The GTE truncated too early, and in too many places and there is a register for adding a fraction to the post-perspective-pre-truncate result to round instead.

The GTE truncated too early, and in too many places and there is a register for adding a fraction to the post-perspective-pre-truncate result to round instead. I didn't implement this back in the day, but thought about scaling the target dimensions in the GTE (watching for overflow) to get 1 bit of fraction back to refine the screen coordinate.