cubek-reduce 0.2.0-pre.2

CubeK: Reduce Kernels
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

use cubecl::{flex32, prelude::Numeric};

/// Precision used for the reduction.
pub trait ReducePrecision {
    /// Precision used for the input tensor.
    type EI: Numeric;
    /// Precision used for the accumulation.
    type EA: Numeric;
}

impl<EI: Numeric, EA: Numeric> ReducePrecision for (EI, EA) {
    type EI = EI;
    type EA = EA;
}

// The below implementations are suggestion for reduction that can accumulate precision errors like
// summations.

impl ReducePrecision for f64 {
    type EI = f64;
    type EA = f64;
}

impl ReducePrecision for f32 {
    type EI = f32;
    type EA = f32;
}

impl ReducePrecision for flex32 {
    type EI = f32;
    type EA = f32;
}

impl ReducePrecision for half::f16 {
    type EI = half::f16;
    type EA = f32;
}

impl ReducePrecision for half::bf16 {
    type EI = half::bf16;
    type EA = f32;
}

impl ReducePrecision for i64 {
    type EI = i64;
    type EA = i64;
}

impl ReducePrecision for i32 {
    type EI = i32;
    type EA = i32;
}

impl ReducePrecision for i16 {
    type EI = i16;
    type EA = i32;
}

impl ReducePrecision for i8 {
    type EI = i8;
    type EA = i32;
}

impl ReducePrecision for u64 {
    type EI = u64;
    type EA = u64;
}

impl ReducePrecision for u32 {
    type EI = u32;
    type EA = u32;
}

impl ReducePrecision for u16 {
    type EI = u16;
    type EA = u32;
}

impl ReducePrecision for u8 {
    type EI = u8;
    type EA = u32;
}