Through work with the Stampede folks we explored the performance of MPI_THREAD_MULTIPLE on 5 series KNC. A repo describing the tests is here:
https://github.com/cwsmith/commKernel
I'm not sure how many folks delved into the dark recess that is MPI_THREAD_MULTIPLE, but this may bring some light to where there is little.
Through work with the Stampede folks we explored the performance of MPI_THREAD_MULTIPLE on 5 series KNC. A repo describing the tests is here:
https://github.com/cwsmith/commKernel
I'm not sure how many folks delved into the dark recess that is MPI_THREAD_MULTIPLE, but this may bring some light to where there is little.