https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92462
--- Comment #16 from Wilco <wilco at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #15)
> I can't find PRE doing anything wrong and on 32bit x86_64 the testcase
> executes
> correctly with GCC 7.3 and GCC 9 (when I add the missing return to
> Bar::cmpxchg).
> So -ftree-pre, if it triggers a bug, triggers it elsewhere, the bug isn't in
> PRE itself AFAICS.
Well the difference is that one of the loads is removed by dse1:
**scanning insn=26
cselib lookup (reg/f:SI 102 sfp) => 8:8
cselib lookup (reg/f:SI 126) => 9:4339
cselib lookup (reg/f:SI 102 sfp) => 8:8
cselib lookup (reg/f:SI 102 sfp) => 8:8
cselib lookup (plus:SI (reg/f:SI 102 sfp)
(const_int -8 [0xfffffffffffffff8])) => 9:4339
mem: (plus:SI (reg/f:SI 102 sfp)
(const_int -8 [0xfffffffffffffff8]))
after canon_rtx address: (plus:SI (reg/f:SI 102 sfp)
(const_int -8 [0xfffffffffffffff8]))
gid=1 offset=-8
processing const load gid=1[-8..-4)
trying to replace SImode load in insn 26 from SImode store in insn 16
deferring rescan insn with uid = 26.
deferring rescan insn with uid = 77.
-- replaced the loaded MEM with (reg 144)
mems_found = 0, cannot_delete = true
cselib lookup (mem/c:SI (plus:SI (reg/f:SI 102 sfp)
(const_int -8 [0xfffffffffffffff8])) [1 curD.6314+0 S4 A64]) => 0:0
