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 "arm_compute/runtime/CL/functions/CLFastCorners.h"
25
26 #include "arm_compute/core/CL/OpenCL.h"
27 #include "arm_compute/core/Error.h"
28 #include "arm_compute/core/TensorInfo.h"
29 #include "arm_compute/core/Validate.h"
30 #include "arm_compute/runtime/CL/CLScheduler.h"
31 #include "arm_compute/runtime/ITensorAllocator.h"
32 #include "src/core/CL/kernels/CLFastCornersKernel.h"
33 #include "src/core/CL/kernels/CLFillBorderKernel.h"
34 #include "support/MemorySupport.h"
35
36 #include <algorithm>
37 #include <cstring>
38
39 using namespace arm_compute;
40
CLFastCorners(std::shared_ptr<IMemoryManager> memory_manager)41 CLFastCorners::CLFastCorners(std::shared_ptr<IMemoryManager> memory_manager)
42 : _memory_group(std::move(memory_manager)),
43 _fast_corners_kernel(support::cpp14::make_unique<CLFastCornersKernel>()),
44 _suppr_func(),
45 _copy_array_kernel(support::cpp14::make_unique<CLCopyToArrayKernel>()),
46 _output(),
47 _suppr(),
48 _win(),
49 _non_max(false),
50 _num_corners(nullptr),
51 _num_buffer(),
52 _corners(nullptr),
53 _constant_border_value(0)
54 {
55 }
56
57 CLFastCorners::~CLFastCorners() = default;
58
configure(const ICLImage * input,float threshold,bool nonmax_suppression,ICLKeyPointArray * corners,unsigned int * num_corners,BorderMode border_mode,uint8_t constant_border_value)59 void CLFastCorners::configure(const ICLImage *input, float threshold, bool nonmax_suppression, ICLKeyPointArray *corners,
60 unsigned int *num_corners, BorderMode border_mode, uint8_t constant_border_value)
61 {
62 configure(CLKernelLibrary::get().get_compile_context(), input, threshold, nonmax_suppression, corners, num_corners, border_mode, constant_border_value);
63 }
64
configure(const CLCompileContext & compile_context,const ICLImage * input,float threshold,bool nonmax_suppression,ICLKeyPointArray * corners,unsigned int * num_corners,BorderMode border_mode,uint8_t constant_border_value)65 void CLFastCorners::configure(const CLCompileContext &compile_context, const ICLImage *input, float threshold, bool nonmax_suppression, ICLKeyPointArray *corners,
66 unsigned int *num_corners, BorderMode border_mode, uint8_t constant_border_value)
67 {
68 ARM_COMPUTE_ERROR_ON_TENSOR_NOT_2D(input);
69 ARM_COMPUTE_ERROR_ON(BorderMode::UNDEFINED != border_mode);
70 ARM_COMPUTE_ERROR_ON(nullptr == corners);
71 ARM_COMPUTE_ERROR_ON(threshold < 1 && threshold > 255);
72
73 TensorInfo tensor_info(input->info()->tensor_shape(), 1, DataType::U8);
74 _output.allocator()->init(tensor_info);
75
76 _non_max = nonmax_suppression;
77 _num_corners = num_corners;
78 _corners = corners;
79 _num_buffer = cl::Buffer(CLScheduler::get().context(), CL_MEM_ALLOC_HOST_PTR | CL_MEM_READ_WRITE, sizeof(unsigned int));
80 _constant_border_value = constant_border_value;
81
82 const bool update_number = (nullptr != _num_corners);
83
84 _memory_group.manage(&_output);
85 _fast_corners_kernel->configure(compile_context, input, &_output, threshold, nonmax_suppression, border_mode);
86
87 if(!_non_max)
88 {
89 _copy_array_kernel->configure(compile_context, &_output, update_number, _corners, &_num_buffer);
90 }
91 else
92 {
93 _suppr.allocator()->init(tensor_info);
94 _memory_group.manage(&_suppr);
95
96 _suppr_func.configure(compile_context, &_output, &_suppr, border_mode);
97 _copy_array_kernel->configure(compile_context, &_suppr, update_number, _corners, &_num_buffer);
98
99 _suppr.allocator()->allocate();
100 }
101
102 // Allocate intermediate tensors
103 _output.allocator()->allocate();
104 }
105
run()106 void CLFastCorners::run()
107 {
108 cl::CommandQueue q = CLScheduler::get().queue();
109
110 MemoryGroupResourceScope scope_mg(_memory_group);
111
112 if(_non_max)
113 {
114 ARM_COMPUTE_ERROR_ON_MSG(_output.cl_buffer().get() == nullptr, "Unconfigured function");
115 const auto out_buffer = static_cast<unsigned char *>(q.enqueueMapBuffer(_output.cl_buffer(), CL_TRUE, CL_MAP_WRITE, 0, _output.info()->total_size()));
116 memset(out_buffer, 0, _output.info()->total_size());
117 q.enqueueUnmapMemObject(_output.cl_buffer(), out_buffer);
118 }
119
120 CLScheduler::get().enqueue(*_fast_corners_kernel, false);
121
122 if(_non_max)
123 {
124 _suppr_func.run();
125 }
126
127 CLScheduler::get().enqueue(*_copy_array_kernel, false);
128
129 unsigned int get_num_corners = 0;
130 q.enqueueReadBuffer(_num_buffer, CL_TRUE, 0, sizeof(unsigned int), &get_num_corners);
131
132 size_t corner_size = std::min(static_cast<size_t>(get_num_corners), _corners->max_num_values());
133
134 _corners->resize(corner_size);
135
136 if(_num_corners != nullptr)
137 {
138 *_num_corners = get_num_corners;
139 }
140
141 q.flush();
142 }
143