[FakeTensor] Reuse flat_args throughout FakeTensorMode.dispatch #112418
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
This function repeatedly flattens and unflattens the `args, kwargs` pair so we get a quite significant perf improvement from saving the `flat_args` and operating directly on those. I see a 15% improvement in dispatch for `empty_strided`. [ghstack-poisoned]
This was referenced Oct 30, 2023
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/112418
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 2cafa24 with merge base 29f3d39 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Oct 30, 2023
This function repeatedly flattens and unflattens the `args, kwargs` pair so we get a quite significant perf improvement from saving the `flat_args` and operating directly on those. I see a 15% improvement in dispatch for `empty_strided`. [ghstack-poisoned]
peterbell10
changed the title
lezcano
approved these changes
Nov 1, 2023
torch/_subclasses/fake_tensor.py
Outdated
|
|
||
| def wrap_meta_outputs_with_default_device_logic(self, r, func, args, kwargs): | ||
| wrap = self.gen_wrap_fn(func, args, kwargs) | ||
| def wrap_meta_outputs_with_default_device_logic(self, r, func, flat_args, kwargs): |
There was a problem hiding this comment.
nit. perhaps just pass `device rather than passing flat_args and kwargs. I found this a bit confusing.
torch/_subclasses/fake_tensor.py
Outdated
|
|
||
| return tree_map(map_out, r) | ||
| flat_out = [map_out(o) for o in flat_out] | ||
| return pytree.tree_unflatten(flat_out, out_spec) |
There was a problem hiding this comment.
Here perhaps we just want to tree_map all the transformation on r rather than unpacking and packing again?
…patch" This function repeatedly flattens and unflattens the `args, kwargs` pair so we get a quite significant perf improvement from saving the `flat_args` and operating directly on those. I see a 15% improvement in dispatch for `empty_strided`. [ghstack-poisoned]
…patch" This function repeatedly flattens and unflattens the `args, kwargs` pair so we get a quite significant perf improvement from saving the `flat_args` and operating directly on those. I see a 15% improvement in dispatch for `empty_strided`. [ghstack-poisoned]
pytorchmergebot
pushed a commit
that referenced
this pull request
Nov 1, 2023
`ShapeEnv` has tons of functionallity that is conditioned on this `translation_validation_enabled()` check, to the point where 8% of time in `empty_strided` is spent just in that function. However, it doesn't really make sense for the value of `translation_validation_enabled()` to change throughout the life of a `ShapeEnv` so we might as well run the check once and store it in the `ShapeEnv`. Pull Request resolved: #112493 Approved by: https://github.com/lezcano ghstack dependencies: #112418
xuhancn
pushed a commit
to xuhancn/pytorch
that referenced
this pull request
Nov 7, 2023
…rch#112418) This function repeatedly flattens and unflattens the `args, kwargs` pair so we get a quite significant perf improvement from saving the `flat_args` and operating directly on those. I see a 15% improvement in dispatch for `empty_strided`. Pull Request resolved: pytorch#112418 Approved by: https://github.com/lezcano
xuhancn
pushed a commit
to xuhancn/pytorch
that referenced
this pull request
Nov 7, 2023
`ShapeEnv` has tons of functionallity that is conditioned on this `translation_validation_enabled()` check, to the point where 8% of time in `empty_strided` is spent just in that function. However, it doesn't really make sense for the value of `translation_validation_enabled()` to change throughout the life of a `ShapeEnv` so we might as well run the check once and store it in the `ShapeEnv`. Pull Request resolved: pytorch#112493 Approved by: https://github.com/lezcano ghstack dependencies: pytorch#112418
Skylion007
pushed a commit
to Skylion007/pytorch
that referenced
this pull request
Nov 14, 2023
…rch#112418) This function repeatedly flattens and unflattens the `args, kwargs` pair so we get a quite significant perf improvement from saving the `flat_args` and operating directly on those. I see a 15% improvement in dispatch for `empty_strided`. Pull Request resolved: pytorch#112418 Approved by: https://github.com/lezcano
Skylion007
pushed a commit
to Skylion007/pytorch
that referenced
this pull request
Nov 14, 2023
`ShapeEnv` has tons of functionallity that is conditioned on this `translation_validation_enabled()` check, to the point where 8% of time in `empty_strided` is spent just in that function. However, it doesn't really make sense for the value of `translation_validation_enabled()` to change throughout the life of a `ShapeEnv` so we might as well run the check once and store it in the `ShapeEnv`. Pull Request resolved: pytorch#112493 Approved by: https://github.com/lezcano ghstack dependencies: pytorch#112418
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):
ShapeEnv#112493This function repeatedly flattens and unflattens the
args, kwargspair so weget a quite significant perf improvement from saving the
flat_argsandoperating directly on those. I see a 15% improvement in dispatch for
empty_strided.