lkml.org 
[lkml]   [2012]   [Jan]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    Patch in this message
    /
    Date
    From
    SubjectRe: [PATCH] Add num_to_str() for speedup /proc/stat
    On Mon, 30 Jan 2012 14:16:19 +0900
    KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:

    > From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
    > Date: Mon, 30 Jan 2012 14:15:12 +0900
    > Subject: [PATCH] Add num_to_str() for speedup /proc/stat
    >
    > At reading /proc/stat, most of time is consumed by vsnprintf() at el.
    >
    > Here is a test script, reading /proc/stat 1000 times..
    >
    > == stat_check.py
    > num = 0
    > with open("/proc/stat") as f:
    > while num < 1000 :
    > data = f.read()
    > f.seek(0, 0)
    > num = num + 1
    > ==
    >
    > perf shows
    >
    > 20.39% stat_check.py [kernel.kallsyms] [k] format_decode
    > 13.41% stat_check.py [kernel.kallsyms] [k] number
    > 12.61% stat_check.py [kernel.kallsyms] [k] vsnprintf
    > 10.85% stat_check.py [kernel.kallsyms] [k] memcpy
    > 4.85% stat_check.py [kernel.kallsyms] [k] radix_tree_lookup
    > 4.43% stat_check.py [kernel.kallsyms] [k] seq_printf
    >
    > This patch removes most of calls to vsnprintf() by adding
    > num_to_str() and seq_print_decimal_ull(), which prints decimal numbers
    > without rich functions provided by printf().
    >
    > On my 8cpu box.
    > == Before patch ==
    > [root@bluextal test]# time ./stat_check.py
    >
    > real 0m0.150s
    > user 0m0.026s
    > sys 0m0.121s
    >
    > == After patch ==
    > [root@bluextal test]# time ./stat_check.py
    >
    > real 0m0.055s
    > user 0m0.022s
    > sys 0m0.030s
    >
    > Maybe it's worth to add this simple function.

    I suppose so - the new infrastructure can be used elsewhere.

    I tried doing the

    if (kstst_irqs(j) == 0) {
    seq_putc(p, ' ');
    seq_putc(p, '0');

    think on top of this and didn't observe any improvement.


    I made some changes - please review. I'm not sure why you did "char
    tmp[66]"?


    From: Andrew Morton <akpm@linux-foundation.org>
    Subject: procfs-add-num_to_str-to-speed-up-proc-stat-fix

    - remove incorrect comment
    - use less stack in num_to_str()
    - move comment from .h to .c
    - simplify seq_put_decimal_ull()

    Cc: Eric Dumazet <eric.dumazet@gmail.com>
    Cc: Glauber Costa <glommer@parallels.com>
    Cc: Ingo Molnar <mingo@elte.hu>
    Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
    Cc: Paul Turner <pjt@google.com>
    Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
    Cc: Russell King <rmk@arm.linux.org.uk>
    Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
    ---

    fs/seq_file.c | 5 ++---
    include/linux/kernel.h | 6 ------
    include/linux/seq_file.h | 4 +---
    lib/vsprintf.c | 8 +++++++-
    4 files changed, 10 insertions(+), 13 deletions(-)

    diff -puN fs/proc/stat.c~procfs-add-num_to_str-to-speed-up-proc-stat-fix fs/proc/stat.c
    diff -puN fs/seq_file.c~procfs-add-num_to_str-to-speed-up-proc-stat-fix fs/seq_file.c
    --- a/fs/seq_file.c~procfs-add-num_to_str-to-speed-up-proc-stat-fix
    +++ a/fs/seq_file.c
    @@ -659,14 +659,13 @@ int seq_put_decimal_ull(struct seq_file
    if (m->count + 2 >= m->size) /* we'll write 2 bytes at least */
    goto overflow;

    + m->buf[m->count++] = delimiter;
    +
    if (num < 10) {
    - m->buf[m->count++] = delimiter;
    m->buf[m->count++] = num + '0';
    return 0;
    }

    - m->buf[m->count++] = delimiter;
    -
    len = num_to_str(m->buf + m->count, m->size - m->count, num);
    if (!len)
    goto overflow;
    diff -puN include/linux/kernel.h~procfs-add-num_to_str-to-speed-up-proc-stat-fix include/linux/kernel.h
    --- a/include/linux/kernel.h~procfs-add-num_to_str-to-speed-up-proc-stat-fix
    +++ a/include/linux/kernel.h
    @@ -299,12 +299,6 @@ extern long long simple_strtoll(const ch
    #define strict_strtoull kstrtoull
    #define strict_strtoll kstrtoll

    -/*
    - * Convert passed number to decimal string.
    - * returns returns the length of string. at buffer overflow, returns 0.
    - *
    - * If speed is not important, use snprintf(). It's easy to read the code.
    - */
    extern int num_to_str(char *buf, int size, unsigned long long num);

    /* lib/printf utilities */
    diff -puN include/linux/seq_file.h~procfs-add-num_to_str-to-speed-up-proc-stat-fix include/linux/seq_file.h
    --- a/include/linux/seq_file.h~procfs-add-num_to_str-to-speed-up-proc-stat-fix
    +++ a/include/linux/seq_file.h
    @@ -121,10 +121,8 @@ int single_release(struct inode *, struc
    void *__seq_open_private(struct file *, const struct seq_operations *, int);
    int seq_open_private(struct file *, const struct seq_operations *, int);
    int seq_release_private(struct inode *, struct file *);
    -
    -/* defined in lib/vsprintf.c */
    int seq_put_decimal_ull(struct seq_file *m, char delimiter,
    - unsigned long long num);
    + unsigned long long num);

    #define SEQ_START_TOKEN ((void *)1)
    /*
    diff -puN lib/vsprintf.c~procfs-add-num_to_str-to-speed-up-proc-stat-fix lib/vsprintf.c
    --- a/lib/vsprintf.c~procfs-add-num_to_str-to-speed-up-proc-stat-fix
    +++ a/lib/vsprintf.c
    @@ -212,9 +212,15 @@ char *put_dec(char *buf, unsigned long l
    }
    }

    +/*
    + * Convert passed number to decimal string.
    + * Returns the length of string. On buffer overflow, returns 0.
    + *
    + * If speed is not important, use snprintf(). It's easy to read the code.
    + */
    int num_to_str(char *buf, int size, unsigned long long num)
    {
    - char tmp[66];
    + char tmp[21]; /* Enough for 2^64 in decimal */
    int idx, len;

    len = put_dec(tmp, num) - tmp;
    _


    \
     
     \ /
      Last update: 2012-01-31 00:23    [W:0.032 / U:0.104 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site