[MLIR][NVGPU] Fix the cga_cluster.mlir test (llvm#112191)

This patch fixes the sm90 cluster test by: * Fixing a typo in LowerGpuOpsToNVVMOps where one of the ClusterDim Op conversion pattern should actually be for the ClusterDimBlocks Op. This addresses the compilation error for this test. * The grid-size should be (4,4,1) instead of (2,2,1). This passes the scf-if check against the threshold of 3 below and actually generates the required prints from the GPU. Signed-off-by: Durgadoss R <[email protected]>
demin-han · Oct 14, 2024 · a8b5115 · a8b5115
1 parent ddb64e6
commit a8b5115
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 3 deletions.
diff --git a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
@@ -373,8 +373,9 @@ void mlir::populateGpuToNVVMConversionPatterns(
       NVVM::BlockInClusterIdYOp, NVVM::BlockInClusterIdZOp>>(
       converter, IndexKind::Other, IntrType::Id);
   patterns.add<gpu::index_lowering::OpLowering<
-      gpu::ClusterDimOp, NVVM::ClusterDimXOp, NVVM::ClusterDimYOp,
-      NVVM::ClusterDimZOp>>(converter, IndexKind::Other, IntrType::Dim);
+      gpu::ClusterDimBlocksOp, NVVM::ClusterDimBlocksXOp,
+      NVVM::ClusterDimBlocksYOp, NVVM::ClusterDimBlocksZOp>>(
+      converter, IndexKind::Other, IntrType::Dim);
   patterns.add<gpu::index_lowering::OpLowering<
       gpu::BlockIdOp, NVVM::BlockIdXOp, NVVM::BlockIdYOp, NVVM::BlockIdZOp>>(
       converter, IndexKind::Grid, IntrType::Id);

diff --git a/mlir/test/Integration/GPU/CUDA/sm90/cga_cluster.mlir b/mlir/test/Integration/GPU/CUDA/sm90/cga_cluster.mlir
@@ -18,7 +18,7 @@ module attributes {gpu.container_module} {
     return
   }
   gpu.module @gpumodule {
-    gpu.func @kernel_cluster() kernel attributes {gpu.known_block_size = array<i32: 1, 1, 1>, gpu.known_grid_size = array<i32: 2, 2, 1>} {
+    gpu.func @kernel_cluster() kernel attributes {gpu.known_block_size = array<i32: 1, 1, 1>, gpu.known_grid_size = array<i32: 4, 4, 1>} {
       %cidX = gpu.cluster_id  x
       %cidY = gpu.cluster_id  y
       %cidZ = gpu.cluster_id  z