• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/CLSpaceToBatchLayerKernel.h"
25 
26 #include "arm_compute/core/CL/CLHelpers.h"
27 #include "arm_compute/core/CL/ICLTensor.h"
28 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
29 #include "src/core/CL/CLValidate.h"
30 #include "src/core/helpers/AutoConfiguration.h"
31 #include "src/core/helpers/WindowHelpers.h"
32 #include "support/StringSupport.h"
33 
34 using namespace arm_compute::misc::shape_calculator;
35 
36 namespace arm_compute
37 {
38 namespace
39 {
validate_arguments(const ITensorInfo * input,const ITensorInfo * block_info,const ITensorInfo * paddings,const ITensorInfo * output)40 Status validate_arguments(const ITensorInfo *input, const ITensorInfo *block_info, const ITensorInfo *paddings, const ITensorInfo *output)
41 {
42     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, block_info, paddings, output);
43     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(block_info, 1, DataType::S32);
44     ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
45     ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
46     ARM_COMPUTE_RETURN_ERROR_ON(block_info->num_dimensions() > 1);
47     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(block_info->tensor_shape(), TensorShape{ 2 });
48     ARM_COMPUTE_RETURN_ERROR_ON(paddings->num_dimensions() > 2);
49     ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(paddings->tensor_shape(), TensorShape{ 2, 2 });
50 
51     // Validate output if initialized
52     if(output->total_size() != 0)
53     {
54         const DataLayout data_layout = input->data_layout();
55         const int        idx_channel = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
56         ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape()[idx_channel] != output->tensor_shape()[idx_channel]);
57         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
58         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
59     }
60 
61     return Status{};
62 }
validate_arguments_static(const ITensorInfo * input,const int block_shape_x,const int block_shape_y,const Size2D & padding_left,const Size2D & padding_right,const ITensorInfo * output)63 Status validate_arguments_static(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
64                                  const ITensorInfo *output)
65 {
66     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
67     ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
68     ARM_COMPUTE_RETURN_ERROR_ON(input->num_dimensions() > 4);
69     ARM_COMPUTE_RETURN_ERROR_ON(block_shape_x < 1 || block_shape_y < 1);
70 
71     // Validate output if initialized
72     if(output->total_size() != 0)
73     {
74         TensorShape expected_output_shape = misc::shape_calculator::compute_space_to_batch_shape(input, block_shape_x, block_shape_y, padding_left, padding_right);
75         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), expected_output_shape);
76         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
77         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
78     }
79 
80     return Status{};
81 }
82 } // namespace
83 
CLSpaceToBatchLayerKernel()84 CLSpaceToBatchLayerKernel::CLSpaceToBatchLayerKernel()
85     : _input(nullptr), _block_shape(nullptr), _paddings(nullptr), _output(nullptr)
86 {
87 }
88 
configure(const ICLTensor * input,const ICLTensor * block_shape,const ICLTensor * paddings,ICLTensor * output)89 void CLSpaceToBatchLayerKernel::configure(const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output)
90 {
91     configure(CLKernelLibrary::get().get_compile_context(), input, block_shape, paddings, output);
92 }
93 
configure(const CLCompileContext & compile_context,const ICLTensor * input,const ICLTensor * block_shape,const ICLTensor * paddings,ICLTensor * output)94 void CLSpaceToBatchLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLTensor *block_shape, const ICLTensor *paddings, ICLTensor *output)
95 {
96     ARM_COMPUTE_ERROR_ON_NULLPTR(input, block_shape, paddings, output);
97     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), block_shape->info(), paddings->info(), output->info()));
98 
99     _input       = input;
100     _block_shape = block_shape;
101     _paddings    = paddings;
102     _output      = output;
103 
104     const DataLayout data_layout = input->info()->data_layout();
105     const int        idx_width   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
106     const int        idx_height  = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
107     const int        idx_batch   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::BATCHES);
108 
109     // Create kernel
110     CLBuildOptions build_opts;
111     build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
112     build_opts.add_option("-DWIDTH_OUT=" + support::cpp11::to_string(output->info()->dimension(idx_width)));
113     build_opts.add_option("-DHEIGHT_OUT=" + support::cpp11::to_string(output->info()->dimension(idx_height)));
114     build_opts.add_option("-DBATCH_SIZE=" + support::cpp11::to_string(output->info()->dimension(idx_batch)));
115     build_opts.add_option("-DWIDTH_IN=" + support::cpp11::to_string(input->info()->dimension(idx_width)));
116     build_opts.add_option("-DHEIGHT_IN=" + support::cpp11::to_string(input->info()->dimension(idx_height)));
117     build_opts.add_option("-DBATCH_IN=" + support::cpp11::to_string(input->info()->dimension(idx_batch)));
118     _kernel = create_kernel(compile_context, "space_to_batch_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
119 
120     // Configure kernel window
121     Window win = calculate_max_window(*output->info(), Steps());
122     ICLKernel::configure_internal(win);
123 }
124 
configure(const ICLTensor * input,const int block_shape_x,const int block_shape_y,const Size2D & padding_left,const Size2D & padding_right,ICLTensor * output)125 void CLSpaceToBatchLayerKernel::configure(const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
126                                           ICLTensor *output)
127 {
128     configure(CLKernelLibrary::get().get_compile_context(), input, block_shape_x, block_shape_y, padding_left, padding_right, output);
129 }
130 
configure(const CLCompileContext & compile_context,const ICLTensor * input,const int block_shape_x,const int block_shape_y,const Size2D & padding_left,const Size2D & padding_right,ICLTensor * output)131 void CLSpaceToBatchLayerKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left,
132                                           const Size2D &padding_right,
133                                           ICLTensor    *output)
134 {
135     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
136 
137     TensorShape output_shape = misc::shape_calculator::compute_space_to_batch_shape(input->info(), block_shape_x, block_shape_y, padding_left, padding_right);
138     auto_init_if_empty(*output->info(), output_shape, 1, input->info()->data_type(), input->info()->quantization_info());
139 
140     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments_static(input->info(), block_shape_x, block_shape_y, padding_left, padding_right, output->info()));
141 
142     _input  = input;
143     _output = output;
144 
145     const DataLayout data_layout = input->info()->data_layout();
146     const int        idx_width   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
147     const int        idx_height  = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
148     const int        idx_batch   = get_data_layout_dimension_index(data_layout, DataLayoutDimension::BATCHES);
149 
150     // Create kernel
151     CLBuildOptions build_opts;
152     build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
153     build_opts.add_option("-DWIDTH_OUT=" + support::cpp11::to_string(output->info()->dimension(idx_width)));
154     build_opts.add_option("-DHEIGHT_OUT=" + support::cpp11::to_string(output->info()->dimension(idx_height)));
155     build_opts.add_option("-DBATCH_SIZE=" + support::cpp11::to_string(output->info()->dimension(idx_batch)));
156     build_opts.add_option("-DWIDTH_IN=" + support::cpp11::to_string(input->info()->dimension(idx_width)));
157     build_opts.add_option("-DHEIGHT_IN=" + support::cpp11::to_string(input->info()->dimension(idx_height)));
158     build_opts.add_option("-DBATCH_IN=" + support::cpp11::to_string(input->info()->dimension(idx_batch)));
159     build_opts.add_option("-DBLOCK_SHAPE_X=" + support::cpp11::to_string(block_shape_x));
160     build_opts.add_option("-DBLOCK_SHAPE_Y=" + support::cpp11::to_string(block_shape_y));
161     build_opts.add_option("-DPAD_LEFT_X=" + support::cpp11::to_string(padding_left.x()));
162     build_opts.add_option("-DPAD_RIGHT_X=" + support::cpp11::to_string(padding_right.x()));
163     build_opts.add_option("-DPAD_LEFT_Y=" + support::cpp11::to_string(padding_left.y()));
164     build_opts.add_option("-DPAD_RIGHT_Y=" + support::cpp11::to_string(padding_right.y()));
165     _kernel = create_kernel(compile_context, "space_to_batch_static_" + lower_string(string_from_data_layout(input->info()->data_layout())), build_opts.options());
166 
167     // Configure kernel window
168     Window win = calculate_max_window(*output->info(), Steps());
169     ICLKernel::configure_internal(win);
170 }
171 
validate(const ITensorInfo * input,const ITensorInfo * block_shape,const ITensorInfo * paddings,const ITensorInfo * output)172 Status CLSpaceToBatchLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *paddings, const ITensorInfo *output)
173 {
174     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, block_shape, paddings, output));
175     return Status{};
176 }
validate(const ITensorInfo * input,const int block_shape_x,const int block_shape_y,const Size2D & padding_left,const Size2D & padding_right,const ITensorInfo * output)177 Status CLSpaceToBatchLayerKernel::validate(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
178                                            const ITensorInfo *output)
179 {
180     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments_static(input, block_shape_x, block_shape_y, padding_left, padding_right, output));
181     return Status{};
182 }
183 
run(const Window & window,cl::CommandQueue & queue)184 void CLSpaceToBatchLayerKernel::run(const Window &window, cl::CommandQueue &queue)
185 {
186     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
187     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
188 
189     Window slice_out = window.first_slice_window_3D();
190 
191     Window slice_in = window.first_slice_window_4D();
192     slice_in.set(Window::DimX, Window::Dimension(0, 0, 0));
193     slice_in.set(Window::DimY, Window::Dimension(0, 0, 0));
194     slice_in.set(Window::DimZ, Window::Dimension(0, 0, 0));
195     slice_in.set(3, Window::Dimension(0, 0, 0));
196 
197     Window vector_slice = window.first_slice_window_1D();
198     vector_slice.set(Window::DimX, Window::Dimension(0, 0, 0));
199 
200     Window padding_slice = window.first_slice_window_2D();
201     padding_slice.set(Window::DimX, Window::Dimension(0, 0, 0));
202     padding_slice.set(Window::DimY, Window::Dimension(0, 0, 0));
203 
204     int batch_id = 0;
205     do
206     {
207         unsigned int idx  = 0;
208         const bool   cond = (_paddings != nullptr && _block_shape != nullptr);
209         add_4D_tensor_argument(idx, _input, slice_in);
210         add_2D_tensor_argument_if(cond, idx, _paddings, padding_slice);
211         add_1D_tensor_argument_if(cond, idx, _block_shape, vector_slice);
212 
213         add_argument(idx, batch_id);
214         add_3D_tensor_argument(idx, _output, slice_out);
215         enqueue(queue, *this, slice_out, lws_hint());
216         ++batch_id;
217     }
218     while(window.slide_window_slice_3D(slice_out));
219 }
220 } // namespace arm_compute
221