| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "posternak_a_crs_mul_complex_matrix/omp/include/ops_omp.hpp" | ||
| 2 | |||
| 3 | #include <algorithm> | ||
| 4 | #include <cmath> | ||
| 5 | #include <complex> | ||
| 6 | #include <cstddef> | ||
| 7 | #include <unordered_map> | ||
| 8 | #include <utility> | ||
| 9 | #include <vector> | ||
| 10 | |||
| 11 | #include "posternak_a_crs_mul_complex_matrix/common/include/common.hpp" | ||
| 12 | |||
| 13 | namespace { | ||
| 14 | |||
| 15 | 40 | size_t ComputeRowNoZeroCount(const posternak_a_crs_mul_complex_matrix::CRSMatrix &a, | |
| 16 | const posternak_a_crs_mul_complex_matrix::CRSMatrix &b, int row, double threshold) { | ||
| 17 | std::unordered_map<int, std::complex<double>> row_sum; | ||
| 18 | |||
| 19 |
2/2✓ Branch 0 taken 48 times.
✓ Branch 1 taken 40 times.
|
88 | for (int idx_a = a.index_row[row]; idx_a < a.index_row[row + 1]; ++idx_a) { |
| 20 | 48 | int col_a = a.index_col[idx_a]; | |
| 21 | 48 | auto val_a = a.values[idx_a]; | |
| 22 | |||
| 23 |
2/2✓ Branch 0 taken 68 times.
✓ Branch 1 taken 48 times.
|
116 | for (int idx_b = b.index_row[col_a]; idx_b < b.index_row[col_a + 1]; ++idx_b) { |
| 24 |
1/2✓ Branch 1 taken 68 times.
✗ Branch 2 not taken.
|
68 | int col_b = b.index_col[idx_b]; |
| 25 |
1/2✓ Branch 1 taken 68 times.
✗ Branch 2 not taken.
|
68 | auto val_b = b.values[idx_b]; |
| 26 | row_sum[col_b] += val_a * val_b; | ||
| 27 | } | ||
| 28 | } | ||
| 29 | |||
| 30 | size_t local = 0; | ||
| 31 |
2/2✓ Branch 0 taken 56 times.
✓ Branch 1 taken 40 times.
|
96 | for (const auto &[col, val] : row_sum) { |
| 32 |
1/2✓ Branch 0 taken 56 times.
✗ Branch 1 not taken.
|
56 | if (std::abs(val) > threshold) { |
| 33 | 56 | ++local; | |
| 34 | } | ||
| 35 | } | ||
| 36 | 40 | return local; | |
| 37 | } | ||
| 38 | |||
| 39 | 16 | void BuildResultStructure(posternak_a_crs_mul_complex_matrix::CRSMatrix &res, std::vector<size_t> &row_prefix) { | |
| 40 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 16 times.
|
40 | for (int i = 1; i < res.rows; ++i) { |
| 41 | 24 | row_prefix[i] += row_prefix[i - 1]; | |
| 42 | } | ||
| 43 | |||
| 44 |
1/2✓ Branch 0 taken 16 times.
✗ Branch 1 not taken.
|
16 | const size_t total = row_prefix.empty() ? 0 : row_prefix.back(); |
| 45 | 16 | res.values.resize(total); | |
| 46 | 16 | res.index_col.resize(total); | |
| 47 | 16 | res.index_row.resize(res.rows + 1); | |
| 48 | |||
| 49 |
2/2✓ Branch 0 taken 56 times.
✓ Branch 1 taken 16 times.
|
72 | for (int i = 0; i <= res.rows; ++i) { |
| 50 |
2/2✓ Branch 0 taken 40 times.
✓ Branch 1 taken 16 times.
|
56 | res.index_row[i] = (i == 0 ? 0 : static_cast<int>(row_prefix[i - 1])); |
| 51 | } | ||
| 52 | 16 | } | |
| 53 | |||
| 54 | 40 | void ComputeAndWriteRow(const posternak_a_crs_mul_complex_matrix::CRSMatrix &a, | |
| 55 | const posternak_a_crs_mul_complex_matrix::CRSMatrix &b, | ||
| 56 | posternak_a_crs_mul_complex_matrix::CRSMatrix &res, int row, double threshold) { | ||
| 57 | std::unordered_map<int, std::complex<double>> row_sum; | ||
| 58 | |||
| 59 |
2/2✓ Branch 0 taken 48 times.
✓ Branch 1 taken 40 times.
|
88 | for (int idx_a = a.index_row[row]; idx_a < a.index_row[row + 1]; ++idx_a) { |
| 60 | 48 | int col_a = a.index_col[idx_a]; | |
| 61 | 48 | auto val_a = a.values[idx_a]; | |
| 62 | |||
| 63 |
2/2✓ Branch 0 taken 68 times.
✓ Branch 1 taken 48 times.
|
116 | for (int idx_b = b.index_row[col_a]; idx_b < b.index_row[col_a + 1]; ++idx_b) { |
| 64 |
1/2✓ Branch 1 taken 68 times.
✗ Branch 2 not taken.
|
68 | int col_b = b.index_col[idx_b]; |
| 65 |
1/2✓ Branch 1 taken 68 times.
✗ Branch 2 not taken.
|
68 | auto val_b = b.values[idx_b]; |
| 66 | row_sum[col_b] += val_a * val_b; | ||
| 67 | } | ||
| 68 | } | ||
| 69 | |||
| 70 |
1/2✓ Branch 1 taken 40 times.
✗ Branch 2 not taken.
|
40 | std::vector<std::pair<int, std::complex<double>>> sorted(row_sum.begin(), row_sum.end()); |
| 71 | |||
| 72 |
1/22✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
✓ Branch 12 taken 16 times.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
✗ Branch 15 not taken.
✗ Branch 16 not taken.
✗ Branch 17 not taken.
✗ Branch 18 not taken.
✗ Branch 19 not taken.
✗ Branch 20 not taken.
✗ Branch 21 not taken.
|
16 | std::ranges::sort(sorted, [](const auto &p1, const auto &p2) { return p1.first < p2.first; }); |
| 73 | |||
| 74 | 40 | size_t pos = res.index_row[row]; | |
| 75 |
2/2✓ Branch 0 taken 56 times.
✓ Branch 1 taken 40 times.
|
96 | for (const auto &[col_idx, value] : sorted) { |
| 76 |
1/2✓ Branch 0 taken 56 times.
✗ Branch 1 not taken.
|
56 | if (std::abs(value) > threshold) { |
| 77 | 56 | res.values[pos] = value; | |
| 78 | 56 | res.index_col[pos] = col_idx; | |
| 79 | 56 | ++pos; | |
| 80 | } | ||
| 81 | } | ||
| 82 | 40 | } | |
| 83 | |||
| 84 | } // namespace | ||
| 85 | |||
| 86 | namespace posternak_a_crs_mul_complex_matrix { | ||
| 87 | |||
| 88 |
1/2✓ Branch 2 taken 24 times.
✗ Branch 3 not taken.
|
24 | PosternakACRSMulComplexMatrixOMP::PosternakACRSMulComplexMatrixOMP(const InType &in) { |
| 89 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 90 | GetInput() = in; | ||
| 91 | 24 | GetOutput() = CRSMatrix{}; | |
| 92 | 24 | } | |
| 93 | |||
| 94 | 24 | bool PosternakACRSMulComplexMatrixOMP::ValidationImpl() { | |
| 95 | const auto &input = GetInput(); | ||
| 96 | 24 | const auto &a = input.first; | |
| 97 | 24 | const auto &b = input.second; | |
| 98 |
3/6✓ Branch 0 taken 24 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 24 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 24 times.
|
24 | return a.IsValid() && b.IsValid() && a.cols == b.rows; |
| 99 | } | ||
| 100 | |||
| 101 | 24 | bool PosternakACRSMulComplexMatrixOMP::PreProcessingImpl() { | |
| 102 | const auto &input = GetInput(); | ||
| 103 | const auto &a = input.first; | ||
| 104 | const auto &b = input.second; | ||
| 105 | auto &res = GetOutput(); | ||
| 106 | |||
| 107 | 24 | res.rows = a.rows; | |
| 108 | 24 | res.cols = b.cols; | |
| 109 | 24 | return true; | |
| 110 | } | ||
| 111 | |||
| 112 | 24 | bool PosternakACRSMulComplexMatrixOMP::RunImpl() { | |
| 113 | const auto &input = GetInput(); | ||
| 114 | 24 | const auto &a = input.first; | |
| 115 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 4 times.
|
24 | const auto &b = input.second; |
| 116 | auto &res = GetOutput(); | ||
| 117 | |||
| 118 |
4/4✓ Branch 0 taken 20 times.
✓ Branch 1 taken 4 times.
✓ Branch 2 taken 4 times.
✓ Branch 3 taken 16 times.
|
24 | if (a.values.empty() || b.values.empty()) { |
| 119 | res.values.clear(); | ||
| 120 | res.index_col.clear(); | ||
| 121 | 8 | res.index_row.assign(res.rows + 1, 0); | |
| 122 | 8 | return true; | |
| 123 | } | ||
| 124 | |||
| 125 | constexpr double kThreshold = 1e-12; | ||
| 126 | 16 | std::vector<size_t> no_zero_rows(res.rows, 0); | |
| 127 | |||
| 128 | // каждый поток определяет количество ненулевых элементов в своих строках | ||
| 129 |
1/2✓ Branch 0 taken 16 times.
✗ Branch 1 not taken.
|
16 | #pragma omp parallel for default(none) shared(a, b, res, no_zero_rows) schedule(dynamic) |
| 130 | for (int row = 0; row < res.rows; ++row) { | ||
| 131 | no_zero_rows[row] = ComputeRowNoZeroCount(a, b, row, kThreshold); | ||
| 132 | } | ||
| 133 | |||
| 134 | // структурируем результат, чтобы избежать конфликта потоков (предотвращаем гонку данных) | ||
| 135 | #pragma omp single | ||
| 136 | { | ||
| 137 | 16 | BuildResultStructure(res, no_zero_rows); | |
| 138 | } | ||
| 139 | |||
| 140 | // записываем результат в итоговый массив параллельно | ||
| 141 | 16 | #pragma omp parallel for default(none) shared(a, b, res) schedule(dynamic) | |
| 142 | for (int row = 0; row < res.rows; ++row) { | ||
| 143 | ComputeAndWriteRow(a, b, res, row, kThreshold); | ||
| 144 | } | ||
| 145 | |||
| 146 |
1/2✓ Branch 0 taken 16 times.
✗ Branch 1 not taken.
|
16 | return res.IsValid(); |
| 147 | } | ||
| 148 | |||
| 149 | 24 | bool PosternakACRSMulComplexMatrixOMP::PostProcessingImpl() { | |
| 150 | 24 | return GetOutput().IsValid(); | |
| 151 | } | ||
| 152 | |||
| 153 | } // namespace posternak_a_crs_mul_complex_matrix | ||
| 154 |