After much pruning, it looks like the baud rate division calculation can be saved. I did remove support for very low baud rates (under 180 baud) which shaved off a few instructions. But I'll admit I'm rather proud of that feature, and am eager to include it in the final submission.

I also looked into removing ping-pong buffering which will reduce the code overhead slightly, however, supporting data from Serial to USB Host will require more code without ping-pong unless I allow for some occasional byte loss when the USB buffer is busy sending to the host. However, the MCU has a mode where all endpoints have ping-pong buffering enabled EXCEPT endpoint 0. This should reduce the code overhead for the more common control-transfer case, but still streamline serial data transfer. I'll explore this in the days to come, a quick look suggests that ping pong on EP0 is responsible for 38 bytes.

Also, the division function for baud rates is not original art. It started from a copy of a 24-bit division routine found online by Tony Nixon. In a later post, I'll also document the process I took in refactoring it.