(git:ec04593)
Loading...
Searching...
No Matches
kpoint_transform Module Reference

Prepared real-space/reciprocal-space transforms and distributed AO adapters. Routes fix geometry and ownership. Captured source rows own numerical values; cached evaluation reads them without communication or SCF state. More...

Data Types

type  k_grid_to_cell_phase_type
type  kp_transform_plan_type
type  rskp_route_type
type  rskp_values_type
type  rskp_prepared_transform_type

Functions/Subroutines

subroutine, public cell_to_k_transform (values_rs, cells, xkp, nkp_grid, values_k, backend, accumulate, profile_fft, profile_label)
 Transform local real matrix panels with the selected RSKP backend. GEMM reuses the prepared parity folding and block contraction. FFT reuses one geometry across bounded matrix tiles. Both accept rectangular panels and regular-grid subsets. Input and output may be strided; only the internal GEMM panels are contiguous. Call outside OpenMP worker regions; this service owns its numerical workspace.
subroutine, public k_to_real_cell_transform (values_k, xkp, nkp_grid, cells, values_rs, backend, profile_fft)
 Real part of the inverse lattice transform, normalized by the number of k points. Local rectangular panels use shared FFT execution or real GEMM contractions with cosine and sine weights. Matrix and cell tiles bound temporary storage. Call outside OpenMP worker regions; this service owns its numerical workspace.
subroutine, public rskp_transform (rmatrix, cmatrix, rsmat, ispin, xkp, cell_to_index, sab_nl, is_complex, rs_sign)
 Transformation of real space matrices to a kpoint.
subroutine, public rskp_transform_prepare (prepared, source, source_rows, source_sab, cell_to_index, target_struct, world, group, inter, backend, xkp, nkp_grid, kp_dist, mirror_sign, scale)
 Prepare source rows for a generic group- or world-distributed target. GEMM retains sparse snapshots and sizes k tiles against a workspace target. Own route, snapshots, singleton scratch and any FFT cache without SCF slots.
subroutine, public rskp_transform_evaluate (prepared, source_index, k, xkp, fm_re, fm_im, cfm)
 Evaluate one k point directly into an FM, split-FM or CFM target. The handle owns mutable scratch; concurrent evaluations need separate handles. Evaluation uses the batch backend with width one and no communication. Groups that do not own the requested global k point leave outputs unchanged.
subroutine, public rskp_transform_release (prepared)
 Release the route, snapshot and cached backend state of a prepared transform.
subroutine, public rskp_dbcsr_transform_batch (source, real_k, imag_k, xkp, cell_to_index, sab_nl, nkp_grid, backend)
 Transform all real-space spin rows to a full k-grid in distributed DBCSR matrices. The real and imaginary output blocks retain their caller-provided sparsity and distribution. GEMM contracts all images of an AO block together; FFT transforms element tiles while reusing one prepared lattice geometry.
subroutine, public kp_transform_plan_create (plan, sab_nl, cell_to_index, nimg, block_template, group_entries)
 Build the immutable neighbor-list traversal shared by R-to-K and K-to-R transforms.
subroutine, public rskp_route_create (source, source_sab, cell_to_index, matrix_struct, world, group, inter, backend, route, ready)
 Cache routing to the FM tiles of one representative group. Equivalent ranks in the other groups receive only their tile's blocks. Both backends receive the union of locally owned direct/mirrored AO elements. Missing source blocks are packed as zero on a fixed geometry.
logical function, public rskp_compatible (route, source, matrix_row, source_sab)
 Check the source AO layout, not its values or sparse block presence.
subroutine, public rskp_capture (source, matrix_row, route, snapshot, mirror_sign, scale)
 Capture current values using cached routing, with no per-k communication. GEMM/FFT snapshots own packed local AO-element/image matrices. A shared cropped receive buffer is reused across slots and operator refreshes.
subroutine, public rskp_values_release (snapshot)
 Release one captured row without invalidating other slots or routing.
subroutine, public rskp_route_release (route)
 Release backend-owned geometry and FFT work buffers from a route.
subroutine, public cell_to_k_sparse_fft (snapshot, route, xkp, nkp_grid, kp_dist, inter, max_bytes, cache, output)
 FFT captured raw-element panels and retain the destination group's k points. Reuse the sparse route and the dense layout consumed by the batched solvers.
subroutine, public cell_to_k_backend_prepare (route, xkp, nkp_grid, max_bytes)
 Prepare the selected R-to-K backend plan once for a fixed route. Both backends retain their geometry with the route; the snapshots then provide only values and use the common evaluate interface.
subroutine, public rskp_batch_size (route, nk, ncomp, nworkers, nmatrices, max_bytes, width, work_size)
 Bound worker scratch and output tiles for either prepared backend. Packed input belongs to the snapshot and is shared by every k tile.
subroutine, public cell_to_k_backend_evaluate_batch (snapshot, route, xkp, first_local, work, batch, output)
 Evaluate a k-point tile through the selected prepared R-to-K backend. GEMM contracts the retained real-space panel; FFT returns a slice of the canonical reciprocal-space snapshot materialized during capture.
subroutine, public k_grid_to_cell_phase_prepare (table, cells, xkp, nworkers)
 Tabulate distinct absolute translation coordinates for a group's local k points. All AO pairs share these unweighted phases. Zero coordinates are exact; negative coordinates use the same cosine and the opposite sine.
subroutine, public k_grid_to_cell_contract (source, tiles, ikp_index, xkp, weight, images, source_pair, block_shape, block_offset, shift, time_reversal, trans, phase_table, sums, gemm_ctx)
 Weighted K-to-R contraction of one AO block, before spatial rotations and MPI. The caller supplies the atom mapping and image multiplicities. Phases include the gauge shift, time reversal and conjugation of a transposed source block.
subroutine, public rskp_phase_factors (shift, xkp, time_reversal, real_only, phase)
 Real two-component representation of the Bloch phase and time reversal.
subroutine, public rskp_work_matrix (work, nrow, ncol)
 Allocate a dense work matrix with the requested shape.

