Accel: GPGPU Framework for Rust

CUDA-based GPGPU framework for Rust

Features

• Compile PTX Kernel from Rust using NVPTX backend of LLVM (demonstrated in japaric/nvptx)
• proc-macro-attribute-based approach like futures-await
• Simple memory management using Unified Memory

Sub Crates

• accel-derive: Define procedual macro #[kernel]
• accel-core: Support crate for writing GPU kernel
• cuda-sys: Rust binding to CUDA Driver/Runtime APIs

Pre-requirements

• Install CUDA on your system
• Install LLVM 6.0 or later (use llc and llvm-link to generate PTX)
• Install Rust using rustup.rs
  • Use the nightly Rust toolchain with rustup override nightly
• Install nvptx, and install accel-nvptx toolchain

cargo install nvptx
nvptx install       # install accel-nvptx toolchain to $XDG_DATA_HOME/accel-nvptx

Or, you can use termoshtt/rust-cuda container whith satisfies these requirements.

docker run -it --rm --runtime=nvidia termoshtt/rust-cuda

See also nvidia-docker

Example

#![feature(proc_macro, proc_macro_gen)]

extern crate accel;
extern crate accel_derive;

use accel_derive::kernel;
use accel::*;

#[kernel]
#[crate("accel-core" = "0.2.0-alpha")]
pub unsafe fn add(a: *const f64, b: *const f64, c: *mut f64, n: usize) {
    let i = accel_core::index();
    if (i as usize) < n {
        *c.offset(i) = *a.offset(i) + *b.offset(i);
    }
}

fn main() {
    let n = 32;
    let mut a = UVec::new(n).unwrap();
    let mut b = UVec::new(n).unwrap();
    let mut c = UVec::new(n).unwrap();

    for i in 0..n {
        a[i] = i as f64;
        b[i] = 2.0 * i as f64;
    }
    println!("a = {:?}", a.as_slice());
    println!("b = {:?}", b.as_slice());

    let grid = Grid::x(1);
    let block = Block::x(n as u32);
    add(grid, block, a.as_ptr(), b.as_ptr(), c.as_mut_ptr(), n);

    device::sync().unwrap();
    println!("c = {:?}", c.as_slice());
}

Licence

MIT-License