Don't uselessly recompute axiom dict every static eval call #135429

ezyang · 2024-09-08T03:34:21Z

Stack from ghstack (oldest at bottom):

-> Don't uselessly recompute axiom dict every static eval call #135429

Signed-off-by: Edward Z. Yang [email protected]

cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames @rec

[ghstack-poisoned]

pytorch-bot · 2024-09-08T03:34:24Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/135429

📄 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 or our office hours

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

❌ 1 New Failure

As of commit 5cbd75f with merge base 94e341c ():

NEW FAILURE - The following job has failed:

trunk / macos-py3-arm64 / test (default, 2, 3, macos-m1-stable) (gh)
export/test_serdes.py::SerDesExportNonStrictTestExport::test_slice_with_floordiv_serdes_non_strict

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

Signed-off-by: Edward Z. Yang <[email protected]> ghstack-source-id: 2dc16a9 Pull Request resolved: #135429

Signed-off-by: Edward Z. Yang <[email protected]> ghstack-source-id: 2dc16a9 Pull Request resolved: pytorch#135429

[ghstack-poisoned]

Signed-off-by: Edward Z. Yang <[email protected]> ghstack-source-id: eee20ea Pull Request resolved: #135429

ezyang · 2024-09-27T03:55:45Z

@pytorchbot merge

pytorchmergebot · 2024-09-27T03:57:56Z

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

malfet · 2024-09-27T15:35:24Z

@ezyang do you want to merge -f it if you are confident it'll fix the problem? (I'm still not sure if it was caused by this or previous PR in the stack)

pytorchmergebot · 2024-09-27T16:01:05Z

Merge failed

Reason: 1 jobs have failed, first few of them are: inductor-periodic / cuda12.1-py3.10-gcc9-sm80 / test (inductor_torchbench_smoketest_perf, 1, 1, linux.gcp.a100)

Details for Dev Infra team

Raised by workflow job

ezyang · 2024-09-27T16:27:42Z

I confirmed the previous commit on the stack broke the particular test case, but I'm not particularly in a hurry for this one so sure let's wait for a validation again

ezyang · 2024-09-27T16:28:04Z

@pytorchbot merge

pytorchmergebot · 2024-09-27T16:29:43Z

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

pytorchmergebot · 2024-09-27T22:28:29Z

The merge job was canceled or timed out. This most often happen if two merge requests were issued for the same PR, or if merge job was waiting for more than 6 hours for tests to finish. In later case, please do not hesitate to reissue the merge command
For more information see pytorch-bot wiki.

ezyang · 2024-09-28T20:58:13Z

@pytorchbot merge -f "looks fine"

pytorchmergebot · 2024-09-28T20:59:51Z

Merge started

Your change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use -f as last resort and instead consider -i/--ignore-current to continue the merge ignoring current failures. This will allow currently pending tests to finish and report signal before the merge.

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

ezyang · 2024-09-30T21:30:07Z

@pytorchbot revert -c nosignal -m "apparently this breaks executorch"

pytorchmergebot · 2024-09-30T21:31:42Z

@pytorchbot successfully started a revert job. Check the current status here.
Questions? Feedback? Please reach out to the PyTorch DevX Team

pytorchmergebot · 2024-09-30T21:31:53Z

Reverting PR 135429 failed

Reason: Command git -C /home/runner/work/pytorch/pytorch revert --no-edit 1d6e0412f5205b1cd709e034526d7f21d6f2d56f returned non-zero exit code 1

Auto-merging test/dynamo/test_misc.py
Auto-merging torch/fx/experimental/symbolic_shapes.py
CONFLICT (content): Merge conflict in torch/fx/experimental/symbolic_shapes.py
error: could not revert 1d6e0412f5... Don't uselessly recompute axiom dict every static eval call (#135429)
hint: After resolving the conflicts, mark them with
hint: "git add/rm <pathspec>", then run
hint: "git revert --continue".
hint: You can instead skip this commit with "git revert --skip".
hint: To abort and get back to the state before "git revert",
hint: run "git revert --abort".
hint: Disable this message with "git config advice.mergeConflict false"

Details for Dev Infra team

Raised by workflow job

ezyang · 2024-09-30T22:27:26Z

@pytorchbot revert -c nosignal -m "try again"

pytorchmergebot · 2024-09-30T22:29:05Z

@pytorchbot successfully started a revert job. Check the current status here.
Questions? Feedback? Please reach out to the PyTorch DevX Team

pytorchmergebot · 2024-09-30T22:29:16Z

@ezyang your PR has been successfully reverted.

…135429)" This reverts commit 1d6e041. Reverted #135429 on behalf of https://github.com/ezyang due to try again ([comment](#135429 (comment)))

ezyang · 2024-10-02T03:27:32Z

Notes on the executorch failure: It's whack. This is what the logs should be:

INFO:torch.fx.experimental.symbolic_shapes:create_symbol s1 = 2 for __meta_utils_unknown_tensor105.size()[0] [2, int_oo] (executorch/exir/pass_base.py:238 in make_val), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1"
INFO:torch.fx.experimental.symbolic_shapes:create_symbol s2 = 3 for __meta_utils_unknown_tensor105.size()[2] [2, int_oo] (executorch/exir/pass_base.py:238 in make_val), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s2"
INFO:torch.fx.experimental.symbolic_shapes:set_replacement s1 = 2 (range_refined_to_singleton) VR[2, 2]
INFO:torch.fx.experimental.symbolic_shapes:eval Eq(2, s1) [guard added] (_subclasses/fake_impls.py:715 in conv), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(2, s1)"
INFO:torch.fx.experimental.symbolic_shapes:eval 6 >= s2 [guard added] (_subclasses/fake_impls.py:715 in conv), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="6 >= s2"
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(7 - s2, 1) [guard added] (_prims_common/__init__.py:457 in compute_elementwise_output_logical_to_physical_perm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(7 - s2, 1)"
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(-2*s0*s2 + 14*s0, 0) [guard added] (_prims_common/__init__.py:454 in compute_elementwise_output_logical_to_physical_perm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(-2*s0*s2 + 14*s0, 0)"
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(7 - s2, 1) [guard added] (_prims/__init__.py:1285 in _broadcast_in_dim_meta), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(7 - s2, 1)"
INFO:torch.fx.experimental.symbolic_shapes:eval 9 - s2 >= s2 [guard added] (_subclasses/fake_impls.py:715 in conv), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="9 - s2 >= s2"
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(-2*s0*s2 + 14*s0, 0) [guard added] (_subclasses/fake_impls.py:715 in conv), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(-2*s0*s2 + 14*s0, 0)"
INFO:torch.fx.experimental.symbolic_shapes:runtime_assert 10 - 2*s2 >= 0 [guard added] (_refs/__init__.py:4758 in new_empty), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="10 - 2*s2 >= 0"
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(10 - 2*s2, 0) [guard added] (_meta_registrations.py:2273 in meta_conv), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(10 - 2*s2, 0)"
INFO:torch.fx.experimental.symbolic_shapes:eval 1 < 20 - 4*s2 [guard added] (_meta_registrations.py:2273 in meta_conv), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="1 < 20 - 4*s2"
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(-4*s0*s2 + 20*s0, 0) [guard added] (_prims_common/__init__.py:454 in compute_elementwise_output_logical_to_physical_perm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(-4*s0*s2 + 20*s0, 0)"
INFO:torch.fx.experimental.symbolic_shapes:runtime_assert Eq(7 - s2, 10 - 2*s2) [guard added] (_subclasses/fake_impls.py:817 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(7 - s2, 10 - 2*s2)"
INFO:torch.fx.experimental.symbolic_shapes:set_replacement s2 = 3 (range_refined_to_singleton) VR[3, 3]
INFO:torch.fx.experimental.symbolic_shapes:eval Eq(s2, 3) [guard added] (executorch/exir/pass_base.py:678 in migrate_meta_val), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(s2, 3)"
E 00:00:17.357053 executorch:method.cpp:941] Output 0 is memory planned, or is a constant. Cannot override         the existing data pointer.

