/*
 * Copyright (c) 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/CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel.h"

#include "arm_compute/core/AccessWindowStatic.h"
#include "arm_compute/core/CL/ICLTensor.h"
#include "arm_compute/core/Error.h"
#include "arm_compute/core/Helpers.h"
#include "arm_compute/core/Types.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/Window.h"
#include "support/ToolchainSupport.h"

using namespace arm_compute;

namespace arm_compute
{
class Coordinates;
} // namespace arm_compute

CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel::CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel()
    : _input(nullptr), _bias(nullptr), _output(nullptr)
{
}

void CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel::configure(const ICLTensor *input, const ICLTensor *bias, ICLTensor *output, int result_offset, int result_mult_int, int result_shift, int min,
                                                              int max)
{
    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::S32);
    ARM_COMPUTE_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(output, 1, DataType::QASYMM8);
    ARM_COMPUTE_ERROR_ON(max > 255);
    ARM_COMPUTE_ERROR_ON(min < 0 || min > max);

    if(bias != nullptr)
    {
        ARM_COMPUTE_ERROR_ON_MISMATCHING_DATA_TYPES(input, bias);
        ARM_COMPUTE_ERROR_ON(bias->info()->num_dimensions() > 1);
        ARM_COMPUTE_ERROR_ON(input->info()->dimension(0) != bias->info()->dimension(0));
    }

    _input  = input;
    _bias   = bias;
    _output = output;

    // Set the arguments to pass at compile time
    CLBuildOptions build_opts;
    build_opts.add_option("-DRESULT_OFFSET=" + support::cpp11::to_string(result_offset));
    build_opts.add_option("-DRESULT_MULT_INT=" + support::cpp11::to_string(result_mult_int));
    build_opts.add_option("-DRESULT_SHIFT=" + support::cpp11::to_string(result_shift));
    build_opts.add_option_if((min != 0) && (min != max), "-DMIN_BOUND=" + support::cpp11::to_string(min));
    build_opts.add_option_if((max != 255) && (min != max), "-DMAX_BOUND=" + support::cpp11::to_string(max));
    build_opts.add_option_if(bias != nullptr, "-DADD_BIAS");

    // Create kernel
    _kernel = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel("gemmlowp_output_stage_quantize_down", build_opts.options()));

    constexpr unsigned int num_elems_processed_per_iteration = 16;

    // Configure kernel window
    Window win = calculate_max_window(*output->info(), Steps(num_elems_processed_per_iteration));

    AccessWindowHorizontal input_access(input->info(), 0, num_elems_processed_per_iteration);
    AccessWindowHorizontal output_result_access(output->info(), 0, num_elems_processed_per_iteration);

    update_window_and_padding(win,
                              input_access,
                              output_result_access);

    if(bias != nullptr)
    {
        AccessWindowStatic bias_access(bias->info(), 0, 0, ceil_to_multiple(bias->info()->dimension(0), num_elems_processed_per_iteration), bias->info()->tensor_shape()[1]);

        update_window_and_padding(win,
                                  bias_access);
    }

    output_result_access.set_valid_region(win, ValidRegion(Coordinates(0, 0), output->info()->tensor_shape()));

    ICLKernel::configure(win);
}

void CLGEMMLowpQuantizeDownInt32ToUint8ScaleKernel::run(const Window &window, cl::CommandQueue &queue)
{
    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);

    Window collapsed = window.collapse_if_possible(ICLKernel::window(), Window::DimZ);
    Window slice     = collapsed.first_slice_window_3D();

    unsigned int idx1 = num_arguments_per_3D_tensor();
    if(_bias != nullptr)
    {
        Window biases_slice(slice);
        biases_slice.set(Window::DimY, Window::Dimension(0, 1, 1));
        biases_slice.set(Window::DimZ, Window::Dimension(0, 1, 1));
        add_1D_tensor_argument(idx1, _bias, biases_slice);
    }

    do
    {
        unsigned int idx = 0;
        add_3D_tensor_argument(idx, _input, slice);
        add_3D_tensor_argument(idx1, _output, slice);
        enqueue(queue, *this, slice);
    }
    while(collapsed.slide_window_slice_3D(slice));
}