SiCong Li | b63b119 | 2022-01-28 18:24:39 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2022 Arm Limited. |
| 3 | * |
| 4 | * SPDX-License-Identifier: MIT |
| 5 | * |
| 6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
| 7 | * of this software and associated documentation files (the "Software"), to |
| 8 | * deal in the Software without restriction, including without limitation the |
| 9 | * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or |
| 10 | * sell copies of the Software, and to permit persons to whom the Software is |
| 11 | * furnished to do so, subject to the following conditions: |
| 12 | * |
| 13 | * The above copyright notice and this permission notice shall be included in all |
| 14 | * copies or substantial portions of the Software. |
| 15 | * |
| 16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
| 17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
| 19 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
| 20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
| 21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE |
| 22 | * SOFTWARE. |
| 23 | */ |
SiCong Li | 4e9f568 | 2022-05-10 10:15:59 +0100 | [diff] [blame] | 24 | #ifdef ENABLE_EXPERIMENTAL_DYNAMIC_FUSION |
SiCong Li | b63b119 | 2022-01-28 18:24:39 +0000 | [diff] [blame] | 25 | #include "arm_compute/core/experimental/DependencyGraph.h" |
| 26 | |
| 27 | #include "tests/framework/Asserts.h" |
| 28 | #include "tests/framework/Macros.h" |
| 29 | |
| 30 | using namespace arm_compute::experimental::dynamic_fusion; |
| 31 | |
| 32 | namespace arm_compute |
| 33 | { |
| 34 | namespace test |
| 35 | { |
| 36 | namespace validation |
| 37 | { |
| 38 | TEST_SUITE(CL) |
| 39 | |
| 40 | TEST_SUITE(UNIT) |
| 41 | TEST_SUITE(DYNAMIC_FUSION) |
| 42 | TEST_SUITE(DependencyGraph) |
| 43 | |
| 44 | TEST_CASE(Correct_Graph_Creation_Should_Pass, framework::DatasetMode::ALL) |
| 45 | { |
| 46 | DependencyGraph graph{}; |
| 47 | const auto t0 = graph.add_tensor(); |
| 48 | const auto t1 = graph.add_tensor(); |
| 49 | const auto t2 = graph.add_tensor(); |
| 50 | const auto t3 = graph.add_tensor(); |
| 51 | const auto t4 = graph.add_tensor(); |
| 52 | |
| 53 | const auto o0 = graph.add_operator({ t0, t1 }, { t2 }).second; |
| 54 | const auto o1 = graph.add_operator({ t3, t2 }, { t4 }).second; |
| 55 | |
| 56 | ARM_COMPUTE_EXPECT_EQUAL(graph.number_of_ops(), 2U, framework::LogLevel::ERRORS); |
| 57 | ARM_COMPUTE_EXPECT_EQUAL(graph.number_of_tensors(), 5U, framework::LogLevel::ERRORS); |
| 58 | |
| 59 | const DependencyGraph ref_graph |
| 60 | { |
| 61 | { |
| 62 | // src_tensors |
| 63 | { o0, { t0, t1 } }, |
| 64 | { o1, { t3, t2 } }, |
| 65 | }, |
| 66 | { |
| 67 | // dst_tensors |
| 68 | { o0, { t2 } }, |
| 69 | { o1, { t4 } }, |
| 70 | }, |
| 71 | { |
| 72 | // src_ops |
| 73 | { t0, {} }, |
| 74 | { t1, {} }, |
| 75 | { t2, { o0 } }, |
| 76 | { t3, {} }, |
| 77 | { t4, { o1 } }, |
| 78 | }, |
| 79 | { |
| 80 | // dst_ops |
| 81 | { t0, { o0 } }, |
| 82 | { t1, { o0 } }, |
| 83 | { t2, { o1 } }, |
| 84 | { t3, { o1 } }, |
| 85 | { t4, {} }, |
| 86 | } |
| 87 | |
| 88 | }; |
| 89 | ARM_COMPUTE_EXPECT(graph == ref_graph, framework::LogLevel::ERRORS); |
| 90 | } |
| 91 | |
| 92 | TEST_CASE(Correct_Merge_Points_Should_Enable_Graph_Expansion, framework::DatasetMode::ALL) |
| 93 | { |
| 94 | // Merge points are a simple way to collapse "graph of graphs" into a single graph |
| 95 | // Suppose we have a top-level graph g0 |
| 96 | DependencyGraph g0{}; |
| 97 | const auto g0_t0 = g0.add_tensor(); |
| 98 | const auto g0_t1 = g0.add_tensor(); |
| 99 | const auto g0_t2 = g0.add_tensor(); |
| 100 | const auto g0_t3 = g0.add_tensor(); |
| 101 | const auto g0_t4 = g0.add_tensor(); |
| 102 | g0.add_operator({ g0_t0, g0_t1 }, { g0_t2 }); // g0_o0 |
| 103 | g0.add_operator({ g0_t3, g0_t2 }, { g0_t4 }); // g0_o1 |
| 104 | |
| 105 | // Then g0 expands into g1, with additional nodes added in-between "merge point tensors" |
| 106 | // Note that the expansion logic may be local to each operator node |
| 107 | DependencyGraph g1{}; |
| 108 | // g0_o0 expands into g1_o0, g1_o1, g1_o2 |
| 109 | const auto g1_t0 = g1.add_tensor(g0_t0); |
| 110 | const auto g1_t1 = g1.add_tensor(g0_t1); |
| 111 | const auto g1_t2 = g1.add_tensor(); |
| 112 | const auto g1_t3 = g1.add_tensor(); |
| 113 | const auto g1_t4 = g1.add_tensor(g0_t2); |
| 114 | const auto g1_o0 = g1.add_operator({ g1_t0 }, { g1_t2 }).second; |
| 115 | const auto g1_o1 = g1.add_operator({ g1_t1 }, { g1_t3 }).second; |
| 116 | const auto g1_o2 = g1.add_operator({ g1_t2, g1_t3 }, { g1_t4 }).second; |
| 117 | |
| 118 | // g0_o1 expands into g1_o3 |
| 119 | const auto g1_t5 = g1.add_tensor(g0_t3); |
| 120 | const auto g1_t6 = g1.add_tensor(g0_t2); |
| 121 | const auto g1_t7 = g1.add_tensor(g0_t4); |
| 122 | ARM_COMPUTE_EXPECT_EQUAL(g1_t4, g1_t6, framework::LogLevel::ERRORS); // both associate with the same merge point g0_t2, thus they should point to the same tensor in g1 |
| 123 | const auto g1_o3 = g1.add_operator({ g1_t5, g1_t6 }, { g1_t7 }).second; |
| 124 | |
| 125 | const DependencyGraph ref_graph |
| 126 | { |
| 127 | { |
| 128 | // src_tensors |
| 129 | { g1_o0, { g1_t0 } }, |
| 130 | { g1_o1, { g1_t1 } }, |
| 131 | { g1_o2, { g1_t2, g1_t3 } }, |
| 132 | { g1_o3, { g1_t5, g1_t4 } }, |
| 133 | }, |
| 134 | { |
| 135 | // dst_tensors |
| 136 | { g1_o0, { g1_t2 } }, |
| 137 | { g1_o1, { g1_t3 } }, |
| 138 | { g1_o2, { g1_t4 } }, |
| 139 | { g1_o3, { g1_t7 } }, |
| 140 | }, |
| 141 | { |
| 142 | // src_ops |
| 143 | { g1_t0, {} }, |
| 144 | { g1_t1, {} }, |
| 145 | { g1_t2, { g1_o0 } }, |
| 146 | { g1_t3, { g1_o1 } }, |
| 147 | { g1_t4, { g1_o2 } }, |
| 148 | { g1_t5, {} }, |
| 149 | { g1_t7, { g1_o3 } }, |
| 150 | }, |
| 151 | { |
| 152 | // dst_ops |
| 153 | { g1_t0, { g1_o0 } }, |
| 154 | { g1_t1, { g1_o1 } }, |
| 155 | { g1_t2, { g1_o2 } }, |
| 156 | { g1_t3, { g1_o2 } }, |
| 157 | { g1_t4, { g1_o3 } }, |
| 158 | { g1_t5, { g1_o3 } }, |
| 159 | { g1_t7, {} }, |
| 160 | }, |
| 161 | { |
| 162 | // merge points |
| 163 | { g0_t0, g1_t0 }, |
| 164 | { g0_t1, g1_t1 }, |
| 165 | { g0_t2, g1_t4 }, |
| 166 | { g0_t3, g1_t5 }, |
| 167 | { g0_t4, g1_t7 }, |
| 168 | } |
| 169 | }; |
| 170 | ARM_COMPUTE_EXPECT(g1 == ref_graph, framework::LogLevel::ERRORS); |
| 171 | } |
| 172 | |
| 173 | TEST_CASE(Path_Existence_Check_0, framework::DatasetMode::ALL) |
| 174 | { |
| 175 | DependencyGraph graph{}; |
| 176 | const auto t0 = graph.add_tensor(); |
| 177 | const auto t1 = graph.add_tensor(); |
| 178 | const auto t2 = graph.add_tensor(); |
| 179 | const auto t3 = graph.add_tensor(); |
| 180 | const auto t4 = graph.add_tensor(); |
| 181 | const auto t5 = graph.add_tensor(); |
| 182 | const auto t6 = graph.add_tensor(); |
| 183 | const auto t7 = graph.add_tensor(); |
| 184 | const auto o0 = graph.add_operator({ t1 }, { t3, t4 }).second; |
| 185 | const auto o1 = graph.add_operator({ t3 }, { t5 }).second; |
| 186 | const auto o2 = graph.add_operator({ t5, t6 }, { t7 }).second; |
| 187 | const auto o3 = graph.add_operator({ t4 }, { t6 }).second; |
| 188 | const auto o4 = graph.add_operator({ t0, t5 }, { t2 }).second; |
| 189 | |
| 190 | ARM_COMPUTE_UNUSED(o1, o3); |
| 191 | |
