| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "shakirova_e_sobel_edge_detection/all/include/ops_all.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | |||
| 5 | #include <algorithm> | ||
| 6 | #include <cmath> | ||
| 7 | #include <cstddef> | ||
| 8 | #include <thread> | ||
| 9 | #include <vector> | ||
| 10 | |||
| 11 | #include "shakirova_e_sobel_edge_detection/common/include/common.hpp" | ||
| 12 | #include "util/include/util.hpp" | ||
| 13 | |||
| 14 | namespace shakirova_e_sobel_edge_detection { | ||
| 15 | |||
| 16 | namespace { | ||
| 17 | |||
| 18 | 24 | void ProcessRowsALL(const int *inp, int *out, int w, int row_begin, int row_end) { | |
| 19 |
2/2✓ Branch 0 taken 1561 times.
✓ Branch 1 taken 24 times.
|
1585 | for (int row = row_begin; row < row_end; ++row) { |
| 20 | 1561 | const int *prev = inp + static_cast<ptrdiff_t>((row)*w); | |
| 21 | 1561 | const int *curr = inp + static_cast<ptrdiff_t>((row + 1) * w); | |
| 22 | 1561 | const int *next = inp + static_cast<ptrdiff_t>((row + 2) * w); | |
| 23 | int *out_row = out + static_cast<ptrdiff_t>(row * w); | ||
| 24 | |||
| 25 |
2/2✓ Branch 0 taken 814589 times.
✓ Branch 1 taken 1561 times.
|
816150 | for (int col = 1; col < w - 1; ++col) { |
| 26 | 814589 | const int gx = | |
| 27 | 814589 | -prev[col - 1] + prev[col + 1] - (2 * curr[col - 1]) + (2 * curr[col + 1]) - next[col - 1] + next[col + 1]; | |
| 28 | 814589 | const int gy = -prev[col - 1] - (2 * prev[col]) - prev[col + 1] + next[col - 1] + (2 * next[col]) + next[col + 1]; | |
| 29 | |||
| 30 | 814589 | const int agx = std::abs(gx); | |
| 31 | 814589 | const int agy = std::abs(gy); | |
| 32 | 814589 | const int magnitude = (((std::max(agx, agy) * 123) + (std::min(agx, agy) * 51)) >> 7); | |
| 33 | 814589 | out_row[col] = magnitude > 255 ? 255 : magnitude; | |
| 34 | } | ||
| 35 | } | ||
| 36 | 24 | } | |
| 37 | |||
| 38 | 12 | void ComputeDistribution(int mpi_size, int inner_rows, std::vector<int> &row_counts, std::vector<int> &row_offsets) { | |
| 39 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 12 times.
|
36 | for (int i = 0; i < mpi_size; ++i) { |
| 40 |
4/4✓ Branch 0 taken 22 times.
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 12 times.
✓ Branch 3 taken 12 times.
|
46 | row_counts[i] = (inner_rows / mpi_size) + (i < inner_rows % mpi_size ? 1 : 0); |
| 41 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | row_offsets[i] = (i == 0) ? 0 : row_offsets[i - 1] + row_counts[i - 1]; |
| 42 | } | ||
| 43 | 12 | } | |
| 44 | |||
| 45 | 12 | void ComputeScatterParams(int mpi_size, int w, const std::vector<int> &row_counts, const std::vector<int> &row_offsets, | |
| 46 | std::vector<int> &send_counts, std::vector<int> &send_displs) { | ||
| 47 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 12 times.
|
36 | for (int i = 0; i < mpi_size; ++i) { |
| 48 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 24 times.
|
24 | if (row_counts[i] == 0) { |
| 49 | ✗ | send_counts[i] = 0; | |
| 50 | ✗ | send_displs[i] = (i == 0) ? 0 : send_displs[i - 1]; | |
| 51 | } else { | ||
| 52 | 24 | send_counts[i] = (row_counts[i] + 2) * w; | |
| 53 | 24 | send_displs[i] = row_offsets[i] * w; | |
| 54 | } | ||
| 55 | } | ||
| 56 | 12 | } | |
| 57 | |||
| 58 | 12 | void RunThreads(const int *lb, int *lo, int w, int local_inner, int num_stl) { | |
| 59 | 12 | std::vector<std::thread> threads; | |
| 60 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | threads.reserve(static_cast<size_t>(num_stl)); |
| 61 | |||
| 62 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 12 times.
|
36 | for (int tid = 0; tid < num_stl; ++tid) { |
| 63 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | threads.emplace_back([lb, lo, w, local_inner, num_stl, tid]() { |
| 64 | 24 | const int chunk = local_inner / num_stl; | |
| 65 | 24 | const int row_begin = (tid * chunk); | |
| 66 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 12 times.
|
24 | const int row_end = (tid == num_stl - 1) ? local_inner : row_begin + chunk; |
| 67 | 24 | ProcessRowsALL(lb, lo, w, row_begin, row_end); | |
| 68 | 24 | }); | |
| 69 | } | ||
| 70 | |||
| 71 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 12 times.
|
36 | for (auto &th : threads) { |
| 72 |
1/2✓ Branch 1 taken 24 times.
