• Ken Martin's avatar
    A number of transparency related issues fixed · 944ee020
    Ken Martin authored
    A few good fixes in here that are all related. The biggest is a
    fix to OpenGLActor to use the more comprehensive GetIsOpaque
    method in determining if an actor is opaque. The old code only
    looked at the property's opacity. This is why you would see folks
    setting somethings opacity to 0.99 to get it treated as translucent
    because the actor code wasn't handling it correctly. This change
    resulted in 3 new valid images for tests that were rendering
    translucent items as if they were opaque. I removed these tests
    from the old backend as I did not want to change the behavior
    for that backend.
    
    Next up is depth peeling. OpenGL2 depth peeling would render
    a number of peels up to the limit and then just stop. Any geometry
    left between the opaque layer and the last translucent layer was
    just thrown out.  This created very noticable artifacts and was bad.
    This change makes it so that when it gets to the last pass, it will
    render all remaining geometry using alpha blending. This is much
    better. Fix example if you get the limit to 20 layers and 20 was
    not enough to render all the geometry, this change renders 19
    layers using depth peeling, then in the 20th render it would add
    in all the remaining geometry as last layer. That last layer may
    have overlapping geometry which is rendered using alpha blending.
    
    Finally some changes to the PointGaussianMapper. It turns out that
    the point gaussian mapper is an translucent mapper which means it
    needs the first change, and it is a very tough case for depth peeling
    resulting in the second change. While at it I modified this mapper
    to render far fewer fragments, resulting is far fewer overlapping
    layers and fewer artifacts. Really when using PointGaussian you
    should turn off depthpeeling as it is not needed and actually makes
    it worse, but these changes will make it better if you happen to have
    it on.  Also cleaned up the mapper so that the radius is equal to
    the standard devisation of the gaussian, previously it was four
    standard deviations.
    
    Change-Id: I62d90b6e424ac0e939033387ed10ba8caac66095
    944ee020
socbal.py 13.1 KB