[Bug c++/94957] New: Compilation slowww for simple code with -O1/2/3 and -g in GCC 8 and 9

[Bug c++/94957] New: Compilation slowww for simple code with -O1/2/3 and -g in GCC 8 and 9

[hehaochen at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94957

            Bug ID: 94957
           Summary: Compilation slowww for simple code with -O1/2/3 and -g
                    in GCC 8 and 9
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hehaochen at hotmail dot com
  Target Milestone: ---

code:

---------------------------------

class MyObject
    {
    public:
        MyObject() {}
    };

    class MyClassSlow
    {
        MyObject array[30000] {}; 
    };


    int main()
    {
        MyClassSlow slow;
    }

---------------------------------


run : 
   g++ -Ox -g -std=c++11 test.cpp
in
  gcc-7 gcc-8 gcc-9 with -O1/2/3 -g


#############
##
##   Hangs "forever"
##   gcc version 9.2.0 (GCC)
##
#############

  -O3

    real    0m1.299s
    user    0m0.910s
    sys 0m0.197s

  -O3 -g

    Forever... (kill after 15min)


#############
##
##   Hangs "forever"
##   gcc version 8.2.0 (GCC)
##
#############

  -O3

    real    0m1.434s
    user    0m1.062s
    sys     0m0.229s

  -O2 -g

    Forever... (kill after 15min)

#############
##
## Acceptable but still slow in
##   gcc version 7.4.0 (GCC)
##
#############


  -O0 -g 

    real    0m 3.892s
    user    0m 3.404s
    sys     0m 0.482s


  -O1 -g

    real    1m 6.103s
    user    1m 5.718s
    sys     0m 0.302s


  -O2 -g

    real    1m 23.666s
    user    1m 23.310s
    sys     0m 0.280s


  -O3

    real    0m 1.451s
    user    0m 1.229s
    sys     0m 0.221s

  -O3 -g

    real    1m 19.463s
    user    1m 19.037s
    sys     0m 0.349s


Execution times (seconds)
 phase opt and generate  :  84.89 (99%) usr   1.88 (90%) sys  86.89 (99%) wall 
139631 kB (65%) ggc
 trivially dead code     :  82.12 (96%) usr   0.00 ( 0%) sys  82.26 (94%) wall 
     0 kB ( 0%) ggc



#############
##
## OK in clang
##
#############

clang version 9.0.0-svn366056

root@c5c7aa8003d9:/downloads# time  clang++-9 -O3 -g test.cpp

  real    0m 0.176s
  user    0m 0.103s
  sys     0m 0.073s

root@c5c7aa8003d9:/downloads# time  clang++-9 -O3 test.cpp

  real    0m 0.165s
  user    0m 0.102s
  sys     0m 0.063s

[Bug c++/94957] Compilation slowww for simple code with -O1/2/3 and -g in GCC 8 and 9

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94957

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |jason at gcc dot gnu.org,
                   |                            |mpolacek at gcc dot gnu.org
            Version|unknown                     |9.0

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Seems var-tracking, we end up with -O2 -g with main containing just debug
stmts,
  # DEBUG BEGIN_STMT
  # DEBUG thisD.212422 => NULL
  # DEBUG thisD.212423 => NULL
  # DEBUG thisD.212424 => NULL
...
  # DEBUG thisD.242421 => NULL
  # DEBUG thisD.212421 => NULL
  return 0;
for those 30000 this pointers of the ctor, and var-tracking is quite slow on
that indeed.
Though, code-generation-wise, I think that for these larger arrays we shouldn't
emit
  _1 = &this->array[0];
  MyObject::MyObject (_1);
  _2 = &this->array[1];
  MyObject::MyObject (_2);
  _3 = &this->array[2];
  MyObject::MyObject (_3);
...
  _29997 = &this->array[29996];
  MyObject::MyObject (_29997);
  _29998 = &this->array[29997];
  MyObject::MyObject (_29998);
  _29999 = &this->array[29998];
  MyObject::MyObject (_29999);
  _30000 = &this->array[29999];
  MyObject::MyObject (_30000);
but instead just a loop over the array elts initializing them, because e.g. if
everything doesn't get inlined, it will be much shorter.

[Bug c++/94957] Compilation slowww for simple code with -O1/2/3 and -g in GCC 8 and 9

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94957

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Plenty of dups for this in bugzilla - but FE folks never get that idea of using
a loop ...

[Bug c++/94957] Compilation slowww for simple code with -O1/2/3 and -g in GCC 8 and 9

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94957

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Some of it changed recently, e.g. when the FEs use ARRAY_RANGE_REFs in the
initializer the gimplifier's gimplify_init_ctor_eval emits a loop.
But in this case I think we need the FE to emit the loop itself.
Marek, would you like to have a look?
I'd use serial code if the array doesn't have too many elts (perhaps compare to
a param, or hardcode something small like 16).

[Bug c++/94957] Compilation slowww for simple code with -O1/2/3 and -g in GCC 8 and 9

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94957

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |mpolacek at gcc dot gnu.org
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2020-05-07

--- Comment #4 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
I'll try to play with this in GCC 11.

[Bug c++/94957] Compilation slowww for simple code with -O1/2/3 and -g in GCC 8 and 9

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94957

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ilord.tiran at yandex dot ru

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 98547 has been marked as a duplicate of this bug. ***

[Bug c++/94957] Compilation slowww for simple code with big array of structs with constructors

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94957

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |DUPLICATE

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Dup of bug 92385.

*** This bug has been marked as a duplicate of bug 92385 ***