Date: Thu, 8 Jun 2000 11:36:34 -0400
From: Thor Lancelot Simon <tls@rek.tjls.com>
To: Michael Sokolov <msokolov@ivan.Harhan.ORG>
Cc: pups@minnie.cs.adfa.edu.au
Subject: Re: Newer BSD thingies....nice but then again....
Message-ID: <20000608113634.A26968@rek.tjls.com>
Reply-To: tls@rek.tjls.com
References: <0006072246.AA16781@ivan.Harhan.ORG>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.2i
In-Reply-To: <0006072246.AA16781@ivan.Harhan.ORG>; from msokolov@ivan.Harhan.ORG on Wed, Jun 07, 2000 at 05:46:55PM -0500
Sender: owner-pups@minnie.cs.adfa.edu.au
Precedence: bulk

On Wed, Jun 07, 2000 at 05:46:55PM -0500, Michael Sokolov wrote:
> jkunz@unixag-kl.fh-kl.de wrote:
> 
> > Oh, yes. My VS4000m60 needs only 36 hours to go through a "make build".
> > This is pure luxury.=20
> 
> And 4.3BSD-Quasijarus completes its make build on my CSRG dev mill, which is a
> KA655 (3.8 VUPs, whereas your KA46 is 12 VUPs), in a little under 4 hours. The
> GENERIC vmunix kernel is another 30 minutes.

My experience with compilers on the VAX leads me to believe that the
substantial "savings" seen over NetBSD or post-4.3 BSD distributions here
is almost entirely due to the compiler and options used.  If Quasijarus
builds like CSRG 4.3 did, with pcc, it can't even use the optimizer *at all*
for the kernel build, due to severe bugs; either way, pcc runs a lot faster
than gcc though it generates code that runs a whole lot slower.

I'd be willing to bet that gcc -O0 would build NetBSD at least ten times
as fast as gcc -O2; the VAX is (as we all know ;-)) a "rather complex"
processor, with "rather complex" instruction patterns, gcc is not the
swiftest of compilers in the first place, and it does a *lot* of work.

Slow machines *are* good for demonstrating how good your compiler is;
I recall that rebuilding "compress" with gcc on my 750, way back when,
pretty much doubled the amount of Usenet news I could handle in a day. :-)

Thor