FFTW 3.3.8: FFTW Execution in Fortran

Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: FFTW 3.3.8: FFTW Execution in Fortran Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82:

Chris@82: Chris@82:

8.3 FFTW Execution in Fortran

Chris@82: Chris@82:

In C, in order to use a plan, one normally calls fftw_execute, Chris@82: which executes the plan to perform the transform on the input/output Chris@82: arrays passed when the plan was created (see Using Plans). The Chris@82: corresponding subroutine call in legacy Fortran is: Chris@82:

Chris@82:

        call dfftw_execute(plan)
Chris@82:

Chris@82: Chris@82: Chris@82:

However, we have had reports that this causes problems with some Chris@82: recent optimizing Fortran compilers. The problem is, because the Chris@82: input/output arrays are not passed as explicit arguments to Chris@82: dfftw_execute, the semantics of Fortran (unlike C) allow the Chris@82: compiler to assume that the input/output arrays are not changed by Chris@82: dfftw_execute. As a consequence, certain compilers end up Chris@82: optimizing out or repositioning the call to dfftw_execute, Chris@82: assuming incorrectly that it does nothing. Chris@82:

Chris@82:

There are various workarounds to this, but the safest and simplest Chris@82: thing is to not use dfftw_execute in Fortran. Instead, use the Chris@82: functions described in New-array Execute Functions, which take Chris@82: the input/output arrays as explicit arguments. For example, if the Chris@82: plan is for a complex-data DFT and was created for the arrays Chris@82: in and out, you would do: Chris@82:

Chris@82:

        call dfftw_execute_dft(plan, in, out)
Chris@82:

Chris@82: Chris@82: Chris@82:

There are a few things to be careful of, however: Chris@82:

Chris@82:

You must use the correct type of execute function, matching the way Chris@82: the plan was created. Complex DFT plans should use Chris@82: dfftw_execute_dft, Real-input (r2c) DFT plans should use use Chris@82: dfftw_execute_dft_r2c, and real-output (c2r) DFT plans should Chris@82: use dfftw_execute_dft_c2r. The various r2r plans should use Chris@82: dfftw_execute_r2r. Chris@82: Chris@82:
You should normally pass the same input/output arrays that were used when Chris@82: creating the plan. This is always safe. Chris@82: Chris@82:
If you pass different input/output arrays compared to Chris@82: those used when creating the plan, you must abide by all the Chris@82: restrictions of the new-array execute functions (see New-array Execute Functions). The most difficult of these, in Fortran, is the Chris@82: requirement that the new arrays have the same alignment as the Chris@82: original arrays, because there seems to be no way in legacy Fortran to obtain Chris@82: guaranteed-aligned arrays (analogous to fftw_malloc in C). You Chris@82: can, of course, use the FFTW_UNALIGNED flag when creating the Chris@82: plan, in which case the plan does not depend on the alignment, but Chris@82: this may sacrifice substantial performance on architectures (like x86) Chris@82: with SIMD instructions (see SIMD alignment and fftw_malloc). Chris@82: Chris@82: Chris@82:

Chris@82: Chris@82:

Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: Chris@82: