1 /*
2 * Copyright (c) 2016-2021 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
25 #include "arm_compute/runtime/NEON/NEFunctions.h"
26
27 #include "arm_compute/core/Types.h"
28 #include "utils/Utils.h"
29
30 #include <cstring>
31 #include <iostream>
32
33 using namespace arm_compute;
34 using namespace utils;
35
36 class NEONCopyObjectsExample : public Example
37 {
38 public:
do_setup(int argc,char ** argv)39 bool do_setup(int argc, char **argv) override
40 {
41 ARM_COMPUTE_UNUSED(argc);
42 ARM_COMPUTE_UNUSED(argv);
43
44 /** [Copy objects example] */
45 constexpr unsigned int width = 4;
46 constexpr unsigned int height = 3;
47 constexpr unsigned int batch = 2;
48
49 src_data = new float[width * height * batch];
50 dst_data = new float[width * height * batch];
51
52 // Fill src_data with pseudo(meaningless) values:
53 for(unsigned int b = 0; b < batch; b++)
54 {
55 for(unsigned int h = 0; h < height; h++)
56 {
57 for(unsigned int w = 0; w < width; w++)
58 {
59 src_data[b * (width * height) + h * width + w] = static_cast<float>(100 * b + 10 * h + w);
60 }
61 }
62 }
63
64 // Initialize the tensors dimensions and type:
65 const TensorShape shape(width, height, batch);
66 input.allocator()->init(TensorInfo(shape, 1, DataType::F32));
67 output.allocator()->init(TensorInfo(shape, 1, DataType::F32));
68
69 // Configure softmax:
70 softmax.configure(&input, &output);
71
72 // Allocate the input / output tensors:
73 input.allocator()->allocate();
74 output.allocator()->allocate();
75
76 // Fill the input tensor:
77 // Simplest way: create an iterator to iterate through each element of the input tensor:
78 Window input_window;
79 input_window.use_tensor_dimensions(input.info()->tensor_shape());
80 std::cout << " Dimensions of the input's iterator:\n";
81 std::cout << " X = [start=" << input_window.x().start() << ", end=" << input_window.x().end() << ", step=" << input_window.x().step() << "]\n";
82 std::cout << " Y = [start=" << input_window.y().start() << ", end=" << input_window.y().end() << ", step=" << input_window.y().step() << "]\n";
83 std::cout << " Z = [start=" << input_window.z().start() << ", end=" << input_window.z().end() << ", step=" << input_window.z().step() << "]\n";
84
85 // Create an iterator:
86 Iterator input_it(&input, input_window);
87
88 // Iterate through the elements of src_data and copy them one by one to the input tensor:
89 // This is equivalent to:
90 // for( unsigned int z = 0; z < batch; ++z)
91 // {
92 // for( unsigned int y = 0; y < height; ++y)
93 // {
94 // for( unsigned int x = 0; x < width; ++x)
95 // {
96 // *reinterpret_cast<float*>( input.buffer() + input.info()->offset_element_in_bytes(Coordinates(x,y,z))) = src_data[ z * (width*height) + y * width + x];
97 // }
98 // }
99 // }
100 // Except it works for an arbitrary number of dimensions
101 execute_window_loop(input_window, [&](const Coordinates & id)
102 {
103 std::cout << "Setting item [" << id.x() << "," << id.y() << "," << id.z() << "]\n";
104 *reinterpret_cast<float *>(input_it.ptr()) = src_data[id.z() * (width * height) + id.y() * width + id.x()];
105 },
106 input_it);
107
108 // More efficient way: create an iterator to iterate through each row (instead of each element) of the output tensor:
109 Window output_window;
110 output_window.use_tensor_dimensions(output.info()->tensor_shape(), /* first_dimension =*/Window::DimY); // Iterate through the rows (not each element)
111 std::cout << " Dimensions of the output's iterator:\n";
112 std::cout << " X = [start=" << output_window.x().start() << ", end=" << output_window.x().end() << ", step=" << output_window.x().step() << "]\n";
113 std::cout << " Y = [start=" << output_window.y().start() << ", end=" << output_window.y().end() << ", step=" << output_window.y().step() << "]\n";
114 std::cout << " Z = [start=" << output_window.z().start() << ", end=" << output_window.z().end() << ", step=" << output_window.z().step() << "]\n";
115
116 // Create an iterator:
117 Iterator output_it(&output, output_window);
118
119 // Iterate through the rows of the output tensor and copy them to dst_data:
120 // This is equivalent to:
121 // for( unsigned int z = 0; z < batch; ++z)
122 // {
123 // for( unsigned int y = 0; y < height; ++y)
124 // {
125 // memcpy( dst_data + z * (width*height) + y * width, input.buffer() + input.info()->offset_element_in_bytes(Coordinates(0,y,z)), width * sizeof(float));
126 // }
127 // }
128 // Except it works for an arbitrary number of dimensions
129 execute_window_loop(output_window, [&](const Coordinates & id)
130 {
131 std::cout << "Copying one row starting from [" << id.x() << "," << id.y() << "," << id.z() << "]\n";
132 // Copy one whole row:
133 memcpy(dst_data + id.z() * (width * height) + id.y() * width, output_it.ptr(), width * sizeof(float));
134 },
135 output_it);
136
137 /** [Copy objects example] */
138
139 return true;
140 }
do_run()141 void do_run() override
142 {
143 // Run softmax:
144 softmax.run();
145 }
do_teardown()146 void do_teardown() override
147 {
148 delete[] src_data;
149 delete[] dst_data;
150 }
151
152 private:
153 Tensor input{}, output{};
154 float *src_data{};
155 float *dst_data{};
156 NESoftmaxLayer softmax{};
157 };
158 /** Main program for the copy objects test
159 *
160 * @param[in] argc Number of arguments
161 * @param[in] argv Arguments
162 */
main(int argc,char ** argv)163 int main(int argc, char **argv)
164 {
165 return utils::run_example<NEONCopyObjectsExample>(argc, argv);
166 }
167