Hi DTolm!

First of all thank you very much for this wonderful package!

I need to take a 1D FFT over an array with a dynamic size.
The array has dimensions (X,N) with the FFT taken over X=constant and the batch number N changes per frame. X is the non-strided axis.

I suppose there are two ways to achieve this:

1. Allocate a large enough buffer to accommodate the largest values of N=Nmax and perform batched FFT.
   -> In this case, is there a way to dynamically change the number of batches that is actually performed? It is a waste if all Nmax FFT's are performed when N < Nmax.

2. Perform N single 1D FFT's and rewrite the command buffer every frame to accommodate changing N.
   -> In this case, the transform size would be set to X and offsets are used to select the right part of the buffer. However, any data past element no. X is returned as zero.

Do you have a suggestion how to solve either (or both) of these issues?

I'm using my own compilation of pyvkfft with Vulkan backend, VkFFT version 1.3.4.