1 /** 2 * Copyright 2021 Huawei Technologies Co., Ltd 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17 #ifndef MINDSPORE_LITE_SRC_RUNTIME_DELEGATE_NPU_NPU_SUBGRAPH_H_ 18 #define MINDSPORE_LITE_SRC_RUNTIME_DELEGATE_NPU_NPU_SUBGRAPH_H_ 19 20 #include <memory> 21 #include <vector> 22 #include <string> 23 #include "include/api/kernel.h" 24 #include "src/litert/delegate/npu/npu_executor.h" 25 26 namespace mindspore::lite { 27 class NPUSubGraph : public kernel::Kernel { 28 public: NPUSubGraph(const std::vector<NPUOp * > & npu_ops,NPUManager * npu_manager)29 NPUSubGraph(const std::vector<NPUOp *> &npu_ops, NPUManager *npu_manager) 30 : npu_ops_(npu_ops), npu_manager_(npu_manager) {} 31 32 ~NPUSubGraph() override; 33 34 int Init(); 35 36 int Prepare() override; 37 38 int Execute() override; 39 ReSize()40 int ReSize() override { 41 MS_LOG(ERROR) << "NPU does not support the resize function temporarily."; 42 return RET_ERROR; 43 } 44 45 void set_input(mindspore::MSTensor in_tensor, int index) override; 46 47 void set_output(mindspore::MSTensor out_tensor, int index) override; 48 49 int GetGraphInOutOps(); 50 51 std::vector<NPUOp *> FindPreOps(NPUOp *cur_op); 52 UpdateInputMappingRelationShip(const std::vector<size_t> & input_index)53 void UpdateInputMappingRelationShip(const std::vector<size_t> &input_index) { input_relationship_ = input_index; } 54 55 private: 56 std::shared_ptr<domi::ModelBufferData> BuildIRModel(); 57 58 int BuildNPUInputOp(); 59 60 int BuildNPUOutputOp(); 61 62 int GetNPUOperators(const std::vector<NPUOp *> &ops); 63 64 bool IsSubGraphInputTensor(const mindspore::MSTensor &input); 65 66 std::string GetOMModelName(); 67 68 std::string GetModelHash(); 69 70 bool is_compiled_ = false; 71 72 std::vector<ge::Operator> subgraph_input_ops_; 73 74 std::vector<ge::Operator> subgraph_output_ops_; 75 76 std::vector<mindspore::MSTensor> out_tensor_sorted_; 77 78 std::vector<mindspore::MSTensor> all_tensors_from_out_ops_; 79 80 std::vector<ge::Operator *> op_buffer_; 81 82 std::vector<NPUOp *> npu_ops_{}; 83 // entry nodes in nodes 84 std::vector<NPUOp *> in_ops_{}; 85 // exit nodes in nodes 86 std::vector<NPUOp *> out_ops_{}; 87 88 NPUExecutor *executor_ = nullptr; 89 90 NPUManager *npu_manager_ = nullptr; 91 92 std::vector<size_t> input_relationship_; 93 }; 94 95 } // namespace mindspore::lite 96 97 #endif // MINDSPORE_LITE_SRC_RUNTIME_DELEGATE_NPU_NPU_SUBGRAPH_H_ 98