1 /* Combine stack adjustments.
2 Copyright (C) 1987-2014 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 /* Track stack adjustments and stack memory references. Attempt to
21 reduce the number of stack adjustments by back-propagating across
22 the memory references.
24 This is intended primarily for use with targets that do not define
25 ACCUMULATE_OUTGOING_ARGS. It is of significantly more value to
26 targets that define PREFERRED_STACK_BOUNDARY more aligned than
27 STACK_BOUNDARY (e.g. x86), or if not all registers can be pushed
28 (e.g. x86 fp regs) which would ordinarily have to be implemented
29 as a sub/mov pair due to restrictions in calls.c.
31 Propagation stops when any of the insns that need adjusting are
32 (a) no longer valid because we've exceeded their range, (b) a
33 non-trivial push instruction, or (c) a call instruction.
35 Restriction B is based on the assumption that push instructions
36 are smaller or faster. If a port really wants to remove all
37 pushes, it should have defined ACCUMULATE_OUTGOING_ARGS. The
38 one exception that is made is for an add immediately followed
43 #include "coretypes.h"
47 #include "insn-config.h"
50 #include "hard-reg-set.h"
54 #include "basic-block.h"
58 #include "tree-pass.h"
61 /* Turn STACK_GROWS_DOWNWARD into a boolean. */
62 #ifdef STACK_GROWS_DOWNWARD
63 #undef STACK_GROWS_DOWNWARD
64 #define STACK_GROWS_DOWNWARD 1
66 #define STACK_GROWS_DOWNWARD 0
69 /* This structure records two kinds of stack references between stack
70 adjusting instructions: stack references in memory addresses for
71 regular insns and all stack references for debug insns. */
75 HOST_WIDE_INT sp_offset
;
78 struct csa_reflist
*next
;
81 static int stack_memref_p (rtx
);
82 static rtx
single_set_for_csa (rtx_insn
*);
83 static void free_csa_reflist (struct csa_reflist
*);
84 static struct csa_reflist
*record_one_stack_ref (rtx_insn
*, rtx
*,
85 struct csa_reflist
*);
86 static int try_apply_stack_adjustment (rtx_insn
*, struct csa_reflist
*,
87 HOST_WIDE_INT
, HOST_WIDE_INT
);
88 static void combine_stack_adjustments_for_block (basic_block
);
89 static int record_stack_refs (rtx
*, void *);
92 /* Main entry point for stack adjustment combination. */
95 combine_stack_adjustments (void)
99 FOR_EACH_BB_FN (bb
, cfun
)
100 combine_stack_adjustments_for_block (bb
);
103 /* Recognize a MEM of the form (sp) or (plus sp const). */
106 stack_memref_p (rtx x
)
112 if (x
== stack_pointer_rtx
)
114 if (GET_CODE (x
) == PLUS
115 && XEXP (x
, 0) == stack_pointer_rtx
116 && CONST_INT_P (XEXP (x
, 1)))
122 /* Recognize either normal single_set or the hack in i386.md for
123 tying fp and sp adjustments. */
126 single_set_for_csa (rtx_insn
*insn
)
129 rtx tmp
= single_set (insn
);
133 if (!NONJUMP_INSN_P (insn
)
134 || GET_CODE (PATTERN (insn
)) != PARALLEL
)
137 tmp
= PATTERN (insn
);
138 if (GET_CODE (XVECEXP (tmp
, 0, 0)) != SET
)
141 for (i
= 1; i
< XVECLEN (tmp
, 0); ++i
)
143 rtx this_rtx
= XVECEXP (tmp
, 0, i
);
145 /* The special case is allowing a no-op set. */
146 if (GET_CODE (this_rtx
) == SET
147 && SET_SRC (this_rtx
) == SET_DEST (this_rtx
))
149 else if (GET_CODE (this_rtx
) != CLOBBER
150 && GET_CODE (this_rtx
) != USE
)
154 return XVECEXP (tmp
, 0, 0);
157 /* Free the list of csa_reflist nodes. */
160 free_csa_reflist (struct csa_reflist
*reflist
)
162 struct csa_reflist
*next
;
163 for (; reflist
; reflist
= next
)
165 next
= reflist
->next
;
170 /* Create a new csa_reflist node from the given stack reference.
