| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "sinev_a_mult_matrix_fox_algorithm/omp/include/ops_omp.hpp" | ||
| 2 | |||
| 3 | #include <omp.h> | ||
| 4 | |||
| 5 | #include <cmath> | ||
| 6 | #include <cstddef> | ||
| 7 | #include <vector> | ||
| 8 | |||
| 9 | #include "sinev_a_mult_matrix_fox_algorithm/common/include/common.hpp" | ||
| 10 | |||
| 11 | namespace sinev_a_mult_matrix_fox_algorithm { | ||
| 12 | |||
| 13 |
1/2✓ Branch 1 taken 52 times.
✗ Branch 2 not taken.
|
52 | SinevAMultMatrixFoxAlgorithmOMP::SinevAMultMatrixFoxAlgorithmOMP(const InType &in) { |
| 14 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 15 | GetInput() = in; | ||
| 16 | GetOutput() = {}; | ||
| 17 | 52 | } | |
| 18 | |||
| 19 | 52 | bool SinevAMultMatrixFoxAlgorithmOMP::ValidationImpl() { | |
| 20 | const auto &[matrix_size, matrix_a, matrix_b] = GetInput(); | ||
| 21 | |||
| 22 |
3/6✓ Branch 0 taken 52 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 52 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 52 times.
|
52 | return matrix_size > 0 && matrix_a.size() == matrix_size * matrix_size && |
| 23 | 52 | matrix_b.size() == matrix_size * matrix_size; | |
| 24 | } | ||
| 25 | |||
| 26 | 52 | bool SinevAMultMatrixFoxAlgorithmOMP::PreProcessingImpl() { | |
| 27 | const auto &[matrix_size, matrix_a, matrix_b] = GetInput(); | ||
| 28 | 52 | GetOutput() = std::vector<double>(matrix_size * matrix_size, 0.0); | |
| 29 | 52 | return true; | |
| 30 | } | ||
| 31 | |||
| 32 | ✗ | void SinevAMultMatrixFoxAlgorithmOMP::SimpleMultiply(size_t n, const std::vector<double> &a, | |
| 33 | const std::vector<double> &b, std::vector<double> &c) { | ||
| 34 | 20 | #pragma omp parallel for default(none) shared(n, a, b, c) collapse(2) | |
| 35 | for (size_t i = 0; i < n; ++i) { | ||
| 36 | for (size_t j = 0; j < n; ++j) { | ||
| 37 | double sum = 0.0; | ||
| 38 | for (size_t k = 0; k < n; ++k) { | ||
| 39 | sum += a[(i * n) + k] * b[(k * n) + j]; | ||
| 40 | } | ||
| 41 | c[(i * n) + j] = sum; | ||
| 42 | } | ||
| 43 | } | ||
| 44 | ✗ | } | |
| 45 | |||
| 46 | ✗ | void SinevAMultMatrixFoxAlgorithmOMP::DecomposeToBlocks(const std::vector<double> &src, std::vector<double> &dst, | |
| 47 | size_t n, size_t bs, int q) { | ||
| 48 | 32 | #pragma omp parallel for default(none) shared(src, dst, n, bs, q) collapse(2) | |
| 49 | for (int bi = 0; bi < q; ++bi) { | ||
| 50 | for (int bj = 0; bj < q; ++bj) { | ||
| 51 | const size_t block_off = (static_cast<size_t>((bi * q) + bj)) * (bs * bs); | ||
| 52 | for (size_t i = 0; i < bs; ++i) { | ||
| 53 | for (size_t j = 0; j < bs; ++j) { | ||
| 54 | const size_t src_idx = ((static_cast<size_t>(bi) * bs + i) * n) + (static_cast<size_t>(bj) * bs + j); | ||
| 55 | const size_t dst_idx = block_off + (i * bs) + j; | ||
| 56 | dst[dst_idx] = src[src_idx]; | ||
| 57 | } | ||
| 58 | } | ||
| 59 | } | ||
| 60 | } | ||
| 61 | ✗ | } | |
| 62 | |||
| 63 | ✗ | void SinevAMultMatrixFoxAlgorithmOMP::AssembleFromBlocks(const std::vector<double> &src, std::vector<double> &dst, | |
| 64 | size_t n, size_t bs, int q) { | ||
| 65 |
1/2✓ Branch 0 taken 32 times.
