tf_device.return %1, %2 : tensor<*xi32>, tensor<*xi1>
    }
    func.return %0#0, %1#0 : tensor<*xi32>, tensor<*xi1>
  }
  func.func @tpu0_func(%arg0: tensor<8xi32>) -> (tensor<*xi32>, tensor<*xi1>) {
    %1, %2 = "tf.A"(%arg0) : (tensor<8xi32>) -> (tensor<*xi32>, tensor<*xi1>)

// Ensure constants roundtrip exactly

func.func @bool() -> tensor<4xi1> {
  // CHECK-LABEL: @bool
  // CHECK: value = dense<[false, true, true, false]> : tensor<4xi1>
  %0 = "tfl.pseudo_const"() { value = dense<[false, true, true, false]> : tensor<4xi1> } : () -> tensor<4xi1>
  func.return %0 : tensor<4xi1>
}

func.func @complex64() -> tensor<4xcomplex<f32>> {
  // CHECK-LABEL: @complex64

func.func @while_cond_10_frozen0(%arg0: tensor<*xi32>, %arg1: tensor<*xi32>, %arg2: tensor<*xi32>, %arg3: tensor<*xf32>, %arg4: tensor<*xf32>) -> tensor<*xi1> {
  %cst = arith.constant dense<10> : tensor<i32>
  %0 = "tf.Less"(%arg2, %cst) {T = i32, device = ""} : (tensor<*xi32>, tensor<i32>) -> tensor<*xi1>
  func.return %0 : tensor<*xi1>
}

// CHECK: tfl.while
// CHECK: ^bb0([[ARGS:.*]]):
// CHECK:   call @while_cond_10_frozen0
// CHECK:   yield

  %outputs_4 =  "tf.LogicalAnd"(%outputs_0, %outputs_2) {device = ""} : (tensor<*xi1>, tensor<*xi1>) -> tensor<*xi1>
  %outputs_6 =  "tf.Identity"(%outputs_4) {device = ""} : (tensor<*xi1>) -> tensor<*xi1>
  return %outputs_6 : tensor<*xi1>
}

// Original input argument list (loop_counter, max_iterations, element_index, tensor_list, read_only_tensor_list, scale)
// CHECK-LABEL: @map_while_body_180

  %0 = "tf.TensorListReserve"(%arg0, %arg1) : (tensor<i32>, tensor<i32>) -> tensor<!tf_type.variant<tensor<*xi1>>>
  // CHECK: %0 = "tfl.custom"(%arg0, %arg1) <{custom_code = "TensorListReserve", custom_option = #tfl<const_bytes : "0x06">}> : (tensor<i32>, tensor<i32>) -> tensor<!tf_type.variant<tensor<*xi1>>>
  func.return %0 : tensor<!tf_type.variant<tensor<*xi1>>>
}

// -----

// CHECK:       %14 = "tfl.custom"(%13#5, %13#2, %13#1) <{custom_code = "FlexSelect", custom_option = #tfl<const_bytes : "0x0653656C6563740031120653656C6563741A001A001A002A070A01541202300732180A052E31323131120F7768696C655F626F64795F3733303200023B341414042801">}> {tac.device = "CPU", tac.inference_type = "FLOAT"} : (tensor<?xi1>, tensor<?x!tf_type.string>, tensor<?x!tf_type.string>)...

  %15 = mhlo.compare  LT, %13, %14,  SIGNED : (tensor<4xi32>, tensor<4xi32>) -> tensor<4xi1>
  %16 = "mhlo.broadcast_in_dim"(%0) <{broadcast_dimensions = dense<> : tensor<0xi64>}> : (tensor<i32>) -> tensor<4xi32>
  %17 = mhlo.add %13, %16 : tensor<4xi32>
  %18 = mhlo.select %15, %17, %13 : tensor<4xi1>, tensor<4xi32>

// -----

// CHECK-LABEL: scatter_nd_i1
func.func @scatter_nd_i1(%arg0: tensor<?xi32>, %arg1: tensor<?xi1>, %arg2: tensor<?xi32>) -> tensor<?xi1> {
  // CHECK: "tfl.scatter_nd"(%arg0, %arg1, %arg2)
  %0 = "tfl.scatter_nd"(%arg0, %arg1, %arg2): (tensor<?xi32>, tensor<?xi1>, tensor<?xi32>) -> tensor<?xi1>
  func.return %0 : tensor<?xi1>
}

// -----

// CHECK-LABEL: testUnsortedSegmentSum

  // CHECK: %[[CONVERT_1:.*]] = mhlo.bitcast_convert %[[QUANTIZE]] : (tensor<2x!quant.uniform<i8:f32, 1.000000e+00:3>>) -> tensor<2xi8>
  // CHECK: %[[CONVERT_2:.*]] = mhlo.bitcast_convert %[[CONVERT_1]] : (tensor<2xi8>) -> tensor<2x!quant.uniform<i8:f32, 1.000000e+00:3>>
  // CHECK: %[[DEQUANTIZE:.*]] = mhlo.uniform_dequantize %[[CONVERT_2]] : (tensor<2x!quant.uniform<i8:f32, 1.000000e+00:3>>) -> tensor<2xf32>

  %5 = "tf.Range"(%4, %3, %2) : (tensor<i32>, tensor<i32>, tensor<i32>) -> tensor<4096xi32>
  %6 = "tf.Equal"(%1, %5) : (tensor<*xi32>, tensor<4096xi32>) -> tensor<*xi1>
  %7 = "tf.Cast"(%6) : (tensor<*xi1>) -> tensor<*xf32>
  %8 = "tf.BatchMatMulV2"(%7, %arg0) {adj_x = false, adj_y = false} : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
  func.return %8 : tensor<*xf32>
}