https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85393
--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Actually, PRE does the right thing; the ~S () call is in bb3 and there is EH edge from there to bb23, which has another EH predecessor, and PRE inserts the initialization of pseudo 123 in that bb23. Then comes bbpart, determines that one of the EH edges is crossing and duplicates the bb (.gcc_except_table contains expressions relative to the start of the partition in the current function, so EH edges need to be non-crossing). The pseudo 123 initialization is not duplicated though. So, likely bbpart bug then.