✗ Branch 1 not taken.
|
32 | #pragma omp parallel for default(none) shared(src, dst, n, bs, q) collapse(2) |
| 66 | for (int bi = 0; bi < q; ++bi) { | ||
| 67 | for (int bj = 0; bj < q; ++bj) { | ||
| 68 | const size_t block_off = (static_cast<size_t>((bi * q) + bj)) * (bs * bs); | ||
| 69 | for (size_t i = 0; i < bs; ++i) { | ||
| 70 | for (size_t j = 0; j < bs; ++j) { | ||
| 71 | const size_t src_idx = block_off + (i * bs) + j; | ||
| 72 | const size_t dst_idx = ((static_cast<size_t>(bi) * bs + i) * n) + (static_cast<size_t>(bj) * bs + j); | ||
| 73 | dst[dst_idx] = src[src_idx]; | ||
| 74 | } | ||
| 75 | } | ||
| 76 | } | ||
| 77 | } | ||
| 78 | ✗ | } | |
| 79 | |||
| 80 | ✗ | void SinevAMultMatrixFoxAlgorithmOMP::FoxStep(const std::vector<double> &blocks_a, const std::vector<double> &blocks_b, | |
| 81 | std::vector<double> &blocks_c, size_t bs, int q, int step) { | ||
| 82 | ✗ | const size_t block_size_bytes = bs * bs; | |
| 83 | 232 | #pragma omp parallel for default(none) shared(blocks_a, blocks_b, blocks_c, bs, q, step, block_size_bytes) collapse(2) | |
| 84 | for (int i = 0; i < q; ++i) { | ||
| 85 | for (int j = 0; j < q; ++j) { | ||
| 86 | const int k = (i + step) % q; | ||
| 87 | |||
| 88 | const size_t a_off = (static_cast<size_t>((i * q) + k)) * block_size_bytes; | ||
| 89 | const size_t b_off = (static_cast<size_t>((k * q) + j)) * block_size_bytes; | ||
| 90 | const size_t c_off = (static_cast<size_t>((i * q) + j)) * block_size_bytes; | ||
| 91 | |||
| 92 | for (size_t ii = 0; ii < bs; ++ii) { | ||
| 93 | for (size_t kk = 0; kk < bs; ++kk) { | ||
| 94 | const double val = blocks_a[a_off + (ii * bs) + kk]; | ||
| 95 | for (size_t jj = 0; jj < bs; ++jj) { | ||
| 96 | blocks_c[c_off + (ii * bs) + jj] += val * blocks_b[b_off + (kk * bs) + jj]; | ||
| 97 | } | ||
| 98 | } | ||
| 99 | } | ||
| 100 | } | ||
| 101 | } | ||
| 102 | ✗ | } | |
| 103 | |||
| 104 | 52 | bool SinevAMultMatrixFoxAlgorithmOMP::RunImpl() { | |
| 105 | const auto &input = GetInput(); | ||
| 106 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 32 times.
|
52 | const size_t n = std::get<0>(input); |
| 107 | const auto &a = std::get<1>(input); | ||
| 108 | const auto &b = std::get<2>(input); | ||
| 109 | auto &c = GetOutput(); | ||
| 110 | |||
| 111 | // Для маленьких матриц используем простое умножение | ||
| 112 |
2/2✓ Branch 0 taken 20 times.
✓ Branch 1 taken 32 times.
|
52 | if (n <= 8) { |
| 113 | SimpleMultiply(n, a, b, c); | ||
| 114 | 20 | return true; | |
| 115 | } | ||
| 116 | |||
| 117 | size_t bs = 1; | ||
| 118 | 32 | auto sqrt_n = static_cast<size_t>(std::sqrt(static_cast<double>(n))); | |
| 119 |
1/2✓ Branch 0 taken 60 times.
✗ Branch 1 not taken.
|
60 | for (size_t div = sqrt_n; div >= 1; --div) { |
| 120 |
2/2✓ Branch 0 taken 28 times.
✓ Branch 1 taken 32 times.
|
60 | if (n % div == 0) { |
| 121 | bs = div; | ||
| 122 | break; | ||
| 123 | } | ||
| 124 | } | ||
| 125 | |||
| 126 | 32 | const int actual_q = static_cast<int>(n / bs); | |
| 127 | |||
| 128 | 32 | auto total_blocks = static_cast<size_t>(actual_q) * static_cast<size_t>(actual_q); | |
| 129 | 32 | auto block_elements = bs * bs; | |
| 130 | |||
| 131 | 32 | std::vector<double> blocks_a(total_blocks * block_elements); | |
| 132 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<double> blocks_b(total_blocks * block_elements); |
| 133 |
1/4✓ Branch 1 taken 32 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
32 | std::vector<double> blocks_c(total_blocks * block_elements, 0.0); |
| 134 | |||
| 135 | DecomposeToBlocks(a, blocks_a, n, bs, actual_q); | ||
| 136 | DecomposeToBlocks(b, blocks_b, n, bs, actual_q); | ||
| 137 | |||
| 138 |
2/2✓ Branch 0 taken 232 times.
✓ Branch 1 taken 32 times.
|
264 | for (int step = 0; step < actual_q; ++step) { |
| 139 | FoxStep(blocks_a, blocks_b, blocks_c, bs, actual_q, step); | ||
| 140 | } | ||
| 141 | |||
| 142 | AssembleFromBlocks(blocks_c, c, n, bs, actual_q); | ||
| 143 | |||
| 144 | return true; | ||
| 145 | } | ||
| 146 | |||
| 147 | 52 | bool SinevAMultMatrixFoxAlgorithmOMP::PostProcessingImpl() { | |
| 148 | 52 | return true; | |
| 149 | } | ||
| 150 | |||
| 151 | } // namespace sinev_a_mult_matrix_fox_algorithm | ||
| 152 |