1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
|
//
// Copyright © 2020 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "NeonSoftmaxWorkload.hpp"
#include "NeonWorkloadUtils.hpp"
#include <armnn/utility/PolymorphicDowncast.hpp>
#include <aclCommon/ArmComputeUtils.hpp>
#include <aclCommon/ArmComputeTensorUtils.hpp>
#include <arm_compute/runtime/NEON/functions/NESoftmaxLayer.h>
namespace armnn
{
arm_compute::Status NeonSoftmaxWorkloadValidate(const TensorInfo& input,
const TensorInfo& output,
const SoftmaxDescriptor& descriptor)
{
const arm_compute::TensorInfo aclInputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(input);
const arm_compute::TensorInfo aclOutputInfo = armcomputetensorutils::BuildArmComputeTensorInfo(output);
int aclAxis = ComputeAclAxis(descriptor.m_Axis, input);
return arm_compute::NESoftmaxLayer::validate(&aclInputInfo,
&aclOutputInfo,
descriptor.m_Beta,
aclAxis);
}
NeonSoftmaxWorkload::NeonSoftmaxWorkload(const SoftmaxQueueDescriptor& descriptor,
const WorkloadInfo& info, std::shared_ptr<arm_compute::MemoryManagerOnDemand>& memoryManager)
: BaseWorkload<SoftmaxQueueDescriptor>(descriptor, info)
{
// Report Profiling Details
ARMNN_REPORT_PROFILING_WORKLOAD_DESC("NeonSoftmaxWorkload_Construct",
descriptor.m_Parameters,
info,
this->GetGuid());
m_Data.ValidateInputsOutputs("NeonSoftmaxWorkload", 1, 1);
// The ArmCompute softmax layer uses 2D input/output tensors, so flatten the first three dimensions.
arm_compute::ITensor& input = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Inputs[0])->GetTensor();
arm_compute::ITensor& output = PolymorphicDowncast<IAclTensorHandle*>(m_Data.m_Outputs[0])->GetTensor();
auto layer = std::make_unique<arm_compute::NESoftmaxLayer>(memoryManager);
int aclAxis = ComputeAclAxis(m_Data.m_Parameters.m_Axis, info.m_InputTensorInfos[0]);
layer->configure(&input, &output, m_Data.m_Parameters.m_Beta, aclAxis);
m_SoftmaxLayer.reset(layer.release());
}
void NeonSoftmaxWorkload::Execute() const
{
ARMNN_SCOPED_PROFILING_EVENT_NEON_GUID("NeonSoftmaxWorkload_Execute", this->GetGuid());
m_SoftmaxLayer->run();
}
} //namespace armnn
|
