FFTW gives wrong results in comparison to MATLAB

Question

After analysing some signal data, I do some resynthesis. For analysing e.g. a wav-file I use CUDA-CUFFT for 1D-complex-FFT-transformation, after some signal processing operations (cutting pieces of signals,...) I do an IFFT-tranformation using FFTWF (FFTW=for float type).

On the CUDA-side I verified the FFT-transformation with MATLAB and get almost the same results (due to round-off errors it's not exactyl the same). After cutting the signal and performin the FFTShift-operation the signals seem to be almost equal.

But after the IFFT-operation I get complete different results (directly after the IFFT-operation I reverse the output, which I also do in MATLAB -> and finally I save the FFTW-output to a binary file for comparison)

P.S:(for additional information. after flipping the output I norm the results with sqrtf(FFTLEN); (on the CUDA-site I norm the output with 1/sqrtf(FFTLEN)

For the IFFT-transformation with FFTW I use the following syntax in the documentation: http://www.fftw.org/doc/Complex-One_002dDimensional-DFTs.html But instead of using the fftw_complex type I use the C++-complex-type and cast it to fftw_complex-type. The operation sign is FFTW_BACKWARD beacuse I use CUFFT_FORAWRD on the CUDA-site:

fftwf_plan localFFTPlan;
localFFTPlan = fftwf_plan_dft_1d((int)sFFTParams.uFFTLength, reinterpret_cast<fftwf_complex*>(&tFFTraw[0]), reinterpret_cast<fftwf_complex*>(&tFFTin[0]), FFTW_BACKWARD, FFTW_MEASURE);
fftwf_execute(localFFTPlan);
fftwf_destroy_plan(localFFTPlan);

In MATLAB the same operation is performed like this:

flipud(ifft(fftshift(st))

Can anyone help me, why the results get wrong directyl after the IFFT-transformation(as shown above, before the operation everything is fine)?

EDIT the array allocation looks like the following statement: TFFTContainer tFFTin(sFFTParams.uFFTLength, gComplexZero);