Skip to content

GitLab

"platforms/opencl/vscode:/vscode.git/clone" did not exist on "a885a6ae6bc14390f9540d383fbad8a58a087998"
Commit 53f770f4 authored by peastman's avatar peastman Committed by GitHub
Browse files

Merge pull request #1557 from andysim/dpme 

Dispersion PME
parents f36f36bf c0850062
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 957 additions and 198 deletions
docs-source/usersguide/application.rst
View file @ 53f770f4
......@@ -916,8 +916,9 @@ Value Meaning
:code:`NoCutoff` No cutoff is applied.
:code:`CutoffNonPeriodic` The reaction field method is used to eliminate all interactions beyond a cutoff distance. Not valid for AMOEBA.
:code:`CutoffPeriodic` The reaction field method is used to eliminate all interactions beyond a cutoff distance. Periodic boundary conditions are applied, so each atom interacts only with the nearest periodic copy of every other atom. Not valid for AMOEBA.
:code:`Ewald` Periodic boundary conditions are applied. Ewald summation is used to compute long range interactions. (This option is rarely used, since PME is much faster for all but the smallest systems.) Not valid for AMOEBA.
:code:`PME` Periodic boundary conditions are applied. The Particle Mesh Ewald method is used to compute long range interactions.
:code:`Ewald` Periodic boundary conditions are applied. Ewald summation is used to compute long range Coulomb interactions. (This option is rarely used, since PME is much faster for all but the smallest systems.) Not valid for AMOEBA.
:code:`PME` Periodic boundary conditions are applied. The Particle Mesh Ewald method is used to compute long range Coulomb interactions.
:code:`LJPME` Periodic boundary conditions are applied. The Particle Mesh Ewald method is used to compute long range interactions for both Coulomb and Lennard-Jones.
========================= ===========================================================================================================================================================================================================================================
......@@ -926,7 +927,7 @@ cutoff distance. Be sure to specify units, as shown in the examples above. For
example, :code:`nonbondedCutoff=1.5*nanometers` or
:code:`nonbondedCutoff=12*angstroms` are legal values.
When using :code:`Ewald` or :code:`PME`\ , you can optionally specify an
When using :code:`Ewald`, :code:`PME`, or :code:`LJPME`\ , you can optionally specify an
error tolerance for the force computation. For example:
::
......
docs-source/usersguide/references.bib
View file @ 53f770f4
......@@ -517,3 +517,15 @@
year = {2014},
type = {Journal Article}
}
@article{Wennberg2015
author = {Wennberg, Christian L. and Murtola, Teemu and Páll, Szilárd and Abraham, Mark J. and Hess, Berk and Lindahl, Erik},
title = {Direct-Space Corrections Enable Fast and Accurate {Lorentz–Berthelot} Combination Rule {Lennard-Jones} Lattice Summation},
journal = {Journal of Chemical Theory and Computation},
volume = {11},
number = {12},
pages = {5737-5746},
year = {2015},
type = {Journal Article}
}
docs-source/usersguide/theory.rst
View file @ 53f770f4
......@@ -412,7 +412,7 @@ and the number of nodes in the mesh along each dimension as
.. math::
n_\mathit{mesh}=\frac{2\alpha d}{{3d}^{1/5}}
n_\mathit{mesh}=\frac{2\alpha d}{{3\delta}^{1/5}}
where *d* is the width of the periodic box along that dimension. Alternatively,
......@@ -432,6 +432,38 @@ to numerical round-off error than Ewald summation. For Platforms that do
calculations in single precision, making :math:`\delta` too small (typically below about
5·10\ :sup:`-5`\ ) can actually cause the error to increase.
Lennard-Jones Interaction With Particle Mesh Ewald
==================================================
The PME algorithm can also be used for Lennard-Jones interactions. Usually this
is not necessary, since Lennard-Jones forces are short ranged, but there are
situations (such as membrane simulations) where neglecting interactions beyond
the cutoff can measurably affect results.
For computational efficiency, certain approximations are made\ :cite:`Wennberg2015`.
Interactions beyond the cutoff distance include only the attractive :math:`1/r^6`
term, not the repulsive :math:`1/r^{12}` term. Since the latter is much smaller
than the former at long distances, this usually has negligible effect. Also,
the interaction between particles farther apart than the cutoff distance is
computed using geometric combination rules:
.. math::
\sigma=\sqrt{\sigma_1 \sigma_2}
The effect of this approximation is also quite small, and it is still far more
accurate than ignoring the interactions altogether (which is what would happen
with PME).
The formula used to compute the number of nodes along each dimension of the mesh
is slightly different from the one used for Coulomb interactions:
.. math::
n_\mathit{mesh}=\frac{\alpha d}{{3\delta}^{1/5}}
As before, this is an empirical formula. It will usually produce an average
relative error in the forces less than or similar to :math:`\delta`\ , but that
is not guaranteed.
.. _gbsaobcforce:
GBSAOBCForce
......
olla/include/openmm/kernels.h
View file @ 53f770f4
......@@ -555,7 +555,8 @@ public:
CutoffNonPeriodic = 1,
CutoffPeriodic = 2,
Ewald = 3,
PME = 4
PME = 4,
LJPME = 5
};
static std::string Name() {
return "CalcNonbondedForce";
......@@ -589,13 +590,22 @@ public:
virtual void copyParametersToContext(ContextImpl& context, const NonbondedForce& force) = 0;
/**
* Get the parameters being used for PME.
*
*
* @param alpha the separation parameter
* @param nx the number of grid points along the X axis
* @param ny the number of grid points along the Y axis
* @param nz the number of grid points along the Z axis
*/
virtual void getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const = 0;
/**
* Get the parameters being used for the dispersion terms in LJPME.
*
* @param alpha the separation parameter
* @param nx the number of grid points along the X axis
* @param ny the number of grid points along the Y axis
* @param nz the number of grid points along the Z axis
*/
virtual void getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const = 0;
};
/**
......@@ -1335,6 +1345,57 @@ public:
};
/**
* This kernel performs the dispersion reciprocal space calculation for LJPME. In most cases, this
* calculation is done directly by CalcNonbondedForceKernel so this kernel is unneeded.
* In some cases it may want to outsource the work to a different kernel. In particular,
* GPU based platforms sometimes use a CPU based implementation provided by a separate
* plugin.
*/
class CalcDispersionPmeReciprocalForceKernel : public KernelImpl {
public:
class IO;
static std::string Name() {
return "CalcDispersionPmeReciprocalForce";
}
CalcDispersionPmeReciprocalForceKernel(std::string name, const Platform& platform) : KernelImpl(name, platform) {
}
/**
* Initialize the kernel.
*
* @param gridx the x size of the PME grid
* @param gridy the y size of the PME grid
* @param gridz the z size of the PME grid
* @param numParticles the number of particles in the system
* @param alpha the Ewald blending parameter
*/
virtual void initialize(int gridx, int gridy, int gridz, int numParticles, double alpha) = 0;
/**
* Begin computing the force and energy.
*
* @param io an object that coordinates data transfer
* @param periodicBoxVectors the vectors defining the periodic box (measured in nm)
* @param includeEnergy true if potential energy should be computed
*/
virtual void beginComputation(IO& io, const Vec3* periodicBoxVectors, bool includeEnergy) = 0;
/**
* Finish computing the force and energy.
*
* @param io an object that coordinates data transfer
* @return the potential energy due to the PME reciprocal space interactions
*/
virtual double finishComputation(IO& io) = 0;
/**
* Get the parameters being used for PME.
*
* @param alpha the separation parameter
* @param nx the number of grid points along the X axis
* @param ny the number of grid points along the Y axis
* @param nz the number of grid points along the Z axis
*/
virtual void getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const = 0;
};
} // namespace OpenMM
#endif /*OPENMM_KERNELS_H_*/
openmmapi/include/openmm/NonbondedForce.h
View file @ 53f770f4
......@@ -101,15 +101,20 @@ public:
*/
CutoffPeriodic = 2,
/**
* Periodic boundary conditions are used, and Ewald summation is used to compute the interaction of each particle
* Periodic boundary conditions are used, and Ewald summation is used to compute the Coulomb interaction of each particle
* with all periodic copies of every other particle.
*/
Ewald = 3,
/**
* Periodic boundary conditions are used, and Particle-Mesh Ewald (PME) summation is used to compute the interaction of each particle
* Periodic boundary conditions are used, and Particle-Mesh Ewald (PME) summation is used to compute the Coulomb interaction of each particle
* with all periodic copies of every other particle.
*/
PME = 4
PME = 4,
/**
* Periodic boundary conditions are used, and Particle-Mesh Ewald (PME) summation is used to compute the interaction of each particle
* with all periodic copies of every other particle for both Coulomb and Lennard-Jones. No switching is used for either interaction.
*/
LJPME = 5
};
/**
* Create a NonbondedForce.
