chore: Remove MPI - #2192
Conversation
…ove RandomBuffer.
31f7c58 to
7c92728
Compare
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
⚠️ Performance Alert ⚠️
Possible performance regression was detected for benchmark.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 2.
| Benchmark suite | Current: 4342376 | Previous: 467a003 | Ratio |
|---|---|---|---|
BM_Box_MinimumDistance<OrthorhombicBox> |
12.91046146999677 ns/iter |
5.29815982374698 ns/iter |
2.44 |
BM_Box_MinimumDistance<MonoclinicAlphaBox> |
11.481027478519398 ns/iter |
5.6044062930106175 ns/iter |
2.05 |
BM_Box_MinimumDistanceSquared<MonoclinicAlphaBox> |
11.956455172902752 ns/iter |
4.977398883100473 ns/iter |
2.40 |
This comment was automatically generated by workflow using github-action-benchmark.
CC: @disorderedmaterials/dissolve-devs
7c92728 to
d3e2857
Compare
d3e2857 to
d17e057
Compare
rprospero
left a comment
rprospero
left a comment
There was a problem hiding this comment.
I've pushed out a PR that should fix the nix flake. You covered most of it - just needed to delete a few more lines we didn't need any more.
|
As I'm reviewing this, it looks like we can get rid of the |
rprospero
left a comment
rprospero
left a comment
There was a problem hiding this comment.
Everything looks much cleaner than before. I've made some comments above on extra code that we could delete, but that might just need to be a new issue. I've already added a couple of other issues from items I noticed during this review.
| for (auto it = cfg->molecules().begin() + offset; it < cfg->molecules().end(); it += nChunks) | ||
| for (auto &mol : cfg->molecules()) |
There was a problem hiding this comment.
This single change was worth reviewing 233 files.
Now removed in this PR. |
Co-authored-by: Tristan Youngs <trisyoungs@googlemail.com> Co-authored-by: Adam Washington <adam.washington@stfc.ac.uk>
Co-authored-by: Tristan Youngs <trisyoungs@googlemail.com> Co-authored-by: Adam Washington <adam.washington@stfc.ac.uk>
2 participants
I don't really know how you go about reviewing this, but this PR removes all trace of MPI-related parallelism from the code. The various
Distributorclasses as well asChangeStoreare now basically redundant, but the effort to remove usage of them is left for when relevant modules are converted toNodes.TODO
Distributorand associated classesChangeStoreModuleContext#ifdef PARALLELandParallel CommscommsTimerCloses #562.