[Mlir-commits] [mlir] [MLIR][MemRef] Fix AllocOp/AllocaOp flattening domination violation (PR #188980)

[Hocky Yudhiono]

================
@@ -271,6 +271,30 @@ func.func @alloca() -> memref<4x8xf32> {
 
 // -----
 
+func.func @alloc_dynamic(%n: index) -> memref<?x4xf32> {
+  %0 = memref.alloc(%n) : memref<?x4xf32>
+  return %0 : memref<?x4xf32>
+}
+
+// CHECK-LABEL: func @alloc_dynamic
+// CHECK-SAME: (%[[N:.*]]: index)
+// CHECK: %[[ALLOC:.*]] = memref.alloc(%{{.*}}) : memref<?xf32, strided<[1]>>
+// CHECK: memref.reinterpret_cast %[[ALLOC]] to offset: [0], sizes: [%[[N]], 4], strides: [4, 1]
+
+// -----
+
+func.func @alloca_dynamic(%n: index) -> memref<?x4xf32> {
+  %0 = memref.alloca(%n) : memref<?x4xf32>
+  return %0 : memref<?x4xf32>
+}
+
+// CHECK-LABEL: func @alloca_dynamic
+// CHECK-SAME: (%[[N:.*]]: index)
+// CHECK: %[[ALLOCA:.*]] = memref.alloca(%{{.*}}) : memref<?xf32, strided<[1]>>
+// CHECK: memref.reinterpret_cast %[[ALLOCA]] to offset: [0], sizes: [%[[N]], 4], strides: [4, 1]
+
+// -----
+
----------------
hockyy wrote:

Add some alloc test case with strided layout?
```suggestion
func.func @alloc_strided_row_major() -> memref<4x8xf32, strided<[8, 1]>> {
  %0 = memref.alloc() : memref<4x8xf32, strided<[8, 1]>>
  return %0 : memref<4x8xf32, strided<[8, 1]>>
}
// -----
```

https://github.com/llvm/llvm-project/pull/188980