[Bug tree-optimization/64024] [5 Regression] gcc.dg/vect/vect-simd-clone-6.c ICEs

["rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64024

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #6)
> (In reply to Richard Biener from comment #5)
> > Generally calling SCEV analysis again during the transform phase asks for
> > trouble
> > (though it may work in most cases).  This means that the simple_iv calls
> > (which
> > are mostly useless as computed stuff isn't actually used in most cases)
> > should
> > be avoided at transform time.
> > 
> > Didn't look into this specific case but can do so (just assign the bug to me
> > then).
> 
> I can handle the storing of the op/linear_step combo somewhere, just don't
> know how to adjust it.  Shall vect_update_inits_of_drs update those, or can
> it e.g. stick the tree somewhere in the LOOP_VINFO (or its caller) and
> vect_simd_clone_call use it?

Not sure, but "linear_step" sounds like sth available already via
STMT_VINFO_LOOP_PHI_EVOLUTION_PART?  That is, you could store it there
(on the IV def stmt vinfo).  Possibly adding a new access macro that doesn't
have "PHI" in its name (and adjusting the comment before the stmt vinfo
field).