Merged fork with latest original master

openmmapi/src/MonteCarloAnisotropicBarostatImpl.cpp

@@ -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) 2010-2013 Stanford University and the Authors.      *
+ * Portions copyright (c) 2010-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman, Lee-Ping Wang                                      *
  * Contributors:                                                              *
  *                                                                            *
@@ -119,7 +119,7 @@ void MonteCarloAnisotropicBarostatImpl::updateContextState(ContextImpl& context)
     // Compute the energy of the modified system.
     
     double finalEnergy = context.getOwner().getState(State::Energy).getPotentialEnergy();
-    double kT = BOLTZ*owner.getTemperature();
+    double kT = BOLTZ*context.getParameter(MonteCarloAnisotropicBarostat::Temperature());
     double w = finalEnergy-initialEnergy + pressure*deltaVolume - context.getMolecules().size()*kT*std::log(newVolume/volume);
     if (w > 0 && genrand_real2(random) > std::exp(-w/kT)) {
         // Reject the step.
@@ -150,6 +150,7 @@ std::map<std::string, double> MonteCarloAnisotropicBarostatImpl::getDefaultParam
     parameters[MonteCarloAnisotropicBarostat::PressureX()] = getOwner().getDefaultPressure()[0];
     parameters[MonteCarloAnisotropicBarostat::PressureY()] = getOwner().getDefaultPressure()[1];
     parameters[MonteCarloAnisotropicBarostat::PressureZ()] = getOwner().getDefaultPressure()[2];
+    parameters[MonteCarloAnisotropicBarostat::Temperature()] = getOwner().getDefaultTemperature();
     return parameters;
 }
 

openmmapi/src/MonteCarloBarostat.cpp

@@ -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) 2010 Stanford University and the Authors.           *
+ * Portions copyright (c) 2010-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -34,8 +34,8 @@
 
 using namespace OpenMM;
 
-MonteCarloBarostat::MonteCarloBarostat(double defaultPressure, double temperature, int frequency) :
-        defaultPressure(defaultPressure), temperature(temperature), frequency(frequency) {
+MonteCarloBarostat::MonteCarloBarostat(double defaultPressure, double defaultTemperature, int frequency) :
+        defaultPressure(defaultPressure), defaultTemperature(defaultTemperature), frequency(frequency) {
     setRandomNumberSeed(0);
 }
 

openmmapi/src/MonteCarloBarostatImpl.cpp

@@ -89,7 +89,7 @@ void MonteCarloBarostatImpl::updateContextState(ContextImpl& context) {
     
     double finalEnergy = context.getOwner().getState(State::Energy).getPotentialEnergy();
     double pressure = context.getParameter(MonteCarloBarostat::Pressure())*(AVOGADRO*1e-25);
-    double kT = BOLTZ*owner.getTemperature();
+    double kT = BOLTZ*context.getParameter(MonteCarloBarostat::Temperature());
     double w = finalEnergy-initialEnergy + pressure*deltaVolume - context.getMolecules().size()*kT*std::log(newVolume/volume);
     if (w > 0 && genrand_real2(random) > std::exp(-w/kT)) {
         // Reject the step.
@@ -118,6 +118,7 @@ void MonteCarloBarostatImpl::updateContextState(ContextImpl& context) {
 std::map<std::string, double> MonteCarloBarostatImpl::getDefaultParameters() {
     std::map<std::string, double> parameters;
     parameters[MonteCarloBarostat::Pressure()] = getOwner().getDefaultPressure();
+    parameters[MonteCarloBarostat::Temperature()] = getOwner().getDefaultTemperature();
     return parameters;
 }
 

openmmapi/src/MonteCarloMembraneBarostat.cpp

@@ -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) 2010-2014 Stanford University and the Authors.      *
+ * Portions copyright (c) 2010-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -34,8 +34,8 @@
 
 using namespace OpenMM;
 
