Repository navigation
Integrate sidereus into master with QC removal and MPI force fix - #67
Conversation
Unify CPU/GPU classification and fix VDS smoke expectations, EAM runtime fixtures, and CUDA CMAP test linking. Validated with Pixi: 38 CPU tests and 39 CUDA tests passed. Fixes #58
…ication fix: make floating-point classification safe under fast math
Keep installation checks for all packages, but only launch CPU executables on hosted runners that are not guaranteed to provide a CUDA driver.
refactor(boundary): unify explicit PBC and NOPBC interfaces
fix(nopbc): decouple CV and full-system SITS from periodic initialization
fix(virtual-atoms): validate dependencies and correct force redistribution
fix(barostat): correct box geometry and MC rejection rollback
Remove experimental quantum chemistry support
|
@codex review |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d04ffecfcd
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if (vatom.has_type_3) | ||
| vatom.Throw_If_Runtime_Error( | ||
| "coordinate refresh or force redistribution"); |
There was a problem hiding this comment.
When a type-3 virtual atom becomes degenerate during dynamics, the kernels only record d_runtime_error, while this check runs solely on Check_Mdout_Step(). With the default write_mdout_interval = 1000—or any short run ending before its next mdout frame—the simulation continues or even exits successfully after substituting the source position and dropping the virtual-atom force, so trajectories and restarts can contain silently corrupted dynamics. Check the flag after each affected force evaluation or otherwise outside the mdout cadence.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Codex Review: Didn't find any major issues. What shall we delve into next?
Reviewed commit: d04ffecfcd
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
2 participants
Integrate lab/sidereus-ai into master without updating the lab branches. This replaces #66 and includes the already-merged Maxwell initial-velocity work from #65.
The integration brings PRs #59–64 into master: floating-point classification, Boundary and NOPBC/CV/SITS updates, virtual-atom and barostat fixes, and removal of experimental QC code, PySCF benchmarks, and QC-specific HDF5 interfaces. ReaxFF and other many-body potential implementations are retained. The pre-removal QC tree remains available on lab/sidereus-ai-QM at 8538c57.
Resolve the initialization conflict by preserving Boundary-based virtual-atom refresh followed by Maxwell initialization, and adapt Maxwell's SETTLE/SHAKE projection calls to Boundary.
Address the Codex review on #66: ReaxFF EEQ skips ordinary PME computation but must not skip receiving the PM/CV force buffer. PP ranks now receive once per force evaluation, matching the sender. A new regression exercises one PP and one PM rank with and without nonzero CV biases, checks serial/MPI force and energy agreement over multiple frames, and enforces a timeout. It is included in vali-cv-mpi.
Validation on Linux:
GPU MPI has not been validated locally. Earlier Windows heap-corruption and Linux custom-force SIGILL observations are not claimed to be fixed by this MPI communication change. The separate historical SITS pressure-tensor concern remains outside this PR's new fix.