From:   Steve Lord <lord@sgi.com>
Message-Id: <200008271414.JAA19505@jen.americas.sgi.com>
To:     Thomas Graichen <graichen@innominate.de>,
        thomas.graichen@innominate.de
cc:     linux-xfs@oss.sgi.com
Subject: Re: xfs & ppc & egcs 
In-reply-to: Your message of "27 Aug 2000 09:07:02 GMT
Date:   Sun, 27 Aug 2000 09:14:15 -0500
Sender: owner-linux-xfs@oss.sgi.com
Precedence: bulk


We do know that there are problems in the generated code for gcc 2.95
on ia32. We do not know if it is the compiler back end or front end
which causes them, and hence will the problems propogate across
architectures. 

We had isolated a specific problem (I cannot off the top of my head
remember it - apart from being shift related). We could probably modify the
XFS code which was affected by this problem, but finding all instances
would not be easy. I will try and dig out the failure case and see if
fixing that makes things better. 

Steve

p.s. those files could not be more important, that is a fair percentage
of the journalling code, and all of the buffering code in XFS!

> just installed an egcs on the ppc machine and rebuilt the xfs kernel
> with it (4 files did not compile due to assembler problems - maybe
> i need to downgrade the binutils too for which i don't have the
> diskspace right now - but lets hope those 4 files are not that
> important ... they are btw.: xfs_log.c, xfs_trans_buf.c, page_buf.c,
> page_buf_io.c - i recompiled those 4 files with gcc 2.95 then)
> 
> the final result of this little experiment was: i get the same
> problems on the ppc as with gcc 2.95 - so it looks like the problems
> are more generic and not directly related to gcc vs. egcs ... will
> try to investigate all this again deeper in the next days (i think
> i will have a closer look at xfs_db and how it works - to see how
> exactly the on disk structure looks like - is this a good idea ?)
> 
> t
> 
> -- 
> thomas.graichen@innominate.de
> technical director                                       innominate AG
> clustering & security                                networking people
> tel: +49.30.308806-13  fax: -77                   http://innominate.de