[PATCH] D105222: [AMDGPU][Docs] Update Code Object V3 example to includes args section

[Scott Linder via Phabricator via llvm-commits]

scott.linder added subscribers: rampitec, scott.linder.
scott.linder added a comment.

Thank you for improving the docs! This definitely just looks like an oversight on my point when implementing the new metadata.

I have a couple small requests, but otherwise LGTM



================
Comment at: llvm/docs/AMDGPUUsage.rst:12191-12194
+       .args:
+         - .size: 8
+           .offset: 0
+           .value_kind: by_value
----------------
The argument in this case it a pointer, which I believe we would describe with `.value_kind: global_buffer`.

We are also using flat store, so this likely should have `.address_space: generic`, although I think in the frontend we coerce pointer arguments to the global address space because we know the arguments are not in e.g. scratch; with that in mind I think `.address_space: global` may actually be correct? The TODO in the description of `.address_space` reads:


```
.. TODO
                                                        Is "global_buffer" only "global"
                                                        or "constant"? Is
                                                        "dynamic_shared_pointer" always
                                                        "local"? Can HCC allow "generic"?
                                                        How can "private" or "region"
                                                        ever happen?
```

Maybe one of @kzhuravl @t-tye @rampitec know what the `.address_space` should be in this case?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105222/new/

https://reviews.llvm.org/D105222