| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "chaschin_v_max_for_each_row/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <cstddef> | ||
| 7 | #include <limits> | ||
| 8 | #include <utility> | ||
| 9 | #include <vector> | ||
| 10 | |||
| 11 | #include "chaschin_v_max_for_each_row/common/include/common.hpp" | ||
| 12 | |||
| 13 | namespace chaschin_v_max_for_each_row { | ||
| 14 | |||
| 15 |
1/2✓ Branch 1 taken 40 times.
✗ Branch 2 not taken.
|
40 | ChaschinVMaxForEachRow::ChaschinVMaxForEachRow(const InType &in) { |
| 16 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 17 |
1/2✓ Branch 1 taken 40 times.
✗ Branch 2 not taken.
|
40 | auto in_copy = in; |
| 18 | 40 | GetInput() = std::move(in_copy); | |
| 19 | this->GetOutput().clear(); | ||
| 20 | 40 | } | |
| 21 | |||
| 22 | 40 | bool ChaschinVMaxForEachRow::ValidationImpl() { | |
| 23 | const auto &in = GetInput(); | ||
| 24 | |||
| 25 | if (in.empty()) { | ||
| 26 | return in.empty(); | ||
| 27 | } | ||
| 28 | |||
| 29 | if (in[0].empty()) { | ||
| 30 | return in[0].empty(); | ||
| 31 | } | ||
| 32 | |||
| 33 | return true; | ||
| 34 | } | ||
| 35 | |||
| 36 | 40 | bool ChaschinVMaxForEachRow::PreProcessingImpl() { | |
| 37 | 40 | return true; | |
| 38 | } | ||
| 39 | |||
| 40 | 20 | void chaschin_v_max_for_each_row::ChaschinVMaxForEachRow::SendRowsToWorkers(const std::vector<std::vector<float>> &mat, | |
| 41 | int size) { | ||
| 42 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 20 times.
|
40 | for (int pi = 1; pi < size; ++pi) { |
| 43 | // Inline ComputeRange | ||
| 44 | 20 | int nrows = static_cast<int>(mat.size()); | |
| 45 | 20 | int base = nrows / size; | |
| 46 | 20 | int rem = nrows % size; | |
| 47 |
1/2✓ Branch 0 taken 20 times.
✗ Branch 1 not taken.
|
20 | int start = (pi * base) + std::min(pi, rem); |
| 48 |
1/2✓ Branch 0 taken 20 times.
✗ Branch 1 not taken.
|
20 | int count = base + (pi < rem ? 1 : 0); |
| 49 | |||
| 50 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 20 times.
|
44 | for (int ii = 0; ii < count; ++ii) { |
| 51 | 24 | const auto &row = mat[start + ii]; | |
| 52 | 24 | int len = static_cast<int>(row.size()); | |
| 53 | |||
| 54 | 24 | MPI_Send(&len, 1, MPI_INT, pi, 100, MPI_COMM_WORLD); | |
| 55 |
1/2✓ Branch 0 taken 24 times.
✗ Branch 1 not taken.
|
24 | if (len > 0) { |
| 56 | 24 | MPI_Send(row.data(), len, MPI_FLOAT, pi, 101, MPI_COMM_WORLD); | |
| 57 | } | ||
| 58 | } | ||
| 59 | } | ||
| 60 | 20 | } | |
| 61 | |||
| 62 | 20 | void chaschin_v_max_for_each_row::ChaschinVMaxForEachRow::ReceiveRowsFromRoot( | |
| 63 | std::vector<std::vector<float>> &local_mat) { | ||
| 64 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 20 times.
|
44 | for (auto &row : local_mat) { |
| 65 | 24 | int len = 0; | |
| 66 | 24 | MPI_Recv(&len, 1, MPI_INT, 0, 100, MPI_COMM_WORLD, MPI_STATUS_IGNORE); | |
| 67 | |||
| 68 | 24 | row.resize(len); | |
| 69 |
1/2✓ Branch 0 taken 24 times.
✗ Branch 1 not taken.
|
24 | if (len > 0) { |
| 70 | 24 | MPI_Recv(row.data(), len, MPI_FLOAT, 0, 101, MPI_COMM_WORLD, MPI_STATUS_IGNORE); | |
| 71 | } | ||
| 72 | } | ||
| 73 | 20 | } | |
| 74 | |||
| 75 | 40 | std::vector<std::vector<float>> chaschin_v_max_for_each_row::ChaschinVMaxForEachRow::DistributeRows( | |
| 76 | const std::vector<std::vector<float>> &mat, int rank, int size, const RowRange &range) { | ||
| 77 | 40 | std::vector<std::vector<float>> local_mat(range.count); | |
| 78 | |||
| 79 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 20 times.
|
40 | if (rank == 0) { |
| 80 |
1/2✓ Branch 1 taken 20 times.
✗ Branch 2 not taken.
|
20 | SendRowsToWorkers(mat, size); |
| 81 | |||
| 82 |
2/2✓ Branch 0 taken 40 times.
✓ Branch 1 taken 20 times.
|
60 | for (int ii = 0; ii < range.count; ++ii) { |
| 83 |
1/2✓ Branch 1 taken 40 times.
✗ Branch 2 not taken.
|
40 | local_mat[ii] = mat[range.start + ii]; |
| 84 | } | ||
| 85 | } else { | ||
| 86 |
1/2✓ Branch 1 taken 20 times.
✗ Branch 2 not taken.
|
20 | ReceiveRowsFromRoot(local_mat); |
| 87 | } | ||
| 88 | |||
| 89 | 40 | return local_mat; | |
| 90 | ✗ | } | |
| 91 | |||
| 92 | 40 | std::vector<float> chaschin_v_max_for_each_row::ChaschinVMaxForEachRow::ComputeLocalMax( | |
| 93 | const std::vector<std::vector<float>> &local_mat) { | ||
| 94 | 40 | std::vector<float> local_out(local_mat.size()); | |
| 95 |
2/2✓ Branch 0 taken 64 times.
✓ Branch 1 taken 40 times.
|
104 | for (size_t ii = 0; ii < local_mat.size(); ++ii) { |
| 96 |
1/2✓ Branch 0 taken 64 times.
✗ Branch 1 not taken.
|
128 | local_out[ii] = local_mat[ii].empty() ? std::numeric_limits<float>::lowest() |
| 97 | : *std::max_element(local_mat[ii].begin(), local_mat[ii].end()); | ||
| 98 | } | ||
| 99 | 40 | return local_out; | |
| 100 | } | ||
| 101 | |||
| 102 | namespace { | ||
| 103 | inline void GetRangeForRank(int rank, int total, int world_size, int &start, int &count) { | ||
| 104 | 20 | int base = total / world_size; | |
| 105 | 20 | int rem = total % world_size; | |
| 106 |
1/2✓ Branch 0 taken 20 times.
✗ Branch 1 not taken.
|
20 | start = (rank * base) + std::min(rank, rem); |
| 107 |
1/2✓ Branch 0 taken 20 times.
✗ Branch 1 not taken.
|
20 | count = base + (rank < rem ? 1 : 0); |
| 108 | } | ||
| 109 | |||
| 110 | 12 | inline void RecvRows(int src_rank, std::vector<float> &out, int start, int count) { | |
| 111 |
1/2✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
|
12 | std::vector<float> tmp(count); |
| 112 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | MPI_Recv(tmp.data(), count, MPI_FLOAT, src_rank, 2, MPI_COMM_WORLD, MPI_STATUS_IGNORE); |
| 113 | std::ranges::copy(tmp, out.begin() + start); | ||
| 114 | 12 | } | |
| 115 | } // namespace | ||
| 116 | |||
| 117 | 40 | void chaschin_v_max_for_each_row::ChaschinVMaxForEachRow::GatherResults(std::vector<float> &out, | |
| 118 | const std::vector<float> &local_out, int rank, | ||
| 119 | int size, const RowRange &range) { | ||
| 120 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 20 times.
|
40 | if (rank != 0) { |
| 121 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 8 times.
|
20 | if (!local_out.empty()) { |
| 122 | 12 | MPI_Send(local_out.data(), static_cast<int>(local_out.size()), MPI_FLOAT, 0, 2, MPI_COMM_WORLD); | |
| 123 | } | ||
| 124 | 20 | return; | |
| 125 | } | ||
| 126 | |||
| 127 |
2/2✓ Branch 0 taken 40 times.
