What is the intention of cufftXtSetWorkAreaPolicy?

Accelerated Computing GPU-Accelerated Libraries
1

I could not tell the actual use case for cufftXtSetWorkAreaPolicy call. How it is different from cufftSetAutoAllocation?
Appreciate any one can shed some light.

2

Hi,

depending on the chosen FFT algorithm, cuFFT needs to allocate a workspace for temporary data. By using cufftXtSetWorkAreaPolicy, you can constrain its size. It can also influence the algorithm choice. Currently there are 3 policies you can apply:

  1. CUFFT_WORKAREA_PERFORMANCE - default, no workspace size restrictions,
  2. CUFFT_WORKAREA_USER - the user specifies maximum workspace size by workSize parameter,
  3. CUFFT_WORKAREA_MINIMAL - the plan will be recreated without workspace memory, but it has to be a single GPU C2C (up to 4096 in any dimension), Z2Z (up to 2048 in any dimension) transformation.

Hope it helps.

Dave