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/CL/kernels/CLFillBorderKernel.cpp | 175 +++++++++++++++++++++++++++++ 1 file changed, 175 insertions(+) create mode 100644 src/core/CL/kernels/CLFillBorderKernel.cpp (limited to 'src/core/CL/kernels/CLFillBorderKernel.cpp') diff --git a/src/core/CL/kernels/CLFillBorderKernel.cpp b/src/core/CL/kernels/CLFillBorderKernel.cpp new file mode 100644 index 0000000000..981aad665a --- /dev/null +++ b/src/core/CL/kernels/CLFillBorderKernel.cpp @@ -0,0 +1,175 @@ +/* + * 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/CL/kernels/CLFillBorderKernel.h" + +#include "arm_compute/core/CL/CLHelpers.h" +#include "arm_compute/core/CL/CLKernelLibrary.h" +#include "arm_compute/core/CL/ICLTensor.h" +#include "arm_compute/core/CL/OpenCL.h" +#include "arm_compute/core/Error.h" +#include "arm_compute/core/TensorInfo.h" +#include "arm_compute/core/Utils.h" +#include "arm_compute/core/Validate.h" +#include "arm_compute/core/Window.h" + +#include +#include +#include +#include + +using namespace arm_compute; + +CLFillBorderKernel::CLFillBorderKernel() + : ICLKernel(), _tensor(nullptr) +{ +} + +bool CLFillBorderKernel::is_parallelisable() const +{ + return false; +} + +template +void CLFillBorderKernel::set_constant_border(unsigned int idx, const PixelValue &constant_border_value) +{ + T value; + constant_border_value.get(value); + ICLKernel::add_argument(idx, static_cast(value)); +} + +void CLFillBorderKernel::configure(ICLTensor *tensor, BorderSize border_size, BorderMode border_mode, const PixelValue &constant_border_value) +{ + ARM_COMPUTE_ERROR_ON(tensor == nullptr); + ARM_COMPUTE_ERROR_ON(tensor->info()->num_channels() != 1); + + border_size.limit(tensor->info()->padding()); + + // If there is no border: early exit + if(border_size.empty() || border_mode == BorderMode::UNDEFINED) + { + return; + } + + // Select appropriate kernel + std::string kernel_name = "fill_image_borders_" + lower_string(string_from_border_mode(border_mode)); + + // Define select type required by replicate border > 1 + const DataType dt = tensor->info()->data_type(); + std::string select_type = get_cl_type_from_data_type(dt); + if(is_data_type_float(dt)) + { + select_type = (DataType::F32 == dt) ? "int" : "short"; + } + + // Define build options + std::set build_opts; + build_opts.emplace(("-DDATA_TYPE=" + get_cl_type_from_data_type(dt))); + build_opts.emplace(("-DSELECT_TYPE=" + select_type)); + build_opts.emplace(("-DBORDER_SIZE_TOP=" + val_to_string(border_size.top))); + build_opts.emplace(("-DBORDER_SIZE_BOTTOM=" + val_to_string(border_size.bottom))); + build_opts.emplace(("-DBORDER_SIZE_LEFT=" + val_to_string(border_size.left))); + build_opts.emplace(("-DBORDER_SIZE_RIGHT=" + val_to_string(border_size.right))); + + // Create kernel + _kernel = static_cast(CLKernelLibrary::get().create_kernel(kernel_name, build_opts)); + _tensor = tensor; + + // Create static kernel arguments + const unsigned int valid_width = tensor->info()->valid_region().shape[0]; + const unsigned int valid_height = tensor->info()->valid_region().shape[1]; + const cl_int2 valid_region_coords = + { + { + static_cast(tensor->info()->valid_region().anchor[0]), + static_cast(tensor->info()->valid_region().anchor[1]), + } + }; + const unsigned int total_valid_width = border_size.left + valid_width + border_size.right; + + // Set static kernel arguments + unsigned int idx = num_arguments_per_2D_tensor(); //Skip the tensor parameters + ICLKernel::add_argument(idx, valid_width); + ICLKernel::add_argument(idx, valid_height); + ICLKernel::add_argument(idx, valid_region_coords); + if(BorderMode::CONSTANT == border_mode) + { + switch(dt) + { + case DataType::U8: + set_constant_border(idx, constant_border_value); + break; + case DataType::U16: + set_constant_border(idx, constant_border_value); + break; + case DataType::S16: + set_constant_border(idx, constant_border_value); + break; + case DataType::U32: + set_constant_border(idx, constant_border_value); + break; + case DataType::S32: + set_constant_border(idx, constant_border_value); + break; + case DataType::F32: + static_assert(sizeof(float) == 4, "Float must be 32 bit"); + set_constant_border(idx, constant_border_value); + break; + case DataType::F16: + static_assert(sizeof(cl_half) == 2, "Half must be 16 bit"); + set_constant_border(idx, constant_border_value); + break; + default: + ARM_COMPUTE_ERROR("Not handled"); + } + } + + // Configure kernel window + Window win; + win.set(Window::DimX, Window::Dimension(0, total_valid_width + valid_height)); + win.set(Window::DimY, Window::Dimension(0, 1, 1)); + win.use_tensor_dimensions(tensor->info(), Window::DimZ); + ICLKernel::configure(win); +} + +void CLFillBorderKernel::run(const Window &window, cl::CommandQueue &queue) +{ + // Border mode undefined or border width == 0 + if(_kernel() == nullptr) + { + return; + } + + ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); + ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window); + + Window slice = window.first_slice_window_2D(); + + do + { + unsigned int idx = 0; + add_2D_tensor_argument(idx, _tensor, slice); + enqueue(queue, *this, slice, cl::NullRange); + } + while(window.slide_window_slice_2D(slice)); +} -- cgit v1.2.1