......@@ -207,6 +212,16 @@ public:
* @param[out] nz the number of grid points along the Z axis
*/
void getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
/**
* Get the parameters to use for dispersion term in LJ-PME calculations. If alpha is 0 (the default),
* these parameters are ignored and instead their values are chosen based on the Ewald error tolerance.
*
* @param[out] alpha the separation parameter
* @param[out] nx the number of dispersion grid points along the X axis
* @param[out] ny the number of dispersion grid points along the Y axis
* @param[out] nz the number of dispersion grid points along the Z axis
*/
void getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
/**
* Set the parameters to use for PME calculations. If alpha is 0 (the default), these parameters are
* ignored and instead their values are chosen based on the Ewald error tolerance.
......@@ -217,6 +232,16 @@ public:
* @param nz the number of grid points along the Z axis
*/
void setPMEParameters(double alpha, int nx, int ny, int nz);
/**
* Set the parameters to use for the dispersion term in LJPME calculations. If alpha is 0 (the default),
* these parameters are ignored and instead their values are chosen based on the Ewald error tolerance.
*
* @param alpha the separation parameter
* @param nx the number of grid points along the X axis
* @param ny the number of grid points along the Y axis
* @param nz the number of grid points along the Z axis
*/
void setLJPMEParameters(double alpha, int nx, int ny, int nz);
/**
* Get the parameters being used for PME in a particular Context. Because some platforms have restrictions
* on the allowed grid sizes, the values that are actually used may be slightly different from those
......@@ -230,6 +255,19 @@ public:
* @param[out] nz the number of grid points along the Z axis
*/
void getPMEParametersInContext(const Context& context, double& alpha, int& nx, int& ny, int& nz) const;
/**
* Get the PME parameters being used for the dispersion term for LJPME in a particular Context. Because some
* platforms have restrictions on the allowed grid sizes, the values that are actually used may be slightly different
* from those specified with setPMEParameters(), or the standard values calculated based on the Ewald error tolerance.
* See the manual for details.
*
* @param context the Context for which to get the parameters
* @param[out] alpha the separation parameter
* @param[out] nx the number of grid points along the X axis
* @param[out] ny the number of grid points along the Y axis
* @param[out] nz the number of grid points along the Z axis
*/
void getLJPMEParametersInContext(const Context& context, double& alpha, int& nx, int& ny, int& nz) const;
/**
* Add the nonbonded force parameters for a particle. This should be called once for each particle
* in the System. When it is called for the i'th time, it specifies the parameters for the i'th particle.
......@@ -374,7 +412,8 @@ public:
bool usesPeriodicBoundaryConditions() const {
return nonbondedMethod == NonbondedForce::CutoffPeriodic ||
nonbondedMethod == NonbondedForce::Ewald ||
nonbondedMethod == NonbondedForce::PME;
nonbondedMethod == NonbondedForce::PME ||
nonbondedMethod == NonbondedForce::LJPME;
}
protected:
ForceImpl* createImpl() const;
......@@ -382,9 +421,9 @@ private:
class ParticleInfo;
class ExceptionInfo;
NonbondedMethod nonbondedMethod;
double cutoffDistance, switchingDistance, rfDielectric, ewaldErrorTol, alpha;
double cutoffDistance, switchingDistance, rfDielectric, ewaldErrorTol, alpha, dalpha;
bool useSwitchingFunction, useDispersionCorrection;
int recipForceGroup, nx, ny, nz;
int recipForceGroup, nx, ny, nz, dnx, dny, dnz;
void addExclusionsToSet(const std::vector<std::set<int> >& bonded12, std::set<int>& exclusions, int baseParticle, int fromParticle, int currentLevel) const;
std::vector<ParticleInfo> particles;
std::vector<ExceptionInfo> exceptions;
......
openmmapi/include/openmm/internal/NonbondedForceImpl.h
View file @ 53f770f4
......@@ -65,6 +65,7 @@ public:
std::vector<std::string> getKernelNames();
void updateParametersInContext(ContextImpl& context);
void getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
void getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
/**
* This is a utility routine that calculates the values to use for alpha and kmax when using
* Ewald summation.
......@@ -74,7 +75,7 @@ public:
* This is a utility routine that calculates the values to use for alpha and grid size when using
* Particle Mesh Ewald.
*/
static void calcPMEParameters(const System& system, const NonbondedForce& force, double& alpha, int& xsize, int& ysize, int& zsize);
static void calcPMEParameters(const System& system, const NonbondedForce& force, double& alpha, int& xsize, int& ysize, int& zsize, bool lj);
/**
* Compute the coefficient which, when divided by the periodic box volume, gives the
* long range dispersion correction to the energy.
......
openmmapi/src/NonbondedForce.cpp
View file @ 53f770f4
......@@ -48,7 +48,8 @@ using std::stringstream;
using std::vector;
NonbondedForce::NonbondedForce() : nonbondedMethod(NoCutoff), cutoffDistance(1.0), switchingDistance(-1.0), rfDielectric(78.3),
ewaldErrorTol(5e-4), alpha(0.0), useSwitchingFunction(false), useDispersionCorrection(true), recipForceGroup(-1), nx(0), ny(0), nz(0) {
ewaldErrorTol(5e-4), alpha(0.0), dalpha(0.0), useSwitchingFunction(false), useDispersionCorrection(true), recipForceGroup(-1),
nx(0), ny(0), nz(0), dnx(0), dny(0), dnz(0) {
}
NonbondedForce::NonbondedMethod NonbondedForce::getNonbondedMethod() const {
......@@ -106,6 +107,13 @@ void NonbondedForce::getPMEParameters(double& alpha, int& nx, int& ny, int& nz)
nz = this->nz;
}
void NonbondedForce::getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const {
alpha = this->dalpha;
nx = this->dnx;
ny = this->dny;
nz = this->dnz;
}
void NonbondedForce::setPMEParameters(double alpha, int nx, int ny, int nz) {
this->alpha = alpha;
this->nx = nx;
......@@ -113,10 +121,21 @@ void NonbondedForce::setPMEParameters(double alpha, int nx, int ny, int nz) {
this->nz = nz;
}
void NonbondedForce::setLJPMEParameters(double alpha, int nx, int ny, int nz) {
this->dalpha = alpha;
this->dnx = nx;
this->dny = ny;
this->dnz = nz;
}
void NonbondedForce::getPMEParametersInContext(const Context& context, double& alpha, int& nx, int& ny, int& nz) const {
dynamic_cast<const NonbondedForceImpl&>(getImplInContext(context)).getPMEParameters(alpha, nx, ny, nz);
}
void NonbondedForce::getLJPMEParametersInContext(const Context& context, double& alpha, int& nx, int& ny, int& nz) const {
dynamic_cast<const NonbondedForceImpl&>(getImplInContext(context)).getLJPMEParameters(alpha, nx, ny, nz);
}
int NonbondedForce::addParticle(double charge, double sigma, double epsilon) {
particles.push_back(ParticleInfo(charge, sigma, epsilon));
return particles.size()-1;
......
openmmapi/src/NonbondedForceImpl.cpp
View file @ 53f770f4
......@@ -90,9 +90,7 @@ void NonbondedForceImpl::initialize(ContextImpl& context) {
exceptions[particle1].insert(particle2);
exceptions[particle2].insert(particle1);
}
if (owner.getNonbondedMethod() == NonbondedForce::CutoffPeriodic ||
owner.getNonbondedMethod() == NonbondedForce::Ewald ||
owner.getNonbondedMethod() == NonbondedForce::PME) {
if (owner.getNonbondedMethod() != NonbondedForce::NoCutoff && owner.getNonbondedMethod() != NonbondedForce::CutoffNonPeriodic) {
Vec3 boxVectors[3];
system.getDefaultPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
double cutoff = owner.getCutoffDistance();
......@@ -151,19 +149,29 @@ void NonbondedForceImpl::calcEwaldParameters(const System& system, const Nonbond
kmaxz++;
}
void NonbondedForceImpl::calcPMEParameters(const System& system, const NonbondedForce& force, double& alpha, int& xsize, int& ysize, int& zsize) {
force.getPMEParameters(alpha, xsize, ysize, zsize);
void NonbondedForceImpl::calcPMEParameters(const System& system, const NonbondedForce& force, double& alpha, int& xsize, int& ysize, int& zsize, bool lj) {
if (lj)
force.getLJPMEParameters(alpha, xsize, ysize, zsize);
else
force.getPMEParameters(alpha, xsize, ysize, zsize);
if (alpha == 0.0) {
Vec3 boxVectors[3];
system.getDefaultPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
double tol = force.getEwaldErrorTolerance();
alpha = (1.0/force.getCutoffDistance())*std::sqrt(-log(2.0*tol));
xsize = (int) ceil(2*alpha*boxVectors[0][0]/(3*pow(tol, 0.2)));
ysize = (int) ceil(2*alpha*boxVectors[1][1]/(3*pow(tol, 0.2)));
zsize = (int) ceil(2*alpha*boxVectors[2][2]/(3*pow(tol, 0.2)));
xsize = max(xsize, 5);
ysize = max(ysize, 5);
zsize = max(zsize, 5);
if (lj) {
xsize = (int) ceil(alpha*boxVectors[0][0]/(3*pow(tol, 0.2)));
ysize = (int) ceil(alpha*boxVectors[1][1]/(3*pow(tol, 0.2)));
zsize = (int) ceil(alpha*boxVectors[2][2]/(3*pow(tol, 0.2)));
}
else {
xsize = (int) ceil(2*alpha*boxVectors[0][0]/(3*pow(tol, 0.2)));
ysize = (int) ceil(2*alpha*boxVectors[1][1]/(3*pow(tol, 0.2)));
zsize = (int) ceil(2*alpha*boxVectors[2][2]/(3*pow(tol, 0.2)));
}
xsize = max(xsize, 6);
ysize = max(ysize, 6);
zsize = max(zsize, 6);
}
}
......@@ -283,3 +291,7 @@ void NonbondedForceImpl::updateParametersInContext(ContextImpl& context) {
void NonbondedForceImpl::getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const {
kernel.getAs<CalcNonbondedForceKernel>().getPMEParameters(alpha, nx, ny, nz);
}
void NonbondedForceImpl::getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const {
kernel.getAs<CalcNonbondedForceKernel>().getLJPMEParameters(alpha, nx, ny, nz);
}
platforms/cpu/include/CpuKernels.h
View file @ 53f770f4
......@@ -249,27 +249,37 @@ public:
void copyParametersToContext(ContextImpl& context, const NonbondedForce& force);
/**
* Get the parameters being used for PME.
