In the following two graphs,
- the red lines are borders of the perspective-projection frustum,
- the blue lines are to be rasterized,
- the green lines are perspective-projection lines, which are wrong.
How do modern GPU pipelines do such clipping given only a perspective-projection matrix?