IVGCVSW-5314 Create OptimizeForExclusiveConnection
* FuseBatchNorm class has been added to facilitate testing
* Only Convolution2D FP32 being fused
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I049c4770946ddca21b08516d4c9f4d0d22bf9b45
diff --git a/src/armnn/Network.cpp b/src/armnn/Network.cpp
index 373f999..6578b84 100644
--- a/src/armnn/Network.cpp
+++ b/src/armnn/Network.cpp
@@ -1054,7 +1054,8 @@
OptimizeConsecutiveReshapes(),
FoldPadIntoConvolution2d(),
PermuteAndBatchToSpaceAsDepthToSpace(),
- TransposeAndBatchToSpaceAsDepthToSpace()));
+ TransposeAndBatchToSpaceAsDepthToSpace(),
+ FuseBatchNormIntoConvolution2D()));
// If Fp32 to Fp16 optimization is set convert Fp32 network to Fp16
if (options.m_ReduceFp32ToFp16)
diff --git a/src/armnn/optimizations/All.hpp b/src/armnn/optimizations/All.hpp
index e89c36b..d042616 100644
--- a/src/armnn/optimizations/All.hpp
+++ b/src/armnn/optimizations/All.hpp
@@ -10,6 +10,7 @@
#include "ConvertFp32NetworkToBf16.hpp"
#include "ConvertFp32NetworkToFp16.hpp"
#include "FoldPadIntoConvolution2d.hpp"
+#include "FuseBatchNorm.hpp"
#include "MovePermuteUp.hpp"
#include "MoveTransposeUp.hpp"
#include "OptimizeConsecutiveReshapes.hpp"
diff --git a/src/armnn/optimizations/FuseBatchNorm.hpp b/src/armnn/optimizations/FuseBatchNorm.hpp
new file mode 100644
index 0000000..e8e8c5d
--- /dev/null
+++ b/src/armnn/optimizations/FuseBatchNorm.hpp
@@ -0,0 +1,152 @@
+//
+// Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
+// SPDX-License-Identifier: MIT
+//
+
+#pragma once
+
+#include "Optimization.hpp"
+#include <armnnUtils/DataLayoutIndexed.hpp>
+
+namespace armnn
+{
+namespace optimizations
+{
+
+template <typename ConvLayer>
+class FuseBatchNorm
+{
+public:
+ /// Run for every exclusive connection between any base Convolution layer and a child BatchNorm layer for not
+ /// quantized layers.
+ /// The child will be removed, the base will be removed if it's left unconnected. A new Convolution layer will
+ /// be added, its weights and bias will be calculated using the weights and bias of the base Convolution layer
+ /// combined with the parameters of the child BatchNorm layer.
+ void Run(Graph& graph, InputSlot& connection) const
+ {
+ Layer& base = connection.GetConnectedOutputSlot()->GetOwningLayer();
+ Layer& child = connection.GetOwningLayer();
+
+ ARMNN_ASSERT(base.GetType() == LayerType::Convolution2d);
+ ARMNN_ASSERT(child.GetType() == LayerType::BatchNormalization);
+
+ if (base.GetDataType() == DataType::Float32 && child.GetDataType() == DataType::Float32)
+ {
+ OutputSlot* parentOut = base.GetInputSlot(0).GetConnectedOutputSlot();
+ auto convLayer = PolymorphicDowncast<ConvLayer*>(&base);
+ auto batchNormLayer = PolymorphicDowncast<BatchNormalizationLayer*>(&child);
+
+ // Read convolution and batch norm parameters
+ BatchNormalizationDescriptor batchNormDescriptor = batchNormLayer->GetParameters();
+ auto epsilon = batchNormDescriptor.m_Eps;
+ IgnoreUnused(epsilon);
+
+ ConstTensor betaTensor(batchNormLayer->m_Beta->GetTensorInfo(), batchNormLayer->m_Beta->Map(true));
+ ConstTensor gammaTensor(batchNormLayer->m_Gamma->GetTensorInfo(), batchNormLayer->m_Gamma->Map(true));
+ ConstTensor meanTensor(batchNormLayer->m_Mean->GetTensorInfo(), batchNormLayer->m_Mean->Map(true));
+ ConstTensor varTensor(batchNormLayer->m_Variance->GetTensorInfo(), batchNormLayer->m_Variance->Map(true));
+
+ auto convDescriptor = convLayer->GetParameters();
+ ConstTensor weightsTensor(convLayer->m_Weight->GetTensorInfo(), convLayer->m_Weight->Map(true));
+
+ armnnUtils::DataLayoutIndexed dataLayout(convDescriptor.m_DataLayout);
+ auto weightsShape = convLayer->m_Weight->GetTensorInfo().GetShape();
+ const unsigned int outputChannels = weightsShape[0];
+ const unsigned int inputChannels = weightsShape[dataLayout.GetChannelsIndex()];
+ const unsigned int weightsHeight = weightsShape[dataLayout.GetHeightIndex()];
+ const unsigned int weightsWidth = weightsShape[dataLayout.GetWidthIndex()];
+
+ const auto* weightsBuffer = static_cast<const float*>(weightsTensor.GetMemoryArea());
+ const auto* betaBuffer = static_cast<const float*>(betaTensor.GetMemoryArea());
+ const auto* gammaBuffer = static_cast<const float*>(gammaTensor.GetMemoryArea());
+ const auto* meanBuffer = static_cast<const float*>(meanTensor.GetMemoryArea());
+ const auto* varBuffer = static_cast<const float*>(varTensor.GetMemoryArea());
+
+ std::vector<float> weightsVector (weightsBuffer, weightsBuffer + weightsTensor.GetNumElements());
+ std::vector<float> betaVector (betaBuffer, betaBuffer + betaTensor.GetNumElements());
+ std::vector<float> gammaVector (gammaBuffer, gammaBuffer + gammaTensor.GetNumElements());
+ std::vector<float> meanVector (meanBuffer, meanBuffer + meanTensor.GetNumElements());
+ std::vector<float> varianceVector(varBuffer, varBuffer + varTensor.GetNumElements());
+
+ // fusedWeights = ( gamma * weights ) / ( std - epsilon);
+ std::vector<float> fusedWeightsVector(weightsVector.size());
+
+ unsigned int i = 0;
+ for (unsigned int cOut = 0; cOut < outputChannels; ++cOut)
+ {
+ auto mult = gammaVector[cOut] / sqrtf (varianceVector[cOut] + epsilon);
+ for (unsigned int cInput = 0; cInput < inputChannels; ++cInput)
+ {
+ for (unsigned int h = 0; h < weightsHeight; ++h)
+ {
+ for (unsigned int w = 0; w < weightsWidth; ++w)
+ {
+ fusedWeightsVector[i] = mult * weightsVector[i];
+ i++;
+ }
+ }
+ }
+ }
+ ConstTensor fusedWeightsTensor(convLayer->m_Weight->GetTensorInfo(), fusedWeightsVector);
+
+ // fusedBias = (gamma * (bias - mean)) / (variance - epsilon) + beta;
+ std::vector<float> fusedBiasVector(outputChannels);
+ if (convDescriptor.m_BiasEnabled)
+ {
+ ARMNN_ASSERT_MSG(convLayer->m_Bias != nullptr,
+ "FuseBatchNorm: Bias data should not be null if bias is enabled.");
+
+ ConstTensor biasTensor(convLayer->m_Bias->GetTensorInfo(), convLayer->m_Bias->Map(true));
+ const auto* biasBuffer = static_cast<const float*>(biasTensor.GetMemoryArea());
+ std::vector<float> biasVector(biasBuffer, biasBuffer + biasTensor.GetNumElements());
+
+ for (unsigned int cOut = 0; cOut < outputChannels; ++cOut)
+ {
+ fusedBiasVector[cOut] = ((gammaVector[cOut] * (biasVector[cOut] - meanVector[cOut])) /
