| 
									
										
										
										
											2019-07-19 17:09:09 +08:00
										 |  |  | //
 | 
					
						
							|  |  |  | //  ShapeCosineSimilarity.cpp
 | 
					
						
							|  |  |  | //  MNN
 | 
					
						
							|  |  |  | //
 | 
					
						
							|  |  |  | //  Created by MNN on 2019/7/17.
 | 
					
						
							|  |  |  | //  Copyright © 2018, Alibaba Group Holding Limited
 | 
					
						
							|  |  |  | //
 | 
					
						
							|  |  |  | 
 | 
					
						
							| 
									
										
										
										
											2019-12-27 22:16:57 +08:00
										 |  |  | #include "core/Macro.h"
 | 
					
						
							|  |  |  | #include "core/SizeComputer.hpp"
 | 
					
						
							|  |  |  | #include "core/TensorUtils.hpp"
 | 
					
						
							| 
									
										
										
										
											2019-07-19 17:09:09 +08:00
										 |  |  | 
 | 
					
						
							|  |  |  | namespace MNN { | 
					
						
							|  |  |  | class CosineSimilaritySize : public SizeComputer { | 
					
						
							|  |  |  |     virtual bool onComputeSize(const MNN::Op* op, const std::vector<Tensor*>& inputs, | 
					
						
							|  |  |  |                                const std::vector<Tensor*>& outputs) const override { | 
					
						
							|  |  |  |         MNN_ASSERT(4 == inputs.size()); | 
					
						
							|  |  |  |         auto x1        = inputs[0]; | 
					
						
							|  |  |  |         auto x2        = inputs[1]; | 
					
						
							|  |  |  |         auto dimTensor = inputs[2]; | 
					
						
							|  |  |  |         const auto dim = dimTensor->host<int32_t>()[0]; | 
					
						
							|  |  |  |         MNN_ASSERT(dim == 1); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         const int dimensions0 = x1->dimensions(); | 
					
						
							|  |  |  |         const int dimensions1 = x2->dimensions(); | 
					
						
							|  |  |  |         MNN_ASSERT(dimensions0 == dimensions1); | 
					
						
							|  |  |  |         for (int i = 0; i < dimensions0; ++i) { | 
					
						
							|  |  |  |             MNN_ASSERT(x1->length(i) == x2->length(i)); | 
					
						
							|  |  |  |         } | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  |         auto output                 = outputs[0]; | 
					
						
							|  |  |  |         output->buffer().dimensions = dimensions0 - 1; | 
					
						
							|  |  |  |         for (int i = 0; i < dimensions0; ++i) { | 
					
						
							|  |  |  |             int index = i; | 
					
						
							|  |  |  |             if (i == dim) { | 
					
						
							|  |  |  |                 continue; | 
					
						
							|  |  |  |             } | 
					
						
							|  |  |  |             if (i > dim) { | 
					
						
							|  |  |  |                 index = i - 1; | 
					
						
							|  |  |  |             } | 
					
						
							|  |  |  |             output->setLength(index, x1->length(i)); | 
					
						
							|  |  |  |         } | 
					
						
							| 
									
										
										
										
											2020-02-26 09:57:17 +08:00
										 |  |  |         output->buffer().type = x1->getType(); | 
					
						
							| 
									
										
										
										
											2019-07-25 13:36:35 +08:00
										 |  |  |         TensorUtils::getDescribe(output)->dimensionFormat = MNN_DATA_FORMAT_NCHW; | 
					
						
							| 
									
										
										
										
											2019-07-19 17:09:09 +08:00
										 |  |  |         return true; | 
					
						
							|  |  |  |     } | 
					
						
							|  |  |  | }; | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | REGISTER_SHAPE(CosineSimilaritySize, OpType_CosineSimilarity); | 
					
						
							|  |  |  | 
 | 
					
						
							|  |  |  | } // namespace MNN
 |