Avoid calling `GPU_indexbuf_add_line_verts` and the triangle version of that function. It's faster to avoid function calls and just write to the data arrays directly. I did some very rough tests and observed about a 10% improvement in runtime for the entire index buffer creation process.