there are more delay slots in SEQ than we
998 : can handle, or if we don't know that it will be a taken branch.
999 : We know that it will be a taken branch if it is either an unconditional
1000 : branch or a conditional branch with a stricter branch condition.
1001 :
1002 : Also, exit if the branch has more than one set, since then it is computing
1003 : other results that can't be ignored, e.g. the HPPA mov&branch instruction.
1004 : ??? It may be possible to move other sets into INSN in addition to
1005 : moving the instructions in the delay slots.
1006 :
1007 : We cannot steal the delay list if one of the instructions in the
1008 : current delay_list modifies the condition codes and the jump in the
1009 : sequence is a conditional jump. We cannot do this because we cannot
1010 : change the direction of the jump because the condition codes
1011 : will effect the direction of the jump in the sequence. */
1012 :
1013 0 : CLEAR_RESOURCE (&cc_set);
1014 :
1015 : rtx_insn *trial;
1016 0 : FOR_EACH_VEC_ELT (*delay_list, i, trial)
1017 : {
1018 0 : mark_set_resources (trial, &cc_set, 0, MARK_SRC_DEST_CALL);
1019 0 : if (insn_references_resource_p (seq->insn (0), &cc_set, false))
1020 : return;
1021 : }
1022 :
1023 0 : if (XVECLEN (seq, 0) - 1 > slots_remaining
1024 0 : || ! condition_dominates_p (condition, seq->insn (0))
1025 0 : || ! single_set (seq->insn (0)))
1026 : return;
1027 :
1028 : /* On some targets, branches with delay slots can have a limited
1029 : displacement. Give the back end a chance to tell us we can't do
1030 : this. */
1031 0 : if (! targetm.can_follow_jump (insn, seq->insn (0)))
1032 : return;
1033 :
1034 0 : redundant = XALLOCAVEC (rtx_insn *, XVECLEN (seq, 0));
1035 0 : for (i = 1; i < seq->len (); i++)
1036 : {
1037 0 : rtx_insn *trial = seq->insn (i);
1038 0 : int flags;
1039 :
1040 0 : if (insn_references_resource_p (trial, sets, false)
1041 0 : || insn_sets_resource_p (trial, needed, false)
1042 0 : || insn_sets_resource_p (trial, sets, false)
1043 : /* If TRIAL is from the fallthrough code of an annulled branch insn
1044 : in SEQ, we cannot use it. */
1045 0 : || (INSN_ANNULLED_BRANCH_P (seq->insn (0))
1046 0 : && ! INSN_FROM_TARGET_P (trial)))
1047 : return;
1048 :
1049 : /* If this insn was already done (usually in a previous delay slot),
1050 : pretend we put it in our delay slot. */
1051 0 : redundant[i] = redundant_insn (trial, insn, new_delay_list);
1052 0 : if (redundant[i])
1053 0 : continue;
1054 :
1055 : /* We will end up re-vectoring this branch, so compute flags
1056 : based on jumping to the new label. */
1057 0 : flags = get_jump_flags (insn, JUMP_LABEL (seq->insn (0)));
1058 :
1059 0 : if (! must_annul
1060 0 : && ((condition == const_true_rtx
1061 0 : || (! insn_sets_resource_p (trial, other_needed, false)
1062 0 : && ! may_trap_or_fault_p (PATTERN (trial)))))
1063 0 : ? eligible_for_delay (insn, total_slots_filled, trial, flags)
1064 0 : : (must_annul || (delay_list->is_empty () && new_delay_list.is_empty ()))
1065 0 : && (must_annul = true,
1066 0 : check_annul_list_true_false (false, *delay_list)
1067 0 : && check_annul_list_true_false (false, new_delay_list)
1068 0 : && eligible_for_annul_false (insn, total_slots_filled,
1069 : trial, flags)))
1070 : {
1071 0 : if (must_annul)
1072 : {
1073 : /* Frame related instructions cannot go into annulled delay
1074 : slots, it messes up the dwarf info. */
1075 0 : if (RTX_FRAME_RELATED_P (trial))
1076 : return;
1077 : used_annul = true;
1078 : }
1079 0 : rtx_insn *temp = copy_delay_slot_insn (trial);
1080 0 : INSN_FROM_TARGET_P (temp) = 1;
1081 0 : add_to_delay_list (temp, &new_delay_list);
1082 0 : total_slots_filled++;
1083 :
1084 0 : if (--slots_remaining == 0)
1085 : break;
1086 : }
1087 : else
1088 : return;
1089 : }
1090 :
1091 : /* Record the effect of the instructions that were redundant and which
1092 : we therefore decided not to copy. */
1093 0 : for (i = 1; i < seq->len (); i++)
1094 0 : if (redundant[i])
1095 : {
1096 0 : fix_reg_dead_note (redundant[i], insn);
1097 0 : update_block (seq->insn (i), insn);
1098 : }
1099 :
1100 : /* Show the place to which we will be branching. */
1101 0 : *pnew_thread = first_active_target_insn (JUMP_LABEL (seq->insn (0)));
1102 :
1103 : /* Add any new insns to the delay list and update the count of the
1104 : number of slots filled. */
1105 0 : *pslots_filled = total_slots_filled;
1106 0 : if (used_annul)
1107 0 : *pannul_p = true;
1108 :
1109 0 : rtx_insn *temp;
1110 0 : FOR_EACH_VEC_ELT (new_delay_list, i, temp)
1111 0 : add_to_delay_list (temp, delay_list);
1112 0 : }
1113 :
1114 : /* Similar to steal_delay_list_from_target except that SEQ is on the
1115 : fallthrough path of INSN. Here we only do something if the delay insn
1116 : of SEQ is an unconditional branch. In that case we steal its delay slot
1117 : for INSN since unconditional branches are much easier to fill. */
1118 :
1119 : static void
1120 0 : steal_delay_list_from_fallthrough (rtx_insn *insn, rtx condition,
1121 : rtx_sequence *seq,
1122 : vec<rtx_insn *> *delay_list,
1123 : struct resources *sets,
1124 : struct resources *needed,
1125 : struct resources *other_needed,
1126 : int slots_to_fill, int *pslots_filled,
1127 : bool *pannul_p)
1128 : {
1129 0 : int i;
1130 0 : int flags;
1131 0 : bool must_annul = *pannul_p;
1132 0 : bool used_annul = false;
1133 :
1134 0 : flags = get_jump_flags (insn, JUMP_LABEL (insn));
1135 :
1136 : /* We can't do anything if SEQ's delay insn isn't an
1137 : unconditional branch. */
1138 :
1139 0 : if (! simplejump_or_return_p (seq->insn (0)))
1140 : return;
1141 :
1142 0 : for (i = 1; i < seq->len (); i++)
1143 : {
1144 0 : rtx_insn *trial = seq->insn (i);
1145 0 : rtx_insn *prior_insn;
1146 :
1147 0 : if (insn_references_resource_p (trial, sets, false)
1148 0 : || insn_sets_resource_p (trial, needed, false)
1149 0 : || insn_sets_resource_p (trial, sets, false))
1150 : break;
1151 :
1152 : /* If this insn was already done, we don't need it. */
1153 0 : if ((prior_insn = redundant_insn (trial, insn, *delay_list)))
1154 : {
1155 0 : fix_reg_dead_note (prior_insn, insn);
1156 0 : update_block (trial, insn);
1157 0 : delete_from_delay_slot (trial);
1158 0 : continue;
1159 : }
1160 :
1161 0 : if (! must_annul
1162 0 : && ((condition == const_true_rtx
1163 0 : || (! insn_sets_resource_p (trial, other_needed, false)
1164 0 : && ! may_trap_or_fault_p (PATTERN (trial)))))
1165 0 : ? eligible_for_delay (insn, *pslots_filled, trial, flags)
1166 0 : : (must_annul || delay_list->is_empty ()) && (must_annul = true,
1167 0 : check_annul_list_true_false (true, *delay_list)
1168 0 : && eligible_for_annul_true (insn, *pslots_filled, trial, flags)))
1169 : {
1170 0 : if (must_annul)
1171 0 : used_annul = true;
1172 0 : delete_from_delay_slot (trial);
1173 0 : add_to_delay_list (trial, delay_list);
1174 :
1175 0 : if (++(*pslots_filled) == slots_to_fill)
1176 : break;
1177 : }
1178 : else
1179 : break;
1180 : }
1181 :
1182 0 : if (used_annul)
1183 0 : *pannul_p = true;
1184 : }
1185 :
1186 : /* Try merging insns starting at THREAD which match exactly the insns in
1187 : INSN's delay list.
1188 :
1189 : If all insns were matched and the insn was previously annulling, the
1190 : annul bit will be cleared.
1191 :
1192 : For each insn that is merged, if the branch is or will be non-annulling,
1193 : we delete the merged insn. */
1194 :
1195 : static void
1196 0 : try_merge_delay_insns (rtx_insn *insn, rtx_insn *thread)
1197 : {
1198 0 : rtx_insn *trial, *next_trial;
1199 0 : rtx_insn *delay_insn = as_a <rtx_insn *> (XVECEXP (PATTERN (insn), 0, 0));
1200 0 : bool annul_p = JUMP_P (delay_insn) && INSN_ANNULLED_BRANCH_P (delay_insn);
1201 0 : int slot_number = 1;
1202 0 : int num_slots = XVECLEN (PATTERN (insn), 0);
1203 0 : rtx next_to_match = XVECEXP (PATTERN (insn), 0, slot_number);
1204 0 : struct resources set, needed, modified;
1205 0 : auto_vec<std::pair<rtx_insn *, bool>, 10> merged_insns;
1206 0 : int flags;
1207 :
1208 0 : flags = get_jump_flags (delay_insn, JUMP_LABEL (delay_insn));
1209 :
1210 0 : CLEAR_RESOURCE (&needed);
1211 0 : CLEAR_RESOURCE (&set);
1212 :
1213 : /* If this is not an annulling branch, take into account anything needed in
1214 : INSN's delay slot. This prevents two increments from being incorrectly
1215 : folded into one. If we are annulling, this would be the correct
1216 : thing to do. (The alternative, looking at things set in NEXT_TO_MATCH
1217 : will essentially disable this optimization. This method is somewhat of
1218 : a kludge, but I don't see a better way.) */
1219 0 : if (! annul_p)
1220 0 : for (int i = 1; i < num_slots; i++)
1221 0 : if (XVECEXP (PATTERN (insn), 0, i))
1222 0 : mark_referenced_resources (XVECEXP (PATTERN (insn), 0, i), &needed,
1223 : true);
1224 :
1225 0 : for (trial = thread; !stop_search_p (trial, true); trial = next_trial)
1226 : {
1227 0 : rtx pat = PATTERN (trial);
1228 0 : rtx oldtrial = trial;
1229 :
1230 0 : next_trial = next_nonnote_insn (trial);
1231 :
1232 : /* TRIAL must be a CALL_INSN or INSN. Skip USE and CLOBBER. */
1233 0 : if (NONJUMP_INSN_P (trial)
1234 0 : && (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER))
1235 0 : continue;
1236 :
1237 0 : if (GET_CODE (next_to_match) == GET_CODE (trial)
1238 0 : && ! insn_references_resource_p (trial, &set, true)
1239 0 : && ! insn_sets_resource_p (trial, &set, true)
1240 0 : && ! insn_sets_resource_p (trial, &needed, true)
1241 0 : && (trial = try_split (pat, trial, 0)) != 0
1242 : /* Update next_trial, in case try_split succeeded. */
1243 0 : && (next_trial = next_nonnote_insn (trial))
1244 : /* Likewise THREAD. */
1245 0 : && (thread = oldtrial == thread ? trial : thread)
1246 0 : && rtx_equal_p (PATTERN (next_to_match), PATTERN (trial))
1247 : /* Have to test this condition if annul condition is different
1248 : from (and less restrictive than) non-annulling one. */
1249 0 : && eligible_for_delay (delay_insn, slot_number - 1, trial, flags))
1250 : {
1251 :
1252 0 : if (! annul_p)
1253 : {
1254 0 : update_block (trial, thread);
1255 0 : if (trial == thread)
1256 0 : thread = next_active_insn (thread);
1257 :
1258 0 : delete_related_insns (trial);
1259 0 : INSN_FROM_TARGET_P (next_to_match) = 0;
1260 : }
1261 : else
1262 0 : merged_insns.safe_push (std::pair<rtx_insn *, bool> (trial, false));
1263 :
1264 0 : if (++slot_number == num_slots)
1265 : break;
1266 :
1267 0 : next_to_match = XVECEXP (PATTERN (insn), 0, slot_number);
1268 : }
1269 :
1270 0 : mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL);
1271 0 : mark_referenced_resources (trial, &needed, true);
1272 : }
1273 :
1274 : /* See if we stopped on a filled insn. If we did, try to see if its
1275 : delay slots match. */
1276 0 : if (slot_number != num_slots
1277 0 : && trial && NONJUMP_INSN_P (trial)
1278 0 : && GET_CODE (PATTERN (trial)) == SEQUENCE
1279 0 : && !(JUMP_P (XVECEXP (PATTERN (trial), 0, 0))
1280 0 : && INSN_ANNULLED_BRANCH_P (XVECEXP (PATTERN (trial), 0, 0))))
1281 : {
1282 0 : rtx_sequence *pat = as_a <rtx_sequence *> (PATTERN (trial));
1283 0 : rtx filled_insn = XVECEXP (pat, 0, 0);
1284 :
1285 : /* Account for resources set/needed by the filled insn. */
1286 0 : mark_set_resources (filled_insn, &set, 0, MARK_SRC_DEST_CALL);
1287 0 : mark_referenced_resources (filled_insn, &needed, true);
1288 :
1289 0 : for (int i = 1; i < pat->len (); i++)
1290 : {
1291 0 : rtx_insn *dtrial = pat->insn (i);
1292 :
1293 0 : CLEAR_RESOURCE (&modified);
1294 : /* Account for resources set by the insn following NEXT_TO_MATCH
1295 : inside INSN's delay list. */
1296 0 : for (int j = 1; slot_number + j < num_slots; j++)
1297 0 : mark_set_resources (XVECEXP (PATTERN (insn), 0, slot_number + j),
1298 : &modified, 0, MARK_SRC_DEST_CALL);
1299 : /* Account for resources set by the insn before DTRIAL and inside
1300 : TRIAL's delay list. */
1301 0 : for (int j = 1; j < i; j++)
1302 0 : mark_set_resources (XVECEXP (pat, 0, j),
1303 : &modified, 0, MARK_SRC_DEST_CALL);
1304 0 : if (! insn_references_resource_p (dtrial, &set, true)
1305 0 : && ! insn_sets_resource_p (dtrial, &set, true)
1306 0 : && ! insn_sets_resource_p (dtrial, &needed, true)
1307 0 : && rtx_equal_p (PATTERN (next_to_match), PATTERN (dtrial))
1308 : /* Check that DTRIAL and NEXT_TO_MATCH does not reference a
1309 : resource modified between them (only dtrial is checked because
1310 : next_to_match and dtrial shall to be equal in order to hit
1311 : this line) */
1312 0 : && ! insn_references_resource_p (dtrial, &modified, true)
1313 0 : && eligible_for_delay (delay_insn, slot_number - 1, dtrial, flags))
1314 : {
1315 0 : if (! annul_p)
1316 : {
1317 0 : rtx_insn *new_rtx;
1318 :
1319 0 : update_block (dtrial, thread);
1320 0 : new_rtx = delete_from_delay_slot (dtrial);
1321 0 : if (thread->deleted ())
1322 0 : thread = new_rtx;
1323 0 : INSN_FROM_TARGET_P (next_to_match) = 0;
1324 : }
1325 : else
1326 0 : merged_insns.safe_push (std::pair<rtx_insn *, bool> (dtrial,
1327 0 : true));
1328 :
1329 0 : if (++slot_number == num_slots)
1330 : break;
1331 :
1332 0 : next_to_match = XVECEXP (PATTERN (insn), 0, slot_number);
1333 : }
1334 : else
1335 : {
1336 : /* Keep track of the set/referenced resources for the delay
1337 : slots of any trial insns we encounter. */
1338 0 : mark_set_resources (dtrial, &set, 0, MARK_SRC_DEST_CALL);
1339 0 : mark_referenced_resources (dtrial, &needed, true);
1340 : }
1341 : }
1342 : }
1343 :
1344 : /* If all insns in the delay slot have been matched and we were previously
1345 : annulling the branch, we need not any more. In that case delete all the
1346 : merged insns. Also clear the INSN_FROM_TARGET_P bit of each insn in
1347 : the delay list so that we know that it isn't only being used at the
1348 : target. */
1349 0 : if (slot_number == num_slots && annul_p)
1350 : {
1351 0 : unsigned int len = merged_insns.length ();
1352 0 : for (unsigned int i = len - 1; i < len; i--)
1353 0 : if (merged_insns[i].second)
1354 : {
1355 0 : update_block (merged_insns[i].first, thread);
1356 0 : rtx_insn *new_rtx = delete_from_delay_slot (merged_insns[i].first);
1357 0 : if (thread->deleted ())
1358 0 : thread = new_rtx;
1359 : }
1360 : else
1361 : {
1362 0 : update_block (merged_insns[i].first, thread);
1363 0 : delete_related_insns (merged_insns[i].first);
1364 : }
1365 :
1366 0 : INSN_ANNULLED_BRANCH_P (delay_insn) = 0;
1367 :
1368 0 : for (int i = 0; i < XVECLEN (PATTERN (insn), 0); i++)
1369 0 : INSN_FROM_TARGET_P (XVECEXP (PATTERN (insn), 0, i)) = 0;
1370 : }
1371 0 : }
1372 :
1373 : /* See if INSN is redundant with an insn in front of TARGET. Often this
1374 : is called when INSN is a candidate for a delay slot of TARGET.
