/*
 * 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 <OperationsUtils.h>
#include <arm_compute/core/TensorShape.h>
#include <arm_compute/core/TensorInfo.h>

#include <cassert>

// TODO: fix include path in CMakeFiles
#include "../IO_accessor.h"
#include "../shape.h"

namespace nnfw {
namespace kernel {
namespace acl {

bool concatenationFloat32(const std::vector<const float*>& inputDataPtrs,
                          const std::vector<nnfw::rt::Shape>& inputShapes, int32_t axis,
                          float* outputData, const nnfw::rt::Shape& outputShape)
{
  if (axis != 3)
  {
    assert("Only support axis=3 for ACL" && 0);
    return false;
  }
  assert(inputDataPtrs.size() == inputShapes.size());

  std::vector<arm_compute::CLTensor*> inputPtrs;
  std::vector<arm_compute::ICLTensor*> inputIptrs;
  arm_compute::CLTensor output;

  // init Tensors
  std::vector<nnfw::rt::Shape>::const_iterator it_inputShape = inputShapes.begin();
  for (auto inputData : inputDataPtrs)
  {
    const nnfw::rt::Shape& inputShape = *it_inputShape;
    arm_compute::TensorShape input_shape = util::fromNNShape(inputShape);
    arm_compute::CLTensor* inputPtr = new arm_compute::CLTensor();

    inputPtr->allocator()->init(arm_compute::TensorInfo(input_shape, arm_compute::Format::F32));
    inputPtrs.push_back(inputPtr);
    inputIptrs.push_back(inputPtr);

    it_inputShape++;
  }
  arm_compute::TensorShape output_shape = util::fromNNShape(outputShape);
  output.allocator()->init(arm_compute::TensorInfo(output_shape, arm_compute::Format::F32));

  // prepare ACL Concatenate and configure tensors
  auto concat = std::make_shared<arm_compute::CLDepthConcatenateLayer>();
  concat->configure(inputIptrs, &output);

  // allocate Tensors
  it_inputShape = inputShapes.begin();
  std::vector<const float*>::const_iterator it_inputData = inputDataPtrs.begin();
  for (auto inputPtr : inputPtrs)
  {
    inputPtr->allocator()->allocate();

    const float* inputData = *it_inputData;
    const nnfw::rt::Shape& inputShape = *it_inputShape;

    TensorAccess<InputAccessor>(*inputPtr, inputData, inputShape);

    it_inputShape++;
    it_inputData++;
  }
  output.allocator()->allocate();

  // run
  concat->run();
  arm_compute::CLScheduler::get().sync();

  // get output
  TensorAccess<OutputAccessor>(output, outputData, outputShape);

  // cleanup
  for (auto inputPtr : inputPtrs)
  {
    inputPtr->allocator()->free();
    delete inputPtr;
  }
  output.allocator()->free();

  return true;
}

} // namespace acl
} // namespace kernel
} // namespace nnfw