| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | #include "tabalaev_a_matrix_mul_strassen/stl/include/ops_stl.hpp" | ||
| 2 | |||
| 3 | #include <algorithm> | ||
| 4 | #include <cstddef> | ||
| 5 | #include <functional> | ||
| 6 | #include <stack> | ||
| 7 | #include <thread> | ||
| 8 | #include <utility> | ||
| 9 | #include <vector> | ||
| 10 | |||
| 11 | #include "tabalaev_a_matrix_mul_strassen/common/include/common.hpp" | ||
| 12 | #include "util/include/util.hpp" | ||
| 13 | |||
| 14 | namespace tabalaev_a_matrix_mul_strassen { | ||
| 15 | |||
| 16 | static constexpr std::size_t kBaseCaseSize = 128; | ||
| 17 | static constexpr std::size_t kParallelThreshold = 65536; | ||
| 18 | |||
| 19 | namespace { | ||
| 20 | template <typename fnc> | ||
| 21 | 704 | void RunParallel(std::size_t begin, std::size_t end, std::size_t threshold, const fnc &func) { | |
| 22 | 704 | std::size_t total = end - begin; | |
| 23 | |||
| 24 |
1/2✓ Branch 0 taken 352 times.
✗ Branch 1 not taken.
|
704 | if (total < threshold) { |
| 25 |
2/2✓ Branch 0 taken 296040 times.
✓ Branch 1 taken 352 times.
|
592784 | for (std::size_t i = begin; i < end; ++i) { |
| 26 | 592080 | func(i); | |
| 27 | } | ||
| 28 | 704 | return; | |
| 29 | } | ||
| 30 | |||
| 31 | ✗ | auto num_threads = static_cast<std::size_t>(ppc::util::GetNumThreads()); | |
| 32 | |||
| 33 | ✗ | num_threads = std::min(num_threads, total); | |
| 34 | |||
| 35 | ✗ | std::vector<std::thread> threads; | |
| 36 | ✗ | std::size_t chunk_size = total / num_threads; | |
| 37 | |||
| 38 | ✗ | for (std::size_t i = 0; i < num_threads; ++i) { | |
| 39 | ✗ | std::size_t current_start = begin + (i * chunk_size); | |
| 40 | |||
| 41 | ✗ | std::size_t current_end = (i == num_threads - 1) ? end : current_start + chunk_size; | |
| 42 | |||
| 43 | ✗ | threads.emplace_back([current_start, current_end, &func]() { | |
| 44 | ✗ | for (std::size_t j = current_start; j < current_end; ++j) { | |
| 45 | ✗ | func(j); | |
| 46 | } | ||
| 47 | }); | ||
| 48 | } | ||
| 49 | |||
| 50 | ✗ | for (auto &t : threads) { | |
| 51 | ✗ | t.join(); | |
| 52 | } | ||
| 53 | ✗ | } | |
| 54 | |||
| 55 | } // namespace | ||
| 56 | |||
| 57 |
1/2✓ Branch 1 taken 48 times.
✗ Branch 2 not taken.
|
48 | TabalaevAMatrixMulStrassenSTL::TabalaevAMatrixMulStrassenSTL(const InType &in) { |
| 58 | SetTypeOfTask(GetStaticTypeOfTask()); | ||
| 59 |
1/2✓ Branch 1 taken 48 times.
✗ Branch 2 not taken.
|
48 | GetInput() = in; |
| 60 | GetOutput() = {}; | ||
| 61 | 48 | } | |
| 62 | |||
| 63 | 48 | bool TabalaevAMatrixMulStrassenSTL::ValidationImpl() { | |
| 64 | const auto &in = GetInput(); | ||
| 65 |
2/4✗ Branch 0 not taken.
