Harvard architecture with 24 bit data bytes and 32 bit code bytes

[Ian Lance Taylor]

Ned <Thomas.Gill@csr.com> writes:

> I'm beginning to hack a binutils port for our in-house DSP core. It's
> a  Harvard architecture, with different sized bytes in instruction and
> data  memory. Instruction memory has a smallest addressable unit of 32
> bits,  while data memory has a smallest addressable unit of 24 bits.
> 
> Does anyone on the list know whether this is going to cause me a lot
> of  pain? Has anyone had to deal with this problem before?
> 
> I'm guessing my first problem will be that there's only one
> OCTETS_PER_BYTE. I'll presumably have to look at the places where
> that's  used and add code to guess whether it should be using 3 or
> 4. Is there  likely to be anything else that trips me up? Any corner
> cases I should  watch out for?

I find it much simpler to avoid the OCTETS_PER_BYTE code and just use
byte addressing for everything.  You compensate in the relocation
processing, which is the only place which really cares.

Ian