*
*
* @param alpha the separation parameter
* @param nx the number of grid points along the X axis
* @param ny the number of grid points along the Y axis
* @param nz the number of grid points along the Z axis
*/
void getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
/**
* Get the parameters being used for the dispersion term in LJPME.
*
* @param alpha the separation parameter
* @param nx the number of grid points along the X axis
* @param ny the number of grid points along the Y axis
* @param nz the number of grid points along the Z axis
*/
void getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
private:
class PmeIO;
CpuPlatform::PlatformData& data;
int numParticles, num14;
int **bonded14IndexArray;
double **bonded14ParamArray;
double nonbondedCutoff, switchingDistance, rfDielectric, ewaldAlpha, ewaldSelfEnergy, dispersionCoefficient;
int kmax[3], gridSize[3];
bool useSwitchingFunction, useOptimizedPme, hasInitializedPme;
double nonbondedCutoff, switchingDistance, rfDielectric, ewaldAlpha, ewaldDispersionAlpha, ewaldSelfEnergy, dispersionCoefficient;
int kmax[3], gridSize[3], dispersionGridSize[3];
bool useSwitchingFunction, useOptimizedPme, hasInitializedPme, hasInitializedDispersionPme;
std::vector<std::set<int> > exclusions;
std::vector<std::pair<float, float> > particleParams;
std::vector<float> C6params;
NonbondedMethod nonbondedMethod;
CpuNonbondedForce* nonbonded;
Kernel optimizedPme;
Kernel optimizedPme, optimizedDispersionPme;
CpuBondForce bondForce;
};
......
platforms/cpu/include/CpuNonbondedForce.h
View file @ 53f770f4
......@@ -103,16 +103,27 @@ class CpuNonbondedForce {
/**---------------------------------------------------------------------------------------
Set the force to use Particle-Mesh Ewald (PME) summation.
@param alpha the Ewald separation parameter
@param gridSize the dimensions of the mesh
--------------------------------------------------------------------------------------- */
void setUsePME(float alpha, int meshSize[3]);
/**---------------------------------------------------------------------------------------
Set the force to use Particle-Mesh Ewald (PME) summation for dispersion.
@param alpha the Ewald separation parameter
@param gridSize the dimensions of the mesh
--------------------------------------------------------------------------------------- */
void setUseLJPME(float alpha, int meshSize[3]);
/**---------------------------------------------------------------------------------------
Calculate Ewald ixn
......@@ -121,16 +132,17 @@ class CpuNonbondedForce {
@param posq atom coordinates and charges
@param atomCoordinates atom coordinates (in format needed by PME)
@param atomParameters atom parameters (sigma/2, 2*sqrt(epsilon))
@param C6Paramrs C6 parameters for multiplicative representation of dispersion
@param exclusions atom exclusion indices
exclusions[atomIndex] contains the list of exclusions for that atom
@param forces force array (forces added)
@param totalEnergy total energy
--------------------------------------------------------------------------------------- */
void calculateReciprocalIxn(int numberOfAtoms, float* posq, const std::vector<RealVec>& atomCoordinates,
const std::vector<std::pair<float, float> >& atomParameters, const std::vector<std::set<int> >& exclusions,
std::vector<RealVec>& forces, double* totalEnergy) const;
const std::vector<std::pair<float, float> >& atomParameters, const std::vector<float> &C6params,
const std::vector<std::set<int> >& exclusions, std::vector<RealVec>& forces, double* totalEnergy) const;
/**---------------------------------------------------------------------------------------
......@@ -149,7 +161,7 @@ class CpuNonbondedForce {
--------------------------------------------------------------------------------------- */
void calculateDirectIxn(int numberOfAtoms, float* posq, const std::vector<RealVec>& atomCoordinates, const std::vector<std::pair<float, float> >& atomParameters,
const std::vector<std::set<int> >& exclusions, std::vector<AlignedArray<float> >& threadForce, double* totalEnergy, ThreadPool& threads);
const std::vector<float>& C6params, const std::vector<std::set<int> >& exclusions, std::vector<AlignedArray<float> >& threadForce, double* totalEnergy, ThreadPool& threads);
/**
* This routine contains the code executed by each thread.
......@@ -162,28 +174,32 @@ protected:
bool periodic;
bool triclinic;
bool ewald;
bool pme;
bool tableIsValid;
bool ljpme, pme;
bool tableIsValid, expTableIsValid;
const CpuNeighborList* neighborList;
float recipBoxSize[3];
RealVec periodicBoxVectors[3];
AlignedArray<fvec4> periodicBoxVec4;
float cutoffDistance, switchingDistance;
float krf, crf;
float alphaEwald;
float alphaEwald, alphaDispersionEwald;
int numRx, numRy, numRz;
int meshDim[3];
int meshDim[3], dispersionMeshDim[3];
std::vector<float> erfcTable, ewaldScaleTable;
float ewaldDX, ewaldDXInv, erfcDXInv;
std::vector<float> exptermsTable, dExptermsTable;
float ewaldDX, ewaldDXInv, erfcDXInv, exptermsDX, exptermsDXInv;
std::vector<double> threadEnergy;
// The following variables are used to make information accessible to the individual threads.
int numberOfAtoms;
float* posq;
RealVec const* atomCoordinates;
std::pair<float, float> const* atomParameters;
std::pair<float, float> const* atomParameters;
float const *C6params;
std::set<int> const* exclusions;
std::vector<AlignedArray<float> >* threadForce;
bool includeEnergy;
float inverseRcut6;
float inverseRcut6Expterm;
void* atomicCounter;
static const float TWO_OVER_SQRT_PI;
......@@ -237,10 +253,29 @@ protected:
*/
void tabulateEwaldScaleFactor();
/**
* Create a lookup table for the scale factor used with dispersion PME.
*/
void tabulateExpTerms();
/**
* Compute a fast approximation to erfc(x).
*/
float erfcApprox(float x);
/**
* Compute a fast approximation to (1.0 - EXP(-dar^2) * (1.0 + dar^2 + 0.5*dar^4))
* where dar = (dispersionAlpha * R)
* needed for LJPME energies.
*/
float exptermsApprox(float R);
/**
* Compute a fast approximation to (1.0 - EXP(-dar^2) * (1.0 + dar^2 + 0.5*dar^4 + dar^6/6.0))
* where dar = (dispersionAlpha * R)
* needed for LJPME forces.
*/
float dExptermsApprox(float R);
};
} // namespace OpenMM
......
platforms/cpu/include/CpuNonbondedForceVec4.h
View file @ 53f770f4
......@@ -88,11 +88,25 @@ protected:
* Compute a fast approximation to erfc(x).
*/
fvec4 erfcApprox(const fvec4& x);
/**
* Evaluate the scale factor used with Ewald and PME: erfc(alpha*r) + 2*alpha*r*exp(-alpha*alpha*r*r)/sqrt(PI)
*/
fvec4 ewaldScaleFunction(const fvec4& x);
/**
* Compute a fast approximation to (1.0 - EXP(-dar^2) * (1.0 + dar^2 + 0.5*dar^4))
* where dar = (dispersionAlpha * R)
* needed for LJPME energies.
*/
fvec4 exptermsApprox(const fvec4& R);
/**
* Compute a fast approximation to (1.0 - EXP(-dar^2) * (1.0 + dar^2 + 0.5*dar^4 + dar^6/6.0))
* where dar = (dispersionAlpha * R)
* needed for LJPME forces.
