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/NEDilateKernel.h"
25
26 #include "arm_compute/core/Helpers.h"
27 #include "arm_compute/core/ITensor.h"
28 #include "arm_compute/core/TensorInfo.h"
29 #include "arm_compute/core/Validate.h"
30 #include "src/core/NEON/INEKernel.h"
31 #include "src/core/helpers/AutoConfiguration.h"
32 #include "src/core/helpers/WindowHelpers.h"
33
34 #include <arm_neon.h>
35
36 namespace arm_compute
37 {
border_size() const38 BorderSize NEDilateKernel::border_size() const
39 {
40 return BorderSize(1);
41 }
42
configure(const ITensor * input,ITensor * output,bool border_undefined)43 void NEDilateKernel::configure(const ITensor *input, ITensor *output, bool border_undefined)
44 {
45 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
46 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8);
47 ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
48
49 _input = input;
50 _output = output;
51
52 constexpr unsigned int num_elems_processed_per_iteration = 8;
53 constexpr unsigned int num_elems_read_per_iteration = 16;
54 constexpr unsigned int num_elems_written_per_iteration = 8;
55 constexpr unsigned int num_rows_read_per_iteration = 3;
56
57 // Configure kernel window
58 Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration), border_undefined, border_size());
59 AccessWindowHorizontal output_access(output->info(), 0, num_elems_written_per_iteration);
60 AccessWindowRectangle input_access(input->info(), -border_size().left, -border_size().top, num_elems_read_per_iteration, num_rows_read_per_iteration);
61
62 update_window_and_padding(win, input_access, output_access);
63
64 output_access.set_valid_region(win, input->info()->valid_region(), border_undefined, border_size());
65
66 INEKernel::configure(win);
67 }
68
run(const Window & window,const ThreadInfo & info)69 void NEDilateKernel::run(const Window &window, const ThreadInfo &info)
70 {
71 ARM_COMPUTE_UNUSED(info);
72 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
73 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INESimpleKernel::window(), window);
74
75 Iterator in(_input, window);
76 Iterator out(_output, window);
77
78 const size_t in_stride = _input->info()->strides_in_bytes()[1];
79
80 execute_window_loop(window, [&](const Coordinates &)
81 {
82 uint8_t *in_ptr = in.ptr() - 1;
83 const uint8x16_t top_data = vld1q_u8(in_ptr - in_stride);
84 const uint8x16_t mid_data = vld1q_u8(in_ptr);
85 const uint8x16_t bot_data = vld1q_u8(in_ptr + in_stride);
86
87 uint8x8_t top_high_data = vget_high_u8(top_data);
88 uint8x8_t top_low_data = vget_low_u8(top_data);
89
90 uint8x8_t mid_high_data = vget_high_u8(mid_data);
91 uint8x8_t mid_low_data = vget_low_u8(mid_data);
92
93 uint8x8_t bot_high_data = vget_high_u8(bot_data);
94 uint8x8_t bot_low_data = vget_low_u8(bot_data);
95
96 uint8x8_t p0;
97 uint8x8_t p1;
98
99 p0 = top_low_data;
100 p1 = vext_u8(top_low_data, top_high_data, 1);
101 p0 = vmax_u8(p0, p1);
102
103 p1 = vext_u8(top_low_data, top_high_data, 2);
104 p0 = vmax_u8(p0, p1);
105
106 p1 = mid_low_data;
107 p0 = vmax_u8(p0, p1);
108
109 p1 = vext_u8(mid_low_data, mid_high_data, 1);
110 p0 = vmax_u8(p0, p1);
111
112 p1 = vext_u8(mid_low_data, mid_high_data, 2);
113 p0 = vmax_u8(p0, p1);
114
115 p1 = bot_low_data;
116 p0 = vmax_u8(p0, p1);
117
118 p1 = vext_u8(bot_low_data, bot_high_data, 1);
119 p0 = vmax_u8(p0, p1);
120
121 p1 = vext_u8(bot_low_data, bot_high_data, 2);
122 p0 = vmax_u8(p0, p1);
123
124 vst1_u8(out.ptr(), p0);
125 },
126 in, out);
127 }
128 } // namespace arm_compute
129