Chris@19: Chris@19:
Chris@19:Chris@19: Next: Thread safety, Chris@19: Previous: Usage of Multi-threaded FFTW, Chris@19: Up: Multi-threaded FFTW Chris@19:
There is a fair amount of overhead involved in synchronizing threads, Chris@19: so the optimal number of threads to use depends upon the size of the Chris@19: transform as well as on the number of processors you have. Chris@19: Chris@19:
As a general rule, you don't want to use more threads than you have Chris@19: processors. (Using more threads will work, but there will be extra Chris@19: overhead with no benefit.) In fact, if the problem size is too small, Chris@19: you may want to use fewer threads than you have processors. Chris@19: Chris@19:
You will have to experiment with your system to see what level of
Chris@19: parallelization is best for your problem size. Typically, the problem
Chris@19: will have to involve at least a few thousand data points before threads
Chris@19: become beneficial. If you plan with FFTW_PATIENT
, it will
Chris@19: automatically disable threads for sizes that don't benefit from
Chris@19: parallelization.
Chris@19:
Chris@19:
Chris@19:
Chris@19:
Chris@19: