Improve TypeMeta #11502
Closed
Improve TypeMeta #11502
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
Differential Revision: D9623916 Differential Version: 56705898
Differential Revision: D9623916 Differential Version: 56858422
Differential Revision: D9644700 Differential Version: 56870160
Differential Revision: D9623916 Differential Version: 56900314
Differential Revision: D9644700 Differential Version: 56900331
Differential Revision: D9652088 Differential Version: 56917655
Differential Revision: D9652088 Differential Version: 56918713
Differential Revision: D9652089 Differential Version: 56918714
Differential Revision: D9623916 Differential Version: 56982469
Differential Revision: D9644700 Differential Version: 56982494
Differential Revision: D9652088 Differential Version: 56982465
Differential Revision: D9652089 Differential Version: 56982470
Differential Revision: D9663476 Differential Version: 56982467
Differential Revision: D9623916 Differential Version: 57101548
Differential Revision: D9644700 Differential Version: 57101652
Differential Revision: D9652088 Differential Version: 57101545
Differential Revision: D9652089 Differential Version: 57101550
Differential Revision: D9663476 Differential Version: 57101556
Differential Revision: D9623916 Differential Version: 57135223
Differential Revision: D9644700 Differential Version: 57135224
Differential Revision: D9652088 Differential Version: 57135218
Differential Revision: D9652089 Differential Version: 57135219
Differential Revision: D9663476 Differential Version: 57135226
Differential Revision: D9694327 Differential Version: 57135227
Differential Revision: D9694326 Differential Version: 57135216
Differential Revision: D9694918 Differential Version: 57140103
Differential Revision: D9623916 Differential Version: 57214033
Differential Revision: D9644700 Differential Version: 57214023
Differential Revision: D9652088 Differential Version: 57214005
Differential Revision: D9652089 Differential Version: 57214030
Contributor
Author
|
@pytorchbot retest this please |
Contributor
Author
|
@pytorchbot re-test this please |
Differential Revision: D9763422 Differential Version: 59814332
Closed
Differential Revision: D9763422 Differential Version: 59857350
Differential Revision: D9763422 Differential Version: 59871561
Differential Revision: D9763422 Differential Version: 59872350
Differential Revision: D9763422 Differential Version: 59872623
Differential Revision: D9763422 Differential Version: 59874179
Differential Revision: D9763422 Differential Version: 59875467
Differential Revision: D9763422 Differential Version: 59876674
Differential Revision: D9763422 Differential Version: 59879239
Differential Revision: D9763422 Differential Version: 59882454
Differential Revision: D9763422 Differential Version: 59883487
Differential Revision: D9763422 Differential Version: 59886110
Differential Revision: D9763422 Differential Version: 59890579
Differential Revision: D9763422 Differential Version: 59898430
Differential Revision: D9763422 Differential Version: 59902014
Differential Revision: D9763422 Differential Version: 59910282
This was referenced Oct 6, 2018
zdevito
pushed a commit
to zdevito/ATen
that referenced
this pull request
Oct 6, 2018
Summary: Pull Request resolved: pytorch/pytorch#11502 TypeMeta now is only a pointer to a TypeMetaData structure, of which there is exactly one global instance per type. This reduces the size of everything storing a TypeMeta (Tensor, Blob, ...) and potentially improves performance. Also, this diff gets rid of the type name registry in favor of static strings. Experiments (summary: 1-3% perf gain) - Service Lab: https://our.intern.facebook.com/intern/servicelab/30712497/ -> No significant results found. - Mobile Lab c10bench.json: https://our.intern.facebook.com/intern/fblearner/details/75984908/ -> 1-3% perf gain - Mobile Lab c10bench default: https://our.intern.facebook.com/intern/fblearner/details/75984999/ -> 2-3% perf gain - adindexer canary: https://our.intern.facebook.com/intern/ads/canary/413002142824203076 -> no significant changes (benchmark too noisy) - adfinder canary: https://our.intern.facebook.com/intern/ads/canary/413002166737860362 -> no significant changes (benchmark too noisy) Reviewed By: dzhulgakov Differential Revision: D9763422 fbshipit-source-id: fc08937f114af5ff9f3ddbe7c7e396942868cdf5
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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:
:black_circle: #11502 Improve TypeMeta 💚
:white_circle: #12412 [noland] try if global variable on msvc works with CAFFE2_API 💛
:white_circle: #12423 [noland] try C10_EXPORT on gcc 💛
TypeMeta now is only a pointer to a TypeMetaData structure, of which there is exactly one global instance per type.
This reduces the size of everything storing a TypeMeta (Tensor, Blob, ...) and potentially improves performance.
Also, this diff gets rid of the type name registry in favor of static strings.
Experiments (summary: 1-3% perf gain)
-> No significant results found.
-> 1-3% perf gain
-> 2-3% perf gain
-> no significant changes (benchmark too noisy)
-> no significant changes (benchmark too noisy)
Differential Revision: D9763422