Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 3 Mar 2021 10:49:27 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Josh Don <joshdon@google.com>
Cc:     Ingo Molnar <mingo@redhat.com>, Juri Lelli <juri.lelli@redhat.com>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        Clement Courbet <courbet@google.com>,
        Oleg Rombakh <olegrom@google.com>
Subject: Re: [PATCH] sched: Optimize __calc_delta.
Message-ID: <YD9bp7C9k6ojjq/5@hirez.programming.kicks-ass.net>
References: <20210226195239.3905966-1-joshdon@google.com>
 <YDliCOPy9s1RdLwd@hirez.programming.kicks-ass.net>
 <CABk29Nv7iJEcDg3rgSvfTkXEM69ZeLByJAsZYuA5qpdj645nZw@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CABk29Nv7iJEcDg3rgSvfTkXEM69ZeLByJAsZYuA5qpdj645nZw@mail.gmail.com>
Precedence: bulk

On Tue, Mar 02, 2021 at 12:55:07PM -0800, Josh Don wrote:
> On Fri, Feb 26, 2021 at 1:03 PM Peter Zijlstra <peterz@infradead.org> wrote:
> >
> > On Fri, Feb 26, 2021 at 11:52:39AM -0800, Josh Don wrote:
> > > From: Clement Courbet <courbet@google.com>
> > >
> > > A significant portion of __calc_delta time is spent in the loop
> > > shifting a u64 by 32 bits. Use a __builtin_clz instead of iterating.
> > >
> > > This is ~7x faster on benchmarks.
> >
> > Have you tried on hardware without such fancy instructions?
> 
> Was not able to find any on hand unfortunately. Clement did rework the
> patch to use fls() instead, and has benchmarks for the generic and asm
> variations. All of which are faster than the loop. In my next reply,
> I'll include the updated patch inline.

Excellent; I have some vague memories where using fls ended up slower
for some ARMs, but I can't seem to remember enough to even Google it :/