*/
fvec4 dExptermsApprox(const fvec4& R);
};
} // namespace OpenMM
......
platforms/cpu/include/CpuNonbondedForceVec8.h
View file @ 53f770f4
......@@ -92,6 +92,21 @@ protected:
* Evaluate the scale factor used with Ewald and PME: erfc(alpha*r) + 2*alpha*r*exp(-alpha*alpha*r*r)/sqrt(PI)
*/
fvec8 ewaldScaleFunction(const fvec8& x);
/**
* Compute a fast approximation to (1.0 - EXP(-dar^2) * (1.0 + dar^2 + 0.5*dar^4))
* where dar = (dispersionAlpha * R)
* needed for LJPME energies.
*/
fvec8 exptermsApprox(const fvec8& R);
/**
* Compute a fast approximation to (1.0 - EXP(-dar^2) * (1.0 + dar^2 + 0.5*dar^4 + dar^6/6.0))
* where dar = (dispersionAlpha * R)
* needed for LJPME forces.
*/
fvec8 dExptermsApprox(const fvec8& R);
};
} // namespace OpenMM
......
platforms/cpu/src/CpuKernels.cpp
View file @ 53f770f4
......@@ -50,6 +50,7 @@
#include "lepton/CustomFunction.h"
#include "lepton/Operation.h"
#include "lepton/Parser.h"
#include <iostream>
#include "lepton/ParsedExpression.h"
using namespace OpenMM;
......@@ -508,7 +509,7 @@ CpuNonbondedForce* createCpuNonbondedForceVec4();
CpuNonbondedForce* createCpuNonbondedForceVec8();
CpuCalcNonbondedForceKernel::CpuCalcNonbondedForceKernel(string name, const Platform& platform, CpuPlatform::PlatformData& data) : CalcNonbondedForceKernel(name, platform),
data(data), bonded14IndexArray(NULL), bonded14ParamArray(NULL), hasInitializedPme(false), nonbonded(NULL) {
data(data), bonded14IndexArray(NULL), bonded14ParamArray(NULL), hasInitializedPme(false), hasInitializedDispersionPme(false), nonbonded(NULL) {
if (isVec8Supported())
nonbonded = createCpuNonbondedForceVec8();
else
......@@ -555,12 +556,14 @@ void CpuCalcNonbondedForceKernel::initialize(const System& system, const Nonbond
for (int i = 0; i < num14; i++)
bonded14ParamArray[i] = new double[3];
particleParams.resize(numParticles);
C6params.resize(numParticles);
double sumSquaredCharges = 0.0;
for (int i = 0; i < numParticles; ++i) {
double charge, radius, depth;
force.getParticleParameters(i, charge, radius, depth);
data.posq[4*i+3] = (float) charge;
particleParams[i] = make_pair((float) (0.5*radius), (float) (2.0*sqrt(depth)));
C6params[i] = 8.0*pow(particleParams[i].first, 3.0) * particleParams[i].second;
sumSquaredCharges += charge*charge;
}
......@@ -596,19 +599,35 @@ void CpuCalcNonbondedForceKernel::initialize(const System& system, const Nonbond
}
else if (nonbondedMethod == PME) {
double alpha;
NonbondedForceImpl::calcPMEParameters(system, force, alpha, gridSize[0], gridSize[1], gridSize[2]);
NonbondedForceImpl::calcPMEParameters(system, force, alpha, gridSize[0], gridSize[1], gridSize[2], false);
ewaldAlpha = alpha;
}
if (nonbondedMethod == Ewald || nonbondedMethod == PME)
else if (nonbondedMethod == LJPME) {
double alpha;
NonbondedForceImpl::calcPMEParameters(system, force, alpha, gridSize[0], gridSize[1], gridSize[2], false);
ewaldAlpha = (RealOpenMM) alpha;
NonbondedForceImpl::calcPMEParameters(system, force, alpha, dispersionGridSize[0], dispersionGridSize[1], dispersionGridSize[2], true);
ewaldDispersionAlpha = (RealOpenMM) alpha;
useSwitchingFunction = false;
}
if (nonbondedMethod == Ewald || nonbondedMethod == PME || nonbondedMethod == LJPME) {
ewaldSelfEnergy = -ONE_4PI_EPS0*ewaldAlpha*sumSquaredCharges/sqrt(M_PI);
else
if(nonbondedMethod == LJPME){
for (int atom = 0; atom < numParticles; atom++) {
// Dispersion self term
ewaldSelfEnergy += pow(ewaldDispersionAlpha, 6.0) * C6params[atom]*C6params[atom] / 12.0;
}
}
} else {
ewaldSelfEnergy = 0.0;
}
rfDielectric = force.getReactionFieldDielectric();
if (force.getUseDispersionCorrection())
dispersionCoefficient = NonbondedForceImpl::calcDispersionCorrection(system, force);
else
dispersionCoefficient = 0.0;
data.isPeriodic = (nonbondedMethod == CutoffPeriodic || nonbondedMethod == Ewald || nonbondedMethod == PME);
data.isPeriodic = (nonbondedMethod == CutoffPeriodic || nonbondedMethod == Ewald || nonbondedMethod == PME || nonbondedMethod == LJPME);
}
double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy, bool includeDirect, bool includeReciprocal) {
......@@ -626,6 +645,20 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
optimizedPme.getAs<CalcPmeReciprocalForceKernel>().initialize(gridSize[0], gridSize[1], gridSize[2], numParticles, ewaldAlpha);
}
}
if (nonbondedMethod == LJPME) {
// If available, use the optimized PME implementation.
vector<string> kernelNames;
kernelNames.push_back("CalcPmeReciprocalForce");
useOptimizedPme = getPlatform().supportsKernels(kernelNames);
if (useOptimizedPme) {
optimizedPme = getPlatform().createKernel(CalcPmeReciprocalForceKernel::Name(), context);
optimizedPme.getAs<CalcPmeReciprocalForceKernel>().initialize(gridSize[0], gridSize[1], gridSize[2], numParticles, ewaldAlpha);
optimizedDispersionPme = getPlatform().createKernel(CalcDispersionPmeReciprocalForceKernel::Name(), context);
optimizedDispersionPme.getAs<CalcDispersionPmeReciprocalForceKernel>().initialize(dispersionGridSize[0], dispersionGridSize[1],
dispersionGridSize[2], numParticles, ewaldDispersionAlpha);
}
}
}
AlignedArray<float>& posq = data.posq;
vector<RealVec>& posData = extractPositions(context);
......@@ -634,6 +667,7 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
double energy = (includeReciprocal ? ewaldSelfEnergy : 0.0);
bool ewald = (nonbondedMethod == Ewald);
bool pme = (nonbondedMethod == PME);
bool ljpme = (nonbondedMethod == LJPME);
if (nonbondedMethod != NoCutoff)
nonbonded->setUseCutoff(nonbondedCutoff, *data.neighborList, rfDielectric);
if (data.isPeriodic) {
......@@ -649,9 +683,13 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
nonbonded->setUsePME(ewaldAlpha, gridSize);
if (useSwitchingFunction)
nonbonded->setUseSwitchingFunction(switchingDistance);
if (ljpme){
nonbonded->setUsePME(ewaldAlpha, gridSize);
nonbonded->setUseLJPME(ewaldDispersionAlpha, dispersionGridSize);
}
double nonbondedEnergy = 0;
if (includeDirect)
nonbonded->calculateDirectIxn(numParticles, &posq[0], posData, particleParams, exclusions, data.threadForce, includeEnergy ? &nonbondedEnergy : NULL, data.threads);
nonbonded->calculateDirectIxn(numParticles, &posq[0], posData, particleParams, C6params, exclusions, data.threadForce, includeEnergy ? &nonbondedEnergy : NULL, data.threads);
if (includeReciprocal) {
if (useOptimizedPme) {
PmeIO io(&posq[0], &data.threadForce[0][0], numParticles);
......@@ -660,13 +698,13 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
nonbondedEnergy += optimizedPme.getAs<CalcPmeReciprocalForceKernel>().finishComputation(io);
}
else
nonbonded->calculateReciprocalIxn(numParticles, &posq[0], posData, particleParams, exclusions, forceData, includeEnergy ? &nonbondedEnergy : NULL);
nonbonded->calculateReciprocalIxn(numParticles, &posq[0], posData, particleParams, C6params, exclusions, forceData, includeEnergy ? &nonbondedEnergy : NULL);
}
energy += nonbondedEnergy;
if (includeDirect) {
ReferenceLJCoulomb14 nonbonded14;
bondForce.calculateForce(posData, bonded14ParamArray, forceData, includeEnergy ? &energy : NULL, nonbonded14);
if (data.isPeriodic)
if (data.isPeriodic && nonbondedMethod != LJPME)
energy += dispersionCoefficient/(boxVectors[0][0]*boxVectors[1][1]*boxVectors[2][2]);
}
return energy;
......@@ -719,7 +757,7 @@ void CpuCalcNonbondedForceKernel::copyParametersToContext(ContextImpl& context,
}
void CpuCalcNonbondedForceKernel::getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const {
if (nonbondedMethod != PME)
if (nonbondedMethod != PME && nonbondedMethod != LJPME)
throw OpenMMException("getPMEParametersInContext: This Context is not using PME");
if (useOptimizedPme)
optimizedPme.getAs<const CalcPmeReciprocalForceKernel>().getPMEParameters(alpha, nx, ny, nz);
......@@ -731,6 +769,19 @@ void CpuCalcNonbondedForceKernel::getPMEParameters(double& alpha, int& nx, int&
}
}
void CpuCalcNonbondedForceKernel::getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const {
if (nonbondedMethod != LJPME)
throw OpenMMException("getPMEParametersInContext: This Context is not using PME");
if (useOptimizedPme)
optimizedDispersionPme.getAs<const CalcPmeReciprocalForceKernel>().getPMEParameters(alpha, nx, ny, nz);
else {
alpha = ewaldDispersionAlpha;
nx = dispersionGridSize[0];
ny = dispersionGridSize[1];
nz = dispersionGridSize[2];
}
}
CpuCalcCustomNonbondedForceKernel::CpuCalcCustomNonbondedForceKernel(string name, const Platform& platform, CpuPlatform::PlatformData& data) :
CalcCustomNonbondedForceKernel(name, platform), data(data), forceCopy(NULL), nonbonded(NULL) {
}
......
