> > I fail to see how replacing incrementation with a branch statement helps > performance or even not making it worse... It may have been coincidental in my setup - I had assumed the test on the local io_stat was cheaper than the cost of following hctx to then increment (or not) poll_invoked.