https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122587

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
So with this we can avoid maybe-zero checks, I think there's still improvements
possible but with any range we already get an appropriate upper bound for the
loops so most interesting bits for vectorization are already handled.

Some power-of-two checks could be improved, likewise expected prolog/epilog
iterations for costing.

We'd want to track and update a value_range niter_range besides the
tree niter variable.  That might be useful to simplify the code
computing and setting upper bounds on epilog loops.

Reply via email to