I think we should just drop draw calls with a 0-sized index buffer. For GL_AMD_pinned_memory, I think we can just map whole pages that intersect the mapped range.