1375 : DELAY_LIST are insns that will be placed in delay slots of TARGET in front
1376 : of INSN. Often INSN will be redundant with an insn in a delay slot of
1377 : some previous insn. This happens when we have a series of branches to the
1378 : same label; in that case the first insn at the target might want to go
1379 : into each of the delay slots.
1380 :
1381 : If we are not careful, this routine can take up a significant fraction
1382 : of the total compilation time (4%), but only wins rarely. Hence we
1383 : speed this routine up by making two passes. The first pass goes back
1384 : until it hits a label and sees if it finds an insn with an identical
1385 : pattern. Only in this (relatively rare) event does it check for
1386 : data conflicts.
1387 :
1388 : We do not split insns we encounter. This could cause us not to find a
1389 : redundant insn, but the cost of splitting seems greater than the possible
1390 : gain in rare cases. */
1391 :
1392 : static rtx_insn *
1393 0 : redundant_insn (rtx insn, rtx_insn *target, const vec<rtx_insn *> &delay_list)
1394 : {
1395 0 : rtx target_main = target;
1396 0 : rtx ipat = PATTERN (insn);
1397 0 : rtx_insn *trial;
1398 0 : rtx pat;
1399 0 : struct resources needed, set;
1400 0 : int i;
1401 0 : unsigned insns_to_search;
1402 :
1403 : /* If INSN has any REG_UNUSED notes, it can't match anything since we
1404 : are allowed to not actually assign to such a register. */
1405 0 : if (find_reg_note (insn, REG_UNUSED, NULL_RTX) != 0)
1406 : return 0;
1407 :
1408 : /* Scan backwards looking for a match. */
1409 0 : for (trial = PREV_INSN (target),
1410 0 : insns_to_search = param_max_delay_slot_insn_search;
1411 0 : trial && insns_to_search > 0;
1412 0 : trial = PREV_INSN (trial))
1413 : {
1414 : /* (use (insn))s can come immediately after a barrier if the
1415 : label that used to precede them has been deleted as dead.
1416 : See delete_related_insns. */
1417 0 : if (LABEL_P (trial) || BARRIER_P (trial))
1418 : return 0;
1419 :
1420 0 : if (!NONDEBUG_INSN_P (trial))
1421 0 : continue;
1422 0 : --insns_to_search;
1423 :
1424 0 : pat = PATTERN (trial);
1425 0 : if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER)
1426 0 : continue;
1427 :
1428 0 : if (rtx_sequence *seq = dyn_cast <rtx_sequence *> (pat))
1429 : {
1430 : /* Stop for a CALL and its delay slots because it is difficult to
1431 : track its resource needs correctly. */
1432 0 : if (CALL_P (seq->element (0)))
1433 : return 0;
1434 :
1435 : /* Stop for an INSN or JUMP_INSN with delayed effects and its delay
1436 : slots because it is difficult to track its resource needs
1437 : correctly. */
1438 :
1439 0 : if (INSN_SETS_ARE_DELAYED (seq->insn (0)))
1440 : return 0;
1441 :
1442 0 : if (INSN_REFERENCES_ARE_DELAYED (seq->insn (0)))
1443 : return 0;
1444 :
1445 : /* See if any of the insns in the delay slot match, updating
1446 : resource requirements as we go. */
1447 0 : for (i = seq->len () - 1; i > 0; i--)
1448 0 : if (GET_CODE (seq->element (i)) == GET_CODE (insn)
1449 0 : && rtx_equal_p (PATTERN (seq->element (i)), ipat)
1450 0 : && ! find_reg_note (seq->element (i), REG_UNUSED, NULL_RTX))
1451 : break;
1452 :
1453 : /* If found a match, exit this loop early. */
1454 0 : if (i > 0)
1455 : break;
1456 : }
1457 :
1458 0 : else if (GET_CODE (trial) == GET_CODE (insn) && rtx_equal_p (pat, ipat)
1459 0 : && ! find_reg_note (trial, REG_UNUSED, NULL_RTX))
1460 : break;
1461 : }
1462 :
1463 : /* If we didn't find an insn that matches, return 0. */
1464 0 : if (trial == 0)
1465 : return 0;
1466 :
1467 : /* See what resources this insn sets and needs. If they overlap, it
1468 : can't be redundant. */
1469 :
1470 0 : CLEAR_RESOURCE (&needed);
1471 0 : CLEAR_RESOURCE (&set);
1472 0 : mark_set_resources (insn, &set, 0, MARK_SRC_DEST_CALL);
1473 0 : mark_referenced_resources (insn, &needed, true);
1474 :
1475 : /* If TARGET is a SEQUENCE, get the main insn. */
1476 0 : if (NONJUMP_INSN_P (target) && GET_CODE (PATTERN (target)) == SEQUENCE)
1477 0 : target_main = XVECEXP (PATTERN (target), 0, 0);
1478 :
1479 0 : if (resource_conflicts_p (&needed, &set)
1480 : /* The insn requiring the delay may not set anything needed or set by
1481 : INSN. */
1482 0 : || insn_sets_resource_p (target_main, &needed, true)
1483 0 : || insn_sets_resource_p (target_main, &set, true))
1484 : return 0;
1485 :
1486 : /* Insns we pass may not set either NEEDED or SET, so merge them for
1487 : simpler tests. */
1488 0 : needed.memory |= set.memory;
1489 0 : needed.regs |= set.regs;
1490 :
1491 : /* This insn isn't redundant if it conflicts with an insn that either is
1492 : or will be in a delay slot of TARGET. */
1493 :
1494 : unsigned int j;
1495 : rtx_insn *temp;
1496 0 : FOR_EACH_VEC_ELT (delay_list, j, temp)
1497 0 : if (insn_sets_resource_p (temp, &needed, true))
1498 : return 0;
1499 :
1500 0 : if (NONJUMP_INSN_P (target) && GET_CODE (PATTERN (target)) == SEQUENCE)
1501 0 : for (i = 1; i < XVECLEN (PATTERN (target), 0); i++)
1502 0 : if (insn_sets_resource_p (XVECEXP (PATTERN (target), 0, i), &needed,
1503 : true))
1504 : return 0;
1505 :
1506 : /* Scan backwards until we reach a label or an insn that uses something
1507 : INSN sets or sets something insn uses or sets. */
1508 :
1509 0 : for (trial = PREV_INSN (target),
1510 0 : insns_to_search = param_max_delay_slot_insn_search;
1511 0 : trial && !LABEL_P (trial) && insns_to_search > 0;
1512 0 : trial = PREV_INSN (trial))
1513 : {
1514 0 : if (!NONDEBUG_INSN_P (trial))
1515 0 : continue;
1516 0 : --insns_to_search;
1517 :
1518 0 : pat = PATTERN (trial);
1519 0 : if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER)
1520 0 : continue;
1521 :
1522 0 : if (rtx_sequence *seq = dyn_cast <rtx_sequence *> (pat))
1523 : {
1524 0 : bool annul_p = false;
1525 0 : rtx_insn *control = seq->insn (0);
1526 :
1527 : /* If this is a CALL_INSN and its delay slots, it is hard to track
1528 : the resource needs properly, so give up. */
1529 0 : if (CALL_P (control))
1530 : return 0;
1531 :
1532 : /* If this is an INSN or JUMP_INSN with delayed effects, it
1533 : is hard to track the resource needs properly, so give up. */
1534 :
1535 0 : if (INSN_SETS_ARE_DELAYED (control))
1536 : return 0;
1537 :
1538 0 : if (INSN_REFERENCES_ARE_DELAYED (control))
1539 : return 0;
1540 :
1541 0 : if (JUMP_P (control))
1542 0 : annul_p = INSN_ANNULLED_BRANCH_P (control);
1543 :
1544 : /* See if any of the insns in the delay slot match, updating
1545 : resource requirements as we go. */
1546 0 : for (i = seq->len () - 1; i > 0; i--)
1547 : {
1548 0 : rtx_insn *candidate = seq->insn (i);
1549 :
1550 : /* If an insn will be annulled if the branch is false, it isn't
1551 : considered as a possible duplicate insn. */
1552 0 : if (rtx_equal_p (PATTERN (candidate), ipat)
1553 0 : && ! (annul_p && INSN_FROM_TARGET_P (candidate)))
1554 : {
1555 : /* Show that this insn will be used in the sequel. */
1556 0 : INSN_FROM_TARGET_P (candidate) = 0;
1557 0 : return candidate;
1558 : }
1559 :
1560 : /* Unless this is an annulled insn from the target of a branch,
1561 : we must stop if it sets anything needed or set by INSN. */
1562 0 : if ((!annul_p || !INSN_FROM_TARGET_P (candidate))
1563 0 : && insn_sets_resource_p (candidate, &needed, true))
1564 : return 0;
1565 : }
1566 :
1567 : /* If the insn requiring the delay slot conflicts with INSN, we
1568 : must stop. */
1569 0 : if (insn_sets_resource_p (control, &needed, true))
1570 : return 0;
1571 : }
1572 : else
1573 : {
1574 : /* See if TRIAL is the same as INSN. */
1575 0 : pat = PATTERN (trial);
1576 0 : if (rtx_equal_p (pat, ipat))
1577 : return trial;
1578 :
1579 : /* Can't go any further if TRIAL conflicts with INSN. */
1580 0 : if (insn_sets_resource_p (trial, &needed, true))
1581 : return 0;
1582 : }
1583 : }
1584 :
1585 : return 0;
1586 : }
1587 :
1588 : /* Return true if THREAD can only be executed in one way. If LABEL is nonzero,
1589 : it is the target of the branch insn being scanned. If ALLOW_FALLTHROUGH
1590 : is true, we are allowed to fall into this thread; otherwise, we are not.
1591 :
1592 : If LABEL is used more than one or we pass a label other than LABEL before
1593 : finding an active insn, we do not own this thread. */
1594 :
1595 : static bool
1596 0 : own_thread_p (rtx thread, rtx label, bool allow_fallthrough)
1597 : {
1598 0 : rtx_insn *active_insn;
1599 0 : rtx_insn *insn;
1600 :
1601 : /* We don't own the function end. */
1602 0 : if (thread == 0 || ANY_RETURN_P (thread))
1603 : return false;
1604 :
1605 : /* We have a non-NULL insn. */
1606 0 : rtx_insn *thread_insn = as_a <rtx_insn *> (thread);
1607 :
1608 : /* Get the first active insn, or THREAD_INSN, if it is an active insn. */
1609 0 : active_insn = next_active_insn (PREV_INSN (thread_insn));
1610 :
1611 0 : for (insn = thread_insn; insn != active_insn; insn = NEXT_INSN (insn))
1612 0 : if (LABEL_P (insn)
1613 0 : && (insn != label || LABEL_NUSES (insn) != 1))
1614 : return false;
1615 :
1616 0 : if (allow_fallthrough)
1617 : return true;
1618 :
1619 : /* Ensure that we reach a BARRIER before any insn or label. */
1620 0 : for (insn = prev_nonnote_insn (thread_insn);
1621 0 : insn == 0 || !BARRIER_P (insn);
1622 0 : insn = prev_nonnote_insn (insn))
1623 0 : if (insn == 0
1624 0 : || LABEL_P (insn)
1625 0 : || (NONJUMP_INSN_P (insn)
1626 0 : && GET_CODE (PATTERN (insn)) != USE
1627 0 : && GET_CODE (PATTERN (insn)) != CLOBBER))
1628 : return false;
1629 :
1630 : return true;
1631 : }
1632 :
1633 : /* Called when INSN is being moved from a location near the target of a jump.
1634 : We leave a marker of the form (use (INSN)) immediately in front of WHERE
1635 : for mark_target_live_regs. These markers will be deleted at the end.
1636 :
1637 : We used to try to update the live status of registers if WHERE is at
1638 : the start of a basic block, but that can't work since we may remove a
1639 : BARRIER in relax_delay_slots. */
1640 :
1641 : static void
1642 0 : update_block (rtx_insn *insn, rtx_insn *where)
1643 : {
1644 0 : emit_insn_before (gen_rtx_USE (VOIDmode, insn), where);
1645 :
1646 : /* INSN might be making a value live in a block where it didn't use to
1647 : be. So recompute liveness information for this block. */
1648 0 : incr_ticks_for_insn (insn);
1649 0 : }
1650 :
1651 : /* Similar to REDIRECT_JUMP except that we update the BB_TICKS entry for
1652 : the basic block containing the jump. */
1653 :
1654 : static bool
1655 0 : reorg_redirect_jump (rtx_jump_insn *jump, rtx nlabel)
1656 : {
1657 0 : incr_ticks_for_insn (jump);
1658 0 : return redirect_jump (jump, nlabel, 1);
1659 : }
1660 :
1661 : /* Called when INSN is being moved forward into a delay slot of DELAYED_INSN.
1662 : We check every instruction between INSN and DELAYED_INSN for REG_DEAD notes
1663 : that reference values used in INSN. If we find one, then we move the
1664 : REG_DEAD note to INSN.
1665 :
1666 : This is needed to handle the case where a later insn (after INSN) has a
1667 : REG_DEAD note for a register used by INSN, and this later insn subsequently
1668 : gets moved before a CODE_LABEL because it is a redundant insn. In this
1669 : case, mark_target_live_regs may be confused into thinking the register
1670 : is dead because it sees a REG_DEAD note immediately before a CODE_LABEL. */
1671 :
1672 : static void
1673 0 : update_reg_dead_notes (rtx_insn *insn, rtx_insn *delayed_insn)
1674 : {
1675 0 : rtx link, next;
1676 0 : rtx_insn *p;
1677 :
1678 0 : for (p = next_nonnote_insn (insn); p != delayed_insn;
1679 0 : p = next_nonnote_insn (p))
1680 0 : for (link = REG_NOTES (p); link; link = next)
1681 : {
1682 0 : next = XEXP (link, 1);
1683 :
1684 0 : if (REG_NOTE_KIND (link) != REG_DEAD
1685 0 : || !REG_P (XEXP (link, 0)))
1686 0 : continue;
1687 :
1688 0 : if (reg_referenced_p (XEXP (link, 0), PATTERN (insn)))
1689 : {
1690 : /* Move the REG_DEAD note from P to INSN. */
1691 0 : remove_note (p, link);
1692 0 : XEXP (link, 1) = REG_NOTES (insn);
1693 0 : REG_NOTES (insn) = link;
1694 : }
1695 : }
1696 0 : }
1697 :
1698 : /* Called when an insn redundant with start_insn is deleted. If there
1699 : is a REG_DEAD note for the target of start_insn between start_insn
1700 : and stop_insn, then the REG_DEAD note needs to be deleted since the
1701 : value no longer dies there.
