For implementing STFT for Intel based MACs we have a solution using Intel’s MKL library (oneAPI -> MKL ->Dfti API).
For M1 architecture we were trying to implement the same functionality on top of the FFTW.
Here we have differences with processed outputs via MKL and FFTW.
For FFTW configuration we used the fftwf_plan_many_dft_r2c() API, as per initial investigation it has the same configurations used in MKL solution, i.e. in/out strides, in/out distances, etc.
We created sample app evaluating the mentioned difference, see attached.
fftw_mkl_test.zip
For implementing STFT for Intel based MACs we have a solution using Intel’s MKL library (oneAPI -> MKL ->Dfti API). For M1 architecture we were trying to implement the same functionality on top of the FFTW. Here we have differences with processed outputs via MKL and FFTW.
For FFTW configuration we used the fftwf_plan_many_dft_r2c() API, as per initial investigation it has the same configurations used in MKL solution, i.e. in/out strides, in/out distances, etc. We created sample app evaluating the mentioned difference, see attached. fftw_mkl_test.zip