src/armnn/backends/NeonWorkloads/NeonConvolution2dBaseWorkload.cpp

   1 //
   2 // Copyright © 2017 Arm Ltd. All rights reserved.
   3 // See LICENSE file in the project root for full license information.
   4 //
   5
   6 #include "backends/CpuTensorHandle.hpp"
   7 #include "backends/ArmComputeTensorUtils.hpp"
   8 #include "backends/NeonLayerSupport.hpp"
   9
  10 #include "NeonConvolution2dBaseWorkload.hpp"
  11
  12 namespace armnn
  13 {
  14
  15 template<armnn::DataType dataType>
  16 NeonConvolution2dBaseWorkload<dataType>::NeonConvolution2dBaseWorkload(const Convolution2dQueueDescriptor& descriptor,
  17                                                                        const WorkloadInfo& info)
  18     : TypedWorkload<Convolution2dQueueDescriptor, dataType>(descriptor, info)
  19 {
  20     using arm_compute::NEDirectConvolutionLayer;
  21     using namespace armcomputetensorutils;
  22
  23     ValidateData();
  24
  25     // todo: check tensor shapes match
  26
  27     arm_compute::ITensor& input = boost::polymorphic_downcast<INeonTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
  28     arm_compute::ITensor& output = boost::polymorphic_downcast<INeonTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
  29
  30     BuildArmComputeTensor(m_KernelTensor, m_Data.m_Weight->GetTensorInfo());
  31
  32     arm_compute::Tensor* optionalBiasTensor = nullptr;
  33     if (m_Data.m_Parameters.m_BiasEnabled)
  34     {
  35         BuildArmComputeTensor(m_BiasTensor, m_Data.m_Bias->GetTensorInfo());
  36         optionalBiasTensor = &m_BiasTensor;
  37     }
  38
  39     arm_compute::PadStrideInfo padStrideInfo(m_Data.m_Parameters.m_StrideX,
  40                                              m_Data.m_Parameters.m_StrideY,
  41                                              m_Data.m_Parameters.m_PadLeft,
  42                                              m_Data.m_Parameters.m_PadRight,
  43                                              m_Data.m_Parameters.m_PadTop,
  44                                              m_Data.m_Parameters.m_PadBottom,
  45                                              arm_compute::DimensionRoundingType::FLOOR);
  46
  47     const bool preferDirectConvolution =
  48         IsNeonDirectConvolutionPreferred(m_Data.m_Weight->GetTensorInfo(),
  49                                          m_Data.m_Parameters);
  50
  51     if (preferDirectConvolution)
  52     {
  53         auto directConvolutionLayer = std::make_unique<arm_compute::NEDirectConvolutionLayer>();
  54         directConvolutionLayer->configure(&input,
  55                                           &m_KernelTensor,
  56                                           optionalBiasTensor,
  57                                           &output,
  58                                           padStrideInfo);
  59         m_ConvolutionLayer.reset(directConvolutionLayer.release());
  60     }
  61     else
  62     {
  63         auto convolutionLayer = std::make_unique<arm_compute::NEConvolutionLayer>();
  64         convolutionLayer->configure(&input,
  65                                     &m_KernelTensor,
  66                                     optionalBiasTensor,
  67                                     &output,
  68                                     padStrideInfo);
  69         m_ConvolutionLayer.reset(convolutionLayer.release());
  70     }
  71     BOOST_ASSERT(m_ConvolutionLayer);
  72
  73     using Type = ResolveType<dataType>;
  74
  75     InitialiseArmComputeTensorData(m_KernelTensor, m_Data.m_Weight->template GetConstTensor<Type>());
  76 }
  77
  78 // Generate known implementations for linker
  79 template class NeonConvolution2dBaseWorkload<DataType::Float32>;
  80 template class NeonConvolution2dBaseWorkload<DataType::QuantisedAsymm8>;
  81
  82 } //namespace armnn
  83
  84