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
|
/*
* 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_EXEC_SINK_H__
#define __NEURUN_EXEC_SINK_H__
#include <cassert>
#include "cpp14/memory.h"
#include "util/feature/nhwc/View.h"
#include "util/feature/nchw/View.h"
#include <misc/feature/IndexIterator.h>
namespace neurun
{
namespace exec
{
struct ISink
{
virtual ~ISink() = default;
virtual void pull(::neurun::backend::operand::ITensor &tensor) const = 0;
};
template <typename T> class Sink final : public ISink
{
public:
Sink(T *base, const size_t size) : _base{base}, _size{size} {}
public:
void pull(::neurun::backend::operand::ITensor &tensor) const override
{
memcpy(_base, tensor.buffer(), _size);
}
private:
T *const _base;
const size_t _size;
};
class PermutateSink final : public ISink
{
public:
PermutateSink(void *output_buffer, const size_t &output_size, const model::operand::Shape &shape)
: _output_buffer{(uint8_t *)output_buffer}, _output_size{output_size}, _shape{shape}
{
}
public:
void pull(neurun::backend::operand::ITensor &tensor) const override
{
// do NCHW_TO_NHWC permutation
auto input_buffer = tensor.buffer();
auto rank = _shape.rank();
switch (rank)
{
case 0:
case 1:
{
memcpy(_output_buffer, input_buffer, _output_size);
break;
}
case 2:
{
auto matrix_shape = _shape.asMatrix();
for (auto h = 0; h < matrix_shape.H; ++h)
{
neurun::util::feature::Coordinate4D coord{0, h, 0, 0};
memcpy(_output_buffer + h * matrix_shape.W, input_buffer + tensor.calcOffset(coord),
matrix_shape.W * sizeof(float));
}
break;
}
case 3:
{
const int32_t depth = _shape.dim(0);
const int32_t height = _shape.dim(1);
const int32_t width = _shape.dim(2);
for (auto c = 0; c < depth; ++c)
{
for (auto h = 0; h < height; ++h)
{
neurun::util::feature::Coordinate4D coord{0, h, 0, c};
memcpy(_output_buffer + c * height * width + h * width,
input_buffer + tensor.calcOffset(coord), width * sizeof(float));
}
}
break;
}
case 4:
{
auto feature = _shape.asFeature();
const util::feature::nchw::View<float> from{&tensor};
util::feature::nhwc::View<float> into{feature, reinterpret_cast<float *>(_output_buffer),
_output_size};
::nnfw::misc::feature::iterate(feature)
<< [&](uint32_t batch, uint32_t ch, uint32_t row, uint32_t col) {
const auto value = from.at(batch, ch, row, col);
into.at(batch, ch, row, col) = value;
};
break;
}
default:
throw "NYI";
break;
}
}
private:
uint8_t *_output_buffer;
const size_t _output_size;
const model::operand::Shape _shape;
};
} // namespace exec
} // namespace neurun
#endif // __NEURUN_EXEC_SINK_H__
|
