[Bug target/67288] [6/7/8 regression] non optimal simple function (useless additional shift/remove/shift/add)

[segher at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67288

--- Comment #13 from Segher Boessenkool <segher at gcc dot gnu.org> ---
This setup code is created *by* the RTL loop optimisers, via
iv_number_of_iterations IIRC.  Combine cannot get rid of it because it does
not know the count cannot be 0 (that is info from a different bb).

Yes I think we want some extra hooks so the target can tune how expensive
loops are, based on exit condition, if it is the inner loop, etc.