Skip to content

GitLab

Commit 3db6b8ee authored by Jason Swails's avatar Jason Swails
Browse files

Merge branch 'master' into gbn-chk

parents 5d8e92ff 3a80b0f1
Expand all Show whitespace changes
Inline Side-by-side
Showing with 848 additions and 343 deletions
platforms/cpu/src/CpuKernels.cpp
View file @ 3db6b8ee
......@@ -73,6 +73,11 @@ static RealVec& extractBoxSize(ContextImpl& context) {
return *(RealVec*) data->periodicBoxSize;
}
static RealVec* extractBoxVectors(ContextImpl& context) {
ReferencePlatform::PlatformData* data = reinterpret_cast<ReferencePlatform::PlatformData*>(context.getPlatformData());
return (RealVec*) data->periodicBoxVectors;
}
static ReferenceConstraints& extractConstraints(ContextImpl& context) {
ReferencePlatform::PlatformData* data = reinterpret_cast<ReferencePlatform::PlatformData*>(context.getPlatformData());
return *(ReferenceConstraints*) data->constraints;
......@@ -147,19 +152,36 @@ public:
AlignedArray<float>& posq = data.posq;
vector<RealVec>& posData = extractPositions(context);
RealVec boxSize = extractBoxSize(context);
double invBoxSize[3] = {1/boxSize[0], 1/boxSize[1], 1/boxSize[2]};
RealVec* boxVectors = extractBoxVectors(context);
double boxSize[3] = {boxVectors[0][0], boxVectors[1][1], boxVectors[2][2]};
double invBoxSize[3] = {1/boxVectors[0][0], 1/boxVectors[1][1], 1/boxVectors[2][2]};
bool triclinic = (boxVectors[0][1] != 0 || boxVectors[0][2] != 0 || boxVectors[1][0] != 0 || boxVectors[1][2] != 0 || boxVectors[2][0] != 0 || boxVectors[2][1] != 0);
int numParticles = context.getSystem().getNumParticles();
int numThreads = threads.getNumThreads();
int start = threadIndex*numParticles/numThreads;
int end = (threadIndex+1)*numParticles/numThreads;
if (data.isPeriodic)
for (int i = start; i < end; i++)
if (data.isPeriodic) {
if (triclinic) {
for (int i = start; i < end; i++) {
RealVec pos = posData[i];
pos -= boxVectors[2]*floor(pos[2]*invBoxSize[2]);
pos -= boxVectors[1]*floor(pos[1]*invBoxSize[1]);
pos -= boxVectors[0]*floor(pos[0]*invBoxSize[0]);
posq[4*i] = (float) pos[0];
posq[4*i+1] = (float) pos[1];
posq[4*i+2] = (float) pos[2];
}
}
else {
for (int i = start; i < end; i++) {
for (int j = 0; j < 3; j++) {
RealOpenMM x = posData[i][j];
double base = floor(x*invBoxSize[j])*boxSize[j];
posq[4*i+j] = (float) (x-base);
}
}
}
}
else
for (int i = start; i < end; i++) {
posq[4*i] = (float) posData[i][0];
......@@ -499,8 +521,7 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
AlignedArray<float>& posq = data.posq;
vector<RealVec>& posData = extractPositions(context);
vector<RealVec>& forceData = extractForces(context);
RealVec boxSize = extractBoxSize(context);
float floatBoxSize[3] = {(float) boxSize[0], (float) boxSize[1], (float) boxSize[2]};
RealVec* boxVectors = extractBoxVectors(context);
double energy = (includeReciprocal ? ewaldSelfEnergy : 0.0);
bool ewald = (nonbondedMethod == Ewald);
bool pme = (nonbondedMethod == PME);
......@@ -546,16 +567,17 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
}
}
if (needRecompute) {
neighborList->computeNeighborList(numParticles, posq, exclusions, floatBoxSize, data.isPeriodic, nonbondedCutoff+padding, data.threads);
neighborList->computeNeighborList(numParticles, posq, exclusions, boxVectors, data.isPeriodic, nonbondedCutoff+padding, data.threads);
lastPositions = posData;
}
nonbonded->setUseCutoff(nonbondedCutoff, *neighborList, rfDielectric);
}
if (data.isPeriodic) {
RealVec* boxVectors = extractBoxVectors(context);
double minAllowedSize = 1.999999*nonbondedCutoff;
if (boxSize[0] < minAllowedSize || boxSize[1] < minAllowedSize || boxSize[2] < minAllowedSize)
if (boxVectors[0][0] < minAllowedSize || boxVectors[1][1] < minAllowedSize || boxVectors[2][2] < minAllowedSize)
throw OpenMMException("The periodic box size has decreased to less than twice the nonbonded cutoff.");
nonbonded->setPeriodic(floatBoxSize);
nonbonded->setPeriodic(boxVectors);
}
if (ewald)
nonbonded->setUseEwald(ewaldAlpha, kmax[0], kmax[1], kmax[2]);
......@@ -569,8 +591,8 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
if (includeReciprocal) {
if (useOptimizedPme) {
PmeIO io(&posq[0], &data.threadForce[0][0], numParticles);
Vec3 periodicBoxSize(boxSize[0], boxSize[1], boxSize[2]);
optimizedPme.getAs<CalcPmeReciprocalForceKernel>().beginComputation(io, periodicBoxSize, includeEnergy);
Vec3 periodicBoxVectors[3] = {boxVectors[0], boxVectors[1], boxVectors[2]};
optimizedPme.getAs<CalcPmeReciprocalForceKernel>().beginComputation(io, periodicBoxVectors, includeEnergy);
nonbondedEnergy += optimizedPme.getAs<CalcPmeReciprocalForceKernel>().finishComputation(io);
}
else
......@@ -582,7 +604,7 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
ReferenceLJCoulomb14 nonbonded14;
refBondForce.calculateForce(num14, bonded14IndexArray, posData, bonded14ParamArray, forceData, includeEnergy ? &energy : NULL, nonbonded14);
if (data.isPeriodic)
energy += dispersionCoefficient/(boxSize[0]*boxSize[1]*boxSize[2]);
energy += dispersionCoefficient/(boxVectors[0][0]*boxVectors[1][1]*boxVectors[2][2]);
}
return energy;
}
......@@ -736,19 +758,18 @@ void CpuCalcCustomNonbondedForceKernel::initialize(const System& system, const C
double CpuCalcCustomNonbondedForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
vector<RealVec>& posData = extractPositions(context);
vector<RealVec>& forceData = extractForces(context);
RealVec& box = extractBoxSize(context);
float floatBoxSize[3] = {(float) box[0], (float) box[1], (float) box[2]};
RealVec* boxVectors = extractBoxVectors(context);
double energy = 0;
bool periodic = (nonbondedMethod == CutoffPeriodic);
if (nonbondedMethod != NoCutoff) {
neighborList->computeNeighborList(numParticles, data.posq, exclusions, floatBoxSize, data.isPeriodic, nonbondedCutoff, data.threads);
neighborList->computeNeighborList(numParticles, data.posq, exclusions, boxVectors, data.isPeriodic, nonbondedCutoff, data.threads);
nonbonded->setUseCutoff(nonbondedCutoff, *neighborList);
}
if (periodic) {
double minAllowedSize = 2*nonbondedCutoff;
if (box[0] < minAllowedSize || box[1] < minAllowedSize || box[2] < minAllowedSize)
if (boxVectors[0][0] < minAllowedSize || boxVectors[1][1] < minAllowedSize || boxVectors[2][2] < minAllowedSize)
throw OpenMMException("The periodic box size has decreased to less than twice the nonbonded cutoff.");
nonbonded->setPeriodic(box);
nonbonded->setPeriodic(boxVectors);
}
bool globalParamsChanged = false;
for (int i = 0; i < (int) globalParameterNames.size(); i++) {
......@@ -767,7 +788,7 @@ double CpuCalcCustomNonbondedForceKernel::execute(ContextImpl& context, bool inc
longRangeCoefficient = CustomNonbondedForceImpl::calcLongRangeCorrection(*forceCopy, context.getOwner());
hasInitializedLongRangeCorrection = true;
}
energy += longRangeCoefficient/(box[0]*box[1]*box[2]);
energy += longRangeCoefficient/(boxVectors[0][0]*boxVectors[1][1]*boxVectors[2][2]);
return energy;
}
......@@ -975,13 +996,12 @@ void CpuCalcCustomGBForceKernel::initialize(const System& system, const CustomGB
double CpuCalcCustomGBForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
vector<RealVec>& forceData = extractForces(context);
RealOpenMM energy = 0;
RealVec& box = extractBoxSize(context);
float floatBoxSize[3] = {(float) box[0], (float) box[1], (float) box[2]};
RealVec* boxVectors = extractBoxVectors(context);
if (data.isPeriodic)
ixn->setPeriodic(extractBoxSize(context));
if (nonbondedMethod != NoCutoff) {
vector<set<int> > noExclusions(numParticles);
neighborList->computeNeighborList(numParticles, data.posq, exclusions, floatBoxSize, data.isPeriodic, nonbondedCutoff, data.threads);
neighborList->computeNeighborList(numParticles, data.posq, exclusions, boxVectors, data.isPeriodic, nonbondedCutoff, data.threads);
ixn->setUseCutoff(nonbondedCutoff, *neighborList);
}
map<string, double> globalParameters;
......@@ -1038,6 +1058,7 @@ void CpuCalcCustomManyParticleForceKernel::initialize(const System& system, cons
ixn = new CpuCustomManyParticleForce(force, data.threads);
nonbondedMethod = CalcCustomManyParticleForceKernel::NonbondedMethod(force.getNonbondedMethod());
cutoffDistance = force.getCutoffDistance();
data.isPeriodic = (nonbondedMethod == CutoffPeriodic);
}
double CpuCalcCustomManyParticleForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
......@@ -1045,11 +1066,11 @@ double CpuCalcCustomManyParticleForceKernel::execute(ContextImpl& context, bool
for (int i = 0; i < (int) globalParameterNames.size(); i++)
globalParameters[globalParameterNames[i]] = context.getParameter(globalParameterNames[i]);
if (nonbondedMethod == CutoffPeriodic) {
RealVec& box = extractBoxSize(context);
RealVec* boxVectors = extractBoxVectors(context);
double minAllowedSize = 2*cutoffDistance;
if (box[0] < minAllowedSize || box[1] < minAllowedSize || box[2] < minAllowedSize)
if (boxVectors[0][0] < minAllowedSize || boxVectors[1][1] < minAllowedSize || boxVectors[2][2] < minAllowedSize)
throw OpenMMException("The periodic box size has decreased to less than twice the nonbonded cutoff.");
ixn->setPeriodic(box);
ixn->setPeriodic(boxVectors);
}
double energy = 0;
ixn->calculateIxn(data.posq, particleParamArray, globalParameters, data.threadForce, includeForces, includeEnergy, energy);
......
platforms/cpu/src/CpuNeighborList.cpp
View file @ 3db6b8ee
This diff is collapsed.
platforms/cpu/src/CpuNonbondedForce.cpp
View file @ 3db6b8ee
......@@ -103,20 +103,30 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
also been set, and the smallest side of the periodic box is at least twice the cutoff
distance.
