#include <fft_grid.hpp>

Inheritance diagram for q3::FFTGrid:

Public Member Functions
virtual	~FFTGrid ()=default

virtual void	assign_sticks (const std::vector< int > &)

virtual bool	co_locates_mirrors () const

virtual int	concurrent_transform_slots () const

virtual void	execute_forward_on_buffers () const =0

virtual void	execute_forward_on_slot (int) const
	Forward transform on slot `slot`, the slot-aware `execute_forward_on_buffers()`.

virtual void	execute_inverse_raw_on_buffers () const =0

virtual void	execute_inverse_raw_on_slot (int) const
	Inverse transform on slot `slot`, the slot-aware `execute_inverse_raw_on_buffers()`.

virtual void	forward (Complex *data) const =0

virtual void	forward_batched (Complex *data, int batch_size) const

virtual int	g_vector_to_fft_grid_index (int miller_1, int miller_2, int miller_3) const =0

virtual void	gather_local_g_space (const Complex full_g_space, Complex local_g_space) const

virtual int	global_real_space_index (int local_index) const

virtual int	grid_dimension_1 () const =0

virtual int	grid_dimension_2 () const =0

virtual int	grid_dimension_3 () const =0

virtual void	init (int dimension_1, int dimension_2, int dimension_3, bool use_measured_planning=false)=0

virtual void	initialize_batched_plans (int, bool=false) const

virtual int	initialized_batched_plan_size () const

virtual Complex *	input_buffer () const =0

virtual void	inverse (Complex *data) const =0

virtual void	inverse_raw (Complex *data) const =0

virtual void	inverse_raw_batched (Complex *data, int batch_size) const

virtual void	inverse_raw_on_slot (Complex *data, int) const

virtual int	local_g_space_points () const

virtual int	local_grid_points () const =0

virtual int	local_offset () const =0

virtual Complex *	output_buffer () const =0

virtual bool	owns_stick (int, int) const

virtual void	prepare_forward_from_output () const =0

virtual void	prepare_forward_from_output_on_slot (int) const
	Output-to-input copy on slot `slot`, the slot-aware `prepare_forward_from_output()`.

virtual void	reduce_g_space (Complex *data, int number_of_elements) const =0

virtual void	scatter_local_g_space (const Complex local_g_space, Complex full_g_space) const
	The reverse: place this rank's reciprocal-space points into a full array.

virtual int	slab_dimension_1 () const

virtual Complex *	slot_input_buffer (int) const

virtual Complex *	slot_output_buffer (int) const

virtual int	total_grid_points () const =0

Detailed Description

FFTGrid — Abstract interface for 3D FFT grids.

Provides a unified contract for real <-> reciprocal space transformations, shielding consumers from the details of memory distribution (serial vs. MPI).

Definition at line 16 of file fft_grid.hpp.

Constructor & Destructor Documentation

◆ ~FFTGrid()

virtual q3::FFTGrid::~FFTGrid ( )

virtualdefault

Member Function Documentation

◆ assign_sticks()

virtual void q3::FFTGrid::assign_sticks ( const std::vector< int > & )

inlinevirtual

Assign reciprocal-space ownership from a weight per stick – the number of G-vectors on it inside the cutoff. Only a layout that distributes by stick has anything to do with it; everything else ignores it.

Reimplemented in q3::FFTGridSticks.

Definition at line 106 of file fft_grid.hpp.

Referenced by q3::PlanewaveBasis::generate(), and q3::PlanewaveBasis::initialize_kpoint_data().

◆ co_locates_mirrors()

virtual bool q3::FFTGrid::co_locates_mirrors ( ) const

inlinevirtual

Whether a G-vector and its negative are always held by the same rank. True for a serial grid and for the stick layout; false for slabs, which is what forced the gamma-only path off under MPI.

Reimplemented in q3::FFTGridSticks.

Definition at line 113 of file fft_grid.hpp.

Referenced by q3::PlanewaveKohnShamSCFIterator::initialize_representation().

◆ concurrent_transform_slots()

virtual int q3::FFTGrid::concurrent_transform_slots ( ) const

inlinevirtual

How many transforms this grid can have in flight at once.

