diff options
author | Mike Kelly <mike.kelly@arm.com> | 2023-07-07 15:43:06 +0100 |
---|---|---|
committer | Mike Kelly <mike.kelly@arm.com> | 2023-07-14 00:00:53 +0100 |
commit | 4cc341cf8b5a6e6bb0543504cbbfde6fa11a2cdb (patch) | |
tree | 7cac128e9ec6f2fd27f1afdb55f44b870f39e0b3 /src/armnn/layers/SplitterLayer.cpp | |
parent | 6963b33221c23af4a8eff19ff4a5773230b0befd (diff) | |
download | armnn-4cc341cf8b5a6e6bb0543504cbbfde6fa11a2cdb.tar.gz |
IVGCVSW-7830 Add backend optimizations to remove Reshapes where possible
* Added optimization to remove reshapes for Neon and Ref Backends
by using overridden TensorInfos
* Added ability to delete Subgraphs during Optimization
* Fixed naming error in NeonEndToEndTests and CLEndToEndTests
* Added LayerNameAndTypeCheck for testing.
* Fixed error where layers were not marked as altered when removed in
CLBackend
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: I1ac25cd4ec9821470d961831ae2c8d24882276cc
Diffstat (limited to 'src/armnn/layers/SplitterLayer.cpp')
-rw-r--r-- | src/armnn/layers/SplitterLayer.cpp | 5 |
1 files changed, 3 insertions, 2 deletions
diff --git a/src/armnn/layers/SplitterLayer.cpp b/src/armnn/layers/SplitterLayer.cpp index 86a42305ff..dc8864a736 100644 --- a/src/armnn/layers/SplitterLayer.cpp +++ b/src/armnn/layers/SplitterLayer.cpp @@ -131,13 +131,14 @@ void SplitterLayer::CreateTensors(const TensorHandleFactoryRegistry& registry, // 2) the same TensorHandleFactory is used for input and split layer output // 3) the output does not go to a Constant layer or input layer // 4) if split along x or y (2 innermost dimensions) and the next layers do not require padding - // 5) none of the outputs have an Overridden TensorInfo + // 5) neither the input nor the outputs have an Overridden TensorInfo if (parentInfo.IsTypeSpaceMatch(info) && //(1) factoryId == slot->GetTensorHandleFactoryId() && //(2) GetOutputSlot(i).GetConnection(0)->GetOwningLayer().GetType() != LayerType::Constant && //(3) GetOutputSlot(i).GetConnection(0)->GetOwningLayer().GetType() != LayerType::Input && //(3) canUseSubTensorOnXorY && //(4) - !GetOutputSlot(i).GetConnection(0)->IsTensorInfoOverridden()) //(5) + !GetOutputSlot(i).GetConnection(0)->IsTensorInfoOverridden() && //(5) + !GetInputSlot(0).IsTensorInfoOverridden()) //(5) { ARMNN_NO_DEPRECATE_WARN_BEGIN return factory.CreateSubTensorHandle(*inputData, |