| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "ermakov_a_spar_mat_mult/stl/include/ops_stl.hpp" | ||
| 2 | |||
| 3 | #include <algorithm> | ||
| 4 | #include <complex> | ||
| 5 | #include <cstddef> | ||
| 6 | #include <numeric> | ||
| 7 | #include <thread> | ||
| 8 | #include <vector> | ||
| 9 | |||
| 10 | #include "ermakov_a_spar_mat_mult/common/include/common.hpp" | ||
| 11 | #include "util/include/util.hpp" | ||
| 12 | |||
| 13 | namespace ermakov_a_spar_mat_mult { | ||
| 14 | |||
| 15 | namespace { | ||
| 16 | |||
| 17 | constexpr std::complex<double> kZero{0.0, 0.0}; | ||
| 18 | constexpr std::size_t kMinWorkPerThread = 4096; | ||
| 19 | |||
| 20 | } // namespace | ||
| 21 | |||
| 22 |
1/2✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 | ErmakovASparMatMultSTL::ErmakovASparMatMultSTL(const InType &in) { |
| 23 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 24 | GetInput() = in; | ||
| 25 | 32 | } | |
| 26 | |||
| 27 | 64 | bool ErmakovASparMatMultSTL::ValidateMatrix(const MatrixCRS &m) { | |
| 28 |
2/4✓ Branch 0 taken 64 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 64 times.
|
64 | if (m.rows < 0 || m.cols < 0) { |
| 29 | return false; | ||
| 30 | } | ||
| 31 | |||
| 32 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 64 times.
|
64 | if (m.row_ptr.size() != static_cast<std::size_t>(m.rows) + 1) { |
| 33 | return false; | ||
| 34 | } | ||
| 35 | |||
| 36 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 64 times.
|
64 | if (m.values.size() != m.col_index.size()) { |
| 37 | return false; | ||
| 38 | } | ||
| 39 | |||
| 40 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 64 times.
|
64 | if (m.row_ptr.empty()) { |
| 41 | return false; | ||
| 42 | } | ||
| 43 | |||
| 44 | 64 | const int nnz = static_cast<int>(m.values.size()); | |
| 45 |
2/4✗ Branch 0 not taken.
✓ Branch 1 taken 64 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 64 times.
|
64 | if (m.row_ptr.front() != 0 || m.row_ptr.back() != nnz) { |
| 46 | return false; | ||
| 47 | } | ||
| 48 | |||
| 49 |
2/2✓ Branch 0 taken 1008 times.
✓ Branch 1 taken 64 times.
|
1072 | for (int i = 0; i < m.rows; ++i) { |
| 50 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 1008 times.
|
1008 | if (m.row_ptr[i] > m.row_ptr[i + 1]) { |
| 51 | return false; | ||
| 52 | } | ||
| 53 | } | ||
| 54 | |||
| 55 |
2/2✓ Branch 0 taken 11384 times.
✓ Branch 1 taken 64 times.
|
11448 | for (int k = 0; k < nnz; ++k) { |
| 56 |
2/4✗ Branch 0 not taken.
✓ Branch 1 taken 11384 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 11384 times.
|
11384 | if (m.col_index[k] < 0 || m.col_index[k] >= m.cols) { |
| 57 | return false; | ||
| 58 | } | ||
| 59 | } | ||
| 60 | |||
| 61 | return true; | ||
| 62 | } | ||
| 63 | |||
| 64 | 32 | bool ErmakovASparMatMultSTL::ValidationImpl() { | |
| 65 | 32 | const auto &a = GetInput().A; | |
| 66 | 32 | const auto &b = GetInput().B; | |
| 67 | |||
| 68 |
2/4✓ Branch 0 taken 32 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 32 times.
✗ Branch 3 not taken.
|
32 | return a.cols == b.rows && ValidateMatrix(a) && ValidateMatrix(b); |
| 69 | } | ||
| 70 | |||
| 71 | 32 | bool ErmakovASparMatMultSTL::PreProcessingImpl() { | |
| 72 | 32 | a_ = GetInput().A; | |
| 73 | 32 | b_ = GetInput().B; | |
| 74 | |||
| 75 | 32 | c_.rows = a_.rows; | |
| 76 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 32 times.
|
32 | c_.cols = b_.cols; |
| 77 | c_.values.clear(); | ||
| 78 | c_.col_index.clear(); | ||
| 79 | 32 | c_.row_ptr.assign(static_cast<std::size_t>(c_.rows) + 1, 0); | |
| 80 | |||
| 81 | 32 | return true; | |
| 82 | } | ||
| 83 | |||
| 84 | 32 | int ErmakovASparMatMultSTL::ResolveThreadCount(int row_count, std::size_t total_work) { | |
| 85 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 24 times.
|
32 | if (row_count <= 1 || total_work < kMinWorkPerThread) { |
| 86 | return 1; | ||
| 87 | } | ||
| 88 | |||
| 89 | 8 | const int hw_threads = std::max(1, ppc::util::GetNumThreads()); | |
| 90 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
|
8 | const std::size_t work_limited_threads = std::max<std::size_t>(1, total_work / kMinWorkPerThread); |
| 91 | 16 | return std::max(1, std::min({hw_threads, row_count, static_cast<int>(work_limited_threads)})); | |
| 92 | } | ||
| 93 | |||
| 94 | 32 | std::vector<int> ErmakovASparMatMultSTL::BuildRowCosts(const MatrixCRS &a, const MatrixCRS &b) { | |
| 95 | 32 | std::vector<int> row_costs(static_cast<std::size_t>(a.rows), 1); | |
| 96 | |||
| 97 |
2/2✓ Branch 0 taken 504 times.
✓ Branch 1 taken 32 times.
|
536 | for (int i = 0; i < a.rows; ++i) { |
| 98 | 504 | int cost = 0; | |
| 99 |
2/2✓ Branch 0 taken 5632 times.
✓ Branch 1 taken 504 times.
|
6136 | for (int ak = a.row_ptr[i]; ak < a.row_ptr[i + 1]; ++ak) { |
| 100 | 5632 | const int b_row = a.col_index[ak]; | |
| 101 | 5632 | cost += b.row_ptr[b_row + 1] - b.row_ptr[b_row]; | |
| 102 | } | ||
| 103 | 504 | row_costs[static_cast<std::size_t>(i)] = std::max(1, cost); | |
| 104 | } | ||
| 105 | |||
| 106 | 32 | return row_costs; | |
| 107 | } | ||
| 108 | |||
| 109 | 6 | std::vector<int> ErmakovASparMatMultSTL::BuildThreadBoundaries(const std::vector<int> &row_costs, int thread_count) { | |
| 110 | const int safe_thread_count = std::max(1, thread_count); | ||
| 111 |
1/2✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 | std::vector<int> boundaries(static_cast<std::size_t>(safe_thread_count) + 1, 0); |
| 112 |
1/2✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
|
6 | boundaries[static_cast<std::size_t>(safe_thread_count)] = static_cast<int>(row_costs.size()); |
| 113 | |||
| 114 |
2/4✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 6 times.
✗ Branch 3 not taken.
|
6 | if (safe_thread_count <= 1 || row_costs.empty()) { |
| 115 | return boundaries; | ||
| 116 | } | ||
| 117 | |||
| 118 | const std::size_t total_work = std::accumulate(row_costs.begin(), row_costs.end(), std::size_t{0}); | ||
| 119 | 6 | const std::size_t target_chunk = std::max<std::size_t>(1, total_work / static_cast<std::size_t>(safe_thread_count)); | |
| 120 | |||
| 121 | std::size_t accumulated = 0; | ||
| 122 | std::size_t next_target = target_chunk; | ||
| 123 | int boundary_index = 1; | ||
| 124 | const auto row_count = row_costs.size(); | ||
| 125 | |||
| 126 |
2/2✓ Branch 0 taken 112 times.
✓ Branch 1 taken 6 times.
|
118 | for (std::size_t row = 0; row < row_count && boundary_index < safe_thread_count; ++row) { |
| 127 | 112 | accumulated += static_cast<std::size_t>(row_costs[row]); | |
| 128 |
2/2✓ Branch 0 taken 10 times.
✓ Branch 1 taken 102 times.
|
112 | if (accumulated >= next_target) { |
| 129 | 10 | boundaries[static_cast<std::size_t>(boundary_index)] = static_cast<int>(row) + 1; | |
| 130 | 10 | ++boundary_index; | |
| 131 | 10 | next_target = target_chunk * static_cast<std::size_t>(boundary_index); | |
| 132 | } | ||
| 133 | } | ||
| 134 | |||
| 135 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
|
6 | for (; boundary_index < safe_thread_count; ++boundary_index) { |
| 136 | ✗ | boundaries[static_cast<std::size_t>(boundary_index)] = boundaries[static_cast<std::size_t>(boundary_index) - 1]; | |
| 137 | } | ||
| 138 | |||
| 139 | return boundaries; | ||
| 140 | } | ||
| 141 | |||
| 142 |
1/2✓ Branch 0 taken 42 times.
✗ Branch 1 not taken.
|
42 | void ErmakovASparMatMultSTL::ResetWorkspace(Workspace &workspace, int cols) { |
| 143 |
1/2✓ Branch 0 taken 42 times.
✗ Branch 1 not taken.
|
42 | if (workspace.accum.size() != static_cast<std::size_t>(cols)) { |
| 144 | 42 | workspace.accum.assign(static_cast<std::size_t>(cols), kZero); | |
| 145 |
1/2✗ Branch 1 not taken.
