| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "sabirov_s_linear_filtering_block_partitioning/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <array> | ||
| 7 | #include <cstddef> | ||
| 8 | #include <cstdint> | ||
| 9 | |||
| 10 | #include "sabirov_s_linear_filtering_block_partitioning/common/include/common.hpp" | ||
| 11 | |||
| 12 | namespace sabirov_s_linear_filtering_block_partitioning { | ||
| 13 | |||
| 14 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | SabirovSLinearFilteringBlockPartitioningMPI::SabirovSLinearFilteringBlockPartitioningMPI(const InType &in) { |
| 15 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 16 | GetInput() = in; | ||
| 17 | 10 | } | |
| 18 | |||
| 19 | 10 | bool SabirovSLinearFilteringBlockPartitioningMPI::ValidationImpl() { | |
| 20 | 10 | int rank = 0; | |
| 21 | 10 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 22 | |||
| 23 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 24 | const auto &input = GetInput(); | ||
| 25 |
4/8✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 5 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 5 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✓ Branch 7 taken 5 times.
|
5 | bool valid = !input.Empty() && input.width > 0 && input.height > 0 && input.channels == 3; |
| 26 | 5 | int valid_flag = valid ? 1 : 0; | |
| 27 | 5 | MPI_Bcast(&valid_flag, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 28 | return valid; | ||
| 29 | } | ||
| 30 | |||
| 31 | 5 | int valid_flag = 0; | |
| 32 | 5 | MPI_Bcast(&valid_flag, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 33 | 5 | return valid_flag == 1; | |
| 34 | } | ||
| 35 | |||
| 36 | 10 | bool SabirovSLinearFilteringBlockPartitioningMPI::PreProcessingImpl() { | |
| 37 | 10 | int rank = 0; | |
| 38 | 10 | int size = 0; | |
| 39 | 10 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 40 | 10 | MPI_Comm_size(MPI_COMM_WORLD, &size); | |
| 41 | |||
| 42 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 43 | const auto &input = GetInput(); | ||
| 44 | 5 | GetOutput() = ImageData(input.width, input.height, input.channels); | |
| 45 | |||
| 46 | // Рассылаем размеры изображения всем процессам | ||
| 47 | 5 | std::array<int, 3> dims = {input.width, input.height, input.channels}; | |
| 48 | 5 | MPI_Bcast(dims.data(), 3, MPI_INT, 0, MPI_COMM_WORLD); | |
| 49 | } else { | ||
| 50 | 5 | std::array<int, 3> dims = {0, 0, 0}; | |
| 51 | 5 | MPI_Bcast(dims.data(), 3, MPI_INT, 0, MPI_COMM_WORLD); | |
| 52 | 5 | GetInput() = ImageData(dims[0], dims[1], dims[2]); | |
| 53 | 10 | GetOutput() = ImageData(dims[0], dims[1], dims[2]); | |
| 54 | } | ||
| 55 | |||
| 56 | 10 | return !GetOutput().Empty(); | |
| 57 | } | ||
| 58 | |||
| 59 | namespace { | ||
| 60 | // Вспомогательная функция для фильтрации блока строк | ||
| 61 | 10 | void FilterRows(const ImageData &input, ImageData &output, int start_row, int end_row) { | |
| 62 | 10 | const int width = input.width; | |
| 63 | 10 | const int channels = input.channels; | |
| 64 | |||
| 65 |
2/2✓ Branch 0 taken 64 times.
✓ Branch 1 taken 10 times.
|
74 | for (int row = start_row; row < end_row; row++) { |
| 66 |
2/2✓ Branch 0 taken 1378 times.
✓ Branch 1 taken 64 times.
|
1442 | for (int col = 0; col < width; col++) { |
| 67 |
2/2✓ Branch 0 taken 4134 times.
✓ Branch 1 taken 1378 times.
