[lld] r257023 - [ELF] Add AMDGPU support

[Rafael Espíndola via llvm-commits]

> I agree with that. The current definition of st_value field for AMDGPU ELF
> files looks pretty odd and not in line with the ELF spec. I'd greatly
> appreciate if you do something before it's baked in to the AMDGPU ELF spec
> forever.

I strongly agree.

A thought that has been growing in my mind for some time: There has to
be a limit to how odd a system we support on trunk. If someone
proposes a backend that use PDP endian, 9 bit bytes and cray floats I
think we should just reject it.

The main thing that started me thinking was MIPS, were there seems to
be an endless series of differences to compensate for some other
difference. IMHO it and LLVM would be better served by having by
having a clean V2 ABI on trunk and all of odd st_info, relocation
packing, hi/lo mapping in a vendor branch.

But this is worse. To fully support this in llvm we would have to
propagate a isAMDGpu everywhere to know if we are looking at offsets
or addresses. I asked Tom on IRC to change this to use addresses and
keep a change or patching tool locally to support existing users while
the psabi is update to conform to the ELF spec.

Cheers,
Rafael