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.

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

Compile speed is slower than 1 month ago (13 minutes -> 22 minutes) #5051

Closed
luotao1 opened this issue Oct 24, 2017 · 6 comments
Closed

Compile speed is slower than 1 month ago (13 minutes -> 22 minutes) #5051

luotao1 opened this issue Oct 24, 2017 · 6 comments

Comments

@luotao1
Copy link
Contributor

luotao1 commented Oct 24, 2017

Following are two teamcity logs:

  • #6469 (24 Oct 17 16:19)
    • It costs 27 minutes totally, 22 minutes for compiler.
    • 6 minutes (16:19-16:25) for compilingWITH_GPU=False
    • 16 minutes (16:25-16:41) for compilingWITH_GPU=true
    • 4 minutes for unittest. (Total Test time (real) = 230.80 sec)
  • #5019 (30 Sep 17 03:49)
    • It costs 17 minutes totally, 13 minutes for compiler.
    • 4 minutes (3:49-3:53) for compilingWITH_GPU=False
    • 9 minutes (3:53-4:02) for compilingWITH_GPU=true
    • 2 minutes for unittest. (Total Test time (real) = 137.64 sec)

From 13 minutes to 22 minutes, current compile speed is almost 1 times slower than 1 month ago. Maybe the reason is that there are a lot of ops merged now.

Hope somebody can help to make the compiler faster, which will make developing faster as well.

@luotao1
Copy link
Contributor Author

luotao1 commented Oct 24, 2017

The reason why teamcity compile both WITH_GPU=False and WITH_GPU=True is in #4476

@lcy-seso
Copy link
Contributor

lcy-seso commented Oct 24, 2017

I felt recent days the compiling time become much slower (but I did not check exactly what happened. I just can clearly feel the changes of compiling speed becomes much slower).

I changed my newly added operator, but many source files (other operators) I did not change re-compiled. Are there any changes in the dependencies (only guess)?

@wangkuiyi
Copy link
Collaborator

Is this due to book chapters being added into paddle/framework/tests/test_*?

@luotao1
Copy link
Contributor Author

luotao1 commented Nov 3, 2017

Is this due to book chapters being added into paddle/framework/tests/test_*?

I think this is not the main reason, since this issue is created on Oct.24, when book chapters haven't been added in.

@wangkuiyi
Copy link
Collaborator

wangkuiyi commented Nov 3, 2017

I noticed that in TeamCity log there are

[21:12:28]	251/256 Test #233: test_seq_conv ............................... Passed 19.33 sec
[21:12:30]	252/256 Test #256: test_recommender_system .....................***Failed 8.88 sec
[21:12:30]	Cache file /root/.cache/paddle/dataset/movielens/ml-1m.zip not found, downloading http://files.grouplens.org/datasets/movielens/ml-1m.zip
[21:12:30]	[ ]
[21:12:30]	[= ]
[21:12:30]	[== ]
[21:12:30]	[=== ]
[21:12:30]

which means that our book chapter tests in python/paddle/v2/framework/tests are downloading datasets.

I understand that this is because our python/datasets package, which is used by above chapter tests, cannot find the local cache of these datasets, so it downloads.

This would definitely cause slow CI tests.

@luotao1
Copy link
Contributor Author

luotao1 commented Nov 20, 2017

close due to #5712 and #5491

@luotao1 luotao1 closed this as completed Nov 20, 2017
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

No branches or pull requests

3 participants