blob: ee4ba6b65fcc883b94b8c73ab60ac7366d8570a7 (
plain)
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
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "ClConcatWorkload.hpp"
#include "ClWorkloadUtils.hpp"
#include <aclCommon/ArmComputeTensorUtils.hpp>
#include <backendsCommon/CpuTensorHandle.hpp>
#include <cl/ClTensorHandle.hpp>
#include <cl/ClLayerSupport.hpp>
#include <arm_compute/core/Types.h>
#include <boost/polymorphic_pointer_cast.hpp>
namespace armnn
{
using namespace armcomputetensorutils;
namespace
{
size_t CalcAxis(const MergerDescriptor& desc)
{
return (desc.GetNumDimensions() - desc.GetConcatAxis()) - 1;
}
} //namespace
arm_compute::Status ClConcatWorkloadValidate(const std::vector<const TensorInfo*>& inputs,
const TensorInfo& output,
const MergerDescriptor& descriptor)
{
std::vector<arm_compute::TensorInfo> aclInputs;
for (const TensorInfo* input : inputs)
{
arm_compute::TensorInfo aclInputInfo = BuildArmComputeTensorInfo(*input, armnn::DataLayout::NCHW);
aclInputs.emplace_back(aclInputInfo);
}
const arm_compute::TensorInfo aclOutputInfo = BuildArmComputeTensorInfo(output);
std::vector<arm_compute::ITensorInfo*> aclInputPtrs;
for (arm_compute::ITensorInfo& input : aclInputs)
{
aclInputPtrs.emplace_back(&input);
}
size_t aclAxis = CalcAxis(descriptor);
return arm_compute::CLConcatenateLayer::validate(aclInputPtrs, &aclOutputInfo, aclAxis);
}
ClConcatWorkload::ClConcatWorkload(const MergerQueueDescriptor& descriptor, const WorkloadInfo& info)
: BaseWorkload<MergerQueueDescriptor>(descriptor, info)
{
bool allInputsAreSubtensors = true;
// Check that all inputs are sub-tensors
for (auto input : descriptor.m_Inputs)
{
if (!input->GetParent())
{
// Non sub-tensor input found so we need to execute the merger function
allInputsAreSubtensors = false;
break;
}
}
if (allInputsAreSubtensors)
{
// Can skip configuring the merger function since it's not executed
return;
}
std::vector<arm_compute::ICLTensor *> aclInputs;
for (auto input : m_Data.m_Inputs)
{
arm_compute::ICLTensor& aclInput = boost::polymorphic_pointer_downcast<IClTensorHandle>(input)->GetTensor();
aclInputs.emplace_back(&aclInput);
}
arm_compute::ICLTensor& output = boost::polymorphic_pointer_downcast<IClTensorHandle>(
m_Data.m_Outputs[0])->GetTensor();
// Create the layer function
m_Layer.reset(new arm_compute::CLConcatenateLayer());
// Configure input and output tensors
size_t aclAxis = CalcAxis(descriptor.m_Parameters);
m_Layer->configure(aclInputs, &output, aclAxis);
// Prepare
m_Layer->prepare();
}
void ClConcatWorkload::Execute() const
{
if (m_Layer)
{
ARMNN_SCOPED_PROFILING_EVENT_CL("ClConcatWorkload_Execute");
m_Layer->run();
}
}
} //namespace armnn
|
