Attribute Macro stencil_kernel 

#[stencil_kernel]

Attribute macro for defining stencil kernels that transpile to CUDA.

This macro generates CUDA C code from Rust stencil kernel functions at compile time. The generated CUDA source is embedded in the binary and can be compiled at runtime using NVRTC.

Attributes

• id (required) - Unique kernel identifier
• grid - Grid dimensionality: “1d”, “2d” (default), or “3d”
• tile_size - Tile/block size (default: 16)
• tile_width / tile_height - Non-square tile dimensions
• halo - Stencil radius / ghost cell width (default: 1)

Supported Rust Subset

• Primitives: f32, f64, i32, u32, i64, u64, bool
• Slices: &[T], &mut [T]
• Arithmetic: +, -, *, /, %
• Comparisons: <, >, <=, >=, ==, !=
• Let bindings: let x = expr;
• If/else: if cond { a } else { b }
• Stencil intrinsics via GridPos

Example

use ringkernel_derive::stencil_kernel;
use ringkernel_cuda_codegen::GridPos;

#[stencil_kernel(id = "fdtd", grid = "2d", tile_size = 16, halo = 1)]
fn fdtd(p: &[f32], p_prev: &mut [f32], c2: f32, pos: GridPos) {
    let curr = p[pos.idx()];
    let lap = pos.north(p) + pos.south(p) + pos.east(p) + pos.west(p) - 4.0 * curr;
    p_prev[pos.idx()] = 2.0 * curr - p_prev[pos.idx()] + c2 * lap;
}

// Access generated CUDA source:
assert!(FDTD_CUDA_SOURCE.contains("__global__"));