| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "papulina_y_simple_iteration/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <cctype> | ||
| 7 | #include <cmath> | ||
| 8 | #include <cstddef> | ||
| 9 | #include <cstdlib> | ||
| 10 | #include <iostream> | ||
| 11 | #include <utility> | ||
| 12 | #include <vector> | ||
| 13 | |||
| 14 | #include "papulina_y_simple_iteration/common/include/common.hpp" | ||
| 15 | |||
| 16 | namespace papulina_y_simple_iteration { | ||
| 17 | |||
| 18 |
1/2✓ Branch 1 taken 36 times.
✗ Branch 2 not taken.
|
36 | PapulinaYSimpleIterationMPI::PapulinaYSimpleIterationMPI(const InType &in) { |
| 19 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 20 | GetInput() = in; | ||
| 21 | ✗ | GetOutput() = std::vector<double>(0); | |
| 22 |
1/2✓ Branch 1 taken 36 times.
✗ Branch 2 not taken.
|
36 | MPI_Comm_size(MPI_COMM_WORLD, &procNum_); |
| 23 | 36 | } | |
| 24 | 17 | bool PapulinaYSimpleIterationMPI::DiagonalDominance(const std::vector<double> &a, const size_t &n) { | |
| 25 | bool flag = true; | ||
| 26 |
2/2✓ Branch 0 taken 69 times.
✓ Branch 1 taken 16 times.
|
85 | for (size_t i = 0; i < n; i++) { |
| 27 | double sum = 0.0; | ||
| 28 |
2/2✓ Branch 0 taken 357 times.
✓ Branch 1 taken 69 times.
|
426 | for (size_t j = 0; j < n; j++) { |
| 29 |
2/2✓ Branch 0 taken 288 times.
✓ Branch 1 taken 69 times.
|
357 | if (j != i) { |
| 30 | 288 | sum += abs(a[(i * n) + j]); | |
| 31 | } | ||
| 32 | } | ||
| 33 |
2/2✓ Branch 0 taken 68 times.
✓ Branch 1 taken 1 times.
|
69 | if (sum > abs(a[(i * n) + i])) { |
| 34 | flag = false; | ||
| 35 | break; | ||
| 36 | } | ||
| 37 | } | ||
| 38 | 17 | return flag; | |
| 39 | } | ||
| 40 | 18 | bool PapulinaYSimpleIterationMPI::DetermChecking(const std::vector<double> &a, const size_t &n) { | |
| 41 | 18 | std::vector<double> tmp = a; | |
| 42 | |||
| 43 |
2/2✓ Branch 0 taken 72 times.
✓ Branch 1 taken 17 times.
|
89 | for (size_t i = 0; i < n; i++) { |
| 44 |
2/2✓ Branch 0 taken 2 times.
✓ Branch 1 taken 70 times.
|
72 | if (std::fabs(tmp[(i * n) + i]) < 1e-10) { |
| 45 |
2/2✓ Branch 1 taken 1 times.
✓ Branch 2 taken 1 times.
|
2 | if (!FindAndSwapRow(tmp, i, n)) { |
| 46 | return false; | ||
| 47 | } | ||
| 48 | } | ||
| 49 | 71 | double pivot = tmp[(i * n) + i]; | |
| 50 |
2/2✓ Branch 0 taken 218 times.
✓ Branch 1 taken 71 times.
|
289 | for (size_t j = i; j < n; j++) { |
| 51 | 218 | tmp[(i * n) + j] /= pivot; | |
| 52 | } | ||
| 53 |
2/2✓ Branch 0 taken 147 times.
✓ Branch 1 taken 71 times.
|
218 | for (size_t j = i + 1; j < n; j++) { |
| 54 | 147 | double factor = tmp[(j * n) + i]; | |
| 55 |
2/2✓ Branch 0 taken 746 times.
✓ Branch 1 taken 147 times.
|
893 | for (size_t k = i; k < n; k++) { |
| 56 | 746 | tmp[(j * n) + k] -= tmp[(i * n) + k] * factor; | |
| 57 | } | ||
| 58 | } | ||
| 59 | } | ||
| 60 | |||
| 61 | return true; | ||
| 62 | } | ||
| 63 | 2 | bool PapulinaYSimpleIterationMPI::FindAndSwapRow(std::vector<double> &tmp, size_t i, size_t n) { | |
| 64 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 1 times.
|
4 | for (size_t j = i + 1; j < n; j++) { |
| 65 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 2 times.
|
3 | if (std::fabs(tmp[(j * n) + i]) > 1e-10) { |
| 66 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 1 times.
|
4 | for (size_t k = i; k < n; k++) { |
| 67 | 3 | std::swap(tmp[(i * n) + k], tmp[(j * n) + k]); | |
| 68 | } | ||
| 69 | return true; | ||
| 70 | } | ||
| 71 | } | ||
| 72 | return false; | ||
| 73 | } | ||
| 74 | 36 | bool PapulinaYSimpleIterationMPI::ValidationImpl() { | |
| 75 | 36 | int flag = 1; | |
| 76 | 36 | int proc_rank = 0; | |
| 77 | 36 | MPI_Comm_rank(MPI_COMM_WORLD, &proc_rank); | |
| 78 | |||
| 79 |
2/2✓ Branch 0 taken 18 times.
✓ Branch 1 taken 18 times.
|
36 | if (proc_rank == 0) { |
| 80 |
1/2✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
|
18 | size_t n = std::get<0>(GetInput()); |
| 81 | const auto &a_matrix = std::get<1>(GetInput()); | ||
| 82 | |||
| 83 |
5/6✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
✓ Branch 3 taken 17 times.
✓ Branch 4 taken 1 times.
✓ Branch 5 taken 1 times.
✓ Branch 6 taken 16 times.
|
18 | if ((n < 1) || (!DetermChecking(a_matrix, n) || (!DiagonalDominance(a_matrix, n)))) { |
| 84 | 2 | flag = 0; // валидация не прошла | |
| 85 | } else { | ||
| 86 | 16 | double norm_b = CalculateNormB(a_matrix, n); | |
| 87 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 16 times.
|
16 | if (norm_b >= 1.0) { |
| 88 | ✗ | std::cout << "WARNING: sufficient condition for convergence may not hold (norm_b = " << norm_b << " >= 1)\n"; | |
| 89 | } | ||
| 90 | } | ||
| 91 | } | ||
| 92 | |||
| 93 | 36 | MPI_Bcast(&flag, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 94 | |||
| 95 | 36 | return (flag == 1); | |
| 96 | } | ||
| 97 | 16 | double PapulinaYSimpleIterationMPI::CalculateNormB(const std::vector<double> &a, size_t n) { | |
| 98 | 16 | double max_row_sum = 0.0; | |
| 99 |
2/2✓ Branch 0 taken 67 times.
✓ Branch 1 taken 16 times.
|
83 | for (size_t i = 0; i < n; i++) { |
| 100 | 67 | double diag = a[(i * n) + i]; | |
| 101 | 67 | double row_sum = 0.0; | |
| 102 | |||
| 103 |
2/2✓ Branch 0 taken 351 times.
✓ Branch 1 taken 67 times.
|
418 | for (size_t j = 0; j < n; j++) { |
| 104 |
2/2✓ Branch 0 taken 284 times.
✓ Branch 1 taken 67 times.
|
351 | if (j != i) { |
| 105 | 284 | row_sum += std::abs(a[(i * n) + j] / diag); | |
| 106 | } | ||
| 107 | } | ||
| 108 | 67 | max_row_sum = std::max(row_sum, max_row_sum); | |
| 109 | } | ||
| 110 | |||
| 111 | 16 | return max_row_sum; | |
| 112 | } | ||
| 113 | 32 | bool PapulinaYSimpleIterationMPI::PreProcessingImpl() { | |
| 114 | 32 | int proc_rank = 0; | |
| 115 | 32 | MPI_Comm_rank(MPI_COMM_WORLD, &proc_rank); | |
| 116 | |||
| 117 |
2/2✓ Branch 0 taken 16 times.
