http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57073
--- Comment #10 from Thomas Koenig <tkoenig at gcc dot gnu.org> --- Created attachment 30142 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30142&action=edit Another patch that doesn't work... This time with the right PR... I tried to follow the advice in comment#9, but I hit a wall (again). With the attached patch, I hit foo.f90:1:0: error: SSA_NAME_DEF_STMT is wrong program main ^ Expected definition statement: _16 = 1.0e+0; Actual definition statement: _16 = _24; so I suspect I need PHI nodes here. Is this the right approach in general? Am I missing something simple here?