Cone rendering regression

[stukowski]

After upgrading from v0.13.0 to next_release (1e19613), I noticed a significant deterioration in the rendering of cones. Here's a before-and-after image rendered using these two versions of VisRTX with identical settings.

v0.13.0:

Current:

[tarcila]

@stukowski Thanks for the report. Shoot in the dark here, to try and narrow down the issue: can you check whether turning AO off works around the bug?

[stukowski]

After setting ambientSample=0 it gets better, but the ring artifact in the projection center is still there, I would say.

[stukowski]

Just in case this is relevant: The height and radius of these cones are on the order of 1.0 world units. I mean if the numerical precision in the cone intersector plays a role here, and the problem is only visible on the right length scale.

[tarcila]

Agree, it looks like a precision issue and would match latest changes. Can you a have a look at #334 and see if this fixes the issue?

[stukowski]

Yes, it looks much better when using your PR version. That would be good enough for our purposes. There’s just a small artifact in the immediate vicinity of the vertex. For cones of size 1, it’s almost invisible; for cones of size 20, as in the following image, it’s visible up close.