[Mlir-commits] [mlir] [mlir][vector] Refactor vector-transfer-flatten.mlir (nfc) (3/n) (PR #95745)

[Hugo Trachino]

================
@@ -131,10 +131,42 @@ func.func @transfer_read_dims_mismatch_non_contiguous_non_zero_indices(
 
 // -----
 
+/// The leading dynamic shapes don't affect whether this example is flattenable
+/// or not as those dynamic shapes are not candidates for flattening anyway.
+
+func.func @transfer_read_leading_dynamic_dims(
+    %arg : memref<?x?x8x4xi8, strided<[?, 32, 4, 1], offset: ?>>,
+    %idx_1 : index,
+    %idx_2 : index) -> vector<8x4xi8> {
+
+  %c0_i8 = arith.constant 0 : i8
+  %c0 = arith.constant 0 : index
+  %result = vector.transfer_read %arg[%idx_1, %idx_2, %c0, %c0], %c0_i8 {in_bounds = [true, true]} : memref<?x?x8x4xi8, strided<[?, 32, 4, 1], offset: ?>>, vector<8x4xi8>
+  return %result : vector<8x4xi8>
+}
+
+// CHECK-LABEL: func @transfer_read_leading_dynamic_dims
+// CHECK-SAME:    %[[ARG0:.+]]: memref<?x?x8x4xi8, {{.+}}>, %[[ARG1:.+]]: index, %[[ARG2:.+]]: index
----------------
nujaa wrote:

NIT: In other tests (e.g. transfer_read_dims_mismatch_non_zero_indices) arg definition are aligned with body,
``` suggestion
// CHECK-SAME:  %[[ARG0:.+]]: memref<?x?x8x4xi8, {{.+}}>, %[[ARG1:.+]]: index, %[[ARG2:.+]]: index
```

https://github.com/llvm/llvm-project/pull/95745