GCC Code Coverage Report

Directory: ./
File: tasks/smyshlaev_a_mat_mul/mpi/src/ops_mpi.cpp
Date: 2026-01-10 02:40:41
Exec Total Coverage
Lines: 139 141 98.6%
Functions: 9 9 100.0%
Branches: 94 148 63.5%
Line Branch Exec Source
1 #include "smyshlaev_a_mat_mul/mpi/include/ops_mpi.hpp"
2
3 #include <mpi.h>
4
5 #include <algorithm>
6 #include <cstddef>
7 #include <vector>
8
9 #include "smyshlaev_a_mat_mul/common/include/common.hpp"
10
11 namespace smyshlaev_a_mat_mul {
12
13 namespace {
14
15 void CalculateDistribution(int total_len, int proc_count, std::vector<int> &counts, std::vector<int> &offsets) {
16 36 const int chunk = total_len / proc_count;
17 36 const int remainder = total_len % proc_count;
18 int offset = 0;
19
4/4
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
✓ Branch 2 taken 36 times.
✓ Branch 3 taken 18 times.
 108 for (int i = 0; i < proc_count; i++) {
20
4/4
✓ Branch 0 taken 30 times.
✓ Branch 1 taken 6 times.
✓ Branch 2 taken 30 times.
✓ Branch 3 taken 6 times.
 132 counts[i] = chunk + (i < remainder ? 1 : 0);
21 72 offsets[i] = offset;
22 72 offset += counts[i];
23 }
24 }
25 } // namespace
26
27
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
 24 SmyshlaevAMatMulMPI::SmyshlaevAMatMulMPI(const InType &in) {
28 SetTypeOfTask(GetStaticTypeOfTask());
29
30 24 int rank = 0;
31
1/2
✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
 24 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
32
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
 24 if (rank == 0) {
33 GetInput() = in;
34 }
35 24 }
36
37 24 bool SmyshlaevAMatMulMPI::ValidationImpl() {
38 24 int rank = 0;
39 24 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
40 24 int error_flag = 0;
41
42
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
 24 if (rank == 0) {
43 const auto &num_rows_a = std::get<0>(GetInput());
44 const auto &mat_a = std::get<1>(GetInput());
45 const auto &num_rows_b = std::get<2>(GetInput());
46 const auto &mat_b = std::get<3>(GetInput());
47
48
3/6
✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 12 times.
✗ Branch 5 not taken.
 12 bool is_invalid = (num_rows_a <= 0 || num_rows_b <= 0) || (mat_a.empty() || mat_b.empty()) ||
49
3/6
✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✓ Branch 5 taken 12 times.
 24 (mat_a.size() % num_rows_a != 0) || (mat_b.size() % num_rows_b != 0);
50
51 if (is_invalid) {
52 ✗ error_flag = 1;
53 } else {
54 12 const auto &num_cols_a = static_cast<int>(mat_a.size()) / num_rows_a;
55
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 12 times.
 12 if (num_cols_a != num_rows_b) {
56 ✗ error_flag = 1;
57 }
58 }
59 }
60 24 MPI_Bcast(&error_flag, 1, MPI_INT, 0, MPI_COMM_WORLD);
61 24 return (error_flag == 0);
62 }
63
64 24 bool SmyshlaevAMatMulMPI::PreProcessingImpl() {
65 24 int rank = 0;
66 24 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
67
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
 24 if (rank == 0) {
68 const auto &num_rows_b = std::get<2>(GetInput());
69 const auto &mat_b = std::get<3>(GetInput());
70 12 const auto num_cols_b = static_cast<int>(mat_b.size()) / num_rows_b;
71
72 12 mat_b_transposed_.resize(mat_b.size());
73
74
2/2
✓ Branch 0 taken 81 times.
✓ Branch 1 taken 12 times.
 93 for (int i = 0; i < num_rows_b; ++i) {
75
2/2
✓ Branch 0 taken 1220 times.
✓ Branch 1 taken 81 times.
 1301 for (int j = 0; j < num_cols_b; ++j) {
76 1220 mat_b_transposed_[(j * num_rows_b) + i] = mat_b[(i * num_cols_b) + j];
77 }
78 }
79 }
80 24 return true;
81 }
82 6 bool SmyshlaevAMatMulMPI::RunSequential() {
83 6 int rank = 0;
84 6 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
85 6 std::vector<double> result;
86 6 int num_elms = 0;
87
2/2
✓ Branch 0 taken 3 times.
✓ Branch 1 taken 3 times.
 6 if (rank == 0) {
88 const auto &num_rows_a = std::get<0>(GetInput());
89 const auto &mat_a = std::get<1>(GetInput());
90 const auto &num_rows_b = std::get<2>(GetInput());
91 const auto &mat_b = std::get<3>(GetInput());
92 3 const int num_cols_b = static_cast<int>(mat_b.size()) / num_rows_b;
93 const int num_cols_a = num_rows_b;
94
95 3 num_elms = num_rows_a * num_cols_b;
96
1/4
✓ Branch 1 taken 3 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 3 result.resize(num_elms, 0.0);
97
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 3 times.
