| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "kruglova_a_vertical_ribbon_matvec/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <cstddef> | ||
| 6 | #include <vector> | ||
| 7 | |||
| 8 | #include "kruglova_a_vertical_ribbon_matvec/common/include/common.hpp" | ||
| 9 | |||
| 10 | namespace kruglova_a_vertical_ribbon_matvec { | ||
| 11 | |||
| 12 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | KruglovaAVerticalRibbMatMPI::KruglovaAVerticalRibbMatMPI(const InType &in) { |
| 13 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 14 | GetInput() = in; | ||
| 15 | GetOutput() = {}; | ||
| 16 | 10 | } | |
| 17 | |||
| 18 | 10 | bool KruglovaAVerticalRibbMatMPI::ValidationImpl() { | |
| 19 | 10 | int rank = 0; | |
| 20 | 10 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 21 | |||
| 22 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 23 | 5 | const int rows = std::get<0>(GetInput()); | |
| 24 | 5 | const int cols = std::get<1>(GetInput()); | |
| 25 | const std::vector<double> &matrix = std::get<2>(GetInput()); | ||
| 26 | const std::vector<double> &vec = std::get<3>(GetInput()); | ||
| 27 | |||
| 28 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | if (rows <= 0 || cols <= 0) { |
| 29 | return false; | ||
| 30 | } | ||
| 31 |
1/2✓ Branch 0 taken 5 times.
✗ Branch 1 not taken.
|
5 | if (matrix.size() != static_cast<size_t>(rows) * static_cast<size_t>(cols)) { |
| 32 | return false; | ||
| 33 | } | ||
| 34 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 5 times.
|
5 | if (vec.size() != static_cast<size_t>(cols)) { |
| 35 | ✗ | return false; | |
| 36 | } | ||
| 37 | } | ||
| 38 | return true; | ||
| 39 | } | ||
| 40 | |||
| 41 | 10 | bool KruglovaAVerticalRibbMatMPI::PreProcessingImpl() { | |
| 42 | 10 | int rank = 0; | |
| 43 | 10 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 44 | |||
| 45 | 10 | int rows = 0; | |
| 46 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 47 | 5 | rows = std::get<0>(GetInput()); | |
| 48 | } | ||
| 49 | |||
| 50 | 10 | MPI_Bcast(&rows, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 51 | |||
| 52 | 10 | GetOutput().assign(rows, 0.0); | |
| 53 | |||
| 54 | 10 | return true; | |
| 55 | } | ||
| 56 | |||
| 57 | namespace { | ||
| 58 | |||
| 59 | 10 | std::vector<int> ComputeSendCountsAndDispls(int cols, int size, std::vector<int> &displs) { | |
| 60 | 10 | std::vector<int> sendcounts(size); | |
| 61 | 10 | const int base_cols = cols / size; | |
| 62 | 10 | const int rem_cols = cols % size; | |
| 63 | int offset = 0; | ||
| 64 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 10 times.
|
30 | for (int i = 0; i < size; ++i) { |
| 65 |
1/2✓ Branch 0 taken 20 times.
✗ Branch 1 not taken.
|
40 | sendcounts[i] = base_cols + (i < rem_cols ? 1 : 0); |
| 66 | 20 | displs[i] = offset; | |
| 67 | 20 | offset += sendcounts[i]; | |
| 68 | } | ||
| 69 | 10 | return sendcounts; | |
| 70 | } | ||
| 71 | |||
| 72 | void TransposeMatrix(const std::vector<double> &matrix, std::vector<double> &transposed, int rows, int cols) { | ||
| 73 |
2/2✓ Branch 0 taken 760 times.
✓ Branch 1 taken 5 times.
|
765 | for (int i = 0; i < rows; ++i) { |
| 74 |
2/2✓ Branch 0 taken 112600 times.
✓ Branch 1 taken 760 times.
|
113360 | for (int j = 0; j < cols; ++j) { |
| 75 | 112600 | transposed[(static_cast<size_t>(j) * rows) + i] = matrix[(static_cast<size_t>(i) * cols) + j]; | |
| 76 | } | ||
| 77 | } | ||
| 78 | } | ||
| 79 | |||
| 80 | 10 | void LocalMatVecMul(const std::vector<double> &local_matrix, const std::vector<double> &local_b, | |
| 81 | std::vector<double> &local_res, int rows, int local_cols) { | ||
| 82 |
2/2✓ Branch 0 taken 760 times.
✓ Branch 1 taken 10 times.
|
770 | for (int j = 0; j < local_cols; ++j) { |
| 83 | 760 | const double b_val = local_b[j]; | |
| 84 |
2/2✓ Branch 0 taken 112600 times.
✓ Branch 1 taken 760 times.
|
113360 | for (int i = 0; i < rows; ++i) { |
| 85 | 112600 | local_res[i] += local_matrix[(static_cast<size_t>(j) * rows) + i] * b_val; | |
| 86 | } | ||
| 87 | } | ||
| 88 | 10 | } | |
| 89 | |||
| 90 | } // namespace | ||
| 91 | |||
| 92 | 10 | bool KruglovaAVerticalRibbMatMPI::RunImpl() { | |
| 93 | 10 | int rank = 0; | |
| 94 | 10 | int size = 0; | |
| 95 | 10 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 96 | 10 | MPI_Comm_size(MPI_COMM_WORLD, &size); | |
| 97 | |||
| 98 | 10 | int rows = 0; | |
| 99 | 10 | int cols = 0; | |
| 100 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 101 | 5 | rows = std::get<0>(GetInput()); | |
| 102 | 5 | cols = std::get<1>(GetInput()); | |
| 103 | } | ||
| 104 | |||
| 105 | 10 | MPI_Bcast(&rows, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 106 | 10 | MPI_Bcast(&cols, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 107 | |||
| 108 | 10 | std::vector<int> displs(size); | |
| 109 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | std::vector<int> sendcounts = ComputeSendCountsAndDispls(cols, size, displs); |
| 110 | |||
| 111 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | int local_cols = sendcounts[rank]; |
| 112 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<double> local_b(local_cols); |
| 113 | |||
| 114 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | const double *global_b_ptr = (rank == 0) ? std::get<3>(GetInput()).data() : nullptr; |
| 115 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | MPI_Scatterv(global_b_ptr, sendcounts.data(), displs.data(), MPI_DOUBLE, local_b.data(), local_cols, MPI_DOUBLE, 0, |
| 116 | MPI_COMM_WORLD); | ||
| 117 | |||
| 118 | 10 | std::vector<double> transposed_matrix; | |
| 119 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | if (rank == 0) { |
| 120 | const auto &matrix = std::get<2>(GetInput()); | ||
| 121 |
1/2✓ Branch 1 taken 5 times.
✗ Branch 2 not taken.
|
5 | transposed_matrix.resize(static_cast<size_t>(rows) * cols); |
| 122 | 5 | TransposeMatrix(matrix, transposed_matrix, rows, cols); | |
| 123 | } | ||
| 124 | |||
| 125 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<int> mat_sendcounts(size); |
| 126 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<int> mat_displs(size); |
| 127 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 10 times.
|
30 | for (int i = 0; i < size; ++i) { |
| 128 | 20 | mat_sendcounts[i] = sendcounts[i] * rows; | |
| 129 | 20 | mat_displs[i] = displs[i] * rows; | |
| 130 | } | ||
| 131 | |||
| 132 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<double> local_matrix(static_cast<size_t>(local_cols) * rows); |
| 133 |
3/4✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
✓ Branch 3 taken 10 times.
✗ Branch 4 not taken.
|
15 | MPI_Scatterv(rank == 0 ? transposed_matrix.data() : nullptr, mat_sendcounts.data(), mat_displs.data(), MPI_DOUBLE, |
| 134 | local_matrix.data(), local_cols * rows, MPI_DOUBLE, 0, MPI_COMM_WORLD); | ||
| 135 | |||
| 136 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<double> local_res(rows, 0.0); |
| 137 | 10 | LocalMatVecMul(local_matrix, local_b, local_res, rows, local_cols); | |
| 138 | |||
| 139 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 5 times.
|
10 | double *global_res_ptr = (rank == 0) ? GetOutput().data() : nullptr; |
| 140 |
1/2✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
|
10 | MPI_Reduce(local_res.data(), global_res_ptr, rows, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD); |
| 141 | |||
| 142 | 10 | return true; | |
| 143 | } | ||
| 144 | |||
| 145 | 10 | bool KruglovaAVerticalRibbMatMPI::PostProcessingImpl() { | |
| 146 | 10 | return true; | |
| 147 | } | ||
| 148 | |||
| 149 | } // namespace kruglova_a_vertical_ribbon_matvec | ||
| 150 |