1 /*
2 * Copyright (c) 2018-2020 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24 #include "src/core/NEON/kernels/NERangeKernel.h"
25
26 #include "arm_compute/core/Error.h"
27 #include "arm_compute/core/Helpers.h"
28 #include "arm_compute/core/IAccessWindow.h"
29 #include "arm_compute/core/ITensor.h"
30 #include "arm_compute/core/TensorInfo.h"
31 #include "arm_compute/core/Validate.h"
32 #include "src/core/NEON/NEAsymm.h"
33 #include "src/core/NEON/wrapper/wrapper.h"
34 #include "src/core/helpers/AutoConfiguration.h"
35 #include "src/core/helpers/WindowHelpers.h"
36
37 #include "arm_compute/core/Utils.h"
38
39 namespace arm_compute
40 {
41 namespace
42 {
43 template <typename T>
range_function(ITensor * output,float start,float step,const Window & window)44 void range_function(ITensor *output, float start, float step, const Window &window)
45 {
46 /** NEON vector tag type. */
47 using ExactTagType = typename wrapper::traits::neon_bitvector<T, wrapper::traits::BitWidth::W128>::tag_type;
48
49 const auto step_vec = wrapper::vdup_n(static_cast<T>(step), ExactTagType{});
50 const auto start_vec = wrapper::vdup_n(static_cast<T>(start), ExactTagType{});
51 auto id_vec = wrapper::vdup_n(static_cast<T>(0.f), ExactTagType{});
52
53 const auto window_start_x = static_cast<int>(window.x().start());
54 const auto window_end_x = static_cast<int>(window.x().end());
55 const int window_step_x = 16 / sizeof(T);
56
57 Window win{ window };
58 win.set(Window::DimX, Window::Dimension(0, 1, 1));
59 Iterator output_it(output, win);
60
61 execute_window_loop(win, [&](const Coordinates &)
62 {
63 int x = window_start_x;
64 const auto out_ptr = reinterpret_cast<T *>(output_it.ptr());
65 for(; x <= (window_end_x - window_step_x); x += window_step_x)
66 {
67 for(int count = 0; count < window_step_x; ++count)
68 {
69 id_vec = wrapper::vsetlane(static_cast<T>(x + count), id_vec, count);
70 }
71
72 // start + step * id
73 const auto res_vec = wrapper::vmla(start_vec, id_vec, step_vec);
74 wrapper::vstore(out_ptr + x, res_vec);
75 }
76
77 // Compute left-over elements
78 for(; x < window_end_x; ++x)
79 {
80 const auto res = start + x * step;
81 *(out_ptr + x) = res;
82 }
83
84 },
85 output_it);
86 }
87
validate_arguments(const ITensorInfo & output,const float start,const float end,const float step)88 Status validate_arguments(const ITensorInfo &output, const float start, const float end, const float step)
89 {
90 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&output,
91 1,
92 DataType::U8, DataType::S8,
93 DataType::U16, DataType::S16,
94 DataType::U32, DataType::S32,
95 DataType::F16, DataType::F32);
96
97 ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end");
98 ARM_COMPUTE_RETURN_ERROR_ON_MSG(((start < end) && (step <= 0)), "step must be greater than 0 when start < end");
99 ARM_COMPUTE_RETURN_ERROR_ON_MSG(((start > end) && (step >= 0)), "step must be less than 0 when start > end");
100
101 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");
102 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");
103 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");
104
105 ARM_COMPUTE_RETURN_ERROR_ON_MSG((start == end), "start of the requested sequence must not be equal to the end");
106
107 ARM_COMPUTE_RETURN_ERROR_ON_MSG(output.num_dimensions() != 1, "Output has to be a 1-D tensor");
108 ARM_COMPUTE_RETURN_ERROR_ON_MSG(output.tensor_shape().total_size() < num_of_elements_in_range(start, end, step), "Output tensor size is incorrect");
109
110 return Status{};
111 }
112 } // namespace
113
NERangeKernel()114 NERangeKernel::NERangeKernel()
115 : _func(nullptr), _start(0), _end(1), _step(1), _output(nullptr)
116 {
117 }
118
configure(ITensor * output,float start,float end,float step)119 void NERangeKernel::configure(ITensor *output, float start, float end, float step)
120 {
121 ARM_COMPUTE_ERROR_ON_NULLPTR(output);
122
123 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*(output->info()), start, end, step));
124
125 // Auto initialize output if not initialized
126 auto_init_if_empty(*output->info(), TensorShape(num_of_elements_in_range(start, end, step)), 1, output->info()->data_type(), output->info()->quantization_info());
127
128 // Configure kernel window
129 Window win = calculate_max_window(*output->info(), Steps());
130 Coordinates coord;
131 coord.set_num_dimensions(output->info()->num_dimensions());
132 output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
133
134 _start = start;
135 _end = end;
136 _step = step;
137 _output = output;
138 switch(_output->info()->data_type())
139 {
140 case DataType::U8:
141 _func = &range_function<uint8_t>;
142 break;
143 case DataType::U16:
144 _func = &range_function<uint16_t>;
145 break;
146 case DataType::U32:
147 _func = &range_function<uint32_t>;
148 break;
149 case DataType::S8:
150 _func = &range_function<int8_t>;
151 break;
152 case DataType::S16:
153 _func = &range_function<int16_t>;
154 break;
155 case DataType::S32:
156 _func = &range_function<int32_t>;
157 break;
158 case DataType::F32:
159 _func = &range_function<float>;
160 break;
161 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
162 case DataType::F16:
163 _func = &range_function<float16_t>;
164 break;
165 #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
166 default:
167 ARM_COMPUTE_ERROR("Unsupported data type.");
168 break;
169 }
170
171 INEKernel::configure(win);
172 }
173
validate(const ITensorInfo * output,float start,float end,float step)174 Status NERangeKernel::validate(const ITensorInfo *output, float start, float end, float step)
175 {
176 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(output);
177
178 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*output, start, end, step));
179
180 return Status{};
181 }
182
run(const Window & window,const ThreadInfo & info)183 void NERangeKernel::run(const Window &window, const ThreadInfo &info)
184 {
185 ARM_COMPUTE_UNUSED(info);
186 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
187 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
188 ARM_COMPUTE_ERROR_ON(_func == nullptr);
189
190 (*_func)(_output, _start, _step, window);
191 }
192 } // namespace arm_compute