fastlanes 0.3.0

Rust implementation of the FastLanes compression layout
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

use crate::{pack, supported_bit_width, unpack, BitPacking, FastLanes};
use paste::paste;

pub trait FoR: BitPacking {
    fn for_pack<const W: usize, const B: usize>(
        input: &[Self; 1024],
        reference: Self,
        output: &mut [Self; B],
    );

    fn unfor_pack<const W: usize, const B: usize>(
        input: &[Self; B],
        reference: Self,
        output: &mut [Self; 1024],
    );
}

macro_rules! impl_for {
    ($T:ty) => {
        impl FoR for $T {
            fn for_pack<const W: usize, const B: usize>(
                input: &[Self; 1024],
                reference: Self,
                output: &mut [Self; B],
            ) {
                const {
                    assert!(supported_bit_width(W, 8 * core::mem::size_of::<$T>()));
                    assert!(B == 1024 * W / Self::T);
                }

                for lane in 0..Self::LANES {
                    pack!($T, W, output, lane, |$idx| {
                        input[$idx].wrapping_sub(reference)
                    });
                }
            }

            fn unfor_pack<const W: usize, const B: usize>(
                input: &[Self; B],
                reference: Self,
                output: &mut [Self; 1024],
            ) {
                const {
                    assert!(supported_bit_width(W, 8 * core::mem::size_of::<$T>()));
                    assert!(B == 1024 * W / Self::T);
                }

                for lane in 0..Self::LANES {
                    unpack!($T, W, input, lane, |$idx, $elem| {
                        output[$idx] = $elem.wrapping_add(reference)
                    });
                }
            }
        }
    };
}

impl_for!(u8);
impl_for!(u16);
impl_for!(u32);
impl_for!(u64);

#[cfg(test)]
mod test {
    use super::*;
    use core::mem::size_of;

    #[test]
    fn test_ffor() {
        const W: usize = 15;
        const B: usize = 1024 * W / u16::T;

        let mut values: [u16; 1024] = [0; 1024];
        for i in 0..1024 {
            values[i] = (i % (1 << W)) as u16;
        }

        let mut packed = [0; 128 * W / size_of::<u16>()];
        FoR::for_pack::<W, B>(&values, 10, &mut packed);

        let mut unpacked = [0; 1024];
        BitPacking::unpack::<W, B>(&packed, &mut unpacked);

        for (i, (a, b)) in values.iter().zip(unpacked.iter()).enumerate() {
            assert_eq!(
                // Check that the unpacked array is 10 less than the original (modulo 2^15)
                a.wrapping_sub(10) & ((1 << W) - 1),
                *b,
                "Mismatch at index {i}"
            );
        }
    }
}