Contents
dfftopt - compute the length of the closest fast FFT
INTEGER FUNCTION DFFTOPT(LEN)
INTEGER LEN
INTEGER*8 FUNCTION DFFTOPT_64(LEN)
INTEGER*8 LEN
F95 INTERFACE
INTEGER FUNCTION DFFTOPT(LEN)
INTEGER :: LEN
INTEGER(8) FUNCTION DFFTOPT_64(LEN)
INTEGER(8) :: LEN
C INTERFACE
#include <sunperf.h>
int dfftopt(int len);
long dfftopt_64(long len);
dfftopt computes the length of the closest fast FFT. Fast
Fourier transform algorithms, including those used in Per-
formance Library, work best with vector lengths that are
products of small primes. For example, an FFT of length
32=2**5 will run faster than an FFT of prime length 31
because 32 is a product of small primes and 31 is not. If
your application is such that you can taper or zero pad your
vector to a larger length then this function may help you
select a better length and run your FFT faster.
DFFTOPT will return an integer no smaller than the input
argument N that is the closest number that is the product of
small primes. DFFTOPT will return 16 for an input of N=16
and return 18=2*3*3 for an input of N=17.
Note that the length computed here is not guaranteed to be
optimal, only to be a
product of small primes. Also, the value returned may
change as the underlying
FFTs become capable of handling larger primes. For exam-
ple, passing in N=51 to day will return 52=2*2*13 rather
than 51=3*17 because the FFTs in Performance Li brary do not
have fast radix 17 code. In the future, radix 17 code may
be added
and then N=51 will return 51.