src/armnn/backends/ClWorkloads/ClMultiplicationFloat32Workload.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 "ClMultiplicationFloat32Workload.hpp"
   7 #include "backends/ClTensorHandle.hpp"
   8 #include "backends/CpuTensorHandle.hpp"
   9
  10 namespace armnn
  11 {
  12
  13 arm_compute::Status ClMultiplicationWorkloadValidate(const TensorInfo& input0,
  14                                                      const TensorInfo& input1,
  15                                                      const TensorInfo& output)
  16 {
  17     const arm_compute::TensorInfo aclInput1 = armcomputetensorutils::BuildArmComputeTensorInfo(input0);
  18     const arm_compute::TensorInfo aclInput2 = armcomputetensorutils::BuildArmComputeTensorInfo(input1);
  19     const arm_compute::TensorInfo aclOutput = armcomputetensorutils::BuildArmComputeTensorInfo(output);
  20
  21     // At the time of writing, configure() will fail if a rounding policy other than TO_ZERO is supplied to it,
  22     // when providing a scale of 1.0 for F32 tensors, even though the provided rounding policy appears to be
  23     // ignored for F32 tensors.
  24     return arm_compute::CLPixelWiseMultiplication::validate(&aclInput1,
  25                                                             &aclInput2,
  26                                                             &aclOutput,
  27                                                             1.0f,
  28                                                             arm_compute::ConvertPolicy::SATURATE,
  29                                                             arm_compute::RoundingPolicy::TO_ZERO);
  30 }
  31
  32
  33 ClMultiplicationFloat32Workload::ClMultiplicationFloat32Workload(const MultiplicationQueueDescriptor& descriptor,
  34                                                                  const WorkloadInfo& info)
  35     : FloatWorkload<MultiplicationQueueDescriptor>(descriptor, info)
  36 {
  37     m_Data.ValidateInputsOutputs("ClMultiplicationFloat32Workload", 2, 1);
  38
  39     arm_compute::ICLTensor& input0 = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
  40     arm_compute::ICLTensor& input1 = static_cast<IClTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
  41     arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
  42     // Construct
  43     m_PixelWiseMultiplication.configure(&input0,
  44                                         &input1,
  45                                         &output,
  46                                         1.0f,
  47                                         arm_compute::ConvertPolicy::SATURATE,
  48                                         arm_compute::RoundingPolicy::TO_NEAREST_EVEN);
  49 }
  50
  51 void ClMultiplicationFloat32Workload::Execute() const
  52 {
  53     ARMNN_SCOPED_PROFILING_EVENT_CL("ClMultiplicationFloat32Workload_Execute");
  54
  55     // Executes the layer.
  56     m_PixelWiseMultiplication.run();
  57 }
  58
  59 } //namespace armnn