[Bug 84104] Ring hang with geometry shaders on SNB

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=84104

--- Comment #6 from Iago Toral <itoral at igalia.com> ---
I think the case that this still hangs even in the presence of an empty GS is
remarkable. So If I use this GS:

#version 150
layout(triangles_adjacency) in;
layout(points, max_vertices=1) out;
void main()
{
}

the GS stage will not emit any vertices. I have even modified the
implementation to eliminate the loop that handles the URB writes so that the
native code we generate is trivial (the loop would not be executed anyway, but
thisway  we see that the resulting program is trivial in this case), and the
hang persists. With this shader, and removing the emission loop, the code
generated in with INTEL_DEBUG=gs is reduced to this and still hangs:

   START B0
   clear r0.2
mov(1)   g0.2<1>UD    0x00000000UD    { align1 WE_all compacted };
   gen6 prolog
mov(8)   m1<1>UD      g0<4,4,1>UD     { align16 WE_all 1Q compacted };
   gen6 thread end: EOT
send(8)  null         m1<4,4,1>F
         urb 0 urb_write interleave complete mlen 1 rlen 0 { align16 1Q EOT };
   END B0

I think this discards that the problem is related to the output primitive type
we emit in the GS stage and generally, any aspect of the code that we generate
for the GS stage.

Kenneth, I recorded an AUB trace for this case too, since in this case we are
not generating any DUAL OWORD BLOCK messages or even generating any output in
the GS at all, I wonder if the emulator still complains about something... if
it does it might be more useful:

http://people.igalia.com/itoral/intel.hangs_empty_gs.aub


Of course, the above setup works fine if any of the conditions that lead to the
hang are removed, that is:

- if we use glDrawArrays instead of glDrawElements or
- if we render a primitive other than GL_TRIANGLE_STRIP_ADJACENCY or
- if the list of indices we use with glDrawElements does not have 8 consecutive
repeating vertices or at least 8 repeated vertices in any subset of 9
consecutive indices.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20140922/afb961b1/attachment.html>