+ sqrtf(varianceVector[cOut] + epsilon)) + betaVector[cOut];
+ }
+ }
+ else
+ {
+ convDescriptor.m_BiasEnabled = true;
+ std::vector<float> biasVector(outputChannels, 0);
+
+ for (unsigned int cOut = 0; cOut < outputChannels; ++cOut)
+ {
+ fusedBiasVector[cOut] = ((gammaVector[cOut] * (biasVector[cOut] - meanVector[cOut])) /
+ sqrtf(varianceVector[cOut] + epsilon)) + betaVector[cOut];
+ }
+ }
+ ConstTensor fusedBiasTensor(TensorInfo({outputChannels}, DataType::Float32), fusedBiasVector);
+
+ // Insert the new convolution layer that has batch norm parameters fused into
+ const std::string name = std::string("fused-") + child.GetName() + std::string("-into-") + base.GetName();
+ auto& newConv2dLayer = *graph.InsertNewLayer<ConvLayer>(base.GetInputSlot(0),
+ convDescriptor,
+ name.c_str());
+ newConv2dLayer.m_Weight = std::make_unique<ScopedCpuTensorHandle>(fusedWeightsTensor);
+ newConv2dLayer.m_Bias = std::make_unique<ScopedCpuTensorHandle>(ConstTensor(fusedBiasTensor));
+
+ // Reconnects with original parent.
+ newConv2dLayer.GetOutputSlot().MoveAllConnections(*parentOut);
+ // Parent is now the new convolution2d layer.
+ parentOut = &newConv2dLayer.GetOutputSlot();
+
+ // Moves connections in child output to parent layer.
+ // Child layer will be removed as it's left unconnected.
+ // Base layer will be removed if left unconnected.
+ child.GetOutputSlot().MoveAllConnections(*parentOut);
+ }
+ }
+protected:
+ FuseBatchNorm() = default;
+ ~FuseBatchNorm() = default;
+};
+
+using FuseBatchNormIntoConvolution2D =
+ OptimizeForExclusiveConnection<Convolution2dLayer,
+ BatchNormalizationLayer,
+ FuseBatchNorm<Convolution2dLayer>>;
+
+} // namespace optimizations
+} // namespace armnn
\ No newline at end of file
diff --git a/src/armnn/optimizations/Optimization.hpp b/src/armnn/optimizations/Optimization.hpp
index 1796ac8..320cae2 100644
--- a/src/armnn/optimizations/Optimization.hpp
+++ b/src/armnn/optimizations/Optimization.hpp
@@ -122,4 +122,60 @@
using OptimizeForTypeImpl<BaseType, OptimizeForConnectionImpl<BaseType, ChildType, Wrapped>>::OptimizeForTypeImpl;
};
+/// Wrapper Optimization class that calls Wrapped::Run for every connection BaseType -> ChildType.
+/// - Wrapped class mustn't remove the base layer. The optimizer will remove it if left unconnected
+/// after applying each optimization.
+/// - Wrapped class mustn't affect existing connections in the same output. It might add new ones.
+/// - Children layers are removed if left unconnected after applying the wrapped optimization.
+template <typename BaseType, typename ChildType, typename Wrapped>
+class OptimizeForExclusiveConnectionImpl : public Wrapped
+{
+public:
+ using Wrapped::Wrapped;
+
+ void Run(Graph& graph, BaseType& base) const
+ {
+ for (auto output = base.BeginOutputSlots(); output != base.EndOutputSlots(); ++output)
+ {
+ if (output->GetNumConnections() == 1)
+ {
+ for (auto&& childInput : output->GetConnections())
+ {
+ if (childInput->GetOwningLayer().GetType() == LayerEnumOf<ChildType>())
+ {
+ Wrapped::Run(graph, *childInput);
+ }
+ }
+
+ // Removes unconnected children.