171 It is already known that the reference is either a MEM satisfying the
172 predicate stack_memref_p or a REG representing the stack pointer. */
174 static struct csa_reflist
*
175 record_one_stack_ref (rtx_insn
*insn
, rtx
*ref
, struct csa_reflist
*next_reflist
)
177 struct csa_reflist
*ml
;
179 ml
= XNEW (struct csa_reflist
);
181 if (REG_P (*ref
) || XEXP (*ref
, 0) == stack_pointer_rtx
)
184 ml
->sp_offset
= INTVAL (XEXP (XEXP (*ref
, 0), 1));
188 ml
->next
= next_reflist
;
193 /* Attempt to apply ADJUST to the stack adjusting insn INSN, as well
194 as each of the memories and stack references in REFLIST. Return true
198 try_apply_stack_adjustment (rtx_insn
*insn
, struct csa_reflist
*reflist
,
199 HOST_WIDE_INT new_adjust
, HOST_WIDE_INT delta
)
201 struct csa_reflist
*ml
;
204 set
= single_set_for_csa (insn
);
205 if (MEM_P (SET_DEST (set
)))
206 validate_change (insn
, &SET_DEST (set
),
207 replace_equiv_address (SET_DEST (set
), stack_pointer_rtx
),
210 validate_change (insn
, &XEXP (SET_SRC (set
), 1), GEN_INT (new_adjust
), 1);
212 for (ml
= reflist
; ml
; ml
= ml
->next
)
214 rtx new_addr
= plus_constant (Pmode
, stack_pointer_rtx
,
215 ml
->sp_offset
- delta
);
218 if (MEM_P (*ml
->ref
))
219 new_val
= replace_equiv_address_nv (*ml
->ref
, new_addr
);
220 else if (GET_MODE (*ml
->ref
) == GET_MODE (stack_pointer_rtx
))
223 new_val
= lowpart_subreg (GET_MODE (*ml
->ref
), new_addr
,
224 GET_MODE (new_addr
));
225 validate_change (ml
->insn
, ml
->ref
, new_val
, 1);
228 if (apply_change_group ())
230 /* Succeeded. Update our knowledge of the stack references. */
231 for (ml
= reflist
; ml
; ml
= ml
->next
)
232 ml
->sp_offset
-= delta
;
240 /* Called via for_each_rtx and used to record all stack memory and other
241 references in the insn and discard all other stack pointer references. */
242 struct record_stack_refs_data
245 struct csa_reflist
*reflist
;
249 record_stack_refs (rtx
*xp
, void *data
)
252 struct record_stack_refs_data
*d
=
253 (struct record_stack_refs_data
*) data
;
256 switch (GET_CODE (x
))
259 if (!reg_mentioned_p (stack_pointer_rtx
, x
))
261 /* We are not able to handle correctly all possible memrefs containing
262 stack pointer, so this check is necessary. */
263 if (stack_memref_p (x
))
265 d
->reflist
= record_one_stack_ref (d
->insn
, xp
, d
->reflist
);
268 /* Try harder for DEBUG_INSNs, handle e.g. (mem (mem (sp + 16) + 4). */
269 return !DEBUG_INSN_P (d
->insn
);
271 /* ??? We want be able to handle non-memory stack pointer
272 references later. For now just discard all insns referring to
273 stack pointer outside mem expressions. We would probably
274 want to teach validate_replace to simplify expressions first.
276 We can't just compare with STACK_POINTER_RTX because the
277 reference to the stack pointer might be in some other mode.
278 In particular, an explicit clobber in an asm statement will
279 result in a QImode clobber.
