Replace empty_affine_quantizer with direct dispatch to at::native::empty_affine.. .#36814
Closed
kimishpatel wants to merge 9 commits intogh/kimishpatel/3/basefrom
Closed
Replace empty_affine_quantizer with direct dispatch to at::native::empty_affine.. .#36814kimishpatel wants to merge 9 commits intogh/kimishpatel/3/basefrom
kimishpatel wants to merge 9 commits intogh/kimishpatel/3/basefrom
Conversation
Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
This was referenced Apr 17, 2020
💊 Build failures summary and remediationsAs of commit 729b708 (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker. This comment has been revised 35 times. |
Contributor
|
Can you update the commit message to explain the motivation for this change? |
Contributor
Author
Updated. |
jerryzh168
reviewed
Apr 23, 2020
| const auto b_scale = qb_contig.q_scale(); | ||
|
|
||
| Tensor qy = at::_empty_affine_quantized( | ||
| Tensor qy = at::new_qtensor_cpu( |
Contributor
There was a problem hiding this comment.
could you use at::native:: empty_affine_quantized?
Contributor
Author
There was a problem hiding this comment.
Whey do you think that would be better? Does it get around the dispatch overhead?
Contributor
There was a problem hiding this comment.
It is better because it has the same API as at::_empty_affine_quantized.
yes, it will get around the dispatch overhead.
Contributor
Author
There was a problem hiding this comment.
Sure. Let me try.
Contributor
Author
There was a problem hiding this comment.
This is still slower, but not a whole lot. So I think it can make for a decent compromise.
From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
jerryzh168
reviewed
Apr 27, 2020
| #include <ATen/native/quantized/cpu/quantized_ops.h> | ||
| #include <ATen/native/quantized/cpu/init_qnnpack.h> | ||
| #include <ATen/native/quantized/cpu/qnnpack_utils.h> | ||
| #include <c10/core/TensorOptions.h> |
Contributor
There was a problem hiding this comment.
can these includes be removed?
jerryzh168
reviewed
Apr 27, 2020
Comment on lines
+10
to
+13
| #include <ATen/quantized/Quantizer.h> | ||
| #include <ATen/native/quantized/cpu/fbgemm_utils.h> | ||
| #include <ATen/native/quantized/cpu/qnnpack_utils.h> | ||
| #include <c10/core/TensorOptions.h> |
Contributor
There was a problem hiding this comment.
can these includes be removed?
Contributor
Author
There was a problem hiding this comment.
Yes, I was gonna do that.
From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
…:native::empty_affine.. ." From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
…:native::empty_affine.. ." From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
…:native::empty_affine.. ." From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable Differential Revision: [D21093840](https://our.internmc.facebook.com/intern/diff/D21093840/) [ghstack-poisoned]
Contributor
|
This pull request has been merged in 1510bdd. |
laurentdupin
pushed a commit
to laurentdupin/pytorch
that referenced
this pull request
Apr 24, 2026
Summary: Pull Request resolved: pytorch#36814 ghstack-source-id: 103218412 From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable {F234432545} Test Plan: Quantized op tests. Reviewed By: jerryzh168 Differential Revision: D21093840 fbshipit-source-id: 1b98b57eae403353596fc31171069d2f43b13385
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:
From the flamegraph it seems 40% the time we are spending going through the dispatch stack. I think in quantized model where compute can take less time, such overheads become noticeable
Differential Revision: D21093840