cubecl-matmul 0.8.1

CubeCL Matrix Multiplication 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

use crate::components::{
    MatmulIdent, MatmulLineSizes, TilingScheme,
    error::MatmulSetupError,
    global::{GlobalConfig, multi_stage::LoadMaxRoundPlaneCount},
};

pub(crate) fn shared_global_config_validation<G: GlobalConfig>(
    config: G,
) -> Result<G, MatmulSetupError> {
    #[cfg(target_os = "macos")]
    {
        let cube_dim = config.cube_dim();
        if cube_dim.num_elems() >= 512 {
            use crate::components::error::MatmulAvailabilityError;

            return Err(MatmulSetupError::Unavailable(
                MatmulAvailabilityError::CubeDimTooBig(cube_dim),
            ));
        }
    }

    Ok(config)
}

/// Maximal number of planes each reader can handle to divide its workload evenly
pub struct MaxGlobalReaderPlanes {
    pub lhs: u32,
    pub rhs: u32,
}

impl MaxGlobalReaderPlanes {
    /// Create a MaxGlobalReaderPlanes
    pub fn new<LL: LoadMaxRoundPlaneCount, RL: LoadMaxRoundPlaneCount>(
        tiling_scheme: &TilingScheme,
        line_sizes: &MatmulLineSizes,
        plane_dim: u32,
    ) -> Self {
        MaxGlobalReaderPlanes {
            lhs: LL::max_round_plane_count(
                tiling_scheme,
                MatmulIdent::Lhs,
                line_sizes.lhs,
                plane_dim,
            ),
            rhs: RL::max_round_plane_count(
                tiling_scheme,
                MatmulIdent::Rhs,
                line_sizes.rhs,
                plane_dim,
            ),
        }
    }
}