/*
 * 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/CLGEMMInterleave4x4Kernel.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/Helpers.h"
#include "arm_compute/core/Types.h"
#include "arm_compute/core/Utils.h"
#include "arm_compute/core/Validate.h"
#include "arm_compute/core/Window.h"
#include "arm_compute/core/utils/misc/ShapeCalculator.h"

using namespace arm_compute;
using namespace arm_compute::misc::shape_calculator;

namespace
{
Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
{
    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 1, DataType::QS8, DataType::QASYMM8, DataType::U8, DataType::S8,
                                                         DataType::QS16, DataType::U16, DataType::S16, DataType::U32, DataType::S32,
                                                         DataType::F16, DataType::F32);
    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);

    if(output->total_size() != 0)
    {
        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), compute_interleaved_shape(*input));
        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_FIXED_POINT(input, output);
    }

    return Status{};
}

std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
{
    unsigned int           num_elems_processed_per_iteration_x = max_cl_vector_width / data_size_from_type(input->data_type());
    constexpr unsigned int num_elems_processed_per_iteration_y = 4;
    const unsigned int     num_elems_written_per_iteration     = num_elems_processed_per_iteration_x * num_elems_processed_per_iteration_y;
    bool                   window_changed                      = false;

    // Configure kernel window
    Window                win = calculate_max_window(*input, Steps(num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y));
    AccessWindowRectangle input_access(input, 0, 0, num_elems_processed_per_iteration_x, num_elems_processed_per_iteration_y);
    window_changed = window_changed || update_window_and_padding(win, input_access);

    // Configure window in case of configured output
    if(output->total_size() != 0)
    {
        AccessWindowRectangle output_access(output, 0, 0, num_elems_written_per_iteration, 1, 4.f, 0.25f);
        window_changed = window_changed || update_window_and_padding(win, output_access);
        output_access.set_valid_region(win, input->valid_region());
    }

    Status err = (window_changed) ? ARM_COMPUTE_CREATE_ERROR(ErrorCode::RUNTIME_ERROR, "Insufficient Padding!") : Status{};
    return std::make_pair(err, win);
}
} // namespace

CLGEMMInterleave4x4Kernel::CLGEMMInterleave4x4Kernel()
    : _input(nullptr), _output(nullptr)
{
}

void CLGEMMInterleave4x4Kernel::configure(const ICLTensor *input, ICLTensor *output)
{
    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);

    // Output auto inizialitation if not yet initialized
    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(compute_interleaved_shape(*input->info())));

    // Perform validate step
    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info()));

    _input  = input;
    _output = output;

    // Create kernel
    std::string kernel_name = "gemm_interleave4x4_" + support::cpp11::to_string(input->info()->element_size() * 8) + "bit";
    _kernel                 = static_cast<cl::Kernel>(CLKernelLibrary::get().create_kernel(kernel_name));

    // Configure kernel window
    auto win_config = validate_and_configure_window(input->info(), output->info());
    ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
    ICLKernel::configure(win_config.second);

    // Set config_id for enabling LWS tuning
    _config_id = "interleave4x4_";
    _config_id += lower_string(string_from_data_type(input->info()->data_type()));
    _config_id += "_";
    _config_id += support::cpp11::to_string(output->info()->dimension(0));
    _config_id += "_";
    _config_id += support::cpp11::to_string(output->info()->dimension(1));
}

Status CLGEMMInterleave4x4Kernel::validate(const ITensorInfo *input, const ITensorInfo *output)
{
    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
    ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first);

    return Status{};
}

void CLGEMMInterleave4x4Kernel::run(const Window &window, cl::CommandQueue &queue)
{
    ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
    ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);

    /*
     * This kernel puts the values in a 4x4 block of Matrix A on the same row (Interleaved values)
     *         |a00 a01 a02 a03|
     *         |a10 a11 a12 a13|
     *         |a20 a21 a22 a23| = | a00 a10 a20 a30 || a01 a11 a21 a31 || a02 a12 a22 a32 || a03 a13 a23 a33 |
     *         |a30 a31 a32 a33|
     *
     * After this operation, the output matrix will have the following shape: [ height * 4, width / 4 ]
     */
    Window in_slice  = window.first_slice_window_2D();
    Window out_slice = window.first_slice_window_2D();

    // Change x and y steps for the slide of output tensor
    out_slice.scale(Window::DimX, 4.f);
    out_slice.scale(Window::DimY, 0.25f);

    do
    {
        unsigned int idx = 0;
        add_2D_tensor_argument(idx, _input, in_slice);
        add_2D_tensor_argument(idx, _output, out_slice);
        enqueue(queue, *this, in_slice, _lws_hint);
    }
    while(window.slide_window_slice_2D(in_slice) && window.slide_window_slice_2D(out_slice));
}