| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "timur_a_cannon/omp/include/ops_omp.hpp" | ||
| 2 | |||
| 3 | #include <omp.h> | ||
| 4 | |||
| 5 | #include <cstddef> | ||
| 6 | #include <utility> | ||
| 7 | #include <vector> | ||
| 8 | |||
| 9 | #include "timur_a_cannon/common/include/common.hpp" | ||
| 10 | |||
| 11 | namespace timur_a_cannon { | ||
| 12 | |||
| 13 |
1/2✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
|
32 | TimurACannonMatrixMultiplicationOMP::TimurACannonMatrixMultiplicationOMP(const InType &in) { |
| 14 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 15 | GetInput() = in; | ||
| 16 | 32 | } | |
| 17 | |||
| 18 | 32 | bool TimurACannonMatrixMultiplicationOMP::ValidationImpl() { | |
| 19 | const auto &input = GetInput(); | ||
| 20 | 32 | int b_size = std::get<0>(input); | |
| 21 | const auto &mat_a = std::get<1>(input); | ||
| 22 | const auto &mat_b = std::get<2>(input); | ||
| 23 | |||
| 24 |
4/8✗ Branch 0 not taken.
✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 32 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 32 times.
✗ Branch 6 not taken.
✓ Branch 7 taken 32 times.
|
32 | return b_size > 0 && !mat_a.empty() && !mat_b.empty() && mat_a.size() == mat_a[0].size() && |
| 25 |
2/4✓ Branch 0 taken 32 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 32 times.
|
64 | mat_a.size() == mat_b.size() && mat_a.size() % static_cast<size_t>(b_size) == 0; |
| 26 | } | ||
| 27 | |||
| 28 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 32 times.
|
32 | bool TimurACannonMatrixMultiplicationOMP::PreProcessingImpl() { |
| 29 | GetOutput().clear(); | ||
| 30 | 32 | return true; | |
| 31 | } | ||
| 32 | |||
| 33 | 304 | void TimurACannonMatrixMultiplicationOMP::BlockMultiplyAccumulate(const std::vector<std::vector<double>> &a, | |
| 34 | const std::vector<std::vector<double>> &b, | ||
| 35 | std::vector<std::vector<double>> &c, int b_size) { | ||
| 36 |
2/2✓ Branch 0 taken 784 times.
✓ Branch 1 taken 304 times.
|
1088 | for (int i = 0; i < b_size; ++i) { |
| 37 |
2/2✓ Branch 0 taken 2224 times.
✓ Branch 1 taken 784 times.
|
3008 | for (int k = 0; k < b_size; ++k) { |
| 38 | 2224 | double temp = a[i][k]; | |
| 39 |
2/2✓ Branch 0 taken 6736 times.
✓ Branch 1 taken 2224 times.
|
8960 | for (int j = 0; j < b_size; ++j) { |
| 40 | 6736 | c[i][j] += temp * b[k][j]; | |
| 41 | } | ||
| 42 | } | ||
| 43 | } | ||
| 44 | 304 | } | |
| 45 | |||
| 46 | ✗ | void TimurACannonMatrixMultiplicationOMP::DistributeData( | |
| 47 | const std::vector<std::vector<double>> &src_a, const std::vector<std::vector<double>> &src_b, | ||
| 48 | std::vector<std::vector<std::vector<std::vector<double>>>> &bl_a, | ||
| 49 | std::vector<std::vector<std::vector<std::vector<double>>>> &bl_b, int b_size, int grid_sz) { | ||
| 50 | 32 | #pragma omp parallel for collapse(2) default(none) shared(src_a, src_b, bl_a, bl_b, b_size, grid_sz) | |
| 51 | for (int i = 0; i < grid_sz; ++i) { | ||
| 52 | for (int j = 0; j < grid_sz; ++j) { | ||
| 53 | int shift = (i + j) % grid_sz; | ||
| 54 | for (int row = 0; row < b_size; ++row) { | ||
| 55 | for (int col = 0; col < b_size; ++col) { | ||
| 56 | bl_a[i][j][row][col] = src_a[(i * b_size) + row][((shift)*b_size) + col]; | ||
| 57 | bl_b[i][j][row][col] = src_b[((shift)*b_size) + row][(j * b_size) + col]; | ||
| 58 | } | ||
| 59 | } | ||
| 60 | } | ||
| 61 | } | ||
| 62 | ✗ | } | |
| 63 | |||
| 64 | ✗ | void TimurACannonMatrixMultiplicationOMP::CollectResult( | |
| 65 | const std::vector<std::vector<std::vector<std::vector<double>>>> &bl_c, std::vector<std::vector<double>> &res, | ||
| 66 | int b_size, int grid_sz) { | ||
| 67 | 32 | #pragma omp parallel for collapse(2) default(none) shared(bl_c, res, b_size, grid_sz) | |
| 68 | for (int i = 0; i < grid_sz; ++i) { | ||
| 69 | for (int j = 0; j < grid_sz; ++j) { | ||
| 70 | for (int row = 0; row < b_size; ++row) { | ||
| 71 | for (int col = 0; col < b_size; ++col) { | ||
