FFTW 3.3.8: Upgrading from FFTW version 2

In this chapter, we outline the process for updating codes designed for Chris@82: the older FFTW 2 interface to work with FFTW 3. The interface for FFTW Chris@82: 3 is not backwards-compatible with the interface for FFTW 2 and earlier Chris@82: versions; codes written to use those versions will fail to link with Chris@82: FFTW 3. Nor is it possible to write “compatibility wrappers” to Chris@82: bridge the gap (at least not efficiently), because FFTW 3 has different Chris@82: semantics from previous versions. However, upgrading should be a Chris@82: straightforward process because the data formats are identical and the Chris@82: overall style of planning/execution is essentially the same. Chris@82:

Unlike FFTW 2, there are no separate header files for real and complex Chris@82: transforms (or even for different precisions) in FFTW 3; all interfaces Chris@82: are defined in the <fftw3.h> header file. Chris@82:

Numeric Types

The main difference in data types is that fftw_complex in FFTW 2 Chris@82: was defined as a struct with macros c_re and c_im Chris@82: for accessing the real/imaginary parts. (This is binary-compatible with Chris@82: FFTW 3 on any machine except perhaps for some older Crays in single Chris@82: precision.) The equivalent macros for FFTW 3 are: Chris@82:

Chris@82:

#define c_re(c) ((c)[0])
Chris@82: #define c_im(c) ((c)[1])
Chris@82:

This does not work if you are using the C99 complex type, however, Chris@82: unless you insert a double* typecast into the above macros Chris@82: (see Complex numbers). Chris@82:

Also, FFTW 2 had an fftw_real typedef that was an alias for Chris@82: double (in double precision). In FFTW 3 you should just use Chris@82: double (or whatever precision you are employing). Chris@82:

Plans

The major difference between FFTW 2 and FFTW 3 is in the Chris@82: planning/execution division of labor. In FFTW 2, plans were found for a Chris@82: given transform size and type, and then could be applied to any Chris@82: arrays and for any multiplicity/stride parameters. In FFTW 3, Chris@82: you specify the particular arrays, stride parameters, etcetera when Chris@82: creating the plan, and the plan is then executed for those arrays Chris@82: (unless the guru interface is used) and those parameters Chris@82: only. (FFTW 2 had “specific planner” routines that planned for Chris@82: a particular array and stride, but the plan could still be used for Chris@82: other arrays and strides.) That is, much of the information that was Chris@82: formerly specified at execution time is now specified at planning time. Chris@82:

Like FFTW 2’s specific planner routines, the FFTW 3 planner overwrites Chris@82: the input/output arrays unless you use FFTW_ESTIMATE. Chris@82:

FFTW 2 had separate data types fftw_plan, fftwnd_plan, Chris@82: rfftw_plan, and rfftwnd_plan for complex and real one- and Chris@82: multi-dimensional transforms, and each type had its own ‘destroy’ Chris@82: function. In FFTW 3, all plans are of type fftw_plan and all are Chris@82: destroyed by fftw_destroy_plan(plan). Chris@82:

Where you formerly used fftw_create_plan and fftw_one to Chris@82: plan and compute a single 1d transform, you would now use Chris@82: fftw_plan_dft_1d to plan the transform. If you used the generic Chris@82: fftw function to execute the transform with multiplicity Chris@82: (howmany) and stride parameters, you would now use the advanced Chris@82: interface fftw_plan_many_dft to specify those parameters. The Chris@82: plans are now executed with fftw_execute(plan), which takes all Chris@82: of its parameters (including the input/output arrays) from the plan. Chris@82:

In-place transforms no longer interpret their output argument as scratch Chris@82: space, nor is there an FFTW_IN_PLACE flag. You simply pass the Chris@82: same pointer for both the input and output arguments. (Previously, the Chris@82: output ostride and odist parameters were ignored for Chris@82: in-place transforms; now, if they are specified via the advanced Chris@82: interface, they are significant even in the in-place case, although they Chris@82: should normally equal the corresponding input parameters.) Chris@82:

The FFTW_ESTIMATE and FFTW_MEASURE flags have the same Chris@82: meaning as before, although the planning time will differ. You may also Chris@82: consider using FFTW_PATIENT, which is like FFTW_MEASURE Chris@82: except that it takes more time in order to consider a wider variety of Chris@82: algorithms. Chris@82:

For multi-dimensional complex DFTs, instead of fftwnd_create_plan Chris@82: (or fftw2d_create_plan or fftw3d_create_plan), followed by Chris@82: fftwnd_one, you would use fftw_plan_dft (or Chris@82: fftw_plan_dft_2d or fftw_plan_dft_3d). followed by Chris@82: fftw_execute. If you used fftwnd to to specify strides Chris@82: etcetera, you would instead specify these via fftw_plan_many_dft. Chris@82:

The analogues to rfftw_create_plan and rfftw_one with Chris@82: FFTW_REAL_TO_COMPLEX or FFTW_COMPLEX_TO_REAL directions Chris@82: are fftw_plan_r2r_1d with kind FFTW_R2HC or Chris@82: FFTW_HC2R, followed by fftw_execute. The stride etcetera Chris@82: arguments of rfftw are now in fftw_plan_many_r2r. Chris@82:

Instead of rfftwnd_create_plan (or rfftw2d_create_plan or Chris@82: rfftw3d_create_plan) followed by Chris@82: rfftwnd_one_real_to_complex or Chris@82: rfftwnd_one_complex_to_real, you now use fftw_plan_dft_r2c Chris@82: (or fftw_plan_dft_r2c_2d or fftw_plan_dft_r2c_3d) or Chris@82: fftw_plan_dft_c2r (or fftw_plan_dft_c2r_2d or Chris@82: fftw_plan_dft_c2r_3d), respectively, followed by Chris@82: fftw_execute. As usual, the strides etcetera of Chris@82: rfftwnd_real_to_complex or rfftwnd_complex_to_real are no Chris@82: specified in the advanced planner routines, Chris@82: fftw_plan_many_dft_r2c or fftw_plan_many_dft_c2r. Chris@82:

Wisdom

In FFTW 2, you had to supply the FFTW_USE_WISDOM flag in order to Chris@82: use wisdom; in FFTW 3, wisdom is always used. (You could simulate the Chris@82: FFTW 2 wisdom-less behavior by calling fftw_forget_wisdom after Chris@82: every planner call.) Chris@82:

The FFTW 3 wisdom import/export routines are almost the same as before Chris@82: (although the storage format is entirely different). There is one Chris@82: significant difference, however. In FFTW 2, the import routines would Chris@82: never read past the end of the wisdom, so you could store extra data Chris@82: beyond the wisdom in the same file, for example. In FFTW 3, the Chris@82: file-import routine may read up to a few hundred bytes past the end of Chris@82: the wisdom, so you cannot store other data just beyond it.¹¹ Chris@82:

Wisdom has been enhanced by additional humility in FFTW 3: whereas FFTW Chris@82: 2 would re-use wisdom for a given transform size regardless of the Chris@82: stride etc., in FFTW 3 wisdom is only used with the strides etc. for Chris@82: which it was created. Unfortunately, this means FFTW 3 has to create Chris@82: new plans from scratch more often than FFTW 2 (in FFTW 2, planning Chris@82: e.g. one transform of size 1024 also created wisdom for all smaller Chris@82: powers of 2, but this no longer occurs). Chris@82:

FFTW 3 also has the new routine fftw_import_system_wisdom to Chris@82: import wisdom from a standard system-wide location. Chris@82:

Memory allocation

In FFTW 3, we recommend allocating your arrays with fftw_malloc Chris@82: and deallocating them with fftw_free; this is not required, but Chris@82: allows optimal performance when SIMD acceleration is used. (Those two Chris@82: functions actually existed in FFTW 2, and worked the same way, but were Chris@82: not documented.) Chris@82:

In FFTW 2, there were fftw_malloc_hook and fftw_free_hook Chris@82: functions that allowed the user to replace FFTW’s memory-allocation Chris@82: routines (e.g. to implement different error-handling, since by default Chris@82: FFTW prints an error message and calls exit to abort the program Chris@82: if malloc returns NULL). These hooks are not supported in Chris@82: FFTW 3; those few users who require this functionality can just Chris@82: directly modify the memory-allocation routines in FFTW (they are defined Chris@82: in kernel/alloc.c). Chris@82:

Fortran interface

In FFTW 2, the subroutine names were obtained by replacing ‘fftw_’ Chris@82: with ‘fftw_f77’; in FFTW 3, you replace ‘fftw_’ with Chris@82: ‘dfftw_’ (or ‘sfftw_’ or ‘lfftw_’, depending upon the Chris@82: precision). Chris@82:

In FFTW 3, we have begun recommending that you always declare the type Chris@82: used to store plans as integer*8. (Too many people didn’t notice Chris@82: our instruction to switch from integer to integer*8 for Chris@82: 64-bit machines.) Chris@82:

In FFTW 3, we provide a fftw3.f “header file” to include in Chris@82: your code (and which is officially installed on Unix systems). (In FFTW Chris@82: 2, we supplied a fftw_f77.i file, but it was not installed.) Chris@82:

Otherwise, the C-Fortran interface relationship is much the same as it Chris@82: was before (e.g. return values become initial parameters, and Chris@82: multi-dimensional arrays are in column-major order). Unlike FFTW 2, we Chris@82: do provide some support for wisdom import/export in Fortran Chris@82: (see Wisdom of Fortran?). Chris@82:

Threads

Like FFTW 2, only the execution routines are thread-safe. All planner Chris@82: routines, etcetera, should be called by only a single thread at a time Chris@82: (see Thread safety). Unlike FFTW 2, there is no special Chris@82: FFTW_THREADSAFE flag for the planner to allow a given plan to be Chris@82: usable by multiple threads in parallel; this is now the case by default. Chris@82:

The multi-threaded version of FFTW 2 required you to pass the number of Chris@82: threads each time you execute the transform. The number of threads is Chris@82: now stored in the plan, and is specified before the planner is called by Chris@82: fftw_plan_with_nthreads. The threads initialization routine used Chris@82: to be called fftw_threads_init and would return zero on success; Chris@82: the new routine is called fftw_init_threads and returns zero on Chris@82: failure. See Multi-threaded FFTW. Chris@82:

There is no separate threads header file in FFTW 3; all the function Chris@82: prototypes are in <fftw3.h>. However, you still have to link to Chris@82: a separate library (-lfftw3_threads -lfftw3 -lm on Unix), as well as Chris@82: to the threading library (e.g. POSIX threads on Unix). Chris@82:

Chris@82:

Footnotes

Chris@82: Chris@82:

(11)

Chris@82:

We Chris@82: do our own buffering because GNU libc I/O routines are horribly slow for Chris@82: single-character I/O, apparently for thread-safety reasons (whether you Chris@82: are using threads or not).

Chris@82:

9 Upgrading from FFTW version 2

Numeric Types

Plans

Wisdom

Memory allocation

Fortran interface

Threads

Footnotes

(11)