Python API Reference

This page provides the generated Python API reference for dtfft with explicit per-object documentation.

Exceptions

dtfft.dtfft_Exception

alias of Exception

Enumerations and Constants

class dtfft.Version

dtFFT version information

CODE = 301005

MAJOR = 3

MINOR = 1

PATCH = 5

static get(*args, **kwargs)

Overloaded function.

1. get() -> int

Get Version Code defined during compilation

2. get(major: typing.SupportsInt | typing.SupportsIndex, minor: typing.SupportsInt | typing.SupportsIndex, patch: typing.SupportsInt | typing.SupportsIndex) -> int

Get Version Code based on input parameters

enum dtfft.Execute(value)

This class lists valid execute_type parameters that can be passed to Plan.execute.

Valid values are as follows:

FORWARD = <Execute.FORWARD: 11>

Perform XYZ –> YZX –> ZXY plan execution (Forward)

BACKWARD = <Execute.BACKWARD: 12>

Perform ZXY –> YZX –> XYZ plan execution (Backward)

enum dtfft.Transpose(value)

This class lists valid transpose_type parameters that can be passed to Plan.transpose.

Valid values are as follows:

X_TO_Y = <Transpose.X_TO_Y: 1>

Transpose from Fortran X aligned to Fortran Y aligned

Y_TO_X = <Transpose.Y_TO_X: -1>

Transpose from Fortran Y aligned to Fortran X aligned

Y_TO_Z = <Transpose.Y_TO_Z: 2>

Transpose from Fortran Y aligned to Fortran Z aligned

Z_TO_Y = <Transpose.Z_TO_Y: -2>

Transpose from Fortran Z aligned to Fortran Y aligned

X_TO_Z = <Transpose.X_TO_Z: 3>

Transpose from Fortran X aligned to Fortran Z aligned. Note: This value is valid to pass only in 3D Plan and value returned by Plan.get_z_slab_enabled must be true

Z_TO_X = <Transpose.Z_TO_X: -3>

Transpose from Fortran Z aligned to Fortran X aligned. Note: This value is valid to pass only in 3D Plan and value returned by Plan.get_z_slab_enabled must be true

enum dtfft.Reshape(value)

This class lists valid reshape types that can be passed to Plan.reshape.

Valid values are as follows:

X_BRICKS_TO_PENCILS = <Reshape.X_BRICKS_TO_PENCILS: 11>

Reshape from X bricks to X pencils

X_PENCILS_TO_BRICKS = <Reshape.X_PENCILS_TO_BRICKS: 12>

Reshape from X pencils to X bricks

Z_BRICKS_TO_PENCILS = <Reshape.Z_BRICKS_TO_PENCILS: 14>

Reshape from Y bricks to Y pencils

Z_PENCILS_TO_BRICKS = <Reshape.Z_PENCILS_TO_BRICKS: 13>

Reshape from Y pencils to Y bricks

enum dtfft.Layout(value)

This class lists valid layout parameters that can be passed to Plan.get_pencil.

Valid values are as follows:

X_BRICKS = <Layout.X_BRICKS: 1>

X-brick layout

X_PENCILS = <Layout.X_PENCILS: 2>

X-pencil layout

X_PENCILS_FOURIER = <Layout.X_PENCILS_FOURIER: 3>

X-pencil Fourier layout for R2C plans

Y_PENCILS = <Layout.Y_PENCILS: 4>

Y-pencil layout

Z_PENCILS = <Layout.Z_PENCILS: 5>

Z-pencil layout

Z_BRICKS = <Layout.Z_BRICKS: 6>

Z-brick layout

enum dtfft.Precision(value)

This class lists valid precision parameters that can be passed to Plan constructors.

Valid values are as follows:

SINGLE = <Precision.SINGLE: 0>

Use Single precision

DOUBLE = <Precision.DOUBLE: 1>

Use Double precision

enum dtfft.Effort(value)

This class lists valid effort parameters that can be passed to Plan constructors.

Valid values are as follows:

ESTIMATE = <Effort.ESTIMATE: 0>

Create plan as fast as possible

MEASURE = <Effort.MEASURE: 1>

Will attempt to find best MPI Grid decomposition. Passing this flag and MPI Communicator with cartesian topology to any Plan Constructor is same as Effort::ESTIMATE.

