https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116214
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> --- It's a false positive, the init gate is 'costing_p' we have if (!costing_p) stride_step = ...; for (...) for (...) { if (costing_p) continue; ... use of stride_step ...