[Bug rtl-optimization/103006] [9/10/11/12 Regression] wrong code at -O1 or -O2 on x86_64-linux-gnu by r7-7101

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103006

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |msebor at gcc dot gnu.org,
                   |                            |rsandifo at gcc dot gnu.org

--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> ---
So I have a patch that adds explicit birth markers (using clobbers specially
marked).  That works well sofar but it conflicts with clobbers (not marked as
birth) that are added for clobbering at the start of variable lifetime like
C++ does at the beginning of CTORs.  I for example see

  inst ={v} {CLOBBER(birth)};
  inst ={v} {CLOBBER};  (**)
  inst.v = 42;
...
  inst ={v} {CLOBBER};

where (**) is inserted by the C++ frontend (with -flifetime-dse which is
the default).  Indeed my life analysis for the purpose of stack slot
sharing now only relies on the birth/death markers so it gets confused
by the extra clobber.

We now also have some use-after-free diagnostic that would likely trip
over this as it assumes that a CLOBBER ends lifetime of storage.

I guess disentangling both use-cases by also marking the
end-of-storage-lifetime
clobbers specially would solve both issues.