func.func @decompose_scalar_updates(%arg0: tensor<!tf_type.resource<tensor<4x3xi32>>>, %arg1: tensor<2xi32>, %arg2: tensor<i32>) {
  "tf_device.cluster"() ({
    // CHECK-NOT: ResourceScatterUpdate
    // CHECK-NOT: BroadcastTo
    // CHECK: TensorScatterUpdate
    "tf.ResourceScatterUpdate"(%arg0, %arg1, %arg2) : (tensor<!tf_type.resource<tensor<4x3xi32>>>, tensor<2xi32>, tensor<i32>) -> ()
    tf_device.return
  }) : () -> ()

}

// CHECK-LABEL: quantize_float_placeholder_only
func.func @quantize_float_placeholder_only(%arg0: tensor<f32>, %arg1: tensor<2x3xi32>, %arg2: tensor<2x3xf32>) -> (tensor<f32>, tensor<2x3xi32>, tensor<2x3xf32>) {
  func.return %arg0, %arg1, %arg2: tensor<f32>, tensor<2x3xi32>, tensor<2x3xf32>

// CHECK-NEXT: %[[q:.*]] = "tfl.quantize"(%arg0)
// CHECK-NEXT: %[[dq:.*]] = "tfl.dequantize"(%[[q]])

//                (tensor<4x2xf32>, tensor<4x2xf32>, tensor<4x2xf32>)
//
// will be converted into:
//
//   %0 = "mhlo.slice"(%input) {
//             limit_indices = dense<[4, 2]> : tensor<2xi64>,
//             start_indices = dense<0> : tensor<2xi64>,
//             strides = dense<1> : tensor<2xi64>} :
//        (tensor<4x6xf32>) -> tensor<4x2xf32>
//   %1 = "mhlo.slice"(%input) {

//
//   %indices0 = "tf.Const"() {value = dense<4> : tensor<i32>}
//   %indices1 = "tf.Const"() {value = dense<[[3, 2], [1, 0]]> :
//   tensor<2x2xi32>} %0 = "tf.DynamicStitch"(%indices0, %indices1, %arg0,
//   %arg1)
//     : (tensor<i32>, tensor<2x2xi32>, tensor<2xf32>, tensor<2x2x2xf32>)
//     -> tensor<5x2xf32>
//
// is lowered to
//
//   %shape = "tf.Const"() {value = dense<[-1, 2]> : tensor<2xi64>}

  %11 = "tf.Reshape"(%10, %0) {device = ""} : (tensor<4x1x1x2x1xi32>, tensor<2xi32>) -> tensor<4x2xi32>
  %12 = "tf.Transpose"(%11, %6) {device = ""} : (tensor<4x2xi32>, tensor<2xi32>) -> tensor<2x4xi32>
  %13 = "tf.Reshape"(%arg0, %1) {device = ""} : (tensor<1x1x2x1xf32>, tensor<1xi32>) -> tensor<2xf32>