[Bug ipa/104303] [12 regression] gnatmake is miscompiled by IPA/modref

[hubicka at kam dot mff.cuni.cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104303

--- Comment #10 from hubicka at kam dot mff.cuni.cz ---
Hi,
> 
> I'm testing this, adding escaped_local_p parameters throughout the API and
> adjusting consumers, throwing away ref_may_access_global_memory_p.
> 
> Following passes adjusted from false to true:
> 
> - DCE in mark_stmt_if_obviously_necessary but that's just compile-time
>   optimization I think
> - trans-mem.cc I think got it wrong, so now fixed (but trans-mem is dead)
> - the modref query in modref_may_conflict
> 
> all others keep passing false.

Thanks!  I was adding the function when introducing global memory logic
to modref. This was targeted to handle fatigue regression at -O2 but it
also helps in quite few extra cases.

Honza