GCC Code Coverage Report

Directory: ./
File: tasks/melnik_i_matrix_mult_ribbon/mpi/src/ops_mpi.cpp
Date: 2026-01-27 01:59:34
Exec Total Coverage
Lines: 121 125 96.8%
Functions: 11 12 91.7%
Branches: 81 130 62.3%
Line Branch Exec Source
1 #include "melnik_i_matrix_mult_ribbon/mpi/include/ops_mpi.hpp"
2
3 #include <mpi.h>
4
5 #include <algorithm>
6 #include <array>
7 #include <cstddef>
8 #include <functional>
9 #include <numeric>
10 #include <ranges>
11 #include <vector>
12
13 #include "melnik_i_matrix_mult_ribbon/common/include/common.hpp"
14
15 namespace melnik_i_matrix_mult_ribbon {
16
17
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MelnikIMatrixMultRibbonMPI::MelnikIMatrixMultRibbonMPI(const InType &in) {
18
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MPI_Comm_rank(MPI_COMM_WORLD, &proc_rank_);
19
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MPI_Comm_size(MPI_COMM_WORLD, &proc_num_);
20
21 SetTypeOfTask(GetStaticTypeOfTask());
22 18 GetOutput() = std::vector<std::vector<double>>();
23
24
2/4
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
 18 GetInput() = (proc_rank_ == 0) ? in : InType{};
25
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (proc_rank_ == 0) {
26
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 matrix_A_ = std::get<0>(GetInput());
27
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 matrix_B_ = std::get<1>(GetInput());
28 }
29 18 }
30
31 18 bool MelnikIMatrixMultRibbonMPI::ValidationImpl() {
32
3/4
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
✓ Branch 3 taken 9 times.
✗ Branch 4 not taken.
 18 bool is_valid = (proc_rank_ == 0) ? ValidateOnRoot() : true;
33
34 18 int valid_flag = is_valid ? 1 : 0;
35 18 MPI_Bcast(&valid_flag, 1, MPI_INT, 0, MPI_COMM_WORLD);
36
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 if (valid_flag == 0) {
37 ✗ rows_a_ = cols_a_ = rows_b_ = cols_b_ = 0;
38 ✗ return false;
39 }
40
41 18 std::array<int, 4> dims = {static_cast<int>(rows_a_), static_cast<int>(rows_b_), static_cast<int>(cols_a_),
42 18 static_cast<int>(cols_b_)};
43 18 MPI_Bcast(dims.data(), 4, MPI_INT, 0, MPI_COMM_WORLD);
44 18 rows_a_ = static_cast<std::size_t>(dims[0]);
45 18 rows_b_ = static_cast<std::size_t>(dims[1]);
46 18 cols_a_ = static_cast<std::size_t>(dims[2]);
47 18 cols_b_ = static_cast<std::size_t>(dims[3]);
48
49 18 return GetOutput().empty();
50 }
51
52
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 bool MelnikIMatrixMultRibbonMPI::PreProcessingImpl() {
53 GetOutput().clear();
54
55
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (proc_rank_ == 0) {
56 const auto &[matrix_a, matrix_b] = GetInput();
57 9 rows_a_ = matrix_a.size();
58 9 rows_b_ = matrix_b.size();
59 9 cols_a_ = matrix_a.front().size();
60
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 cols_b_ = matrix_b.front().size();
61
62 flat_a_.clear();
63 9 flat_a_.reserve(rows_a_ * cols_a_);
64
2/2
✓ Branch 0 taken 44 times.
✓ Branch 1 taken 9 times.
 53 for (const auto &row : matrix_a) {
65 44 flat_a_.insert(flat_a_.end(), row.begin(), row.end());
66 }
67
68 9 flat_b_transposed_.assign(cols_b_ * cols_a_, 0.0);
69
2/2
✓ Branch 0 taken 46 times.
✓ Branch 1 taken 9 times.
 55 for (std::size_t row_idx = 0; row_idx < cols_a_; ++row_idx) {
70
2/2
✓ Branch 0 taken 256 times.
✓ Branch 1 taken 46 times.
 302 for (std::size_t col_idx = 0; col_idx < cols_b_; ++col_idx) {
71 256 flat_b_transposed_[(col_idx * cols_a_) + row_idx] = matrix_b[row_idx][col_idx];
72 }
73 }
74 }
75 18 return true;
76 }
77
78 18 bool MelnikIMatrixMultRibbonMPI::RunImpl() {
79
3/6
✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 18 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 18 times.
✗ Branch 5 not taken.
 18 if (rows_a_ == 0 || cols_a_ == 0 || cols_b_ == 0) {
80 return false;
81 }
82
83 18 ShareSizes();
84 18 ShareMatrixB();
85
86 18 std::vector<double> local_a;
87 18 std::vector<int> rows_per_rank;
88
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 const std::size_t local_rows = ScatterRows(local_a, rows_per_rank);
89
90
3/6
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 16 times.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
 18 std::vector<double> local_c(local_rows * cols_b_, 0.0);
91
3/4
✓ Branch 0 taken 16 times.
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 16 times.
✗ Branch 3 not taken.
 18 if (!local_a.empty() && local_rows > 0) {
92 16 MultiplyLocal(local_a, local_c);
93 }
94
95
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 GatherAndDistribute(rows_per_rank, local_c);
96 return true;
97 }
98
99
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 bool MelnikIMatrixMultRibbonMPI::PostProcessingImpl() {
100 auto &output = GetOutput();
101 output.clear();
102 18 output.resize(rows_a_);
103
2/2
✓ Branch 0 taken 88 times.
✓ Branch 1 taken 18 times.
 106 for (std::size_t i = 0; i < rows_a_; ++i) {
104 88 output[i].assign(cols_b_, 0.0);
105 }
106
107
1/2
✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
 18 const std::size_t total = rows_a_ * cols_b_;
108
1/2
✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
 18 if (flat_c_.size() >= total) {
109
2/2
✓ Branch 0 taken 508 times.
✓ Branch 1 taken 18 times.
 526 for (std::size_t idx = 0; idx < total; ++idx) {
110 508 const std::size_t r = idx / cols_b_;
111 508 const std::size_t c = idx % cols_b_;
112 508 output[r][c] = flat_c_[idx];
113 }
114 }
115
116 18 return true;
117 }
118
119
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 9 bool MelnikIMatrixMultRibbonMPI::ValidateOnRoot() {
120 const auto &matrix_a = std::get<0>(GetInput());
121 const auto &matrix_b = std::get<1>(GetInput());
122
123
2/4
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 9 times.
✗ Branch 3 not taken.
 9 if (matrix_a.empty() || matrix_b.empty()) {
124 return false;
125 }
126
127 const std::size_t width_a = matrix_a.front().size();
128 const std::size_t width_b = matrix_b.front().size();
129 const std::size_t height_b = matrix_b.size();
130
131
2/4
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 9 times.
✗ Branch 3 not taken.
 9 if (width_a == 0 || width_b == 0 || width_a != height_b) {
132 return false;
133 }
134
135
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 9 if (!HasUniformRowWidth(matrix_a, width_a)) {
136 return false;
137 }
138
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 9 if (!HasUniformRowWidth(matrix_b, width_b)) {
139 return false;
140 }
141
142 9 rows_a_ = matrix_a.size();
143 9 cols_a_ = width_a;
144 9 rows_b_ = height_b;
145 9 cols_b_ = width_b;
146 9 return true;
147 }
148
149 ✗ bool MelnikIMatrixMultRibbonMPI::HasUniformRowWidth(const std::vector<std::vector<double>> &matrix,
150 std::size_t expected_width) {
151 return std::ranges::all_of(matrix,
152 ✗ [expected_width](const std::vector<double> &row) { return row.size() == expected_width; });
153 }
154
155 18 void MelnikIMatrixMultRibbonMPI::ShareSizes() {
156 18 std::array<int, 4> sizes = {static_cast<int>(rows_a_), static_cast<int>(rows_b_), static_cast<int>(cols_a_),
157 18 static_cast<int>(cols_b_)};
158 18 MPI_Bcast(sizes.data(), 4, MPI_INT, 0, MPI_COMM_WORLD);
159 18 rows_a_ = static_cast<std::size_t>(sizes[0]);
160 18 rows_b_ = static_cast<std::size_t>(sizes[1]);
161 18 cols_a_ = static_cast<std::size_t>(sizes[2]);
162 18 cols_b_ = static_cast<std::size_t>(sizes[3]);
163 18 }
164
165 18 void MelnikIMatrixMultRibbonMPI::ShareMatrixB() {
166 18 const std::size_t total = cols_b_ * cols_a_;
167
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (proc_rank_ != 0) {
168 9 flat_b_transposed_.assign(total, 0.0);
169 }
170 18 MPI_Bcast(flat_b_transposed_.data(), static_cast<int>(total), MPI_DOUBLE, 0, MPI_COMM_WORLD);
171 18 }
172
173 18 std::size_t MelnikIMatrixMultRibbonMPI::ScatterRows(std::vector<double> &local_a, std::vector<int> &rows_per_rank) {
174 18 rows_per_rank.resize(proc_num_);
175 18 const int base = static_cast<int>(rows_a_ / static_cast<std::size_t>(proc_num_));
176 18 const int remainder = static_cast<int>(rows_a_ % static_cast<std::size_t>(proc_num_));
177
2/2
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
 54 for (int i = 0; i < proc_num_; ++i) {
178
2/2
✓ Branch 0 taken 32 times.
