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
|
/*
* Copyright (c) 2018-2020 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 "src/core/NEON/kernels/NERangeKernel.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/TensorInfo.h"
#include "arm_compute/core/Validate.h"
#include "src/core/NEON/NEAsymm.h"
#include "src/core/NEON/wrapper/wrapper.h"
#include "src/core/helpers/AutoConfiguration.h"
#include "src/core/helpers/WindowHelpers.h"
#include "arm_compute/core/Utils.h"
namespace arm_compute
{
namespace
{
template <typename T>
void range_function(ITensor *output, float start, float step, const Window &window)
{
/** NEON vector tag type. */
using ExactTagType = typename wrapper::traits::neon_bitvector<T, wrapper::traits::BitWidth::W128>::tag_type;
const auto step_vec = wrapper::vdup_n(static_cast<T>(step), ExactTagType{});
const auto start_vec = wrapper::vdup_n(static_cast<T>(start), ExactTagType{});
auto id_vec = wrapper::vdup_n(static_cast<T>(0.f), ExactTagType{});
const auto window_start_x = static_cast<int>(window.x().start());
const auto window_end_x = static_cast<int>(window.x().end());
const int window_step_x = 16 / sizeof(T);
Window win{ window };
win.set(Window::DimX, Window::Dimension(0, 1, 1));
Iterator output_it(output, win);
execute_window_loop(win, [&](const Coordinates &)
{
int x = window_start_x;
const auto out_ptr = reinterpret_cast<T *>(output_it.ptr());
for(; x <= (window_end_x - window_step_x); x += window_step_x)
{
for(int count = 0; count < window_step_x; ++count)
{
id_vec = wrapper::vsetlane(static_cast<T>(x + count), id_vec, count);
}
// start + step * id
const auto res_vec = wrapper::vmla(start_vec, id_vec, step_vec);
wrapper::vstore(out_ptr + x, res_vec);
}
// Compute left-over elements
for(; x < window_end_x; ++x)
{
const auto res = start + x * step;
*(out_ptr + x) = res;
}
},
output_it);
}
Status validate_arguments(const ITensorInfo &output, const float start, const float end, const float step)
{
ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&output,
1,
DataType::U8, DataType::S8,
DataType::U16, DataType::S16,
DataType::U32, DataType::S32,
DataType::F16, DataType::F32);
ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(((start < end) && (step <= 0)), "step must be greater than 0 when start < end");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(((start > end) && (step >= 0)), "step must be less than 0 when start > end");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(start, output.data_type(), output.quantization_info()), "start value is outside the range of the data type");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(end, output.data_type(), output.quantization_info()), "end value is outside the range of the data type");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(!check_value_range(step, output.data_type(), output.quantization_info()), "step value is outside the range of the data type");
ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(output.num_dimensions() != 1, "Output has to be a 1-D tensor");
ARM_COMPUTE_RETURN_ERROR_ON_MSG(output.tensor_shape().total_size() < num_of_elements_in_range(start, end, step), "Output tensor size is incorrect");
return Status{};
}
} // namespace
NERangeKernel::NERangeKernel()
: _func(nullptr), _start(0), _end(1), _step(1), _output(nullptr)
{
}
void NERangeKernel::configure(ITensor *output, float start, float end, float step)
{
ARM_COMPUTE_ERROR_ON_NULLPTR(output);
ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*(output->info()), start, end, step));
// Auto initialize output if not initialized
auto_init_if_empty(*output->info(), TensorShape(num_of_elements_in_range(start, end, step)), 1, output->info()->data_type(), output->info()->quantization_info());
// Configure kernel window
Window win = calculate_max_window(*output->info(), Steps());
Coordinates coord;
coord.set_num_dimensions(output->info()->num_dimensions());
output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
_start = start;
_end = end;
_step = step;
_output = output;
switch(_output->info()->data_type())
{
case DataType::U8:
_func = &range_function<uint8_t>;
break;
case DataType::U16:
_func = &range_function<uint16_t>;
break;
case DataType::U32:
_func = &range_function<uint32_t>;
break;
case DataType::S8:
_func = &range_function<int8_t>;
break;
case DataType::S16:
_func = &range_function<int16_t>;
break;
case DataType::S32:
_func = &range_function<int32_t>;
break;
case DataType::F32:
_func = &range_function<float>;
break;
#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
case DataType::F16:
_func = &range_function<float16_t>;
break;
#endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
default:
ARM_COMPUTE_ERROR("Unsupported data type.");
break;
}
INEKernel::configure(win);
}
Status NERangeKernel::validate(const ITensorInfo *output, float start, float end, float step)
{
ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*output, start, end, step));
return Status{};
}
void NERangeKernel::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(_func == nullptr);
(*_func)(_output, _start, _step, window);
}
} // namespace arm_compute
|