 15 for (int i = 0; i < num_rows_a; ++i) {
98
2/2
✓ Branch 0 taken 16 times.
✓ Branch 1 taken 12 times.
 28 for (int j = 0; j < num_cols_b; ++j) {
99 double sum = 0.0;
100
2/2
✓ Branch 0 taken 126 times.
✓ Branch 1 taken 16 times.
 142 for (int k = 0; k < num_cols_a; ++k) {
101 126 sum += mat_a[(i * num_cols_a) + k] * mat_b_transposed_[(j * num_cols_a) + k];
102 }
103 16 result[(i * num_cols_b) + j] = sum;
104 }
105 }
106 }
107
1/2
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
 6 MPI_Bcast(&num_elms, 1, MPI_INT, 0, MPI_COMM_WORLD);
108
109
2/2
✓ Branch 0 taken 3 times.
✓ Branch 1 taken 3 times.
 6 if (rank != 0) {
110
1/2
✓ Branch 1 taken 3 times.
✗ Branch 2 not taken.
 3 result.resize(num_elms);
111 }
112
113
1/2
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
 6 MPI_Bcast(result.data(), num_elms, MPI_DOUBLE, 0, MPI_COMM_WORLD);
114
1/2
✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
 6 GetOutput() = result;
115
116 6 return true;
117 }
118
119 24 void SmyshlaevAMatMulMPI::BroadcastDimensions(int &rows_a, int &cols_a, int &cols_b) {
120 24 int rank = 0;
121 24 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
122
2/2
✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
 24 if (rank == 0) {
123 12 rows_a = std::get<0>(GetInput());
124 const auto &mat_a = std::get<1>(GetInput());
125 12 cols_a = static_cast<int>(mat_a.size()) / rows_a;
126 const auto &num_rows_b = std::get<2>(GetInput());
127 const auto &mat_b = std::get<3>(GetInput());
128 12 cols_b = static_cast<int>(mat_b.size()) / num_rows_b;
129 }
130 24 MPI_Bcast(&rows_a, 1, MPI_INT, 0, MPI_COMM_WORLD);
131 24 MPI_Bcast(&cols_a, 1, MPI_INT, 0, MPI_COMM_WORLD);
132 24 MPI_Bcast(&cols_b, 1, MPI_INT, 0, MPI_COMM_WORLD);
133 24 }
134
135 18 void SmyshlaevAMatMulMPI::RingShiftAlgorithm(int rank, int size, int my_rows_a, int num_cols_a, int num_cols_b,
136 const std::vector<int> &counts_b, const std::vector<int> &disps_b,
137 std::vector<double> &local_a, std::vector<double> &local_b,
138 std::vector<double> &local_c) {
139 18 int max_elem_count_b = 0;
140
2/2
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
 54 for (int c : counts_b) {
141 36 max_elem_count_b = std::max(max_elem_count_b, c);
142 }
143 18 std::vector<double> local_b_next(max_elem_count_b);
144
145 18 int left_neighbor = (rank - 1 + size) % size;
146 18 int right_neighbor = (rank + 1) % size;
147
148
2/2
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
 54 for (int step = 0; step < size; ++step) {
149 36 int b_owner_rank = (rank - step + size) % size;
150 36 int current_cols_b_count = counts_b[b_owner_rank] / num_cols_a;
151 36 int global_col_shift = disps_b[b_owner_rank] / num_cols_a;
152
153
2/2
✓ Branch 0 taken 130 times.
✓ Branch 1 taken 36 times.
 166 for (int i = 0; i < my_rows_a; ++i) {
154
2/2
✓ Branch 0 taken 1208 times.
✓ Branch 1 taken 130 times.
 1338 for (int j = 0; j < current_cols_b_count; ++j) {
155 double sum = 0.0;
156
2/2
✓ Branch 0 taken 33984 times.
✓ Branch 1 taken 1208 times.
 35192 for (int k = 0; k < num_cols_a; ++k) {
157 33984 sum += local_a[(i * num_cols_a) + k] * local_b[(j * num_cols_a) + k];
158 }
159 1208 local_c[(i * num_cols_b) + (global_col_shift + j)] = sum;
160 }
161 }
162
163 int send_count = counts_b[b_owner_rank];
164 36 int recv_owner_rank = (rank - (step + 1) + size) % size;
165
1/2
✓ Branch 1 taken 36 times.
✗ Branch 2 not taken.
 36 int recv_count = counts_b[recv_owner_rank];
166
167
1/2
✓ Branch 1 taken 36 times.
✗ Branch 2 not taken.
 36 MPI_Sendrecv(local_b.data(), send_count, MPI_DOUBLE, right_neighbor, 0, local_b_next.data(), recv_count, MPI_DOUBLE,
168 left_neighbor, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
169
170 36 local_b.assign(local_b_next.begin(), local_b_next.begin() + recv_count);
171 }
172 18 }
173
174 18 void SmyshlaevAMatMulMPI::GatherAndBroadcastResults(int rank, int size, int rows_a, int cols_a, int cols_b,
175 const std::vector<int> &counts_a,
176 const std::vector<double> &local_c) {
177 18 std::vector<double> final_res;
178
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<int> recvcounts_c(size);
179
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<int> disps_c(size);
180
181
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (rank == 0) {
182
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 final_res.resize(static_cast<size_t>(rows_a) * cols_b);
183 int offset = 0;
184
2/2
✓ Branch 0 taken 18 times.
✓ Branch 1 taken 9 times.
 27 for (int i = 0; i < size; ++i) {
185 18 int r_rows = counts_a[i] / cols_a;
186 18 recvcounts_c[i] = r_rows * cols_b;
187 18 disps_c[i] = offset;
188 18 offset += recvcounts_c[i];
189 }
190 }
191
192
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MPI_Gatherv(local_c.data(), static_cast<int>(local_c.size()), MPI_DOUBLE, final_res.data(), recvcounts_c.data(),
193 disps_c.data(), MPI_DOUBLE, 0, MPI_COMM_WORLD);
194
195
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (rank != 0) {
196
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 final_res.resize(static_cast<size_t>(rows_a) * cols_b);
197 }
198
199
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MPI_Bcast(final_res.data(), rows_a * cols_b, MPI_DOUBLE, 0, MPI_COMM_WORLD);
200 GetOutput() = final_res;
201 18 }
202
203 24 bool SmyshlaevAMatMulMPI::RunImpl() {
204 24 int rank = 0;
205 24 int size = 0;
206 24 MPI_Comm_rank(MPI_COMM_WORLD, &rank);
207 24 MPI_Comm_size(MPI_COMM_WORLD, &size);
208
209 24 int num_rows_a = 0;
210 24 int num_cols_a = 0;
211 24 int num_cols_b = 0;
212
213 24 BroadcastDimensions(num_rows_a, num_cols_a, num_cols_b);
214
215
4/4
✓ Branch 0 taken 20 times.
✓ Branch 1 taken 4 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 18 times.
 24 if (size > num_rows_a || size > num_cols_b) {
216 6 return RunSequential();
217 }
218
219 18 std::vector<int> sendcounts_a(size);
220
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<int> offsets_a(size);
221
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<int> sendcounts_b(size);
222
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<int> offsets_b(size);
223
224 18 CalculateDistribution(num_rows_a, size, sendcounts_a, offsets_a);
225
226
2/2
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
 54 for (int i = 0; i < size; ++i) {
227 36 sendcounts_a[i] *= num_cols_a;
228 36 offsets_a[i] *= num_cols_a;
229 }
230
231 18 CalculateDistribution(num_cols_b, size, sendcounts_b, offsets_b);
232
233
2/2
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
 54 for (int i = 0; i < size; ++i) {
234 36 sendcounts_b[i] *= num_cols_a;
235 36 offsets_b[i] *= num_cols_a;
236 }
237
238
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 int my_rows_a = sendcounts_a[rank] / num_cols_a;
239
240
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<double> local_a(sendcounts_a[rank]);
241
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<double> local_b(sendcounts_b[rank]);
242
243 18 int max_elem_count_b = 0;
244
2/2
✓ Branch 0 taken 36 times.
✓ Branch 1 taken 18 times.
 54 for (int c : sendcounts_b) {
245 36 max_elem_count_b = std::max(max_elem_count_b, c);
246 }
247
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<double> local_b_next(max_elem_count_b);
248
249
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 const double *sendbuf_a = (rank == 0) ? std::get<1>(GetInput()).data() : nullptr;
250
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MPI_Scatterv(sendbuf_a, sendcounts_a.data(), offsets_a.data(), MPI_DOUBLE, local_a.data(), sendcounts_a[rank],
251 MPI_DOUBLE, 0, MPI_COMM_WORLD);
252
253
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 const double *sendbuf_b = (rank == 0) ? mat_b_transposed_.data() : nullptr;
254
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MPI_Scatterv(sendbuf_b, sendcounts_b.data(), offsets_b.data(), MPI_DOUBLE, local_b.data(), sendcounts_b[rank],
255 MPI_DOUBLE, 0, MPI_COMM_WORLD);
256
257
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 18 std::vector<double> local_c(static_cast<size_t>(my_rows_a) * num_cols_b, 0.0);
258
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 RingShiftAlgorithm(rank, size, my_rows_a, num_cols_a, num_cols_b, sendcounts_b, offsets_b, local_a, local_b, local_c);
259
260
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 GatherAndBroadcastResults(rank, size, num_rows_a, num_cols_a, num_cols_b, sendcounts_a, local_c);
261 return true;
262 }
263
264 24 bool SmyshlaevAMatMulMPI::PostProcessingImpl() {
265 24 return true;
266 }
267
268 } // namespace smyshlaev_a_mat_mul
269