Richard Earnshaw wrote:
> 
> Leaving asside the issue of the correctness of write_register_bytes (note
> to self, must finish of my register patches), I don't think this is
> correct -- in fact, I think it's also wrong for little-endian as well.
> 
> What should happen is that the smaller-than-word value should be
> zero/sign-extended to 32 bits and then the whole thing stored in A1_REGNUM.

Ah, thanks.  OK, how about this?