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
|
/*
* Copyright (C) 2020 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.
*/
#include "PreparedModel.h"
#include "Burst.h"
#include "BurstUtils.h"
#include "Callbacks.h"
#include "Conversions.h"
#include "Execution.h"
#include "Utils.h"
#include <android/hardware/neuralnetworks/1.0/types.h>
#include <android/hardware/neuralnetworks/1.1/types.h>
#include <android/hardware/neuralnetworks/1.2/IPreparedModel.h>
#include <android/hardware/neuralnetworks/1.2/types.h>
#include <nnapi/IPreparedModel.h>
#include <nnapi/Result.h>
#include <nnapi/Types.h>
#include <nnapi/hal/1.0/Conversions.h>
#include <nnapi/hal/1.0/HandleError.h>
#include <nnapi/hal/1.0/ProtectCallback.h>
#include <nnapi/hal/CommonUtils.h>
#include <chrono>
#include <memory>
#include <tuple>
#include <utility>
#include <vector>
// See hardware/interfaces/neuralnetworks/utils/README.md for more information on HIDL interface
// lifetimes across processes and for protecting asynchronous calls across HIDL.
namespace android::hardware::neuralnetworks::V1_2::utils {
nn::GeneralResult<std::shared_ptr<const PreparedModel>> PreparedModel::create(
sp<V1_2::IPreparedModel> preparedModel, bool executeSynchronously) {
if (preparedModel == nullptr) {
return NN_ERROR() << "V1_2::utils::PreparedModel::create must have non-null preparedModel";
}
auto deathHandler = NN_TRY(hal::utils::DeathHandler::create(preparedModel));
return std::make_shared<const PreparedModel>(PrivateConstructorTag{}, executeSynchronously,
std::move(preparedModel), std::move(deathHandler));
}
PreparedModel::PreparedModel(PrivateConstructorTag /*tag*/, bool executeSynchronously,
sp<V1_2::IPreparedModel> preparedModel,
hal::utils::DeathHandler deathHandler)
: kExecuteSynchronously(executeSynchronously),
kPreparedModel(std::move(preparedModel)),
kDeathHandler(std::move(deathHandler)) {}
nn::ExecutionResult<std::pair<std::vector<nn::OutputShape>, nn::Timing>>
PreparedModel::executeSynchronously(const V1_0::Request& request, MeasureTiming measure) const {
auto cb = hal::utils::CallbackValue(executionCallback);
const auto ret = kPreparedModel->executeSynchronously(request, measure, cb);
HANDLE_TRANSPORT_FAILURE(ret);
return cb.take();
}
nn::ExecutionResult<std::pair<std::vector<nn::OutputShape>, nn::Timing>>
PreparedModel::executeAsynchronously(const V1_0::Request& request, MeasureTiming measure) const {
const auto cb = sp<ExecutionCallback>::make();
const auto scoped = kDeathHandler.protectCallback(cb.get());
const auto ret = kPreparedModel->execute_1_2(request, measure, cb);
const auto status = HANDLE_TRANSPORT_FAILURE(ret);
if (status != V1_0::ErrorStatus::OUTPUT_INSUFFICIENT_SIZE) {
HANDLE_STATUS_HIDL(status) << "execution failed with " << toString(status);
}
return cb->get();
}
nn::ExecutionResult<std::pair<std::vector<nn::OutputShape>, nn::Timing>> PreparedModel::execute(
const nn::Request& request, nn::MeasureTiming measure,
const nn::OptionalTimePoint& /*deadline*/,
const nn::OptionalDuration& /*loopTimeoutDuration*/,
const std::vector<nn::TokenValuePair>& /*hints*/,
const std::vector<nn::ExtensionNameAndPrefix>& /*extensionNameToPrefix*/) const {
// Ensure that request is ready for IPC.
std::optional<nn::Request> maybeRequestInShared;
hal::utils::RequestRelocation relocation;
const nn::Request& requestInShared = NN_TRY(hal::utils::convertRequestFromPointerToShared(
&request, nn::kDefaultRequestMemoryAlignment, nn::kMinMemoryPadding,
&maybeRequestInShared, &relocation));
const auto hidlRequest = NN_TRY(convert(requestInShared));
const auto hidlMeasure = NN_TRY(convert(measure));
return executeInternal(hidlRequest, hidlMeasure, relocation);
}
nn::ExecutionResult<std::pair<std::vector<nn::OutputShape>, nn::Timing>>
PreparedModel::executeInternal(const V1_0::Request& request, MeasureTiming measure,
const hal::utils::RequestRelocation& relocation) const {
if (relocation.input) {
relocation.input->flush();
}
auto result = kExecuteSynchronously ? executeSynchronously(request, measure)
: executeAsynchronously(request, measure);
auto [outputShapes, timing] = NN_TRY(std::move(result));
if (relocation.output) {
relocation.output->flush();
}
return std::make_pair(std::move(outputShapes), timing);
}
nn::GeneralResult<std::pair<nn::SyncFence, nn::ExecuteFencedInfoCallback>>
PreparedModel::executeFenced(
const nn::Request& /*request*/, const std::vector<nn::SyncFence>& /*waitFor*/,
nn::MeasureTiming /*measure*/, const nn::OptionalTimePoint& /*deadline*/,
const nn::OptionalDuration& /*loopTimeoutDuration*/,
const nn::OptionalDuration& /*timeoutDurationAfterFence*/,
const std::vector<nn::TokenValuePair>& /*hints*/,
const std::vector<nn::ExtensionNameAndPrefix>& /*extensionNameToPrefix*/) const {
return NN_ERROR(nn::ErrorStatus::GENERAL_FAILURE)
<< "IPreparedModel::executeFenced is not supported on 1.2 HAL service";
}
nn::GeneralResult<nn::SharedExecution> PreparedModel::createReusableExecution(
const nn::Request& request, nn::MeasureTiming measure,
const nn::OptionalDuration& /*loopTimeoutDuration*/,
const std::vector<nn::TokenValuePair>& /*hints*/,
const std::vector<nn::ExtensionNameAndPrefix>& /*extensionNameToPrefix*/) const {
// Ensure that request is ready for IPC.
std::optional<nn::Request> maybeRequestInShared;
hal::utils::RequestRelocation relocation;
const nn::Request& requestInShared = NN_TRY(hal::utils::convertRequestFromPointerToShared(
&request, nn::kDefaultRequestMemoryAlignment, nn::kMinMemoryPadding,
&maybeRequestInShared, &relocation));
auto hidlRequest = NN_TRY(convert(requestInShared));
auto hidlMeasure = NN_TRY(convert(measure));
return Execution::create(shared_from_this(), std::move(hidlRequest), std::move(relocation),
hidlMeasure);
}
nn::GeneralResult<nn::SharedBurst> PreparedModel::configureExecutionBurst() const {
const auto pollingTimeWindow = getBurstControllerPollingTimeWindow();
return Burst::create(shared_from_this(), kPreparedModel, pollingTimeWindow);
}
std::any PreparedModel::getUnderlyingResource() const {
sp<V1_2::IPreparedModel> resource = kPreparedModel;
return resource;
}
} // namespace android::hardware::neuralnetworks::V1_2::utils
|
