Skip executing large aliased constant folds that 4531 will not install. - #4597
Draft
micwill755 wants to merge 1 commit into
Draft
Skip executing large aliased constant folds that 4531 will not install.#4597micwill755 wants to merge 1 commit into
micwill755 wants to merge 1 commit into
Conversation
github-actions
Bot
added
component: tests
2 tasks
micwill755
force-pushed
the
skip-aliased-fold-execute
branch
from
September 1, 2026 18:30
d844300 to
b61b68c
Compare
Inductor's skip_folding_node_fn never fires on our cf.run() path; honor it so large weight permutes/views are not computed, then discarded.
micwill755
force-pushed
the
skip-aliased-fold-execute
branch
from
September 2, 2026 07:06
b61b68c to
cacfbc5
Compare
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
skip_folding_node_fnduringconstant_fold()so large aliased weight permutes/views are not executed, then thrown away.cpu().contiguous()). This skips computing them:_TorchTensorRTConstantFolder.run_nodereturns unknown without calling the op when the node is a true view of a large module tensor.aten.contiguousand other materializing ops still run. The 4531 install-time skip stays as a safety net.PR 6 of the compile-startup stack; based on #4596.
Test plan
tests/py/dynamo/lowering/test_skip_aliased_fold_execute.py(large permute skipped and not executed, small permute still folds, large contiguous still executes, materializedweight + weightstill folds)ConstantFolder.runis ~50 ms, of which the aliased permutes are ~2 ms (4531 already took the ~0.25 scpu().contiguous()copies)Stack created with GitHub Stacks CLI • Give Feedback 💬