281 In DEBUG_INSNs, we want to replace all occurrences, otherwise
282 they will cause -fcompare-debug failures. */
283 if (REGNO (x
) == STACK_POINTER_REGNUM
)
285 if (!DEBUG_INSN_P (d
->insn
))
287 d
->reflist
= record_one_stack_ref (d
->insn
, xp
, d
->reflist
);
297 /* If INSN has a REG_ARGS_SIZE note, move it to LAST.
298 AFTER is true iff LAST follows INSN in the instruction stream. */
301 maybe_move_args_size_note (rtx_insn
*last
, rtx_insn
*insn
, bool after
)
305 note
= find_reg_note (insn
, REG_ARGS_SIZE
, NULL_RTX
);
309 last_note
= find_reg_note (last
, REG_ARGS_SIZE
, NULL_RTX
);
312 /* The ARGS_SIZE notes are *not* cumulative. They represent an
313 absolute value, and the "most recent" note wins. */
315 XEXP (last_note
, 0) = XEXP (note
, 0);
318 add_reg_note (last
, REG_ARGS_SIZE
, XEXP (note
, 0));
321 /* Return the next (or previous) active insn within BB. */
324 prev_active_insn_bb (basic_block bb
, rtx_insn
*insn
)
326 for (insn
= PREV_INSN (insn
);
327 insn
!= PREV_INSN (BB_HEAD (bb
));
328 insn
= PREV_INSN (insn
))
329 if (active_insn_p (insn
))
335 next_active_insn_bb (basic_block bb
, rtx_insn
*insn
)
337 for (insn
= NEXT_INSN (insn
);
338 insn
!= NEXT_INSN (BB_END (bb
));
339 insn
= NEXT_INSN (insn
))
340 if (active_insn_p (insn
))
345 /* If INSN has a REG_ARGS_SIZE note, if possible move it to PREV. Otherwise
346 search for a nearby candidate within BB where we can stick the note. */
349 force_move_args_size_note (basic_block bb
, rtx_insn
*prev
, rtx_insn
*insn
)
352 rtx_insn
*test
, *next_candidate
, *prev_candidate
;
354 /* If PREV exists, tail-call to the logic in the other function. */
357 maybe_move_args_size_note (prev
, insn
, false);
361 /* First, make sure there's anything that needs doing. */
362 note
= find_reg_note (insn
, REG_ARGS_SIZE
, NULL_RTX
);
366 /* We need to find a spot between the previous and next exception points
367 where we can place the note and "properly" deallocate the arguments. */
368 next_candidate
= prev_candidate
= NULL
;
370 /* It is often the case that we have insns in the order:
372 add sp (previous deallocation)
373 sub sp (align for next arglist)
375 and the add/sub cancel. Therefore we begin by searching forward. */
378 while ((test
= next_active_insn_bb (bb
, test
)) != NULL
)
380 /* Found an existing note: nothing to do. */
381 if (find_reg_note (test
, REG_ARGS_SIZE
, NULL_RTX
))
383 /* Found something that affects unwinding. Stop searching. */
384 if (CALL_P (test
) || !insn_nothrow_p (test
))
386 if (next_candidate
== NULL
)
387 next_candidate
= test
;
391 while ((test
= prev_active_insn_bb (bb
, test
)) != NULL
)
394 /* Found a place that seems logical to adjust the stack. */
395 tnote
= find_reg_note (test
, REG_ARGS_SIZE
, NULL_RTX
);
398 XEXP (tnote
, 0) = XEXP (note
, 0);
401 if (prev_candidate
== NULL
)
402 prev_candidate
= test
;
403 /* Found something that affects unwinding. Stop searching. */
404 if (CALL_P (test
) || !insn_nothrow_p (test
))
409 test
= prev_candidate
;
410 else if (next_candidate
)
411 test
= next_candidate
;
414 /* ??? We *must* have a place, lest we ICE on the lost adjustment.