1702 :
1703 : If the REG_DEAD note isn't deleted, then mark_target_live_regs may be
1704 : confused into thinking the register is dead. */
1705 :
1706 : static void
1707 0 : fix_reg_dead_note (rtx_insn *start_insn, rtx stop_insn)
1708 : {
1709 0 : rtx link, next;
1710 0 : rtx_insn *p;
1711 :
1712 0 : for (p = next_nonnote_insn (start_insn); p != stop_insn;
1713 0 : p = next_nonnote_insn (p))
1714 0 : for (link = REG_NOTES (p); link; link = next)
1715 : {
1716 0 : next = XEXP (link, 1);
1717 :
1718 0 : if (REG_NOTE_KIND (link) != REG_DEAD
1719 0 : || !REG_P (XEXP (link, 0)))
1720 0 : continue;
1721 :
1722 0 : if (reg_set_p (XEXP (link, 0), PATTERN (start_insn)))
1723 : {
1724 0 : remove_note (p, link);
1725 0 : return;
1726 : }
1727 : }
1728 : }
1729 :
1730 : /* Delete any REG_UNUSED notes that exist on INSN but not on OTHER_INSN.
1731 :
1732 : This handles the case of udivmodXi4 instructions which optimize their
1733 : output depending on whether any REG_UNUSED notes are present. We must
1734 : make sure that INSN calculates as many results as OTHER_INSN does. */
1735 :
1736 : static void
1737 0 : update_reg_unused_notes (rtx_insn *insn, rtx other_insn)
1738 : {
1739 0 : rtx link, next;
1740 :
1741 0 : for (link = REG_NOTES (insn); link; link = next)
1742 : {
1743 0 : next = XEXP (link, 1);
1744 :
1745 0 : if (REG_NOTE_KIND (link) != REG_UNUSED
1746 0 : || !REG_P (XEXP (link, 0)))
1747 0 : continue;
1748 :
1749 0 : if (!find_regno_note (other_insn, REG_UNUSED, REGNO (XEXP (link, 0))))
1750 0 : remove_note (insn, link);
1751 : }
1752 0 : }
1753 :
1754 : static vec <rtx> sibling_labels;
1755 :
1756 : /* Return the label before INSN, or put a new label there. If SIBLING is
1757 : non-zero, it is another label associated with the new label (if any),
1758 : typically the former target of the jump that will be redirected to
1759 : the new label. */
1760 :
1761 : static rtx_insn *
1762 0 : get_label_before (rtx_insn *insn, rtx sibling)
1763 : {
1764 0 : rtx_insn *label;
1765 :
1766 : /* Find an existing label at this point
1767 : or make a new one if there is none. */
1768 0 : label = prev_nonnote_insn (insn);
1769 :
1770 0 : if (label == 0 || !LABEL_P (label))
1771 : {
1772 0 : rtx_insn *prev = PREV_INSN (insn);
1773 :
1774 0 : label = gen_label_rtx ();
1775 0 : emit_label_after (label, prev);
1776 0 : LABEL_NUSES (label) = 0;
1777 0 : if (sibling)
1778 : {
1779 0 : sibling_labels.safe_push (label);
1780 0 : sibling_labels.safe_push (sibling);
1781 : }
1782 : }
1783 0 : return label;
1784 : }
1785 :
1786 : /* Scan a function looking for insns that need a delay slot and find insns to
1787 : put into the delay slot.
1788 :
1789 : NON_JUMPS_P is true if we are to only try to fill non-jump insns (such
1790 : as calls). We do these first since we don't want jump insns (that are
1791 : easier to fill) to get the only insns that could be used for non-jump insns.
1792 : When it is zero, only try to fill JUMP_INSNs.
1793 :
1794 : When slots are filled in this manner, the insns (including the
1795 : delay_insn) are put together in a SEQUENCE rtx. In this fashion,
1796 : it is possible to tell whether a delay slot has really been filled
1797 : or not. `final' knows how to deal with this, by communicating
1798 : through FINAL_SEQUENCE. */
1799 :
1800 : static void
1801 0 : fill_simple_delay_slots (bool non_jumps_p)
1802 : {
1803 0 : rtx_insn *insn, *trial, *next_trial;
1804 0 : rtx pat;
1805 0 : int i;
1806 0 : int num_unfilled_slots = unfilled_slots_next - unfilled_slots_base;
1807 0 : struct resources needed, set;
1808 0 : int slots_to_fill, slots_filled;
1809 0 : auto_vec<rtx_insn *, 5> delay_list;
1810 :
1811 0 : for (i = 0; i < num_unfilled_slots; i++)
1812 : {
1813 0 : int flags;
1814 : /* Get the next insn to fill. If it has already had any slots assigned,
1815 : we can't do anything with it. Maybe we'll improve this later. */
1816 :
1817 0 : insn = unfilled_slots_base[i];
1818 0 : if (insn == 0
1819 0 : || insn->deleted ()
1820 0 : || (NONJUMP_INSN_P (insn)
1821 0 : && GET_CODE (PATTERN (insn)) == SEQUENCE)
1822 0 : || (JUMP_P (insn) && non_jumps_p)
1823 0 : || (!JUMP_P (insn) && ! non_jumps_p))
1824 0 : continue;
1825 :
1826 : /* It may have been that this insn used to need delay slots, but
1827 : now doesn't; ignore in that case. This can happen, for example,
1828 : on the HP PA RISC, where the number of delay slots depends on
1829 : what insns are nearby. */
1830 0 : slots_to_fill = num_delay_slots (insn);
1831 :
1832 : /* Some machine description have defined instructions to have
1833 : delay slots only in certain circumstances which may depend on
1834 : nearby insns (which change due to reorg's actions).
1835 :
1836 : For example, the PA port normally has delay slots for unconditional
1837 : jumps.
1838 :
1839 : However, the PA port claims such jumps do not have a delay slot
1840 : if they are immediate successors of certain CALL_INSNs. This
1841 : allows the port to favor filling the delay slot of the call with
1842 : the unconditional jump. */
1843 0 : if (slots_to_fill == 0)
1844 0 : continue;
1845 :
1846 : /* This insn needs, or can use, some delay slots. SLOTS_TO_FILL
1847 : says how many. After initialization, first try optimizing
1848 :
1849 : call _foo call _foo
1850 : nop add %o7,.-L1,%o7
1851 : b,a L1
1852 : nop
1853 :
1854 : If this case applies, the delay slot of the call is filled with
1855 : the unconditional jump. This is done first to avoid having the
1856 : delay slot of the call filled in the backward scan. Also, since
1857 : the unconditional jump is likely to also have a delay slot, that
1858 : insn must exist when it is subsequently scanned.
1859 :
1860 : This is tried on each insn with delay slots as some machines
1861 : have insns which perform calls, but are not represented as
1862 : CALL_INSNs. */
1863 :
1864 0 : slots_filled = 0;
1865 0 : delay_list.truncate (0);
1866 :
1867 0 : if (JUMP_P (insn))
1868 0 : flags = get_jump_flags (insn, JUMP_LABEL (insn));
1869 : else
1870 0 : flags = get_jump_flags (insn, NULL_RTX);
1871 :
1872 0 : if ((trial = next_active_insn (insn))
1873 0 : && JUMP_P (trial)
1874 0 : && simplejump_p (trial)
1875 0 : && eligible_for_delay (insn, slots_filled, trial, flags)
1876 0 : && no_labels_between_p (insn, trial)
1877 0 : && ! can_throw_internal (trial))
1878 : {
1879 0 : rtx_insn **tmp;
1880 0 : slots_filled++;
1881 0 : add_to_delay_list (trial, &delay_list);
1882 :
1883 : /* TRIAL may have had its delay slot filled, then unfilled. When
1884 : the delay slot is unfilled, TRIAL is placed back on the unfilled
1885 : slots obstack. Unfortunately, it is placed on the end of the
1886 : obstack, not in its original location. Therefore, we must search
1887 : from entry i + 1 to the end of the unfilled slots obstack to
1888 : try and find TRIAL. */
1889 0 : tmp = &unfilled_slots_base[i + 1];
1890 0 : while (*tmp != trial && tmp != unfilled_slots_next)
1891 0 : tmp++;
1892 :
1893 : /* Remove the unconditional jump from consideration for delay slot
1894 : filling and unthread it. */
1895 0 : if (*tmp == trial)
1896 0 : *tmp = 0;
1897 0 : {
1898 0 : rtx_insn *next = NEXT_INSN (trial);
1899 0 : rtx_insn *prev = PREV_INSN (trial);
1900 0 : if (prev)
1901 0 : SET_NEXT_INSN (prev) = next;
1902 0 : if (next)
1903 0 : SET_PREV_INSN (next) = prev;
1904 : }
1905 : }
1906 :
1907 : /* Now, scan backwards from the insn to search for a potential
1908 : delay-slot candidate. Stop searching when a label or jump is hit.
1909 :
1910 : For each candidate, if it is to go into the delay slot (moved
1911 : forward in execution sequence), it must not need or set any resources
1912 : that were set by later insns and must not set any resources that
1913 : are needed for those insns.
1914 :
1915 : The delay slot insn itself sets resources unless it is a call
1916 : (in which case the called routine, not the insn itself, is doing
1917 : the setting). */
1918 :
1919 0 : if (slots_filled < slots_to_fill)
1920 : {
1921 : /* If the flags register is dead after the insn, then we want to be
1922 : able to accept a candidate that clobbers it. For this purpose,
1923 : we need to filter the flags register during life analysis, so
1924 : that it doesn't create RAW and WAW dependencies, while still
1925 : creating the necessary WAR dependencies. */
1926 0 : bool filter_flags
1927 : = (slots_to_fill == 1
1928 0 : && targetm.flags_regnum != INVALID_REGNUM
1929 0 : && find_regno_note (insn, REG_DEAD, targetm.flags_regnum));
1930 0 : struct resources fset;
1931 0 : CLEAR_RESOURCE (&needed);
1932 0 : CLEAR_RESOURCE (&set);
1933 0 : mark_set_resources (insn, &set, 0, MARK_SRC_DEST);
1934 0 : if (filter_flags)
1935 : {
1936 0 : CLEAR_RESOURCE (&fset);
1937 0 : mark_set_resources (insn, &fset, 0, MARK_SRC_DEST);
1938 : }
1939 0 : mark_referenced_resources (insn, &needed, false);
1940 :
1941 0 : for (trial = prev_nonnote_insn (insn); ! stop_search_p (trial, true);
1942 0 : trial = next_trial)
1943 : {
1944 0 : next_trial = prev_nonnote_insn (trial);
1945 :
1946 : /* This must be an INSN or CALL_INSN. */
1947 0 : pat = PATTERN (trial);
1948 :
1949 : /* Stand-alone USE and CLOBBER are just for flow. */
1950 0 : if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER)
1951 0 : continue;
1952 :
1953 : /* And DEBUG_INSNs never go into delay slots. */
1954 0 : if (GET_CODE (trial) == DEBUG_INSN)
1955 0 : continue;
1956 :
1957 : /* Check for resource conflict first, to avoid unnecessary
1958 : splitting. */
1959 0 : if (! insn_references_resource_p (trial, &set, true)
1960 0 : && ! insn_sets_resource_p (trial,
1961 : filter_flags ? &fset : &set,
1962 : true)
1963 0 : && ! insn_sets_resource_p (trial, &needed, true)
1964 0 : && ! can_throw_internal (trial))
1965 : {
1966 0 : trial = try_split (pat, trial, 1);
1967 0 : next_trial = prev_nonnote_insn (trial);
1968 0 : if (eligible_for_delay (insn, slots_filled, trial, flags))
1969 : {
1970 : /* In this case, we are searching backward, so if we
1971 : find insns to put on the delay list, we want
1972 : to put them at the head, rather than the
1973 : tail, of the list. */
1974 :
1975 0 : update_reg_dead_notes (trial, insn);
1976 0 : delay_list.safe_insert (0, trial);
1977 0 : update_block (trial, trial);
1978 0 : delete_related_insns (trial);
1979 0 : if (slots_to_fill == ++slots_filled)
1980 : break;
1981 0 : continue;
1982 : }
1983 : }
1984 :
1985 0 : mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL);
1986 0 : if (filter_flags)
1987 : {
1988 0 : mark_set_resources (trial, &fset, 0, MARK_SRC_DEST_CALL);
1989 : /* If the flags register is set, then it doesn't create RAW
1990 : dependencies any longer and it also doesn't create WAW
1991 : dependencies since it's dead after the original insn. */
1992 0 : if (TEST_HARD_REG_BIT (fset.regs, targetm.flags_regnum))
1993 : {
1994 0 : CLEAR_HARD_REG_BIT (needed.regs, targetm.flags_regnum);
1995 0 : CLEAR_HARD_REG_BIT (fset.regs, targetm.flags_regnum);
1996 : }
1997 : }
1998 0 : mark_referenced_resources (trial, &needed, true);
1999 : }
2000 : }
2001 :
2002 : /* If all needed slots haven't been filled, we come here. */
2003 :
2004 : /* Try to optimize case of jumping around a single insn. */
2005 0 : if ((ANNUL_IFTRUE_SLOTS || ANNUL_IFFALSE_SLOTS)
2006 : && slots_filled != slots_to_fill
2007 : && delay_list.is_empty ()
2008 : && JUMP_P (insn)
2009 : && (condjump_p (insn) || condjump_in_parallel_p (insn))
2010 : && !ANY_RETURN_P (JUMP_LABEL (insn)))
2011 : {
2012 : optimize_skip (as_a <rtx_jump_insn *> (insn), &delay_list);
2013 : if (!delay_list.is_empty ())
2014 : slots_filled += 1;
2015 : }
2016 :
2017 : /* Try to get insns from beyond the insn needing the delay slot.
2018 : These insns can neither set or reference resources set in insns being
2019 : skipped, cannot set resources in the insn being skipped, and, if this
2020 : is a CALL_INSN (or a CALL_INSN is passed), cannot trap (because the
2021 : call might not return).
2022 :
2023 : There used to be code which continued past the target label if
2024 : we saw all uses of the target label. This code did not work,
2025 : because it failed to account for some instructions which were
2026 : both annulled and marked as from the target. This can happen as a
2027 : result of optimize_skip. Since this code was redundant with
2028 : fill_eager_delay_slots anyways, it was just deleted. */
2029 :
2030 0 : if (slots_filled != slots_to_fill
2031 : /* If this instruction could throw an exception which is
2032 : caught in the same function, then it's not safe to fill
2033 : the delay slot with an instruction from beyond this
2034 : point. For example, consider:
2035 :
2036 : int i = 2;
2037 :
2038 : try {
2039 : f();
2040 : i = 3;
2041 : } catch (...) {}
2042 :
2043 : return i;
2044 :
2045 : Even though `i' is a local variable, we must be sure not
2046 : to put `i = 3' in the delay slot if `f' might throw an
2047 : exception.
2048 :
2049 : Presumably, we should also check to see if we could get
2050 : back to this function via `setjmp'. */
2051 0 : && ! can_throw_internal (insn)
2052 0 : && !JUMP_P (insn))
2053 : {
2054 0 : bool maybe_never = false;
2055 0 : rtx pat, trial_delay;
2056 :
2057 0 : CLEAR_RESOURCE (&needed);
2058 0 : CLEAR_RESOURCE (&set);
2059 0 : mark_set_resources (insn, &set, 0, MARK_SRC_DEST_CALL);
2060 0 : mark_referenced_resources (insn, &needed, true);
2061 :
2062 0 : if (CALL_P (insn))
2063 0 : maybe_never = true;
2064 :
2065 0 : for (trial = next_nonnote_insn (insn); !stop_search_p (trial, true);
2066 0 : trial = next_trial)
2067 : {
2068 0 : next_trial = next_nonnote_insn (trial);
2069 :
2070 : /* This must be an INSN or CALL_INSN. */
2071 0 : pat = PATTERN (trial);
2072 :
2073 : /* Stand-alone USE and CLOBBER are just for flow. */
2074 0 : if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER)
2075 0 : continue;
2076 :
2077 : /* And DEBUG_INSNs do not go in delay slots. */
2078 0 : if (GET_CODE (trial) == DEBUG_INSN)
2079 0 : continue;
2080 :
2081 : /* If this already has filled delay slots, get the insn needing
2082 : the delay slots. */
2083 0 : if (GET_CODE (pat) == SEQUENCE)
2084 0 : trial_delay = XVECEXP (pat, 0, 0);
2085 : else
2086 : trial_delay = trial;
2087 :
2088 : /* Stop our search when seeing a jump. */
2089 0 : if (JUMP_P (trial_delay))
2090 : break;
2091 :
2092 : /* See if we have a resource problem before we try to split. */
2093 0 : if (GET_CODE (pat) != SEQUENCE
2094 0 : && ! insn_references_resource_p (trial, &set, true)
2095 0 : && ! insn_sets_resource_p (trial, &set, true)
2096 0 : && ! insn_sets_resource_p (trial, &needed, true)
2097 0 : && ! (maybe_never && may_trap_or_fault_p (pat))
2098 0 : && (trial = try_split (pat, trial, 0))
2099 0 : && eligible_for_delay (insn, slots_filled, trial, flags)
2100 0 : && ! can_throw_internal (trial))
2101 : {
2102 0 : next_trial = next_nonnote_insn (trial);
2103 0 : add_to_delay_list (trial, &delay_list);
2104 :
2105 0 : delete_related_insns (trial);
2106 0 : if (slots_to_fill == ++slots_filled)
2107 : break;
2108 0 : continue;
2109 : }
2110 :
2111 0 : mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL);
2112 0 : mark_referenced_resources (trial, &needed, true);
2113 :
2114 : /* Ensure we don't put insns between the setting of cc and the
2115 : comparison by moving a setting of cc into an earlier delay
2116 : slot since these insns could clobber the condition code. */
2117 0 : set.cc = 1;
2118 :
2119 : /* If this is a call, we might not get here. */
2120 0 : if (CALL_P (trial_delay))
2121 0 : maybe_never = true;
2122 : }
2123 :
2124 : /* If there are slots left to fill and our search was stopped by an
2125 : unconditional branch, try the insn at the branch target. We can
2126 : redirect the branch if it works.
2127 :
2128 : Don't do this if the insn at the branch target is a branch. */
2129 0 : if (slots_to_fill != slots_filled
2130 0 : && trial
2131 0 : && jump_to_label_p (trial)
2132 0 : && simplejump_p (trial)
2133 0 : && (next_trial = next_active_insn (JUMP_LABEL_AS_INSN (trial))) != 0
2134 0 : && ! (NONJUMP_INSN_P (next_trial)
2135 0 : && GET_CODE (PATTERN (next_trial)) == SEQUENCE)
2136 0 : && !JUMP_P (next_trial)
2137 0 : && ! insn_references_resource_p (next_trial, &set, true)
2138 0 : && ! insn_sets_resource_p (next_trial, &set, true)
2139 0 : && ! insn_sets_resource_p (next_trial, &needed, true)
2140 0 : && ! (maybe_never && may_trap_or_fault_p (PATTERN (next_trial)))
2141 0 : && (next_trial = try_split (PATTERN (next_trial), next_trial, 0))
2142 0 : && eligible_for_delay (insn, slots_filled, next_trial, flags)
2143 0 : && ! can_throw_internal (trial))
2144 : {
2145 : /* See comment in relax_delay_slots about necessity of using
2146 : next_real_nondebug_insn here. */
2147 0 : rtx_insn *new_label = next_real_nondebug_insn (next_trial);
2148 :
2149 0 : if (new_label != 0)
2150 0 : new_label = get_label_before (new_label, JUMP_LABEL (trial));
2151 : else
2152 0 : new_label = find_end_label (simple_return_rtx);
2153 :
2154 0 : if (new_label)
2155 : {
2156 0 : add_to_delay_list (copy_delay_slot_insn (next_trial),
2157 : &delay_list);
2158 0 : slots_filled++;
2159 0 : reorg_redirect_jump (as_a <rtx_jump_insn *> (trial),
2160 : new_label);
2161 : }
2162 : }
2163 : }
2164 :
2165 : /* If this is an unconditional jump, then try to get insns from the
2166 : target of the jump. */
2167 0 : rtx_jump_insn *jump_insn;
2168 0 : if ((jump_insn = dyn_cast <rtx_jump_insn *> (insn))
2169 0 : && simplejump_p (jump_insn)
2170 0 : && slots_filled != slots_to_fill)
2171 0 : fill_slots_from_thread (jump_insn, const_true_rtx,
2172 0 : next_active_insn (JUMP_LABEL_AS_INSN (insn)),
2173 : NULL, 1, 1, own_thread_p (JUMP_LABEL (insn),
2174 : JUMP_LABEL (insn), false),
2175 : slots_to_fill, &slots_filled, &delay_list);
2176 :
2177 0 : if (!delay_list.is_empty ())
2178 0 : unfilled_slots_base[i]
2179 0 : = emit_delay_sequence (insn, delay_list, slots_filled);
2180 :
2181 0 : if (slots_to_fill == slots_filled)
2182 0 : unfilled_slots_base[i] = 0;
2183 :
2184 0 : note_delay_statistics (slots_filled, 0);
2185 : }
2186 0 : }
2187 :
2188 : /* Follow any unconditional jump at LABEL, for the purpose of redirecting JUMP;
2189 : return the ultimate label reached by any such chain of jumps.
2190 : Return a suitable return rtx if the chain ultimately leads to a
2191 : return instruction.
2192 : If LABEL is not followed by a jump, return LABEL.
2193 : If the chain loops or we can't find end, return LABEL,
2194 : since that tells caller to avoid changing the insn.
2195 : If the returned label is obtained by following a crossing jump,
2196 : set *CROSSING to true, otherwise set it to false. */
2197 :
2198 : static rtx
2199 0 : follow_jumps (rtx label, rtx_insn *jump, bool *crossing)
2200 : {
2201 0 : rtx_insn *insn;
2202 0 : rtx_insn *next;
2203 0 : int depth;
2204 :
2205 0 : *crossing = false;
2206 0 : if (ANY_RETURN_P (label))
2207 : return label;
2208 :
2209 0 : rtx_insn *value = as_a <rtx_insn *> (label);
2210 :
2211 0 : for (depth = 0;
2212 : (depth < 10
2213 0 : && (insn = next_active_insn (value)) != 0
2214 0 : && JUMP_P (insn)
2215 0 : && JUMP_LABEL (insn) != NULL_RTX
2216 0 : && ((any_uncondjump_p (insn) && onlyjump_p (insn))
2217 0 : || ANY_RETURN_P (PATTERN (insn)))
2218 0 : && (next = NEXT_INSN (insn))
2219 0 : && BARRIER_P (next));
2220 : depth++)
2221 : {
2222 0 : rtx this_label_or_return = JUMP_LABEL (insn);
2223 :
2224 : /* If we have found a cycle, make the insn jump to itself. */
2225 0 : if (this_label_or_return == label)
2226 : return label;
2227 :
2228 : /* Cannot follow returns and cannot look through tablejumps. */
2229 0 : if (ANY_RETURN_P (this_label_or_return))
2230 : return this_label_or_return;
2231 :
2232 0 : rtx_insn *this_label = as_a <rtx_insn *> (this_label_or_return);
2233 0 : if (NEXT_INSN (this_label)
2234 0 : && JUMP_TABLE_DATA_P (NEXT_INSN (this_label)))
2235 : break;
2236 :
2237 0 : if (!targetm.can_follow_jump (jump, insn))
2238 : break;
2239 0 : if (!*crossing)
2240 0 : *crossing = CROSSING_JUMP_P (jump);
2241 0 : value = this_label;
2242 : }
2243 0 : if (depth == 10)
2244 0 : return label;
2245 : return value;
2246 : }
2247 :
2248 : /* Try to find insns to place in delay slots.
2249 :
2250 : INSN is the jump needing SLOTS_TO_FILL delay slots. It tests CONDITION
2251 : or is an unconditional branch if CONDITION is const_true_rtx.
2252 : *PSLOTS_FILLED is updated with the number of slots that we have filled.
2253 :
2254 : THREAD is a flow-of-control, either the insns to be executed if the
2255 : branch is true or if the branch is false, THREAD_IF_TRUE says which.
2256 :
2257 : OPPOSITE_THREAD is the thread in the opposite direction. It is used
2258 : to see if any potential delay slot insns set things needed there.
2259 :
2260 : LIKELY is true if it is extremely likely that the branch will be
2261 : taken and THREAD_IF_TRUE is set. This is used for the branch at the
2262 : end of a loop back up to the top.
2263 :
2264 : OWN_THREAD is true if we are the only user of the thread, i.e. it is
2265 : the target of the jump when we are the only jump going there.
2266 :
2267 : If OWN_THREAD is false, it must be the "true" thread of a jump. In that
2268 : case, we can only take insns from the head of the thread for our delay
2269 : slot. We then adjust the jump to point after the insns we have taken. */
2270 :
2271 : static void
2272 0 : fill_slots_from_thread (rtx_jump_insn *insn, rtx condition,
2273 : rtx thread_or_return, rtx opposite_thread, bool likely,
2274 : bool thread_if_true, bool own_thread, int slots_to_fill,
2275 : int *pslots_filled, vec<rtx_insn *> *delay_list)
2276 : {
2277 0 : rtx new_thread;
2278 0 : struct resources opposite_needed, set, needed;
2279 0 : rtx_insn *trial;
2280 0 : bool lose = false;
2281 0 : bool must_annul = false;
2282 0 : int flags;
2283 :
2284 : /* Validate our arguments. */
2285 0 : gcc_assert (condition != const_true_rtx || thread_if_true);
2286 0 : gcc_assert (own_thread || thread_if_true);
2287 :
2288 0 : flags = get_jump_flags (insn, JUMP_LABEL (insn));
2289 :
2290 : /* If our thread is the end of subroutine, we can't get any delay
2291 : insns from that. */
2292 0 : if (thread_or_return == NULL_RTX || ANY_RETURN_P (thread_or_return))
2293 0 : return;
2294 :
2295 0 : rtx_insn *thread = as_a <rtx_insn *> (thread_or_return);
2296 :
2297 : /* If this is an unconditional branch, nothing is needed at the
2298 : opposite thread. Otherwise, compute what is needed there. */
2299 0 : if (condition == const_true_rtx)
2300 0 : CLEAR_RESOURCE (&opposite_needed);
2301 : else
2302 0 : mark_target_live_regs (get_insns (), opposite_thread, &opposite_needed);
2303 :
2304 : /* If the insn at THREAD can be split, do it here to avoid having to
2305 : update THREAD and NEW_THREAD if it is done in the loop below. Also
2306 : initialize NEW_THREAD. */
2307 :
2308 0 : new_thread = thread = try_split (PATTERN (thread), thread, 0);
2309 :
2310 : /* Scan insns at THREAD. We are looking for an insn that can be removed
2311 : from THREAD (it neither sets nor references resources that were set
2312 : ahead of it and it doesn't set anything needs by the insns ahead of
2313 : it) and that either can be placed in an annulling insn or aren't
2314 : needed at OPPOSITE_THREAD. */
2315 :
2316 0 : CLEAR_RESOURCE (&needed);
2317 0 : CLEAR_RESOURCE (&set);
2318 :
2319 : /* Handle the flags register specially, to be able to accept a
2320 : candidate that clobbers it. See also fill_simple_delay_slots. */
2321 0 : bool filter_flags
2322 : = (slots_to_fill == 1
2323 0 : && targetm.flags_regnum != INVALID_REGNUM
2324 0 : && find_regno_note (insn, REG_DEAD, targetm.flags_regnum));
2325 0 : struct resources fset;
2326 0 : struct resources flags_res;
2327 0 : if (filter_flags)
2328 : {
2329 0 : CLEAR_RESOURCE (&fset);
2330 0 : CLEAR_RESOURCE (&flags_res);
2331 0 : SET_HARD_REG_BIT (flags_res.regs, targetm.flags_regnum);
2332 : }
2333 :
2334 : /* If we do not own this thread, we must stop as soon as we find
2335 : something that we can't put in a delay slot, since all we can do
2336 : is branch into THREAD at a later point. Therefore, labels stop
2337 : the search if this is not the `true' thread. */
2338 :
2339 0 : for (trial = thread;
2340 0 : ! stop_search_p (trial, ! thread_if_true) && (! lose || own_thread);
2341 0 : trial = next_nonnote_insn (trial))
2342 : {
2343 0 : rtx pat, old_trial;
2344 :
2345 : /* If we have passed a label, we no longer own this thread. */
2346 0 : if (LABEL_P (trial))
2347 : {
2348 0 : own_thread = 0;
2349 0 : continue;
2350 : }
2351 :
2352 0 : pat = PATTERN (trial);
2353 0 : if (GET_CODE (pat) == USE || GET_CODE (pat) == CLOBBER)
2354 0 : continue;
2355 :
2356 0 : if (GET_CODE (trial) == DEBUG_INSN)
2357 0 : continue;
2358 :
2359 : /* If TRIAL conflicts with the insns ahead of it, we lose. */
2360 0 : if (! insn_references_resource_p (trial, &set, true)
2361 0 : && ! insn_sets_resource_p (trial, filter_flags ? &fset : &set, true)
2362 0 : && ! insn_sets_resource_p (trial, &needed, true)
2363 : /* If we're handling sets to the flags register specially, we
2364 : only allow an insn into a delay-slot, if it either:
2365 : - doesn't set the flags register,
2366 : - the "set" of the flags register isn't used (clobbered),
2367 : - insns between the delay-slot insn and the trial-insn
2368 : as accounted in "set", have not affected the flags register. */
2369 0 : && (! filter_flags
2370 0 : || ! insn_sets_resource_p (trial, &flags_res, true)
2371 0 : || find_regno_note (trial, REG_UNUSED, targetm.flags_regnum)
2372 0 : || ! TEST_HARD_REG_BIT (set.regs, targetm.flags_regnum))
2373 0 : && ! can_throw_internal (trial))
2374 : {
2375 0 : rtx_insn *prior_insn;
2376 :
2377 : /* If TRIAL is redundant with some insn before INSN, we don't
2378 : actually need to add it to the delay list; we can merely pretend
2379 : we did. */
2380 0 : if ((prior_insn = redundant_insn (trial, insn, *delay_list)))
2381 : {
2382 0 : fix_reg_dead_note (prior_insn, insn);
2383 0 : if (own_thread)
2384 : {
2385 0 : update_block (trial, thread);
2386 0 : if (trial == thread)
2387 : {
2388 0 : thread = next_active_insn (thread);
2389 0 : if (new_thread == trial)
2390 0 : new_thread = thread;
2391 : }
2392 :
2393 0 : delete_related_insns (trial);
2394 : }
2395 : else
2396 : {
2397 0 : update_reg_unused_notes (prior_insn, trial);
2398 0 : new_thread = next_active_insn (trial);
2399 : }
2400 :
2401 0 : continue;
2402 : }
2403 :
2404 : /* There are two ways we can win: If TRIAL doesn't set anything
2405 : needed at the opposite thread and can't trap, or if it can
2406 : go into an annulled delay slot. But we want neither to copy
2407 : nor to speculate frame-related insns. */
2408 0 : if (!must_annul
2409 0 : && ((condition == const_true_rtx
2410 0 : && (own_thread || !RTX_FRAME_RELATED_P (trial)))
2411 0 : || (! insn_sets_resource_p (trial, &opposite_needed, true)
2412 0 : && ! may_trap_or_fault_p (pat)
2413 0 : && ! RTX_FRAME_RELATED_P (trial))))
2414 : {
2415 0 : old_trial = trial;
2416 0 : trial = try_split (pat, trial, 0);
2417 0 : if (new_thread == old_trial)
2418 0 : new_thread = trial;
2419 0 : if (thread == old_trial)
2420 0 : thread = trial;
2421 0 : pat = PATTERN (trial);
2422 0 : if (eligible_for_delay (insn, *pslots_filled, trial, flags))
2423 0 : goto winner;
2424 : }
2425 0 : else if (!RTX_FRAME_RELATED_P (trial)
2426 : && ((ANNUL_IFTRUE_SLOTS && ! thread_if_true)
2427 : || (ANNUL_IFFALSE_SLOTS && thread_if_true)))
2428 : {
2429 : old_trial = trial;
2430 : trial = try_split (pat, trial, 0);
2431 : if (new_thread == old_trial)
2432 : new_thread = trial;
2433 : if (thread == old_trial)
2434 : thread = trial;
2435 : pat = PATTERN (trial);
2436 : if ((must_annul || delay_list->is_empty ()) && (thread_if_true
2437 : ? check_annul_list_true_false (false, *delay_list)
2438 : && eligible_for_annul_false (insn, *pslots_filled, trial, flags)
2439 : : check_annul_list_true_false (true, *delay_list)
2440 : && eligible_for_annul_true (insn, *pslots_filled, trial, flags)))
2441 : {
2442 : rtx_insn *temp;
2443 :
2444 : must_annul = true;
2445 0 : winner:
2446 :
2447 : /* If we own this thread, delete the insn. If this is the
2448 : destination of a branch, show that a basic block status
