[Bug 103632] [i965 bisected] commit 6ee4b352c90e93c82d900328716c1068a7188b91 regress ASTROKILL and EVERSPACE on Haswell

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=103632

Kenneth Graunke <kenneth at whitecape.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|NEW                         |RESOLVED

--- Comment #4 from Kenneth Graunke <kenneth at whitecape.org> ---
Thanks for tracking that down, Darius!  Apparently there was a pretty nasty
bug...the L3 config code was doing:

      const struct gl_program *prog =
         brw->ctx._Shader->CurrentProgram[stage_states[i]->stage];

but because we'd failed to set brw->cs.stage = MESA_SHADER_COMPUTE, it was
initialized to 0 aka MESA_SHADER_VERTEX.  So, when calculating the data cache
needs, prog_data properly pointed at the compute shader, but prog incorrectly
was the vertex shader.

That means Jason's commit (the one regressing things) caused us to start
considering the vertex shader's number of images rather than the compute
shader.  That was probably near 0, so we thought we didn't need to bother
allocating any cache to DC.  But, his patch was fine...and in fact, the bogus
prog had been around for quite a while.  The real bug was the one I fixed.

While looking at this, I found another bug in the area, and sent a patch:
https://patchwork.freedesktop.org/patch/188561/

Again, thanks for helping us catch these issues!

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20171116/6b2b610a/attachment-0001.html>