From 33a6990ee6ba7bf85b88822d9723060262d00785 Mon Sep 17 00:00:00 2001 From: Michalis Spyrou Date: Fri, 23 Feb 2018 15:01:52 +0000 Subject: COMPMID-540 Replace NEDeconvolutionLayerUpsampleKernel with NEScaleKernel Change-Id: Ic29557cca24447ef40fa2cfca84f208b4d43f8de Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/122180 Tested-by: Jenkins Reviewed-by: Michele DiGiorgio Reviewed-by: Pablo Tello --- src/runtime/CPP/functions/CPPUpsample.cpp | 36 +++++++++++++++++++++ .../NEON/functions/NEDeconvolutionLayer.cpp | 37 ++++------------------ 2 files changed, 43 insertions(+), 30 deletions(-) create mode 100644 src/runtime/CPP/functions/CPPUpsample.cpp (limited to 'src/runtime') diff --git a/src/runtime/CPP/functions/CPPUpsample.cpp b/src/runtime/CPP/functions/CPPUpsample.cpp new file mode 100644 index 0000000000..619b7e1f43 --- /dev/null +++ b/src/runtime/CPP/functions/CPPUpsample.cpp @@ -0,0 +1,36 @@ +/* + * Copyright (c) 2017-2018 ARM Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include "arm_compute/runtime/CPP/functions/CPPUpsample.h" + +#include "arm_compute/core/CPP/kernels/CPPUpsampleKernel.h" +#include "support/ToolchainSupport.h" + +using namespace arm_compute; + +void CPPUpsample::configure(const ITensor *input, ITensor *output, const PadStrideInfo &info, unsigned int inner_border_right, unsigned int inner_border_top) +{ + auto k = arm_compute::support::cpp14::make_unique(); + k->configure(input, output, info, inner_border_right, inner_border_top); + _kernel = std::move(k); +} \ No newline at end of file diff --git a/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp b/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp index c1ba5dd36e..693d7a4f70 100644 --- a/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp +++ b/src/runtime/NEON/functions/NEDeconvolutionLayer.cpp @@ -34,6 +34,7 @@ using namespace arm_compute::misc::shape_calculator; NEDeconvolutionLayer::NEDeconvolutionLayer(std::shared_ptr memory_manager) // NOLINT : _memory_group(std::move(memory_manager)), _conv_f(), + _upsample_f(), _scaled_output(), _input(nullptr), _info(), @@ -79,44 +80,20 @@ void NEDeconvolutionLayer::configure(ITensor *input, const ITensor *weights, con // Allocate auxiliary tensors _scaled_output.allocator()->allocate(); + + // configure upsample function + _upsample_f.configure(input, &_scaled_output, info, inner_border_right, inner_border_top); } void NEDeconvolutionLayer::run() { _memory_group.acquire(); - // Initialize _scaled_output buffer - const int width_in = _input->info()->dimension(0); - const int height_in = _input->info()->dimension(1); - const int width_scaled = _scaled_output.info()->dimension(0); - const int height_scaled = _scaled_output.info()->dimension(1); - const int num_2d_slices = _input->info()->tensor_shape().total_size() / (width_in * height_in); - const int stride_x = _info.stride().first; - const int stride_y = _info.stride().second; - - std::fill_n(_scaled_output.buffer(), _scaled_output.info()->total_size(), 0); - - // scaled_output is the input for the forward convolution. We copy the input elements to scaled_output - // and insert rows and columns with zeroes depending on the stride values. - for(int slice = 0; slice < num_2d_slices; ++slice) - { - const int start_x = _info.pad().first; - const int start_y = _inner_border.second + _info.pad().second; - const int end_y = height_scaled - _info.pad().second; - const int end_x = width_scaled - _inner_border.first - _info.pad().first; - - for(int yi = start_y, in_y = 0; yi < end_y; yi += stride_y, in_y++) - { - for(int xi = start_x, in_x = 0; xi < end_x; xi += stride_x, in_x++) - { - const auto in = *(reinterpret_cast(_input->buffer() + _input->info()->offset_element_in_bytes(Coordinates(in_x, in_y, slice)))); - *(reinterpret_cast(_scaled_output.buffer() + _scaled_output.info()->offset_element_in_bytes(Coordinates(xi, yi, slice)))) = in; - } - } - } + // Run upsample kernel + _upsample_f.run(); // Run convolution layer _conv_f.run(); _memory_group.release(); -} +} \ No newline at end of file -- cgit v1.2.1