[Bug libstdc++/100466] New: compilation of assignment from initialization list to std::array<T, N> with non-trivial constructor of T is very slow

[Bug libstdc++/100466] New: compilation of assignment from initialization list to std::array<T, N> with non-trivial constructor of T is very slow

[xry111 at mengyan1223 dot wang]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100466

            Bug ID: 100466
           Summary: compilation of assignment from initialization list to
                    std::array<T, N> with non-trivial constructor of T is
                    very slow
           Product: gcc
           Version: 11.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libstdc++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: xry111 at mengyan1223 dot wang
  Target Milestone: ---

Code:

#include <array>

struct t
{
        int a;
        t() : a(0) {}
};

std::array<t, X> g;

int main()
{
        g = {};
}

$ for i in 10 100 1000 10000 100000; do
>   time g++ t.cc -DX=$i
> done
g++ t.cc -DX=$i  0.74s user 0.06s system 99% cpu 0.798 total
g++ t.cc -DX=$i  0.74s user 0.05s system 99% cpu 0.794 total
g++ t.cc -DX=$i  0.78s user 0.06s system 99% cpu 0.833 total
g++ t.cc -DX=$i  1.08s user 0.08s system 99% cpu 1.167 total
g++ t.cc -DX=$i  4.55s user 0.25s system 99% cpu 4.809 total

With -O2, it's much worse:

$ for i in 10 100 1000 10000 100000; do
>   time g++ t.cc -DX=$i -O2
> done
g++ t.cc -DX=$i -O2  0.07s user 0.01s system 99% cpu 0.089 total
g++ t.cc -DX=$i -O2  0.08s user 0.01s system 99% cpu 0.087 total
g++ t.cc -DX=$i -O2  0.49s user 0.01s system 99% cpu 0.503 total
g++ t.cc -DX=$i -O2  6.34s user 0.03s system 99% cpu 6.376 total
g++ t.cc -DX=$i -O2  68.91s user 0.18s system 99% cpu 1:09.09 total

And the resulted code seems "unrolling a loop" too excessively:

  401040:       48 c7 05 95 aa 0e 00    movq   $0x0,0xeaa95(%rip)        #
4ebae0 <g>
  401047:       00 00 00 00 
  40104b:       31 c0                   xor    %eax,%eax
  40104d:       48 c7 05 90 aa 0e 00    movq   $0x0,0xeaa90(%rip)        #
4ebae8 <g+0x8>
  401054:       00 00 00 00 
  401058:       48 c7 05 8d aa 0e 00    movq   $0x0,0xeaa8d(%rip)        #
4ebaf0 <g+0x10>
  40105f:       00 00 00 00 
  401063:       48 c7 05 8a aa 0e 00    movq   $0x0,0xeaa8a(%rip)        #
4ebaf8 <g+0x18>
  40106a:       00 00 00 00 
  ... ...
  4874a7:       48 c7 05 a6 60 0c 00    movq   $0x0,0xc60a6(%rip)        #
54d558 <g+0x61a78>
  4874ae:       00 00 00 00

[Bug libstdc++/100466] compilation of assignment from initialization list to std::array<T, N> with non-trivial constructor of T is very slow

[xry111 at mengyan1223 dot wang]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100466

--- Comment #1 from Xi Ruoyao <xry111 at mengyan1223 dot wang> ---
clang-12 handles this correctly.

[Bug libstdc++/100466] compilation of assignment from initialization list to std::array<T, N> with non-trivial constructor of T is very slow

[xry111 at mengyan1223 dot wang]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100466

Xi Ruoyao <xry111 at mengyan1223 dot wang> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|11.1.0                      |12.0
      Known to fail|                            |10.3.0, 11.1.0, 12.0,
                   |                            |8.3.0, 9.3.0

--- Comment #2 from Xi Ruoyao <xry111 at mengyan1223 dot wang> ---
Updated version, based on the test result on godbolt.

[Bug libstdc++/100466] compilation of assignment from initialization list to std::array<T, N> with non-trivial constructor of T is very slow

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100466

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |compile-time-hog

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think that's an old known C++ FE issue which should emit a loop for inits
instead of repeating assignments for each element.

[Bug c++/100466] compilation of assignment from initialization list to an object with array member T[N] and non-trivial constructor of T is very slow

[xry111 at mengyan1223 dot wang]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100466

Xi Ruoyao <xry111 at mengyan1223 dot wang> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|libstdc++                   |c++
            Summary|compilation of assignment   |compilation of assignment
                   |from initialization list to |from initialization list to
                   |std::array<T, N> with       |an object with array member
                   |non-trivial constructor of  |T[N] and non-trivial
                   |T is very slow              |constructor of T is very
                   |                            |slow

--- Comment #4 from Xi Ruoyao <xry111 at mengyan1223 dot wang> ---
(In reply to Richard Biener from comment #3)
> I think that's an old known C++ FE issue which should emit a loop for inits
> instead of repeating assignments for each element.

Change summary and component then.

Is this a dup?  I tried to search an existing PR but couldn't find it.

[Bug c++/100466] compilation of assignment from initialization list to an object with array member T[N] and non-trivial constructor of T is very slow

[xry111 at mengyan1223 dot wang]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100466

Xi Ruoyao <xry111 at mengyan1223 dot wang> changed:

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

--- Comment #5 from Xi Ruoyao <xry111 at mengyan1223 dot wang> ---
Alright, there is PR77443 and PR84281.  But the "fix" to PR84281 only fixed the
memory issue, the compile time is still too excessive.

I'll mark it as a dup of PR77443.

*** This bug has been marked as a duplicate of bug 77443 ***