%0:6 = tf_device.replicate([%arg0, %arg1] as %ri: tensor<*x!tf_type.resource>) {n = 2: i32} {
    %1 = "tf.ReadVariableOp"(%ri) {dtype = "tfdtype$DT_FLOAT"} : (tensor<*x!tf_type.resource>) -> tensor<*xf32>
    %2 = "tf.Shape"(%1) {device = "", T = "tfdtype$DT_FLOAT", out_type = "tfdtype$DT_INT32"} : (tensor<*xf32>) -> tensor<?xi32>

                                                void* data, size_t len,
                                                TF_Status* status) {
  auto dtype = static_cast<tensorflow::DataType>(data_type);
  DCHECK(tensorflow::DataTypeCanUseMemcpy(dtype));

  tensorflow::Tensor tensor(dtype, tensorflow::TensorShape({}));
  std::memcpy(tensorflow::TensorCApi::Buffer(tensor)->data(), data, len);

  status->status = absl::OkStatus();

  std::vector<int64_t> size_as_vector;
};

std::vector<int64_t> IntTensorAsVector(const Tensor& t) {
  DCHECK(t.dtype() == DT_INT32 || t.dtype() == DT_INT64);
  std::vector<int64_t> result;
  result.reserve(t.NumElements());
  for (int i = 0; i < t.NumElements(); i++) {
    int64_t element = t.dtype() == DT_INT32
                          ? static_cast<int64_t>(t.flat<int32>()(i))
                          : t.flat<int64_t>()(i);

                        AttrValueMap* values) {
  AttrValue value;
  auto& type_list = *value.mutable_list();
  for (auto type : types) {
    DataType dtype;
    TF_RETURN_IF_ERROR(ConvertScalarTypeToDataType(type, &dtype));
    type_list.add_type(dtype);
  }

  auto result = values->insert({string(name), value});
  assert(result.second && "cannot have multiple attributes with the same name");
  (void)result;

	if done {
		return
	}

	code := StatusOK

	// If Content-Type isn't set, use the file's extension to find it, but
	// if the Content-Type is unset explicitly, do not sniff the type.
	ctypes, haveType := w.Header()["Content-Type"]
	var ctype string
	if !haveType {
		ctype = mime.TypeByExtension(filepath.Ext(name))
		if ctype == "" {
			// read a chunk to decide between utf-8 text and binary

// depending on indir. The top parameter is whether we are at the top
// level, where Go pointers are allowed. Go pointers to pinned objects are
// allowed as long as they don't reference other unpinned pointers.
func cgoCheckArg(t *_type, p unsafe.Pointer, indir, top bool, msg string) {
	if !t.Pointers() || p == nil {
		// If the type has no pointers there is nothing to do.
		return
	}

      shape_determination_fns_.layout_preference_fn(
          device_tensor->shape(), device_tensor->dtype(), std::nullopt);
  Status status = [&]() -> Status {
    TF_ASSIGN_OR_RETURN(xla::Shape shape,
                        shape_determination_fns_.shape_representation_fn(
                            device_tensor->shape(), device_tensor->dtype(),
                            /*fast_mem=*/false, layout_preference));

// have been serialized so far, most recently used first.
// Note: when a bucket overflows we may end up
// serializing a type more than once. That's ok.
const (
	typeCacheBuckets = 256
	typeCacheAssoc   = 4
)

type typeCacheBucket struct {
	t [typeCacheAssoc]*_type
}

var typecache [typeCacheBuckets]typeCacheBucket

// dump a uint64 in a varint format parseable by encoding/binary.

        operation->SetOpName(absl::StrCat("_input_", inputs_.size()).c_str()));
    TF_RETURN_IF_ERROR(operation->SetAttrType("dtype", dtype));
    if (!shape.unknown_rank()) {
      TF_RETURN_IF_ERROR(operation->SetAttrShape(
          "shape", reinterpret_cast<int64_t*>(shape.dim_sizes().data()),
          shape.dims()));
    }
    int num_outputs = 1;
    std::vector<AbstractTensorHandle*> outputs(num_outputs);

}

// tryRegAssignArg tries to register-assign a value of type t.
// If this type is nested in an aggregate type, then offset is the
// offset of this type within its parent type.
// Assumes t.size <= goarch.PtrSize and t.size != 0.
//
// Returns whether the assignment succeeded.
func (p *abiDesc) tryRegAssignArg(t *_type, offset uintptr) bool {
	switch k := t.Kind_ & abi.KindMask; k {