| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "moskaev_v_lin_filt_block_gauss_3/omp/include/ops_omp.hpp" | ||
| 2 | |||
| 3 | #include <algorithm> | ||
| 4 | #include <cmath> | ||
| 5 | #include <cstddef> | ||
| 6 | #include <cstdint> | ||
| 7 | #include <vector> | ||
| 8 | |||
| 9 | #include "moskaev_v_lin_filt_block_gauss_3/common/include/common.hpp" | ||
| 10 | |||
| 11 | #ifdef _OPENMP | ||
| 12 | # include <omp.h> | ||
| 13 | #endif | ||
| 14 | |||
| 15 | namespace moskaev_v_lin_filt_block_gauss_3 { | ||
| 16 | |||
| 17 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | MoskaevVLinFiltBlockGauss3OMP::MoskaevVLinFiltBlockGauss3OMP(const InType &in) { |
| 18 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 19 | GetInput() = in; | ||
| 20 | 16 | GetOutput() = OutType(); | |
| 21 | 16 | } | |
| 22 | |||
| 23 | 16 | bool MoskaevVLinFiltBlockGauss3OMP::ValidationImpl() { | |
| 24 | const auto &input = GetInput(); | ||
| 25 | 16 | return !std::get<4>(input).empty(); | |
| 26 | } | ||
| 27 | |||
| 28 | 16 | bool MoskaevVLinFiltBlockGauss3OMP::PreProcessingImpl() { | |
| 29 | 16 | return true; | |
| 30 | } | ||
| 31 | |||
| 32 | ✗ | void MoskaevVLinFiltBlockGauss3OMP::ApplyGaussianFilterToBlock(const std::vector<uint8_t> &input_block, | |
| 33 | std::vector<uint8_t> &output_block, int block_width, | ||
| 34 | int block_height, int channels) { | ||
| 35 | ✗ | int inner_width = block_width - 2; | |
| 36 | ✗ | int inner_height = block_height - 2; | |
| 37 | |||
| 38 | ✗ | #pragma omp parallel for collapse(3) schedule(static) default(none) \ | |
| 39 | shared(input_block, output_block, inner_width, inner_height, channels, block_width, kGaussianKernel) | ||
| 40 | for (int row = 0; row < inner_height; ++row) { | ||
| 41 | for (int col = 0; col < inner_width; ++col) { | ||
| 42 | for (int channel = 0; channel < channels; ++channel) { | ||
| 43 | float sum = 0.0F; | ||
| 44 | |||
| 45 | for (int ky = -1; ky <= 1; ++ky) { | ||
| 46 | for (int kx = -1; kx <= 1; ++kx) { | ||
| 47 | int ny = row + 1 + ky; | ||
| 48 | int nx = col + 1 + kx; | ||
| 49 | |||
| 50 | int idx = (((ny * block_width) + nx) * channels) + channel; | ||
| 51 | sum += static_cast<float>(input_block[idx]) * kGaussianKernel[((ky + 1) * 3) + (kx + 1)]; | ||
| 52 | } | ||
| 53 | } | ||
| 54 | |||
| 55 | int out_idx = (((row * inner_width) + col) * channels) + channel; | ||
| 56 | output_block[out_idx] = static_cast<uint8_t>(std::round(sum)); | ||
| 57 | } | ||
| 58 | } | ||
| 59 | } | ||
| 60 | ✗ | } | |
| 61 | |||
| 62 | namespace { | ||
| 63 | void CopyBlockWithPadding(const std::vector<uint8_t> &source_image, std::vector<uint8_t> &padded_block, int width, | ||
| 64 | int height, int channels, int block_x, int block_y, int current_block_width, | ||
| 65 | int current_block_height, int block_with_padding_width) { | ||
| 66 | #pragma omp parallel for collapse(2) schedule(static) default(none) \ | ||
| 67 | shared(source_image, padded_block, width, height, channels, block_x, block_y, current_block_width, \ | ||
| 68 | current_block_height, block_with_padding_width) | ||
| 69 | for (int row = -1; row <= current_block_height; ++row) { | ||
| 70 | for (int col = -1; col <= current_block_width; ++col) { | ||
| 71 | int src_y = std::clamp(block_y + row, 0, height - 1); | ||
| 72 | int src_x = std::clamp(block_x + col, 0, width - 1); | ||
| 73 | int dst_y = row + 1; | ||
| 74 | int dst_x = col + 1; | ||
| 75 | |||
| 76 | for (int channel = 0; channel < channels; ++channel) { | ||
| 77 | int src_idx = (((src_y * width) + src_x) * channels) + channel; | ||
| 78 | int dst_idx = (((dst_y * block_with_padding_width) + dst_x) * channels) + channel; | ||
| 79 | padded_block[dst_idx] = source_image[src_idx]; | ||
