Eager module attribution in profiler stack traces#48433
Merged
ilia-cher merged 4 commits intogh/ilia-cher/86/basefrom Dec 16, 2020
Merged
Eager module attribution in profiler stack traces#48433ilia-cher merged 4 commits intogh/ilia-cher/86/basefrom
ilia-cher merged 4 commits intogh/ilia-cher/86/basefrom
Conversation
Summary: Adding classnames into profiler stack traces (eager mode only atm) Test Plan: python test/test_profiler.py -k test_module_attrib_eager [ghstack-poisoned]
This was referenced Nov 25, 2020
Closed
Closed
💊 CI failures summary and remediationsAs of commit 045c0c2 (more details on the Dr. CI page):
🕵️ 6 new failures recognized by patternsThe following CI failures do not appear to be due to upstream breakages:
|
added 3 commits
November 24, 2020 20:04
Summary: Adding classnames into profiler stack traces (eager mode only atm) Test Plan: python test/test_profiler.py -k test_module_attrib_eager output: https://gist.github.com/ilia-cher/e988a43dc9a444ae8caa68f3e6b0a294 e.g.: ``` ---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------------------------------------- Name Self CPU % Self CPU CPU total % CPU total CPU time avg # of Calls Source Location ---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------------------------------------- aten::mkldnn_convolution 98.47% 30.425ms 99.07% 30.610ms 30.610ms 1 ...s/iliacher/pytorch/torch/nn/modules/conv.py(389): _conv_forward (Conv2d) ...a/users/iliacher/pytorch/torch/nn/modules/conv.py(393): forward (Conv2d) ...rs/iliacher/pytorch/torch/nn/modules/module.py(744): _call_impl (Conv2d) test/test_profiler.py(172): forward (DummyModule_1) ...cher/pytorch/torch/nn/modules/module.py(744): _call_impl (DummyModule_1) test/test_profiler.py(180): forward (DummyModule_2) ...cher/pytorch/torch/nn/modules/module.py(744): _call_impl (DummyModule_2) test/test_profiler.py(185): test_module_attrib_eager (TestProfiler) ...orch/lib/python3.8/unittest/case.py(633): _callTestMethod (TestProfiler) ...da3/envs/pytorch/lib/python3.8/unittest/case.py(676): run (TestProfiler) ``` [ghstack-poisoned]
Summary: Adding classnames into profiler stack traces (eager mode only atm) Test Plan: python test/test_profiler.py -k test_module_attrib_eager output: https://gist.github.com/ilia-cher/e988a43dc9a444ae8caa68f3e6b0a294 e.g.: ``` ---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------------------------------------- Name Self CPU % Self CPU CPU total % CPU total CPU time avg # of Calls Source Location ---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------------------------------------- aten::mkldnn_convolution 98.47% 30.425ms 99.07% 30.610ms 30.610ms 1 ...s/iliacher/pytorch/torch/nn/modules/conv.py(389): _conv_forward (Conv2d) ...a/users/iliacher/pytorch/torch/nn/modules/conv.py(393): forward (Conv2d) ...rs/iliacher/pytorch/torch/nn/modules/module.py(744): _call_impl (Conv2d) test/test_profiler.py(172): forward (DummyModule_1) ...cher/pytorch/torch/nn/modules/module.py(744): _call_impl (DummyModule_1) test/test_profiler.py(180): forward (DummyModule_2) ...cher/pytorch/torch/nn/modules/module.py(744): _call_impl (DummyModule_2) test/test_profiler.py(185): test_module_attrib_eager (TestProfiler) ...orch/lib/python3.8/unittest/case.py(633): _callTestMethod (TestProfiler) ...da3/envs/pytorch/lib/python3.8/unittest/case.py(676): run (TestProfiler) ``` Differential Revision: [D25174271](https://our.internmc.facebook.com/intern/diff/D25174271) [ghstack-poisoned]
Summary: Adding classnames into profiler stack traces (eager mode only atm) Test Plan: python test/test_profiler.py -k test_module_attrib_eager output: https://gist.github.com/ilia-cher/e988a43dc9a444ae8caa68f3e6b0a294 e.g.: ``` ---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------------------------------------- Name Self CPU % Self CPU CPU total % CPU total CPU time avg # of Calls Source Location ---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------------------------------------- aten::mkldnn_convolution 98.47% 30.425ms 99.07% 30.610ms 30.610ms 1 ...s/iliacher/pytorch/torch/nn/modules/conv.py(389): _conv_forward (Conv2d) ...a/users/iliacher/pytorch/torch/nn/modules/conv.py(393): forward (Conv2d) ...rs/iliacher/pytorch/torch/nn/modules/module.py(744): _call_impl (Conv2d) test/test_profiler.py(172): forward (DummyModule_1) ...cher/pytorch/torch/nn/modules/module.py(744): _call_impl (DummyModule_1) test/test_profiler.py(180): forward (DummyModule_2) ...cher/pytorch/torch/nn/modules/module.py(744): _call_impl (DummyModule_2) test/test_profiler.py(185): test_module_attrib_eager (TestProfiler) ...orch/lib/python3.8/unittest/case.py(633): _callTestMethod (TestProfiler) ...da3/envs/pytorch/lib/python3.8/unittest/case.py(676): run (TestProfiler) ``` Differential Revision: [D25174271](https://our.internmc.facebook.com/intern/diff/D25174271) [ghstack-poisoned]
Contributor
Author
|
will add usage of named_modules |
Contributor
Author
|
@jamesr66a @suo on support for TorchScript |
dzhulgakov
reviewed
Dec 7, 2020
Collaborator
dzhulgakov
left a comment
There was a problem hiding this comment.
Looks in the right direction. Creating a separate stack trace with module names (not just class names) would be the most helpful, but it can happen in the follow up PR
| header_sep_lst = [""] | ||
| line_length_lst = [-SPACING_SIZE] | ||
| MAX_STACK_ENTRY = 5 | ||
| MAX_STACK_ENTRY = 10 |
Collaborator
There was a problem hiding this comment.
should we make it overridable too?
| if (PyDict_Check(locals)) { | ||
| PyObject *key, *value; | ||
| ssize_t pos = 0; | ||
| while (PyDict_Next(locals, &pos, &key, &value)) { |
Collaborator
There was a problem hiding this comment.
there should be an api for looking up in the dict, instead of iteration
| auto source = std::make_shared<Source>(funcname, filename, line); | ||
| std::string classname = ""; | ||
|
|
||
| if (PyFrame_FastToLocalsWithError(frame) >= 0) { |
Collaborator
There was a problem hiding this comment.
you probably want to trigger it only when you encounter the frame that belongs to torch.nn.Module._call_impl. You can do it by looking up the code object for that method once and then just compare it with frame->f_code. That's how https://colab.research.google.com/drive/1NGRzvgzCZR6EEM3RVPvPAuMXfHsyC6eN?usp=sharing worked
Contributor
Author
There was a problem hiding this comment.
yeah, I went with all class names, not only Module specific; we could limit this to Module-inherited ones
Contributor
Author
|
this shouldn't be marked as 'merged', strange i only reordered commits and reran ghstack (cc. @ezyang) |
Contributor
Author
|
will resend as a new PR |
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
3 participants
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:
Summary:
Adding classnames into profiler stack traces (eager mode only atm)
Test Plan:
python test/test_profiler.py -k test_module_attrib_eager
output: https://gist.github.com/ilia-cher/e988a43dc9a444ae8caa68f3e6b0a294
e.g.:
Differential Revision: D25174271