2449 : may have been updated. In any case, mark the new
2450 : starting point of this thread. */
2451 0 : if (own_thread)
2452 : {
2453 0 : rtx note;
2454 :
2455 0 : update_block (trial, thread);
2456 0 : if (trial == thread)
2457 : {
2458 0 : thread = next_active_insn (thread);
2459 0 : if (new_thread == trial)
2460 0 : new_thread = thread;
2461 : }
2462 :
2463 : /* We are moving this insn, not deleting it. We must
2464 : temporarily increment the use count on any referenced
2465 : label lest it be deleted by delete_related_insns. */
2466 0 : for (note = REG_NOTES (trial);
2467 0 : note != NULL_RTX;
2468 0 : note = XEXP (note, 1))
2469 0 : if (REG_NOTE_KIND (note) == REG_LABEL_OPERAND
2470 0 : || REG_NOTE_KIND (note) == REG_LABEL_TARGET)
2471 : {
2472 : /* REG_LABEL_OPERAND could be
2473 : NOTE_INSN_DELETED_LABEL too. */
2474 0 : if (LABEL_P (XEXP (note, 0)))
2475 0 : LABEL_NUSES (XEXP (note, 0))++;
2476 : else
2477 0 : gcc_assert (REG_NOTE_KIND (note)
2478 : == REG_LABEL_OPERAND);
2479 : }
2480 0 : if (jump_to_label_p (trial))
2481 0 : LABEL_NUSES (JUMP_LABEL (trial))++;
2482 :
2483 0 : delete_related_insns (trial);
2484 :
2485 0 : for (note = REG_NOTES (trial);
2486 0 : note != NULL_RTX;
2487 0 : note = XEXP (note, 1))
2488 0 : if (REG_NOTE_KIND (note) == REG_LABEL_OPERAND
2489 0 : || REG_NOTE_KIND (note) == REG_LABEL_TARGET)
2490 : {
2491 : /* REG_LABEL_OPERAND could be
2492 : NOTE_INSN_DELETED_LABEL too. */
2493 0 : if (LABEL_P (XEXP (note, 0)))
2494 0 : LABEL_NUSES (XEXP (note, 0))--;
2495 : else
2496 0 : gcc_assert (REG_NOTE_KIND (note)
2497 : == REG_LABEL_OPERAND);
2498 : }
2499 0 : if (jump_to_label_p (trial))
2500 0 : LABEL_NUSES (JUMP_LABEL (trial))--;
2501 : }
2502 : else
2503 0 : new_thread = next_active_insn (trial);
2504 :
2505 0 : temp = own_thread ? trial : copy_delay_slot_insn (trial);
2506 0 : if (thread_if_true)
2507 0 : INSN_FROM_TARGET_P (temp) = 1;
2508 :
2509 0 : add_to_delay_list (temp, delay_list);
2510 :
2511 0 : if (slots_to_fill == ++(*pslots_filled))
2512 : {
2513 : /* Even though we have filled all the slots, we
2514 : may be branching to a location that has a
2515 : redundant insn. Skip any if so. */
2516 0 : while (new_thread && ! own_thread
2517 0 : && ! insn_sets_resource_p (new_thread, &set, true)
2518 0 : && ! insn_sets_resource_p (new_thread, &needed,
2519 : true)
2520 0 : && ! insn_references_resource_p (new_thread,
2521 : &set, true)
2522 0 : && (prior_insn
2523 0 : = redundant_insn (new_thread, insn,
2524 : *delay_list)))
2525 : {
2526 : /* We know we do not own the thread, so no need
2527 : to call update_block and delete_insn. */
2528 0 : fix_reg_dead_note (prior_insn, insn);
2529 0 : update_reg_unused_notes (prior_insn, new_thread);
2530 0 : new_thread
2531 0 : = next_active_insn (as_a<rtx_insn *> (new_thread));
2532 : }
2533 : break;
2534 : }
2535 :
2536 0 : continue;
2537 0 : }
2538 : }
2539 : }
2540 :
2541 : /* This insn can't go into a delay slot. */
2542 0 : lose = true;
2543 0 : mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL);
2544 0 : mark_referenced_resources (trial, &needed, true);
2545 0 : if (filter_flags)
2546 : {
2547 0 : mark_set_resources (trial, &fset, 0, MARK_SRC_DEST_CALL);
2548 :
2549 : /* Groups of flags-register setters with users should not
2550 : affect opportunities to move flags-register-setting insns
2551 : (clobbers) into the delay-slot. */
2552 0 : CLEAR_HARD_REG_BIT (needed.regs, targetm.flags_regnum);
2553 0 : CLEAR_HARD_REG_BIT (fset.regs, targetm.flags_regnum);
2554 : }
2555 :
2556 : /* Ensure we don't put insns between the setting of cc and the comparison
2557 : by moving a setting of cc into an earlier delay slot since these insns
2558 : could clobber the condition code. */
2559 0 : set.cc = 1;
2560 :
2561 : /* If this insn is a register-register copy and the next insn has
2562 : a use of our destination, change it to use our source. That way,
2563 : it will become a candidate for our delay slot the next time
2564 : through this loop. This case occurs commonly in loops that
2565 : scan a list.
2566 :
2567 : We could check for more complex cases than those tested below,
2568 : but it doesn't seem worth it. It might also be a good idea to try
2569 : to swap the two insns. That might do better.
2570 :
2571 : We can't do this if the next insn modifies our destination, because
2572 : that would make the replacement into the insn invalid. We also can't
2573 : do this if it modifies our source, because it might be an earlyclobber
2574 : operand. This latter test also prevents updating the contents of
2575 : a PRE_INC. We also can't do this if there's overlap of source and
2576 : destination. Overlap may happen for larger-than-register-size modes. */
2577 :
2578 0 : if (NONJUMP_INSN_P (trial) && GET_CODE (pat) == SET
2579 0 : && REG_P (SET_SRC (pat))
2580 0 : && REG_P (SET_DEST (pat))
2581 0 : && !reg_overlap_mentioned_p (SET_DEST (pat), SET_SRC (pat)))
2582 : {
2583 0 : rtx_insn *next = next_nonnote_insn (trial);
2584 :
2585 0 : if (next && NONJUMP_INSN_P (next)
2586 0 : && GET_CODE (PATTERN (next)) != USE
2587 0 : && ! reg_set_p (SET_DEST (pat), next)
2588 0 : && ! reg_set_p (SET_SRC (pat), next)
2589 0 : && reg_referenced_p (SET_DEST (pat), PATTERN (next))
2590 0 : && ! modified_in_p (SET_DEST (pat), next))
2591 0 : validate_replace_rtx (SET_DEST (pat), SET_SRC (pat), next);
2592 : }
2593 : }
2594 :
2595 : /* If we stopped on a branch insn that has delay slots, see if we can
2596 : steal some of the insns in those slots. */
2597 0 : if (trial && NONJUMP_INSN_P (trial)
2598 0 : && GET_CODE (PATTERN (trial)) == SEQUENCE
2599 0 : && JUMP_P (XVECEXP (PATTERN (trial), 0, 0)))
2600 : {
2601 0 : rtx_sequence *sequence = as_a <rtx_sequence *> (PATTERN (trial));
2602 : /* If this is the `true' thread, we will want to follow the jump,
2603 : so we can only do this if we have taken everything up to here. */
2604 0 : if (thread_if_true && trial == new_thread)
2605 : {
2606 0 : steal_delay_list_from_target (insn, condition, sequence,
2607 : delay_list, &set, &needed,
2608 : &opposite_needed, slots_to_fill,
2609 : pslots_filled, &must_annul,
2610 : &new_thread);
2611 : /* If we owned the thread and are told that it branched
2612 : elsewhere, make sure we own the thread at the new location. */
2613 0 : if (own_thread && trial != new_thread)
2614 0 : own_thread = own_thread_p (new_thread, new_thread, false);
2615 : }
2616 : else if (! thread_if_true)
2617 0 : steal_delay_list_from_fallthrough (insn, condition, sequence,
2618 : delay_list, &set, &needed,
2619 : &opposite_needed, slots_to_fill,
2620 : pslots_filled, &must_annul);
2621 : }
2622 :
2623 : /* If we haven't found anything for this delay slot and it is very
2624 : likely that the branch will be taken, see if the insn at our target
2625 : increments or decrements a register with an increment that does not
2626 : depend on the destination register. If so, try to place the opposite
2627 : arithmetic insn after the jump insn and put the arithmetic insn in the
2628 : delay slot. If we can't do this, return. */
2629 0 : if (delay_list->is_empty () && likely
2630 0 : && new_thread
2631 0 : && !ANY_RETURN_P (new_thread)
2632 0 : && NONJUMP_INSN_P (new_thread)
2633 0 : && !RTX_FRAME_RELATED_P (new_thread)
2634 0 : && GET_CODE (PATTERN (new_thread)) != ASM_INPUT
2635 0 : && asm_noperands (PATTERN (new_thread)) < 0)
2636 : {
2637 0 : rtx dest;
2638 0 : rtx src;
2639 :
2640 : /* We know "new_thread" is an insn due to NONJUMP_INSN_P (new_thread)
2641 : above. */
2642 0 : trial = as_a <rtx_insn *> (new_thread);
2643 0 : rtx pat = PATTERN (trial);
2644 :
2645 0 : if (!NONJUMP_INSN_P (trial)
2646 0 : || GET_CODE (pat) != SET
2647 0 : || ! eligible_for_delay (insn, 0, trial, flags)
2648 0 : || can_throw_internal (trial))
2649 : return;
2650 :
2651 0 : dest = SET_DEST (pat), src = SET_SRC (pat);
2652 0 : if ((GET_CODE (src) == PLUS || GET_CODE (src) == MINUS)
2653 0 : && rtx_equal_p (XEXP (src, 0), dest)
2654 0 : && (!FLOAT_MODE_P (GET_MODE (src))
2655 0 : || flag_unsafe_math_optimizations)
2656 0 : && ! reg_overlap_mentioned_p (dest, XEXP (src, 1))
2657 0 : && ! side_effects_p (pat))
2658 : {
2659 0 : rtx other = XEXP (src, 1);
2660 0 : rtx new_arith;
2661 0 : rtx_insn *ninsn;
2662 :
2663 : /* If this is a constant adjustment, use the same code with
2664 : the negated constant. Otherwise, reverse the sense of the
2665 : arithmetic. */
2666 0 : if (CONST_INT_P (other))
2667 0 : new_arith = gen_rtx_fmt_ee (GET_CODE (src), GET_MODE (src), dest,
2668 : negate_rtx (GET_MODE (src), other));
2669 : else
2670 0 : new_arith = gen_rtx_fmt_ee (GET_CODE (src) == PLUS ? MINUS : PLUS,
2671 : GET_MODE (src), dest, other);
2672 :
2673 0 : ninsn = emit_insn_after (gen_rtx_SET (dest, new_arith), insn);
2674 :
2675 0 : if (recog_memoized (ninsn) < 0
2676 0 : || (extract_insn (ninsn),
2677 0 : !constrain_operands (1, get_preferred_alternatives (ninsn))))
2678 : {
2679 0 : delete_related_insns (ninsn);
2680 0 : return;
2681 : }
2682 :
2683 0 : if (own_thread)
2684 : {
2685 0 : update_block (trial, thread);
2686 0 : if (trial == thread)
2687 : {
2688 0 : thread = next_active_insn (thread);
2689 0 : if (new_thread == trial)
2690 0 : new_thread = thread;
2691 : }
2692 0 : delete_related_insns (trial);
2693 : }
2694 : else
2695 0 : new_thread = next_active_insn (trial);
2696 :
2697 0 : ninsn = own_thread ? trial : copy_delay_slot_insn (trial);
2698 0 : if (thread_if_true)
2699 0 : INSN_FROM_TARGET_P (ninsn) = 1;
2700 :
2701 0 : add_to_delay_list (ninsn, delay_list);
2702 0 : (*pslots_filled)++;
2703 : }
2704 : }
2705 :
2706 0 : if (!delay_list->is_empty () && must_annul)
2707 0 : INSN_ANNULLED_BRANCH_P (insn) = 1;
2708 :
2709 : /* If we are to branch into the middle of this thread, find an appropriate
2710 : label or make a new one if none, and redirect INSN to it. If we hit the
2711 : end of the function, use the end-of-function label. */
2712 0 : if (new_thread != thread)
2713 : {
2714 0 : rtx label;
2715 0 : bool crossing = false;
2716 :
2717 0 : gcc_assert (thread_if_true);
2718 :
2719 0 : if (new_thread
2720 0 : && simplejump_or_return_p (new_thread)
2721 0 : && redirect_with_delay_list_safe_p (insn,
2722 : JUMP_LABEL (new_thread),
2723 : *delay_list))
2724 0 : new_thread = follow_jumps (JUMP_LABEL (new_thread), insn, &crossing);
2725 :
2726 0 : if (!new_thread)
2727 0 : label = find_end_label (simple_return_rtx);
2728 0 : else if (ANY_RETURN_P (new_thread))
2729 0 : label = find_end_label (new_thread);
2730 0 : else if (LABEL_P (new_thread))
2731 : label = new_thread;
2732 : else
2733 0 : label = get_label_before (as_a <rtx_insn *> (new_thread),
2734 : JUMP_LABEL (insn));
2735 :
2736 0 : if (label)
2737 : {
2738 0 : reorg_redirect_jump (insn, label);
2739 0 : if (crossing)
2740 0 : CROSSING_JUMP_P (insn) = 1;
2741 : }
2742 : }
2743 : }
2744 :
2745 : /* Make another attempt to find insns to place in delay slots.
2746 :
2747 : We previously looked for insns located in front of the delay insn
2748 : and, for non-jump delay insns, located behind the delay insn.
2749 :
2750 : Here only try to schedule jump insns and try to move insns from either
2751 : the target or the following insns into the delay slot. If annulling is
2752 : supported, we will be likely to do this. Otherwise, we can do this only
2753 : if safe. */
2754 :
2755 : static void
2756 0 : fill_eager_delay_slots (void)
2757 : {
2758 0 : rtx_insn *insn;
2759 0 : int i;
2760 0 : int num_unfilled_slots = unfilled_slots_next - unfilled_slots_base;
2761 :
2762 0 : for (i = 0; i < num_unfilled_slots; i++)
2763 : {
2764 0 : rtx condition;
2765 0 : rtx target_label, insn_at_target;
2766 0 : rtx_insn *fallthrough_insn;
2767 0 : auto_vec<rtx_insn *, 5> delay_list;
2768 0 : rtx_jump_insn *jump_insn;
2769 0 : bool own_target;
2770 0 : bool own_fallthrough;
2771 0 : int prediction, slots_to_fill, slots_filled;
2772 :
2773 0 : insn = unfilled_slots_base[i];
2774 0 : if (insn == 0
2775 0 : || insn->deleted ()
2776 0 : || ! (jump_insn = dyn_cast <rtx_jump_insn *> (insn))
2777 0 : || ! (condjump_p (jump_insn) || condjump_in_parallel_p (jump_insn)))
2778 0 : continue;
2779 :
2780 0 : slots_to_fill = num_delay_slots (jump_insn);
2781 : /* Some machine description have defined instructions to have
2782 : delay slots only in certain circumstances which may depend on
2783 : nearby insns (which change due to reorg's actions).
2784 :
2785 : For example, the PA port normally has delay slots for unconditional
2786 : jumps.
