MATRIX_DIAG = 113,
  QUANTIZE = 114,
  MATRIX_SET_DIAG = 115,
  ROUND = 116,
  HARD_SWISH = 117,
  IF = 118,
  WHILE = 119,
  NON_MAX_SUPPRESSION_V4 = 120,
  NON_MAX_SUPPRESSION_V5 = 121,
  SCATTER_ND = 122,
  SELECT_V2 = 123,
  DENSIFY = 124,
  SEGMENT_SUM = 125,
  BATCH_MATMUL = 126
}

// Options for the builtin operators.
union BuiltinOptions {
  Conv2DOptions,
  DepthwiseConv2DOptions,

  MATRIX_DIAG = 113,
  QUANTIZE = 114,
  MATRIX_SET_DIAG = 115,
  ROUND = 116,
  HARD_SWISH = 117,
  IF = 118,
  WHILE = 119,
  NON_MAX_SUPPRESSION_V4 = 120,
  NON_MAX_SUPPRESSION_V5 = 121,
  SCATTER_ND = 122,
  SELECT_V2 = 123,
  DENSIFY = 124,
  SEGMENT_SUM = 125,
  BATCH_MATMUL = 126,
  PLACEHOLDER_FOR_GREATER_OP_CODES = 127,
  CUMSUM = 128,
  CALL_ONCE = 129,
  BROADCAST_TO = 130,

  MATRIX_DIAG = 113,
  QUANTIZE = 114,
  MATRIX_SET_DIAG = 115,
  ROUND = 116,
  HARD_SWISH = 117,
  IF = 118,
  WHILE = 119,
  NON_MAX_SUPPRESSION_V4 = 120,
  NON_MAX_SUPPRESSION_V5 = 121,
  SCATTER_ND = 122,
  SELECT_V2 = 123,
  DENSIFY = 124,
  SEGMENT_SUM = 125,
  BATCH_MATMUL = 126,
  PLACEHOLDER_FOR_GREATER_OP_CODES = 127,
  CUMSUM = 128,
  CALL_ONCE = 129,
  BROADCAST_TO = 130,

  func.return %1 : tensor<10x3x2xf32>

// CHECK-LABEL:scatterNdVectorIndices
// CHECK: %[[CST:.*]] = arith.constant dense<[10, 3, 2]> : tensor<3xi32>
// CHECK: %[[RES:.*]] = "tfl.scatter_nd"(%arg0, %arg1, %[[CST]]) : (tensor<5x1xi32>, tensor<5x3x2xf32>, tensor<3xi32>) -> tensor<10x3x2xf32>
// CHECK: return %[[RES]]
}

  %shape = arith.constant dense<[2]> : tensor<1xi32>
  %0 = "tfl.scatter_nd"(%cst, %arg0, %shape) : (tensor<1x1xi32>, tensor<1xf32>, tensor<1xi32>) -> tensor<2xf32>
  func.return %0 : tensor<2xf32>

// CHECK-LABEL: DontConvertIdentityScatterNdWithLargerOutputDimSize
// CHECK: %[[SCATTER_ND:.*]] = "tfl.scatter_nd"(%{{.*}}, %arg0, %{{.*}})
// CHECK-NEXT: return %[[SCATTER_ND]] : tensor<2xf32>
}

      indices_type.getDimSize(1) != 1)
    return false;

  // Checks the `params_type` is equal to `output_type`. If not equal, we
  // cannot replace the scatter_nd/gather_nd op with `params`.
  if (params_type != output_type) return false;

  // Checks the value in `indices` is from 0 to n-1.
  int cur_value = 0;
  for (const auto &v : indices.getValues<APInt>()) {

  );
}

def TFL_ScatterNdOp : TFL_Op<"scatter_nd", [
    Pure,
    QuantizableResult,
    SameOperandsAndResultsScale,
    TFL_OperandHasAtleastRank<0, 1>,
    TFL_OperandHasAtleastRank<1, 1>,
    PredOpTrait<"updates and output must have same element type",
      TFL_TCresVTEtIsSameAsOp<0, 1>>
  ]> {
  let summary = "Scatter_nd operator";

  let description = [{

}

// -----

// 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

with tf.Session() as sess:
  print sess.run(add)
```

The resulting update to ref would look like this:

    [1, 13, 3, 14, 14, 6, 7, 20]

See `tf.scatter_nd` for more details about how to make updates to
slices.
  }];

  let arguments = (ins
    Arg<TF_ResourceTensor, [{A resource handle. Must be from a VarHandleOp.}], [TF_VariableRead, TF_VariableWrite]>:$ref,

        *   `tf.unsortedsegmentprod` op is supported.
        *   `tf.unsortedsegmentmax` op is supported.
        *   `tf.unsortedsegmentsum` op is supported.
    *   Updates to existing operations:
        *   `tfl.scatter_nd` now supports I1 for the `update` arg.
    *   Upgrade Flatbuffers v2.0.5 from v1.12.0

*   `tf.keras`:

    *   `EinsumDense` layer is moved from experimental to core. Its import path