Am 26.01.2010 um 19:23 schrieb Henri Verbeet:
On 26 January 2010 18:56, Stefan Dösinger stefandoesinger@gmx.at wrote:
It would be interesting to see how D3DCMP_GREATER behaves with clipping off. Does the clamping occur before the depth test, or after the test but before writing the values into the depth buffer. (ie, is 3.0 > 1.0 or is 3.0 == 1.0)
It's not completely trivial to reliably test that due to precision differences between the vertex data and the depth buffer
Why? I'd expect that if you first draw something with depth clamping enabled with z=2(so it ends up at 1.0) then it really ends up at 1.0 and not 0.999998. If the Z buffer is cleared without drawing anything I'd expect it to contain an exact 1.0 as well. If there are precision problems in either situation I think the card/driver would be pretty broken.
ARB_depth_buffer_float doesn't describe interactions with ARB_depth_clamp, which might imply we'll have to disable depth clamping for floating point depth buffers, but I doubt floating point depth buffers are particularly useful without shaders anyway.
Did you try depth clamping with shaders? Wrt vertex shaders I don't really expect surprises on either side, and wrt pixel shaders there are at least no surprises on the GL side(since we have that replacement pipeline and are using shaders anyway)