https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107690
--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> --- We're vectorizing it as uncounted loop, that's hardly a progression. niter analysis only handles INTEGER_TYPE and POINTER_TYPE_P exit conditions, so we probably want to disable if-combining into loop exit conditions. Alternatively niter analysis could treat bit-and of two conditions as two separate analyses and then attempt to merge the niter results somehow (not exactly sure how or how useful that will be). There's the possibility the user wrote things this way. We probably want to have a separate but tracking the if-combining issue and one tracking niter analysis enhancement. I'm testing a fix for ifcombine as heuristic.
