8000 Testsuite hangs when built with MKL 2018.2 (but not 2017.1) · Issue #10993 · numpy/numpy · GitHub
[go: up one dir, main page]

Skip to content

Testsuite hangs when built with MKL 2018.2 (but not 2017.1) #10993

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

Closed
bodgerer opened this issue Apr 27, 2018 · 5 comments
Closed

Testsuite hangs when built with MKL 2018.2 (but not 2017.1) #10993

bodgerer opened this issue Apr 27, 2018 · 5 comments

Comments

@bodgerer
Copy link

Hi,

I've noticed that numpy's testsuite hangs when numpy is built against MKL 2018.2 - but runs to completion when built against MKL 2017.1.

This is the case for numpy v1.14.2, 1.14.x HEAD and master branches.

I'm on CentOS 7.4 x86_64 on a Broadwell processor, python 2.7.5 (but also affects other pythons), Cython 0.28.2, nose 1.3.7.

By executing this: python -c "import numpy; numpy.test(verbose=2)"

I can see that the hang occurs while running this test: numpy.distutils.tests.test_exec_command.TestExecCommand.test_basic

At this point, the python interpreter is consuming an entire CPU core and, if I strace it, I see a stream of identical lines, all showing:

sched_yield() = 0

Any ideas, please?

Thanks!

@mattip
Copy link
Member
mattip commented Apr 27, 2018

Is it a duplicate of #10060, where it is suggested to use this work around?

# work-around 1:
KMP_INIT_AT_FORK=FALSE python script.py
# work-around 2:
MKL_THREADING_LAYER=TBB python script.py` 

Edit: add question mark

@bodgerer
Copy link
Author

Got it: sorry for the noise. I did search the issues list, just not well enough - honest!

Thanks

@mattip
Copy link
Member
mattip commented Apr 27, 2018

Our fault, that issue was closed but a new MKL has not been released yet. Please leave this open so others can find it

@oleksandr-pavlyk
Copy link
Contributor
oleksandr-pavlyk commented Jan 7, 2019

The fix in Intel OpenMP was released in 2018 update 3 of the library. New MKL is available on defaults, and Intel conda channels, as well as on PyPI.

Please close the ticket.

@charris
Copy link
Member
charris commented Jan 7, 2019

Thanks for the heads up.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

4 participants
0