[Bug libstdc++/103621] New: stable_sort could call std::__merge_sort_with_buffer directly in typical case

[Bug libstdc++/103621] New: stable_sort could call std::__merge_sort_with_buffer directly in typical case

[riad93 at mail dot ru]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103621

            Bug ID: 103621
           Summary: stable_sort could call std::__merge_sort_with_buffer
                    directly in typical case
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libstdc++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: riad93 at mail dot ru
  Target Milestone: ---

So, currently __stable_sort is implemented this way:

if (__buf.begin() == 0)
  std::__inplace_stable_sort(__first, __last, __comp);
else
  std::__stable_sort_adaptive(__first, __last, __buf.begin(),
                                    _DistanceType(__buf.size()), __comp);


which means that on the topmost iteration the merge the split will be done in a
naive way, and the copy+merge will happen even in typical case when there's
enough memory for everything. 

Subjectively, it makes sense that in general calling
std::__merge_sort_with_buffer should be faster (otherwise why would it exist),
but I wasn't able to measure any difference yet.

[Bug libstdc++/103621] stable_sort could call std::__merge_sort_with_buffer directly in typical case

[fdumont at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103621

François Dumont <fdumont at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |13.0
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |fdumont at gcc dot gnu.org
         Resolution|---                         |FIXED

--- Comment #1 from François Dumont <fdumont at gcc dot gnu.org> ---
I think it has been fixed by:
commit 63d182fb86e47323ac50d9368845d712e1f7da89
Author: François Dumont <fdumont@gcc.gnu.org>
Date:   Thu Jan 21 19:30:47 2021 +0100

    libstdc++: Enhance branching in std::inplace_merge and std::stable_sort

    When we manage to allocate a buffer of the expected size we can simplify
the code to
    perform the expected algorithm.

    libstdc++-v3/ChangeLog:

            * include/bits/stl_algo.h
            (__merge_adaptive): Adapt to merge only when buffer is large
enough..
            (__merge_adaptive_resize): New, adapt merge when buffer is too
small.
            (__inplace_merge): Adapt, use latter.
            (__stable_sort_adaptive): Adapt to sort only when buffer is large
enough.
            (__stable_sort_adaptive_resize): New, adapt sort when buffer is too
small.
            (__stable_sort): Adapt, use latter.


Let us know otherwise.