ITLabAI/src/layers/SplitLayer.cpp at a3a29fca1956695ef875c0521611d5707faf6360 · embedded-dev-research/ITLabAI · GitHub

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
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
#include "layers/SplitLayer.hpp"

#include <algorithm>
#include <cstring>

namespace it_lab_ai {

void SplitLayer::run(const Tensor& input, Tensor& output) { output = input; }

void SplitLayer::run(const Tensor& input, std::vector<Tensor>& outputs) {
  validate(input);

  switch (input.get_type()) {
    case Type::kFloat:
      split_impl<float>(input, outputs);
      break;
    case Type::kInt:
      split_impl<int>(input, outputs);
      break;
    default:
      throw std::runtime_error("Unsupported tensor data type");
  }
}

template <typename T>
void SplitLayer::split_impl(const Tensor& input,
                            std::vector<Tensor>& outputs) const {
  const auto& input_data = *input.as<T>();
  const Shape& shape = input.get_shape();
  const int axis = get_normalized_axis(static_cast<int>(shape.dims()));

  std::vector<int> part_sizes;
  if (splits_) {
    part_sizes = *splits_;
  } else {
    const int total_size = static_cast<int>(shape[axis]);
    const int base_size = total_size / *num_outputs_;
    const int remainder = total_size % *num_outputs_;

    part_sizes.reserve(*num_outputs_);
    for (int i = 0; i < *num_outputs_; ++i) {
      part_sizes.push_back(i < remainder ? base_size + 1 : base_size);
    }
  }

  size_t outer_size = 1;
  for (int i = 0; i < axis; ++i) {
    outer_size *= shape[i];
  }

  size_t inner_size = 1;
  for (size_t i = axis + 1; i < shape.dims(); ++i) {
    inner_size *= shape[i];
  }

  const size_t input_axis_stride = shape[axis] * inner_size;

  outputs.clear();
  outputs.reserve(part_sizes.size());

  size_t input_offset = 0;
  for (const auto part_size : part_sizes) {
    const auto output_axis_size = static_cast<size_t>(part_size);

    std::vector<size_t> output_shape_vec(shape.dims());
    for (size_t i = 0; i < shape.dims(); ++i) {
      output_shape_vec[i] =
          (static_cast<int>(i) == axis) ? output_axis_size : shape[i];
    }
    Shape output_shape(output_shape_vec);

    outputs.emplace_back(output_shape, input.get_type());
    auto& output_data = *outputs.back().as<T>();

    const size_t output_part_size = output_axis_size * inner_size;

    for (size_t outer = 0; outer < outer_size; ++outer) {
      const T* input_start =
          &input_data[outer * input_axis_stride + input_offset * inner_size];
      T* output_start = &output_data[outer * output_part_size];

      std::copy_n(input_start, output_part_size, output_start);
    }

    input_offset += output_axis_size;
  }
}

void SplitLayer::validate(const Tensor& input) const {
  if (input.get_shape().dims() == 0) {
    throw std::runtime_error("Cannot split scalar tensor");
  }

  const int axis =
      get_normalized_axis(static_cast<int>(input.get_shape().dims()));
  const int axis_size = static_cast<int>(input.get_shape()[axis]);

  if (splits_) {
    int sum = 0;
    for (int s : *splits_) {
      if (s <= 0) throw std::runtime_error("Split size must be positive");
      sum += s;
    }
    if (sum != axis_size) {
      throw std::runtime_error("Sum of splits must match axis size");
    }
  } else {
    if (*num_outputs_ <= 0) {
      throw std::runtime_error("num_outputs must be positive");
    }
    if (*num_outputs_ > axis_size) {
      throw std::runtime_error("num_outputs (" + std::to_string(*num_outputs_) +
                               ") cannot be greater than axis size (" +
                               std::to_string(axis_size) + ")");
    }
  }
}

int SplitLayer::get_normalized_axis(int rank) const {
  if (axis_ < -rank || axis_ >= rank) {
    throw std::runtime_error("Axis out of bounds");
  }
  return (axis_ < 0) ? axis_ + rank : axis_;
}

template void SplitLayer::split_impl<float>(const Tensor&,
                                            std::vector<Tensor>&) const;
template void SplitLayer::split_impl<int>(const Tensor&,
                                          std::vector<Tensor>&) const;

}  // namespace it_lab_ai