cubecl-matmul 0.8.1

use std::marker::PhantomData;

use crate::components::global::read::SyncPartialLoadingStrategy;
use crate::components::global::{RoleRule, read::tiled::TiledLayout};
use crate::components::stage::TilingOrderEnum;
use crate::components::{
    FormattedConfigError, InvalidConfigError, MatmulIdent, MatrixPrecision, TilingScheme,
};
use crate::components::{global::multi_stage::LoadMaxRoundPlaneCount, stage::TilingValidation};
use crate::components::{
    global::{GlobalConfig, memory::GlobalIterator},
    stage::{ContiguousTilingLayout, StridedStage, TilingOrder},
};
use cubecl_core as cubecl;
use cubecl_core::prelude::*;
use cubecl_std::tensor::layout::Coords2d;

use super::{LoadingJob, LoadingValidation};

#[derive(CubeType, Clone, Copy)]
/// Each tile is guaranteed to be loaded entirely by the same plane.
/// Each plane can load multiple tiles, provided the number of planes evenly divides the number of tiles.
/// In this case, a plane loads contiguous tiles following the `TilingOrder`,
/// until it would otherwise write to the opposite stage. At that point, it continues on the next
/// row or column of the same stage, skipping over the memory region of the other stage.
///
/// Only supports RowMajorTilingOrder for Lhs and ColMajorTilingOrder for Rhs
pub struct SyncPartialTilewiseLoading<T: TilingOrder> {
    #[cube(comptime)]
    tiling_order: PhantomData<T>,
}

impl<TO: TilingOrder> LoadMaxRoundPlaneCount for SyncPartialTilewiseLoading<TO> {
    fn max_round_plane_count(
        tiling_scheme: &TilingScheme,
        ident: MatmulIdent,
        _line_size: u8,
        _plane_dim: u32,
    ) -> u32 {
        tiling_scheme.tiles_in_stage(ident)
    }
}

impl<T: TilingOrder> LoadingValidation for SyncPartialTilewiseLoading<T> {
    fn check<C: GlobalConfig>(config: &C, ident: MatmulIdent) -> Result<(), InvalidConfigError> {
        let line_size = config.global_line_size(ident);
        let num_planes = config.num_loading_planes(ident);
        let num_tiles = config.tiling_scheme().tiles_in_stage(ident);

        if !num_tiles.is_multiple_of(num_planes) {
            return Err(FormattedConfigError::new(move || {
                "Number of planes {num_planes:?} must divide number of tiles {num_tiles:?} for tilewise loading.".to_string()
            }));
        }

        let num_tiles_per_plane = comptime!(num_tiles / num_planes);
        let num_lines_per_tile =
            comptime!(config.tiling_scheme().elements_in_tile(ident) / line_size);
        let num_lines_per_plane = num_lines_per_tile * num_tiles_per_plane;
        let num_planes = config.plane_dim();

        if num_lines_per_plane % num_planes != 0 {
            return Err(FormattedConfigError::new(move || {
                "Number of planes {num_planes:?} must divide number of lines per plane {num_lines_per_plane:?} for tilewise loading.".to_string()
            }));
        }

        match ident {
            MatmulIdent::Lhs => {
                if !matches!(T::to_enum(), TilingOrderEnum::RowMajor) {
                    return Err(FormattedConfigError::new(move || {
                        "Sync partial tilewise on Lhs is only supported with RowMajor tiling order"
                            .to_string()
                    }));
                }
            }
            MatmulIdent::Rhs => {
                if !matches!(T::to_enum(), TilingOrderEnum::ColMajor) {
                    return Err(FormattedConfigError::new(move || {
                        "Sync partial tilewise on Rhs is only supported with ColMajor tiling order"
                            .to_string()
                    }));
                }
            }
            MatmulIdent::Out => unreachable!(),
        }

        ContiguousTilingLayout::<T>::check(config.global_memory_config(ident))?;

        Ok(())
    }
}

#[cube]
impl<TO: TilingOrder> SyncPartialLoadingStrategy for SyncPartialTilewiseLoading<TO> {
    type TilingLayout = ContiguousTilingLayout<TO>;
    type Job<IP: MatrixPrecision> = SyncPartialTilewiseJob;

    fn new_job<IP: MatrixPrecision, G: GlobalConfig>(
        #[comptime] stage_index: u32,
        #[comptime] ident: MatmulIdent,
        #[comptime] line_size: u32,
        #[comptime] config: G,
    ) -> SyncPartialTilewiseJob {
        let num_planes = config.num_loading_planes(ident);
        let num_tiles = config.tiling_scheme().tiles_in_stage(ident);
        let plane_dim = config.plane_dim();

        let num_tiles_per_plane = comptime!(num_tiles / num_planes);
        let num_lines_per_tile =
            comptime!(config.tiling_scheme().elements_in_tile(ident) / line_size);
        let num_lines_per_plane = num_lines_per_tile * num_tiles_per_plane;
        let num_lines_per_unit = num_lines_per_plane / plane_dim;

