> Both tests are protected by there being exactly one latch in the loop (we > exit early if there's not one latch). I'm not sure what makes the > LoopContinuePredicate check safer?
You can have a loop with a single latch but an arbitrary number of assumes (= n). In such a loop, you'll still end up calling isImpliedCond O(n!) times because of the assumes (and the guard on there being only one latch won't help). OTOH if the loop has multiple latches = m > 1, then we exit early we won't recurse at all, we'll just bail early from isLoopBackedgeGuardedByCond because of the check you mention (not ideal, but safe). -- Sanjoy _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