| 192 | ARM_COMPUTE_EXPECT((graph.path_exists_from_tensor_to_op(t3, o2)), framework::LogLevel::ERRORS); |
| 193 | ARM_COMPUTE_EXPECT((graph.path_exists_from_tensor_to_op(t1, o4)), framework::LogLevel::ERRORS); |
| 194 | ARM_COMPUTE_EXPECT(!(graph.path_exists_from_tensor_to_op(t2, o4)), framework::LogLevel::ERRORS); |
| 195 | ARM_COMPUTE_EXPECT(!(graph.path_exists_from_tensor_to_op(t0, o2)), framework::LogLevel::ERRORS); |
| 196 | |
| 197 | ARM_COMPUTE_EXPECT((graph.path_exists_from_op_to_op(o0, o2)), framework::LogLevel::ERRORS); |
| 198 | ARM_COMPUTE_EXPECT(!(graph.path_exists_from_op_to_op(o2, o0)), framework::LogLevel::ERRORS); |
| 199 | |
| 200 | ARM_COMPUTE_EXPECT(!(graph.path_exists_from_op_to_op(o2, o4)), framework::LogLevel::ERRORS); |
| 201 | } |
| 202 | |
| 203 | TEST_CASE(Correct_Topological_Sort_Should_Pass, framework::DatasetMode::ALL) |
| 204 | { |
| 205 | DependencyGraph graph{}; |
| 206 | const auto t0 = graph.add_tensor(); |
| 207 | const auto t1 = graph.add_tensor(); |
| 208 | const auto t2 = graph.add_tensor(); |
| 209 | const auto t3 = graph.add_tensor(); |
| 210 | const auto t4 = graph.add_tensor(); |
| 211 | const auto t5 = graph.add_tensor(); |
| 212 | const auto t6 = graph.add_tensor(); |
| 213 | const auto t7 = graph.add_tensor(); |
| 214 | const auto o0 = graph.add_operator({ t1 }, { t3, t4 }).second; |
| 215 | const auto o1 = graph.add_operator({ t3 }, { t5 }).second; |
| 216 | const auto o2 = graph.add_operator({ t5, t6 }, { t7 }).second; |
| 217 | const auto o3 = graph.add_operator({ t4 }, { t6 }).second; |
| 218 | const auto o4 = graph.add_operator({ t0, t5 }, { t2 }).second; |
| 219 | |
| 220 | const auto res = graph.topological_sort(); |
| 221 | ARM_COMPUTE_EXPECT(bool(res.first), framework::LogLevel::ERRORS); |
| 222 | std::vector<DependencyGraph::OpPack> ref_sorted_op_packs |
| 223 | { |
| 224 | { o0, { t1 }, { t3, t4 } }, |
| 225 | { o1, { t3 }, { t5 } }, |
| 226 | { o3, { t4 }, { t6 } }, |
| 227 | { o4, { t0, t5 }, { t2 } }, |
| 228 | { o2, { t5, t6 }, { t7 } }, |
| 229 | |
| 230 | }; |
| 231 | ARM_COMPUTE_EXPECT((res.second == ref_sorted_op_packs), framework::LogLevel::ERRORS); |
| 232 | } |
| 233 | |
| 234 | TEST_CASE(Cycles_Should_Fail, framework::DatasetMode::ALL) |
| 235 | { |
| 236 | DependencyGraph graph{}; |
| 237 | const auto t0 = graph.add_tensor(); |
| 238 | const auto t1 = graph.add_tensor(); |
| 239 | const auto t2 = graph.add_tensor(); |
| 240 | const auto t3 = graph.add_tensor(); |
| 241 | |
| 242 | graph.add_operator({ t0, t1 }, { t2 }); |
| 243 | graph.add_operator({ t2 }, { t1, t3 }); // Ideally error should occur here |
| 244 | |
| 245 | const auto res = graph.topological_sort(); |
| 246 | ARM_COMPUTE_EXPECT(!bool(res.first), framework::LogLevel::ERRORS); |
| 247 | } |
| 248 | TEST_CASE(Loops_Should_Fail, framework::DatasetMode::ALL) |
| 249 | { |
| 250 | DependencyGraph graph{}; |
| 251 | const auto t0 = graph.add_tensor(); |
| 252 | const auto t1 = graph.add_tensor(); |
| 253 | const auto t2 = graph.add_tensor(); |
| 254 | |
| 255 | ARM_COMPUTE_EXPECT_THROW(graph.add_operator({ t0, t2 }, { t1, t2 }).first, framework::LogLevel::ERRORS); |
| 256 | ARM_COMPUTE_UNUSED(t0, t1, t2); |
| 257 | } |
| 258 | TEST_SUITE_END() // DependencyGraph |
| 259 | TEST_SUITE_END() // DYNAMIC_FUSION |
| 260 | TEST_SUITE_END() // UNIT |
| 261 | |
| 262 | TEST_SUITE_END() // CL |
| 263 | } // namespace validation |
| 264 | } // namespace test |
SiCong Li | 4e9f568 | 2022-05-10 10:15:59 +0100 | [diff] [blame] | 265 | } // namespace arm_compute |
| 266 | #endif /* ENABLE_EXPERIMENTAL_DYNAMIC_FUSION */ |