%const = "tf.Const"() {value = dense<1> : tensor<i1>} : () -> tensor<*xi1>
      %print = "tf.Print"(%const) { message = "bla" } : (tensor<*xi1>) -> (tensor<*xi1>)
      tf_executor.yield %const : tensor<*xi1>
    }
    %loop_cond:2 = tf_executor.LoopCond %island#0 : tensor<*xi1>
    tf_executor.fetch %loop_cond#0 : tensor<*xi1>
  }
  func.return
}

// 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>)...

# CHECK-DAG: tf.InfeedDequeue{{.+}}-> tensor<?x8x?xi8>
# CHECK-DAG: tf.InfeedDequeue{{.+}}-> tensor<8x64xui8>

    %5:2 = tf_executor.island(%4#2) wraps "tf.opC"() : () -> tensor<i32>
    %6:2 = tf_executor.island wraps "tf.opD"(%4#0, %5#0) : (tensor<*xi32>, tensor<i32>) -> tensor<*xi1>
    %7:2 = tf_executor.LoopCond %6#0 : (tensor<*xi1>) -> (tensor<i1>, !tf_executor.control)
    %8:3 = tf_executor.Switch %4#0, %7#0 : tensor<*xi32>
    %9:2 = tf_executor.Exit %8#0 : tensor<*xi32>

  %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>

  %0 = "tf.Const"() { value = #tf_type<tensor_proto : "0x746674656e736f722464747970653a2044545f51494e54382074656e736f725f7368617065207b2064696d207b2073697a653a2032207d207d2074656e736f725f636f6e74656e743a20225c3137375c30323222"> : tensor<2x!tf_type.qint8> } : () -> tensor<2x!tf_type.qint8>
  // CHECK: return %[[result]] : tensor<2xi8>
  func.return %0 :  tensor<2x!tf_type.qint8>
}

  func.return %2 : tensor<1x2x2x1xf32>
// CHECK: tf__max_pool
}

// CHECK-LABEL: no_tf_canonicalization
func.func @no_tf_canonicalization(%arg0: tensor<8xi1>, %arg1: tensor<8x3xf32>, %arg2: tensor<8x3xf32>) -> tensor<8x3xf32> {
  %0 = "tf.Select"(%arg0, %arg1, %arg2) : (tensor<8xi1>, tensor<8x3xf32>, tensor<8x3xf32>) -> tensor<8x3xf32>
  func.return %0: tensor<8x3xf32>

// CHECK:   "tf.Select"
}

// CHECK-LABEL: denied_attribute

    %cst = arith.constant dense<-2.0>: tensor<0xf32>
    %cst_1 = arith.constant dense<0> : tensor<i32>
    %greater = "tf.Greater"(%val, %cst) : (tensor<0xf32>, tensor<0xf32>) -> tensor<0xi1>
    %res = "tf.Any"(%greater, %cst_1) : (tensor<0xi1>, tensor<i32>) -> (tensor<i1>)
    func.return %res : tensor<i1>
  }

  // CHECK-LABEL: @body(%arg0: tensor<0xf32>) -> tensor<0xf32>

    // CHECK: tf.Equal
    // CHECK-SAME: (tensor<4xf32>, tensor<4xf32>) -> tensor<*xi1>
    // TODO(ycao): Investigate why result type of tf.Equal is not inferred.
    %2 = "tf.Equal"(%0, %arg0) : (tensor<4xf32>, tensor<*xf32>) -> tensor<*xi1>
    %3 = "tf.Any"(%2, %1) : (tensor<*xi1>, tensor<i32>) -> (tensor<i1>)
    func.return %3 : tensor<i1>
  }

  // CHECK-LABEL: func @while_body_func

// -----

// 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