feat: add indexed fabric transform kernels for local/world matrix propagation

[pv-nvidia]

Motivation

FabricFrameView currently only accelerates world poses via Fabric. To support Fabric-accelerated get/set_local_poses, we need kernels that can propagate transforms between parent world matrices and child local/world matrices directly in Fabric storage — without round-tripping through USD.

Changes

New Warp kernels (operating on wp.indexedfabricarray):

Kernel	Purpose
decompose_indexed_fabric_transforms	Extract position/quaternion/scale from indexed fabric matrix
compose_indexed_fabric_transforms	Write position/quaternion/scale into indexed fabric matrix
update_indexed_local_matrix_from_world	local = inv(parent) * world
update_indexed_world_matrix_from_local	world = parent * local

Tests:

• Smoke test verifying all kernels are importable and are valid wp.Kernel instances
• Full integration tests (requiring Fabric/USDRT) will follow in the local-poses PR

Docs:

• Added API docs section for isaaclab.utils.warp.fabric module

Note

These kernels are purely additive — no existing behavior changes. They form the foundation for the upcoming Fabric-accelerated local-poses feature in FabricFrameView.

[greptile-apps]

Greptile Summary

This PR adds four new @wp.kernel functions and two @wp.func helpers to isaaclab.utils.warp.fabric, enabling Fabric-accelerated local/world matrix propagation via wp.indexedfabricarray without USD round-trips. The changes are purely additive and form a foundation for the upcoming FabricFrameView.get/set_local_poses feature.

• decompose_indexed_fabric_transforms / compose_indexed_fabric_transforms mirror the existing non-indexed kernels but skip the explicit mapping argument, since the view-to-fabric mapping is baked into the IndexedFabricArrayMat44d.
• update_indexed_local_matrix_from_world / update_indexed_world_matrix_from_local compute local = inv(parent) * world and world = parent * local directly on transposed Fabric storage matrices using the identity (A·B)ᵀ = Bᵀ·Aᵀ, avoiding explicit transposes.
• A new test file exercises the decompose round-trip and both matrix-propagation helpers via plain wp.array wrappers so no Fabric/USDRT runtime is required.

Confidence Score: 5/5

Safe to merge — all four kernels are purely additive and no existing behaviour is altered.

The change introduces new kernels that are never called by existing code paths, so there is no regression risk. The matrix math (transposed storage convention, float32 intermediate precision) is consistent with the patterns already established in the file. The helper @wp.func functions are directly exercised by the new tests.

No files require special attention; the new kernels in source/isaaclab/isaaclab/utils/warp/fabric.py follow the existing patterns closely.

Important Files Changed

Filename	Overview
source/isaaclab/isaaclab/utils/warp/fabric.py	Adds four new Warp kernels and two helper @wp.func functions for indexed Fabric matrix operations; logic is consistent with existing non-indexed kernels but has a minor docstring inaccuracy in update_indexed_local_matrix_from_world and the compose kernel reads the existing fabric matrix unnecessarily when all components are supplied.
source/isaaclab/test/utils/warp/test_fabric_kernels.py	New test file covering decompose round-trip and local↔world matrix propagation math via plain wp.array wrappers; compose_indexed_fabric_transforms is not exercised (noted in a prior review thread).
source/isaaclab/changelog.d/indexed-fabric-kernels.rst	Changelog entry correctly describes the new kernels but reverses the matrix multiplication order for the local-from-world formula (previously flagged in review).
docs/source/api/lab/isaaclab.utils.rst	Adds an automodule section for the new isaaclab.utils.warp.fabric module; straightforward documentation change with no issues.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    subgraph "set_local_poses (upcoming)"
        SLP[Caller writes local pos/quat/scale]
    end

    subgraph "compose_indexed_fabric_transforms"
        CIF1[Read current fabric_matrices view_index]
        CIF2[_decompose_transformation_matrix]
        CIF3[Overwrite pos / rot / scale from input arrays]
        CIF4["Write wp.transpose(transform_compose) → fabric_matrices[view_index]"]
        CIF1 --> CIF2 --> CIF3 --> CIF4
    end

    subgraph "update_indexed_world_matrix_from_local"
        UW1["Read child_local_matrices[view_index] → mat44f"]
        UW2["Read parent_world_matrices[view_index] → mat44f"]
        UW3["_world_from_local_transposed: world^T = local^T × parent^T"]
        UW4["Write → child_world_matrices[view_index]"]
        UW1 --> UW3
        UW2 --> UW3 --> UW4
    end

    subgraph "update_indexed_local_matrix_from_world"
        UL1["Read child_world_matrices[view_index] → mat44f"]
        UL2["Read parent_world_matrices[view_index] → mat44f"]
        UL3["_local_from_world_transposed: local^T = world^T × inv(parent^T)"]
        UL4["Write → child_local_matrices[view_index]"]
        UL1 --> UL3
        UL2 --> UL3 --> UL4
    end

    subgraph "decompose_indexed_fabric_transforms"
        DIF1["Read fabric_matrices[view_index] → mat44f"]
        DIF2[_decompose_transformation_matrix]
        DIF3["Write pos/rot/scale → output arrays[output_index]"]
        DIF1 --> DIF2 --> DIF3
    end

    SLP --> CIF1
    CIF4 --> UW1
    UW4 -->|"consistent worldMatrix"| DIF1

Loading

_{Reviews (2): Last reviewed commit: "test: replace scipy with warp builtins i..." | Re-trigger Greptile}