HAL interface for compilation and execution hints
The following AIDL types are added:
- TokenValuePair
- PrepareModelConfig
- ExecutionConfig
The following AIDL methods are added:
- IDevice::prepareModelWithConfig
- IPreparedModel::executeSynchronouslyWithConfig
- IPreparedModel::executeFencedWithConfig
- IBurst::executeSynchronouslyWithConfig
The compilation and execution hints are being stored as a list of
token-value pairs as part of the PrepareModelConfig / ExecutionConfig.
And the PrepareModelConfig / ExecutionConfig parcelables are created in
order to make future extensions to the execution related interfaces
easier.
It is the drivers responsibility to verify the hints, and it is allowed
for the driver to ignore them.
Bug: 203248587
Test: neuralnetworks_utils_hal_aidl_test
Change-Id: I98240fd75089fc85cdfcaa0be28aab8a6f0dfca5
Merged-In: I98240fd75089fc85cdfcaa0be28aab8a6f0dfca5
(cherry picked from commit 0e671f3edb9d2c78658a4ef4169e3211e3f9bb00)
diff --git a/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Burst.h b/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Burst.h
index 8bd2fbe..cef76c6 100644
--- a/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Burst.h
+++ b/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Burst.h
@@ -45,12 +45,15 @@
nn::ExecutionResult<std::pair<std::vector<nn::OutputShape>, nn::Timing>> execute(
const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalTimePoint& deadline,
- const nn::OptionalDuration& loopTimeoutDuration) const override;
+ const nn::OptionalTimePoint& deadline, const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const override;
nn::GeneralResult<nn::SharedExecution> createReusableExecution(
const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalDuration& loopTimeoutDuration) const override;
+ const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const override;
private:
const nn::SharedPreparedModel kPreparedModel;
diff --git a/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Device.h b/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Device.h
index 0a6ca3e..d7c43ef 100644
--- a/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Device.h
+++ b/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Device.h
@@ -65,8 +65,9 @@
nn::GeneralResult<nn::SharedPreparedModel> prepareModel(
const nn::Model& model, nn::ExecutionPreference preference, nn::Priority priority,
nn::OptionalTimePoint deadline, const std::vector<nn::SharedHandle>& modelCache,
- const std::vector<nn::SharedHandle>& dataCache,
- const nn::CacheToken& token) const override;
+ const std::vector<nn::SharedHandle>& dataCache, const nn::CacheToken& token,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const override;
nn::GeneralResult<nn::SharedPreparedModel> prepareModelFromCache(
nn::OptionalTimePoint deadline, const std::vector<nn::SharedHandle>& modelCache,
diff --git a/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/PreparedModel.h b/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/PreparedModel.h
index bdb5b54..337c132 100644
--- a/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/PreparedModel.h
+++ b/neuralnetworks/1.0/utils/include/nnapi/hal/1.0/PreparedModel.h
@@ -49,18 +49,23 @@
nn::ExecutionResult<std::pair<std::vector<nn::OutputShape>, nn::Timing>> execute(
const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalTimePoint& deadline,
- const nn::OptionalDuration& loopTimeoutDuration) const override;
+ const nn::OptionalTimePoint& deadline, const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const override;
nn::GeneralResult<std::pair<nn::SyncFence, nn::ExecuteFencedInfoCallback>> 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 override;
+ const nn::OptionalDuration& timeoutDurationAfterFence,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const override;
nn::GeneralResult<nn::SharedExecution> createReusableExecution(
const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalDuration& loopTimeoutDuration) const override;
+ const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const override;
nn::GeneralResult<nn::SharedBurst> configureExecutionBurst() const override;
diff --git a/neuralnetworks/1.0/utils/src/Burst.cpp b/neuralnetworks/1.0/utils/src/Burst.cpp
index 1284721..3642bc6 100644
--- a/neuralnetworks/1.0/utils/src/Burst.cpp
+++ b/neuralnetworks/1.0/utils/src/Burst.cpp
@@ -50,15 +50,20 @@
nn::ExecutionResult<std::pair<std::vector<nn::OutputShape>, nn::Timing>> Burst::execute(
const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalTimePoint& deadline,
- const nn::OptionalDuration& loopTimeoutDuration) const {
- return kPreparedModel->execute(request, measure, deadline, loopTimeoutDuration);
+ const nn::OptionalTimePoint& deadline, const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const {
+ return kPreparedModel->execute(request, measure, deadline, loopTimeoutDuration, hints,
+ extensionNameToPrefix);
}
nn::GeneralResult<nn::SharedExecution> Burst::createReusableExecution(
const nn::Request& request, nn::MeasureTiming measure,
- const nn::OptionalDuration& loopTimeoutDuration) const {
- return kPreparedModel->createReusableExecution(request, measure, loopTimeoutDuration);
+ const nn::OptionalDuration& loopTimeoutDuration,
+ const std::vector<nn::TokenValuePair>& hints,
+ const std::vector<nn::ExtensionNameAndPrefix>& extensionNameToPrefix) const {
+ return kPreparedModel->createReusableExecution(request, measure, loopTimeoutDuration, hints,
+ extensionNameToPrefix);
}
} // namespace android::hardware::neuralnetworks::V1_0::utils
diff --git a/neuralnetworks/1.0/utils/src/Device.cpp b/neuralnetworks/1.0/utils/src/Device.cpp
index b0c236e..620d040 100644
--- a/neuralnetworks/1.0/utils/src/Device.cpp
+++ b/neuralnetworks/1.0/utils/src/Device.cpp
@@ -143,7 +143,9 @@
nn::GeneralResult<nn::SharedPreparedModel> Device::prepareModel(
const nn::Model& model, nn::ExecutionPreference /*preference*/, nn::Priority /*priority*/,
nn::OptionalTimePoint /*deadline*/, const std::vector<nn::SharedHandle>& /*modelCache*/,
- const std::vector<nn::SharedHandle>& /*dataCache*/, const nn::CacheToken& /*token*/) const {
+ const std::vector<nn::SharedHandle>& /*dataCache*/, const nn::CacheToken& /*token*/,
+ const std::vector<nn::TokenValuePair>& /*hints*/,
+ const std::vector<nn::ExtensionNameAndPrefix>& /*extensionNameToPrefix*/) const {
// Ensure that model is ready for IPC.
std::optional<nn::Model> maybeModelInShared;
const nn::Model& modelInShared =
diff --git a/neuralnetworks/1.0/utils/src/PreparedModel.cpp b/neuralnetworks/1.0/utils/src/PreparedModel.cpp
index 00e7d22..b8055fc 100644
--- a/neuralnetworks/1.0/utils/src/PreparedModel.cpp
+++ b/neuralnetworks/1.0/utils/src/PreparedModel.cpp
@@ -59,7 +59,9 @@
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 {
+ 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;
@@ -94,19 +96,22 @@
}
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 {
+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 1.0 HAL service";
}
nn::GeneralResult<nn::SharedExecution> PreparedModel::createReusableExecution(
const nn::Request& request, nn::MeasureTiming /*measure*/,
- const nn::OptionalDuration& /*loopTimeoutDuration*/) const {
+ 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;
diff --git a/neuralnetworks/1.0/utils/test/DeviceTest.cpp b/neuralnetworks/1.0/utils/test/DeviceTest.cpp
index 83e555f..9e9db16 100644
--- a/neuralnetworks/1.0/utils/test/DeviceTest.cpp
+++ b/neuralnetworks/1.0/utils/test/DeviceTest.cpp
@@ -380,7 +380,7 @@
// run test
const auto result = device->prepareModel(kSimpleModel, nn::ExecutionPreference::DEFAULT,
- nn::Priority::DEFAULT, {}, {}, {}, {});
+ nn::Priority::DEFAULT, {}, {}, {}, {}, {}, {});
// verify result
ASSERT_TRUE(result.has_value())
@@ -399,7 +399,7 @@
// run test
const auto result = device->prepareModel(kSimpleModel, nn::ExecutionPreference::DEFAULT,
- nn::Priority::DEFAULT, {}, {}, {}, {});
+ nn::Priority::DEFAULT, {}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -417,7 +417,7 @@
// run test
const auto result = device->prepareModel(kSimpleModel, nn::ExecutionPreference::DEFAULT,
- nn::Priority::DEFAULT, {}, {}, {}, {});
+ nn::Priority::DEFAULT, {}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -435,7 +435,7 @@
// run test
const auto result = device->prepareModel(kSimpleModel, nn::ExecutionPreference::DEFAULT,
- nn::Priority::DEFAULT, {}, {}, {}, {});
+ nn::Priority::DEFAULT, {}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -452,7 +452,7 @@
// run test
const auto result = device->prepareModel(kSimpleModel, nn::ExecutionPreference::DEFAULT,
- nn::Priority::DEFAULT, {}, {}, {}, {});
+ nn::Priority::DEFAULT, {}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -469,7 +469,7 @@
// run test
const auto result = device->prepareModel(kSimpleModel, nn::ExecutionPreference::DEFAULT,
- nn::Priority::DEFAULT, {}, {}, {}, {});
+ nn::Priority::DEFAULT, {}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -488,7 +488,7 @@
// run test
const auto result = device->prepareModel(kSimpleModel, nn::ExecutionPreference::DEFAULT,
- nn::Priority::DEFAULT, {}, {}, {}, {});
+ nn::Priority::DEFAULT, {}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
diff --git a/neuralnetworks/1.0/utils/test/PreparedModelTest.cpp b/neuralnetworks/1.0/utils/test/PreparedModelTest.cpp
index 7820c06..e03a98d 100644
--- a/neuralnetworks/1.0/utils/test/PreparedModelTest.cpp
+++ b/neuralnetworks/1.0/utils/test/PreparedModelTest.cpp
@@ -121,7 +121,7 @@
.WillOnce(Invoke(makeExecute(V1_0::ErrorStatus::NONE, V1_0::ErrorStatus::NONE)));
// run test
- const auto result = preparedModel->execute({}, {}, {}, {});
+ const auto result = preparedModel->execute({}, {}, {}, {}, {}, {});
// verify result
EXPECT_TRUE(result.has_value())
@@ -138,7 +138,7 @@
V1_0::ErrorStatus::GENERAL_FAILURE)));
// run test
- const auto result = preparedModel->execute({}, {}, {}, {});
+ const auto result = preparedModel->execute({}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -155,7 +155,7 @@
makeExecute(V1_0::ErrorStatus::NONE, V1_0::ErrorStatus::GENERAL_FAILURE)));
// run test
- const auto result = preparedModel->execute({}, {}, {}, {});
+ const auto result = preparedModel->execute({}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -171,7 +171,7 @@
.WillOnce(InvokeWithoutArgs(makeGeneralTransportFailure));
// run test
- const auto result = preparedModel->execute({}, {}, {}, {});
+ const auto result = preparedModel->execute({}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -187,7 +187,7 @@
.WillOnce(InvokeWithoutArgs(makeDeadObjectFailure));
// run test
- const auto result = preparedModel->execute({}, {}, {}, {});
+ const auto result = preparedModel->execute({}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -205,7 +205,7 @@
EXPECT_CALL(*mockPreparedModel, execute(_, _)).Times(1).WillOnce(InvokeWithoutArgs(ret));
// run test
- const auto result = preparedModel->execute({}, {}, {}, {});
+ const auto result = preparedModel->execute({}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -218,7 +218,7 @@
const auto preparedModel = PreparedModel::create(mockPreparedModel).value();
// run test
- const auto result = preparedModel->executeFenced({}, {}, {}, {}, {}, {});
+ const auto result = preparedModel->executeFenced({}, {}, {}, {}, {}, {}, {}, {});
// verify result
ASSERT_FALSE(result.has_value());
@@ -235,7 +235,7 @@
.WillRepeatedly(Invoke(makeExecute(V1_0::ErrorStatus::NONE, V1_0::ErrorStatus::NONE)));
// create execution
- const auto createResult = preparedModel->createReusableExecution({}, {}, {});
+ const auto createResult = preparedModel->createReusableExecution({}, {}, {}, {}, {});
ASSERT_TRUE(createResult.has_value())
<< "Failed with " << createResult.error().code << ": " << createResult.error().message;
ASSERT_NE(createResult.value(), nullptr);
@@ -258,7 +258,7 @@
V1_0::ErrorStatus::GENERAL_FAILURE)));
// create execution
- const auto createResult = preparedModel->createReusableExecution({}, {}, {});
+ const auto createResult = preparedModel->createReusableExecution({}, {}, {}, {}, {});
ASSERT_TRUE(createResult.has_value())
<< "Failed with " << createResult.error().code << ": " << createResult.error().message;
ASSERT_NE(createResult.value(), nullptr);
@@ -279,7 +279,7 @@
makeExecute(V1_0::ErrorStatus::NONE, V1_0::ErrorStatus::GENERAL_FAILURE)));
// create execution
- const auto createResult = preparedModel->createReusableExecution({}, {}, {});
+ const auto createResult = preparedModel->createReusableExecution({}, {}, {}, {}, {});
ASSERT_TRUE(createResult.has_value())
<< "Failed with " << createResult.error().code << ": " << createResult.error().message;
ASSERT_NE(createResult.value(), nullptr);
@@ -299,7 +299,7 @@
.WillOnce(InvokeWithoutArgs(makeGeneralTransportFailure));
// create execution
- const auto createResult = preparedModel->createReusableExecution({}, {}, {});
+ const auto createResult = preparedModel->createReusableExecution({}, {}, {}, {}, {});
ASSERT_TRUE(createResult.has_value())
<< "Failed with " << createResult.error().code << ": " << createResult.error().message;
ASSERT_NE(createResult.value(), nullptr);
@@ -319,7 +319,7 @@
.WillOnce(InvokeWithoutArgs(makeDeadObjectFailure));
// create execution
- const auto createResult = preparedModel->createReusableExecution({}, {}, {});
+ const auto createResult = preparedModel->createReusableExecution({}, {}, {}, {}, {});
ASSERT_TRUE(createResult.has_value())
<< "Failed with " << createResult.error().code << ": " << createResult.error().message;
ASSERT_NE(createResult.value(), nullptr);
@@ -341,7 +341,7 @@
EXPECT_CALL(*mockPreparedModel, execute(_, _)).Times(1).WillOnce(InvokeWithoutArgs(ret));
// create execution
- const auto createResult = preparedModel->createReusableExecution({}, {}, {});
+ const auto createResult = preparedModel->createReusableExecution({}, {}, {}, {}, {});
ASSERT_TRUE(createResult.has_value())
<< "Failed with " << createResult.error().code << ": " << createResult.error().message;
ASSERT_NE(createResult.value(), nullptr);
@@ -358,7 +358,7 @@
const auto preparedModel = PreparedModel::create(mockPreparedModel).value();
// create execution
- const auto createResult = preparedModel->createReusableExecution({}, {}, {});
+ const auto createResult = preparedModel->createReusableExecution({}, {}, {}, {}, {});
ASSERT_TRUE(createResult.has_value())
<< "Failed with " << createResult.error().code << ": " << createResult.error().message;
ASSERT_NE(createResult.value(), nullptr);