✓ Branch 1 taken 16 times.
|
32 | if (proc_rank == 0) { |
| 118 | 16 | n_ = static_cast<size_t>(std::get<0>(GetInput())); | |
| 119 | 16 | A_.assign(n_ * n_, 0.0); | |
| 120 |
2/2✓ Branch 1 taken 15 times.
✓ Branch 2 taken 1 times.
|
16 | b_.assign(n_, 0.0); |
| 121 | |||
| 122 |
2/2✓ Branch 0 taken 15 times.
✓ Branch 1 taken 1 times.
|
16 | std::copy(std::get<1>(GetInput()).data(), std::get<1>(GetInput()).data() + (n_ * n_), A_.data()); |
| 123 |
2/2✓ Branch 0 taken 15 times.
✓ Branch 1 taken 1 times.
|
16 | std::copy(std::get<2>(GetInput()).data(), std::get<2>(GetInput()).data() + n_, b_.data()); |
| 124 | } | ||
| 125 | |||
| 126 | 32 | MPI_Bcast(&n_, 1, MPI_UNSIGNED_LONG, 0, MPI_COMM_WORLD); | |
| 127 | 32 | int rows_for_proc = static_cast<int>(n_) / procNum_; | |
| 128 | 32 | int remainder = static_cast<int>(n_) % procNum_; | |
| 129 | |||
| 130 |
2/2✓ Branch 0 taken 25 times.
✓ Branch 1 taken 7 times.
|
32 | int start_row = (proc_rank * rows_for_proc) + std::min(proc_rank, remainder); |
| 131 |
2/2✓ Branch 0 taken 25 times.
✓ Branch 1 taken 7 times.
|
32 | int last_row = start_row + rows_for_proc + (proc_rank < remainder ? 1 : 0); |
| 132 | 32 | int local_rows_count = last_row - start_row; | |
| 133 | |||
| 134 | 32 | local_a_.resize(local_rows_count * n_); | |
| 135 | 32 | local_b_.resize(local_rows_count); | |
| 136 | |||
| 137 | 32 | std::vector<int> proc_count_elements_a(procNum_, 0); | |
| 138 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<int> proc_count_elements_b(procNum_, 0); |
| 139 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<int> a_displs(procNum_, 0); |
| 140 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<int> b_displs(procNum_, 0); |
| 141 | |||
| 142 |
2/2✓ Branch 0 taken 64 times.
✓ Branch 1 taken 32 times.
|
96 | for (int i = 0; i < procNum_; i++) { |
| 143 |
2/2✓ Branch 0 taken 50 times.
✓ Branch 1 taken 14 times.
|
64 | int start = (i * rows_for_proc) + std::min(i, remainder); |
| 144 |
2/2✓ Branch 0 taken 50 times.
✓ Branch 1 taken 14 times.
|
64 | int end = start + rows_for_proc + (i < remainder ? 1 : 0); |
| 145 | 64 | int count = end - start; | |
| 146 | |||
| 147 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 32 times.
|
64 | proc_count_elements_a[i] = count * static_cast<int>(n_); |
| 148 | 64 | proc_count_elements_b[i] = count; | |
| 149 | |||
| 150 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 32 times.
|
64 | if (i > 0) { |
| 151 | 32 | a_displs[i] = a_displs[i - 1] + proc_count_elements_a[i - 1]; | |
| 152 | 32 | b_displs[i] = b_displs[i - 1] + proc_count_elements_b[i - 1]; | |
| 153 | } | ||
| 154 | } | ||
| 155 | |||
| 156 |
1/2✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 | MPI_Scatterv((proc_rank == 0) ? A_.data() : nullptr, proc_count_elements_a.data(), a_displs.data(), MPI_DOUBLE, |
| 157 |
2/2✓ Branch 0 taken 16 times.
✓ Branch 1 taken 16 times.
|
32 | local_a_.data(), local_rows_count * static_cast<int>(n_), MPI_DOUBLE, 0, MPI_COMM_WORLD); |
| 158 | |||
| 159 |
3/4✓ Branch 0 taken 16 times.
✓ Branch 1 taken 16 times.
✓ Branch 3 taken 32 times.
