Rust Engine API (sc_neurocore_engine)¶

The high-performance Rust engine provides SIMD-accelerated stochastic computing with SIMD-accelerated bitstream operations, IR compilation, and HDC support.

Browse the full Rust API documentation →

Key Modules¶

Module	Description
`bitstream`	Packed bitstream types and SIMD operations (AND, popcount, rotate)
`encoder`	LFSR-based stochastic encoders with decorrelated seeds
`neuron`	Fixed-point LIF neuron with Q8.8 arithmetic
`layer`	Dense layer pipeline with vectorised forward pass
`ir`	Intermediate representation for graph compilation
`graph`	Computational graph builder and verifier
`attention`	Stochastic attention mechanism
`grad`	Surrogate gradient training support
`scpn`	SCPN layer primitives (Petri net places/transitions)
`simd`	Platform-adaptive SIMD kernels (AVX2, SSE4.1, NEON, portable)
`analysis`	22 spike train analysis modules (see Rust Analysis Engine)
`neurons`	100+ neuron models — biophysical, maps, hardware, interneurons

Criterion Benchmarks → — all measured latencies for engine, neurons, and analysis.

Building from Source¶

Bash

cd engine
cargo build --release
cargo test
cargo doc --open

Python Bindings (PyO3)¶

The engine is exposed to Python via the local sc_neurocore_engine bridge. For source checkouts:

Bash

cd bridge
maturin develop --release

Python

import sc_neurocore_engine as engine

# Compile an IR graph
graph = engine.IRGraph()
graph.add_encode(0, 1024, 0xACE1)
graph.verify()
sv_code = graph.emit_sv()

Installed-wheel interface evidence¶

The engine bridge contracts document tools/engine_abi_inventory.py, which captures the facade and compiled module after wheel installation. Its installed-origin guard refuses a checkout facade; retained captures include module hashes, complete interface metadata and exact alias identities. Compare inventories under an equal Python/NumPy/feature profile, then use per-binding behavioral tests for input layouts, exception and mutation contracts, numerical goldens and supported pickle state.

The qualified default interface reference is tests/fixtures/engine_abi_default.json, for Linux x86-64, CPython 3.12 and NumPy 2.2.3 with default engine features. Its matching wheel job compares all interface fields; other runtime profiles retain separate measured captures. See model compatibility and migration for canonical source models and their explicit retained project profiles.