✓ Branch 2 taken 42 times.
|
42 | workspace.marks.assign(static_cast<std::size_t>(cols), -1); |
| 146 | workspace.touched_cols.clear(); | ||
| 147 | 42 | workspace.touched_cols.reserve(256); | |
| 148 | } | ||
| 149 | 42 | } | |
| 150 | |||
| 151 |
2/2✓ Branch 0 taken 406 times.
✓ Branch 1 taken 98 times.
|
504 | void ErmakovASparMatMultSTL::MultiplyRow(int row_index, Workspace &workspace, RowData &row_data) const { |
| 152 | workspace.touched_cols.clear(); | ||
| 153 | |||
| 154 |
2/2✓ Branch 0 taken 5632 times.
✓ Branch 1 taken 504 times.
|
6136 | for (int ak = a_.row_ptr[row_index]; ak < a_.row_ptr[row_index + 1]; ++ak) { |
| 155 | 5632 | const int b_row = a_.col_index[ak]; | |
| 156 | 5632 | const auto a_value = a_.values[ak]; | |
| 157 | |||
| 158 |
2/2✓ Branch 0 taken 108312 times.
✓ Branch 1 taken 5632 times.
|
113944 | for (int bk = b_.row_ptr[b_row]; bk < b_.row_ptr[b_row + 1]; ++bk) { |
| 159 |
2/2✓ Branch 0 taken 8640 times.
✓ Branch 1 taken 99672 times.
|
108312 | const int col = b_.col_index[bk]; |
| 160 | 108312 | const auto product = a_value * b_.values[bk]; | |
| 161 | |||
| 162 |
2/2✓ Branch 0 taken 8640 times.
✓ Branch 1 taken 99672 times.
|
108312 | if (workspace.marks[static_cast<std::size_t>(col)] != row_index) { |
| 163 |
1/2✓ Branch 0 taken 8640 times.
✗ Branch 1 not taken.
|
8640 | workspace.marks[static_cast<std::size_t>(col)] = row_index; |
| 164 | 8640 | workspace.accum[static_cast<std::size_t>(col)] = product; | |
| 165 |
1/2✓ Branch 0 taken 8640 times.
✗ Branch 1 not taken.
|
8640 | workspace.touched_cols.push_back(col); |
| 166 | } else { | ||
| 167 | workspace.accum[static_cast<std::size_t>(col)] += product; | ||
| 168 | } | ||
| 169 | } | ||
| 170 | } | ||
| 171 | |||
| 172 | std::ranges::sort(workspace.touched_cols); | ||
| 173 | |||
| 174 | row_data.cols.clear(); | ||
| 175 | row_data.vals.clear(); | ||
| 176 | 504 | row_data.cols.reserve(workspace.touched_cols.size()); | |
| 177 | 504 | row_data.vals.reserve(workspace.touched_cols.size()); | |
| 178 | |||
| 179 |
2/2✓ Branch 0 taken 8640 times.
✓ Branch 1 taken 504 times.
|
9144 | for (int col : workspace.touched_cols) { |
| 180 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 8640 times.
|
8640 | const auto &value = workspace.accum[static_cast<std::size_t>(col)]; |
| 181 | if (value != kZero) { | ||
| 182 | row_data.cols.push_back(col); | ||
| 183 | row_data.vals.push_back(value); | ||
| 184 | } | ||
| 185 | } | ||
| 186 | 504 | } | |
| 187 | |||
| 188 | 32 | void ErmakovASparMatMultSTL::FinalizeResult(const std::vector<RowData> &rows_data) { | |
| 189 | int total_nnz = 0; | ||
| 190 | |||
| 191 |
2/2✓ Branch 0 taken 504 times.
✓ Branch 1 taken 32 times.
|
536 | for (int i = 0; i < c_.rows; ++i) { |
| 192 | 504 | c_.row_ptr[static_cast<std::size_t>(i)] = total_nnz; | |
| 193 | 504 | total_nnz += static_cast<int>(rows_data[static_cast<std::size_t>(i)].vals.size()); | |
| 194 | } | ||
| 195 | |||
| 196 | 32 | c_.row_ptr[static_cast<std::size_t>(c_.rows)] = total_nnz; | |
| 197 | 32 | c_.values.resize(static_cast<std::size_t>(total_nnz)); | |
| 198 | 32 | c_.col_index.resize(static_cast<std::size_t>(total_nnz)); | |
| 199 | |||
| 200 |
2/2✓ Branch 0 taken 504 times.
