nn_gpu/prepare_model.cpp at master · projectceladon/nn_gpu · GitHub

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
#include <algorithm>
#include <memory.h>
#include <string.h>

#include <hidl/LegacySupport.h>
#include <thread>

#include "prepare_model.h"
#include "executor_manager.h"
#include "ValidateHal.h"

NAME_SPACE_BEGIN

using namespace android::nn;

static const Timing kNoTiming = {.timeOnDevice = UINT64_MAX, .timeInDriver = UINT64_MAX};

auto now()
{
    return std::chrono::steady_clock::now();
};

PreparedModel::PreparedModel(const Model& model)
      : // Make a copy of the model, as we need to preserve it.
        mModel(model)
{
    NN_GPU_CALL();
    exec = ExecutorManager::createExecutor(mModel);
}

bool PreparedModel::initialize()
{
    NN_GPU_CALL();
    return exec->initPerModel();
}

void PreparedModel::asyncExecute_1_2(const Request& request,
                                     const sp<V1_2::IExecutionCallback>& callback)
{
    NN_GPU_CALL();
    exec->initPerExecThread();
    bool succ = exec->run(request);
    exec->deinitPerExecThread();
    if (succ)
    {
        callback->notify_1_2(ErrorStatus::NONE, {}, kNoTiming);
    }
    else
    {
        callback->notify_1_2(ErrorStatus::GENERAL_FAILURE, {}, kNoTiming);
    }
}

void PreparedModel::asyncExecute(const Request& request,
                                 const sp<V1_0::IExecutionCallback>& callback)
{
    NN_GPU_CALL();

    exec->initPerExecThread();
    bool succ = exec->run(request);
    exec->deinitPerExecThread();

    if (succ)
    {
        callback->notify(ErrorStatus::NONE);
    }
    else
    {
        callback->notify(ErrorStatus::GENERAL_FAILURE);
    }
}

Return<ErrorStatus> PreparedModel::execute(const Request& request,
                                           const sp<V1_0::IExecutionCallback>& callback)
{
    NN_GPU_CALL();

    if (callback.get() == nullptr)
    {
        LOGE("invalid callback passed to execute");
        return ErrorStatus::INVALID_ARGUMENT;
    }

    if (!validateRequest(request, mModel))
    {
        callback->notify(ErrorStatus::INVALID_ARGUMENT);
        return ErrorStatus::INVALID_ARGUMENT;
    }

    execThreads.push_back(std::thread([this, request, callback]{ asyncExecute(request, callback); }));

    return ErrorStatus::NONE;
}

Return<ErrorStatus> PreparedModel::execute_1_2(const Request& request,
                                               MeasureTiming measure,
                                               const sp<V1_2::IExecutionCallback>& callback)
{
    NN_GPU_CALL();

    time_point driverStart;
    if (measure == MeasureTiming::YES)
    {
        driverStart = now();
    }

    if (callback.get() == nullptr)
    {
        LOGE("invalid callback passed to execute");
        return ErrorStatus::INVALID_ARGUMENT;
    }

    if (!validateRequest(request, mModel))
    {
        callback->notify_1_2(ErrorStatus::INVALID_ARGUMENT, {}, kNoTiming);
        return ErrorStatus::INVALID_ARGUMENT;
    }

    execThreads.push_back(std::thread([this, request, callback]{ asyncExecute(request, callback); }));

    return ErrorStatus::NONE;
}

Return<void> PreparedModel::executeSynchronously(const Request& request,
                                                 MeasureTiming measure,
                                                 executeSynchronously_cb cb)
{
    NN_GPU_CALL();
    time_point driverStart;

    if (measure == MeasureTiming::YES)
    {
        driverStart = now();
    }

    if (!validateRequest(request, mModel))
    {
        cb(ErrorStatus::INVALID_ARGUMENT, {}, kNoTiming);
        return Void();
    }

    exec->initPerExecThread();
    bool succ = exec->run(request);
    exec->deinitPerExecThread();

    if (succ)
    {
        cb(ErrorStatus::NONE, {}, kNoTiming);
    }
    else
    {
        cb(ErrorStatus::GENERAL_FAILURE, {}, kNoTiming);
    }

    return Void();
}

Return<void> PreparedModel::configureExecutionBurst(const sp<V1_2::IBurstCallback>& callback,
                                                    const MQDescriptorSync<V1_2::FmqRequestDatum>& requestChannel,
                                                    const MQDescriptorSync<V1_2::FmqResultDatum>& resultChannel,
                                                    configureExecutionBurst_cb cb)
{
    NN_GPU_CALL();

    UNUSED(callback);
    UNUSED(requestChannel);
    UNUSED(resultChannel);
    UNUSED(cb);

    return Void();
}

PreparedModel::~PreparedModel()
{
    NN_GPU_CALL();
    for (auto& th : execThreads) th.join();
    exec->deinitPerModel();
}

NAME_SPACE_STOP