| 80 | } | ||
| 81 | } | ||
| 82 | } | ||
| 83 | } | ||
| 84 | |||
| 85 | void CopyProcessedBlockToOutput(const std::vector<uint8_t> &processed_block, std::vector<uint8_t> &output_image, | ||
| 86 | int width, int channels, int block_x, int block_y, int current_block_width, | ||
| 87 | int current_block_height) { | ||
| 88 | #pragma omp parallel for collapse(2) schedule(static) default(none) shared( \ | ||
| 89 | processed_block, output_image, width, channels, block_x, block_y, current_block_width, current_block_height) | ||
| 90 | for (int row = 0; row < current_block_height; ++row) { | ||
| 91 | for (int col = 0; col < current_block_width; ++col) { | ||
| 92 | for (int channel = 0; channel < channels; ++channel) { | ||
| 93 | int src_idx = (((row * current_block_width) + col) * channels) + channel; | ||
| 94 | int dst_idx = ((((block_y + row) * width) + (block_x + col)) * channels) + channel; | ||
| 95 | output_image[dst_idx] = processed_block[src_idx]; | ||
| 96 | } | ||
| 97 | } | ||
| 98 | } | ||
| 99 | } | ||
| 100 | } // namespace | ||
| 101 | |||
| 102 | 16 | bool MoskaevVLinFiltBlockGauss3OMP::RunImpl() { | |
| 103 | const auto &input = GetInput(); | ||
| 104 | |||
| 105 | 16 | int width = std::get<0>(input); | |
| 106 | 16 | int height = std::get<1>(input); | |
| 107 |
1/2✓ Branch 0 taken 16 times.
✗ Branch 1 not taken.
|
16 | int channels = std::get<2>(input); |
| 108 | const auto &image_data = std::get<4>(input); | ||
| 109 | |||
| 110 |
1/2✓ Branch 0 taken 16 times.
✗ Branch 1 not taken.
|
16 | if (image_data.empty()) { |
| 111 | return false; | ||
| 112 | } | ||
| 113 | |||
| 114 | 16 | block_size_ = 64; | |
| 115 | 16 | int block_size = block_size_; | |
| 116 | |||
| 117 | 16 | GetOutput().resize(static_cast<size_t>(width) * static_cast<size_t>(height) * static_cast<size_t>(channels)); | |
| 118 | |||
| 119 | 16 | #pragma omp parallel for collapse(2) schedule(dynamic) default(none) \ | |
| 120 | shared(height, width, channels, image_data, block_size) | ||
| 121 | for (int block_y = 0; block_y < height; block_y += block_size) { | ||
| 122 | for (int block_x = 0; block_x < width; block_x += block_size) { | ||
| 123 | int current_block_width = std::min(block_size, width - block_x); | ||
| 124 | int current_block_height = std::min(block_size, height - block_y); | ||
| 125 | |||
| 126 | int block_with_padding_width = current_block_width + 2; | ||
| 127 | int block_with_padding_height = current_block_height + 2; | ||
| 128 | |||
| 129 | std::vector<uint8_t> input_block(static_cast<size_t>(block_with_padding_width) * | ||
| 130 | static_cast<size_t>(block_with_padding_height) * | ||
| 131 | static_cast<size_t>(channels), | ||
| 132 | 0); | ||
| 133 | |||
| 134 | std::vector<uint8_t> output_block(static_cast<size_t>(current_block_width) * | ||
| 135 | static_cast<size_t>(current_block_height) * static_cast<size_t>(channels), | ||
| 136 | 0); | ||
| 137 | |||
| 138 | CopyBlockWithPadding(image_data, input_block, width, height, channels, block_x, block_y, current_block_width, | ||
| 139 | current_block_height, block_with_padding_width); | ||
| 140 | |||
| 141 | ApplyGaussianFilterToBlock(input_block, output_block, block_with_padding_width, block_with_padding_height, | ||
| 142 | channels); | ||
| 143 | |||
| 144 | CopyProcessedBlockToOutput(output_block, GetOutput(), width, channels, block_x, block_y, current_block_width, | ||
| 145 | current_block_height); | ||
| 146 | } | ||
| 147 | } | ||
| 148 | |||
| 149 | 16 | return true; | |
| 150 | } | ||
| 151 | |||
| 152 | 16 | bool MoskaevVLinFiltBlockGauss3OMP::PostProcessingImpl() { | |
| 153 | 16 | return !GetOutput().empty(); | |
| 154 | } | ||
| 155 | |||
| 156 | } // namespace moskaev_v_lin_filt_block_gauss_3 | ||
| 157 |