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

use std::sync::Arc;
use std::time::Instant;

use ha_ndarray::*;

const ITERATIONS: usize = 10;

fn broadcast_and_multiply(context: &Context) -> Result<(), Error> {
    for m in 0..4 {
        let dim = 10usize.pow(m);
        let shape = vec![3, dim, 5, 10];
        let size = shape.iter().product::<usize>();
        let queue = Queue::new(context.clone(), size)?;

        let left = ArrayBase::<Vec<_>>::with_context(
            context.clone(),
            vec![dim, 5, 10],
            vec![1.0f64; dim * 5 * 10],
        )?;

        let right = ArrayBase::<Vec<_>>::with_context(
            context.clone(),
            vec![3, dim, 1, 10],
            vec![1.0f64; 3 * dim * 10],
        )?;

        println!(
            "broadcast and multiply {:?} and {:?} (size {})...",
            left, right, size
        );

        let product = left.broadcast(shape.to_vec())? * right.broadcast(shape)?;

        for _ in 0..ITERATIONS {
            let start = Instant::now();
            product.read(&queue)?;
            let duration = start.elapsed();
            println!("{:?} us", duration.as_micros());
        }
    }

    Ok(())
}

fn matmul(context: &Context) -> Result<(), Error> {
    for m in 1..16usize {
        let dim = 2usize.pow(m as u32);

        let l = ArrayBase::<Vec<_>>::with_context(
            context.clone(),
            vec![16 * m, dim],
            vec![1.0f32; 16 * m * dim],
        )?;

        let r = ArrayBase::<Vec<_>>::with_context(
            context.clone(),
            vec![dim, m * 32],
            vec![1.0f32; dim * m * 32],
        )?;

        let num_ops = 16 * 32 * dim;
        println!("matmul {:?} with {:?} ({} ops)", l, r, num_ops);

        let x = l.matmul(r)?;
        let queue = Queue::new(context.clone(), x.size())?;

        for _ in 0..ITERATIONS {
            let start = Instant::now();
            let _output = x.read(&queue)?;
            let duration = start.elapsed();
            let rate = num_ops as f32 / duration.as_secs_f32();
            println!("{:?} us @ {} M/s", duration.as_micros(), rate / 1_000_000.);
        }
    }

    Ok(())
}

fn reduce_sum_axis(context: &Context) -> Result<(), Error> {
    let shape = vec![10, 20, 30, 40, 50];
    let size = shape.iter().product();
    let queue = Queue::new(context.clone(), size)?;
    let x = ArrayBase::<Vec<_>>::with_context(context.clone(), shape, vec![1; size])?;

    println!("reduce axis {} of {:?} (size {})", 2, x, x.size());

    let reduced = x.sum(vec![2], false)?;

    for _ in 0..ITERATIONS {
        let start = Instant::now();
        let _output = reduced.read(&queue)?;
        let duration = start.elapsed();
        println!("{:?} ms", duration.as_millis());
    }

    Ok(())
}

fn reduce_sum_all(context: &Context) -> Result<(), Error> {
    for m in 2..8 {
        let shape = (1..m).map(|dim| dim * 10).collect::<Vec<usize>>();
        let size = shape.iter().product();
        let x = ArrayBase::<Arc<Vec<_>>>::with_context(
            context.clone(),
            shape,
            Arc::new(vec![1; size]),
        )?;

        println!("reduce {:?} (size {})...", x, x.size());

        for _ in 0..ITERATIONS {
            let start = Instant::now();
            let _x = x.clone().sum_all()?;
            let duration = start.elapsed();
            println!("{:?} us", duration.as_micros());
        }
    }

    Ok(())
}

fn transpose(context: &Context) -> Result<(), Error> {
    let shape = vec![10, 20, 30, 40, 50];
    let size = shape.iter().product();
    let permutation = vec![2, 4, 3, 0, 1];

    let queue = Queue::new(context.clone(), size)?;
    let x = ArrayBase::<Vec<_>>::with_context(context.clone(), shape, vec![1; size])?;

    println!("transpose axes {permutation:?} of {x:?}...");

    let transposed = x.transpose(Some(permutation))?;
    for _ in 0..ITERATIONS {
        let start = Instant::now();
        transposed.read(&queue)?;
        let duration = start.elapsed();
        println!("{:?} ms", duration.as_millis());
    }

    Ok(())
}

fn main() -> Result<(), Error> {
    let context = Context::new(0, 0, None)?;

    broadcast_and_multiply(&context)?;
    matmul(&context)?;
    reduce_sum_axis(&context)?;
    reduce_sum_all(&context)?;
    transpose(&context)?;

    Ok(())
}