Subject: bin/14199: gcc optimisation bug
To: None <gnats-bugs@gnats.netbsd.org>
From: Mario Kemper <magick@bundy.zhadum.de>
List: netbsd-bugs
Date: 10/09/2001 22:52:01
>Number:         14199
>Category:       bin
>Synopsis:       gcc optimisation bug
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    bin-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Tue Oct 09 13:53:00 PDT 2001
>Closed-Date:
>Last-Modified:
>Originator:     Mario Kemper
>Release:        -current as of oct 3
>Organization:
Mario Kemper
magick@zhadum.de
>Environment:
	
System: NetBSD bundy 1.5Y NetBSD 1.5Y (BUNDY) #1: Wed Oct 3 16:19:30 CEST 2001 root@bundy:/usr/netbsd-src/src/sys/arch/i386/compile/BUNDY i386
Architecture: i386
Machine: i386
>Description:

gcc 2.95.3 (aka NEW_TOOLCHAIN) has an optimisation bug when using -O2 together 
with long long. I first noticed that when i compiled squid with the new gcc 
and it crashed on every page to be loaded. As i'm obviously not the only one 
who found that out, you can read a full analysis on:

http://www.geocrawler.com/archives/3/145/2001/8/50/6506067/

This affect IMHO every program that uses long long not only squid

>How-To-Repeat:
Compile squid with -O2 ans see it crash.

>Fix:
The *fix* for squid in 2.4.2 turns out to be a simple downgrade of the 
optimization. Maybe we should revert to -O in the NetBSD-tree to be on 
the safe side.

>Release-Note:
>Audit-Trail:
>Unformatted: