module_rt::gpu Namespace Reference

Functions
void	initialize_gpu_resources ()
	Initialize GPU resources (copy grids to constant memory) Should be called at the start of each calculate_HR.
void	finalize_gpu_resources ()
	Release GPU resources (clear any error states) Should be called at the end of each calculate_HR.
void	snap_psibeta_atom_batch_gpu (const LCAO_Orbitals &orb, const InfoNonlocal &infoNL_, const int T0, const ModuleBase::Vector3< double > &R0, const ModuleBase::Vector3< double > &A, const AdjacentAtomInfo &adjs, const UnitCell *ucell, const Parallel_Orbitals *paraV, const int npol, const int nlm_dim, std::vector< std::vector< std::unordered_map< int, std::vector< std::complex< double > > > > > &nlm_tot)
	Atom-level GPU batch processing interface.

Function Documentation

finalize_gpu_resources()

void module_rt::gpu::finalize_gpu_resources

(

)

Release GPU resources (clear any error states) Should be called at the end of each calculate_HR.

initialize_gpu_resources()

void module_rt::gpu::initialize_gpu_resources

(

)

Initialize GPU resources (copy grids to constant memory) Should be called at the start of each calculate_HR.

snap_psibeta_atom_batch_gpu()

void module_rt::gpu::snap_psibeta_atom_batch_gpu	(	const LCAO_Orbitals &	orb,
		const InfoNonlocal &	infoNL_,
		const int	T0,
		const ModuleBase::Vector3< double > &	R0,
		const ModuleBase::Vector3< double > &	A,
		const AdjacentAtomInfo &	adjs,
		const UnitCell *	ucell,
		const Parallel_Orbitals *	paraV,
		const int	npol,
		const int	nlm_dim,
		std::vector< std::vector< std::unordered_map< int, std::vector< std::complex< double > > > > > &	nlm_tot
	)

Atom-level GPU batch processing interface.

Processes ALL neighbors for a center atom in a SINGLE kernel launch. This significantly reduces kernel launch overhead compared to neighbor-level batching.

Parameters

orb	Orbital information
infoNL_	Non-local pseudopotential information
T0	Center atom type (projector location)
R0	Center atom position (already multiplied by lat0)
A	Vector potential
adjs	Adjacent atom information for this center atom
ucell	Unit cell pointer
paraV	Parallel orbitals information
npol	Polarization number
nlm_dim	1 for no current, 4 for current calculation
nlm_tot	Output: nlm_tot[ad][dir][iw_index] = nlm_vector