qt_D_GGX called from qt_specularGGXBSDF returns INF in some cases. When the color buffer is a floating point one, this gets written out as-is.

While these white pixels are not necessarily noticable on screen, this then causes vastly amplified problems when there is a postprocessing effect. With SSGI for example, we may end up with NaN values over a much larger area.

With SSGI on this can lead to results such as this in some frames:

d is zero in the following code: