[dynamo] Add min/max bounds support to mark_unbacked by laithsakka · Pull Request #176313 · pytorch/pytorch

laithsakka · 2026-03-03T17:42:27Z

Stack from ghstack (oldest at bottom):

Adds min and max parameters to mark_unbacked() to allow specifying
value range constraints for unbacked dimensions.

When an unbacked symbol is created with min/max bounds, torch._check()
calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)).

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @kadeng @chauhang @amjames @Lucaskabela @jataylo

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. [ghstack-poisoned]

pytorch-bot · 2026-03-03T17:42:32Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/176313

📄 Preview Python docs built from this PR
📄 Preview C++ docs built from this PR
❓ Need help or want to give feedback on the CI? Visit the bot commands wiki

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (1 Unrelated Failure)

As of commit 2e74e51 with merge base 07efc60 ():

UNSTABLE - The following job is marked as unstable, possibly due to flakiness on trunk:

inductor / inductor-cpu-test / test (cpu_inductor_torchbench, 1, 2, linux.2xlarge.amx, unstable) (gh) (#174929)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. ghstack-source-id: 5f2f610 Pull Request resolved: #176313

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. [ghstack-poisoned]

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. ghstack-source-id: 5f2f610 Pull Request resolved: #176313

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. [ghstack-poisoned]

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. ghstack-source-id: 0fbc823 Pull Request resolved: #176313

Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). [ghstack-poisoned]

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. ghstack-source-id: e057ce9 Pull Request resolved: #176313

Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). [ghstack-poisoned]

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. ghstack-source-id: 97416af Pull Request resolved: #176313

Lucaskabela · 2026-03-04T16:33:52Z

torch/_dynamo/guards.py

+                        guard.user_stack,
+                    )
+
+                # TODO we dont have guards on _dynamo_unbacked_indices like those of _dynamo_dynamic_indices this seems wrong!!


Do we still need this todo now that we have guards? Or is this particularly saying these guards differ from dynamic_indices so need to change in the future?

yeh we need those i am doing it in the PR on top of this basically
what this comment say

if we do mark_unbacked(x, [0,1]) then mark_unbacked(x, [0,1,2]), we do not recompile since we do not add any guards on _dynamo_unbacked_indices value.

Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). [ghstack-poisoned]

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. ghstack-source-id: 9eb449b Pull Request resolved: #176313

Lucaskabela

Thanks for the change - this looks good to me

laithsakka · 2026-03-09T21:36:48Z

@pytorchbot merge

pytorchmergebot · 2026-03-09T21:39:00Z

Merge started

Your change will be merged once all checks pass (ETA 0-4 Hours).

Learn more about merging in the wiki.

Questions? Feedback? Please reach out to the PyTorch DevX Team

Advanced Debugging

Check the merge workflow status
here

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). X-link: pytorch/pytorch#176313 Approved by: https://github.com/Lucaskabela, https://github.com/mlazos ghstack dependencies: #175946, #175956 Reviewed By: huydhn Differential Revision: D95914182 fbshipit-source-id: 0acdb0d80e3c4710a57deb70ebfbefe32af86f93

Summary: Adds min and max parameters to mark_unbacked() to allow specifying value range constraints for unbacked dimensions. When an unbacked symbol is created with min/max bounds, torch._check() calls are emitted to establish the constraints (e.g., torch._check(u0 >= 1)). This allows the optimizer to reason about unbacked symbols the same way it does for backed symbols, enabling constant folding and other optimizations that require knowing value ranges. Test Plan: Added test_mark_unbacked_min_max_bounds in test_dynamic_shapes.py. ghstack-source-id: 2304ec2 Pull Request resolved: pytorch/pytorch#176313

laithsakka requested a review from bobrenjc93 as a code owner March 3, 2026 17:42

This was referenced Mar 3, 2026

symbolic shapes guarding_hint_or_throw and optimization_hint #175497

Closed

Handle unabcked better in pad_mm pass #175824

Open

pytorch-bot bot added ciflow/inductor module: dynamo release notes: fx release notes category labels Mar 3, 2026

laithsakka mentioned this pull request Mar 3, 2026

[inductor] Use all_unbacked_explicitly_hinted in SIMD scheduling #176319

Open

laithsakka added 2 commits March 3, 2026 10:37

laithsakka added 2 commits March 3, 2026 11:26

laithsakka marked this pull request as draft March 3, 2026 19:28

laithsakka mentioned this pull request Mar 3, 2026

revisit guarding in mark dynamic APIs #176341

Open

laithsakka requested review from anijain2305, aorenste and ezyang March 4, 2026 00:43

laithsakka mentioned this pull request Mar 4, 2026

add missing guards for _dynamo_hint_overrides #176360

Open

laithsakka marked this pull request as ready for review March 4, 2026 04:27

Lucaskabela reviewed Mar 4, 2026

View reviewed changes

laithsakka requested review from Lucaskabela and zou3519 March 5, 2026 00:33

Lucaskabela approved these changes Mar 5, 2026

View reviewed changes

mlazos approved these changes Mar 8, 2026

View reviewed changes

pytorch-bot bot added the ciflow/trunk Trigger trunk jobs on your pull request label Mar 9, 2026

pytorchmergebot added the merging label Mar 9, 2026

pytorchmergebot closed this in 4ae45dd Mar 10, 2026

pytorchmergebot added Merged and removed merging labels Mar 10, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[dynamo] Add min/max bounds support to mark_unbacked#176313

[dynamo] Add min/max bounds support to mark_unbacked#176313
laithsakka wants to merge 8 commits intogh/laithsakka/414/basefrom
gh/laithsakka/414/head

laithsakka commented Mar 3, 2026 •

edited by pytorch-bot bot

Loading

Uh oh!

pytorch-bot bot commented Mar 3, 2026 •

edited

Loading

Uh oh!

Lucaskabela Mar 4, 2026

Uh oh!

laithsakka Mar 5, 2026

Uh oh!

Lucaskabela left a comment

Uh oh!

laithsakka commented Mar 9, 2026

Uh oh!

pytorchmergebot commented Mar 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

laithsakka commented Mar 3, 2026 • edited by pytorch-bot bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Mar 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/176313

✅ You can merge normally! (1 Unrelated Failure)

Uh oh!

Lucaskabela Mar 4, 2026

Choose a reason for hiding this comment

Uh oh!

laithsakka Mar 5, 2026

Choose a reason for hiding this comment

Uh oh!

Lucaskabela left a comment

Choose a reason for hiding this comment

Uh oh!

laithsakka commented Mar 9, 2026

Uh oh!

pytorchmergebot commented Mar 9, 2026

Merge started

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

laithsakka commented Mar 3, 2026 •

edited by pytorch-bot bot

Loading

pytorch-bot bot commented Mar 3, 2026 •

edited

Loading