On Fri, Mar 11, 2022 at 08:50:47AM +0900, Damien Le Moal wrote: > On 3/11/22 05:35, Luis Chamberlain wrote: > > On Thu, Mar 10, 2022 at 06:43:47AM +0900, Damien Le Moal wrote: > >> The indirect call using a function pointer makes the code obscure. And > >> the cost of that call is far greater and always present compared to the > >> CPU branch prediction which will luckily avoid most of the time taking > >> the wrong branch of an if. > > > > The goal was to ensure no performance impact, and given a hot path > > was involved and we simply cannot microbench append as there is no > > way / API to do that, we can't be sure. But if you are certain that > > there is no perf impact, it would be wonderful to live without it. > > Use zonefs. It uses zone append. That'd be a microbench on zonefs append, not raw append, and so I suspect a simple branch cannot make a difference and would get lost as noise. So I don't think a perf test is required here. Please let me know. Luis
