Re: [RFC][PATCH] Faster generic_fls

[hugang <hugang@soulinfo.com>]

On Thu, 1 May 2003 15:52:04 +0200
Willy TARREAU <willy@w.ods.org> wrote:

>  However, I have good news for you, the table code is 15% faster than my tree
>  on an Alpha EV6 ;-)
>  It may be because gcc can use different tricks on this arch.
> 
>  Cheers
>  Willy

However, The most code changed has increase a little peformance, Do we really need it?

Here is test in my machine, The faster is still table.

==== -Wall -Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common -fomit-frame-pointer -pipe -mpreferred-stack-boundary=2 -march=i686 ====
test_fls.c: In function `fls_tree_fls':
test_fls.c:78: warning: type defaults to `int' in declaration of `t'
4294967295 iterations of nil... checksum = 4294967295

real    0m21.852s
user    0m21.500s
sys     0m0.300s
4294967295 iterations of old... checksum = 4294967265

real    0m45.206s
user    0m44.800s
sys     0m0.310s
4294967295 iterations of new... checksum = 4294967265

real    0m45.235s
user    0m44.720s
sys     0m0.420s
4294967295 iterations of new2... checksum = 4294967265

real    0m59.615s
user    0m58.960s
sys     0m0.540s
4294967295 iterations of table... checksum = 4294967265

real    0m41.784s
user    0m41.420s
sys     0m0.280s
4294967295 iterations of tree... checksum = 4294967265

real    0m44.874s
user    0m44.410s
sys     0m0.380s
------------------


-- 
Hu Gang / Steve
Email        : huagng@soulinfo.com, steve@soulinfo.com
GPG FinePrint: 4099 3F1D AE01 1817 68F7  D499 A6C2 C418 86C8 610E
ICQ#         : 205800361
Registered Linux User : 204016