[RF] Speed up CodegenContext::beginLoop() with single graph traversal - #23313
Merged
Merged
Conversation
To figure out which vector observables to loop over, beginLoop() called RooAbsArg::dependsOn() once per vector observable. These dependsOn() walks don't deduplicate visited nodes, so their cost scales with the number of paths in the computation graph instead of the number of nodes, which gets very expensive for large models with many shared nodes. Replace the per-observable walks with a single depth-first traversal that visits every node only once and collects the reachable vector observables, which is equivalent. For the ATLAS VHbb benchmark workspace from rootbench, this reduces the "Function JIT time" of the codegen backend (which includes the code generation itself) from 9.4 s to 5.3 s. 🤖 Done with the help of AI
vgvassilev
approved these changes
Sep 9, 2026
vgvassilev
left a comment
vgvassilev
left a comment
Member
There was a problem hiding this comment.
Oh, wow, pretty significant improvement. Can we put some visitation traces so that we can detect if we are doing something else suboptimal elsewhere? LGTM!
Contributor
Author
Right now I think we're good. This was the one big overhead that the profiling found. Otherwise, the time spent to setup the codegen likelihood is mostly compiling the code. |
Contributor
Author
|
/backport to 6.40 |
|
Preparing to backport PR #23313 to branch 6.40 requested by guitargeek |
|
This PR has been backported to branch 6.40: #23326 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
To figure out which vector observables to loop over, beginLoop() called RooAbsArg::dependsOn() once per vector observable. These dependsOn() walks don't deduplicate visited nodes, so their cost scales with the number of paths in the computation graph instead of the number of nodes, which gets very expensive for large models with many shared nodes.
Replace the per-observable walks with a single depth-first traversal that visits every node only once and collects the reachable vector observables, which is equivalent.
For the ATLAS VHbb benchmark workspace from rootbench, this reduces the "Function JIT time" of the codegen backend (which includes the code generation itself) from 9.4 s to 5.3 s.
🤖 Done with the help of AI