summaryrefslogtreecommitdiff
path: root/runtimes/pure_arm_compute/src/internal/layers/SimplePackLayer.cc
blob: 2a0a25f0c07b0b5f1ce51ac9196f7164f0c72c55 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
/*
 * 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/arm_compute.h"
#include "SimplePackLayer.h"

void SimplePackLayer::configure(const std::vector<::arm_compute::ICLTensor *> &input_vector,
                                ::arm_compute::ICLTensor *output, int32_t axis)
{
  uint32_t nr_inputs = input_vector.size();
  uint32_t output_rank = output->info()->num_dimensions();
  const ::arm_compute::PermutationVector pv{1, 2, 0};
  _cl_permuted_vector.resize(nr_inputs);
  _cl_permute_vector.resize(nr_inputs);

  _output = output;
  // A negative axis implies axis from the end.
  // For example, axis = -1 implies the first axis from the end, i.e. axis = Rank - 1.
  // Similarly, axis = -2 imples second axis from the end, i.e. axis = Rank - 2.
  if (axis < 0)
  {
    axis += output_rank;
  }
  _axis = ToARMComputeAxis(output_rank, axis).value();
  _cl_reshape_vector.resize(nr_inputs);

  ::arm_compute::TensorShape subTensor_shape{};
  for (int i = 0; i < output_rank; i++)
  {
    if (i != _axis)
    {
      subTensor_shape.set(i, _output->info()->tensor_shape()[i]);
    }
    else
    {
      subTensor_shape.set(i, 1);
    }
  }

  auto subTensor_offset = ::arm_compute::Coordinates{};
  subTensor_offset.set_num_dimensions(output_rank);

  for (int i = 0; i < input_vector.size(); i++)
  {
    _input_vector.push_back(input_vector[i]);
    subTensor_offset[_axis] = i;
    auto temp_tensor = std::make_shared<::arm_compute::CLSubTensor>(
        CAST_CL(_output), subTensor_shape, subTensor_offset, true);
    _sub_tensor_vector.push_back(temp_tensor);
    // configure to resize of input tensor in sub tensor offseted, dimension expansion will be
    // automatic
    _cl_permute_vector[i].configure(CAST_CL(_input_vector[i]), &_cl_permuted_vector[i], pv);
    _cl_reshape_vector[i].configure(&_cl_permuted_vector[i], _sub_tensor_vector[i].get());
    _cl_permuted_vector[i].allocator()->allocate();
  }
}

void SimplePackLayer::run(void)
{
  for (int i = 0; i < _input_vector.size(); i++)
  {
    _cl_permute_vector[i].run();
    _cl_reshape_vector[i].run();
  }
}