From owner-freebsd-hackers  Sun Jan 12 21:42:10 1997
Return-Path: <owner-hackers>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.4/8.8.4) id VAA20399
          for hackers-outgoing; Sun, 12 Jan 1997 21:42:10 -0800 (PST)
Received: from rocky.mt.sri.com (rocky.mt.sri.com [206.127.76.100])
          by freefall.freebsd.org (8.8.4/8.8.4) with ESMTP id VAA20394
          for <hackers@freebsd.org>; Sun, 12 Jan 1997 21:42:08 -0800 (PST)
Received: (from nate@localhost) by rocky.mt.sri.com (8.7.5/8.7.3) id WAA23845; Sun, 12 Jan 1997 22:41:45 -0700 (MST)
Date: Sun, 12 Jan 1997 22:41:45 -0700 (MST)
Message-Id: <199701130541.WAA23845@rocky.mt.sri.com>
From: Nate Williams <nate@mt.sri.com>
To: Joao Carlos Mendes Luis <jonny@mailhost.coppe.ufrj.br>
Cc: dg@root.com, hackers@freebsd.org
Subject: Re: Compiling kernel with optimisation
In-Reply-To: <199701122231.UAA22599@gaia.coppe.ufrj.br>
References: <199701121724.JAA25580@root.com>
	<199701122231.UAA22599@gaia.coppe.ufrj.br>
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Joao Carlos Mendes Luis writes:
> #define quoting(David Greenman)
> // >I just noticed that when compiling a kernel it is done with the -O flag.
> // >Would there be much speed improvement in the sytem if it was done with
> // >-O3? Would this break the kernel or is the added time it takes to compile
> // >not worth the benfits?
> // 
> //    It has very little effect on performance and optimizations levels > "-O"
> // have traditionally been broken in gcc.
> 
> Well, the NetBSD team has managed to compile their kernel with -O6
> and -Wall, but they had to change lots of things..
> 
> I don't know what do you call "little effect on performance", but
> 5% gain would be enough to make me think about.

5% kernel improvement would end up being lost in the noise for 'overall'
system improvement.  And, the possible (probably in this case) loss in
stability due to a buggy optimizer and/or x86 support is simply not
worth it.  Even a 15% performance increase for the kernel might not even
show up.  (Sometimes kernel improvements can be losses in performance
due to cache busting and other assorted details.)


Nate