Add realtime AI decoder / predecoder infrastructure (GPU + Host) w/ host dispatcher#457

Draft

wsttiger wants to merge 43 commits intoNVIDIA:mainfrom

wsttiger:add_realtime_ai_predecoder_host_side_gb200

Commits on Feb 18, 2026

Add realtime GPU dispatch kernel library with RPC-based function dispatch
wsttiger
committed
Add AI predecoder service with hybrid GPU-CPU decoding pipeline
wsttiger
committed

Commits on Feb 19, 2026

Enable real ONNX model inference in AI predecoder pipeline
wsttiger
committed
Integrate real PyMatching MWPM decoder into AI predecoder pipeline
wsttiger
committed
Refactor predecoder test into multi-distance PipelineConfig architecture
wsttiger
committed
Add PipelineBenchmark utility for realtime decoding latency measurement
wsttiger
committed
Integrate PipelineBenchmark into predecoder test and track incomplete requests
wsttiger
committed
Add per-worker timing breakdown to predecoder pipeline test
wsttiger
committed
Cache TRT engines to disk and use per-worker decoder pool
wsttiger
committed
Add streaming test mode with continuous syndrome arrival simulation
wsttiger
committed

Commits on Feb 20, 2026

Commits on Feb 21, 2026

realtime: host-side dynamic worker pool dispatcher and predecoder refactor
wsttiger
committed

Commits on Feb 22, 2026

Commits on Feb 26, 2026

Commits on Feb 27, 2026

Add pre-launch DMA input copy callback and d13_r104 config
wsttiger
committed

Commits on Mar 2, 2026

Add RealtimePipeline scaffolding; refactor benchmark to use it
wsttiger
committed

Commits on Mar 6, 2026

Commits on Mar 9, 2026

Commits on Mar 11, 2026

Fix uint8 model I/O and enable correctness verification with Stim data
wsttiger
committed

Commits on Mar 12, 2026

Add syndrome density diagnostic and fix X-basis config labels
wsttiger
committed

Commits on Mar 16, 2026

Commits on Mar 17, 2026

Decouple PyMatching workers from predecoder workers and update docs
wsttiger
committed