1 /*
2 * Copyright (c) 2018-2020 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/CLReorgLayerKernel.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 "arm_compute/core/utils/misc/ShapeCalculator.h"
33 #include "src/core/helpers/AutoConfiguration.h"
34 #include "src/core/helpers/WindowHelpers.h"
35 #include "support/StringSupport.h"
36
37 #include <string>
38
39 namespace arm_compute
40 {
41 namespace
42 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * output,int32_t stride)43 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, int32_t stride)
44 {
45 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
46 ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
47 ARM_COMPUTE_RETURN_ERROR_ON(input->data_layout() == DataLayout::UNKNOWN);
48
49 const size_t idx_width = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::WIDTH);
50 const size_t idx_height = get_data_layout_dimension_index(input->data_layout(), DataLayoutDimension::HEIGHT);
51
52 ARM_COMPUTE_RETURN_ERROR_ON(stride <= 0);
53 ARM_COMPUTE_RETURN_ERROR_ON_MSG((input->tensor_shape()[idx_width] % stride) != 0, "The width of the input tensor must be a multiple of stride");
54 ARM_COMPUTE_RETURN_ERROR_ON_MSG((input->tensor_shape()[idx_height] % stride) != 0, "The height of the input tensor must be a multiple of stride");
55
56 // Validate output if initialized
57 if(output->total_size() != 0)
58 {
59 const TensorInfo tensor_info_output = output->clone()->set_tensor_shape(misc::shape_calculator::compute_reorg_output_shape(*input, stride));
60 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output, &tensor_info_output);
61 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
62 }
63
64 return Status{};
65 }
66 } // namespace
67
CLReorgLayerKernel()68 CLReorgLayerKernel::CLReorgLayerKernel()
69 : _input(nullptr), _output(nullptr)
70 {
71 }
72
configure(const ICLTensor * input,ICLTensor * output,int32_t stride)73 void CLReorgLayerKernel::configure(const ICLTensor *input, ICLTensor *output, int32_t stride)
74 {
75 configure(CLKernelLibrary::get().get_compile_context(), input, output, stride);
76 }
77
configure(const CLCompileContext & compile_context,const ICLTensor * input,ICLTensor * output,int32_t stride)78 void CLReorgLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t stride)
79 {
80 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
81 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), stride));
82
83 _input = input;
84 _output = output;
85
86 std::string kernel_name = std::string("reorg_layer_") + lower_string(string_from_data_layout(input->info()->data_layout()));
87 const size_t idx_channel = get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::CHANNEL);
88
89 // Create kernel
90 CLBuildOptions build_opts;
91 build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
92 build_opts.add_option("-DSRC_DEPTH=" + support::cpp11::to_string(input->info()->dimension(idx_channel)));
93 build_opts.add_option("-DSTRIDE=" + support::cpp11::to_string(stride));
94 _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
95
96 // Configure window
97 // auto inizialize the output tensor if not yet initialized
98 auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(misc::shape_calculator::compute_reorg_output_shape(*input->info(), stride)));
99
100 Window win = calculate_max_window(*output->info(), Steps());
101
102 // The CLWeightsReshapeKernel doesn't need padding so update_window_and_padding() can be skipped
103 output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
104 ICLKernel::configure_internal(win);
105
106 _config_id = kernel_name;
107 _config_id += "_";
108 _config_id += string_from_data_type(input->info()->data_type());
109 _config_id += "_";
110 _config_id += support::cpp11::to_string(input->info()->dimension(0));
111 _config_id += "_";
112 _config_id += support::cpp11::to_string(input->info()->dimension(1));
113 _config_id += "_";
114 _config_id += support::cpp11::to_string(input->info()->dimension(2));
115 _config_id += "_";
116 _config_id += support::cpp11::to_string(stride);
117 }
118
validate(const arm_compute::ITensorInfo * input,const arm_compute::ITensorInfo * output,int32_t stride)119 Status CLReorgLayerKernel::validate(const arm_compute::ITensorInfo *input, const arm_compute::ITensorInfo *output, int32_t stride)
120 {
121 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, stride));
122
123 return Status{};
124 }
125
run(const Window & window,cl::CommandQueue & queue)126 void CLReorgLayerKernel::run(const Window &window, cl::CommandQueue &queue)
127 {
128 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
129 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
130
131 Window slice = window.first_slice_window_3D();
132
133 do
134 {
135 unsigned int idx = 0;
136 add_3D_tensor_argument(idx, _input, slice);
137 add_3D_tensor_argument(idx, _output, slice);
138 enqueue(queue, *this, slice, lws_hint());
139 }
140 while(window.slide_window_slice_3D(slice));
141 }
142 } // namespace arm_compute
143