[Mlir-commits] [mlir] [MLIR][XeGPU] Allow create mem desc from 2d memref (PR #167767)

[Jianhui Li]

================
@@ -150,25 +175,24 @@ gpu.module @test_kernel [#xevm.target<chip = "pvc">] {
  
   // e.g. for mem_desc<32x64xf16, @block=[16, 16]>
   // its memory layout tuple is ([2,4,16,16],[1024,256,16,1])
-  //CHECK-LABEL: load_store_matrix_5
-  gpu.func @load_store_matrix_5(%arg0: memref<4096xi8, 3>) -> vector<8xf16> {
-    //CHECK: %[[c0:.*]] = arith.constant 0 : index
-    //CHECK: %[[view:.*]] = memref.view %arg0[%[[c0]]][] : memref<4096xi8, 3> to memref<2048xf16, 3>
- 
-    %0 = xegpu.create_mem_desc %arg0 : memref<4096xi8, 3> -> !xegpu.mem_desc<32x64xf16, #xegpu.mem_layout<block = [16, 16]>>
- 
+  //CHECK-LABEL: load_store_matrix_blocked_subgroupblockio
+  gpu.func @load_store_matrix_blocked_subgroupblockio(%arg0: memref<4096xi8, 3>) -> vector<8xf16> {
+
+    //CHECK: %[[intptr:.*]] = memref.extract_aligned_pointer_as_index %arg0 : memref<4096xi8, 3> -> index
+    //CHECK: %[[basePtrI64:.*]] = arith.index_castui %[[intptr]] : index to i32
----------------
Jianhui-Li wrote:

fixed.

https://github.com/llvm/llvm-project/pull/167767