415 Options are: dummy clobber insn, nop, or prevent the removal of
417 /* TODO: Find another way to indicate to the dwarf2 code that we
418 have not in fact lost an adjustment. */
419 test
= emit_insn_before (gen_rtx_CLOBBER (VOIDmode
, const0_rtx
), insn
);
421 add_reg_note (test
, REG_ARGS_SIZE
, XEXP (note
, 0));
424 /* Subroutine of combine_stack_adjustments, called for each basic block. */
427 combine_stack_adjustments_for_block (basic_block bb
)
429 HOST_WIDE_INT last_sp_adjust
= 0;
430 rtx_insn
*last_sp_set
= NULL
;
431 rtx_insn
*last2_sp_set
= NULL
;
432 struct csa_reflist
*reflist
= NULL
;
433 rtx_insn
*insn
, *next
;
435 struct record_stack_refs_data data
;
436 bool end_of_block
= false;
438 for (insn
= BB_HEAD (bb
); !end_of_block
; insn
= next
)
440 end_of_block
= insn
== BB_END (bb
);
441 next
= NEXT_INSN (insn
);
446 set
= single_set_for_csa (insn
);
449 rtx dest
= SET_DEST (set
);
450 rtx src
= SET_SRC (set
);
452 /* Find constant additions to the stack pointer. */
453 if (dest
== stack_pointer_rtx
454 && GET_CODE (src
) == PLUS
455 && XEXP (src
, 0) == stack_pointer_rtx
456 && CONST_INT_P (XEXP (src
, 1)))
458 HOST_WIDE_INT this_adjust
= INTVAL (XEXP (src
, 1));
460 /* If we've not seen an adjustment previously, record
461 it now and continue. */
465 last_sp_adjust
= this_adjust
;
469 /* If not all recorded refs can be adjusted, or the
470 adjustment is now too large for a constant addition,
471 we cannot merge the two stack adjustments.
473 Also we need to be careful to not move stack pointer
474 such that we create stack accesses outside the allocated
475 area. We can combine an allocation into the first insn,
476 or a deallocation into the second insn. We can not
477 combine an allocation followed by a deallocation.
479 The only somewhat frequent occurrence of the later is when
480 a function allocates a stack frame but does not use it.
481 For this case, we would need to analyze rtl stream to be
482 sure that allocated area is really unused. This means not
483 only checking the memory references, but also all registers
484 or global memory references possibly containing a stack
487 Perhaps the best way to address this problem is to teach
488 gcc not to allocate stack for objects never used. */
490 /* Combine an allocation into the first instruction. */
491 if (STACK_GROWS_DOWNWARD
? this_adjust
<= 0 : this_adjust
>= 0)
493 if (try_apply_stack_adjustment (last_sp_set
, reflist
,
494 last_sp_adjust
+ this_adjust
,
498 maybe_move_args_size_note (last_sp_set
, insn
, false);
500 last_sp_adjust
+= this_adjust
;
505 /* Otherwise we have a deallocation. Do not combine with
506 a previous allocation. Combine into the second insn. */
507 else if (STACK_GROWS_DOWNWARD
508 ? last_sp_adjust
>= 0 : last_sp_adjust
<= 0)
510 if (try_apply_stack_adjustment (insn
, reflist
,
511 last_sp_adjust
+ this_adjust
,
515 maybe_move_args_size_note (insn
, last_sp_set
, true);
516 delete_insn (last_sp_set
);
518 last_sp_adjust
+= this_adjust
;
519 free_csa_reflist (reflist
);
525 /* Combination failed. Restart processing from here. If
526 deallocation+allocation conspired to cancel, we can
527 delete the old deallocation insn. */
530 if (last_sp_adjust
== 0)
532 maybe_move_args_size_note (insn
, last_sp_set
, true);
533 delete_insn (last_sp_set
);
