[Mesa-dev] [Bug 27512] Illegal instruction _mesa_x86_64_transform_points4_general

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Wed Feb 3 17:11:55 UTC 2016


https://bugs.freedesktop.org/show_bug.cgi?id=27512

--- Comment #18 from Roland Scheidegger <sroland at vmware.com> ---
(In reply to Timothy Arceri from comment #17)
> (In reply to Michael Harder from comment #15)
> > It worked for a few days but now I get this:
> > 
> > [Thread debugging using libthread_db enabled]
> > Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
> > Core was generated by `/usr/lib/x86_64-linux-gnu/kodi/kodi.bin --standalone'.
> > Program terminated with signal SIGILL, Illegal instruction.
> > #0  _mesa_x86_64_transform_points4_general () at x86-64/xform4.S:72
> > 72              prefetcht1 16(%rdx)
> > [Current thread is 1 (Thread 0x7fd24af779c0 (LWP 797))]
> 
> I ran into this problem with my new old hardware I've been playing with
> recently.
> 
> The problem can be reproduced running a number of piglit tests such as:
> 
> ./bin/fbo-stencil readpixels GL_DEPTH24_STENCIL8 -auto -fbo
> 
> 
> The patch doesn't fix the problem as it seems prefetcht1 doesn't like
> offsets.
> If I change for example prefetcht1 16(%rdx) -> prefetcht1 (%rdx) removing
> the offset for all instances the piglit will now pass. Not sure how to work
> around this problem.

That doesn't make sense to me. The offset is just part of the memory operand.
Unless the assembler encodes it wrong I can't see why that wouldn't work (which
I would think to be unlikely, but the locality hints are also encoded into the
mod r/m byte - what's the encoding of the instruction?)
I suppose a solution would just be to ditch prefetch - as was pointed out it's
not really far ahead enough in any case, even k8 and p4 had primitive hw
prefetchers which should make such a simple software prefetch completely
unnecessary.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20160203/de7a866c/attachment-0001.html>


More information about the mesa-dev mailing list