diff options
| author | Matthew Bentham <matthew.bentham@arm.com> | 2019-04-01 17:17:58 +0100 |
|---|---|---|
| committer | Matthew Bentham <matthew.bentham@arm.com> | 2019-04-02 10:22:16 +0100 |
| commit | 16196e267833178dae62926ff090c50ec6813ad4 (patch) | |
| tree | 03c0f7cd2490a1e68a63f71ef08c6a6c591f363c /ArmnnDriverImpl.cpp | |
| parent | a6542c53e929172a36b4d3787686c13a7c569386 (diff) | |
| download | android-nn-driver-16196e267833178dae62926ff090c50ec6813ad4.tar.gz | |
MLCE-110 Propagate error from armnn EnqueueWorkload
Change-Id: Ic53b1cdbdd3a7d656932651c74911940affc09b6
Signed-off-by: Matthew Bentham <matthew.bentham@arm.com>
Diffstat (limited to 'ArmnnDriverImpl.cpp')
| -rw-r--r-- | ArmnnDriverImpl.cpp | 5 |
1 files changed, 4 insertions, 1 deletions
diff --git a/ArmnnDriverImpl.cpp b/ArmnnDriverImpl.cpp index a3c2e10f..f6456ee1 100644 --- a/ArmnnDriverImpl.cpp +++ b/ArmnnDriverImpl.cpp @@ -208,7 +208,10 @@ Return<ErrorStatus> ArmnnDriverImpl<HalPolicy>::prepareModel( // Run a single 'dummy' inference of the model. This means that CL kernels will get compiled (and tuned if // this is enabled) before the first 'real' inference which removes the overhead of the first inference. - preparedModel->ExecuteWithDummyInputs(); + if (!preparedModel->ExecuteWithDummyInputs()) + { + return FailPrepareModel(ErrorStatus::GENERAL_FAILURE, "Network could not be executed", cb); + } if (clTunedParameters && options.GetClTunedParametersMode() == armnn::IGpuAccTunedParameters::Mode::UpdateTunedParameters) |