• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //
2 // Copyright © 2020 Arm Ltd and Contributors. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5 
6 #include "ClGatherWorkload.hpp"
7 #include "ClWorkloadUtils.hpp"
8 #include <aclCommon/ArmComputeUtils.hpp>
9 #include <cl/ClTensorHandle.hpp>
10 
11 using namespace armnn::armcomputetensorutils;
12 
13 namespace armnn
14 {
ClGatherWorkloadValidate(const TensorInfo & input,const TensorInfo & indices,const TensorInfo & output,const GatherDescriptor & descriptor)15 arm_compute::Status ClGatherWorkloadValidate(const TensorInfo& input,
16                                              const TensorInfo& indices,
17                                              const TensorInfo& output,
18                                              const GatherDescriptor& descriptor)
19 {
20     const arm_compute::TensorInfo aclInput   = BuildArmComputeTensorInfo(input);
21     const arm_compute::TensorInfo aclIndices = BuildArmComputeTensorInfo(indices);
22     const arm_compute::TensorInfo aclOutput  = BuildArmComputeTensorInfo(output);
23 
24     int aclAxis = ComputeAclAxis(descriptor.m_Axis, input);
25 
26     return arm_compute::CLGather::validate(&aclInput, &aclIndices, &aclOutput, aclAxis);
27 }
28 
ClGatherWorkload(const GatherQueueDescriptor & descriptor,const WorkloadInfo & info)29 ClGatherWorkload::ClGatherWorkload(const GatherQueueDescriptor& descriptor,
30                                    const WorkloadInfo& info)
31         : BaseWorkload<GatherQueueDescriptor>(descriptor, info)
32 {
33     m_Data.ValidateInputsOutputs("ClGatherWorkload", 1, 1);
34 
35     arm_compute::ICLTensor& input    = static_cast<IClTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
36     arm_compute::ICLTensor& indices  = static_cast<IClTensorHandle*>(m_Data.m_Inputs[1])->GetTensor();
37     arm_compute::ICLTensor& output   = static_cast<IClTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
38 
39     int aclAxis = ComputeAclAxis(descriptor.m_Parameters.m_Axis, info.m_InputTensorInfos[0]);
40 
41     m_Layer.configure(&input, &indices, &output, aclAxis);
42 };
43 
Execute() const44 void ClGatherWorkload::Execute() const
45 {
46     ARMNN_SCOPED_PROFILING_EVENT_CL("ClGatherWorkload_Execute");
47     RunClFunction(m_Layer, CHECK_LOCATION());
48 }
49 } // namespace armnn
50