✓ Branch 1 taken 48 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 48 times.
|
48 | return in.a_rows > 0 && in.a_cols_b_rows > 0 && in.b_cols > 0 && |
| 66 |
2/4✓ Branch 0 taken 48 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 48 times.
|
96 | in.a.size() == static_cast<size_t>(in.a_rows * in.a_cols_b_rows) && |
| 67 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 48 times.
|
48 | in.b.size() == static_cast<size_t>(in.a_cols_b_rows * in.b_cols); |
| 68 | } | ||
| 69 | |||
| 70 | 48 | bool TabalaevAMatrixMulStrassenSTL::PreProcessingImpl() { | |
| 71 | GetOutput() = {}; | ||
| 72 | const auto &in = GetInput(); | ||
| 73 | |||
| 74 | 48 | a_rows_ = in.a_rows; | |
| 75 | 48 | a_cols_b_rows_ = in.a_cols_b_rows; | |
| 76 | 48 | b_cols_ = in.b_cols; | |
| 77 | |||
| 78 | 48 | std::size_t max_dim = std::max({a_rows_, a_cols_b_rows_, b_cols_}); | |
| 79 | 48 | padded_n_ = 1; | |
| 80 |
2/2✓ Branch 0 taken 224 times.
✓ Branch 1 taken 48 times.
|
272 | while (padded_n_ < max_dim) { |
| 81 | 224 | padded_n_ *= 2; | |
| 82 | } | ||
| 83 | |||
| 84 | 48 | padded_a_.assign(padded_n_ * padded_n_, 0.0); | |
| 85 | 48 | padded_b_.assign(padded_n_ * padded_n_, 0.0); | |
| 86 | |||
| 87 | 48 | RunParallel(0, a_rows_, kParallelThreshold, [&](std::size_t i) { | |
| 88 | 4392 | auto src_start = in.a.begin() + static_cast<ptrdiff_t>(i * a_cols_b_rows_); | |
| 89 | 4392 | auto dst_start = padded_a_.begin() + static_cast<ptrdiff_t>(i * padded_n_); | |
| 90 | 4392 | std::copy(src_start, src_start + static_cast<ptrdiff_t>(a_cols_b_rows_), dst_start); | |
| 91 | 4392 | }); | |
| 92 | |||
| 93 | 48 | RunParallel(0, a_cols_b_rows_, kParallelThreshold, [&](std::size_t i) { | |
| 94 | 4312 | auto src_start = in.b.begin() + static_cast<ptrdiff_t>(i * b_cols_); | |
| 95 | 4312 | auto dst_start = padded_b_.begin() + static_cast<ptrdiff_t>(i * padded_n_); | |
| 96 | 4312 | std::copy(src_start, src_start + static_cast<ptrdiff_t>(b_cols_), dst_start); | |
| 97 | 4312 | }); | |
| 98 | |||
| 99 | 48 | return true; | |
| 100 | } | ||
| 101 | |||
| 102 | 48 | bool TabalaevAMatrixMulStrassenSTL::RunImpl() { | |
| 103 | 96 | result_c_ = StrassenMultiply(padded_a_, padded_b_, padded_n_); | |
| 104 | |||
| 105 | auto &out = GetOutput(); | ||
| 106 | 48 | out.assign(a_rows_ * b_cols_, 0.0); | |
| 107 | |||
| 108 | 48 | RunParallel(0, a_rows_, kParallelThreshold, [&](std::size_t i) { | |
| 109 | 4392 | auto src_start = result_c_.begin() + static_cast<ptrdiff_t>(i * padded_n_); | |
| 110 | 4392 | auto dst_start = out.begin() + static_cast<ptrdiff_t>(i * b_cols_); | |
| 111 | 4392 | std::copy(src_start, src_start + static_cast<ptrdiff_t>(b_cols_), dst_start); | |
| 112 | 4392 | }); | |
| 113 | |||
| 114 | 48 | return true; | |
| 115 | } | ||
| 116 | |||
| 117 | 48 | bool TabalaevAMatrixMulStrassenSTL::PostProcessingImpl() { | |
| 118 | 48 | return true; | |
| 119 | } | ||
| 120 | |||
| 121 | 96 | std::vector<double> TabalaevAMatrixMulStrassenSTL::Add(const std::vector<double> &mat_a, | |
| 122 | const std::vector<double> &mat_b) { | ||
| 123 | 96 | std::vector<double> res(mat_a.size()); | |
| 124 | std::ranges::transform(mat_a, mat_b, res.begin(), std::plus<>()); | ||
| 125 | 96 | return res; | |
| 126 | } | ||
| 127 | |||
| 128 | 64 | std::vector<double> TabalaevAMatrixMulStrassenSTL::Subtract(const std::vector<double> &mat_a, | |
| 129 | const std::vector<double> &mat_b) { | ||
| 130 | 64 | std::vector<double> res(mat_a.size()); | |
| 131 | std::ranges::transform(mat_a, mat_b, res.begin(), std::minus<>()); | ||
| 132 | 64 | return res; | |
| 133 | } | ||
| 134 | |||
| 135 | 144 | std::vector<double> TabalaevAMatrixMulStrassenSTL::BaseMultiply(const std::vector<double> &mat_a, | |
| 136 | const std::vector<double> &mat_b, std::size_t n) { | ||
| 137 |
1/2✓ Branch 2 taken 144 times.