+ for (unsigned int i = 0; i < output->GetNumConnections();)
+ {
+ Layer* child = &output->GetConnection(i)->GetOwningLayer();
+
+ if (child->IsOutputUnconnected())
+ {
+ graph.EraseLayer(child);
+ }
+ else
+ {
+ ++i;
+ }
+ }
+ }
+ }
+ }
+
+protected:
+ ~OptimizeForExclusiveConnectionImpl() = default;
+};
+
+template <typename BaseType, typename ChildType, typename Wrapped>
+class OptimizeForExclusiveConnection final
+ : public OptimizeForTypeImpl<BaseType, OptimizeForExclusiveConnectionImpl<BaseType, ChildType, Wrapped>>
+{
+public:
+ using OptimizeForTypeImpl<BaseType,
+ OptimizeForExclusiveConnectionImpl<BaseType, ChildType, Wrapped>>::OptimizeForTypeImpl;
+};
+
} // namespace armnn
diff --git a/src/armnn/test/OptimizerTests.cpp b/src/armnn/test/OptimizerTests.cpp
index 3af50ec..879905b 100644
--- a/src/armnn/test/OptimizerTests.cpp
+++ b/src/armnn/test/OptimizerTests.cpp
@@ -597,11 +597,11 @@
};
BOOST_TEST(CheckSequence(graph.cbegin(),
- graph.cend(),
- &IsLayerOfType<armnn::InputLayer>,
- &IsLayerOfType<armnn::PadLayer>,
- checkSimpleConv2d,
- &IsLayerOfType<armnn::OutputLayer>));
+ graph.cend(),
+ &IsLayerOfType<armnn::InputLayer>,
+ &IsLayerOfType<armnn::PadLayer>,
+ checkSimpleConv2d,
+ &IsLayerOfType<armnn::OutputLayer>));
armnn::Optimizer::Pass(graph, armnn::MakeOptimizations(FoldPadIntoConvolution2d()));
@@ -622,10 +622,10 @@
};
BOOST_TEST(CheckSequence(graph.cbegin(),
- graph.cend(),
- &IsLayerOfType<armnn::InputLayer>,
- checkPadFoldedIntoConv2d,
- &IsLayerOfType<armnn::OutputLayer>));
+ graph.cend(),
+ &IsLayerOfType<armnn::InputLayer>,
+ checkPadFoldedIntoConv2d,
+ &IsLayerOfType<armnn::OutputLayer>));
}
@@ -798,4 +798,294 @@
}
}
+BOOST_AUTO_TEST_CASE(OptimizeForExclusiveConnections_fuse_Test)
+{
+ using namespace armnn;
+ // Define layers information
+ Convolution2dDescriptor convolution2dDescriptor;
+ convolution2dDescriptor.m_BiasEnabled = false;
+ convolution2dDescriptor.m_DataLayout = DataLayout::NHWC;
+ BatchNormalizationDescriptor batchNormDescriptor;
+ batchNormDescriptor.m_DataLayout = DataLayout::NHWC;
+
+ const unsigned int inputDimensionSizes[] = {1, 4, 4, 3}; // NHWCin
+ const unsigned int weightsDimensionSizes[] = {1, 2, 2, 3}; // CoutHWCin
+ const unsigned int outputDimensionSizes[] = {1, 3, 3, 1}; // NHWCout
+ const unsigned int outputChannelSize[] = {outputDimensionSizes[3]}; // Cout
+
+ TensorInfo inputInfo (4, inputDimensionSizes, DataType::Float32);
+ TensorInfo outputInfo(4, outputDimensionSizes, DataType::Float32);
+
+ std::vector<float> weightsVector = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12};
+ ConstTensor weights (TensorInfo(4, weightsDimensionSizes, DataType::Float32), weightsVector);
+
+
+ std::vector<float> betaVector = {0.1f};
+ std::vector<float> gammaVector = {0.5f};
+ std::vector<float> meanVector = {0};
+ std::vector<float> varianceVector = {1};
+ ConstTensor beta (TensorInfo(1, outputChannelSize, DataType::Float32), betaVector);
+ ConstTensor gamma (TensorInfo(1, outputChannelSize, DataType::Float32), gammaVector);
+ ConstTensor mean (TensorInfo(1, outputChannelSize, DataType::Float32), meanVector);
+ ConstTensor variance(TensorInfo(1, outputChannelSize, DataType::Float32), varianceVector);
+
+ // Define the network
+ Graph graph;
+ auto input = graph.AddLayer<InputLayer>(0, "input");
+ auto conv = graph.AddLayer<Convolution2dLayer>(convolution2dDescriptor, "convolution");
+ auto batchNorm = graph.AddLayer<BatchNormalizationLayer>(batchNormDescriptor, "batchNorm");
+ auto output = graph.AddLayer<OutputLayer>(0, "output");
+
+ // Set layer information
+ input ->GetOutputSlot().SetTensorInfo(inputInfo);
+ conv ->GetOutputSlot().SetTensorInfo(outputInfo);
+ batchNorm->GetOutputSlot().SetTensorInfo(outputInfo);
+ conv ->m_Weight = std::make_unique<ScopedCpuTensorHandle>(weights);
+ batchNorm->m_Beta = std::make_unique<ScopedCpuTensorHandle>(beta);
+ batchNorm->m_Gamma = std::make_unique<ScopedCpuTensorHandle>(gamma);
+ batchNorm->m_Mean = std::make_unique<ScopedCpuTensorHandle>(mean);
+ batchNorm->m_Variance = std::make_unique<ScopedCpuTensorHandle>(variance);
+ if (convolution2dDescriptor.m_BiasEnabled)
+ {
+ std::vector<float> biasVector = {11};
+ ConstTensor bias (TensorInfo(1, outputChannelSize, DataType::Float32), biasVector);
+ conv->m_Bias = std::make_unique<ScopedCpuTensorHandle>(bias);
+ }
+
+ // Connect layers
+ input ->GetOutputSlot(0).Connect(conv ->GetInputSlot(0));
+ conv ->GetOutputSlot(0).Connect(batchNorm->GetInputSlot(0));
+ batchNorm ->GetOutputSlot(0).Connect(output ->GetInputSlot(0));
+
+ BOOST_CHECK(4 == graph.GetNumLayers());
+ BOOST_TEST(CheckSequence(graph.cbegin(),
+ graph.cend(),
+ &IsLayerOfType<InputLayer>,
+ &IsLayerOfType<Convolution2dLayer>,
+ &IsLayerOfType<BatchNormalizationLayer>,
+ &IsLayerOfType<OutputLayer>));
+
+ // Optimize graph
+ armnn::Optimizer::Pass(graph, MakeOptimizations(FuseBatchNormIntoConvolution2D()));
+
+ auto checkFusedConv2d = [ ](const armnn::Layer* const layer) -> bool
+ {
+ return IsLayerOfType<armnn::Convolution2dLayer>(layer) &&
+ (layer->GetNameStr() == "fused-batchNorm-into-convolution");
+ };
+
+ BOOST_CHECK(3 == graph.GetNumLayers());
+ BOOST_TEST(CheckSequence(graph.cbegin(),
+ graph.cend(),
+ &IsLayerOfType<InputLayer>,
+ checkFusedConv2d,
+ &IsLayerOfType<OutputLayer>));
+}
+
+BOOST_AUTO_TEST_CASE(OptimizeForExclusiveConnections_notFuse_Test)
+{
+ // Define the network
+ Graph graph;
+ Convolution2dDescriptor convolution2dDescriptor;
+ BatchNormalizationDescriptor batchNormDescriptor;
+
+ auto input = graph.AddLayer<InputLayer>(0, "input");
+ auto conv = graph.AddLayer<Convolution2dLayer>(convolution2dDescriptor, "convolution");
+ auto batchNorm = graph.AddLayer<BatchNormalizationLayer>(batchNormDescriptor, "batchNorm");
+ auto output = graph.AddLayer<OutputLayer>(0, "output");
+ auto output2 = graph.AddLayer<OutputLayer>(1, "output2");
+
+ // Connect layers
+ input ->GetOutputSlot(0).Connect(conv ->GetInputSlot(0));
+ conv ->GetOutputSlot(0).Connect(batchNorm->GetInputSlot(0));
+ batchNorm ->GetOutputSlot(0).Connect(output ->GetInputSlot(0));
+ conv ->GetOutputSlot(0).Connect(output2 ->GetInputSlot(0));
+
+ BOOST_CHECK(5 == graph.GetNumLayers());
+ BOOST_TEST(CheckSequence(graph.cbegin(),
+ graph.cend(),
+ &IsLayerOfType<armnn::InputLayer>,
+ &IsLayerOfType<armnn::Convolution2dLayer>,
+ &IsLayerOfType<armnn::BatchNormalizationLayer>,
+ &IsLayerOfType<armnn::OutputLayer>,
+ &IsLayerOfType<armnn::OutputLayer>));
+ // Optimize graph
+ armnn::Optimizer::Pass(graph, armnn::MakeOptimizations(FuseBatchNormIntoConvolution2D()));
+
+ BOOST_CHECK(5 == graph.GetNumLayers());
+ BOOST_TEST(CheckSequence(graph.cbegin(),
+ graph.cend(),
+ &IsLayerOfType<armnn::InputLayer>,
+ &IsLayerOfType<armnn::Convolution2dLayer>,
+ &IsLayerOfType<armnn::BatchNormalizationLayer>,
+ &IsLayerOfType<armnn::OutputLayer>,
+ &IsLayerOfType<armnn::OutputLayer>));
+}
+
+BOOST_AUTO_TEST_CASE(Fuse_batchNorm_into_Conv2D_Float32_Test)
+{
+ using namespace armnn;
+
+ // Define layers information
+ Convolution2dDescriptor convolution2dDescriptor;
+ convolution2dDescriptor.m_BiasEnabled = false;
+ convolution2dDescriptor.m_DataLayout = DataLayout::NHWC;
+ convolution2dDescriptor.m_StrideX = 1;
+ convolution2dDescriptor.m_StrideY = 1;
+ BatchNormalizationDescriptor batchNormDescriptor;
+ batchNormDescriptor.m_DataLayout = DataLayout::NHWC;
+
+ const unsigned int inputDimensionSizes[] = {1, 4, 4, 3}; // NHWCin
+ const unsigned int weightsDimensionSizes[] = {4, 2, 2, 3}; // CoutHWCin
+ const unsigned int outputDimensionSizes[] = {1, 3, 3, 4}; // NHWCout
+ const unsigned int outputChannelSize[] = {outputDimensionSizes[3]}; // Cout
+
+ TensorInfo inputInfo (4, inputDimensionSizes, DataType::Float32);
+ TensorInfo outputInfo(4, outputDimensionSizes, DataType::Float32);
+
+ std::vector<float> weightsVector = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12,
+ 11, 12, 13, 14, 15, 16, 17, 18, 19, 110, 111, 112,
+ 21, 22, 23, 24, 25, 26, 27, 28, 29, 210, 211, 212,
+ 31, 32, 33, 34, 35, 36, 37, 38, 39, 310, 311, 312};
+ TensorInfo weightsInfo(4, weightsDimensionSizes, DataType::Float32);
+ ConstTensor weights (weightsInfo, weightsVector);
