Fix copy kernel speed regression introduced in #29631 #31279
Closed
+16
−1
Conversation
This file contains 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
There was a problem hiding this comment.
@ngimel has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
CircleCI build failures summaryAs of commit f734bbe:
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. |
zasdfgbnm
added a commit
to zasdfgbnm/pytorch
that referenced
this pull request
Dec 16, 2019
…#31279) Summary: Fixes pytorch#31271 This fixes copy kernel speed regression introduced in pytorch#29631. The previous implementation forces the compiler to instantiate `static_cast_with_inter_type` because it is passed as an argument of a function. This behavior makes it impossible for compilers to do optimizations like automatic vectorization, and, function call itself is expensive compared to a single casting instruction. To check the change, run ``` readelf -Ws /home/xgao/anaconda3/lib/python3.7/site-packages/torch/lib/libtorch_cpu.so | grep static_cast_with_inter_type ``` On nightly build, we have output ``` 168217: 0000000001852bf0 5 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsdE5applyEd 168816: 0000000001852d30 33 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIfEaE5applyEa 168843: 00000000018531f0 7 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIblE5applyEl 168930: 0000000001852c20 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIslE5applyEl 168935: 00000000018528d0 124 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIfNS_4HalfEE5applyES1_ 169023: 0000000001852f30 17 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdEhE5applyEh 169713: 00000000018525c0 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIahE5applyEh 170033: 0000000001852c10 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsiE5applyEi 170105: 0000000001852bd0 5 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIshE5applyEh 170980: 0000000001852fc0 27 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdES1_IfEE5applyES3_ 171398: 0000000001852810 13 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIdbE5applyEb 171574: 00000000018532e0 35 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIbNS_8BFloat16EE5applyES1_ 171734: 0000000001852b20 6 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIlSt7complexIdEE5applyES2_ 172422: 0000000001853350 54 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_8BFloat16EaE5applyEa 172704: 00000000018533c0 38 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_8BFloat16EfE5applyEf 172976: 0000000001852890 10 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIflE5applyEl 173038: 0000000001852f80 9 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdEfE5applyEf 173329: 00000000018531c0 20 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIbfE5applyEf 173779: 00000000018524d0 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIhiE5applyEi 174032: 0000000001852960 14 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIfNS_8BFloat16EE5applyES1_ 174334: 0000000001852d60 29 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIfEdE5applyEd 174470: 0000000001852c60 124 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsNS_4HalfEE5applyES1_ 174770: 0000000001852bc0 15 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIlNS_8BFloat16EE5applyES1_ 176408: 0000000001853980 144 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_4HalfEbE5applyEb 176475: 0000000001852790 128 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIdNS_4HalfEE5applyES1_ .... ``` And after this PR, we get empty output ``` ``` Pull Request resolved: pytorch#31279 Differential Revision: D19075587 Pulled By: ngimel fbshipit-source-id: c20088241f39fa40c1d055f0a46eb5b9ece52e71
mingbowan
pushed a commit
that referenced
this pull request
Dec 20, 2019
Summary: Fixes #31271 This fixes copy kernel speed regression introduced in #29631. The previous implementation forces the compiler to instantiate `static_cast_with_inter_type` because it is passed as an argument of a function. This behavior makes it impossible for compilers to do optimizations like automatic vectorization, and, function call itself is expensive compared to a single casting instruction. To check the change, run ``` readelf -Ws /home/xgao/anaconda3/lib/python3.7/site-packages/torch/lib/libtorch_cpu.so | grep static_cast_with_inter_type ``` On nightly build, we have output ``` 168217: 0000000001852bf0 5 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsdE5applyEd 168816: 0000000001852d30 33 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIfEaE5applyEa 168843: 00000000018531f0 7 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIblE5applyEl 168930: 0000000001852c20 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIslE5applyEl 168935: 00000000018528d0 124 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIfNS_4HalfEE5applyES1_ 169023: 0000000001852f30 17 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdEhE5applyEh 169713: 00000000018525c0 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIahE5applyEh 170033: 0000000001852c10 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsiE5applyEi 170105: 0000000001852bd0 5 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIshE5applyEh 170980: 0000000001852fc0 27 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdES1_IfEE5applyES3_ 171398: 0000000001852810 13 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIdbE5applyEb 171574: 00000000018532e0 35 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIbNS_8BFloat16EE5applyES1_ 171734: 0000000001852b20 6 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIlSt7complexIdEE5applyES2_ 172422: 0000000001853350 54 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_8BFloat16EaE5applyEa 172704: 00000000018533c0 38 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_8BFloat16EfE5applyEf 172976: 0000000001852890 10 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIflE5applyEl 173038: 0000000001852f80 9 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdEfE5applyEf 173329: 00000000018531c0 20 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIbfE5applyEf 173779: 00000000018524d0 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIhiE5applyEi 174032: 0000000001852960 14 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIfNS_8BFloat16EE5applyES1_ 174334: 0000000001852d60 29 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIfEdE5applyEd 174470: 0000000001852c60 124 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsNS_4HalfEE5applyES1_ 174770: 0000000001852bc0 15 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIlNS_8BFloat16EE5applyES1_ 176408: 0000000001853980 144 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_4HalfEbE5applyEb 176475: 0000000001852790 128 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIdNS_4HalfEE5applyES1_ .... ``` And after this PR, we get empty output ``` ``` Pull Request resolved: #31279 Differential Revision: D19075587 Pulled By: ngimel fbshipit-source-id: c20088241f39fa40c1d055f0a46eb5b9ece52e71
wuhuikx
pushed a commit
to wuhuikx/pytorch
that referenced
this pull request
Jan 30, 2020
…#31279) Summary: Fixes pytorch#31271 This fixes copy kernel speed regression introduced in pytorch#29631. The previous implementation forces the compiler to instantiate `static_cast_with_inter_type` because it is passed as an argument of a function. This behavior makes it impossible for compilers to do optimizations like automatic vectorization, and, function call itself is expensive compared to a single casting instruction. To check the change, run ``` readelf -Ws /home/xgao/anaconda3/lib/python3.7/site-packages/torch/lib/libtorch_cpu.so | grep static_cast_with_inter_type ``` On nightly build, we have output ``` 168217: 0000000001852bf0 5 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsdE5applyEd 168816: 0000000001852d30 33 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIfEaE5applyEa 168843: 00000000018531f0 7 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIblE5applyEl 168930: 0000000001852c20 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIslE5applyEl 168935: 00000000018528d0 124 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIfNS_4HalfEE5applyES1_ 169023: 0000000001852f30 17 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdEhE5applyEh 169713: 00000000018525c0 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIahE5applyEh 170033: 0000000001852c10 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsiE5applyEi 170105: 0000000001852bd0 5 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIshE5applyEh 170980: 0000000001852fc0 27 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdES1_IfEE5applyES3_ 171398: 0000000001852810 13 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIdbE5applyEb 171574: 00000000018532e0 35 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIbNS_8BFloat16EE5applyES1_ 171734: 0000000001852b20 6 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIlSt7complexIdEE5applyES2_ 172422: 0000000001853350 54 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_8BFloat16EaE5applyEa 172704: 00000000018533c0 38 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_8BFloat16EfE5applyEf 172976: 0000000001852890 10 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIflE5applyEl 173038: 0000000001852f80 9 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIdEfE5applyEf 173329: 00000000018531c0 20 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIbfE5applyEf 173779: 00000000018524d0 3 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIhiE5applyEi 174032: 0000000001852960 14 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIfNS_8BFloat16EE5applyES1_ 174334: 0000000001852d60 29 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeISt7complexIfEdE5applyEd 174470: 0000000001852c60 124 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIsNS_4HalfEE5applyES1_ 174770: 0000000001852bc0 15 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIlNS_8BFloat16EE5applyES1_ 176408: 0000000001853980 144 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeINS_4HalfEbE5applyEb 176475: 0000000001852790 128 FUNC LOCAL DEFAULT 9 _ZN3c1027static_cast_with_inter_typeIdNS_4HalfEE5applyES1_ .... ``` And after this PR, we get empty output ``` ``` Pull Request resolved: pytorch#31279 Differential Revision: D19075587 Pulled By: ngimel fbshipit-source-id: c20088241f39fa40c1d055f0a46eb5b9ece52e71
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.
Fixes #31271
This fixes copy kernel speed regression introduced in #29631.
The previous implementation forces the compiler to instantiate
static_cast_with_inter_typebecause it is passed as an argument of a function. This behavior makes it impossible for compilers to do optimizations like automatic vectorization, and, function call itself is expensive compared to a single casting instruction.To check the change, run
On nightly build, we have output
And after this PR, we get empty output