Subversion Repositories shark

Rev

Details | Last modification | View Log | RSS feed

Rev Author Line No. Line
2 pj 1
This directory contains a test program, fftw_test, for the one- and
2
multi-dimensional transforms.  There is also a corresponding program,
3
rfftw_test, for the real-complex transforms, which takes the same
4
command-line options as fftw_test.  (See the FFTW manual for
5
compilation instructions.)
6
 
7
They take a number of command-line options that determine what to
8
tests.  (On systems lacking a command-line, they will prompt the user
9
interactively for this information.)  We use the following
10
conventions.
11
 
12
   <n> is a size parameter, and it has the form N1xN2xN3... .  For
13
example, 3x8 denotes a two-dimensional array with 3 rows and 8
14
columns.  A single number N denotes a 1D transform of size N.  For the
15
cases where it matters, the syntax xN denotes a ND-transform of rank
16
1.  For example, x5 denotes a n-dimensional array of rank 1 with 5
17
elements.  (This matters for real->complex transforms)
18
 
19
   <rank> is an integer.  Rank 0 denotes 1D transforms.  Rank >= 1
20
denotes n-dimensional transforms.  Again, rank 0 uses [R]FFTW, but
21
rank >= 1 uses [R]FFTWND.
22
 
23
   -r <rank>           Tests the forward and backward transforms of
24
                       random dimensions and sizes.  Does not terminate
25
                       (press ^C when you are tired).  We use the FFT
26
                       testing algorithm described in:
27
 
28
                       Funda Ergün, "Testing multivariate linear functions:
29
                       Overcoming the generator bottleneck." Proceedings of
30
                       the Twenty-Seventh Annual ACM Symposium on the Theory
31
                       of Computing, pp. 407-416 (1995).
32
 
33
   -c <n>              Checks the forward and backward transform of size <n>
34
 
35
 
36
   -s <n>              Test the speed of the transform of size <n>
37
                       Times both in-place and out-of-place transforms.
38
 
39
   -a <rank>           Like fftw_test -c <n>  for all n = 1 ... infinity.
40
	               Does not terminate (press ^C when you are tired)
41
 
42
   -p <rank>           Exercises the planner to make sure there are no
43
                       memory leaks.
44
 
45
   -m                  Uses the FFTW_MEASURE flag (instead of FFTW_ESTIMATE)
46
                       during correctness tests (this will slow things down
47
                       considerably).
48
 
49
   -w <file>           Tests the "wisdom" mechanism (wisdom is used
50
                       in all transforms computed...this option should
51
                       precede one of the other options above).  Also,
52
                       loads wisdom from <file> and stores accumulated
53
                       wisdom in <file> (after tests are completed).
54
                       (<file> is created if it does not exist.)
55
 
56
   -t                  Tests the resolution of the timer you are currently
57
                       using.  This is done by measuring the time per
58
                       iteration of a simple loop--the time/iter should
59
                       converge to some value for large numbers of iterations,
60
                       and the rate of convergence can be used to infer
61
                       something about the timer resolution.
62
 
63
                       The minimum timing interval required to achieve
64
                       various levels of consistency/accuracy is outputted.
65
                       You can compare this to the FFTW_TIME_MIN setting
66
                       in fftw.h to see how appropriate it is for your
67
                       machine's clock.
68
 
69
   -v                  Verbose output (should precede other options).
70
 
71
   -h                  Prints help on these and other command-line flags