Skip to content

GitLab

Commit 047934e2 authored by Rafal P. Wiewiora's avatar Rafal P. Wiewiora
Browse files

Merge remote-tracking branch 'upstream/master'

parents ce3a5dc0 d12c9bd1
Expand all Show whitespace changes
Inline Side-by-side
Showing with 808 additions and 536 deletions
platforms/cpu/src/CpuCustomManyParticleForce.cpp
View file @ 047934e2
/* Portions copyright (c) 2009-2014 Stanford University and Simbios. /* Portions copyright (c) 2009-2017 Stanford University and Simbios.
* Contributors: Peter Eastman * Contributors: Peter Eastman
* *
* Permission is hereby granted, free of charge, to any person obtaining * Permission is hereby granted, free of charge, to any person obtaining
...@@ -37,16 +37,6 @@ ...@@ -37,16 +37,6 @@
using namespace OpenMM; using namespace OpenMM;
using namespace std; using namespace std;
class CpuCustomManyParticleForce::ComputeForceTask : public ThreadPool::Task {
public:
ComputeForceTask(CpuCustomManyParticleForce& owner) : owner(owner) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadComputeForce(threads, threadIndex);
}
CpuCustomManyParticleForce& owner;
};
CpuCustomManyParticleForce::CpuCustomManyParticleForce(const CustomManyParticleForce& force, ThreadPool& threads) : CpuCustomManyParticleForce::CpuCustomManyParticleForce(const CustomManyParticleForce& force, ThreadPool& threads) :
threads(threads), useCutoff(false), usePeriodic(false), neighborList(NULL) { threads(threads), useCutoff(false), usePeriodic(false), neighborList(NULL) {
numParticles = force.getNumParticles(); numParticles = force.getNumParticles();
...@@ -98,7 +88,7 @@ CpuCustomManyParticleForce::~CpuCustomManyParticleForce() { ...@@ -98,7 +88,7 @@ CpuCustomManyParticleForce::~CpuCustomManyParticleForce() {
delete threadData[i]; delete threadData[i];
} }
void CpuCustomManyParticleForce::calculateIxn(AlignedArray<float>& posq, RealOpenMM** particleParameters, void CpuCustomManyParticleForce::calculateIxn(AlignedArray<float>& posq, double** particleParameters,
const map<string, double>& globalParameters, vector<AlignedArray<float> >& threadForce, const map<string, double>& globalParameters, vector<AlignedArray<float> >& threadForce,
bool includeForces, bool includeEnergy, double& energy) { bool includeForces, bool includeEnergy, double& energy) {
// Record the parameters for the threads. // Record the parameters for the threads.
...@@ -141,8 +131,7 @@ void CpuCustomManyParticleForce::calculateIxn(AlignedArray<float>& posq, RealOpe ...@@ -141,8 +131,7 @@ void CpuCustomManyParticleForce::calculateIxn(AlignedArray<float>& posq, RealOpe
// Signal the threads to start running and wait for them to finish. // Signal the threads to start running and wait for them to finish.
ComputeForceTask task(*this); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadComputeForce(threads, threadIndex); });
threads.execute(task);
threads.waitForThreads(); threads.waitForThreads();
// Combine the energies from all the threads. // Combine the energies from all the threads.
...@@ -191,14 +180,14 @@ void CpuCustomManyParticleForce::threadComputeForce(ThreadPool& threads, int thr ...@@ -191,14 +180,14 @@ void CpuCustomManyParticleForce::threadComputeForce(ThreadPool& threads, int thr
} }
} }
void CpuCustomManyParticleForce::setUseCutoff(RealOpenMM distance) { void CpuCustomManyParticleForce::setUseCutoff(double distance) {
useCutoff = true; useCutoff = true;
cutoffDistance = distance; cutoffDistance = distance;
if (neighborList == NULL) if (neighborList == NULL)
neighborList = new CpuNeighborList(4); neighborList = new CpuNeighborList(4);
} }
void CpuCustomManyParticleForce::setPeriodic(RealVec* periodicBoxVectors) { void CpuCustomManyParticleForce::setPeriodic(Vec3* periodicBoxVectors) {
assert(useCutoff); assert(useCutoff);
assert(periodicBoxVectors[0][0] >= 2.0*cutoffDistance); assert(periodicBoxVectors[0][0] >= 2.0*cutoffDistance);
assert(periodicBoxVectors[1][1] >= 2.0*cutoffDistance); assert(periodicBoxVectors[1][1] >= 2.0*cutoffDistance);
...@@ -220,7 +209,7 @@ void CpuCustomManyParticleForce::setPeriodic(RealVec* periodicBoxVectors) { ...@@ -220,7 +209,7 @@ void CpuCustomManyParticleForce::setPeriodic(RealVec* periodicBoxVectors) {
} }
void CpuCustomManyParticleForce::loopOverInteractions(vector<int>& availableParticles, vector<int>& particleSet, int loopIndex, int startIndex, void CpuCustomManyParticleForce::loopOverInteractions(vector<int>& availableParticles, vector<int>& particleSet, int loopIndex, int startIndex,
RealOpenMM** particleParameters, float* forces, ThreadData& data, const fvec4& boxSize, const fvec4& invBoxSize) { double** particleParameters, float* forces, ThreadData& data, const fvec4& boxSize, const fvec4& invBoxSize) {
int numParticles = availableParticles.size(); int numParticles = availableParticles.size();
double cutoff2 = cutoffDistance*cutoffDistance; double cutoff2 = cutoffDistance*cutoffDistance;
int checkRange = (centralParticleMode ? 1 : loopIndex); int checkRange = (centralParticleMode ? 1 : loopIndex);
...@@ -254,7 +243,7 @@ void CpuCustomManyParticleForce::loopOverInteractions(vector<int>& availablePart ...@@ -254,7 +243,7 @@ void CpuCustomManyParticleForce::loopOverInteractions(vector<int>& availablePart
} }
} }
void CpuCustomManyParticleForce::calculateOneIxn(vector<int>& particleSet, RealOpenMM** particleParameters, float* forces, ThreadData& data, const fvec4& boxSize, const fvec4& invBoxSize) { void CpuCustomManyParticleForce::calculateOneIxn(vector<int>& particleSet, double** particleParameters, float* forces, ThreadData& data, const fvec4& boxSize, const fvec4& invBoxSize) {
// Select the ordering to use for the particles. // Select the ordering to use for the particles.
vector<int>& permutedParticles = data.permutedParticles; vector<int>& permutedParticles = data.permutedParticles;
......
platforms/cpu/src/CpuCustomNonbondedForce.cpp
View file @ 047934e2
/* Portions copyright (c) 2009-2016 Stanford University and Simbios. /* Portions copyright (c) 2009-2017 Stanford University and Simbios.
* Contributors: Peter Eastman * Contributors: Peter Eastman
* *
* Permission is hereby granted, free of charge, to any person obtaining * Permission is hereby granted, free of charge, to any person obtaining
...@@ -33,16 +33,6 @@ ...@@ -33,16 +33,6 @@
using namespace OpenMM; using namespace OpenMM;
using namespace std; using namespace std;
class CpuCustomNonbondedForce::ComputeForceTask : public ThreadPool::Task {
public:
ComputeForceTask(CpuCustomNonbondedForce& owner) : owner(owner) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadComputeForce(threads, threadIndex);
}
CpuCustomNonbondedForce& owner;
};
CpuCustomNonbondedForce::ThreadData::ThreadData(const Lepton::CompiledExpression& energyExpression, const Lepton::CompiledExpression& forceExpression, CpuCustomNonbondedForce::ThreadData::ThreadData(const Lepton::CompiledExpression& energyExpression, const Lepton::CompiledExpression& forceExpression,
const vector<string>& parameterNames, const std::vector<Lepton::CompiledExpression> energyParamDerivExpressions) : const vector<string>& parameterNames, const std::vector<Lepton::CompiledExpression> energyParamDerivExpressions) :
energyExpression(energyExpression), forceExpression(forceExpression), energyParamDerivExpressions(energyParamDerivExpressions) { energyExpression(energyExpression), forceExpression(forceExpression), energyParamDerivExpressions(energyParamDerivExpressions) {
...@@ -70,7 +60,7 @@ CpuCustomNonbondedForce::ThreadData::ThreadData(const Lepton::CompiledExpression ...@@ -70,7 +60,7 @@ CpuCustomNonbondedForce::ThreadData::ThreadData(const Lepton::CompiledExpression
CpuCustomNonbondedForce::CpuCustomNonbondedForce(const Lepton::CompiledExpression& energyExpression, CpuCustomNonbondedForce::CpuCustomNonbondedForce(const Lepton::CompiledExpression& energyExpression,
const Lepton::CompiledExpression& forceExpression, const vector<string>& parameterNames, const vector<set<int> >& exclusions, const Lepton::CompiledExpression& forceExpression, const vector<string>& parameterNames, const vector<set<int> >& exclusions,
const std::vector<Lepton::CompiledExpression> energyParamDerivExpressions, ThreadPool& threads) : const std::vector<Lepton::CompiledExpression> energyParamDerivExpressions, ThreadPool& threads) :
cutoff(false), useSwitch(false), periodic(false), paramNames(parameterNames), exclusions(exclusions), threads(threads) { cutoff(false), useSwitch(false), periodic(false), useInteractionGroups(false), paramNames(parameterNames), exclusions(exclusions), threads(threads) {
for (int i = 0; i < threads.getNumThreads(); i++) for (int i = 0; i < threads.getNumThreads(); i++)
threadData.push_back(new ThreadData(energyExpression, forceExpression, parameterNames, energyParamDerivExpressions)); threadData.push_back(new ThreadData(energyExpression, forceExpression, parameterNames, energyParamDerivExpressions));
} }
...@@ -80,13 +70,14 @@ CpuCustomNonbondedForce::~CpuCustomNonbondedForce() { ...@@ -80,13 +70,14 @@ CpuCustomNonbondedForce::~CpuCustomNonbondedForce() {
delete threadData[i]; delete threadData[i];
} }
void CpuCustomNonbondedForce::setUseCutoff(RealOpenMM distance, const CpuNeighborList& neighbors) { void CpuCustomNonbondedForce::setUseCutoff(double distance, const CpuNeighborList& neighbors) {
cutoff = true; cutoff = true;
cutoffDistance = distance; cutoffDistance = distance;
neighborList = &neighbors; neighborList = &neighbors;
} }
void CpuCustomNonbondedForce::setInteractionGroups(const vector<pair<set<int>, set<int> > >& groups) { void CpuCustomNonbondedForce::setInteractionGroups(const vector<pair<set<int>, set<int> > >& groups) {
useInteractionGroups = true;
for (int group = 0; group < (int) groups.size(); group++) { for (int group = 0; group < (int) groups.size(); group++) {
const set<int>& set1 = groups[group].first; const set<int>& set1 = groups[group].first;
const set<int>& set2 = groups[group].second; const set<int>& set2 = groups[group].second;
...@@ -102,12 +93,12 @@ void CpuCustomNonbondedForce::setInteractionGroups(const vector<pair<set<int>, s ...@@ -102,12 +93,12 @@ void CpuCustomNonbondedForce::setInteractionGroups(const vector<pair<set<int>, s
} }
} }
void CpuCustomNonbondedForce::setUseSwitchingFunction(RealOpenMM distance) { void CpuCustomNonbondedForce::setUseSwitchingFunction(double distance) {
useSwitch = true; useSwitch = true;
switchingDistance = distance; switchingDistance = distance;
} }
void CpuCustomNonbondedForce::setPeriodic(RealVec* periodicBoxVectors) { void CpuCustomNonbondedForce::setPeriodic(Vec3* periodicBoxVectors) {
assert(cutoff); assert(cutoff);
assert(periodicBoxVectors[0][0] >= 2.0*cutoffDistance); assert(periodicBoxVectors[0][0] >= 2.0*cutoffDistance);
assert(periodicBoxVectors[1][1] >= 2.0*cutoffDistance); assert(periodicBoxVectors[1][1] >= 2.0*cutoffDistance);
...@@ -129,8 +120,8 @@ void CpuCustomNonbondedForce::setPeriodic(RealVec* periodicBoxVectors) { ...@@ -129,8 +120,8 @@ void CpuCustomNonbondedForce::setPeriodic(RealVec* periodicBoxVectors) {
} }
void CpuCustomNonbondedForce::calculatePairIxn(int numberOfAtoms, float* posq, vector<RealVec>& atomCoordinates, RealOpenMM** atomParameters, void CpuCustomNonbondedForce::calculatePairIxn(int numberOfAtoms, float* posq, vector<Vec3>& atomCoordinates, double** atomParameters,
RealOpenMM* fixedParameters, const map<string, double>& globalParameters, double* fixedParameters, const map<string, double>& globalParameters,
vector<AlignedArray<float> >& threadForce, bool includeForce, bool includeEnergy, double& totalEnergy, double* energyParamDerivs) { vector<AlignedArray<float> >& threadForce, bool includeForce, bool includeEnergy, double& totalEnergy, double* energyParamDerivs) {
// Record the parameters for the threads. // Record the parameters for the threads.
...@@ -149,8 +140,7 @@ void CpuCustomNonbondedForce::calculatePairIxn(int numberOfAtoms, float* posq, v ...@@ -149,8 +140,7 @@ void CpuCustomNonbondedForce::calculatePairIxn(int numberOfAtoms, float* posq, v
// Signal the threads to start running and wait for them to finish. // Signal the threads to start running and wait for them to finish.
ComputeForceTask task(*this); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadComputeForce(threads, threadIndex); });
threads.execute(task);
threads.waitForThreads(); threads.waitForThreads();
// Combine the energies from all the threads. // Combine the energies from all the threads.
...@@ -183,7 +173,7 @@ void CpuCustomNonbondedForce::threadComputeForce(ThreadPool& threads, int thread ...@@ -183,7 +173,7 @@ void CpuCustomNonbondedForce::threadComputeForce(ThreadPool& threads, int thread
data.energyParamDerivs[i] = 0.0; data.energyParamDerivs[i] = 0.0;
fvec4 boxSize(periodicBoxVectors[0][0], periodicBoxVectors[1][1], periodicBoxVectors[2][2], 0); fvec4 boxSize(periodicBoxVectors[0][0], periodicBoxVectors[1][1], periodicBoxVectors[2][2], 0);
fvec4 invBoxSize(recipBoxSize[0], recipBoxSize[1], recipBoxSize[2], 0); fvec4 invBoxSize(recipBoxSize[0], recipBoxSize[1], recipBoxSize[2], 0);
if (groupInteractions.size() > 0) { if (useInteractionGroups) {
// The user has specified interaction groups, so compute only the requested interactions. // The user has specified interaction groups, so compute only the requested interactions.
while (true) { while (true) {
......
platforms/cpu/src/CpuGBSAOBCForce.cpp
View file @ 047934e2
/* Portions copyright (c) 2006-2017 Stanford University and Simbios.
/* Portions copyright (c) 2006-2013 Stanford University and Simbios.
* Contributors: Pande Group * Contributors: Pande Group
* *
* Permission is hereby granted, free of charge, to any person obtaining * Permission is hereby granted, free of charge, to any person obtaining
...@@ -37,16 +36,6 @@ const int CpuGBSAOBCForce::NUM_TABLE_POINTS = 4096; ...@@ -37,16 +36,6 @@ const int CpuGBSAOBCForce::NUM_TABLE_POINTS = 4096;
const float CpuGBSAOBCForce::TABLE_MIN = 0.25f; const float CpuGBSAOBCForce::TABLE_MIN = 0.25f;
const float CpuGBSAOBCForce::TABLE_MAX = 1.5f; const float CpuGBSAOBCForce::TABLE_MAX = 1.5f;
class CpuGBSAOBCForce::ComputeTask : public ThreadPool::Task {
public:
ComputeTask(CpuGBSAOBCForce& owner) : owner(owner) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadComputeForce(threads, threadIndex);
}
CpuGBSAOBCForce& owner;
};
CpuGBSAOBCForce::CpuGBSAOBCForce() : cutoff(false), periodic(false) { CpuGBSAOBCForce::CpuGBSAOBCForce() : cutoff(false), periodic(false) {
logDX = (TABLE_MAX-TABLE_MIN)/NUM_TABLE_POINTS; logDX = (TABLE_MAX-TABLE_MIN)/NUM_TABLE_POINTS;
logDXInv = 1.0f/logDX; logDXInv = 1.0f/logDX;
...@@ -89,6 +78,10 @@ void CpuGBSAOBCForce::setParticleParameters(const std::vector<std::pair<float, f ...@@ -89,6 +78,10 @@ void CpuGBSAOBCForce::setParticleParameters(const std::vector<std::pair<float, f
particleParams = params; particleParams = params;
bornRadii.resize(params.size()+3); bornRadii.resize(params.size()+3);
obcChain.resize(params.size()+3); obcChain.resize(params.size()+3);
for (int i = bornRadii.size()-3; i < bornRadii.size(); i++) {
bornRadii[i] = 0;
obcChain[i] = 0;
}
} }
void CpuGBSAOBCForce::computeForce(const AlignedArray<float>& posq, vector<AlignedArray<float> >& threadForce, double* totalEnergy, ThreadPool& threads) { void CpuGBSAOBCForce::computeForce(const AlignedArray<float>& posq, vector<AlignedArray<float> >& threadForce, double* totalEnergy, ThreadPool& threads) {
...@@ -107,9 +100,8 @@ void CpuGBSAOBCForce::computeForce(const AlignedArray<float>& posq, vector<Align ...@@ -107,9 +100,8 @@ void CpuGBSAOBCForce::computeForce(const AlignedArray<float>& posq, vector<Align
// Signal the threads to start running and wait for them to finish. // Signal the threads to start running and wait for them to finish.
ComputeTask task(*this);
gmx_atomic_set(&counter, 0); gmx_atomic_set(&counter, 0);
threads.execute(task); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadComputeForce(threads, threadIndex); });
threads.waitForThreads(); // Compute Born radii threads.waitForThreads(); // Compute Born radii
gmx_atomic_set(&counter, 0); gmx_atomic_set(&counter, 0);
threads.resumeThreads(); threads.resumeThreads();
......
platforms/cpu/src/CpuGayBerneForce.cpp
View file @ 047934e2
...@@ -6,7 +6,7 @@ ...@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for * * Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. * * Medical Research, grant U54 GM072970. See https://simtk.org. *
* * * *
* Portions copyright (c) 2016 Stanford University and the Authors. * * Portions copyright (c) 2016-2017 Stanford University and the Authors. *
* Authors: Peter Eastman * * Authors: Peter Eastman *
* Contributors: * * Contributors: *
* * * *
...@@ -44,17 +44,6 @@ ...@@ -44,17 +44,6 @@
using namespace OpenMM; using namespace OpenMM;
using namespace std; using namespace std;
class CpuGayBerneForce::ComputeTask : public ThreadPool::Task {
public:
ComputeTask(CpuGayBerneForce& owner, CpuNeighborList* neighborList) : owner(owner), neighborList(neighborList) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadComputeForce(threads, threadIndex, neighborList);
}
CpuGayBerneForce& owner;
CpuNeighborList* neighborList;
};
CpuGayBerneForce::CpuGayBerneForce(const GayBerneForce& force) { CpuGayBerneForce::CpuGayBerneForce(const GayBerneForce& force) {
// Record the force parameters. // Record the force parameters.
...@@ -111,7 +100,7 @@ const vector<set<int> >& CpuGayBerneForce::getExclusions() const { ...@@ -111,7 +100,7 @@ const vector<set<int> >& CpuGayBerneForce::getExclusions() const {
return particleExclusions; return particleExclusions;
} }
RealOpenMM CpuGayBerneForce::calculateForce(const vector<RealVec>& positions, std::vector<RealVec>& forces, std::vector<AlignedArray<float> >& threadForce, RealVec* boxVectors, CpuPlatform::PlatformData& data) { double CpuGayBerneForce::calculateForce(const vector<Vec3>& positions, std::vector<Vec3>& forces, std::vector<AlignedArray<float> >& threadForce, Vec3* boxVectors, CpuPlatform::PlatformData& data) {
if (nonbondedMethod == GayBerneForce::CutoffPeriodic) { if (nonbondedMethod == GayBerneForce::CutoffPeriodic) {
double minAllowedSize = 1.999999*cutoffDistance; double minAllowedSize = 1.999999*cutoffDistance;
if (boxVectors[0][0] < minAllowedSize || boxVectors[1][1] < minAllowedSize || boxVectors[2][2] < minAllowedSize) if (boxVectors[0][0] < minAllowedSize || boxVectors[1][1] < minAllowedSize || boxVectors[2][2] < minAllowedSize)
...@@ -137,8 +126,7 @@ RealOpenMM CpuGayBerneForce::calculateForce(const vector<RealVec>& positions, st ...@@ -137,8 +126,7 @@ RealOpenMM CpuGayBerneForce::calculateForce(const vector<RealVec>& positions, st
// Signal the threads to compute the pairwise interactions. // Signal the threads to compute the pairwise interactions.
ComputeTask task(*this, data.neighborList); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadComputeForce(threads, threadIndex, data.neighborList); });
threads.execute(task);
threads.waitForThreads(); threads.waitForThreads();
// Signal the threads to compute exceptions. // Signal the threads to compute exceptions.
...@@ -164,10 +152,10 @@ void CpuGayBerneForce::threadComputeForce(ThreadPool& threads, int threadIndex, ...@@ -164,10 +152,10 @@ void CpuGayBerneForce::threadComputeForce(ThreadPool& threads, int threadIndex,
int numThreads = threads.getNumThreads(); int numThreads = threads.getNumThreads();
threadEnergy[threadIndex] = 0; threadEnergy[threadIndex] = 0;
float* forces = &(*threadForce)[threadIndex][0]; float* forces = &(*threadForce)[threadIndex][0];
vector<RealVec>& torques = threadTorque[threadIndex]; vector<Vec3>& torques = threadTorque[threadIndex];
torques.resize(numParticles); torques.resize(numParticles);
for (int i = 0; i < numParticles; i++) for (int i = 0; i < numParticles; i++)
torques[i] = RealVec(); torques[i] = Vec3();
double energy = 0.0; double energy = 0.0;
// Compute this thread's subset of interactions. // Compute this thread's subset of interactions.
...@@ -184,8 +172,8 @@ void CpuGayBerneForce::threadComputeForce(ThreadPool& threads, int threadIndex, ...@@ -184,8 +172,8 @@ void CpuGayBerneForce::threadComputeForce(ThreadPool& threads, int threadIndex,
continue; continue;
if (particleExclusions[i].find(j) != particleExclusions[i].end()) if (particleExclusions[i].find(j) != particleExclusions[i].end())
continue; // This interaction will be handled by an exception. continue; // This interaction will be handled by an exception.
RealOpenMM sigma = particles[i].sigmaOver2+particles[j].sigmaOver2; double sigma = particles[i].sigmaOver2+particles[j].sigmaOver2;
RealOpenMM epsilon = particles[i].sqrtEpsilon*particles[j].sqrtEpsilon; double epsilon = particles[i].sqrtEpsilon*particles[j].sqrtEpsilon;
energy += computeOneInteraction(i, j, sigma, epsilon, positions, forces, torques, boxVectors); energy += computeOneInteraction(i, j, sigma, epsilon, positions, forces, torques, boxVectors);
} }
} }
...@@ -208,8 +196,8 @@ void CpuGayBerneForce::threadComputeForce(ThreadPool& threads, int threadIndex, ...@@ -208,8 +196,8 @@ void CpuGayBerneForce::threadComputeForce(ThreadPool& threads, int threadIndex,
int second = blockAtom[k]; int second = blockAtom[k];
if (particles[second].sqrtEpsilon == 0.0f) if (particles[second].sqrtEpsilon == 0.0f)
continue; continue;
RealOpenMM sigma = particles[first].sigmaOver2+particles[second].sigmaOver2; double sigma = particles[first].sigmaOver2+particles[second].sigmaOver2;
RealOpenMM epsilon = particles[first].sqrtEpsilon*particles[second].sqrtEpsilon; double epsilon = particles[first].sqrtEpsilon*particles[second].sqrtEpsilon;
energy += computeOneInteraction(first, second, sigma, epsilon, positions, forces, torques, boxVectors); energy += computeOneInteraction(first, second, sigma, epsilon, positions, forces, torques, boxVectors);
} }
} }
...@@ -235,39 +223,39 @@ void CpuGayBerneForce::threadComputeForce(ThreadPool& threads, int threadIndex, ...@@ -235,39 +223,39 @@ void CpuGayBerneForce::threadComputeForce(ThreadPool& threads, int threadIndex,
threadEnergy[threadIndex] = energy; threadEnergy[threadIndex] = energy;
} }
void CpuGayBerneForce::computeEllipsoidFrames(const vector<RealVec>& positions) { void CpuGayBerneForce::computeEllipsoidFrames(const vector<Vec3>& positions) {
int numParticles = particles.size(); int numParticles = particles.size();
for (int particle = 0; particle < numParticles; particle++) { for (int particle = 0; particle < numParticles; particle++) {
ParticleInfo& p = particles[particle]; ParticleInfo& p = particles[particle];
// Compute the local coordinate system of the ellipsoid; // Compute the local coordinate system of the ellipsoid;
RealVec xdir, ydir, zdir; Vec3 xdir, ydir, zdir;
if (p.xparticle == -1) { if (p.xparticle == -1) {
xdir = RealVec(1, 0, 0); xdir = Vec3(1, 0, 0);
ydir = RealVec(0, 1, 0); ydir = Vec3(0, 1, 0);
} }
else { else {
xdir = positions[particle]-positions[p.xparticle]; xdir = positions[particle]-positions[p.xparticle];
xdir /= SQRT(xdir.dot(xdir)); xdir /= sqrt(xdir.dot(xdir));
if (p.yparticle == -1) { if (p.yparticle == -1) {
if (xdir[1] > -0.5 && xdir[1] < 0.5) if (xdir[1] > -0.5 && xdir[1] < 0.5)
ydir = RealVec(0, 1, 0); ydir = Vec3(0, 1, 0);
else else
ydir = RealVec(1, 0, 0); ydir = Vec3(1, 0, 0);
} }
else else
ydir = positions[particle]-positions[p.yparticle]; ydir = positions[particle]-positions[p.yparticle];
ydir -= xdir*(xdir.dot(ydir)); ydir -= xdir*(xdir.dot(ydir));
ydir /= SQRT(ydir.dot(ydir)); ydir /= sqrt(ydir.dot(ydir));
} }
zdir = xdir.cross(ydir); zdir = xdir.cross(ydir);
// Compute matrices we will need later. // Compute matrices we will need later.
RealOpenMM (&a)[3][3] = A[particle].v; double (&a)[3][3] = A[particle].v;
RealOpenMM (&b)[3][3] = B[particle].v; double (&b)[3][3] = B[particle].v;
RealOpenMM (&g)[3][3] = G[particle].v; double (&g)[3][3] = G[particle].v;
a[0][0] = xdir[0]; a[0][0] = xdir[0];
a[0][1] = xdir[1]; a[0][1] = xdir[1];
a[0][2] = xdir[2]; a[0][2] = xdir[2];
...@@ -277,8 +265,8 @@ void CpuGayBerneForce::computeEllipsoidFrames(const vector<RealVec>& positions) ...@@ -277,8 +265,8 @@ void CpuGayBerneForce::computeEllipsoidFrames(const vector<RealVec>& positions)
a[2][0] = zdir[0]; a[2][0] = zdir[0];
a[2][1] = zdir[1]; a[2][1] = zdir[1];
a[2][2] = zdir[2]; a[2][2] = zdir[2];
RealVec r2(p.rx*p.rx, p.ry*p.ry, p.rz*p.rz); Vec3 r2(p.rx*p.rx, p.ry*p.ry, p.rz*p.rz);
RealVec e2(1/sqrt(p.ex), 1/sqrt(p.ey), 1/sqrt(p.ez)); Vec3 e2(1/sqrt(p.ex), 1/sqrt(p.ey), 1/sqrt(p.ez));
for (int i = 0; i < 3; i++) for (int i = 0; i < 3; i++)
for (int j = 0; j < 3; j++) { for (int j = 0; j < 3; j++) {
b[i][j] = 0; b[i][j] = 0;
...@@ -291,33 +279,33 @@ void CpuGayBerneForce::computeEllipsoidFrames(const vector<RealVec>& positions) ...@@ -291,33 +279,33 @@ void CpuGayBerneForce::computeEllipsoidFrames(const vector<RealVec>& positions)
} }
} }
void CpuGayBerneForce::applyTorques(const vector<RealVec>& positions, vector<RealVec>& forces) { void CpuGayBerneForce::applyTorques(const vector<Vec3>& positions, vector<Vec3>& forces) {
int numParticles = particles.size(); int numParticles = particles.size();
int numThreads = threadTorque.size(); int numThreads = threadTorque.size();
for (int particle = 0; particle < numParticles; particle++) { for (int particle = 0; particle < numParticles; particle++) {
ParticleInfo& p = particles[particle]; ParticleInfo& p = particles[particle];
RealVec pos = positions[particle]; Vec3 pos = positions[particle];
if (p.xparticle != -1) { if (p.xparticle != -1) {
// Add up the torques from the individual threads. // Add up the torques from the individual threads.
RealVec torque; Vec3 torque;
for (int i = 0; i < numThreads; i++) for (int i = 0; i < numThreads; i++)
torque += threadTorque[i][particle]; torque += threadTorque[i][particle];
// Apply a force to the x particle. // Apply a force to the x particle.
RealVec dx = positions[p.xparticle]-pos; Vec3 dx = positions[p.xparticle]-pos;
double dx2 = dx.dot(dx); double dx2 = dx.dot(dx);
RealVec f = torque.cross(dx)/dx2; Vec3 f = torque.cross(dx)/dx2;
forces[p.xparticle] += f; forces[p.xparticle] += f;
forces[particle] -= f; forces[particle] -= f;
if (p.yparticle != -1) { if (p.yparticle != -1) {
// Apply a force to the y particle. This is based on the component of the torque // Apply a force to the y particle. This is based on the component of the torque
// that was not already applied to the x particle. // that was not already applied to the x particle.
RealVec dy = positions[p.yparticle]-pos; Vec3 dy = positions[p.yparticle]-pos;
double dy2 = dy.dot(dy); double dy2 = dy.dot(dy);
RealVec torque2 = dx*(torque.dot(dx)/dx2); Vec3 torque2 = dx*(torque.dot(dx)/dx2);
f = torque2.cross(dy)/dy2; f = torque2.cross(dy)/dy2;
forces[p.yparticle] += f; forces[p.yparticle] += f;
forces[particle] -= f; forces[particle] -= f;
...@@ -326,27 +314,27 @@ void CpuGayBerneForce::applyTorques(const vector<RealVec>& positions, vector<Rea ...@@ -326,27 +314,27 @@ void CpuGayBerneForce::applyTorques(const vector<RealVec>& positions, vector<Rea
} }
} }
RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2, RealOpenMM sigma, RealOpenMM epsilon, const RealVec* positions, double CpuGayBerneForce::computeOneInteraction(int particle1, int particle2, double sigma, double epsilon, const Vec3* positions,
float* forces, vector<RealVec>& torques, const RealVec* boxVectors) { float* forces, vector<Vec3>& torques, const Vec3* boxVectors) {
// Compute the displacement and check against the cutoff. // Compute the displacement and check against the cutoff.
RealOpenMM deltaR[ReferenceForce::LastDeltaRIndex]; double deltaR[ReferenceForce::LastDeltaRIndex];
if (nonbondedMethod == GayBerneForce::CutoffPeriodic) if (nonbondedMethod == GayBerneForce::CutoffPeriodic)
ReferenceForce::getDeltaRPeriodic(positions[particle2], positions[particle1], boxVectors, deltaR); ReferenceForce::getDeltaRPeriodic(positions[particle2], positions[particle1], boxVectors, deltaR);
else else
ReferenceForce::getDeltaR(positions[particle2], positions[particle1], deltaR); ReferenceForce::getDeltaR(positions[particle2], positions[particle1], deltaR);
RealOpenMM r = deltaR[ReferenceForce::RIndex]; double r = deltaR[ReferenceForce::RIndex];
if (nonbondedMethod != GayBerneForce::NoCutoff && r >= cutoffDistance) if (nonbondedMethod != GayBerneForce::NoCutoff && r >= cutoffDistance)
return 0; return 0;
RealOpenMM rInv = 1/r; double rInv = 1/r;
RealVec dr(deltaR[ReferenceForce::XIndex], deltaR[ReferenceForce::YIndex], deltaR[ReferenceForce::ZIndex]); Vec3 dr(deltaR[ReferenceForce::XIndex], deltaR[ReferenceForce::YIndex], deltaR[ReferenceForce::ZIndex]);
RealVec drUnit = dr*rInv; Vec3 drUnit = dr*rInv;
// Compute the switching function. // Compute the switching function.
RealOpenMM switchValue = 1, switchDeriv = 0; double switchValue = 1, switchDeriv = 0;
if (useSwitchingFunction && r > switchingDistance) { if (useSwitchingFunction && r > switchingDistance) {
RealOpenMM t = (r-switchingDistance)/(cutoffDistance-switchingDistance); double t = (r-switchingDistance)/(cutoffDistance-switchingDistance);
switchValue = 1+t*t*t*(-10+t*(15-t*6)); switchValue = 1+t*t*t*(-10+t*(15-t*6));
switchDeriv = t*t*(-30+t*(60-t*30))/(cutoffDistance-switchingDistance); switchDeriv = t*t*(-30+t*(60-t*30))/(cutoffDistance-switchingDistance);
} }
...@@ -354,11 +342,11 @@ RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2, ...@@ -354,11 +342,11 @@ RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2,
// Interactions between two point particles can be computed more easily. // Interactions between two point particles can be computed more easily.
if (particles[particle1].isPointParticle && particles[particle2].isPointParticle) { if (particles[particle1].isPointParticle && particles[particle2].isPointParticle) {
RealOpenMM sig = sigma*rInv; double sig = sigma*rInv;
RealOpenMM sig2 = sig*sig; double sig2 = sig*sig;
RealOpenMM sig6 = sig2*sig2*sig2; double sig6 = sig2*sig2*sig2;
RealOpenMM energy = 4*epsilon*(sig6-1)*sig6; double energy = 4*epsilon*(sig6-1)*sig6;
RealVec force = drUnit*(switchValue*4*epsilon*(12*sig6 - 6)*sig6*rInv - energy*switchDeriv); Vec3 force = drUnit*(switchValue*4*epsilon*(12*sig6 - 6)*sig6*rInv - energy*switchDeriv);
forces[4*particle1] += force[0]; forces[4*particle1] += force[0];
forces[4*particle1+1] += force[1]; forces[4*particle1+1] += force[1];
forces[4*particle1+2] += force[2]; forces[4*particle1+2] += force[2];
...@@ -374,31 +362,31 @@ RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2, ...@@ -374,31 +362,31 @@ RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2,
Matrix G12 = G[particle1]+G[particle2]; Matrix G12 = G[particle1]+G[particle2];
Matrix B12inv = B12.inverse(); Matrix B12inv = B12.inverse();
Matrix G12inv = G12.inverse(); Matrix G12inv = G12.inverse();
RealOpenMM detG12 = G12.determinant(); double detG12 = G12.determinant();
// Estimate the distance between the ellipsoids and compute the first terms needed for the energy. // Estimate the distance between the ellipsoids and compute the first terms needed for the energy.
RealOpenMM sigma12 = 1/SQRT(0.5*drUnit.dot(G12inv*drUnit)); double sigma12 = 1/sqrt(0.5*drUnit.dot(G12inv*drUnit));
RealOpenMM h12 = r - sigma12; double h12 = r - sigma12;
RealOpenMM rho = sigma/(h12+sigma); double rho = sigma/(h12+sigma);
RealOpenMM rho2 = rho*rho; double rho2 = rho*rho;
RealOpenMM rho6 = rho2*rho2*rho2; double rho6 = rho2*rho2*rho2;
RealOpenMM u = 4*epsilon*(rho6*rho6-rho6); double u = 4*epsilon*(rho6*rho6-rho6);
RealOpenMM eta = SQRT(2*s[particle1]*s[particle2]/detG12); double eta = sqrt(2*s[particle1]*s[particle2]/detG12);
RealOpenMM chi = 2*drUnit.dot(B12inv*drUnit); double chi = 2*drUnit.dot(B12inv*drUnit);
chi *= chi; chi *= chi;
RealOpenMM energy = u*eta*chi; double energy = u*eta*chi;
// Compute the terms needed for the force. // Compute the terms needed for the force.
RealVec kappa = G12inv*dr; Vec3 kappa = G12inv*dr;
RealVec iota = B12inv*dr; Vec3 iota = B12inv*dr;
RealOpenMM rInv2 = rInv*rInv; double rInv2 = rInv*rInv;
RealOpenMM dUSLJdr = 24*epsilon*(2*rho6-1)*rho6*rho/sigma; double dUSLJdr = 24*epsilon*(2*rho6-1)*rho6*rho/sigma;
RealOpenMM temp = 0.5*sigma12*sigma12*sigma12*rInv2; double temp = 0.5*sigma12*sigma12*sigma12*rInv2;
RealVec dudr = (drUnit + (kappa-drUnit*kappa.dot(drUnit))*temp)*dUSLJdr; Vec3 dudr = (drUnit + (kappa-drUnit*kappa.dot(drUnit))*temp)*dUSLJdr;
RealVec dchidr = (iota-drUnit*iota.dot(drUnit))*(-8*rInv2*SQRT(chi)); Vec3 dchidr = (iota-drUnit*iota.dot(drUnit))*(-8*rInv2*sqrt(chi));
RealVec force = (dchidr*u + dudr*chi)*(eta*switchValue) - drUnit*(energy*switchDeriv); Vec3 force = (dchidr*u + dudr*chi)*(eta*switchValue) - drUnit*(energy*switchDeriv);
forces[4*particle1] += force[0]; forces[4*particle1] += force[0];
forces[4*particle1+1] += force[1]; forces[4*particle1+1] += force[1];
forces[4*particle1+2] += force[2]; forces[4*particle1+2] += force[2];
...@@ -413,13 +401,13 @@ RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2, ...@@ -413,13 +401,13 @@ RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2,
ParticleInfo& p = particles[particle]; ParticleInfo& p = particles[particle];
if (p.isPointParticle) if (p.isPointParticle)
continue; continue;
RealVec dudq = (kappa*G[particle]).cross(kappa*(temp*dUSLJdr)); Vec3 dudq = (kappa*G[particle]).cross(kappa*(temp*dUSLJdr));
RealVec dchidq = (iota*B[particle]).cross(iota)*(-4*rInv2); Vec3 dchidq = (iota*B[particle]).cross(iota)*(-4*rInv2);
RealOpenMM (&g12)[3][3] = G12.v; double (&g12)[3][3] = G12.v;
RealOpenMM (&a)[3][3] = A[particle].v; double (&a)[3][3] = A[particle].v;
RealVec scale = RealVec(p.rx*p.rx, p.ry*p.ry, p.rz*p.rz)*(-0.5*eta/detG12); Vec3 scale = Vec3(p.rx*p.rx, p.ry*p.ry, p.rz*p.rz)*(-0.5*eta/detG12);
Matrix D; Matrix D;
RealOpenMM (&d)[3][3] = D.v; double (&d)[3][3] = D.v;
d[0][0] = scale[0]*(2*a[0][0]*(g12[1][1]*g12[2][2] - g12[1][2]*g12[2][1]) + d[0][0] = scale[0]*(2*a[0][0]*(g12[1][1]*g12[2][2] - g12[1][2]*g12[2][1]) +
a[0][2]*(g12[1][2]*g12[0][1] + g12[1][0]*g12[2][1] - g12[1][1]*(g12[0][2] + g12[2][0])) + a[0][2]*(g12[1][2]*g12[0][1] + g12[1][0]*g12[2][1] - g12[1][1]*(g12[0][2] + g12[2][0])) +
a[0][1]*(g12[0][2]*g12[2][1] + g12[2][0]*g12[1][2] - g12[2][2]*(g12[0][1] + g12[1][0]))); a[0][1]*(g12[0][2]*g12[2][1] + g12[2][0]*g12[1][2] - g12[2][2]*(g12[0][1] + g12[1][0])));
...@@ -447,10 +435,10 @@ RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2, ...@@ -447,10 +435,10 @@ RealOpenMM CpuGayBerneForce::computeOneInteraction(int particle1, int particle2,
d[2][2] = scale[2]*( a[2][0]*(g12[0][1]*g12[1][2] + g12[2][1]*g12[1][0] - g12[1][1]*(g12[0][2] + g12[2][0])) + d[2][2] = scale[2]*( a[2][0]*(g12[0][1]*g12[1][2] + g12[2][1]*g12[1][0] - g12[1][1]*(g12[0][2] + g12[2][0])) +
a[2][1]*(g12[1][0]*g12[0][2] + g12[2][0]*g12[0][1] - g12[0][0]*(g12[1][2] + g12[2][1])) + a[2][1]*(g12[1][0]*g12[0][2] + g12[2][0]*g12[0][1] - g12[0][0]*(g12[1][2] + g12[2][1])) +
2*a[2][2]*(g12[1][1]*g12[0][0] - g12[1][0]*g12[0][1])); 2*a[2][2]*(g12[1][1]*g12[0][0] - g12[1][0]*g12[0][1]));
RealVec detadq; Vec3 detadq;
for (int i = 0; i < 3; i++) for (int i = 0; i < 3; i++)
detadq += RealVec(a[i][0], a[i][1], a[i][2]).cross(RealVec(d[i][0], d[i][1], d[i][2])); detadq += Vec3(a[i][0], a[i][1], a[i][2]).cross(Vec3(d[i][0], d[i][1], d[i][2]));
RealVec torque = (dchidq*(u*eta) + detadq*(u*chi) + dudq*(eta*chi))*switchValue; Vec3 torque = (dchidq*(u*eta) + detadq*(u*chi) + dudq*(eta*chi))*switchValue;
torques[particle] -= torque; torques[particle] -= torque;
} }
return switchValue*energy; return switchValue*energy;
......
platforms/cpu/src/CpuKernels.cpp
View file @ 047934e2
This diff is collapsed.
platforms/cpu/src/CpuLangevinDynamics.cpp
View file @ 047934e2
/* Portions copyright (c) 2006-2015 Stanford University and Simbios. /* Portions copyright (c) 2006-2017 Stanford University and Simbios.
