[Mlir-commits] [mlir] [MLIR][GPU] Add gpu.cluster_dim_blocks and gpu.cluster_block_id Ops (PR #95245)

[Guray Ozen]

================
@@ -95,6 +100,12 @@ void ClusterIdOp::inferResultRanges(ArrayRef<ConstantIntRanges>,
   setResultRange(getResult(), getIndexRange(0, max - 1ULL));
 }
 
+void ClusterBlockIdOp::inferResultRanges(ArrayRef<ConstantIntRanges>,
+                                         SetIntRangeFn setResultRange) {
+  uint64_t max = kMaxClusterDim;
+  setResultRange(getResult(), getIndexRange(0, max - 1ULL));
----------------
grypp wrote:

IMHO, Let's not add this. What do you think?

Because non_portable_cluster_size attribute could be set on the kernel launch, but this verifier won't let you use more than 8 cluster size. 

https://github.com/llvm/llvm-project/pull/95245