| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "lopatin_a_sobel_operator/all/include/ops_all.hpp" | ||
| 2 | |||
| 3 | #include <mpi.h> | ||
| 4 | #include <omp.h> | ||
| 5 | |||
| 6 | #include <algorithm> | ||
| 7 | #include <array> | ||
| 8 | #include <cmath> | ||
| 9 | #include <cstddef> | ||
| 10 | #include <cstdint> | ||
| 11 | #include <utility> | ||
| 12 | #include <vector> | ||
| 13 | |||
| 14 | #include "lopatin_a_sobel_operator/common/include/common.hpp" | ||
| 15 | #include "util/include/util.hpp" | ||
| 16 | |||
| 17 | namespace lopatin_a_sobel_operator { | ||
| 18 | |||
| 19 | const std::array<std::array<int, 3>, 3> kSobelX = {std::array<int, 3>{-1, 0, 1}, std::array<int, 3>{-2, 0, 2}, | ||
| 20 | std::array<int, 3>{-1, 0, 1}}; | ||
| 21 | |||
| 22 | const std::array<std::array<int, 3>, 3> kSobelY = {std::array<int, 3>{-1, -2, -1}, std::array<int, 3>{0, 0, 0}, | ||
| 23 | std::array<int, 3>{1, 2, 1}}; | ||
| 24 | |||
| 25 | 7 | void LopatinASobelOperatorALL::RunSobel(const Image &img, std::size_t start, std::size_t end, | |
| 26 | lopatin_a_sobel_operator::OutType &output, std::size_t shift) { | ||
| 27 | 7 | const auto &input_data = img.pixels; | |
| 28 | |||
| 29 | 7 | int proc_num = 0; | |
| 30 | 7 | MPI_Comm_size(MPI_COMM_WORLD, &proc_num); | |
| 31 | |||
| 32 | 7 | #pragma omp parallel for num_threads(ppc::util::GetNumThreads() / proc_num) schedule(static) default(none) \ | |
| 33 | shared(kSobelX, kSobelY, img, input_data, output, start, end, shift) | ||
| 34 | for (std::size_t j = start; j < end; ++j) { // processing only pixels with a full 3 x 3 neighborhood size | ||
| 35 | std::size_t out_row = j - shift; | ||
| 36 | for (std::size_t i = 1; i < img.width - 1; ++i) { | ||
| 37 | int gx = 0; | ||
| 38 | int gy = 0; | ||
| 39 | |||
| 40 | for (int ky = -1; ky <= 1; ++ky) { | ||
| 41 | for (int kx = -1; kx <= 1; ++kx) { | ||
| 42 | std::uint8_t pixel = input_data[((j + ky) * img.width) + (i + kx)]; | ||
| 43 | gx += pixel * kSobelX.at(ky + 1).at(kx + 1); | ||
| 44 | gy += pixel * kSobelY.at(ky + 1).at(kx + 1); | ||
| 45 | } | ||
| 46 | } | ||
| 47 | |||
| 48 | auto magnitude = static_cast<int>(round(std::sqrt((gx * gx) + (gy * gy)))); | ||
| 49 | output[(out_row * img.width) + i] = (magnitude > img.threshold) ? magnitude : 0; | ||
| 50 | } | ||
| 51 | } | ||
| 52 | 7 | } | |
| 53 | |||
| 54 |
1/2✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 | LopatinASobelOperatorALL::LopatinASobelOperatorALL(const InType &in) : h_(in.height), w_(in.width) { |
| 55 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 56 | GetInput() = in; | ||
| 57 | GetOutput().clear(); | ||
| 58 | 6 | } | |
| 59 | |||
| 60 | 6 | bool LopatinASobelOperatorALL::ValidationImpl() { | |
| 61 | const auto &input = GetInput(); | ||
| 62 | 6 | return h_ * w_ == input.pixels.size(); | |
| 63 | } | ||
| 64 | |||
| 65 | 6 | bool LopatinASobelOperatorALL::PreProcessingImpl() { | |
| 66 | 6 | GetOutput().resize(h_ * w_); | |
