Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Improve the performance of fake quantize OP #40491

Merged
merged 2 commits into from
Mar 17, 2022
Merged

Improve the performance of fake quantize OP #40491

merged 2 commits into from
Mar 17, 2022

Conversation

leo0519
Copy link
Collaborator

@leo0519 leo0519 commented Mar 13, 2022 •
edited
Loading

PR types

Performance optimization

PR changes

OPs

Describe

The modifications of fake quantize op are as follow.

  1. (moving average) Move the computation of moving average scale to device (80us -> 3.5us)
  2. (find max abs) Use register to save the local maximum in a thread

@paddle-bot-old
Copy link

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

@leo0519 leo0519 added the NVIDIA label Mar 13, 2022
@leo0519 leo0519 changed the title [Paddle-TRT] Improve the performance of fake quantize OP Improve the performance of fake quantize OP Mar 13, 2022
qingqing01
qingqing01 approved these changes Mar 14, 2022
View reviewed changes
Copy link
Contributor

@qingqing01 qingqing01 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

ceci3
ceci3 approved these changes Mar 14, 2022
View reviewed changes
Copy link
Contributor

@ceci3 ceci3 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@Wangzheee Wangzheee merged commit 827b6a0 into PaddlePaddle:develop Mar 17, 2022
liqitong-a pushed a commit to liqitong-a/Paddle that referenced this pull request Mar 17, 2022
@leo0519 @liqitong-a
Improve the performance of fake quantize OP (PaddlePaddle#40491)
925ebb4 
* Move the computation of moving average scale to device

* Use register to save local maximum in a thread
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@qingqing01 qingqing01 qingqing01 approved these changes

@ceci3 ceci3 ceci3 approved these changes

@wanghaoshuang wanghaoshuang Awaiting requested review from wanghaoshuang

@yghstill yghstill Awaiting requested review from yghstill

Assignees
No one assigned
Labels
NVIDIA
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@leo0519 @qingqing01 @ceci3 @Wangzheee