Message-Id: <199910280735.JAA27462@w9205mbl.fmis.hypovereinsbank.de>
X-Exmh-Isig-Folder: undelivMail
X-Exmh-Isig-CompType: forw
To: pgcc AT delorie DOT com
cc: dieter DOT nuetzel AT myokay DOT net, postmaster AT delorie DOT com
Subject: delorie.com is a sissy! - Was: Re: Q: I plan to buy me an Athlon (aka AMD K7). What can I expect from the
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Thu, 28 Oct 1999 09:35:08 MET DST
From: Herbert Wengatz 42850 <herbert DOT wengatz DOT extern AT ae3 DOT hypo DOT de>
Reply-To: pgcc AT delorie DOT com

Hey folks, delorie.com wanted it. - Now you get it all!

Sorry, but I don't intend to change my language. Is delorie under the
thumbs of Scientology, or what? - IMHO you should change the provider.

I'm not overly agressive, nor was I insulting anybody. I feel this is
a plain overexaggeration. And such measures lead to "doublketalk" and
I personally feel like "big brother is watching me" - Ouch!

We here in germany have a quite different feeling about such oppressive
behaviour... We already have made our experiences with people trying
to tell people what they shall think and say.

Is america no longer the land of the free? The land where everything
is possible?

Are you trying to implement some sort of thought police?

I was just using a word every little child can find in the holy bible.
So what!?

Honestly, heavily disgusted,

	Herbert

PS: There is a mailinglist-related mail enclosed, at the end of the 
following garbage... ;)

------- Forwarded Message

Return-Path: <Mailer-Daemon>
Received: from mailgate2.hypovereinsbank.de by ae3.hypo.de (SMI-8.6/SMI-SVR4)
	id IAA19499; Thu, 28 Oct 1999 08:40:04 +0200
Received: from fwproi1.HypoVereinsbank.DE (mailrelay.hypovereinsbank.de [192.168.254.10])
	by mailgate2.hypovereinsbank.de ($Revision: 1.13 $)  with ESMTP id IAA28987
	for <herbert DOT wengatz DOT extern AT ae3 DOT hypo DOT de>; Thu, 28 Oct 1999 08:39:07 +0200 (MET DST)
Received: from vwmail.HypoVereinsbank.de (fwvw1 [193.158.192.130])
	by fwproi1.HypoVereinsbank.DE ($Revision: 1.1 $)  with ESMTP id GAA29887
	for <herbert DOT wengatz DOT extern AT ae3 DOT hypo DOT de>; Thu, 28 Oct 1999 06:39:06 GMT
Received: from localhost (localhost)
	by vwmail.HypoVereinsbank.de ($Revision: 1.1 $)  with internal id GAA00730;
	Thu, 28 Oct 1999 06:38:45 GMT
Date: Thu, 28 Oct 1999 06:38:45 GMT
From: Mail Delivery Subsystem <MAILER-DAEMON AT HypoVereinsbank DOT DE>
Message-Id: <199910280638 DOT GAA00730 AT vwmail DOT HypoVereinsbank DOT de>
To: <herbert DOT wengatz DOT extern AT ae3 DOT hypo DOT de>
MIME-Version: 1.0
Content-Type: multipart/report; report-type=delivery-status;
	boundary="GAA00730.941092725/vwmail.HypoVereinsbank.de"
Subject: Returned mail: Remote protocol error
Auto-Submitted: auto-generated (failure)
content-length: 8926
X-Filter: mailagent [version 3.0 PL44] for ts37999 AT w9205mbl DOT ae3 DOT hypo DOT de

This is a MIME-encapsulated message

- --GAA00730.941092725/vwmail.HypoVereinsbank.de

The original message was received at Thu, 28 Oct 1999 06:37:31 GMT
from localhost [127.0.0.1]

   ----- The following addresses had permanent fatal errors -----
<pgcc AT delorie DOT com>

   ----- Transcript of session follows -----
... while talking to delorie.com.:
>>> DATA
<<< 500 Hey, offensive language ("h?ll") is not allowed.  <dj AT delorie DOT com>
554 <pgcc AT delorie DOT com>... Remote protocol error

- --GAA00730.941092725/vwmail.HypoVereinsbank.de
Content-Type: message/delivery-status

Reporting-MTA: dns; vwmail.HypoVereinsbank.de
Received-From-MTA: DNS; localhost
Arrival-Date: Thu, 28 Oct 1999 06:37:31 GMT

Final-Recipient: RFC822; pgcc AT delorie DOT com
Action: failed
Status: 5.5.2
Remote-MTA: DNS; delorie.com
Diagnostic-Code: SMTP; 500 Hey, offensive language ("h?ll") is not allowed.  <dj AT delorie DOT com>
Last-Attempt-Date: Thu, 28 Oct 1999 06:38:45 GMT

- --GAA00730.941092725/vwmail.HypoVereinsbank.de
Content-Type: message/rfc822

Return-Path: <herbert DOT wengatz DOT extern AT ae3 DOT hypo DOT de>
Received: from fwproi1.HypoVereinsbank.DE (localhost [127.0.0.1])
	by vwmail.HypoVereinsbank.de ($Revision: 1.1 $)  with ESMTP id GAA00724;
	Thu, 28 Oct 1999 06:37:31 GMT
Received: from mailgate2.hypovereinsbank.de (mailgate2.fmis.de [129.0.1.149])
	by fwproi1.HypoVereinsbank.DE ($Revision: 1.1 $)  with ESMTP id GAA29539;
	Thu, 28 Oct 1999 06:37:51 GMT
Received: from ae3.hypo.de (sv002mz3.ae3.hypo.de [129.22.4.2])
	by mailgate2.hypovereinsbank.de ($Revision: 1.13 $)  with SMTP id IAA28729;
	Thu, 28 Oct 1999 08:37:47 +0200 (MET DST)
Received: from w9205mbl.fmis.hypovereinsbank.de by ae3.hypo.de (SMI-8.6/SMI-SVR4)
	id IAA19320; Thu, 28 Oct 1999 08:38:39 +0200
Received: from w9205mbl by w9205mbl.fmis.hypovereinsbank.de (8.8.8+Sun/SMI-SVR4)
	id IAA26337; Thu, 28 Oct 1999 08:37:41 +0200 (MET DST)
Message-Id: <199910280637 DOT IAA26337 AT w9205mbl DOT fmis DOT hypovereinsbank DOT de>
X-Exmh-Isig-Folder: inbox
X-Exmh-Isig-CompType: repl
To: Dieter =?iso-8859-1?Q?N=FCtzel?= <dieter DOT nuetzel AT myokay DOT net>
cc: Herbert DOT Wengatz AT consol DOT de, pgcc AT delorie DOT com,
        herbert DOT wengatz DOT extern AT ae3 DOT hypo DOT de
Subject: Re: Q: I plan to buy me an Athlon (aka AMD K7). What can I expect from the 
In-reply-to: Your message of "Wed, 27 Oct 1999 23:03:40 MET DST."
             <381776BC DOT 74D6E9F2 AT myokay DOT net> 
Mime-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
Date: Thu, 28 Oct 1999 08:37:41 MET DST
From: Herbert Wengatz 42850 <herbert DOT wengatz DOT extern AT ae3 DOT hypo DOT de>

+> <!doctype html public "-//w3c//dtd html 4.0 transitional//en">
+> <html>
+> Hello Herbert,

Hello Dieter,

first things first: Sorry, I don't take the time to remove all the HTML-garbage
from your posting. - Sorry for all the other who have to read this now...
(perhaps you should better decide to keep mails as plain ASCII, without HTML?)

+> <p>I own an Athlon 500 with MSI MS-6167 since mid August.
Great!

+> <br>So you are not the first, here...:-)
Even better! - So I'm the second? ;)

+> <p>The Athlon is more PPro/PII/PIII compatible than K6/K6-2/K6-III, I think.

I guess that's difficult to say. - Yesterday I started reading the specs from
AMD (you can download them right from their site and even order a free CD, which
can be updated on a regular basis, which contain a lot of stuff for developing
software espeially for the Athlons - optimization hints, you know..)

>From what I read until now, the major improvements in speed (comparing to, I guess,
an 486?) can be made by (according to AMD Document #22007):

- - Memory Size and alignment issues - Avoid Memory size mismatches and align data
  where possible 

	QWORD adresses are aligned by adresses divisible by 8
	DWORD ...... div by 4
	WORD .. div by 2
	TBYTE .. div by 8

	(Was that news for anybody? ;-) )

- - Use the 3Dnow! PREFETCH and PREFETCHW instructions (is that already in the egcs
  available? - Should be in the 3dnow!-extensions...) This boosts the effective
  bandwidth to the Athlon CPU and should be preferred over their MMX-Counterparts!

- - Select DirectPath over VectorPath Instructions

These three are the official "essential" optimizations for the Athlon.
And here are the "secondary" optimizations:

- - Load-Execute Instruction usage -- *Use* Load-Execute instructions - Avoid load-
  execute floatingpoint instructions with integer operands (Load-execute integer
  instructions can be decoded at a rate of three per cycle...) If this has to be done
  nevertheless, better use separate FILD and arithmetic instructions.

- - Take advantage of write-combining (goes especially to writers of OSes, device
  drivers and BIOSes...) - The Athlon has a very agressive write-combining
  algorithm, which improves performance significantly...

- - Use 3Dnow! Instructions (They are twice as fast as their x87 counterpart...)

- - Avoid Branches dependent on Random Data. (Possible misprediction up to 50% of the
  time! So better think of "branchfree" alternative code.

- - Avoid placing Code and Data in the SAME 64-Byte Cache line. (Self modifying
  code should be avoided...) Sharing code and Data in the same 64-byte cache-line
  may cause the L1-Cache to thrash. Since all other CPUs before had only 32-byte
  Cache at this place, this has to be taken care of!

+> <p>Normally I only use
+> <p>'-O -funroll-loops'
+> <p>because I found with 'dgemm.c' (little bench from QuantX, the cool Alpha
+> company) that it is the fastest for the moment.

Unrolling loops is also essential for the modern, superscalar CPUs. They are
faster with unrolled code, but the binaries tend to get bigger...
Small loops should be unrolled completely.

Other hints: 
- - Avoid unnecessary integer division like h?ll. <- Here it is! Was I playing Quake or what? :-(((

What I also found rather interesting is, that K6 and K7 code can be blended quite well.
I guess this was already supposed to be... ;)

But I still would like to have the possibility to decide, whether I want pure K6 code,
pure K7 code or mixed.

+> <p>For Mesa-3.1 beta3/CVS I use
+> <p>setenv CFLAGS '-O6 -march=pentiumpro -pipe -malign-double -mstack-align-double
+> -funroll-loops -DSHM'

arch=pentiumpro should be OK. As you said already, the K7 is closer to the PII+
CPUs. The alignment was already mentioned. Should be OK. But I read that stack-
alignment should go to 32 bits. - Does that match with "-mstack-align-double" ?

+> <p>for the libraries and then '-O -funroll-loops', again for the demo programs.
+> <br>Marc?
+> <p>Both work with 2.2.xx and 2.3.xx for me.

That's great news! :-)

+> <p>SunWave1>cat /proc/version<br>
+> Linux version 2.3.24 (root AT SunWave1) (gcc version pgcc-2.95.1 19990816
+> (release)) #1 Wed Oct 27 04:03:48 MET 1999
+> <pre>SunWave1>cat /proc/cpuinfo
+> processor&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : 0
+> vendor_id&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : AuthenticAMD
+> cpu family&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : 6
+> model&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : 1
+> model name&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : AMD-K7(tm) Processor
+> stepping&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : 2
+> cpu MHz&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : 499.040637
+> cache size&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : 512 KB
+> fdiv_bug&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : no
+> hlt_bug&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : no
+> sep_bug&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : no
+> f00f_bug&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : no
+> coma_bug&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : no
+> fpu&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : yes
+> fpu_exception&nbsp;&nbsp; : yes
+> cpuid level&nbsp;&nbsp;&nbsp;&nbsp; : 1
+> wp&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : yes
+> flags&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : fpu vme de pse tsc msr 6 mce cx8 sep mtrr pge 14 cmov fcmov 22 mmx 30 3dnow
+> bogomips&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; : 498.07</pre>
+> I&nbsp;do 3D medical visualization as my thesis.

You mean you are *NOT* doing &nbsp; for your thesis? ;-)

Sorry, I couldn't resist. - Of course my mailtool displays your mail correct.
But look, what you made me reply it!

+> <p>Regards,
+> <br>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; Dieter
+> <pre>--&nbsp;

Best regards also,

&nbsp;&nbsp;&nbsp;Herbert

+> Dieter N&uuml;tzel
+> Graduate Student, Computer Science
+> 
+> University of Hamburg
+> Department of Computer Science
+> Cognitive Systems Group
+> Vogt-K&ouml;lln-Stra&szlig;e 30
+> D-22527 Hamburg, Germany
+> 
+> email: nuetzel AT kogs DOT informatik DOT uni-hamburg DOT de
+> @home: dieter DOT nuetzel AT myokay DOT net</pre>
+> </html>
</ascii>
 
PS: Please *all* reply to me directly! - I'm NOT on this Mailinglist!

German: n., A veller dot vas mighty broud (and mighty flat) to coom vrom
Deutschland, don't it? - Ambrose Bierce, The Devils Dictionary


- --GAA00730.941092725/vwmail.HypoVereinsbank.de--


------- End of Forwarded Message


"Never ascribe to malice what can be explained by human stupidity."
Robert A. Heinlein