The in memory version is maintained as 16 bits to allow
some overflow in calculations, such as horizontal counts.

The original 6845 core was extended in the design of the
vga to allow the generation of wider displays. The original
width register was too small, so bits in other registers were
borrowed to extend the basic register.

By using a 16 bit in-memory register, the standard
calculation could be done, and then the overflow bits
copied to the right place.

This explains what you are seeing.

If you have not done this before, I strongly recommend
getting a working system you can play with. This is not
known as vga Hell for nothing. You should also read a
good book on the history of the vga to understand the
full horror.

Which chip is it that is not supported? I would have thought
all the Cirrus CL-GD chips were supported.