Skip to content

gpu speed up for correct_velocity - #1905

Merged
brucefan1983 merged 1 commit into
masterfrom
correct_velocity_gpu
Nov 16, 2026
Merged

brucefan1983 merged 1 commit into
masterfrom
correct_velocity_gpu

Conversation

@brucefan1983

@brucefan1983 brucefan1983 commented Nov 16, 2026 •

Copy link
Copy Markdown
Owner

Summary

Move periodic correct_velocity calculations to the GPU while preserving the existing correction algorithm.

Changes

  • Use double-precision GPU reductions and reusable buffers for global and group-wise correction, eliminating per-atom host/device transfers.
  • Share the angular-velocity solver between CPU and GPU, checking singular inertia before division.

Validation

  • All 255 regression tests passed.
  • Overall MD throughput improved by approximately 2.8% at 128,000 atoms.

@brucefan1983
brucefan1983 marked this pull request as ready for review November 16, 2026 06:06
@brucefan1983
brucefan1983 merged commit 83ab561 into master Nov 16, 2026
2 checks passed
@brucefan1983
brucefan1983 deleted the correct_velocity_gpu branch November 16, 2026 06:20
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant