X-pop3-spooler: POP3MAIL 2.1.0 b 3 961213 -bs- Delivered-To: pcg AT goof DOT com Date: Tue, 14 Apr 1998 13:07:57 +0200 (CEST) From: Ronald Wahl X-Sender: rwa AT goliath DOT csn DOT tu-chemnitz DOT de To: beastium-list AT Desk DOT nl Subject: Re: [performance] newer binutils / pgcc / K6 In-Reply-To: Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: Marc Lehmann Status: RO Content-Length: 1545 Lines: 30 On Tue, 14 Apr 1998, Ronald Wahl wrote: > Hi, > > I noticed a performance problem starting with release 2.8.1.0.26 of > binutils. If I run nbench on my K6 with binutils 2.8.1.0.26 (or higher) > some tests will be slower. My first thoughts were that it had something to > do with the changes hjl made from 2.8.1.0.25 to 2.8.1.0.26 but after some > further testing I found out that it is a code alignment issue. If I use > -malign-loops=2 the tests run nearly at the same speed as with the older > versions of binutils (gas). Some tests are a bit slower but not much > (--> see my appended nbench results). Other alignments will cause > slowdowns. Before changing any defaults for loop alignment on a K6 in pgcc > - is someone willing to play a bit with old and new releases of binutils > and some other benchmarks or real world applications? I tested around by myself and found out that real world applications will often run slower with -malign-loops=2. My guess is that in nested loops dummy code (from the alignment) consumes decode slots. This will result in a loss of performance. So my question: Is it possible to align loops only if these are not nested in another loop (with a high loop count) or if the alignment code will never be executed? ron -- \ Ronald Wahl --- rwahl AT gmx DOT net \ Gib Gates keine Chance! / \ WWW: http://www.tu-chemnitz.de/~row/ \ / \ Talk: rwa AT goliath DOT csn DOT tu-chemnitz DOT de \ Pinguine schuetzen. / \ PGP key available \ /