platforms/cpu/src/CpuNonbondedForce.cpp
View file @ 53f770f4
......@@ -30,6 +30,7 @@
#include "ReferencePME.h"
#include "openmm/internal/gmx_atomic.h"
#include <algorithm>
#include <iostream>
// In case we're using some primitive version of Visual Studio this will
// make sure that erf() and erfc() are defined.
......@@ -47,7 +48,8 @@ const int CpuNonbondedForce::NUM_TABLE_POINTS = 2048;
--------------------------------------------------------------------------------------- */
CpuNonbondedForce::CpuNonbondedForce() : cutoff(false), useSwitch(false), periodic(false), ewald(false), pme(false), tableIsValid(false), cutoffDistance(0.0f), alphaEwald(0.0f) {
CpuNonbondedForce::CpuNonbondedForce() : cutoff(false), useSwitch(false), periodic(false), ewald(false), pme(false), ljpme(false), tableIsValid(false), expTableIsValid(false),
cutoffDistance(0.0f), alphaDispersionEwald(0.0f), alphaEwald(0.0f) {
}
CpuNonbondedForce::~CpuNonbondedForce() {
......@@ -68,10 +70,21 @@ void CpuNonbondedForce::setUseCutoff(float distance, const CpuNeighborList& neig
tableIsValid = false;
cutoff = true;
cutoffDistance = distance;
inverseRcut6 = pow(cutoffDistance, -6);
neighborList = &neighbors;
krf = pow(cutoffDistance, -3.0f)*(solventDielectric-1.0)/(2.0*solventDielectric+1.0);
crf = (1.0/cutoffDistance)*(3.0*solventDielectric)/(2.0*solventDielectric+1.0);
}
if(alphaDispersionEwald != 0.0f){
// We set this here, in case setUseCutoff is called after the dispersion alpha is set.
double dalphaR = alphaDispersionEwald*cutoffDistance;
double dar2 = dalphaR * dalphaR;
double dar4 = dar2*dar2;
double dar6 = dar4*dar2;
double expterm = EXP(-dar2);
inverseRcut6Expterm = inverseRcut6*(1.0 - expterm * (1.0 + dar2 + 0.5*dar4));
}
}
/**---------------------------------------------------------------------------------------
......@@ -86,7 +99,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
switchingDistance = distance;
}
/**---------------------------------------------------------------------------------------
/**---------------------------------------------------------------------------------------
Set the force to use periodic boundary conditions. This requires that a cutoff has
also been set, and the smallest side of the periodic box is at least twice the cutoff
......@@ -96,7 +109,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
--------------------------------------------------------------------------------------- */
void CpuNonbondedForce::setPeriodic(RealVec* periodicBoxVectors) {
void CpuNonbondedForce::setPeriodic(RealVec* periodicBoxVectors) {
assert(cutoff);
assert(periodicBoxVectors[0][0] >= 2.0*cutoffDistance);
......@@ -114,11 +127,11 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
periodicBoxVec4[1] = fvec4(periodicBoxVectors[1][0], periodicBoxVectors[1][1], periodicBoxVectors[1][2], 0);
periodicBoxVec4[2] = fvec4(periodicBoxVectors[2][0], periodicBoxVectors[2][1], periodicBoxVectors[2][2], 0);
triclinic = (periodicBoxVectors[0][1] != 0.0 || periodicBoxVectors[0][2] != 0.0 ||
periodicBoxVectors[1][0] != 0.0 || periodicBoxVectors[1][2] != 0.0 ||
periodicBoxVectors[2][0] != 0.0 || periodicBoxVectors[2][1] != 0.0);
}
periodicBoxVectors[1][0] != 0.0 || periodicBoxVectors[1][2] != 0.0 ||
periodicBoxVectors[2][0] != 0.0 || periodicBoxVectors[2][1] != 0.0);
}
/**---------------------------------------------------------------------------------------
/**---------------------------------------------------------------------------------------
Set the force to use Ewald summation.
......@@ -129,18 +142,18 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
--------------------------------------------------------------------------------------- */
void CpuNonbondedForce::setUseEwald(float alpha, int kmaxx, int kmaxy, int kmaxz) {
if (alpha != alphaEwald)
tableIsValid = false;
alphaEwald = alpha;
numRx = kmaxx;
numRy = kmaxy;
numRz = kmaxz;
ewald = true;
tabulateEwaldScaleFactor();
}
void CpuNonbondedForce::setUseEwald(float alpha, int kmaxx, int kmaxy, int kmaxz) {
if (alpha != alphaEwald)
tableIsValid = false;
alphaEwald = alpha;
numRx = kmaxx;
numRy = kmaxy;
numRz = kmaxz;
ewald = true;
tabulateEwaldScaleFactor();
}
/**---------------------------------------------------------------------------------------
/**---------------------------------------------------------------------------------------
Set the force to use Particle-Mesh Ewald (PME) summation.
......@@ -149,19 +162,49 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
--------------------------------------------------------------------------------------- */
void CpuNonbondedForce::setUsePME(float alpha, int meshSize[3]) {
if (alpha != alphaEwald)
tableIsValid = false;
alphaEwald = alpha;
meshDim[0] = meshSize[0];
meshDim[1] = meshSize[1];
meshDim[2] = meshSize[2];
pme = true;
tabulateEwaldScaleFactor();
}
void CpuNonbondedForce::tabulateEwaldScaleFactor() {
void CpuNonbondedForce::setUsePME(float alpha, int meshSize[3]) {
if (alpha != alphaEwald)
tableIsValid = false;
alphaEwald = alpha;
meshDim[0] = meshSize[0];
meshDim[1] = meshSize[1];
meshDim[2] = meshSize[2];
pme = true;
tabulateEwaldScaleFactor();
}
/**---------------------------------------------------------------------------------------
Set the force to use Particle-Mesh Ewald (PME) summation for dispersion.
