I am more fond of banking Data Memory than Code Memory. Since the originating post regarded Code Memory it's easy to understand Andy's reluctance.


When implementing loggers or waveform recorder functionality it's easy to require 256K to 4M of memory. The storing of quantized data into a banked data store is as trivial task.

Folks readily implement serial E2ROM and I2C periphs which require a "front end" driver to "regularize" it for general use within a program. Banking data is no more complicated.

Now banking code is a little more different and requires either extremely simple code execution threads or some form of memory managment (no matter how primitive).

I have so far alway's favored the former placing "self-contained" routines into the banked code space (which occupies a fixed location in the Code Mem Map - typically i bank the upper 32K). Each module in the banked Code Space has an identifier which i check before calling.

regards,
p