2787 :
2788 : However, the PA port claims such jumps do not have a delay slot
2789 : if they are immediate successors of certain CALL_INSNs. This
2790 : allows the port to favor filling the delay slot of the call with
2791 : the unconditional jump. */
2792 0 : if (slots_to_fill == 0)
2793 0 : continue;
2794 :
2795 0 : slots_filled = 0;
2796 0 : target_label = JUMP_LABEL (jump_insn);
2797 0 : condition = get_branch_condition (jump_insn, target_label);
2798 :
2799 0 : if (condition == 0)
2800 0 : continue;
2801 :
2802 : /* Get the next active fallthrough and target insns and see if we own
2803 : them. Then see whether the branch is likely true. We don't need
2804 : to do a lot of this for unconditional branches. */
2805 :
2806 0 : insn_at_target = first_active_target_insn (target_label);
2807 0 : own_target = own_thread_p (target_label, target_label, false);
2808 :
2809 0 : if (condition == const_true_rtx)
2810 : {
2811 : own_fallthrough = false;
2812 : fallthrough_insn = 0;
2813 : prediction = 2;
2814 : }
2815 : else
2816 : {
2817 0 : fallthrough_insn = next_active_insn (jump_insn);
2818 0 : own_fallthrough = own_thread_p (NEXT_INSN (jump_insn),
2819 : NULL_RTX, true);
2820 0 : prediction = mostly_true_jump (jump_insn);
2821 : }
2822 :
2823 : /* If this insn is expected to branch, first try to get insns from our
2824 : target, then our fallthrough insns. If it is not expected to branch,
2825 : try the other order. */
2826 :
2827 0 : if (prediction > 0)
2828 : {
2829 0 : fill_slots_from_thread (jump_insn, condition, insn_at_target,
2830 : fallthrough_insn, prediction == 2, true,
2831 : own_target, slots_to_fill,
2832 : &slots_filled, &delay_list);
2833 :
2834 0 : if (delay_list.is_empty () && own_fallthrough)
2835 : {
2836 : /* Even though we didn't find anything for delay slots,
2837 : we might have found a redundant insn which we deleted
2838 : from the thread that was filled. So we have to recompute
2839 : the next insn at the target. */
2840 0 : target_label = JUMP_LABEL (jump_insn);
2841 0 : insn_at_target = first_active_target_insn (target_label);
2842 :
2843 0 : fill_slots_from_thread (jump_insn, condition, fallthrough_insn,
2844 : insn_at_target, false, false,
2845 : own_fallthrough, slots_to_fill,
2846 : &slots_filled, &delay_list);
2847 : }
2848 : }
2849 : else
2850 : {
2851 0 : if (own_fallthrough)
2852 0 : fill_slots_from_thread (jump_insn, condition, fallthrough_insn,
2853 : insn_at_target, false, false,
2854 : own_fallthrough, slots_to_fill,
2855 : &slots_filled, &delay_list);
2856 :
2857 0 : if (delay_list.is_empty ())
2858 0 : fill_slots_from_thread (jump_insn, condition, insn_at_target,
2859 0 : next_active_insn (insn), false, true,
2860 : own_target, slots_to_fill,
2861 : &slots_filled, &delay_list);
2862 : }
2863 :
2864 0 : if (!delay_list.is_empty ())
2865 0 : unfilled_slots_base[i]
2866 0 : = emit_delay_sequence (jump_insn, delay_list, slots_filled);
2867 :
2868 0 : if (slots_to_fill == slots_filled)
2869 0 : unfilled_slots_base[i] = 0;
2870 :
2871 0 : note_delay_statistics (slots_filled, 1);
2872 0 : }
2873 0 : }
2874 :
2875 : static void delete_computation (rtx_insn *insn);
2876 :
2877 : /* Recursively delete prior insns that compute the value (used only by INSN
2878 : which the caller is deleting) stored in the register mentioned by NOTE
2879 : which is a REG_DEAD note associated with INSN. */
2880 :
2881 : static void
2882 0 : delete_prior_computation (rtx note, rtx_insn *insn)
2883 : {
2884 0 : rtx_insn *our_prev;
2885 0 : rtx reg = XEXP (note, 0);
2886 :
2887 0 : for (our_prev = prev_nonnote_insn (insn);
2888 0 : our_prev && (NONJUMP_INSN_P (our_prev)
2889 0 : || CALL_P (our_prev));
2890 0 : our_prev = prev_nonnote_insn (our_prev))
2891 : {
2892 0 : rtx pat = PATTERN (our_prev);
2893 :
2894 : /* If we reach a CALL which is not calling a const function
2895 : or the callee pops the arguments, then give up. */
2896 0 : if (CALL_P (our_prev)
2897 0 : && (! RTL_CONST_CALL_P (our_prev)
2898 0 : || GET_CODE (pat) != SET || GET_CODE (SET_SRC (pat)) != CALL))
2899 : break;
2900 :
2901 : /* If we reach a SEQUENCE, it is too complex to try to
2902 : do anything with it, so give up. We can be run during
2903 : and after reorg, so SEQUENCE rtl can legitimately show
2904 : up here. */
2905 0 : if (GET_CODE (pat) == SEQUENCE)
2906 : break;
2907 :
2908 0 : if (GET_CODE (pat) == USE
2909 0 : && NONJUMP_INSN_P (XEXP (pat, 0)))
2910 : /* reorg creates USEs that look like this. We leave them
2911 : alone because reorg needs them for its own purposes. */
2912 : break;
2913 :
2914 0 : if (reg_set_p (reg, pat))
2915 : {
2916 0 : if (side_effects_p (pat) && !CALL_P (our_prev))
2917 : break;
2918 :
2919 0 : if (GET_CODE (pat) == PARALLEL)
2920 : {
2921 : /* If we find a SET of something else, we can't
2922 : delete the insn. */
2923 :
2924 : int i;
2925 :
2926 0 : for (i = 0; i < XVECLEN (pat, 0); i++)
2927 : {
2928 0 : rtx part = XVECEXP (pat, 0, i);
2929 :
2930 0 : if (GET_CODE (part) == SET
2931 0 : && SET_DEST (part) != reg)
2932 : break;
2933 : }
2934 :
2935 0 : if (i == XVECLEN (pat, 0))
2936 0 : delete_computation (our_prev);
2937 : }
2938 0 : else if (GET_CODE (pat) == SET
2939 0 : && REG_P (SET_DEST (pat)))
2940 : {
2941 0 : int dest_regno = REGNO (SET_DEST (pat));
2942 0 : int dest_endregno = END_REGNO (SET_DEST (pat));
2943 0 : int regno = REGNO (reg);
2944 0 : int endregno = END_REGNO (reg);
2945 :
2946 0 : if (dest_regno >= regno
2947 0 : && dest_endregno <= endregno)
2948 0 : delete_computation (our_prev);
2949 :
2950 : /* We may have a multi-word hard register and some, but not
2951 : all, of the words of the register are needed in subsequent
2952 : insns. Write REG_UNUSED notes for those parts that were not
2953 : needed. */
2954 0 : else if (dest_regno <= regno
2955 0 : && dest_endregno >= endregno)
2956 : {
2957 0 : int i;
2958 :
2959 0 : add_reg_note (our_prev, REG_UNUSED, reg);
2960 :
2961 0 : for (i = dest_regno; i < dest_endregno; i++)
2962 0 : if (! find_regno_note (our_prev, REG_UNUSED, i))
2963 : break;
2964 :
2965 0 : if (i == dest_endregno)
2966 0 : delete_computation (our_prev);
2967 : }
2968 : }
2969 :
2970 : break;
2971 : }
2972 :
2973 : /* If PAT references the register that dies here, it is an
2974 : additional use. Hence any prior SET isn't dead. However, this
2975 : insn becomes the new place for the REG_DEAD note. */
2976 0 : if (reg_overlap_mentioned_p (reg, pat))
2977 : {
2978 0 : XEXP (note, 1) = REG_NOTES (our_prev);
2979 0 : REG_NOTES (our_prev) = note;
2980 0 : break;
2981 : }
2982 : }
2983 0 : }
2984 :
2985 : /* Delete INSN and recursively delete insns that compute values used only
2986 : by INSN. This uses the REG_DEAD notes computed during flow analysis.
2987 :
2988 : Look at all our REG_DEAD notes. If a previous insn does nothing other
2989 : than set a register that dies in this insn, we can delete that insn
2990 : as well. */
2991 :
2992 : static void
2993 0 : delete_computation (rtx_insn *insn)
2994 : {
2995 0 : rtx note, next;
2996 :
2997 0 : for (note = REG_NOTES (insn); note; note = next)
2998 : {
2999 0 : next = XEXP (note, 1);
3000 :
3001 0 : if (REG_NOTE_KIND (note) != REG_DEAD
3002 : /* Verify that the REG_NOTE is legitimate. */
3003 0 : || !REG_P (XEXP (note, 0)))
3004 0 : continue;
3005 :
3006 0 : delete_prior_computation (note, insn);
3007 : }
3008 :
3009 0 : delete_related_insns (insn);
3010 0 : }
3011 :
3012 : /* If all INSN does is set the pc, delete it,
3013 : and delete the insn that set the condition codes for it
3014 : if that's what the previous thing was. */
3015 :
3016 : static void
3017 0 : delete_jump (rtx_insn *insn)
3018 : {
3019 0 : rtx set = single_set (insn);
3020 :
3021 0 : if (set && GET_CODE (SET_DEST (set)) == PC)
3022 0 : delete_computation (insn);
3023 0 : }
3024 :
3025 : static rtx_insn *
3026 0 : label_before_next_insn (rtx_insn *x, rtx scan_limit)
3027 : {
3028 0 : rtx_insn *insn = next_active_insn (x);
3029 0 : while (insn)
3030 : {
3031 0 : insn = PREV_INSN (insn);
3032 0 : if (insn == scan_limit || insn == NULL_RTX)
3033 : return NULL;
3034 0 : if (LABEL_P (insn))
3035 : break;
3036 : }
3037 : return insn;
3038 : }
3039 :
3040 : /* Return TRUE if there is a NOTE_INSN_SWITCH_TEXT_SECTIONS note in between
3041 : BEG and END. */
3042 :
3043 : static bool
3044 0 : switch_text_sections_between_p (const rtx_insn *beg, const rtx_insn *end)
3045 : {
3046 0 : const rtx_insn *p;
3047 0 : for (p = beg; p != end; p = NEXT_INSN (p))
3048 0 : if (NOTE_P (p) && NOTE_KIND (p) == NOTE_INSN_SWITCH_TEXT_SECTIONS)
3049 : return true;
3050 : return false;
3051 : }
3052 :
3053 :
3054 : /* Once we have tried two ways to fill a delay slot, make a pass over the
3055 : code to try to improve the results and to do such things as more jump
3056 : threading. */
3057 :
3058 : static void
3059 0 : relax_delay_slots (rtx_insn *first)
3060 : {
3061 0 : rtx_insn *insn, *next;
3062 0 : rtx_sequence *pat;
3063 0 : rtx_insn *delay_insn;
3064 0 : rtx target_label;
3065 :
3066 : /* Look at every JUMP_INSN and see if we can improve it. */
3067 0 : for (insn = first; insn; insn = next)
3068 : {
3069 0 : rtx_insn *other, *prior_insn;
3070 0 : bool crossing;
3071 :
3072 0 : next = next_active_insn (insn);
3073 :
3074 : /* If this is a jump insn, see if it now jumps to a jump, jumps to
3075 : the next insn, or jumps to a label that is not the last of a
3076 : group of consecutive labels. */
3077 0 : if (is_a <rtx_jump_insn *> (insn)
3078 0 : && (condjump_p (insn) || condjump_in_parallel_p (insn))
3079 0 : && !ANY_RETURN_P (target_label = JUMP_LABEL (insn)))
3080 : {
3081 0 : rtx_jump_insn *jump_insn = as_a <rtx_jump_insn *> (insn);
3082 0 : target_label
3083 0 : = skip_consecutive_labels (follow_jumps (target_label, jump_insn,
3084 : &crossing));
3085 0 : if (ANY_RETURN_P (target_label))
3086 0 : target_label = find_end_label (target_label);
3087 :
3088 0 : if (target_label
3089 0 : && next_active_insn (as_a<rtx_insn *> (target_label)) == next
3090 0 : && ! condjump_in_parallel_p (jump_insn)
3091 0 : && ! (next && switch_text_sections_between_p (jump_insn, next)))
3092 : {
3093 0 : rtx_insn *direct_label = as_a<rtx_insn *> (JUMP_LABEL (insn));
3094 0 : rtx_insn *prev = prev_nonnote_insn (direct_label);
3095 :
3096 : /* If the insn jumps over a BARRIER and is the only way to reach
3097 : its target, then we need to delete the BARRIER before the jump
3098 : because, otherwise, the target may end up being considered as
3099 : unreachable and thus also deleted. */
3100 0 : if (BARRIER_P (prev) && LABEL_NUSES (direct_label) == 1)
3101 : {
3102 0 : delete_related_insns (prev);
3103 :
3104 : /* We have just removed a BARRIER, which means that the block
3105 : number of the next insns has effectively been changed (see
3106 : find_basic_block in resource.cc), so clear it. */
3107 0 : clear_hashed_info_until_next_barrier (direct_label);
3108 : }
3109 :
3110 0 : delete_jump (jump_insn);
3111 0 : continue;
3112 0 : }
3113 :
3114 0 : if (target_label && target_label != JUMP_LABEL (jump_insn))
3115 : {
3116 0 : reorg_redirect_jump (jump_insn, target_label);
3117 0 : if (crossing)
3118 0 : CROSSING_JUMP_P (jump_insn) = 1;
3119 : }
3120 :
3121 : /* See if this jump conditionally branches around an unconditional
3122 : jump. If so, invert this jump and point it to the target of the
3123 : second jump. Check if it's possible on the target. */
3124 0 : if (next && simplejump_or_return_p (next)
3125 0 : && any_condjump_p (jump_insn)
3126 0 : && target_label
3127 0 : && (next_active_insn (as_a<rtx_insn *> (target_label))
3128 0 : == next_active_insn (next))
3129 0 : && no_labels_between_p (jump_insn, next)
3130 0 : && targetm.can_follow_jump (jump_insn, next))
3131 : {
3132 0 : rtx label = JUMP_LABEL (next);
3133 :
3134 : /* Be careful how we do this to avoid deleting code or
3135 : labels that are momentarily dead. See similar optimization
3136 : in jump.cc.
3137 :
3138 : We also need to ensure we properly handle the case when
3139 : invert_jump fails. */
3140 :
3141 0 : ++LABEL_NUSES (target_label);
3142 0 : if (!ANY_RETURN_P (label))
3143 0 : ++LABEL_NUSES (label);
3144 :
3145 0 : if (invert_jump (jump_insn, label, 1))
3146 : {
3147 0 : rtx_insn *from = delete_related_insns (next);
3148 :
3149 : /* We have just removed a BARRIER, which means that the block
3150 : number of the next insns has effectively been changed (see
3151 : find_basic_block in resource.cc), so clear it. */
3152 0 : if (from)
3153 0 : clear_hashed_info_until_next_barrier (from);
3154 :
3155 : next = jump_insn;
3156 : }
3157 :
3158 0 : if (!ANY_RETURN_P (label))
3159 0 : --LABEL_NUSES (label);
3160 :
3161 0 : if (--LABEL_NUSES (target_label) == 0)
3162 0 : delete_related_insns (target_label);
3163 :
3164 0 : continue;
3165 0 : }
3166 : }
3167 :
3168 : /* If this is an unconditional jump and the previous insn is a
3169 : conditional jump, try reversing the condition of the previous
3170 : insn and swapping our targets. The next pass might be able to
3171 : fill the slots.
3172 :
3173 : Don't do this if we expect the conditional branch to be true, because
3174 : we would then be making the more common case longer. */
3175 :
3176 0 : if (simplejump_or_return_p (insn)
3177 0 : && (other = prev_active_insn (insn)) != 0
3178 0 : && any_condjump_p (other)
3179 0 : && no_labels_between_p (other, insn)
3180 0 : && mostly_true_jump (other) < 0)
3181 : {
3182 0 : rtx other_target = JUMP_LABEL (other);
3183 0 : target_label = JUMP_LABEL (insn);
3184 :
3185 0 : if (invert_jump (as_a <rtx_jump_insn *> (other), target_label, 0))
3186 0 : reorg_redirect_jump (as_a <rtx_jump_insn *> (insn), other_target);
3187 : }
3188 :
3189 : /* Now look only at cases where we have a filled delay slot. */
3190 0 : if (!NONJUMP_INSN_P (insn) || GET_CODE (PATTERN (insn)) != SEQUENCE)
3191 0 : continue;
3192 :
3193 0 : pat = as_a <rtx_sequence *> (PATTERN (insn));
3194 0 : delay_insn = pat->insn (0);
3195 :
3196 : /* See if the first insn in the delay slot is redundant with some
3197 : previous insn. Remove it from the delay slot if so; then set up
3198 : to reprocess this insn. */
3199 0 : if ((prior_insn = redundant_insn (pat->insn (1), delay_insn, vNULL)))
3200 : {
3201 0 : fix_reg_dead_note (prior_insn, insn);
3202 0 : update_block (pat->insn (1), insn);
3203 0 : delete_from_delay_slot (pat->insn (1));
3204 0 : next = prev_active_insn (next);
3205 0 : continue;
3206 : }
3207 :
3208 : /* See if we have a RETURN insn with a filled delay slot followed
3209 : by a RETURN insn with an unfilled a delay slot. If so, we can delete
3210 : the first RETURN (but not its delay insn). This gives the same
3211 : effect in fewer instructions.
