| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "rozenberg_a_radix_simple_merge/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <array> | ||
| 7 | #include <cstddef> | ||
| 8 | #include <cstdint> | ||
| 9 | #include <cstring> | ||
| 10 | #include <utility> | ||
| 11 | #include <vector> | ||
| 12 | |||
| 13 | #include "rozenberg_a_radix_simple_merge/common/include/common.hpp" | ||
| 14 | |||
| 15 | namespace rozenberg_a_radix_simple_merge { | ||
| 16 | |||
| 17 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 23 times.
|
24 | void RozenbergARadixSimpleMergeMPI::LocalRadixSort(InType &data) { |
| 18 |
2/2✓ Branch 0 taken 1 times.
✓ Branch 1 taken 23 times.
|
24 | if (data.empty()) { |
| 19 | 1 | return; | |
| 20 | } | ||
| 21 | size_t n = data.size(); | ||
| 22 | 23 | InType buffer(n); | |
| 23 | |||
| 24 |
2/2✓ Branch 0 taken 184 times.
✓ Branch 1 taken 23 times.
|
207 | for (int shift = 0; shift < 64; shift += 8) { |
| 25 | 184 | std::array<size_t, 256> count = {0}; | |
| 26 | |||
| 27 |
4/4✓ Branch 0 taken 576 times.
✓ Branch 1 taken 2200 times.
✓ Branch 2 taken 2776 times.
✓ Branch 3 taken 184 times.
|
2960 | for (double val : data) { |
| 28 | uint64_t u = 0; | ||
| 29 | std::memcpy(&u, &val, sizeof(double)); | ||
| 30 | |||
| 31 |
2/2✓ Branch 0 taken 576 times.
✓ Branch 1 taken 2200 times.
|
2776 | u = (u >> 63 != 0U) ? ~u : (u ^ 0x8000000000000000); |
| 32 | |||
| 33 | 2776 | auto byte = static_cast<uint8_t>((u >> shift) & 0xFF); | |
| 34 | 2776 | count.at(byte)++; | |
| 35 | } | ||
| 36 | |||
| 37 |
2/2✓ Branch 0 taken 46920 times.
✓ Branch 1 taken 184 times.
|
47104 | for (int i = 1; i < 256; ++i) { |
| 38 | 46920 | count.at(i) += count.at(i - 1); | |
| 39 | } | ||
| 40 | |||
| 41 |
2/2✓ Branch 0 taken 2776 times.
✓ Branch 1 taken 184 times.
|
2960 | for (int i = static_cast<int>(n) - 1; i >= 0; --i) { |
| 42 | uint64_t u = 0; | ||
| 43 |
2/2✓ Branch 0 taken 576 times.
✓ Branch 1 taken 2200 times.
|
2776 | std::memcpy(&u, &data[i], 8); |
| 44 | |||
| 45 |
2/2✓ Branch 0 taken 576 times.
✓ Branch 1 taken 2200 times.
|
2776 | u = (u >> 63 != 0U) ? ~u : (u ^ 0x8000000000000000); |
| 46 | |||
| 47 | 2776 | auto byte = static_cast<uint8_t>((u >> shift) & 0xFF); | |
| 48 | 2776 | buffer[--count.at(byte)] = data[i]; | |
| 49 | } | ||
| 50 |
1/2✓ Branch 1 taken 184 times.
✗ Branch 2 not taken.
|
184 | data = buffer; |
| 51 | } | ||
| 52 | } | ||
| 53 | |||
| 54 | 24 | void RozenbergARadixSimpleMergeMPI::ExchangeAndMerge(InType &local_buf, int rank, int size) { | |
| 55 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 12 times.
|
36 | for (int step = 1; step < size; step *= 2) { |
| 56 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | if (rank % (2 * step) == 0) { |
| 57 | 12 | int neighbor = rank + step; | |
| 58 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | if (neighbor < size) { |
| 59 | 12 | int neighbor_size = 0; | |
| 60 | 12 | MPI_Recv(&neighbor_size, 1, MPI_INT, neighbor, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE); | |
| 61 | |||
| 62 | 12 | std::vector<double> neighbor_data(neighbor_size); | |
| 63 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | MPI_Recv(neighbor_data.data(), neighbor_size, MPI_DOUBLE, neighbor, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE); |
| 64 | |||
| 65 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
|
12 | InType merged(local_buf.size() + neighbor_size); |
| 66 | 12 | std::ranges::merge(local_buf, neighbor_data, merged.begin()); | |
| 67 | local_buf = std::move(merged); | ||
| 68 | } | ||
| 69 | } else { | ||
| 70 | 12 | int target = rank - step; | |
| 71 | 12 | int my_size = static_cast<int>(local_buf.size()); | |
| 72 | 12 | MPI_Send(&my_size, 1, MPI_INT, target, 0, MPI_COMM_WORLD); | |
| 73 | 12 | MPI_Send(local_buf.data(), my_size, MPI_DOUBLE, target, 0, MPI_COMM_WORLD); | |
| 74 | local_buf.clear(); | ||
| 75 | break; | ||
| 76 | } | ||
| 77 | } | ||
| 78 | 24 | } | |
| 79 | |||
| 80 | 24 | RozenbergARadixSimpleMergeMPI::RozenbergARadixSimpleMergeMPI(const InType &in) { | |
| 81 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 82 | |||
| 83 | InType empty; | ||
| 84 | GetInput().swap(empty); | ||
| 85 | |||
| 86 |
2/2✓ Branch 0 taken 347 times.
✓ Branch 1 taken 24 times.
|
371 | for (const auto &elem : in) { |
| 87 | GetInput().push_back(elem); | ||
| 88 | } | ||
| 89 | |||
| 90 | GetOutput().clear(); | ||
| 91 | 24 | } | |
| 92 | |||
| 93 | 24 | bool RozenbergARadixSimpleMergeMPI::ValidationImpl() { | |
| 94 | 24 | int rank = 0; | |
| 95 | 24 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 96 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | if (rank == 0) { |
| 97 |
2/4✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
|
12 | return (!(GetInput().empty())) && (GetOutput().empty()); |
| 98 | } | ||
| 99 | return true; | ||
| 100 | } | ||
| 101 | |||
| 102 | 24 | bool RozenbergARadixSimpleMergeMPI::PreProcessingImpl() { | |
| 103 | 24 | int rank = 0; | |
| 104 | 24 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 105 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | if (rank == 0) { |
| 106 | 12 | GetOutput().resize(GetInput().size()); | |
| 107 | } | ||
| 108 | 24 | return true; | |
| 109 | } | ||
| 110 | |||
| 111 | 24 | bool RozenbergARadixSimpleMergeMPI::RunImpl() { | |
| 112 | 24 | int rank = 0; | |
| 113 | 24 | int size = 0; | |
| 114 | 24 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 115 | 24 | MPI_Comm_size(MPI_COMM_WORLD, &size); | |
| 116 | |||
| 117 | 24 | int n = 0; | |
| 118 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | if (rank == 0) { |
| 119 | 12 | n = static_cast<int>(GetInput().size()); | |
| 120 | } | ||
| 121 | 24 | MPI_Bcast(&n, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 122 | |||
| 123 | 24 | std::vector<int> sendcounts(size); | |
| 124 |
1/4✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
24 | std::vector<int> displs(size); |
| 125 | |||
| 126 | int sum = 0; | ||
| 127 |
2/2✓ Branch 0 taken 48 times.
✓ Branch 1 taken 24 times.
|
72 | for (int i = 0; i < size; i++) { |
| 128 |
2/2✓ Branch 0 taken 34 times.
✓ Branch 1 taken 14 times.
|
82 | sendcounts[i] = (n / size) + (i < (n % size) ? 1 : 0); |
| 129 | 48 | displs[i] = sum; | |
| 130 | 48 | sum += sendcounts[i]; | |
| 131 | } | ||
| 132 | |||
| 133 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | int chunk = sendcounts[rank]; |
| 134 |
2/6✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 24 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
|
24 | InType local_buf(static_cast<size_t>(chunk)); |
| 135 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | MPI_Scatterv(GetInput().data(), sendcounts.data(), displs.data(), MPI_DOUBLE, local_buf.data(), chunk, MPI_DOUBLE, 0, |
| 136 | MPI_COMM_WORLD); | ||
| 137 | |||
| 138 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | LocalRadixSort(local_buf); |
| 139 | |||
| 140 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | ExchangeAndMerge(local_buf, rank, size); |
| 141 | |||
| 142 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | if (rank == 0) { |
| 143 | GetOutput() = std::move(local_buf); | ||
| 144 | } | ||
| 145 | 24 | return true; | |
| 146 | } | ||
| 147 | |||
| 148 | 24 | bool RozenbergARadixSimpleMergeMPI::PostProcessingImpl() { | |
| 149 | 24 | return true; | |
| 150 | } | ||
| 151 | |||
| 152 | } // namespace rozenberg_a_radix_simple_merge | ||
| 153 |