mnn 0.2.0

Rust bindings for MNN, a lightweight deep neural network inference 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

use mnn::*;
use std::path::PathBuf;

#[derive(Debug, clap::Parser, Clone)]
pub struct Cli {
    model: PathBuf,
    #[clap(short, long)]
    forward: ForwardType,
    #[clap(short, long, default_value = "high")]
    power: PowerMode,
    #[clap(short = 'P', long, default_value = "high")]
    precision: PrecisionMode,
    #[clap(short, long, default_value = "high")]
    memory: MemoryMode,
    #[clap(short, long, default_value = "f32")]
    output_data_type: DataType,
    #[clap(short, long, default_value = "f32")]
    input_data_type: DataType,
    #[clap(short, long, default_value = "1")]
    loops: usize,
    #[clap(short, long)]
    no_cache: bool,
}

#[derive(Debug, Clone, clap::ValueEnum)]
pub enum DataType {
    F32,
    U8,
    I8,
}

macro_rules! time {
    ($($x:expr),+ ; $text: expr) => {
        {
            let start = std::time::Instant::now();
            let result = { $($x);+ };
            let elapsed = start.elapsed();
            println!("{}: took: {:?}", $text,elapsed );
            result
        }
    };
    ($($x:expr),+) => {
        time!($($x),+; "")
    };
}

pub fn main() -> anyhow::Result<()> {
    use clap::Parser;
    let cli = Cli::parse();
    let mut interpreter = Interpreter::from_file(&cli.model)?;
    if !cli.no_cache {
        interpreter.set_cache_file(cli.model.with_extension("cache"), 128)?;
    }

    tracing_subscriber::fmt()
        .event_format(
            tracing_subscriber::fmt::format()
                .with_file(true)
                .with_line_number(true),
        )
        .init();

    let mut config = ScheduleConfig::new();
    config.set_type(cli.forward);
    let mut session = time!(interpreter.create_session(config)?; "create session");
    if !cli.no_cache {
        interpreter.update_cache_file(&mut session)?;
    }

    let mut current = 0;
    println!("--------------------------------Info--------------------------------");
    let mem = interpreter.memory(&session)?;
    let flops = interpreter.flops(&session)?;
    println!("Memory: {:?}MiB", mem);
    println!("Flops : {:?}M", flops);
    println!("ResizeStatus : {:?}", interpreter.resize_status(&session)?);

    time!(loop {
        println!("--------------------------------Inputs--------------------------------");
        interpreter.inputs(&session).iter().for_each(|x| {
            match cli.input_data_type {
                DataType::F32 => {
                    let mut tensor = x.tensor::<f32>().expect("No tensor");
                    println!("{}: {:?}", x.name(), tensor.shape());
                    tensor.fill(1.0f32);
                },
                DataType::U8 => {
                    let mut tensor = x.tensor::<u8>().expect("No tensor");
                    println!("{}: {:?}", x.name(), tensor.shape());
                    tensor.fill(1u8);
                },
                DataType::I8 => {
                    let mut tensor = x.tensor::<i8>().expect("No tensor");
                    println!("{}: {:?}", x.name(), tensor.shape());
                    tensor.fill(1i8);
                },
            };
        });

        println!("Running session");
        interpreter.run_session(&session)?;
        println!("--------------------------------Outputs--------------------------------");
        let outputs = interpreter.outputs(&session);
        outputs.iter().for_each(|x| {
            match cli.output_data_type {
                DataType::F32 => {
                    let tensor = x.tensor::<f32>().expect("No tensor");
                    println!("{}: {:?}", x.name(), tensor.shape());
                    time!(tensor.wait(MapType::MAP_TENSOR_READ, true); format!("Waiting for tensor: {}", x.name()));
                },
                DataType::U8 => {
                    let tensor = x.tensor::<u8>().expect("No tensor");
                    println!("{}: {:?}", x.name(), tensor.shape());
                    time!(tensor.wait(MapType::MAP_TENSOR_READ, true); format!("Waiting for tensor: {}", x.name()));
                },
                DataType::I8 => {
                    let tensor = x.tensor::<i8>().expect("No tensor");
                    println!("{}: {:?}", x.name(), tensor.shape());
                    time!(tensor.wait(MapType::MAP_TENSOR_READ, true); format!("Waiting for tensor: {}", x.name()));
                },
            };
    
        });
        current += 1;
        if current >= cli.loops {
            break;
        }
    }; "run loop");
    Ok(())
}