3212 :
3213 : Only do so if optimizing for size since this results in slower, but
3214 : smaller code. */
3215 0 : if (optimize_function_for_size_p (cfun)
3216 0 : && ANY_RETURN_P (PATTERN (delay_insn))
3217 0 : && next
3218 0 : && JUMP_P (next)
3219 0 : && PATTERN (next) == PATTERN (delay_insn))
3220 : {
3221 : rtx_insn *after;
3222 : int i;
3223 :
3224 : /* Delete the RETURN and just execute the delay list insns.
3225 :
3226 : We do this by deleting the INSN containing the SEQUENCE, then
3227 : re-emitting the insns separately, and then deleting the RETURN.
3228 : This allows the count of the jump target to be properly
3229 : decremented.
3230 :
3231 : Note that we need to change the INSN_UID of the re-emitted insns
3232 : since it is used to hash the insns for mark_target_live_regs and
3233 : the re-emitted insns will no longer be wrapped up in a SEQUENCE.
3234 :
3235 : Clear the from target bit, since these insns are no longer
3236 : in delay slots. */
3237 0 : for (i = 0; i < XVECLEN (pat, 0); i++)
3238 0 : INSN_FROM_TARGET_P (XVECEXP (pat, 0, i)) = 0;
3239 :
3240 0 : rtx_insn *prev = PREV_INSN (insn);
3241 0 : delete_related_insns (insn);
3242 0 : gcc_assert (GET_CODE (pat) == SEQUENCE);
3243 0 : add_insn_after (delay_insn, prev, NULL);
3244 0 : after = delay_insn;
3245 0 : for (i = 1; i < pat->len (); i++)
3246 0 : after = emit_copy_of_insn_after (pat->insn (i), after);
3247 0 : delete_scheduled_jump (delay_insn);
3248 0 : continue;
3249 0 : }
3250 :
3251 : /* Now look only at the cases where we have a filled JUMP_INSN. */
3252 0 : rtx_jump_insn *delay_jump_insn =
3253 0 : dyn_cast <rtx_jump_insn *> (delay_insn);
3254 0 : if (! delay_jump_insn || !(condjump_p (delay_jump_insn)
3255 0 : || condjump_in_parallel_p (delay_jump_insn)))
3256 0 : continue;
3257 :
3258 0 : target_label = JUMP_LABEL (delay_jump_insn);
3259 0 : if (target_label && ANY_RETURN_P (target_label))
3260 0 : continue;
3261 :
3262 : /* If this jump goes to another unconditional jump, thread it, but
3263 : don't convert a jump into a RETURN here. */
3264 0 : rtx trial = skip_consecutive_labels (follow_jumps (target_label,
3265 : delay_jump_insn,
3266 : &crossing));
3267 0 : if (ANY_RETURN_P (trial))
3268 0 : trial = find_end_label (trial);
3269 :
3270 0 : if (trial && trial != target_label
3271 0 : && redirect_with_delay_slots_safe_p (delay_jump_insn, trial, insn))
3272 : {
3273 0 : reorg_redirect_jump (delay_jump_insn, trial);
3274 0 : target_label = trial;
3275 0 : if (crossing)
3276 0 : CROSSING_JUMP_P (delay_jump_insn) = 1;
3277 : }
3278 :
3279 : /* If the first insn at TARGET_LABEL is redundant with a previous
3280 : insn, redirect the jump to the following insn and process again.
3281 : We use next_real_nondebug_insn instead of next_active_insn so we
3282 : don't skip USE-markers, or we'll end up with incorrect
3283 : liveness info. */
3284 0 : trial = next_real_nondebug_insn (target_label);
3285 0 : if (trial && GET_CODE (PATTERN (trial)) != SEQUENCE
3286 0 : && redundant_insn (trial, insn, vNULL)
3287 0 : && ! can_throw_internal (trial))
3288 : {
3289 : /* Figure out where to emit the special USE insn so we don't
3290 : later incorrectly compute register live/death info. */
3291 0 : rtx_insn *tmp = next_active_insn (as_a<rtx_insn *> (trial));
3292 0 : if (tmp == 0)
3293 0 : tmp = find_end_label (simple_return_rtx);
3294 :
3295 0 : if (tmp)
3296 : {
3297 : /* Insert the special USE insn and update dataflow info.
3298 : We know "trial" is an insn here as it is the output of
3299 : next_real_nondebug_insn () above. */
3300 0 : update_block (as_a <rtx_insn *> (trial), tmp);
3301 :
3302 : /* Now emit a label before the special USE insn, and
3303 : redirect our jump to the new label. */
3304 0 : target_label = get_label_before (PREV_INSN (tmp), target_label);
3305 0 : reorg_redirect_jump (delay_jump_insn, target_label);
3306 0 : next = insn;
3307 0 : continue;
3308 : }
3309 : }
3310 :
3311 : /* Similarly, if it is an unconditional jump with one insn in its
3312 : delay list and that insn is redundant, thread the jump. */
3313 0 : rtx_sequence *trial_seq =
3314 0 : trial ? dyn_cast <rtx_sequence *> (PATTERN (trial)) : NULL;
3315 0 : if (trial_seq
3316 0 : && trial_seq->len () == 2
3317 0 : && JUMP_P (trial_seq->insn (0))
3318 0 : && simplejump_or_return_p (trial_seq->insn (0))
3319 0 : && redundant_insn (trial_seq->insn (1), insn, vNULL))
3320 : {
3321 0 : rtx temp_label = JUMP_LABEL (trial_seq->insn (0));
3322 0 : if (ANY_RETURN_P (temp_label))
3323 0 : temp_label = find_end_label (temp_label);
3324 :
3325 0 : if (temp_label
3326 0 : && redirect_with_delay_slots_safe_p (delay_jump_insn,
3327 : temp_label, insn))
3328 : {
3329 0 : update_block (trial_seq->insn (1), insn);
3330 0 : reorg_redirect_jump (delay_jump_insn, temp_label);
3331 0 : next = insn;
3332 0 : continue;
3333 : }
3334 : }
3335 :
3336 : /* See if we have a simple (conditional) jump that is useless. */
3337 0 : if (!CROSSING_JUMP_P (delay_jump_insn)
3338 0 : && !INSN_ANNULLED_BRANCH_P (delay_jump_insn)
3339 0 : && !condjump_in_parallel_p (delay_jump_insn)
3340 0 : && prev_active_insn (as_a<rtx_insn *> (target_label)) == insn
3341 0 : && !BARRIER_P (prev_nonnote_insn (as_a<rtx_insn *> (target_label))))
3342 : {
3343 : rtx_insn *after;
3344 : int i;
3345 :
3346 : /* All this insn does is execute its delay list and jump to the
3347 : following insn. So delete the jump and just execute the delay
3348 : list insns.
3349 :
3350 : We do this by deleting the INSN containing the SEQUENCE, then
3351 : re-emitting the insns separately, and then deleting the jump.
3352 : This allows the count of the jump target to be properly
3353 : decremented.
3354 :
3355 : Note that we need to change the INSN_UID of the re-emitted insns
3356 : since it is used to hash the insns for mark_target_live_regs and
3357 : the re-emitted insns will no longer be wrapped up in a SEQUENCE.
3358 :
3359 : Clear the from target bit, since these insns are no longer
3360 : in delay slots. */
3361 0 : for (i = 0; i < XVECLEN (pat, 0); i++)
3362 0 : INSN_FROM_TARGET_P (XVECEXP (pat, 0, i)) = 0;
3363 :
3364 0 : rtx_insn *prev = PREV_INSN (insn);
3365 0 : delete_related_insns (insn);
3366 0 : gcc_assert (GET_CODE (pat) == SEQUENCE);
3367 0 : add_insn_after (delay_jump_insn, prev, NULL);
3368 0 : after = delay_jump_insn;
3369 0 : for (i = 1; i < pat->len (); i++)
3370 0 : after = emit_copy_of_insn_after (pat->insn (i), after);
3371 0 : delete_scheduled_jump (delay_jump_insn);
3372 0 : continue;
3373 0 : }
3374 :
3375 : /* See if this is an unconditional jump around a single insn which is
3376 : identical to the one in its delay slot. In this case, we can just
3377 : delete the branch and the insn in its delay slot. */
3378 0 : if (next && NONJUMP_INSN_P (next)
3379 0 : && label_before_next_insn (next, insn) == target_label
3380 0 : && simplejump_p (insn)
3381 0 : && XVECLEN (pat, 0) == 2
3382 0 : && rtx_equal_p (PATTERN (next), PATTERN (pat->insn (1))))
3383 : {
3384 0 : delete_related_insns (insn);
3385 0 : continue;
3386 : }
3387 :
3388 : /* See if this jump (with its delay slots) conditionally branches
3389 : around an unconditional jump (without delay slots). If so, invert
3390 : this jump and point it to the target of the second jump. We cannot
3391 : do this for annulled jumps, though. Again, don't convert a jump to
3392 : a RETURN here. */
3393 0 : if (! INSN_ANNULLED_BRANCH_P (delay_jump_insn)
3394 0 : && any_condjump_p (delay_jump_insn)
3395 0 : && next && simplejump_or_return_p (next)
3396 0 : && (next_active_insn (as_a<rtx_insn *> (target_label))
3397 0 : == next_active_insn (next))
3398 0 : && no_labels_between_p (insn, next)
3399 0 : && !switch_text_sections_between_p (insn, next_active_insn (next)))
3400 : {
3401 0 : rtx label = JUMP_LABEL (next);
3402 0 : rtx old_label = JUMP_LABEL (delay_jump_insn);
3403 :
3404 0 : if (ANY_RETURN_P (label))
3405 0 : label = find_end_label (label);
3406 :
3407 : /* find_end_label can generate a new label. Check this first. */
3408 0 : if (label
3409 0 : && no_labels_between_p (insn, next)
3410 0 : && redirect_with_delay_slots_safe_p (delay_jump_insn,
3411 : label, insn))
3412 : {
3413 : /* Be careful how we do this to avoid deleting code or labels
3414 : that are momentarily dead. See similar optimization in
3415 : jump.cc */
3416 0 : if (old_label)
3417 0 : ++LABEL_NUSES (old_label);
3418 :
3419 0 : if (invert_jump (delay_jump_insn, label, 1))
3420 : {
3421 : /* Must update the INSN_FROM_TARGET_P bits now that
3422 : the branch is reversed, so that mark_target_live_regs
3423 : will handle the delay slot insn correctly. */
3424 0 : for (int i = 1; i < XVECLEN (PATTERN (insn), 0); i++)
3425 : {
3426 0 : rtx slot = XVECEXP (PATTERN (insn), 0, i);
3427 0 : INSN_FROM_TARGET_P (slot) = ! INSN_FROM_TARGET_P (slot);
3428 : }
3429 :
3430 : /* We have just removed a BARRIER, which means that the block
3431 : number of the next insns has effectively been changed (see
3432 : find_basic_block in resource.cc), so clear it. */
3433 0 : rtx_insn *from = delete_related_insns (next);
3434 0 : if (from)
3435 0 : clear_hashed_info_until_next_barrier (from);
3436 :
3437 : next = insn;
3438 : }
3439 :
3440 0 : if (old_label && --LABEL_NUSES (old_label) == 0)
3441 0 : delete_related_insns (old_label);
3442 0 : continue;
3443 0 : }
3444 : }
3445 :
3446 : /* If we own the thread opposite the way this insn branches, see if we
3447 : can merge its delay slots with following insns. */
3448 0 : if (INSN_FROM_TARGET_P (pat->insn (1))
3449 0 : && own_thread_p (NEXT_INSN (insn), 0, true))
3450 0 : try_merge_delay_insns (insn, next);
3451 0 : else if (! INSN_FROM_TARGET_P (pat->insn (1))
3452 0 : && own_thread_p (target_label, target_label, false))
3453 0 : try_merge_delay_insns (insn,
3454 : next_active_insn (as_a<rtx_insn *> (target_label)));
3455 :
3456 : /* If we get here, we haven't deleted INSN. But we may have deleted
3457 : NEXT, so recompute it. */
3458 0 : next = next_active_insn (insn);
3459 : }
3460 0 : }
3461 :
3462 :
3463 : /* Look for filled jumps to the end of function label. We can try to convert
3464 : them into RETURN insns if the insns in the delay slot are valid for the
3465 : RETURN as well. */
3466 :
3467 : static void
3468 0 : make_return_insns (rtx_insn *first)
3469 : {
3470 0 : rtx_insn *insn;
3471 0 : rtx_jump_insn *jump_insn;
3472 0 : rtx real_return_label = function_return_label;
3473 0 : rtx real_simple_return_label = function_simple_return_label;
3474 0 : int slots, i;
3475 :
3476 : /* See if there is a RETURN insn in the function other than the one we
3477 : made for END_OF_FUNCTION_LABEL. If so, set up anything we can't change
3478 : into a RETURN to jump to it. */
3479 0 : for (insn = first; insn; insn = NEXT_INSN (insn))
3480 0 : if (JUMP_P (insn) && ANY_RETURN_P (PATTERN (insn)))
3481 : {
3482 0 : rtx t = get_label_before (insn, NULL_RTX);
3483 0 : if (PATTERN (insn) == ret_rtx)
3484 : real_return_label = t;
3485 : else
3486 : real_simple_return_label = t;
3487 : break;
3488 : }
3489 :
3490 : /* Show an extra usage of REAL_RETURN_LABEL so it won't go away if it
3491 : was equal to END_OF_FUNCTION_LABEL. */
3492 0 : if (real_return_label)
3493 0 : LABEL_NUSES (real_return_label)++;
3494 0 : if (real_simple_return_label)
3495 0 : LABEL_NUSES (real_simple_return_label)++;
3496 :
3497 : /* Clear the list of insns to fill so we can use it. */
3498 0 : obstack_free (&unfilled_slots_obstack, unfilled_firstobj);
3499 :
3500 0 : for (insn = first; insn; insn = NEXT_INSN (insn))
3501 : {
3502 0 : int flags;
3503 0 : rtx kind, real_label;
3504 :
3505 : /* Only look at filled JUMP_INSNs that go to the end of function
3506 : label. */
3507 0 : if (!NONJUMP_INSN_P (insn))
3508 0 : continue;
3509 :
3510 0 : if (GET_CODE (PATTERN (insn)) != SEQUENCE)
3511 0 : continue;
3512 :
3513 0 : rtx_sequence *pat = as_a <rtx_sequence *> (PATTERN (insn));
3514 :
3515 0 : if (!jump_to_label_p (pat->insn (0)))
3516 0 : continue;
3517 :
3518 0 : if (JUMP_LABEL (pat->insn (0)) == function_return_label)
3519 : {
3520 0 : kind = ret_rtx;
3521 0 : real_label = real_return_label;
3522 : }
3523 0 : else if (JUMP_LABEL (pat->insn (0)) == function_simple_return_label)
3524 : {
3525 0 : kind = simple_return_rtx;
3526 0 : real_label = real_simple_return_label;
3527 : }
3528 : else
3529 0 : continue;
3530 :
3531 0 : jump_insn = as_a <rtx_jump_insn *> (pat->insn (0));
3532 :
3533 : /* If we can't make the jump into a RETURN, try to redirect it to the best
3534 : RETURN and go on to the next insn. */
3535 0 : if (!reorg_redirect_jump (jump_insn, kind))
3536 : {
3537 : /* Make sure redirecting the jump will not invalidate the delay
3538 : slot insns. */
3539 0 : if (redirect_with_delay_slots_safe_p (jump_insn, real_label, insn))
3540 0 : reorg_redirect_jump (jump_insn, real_label);
3541 0 : continue;
3542 : }
3543 :
3544 : /* See if this RETURN can accept the insns current in its delay slot.
