1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
|
/*
* Copyright (c) 2020 Samsung Electronics Co., Ltd. All Rights Reserved
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "WhileLayer.h"
#include <backend/ITensor.h>
#include "exec/ExecutorBase.h"
#include <misc/polymorphic_downcast.h>
#include "PermuteLayer.h"
namespace onert
{
namespace backend
{
namespace controlflow
{
namespace kernel
{
WhileLayer::WhileLayer(const std::vector<std::shared_ptr<backend::ITensor>> &input_tensors,
const std::vector<std::shared_ptr<backend::ITensor>> &output_tensors,
const ir::OperandIndexSequence &output_indices, const ir::Graph &graph,
const exec::DynAllocInfoMap &outputs_dyn_alloc_info,
const ir::SubgraphIndex &cond_subg_index,
const ir::SubgraphIndex &body_subg_index, exec::ExecutorMap *executor_map)
: _cond_subg_index{cond_subg_index}, _body_subg_index{body_subg_index},
_output_indices{output_indices}, _graph{graph}, _input_tensors{input_tensors},
_output_tensors{output_tensors}, _outputs_dyn_alloc_info{outputs_dyn_alloc_info},
_executor_map{executor_map}
{
// At this point, executor_map may not have executors of cond subg and body subg
}
void WhileLayer::run()
{
// Copy "_input_tensors" -> "cond subg inputs"
// Run cond subg
// Start loop while output of cond subg is ture
// // Copy "_input_tensors" -> "body subg inputs" in the first iteration, then copy "body subg
// outputs" -> "body subg inputs" in the second or more iterations
// // Run body subg
// // Copy "body subg outputs" -> "cond subg inputs"
// // Run cond subg
// If there is no loop copy "_input_tensors" -> "_dst_tensors", else copy "cond subg inputs" ->
// "_dst_tensors"
auto cond_exec = nnfw::misc::polymorphic_downcast<exec::ExecutorBase *>(
_executor_map->at(_cond_subg_index).get());
auto body_exec = nnfw::misc::polymorphic_downcast<exec::ExecutorBase *>(
_executor_map->at(_body_subg_index).get());
const auto &cond_graph = cond_exec->graph();
const auto &cond_inputs_dyn_alloc = cond_exec->getInputsDynamicAllocInfo();
const auto &body_graph = body_exec->graph();
const auto &body_inputs_dyn_alloc = body_exec->getInputsDynamicAllocInfo();
std::vector<std::shared_ptr<backend::ITensor>> input_tensors;
std::vector<std::shared_ptr<backend::ITensor>> cond_input_tensors;
std::vector<std::shared_ptr<backend::ITensor>> body_input_tensors;
std::vector<std::shared_ptr<backend::ITensor>> body_output_tensors;
std::vector<std::shared_ptr<backend::ITensor>> output_tensors;
// Add only used tensors in cond subgraph
assert(cond_graph.getInputs().size() == _input_tensors.size());
assert(cond_graph.getInputs().size() == cond_exec->getInputTensors().size());
for (uint32_t i = 0; i < cond_graph.getInputs().size(); ++i)
{
const auto &cond_input = cond_graph.operands().at(cond_graph.getInputs().at(i));
if (cond_input.getUses().size() > 0)
{
input_tensors.emplace_back(_input_tensors.at(i));
cond_input_tensors.emplace_back(cond_exec->getInputTensors().at(i));
}
}
const auto permute_op_input_to_cond_input =
std::make_shared<PermuteLayer>(input_tensors, cond_input_tensors, cond_inputs_dyn_alloc);
// Add only used tensors among outputs of while operation
assert(_output_indices.size() == _input_tensors.size());
assert(_output_indices.size() == _output_tensors.size());
input_tensors.clear();
output_tensors.clear();
for (size_t i = 0; i < _output_indices.size(); ++i)
{
const auto &output_index = _output_indices.at(i);
const auto &output = _graph.operands().at(output_index);
if (output.getUses().size() > 0 || _graph.getOutputs().contains(output_index))
{
input_tensors.emplace_back(_input_tensors.at(i));
output_tensors.emplace_back(_output_tensors.at(i));
}
}
const auto permute_op_input_to_op_output =
std::make_shared<PermuteLayer>(input_tensors, output_tensors, _outputs_dyn_alloc_info);
// Add all tensors with unused tensors in body subgraph because unused input tensors will be
// copied output tensors in body subgraph
assert(_input_tensors.size() == body_exec->getInputTensors().size());
input_tensors = _input_tensors;
