[Mlir-commits] [mlir] [MLIR][GPU] Add gpu.cluster_dim_blocks Op to represent number of blocks per cluster (PR #95245)

[Pradeep Kumar]

================
@@ -86,6 +86,11 @@ static std::optional<uint64_t> getKnownLaunchDim(Op op, LaunchDims type) {
 
 void ClusterDimOp::inferResultRanges(ArrayRef<ConstantIntRanges>,
                                      SetIntRangeFn setResultRange) {
+  setResultRange(getResult(), getIndexRange(1, kMaxDim));
----------------
schwarzschild-radius wrote:

The PTX ISA or the CUDA Programming Guide does not mention of any upper bound other than the maximum portable cluster size which can be overriden using `CU_FUNC_ATTRIBUTE_NON_PORTABLE_CLUSTER_SIZE_ALLOWED`

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