Hi Dave, Am Mittwoch, den 10.01.2018, 16:48 +1000 schrieb Dave Airlie: > This is an attempt to add tessellation support to the SB backend. >
I tried to dig a bit more in the failing piglits, specifically "1in-1out" that passed with your WIP branch form Jan/9. Now, with sb it fails by drawing some pseudo-random patterns that would indicate that for some of the patches uninitialized memory is accessed when drawing the upper left sub-quad of each patch. The pattern may change when pressing some key. However, by inspecting disassembly I was not able to detect any change in the LDS addresses accessed by the optimized shaders vs. the original ones, only the access order is (mostly) reversed (which is probably the most notable change w.r.t. the WIP branch). When I change the tess-factors to 2.0 (instead of 3.0) and also the related values in the vertex and tes shader, the test passes, which should confirm that the LDS access is correct. Do I change the 3 only in tes to a positive value below 2.5 then the drawn pattern doesn't show random behaviour. Also if in the assignment to the input[].z value in the vertex shader the 64 is replaced by 64.0, making the evaluation there a floating point operation, then the random patterns in the piglit screen output disappear. Since this z-value is not used for the coordinate evaluation it should have no influence on these patters. Best, Gert _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
