https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128357

Jerry DeLisle <jvdelisle at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |jvdelisle at gcc dot 
gnu.org

--- Comment #5 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> ---
I had to do some head scratching on this.

Steve, you are right that two problems are mixed here; I will split
them.  The wrong-code part is independent of any leak:

  call sq ((q), zs)

passes q itself rather than a copy, so the dummy x aliases z and
changes when z is assigned (STOP 4 on trunk).  Making a parenthesized
scalar CLASS variable produce a copy will get its own PR.

On the leaks I agree with Mikael: these are compiler temporaries that
live only for the call, not variables of the main program.  With each
statement moved into a subroutine whose variables are all local, and
the subroutine called 1000 times, trunk loses:

  call s ((ta))           144,000 bytes in 1,000 blocks
                          (the array temporary itself)
  call sc ((qa))           12,000 bytes in 2,000 blocks
  associate (b => (qa))    12,000 bytes in 2,000 blocks
                          (allocatable components of the deep copies)

I will submit a patch for this one and open a new PR for the CLASS problem

Reply via email to