@param alpha the Ewald separation parameter
@param gridSize the dimensions of the mesh
--------------------------------------------------------------------------------------- */
void CpuNonbondedForce::setUseLJPME(float alpha, int meshSize[3]) {
if (alpha != alphaDispersionEwald)
expTableIsValid = false;
alphaDispersionEwald = alpha;
dispersionMeshDim[0] = meshSize[0];
dispersionMeshDim[1] = meshSize[1];
dispersionMeshDim[2] = meshSize[2];
ljpme = true;
tabulateExpTerms();
if(cutoffDistance != 0.0f){
// We set this here, in case setUseLJPME is called after the cutoff is set
double dalphaR = alphaDispersionEwald*cutoffDistance;
double dar2 = dalphaR * dalphaR;
double dar4 = dar2*dar2;
double dar6 = dar4*dar2;
double expterm = EXP(-dar2);
inverseRcut6Expterm = inverseRcut6*(1.0 - expterm * (1.0 + dar2 + 0.5*dar4));
}
}
void CpuNonbondedForce::tabulateEwaldScaleFactor() {
if (tableIsValid)
return;
tableIsValid = true;
......@@ -177,10 +220,30 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
ewaldScaleTable[i] = erfcTable[i] + TWO_OVER_SQRT_PI*alphaR*exp(-alphaR*alphaR);
}
}
void CpuNonbondedForce::tabulateExpTerms() {
if (expTableIsValid)
return;
expTableIsValid = true;
exptermsDX = cutoffDistance/NUM_TABLE_POINTS;
exptermsDXInv = 1.0f/exptermsDX;
exptermsTable.resize(NUM_TABLE_POINTS+4);
dExptermsTable.resize(NUM_TABLE_POINTS+4);
for (int i = 0; i < NUM_TABLE_POINTS+4; i++) {
double r = i*ewaldDX;
double dalphaR = alphaDispersionEwald*r;
double dar2 = dalphaR * dalphaR;
double dar4 = dar2*dar2;
double dar6 = dar4*dar2;
double expterm = EXP(-dar2);
exptermsTable[i] = (1.0 - expterm * (1.0 + dar2 + 0.5*dar4));
dExptermsTable[i] = (1.0 - expterm * (1.0 + dar2 + 0.5*dar4 + dar6/6.0));
}
}
void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, const vector<RealVec>& atomCoordinates,
const vector<pair<float, float> >& atomParameters, const vector<set<int> >& exclusions,
vector<RealVec>& forces, double* totalEnergy) const {
const vector<pair<float, float> >& atomParameters, const vector<float> &C6params, const vector<set<int> >& exclusions,
vector<RealVec>& forces, double* totalEnergy) const {
typedef std::complex<float> d_complex;
static const float epsilon = 1.0;
......@@ -201,6 +264,29 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
if (totalEnergy)
*totalEnergy += recipEnergy;
pme_destroy(pmedata);
if (ljpme) {
// Dispersion reciprocal space terms
pme_init(&pmedata,alphaDispersionEwald,numberOfAtoms,dispersionMeshDim,5,1);
std::vector<RealVec> dpmeforces;
for (int i = 0; i < numberOfAtoms; i++){
charges[i] = (RealOpenMM)C6params[i];
dpmeforces.push_back(RealVec());
}
RealOpenMM recipDispersionEnergy = 0.0;
pme_exec_dpme(pmedata,atomCoordinates,dpmeforces,charges,periodicBoxVectors,&recipDispersionEnergy);
for (int i = 0; i < numberOfAtoms; i++){
forces[i][0] -= 2.0*dpmeforces[i][0];
forces[i][1] -= 2.0*dpmeforces[i][1];
forces[i][2] -= 2.0*dpmeforces[i][2];
}
if (totalEnergy)
*totalEnergy += recipDispersionEnergy;
pme_destroy(pmedata);
}
}
// Ewald method
......@@ -214,7 +300,7 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
// setup K-vectors
#define EIR(x, y, z) eir[(x)*numberOfAtoms*3+(y)*3+z]
#define EIR(x, y, z) eir[(x)*numberOfAtoms*3+(y)*3+z]
vector<d_complex> eir(kmax*numberOfAtoms*3);
vector<d_complex> tab_xy(numberOfAtoms);
vector<d_complex> tab_qxyz(numberOfAtoms);
......@@ -222,15 +308,15 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
for (int i = 0; (i < numberOfAtoms); i++) {
float* pos = posq+4*i;
for (int m = 0; (m < 3); m++)
EIR(0, i, m) = d_complex(1,0);
EIR(0, i, m) = d_complex(1,0);
for (int m=0; (m<3); m++)
EIR(1, i, m) = d_complex(cos(pos[m]*recipBoxSize[m]),
sin(pos[m]*recipBoxSize[m]));
EIR(1, i, m) = d_complex(cos(pos[m]*recipBoxSize[m]),
sin(pos[m]*recipBoxSize[m]));
for (int j=2; (j<kmax); j++)
for (int m=0; (m<3); m++)
EIR(j, i, m) = EIR(j-1, i, m) * EIR(1, i, m);
for (int m=0; (m<3); m++)
EIR(j, i, m) = EIR(j-1, i, m) * EIR(1, i, m);
}
// calculate reciprocal space energy and forces
......@@ -244,11 +330,11 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
float ky = ry * recipBoxSize[1];
if (ry >= 0) {
for (int n = 0; n < numberOfAtoms; n++)
tab_xy[n] = EIR(rx, n, 0) * EIR(ry, n, 1);
tab_xy[n] = EIR(rx, n, 0) * EIR(ry, n, 1);
}
else {
for (int n = 0; n < numberOfAtoms; n++)
tab_xy[n]= EIR(rx, n, 0) * conj (EIR(-ry, n, 1));
tab_xy[n]= EIR(rx, n, 0) * conj (EIR(-ry, n, 1));
}
for (int rz = lowrz; rz < numRz; rz++) {
if (rz >= 0) {
......@@ -291,13 +377,14 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
void CpuNonbondedForce::calculateDirectIxn(int numberOfAtoms, float* posq, const vector<RealVec>& atomCoordinates, const vector<pair<float, float> >& atomParameters,
const vector<set<int> >& exclusions, vector<AlignedArray<float> >& threadForce, double* totalEnergy, ThreadPool& threads) {
const vector<float>& C6params, const vector<set<int> >& exclusions, vector<AlignedArray<float> >& threadForce, double* totalEnergy, ThreadPool& threads) {
// Record the parameters for the threads.
this->numberOfAtoms = numberOfAtoms;
this->posq = posq;
this->atomCoordinates = &atomCoordinates[0];
this->atomParameters = &atomParameters[0];
this->C6params = &C6params[0];
this->exclusions = &exclusions[0];
this->threadForce = &threadForce;
includeEnergy = (totalEnergy != NULL);
......@@ -339,9 +426,8 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex
float* forces = &(*threadForce)[threadIndex][0];
fvec4 boxSize(periodicBoxVectors[0][0], periodicBoxVectors[1][1], periodicBoxVectors[2][2], 0);
fvec4 invBoxSize(recipBoxSize[0], recipBoxSize[1], recipBoxSize[2], 0);
if (ewald || pme) {
if (ewald || pme || ljpme) {
// Compute the interactions from the neighbor list.
while (true) {
int nextBlock = gmx_atomic_fetch_add(reinterpret_cast<gmx_atomic_t*>(atomicCounter), 1);
if (nextBlock >= neighborList->getNumBlocks())
......@@ -359,7 +445,7 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex
break;
int end = min(start+groupSize, numberOfAtoms);
for (int i = start; i < end; i++) {
fvec4 posI((float) atomCoordinates[i][0], (float) atomCoordinates[i][1], (float) atomCoordinates[i][2], 0.0f);
fvec4 posI((float) atomCoordinates[i][0], (float) atomCoordinates[i][1], (float) atomCoordinates[i][2], 0.0f);
float scaledChargeI = (float) (ONE_4PI_EPS0*posq[4*i+3]);
for (set<int>::const_iterator iter = exclusions[i].begin(); iter != exclusions[i].end(); ++iter) {
if (*iter > i) {
......@@ -383,7 +469,18 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex
threadEnergy[threadIndex] -= chargeProdOverR*erfAlphaR;
}
else if (includeEnergy)
threadEnergy[threadIndex] -= alphaEwald*TWO_OVER_SQRT_PI*scaledChargeI*posq[4*j+3];
threadEnergy[threadIndex] -= alphaEwald*TWO_OVER_SQRT_PI*scaledChargeI*posq[4*j+3];
if (ljpme) {
float C6ij = C6params[i]*C6params[j];
float inverseR2 = 1.0f/r2;
float emult = C6ij*inverseR2*inverseR2*inverseR2*exptermsApprox(r);
if(includeEnergy)
threadEnergy[threadIndex] += emult;
float dEdR = -6.0f*C6ij*inverseR2*inverseR2*inverseR2*inverseR2*dExptermsApprox(r);
fvec4 result = deltaR*dEdR;
(fvec4(forces+4*i)-result).store(forces+4*i);
(fvec4(forces+4*j)+result).store(forces+4*j);
}
}
}
}
......@@ -433,7 +530,7 @@ void CpuNonbondedForce::calculateOneIxn(int ii, int jj, float* forces, double* t
}
float sig = atomParameters[ii].first + atomParameters[jj].first;
float sig2 = inverseR*sig;
sig2 *= sig2;
sig2 *= sig2;
float sig6 = sig2*sig2*sig2;
float eps = atomParameters[ii].second*atomParameters[jj].second;
......@@ -465,7 +562,7 @@ void CpuNonbondedForce::calculateOneIxn(int ii, int jj, float* forces, double* t
fvec4 result = deltaR*dEdR;
(fvec4(forces+4*ii)+result).store(forces+4*ii);
(fvec4(forces+4*jj)-result).store(forces+4*jj);
}
}
void CpuNonbondedForce::getDeltaR(const fvec4& posI, const fvec4& posJ, fvec4& deltaR, float& r2, bool periodic, const fvec4& boxSize, const fvec4& invBoxSize) const {
deltaR = posJ-posI;
......@@ -491,3 +588,18 @@ float CpuNonbondedForce::erfcApprox(float x) {
return coeff1*erfcTable[index] + coeff2*erfcTable[index+1];
}
float CpuNonbondedForce::exptermsApprox(float x) {
float x1 = x*exptermsDXInv;
int index = min((int) floor(x1), NUM_TABLE_POINTS);
float coeff2 = x1-index;
float coeff1 = 1.0f-coeff2;
return coeff1*exptermsTable[index] + coeff2*exptermsTable[index+1];
}
float CpuNonbondedForce::dExptermsApprox(float x) {
float x1 = x*exptermsDXInv;
int index = min((int) floor(x1), NUM_TABLE_POINTS);
float coeff2 = x1-index;
float coeff1 = 1.0f-coeff2;
return coeff1*dExptermsTable[index] + coeff2*dExptermsTable[index+1];
}
platforms/cpu/src/CpuNonbondedForceVec4.cpp
View file @ 53f770f4
......@@ -25,6 +25,7 @@
#include "SimTKOpenMMUtilities.h"
#include "CpuNonbondedForceVec4.h"
#include <algorithm>
#include <iostream>
using namespace std;
using namespace OpenMM;
......@@ -213,7 +214,6 @@ void CpuNonbondedForceVec4::calculateBlockIxnImpl(int blockIndex, float* forces,
void CpuNonbondedForceVec4::calculateBlockEwaldIxn(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
// Determine whether we need to apply periodic boundary conditions.
