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
|
/*
* Copyright (c) 2019 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.
*/
#ifndef __NEURUN_BACKEND_EXEC_TIME_H__
#define __NEURUN_BACKEND_EXEC_TIME_H__
#include "backend/Backend.h"
#include "backend/IConfig.h"
#include "JSONExecTime.h"
#include <memory>
#include <limits>
#include <map>
#include <unordered_map>
#include <vector>
namespace neurun
{
namespace backend
{
class ExecTime
{
public:
explicit ExecTime(const std::vector<const Backend *> &backends) : _json(backends, _measurements)
{
}
public:
/**
* @brief Get exec time of an operation with input size
* or linearly interpolated value based on size if there is no record for given size
*
* @param[in] backend id of a backend
* @param[in] operation name of an operation
* @param[in] quant if input type quantized
* @param[in] op_size sum of operation's flattened sizes of inputs and outputs
* @return execution time for given input sizes
* -1 if there are no records for given parameters (backend, op, quantization).
*/
int64_t getOperationExecTime(const Backend *backend, const std::string &operation, bool quant,
uint32_t op_size) const;
/**
* @brief Update exec time of the operation on a backend with given input size or
* add new entity if there is no one.
*
* @param[in] backend id of a backend
* @param[in] operation name of an operation
* @param[in] quant if input type quantized
* @param[in] op_size sum of operation's flattened sizes of inputs and outputs
* @param[in] time real measured value
*/
void updateOperationExecTime(const Backend *backend, const std::string &operation, bool quant,
uint32_t op_size, int64_t time);
/**
* @brief Get the permute time from one backend to another
*
* @param[in] from_backend
* @param[in] to_backend
* @param[in] quant if input type quantized
* @param[in] op_size sum of operation's flattened sizes of inputs and outputs
* @return permutation time for operation size
*/
int64_t getPermuteTime(const Backend *from_backend, const Backend *to_backend, bool quant,
uint32_t op_size) const;
/**
* @brief Update permute time from one backend to another
*
* @param[in] from_backend
* @param[in] to_backend
* @param[in] quant if input type quantized
* @param[in] time measured permutation time
* @param[in] op_size sum of operation's flattened sizes of inputs and outputs
*/
void updatePermuteTime(const Backend *from_backend, const Backend *to_backend, bool quant,
uint32_t op_size, int64_t time);
/**
* @brief Get the max value of int32_t in int64_t
* @return max value
*/
static int64_t getMax() { return _MAX; }
/**
* @brief Update metrics file with new data.
*/
void uploadOperationsExecTime() const { _json.uploadOperationsExecTime(); }
static const int64_t NOT_FOUND = -1;
private:
/// @brief Measurement data, which is shared with serializer
MeasurementData _measurements;
// int64_t::max may cause integer overflow
static const int64_t _MAX = std::numeric_limits<int32_t>::max();
/// @brief Serializer
JSON _json;
};
} // namespace backend
} // namespace neurun
#endif // __NEURUN_BACKEND_EXEC_TIME_H__
|
