Plan execution in Fortran

Chris@19: Chris@19: Chris@19: Plan execution in Fortran - FFTW 3.3.4 Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19: Chris@19:

Chris@19: Chris@19:

Chris@19: Next: Allocating aligned memory in Fortran, Chris@19: Previous: FFTW Fortran type reference, Chris@19: Up: Calling FFTW from Modern Fortran Chris@19:

Chris@19:

Chris@19: Chris@19:

7.4 Plan execution in Fortran

Chris@19: Chris@19:

In C, in order to use a plan, one normally calls fftw_execute, Chris@19: which executes the plan to perform the transform on the input/output Chris@19: arrays passed when the plan was created (see Using Plans). The Chris@19: corresponding subroutine call in modern Fortran is: Chris@19:

      call fftw_execute(plan)
Chris@19:

Chris@19:

Chris@19: However, we have had reports that this causes problems with some Chris@19: recent optimizing Fortran compilers. The problem is, because the Chris@19: input/output arrays are not passed as explicit arguments to Chris@19: fftw_execute, the semantics of Fortran (unlike C) allow the Chris@19: compiler to assume that the input/output arrays are not changed by Chris@19: fftw_execute. As a consequence, certain compilers end up Chris@19: repositioning the call to fftw_execute, assuming incorrectly Chris@19: that it does nothing to the arrays. Chris@19: Chris@19:

There are various workarounds to this, but the safest and simplest Chris@19: thing is to not use fftw_execute in Fortran. Instead, use the Chris@19: functions described in New-array Execute Functions, which take Chris@19: the input/output arrays as explicit arguments. For example, if the Chris@19: plan is for a complex-data DFT and was created for the arrays Chris@19: in and out, you would do: Chris@19:

      call fftw_execute_dft(plan, in, out)
Chris@19:

Chris@19:

Chris@19: There are a few things to be careful of, however: Chris@19: Chris@19:

You must use the correct type of execute function, matching the way Chris@19: the plan was created. Complex DFT plans should use Chris@19: fftw_execute_dft, Real-input (r2c) DFT plans should use use Chris@19: fftw_execute_dft_r2c, and real-output (c2r) DFT plans should Chris@19: use fftw_execute_dft_c2r. The various r2r plans should use Chris@19: fftw_execute_r2r. Fortunately, if you use the wrong one you Chris@19: will get a compile-time type-mismatch error (unlike legacy Fortran). Chris@19: Chris@19:
You should normally pass the same input/output arrays that were used when Chris@19: creating the plan. This is always safe. Chris@19: Chris@19:
If you pass different input/output arrays compared to Chris@19: those used when creating the plan, you must abide by all the Chris@19: restrictions of the new-array execute functions (see New-array Execute Functions). The most tricky of these is the Chris@19: requirement that the new arrays have the same alignment as the Chris@19: original arrays; the best (and possibly only) way to guarantee this Chris@19: is to use the ‘fftw_alloc’ functions to allocate your arrays (see Allocating aligned memory in Fortran). Alternatively, you can Chris@19: use the FFTW_UNALIGNED flag when creating the Chris@19: plan, in which case the plan does not depend on the alignment, but Chris@19: this may sacrifice substantial performance on architectures (like x86) Chris@19: with SIMD instructions (see SIMD alignment and fftw_malloc). Chris@19: Chris@19:

Chris@19: Chris@19: Chris@19: Chris@19: