Struct CompiledWave 

pub(crate) struct CompiledWave {
    pub(crate) init: CompiledBlock,
    pub(crate) per_frame: CompiledBlock,
    pub(crate) per_point: Vec<Node>,
    pub(crate) per_point_parallelism: PerPointParallelism,
    pub(crate) per_point_bytecode: Option<CompiledBytecode>,
}

Pre-compiled equation Nodes for one wavecode_N block. We compile each phase exactly once per load_preset; per-frame and per-point evaluation just runs the cached Nodes, skipping the regex-based preprocess that MilkEvaluator::eval performs every call. This is load-bearing for performance: a single 512-sample wave with 5 per-point statements runs the loop 2 560 times per frame.

init and per_frame are wrapped in CompiledBlock so they transparently pick the bytecode VM when the block lowers cleanly (~80 % of corpus presets) and fall back to evalexpr nodes otherwise. per_point keeps its raw Vec<Node> + separate Option<CompiledBytecode> because the parallel-samples branch in compute_one_wave needs them addressable individually (carry-safe presets fan samples out across rayon workers, with each worker running either path).

Fields

init: CompiledBlock

per_frame: CompiledBlock

per_point: Vec<Node>

per_point_parallelism: PerPointParallelism

Cached carry-analysis verdict — Safe means samples can be evaluated in parallel via rayon (no cross-sample state), Sequential means we must thread the previous sample’s output into the next call (MD2 default semantics).

per_point_bytecode: Option<CompiledBytecode>

Bytecode-VM compilation of per_point. Some when every node in the block lowered cleanly; None when at least one operator or function falls outside the VM’s supported set (rand, gmegabuf, loop, …) — in that case compute_one_wave falls back to the evalexpr Node path so behaviour is preserved.

Trait Implementations

impl Debug for CompiledWave

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for CompiledWave

fn default() -> Self

Returns the “default value” for a type.