[Bug tree-optimization/69873] Vectorizer fails to emit runtime profitability check if no peeling/versioning is done

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69873

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
It's difficult to find a testcase that we'd vectorize with a runtime
profitability check but whithout any versioning or peeling because we should
always be able to statically decide profitability there - more vector
iterations
shouldn't be able to offset any outside cost on the vector path because there
is none (well, an "excessive" number of emitted splats maybe - but that starts
to be quite a bit tricky).