* Authors: Peter Eastman * Authors: Peter Eastman
* Contributors: * Contributors:
* *
...@@ -29,45 +29,15 @@ ...@@ -29,45 +29,15 @@
using namespace OpenMM; using namespace OpenMM;
using namespace std; using namespace std;
class CpuLangevinDynamics::Update1Task : public ThreadPool::Task { CpuLangevinDynamics::CpuLangevinDynamics(int numberOfAtoms, double deltaT, double friction, double temperature, ThreadPool& threads, CpuRandom& random) :
public: ReferenceStochasticDynamics(numberOfAtoms, deltaT, friction, temperature), threads(threads), random(random) {
Update1Task(CpuLangevinDynamics& owner) : owner(owner) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadUpdate1(threadIndex);
}
CpuLangevinDynamics& owner;
};
class CpuLangevinDynamics::Update2Task : public ThreadPool::Task {
public:
Update2Task(CpuLangevinDynamics& owner) : owner(owner) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadUpdate2(threadIndex);
}
CpuLangevinDynamics& owner;
};
class CpuLangevinDynamics::Update3Task : public ThreadPool::Task {
public:
Update3Task(CpuLangevinDynamics& owner) : owner(owner) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadUpdate3(threadIndex);
}
CpuLangevinDynamics& owner;
};
CpuLangevinDynamics::CpuLangevinDynamics(int numberOfAtoms, RealOpenMM deltaT, RealOpenMM tau, RealOpenMM temperature, ThreadPool& threads, CpuRandom& random) :
ReferenceStochasticDynamics(numberOfAtoms, deltaT, tau, temperature), threads(threads), random(random) {
} }
CpuLangevinDynamics::~CpuLangevinDynamics() { CpuLangevinDynamics::~CpuLangevinDynamics() {
} }
void CpuLangevinDynamics::updatePart1(int numberOfAtoms, vector<RealVec>& atomCoordinates, vector<RealVec>& velocities, void CpuLangevinDynamics::updatePart1(int numberOfAtoms, vector<Vec3>& atomCoordinates, vector<Vec3>& velocities,
vector<RealVec>& forces, vector<RealOpenMM>& inverseMasses, vector<RealVec>& xPrime) { vector<Vec3>& forces, vector<double>& inverseMasses, vector<Vec3>& xPrime) {
// Record the parameters for the threads. // Record the parameters for the threads.
this->numberOfAtoms = numberOfAtoms; this->numberOfAtoms = numberOfAtoms;
...@@ -79,13 +49,12 @@ void CpuLangevinDynamics::updatePart1(int numberOfAtoms, vector<RealVec>& atomCo ...@@ -79,13 +49,12 @@ void CpuLangevinDynamics::updatePart1(int numberOfAtoms, vector<RealVec>& atomCo
// Signal the threads to start running and wait for them to finish. // Signal the threads to start running and wait for them to finish.
Update1Task task(*this); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadUpdate1(threadIndex); });
threads.execute(task);
threads.waitForThreads(); threads.waitForThreads();
} }
void CpuLangevinDynamics::updatePart2(int numberOfAtoms, vector<RealVec>& atomCoordinates, vector<RealVec>& velocities, void CpuLangevinDynamics::updatePart2(int numberOfAtoms, vector<Vec3>& atomCoordinates, vector<Vec3>& velocities,
vector<RealVec>& forces, vector<RealOpenMM>& inverseMasses, vector<RealVec>& xPrime) { vector<Vec3>& forces, vector<double>& inverseMasses, vector<Vec3>& xPrime) {
// Record the parameters for the threads. // Record the parameters for the threads.
this->numberOfAtoms = numberOfAtoms; this->numberOfAtoms = numberOfAtoms;
...@@ -97,13 +66,12 @@ void CpuLangevinDynamics::updatePart2(int numberOfAtoms, vector<RealVec>& atomCo ...@@ -97,13 +66,12 @@ void CpuLangevinDynamics::updatePart2(int numberOfAtoms, vector<RealVec>& atomCo
// Signal the threads to start running and wait for them to finish. // Signal the threads to start running and wait for them to finish.
Update2Task task(*this); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadUpdate2(threadIndex); });
threads.execute(task);
threads.waitForThreads(); threads.waitForThreads();
} }
void CpuLangevinDynamics::updatePart3(int numberOfAtoms, vector<RealVec>& atomCoordinates, vector<RealVec>& velocities, void CpuLangevinDynamics::updatePart3(int numberOfAtoms, vector<Vec3>& atomCoordinates, vector<Vec3>& velocities,
vector<RealOpenMM>& inverseMasses, vector<RealVec>& xPrime) { vector<double>& inverseMasses, vector<Vec3>& xPrime) {
// Record the parameters for the threads. // Record the parameters for the threads.
this->numberOfAtoms = numberOfAtoms; this->numberOfAtoms = numberOfAtoms;
...@@ -114,44 +82,44 @@ void CpuLangevinDynamics::updatePart3(int numberOfAtoms, vector<RealVec>& atomCo ...@@ -114,44 +82,44 @@ void CpuLangevinDynamics::updatePart3(int numberOfAtoms, vector<RealVec>& atomCo
// Signal the threads to start running and wait for them to finish. // Signal the threads to start running and wait for them to finish.
Update3Task task(*this); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadUpdate3(threadIndex); });
threads.execute(task);
threads.waitForThreads(); threads.waitForThreads();
} }
void CpuLangevinDynamics::threadUpdate1(int threadIndex) { void CpuLangevinDynamics::threadUpdate1(int threadIndex) {
const RealOpenMM tau = getTau(); double dt = getDeltaT();
const RealOpenMM vscale = EXP(-getDeltaT()/tau); double friction = getFriction();
const RealOpenMM fscale = (1-vscale)*tau; const double vscale = exp(-dt*friction);
const RealOpenMM kT = BOLTZ*getTemperature(); const double fscale = (friction == 0 ? dt : (1-vscale)/friction);
const RealOpenMM noisescale = SQRT(2*kT/tau)*SQRT(0.5*(1-vscale*vscale)*tau); const double kT = BOLTZ*getTemperature();
const double noisescale = sqrt(kT*(1-vscale*vscale));
int start = threadIndex*numberOfAtoms/threads.getNumThreads(); int start = threadIndex*numberOfAtoms/threads.getNumThreads();
int end = (threadIndex+1)*numberOfAtoms/threads.getNumThreads(); int end = (threadIndex+1)*numberOfAtoms/threads.getNumThreads();
for (int i = start; i < end; i++) { for (int i = start; i < end; i++) {
if (inverseMasses[i] != 0.0) { if (inverseMasses[i] != 0.0) {
RealOpenMM sqrtInvMass = SQRT(inverseMasses[i]); double sqrtInvMass = sqrt(inverseMasses[i]);
RealVec noise(random.getGaussianRandom(threadIndex), random.getGaussianRandom(threadIndex), random.getGaussianRandom(threadIndex)); Vec3 noise(random.getGaussianRandom(threadIndex), random.getGaussianRandom(threadIndex), random.getGaussianRandom(threadIndex));
velocities[i] = velocities[i]*vscale + forces[i]*(fscale*inverseMasses[i]) + noise*(noisescale*sqrtInvMass); velocities[i] = velocities[i]*vscale + forces[i]*(fscale*inverseMasses[i]) + noise*(noisescale*sqrtInvMass);
} }
} }
} }
void CpuLangevinDynamics::threadUpdate2(int threadIndex) { void CpuLangevinDynamics::threadUpdate2(int threadIndex) {
const RealOpenMM dt = getDeltaT(); const double dt = getDeltaT();
int start = threadIndex*numberOfAtoms/threads.getNumThreads(); int start = threadIndex*numberOfAtoms/threads.getNumThreads();
int end = (threadIndex+1)*numberOfAtoms/threads.getNumThreads(); int end = (threadIndex+1)*numberOfAtoms/threads.getNumThreads();
for (int i = start; i < end; i++) { for (int i = start; i < end; i++) {
if (inverseMasses[i] != 0.0) { if (inverseMasses[i] != 0.0) {
RealOpenMM sqrtInvMass = SQRT(inverseMasses[i]); double sqrtInvMass = sqrt(inverseMasses[i]);
xPrime[i] = atomCoordinates[i]+velocities[i]*dt; xPrime[i] = atomCoordinates[i]+velocities[i]*dt;
} }
} }
} }
void CpuLangevinDynamics::threadUpdate3(int threadIndex) { void CpuLangevinDynamics::threadUpdate3(int threadIndex) {
const RealOpenMM invStepSize = 1.0/getDeltaT(); const double invStepSize = 1.0/getDeltaT();
int start = threadIndex*numberOfAtoms/threads.getNumThreads(); int start = threadIndex*numberOfAtoms/threads.getNumThreads();
int end = (threadIndex+1)*numberOfAtoms/threads.getNumThreads(); int end = (threadIndex+1)*numberOfAtoms/threads.getNumThreads();
......
platforms/cpu/src/CpuNeighborList.cpp
View file @ 047934e2
...@@ -6,7 +6,7 @@ ...@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for * * Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. * * Medical Research, grant U54 GM072970. See https://simtk.org. *
* * * *
* Portions copyright (c) 2013-2016 Stanford University and the Authors. * * Portions copyright (c) 2013-2017 Stanford University and the Authors. *
* Authors: Peter Eastman * * Authors: Peter Eastman *
* Contributors: * * Contributors: *
* * * *
...@@ -59,7 +59,7 @@ public: ...@@ -59,7 +59,7 @@ public:
*/ */
class CpuNeighborList::Voxels { class CpuNeighborList::Voxels {
public: public:
Voxels(int blockSize, float vsy, float vsz, float miny, float maxy, float minz, float maxz, const RealVec* boxVectors, bool usePeriodic) : Voxels(int blockSize, float vsy, float vsz, float miny, float maxy, float minz, float maxz, const Vec3* boxVectors, bool usePeriodic) :
blockSize(blockSize), voxelSizeY(vsy), voxelSizeZ(vsz), miny(miny), maxy(maxy), minz(minz), maxz(maxz), usePeriodic(usePeriodic) { blockSize(blockSize), voxelSizeY(vsy), voxelSizeZ(vsz), miny(miny), maxy(maxy), minz(minz), maxz(maxz), usePeriodic(usePeriodic) {
for (int i = 0; i < 3; i++) for (int i = 0; i < 3; i++)
for (int j = 0; j < 3; j++) for (int j = 0; j < 3; j++)
...@@ -409,21 +409,11 @@ private: ...@@ -409,21 +409,11 @@ private:
vector<vector<vector<pair<float, int> > > > bins; vector<vector<vector<pair<float, int> > > > bins;
}; };
class CpuNeighborList::ThreadTask : public ThreadPool::Task {
public:
ThreadTask(CpuNeighborList& owner) : owner(owner) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadComputeNeighborList(threads, threadIndex);
}
CpuNeighborList& owner;
};
CpuNeighborList::CpuNeighborList(int blockSize) : blockSize(blockSize) { CpuNeighborList::CpuNeighborList(int blockSize) : blockSize(blockSize) {
} }
void CpuNeighborList::computeNeighborList(int numAtoms, const AlignedArray<float>& atomLocations, const vector<set<int> >& exclusions, void CpuNeighborList::computeNeighborList(int numAtoms, const AlignedArray<float>& atomLocations, const vector<set<int> >& exclusions,
const RealVec* periodicBoxVectors, bool usePeriodic, float maxDistance, ThreadPool& threads) { const Vec3* periodicBoxVectors, bool usePeriodic, float maxDistance, ThreadPool& threads) {
int numBlocks = (numAtoms+blockSize-1)/blockSize; int numBlocks = (numAtoms+blockSize-1)/blockSize;
blockNeighbors.resize(numBlocks); blockNeighbors.resize(numBlocks);
blockExclusions.resize(numBlocks); blockExclusions.resize(numBlocks);
...@@ -460,8 +450,7 @@ void CpuNeighborList::computeNeighborList(int numAtoms, const AlignedArray<float ...@@ -460,8 +450,7 @@ void CpuNeighborList::computeNeighborList(int numAtoms, const AlignedArray<float
// Sort the atoms based on a Hilbert curve. // Sort the atoms based on a Hilbert curve.
