The following program causes an ICE. With an older version of the 4.5 trunk it worked thus it is a true regression.
$ gfortran -Wall -c -flto -O1 test.f $ ${PATH}/libexec/gcc/x86_64-unknown-linux-gnu/4.5.0/lto1 -o /dev/null ylm4.o Performing interprocedural optimizations <whole-program> <inline>Assembling functions: ylm4 In file included from test.f:5:0, from :0: test.f: In function 'ylm4': test.f:9:0: internal compiler error: in expand_mult, at expmed.c:3253 SUBROUTINE ylm4(ylm) COMPLEX, INTENT (OUT):: ylm(1) INTEGER l,m COMPLEX ylms REAL, ALLOCATABLE, SAVE :: ynorm(:) ylms = 0 DO m = 1, 1 DO l = m, 1 ylm(m) = conjg(ylms)*ynorm(m) ENDDO ENDDO END SUBROUTINE ylm4 -- Summary: [4.5 Regression] ICE with lto1 in expand_mult, at expmed.c:3253 Product: gcc Version: 4.5.0 Status: UNCONFIRMED Keywords: ice-on-valid-code Severity: normal Priority: P3 Component: middle-end AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: burnus at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42678