From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S1754531AbbCMM4H (ORCPT <rfc822;w@1wt.eu>);
	Fri, 13 Mar 2015 08:56:07 -0400
Received: from mail-lb0-f175.google.com ([209.85.217.175]:33134 "EHLO
	mail-lb0-f175.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1751391AbbCMM4D (ORCPT
	<rfc822;linux-kernel@vger.kernel.org>);
	Fri, 13 Mar 2015 08:56:03 -0400
MIME-Version: 1.0
In-Reply-To: <CACVxJT_enpRNoBf3TizQ2U1Z77viP6sCy3CGTMnhpzpkjDpqcw@mail.gmail.com>
References: <5500b987.kerYYCYfIffruy3Z%akpm@linux-foundation.org>
	<CACVxJT_enpRNoBf3TizQ2U1Z77viP6sCy3CGTMnhpzpkjDpqcw@mail.gmail.com>
Date: Fri, 13 Mar 2015 15:56:01 +0300
Message-ID: <CACVxJT-vtOVbM9wTFoJe01i2CxxcJYsMdg5kCR4S9tfzxe883Q@mail.gmail.com>
Subject: Re: + lib-vsprintfc-even-faster-decimal-conversion.patch added to -mm tree
From: Alexey Dobriyan <adobriyan@gmail.com>
To: Linux Kernel <linux-kernel@vger.kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>
Cc: linux@rasmusvillemoes.dk, Peter Zijlstra <peterz@infradead.org>,
        Tejun Heo <tj@kernel.org>, Denis Vlasenko <vda.linux@googlemail.com>,
        KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Content-Type: text/plain; charset=UTF-8
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Fri, Mar 13, 2015 at 3:49 PM, Alexey Dobriyan <adobriyan@gmail.com> wrote:
> Legend is "number avg+-1sigma min-max". Every number is CPU cycles.
> Great care was taken to remove interrupt noise.

To reproduce numbers:
* apply the patch which adds sys_nr_irq() system call.
* recompile without cpufreq, reboot with isolcpus=,
* shift interrupts off free CPU

* use new system call
static inline uint64_t sys_nr_irq(void)
{
        uint64_t rv;

        asm volatile ("syscall" : "=a" (rv) : "0" (323) : "memory",
"cc", "rcx", "r11");
        return rv;
}

* use measurement code
static inline void rdtsc(uint32_t *edx, uint32_t *eax)
{
        asm volatile ("rdtsc" : "=d" (*edx), "=a" (*eax));
}
static inline void lfence(void)
{
        asm volatile ("lfence" ::: "memory");
}
static inline uint64_t time_init(uint64_t *x)
{
        uint32_t edx, eax;

        *x = sys_nr_irq();

        lfence();
        rdtsc(&edx, &eax);
        return ((uint64_t)edx << 32) | eax;
}

static inline uint64_t time_fini(uint64_t *x)
{
        uint32_t edx, eax;

        lfence();
        rdtsc(&edx, &eax);

        *x = sys_nr_irq();

        return ((uint64_t)edx << 32) | eax;
}
----------------------------------------
* drop measurement where interrupt interfered
        i = 0;
        while (i < N) {
                uint64_t t0, t1;
                uint64_t x0, x1;

                t0 = time_init(&x0);
                f();
                t1 = time_fini(&x1);

                if (x1 != x0)
                        continue;

                T[i] = t1 - t0;
                i++;
        }
------------------------------------
* average results how you were taught in school :-)