Skip to content

Conversation

@BramOtte
Copy link
Contributor

@BramOtte BramOtte commented Dec 8, 2025

Just a draft so people know this exists, its been quite a while since I have worked on this so I don't think its ready to merge.
It removes quite a few nodes and also improves runtime performance but significantly increases compile times.

@BramOtte BramOtte force-pushed the graph-optimizations-analysis-info branch from 883934b to e0f52dc Compare December 27, 2025 23:32
@BramOtte
Copy link
Contributor Author

I resolved the compile time issues with this PR now.
Previously for iris Mandelbrot it took 5 to 6 seconds to run the coalesce2 pass now its just ~350ms.
The ss_range_analysis pass also has a more reasonable compile time of 280ms which I think is still slower than master but it was closer to a whole second before and this pass is more effective than master.
I should be able to get this merge ready soon.

There are still a few things that keep this a draft:

  • The coalesce2 pass uses unsafe to reduce copies and allocations, this brings its compile time down from 700ms to 350ms but if there is a safe way to achieve similar results I'd prefer that, if anyone has any ideas I would like to know.
  • The coalesce2 pass currently does signal strength distance normalization, this could maybe be done in an earlier pass instead
  • The ss_range_analysis pass could be optimized further it currently iterates over all the nodes repeatedly
  • Ideally more than just iris is tested

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