Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit b5c8a82f authored by Miao Wang's avatar Miao Wang
Browse files

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 0e671f3e)
parent 63a4e9b2

neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Burst.h

+6 −3
Original line number Diff line number Diff line
@@ -45,12 +45,15 @@ class Burst final : public nn::IBurst { 


    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;

neuralnetworks/1.0/utils/include/nnapi/hal/1.0/Device.h

+3 −2
Original line number Diff line number Diff line
@@ -65,8 +65,9 @@ class Device final : public nn::IDevice { 
    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,

neuralnetworks/1.0/utils/include/nnapi/hal/1.0/PreparedModel.h

+9 −4
Original line number Diff line number Diff line
@@ -49,18 +49,23 @@ class PreparedModel final : public nn::IPreparedModel, 


    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;

neuralnetworks/1.0/utils/src/Burst.cpp

+10 −5
Original line number Diff line number Diff line
@@ -50,15 +50,20 @@ Burst::OptionalCacheHold Burst::cacheMemory(const nn::SharedMemory& /*memory*/) 


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

neuralnetworks/1.0/utils/src/Device.cpp

+3 −1
Original line number Diff line number Diff line
@@ -143,7 +143,9 @@ nn::GeneralResult<std::vector<bool>> Device::getSupportedOperations(const nn::Mo 
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 =