| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "remizov_k_dense_matrix_multiplication_cannon_algorithm/seq/include/ops_seq.hpp" | ||
| 2 | |||
| 3 | #include <cstddef> | ||
| 4 | #include <utility> | ||
| 5 | #include <vector> | ||
| 6 | |||
| 7 | #include "remizov_k_dense_matrix_multiplication_cannon_algorithm/common/include/common.hpp" | ||
| 8 | |||
| 9 | namespace remizov_k_dense_matrix_multiplication_cannon_algorithm { | ||
| 10 | |||
| 11 |
1/2✓ Branch 1 taken 64 times.
✗ Branch 2 not taken.
|
64 | RemizovKDenseMatrixMultiplicationCannonAlgorithm::RemizovKDenseMatrixMultiplicationCannonAlgorithm(const InType &in) { |
| 12 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 13 | GetInput() = in; | ||
| 14 | 64 | } | |
| 15 | |||
| 16 | 64 | bool RemizovKDenseMatrixMultiplicationCannonAlgorithm::ValidationImpl() { | |
| 17 | const auto &input_data = GetInput(); | ||
| 18 | |||
| 19 | 64 | int block_dim = std::get<0>(input_data); | |
| 20 | const auto &mat_a = std::get<1>(input_data); | ||
| 21 | const auto &mat_b = std::get<2>(input_data); | ||
| 22 | |||
| 23 |
1/2✓ Branch 0 taken 64 times.
✗ Branch 1 not taken.
|
64 | if (block_dim <= 0) { |
| 24 | return false; | ||
| 25 | } | ||
| 26 |
2/4✓ Branch 0 taken 64 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 64 times.
✗ Branch 3 not taken.
|
64 | if (mat_a.empty() || mat_b.empty()) { |
| 27 | return false; | ||
| 28 | } | ||
| 29 | |||
| 30 | size_t n = mat_a.size(); | ||
| 31 |
1/2✓ Branch 0 taken 64 times.
✗ Branch 1 not taken.
|
64 | if (n != mat_a[0].size()) { |
| 32 | return false; | ||
| 33 | } | ||
| 34 |
2/4✓ Branch 0 taken 64 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 64 times.
✗ Branch 3 not taken.
|
64 | if (n != mat_b.size() || n != mat_b[0].size()) { |
| 35 | return false; | ||
| 36 | } | ||
| 37 | |||
| 38 | 64 | return (n % static_cast<size_t>(block_dim) == 0); | |
| 39 | } | ||
| 40 | |||
| 41 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 64 times.
|
64 | bool RemizovKDenseMatrixMultiplicationCannonAlgorithm::PreProcessingImpl() { |
| 42 | GetOutput().clear(); | ||
| 43 | 64 | return true; | |
| 44 | } | ||
| 45 | |||
| 46 | 456 | void RemizovKDenseMatrixMultiplicationCannonAlgorithm::MultiplyBlock(const std::vector<std::vector<double>> &a, | |
| 47 | const std::vector<std::vector<double>> &b, | ||
| 48 | std::vector<std::vector<double>> &c, | ||
| 49 | int block_size) { | ||
| 50 |
2/2✓ Branch 0 taken 984 times.
✓ Branch 1 taken 456 times.
|
1440 | for (int i = 0; i < block_size; ++i) { |
| 51 |
2/2✓ Branch 0 taken 2568 times.
✓ Branch 1 taken 984 times.
|
3552 | for (int j = 0; j < block_size; ++j) { |
| 52 | double accumulator = 0.0; | ||
| 53 |
2/2✓ Branch 0 taken 7704 times.
✓ Branch 1 taken 2568 times.
|
10272 | for (int k = 0; k < block_size; ++k) { |
| 54 | 7704 | accumulator += a[i][k] * b[k][j]; | |
| 55 | } | ||
| 56 | 2568 | c[i][j] += accumulator; | |
| 57 | } | ||
| 58 | } | ||
| 59 | 456 | } | |
| 60 | |||
| 61 | 56 | void RemizovKDenseMatrixMultiplicationCannonAlgorithm::ShiftBlocksLeft( | |
| 62 | std::vector<std::vector<std::vector<std::vector<double>>>> &matrix_blocks, int block_count) { | ||
| 63 |
2/2✓ Branch 0 taken 112 times.
✓ Branch 1 taken 56 times.
|
168 | for (int i = 0; i < block_count; ++i) { |
| 64 | 112 | auto first_element = std::move(matrix_blocks[i][0]); | |
| 65 | |||
| 66 |
2/2✓ Branch 0 taken 112 times.
✓ Branch 1 taken 112 times.
|
224 | for (int j = 1; j < block_count; ++j) { |
| 67 | 112 | matrix_blocks[i][j - 1] = std::move(matrix_blocks[i][j]); | |
| 68 | } | ||
| 69 | |||
| 70 | 112 | matrix_blocks[i][block_count - 1] = std::move(first_element); | |
| 71 | 112 | } | |
| 72 | 56 | } | |
| 73 | |||
| 74 | 56 | void RemizovKDenseMatrixMultiplicationCannonAlgorithm::ShiftBlocksUp( | |
| 75 | std::vector<std::vector<std::vector<std::vector<double>>>> &matrix_blocks, int block_count) { | ||
| 76 |
2/2✓ Branch 0 taken 112 times.
✓ Branch 1 taken 56 times.
|
168 | for (int j = 0; j < block_count; ++j) { |
| 77 | 112 | auto first_element = std::move(matrix_blocks[0][j]); | |
| 78 | |||
| 79 |
2/2✓ Branch 0 taken 112 times.
✓ Branch 1 taken 112 times.
|
224 | for (int i = 1; i < block_count; ++i) { |
| 80 | 112 | matrix_blocks[i - 1][j] = std::move(matrix_blocks[i][j]); | |
| 81 | } | ||
| 82 | |||
| 83 | 112 | matrix_blocks[block_count - 1][j] = std::move(first_element); | |
| 84 | 112 | } | |
| 85 | 56 | } | |
| 86 | |||
| 87 | 64 | void RemizovKDenseMatrixMultiplicationCannonAlgorithm::RunCannonCycle( | |
| 88 | std::vector<std::vector<std::vector<std::vector<double>>>> &a_blocks, | ||
| 89 | std::vector<std::vector<std::vector<std::vector<double>>>> &b_blocks, | ||
| 90 | std::vector<std::vector<std::vector<std::vector<double>>>> &c_blocks, int block_size, int block_count) { | ||
| 91 |
2/2✓ Branch 0 taken 120 times.
✓ Branch 1 taken 64 times.
|
184 | for (int step = 0; step < block_count; ++step) { |
| 92 |
2/2✓ Branch 0 taken 232 times.
✓ Branch 1 taken 120 times.
|
352 | for (int i = 0; i < block_count; ++i) { |
| 93 |
2/2✓ Branch 0 taken 456 times.
✓ Branch 1 taken 232 times.
|
688 | for (int j = 0; j < block_count; ++j) { |
| 94 | 456 | RemizovKDenseMatrixMultiplicationCannonAlgorithm::MultiplyBlock(a_blocks[i][j], b_blocks[i][j], c_blocks[i][j], | |
| 95 | block_size); | ||
| 96 | } | ||
| 97 | } | ||
| 98 | |||
| 99 |
2/2✓ Branch 0 taken 56 times.
✓ Branch 1 taken 64 times.
|
120 | if (step < block_count - 1) { |
| 100 | 56 | RemizovKDenseMatrixMultiplicationCannonAlgorithm::ShiftBlocksLeft(a_blocks, block_count); | |
| 101 | 56 | RemizovKDenseMatrixMultiplicationCannonAlgorithm::ShiftBlocksUp(b_blocks, block_count); | |
| 102 | } | ||
| 103 | } | ||
| 104 | 64 | } | |
| 105 | |||
| 106 | 64 | void RemizovKDenseMatrixMultiplicationCannonAlgorithm::InitializeBlocks( | |
| 107 | const std::vector<std::vector<double>> &matrix_a, const std::vector<std::vector<double>> &matrix_b, | ||
| 108 | std::vector<std::vector<std::vector<std::vector<double>>>> &a_blocks, | ||
| 109 | std::vector<std::vector<std::vector<std::vector<double>>>> &b_blocks, int block_size, int block_count) { | ||
| 110 |
2/2✓ Branch 0 taken 120 times.
✓ Branch 1 taken 64 times.
|
184 | for (int i = 0; i < block_count; ++i) { |
| 111 |
2/2✓ Branch 0 taken 232 times.
✓ Branch 1 taken 120 times.
|
352 | for (int j = 0; j < block_count; ++j) { |
| 112 | 232 | int shift_value = (i + j) % block_count; | |
| 113 | |||
| 114 |
2/2✓ Branch 0 taken 504 times.
✓ Branch 1 taken 232 times.
|
736 | for (int bi = 0; bi < block_size; ++bi) { |
| 115 |
2/2✓ Branch 0 taken 1320 times.
✓ Branch 1 taken 504 times.
|
1824 | for (int bj = 0; bj < block_size; ++bj) { |
| 116 | 1320 | a_blocks[i][j][bi][bj] = matrix_a[(i * block_size) + bi][(shift_value * block_size) + bj]; | |
| 117 | 1320 | b_blocks[i][j][bi][bj] = matrix_b[(shift_value * block_size) + bi][(j * block_size) + bj]; | |
| 118 | } | ||
| 119 | } | ||
| 120 | } | ||
| 121 | } | ||
| 122 | 64 | } | |
| 123 | |||
| 124 | 64 | void RemizovKDenseMatrixMultiplicationCannonAlgorithm::AssembleOutput( | |
| 125 | std::vector<std::vector<std::vector<std::vector<double>>>> &c_blocks, std::vector<std::vector<double>> &output, | ||
| 126 | int block_size, int block_count) { | ||
| 127 |
2/2✓ Branch 0 taken 120 times.