atomBins.resize(numAtoms); atomBins.resize(numAtoms);
ThreadTask task(*this); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadComputeNeighborList(threads, threadIndex); });
threads.execute(task);
threads.waitForThreads(); threads.waitForThreads();
sort(atomBins.begin(), atomBins.end()); sort(atomBins.begin(), atomBins.end());
......
platforms/cpu/src/CpuNonbondedForce.cpp
View file @ 047934e2
/* Portions copyright (c) 2006-2015 Stanford University and Simbios. /* Portions copyright (c) 2006-2017 Stanford University and Simbios.
* Contributors: Pande Group * Contributors: Pande Group
* *
* Permission is hereby granted, free of charge, to any person obtaining * Permission is hereby granted, free of charge, to any person obtaining
...@@ -30,6 +30,7 @@ ...@@ -30,6 +30,7 @@
#include "ReferencePME.h" #include "ReferencePME.h"
#include "openmm/internal/gmx_atomic.h" #include "openmm/internal/gmx_atomic.h"
#include <algorithm> #include <algorithm>
#include <iostream>
// In case we're using some primitive version of Visual Studio this will // In case we're using some primitive version of Visual Studio this will
// make sure that erf() and erfc() are defined. // make sure that erf() and erfc() are defined.
...@@ -41,23 +42,14 @@ using namespace OpenMM; ...@@ -41,23 +42,14 @@ using namespace OpenMM;
const float CpuNonbondedForce::TWO_OVER_SQRT_PI = (float) (2/sqrt(PI_M)); const float CpuNonbondedForce::TWO_OVER_SQRT_PI = (float) (2/sqrt(PI_M));
const int CpuNonbondedForce::NUM_TABLE_POINTS = 2048; const int CpuNonbondedForce::NUM_TABLE_POINTS = 2048;
class CpuNonbondedForce::ComputeDirectTask : public ThreadPool::Task {
public:
ComputeDirectTask(CpuNonbondedForce& owner) : owner(owner) {
}
void execute(ThreadPool& threads, int threadIndex) {
owner.threadComputeDirect(threads, threadIndex);
}
CpuNonbondedForce& owner;
};
/**--------------------------------------------------------------------------------------- /**---------------------------------------------------------------------------------------
CpuNonbondedForce constructor CpuNonbondedForce constructor
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
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() { CpuNonbondedForce::~CpuNonbondedForce() {
...@@ -78,11 +70,22 @@ void CpuNonbondedForce::setUseCutoff(float distance, const CpuNeighborList& neig ...@@ -78,11 +70,22 @@ void CpuNonbondedForce::setUseCutoff(float distance, const CpuNeighborList& neig
tableIsValid = false; tableIsValid = false;
cutoff = true; cutoff = true;
cutoffDistance = distance; cutoffDistance = distance;
inverseRcut6 = pow(cutoffDistance, -6);
neighborList = &neighbors; neighborList = &neighbors;
krf = pow(cutoffDistance, -3.0f)*(solventDielectric-1.0)/(2.0*solventDielectric+1.0); 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); 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));
} }
}
/**--------------------------------------------------------------------------------------- /**---------------------------------------------------------------------------------------
Set the force to use a switching function on the Lennard-Jones interaction. Set the force to use a switching function on the Lennard-Jones interaction.
...@@ -96,7 +99,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) { ...@@ -96,7 +99,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
switchingDistance = distance; switchingDistance = distance;
} }
/**--------------------------------------------------------------------------------------- /**---------------------------------------------------------------------------------------
Set the force to use periodic boundary conditions. This requires that a cutoff has 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 also been set, and the smallest side of the periodic box is at least twice the cutoff
...@@ -106,7 +109,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) { ...@@ -106,7 +109,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
void CpuNonbondedForce::setPeriodic(RealVec* periodicBoxVectors) { void CpuNonbondedForce::setPeriodic(Vec3* periodicBoxVectors) {
assert(cutoff); assert(cutoff);
assert(periodicBoxVectors[0][0] >= 2.0*cutoffDistance); assert(periodicBoxVectors[0][0] >= 2.0*cutoffDistance);
...@@ -126,9 +129,9 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) { ...@@ -126,9 +129,9 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
triclinic = (periodicBoxVectors[0][1] != 0.0 || periodicBoxVectors[0][2] != 0.0 || triclinic = (periodicBoxVectors[0][1] != 0.0 || periodicBoxVectors[0][2] != 0.0 ||
periodicBoxVectors[1][0] != 0.0 || periodicBoxVectors[1][2] != 0.0 || periodicBoxVectors[1][0] != 0.0 || periodicBoxVectors[1][2] != 0.0 ||
periodicBoxVectors[2][0] != 0.0 || periodicBoxVectors[2][1] != 0.0); periodicBoxVectors[2][0] != 0.0 || periodicBoxVectors[2][1] != 0.0);
} }
/**--------------------------------------------------------------------------------------- /**---------------------------------------------------------------------------------------
Set the force to use Ewald summation. Set the force to use Ewald summation.
...@@ -139,7 +142,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) { ...@@ -139,7 +142,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
void CpuNonbondedForce::setUseEwald(float alpha, int kmaxx, int kmaxy, int kmaxz) { void CpuNonbondedForce::setUseEwald(float alpha, int kmaxx, int kmaxy, int kmaxz) {
if (alpha != alphaEwald) if (alpha != alphaEwald)
tableIsValid = false; tableIsValid = false;
alphaEwald = alpha; alphaEwald = alpha;
...@@ -148,9 +151,9 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) { ...@@ -148,9 +151,9 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
numRz = kmaxz; numRz = kmaxz;
ewald = true; ewald = true;
tabulateEwaldScaleFactor(); tabulateEwaldScaleFactor();
} }
/**--------------------------------------------------------------------------------------- /**---------------------------------------------------------------------------------------
Set the force to use Particle-Mesh Ewald (PME) summation. Set the force to use Particle-Mesh Ewald (PME) summation.
...@@ -159,7 +162,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) { ...@@ -159,7 +162,7 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
--------------------------------------------------------------------------------------- */ --------------------------------------------------------------------------------------- */
void CpuNonbondedForce::setUsePME(float alpha, int meshSize[3]) { void CpuNonbondedForce::setUsePME(float alpha, int meshSize[3]) {
if (alpha != alphaEwald) if (alpha != alphaEwald)
tableIsValid = false; tableIsValid = false;
alphaEwald = alpha; alphaEwald = alpha;
...@@ -168,10 +171,40 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) { ...@@ -168,10 +171,40 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
meshDim[2] = meshSize[2]; meshDim[2] = meshSize[2];
pme = true; pme = true;
tabulateEwaldScaleFactor(); 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() { void CpuNonbondedForce::tabulateEwaldScaleFactor() {
if (tableIsValid) if (tableIsValid)
return; return;
tableIsValid = true; tableIsValid = true;
...@@ -188,9 +221,29 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) { ...@@ -188,9 +221,29 @@ void CpuNonbondedForce::setUseSwitchingFunction(float distance) {
} }
} }
void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, const vector<RealVec>& atomCoordinates, void CpuNonbondedForce::tabulateExpTerms() {
const vector<pair<float, float> >& atomParameters, const vector<set<int> >& exclusions, if (expTableIsValid)
vector<RealVec>& forces, double* totalEnergy) const { 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<Vec3>& atomCoordinates,
const vector<pair<float, float> >& atomParameters, const vector<float> &C6params, const vector<set<int> >& exclusions,
vector<Vec3>& forces, double* totalEnergy) const {
typedef std::complex<float> d_complex; typedef std::complex<float> d_complex;
static const float epsilon = 1.0; static const float epsilon = 1.0;
...@@ -203,14 +256,37 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c ...@@ -203,14 +256,37 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
if (pme) { if (pme) {
pme_t pmedata; pme_t pmedata;
pme_init(&pmedata, alphaEwald, numberOfAtoms, meshDim, 5, 1); pme_init(&pmedata, alphaEwald, numberOfAtoms, meshDim, 5, 1);
vector<RealOpenMM> charges(numberOfAtoms); vector<double> charges(numberOfAtoms);
for (int i = 0; i < numberOfAtoms; i++) for (int i = 0; i < numberOfAtoms; i++)
charges[i] = posq[4*i+3]; charges[i] = posq[4*i+3];
RealOpenMM recipEnergy = 0.0; double recipEnergy = 0.0;
pme_exec(pmedata, atomCoordinates, forces, charges, periodicBoxVectors, &recipEnergy); pme_exec(pmedata, atomCoordinates, forces, charges, periodicBoxVectors, &recipEnergy);
if (totalEnergy) if (totalEnergy)
*totalEnergy += recipEnergy; *totalEnergy += recipEnergy;
pme_destroy(pmedata); pme_destroy(pmedata);
if (ljpme) {
// Dispersion reciprocal space terms
pme_init(&pmedata,alphaDispersionEwald,numberOfAtoms,dispersionMeshDim,5,1);
std::vector<Vec3> dpmeforces;
for (int i = 0; i < numberOfAtoms; i++){
charges[i] = C6params[i];
dpmeforces.push_back(Vec3());
}
double 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 // Ewald method
...@@ -224,7 +300,7 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c ...@@ -224,7 +300,7 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
// setup K-vectors // 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> eir(kmax*numberOfAtoms*3);
vector<d_complex> tab_xy(numberOfAtoms); vector<d_complex> tab_xy(numberOfAtoms);
vector<d_complex> tab_qxyz(numberOfAtoms); vector<d_complex> tab_qxyz(numberOfAtoms);
...@@ -300,14 +376,15 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c ...@@ -300,14 +376,15 @@ 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, void CpuNonbondedForce::calculateDirectIxn(int numberOfAtoms, float* posq, const vector<Vec3>& 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. // Record the parameters for the threads.
this->numberOfAtoms = numberOfAtoms; this->numberOfAtoms = numberOfAtoms;
this->posq = posq; this->posq = posq;
this->atomCoordinates = &atomCoordinates[0]; this->atomCoordinates = &atomCoordinates[0];
this->atomParameters = &atomParameters[0]; this->atomParameters = &atomParameters[0];
this->C6params = &C6params[0];
this->exclusions = &exclusions[0]; this->exclusions = &exclusions[0];
this->threadForce = &threadForce; this->threadForce = &threadForce;
includeEnergy = (totalEnergy != NULL); includeEnergy = (totalEnergy != NULL);
...@@ -318,8 +395,7 @@ void CpuNonbondedForce::calculateDirectIxn(int numberOfAtoms, float* posq, const ...@@ -318,8 +395,7 @@ void CpuNonbondedForce::calculateDirectIxn(int numberOfAtoms, float* posq, const
// Signal the threads to start running and wait for them to finish. // Signal the threads to start running and wait for them to finish.
