https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117974
--- Comment #6 from Vineet Gupta <vineetg at gcc dot gnu.org> --- (In reply to Jeffrey A. Law from comment #5) > What's starting to rattle around in my brain is the for a loop, if the count > is unknown, then we probably don't want to unroll as that's much more likely > to trigger the VL=0 problem you're trying to avoid. Sure, but the unrolling is contrived/deliberate/forced in the test, just to expose the codegen. This is not so much about VL=0 per-se but to delay the VSETVL to after the branch. IOW this is a less deep issue :-) just an implementation tweak.
