Targs = [tensor<32xf32>, tensor<64xf32>, tensor<8xf32>, tensor<2xf32>],
      Tresults = [tensor<32xf32>, tensor<64xf32>, tensor<8xf32>]}
      : (tensor<32xf32>, tensor<64xf32>, tensor<8xf32>, tensor<2xf32>, tensor<2x!tf_type.string>)
        -> (tensor<32xf32>, tensor<64xf32>, tensor<8xf32>)
    tf_device.return %0#0, %0#1, %0#2 : tensor<32xf32>, tensor<64xf32>, tensor<8xf32>

      %read1 = "tf.ReadVariableOp"(%arg1) : (tensor<*x!tf_type.resource<tensor<32xf32>>>) -> tensor<32xf32>
      %var_handle = "tf.VarHandleOp"() {container = "c", shared_name = "v0"} : () -> tensor<*x!tf_type.resource<tensor<32xf32>>>
      %read2 = "tf.ReadVariableOp"(%var_handle) : (tensor<*x!tf_type.resource<tensor<32xf32>>>) -> tensor<32xf32>

      "tf.AssignVariableOp"(%arg0, %read2) : (!tf_res, tensor<32xf32>) -> ()
      %read3 = "tf.ReadVariableOp"(%arg0) : (!tf_res) -> tensor<32xf32>
      tf_executor.yield %read3 : tensor<32xf32>
    }
    tf_executor.fetch %island#0 : tensor<32xf32>
  }
  func.return %graph : tensor<32xf32>
}

// -----
// Tests aliasing of the two resource handles that refer to the same variable.

          [%u0#0, %u0#1] as %u : tensor<32xf32>)
          {n = 2 : i32, devices = {CORE_0 = ["/CPU:0", "/GPU:1"]}} {
        %read0 = "tf.ReadVariableOp"(%r0) : (tensor<*x!tf_type.resource<tensor<32xf32>>>) -> tensor<32xf32>
        // expected-remark@above {{ID: 1}}
        "tf.AssignVariableOp"(%r1, %u) : (tensor<*x!tf_type.resource<tensor<32xf32>>>, tensor<32xf32>) -> ()
        // expected-remark@above {{ID: 2}}

  %4 = "tfl.minimum"(%0, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %5 = "tfl.minimum"(%1, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %6 = "tfl.minimum"(%2, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %7 = "tfl.minimum"(%3, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>

        -> !tf_res
      %read = "tf.ReadVariableOp"(%id2) : (!tf_res) -> tensor<32xf32>
      %id3 = "tf.Identity"(%read) : (tensor<32xf32>) -> tensor<32xf32>
      tf_executor.yield
    }
    tf_executor.fetch %island : !tf_executor.control
  }
  func.return
}

// -----
!tf_res = tensor<*x!tf_type.resource<tensor<32xf32>>>

// Tesets that the pass can propagate through tf.If's branches.

      %read1 = "tf.ReadVariableOp"(%arg1) : (tensor<*x!tf_type.resource<tensor<32xf32>>>) -> tensor<32xf32>
      %var_handle = "tf.VarHandleOp"() {container = "c", shared_name = "v0"} : () -> tensor<*x!tf_type.resource<tensor<32xf32>>>
      %read2 = "tf.ReadVariableOp"(%var_handle) : (tensor<*x!tf_type.resource<tensor<32xf32>>>) -> tensor<32xf32>

    %assign1_control = tf_executor.island wraps "tf.AssignVariableOp"(%arg1, %read0) : (tensor<*x!tf_type.resource<tensor<32xf32>>>, tensor<32xf32>) -> ()
    %assign2_control = tf_executor.island wraps "tf.AssignVariableOp"(%arg0, %read2) : (tensor<*x!tf_type.resource<tensor<32xf32>>>, tensor<32xf32>) -> ()

  %1:2 = "tfl.split"(%cst, %0) {num_splits = 2 : i32} : (tensor<i32>, tensor<4xf32>) -> (tensor<2xf32>, tensor<2xf32>)
  %2 = "tfl.quantize"(%1#0) {qtype = tensor<2x!quant.uniform<u8:f32, 1.0>>} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.0>>
  %3 = "tfl.quantize"(%1#1) {qtype = tensor<2x!quant.uniform<u8:f32, 1.0>>} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.0>>

    func.return
}

func.func private @__reduce_func_2(%arg0: tensor<i64> {tf._user_specified_name = "args_0"}, %arg1: tensor<32xf32> {tf._user_specified_name = "args_1"}, %arg2: tensor<!tf_type.resource<tensor<64xf32>>>, %arg3: tensor<!tf_type.resource<tensor<128xf32>>>)