Notes and changes
Since the machine description is a shell script, and the Raspberry Pi on which I was running this was quite slow, it has a small modification to cache the output the machine description shell script. commit
Some syntax fixes in generated code
Some hacks to overcome register allocator limitations (>32 bit loads have restrictions the allocator doesn't understand) e.g. commit
Some bespoke modifications to have even wider stores.
There is also a 'stackless' branch, where the stack pointer is made available to the register allocator. This is the version used by my Prøst implementation. This modification does of course mean that you can't use the stack anymore, and need to store/load the stack pointer somewhere (as in this line).
If you have any further questions, feel free to ask me.