Add C++ function for torch.distributed.tensor._op_schema.is_view_op#161595
Closed
swolchok wants to merge 6 commits intogh/swolchok/814/basefrom
Closed
Add C++ function for torch.distributed.tensor._op_schema.is_view_op#161595swolchok wants to merge 6 commits intogh/swolchok/814/basefrom
swolchok wants to merge 6 commits intogh/swolchok/814/basefrom
Conversation
This seems to have been an especially slow one because of the repeated pybind access (schema is a pybind, as is arguments, and then we hit each argument). It's still ~~1% of total benchmark runtime because of the repeated single pybind function call, but that's a lot better. [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/161595
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (2 Unrelated Failures)As of commit 90190d0 with merge base dcf3853 ( BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Aug 27, 2025
markc-614
pushed a commit
to markc-614/pytorch
that referenced
this pull request
Sep 17, 2025
…orch#162218) It returns a const reference to a vector. Pull Request resolved: pytorch#162218 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220
markc-614
pushed a commit
to markc-614/pytorch
that referenced
this pull request
Sep 17, 2025
These seem to have been costing us 5-10 usec per detach (out of ~~95 usec total). If they need to ship let's talk about requirements and how we can make this more efficient given that we would prefer if an entire DTensor op could finish in 10 usec. Differential Revision: [D81530106](https://our.internmc.facebook.com/intern/diff/D81530106) Pull Request resolved: pytorch#161596 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220, pytorch#162218
markc-614
pushed a commit
to markc-614/pytorch
that referenced
this pull request
Sep 17, 2025
…162337) We control DTensor, so we can just guarantee there isn't a programming error with __torch_dispatch__. (The guard is already less-than-perfect; see the note that the deleted comment refers to.) Pull Request resolved: pytorch#162337 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220, pytorch#162218, pytorch#161596
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
…ytorch#161595) This seems to have been an especially slow one because of the repeated pybind access (schema is a pybind, as is arguments, and then we hit each argument). It's still ~~1% of total benchmark runtime because of the repeated single pybind function call, but that's a lot better. Differential Revision: [D81530095](https://our.internmc.facebook.com/intern/diff/D81530095) Pull Request resolved: pytorch#161595 Approved by: https://github.com/ezyang, https://github.com/bdhirsh ghstack dependencies: pytorch#161466, pytorch#161586, pytorch#161590, pytorch#161591
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
…ytorch#161633) py::args::size() calls PyTuple_GetSize. Compiler can't know the two calls will always return the same result, so we have to consolidate them ourselves. Differential Revision: [D81530096](https://our.internmc.facebook.com/intern/diff/D81530096) Pull Request resolved: pytorch#161633 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
It is cheap to do an exact check against Tensor and much faster when it works (PyType_IsSubtype does not have this fastpath, I checked [source](https://github.com/python/cpython/blob/9ee0214b5dd982ac9fbe18dcce0e8787456e29af/Objects/typeobject.c#L2889)). Spot-checked in perf on detach-DTensor-in-a-loop benchmark; small win but clear. Differential Revision: [D81530101](https://our.internmc.facebook.com/intern/diff/D81530101) Pull Request resolved: pytorch#161634 Approved by: https://github.com/Skylion007, https://github.com/albanD ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
Calling this first minimizes overhead for plain old ints, making cheap things cheap. Differential Revision: [D81530098](https://our.internmc.facebook.com/intern/diff/D81530098) Pull Request resolved: pytorch#161692 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
…rayRef (pytorch#162219) Avoids requiring vector allocation to call this. Pull Request resolved: pytorch#162219 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
Optimize for common case and remove a pair of refcount operations (see new comments.) Pull Request resolved: pytorch#162220 Approved by: https://github.com/jansel, https://github.com/williamwen42 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
…orch#162218) It returns a const reference to a vector. Pull Request resolved: pytorch#162218 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
These seem to have been costing us 5-10 usec per detach (out of ~~95 usec total). If they need to ship let's talk about requirements and how we can make this more efficient given that we would prefer if an entire DTensor op could finish in 10 usec. Differential Revision: [D81530106](https://our.internmc.facebook.com/intern/diff/D81530106) Pull Request resolved: pytorch#161596 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220, pytorch#162218
mansiag05
pushed a commit
to mansiag05/pytorch
that referenced
this pull request
Sep 22, 2025
…162337) We control DTensor, so we can just guarantee there isn't a programming error with __torch_dispatch__. (The guard is already less-than-perfect; see the note that the deleted comment refers to.) Pull Request resolved: pytorch#162337 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220, pytorch#162218, pytorch#161596
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
…ytorch#161595) This seems to have been an especially slow one because of the repeated pybind access (schema is a pybind, as is arguments, and then we hit each argument). It's still ~~1% of total benchmark runtime because of the repeated single pybind function call, but that's a lot better. Differential Revision: [D81530095](https://our.internmc.facebook.com/intern/diff/D81530095) Pull Request resolved: pytorch#161595 Approved by: https://github.com/ezyang, https://github.com/bdhirsh ghstack dependencies: pytorch#161466, pytorch#161586, pytorch#161590, pytorch#161591
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
…ytorch#161633) py::args::size() calls PyTuple_GetSize. Compiler can't know the two calls will always return the same result, so we have to consolidate them ourselves. Differential Revision: [D81530096](https://our.internmc.facebook.com/intern/diff/D81530096) Pull Request resolved: pytorch#161633 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
It is cheap to do an exact check against Tensor and much faster when it works (PyType_IsSubtype does not have this fastpath, I checked [source](https://github.com/python/cpython/blob/9ee0214b5dd982ac9fbe18dcce0e8787456e29af/Objects/typeobject.c#L2889)). Spot-checked in perf on detach-DTensor-in-a-loop benchmark; small win but clear. Differential Revision: [D81530101](https://our.internmc.facebook.com/intern/diff/D81530101) Pull Request resolved: pytorch#161634 Approved by: https://github.com/Skylion007, https://github.com/albanD ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
Calling this first minimizes overhead for plain old ints, making cheap things cheap. Differential Revision: [D81530098](https://our.internmc.facebook.com/intern/diff/D81530098) Pull Request resolved: pytorch#161692 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
…rayRef (pytorch#162219) Avoids requiring vector allocation to call this. Pull Request resolved: pytorch#162219 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
Optimize for common case and remove a pair of refcount operations (see new comments.) Pull Request resolved: pytorch#162220 Approved by: https://github.com/jansel, https://github.com/williamwen42 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
…orch#162218) It returns a const reference to a vector. Pull Request resolved: pytorch#162218 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
These seem to have been costing us 5-10 usec per detach (out of ~~95 usec total). If they need to ship let's talk about requirements and how we can make this more efficient given that we would prefer if an entire DTensor op could finish in 10 usec. Differential Revision: [D81530106](https://our.internmc.facebook.com/intern/diff/D81530106) Pull Request resolved: pytorch#161596 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220, pytorch#162218
cleonard530
pushed a commit
to cleonard530/pytorch
that referenced
this pull request
Sep 22, 2025
…162337) We control DTensor, so we can just guarantee there isn't a programming error with __torch_dispatch__. (The guard is already less-than-perfect; see the note that the deleted comment refers to.) Pull Request resolved: pytorch#162337 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220, pytorch#162218, pytorch#161596
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
…ytorch#161595) This seems to have been an especially slow one because of the repeated pybind access (schema is a pybind, as is arguments, and then we hit each argument). It's still ~~1% of total benchmark runtime because of the repeated single pybind function call, but that's a lot better. Differential Revision: [D81530095](https://our.internmc.facebook.com/intern/diff/D81530095) Pull Request resolved: pytorch#161595 Approved by: https://github.com/ezyang, https://github.com/bdhirsh ghstack dependencies: pytorch#161466, pytorch#161586, pytorch#161590, pytorch#161591
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
…ytorch#161633) py::args::size() calls PyTuple_GetSize. Compiler can't know the two calls will always return the same result, so we have to consolidate them ourselves. Differential Revision: [D81530096](https://our.internmc.facebook.com/intern/diff/D81530096) Pull Request resolved: pytorch#161633 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
It is cheap to do an exact check against Tensor and much faster when it works (PyType_IsSubtype does not have this fastpath, I checked [source](https://github.com/python/cpython/blob/9ee0214b5dd982ac9fbe18dcce0e8787456e29af/Objects/typeobject.c#L2889)). Spot-checked in perf on detach-DTensor-in-a-loop benchmark; small win but clear. Differential Revision: [D81530101](https://our.internmc.facebook.com/intern/diff/D81530101) Pull Request resolved: pytorch#161634 Approved by: https://github.com/Skylion007, https://github.com/albanD ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
Calling this first minimizes overhead for plain old ints, making cheap things cheap. Differential Revision: [D81530098](https://our.internmc.facebook.com/intern/diff/D81530098) Pull Request resolved: pytorch#161692 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
…rayRef (pytorch#162219) Avoids requiring vector allocation to call this. Pull Request resolved: pytorch#162219 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
Optimize for common case and remove a pair of refcount operations (see new comments.) Pull Request resolved: pytorch#162220 Approved by: https://github.com/jansel, https://github.com/williamwen42 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
…orch#162218) It returns a const reference to a vector. Pull Request resolved: pytorch#162218 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
These seem to have been costing us 5-10 usec per detach (out of ~~95 usec total). If they need to ship let's talk about requirements and how we can make this more efficient given that we would prefer if an entire DTensor op could finish in 10 usec. Differential Revision: [D81530106](https://our.internmc.facebook.com/intern/diff/D81530106) Pull Request resolved: pytorch#161596 Approved by: https://github.com/ezyang, https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220, pytorch#162218
dsashidh
pushed a commit
to dsashidh/pytorch
that referenced
this pull request
Sep 26, 2025
…162337) We control DTensor, so we can just guarantee there isn't a programming error with __torch_dispatch__. (The guard is already less-than-perfect; see the note that the deleted comment refers to.) Pull Request resolved: pytorch#162337 Approved by: https://github.com/Skylion007 ghstack dependencies: pytorch#161591, pytorch#161595, pytorch#161633, pytorch#161634, pytorch#161692, pytorch#162219, pytorch#162220, pytorch#162218, pytorch#161596
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):
This seems to have been an especially slow one because of the repeated pybind access (schema is a pybind, as is arguments, and then we hit each argument). It's still ~~1% of total benchmark runtime because of the repeated single pybind function call, but that's a lot better.
Differential Revision: D81530095
cc @H-Huang @awgu @wanchaol @fegin @fduwjj @wz337 @wconstab @d4l3k @pragupta @ezyang @msaroufim @EikanWang @jgong5 @wenzhe-nrv @sanchitintel