✓ Branch 1 taken 32 times.
|
536 | for (int i = 0; i < c_.rows; ++i) { |
| 201 | 504 | const auto &row = rows_data[static_cast<std::size_t>(i)]; | |
| 202 | 504 | const auto offset = static_cast<std::size_t>(c_.row_ptr[static_cast<std::size_t>(i)]); | |
| 203 | |||
| 204 | std::ranges::copy(row.cols, c_.col_index.begin() + static_cast<std::ptrdiff_t>(offset)); | ||
| 205 | std::ranges::copy(row.vals, c_.values.begin() + static_cast<std::ptrdiff_t>(offset)); | ||
| 206 | } | ||
| 207 | 32 | } | |
| 208 | |||
| 209 | 32 | bool ErmakovASparMatMultSTL::RunImpl() { | |
| 210 |
1/2✓ Branch 0 taken 32 times.
✗ Branch 1 not taken.
|
32 | if (a_.cols != b_.rows) { |
| 211 | return false; | ||
| 212 | } | ||
| 213 | |||
| 214 | c_.values.clear(); | ||
| 215 | c_.col_index.clear(); | ||
| 216 | std::ranges::fill(c_.row_ptr, 0); | ||
| 217 | |||
| 218 |
2/4✓ Branch 0 taken 32 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 32 times.
|
32 | if (a_.rows == 0 || b_.cols == 0) { |
| 219 | return true; | ||
| 220 | } | ||
| 221 | |||
| 222 | 32 | const std::vector<int> row_costs = BuildRowCosts(a_, b_); | |
| 223 | const std::size_t total_work = std::accumulate(row_costs.begin(), row_costs.end(), std::size_t{0}); | ||
| 224 |
1/2✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 | const int thread_count = ResolveThreadCount(a_.rows, total_work); |
| 225 | |||
| 226 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<RowData> rows_data(static_cast<std::size_t>(a_.rows)); |
| 227 | |||
| 228 |
2/2✓ Branch 0 taken 26 times.
✓ Branch 1 taken 6 times.
|
32 | if (thread_count == 1) { |
| 229 | 26 | Workspace workspace; | |
| 230 |
1/2✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
|
26 | ResetWorkspace(workspace, b_.cols); |
| 231 |
2/2✓ Branch 0 taken 324 times.
✓ Branch 1 taken 26 times.
|
350 | for (int row = 0; row < a_.rows; ++row) { |
| 232 |
1/2✓ Branch 1 taken 324 times.
✗ Branch 2 not taken.
|
324 | MultiplyRow(row, workspace, rows_data[static_cast<std::size_t>(row)]); |
| 233 | } | ||
| 234 |
1/2✓ Branch 1 taken 26 times.
✗ Branch 2 not taken.
|
26 | FinalizeResult(rows_data); |
| 235 | return true; | ||
| 236 | 26 | } | |
| 237 | |||
| 238 |
1/2✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 | const std::vector<int> boundaries = BuildThreadBoundaries(row_costs, thread_count); |
| 239 | 6 | std::vector<std::thread> workers; | |
| 240 |
1/2✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 | workers.reserve(static_cast<std::size_t>(thread_count)); |
| 241 | |||
| 242 |
2/2✓ Branch 0 taken 16 times.
✓ Branch 1 taken 6 times.
|
22 | for (int thread_id = 0; thread_id < thread_count; ++thread_id) { |
| 243 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | workers.emplace_back([&, thread_id] { |
| 244 | 16 | Workspace workspace; | |
| 245 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | ResetWorkspace(workspace, b_.cols); |
| 246 | |||
| 247 | 16 | const int row_begin = boundaries[static_cast<std::size_t>(thread_id)]; | |
| 248 | 16 | const int row_end = boundaries[static_cast<std::size_t>(thread_id) + 1]; | |
| 249 | |||
| 250 |
2/2✓ Branch 0 taken 180 times.
✓ Branch 1 taken 16 times.
|
196 | for (int row = row_begin; row < row_end; ++row) { |
| 251 |
1/2✓ Branch 1 taken 180 times.
✗ Branch 2 not taken.
|
180 | MultiplyRow(row, workspace, rows_data[static_cast<std::size_t>(row)]); |
| 252 | } | ||
| 253 | 16 | }); | |
| 254 | } | ||
| 255 | |||
| 256 |
2/2✓ Branch 0 taken 16 times.
✓ Branch 1 taken 6 times.
|
22 | for (auto &worker : workers) { |
| 257 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | worker.join(); |
| 258 | } | ||
| 259 | |||
| 260 |
1/2✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 | FinalizeResult(rows_data); |
| 261 | return true; | ||
| 262 | 38 | } | |
| 263 | |||
| 264 | 32 | bool ErmakovASparMatMultSTL::PostProcessingImpl() { | |
| 265 | 32 | GetOutput() = c_; | |
| 266 | 32 | return true; | |
| 267 | } | ||
| 268 | |||
| 269 | } // namespace ermakov_a_spar_mat_mult | ||
| 270 |