From owner-freebsd-alpha Sat Jul 1 18:10: 8 2000 Delivered-To: freebsd-alpha@freebsd.org Received: from teapot21.domain3.bigpond.com (teapot21.domain3.bigpond.com [139.134.5.159]) by hub.freebsd.org (Postfix) with SMTP id 3FDFD37BA7B for ; Sat, 1 Jul 2000 18:10:04 -0700 (PDT) (envelope-from tpnelson@student.cowan.edu.au) Received: from localhost (localhost [127.0.0.1]) by teapot21.domain3.bigpond.com (NTMail 3.02.13) with ESMTP id ya002104 for ; Sun, 2 Jul 2000 00:07:23 +1000 Received: from WEBH-T-009-p-99-204.tmns.net.au ([139.134.99.204]) by mail3.bigpond.com (Claudes-Colourful-MailRouter V2.8a 5/476039); 02 Jul 2000 00:07:19 Message-ID: <395DFB2D.4D2139C7@student.cowan.edu.au> Date: Sat, 01 Jul 2000 22:07:41 +0800 From: Trent Nelson X-Mailer: Mozilla 4.7 [en] (WinNT; I) X-Accept-Language: en MIME-Version: 1.0 To: John Polstra Cc: alpha@FreeBSD.ORG Subject: Re: Cache line size for the Alpha References: <395A0EE8.28E6C524@student.cowan.edu.au> <200006291628.JAA23243@vashon.polstra.com> <395B0487.9D94D2A8@student.cowan.edu.au> <200006301717.KAA25927@vashon.polstra.com> Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Sender: owner-freebsd-alpha@FreeBSD.ORG Precedence: bulk X-Loop: FreeBSD.org John Polstra wrote: > > In article <395B0487.9D94D2A8@student.cowan.edu.au>, > Trent Nelson wrote: > > From what I've read, I think the biggest performance > > consideration is the locality of the resource in contention. For > > high contention locks, it seems desirable to ensure that the > > spinlock code and the resource in contention are placed on separate > > 128-byte cache block boundaries. > > The 128-byte figure confuses me. I thought the biggest cache line > size you mentioned before was 64 bytes. Is a cache "block" different > from a cache line? The 128-byte figure simply derives from Compaq's encouragement for Software Developers to "over-align" code in anticipation of future Alpha implementations with a 128-bit wide data bus. Effectively, in these environments, code that is octaword aligned will be given a higher execution priority in the pipeline than quadword aligned code. Aligning code to 128-byte boundaries will have no performance advantages in current Alpha implementations. That said, it will also incur no performance penalties and will be advantageous in later Alpha architectures. Otherwise, quadword alignment is fine. Your call. On a separate note, would it be possible to take a look at the code when you're finished? I don't have access to an Alpha so it makes it a little hard to keep track of the source in this particular branch. > John Regards, Trent. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-alpha" in the body of the message