xref: /aosp_15_r20/external/armnn/src/backends/neon/workloads/NeonAdditionWorkload.cpp (revision 89c4ff92f2867872bb9e2354d150bf0c8c502810)
1 //
2 // Copyright © 2017 Arm Ltd and Contributors. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5 
6 #include "NeonAdditionWorkload.hpp"
7 #include "NeonWorkloadUtils.hpp"
8 
9 #include <aclCommon/ArmComputeTensorUtils.hpp>
10 #include <aclCommon/ArmComputeUtils.hpp>
11 
12 #include <armnn/utility/PolymorphicDowncast.hpp>
13 #include <armnn/backends/TensorHandle.hpp>
14 
15 #include <arm_compute/runtime/NEON/functions/NEArithmeticAddition.h>
16 
17 namespace armnn
18 {
19 
NeonAdditionWorkloadValidate(const TensorInfo & input0,const TensorInfo & input1,const TensorInfo & output,const ActivationDescriptor * activationDescriptor)20 arm_compute::Status NeonAdditionWorkloadValidate(const TensorInfo& input0,
21                                                  const TensorInfo& input1,
22                                                  const TensorInfo& output,
23                                                  const ActivationDescriptor* activationDescriptor)
24 {
25     const arm_compute::TensorInfo aclInput0 = armcomputetensorutils::BuildArmComputeTensorInfo(input0);
26     const arm_compute::TensorInfo aclInput1 = armcomputetensorutils::BuildArmComputeTensorInfo(input1);
27     const arm_compute::TensorInfo aclOutput = armcomputetensorutils::BuildArmComputeTensorInfo(output);
28 
29     const arm_compute::ActivationLayerInfo activationInfo = ConvertActivationDescriptorToAclActivationLayerInfo(
30             activationDescriptor);
31 
32     return arm_compute::NEArithmeticAddition::validate(&aclInput0,
33                                                        &aclInput1,
34                                                        &aclOutput,
35                                                        arm_compute::ConvertPolicy::SATURATE,
36                                                        activationInfo);
37 }
38 
39 
NeonAdditionWorkload(const AdditionQueueDescriptor & descriptor,const WorkloadInfo & info)40 NeonAdditionWorkload::NeonAdditionWorkload(const AdditionQueueDescriptor& descriptor,
41                                            const WorkloadInfo& info)
42     : NeonBaseWorkload<AdditionQueueDescriptor>(descriptor, info)
43 {
44     m_Data.ValidateInputsOutputs("NeonAdditionWorkload", 2, 1);
45 
46     arm_compute::ITensor& input1 = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
47     arm_compute::ITensor& input2 = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
48     arm_compute::ITensor& output = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
49 
50     const arm_compute::ActivationLayerInfo activationInfo = ConvertAdditionalInfoToAclActivationLayerInfo(descriptor);
51 
52     auto layer = std::make_unique<arm_compute::NEArithmeticAddition>();
53     layer->configure(&input1, &input2, &output, arm_compute::ConvertPolicy::SATURATE, activationInfo);
54     m_AddLayer.reset(layer.release());
55 }
56 
Execute() const57 void NeonAdditionWorkload::Execute() const
58 {
59     ARMNN_SCOPED_PROFILING_EVENT_NEON_GUID("NeonAdditionWorkload_Execute", this->GetGuid());
60     m_AddLayer->run();
61 }
62 
63 } //namespace armnn
64 
65