src/core/NEON/kernels/assembly/depthwise.hpp

/*
 * Copyright (c) 2021-2023 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.
 */

#pragma once

#include "arm_gemm.hpp"
#include "arm_gemm_local.hpp"
#include "depthwise_common.hpp"
#include "premultiply.hpp"

namespace arm_conv
{
namespace depthwise
{
struct DepthwiseConfig
{
    DepthwiseMethod method = DepthwiseMethod::DEFAULT;
    std::string     filter = "";

    DepthwiseConfig(DepthwiseMethod method)
        : method(method) {};
    DepthwiseConfig() {};
};

struct DepthwiseArgs
{
    const CPUInfo *cpu_info;

    unsigned int kernel_rows, kernel_cols;
    unsigned int stride_rows, stride_cols;
    unsigned int dilation_rows, dilation_cols;

    unsigned int n_batches, input_rows, input_cols, input_channels;
    unsigned int output_rows, output_cols;
    unsigned int channel_multiplier;

    PaddingValues padding;

    arm_gemm::Activation activation;

    const DepthwiseConfig *config;

    bool fast_mode = false;

    DepthwiseArgs(
        const CPUInfo *cpu_info,
        unsigned int kernel_rows, unsigned int kernel_cols,
        unsigned int stride_rows, unsigned int stride_cols,
        unsigned int dilation_rows, unsigned int dilation_cols,
        unsigned int n_batches, unsigned int input_rows, unsigned int input_cols,
        unsigned int input_channels,
        unsigned int output_rows, unsigned int output_cols,
        unsigned int  channel_multiplier,
        PaddingValues padding, arm_gemm::Activation activation,

        const DepthwiseConfig *config)
        : cpu_info(cpu_info),
          kernel_rows(kernel_rows),
          kernel_cols(kernel_cols),
          stride_rows(stride_rows),
          stride_cols(stride_cols),
          dilation_rows(dilation_rows),
          dilation_cols(dilation_cols),
          n_batches(n_batches),
          input_rows(input_rows),
          input_cols(input_cols),
          input_channels(input_channels),
          output_rows(output_rows),
          output_cols(output_cols),
          channel_multiplier(channel_multiplier),
          padding(padding),
          activation(activation),
          config(config)
    {
    }

    DepthwiseArgs(
        const CPUInfo *cpu_info,
        unsigned int kernel_rows, unsigned int kernel_cols,
        unsigned int stride_rows, unsigned int stride_cols,
        unsigned int n_batches, unsigned int input_rows, unsigned int input_cols,
        unsigned int input_channels,
        unsigned int output_rows, unsigned int output_cols,
        unsigned int  channel_multiplier,
        PaddingValues padding, arm_gemm::Activation activation,
        const DepthwiseConfig *config)
        : DepthwiseArgs(cpu_info, kernel_rows, kernel_cols, stride_rows,
                        stride_cols, 1, 1, n_batches, input_rows, input_cols,
                        input_channels, output_rows, output_cols,
                        channel_multiplier, padding, activation, config)
    {
    }
};

template <typename TInput>
struct Tile
{
    TInput *array;

    unsigned int tile_rows     = 0;
    unsigned int tile_cols     = 0;
    unsigned int tile_channels = 0;

    Tile(TInput *array, unsigned int tile_rows, unsigned int tile_cols, unsigned int tile_channels)
        : array(array), tile_rows(tile_rows), tile_cols(tile_cols), tile_channels(tile_channels)
    {
    }

    Tile()
        : Tile(nullptr, 0, 0, 0)
    {
    }

    void load_from(
        const TInput      *input,
        const unsigned int ld_row, const unsigned int ld_col,
        const unsigned int n_rows, const unsigned int n_cols,
        const int input_i, const int input_j,
        const unsigned int channel_multiplier) const
    {
        const auto pad_top  = input_i < 0 ? -input_i : 0;
        const auto pad_left = input_j < 0 ? -input_j : 0;

        const auto padded_rows = std::min(n_rows - input_i, tile_rows) - pad_top;
        const auto padded_cols = std::min(n_cols - input_j, tile_cols) - pad_left;

        if(padded_rows < tile_rows || padded_cols < tile_cols)
        {
            memset(array, 0, tile_rows * tile_cols * tile_channels * sizeof(TInput));
        }

        do_premultiply<TInput>(
            (TInput *)input + std::max(input_i, 0) * ld_row + std::max(input_j, 0) * ld_col,
            ld_row, ld_col,
            array + pad_top * tile_cols * tile_channels + pad_left * tile_channels,
            tile_cols * tile_channels, tile_channels,
            padded_rows, padded_cols, tile_channels / channel_multiplier,
            channel_multiplier);
    }
};

template <typename TInput, typename TWeight, typename TOutput>
class DepthwiseCommon : public IDepthwiseCommon
{
protected:
    const DepthwiseArgs m_args; // Copy of arguments
    std::string         m_name{};

public:
    DepthwiseCommon(const DepthwiseArgs &args)
        : m_args(args) {};
    DepthwiseCommon(DepthwiseCommon &) = delete;
    DepthwiseCommon &operator=(DepthwiseCommon &) = delete;

    std::string name() const override
    {
        return m_name;
    }

    void set_name(std::string name)
    {
        // Only allow the name to be set once
        if(m_name.empty())
        {
            m_name = name;
        }
    }