The zero-copy buffer API above is a single set of scratch arrays, so two threads transforming through it at the same time would overwrite each other. A grid that can hand out independent scratch reports more than one slot here and honours the *_on_slot calls below; a grid that cannot leaves this at 1 and callers keep their band loops serial.

Slot 0 is always the buffers input_buffer() and output_buffer() return, so a caller that asks for one slot and a caller that never heard of slots drive exactly the same arrays through exactly the same plan. That is what keeps a one-thread run bit-identical to a build with no threading at all.

Reimplemented in q3::SerialFFTGrid.

Definition at line 235 of file fft_grid.hpp.

◆ execute_forward_on_buffers()

virtual void q3::FFTGrid::execute_forward_on_buffers ( ) const

pure virtual

Execute a forward FFT on the internal buffers (no copy).

Caller must have populated input_buffer() before calling. Result is available in output_buffer() after this call.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by execute_forward_on_slot().

◆ execute_forward_on_slot()

virtual void q3::FFTGrid::execute_forward_on_slot ( int ) const

inlinevirtual

Forward transform on slot slot, the slot-aware execute_forward_on_buffers().

Reimplemented in q3::SerialFFTGrid.

Definition at line 261 of file fft_grid.hpp.

References execute_forward_on_buffers().

◆ execute_inverse_raw_on_buffers()

virtual void q3::FFTGrid::execute_inverse_raw_on_buffers ( ) const

pure virtual

Execute a raw inverse FFT on the internal buffers (no copy, no normalization).

Caller must have populated input_buffer() before calling. Result is available in output_buffer() after this call.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by execute_inverse_raw_on_slot().

◆ execute_inverse_raw_on_slot()

virtual void q3::FFTGrid::execute_inverse_raw_on_slot ( int ) const

inlinevirtual

Inverse transform on slot slot, the slot-aware execute_inverse_raw_on_buffers().

Reimplemented in q3::SerialFFTGrid.

Definition at line 256 of file fft_grid.hpp.

References execute_inverse_raw_on_buffers().

◆ forward()

virtual void q3::FFTGrid::forward ( Complex * data ) const

pure virtual

Forward FFT: real-space -> reciprocal-space (unnormalized).

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::compute_smooth_grid_potential_from_dense(), forward_batched(), q3::gpu::GPUFFTGrid::forward_transform(), q3::WavefunctionTransformer::r_to_g(), q3::ChargeDensityTransformer::rho_r_to_g(), and q3::PlanewaveDirectDiagonalizer::solve().

◆ forward_batched()

virtual void q3::FFTGrid::forward_batched	(	Complex *	data,
		int	batch_size
	)		const

inlinevirtual

Batched forward FFT: real-space -> reciprocal-space (unnormalized).

Transforms batch_size independent 3D grids packed contiguously: data[0 .. local_grid_points-1] = transform 0 data[local_grid_points .. 2*local_grid_points-1] = transform 1 ...

The default implementation loops over forward() for each transform.

Definition at line 319 of file fft_grid.hpp.

References forward(), and local_grid_points().

◆ g_vector_to_fft_grid_index()

virtual int q3::FFTGrid::g_vector_to_fft_grid_index	(	int	miller_1,
		int	miller_2,
		int	miller_3
	)		const

pure virtual

Map a G-vector (integer Miller indices) to FFT array index. Handles negative indices via wrapping.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::compute_g_to_fft_index_mapping(), q3::compute_smooth_grid_potential_from_dense(), q3::WavefunctionTransformer::g_to_r(), q3::WavefunctionTransformer::r_to_g(), q3::ChargeDensityTransformer::rho_g_to_r(), q3::ChargeDensityTransformer::rho_r_to_g(), and q3::PlanewaveDirectDiagonalizer::solve().

◆ gather_local_g_space()

virtual void q3::FFTGrid::gather_local_g_space	(	const Complex *	full_g_space,
		Complex *	local_g_space
	)		const

inlinevirtual

Copy this rank's reciprocal-space points out of a full, replicated G-space array laid out [i1][i2][i3], into the local G-space layout. The slab layout is a contiguous run at local_offset(); the stick layout is not, so consumers must go through this rather than a memcpy.

Reimplemented in q3::FFTGridSticks.

Definition at line 123 of file fft_grid.hpp.

References grid_dimension_2(), grid_dimension_3(), local_g_space_points(), and local_offset().

