FFTW 3.3.8: Using Plans

Plans for all transform types in FFTW are stored as type Chris@82: fftw_plan (an opaque pointer type), and are created by one of the Chris@82: various planning routines described in the following sections. Chris@82: Chris@82: An fftw_plan contains all information necessary to compute the Chris@82: transform, including the pointers to the input and output arrays. Chris@82:

Chris@82:

void fftw_execute(const fftw_plan plan);
Chris@82:

This executes the plan, to compute the corresponding transform on Chris@82: the arrays for which it was planned (which must still exist). The plan Chris@82: is not modified, and fftw_execute can be called as many times as Chris@82: desired. Chris@82:

To apply a given plan to a different array, you can use the new-array execute Chris@82: interface. See New-array Execute Functions. Chris@82:

fftw_execute (and equivalents) is the only function in FFTW Chris@82: guaranteed to be thread-safe; see Thread safety. Chris@82:

Chris@82:

void fftw_destroy_plan(fftw_plan plan);
Chris@82:

FFTW’s planner saves some other persistent data, such as the Chris@82: accumulated wisdom and a list of algorithms available in the current Chris@82: configuration. If you want to deallocate all of that and reset FFTW Chris@82: to the pristine state it was in when you started your program, you can Chris@82: call: Chris@82:

Chris@82:

void fftw_cleanup(void);
Chris@82:

After calling fftw_cleanup, all existing plans become undefined, Chris@82: and you should not attempt to execute them nor to destroy them. You can Chris@82: however create and execute/destroy new plans, in which case FFTW starts Chris@82: accumulating wisdom information again. Chris@82:

fftw_cleanup does not deallocate your plans, however. To prevent Chris@82: memory leaks, you must still call fftw_destroy_plan before Chris@82: executing fftw_cleanup. Chris@82:

Occasionally, it may useful to know FFTW’s internal “cost” metric Chris@82: that it uses to compare plans to one another; this cost is Chris@82: proportional to an execution time of the plan, in undocumented units, Chris@82: if the plan was created with the FFTW_MEASURE or other Chris@82: timing-based options, or alternatively is a heuristic cost function Chris@82: for FFTW_ESTIMATE plans. (The cost values of measured and Chris@82: estimated plans are not comparable, being in different units. Also, Chris@82: costs from different FFTW versions or the same version compiled Chris@82: differently may not be in the same units. Plans created from wisdom Chris@82: have a cost of 0 since no timing measurement is performed for them. Chris@82: Finally, certain problems for which only one top-level algorithm was Chris@82: possible may have required no measurements of the cost of the whole Chris@82: plan, in which case fftw_cost will also return 0.) The cost Chris@82: metric for a given plan is returned by: Chris@82:

Chris@82:

double fftw_cost(const fftw_plan plan);
Chris@82:

The following two routines are provided purely for academic purposes Chris@82: (that is, for entertainment). Chris@82:

Chris@82:

void fftw_flops(const fftw_plan plan, 
Chris@82:                 double *add, double *mul, double *fma);
Chris@82:

Given a plan, set add, mul, and fma to an Chris@82: exact count of the number of floating-point additions, multiplications, Chris@82: and fused multiply-add operations involved in the plan’s execution. The Chris@82: total number of floating-point operations (flops) is

add + mul +
Chris@82: 2*fma

, or add + mul + fma if the hardware supports fused Chris@82: multiply-add instructions (although the number of FMA operations is only Chris@82: approximate because of compiler voodoo). (The number of operations Chris@82: should be an integer, but we use double to avoid overflowing Chris@82: int for large transforms; the arguments are of type double Chris@82: even for single and long-double precision versions of FFTW.) Chris@82:

Chris@82:

void fftw_fprint_plan(const fftw_plan plan, FILE *output_file);
Chris@82: void fftw_print_plan(const fftw_plan plan);
Chris@82: char *fftw_sprint_plan(const fftw_plan plan);
Chris@82:

This outputs a “nerd-readable” representation of the plan to Chris@82: the given file, to stdout, or two a newly allocated Chris@82: NUL-terminated string (which the caller is responsible for deallocating Chris@82: with free), respectively. Chris@82:

4.2 Using Plans