@param boxSize the X, Y, and Z widths of the periodic box
@param periodicBoxVectors the vectors defining the periodic box
--------------------------------------------------------------------------------------- */
void CpuNonbondedForce::setPeriodic(float* periodicBoxSize) {
void CpuNonbondedForce::setPeriodic(RealVec* periodicBoxVectors) {
assert(cutoff);
assert(periodicBoxSize[0] >= 2*cutoffDistance);
assert(periodicBoxSize[1] >= 2*cutoffDistance);
assert(periodicBoxSize[2] >= 2*cutoffDistance);
assert(periodicBoxVectors[0][0] >= 2.0*cutoffDistance);
assert(periodicBoxVectors[1][1] >= 2.0*cutoffDistance);
assert(periodicBoxVectors[2][2] >= 2.0*cutoffDistance);
periodic = true;
this->periodicBoxSize[0] = periodicBoxSize[0];
this->periodicBoxSize[1] = periodicBoxSize[1];
this->periodicBoxSize[2] = periodicBoxSize[2];
this->periodicBoxVectors[0] = periodicBoxVectors[0];
this->periodicBoxVectors[1] = periodicBoxVectors[1];
this->periodicBoxVectors[2] = periodicBoxVectors[2];
recipBoxSize[0] = (float) (1.0/periodicBoxVectors[0][0]);
recipBoxSize[1] = (float) (1.0/periodicBoxVectors[1][1]);
recipBoxSize[2] = (float) (1.0/periodicBoxVectors[2][2]);
periodicBoxVec4.resize(3);
periodicBoxVec4[0] = fvec4(periodicBoxVectors[0][0], periodicBoxVectors[0][1], periodicBoxVectors[0][2], 0);
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);
}
/**---------------------------------------------------------------------------------------
......@@ -186,18 +196,16 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
int kmax = (ewald ? max(numRx, max(numRy,numRz)) : 0);
float factorEwald = -1 / (4*alphaEwald*alphaEwald);
float TWO_PI = 2.0 * PI_M;
float recipCoeff = (float)(ONE_4PI_EPS0*4*PI_M/(periodicBoxSize[0] * periodicBoxSize[1] * periodicBoxSize[2]) /epsilon);
float recipCoeff = (float)(ONE_4PI_EPS0*4*PI_M/(periodicBoxVectors[0][0] * periodicBoxVectors[1][1] * periodicBoxVectors[2][2]) /epsilon);
if (pme) {
pme_t pmedata;
RealOpenMM virial[3][3];
pme_init(&pmedata, alphaEwald, numberOfAtoms, meshDim, 5, 1);
vector<RealOpenMM> charges(numberOfAtoms);
for (int i = 0; i < numberOfAtoms; i++)
charges[i] = posq[4*i+3];
RealOpenMM boxSize[3] = {periodicBoxSize[0], periodicBoxSize[1], periodicBoxSize[2]};
RealOpenMM recipEnergy = 0.0;
pme_exec(pmedata, atomCoordinates, forces, charges, boxSize, &recipEnergy, virial);
pme_exec(pmedata, atomCoordinates, forces, charges, periodicBoxVectors, &recipEnergy);
if (totalEnergy)
*totalEnergy += recipEnergy;
pme_destroy(pmedata);
......@@ -209,7 +217,7 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
// setup reciprocal box
float recipBoxSize[3] = { TWO_PI / periodicBoxSize[0], TWO_PI / periodicBoxSize[1], TWO_PI / periodicBoxSize[2]};
float recipBoxSize[3] = { TWO_PI / periodicBoxVectors[0][0], TWO_PI / periodicBoxVectors[1][1], TWO_PI / periodicBoxVectors[2][2]};
// setup K-vectors
......@@ -330,8 +338,8 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex
threadEnergy[threadIndex] = 0;
double* energyPtr = (includeEnergy ? &threadEnergy[threadIndex] : NULL);
float* forces = &(*threadForce)[threadIndex][0];
fvec4 boxSize(periodicBoxSize[0], periodicBoxSize[1], periodicBoxSize[2], 0);
fvec4 invBoxSize((1/periodicBoxSize[0]), (1/periodicBoxSize[1]), (1/periodicBoxSize[2]), 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) {
// Compute the interactions from the neighbor list.
......@@ -344,8 +352,6 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex
// Now subtract off the exclusions, since they were implicitly included in the reciprocal space sum.
fvec4 boxSize(periodicBoxSize[0], periodicBoxSize[1], periodicBoxSize[2], 0);
fvec4 invBoxSize((1/periodicBoxSize[0]), (1/periodicBoxSize[1]), (1/periodicBoxSize[2]), 0);
for (int i = threadIndex; i < numberOfAtoms; i += numThreads) {
fvec4 posI((float) atomCoordinates[i][0], (float) atomCoordinates[i][1], (float) atomCoordinates[i][2], 0.0f);
for (set<int>::const_iterator iter = exclusions[i].begin(); iter != exclusions[i].end(); ++iter) {
......@@ -454,9 +460,16 @@ void CpuNonbondedForce::calculateOneIxn(int ii, int jj, float* forces, double* t
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;
if (periodic) {
if (triclinic) {
deltaR -= periodicBoxVec4[2]*floorf(deltaR[2]*recipBoxSize[2]+0.5f);
deltaR -= periodicBoxVec4[1]*floorf(deltaR[1]*recipBoxSize[1]+0.5f);
deltaR -= periodicBoxVec4[0]*floorf(deltaR[0]*recipBoxSize[0]+0.5f);
}
else {
fvec4 base = round(deltaR*invBoxSize)*boxSize;
deltaR = deltaR-base;
}
}
r2 = dot3(deltaR, deltaR);
}
......
platforms/cpu/src/CpuNonbondedForceVec4.cpp
View file @ 3db6b8ee
/* Portions copyright (c) 2006-2013 Stanford University and Simbios.
/* Portions copyright (c) 2006-2014 Stanford University and Simbios.
* Contributors: Pande Group
*
* Permission is hereby granted, free of charge, to any person obtaining
......@@ -46,6 +46,14 @@ CpuNonbondedForceVec4::CpuNonbondedForceVec4() {
}
void CpuNonbondedForceVec4::calculateBlockIxn(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
if (triclinic)
calculateBlockIxnImpl<true>(blockIndex, forces, totalEnergy, boxSize, invBoxSize);
else
calculateBlockIxnImpl<false>(blockIndex, forces, totalEnergy, boxSize, invBoxSize);
}
template <bool TRICLINIC>
void CpuNonbondedForceVec4::calculateBlockIxnImpl(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
// Load the positions and parameters of the atoms in the block.
const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex];
......@@ -75,7 +83,7 @@ void CpuNonbondedForceVec4::calculateBlockIxn(int blockIndex, float* forces, dou
// Compute the distances to the block atoms.
fvec4 dx, dy, dz, r2;
getDeltaR(posq+4*atom, blockAtomX, blockAtomY, blockAtomZ, dx, dy, dz, r2, needPeriodic, boxSize, invBoxSize);
getDeltaR<TRICLINIC>(posq+4*atom, blockAtomX, blockAtomY, blockAtomZ, dx, dy, dz, r2, needPeriodic, boxSize, invBoxSize);
ivec4 include;
char excl = exclusions[i];
if (excl == 0)
......@@ -155,6 +163,14 @@ void CpuNonbondedForceVec4::calculateBlockIxn(int blockIndex, float* forces, dou
}
void CpuNonbondedForceVec4::calculateBlockEwaldIxn(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
if (triclinic)
calculateBlockEwaldIxnImpl<true>(blockIndex, forces, totalEnergy, boxSize, invBoxSize);
else
calculateBlockEwaldIxnImpl<false>(blockIndex, forces, totalEnergy, boxSize, invBoxSize);
}
template <bool TRICLINIC>
void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
// Load the positions and parameters of the atoms in the block.
const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex];
......@@ -184,7 +200,7 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxn(int blockIndex, float* forces
// Compute the distances to the block atoms.
fvec4 dx, dy, dz, r2;
getDeltaR(posq+4*atom, blockAtomX, blockAtomY, blockAtomZ, dx, dy, dz, r2, needPeriodic, boxSize, invBoxSize);
getDeltaR<TRICLINIC>(posq+4*atom, blockAtomX, blockAtomY, blockAtomZ, dx, dy, dz, r2, needPeriodic, boxSize, invBoxSize);
ivec4 include;
char excl = exclusions[i];
if (excl == 0)
......@@ -257,15 +273,29 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxn(int blockIndex, float* forces
(fvec4(forces+4*blockAtom[j])+f[j]).store(forces+4*blockAtom[j]);
}
template <bool TRICLINIC>
void CpuNonbondedForceVec4::getDeltaR(const float* posI, const fvec4& x, const fvec4& y, const fvec4& z, fvec4& dx, fvec4& dy, fvec4& dz, fvec4& r2, bool periodic, const fvec4& boxSize, const fvec4& invBoxSize) const {
dx = x-posI[0];
dy = y-posI[1];
dz = z-posI[2];
if (periodic) {
if (TRICLINIC) {
fvec4 scale3 = floor(dz*recipBoxSize[2]+0.5f);
dx -= scale3*periodicBoxVectors[2][0];
dy -= scale3*periodicBoxVectors[2][1];
dz -= scale3*periodicBoxVectors[2][2];
fvec4 scale2 = floor(dy*recipBoxSize[1]+0.5f);
dx -= scale2*periodicBoxVectors[1][0];
dy -= scale2*periodicBoxVectors[1][1];
fvec4 scale1 = floor(dx*recipBoxSize[0]+0.5f);
dx -= scale1*periodicBoxVectors[0][0];
}
else {
dx -= round(dx*invBoxSize[0])*boxSize[0];
dy -= round(dy*invBoxSize[1])*boxSize[1];
dz -= round(dz*invBoxSize[2])*boxSize[2];
}
}
r2 = dx*dx + dy*dy + dz*dz;
}
......
platforms/cpu/src/CpuNonbondedForceVec8.cpp
View file @ 3db6b8ee
/* Portions copyright (c) 2006-2013 Stanford University and Simbios.