        let num_stages = config.num_stages(ident);
        let stage_width = comptime!(match ident {
            MatmulIdent::Lhs => config.tiling_scheme().tiles_in_stage_col(ident),
            MatmulIdent::Rhs => config.tiling_scheme().tiles_in_stage_row(ident),
            MatmulIdent::Out => unreachable!(),
        });
        let row_col_stride = num_stages * stage_width;
        let stage_offset = stage_width * stage_index;

        let starting_tile_within_stage = RoleRule::new(config.role_rule_config())
            .load_index(ident, config.specialized_loading_sides())
            * num_tiles_per_plane;
        let row_col_index = starting_tile_within_stage / stage_width;
        let inner_offset = starting_tile_within_stage % stage_width;
        let num_tiles_to_skip = row_col_index * row_col_stride + inner_offset + stage_offset;

        SyncPartialTilewiseJob {
            num_tiles_to_skip,
            row_col_stride,
            stage_width,
            num_lines_per_tile,
            num_lines_per_unit,
            plane_dim: config.plane_dim(),
            line_size,
            ident,
        }
    }
}

#[derive(CubeType, Clone, Copy)]
pub struct SyncPartialTilewiseJob {
    num_tiles_to_skip: u32,

    #[cube(comptime)]
    row_col_stride: u32,
    #[cube(comptime)]
    stage_width: u32,
    #[cube(comptime)]
    num_lines_per_tile: u32,
    #[cube(comptime)]
    num_lines_per_unit: u32,
    #[cube(comptime)]
    plane_dim: u32,
    #[cube(comptime)]
    line_size: u32,
    #[cube(comptime)]
    ident: MatmulIdent,
}

#[cube]
impl<IP: MatrixPrecision, TO: TilingOrder> LoadingJob<IP, ContiguousTilingLayout<TO>>
    for SyncPartialTilewiseJob
{
    fn execute_task<G: GlobalConfig>(
        this: &mut Self,
        #[comptime] task_id: u32,
        global_iter: &GlobalIterator<Line<IP::Global>>,
        stage: &mut StridedStage<IP::Stage, ContiguousTilingLayout<TO>>,
        #[comptime] config: G,
    ) {
        let pos_across_tiles = task_id * this.plane_dim + UNIT_POS_X;
        let nth_tile_for_this_plane = pos_across_tiles / this.num_lines_per_tile;
        let line_index_within_tile = pos_across_tiles % this.num_lines_per_tile;

        let row_col_index_local = nth_tile_for_this_plane / this.stage_width;
        let inner_offset = nth_tile_for_this_plane % this.stage_width;
        let num_tiles_to_skip_local = row_col_index_local * this.row_col_stride + inner_offset;
        let nth_tile_global = this.num_tiles_to_skip + num_tiles_to_skip_local;

        let (total_tile_count_row, total_tile_count_col) = match comptime!(this.ident) {
            MatmulIdent::Lhs => (
                comptime!(config.tiling_scheme().tiles_in_stage_m()),
                comptime!(
                    config.tiling_scheme().tiles_in_stage_k() * config.num_stages(MatmulIdent::Lhs)
                ),
            ),
            MatmulIdent::Rhs => (
                comptime!(
                    config.tiling_scheme().tiles_in_stage_k() * config.num_stages(MatmulIdent::Rhs)
                ),
                comptime!(config.tiling_scheme().tiles_in_stage_n()),
            ),
            MatmulIdent::Out => comptime!(unreachable!()),
        };

        let tile = TO::to_row_col(
            nth_tile_global,
            total_tile_count_row,
            total_tile_count_col,
            comptime!(config.stage_memory_config(this.ident)),
        );

        let num_lines_to_skip_global = nth_tile_global * this.num_lines_per_tile;

        SyncPartialTilewiseJob::load_and_store_line::<IP, TO, G>(
            this,
            tile,
            line_index_within_tile,
            num_lines_to_skip_global,
            global_iter,
            stage,
            config,
        );
    }

    fn task_count(this: &Self) -> comptime_type!(u32) {
        comptime!(this.num_lines_per_unit)
    }
}

#[cube]
impl SyncPartialTilewiseJob {
    #[allow(clippy::too_many_arguments)]
    fn load_and_store_line<IP: MatrixPrecision, TO: TilingOrder, G: GlobalConfig>(
        this: &Self,
        tile: Coords2d,
        line_index_within_tile: u32,
        num_lines_to_skip_global: u32,
        global_iter: &GlobalIterator<Line<IP::Global>>,
        stage: &mut StridedStage<IP::Stage, ContiguousTilingLayout<TO>>,
        #[comptime] config: G,
    ) {
        let layout = TiledLayout::new(comptime!(config.global_memory_config(this.ident)));
        let view = global_iter.view().view(layout);

        let line_read = view.read_checked((tile, line_index_within_tile * this.line_size));

        let offset = line_index_within_tile + num_lines_to_skip_global;

        stage.as_slice_mut(this.line_size)[offset] = Line::cast_from(line_read);
    }
}