|
5512 | for (int ch = 0; ch < channels; ch++) { |
| 68 | 4134 | float sum = ApplyGaussianKernel(input, row, col, ch); | |
| 69 | 4134 | auto output_idx = | |
| 70 | 4134 | ((static_cast<std::size_t>(row) * static_cast<std::size_t>(width) + static_cast<std::size_t>(col)) * | |
| 71 | 4134 | static_cast<std::size_t>(channels)) + | |
| 72 | 4134 | static_cast<std::size_t>(ch); | |
| 73 | 4134 | output.pixels[output_idx] = static_cast<uint8_t>(std::clamp(sum, 0.0F, 255.0F)); | |
| 74 | } | ||
| 75 | } | ||
| 76 | } | ||
| 77 | 10 | } | |
| 78 | |||
| 79 | // Вспомогательная функция для рассылки изображения | ||
| 80 | 10 | void BroadcastImage(ImageData &input, int rank, int size) { | |
| 81 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 82 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | for (int dest = 1; dest < size; dest++) { |
| 83 | 5 | MPI_Send(input.pixels.data(), static_cast<int>(input.pixels.size()), MPI_UNSIGNED_CHAR, dest, 0, MPI_COMM_WORLD); | |
| 84 | } | ||
| 85 | } else { | ||
| 86 | 5 | MPI_Recv(input.pixels.data(), static_cast<int>(input.pixels.size()), MPI_UNSIGNED_CHAR, 0, 0, MPI_COMM_WORLD, | |
| 87 | MPI_STATUS_IGNORE); | ||
| 88 | } | ||
| 89 | 10 | } | |
| 90 | |||
| 91 | // Вспомогательная функция для сбора результатов | ||
| 92 | 10 | void GatherResults(ImageData &output, int rank, int size, int rows_per_proc, int remainder, int width, int channels) { | |
| 93 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 94 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | for (int src = 1; src < size; src++) { |
| 95 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | int src_start = (src * rows_per_proc) + std::min(src, remainder); |
| 96 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | int src_end = src_start + rows_per_proc + (src < remainder ? 1 : 0); |
| 97 | 5 | int src_size = (src_end - src_start) * width * channels; | |
| 98 | |||
| 99 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | if (src_size > 0) { |
| 100 | 5 | auto offset = | |
| 101 | 5 | static_cast<std::size_t>(src_start) * static_cast<std::size_t>(width) * static_cast<std::size_t>(channels); | |
| 102 | 5 | MPI_Recv(output.pixels.data() + offset, src_size, MPI_UNSIGNED_CHAR, src, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE); | |
| 103 | } | ||
| 104 | } | ||
| 105 | } else { | ||
| 106 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | int start_row = (rank * rows_per_proc) + std::min(rank, remainder); |
| 107 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | int end_row = start_row + rows_per_proc + (rank < remainder ? 1 : 0); |
| 108 | 5 | int local_size = (end_row - start_row) * width * channels; | |
| 109 | |||
| 110 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | if (local_size > 0) { |
| 111 | 5 | auto offset = | |
| 112 | 5 | static_cast<std::size_t>(start_row) * static_cast<std::size_t>(width) * static_cast<std::size_t>(channels); | |
| 113 | 5 | MPI_Send(output.pixels.data() + offset, local_size, MPI_UNSIGNED_CHAR, 0, 1, MPI_COMM_WORLD); | |
| 114 | } | ||
| 115 | } | ||
| 116 | 10 | } | |
| 117 | } // namespace | ||
| 118 | |||
| 119 | 10 | bool SabirovSLinearFilteringBlockPartitioningMPI::RunImpl() { | |
| 120 | 10 | int rank = 0; | |
| 121 | 10 | int size = 0; | |
| 122 | 10 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 123 | 10 | MPI_Comm_size(MPI_COMM_WORLD, &size); | |
| 124 | |||
| 125 | auto &input = GetInput(); | ||
| 126 | auto &output = GetOutput(); | ||
| 127 | |||
| 128 |
2/4✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
|
10 | if (input.width < 3 || input.height < 3) { |
| 129 | return false; | ||
| 130 | } | ||
| 131 | |||
| 132 | const int width = input.width; | ||
| 133 | const int height = input.height; | ||
| 134 | 10 | const int channels = input.channels; | |
| 135 | |||
| 136 | // Рассылаем входное изображение от процесса 0 всем процессам | ||
| 137 | 10 | BroadcastImage(input, rank, size); | |
| 138 | |||
| 139 | // Вычисляем блочное разбиение (распределяем строки между процессами) | ||
| 140 | 10 | int rows_per_proc = height / size; | |
| 141 | 10 | int remainder = height % size; | |
| 142 | |||
| 143 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 | int start_row = (rank * rows_per_proc) + std::min(rank, remainder); |
| 144 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 | int end_row = start_row + rows_per_proc + (rank < remainder ? 1 : 0); |
| 145 | |||
| 146 | // Каждый процесс фильтрует свои назначенные строки | ||
| 147 | 10 | FilterRows(input, output, start_row, end_row); | |
| 148 | |||
| 149 | // Собираем результаты на процессе 0 | ||
| 150 | 10 | GatherResults(output, rank, size, rows_per_proc, remainder, width, channels); | |
| 151 | |||
| 152 | // Рассылаем полный результат от процесса 0 всем процессам | ||
| 153 | 10 | MPI_Bcast(output.pixels.data(), static_cast<int>(output.pixels.size()), MPI_UNSIGNED_CHAR, 0, MPI_COMM_WORLD); | |
| 154 | |||
| 155 | return true; | ||
| 156 | } | ||
| 157 | |||
| 158 | 10 | bool SabirovSLinearFilteringBlockPartitioningMPI::PostProcessingImpl() { | |
| 159 | 10 | return !GetOutput().Empty(); | |
| 160 | } | ||
| 161 | |||
| 162 | } // namespace sabirov_s_linear_filtering_block_partitioning | ||
| 163 |