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
|
/*
* Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright (c) 2016, 2017 ARM Limited.
*
* 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.
*/
#include "helpers.h"
#if defined(DATA_TYPE) && defined(DEPTH_OUT) && defined(OP_CODE)
/** Perform reduce max/min
*
* @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=short
* @attention Output tensor depth should be given as a preprocessor argument using -DDEPTH_OUT=size. e.g. -DDEPTH_OUT=16
* @attention Operation type(code) specifying which operation to perform should be passed as preprocessor argument using
* -DOP_CODE = number. e.g. -DOP_CODE=1
*
* @param[in] input_ptr Pointer to the source image. Supported data types: U8/S8/QASYMM8/U16/S16/F16/U32/S32/F32
* @param[in] input_stride_x Stride of the source image in X dimension (in bytes)
* @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] input_stride_y Stride of the source image in Y dimension (in bytes)
* @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes)
* @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes)
* @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image
* @param[in] input_stride_w Stride of the source tensor in W dimension (in bytes)
* @param[in] input_step_w output_stride_w * number of elements along W processed per workitem(in bytes)
* @param[out] output_ptr Pointer to the destination image. Supported data types: same as @p input_ptr
* @param[in] output_stride_x Stride of the destination image in X dimension (in bytes)
* @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes)
* @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] output_stride_z Stride of the source tensor in Z dimension (in bytes)
* @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes)
* @param[in] output_stride_w Stride of the source tensor in W dimension (in bytes)
* @param[in] output_step_w output_stride_w * number of elements along W processed per workitem(in bytes)
* @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image
* @param[in] axis Axis through which reduction occurs
* @param[in] dim Dimension across the axis to be reduced.
*/
__kernel void reduce_min_max(TENSOR4D_DECLARATION(input),
TENSOR4D_DECLARATION(output),
const int axis,
const int dim)
{
Tensor4D in = CONVERT_TO_TENSOR4D_STRUCT(input, 0);
Tensor4D out = CONVERT_TO_TENSOR4D_STRUCT(output, DEPTH_OUT);
int indices[4] =
{
get_global_id(0),
get_global_id(1),
get_global_id(2) % DEPTH_OUT,
get_global_id(2) / DEPTH_OUT,
};
DATA_TYPE value = *((__global DATA_TYPE *)tensor4D_offset(&in, indices[0], indices[1], indices[2], indices[3]));
for(int i = 1; i < dim; ++i)
{
indices[axis] = i;
#if OP_CODE == 1 // REDUCE_MAX
value = max(value, *((__global DATA_TYPE *)
tensor4D_offset(&in, indices[0], indices[1], indices[2], indices[3])));
#elif OP_CODE == 2 // REDUCE_MIN
value = min(value, *((__global DATA_TYPE *)
tensor4D_offset(&in, indices[0], indices[1], indices[2], indices[3])));
#else // OP NOT SUPPORTED
return;
#endif
}
*((__global DATA_TYPE *)out.ptr) = value;
}
/** Perform reduce sum/mean
*
* @note Datatype should be given as a preprocessor argument using -DDATA_TYPE=type. e.g. -DDATA_TYPE=short
* @attention Output tensor depth should be given as a preprocessor argument using -DDEPTH_OUT=size. e.g. -DDEPTH_OUT=16
* @attention Operation type(code) specifying which operation to perform should be passed as preprocessor argument using
* -DOP_CODE = number. e.g. -DOP_CODE=1
*
* @param[in] input_ptr Pointer to the source image. Supported data types: U8/S8/U16/S16/F16/U32/S32/F32
* @param[in] input_stride_x Stride of the source image in X dimension (in bytes)
* @param[in] input_step_x input_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] input_stride_y Stride of the source image in Y dimension (in bytes)
* @param[in] input_step_y input_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] input_stride_z Stride of the source tensor in Z dimension (in bytes)
* @param[in] input_step_z input_stride_z * number of elements along Z processed per workitem(in bytes)
* @param[in] input_offset_first_element_in_bytes The offset of the first element in the source image
* @param[in] input_stride_w Stride of the source tensor in W dimension (in bytes)
* @param[in] input_step_w output_stride_w * number of elements along W processed per workitem(in bytes)
* @param[out] output_ptr Pointer to the destination image. Supported data types: same as @p input_ptr
* @param[in] output_stride_x Stride of the destination image in X dimension (in bytes)
* @param[in] output_step_x output_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] output_stride_y Stride of the destination image in Y dimension (in bytes)
* @param[in] output_step_y output_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] output_stride_z Stride of the source tensor in Z dimension (in bytes)
* @param[in] output_step_z output_stride_z * number of elements along Z processed per workitem(in bytes)
* @param[in] output_stride_w Stride of the source tensor in W dimension (in bytes)
* @param[in] output_step_w output_stride_w * number of elements along W processed per workitem(in bytes)
* @param[in] output_offset_first_element_in_bytes The offset of the first element in the destination image
* @param[in] axis Axis through which reduction occurs
* @param[in] dim Dimension across the axis to be reduced.
*/
__kernel void reduce_sum_mean(TENSOR4D_DECLARATION(input),
TENSOR4D_DECLARATION(output),
const int axis,
const int dim)
{
Tensor4D in = CONVERT_TO_TENSOR4D_STRUCT(input, 0);
Tensor4D out = CONVERT_TO_TENSOR4D_STRUCT(output, DEPTH_OUT);
int indices[4] =
{
get_global_id(0),
get_global_id(1),
get_global_id(2) % DEPTH_OUT,
get_global_id(2) / DEPTH_OUT,
};
DATA_TYPE sum_value = (DATA_TYPE)0;
for(int i = 0; i < dim; ++i)
{
indices[axis] = i;
sum_value += *((__global DATA_TYPE *)tensor4D_offset(&in, indices[0], indices[1], indices[2], indices[3]));
}
#if OP_CODE == 3 // REDUCE_SUM
*((__global DATA_TYPE *)out.ptr) = sum_value;
#elif OP_CODE == 4 // REDUCE_MEAN
*((__global DATA_TYPE *)out.ptr) = sum_value / CONVERT(dim, DATA_TYPE);
#else // OP NOT SUPPORTED
return;
#endif
}
#endif // defined(DATA_TYPE) && defined(DEPTH_OUT) && defined(OP_CODE)
|
