[GitHub] fhieber commented on issue #8532: mxnet-mkl (v0.12.0) crash when using (conda-installed) numpy with MKL

[GitBox <gi...@apache.org>]

fhieber commented on issue #8532: mxnet-mkl (v0.12.0) crash when using (conda-installed) numpy with MKL
URL: https://github.com/apache/incubator-mxnet/issues/8532#issuecomment-425350410
 
 
   I finally got back to this after a while. I no longer observe the libomp-related error mentioned in the original issue, but I am observing process deadlocks with the following numpy/mxnet configuration in Sockeye:
   ```
   conda list | grep numpy
   numpy                     1.15.1           py36h6a91979_0
   numpy-base                1.15.1           py36h8a80b8c_0
   conda list | grep mkl
   blas                      1.0                         mkl
   mkl                       2019.0                      118
   mkl_fft                   1.0.4            py36h5d10147_1
   mkl_random                1.0.1            py36h5d10147_1
   mxnet-mkl                 1.3.0.post0               <pip>
   ```
   If mkl-optimized numpy is installed via anaconda (as shown above) and using mxnet-mkl==1.3.0.post0 on a Mac laptop, the Sockeye subprocess spawned at a checkpoint (to decode the validation data set), is unable to spawn and the main process deterministically hangs. When debugging, it seems that it fails to spawn the subprocess.
   However, when using either mxnet==1.3.0.post0 (no mkl) or pip-installed numpy (no mkl), everything works just fine.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services