ComputeDirectTask task(*this); threads.execute([&] (ThreadPool& threads, int threadIndex) { threadComputeDirect(threads, threadIndex); });
threads.execute(task);
threads.waitForThreads(); threads.waitForThreads();
// Signal the threads to subtract the exclusions. // Signal the threads to subtract the exclusions.
...@@ -350,9 +426,8 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex ...@@ -350,9 +426,8 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex
float* forces = &(*threadForce)[threadIndex][0]; float* forces = &(*threadForce)[threadIndex][0];
fvec4 boxSize(periodicBoxVectors[0][0], periodicBoxVectors[1][1], periodicBoxVectors[2][2], 0); fvec4 boxSize(periodicBoxVectors[0][0], periodicBoxVectors[1][1], periodicBoxVectors[2][2], 0);
fvec4 invBoxSize(recipBoxSize[0], recipBoxSize[1], recipBoxSize[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. // Compute the interactions from the neighbor list.
while (true) { while (true) {
int nextBlock = gmx_atomic_fetch_add(reinterpret_cast<gmx_atomic_t*>(atomicCounter), 1); int nextBlock = gmx_atomic_fetch_add(reinterpret_cast<gmx_atomic_t*>(atomicCounter), 1);
if (nextBlock >= neighborList->getNumBlocks()) if (nextBlock >= neighborList->getNumBlocks())
...@@ -395,6 +470,17 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex ...@@ -395,6 +470,17 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex
} }
else if (includeEnergy) 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);
}
} }
} }
} }
...@@ -476,7 +562,7 @@ void CpuNonbondedForce::calculateOneIxn(int ii, int jj, float* forces, double* t ...@@ -476,7 +562,7 @@ void CpuNonbondedForce::calculateOneIxn(int ii, int jj, float* forces, double* t
fvec4 result = deltaR*dEdR; fvec4 result = deltaR*dEdR;
(fvec4(forces+4*ii)+result).store(forces+4*ii); (fvec4(forces+4*ii)+result).store(forces+4*ii);
(fvec4(forces+4*jj)-result).store(forces+4*jj); (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 { 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; deltaR = posJ-posI;
...@@ -502,3 +588,18 @@ float CpuNonbondedForce::erfcApprox(float x) { ...@@ -502,3 +588,18 @@ float CpuNonbondedForce::erfcApprox(float x) {
return coeff1*erfcTable[index] + coeff2*erfcTable[index+1]; 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 @ 047934e2
...@@ -25,6 +25,7 @@ ...@@ -25,6 +25,7 @@
#include "SimTKOpenMMUtilities.h" #include "SimTKOpenMMUtilities.h"
#include "CpuNonbondedForceVec4.h" #include "CpuNonbondedForceVec4.h"
#include <algorithm> #include <algorithm>
#include <iostream>
using namespace std; using namespace std;
using namespace OpenMM; using namespace OpenMM;
...@@ -213,7 +214,6 @@ void CpuNonbondedForceVec4::calculateBlockIxnImpl(int blockIndex, float* forces, ...@@ -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) { void CpuNonbondedForceVec4::calculateBlockEwaldIxn(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; PeriodicType periodicType;
fvec4 blockCenter; fvec4 blockCenter;
if (!periodic) { if (!periodic) {
...@@ -263,7 +263,6 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxn(int blockIndex, float* forces ...@@ -263,7 +263,6 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxn(int blockIndex, float* forces
template <int PERIODIC_TYPE> template <int PERIODIC_TYPE>
void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize, const fvec4& blockCenter) { 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. // Load the positions and parameters of the atoms in the block.
const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex]; const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex];
fvec4 blockAtomPosq[4]; fvec4 blockAtomPosq[4];
fvec4 blockAtomForceX(0.0f), blockAtomForceY(0.0f), blockAtomForceZ(0.0f); fvec4 blockAtomForceX(0.0f), blockAtomForceY(0.0f), blockAtomForceZ(0.0f);
...@@ -278,6 +277,7 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo ...@@ -278,6 +277,7 @@ 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 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 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 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 bool needPeriodic = (PERIODIC_TYPE == PeriodicPerInteraction || PERIODIC_TYPE == PeriodicTriclinic);
const float invSwitchingInterval = 1/(cutoffDistance-switchingDistance); const float invSwitchingInterval = 1/(cutoffDistance-switchingDistance);
...@@ -318,7 +318,8 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo ...@@ -318,7 +318,8 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
fvec4 sig2 = inverseR*sig; fvec4 sig2 = inverseR*sig;
sig2 *= sig2; sig2 *= sig2;
fvec4 sig6 = sig2*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); dEdR = epsSig6*(12.0f*sig6 - 6.0f);
energy = epsSig6*(sig6-1.0f); energy = epsSig6*(sig6-1.0f);
if (useSwitch) { if (useSwitch) {
...@@ -328,6 +329,17 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo ...@@ -328,6 +329,17 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
dEdR = switchValue*dEdR - energy*switchDeriv*r; dEdR = switchValue*dEdR - energy*switchDeriv*r;
energy *= switchValue; 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 { else {
energy = 0.0f; energy = 0.0f;
...@@ -420,3 +432,30 @@ fvec4 CpuNonbondedForceVec4::ewaldScaleFunction(const fvec4& x) { ...@@ -420,3 +432,30 @@ fvec4 CpuNonbondedForceVec4::ewaldScaleFunction(const fvec4& x) {
transpose(t1, t2, t3, t4); transpose(t1, t2, t3, t4);
return coeff1*t1 + coeff2*t2; 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 @ 047934e2
...@@ -27,6 +27,7 @@ ...@@ -27,6 +27,7 @@
#include "openmm/OpenMMException.h" #include "openmm/OpenMMException.h"
#include "openmm/internal/hardware.h" #include "openmm/internal/hardware.h"
#include <algorithm> #include <algorithm>
#include <iostream>
using namespace std; using namespace std;
using namespace OpenMM; using namespace OpenMM;
...@@ -81,7 +82,6 @@ enum PeriodicType {NoPeriodic, PeriodicPerAtom, PeriodicPerInteraction, Periodic ...@@ -81,7 +82,6 @@ enum PeriodicType {NoPeriodic, PeriodicPerAtom, PeriodicPerInteraction, Periodic
void CpuNonbondedForceVec8::calculateBlockIxn(int blockIndex, float* forces, double* totalEnergy, const fvec4& boxSize, const fvec4& invBoxSize) { 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; PeriodicType periodicType;
fvec4 blockCenter; fvec4 blockCenter;
if (!periodic) { if (!periodic) {
...@@ -308,6 +308,7 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo ...@@ -308,6 +308,7 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
blockAtomCharge *= ONE_4PI_EPS0; 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 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 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 bool needPeriodic = (PERIODIC_TYPE == PeriodicPerInteraction || PERIODIC_TYPE == PeriodicTriclinic);
const float invSwitchingInterval = 1/(cutoffDistance-switchingDistance); const float invSwitchingInterval = 1/(cutoffDistance-switchingDistance);
...@@ -348,7 +349,8 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo ...@@ -348,7 +349,8 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
fvec8 sig2 = inverseR*sig; fvec8 sig2 = inverseR*sig;
sig2 *= sig2; sig2 *= sig2;
fvec8 sig6 = sig2*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); dEdR = epsSig6*(12.0f*sig6 - 6.0f);
energy = epsSig6*(sig6-1.0f); energy = epsSig6*(sig6-1.0f);
if (useSwitch) { if (useSwitch) {
...@@ -358,6 +360,17 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo ...@@ -358,6 +360,17 @@ void CpuNonbondedForceVec8::calculateBlockEwaldIxnImpl(int blockIndex, float* fo
dEdR = switchValue*dEdR - energy*switchDeriv*r; dEdR = switchValue*dEdR - energy*switchDeriv*r;
energy *= switchValue; 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 { else {
energy = 0.0f; energy = 0.0f;
...@@ -464,4 +477,45 @@ fvec8 CpuNonbondedForceVec8::ewaldScaleFunction(const fvec8& x) { ...@@ -464,4 +477,45 @@ fvec8 CpuNonbondedForceVec8::ewaldScaleFunction(const fvec8& x) {
transpose(t1, t2, t3, t4, t5, t6, t7, t8, s1, s2, s3, s4); transpose(t1, t2, t3, t4, t5, t6, t7, t8, s1, s2, s3, s4);
return coeff1*s1 + coeff2*s2; 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 #endif
platforms/cpu/src/CpuPlatform.cpp
View file @ 047934e2
...@@ -127,6 +127,8 @@ void CpuPlatform::contextDestroyed(ContextImpl& context) const { ...@@ -127,6 +127,8 @@ void CpuPlatform::contextDestroyed(ContextImpl& context) const {
PlatformData* data = contextData[&context]; PlatformData* data = contextData[&context];
delete data; delete data;
contextData.erase(&context); contextData.erase(&context);
ReferencePlatform::PlatformData* refPlatformData = reinterpret_cast<ReferencePlatform::PlatformData*>(context.getPlatformData());
delete refPlatformData;
} }
CpuPlatform::PlatformData& CpuPlatform::getPlatformData(ContextImpl& context) { CpuPlatform::PlatformData& CpuPlatform::getPlatformData(ContextImpl& context) {
......
platforms/cpu/src/CpuSETTLE.cpp
View file @ 047934e2
...@@ -6,7 +6,7 @@ ...@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for * * Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. * * Medical Research, grant U54 GM072970. See https://simtk.org. *
* * * *
* Portions copyright (c) 2013-2015 Stanford University and the Authors. * * Portions copyright (c) 2013-2017 Stanford University and the Authors. *
* Authors: Peter Eastman * * Authors: Peter Eastman *
* Contributors: * * Contributors: *
* * * *
...@@ -35,56 +35,10 @@ ...@@ -35,56 +35,10 @@
using namespace OpenMM; using namespace OpenMM;
using namespace std; using namespace std;
class CpuSETTLE::ApplyToPositionsTask : public ThreadPool::Task {
public:
ApplyToPositionsTask(vector<OpenMM::RealVec>& atomCoordinates, vector<OpenMM::RealVec>& atomCoordinatesP, vector<RealOpenMM>& inverseMasses,
RealOpenMM tolerance, vector<ReferenceSETTLEAlgorithm*>& threadSettle) : atomCoordinates(atomCoordinates), atomCoordinatesP(atomCoordinatesP),
inverseMasses(inverseMasses), tolerance(tolerance), threadSettle(threadSettle) {
gmx_atomic_set(&atomicCounter, 0);
}
void execute(ThreadPool& threads, int threadIndex) {
while (true) {
int index = gmx_atomic_fetch_add(&atomicCounter, 1);
if (index >= threadSettle.size())
break;
threadSettle[index]->apply(atomCoordinates, atomCoordinatesP, inverseMasses, tolerance);
}
}
vector<OpenMM::RealVec>& atomCoordinates;
vector<OpenMM::RealVec>& atomCoordinatesP;
vector<RealOpenMM>& inverseMasses;
RealOpenMM tolerance;
vector<ReferenceSETTLEAlgorithm*>& threadSettle;
gmx_atomic_t atomicCounter;
};
class CpuSETTLE::ApplyToVelocitiesTask : public ThreadPool::Task {
public:
ApplyToVelocitiesTask(vector<OpenMM::RealVec>& atomCoordinates, vector<OpenMM::RealVec>& velocities, vector<RealOpenMM>& inverseMasses,
RealOpenMM tolerance, vector<ReferenceSETTLEAlgorithm*>& threadSettle) : atomCoordinates(atomCoordinates), velocities(velocities),
inverseMasses(inverseMasses), tolerance(tolerance), threadSettle(threadSettle) {
gmx_atomic_set(&atomicCounter, 0);
}
void execute(ThreadPool& threads, int threadIndex) {
while (true) {
int index = gmx_atomic_fetch_add(&atomicCounter, 1);
if (index >= threadSettle.size())
break;
threadSettle[index]->applyToVelocities(atomCoordinates, velocities, inverseMasses, tolerance);
}
}
vector<OpenMM::RealVec>& atomCoordinates;
vector<OpenMM::RealVec>& velocities;
vector<RealOpenMM>& inverseMasses;
RealOpenMM tolerance;
vector<ReferenceSETTLEAlgorithm*>& threadSettle;
gmx_atomic_t atomicCounter;
};
CpuSETTLE::CpuSETTLE(const System& system, const ReferenceSETTLEAlgorithm& settle, ThreadPool& threads) : threads(threads) { CpuSETTLE::CpuSETTLE(const System& system, const ReferenceSETTLEAlgorithm& settle, ThreadPool& threads) : threads(threads) {
int numBlocks = 10*threads.getNumThreads(); int numBlocks = 10*threads.getNumThreads();
int numClusters = settle.getNumClusters(); int numClusters = settle.getNumClusters();
vector<RealOpenMM> mass(system.getNumParticles()); vector<double> mass(system.getNumParticles());
for (int i = 0; i < system.getNumParticles(); i++) for (int i = 0; i < system.getNumParticles(); i++)
mass[i] = system.getParticleMass(i); mass[i] = system.getParticleMass(i);
for (int i = 0; i < numBlocks; i++) { for (int i = 0; i < numBlocks; i++) {
...@@ -93,7 +47,7 @@ CpuSETTLE::CpuSETTLE(const System& system, const ReferenceSETTLEAlgorithm& settl ...@@ -93,7 +47,7 @@ CpuSETTLE::CpuSETTLE(const System& system, const ReferenceSETTLEAlgorithm& settl
if (start != end) { if (start != end) {
int numThreadClusters = end-start; int numThreadClusters = end-start;
vector<int> atom1(numThreadClusters), atom2(numThreadClusters), atom3(numThreadClusters); vector<int> atom1(numThreadClusters), atom2(numThreadClusters), atom3(numThreadClusters);
vector<RealOpenMM> distance1(numThreadClusters), distance2(numThreadClusters); vector<double> distance1(numThreadClusters), distance2(numThreadClusters);
for (int j = 0; j < numThreadClusters; j++) for (int j = 0; j < numThreadClusters; j++)
settle.getClusterParameters(start+j, atom1[j], atom2[j], atom3[j], distance1[j], distance2[j]); settle.getClusterParameters(start+j, atom1[j], atom2[j], atom3[j], distance1[j], distance2[j]);
threadSettle.push_back(new ReferenceSETTLEAlgorithm(atom1, atom2, atom3, distance1, distance2, mass)); threadSettle.push_back(new ReferenceSETTLEAlgorithm(atom1, atom2, atom3, distance1, distance2, mass));
...@@ -106,14 +60,30 @@ CpuSETTLE::~CpuSETTLE() { ...@@ -106,14 +60,30 @@ CpuSETTLE::~CpuSETTLE() {
delete threadSettle[i]; delete threadSettle[i];
} }
void CpuSETTLE::apply(vector<OpenMM::RealVec>& atomCoordinates, vector<OpenMM::RealVec>& atomCoordinatesP, vector<RealOpenMM>& inverseMasses, RealOpenMM tolerance) { void CpuSETTLE::apply(vector<OpenMM::Vec3>& atomCoordinates, vector<OpenMM::Vec3>& atomCoordinatesP, vector<double>& inverseMasses, double tolerance) {
ApplyToPositionsTask task(atomCoordinates, atomCoordinatesP, inverseMasses, tolerance, threadSettle); gmx_atomic_t atomicCounter;
threads.execute(task); gmx_atomic_set(&atomicCounter, 0);
threads.execute([&] (ThreadPool& threads, int threadIndex) {
while (true) {
int index = gmx_atomic_fetch_add(&atomicCounter, 1);
if (index >= threadSettle.size())
break;
threadSettle[index]->apply(atomCoordinates, atomCoordinatesP, inverseMasses, tolerance);
}
});
threads.waitForThreads(); threads.waitForThreads();
} }
void CpuSETTLE::applyToVelocities(vector<OpenMM::RealVec>& atomCoordinates, vector<OpenMM::RealVec>& velocities, vector<RealOpenMM>& inverseMasses, RealOpenMM tolerance) { void CpuSETTLE::applyToVelocities(vector<OpenMM::Vec3>& atomCoordinates, vector<OpenMM::Vec3>& velocities, vector<double>& inverseMasses, double tolerance) {
ApplyToVelocitiesTask task(atomCoordinates, velocities, inverseMasses, tolerance, threadSettle); gmx_atomic_t atomicCounter;
threads.execute(task); gmx_atomic_set(&atomicCounter, 0);
threads.execute([&] (ThreadPool& threads, int threadIndex) {
while (true) {
int index = gmx_atomic_fetch_add(&atomicCounter, 1);
if (index >= threadSettle.size())
break;
threadSettle[index]->applyToVelocities(atomCoordinates, velocities, inverseMasses, tolerance);
}
});
threads.waitForThreads(); threads.waitForThreads();
} }
platforms/cpu/staticTarget/CMakeLists.txt
View file @ 047934e2
...@@ -16,7 +16,6 @@ ENDFOREACH(file) ...@@ -16,7 +16,6 @@ ENDFOREACH(file)
ADD_LIBRARY(${STATIC_TARGET} STATIC ${SOURCE_FILES} ${SOURCE_INCLUDE_FILES} ${API_ABS_INCLUDE_FILES}) ADD_LIBRARY(${STATIC_TARGET} STATIC ${SOURCE_FILES} ${SOURCE_INCLUDE_FILES} ${API_ABS_INCLUDE_FILES})
TARGET_LINK_LIBRARIES(${STATIC_TARGET} ${OPENMM_LIBRARY_NAME}_static ${PTHREADS_LIB_STATIC}) TARGET_LINK_LIBRARIES(${STATIC_TARGET} ${OPENMM_LIBRARY_NAME}_static ${PTHREADS_LIB_STATIC})
#-DPTW32_STATIC_LIB only works for the windows pthreads. SET_TARGET_PROPERTIES(${STATIC_TARGET} PROPERTIES LINK_FLAGS "${EXTRA_LINK_FLAGS}" COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -DOPENMM_CPU_BUILDING_STATIC_LIBRARY")
SET_TARGET_PROPERTIES(${STATIC_TARGET} PROPERTIES LINK_FLAGS "${EXTRA_LINK_FLAGS}" COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -DOPENMM_CPU_BUILDING_STATIC_LIBRARY -DPTW32_STATIC_LIB")
INSTALL_TARGETS(/lib/plugins RUNTIME_DIRECTORY /lib/plugins ${STATIC_TARGET}) INSTALL_TARGETS(/lib/plugins RUNTIME_DIRECTORY /lib/plugins ${STATIC_TARGET})
platforms/cpu/tests/TestCpuDispersionPME.cpp 0 → 100644
View file @ 047934e2
/* -------------------------------------------------------------------------- *
* 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/cpu/tests/TestCpuNeighborList.cpp
View file @ 047934e2
...@@ -51,16 +51,16 @@ using namespace std; ...@@ -51,16 +51,16 @@ using namespace std;
void testNeighborList(bool periodic, bool triclinic) { void testNeighborList(bool periodic, bool triclinic) {
const int numParticles = 500; const int numParticles = 500;
const float cutoff = 2.0f; const float cutoff = 2.0f;
RealVec boxVectors[3]; Vec3 boxVectors[3];
if (triclinic) { if (triclinic) {
boxVectors[0] = RealVec(10, 0, 0); boxVectors[0] = Vec3(10, 0, 0);
boxVectors[1] = RealVec(4, 9, 0); boxVectors[1] = Vec3(4, 9, 0);
boxVectors[2] = RealVec(-3, -3.5, 11); boxVectors[2] = Vec3(-3, -3.5, 11);
} }
else { else {
boxVectors[0] = RealVec(10, 0, 0); boxVectors[0] = Vec3(10, 0, 0);
boxVectors[1] = RealVec(0, 9, 0); boxVectors[1] = Vec3(0, 9, 0);
boxVectors[2] = RealVec(0, 0, 11); boxVectors[2] = Vec3(0, 0, 11);
} }
const float boxSize[3] = {(float) boxVectors[0][0], (float) boxVectors[1][1], (float) boxVectors[2][2]}; const float boxSize[3] = {(float) boxVectors[0][0], (float) boxVectors[1][1], (float) boxVectors[2][2]};
const int blockSize = 8; const int blockSize = 8;
......
platforms/cuda/include/CudaContext.h
View file @ 047934e2
This diff is collapsed.
platforms/cuda/include/CudaKernels.h
View file @ 047934e2
This diff is collapsed.
platforms/cuda/src/CudaContext.cpp
View file @ 047934e2
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