[Mlir-commits] [mlir] [MLIR][XeGPU] Add LANE level integration test without XeGPU ops. (PR #181891)

[Sang Ik Lee]

https://github.com/silee2 created https://github.com/llvm/llvm-project/pull/181891

XeGPU LANE level integration test lacks a test without usage of any XeGPU dialect ops.
Add an integration test without XeGPU dialect ops.

>From 2e8e2968d5d7bcab1372336ed285a980ba3004a4 Mon Sep 17 00:00:00 2001
From: "Lee, Sang Ik" <sang.ik.lee at intel.com>
Date: Tue, 17 Feb 2026 19:56:43 +0000
Subject: [PATCH] [MLIR][XeGPU] Add LANE level integration test without XeGPU
 ops.

---
 .../Dialect/XeGPU/LANE/no-xegpu-ops.mlir      | 53 +++++++++++++++++++
 1 file changed, 53 insertions(+)
 create mode 100644 mlir/test/Integration/Dialect/XeGPU/LANE/no-xegpu-ops.mlir

diff --git a/mlir/test/Integration/Dialect/XeGPU/LANE/no-xegpu-ops.mlir b/mlir/test/Integration/Dialect/XeGPU/LANE/no-xegpu-ops.mlir
new file mode 100644
index 0000000000000..548e1864b1a05
--- /dev/null
+++ b/mlir/test/Integration/Dialect/XeGPU/LANE/no-xegpu-ops.mlir
@@ -0,0 +1,53 @@
+// RUN: mlir-opt %s --gpu-async-region --gpu-lower-to-xevm-pipeline="xegpu-op-level=lane" \
+// RUN: | mlir-runner \
+// RUN:   --shared-libs=%mlir_levelzero_runtime \
+// RUN:   --shared-libs=%mlir_runner_utils \
+// RUN:   --entry-point-result=void \
+// RUN: | FileCheck %s
+
+module @add attributes {gpu.container_module} {
+  memref.global "private" constant @__constant_2x2x2xf32_0 : memref<2x2x2xf32> = dense<[[[1.1, 2.2], [3.3, 4.4]], [[5.5, 6.6], [7.7, 8.8 ]]]>
+  memref.global "private" constant @__constant_2x2x2xf32 : memref<2x2x2xf32> = dense<[[[1.2, 2.3], [4.5, 5.8]], [[7.2, 8.3], [10.5, 11.8]]]>
+  func.func @main() {
+    %0 = memref.get_global @__constant_2x2x2xf32 : memref<2x2x2xf32>
+    %1 = memref.get_global @__constant_2x2x2xf32_0 : memref<2x2x2xf32>
+    %2 = call @test(%0, %1) : (memref<2x2x2xf32>, memref<2x2x2xf32>) -> memref<2x2x2xf32>
+    %cast = memref.cast %2 : memref<2x2x2xf32> to memref<*xf32>
+    call @printMemrefF32(%cast) : (memref<*xf32>) -> ()
+    return
+  }
+  func.func private @printMemrefF32(memref<*xf32>)
+  func.func @test(%arg0: memref<2x2x2xf32>, %arg1: memref<2x2x2xf32>) -> memref<2x2x2xf32> {
+    %c2 = arith.constant 2 : index
+    %c1 = arith.constant 1 : index
+    %mem = gpu.alloc () : memref<2x2x2xf32>
+    gpu.memcpy %mem, %arg1 : memref<2x2x2xf32>, memref<2x2x2xf32>
+    %memref_0 = gpu.alloc () : memref<2x2x2xf32>
+    gpu.memcpy %memref_0, %arg0 : memref<2x2x2xf32>, memref<2x2x2xf32>
+    %memref_2 = gpu.alloc () : memref<2x2x2xf32>
+    gpu.launch_func @test_kernel::@test_kernel blocks in (%c2, %c2, %c2) threads in (%c1, %c1, %c1)
+          args(%memref_0 : memref<2x2x2xf32>, %mem : memref<2x2x2xf32>, %memref_2 : memref<2x2x2xf32>)
+    %alloc = memref.alloc() : memref<2x2x2xf32>
+    gpu.memcpy %alloc, %memref_2 : memref<2x2x2xf32>, memref<2x2x2xf32>
+    gpu.dealloc %memref_2 : memref<2x2x2xf32>
+    gpu.dealloc %memref_0 : memref<2x2x2xf32>
+    gpu.dealloc %mem : memref<2x2x2xf32>
+    return %alloc : memref<2x2x2xf32>
+  }
+  gpu.module @test_kernel {
+    gpu.func @test_kernel(%arg0: memref<2x2x2xf32>, %arg1: memref<2x2x2xf32>, %arg2: memref<2x2x2xf32>) kernel {
+      %0 = gpu.block_id  x
+      %1 = gpu.block_id  y
+      %2 = gpu.block_id  z
+      %3 = memref.load %arg0[%0, %1, %2] : memref<2x2x2xf32>
+      %4 = memref.load %arg1[%0, %1, %2] : memref<2x2x2xf32>
+      %5 = arith.addf %3, %4 : f32
+      memref.store %5, %arg2[%0, %1, %2] : memref<2x2x2xf32>
+      gpu.return
+    }
+  }
+  // CHECK: [2.3, 4.5]
+  // CHECK: [7.8, 10.2]
+  // CHECK: [12.7, 14.9]
+  // CHECK: [18.2, 20.6]
+}