+ std::vector<float> biasVector = {3.3f, 3.2f, 3.1f, 3.0f};
+ TensorInfo biasInfo(1, outputChannelSize, DataType::Float32);
+ ConstTensor bias (biasInfo, biasVector);
+ Optional<ConstTensor> optionalBias = Optional<ConstTensor>(bias);
+
+ std::vector<float> betaVector = {0.0f, 0.2f, 0.3f, 0.4f};
+ std::vector<float> gammaVector = {0.5f, 0.6f, 0.7f, 0.8f};
+ std::vector<float> meanVector = {0.1f, 0.2f, 0.3f, 0.4f};
+ std::vector<float> varianceVector = {1.0f, 1.1f, 1.2f, 1.3f};
+ ConstTensor beta (TensorInfo(1, outputChannelSize, DataType::Float32), betaVector);
+ ConstTensor gamma (TensorInfo(1, outputChannelSize, DataType::Float32), gammaVector);
+ ConstTensor mean (TensorInfo(1, outputChannelSize, DataType::Float32), meanVector);
+ ConstTensor variance(TensorInfo(1, outputChannelSize, DataType::Float32), varianceVector);
+
+ auto inputSize = inputDimensionSizes[0]*inputDimensionSizes[1]*inputDimensionSizes[2]*inputDimensionSizes[3];
+ auto outputSize = outputDimensionSizes[0]*outputDimensionSizes[1]*outputDimensionSizes[2]*outputDimensionSizes[3];
+
+ // FIRST NETWORK: Fused
+
+ // Construct ArmNN network
+ NetworkId networkIdentifier;
+ INetworkPtr network = INetwork::Create();
+ IConnectableLayer *inputLayer = network->AddInputLayer(0);
+ IConnectableLayer *convLayer = network->AddConvolution2dLayer(convolution2dDescriptor,
+ weights,
+ optionalBias,
+ "convolution");
+ IConnectableLayer *batchNormLayer = network->AddBatchNormalizationLayer(batchNormDescriptor,
+ mean,
+ variance,
+ beta,
+ gamma,
+ "batchNorm");
+ IConnectableLayer *outputLayer = network->AddOutputLayer(0);
+
+ inputLayer ->GetOutputSlot(0).Connect(convLayer ->GetInputSlot(0));
+ convLayer ->GetOutputSlot(0).Connect(batchNormLayer->GetInputSlot(0));
+ batchNormLayer ->GetOutputSlot(0).Connect(outputLayer ->GetInputSlot(0));
+
+ // Create ArmNN runtime
+ IRuntime::CreationOptions options; // default options
+ IRuntimePtr run = IRuntime::Create(options);
+
+ //Set the tensors in the network.
+ inputLayer ->GetOutputSlot(0).SetTensorInfo(inputInfo);
+ convLayer ->GetOutputSlot(0).SetTensorInfo(outputInfo);
+ batchNormLayer ->GetOutputSlot(0).SetTensorInfo(outputInfo);
+
+ // Optimise ArmNN network
+ IOptimizedNetworkPtr optNet = Optimize(*network, {Compute::CpuRef}, run->GetDeviceSpec());
+ if (!optNet)
+ {
+ // This shouldn't happen for this simple sample, with reference backend.
+ // But in general usage Optimize could fail if the hardware at runtime cannot
+ // support the model that has been provided.
+ std::cerr << "Error: Failed to optimise the input network." << std::endl;
+ }
+
+ // Load graph into runtime
+ run->LoadNetwork(networkIdentifier, std::move(optNet));
+
+ //Creates structures for inputs and outputs.
