svn commit: r238907 - projects/calloutng/sys/kern

[Attilio Rao]

On 10/29/12, Bruce Evans <brde at optusnet.com.au> wrote:
> On Mon, 29 Oct 2012, Attilio Rao wrote:
>

[ trimm ]
>
> The above is mostly about -current.  My previous mail discussed why
> uninlining mtx_lock/unlock() doesn't work so well.  It is because these
> macros are shorter and don't make unconditional function calls, so
> inlining of them has better chances of improving their efficiency, and
> in fact does.

Again you are mentioning mtx_lock/unlock() but this has nothing to do
with them, this is about mtx_lock_spin().
However, I quite see your point and honestly I always thought that
inlining mtx_lock_spin() wasn't really useful because of the function
call themselves from spinlock_enter().

However, in order to get to a version which *doesn't* inline them we
should have a clear win which I don't think it is there (this is too
dependent by the arch), so I would say to stick with what the current
code does.

Attilio


-- 
Peace can only be achieved by understanding - A. Einstein