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
|
/*
* Copyright (c) 2019 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright (c) 2018-2019 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 "arm_compute/core/NEON/kernels/NEElementwiseUnaryKernelEx.h"
#include "arm_compute/core/CPP/Validate.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/IAccessWindow.h"
#include "arm_compute/core/ITensor.h"
#include "arm_compute/core/NEON/NEAsymm.h"
#include "arm_compute/core/NEON/NEFixedPoint.h"
#include "arm_compute/core/NEON/wrapper/wrapper.h"
#include "arm_compute/core/TensorInfo.h"
#include "arm_compute/core/Validate.h"
#include <algorithm>
#include <arm_neon.h>
#include <cstdint>
#include <map>
#include <string>
namespace arm_compute
{
class Coordinates;
namespace
{
template <ElementWiseUnaryEx op, typename ScalarType>
inline ScalarType elementwise_op_scalar(const ScalarType &a)
{
switch (op)
{
case ElementWiseUnaryEx::NEG:
return -a;
default:
ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
}
}
template <ElementWiseUnaryEx op, typename VectorType>
inline VectorType elementwise_op(const VectorType &a)
{
switch (op)
{
case ElementWiseUnaryEx::NEG:
return wrapper::vneg(a);
default:
ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
}
}
template <ElementWiseUnaryEx op, typename ScalarType>
void elementwise_op(const ITensor *in, ITensor *out, const Window &window)
{
const int window_step_x = 16 / sizeof(ScalarType);
const auto window_start_x = static_cast<int>(window.x().start());
const auto window_end_x = static_cast<int>(window.x().end());
Window win = window;
win.set(Window::DimX, Window::Dimension(0, 1, 1));
Iterator input(in, win);
Iterator output(out, win);
execute_window_loop(win,
[&](const Coordinates &) {
auto output_ptr = reinterpret_cast<ScalarType *>(output.ptr());
const auto input_ptr = reinterpret_cast<const ScalarType *>(input.ptr());
int x = window_start_x;
for (; x <= window_end_x - window_step_x; x += window_step_x)
{
wrapper::vstore(output_ptr + x,
elementwise_op<op>(wrapper::vloadq(input_ptr + x)));
}
for (; x < window_end_x; ++x)
{
*(output_ptr + x) = elementwise_op_scalar<op>(*(input_ptr + x));
}
},
input, output);
}
template <ElementWiseUnaryEx op>
std::function<void(const ITensor *input, ITensor *output, const Window &window)>
configure_func(const ITensor *input, ITensor *output)
{
std::string function_to_call("op_");
function_to_call += string_from_data_type(input->info()->data_type()) + "_";
function_to_call += string_from_data_type(output->info()->data_type());
static std::map<std::string, NEElementwiseUnaryKernelEx::ElementwiseUnaryFunction *>
map_function = {
{"op_F32_F32", &elementwise_op<op, float>}, {"op_S32_S32", &elementwise_op<op, int32_t>},
};
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
map_function["op_F16_F16"] = &elementwise_op<op, float16_t>;
#endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
auto it = map_function.find(function_to_call);
if (it != map_function.end())
{
auto func = it->second;
return [func](const ITensor *input, ITensor *output, const Window &window) {
func(input, output, window);
};
}
return nullptr;
}
} // namespace
NEElementwiseUnaryKernelEx::NEElementwiseUnaryKernelEx()
: _function(nullptr), _input(nullptr), _output(nullptr)
{
}
void NEElementwiseUnaryKernelEx::configure(ElementWiseUnaryEx op, const ITensor *input,
ITensor *output)
{
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*input->info(), *output->info()));
ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
// Configure kernel window
const std::pair<TensorShape, ValidRegion> broadcast_pair =
ITensorInfo::broadcast_shape_and_valid_region(*input->info());
const TensorShape &out_shape = broadcast_pair.first;
const ValidRegion &valid_region = broadcast_pair.second;
// Auto initialize output if not initialized
auto_init_if_empty(*output->info(), out_shape, 1, input->info()->data_type());
Window win = calculate_max_window(valid_region);
_input = input;
_output = output;
INEKernel::configure(win);
switch (op)
{
case ElementWiseUnaryEx::NEG:
_function = configure_func<ElementWiseUnaryEx::NEG>(input, output);
break;
default:
ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
}
}
Status NEElementwiseUnaryKernelEx::validate_arguments(const ITensorInfo &input,
const ITensorInfo &output)
{
ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(&input);
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input, 1, DataType::F16, DataType::F32,
DataType::S32);
// Validate in case of configured output
if (output.total_size() > 0)
{
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(&input, &output);
}
return Status{};
}
Status NEElementwiseUnaryKernelEx::validate(ElementWiseUnaryEx op, const ITensorInfo *input,
const ITensorInfo *output)
{
ARM_COMPUTE_UNUSED(op);
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*input, *output));
return Status{};
}
void NEElementwiseUnaryKernelEx::run(const Window &window, const ThreadInfo &info)
{
ARM_COMPUTE_UNUSED(info);
ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
ARM_COMPUTE_ERROR_ON(_function == nullptr);
_function(_input, _output, window);
}
} // namespace arm_compute
|
