Struct PersistentOutputs 

pub struct PersistentOutputs { /* private fields */ }

Persistent per-node output buffers for CUDA-graph capture (the “CapturedRun executor” build-out). Maps each compute node’s NodeId to a FIXED-ADDRESS output Storage Arc that survives across realizes, so a captured graph’s baked kernel-argument addresses stay valid on replay.

Two modes, both driven by the executor’s Kernel arm:

• Record (warm run): the executor allocates each kernel output as usual AND records the Arc here. Produces the fixed buffers.
• Reuse (capture / replay run): the executor REUSES the recorded Arc for each node — no allocation (mandatory inside a capture scope) — and the kernel’s write-into path writes into it in place.

None (the param default at every existing realize call site) preserves byte-identical behavior: fresh alloc every realize, nothing recorded.

Implementations

impl PersistentOutputs

pub fn recording() -> Self

A fresh recording map for the warm run.

pub fn reuse(map: HashMap<NodeId, Arc<RwLock<Storage>>>) -> Self

A reuse map for the capture / replay run, seeded with the buffers a prior warm run recorded.

pub fn into_map(self) -> HashMap<NodeId, Arc<RwLock<Storage>>>

Consume, yielding the recorded NodeId→buffer map (to seed a reuse map for the capture run).