In current code, no matter what number of threads specified, all
available CPU count is used when invoking OMP, which leads to very bad
performance if the workload is small while all available CPUs are big.
Lots of time are wasted on inter-thread sync. Fix this issue by really
using the number specified by the variable 'num' from calling API.
Signed-off-by: Chen, Guobing <guobing.chen@intel.com>
* Allow choosing the OpenMP scheduler and add range hint for GEMM_MULTITHREAD_THRESHOLD
* Amended description of GEMM_MULTITHREAD_THRESHOLD
to reflect #742 making it track floating point operations rather than matrix size
The problem is the mutex we used in blas_server. Thus, we must clear
the mutex before the fork and re-init them at parent and child process.
If you used OpenMP, GOMP has the same problem by now. Please try other OpenMP
implemantation.
Chen, Guobing
Martin Kroeker
zhiyong.dang
Timothy Gu
Olivier Grisel
Zhang Xianyi
Xianyi Zhang