PeriodicType periodicType;
fvec4 blockCenter;
if (!periodic) {
......@@ -263,7 +263,6 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxn(int blockIndex, float* forces
template <int PERIODIC_TYPE>
void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize, const fvec4& blockCenter) {
// Load the positions and parameters of the atoms in the block.
const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex];
fvec4 blockAtomPosq[4];
fvec4 blockAtomForceX(0.0f), blockAtomForceY(0.0f), blockAtomForceZ(0.0f);
......@@ -278,9 +277,10 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
fvec4 blockAtomCharge = fvec4(ONE_4PI_EPS0)*fvec4(blockAtomPosq[0][3], blockAtomPosq[1][3], blockAtomPosq[2][3], blockAtomPosq[3][3]);
fvec4 blockAtomSigma(atomParameters[blockAtom[0]].first, atomParameters[blockAtom[1]].first, atomParameters[blockAtom[2]].first, atomParameters[blockAtom[3]].first);
fvec4 blockAtomEpsilon(atomParameters[blockAtom[0]].second, atomParameters[blockAtom[1]].second, atomParameters[blockAtom[2]].second, atomParameters[blockAtom[3]].second);
fvec4 C6s(C6params[blockAtom[0]], C6params[blockAtom[1]], C6params[blockAtom[2]], C6params[blockAtom[3]]);
const bool needPeriodic = (PERIODIC_TYPE == PeriodicPerInteraction || PERIODIC_TYPE == PeriodicTriclinic);
const float invSwitchingInterval = 1/(cutoffDistance-switchingDistance);
// Loop over neighbors for this block.
const vector<int>& neighbors = neighborList->getBlockNeighbors(blockIndex);
......@@ -318,7 +318,8 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
fvec4 sig2 = inverseR*sig;
sig2 *= sig2;
fvec4 sig6 = sig2*sig2*sig2;
fvec4 epsSig6 = blockAtomEpsilon*atomEpsilon*sig6;
fvec4 eps = blockAtomEpsilon*atomEpsilon;
fvec4 epsSig6 = eps*sig6;
dEdR = epsSig6*(12.0f*sig6 - 6.0f);
energy = epsSig6*(sig6-1.0f);
if (useSwitch) {
......@@ -328,6 +329,17 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
dEdR = switchValue*dEdR - energy*switchDeriv*r;
energy *= switchValue;
}
if (ljpme) {
fvec4 C6ij = C6s*C6params[atom];
fvec4 inverseR2 = inverseR*inverseR;
fvec4 mysig2 = sig*sig;
fvec4 mysig6 = mysig2*mysig2*mysig2;
fvec4 emult = C6ij*inverseR2*inverseR2*inverseR2*exptermsApprox(r);
fvec4 potentialShift = eps*(1.0f-mysig6*inverseRcut6)*mysig6*inverseRcut6 - C6ij*inverseRcut6Expterm;
dEdR += 6.0f*C6ij*inverseR2*inverseR2*inverseR2*dExptermsApprox(r);
energy += emult + potentialShift;
}
}
else {
energy = 0.0f;
......@@ -362,7 +374,7 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
}
// Record the forces on the block atoms.
fvec4 f[4] = {blockAtomForceX, blockAtomForceY, blockAtomForceZ, 0.0f};
transpose(f[0], f[1], f[2], f[3]);
for (int j = 0; j < 4; j++)
......@@ -420,3 +432,30 @@ fvec4 CpuNonbondedForceVec4::ewaldScaleFunction(const fvec4& x) {
transpose(t1, t2, t3, t4);
return coeff1*t1 + coeff2*t2;
}
fvec4 CpuNonbondedForceVec4::exptermsApprox(const fvec4& r) {
fvec4 r1 = r*exptermsDXInv;
ivec4 index = min(floor(r1), NUM_TABLE_POINTS);
fvec4 coeff2 = r1-index;
fvec4 coeff1 = 1.0f-coeff2;
fvec4 t1(&exptermsTable[index[0]]);
fvec4 t2(&exptermsTable[index[1]]);
fvec4 t3(&exptermsTable[index[2]]);
fvec4 t4(&exptermsTable[index[3]]);
transpose(t1, t2, t3, t4);
return coeff1*t1 + coeff2*t2;
}
fvec4 CpuNonbondedForceVec4::dExptermsApprox(const fvec4& r) {
fvec4 r1 = r*exptermsDXInv;
ivec4 index = min(floor(r1), NUM_TABLE_POINTS);
fvec4 coeff2 = r1-index;
fvec4 coeff1 = 1.0f-coeff2;
fvec4 t1(&dExptermsTable[index[0]]);
fvec4 t2(&dExptermsTable[index[1]]);
fvec4 t3(&dExptermsTable[index[2]]);
fvec4 t4(&dExptermsTable[index[3]]);
transpose(t1, t2, t3, t4);
return coeff1*t1 + coeff2*t2;
}
platforms/cpu/src/CpuNonbondedForceVec8.cpp
View file @ 53f770f4
......@@ -27,6 +27,7 @@
#include "openmm/OpenMMException.h"
#include "openmm/internal/hardware.h"
#include <algorithm>
#include <iostream>
using namespace std;
using namespace OpenMM;
......@@ -80,8 +81,7 @@ CpuNonbondedForceVec8::CpuNonbondedForceVec8() {
enum PeriodicType {NoPeriodic, PeriodicPerAtom, PeriodicPerInteraction, PeriodicTriclinic};
void CpuNonbondedForceVec8::calculateBlockIxn(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
// Determine whether we need to apply periodic boundary conditions.
// Determine whether we need to apply periodic boundary conditions.
