[Bug c++/120955] [15/16 Regression] 50 % increase in data segment size on avr-gcc for -Os

[gjl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120955

--- Comment #36 from Georg-Johann Lay <gjl at gcc dot gnu.org> ---
(In reply to Georg-Johann Lay from comment #35)
> (In reply to fiesh from comment #34)
> > (In reply to Georg-Johann Lay from comment #33)
> > > avr.cc::avr_use_by_pieces_infrastructure_p()
> > 
> > Thank you very much, that's precisely what I was looking for.
> The problem with that hook is that it is not telling whether the source
> values are known at link time, in which case no load+store is needed but a
> load-immediate is enough.  So that hook cannot return correct costs.
That is, the hook is called with op=MOVE_BY_PIECES even though all values can
be loaded with load-immediate.  So why isn't the hook invoked with
op=SET_BY_PIECES?