fix(dynamics): use inv_ex with contiguous in NPT.pre_update

[Ryan-Reese]

ALCHEMI Toolkit Pull Request

Description

NPT.pre_update() was the only cell inversion call in the dynamics codebase using torch.linalg.inv() instead of inv_ex()[0].contiguous(). inv() returns a non-contiguous tensor (column-major strides (9, 1, 3)) which can cause incorrect results in downstream Warp kernels that expect row-major layout.

Type of Change

• Bug fix (non-breaking change that fixes an issue)

Changes Made

• Changed npt.py:292 from torch.linalg.inv(batch.cell) to torch.linalg.inv_ex(batch.cell)[0].contiguous(), matching every other cell inversion call site (NPH, NPT.post_update, FIREVariableCell, NeighborListHook).

Testing

• Unit tests pass locally (make pytest)
• Linting passes (make lint)
• Ran test_npt_nph.py (20/20 pass) and test_ops.py (80/80 pass) on CUDA

Checklist

• I have read and understand the Contributing Guidelines
• I have performed a self-review of my code

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[greptile-apps]

Greptile Summary

This PR fixes a one-line inconsistency in NPT.pre_update() where torch.linalg.inv(batch.cell) was the only remaining call site not using torch.linalg.inv_ex(batch.cell)[0].contiguous(). The fix aligns it with all other cell inversions in the codebase (NPH pre/post, NPT post_update, FIREVariableCell, NeighborListHook), ensuring the result tensor is contiguous and row-major as required by downstream Warp kernels.

Important Files Changed

Filename	Overview
nvalchemi/dynamics/integrators/npt.py	Single-line fix in pre_update(): replaces torch.linalg.inv() with inv_ex()[0].contiguous() to match every other call site and ensure row-major layout for downstream Warp kernels.

_{Reviews (3): Last reviewed commit: "Merge branch 'main' into fix/npt-inv-con..." | Re-trigger Greptile}