GCC Code Coverage Report

Directory: ./
File: tasks/makoveeva_s_cannon_algorithm/mpi/src/ops_mpi.cpp
Date: 2026-01-27 01:59:34
Exec Total Coverage
Lines: 162 168 96.4%
Functions: 17 17 100.0%
Branches: 67 120 55.8%
Line Branch Exec Source
1 #include "makoveeva_s_cannon_algorithm/mpi/include/ops_mpi.hpp"
2
3 #include <mpi.h>
4
5 #include <array>
6 #include <cmath>
7 #include <cstddef>
8 #include <tuple>
9 #include <utility>
10 #include <vector>
11
12 #include "makoveeva_s_cannon_algorithm/common/include/common.hpp"
13
14 namespace makoveeva_s_cannon_algorithm {
15
16 namespace {
17
18 bool CheckInputLocal(const std::vector<double> &a, const std::vector<double> &b, int n) {
19
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 if (n <= 0) {
20 return false;
21 }
22 9 const auto n_sz = static_cast<std::size_t>(n);
23
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 const auto expected = n_sz * n_sz;
24
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 9 times.
 9 return (a.size() == expected) && (b.size() == expected);
25 }
26
27 36 int ChooseGridQ(int size, int n) {
28
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 36 times.
 36 if (size <= 0 || n <= 0) {
29 return 0;
30 }
31 36 int q = static_cast<int>(std::floor(std::sqrt(static_cast<double>(size))));
32
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 36 times.
 36 while (q > 0) {
33
2/4
✓ Branch 0 taken 36 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 36 times.
 36 if ((q * q) <= size && (n % q == 0)) {
34 return q;
35 }
36 ✗ --q;
37 }
38 return 0;
39 }
40
41 54 int BroadcastN(const InType &input, int world_rank) {
42 54 int n = 0;
43
2/2
✓ Branch 0 taken 27 times.
✓ Branch 1 taken 27 times.
 54 if (world_rank == 0) {
44 27 n = std::get<2>(input);
45 }
46 54 MPI_Bcast(&n, 1, MPI_INT, 0, MPI_COMM_WORLD);
47 54 return n;
48 }
49
50 9 MPI_Datatype MakeBlockType(int n, int bs) {
51 9 MPI_Datatype block = MPI_DATATYPE_NULL;
52 9 MPI_Type_vector(bs, bs, n, MPI_DOUBLE, &block);
53
54 9 MPI_Datatype resized = MPI_DATATYPE_NULL;
55 9 MPI_Type_create_resized(block, 0, static_cast<MPI_Aint>(sizeof(double)), &resized);
56
57 9 MPI_Type_commit(&resized);
58 9 MPI_Type_free(&block);
59 9 return resized;
60 }
61
62 9 void LocalMatMulAcc(const std::vector<double> &a, const std::vector<double> &b, int bs, std::vector<double> *c) {
63 9 const auto bs_sz = static_cast<std::size_t>(bs);
64
2/2
✓ Branch 0 taken 93 times.
✓ Branch 1 taken 9 times.
 102 for (int i = 0; i < bs; ++i) {
65 93 const auto i_sz = static_cast<std::size_t>(i);
66
2/2
✓ Branch 0 taken 1823 times.
✓ Branch 1 taken 93 times.
 1916 for (int k = 0; k < bs; ++k) {
67 1823 const auto k_sz = static_cast<std::size_t>(k);
68 1823 const double a_ik = a[(i_sz * bs_sz) + k_sz];
69
2/2
✓ Branch 0 taken 45819 times.
✓ Branch 1 taken 1823 times.
 47642 for (int j = 0; j < bs; ++j) {
70 45819 const auto j_sz = static_cast<std::size_t>(j);
71 45819 (*c)[(i_sz * bs_sz) + j_sz] += a_ik * b[(k_sz * bs_sz) + j_sz];
72 }
73 }
74 }
75 9 }
76
77 MPI_Comm MakeActiveComm(int world_rank, int active_p, bool *is_active) {
78 *is_active = (world_rank < active_p);
79 18 MPI_Comm active_comm = MPI_COMM_NULL;
80 18 const int color = (*is_active) ? 0 : MPI_UNDEFINED;
81 18 MPI_Comm_split(MPI_COMM_WORLD, color, world_rank, &active_comm);
82 18 return active_comm;
83 }
84
85 9 MPI_Comm MakeCartComm(MPI_Comm active_comm, int q) {
86 9 const std::array<int, 2> dims = {q, q};
87 9 const std::array<int, 2> periods = {1, 1};
88
89 9 MPI_Comm cart_comm = MPI_COMM_NULL;
90 9 MPI_Cart_create(active_comm, 2, dims.data(), periods.data(), 0 /*reorder*/, &cart_comm);
91 9 return cart_comm;
92 }
93
94 9 std::array<int, 2> GetCoords(MPI_Comm cart_comm) {
95 9 int cart_rank = 0;
96 9 MPI_Comm_rank(cart_comm, &cart_rank);
97
98 9 std::array<int, 2> coords = {0, 0};
99 9 MPI_Cart_coords(cart_comm, cart_rank, 2, coords.data());
100 9 return coords;
101 }
102
103 9 void BuildScatterMeta(int q, int bs, int n, int active_p, std::vector<int> *sendcounts, std::vector<int> *displs,
104 MPI_Comm cart_comm) {
105 9 int cart_rank = 0;
106 9 MPI_Comm_rank(cart_comm, &cart_rank);
107
108
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 if (cart_rank != 0) {
109 ✗ return;
110 }
111
112 9 sendcounts->assign(static_cast<std::size_t>(active_p), 1);
113 9 displs->assign(static_cast<std::size_t>(active_p), 0);
114
115
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 for (int row_idx = 0; row_idx < q; ++row_idx) {
116
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 for (int col_idx = 0; col_idx < q; ++col_idx) {
117 9 const int proc = (row_idx * q) + col_idx;
118 9 (*displs)[static_cast<std::size_t>(proc)] = ((row_idx * bs) * n) + (col_idx * bs);
119 }
120 }
121 }
122
123 9 void ScatterBlocks(const double *a_full_ptr, const double *b_full_ptr, const int *sendcounts_ptr, const int *displs_ptr,
124 MPI_Datatype block_type, int bs, MPI_Comm cart_comm, std::vector<double> *a_block,
125 std::vector<double> *b_block) {
126 9 MPI_Scatterv(a_full_ptr, sendcounts_ptr, displs_ptr, block_type, a_block->data(), bs * bs, MPI_DOUBLE, 0, cart_comm);
127 9 MPI_Scatterv(b_full_ptr, sendcounts_ptr, displs_ptr, block_type, b_block->data(), bs * bs, MPI_DOUBLE, 0, cart_comm);
128 9 }
129
130 9 void InitialShift(MPI_Comm cart_comm, int row, int col, int bs, std::vector<double> *a_block,
131 std::vector<double> *b_block) {
132 9 int src = 0;
133 9 int dst = 0;
134
135 9 MPI_Cart_shift(cart_comm, 1, -row, &src, &dst);
136 9 MPI_Sendrecv_replace(a_block->data(), bs * bs, MPI_DOUBLE, dst, 0, src, 0, cart_comm, MPI_STATUS_IGNORE);
137
138 9 MPI_Cart_shift(cart_comm, 0, -col, &src, &dst);
139 9 MPI_Sendrecv_replace(b_block->data(), bs * bs, MPI_DOUBLE, dst, 1, src, 1, cart_comm, MPI_STATUS_IGNORE);
140 9 }
141
142 9 void CannonLoop(MPI_Comm cart_comm, int q, int bs, std::vector<double> *a_block, std::vector<double> *b_block,
143 std::vector<double> *c_block) {
144 9 int src = 0;
145 9 int dst = 0;
146
147
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 for (int step = 0; step < q; ++step) {
148 9 LocalMatMulAcc(*a_block, *b_block, bs, c_block);
149
150
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
 9 if (step + 1 < q) {
151 ✗ MPI_Cart_shift(cart_comm, 1, -1, &src, &dst);
152 ✗ MPI_Sendrecv_replace(a_block->data(), bs * bs, MPI_DOUBLE, dst, 2, src, 2, cart_comm, MPI_STATUS_IGNORE);
153
154 ✗ MPI_Cart_shift(cart_comm, 0, -1, &src, &dst);
155 ✗ MPI_Sendrecv_replace(b_block->data(), bs * bs, MPI_DOUBLE, dst, 3, src, 3, cart_comm, MPI_STATUS_IGNORE);
156 }
157 }
158 9 }
159
160 9 std::vector<double> GatherResult(MPI_Comm cart_comm, int n, int bs, MPI_Datatype block_type, const int *sendcounts_ptr,
161 const int *displs_ptr, const std::vector<double> &c_block) {
162 9 int cart_rank = 0;
163 9 MPI_Comm_rank(cart_comm, &cart_rank);
164
165 9 std::vector<double> c_full;
166 double *recv_ptr = nullptr;
167
168
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 9 if (cart_rank == 0) {
169
1/4
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 9 c_full.assign(static_cast<std::size_t>(n) * static_cast<std::size_t>(n), 0.0);
170 recv_ptr = c_full.data();
171 }
172
173
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 MPI_Gatherv(c_block.data(), bs * bs, MPI_DOUBLE, recv_ptr, sendcounts_ptr, displs_ptr, block_type, 0, cart_comm);
174 9 return c_full;
175 }
176
177 18 std::vector<double> BroadcastFullToAll(std::vector<double> c_full, int world_rank) {
178 18 int out_size = 0;
179
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (world_rank == 0) {
180 9 out_size = static_cast<int>(c_full.size());
181 }
182 18 MPI_Bcast(&out_size, 1, MPI_INT, 0, MPI_COMM_WORLD);
183
184
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (world_rank != 0) {
185 9 c_full.assign(static_cast<std::size_t>(out_size), 0.0);
186 }
187
188
1/2
✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
 18 if (out_size > 0) {
189 18 MPI_Bcast(c_full.data(), out_size, MPI_DOUBLE, 0, MPI_COMM_WORLD);
190 }
191 18 return c_full;
192 }
193
194 } // namespace
195
196
1/2
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
 18 MakoveevaSCannonAlgorithmMPI::MakoveevaSCannonAlgorithmMPI(const InType &in) {
197 SetTypeOfTask(GetStaticTypeOfTask());
198 GetInput() = in;
199 18 GetOutput() = OutType{};
200 18 }
201
202 18 bool MakoveevaSCannonAlgorithmMPI::ValidationImpl() {
203 18 int world_rank = 0;
204 18 int world_size = 0;
205 18 MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
206 18 MPI_Comm_size(MPI_COMM_WORLD, &world_size);
207
208 18 const int n = BroadcastN(GetInput(), world_rank);
209
210 18 int ok_input = 0;
211
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (world_rank == 0) {
212 const auto &a = std::get<0>(GetInput());
213 const auto &b = std::get<1>(GetInput());
214
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 18 ok_input = (GetOutput().empty() && CheckInputLocal(a, b, n)) ? 1 : 0;
215 }
216 18 MPI_Bcast(&ok_input, 1, MPI_INT, 0, MPI_COMM_WORLD);
217
218 18 const int q = ChooseGridQ(world_size, n);
219
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 const int ok_grid = (q > 0) ? 1 : 0;
220
221
2/4
✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 18 times.
 18 return (ok_input != 0) && (ok_grid != 0);
222 }
223
224
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 bool MakoveevaSCannonAlgorithmMPI::PreProcessingImpl() {
225
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 if (!GetOutput().empty()) {
226 GetOutput().clear();
227 }
228 18 return true;
229 }
230
231 18 bool MakoveevaSCannonAlgorithmMPI::RunImpl() {
232 18 int world_rank = 0;
233 18 int world_size = 0;
234 18 MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
235 18 MPI_Comm_size(MPI_COMM_WORLD, &world_size);
236
237 18 const int n = BroadcastN(GetInput(), world_rank);
238
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 if (n <= 0) {
239 return false;
240 }
241
242 18 const int q = ChooseGridQ(world_size, n);
243
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 18 times.
