On Thu, May 10, 2012 at 7:10 AM, Brian Paul <[email protected]> wrote: > On 05/09/2012 05:56 PM, Jordan Justen wrote: >> >> One additional note: >> This test passed with NVidia's driver, but failed with llvmpipe. >> >> I'm not certain if the test or llvmpipe is to blame. So, feedback >> either pointing out a test flaw, or known issue with llvmpipe would be >> appreciated. > > > It looks like you've exposed a new bug with VBOs. You're storing both the > vertex arrays and vertex indexes in a single VBO (which is a little > unusual). Looks like something in Gallium is choking on that. > > I modified your code to use separate VBOs for vertex data and indexes and > then the test passes with softpipe/llvmpipe. Perhaps you can roll the > attached patch into your change.
Sure. Can you provide your Signed-off-by for the patch? > I'd like to have a separate test to check the case of one VBO containing > both vertex data and index data. Would you be interested in writing such a > test? Sure. Something like tests/spec/arb_vertex_buffer_object/combined-vbo.c? > Also, the primitive-restart test could be improved in another way. After > drawing the test pattern, we should probe pixel (0,0) to check that it's > black. As it turns out, when there are primitive restart failures/bugs, we > often get a bogus vertex at (0,0) and the triangles/lines are > distorted/stretched to (0,0). When I was debugging the llvmpipe failure I > noticed that one of the line strip tests was passing the test, but was in > fact drawn incorrectly. Probing (0,0) would have caught the false positive. Okay, I'll add this. -Jordan _______________________________________________ Piglit mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/piglit
