xref: /aosp_15_r20/external/armnn/src/backends/neon/workloads/NeonGatherWorkload.cpp (revision 89c4ff92f2867872bb9e2354d150bf0c8c502810)
1 //
2 // Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5 
6 #include "NeonGatherWorkload.hpp"
7 #include "NeonWorkloadUtils.hpp"
8 #include <armnn/utility/PolymorphicDowncast.hpp>
9 #include <aclCommon/ArmComputeUtils.hpp>
10 
11 namespace armnn
12 {
NeonGatherWorkloadValidate(const TensorInfo & input,const TensorInfo & indices,const TensorInfo & output,const GatherDescriptor & descriptor)13 arm_compute::Status NeonGatherWorkloadValidate(const TensorInfo& input,
14                                                const TensorInfo& indices,
15                                                const TensorInfo& output,
16                                                const GatherDescriptor& descriptor)
17 {
18     const arm_compute::TensorInfo aclInput   = BuildArmComputeTensorInfo(input);
19     const arm_compute::TensorInfo aclIndices = BuildArmComputeTensorInfo(indices);
20     const arm_compute::TensorInfo aclOutput  = BuildArmComputeTensorInfo(output);
21 
22     int aclAxis = ComputeAclAxis(descriptor.m_Axis, input);
23 
24     return arm_compute::NEGather::validate(&aclInput, &aclIndices, &aclOutput, aclAxis);
25 }
26 
NeonGatherWorkload(const GatherQueueDescriptor & descriptor,const WorkloadInfo & info)27 NeonGatherWorkload::NeonGatherWorkload(const GatherQueueDescriptor& descriptor,
28                                        const WorkloadInfo& info)
29         : NeonBaseWorkload<GatherQueueDescriptor>(descriptor, info)
30 {
31     // Report Profiling Details
32     ARMNN_REPORT_PROFILING_WORKLOAD_DESC("NeonGatherWorkload_Construct",
33                                          descriptor.m_Parameters,
34                                          info,
35                                          this->GetGuid());
36 
37     m_Data.ValidateInputsOutputs("NeonGatherWorkload", 2, 1);
38 
39     arm_compute::ITensor& input   = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
40     arm_compute::ITensor& indices = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
41     arm_compute::ITensor& output  = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
42 
43     int aclAxis = ComputeAclAxis(descriptor.m_Parameters.m_Axis, info.m_InputTensorInfos[0]);
44 
45     m_Layer.configure(&input, &indices, &output, aclAxis);
46 }
47 
Execute() const48 void NeonGatherWorkload::Execute() const
49 {
50     ARMNN_SCOPED_PROFILING_EVENT_NEON_GUID("NeonGatherWorkload_Execute", this->GetGuid());
51     m_Layer.run();
52 }
53 } //namespace armnn