| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "terekhov_d_gauss_vert/tbb/include/ops_tbb.hpp" | ||
| 2 | |||
| 3 | #include <oneapi/tbb/blocked_range.h> | ||
| 4 | #include <oneapi/tbb/parallel_for.h> | ||
| 5 | |||
| 6 | #include <algorithm> | ||
| 7 | #include <cmath> | ||
| 8 | #include <cstddef> | ||
| 9 | #include <vector> | ||
| 10 | |||
| 11 | #include "terekhov_d_gauss_vert/common/include/common.hpp" | ||
| 12 | |||
| 13 | namespace terekhov_d_gauss_vert { | ||
| 14 | |||
| 15 | namespace { | ||
| 16 | |||
| 17 | 5184 | inline void ProcessPixel(OutType &output, const std::vector<int> &padded_image, int padded_width, int width, int row, | |
| 18 | int col) { | ||
| 19 | 5184 | size_t idx = (static_cast<size_t>(row) * static_cast<size_t>(width)) + static_cast<size_t>(col); | |
| 20 | float sum = 0.0F; | ||
| 21 |
2/2✓ Branch 0 taken 15552 times.
✓ Branch 1 taken 5184 times.
|
20736 | for (int ky = -1; ky <= 1; ++ky) { |
| 22 |
2/2✓ Branch 0 taken 46656 times.
✓ Branch 1 taken 15552 times.
|
62208 | for (int kx = -1; kx <= 1; ++kx) { |
| 23 | 46656 | int px = col + kx + 1; | |
| 24 | 46656 | int py = row + ky + 1; | |
| 25 | 46656 | int kernel_idx = ((ky + 1) * 3) + (kx + 1); | |
| 26 | 46656 | size_t padded_idx = (static_cast<size_t>(py) * static_cast<size_t>(padded_width)) + static_cast<size_t>(px); | |
| 27 | 46656 | sum += static_cast<float>(padded_image[padded_idx]) * kGaussKernel[static_cast<size_t>(kernel_idx)]; | |
| 28 | } | ||
| 29 | } | ||
| 30 | 5184 | output.data[idx] = static_cast<int>(std::lround(sum)); | |
| 31 | 5184 | } | |
| 32 | |||
| 33 | 48 | inline void ProcessBand(OutType &output, const std::vector<int> &padded_image, int padded_width, int width, int height, | |
| 34 | int band, int band_width, int num_bands) { | ||
| 35 | 48 | int start_x = band * band_width; | |
| 36 |
2/2✓ Branch 0 taken 36 times.
✓ Branch 1 taken 12 times.
|
48 | int end_x = (band == num_bands - 1) ? width : ((band + 1) * band_width); |
| 37 |
2/2✓ Branch 0 taken 832 times.
✓ Branch 1 taken 48 times.
|
880 | for (int row = 0; row < height; ++row) { |
| 38 |
2/2✓ Branch 0 taken 5184 times.
✓ Branch 1 taken 832 times.
|
6016 | for (int col = start_x; col < end_x; ++col) { |
| 39 | 5184 | ProcessPixel(output, padded_image, padded_width, width, row, col); | |
| 40 | } | ||
| 41 | } | ||
| 42 | 48 | } | |
| 43 | |||
| 44 | 12 | inline OutType SolveTBB(const std::vector<int> &padded_image, int width, int height) { | |
| 45 | 12 | OutType output; | |
| 46 | 12 | output.width = width; | |
| 47 | 12 | output.height = height; | |
| 48 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | output.data.resize(static_cast<size_t>(width) * static_cast<size_t>(height)); |
| 49 | |||
| 50 | 12 | const int padded_width = width + 2; | |
| 51 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | const int band_width = std::max(width / 4, 1); |
| 52 | const int num_bands = 4; | ||
| 53 | |||
| 54 | 12 | oneapi::tbb::parallel_for(oneapi::tbb::blocked_range<int>(0, num_bands), | |
| 55 |
1/4✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
60 | [&](const oneapi::tbb::blocked_range<int> &range) { |
| 56 |
2/2✓ Branch 0 taken 48 times.
✓ Branch 1 taken 48 times.
|
96 | for (int band = range.begin(); band != range.end(); ++band) { |
| 57 | 48 | ProcessBand(output, padded_image, padded_width, width, height, band, band_width, num_bands); | |
| 58 | } | ||
| 59 | 48 | }); | |
| 60 | |||
| 61 | 12 | return output; | |
| 62 | } | ||
| 63 | |||
| 64 | } // namespace | ||
| 65 | |||
| 66 |
1/2✓ Branch 1 taken 12 times.
✗ Branch 2 not taken.
|
12 | TerekhovDGaussVertTBB::TerekhovDGaussVertTBB(const InType &in) { |
| 67 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 68 | GetInput() = in; | ||
| 69 | 12 | } | |
| 70 | |||
| 71 | 12 | bool TerekhovDGaussVertTBB::ValidationImpl() { | |
| 72 | const auto &input = GetInput(); | ||
| 73 |
2/4✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
|
12 | if (input.width <= 0 || input.height <= 0) { |
| 74 | return false; | ||
| 75 | } | ||
| 76 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 12 times.
|
12 | if (static_cast<int>(input.data.size()) != input.width * input.height) { |
| 77 | ✗ | return false; | |
| 78 | } | ||
| 79 | return true; | ||
| 80 | } | ||
| 81 | |||
| 82 | 12 | bool TerekhovDGaussVertTBB::PreProcessingImpl() { | |
| 83 | const auto &input = GetInput(); | ||
| 84 | 12 | width_ = input.width; | |
| 85 | 12 | height_ = input.height; | |
| 86 | |||
| 87 | 12 | int padded_width = width_ + 2; | |
| 88 | 12 | int padded_height = height_ + 2; | |
| 89 | 12 | padded_image_.resize(static_cast<size_t>(padded_width) * static_cast<size_t>(padded_height)); | |
| 90 | |||
| 91 |
2/2✓ Branch 0 taken 232 times.
✓ Branch 1 taken 12 times.
|
244 | for (int row = 0; row < padded_height; ++row) { |
| 92 |
2/2✓ Branch 0 taken 6064 times.
✓ Branch 1 taken 232 times.
|
6296 | for (int col = 0; col < padded_width; ++col) { |
| 93 | 6064 | int src_x = col - 1; | |
| 94 | 6064 | int src_y = row - 1; | |
| 95 | |||
| 96 | 6064 | if (src_x < 0) { | |
| 97 | src_x = -src_x - 1; | ||
| 98 | } | ||
| 99 |
2/2✓ Branch 0 taken 232 times.
✓ Branch 1 taken 5832 times.
|
6064 | if (src_x >= width_) { |
| 100 | 232 | src_x = (2 * width_) - src_x - 1; | |
| 101 | } | ||
| 102 | 6064 | if (src_y < 0) { | |
| 103 | src_y = -src_y - 1; | ||
| 104 | } | ||
| 105 |
2/2✓ Branch 0 taken 232 times.
✓ Branch 1 taken 5832 times.
|
6064 | if (src_y >= height_) { |
| 106 | 232 | src_y = (2 * height_) - src_y - 1; | |
| 107 | } | ||
| 108 | |||
| 109 | 6064 | size_t padded_idx = (static_cast<size_t>(row) * static_cast<size_t>(padded_width)) + static_cast<size_t>(col); | |
| 110 | 6064 | size_t src_idx = (static_cast<size_t>(src_y) * static_cast<size_t>(width_)) + static_cast<size_t>(src_x); | |
| 111 | 6064 | padded_image_[padded_idx] = input.data[src_idx]; | |
| 112 | } | ||
| 113 | } | ||
| 114 | 12 | return true; | |
| 115 | } | ||
| 116 | |||
| 117 |
1/2✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
|
12 | bool TerekhovDGaussVertTBB::RunImpl() { |
| 118 | const auto &input = GetInput(); | ||
| 119 |
3/6✓ Branch 0 taken 12 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 12 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 12 times.
✗ Branch 5 not taken.
|
12 | if (input.data.empty() || width_ <= 0 || height_ <= 0) { |
| 120 | return false; | ||
| 121 | } | ||
| 122 | 12 | GetOutput() = SolveTBB(padded_image_, width_, height_); | |
| 123 | 12 | return true; | |
| 124 | } | ||
| 125 | |||
| 126 | 12 | bool TerekhovDGaussVertTBB::PostProcessingImpl() { | |
| 127 | 12 | return GetOutput().data.size() == (static_cast<size_t>(GetOutput().width) * static_cast<size_t>(GetOutput().height)); | |
| 128 | } | ||
| 129 | |||
| 130 | } // namespace terekhov_d_gauss_vert | ||
| 131 |