CUFFT with UINT16 Using CUFFT for UINT16 ???

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

Our problem is as following : We have a very large dataset , which limits us to work only with UINT16 numbers.
At a certain point, we have to take FFT of this data and that’s where we get stuck . The problem is not
a memory, but the variable type UINT16 .

Our questions would be :

  1. Is it possible to use cufftExecR2C with UINT16 data instead of float ( real ) ?
  2. If yes , how ? If no, are there any other tricks ( ways ) to overcome this obstacle ?

BR,

Coll

2

I’m curious what you consider a very large data set.
You say memory isn’t an issue, so why not convert to float?
If memory is an issue, you can always convert to float and compute the fft in pieces.