Date: Mon, 12 Jan 2009 22:53:52 -0800 From: Xin LI <delphij@delphij.net> To: =?UTF-8?B?RGFnLUVybGluZyBTbcO4cmdyYXY=?= <des@des.no> Cc: d@delphij.net, freebsd-arch@FreeBSD.org Subject: Re: RFC: MI strlen() Message-ID: <496C3A80.5040007@delphij.net> In-Reply-To: <86zlhw5zsr.fsf@ds4.des.no> References: <4966B5D4.7040709@delphij.net> <86zlhw5zsr.fsf@ds4.des.no>
next in thread | previous in thread | raw e-mail | index | archive | help
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Dag-Erling Smørgrav wrote:
> Xin LI <delphij@delphij.net> writes:
>> Here is a new implementation of strlen() which employed the bitmask
>> skill in order to achieve better performance on modern hardware.
>
> Why bother? Do we have code that uses strlen() heavily in performance-
> critical regions? Does anybody?
I agree that strlen() should never be used in performance-sensitive
regions, but given we do not have assembly optimized versions of these
functions for amd64 and almost everybody else has it, having a C version
that gives similar performance is valuable. Also, worldstone with -j9
on 2x4core machine with both src/ and obj/ in tmpfs, seems to have
small, but positive effect:
Unpatched libc:
1400.97 real 4159.34 user 2901.08 sys
1396.73 real 4159.06 user 2906.16 sys
1380.27 real 4158.20 user 2803.22 sys
Patched libc:
1363.29 real 4154.89 user 2749.94 sys
1373.96 real 4150.45 user 2830.46 sys
1368.62 real 4152.48 user 2838.52 sys
(with 'make cleanworld' between builds)
Cheers,
- --
Xin LI <delphij@delphij.net> http://www.delphij.net/
FreeBSD - The Power to Serve!
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.9 (FreeBSD)
iEYEARECAAYFAklsOoAACgkQi+vbBBjt66COQgCfTmpQK9YliCxpdJkckJ2/cZim
NzEAoKRqC2HN1FtKRWaZhstYyVjYeewr
=eeea
-----END PGP SIGNATURE-----
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?496C3A80.5040007>