}

func.func @while_body(%arg0: tensor<*xi32>) -> (tensor<*xi32>) {
  %0 = "tf.Rank"(%arg0) : (tensor<*xi32>) -> tensor<i32>
  %1 = "tf.Add"(%0, %arg0): (tensor<i32>, tensor<*xi32>) -> tensor<*xi32>
  func.return %1: tensor<*xi32>
}

func.func @while_cond(%arg0: tensor<*xi32>) -> tensor<*xi1> {
  %cst = arith.constant dense<10> : tensor<i32>
  %0 = "tf.Less"(%arg0, %cst) {T = i32, device = ""} : (tensor<*xi32>, tensor<i32>) -> tensor<*xi1>

func.func @dead_island(%arg0 : i32) -> i32 {
// CHECK: tf_executor.island
// CHECK-NOT: tf_executor.island
  %0 = tf_executor.graph {
    %1:2 = tf_executor.island {
      %a = "tf.opA"(%arg0) : (i32) -> i32
      %b = "tf.opB"(%a) : (i32) -> i32
      tf_executor.yield %b : i32
    }
    %2:2 = tf_executor.island {
      %a = "tf.opA"(%1#0) : (i32) -> i32
      tf_executor.yield %a : i32
    }
    tf_executor.fetch %1#0 : i32

      %0 = "arith.constant" () {value = dense<0> : tensor<i32>} : () -> tensor<i32> loc("Const")
      %1 = "tf.NotEqual"(%condArg0, %0) : (tensor<*xi32>, tensor<i32>) -> tensor<i1>
      "tf.Yield"(%1) : (tensor<i1>) -> ()
    },
    // body
    {
    ^bb0(%bodyArg0: tensor<*xi32>, %bodyArg1: tensor<*xf32>):
      %0 = "arith.constant" () {value = dense<1> : tensor<i32>} : () -> tensor<i32> loc("Const")

  %outputs_0 = "tf.Const"() {value = dense<0> : tensor<i32>} : () -> tensor<i32>
  %outputs_2 = "tf.Const"() {value = dense<1> : tensor<i32>} : () -> tensor<i32>
  %outputs_4 = "tf.Identity"(%arg0) {device = ""} : (tensor<*xi32>) -> tensor<*xi32>
  %outputs_6 = "tf.AddV2"(%outputs_4, %outputs_2) {device = ""} : (tensor<*xi32>, tensor<i32>) -> tensor<*xi32>
  %outputs_8 = "tf.Identity"(%arg1) {device = ""} : (tensor<*xi32>) -> tensor<*xi32>

      %replicated_output:2 = "tf.TPUReplicatedOutput(%identity) : (tensor<i32>) -> (tensor<i32>, tensor<i32>)
      return %replicated_output#0, %replicated_output#1 : tensor<i32>, tensor<i32>
    }
    ```

    will be transformed into:

    ```mlir
    func @tpu_computation(%arg0: tensor<i32>, %arg1: tensor<i32>) -> (tensor<i32>, tensor<i32>) {

      %0 = "arith.constant" () {value = dense<0> : tensor<i32>} : () -> tensor<i32> loc("Const")
      %1 = "tfl.greater"(%condArg0, %0) : (tensor<*xi32>, tensor<i32>) -> tensor<i1>
      "tfl.yield"(%1) : (tensor<i1>) -> ()
    },
    // body
    {
    ^bb0(%bodyArg0: tensor<*xi32>, %bodyArg1: tensor<*xf32>):
      %0 = "arith.constant" () {value = dense<1> : tensor<i32>} : () -> tensor<i32> loc("Const")

    %6:2 = tf_executor.LoopCond %5#0 : (tensor<*xi1>) -> (tensor<*xi1>, !tf_executor.control) {device = ""} loc("while/LoopCond")
    %7:3 = tf_executor.Switch %3#0, %6#0 : (tensor<*xi32>, tensor<*xi1>) -> (tensor<*xi32>, tensor<*xi32>, !tf_executor.control) {device = "", T = "tfdtype$DT_INT32", _class = ["loc:@while/Merge"]} loc("while/Switch")

  %3:2 = "tf.TPUReplicatedOutput"(%2#0) : (tensor<?xi32>) -> (tensor<?xi32>, tensor<?xi32>)
  %5:2 = "tf.TPUPartitionedOutputV2"(%3#0) {_XlaSharding = "", partition_dims = []} : (tensor<?xi32>) -> (tensor<?xi32>, tensor<?xi32>)
  %7:2 = "tf.TPUPartitionedOutputV2"(%3#1) {_XlaSharding = "", partition_dims = []} : (tensor<?xi32>) -> (tensor<?xi32>, tensor<?xi32>)

		case ti32*100 + ti8:
			i32 = int32(v); i8 = int8(i32); w = big(i8)
		case ti32*100 + tu8:
			i32 = int32(v); u8 = uint8(i32); w = big(u8)
		case ti32*100 + ti16:
			i32 = int32(v); i16 = int16(i32); w = big(i16)
		case ti32*100 + tu16:
			i32 = int32(v); u16 = uint16(i32); w = big(u16)
		case ti32*100 + ti32:
			i32 = int32(v); i32 = int32(i32); w = big(i32)
		case ti32*100 + tu32:

    }) {tac.device = "CPU", tac.inference_type = "FLOAT"} : (tensor<i32>, tensor<i32>, tensor<?xi32>, tensor<?xi32>, tensor<!tf_type.variant<tensor<?x!tf_type.string>>>, tensor<!tf_type.variant<tensor<?x!tf_type.string>>>, tensor<!tf_type.variant<tensor<?xi32>>>, tensor<!tf_type.variant<tensor<?xi32>>>) -> (tensor<i32>, tensor<i32>, tensor<?xi32>, tensor<?xi32>, tensor<!tf_type.variant<tensor<?x!tf_type.string>>>,...