after this PR, the logs become

INFO:torch.fx.experimental.symbolic_shapes:create_symbol s0 = 8 for __meta_utils_unknown_tensor12.size()[0] [2, int_oo] (utils/_pytree.py:803 in unflatten), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s0"      
INFO:torch.fx.experimental.symbolic_shapes:create_symbol s1 = 16 for __meta_utils_unknown_tensor12.size()[1] [2, int_oo] (utils/_pytree.py:803 in unf
latten), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1"                                                                        INFO:torch.fx.experimental.symbolic_shapes:create_symbol s2 = 32 for __meta_utils_unknown_tensor12.size()[2] [2, int_oo] (utils/_pytree.py:803 in unf
latten), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s2"       
INFO:torch.fx.experimental.symbolic_shapes:create_symbol s3 = 64 for __meta_utils_unknown_tensor12.size()[3] [2, int_oo] (utils/_pytree.py:803 in unflatten), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s3"      
INFO:torch.fx.experimental.symbolic_shapes:set_replacement s2 = 32 (range_refined_to_singleton) VR[32, 32]
INFO:torch.fx.experimental.symbolic_shapes:eval Eq(32, s2) [guard added] (<eval_with_key>.560:9 in forward), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(32, s2)"                                                                                                                   
INFO:torch.fx.experimental.symbolic_shapes:create_symbol s1 = 2 for __meta_utils_unknown_tensor105.size()[0] [2, int_oo] (executorch/exir/pass_base.p
y:238 in make_val), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s1"                                                             INFO:torch.fx.experimental.symbolic_shapes:create_symbol s2 = 3 for __meta_utils_unknown_tensor105.size()[2] [2, int_oo] (executorch/exir/pass_base.p
y:238 in make_val), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL="s2"
INFO:torch.fx.experimental.symbolic_shapes:set_replacement s1 = 2 (range_refined_to_singleton) VR[2, 2]                                              INFO:torch.fx.experimental.symbolic_shapes:eval Eq(2, s1) [guard added] (_subclasses/fake_impls.py:715 in conv), for more info run with TORCHDYNAMO_E
XTENDED_DEBUG_GUARD_ADDED="Eq(2, s1)"                               
INFO:torch.fx.experimental.symbolic_shapes:eval 6 >= s2 [guard added] (_subclasses/fake_impls.py:715 in conv), for more info run with TORCHDYNAMO_EXT
ENDED_DEBUG_GUARD_ADDED="6 >= s2"                                         
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(7 - s2, 1) [guard added] (_prims_common/__init__.py:457 in compute_elementwise_output_logical_to_physical_perm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(7 - s2, 1)"
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(-2*s0*s2 + 14*s0, 0) [guard added] (_prims_common/__init__.py:454 in compute_elementwise_output_lo
gical_to_physical_perm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(-2*s0*s2 + 14*s0, 0)"                                     INFO:torch.fx.experimental.symbolic_shapes:eval 9 - s2 >= s2 [guard added] (_subclasses/fake_impls.py:715 in conv), for more info run with TORCHDYNAM
O_EXTENDED_DEBUG_GUARD_ADDED="9 - s2 >= s2"                    
INFO:torch.fx.experimental.symbolic_shapes:runtime_assert 10 - 2*s2 >= 0 [guard added] (_refs/__init__.py:4758 in new_empty), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="10 - 2*s2 >= 0"                                                                                              
INFO:torch.fx.experimental.symbolic_shapes:eval Ne(10 - 2*s2, 0) [guard added] (_meta_registrations.py:2273 in meta_conv), for more info run with TOR
CHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(10 - 2*s2, 0)"                                                                                               
INFO:torch.fx.experimental.symbolic_shapes:eval 1 < 20 - 4*s2 [guard added] (_meta_registrations.py:2273 in meta_conv), for more info run with TORCHD
YNAMO_EXTENDED_DEBUG_GUARD_ADDED="1 < 20 - 4*s2"                                                                                                     INFO:torch.fx.experimental.symbolic_shapes:eval Ne(-4*s0*s2 + 20*s0, 0) [guard added] (_prims_common/__init__.py:454 in compute_elementwise_output_lo
gical_to_physical_perm), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Ne(-4*s0*s2 + 20*s0, 0)"
INFO:torch.fx.experimental.symbolic_shapes:runtime_assert Eq(7 - s2, 10 - 2*s2) [guard added] (_subclasses/fake_impls.py:817 in infer_size), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="Eq(7 - s2, 10 - 2*s2)"      
INFO:torch.fx.experimental.symbolic_shapes:eval -2*s0*s2 + 14*s0 >= 2 [guard added] (_prims_common/__init__.py:236 in is_contiguous), for more info r
un with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="-2*s0*s2 + 14*s0 >= 2"                                                                               INFO:torch.fx.experimental.symbolic_shapes:eval -4*s0*s2 + 20*s0 >= 2 [guard added] (_prims_common/__init__.py:236 in is_contiguous), for more info r
un with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="-4*s0*s2 + 20*s0 >= 2"
INFO:torch.fx.experimental.symbolic_shapes:eval 20 - 4*s2 > 2 [guard added] (_prims_common/__init__.py:479 in should_swap), for more info run with TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED="20 - 4*s2 > 2"

and then it fails because we never manage the replacement on s2. Strange strange...

https://fb.workplace.com/groups/pytorch.edge.users/permalink/1605864123617208/

…ytorch#135429)" This reverts commit 1d6e041. Reverted pytorch#135429 on behalf of https://github.com/ezyang due to try again ([comment](pytorch#135429 (comment)))

ezyang · 2024-10-08T14:37:41Z

gonna chat with @angelayi about the high level pass structure

Signed-off-by: Edward Z. Yang <ezyangmeta.com> cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames rec [ghstack-poisoned]

Signed-off-by: Edward Z. Yang <[email protected]> ghstack-source-id: 81ee6b3 Pull Request resolved: #135429

laithsakka · 2024-10-26T16:26:59Z

looking at this now

Update

a22adbe

[ghstack-poisoned]

pytorch-bot bot added ciflow/inductor release notes: fx release notes category labels Sep 8, 2024

ezyang added a commit that referenced this pull request Sep 8, 2024

Don't uselessly recompute axiom dict every static eval call

ffc33d5

Signed-off-by: Edward Z. Yang <[email protected]> ghstack-source-id: 2dc16a9 Pull Request resolved: #135429

github-actions bot requested review from SherlockNoMad, albanD, antoniojkim, bdhirsh and miladm September 8, 2024 03:34

ezyang mentioned this pull request Sep 11, 2024

Simplify expr before getting implications in _maybe_evaluate_static #135499

Closed

albanD removed their request for review September 13, 2024 19:38

ezyang mentioned this pull request Sep 8, 2024

Refactor maybe_evaluate_static into a worker function off of ShapeEnv #135107

Closed

isuruf pushed a commit to isuruf/pytorch that referenced this pull request Sep 23, 2024

Don't uselessly recompute axiom dict every static eval call

c809c09

Signed-off-by: Edward Z. Yang <[email protected]> ghstack-source-id: 2dc16a9 Pull Request resolved: pytorch#135429

ezyang mentioned this pull request Sep 8, 2024

Deal with size oblivious before going into worker #135137

Closed

Update

b32194e

[ghstack-poisoned]

pytorch-bot bot added the module: dynamo label Sep 25, 2024

Update

b6f8fbc

[ghstack-poisoned]

ezyang added a commit that referenced this pull request Sep 25, 2024

Don't uselessly recompute axiom dict every static eval call

66ac7e5

Signed-off-by: Edward Z. Yang <[email protected]> ghstack-source-id: eee20ea Pull Request resolved: #135429

ezyang requested review from avikchaudhuri, isuruf and pianpwk September 25, 2024 23:36

ezyang added topic: not user facing topic category ciflow/trunk Trigger trunk jobs on your pull request labels Sep 25, 2024

isuruf approved these changes Sep 26, 2024

View reviewed changes

pytorchmergebot added the merging label Sep 27, 2024

pytorchmergebot added the Merged label Sep 27, 2024

pytorchmergebot removed the merging label Sep 27, 2024

pytorchmergebot added the merging label Sep 27, 2024

pytorchmergebot closed this in 1d6e041 Sep 28, 2024

pytorchmergebot removed the merging label Sep 28, 2024

pytorchmergebot reopened this Sep 30, 2024

Update on "Don't uselessly recompute axiom dict every static eval call"

5cbd75f

Signed-off-by: Edward Z. Yang <ezyangmeta.com> cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx chenyang78 kadeng chauhang amjames rec [ghstack-poisoned]

laithsakka added a commit that referenced this pull request Oct 26, 2024

Don't uselessly recompute axiom dict every static eval call

d6bdaf9

Signed-off-by: Edward Z. Yang <[email protected]> ghstack-source-id: 81ee6b3 Pull Request resolved: #135429

ezyang mentioned this pull request Sep 8, 2024

Add scribe logging for maybe_evaluate_static_worker #135138

Closed

ezyang closed this Oct 28, 2024

github-actions bot deleted the gh/ezyang/2926/head branch November 28, 2024 02:13

Don't uselessly recompute axiom dict every static eval call #135429

Don't uselessly recompute axiom dict every static eval call #135429

Uh oh!

Conversation

ezyang commented Sep 8, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Sep 8, 2024 • 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/135429

❌ 1 New Failure

Uh oh!

ezyang commented Sep 27, 2024

Uh oh!

pytorchmergebot commented Sep 27, 2024

Merge started

Uh oh!

malfet commented Sep 27, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorchmergebot commented Sep 27, 2024

Merge failed

Uh oh!

ezyang commented Sep 27, 2024

Uh oh!

ezyang commented Sep 27, 2024

Uh oh!

pytorchmergebot commented Sep 27, 2024

Merge started

Uh oh!

pytorchmergebot commented Sep 27, 2024

Uh oh!

ezyang commented Sep 28, 2024

Uh oh!

pytorchmergebot commented Sep 28, 2024

Merge started

Uh oh!

ezyang commented Sep 30, 2024

Uh oh!

pytorchmergebot commented Sep 30, 2024

Uh oh!

pytorchmergebot commented Sep 30, 2024

Reverting PR 135429 failed

Uh oh!

ezyang commented Sep 30, 2024

Uh oh!

pytorchmergebot commented Sep 30, 2024

Uh oh!

pytorchmergebot commented Sep 30, 2024

Uh oh!

ezyang commented Oct 2, 2024

Uh oh!

ezyang commented Oct 8, 2024

Uh oh!

laithsakka commented Oct 26, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

ezyang commented Sep 8, 2024 •

edited

Loading

pytorch-bot bot commented Sep 8, 2024 •

edited

Loading

malfet commented Sep 27, 2024 •

edited

Loading