GCC 3.2.x miscompiles X

GCC 3.2.x miscompiles X

[Aaron Lehmann]

In the unstable branch of Debian, a bug has recently been introduced
which breaks hardware-accelerated GL for Radeon users on ia32. The bug
report is at http://bugs.debian.org/178242. This one hit me particularly
hard, as I have a Radeon in both my laptop and my desktop.

A comment in the bug report linked the problem to GCC, so I attempted
to compile Debian's XFree86 packages with today's GCC mainline. Sure
enough, this fixed the bug.
http://marc.theaimsgroup.com/?l=dri-devel&m=103851238913228 suggests
that the problem is the fault of GCC mixing FPU code with MMX code.
The same message implies that this has been fixed in GCC 3.2.1, but I
was able to reproduce this bug in GCC 3.2.2 20030124 (Debian prerelease).
Since this problem seems to have been fixed in HEAD, and possibly in
the 3.3 branch (I haven't tested it), I'm wondering if anyone can
point me towards the patch that fixed the bug. I looked in GNATS but
was not able to find an associated PR. I'm sorry that I can't provide
a worthwhile testcase at this point, only a vague hypothesis about the
problem.


Thanks,
Aaron Lehmann
Debian Developer

Re: GCC 3.2.x miscompiles X

[Aaron Lehmann]

On Wed, Jan 29, 2003 at 08:24:33PM -0800, Aaron Lehmann wrote:
> Since this problem seems to have been fixed in HEAD, and possibly in
> the 3.3 branch (I haven't tested it)

Update - 3.3 DOES have the bug.

Re: GCC 3.2.x miscompiles X

[Andreas Jaeger]

Aaron Lehmann <aaronl@vitelus.com> writes:

> On Wed, Jan 29, 2003 at 08:24:33PM -0800, Aaron Lehmann wrote:
>> Since this problem seems to have been fixed in HEAD, and possibly in
>> the 3.3 branch (I haven't tested it)
>
> Update - 3.3 DOES have the bug.

Do you have a small, self contained testcase?  Please file this
together with the bugreport using the gccbug script,

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj