Is there a site for unofficail patches?

Is there a site for unofficail patches?

[jfm2]

I have found that gcc 2.95.2 tends to coredump in Linux (RedHat 6.2)
when compiling with -O3 -mcpu=i686 -mpreferred-stack-boundary=2

The most notorious example is that it fails to recompile itself.  The
culprit is the preferred-stack-boundary flag since after removing it
then everything works fine.  Is there a place where I could look for a
patch?

-- 
			Jean Francois Martinez

Project Independence: Linux for the Masses
http://www.independence.seul.org

Re: Is there a site for unofficail patches?

[Joe Buck]

> I have found that gcc 2.95.2 tends to coredump in Linux (RedHat 6.2)
> when compiling with -O3 -mcpu=i686 -mpreferred-stack-boundary=2
> 
> The most notorious example is that it fails to recompile itself.  The
> culprit is the preferred-stack-boundary flag since after removing it
> then everything works fine.  Is there a place where I could look for a
> patch?

The problem with unusual flag combinations and high optimization levels
is that you're in untested territory.  It may well be that no one besides
you ever tried this combination before, so it is very unlikely that you'll
find a patch anywhere.

Re: Is there a site for unofficail patches?

[jfm2]

> 
> > I have found that gcc 2.95.2 tends to coredump in Linux (RedHat 6.2)
> > when compiling with -O3 -mcpu=i686 -mpreferred-stack-boundary=2
> > 
> > The most notorious example is that it fails to recompile itself.  The
> > culprit is the preferred-stack-boundary flag since after removing it
> > then everything works fine.  Is there a place where I could look for a
> > patch?
> 
> The problem with unusual flag combinations and high optimization levels
> is that you're in untested territory.  It may well be that no one besides
> you ever tried this combination before, so it is very unlikely that you'll
> find a patch anywhere.
> 
> 
> 

In fact I was using a subset of Linux Mandrake's flags with the only
difference that Mandrake uses -march=i586 while I use -mcpu=i686.

But using plain -03 -mcpu=i686 -mpreferred-stack-boundary crashed gcc
while rebuilding itself.

-- 
			Jean Francois Martinez

Project Independence: Linux for the Masses
http://www.independence.seul.org

Re: Is there a site for unofficail patches?

[Philipp Thomas]

* jfm2@club-internet.fr (jfm2@club-internet.fr) [20000822 00:32]:

> I have found that gcc 2.95.2 tends to coredump in Linux (RedHat 6.2)
> when compiling with -O3 -mcpu=i686 -mpreferred-stack-boundary=2
                      ^^^
This won't buy you anything! -O3 only gets you automatic inlining which is
useless for the Linux kernel as functions are explicitely marked inline
where necessary for speed. There's a reason why the kernel makefiles use -O2
by default.

Philipp

-- 
Philipp Thomas <pthomas@suse.de>
Development, SuSE GmbH, Schanzaecker Str. 10, D-90443 Nuremberg, Germany

#define NINODE  50              /* number of in core inodes */
#define NPROC   30              /* max number of processes */
 	-- Version 7 UNIX for PDP 11, /usr/include/sys/param.h

Re: Is there a site for unofficail patches?

[jfm2]

> * jfm2@club-internet.fr (jfm2@club-internet.fr) [20000822 00:32]:
> 
> > I have found that gcc 2.95.2 tends to coredump in Linux (RedHat 6.2)
> > when compiling with -O3 -mcpu=i686 -mpreferred-stack-boundary=2
>                       ^^^
> This won't buy you anything! -O3 only gets you automatic inlining which is
> useless for the Linux kernel as functions are explicitely marked inline
> where necessary for speed. There's a reason why the kernel makefiles use -O2
> by default.
> 

I knew this.  I was using the above flags to recompile gcc 2.95.2 with
itself not the kernel.  And it coredumped.  Since it works for
Mnadrake people with difference that they use -march=i586 then
conclusion is that the crash is due to use of -mcpu=i686 in conjuction
with preferred-stack-boundary.

-- 
			Jean Francois Martinez

Project Independence: Linux for the Masses
http://www.independence.seul.org