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
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
|
//
// Copyright © 2017 Arm Ltd. All rights reserved.
// SPDX-License-Identifier: MIT
//
#include "TensorCopyUtils.hpp"
#include <Half.hpp>
#ifdef ARMCOMPUTECL_ENABLED
#include <cl/ClTensorHandle.hpp>
#endif
#if ARMCOMPUTENEON_ENABLED
#include <neon/NeonTensorHandle.hpp>
#endif
#if ARMCOMPUTECLENABLED || ARMCOMPUTENEON_ENABLED
#include <aclCommon/ArmComputeTensorUtils.hpp>
#endif
#include <backendsCommon/CpuTensorHandle.hpp>
#include <boost/cast.hpp>
#include <algorithm>
#include <cstring>
void CopyDataToITensorHandle(armnn::ITensorHandle* tensorHandle, const void* mem)
{
switch (tensorHandle->GetType())
{
case armnn::ITensorHandle::Cpu:
{
auto handle = boost::polymorphic_downcast<armnn::ScopedCpuTensorHandle*>(tensorHandle);
memcpy(handle->GetTensor<void>(), mem, handle->GetTensorInfo().GetNumBytes());
break;
}
#ifdef ARMCOMPUTECL_ENABLED
case armnn::ITensorHandle::CL:
{
using armnn::armcomputetensorutils::CopyArmComputeITensorData;
auto handle = boost::polymorphic_downcast<armnn::IClTensorHandle*>(tensorHandle);
handle->Map(true);
switch(handle->GetDataType())
{
case arm_compute::DataType::F32:
CopyArmComputeITensorData(static_cast<const float*>(mem), handle->GetTensor());
break;
case arm_compute::DataType::QASYMM8:
CopyArmComputeITensorData(static_cast<const uint8_t*>(mem), handle->GetTensor());
break;
case arm_compute::DataType::F16:
CopyArmComputeITensorData(static_cast<const armnn::Half*>(mem), handle->GetTensor());
break;
default:
{
throw armnn::UnimplementedException();
}
}
handle->Unmap();
break;
}
#endif
#if ARMCOMPUTENEON_ENABLED
case armnn::ITensorHandle::Neon:
{
using armnn::armcomputetensorutils::CopyArmComputeITensorData;
auto handle = boost::polymorphic_downcast<armnn::INeonTensorHandle*>(tensorHandle);
switch (handle->GetDataType())
{
case arm_compute::DataType::F32:
CopyArmComputeITensorData(static_cast<const float*>(mem), handle->GetTensor());
break;
case arm_compute::DataType::QASYMM8:
CopyArmComputeITensorData(static_cast<const uint8_t*>(mem), handle->GetTensor());
break;
default:
{
throw armnn::UnimplementedException();
}
}
break;
}
#endif
default:
{
throw armnn::UnimplementedException();
}
}
}
void CopyDataFromITensorHandle(void* mem, const armnn::ITensorHandle* tensorHandle)
{
switch (tensorHandle->GetType())
{
case armnn::ITensorHandle::Cpu:
{
auto handle = boost::polymorphic_downcast<const armnn::ScopedCpuTensorHandle*>(tensorHandle);
memcpy(mem, handle->GetTensor<void>(), handle->GetTensorInfo().GetNumBytes());
break;
}
#ifdef ARMCOMPUTECL_ENABLED
case armnn::ITensorHandle::CL:
{
using armnn::armcomputetensorutils::CopyArmComputeITensorData;
auto handle = boost::polymorphic_downcast<const armnn::IClTensorHandle*>(tensorHandle);
const_cast<armnn::IClTensorHandle*>(handle)->Map(true);
switch(handle->GetDataType())
{
case arm_compute::DataType::F32:
CopyArmComputeITensorData(handle->GetTensor(), static_cast<float*>(mem));
break;
case arm_compute::DataType::QASYMM8:
CopyArmComputeITensorData(handle->GetTensor(), static_cast<uint8_t*>(mem));
break;
case arm_compute::DataType::F16:
CopyArmComputeITensorData(handle->GetTensor(), static_cast<armnn::Half*>(mem));
break;
default:
{
throw armnn::UnimplementedException();
}
}
const_cast<armnn::IClTensorHandle*>(handle)->Unmap();
break;
}
#endif
#if ARMCOMPUTENEON_ENABLED
case armnn::ITensorHandle::Neon:
{
using armnn::armcomputetensorutils::CopyArmComputeITensorData;
auto handle = boost::polymorphic_downcast<const armnn::INeonTensorHandle*>(tensorHandle);
switch (handle->GetDataType())
{
case arm_compute::DataType::F32:
CopyArmComputeITensorData(handle->GetTensor(), static_cast<float*>(mem));
break;
case arm_compute::DataType::QASYMM8:
CopyArmComputeITensorData(handle->GetTensor(), static_cast<uint8_t*>(mem));
break;
default:
{
throw armnn::UnimplementedException();
}
}
break;
}
#endif
default:
{
throw armnn::UnimplementedException();
}
}
}
void AllocateAndCopyDataToITensorHandle(armnn::ITensorHandle* tensorHandle, const void* mem)
{
tensorHandle->Allocate();
CopyDataToITensorHandle(tensorHandle, mem);
}
|
