Date: Wed, 19 Oct 2016 15:07:23 +0200
From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
To: Davidlohr Bueso <dave@stgolabs.net>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>, Ingo Molnar <mingo@redhat.com>,
        linux-kernel@vger.kernel.org, Davidlohr Bueso <dbueso@suse.de>
Subject: Re: [PATCH 1/2] perf bench futex: cache align the worer struct
Message-ID: <20161019130722.t7viruflpg2xu5sx@linutronix.de>
References: <20161016190803.3392-1-bigeasy@linutronix.de>
 <20161018010949.GD29373@linux-80c1.suse>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8BIT
In-Reply-To: <20161018010949.GD29373@linux-80c1.suse>
User-Agent: NeoMutt/20161014 (1.7.1)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 965
Lines: 27

On 2016-10-17 18:09:49 [-0700], Davidlohr Bueso wrote:
> On Sun, 16 Oct 2016, Sebastian Andrzej Siewior wrote:
> 
> > It popped up in perf testing that the worker consumes some amount of
> > CPU. It boils down to the increment of `ops` which causes cache line
> > bouncing between the individual threads.
> 
> Are you referring to this?
> 
>       â         for (i = 0; i < nfutexes; i++, w->ops++) {
>       â be:   add    $0x1,%ebx
> 65.87 â       addq   $0x1,0x18(%r12)
> 
> (which is like 65% of 13% on my box with a default futex-hash run).
correct.

> Even better, could we get rid entirely of the ops increments and just
> use a local variable, then update the worker at the end of the function.
> The following makes 'perf' pretty much disappear in the profile.

this should do it, too. So what remains is the read access for w->futex
but since it does not pop up in perf, it is probably not that important.

> Thanks,
> Davidlohr

Sebastian