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/NEON/kernels/NEReorgLayerKernel.h"
25
26 #include "arm_compute/core/Error.h"
27 #include "arm_compute/core/Helpers.h"
28 #include "arm_compute/core/ITensor.h"
29 #include "arm_compute/core/TensorInfo.h"
30 #include "arm_compute/core/Types.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
36 #include <cstddef>
37 #include <cstdint>
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 //Note: ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input) is not needed here as this kernel doesn't use NEON FP16 instructions.
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
NEReorgLayerKernel()68 NEReorgLayerKernel::NEReorgLayerKernel()
69 : _input(nullptr), _output(nullptr), _stride(1)
70 {
71 }
72
configure(const ITensor * input,ITensor * output,int32_t stride)73 void NEReorgLayerKernel::configure(const ITensor *input, ITensor *output, int32_t stride)
74 {
75 ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
76
77 // Output auto inizialitation if not yet initialized
78 const TensorShape output_shape = misc::shape_calculator::compute_reorg_output_shape(*input->info(), stride);
79 auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
80
81 // Perform validation step
82 ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), stride));
83
84 _input = input;
85 _output = output;
86 _stride = stride;
87
88 // The NEReorgLayerKernel doesn't need padding so update_window_and_padding() can be skipped
89 output->info()->set_valid_region(ValidRegion(Coordinates(), output->info()->tensor_shape()));
90
91 // Configure kernel window
92 Window win = calculate_max_window(*output->info(), Steps());
93
94 ICPPKernel::configure(win);
95 }
96
validate(const ITensorInfo * input,const ITensorInfo * output,int32_t stride)97 Status NEReorgLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output, int32_t stride)
98 {
99 ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, stride));
100 return Status{};
101 }
102
run(const Window & window,const ThreadInfo & info)103 void NEReorgLayerKernel::run(const Window &window, const ThreadInfo &info)
104 {
105 ARM_COMPUTE_UNUSED(info);
106 ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
107 ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICPPKernel::window(), window);
108
109 const DataLayout data_layout = _input->info()->data_layout();
110 const size_t idx_w = get_data_layout_dimension_index(data_layout, DataLayoutDimension::WIDTH);
111 const size_t idx_h = get_data_layout_dimension_index(data_layout, DataLayoutDimension::HEIGHT);
112 const size_t idx_c = get_data_layout_dimension_index(data_layout, DataLayoutDimension::CHANNEL);
113
114 const unsigned int stride = _stride;
115 const unsigned int out_c = _output->info()->tensor_shape()[idx_c] / (stride * stride);
116 const uint8_t *in_ptr = _input->buffer();
117
118 // Collapse
119 Window collapsed_window = window.collapse_if_possible(window, 4);
120
121 // Create Iterator
122 Iterator out(_output, collapsed_window);
123
124 // Perform reorg
125 execute_window_loop(collapsed_window, [&](const Coordinates & id)
126 {
127 // Get spatial coords and channels
128 const unsigned int w = id[idx_w];
129 const unsigned int h = id[idx_h];
130 const unsigned int c = id[idx_c];
131
132 // Calculate mapping
133 const unsigned int offset = c / out_c;
134 Coordinates map_coords = id;
135 map_coords.set(idx_w, w * stride + offset % stride);
136 map_coords.set(idx_h, h * stride + offset / stride);
137 map_coords.set(idx_c, c % out_c);
138
139 // Perform mapping
140 std::memcpy(out.ptr(), in_ptr + _input->info()->offset_element_in_bytes(map_coords), _input->info()->element_size());
141 },
142 out);
143 }
144 } // namespace arm_compute
145