Re: [RFC][PATCH] Faster generic_fls

From: Daniel Phillips <dphillips@sistina.com>
To: Thomas Schlichter <schlicht@uni-mannheim.de>,
	Willy TARREAU <willy@w.ods.org>, hugang <hugang@soulinfo.com>
Cc: linux-kernel@vger.kernel.org, akpm@digeo.com
Subject: Re: [RFC][PATCH] Faster generic_fls
Date: Fri, 2 May 2003 02:13:41 +0200	[thread overview]
Message-ID: <200305020213.41721.dphillips@sistina.com> (raw)
In-Reply-To: <200305020127.26279.schlicht@uni-mannheim.de>

On Friday 02 May 2003 01:27, Thomas Schlichter wrote:
> ...So for me the table version seems to be the slowest one. The BSRL
> instruction on the K6-III seems to be very slow, too. The tree and my shift
> version are faster than the original version here...
>
> That someone else can test my fls_shift version I'll provide it here again:
> static inline int fls_shift(int x)
> {
> 	int bit = 32;
>
> 	while(x > 0) {
> 		--bit;
> 		x <<= 1;
> 	}
>
> 	return x ? bit : 0;
> }

Your shift version is the fastest on the PIII as well, finishing in 45.3 
seconds vs 53.4 for my original, and using only 12 bytes of text.  This was a 
big surprise.  The time was the same, whether I inlined it or not.  I take 
this to mean that -O3 inlines such a short function in either case.

Regards,

Daniel