https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117738
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Last reconfirmed| |2024-11-25
Status|UNCONFIRMED |NEW
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
One of the implementation reasons is that we do not support niter peeling for
the outer loop at this point. The other reason is that sum[j] in the inner
loop conflicts with different iterations in the outer loop - we do not
currently handle this situation specially.
Then there's the issue that with constant outer bound
we apply unroll-and-jam and that confuses us with a duplicate store to
sum[j].
So the biggest roadblock is to incrementally relax the restrictions
on grouped accesses for outer loop vect in vect_analyze_data_ref_access.
IIRC the actual issues are subtle.
