fix _unsafe_view composite compliance#77127
Closed
bdhirsh wants to merge 3 commits intogh/bdhirsh/225/basefrom
Closed
fix _unsafe_view composite compliance#77127bdhirsh wants to merge 3 commits intogh/bdhirsh/225/basefrom
bdhirsh wants to merge 3 commits intogh/bdhirsh/225/basefrom
Conversation
[ghstack-poisoned]
Contributor
🔗 Helpful links
❌ 5 New Failures, 5 Flaky FailuresAs of commit b4be12a (more details on the Dr. CI page): Expand to see more
🕵️ 5 new failures recognized by patternsThe following CI failures do not appear to be due to upstream breakages
|
This was referenced May 10, 2022
albanD
reviewed
May 10, 2022
| "not compatible with input tensor's size and stride (at least one dimension" | ||
| " spans across two contiguous subspaces). Use .reshape(...) instead."); | ||
| return alias_with_sizes_and_strides(self, inferred_size, *stride); | ||
| return self.as_strided(inferred_size, *stride); |
Collaborator
There was a problem hiding this comment.
The whole point of unsafe_view is to be faster than doing a proper view that goes through the dispatcher.
So if we do this, we might as well remove unsafe_view altogether.
Collaborator
Author
There was a problem hiding this comment.
oh haha. Then maybe the CompositeExplicitAutograd label is just false advertising.
Ok, I'm actually going to just remove this PR. I guess we could replace CompositeExplicitAutograd with CPU, CUDA, QuantizedCPU, QuantizedCUDA, etc, but we risk breaking 3rd party backends that happened to be relying on it.
I have another PR later that will add a new alias key for basically "all backends except functional backends", which might be a good candidate for this kernel (since functional backends like XLA/LTC won't like the alias_with_sizes_and_strides call).
Contributor
|
I'm wondering what the right thing to do here is. @albanD isn't the goal of |
Collaborator
|
Ho I forgot we added full autograd support for it. It is already going through the dispatcher once I gues (unless it is called via at::native). On a side note, isn't composite compliance only for the Implicit autograd keys? |
Contributor
Yes? Wasn't this the whole point of inference mode |
Collaborator
Author
I'm probably overloading the term, yeah 😛. "Ops advertising as |
`_unsafe_view` is listed as `CompositeExplicitAutograd`, but internally is calls `alias_with_sizes_and_strides` which isn't supported by non-eager backends. [ghstack-poisoned]
`_unsafe_view` is listed as `CompositeExplicitAutograd`, but internally is calls `alias_with_sizes_and_strides` which isn't supported by non-eager backends. [ghstack-poisoned]
Contributor
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
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
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.
_unsafe_viewis listed asCompositeExplicitAutograd, but internally is callsalias_with_sizes_and_strideswhich isn't supported by non-eager backends.Stack from ghstack: