is FFTW wrapper call thread safe inside OpenMP region?

mohanmuthu · ‎06-23-2023

Good Day!

I am using the FFTW compatible wrapper for FFT and compiling the code with /Qmkl:sequential option. It works well in serial execution. The question I have is,

(A) if I do the same inside the OpenMP parallel region as mentioned below, is it thread safe?

Also,

(B) do I need to change the compiler option from /Qmkl:sequential to /Qmkl:parallel, or its not required since I am not going to use parallel threads within FFTW.

!$OMP PARALLEL DEFAULT(NONE) PRIVATE(plan, x, y) SHARED(n, FFTW_ESTIMATE)
!$OMP DO
...
    CALL sfftw_plan_dft_r2c_1d(plan, n, x, y, FFTW_ESTIMATE)
    CALL sfftw_execute_dft(plan, x, y)
    CALL sfftw_destroy_plan(plan)
...
!$OMP END DO
!$OMP PARALLEL

Appreciate any help I can get on this topic.

Barbara_P_Intel · ‎06-23-2023

This is best answered by those on the MKL Forum. I'll move this thread there.