Expand description
Layer 3 complete compute engines.
Each engine is a self-contained GPU compute pipeline: structured input in, compute passes on a real GPU backend, typed output back.
This module owns backend execution helpers only: command graphs, work
partitioning policy, persistent queues, host-ingress compatibility, and
shared record/readback. Domain algorithms live in vyre-libs or
wrappers that consume vyre.
Modulesยง
- graph
- GPU-resident command graph execution. GPU-resident dispatch graph execution (Innovation I.14).
- multi_
gpu - Mockable multi-GPU work partitioning. Multi-GPU adapter probing, backend acquisition, and work partitioning.
- persistent
- Resident persistent-kernel queue engine. Persistent-kernel queue execution for wgpu compute pipelines.
- streaming
- Host-ingress chunk bridge for callers that still receive bytes through CPU memory. Host-ingress compatibility stream for chunked inputs.