syntax = "proto3";

package third_party.tensorflow.compiler.mlir.lite.experimental.tac;

// A list of filters for TAC users to run ops/functions on ML hardwares. The
// intuition is that, for ops/functions that can be run on ML hardware (e.g.
// EdgeTPU) and TFLite CPU, TAC users give a hint that they're more performant
// to run on TFLite CPU. These filters give the TAC users freedom to specify the
// parts that they want to use other hardware to accelerate.

// variables should be legalized to TFLite or not.
const char kLegalizeTflVariables[] = "tfl._legalize_tfl_variables";

// Returns true if 'op' is TF op that accepts resource type, but is
// supported by TFLite.
bool IsSupportedTFLiteResourceOp(Operation* op) {
  return llvm::isa<TF::ReadVariableOp, TF::AssignVariableOp, TF::VarHandleOp,
                   TF::LookupTableFindV2Op, TF::LookupTableImportV2Op,
                   TF::LookupTableSizeV2Op>(op);

               << ", input_inference_type: "
               << tflite::EnumNameTensorType(input_type)
               << ", output_inference_type: "
               << tflite::EnumNameTensorType(output_type) << "\n";
  mlir::Builder mlir_builder(&context);
  mlir::Type input_mlir_type =
      tflite::ConvertElementType(input_type, mlir_builder);
  mlir::Type output_mlir_type =
      tflite::ConvertElementType(output_type, mlir_builder);

        return tensorflow::PyoOrThrow(
            tflite::MlirSparsifyModel(input_contents_txt_raw.ptr()));
      },
      py::arg("input_contents_txt_raw"),
      R"pbdoc(
      Returns a sparsified model.
    )pbdoc");
  m.def(
      "RegisterCustomOpdefs",
      [](py::object custom_opdefs_txt_raw) {
        return tensorflow::PyoOrThrow(
            tflite::RegisterCustomOpdefs(custom_opdefs_txt_raw.ptr()));
      },

  // Load input model
  auto input_fbm = tflite::FlatBufferModel::BuildFromFile(
      "tensorflow/lite/testdata/sparse_tensor.bin");
  tflite::ModelT input_model;
  input_fbm->GetModel()->UnPackTo(&input_model);

  // Populate input metadata
  auto model_metadata_buffer = std::make_unique<tflite::BufferT>();
  model_metadata_buffer->data =
      std::vector<uint8_t>(expected_value.begin(), expected_value.end());

#include "tensorflow/compiler/mlir/lite/schema/schema_generated.h"
#include "tensorflow/core/framework/tensor.pb.h"
#include "tensorflow/core/framework/tensor_shape.pb.h"

namespace mlir {
namespace TFL {

bool IsQuantized(const tflite::TensorT& tensor);

absl::StatusOr<mlir::quant::QuantizedType> GetQuantizedType(
    const tflite::TensorT& tensor, mlir::Builder builder,

bool use_buffer_offset;
bool emit_stablehlo_ops;
bool disable_vhlo_to_stablehlo;

// NOLINTNEXTLINE
static opt<bool, true> emit_builtin_tflite_ops_flag(
    "emit-builtin-tflite-ops",
    llvm::cl::desc(
        "Emit TFLite built in operations in the generated TFLite model"),
    llvm::cl::location(emit_builtin_tflite_ops), llvm::cl::init(true));

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

              [&]() {});
        });
  }
};
}  // namespace

tflite::metrics::ConverterErrorData NewConverterErrorData(
    const std ::string& pass_name, const std::string& error_message,
    tflite::metrics::ConverterErrorData::ErrorCode error_code,
    const std::string& op_name, const Location& location) {
  using tflite::metrics::ConverterErrorData;
  ConverterErrorData error;
  if (!pass_name.empty()) {

#include <string>
#include <vector>

#include "absl/status/status.h"
#include "llvm/Support/SourceMgr.h"
#include "tensorflow/compiler/mlir/lite/experimental/tac/tac_importer_exporter.h"

namespace mlir {
namespace TFL {
namespace tac {
// TAC Importer for TFLite.
// This import to MLIR from tflite file or MLIR
class TfLiteImporter : public mlir::TFL::tac::TacImporter {
 public:
  // Options for configuring the importer.

struct ConverterErrorDataComparison {
  std::size_t operator()(
      const tflite::metrics::ConverterErrorData& a,
      const tflite::metrics::ConverterErrorData& b) const noexcept {
    return ConverterErrorDataHash()(a) == ConverterErrorDataHash()(b);
  }
};

// Helper function to create a new ConverterErrorData.
tflite::metrics::ConverterErrorData NewConverterErrorData(