[Bug fortran/60549] [4.9 Regression] Run time doubled for fatigue.f90 due to SAVE changes

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60549

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Can't you just invent some other SAVE flag for the MAIN__ variables, such that
it would make TREE_STATIC only very large vars or where it is absolutely
required that they are TREE_STATIC?  If there is no way how a program could
find out (other than just comparing addresses and seeing if it falls into the
stack segment) whether a MAIN__ variable is really SAVEd or not, I think it
would be better to allocate them as automatic variables.