vyre-libs

Category A composition ecosystem over vyre's hardware-intrinsic primitives. Every function here returns a vyre::Program (or a list of vyre::ir::Node a consumer can embed in a larger Program) built entirely from existing vyre-ops primitives. No shader source, no new inventory::submit!, no backend-specific code.

Modules

vyre_libs::math: linear algebra, scans, broadcasts
vyre_libs::nn: neural-net primitives (linear, ReLU, softmax, layer_norm, attention)
vyre_libs::matching: string scanning (substring, DFA, multi-string): one building block inside arbitrary programs
vyre_libs::crypto: hashing (FNV-1a, BLAKE3, SHA-256, CRC32)

Design

Every public function wraps its IR body in a Node::Region with a stable generator name. The optimizer treats Regions as opaque by default (preserves source-mapping + debuggability); explicit inline passes can unroll. This is LLVM's function-vs-always-inline split at IR level.

One crate with four public modules today; each module promotes to its own crates.io identity (vyre-nn, vyre-math, vyre-scan, vyre-crypto) when its consumer base justifies the fragmentation.

Usage

use vyre::VyreBackend;
use vyre_driver::backend::acquire_preferred_dispatch_backend;
use vyre_libs::math::dot;

let program = dot("x", "y", "result");
let backend = acquire_preferred_dispatch_backend()?;
let result = backend.dispatch(
    &program,
    &[vec![1, 2, 3, 4], vec![5, 6, 7, 8], vec![0]],
    &vyre::DispatchConfig::default(),
)?;

Feature flags

[dependencies]
vyre-libs = { version = "0.1", default-features = false, features = ["nn"] }

math (default): linear algebra
nn (default, implies math): neural-net primitives
matching (default): string scanning primitives
crypto (default): hashing

License