svn commit: r266974 - in head/sys: dev/dc dev/fxp dev/mii dev/netmap kern net

[Gleb Smirnoff]

On Tue, Jun 03, 2014 at 12:59:41PM +0400, Gleb Smirnoff wrote:
T> On Mon, Jun 02, 2014 at 12:27:47PM -0700, Adrian Chadd wrote:
T> A> >> > .. and actually, bikeshedding for a moment, would we be able to move a
T> A> >> > lot of these accessor methods over to inlines? Would that break the
T> A> >> > Juniper way of doing things?
T> A> >>
T> A> >> That would definitely break Juniper as it doesn't give a stable
T> A> >> ABI.
T> A> >>
T> A> >> I've suggested an approach that allows for both, but it was deemed
T> A> >> unnecessary. The argument being that the function call overhead is
T> A> >> negligible.
T> A> >>
T> A> >> We can always revisit that decision if needed...
T> A> >>
T> A> >
T> A> > In my experience, function call overhead is anything but minimal,
T> A> > especially on ARM platforms.
T> A> 
T> A> Same on MIPS.
T> 
T> And same on amd64. While benchmarking the counter(9), I even encountered
T> artifacts when function call was faster than inline.
T> 
T> Of course, function call shouldn't be via a pointer.

Sorry, I misunderstood the "anything but minimal" phrase. That's why
my reply Adrian and Ian looks odd.

-- 
Totus tuus, Glebius.