[Mlir-commits] [mlir] [mlir][gpu] Add Support for Cluster of Thread Blocks in `gpu.launch` (PR #76924)

Adam Paszke llvmlistbot at llvm.org
Thu Jan 4 02:33:56 PST 2024


================
@@ -660,6 +661,8 @@ void LaunchOp::build(OpBuilder &builder, OperationState &result,
   // Add grid and block sizes as op operands, followed by the data operands.
   result.addOperands({gridSizeX, gridSizeY, gridSizeZ, getBlockSizeX,
                       getBlockSizeY, getBlockSizeZ});
+  if (clusterSizeX && clusterSizeY && clusterSizeZ)
----------------
apaszke wrote:

If only one is passed in... it seems wrong to ignore them, I think? Perhaps it would be better to construct the op as given and let the verifier fail?

https://github.com/llvm/llvm-project/pull/76924


More information about the Mlir-commits mailing list