[Mlir-commits] [mlir] [mlir][gpu] Support Cluster of Thread Blocks in `gpu.launch_func` (PR #72871)

[llvmlistbot at llvm.org]

================
@@ -1065,6 +1081,11 @@ KernelDim3 LaunchFuncOp::getBlockSizeOperandValues() {
   return KernelDim3{operands[3], operands[4], operands[5]};
 }
 
+KernelDim3 LaunchFuncOp::getClusterSizeOperandValues() {
+  auto operands = getOperands().drop_front(getAsyncDependencies().size());
+  return KernelDim3{operands[6], operands[7], operands[8]};
----------------
qcolombet wrote:

Could we use `getClusterSizeX`, ..., here instead of hardcoding the indices?

**edit**: I see this is a common pattern in this file so fell free to ignore.

https://github.com/llvm/llvm-project/pull/72871