1 /*
2 * Copyright (c) 2017-2020 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/CLRemapKernel.h"
25
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/CLKernelLibrary.h"
28 #include "arm_compute/core/CL/ICLTensor.h"
29 #include "arm_compute/core/Helpers.h"
30 #include "arm_compute/core/TensorInfo.h"
31 #include "arm_compute/core/Validate.h"
32 #include "src/core/AccessWindowStatic.h"
33 #include "src/core/helpers/WindowHelpers.h"
34
35 #include <algorithm>
36
37 using namespace arm_compute;
38
CLRemapKernel()39 CLRemapKernel::CLRemapKernel()
40 : _input(nullptr), _output(nullptr), _map_x(nullptr), _map_y(nullptr)
41 {
42 }
43
border_size() const44 BorderSize CLRemapKernel::border_size() const
45 {
46 return BorderSize(1);
47 }
48
configure(const ICLTensor * input,const ICLTensor * map_x,const ICLTensor * map_y,ICLTensor * output,InterpolationPolicy policy,bool border_undefined)49 void CLRemapKernel::configure(const ICLTensor *input, const ICLTensor *map_x, const ICLTensor *map_y, ICLTensor *output, InterpolationPolicy policy, bool border_undefined)
50 {
51 configure(CLKernelLibrary::get().get_compile_context(), input, map_x, map_y, output, policy, border_undefined);
52 }
53
configure(const CLCompileContext & compile_context,const ICLTensor * input,const ICLTensor * map_x,const ICLTensor * map_y,ICLTensor * output,InterpolationPolicy policy,bool border_undefined)54 void CLRemapKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *map_x, const ICLTensor *map_y, ICLTensor *output, InterpolationPolicy policy,
55 bool border_undefined)
56 {
57 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::U8);
58 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::U8);
59 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(map_x, 1, DataType::F32);
60 ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(map_y, 1, DataType::F32);
61 ARM_COMPUTE_ERROR_ON_MSG(policy == InterpolationPolicy::AREA, "Area interpolation is not supported!");
62 ARM_COMPUTE_UNUSED(border_undefined);
63
64 _input = input;
65 _output = output;
66 _map_x = map_x;
67 _map_y = map_y;
68
69 // Create kernel
70 std::set<std::string> build_opts = { ("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type())) };
71 std::string interpolation_name = string_from_interpolation_policy(policy);
72 std::transform(interpolation_name.begin(), interpolation_name.end(), interpolation_name.begin(), ::tolower);
73 std::string kernel_name = "remap_" + interpolation_name;
74 _kernel = create_kernel(compile_context, kernel_name, build_opts);
75
76 // Configure window
77 constexpr unsigned int num_elems_processed_per_iteration = 4;
78
79 const int total_right = ceil_to_multiple(input->info()->dimension(0), num_elems_processed_per_iteration);
80 const int access_right = total_right + (((total_right - input->info()->dimension(0)) == 0) ? border_size().right : 0);
81
82 Window win = calculate_max_window(*_output->info(), Steps(num_elems_processed_per_iteration));
83 AccessWindowStatic input_access(input->info(), -border_size().left, -border_size().top, access_right, input->info()->dimension(1) + border_size().bottom);
84
85 AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration);
86
87 update_window_and_padding(win, input_access, output_access);
88
89 output_access.set_valid_region(win, ValidRegion(Coordinates(), output->info()->tensor_shape()));
90
91 ICLKernel::configure_internal(win);
92
93 // Set static arguments
94 unsigned int idx = 4 * num_arguments_per_2D_tensor(); //Skip the input and output parameters
95 _kernel.setArg<cl_float>(idx++, input->info()->dimension(0));
96 _kernel.setArg<cl_float>(idx++, input->info()->dimension(1));
97 }
98
run(const Window & window,cl::CommandQueue & queue)99 void CLRemapKernel::run(const Window &window, cl::CommandQueue &queue)
100 {
101 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
102 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
103
104 Window slice = window.first_slice_window_2D();
105
106 do
107 {
108 unsigned int idx = 0;
109 add_2D_tensor_argument(idx, _input, slice);
110 add_2D_tensor_argument(idx, _output, slice);
111 add_2D_tensor_argument(idx, _map_x, slice);
112 add_2D_tensor_argument(idx, _map_y, slice);
113 enqueue(queue, *this, slice, lws_hint());
114 }
115 while(window.slide_window_slice_2D(slice));
116 }
117