src/armnn/backends/NeonWorkloads/NeonAdditionFloat32Workload.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 "NeonAdditionFloat32Workload.hpp"
   7 #include "backends/ArmComputeTensorUtils.hpp"
   8 #include "backends/CpuTensorHandle.hpp"
   9
  10 namespace armnn
  11 {
  12
  13 arm_compute::Status NeonAdditionWorkloadValidate(const TensorInfo& input0,
  14                                                  const TensorInfo& input1,
  15                                                  const TensorInfo& output)
  16 {
  17     const arm_compute::TensorInfo aclInput0 = armcomputetensorutils::BuildArmComputeTensorInfo(input0);
  18     const arm_compute::TensorInfo aclInput1 = armcomputetensorutils::BuildArmComputeTensorInfo(input1);
  19     const arm_compute::TensorInfo aclOutput = armcomputetensorutils::BuildArmComputeTensorInfo(output);
  20
  21     return arm_compute::NEArithmeticAddition::validate(&aclInput0,
  22                                                        &aclInput1,
  23                                                        &aclOutput,
  24                                                        arm_compute::ConvertPolicy::SATURATE);
  25 }
  26
  27
  28 NeonAdditionFloat32Workload::NeonAdditionFloat32Workload(const AdditionQueueDescriptor& descriptor,
  29                                                          const WorkloadInfo& info)
  30     : FloatWorkload<AdditionQueueDescriptor>(descriptor, info)
  31 {
  32     m_Data.ValidateInputsOutputs("NeonAdditionFloat32Workload", 2, 1);
  33
  34     arm_compute::ITensor& input1 = boost::polymorphic_downcast<INeonTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
  35     arm_compute::ITensor& input2 = boost::polymorphic_downcast<INeonTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
  36     arm_compute::ITensor& output = boost::polymorphic_downcast<INeonTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
  37
  38     m_AddLayer.configure(&input1, &input2, &output, arm_compute::ConvertPolicy::SATURATE);
  39 }
  40
  41 void NeonAdditionFloat32Workload::Execute() const
  42 {
  43     ARMNN_SCOPED_PROFILING_EVENT_NEON("NeonAdditionFloat32Workload_Execute");
  44     m_AddLayer.run();
  45 }
  46
  47 } //namespace armnn
  48