[Inductor] Fix the High Order Op layout issue #128275
Closed
+29
−3
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
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/128275
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (3 Unrelated Failures)As of commit 9d761e1 with merge base a3af32c ( FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Fix the issue: #127995 - In current implementation, the device of the `NoneLayout` will be None when `example_output` returns from `cls.process_kernel` is None. The test reported in the issue is the case when `ExternalKernel` returns None. https://github.com/pytorch/pytorch/blob/921aa194c77f5279b15415eaa213813ddcdb3b29/torch/_inductor/ir.py#L5632-L5649 - If a `ExternalKernel schedulerNode` has None device, the previous buffer will not flush before codegen this `ExternalKernel schedulerNode` which causes the wrong generated code. https://github.com/pytorch/pytorch/blob/ef2b5ed500cba0b8b2bf04e6006a0d64c910f440/torch/_inductor/scheduler.py#L2701-L2709 cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
eellison
reviewed
Jun 13, 2024
| @@ -198,6 +198,38 @@ def f(x): | |||
| res = torch.compile(f, backend="inductor")(*inputs) | |||
| self.assertTrue(torch.allclose(res, f(*inputs))) | |||
|
|
|||
| @unittest.skipIf(IS_WINDOWS, "triton") | |||
| @unittest.skipIf(TEST_WITH_ROCM, "triton") | |||
| @unittest.skipIf(_get_torch_cuda_version() >= (11, 7), "triton") | |||
There was a problem hiding this comment.
This doesn't test cuda. why are skipping here ? also, you can test directly torch.utils._triton import has_triton
There was a problem hiding this comment.
I think I need to skip this test on windows, the preCI reports error in X win-vs2019-cpu-py3 / test (default, 1, 3, windows.4xlarge.nonephemeral) as
2024-06-14T03:11:36.4507249Z torch._dynamo.exc.BackendCompilerFailed: backend='inductor' raised:
2024-06-14T03:11:36.4508166Z InvalidCxxCompiler: No working C++ compiler found in torch._inductor.config.cpp.cxx: (None, 'g++')
There was a problem hiding this comment.
I think you just need to skip on windows
The 8000 re was a problem hiding this comment.
Thanks, skip this test on Windows.
| if example_output is None: | ||
| packed = cls( | ||
| NoneLayout(None), | ||
| NoneLayout(device), |
There was a problem hiding this comment.
Now that we're propagating device here, can we remove some of the if device is not None checks from the original pr:
493478d#diff-d2539c9c8dc6a3d7e457767a880612e96d3c85752a77ead49a9e4e00a3e4c3c7R2443
There was a problem hiding this comment.
From the implementation of FallbackKernel.find_device,
Lines 5455 to 5471 in 685fcfb
None.
eellison
approved these changes
Jun 13, 2024
Fix the issue: #127995 - In current implementation of creating `FallbackKernel`, the `device` of the `NoneLayout` is set to `None` when `example_output` returns from `cls.process_kernel` is `None`. https://github.com/pytorch/pytorch/blob/921aa194c77f5279b15415eaa213813ddcdb3b29/torch/_inductor/ir.py#L5632-L5649 - If a `ExternalKernel schedulerNode` has None device, the previous buffer will not flush before codegen this `ExternalKernel schedulerNode` which causes the wrong generated code. https://github.com/pytorch/pytorch/blob/ef2b5ed500cba0b8b2bf04e6006a0d64c910f440/torch/_inductor/scheduler.py#L2701-L2709 **Test Plan** ``` python -u -m pytest -s -v test/higher_order_ops/test_with_effects.py -k test_compile_inductor_external_op_return_none ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Fix the issue: #127995 - In current implementation of creating `FallbackKernel`, the `device` of the `NoneLayout` is set to `None` when `example_output` returns from `cls.process_kernel` is `None`. https://github.com/pytorch/pytorch/blob/921aa194c77f5279b15415eaa213813ddcdb3b29/torch/_inductor/ir.py#L5632-L5649 - If a `ExternalKernel schedulerNode` has None device, the previous buffer will not flush before codegen this `ExternalKernel schedulerNode` which causes the wrong generated code. https://github.com/pytorch/pytorch/blob/ef2b5ed500cba0b8b2bf04e6006a0d64c910f440/torch/_inductor/scheduler.py#L2701-L2709 **Test Plan** ``` python -u -m pytest -s -v test/higher_order_ops/test_with_effects.py -k test_compile_inductor_external_op_return_none ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Fix the issue: #127995 - In current implementation of creating `FallbackKernel`, the `device` of the `NoneLayout` is set to `None` when `example_output` returns from `cls.process_kernel` is `None`. https://github.com/pytorch/pytorch/blob/921aa194c77f5279b15415eaa213813ddcdb3b29/torch/_inductor/ir.py#L5632-L5649 - If a `ExternalKernel schedulerNode` has None device, the previous buffer will not flush before codegen this `ExternalKernel schedulerNode` which causes the wrong generated code. https://github.com/pytorch/pytorch/blob/ef2b5ed500cba0b8b2bf04e6006a0d64c910f440/torch/_inductor/scheduler.py#L2701-L2709 **Test Plan** ``` python -u -m pytest -s -v test/higher_order_ops/test_with_effects.py -k test_compile_inductor_external_op_return_none ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Fix the issue: #127995 - In current implementation of creating `FallbackKernel`, the `device` of the `NoneLayout` is set to `None` when `example_output` returns from `cls.process_kernel` is `None`. https://github.com/pytorch/pytorch/blob/921aa194c77f5279b15415eaa213813ddcdb3b29/torch/_inductor/ir.py#L5632-L5649 - If a `ExternalKernel schedulerNode` has None device, the previous buffer will not flush before codegen this `ExternalKernel schedulerNode` which causes the wrong generated code. https://github.com/pytorch/pytorch/blob/ef2b5ed500cba0b8b2bf04e6006a0d64c910f440/torch/_inductor/scheduler.py#L2701-L2709 **Test Plan** ``` python -u -m pytest -s -v test/higher_order_ops/test_with_effects.py -k test_compile_inductor_external_op_return_none ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
Fix the issue: #127995 - In current implementation of creating `FallbackKernel`, the `device` of the `NoneLayout` is set to `None` when `example_output` returns from `cls.process_kernel` is `None`. https://github.com/pytorch/pytorch/blob/921aa194c77f5279b15415eaa213813ddcdb3b29/torch/_inductor/ir.py#L5632-L5649 - If a `ExternalKernel schedulerNode` has None device, the previous buffer will not flush before codegen this `ExternalKernel schedulerNode` which causes the wrong generated code. https://github.com/pytorch/pytorch/blob/ef2b5ed500cba0b8b2bf04e6006a0d64c910f440/torch/_inductor/scheduler.py#L2701-L2709 **Test Plan** ``` python -u -m pytest -s -v test/higher_order_ops/test_with_effects.py -k test_compile_inductor_external_op_return_none ``` cc voznesenskym penguinwu EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng wenzhe-nrv jiayisunx peterbell10 ipiszy yf225 chenyang78 kadeng muchulee8 ColinPeppler amjames desertfire chauhang [ghstack-poisoned]
|
@leslie-fang-intel Yes! It is fine to land this PR! |
|
@pytorchbot merge |
Merge startedYour 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 |
|
@leslie-fang-intel thank you for the fix. I'll submit this for cherry-pick onto the 2.4 branch |
zou3519
pushed a commit
that referenced
this pull request
Jun 17, 2024
Fix the issue: #127995 - In current implementation of creating `FallbackKernel`, the `device` of the `NoneLayout` is set to `None` when `example_output` returns from `cls.process_kernel` is `None`. https://github.com/pytorch/pytorch/blob/921aa194c77f5279b15415eaa213813ddcdb3b29/torch/_inductor/ir.py#L5632-L5649 - If a `ExternalKernel schedulerNode` has None device, the previous buffer will not flush before codegen this `ExternalKernel schedulerNode` which causes the wrong generated code. https://github.com/pytorch/pytorch/blob/ef2b5ed500cba0b8b2bf04e6006a0d64c910f440/torch/_inductor/scheduler.py#L2701-L2709 **Test Plan** ``` python -u -m pytest -s -v test/higher_order_ops/test_with_effects.py -k test_compile_inductor_external_op_return_none ``` Pull Request resolved: #128275 Approved by: https://github.com/eellison
This was referenced Jun 17, 2024
atalman
pushed a commit
that referenced
this pull request
Jun 19, 2024
Fix the issue: #127995 - In current implementation of creating `FallbackKernel`, the `device` of the `NoneLayout` is set to `None` when `example_output` returns from `cls.process_kernel` is `None`. https://github.com/pytorch/pytorch/blob/921aa194c77f5279b15415eaa213813ddcdb3b29/torch/_inductor/ir.py#L5632-L5649 - If a `ExternalKernel schedulerNode` has None device, the previous buffer will not flush before codegen this `ExternalKernel schedulerNode` which causes the wrong generated code. https://github.com/pytorch/pytorch/blob/ef2b5ed500cba0b8b2bf04e6006a0d64c910f440/torch/_inductor/scheduler.py#L2701-L2709 **Test Plan** ``` python -u -m pytest -s -v test/higher_order_ops/test_with_effects.py -k test_compile_inductor_external_op_return_none ``` Pull Request resolved: #128275 Approved by: https://github.com/eellison Co-authored-by: leslie-fang-intel <leslie.fang@intel.com>
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.
Stack from ghstack (oldest at bottom):
Fix the issue: #127995
FallbackKernel, thedeviceof theNoneLayoutis set toNonewhenexample_outputreturns fromcls.process_kernelisNone.pytorch/torch/_inductor/ir.py
Lines 5632 to 5649 in 921aa19
ExternalKernel schedulerNodehas None device, the previous buffer will not flush before codegen thisExternalKernel schedulerNodewhich causes the wrong generated code.pytorch/torch/_inductor/scheduler.py
Lines 2701 to 2709 in ef2b5ed
Test Plan
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang