// NOLINTNEXTLINE
opt<bool> import_hlo("import-hlo",
                     llvm::cl::desc("Whether the input file is hlo file."),
                     llvm::cl::init(false));

// NOLINTNEXTLINE
opt<HloImportType> hlo_import_type(
    "hlo-import-type", llvm::cl::desc("The file type of the hlo."),
    llvm::cl::values(clEnumVal(proto, "Import hlo in proto binary format"),

// RUN: tf-mlir-translate -mlir-tf-to-hlo-text %s -tf-input-shapes=10,17:17,19 -tf-xla-emit-use-tuple-args -tf-xla-emit-return-tuple | FileCheck %s
// RUN: tf-mlir-translate -mlir-tf-to-hlo-text %s -tf-input-shapes=10,17:17,19 | FileCheck -check-prefix=NO_TUPLES %s
// RUN: tf-mlir-translate -mlir-tf-to-hlo-text-via-builder %s -tf-input-shapes=10,17:17,19 | FileCheck -check-prefix=NO_TUPLES %s

module attributes {tf.versions = {producer = 179 : i32}} {

// RUN: tf-mlir-translate -mlir-tf-to-hlo-text %s -tf-input-shapes=: -tf-xla-emit-return-tuple | FileCheck %s
// RUN: tf-mlir-translate -mlir-tf-to-hlo-text %s -tf-input-shapes=: -tf-xla-emit-use-tuple-args -tf-xla-emit-return-tuple | FileCheck -check-prefix=TUPLE-ARGS %s
// RUN: tf-mlir-translate -mlir-tf-to-hlo-text %s -tf-input-shapes=: | FileCheck -check-prefix=NO_RET_TUPLE %s

// Declare passes used in xla_legalize_tf.

include "mlir/Pass/PassBase.td"

def LegalizeTF : Pass<"xla-legalize-tf", "ModuleOp"> {
  let summary = "Legalize from TF dialect's or HLO dialect's control flow.";

  let description = [{
    Legalizes from TF dialect to HLO dialect. When allow_partial_conversion is
    false, emits an error if there is any operation that can't be legalized.

                                            RewritePatternSet* patterns);

/// Converts the provided Operation as well as all nested operations into HLO
/// dialect using the conversion patterns registered by the HLO dialect. When
/// allow_partial_conversion is false, emits an error if there is any operation
/// that can't be legalized.
/// When `tf2xla_fallback_device_type` is not `None`, also uses legalization

// RUN: tf-mlir-translate -mlir-tf-to-hlo-text %s -tf-input-shapes=8,16,16,64:64 -tf-xla-emit-use-tuple-args -tf-xla-emit-return-tuple | FileCheck %s
// RUN: tf-mlir-translate -mlir-tf-to-hlo-text-via-builder %s -tf-input-shapes=8,16,16,64:64 | FileCheck %s

module attributes {tf.versions = {producer = 179 : i32}} {

extern llvm::cl::opt<std::string> saved_model_exported_names;

// Import HLO.
enum HloImportType { proto, hlotxt, mlir_text };

extern llvm::cl::opt<bool> import_hlo;
extern llvm::cl::opt<HloImportType> hlo_import_type;

// enable_hlo_to_tf_conversion and disable_hlo_to_tfl_conversion are used to
// control the HLO to TF and HLO to TFLite conversion while debugging an

namespace tensorflow {

template <typename ExecutableType, typename ClientType>
class DeviceCompilerClient {
 public:
  DeviceCompilerClient() = default;
  virtual ~DeviceCompilerClient() = default;

  // Compiles `result` (HLO) to an `ExecutableType` using `ClientType` and
  // returns it.
  virtual StatusOr<std::unique_ptr<ExecutableType>> BuildExecutable(
      const XlaCompiler::Options& options,

                                OpBuilder* builder) {
  return builder->create<ConstantOp>(loc, hlo::getScalarOfType(ty, raw_value));
}

ConstantOp GetScalarNegZeroOfType(Type ty, Location loc, OpBuilder* builder) {
  return builder->create<ConstantOp>(loc, hlo::getScalarNegZeroOfType(ty));
}

DenseIntElementsAttr GetI64ElementsAttr(ArrayAttr attr) {
  RankedTensorType ty =

// RUN: tf-mlir-translate -mlir-tf-to-hlo-text %s -tf-input-shapes=: -tf-xla-emit-use-tuple-args -tf-xla-emit-return-tuple | FileCheck %s
// RUN: tf-mlir-translate -mlir-tf-to-hlo-text %s -tf-input-shapes=: | FileCheck -check-prefix=NO_TUPLES %s
// RUN: tf-mlir-translate -mlir-tf-to-hlo-text-via-builder %s -tf-input-shapes=: | FileCheck -check-prefix=NO_TUPLES %s

module attributes {tf.versions = {producer = 179 : i32}} {
  func.func @main() -> (tensor<0xi32>, tensor<0xi32>) {