public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
* [Bug middle-end/110378] New: SRA for destructors
@ 2023-06-23 16:14 hubicka at gcc dot gnu.org
  2023-06-23 16:15 ` [Bug middle-end/110378] " hubicka at gcc dot gnu.org
                   ` (9 more replies)
  0 siblings, 10 replies; 11+ messages in thread
From: hubicka at gcc dot gnu.org @ 2023-06-23 16:14 UTC (permalink / raw)
  To: gcc-bugs

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110378

            Bug ID: 110378
           Summary: SRA for destructors
           Product: gcc
           Version: 13.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hubicka at gcc dot gnu.org
  Target Milestone: ---

In thestcase from PR109849 built with profile feedback we get the following
cleanup:

<L10>:
  D.28322 ={v} {CLOBBER};
  cur ={v} {CLOBBER(eol)};
  std::vector<std::pair<unsigned int, unsigned int> >::~vector (&stack);
  _28 = __builtin_eh_pointer (2);
  __builtin_unwind_resume (_28);

which calls destructor of std::vector. This destructor compiles to:

void std::vector<std::pair<unsigned int, unsigned int> >::~vector (struct
vector * const this)
{
  struct pair * _6;
  struct pair * _7;
  long int _8;
  long unsigned int _15;

;;   basic block 2, loop depth 0, count 10000 (precise)
;;    prev block 0, next block 3, flags: (NEW, REACHABLE, VISITED)
;;    pred:       ENTRY [always]  count:10000 (precise) (FALLTHRU,EXECUTABLE)
  _6 = MEM[(struct _Vector_base *)this_3(D)]._M_impl.D.25963._M_end_of_storage;
  _7 = MEM[(struct _Vector_base *)this_3(D)]._M_impl.D.25963._M_start;
  _8 = _6 - _7;
  if (_7 != 0B)
    goto <bb 3>; [100.00%]
  else
    goto <bb 4>; [0.00%]
;;    succ:       3 [always]  count:10000 (precise) (TRUE_VALUE,EXECUTABLE)
;;                4 [never]  count:0 (precise) (FALSE_VALUE,EXECUTABLE)

;;   basic block 3, loop depth 0, count 10000 (precise)
;;    prev block 2, next block 4, flags: (NEW, REACHABLE, VISITED)
;;    pred:       2 [always]  count:10000 (precise) (TRUE_VALUE,EXECUTABLE)
  _15 = (long unsigned int) _8;
  operator delete (_7, _15); [tail call]
;;    succ:       4 [always]  count:10000 (precise) (FALLTHRU,EXECUTABLE)

;;   basic block 4, loop depth 0, count 10000 (precise)
;;    prev block 3, next block 1, flags: (NEW, REACHABLE, VISITED)
;;    pred:       2 [never]  count:0 (precise) (FALSE_VALUE,EXECUTABLE)
;;                3 [always]  count:10000 (precise) (FALLTHRU,EXECUTABLE)
  return;
;;    succ:       EXIT [always]  count:10000 (precise) (EXECUTABLE)
/usr/include/c++/13/bits/stl_vector.h:735:7

}

and it is used twice in code that is considered cold. Since the code itself is
longer than the call of operator delete, it makes sense to not inline it.

Before ipa-cp it is:
Released 8 names, 53.33%, removed 8 holes
void std::vector<std::pair<unsigned int, unsigned int> >::~vector (struct
vector * const this)
{
  struct pair * _1;
  struct pair * _2;
  struct pair * _6;
  struct pair * _7;
  long int _8;
  long int _9;
  long unsigned int _10;
  struct _Vector_impl * _11;

;;   basic block 2, loop depth 0, count 1073741824 (estimated locally), maybe
hot
;;    prev block 0, next block 3, flags: (NEW, REACHABLE, VISITED)
;;    pred:       ENTRY [always]  count:1073741824 (estimated locally)
(FALLTHRU,EXECUTABLE)
  _1 = this_3(D)->D.26656._M_impl.D.25963._M_finish;
  _2 = this_3(D)->D.26656._M_impl.D.25963._M_start;
  std::_Destroy<std::pair<unsigned int, unsigned int>*> (_2, _1);
  _6 = MEM[(struct _Vector_base *)this_3(D)]._M_impl.D.25963._M_end_of_storage;
  _7 = MEM[(struct _Vector_base *)this_3(D)]._M_impl.D.25963._M_start;
  _8 = _6 - _7;
  _9 = _8 /[ex] 8;
  _10 = (long unsigned int) _9;
  if (_7 != 0B)
    goto <bb 3>; [53.47%]
  else
    goto <bb 4>; [46.53%]
;;    succ:       3 [53.5% (guessed)]  count:574129752 (estimated locally)
(TRUE_VALUE,EXECUTABLE)
;;                4 [46.5% (guessed)]  count:499612072 (estimated locally)
(FALSE_VALUE,EXECUTABLE)

;;   basic block 3, loop depth 0, count 574129753 (estimated locally), maybe
hot
;;    prev block 2, next block 4, flags: (NEW, REACHABLE, VISITED)
;;    pred:       2 [53.5% (guessed)]  count:574129752 (estimated locally)
(TRUE_VALUE,EXECUTABLE)
  _11 = &MEM[(struct _Vector_base *)this_3(D)]._M_impl;
  std::__new_allocator<std::pair<unsigned int, unsigned int> >::deallocate
(_11, _7, _10);
;;    succ:       4 [always]  count:574129753 (estimated locally)
(FALLTHRU,EXECUTABLE) /usr/include/c++/13/bits/alloc_traits.h:516:35

;;   basic block 4, loop depth 0, count 1073741824 (estimated locally), maybe
hot
;;    prev block 3, next block 1, flags: (NEW, REACHABLE, VISITED)
;;    pred:       2 [46.5% (guessed)]  count:499612072 (estimated locally)
(FALSE_VALUE,EXECUTABLE)
;;                3 [always]  count:574129753 (estimated locally)
(FALLTHRU,EXECUTABLE) /usr/include/c++/13/bits/alloc_traits.h:516:35
  *this_3(D) ={v} {CLOBBER};
  return;
;;    succ:       EXIT [always]  count:1073741824 (estimated locally)
(EXECUTABLE) /usr/include/c++/13/bits/stl_vector.h:735:7

}

I think it is the clobber that prevents ipa-sra :)

^ permalink raw reply	[flat|nested] 11+ messages in thread
end of thread, other threads:[~2023-10-03 16:53 UTC | newest]

Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-06-23 16:14 [Bug middle-end/110378] New: SRA for destructors hubicka at gcc dot gnu.org
2023-06-23 16:15 ` [Bug middle-end/110378] " hubicka at gcc dot gnu.org
2023-07-31 15:25 ` [Bug middle-end/110378] IPA-SRA " jamborm at gcc dot gnu.org
2023-07-31 16:02 ` [Bug ipa/110378] " jamborm at gcc dot gnu.org
2023-07-31 16:06 ` jamborm at gcc dot gnu.org
2023-07-31 16:08 ` jamborm at gcc dot gnu.org
2023-08-01 12:45 ` jamborm at gcc dot gnu.org
2023-08-07 17:15 ` cvs-commit at gcc dot gnu.org
2023-08-08 16:20 ` clyon at gcc dot gnu.org
2023-08-08 16:20 ` clyon at gcc dot gnu.org
2023-10-03 16:53 ` cvs-commit at gcc dot gnu.org

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).