tensor<1x1x5xf32>,
        tensor<2x5xf32>, tensor<2x5xf32>, tensor<2x5xf32>, tensor<2x5xf32>,
        tensor<2x4xf32>, tensor<2x4xf32>, tensor<2x4xf32>, tensor<2x4xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<4x2xf32>, tensor<4xf32>,
        tensor<1x4xf32>, tensor<1x2xf32>,
        none, none, none, none) -> tensor<*xf32>

    // CHECK-NEXT:  %[[RES:.*]] = "tf.SelectV2"(%[[PRED]], %[[SCALED_GRADIENTS]], %[[SELU_GRAD_VALUE]]) : (tensor<4x8xi1>, tensor<4x8xf32>, tensor<4x8xf32>) -> tensor<4x8xf32>
    // CHECK-NEXT:  return %[[RES]] : tensor<4x8xf32>
    %2 = "tf.SeluGrad"(%gradients, %features) : (tensor<4x8xf32>, tensor<4x8xf32>) -> tensor<4x8xf32>
    func.return %2 : tensor<4x8xf32>
}

  // CHECK: return %[[MATMUL_1]] : tensor<4x6xf32>
}

// -----

func.func @batchMatMulMatrixAdjXY(%arg0: tensor<5x4xf32>, %arg1: tensor<6x5xf32>) -> tensor<4x6xf32> {
  %0 = "tf.BatchMatMul"(%arg0, %arg1) {adj_x = true, adj_y = true} : (tensor<5x4xf32>, tensor<6x5xf32>) -> tensor<4x6xf32>
  func.return %0 : tensor<4x6xf32>

  %4 = "tf.MatMul"(%arg0, %3) {device = "", transpose_a = false, transpose_b = false} : (tensor<2x3xf32>, tensor<3x4xf32>) -> tensor<2x4xf32>
  %5 = "tf.Identity"(%4) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
  %6 = "tf.Identity"(%5) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
  func.return %6 : tensor<2x4xf32>

  // CHECK-LABEL: QuantDequantTranspose

  %8 = "tfl.concatenation"(%2, %0) {axis = -1 : i32, fused_activation_function = "NONE"} : (tensor<1x1xf32>, tensor<1x1xf32>) -> tensor<1x2xf32>
  %9 = "quantfork.stats"(%8) {layerStats = dense<[-0.488159984, 0.189515018]> : tensor<2xf32>} : (tensor<1x2xf32>) -> tensor<1x2xf32>
  %10 = "tfl.concatenation"(%9, %7) {axis = -1 : i32, fused_activation_function = "NONE"} : (tensor<1x2xf32>, tensor<1x2xf32>) -> tensor<1x4xf32>

  %outputs =  "tf.Const"() {device = "/job:localhost/replica:0/task:0/device:CPU:0", value = dense<0> : tensor<i32>} : () -> tensor<i32>...

    %11 = "tf.Einsum"(%8, %10) {device = "", equation = "ab,bc->ac"} : (tensor<2x3xi32>, tensor<3x4xi32>) -> tensor<2x4xi32>
    %12 = "tf.Cast"(%11) {Truncate = false, device = ""} : (tensor<2x4xi32>) -> tensor<2x4xf32>
    %13 = "tf.Mul"(%12, %cst_0) {device = ""} : (tensor<2x4xf32>, tensor<f32>) -> tensor<2x4xf32>
    %14 = "tf.Relu"(%13) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>

    %0 = "tf.MatMul"(%arg0, %arg1) {attr_map = "0:transpose_a,1:transpose_b", device = "", transpose_a = false, transpose_b = false} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
    return %0 : tensor<2x2xf32>
  }

For example, if we have the code

```mlir
  %0 = "tf.Const"() {value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
  %1 = "tf.Const"() {device = "", value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
  %2 = "tf.Const"() {device = "baz", value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
```

then running this pass with 'default-device=foobar', we get:

```mlir

//     -> tensor<5x2xf32>
//
// is lowered to
//
//   %shape = "tf.Const"() {value = dense<[-1, 2]> : tensor<2xi64>}
//   %inp0 = "tf.Reshape"(%arg0, %shape)
//     : (tensor<2xf32>, tensor<2xi64>) -> tensor<1x2xf32>
//   %inp1 = "tf.Reshape"(%arg1, %shape)
//     : (tensor<2x2x2xf32>, tensor<2xi64>) -> tensor<4x2xf32>
//   %items0 = "tf.Unpack"(%[[INP0]]) {axis = 0 : i64}