linux Erlang/OTP status

[Daniel Luna]

On Sat, 5 Nov 2005, Marthin Laubscher wrote:
>Daniel Luna wrote:
>> On Sat, 5 Nov 2005, Marthin Laubscher wrote:
>>> Brian's announcement did mention the exclusion of HiPE. I've not seen
>>> HiPE code, but I suspect that the trouble is that very tight
>>> optimisation is often opportunistically using specific platform and
>>> processor capabilities.
>>
>> If you are running Linux then HiPE should work out-of-the-box for x86 and
>> amd64 on most kernels. It also runs on sparc/solaris(V8+) and powerPC
>> (both Linux and MacOSX).
>
> Sorry, no x86/amd64, we were talking ARM/XSCALE/Intel PXA255

Aha! Then it's some work to port HiPE.

Assuming that OTP in general runs on the particular platform, what has to 
be done is:

Translate to arch-specific represention -- Steal the ppc stuff for RISC 
and the x86 stuff for CISC. Then make the changes necessary. Easy, but you 
need to know the instruction set for the architecture.

Register allocation -- Just plug it in. (sort of)

Frame allocation (non-register temps to stack slots, tailcall stack 
management...) -- Copy and change. Easy.

Assemble and encode -- This part is more work. You need to write a 
translator from the internal represenation to machine instructions. 
Potentially tricky (and somewhat boring).

Port the runtime system parts (loader and stuff) -- Should be easy if 
Linux is used. Most parts in the runtime system are generic and the 
specific parts should be easy to fix. Some things here could be tricky.

I think that should be it.

Feel free to contact me for further information.

/Luna
-- 
Daniel Luna                           | Top reasons that I have a beard:
luna@REDACTED                     |  a) Laziness.
http://www.update.uu.se/~luna/        |  b) I can.
Don't look at my homepage (it stinks).|  c) I can get away with it.