"tf.opC"(%4) {_xla_compile_device_type = "TPU", _replication_info = "replicate", device = "/device:TPU:0", name = "name", is_stateless = true} : (tensor<i1>) -> ()
  %5 = "tf.TPUReplicatedInput"(%arg5) {index = 4 : i64, is_packed = true} : (tensor<i1>) -> tensor<i1>

//    "tf_device.parallel_execute"() ({
//      "tf.OpC"
//    }, {
//      "tf.OpD"
//    })
//  }
//
// The above IR will be flattened after `ReplicateToIslandPass` and
// `ParallelExecuteToIslandsPass` as follows:
//
// tf_executor.island wraps "tf.OpA"
// tf_executor.island {_parallel_execution_ids=r0:0} wraps "tf.OpB"
// tf_executor.island {_parallel_execution_ids=r0:0,p0:0} wraps "tf.OpC"

}

// -----

// CHECK-LABEL: func @replicate_with_multiple_operands
func.func @replicate_with_multiple_operands() {
  %0 = "tf.opA"() : () -> tensor<*xi1>
  %1 = "tf.opB"() : () -> tensor<*xi1>
  %2 = "tf.opC"() : () -> tensor<*xi1>
  %3 = "tf.opD"() : () -> tensor<*xi32>
  %4 = "tf.opE"() : () -> tensor<*xi32>
  %5 = "tf.opF"() : () -> tensor<*xi32>
  %6 = "tf.opG"() : () -> tensor<*xf32>
  %7 = "tf.opH"() : () -> tensor<*xf32>

    ```mlir
      %island:5 = tf_executor.island {
        %a = "tf.opA"() : () -> tensor<i1>
        %launch:2 = "tf_device.launch"() ( {
          %b = "tf.opB"() : () -> tensor<i32>
          %c = "tf.opC"() : () -> tensor<f32>
          tf_device.return %c, %b : tensor<f32>, tensor<i32>
        }) {device = "CPU:0"} : () -> (tensor<f32>, tensor<i32>)
        %d = "tf.opD"() : () -> tensor<i1>

are replicated each from the regions terminator.

For example:
```
%0 = "tf.opA"() : () -> tensor<i32>
%1 = "tf.opB"() : () -> tensor<i32>
%2 = "tf.opC"() : () -> tensor<f32>
%3 = "tf.opD"() : () -> tensor<f32>
%4 = "tf.opE"() : () -> tensor<!tf_type.resource>
%5 = "tf.opF"() : () -> tensor<!tf_type.resource>
%6 = "tf.opG"() : () -> tensor<!tf_type.string>

    %2 = "tf.Identity"(%1) : (tensor<!tf_type.variant<tensor<f32>>>) -> (tensor<!tf_type.variant<tensor<f32>>>)
    "tf.opC"(%2) {_xla_outside_compilation = "0"} : (tensor<!tf_type.variant<tensor<f32>>>) -> ()
    tf_device.return
  }) {cluster_attr = "cluster_attr"} : () -> ()
  func.return
}

// CHECK-LABEL: func @single_variant_output

            %1 = "tf.OpA"() {is_stateless = true}
            "tf.Yield"(%1)
          }, {
          ^bb0(%arg2, %arg3):
            %1 = "tf.OpB"(%arg2) {is_stateless = true}
            %2 = "tf.OpC"(%arg3) {is_stateless = true}
            "tf.Yield"(%1, %2)
          }) {is_stateless = true}
          return %0#1
        }
      ```
      would be transformed to
      ```mlir

}

func SYSHINT(x uint32) uint32 {
	return SYSOP(0, 0, 3, 2, 0, x, 0x1F)
}

func LDSTR(sz uint32, v uint32, opc uint32) uint32 {
	return sz<<30 | 7<<27 | v<<26 | opc<<22
}

func LD2STR(o uint32) uint32 {
	return o &^ (3 << 22)
}

func LDSTX(sz uint32, o2 uint32, l uint32, o1 uint32, o0 uint32) uint32 {