topic Re: MKL FFT functions - compute the N-point fft? in Intel® oneAPI Math Kernel Library

MKL FFT functions - compute the N-point fft?

andreasjh — Thu, 22 Apr 2021 13:24:25 GMT

Hello,

Is there a way to compute the n-point fft of an 1D array in MKL (C-language, Linux) for an arbitrary number of n?
The problem is that i need to translate the matlab function fft(X,N) to use with the MKL-lib. The function inside matlab is described as : "fft(X,N) is the N-point fft, padded with zeros if X has less
than N points and truncated if it has more." (see ref.)

As an example, lets say i have a:
Input vector, I_vec, with size of 100 (can vary from 1 to 1500) elements and a output vector, O_vec, with size of 2048 (can vary between the fifth closest power of 2 of input vec).
I want to compute the values on these 100 elements to "fill up" the 2048 (1D) vector.
Let's say that i have computed the values by the DftiComputeForward. I am then expecting the output to be:
inplace fft for I_vec: I_vec[0] = 0.1, i_vec[1] = 0.2, ....
out of place fft for I_vec to O_vec: O_vec[0] = 0.1 ... O_vec[~10] = 0.15 ... O_vec[~20] = 0.2.

Is this possible? I have tried to read the reference document and looked at the examples - but cant find anything that does this kind of computation straight away. Maybe do a number of transformations and try to pad inside the vector? If you have any tips to where to look or can give an example of how to do this, it would be highly appreciated.

Thanks in advance,
Andreas

More info of the n-point fft can be found on: https://www.dsprelated.com/showthread/comp.dsp/56353-1.php and the matlab api: https://se.mathworks.com/help/matlab/ref/fft.html#d123e392801

Specs:
Ubuntu 20.04 (linux)
Intel oneAPI integrated in Eclipse IDE (v. 2021.2.0)
Intel C++ compiler
Intel MKL - static lib
C-languge
Intel core i7-8750h

Re: MKL FFT functions - compute the N-point fft?

MRajesh_intel — Mon, 17 May 2021 11:45:53 GMT

Hi,

The output 2048 array will be padded with zeros after the computation is completed. You may refer to this sample.

#include <stdio.h> #include <stdlib.h> #include <complex.h> #include "mkl.h" #include "mkl_dfti.h" float _Complex c2c_input[100]; float _Complex c2c_output[2048]; DFTI_DESCRIPTOR_HANDLE my_desc_handle = NULL; MKL_LONG status; int main(int argc, char* argv[]) { int i; for(i=0;i<100;i++) c2c_input[i]=i; status = DftiCreateDescriptor(&my_desc_handle, DFTI_SINGLE,DFTI_COMPLEX, 1, 100); status = DftiSetValue(my_desc_handle, DFTI_PLACEMENT, DFTI_NOT_INPLACE); status = DftiCommitDescriptor(my_desc_handle); status = DftiComputeForward(my_desc_handle, c2c_input,c2c_output); status = DftiFreeDescriptor(&my_desc_handle); for(i=0;i<110;i++) printf("%f\t",creal(c2c_output[i])); return 0; }

If this is not the kind of computation you were looking for, can you please elaborate more?

Regards

Rajesh.

Re:MKL FFT functions - compute the N-point fft?

MRajesh_intel — Mon, 24 May 2021 14:36:36 GMT

Hi,

Can you please give us any updates regarding the query?

Regards

Rajesh.

Re: MKL FFT functions - compute the N-point fft?

andreasjh — Tue, 25 May 2021 08:52:24 GMT

Hi Rajesh,

I managed to get my wanted result working by trial and error principle but your reply was really spot on.

The reply is more then satisfying.
Thanks you so much for the help, really appreciated!

MVH,
Andreas

Re:MKL FFT functions - compute the N-point fft?

MRajesh_intel — Tue, 25 May 2021 09:03:31 GMT

Hi,

Thanks for the confirmation!

As this issue has been resolved, we will no longer respond to this thread. If you require any additional assistance from Intel, please start a new thread. Any further interaction in this thread will be considered community only.

Have a Good day.

Regards

Rajesh