Thursday, April 9, 2015

P3DFFT timers

P3DFFT is one of the fastest parallel Fourier transform (https://code.google.com/p/p3dfft/). In output of P3DFFT, 12 timers are printed. These are time takes by various parts of it. The following table lists the timer number and corresponding function.


timer[1]Alltoallv in XY transpose
timer[2]Alltoallv in YZ transpose
timer[3]Alltoallv in ZY transpose
timer[4]Alltoallv in YX transpose
timer[5]FFT transform (R2C) in X for all z and y
timer[6]local rearrangement of data in XY transpose
timer[7]FFT transform (C2C) in Y for all x and z, one Z plane at a time
timer[8]local rearrangement of data in YZ transpose
timer[9]local rearrangement of data in ZY transpose
timer[10]FFT Transform (C2C) in y dimension for all x, one z-plane at a time
timer[11]local rearrangement of data in YX transpose
timer[12]Perform Complex-to-real FFT in x dimension for all y and z

Note: Description of timer[8], and timer[9] may not be accurate.

No comments: