Gromacs  2025-dev-20240718-982dfdc
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
List of all members | Classes | Public Member Functions | Public Attributes
nbnxn_atomdata_t Struct Reference

#include <gromacs/nbnxm/atomdata.h>


Struct that stores atom related data for the nbnxn module.

Note: performance would improve slightly when all std::vector containers in this struct would not initialize during resize().


struct  Params
 The actual atom data parameter values. More...
struct  SimdMasks
 Diagonal and topology exclusion helper data for all SIMD kernels. More...

Public Member Functions

 nbnxn_atomdata_t (gmx::PinningPolicy pinningPolicy)
 nbnxn_atomdata_t (gmx::PinningPolicy pinningPolicy, const gmx::MDLogger &mdlog, Nbnxm::KernelType kernelType, const std::optional< LJCombinationRule > &ljCombinationRule, LJCombinationRule pmeLJCombinationRule, gmx::ArrayRef< const real > nbfp, bool addFillerAtomType, int numEnergyGroups, int numOutputBuffers)
 Constructor. More...
const Paramsparams () const
 Returns a const reference to the parameters.
ParamsparamsDeprecated ()
 Returns a non-const reference to the parameters.
int numAtoms () const
 Returns the current total number of atoms stored.
int numLocalAtoms () const
 Returns the number of local atoms.
gmx::ArrayRef< const realx () const
 Return the coordinate buffer, and q with xFormat==nbatXYZQ.
gmx::ArrayRef< realx ()
 Return the coordinate buffer, and q with xFormat==nbatXYZQ.
const SimdMaskssimdMasks () const
 Masks for handling exclusions in the SIMD kernels.
void resizeCoordinateBuffer (int numAtoms, int domainDecompositionZone)
 Resizes the coordinate buffer and sets the number of atoms. More...
void resizeForceBuffers ()
 Resizes the force buffers for the current number of atoms.
nbnxn_atomdata_output_toutputBuffer (int thread)
 Returns the output buffer for the given thread.
const nbnxn_atomdata_output_toutputBuffer (int thread) const
 Returns the output buffer for the given thread.
gmx::ArrayRef< const
outputBuffers () const
 Returns the list of output buffers.
bool useBufferFlags () const
 Returns whether buffer flags are used.
std::vector< gmx_bitmask_t > & bufferFlags ()
 Returns the vector of buffer flags.
void reduceForces (gmx::AtomLocality locality, const Nbnxm::GridSet &gridSet, rvec *totalForce)
 Add the computed forces to f, an internal reduction might be performed as well. More...
void clearForceBuffer (int outputIndex)
 Clears the force buffer. More...

Public Attributes

int XFormat
 The format of x (and q), enum.
int FFormat
 The format of f, enum.
bool bDynamicBox
 Do we need to update shift_vec every step?
gmx::HostVector< gmx::RVecshift_vec
 Shift vectors, copied from t_forcerec.
int xstride
 stride for a coordinate in x (usually 3 or 4)
int fstride
 stride for a coordinate in f (usually 3 or 4)

Constructor & Destructor Documentation

nbnxn_atomdata_t::nbnxn_atomdata_t ( gmx::PinningPolicy  pinningPolicy,
const gmx::MDLogger mdlog,
Nbnxm::KernelType  kernelType,
const std::optional< LJCombinationRule > &  ljCombinationRule,
LJCombinationRule  pmeLJCombinationRule,
gmx::ArrayRef< const real nbfp,
bool  addFillerAtomType,
int  numEnergyGroups,
int  numOutputBuffers 


This class only stores one, or no, LJ combination rule parameter list. With LJ-PME the rule for the LJ PME-grid must match the PME grid combination rule and there can be no combination rule for the LJ pair parameters. Without LJ-PME the combination rule should match the combination rule for the LJ parameters. An (release) assertion failure will occur when these conditions are not met.

The non-bonded force parameter matrix nbfp is a serialized version of a square parameter matrix with a pair of parameters 6*C6, 12*C12 for every atom type pair.

[in]pinningPolicySets the pinning policy for all data that might be transferred to a GPU
[in]mdlogThe logger
[in]kernelTypeNonbonded NxN kernel type
[in]ljCombinationRuleThe LJ combination rule parameters to generate, empty is detect from the LJ parameters
[in]pmeLJCombinationRuleThe LJ combination rule parameters to generate for the LJ PME-grid part
[in]nbfpNon-bonded force parameter matrix
[in]addFillerAtomTypeWhen true, add a filler atom type, when false, nbfp should have atom-type with index numTypes-1 with all parameters zero so that that row and column have only zero values.
[in]numEnergyGroupsNumber of energy groups
[in]numOutputBuffersNumber of output data structures

Member Function Documentation

void nbnxn_atomdata_t::clearForceBuffer ( int  outputIndex)

Clears the force buffer.

Either the whole buffer is cleared or only the parts used by thread/task outputIndex when useBufferFlags() returns true.

[in]outputIndexThe index of the output object to clear
void nbnxn_atomdata_t::reduceForces ( gmx::AtomLocality  locality,
const Nbnxm::GridSet gridSet,
rvec totalForce 

Add the computed forces to f, an internal reduction might be performed as well.

[in]localityIf the reduction should be performed on local or non-local atoms.
[in]gridSetThe grids data.
[out]totalForceBuffer to accumulate resulting force
void nbnxn_atomdata_t::resizeCoordinateBuffer ( int  numAtoms,
int  domainDecompositionZone 

Resizes the coordinate buffer and sets the number of atoms.

numAtomsThe new number of atoms
domainDecompositionZoneThe domain decomposition zone index

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