Am 21.03.2013 17:17, schrieb Michel Dänzer:
On Don, 2013-03-21 at 17:12 +0100, Christian König wrote:
Am 21.03.2013 15:06, schrieb Michel Dänzer:
When I tried this earlier, something broke. Unfortunately, I can't seem
to remember or dig up if it was a piglit test or e.g. one of the sampler
demos in mesa/demos/src/glsl/. Did you test the latter with this change?
Similar concern for patch 4, the rest of the series looks good to me.
It breaks "texrectmany" or something like that, and that was actually my
testcase that it now works properly.
You mean this change used to break that test but no longer does?
Yes, exactly. The basic problem was that LLVM didn't recognized our
loads as loads from constant memory, and so refused to move them with
respect to any intrinsic that's not marked as readonly (especially KILL
and EXP).
I haven't tested the sampler demos with it, but lightsmark stresses that
quite a bit and I'm pretty sure that it works now.
Please just test them and be 100% sure. :)
I've just did so and they still seems to work (well lets say that they
still doesn't crash).
Do I have your rb for this series?
Cheers,
Christian.
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
