I’d like to do the following in cuda. I have two arrays, A1, A2, of ‘oversampled’ time-series, whose convolution I require. I would like to filter + decimate, giving B1 and B2, and then compute their convolution in cufft. My question is whether I can use the the cufft callback api for doing the filter + decimate step? I have not been able to find an example online in which the load callback routine changes the size of the input arrays before computing the fft, as I am proposing to do. Is it possible?