[Bug 89580] Implement a NIR -> vec4 pass

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=89580

--- Comment #30 from Kenneth Graunke <kenneth at whitecape.org> ---
Hi Eduardo, Antía,

FYI, I've mostly gotten SIMD8 Geometry Shaders (new on Broadwell) working -
which is the first backend using NIR for geometry shaders.

You can find my work-in-progress branch here:
http://cgit.freedesktop.org/~kwg/mesa/log/?h=simd8gs

I've done things a little differently than the old vec4 backend - namely, I
moved the vertex count (how many times EmitVertex() has been called) into NIR,
introducing new intrinsics.

My thinking is that NIR's optimizations should be able to constant
fold/propagate the counter values, so in common/simple cases, NIR will simply
eliminate the "safety check" if statements, as they'll be if (true).  Knowing
that the vertex count is an immediate may allow us to avoid per-slot-offset
messages as well, and should also help us use the "Static Vertex Count" feature
on Broadwell.

It's not quite done yet: I haven't actually added code to do safety checks yet;
there are a few Piglit regressions still, and a few tests try to use 700
registers for the payload data :)  But I figured I'd share it so you can see
where I'm going with this, and in case you wanted to reuse the NIR intrinsics.

-- 
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/20150713/06e428a1/attachment.html>