✗ Branch 3 not taken.
|
144 | std::vector<double> res(n * n, 0.0); |
| 138 | |||
| 139 | 144 | const double *a_ptr = mat_a.data(); | |
| 140 |
1/2✓ Branch 1 taken 144 times.
✗ Branch 2 not taken.
|
144 | const double *b_ptr = mat_b.data(); |
| 141 | 144 | double *res_ptr = res.data(); | |
| 142 | |||
| 143 |
1/4✓ Branch 1 taken 144 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
144 | RunParallel(0, n, kParallelThreshold, [&](std::size_t i) { |
| 144 | 14656 | std::size_t i_n = i * n; | |
| 145 |
2/2✓ Branch 0 taken 1839360 times.
✓ Branch 1 taken 14656 times.
|
1854016 | for (std::size_t k = 0; k < n; ++k) { |
| 146 | 1839360 | double temp = a_ptr[i_n + k]; | |
| 147 |
2/2✓ Branch 0 taken 7384 times.
✓ Branch 1 taken 1831976 times.
|
1839360 | if (temp == 0.0) { |
| 148 | 7384 | continue; | |
| 149 | } | ||
| 150 | 1831976 | std::size_t k_n = k * n; | |
| 151 |
2/2✓ Branch 0 taken 234176224 times.
✓ Branch 1 taken 1831976 times.
|
236008200 | for (std::size_t j = 0; j < n; ++j) { |
| 152 | 234176224 | res_ptr[i_n + j] += temp * b_ptr[k_n + j]; | |
| 153 | } | ||
| 154 | } | ||
| 155 | 14656 | }); | |
| 156 | |||
| 157 | 144 | return res; | |
| 158 | } | ||
| 159 | |||
| 160 | 32 | void TabalaevAMatrixMulStrassenSTL::SplitMatrix(const std::vector<double> &src, std::size_t n, std::vector<double> &c11, | |
| 161 | std::vector<double> &c12, std::vector<double> &c21, | ||
| 162 | std::vector<double> &c22) { | ||
| 163 | 32 | std::size_t h = n / 2; | |
| 164 | 32 | std::size_t sz = h * h; | |
| 165 | 32 | c11.resize(sz); | |
| 166 | 32 | c12.resize(sz); | |
| 167 | 32 | c21.resize(sz); | |
| 168 | 32 | c22.resize(sz); | |
| 169 | |||
| 170 | 32 | RunParallel(0, h, kParallelThreshold, [&](std::size_t i) { | |
| 171 | 4096 | auto src_row1 = src.begin() + static_cast<ptrdiff_t>(i * n); | |
| 172 | 4096 | auto src_row2 = src.begin() + static_cast<ptrdiff_t>((i + h) * n); | |
| 173 | 4096 | auto dst_row = static_cast<ptrdiff_t>(i * h); | |
| 174 | |||
| 175 | 4096 | std::copy(src_row1, src_row1 + static_cast<ptrdiff_t>(h), c11.begin() + dst_row); | |
| 176 | 4096 | std::copy(src_row1 + static_cast<ptrdiff_t>(h), src_row1 + static_cast<ptrdiff_t>(n), c12.begin() + dst_row); | |
| 177 | |||
| 178 | 4096 | std::copy(src_row2, src_row2 + static_cast<ptrdiff_t>(h), c21.begin() + dst_row); | |
| 179 | 4096 | std::copy(src_row2 + static_cast<ptrdiff_t>(h), src_row2 + static_cast<ptrdiff_t>(n), c22.begin() + dst_row); | |
| 180 | 4096 | }); | |
| 181 | 32 | } | |
| 182 | |||
