src/backends/cl/workloads/ClMultiplicationWorkload.cpp

   1 //
   2 // Copyright © 2017 Arm Ltd. All rights reserved.
   3 // SPDX-License-Identifier: MIT
   4 //
   5
   6 #include "ClMultiplicationWorkload.hpp"
   7 #include <cl/ClTensorHandle.hpp>
   8 #include <backendsCommon/CpuTensorHandle.hpp>
   9 #include "ClWorkloadUtils.hpp"
  10
  11 namespace armnn
  12 {
  13
  14 arm_compute::Status ClMultiplicationWorkloadValidate(const TensorInfo& input0,
  15                                                      const TensorInfo& input1,
  16                                                      const TensorInfo& output)
  17 {
  18     const arm_compute::TensorInfo aclInput1 = armcomputetensorutils::BuildArmComputeTensorInfo(input0);
  19     const arm_compute::TensorInfo aclInput2 = armcomputetensorutils::BuildArmComputeTensorInfo(input1);
  20     const arm_compute::TensorInfo aclOutput = armcomputetensorutils::BuildArmComputeTensorInfo(output);
  21
  22     // At the time of writing, configure() will fail if a rounding policy other than TO_ZERO is supplied to it,
  23     // when providing a scale of 1.0 for F32 tensors, even though the provided rounding policy appears to be
  24     // ignored for F32 tensors.
  25     return arm_compute::CLPixelWiseMultiplication::validate(&aclInput1,
  26                                                             &aclInput2,
  27                                                             &aclOutput,
  28                                                             1.0f,
  29                                                             arm_compute::ConvertPolicy::SATURATE,
  30                                                             arm_compute::RoundingPolicy::TO_ZERO);
  31 }
  32
  33
  34 ClMultiplicationWorkload::ClMultiplicationWorkload(const MultiplicationQueueDescriptor& descriptor,
  35                                                    const WorkloadInfo& info)
  36     : BaseWorkload<MultiplicationQueueDescriptor>(descriptor, info)
  37 {
  38     m_Data.ValidateInputsOutputs("ClMultiplicationWorkload", 2, 1);
  39
  40     arm_compute::ICLTensor& input0 = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
  41     arm_compute::ICLTensor& input1 = static_cast<IClTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
  42     arm_compute::ICLTensor& output = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
  43     // Construct
  44     m_PixelWiseMultiplication.configure(&input0,
  45                                         &input1,
  46                                         &output,
  47                                         1.0f,
  48                                         arm_compute::ConvertPolicy::SATURATE,
  49                                         arm_compute::RoundingPolicy::TO_NEAREST_EVEN);
  50 }
  51
  52 void ClMultiplicationWorkload::Execute() const
  53 {
  54     ARMNN_SCOPED_PROFILING_EVENT_CL("ClMultiplicationWorkload_Execute");
  55     RunClFunction(m_PixelWiseMultiplication, CHECK_LOCATION());
  56 }
  57
  58 } //namespace armnn