Fix _del_library #150495

zou3519 · 2025-04-02T01:44:49Z

Stack from ghstack (oldest at bottom):

On library deletion, we need to clear fx's schema cache.

Test Plan:

top PR in the stack, I don't have a good test case for this PR.

On library deletion, we need to clear fx's schema cache. Test Plan: - next PR up in the stack, I don't have a good test case for this PR. [ghstack-poisoned]

pytorch-bot · 2025-04-02T01:44:52Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/150495

📄 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.

✅ No Failures

As of commit 6a0c206 with merge base 58ede0c ():
💚 Looks good so far! There are no failures yet. 💚

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

On library deletion, we need to clear fx's schema cache. Test Plan: - top PR in the stack, I don't have a good test case for this PR. [ghstack-poisoned]

pytorchmergebot · 2025-04-09T02:08:56Z

Starting merge as part of PR stack under #148104

This PR: - cleans up some existing comments that don't make sense anymore - hooks up the "custom_op_default_layout_constraint" back (that seems to have broken) - cleans up the "lazy registration path" which seems to never get hit anymore - adds dislike_padding to nodes that require exact strides Test Plan: - tests + CI disable padding Pull Request resolved: #148104 Approved by: https://github.com/shunting314, https://github.com/eellison ghstack dependencies: #150495

If a tag is not specified on a custom operator, then inductor will assume that it needs exact strides. Test Plan: - tests + CI Pull Request resolved: #150511 Approved by: https://github.com/eellison, https://github.com/shunting314 ghstack dependencies: #150495, #148104

On library deletion, we need to clear fx's schema cache. Test Plan: - top PR in the stack, I don't have a good test case for this PR. Pull Request resolved: pytorch#150495 Approved by: https://github.com/eellison

This PR: - cleans up some existing comments that don't make sense anymore - hooks up the "custom_op_default_layout_constraint" back (that seems to have broken) - cleans up the "lazy registration path" which seems to never get hit anymore - adds dislike_padding to nodes that require exact strides Test Plan: - tests + CI disable padding Pull Request resolved: pytorch#148104 Approved by: https://github.com/shunting314, https://github.com/eellison ghstack dependencies: pytorch#150495

) If a tag is not specified on a custom operator, then inductor will assume that it needs exact strides. Test Plan: - tests + CI Pull Request resolved: pytorch#150511 Approved by: https://github.com/eellison, https://github.com/shunting314 ghstack dependencies: pytorch#150495, pytorch#148104

On library deletion, we need to clear fx's schema cache. Test Plan: - top PR in the stack, I don't have a good test case for this PR. Pull Request resolved: pytorch#150495 Approved by: https://github.com/eellison

This PR: - cleans up some existing comments that don't make sense anymore - hooks up the "custom_op_default_layout_constraint" back (that seems to have broken) - cleans up the "lazy registration path" which seems to never get hit anymore - adds dislike_padding to nodes that require exact strides Test Plan: - tests + CI disable padding Pull Request resolved: pytorch#148104 Approved by: https://github.com/shunting314, https://github.com/eellison ghstack dependencies: pytorch#150495

) If a tag is not specified on a custom operator, then inductor will assume that it needs exact strides. Test Plan: - tests + CI Pull Request resolved: pytorch#150511 Approved by: https://github.com/eellison, https://github.com/shunting314 ghstack dependencies: pytorch#150495, pytorch#148104

aorenste · 2025-04-21T23:25:32Z

@zou3519 This seems to have broken something with TORCH_LOGS=help:

(py39) $ TORCH_LOGS=help python test/inductor/test_codecache.py -k test_cache_hot_load_pgo_swap_file_names
<stuff snipped>
Traceback (most recent call last):
  File "/home/aorenste/.conda/envs/py39/lib/python3.9/weakref.py", line 667, in _exitfunc
    f()
  File "/home/aorenste/.conda/envs/py39/lib/python3.9/weakref.py", line 591, in __call__
    return info.func(*info.args, **(info.kwargs or {}))
  File "/home/aorenste/local/pytorch/torch/library.py", line 446, in _del_library
    import torch.fx
  File "/home/aorenste/local/pytorch/torch/__init__.py", line 2643, in <module>
    class _TritonLibrary:
  File "/home/aorenste/local/pytorch/torch/__init__.py", line 2644, in _TritonLibrary
    lib = torch.library.Library("triton", "DEF")
  File "/home/aorenste/local/pytorch/torch/library.py", line 109, in __init__
    self.m: Optional[Any] = torch._C._dispatch_library(
RuntimeError: Only a single TORCH_LIBRARY can be used to register the namespace triton; please put all of your definitions in a single TORCH_LIBRARY block.  If you were trying to specify implementations,
consider using TORCH_LIBRARY_IMPL (which can be duplicated).  If you really intended to define operators for a single namespace in a distributed way, you can use TORCH_LIBRARY_FRAGMENT to explicitly
indicate this.  Previous registration of TORCH_LIBRARY was registered at /dev/null:2643; latest registration was registered at /dev/null:2643

(the error is repeated a bunch of times)

zou3519 · 2025-04-21T23:49:53Z

We are forward fixing this in #151678

On library deletion, we need to clear fx's schema cache. Test Plan: - top PR in the stack, I don't have a good test case for this PR. ghstack-source-id: bd75d3e Pull Request resolved: pytorch/pytorch#150495

If a tag is not specified on a custom operator, then inductor will assume that it needs exact strides. Test Plan: - tests + CI Pull Request resolved: #150511 Approved by: https://github.com/eellison, https://github.com/shunting314 ghstack dependencies: #150495, #148104 ghstack-source-id: 7192ed4

If a tag is not specified on a custom operator, then inductor will assume that it needs exact strides. Test Plan: - tests + CI Pull Request resolved: #150511 Approved by: https://github.com/eellison, https://github.com/shunting314 ghstack dependencies: #150495, #148104 ghstack-source-id: 923efe0

If a tag is not specified on a custom operator, then inductor will assume that it needs exact strides. Test Plan: - tests + CI Pull Request resolved: #150511 Approved by: https://github.com/eellison, https://github.com/shunting314 ghstack dependencies: #150495, #148104 ghstack-source-id: f095c0a

Fix _del_library

dcc88a7

On library deletion, we need to clear fx's schema cache. Test Plan: - next PR up in the stack, I don't have a good test case for this PR. [ghstack-poisoned]

This was referenced Apr 2, 2025

Change arg_kwarg_vals propagation strategy #148046

Closed

Add needs_exact_strides operator tag for Inductor to force exact strides #148063

Closed

This was referenced Apr 2, 2025

Implement needs_exact_strides for mutable custom operators #148091

Closed

Rename node.meta["arg_kwarg_vals"] to node.meta["eager_input_vals"] #148092

Closed

Refactor layout constraint selection logic #148104

Closed

Update on "Fix _del_library"

f1a04dc

On library deletion, we need to clear fx's schema cache. Test Plan: - top PR in the stack, I don't have a good test case for this PR. [ghstack-poisoned]

zou3519 mentioned this pull request Apr 2, 2025

Inductor respects exact strides on custom ops by default #150511

Closed

zou3519 added the topic: not user facing topic category label Apr 2, 2025

zou3519 requested review from eellison and williamwen42 April 2, 2025 13:23

Update on "Fix _del_library"

ae8e07f

On library deletion, we need to clear fx's schema cache. Test Plan: - top PR in the stack, I don't have a good test case for this PR. [ghstack-poisoned]

eellison approved these changes Apr 2, 2025

View reviewed changes

zou3519 added 2 commits April 2, 2025 11:08

Update on "Fix _del_library"

ac35d46

On library deletion, we need to clear fx's schema cache. Test Plan: - top PR in the stack, I don't have a good test case for this PR. [ghstack-poisoned]

Update on "Fix _del_library"

6a0c206

On library deletion, we need to clear fx's schema cache. Test Plan: - top PR in the stack, I don't have a good test case for this PR. [ghstack-poisoned]

pytorchmergebot closed this in 44deb67 Apr 9, 2025

pytorchmergebot added the Merged label Apr 9, 2025

github-actions bot deleted the gh/zou3519/1151/head branch May 25, 2025 02:21

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix _del_library #150495

Fix _del_library #150495

Uh oh!

zou3519 commented Apr 2, 2025 •

edited

Loading

Uh oh!

pytorch-bot bot commented Apr 2, 2025 •

edited

Loading

Uh oh!

pytorchmergebot commented Apr 9, 2025

Uh oh!

aorenste commented Apr 21, 2025 •

edited

Loading

Uh oh!

zou3519 commented Apr 21, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Fix _del_library #150495

Fix _del_library #150495

Uh oh!

Conversation

zou3519 commented Apr 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Apr 2, 2025 • 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/150495

✅ No Failures

Uh oh!

pytorchmergebot commented Apr 9, 2025

Uh oh!

aorenste commented Apr 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

zou3519 commented Apr 21, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

zou3519 commented Apr 2, 2025 •

edited

Loading

pytorch-bot bot commented Apr 2, 2025 •

edited

Loading

aorenste commented Apr 21, 2025 •

edited

Loading