Re: optimalisation for strlcpy (lib/string.c)

[Adrian Bunk <bunk@stusta.de>]

On Sun, Dec 10, 2006 at 10:23:51PM +0100, Folkert van Heusden wrote:
> Hi,
> 
> Like the other patch (by that other person), I think it is faster to not
> do a strlen first.
>...
> --- lib/string.c        2006-11-04 02:33:58.000000000 +0100
> +++ string-new.c        2006-12-10 22:22:08.000000000 +0100
> @@ -121,14 +121,24 @@
>   */
>  size_t strlcpy(char *dest, const char *src, size_t size)
>  {
> -       size_t ret = strlen(src);
> +        char *tmp = dest;
> 
> -       if (size) {
> -               size_t len = (ret >= size) ? size - 1 : ret;
> -               memcpy(dest, src, len);
> -               dest[len] = '\0';
> +        for(;;)
> +        {
> +                *dest = *src;
> +                if (!*src)
> +                        break;
> +
> +                if (--size == 0)
> +                        break;
> +
> +                dest++;
> +                src++;
>         }
> -       return ret;
> +
> +        *dest = 0x00;
> +
> +        return dest - tmp;
>...

Two bugs in your code:
- you copy a maximum of size bytes _plus_ \0
- size == 0 is no longer handled correctly

> I've tested the speed difference with this:
> http://www.vanheusden.com/misc/kernel-strlcpy-opt-test.c
> and the speed difference is quite a bit on a P4: 28% faster.
>...

My Athlon says:
org: 2.400000
new: 6.710000

IOW, your version is much slower.

But the main question is actually:
Does the performance of this function matter anywhere inside the kernel?
Is strlcpy() used in any fast path?
If not, there's no point in trying to optimize it.

> Folkert van Heusden

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed