[Backport] PR #7347 to release/3.0.0

[isaaclab-bot]

Backports #7347 to release/3.0.0.

The original cherry-pick conflicted. An NVIDIA inference model proposed this resolution, and deterministic validation confirmed that it changes no paths outside the original PR. Because conflict resolution cannot be certified as an exact patch replay, this PR is intentionally a draft and requires release-maintainer review.

Field	Commit
Original merged change	a2db45885453779bdec71f57b3c5d91f958b713f
Release base used	def4b80288de2a8bac3abad4d2e7cdb80b849a4f
Proposed backport	7db3a4902c045a56b31345c231e4afb430f2da29

[isaaclab-review-bot]

Isaac Lab Review Bot

This backport adds optional OVRTX render-product device pinning through the USD deviceIds attribute, shares CUDA ordinal parsing with attribute mappings, and includes focused USD-generation and renderer-path tests plus a changelog fragment.

• Design and architecture: The reviewed design extends the render-scope builders with an optional device ordinal while preserving OVRTX automatic device selection when none is supplied. Reusing cuda_device_id for mapping and render-product selection is intended to keep their CUDA ordinals aligned; the supplied renderer test demonstrates that a cuda:1 render spec produces deviceIds = [1].
• API: Both USD-builder parameters are appended with None defaults, preserving existing positional and keyword callers. The omission test confirms that default calls do not author deviceIds. Renaming _cuda_device_id exposes a formerly private helper under a public-looking name, but no unsupported compatibility break is demonstrated.
• Implementation: The generated property is valid USD uint[] according to the Sdf-based test, and the renderer-path test checks propagation from a cuda:1 spec into the combined stage. Bare "cuda" still resolves to ordinal 0 as explicitly documented pre-existing behavior. The large ovrtx_renderer.py change was supplied without patch or current-file content, limiting direct inspection of that implementation path, but the available evidence does not support a concrete finding.

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

The backport pins OVRTX render-product buffers to the CUDA device used by the renderer, preventing cross-device access during tile extraction.

• Exposes a shared CUDA-device-index parser for mappings and render-product creation.
• Authors the selected index as the render product’s deviceIds attribute.
• Adds USD-generation and renderer-contract coverage for indexed and automatic device selection.
• Includes the corresponding changelog entry.

Confidence Score: 5/5

The PR appears safe to merge, with the render product and consuming Warp kernels consistently assigned to the normalized simulation CUDA device.

The changed initialization passes the same indexed renderer device used by runtime kernels into OVRTX buffer allocation, and no reachable regression remains in the reviewed renderer lifecycle or USD contract.

Important Files Changed

Filename	Overview
source/isaaclab_ov/isaaclab_ov/renderers/ovrtx_mapping.py	Makes CUDA-device parsing public so mapping and render-product allocation use the same indexed device.
source/isaaclab_ov/isaaclab_ov/renderers/ovrtx_renderer.py	Passes the renderer’s normalized CUDA index into both render-product initialization paths; no actionable lifecycle regression was established.
source/isaaclab_ov/isaaclab_ov/renderers/ovrtx_usd.py	Optionally authors a UIntArray deviceIds attribute while preserving automatic OVRTX assignment for callers that omit the device.
source/isaaclab_ov/test/test_ovrtx_clone_plan.py	Verifies create_render_data propagates an indexed specification device into the generated render-product USD.
source/isaaclab_ov/test/test_ovrtx_usd.py	Verifies the deviceIds attribute’s USD type and value, plus omission when no device is supplied.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
    Spec[CameraRenderSpec device] --> Parse[cuda_device_id]
    Parse --> USD[RenderProduct deviceIds]
    USD --> Buffers[OVRTX render-var buffers]
    Spec --> Warp[Warp extraction kernels]
    Buffers --> Warp

Loading

_{Reviews (1): Last reviewed commit: "Fix OVRTX renderer formatting" | Re-trigger Greptile}