+ std::vector<float> inputData(inputSize, 128);
+ std::vector<float> outputData(outputSize);
+
+ InputTensors inputTensors {{0, ConstTensor(run->GetInputTensorInfo (networkIdentifier, 0), inputData.data())}};
+ OutputTensors outputTensors{{0, Tensor(run->GetOutputTensorInfo(networkIdentifier, 0), outputData.data())}};
+
+
+ // Execute network
+ run->EnqueueWorkload(networkIdentifier, inputTensors, outputTensors);
+
+ // SECOND NETWORK: NotFused
+
+ // Construct ArmNN network
+ NetworkId networkIdentifierNotFused;
+ INetworkPtr networkNotFused = INetwork::Create();
+ IConnectableLayer *inputLayerNotFused = networkNotFused->AddInputLayer(0);
+ IConnectableLayer *convLayerNotFused = networkNotFused->AddConvolution2dLayer(convolution2dDescriptor,
+ weights,
+ optionalBias,
+ "convolution");
+ IConnectableLayer *batchNormLayerNotFused = networkNotFused->AddBatchNormalizationLayer(batchNormDescriptor,
+ mean,
+ variance,
+ beta,
+ gamma,
+ "batchNorm");
+ IConnectableLayer *outputLayerNotFused = networkNotFused->AddOutputLayer(0);
+ IConnectableLayer *output2LayerNotFused = networkNotFused->AddOutputLayer(1);
+
+
+ inputLayerNotFused ->GetOutputSlot(0).Connect(convLayerNotFused ->GetInputSlot(0));
+ convLayerNotFused ->GetOutputSlot(0).Connect(batchNormLayerNotFused->GetInputSlot(0));
+ batchNormLayerNotFused ->GetOutputSlot(0).Connect(outputLayerNotFused ->GetInputSlot(0));
+ convLayerNotFused ->GetOutputSlot(0).Connect(output2LayerNotFused ->GetInputSlot(0));
+
+ // Create ArmNN runtime
+ IRuntimePtr runNotFused = IRuntime::Create(options);
+
+ //Set the tensors in the network.
+ inputLayerNotFused ->GetOutputSlot(0).SetTensorInfo(inputInfo);
+ convLayerNotFused ->GetOutputSlot(0).SetTensorInfo(outputInfo);
+ batchNormLayerNotFused ->GetOutputSlot(0).SetTensorInfo(outputInfo);
+
+ // Optimise ArmNN network
+ IOptimizedNetworkPtr optNetNotFused = Optimize(*networkNotFused, {Compute::CpuRef}, runNotFused->GetDeviceSpec());
+ if (!optNetNotFused)
+ {
+ // This shouldn't happen for this simple sample, with reference backend.
+ // But in general usage Optimize could fail if the hardware at runtime cannot
+ // support the model that has been provided.
+ std::cerr << "Error: Failed to optimise the input network." << std::endl;
+ }
+
+ // Load graph into runtime
+ runNotFused->LoadNetwork(networkIdentifierNotFused, std::move(optNetNotFused));
+
+ //Creates structures for inputs and outputs.
+ std::vector<float> inputDataNotFused(inputSize, 128);
+ std::vector<float> outputDataNotFused(outputSize);
+ std::vector<float> outputData2NotFused(outputSize);
+
+ InputTensors inputTensorsNotFused{
+ {0, ConstTensor(runNotFused->GetInputTensorInfo(networkIdentifierNotFused, 0), inputDataNotFused.data())}};
+ OutputTensors outputTensorsNotFused{
+ {0, Tensor(runNotFused->GetOutputTensorInfo(networkIdentifierNotFused, 0), outputDataNotFused.data())},
+ {1, Tensor(runNotFused->GetOutputTensorInfo(networkIdentifierNotFused, 1), outputData2NotFused.data())}};
+
+ // Execute network
+ runNotFused->EnqueueWorkload(networkIdentifierNotFused, inputTensorsNotFused, outputTensorsNotFused);
+
+ // Check the output of the fused-convolution matches with the output of the batchNormm in the "NotFused" network
+ for (unsigned int n = 0; n < outputData.size(); ++n)
+ {
+ BOOST_CHECK_CLOSE(outputData[n], outputDataNotFused[n], 0.001);
+ }
+}
+
BOOST_AUTO_TEST_SUITE_END()