Hi @azai91 , you can try to use llvm from `brew install llvm` instead of apple llvm built in xcode to enable OPENMP on both mxnet and mkldnn level. And I found this llvm also supports OPENCV well compared with `brew install gcc`. Details are as below: ``` brew install llvm # .bash_profile export LIBRARY_PATH=/usr/local/Cellar/llvm/6.0.1/lib/ # config.mk and mkldnn.mk CC=/usr/local/Cellar/llvm/6.0.1/bin/clang CXX=/usr/local/Cellar/llvm/6.0.1/bin/clang++ ``` And I get better performance on our local iMac Pro (Xeon W 8 cores with AVX-512): ``` INFO:root:network: vgg-16 INFO:root:device: cpu(0) INFO:root:batch size 1, image/sec: 20.913986 INFO:root:batch size 16, image/sec: 24.273071 INFO:root:batch size 32, image/sec: 24.704907 INFO:root:batch size 64, image/sec: 24.946080 INFO:root:batch size 128, image/sec: 25.074148 INFO:root:batch size 256, image/sec: 25.262900 INFO:root:network: inception-v3 INFO:root:device: cpu(0) INFO:root:batch size 1, image/sec: 41.431404 INFO:root:batch size 16, image/sec: 54.312317 INFO:root:batch size 32, image/sec: 54.604119 INFO:root:batch size 64, image/sec: 54.395680 INFO:root:batch size 128, image/sec: 54.410785 INFO:root:batch size 256, image/sec: 54.614424 INFO:root:network: inception-v4 INFO:root:device: cpu(0) INFO:root:batch size 1, image/sec: 20.715221 INFO:root:batch size 16, image/sec: 26.249734 INFO:root:batch size 32, image/sec: 26.197659 INFO:root:batch size 64, image/sec: 26.161530 INFO:root:batch size 128, image/sec: 26.247461 INFO:root:batch size 256, image/sec: 26.313875 INFO:root:network: resnet-50 INFO:root:device: cpu(0) INFO:root:batch size 1, image/sec: 40.505148 [23:07:34] src/operator/nn/mkldnn/mkldnn_base.cc:74: Allocate 51380224 bytes with malloc directly INFO:root:batch size 16, image/sec: 41.431191 INFO:root:batch size 32, image/sec: 40.563484 INFO:root:batch size 64, image/sec: 37.070624 INFO:root:batch size 128, image/sec: 37.709096 INFO:root:batch size 256, image/sec: 37.448483 INFO:root:network: resnetv1-50 INFO:root:device: cpu(0) INFO:root:batch size 1, image/sec: 41.701090 INFO:root:batch size 16, image/sec: 43.132788 INFO:root:batch size 32, image/sec: 41.613291 INFO:root:batch size 64, image/sec: 38.133290 INFO:root:batch size 128, image/sec: 38.839577 INFO:root:batch size 256, image/sec: 38.853521 INFO:root:network: mobilenet INFO:root:device: cpu(0) INFO:root:batch size 1, image/sec: 200.916080 INFO:root:batch size 16, image/sec: 287.614019 INFO:root:batch size 32, image/sec: 277.838051 INFO:root:batch size 64, image/sec: 274.474078 INFO:root:batch size 128, image/sec: 273.622323 INFO:root:batch size 256, image/sec: 273.445636 ```
[ Full content available at: https://github.com/apache/incubator-mxnet/pull/12724 ] This message was relayed via gitbox.apache.org for devnull@infra.apache.org
