commit 04a729708f986b1a69c1efc42d5cf18271cfae1e
author Sadik Armagan <sadik.armagan@arm.com> Mon Sep 14 15:44:18 2020 +0100
committer Sadik Armagan <sadik.armagan@arm.com> Mon Sep 14 14:42:59 2020 +0000
tree d39be642c3b5bfe9f9520ae69889b20dade10406
parent 80c6b146046252af153af27025a28fb59d33c5e6

IVGCVSW-5157 'Pipe ModelOption through Network::LoadNetwork() to Workload factory'

* Pass ModelOptions to WorkloadFactory
* Updated signature of CL and NEON Convolution2d workloads added FastMathEnabled param.


Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I536178be8e4dd4083489e69febadaf0feeba46d2

src/backends/neon/workloads/NeonConvolution2dWorkload.cpp

diff --git a/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp b/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp
index 83f7611..d35b968 100644
--- a/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp
+++ b/src/backends/neon/workloads/NeonConvolution2dWorkload.cpp
@@ -59,8 +59,10 @@
 }
 
 NeonConvolution2dWorkload::NeonConvolution2dWorkload(
-    const Convolution2dQueueDescriptor& descriptor, const WorkloadInfo& info,
-    std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager)
+    const Convolution2dQueueDescriptor& descriptor,
+    const WorkloadInfo& info,
+    std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager,
+    const bool isFastMathEnabled)
     : BaseWorkload<Convolution2dQueueDescriptor>(descriptor, info)
 {
     using arm_compute::NEDirectConvolutionLayer;
@@ -97,7 +99,19 @@
                                 &output,
                                 padStrideInfo,
                                 arm_compute::WeightsInfo(),
-                                aclDilationInfo);
+                                aclDilationInfo,
+                                arm_compute::ActivationLayerInfo(),
+                                isFastMathEnabled);
+
+    m_ConvolutionMethod =
+        convolutionLayer->get_convolution_method(input.info(),
+                                                 m_KernelTensor->info(),
+                                                 output.info(),
+                                                 padStrideInfo,
+                                                 arm_compute::WeightsInfo(),
+                                                 aclDilationInfo,
+                                                 arm_compute::ActivationLayerInfo(),
+                                                 isFastMathEnabled);
 
     m_ConvolutionLayer.reset(convolutionLayer.release());
 
@@ -120,6 +134,11 @@
     m_ConvolutionLayer->run();
 }
 
+arm_compute::ConvolutionMethod NeonConvolution2dWorkload::GetConvolutionMethod() const
+{
+    return m_ConvolutionMethod;
+}
+
 void NeonConvolution2dWorkload::FreeUnusedTensors()
 {
     FreeTensorIfUnused(m_KernelTensor);

src/backends/neon/workloads/NeonConvolution2dWorkload.hpp

diff --git a/src/backends/neon/workloads/NeonConvolution2dWorkload.hpp b/src/backends/neon/workloads/NeonConvolution2dWorkload.hpp
index 54e08a2..860d78b 100644
--- a/src/backends/neon/workloads/NeonConvolution2dWorkload.hpp
+++ b/src/backends/neon/workloads/NeonConvolution2dWorkload.hpp
@@ -28,17 +28,23 @@
 public:
     using BaseWorkload<Convolution2dQueueDescriptor>::m_Data;
 
-    NeonConvolution2dWorkload(const Convolution2dQueueDescriptor& descriptor, const WorkloadInfo& info,
-                              std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager);
+    NeonConvolution2dWorkload(const Convolution2dQueueDescriptor& descriptor,
+                              const WorkloadInfo& info,
+                              std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager,
+                              const bool isFastMathENabled = false);
 
     void Execute() const override;
 
+    arm_compute::ConvolutionMethod GetConvolutionMethod() const;
+
 private:
     std::unique_ptr<arm_compute::IFunction> m_ConvolutionLayer;
 
     std::unique_ptr<arm_compute::Tensor> m_KernelTensor;
     std::unique_ptr<arm_compute::Tensor> m_BiasTensor;
 
+    arm_compute::ConvolutionMethod m_ConvolutionMethod;
+
     void FreeUnusedTensors();
 
 };