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
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175

use burn_compute::tune::{AutotuneOperation, AutotuneOperationSet};
use burn_tensor::{Element, ElementConversion};

use crate::{
    compute::WgpuAutotuneKey,
    element::WgpuElement,
    kernel::{matmul::utils::init_matmul_output, prng::random_like_uniform},
    ops::numeric::empty_device,
    tensor::WgpuTensor,
};

use super::key::MatmulAutotuneKey;

/// Set of matmul implementations available for autotune
/// Autotune key is given by concatenating the closest upper power of 2 of m, k and n
pub struct MatmulAutotuneOperationSet<E: WgpuElement, const D: usize> {
    key: WgpuAutotuneKey,
    lhs: WgpuTensor<E, D>,
    rhs: WgpuTensor<E, D>,
    out: WgpuTensor<E, D>,
}
impl<E: WgpuElement, const D: usize> MatmulAutotuneOperationSet<E, D> {
    fn new(lhs: WgpuTensor<E, D>, rhs: WgpuTensor<E, D>, out: WgpuTensor<E, D>) -> Self {
        Self {
            key: WgpuAutotuneKey::Matmul(MatmulAutotuneKey::new(&lhs.shape, &rhs.shape)),
            lhs,
            rhs,
            out,
        }
    }
}

impl<E: WgpuElement + Element, const D: usize> AutotuneOperationSet<WgpuAutotuneKey>
    for MatmulAutotuneOperationSet<E, D>
{
    fn key(&self) -> WgpuAutotuneKey {
        self.key.clone()
    }

    fn autotunables(&self) -> Vec<Box<dyn AutotuneOperation>> {
        let random_bounds: (E, E) = ((-10.0).elem::<E>(), (10.0).elem::<E>());
        let lhs = random_like_uniform(&self.lhs, random_bounds.0, random_bounds.1);
        let rhs = random_like_uniform(&self.rhs, random_bounds.0, random_bounds.1);

        let out = empty_device(
            self.out.client.clone(),
            self.out.device.clone(),
            self.out.shape.clone(),
        );

        vec![
            Box::new(MemoryCoalescingMatmulDefault::<E, D>::new(
                lhs.clone(),
                rhs.clone(),
                out.clone(),
            )),
            Box::new(MemoryCoalescingMatmulW16x16::<E, D>::new(
                lhs.clone(),
                rhs.clone(),
                out.clone(),
            )),
            Box::new(Vec4TilingMatmulDefault::<E, D>::new(
                lhs.clone(),
                rhs.clone(),
                out.clone(),
            )),
            Box::new(Vec4TilingMatmulUnpaddedDefault::<E, D>::new(
                lhs.clone(),
                rhs.clone(),
                out.clone(),
            )),
            Box::new(Vec4LhsOnlyTilingMatmulDefault::<E, D>::new(
                lhs.clone(),
                rhs.clone(),
                out.clone(),
            )),
        ]
    }

    fn fastest(self: Box<Self>, fastest_index: usize) -> Box<dyn AutotuneOperation> {
        match fastest_index {
            0 => Box::new(MemoryCoalescingMatmulDefault::<E, D>::new(
                self.lhs, self.rhs, self.out,
            )),
            1 => Box::new(MemoryCoalescingMatmulW16x16::<E, D>::new(
                self.lhs, self.rhs, self.out,
            )),
            2 => Box::new(Vec4TilingMatmulDefault::<E, D>::new(
                self.lhs, self.rhs, self.out,
            )),
            3 => Box::new(Vec4TilingMatmulUnpaddedDefault::<E, D>::new(
                self.lhs, self.rhs, self.out,
            )),
            4 => Box::new(Vec4LhsOnlyTilingMatmulDefault::<E, D>::new(
                self.lhs, self.rhs, self.out,
            )),
            _ => panic!("Fastest index is out of bound"),
        }
    }
}

/// Executes autotune on matmul operations
pub fn matmul_autotune<E: WgpuElement + Element, const D: usize>(
    lhs: WgpuTensor<E, D>,
    rhs: WgpuTensor<E, D>,
) -> WgpuTensor<E, D> {
    let client = lhs.client.clone();

    let output = init_matmul_output(&lhs, &rhs);

    let operation_set = Box::new(MatmulAutotuneOperationSet::<E, D>::new(
        lhs,
        rhs,
        output.clone(),
    ));

    client.execute_autotune(operation_set);

    output
}

macro_rules! matmul_tune_ops {
    ($name:ident, $func:expr) => {
        #[derive(new)]
        pub(crate) struct $name<E: WgpuElement, const D: usize> {
            lhs: WgpuTensor<E, D>,
            rhs: WgpuTensor<E, D>,
            out: WgpuTensor<E, D>,
        }

        impl<E: WgpuElement, const D: usize> AutotuneOperation for $name<E, D> {
            fn execute(self: Box<Self>) {
                #[allow(clippy::redundant_closure_call)]
                $func(self.lhs, self.rhs, self.out);
            }

            fn clone(&self) -> Box<dyn AutotuneOperation> {
                Box::new(Self {
                    lhs: self.lhs.clone(),
                    rhs: self.rhs.clone(),
                    out: self.out.clone(),
                })
            }
        }
    };
}

// Potentially better for small matrices.
matmul_tune_ops!(
    MemoryCoalescingMatmulDefault,
    crate::kernel::matmul::matmul_mem_coalescing_default
);

// Potentially better for small matrices.
matmul_tune_ops!(MemoryCoalescingMatmulW16x16, |lhs, rhs, out| {
    crate::kernel::matmul::matmul_mem_coalescing(lhs, rhs, out, 16, 16)
});

// Maybe the fastest on MacOS.
matmul_tune_ops!(
    Vec4LhsOnlyTilingMatmulDefault,
    crate::kernel::matmul::vec4_lhs::matmul_tiling_2d_vec4_lhs
);

// Probably the fastest when fixed sizes.
matmul_tune_ops!(
    Vec4TilingMatmulDefault,
    crate::kernel::matmul::vec4::matmul_tiling_2d_vec4
);

// Probably the fastest otherwise.
matmul_tune_ops!(
    Vec4TilingMatmulUnpaddedDefault,
    crate::kernel::matmul::unpadded::matmul_tiling_2d_unpadded
);