PmeShared Struct Reference

#include <gromacs/ewald/pme_gpu_types_host.h>

Collaboration diagram for PmeShared:

Description

The PME GPU structure for all the data copied directly from the CPU PME structure. The copying is done when the CPU PME structure is already (re-)initialized (pme_gpu_reinit is called at the end of gmx_pme_init). All the variables here are named almost the same way as in gmx_pme_t. The types are different: pointers are replaced by vectors. TODO: use the shared data with the PME CPU. Included in the main PME GPU structure by value.

Public Attributes
int	ngrids
	Grid count.
int	nk [DIM]
	Grid dimensions - nkx, nky, nkz.
int	pme_order
	PME interpolation order.
real	ewaldcoeff_q
	Ewald splitting coefficient for Coulomb.
real	epsilon_r
	Electrostatics parameter.
std::vector< int >	nn
	Gridline indices - nnx, nny, nnz.
std::vector< real >	fsh
	Fractional shifts - fshx, fshy, fshz.
std::vector< real >	bsp_mod [DIM]
	Precomputed B-spline values.
PmeRunMode	runMode
	The PME codepath being taken.
bool	isRankPmeOnly
	Whether PME execution is happening on a PME-only rank (from gmx_pme_t.bPPnode).
class EwaldBoxZScaler *	boxScaler
	The box scaler based on inputrec - created in pme_init and managed by CPU structure.
matrix	previousBox
	The previous computation box to know if we even need to update the current box params. More...
int	ndecompdim
	The The number of decomposition dimensions.
int	nodeid
	MPI rank within communicator.
int	nnodes
	Number of MPI ranks doing PME.
int	nodeidX
	MPI rank within communicator for PME X-decomposition.
int	nodeidY
	MPI rank within communicator for PME Y-decomposition.
int	nnodesX
	Number of MPI ranks in X-decomposition.
int	nnodesY
	Number of MPI ranks in Y-decomposition.
MPI_Comm	mpiComm
	MPI communicator for PME ranks.
MPI_Comm	mpiCommX
	MPI communicator for ranks in X-decomposition.
MPI_Comm	mpiCommY
	MPI communicator for ranks in Y-decomposition.
std::vector< int >	s2g0X
	local interpolation grid start values in x-dimension
std::vector< int >	s2g1X
	local interpolation grid end values in x-dimension
std::vector< int >	s2g0Y
	local interpolation grid start values in y-dimension
std::vector< int >	s2g1Y
	local interpolation grid end values in y-dimension
std::array< int, DIM >	pmegridNk
	local grid size
int	gridHalo
	Size of the grid halo region.

Member Data Documentation

matrix PmeShared::previousBox

The previous computation box to know if we even need to update the current box params.

Todo:

Manage this on higher level.

Alternatively, when this structure is used by CPU PME code, make use of this field there as well.

---

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

• src/gromacs/ewald/pme_gpu_types_host.h