commit 07810fc2fcdd34db74222d90cc73ef12a88e7b78
author Mike Kelly <mike.kelly@arm.com> Thu Nov 12 10:58:48 2020 +0000
committer Jim Flynn <jim.flynn@arm.com> Fri Nov 13 14:25:30 2020 +0000
tree 8becef8453674822d079815b06ae37310b97d2cf
parent 8502adeafbbb1db0acefa62560d93453e38dcadb

IVGCVSW-5328-5329 Fuse Activation

 * Added Fused Activation Optimization to both CL and Neon backends.
 * Added Fused Activation support to all the CL and Neon workloads
   that support it.
 * Changed ProfilingTest network to be a Convolution layer
   followed by an Abs layer rather than an Activation layer.
 * Added IBackendInternal::OptimizeSubgraphView function that can accept a
   ModelOptions.
 * Network will now call OptimizeSubgraphView passing in the ModelOptions.

Signed-off-by: Keith Davis <keith.davis@arm.com>
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Ib536ac3cbafc7d9b35c139ad9a65b7735262cd9d

src/backends/cl/workloads/ClFullyConnectedWorkload.cpp

diff --git a/src/backends/cl/workloads/ClFullyConnectedWorkload.cpp b/src/backends/cl/workloads/ClFullyConnectedWorkload.cpp
index 60eb138..eaec639 100644
--- a/src/backends/cl/workloads/ClFullyConnectedWorkload.cpp
+++ b/src/backends/cl/workloads/ClFullyConnectedWorkload.cpp
@@ -20,7 +20,8 @@
                                                      const TensorInfo& output,
                                                      const TensorInfo& weights,
                                                      const TensorInfo& biases,
-                                                     const FullyConnectedDescriptor& descriptor)
+                                                     const FullyConnectedDescriptor& descriptor,
+                                                     const ActivationDescriptor* activationDescriptor)
 {
     const arm_compute::TensorInfo aclInput = BuildArmComputeTensorInfo(input);
     const arm_compute::TensorInfo aclOutput = BuildArmComputeTensorInfo(output);
@@ -35,7 +36,7 @@
     }
 
     const arm_compute::FullyConnectedLayerInfo fullyConnectedLayerInfo =
-        ConvertFullyConnectedDescriptorToAclFullyConnectedLayerInfo(descriptor);
+        ConvertFullyConnectedDescriptorToAclFullyConnectedLayerInfo(descriptor, activationDescriptor);
 
     return arm_compute::CLFullyConnectedLayer::validate(&aclInput,
                                                         &aclWeights,
@@ -63,9 +64,11 @@
     arm_compute::ICLTensor& input  = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
     arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
 
-    // Construct
-    arm_compute::FullyConnectedLayerInfo fc_info;
-    fc_info.transpose_weights = m_Data.m_Parameters.m_TransposeWeightMatrix;
+    const arm_compute::ActivationLayerInfo activationInfo = ConvertAdditionalInfoToAclActivationLayerInfo(descriptor);
+
+    arm_compute::FullyConnectedLayerInfo fc_info =
+            ConvertFullyConnectedDescriptorToAclFullyConnectedLayerInfo(descriptor.m_Parameters, activationInfo);
+
     m_FullyConnectedLayer.configure(&input, m_WeightsTensor.get(), m_BiasesTensor.get(), &output, fc_info);
 
     InitializeArmComputeClTensorData(*m_WeightsTensor, m_Data.m_Weight);