(replaceWithValue $params),
  [(CanOptimizeIdentityGatherNdOrScatterNdOp $params, $indices, $output)]>;

def OptimizeIdentityScatterNdOp : Pat<
  (TFL_ScatterNdOp:$output (Arith_ConstantOp I32ElementsAttr: $indices), $params, $ignored),
  (replaceWithValue $params),
  [(CanOptimizeIdentityGatherNdOrScatterNdOp $params, $indices, $output)]>;

    }
  }

  return indices;
}

// Given a list of tensor indices, returns true if any of the tensors have
// non-empty name strings.
bool HasNonEmptyNames(const tflite::SubGraphT& subgraph,
                      ArrayRef<int32_t> indices) {
  return llvm::any_of(
      indices, [&](int i) { return !subgraph.tensors.at(i)->name.empty(); });
}

      op->setAttr(attr.getName(),
                  FlatSymbolRefAttr::get(op->getContext(), cloned.getName()));
      break;
    }
  }

  // Traverse results backward so that indices to be deleted stay unchanged.
  for (OpResult result : llvm::reverse(op->getResults())) {
    if (!result.use_empty()) continue;
    int result_idx = result.getResultNumber();
    for (func::FuncOp func : cloned_branches)

    // final_op = "Div"
    llvm::SmallVector<int64_t> indices(group_size.getInt());
    std::iota(indices.begin(), indices.end(), 0);

    auto replica_groups = mlir::DenseIntElementsAttr::get(
        mlir::RankedTensorType::get({1, group_size.getInt()},
                                    rewriter.getI64Type()),
        indices);

    {

  let description = [{
    Get a tensor in resource array.

    indices: the indices in the resource array.
    results: the tensor values for the corresponding indices.

    in_ch and out_ch are for control dependencies.
  }];

  let arguments = (ins
    TFRT_ChainType:$in_ch,
    StrAttr:$device,
    I64ArrayAttr:$indices
  );

  let results = (outs
    TFRT_ChainType:$out_ch,

}

// CHECK-LABEL: func @tensor_array_while_test
// CHECK-SAME: ([[in_chain:%.*]]: !tfrt.chain
func.func @tensor_array_while_test(%indices: tensor<?xi32>, %input_0: tensor<?x?x?xf32>, %input_1: tensor<?x?x?xf32>) -> (tensor<?x?x512xf32>, tensor<?x?x512xf32>) {
  %index = "tf.Const"() {device = "/device:CPU:0", value = dense<0> : tensor<i32>} : () -> (tensor<i32>)

  let summary = "get a tensor in resource array";

  let description = [{
    Get a tensor in resource array.

    indices: the indices in the resource array.
    results: the tensor values for the corresponding indices.
  }];

  let arguments = (ins
    I64ArrayAttr:$indices
  );

  let results = (outs
    Variadic<TFTensorType>:$results
  );

	trustAD       bool          // add AD flag to queries
	noReload      bool          // do not check for config file updates
}

// serverOffset returns an offset that can be used to determine
// indices of servers in c.servers when making queries.
// When the rotate option is enabled, this offset increases.
// Otherwise it is always 0.
func (c *dnsConfig) serverOffset() uint32 {
	if c.rotate {

                "of indices");
    }
  }

  if ((indices_rank == 0) || (indices_rank == batch_dims_i)) {
    // Scalar indices (output is rank(params) - 1).
    // Erase shape[axis]
    shape.erase(shape.begin() + axis_i);
  } else if (indices_rank == 1) {
    // Vector indices (output is rank(params)).
    // Copy indices.shape into params.shape[axis]

    $constant_operand_indices are the indices of the inputs that are constant to the TPU program (e.g. weights in inference), the rest of the inputs are input tensors.
    constant_operand_indices is sorted in ascending order.
    $operands_with_static_shape are indices of operands that are tagged with a maximum static shape.