gemmkit 0.1.0

A clean, extensible, high-performance GEMM (general matrix multiply) 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
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141

//! Tiny 2x2 GEMM entry point per element-type dispatch ladder, shared by the
//! `GEMMKIT_REQUIRE_ISA` cross-architecture pin binaries (`env_isa_neon`, `env_isa_wasm`)
//!
//! Each function below drives one dtype's public GEMM entry through a fixed 2x2 product; the
//! pin binaries wrap them in `catch_unwind` to observe whether the pinned ISA's `select_*`
//! ladder accepts the current target (returns normally) or rejects it (panics). A dtype whose
//! feature is off is simply missing from [`dtype_cases`]. Lives in a subdirectory, reached with
//! a plain `mod isa_dtypes;`, so cargo's default test-discovery does not also build it as its
//! own top-level test binary

use gemmkit::{MatMut, MatRef, Parallelism};

fn gemm_f32() {
    let a = [1.0f32, 2.0, 3.0, 4.0];
    let b = [5.0f32, 6.0, 7.0, 8.0];
    let mut c = [0.0f32; 4];
    gemmkit::gemm(
        1.0,
        MatRef::from_row_major(&a, 2, 2),
        MatRef::from_row_major(&b, 2, 2),
        0.0,
        MatMut::from_row_major(&mut c, 2, 2),
        Parallelism::Serial,
    );
}

fn gemm_f64() {
    let a = [1.0f64, 2.0, 3.0, 4.0];
    let b = [5.0f64, 6.0, 7.0, 8.0];
    let mut c = [0.0f64; 4];
    gemmkit::gemm(
        1.0,
        MatRef::from_row_major(&a, 2, 2),
        MatRef::from_row_major(&b, 2, 2),
        0.0,
        MatMut::from_row_major(&mut c, 2, 2),
        Parallelism::Serial,
    );
}

#[cfg(feature = "half")]
fn gemm_f16() {
    use gemmkit::f16;
    let a = [f16::from_f32(1.0); 4];
    let b = [f16::from_f32(2.0); 4];
    let mut c = [f16::from_f32(0.0); 4];
    gemmkit::gemm(
        f16::from_f32(1.0),
        MatRef::from_row_major(&a, 2, 2),
        MatRef::from_row_major(&b, 2, 2),
        f16::from_f32(0.0),
        MatMut::from_row_major(&mut c, 2, 2),
        Parallelism::Serial,
    );
}

#[cfg(feature = "half")]
fn gemm_bf16() {
    use gemmkit::bf16;
    let a = [bf16::from_f32(1.0); 4];
    let b = [bf16::from_f32(2.0); 4];
    let mut c = [bf16::from_f32(0.0); 4];
    gemmkit::gemm(
        bf16::from_f32(1.0),
        MatRef::from_row_major(&a, 2, 2),
        MatRef::from_row_major(&b, 2, 2),
        bf16::from_f32(0.0),
        MatMut::from_row_major(&mut c, 2, 2),
        Parallelism::Serial,
    );
}

#[cfg(feature = "int8")]
fn gemm_i8() {
    let a = [1i8, 2, 3, 4];
    let b = [5i8, 6, 7, 8];
    let mut c = [0i32; 4];
    gemmkit::gemm_i8(
        1,
        MatRef::from_row_major(&a, 2, 2),
        MatRef::from_row_major(&b, 2, 2),
        0,
        MatMut::from_row_major(&mut c, 2, 2),
        Parallelism::Serial,
    );
}

#[cfg(feature = "complex")]
fn gemm_c32() {
    use gemmkit::Complex;
    let a = [Complex::new(1.0f32, 0.5); 4];
    let b = [Complex::new(2.0f32, -0.5); 4];
    let mut c = [Complex::new(0.0f32, 0.0); 4];
    gemmkit::gemm_cplx(
        Complex::new(1.0, 0.0),
        MatRef::from_row_major(&a, 2, 2),
        false,
        MatRef::from_row_major(&b, 2, 2),
        false,
        Complex::new(0.0, 0.0),
        MatMut::from_row_major(&mut c, 2, 2),
        Parallelism::Serial,
    );
}

#[cfg(feature = "complex")]
fn gemm_c64() {
    use gemmkit::Complex;
    let a = [Complex::new(1.0f64, 0.5); 4];
    let b = [Complex::new(2.0f64, -0.5); 4];
    let mut c = [Complex::new(0.0f64, 0.0); 4];
    gemmkit::gemm_cplx(
        Complex::new(1.0, 0.0),
        MatRef::from_row_major(&a, 2, 2),
        false,
        MatRef::from_row_major(&b, 2, 2),
        false,
        Complex::new(0.0, 0.0),
        MatMut::from_row_major(&mut c, 2, 2),
        Parallelism::Serial,
    );
}

/// Every element-type dispatch ladder available under the active feature set, paired with its
/// entry point. `f32` and `f64` are unconditional; the rest are gated by their own feature
pub fn dtype_cases() -> Vec<(&'static str, fn())> {
    vec![
        ("f32", gemm_f32 as fn()),
        ("f64", gemm_f64 as fn()),
        #[cfg(feature = "half")]
        ("f16", gemm_f16 as fn()),
        #[cfg(feature = "half")]
        ("bf16", gemm_bf16 as fn()),
        #[cfg(feature = "int8")]
        ("i8", gemm_i8 as fn()),
        #[cfg(feature = "complex")]
        ("c32", gemm_c32 as fn()),
        #[cfg(feature = "complex")]
        ("c64", gemm_c64 as fn()),
    ]
}