commit d80a7126b0abdd532a9f731559827a23f2e565e0
author Matthew Bentham <matthew.bentham@arm.com> Tue Jan 08 17:52:37 2019 +0000
committer Matthew Bentham <matthew.bentham@arm.com> Tue Jan 08 18:00:12 2019 +0000
tree e04f6454e2353469dd9806b2a589c54b61dd777d
parent a1d3c6a49f35d7d3f11cc7e1b588d1d5401bdbf1

Refactor: Don't include all ComputeLibrary function definitions everywhere.

Just include the function definition that is specifically needed for each workload.
Also, tighten up the scope where Compute Library functions are available.

Knocks about 30seconds off a 4m30s single-threaded compile of the Neon workloads.

Change-Id: Idac438f3bc77ff978295fbc9505cb42447def145

src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp

diff --git a/src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp b/src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp
index 44d5035..fc80f41 100644
--- a/src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp
+++ b/src/backends/neon/workloads/NeonBatchNormalizationWorkload.cpp
@@ -4,9 +4,13 @@
 //
 
 #include "NeonBatchNormalizationWorkload.hpp"
+
+#include "NeonWorkloadUtils.hpp"
+
 #include <backendsCommon/CpuTensorHandle.hpp>
 #include <aclCommon/ArmComputeTensorUtils.hpp>
-#include <armnn/ArmNN.hpp>
+
+#include <arm_compute/runtime/NEON/functions/NEBatchNormalizationLayer.h>
 
 namespace armnn
 {
@@ -68,13 +72,15 @@
     m_Beta = std::make_unique<arm_compute::Tensor>();
     BuildArmComputeTensor(*m_Beta, m_Data.m_Beta->GetTensorInfo());
 
-    m_Layer.configure(&input,
-                      &output,
-                      m_Mean.get(),
-                      m_Variance.get(),
-                      m_Beta.get(),
-                      m_Gamma.get(),
-                      m_Data.m_Parameters.m_Eps);
+    auto layer = std::make_unique<arm_compute::NEBatchNormalizationLayer>();
+    layer->configure(&input,
+                     &output,
+                     m_Mean.get(),
+                     m_Variance.get(),
+                     m_Beta.get(),
+                     m_Gamma.get(),
+                     m_Data.m_Parameters.m_Eps);
+    m_Layer.reset(layer.release());
 
     InitializeArmComputeTensorData(*m_Mean, m_Data.m_Mean);
     InitializeArmComputeTensorData(*m_Variance, m_Data.m_Variance);
@@ -83,14 +89,14 @@
 
     // Force Compute Library to perform the necessary copying and reshaping, after which
     // delete all the input tensors that will no longer be needed
-    m_Layer.prepare();
+    m_Layer->prepare();
     FreeUnusedTensors();
 }
 
 void NeonBatchNormalizationWorkload::Execute() const
 {
     ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonBatchNormalizationWorkload_Execute");
-    m_Layer.run();
+    m_Layer->run();
 }
 
 void NeonBatchNormalizationWorkload::FreeUnusedTensors()