Re: Package building

[David Chisnall]

On 23 Nov 2019, at 14:48, Riccardo Mottola <riccardo.mottola@libero.it> wrote:
> 
> Hi,
> 
> 
> On 11/20/19 12:28 PM, David Chisnall wrote:
>> 
>> I'm happy to take contributions for the assembly paths for other 
>> architectures.  We currently have:
>> 
>> - x86-32
>> - x86-64
>> - AArch32
>> - AArch64
>> - MIPS (at least n64 works, I think o32 and n32 are there but untested).
>> 
>> I passionately hate PowerPC assembly, so will definitely not write that 
>> myself, but anyone who wants to add it is very welcome to do so.  I will 
>> probably get around to adding RISC-V support at some point.
> 
> 
> this list only lacks in my interest:
> 
> - MIPS LittleEndian (as soon as I get the small project with Nikolaus 
> actually do something useful) or is your implementation already 
> endian-independent?

I’ve only tested it on BE, I think the logic that pulls bytes out of the 
selector index will need tweaking for LE, but that should be about it.

SPARC and POWER are completely missing, but I don’t see any fundamental problem 
with someone adding them.  POWER may be fun because of the function descriptor 
model.  I’m not sure how well LLVM’s SPARC back end works.  ESA did some work 
on it for LEON, so SPARC32 may be stable.

There’s also System/Z, which is supported by LLVM, if someone wants to run 
Objective-C on a mainframe...

My somewhat biased view is that if there aren’t enough engaged users for a 
platform that we can get around 100 lines of assembly written for that 
platform, then it’s hard to justify the investment in that platform.

David