}

def BatchFunctionOp : TensorflowMlrt_Op<"batch_function", [Pure]> {
  let summary = "Fallback ExecuteOp specialized for tf.BatchFunction.";

  let description = [{
    This kernel executes a variant tf.BatchFunction kernel that supports having
    the `f` attribute as a bytecode function.

    Example:
      %res = tf_mlrt.batch_function(%input, %captured_input)  {
          device = "/device:CPU:0",

func.func @callee(%arg0: tensor<i32>, %arg1: tensor<i32>) -> (tensor<i32>) {
  func.return %arg0: tensor<i32>
}

// CHECK-LABEL: func @batch_function
func.func @batch_function(%arg0: tensor<i32>) -> (tensor<i32>) {
  // CHECK: [[batch_result_future:%.*]] = tf_mlrt.batch_function
  // CHECK: [[batch_result:%.*]] = tf_mlrt.await [[batch_result_future]]

}

// -----

// Test lowering BatchFunction op.

func.func @batched_function(%x: tensor<?xi32>) -> tensor<?xi32> {
  return %x : tensor<?xi32>
}

// CHECK-LABEL: func @main
func.func @main(%x: tensor<1xi32>) -> (tensor<1xi32>, tensor<1xi32>, tensor<1xi32>) {
  // CHECK: [[y_future:%.*]] = tf_mlrt.batch_function
  // CHECK-SAME: f = @batched_function
  // CHECK-SAME: \22batch_function\22

}

def BatchFunctionOp : FallbackAsync_Op<"batch_function",
    [Pure, CoreRT_TypedAttributeTrait]> {
  let summary = "Fallback ExecuteOp specialized for tf.BatchFunction.";
  let description = [{
    This kernel executes a variant tf.BatchFunction kernel that supports having
    the `f` attribute as a BEF function.

    Example:
      %res = tfrt_fallback_async.batch_function device("/CPU:0") @f(%input, %captured_input)  {

              !tfrt_fallback.tf_tensor, !tfrt_fallback.tf_tensor, !tfrt_fallback.tf_tensor, !tfrt_fallback.tf_tensor, !tfrt_fallback.tf_tensor, !tfrt_fallback.tf_tensor, !tfrt_fallback.tf_tensor) {

  %res:112 = tfrt_fallback_async.batch_function device("/device:CPU:0") @f (%arg1, %arg2, %arg3, %arg4, %arg5, %arg6, %arg7, %arg8, %arg9, %arg10,%arg11, %arg12, %arg13, %arg14,

  // CHECK: "tf.BatchFunction"(%arg0, %0)
  // CHECK: operandSegmentSizes = array<i32: 1, 1>

  // CHECK: "tf.BatchFunction"(%arg0, %0)
  // CHECK: operandSegmentSizes = array<i32: 1, 1>

  CopyDeviceAndUnderscoredAttributes(case_op, case_region);

  for (const auto& item : llvm::enumerate(case_region.getBranches())) {
    CreateCall(case_op, case_op.branch_function(item.index()),
               /*caller_region=*/item.value(), case_op.getInput(),
               /*use_region_args=*/false,
               /*forward_block_args=*/false);
  }

  func.func @f(%arg: tensor<1xf32>) -> (tensor<*xf32>, tensor<*xf32>) {
    %handle = "tf.VarHandleOp"() {shared_name = "var1"} : () -> tensor<!tf_type.resource<tensor<0xf32>>>