Referenced by q3::inverse_fft_smooth_potential().

◆ global_real_space_index()

virtual int q3::FFTGrid::global_real_space_index ( int local_index ) const

inlinevirtual

Where this rank's real-space point local_index sits in the whole grid, as the flat index i1·d2·d3 + i2·d3 + i3 that a serial grid stores it at.

Code that needs the position of a grid point – a multipole moment, a kernel evaluated at r – must ask this rather than add local_offset() planes: that is right for a serial grid and a slab along the first dimension, which is this default, and wrong for the stick layout, which holds planes along the third dimension and overrides it.

Reimplemented in q3::FFTGridSticks.

Definition at line 64 of file fft_grid.hpp.

References grid_dimension_2(), grid_dimension_3(), and local_offset().

Referenced by q3::MartynaTuckermanCorrectionCalculator::compute().

◆ grid_dimension_1()

virtual int q3::FFTGrid::grid_dimension_1 ( ) const

pure virtual

Grid size along the first dimension.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::MakovPayneCorrectionCalculator::compute(), q3::MartynaTuckermanCorrectionCalculator::compute(), q3::compute_g_to_fft_index_mapping(), q3::compute_smooth_grid_potential_from_dense(), q3::PlanewaveBasis::generate(), q3::PlanewaveKohnShamSCFIterator::log_representation_summary(), slab_dimension_1(), q3::PlanewaveDirectDiagonalizer::solve(), and q3::PlanewaveBasis::stick_weights_for_grid().

◆ grid_dimension_2()

virtual int q3::FFTGrid::grid_dimension_2 ( ) const

pure virtual

Grid size along the second dimension.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::MartynaTuckermanCorrectionCalculator::compute(), q3::compute_g_to_fft_index_mapping(), q3::compute_smooth_grid_potential_from_dense(), gather_local_g_space(), q3::PlanewaveBasis::generate(), global_real_space_index(), q3::PlanewaveKohnShamSCFIterator::log_representation_summary(), scatter_local_g_space(), q3::PlanewaveDirectDiagonalizer::solve(), and q3::PlanewaveBasis::stick_weights_for_grid().

◆ grid_dimension_3()

virtual int q3::FFTGrid::grid_dimension_3 ( ) const

pure virtual

Grid size along the third dimension.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::MartynaTuckermanCorrectionCalculator::compute(), q3::compute_g_to_fft_index_mapping(), q3::compute_smooth_grid_potential_from_dense(), gather_local_g_space(), q3::PlanewaveBasis::generate(), global_real_space_index(), q3::PlanewaveKohnShamSCFIterator::log_representation_summary(), scatter_local_g_space(), q3::PlanewaveDirectDiagonalizer::solve(), and q3::PlanewaveBasis::stick_weights_for_grid().

◆ init()

virtual void q3::FFTGrid::init	(	int	dimension_1,
		int	dimension_2,
		int	dimension_3,
		bool	use_measured_planning = `false`
	)

pure virtual

Initialize with grid dimensions (dimension_1 x dimension_2 x dimension_3).

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::GPUPlanewaveHamiltonianApplicator::GPUPlanewaveHamiltonianApplicator().

◆ initialize_batched_plans()

virtual void q3::FFTGrid::initialize_batched_plans	(	int	,
		bool	= `false`
	)		const

inlinevirtual

Pre-create FFTW plans for batched transforms of a given batch size.

Must be called before forward_batched() or inverse_raw_batched() with the corresponding batch_size. Calling with a different batch_size replaces the existing batched plans.

The default implementation is a no-op (the fallback batched methods loop over single-transform calls, which need no separate plan).

Definition at line 292 of file fft_grid.hpp.

◆ initialized_batched_plan_size()

virtual int q3::FFTGrid::initialized_batched_plan_size ( ) const

inlinevirtual

Returns the batch_size for which batched plans are currently initialized, or 0 if no batched plans have been created.

Callers can use this to determine whether to zero-pad a smaller batch to the pre-initialized plan size (avoiding a plan recreation that would require a collective MPI call).

Definition at line 305 of file fft_grid.hpp.

◆ input_buffer()

virtual Complex * q3::FFTGrid::input_buffer ( ) const

pure virtual

Access the FFT grid's internal input buffer for zero-copy transforms.

Callers scatter G-space data directly into this buffer, then call execute_inverse_raw_on_buffers() to transform in-place. This eliminates the memcpy overhead of inverse_raw() / forward().

The buffer has at least local_grid_points() elements of usable space.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by slot_input_buffer().

◆ inverse()

virtual void q3::FFTGrid::inverse ( Complex * data ) const

pure virtual

Inverse FFT: reciprocal-space -> real-space (includes 1/N normalization).

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::WavefunctionTransformer::g_to_r(), q3::ChargeDensityTransformer::rho_g_to_r(), and q3::PlanewaveDirectDiagonalizer::solve().

◆ inverse_raw()

virtual void q3::FFTGrid::inverse_raw ( Complex * data ) const

pure virtual

Inverse FFT without 1/N normalization (raw FFTW backward).

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::inverse_fft_smooth_potential(), inverse_raw_batched(), inverse_raw_on_slot(), and q3::gpu::GPUFFTGrid::inverse_transform().

◆ inverse_raw_batched()

virtual void q3::FFTGrid::inverse_raw_batched	(	Complex *	data,
		int	batch_size
	)		const

inlinevirtual

Batched inverse FFT without 1/N normalization (raw FFTW backward).

Transforms batch_size independent 3D grids packed contiguously (same layout as forward_batched).

The default implementation loops over inverse_raw() for each transform.

Definition at line 334 of file fft_grid.hpp.

References inverse_raw(), and local_grid_points().

◆ inverse_raw_on_slot()

virtual void q3::FFTGrid::inverse_raw_on_slot	(	Complex *	data,
		int
	)		const

inlinevirtual

In-place inverse transform of a caller-owned array through slot slot, the slot-aware form of inverse_raw().

The array belongs to the caller, so two threads passing different arrays collide only over the grid's own scratch – which is what the slot separates.

Reimplemented in q3::SerialFFTGrid.

Definition at line 278 of file fft_grid.hpp.

References inverse_raw().

◆ local_g_space_points()

virtual int q3::FFTGrid::local_g_space_points ( ) const

inlinevirtual

Number of reciprocal-space points this rank holds.

Equal to local_grid_points() for every layout in which G-space and real space share a buffer. The stick decomposition holds G-space as whole sticks and real space as whole planes, so the two counts differ, and a caller sizing a G-space scatter must ask this rather than assume.

Reimplemented in q3::FFTGridSticks.

Definition at line 88 of file fft_grid.hpp.

References local_grid_points().

Referenced by q3::compute_smooth_grid_potential_from_dense(), q3::WavefunctionTransformer::g_to_r(), gather_local_g_space(), q3::inverse_fft_smooth_potential(), q3::WavefunctionTransformer::r_to_g(), q3::ChargeDensityTransformer::rho_g_to_r(), q3::ChargeDensityTransformer::rho_r_to_g(), and scatter_local_g_space().

◆ local_grid_points()

virtual int q3::FFTGrid::local_grid_points ( ) const

pure virtual

Number of real-space grid points stored on the local rank.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::PlanewaveChargeDensityAccumulator::accumulate(), q3::GPUPlanewaveHamiltonianApplicator::apply_batch(), q3::GPUPlanewaveHamiltonianApplicator::apply_batch_single_precision(), q3::PlanewaveEffectivePotentialCalculator::build_exchange_correlation_density(), q3::NormConservingPlanewaveEffectivePotentialCalculator::build_exchange_correlation_density(), q3::MartynaTuckermanCorrectionCalculator::compute(), q3::PlanewaveEffectivePotentialCalculator::compute(), q3::PlanewaveKohnShamSCFIterator::compute_delta_escf_correction(), q3::PlanewaveKohnShamSCFIterator::compute_double_counting_correction(), q3::compute_smooth_grid_potential_from_dense(), q3::GradientCorrectedExchangeCorrelationPotentialCalculator::evaluate_spin_polarized(), q3::GradientCorrectedExchangeCorrelationPotentialCalculator::evaluate_unpolarized(), forward_batched(), q3::WavefunctionTransformer::g_to_r(), q3::inverse_fft_smooth_potential(), inverse_raw_batched(), q3::PlanewaveKohnShamSCFIterator::loaded_charge_density_fits_grid(), local_g_space_points(), q3::WavefunctionTransformer::r_to_g(), q3::ChargeDensityTransformer::rho_g_to_r(), q3::ChargeDensityTransformer::rho_r_to_g(), q3::GPUPlanewaveHamiltonianApplicator::update_potential(), q3::SerialPlanewaveHamiltonianApplicator::update_potential(), and q3::PlanewaveKohnShamSCFIterator::warn_if_total_charge_deviates().

