https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69368
--- Comment #48 from rguenther at suse dot de <rguenther at suse dot de> --- On Thu, 18 Feb 2016, rguenther at suse dot de wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69368 > > --- Comment #46 from rguenther at suse dot de <rguenther at suse dot de> --- > On Thu, 18 Feb 2016, jakub at gcc dot gnu.org wrote: > > > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69368 > > > > --- Comment #45 from Jakub Jelinek <jakub at gcc dot gnu.org> --- > > Consider e.g. one source file > > INTEGER FOO > > COMMON /BLK/ K(64) > > DO I=1,64 > > K(I)=1 > > END DO > > IF (FOO(5,12).NE.51) CALL ABORT > > END > > and another one: > > FUNCTION FOO(I, J) > > COMMON /BLK/ K(1) > > FOO = K(I) + K(J) + K(2*I) + K(2*J) > > END FUNCTION > > If the second source file is compiled with -O2, it will return "wrong" > > value, > > as if all accesses to K were using the same index (because the only valid > > index > > is 1). If you compile with -O2 -fno-tree-dominator-opts, then it "works". > > Which is odd since FRE already should optimize it that way. Ok, it does not. Supposedly more complicated situations are needed to trigger sth wrong in FRE.
