diff options
author | Jan Eilers <jan.eilers@arm.com> | 2021-10-26 14:56:47 +0100 |
---|---|---|
committer | TeresaARM <teresa.charlinreyes@arm.com> | 2021-10-29 10:20:16 +0000 |
commit | b1c62f11881e0d528bea5b3664a8f36e4c03b508 (patch) | |
tree | 970be1e2fde222603fa3aa05c87b22cb30035be2 /include/armnn/INetwork.hpp | |
parent | 05eccac1433a6637ef09bb30811c8c842116456d (diff) | |
download | armnn-b1c62f11881e0d528bea5b3664a8f36e4c03b508.tar.gz |
Fix armnn_external_delegate option parsing
* Fixed the mechanism in armnn_delegate.cpp which creates the runtime
options by moving it to DelegateOptions. BackendOptions are used in
the runtime options as well as in the optimizer options but need
to be processed separately.
* Added runtime options to DelegateOptions to make it easier to differ
between backend options of the runtime and the optimization
* Updated armnn_external_delegate to use runtime options
* Added explanations to OptimizerOptions
Signed-off-by: Jan Eilers <jan.eilers@arm.com>
Change-Id: I34235474147def261821e4576dd41a83ad492258
Diffstat (limited to 'include/armnn/INetwork.hpp')
-rw-r--r-- | include/armnn/INetwork.hpp | 14 |
1 files changed, 12 insertions, 2 deletions
diff --git a/include/armnn/INetwork.hpp b/include/armnn/INetwork.hpp index f85b29ee81..a79afead95 100644 --- a/include/armnn/INetwork.hpp +++ b/include/armnn/INetwork.hpp @@ -115,6 +115,8 @@ protected: }; +/// ArmNN performs an optimization on each model/network before it gets loaded for execution. OptimizerOptions provides +/// a set of features that allows the user to customize this optimization on a per model basis. struct OptimizerOptions { OptimizerOptions() @@ -160,13 +162,21 @@ struct OptimizerOptions } } - // Reduce Fp32 data to Fp16 for faster processing + /// Reduces all Fp32 operators in the model to Fp16 for faster processing. + /// @Note This feature works best if all operators of the model are in Fp32. ArmNN will add conversion layers + /// between layers that weren't in Fp32 in the first place or if the operator is not supported in Fp16. + /// The overhead of these conversions can lead to a slower overall performance if too many conversions are + /// required. bool m_ReduceFp32ToFp16; // Add debug data for easier troubleshooting bool m_Debug; - // Reduce Fp32 data to Bf16 for faster processing + /// Reduces all Fp32 operators in the model to Bf16 for faster processing. + /// @Note This feature works best if all operators of the model are in Fp32. ArmNN will add conversion layers + /// between layers that weren't in Fp32 in the first place or if the operator is not supported in Bf16. + /// The overhead of these conversions can lead to a slower overall performance if too many conversions are + /// required. bool m_ReduceFp32ToBf16; // Infer output size when not available |