[llvm-branch-commits] [llvm] [LangRef][AMDGPU] Specify that syncscope can cause atomic operations to race (PR #189015)

[Fabian Ritter via llvm-branch-commits]

================
@@ -3934,7 +3934,8 @@ Given that definition, R\ :sub:`byte` is defined as follows:
 -  Otherwise, if R is atomic, and all the writes R\ :sub:`byte` may
    see are atomic, it chooses one of the values written. See the :ref:`Atomic
    Memory Ordering Constraints <ordering>` section for additional
-   constraints on how the choice is made.
+   constraints on how the choice is made. Targets may impose additional
+   requirements on R and the writes it may see based on their ``syncscope``.
----------------
ritter-x2a wrote:

Thanks for the discussion! My takeaway is that reviewers agree that this PR is good to land since it's a gradual improvement of the current state, and that we should look separately into properly defining the semantics of the default syncscope that allows relevant transformations and works for existing front- and backends (I agree, I'll put looking into that on my to-do list).
Since it's already approved, I'll merge this PR in a couple of days unless I hear otherwise.

https://github.com/llvm/llvm-project/pull/189015