https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102510
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2021-09-28
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Known to fail| |12.0
Keywords| |missed-optimization
Summary|Function call has |Function call has
|unnecessary aliasing check |unnecessary stride check
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
It's not an alias but we seem to not know that the array stride of the result
is one and thus we version on that. I'm not sure whether fortran allows
res(/1, 16, 2/) = add2vecs2(a, b)
or so or why we know the inputs are not strided. The result comes in as
array descriptor while the other arguments are pointers to a flat [8] array.
Adding 'contiguous' yields
Error: 'add2vecs2' at (1) has the CONTIGUOUS attribute but is not an array
pointer or an assumed-shape or assumed-rank array
so maybe it's really just a missed optimization in the Frontend?