✓ Branch 1 taken 64 times.
|
184 | for (int i = 0; i < block_count; ++i) { |
| 128 |
2/2✓ Branch 0 taken 232 times.
✓ Branch 1 taken 120 times.
|
352 | for (int j = 0; j < block_count; ++j) { |
| 129 |
2/2✓ Branch 0 taken 504 times.
✓ Branch 1 taken 232 times.
|
736 | for (int bi = 0; bi < block_size; ++bi) { |
| 130 |
2/2✓ Branch 0 taken 1320 times.
✓ Branch 1 taken 504 times.
|
1824 | for (int bj = 0; bj < block_size; ++bj) { |
| 131 | 1320 | output[(i * block_size) + bi][(j * block_size) + bj] = c_blocks[i][j][bi][bj]; | |
| 132 | } | ||
| 133 | } | ||
| 134 | } | ||
| 135 | } | ||
| 136 | 64 | } | |
| 137 | |||
| 138 | 64 | bool RemizovKDenseMatrixMultiplicationCannonAlgorithm::RunImpl() { | |
| 139 | const auto ¶ms = GetInput(); | ||
| 140 | |||
| 141 | 64 | int block_dim = std::get<0>(params); | |
| 142 | const auto &source_a = std::get<1>(params); | ||
| 143 | const auto &source_b = std::get<2>(params); | ||
| 144 | |||
| 145 | 64 | int matrix_size = static_cast<int>(source_a.size()); | |
| 146 | 64 | int blocks_per_dim = matrix_size / block_dim; | |
| 147 | |||
| 148 | std::vector<std::vector<std::vector<std::vector<double>>>> blocks_a( | ||
| 149 | blocks_per_dim, | ||
| 150 | 64 | std::vector<std::vector<std::vector<double>>>( | |
| 151 |
3/6✓ Branch 2 taken 64 times.
✗ Branch 3 not taken.
✓ Branch 5 taken 64 times.
✗ Branch 6 not taken.
✓ Branch 8 taken 64 times.
✗ Branch 9 not taken.
|
128 | blocks_per_dim, std::vector<std::vector<double>>(block_dim, std::vector<double>(block_dim, 0.0)))); |
| 152 | |||
| 153 | std::vector<std::vector<std::vector<std::vector<double>>>> blocks_b( | ||
| 154 | blocks_per_dim, | ||
| 155 | 64 | std::vector<std::vector<std::vector<double>>>( | |
| 156 |
4/8✓ Branch 1 taken 64 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 64 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 64 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 64 times.
✗ Branch 11 not taken.
|
128 | blocks_per_dim, std::vector<std::vector<double>>(block_dim, std::vector<double>(block_dim, 0.0)))); |
| 157 | |||
| 158 | std::vector<std::vector<std::vector<std::vector<double>>>> blocks_c( | ||
| 159 | blocks_per_dim, | ||
| 160 | 64 | std::vector<std::vector<std::vector<double>>>( | |
| 161 |
4/8✓ Branch 1 taken 64 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 64 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 64 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 64 times.
✗ Branch 11 not taken.
|
128 | blocks_per_dim, std::vector<std::vector<double>>(block_dim, std::vector<double>(block_dim, 0.0)))); |
| 162 | |||
| 163 | 64 | InitializeBlocks(source_a, source_b, blocks_a, blocks_b, block_dim, blocks_per_dim); | |
| 164 | 64 | RunCannonCycle(blocks_a, blocks_b, blocks_c, block_dim, blocks_per_dim); | |
| 165 | |||
| 166 |
2/4✓ Branch 1 taken 64 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 64 times.
✗ Branch 5 not taken.
|
64 | std::vector<std::vector<double>> result(matrix_size, std::vector<double>(matrix_size, 0.0)); |
| 167 | 64 | AssembleOutput(blocks_c, result, block_dim, blocks_per_dim); | |
| 168 | |||
| 169 | 64 | GetOutput() = std::move(result); | |
| 170 | 64 | return true; | |
| 171 | 64 | } | |
| 172 | |||
| 173 | 64 | bool RemizovKDenseMatrixMultiplicationCannonAlgorithm::PostProcessingImpl() { | |
| 174 | 64 | return true; | |
| 175 | } | ||
| 176 | |||
| 177 | } // namespace remizov_k_dense_matrix_multiplication_cannon_algorithm | ||
| 178 |