✗ Branch 4 not taken.
|
48 | MPI_Scatterv((proc_rank == 0) ? b_.data() : nullptr, proc_count_elements_b.data(), b_displs.data(), MPI_DOUBLE, |
| 160 | local_b_.data(), local_rows_count, MPI_DOUBLE, 0, MPI_COMM_WORLD); | ||
| 161 | |||
| 162 | 32 | return true; | |
| 163 | } | ||
| 164 | |||
| 165 | 32 | bool PapulinaYSimpleIterationMPI::RunImpl() { | |
| 166 | 32 | int proc_rank = 0; | |
| 167 | 32 | MPI_Comm_rank(MPI_COMM_WORLD, &proc_rank); | |
| 168 | 32 | int rows_for_proc = static_cast<int>(n_) / procNum_; | |
| 169 | 32 | int remainder = static_cast<int>(n_) % procNum_; | |
| 170 | |||
| 171 |
2/2✓ Branch 0 taken 25 times.
✓ Branch 1 taken 7 times.
|
32 | int start_row = (proc_rank * rows_for_proc) + std::min(proc_rank, remainder); |
| 172 |
2/2✓ Branch 0 taken 25 times.
✓ Branch 1 taken 7 times.
|
32 | int last_row = start_row + rows_for_proc + (proc_rank < remainder ? 1 : 0); |
| 173 | 32 | int local_rows_count = last_row - start_row; | |
| 174 | 32 | std::vector<double> local_b_matrix(local_rows_count * n_, 0); | |
| 175 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<double> local_d(local_rows_count, 0); |
| 176 | |||
| 177 | 32 | PrepareLocalMatrices(local_b_matrix, local_d, start_row, local_rows_count, static_cast<int>(n_)); | |
| 178 | |||
| 179 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<double> x(n_, 0.0); |
| 180 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<double> x_new(n_, 0.0); |
| 181 | |||
| 182 | 32 | double diff = 0.0; | |
| 183 | |||
| 184 |
1/2✓ Branch 0 taken 526 times.
✗ Branch 1 not taken.
|
526 | for (unsigned int step = 0; step < steps_count_; step++) { |
| 185 |
1/4✓ Branch 1 taken 526 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
526 | std::vector<double> local_x_new(local_rows_count, 0.0); |
| 186 | 526 | for (size_t i = 0; std::cmp_less(i, local_rows_count); i++) { | |
| 187 |
2/2✓ Branch 0 taken 8845 times.
✓ Branch 1 taken 1379 times.
|
10224 | for (size_t j = 0; j < n_; j++) { |
| 188 | 8845 | local_x_new[i] += local_b_matrix[(i * n_) + j] * x[j]; | |
| 189 | } | ||
| 190 | 1379 | local_x_new[i] += local_d[i]; | |
| 191 | } | ||
| 192 | |||
| 193 |
1/4✓ Branch 1 taken 526 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
526 | std::vector<int> proc_count_elemts_x(procNum_, 0); |
| 194 |
1/4✓ Branch 1 taken 526 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
526 | std::vector<int> x_displs(procNum_, 0); |
| 195 | 526 | CalculateGatherParameters(proc_count_elemts_x, x_displs, rows_for_proc, remainder); | |
| 196 | // MPI_Gatherv(local_x_new.data(), local_rows_count, MPI_DOUBLE, x_new.data(), proc_count_elemts_x.data(), | ||
| 197 | // x_displs.data(), MPI_DOUBLE, 0, MPI_COMM_WORLD); MPI_Bcast(x_new.data(), static_cast<int>(n_), MPI_DOUBLE, 0, | ||
| 198 | // MPI_COMM_WORLD); | ||
| 199 |
1/2✓ Branch 1 taken 526 times.
✗ Branch 2 not taken.
|
526 | MPI_Allgatherv(local_x_new.data(), local_rows_count, MPI_DOUBLE, x_new.data(), proc_count_elemts_x.data(), |
| 200 | x_displs.data(), MPI_DOUBLE, MPI_COMM_WORLD); | ||
| 201 | 526 | double local_sum_for_norm = 0.0; | |
| 202 |
2/2✓ Branch 0 taken 1379 times.
✓ Branch 1 taken 526 times.
|
1905 | for (int i = 0; i < local_rows_count; i++) { |
| 203 | 1379 | int global_i = start_row + i; | |
| 204 | 1379 | double diff_i = x_new[global_i] - x[global_i]; | |
| 205 | 1379 | local_sum_for_norm += diff_i * diff_i; | |
| 206 | } | ||
| 207 | |||
| 208 |
1/2✓ Branch 1 taken 526 times.
✗ Branch 2 not taken.
|
526 | MPI_Allreduce(&local_sum_for_norm, &diff, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD); |
| 209 | 526 | diff = std::sqrt(diff); | |
| 210 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 494 times.
|
526 | if (diff < eps_) { |
| 211 |
1/2✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 | x = x_new; |
| 212 | break; | ||
| 213 | } | ||
| 214 |
1/2✓ Branch 1 taken 494 times.
✗ Branch 2 not taken.
|
494 | x = x_new; |
| 215 | } | ||
| 216 |
1/2✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 | GetOutput() = x; |
| 217 | 32 | return true; | |
| 218 | } | ||
| 219 | 32 | bool PapulinaYSimpleIterationMPI::PostProcessingImpl() { | |
| 220 | 32 | return true; | |
| 221 | } | ||
| 222 | 526 | void PapulinaYSimpleIterationMPI::CalculateGatherParameters(std::vector<int> &proc_count_elemts_x, | |
| 223 | std::vector<int> &x_displs, int rows_for_proc, | ||
| 224 | int remainder) const { | ||
| 225 | // if (proc_rank != 0) { | ||
| 226 | // return; | ||
| 227 | // } | ||
| 228 |
2/2✓ Branch 0 taken 1052 times.
✓ Branch 1 taken 526 times.
|
1578 | for (int i = 0; i < procNum_; i++) { |
| 229 |
2/2✓ Branch 0 taken 842 times.
✓ Branch 1 taken 210 times.
|
1052 | unsigned int start = (i * rows_for_proc) + std::min(i, remainder); |
| 230 |
2/2✓ Branch 0 taken 842 times.
✓ Branch 1 taken 210 times.
|
1052 | unsigned int end = start + rows_for_proc + (i < remainder ? 1 : 0); |
| 231 | 1052 | unsigned int count = end - start; | |
| 232 |
2/2✓ Branch 0 taken 526 times.
✓ Branch 1 taken 526 times.
|
1052 | proc_count_elemts_x[i] = static_cast<int>(count); |
| 233 | |||
| 234 |
2/2✓ Branch 0 taken 526 times.
✓ Branch 1 taken 526 times.
|
1052 | if (i > 0) { |
| 235 | 526 | x_displs[i] = x_displs[i - 1] + proc_count_elemts_x[i - 1]; | |
| 236 | } | ||
| 237 | } | ||
| 238 | 526 | } | |
| 239 | 32 | void PapulinaYSimpleIterationMPI::PrepareLocalMatrices(std::vector<double> &local_b_matrix, | |
| 240 | std::vector<double> &local_d, int start_row, | ||
| 241 | int local_rows_count, int n) { | ||
| 242 |
2/2✓ Branch 0 taken 67 times.
✓ Branch 1 taken 32 times.
|
99 | for (int i = 0; i < local_rows_count; i++) { |
| 243 | 67 | int global_i = start_row + i; | |
| 244 | 67 | double diag = local_a_[(i * n) + global_i]; | |
| 245 | 67 | double inv_diag = 1.0 / diag; | |
| 246 |
2/2✓ Branch 0 taken 351 times.
✓ Branch 1 taken 67 times.
|
418 | for (int j = 0; j < n; j++) { |
| 247 |
2/2✓ Branch 0 taken 284 times.
✓ Branch 1 taken 67 times.
|
351 | if (j != global_i) { |
| 248 | 284 | local_b_matrix[(i * n) + j] = -local_a_[(i * n) + j] * inv_diag; | |
| 249 | } | ||
| 250 | } | ||
| 251 | 67 | local_d[i] = local_b_[i] * inv_diag; | |
| 252 | } | ||
| 253 | 32 | } | |
| 254 | } // namespace papulina_y_simple_iteration | ||
| 255 |