✓ Branch 1 taken 20 times.
|
60 | for (int i = 0; i < range.count; ++i) { |
| 128 | 40 | out[range.start + i] = local_out[i]; | |
| 129 | } | ||
| 130 | |||
| 131 | 20 | int total = static_cast<int>(out.size()); | |
| 132 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 20 times.
|
40 | for (int pi = 1; pi < size; ++pi) { |
| 133 | int start = 0; | ||
| 134 | int count = 0; | ||
| 135 | |||
| 136 | GetRangeForRank(pi, total, size, start, count); | ||
| 137 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 8 times.
|
20 | if (count > 0) { |
| 138 | 12 | RecvRows(pi, out, start, count); | |
| 139 | } | ||
| 140 | } | ||
| 141 | } | ||
| 142 | |||
| 143 | 40 | bool ChaschinVMaxForEachRow::RunImpl() { | |
| 144 | 40 | int rank = 0; | |
| 145 | 40 | int size = 0; | |
| 146 | 40 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 147 | 40 | MPI_Comm_size(MPI_COMM_WORLD, &size); | |
| 148 | |||
| 149 | const auto &mat = GetInput(); | ||
| 150 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 20 times.
|
40 | int nrows = (rank == 0) ? static_cast<int>(mat.size()) : 0; |
| 151 | 40 | MPI_Bcast(&nrows, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 152 | |||
| 153 | 40 | int base = nrows / size; | |
| 154 | 40 | int rem = nrows % size; | |
| 155 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 16 times.
|
40 | int start = (rank * base) + std::min(rank, rem); |
| 156 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 16 times.
|
40 | int count = base + (rank < rem ? 1 : 0); |
| 157 | 40 | RowRange range{.start = start, .count = count}; | |
| 158 | |||
| 159 | 40 | auto local_mat = DistributeRows(mat, rank, size, range); | |
| 160 |
1/2✓ Branch 1 taken 40 times.
✗ Branch 2 not taken.
|
40 | auto local_out = ComputeLocalMax(local_mat); |
| 161 | |||
| 162 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 20 times.
|
40 | if (rank == 0) { |
| 163 |
1/2✓ Branch 1 taken 20 times.
✗ Branch 2 not taken.
|
20 | GetOutput().resize(nrows); |
| 164 | } | ||
| 165 |
1/2✓ Branch 1 taken 40 times.
✗ Branch 2 not taken.
|
40 | GatherResults(GetOutput(), local_out, rank, size, range); |
| 166 | |||
| 167 | auto &out = GetOutput(); | ||
| 168 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 20 times.
|
40 | if (rank != 0) { |
| 169 |
1/2✓ Branch 1 taken 20 times.
✗ Branch 2 not taken.
|
20 | out.resize(nrows); |
| 170 | } | ||
| 171 | |||
| 172 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 8 times.
|
40 | if (nrows > 0) { |
| 173 |
1/2✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 | MPI_Bcast(out.data(), nrows, MPI_FLOAT, 0, MPI_COMM_WORLD); |
| 174 | } | ||
| 175 | |||
| 176 | 40 | return true; | |
| 177 | 40 | } | |
| 178 | |||
| 179 | 40 | bool ChaschinVMaxForEachRow::PostProcessingImpl() { | |
| 180 | 40 | int rank = 0; | |
| 181 | 40 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 182 | 40 | return true; | |
| 183 | } | ||
| 184 | } // namespace chaschin_v_max_for_each_row | ||
| 185 |