PATIENT = <Effort.PATIENT: 2>

Same as Effort::MEASURE plus cycle through various send and receive MPI_Datatypes. For GPU Build this flag will run autotune procedure to find best backend

EXHAUSTIVE = <Effort.EXHAUSTIVE: 3>

Same as Effort::PATIENT plus autotune all possible kernels and reshape backends.

enum dtfft.Executor(value)

This class lists available FFT executors.

Valid values are as follows:

NONE = <Executor.NONE: 0>

Do not create any FFT plans. Creates transpose only plan.

FFTW3 = <Executor.FFTW3: 1>

FFTW3 Executor (Host only)

MKL = <Executor.MKL: 2>

MKL DFTI Executor (Host only)

CUFFT = <Executor.CUFFT: 3>

CUFFT Executor (GPU Only)

VKFFT = <Executor.VKFFT: 4>

VkFFT Executor (GPU Only)

enum dtfft.R2RKind(value)

Real-to-Real FFT kinds available in dtFFT

Valid values are as follows:

DCT_1 = <R2RKind.DCT_1: 3>

DCT-I (Logical N=2*(n-1), inverse is R2RKind::DCT_1)

DCT_2 = <R2RKind.DCT_2: 5>

DCT-II (Logical N=2*n, inverse is R2RKind::DCT_3)

DCT_3 = <R2RKind.DCT_3: 4>

DCT-III (Logical N=2*n, inverse is R2RKind::DCT_2)

DCT_4 = <R2RKind.DCT_4: 6>

DCT-IV (Logical N=2*n, inverse is R2RKind::DCT_4)

DST_1 = <R2RKind.DST_1: 7>

DST-I (Logical N=2*(n+1), inverse is R2RKind::DST_1)

DST_2 = <R2RKind.DST_2: 9>

DST-II (Logical N=2*n, inverse is R2RKind::DST_3)

DST_3 = <R2RKind.DST_3: 8>

DST-III (Logical N=2*n, inverse is R2RKind::DST_2)

DST_4 = <R2RKind.DST_4: 10>

DST-IV (Logical N=2*n, inverse is R2RKind::DST_4)

enum dtfft.TransposeMode(value)

This class lists possible local transposition modes for generic backends.

Valid values are as follows:

PACK = <TransposeMode.PACK: 15>

Perform transposition during packing

UNPACK = <TransposeMode.UNPACK: 16>

Perform transposition during unpacking

enum dtfft.AccessMode(value)

This class lists memory access modes for local transposition in generic backends.

Valid values are as follows:

WRITE = <AccessMode.WRITE: -1>

Write-aligned access

READ = <AccessMode.READ: 1>

Read-aligned access

enum dtfft.Backend(value)

Various backends available in dtFFT.

Valid values are as follows:

MPI_DATATYPE = <Backend.MPI_DATATYPE: 21>

Backend that uses MPI datatypes.

MPI_P2P = <Backend.MPI_P2P: 22>

MPI peer-to-peer algorithm

MPI_P2P_PIPELINED = <Backend.MPI_P2P_PIPELINED: 26>

MPI peer-to-peer algorithm with overlapping data copying and unpacking

MPI_A2A = <Backend.MPI_A2A: 23>

MPI backend using MPI_Alltoall[v]

MPI_RMA = <Backend.MPI_RMA: 29>

MPI backend using one-sided communications

MPI_RMA_PIPELINED = <Backend.MPI_RMA_PIPELINED: 30>

MPI backend using pipelined one-sided communications

MPI_P2P_SCHEDULED = <Backend.MPI_P2P_SCHEDULED: 31>

MPI peer-to-peer algorithm with scheduled communication

MPI_P2P_FUSED = <Backend.MPI_P2P_FUSED: 32>

MPI peer-to-peer pipelined algorithm with overlapping packing, exchange and unpacking with scheduled communication

MPI_RMA_FUSED = <Backend.MPI_RMA_FUSED: 33>

MPI RMA pipelined algorithm with overlapping packing, exchange and unpacking with scheduled communication

MPI_P2P_COMPRESSED = <Backend.MPI_P2P_COMPRESSED: 34>

MPI peer-to-peer fused backend with compression

MPI_RMA_COMPRESSED = <Backend.MPI_RMA_COMPRESSED: 35>

MPI RMA fused backend with compression

NCCL = <Backend.NCCL: 24>

NCCL backend

NCCL_PIPELINED = <Backend.NCCL_PIPELINED: 27>

NCCL backend with overlapping data copying and unpacking

NCCL_COMPRESSED = <Backend.NCCL_COMPRESSED: 37>

NCCL backend with compression

CUFFTMP = <Backend.CUFFTMP: 25>

cuFFTMp backend

CUFFTMP_PIPELINED = <Backend.CUFFTMP_PIPELINED: 28>

cuFFTMp backend with an additional internal buffer

ADAPTIVE = <Backend.ADAPTIVE: 36>

Adaptive backend selection

NONE = <Backend.NONE: -111>

Backend is not defined

enum dtfft.Platform(value)

Runtime platform enum used by dtFFT plans.

Valid values are as follows:

HOST = <Platform.HOST: 0>

CUDA = <Platform.CUDA: 1>

enum dtfft.CompressionMode(value)

Compression mode.

Valid values are as follows:

LOSSLESS = <CompressionMode.LOSSLESS: 1>

Lossless compression mode.

FIXED_RATE = <CompressionMode.FIXED_RATE: 2>

Fixed rate compression mode.

FIXED_PRECISION = <CompressionMode.FIXED_PRECISION: 3>

Fixed precision compression mode.

FIXED_ACCURACY = <CompressionMode.FIXED_ACCURACY: 4>

Fixed accuracy compression mode.

enum dtfft.CompressionLib(value)

Compression backend library.

Valid values are as follows:

ZFP = <CompressionLib.ZFP: 1>

Use ZFP compression library.

Functions

dtfft.get_backend_string(backend: Backend) → str

Return string representation of a Backend value.

dtfft.is_fftw_enabled() → bool

Return True if dtFFT was built with FFTW3 support.

dtfft.is_mkl_enabled() → bool

Return True if dtFFT was built with Intel MKL support.

dtfft.is_cufft_enabled() → bool

Return True if dtFFT was built with cuFFT support.

dtfft.is_vkfft_enabled() → bool

Return True if dtFFT was built with VkFFT support.

dtfft.is_cuda_enabled() → bool

Return True if CUDA platform support is enabled in dtFFT.

dtfft.is_transpose_only_enabled() → bool

Return True if dtFFT was built in transpose-only mode.

dtfft.is_nccl_enabled() → bool

Return True if NCCL support is enabled in dtFFT.

dtfft.is_nvshmem_enabled() → bool

Return True if NVSHMEM support is enabled in dtFFT.

dtfft.is_compression_enabled() → bool

Return True if compression backends are enabled in dtFFT.

Core Classes

class dtfft.Config(*, enable_log: bool | None = None, enable_z_slab: bool | None = None, enable_y_slab: bool | None = None, measure_warmup_iters: int | None = None, measure_iters: int | None = None, backend: Backend | None = None, reshape_backend: Backend | None = None, enable_datatype_backend: bool | None = None, enable_mpi_backends: bool | None = None, enable_pipelined_backends: bool | None = None, enable_rma_backends: bool | None = None, enable_fused_backends: bool | None = None, enable_kernel_autotune: bool | None = None, enable_fourier_reshape: bool | None = None, transpose_mode: TransposeMode | None = None, access_mode: AccessMode | None = None, platform: Platform | None = None, stream: cp.cuda.Stream | None = None, enable_nccl_backends: bool | None = None, enable_nvshmem_backends: bool | None = None, enable_compressed_backends: bool | None = None, compression_config_transpose: CompressionConfig | None = None, compression_config_reshape: CompressionConfig | None = None)

Configuration object applied before plan creation.

Parameters:

• enable_log (bool, optional) – Whether dtFFT should print additional information (default: False).

• enable_z_slab (bool, optional) – Whether dtFFT uses Z-slab optimisation (default: True).

• enable_y_slab (bool, optional) – Whether dtFFT uses Y-slab optimisation (default: False).

• measure_warmup_iters (int, optional) – Number of warmup iterations used during autotuning.

• measure_iters (int, optional) – Number of measurement iterations used during autotuning.

• backend (Backend, optional) – Backend used when effort is ESTIMATE or MEASURE.

• reshape_backend (Backend, optional) – Backend used for reshape operations when effort is ESTIMATE or MEASURE.

• enable_datatype_backend (bool, optional) – Whether MPI datatype backend is enabled for autotuning (default: True).

• enable_mpi_backends (bool, optional) – Whether MPI backends are enabled for PATIENT effort (default: False).

• enable_pipelined_backends (bool, optional) – Whether pipelined backends are enabled for PATIENT effort (default: True).

• enable_rma_backends (bool, optional) – Whether RMA backends are enabled for PATIENT effort (default: True).

• enable_fused_backends (bool, optional) – Whether fused backends are enabled for PATIENT effort (default: True).

• enable_kernel_autotune (bool, optional) – Whether kernel launch parameter autotuning is enabled below EXHAUSTIVE (default: False).

• enable_fourier_reshape (bool, optional) – Whether Fourier-space reshapes are executed during Plan.execute (default: False).

• transpose_mode (TransposeMode, optional) – At which stage local transposition is performed for generic backends.

• access_mode (AccessMode, optional) – Memory access mode for local transposition in generic backends.

• platform (Platform, optional) – Execution platform (default: Platform.HOST). CUDA builds only.

• stream (cp.cuda.Stream, optional) – CUDA stream to use for all operations. If not specified, dtFFT will create stream internally. Such stream can be retrieved via the stream property of the plan after creation. CUDA builds only.

• enable_nccl_backends (bool, optional) – Whether NCCL backends are enabled for PATIENT effort (default: True). CUDA builds only.

• enable_nvshmem_backends (bool, optional) – Whether NVSHMEM backends are enabled for PATIENT effort (default: True). CUDA builds only.

• enable_compressed_backends (bool, optional) – Whether compressed backends are enabled for autotuning. Compression builds only.

• compression_config_transpose (CompressionConfig, optional) – Compression configuration for transpose operations. Compression builds only.

• compression_config_reshape (CompressionConfig, optional) – Compression configuration for reshape operations. Compression builds only.

Raises:

AttributeError – If an option is not available in the current build.

Notes

Values are passed to dtFFT via set_config() right before plan construction when a config is provided to a plan constructor.

class dtfft.Pencil(starts: list[int] = None, counts: list[int] = None, pencil: Pencil = None)

Create a Pencil decomposition object that can be passed to plan constructors.

Parameters:

• starts (list[int], optional) – Starting indices of the local data block in each dimension.

• counts (list[int], optional) – Number of elements in the local data block in each dimension.

• pencil (_dtfft.Pencil, optional) – An existing _dtfft.Pencil object to wrap. If provided, starts and counts are ignored. This parameter is intended for internal use when a Pencil object is returned by a plan method, and allows wrapping an existing C++ Pencil without copying. For user-created Pencil objects, it is more common to specify starts and counts directly.

classmethod from_pencil(pencil: Pencil)

Creates Pencil wrapper from an existing _dtfft.Pencil instance. This is used internally when a plan method returns a Pencil object.

class dtfft.Request(handle: int, kind: str)

Python wrapper for internal async request handle (dtfft_request_t).

class dtfft.CompressionConfig(compression_lib: CompressionLib = CompressionLib.ZFP, compression_mode: CompressionMode = CompressionMode.LOSSLESS, rate: float = 0.0, precision: int = 0, tolerance: float = 0.0)

Compression configuration for transpose/reshape paths.

Parameters:

• compression_lib (CompressionLib, optional) – Compression library to use. Default is CompressionLib.ZFP.

• compression_mode (CompressionMode, optional) – Compression mode to use. Default is CompressionMode.LOSSLESS.

• rate (float, optional) – Target rate for fixed rate mode. Ignored for other modes. Default is 0.0.

• precision (int, optional) – Target precision for fixed precision mode. Ignored for other modes. Default is 0.0.

• tolerance (float, optional) – Target tolerance for fixed accuracy mode. Ignored for other modes. Default is 0.0.

Plan Classes

class dtfft.Plan(plan: Plan)

Python wrapper around a concrete dtFFT plan instance.

This class delegates most of its informational properties to the underlying C++ plan. The following properties are dynamically available via __getattr__:

Properties

z_slab_enabledbool

Return whether Z-slab optimization is enabled for this plan.

y_slab_enabledbool

Return whether Y-slab optimization is enabled for this plan.

alloc_sizeint

Return minimum number of elements required for main buffers.

aux_sizeint

Return auxiliary buffer size (in elements) for execute.

aux_bytesint

Return auxiliary buffer size (in bytes) for execute.

aux_size_reshapeint

Return auxiliary buffer size (in elements) for reshape.

aux_bytes_reshapeint

Return auxiliary buffer size (in bytes) for reshape.

aux_size_transposeint

Return auxiliary buffer size (in elements) for transpose.

aux_bytes_transposeint

Return auxiliary buffer size (in bytes) for transpose.

local_sizestuple[list[int], list[int], list[int], list[int], int]

Return local decomposition: (in_starts, in_counts, out_starts, out_counts, alloc_size).

dimslist[int]

Return global plan dimensions in natural Fortran order.

grid_dimslist[int]

Return MPI grid dimensions in natural Fortran order.

element_sizeint

Return size (in bytes) of one data element for this plan.

alloc_bytesint

Return minimum bytes required for input/output buffers.

executorExecutor

Return executor selected for this plan.

precisionPrecision

Return precision selected for this plan.

backendBackend

Return communication backend selected for this plan.

reshape_backendBackend

Return backend selected for reshape operations.

streamcupy.cuda.Stream | cupy.cuda.ExternalStream

Return CUDA stream associated with this plan.

platformPlatform

Return execution platform (HOST/CUDA) for this plan.

execute(input: np.ndarray | cp.ndarray, output: np.ndarray | cp.ndarray, execute_type: Execute, aux: np.ndarray | cp.ndarray = None) → None

Execute full plan in FORWARD or BACKWARD direction.

Parameters:

• input (np.ndarray | cp.ndarray) – Input array containing the source data (e.g., numpy.ndarray or cupy.ndarray).

• output (np.ndarray | cp.ndarray) – Output array where the result will be written.

• execute_type (Execute) – Direction of execution (e.g., Execute.FORWARD or Execute.BACKWARD).

• aux (np.ndarray | cp.ndarray, optional) – Optional auxiliary array for intermediate data.

transpose(input: np.ndarray | cp.ndarray, output: np.ndarray | cp.ndarray, transpose_type: Transpose, aux: np.ndarray | cp.ndarray = None) → None

Transpose data in a single dimension (e.g., X->Y, Y->Z).

Parameters:

• input (np.ndarray | cp.ndarray) – Input array containing the source data.

• output (np.ndarray | cp.ndarray) – Output array where transposed data will be written. Must be different from input.

• transpose_type (Transpose) – Type of transpose operation to perform.

• aux (np.ndarray | cp.ndarray, optional) – Optional auxiliary array for intermediate data.

reshape(input: np.ndarray | cp.ndarray, output: np.ndarray | cp.ndarray, reshape_type: Reshape, aux: np.ndarray | cp.ndarray = None) → None

Reshape data between brick and pencil layouts and vice versa.

Parameters:

• input (np.ndarray | cp.ndarray) – Input array containing the source data.

• output (np.ndarray | cp.ndarray) – Output array where reshaped data will be written. Must be different from input.

• reshape_type (Reshape) – Type of reshape operation to perform.

• aux (np.ndarray | cp.ndarray, optional) – Optional auxiliary array for intermediate data.

transpose_start(input: np.ndarray | cp.ndarray, output: np.ndarray | cp.ndarray, transpose_type: Transpose, aux: np.ndarray | cp.ndarray = None) → Request

Start asynchronous transpose and return request handle.

Parameters:

• input (np.ndarray | cp.ndarray) – Input array containing the source data.

• output (np.ndarray | cp.ndarray) – Output array where transposed data will be written. Must be different from input.

• transpose_type (Transpose) – Type of transpose operation to perform.

• aux (np.ndarray | cp.ndarray, optional) – Optional auxiliary array for intermediate data.

Returns:

A handle to the asynchronous operation, which must be passed to transpose_end.

Return type:

Request

transpose_end(request: Request) → None

Finalize asynchronous transpose operation.

Parameters:

request (Request) – The operation handle previously returned by transpose_start.

reshape_start(input: np.ndarray | cp.ndarray, output: np.ndarray | cp.ndarray, reshape_type: Reshape, aux: np.ndarray | cp.ndarray = None) → Request

Start asynchronous reshape and return request handle.

Parameters:

• input (np.ndarray | cp.ndarray) – Input array containing the source data.

• output (np.ndarray | cp.ndarray) – Output array where reshaped data will be written. Must be different from input.

• reshape_type (Reshape) – Type of reshape operation to perform.

• aux (np.ndarray | cp.ndarray, optional) – Optional auxiliary array for intermediate data.

Returns:

A handle to the asynchronous operation, which must be passed to reshape_end.

Return type:

Request

reshape_end(request: Request) → None

Finalize asynchronous reshape operation.

Parameters:

request (Request) – The operation handle previously returned by reshape_start.

report() → None

Print plan-related information to stdout.

report_compression() → None

Print compression-related information to stdout (if available).

get_pencil(layout: Layout) → Pencil

Return Pencil metadata for the requested Layout.

destroy() → None

Destroy underlying dtFFT plan resources explicitly.

get_ndarray(size: int, shape: int | tuple[int, ...] | None = None, dtype: Any = None, order: str = 'C') → np.ndarray | cp.ndarray

Allocate an ndarray backed by plan-managed memory.

Parameters:

• size (int) – Number of elements to allocate.

• shape (int | tuple[int, ...] | None, optional) – Desired array shape. Note, that product of shape might be smaller then size.

• dtype (numpy.dtype | type | None, optional) – Element type. If None, uses plan-native dtype.

• order (str, optional) – Memory order passed to NumPy/CuPy ndarray constructor.

Returns:

Array view that uses dtFFT allocation and is automatically freed when no longer referenced.

Return type:

numpy.ndarray or cupy.ndarray

property dtype: dtype

class dtfft.PlanC2C(dims_or_pencil: list[int] | ~dtfft.Pencil, comm: ~mpi4py.MPI.Comm = <mpi4py.MPI.Intracomm object>, precision: ~dtfft._dtfft.Precision = Precision.DOUBLE, effort: ~dtfft._dtfft.Effort = Effort.ESTIMATE, executor: ~dtfft._dtfft.Executor = Executor.NONE, config: ~dtfft.Config | None = None)

Bases: Plan

Create a complex-to-complex dtFFT plan.

Parameters:

• dims_or_pencil (list[int] | Pencil) – Global dimensions (natural Fortran order) or already prepared input Pencil.

• comm (mpi4py.MPI.Comm | int, optional) – MPI communicator object or raw handle.

• precision (Precision, optional) – Scalar precision used by the plan.

• effort (Effort, optional) – Planning effort level.

• executor (Executor, optional) – Preferred FFT executor.

• config (Config | None, optional) – Optional configuration applied before plan creation.

class dtfft.PlanR2C(dims_or_pencil: list[int] | ~dtfft.Pencil, comm: ~mpi4py.MPI.Comm = <mpi4py.MPI.Intracomm object>, precision: ~dtfft._dtfft.Precision = Precision.DOUBLE, effort: ~dtfft._dtfft.Effort = Effort.ESTIMATE, executor: ~dtfft._dtfft.Executor = Executor.NONE, config: ~dtfft.Config | None = None)

Bases: Plan

Create a real-to-complex dtFFT plan.

Parameters:

• dims_or_pencil (list[int] | Pencil) – Global dimensions (natural Fortran order) or already prepared input Pencil.

• executor (Executor) – FFT executor for local real/complex transforms.

• comm (mpi4py.MPI.Comm | int, optional) – MPI communicator object or raw handle.

• precision (Precision, optional) – Scalar precision used by the plan.

• effort (Effort, optional) – Planning effort level.

• config (Config | None, optional) – Optional configuration applied before plan creation.

class dtfft.PlanR2R(dims_or_pencil: list[int] | ~dtfft.Pencil, kinds: list[~dtfft._dtfft.R2RKind] | None = None, comm: ~mpi4py.MPI.Comm = <mpi4py.MPI.Intracomm object>, precision: ~dtfft._dtfft.Precision = Precision.DOUBLE, effort: ~dtfft._dtfft.Effort = Effort.ESTIMATE, executor: ~dtfft._dtfft.Executor = Executor.NONE, config: ~dtfft.Config | None = None)

Bases: Plan

Create a real-to-real dtFFT plan.

Parameters:

• dims_or_pencil (list[int] | Pencil) – Global dimensions (natural Fortran order) or already prepared input Pencil.

• kinds (list[R2RKind] | None, optional) – Transform kinds for each transformed axis.

• comm (mpi4py.MPI.Comm | int, optional) – MPI communicator object or raw handle.

• precision (Precision, optional) – Scalar precision used by the plan.

• effort (Effort, optional) – Planning effort level.

• executor (Executor, optional) – Preferred FFT executor.

• config (Config | None, optional) – Optional configuration applied before plan creation.