src/core/CL/kernels/CLFloorKernel.cpp

   1 /*
   2  * Copyright (c) 2017 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 "arm_compute/core/CL/kernels/CLFloorKernel.h"
  25
  26 #include "arm_compute/core/CL/CLHelpers.h"
  27 #include "arm_compute/core/CL/CLKernelLibrary.h"
  28 #include "arm_compute/core/CL/ICLTensor.h"
  29 #include "arm_compute/core/Helpers.h"
  30 #include "arm_compute/core/IAccessWindow.h"
  31 #include "arm_compute/core/TensorInfo.h"
  32 #include "arm_compute/core/Utils.h"
  33 #include "arm_compute/core/Validate.h"
  34 #include "arm_compute/core/Window.h"
  35
  36 using namespace arm_compute;
  37
  38 CLFloorKernel::CLFloorKernel()
  39     : _input(nullptr), _output(nullptr)
  40 {
  41 }
  42
  43 void CLFloorKernel::configure(const ICLTensor *input, ICLTensor *output)
  44 {
  45     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
  46
  47     set_shape_if_empty(*output->info(), input->info()->tensor_shape());
  48
  49     set_data_type_if_unknown(*input->info(), DataType::F32);
  50     set_data_type_if_unknown(*output->info(), DataType::F32);
  51
  52     ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::F32);
  53     ARM_COMPUTE_ERROR_ON_MISMATCHING_SHAPES(input, output);
  54     ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
  55
  56     _input  = input;
  57     _output = output;
  58
  59     constexpr unsigned int num_elems_processed_per_iteration = 4;
  60
  61     // Create kernel
  62     std::set<std::string> build_opts;
  63     build_opts.emplace(("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type())));
  64     build_opts.emplace(("-DVEC_SIZE=" + support::cpp11::to_string(num_elems_processed_per_iteration)));
  65     _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("floor_layer", build_opts));
  66
  67     // Configure kernel window
  68     Window                 win = calculate_max_window(*input->info(), Steps(num_elems_processed_per_iteration));
  69     AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration);
  70     AccessWindowHorizontal output_access(output->info(), 0, num_elems_processed_per_iteration);
  71     update_window_and_padding(win, input_access, output_access);
  72     output_access.set_valid_region(win, input->info()->valid_region());
  73
  74     ICLKernel::configure(win);
  75 }
  76
  77 void CLFloorKernel::run(const Window &window, cl::CommandQueue &queue)
  78 {
  79     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
  80     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
  81
  82     Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
  83     Window slice     = collapsed.first_slice_window_3D();
  84
  85     do
  86     {
  87         unsigned int idx = 0;
  88         add_3D_tensor_argument(idx, _input, slice);
  89         add_3D_tensor_argument(idx, _output, slice);
  90         enqueue(queue, *this, slice);
  91     }
  92     while(collapsed.slide_window_slice_3D(slice));
  93 }