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
|
/*
* Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright (C) 2017 The Android Open Source Project
*
* 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.
*/
#ifndef __NNFW_RT_CPU_EXECUTOR_H__
#define __NNFW_RT_CPU_EXECUTOR_H__
#include "HalInterfaces.h"
#include "OperationsUtils.h"
#include "Utils.h"
#include <algorithm>
#include <vector>
namespace nnfw {
namespace rt {
// Information we maintain about each operand during execution that
// may change during execution.
struct RunTimeOperandInfo {
// TODO Storing the type here is redundant, as it won't change during execution.
OperandType type;
// The type and dimensions of the operand. The dimensions can
// change at runtime. We include the type because it's useful
// to pass together with the dimension to the functions implementing
// the operators.
std::vector<uint32_t> dimensions;
float scale;
int32_t zeroPoint;
// Where the operand's data is stored. Check the corresponding
// location information in the model to figure out if this points
// to memory we have allocated for an temporary operand.
uint8_t* buffer;
// The length of the buffer.
uint32_t length;
// Whether this is a temporary variable, a model input, a constant, etc.
OperandLifeTime lifetime;
// Keeps track of how many operations have yet to make use
// of this temporary variable. When the count is decremented to 0,
// we free the buffer. For non-temporary variables, this count is
// always 0.
uint32_t numberOfUsesLeft;
Shape shape() const {
return Shape{.type = type, .dimensions = dimensions, .scale = scale, .offset = zeroPoint};
}
};
// Used to keep a pointer to each of the memory pools.
struct RunTimePoolInfo {
#if 0 // REF-ANN
sp<IMemory> memory;
#endif
hidl_memory hidlMemory;
uint8_t* buffer;
bool set(const hidl_memory& hidlMemory);
bool update();
};
bool setRunTimePoolInfosFromHidlMemories(std::vector<RunTimePoolInfo>* poolInfos,
const hidl_vec<hidl_memory>& pools);
// This class is used to execute a model on the CPU.
class CpuExecutor {
public:
// Executes the model. The results will be stored at the locations
// specified in the constructor.
// The model must outlive the executor. We prevent it from being modified
// while this is executing.
int run(const Model& model, const Request& request,
const std::vector<RunTimePoolInfo>& modelPoolInfos,
const std::vector<RunTimePoolInfo>& requestPoolInfos);
private:
bool initializeRunTimeInfo(const std::vector<RunTimePoolInfo>& modelPoolInfos,
const std::vector<RunTimePoolInfo>& requestPoolInfos);
// Runs one operation of the graph.
int executeOperation(const Operation& entry);
// Decrement the usage count for the operands listed. Frees the memory
// allocated for any temporary variable with a count of zero.
void freeNoLongerUsedOperands(const std::vector<uint32_t>& inputs);
// The model and the request that we'll execute. Only valid while run()
// is being executed.
const Model* mModel = nullptr;
const Request* mRequest = nullptr;
// We're copying the list of all the dimensions from the model, as
// these may be modified when we run the operatins. Since we're
// making a full copy, the indexes used in the operand description
// stay valid.
// std::vector<uint32_t> mDimensions;
// Runtime information about all the operands.
std::vector<RunTimeOperandInfo> mOperands;
};
namespace {
template <typename T>
T getScalarData(const RunTimeOperandInfo& info) {
// TODO: Check buffer is at least as long as size of data.
T* data = reinterpret_cast<T*>(info.buffer);
return data[0];
}
inline bool IsNullInput(const RunTimeOperandInfo *input) {
return input->lifetime == OperandLifeTime::NO_VALUE;
}
#if 0 // REF-ANN
inline int NumInputsWithValues(const Operation &operation,
std::vector<RunTimeOperandInfo> &operands) {
const std::vector<uint32_t> &inputs = operation.inputs;
return std::count_if(inputs.begin(), inputs.end(),
[&operands](uint32_t i) {
return !IsNullInput(&operands[i]);
});
}
inline int NumOutputs(const Operation &operation) {
return operation.outputs.size();
}
inline size_t NumDimensions(const RunTimeOperandInfo *operand) {
return operand->shape().dimensions.size();
}
inline uint32_t SizeOfDimension(const RunTimeOperandInfo *operand, int i) {
return operand->shape().dimensions[i];
}
inline RunTimeOperandInfo *GetInput(const Operation &operation,
std::vector<RunTimeOperandInfo> &operands,
int index) {
return &operands[operation.inputs[index]];
}
inline RunTimeOperandInfo *GetOutput(const Operation &operation,
std::vector<RunTimeOperandInfo> &operands,
int index) {
return &operands[operation.outputs[index]];
}
#endif
} // anonymous namespace
} // namespace rt
} // namespace nnfw
#endif // __NNFW_RT_CPU_EXECUTOR_H__
|
