| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "ovchinnikov_m_max_values_in_matrix_rows/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <cstddef> | ||
| 7 | #include <limits> | ||
| 8 | #include <vector> | ||
| 9 | |||
| 10 | #include "ovchinnikov_m_max_values_in_matrix_rows/common/include/common.hpp" | ||
| 11 | |||
| 12 | namespace ovchinnikov_m_max_values_in_matrix_rows { | ||
| 13 | |||
| 14 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | OvchinnikovMMaxValuesInMatrixRowsMPI::OvchinnikovMMaxValuesInMatrixRowsMPI(const InType &in) { |
| 15 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 16 | GetInput() = in; | ||
| 17 | static_cast<void>(GetOutput()); | ||
| 18 | 12 | } | |
| 19 | |||
| 20 | 12 | bool OvchinnikovMMaxValuesInMatrixRowsMPI::ValidationImpl() { | |
| 21 | 12 | return true; | |
| 22 | } | ||
| 23 | |||
| 24 | 12 | bool OvchinnikovMMaxValuesInMatrixRowsMPI::PreProcessingImpl() { | |
| 25 | 12 | return true; | |
| 26 | } | ||
| 27 | |||
| 28 | 12 | bool OvchinnikovMMaxValuesInMatrixRowsMPI::RunImpl() { | |
| 29 | 12 | int tmp_rank = 0; | |
| 30 | 12 | int tmp_proc_amount = 0; | |
| 31 | 12 | MPI_Comm_rank(MPI_COMM_WORLD, &tmp_rank); | |
| 32 | 12 | MPI_Comm_size(MPI_COMM_WORLD, &tmp_proc_amount); | |
| 33 | 12 | auto rank = static_cast<size_t>(tmp_rank); | |
| 34 | 12 | auto proc_amount = static_cast<size_t>(tmp_proc_amount); | |
| 35 | |||
| 36 | 12 | const size_t lines = std::get<0>(GetInput()); | |
| 37 | 12 | const size_t cols = std::get<1>(GetInput()); | |
| 38 |
2/2✓ Branch 0 taken 10 times.
✓ Branch 1 taken 2 times.
|
12 | if (lines == 0 || cols == 0) { |
| 39 | return true; | ||
| 40 | } | ||
| 41 | const auto &matrix = std::get<2>(GetInput()); | ||
| 42 | |||
| 43 | 10 | std::vector<int> elem_count(proc_amount); | |
| 44 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<int> elem_offset(proc_amount); |
| 45 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 46 | 5 | const size_t chunk_base = lines / proc_amount; | |
| 47 | 5 | const size_t chunk_extra = lines % proc_amount; | |
| 48 |
2/2✓ Branch 0 taken 10 times.
✓ Branch 1 taken 5 times.
|
15 | for (size_t process = 0; process < proc_amount; process++) { |
| 49 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 7 times.
|
10 | const size_t line_begin = (process * chunk_base) + std::min(process, chunk_extra); |
| 50 | 10 | size_t line_end = line_begin + chunk_base; | |
| 51 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 7 times.
|
10 | if (process < chunk_extra) { |
| 52 | 3 | line_end++; | |
| 53 | } | ||
| 54 | |||
| 55 | 10 | size_t elems = (line_end - line_begin) * cols; | |
| 56 | 10 | size_t offset = line_begin * cols; | |
| 57 | 10 | elem_count[process] = static_cast<int>(elems); | |
| 58 | 10 | elem_offset[process] = static_cast<int>(offset); | |
| 59 | } | ||
| 60 | } | ||
| 61 | |||
| 62 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | MPI_Bcast(elem_count.data(), static_cast<int>(proc_amount), MPI_INT, 0, MPI_COMM_WORLD); |
| 63 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | MPI_Bcast(elem_offset.data(), static_cast<int>(proc_amount), MPI_INT, 0, MPI_COMM_WORLD); |
| 64 | |||
| 65 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<int> local_data(elem_count[rank]); |
| 66 | |||
| 67 | const int *matrix_buffer = nullptr; | ||
| 68 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 69 | matrix_buffer = matrix.data(); | ||
| 70 | } | ||
| 71 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | MPI_Scatterv(matrix_buffer, // only for rank 0 |
| 72 | elem_count.data(), elem_offset.data(), MPI_INT, local_data.data(), elem_count[rank], MPI_INT, 0, | ||
| 73 | MPI_COMM_WORLD); | ||
| 74 | |||
| 75 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<int> local_max(cols, std::numeric_limits<int>::min()); |
| 76 | |||
| 77 | 10 | size_t local_lines = local_data.size() / cols; | |
| 78 | |||
| 79 |
2/2✓ Branch 0 taken 13 times.
✓ Branch 1 taken 10 times.
|
23 | for (size_t i = 0; i < local_lines; i++) { |
| 80 | 13 | const size_t local_base = i * cols; | |
| 81 |
2/2✓ Branch 0 taken 65 times.
✓ Branch 1 taken 13 times.
|
78 | for (size_t j = 0; j < cols; j++) { |
| 82 |
2/2✓ Branch 0 taken 44 times.
✓ Branch 1 taken 21 times.
|
109 | local_max[j] = std::max(local_max[j], local_data[local_base + j]); |
| 83 | } | ||
| 84 | } | ||
| 85 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<int> global_max(cols); |
| 86 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | MPI_Allreduce(local_max.data(), global_max.data(), static_cast<int>(cols), MPI_INT, MPI_MAX, MPI_COMM_WORLD); |
| 87 | |||
| 88 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | GetOutput() = global_max; |
| 89 | return true; | ||
| 90 | } | ||
| 91 | |||
| 92 | 12 | bool OvchinnikovMMaxValuesInMatrixRowsMPI::PostProcessingImpl() { | |
| 93 | 12 | return true; | |
| 94 | } | ||
| 95 | |||
| 96 | } // namespace ovchinnikov_m_max_values_in_matrix_rows | ||
| 97 |