https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117113
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
Status|NEW |ASSIGNED
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
So the issue is that we have an outer loop with a virtual PHI and an inner loop
with just a load. When unrolling this there's virtual definitions
(from PHIs of the outer loop, but in general also from stores if there
were any) between the two inner loop copies. fuse_loops () then goes along
merging those blocks (with the VDEFs) but it does not add a loop PHI node
to the inner loop covering the virtual definitions - and that's what is
missing.
Probably a bit artificial because of -fno-tree-dce but not entirely
impossible that a stray loop header PHI for virtuals is around.
I have a patch to reject unroll-and-jam in this situation.
