cubecl-convolution 0.7.0

CubeCL Convolution Kernels Engine
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118

use cubecl::prelude::*;
use cubecl_core::{self as cubecl};
use cubecl_matmul::components::{
    MatmulIdent,
    global::{GlobalConfig, memory::GlobalMemoryConfig},
};
use cubecl_std::{
    FastDivmod, FastDivmodArgs,
    tensor::layout::{Coords3d, Layout, LayoutExpand},
};

use crate::{
    components::{
        ConvGemmConfig, ConvolutionConfig, ConvolutionParams, ConvolutionProblem,
        global::layout::NhwcCoords,
    },
    kernels::layered::selector::RuntimeArgs,
};

/// Maps a 4D weight tensor of shape `(out_c, (k_h, k_w, in_c))` to a col-major 2D matmul tile with
/// shape `(n, k)`
#[derive(CubeType, CubeLaunch, Clone)]
pub struct WeightLayout {
    /// Number of channels, including padding, used for decomposing `k`
    pub channels: FastDivmod,

    /// Shape of the conceptual `k` size, including padding
    pub shape_k: u32,
    /// Shape of the conceptual `n` size, or `out_c`
    pub shape_n: u32,

    /// Size of the convolution kernel
    #[cube(comptime)]
    pub params: ConvolutionParams,
    /// Global memory config for the backing tensor
    #[cube(comptime)]
    pub config: GlobalMemoryConfig,
}

#[cube]
impl WeightLayout {
    pub fn new<E: Numeric, G: GlobalConfig>(
        args: &RuntimeArgs,
        #[comptime] config: ConvolutionConfig<G>,
    ) -> WeightLayout {
        WeightLayout {
            shape_k: args.shape_k,
            shape_n: args.shape_n,
            channels: args.padded_channels,
            params: config.convolution_params(),
            config: config.global_memory_config(MatmulIdent::Rhs),
        }
    }
}

#[cube]
impl Layout for WeightLayout {
    type Coordinates = Coords3d;
    type SourceCoordinates = NhwcCoords;

    fn to_source_pos(&self, coords: Self::Coordinates) -> NhwcCoords {
        let params = comptime![self.params];
        let (_, k, n) = coords;

        let (mut rem, in_c) = self.channels.div_mod(k);

        let spatial_dims = comptime![params.dimensionality.num_dims()];
        let mut kernel_pos = Sequence::<i32>::new();

        #[unroll]
        for i in 0..spatial_dims {
            let dim = comptime![spatial_dims - i - 1];
            let ksize = comptime![params.kernel_size[dim as usize]];
            let k_pos = rem % ksize;
            rem /= ksize;

            kernel_pos.push(k_pos as i32);
        }

        let kernel_pos = kernel_pos.rev();

        NhwcCoords {
            batch: n,
            spatial: kernel_pos,
            channel: in_c,
        }
    }

    fn to_source_pos_checked(&self, coords: Self::Coordinates) -> (NhwcCoords, bool) {
        (self.to_source_pos(coords), self.is_in_bounds(coords))
    }

    fn shape(&self) -> Self::Coordinates {
        (1, self.shape_k, self.shape_n)
    }

    fn is_in_bounds(&self, pos: Self::Coordinates) -> bool {
        let (_, k, n) = pos;
        let check_k = comptime![self.config.check_row_bounds];
        let check_n = comptime![self.config.check_col_bounds];
        (!check_k || k < self.shape_k) && (!check_n || n < self.shape_n)
    }
}

impl<'a, R: Runtime> WeightLayoutLaunch<'a, R> {
    pub fn from_args(
        client: &ComputeClient<R::Server>,
        problem: &ConvolutionProblem,
        params: ConvolutionParams,
        config: GlobalMemoryConfig,
    ) -> Self {
        let channels = FastDivmodArgs::new(client, problem.channels as u32);
        let shape_k = ScalarArg::new(problem.k as u32);
        let shape_n = ScalarArg::new(problem.n as u32);

        WeightLayoutLaunch::new(channels, shape_k, shape_n, params, config)
    }
}