amrex::FFT::Info Struct Reference

#include <AMReX_FFT_Helper.H>

Public Member Functions
Info &	setDomainStrategy (DomainStrategy s)
	Select how the domain is decomposed across MPI ranks.
Info &	setPencilThreshold (int t)
	Override the slab→pencil break-even threshold for the automatic strategy.
Info &	setTwoDMode (bool x)
	Restrict transforms to the first two dimensions (3-D problems only).
Info &	setOneDMode (bool x)
	Flag the degenerate 2-D mode (nx==1 or ny==1) that still batches along z.
Info &	setBatchSize (int bsize)
	Specify the batch size for FFT.
Info &	setNumProcs (int n)
	Cap the number of MPI ranks used by FFT.

Public Attributes
DomainStrategy	domain_strategy = DomainStrategy::automatic
	Domain composition strategy.
int	pencil_threshold = 4
bool	twod_mode = false
bool	oned_mode = false
	We might have a special twod_mode: nx or ny == 1 && nz > 1.
int	batch_size = 1
	Batched FFT size. Only support in R2C, not R2X.
int	nprocs = std::numeric_limits<int>::max()
	Max number of processes to use.

Member Function Documentation

setBatchSize()

Info & amrex::FFT::Info::setBatchSize ( int  bsize )

inline

Specify the batch size for FFT.

Parameters

bsize

Number of independent transforms executed together.

Returns

Reference to this Info.

setDomainStrategy()

Info & amrex::FFT::Info::setDomainStrategy ( DomainStrategy  s )

inline

Select how the domain is decomposed across MPI ranks.

Parameters

s	Strategy (automatic|slab|pencil).

Returns

Reference to this Info.

setNumProcs()

Info & amrex::FFT::Info::setNumProcs ( int  n )

inline

Cap the number of MPI ranks used by FFT.

Parameters

n	Maximum participating ranks.

Returns

Reference to this Info.

setOneDMode()

Info & amrex::FFT::Info::setOneDMode ( bool  x )

inline

Flag the degenerate 2-D mode (nx==1 or ny==1) that still batches along z.

This keeps the solver in “two-dimensional” mode but makes it clear that only one transverse direction has more than one cell; the z-direction remains the batch axis.

Parameters

x	True when running on effectively 1-D x/y planes while batching along z.

Returns

Reference to this Info.

setPencilThreshold()

Info & amrex::FFT::Info::setPencilThreshold ( int  t )

inline

Override the slab→pencil break-even threshold for the automatic strategy.

Parameters

t	Minimum cells per rank before switching to pencil decomposition.

Returns

Reference to this Info.

setTwoDMode()

Info & amrex::FFT::Info::setTwoDMode ( bool  x )

inline

Restrict transforms to the first two dimensions (3-D problems only).

Parameters

x	True to operate on x/y planes while batching along z.

Returns

Reference to this Info.

Member Data Documentation

batch_size

int amrex::FFT::Info::batch_size = 1

Batched FFT size. Only support in R2C, not R2X.

domain_strategy

DomainStrategy amrex::FFT::Info::domain_strategy = DomainStrategy::automatic

Domain composition strategy.

nprocs

int amrex::FFT::Info::nprocs = std::numeric_limits<int>::max()

Max number of processes to use.

oned_mode

bool amrex::FFT::Info::oned_mode = false

We might have a special twod_mode: nx or ny == 1 && nz > 1.

pencil_threshold

int amrex::FFT::Info::pencil_threshold = 4

For automatic strategy, this is the size per process below which we switch from slab to pencil.

twod_mode

bool amrex::FFT::Info::twod_mode = false

Supported only in 3D. When twod_mode is true, FFT is performed on the first two dimensions only and the third dimension size is the batch size.

---

The documentation for this struct was generated from the following file:

• /home/runner/work/amrex/amrex/Src/FFT/AMReX_FFT_Helper.H