Constant fold branches of variadic add/mul #1422
Open
+104
−29
Conversation
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
lucianopaz
reviewed
May 29, 2025
| nb_inputs = len(node.inputs) | ||
| max_inputs = float("inf") | ||
| if hasattr(node.op, "max_inputs"): | ||
| max_inputs = node.op.max_inputs(node) |
| # Do not duplicate the operation. | ||
| len(fgraph.clients[inp]) == 1 | ||
| and (nb_inputs + len(inp.owner.inputs) - 1) <= max_inputs |
|
|
||
| fuse_seqopt.register( | ||
| "local_add_mul_fusion", | ||
| EquilibriumGraphRewriter(rewriters=[local_add_mul_fusion], max_use_ratio=1000), |
| ) | ||
| add_mul_flat_seqopt.register( | ||
| constant_fold_branches_of_add_mul.__name__, | ||
| in2out(constant_fold_branches_of_add_mul, ignore_newtrees=True), |
There was a problem hiding this comment.
Why should ignore new trees be true?
|
@ricardoV94, I just went through your branch's code and found that the error is coming from the fact that the |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Refactoring and renaming:
local_add_mul_fusionfunction toflatten_nested_add_multo better reflect its purpose of flattening nested add/mul operations. The function now explicitly tracksaddandmuloperations instead of relying on genericElemwisechecks. [1] [2] [3]New optimization for constant folding:
Introduced a new rewrite function,
constant_fold_branches_of_add_mul, which folds constants in add/mul operations when it does not result in higher intermediate memory usage. This optimization is registered in a new sequence database,add_mul_flat_seqopt, which runs before generic elementwise fusion.This is pulled out of a separate database so it's included in JAX rewrites (JAX does not include fusion rewrites). We've found this could help avoding XLA constant fold (CC @lucianopaz)
📚 Documentation preview 📚: https://pytensor--1422.org.readthedocs.build/en/1422/