Detailed Description

Prepared real-space/reciprocal-space transforms and distributed AO adapters. Routes fix geometry and ownership. Captured source rows own numerical values; cached evaluation reads them without communication or SCF state.

Function/Subroutine Documentation

◆ cell_to_k_transform()

subroutine, public kpoint_transform::cell_to_k_transform ( real(kind=dp), dimension(:, :, :), intent(in) values_rs,
integer, dimension(:, :), intent(in) cells,
real(kind=dp), dimension(:, :), intent(in) xkp,
integer, dimension(3), intent(in) nkp_grid,
complex(kind=dp), dimension(:, :, :), intent(inout) values_k,
integer, intent(in) backend,
logical, intent(in), optional accumulate,
logical, intent(in), optional profile_fft,
character(len=*), intent(in), optional profile_label )

Transform local real matrix panels with the selected RSKP backend. GEMM reuses the prepared parity folding and block contraction. FFT reuses one geometry across bounded matrix tiles. Both accept rectangular panels and regular-grid subsets. Input and output may be strided; only the internal GEMM panels are contiguous. Call outside OpenMP worker regions; this service owns its numerical workspace.

Parameters
values_rsreal matrices, with one image per third-dimension column
cellslattice translations of the images
xkprequested reciprocal coordinates, in output order
nkp_gridmesh dimensions for FFT
values_kcomplex transformed matrices
backendGEMM or FFT
accumulateadd this image panel to an existing output
profile_fftprint local FFT stage times on the logger rank, without MPI collectives
profile_labelidentify the caller operation in the optional FFT profile

Definition at line 158 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ k_to_real_cell_transform()

subroutine, public kpoint_transform::k_to_real_cell_transform ( complex(kind=dp), dimension(:, :, :), intent(in), target, contiguous values_k,
real(kind=dp), dimension(:, :), intent(in) xkp,
integer, dimension(3), intent(in) nkp_grid,
integer, dimension(:, :), intent(in) cells,
real(kind=dp), dimension(:, :, :), intent(out), target, contiguous values_rs,
integer, intent(in) backend,
logical, intent(in), optional profile_fft )

Real part of the inverse lattice transform, normalized by the number of k points. Local rectangular panels use shared FFT execution or real GEMM contractions with cosine and sine weights. Matrix and cell tiles bound temporary storage. Call outside OpenMP worker regions; this service owns its numerical workspace.

