Hi, On Friday, October 24, 2014 17:17:34 Ilia Mirkin wrote: > Can you verify that your test fails with an incorrect implementation? > I just set nvc0 so that it always acts as if clip_halfz == 0, and the > test passed... That's interesting.
I can't test with the nvc0, but when I hard set clip_halfz to zero in st_atom_rasterizer.c, then I get a failing clip-control test with llvmpipe and radeonsi. Also interrestingly softpipe does not fail either. Probably nvc0 and softpipe are clipping based on the final depth buffer values in [0, 1] instead of the clip space z values either in [-1, 1] or [0, 1] depending on clip control depth state? The unclipped NV_depth_buffer_float extension would have made this much more sensitive to what is clipped exactly when. Without the unclipped depth buffer values from NV_... I would think it's not important which of the above variants is taken for clipping or clamping. The clip-control-depth-precision test is supposed to be independent of clip_halfz. This tests if the roundoff characteristics are preserved as targeted in the spec. > Also, you sent the patches inline just fine for mesa... can you do it > here as well? It enables people to review your patches... Will do ... Greetings Mathias _______________________________________________ Piglit mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/piglit
