[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <da676f991cd2818bcb234ac4e70c3e56c5407167.camel@linux.intel.com>
Date: Wed, 06 Sep 2023 08:56:55 -0700
From: Tim Chen <tim.c.chen@...ux.intel.com>
To: Shrikanth Hegde <sshegde@...ux.vnet.ibm.com>
Cc: dietmar.eggemann@....com, vschneid@...hat.com,
linux-kernel@...r.kernel.org, srikar@...ux.vnet.ibm.com,
mgorman@...hsingularity.net, mingo@...nel.org, yu.c.chen@...el.com,
ricardo.neri-calderon@...ux.intel.com, iamjoonsoo.kim@....com,
juri.lelli@...hat.com, rocking@...ux.alibaba.com,
joshdon@...gle.com, mingo@...hat.com, peterz@...radead.org,
vincent.guittot@...aro.org
Subject: Re: [PATCH] sched/fair: optimize should_we_balance for higher SMT
systems
On Wed, 2023-09-06 at 07:36 +0530, Shrikanth Hegde wrote:
>
> Hi Tim,
>
> Thanks for taking a look at this patch.
>
> > Wonder if we can avoid allocating the
> > should_we_balance_tmpmask for SMT2 case to save memory
> > for system with large number of cores.
> >
> > The new mask and logic I think is only needed for more than 2 threads in a core.
>
> Code would have to be refactored quite a bit if one needs to take
> different approach for specific SMT setting.
>
> I think there would some cases in SMT2 that will benefit as well.
> Lets say 1 cpu in each core is busy. the busy CPU happens to be second
> CPU in the core. In that case, this approach would skip that instead of
> checking if that is idle or not.
>
>
That's true. This change could skip the sibling for SMT2.
Tim
Powered by blists - more mailing lists