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 (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef __NEURUN_BACKEND_ACL_COMMON_SWIZZLE_H__
#define __NEURUN_BACKEND_ACL_COMMON_SWIZZLE_H__
#include <cassert>
#include <model/Layout.h>
namespace neurun
{
namespace backend
{
namespace acl_common
{
class ARMComputeAxis
{
public:
ARMComputeAxis() = default;
public:
explicit ARMComputeAxis(uint32_t value) : _value{value}
{
// DO NOTHING
}
public:
uint32_t value(void) const { return _value; }
private:
uint32_t _value;
};
// Convert axis in acl order
inline ARMComputeAxis ToARMComputeAxis(uint32_t rank, uint32_t axis,
const model::Layout org_layout = model::Layout::UNKNOWN,
const model::Layout acl_layout = model::Layout::UNKNOWN)
{
assert(rank > axis);
const ARMComputeAxis reversed{(rank - axis) - 1};
if (rank >= 4 && org_layout == model::Layout::NHWC && acl_layout == model::Layout::NCHW)
{
// NHWC -> WHCN
// DEPTH
if (0 == reversed.value())
{
return ARMComputeAxis{2};
}
// WIDTH
if (1 == reversed.value())
{
return ARMComputeAxis{0};
}
// HEIGHT
if (2 == reversed.value())
{
return ARMComputeAxis{1};
}
}
if (rank >= 4 && org_layout == model::Layout::NCHW && acl_layout == model::Layout::NHWC)
{
// NCHW -> CWHN
// WIDTH
if (0 == reversed.value())
{
return ARMComputeAxis{1};
}
// HEIGHT
if (1 == reversed.value())
{
return ARMComputeAxis{2};
}
// DEPTH
if (2 == reversed.value())
{
return ARMComputeAxis{0};
}
}
return reversed;
}
inline ::arm_compute::Coordinates
getARMComputeAxises(uint32_t rank, const model::Layout org_layout = model::Layout::UNKNOWN,
const model::Layout acl_layout = model::Layout::UNKNOWN)
{
::arm_compute::Coordinates res{};
res.set_num_dimensions(rank);
for (uint32_t axis = 0; axis < rank; ++axis)
{
res.set(axis, ToARMComputeAxis(rank, axis, org_layout, acl_layout).value());
}
return res;
}
// Restructure runtime_permutationVector to ACL_permutationVector
inline ::arm_compute::PermutationVector
getARMComputePermutationVector(uint32_t rank, const std::vector<int32_t> runtime_pv,
const model::Layout org_layout = model::Layout::UNKNOWN,
const model::Layout acl_layout = model::Layout::UNKNOWN)
{
// rank upto 4 is supported
assert(rank <= 4);
assert(runtime_pv.size() > 0);
int new_pv[4] = {0};
::arm_compute::Coordinates axises = getARMComputeAxises(rank, org_layout, acl_layout);
for (uint32_t i = 0; i < rank; ++i)
{
new_pv[axises[i]] = ToARMComputeAxis(rank, runtime_pv[i], org_layout, acl_layout).value();
}
::arm_compute::PermutationVector ACL_PV =
::arm_compute::PermutationVector{new_pv[0], new_pv[1], new_pv[2], new_pv[3]};
ACL_PV.set_num_dimensions(rank);
return ACL_PV;
}
template <typename T>
inline T ReorderBits(T in, size_t numOfBits,
const model::Layout org_layout = model::Layout::UNKNOWN,
const model::Layout acl_layout = model::Layout::UNKNOWN)
{
assert(numOfBits > 0);
T out = 0;
for (int32_t i = numOfBits - 1; i >= 0; --i)
{
const uint32_t toShift =
numOfBits - ToARMComputeAxis(numOfBits, i, org_layout, acl_layout).value() - 1;
out += ((in & 1) << toShift);
in >>= 1;
}
return out;
}
} // namespace acl_common
} // namespace backend
} // namespace neurun
#endif // __NEURUN_BACKEND_ACL_COMMON_SWIZZLE_H__
|
