| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "dergynov_s_radix_sort_double_simple_merge/mpi/include/ops_mpi.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <cstddef> | ||
| 6 | #include <cstdint> | ||
| 7 | #include <cstring> | ||
| 8 | #include <utility> | ||
| 9 | #include <vector> | ||
| 10 | |||
| 11 | #include "dergynov_s_radix_sort_double_simple_merge/common/include/common.hpp" | ||
| 12 | |||
| 13 | namespace dergynov_s_radix_sort_double_simple_merge { | ||
| 14 | namespace { | ||
| 15 | |||
| 16 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 10 times.
|
14 | void RadixSortDoubles(std::vector<double> &data) { |
| 17 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 10 times.
|
14 | if (data.size() <= 1) { |
| 18 | 4 | return; | |
| 19 | } | ||
| 20 | |||
| 21 | 10 | std::vector<uint64_t> keys(data.size()); | |
| 22 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 10 times.
|
42 | for (size_t i = 0; i < data.size(); ++i) { |
| 23 |
2/2✓ Branch 0 taken 9 times.
✓ Branch 1 taken 23 times.
|
64 | keys[i] = DoubleToSortableUint64(data[i]); |
| 24 | } | ||
| 25 | |||
| 26 | const int k_radix = 256; | ||
| 27 |
1/4✓ Branch 1 taken 10 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
10 | std::vector<uint64_t> temp(data.size()); |
| 28 | |||
| 29 |
2/2✓ Branch 0 taken 80 times.
✓ Branch 1 taken 10 times.
|
90 | for (int shift = 0; shift < 64; shift += 8) { |
| 30 |
1/4✓ Branch 1 taken 80 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
80 | std::vector<size_t> count(k_radix + 1, 0); |
| 31 | |||
| 32 |
2/2✓ Branch 0 taken 256 times.
✓ Branch 1 taken 80 times.
|
336 | for (uint64_t key : keys) { |
| 33 | 256 | uint8_t digit = (key >> shift) & 0xFF; | |
| 34 | 256 | ++count[digit + 1]; | |
| 35 | } | ||
| 36 | |||
| 37 |
2/2✓ Branch 0 taken 20480 times.
✓ Branch 1 taken 80 times.
|
20560 | for (int i = 0; i < k_radix; ++i) { |
| 38 | 20480 | count[i + 1] += count[i]; | |
| 39 | } | ||
| 40 | |||
| 41 |
2/2✓ Branch 0 taken 256 times.
✓ Branch 1 taken 80 times.
|
336 | for (uint64_t key : keys) { |
| 42 | 256 | uint8_t digit = (key >> shift) & 0xFF; | |
| 43 | 256 | size_t pos = count[digit]; | |
| 44 | 256 | temp[pos] = key; | |
| 45 | 256 | ++count[digit]; | |
| 46 | } | ||
| 47 | |||
| 48 | keys.swap(temp); | ||
| 49 | } | ||
| 50 | |||
| 51 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 10 times.
|
42 | for (size_t i = 0; i < data.size(); ++i) { |
| 52 |
2/2✓ Branch 0 taken 23 times.
✓ Branch 1 taken 9 times.
|
64 | data[i] = SortableUint64ToDouble(keys[i]); |
| 53 | } | ||
| 54 | } | ||
| 55 | |||
| 56 | 7 | std::vector<double> MergeSorted(const std::vector<double> &a, const std::vector<double> &b) { | |
| 57 |
1/2✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
|
7 | std::vector<double> result; |
| 58 |
1/2✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
|
7 | result.reserve(a.size() + b.size()); |
| 59 | size_t i = 0; | ||
| 60 | size_t j = 0; | ||
| 61 |
4/4✓ Branch 0 taken 3 times.
✓ Branch 1 taken 22 times.
✓ Branch 2 taken 4 times.
✓ Branch 3 taken 18 times.
|
25 | while (i < a.size() && j < b.size()) { |
| 62 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 10 times.
|
18 | if (a[i] < b[j]) { |
| 63 |
1/2✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
|
8 | result.push_back(a[i++]); |
| 64 | } else { | ||
| 65 |
1/2✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
|
10 | result.push_back(b[j++]); |
| 66 | } | ||
| 67 | } | ||
| 68 |
2/2✓ Branch 0 taken 11 times.
✓ Branch 1 taken 7 times.
|
18 | while (i < a.size()) { |
| 69 |
1/2✓ Branch 0 taken 11 times.
✗ Branch 1 not taken.
|
11 | result.push_back(a[i++]); |
| 70 | } | ||
| 71 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 7 times.
|
11 | while (j < b.size()) { |
| 72 |
1/2✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
|
4 | result.push_back(b[j++]); |
| 73 | } | ||
| 74 | 7 | return result; | |
| 75 | } | ||
| 76 | |||
| 77 | } // namespace | ||
| 78 | |||
| 79 |
1/2✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 | DergynovSRadixSortDoubleSimpleMergeMPI::DergynovSRadixSortDoubleSimpleMergeMPI(const InType &in) { |
| 80 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 81 |
1/2✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 | GetInput() = in; |
| 82 | result_.clear(); | ||
| 83 | 14 | std::get<1>(GetOutput()) = -1; | |
| 84 | 14 | } | |
| 85 | |||
| 86 | 14 | bool DergynovSRadixSortDoubleSimpleMergeMPI::ValidationImpl() { | |
| 87 | 14 | int initialized = 0; | |
| 88 | 14 | MPI_Initialized(&initialized); | |
| 89 | 14 | return initialized != 0; | |
| 90 | } | ||
| 91 | |||
| 92 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 14 times.
|
14 | bool DergynovSRadixSortDoubleSimpleMergeMPI::PreProcessingImpl() { |
| 93 | result_.clear(); | ||
| 94 | 14 | return true; | |
| 95 | } | ||
| 96 | |||
| 97 | 14 | bool DergynovSRadixSortDoubleSimpleMergeMPI::RunImpl() { | |
| 98 | 14 | int rank = 0; | |
| 99 | 14 | int size = 0; | |
| 100 | 14 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 101 | 14 | MPI_Comm_size(MPI_COMM_WORLD, &size); | |
| 102 | |||
| 103 | const auto &input = GetInput(); | ||
| 104 | 14 | int n = static_cast<int>(input.size()); | |
| 105 | |||
| 106 | 14 | MPI_Bcast(&n, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 107 | |||
| 108 | 14 | std::vector<int> counts(size); | |
| 109 |
1/4✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
14 | std::vector<int> displs(size); |
| 110 | |||
| 111 | 14 | int base = n / size; | |
| 112 | 14 | int rem = n % size; | |
| 113 | int offset = 0; | ||
| 114 | |||
| 115 |
2/2✓ Branch 0 taken 28 times.
✓ Branch 1 taken 14 times.
|
42 | for (int i = 0; i < size; ++i) { |
| 116 |
2/2✓ Branch 0 taken 18 times.
✓ Branch 1 taken 10 times.
|
46 | counts[i] = base + (i < rem ? 1 : 0); |
| 117 | 28 | displs[i] = offset; | |
| 118 | 28 | offset += counts[i]; | |
| 119 | } | ||
| 120 | |||
| 121 |
1/4✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
14 | std::vector<double> local_data(counts[rank]); |
| 122 | |||
| 123 |
3/4✓ Branch 0 taken 7 times.
✓ Branch 1 taken 7 times.
✓ Branch 3 taken 14 times.
✗ Branch 4 not taken.
|
21 | MPI_Scatterv(rank == 0 ? input.data() : nullptr, counts.data(), displs.data(), MPI_DOUBLE, local_data.data(), |
| 124 |
2/2✓ Branch 0 taken 7 times.
✓ Branch 1 taken 7 times.
|
14 | counts[rank], MPI_DOUBLE, 0, MPI_COMM_WORLD); |
| 125 | |||
| 126 |
1/2✓ Branch 1 taken 14 times.
✗ Branch 2 not taken.
|
14 | RadixSortDoubles(local_data); |
| 127 | |||
| 128 |
2/2✓ Branch 0 taken 7 times.
✓ Branch 1 taken 7 times.
|
14 | if (rank == 0) { |
| 129 | 7 | result_ = std::move(local_data); | |
| 130 | |||
| 131 |
2/2✓ Branch 0 taken 7 times.
✓ Branch 1 taken 7 times.
|
14 | for (int proc = 1; proc < size; ++proc) { |
| 132 | 7 | int recv_count = 0; | |
| 133 |
1/2✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
|
7 | MPI_Recv(&recv_count, 1, MPI_INT, proc, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE); |
| 134 |
1/4✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
7 | std::vector<double> part(recv_count); |
| 135 |
1/2✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
|
7 | MPI_Recv(part.data(), recv_count, MPI_DOUBLE, proc, 1, MPI_COMM_WORLD, MPI_STATUS_IGNORE); |
| 136 |
3/6✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 5 times.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
14 | result_ = MergeSorted(result_, part); |
| 137 | } | ||
| 138 | } else { | ||
| 139 | 7 | int send_count = static_cast<int>(local_data.size()); | |
| 140 |
1/2✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
|
7 | MPI_Send(&send_count, 1, MPI_INT, 0, 0, MPI_COMM_WORLD); |
| 141 |
1/2✓ Branch 1 taken 7 times.
✗ Branch 2 not taken.
|
7 | MPI_Send(local_data.data(), send_count, MPI_DOUBLE, 0, 1, MPI_COMM_WORLD); |
| 142 | } | ||
| 143 | |||
| 144 |
2/2✓ Branch 0 taken 5 times.
✓ Branch 1 taken 9 times.
|
14 | std::get<1>(GetOutput()) = rank; |
| 145 | 14 | return true; | |
| 146 | } | ||
| 147 | |||
| 148 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 8 times.
|
14 | bool DergynovSRadixSortDoubleSimpleMergeMPI::PostProcessingImpl() { |
| 149 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 8 times.
|
14 | if (!result_.empty()) { |
| 150 | 6 | std::get<0>(GetOutput()) = result_; | |
| 151 | } | ||
| 152 | 14 | return true; | |
| 153 | } | ||
| 154 | |||
| 155 | } // namespace dergynov_s_radix_sort_double_simple_merge | ||
| 156 |