// meaningless value -1 for non resource outputs.
    out_desc.input_index =
        it != output_to_input_alias.end() ? it->getSecond() : -1;
    // MLIR-based TF-Compiler bridge doesn't support tensorlist output yet.
    // TODO(ycao): Support tensorlist-type output.
    out_desc.is_tensor_list = false;
  }

  // XLA computation always uses Tuple shape.
  *xla_output_shape = xla::ShapeUtil::MakeTupleShape(shapes);

  let summary =
    "Replicate TensorList init ops for correct shape assignments in shape inference";

  let description = [{
    If we pass same TensorList to a while op as multiple arguments or just use
    the same TensorList at multiple places and assign different
    TensorListSetItem to elements of TensorList, the shape inference is then

    llvm::cl::init(false));

// NOLINTNEXTLINE
opt<bool> legalize_custom_tensor_list_ops(
    "legalize-custom-tensor-list-ops",
    llvm::cl::desc("Convert \"tf.TensorList*\" ops to \"tfl.custom_op\""
                   "if they can all be supported."),
    llvm::cl::init(false));

// NOLINTNEXTLINE
opt<bool> serialize_stablehlo_ops(
    "serialize-stablehlo-ops",

_Replicate TensorList init ops for correct shape assignments in shape inference_

If we pass same TensorList to a while op as multiple arguments or just use
the same TensorList at multiple places and assign different
TensorListSetItem to elements of TensorList, the shape inference is then
unable to identify the Shape of these args and thus the input TensorList
shape is unidentifiable.

    llvm::cl::location(emit_custom_ops), llvm::cl::init(false));

// NOLINTNEXTLINE
static opt<bool, true> lower_tensor_list_ops_flag(
    "lower-tensor-list-ops",
    llvm::cl::desc("Lower the TensorList ops within the TFLite dialect"),
    llvm::cl::location(lower_tensor_list_ops), llvm::cl::init(false));

// NOLINTNEXTLINE
static opt<bool, true> strip_debug_info_flag(

// Creates a pass that adds control dependencies to keep the relative execution
// order of operations with side effects frozen.
std::unique_ptr<OperationPass<func::FuncOp>> CreatePinOpsWithSideEffectsPass();

// Legalize TensorList Ops iff all of them are supported.
std::unique_ptr<OperationPass<ModuleOp>> CreateLegalizeTensorListPass();

// Reduce the type precision of some tensor types if all values within that

// RUN: tf-opt "-tfl-lower-static-tensor-list=allow-tensorlist-pass-through default-to-single-batch" -split-input-file %s | FileCheck %s

// -----

// CHECK-LABEL: tensorlistConst
func.func @tensorlistConst(%arg0 : tensor<1xi32>) -> tensor<2x3xi32> {
  // CHECK-DAG: %[[ELEMENT0:.*]] = "tf.Const"() <{value = dense<[0, 1, 2]> : tensor<3xi32>}> : () -> tensor<3xi32>

  }

  for (const XlaOutputDescription& descr : compilation_result->outputs) {
    if (descr.type == DT_VARIANT) {
      return errors::Unimplemented(
          "Support for TensorList crossing the XLA/TF boundary "
          "is not implemented");
    }
  }

  std::vector<TensorShape> output_tensor_shapes;
  output_tensor_shapes.reserve(ctx->num_outputs());

      mlir::TF::RecvTPUEmbeddingActivationsOp::getOperationName(), context));
  supported_ops->insert(OperationName(
      mlir::TF::SendTPUEmbeddingGradientsOp::getOperationName(), context));
}

// Stack, TensorList and TensorArray ops are rewritten during the second phase
// of the bridge (compilation of TPUCompile op). They would not match any
// legalization/canonicalization pattern and have to be manually added to the

// Materialize the MlirPassthroughOp by replacing it with the MLIR module
// attached as an attribute.
std::unique_ptr<OperationPass<func::FuncOp>>
CreateMaterializePassthroughOpPass();

// Replicates the TensorList init op by undoing some CSE needed for correct
// shape assignment in shape_inference.
std::unique_ptr<OperationPass<func::FuncOp>>
CreateReplicateTensorListInitOpsPass();