3545 : It can if it has more or an equal number of slots and the contents
3546 : of each is valid. */
3547 :
3548 0 : flags = get_jump_flags (jump_insn, JUMP_LABEL (jump_insn));
3549 0 : slots = num_delay_slots (jump_insn);
3550 0 : if (slots >= XVECLEN (pat, 0) - 1)
3551 : {
3552 0 : for (i = 1; i < XVECLEN (pat, 0); i++)
3553 0 : if (! (
3554 : #if ANNUL_IFFALSE_SLOTS
3555 : (INSN_ANNULLED_BRANCH_P (jump_insn)
3556 : && INSN_FROM_TARGET_P (pat->insn (i)))
3557 : ? eligible_for_annul_false (jump_insn, i - 1,
3558 : pat->insn (i), flags) :
3559 : #endif
3560 : #if ANNUL_IFTRUE_SLOTS
3561 : (INSN_ANNULLED_BRANCH_P (jump_insn)
3562 : && ! INSN_FROM_TARGET_P (pat->insn (i)))
3563 : ? eligible_for_annul_true (jump_insn, i - 1,
3564 : pat->insn (i), flags) :
3565 : #endif
3566 0 : eligible_for_delay (jump_insn, i - 1,
3567 : pat->insn (i), flags)))
3568 : break;
3569 : }
3570 : else
3571 : i = 0;
3572 :
3573 0 : if (i == XVECLEN (pat, 0))
3574 0 : continue;
3575 :
3576 : /* We have to do something with this insn. If it is an unconditional
3577 : RETURN, delete the SEQUENCE and output the individual insns,
3578 : followed by the RETURN. Then set things up so we try to find
3579 : insns for its delay slots, if it needs some. */
3580 0 : if (ANY_RETURN_P (PATTERN (jump_insn)))
3581 : {
3582 0 : rtx_insn *after = PREV_INSN (insn);
3583 :
3584 0 : delete_related_insns (insn);
3585 0 : insn = jump_insn;
3586 0 : for (i = 1; i < pat->len (); i++)
3587 0 : after = emit_copy_of_insn_after (pat->insn (i), after);
3588 0 : add_insn_after (insn, after, NULL);
3589 0 : emit_barrier_after (insn);
3590 :
3591 0 : if (slots)
3592 0 : obstack_ptr_grow (&unfilled_slots_obstack, insn);
3593 : }
3594 : else
3595 : /* It is probably more efficient to keep this with its current
3596 : delay slot as a branch to a RETURN. */
3597 0 : reorg_redirect_jump (jump_insn, real_label);
3598 : }
3599 :
3600 : /* Now delete REAL_RETURN_LABEL if we never used it. Then try to fill any
3601 : new delay slots we have created. */
3602 0 : if (real_return_label != NULL_RTX && --LABEL_NUSES (real_return_label) == 0)
3603 0 : delete_related_insns (real_return_label);
3604 0 : if (real_simple_return_label != NULL_RTX
3605 0 : && --LABEL_NUSES (real_simple_return_label) == 0)
3606 0 : delete_related_insns (real_simple_return_label);
3607 :
3608 0 : fill_simple_delay_slots (true);
3609 0 : fill_simple_delay_slots (false);
3610 0 : }
3611 :
3612 : /* Try to find insns to place in delay slots. */
3613 :
3614 : static void
3615 0 : dbr_schedule (rtx_insn *first)
3616 : {
3617 0 : rtx_insn *insn, *next, *epilogue_insn = 0;
3618 0 : bool need_return_insns;
3619 0 : int i;
3620 :
3621 : /* If the current function has no insns other than the prologue and
3622 : epilogue, then do not try to fill any delay slots. */
3623 0 : if (n_basic_blocks_for_fn (cfun) == NUM_FIXED_BLOCKS)
3624 : return;
3625 :
3626 : /* Find the highest INSN_UID and allocate and initialize our map from
3627 : INSN_UID's to position in code. */
3628 0 : for (max_uid = 0, insn = first; insn; insn = NEXT_INSN (insn))
3629 : {
3630 0 : if (INSN_UID (insn) > max_uid)
3631 0 : max_uid = INSN_UID (insn);
3632 0 : if (NOTE_P (insn)
3633 0 : && NOTE_KIND (insn) == NOTE_INSN_EPILOGUE_BEG)
3634 0 : epilogue_insn = insn;
3635 : }
3636 :
3637 0 : uid_to_ruid = XNEWVEC (int, max_uid + 1);
3638 0 : for (i = 0, insn = first; insn; i++, insn = NEXT_INSN (insn))
3639 0 : uid_to_ruid[INSN_UID (insn)] = i;
3640 :
3641 : /* Initialize the list of insns that need filling. */
3642 0 : if (unfilled_firstobj == 0)
3643 : {
3644 0 : gcc_obstack_init (&unfilled_slots_obstack);
3645 0 : unfilled_firstobj = XOBNEWVAR (&unfilled_slots_obstack, rtx, 0);
3646 : }
3647 :
3648 0 : for (insn = next_active_insn (first); insn; insn = next_active_insn (insn))
3649 : {
3650 0 : rtx target;
3651 :
3652 : /* Skip vector tables. We can't get attributes for them. */
3653 0 : if (JUMP_TABLE_DATA_P (insn))
3654 0 : continue;
3655 :
3656 0 : if (JUMP_P (insn))
3657 0 : INSN_ANNULLED_BRANCH_P (insn) = 0;
3658 0 : INSN_FROM_TARGET_P (insn) = 0;
3659 :
3660 0 : if (num_delay_slots (insn) > 0)
3661 0 : obstack_ptr_grow (&unfilled_slots_obstack, insn);
3662 :
3663 : /* Ensure all jumps go to the last of a set of consecutive labels. */
3664 0 : if (JUMP_P (insn)
3665 0 : && (condjump_p (insn) || condjump_in_parallel_p (insn))
3666 0 : && !ANY_RETURN_P (JUMP_LABEL (insn))
3667 0 : && ((target = skip_consecutive_labels (JUMP_LABEL (insn)))
3668 : != JUMP_LABEL (insn)))
3669 0 : redirect_jump (as_a <rtx_jump_insn *> (insn), target, 1);
3670 : }
3671 :
3672 0 : init_resource_info (epilogue_insn);
3673 :
3674 : /* Show we haven't computed an end-of-function label yet. */
3675 0 : function_return_label = function_simple_return_label = NULL;
3676 :
3677 : /* Initialize the statistics for this function. */
3678 0 : memset (num_insns_needing_delays, 0, sizeof num_insns_needing_delays);
3679 0 : memset (num_filled_delays, 0, sizeof num_filled_delays);
3680 :
3681 : /* Now do the delay slot filling. Try everything twice in case earlier
3682 : changes make more slots fillable. */
3683 :
3684 0 : for (reorg_pass_number = 0;
3685 0 : reorg_pass_number < MAX_REORG_PASSES;
3686 : reorg_pass_number++)
3687 : {
3688 0 : fill_simple_delay_slots (true);
3689 0 : fill_simple_delay_slots (false);
3690 0 : if (!targetm.no_speculation_in_delay_slots_p ())
3691 0 : fill_eager_delay_slots ();
3692 0 : relax_delay_slots (first);
3693 : }
3694 :
3695 : /* If we made an end of function label, indicate that it is now
3696 : safe to delete it by undoing our prior adjustment to LABEL_NUSES.
3697 : If it is now unused, delete it. */
3698 0 : if (function_return_label && --LABEL_NUSES (function_return_label) == 0)
3699 0 : delete_related_insns (function_return_label);
3700 0 : if (function_simple_return_label
3701 0 : && --LABEL_NUSES (function_simple_return_label) == 0)
3702 0 : delete_related_insns (function_simple_return_label);
3703 :
3704 0 : need_return_insns = false;
3705 0 : need_return_insns |= targetm.have_return () && function_return_label != 0;
3706 0 : need_return_insns |= (targetm.have_simple_return ()
3707 0 : && function_simple_return_label != 0);
3708 0 : if (need_return_insns)
3709 0 : make_return_insns (first);
3710 :
3711 : /* Delete any USE insns made by update_block; subsequent passes don't need
3712 : them or know how to deal with them. */
3713 0 : for (insn = first; insn; insn = next)
3714 : {
3715 0 : next = NEXT_INSN (insn);
3716 :
3717 0 : if (NONJUMP_INSN_P (insn) && GET_CODE (PATTERN (insn)) == USE
3718 0 : && INSN_P (XEXP (PATTERN (insn), 0)))
3719 0 : next = delete_related_insns (insn);
3720 : }
3721 :
3722 0 : obstack_free (&unfilled_slots_obstack, unfilled_firstobj);
3723 :
3724 : /* It is not clear why the line below is needed, but it does seem to be. */
3725 0 : unfilled_firstobj = XOBNEWVAR (&unfilled_slots_obstack, rtx, 0);
3726 :
3727 0 : if (dump_file)
3728 : {
3729 0 : int i, j, need_comma;
3730 0 : int total_delay_slots[MAX_DELAY_HISTOGRAM + 1];
3731 0 : int total_annul_slots[MAX_DELAY_HISTOGRAM + 1];
3732 :
3733 0 : for (reorg_pass_number = 0;
3734 0 : reorg_pass_number < MAX_REORG_PASSES;
3735 : reorg_pass_number++)
3736 : {
3737 0 : fprintf (dump_file, ";; Reorg pass #%d:\n", reorg_pass_number + 1);
3738 0 : for (i = 0; i < NUM_REORG_FUNCTIONS; i++)
3739 : {
3740 0 : need_comma = 0;
3741 0 : fprintf (dump_file, ";; Reorg function #%d\n", i);
3742 :
3743 0 : fprintf (dump_file, ";; %d insns needing delay slots\n;; ",
3744 : num_insns_needing_delays[i][reorg_pass_number]);
3745 :
3746 0 : for (j = 0; j < MAX_DELAY_HISTOGRAM + 1; j++)
3747 0 : if (num_filled_delays[i][j][reorg_pass_number])
3748 : {
3749 0 : if (need_comma)
3750 0 : fprintf (dump_file, ", ");
3751 0 : need_comma = 1;
3752 0 : fprintf (dump_file, "%d got %d delays",
3753 : num_filled_delays[i][j][reorg_pass_number], j);
3754 : }
3755 0 : fprintf (dump_file, "\n");
3756 : }
3757 : }
3758 0 : memset (total_delay_slots, 0, sizeof total_delay_slots);
3759 0 : memset (total_annul_slots, 0, sizeof total_annul_slots);
3760 0 : for (insn = first; insn; insn = NEXT_INSN (insn))
3761 : {
3762 0 : if (! insn->deleted ()
3763 0 : && NONJUMP_INSN_P (insn)
3764 0 : && GET_CODE (PATTERN (insn)) != USE
3765 0 : && GET_CODE (PATTERN (insn)) != CLOBBER)
3766 : {
3767 0 : if (GET_CODE (PATTERN (insn)) == SEQUENCE)
3768 : {
3769 0 : rtx control;
3770 0 : j = XVECLEN (PATTERN (insn), 0) - 1;
3771 0 : if (j > MAX_DELAY_HISTOGRAM)
3772 : j = MAX_DELAY_HISTOGRAM;
3773 0 : control = XVECEXP (PATTERN (insn), 0, 0);
3774 0 : if (JUMP_P (control) && INSN_ANNULLED_BRANCH_P (control))
3775 0 : total_annul_slots[j]++;
3776 : else
3777 0 : total_delay_slots[j]++;
3778 : }
3779 0 : else if (num_delay_slots (insn) > 0)
3780 0 : total_delay_slots[0]++;
3781 : }
3782 : }
3783 0 : fprintf (dump_file, ";; Reorg totals: ");
3784 0 : need_comma = 0;
3785 0 : for (j = 0; j < MAX_DELAY_HISTOGRAM + 1; j++)
3786 : {
3787 0 : if (total_delay_slots[j])
3788 : {
3789 0 : if (need_comma)
3790 0 : fprintf (dump_file, ", ");
3791 0 : need_comma = 1;
3792 0 : fprintf (dump_file, "%d got %d delays", total_delay_slots[j], j);
3793 : }
3794 : }
3795 0 : fprintf (dump_file, "\n");
3796 :
3797 0 : if (ANNUL_IFTRUE_SLOTS || ANNUL_IFFALSE_SLOTS)
3798 : {
3799 : fprintf (dump_file, ";; Reorg annuls: ");
3800 : need_comma = 0;
3801 : for (j = 0; j < MAX_DELAY_HISTOGRAM + 1; j++)
3802 : {
3803 : if (total_annul_slots[j])
3804 : {
3805 : if (need_comma)
3806 : fprintf (dump_file, ", ");
3807 : need_comma = 1;
3808 : fprintf (dump_file, "%d got %d delays", total_annul_slots[j], j);
3809 : }
3810 : }
3811 : fprintf (dump_file, "\n");
3812 : }
3813 :
3814 0 : fprintf (dump_file, "\n");
3815 : }
3816 :
3817 0 : if (!sibling_labels.is_empty ())
3818 : {
3819 0 : update_alignments (sibling_labels);
3820 0 : sibling_labels.release ();
3821 : }
3822 :
3823 0 : free_resource_info ();
3824 0 : free (uid_to_ruid);
3825 0 : crtl->dbr_scheduled_p = true;
3826 : }
3827 :
3828 : /* Run delay slot optimization. */
3829 : static void
3830 0 : rest_of_handle_delay_slots (void)
3831 : {
3832 0 : if (DELAY_SLOTS)
3833 : dbr_schedule (get_insns ());
3834 0 : }
3835 :
3836 : namespace {
3837 :
3838 : const pass_data pass_data_delay_slots =
3839 : {
3840 : RTL_PASS, /* type */
3841 : "dbr", /* name */
3842 : OPTGROUP_NONE, /* optinfo_flags */
3843 : TV_DBR_SCHED, /* tv_id */
3844 : 0, /* properties_required */
3845 : 0, /* properties_provided */
3846 : 0, /* properties_destroyed */
3847 : 0, /* todo_flags_start */
3848 : 0, /* todo_flags_finish */
3849 : };
3850 :
3851 : class pass_delay_slots : public rtl_opt_pass
3852 : {
3853 : public:
3854 294587 : pass_delay_slots (gcc::context *ctxt)
3855 589174 : : rtl_opt_pass (pass_data_delay_slots, ctxt)
3856 : {}
3857 :
3858 : /* opt_pass methods: */
3859 : bool gate (function *) final override;
3860 0 : unsigned int execute (function *) final override
3861 : {
3862 0 : rest_of_handle_delay_slots ();
3863 0 : return 0;
3864 : }
3865 :
3866 : }; // class pass_delay_slots
3867 :
3868 : bool
3869 1511392 : pass_delay_slots::gate (function *)
3870 : {
3871 : /* At -O0 dataflow info isn't updated after RA. */
3872 1511392 : if (DELAY_SLOTS)
3873 : return optimize > 0 && flag_delayed_branch && !crtl->dbr_scheduled_p;
3874 :
3875 1511392 : return false;
3876 : }
3877 :
3878 : } // anon namespace
3879 :
3880 : rtl_opt_pass *
3881 294587 : make_pass_delay_slots (gcc::context *ctxt)
3882 : {
3883 294587 : return new pass_delay_slots (ctxt);
3884 : }
3885 :
3886 : /* Machine dependent reorg pass. */
3887 :
3888 : namespace {
3889 :
3890 : const pass_data pass_data_machine_reorg =
3891 : {
3892 : RTL_PASS, /* type */
3893 : "mach", /* name */
3894 : OPTGROUP_NONE, /* optinfo_flags */
3895 : TV_MACH_DEP, /* tv_id */
3896 : 0, /* properties_required */
3897 : 0, /* properties_provided */
3898 : 0, /* properties_destroyed */
3899 : 0, /* todo_flags_start */
3900 : 0, /* todo_flags_finish */
3901 : };
3902 :
3903 : class pass_machine_reorg : public rtl_opt_pass
3904 : {
3905 : public:
3906 294587 : pass_machine_reorg (gcc::context *ctxt)
3907 589174 : : rtl_opt_pass (pass_data_machine_reorg, ctxt)
3908 : {}
3909 :
3910 : /* opt_pass methods: */
3911 1511392 : bool gate (function *) final override
3912 : {
3913 1511392 : return targetm.machine_dependent_reorg != 0;
3914 : }
3915 :
3916 1511385 : unsigned int execute (function *) final override
3917 : {
3918 1511385 : targetm.machine_dependent_reorg ();
3919 1511385 : return 0;
3920 : }
3921 :
3922 : }; // class pass_machine_reorg
3923 :
3924 : } // anon namespace
3925 :
3926 : rtl_opt_pass *
3927 294587 : make_pass_machine_reorg (gcc::context *ctxt)
3928 : {
3929 294587 : return new pass_machine_reorg (ctxt);
3930 : }
|
