/*
 * Copyright (c) 2018 Samsung Electronics Co., Ltd. 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 "internal/layers/SimpleSpaceToDepth.h"

#include <arm_compute/runtime/CL/CLScheduler.h>

void SimpleSpaceToDepth::configure(::arm_compute::ITensor *input, ::arm_compute::ITensor *output,
                                   int32_t block_size, const ::arm_compute::Coordinates &axises)
{
  const auto rank = axises.num_dimensions();
  assert(rank == 4);
  for (int i = 0; i < rank; ++i)
  {
    assert(axises[i] >= 0);
    assert(axises[i] < rank);
  }

  _input = input;
  _output = output;
  _block_size = block_size;
  _axises = axises;
}

template <typename T>
inline void SpaceToDepth(const ::arm_compute::ITensor *input,
                         const ::arm_compute::TensorShape &input_shape, int32_t block_size,
                         ::arm_compute::ITensor *output,
                         const ::arm_compute::TensorShape &output_shape,
                         const ::arm_compute::Coordinates &axises)
{
  const int input_batch = input_shape[axises[0]];
  const int input_height = input_shape[axises[1]];
  const int input_width = input_shape[axises[2]];
  const int input_depth = input_shape[axises[3]];

  for (int in_b = 0; in_b < input_batch; ++in_b)
  {
    for (int in_h = 0; in_h < input_height; ++in_h)
    {
      for (int in_w = 0; in_w < input_width; ++in_w)
      {
        for (int in_d = 0; in_d < input_depth; ++in_d)
        {
          const int out_b = in_b;
          const int out_h = in_h / block_size;
          const int out_w = in_w / block_size;
          const int out_d =
              in_d + ((in_h % block_size) * block_size + in_w % block_size) * input_depth;

          auto input_id =
              asARMComputeCoordinates(::arm_compute::Coordinates{in_b, in_h, in_w, in_d}, axises);
          auto output_id = asARMComputeCoordinates(
              ::arm_compute::Coordinates{out_b, out_h, out_w, out_d}, axises);

          *reinterpret_cast<T *>(output->ptr_to_element(output_id)) =
              *reinterpret_cast<T *>(input->ptr_to_element(input_id));
        }
      }
    }
  }
}

void SimpleSpaceToDepth::run()
{
  if (::internal::arm_compute::isGpuMode())
  {
    auto &q = ::arm_compute::CLScheduler::get().queue();

    CAST_CL(_input)->map(q);
    CAST_CL(_output)->map(q);
  }

  switch (_input->info()->data_type())
  {
    case ::arm_compute::DataType::U8:
    case ::arm_compute::DataType::QASYMM8:
      SpaceToDepth<uint8_t>(_input, _input->info()->tensor_shape(), _block_size, _output,
                            _output->info()->tensor_shape(), _axises);
      break;
    case ::arm_compute::DataType::F32:
      SpaceToDepth<float>(_input, _input->info()->tensor_shape(), _block_size, _output,
                          _output->info()->tensor_shape(), _axises);
      break;
    default:
      ARM_COMPUTE_ERROR("DataType not supported");
      break;
  }

  if (::internal::arm_compute::isGpuMode())
  {
    auto &q = ::arm_compute::CLScheduler::get().queue();

    CAST_CL(_input)->unmap(q);
    CAST_CL(_output)->unmap(q);
  }
}