I've discovered by accident that the mesa state tracker is trying to use REPEAT wrap mode with texture rectangles - this happens with piglit tex-miplevel-selection texture() 2DRectShadow -fbo -auto for instance though I guess others as well. Needless to say the test fails in llvmpipe because I was lazy and didn't make repeat wrap mode work with non-normalized coords (ok it's quite possible actually the test fails due to a completely unrelated reason but this just seems bogus). This would happen at least with any gallium based driver, but I wonder who's fault this really is. The problem here seems to be that while texture objects get initialized correctly for texture rectangles to CLAMP_TO_EDGE, sampler_objects do not (and since they aren't tied to a target, I guess they cannot be correctly initialized).
Now after some digging, there's actually some bit in the spec about exactly this case in the section about sampler objects: "If state is present in a sampler object bound to a texture unit that would have been rejected by a call to TexParameter* for the texture bound to that unit, the behavior of the implementation is as if the texture were incomplete. For example, if TEXTURE_WRAP_S or TEXTURE_WRAP_T is set to REPEAT, MIRRORED_REPEAT, or MIRROR_CLAMP_TO_EDGE on the sampler object bound to a texture unit and the texture bound to that unit is a rectangle texture, the texture will be considered incomplete." So my guess is this isn't actually supposed to work and the test is bogus? It also looks like it isn't treated in mesa as incomplete. Roland _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