✓ Branch 1 taken 4 times.
 68 rows_per_rank[i] = base + (i < remainder ? 1 : 0);
179 }
180
181 18 std::vector<int> counts(proc_num_);
182
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<int> displs(proc_num_);
183
2/2
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
 54 for (int i = 0, offset = 0; i < proc_num_; ++i) {
184 36 counts[i] = rows_per_rank[i] * static_cast<int>(cols_a_);
185 36 displs[i] = offset;
186 36 offset += counts[i];
187 }
188
189
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 const int local_count = rows_per_rank[proc_rank_] * static_cast<int>(cols_a_);
190
3/6
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 9 times.
✓ Branch 4 taken 9 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
 18 local_a.assign(static_cast<std::size_t>(local_count), 0.0);
191
192
3/4
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
✓ Branch 3 taken 18 times.
✗ Branch 4 not taken.
 27 MPI_Scatterv(proc_rank_ == 0 ? flat_a_.data() : nullptr, counts.data(), displs.data(), MPI_DOUBLE, local_a.data(),
193 local_count, MPI_DOUBLE, 0, MPI_COMM_WORLD);
194
195
1/2
✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
 36 return static_cast<std::size_t>(rows_per_rank[proc_rank_]);
196 }
197
198 16 void MelnikIMatrixMultRibbonMPI::MultiplyLocal(const std::vector<double> &local_a, std::vector<double> &local_c) const {
199 using DiffT = std::ptrdiff_t;
200
2/2
✓ Branch 0 taken 44 times.
✓ Branch 1 taken 16 times.
 60 for (std::size_t row_idx = 0; row_idx < local_c.size() / cols_b_; ++row_idx) {
201 44 const auto a_begin = local_a.begin() + static_cast<DiffT>(row_idx * cols_a_);
202 const auto a_end = a_begin + static_cast<DiffT>(cols_a_);
203
2/2
✓ Branch 0 taken 254 times.
✓ Branch 1 taken 44 times.
 298 for (std::size_t col_idx = 0; col_idx < cols_b_; ++col_idx) {
204 254 const auto b_begin = flat_b_transposed_.begin() + static_cast<DiffT>(col_idx * cols_a_);
205 254 local_c[(row_idx * cols_b_) + col_idx] =
206 254 std::transform_reduce(a_begin, a_end, b_begin, 0.0, std::plus<>(), std::multiplies<>());
207 }
208 }
209 16 }
210
211 18 void MelnikIMatrixMultRibbonMPI::GatherAndDistribute(const std::vector<int> &rows_per_rank,
212 const std::vector<double> &local_c) {
213 18 std::vector<int> counts(proc_num_);
214
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<int> displs(proc_num_);
215
2/2
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
 54 for (int i = 0, offset = 0; i < proc_num_; ++i) {
216 36 counts[i] = rows_per_rank[i] * static_cast<int>(cols_b_);
217 36 displs[i] = offset;
218 36 offset += counts[i];
219 }
220
221
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (proc_rank_ == 0) {
222
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 flat_c_.assign(rows_a_ * cols_b_, 0.0);
223 } else {
224 flat_c_.clear();
225
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 flat_c_.shrink_to_fit();
226
1/4
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 9 flat_c_.assign(rows_a_ * cols_b_, 0.0);
227 }
228
229
3/4
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
✓ Branch 3 taken 18 times.
✗ Branch 4 not taken.
 27 MPI_Gatherv(local_c.data(), counts[proc_rank_], MPI_DOUBLE, proc_rank_ == 0 ? flat_c_.data() : nullptr, counts.data(),
230 displs.data(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
231
232 // Broadcast assembled result so PostProcessing can run uniformly
233
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MPI_Bcast(flat_c_.data(), static_cast<int>(rows_a_ * cols_b_), MPI_DOUBLE, 0, MPI_COMM_WORLD);
234 18 }
235
236 } // namespace melnik_i_matrix_mult_ribbon
237