 18 if (q <= 0) {
244 return false;
245 }
246
247 18 const int active_p = q * q;
248 bool is_active = false;
249
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 MPI_Comm active_comm = MakeActiveComm(world_rank, active_p, &is_active);
250
251 18 std::vector<double> c_full;
252
253
2/2
✓ Branch 0 taken 9 times.
✓ Branch 1 taken 9 times.
 18 if (is_active) {
254 9 const int bs = n / q;
255 9 const auto bs_sz = static_cast<std::size_t>(bs);
256
257
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 MPI_Comm cart_comm = MakeCartComm(active_comm, q);
258
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 const std::array<int, 2> coords = GetCoords(cart_comm);
259 9 const int row = coords[0];
260 9 const int col = coords[1];
261
262
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 std::vector<double> a_block(bs_sz * bs_sz, 0.0);
263
1/4
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 9 std::vector<double> b_block(bs_sz * bs_sz, 0.0);
264
1/4
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 9 std::vector<double> c_block(bs_sz * bs_sz, 0.0);
265
266
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 MPI_Datatype block_type = MakeBlockType(n, bs);
267
268 9 std::vector<int> sendcounts;
269 9 std::vector<int> displs;
270
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 BuildScatterMeta(q, bs, n, active_p, &sendcounts, &displs, cart_comm);
271
272 9 int cart_rank = 0;
273
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 MPI_Comm_rank(cart_comm, &cart_rank);
274
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 9 const int *sendcounts_ptr = (cart_rank == 0) ? sendcounts.data() : nullptr;
275
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 9 const int *displs_ptr = (cart_rank == 0) ? displs.data() : nullptr;
276
277 const double *a_full_ptr = nullptr;
278 const double *b_full_ptr = nullptr;
279
1/2
✓ Branch 0 taken 9 times.
✗ Branch 1 not taken.
 9 if (world_rank == 0) {
280 a_full_ptr = std::get<0>(GetInput()).data();
281 b_full_ptr = std::get<1>(GetInput()).data();
282 }
283
284
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 ScatterBlocks(a_full_ptr, b_full_ptr, sendcounts_ptr, displs_ptr, block_type, bs, cart_comm, &a_block, &b_block);
285
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 InitialShift(cart_comm, row, col, bs, &a_block, &b_block);
286
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 CannonLoop(cart_comm, q, bs, &a_block, &b_block, &c_block);
287
288
1/4
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
 9 c_full = GatherResult(cart_comm, n, bs, block_type, sendcounts_ptr, displs_ptr, c_block);
289
290
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 MPI_Type_free(&block_type);
291
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 MPI_Comm_free(&cart_comm);
292
1/2
✓ Branch 1 taken 9 times.
✗ Branch 2 not taken.
 9 MPI_Comm_free(&active_comm);
293 }
294
295
1/4
✓ Branch 1 taken 18 times.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
 36 c_full = BroadcastFullToAll(std::move(c_full), world_rank);
296
297 GetOutput() = std::move(c_full);
298 return true;
299 }
300
301 18 bool MakoveevaSCannonAlgorithmMPI::PostProcessingImpl() {
302 18 int world_rank = 0;
303 18 MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
304
305 18 const int n = BroadcastN(GetInput(), world_rank);
306
1/2
✓ Branch 0 taken 18 times.
✗ Branch 1 not taken.
 18 if (n <= 0) {
307 return false;
308 }
309
310 18 const auto n_sz = static_cast<std::size_t>(n);
311 18 return GetOutput().size() == (n_sz * n_sz);
312 }
313
314 } // namespace makoveeva_s_cannon_algorithm
315