1 Reply Latest reply on May 31, 2018 5:41 PM by maryt_intel

    pthread and threaded mkl




      I have a parallel code which uses both pthread and threaded mkl.

      Attached is a minimal working example. I set the environment

      variable MKL_NUM_THREADS equal to 2. I run this code and

      the ratio walltime/cputime is 2.5.


      When I comment out the pthread part in the for loop, compile and run,

      the ratio is 1.96. When I comment out the dgemv_ and compile, the

      ratio is 1.81.


      If my understanding is correct, the ratio for the first case should be

      below 2? Or there are some conflicts between pthread and threaded mkl?

      Please correct me if I was wrong.