PeriodicType periodicType;
fvec4 blockCenter;
if (!periodic) {
......@@ -308,6 +308,7 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
blockAtomCharge *= ONE_4PI_EPS0;
fvec8 blockAtomSigma(atomParameters[blockAtom[0]].first, atomParameters[blockAtom[1]].first, atomParameters[blockAtom[2]].first, atomParameters[blockAtom[3]].first, atomParameters[blockAtom[4]].first, atomParameters[blockAtom[5]].first, atomParameters[blockAtom[6]].first, atomParameters[blockAtom[7]].first);
fvec8 blockAtomEpsilon(atomParameters[blockAtom[0]].second, atomParameters[blockAtom[1]].second, atomParameters[blockAtom[2]].second, atomParameters[blockAtom[3]].second, atomParameters[blockAtom[4]].second, atomParameters[blockAtom[5]].second, atomParameters[blockAtom[6]].second, atomParameters[blockAtom[7]].second);
fvec8 C6s(C6params[blockAtom[0]], C6params[blockAtom[1]], C6params[blockAtom[2]], C6params[blockAtom[3]], C6params[blockAtom[4]], C6params[blockAtom[5]], C6params[blockAtom[6]], C6params[blockAtom[7]]);
const bool needPeriodic = (PERIODIC_TYPE == PeriodicPerInteraction || PERIODIC_TYPE == PeriodicTriclinic);
const float invSwitchingInterval = 1/(cutoffDistance-switchingDistance);
......@@ -348,7 +349,8 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
fvec8 sig2 = inverseR*sig;
sig2 *= sig2;
fvec8 sig6 = sig2*sig2*sig2;
fvec8 epsSig6 = blockAtomEpsilon*atomEpsilon*sig6;
fvec8 eps = blockAtomEpsilon*atomEpsilon;
fvec8 epsSig6 = eps*sig6;
dEdR = epsSig6*(12.0f*sig6 - 6.0f);
energy = epsSig6*(sig6-1.0f);
if (useSwitch) {
......@@ -358,6 +360,17 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
dEdR = switchValue*dEdR - energy*switchDeriv*r;
energy *= switchValue;
}
if (ljpme) {
fvec8 C6ij = C6s*C6params[atom];
fvec8 inverseR2 = inverseR*inverseR;
fvec8 mysig2 = sig*sig;
fvec8 mysig6 = mysig2*mysig2*mysig2;
fvec8 emult = C6ij*inverseR2*inverseR2*inverseR2*exptermsApprox(r);
fvec8 potentialShift = eps*(1.0f-mysig6*inverseRcut6)*mysig6*inverseRcut6 - C6ij*inverseRcut6Expterm;
dEdR += 6.0f*C6ij*inverseR2*inverseR2*inverseR2*dExptermsApprox(r);
energy += emult + potentialShift;
}
}
else {
energy = 0.0f;
......@@ -464,4 +477,45 @@ fvec8 CpuNonbondedForceVec8::ewaldScaleFunction(const fvec8& x) {
transpose(t1, t2, t3, t4, t5, t6, t7, t8, s1, s2, s3, s4);
return coeff1*s1 + coeff2*s2;
}
fvec8 CpuNonbondedForceVec8::exptermsApprox(const fvec8& r) {
fvec8 r1 = r*exptermsDXInv;
ivec8 index = min(floor(r1), NUM_TABLE_POINTS);
fvec8 coeff2 = r1-index;
fvec8 coeff1 = 1.0f-coeff2;
ivec4 indexLower = index.lowerVec();
ivec4 indexUpper = index.upperVec();
fvec4 t1(&exptermsTable[indexLower[0]]);
fvec4 t2(&exptermsTable[indexLower[1]]);
fvec4 t3(&exptermsTable[indexLower[2]]);
fvec4 t4(&exptermsTable[indexLower[3]]);
fvec4 t5(&exptermsTable[indexUpper[0]]);
fvec4 t6(&exptermsTable[indexUpper[1]]);
fvec4 t7(&exptermsTable[indexUpper[2]]);
fvec4 t8(&exptermsTable[indexUpper[3]]);
fvec8 s1, s2, s3, s4;
transpose(t1, t2, t3, t4, t5, t6, t7, t8, s1, s2, s3, s4);
return coeff1*s1 + coeff2*s2;
}
fvec8 CpuNonbondedForceVec8::dExptermsApprox(const fvec8& r) {
fvec8 r1 = r*exptermsDXInv;
ivec8 index = min(floor(r1), NUM_TABLE_POINTS);
fvec8 coeff2 = r1-index;
fvec8 coeff1 = 1.0f-coeff2;
ivec4 indexLower = index.lowerVec();
ivec4 indexUpper = index.upperVec();
fvec4 t1(&dExptermsTable[indexLower[0]]);
fvec4 t2(&dExptermsTable[indexLower[1]]);
fvec4 t3(&dExptermsTable[indexLower[2]]);
fvec4 t4(&dExptermsTable[indexLower[3]]);
fvec4 t5(&dExptermsTable[indexUpper[0]]);
fvec4 t6(&dExptermsTable[indexUpper[1]]);
fvec4 t7(&dExptermsTable[indexUpper[2]]);
fvec4 t8(&dExptermsTable[indexUpper[3]]);
fvec8 s1, s2, s3, s4;
transpose(t1, t2, t3, t4, t5, t6, t7, t8, s1, s2, s3, s4);
return coeff1*s1 + coeff2*s2;
}
#endif
platforms/cpu/tests/TestCpuDispersionPME.cpp 0 → 100644
View file @ 53f770f4
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2017 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
* Permission is hereby granted, free of charge, to any person obtaining a *
* copy of this software and associated documentation files (the "Software"), *
* to deal in the Software without restriction, including without limitation *
* the rights to use, copy, modify, merge, publish, distribute, sublicense, *
* and/or sell copies of the Software, and to permit persons to whom the *
* Software is furnished to do so, subject to the following conditions: *
* *
* The above copyright notice and this permission notice shall be included in *
* all copies or substantial portions of the Software. *
* *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, *
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL *
* THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, *
* DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR *
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE *
* USE OR OTHER DEALINGS IN THE SOFTWARE. *
* -------------------------------------------------------------------------- */
#include "CpuTests.h"
#include "TestDispersionPME.h"
void runPlatformTests() {
}
platforms/cuda/include/CudaKernels.h
View file @ 53f770f4
......@@ -614,7 +614,8 @@ class CudaCalcNonbondedForceKernel : public CalcNonbondedForceKernel {
public:
CudaCalcNonbondedForceKernel(std::string name, const Platform& platform, CudaContext& cu, const System& system) : CalcNonbondedForceKernel(name, platform),
cu(cu), hasInitializedFFT(false), sigmaEpsilon(NULL), exceptionParams(NULL), cosSinSums(NULL), directPmeGrid(NULL), reciprocalPmeGrid(NULL),
pmeBsplineModuliX(NULL), pmeBsplineModuliY(NULL), pmeBsplineModuliZ(NULL), pmeAtomRange(NULL), pmeAtomGridIndex(NULL), pmeEnergyBuffer(NULL), sort(NULL), fft(NULL), pmeio(NULL) {
pmeBsplineModuliX(NULL), pmeBsplineModuliY(NULL), pmeBsplineModuliZ(NULL), pmeDispersionBsplineModuliX(NULL), pmeDispersionBsplineModuliY(NULL),
pmeDispersionBsplineModuliZ(NULL), pmeAtomRange(NULL), pmeAtomGridIndex(NULL), pmeEnergyBuffer(NULL), sort(NULL), dispersionFft(NULL), fft(NULL), pmeio(NULL) {
}
~CudaCalcNonbondedForceKernel();
/**
......@@ -651,6 +652,15 @@ public:
* @param nz the number of grid points along the Z axis
*/
void getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
/**
* Get the dispersion parameters being used for the dispersion term in LJPME.
*
* @param alpha the separation parameter
* @param nx the number of grid points along the X axis
* @param ny the number of grid points along the Y axis
* @param nz the number of grid points along the Z axis
*/
void getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
private:
class SortTrait : public CudaSort::SortTrait {
int getDataSize() const {return 8;}
......@@ -679,6 +689,9 @@ private:
CudaArray* pmeBsplineModuliX;
CudaArray* pmeBsplineModuliY;
CudaArray* pmeBsplineModuliZ;
CudaArray* pmeDispersionBsplineModuliX;
CudaArray* pmeDispersionBsplineModuliY;
CudaArray* pmeDispersionBsplineModuliZ;
CudaArray* pmeAtomRange;
CudaArray* pmeAtomGridIndex;
CudaArray* pmeEnergyBuffer;
......@@ -690,20 +703,29 @@ private:
CudaFFT3D* fft;
cufftHandle fftForward;
cufftHandle fftBackward;
CudaFFT3D* dispersionFft;
cufftHandle dispersionFftForward;
cufftHandle dispersionFftBackward;
CUfunction ewaldSumsKernel;
CUfunction ewaldForcesKernel;
CUfunction pmeGridIndexKernel;
CUfunction pmeDispersionGridIndexKernel;
CUfunction pmeSpreadChargeKernel;
CUfunction pmeDispersionSpreadChargeKernel;
CUfunction pmeFinishSpreadChargeKernel;
CUfunction pmeDispersionFinishSpreadChargeKernel;
CUfunction pmeEvalEnergyKernel;
CUfunction pmeEvalDispersionEnergyKernel;
CUfunction pmeConvolutionKernel;
CUfunction pmeDispersionConvolutionKernel;
CUfunction pmeInterpolateForceKernel;
std::map<std::string, std::string> pmeDefines;
CUfunction pmeInterpolateDispersionForceKernel;
std::vector<std::pair<int, int> > exceptionAtoms;
double ewaldSelfEnergy, dispersionCoefficient, alpha;
double ewaldSelfEnergy, dispersionCoefficient, alpha, dispersionAlpha;
int interpolateForceThreads;
int gridSizeX, gridSizeY, gridSizeZ;
bool hasCoulomb, hasLJ, usePmeStream, useCudaFFT;
int dispersionGridSizeX, dispersionGridSizeY, dispersionGridSizeZ;
bool hasCoulomb, hasLJ, usePmeStream, useCudaFFT, doLJPME;
NonbondedMethod nonbondedMethod;
static const int PmeOrder = 5;
};
......
platforms/cuda/include/CudaParallelKernels.h
View file @ 53f770f4
......@@ -439,6 +439,15 @@ public:
* @param nz the number of grid points along the Z axis
*/
void getPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
/**
* Get the dispersion parameters being used for the dispersion term in LJPME.
*
* @param alpha the separation parameter
* @param nx the number of grid points along the X axis
* @param ny the number of grid points along the Y axis
* @param nz the number of grid points along the Z axis
*/
void getLJPMEParameters(double& alpha, int& nx, int& ny, int& nz) const;
private:
class Task;
CudaPlatform::PlatformData& data;
......
platforms/cuda/src/CudaKernels.cpp
View file @ 53f770f4
This diff is collapsed.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment