diff options
author | Xusong Wang <xusongw@google.com> | 2021-03-03 16:20:37 -0800 |
---|---|---|
committer | Xusong Wang <xusongw@google.com> | 2021-05-06 13:56:32 -0700 |
commit | 727a7b2104b0962509fedffe720eec508b2ee6de (patch) | |
tree | db56f9f27626747458a78fe1a1c5454fc666794a /neuralnetworks/utils/common/src/ResilientPreparedModel.cpp | |
parent | 57c9114315650206adab4beaa97e07d27e3e4b10 (diff) |
Introduce reusable execution to canonical interface -- HAL.
This CL modifies the canonical interface for reusable executions:
- Add new interface: IExecution with compute and computeFenced methods
- Add new method IPreparedModel::createExecution
In NNAPI runtime, the new interface IExecution is used to
memoize request-specific execution resources (e.g. converted HAL
request). The expected usage is that, IPreparedModel::createExecution
will be invoked in the first computation of a reusable NDK ANNExecution
object, and IExecution::compute* will be invoked repeatedly.
The IPreparedModel::execute* methods are preserved to avoid redundant
object creation and memoization overhead for a single-time
(non-reusable) execution.
For a vendor implementing the canonical interfaces, only the
IPreparedModel::execute* methods will be called because there is
currently no reusable execution at HAL interface. A DefaultExecution
implementation is provided to reduce the work needed on the vendor side.
Bug: 184073769
Test: NNT_static
Test: neuralnetworks_utils_hal_1_0_test
Test: neuralnetworks_utils_hal_1_1_test
Test: neuralnetworks_utils_hal_1_2_test
Test: neuralnetworks_utils_hal_1_3_test
Test: neuralnetworks_utils_hal_common_test
Test: neuralnetworks_utils_hal_aidl_test
Change-Id: I91790bb5ccf5ae648687fe603f88ffda2c9fd2b2
Diffstat (limited to 'neuralnetworks/utils/common/src/ResilientPreparedModel.cpp')
-rw-r--r-- | neuralnetworks/utils/common/src/ResilientPreparedModel.cpp | 30 |
1 files changed, 30 insertions, 0 deletions
diff --git a/neuralnetworks/utils/common/src/ResilientPreparedModel.cpp b/neuralnetworks/utils/common/src/ResilientPreparedModel.cpp index 5dd5f99f5f..1ae19bc6ca 100644 --- a/neuralnetworks/utils/common/src/ResilientPreparedModel.cpp +++ b/neuralnetworks/utils/common/src/ResilientPreparedModel.cpp @@ -17,7 +17,9 @@ #include "ResilientPreparedModel.h" #include "InvalidBurst.h" +#include "InvalidExecution.h" #include "ResilientBurst.h" +#include "ResilientExecution.h" #include <android-base/logging.h> #include <android-base/thread_annotations.h> @@ -127,6 +129,21 @@ ResilientPreparedModel::executeFenced(const nn::Request& request, return protect(*this, fn); } +nn::GeneralResult<nn::SharedExecution> ResilientPreparedModel::createReusableExecution( + const nn::Request& request, nn::MeasureTiming measure, + const nn::OptionalDuration& loopTimeoutDuration) const { +#if 0 + auto self = shared_from_this(); + ResilientExecution::Factory makeExecution = + [preparedModel = std::move(self), request, measure, loopTimeoutDuration] { + return preparedModel->createReusableExecutionInternal(request, measure, loopTimeoutDuration); + }; + return ResilientExecution::create(std::move(makeExecution)); +#else + return createReusableExecutionInternal(request, measure, loopTimeoutDuration); +#endif +} + nn::GeneralResult<nn::SharedBurst> ResilientPreparedModel::configureExecutionBurst() const { #if 0 auto self = shared_from_this(); @@ -140,6 +157,19 @@ nn::GeneralResult<nn::SharedBurst> ResilientPreparedModel::configureExecutionBur #endif } +nn::GeneralResult<nn::SharedExecution> ResilientPreparedModel::createReusableExecutionInternal( + const nn::Request& request, nn::MeasureTiming measure, + const nn::OptionalDuration& loopTimeoutDuration) const { + if (!isValidInternal()) { + return std::make_shared<const InvalidExecution>(); + } + const auto fn = [&request, measure, + &loopTimeoutDuration](const nn::IPreparedModel& preparedModel) { + return preparedModel.createReusableExecution(request, measure, loopTimeoutDuration); + }; + return protect(*this, fn); +} + std::any ResilientPreparedModel::getUnderlyingResource() const { return getPreparedModel()->getUnderlyingResource(); } |