| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "baranov_a_mult_matrix_fox_algorithm/tbb/include/ops_tbb.hpp" | ||
| 2 | |||
| 3 | #include <algorithm> | ||
| 4 | #include <cmath> | ||
| 5 | #include <cstddef> | ||
| 6 | #include <vector> | ||
| 7 | |||
| 8 | #include "baranov_a_mult_matrix_fox_algorithm/common/include/common.hpp" | ||
| 9 | #include "oneapi/tbb.h" | ||
| 10 | |||
| 11 | namespace baranov_a_mult_matrix_fox_algorithm_tbb { | ||
| 12 | |||
| 13 | 80 | BaranovAMultMatrixFoxAlgorithmTBB::BaranovAMultMatrixFoxAlgorithmTBB( | |
| 14 |
1/2✓ Branch 1 taken 80 times.
✗ Branch 2 not taken.
|
80 | const baranov_a_mult_matrix_fox_algorithm::InType &in) { |
| 15 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 16 | GetInput() = in; | ||
| 17 | 80 | GetOutput() = std::vector<double>(); | |
| 18 | 80 | } | |
| 19 | |||
| 20 | 80 | bool BaranovAMultMatrixFoxAlgorithmTBB::ValidationImpl() { | |
| 21 | const auto &[matrix_size, matrix_a, matrix_b] = GetInput(); | ||
| 22 |
3/6✓ Branch 0 taken 80 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 80 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 80 times.
|
80 | return matrix_size > 0 && matrix_a.size() == matrix_size * matrix_size && |
| 23 | 80 | matrix_b.size() == matrix_size * matrix_size; | |
| 24 | } | ||
| 25 | |||
| 26 | 80 | bool BaranovAMultMatrixFoxAlgorithmTBB::PreProcessingImpl() { | |
| 27 | const auto &[matrix_size, matrix_a, matrix_b] = GetInput(); | ||
| 28 | 80 | GetOutput() = std::vector<double>(matrix_size * matrix_size, 0.0); | |
| 29 | 80 | return true; | |
| 30 | } | ||
| 31 | |||
| 32 | ✗ | void BaranovAMultMatrixFoxAlgorithmTBB::StandardMultiplication(size_t n) { | |
| 33 | const auto &[matrix_size, matrix_a, matrix_b] = GetInput(); | ||
| 34 | auto &output = GetOutput(); | ||
| 35 | |||
| 36 | 72 | tbb::parallel_for(static_cast<size_t>(0), n, [&](size_t i) { | |
| 37 |
2/2✓ Branch 0 taken 6704 times.
✓ Branch 1 taken 480 times.
|
7184 | for (size_t j = 0; j < n; ++j) { |
| 38 | double sum = 0.0; | ||
| 39 |
2/2✓ Branch 0 taken 157320 times.
✓ Branch 1 taken 6704 times.
|
164024 | for (size_t k = 0; k < n; ++k) { |
| 40 | 157320 | sum += matrix_a[(i * n) + k] * matrix_b[(k * n) + j]; | |
| 41 | } | ||
| 42 | 6704 | output[(i * n) + j] = sum; | |
| 43 | } | ||
| 44 | 480 | }); | |
| 45 | ✗ | } | |
| 46 | |||
| 47 | 8 | void BaranovAMultMatrixFoxAlgorithmTBB::FoxBlockMultiplication(size_t n, size_t block_size) { | |
| 48 | const auto &[matrix_size, matrix_a, matrix_b] = GetInput(); | ||
| 49 | auto &output = GetOutput(); | ||
| 50 | |||
| 51 | 8 | size_t num_blocks = (n + block_size - 1) / block_size; | |
| 52 | |||
| 53 | 81928 | tbb::parallel_for(static_cast<size_t>(0), n * n, [&](size_t idx) { output[idx] = 0.0; }); | |
| 54 | |||
| 55 |
2/2✓ Branch 0 taken 12 times.
✓ Branch 1 taken 8 times.
|
20 | for (size_t bk = 0; bk < num_blocks; ++bk) { |
| 56 | 12 | tbb::parallel_for(static_cast<size_t>(0), num_blocks * num_blocks, [&](size_t linear_idx) { | |
| 57 | 36 | size_t bi = linear_idx / num_blocks; | |
| 58 | 36 | size_t bj = linear_idx % num_blocks; | |
| 59 | |||
| 60 | 36 | size_t broadcast_block = (bi + bk) % num_blocks; | |
| 61 | |||
| 62 | 36 | size_t i_start = bi * block_size; | |
| 63 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 36 times.
|
36 | size_t i_end = std::min(i_start + block_size, n); |
| 64 | 36 | size_t j_start = bj * block_size; | |
| 65 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 36 times.
|
36 | size_t j_end = std::min(j_start + block_size, n); |
| 66 | 36 | size_t k_start = broadcast_block * block_size; | |
| 67 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 36 times.
|
36 | size_t k_end = std::min(k_start + block_size, n); |
| 68 | |||
| 69 |
2/2✓ Branch 0 taken 2304 times.
✓ Branch 1 taken 36 times.
|
2340 | for (size_t i = i_start; i < i_end; ++i) { |
| 70 |
2/2✓ Branch 0 taken 147456 times.
✓ Branch 1 taken 2304 times.
|
149760 | for (size_t j = j_start; j < j_end; ++j) { |
| 71 | double sum = 0.0; | ||
| 72 |
2/2✓ Branch 0 taken 9437184 times.
✓ Branch 1 taken 147456 times.
|
9584640 | for (size_t k = k_start; k < k_end; ++k) { |
| 73 | 9437184 | sum += matrix_a[(i * n) + k] * matrix_b[(k * n) + j]; | |
| 74 | } | ||
| 75 | 147456 | output[(i * n) + j] += sum; | |
| 76 | } | ||
| 77 | } | ||
| 78 | 36 | }); | |
| 79 | } | ||
| 80 | 8 | } | |
| 81 | |||
| 82 | 80 | bool BaranovAMultMatrixFoxAlgorithmTBB::RunImpl() { | |
| 83 | const auto &[matrix_size, matrix_a, matrix_b] = GetInput(); | ||
| 84 | 80 | size_t n = matrix_size; | |
| 85 | size_t block_size = 64; | ||
| 86 |
2/2✓ Branch 0 taken 72 times.
✓ Branch 1 taken 8 times.
|
80 | if (n < block_size) { |
| 87 | 72 | StandardMultiplication(n); | |
| 88 | } else { | ||
| 89 | 8 | FoxBlockMultiplication(n, block_size); | |
| 90 | } | ||
| 91 | |||
| 92 | 80 | return true; | |
| 93 | } | ||
| 94 | |||
| 95 | 80 | bool BaranovAMultMatrixFoxAlgorithmTBB::PostProcessingImpl() { | |
| 96 | 80 | return true; | |
| 97 | } | ||
| 98 | |||
| 99 | } // namespace baranov_a_mult_matrix_fox_algorithm_tbb | ||
| 100 |