/* Portions copyright (c) 2006-2014 Stanford University and Simbios.
* Contributors: Pande Group
*
* Permission is hereby granted, free of charge, to any person obtaining
......@@ -78,6 +78,14 @@ CpuNonbondedForceVec8::CpuNonbondedForceVec8() {
}
void CpuNonbondedForceVec8::calculateBlockIxn(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
if (triclinic)
calculateBlockIxnImpl<true>(blockIndex, forces, totalEnergy, boxSize, invBoxSize);
else
calculateBlockIxnImpl<false>(blockIndex, forces, totalEnergy, boxSize, invBoxSize);
}
template <bool TRICLINIC>
void CpuNonbondedForceVec8::calculateBlockIxnImpl(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
// Load the positions and parameters of the atoms in the block.
const int* blockAtom = &neighborList->getSortedAtoms()[8*blockIndex];
......@@ -106,7 +114,7 @@ void CpuNonbondedForceVec8::calculateBlockIxn(int blockIndex, float* forces, dou
// Compute the distances to the block atoms.
fvec8 dx, dy, dz, r2;
getDeltaR(&posq[4*atom], blockAtomX, blockAtomY, blockAtomZ, dx, dy, dz, r2, needPeriodic, boxSize, invBoxSize);
getDeltaR<TRICLINIC>(&posq[4*atom], blockAtomX, blockAtomY, blockAtomZ, dx, dy, dz, r2, needPeriodic, boxSize, invBoxSize);
ivec8 include;
char excl = exclusions[i];
if (excl == 0)
......@@ -186,6 +194,14 @@ void CpuNonbondedForceVec8::calculateBlockIxn(int blockIndex, float* forces, dou
}
void CpuNonbondedForceVec8::calculateBlockEwaldIxn(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
if (triclinic)
calculateBlockEwaldIxnImpl<true>(blockIndex, forces, totalEnergy, boxSize, invBoxSize);
else
calculateBlockEwaldIxnImpl<false>(blockIndex, forces, totalEnergy, boxSize, invBoxSize);
}
template <bool TRICLINIC>
void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) {
// Load the positions and parameters of the atoms in the block.
const int* blockAtom = &neighborList->getSortedAtoms()[8*blockIndex];
......@@ -214,7 +230,7 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxn(int blockIndex, float* forces
// Compute the distances to the block atoms.
fvec8 dx, dy, dz, r2;
getDeltaR(&posq[4*atom], blockAtomX, blockAtomY, blockAtomZ, dx, dy, dz, r2, needPeriodic, boxSize, invBoxSize);
getDeltaR<TRICLINIC>(&posq[4*atom], blockAtomX, blockAtomY, blockAtomZ, dx, dy, dz, r2, needPeriodic, boxSize, invBoxSize);
ivec8 include;
char excl = exclusions[i];
if (excl == 0)
......@@ -287,15 +303,29 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxn(int blockIndex, float* forces
(fvec4(forces+4*blockAtom[j])+f[j]).store(forces+4*blockAtom[j]);
}
template <bool TRICLINIC>
void CpuNonbondedForceVec8::getDeltaR(const float* posI, const fvec8& x, const fvec8& y, const fvec8& z, fvec8& dx, fvec8& dy, fvec8& dz, fvec8& r2, bool periodic, const fvec4& boxSize, const fvec4& invBoxSize) const {
dx = x-posI[0];
dy = y-posI[1];
dz = z-posI[2];
if (periodic) {
if (TRICLINIC) {
fvec8 scale3 = floor(dz*recipBoxSize[2]+0.5f);
dx -= scale3*periodicBoxVectors[2][0];
dy -= scale3*periodicBoxVectors[2][1];
dz -= scale3*periodicBoxVectors[2][2];
fvec8 scale2 = floor(dy*recipBoxSize[1]+0.5f);
dx -= scale2*periodicBoxVectors[1][0];
dy -= scale2*periodicBoxVectors[1][1];
fvec8 scale1 = floor(dx*recipBoxSize[0]+0.5f);
dx -= scale1*periodicBoxVectors[0][0];
}
else {
dx -= round(dx*invBoxSize[0])*boxSize[0];
dy -= round(dy*invBoxSize[1])*boxSize[1];
dz -= round(dz*invBoxSize[2])*boxSize[2];
}
}
r2 = dx*dx + dy*dy + dz*dz;
}
......
platforms/cpu/tests/TestCpuCustomManyParticleForce.cpp
View file @ 3db6b8ee
......@@ -53,7 +53,17 @@ using namespace std;
const double TOL = 1e-5;
void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& positions, const vector<const int*>& expectedSets, double boxSize) {
Vec3 computeDelta(const Vec3& pos1, const Vec3& pos2, bool periodic, const Vec3* periodicBoxVectors) {
Vec3 diff = pos1-pos2;
if (periodic) {
diff -= periodicBoxVectors[2]*floor(diff[2]/periodicBoxVectors[2][2]+0.5);
diff -= periodicBoxVectors[1]*floor(diff[1]/periodicBoxVectors[1][1]+0.5);
diff -= periodicBoxVectors[0]*floor(diff[0]/periodicBoxVectors[0][0]+0.5);
}
return diff;
}
void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& positions, const vector<const int*>& expectedSets, double boxSize, bool triclinic) {
// Create a System and Context.
int numParticles = force->getNumParticles();
......@@ -61,7 +71,18 @@ void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& p
System system;
for (int i = 0; i < numParticles; i++)
system.addParticle(1.0);
system.setDefaultPeriodicBoxVectors(Vec3(boxSize, 0, 0), Vec3(0, boxSize, 0), Vec3(0, 0, boxSize));
Vec3 boxVectors[3];
if (triclinic) {
boxVectors[0] = Vec3(boxSize, 0, 0);
boxVectors[1] = Vec3(0.2*boxSize, boxSize, 0);
boxVectors[2] = Vec3(-0.3*boxSize, -0.1*boxSize, boxSize);
}
else {
boxVectors[0] = Vec3(boxSize, 0, 0);
boxVectors[1] = Vec3(0, boxSize, 0);
boxVectors[2] = Vec3(0, 0, boxSize);
}
system.setDefaultPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
system.addForce(force);
VerletIntegrator integrator(0.001);
CpuPlatform platform;
......@@ -73,20 +94,14 @@ void validateAxilrodTeller(CustomManyParticleForce* force, const vector<Vec3>& p
// See if the energy matches the expected value.
double expectedEnergy = 0;
bool periodic = (nonbondedMethod == CustomManyParticleForce::CutoffPeriodic);
for (int i = 0; i < (int) expectedSets.size(); i++) {
int p1 = expectedSets[i][0];
int p2 = expectedSets[i][1];
int p3 = expectedSets[i][2];
Vec3 d12 = positions[p2]-positions[p1];
Vec3 d13 = positions[p3]-positions[p1];
Vec3 d23 = positions[p3]-positions[p2];
if (nonbondedMethod == CustomManyParticleForce::CutoffPeriodic) {
for (int j = 0; j < 3; j++) {
d12[j] -= floor(d12[j]/boxSize+0.5f)*boxSize;
d13[j] -= floor(d13[j]/boxSize+0.5f)*boxSize;
d23[j] -= floor(d23[j]/boxSize+0.5f)*boxSize;
}
}
Vec3 d12 = computeDelta(positions[p2], positions[p1], periodic, boxVectors);
Vec3 d13 = computeDelta(positions[p3], positions[p1], periodic, boxVectors);
Vec3 d23 = computeDelta(positions[p3], positions[p2], periodic, boxVectors);
double r12 = sqrt(d12.dot(d12));
double r13 = sqrt(d13.dot(d13));
double r23 = sqrt(d23.dot(d23));
......@@ -210,7 +225,7 @@ void testNoCutoff() {
positions.push_back(Vec3(0.4, 0, -0.8));
int sets[4][3] = {{0,1,2}, {1,2,3}, {2,3,0}, {3,0,1}};
vector<const int*> expectedSets(&sets[0], &sets[4]);
validateAxilrodTeller(force, positions, expectedSets, 2.0);
validateAxilrodTeller(force, positions, expectedSets, 2.0, false);
}
void testCutoff() {
......@@ -235,7 +250,7 @@ void testCutoff() {
positions.push_back(Vec3(0.2, 0.5, -0.1));
int sets[7][3] = {{0,1,2}, {0,1,3}, {0,1,4}, {0,2,4}, {0,3,4}, {1,2,4}, {1,3,4}};
vector<const int*> expectedSets(&sets[0], &sets[7]);
validateAxilrodTeller(force, positions, expectedSets, 2.0);
validateAxilrodTeller(force, positions, expectedSets, 2.0, false);
}
void testPeriodic() {
......@@ -261,7 +276,33 @@ void testPeriodic() {
double boxSize = 2.1;
int sets[5][3] = {{0,1,3}, {0,1,4}, {0,2,4}, {0,3,4}, {1,3,4}};
vector<const int*> expectedSets(&sets[0], &sets[5]);
validateAxilrodTeller(force, positions, expectedSets, boxSize);
validateAxilrodTeller(force, positions, expectedSets, boxSize, false);
}
void testTriclinic() {
CustomManyParticleForce* force = new CustomManyParticleForce(3,
"C*(1+3*cos(theta1)*cos(theta2)*cos(theta3))/(r12*r13*r23)^3;"
"theta1=angle(p1,p2,p3); theta2=angle(p2,p3,p1); theta3=angle(p3,p1,p2);"
"r12=distance(p1,p2); r13=distance(p1,p3); r23=distance(p2,p3)");
force->addGlobalParameter("C", 1.5);
force->setNonbondedMethod(CustomManyParticleForce::CutoffPeriodic);
force->setCutoffDistance(1.05);
vector<double> params;
force->addParticle(params);
force->addParticle(params);
force->addParticle(params);
force->addParticle(params);
force->addParticle(params);
vector<Vec3> positions;
positions.push_back(Vec3(0, 0, 0));
positions.push_back(Vec3(1, 0, 0));
positions.push_back(Vec3(0, 1.1, 0.3));
positions.push_back(Vec3(0.4, 0, -0.8));
positions.push_back(Vec3(0.2, 0.5, -0.1));
double boxSize = 2.1;
int sets[4][3] = {{0,1,3}, {0,1,4}, {0,3,4}, {1,3,4}};
vector<const int*> expectedSets(&sets[0], &sets[4]);
validateAxilrodTeller(force, positions, expectedSets, boxSize, true);
}
void testExclusions() {
......@@ -286,7 +327,7 @@ void testExclusions() {
force->addExclusion(0, 3);
int sets[5][3] = {{0,1,4}, {1,2,3}, {1,2,4}, {1,3,4}, {2,3,4}};
vector<const int*> expectedSets(&sets[0], &sets[5]);
validateAxilrodTeller(force, positions, expectedSets, 2.0);
validateAxilrodTeller(force, positions, expectedSets, 2.0, false);
}
void testAllTerms() {
......@@ -680,6 +721,7 @@ int main() {
testNoCutoff();
testCutoff();
testPeriodic();
testTriclinic();
testExclusions();
testAllTerms();
testParameters();
......
