Skip to content

Better local group selection#1162

Merged
dfellis merged 4 commits into
mainfrom
better-local-group-selection
May 1, 2026
Merged

Better local group selection#1162
dfellis merged 4 commits into
mainfrom
better-local-group-selection

Conversation

@dfellis
Copy link
Copy Markdown
Member

@dfellis dfellis commented May 1, 2026

  • Better local group selection based on GPU capabilities and global workgroup size
  • Attempt to remove the hardwired local group sizing and the sentinel hackery for the window and frame

@dfellis dfellis self-assigned this May 1, 2026
Comment thread alan_std/src/lib.rs
// Clamp to [8, maxInvocations]
target = target.max(8).min(max_invocations);
// Snap to nearest multiple of 8 (hardware alignment)
target = ((target + 7) / 8) * 8;
Comment thread alan_std/src/lib.rs
if sqrt >= 8 && sqrt * sqrt == target {
return [sqrt as i64, sqrt as i64, 1];
}
if target % 8 == 0 {
@dfellis dfellis merged commit 1ccde42 into main May 1, 2026
13 checks passed
@dfellis dfellis deleted the better-local-group-selection branch May 1, 2026 06:02
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.

2 participants