Parameters
values_kcomplex reciprocal-space matrices
xkpreciprocal coordinates
nkp_gridmesh dimensions for FFT
cellsrequested lattice translations
values_rsreal part of the reconstructed matrices
backendFFT backend (required)
profile_fftprint local FFT stage times on the logger rank, without MPI collectives

Definition at line 327 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_transform()

subroutine, public kpoint_transform::rskp_transform ( type(dbcsr_type) rmatrix,
type(dbcsr_type), optional cmatrix,
type(dbcsr_p_type), dimension(:, :), pointer rsmat,
integer, intent(in) ispin,
real(kind=dp), dimension(3), intent(in) xkp,
integer, dimension(:, :, :), pointer cell_to_index,
type(neighbor_list_set_p_type), dimension(:), pointer sab_nl,
logical, intent(in), optional is_complex,
real(kind=dp), intent(in), optional rs_sign )

Transformation of real space matrices to a kpoint.

Parameters
rmatrixReal part of kpoint matrix
cmatrixComplex part of kpoint matrix (optional)
rsmatReal space matrices
ispinSpin index
xkpKpoint coordinates
cell_to_indexmapping of cell indices to RS index
sab_nlDefining neighbor list
is_complexMatrix to be transformed is imaginary
rs_signMatrix to be transformed is csaled by rs_sign

Definition at line 407 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_transform_prepare()

subroutine, public kpoint_transform::rskp_transform_prepare ( type(rskp_prepared_transform_type), intent(inout) prepared,
type(dbcsr_p_type), dimension(:, :), pointer source,
integer, dimension(:), intent(in) source_rows,
type(neighbor_list_set_p_type), dimension(:), pointer source_sab,
integer, dimension(:, :, :), pointer cell_to_index,
type(cp_fm_struct_type), pointer target_struct,
type(mp_para_env_type), pointer world,
type(mp_para_env_type), optional, pointer group,
type(mp_para_env_type), optional, pointer inter,
integer, intent(in) backend,
real(kind=dp), dimension(:, :), intent(in) xkp,
integer, dimension(3), intent(in) nkp_grid,
integer, dimension(:, :), intent(in), optional kp_dist,
integer, intent(in), optional mirror_sign,
complex(kind=dp), intent(in), optional scale )

Prepare source rows for a generic group- or world-distributed target. GEMM retains sparse snapshots and sizes k tiles against a workspace target. Own route, snapshots, singleton scratch and any FFT cache without SCF slots.

Parameters
preparedtransform handle to initialize
sourcereal-space source rows and images
source_rowssource rows sharing the route/backend plan
source_sabneighbor-list geometry of source
cell_to_indextranslation-to-image map
target_structoutput matrix layout
worldsource-matrix communicator
groupoptional target-group communicator; omit for a world target
interoptional communicator joining equivalent group ranks; omit for a world target
backendselected GEMM or FFT backend
xkpfull global k-point coordinates
nkp_gridregular-grid dimensions
kp_distoptional global k ranges by inter-group rank; omit for a world target
mirror_signtranspose parity (+1 Hermitian, -1 anti-Hermitian)
scalecomplex prefactor applied to direct and mirrored contributions

Definition at line 522 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_transform_evaluate()

subroutine, public kpoint_transform::rskp_transform_evaluate ( type(rskp_prepared_transform_type), intent(inout) prepared,
integer, intent(in) source_index,
integer, intent(in) k,
real(kind=dp), dimension(:, :), intent(in) xkp,
type(cp_fm_type), intent(inout), optional, target fm_re,
type(cp_fm_type), intent(inout), optional, target fm_im,
type(cp_cfm_type), intent(inout), optional, target cfm )

Evaluate one k point directly into an FM, split-FM or CFM target. The handle owns mutable scratch; concurrent evaluations need separate handles. Evaluation uses the batch backend with width one and no communication. Groups that do not own the requested global k point leave outputs unchanged.

Parameters
preparedinitialized transform handle
source_indexone-based source row index supplied at preparation
kglobal k-point index
xkpfull global k-point coordinates
fm_rereal output matrix for split-FM storage
fm_imoptional imaginary output matrix for split-FM storage
cfmoptional complex full-matrix output

Definition at line 631 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_transform_release()

subroutine, public kpoint_transform::rskp_transform_release ( type(rskp_prepared_transform_type), intent(inout) prepared)

Release the route, snapshot and cached backend state of a prepared transform.

Parameters
preparedtransform handle to release

Definition at line 677 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_dbcsr_transform_batch()

subroutine, public kpoint_transform::rskp_dbcsr_transform_batch ( type(dbcsr_p_type), dimension(:, :), pointer source,
type(dbcsr_p_type), dimension(:, :), pointer real_k,
type(dbcsr_p_type), dimension(:, :), pointer imag_k,
real(kind=dp), dimension(:, :), intent(in) xkp,
integer, dimension(:, :, :), pointer cell_to_index,
type(neighbor_list_set_p_type), dimension(:), pointer sab_nl,
integer, dimension(3), intent(in) nkp_grid,
integer, intent(in) backend )

Transform all real-space spin rows to a full k-grid in distributed DBCSR matrices. The real and imaginary output blocks retain their caller-provided sparsity and distribution. GEMM contracts all images of an AO block together; FFT transforms element tiles while reusing one prepared lattice geometry.

Parameters
sourcereal-space image matrices, by spin and image
real_ksymmetric real output matrices, by spin and k point
imag_kantisymmetric imaginary output matrices, by spin and k point
xkpfull reciprocal grid
cell_to_indextranslation-to-image map
sab_nlneighbor-list geometry of source
nkp_griddimensions of the regular reciprocal grid
backendselected bulk R-to-K backend

Definition at line 707 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ kp_transform_plan_create()

subroutine, public kpoint_transform::kp_transform_plan_create ( type(kp_transform_plan_type), intent(out) plan,
type(neighbor_list_set_p_type), dimension(:), pointer sab_nl,
integer, dimension(:, :, :), pointer cell_to_index,
integer, intent(in) nimg,
type(dbcsr_type), intent(in), optional block_template,
logical, intent(in), optional group_entries )

Build the immutable neighbor-list traversal shared by R-to-K and K-to-R transforms.

Parameters
plan...
sab_nl...
cell_to_index...
nimg...
block_template...
group_entries...

Definition at line 943 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_route_create()

subroutine, public kpoint_transform::rskp_route_create ( type(dbcsr_p_type), dimension(:, :), pointer source,
type(neighbor_list_set_p_type), dimension(:), pointer source_sab,
integer, dimension(:, :, :), pointer cell_to_index,
type(cp_fm_struct_type), pointer matrix_struct,
type(mp_para_env_type), pointer world,
type(mp_para_env_type), pointer group,
type(mp_para_env_type), pointer inter,
integer, intent(in) backend,
type(rskp_route_type), intent(out) route,
logical, intent(out) ready )

Cache routing to the FM tiles of one representative group. Equivalent ranks in the other groups receive only their tile's blocks. Both backends receive the union of locally owned direct/mirrored AO elements. Missing source blocks are packed as zero on a fixed geometry.

Parameters
sourceoperator layout template
source_sabneighbor lists belonging to source
cell_to_indexmap from lattice translations to source image indices
matrix_structtarget group or world FM layout
worldcommunicator distributing the source DBCSR matrices
groupcommunicator of group-local FM targets
intercommunicator joining equivalent FM ranks across groups
backendGEMM or FFT
routecontext-owned structural route
readyfalse collectively if source or target layout is incompatible

Definition at line 1091 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_compatible()

logical function, public kpoint_transform::rskp_compatible ( type(rskp_route_type), intent(in) route,
type(dbcsr_p_type), dimension(:, :), pointer source,
integer, intent(in) matrix_row,
type(neighbor_list_set_p_type), dimension(:), pointer source_sab )

Check the source AO layout, not its values or sparse block presence.

Parameters
routefixed context route
sourcecurrent operator images
matrix_rowrow to capture
source_sabsource neighbor list, required to match the route's geometry
Returns
whether the row has the route's image count and AO layout

Definition at line 1473 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ rskp_capture()

subroutine, public kpoint_transform::rskp_capture ( type(dbcsr_p_type), dimension(:, :), pointer source,
integer, intent(in) matrix_row,
type(rskp_route_type), intent(inout), target route,
type(rskp_values_type), intent(out), target snapshot,
integer, intent(in), optional mirror_sign,
complex(kind=dp), intent(in), optional scale )

Capture current values using cached routing, with no per-k communication. GEMM/FFT snapshots own packed local AO-element/image matrices. A shared cropped receive buffer is reused across slots and operator refreshes.

