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
|
// Copyright (C) 2018-2023 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
// Description: The file defines functions to dump inference performance statistics
#pragma once
#include <map>
#include <string>
#include <utility>
#include <vector>
#include <openvino/openvino.hpp>
#include <samples/common.hpp>
#include <samples/csv_dumper.hpp>
#include <samples/slog.hpp>
#include "utils.hpp"
#include "dla_defines.h"
// @brief statistics reports types
static constexpr char noCntReport[] = "no_counters";
static constexpr char averageCntReport[] = "average_counters";
static constexpr char detailedCntReport[] = "detailed_counters";
/// @brief Responsible for collecting of statistics and dumping to .csv file
class StatisticsReport {
public:
typedef std::vector<ov::ProfilingInfo> PerformanceCounters;
typedef std::vector<std::pair<std::string, std::string>> Parameters;
struct Config {
bool save_report;
std::string report_folder;
};
enum class Category {
COMMAND_LINE_PARAMETERS,
RUNTIME_CONFIG,
EXECUTION_RESULTS,
};
explicit StatisticsReport(Config config) : _config(std::move(config)) {
_separator = dla::util::path_separator;
if (_config.report_folder.empty()) _separator = "";
}
void addParameters(const Category &category, const Parameters ¶meters);
void dump();
/// print the performance counters for neural net layers executed on the CPU.
/// @param perfCounts vector of map of layer name and InferenceEngineProfileInfo.
/// @param sortFlag One of "sort", "no_sort", "simple_sort".
/// "sort": sort by execution RealTime. Default value.
/// "no_sort": no sort.
/// "simple_sort": sort by execution RealTime after removing nodes with "NOT_RUN"
/// status.
void printPerfCountersSort(const std::vector<PerformanceCounters> &perfCounts, std::string sortFlag = "sort");
/// Helper function used by printPerfCountersSort that prints a row of performance count info.
/// prints the following info for a layer from left to right:
/// 0. nodeName: name of the layer
/// 1. LayerStatus: NOT_RUN, OPTIMIZED_OUT, or EXECUTED
/// 2. LayerType: type of layer, such as Convolution.
/// 3. RealTime (ms): The absolute time that the layer ran (in total), including CPU processing time + any potential
/// wait time.
/// 4. CPUTime (ms): The net host cpu time that the layer ran, i.e. CPU processing time.
/// 5. Proportion: RealTime of the node / RealTime in total
/// 6. ExecType: An execution type of unit. e.g., jit_avx2_FP32 (executed using just-in-time (JIT) compilation with
/// AVX2 instructions for FP32 data)
/// @param result_list vector of per-node info, where each per-node info is a vector of formatted string.
void printDetailResult(std::vector<std::vector<std::string>> result_list);
private:
// configuration of current benchmark execution
const Config _config;
// parameters
std::map<Category, Parameters> _parameters;
// csv separator
std::string _separator;
};
|
