Crate aether_core 

aether-core

Hard real-time modular DSP engine — lock-free graph scheduler, generational arena, and zero-allocation buffer pool.

64-sample buffer · 48 kHz · ≤1.33 ms deadline · Zero allocations · Lock-free

Architecture

Control thread                    RT audio thread
──────────────                    ───────────────
AudioGraph (add/remove nodes)     Scheduler::process_block()
    │                                 │
    │  CommandRing (SPSC)             ├─ drain CommandRing
    └────────────────────────────────►│
                                      ├─ iterate sorted NodeArena
                                      ├─ borrow buffers from BufferPool
                                      └─ call DspNode::process() per node

Real-time guarantees

Rule	Enforcement
No heap allocation	Pre-allocated arena + buffer pool
No locks	SPSC ring buffer (ringbuf)
No I/O	All I/O on control/tokio threads
Bounded execution	Flat topo-sorted array, ≤32 commands/tick

Quick start

See node::DspNode to implement a processing node, [graph::AudioGraph] to build a patch, and scheduler::Scheduler to drive the RT loop.

Modules

arena

Generational arena allocator.

buffer_pool

Pre-allocated audio buffer pool.

command

Lock-free command protocol between the control thread and the RT audio thread.

graph

Directed Acyclic Graph (DAG) for DSP routing.

node

Node abstraction for the DSP graph.

param

Sample-accurate parameter automation.

scheduler

Real-time audio scheduler.

state

State continuity framework.

Constants

BUFFER_SIZE

Audio buffer size in samples. Hard real-time constraint: 64 samples @ 48kHz = 1.33ms deadline.

COMMAND_RING_CAPACITY

Command ring buffer capacity.

MAX_BUFFERS

Maximum number of audio buffers in the pool.

MAX_COMMANDS_PER_TICK

Maximum commands processed per audio callback. Bounds mutation cost.

MAX_INPUTS

Maximum number of inputs per node. Kept small to fit in cache lines.

MAX_NODES

Maximum number of nodes in the arena. Pre-allocated at startup.