krnl

Safe, portable, high performance compute (GPGPU) kernels.

Developed for autograph.

• Similar functionality to CUDA and OpenCL.
• Supports GPU's and other Vulkan 1.2 capable devices.
• MacOS / iOS supported via MoltenVK.
• Kernels are written inline, entirely in Rust.
• Simple iterator patterns can be implemented without unsafe.
• Buffers on the host can be accessed natively as Vecs and slices.

krnlc

Kernel compiler for krnl.

• Built on RustGPU's spirv-builder.
• Supports dependencies defined in Cargo.toml.
• Uses spirv-tools to validate and optimize.
• Compiles to "krnl-cache.rs", so the crate will build on stable Rust.

Example

use krnl::{
    anyhow::Result,
    buffer::{Buffer, Slice, SliceMut},
    device::Device,
    macros::module,
};

#[module]
mod kernels {
    #[cfg(not(target_arch = "spirv"))]
    use krnl::krnl_core;
    use krnl_core::macros::kernel;

    pub fn saxpy_impl(x: f32, alpha: f32, y: &mut f32) {
        *y += alpha * x;
    }

    // Item kernels for iterator patterns.
    #[kernel(threads(256))]
    pub fn saxpy(#[item] x: f32, alpha: f32, #[item] y: &mut f32) {
        saxpy_impl(x, alpha, y);
    }

    // General purpose kernels like CUDA / OpenCL.
    #[kernel(threads(256))]
    pub fn saxpy_global(#[global] x: Slice<f32>, alpha: f32, #[global] y: UnsafeSlice<f32>) {
        use krnl_core::buffer::UnsafeIndex;
        let mut index = kernel.global_index() as usize;
        while index < x.len().min(y.len()) {
            saxpy_impl(x[index], alpha, unsafe { y.unsafe_index_mut(index) });
            index += kernel.global_threads() as usize;
        }
    }
}

fn saxpy(x: Slice<f32>, alpha: f32, mut y: SliceMut<f32>) -> Result<()> {
    if let Some((x, y)) = x.as_host_slice().zip(y.as_host_slice_mut()) {
        for (x, y) in x.iter().copied().zip(y) {
            kernels::saxpy_impl(x, alpha, y);
        }
        return Ok(());
    } 
    kernels::saxpy::builder()?
        .build(y.device())?
        .dispatch(x, alpha, y) 
}

fn main() -> Result<()> {
    let x = vec![1f32];
    let alpha = 2f32;
    let y = vec![0f32];
    let device = Device::builder().build().ok().unwrap_or(Device::host());
    let x = Buffer::from(x).into_device(device.clone())?;
    let mut y = Buffer::from(y).into_device(device.clone())?;
    saxpy(x.as_slice(), alpha, y.as_slice_mut())?;
    let y = y.into_vec()?;
    println!("{y:?}");
    Ok(())
}

Performance

NVIDIA GeForce GTX 1060 with Max-Q Design

Copying from Host to Device

Copying from Device to Host

Compute

Recent Changes

See Releases.md

License

Dual-licensed to be compatible with the Rust project.

Licensed under the Apache License, Version 2.0 http://www.apache.org/licenses/LICENSE-2.0 or the MIT license http://opensource.org/licenses/MIT, at your option. This file may not be copied, modified, or distributed except according to those terms.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions