return -1, -1, -1
	}
	var namebuf [1024]byte
	sz := int32(copy(namebuf[:], path))
	A2e(namebuf[:sz])
	attr_sz := uint32(unsafe.Sizeof(*attr))
	var parms [7]unsafe.Pointer
	parms[0] = unsafe.Pointer(&sz)
	parms[1] = unsafe.Pointer(&namebuf[0])
	parms[2] = unsafe.Pointer(&attr_sz)
	parms[3] = unsafe.Pointer(attr)
	parms[4] = unsafe.Pointer(&rv)
	parms[5] = unsafe.Pointer(&rc)
	parms[6] = unsafe.Pointer(&rn)

  string c_base_type = attr_.base_type();
  if (attr_.base_type() == "type") {
    c_base_type = "DataType";
  } else if (attr_.base_type() == "shape") {
    c_base_type = "const PartialTensorShape";
  }

  // ... and wrap in a Span<> if it's a list.
  if (attr_.is_list()) {
    return absl::Substitute("absl::Span<$0>", c_base_type);
  } else {
    return c_base_type;
  }

  return attr_.full_type();
}

  std::srand(0);
  for (auto fn : module->getOps<mlir::func::FuncOp>()) {
    for (auto& bb : fn) {
      for (auto& inst : bb) {
        auto attr_id = mlir::StringAttr::get(context, "value");
        if (auto attr = inst.getAttrOfType<mlir::ElementsAttr>(attr_id)) {
          mlir::Attribute rand_val;
          mlir::Type element_type = attr.getShapedType().getElementType();

    return func_.emitWarning()
           << "Invalid number of results from " << kMaxUnpooling << ": "
           << func_.getFunctionType().getNumResults();
  }

  auto attrs = attr_.getAttrs();

  if (failed(HasIntegerArrayWithSize(&func_, attrs, "pool_size", 2))) {
    return failure();
  }

  if (failed(HasIntegerArrayWithSize(&func_, attrs, "strides", 2))) {
    return failure();

                 StringAttr::get(func_.getContext(), kCustomSSDPostprocessing));

  OpBuilder builder(func_.getBody());
  std::string custom_option_buffer;
  if (failed(CreateNMSCustomOptions(func_, attr_.getAttrs(),
                                    custom_option_buffer))) {
    return failure();
  }
  auto op = builder.create<CustomOp>(
      func_.getLoc(), func_.getFunctionType().getResults(),

                       .getLeafReference()
                       .getValue());
    // Remove the attribute from the instruction as it is already converted to
    // op_name.
    auto attr_id = mlir::StringAttr::get(inst->getContext(), "f");
    inst->removeAttr(attr_id);
  } else {
    // Some control flow ops in TensorFlow Graph have their respective "Ref" ops
    // as well. For example there is Enter and RefEnter op. RefEnter forwards

LogicalResult ConvertToI32Attr(IntegerAttr attr, IntegerAttr* attr_i32) {
  if (attr.getType().isInteger(/*width=*/32)) {
    *attr_i32 = attr;
    return success();
  }

  int64_t value = attr.getInt();
  if (value > std::numeric_limits<int>::max() ||
      value < std::numeric_limits<int>::min()) {
    return failure();
  }

  *attr_i32 = IntegerAttr::get(

        string_attr.getValue() == kNullAttributeValue) {
      continue;
    }

    if (std::find_if(
            attr_to_op_map.begin(), attr_to_op_map.end(), [&](auto attr_op) {
              return std::get<0>(attr_op).getName() == attribute.getName();
            }) == attr_to_op_map.end()) {
      emitError(UnknownLoc::get(&context),
                "Could not find attribute: " + attribute.getName().str());

      tensorflow::Tensor tensor(list->element_dtype,
                                tensorflow::TensorShape(result_shape));
      auto attr_or = tensorflow::ConvertTensor(tensor, &rewriter);
      if (!attr_or.ok()) return failure();
      rewriter.replaceOpWithNewOp<TF::ConstOp>(op, attr_or.value());
      return success();
    }

    // Extract individual tensor list element and combine them using the tf.Pack
    // op.

LogicalResult ConvertToI32Attr(IntegerAttr attr, IntegerAttr* attr_i32) {
  if (attr.getType().isInteger(/*width=*/32)) {
    *attr_i32 = attr;
    return success();
  }

  int64_t value = attr.getInt();
  if (value > std::numeric_limits<int>::max() ||
      value < std::numeric_limits<int>::min()) {
    return failure();
  }

  *attr_i32 = IntegerAttr::get(