%12 = "tf.Cast"(%11) {Truncate = false, device = ""} : (tensor<2x4xi32>) -> tensor<2x4xf32>
    %13 = "tf.Mul"(%12, %cst_0) {device = ""} : (tensor<2x4xf32>, tensor<f32>) -> tensor<2x4xf32>
    %14 = "tf.Relu"(%13) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
    %15 = "tf.Minimum"(%14, %cst) {device = ""} : (tensor<2x4xf32>, tensor<f32>) -> tensor<2x4xf32>

  %1 = "tf.ReadVariableOp"(%arg1) {_output_shapes = ["tfshape$dim { size: 1 } dim { size: 3 }"], device = "/device:CPU:0", dtype = f32} : (tensor<!tf_type.resource<tensor<1x3xf32>>>) -> tensor<1x3xf32>

  // CHECK-NEXT: [[ready_ch:%.*]] = tfrt.new.chain

  // CHECK: return %arg0
  func.return %0 : tensor<2x4xf32>
}

// CHECK-LABEL: func @testBroadcastToNoOp
func.func @testBroadcastToNoOp(%arg0: tensor<2x4xf32>, %arg1: tensor<2xi32>) -> tensor<2x4xf32> {
  %0 = "tf.BroadcastTo"(%arg0, %arg1) : (tensor<2x4xf32>, tensor<2xi32>) -> tensor<2x4xf32>

  // CHECK: return %arg0
  func.return %0 : tensor<2x4xf32>
}

  %2 = "tf.Cast"(%1) {Truncate = false} : (tensor<1x2xbf16>) -> tensor<1x2xf32>
  %3 = "tf.IdentityN"(%2) {device = ""} : (tensor<1x2xf32>) -> tensor<1x2xf32>
  return %3 : tensor<1x2xf32>
}

// CHECK: func @cast_bf16_matmul_to_fp32
// CHECK-DAG: %[[cst:.*]] = "tf.Const"() <{value = dense<{{.*}}> : tensor<10x2xf32>}> : () -> tensor<10x2xf32>

  %3 = "tfl.pack"(%1, %2) {axis = 0 : i32, per_device_costs = {CPU = 2.0 : f32, GPU = -1.0 : f32}, values_count = 2 : i32, tac.device = "CPU"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<2x1xf32>
  func.return %3 : tensor<2x1xf32>
})";
  const std::string kExpectedFB = CreateRuntimeMetadata();
  mlir::DialectRegistry registry;
  registry.insert<mlir::TFL::TensorFlowLiteDialect, mlir::arith::ArithDialect,

// NOFALLBACK-LABEL: @xla_svd
func.func @xla_svd(%arg0: tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>) {
  // NOFALLBACK: XlaSvd
  %s, %u, %v = "tf.XlaSvd"(%arg0) {max_iter = 1, epsilon = 1.0E-09 : f32, precision_config = ""} : (tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>)
  func.return %s, %u, %v : tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>
}

  // CHECK: _arithmetic_count = 3000 : i64
  %0 = "tfl.fully_connected"(%arg0, %arg1, %arg2) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<1x37xf32>, tensor<40x37xf32>, tensor<40xf32>) -> tensor<1x40xf32>
  func.return %0 : tensor<1x40xf32>
}

  %1 = "tfl.neg"(%arg0) : (tensor<2x3xf32>) -> tensor<2x3xf32>
  %2 = "tfl.relu"(%1) : (tensor<2x3xf32>) -> tensor<2x3xf32>
  %3 = "tfl.mul"(%alpha, %2) {fused_activation_function = "NONE"} : (tensor<f32>, tensor<2x3xf32>) -> tensor<2x3xf32>
  %4 = "tfl.add"(%0, %3) {fused_activation_function = "NONE"} : (tensor<2x3xf32>, tensor<2x3xf32>) -> tensor<2x3xf32>
  func.return %4 : tensor<2x3xf32>

  %4 = "tf.MatMul"(%arg0, %3) {device = "", transpose_a = false, transpose_b = false} : (tensor<2x3xf32>, tensor<3x4xf32>) -> tensor<2x4xf32>
  %5 = "tf.Identity"(%4) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
  %6 = "tf.Identity"(%5) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
  func.return %6 : tensor<2x4xf32>

      %arg0 = "tf.Const"() {value = dense<-3.0> : tensor<5x7xf32>} : () -> tensor<5x7xf32>
      %arg1 = "tf.Const"() {value = dense<-3.0> : tensor<11x7xf32>} : () -> tensor<11x7xf32>

      %1 = "tf.MatMul"(%arg0, %arg1) {transpose_a = false, transpose_b = true} : (tensor<5x7xf32>, tensor<11x7xf32>) -> tensor<5x11xf32>

      func.return %1 : tensor<5x11xf32>
    }
  })";
  TF_ASSERT_OK_AND_ASSIGN(