func.func @QuantizedCatsAddRequantsTest(%arg0: tensor<1x1xf32>, %arg1: tensor<1x1xf32>, %arg2: tensor<1x1xf32>, %arg3: tensor<1x1xf32>) -> (tensor<1x4xf32>, tensor<1x3xf32>) {
  %0 = "quantfork.stats"(%arg0) {layerStats = dense<[-0.440728068, 0.189515018]> : tensor<2xf32>} : (tensor<1x1xf32>) -> tensor<1x1xf32>
  %1 = "quantfork.stats"(%arg1) {layerStats = dense<[-0.154693216, 0.26483655]> : tensor<2xf32>} : (tensor<1x1xf32>) -> tensor<1x1xf32>

    %2 = "quantfork.stats"(%1) {layerStats = dense<[5.00000000e-6, 7.00000000e-1]> : tensor<2xf32>} : (tensor<1x3xf32>) -> tensor<1x3xf32>
    return %2 : tensor<1x3xf32>
  }
// CHECK: func.func private @quantize_dot_general_with_bias_same_shape_fn(%[[ARG_0:.+]]: tensor<1x2xf32>) -> tensor<1x3xf32> attributes {tf._original_func_name = "main_0"}

  // CHECK-DAG: %[[EXP:.*]] = "tf.Exp"(%[[SHIFTED]]) : (tensor<2x3xf32>) -> tensor<2x3xf32>
  // CHECK-DAG: %[[SUM:.*]] = "tf.Sum"(%[[EXP]], %[[AXIS]]) <{keep_dims = true}> : (tensor<2x3xf32>, tensor<1xi64>) -> tensor<2x1xf32>
  // CHECK-DAG: %[[LOG:.*]] = "tf.Log"(%[[SUM]]) : (tensor<2x1xf32>) -> tensor<2x1xf32>

}
// CHECK: func private @callee_tensorarray_decomposed(%[[CARG0:.*]]: tensor<!tf_type.resource<tensor<5x3xf32>>>, %[[CARG1:.*]]: tensor<!tf_type.resource<tensor<5x3xf32>>>, %[[CARG2:.*]]: tensor<!tf_type.resource<tensor<5x3xf32>>>)
// CHECK: %[[READ1:.*]] = "tf.ReadVariableOp"(%[[CARG1]]) : (tensor<!tf_type.resource<tensor<5x3xf32>>>) -> tensor<5x3xf32>
// CHECK: %[[UPDATE1:.*]] = "tf.XlaDynamicUpdateSlice"(%[[READ1]],

  // 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>

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

//     -> 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}

  %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:           %[[VAL_4:.*]]:2 = call @func_0_GPU_FLOAT(%[[VAL_0]], %[[VAL_1]], %[[VAL_2]], %[[VAL_3]]) {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_0"} : (tensor<1xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xf32>) -> (tensor<1xf32>, tensor<1xf32>)

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