mkl_serv_set_parallel and mkl_vml_core_t7_vmlreduce_pio2d What are these?

[tbruegge]

I have an application where I make calls to the Vector Math Library routines to compute Sin and Cos (in a single call) (using the c interface; vdSinCos) and to compute the exponential function (vdExp). When I run VTune,I see a large amount of time being spent in the two functions mkl_serv_set_parallel and mkl_vml_core_t7_vmlreduce_pio2d. Can anyone tell me what these functions do? Is the fact that my job is taking longer in these functions than it is apparently taking in the SinCos and Exp routines an indication of a problem?

Thanks for any help you can give. I am using MKL for the first time.

[Andrey_N_Intel]

Thank you for your question. Itwould be nice if you give us a bit more details on your application. Could youclarifyversion of MKL you use? Also,it is useful to know length of input vector a you pass into vdExp and vdSinCos functions and range of the argument values. The fact that theapplication spends significant amount of time in the function mkl_vml_core_t7_vmlreduce_pio2dis the indicator that vdSinCos is called on large argument values. If this is the case, is it really important for your application to process large values of parameters and calculate Sin and Cosfor them?As for function mkl_serv_set_parallel it could be nice if you providea test case which will demonstrate this issue.

[tbruegge]

Thanks for your reply. You were correct, the arguments to the vdSinCos function were extremely large, and reducing these arguments is helping. My application is large, but I may be able to generate a simpler case that demonstrates the large amount of time spent in mkl_serv_set_parallel. However, if you could tell me what this function does, that might be a clue for me to track down the problem.

Thanks

[Andrey_N_Intel]

This is the service function intended for setting the environmental variable of the library. The test case couldbe useful for clarification of the reasons of the issue.Thanks.