Expand description
In-memory NIR graph model.
A graph is a named set of computational nodes plus a list of directed
identity edges, matching neuromorphs/NIR (nodes, edges, metadata,
optional version). Cycles are allowed; structure validation only checks
edge endpoints, duplicate directed edges, and nested NirNode::Graph
subgraphs. Validation walks those subgraphs on a heap-allocated work
list so process-stack usage does not grow with nesting depth.
Convolution and pooling parameter invariants are a separate opt-in check
(NirGraph::validate_parameters).
Structs§
- NirGraph
- Directed NIR computation graph.