platforms/cpu/tests/TestCpuCustomNonbondedForce.cpp
View file @ 3db6b8ee
......@@ -224,6 +224,65 @@ void testPeriodic() {
ASSERT_EQUAL_TOL(1.9+1+0.9, state.getPotentialEnergy(), TOL);
}
void testTriclinic() {
System system;
system.addParticle(1.0);
system.addParticle(1.0);
Vec3 a(3.1, 0, 0);
Vec3 b(0.4, 3.5, 0);
Vec3 c(-0.1, -0.5, 4.0);
system.setDefaultPeriodicBoxVectors(a, b, c);
VerletIntegrator integrator(0.01);
CustomNonbondedForce* nonbonded = new CustomNonbondedForce("r");
nonbonded->addParticle(vector<double>());
nonbonded->addParticle(vector<double>());
nonbonded->setNonbondedMethod(CustomNonbondedForce::CutoffPeriodic);
const double cutoff = 1.5;
nonbonded->setCutoffDistance(cutoff);
system.addForce(nonbonded);
Context context(system, integrator, platform);
vector<Vec3> positions(2);
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
for (int iteration = 0; iteration < 50; iteration++) {
// Generate random positions for the two particles.
positions[0] = a*genrand_real2(sfmt) + b*genrand_real2(sfmt) + c*genrand_real2(sfmt);
positions[1] = a*genrand_real2(sfmt) + b*genrand_real2(sfmt) + c*genrand_real2(sfmt);
context.setPositions(positions);
// Loop over all possible periodic copies and find the nearest one.
Vec3 delta;
double distance2 = 100.0;
for (int i = -1; i < 2; i++)
for (int j = -1; j < 2; j++)
for (int k = -1; k < 2; k++) {
Vec3 d = positions[1]-positions[0]+a*i+b*j+c*k;
if (d.dot(d) < distance2) {
delta = d;
distance2 = d.dot(d);
}
}
double distance = sqrt(distance2);
// See if the force and energy are correct.
State state = context.getState(State::Forces | State::Energy);
if (distance >= cutoff) {
ASSERT_EQUAL(0.0, state.getPotentialEnergy());
ASSERT_EQUAL_VEC(Vec3(0, 0, 0), state.getForces()[0], 0);
ASSERT_EQUAL_VEC(Vec3(0, 0, 0), state.getForces()[1], 0);
}
else {
const Vec3 force = delta/sqrt(delta.dot(delta));
ASSERT_EQUAL_TOL(distance, state.getPotentialEnergy(), TOL);
ASSERT_EQUAL_VEC(force, state.getForces()[0], TOL);
ASSERT_EQUAL_VEC(-force, state.getForces()[1], TOL);
}
}
}
void testContinuous1DFunction() {
System system;
system.addParticle(1.0);
......@@ -852,6 +911,7 @@ int main() {
testExclusions();
testCutoff();
testPeriodic();
testTriclinic();
testContinuous1DFunction();
testContinuous2DFunction();
testContinuous3DFunction();
......
platforms/cpu/tests/TestCpuEwald.cpp
View file @ 3db6b8ee
......@@ -201,6 +201,57 @@ void testEwald2Ions() {
ASSERT_EQUAL_TOL(-217.276, state.getPotentialEnergy(), 0.01/*10*TOL*/);
}
void testTriclinic() {
// Create a triclinic box containing eight particles.
System system;
system.setDefaultPeriodicBoxVectors(Vec3(2.5, 0, 0), Vec3(0.5, 3.0, 0), Vec3(0.7, 0.9, 3.5));
for (int i = 0; i < 8; i++)
system.addParticle(1.0);
NonbondedForce* force = new NonbondedForce();
system.addForce(force);
force->setNonbondedMethod(NonbondedForce::PME);
force->setCutoffDistance(1.0);
force->setPMEParameters(3.45891, 32, 40, 48);
for (int i = 0; i < 4; i++)
force->addParticle(-1, 0.440104, 0.4184); // Cl parameters
for (int i = 0; i < 4; i++)
force->addParticle(1, 0.332840, 0.0115897); // Na parameters
vector<Vec3> positions(8);
positions[0] = Vec3(1.744, 2.788, 3.162);
positions[1] = Vec3(1.048, 0.762, 2.340);
positions[2] = Vec3(2.489, 1.570, 2.817);
positions[3] = Vec3(1.027, 1.893, 3.271);
positions[4] = Vec3(0.937, 0.825, 0.009);
positions[5] = Vec3(2.290, 1.887, 3.352);
positions[6] = Vec3(1.266, 1.111, 2.894);
positions[7] = Vec3(0.933, 1.862, 3.490);
// Compute the forces and energy.
VerletIntegrator integ(0.001);
Context context(system, integ, platform);
context.setPositions(positions);
State state = context.getState(State::Forces | State::Energy);
// Compare them to values computed by Gromacs.
double expectedEnergy = -963.370;
vector<Vec3> expectedForce(8);
expectedForce[0] = Vec3(4.25253e+01, -1.23503e+02, 1.22139e+02);
expectedForce[1] = Vec3(9.74752e+01, 1.68213e+02, 1.93169e+02);
expectedForce[2] = Vec3(-1.50348e+02, 1.29165e+02, 3.70435e+02);
expectedForce[3] = Vec3(9.18644e+02, -3.52571e+00, -1.34772e+03);
expectedForce[4] = Vec3(-1.61193e+02, 9.01528e+01, -7.12904e+01);
expectedForce[5] = Vec3(2.82630e+02, 2.78029e+01, -3.72864e+02);
expectedForce[6] = Vec3(-1.47454e+02, -2.14448e+02, -3.55789e+02);
expectedForce[7] = Vec3(-8.82195e+02, -7.39132e+01, 1.46202e+03);
for (int i = 0; i < 8; i++) {
ASSERT_EQUAL_VEC(expectedForce[i], state.getForces()[i], 1e-4);
}
ASSERT_EQUAL_TOL(expectedEnergy, state.getPotentialEnergy(), 1e-4);
}
void testErrorTolerance(NonbondedForce::NonbondedMethod method) {
// Create a cloud of random point charges.
......@@ -261,6 +312,7 @@ int main(int argc, char* argv[]) {
testEwaldPME(false);
testEwaldPME(true);
// testEwald2Ions();
testTriclinic();
testErrorTolerance(NonbondedForce::Ewald);
testErrorTolerance(NonbondedForce::PME);
}
......
platforms/cpu/tests/TestCpuNeighborList.cpp
View file @ 3db6b8ee
......@@ -48,10 +48,21 @@
using namespace OpenMM;
using namespace std;
void testNeighborList(bool periodic) {
void testNeighborList(bool periodic, bool triclinic) {
const int numParticles = 500;
const float cutoff = 2.0f;
const float boxSize[3] = {20.0f, 15.0f, 22.0f};
RealVec boxVectors[3];
if (triclinic) {
boxVectors[0] = RealVec(20, 0, 0);
boxVectors[1] = RealVec(5, 15, 0);
boxVectors[2] = RealVec(-3, -7, 22);
}
else {
boxVectors[0] = RealVec(20, 0, 0);
boxVectors[1] = RealVec(0, 15, 0);
boxVectors[2] = RealVec(0, 0, 22);
}
const float boxSize[3] = {boxVectors[0][0], boxVectors[1][1], boxVectors[2][2]};
const int blockSize = 8;
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
......@@ -69,7 +80,7 @@ void testNeighborList(bool periodic) {
}
ThreadPool threads;
CpuNeighborList neighborList(blockSize);
neighborList.computeNeighborList(numParticles, positions, exclusions, boxSize, periodic, cutoff, threads);
neighborList.computeNeighborList(numParticles, positions, exclusions, boxVectors, periodic, cutoff, threads);
// Convert the neighbor list to a set for faster lookup.
