Skip to content

Optimize gelu operation with mkl erf#15931

Merged
luotao1 merged 1 commit intoPaddlePaddle:developfrom
yihuaxu:develop_2c5c7b2a7_gelu_mkl_opt
Feb 28, 2019
Merged

Optimize gelu operation with mkl erf#15931
luotao1 merged 1 commit intoPaddlePaddle:developfrom
yihuaxu:develop_2c5c7b2a7_gelu_mkl_opt

Conversation

@yihuaxu
Copy link
Contributor

@yihuaxu yihuaxu commented Feb 26, 2019

According to the performance status of Bert model, optimized GELU operator to accelerate the data processing.

Platform: Intel(R) Xeon(R) CPU E5-2650 v4 @ 2.20GHz
Model Path: third_party/inference_demo/bert_emb128/model
Batch Size: 1
Command: ./paddle/fluid/inference/tests/api/test_analyzer_bert --infer_model=third_party/inference_demo/bert_emb128/model/ --infer_data=third_party/inference_demo/bert_emb128/data.txt --gtest_filter=Analyzer_bert.profile --paddle_num_threads=1 --repeat=1 --batch_size=1 --test_all_data --profile
Data Source: third_party/inference_demo/bert_emb128/data.txt.

Associated with: PR#15770
PR#15871
Fix the memcpy@GLIBC_2.14 link issue of mklml library.

The following is the comparison with the different scenarios.

image

@luotao1 luotao1 added the Intel label Feb 26, 2019
@yihuaxu
Copy link
Contributor Author

yihuaxu commented Feb 27, 2019

start a review

@yihuaxu
Copy link
Contributor Author

yihuaxu commented Feb 27, 2019

@panyx0718 Please help us review this PR since the cmake file was changed. so it need your approval.

[09:42:28] ++ grep cmake/external
[09:42:28] + API_CHANGE=cmake/external/mklml.cmake
[09:42:28] + echo 'checking cmake/external change, PR: 15931, changes: cmake/external/mklml.cmake'
[09:42:28] + '[' cmake/external/mklml.cmake ']'
[09:42:28] + '[' 15931 '!=' '' ']'
[09:42:28] ++ curl -H 'Authorization: token 95a329e820eeae12c4c5c8a0b6e3e39654dc01a4' 'https://api.github.com/repos/PaddlePaddle/Paddle/pulls/15931/reviews?per_page=10000'
[09:42:28] ++ python /paddle/tools/check_pr_approval.py 1 2887803
[09:42:28] % Total % Received % Xferd Average Speed Time Time Time Current
[09:42:28] Dload Upload Total Spent Left Speed
[09:42:28]
[09:42:28] 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
[09:42:29] 0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
[09:42:29] 100 5 100 5 0 0 3 0 0:00:01 0:00:01 --:--:-- 3
[09:42:29] 100 5 100 5 0 0 3 0 0:00:01 0:00:01 --:--:-- 3
[09:42:29] + APPROVALS=FALSE
[09:42:29] + echo 'current pr 15931 got approvals: FALSE'
[09:42:29] + '[' FALSE == FALSE ']'
[09:42:29] + echo 'You must have panyx0718 approval for the api change! cmake/external'
[09:42:29] + exit 1
[09:42:43] Process exited with code 1

@panyx0718
Copy link
Contributor

panyx0718 commented Feb 27, 2019

rerun? cmake file no longer need my review

@yihuaxu
Copy link
Contributor Author

yihuaxu commented Feb 28, 2019

rerun? cmake file no longer need my review

Yes, as mklml library's compatible issue, so update it. thanks for your help!

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 4efdebc into PaddlePaddle:develop Feb 28, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants