You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2022/09/16 00:06:52 UTC

[GitHub] [tvm] tkonolige opened a new pull request, #12806: [Fix,MetaSchedule] Correct log usage in arithmetic intensity feature

tkonolige opened a new pull request, #12806:
URL: https://github.com/apache/tvm/pull/12806

   In meta schedule's featurization, arithmetic intensity was incorrectly calculated as log(FLOPs) / log(bytes). This change corrects it to log(FLOPs/bytes). Note that this is the same issue as in #12079.
   
   @vinx13 @junrushao 


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


[GitHub] [tvm] junrushao merged pull request #12806: [Fix,MetaSchedule] Correct log usage in arithmetic intensity feature

Posted by GitBox <gi...@apache.org>.
junrushao merged PR #12806:
URL: https://github.com/apache/tvm/pull/12806


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


[GitHub] [tvm] junrushao commented on pull request #12806: [Fix,MetaSchedule] Correct log usage in arithmetic intensity feature

Posted by GitBox <gi...@apache.org>.
junrushao commented on PR #12806:
URL: https://github.com/apache/tvm/pull/12806#issuecomment-1270857743

   I'm not sure I fully catch that, but If so, I'm happy to let it in. Tweaking this or not doesn't really matter after all in terms of performance.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


[GitHub] [tvm] junrushao commented on pull request #12806: [Fix,MetaSchedule] Correct log usage in arithmetic intensity feature

Posted by GitBox <gi...@apache.org>.
junrushao commented on PR #12806:
URL: https://github.com/apache/tvm/pull/12806#issuecomment-1269131468

   If there is no impact to overall performance at all, as we have some initial data, I'm happy to merge it in mainly because you prefer, i.e. more textualism in arithmetic intensity definition, meanwhile, Lianmin and I prefer numeric stability, both of which are personal preference, but given your very strong personal preference, I don't want to block it anyways.
   
   > the additional data showed relatively high natural variability in latencies, and no significant evidence of a performance regression
   
   Just to clarify, "relatively high natural variability" means after the PR, the performance varies more?


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org


[GitHub] [tvm] tkonolige commented on pull request #12806: [Fix,MetaSchedule] Correct log usage in arithmetic intensity feature

Posted by GitBox <gi...@apache.org>.
tkonolige commented on PR #12806:
URL: https://github.com/apache/tvm/pull/12806#issuecomment-1270405414

   > Just to clarify, "relatively high natural variability" means after the PR, the performance varies more?
   No. This is just saying that tuning has a lot of variance (regardless of with this PR or not).


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org