TaoLv commented on issue #17980: When compiled with MKL, fully_connected calls DNNL while dot and batch_dot call MKL URL: https://github.com/apache/incubator-mxnet/issues/17980#issuecomment-609860069 Thank you @kpuatamazon. I will try to reproduce the performance issue and get back to you later. > (Which DNNL is in master?) It should be v1.2.2 but was down graded to v1.1.2 by https://github.com/apache/incubator-mxnet/pull/17084 (maybe by accident). > But choosing to use MKL should be 1) atomic and 2) not require undocumented compilation options. Not sure I understand what do you mean. But using MKL BLAS is documented in: - makefile: https://github.com/apache/incubator-mxnet/blob/master/make/config.mk#L120 - cmake: https://github.com/apache/incubator-mxnet/blob/master/cmake/ChooseBlas.cmake#L19 - website: https://mxnet.incubator.apache.org/get_started/build_from_source#math-library-selection
---------------------------------------------------------------- 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
