Expand description
Hybrid kernel: per-node optimal compilation level.
Splits the DAG into segments based on each node’s compilation capability. JIT-able nodes are batched into native code segments. Non-JIT-able nodes run as Phase 2 closures. All segments share the same flat u64 buffer.
This is the “best of all worlds” kernel — no node pays more overhead than it needs to.
Three monomorphic kernel types, each with no runtime branching:
| Type | Push (per-step skip) | Pull (cone guard) |
|---|---|---|
HybridKernelRaw | — | — |
HybridKernelPull | — | yes |
HybridKernelPushPull | yes | yes |
Structs§
- Hybrid
Kernel Pull - Hybrid kernel with pull-side cone guard.
- Hybrid
Kernel Push Pull - Hybrid kernel with both push-side per-step skip and pull-side cone guard.
- Hybrid
Kernel Raw - Hybrid kernel with no provenance tracking.
Type Aliases§
- Hybrid
Kernel - Type alias for the default hybrid kernel (PushPull — full optimization).