536 last2_sp_set
= last_sp_set
;
538 free_csa_reflist (reflist
);
541 last_sp_adjust
= this_adjust
;
545 /* Find a store with pre-(dec|inc)rement or pre-modify of exactly
546 the previous adjustment and turn it into a simple store. This
547 is equivalent to anticipating the stack adjustment so this must
550 && ((STACK_GROWS_DOWNWARD
551 ? (GET_CODE (XEXP (dest
, 0)) == PRE_DEC
553 == (HOST_WIDE_INT
) GET_MODE_SIZE (GET_MODE (dest
)))
554 : (GET_CODE (XEXP (dest
, 0)) == PRE_INC
556 == -(HOST_WIDE_INT
) GET_MODE_SIZE (GET_MODE (dest
))))
557 || ((STACK_GROWS_DOWNWARD
558 ? last_sp_adjust
>= 0 : last_sp_adjust
<= 0)
559 && GET_CODE (XEXP (dest
, 0)) == PRE_MODIFY
560 && GET_CODE (XEXP (XEXP (dest
, 0), 1)) == PLUS
561 && XEXP (XEXP (XEXP (dest
, 0), 1), 0)
563 && GET_CODE (XEXP (XEXP (XEXP (dest
, 0), 1), 1))
565 && INTVAL (XEXP (XEXP (XEXP (dest
, 0), 1), 1))
567 && XEXP (XEXP (dest
, 0), 0) == stack_pointer_rtx
568 && !reg_mentioned_p (stack_pointer_rtx
, src
)
569 && memory_address_p (GET_MODE (dest
), stack_pointer_rtx
)
570 && try_apply_stack_adjustment (insn
, reflist
, 0,
574 maybe_move_args_size_note (last2_sp_set
, last_sp_set
, false);
576 maybe_move_args_size_note (insn
, last_sp_set
, true);
577 delete_insn (last_sp_set
);
578 free_csa_reflist (reflist
);
587 data
.reflist
= reflist
;
588 if (!CALL_P (insn
) && last_sp_set
589 && !for_each_rtx (&PATTERN (insn
), record_stack_refs
, &data
))
591 reflist
= data
.reflist
;
594 reflist
= data
.reflist
;
596 /* Otherwise, we were not able to process the instruction.
597 Do not continue collecting data across such a one. */
600 || reg_mentioned_p (stack_pointer_rtx
, PATTERN (insn
))))
602 if (last_sp_set
&& last_sp_adjust
== 0)
604 force_move_args_size_note (bb
, last2_sp_set
, last_sp_set
);
605 delete_insn (last_sp_set
);
607 free_csa_reflist (reflist
);
615 if (last_sp_set
&& last_sp_adjust
== 0)
617 force_move_args_size_note (bb
, last2_sp_set
, last_sp_set
);
618 delete_insn (last_sp_set
);
622 free_csa_reflist (reflist
);
626 rest_of_handle_stack_adjustments (void)
628 df_note_add_problem ();
630 combine_stack_adjustments ();
636 const pass_data pass_data_stack_adjustments
=
640 OPTGROUP_NONE
, /* optinfo_flags */
641 TV_COMBINE_STACK_ADJUST
, /* tv_id */
642 0, /* properties_required */
643 0, /* properties_provided */
644 0, /* properties_destroyed */
645 0, /* todo_flags_start */
646 TODO_df_finish
, /* todo_flags_finish */
649 class pass_stack_adjustments
: public rtl_opt_pass
652 pass_stack_adjustments (gcc::context
*ctxt
)
653 : rtl_opt_pass (pass_data_stack_adjustments
, ctxt
)
656 /* opt_pass methods: */
657 virtual bool gate (function
*);
658 virtual unsigned int execute (function
*)
660 return rest_of_handle_stack_adjustments ();
663 }; // class pass_stack_adjustments
666 pass_stack_adjustments::gate (function
*)
668 /* This is kind of a heuristic. We need to run combine_stack_adjustments
669 even for machines with possibly nonzero TARGET_RETURN_POPS_ARGS
670 and ACCUMULATE_OUTGOING_ARGS. We expect that only ports having
671 push instructions will have popping returns. */
672 #ifndef PUSH_ROUNDING
673 if (ACCUMULATE_OUTGOING_ARGS
)
676 return flag_combine_stack_adjustments
;
682 make_pass_stack_adjustments (gcc::context
*ctxt
)
684 return new pass_stack_adjustments (ctxt
);