[SymmMem] Add MemPool support to CUDA backend#169740
Closed
kwen2501 wants to merge 5 commits intogh/kwen2501/290/basefrom
Closed
[SymmMem] Add MemPool support to CUDA backend#169740kwen2501 wants to merge 5 commits intogh/kwen2501/290/basefrom
kwen2501 wants to merge 5 commits intogh/kwen2501/290/basefrom
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/169740
Note: Links to docs will display an error until the docs builds have been completed. ❌ 3 New Failures, 3 PendingAs of commit 1082ad3 with merge base 04ae0e1 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
ciflow/h100-symm-mem
release notes: distributed (c10d)
kwen2501
added
release notes: distributed (symm_mem)
Skylion007
reviewed
Dec 8, 2025
ngimel
approved these changes
Dec 10, 2025
| } | ||
|
|
||
| /* Search for a block that covers the given ptr, and write back the offset to | ||
| * the base ptr; error out if not found */ |
Collaborator
There was a problem hiding this comment.
comment seems wrong, returns nullptr if not found, not error
Collaborator
Author
|
@pytorchbot merge |
Collaborator
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 |
Collaborator
|
Starting merge as part of PR stack under #170008 |
Collaborator
Merge failedReason: 1 mandatory check(s) failed. The first few are: Dig deeper by viewing the failures on hud |
Collaborator
Author
|
@pytorchbot merge -f "RoCM failure comes from runner: Available diskspace is less than 30 percent. Not enough diskspace" |
Collaborator
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
pytorchmergebot
pushed a commit
that referenced
this pull request
Dec 10, 2025
We'd like to default some flags for SymmetricMemory pools, e.g. - use_on_oom=False - no_split=True to improve UX and to fortify safety. We thus provide a wrapper with the above flags preset: ``` pool = torch.distributed._symmetric_memory.get_mem_pool(device) ``` Since these flags are internal to the wrapper, we also maintain the flexibility to vary it in the future. Pull Request resolved: #170008 Approved by: https://github.com/ngimel ghstack dependencies: #169739, #169740
skpark-rh
pushed a commit
to skpark-rh/pytorch
that referenced
this pull request
Dec 10, 2025
[1/N] Extended rendezvous matching condition from exact address match to case where tensor falls in allocation range. [2/N] Shifted all heavy stuff (involving cudaMalloc) from `cudaSymmetricMemory` to `cudaPeerAllocInfo`. The former now corresponds to a tensor, while the letter corresponds to an allocation. Tensors on the same allocation share the same `cudaPeerAllocInfo`. [3/N] Added tests. Pull Request resolved: pytorch#169740 Approved by: https://github.com/ngimel ghstack dependencies: pytorch#169739
skpark-rh
pushed a commit
to skpark-rh/pytorch
that referenced
this pull request
Dec 10, 2025
We'd like to default some flags for SymmetricMemory pools, e.g. - use_on_oom=False - no_split=True to improve UX and to fortify safety. We thus provide a wrapper with the above flags preset: ``` pool = torch.distributed._symmetric_memory.get_mem_pool(device) ``` Since these flags are internal to the wrapper, we also maintain the flexibility to vary it in the future. Pull Request resolved: pytorch#170008 Approved by: https://github.com/ngimel ghstack dependencies: pytorch#169739, pytorch#169740
daisyden
pushed a commit
to daisyden/pytorch
that referenced
this pull request
Dec 11, 2025
We'd like to default some flags for SymmetricMemory pools, e.g. - use_on_oom=False - no_split=True to improve UX and to fortify safety. We thus provide a wrapper with the above flags preset: ``` pool = torch.distributed._symmetric_memory.get_mem_pool(device) ``` Since these flags are internal to the wrapper, we also maintain the flexibility to vary it in the future. Pull Request resolved: pytorch#170008 Approved by: https://github.com/ngimel ghstack dependencies: pytorch#169739, pytorch#169740
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
5 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):
[1/N] Extended rendezvous matching condition from exact address match to case where tensor falls in allocation range.
[2/N] Shifted all heavy stuff (involving cudaMalloc) from
cudaSymmetricMemorytocudaPeerAllocInfo. The former now corresponds to a tensor, while the letter corresponds to an allocation. Tensors on the same allocation share the samecudaPeerAllocInfo.[3/N] Added tests.