| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "agafonov_i_matrix_ccs_seq/omp/include/ops_omp.hpp" | ||
| 2 | |||
| 3 | #include <omp.h> | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <cmath> | ||
| 7 | #include <cstddef> | ||
| 8 | #include <vector> | ||
| 9 | |||
| 10 | #include "agafonov_i_matrix_ccs_seq/common/include/common.hpp" | ||
| 11 | #include "util/include/util.hpp" | ||
| 12 | |||
| 13 | namespace agafonov_i_matrix_ccs_seq { | ||
| 14 | |||
| 15 | ✗ | AgafonovIMatrixCCSOMP::AgafonovIMatrixCCSOMP(const InType &in) { | |
| 16 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 17 | GetInput() = in; | ||
| 18 | ✗ | } | |
| 19 | |||
| 20 | ✗ | bool AgafonovIMatrixCCSOMP::ValidationImpl() { | |
| 21 | const auto &left = GetInput().first; | ||
| 22 | const auto &right = GetInput().second; | ||
| 23 | |||
| 24 | ✗ | return (left.cols_num == right.rows_num) && (left.col_ptrs.size() == left.cols_num + 1) && | |
| 25 | ✗ | (right.col_ptrs.size() == right.cols_num + 1); | |
| 26 | } | ||
| 27 | |||
| 28 | ✗ | bool AgafonovIMatrixCCSOMP::PreProcessingImpl() { | |
| 29 | GetOutput().vals.clear(); | ||
| 30 | GetOutput().row_inds.clear(); | ||
| 31 | ✗ | return true; | |
| 32 | } | ||
| 33 | |||
| 34 | ✗ | void AgafonovIMatrixCCSOMP::ProcessColumn(size_t j, const InType::first_type &a, const InType::second_type &b, | |
| 35 | std::vector<double> &accumulator, std::vector<size_t> &active_rows, | ||
| 36 | std::vector<bool> &row_mask, std::vector<double> &local_v, | ||
| 37 | std::vector<int> &local_r) { | ||
| 38 | ✗ | const auto b_col_start = static_cast<size_t>(b.col_ptrs[j]); | |
| 39 | ✗ | const auto b_col_end = static_cast<size_t>(b.col_ptrs[j + 1]); | |
| 40 | |||
| 41 | ✗ | if (b_col_start == b_col_end) { | |
| 42 | return; | ||
| 43 | } | ||
| 44 | |||
| 45 | ✗ | for (size_t kb = b_col_start; kb < b_col_end; ++kb) { | |
| 46 | ✗ | const auto k = static_cast<size_t>(b.row_inds[kb]); | |
| 47 | ✗ | const double v_b = b.vals[kb]; | |
| 48 | ✗ | const auto a_col_start = static_cast<size_t>(a.col_ptrs[k]); | |
| 49 | ✗ | const auto a_col_end = static_cast<size_t>(a.col_ptrs[k + 1]); | |
| 50 | |||
| 51 | ✗ | for (size_t ka = a_col_start; ka < a_col_end; ++ka) { | |
| 52 | ✗ | const auto i = static_cast<size_t>(a.row_inds[ka]); | |
| 53 | ✗ | if (!row_mask[i]) { | |
| 54 | row_mask[i] = true; | ||
| 55 | active_rows.push_back(i); | ||
| 56 | } | ||
| 57 | ✗ | accumulator[i] += a.vals[ka] * v_b; | |
| 58 | } | ||
| 59 | } | ||
| 60 | |||
| 61 | std::ranges::sort(active_rows); | ||
| 62 | |||
| 63 | ✗ | for (const auto row_idx : active_rows) { | |
| 64 | ✗ | if (std::abs(accumulator[row_idx]) > 1e-15) { | |
| 65 | local_v.push_back(accumulator[row_idx]); | ||
| 66 | ✗ | local_r.push_back(static_cast<int>(row_idx)); | |
| 67 | } | ||
| 68 | row_mask[row_idx] = false; | ||
| 69 | ✗ | accumulator[row_idx] = 0.0; | |
| 70 | } | ||
| 71 | active_rows.clear(); | ||
| 72 | } | ||
| 73 | |||
| 74 | // Обновленный RunImpl | ||
| 75 | ✗ | bool AgafonovIMatrixCCSOMP::RunImpl() { | |
| 76 | ✗ | const auto &a = GetInput().first; | |
| 77 | ✗ | const auto &b = GetInput().second; | |
| 78 | auto &c = GetOutput(); | ||
| 79 | |||
| 80 | ✗ | c.rows_num = a.rows_num; | |
| 81 | ✗ | c.cols_num = b.cols_num; | |
| 82 | ✗ | c.col_ptrs.assign(c.cols_num + 1, 0); | |
| 83 | |||
| 84 | ✗ | std::vector<std::vector<double>> local_vals(b.cols_num); | |
| 85 | ✗ | std::vector<std::vector<int>> local_rows(b.cols_num); | |
| 86 | |||
| 87 | ✗ | #pragma omp parallel num_threads(ppc::util::GetNumThreads()) default(none) \ | |
| 88 | shared(a, b, local_vals, local_rows, std::ranges::sort) | ||
| 89 | { | ||
| 90 | std::vector<double> accumulator(a.rows_num, 0.0); | ||
| 91 | std::vector<size_t> active_rows; | ||
| 92 | std::vector<bool> row_mask(a.rows_num, false); | ||
| 93 | |||
| 94 | #pragma omp for | ||
| 95 | for (size_t j = 0; j < b.cols_num; ++j) { | ||
| 96 | ProcessColumn(j, a, b, accumulator, active_rows, row_mask, local_vals[j], local_rows[j]); | ||
| 97 | } | ||
| 98 | } | ||
| 99 | |||
| 100 | // ... (дальше код без изменений до конца функции) | ||
| 101 | size_t total_nnz = 0; | ||
| 102 | ✗ | for (const auto &v : local_vals) { | |
| 103 | ✗ | total_nnz += v.size(); | |
| 104 | } | ||
| 105 | ✗ | c.vals.reserve(total_nnz); | |
| 106 | ✗ | c.row_inds.reserve(total_nnz); | |
| 107 | |||
| 108 | int current_nnz = 0; | ||
| 109 | ✗ | for (size_t j = 0; j < b.cols_num; ++j) { | |
| 110 | ✗ | c.col_ptrs[j] = current_nnz; | |
| 111 | ✗ | c.vals.insert(c.vals.end(), local_vals[j].begin(), local_vals[j].end()); | |
| 112 | ✗ | c.row_inds.insert(c.row_inds.end(), local_rows[j].begin(), local_rows[j].end()); | |
| 113 | ✗ | current_nnz += static_cast<int>(local_vals[j].size()); | |
| 114 | } | ||
| 115 | ✗ | c.col_ptrs[b.cols_num] = current_nnz; | |
| 116 | ✗ | c.nnz = current_nnz; | |
| 117 | ✗ | return true; | |
| 118 | ✗ | } | |
| 119 | |||
| 120 | ✗ | bool AgafonovIMatrixCCSOMP::PostProcessingImpl() { | |
| 121 | ✗ | return true; | |
| 122 | } | ||
| 123 | |||
| 124 | } // namespace agafonov_i_matrix_ccs_seq | ||
| 125 |