feat: Full Fabric acceleration stack — local poses, stage cache, fused compose#5728

Draft

pv-nvidia wants to merge 23 commits into

isaac-sim:developfrom

pv-nvidia:pv/fabric-full-stack

Commits on May 20, 2026

feat: add indexed fabric transform kernels for local/world matrix propagation
pv-nvidia
committed
refactor: move Fabric/USD dispatch from FabricFrameView to FrameView factory
pv-nvidia
committed
refactor: fuse set_world_poses/set_scales into single kernel launch
pv-nvidia
committed
feat: add FabricStageCache service for shared hierarchy handles
pv-nvidia
committed
Merge branch 'pv/indexed-fabric-kernels' into pv/fabric-local-poses-base
pv-nvidia
committed
Merge branch 'pv/fabric-stage-cache' into pv/fabric-local-poses-base
pv-nvidia
committed
feat: Fabric-accelerated get/set_local_poses via indexedfabricarray
pv-nvidia
committed
Enable FabricFrameView on non-primary GPUs
pv-nvidia
committed
Skip cuda:1 fabric tests on single-GPU runners instead of failing
pv-nvidia
committed
Strip sys.argv before AppLauncher boot in fabric test
pv-nvidia
committed
Use ProxyArray.torch accessor in fabric view tests
pv-nvidia
committed
Address review feedback on the multi-GPU branch
pv-nvidia
committed
Update FabricFrameView docstrings for multi-GPU support
pv-nvidia
committed
Split FabricFrameView multi-GPU tests into their own workflow
pv-nvidia
committed
ci: disable multi-GPU workflow (no runner available)
pv-nvidia
committed

Commits on May 22, 2026