| 183 | 16 | std::vector<double> TabalaevAMatrixMulStrassenSTL::CombineMatrix(const std::vector<double> &c11, | |
| 184 | const std::vector<double> &c12, | ||
| 185 | const std::vector<double> &c21, | ||
| 186 | const std::vector<double> &c22, std::size_t n) { | ||
| 187 | 16 | std::size_t h = n / 2; | |
| 188 | 16 | std::vector<double> res(n * n); | |
| 189 | |||
| 190 |
1/4✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
16 | RunParallel(0, h, kParallelThreshold, [&](std::size_t i) { |
| 191 | 2048 | auto res_row1 = res.begin() + static_cast<ptrdiff_t>(i * n); | |
| 192 | 2048 | auto res_row2 = res.begin() + static_cast<ptrdiff_t>((i + h) * n); | |
| 193 | 2048 | auto src_row = static_cast<ptrdiff_t>(i * h); | |
| 194 | |||
| 195 | 2048 | std::copy(c11.begin() + src_row, c11.begin() + src_row + static_cast<ptrdiff_t>(h), res_row1); | |
| 196 | 2048 | std::copy(c12.begin() + src_row, c12.begin() + src_row + static_cast<ptrdiff_t>(h), | |
| 197 | 2048 | res_row1 + static_cast<ptrdiff_t>(h)); | |
| 198 | |||
| 199 | 2048 | std::copy(c21.begin() + src_row, c21.begin() + src_row + static_cast<ptrdiff_t>(h), res_row2); | |
| 200 | 2048 | std::copy(c22.begin() + src_row, c22.begin() + src_row + static_cast<ptrdiff_t>(h), | |
| 201 | 2048 | res_row2 + static_cast<ptrdiff_t>(h)); | |
| 202 | 2048 | }); | |
| 203 | 16 | return res; | |
| 204 | } | ||
| 205 | |||
| 206 | 48 | std::vector<double> TabalaevAMatrixMulStrassenSTL::StrassenMultiply(const std::vector<double> &mat_a, | |
| 207 | const std::vector<double> &mat_b, std::size_t n) { | ||
| 208 | std::stack<StrassenFrameSTL> frames; | ||
| 209 | std::stack<std::vector<double>> results; | ||
| 210 | |||
| 211 | 48 | frames.push({mat_a, mat_b, n, 0}); | |
| 212 | |||
| 213 |
2/2✓ Branch 0 taken 176 times.
✓ Branch 1 taken 48 times.
|
224 | while (!frames.empty()) { |
| 214 | StrassenFrameSTL current = std::move(frames.top()); | ||
| 215 | frames.pop(); | ||
| 216 | |||
| 217 |
2/2✓ Branch 0 taken 144 times.
✓ Branch 1 taken 32 times.
|
176 | if (current.n <= kBaseCaseSize) { |
| 218 |
1/2✓ Branch 1 taken 144 times.
✗ Branch 2 not taken.
|
144 | results.push(BaseMultiply(current.mat_a, current.mat_b, current.n)); |
| 219 | continue; | ||
| 220 | } | ||
| 221 | |||
| 222 |
2/2✓ Branch 0 taken 16 times.
✓ Branch 1 taken 16 times.
|
32 | if (current.stage == 8) { |
| 223 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | std::vector<std::vector<double>> p(7); |
| 224 | |||
| 225 |
2/2✓ Branch 0 taken 112 times.
✓ Branch 1 taken 16 times.
|
128 | for (int i = 6; i >= 0; --i) { |
| 226 | 112 | p[i] = std::move(results.top()); | |
| 227 | results.pop(); | ||
| 228 | } | ||
| 229 | |||
| 230 | 16 | std::size_t h = current.n / 2; | |
| 231 | 16 | std::size_t sz = h * h; | |
| 232 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | std::vector<double> c11(sz); |
| 233 |
1/4✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
16 | std::vector<double> c12(sz); |
| 234 |
1/4✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
|
16 | std::vector<double> c21(sz); |
| 235 |
2/6✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 16 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
|
16 | std::vector<double> c22(sz); |
| 236 | |||
| 237 | 16 | double *c11_ptr = c11.data(); | |
| 238 | 16 | double *c12_ptr = c12.data(); | |
| 239 | 16 | double *c21_ptr = c21.data(); | |
| 240 | 16 | double *c22_ptr = c22.data(); | |
| 241 | |||
| 242 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | RunParallel(0, sz, kParallelThreshold, [&](std::size_t i) { |
| 243 | 262144 | c11_ptr[i] = p[0][i] + p[3][i] - p[4][i] + p[6][i]; | |
| 244 | 262144 | c12_ptr[i] = p[2][i] + p[4][i]; | |
| 245 | 262144 | c21_ptr[i] = p[1][i] + p[3][i]; | |
| 246 | 262144 | c22_ptr[i] = p[0][i] - p[1][i] + p[2][i] + p[5][i]; | |
| 247 | 262144 | }); | |
| 248 | |||
| 249 |
2/6✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 16 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
|
32 | results.push(CombineMatrix(c11, c12, c21, c22, current.n)); |
| 250 | 16 | } else { | |
| 251 | 16 | std::size_t h = current.n / 2; | |
| 252 | 16 | std::vector<double> a11; | |
| 253 | 16 | std::vector<double> a12; | |
| 254 | 16 | std::vector<double> a21; | |
| 255 | 16 | std::vector<double> a22; | |
| 256 | 16 | std::vector<double> b11; | |
| 257 | 16 | std::vector<double> b12; | |
| 258 | 16 | std::vector<double> b21; | |
| 259 | 16 | std::vector<double> b22; | |
| 260 | |||
| 261 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | SplitMatrix(current.mat_a, current.n, a11, a12, a21, a22); |
| 262 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
16 | SplitMatrix(current.mat_b, current.n, b11, b12, b21, b22); |
| 263 | |||
| 264 |
1/2✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
|
32 | frames.push({{}, {}, current.n, 8}); |
| 265 | |||
| 266 | 16 | frames.push({Subtract(a12, a22), Add(b21, b22), h, 0}); | |
| 267 | 16 | frames.push({Subtract(a21, a11), Add(b11, b12), h, 0}); | |
| 268 | 16 | frames.push({Add(a11, a12), b22, h, 0}); | |
| 269 | 16 | frames.push({a22, Subtract(b21, b11), h, 0}); | |
| 270 | 16 | frames.push({a11, Subtract(b12, b22), h, 0}); | |
| 271 | 16 | frames.push({Add(a21, a22), b11, h, 0}); | |
| 272 |
1/4✓ Branch 1 taken 16 times.
✗ Branch 2 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
|
16 | frames.push({Add(a11, a22), Add(b11, b22), h, 0}); |
| 273 | } | ||
| 274 | 176 | } | |
| 275 | |||
| 276 | 48 | return std::move(results.top()); | |
| 277 |
24/48✓ Branch 1 taken 48 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 48 times.
✗ Branch 5 not taken.
✓ Branch 7 taken 48 times.
✗ Branch 8 not taken.
✓ Branch 10 taken 16 times.
✗ Branch 11 not taken.
✓ Branch 13 taken 16 times.
✗ Branch 14 not taken.
✓ Branch 16 taken 16 times.
✗ Branch 17 not taken.
✓ Branch 19 taken 16 times.
✗ Branch 20 not taken.
✓ Branch 22 taken 16 times.
✗ Branch 23 not taken.
✓ Branch 25 taken 16 times.
✗ Branch 26 not taken.
✓ Branch 28 taken 16 times.
✗ Branch 29 not taken.
✓ Branch 31 taken 16 times.
✗ Branch 32 not taken.
✓ Branch 34 taken 16 times.
✗ Branch 35 not taken.
✓ Branch 37 taken 16 times.
✗ Branch 38 not taken.
✓ Branch 40 taken 16 times.
✗ Branch 41 not taken.
✓ Branch 43 taken 16 times.
✗ Branch 44 not taken.
✓ Branch 46 taken 16 times.
✗ Branch 47 not taken.
✓ Branch 49 taken 16 times.
✗ Branch 50 not taken.
✓ Branch 52 taken 16 times.
✗ Branch 53 not taken.
✓ Branch 55 taken 16 times.
✗ Branch 56 not taken.
✓ Branch 58 taken 16 times.
✗ Branch 59 not taken.
✓ Branch 61 taken 16 times.
✗ Branch 62 not taken.
✓ Branch 64 taken 16 times.
✗ Branch 65 not taken.
✓ Branch 67 taken 16 times.
✗ Branch 68 not taken.
✓ Branch 70 taken 16 times.
✗ Branch 71 not taken.
|
160 | } |
| 278 | |||
| 279 | } // namespace tabalaev_a_matrix_mul_strassen | ||
| 280 |