-MonteCarloMembraneBarostat::MonteCarloMembraneBarostat(double defaultPressure, double defaultSurfaceTension, double temperature, XYMode xymode, ZMode zmode, int frequency) :
-        defaultPressure(defaultPressure), defaultSurfaceTension(defaultSurfaceTension), temperature(temperature),
+MonteCarloMembraneBarostat::MonteCarloMembraneBarostat(double defaultPressure, double defaultSurfaceTension, double defaultTemperature, XYMode xymode, ZMode zmode, int frequency) :
+        defaultPressure(defaultPressure), defaultSurfaceTension(defaultSurfaceTension), defaultTemperature(defaultTemperature),
         xymode(xymode), zmode(zmode), frequency(frequency) {
     setRandomNumberSeed(0);
 }

openmmapi/src/MonteCarloMembraneBarostatImpl.cpp

@@ -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) 2010-2014 Stanford University and the Authors.      *
+ * Portions copyright (c) 2010-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman, Lee-Ping Wang                                      *
  * Contributors:                                                              *
  *                                                                            *
@@ -120,7 +120,7 @@ void MonteCarloMembraneBarostatImpl::updateContextState(ContextImpl& context) {
     // Compute the energy of the modified system.
     
     double finalEnergy = context.getOwner().getState(State::Energy).getPotentialEnergy();
-    double kT = BOLTZ*owner.getTemperature();
+    double kT = BOLTZ*context.getParameter(MonteCarloMembraneBarostat::Temperature());
     double w = finalEnergy-initialEnergy + pressure*deltaVolume - tension*deltaArea - context.getMolecules().size()*kT*std::log(newVolume/volume);
     if (w > 0 && genrand_real2(random) > std::exp(-w/kT)) {
         // Reject the step.
@@ -150,6 +150,7 @@ std::map<std::string, double> MonteCarloMembraneBarostatImpl::getDefaultParamete
     std::map<std::string, double> parameters;
     parameters[MonteCarloMembraneBarostat::Pressure()] = getOwner().getDefaultPressure();
     parameters[MonteCarloMembraneBarostat::SurfaceTension()] = getOwner().getDefaultSurfaceTension();
+    parameters[MonteCarloMembraneBarostat::Temperature()] = getOwner().getDefaultTemperature();
     return parameters;
 }
 

openmmapi/src/NonbondedForce.cpp

@@ -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) 2008-2014 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -184,6 +184,9 @@ ForceImpl* NonbondedForce::createImpl() const {
 }
 
 void NonbondedForce::createExceptionsFromBonds(const vector<pair<int, int> >& bonds, double coulomb14Scale, double lj14Scale) {
+    for (int i = 0; i < (int) bonds.size(); ++i)
+        if (bonds[i].first < 0 || bonds[i].second < 0 || bonds[i].first >= particles.size() || bonds[i].second >= particles.size())
+            throw OpenMMException("createExceptionsFromBonds: Illegal particle index in list of bonds");
 
     // Find particles separated by 1, 2, or 3 bonds.
 

openmmapi/src/PeriodicTorsionForce.cpp

@@ -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) 2008-2012 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -37,7 +37,7 @@
 
 using namespace OpenMM;
 
-PeriodicTorsionForce::PeriodicTorsionForce() {
+PeriodicTorsionForce::PeriodicTorsionForce() : usePeriodic(false) {
 }
 
 int PeriodicTorsionForce::addTorsion(int particle1, int particle2, int particle3, int particle4, int periodicity, double phase, double k) {
@@ -74,3 +74,11 @@ ForceImpl* PeriodicTorsionForce::createImpl() const {
 void PeriodicTorsionForce::updateParametersInContext(Context& context) {
     dynamic_cast<PeriodicTorsionForceImpl&>(getImplInContext(context)).updateParametersInContext(getContextImpl(context));
 }
+
+void PeriodicTorsionForce::setUsesPeriodicBoundaryConditions(bool periodic) {
+    usePeriodic = periodic;
+}
+
+bool PeriodicTorsionForce::usesPeriodicBoundaryConditions() const {
+    return usePeriodic;
+}

openmmapi/src/RBTorsionForce.cpp

@@ -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) 2008-2012 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -37,7 +37,7 @@
 
 using namespace OpenMM;
 
-RBTorsionForce::RBTorsionForce() {
+RBTorsionForce::RBTorsionForce() : usePeriodic(false) {
 }
 
 int RBTorsionForce::addTorsion(int particle1, int particle2, int particle3, int particle4, double c0, double c1, double c2, double c3, double c4, double c5) {
@@ -80,3 +80,11 @@ ForceImpl* RBTorsionForce::createImpl() const {
 void RBTorsionForce::updateParametersInContext(Context& context) {
     dynamic_cast<RBTorsionForceImpl&>(getImplInContext(context)).updateParametersInContext(getContextImpl(context));
 }
+
+void RBTorsionForce::setUsesPeriodicBoundaryConditions(bool periodic) {
+    usePeriodic = periodic;
+}
+
+bool RBTorsionForce::usesPeriodicBoundaryConditions() const {
+    return usePeriodic;
+}

openmmapi/src/System.cpp

@@ -133,7 +133,7 @@ void System::setDefaultPeriodicBoxVectors(const Vec3& a, const Vec3& b, const Ve
     periodicBoxVectors[2] = c;
 }
 
-bool System::usesPeriodicBoundaryConditions() {
+bool System::usesPeriodicBoundaryConditions() const {
 
     bool uses_pbc = false;
     bool all_forces_implement = true;

platforms/cpu/include/CpuKernels.h

@@ -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) 2013 Stanford University and the Authors.           *
+ * Portions copyright (c) 2013-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -90,6 +90,7 @@ public:
 private:
     CpuPlatform::PlatformData& data;
     Kernel referenceKernel;
+    std::vector<RealVec> lastPositions;
 };
 
 /**
@@ -98,7 +99,7 @@ private:
 class CpuCalcHarmonicAngleForceKernel : public CalcHarmonicAngleForceKernel {
 public:
     CpuCalcHarmonicAngleForceKernel(std::string name, const Platform& platform, CpuPlatform::PlatformData& data) :
-            CalcHarmonicAngleForceKernel(name, platform), data(data), angleIndexArray(NULL), angleParamArray(NULL) {
+            CalcHarmonicAngleForceKernel(name, platform), data(data), angleIndexArray(NULL), angleParamArray(NULL), usePeriodic(false) {
     }
     ~CpuCalcHarmonicAngleForceKernel();
     /**
@@ -130,6 +131,7 @@ private:
     int **angleIndexArray;
     RealOpenMM **angleParamArray;
     CpuBondForce bondForce;
+    bool usePeriodic;
 };
 
 /**
@@ -138,7 +140,7 @@ private:
 class CpuCalcPeriodicTorsionForceKernel : public CalcPeriodicTorsionForceKernel {
 public:
     CpuCalcPeriodicTorsionForceKernel(std::string name, const Platform& platform, CpuPlatform::PlatformData& data) :
-            CalcPeriodicTorsionForceKernel(name, platform), data(data), torsionIndexArray(NULL), torsionParamArray(NULL) {
+            CalcPeriodicTorsionForceKernel(name, platform), data(data), torsionIndexArray(NULL), torsionParamArray(NULL), usePeriodic(false) {
     }
     ~CpuCalcPeriodicTorsionForceKernel();
     /**
@@ -170,6 +172,7 @@ private:
     int **torsionIndexArray;
     RealOpenMM **torsionParamArray;
     CpuBondForce bondForce;
+    bool usePeriodic;
 };
 
 /**
@@ -178,7 +181,7 @@ private:
 class CpuCalcRBTorsionForceKernel : public CalcRBTorsionForceKernel {
 public:
     CpuCalcRBTorsionForceKernel(std::string name, const Platform& platform, CpuPlatform::PlatformData& data) :
-            CalcRBTorsionForceKernel(name, platform), data(data), torsionIndexArray(NULL), torsionParamArray(NULL) {
+            CalcRBTorsionForceKernel(name, platform), data(data), torsionIndexArray(NULL), torsionParamArray(NULL), usePeriodic(false) {
     }
     ~CpuCalcRBTorsionForceKernel();
     /**
@@ -210,6 +213,7 @@ private:
     int **torsionIndexArray;
     RealOpenMM **torsionParamArray;
     CpuBondForce bondForce;
+    bool usePeriodic;
 };
 
 /**
@@ -264,9 +268,7 @@ private:
     bool useSwitchingFunction, useOptimizedPme, hasInitializedPme;
     std::vector<std::set<int> > exclusions;
     std::vector<std::pair<float, float> > particleParams;
-    std::vector<RealVec> lastPositions;
     NonbondedMethod nonbondedMethod;
-    CpuNeighborList* neighborList;
     CpuNonbondedForce* nonbonded;
     Kernel optimizedPme;
     CpuBondForce bondForce;
@@ -314,7 +316,6 @@ private:
     std::vector<std::string> parameterNames, globalParameterNames;
     std::vector<std::pair<std::set<int>, std::set<int> > > interactionGroups;
     NonbondedMethod nonbondedMethod;
-    CpuNeighborList* neighborList;
     CpuCustomNonbondedForce* nonbonded;
 };
 
