Replace map transforms with pre-owned and reused buffer cherry-pick

[pbarejko]

Description

Cherry-pick #7233

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (existing functionality will not work without user modification)
• Documentation update

Screenshots

Please attach before and after screenshots of the change if applicable.

Checklist

• I have read and understood the contribution guidelines
• I have run the pre-commit checks with ./isaaclab.sh --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have added a changelog fragment under source/<pkg>/changelog.d/ for every touched package (do not edit CHANGELOG.rst or bump extension.toml — CI handles that)
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[isaaclab-review-bot]

Isaac Lab Review Bot

The legacy OVRTX transform paths now use binding.write(..., DataAccess.ASYNC, cuda_stream=...) instead of mapped OVRTX memory. Object transforms use a persistent caller-owned Warp buffer, while camera transforms continue using a per-frame Warp allocation.

• Design and architecture: Object-buffer ownership is consistently contained within the legacy renderer: it is initialized with the legacy fields, allocated alongside the object binding, required by the update guard, and released during legacy shutdown. The ovstage path remains unchanged.
• API: The modified renderer symbols are private, with no demonstrated change to public exports, signatures, tensor shapes, dtypes, or devices. A package changelog fragment documents the behavior change.
• Implementation: The object buffer length and wp.mat44d dtype align with the binding order and kernel launch dimension. Both transform paths pass the producing Warp CUDA stream to the asynchronous OVRTX write, and focused tests verify the selected buffer, access mode, stream forwarding, and cleanup. The camera path still allocates its transform buffer per frame, so this change removes mapping overhead there but does not introduce buffer reuse.

No blocking issues. No inline issue met the actionable-evidence threshold; the assessment above records the review feedback.

Automated review; human maintainers own approval decisions.

[greptile-apps]

Greptile Summary

This PR replaces mapped legacy OVRTX transform updates with caller-owned GPU-buffer writes.

• Allocates and reuses a persistent object-transform buffer.
• Writes object and camera transforms asynchronously on the producing Warp CUDA stream.
• Releases the persistent buffer during renderer cleanup.
• Adds contract tests and a changelog fragment for the new write behavior.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete correctness, security, or lifecycle failure identified in the changed paths.

Object and camera transform writes preserve producer-stream ordering, the persistent object buffer matches the collected object-index set, and cleanup releases the newly owned state.

Important Files Changed

Filename	Overview
source/isaaclab_ov/isaaclab_ov/renderers/ovrtx_renderer.py	Replaces mapped object and camera transform updates with stream-aware caller-owned writes and adds persistent object-buffer lifecycle management.
source/isaaclab_ov/test/test_ovrtx_deformable_bindings.py	Adds focused tests confirming the object and camera paths pass the intended buffers, asynchronous access mode, and CUDA streams.
source/isaaclab_ov/test/test_ovrtx_renderer_contract.py	Extends renderer cleanup coverage to require release of the persistent object-transform buffer.
source/isaaclab_ov/changelog.d/ovrtx-caller-owned-transform-write.rst	Documents the replacement of per-frame mapped transform updates with caller-owned GPU-buffer writes.

Sequence Diagram

sequenceDiagram
    participant Render as Render loop
    participant Warp as Warp CUDA stream
    participant Buffer as Caller-owned transform buffer
    participant OVRTX as OVRTX binding
    Render->>Warp: Launch transform kernel
    Warp->>Buffer: Populate matrices
    Render->>OVRTX: write(buffer, ASYNC, cuda_stream)
    OVRTX->>Warp: Wait on producer stream
    OVRTX->>Buffer: Read transform matrices

Loading

_{Reviews (1): Last reviewed commit: "Reformat" | Re-trigger Greptile}