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/CL/kernels/CLDerivativeKernel.h"
25
26 #include "arm_compute/core/CL/CLKernelLibrary.h"
27 #include "arm_compute/core/CL/ICLTensor.h"
28 #include "arm_compute/core/CL/OpenCL.h"
29 #include "arm_compute/core/Helpers.h"
30 #include "arm_compute/core/Validate.h"
31 #include "src/core/helpers/WindowHelpers.h"
32 #include "support/StringSupport.h"
33
34 #include <set>
35 #include <string>
36
37 using namespace arm_compute;
38
CLDerivativeKernel()39 CLDerivativeKernel::CLDerivativeKernel()
40 : _input(nullptr), _output_x(nullptr), _output_y(nullptr), _run_derivative_x(false), _run_derivative_y(false)
41 {
42 }
43
border_size() const44 BorderSize CLDerivativeKernel::border_size() const
45 {
46 return BorderSize(1);
47 }
48
configure(const ICLTensor * input,ICLTensor * output_x,ICLTensor * output_y,bool border_undefined)49 void CLDerivativeKernel::configure(const ICLTensor *input, ICLTensor *output_x, ICLTensor *output_y, bool border_undefined)
50 {
51 configure(CLKernelLibrary::get().get_compile_context(), input, output_x, output_y, border_undefined);
52 }
53
configure(const CLCompileContext & compile_context,const ICLTensor * input,ICLTensor * output_x,ICLTensor * output_y,bool border_undefined)54 void CLDerivativeKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output_x, ICLTensor *output_y, bool border_undefined)
55 {
56 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8);
57 ARM_COMPUTE_ERROR_ON((output_x == nullptr) && (output_y == nullptr));
58
59 _run_derivative_x = output_x != nullptr;
60 _run_derivative_y = output_y != nullptr;
61
62 if(_run_derivative_x)
63 {
64 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output_x, 1, DataType::S16);
65 }
66
67 if(_run_derivative_y)
68 {
69 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output_y, 1, DataType::S16);
70 }
71
72 _input = input;
73 _output_x = output_x;
74 _output_y = output_y;
75
76 // Set build options
77 std::set<std::string> build_opts;
78
79 if(_run_derivative_x)
80 {
81 build_opts.insert("-DGRAD_X");
82 }
83
84 if(_run_derivative_y)
85 {
86 build_opts.insert("-DGRAD_Y");
87 }
88
89 // Create kernel
90 const std::string kernel_name = std::string("derivative");
91 _kernel = create_kernel(compile_context, kernel_name, build_opts);
92
93 // Configure kernel window
94 constexpr unsigned int num_elems_processed_per_iteration = 16;
95 constexpr unsigned int num_read_rows_per_iteration = 3;
96
97 Window win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration), border_undefined, border_size());
98
99 AccessWindowRectangle input_access(input->info(), 0, 0, 0, 0);
100 AccessWindowHorizontal output_x_access(output_x == nullptr ? nullptr : output_x->info(), 0, num_elems_processed_per_iteration);
101 AccessWindowHorizontal output_y_access(output_y == nullptr ? nullptr : output_y->info(), 0, num_elems_processed_per_iteration);
102 if(_run_derivative_x && _run_derivative_y)
103 {
104 // TODO(COMPMID-415) Fix x-access input bug in CL kernel instead of '+2'
105 input_access = AccessWindowRectangle(input->info(), -border_size().left, -border_size().top, num_elems_processed_per_iteration + 2, num_read_rows_per_iteration);
106 }
107 else if(_run_derivative_x)
108 {
109 // TODO(COMPMID-415) Fix x-access input bug in CL kernel instead of '+2'
110 input_access = AccessWindowHorizontal(input->info(), -border_size().left, num_elems_processed_per_iteration + 2);
111 }
112 else if(_run_derivative_y)
113 {
114 input_access = AccessWindowRectangle(input->info(), 0, -border_size().top, num_elems_processed_per_iteration, num_read_rows_per_iteration);
115 }
116
117 update_window_and_padding(win,
118 input_access,
119 output_x_access,
120 output_y_access);
121
122 output_x_access.set_valid_region(win, input->info()->valid_region(), border_undefined, border_size());
123 output_y_access.set_valid_region(win, input->info()->valid_region(), border_undefined, border_size());
124
125 ICLKernel::configure_internal(win);
126
127 // Set config_id for enabling LWS tuning
128 _config_id = kernel_name;
129 _config_id += "_";
130 _config_id += lower_string(string_from_data_type(input->info()->data_type()));
131 _config_id += "_";
132 _config_id += support::cpp11::to_string(input->info()->dimension(0));
133 _config_id += "_";
134 _config_id += support::cpp11::to_string(input->info()->dimension(1));
135 _config_id += "_";
136 _config_id += support::cpp11::to_string(border_undefined);
137 }
138
run(const Window & window,cl::CommandQueue & queue)139 void CLDerivativeKernel::run(const Window &window, cl::CommandQueue &queue)
140 {
141 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
142 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
143
144 Window slice = window.first_slice_window_2D();
145 do
146 {
147 unsigned int idx = 0;
148 add_2D_tensor_argument(idx, _input, slice);
149 add_2D_tensor_argument_if((_run_derivative_x), idx, _output_x, slice);
150 add_2D_tensor_argument_if((_run_derivative_y), idx, _output_y, slice);
151
152 enqueue(queue, *this, slice, lws_hint());
153 }
154 while(window.slide_window_slice_2D(slice));
155 }
156