truce-simd

Portable SIMD primitives + canonical block-rate audio ops for truce plugins.

Overview

Two backends, feature-gated behind a stable public API:

• wide-backend (default) — stable Rust, uses the wide crate. Maps f32x4 / f32x8 / f64x2 / f64x4 onto SSE2 / AVX2 on x86, NEON on AArch64, scalar fallback elsewhere.
• portable-simd-backend (planned) — nightly-only, swaps in core::simd when it stabilizes. Same public type aliases, so consumers don't have to rewrite.

use truce_simd::{ops, math};

// In `process()`:
ops::scale_block(out, src, gain);                // out = src * gain
ops::mac_block(out, src, gain);                  // out += src * gain
ops::mix_block(out, a, gain_a, b, gain_b);       // dry/wet workhorse
math::tanh_block(out, src);                      // vectorized tanh
math::db_to_linear_block(out, src);              // vectorized dB → lin

All ops are pure math — no atomics, no parameter reads, no audio-thread allocation. They're the inner-loop complement to a truce_params::FloatParam::read_block::<N>() / truce_core::AudioBuffer::chunks_mut::<N>() driver. See the block-gain example for the canonical fast-path / slow-path shape.

Part of truce.