onert-micro/luci-interpreter/src/kernels/BinaryOpCommon.h

   1 /*
   2  * Copyright (c) 2020 Samsung Electronics Co., Ltd. All Rights Reserved
   3  * Copyright 2017 The TensorFlow Authors. All Rights Reserved.
   4  *
   5  * Licensed under the Apache License, Version 2.0 (the "License");
   6  * you may not use this file except in compliance with the License.
   7  * You may obtain a copy of the License at
   8  *
   9  *    http://www.apache.org/licenses/LICENSE-2.0
  10  *
  11  * Unless required by applicable law or agreed to in writing, software
  12  * distributed under the License is distributed on an "AS IS" BASIS,
  13  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  14  * See the License for the specific language governing permissions and
  15  * limitations under the License.
  16  */
  17
  18 #ifndef LUCI_INTERPRETER_KERNELS_BINARYOPUTILS_H
  19 #define LUCI_INTERPRETER_KERNELS_BINARYOPUTILS_H
  20
  21 #include "tensorflow/lite/kernels/internal/common.h"
  22 #include "tensorflow/lite/kernels/internal/types.h"
  23
  24 namespace luci_interpreter
  25 {
  26 namespace kernels
  27 {
  28
  29 // Derived from tensorflow/lite/kernels/internal/reference/maximum_minimum.h (v2.3.0).
  30 template <typename T, typename Op, int N = 5>
  31 void BinaryOpBroadcastSlow(const tflite::RuntimeShape &unextended_input1_shape,
  32                            const T *input1_data,
  33                            const tflite::RuntimeShape &unextended_input2_shape,
  34                            const T *input2_data,
  35                            const tflite::RuntimeShape &unextended_output_shape, T *output_data,
  36                            Op op)
  37 {
  38   if (unextended_input1_shape == unextended_input2_shape)
  39   {
  40     const int flat_size = tflite::MatchingElementsSize(
  41       unextended_input1_shape, unextended_input2_shape, unextended_output_shape);
  42     for (int i = 0; i < flat_size; ++i)
  43     {
  44       output_data[i] = op(input1_data[i], input2_data[i]);
  45     }
  46   }
  47   else
  48   {
  49     assert(unextended_input1_shape.DimensionsCount() <= N);
  50     assert(unextended_input2_shape.DimensionsCount() <= N);
  51     assert(unextended_output_shape.DimensionsCount() <= N);
  52
  53     tflite::NdArrayDesc<N> desc1{};
  54     tflite::NdArrayDesc<N> desc2{};
  55     tflite::NdArrayDesc<N> output_desc{};
  56     tflite::NdArrayDescsForElementwiseBroadcast(unextended_input1_shape, unextended_input2_shape,
  57                                                 &desc1, &desc2);
  58     tflite::CopyDimsToDesc(tflite::RuntimeShape::ExtendedShape(N, unextended_output_shape),
  59                            &output_desc);
  60
  61     auto fn = [&](int indexes[N]) {
  62       output_data[SubscriptToIndex(output_desc, indexes)] =
  63         op(input1_data[SubscriptToIndex(desc1, indexes)],
  64            input2_data[SubscriptToIndex(desc2, indexes)]);
  65     };
  66     tflite::NDOpsHelper<N>(output_desc, fn);
  67   }
  68 }
  69
  70 } // namespace kernels
  71 } // namespace luci_interpreter
  72
  73 #endif // LUCI_INTERPRETER_KERNELS_BINARYOPUTILS_H