Skip to content

Conversation

@Shixiaowei02
Copy link
Contributor

@Shixiaowei02 Shixiaowei02 commented Apr 19, 2021

PR types

Others

PR changes

Others

Describe

flush denormal in the tracer op, test=develop
Reference: #29924
fix PaddlePaddle/PaddleSeg#955

@paddle-bot-old
Copy link

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

@Shixiaowei02 Shixiaowei02 requested a review from luotao1 April 19, 2021 07:50
Copy link
Contributor

@luotao1 luotao1 left a comment

Choose a reason for hiding this comment

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

需要在paddle/fluid/imperative/CMakeLists.txt加依赖
cc_library(tracer SRCS tracer.cc DEPS layer engine program_desc_tracer amp denormal)

@Shixiaowei02
Copy link
Contributor Author

需要在paddle/fluid/imperative/CMakeLists.txt加依赖
cc_library(tracer SRCS tracer.cc DEPS layer engine program_desc_tracer amp denormal)

done


#if !defined(GCC_WITHOUT_INTRINSICS) && !defined(PADDLE_WITH_ARM) && \
!defined(PADDLE_WITH_SW) && !defined(PADDLE_WITH_MIPS)
#if !defined(_WIN32) || defined(PADDLE_WITH_MKL)
Copy link
Contributor

Choose a reason for hiding this comment

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

原因是在windows+openblas上,会出现某些单测精度diff的问题。由于windows平台上默认是发MKL的包,因此先跳过这个组合。

Copy link
Contributor Author

Choose a reason for hiding this comment

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

好的,已修改

@Shixiaowei02 Shixiaowei02 requested a review from luotao1 April 20, 2021 07:26
Copy link
Contributor

@luotao1 luotao1 left a comment

Choose a reason for hiding this comment

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

LGTM

@luotao1 luotao1 merged commit 9ff8556 into PaddlePaddle:develop Apr 21, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

iog模型相同的网络结构和输入图片,不同的网络参数,在跑前向计算时间差异大

2 participants