[GitHub] TaoLv commented on issue #9545: Profiling discussion

2018-02-05 Thread GitBox
TaoLv commented on issue #9545: Profiling discussion URL: https://github.com/apache/incubator-mxnet/issues/9545#issuecomment-363281815 > If you have better code to determine the number of "real" cores, I will use it. > Right now, we use Intel OMP (at least when building with cmake on

[GitHub] TaoLv commented on issue #9545: Profiling discussion

2018-01-31 Thread GitBox
TaoLv commented on issue #9545: Profiling discussion URL: https://github.com/apache/incubator-mxnet/issues/9545#issuecomment-361874591 > "I think, for mordern deep learning workloads and stencil computation, it would be better to keep a worker thread pool and bind one worker thread to one

[GitHub] TaoLv commented on issue #9545: Profiling discussion

2018-01-31 Thread GitBox
TaoLv commented on issue #9545: Profiling discussion URL: https://github.com/apache/incubator-mxnet/issues/9545#issuecomment-361874591 > "I think, for mordern deep learning workloads and stencil computation, it would be better to keep a worker thread pool and bind one worker thread to one

[GitHub] TaoLv commented on issue #9545: Profiling discussion

2018-01-30 Thread GitBox
TaoLv commented on issue #9545: Profiling discussion URL: https://github.com/apache/incubator-mxnet/issues/9545#issuecomment-361808896 > " if there is no operator parallelism, maybe we should give as many threads as possible to the running operator, especially for mkldnn operator" >

[GitHub] TaoLv commented on issue #9545: Profiling discussion

2018-01-30 Thread GitBox
TaoLv commented on issue #9545: Profiling discussion URL: https://github.com/apache/incubator-mxnet/issues/9545#issuecomment-361807553 > "After set USE_OPERATOR_TUNING=0 and rebuilt mxnet, the first 56 threads were not there anymore. Do you think this behaviour is as expectaton?" >

[GitHub] TaoLv commented on issue #9545: Profiling discussion

2018-01-30 Thread GitBox
TaoLv commented on issue #9545: Profiling discussion URL: https://github.com/apache/incubator-mxnet/issues/9545#issuecomment-361807553 > "After set USE_OPERATOR_TUNING=0 and rebuilt mxnet, the first 56 threads were not there anymore. Do you think this behaviour is as expectaton?" >

[GitHub] TaoLv commented on issue #9545: Profiling discussion

2018-01-30 Thread GitBox
TaoLv commented on issue #9545: Profiling discussion URL: https://github.com/apache/incubator-mxnet/issues/9545#issuecomment-361519560 Hi Chris, thanks for your detailed explaination for the omp threads in operator. I used vtune to see the running state of all omp threads in a mxnet

[GitHub] TaoLv commented on issue #9545: Profiling discussion

2018-01-25 Thread GitBox
TaoLv commented on issue #9545: Profiling discussion URL: https://github.com/apache/incubator-mxnet/issues/9545#issuecomment-360400949 Hi Chris, thank you for creating this discussion issue. Some questions from my side: 1. Can I get some statistical infomation from profiling