Parameters
sourcecurrent real-space operator images
matrix_rowrow to snapshot
routecontext-owned metadata, counts and reusable packing buffer
snapshotimmutable tile-local values
mirror_signtranspose parity of the unscaled operator (+1 or -1)
scalecomplex prefactor, applied after forming direct and mirrored values

Definition at line 1509 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_values_release()

subroutine, public kpoint_transform::rskp_values_release ( type(rskp_values_type), intent(inout) snapshot)

Release one captured row without invalidating other slots or routing.

Parameters
snapshotrow to clear

Definition at line 1632 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ rskp_route_release()

subroutine, public kpoint_transform::rskp_route_release ( type(rskp_route_type), intent(inout) route)

Release backend-owned geometry and FFT work buffers from a route.

Parameters
routeroute whose backend plan is no longer needed

Definition at line 1650 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ cell_to_k_sparse_fft()

subroutine, public kpoint_transform::cell_to_k_sparse_fft ( type(rskp_values_type), intent(in), target snapshot,
type(rskp_route_type), intent(inout) route,
real(kind=dp), dimension(:, :), intent(in) xkp,
integer, dimension(3), intent(in) nkp_grid,
integer, dimension(:, :), intent(in) kp_dist,
type(mp_para_env_type), intent(in) inter,
integer(kind=int_8), intent(in) max_bytes,
complex(kind=dp), dimension(:, :), intent(out), optional, contiguous cache,
type(cp_fm_p_type), dimension(:, :), intent(in), optional output )

FFT captured raw-element panels and retain the destination group's k points. Reuse the sparse route and the dense layout consumed by the batched solvers.

Parameters
snapshotimmutable packed raw-image panels
routefixed element/image layout and equivalent target tiles across groups
xkpfull reciprocal grid
nkp_gridreciprocal-grid dimensions
kp_distglobal k-point range of each group
intercommunicator joining equivalent FM ranks across groups
max_bytessoft workspace target, excluding retained inputs and destinations
cacheoptional unscaled canonical elements by local k point
outputoptional existing split-FM destinations, instead of cache

Definition at line 1671 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ cell_to_k_backend_prepare()

subroutine, public kpoint_transform::cell_to_k_backend_prepare ( type(rskp_route_type), intent(inout) route,
real(kind=dp), dimension(:, :), intent(in) xkp,
integer, dimension(3), intent(in) nkp_grid,
integer(kind=int_8), intent(in) max_bytes )

Prepare the selected R-to-K backend plan once for a fixed route. Both backends retain their geometry with the route; the snapshots then provide only values and use the common evaluate interface.

Parameters
routefixed sparse AO route and backend workspace
xkpreciprocal coordinates for the route's k-point domain
nkp_gridregular-grid dimensions required by the FFT backend
max_bytesphase-cache budget for the GEMM backend

Definition at line 1966 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ rskp_batch_size()

subroutine, public kpoint_transform::rskp_batch_size ( type(rskp_route_type), intent(in) route,
integer, intent(in) nk,
integer, intent(in) ncomp,
integer, intent(inout) nworkers,
integer, intent(in) nmatrices,
integer(kind=int_8), intent(in) max_bytes,
integer, intent(out) width,
integer(kind=int_8), intent(out) work_size )

Bound worker scratch and output tiles for either prepared backend. Packed input belongs to the snapshot and is shared by every k tile.

Parameters
routeborrowed local AO-element layout
nklocal k-point count
ncompcontraction components for split-FM outputs (two for a complex source scale)
nworkersrequested workers, reduced to fit the workspace
nmatricesretained complex operator tiles; zero writes existing split-FM outputs
max_bytesshared-phase, worker-scratch and output-tile budget
widthk points per worker
work_sizereal scratch elements per worker

Definition at line 2099 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ cell_to_k_backend_evaluate_batch()

subroutine, public kpoint_transform::cell_to_k_backend_evaluate_batch ( type(rskp_values_type), intent(in), target snapshot,
type(rskp_route_type), intent(in) route,
real(kind=dp), dimension(:, :), intent(in) xkp,
integer, intent(in) first_local,
real(kind=dp), dimension(:), intent(inout), target, contiguous work,
complex(kind=dp), dimension(:, :, :), intent(out), optional, contiguous batch,
type(cp_fm_p_type), dimension(:, :), intent(in), optional output )

Evaluate a k-point tile through the selected prepared R-to-K backend. GEMM contracts the retained real-space panel; FFT returns a slice of the canonical reciprocal-space snapshot materialized during capture.

Parameters
snapshotbackend input or prepared reciprocal-space values
routefixed sparse layout and backend plan
xkpreciprocal coordinates for this k tile
first_localfirst k index in the route-local domain
workcaller-owned GEMM scratch; unused by the FFT backend
batchcomplex output tile (row, column, k point)
outputoptional split-FM output tile (k point, component), instead of batch

Definition at line 2243 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ k_grid_to_cell_phase_prepare()

subroutine, public kpoint_transform::k_grid_to_cell_phase_prepare ( type(k_grid_to_cell_phase_type), intent(out) table,
integer, dimension(:, :), intent(in) cells,
real(kind=dp), dimension(:, :), intent(in) xkp,
integer, intent(in) nworkers )

Tabulate distinct absolute translation coordinates for a group's local k points. All AO pairs share these unweighted phases. Zero coordinates are exact; negative coordinates use the same cosine and the opposite sine.

Parameters
tableowned axis-coordinate dictionary and phases
cellsglobally available image translations
xkpthis symmetry group's local contribution coordinates
nworkersmaximum number of OpenMP workers

Definition at line 2303 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ k_grid_to_cell_contract()

subroutine, public kpoint_transform::k_grid_to_cell_contract ( type(cp_fm_p_type), dimension(:, :, :), intent(in) source,
integer, dimension(:, :, :, :, :), intent(in) tiles,
integer, dimension(:), intent(in) ikp_index,
real(kind=dp), dimension(:, :), intent(in) xkp,
real(kind=dp), dimension(:), intent(in) weight,
integer, dimension(:, :), intent(in) images,
integer, dimension(2), intent(in) source_pair,
integer, dimension(2), intent(in) block_shape,
integer, dimension(2), intent(in) block_offset,
integer, dimension(3), intent(in) shift,
logical, intent(in) time_reversal,
logical, intent(in) trans,
type(k_grid_to_cell_phase_type), intent(in) phase_table,
real(kind=dp), dimension(:, :, :), intent(inout), allocatable sums,
type(local_gemm_ctxt_type), dimension(:), intent(inout) gemm_ctx )

Weighted K-to-R contraction of one AO block, before spatial rotations and MPI. The caller supplies the atom mapping and image multiplicities. Phases include the gauge shift, time reversal and conjugation of a transposed source block.

Parameters
sourceborrowed local matrices by k point, real/imaginary component and spin
tileslocal FM intervals (first/last row and column) by atom, k, component and spin
ikp_indexlocal source indices for this symmetry contribution
xkpreciprocal coordinates for these contributions
weightoriginal k-point weights, including symmetry multiplicity
imagestranslations, real multiplicities and imaginary sign sums
source_pairsource AO atom pair
block_shapesource AO row and column sizes
block_offsetglobal first row and column of the source block
shiftcolumn-minus-row gauge shift
time_reversalconjugate the source before applying its gauge phase
transconjugate the imaginary contribution on transposition
phase_tablegroup-local unweighted one-dimensional phases
sumsallocated flattened source blocks by image and spin
gemm_ctxworker-owned host GEMM contexts

Definition at line 2381 of file kpoint_transform.F.

Here is the call graph for this function:
Here is the caller graph for this function:

◆ rskp_phase_factors()

subroutine, public kpoint_transform::rskp_phase_factors ( integer, dimension(3), intent(in) shift,
real(kind=dp), dimension(3), intent(in) xkp,
logical, intent(in) time_reversal,
logical, intent(in) real_only,
real(kind=dp), dimension(2, 2), intent(out) phase )

Real two-component representation of the Bloch phase and time reversal.

Parameters
shiftcolumn-minus-row cell shift with the selected phase direction
xkptarget k-point coordinates
time_reversalconjugate the source before applying the phase
real_onlyreject phases not representable by real wavefunctions
phasecoefficients mapping source [Re,Im] to transformed [Re,Im]

Definition at line 2656 of file kpoint_transform.F.

Here is the caller graph for this function:

◆ rskp_work_matrix()

subroutine, public kpoint_transform::rskp_work_matrix ( real(kind=dp), dimension(:, :), intent(inout), allocatable work,
integer, intent(in) nrow,
integer, intent(in) ncol )

Allocate a dense work matrix with the requested shape.

Parameters
workdense work matrix
nrownumber of rows
ncolnumber of columns

Definition at line 2680 of file kpoint_transform.F.

Here is the caller graph for this function: