leezu edited a comment on issue #17767: Fix and optimize handling of vectorized memory accesses URL: https://github.com/apache/incubator-mxnet/pull/17767#issuecomment-615419545 @ptrendx the built on master branch (`git clean -ffxd; python ci/build.py --platform publish.centos7_gpu_cu101 /work/runtime_functions.sh build_static_python_cu101_cmake --cache-intermediate` will produce: ``` -rw-r--r-- 1 ubuntu ubuntu 2.4G Apr 17 18:46 libmxnet.a -rwxr-xr-x 1 ubuntu ubuntu 2.0G Apr 17 18:46 libmxnet.so ``` On the other hand, with your PR ``` -rw-r--r-- 1 ubuntu ubuntu 2.8G Apr 17 19:12 libmxnet.a ``` So there is a significant binary size increase. I have some ideas for reducing the size of `.so`, but the most immediate (and easy to backport) solution is to reduce the number of architectures in https://github.com/apache/incubator-mxnet/tree/master/config/distribution For example instead of building for 5.0 and 5.2 or 7.0 and 7.5, just build for 5.0 and 7.0. Then MXNet may not make use of new features on 5.2 / 7.5 respectively, but the improvements by your PR are likely more important than the difference between 5.0 and 5.2 / 7.0 and 7.5
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to 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
