Hmm, this seems like an idea worth thinking about. The D3D behavior the
proposed extension addresses is part of the D3DDECLUSAGE_POSITIONT /
D3DFVF_XYZRHW vertex input semantics.

For now I'm opposed to making this a vertex shader control though. The point of
POSITIONT / XYZRHW is to skip vertex processing altogether, so handling
POSITIONT semantics in a shader seems a bit off. Furthermore, the clipping
behavior of POSITIONT depends on the depth test, so this needs a separate
control anyway (or replicating the depth test interaction, which seems somewhat
ugly to me because its done by a different stage in the rendering pipeline).