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
|
/*
* Copyright (c) 2017-2018 ARM Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include "helpers.h"
#if defined(DATA_TYPE)
/** This kernel reshapes the tensor's low three dimensions to single column
*
* @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=short
*
* @param[in] src_ptr Pointer to the source tensor. Supported data types: F16/F32
* @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)
* @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)
* @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)
* @param[in] src_step_z src_stride_z * number of elements along Y processed per workitem(in bytes)
* @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor
* @param[out] dst_ptr Pointer to the destination tensor. Same as @p src_ptr
* @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)
* @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)
* @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor
* @param[in] bias_ptr Pointer to the bias tensor. Same as @p src_ptr
* @param[in] bias_stride_x Stride of the bias tensor in X dimension (in bytes)
* @param[in] bias_step_x bias_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] bias_offset_first_element_in_bytes The offset of the first element in the source tensor
* @param[in] width The width of the input tensor
* @param[in] height The height of the input tensor
* @param[in] depth The depth of the input tensor
* @param[in] total_filters Total number of filters. 4th dimension of the weights matrix
*/
__kernel void reshape_to_columns_nchw(
TENSOR3D_DECLARATION(src),
IMAGE_DECLARATION(dst),
#ifdef HAS_BIAS
VECTOR_DECLARATION(bias),
#endif /* HAS_BIAS */
uint width, uint height, uint depth, uint total_filters)
{
Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
bool is_last_thread = (get_global_id(0) == (get_global_size(0) - 1) && get_global_id(1) == (get_global_size(1) - 1) && get_global_id(2) == (get_global_size(2) - 1));
__global uchar *tmp_src_ptr = src.ptr;
__global uchar *tmp_dst_ptr = dst_ptr + dst_offset_first_element_in_bytes + get_global_id(0) * dst_stride_y + get_global_id(1) * width * dst_stride_y + get_global_id(
2) * width * height * dst_stride_y;
#ifdef HAS_BIAS
__global uchar *tmp_bias_ptr = bias_ptr + bias_offset_first_element_in_bytes;
#endif /* HAS_BIAS */
if(is_last_thread)
{
for(uint i = 0; i < total_filters; ++i)
{
*((__global DATA_TYPE *)tmp_dst_ptr) = *((__global DATA_TYPE *)tmp_src_ptr);
#ifdef HAS_BIAS
*((__global DATA_TYPE *)(tmp_dst_ptr + dst_stride_y)) = *((__global DATA_TYPE *)(tmp_bias_ptr));
tmp_bias_ptr += bias_stride_x;
#endif /* HAS_BIAS */
tmp_src_ptr += depth * src_stride_z;
tmp_dst_ptr += dst_stride_x;
}
}
else
{
for(uint i = 0; i < total_filters; ++i)
{
*((__global DATA_TYPE *)tmp_dst_ptr) = *((__global DATA_TYPE *)tmp_src_ptr);
tmp_src_ptr += depth * src_stride_z;
tmp_dst_ptr += dst_stride_x;
}
}
}
/** This kernel reshapes the tensor's low three dimensions to single column
*
* @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=short
*
* @param[in] src_ptr Pointer to the source tensor. Supported data types: F16/F32
* @param[in] src_stride_x Stride of the source tensor in X dimension (in bytes)
* @param[in] src_step_x src_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] src_stride_y Stride of the source tensor in Y dimension (in bytes)
* @param[in] src_step_y src_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] src_stride_z Stride of the source tensor in Z dimension (in bytes)
* @param[in] src_step_z src_stride_z * number of elements along Y processed per workitem(in bytes)
* @param[in] src_offset_first_element_in_bytes The offset of the first element in the source tensor
* @param[out] dst_ptr Pointer to the destination tensor. Same as @p src_ptr
* @param[in] dst_stride_x Stride of the destination tensor in X dimension (in bytes)
* @param[in] dst_step_x dst_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] dst_stride_y Stride of the destination tensor in Y dimension (in bytes)
* @param[in] dst_step_y dst_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] dst_offset_first_element_in_bytes The offset of the first element in the destination tensor
* @param[in] bias_ptr Pointer to the bias tensor. Same as @p src_ptr
* @param[in] bias_stride_x Stride of the bias tensor in X dimension (in bytes)
* @param[in] bias_step_x bias_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] bias_offset_first_element_in_bytes The offset of the first element in the source tensor
* @param[in] depth The depth of the input tensor
* @param[in] width The width of the input tensor
* @param[in] height The height of the input tensor
* @param[in] total_filters Total number of filters. 4th dimension of the weights matrix
*/
__kernel void reshape_to_columns_nhwc(
TENSOR3D_DECLARATION(src),
IMAGE_DECLARATION(dst),
#ifdef HAS_BIAS
VECTOR_DECLARATION(bias),
#endif /* HAS_BIAS */
uint depth, uint width, uint height, uint total_filters)
{
Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
bool is_last_thread = (get_global_id(0) == (get_global_size(0) - 1) && get_global_id(1) == (get_global_size(1) - 1) && get_global_id(2) == (get_global_size(2) - 1));
__global uchar *tmp_src_ptr = src.ptr;
__global uchar *tmp_dst_ptr = dst_ptr + dst_offset_first_element_in_bytes + get_global_id(1) * dst_stride_y + get_global_id(2) * width * dst_stride_y + get_global_id(
0) * width * height * dst_stride_y;
#ifdef HAS_BIAS
__global uchar *tmp_bias_ptr = bias_ptr + bias_offset_first_element_in_bytes;
#endif /* HAS_BIAS */
if(is_last_thread)
{
for(uint i = 0; i < total_filters; ++i)
{
*((__global DATA_TYPE *)tmp_dst_ptr) = *((__global DATA_TYPE *)tmp_src_ptr);
#ifdef HAS_BIAS
*((__global DATA_TYPE *)(tmp_dst_ptr + dst_stride_y)) = *((__global DATA_TYPE *)(tmp_bias_ptr));
tmp_bias_ptr += bias_stride_x;
#endif /* HAS_BIAS */
tmp_src_ptr += height * src_stride_z;
tmp_dst_ptr += dst_stride_x;
}
}
else
{
for(uint i = 0; i < total_filters; ++i)
{
*((__global DATA_TYPE *)tmp_dst_ptr) = *((__global DATA_TYPE *)tmp_src_ptr);
tmp_src_ptr += height * src_stride_z;
tmp_dst_ptr += dst_stride_x;
}
}
}
#endif // defined(DATA_TYPE)
|