body_input_tensors = body_exec->getInputTensors();
const auto permute_op_input_to_body_input =
std::make_shared<PermuteLayer>(input_tensors, body_input_tensors, body_inputs_dyn_alloc);
// Add only used tensors in cond subgraph
assert(cond_graph.getInputs().size() == body_exec->getOutputTensors().size());
assert(cond_graph.getInputs().size() == cond_exec->getInputTensors().size());
body_output_tensors.clear();
cond_input_tensors.clear();
for (uint32_t i = 0; i < cond_graph.getInputs().size(); ++i)
{
const auto &cond_input = cond_graph.operands().at(cond_graph.getInputs().at(i));
if (cond_input.getUses().size() > 0)
{
body_output_tensors.emplace_back(body_exec->getOutputTensors().at(i));
cond_input_tensors.emplace_back(cond_exec->getInputTensors().at(i));
}
}
const auto permute_body_output_to_cond_input = std::make_shared<PermuteLayer>(
body_output_tensors, cond_input_tensors, cond_inputs_dyn_alloc);
// Add only used tensors in body subgraph
assert(body_graph.getInputs().size() == body_exec->getOutputTensors().size());
assert(body_graph.getInputs().size() == body_exec->getInputTensors().size());
body_output_tensors.clear();
body_input_tensors.clear();
for (uint32_t i = 0; i < body_graph.getInputs().size(); ++i)
{
const auto &body_input_index = body_graph.getInputs().at(i);
const auto &body_input = body_graph.operands().at(body_input_index);
if (body_input.getUses().size() > 0 &&
!body_exec->graph().getOutputs().contains(body_input_index))
{
body_output_tensors.emplace_back(body_exec->getOutputTensors().at(i));
body_input_tensors.emplace_back(body_exec->getInputTensors().at(i));
}
}
const auto permute_body_output_to_body_input = std::make_shared<PermuteLayer>(
body_output_tensors, body_input_tensors, body_inputs_dyn_alloc);
// Add only used tensors among outputs of while operation
assert(_output_indices.size() == body_exec->getOutputTensors().size());
assert(_output_indices.size() == _output_tensors.size());
body_output_tensors.clear();
output_tensors.clear();
for (size_t i = 0; i < _output_indices.size(); ++i)
{
const auto &output_index = _output_indices.at(i);
const auto &output = _graph.operands().at(output_index);
if (output.getUses().size() > 0 || _graph.getOutputs().contains(output_index))
{
body_output_tensors.emplace_back(body_exec->getOutputTensors().at(i));
output_tensors.emplace_back(_output_tensors.at(i));
}
}
const auto permute_body_output_to_op_output =
std::make_shared<PermuteLayer>(body_output_tensors, output_tensors, _outputs_dyn_alloc_info);
// Remove copying of unused tensor
permute_op_input_to_cond_input->prepare();
permute_op_input_to_op_output->prepare();
permute_op_input_to_body_input->prepare();
permute_body_output_to_cond_input->prepare();
permute_body_output_to_body_input->prepare();
permute_body_output_to_op_output->prepare();
cond_exec->execute(_input_tensors, permute_op_input_to_cond_input);
assert(cond_exec->getOutputTensors().size() == 1);
auto &cond_output_tensor = cond_exec->getOutputTensors().at(0);
auto getResultCond = [](backend::ITensor *tensor) -> bool {
bool ret = false;
tensor->access([&](ITensor &tensor) { ret = *reinterpret_cast<bool *>(tensor.buffer()); });
return ret;
};
const auto body_execute_with_op_inputs = [&]() {
body_exec->execute(_input_tensors, permute_op_input_to_body_input);
};
const auto body_execute_with_body_outputs = [&]() {
body_exec->execute(body_exec->getOutputTensors(), permute_body_output_to_body_input);
};
std::function<void()> body_execute = body_execute_with_op_inputs;
const auto cond_execute = [&]() {
cond_exec->execute(body_exec->getOutputTensors(), permute_body_output_to_cond_input);
};
auto permute_to_outputs_fn = permute_op_input_to_op_output;
// Loop while Cond subgraph's output is true
while (getResultCond(cond_output_tensor.get()))
{
body_execute();
cond_execute();
body_execute = body_execute_with_body_outputs;
permute_to_outputs_fn = permute_body_output_to_op_output;
}
permute_to_outputs_fn->run();
}
} // namespace kernel
} // namespace controlflow
} // namespace backend
} // namespace onert
|
