This method is more flexible and somewhat simpler.

DP2ADD returns a scalar, and fglrx (correctly) complains that it can't
assign that to a 3 component vector.

glStencilFuncSeparateATI does not take a face argument, instead it
sets the front and back facing functions at once. This means the
renderstate_stencil_twosided helper function is somewhat pointless for
this extension.