| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "kamalagin_a_vec_mat_mult/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <cstddef> | ||
| 6 | #include <cstdint> | ||
| 7 | #include <vector> | ||
| 8 | |||
| 9 | #include "kamalagin_a_vec_mat_mult/common/include/common.hpp" | ||
| 10 | |||
| 11 | namespace kamalagin_a_vec_mat_mult { | ||
| 12 | |||
| 13 | namespace { | ||
| 14 | |||
| 15 | 16 | void BuildCountsDispls(int n, int size, std::vector<int> *counts, std::vector<int> *displs) { | |
| 16 | 16 | counts->assign(size, 0); | |
| 17 | 16 | displs->assign(size, 0); | |
| 18 | |||
| 19 | 16 | const int base = n / size; | |
| 20 | 16 | const int rem = n % size; | |
| 21 | |||
| 22 | int offset = 0; | ||
| 23 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 16 times.
|
48 | for (int proc = 0; proc < size; ++proc) { |
| 24 |
2/2✓ Branch 0 taken 22 times.
✓ Branch 1 taken 10 times.
|
32 | const int cnt = base + ((proc < rem) ? 1 : 0); |
| 25 | 32 | (*counts)[proc] = cnt; | |
| 26 | 32 | (*displs)[proc] = offset; | |
| 27 | 32 | offset += cnt; | |
| 28 | } | ||
| 29 | 16 | } | |
| 30 | |||
| 31 | } // namespace | ||
| 32 | |||
| 33 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | KamalaginAVecMatMultMPI::KamalaginAVecMatMultMPI(const InType &in) { |
| 34 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 35 | GetInput() = in; | ||
| 36 | GetOutput().clear(); | ||
| 37 | 16 | } | |
| 38 | |||
| 39 | 16 | bool KamalaginAVecMatMultMPI::ValidationImpl() { | |
| 40 | const auto &[n, m, a_flat, x] = GetInput(); | ||
| 41 |
2/4✓ Branch 0 taken 16 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 16 times.
✗ Branch 3 not taken.
|
16 | if (n < 0 || m < 0) { |
| 42 | return false; | ||
| 43 | } | ||
| 44 |
1/2✓ Branch 0 taken 16 times.
✗ Branch 1 not taken.
|
16 | if (static_cast<std::size_t>(n) * static_cast<std::size_t>(m) != a_flat.size()) { |
| 45 | return false; | ||
| 46 | } | ||
| 47 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 16 times.
|
16 | if (static_cast<std::size_t>(m) != x.size()) { |
| 48 | ✗ | return false; | |
| 49 | } | ||
| 50 | return true; | ||
| 51 | } | ||
| 52 | |||
| 53 | 16 | bool KamalaginAVecMatMultMPI::PreProcessingImpl() { | |
| 54 | const auto &[n, m, a_flat, x] = GetInput(); | ||
| 55 | (void)m; | ||
| 56 | (void)a_flat; | ||
| 57 | (void)x; | ||
| 58 | 16 | GetOutput().assign(static_cast<std::size_t>(n), 0); | |
| 59 | 16 | return true; | |
| 60 | } | ||
| 61 | |||
| 62 | 16 | bool KamalaginAVecMatMultMPI::RunImpl() { | |
| 63 | 16 | int rank = 0; | |
| 64 | 16 | int size = 1; | |
| 65 | 16 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 66 | 16 | MPI_Comm_size(MPI_COMM_WORLD, &size); | |
| 67 | |||
| 68 | 16 | int n = 0; | |
| 69 | 16 | int m = 0; | |
| 70 | 16 | std::vector<int> a_root; | |
| 71 | 16 | std::vector<int> x; | |
| 72 | |||
| 73 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 8 times.
|
16 | if (rank == 0) { |
| 74 | const auto &input = GetInput(); | ||
| 75 | 8 | n = std::get<0>(input); | |
| 76 |
1/2✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 | m = std::get<1>(input); |
| 77 |
1/2✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 | a_root = std::get<2>(input); |
| 78 |
1/2✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
|
8 | x = std::get<3>(input); |
| 79 | } | ||
| 80 | |||
| 81 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | MPI_Bcast(&n, 1, MPI_INT, 0, MPI_COMM_WORLD); |
| 82 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | MPI_Bcast(&m, 1, MPI_INT, 0, MPI_COMM_WORLD); |
| 83 | |||
| 84 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 4 times.
|
16 | if (m > 0) { |
| 85 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 6 times.
|
12 | if (rank != 0) { |
| 86 |
1/2✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 | x.resize(static_cast<std::size_t>(m)); |
| 87 | } | ||
| 88 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | MPI_Bcast(x.data(), m, MPI_INT, 0, MPI_COMM_WORLD); |
| 89 | } | ||
| 90 | |||
| 91 | 16 | std::vector<int> rows_counts; | |
| 92 | 16 | std::vector<int> rows_displs; | |
| 93 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | BuildCountsDispls(n, size, &rows_counts, &rows_displs); |
| 94 | |||
| 95 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | const int local_rows = rows_counts[rank]; |
| 96 |
1/4✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
16 | std::vector<int> a_local(static_cast<std::size_t>(local_rows) * static_cast<std::size_t>(m)); |
| 97 | |||
| 98 |
1/4✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
16 | std::vector<int> sendcounts_a(static_cast<std::size_t>(size)); |
| 99 |
1/4✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
16 | std::vector<int> displs_a(static_cast<std::size_t>(size)); |
| 100 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 16 times.
|
48 | for (int proc = 0; proc < size; ++proc) { |
| 101 | 32 | sendcounts_a[proc] = rows_counts[proc] * m; | |
| 102 | 32 | displs_a[proc] = rows_displs[proc] * m; | |
| 103 | } | ||
| 104 | |||
| 105 |
3/4✓ Branch 0 taken 8 times.
✓ Branch 1 taken 8 times.
✓ Branch 3 taken 16 times.
✗ Branch 4 not taken.
|
24 | MPI_Scatterv(rank == 0 ? a_root.data() : nullptr, sendcounts_a.data(), displs_a.data(), MPI_INT, a_local.data(), |
| 106 | local_rows * m, MPI_INT, 0, MPI_COMM_WORLD); | ||
| 107 | |||
| 108 |
1/4✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
16 | std::vector<int> local_y(static_cast<std::size_t>(local_rows), 0); |
| 109 |
2/2✓ Branch 0 taken 279 times.
✓ Branch 1 taken 16 times.
|
295 | for (int i = 0; i < local_rows; ++i) { |
| 110 | std::int64_t sum = 0; | ||
| 111 |
2/2✓ Branch 0 taken 2606 times.
✓ Branch 1 taken 279 times.
|
2885 | for (int j = 0; j < m; ++j) { |
| 112 | 2606 | sum += | |
| 113 | 2606 | static_cast<std::int64_t>(a_local[(i * m) + j]) * static_cast<std::int64_t>(x[static_cast<std::size_t>(j)]); | |
| 114 | } | ||
| 115 | 279 | local_y[static_cast<std::size_t>(i)] = static_cast<int>(sum); | |
| 116 | } | ||
| 117 | |||
| 118 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | const std::vector<int> recvcounts_y = rows_counts; |
| 119 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | const std::vector<int> displs_y = rows_displs; |
| 120 | |||
| 121 |
3/4✓ Branch 0 taken 8 times.
✓ Branch 1 taken 8 times.
✓ Branch 3 taken 16 times.
✗ Branch 4 not taken.
|
24 | MPI_Gatherv(local_y.data(), local_rows, MPI_INT, rank == 0 ? GetOutput().data() : nullptr, recvcounts_y.data(), |
| 122 | displs_y.data(), MPI_INT, 0, MPI_COMM_WORLD); | ||
| 123 | |||
| 124 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 4 times.
|
16 | if (n > 0) { |
| 125 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | MPI_Bcast(GetOutput().data(), n, MPI_INT, 0, MPI_COMM_WORLD); |
| 126 | } | ||
| 127 | |||
| 128 | 16 | return true; | |
| 129 | } | ||
| 130 | |||
| 131 | 16 | bool KamalaginAVecMatMultMPI::PostProcessingImpl() { | |
| 132 | 16 | return true; | |
| 133 | } | ||
| 134 | |||
| 135 | } // namespace kamalagin_a_vec_mat_mult | ||
| 136 |