......@@ -94,15 +105,13 @@ void testNeighborList(bool periodic) {
for (int i = 0; i < numParticles; i++)
for (int j = 0; j <= i; j++) {
bool shouldInclude = (exclusions[i].find(j) == exclusions[i].end());
float dx = positions[4*i]-positions[4*j];
float dy = positions[4*i+1]-positions[4*j+1];
float dz = positions[4*i+2]-positions[4*j+2];
Vec3 diff(positions[4*i]-positions[4*j], positions[4*i+1]-positions[4*j+1], positions[4*i+2]-positions[4*j+2]);
if (periodic) {
dx -= floor(dx/boxSize[0]+0.5f)*boxSize[0];
dy -= floor(dy/boxSize[1]+0.5f)*boxSize[1];
dz -= floor(dz/boxSize[2]+0.5f)*boxSize[2];
diff -= boxVectors[2]*floor(diff[2]/boxSize[2]+0.5);
diff -= boxVectors[1]*floor(diff[1]/boxSize[1]+0.5);
diff -= boxVectors[0]*floor(diff[0]/boxSize[0]+0.5);
}
if (dx*dx + dy*dy + dz*dz > cutoff*cutoff)
if (diff.dot(diff) > cutoff*cutoff)
shouldInclude = false;
bool isIncluded = (neighbors.find(make_pair(i, j)) != neighbors.end() || neighbors.find(make_pair(j, i)) != neighbors.end());
if (shouldInclude)
......@@ -116,8 +125,9 @@ int main() {
cout << "CPU is not supported. Exiting." << endl;
return 0;
}
testNeighborList(false);
testNeighborList(true);
testNeighborList(false, false);
testNeighborList(true, false);
testNeighborList(true, true);
}
catch(const exception& e) {
cout << "exception: " << e.what() << endl;
......
platforms/cpu/tests/TestCpuNonbondedForce.cpp
View file @ 3db6b8ee
......@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2008-2013 Stanford University and the Authors. *
* Portions copyright (c) 2008-2015 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -353,6 +353,67 @@ void testPeriodic() {
ASSERT_EQUAL_TOL(2*ONE_4PI_EPS0*(1.0)*(1.0+krf*1.0-crf), state.getPotentialEnergy(), TOL);
}
void testTriclinic() {
System system;
system.addParticle(1.0);
system.addParticle(1.0);
Vec3 a(3.1, 0, 0);
Vec3 b(0.4, 3.5, 0);
Vec3 c(-0.1, -0.5, 4.0);
system.setDefaultPeriodicBoxVectors(a, b, c);
VerletIntegrator integrator(0.01);
NonbondedForce* nonbonded = new NonbondedForce();
nonbonded->addParticle(1.0, 1, 0);
nonbonded->addParticle(1.0, 1, 0);
nonbonded->setNonbondedMethod(NonbondedForce::CutoffPeriodic);
const double cutoff = 1.5;
nonbonded->setCutoffDistance(cutoff);
system.addForce(nonbonded);
Context context(system, integrator, platform);
vector<Vec3> positions(2);
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
const double eps = 78.3;
const double krf = (1.0/(cutoff*cutoff*cutoff))*(eps-1.0)/(2.0*eps+1.0);
const double crf = (1.0/cutoff)*(3.0*eps)/(2.0*eps+1.0);
for (int iteration = 0; iteration < 50; iteration++) {
// Generate random positions for the two particles.
positions[0] = a*genrand_real2(sfmt) + b*genrand_real2(sfmt) + c*genrand_real2(sfmt);
positions[1] = a*genrand_real2(sfmt) + b*genrand_real2(sfmt) + c*genrand_real2(sfmt);
context.setPositions(positions);
// Loop over all possible periodic copies and find the nearest one.
Vec3 delta;
double distance2 = 100.0;
for (int i = -1; i < 2; i++)
for (int j = -1; j < 2; j++)
for (int k = -1; k < 2; k++) {
Vec3 d = positions[1]-positions[0]+a*i+b*j+c*k;
if (d.dot(d) < distance2) {
delta = d;
distance2 = d.dot(d);
}
}
double distance = sqrt(distance2);
// See if the force and energy are correct.
State state = context.getState(State::Forces | State::Energy);
if (distance >= cutoff) {
ASSERT_EQUAL(0.0, state.getPotentialEnergy());
ASSERT_EQUAL_VEC(Vec3(0, 0, 0), state.getForces()[0], 0);
ASSERT_EQUAL_VEC(Vec3(0, 0, 0), state.getForces()[1], 0);
}
else {
const Vec3 force = delta*ONE_4PI_EPS0*(-1.0/(distance*distance*distance)+2.0*krf);
ASSERT_EQUAL_TOL(ONE_4PI_EPS0*(1.0/distance+krf*distance*distance-crf), state.getPotentialEnergy(), 1e-4);
ASSERT_EQUAL_VEC(force, state.getForces()[0], 2e-5);
ASSERT_EQUAL_VEC(-force, state.getForces()[1], 2e-5);
}
}
}
void testLargeSystem() {
const int numMolecules = 600;
......@@ -635,6 +696,7 @@ int main(int argc, char* argv[]) {
testCutoff();
testCutoff14();
testPeriodic();
testTriclinic();
testLargeSystem();
testDispersionCorrection();
testChangingParameters();
......
platforms/cuda/include/CudaContext.h
View file @ 3db6b8ee
......@@ -9,7 +9,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009-2013 Stanford University and the Authors. *
* Portions copyright (c) 2009-2015 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -352,43 +352,62 @@ public:
/**
* Get whether double precision is being used.
*/
bool getUseDoublePrecision() {
bool getUseDoublePrecision() const {
return useDoublePrecision;
}
/**
* Get whether mixed precision is being used.
*/
bool getUseMixedPrecision() {
bool getUseMixedPrecision() const {
return useMixedPrecision;
}
/**
* Get whether the periodic box is triclinic.
*/
bool getBoxIsTriclinic() const {
return boxIsTriclinic;
}
/**
* Convert a number to a string in a format suitable for including in a kernel.
* This takes into account whether the context uses single or double precision.
*/
std::string doubleToString(double value);
std::string doubleToString(double value) const;
/**
* Convert a number to a string in a format suitable for including in a kernel.
*/
std::string intToString(int value);
std::string intToString(int value) const;
/**
* Convert a CUDA result code to the corresponding string description.
*/
static std::string getErrorString(CUresult result);
/**
* Get the size of the periodic box.
* Get the vectors defining the periodic box.
*/
double4 getPeriodicBoxSize() const {
return periodicBoxSize;
void getPeriodicBoxVectors(Vec3& a, Vec3& b, Vec3& c) const {
a = Vec3(periodicBoxVecX.x, periodicBoxVecX.y, periodicBoxVecX.z);
b = Vec3(periodicBoxVecY.x, periodicBoxVecY.y, periodicBoxVecY.z);
c = Vec3(periodicBoxVecZ.x, periodicBoxVecZ.y, periodicBoxVecZ.z);
}
/**
* Set the size of the periodic box.
* Set the vectors defining the periodic box.
*/
void setPeriodicBoxSize(double xsize, double ysize, double zsize) {
periodicBoxSize = make_double4(xsize, ysize, zsize, 0.0);
invPeriodicBoxSize = make_double4(1.0/xsize, 1.0/ysize, 1.0/zsize, 0.0);
periodicBoxSizeFloat = make_float4((float) xsize, (float) ysize, (float) zsize, 0.0f);
invPeriodicBoxSizeFloat = make_float4(1.0f/(float) xsize, 1.0f/(float) ysize, 1.0f/(float) zsize, 0.0f);
void setPeriodicBoxVectors(const Vec3& a, const Vec3& b, const Vec3& c) {
periodicBoxVecX = make_double4(a[0], a[1], a[2], 0.0);
periodicBoxVecY = make_double4(b[0], b[1], b[2], 0.0);
periodicBoxVecZ = make_double4(c[0], c[1], c[2], 0.0);
periodicBoxVecXFloat = make_float4((float) a[0], (float) a[1], (float) a[2], 0.0f);
periodicBoxVecYFloat = make_float4((float) b[0], (float) b[1], (float) b[2], 0.0f);
periodicBoxVecZFloat = make_float4((float) c[0], (float) c[1], (float) c[2], 0.0f);
periodicBoxSize = make_double4(a[0], b[1], c[2], 0.0);
invPeriodicBoxSize = make_double4(1.0/a[0], 1.0/b[1], 1.0/c[2], 0.0);
periodicBoxSizeFloat = make_float4((float) a[0], (float) b[1], (float) c[2], 0.0f);
invPeriodicBoxSizeFloat = make_float4(1.0f/(float) a[0], 1.0f/(float) b[1], 1.0f/(float) c[2], 0.0f);
}
/**
* Get the size of the periodic box.
*/
double4 getPeriodicBoxSize() const {
return periodicBoxSize;
}
/**
* Get the inverse of the size of the periodic box.
......@@ -410,6 +429,27 @@ public:
void* getInvPeriodicBoxSizePointer() {
return (useDoublePrecision ? reinterpret_cast<void*>(&invPeriodicBoxSize) : reinterpret_cast<void*>(&invPeriodicBoxSizeFloat));
}
/**
* Get a pointer to the first periodic box vector, represented as either a float4 or double4 depending on
* this context's precision. This value is suitable for passing to kernels as an argument.
*/
void* getPeriodicBoxVecXPointer() {
return (useDoublePrecision ? reinterpret_cast<void*>(&periodicBoxVecX) : reinterpret_cast<void*>(&periodicBoxVecXFloat));
}
/**
* Get a pointer to the second periodic box vector, represented as either a float4 or double4 depending on
* this context's precision. This value is suitable for passing to kernels as an argument.
*/
void* getPeriodicBoxVecYPointer() {
return (useDoublePrecision ? reinterpret_cast<void*>(&periodicBoxVecY) : reinterpret_cast<void*>(&periodicBoxVecYFloat));
}
/**
* Get a pointer to the third periodic box vector, represented as either a float4 or double4 depending on
* this context's precision. This value is suitable for passing to kernels as an argument.
*/
void* getPeriodicBoxVecZPointer() {
return (useDoublePrecision ? reinterpret_cast<void*>(&periodicBoxVecZ) : reinterpret_cast<void*>(&periodicBoxVecZFloat));
}
/**
* Get the CudaIntegrationUtilities for this context.
*/
......@@ -525,10 +565,10 @@ private:
int paddedNumAtoms;
int numAtomBlocks;
int numThreadBlocks;
bool useBlockingSync, useDoublePrecision, useMixedPrecision, contextIsValid, atomsWereReordered;
bool useBlockingSync, useDoublePrecision, useMixedPrecision, contextIsValid, atomsWereReordered, boxIsTriclinic;
std::string compiler, tempDir, cacheDir, gpuArchitecture;
float4 periodicBoxSizeFloat, invPeriodicBoxSizeFloat;
double4 periodicBoxSize, invPeriodicBoxSize;
float4 periodicBoxVecXFloat, periodicBoxVecYFloat, periodicBoxVecZFloat, periodicBoxSizeFloat, invPeriodicBoxSizeFloat;
double4 periodicBoxVecX, periodicBoxVecY, periodicBoxVecZ, periodicBoxSize, invPeriodicBoxSize;
std::string defaultOptimizationOptions;
std::map<std::string, std::string> compilationDefines;
CUcontext context;
......
platforms/cuda/src/CudaContext.cpp
View file @ 3db6b8ee
......@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009-2013 Stanford University and the Authors. *
* Portions copyright (c) 2009-2015 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -233,6 +233,66 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
compilationDefines["ERF"] = useDoublePrecision ? "erf" : "erff";
compilationDefines["ERFC"] = useDoublePrecision ? "erfc" : "erfcf";
// Set defines for applying periodic boundary conditions.
