asitstands commented on issue #10951: [MXNET-545] Fix broken cython build
URL: https://github.com/apache/incubator-mxnet/pull/10951#issuecomment-446253481
 
 
   @larroy This PR just fixes a broken existing feature. Why do we need cython? 
We may have some performance gain. Here is the copy of a quick benchmark result 
I reported before. For a quick benchmark I have run 
`example/gluon/image_classification.py` with `--model=resnet50_v1 
--dataset=cifar10` on a machine with two Xeon E5-2680 and 1080ti. The mean time 
consumption per epoch was 72.2s with cython and 78.6s without cython. So the 
the running time with cython is about 8% shorter. With `--model=resnet101_v1`, 
the gain was about 10% (135.3s vs 150.2s).
   
   For cmake, I added `BUILD_CYTHON_MODULES` option. 'GPU: CMake' environment 
builds the cython modules via cmake but no unit test runs with the artifact 
from the build. I left a comment about the llvm openmp issue at the unit test 
'Python3: GPU'. 
   
   
https://github.com/apache/incubator-mxnet/blob/adce15e25afdf3a47f5dc26bda9db7371aa4882b/ci/docker/runtime_functions.sh#L665-L682
   
   
https://github.com/apache/incubator-mxnet/blob/10d817cea15ec0b3253f3b78a30052f764931026/ci/jenkins/Jenkins_steps.groovy#L599-L612
   

----------------------------------------------------------------
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:
[email protected]


With regards,
Apache Git Services

Reply via email to