On Thu, Sep 26, 2019 at 10:14:17PM +0200, Johannes Schindelin wrote: > I mean, if there is _any_ performance-critical code path hitting this > unrolled loop, we may want to keep it unrolled.
The loop in question is maybe a few dozen instructions, and then it immediately makes an open() syscall, which is probably multiple orders of magnitude more expensive. So I'd be very surprised if it was a problem no matter what the generated code looks like. But... > However, I think that this patch should at least be accompanied by a > commit message that suggests that some thought was put into it, and that > concerns like mine were considered carefully. ...this I would definitely agree with. -Peff