Vec3 boxVectors[3];
system.getDefaultPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
boxIsTriclinic = (boxVectors[0][1] != 0.0 || boxVectors[0][2] != 0.0 ||
boxVectors[1][0] != 0.0 || boxVectors[1][2] != 0.0 ||
boxVectors[2][0] != 0.0 || boxVectors[2][1] != 0.0);
if (boxIsTriclinic) {
compilationDefines["APPLY_PERIODIC_TO_DELTA(delta)"] =
"{"
"real scale3 = floor(delta.z*invPeriodicBoxSize.z+0.5f); \\\n"
"delta.x -= scale3*periodicBoxVecZ.x; \\\n"
"delta.y -= scale3*periodicBoxVecZ.y; \\\n"
"delta.z -= scale3*periodicBoxVecZ.z; \\\n"
"real scale2 = floor(delta.y*invPeriodicBoxSize.y+0.5f); \\\n"
"delta.x -= scale2*periodicBoxVecY.x; \\\n"
"delta.y -= scale2*periodicBoxVecY.y; \\\n"
"real scale1 = floor(delta.x*invPeriodicBoxSize.x+0.5f); \\\n"
"delta.x -= scale1*periodicBoxVecX.x;}";
compilationDefines["APPLY_PERIODIC_TO_POS(pos)"] =
"{"
"real scale3 = floor(pos.z*invPeriodicBoxSize.z); \\\n"
"pos.x -= scale3*periodicBoxVecZ.x; \\\n"
"pos.y -= scale3*periodicBoxVecZ.y; \\\n"
"pos.z -= scale3*periodicBoxVecZ.z; \\\n"
"real scale2 = floor(pos.y*invPeriodicBoxSize.y); \\\n"
"pos.x -= scale2*periodicBoxVecY.x; \\\n"
"pos.y -= scale2*periodicBoxVecY.y; \\\n"
"real scale1 = floor(pos.x*invPeriodicBoxSize.x); \\\n"
"pos.x -= scale1*periodicBoxVecX.x;}";
compilationDefines["APPLY_PERIODIC_TO_POS_WITH_CENTER(pos, center)"] =
"{"
"real scale3 = floor((pos.z-center.z)*invPeriodicBoxSize.z+0.5f); \\\n"
"pos.x -= scale3*periodicBoxVecZ.x; \\\n"
"pos.y -= scale3*periodicBoxVecZ.y; \\\n"
"pos.z -= scale3*periodicBoxVecZ.z; \\\n"
"real scale2 = floor((pos.y-center.y)*invPeriodicBoxSize.y+0.5f); \\\n"
"pos.x -= scale2*periodicBoxVecY.x; \\\n"
"pos.y -= scale2*periodicBoxVecY.y; \\\n"
"real scale1 = floor((pos.x-center.x)*invPeriodicBoxSize.x+0.5f); \\\n"
"pos.x -= scale1*periodicBoxVecX.x;}";
}
else {
compilationDefines["APPLY_PERIODIC_TO_DELTA(delta)"] =
"{"
"delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; \\\n"
"delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y; \\\n"
"delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;}";
compilationDefines["APPLY_PERIODIC_TO_POS(pos)"] =
"{"
"pos.x -= floor(pos.x*invPeriodicBoxSize.x)*periodicBoxSize.x; \\\n"
"pos.y -= floor(pos.y*invPeriodicBoxSize.y)*periodicBoxSize.y; \\\n"
"pos.z -= floor(pos.z*invPeriodicBoxSize.z)*periodicBoxSize.z;}";
compilationDefines["APPLY_PERIODIC_TO_POS_WITH_CENTER(pos, center)"] =
"{"
"pos.x -= floor((pos.x-center.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x; \\\n"
"pos.y -= floor((pos.y-center.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y; \\\n"
"pos.z -= floor((pos.z-center.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;}";
}
// Create the work thread used for parallelization when running on multiple devices.
thread = new WorkThread();
......@@ -534,7 +594,7 @@ void CudaContext::restoreDefaultStream() {
setCurrentStream(0);
}
string CudaContext::doubleToString(double value) {
string CudaContext::doubleToString(double value) const {
stringstream s;
s.precision(useDoublePrecision ? 16 : 8);
s << scientific << value;
......@@ -543,7 +603,7 @@ string CudaContext::doubleToString(double value) {
return s.str();
}
string CudaContext::intToString(int value) {
string CudaContext::intToString(int value) const {
stringstream s;
s << value;
return s.str();
......@@ -1078,16 +1138,21 @@ void CudaContext::reorderAtomsImpl() {
// Move each molecule position into the same box.
for (int i = 0; i < numMolecules; i++) {
int xcell = (int) floor(molPos[i].x*invPeriodicBoxSize.x);
int ycell = (int) floor(molPos[i].y*invPeriodicBoxSize.y);
int zcell = (int) floor(molPos[i].z*invPeriodicBoxSize.z);
Real dx = xcell*periodicBoxSize.x;
Real dy = ycell*periodicBoxSize.y;
Real dz = zcell*periodicBoxSize.z;
if (dx != 0.0f || dy != 0.0f || dz != 0.0f) {
molPos[i].x -= dx;
molPos[i].y -= dy;
molPos[i].z -= dz;
Real4 center = molPos[i];
int zcell = (int) floor(center.z*invPeriodicBoxSize.z);
center.x -= zcell*periodicBoxVecZ.x;
center.y -= zcell*periodicBoxVecZ.y;
center.z -= zcell*periodicBoxVecZ.z;
int ycell = (int) floor(center.y*invPeriodicBoxSize.y);
center.x -= ycell*periodicBoxVecY.x;
center.y -= ycell*periodicBoxVecY.y;
int xcell = (int) floor(center.x*invPeriodicBoxSize.x);
center.x -= xcell*periodicBoxVecX.x;
if (xcell != 0 || ycell != 0 || zcell != 0) {
Real dx = molPos[i].x-center.x;
Real dy = molPos[i].y-center.y;
Real dz = molPos[i].z-center.z;
molPos[i] = center;
for (int j = 0; j < (int) atoms.size(); j++) {
int atom = atoms[j]+mol.offsets[i];
Real4 p = oldPosq[atom];
......
platforms/cuda/src/CudaKernels.cpp
View file @ 3db6b8ee
This diff is collapsed.
platforms/cuda/src/CudaNonbondedUtilities.cpp
View file @ 3db6b8ee
......@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009-2013 Stanford University and the Authors. *
* Portions copyright (c) 2009-2015 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -304,6 +304,9 @@ void CudaNonbondedUtilities::initialize(const System& system) {
findBlockBoundsArgs.push_back(&numAtoms);
findBlockBoundsArgs.push_back(context.getPeriodicBoxSizePointer());
findBlockBoundsArgs.push_back(context.getInvPeriodicBoxSizePointer());
findBlockBoundsArgs.push_back(context.getPeriodicBoxVecXPointer());
findBlockBoundsArgs.push_back(context.getPeriodicBoxVecYPointer());
findBlockBoundsArgs.push_back(context.getPeriodicBoxVecZPointer());
findBlockBoundsArgs.push_back(&context.getPosq().getDevicePointer());
findBlockBoundsArgs.push_back(&blockCenter->getDevicePointer());
findBlockBoundsArgs.push_back(&blockBoundingBox->getDevicePointer());
......@@ -322,6 +325,9 @@ void CudaNonbondedUtilities::initialize(const System& system) {
findInteractingBlocksKernel = context.getKernel(interactingBlocksProgram, "findBlocksWithInteractions");
findInteractingBlocksArgs.push_back(context.getPeriodicBoxSizePointer());
findInteractingBlocksArgs.push_back(context.getInvPeriodicBoxSizePointer());
findInteractingBlocksArgs.push_back(context.getPeriodicBoxVecXPointer());
findInteractingBlocksArgs.push_back(context.getPeriodicBoxVecYPointer());
findInteractingBlocksArgs.push_back(context.getPeriodicBoxVecZPointer());
findInteractingBlocksArgs.push_back(&interactionCount->getDevicePointer());
findInteractingBlocksArgs.push_back(&interactingTiles->getDevicePointer());
findInteractingBlocksArgs.push_back(&interactingAtoms->getDevicePointer());
......@@ -390,10 +396,10 @@ void CudaNonbondedUtilities::updateNeighborListSize() {
interactingAtoms = CudaArray::create<int>(context, CudaContext::TileSize*maxTiles, "interactingAtoms");
if (forceArgs.size() > 0)
forceArgs[7] = &interactingTiles->getDevicePointer();
findInteractingBlocksArgs[3] = &interactingTiles->getDevicePointer();
findInteractingBlocksArgs[6] = &interactingTiles->getDevicePointer();
if (forceArgs.size() > 0)
forceArgs[14] = &interactingAtoms->getDevicePointer();
findInteractingBlocksArgs[4] = &interactingAtoms->getDevicePointer();
forceArgs[17] = &interactingAtoms->getDevicePointer();
findInteractingBlocksArgs[7] = &interactingAtoms->getDevicePointer();
if (context.getUseDoublePrecision()) {
vector<double4> oldPositionsVec(numAtoms, make_double4(1e30, 1e30, 1e30, 0));
oldPositions->upload(oldPositionsVec);
......@@ -627,6 +633,9 @@ CUfunction CudaNonbondedUtilities::createInteractionKernel(const string& source,
forceArgs.push_back(&interactionCount->getDevicePointer());
forceArgs.push_back(context.getPeriodicBoxSizePointer());
forceArgs.push_back(context.getInvPeriodicBoxSizePointer());
forceArgs.push_back(context.getPeriodicBoxVecXPointer());
forceArgs.push_back(context.getPeriodicBoxVecYPointer());
forceArgs.push_back(context.getPeriodicBoxVecZPointer());
forceArgs.push_back(&maxTiles);
forceArgs.push_back(&blockCenter->getDevicePointer());
forceArgs.push_back(&blockBoundingBox->getDevicePointer());
......
