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 #ifndef MINDSPORE_LITE_SRC_RUNTIME_DELEGATE_DELEGATE_UTILS_H_
17 #define MINDSPORE_LITE_SRC_RUNTIME_DELEGATE_DELEGATE_UTILS_H_
18 #include <vector>
19 #include <map>
20 #include <set>
21 #include "src/common/log_adapter.h"
22 #include "include/errorcode.h"
23 #include "core/base/base.h"
24 #include "src/extendrt/delegate/tensorrt/tensor_info.h"
25
26 namespace mindspore::lite {
27 bool IsSubGraphInputTensor(const std::vector<TensorInfo> &inputs, const TensorInfo &input);
28
29 template <typename T>
FindPreOps(T * cur_op,std::vector<T * > all_ops)30 std::vector<T *> FindPreOps(T *cur_op, std::vector<T *> all_ops) {
31 std::vector<T *> in_ops;
32 for (auto in_tensor : cur_op->inputs()) {
33 for (auto op : all_ops) {
34 if (std::find(op->outputs().begin(), op->outputs().end(), in_tensor) != op->outputs().end()) {
35 in_ops.push_back(op);
36 }
37 }
38 }
39 return in_ops;
40 }
41
42 template <typename T>
FindNextOps(T * cur_op,std::vector<T * > all_ops)43 std::vector<T *> FindNextOps(T *cur_op, std::vector<T *> all_ops) {
44 std::vector<T *> out_ops;
45 for (auto out_tensor : cur_op->outputs()) {
46 for (auto op : all_ops) {
47 if (std::find(op->inputs().begin(), op->inputs().end(), out_tensor) != op->inputs().end()) {
48 out_ops.push_back(op);
49 }
50 }
51 }
52 return out_ops;
53 }
54
55 template <typename T>
FindPreNextOps(std::vector<T * > all_ops)56 void FindPreNextOps(std::vector<T *> all_ops) {
57 std::map<TensorInfo, std::set<T *>> in_tensor_op;
58 std::map<TensorInfo, std::set<T *>> out_tensor_op;
59 for (auto op : all_ops) {
60 for (auto in_tensor : op->inputs()) {
61 in_tensor_op[in_tensor].insert(op);
62 }
63 for (auto out_tensor : op->outputs()) {
64 out_tensor_op[out_tensor].insert(op);
65 }
66 }
67 for (auto op : all_ops) {
68 std::set<T *> in_ops_set;
69 for (auto in_tensor : op->inputs()) {
70 auto in_ops = out_tensor_op[in_tensor];
71 in_ops_set.insert(in_ops.begin(), in_ops.end());
72 }
73 std::vector<T *> in_ops_vec;
74 in_ops_vec.assign(in_ops_set.begin(), in_ops_set.end());
75 op->set_in_ops(in_ops_vec);
76
77 std::set<T *> out_ops_set;
78 for (auto out_tensor : op->outputs()) {
79 auto out_ops = in_tensor_op[out_tensor];
80 out_ops_set.insert(out_ops.begin(), out_ops.end());
81 }
82 std::vector<T *> out_ops_vec;
83 out_ops_vec.assign(out_ops_set.begin(), out_ops_set.end());
84 op->set_out_ops(out_ops_vec);
85 }
86 }
87
88 template <typename T>
GetGraphInOutOps(const std::vector<TensorInfo> & inputs,const std::vector<TensorInfo> & outputs,std::vector<T * > * in_ops,std::vector<T * > * out_ops,const std::vector<T * > & all_ops)89 int GetGraphInOutOps(const std::vector<TensorInfo> &inputs, const std::vector<TensorInfo> &outputs,
90 std::vector<T *> *in_ops, std::vector<T *> *out_ops, const std::vector<T *> &all_ops) {
91 for (auto in_tensor : inputs) {
92 for (auto op : all_ops) {
93 if (std::find(op->inputs().begin(), op->inputs().end(), in_tensor) != op->inputs().end() &&
94 std::find(in_ops->begin(), in_ops->end(), op) == in_ops->end()) {
95 in_ops->push_back(op);
96 }
97 }
98 }
99 if (in_ops->empty()) {
100 MS_LOG(ERROR) << "Can't find the input ops for npu sub graph.";
101 return RET_ERROR;
102 }
103
104 for (auto out_tensor : outputs) {
105 for (auto op : all_ops) {
106 if (std::find(op->outputs().begin(), op->outputs().end(), out_tensor) != op->outputs().end() &&
107 std::find(out_ops->begin(), out_ops->end(), op) == out_ops->end()) {
108 out_ops->push_back(op);
109 }
110 }
111 }
112 if (out_ops->empty()) {
113 MS_LOG(ERROR) << "Can't find the output ops for npu sub graph.";
114 return RET_ERROR;
115 }
116 return RET_OK;
117 }
118 } // namespace mindspore::lite
119
120 #endif // MINDSPORE_LITE_SRC_RUNTIME_DELEGATE_DELEGATE_UTILS_H_
121