xref: /aosp_15_r20/external/ComputeLibrary/src/core/CL/kernels/CLReorgLayerKernel.cpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2018-2021 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     _type = CLKernelType::ELEMENTWISE;
72 }
73 
configure(const ICLTensor * input,ICLTensor * output,int32_t stride)74 void CLReorgLayerKernel::configure(const ICLTensor *input, ICLTensor *output, int32_t stride)
75 {
76     configure(CLKernelLibrary::get().get_compile_context(), input, output, stride);
77 }
78 
configure(const CLCompileContext & compile_context,const ICLTensor * input,ICLTensor * output,int32_t stride)79 void CLReorgLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, int32_t stride)
80 {
81     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
82     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), stride));
83     auto padding_info = get_padding_info({ input, output });
84 
85     _input  = input;
86     _output = output;
87 
88     std::string  kernel_name = std::string("reorg_layer_") + lower_string(string_from_data_layout(input->info()->data_layout()));
89     const size_t idx_channel = get_data_layout_dimension_index(input->info()->data_layout(), DataLayoutDimension::CHANNEL);
90 
91     // Create kernel
92     CLBuildOptions build_opts;
93     build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
94     build_opts.add_option("-DSRC_DEPTH=" + support::cpp11::to_string(input->info()->dimension(idx_channel)));
95     build_opts.add_option("-DSTRIDE=" + support::cpp11::to_string(stride));
96     _kernel = create_kernel(compile_context, kernel_name, build_opts.options());
97 
98     // Configure window
99     // auto inizialize the output tensor if not yet initialized
100     auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(misc::shape_calculator::compute_reorg_output_shape(*input->info(), stride)));
101 
102     Window win = calculate_max_window(*output->info(), Steps());
103 
104     // The CLWeightsReshapeKernel doesn't need padding so update_window_and_padding() can be skipped
105     ICLKernel::configure_internal(win);
106 
107     _config_id = kernel_name;
108     _config_id += "_";
109     _config_id += string_from_data_type(input->info()->data_type());
110     _config_id += "_";
111     _config_id += support::cpp11::to_string(input->info()->dimension(0));
112     _config_id += "_";
113     _config_id += support::cpp11::to_string(input->info()->dimension(1));
114     _config_id += "_";
115     _config_id += support::cpp11::to_string(input->info()->dimension(2));
116     _config_id += "_";
117     _config_id += support::cpp11::to_string(stride);
118     ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
119 }
120 
validate(const arm_compute::ITensorInfo * input,const arm_compute::ITensorInfo * output,int32_t stride)121 Status CLReorgLayerKernel::validate(const arm_compute::ITensorInfo *input, const arm_compute::ITensorInfo *output, int32_t stride)
122 {
123     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, stride));
124 
125     return Status{};
126 }
127 
run(const Window & window,cl::CommandQueue & queue)128 void CLReorgLayerKernel::run(const Window &window, cl::CommandQueue &queue)
129 {
130     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
131     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
132 
133     Window slice = window.first_slice_window_3D();
134 
135     do
136     {
137         unsigned int idx = 0;
138         add_3D_tensor_argument(idx, _input, slice);
139         add_3D_tensor_argument(idx, _output, slice);
140         enqueue(queue, *this, slice, lws_hint());
141     }
142     while(window.slide_window_slice_3D(slice));
143 }
144 } // namespace arm_compute
145