Parser#

configuration bbfft::parse_fft_descriptor(std::string_view desc)#

Parses an FFT descriptor.

See user manual for a description of the input format

Parameters:

desc – descriptor

Returns:

configuration

device_info bbfft::parse_device_info(std::string_view desc)#

Parses a device info descriptor.

See user manual for a description of the input format

Parameters:

desc – descriptor

Returns:

device info

Tensor indexer#

template<typename IdxT, unsigned int D, layout L = layout::row_major>
class tensor_indexer#

Utility class to compute addresses for multi-dimensional data.

Template Parameters:

• IdxT – Index type

• D – Tensor dimension

• L – Storage layout

Public Types

using multi_idx_t = std::array<IdxT, D>#

Data type of multi-indices.

Public Functions

inline tensor_indexer()#

Construct empty indexer.

inline tensor_indexer(multi_idx_t shape)#

Construct indexer for $N_1\times \cdots \times N_D$ tensor.

Strides are computed automatically assuming packed data as following:

Row-major layout: $s=(N_{D-1}\cdot \dots \cdot N_2,\dots ,N_{D-1},1)$

Column-major layout: $s=(1,N_1,\dots ,N_1\cdot \dots \cdot N_{D-1})$

Parameters:

shape – The numbers $N_1,\dots ,N_D$

inline tensor_indexer(multi_idx_t shape, multi_idx_t stride)#

Construct indexer for $N_1\times \cdots \times N_D$ tensor with manual strides.

Parameters:

• shape – The numbers $N_1,\dots ,N_D$

• stride – Custom strides

template<typename ...Indices, typename = std::enable_if_t<sizeof...(Indices) == D, int>>
inline IdxT operator()(Indices... is) const#

Compute linear index for entry $(i_1,...,i_D)$.

Indices are computed as following:

$$a=\sum _{j=1}^D{i}_j{s}_j,$$

where $s_j$ are the entries of the stride array.

Template Parameters:

Indices – Index types of arguments

Parameters:

...is – Indices

Returns:

Linear index

inline IdxT operator()(multi_idx_t const &idx) const#

Compute linear index.

Parameters:

idx – Multi-index

Returns:

Linear index

inline auto shape() const -> std::conditional_t<L == layout::row_major, multi_idx_t, multi_idx_t const&>#

Tensor shape.

Returns:

Numbers $N_1,\dots ,N_D$

inline auto shape(unsigned int d) const#

Tensor shape.

Parameters:

d – mode

Returns:

Number $N_d$

inline auto stride() const -> std::conditional_t<L == layout::row_major, multi_idx_t, multi_idx_t const&>#

Strides.

Returns:

Stride array

inline auto stride(unsigned int d) const#

Stride for d-th mode.

Parameters:

d – mode

Returns:

Stride

inline IdxT size() const#

Compute number of elements in tensor.

Returns:

Size (multiply with element type to get number of bytes)

inline constexpr auto dim() const#

Dimension.

Returns:

Dimension

template<unsigned int Dfrom = 0, unsigned int Dto = D - 1>
inline bool may_fuse() const#

Checks whether indices may be fused.

“Fusing” means that one may treat 2 or more neighbouring indices as a single super-index. Modes may only be fused it they are packed in memory.

Template Parameters:

• Dfrom – First fused mode

• Dto – Last fused mode

Returns:

True if modes may be fused

template<unsigned int Dfrom = 0, unsigned int Dto = D - 1>
inline auto fused() const#

Returns a tensor indexer with fused modes.

See also

may_fuse()

Template Parameters:

• Dfrom – First fused mode

• Dto – Last fused mode

Returns:

Fused tensor indexer

template<std::size_t E>
inline bool may_reshape_mode(int mode, std::array<IdxT, E> const &mode_shape) const#

Checks whether a mode may be reshaped.

“Reshaping a mode” means that one views a 1-D mode as a E-D tensor. E.g. for the tensor X_{i,j,k} of size N1 x N2 x N3 a reshape of mode 1 (= index j) with mode shape M1 x M2 means that we view the data of tensor X as the tensor X’_{i,j1,j2,k} of size N1 x M1 x M2 x N3.

Template Parameters:

E – reshape dimension

Parameters:

• mode – The mode number to reshape; counting starts from 0

• mode_shape – The E-D shape of the mode

Returns:

True if the mode may be reshaped

template<std::size_t E>
inline auto reshaped_mode(int mode, std::array<IdxT, E> mode_shape) const#

Returns a tensor indexer with reshaped mode.

See also

may_reshape_mode()

Template Parameters:

E – reshape dimension

Parameters:

• mode – The mode number to reshape; counting starts from 0

• mode_shape – The E-D shape of the mode

Returns:

Reshaped tensor indexer