func.return %0#0 : tensor<2xi32>

// CHECK-LABEL: unpack3Tensors
// CHECK: "tfl.unpack"(%arg0) <{axis = 1 : i32, num = 3 : i32}> : (tensor<2x3xi32>) -> (tensor<2xi32>, tensor<2xi32>, tensor<2xi32>)
}

func.func @unpackNegAxis(%arg0: tensor<2x3xi32>) -> tensor<2xi32> {
  %0:3 = "tf.Unpack"(%arg0) {axis = -1 : i64} : (tensor<2x3xi32>) -> (tensor<2xi32>, tensor<2xi32>, tensor<2xi32>)

  %0 = "tf.BroadcastTo"(%input, %shape) : (tensor<3xi32>, tensor<2xi32>) -> tensor<*xi32>
  func.return %0: tensor<*xi32>

// CHECK-LABEL: broadcast_to_i32_low_dim_with_unknown_output
// CHECK:  %0 = "tf.BroadcastTo"(%arg0, %arg1) : (tensor<3xi32>, tensor<2xi32>) -> tensor<*xi32>
// CHECK:  return %0 : tensor<*xi32>
}

  %8 = "tfl.mul"(%2, %3) {fused_activation_function = "NONE"} : (tensor<4xi32>, tensor<4xi32>) -> tensor<4xi32>

  func.return %5, %6, %7, %8 : tensor<i32>, tensor<4xi32>, tensor<4xi32>, tensor<4xi32>
}

// CHECK-LABEL: @add_dense_splat_int
func.func @add_dense_splat_int() -> tensor<4xi32> {
  %0 = arith.constant dense<[-10, -1, 42, 100]> : tensor<4xi32>

func.func @reverse_func_32(%arg0: tensor<5xi32>) -> tensor<5xi32> {
  %axis = "tf.Const"() {value = dense<0> : tensor<1xi32>} : () -> (tensor<1xi32>)

  // CHECK: [[VAL:%.+]] = "mhlo.reverse"(%arg0) <{dimensions = dense<0> : tensor<1xi64>}>
  %reversed = "tf.ReverseV2"(%arg0, %axis) : (tensor<5xi32>, tensor<1xi32>) -> tensor<5xi32>

  // CHECK: return [[VAL]] : tensor<5xi32>
  func.return %reversed : tensor<5xi32>
}

  %2 = "tf.AddN"(%arg0, %1, %1) : (tensor<2xi32>, tensor<2xi32>, tensor<2xi32>) -> tensor<2xi32>
  %3 = "tf.AddN"(%1, %arg0, %1) : (tensor<2xi32>, tensor<2xi32> , tensor<2xi32>) -> tensor<2xi32>
  %4 = "tf.AddN"(%1, %1) : (tensor<2xi32>, tensor<2xi32>) -> tensor<2xi32>
  %5 = "tf.AddN"(%arg0, %1, %0) : (tensor<2xi32>, tensor<2xi32>, tensor<2xi32>) -> tensor<2xi32>
  func.return %2, %3, %4, %5: tensor<2xi32>, tensor<2xi32>, tensor<2xi32>, tensor<2xi32>
}

  func.return %0#0 : tensor<2xi32>
}

// -----

func.func @unpack(%arg0: tensor<*xi32>) -> (tensor<*xi32>, tensor<*xi32>) {
  %0:2 = "tfl.unpack"(%arg0) {axis = 1 : i32, num = 2 : i32} : (tensor<*xi32>) -> (tensor<*xi32>, tensor<*xi32>)
  func.return %0#0, %0#1 : tensor<*xi32>, tensor<*xi32>
}

// -----

      %17 = "tfl.reduce_prod"(%16, %cst_14) {keep_dims = true, tac.device = "CPU", tac.inference_type = "FLOAT"} : (tensor<2xi32>, tensor<1xi32>) -> tensor<1xi32>
      %18 = "tfl.reshape"(%arg1, %17) {tac.device = "DARWINN", tac.inference_type = "FLOAT"} : (tensor<?x?x!tf_type.string>, tensor<1xi32>) -> tensor<?x!tf_type.string>

//     (tensor<i32>, tensor<i32>, tensor<i32>) -> tensor<5xi32>
//   %shape = "tf.Const"() {value = dense<[5, 1]> : tensor<2xi32>}
//   %indices = "tf.Reshape"(%x, %shape) : (tensor<5xi32, tensor<2xi32) ->
//     tensor<5x1xi32>
//   "tf.TensorScatterUpdate"(%x, %indices, %updates) :
//     (tensor<5xi32>, tensor<5x1xi32>, tensor<5xi32>) -> tensor<5xi32>
//
class LowerInvertPermutationOp : public RewritePattern {
 public:

// CHECK-NEXT:      %6 = "tfl.pseudo_const"() <{value = dense<1> : tensor<i32>}> : () -> tensor<i32>
// CHECK-NEXT:      %7 = "tfl.unsorted_segment_prod"(%3, %4, %6) : (tensor<4xi32>, tensor<4xi32>, tensor<i32>) -> tensor<1xi32>
// CHECK-NEXT:      %8 = "tfl.unsorted_segment_prod"(%3, %5, %6) : (tensor<4xi32>, tensor<4xi32>, tensor<i32>) -> tensor<1xi32>

    %cst_11 = arith.constant dense<[0, 2, 3, 1]> : tensor<4xi32>
    %cst_12 = arith.constant dense<[0, 1, 3, 2]> : tensor<4xi32>
    %cst_18 = arith.constant dense<[0, 2, 1, 3]> : tensor<4xi32>
    %2112 = "tfl.transpose"(%arg0, %cst_11) : (tensor<1x4x1440x256xf32>, tensor<4xi32>) -> tensor<1x1440x256x4xf32>
    %2113 = "tfl.transpose"(%2112, %cst_18) : (tensor<1x1440x256x4xf32>, tensor<4xi32>) -> tensor<1x256x1440x4xf32>