[clang] [Clang][HIP][CUDA] Add `__cluster_dims__` and `__no_cluster__` attribute (PR #156686)

[Shilei Tian via cfe-commits]

================
@@ -13067,6 +13067,14 @@ def warn_cuda_maxclusterrank_sm_90 : Warning<
   "maxclusterrank requires sm_90 or higher, CUDA arch provided: %0, ignoring "
   "%1 attribute">, InGroup<IgnoredAttributes>;
 
+def err_cuda_cluster_attr_not_supported : Error<
+  "%0 is not supported for this GPU architecture"
+>;
+
+def err_cuda_cluster_dims_too_large : Error<
+  "only a maximum of %0 thread blocks in a cluster is supported"
----------------
shiltian wrote:

I'd not add "CUDA" here, since this is for both CUDA and HIP. It also doesn't look well if we write "CUDA/HIP". The best would be to select "CUDA" or "HIP" accordingly, but I don't see it is really necessary.

https://github.com/llvm/llvm-project/pull/156686