shiltian wrote:

> Are there any links to ABI or other documentation that covers the format and 
> ideally design of AMD specific things like `.amdgpu.info`, `SHN_AMDGPU_LDS` 
> and `named-barrier` symbols? Would be good to post links to these in the PR 
> description.

Yes, will do.

> I have similar concerns to MaskRay about maintainability. If someone 
> complained that there was a bug, or some refactoring was needed would anyone 
> outside of AMD have a chance of resolving it?
>
> I think it would be useful to describe how this code interacts with the 
> general parts of LLD in terms of inputs and outputs, and what invariants need 
> to be followed. The large pass could then be treated by LLD developers as a 
> kind of black box. Would likely need an owner from AMD to look after it.

That's a very good point. I understand the concern that, even though this code 
lives in the linker, most of it isn't really about linking itself. It's mainly 
AMDGPU backend logic. Unfortunately, that's unavoidable given the nature of GPU 
object linking support. I also believe this is the first implementation of GPU 
object linking in the entire open source community. I'll make sure to explain 
this clearly in the PR description once I finish writing it. That said, I'd be 
happy to own this part of the code.

https://github.com/llvm/llvm-project/pull/206787
_______________________________________________
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

Reply via email to