using mlir::func::FuncOp;
using tflite::OperatorT;
using tflite::TensorT;

namespace errors = tensorflow::errors;
namespace tfl = mlir::TFL;

namespace {

constexpr char kScatterRegionFuncName[] = "update_computation_func_name";

using ::mlir::tf_saved_model::kTfSavedModelExportedNamesAttr;
using ::mlir::tf_saved_model::kTfSavedModelIndexPathAttr;
using ::tflite::IsValidBufferOffset;

// HWIO data format.
//
// Motivation to prepare for the TFLite legalization before the actual
// legalization is to exploit constant folding opportunities in any newly
// created ops by leveraging constant folding support for the TensorFlow ops.
// This way TFLite can be used as a serialization format only and does not
// require access to the TFLite runtime for optimizations as required by the
// TFLite team.

#include <climits>
#include <cstdint>

  TensorType inference_tensor_type = activations_type;
  const bool fully_quantize = !allow_float;

  flatbuffers::FlatBufferBuilder input_builder;
  tflite::FinishModelBuffer(input_builder,
                            tflite::Model::Pack(input_builder, model));

  const std::string input_buffer(
      reinterpret_cast<const char*>(input_builder.GetBufferPointer()),
      input_builder.GetSize());

def L2NormValidReduceIndex : Constraint<CPred<
  "L2NormalizeReduceAxis($0, $1.cast<DenseElementsAttr>())">>;

// Currently L2Normalization doesn't support activation function
// in TFLite.
// TODO(karimnosseir): Add constraints that the kernel code assumes.
// constraint on axis and depth.
multiclass L2NormalizePatterns<Op FirstOp, Op SecondOp> {
  // This pattern constructs L2NormalizationOp from

    default_visibility = ["//visibility:public"],
    licenses = ["notice"],
)

package_group(
    name = "friends",
    packages = [
        "//learning/brain/experimental/mlir/tflite/tfmrt/...",
        "//learning/brain/mlir/...",
        "//third_party/iree/...",
        "//third_party/odml/infra/...",
        "//tensorflow/compiler/mlir/...",
        "//tensorflow/lite/python/...",

limitations under the License.
==============================================================================*/

// This transformation pass prepares for legalization to the TFLite dialect by
// converting Tensorlist operations in TensorFlow dialect into operations that
// can be legalized to TensorFlow Lite dialect with simple replacements.  The

#     rbe_linux_cuda_nvcc:            RBE options to build with GPU support using nvcc.
#
# Embedded Linux options (experimental and only tested with TFLite build yet)
#     elinux:          General Embedded Linux options shared by all flavors.
#     elinux_aarch64:  Embedded Linux options for aarch64 (ARM64) CPU support.