Skip to content

Rust implementation of the FastLanes compression library

License

Notifications You must be signed in to change notification settings

spiraldb/fastlanes

Repository files navigation

FastLanes Rust

A Rust implementation of the FastLanes compression library

Azim Afroozeh and Peter Boncz. 2023. The FastLanes Compression Layout: Decoding > 100 Billion Integers per Second with Scalar Code. Proc. VLDB Endow. 16, 9 (May 2023), 2132–2144. https://doi.org/10.14778/3598581.3598587

FastLanes is a compression framework that can leverage LLVM's auto-vectorization to achieve high-performance SIMD decoding without intrinsics or other explicit SIMD code.

Usage

use fastlanes::BitPacking;

fn pack_u16_into_u3() {
    const WIDTH: usize = 3;

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

    // Pack the values.
    let mut packed = [0; 128 * WIDTH / size_of::<u16>()];
    // prefer using the safe pack/unpack functions unless you have a specific reason to use the unchecked versions
    // e.g. `BitPacking::pack::<WIDTH>(&values, &mut packed);`
    // unfortunately the safe versions don't work in doctests since they depend on the unstable generic_const_exprs feature
    // see a version of this example without unsafe in `src/lib.rs`
    unsafe { BitPacking::unchecked_pack(WIDTH, &values, &mut packed); }

    // Unpack the values.
    let mut unpacked = [0u16; 1024];
    // e.g., `BitPacking::unpack::<WIDTH>(&packed, &mut unpacked);`
    unsafe { BitPacking::unchecked_unpack(WIDTH, &packed, &mut unpacked); }
    assert_eq!(values, unpacked);

    // Note that for more than ~10 values, it is typically faster to unpack all values and then 
    // access the desired one.
    for i in 0..1024 {
        // e.g., `BitPacking::unpack_single::<WIDTH>(&packed, i)`
        assert_eq!(unsafe { BitPacking::unchecked_unpack_single(WIDTH, &packed, i) }, values[i]);
    }
}

Differences to original FastLanes

Caution

Rust FastLanes is not binary compatible with original FastLanes

The BitPacking implementation in this library is reordered vs the original to enable fused kernels for transposed encodings (like Delta and RLE) in addition to the linear kernels such as FoR.

Verifying ASM

To validate the correctness of the generated assembly and ensure it is vectorized, you can use the following command:

RUSTFLAGS='-C target-cpu=native' cargo asm --profile release --bench bitpacking --rust BitPacking

Note, it requires cargo install cargo-show-asm.

Benchmarking

RUSTFLAGS='-C target-cpu=native' cargo bench --profile release

License

Licensed under the Apache 2.0 license.