◆ local_offset()

virtual int q3::FFTGrid::local_offset ( ) const

pure virtual

Offset into the global grid for the local segment (e.g., start of slab).

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by gather_local_g_space(), global_real_space_index(), and scatter_local_g_space().

◆ output_buffer()

virtual Complex * q3::FFTGrid::output_buffer ( ) const

pure virtual

Access the FFT grid's internal output buffer for zero-copy transforms.

After execute_inverse_raw_on_buffers(), real-space data is here. After execute_forward_on_buffers(), G-space data is here.

For in-place plans (MPI FFT), this returns the same pointer as input_buffer(). For out-of-place plans (Serial FFT), this is a separate buffer.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by slot_output_buffer().

◆ owns_stick()

virtual bool q3::FFTGrid::owns_stick	(	int	,
		int
	)		const

inlinevirtual

Whether this rank holds the G-vectors sharing the first two Miller indices. True everywhere for a layout that does not distribute G-space by stick; the plane-wave selector asks this to decide what to keep.

Reimplemented in q3::FFTGridSticks.

Definition at line 97 of file fft_grid.hpp.

◆ prepare_forward_from_output()

virtual void q3::FFTGrid::prepare_forward_from_output ( ) const

pure virtual

Prepare the input buffer for a forward transform after modifying the output buffer.

For out-of-place plans (Serial FFT), this copies output_buffer → input_buffer. For in-place plans (MPI FFT), this is a no-op (same buffer).

Call this between the real-space multiply and execute_forward_on_buffers().

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by prepare_forward_from_output_on_slot().

◆ prepare_forward_from_output_on_slot()

virtual void q3::FFTGrid::prepare_forward_from_output_on_slot ( int ) const

inlinevirtual

Output-to-input copy on slot slot, the slot-aware prepare_forward_from_output().

Reimplemented in q3::SerialFFTGrid.

Definition at line 266 of file fft_grid.hpp.

References prepare_forward_from_output().

◆ reduce_g_space()

virtual void q3::FFTGrid::reduce_g_space	(	Complex *	data,
		int	number_of_elements
	)		const

pure virtual

Reduce G-space data across the pool to make it redundant on all ranks.

Serial: no-op. MPI: MPI_Allreduce(MPI_IN_PLACE, ..., MPI_SUM, pool_comm).

Implemented in q3::SerialFFTGrid, q3::SerialPocketFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, and q3::gpu::SYCLGPUFFTGrid.

Referenced by q3::ChargeDensityTransformer::rho_r_to_g().

◆ scatter_local_g_space()

virtual void q3::FFTGrid::scatter_local_g_space	(	const Complex *	local_g_space,
		Complex *	full_g_space
	)		const

inlinevirtual

The reverse: place this rank's reciprocal-space points into a full array.

Reimplemented in q3::FFTGridSticks.

Definition at line 133 of file fft_grid.hpp.

References grid_dimension_2(), grid_dimension_3(), local_g_space_points(), and local_offset().

◆ slab_dimension_1()

virtual int q3::FFTGrid::slab_dimension_1 ( ) const

inlinevirtual

Number of first-dimension planes owned by this rank for G-vector partitioning.

In serial mode this equals grid_dimension_1(). In MPI mode this returns the slab size (local_n1 from FFTW's decomposition), which may differ from local_grid_points() / (dim2*dim3) when the FFT operates on the full grid (Allreduce mode).

Reimplemented in q3::FFTGridSticks.

Definition at line 76 of file fft_grid.hpp.

References grid_dimension_1().

◆ slot_input_buffer()

virtual Complex * q3::FFTGrid::slot_input_buffer ( int ) const

inlinevirtual

The input buffer of transform slot slot, the slot-aware form of input_buffer().

Reimplemented in q3::SerialFFTGrid.

Definition at line 243 of file fft_grid.hpp.

References input_buffer().

◆ slot_output_buffer()

virtual Complex * q3::FFTGrid::slot_output_buffer ( int ) const

inlinevirtual

The output buffer of transform slot slot, the slot-aware form of output_buffer().

Reimplemented in q3::SerialFFTGrid.

Definition at line 251 of file fft_grid.hpp.

References output_buffer().

◆ total_grid_points()

virtual int q3::FFTGrid::total_grid_points ( ) const

pure virtual

Total number of real-space grid points across all processes.

Implemented in q3::SerialFFTGrid, q3::FFTGridSticks, q3::gpu::CudaGPUFFTGrid, q3::gpu::HIPGPUFFTGrid, q3::gpu::MetalGPUFFTGrid, q3::gpu::SYCLGPUFFTGrid, and q3::SerialPocketFFTGrid.

Referenced by q3::PlanewaveChargeDensity::PlanewaveChargeDensity(), q3::PlanewaveEffectivePotential::PlanewaveEffectivePotential(), q3::PlanewaveChargeDensityAccumulator::accumulate(), q3::PlanewaveEffectivePotentialCalculator::compute(), q3::PlanewaveKohnShamSCFIterator::compute_delta_escf_correction(), q3::PlanewaveKohnShamSCFIterator::compute_double_counting_correction(), q3::compute_smooth_grid_potential_from_dense(), q3::gpu::GPUFFTGrid::forward_single_batch(), q3::gpu::GPUFFTGrid::forward_transform_batch(), q3::PlanewaveChargeDensity::get_total_charge(), q3::PlanewaveChargeDensity::grid_size(), q3::PlanewaveEffectivePotential::grid_size(), q3::PlanewaveEffectivePotential::initialize_kinetic_potential(), q3::PlanewavePseudopotentialKohnShamSCFIterator::initialize_single_particle_states(), q3::gpu::GPUFFTGrid::inverse_raw_single_batch(), q3::gpu::GPUFFTGrid::inverse_transform_batch(), q3::PlanewaveKohnShamSCFIterator::preload_charge_density(), q3::PlanewaveKohnShamSCFIterator::report_convergence_details(), q3::PlanewaveChargeDensity::reset(), q3::ChargeDensityTransformer::rho_g_to_r(), q3::ChargeDensityTransformer::rho_r_to_g(), q3::PlanewaveKohnShamSCFIterator::scale_loaded_charge_density_to_electron_count(), q3::PlanewaveDirectDiagonalizer::solve(), q3::PlanewaveChargeDensity::to_reciprocal_space(), q3::PlanewaveEffectivePotential::to_reciprocal_space(), q3::GPUPlanewaveHamiltonianApplicator::update_potential(), q3::SerialPlanewaveHamiltonianApplicator::update_potential(), and q3::PlanewaveKohnShamSCFIterator::warn_if_total_charge_deviates().

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

src/layer_3_representation/basis/planewave/internal/fft/fft_grid.hpp

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ ~FFTGrid()

Member Function Documentation

◆ assign_sticks()

◆ co_locates_mirrors()

◆ concurrent_transform_slots()

◆ execute_forward_on_buffers()

◆ execute_forward_on_slot()

◆ execute_inverse_raw_on_buffers()

◆ execute_inverse_raw_on_slot()

◆ forward()

◆ forward_batched()

◆ g_vector_to_fft_grid_index()

◆ gather_local_g_space()

◆ global_real_space_index()

◆ grid_dimension_1()

◆ grid_dimension_2()

◆ grid_dimension_3()

◆ init()

◆ initialize_batched_plans()

◆ initialized_batched_plan_size()

◆ input_buffer()

◆ inverse()

◆ inverse_raw()

◆ inverse_raw_batched()

◆ inverse_raw_on_slot()

◆ local_g_space_points()

◆ local_grid_points()

◆ local_offset()

◆ output_buffer()

◆ owns_stick()

◆ prepare_forward_from_output()

◆ prepare_forward_from_output_on_slot()

◆ reduce_g_space()

◆ scatter_local_g_space()

◆ slab_dimension_1()

◆ slot_input_buffer()

◆ slot_output_buffer()

◆ total_grid_points()