platforms/cuda/src/kernels/customGBEnergyN2.cu
View file @ 3db6b8ee
......@@ -17,7 +17,8 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
const real4* __restrict__ posq, const unsigned int* __restrict__ exclusions, const ushort2* __restrict__ exclusionTiles,
#ifdef USE_CUTOFF
const int* __restrict__ tiles, const unsigned int* __restrict__ interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, const real4* __restrict__ blockCenter, const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms
real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, const real4* __restrict__ blockCenter,
const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms
#else
unsigned int numTiles
#endif
......@@ -56,9 +57,7 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
real3 pos2 = localData[atom2].pos;
real3 delta = make_real3(pos2.x-pos1.x, pos2.y-pos1.y, pos2.z-pos1.z);
#ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(delta)
#endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF
......@@ -109,9 +108,7 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
real3 pos2 = localData[atom2].pos;
real3 delta = make_real3(pos2.x-pos1.x, pos2.y-pos1.y, pos2.z-pos1.z);
#ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(delta)
#endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF
......@@ -254,12 +251,8 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
// box, then skip having to apply periodic boundary conditions later.
real4 blockCenterX = blockCenter[x];
pos1.x -= floor((pos1.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
pos1.y -= floor((pos1.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
pos1.z -= floor((pos1.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
localData[threadIdx.x].pos.x -= floor((localData[threadIdx.x].pos.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
localData[threadIdx.x].pos.y -= floor((localData[threadIdx.x].pos.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
localData[threadIdx.x].pos.z -= floor((localData[threadIdx.x].pos.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS_WITH_CENTER(pos1, blockCenterX)
APPLY_PERIODIC_TO_POS_WITH_CENTER(localData[threadIdx.x].pos, blockCenterX)
unsigned int tj = tgx;
for (j = 0; j < TILE_SIZE; j++) {
int atom2 = tbx+tj;
......@@ -306,9 +299,7 @@ extern "C" __global__ void computeN2Energy(unsigned long long* __restrict__ forc
real3 pos2 = localData[atom2].pos;
real3 delta = make_real3(pos2.x-pos1.x, pos2.y-pos1.y, pos2.z-pos1.z);
#ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(delta)
#endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF
......
platforms/cuda/src/kernels/customGBValueN2.cu
View file @ 3db6b8ee
......@@ -14,7 +14,8 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
const ushort2* __restrict__ exclusionTiles, unsigned long long* __restrict__ global_value,
#ifdef USE_CUTOFF
const int* __restrict__ tiles, const unsigned int* __restrict__ interactionCount, real4 periodicBoxSize, real4 invPeriodicBoxSize,
unsigned int maxTiles, const real4* __restrict__ blockCenter, const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms
real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, unsigned int maxTiles, const real4* __restrict__ blockCenter,
const real4* __restrict__ blockSize, const unsigned int* __restrict__ interactingAtoms
#else
unsigned int numTiles
#endif
......@@ -51,9 +52,7 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
real3 pos2 = localData[atom2].pos;
real3 delta = make_real3(pos2.x-pos1.x, pos2.y-pos1.y, pos2.z-pos1.z);
#ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(delta)
#endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF
......@@ -100,9 +99,7 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
real3 pos2 = localData[atom2].pos;
real3 delta = make_real3(pos2.x-pos1.x, pos2.y-pos1.y, pos2.z-pos1.z);
#ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(delta)
#endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF
......@@ -229,12 +226,8 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
// box, then skip having to apply periodic boundary conditions later.
real4 blockCenterX = blockCenter[x];
pos1.x -= floor((pos1.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
pos1.y -= floor((pos1.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
pos1.z -= floor((pos1.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
localData[threadIdx.x].pos.x -= floor((localData[threadIdx.x].pos.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
localData[threadIdx.x].pos.y -= floor((localData[threadIdx.x].pos.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
localData[threadIdx.x].pos.z -= floor((localData[threadIdx.x].pos.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS_WITH_CENTER(pos1, blockCenterX)
APPLY_PERIODIC_TO_POS_WITH_CENTER(localData[threadIdx.x].pos, blockCenterX)
unsigned int tj = tgx;
for (unsigned int j = 0; j < TILE_SIZE; j++) {
int atom2 = tbx+tj;
......@@ -268,9 +261,7 @@ extern "C" __global__ void computeN2Value(const real4* __restrict__ posq, const
real3 pos2 = localData[atom2].pos;
real3 delta = make_real3(pos2.x-pos1.x, pos2.y-pos1.y, pos2.z-pos1.z);
#ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(delta)
#endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF
......
platforms/cuda/src/kernels/customHbondForce.cu
View file @ 3db6b8ee
......@@ -25,12 +25,10 @@ inline __device__ real4 delta(real4 vec1, real4 vec2) {
* Compute the difference between two vectors, taking periodic boundary conditions into account
* and setting the fourth component to the squared magnitude.
*/
inline __device__ real4 deltaPeriodic(real4 vec1, real4 vec2, real4 periodicBoxSize, real4 invPeriodicBoxSize) {
inline __device__ real4 deltaPeriodic(real4 vec1, real4 vec2, real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ) {
real4 result = make_real4(vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0.0f);
#ifdef USE_PERIODIC
result.x -= floor(result.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
result.y -= floor(result.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
result.z -= floor(result.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(result)
#endif
result.w = result.x*result.x + result.y*result.y + result.z*result.z;
return result;
......@@ -69,7 +67,8 @@ inline __device__ real4 computeCross(real4 vec1, real4 vec2) {
* Compute forces on donors.
*/
extern "C" __global__ void computeDonorForces(unsigned long long* __restrict__ force, real* __restrict__ energyBuffer, const real4* __restrict__ posq,
const int4* __restrict__ exclusions, const int4* __restrict__ donorAtoms, const int4* __restrict__ acceptorAtoms, real4 periodicBoxSize, real4 invPeriodicBoxSize
const int4* __restrict__ exclusions, const int4* __restrict__ donorAtoms, const int4* __restrict__ acceptorAtoms, real4 periodicBoxSize, real4 invPeriodicBoxSize,
real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
PARAMETER_ARGUMENTS) {
extern __shared__ real4 posBuffer[];
real energy = 0;
......@@ -116,7 +115,7 @@ extern "C" __global__ void computeDonorForces(unsigned long long* __restrict__ f
real4 a1 = posBuffer[3*index];
real4 a2 = posBuffer[3*index+1];
real4 a3 = posBuffer[3*index+2];
real4 deltaD1A1 = deltaPeriodic(d1, a1, periodicBoxSize, invPeriodicBoxSize);
real4 deltaD1A1 = deltaPeriodic(d1, a1, periodicBoxSize, invPeriodicBoxSize, periodicBoxVecX, periodicBoxVecY, periodicBoxVecZ);
#ifdef USE_CUTOFF
if (deltaD1A1.w < CUTOFF_SQUARED) {
#endif
......@@ -157,7 +156,8 @@ extern "C" __global__ void computeDonorForces(unsigned long long* __restrict__ f
* Compute forces on acceptors.
*/
extern "C" __global__ void computeAcceptorForces(unsigned long long* __restrict__ force, real* __restrict__ energyBuffer, const real4* __restrict__ posq,
const int4* __restrict__ exclusions, const int4* __restrict__ donorAtoms, const int4* __restrict__ acceptorAtoms, real4 periodicBoxSize, real4 invPeriodicBoxSize
const int4* __restrict__ exclusions, const int4* __restrict__ donorAtoms, const int4* __restrict__ acceptorAtoms, real4 periodicBoxSize, real4 invPeriodicBoxSize,
real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
PARAMETER_ARGUMENTS) {
extern __shared__ real4 posBuffer[];
real3 f1 = make_real3(0);
......@@ -203,7 +203,7 @@ extern "C" __global__ void computeAcceptorForces(unsigned long long* __restrict_
real4 d1 = posBuffer[3*index];
real4 d2 = posBuffer[3*index+1];
real4 d3 = posBuffer[3*index+2];
real4 deltaD1A1 = deltaPeriodic(d1, a1, periodicBoxSize, invPeriodicBoxSize);
real4 deltaD1A1 = deltaPeriodic(d1, a1, periodicBoxSize, invPeriodicBoxSize, periodicBoxVecX, periodicBoxVecY, periodicBoxVecZ);
#ifdef USE_CUTOFF
if (deltaD1A1.w < CUTOFF_SQUARED) {
#endif
......
platforms/cuda/src/kernels/customManyParticle.cu
View file @ 3db6b8ee
......@@ -18,12 +18,10 @@ inline __device__ real3 trim(real4 v) {
* Compute the difference between two vectors, taking periodic boundary conditions into account
* and setting the fourth component to the squared magnitude.
*/
inline __device__ real4 delta(real3 vec1, real3 vec2, real4 periodicBoxSize, real4 invPeriodicBoxSize) {
inline __device__ real4 delta(real3 vec1, real3 vec2, real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ) {
real4 result = make_real4(vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0.0f);
#ifdef USE_PERIODIC
result.x -= floor(result.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
result.y -= floor(result.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
result.z -= floor(result.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(result)
#endif
result.w = result.x*result.x + result.y*result.y + result.z*result.z;
return result;
......@@ -81,7 +79,7 @@ __constant__ float globals[NUM_GLOBALS];
*/
extern "C" __global__ void computeInteraction(
unsigned long long* __restrict__ forceBuffers, real* __restrict__ energyBuffer, const real4* __restrict__ posq,
real4 periodicBoxSize, real4 invPeriodicBoxSize
real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
#ifdef USE_CUTOFF
, const int* __restrict__ neighbors, const int* __restrict__ neighborStartIndex
#endif
......@@ -144,16 +142,14 @@ extern "C" __global__ void computeInteraction(
/**
* Find a bounding box for the atoms in each block.
*/
extern "C" __global__ void findBlockBounds(real4 periodicBoxSize, real4 invPeriodicBoxSize, const real4* __restrict__ posq,
real4* __restrict__ blockCenter, real4* __restrict__ blockBoundingBox, int* __restrict__ numNeighborPairs) {
extern "C" __global__ void findBlockBounds(real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
const real4* __restrict__ posq, real4* __restrict__ blockCenter, real4* __restrict__ blockBoundingBox, int* __restrict__ numNeighborPairs) {
int index = blockIdx.x*blockDim.x+threadIdx.x;
int base = index*TILE_SIZE;
while (base < NUM_ATOMS) {
real4 pos = posq[base];
#ifdef USE_PERIODIC
pos.x -= floor(pos.x*invPeriodicBoxSize.x)*periodicBoxSize.x;
pos.y -= floor(pos.y*invPeriodicBoxSize.y)*periodicBoxSize.y;
pos.z -= floor(pos.z*invPeriodicBoxSize.z)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS(pos)
#endif
real4 minPos = pos;
real4 maxPos = pos;
......@@ -162,9 +158,7 @@ extern "C" __global__ void findBlockBounds(real4 periodicBoxSize, real4 invPerio
pos = posq[i];
#ifdef USE_PERIODIC
real4 center = 0.5f*(maxPos+minPos);
pos.x -= floor((pos.x-center.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
pos.y -= floor((pos.y-center.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
pos.z -= floor((pos.z-center.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS_WITH_CENTER(pos, center)
#endif
minPos = make_real4(min(minPos.x,pos.x), min(minPos.y,pos.y), min(minPos.z,pos.z), 0);
maxPos = make_real4(max(maxPos.x,pos.x), max(maxPos.y,pos.y), max(maxPos.z,pos.z), 0);
......@@ -182,8 +176,8 @@ extern "C" __global__ void findBlockBounds(real4 periodicBoxSize, real4 invPerio
/**
* Find a list of neighbors for each atom.
