<html> <head> <base href="https://bugs.freedesktop.org/" /> </head> <body> <div> <a class="bz_bug_link bz_status_ASSIGNED " title="ASSIGNED - Illegal instruction _mesa_x86_64_transform_points4_general" href="https://bugs.freedesktop.org/show_bug.cgi?id=27512#c18">Comment # 18</a> on <a class="bz_bug_link bz_status_ASSIGNED " title="ASSIGNED - Illegal instruction _mesa_x86_64_transform_points4_general" href="https://bugs.freedesktop.org/show_bug.cgi?id=27512">bug 27512</a> from <a class="email" href="mailto:sroland@vmware.com" title="Roland Scheidegger <sroland@vmware.com>"> Roland Scheidegger</a> <pre>(In reply to Timothy Arceri from <a href="show_bug.cgi?id=27512#c17">comment #17</a>) > (In reply to Michael Harder from <a href="show_bug.cgi?id=27512#c15">comment #15</a>) > > 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.</pre> </div> <hr> You are receiving this mail because: <ul> <li>You are the assignee for the bug.</li> </ul> </body> </html>