Re: [Qemu-devel] [PATCH for-1.5 0/9] Disable expensive QOM cast debugging for official releases

[Aurelien Jarno <aurelien@aurel32.net>]

On Fri, May 10, 2013 at 10:59:22PM +0200, Paolo Bonzini wrote:
> Il 10/05/2013 19:41, Anthony Liguori ha scritto:
> > Paolo Bonzini <pbonzini@redhat.com> writes:
> > 
> >> Il 10/05/2013 16:39, Anthony Liguori ha scritto:
> >>> I just oppose the notion of disabling casts and *especially* only
> >>> disabling casts for official builds.
> >>
> >> This actually happens all the time.  Exactly this kind of type-safe cast
> >> is disabled in releases of GCC, but enabled when building from svn trunk.
> > 
> > Let's assume for a moment that you are right and this behavior is what
> > we should have.  Let's also assume there is a real regression here
> > which has yet to have been established.
> 
> Aurelien timed the effect of my patch two hours before you sent this
> message.  If it's not a regression in 1.5 (which is quite obvious from
> the profile), it is a regression from the introduction of CPU classes
> (1.3 or 1.4), when this code didn't exist at all.
> 
> And in 1.5 we introduced virtio-net casts as well (or did mst sneak in
> his change anyway? ;)). If 10% is the effect of a few hundred
> interrupts/sec, perhaps the same effect is visible on a few thousand
> packets/sec.  I wouldn't bet against that one week from release.

I have run perf top on a KVM x86-64 instance running iperf. I measured
around 3% of the CPU taken by g_hash_table_lookup + g_str_hash +
__strcmp_sse42. I don't know how it was before though.

-- 
Aurelien Jarno                          GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net