From 07810fc2fcdd34db74222d90cc73ef12a88e7b78 Mon Sep 17 00:00:00 2001
From: Mike Kelly <mike.kelly@arm.com>
Date: Thu, 12 Nov 2020 10:58:48 +0000
Subject: 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
---
 .../neon/workloads/NeonBatchNormalizationWorkload.cpp   | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

(limited to 'src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp')

diff --git a/src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp b/src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp
index ff777dbf9b..33480faf69 100644
--- a/src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp
+++ b/src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp
@@ -8,7 +8,10 @@
 #include "NeonWorkloadUtils.hpp"
 
 #include <aclCommon/ArmComputeTensorUtils.hpp>
+#include <aclCommon/ArmComputeUtils.hpp>
+
 #include <armnn/utility/PolymorphicDowncast.hpp>
+
 #include <backendsCommon/CpuTensorHandle.hpp>
 
 #include <arm_compute/runtime/NEON/functions/NEBatchNormalizationLayer.h>
@@ -24,7 +27,8 @@ arm_compute::Status NeonBatchNormalizationValidate(const TensorInfo& input,
                                                    const TensorInfo& var,
                                                    const TensorInfo& beta,
                                                    const TensorInfo& gamma,
-                                                   const BatchNormalizationDescriptor& descriptor)
+                                                   const BatchNormalizationDescriptor& descriptor,
+                                                   const ActivationDescriptor* activationDescriptor)
 {
     const arm_compute::TensorInfo aclInputInfo =
           armcomputetensorutils::BuildArmComputeTensorInfo(input, descriptor.m_DataLayout);
@@ -39,13 +43,17 @@ arm_compute::Status NeonBatchNormalizationValidate(const TensorInfo& input,
     const arm_compute::TensorInfo aclGammaInfo =
           armcomputetensorutils::BuildArmComputeTensorInfo(gamma, descriptor.m_DataLayout);
 
+    const arm_compute::ActivationLayerInfo activationInfo = ConvertActivationDescriptorToAclActivationLayerInfo(
+            activationDescriptor);
+
     return arm_compute::NEBatchNormalizationLayer::validate(&aclInputInfo,
                                                             &aclOutputInfo,
                                                             &aclMeanInfo,
                                                             &aclVarInfo,
                                                             &aclBetaInfo,
                                                             &aclGammaInfo,
-                                                            descriptor.m_Eps);
+                                                            descriptor.m_Eps,
+                                                            activationInfo);
 }
 
 NeonBatchNormalizationWorkload::NeonBatchNormalizationWorkload(
@@ -73,6 +81,8 @@ NeonBatchNormalizationWorkload::NeonBatchNormalizationWorkload(
     m_Beta = std::make_unique<arm_compute::Tensor>();
     BuildArmComputeTensor(*m_Beta, m_Data.m_Beta->GetTensorInfo());
 
+    const arm_compute::ActivationLayerInfo activationInfo = ConvertAdditionalInfoToAclActivationLayerInfo(descriptor);
+
     auto layer = std::make_unique<arm_compute::NEBatchNormalizationLayer>();
     layer->configure(&input,
                      &output,
@@ -80,7 +90,8 @@ NeonBatchNormalizationWorkload::NeonBatchNormalizationWorkload(
                      m_Variance.get(),
                      m_Beta.get(),
                      m_Gamma.get(),
-                     m_Data.m_Parameters.m_Eps);
+                     m_Data.m_Parameters.m_Eps,
+                     activationInfo);
     m_Layer.reset(layer.release());
 
     InitializeArmComputeTensorData(*m_Mean, m_Data.m_Mean);
-- 
cgit v1.2.1