| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "mityaeva_radix/all/include/ops_all.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <cstddef> | ||
| 6 | #include <cstring> | ||
| 7 | #include <utility> | ||
| 8 | #include <vector> | ||
| 9 | |||
| 10 | #include "mityaeva_radix/common/include/common.hpp" | ||
| 11 | #include "mityaeva_radix/omp/include/sorter_omp.hpp" | ||
| 12 | |||
| 13 | namespace mityaeva_radix { | ||
| 14 | |||
| 15 | namespace { | ||
| 16 | |||
| 17 | 24 | void ComputeChunkParams(size_t total_size, int mpi_size, std::vector<size_t> &chunk_sizes, | |
| 18 | std::vector<size_t> &offsets) { | ||
| 19 | 24 | size_t base_chunk = total_size / mpi_size; | |
| 20 | 24 | size_t remainder = total_size % mpi_size; | |
| 21 | |||
| 22 |
2/2✓ Branch 0 taken 48 times.
✓ Branch 1 taken 24 times.
|
72 | for (int i = 0; i < mpi_size; ++i) { |
| 23 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 24 times.
|
48 | chunk_sizes[i] = base_chunk + (std::cmp_less(i, remainder) ? 1 : 0); |
| 24 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 24 times.
|
48 | offsets[i] = (i == 0) ? 0 : offsets[i - 1] + chunk_sizes[i - 1]; |
| 25 | } | ||
| 26 | 24 | } | |
| 27 | |||
| 28 | 24 | void ScatterData(const std::vector<double> &array, std::vector<double> &local_data, | |
| 29 | const std::vector<size_t> &chunk_sizes, const std::vector<size_t> &offsets) { | ||
| 30 | 24 | int mpi_size = static_cast<int>(chunk_sizes.size()); | |
| 31 | 24 | std::vector<int> send_counts(mpi_size); | |
| 32 |
1/4✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
24 | std::vector<int> send_displs(mpi_size); |
| 33 | |||
| 34 |
2/2✓ Branch 0 taken 48 times.
✓ Branch 1 taken 24 times.
|
72 | for (int i = 0; i < mpi_size; ++i) { |
| 35 | 48 | send_counts[i] = static_cast<int>(chunk_sizes[i]); | |
| 36 | 48 | send_displs[i] = static_cast<int>(offsets[i]); | |
| 37 | } | ||
| 38 | |||
| 39 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | MPI_Scatterv(array.data(), send_counts.data(), send_displs.data(), MPI_DOUBLE, local_data.data(), |
| 40 | static_cast<int>(local_data.size()), MPI_DOUBLE, 0, MPI_COMM_WORLD); | ||
| 41 | 24 | } | |
| 42 | |||
| 43 | 24 | std::vector<double> MergeTwoSorted(const std::vector<double> &left, const std::vector<double> &right) { | |
| 44 | 24 | std::vector<double> result(left.size() + right.size()); | |
| 45 | size_t i = 0; | ||
| 46 | size_t j = 0; | ||
| 47 | size_t k = 0; | ||
| 48 | |||
| 49 |
4/4✓ Branch 0 taken 1074 times.
✓ Branch 1 taken 12 times.
✓ Branch 2 taken 12 times.
✓ Branch 3 taken 1062 times.
|
1086 | while (i < left.size() && j < right.size()) { |
| 50 |
2/2✓ Branch 0 taken 531 times.
✓ Branch 1 taken 531 times.
|
1062 | if (left[i] <= right[j]) { |
| 51 | 531 | result[k++] = left[i++]; | |
| 52 | } else { | ||
| 53 | 531 | result[k++] = right[j++]; | |
| 54 | } | ||
| 55 | } | ||
| 56 | |||
| 57 |
2/2✓ Branch 0 taken 27 times.
✓ Branch 1 taken 24 times.
|
51 | while (i < left.size()) { |
| 58 | 27 | result[k++] = left[i++]; | |
| 59 | } | ||
| 60 | |||
| 61 |
2/2✓ Branch 0 taken 27 times.
✓ Branch 1 taken 24 times.
|
51 | while (j < right.size()) { |
| 62 | 27 | result[k++] = right[j++]; | |
| 63 | } | ||
| 64 | |||
| 65 | 24 | return result; | |
| 66 | } | ||
| 67 | |||
| 68 | 24 | void ExchangeAndMerge(int partner, std::vector<double> &merged_data) { | |
| 69 | 24 | size_t my_size = merged_data.size(); | |
| 70 | 24 | size_t partner_size = 0; | |
| 71 | |||
| 72 | 24 | MPI_Sendrecv(&my_size, 1, MPI_UNSIGNED_LONG, partner, 0, &partner_size, 1, MPI_UNSIGNED_LONG, partner, 0, | |
| 73 | MPI_COMM_WORLD, MPI_STATUS_IGNORE); | ||
| 74 | |||
| 75 | 24 | std::vector<double> partner_data(partner_size); | |
| 76 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | MPI_Sendrecv(merged_data.data(), static_cast<int>(my_size), MPI_DOUBLE, partner, 1, partner_data.data(), |
| 77 | static_cast<int>(partner_size), MPI_DOUBLE, partner, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE); | ||
| 78 | |||
| 79 |
3/6✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 23 times.
✓ Branch 4 taken 1 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
48 | merged_data = MergeTwoSorted(merged_data, partner_data); |
| 80 | 24 | } | |
| 81 | |||
| 82 | void ParallelHypercubeMerge(std::vector<double> &merged_data, int mpi_rank, int mpi_size) { | ||
| 83 | int step = 1; | ||
| 84 | |||
| 85 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 24 times.
|
48 | while (step < mpi_size) { |
| 86 | 24 | int partner = mpi_rank ^ step; | |
| 87 | |||
| 88 |
1/2✓ Branch 0 taken 24 times.
✗ Branch 1 not taken.
|
24 | if (partner < mpi_size) { |
| 89 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | ExchangeAndMerge(partner, merged_data); |
| 90 | } | ||
| 91 | |||
| 92 | 24 | step <<= 1; | |
| 93 | } | ||
| 94 | } | ||
| 95 | |||
| 96 | } // namespace | ||
| 97 | |||
| 98 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | MityaevaRadixAll::MityaevaRadixAll(const InType &in) { |
| 99 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 100 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | GetInput() = in; |
| 101 | GetOutput() = {}; | ||
| 102 | 24 | } | |
| 103 | |||
| 104 | 24 | bool MityaevaRadixAll::ValidationImpl() { | |
| 105 | 24 | return !GetInput().empty(); | |
| 106 | } | ||
| 107 | |||
| 108 | 24 | bool MityaevaRadixAll::PreProcessingImpl() { | |
| 109 | 24 | return true; | |
| 110 | } | ||
| 111 | |||
| 112 | 24 | bool MityaevaRadixAll::RunImpl() { | |
| 113 | 24 | int mpi_rank = 0; | |
| 114 | 24 | int mpi_size = 1; | |
| 115 | |||
| 116 | 24 | MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank); | |
| 117 | 24 | MPI_Comm_size(MPI_COMM_WORLD, &mpi_size); | |
| 118 | |||
| 119 | auto &array = GetInput(); | ||
| 120 | size_t total_size = array.size(); | ||
| 121 | |||
| 122 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 24 times.
|
24 | if (total_size == 0) { |
| 123 | ✗ | GetOutput() = array; | |
| 124 | ✗ | return true; | |
| 125 | } | ||
| 126 | |||
| 127 | 24 | std::vector<size_t> chunk_sizes(mpi_size); | |
| 128 |
1/4✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
24 | std::vector<size_t> offsets(mpi_size); |
| 129 | 24 | ComputeChunkParams(total_size, mpi_size, chunk_sizes, offsets); | |
| 130 | |||
| 131 |
1/4✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
24 | std::vector<double> local_data(chunk_sizes[mpi_rank]); |
| 132 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | ScatterData(array, local_data, chunk_sizes, offsets); |
| 133 | |||
| 134 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | SorterOmp::Sort(local_data); |
| 135 | |||
| 136 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 24 times.
|
24 | if (mpi_size == 1) { |
| 137 | ✗ | GetOutput() = local_data; | |
| 138 | return true; | ||
| 139 | } | ||
| 140 | |||
| 141 | std::vector<double> merged_data = std::move(local_data); | ||
| 142 | 24 | ParallelHypercubeMerge(merged_data, mpi_rank, mpi_size); | |
| 143 | |||
| 144 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | if (mpi_rank == 0) { |
| 145 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | GetOutput() = merged_data; |
| 146 | } else { | ||
| 147 | GetOutput() = {}; | ||
| 148 | } | ||
| 149 | |||
| 150 | return true; | ||
| 151 | } | ||
| 152 | |||
| 153 | 24 | bool MityaevaRadixAll::PostProcessingImpl() { | |
| 154 | 24 | return true; | |
| 155 | } | ||
| 156 | |||
| 157 | } // namespace mityaeva_radix | ||
| 158 |