*/
extern "C" __global__ void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, const real4* __restrict__ posq,
const real4* __restrict__ blockCenter, const real4* __restrict__ blockBoundingBox, int2* __restrict__ neighborPairs,
extern "C" __global__ void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
const real4* __restrict__ posq, const real4* __restrict__ blockCenter, const real4* __restrict__ blockBoundingBox, int2* __restrict__ neighborPairs,
int* __restrict__ numNeighborPairs, int* __restrict__ numNeighborsForAtom, int maxNeighborPairs
#ifdef USE_EXCLUSIONS
, int* __restrict__ exclusions, int* __restrict__ exclusionStartIndex
......@@ -216,9 +210,7 @@ extern "C" __global__ void findNeighbors(real4 periodicBoxSize, real4 invPeriodi
real4 blockSize2 = blockBoundingBox[block2];
real4 blockDelta = blockCenter1-blockCenter2;
#ifdef USE_PERIODIC
blockDelta.x -= floor(blockDelta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
blockDelta.y -= floor(blockDelta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
blockDelta.z -= floor(blockDelta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(blockDelta)
#endif
blockDelta.x = max(0.0f, fabs(blockDelta.x)-blockSize1.x-blockSize2.x);
blockDelta.y = max(0.0f, fabs(blockDelta.y)-blockSize1.y-blockSize2.y);
......@@ -247,7 +239,7 @@ extern "C" __global__ void findNeighbors(real4 periodicBoxSize, real4 invPeriodi
// Decide whether to include this atom pair in the neighbor list.
real4 atomDelta = delta(pos1, pos2, periodicBoxSize, invPeriodicBoxSize);
real4 atomDelta = delta(pos1, pos2, periodicBoxSize, invPeriodicBoxSize, periodicBoxVecX, periodicBoxVecY, periodicBoxVecZ);
#ifdef USE_CENTRAL_PARTICLE
bool includeAtom = (atom2 != atom1 && atom2 < NUM_ATOMS && atomDelta.w < CUTOFF_SQUARED);
#else
......
platforms/cuda/src/kernels/customNonbondedGroups.cu
View file @ 3db6b8ee
......@@ -10,7 +10,7 @@ typedef struct {
extern "C" __global__ void computeInteractionGroups(
unsigned long long* __restrict__ forceBuffers, real* __restrict__ energyBuffer, const real4* __restrict__ posq, const int4* __restrict__ groupData,
real4 periodicBoxSize, real4 invPeriodicBoxSize
real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
PARAMETER_ARGUMENTS) {
const unsigned int totalWarps = (blockDim.x*gridDim.x)/TILE_SIZE;
const unsigned int warp = (blockIdx.x*blockDim.x+threadIdx.x)/TILE_SIZE; // global warpIndex
......@@ -47,9 +47,7 @@ extern "C" __global__ void computeInteractionGroups(
posq2 = make_real4(localData[localIndex].x, localData[localIndex].y, localData[localIndex].z, localData[localIndex].q);
real3 delta = make_real3(posq2.x-posq1.x, posq2.y-posq1.y, posq2.z-posq1.z);
#ifdef USE_PERIODIC
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(delta)
#endif
real r2 = delta.x*delta.x + delta.y*delta.y + delta.z*delta.z;
#ifdef USE_CUTOFF
......
platforms/cuda/src/kernels/findInteractingBlocks.cu
View file @ 3db6b8ee
......@@ -4,16 +4,15 @@
/**
* Find a bounding box for the atoms in each block.
*/
extern "C" __global__ void findBlockBounds(int numAtoms, real4 periodicBoxSize, real4 invPeriodicBoxSize, const real4* __restrict__ posq,
real4* __restrict__ blockCenter, real4* __restrict__ blockBoundingBox, int* __restrict__ rebuildNeighborList, real2* __restrict__ sortedBlocks) {
extern "C" __global__ void findBlockBounds(int numAtoms, real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
const real4* __restrict__ posq, real4* __restrict__ blockCenter, real4* __restrict__ blockBoundingBox, int* __restrict__ rebuildNeighborList,
real2* __restrict__ sortedBlocks) {
int index = blockIdx.x*blockDim.x+threadIdx.x;
int base = index*TILE_SIZE;
while (base < numAtoms) {
real4 pos = posq[base];
#ifdef USE_PERIODIC
pos.x -= floor(pos.x*invPeriodicBoxSize.x)*periodicBoxSize.x;
pos.y -= floor(pos.y*invPeriodicBoxSize.y)*periodicBoxSize.y;
pos.z -= floor(pos.z*invPeriodicBoxSize.z)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS(pos)
#endif
real4 minPos = pos;
real4 maxPos = pos;
......@@ -22,9 +21,7 @@ extern "C" __global__ void findBlockBounds(int numAtoms, real4 periodicBoxSize,
pos = posq[i];
#ifdef USE_PERIODIC
real4 center = 0.5f*(maxPos+minPos);
pos.x -= floor((pos.x-center.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
pos.y -= floor((pos.y-center.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
pos.z -= floor((pos.z-center.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS_WITH_CENTER(pos, center)
#endif
minPos = make_real4(min(minPos.x,pos.x), min(minPos.y,pos.y), min(minPos.z,pos.z), 0);
maxPos = make_real4(max(maxPos.x,pos.x), max(maxPos.y,pos.y), max(maxPos.z,pos.z), 0);
......@@ -116,11 +113,11 @@ extern "C" __global__ void sortBoxData(const real2* __restrict__ sortedBlock, co
* [in] rebuildNeighbourList - whether or not to execute this kernel
*
*/
extern "C" __global__ void findBlocksWithInteractions(real4 periodicBoxSize, real4 invPeriodicBoxSize, unsigned int* __restrict__ interactionCount,
int* __restrict__ interactingTiles, unsigned int* __restrict__ interactingAtoms, const real4* __restrict__ posq, unsigned int maxTiles, unsigned int startBlockIndex,
unsigned int numBlocks, real2* __restrict__ sortedBlocks, const real4* __restrict__ sortedBlockCenter, const real4* __restrict__ sortedBlockBoundingBox,
const unsigned int* __restrict__ exclusionIndices, const unsigned int* __restrict__ exclusionRowIndices, real4* __restrict__ oldPositions,
const int* __restrict__ rebuildNeighborList) {
extern "C" __global__ void findBlocksWithInteractions(real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ,
unsigned int* __restrict__ interactionCount, int* __restrict__ interactingTiles, unsigned int* __restrict__ interactingAtoms, const real4* __restrict__ posq,
unsigned int maxTiles, unsigned int startBlockIndex, unsigned int numBlocks, real2* __restrict__ sortedBlocks, const real4* __restrict__ sortedBlockCenter,
const real4* __restrict__ sortedBlockBoundingBox, const unsigned int* __restrict__ exclusionIndices, const unsigned int* __restrict__ exclusionRowIndices,
real4* __restrict__ oldPositions, const int* __restrict__ rebuildNeighborList) {
if (rebuildNeighborList[0] == 0)
return; // The neighbor list doesn't need to be rebuilt.
......@@ -157,9 +154,7 @@ extern "C" __global__ void findBlocksWithInteractions(real4 periodicBoxSize, rea
// The box is small enough that we can just translate all the atoms into a single periodic
// box, then skip having to apply periodic boundary conditions later.
pos1.x -= floor((pos1.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
pos1.y -= floor((pos1.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
pos1.z -= floor((pos1.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS_WITH_CENTER(pos1, blockCenterX)
}
#endif
posBuffer[threadIdx.x] = pos1;
......@@ -185,9 +180,7 @@ extern "C" __global__ void findBlocksWithInteractions(real4 periodicBoxSize, rea
real4 blockSizeY = (block2 < NUM_BLOCKS ? sortedBlockBoundingBox[block2] : make_real4(0));
real4 blockDelta = blockCenterX-blockCenterY;
#ifdef USE_PERIODIC
blockDelta.x -= floor(blockDelta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
blockDelta.y -= floor(blockDelta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
blockDelta.z -= floor(blockDelta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(blockDelta)
#endif
blockDelta.x = max(0.0f, fabs(blockDelta.x)-blockSizeX.x-blockSizeY.x);
blockDelta.y = max(0.0f, fabs(blockDelta.y)-blockSizeX.y-blockSizeY.y);
......@@ -215,9 +208,7 @@ extern "C" __global__ void findBlocksWithInteractions(real4 periodicBoxSize, rea
real3 pos2 = trimTo3(posq[atom2]);
#ifdef USE_PERIODIC
if (singlePeriodicCopy) {
pos2.x -= floor((pos2.x-blockCenterX.x)*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
pos2.y -= floor((pos2.y-blockCenterX.y)*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
pos2.z -= floor((pos2.z-blockCenterX.z)*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_POS_WITH_CENTER(pos2, blockCenterX)
}
#endif
bool interacts = false;
......@@ -226,9 +217,7 @@ extern "C" __global__ void findBlocksWithInteractions(real4 periodicBoxSize, rea
if (!singlePeriodicCopy) {
for (int j = 0; j < TILE_SIZE; j++) {
real3 delta = pos2-posBuffer[warpStart+j];
delta.x -= floor(delta.x*invPeriodicBoxSize.x+0.5f)*periodicBoxSize.x;
delta.y -= floor(delta.y*invPeriodicBoxSize.y+0.5f)*periodicBoxSize.y;
delta.z -= floor(delta.z*invPeriodicBoxSize.z+0.5f)*periodicBoxSize.z;
APPLY_PERIODIC_TO_DELTA(delta)
interacts |= (delta.x*delta.x+delta.y*delta.y+delta.z*delta.z < PADDED_CUTOFF_SQUARED);
}
}
......
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