1 /*
2 * Copyright (c) 2016-2020 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/NEMedian3x3Kernel.h"
25
26 #include "arm_compute/core/Coordinates.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/Validate.h"
31 #include "src/core/NEON/INEKernel.h"
32 #include "src/core/helpers/AutoConfiguration.h"
33 #include "src/core/helpers/WindowHelpers.h"
34
35 #include <arm_neon.h>
36 #include <utility>
37
38 using namespace arm_compute;
39
40 namespace
41 {
sort(uint8x8_t & a,uint8x8_t & b)42 inline void sort(uint8x8_t &a, uint8x8_t &b)
43 {
44 const uint8x8_t min = vmin_u8(a, b);
45 const uint8x8_t max = vmax_u8(a, b);
46 a = min;
47 b = max;
48 }
49 } // namespace
50
border_size() const51 BorderSize NEMedian3x3Kernel::border_size() const
52 {
53 return BorderSize(1);
54 }
55
configure(const ITensor * input,ITensor * output,bool border_undefined)56 void NEMedian3x3Kernel::configure(const ITensor *input, ITensor *output, bool border_undefined)
57 {
58 _input = input;
59 _output = output;
60
61 // Configure kernel window
62 constexpr unsigned int num_elems_processed_per_iteration = 8;
63 constexpr unsigned int num_elems_read_per_iteration = 16;
64 constexpr unsigned int num_elems_written_per_iteration = 8;
65 constexpr unsigned int num_rows_read_per_iteration = 3;
66
67 Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration), border_undefined, border_size());
68 AccessWindowHorizontal output_access(output->info(), 0, num_elems_written_per_iteration);
69
70 update_window_and_padding(win,
71 AccessWindowRectangle(input->info(), -border_size().left, -border_size().top, num_elems_read_per_iteration, num_rows_read_per_iteration),
72 output_access);
73
74 output_access.set_valid_region(win, input->info()->valid_region(), border_undefined, border_size());
75
76 INEKernel::configure(win);
77 }
78
run(const Window & window,const ThreadInfo & info)79 void NEMedian3x3Kernel::run(const Window &window, const ThreadInfo &info)
80 {
81 ARM_COMPUTE_UNUSED(info);
82 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
83 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INESimpleKernel::window(), window);
84
85 const unsigned char *input_bot_ptr = _input->ptr_to_element(Coordinates(-1, -1));
86 const unsigned char *input_mid_ptr = _input->ptr_to_element(Coordinates(-1, 0));
87 const unsigned char *input_top_ptr = _input->ptr_to_element(Coordinates(-1, +1));
88
89 Iterator input(_input, window);
90 Iterator output(_output, window);
91
92 execute_window_loop(window, [&](const Coordinates &)
93 {
94 const uint8x16_t top_data = vld1q_u8(input_top_ptr + input.offset());
95 const uint8x16_t mid_data = vld1q_u8(input_mid_ptr + input.offset());
96 const uint8x16_t bot_data = vld1q_u8(input_bot_ptr + input.offset());
97
98 uint8x8_t p0 = vget_low_u8(top_data);
99 uint8x8_t p1 = vext_u8(vget_low_u8(top_data), vget_high_u8(top_data), 1);
100 uint8x8_t p2 = vext_u8(vget_low_u8(top_data), vget_high_u8(top_data), 2);
101 uint8x8_t p3 = vget_low_u8(mid_data);
102 uint8x8_t p4 = vext_u8(vget_low_u8(mid_data), vget_high_u8(mid_data), 1);
103 uint8x8_t p5 = vext_u8(vget_low_u8(mid_data), vget_high_u8(mid_data), 2);
104 uint8x8_t p6 = vget_low_u8(bot_data);
105 uint8x8_t p7 = vext_u8(vget_low_u8(bot_data), vget_high_u8(bot_data), 1);
106 uint8x8_t p8 = vext_u8(vget_low_u8(bot_data), vget_high_u8(bot_data), 2);
107
108 sort(p1, p2);
109 sort(p4, p5);
110 sort(p7, p8);
111
112 sort(p0, p1);
113 sort(p3, p4);
114 sort(p6, p7);
115
116 sort(p1, p2);
117 sort(p4, p5);
118 sort(p7, p8);
119
120 sort(p0, p3);
121 sort(p5, p8);
122 sort(p4, p7);
123
124 sort(p3, p6);
125 sort(p1, p4);
126 sort(p2, p5);
127
128 sort(p4, p7);
129 sort(p4, p2);
130 sort(p6, p4);
131
132 sort(p4, p2);
133
134 vst1_u8(output.ptr(), p4);
135 },
136 input, output);
137 }
138