tensorflow/lite/micro/kernels/sub_common.cc

/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/

#include "tensorflow/lite/c/builtin_op_data.h"
#include "tensorflow/lite/c/common.h"
#include "tensorflow/lite/kernels/internal/common.h"
#include "tensorflow/lite/kernels/internal/quantization_util.h"
#include "tensorflow/lite/kernels/internal/reference/add.h"
#include "tensorflow/lite/kernels/internal/reference/process_broadcast_shapes.h"
#include "tensorflow/lite/kernels/internal/reference/sub.h"
#include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
#include "tensorflow/lite/kernels/internal/types.h"
#include "tensorflow/lite/kernels/kernel_util.h"
#include "tensorflow/lite/kernels/op_macros.h"
#include "tensorflow/lite/micro/kernels/kernel_util.h"
#include "tensorflow/lite/micro/kernels/sub.h"

namespace tflite {

const int kSubInputTensor1 = 0;
const int kSubInputTensor2 = 1;
const int kSubOutputTensor = 0;

TfLiteStatus CalculateOpDataSub(TfLiteContext* context, TfLiteSubParams* params,
                                const TfLiteTensor* input1,
                                const TfLiteTensor* input2,
                                TfLiteTensor* output, OpDataSub* data) {
  data->requires_broadcast = !HaveSameShapes(input1, input2);

  if (output->type == kTfLiteInt8 || output->type == kTfLiteInt16) {
    // 8bit -> 8bit general quantized path, with general rescalings
    data->input1_offset = -input1->params.zero_point;
    data->input2_offset = -input2->params.zero_point;
    data->output_offset = output->params.zero_point;

    // The shift is set to 15 in case of 16-bit and 20 in case of 8-bit,
    // accordingly. In case of 16-bit we have 65535 << 15 which is less than 1
    // << 31, therefore the addition will still fit in a 32 bit accumulator.
    data->left_shift = output->type == kTfLiteInt16 ? 15 : 20;
    const float twice_max_input_scale =
        2 * std::max(input1->params.scale, input2->params.scale);
    const double real_input1_multiplier =
        static_cast<double>(input1->params.scale / twice_max_input_scale);
    const double real_input2_multiplier =
        static_cast<double>(input2->params.scale / twice_max_input_scale);
    const double real_output_multiplier =
        static_cast<double>(twice_max_input_scale /
                            ((1 << data->left_shift) * output->params.scale));

    QuantizeMultiplierSmallerThanOneExp(
        real_input1_multiplier, &data->input1_multiplier, &data->input1_shift);

    QuantizeMultiplierSmallerThanOneExp(
        real_input2_multiplier, &data->input2_multiplier, &data->input2_shift);

    QuantizeMultiplierSmallerThanOneExp(
        real_output_multiplier, &data->output_multiplier, &data->output_shift);

    TF_LITE_ENSURE_STATUS(CalculateActivationRangeQuantized(
        context, params->activation, output, &data->output_activation_min,
        &data->output_activation_max));
  }

  return kTfLiteOk;
}

TfLiteStatus SubPrepare(TfLiteContext* context, TfLiteNode* node) {
  TFLITE_DCHECK(node->user_data != nullptr);
  TFLITE_DCHECK(node->builtin_data != nullptr);

  OpDataSub* data = static_cast<OpDataSub*>(node->user_data);
  auto* params = reinterpret_cast<TfLiteSubParams*>(node->builtin_data);

  MicroContext* micro_context = GetMicroContext(context);

  TfLiteTensor* input1 =
      micro_context->AllocateTempInputTensor(node, kSubInputTensor1);
  TF_LITE_ENSURE(context, input1 != nullptr);
  TfLiteTensor* input2 =
      micro_context->AllocateTempInputTensor(node, kSubInputTensor2);
  TF_LITE_ENSURE(context, input2 != nullptr);
  TfLiteTensor* output =
      micro_context->AllocateTempOutputTensor(node, kSubOutputTensor);
  TF_LITE_ENSURE(context, output != nullptr);

  TF_LITE_ENSURE_STATUS(
      CalculateOpDataSub(context, params, input1, input2, output, data));

  micro_context->DeallocateTempTfLiteTensor(input1);
  micro_context->DeallocateTempTfLiteTensor(input2);
  micro_context->DeallocateTempTfLiteTensor(output);
  return kTfLiteOk;
}

}  // namespace tflite