mnemosyne-graph-core 0.1.0

Shared graph kernels for the Mnemosyne memory substrate: force-directed simulation, R-tree viewport index, and the SIMD primitives both the native PyO3 crate and the WASM sub-crate depend on.
Documentation

Shared graph kernels for Mnemosyne.

This crate is reusable between the native PyO3 crate (mnemosyne_rs) and the forthcoming graph_wasm sub-crate (Wave 2 of Phase 223). It contains:

  • [util] — runtime-dispatched AVX2+FMA dot products. Moved here from mnemosyne_rs::util so both surfaces can share one SIMD implementation.
  • [force] — Barnes-Hut 2D force simulation (Simulation, SimulationConfig).
  • [viewport] — R-tree viewport spatial index with lazy incremental updates (ViewportIndex, IndexPoint, ScoreKey).

Features

  • native (default) — enables rayon. Intended for the host-side PyO3 crate and any integration tests that want parallelism.
  • wasm — opts out of rayon and any other host-only deps. The WASM sub-crate sets default-features = false, features = ["wasm"].

Rayon is optional because the Wave 1 force simulation is single- threaded (see the note in force.rs); the feature flag reserves the parallel code path for when we need it.