ug 0.3.0

Micro compiler for tensor operations.
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

use ug::Result;

#[test]
fn eval_add() -> Result<()> {
    let kernel = ug::samples::ssa::simple_add(2)?;
    let mut a = ug::interpreter::Buffer::I32(vec![0i32, 0]);
    let mut b = ug::interpreter::Buffer::I32(vec![3i32, 4]);
    let mut c = ug::interpreter::Buffer::I32(vec![1i32, 2]);
    ug::interpreter::eval_ssa::<1>(&kernel, vec![&mut a, &mut b, &mut c], &[], 0)?;
    let a = a.as_i32()?;
    assert_eq!(a, [4, 6]);
    Ok(())
}

#[test]
fn eval_dotprod() -> Result<()> {
    let kernel = ug::samples::ssa::simple_dotprod(2)?;
    let mut a = ug::interpreter::Buffer::F32(vec![0f32]);
    let mut b = ug::interpreter::Buffer::F32(vec![3f32, 4.5]);
    let mut c = ug::interpreter::Buffer::F32(vec![1f32, 2.]);
    ug::interpreter::eval_ssa::<1>(&kernel, vec![&mut a, &mut b, &mut c], &[], 0)?;
    let a = a.as_f32()?;
    assert_eq!(a, [12.]);
    Ok(())
}

#[test]
fn lower_add() -> Result<()> {
    let kernel = ug::samples::simple_add(2)?;
    let ssa_kernel = kernel.lower()?;
    let mut c = ug::interpreter::Buffer::F32(vec![0f32, 0.]);
    let mut b = ug::interpreter::Buffer::F32(vec![3f32, 4.]);
    let mut a = ug::interpreter::Buffer::F32(vec![1f32, 2.]);
    ug::interpreter::eval_ssa::<1>(&ssa_kernel, vec![&mut a, &mut b, &mut c], &[], 0)?;
    let c = c.as_f32()?;
    assert_eq!(c, [4., 6.]);
    Ok(())
}

#[test]
fn softmax() -> Result<()> {
    let kernel = ug::samples::op::softmax(2, 4)?;
    let ssa_kernel = kernel.lower(&Default::default())?;
    let mut a = ug::interpreter::Buffer::F32(vec![0., 1., 2., 3., 2., 1., 2., 1.]);
    let mut b = ug::interpreter::Buffer::F32(vec![0f32; 8]);
    ug::interpreter::eval_ssa::<1>(&ssa_kernel, vec![&mut a, &mut b], &[], 0)?;
    let b = b.as_f32()?;
    assert_eq!(
        b,
        [
            0.032058604,
            0.08714432,
            0.23688284,
            0.6439143,
            0.3655293,
            0.13447072,
            0.3655293,
            0.13447072
        ]
    );
    Ok(())
}

#[test]
fn ssa_softmax() -> Result<()> {
    const N_COLS: usize = 4;
    const N_ROWS: usize = 2;
    let ssa_kernel = ug::samples::ssa::softmax(N_ROWS, N_COLS)?;
    let mut a = ug::interpreter::Buffer::F32(vec![0., 1., 2., 3., 2., 1., 2., 1.]);
    let mut b = ug::interpreter::Buffer::F32(vec![0f32; N_ROWS * N_COLS]);
    for group_idx in 0..N_ROWS {
        ug::interpreter::eval_ssa::<N_COLS>(&ssa_kernel, vec![&mut a, &mut b], &[], group_idx)?;
    }
    let b = b.as_f32()?;
    assert_eq!(
        b,
        [
            0.032058604,
            0.08714432,
            0.23688284,
            0.6439143,
            0.3655293,
            0.13447072,
            0.3655293,
            0.13447072
        ]
    );
    Ok(())
}

#[test]
fn ssa_softmax_reduce() -> Result<()> {
    const N_COLS: usize = 4;
    const N_ROWS: usize = 2;
    let ssa_kernel = ug::samples::ssa::softmax_reduce(N_ROWS, N_COLS)?;
    let mut a = ug::interpreter::Buffer::F32(vec![0., 1., 2., 3., 2., 1., 2., 1.]);
    let mut b = ug::interpreter::Buffer::F32(vec![0f32; N_ROWS * N_COLS]);
    for group_idx in 0..N_ROWS {
        ug::interpreter::eval_ssa::<N_COLS>(&ssa_kernel, vec![&mut a, &mut b], &[], group_idx)?;
    }
    let b = b.as_f32()?;
    assert_eq!(
        b,
        [
            0.032058604,
            0.08714432,
            0.23688284,
            0.6439143,
            0.3655293,
            0.13447072,
            0.3655293,
            0.13447072
        ]
    );
    Ok(())
}

#[test]
fn affine() -> Result<()> {
    let kernel = {
        use ug::{lang::op, DType, Layout};
        let layout = Layout::from_shape(4);
        let ptr = ug::lang::Arg::ptr(DType::F32);
        let src = op::load(ptr.id(), layout.clone(), DType::F32)?;
        let sub = op::broadcast_binary(op::BinaryOp::Sub, src, op::cst(0.5)?)?;
        let mul = op::broadcast_binary(op::BinaryOp::Mul, sub, op::cst(3.)?)?;
        let st = op::store(ptr.id(), layout, mul)?;
        op::Kernel::new("affine".to_string(), vec![ptr], vec![st])
    };
    let ssa_kernel = kernel.lower(&Default::default())?;
    let mut a = ug::interpreter::Buffer::F32(vec![1., 2., 3., 4.]);
    ug::interpreter::eval_ssa::<4>(&ssa_kernel, vec![&mut a], &[], 0)?;
    let a = a.as_f32()?;
    assert_eq!(a, [1.5, 4.5, 7.5, 10.5]);
    Ok(())
}