Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 9 Sep 2019 14:18:12 +0300
From:   Cyrill Gorcunov <gorcunov@gmail.com>
To:     Dmitry Safonov <dima@arista.com>
Cc:     linux-kernel@vger.kernel.org,
        Dmitry Safonov <0x7f454c46@gmail.com>,
        Adrian Reber <adrian@lisas.de>,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        Andrei Vagin <avagin@openvz.org>,
        Andy Lutomirski <luto@kernel.org>,
        Ingo Molnar <mingo@redhat.com>,
        Oleg Nesterov <oleg@redhat.com>,
        Pavel Emelyanov <xemul@virtuozzo.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        containers@lists.linux-foundation.org,
        linux-fsdevel@vger.kernel.org
Subject: Re: [PATCH 4/9] select: Micro-optimise __estimate_accuracy()
Message-ID: <20190909111812.GB1508@uranus>
References: <20190909102340.8592-1-dima@arista.com>
 <20190909102340.8592-5-dima@arista.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190909102340.8592-5-dima@arista.com>
User-Agent: Mutt/1.12.0 (2019-05-25)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Sep 09, 2019 at 11:23:35AM +0100, Dmitry Safonov wrote:
> Shift on s64 is faster than division, use it instead.
> 
> As the result of the patch there is a hardly user-visible effect:
> poll(), select(), etc syscalls will be a bit more precise on ~2.3%
> than before because 1000 != 1024 :)
> 
> Signed-off-by: Dmitry Safonov <dima@arista.com>

> ---
>  fs/select.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/select.c b/fs/select.c
> index 12cdefd3be2d..2477c202631e 100644
> --- a/fs/select.c
> +++ b/fs/select.c
> @@ -51,15 +51,14 @@
>  
>  static long __estimate_accuracy(ktime_t slack)
>  {
> -	int divfactor = 1000;
> -
>  	if (slack < 0)
>  		return 0;
>  
> -	if (task_nice(current) > 0)
> -		divfactor = divfactor / 5;
> +	/* A bit more precise than 0.1% */
> +	slack = slack >> 10;
>  
> -	slack = ktime_divns(slack, divfactor);
> +	if (task_nice(current) > 0)
> +		slack = slack * 5;
>  
>  	if (slack > MAX_SLACK)
>  		return MAX_SLACK;

Compiler precompute constants so it doesn't do division here.
But I didn't read the series yet so I might be missing
something obvious.