| 72 | res[(i * b_size) + row][(j * b_size) + col] = bl_c[i][j][row][col]; | ||
| 73 | } | ||
| 74 | } | ||
| 75 | } | ||
| 76 | } | ||
| 77 | ✗ | } | |
| 78 | |||
| 79 | ✗ | void TimurACannonMatrixMultiplicationOMP::RotateBlocksA( | |
| 80 | std::vector<std::vector<std::vector<std::vector<double>>>> &blocks, int grid_sz) { | ||
| 81 | 32 | #pragma omp parallel for default(none) shared(blocks, grid_sz) | |
| 82 | for (int i = 0; i < grid_sz; ++i) { | ||
| 83 | auto first_block = std::move(blocks[i][0]); | ||
| 84 | for (int j = 0; j < grid_sz - 1; ++j) { | ||
| 85 | blocks[i][j] = std::move(blocks[i][j + 1]); | ||
| 86 | } | ||
| 87 | blocks[i][grid_sz - 1] = std::move(first_block); | ||
| 88 | } | ||
| 89 | ✗ | } | |
| 90 | |||
| 91 | ✗ | void TimurACannonMatrixMultiplicationOMP::RotateBlocksB( | |
| 92 | std::vector<std::vector<std::vector<std::vector<double>>>> &blocks, int grid_sz) { | ||
| 93 | 32 | #pragma omp parallel for default(none) shared(blocks, grid_sz) | |
| 94 | for (int j = 0; j < grid_sz; ++j) { | ||
| 95 | auto first_block = std::move(blocks[0][j]); | ||
| 96 | for (int i = 0; i < grid_sz - 1; ++i) { | ||
| 97 | blocks[i][j] = std::move(blocks[i + 1][j]); | ||
| 98 | } | ||
| 99 | blocks[grid_sz - 1][j] = std::move(first_block); | ||
| 100 | } | ||
| 101 | 32 | } | |
| 102 | |||
| 103 | 32 | bool TimurACannonMatrixMultiplicationOMP::RunImpl() { | |
| 104 | const auto &input = GetInput(); | ||
| 105 | 32 | int b_size = std::get<0>(input); | |
| 106 | 32 | int n = static_cast<int>(std::get<1>(input).size()); | |
| 107 | 32 | int grid_sz = n / b_size; | |
| 108 | |||
| 109 | using Matrix = std::vector<std::vector<double>>; | ||
| 110 | using BlockGrid = std::vector<std::vector<Matrix>>; | ||
| 111 | |||
| 112 |
3/6✓ Branch 2 taken 32 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 32 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 32 times.
✗ Branch 9 not taken.
|
32 | BlockGrid bl_a(grid_sz, std::vector<Matrix>(grid_sz, Matrix(b_size, std::vector<double>(b_size)))); |
| 113 |
4/8✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 32 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 32 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 32 times.
✗ Branch 11 not taken.
|
32 | BlockGrid bl_b(grid_sz, std::vector<Matrix>(grid_sz, Matrix(b_size, std::vector<double>(b_size)))); |
| 114 |
4/8✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 32 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 32 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 32 times.
✗ Branch 11 not taken.
|
64 | BlockGrid bl_c(grid_sz, std::vector<Matrix>(grid_sz, Matrix(b_size, std::vector<double>(b_size, 0.0)))); |
| 115 | |||
| 116 | DistributeData(std::get<1>(input), std::get<2>(input), bl_a, bl_b, b_size, grid_sz); | ||
| 117 | |||
| 118 |
2/2✓ Branch 0 taken 64 times.
✓ Branch 1 taken 32 times.
|
96 | for (int step = 0; step < grid_sz; ++step) { |
| 119 | 64 | #pragma omp parallel for collapse(2) default(none) shared(bl_a, bl_b, bl_c, b_size, grid_sz) | |
| 120 | for (int i = 0; i < grid_sz; ++i) { | ||
| 121 | for (int j = 0; j < grid_sz; ++j) { | ||
| 122 | BlockMultiplyAccumulate(bl_a[i][j], bl_b[i][j], bl_c[i][j], b_size); | ||
| 123 | } | ||
| 124 | } | ||
| 125 |
2/2✓ Branch 0 taken 32 times.
✓ Branch 1 taken 32 times.
|
64 | if (step < grid_sz - 1) { |
| 126 | RotateBlocksA(bl_a, grid_sz); | ||
| 127 | RotateBlocksB(bl_b, grid_sz); | ||
| 128 | } | ||
| 129 | } | ||
| 130 | |||
| 131 |
2/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 32 times.
✗ Branch 5 not taken.
|
64 | Matrix res_mat(n, std::vector<double>(n)); |
| 132 | CollectResult(bl_c, res_mat, b_size, grid_sz); | ||
| 133 | |||
| 134 | 32 | GetOutput() = std::move(res_mat); | |
| 135 | 32 | return true; | |
| 136 | 32 | } | |
| 137 | |||
| 138 | 32 | bool TimurACannonMatrixMultiplicationOMP::PostProcessingImpl() { | |
| 139 | 32 | return true; | |
| 140 | } | ||
| 141 | |||
| 142 | } // namespace timur_a_cannon | ||
| 143 |