[llvm] [NVPTX] Add syncscope support for cmpxchg (PR #140812)

[Akshay Deodhar via llvm-commits]

================
@@ -38,9 +38,12 @@
     for sm, ptx in TESTS:
         with open("cmpxchg-sm{}.ll".format(str(sm)), "w") as fp:
             print(run_statement.substitute(sm=sm, ptx=ptx), file=fp)
-            for size, success, failure, addrspace in product(
-                SIZES, SUCCESS_ORDERINGS, FAILURE_ORDERINGS, ADDRSPACES
+            for size, success, failure, addrspace, llvm_scope in product(
+                SIZES, SUCCESS_ORDERINGS, FAILURE_ORDERINGS, ADDRSPACES, LLVM_SCOPES
             ):
+                # cluster ordering is supported from SM90 onwards
+                if sm != 90 and llvm_scope == "cluster":
----------------
akshayrdeodhar wrote:

I take my words back- we can do this cleanly without the extra parameter in the API. Updated, thanks!

https://github.com/llvm/llvm-project/pull/140812