vyre-wgpu

wgpu backend for vyre IR — implements vyre::VyreBackend on any wgpu-capable GPU (Vulkan, DX12, Metal, WebGPU).

cargo add vyre vyre-wgpu

Example

use vyre::ir::Program;
use vyre::{DispatchConfig, VyreBackend};
use vyre_wgpu::WgpuBackend;

let backend = WgpuBackend::acquire().await?;
let program: Program = my_program();
let inputs: &[&[u8]] = &[b"input data"];
let config = DispatchConfig::default();

let outputs: Vec<Vec<u8>> = backend.dispatch(&program, inputs, &config)?;

Features

• Pipeline cache — reuses compiled WGSL pipelines across dispatches by content hash.
• Buffer pool — reuses GPU buffer allocations across dispatches.
• Tiered validation cache — three-level cache for repeated shader validation.
• Lowering happens internally. Consumers never pass WGSL strings; the crate lowers Program to WGSL via vyre::lower::wgsl::emit.

Requirements

• A wgpu-capable GPU. This crate does NOT silently fall back to CPU. Absence of a GPU is surfaced as an error, not a degradation.
• wgpu = 24.x. Pinned — major wgpu version bumps are a vyre-wgpu major bump.

MSRV

Rust 1.85.

License

MIT OR Apache-2.0.