very different results from FFT when using MKL internal threading

[cgg_distribution]

The following code returns different results (that do not seem to be down to numerical precision). depending on whether I set the mkl internal threading.

I have attached the input (data1.txt) to reproduce the issue.

And also the outputs:

data1_out_4f.txt ( with  mkl_set_num_threads(4) )

data1_out_1f.txt ( with  mkl_set_num_threads(1) )

program fft_issue

        USE MKL_DFTI

        implicit none

        include  'mkl.fi'

        integer length
        integer ntr,i,j,itr,nfft
        real, allocatable :: buffer(:)
        complex, allocatable :: buffer_out(:)
        complex, allocatable :: s1(:)
        integer status

        type(dfti_descriptor), pointer :: My_Desc1_Handle

        length = 2051
        nfft = 3200

        allocate(buffer(length))
        allocate(buffer_out(nfft))
        allocate(s1(nfft))

        open(1, file = 'data1.txt')

        do j = 1 , 2051
           read(1,*), buffer(j)
        enddo
        call mkl_set_num_threads(4)

        s1(:length) = cmplx(buffer(:))
        s1(length+1:) = 0

        status = DftiCreateDescriptor(My_Desc1_Handle,DFTI_SINGLE,DFTI_COMPLEX,1,3200)
        status = DftiSetValue(My_Desc1_Handle, DFTI_PLACEMENT, DFTI_NOT_INPLACE);
        status = DftiSetValue(My_Desc1_Handle, DFTI_NUMBER_OF_TRANSFORMS, 1);
        status = DftiSetValue(My_Desc1_Handle, DFTI_BACKWARD_SCALE, 1.0/nfft);
        status = DftiCommitDescriptor(My_Desc1_Handle);
        status = DftiComputeForward( My_Desc1_Handle, s1, buffer_out )
        status = DftiComputeBackward( My_Desc1_Handle, buffer_out, s1 )

        open(1, file = 'data1_out_4f.txt')

        do j = 1 , nfft
           write(1,*), s1(j)
        enddo


      end program

I am running the program on a 4-core Intel(R) Xeon(R) CPU E3-1240 v3 @ 3.40GHz

I am using mkl 19 update 5

My compile command as follows:

/opt/intel/19/bin/ifort reproduce_intel.f90 -Wl,--start-group ${MKLROOT}/lib/intel64/libmkl_intel_lp64.a ${MKLROOT}/lib/intel64/libmkl_intel_thread.a ${MKLROOT}/lib/intel64/libmkl_core.a -Wl,--end-group -liomp5 -lpthread -lm -ldl -I${MKLROOT}/include

[Gennady_F_Intel]

thanks, Marie for the case. At the very first glance, everything is correct with this case. We will check and get back asap.

[Gennady_F_Intel]

the issue has been reproduced and escalated against the MKL FFT team. We will keep this thread informed when the fix will be ready to release.