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
|
/*
* Copyright (c) 2016, 2017 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(FIXED_POINT_POSITION)
#include "fixed_point.h"
#if defined(SATURATE)
#define MUL_OP(x, y, scale, type, size) MUL_SAT_OP_EXPAND((x), (y), type, size, FIXED_POINT_POSITION)
#else // SATURATE
#define MUL_OP(x, y, scale, type, size) MUL_OP_EXPAND((x), (y), type, size, FIXED_POINT_POSITION)
#endif // SATURATE
#else // FIXED_POINT_POSITION
#if defined(SATURATE)
#define CONVERT_OP_INT_STR(x, type, size) (convert_##type##size##_sat(x))
#else // SATURATE
#define CONVERT_OP_INT_STR(x, type, size) (convert_##type##size(x))
#endif // SATURATE
#define CONVERT_OP_INT(x, type, size) CONVERT_OP_INT_STR(x, type, size)
#define MUL_OP(x, y, scale, type, size) CONVERT_OP_INT((x) * (y) >> scale, type, size)
#endif // FIXED_POINT_POSITION
/** Performs a pixelwise multiplication with integer scale of integer inputs.
*
* @attention The inputs and output data types need to be passed at compile time using -DDATA_TYPE_IN1, -DDATA_TYPE_IN2 and -DDATA_TYPE_OUT:
* e.g. -DDATA_TYPE_IN1=uchar -DDATA_TYPE_IN2=ushort -DDATA_TYPE_OUT=short
* @attention The data_type of the intermediate result of the multiplication should passed as well using -DDATA_TYPE_RES.
* e.g. If one of inputs is S16 -DDATA_TYPE_RES=int should be passed else -DDATA_TYPE_RES=short.
* @note In case of fixed-point operation -DFIXED_POINT_POSITION=fixed_point_position must be provided: e.g. -DFIXED_POINT_POSITION=3
*
* @param[in] in1_ptr Pointer to the source image. Supported data types: U8/QS8/QS16/S16
* @param[in] in1_stride_x Stride of the source image in X dimension (in bytes)
* @param[in] in1_step_x in1_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] in1_stride_y Stride of the source image in Y dimension (in bytes)
* @param[in] in1_step_y in1_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] in1_stride_z Stride of the source image in Y dimension (in bytes)
* @param[in] in1_step_z in1_stride_z * number of elements along Y processed per workitem(in bytes)
* @param[in] in1_offset_first_element_in_bytes The offset of the first element in the source image
* @param[in] in2_ptr Pointer to the source image. Supported data types: same as @p in1_ptr
* @param[in] in2_stride_x Stride of the source image in X dimension (in bytes)
* @param[in] in2_step_x in2_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] in2_stride_y Stride of the source image in Y dimension (in bytes)
* @param[in] in2_step_y in2_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] in2_stride_z Stride of the source image in Y dimension (in bytes)
* @param[in] in2_step_z in2_stride_z * number of elements along Y processed per workitem(in bytes)
* @param[in] in2_offset_first_element_in_bytes The offset of the first element in the source image
* @param[out] out_ptr Pointer to the destination image. Supported data types: same as @p in1_ptr
* @param[in] out_stride_x Stride of the destination image in X dimension (in bytes)
* @param[in] out_step_x out_stride_x * number of elements along X processed per workitem(in bytes)
* @param[in] out_stride_y Stride of the destination image in Y dimension (in bytes)
* @param[in] out_step_y out_stride_y * number of elements along Y processed per workitem(in bytes)
* @param[in] out_stride_z Stride of the destination image in Y dimension (in bytes)
* @param[in] out_step_z out_stride_z * number of elements along Y processed per workitem(in bytes)
* @param[in] out_offset_first_element_in_bytes The offset of the first element in the destination image
* @param[in] scale Integer scaling factor. Supported data types: S32 (ignored for QS8 and QS16 as the assumption is scale = 1).
*/
__kernel void pixelwise_mul_int(
TENSOR3D_DECLARATION(in1),
TENSOR3D_DECLARATION(in2),
TENSOR3D_DECLARATION(out),
const uint scale)
{
// Get pixels pointer
Tensor3D in1 = CONVERT_TO_TENSOR3D_STRUCT(in1);
Tensor3D in2 = CONVERT_TO_TENSOR3D_STRUCT(in2);
Tensor3D out = CONVERT_TO_TENSOR3D_STRUCT(out);
// Load data
VEC_DATA_TYPE(DATA_TYPE_RES, 16)
in1_data = CONVERT(vload16(0, (__global DATA_TYPE_IN1 *)in1.ptr), VEC_DATA_TYPE(DATA_TYPE_RES, 16));
VEC_DATA_TYPE(DATA_TYPE_RES, 16)
in2_data = CONVERT(vload16(0, (__global DATA_TYPE_IN2 *)in2.ptr), VEC_DATA_TYPE(DATA_TYPE_RES, 16));
// Perform multiplication and store result
vstore16(MUL_OP(in1_data, in2_data, scale, DATA_TYPE_OUT, 16), 0, (__global DATA_TYPE_OUT *)out.ptr);
}
|
