inline constexpr char kSavedModelLegacyInitOpKey[] = "legacy_init_op";

/// SavedModel main op collection key. Used in v1 SavedModels.
inline constexpr char kSavedModelMainOpKey[] = "saved_model_main_op";

// CollectionDef key for the SavedModel train op.
// Not exported while export_all_saved_models is experimental.
inline constexpr char kSavedModelTrainOpKey[] = "saved_model_train_op";

// Schema version for SavedModel.

struct FloatType {
  using type = float;
  static constexpr TF_DataType kDType = TF_FLOAT;
};

struct DoubleType {
  using type = double;
  static constexpr TF_DataType kDType = TF_DOUBLE;
};

struct Int32Type {
  using type = int32_t;
  static constexpr TF_DataType kDType = TF_INT32;
};

struct UINT8Type {
  using type = uint8_t;
  static constexpr TF_DataType kDType = TF_UINT8;
};

struct INT8Type {

}

TEST(TestUtil, MlirModuleDoesNotHave) {
  constexpr char arbirary_mlir[] = "arbirary_mlir";
  constexpr char bad_mlir[] = "bad_mlir";

  ASSERT_THAT(success(arbirary_mlir), Not(HasMlirModuleWith(bad_mlir)));
}

TEST(TestUtil, MlirModuleDoesNotHaveFiltered) {
  constexpr char arbirary_mlir[] = "arbirary_mlir";
  constexpr char bad_mlir[] = "bad_mlir";

static constexpr char kClassifyOutputClasses[] = "classes";

/// Classification scores output.
static constexpr char kClassifyOutputScores[] = "scores";

////////////////////////////////////////////////////////////////////////////////
/// Predict API constants.

/// Predict inputs.
static constexpr char kPredictInputs[] = "inputs";

/// Predict method name used in a SignatureDef.

namespace tensorflow {

/// Tag for the `gpu` graph.
constexpr char kSavedModelTagGpu[] = "gpu";

/// Tag for the `tpu` graph.
constexpr char kSavedModelTagTpu[] = "tpu";

/// Tag for the `serving` graph.
constexpr char kSavedModelTagServe[] = "serve";

/// Tag for the `training` graph.
constexpr char kSavedModelTagTrain[] = "train";

}  // namespace tensorflow

  auto const_op = dyn_cast_or_null<TF::ConstOp>(block.front());
  EXPECT_TRUE(const_op);

  return const_op;
}

TEST_F(GetSizeInBytesTest, Int32ScalarConstOpSizeInBytes) {
  constexpr absl::string_view kConstOpExpr =
      R"mlir(%cst = "tf.Const"() {value = dense<1> : tensor<i32>} : () -> tensor<i32>)mlir";

  Block block{};
  TF::ConstOp int_tensor_const_op = ParseConstOp(kConstOpExpr, block, ctx_);

using tpu::TPUCompileMetadataProto;

static constexpr char kMlirLegalizeCount[] =
    "/tensorflow/core/tf2xla/v1/mlir_failed_xla_legalize_tf_count";
static constexpr char kMlirLegalizeErrors[] =
    "/tensorflow/core/tf2xla/v1/mlir_failed_xla_legalize_tf_pass_count";
static constexpr char kBridgeStatusCounter[] =
    "/tensorflow/core/tf2xla/api/v2/phase2_compilation_status";
constexpr char kMlirCombinedMlirSuccess[] = "kMlirCombinedMlirSuccess";

constexpr char kTestDataV2DebugInfo[] =
    "cc/saved_model/testdata/x_plus_y_v2_debuginfo";
constexpr char kTestFuzzGeneratedNegativeShape[] =
    "cc/saved_model/testdata/fuzz_generated/negative_shape";
constexpr char kTestFuzzGeneratedConstWithNoValue[] =
    "cc/saved_model/testdata/fuzz_generated/const_with_no_value";
constexpr char kTestFuzzGeneratedBadNodeAttr[] =
    "cc/saved_model/testdata/fuzz_generated/bad_node_attr";

using ops::Square;
using ops::Variable;

constexpr char kAssignOpName[] = "assign";
constexpr char kCancelOp0[] = "cancel0";
constexpr char kCancelOp1[] = "cancel1";
constexpr char kCloseOp0[] = "close0";
constexpr char kCloseOp1[] = "close1";
constexpr char kCountUpToOpName[] = "count";
constexpr char kDequeueOp0[] = "dequeue0";
constexpr char kDequeueOp1[] = "dequeue1";
constexpr char kEnqueueOp0[] = "enqueue0";

namespace mlir::quant {
namespace {

using ::testing::IsNull;
using ::testing::NotNull;

using FindMainFuncOpTest = ::mlir::quant::QuantizationTestBase;

TEST_F(FindMainFuncOpTest, ReturnsMainFuncOp) {
  constexpr absl::string_view kModuleWithMainFunc = R"mlir(
    module {
      func.func @main() -> () {
        return
      }
    }
  )mlir";

  OwningOpRef<ModuleOp> module_op = ParseModuleOpString(kModuleWithMainFunc);