✗ Branch 2 not taken.
|
24 | th.join(); |
| 73 | } | ||
| 74 | 12 | } | |
| 75 | |||
| 76 | } // namespace | ||
| 77 | |||
| 78 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | ShakirovaESobelEdgeDetectionALL::ShakirovaESobelEdgeDetectionALL(const InType &in) { |
| 79 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 80 | GetInput() = in; | ||
| 81 | GetOutput().clear(); | ||
| 82 | 12 | } | |
| 83 | |||
| 84 | 12 | bool ShakirovaESobelEdgeDetectionALL::ValidationImpl() { | |
| 85 | 12 | int initialized = 0; | |
| 86 | 12 | MPI_Initialized(&initialized); | |
| 87 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | if (initialized != 0) { |
| 88 | 12 | int rank = 0; | |
| 89 | 12 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 90 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 6 times.
|
12 | if (rank == 0) { |
| 91 | return GetInput().IsValid(); | ||
| 92 | } | ||
| 93 | return true; | ||
| 94 | } | ||
| 95 | return GetInput().IsValid(); | ||
| 96 | } | ||
| 97 | |||
| 98 | 12 | bool ShakirovaESobelEdgeDetectionALL::PreProcessingImpl() { | |
| 99 | 12 | int rank = 0; | |
| 100 | 12 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 101 | |||
| 102 | 12 | int w = 0; | |
| 103 | 12 | int h = 0; | |
| 104 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 6 times.
|
12 | if (rank == 0) { |
| 105 | 6 | w = GetInput().width; | |
| 106 | 6 | h = GetInput().height; | |
| 107 | } | ||
| 108 | 12 | MPI_Bcast(&w, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 109 | 12 | MPI_Bcast(&h, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 110 | |||
| 111 | 12 | GetOutput().assign(static_cast<size_t>(w) * static_cast<size_t>(h), 0); | |
| 112 | 12 | return true; | |
| 113 | } | ||
| 114 | |||
| 115 | 12 | bool ShakirovaESobelEdgeDetectionALL::RunImpl() { | |
| 116 | 12 | int rank = 0; | |
| 117 | 12 | int mpi_size = 1; | |
| 118 | 12 | MPI_Comm_rank(MPI_COMM_WORLD, &rank); | |
| 119 | 12 | MPI_Comm_size(MPI_COMM_WORLD, &mpi_size); | |
| 120 | |||
| 121 | 12 | int w = 0; | |
| 122 | 12 | int h = 0; | |
| 123 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 6 times.
|
12 | if (rank == 0) { |
| 124 | 6 | w = GetInput().width; | |
| 125 | 6 | h = GetInput().height; | |
| 126 | } | ||
| 127 | 12 | MPI_Bcast(&w, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 128 | 12 | MPI_Bcast(&h, 1, MPI_INT, 0, MPI_COMM_WORLD); | |
| 129 | |||
| 130 |
2/4✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
|
12 | if (h < 3 || w < 3) { |
| 131 | return true; | ||
| 132 | } | ||
| 133 | |||
| 134 | 12 | const int inner_rows = h - 2; | |
| 135 | 12 | const int num_stl = ppc::util::GetNumThreads(); | |
| 136 | |||
| 137 | 12 | std::vector<int> row_counts(mpi_size); | |
| 138 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
12 | std::vector<int> row_offsets(mpi_size); |
| 139 | 12 | ComputeDistribution(mpi_size, inner_rows, row_counts, row_offsets); | |
| 140 | |||
| 141 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | const int local_inner = row_counts[rank]; |
| 142 | |||
| 143 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
12 | std::vector<int> send_counts(mpi_size); |
| 144 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
12 | std::vector<int> send_displs(mpi_size); |
| 145 | 12 | ComputeScatterParams(mpi_size, w, row_counts, row_offsets, send_counts, send_displs); | |
| 146 | |||
| 147 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | const int local_buf_rows = (local_inner > 0) ? (local_inner + 2) : 0; |
| 148 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
12 | std::vector<int> local_buf(static_cast<size_t>(local_buf_rows) * static_cast<size_t>(w), 0); |
| 149 | |||
| 150 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 6 times.
|
12 | const int *inp_ptr = (rank == 0) ? GetInput().pixels.data() : nullptr; |
| 151 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | MPI_Scatterv(inp_ptr, send_counts.data(), send_displs.data(), MPI_INT, local_buf.data(), local_buf_rows * w, MPI_INT, |
| 152 | 0, MPI_COMM_WORLD); | ||
| 153 | |||
| 154 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
12 | std::vector<int> local_out(static_cast<size_t>(local_inner) * static_cast<size_t>(w), 0); |
| 155 | |||
| 156 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | if (local_inner > 0) { |
| 157 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | RunThreads(local_buf.data(), local_out.data(), w, local_inner, num_stl); |
| 158 | } | ||
| 159 | |||
| 160 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
12 | std::vector<int> recv_counts(mpi_size); |
| 161 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
12 | std::vector<int> recv_displs(mpi_size); |
| 162 |
2/2✓ Branch 0 taken 24 times.
✓ Branch 1 taken 12 times.
|
36 | for (int i = 0; i < mpi_size; ++i) { |
| 163 | 24 | recv_counts[i] = row_counts[i] * w; | |
| 164 | 24 | recv_displs[i] = (row_offsets[i] + 1) * w; | |
| 165 | } | ||
| 166 | |||
| 167 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | MPI_Gatherv(local_out.data(), local_inner * w, MPI_INT, GetOutput().data(), recv_counts.data(), recv_displs.data(), |
| 168 | MPI_INT, 0, MPI_COMM_WORLD); | ||
| 169 | |||
| 170 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | MPI_Bcast(GetOutput().data(), static_cast<int>(GetOutput().size()), MPI_INT, 0, MPI_COMM_WORLD); |
| 171 | |||
| 172 | return true; | ||
| 173 | } | ||
| 174 | |||
| 175 | 12 | bool ShakirovaESobelEdgeDetectionALL::PostProcessingImpl() { | |
| 176 | 12 | return true; | |
| 177 | } | ||
| 178 | |||
| 179 | } // namespace shakirova_e_sobel_edge_detection | ||
| 180 |