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
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
|
/*
* Copyright (c) 2019 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright 2017 The TensorFlow Authors. 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 __NNFW_CKER_CONV_H__
#define __NNFW_CKER_CONV_H__
#include "cker/Types.h"
#include "cker/Shape.h"
#include "cker/Utils.h"
#include "cker/operation/reference/Conv.h"
#include "cker/operation/optimized/Conv.h"
#include <iostream>
#include <vector>
namespace nnfw
{
namespace cker
{
namespace
{
// Naive implementation of transpose for floats. Could be optimized to be more
// cache friendly, but for now it's a one-time cost on first run, and we would
// prefer to remove the need to do this at all eventually.
inline void TransposeFloatTensor(const float *input_data, const nnfw::cker::Shape &output_shape,
float *output_data)
{
const int rows = output_shape.Dims(1);
const int cols = output_shape.Dims(0);
for (int i = 0; i < rows; ++i)
{
for (int j = 0; j < cols; ++j)
{
const float in_value = input_data[i * cols + j];
output_data[j * rows + i] = in_value;
}
}
}
} // namespace
class Conv
{
public:
Conv() : _modified_filter_data(), _im2col_shape(4), _need_im2col(false), _prepared(false) {}
void prepareF32(const Shape &filter_shape, const float *filter_data, PaddingType padding_type,
bool &is_replaced_weights, uint32_t dilationWidthFactor,
uint32_t dilationHeightFactor)
{
if (!_prepared)
{
if (usableMultiThreaded(padding_type, dilationWidthFactor, dilationHeightFactor))
{
transposeFilter(filter_shape, filter_data, is_replaced_weights);
}
_prepared = true;
}
}
void prepareQ8uPerTensor(const Shape &input_shape, const Shape &kernel_shape,
const Shape &output_shape, uint32_t stride_width, uint32_t stride_height,
uint32_t dilation_width_factor, uint32_t dilation_height_factor)
{
if (!_prepared)
{
IsRequiredIm2col(input_shape, kernel_shape, output_shape, stride_width, stride_height,
dilation_width_factor, dilation_height_factor);
_prepared = true;
}
}
void operator()(const ConvParams ¶ms, const Shape &input_shape, const float *input_data,
const Shape &filter_shape, const float *filter_data, const Shape &bias_shape,
const float *bias_data, const Shape &output_shape, float *output_data)
{
if (usableMultiThreaded(params.padding_type, params.dilation_width_factor,
params.dilation_height_factor))
{
bool transposed_in_execution = false;
if (!_prepared)
{
// This means that filter is not constant
// TODO Apply optimized kernel if multithreaded kernel is slower than optimized kernel by
// transposing filter data
transposeFilter(filter_shape, filter_data, transposed_in_execution);
}
multithreaded::Conv(params, input_shape, input_data, filter_shape, &_modified_filter_data[0],
bias_shape, bias_data, output_shape, output_data);
}
else
{
// TODO Support optimized kernel
reference::Conv(params, input_shape, input_data, filter_shape, filter_data, bias_shape,
bias_data, output_shape, output_data);
}
}
void operator()(const ConvParams ¶ms, const Shape &input_shape, const uint8_t *input_data,
const Shape &filter_shape, const uint8_t *filter_data, const Shape &bias_shape,
const int32_t *bias_data, const Shape &output_shape, uint8_t *output_data)
{
if (!_prepared)
{
// This means that input or output are dynamic or filter is not constant
IsRequiredIm2col(input_shape, filter_shape, output_shape, params.stride_width,
params.stride_height, params.dilation_width_factor,
params.dilation_height_factor);
}
int im2col_size = _need_im2col ? _im2col_shape.FlatSize() : 1;
// Use heap if size is larger than 8MB
if (im2col_size > 8 * 1024 * 1024)
{
std::unique_ptr<uint8_t[]> im2col_data = std::make_unique<uint8_t[]>(im2col_size);
optimized::Conv(params, input_shape, input_data, filter_shape, filter_data, bias_shape,
bias_data, output_shape, output_data, _im2col_shape, im2col_data.get());
}
else
{
uint8_t im2col_data[im2col_size];
optimized::Conv(params, input_shape, input_data, filter_shape, filter_data, bias_shape,
bias_data, output_shape, output_data, _im2col_shape, im2col_data);
}
}
void operator()(const ConvParams ¶ms, const Shape &input_shape, const uint8_t *input_data,
const Shape &filter_shape, const uint8_t *filter_data,
const int32_t *filter_zero_point, const Shape &bias_shape,
const int32_t *bias_data, const Shape &output_shape, uint8_t *output_data)
{
reference::Conv<uint8_t, true>(params, _per_channel_output_multiplier.data(),
_per_channel_output_shift.data(), input_shape, input_data,
filter_shape, filter_data, filter_zero_point, bias_shape,
bias_data, output_shape, output_data);
}
void operator()(const ConvParams ¶ms, const Shape &input_shape, const int8_t *input_data,
const Shape &filter_shape, const int8_t *filter_data, const Shape &bias_shape,
const int32_t *bias_data, const Shape &output_shape, int8_t *output_data)
{
reference::Conv<int8_t, false>(params, _per_channel_output_multiplier.data(),
_per_channel_output_shift.data(), input_shape, input_data,
filter_shape, filter_data, nullptr /* filter_zero_point */,
bias_shape, bias_data, output_shape, output_data);
}
std::vector<int32_t> &per_channel_output_multiplier() { return _per_channel_output_multiplier; }
std::vector<int> &per_channel_output_shift() { return _per_channel_output_shift; }
private:
bool usableMultiThreaded(PaddingType padding_type, uint32_t dilation_width_factor,
int32_t dilation_height_factor)
{
return padding_type != PaddingType::kNone && std::thread::hardware_concurrency() > 1 &&
dilation_width_factor == 1 && dilation_height_factor == 1;
}
void transposeFilter(const Shape &filter_shape, const float *filter_data,
bool &is_replaced_weights)
{
const auto output_depth = filter_shape.Dims(0);
const Shape hwcn_filter_shape{filter_shape.FlatSize() / output_depth, output_depth};
_modified_filter_data.resize(hwcn_filter_shape.FlatSize());
TransposeFloatTensor(filter_data, hwcn_filter_shape, &_modified_filter_data[0]);
is_replaced_weights = true;
}
void IsRequiredIm2col(const Shape &input_shape, const Shape &kernel_shape,
const Shape &output_shape, uint32_t stride_width, uint32_t stride_height,
uint32_t dilation_width_factor, uint32_t dilation_height_factor)
{
const bool need_dilated_im2col = dilation_width_factor != 1 || dilation_height_factor != 1;
const bool need_non_dilated_im2col = stride_width != 1 || stride_height != 1 ||
kernel_shape.Dims(1) != 1 || kernel_shape.Dims(2) != 1;
_need_im2col = need_dilated_im2col || need_non_dilated_im2col;
if (_need_im2col)
{
_im2col_shape.SetDim(0, output_shape.Dims(0));
_im2col_shape.SetDim(1, output_shape.Dims(1));
_im2col_shape.SetDim(2, output_shape.Dims(2));
_im2col_shape.SetDim(3, input_shape.Dims(3) * kernel_shape.Dims(1) * kernel_shape.Dims(2));
}
}
private:
std::vector<float> _modified_filter_data;
Shape _im2col_shape;
bool _need_im2col;
bool _prepared;
// Per channel output multiplier and shift.
std::vector<int32_t> _per_channel_output_multiplier;
std::vector<int> _per_channel_output_shift;
};
struct ConvHybridTempArena
{
ConvHybridTempArena(int batch_size, int input_size)
{
input_quantized.resize(input_size);
// TODO: Optimize the case of batch_size = 1
input_scaling_factors.resize(batch_size);
input_offsets.resize(batch_size);
}
std::vector<int8_t> input_quantized;
std::vector<float> input_scaling_factors;
std::vector<int32_t> input_offsets;
};
} // namespace cker
} // namespace nnfw
#endif // __NNFW_CKER_CONCATENATION_H_
|
