Improve detection of workspace/non-output allocations in cudagraphs #99985
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/99985
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit c7c6d51: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
…udagraphs" cc soumith voznesenskym penguinwu anijain2305 EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 desertfire [ghstack-poisoned]
…udagraphs" cc soumith voznesenskym penguinwu anijain2305 EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 desertfire [ghstack-poisoned]
…udagraphs" When we run cudagraph trees we are not allowed to have permanent workspace allocations like in cublas because we might need to reclaim that memory for a previous cudagraph recording, and it is memory that is not accounted for in output weakrefs so it does not work with checkpointing. Previously, I would check that we didn't have any additional allocations through snapshotting. This was extremely slow so I had to turn it off. This PR first does the quick checking to see if we are in an error state, then if we are does the slow logic of creating snapshot. Also turns on history recording so we get a stacktrace of where the bad allocation came from. cc soumith voznesenskym penguinwu anijain2305 EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 desertfire [ghstack-poisoned]
zdevito
approved these changes
Apr 26, 2023
torch/_inductor/cudagraph_trees.py
Outdated
| def format_tb(caching_allocator_trace): | ||
| formatted_traceback = [] | ||
|
|
||
| MAX_LENGHTH = 20 |
| @contextlib.contextmanager | ||
| def enable_history_recording(): | ||
| "Turns on history recording in the CUDA Caching Allocator" | ||
| enabled = torch._C._cuda_isHistoryEnabled() |
There was a problem hiding this comment.
There are caveats when simple recording is enabled but stack traces are not. This function take like 4 arguments, I wonder if when _record_memory_history is called, can we return the current settings as a dict. Maybe for now, we can just have _record_memory_history return whether it was enabled before. I guess I am trying to avoid having to add more virtual functions to the allocator interface than necessary.
There was a problem hiding this comment.
At least I have a default impl here, so you don't need to override on subclasses
…udagraphs" When we run cudagraph trees we are not allowed to have permanent workspace allocations like in cublas because we might need to reclaim that memory for a previous cudagraph recording, and it is memory that is not accounted for in output weakrefs so it does not work with checkpointing. Previously, I would check that we didn't have any additional allocations through snapshotting. This was extremely slow so I had to turn it off. This PR first does the quick checking to see if we are in an error state, then if we are does the slow logic of creating snapshot. Also turns on history recording so we get a stacktrace of where the bad allocation came from. cc soumith voznesenskym penguinwu anijain2305 EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 desertfire [ghstack-poisoned]
…udagraphs" When we run cudagraph trees we are not allowed to have permanent workspace allocations like in cublas because we might need to reclaim that memory for a previous cudagraph recording, and it is memory that is not accounted for in output weakrefs so it does not work with checkpointing. Previously, I would check that we didn't have any additional allocations through snapshotting. This was extremely slow so I had to turn it off. This PR first does the quick checking to see if we are in an error state, then if we are does the slow logic of creating snapshot. Also turns on history recording so we get a stacktrace of where the bad allocation came from. cc soumith voznesenskym penguinwu anijain2305 EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 desertfire [ghstack-poisoned]
…udagraphs" When we run cudagraph trees we are not allowed to have permanent workspace allocations like in cublas because we might need to reclaim that memory for a previous cudagraph recording, and it is memory that is not accounted for in output weakrefs so it does not work with checkpointing. Previously, I would check that we didn't have any additional allocations through snapshotting. This was extremely slow so I had to turn it off. This PR first does the quick checking to see if we are in an error state, then if we are does the slow logic of creating snapshot. Also turns on history recording so we get a stacktrace of where the bad allocation came from. cc soumith voznesenskym penguinwu anijain2305 EikanWang jgong5 Guobing-Chen XiaobingSuper zhuhaozhe blzheng Xia-Weiwen wenzhe-nrv jiayisunx peterbell10 desertfire [ghstack-poisoned]
|
@pytorchbot merge -f "unrelated error" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
valentinandrei
pushed a commit
to valentinandrei/pytorch
that referenced
this pull request
May 2, 2023
…ytorch#99985) When we run cudagraph trees we are not allowed to have permanent workspace allocations like in cublas because we might need to reclaim that memory for a previous cudagraph recording, and it is memory that is not accounted for in output weakrefs so it does not work with checkpointing. Previously, I would check that we didn't have any additional allocations through snapshotting. This was extremely slow so I had to turn it off. This PR first does the quick checking to see if we are in an error state, then if we are does the slow logic of creating snapshot. Also turns on history recording so we get a stacktrace of where the bad allocation came from. Pull Request resolved: pytorch#99985 Approved by: https://github.com/zdevito
xwang233
reviewed
Feb 29, 2024
| lambda: f"These live storage data ptrs are in the cudagraph pool but not " | ||
| f"accounted for as an output of cudagraph trees {allocated_not_in_live_storages}", | ||
| ) | ||
| if allocated_not_in_live_storages != 0: |
There was a problem hiding this comment.
Just curious, wouldn't comparing a dictionary with an int always return not equal, thus this turns into if True: effectively?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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 (oldest at bottom):
When we run cudagraph trees we are not allowed to have permanent workspace allocations like in cublas because we might need to reclaim that memory for a previous cudagraph recording, and it is memory that is not accounted for in output weakrefs so it does not work with checkpointing. Previously, I would check that we didn't have any additional allocations through snapshotting. This was extremely slow so I had to turn it off.
This PR first does the quick checking to see if we are in an error state, then if we are does the slow logic of creating snapshot. Also turns on history recording so we get a stacktrace of where the bad allocation came from.
cc @soumith @voznesenskym @penguinwu @anijain2305 @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @peterbell10 @desertfire