[ros-kernel] Different Hardware Architectures
Mike Nordell
tamlin at algonet.se
Fri Apr 23 20:37:10 CEST 2004
Steven Edwards wrote:
> The big issue is just getting a Mingw and Binutils that can spit out PE
> formats.
Is not MinGW binutil's as already emitting COFF, and ld parsing that and
emitting PE?
If GCC and binutils are reasonably modular, I don't see how it could be a
large problem putting the PPC code generator into current x86 MinGW GCC,
combine the Win32 COFF generator of as with a PPC parser (to parse PPC
assembler, but emit COFF instead of ELF object files). ld should require
very little change, as AFAIK PE is PE is... no matter what the target CPU.
Anyone know if it's possible to cross-compile from any host CPU architecture
to x86 Win32, or is the cross-compilation in that respect limited to x86
hosts? Is it's not limited, than it seems almost all work has already been
done.
> Would it be possible to make ReactOS understand ELF as well
> for platforms that dont have compiler support for a PE target?
Theoretically; mnyes (not a kbd slip). Practically, I don't see this ever
happening. I believe the effort to make that happen could be one to three
orders of magnitudes larger than make the tools emit and parse COFF and PE.
Some things that might be worth considering:
- Do the Power CPU's support a 32-bit mode too? I'm not suggesting this, but
perhaps it could be a first step to make the port 32-bit?
- Do the Power CPU's support endianness selection (on startup), or am I
confusing them with MIPS?
- What Power CPU's are we really talking about here? The older 32-bit ones
or the current 64-bit version(s) (970/G5's I believe)?
To round off: Making the GCC tools emit COFF and PE instead of ELF I don't
see as a problem. If it is, then those tools are badly designed and would
anyway benefit from an overhaul. :-) I still think the target platfrom,
hardware, is the hard part.
/Mike
More information about the Ros-kernel
mailing list