| 67 | 6 | return true; | |
| 68 | } | ||
| 69 | |||
| 70 | 6 | bool LopatinASobelOperatorALL::RunImpl() { | |
| 71 | 6 | int proc_num = 0; | |
| 72 | 6 | MPI_Comm_size(MPI_COMM_WORLD, &proc_num); | |
| 73 | 6 | int proc_rank = 0; | |
| 74 | 6 | MPI_Comm_rank(MPI_COMM_WORLD, &proc_rank); | |
| 75 | |||
| 76 | const auto &input = GetInput(); | ||
| 77 | |||
| 78 | 6 | const auto rows = input.height - 2; | |
| 79 | 6 | const auto n_procs = std::min(static_cast<std::size_t>(proc_num), rows); | |
| 80 | |||
| 81 | 6 | const std::size_t portion = rows / n_procs; | |
| 82 | |||
| 83 | 6 | lopatin_a_sobel_operator::OutType local_output(portion * input.width, 0); | |
| 84 | |||
| 85 |
1/2✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
|
6 | if (std::cmp_less(proc_rank, n_procs)) { |
| 86 | 5 | std::size_t start = 1 + (proc_rank * portion); | |
| 87 | 5 | std::size_t end = 1 + ((proc_rank + 1) * portion); | |
| 88 |
1/2✓ Branch 1 taken 5 times.
✗ Branch 2 not taken.
|
5 | RunSobel(input, start, end, local_output, start); |
| 89 | } | ||
| 90 | |||
| 91 | auto &output = GetOutput(); | ||
| 92 | |||
| 93 |
1/4✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
6 | std::vector<int> recvcounts(proc_num, 0); |
| 94 |
1/4✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
6 | std::vector<int> displs(proc_num, 0); |
| 95 | |||
| 96 |
2/2✓ Branch 0 taken 10 times.
✓ Branch 1 taken 6 times.
|
16 | for (size_t i = 0; i < n_procs; ++i) { |
| 97 | 10 | recvcounts[i] = static_cast<int>(portion * input.width); | |
| 98 | 10 | displs[i] = static_cast<int>((1 + i * portion) * input.width); | |
| 99 | } | ||
| 100 | |||
| 101 |
1/2✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
|
11 | int sendcount = std::cmp_less(proc_rank, n_procs) ? static_cast<int>(local_output.size()) : 0; |
| 102 | |||
| 103 |
1/2✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 | MPI_Gatherv(local_output.data(), sendcount, MPI_INT, output.data(), recvcounts.data(), displs.data(), MPI_INT, 0, |
| 104 | MPI_COMM_WORLD); | ||
| 105 | |||
| 106 |
2/2✓ Branch 0 taken 3 times.
✓ Branch 1 taken 3 times.
|
6 | if (proc_rank == 0) { |
| 107 | 3 | const std::size_t tail = rows % n_procs; | |
| 108 | |||
| 109 |
2/2✓ Branch 0 taken 2 times.
✓ Branch 1 taken 1 times.
|
3 | if (tail != 0) { |
| 110 | 2 | std::size_t start = input.height - 1 - tail; | |
| 111 |
1/2✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
|
2 | RunSobel(input, start, input.height - 1, output, 0); |
| 112 | } | ||
| 113 | } | ||
| 114 | |||
| 115 |
1/2✓ Branch 1 taken 6 times.
✗ Branch 2 not taken.
|
6 | MPI_Bcast(output.data(), static_cast<int>(output.size()), MPI_INT, 0, MPI_COMM_WORLD); |
| 116 | |||
| 117 | 6 | return true; | |
| 118 | } | ||
| 119 | |||
| 120 | 6 | bool LopatinASobelOperatorALL::PostProcessingImpl() { | |
| 121 | 6 | return true; | |
| 122 | } | ||
| 123 | |||
| 124 | } // namespace lopatin_a_sobel_operator | ||
| 125 |