Function Prepare

tensorflow/lite/kernels/div.cc:63–98

}

TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
  auto* params = reinterpret_cast<TfLiteDivParams*>(node->builtin_data);
  OpData* data = reinterpret_cast<OpData*>(node->user_data);

  TF_LITE_ENSURE_EQ(context, NumInputs(node), 2);
  TF_LITE_ENSURE_EQ(context, NumOutputs(node), 1);

  const TfLiteTensor* input1 = GetInput(context, node, kInputTensor1);
  const TfLiteTensor* input2 = GetInput(context, node, kInputTensor2);
  TfLiteTensor* output = GetOutput(context, node, kOutputTensor);

  TF_LITE_ENSURE_EQ(context, input1->type, input2->type);
  output->type = input2->type;

  data->requires_broadcast = !HaveSameShapes(input1, input2);

  TfLiteIntArray* output_size = nullptr;
  if (data->requires_broadcast) {
    TF_LITE_ENSURE_OK(context, CalculateShapeForBroadcast(
                                   context, input1, input2, &output_size));
  } else {
    output_size = TfLiteIntArrayCopy(input1->dims);
  }

  if (output->type == kTfLiteUInt8) {
    CalculateActivationRangeUint8(params->activation, output,
                                  &data->output_activation_min,
                                  &data->output_activation_max);
    const double real_multiplier =
        input1->params.scale / (input2->params.scale * output->params.scale);
    QuantizeMultiplier(real_multiplier, &data->output_multiplier,
                       &data->output_shift);
  }

  return context->ResizeTensor(context, output, output_size);
}

template <KernelType kernel_type>
void EvalDiv(TfLiteContext* context, TfLiteNode* node, TfLiteDivParams* params,