2 * Copyright (c) 2017-2018 ARM Limited.
4 * SPDX-License-Identifier: MIT
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:
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
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
24 #include "arm_compute/runtime/NEON/functions/NEDeconvolutionLayer.h"
26 #include "arm_compute/core/Helpers.h"
27 #include "arm_compute/core/Utils.h"
28 #include "arm_compute/core/Validate.h"
29 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
31 using namespace arm_compute;
32 using namespace arm_compute::misc::shape_calculator;
34 NEDeconvolutionLayer::NEDeconvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager) // NOLINT
35 : _memory_group(std::move(memory_manager)),
45 Status NEDeconvolutionLayer::validate(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *bias, const ITensorInfo *output, const PadStrideInfo &info,
46 unsigned int inner_border_right, unsigned int inner_border_top)
48 ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, weights, output);
49 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
50 ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(weights, 1, DataType::F32);
51 ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(0) != weights->dimension(1));
52 ARM_COMPUTE_RETURN_ERROR_ON(weights->dimension(0) < 1);
53 ARM_COMPUTE_RETURN_ERROR_ON(!info.padding_is_symmetric());
55 const unsigned int stride_x = info.stride().first;
56 const unsigned int stride_y = info.stride().second;
58 ARM_COMPUTE_RETURN_ERROR_ON_MSG(inner_border_right > stride_x - 1, "inner_border_right must be smaller than stride_x");
59 ARM_COMPUTE_RETURN_ERROR_ON_MSG(inner_border_top > stride_y - 1, "inner_border_top must be smaller than stride_y");
61 auto out_dims = deconvolution_output_dimensions(input->dimension(0), input->dimension(1), weights->dimension(0), weights->dimension(1),
62 info.pad().first, info.pad().second, inner_border_right, inner_border_top, stride_x, stride_y);
64 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, weights, bias);
65 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, weights, bias);
69 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, bias);
70 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, bias);
73 if(output->tensor_shape().total_size() > 0)
75 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
76 ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
78 const TensorShape output_shape = deconvolution_output_shape(out_dims, input->tensor_shape(), weights->tensor_shape());
80 ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->dimension(Window::DimX) != output_shape.x(), "Output's width is invalid.");
81 ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->dimension(Window::DimY) != output_shape.y(), "Output's height is invalid.");
82 ARM_COMPUTE_RETURN_ERROR_ON_MSG(output->dimension(Window::DimZ) != output_shape.z(), "Output's depth is invalid.");
85 TensorInfo scale_out_info(input->clone()->set_is_resizable(true).reset_padding().set_tensor_shape(compute_deconvolution_shape(*input, stride_x, stride_y, inner_border_right, inner_border_top,
87 const PadStrideInfo conv_info(1, 1, 0, 0, 0, 0, DimensionRoundingType::CEIL);
89 for(size_t i = 2; i < Coordinates::num_max_dimensions; ++i)
91 ARM_COMPUTE_RETURN_ERROR_ON(input->dimension(i) != scale_out_info.dimension(i));
94 ARM_COMPUTE_RETURN_ON_ERROR(NEConvolutionLayer::validate(&scale_out_info, weights, bias, output, conv_info, WeightsInfo()));
99 void NEDeconvolutionLayer::configure(ITensor *input, const ITensor *weights, const ITensor *bias, ITensor *output, const PadStrideInfo &info,
100 unsigned int inner_border_right, unsigned int inner_border_top)
102 ARM_COMPUTE_ERROR_ON_NULLPTR(input, weights, output);
106 _inner_border = std::make_pair(inner_border_right, inner_border_top);
108 const unsigned int stride_x = info.stride().first;
109 const unsigned int stride_y = info.stride().second;
111 // Perform validation step
112 ARM_COMPUTE_ERROR_THROW_ON(NEDeconvolutionLayer::validate(input->info(), weights->info(), bias == nullptr ? nullptr : bias->info(), output->info(), info, inner_border_right, inner_border_top));
114 _memory_group.manage(&_scaled_output);
116 // configure scale function
117 // Init and allocate intermmidiate tensor for output, same size as input but the first two axis are the same as the output tensor
118 const TensorInfo scale_out_info(compute_deconvolution_shape(*input->info(), stride_x, stride_y, inner_border_right, inner_border_top, info), 1, input->info()->data_type(),
119 input->info()->fixed_point_position());
120 _scaled_output.allocator()->init(scale_out_info);
122 // setup the function to convolve the upscaled output
123 const PadStrideInfo conv_info(1, 1, 0, 0, 0, 0, DimensionRoundingType::CEIL);
124 _conv_f.configure(&_scaled_output, weights, bias, output, conv_info);
126 // Allocate auxiliary tensors
127 _scaled_output.allocator()->allocate();
129 // configure upsample function
130 _upsample_f.configure(input, &_scaled_output, info, inner_border_right, inner_border_top);
133 void NEDeconvolutionLayer::run()
135 _memory_group.acquire();
137 // Run upsample kernel
140 // Run convolution layer
143 _memory_group.release();