https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128071
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2026-10-23
Status|UNCONFIRMED |NEW
Ever confirmed|0 |1
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
The issue is that for BB vectorization we cannot determine
t.f90:11:30: missed: can't determine dependence between MEM[(real(kind=8)[8]
*)_37][0] and MEM[(real(kind=8)[8] *)_38][1]
t.f90:11:30: note: removing SLP instance operations starting from:
MEM[(real(kind=8)[8] *)_37][0] = _47;
that's the a(i)%v store vs the b(i)%v load when attempting to sink the
store across the loads:
MEM[(real(kind=8)[8] *)_37][0] = _47;
_53 = MEM[(real(kind=8)[8] *)_37][1];
_54 = MEM[(real(kind=8)[8] *)_38][1];
but when vectorizing the loop
_17 = MEM[(real(kind=8)[8] *)_37][_16];
_18 = MEM[(real(kind=8)[8] *)_38][_16];
_19 = _17 + _18;
MEM[(real(kind=8)[8] *)_37][_16] = _19;
we can (and do) add a runtime alias check. That's not possible when
doing BB vectorization.
So the issue is indeed that early unrolling defeats vectorization here.
