From 6ff3b19ee6120edf015fad8caab2991faa3070af Mon Sep 17 00:00:00 2001 From: Anthony Barbier Date: Mon, 4 Sep 2017 18:44:23 +0100 Subject: COMPMID-344 Updated doxygen Change-Id: I32f7b84daa560e460b77216add529c8fa8b327ae --- src/core/NEON/kernels/NEFillBorderKernel.cpp | 259 +++++++++++++++++++++++++++ 1 file changed, 259 insertions(+) create mode 100644 src/core/NEON/kernels/NEFillBorderKernel.cpp (limited to 'src/core/NEON/kernels/NEFillBorderKernel.cpp') diff --git a/src/core/NEON/kernels/NEFillBorderKernel.cpp b/src/core/NEON/kernels/NEFillBorderKernel.cpp new file mode 100644 index 0000000000..bd99242b11 --- /dev/null +++ b/src/core/NEON/kernels/NEFillBorderKernel.cpp @@ -0,0 +1,259 @@ +/* + * Copyright (c) 2016, 2017 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/core/NEON/kernels/NEFillBorderKernel.h" + +#include "arm_compute/core/Error.h" +#include "arm_compute/core/Helpers.h" +#include "arm_compute/core/ITensor.h" +#include "arm_compute/core/TensorInfo.h" +#include "arm_compute/core/Validate.h" +#include "arm_compute/core/Window.h" + +#include +#include + +using namespace arm_compute; + +namespace arm_compute +{ +class Coordinates; +} // namespace arm_compute + +NEFillBorderKernel::NEFillBorderKernel() + : _tensor(nullptr), _border_size(0), _mode(BorderMode::UNDEFINED), _constant_border_value(0) +{ +} + +void NEFillBorderKernel::configure(ITensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value) +{ + ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(tensor, 1, DataType::U8, DataType::QS8, DataType::QS16, DataType::U16, DataType::S16, DataType::U32, DataType::S32, DataType::F32); + + _tensor = tensor; + _border_size = border_size; + _mode = border_mode; + _constant_border_value = constant_border_value; + + _border_size.limit(tensor->info()->padding()); + + Window win; + win.set(Window::DimX, Window::Dimension(0, 1, 1)); + win.set(Window::DimY, Window::Dimension(0, 1, 1)); + win.use_tensor_dimensions(_tensor->info(), Window::DimZ); + INEKernel::configure(win); +} + +void NEFillBorderKernel::run(const Window &window) +{ + // If there is no border: early exit + if(_border_size.empty()) + { + return; + } + + ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); + ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); + + switch(_mode) + { + case BorderMode::CONSTANT: + { + switch(_tensor->info()->data_type()) + { + case DataType::U8: + fill_constant_value_single_channel(window); + break; + case DataType::QS8: + case DataType::S8: + fill_constant_value_single_channel(window); + break; + case DataType::U16: + fill_constant_value_single_channel(window); + break; + case DataType::S16: + case DataType::QS16: + fill_constant_value_single_channel(window); + break; + case DataType::U32: + fill_constant_value_single_channel(window); + break; + case DataType::S32: + fill_constant_value_single_channel(window); + break; + case DataType::F32: + static_assert(sizeof(float) == 4, "Float must be 32 bit"); + fill_constant_value_single_channel(window); + break; + default: + ARM_COMPUTE_ERROR("Not handled"); + } + break; + } + case BorderMode::REPLICATE: + { + switch(_tensor->info()->data_type()) + { + case DataType::U8: + fill_replicate_single_channel(window); + break; + case DataType::QS8: + case DataType::S8: + fill_replicate_single_channel(window); + break; + case DataType::U16: + fill_replicate_single_channel(window); + break; + case DataType::S16: + case DataType::QS16: + fill_replicate_single_channel(window); + break; + case DataType::U32: + fill_replicate_single_channel(window); + break; + case DataType::S32: + fill_replicate_single_channel(window); + break; + case DataType::F32: + static_assert(sizeof(float) == 4, "Float must be 32 bit"); + fill_replicate_single_channel(window); + break; + default: + ARM_COMPUTE_ERROR("Not handled"); + } + break; + } + case BorderMode::UNDEFINED: + break; // Nothing to do here + default: + ARM_COMPUTE_ERROR("Unknown border mode"); + } +} + +template +void NEFillBorderKernel::fill_replicate_single_channel(const Window &window) +{ + uint8_t *const start_valid_region = _tensor->ptr_to_element(_tensor->info()->valid_region().anchor); + const size_t &width = _tensor->info()->valid_region().shape[0]; + const size_t &height = _tensor->info()->valid_region().shape[1]; + + // Left and right border + Window vertical(window); + vertical.set(Window::DimY, Window::Dimension(0, height, 1)); + + Iterator vertical_it(_tensor, vertical); + + execute_window_loop(vertical, [&](const Coordinates & id) + { + const auto row_start = reinterpret_cast(start_valid_region + vertical_it.offset()); + const auto left_val = *reinterpret_cast(vertical_it.ptr()); + const auto right_val = *(reinterpret_cast(vertical_it.ptr()) + width - 1); + + // Fill left and right borders + std::fill_n(row_start - _border_size.left, _border_size.left, left_val); + std::fill_n(row_start + width, _border_size.right, right_val); + }, + vertical_it); + + // Top and bottom border + Iterator plane_it(_tensor, window); + + // Iterate over all XY planes + execute_window_loop(window, [&](const Coordinates & id) + { + const auto first_row = reinterpret_cast(start_valid_region + plane_it.offset()); + + // Top border + for(int i = -_border_size.top; i < 0; ++i) + { + const auto row_start = reinterpret_cast(start_valid_region + plane_it.offset() + i * _tensor->info()->strides_in_bytes()[1]); + + // Copy top rows including left/right borders + std::copy_n(first_row - _border_size.left, _border_size.left + width + _border_size.right, row_start - _border_size.left); + } + + const auto last_row = reinterpret_cast(start_valid_region + plane_it.offset() + (height - 1) * _tensor->info()->strides_in_bytes()[1]); + + // Bottom border + for(unsigned int i = height; i < height + _border_size.bottom; ++i) + { + const auto row_start = reinterpret_cast(start_valid_region + plane_it.offset() + i * _tensor->info()->strides_in_bytes()[1]); + + // Copy bottom rows including left/right borders + std::copy_n(last_row - _border_size.left, _border_size.left + width + _border_size.right, row_start - _border_size.left); + } + }, + plane_it); +} + +template +void NEFillBorderKernel::fill_constant_value_single_channel(const Window &window) +{ + T constant_border_value; + _constant_border_value.get(constant_border_value); + + uint8_t *const start_valid_region = _tensor->ptr_to_element(_tensor->info()->valid_region().anchor); + const size_t &width = _tensor->info()->valid_region().shape[0]; + const size_t &height = _tensor->info()->valid_region().shape[1]; + + // Left and right border + Window vertical(window); + vertical.set(Window::DimY, Window::Dimension(0, height, 1)); + + Iterator vertical_it(_tensor, vertical); + + execute_window_loop(vertical, [&](const Coordinates & id) + { + const auto row_start = reinterpret_cast(start_valid_region + vertical_it.offset()); + + // Fill left and right borders + std::fill_n(row_start - _border_size.left, _border_size.left, constant_border_value); + std::fill_n(row_start + width, _border_size.right, constant_border_value); + }, + vertical_it); + + // Top and bottom border + Iterator plane_it(_tensor, window); + + // Iterate over all XY planes + execute_window_loop(window, [&](const Coordinates & id) + { + // Top border + for(int i = -_border_size.top; i < 0; ++i) + { + const auto row_start = reinterpret_cast(start_valid_region + plane_it.offset() + i * _tensor->info()->strides_in_bytes()[1]); + + // Fill top rows including left/right borders + std::fill_n(row_start - _border_size.left, _border_size.left + width + _border_size.right, constant_border_value); + } + + // Bottom border + for(unsigned int i = height; i < height + _border_size.bottom; ++i) + { + const auto row_start = reinterpret_cast(start_valid_region + plane_it.offset() + i * _tensor->info()->strides_in_bytes()[1]); + + // Fill bottom rows including left/right borders + std::fill_n(row_start - _border_size.left, _border_size.left + width + _border_size.right, constant_border_value); + } + }, + plane_it); +} -- cgit v1.2.1