    void execute(
        const void *const  input,
        const void *const  parameters,
        void *const        output,
        void *const        working_space,
        const unsigned int thread_id,
        const unsigned int n_threads) const override final
    {
        const size_t ld_input_col    = m_args.input_channels;
        const size_t ld_input_row    = ld_input_col * m_args.input_cols;
        const size_t ld_input_batch  = ld_input_row * m_args.input_rows;
        const size_t ld_output_col   = m_args.input_channels * m_args.channel_multiplier;
        const size_t ld_output_row   = ld_output_col * m_args.output_cols;
        const size_t ld_output_batch = ld_output_row * m_args.output_rows;

        execute(
            input, ld_input_col, ld_input_row, ld_input_batch,
            parameters, output, ld_output_col, ld_output_row, ld_output_batch,
            working_space, thread_id, n_threads);
    }

    void execute(
        const void *const  input,
        size_t             ld_input_col,
        size_t             ld_input_row,
        size_t             ld_input_batch,
        const void *const  parameters,
        void *const        output,
        size_t             ld_output_col,
        size_t             ld_output_row,
        size_t             ld_output_batch,
        void *const        working_space,
        const unsigned int thread_id,
        const unsigned int n_threads) const override final
    {
        execute(
            m_args.n_batches, m_args.input_rows, m_args.input_cols,
            m_args.input_channels, m_args.padding,
            input, ld_input_col, ld_input_row, ld_input_batch,
            parameters,
            m_args.output_rows, m_args.output_cols,
            output, ld_output_col, ld_output_row, ld_output_batch,
            working_space, thread_id, n_threads);
    }

    void execute(
        unsigned int         batches,
        unsigned int         input_height,
        unsigned int         input_width,
        unsigned int         channels,
        const PaddingValues &padding,
        const void          *input,
        size_t               ld_input_col,
        size_t               ld_input_row,
        size_t               ld_input_batch,
        const void          *parameters,
        unsigned int         output_height,
        unsigned int         output_width,
        void                *output,
        size_t               ld_output_col,
        size_t               ld_output_row,
        size_t               ld_output_batch,
        void                *working_space,
        unsigned int         thread_id,
        unsigned int         n_threads) const override final
    {
        // Construct a new set of arguments to reflect that we might have been
        // passed different input/output tensors. Dilation is handled at this
        // level; so we set the dilation in the arguments to zero.
        DepthwiseArgs args(this->m_args);
        args.n_batches      = batches;
        args.input_rows     = input_height;
        args.input_cols     = input_width;
        args.input_channels = channels;
        args.output_rows    = output_height;
        args.output_cols    = output_width;
        args.padding        = padding;
        args.dilation_rows = args.dilation_cols = 1;

        auto ld_input_col_d  = ld_input_col * m_args.dilation_cols;
        auto ld_input_row_d  = ld_input_row * m_args.dilation_rows;
        auto ld_output_col_d = ld_output_col * m_args.dilation_cols;
        auto ld_output_row_d = ld_output_row * m_args.dilation_rows;

        for(size_t drow = 0; drow < m_args.dilation_rows; drow++)
        {
            size_t start_i;
            std::tie(args.output_rows, args.input_rows, start_i,
                     args.padding.top, args.padding.bottom) =
                         get_reduced_view_for_dilation(
                             output_height, input_height, drow, m_args.dilation_rows,
                             m_args.kernel_rows, m_args.stride_rows, padding.top);

            auto input_row  = static_cast<const TInput *>(input) + start_i * ld_input_row;
            auto output_row = static_cast<TOutput *>(output) + drow * ld_output_row;

            if(args.output_rows)
            {
                for(size_t dcol = 0; dcol < m_args.dilation_cols; dcol++)
                {
                    size_t start_j;
                    std::tie(args.output_cols, args.input_cols, start_j,
                             args.padding.left, args.padding.right) =
                                 get_reduced_view_for_dilation(
                                     output_width, input_width, dcol, m_args.dilation_cols,
                                     m_args.kernel_cols, m_args.stride_cols, padding.left);

                    const TInput *input_col  = input_row + start_j * ld_input_col;
                    TOutput      *output_col = output_row + dcol * ld_output_col;

                    if(args.output_cols)
                    {
                        this->execute_internal(
                            args, input_col, ld_input_col_d, ld_input_row_d, ld_input_batch, parameters,
                            output_col, ld_output_col_d, ld_output_row_d, ld_output_batch,
                            working_space, thread_id, n_threads);
                    }
                }
            }
        }
    }

protected:
    virtual void execute_internal(
        const DepthwiseArgs &instance_args,
        const void          *input,
        size_t               ld_input_col,
        size_t               ld_input_row,
        size_t               ld_input_batch,
        const void          *parameters,
        void                *output,
        size_t               ld_output_col,
        size_t               ld_output_row,
        size_t               ld_output_batch,
        void                *working_space,
        unsigned int         thread_id,
        unsigned int         n_threads) const = 0;

    virtual bool uses_premultiply() const
    {
        return true;
    }
};

template <typename TInput, typename TWeight = TInput, typename TOutput = TInput>
using UniqueDepthwiseCommon = std::unique_ptr<DepthwiseCommon<TInput, TWeight, TOutput>>;

template <typename TInput, typename TWeight = TInput, typename TOutput = TInput, class OutputStage = Nothing>
KernelDescription get_depthwise_method(const DepthwiseArgs &, const OutputStage & = {});

template <typename TInput, typename TWeight = TInput, typename TOutput = TInput, class OutputStage = Nothing>
UniqueDepthwiseCommon<TInput, TWeight, TOutput> depthwise(const DepthwiseArgs &, const OutputStage & = {});

template <typename TInput, typename TWeight = TInput, typename TOutput = TInput, class OutputStage = Nothing>
std::vector<KernelDescription> get_compatible_kernels(const DepthwiseArgs &, const OutputStage & = {});

} // namespace depthwise
} // namespace arm_conv