@@ -400,7 +401,6 @@ private:
     std::vector<OpenMM::CustomGBForce::ComputationType> valueTypes;
     std::vector<OpenMM::CustomGBForce::ComputationType> energyTypes;
     NonbondedMethod nonbondedMethod;
-    CpuNeighborList* neighborList;
 };
 
 /**

platforms/cpu/include/CpuNeighborList.h

@@ -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) 2013-2015 Stanford University and the Authors.      *
+ * Portions copyright (c) 2013-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -51,6 +51,7 @@ public:
     void computeNeighborList(int numAtoms, const AlignedArray<float>& atomLocations, const std::vector<std::set<int> >& exclusions,
             const RealVec* periodicBoxVectors, bool usePeriodic, float maxDistance, ThreadPool& threads);
     int getNumBlocks() const;
+    int getBlockSize() const;
     const std::vector<int>& getSortedAtoms() const;
     const std::vector<int>& getBlockNeighbors(int blockIndex) const;
     const std::vector<char>& getBlockExclusions(int blockIndex) const;

platforms/cpu/include/CpuPlatform.h

@@ -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) 2013 Stanford University and the Authors.           *
+ * Portions copyright (c) 2013-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -34,6 +34,7 @@
 
 #include "AlignedArray.h"
 #include "CpuRandom.h"
+#include "CpuNeighborList.h"
 #include "ReferencePlatform.h"
 #include "openmm/internal/ContextImpl.h"
 #include "openmm/internal/ThreadPool.h"
@@ -64,7 +65,7 @@ public:
      * This is the name of the parameter for selecting the number of threads to use.
      */
     static const std::string& CpuThreads() {
-        static const std::string key = "CpuThreads";
+        static const std::string key = "Threads";
         return key;
     }
     /**
@@ -80,12 +81,18 @@ private:
 class CpuPlatform::PlatformData {
 public:
     PlatformData(int numParticles, int numThreads);
+    ~PlatformData();
+    void requestNeighborList(double cutoffDistance, double padding, bool useExclusions, std::vector<std::set<int> >& exclusionList);
     AlignedArray<float> posq;
     std::vector<AlignedArray<float> > threadForce;
     ThreadPool threads;
     bool isPeriodic;
     CpuRandom random;
     std::map<std::string, std::string> propertyValues;
+    CpuNeighborList* neighborList;
+    double cutoff, paddedCutoff;
+    bool anyExclusions;
+    std::vector<std::set<int> > exclusions;
 };
 
 } // namespace OpenMM

platforms/cpu/src/CpuCustomGBForce.cpp

 
-/* Portions copyright (c) 2009-2014 Stanford University and Simbios.
+/* Portions copyright (c) 2009-2016 Stanford University and Simbios.
  * Contributors: Peter Eastman
  *
  * Permission is hereby granted, free of charge, to any person obtaining
@@ -294,12 +294,13 @@ void CpuCustomGBForce::calculateParticlePairValue(int index, ThreadData& data, i
             int blockIndex = gmx_atomic_fetch_add(reinterpret_cast<gmx_atomic_t*>(atomicCounter), 1);
             if (blockIndex >= neighborList->getNumBlocks())
                 break;
-            const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex];
+            const int blockSize = neighborList->getBlockSize();
+            const int* blockAtom = &neighborList->getSortedAtoms()[blockSize*blockIndex];
             const vector<int>& neighbors = neighborList->getBlockNeighbors(blockIndex);
             const vector<char>& blockExclusions = neighborList->getBlockExclusions(blockIndex);
             for (int i = 0; i < (int) neighbors.size(); i++) {
                 int first = neighbors[i];
-                for (int k = 0; k < 4; k++) {
+                for (int k = 0; k < blockSize; k++) {
                     if ((blockExclusions[i] & (1<<k)) == 0) {
                         int second = blockAtom[k];
                         if (useExclusions && exclusions[first].find(second) != exclusions[first].end())
@@ -379,12 +380,13 @@ void CpuCustomGBForce::calculateParticlePairEnergyTerm(int index, ThreadData& da
             int blockIndex = gmx_atomic_fetch_add(reinterpret_cast<gmx_atomic_t*>(atomicCounter), 1);
             if (blockIndex >= neighborList->getNumBlocks())
                 break;
-            const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex];
+            const int blockSize = neighborList->getBlockSize();
+            const int* blockAtom = &neighborList->getSortedAtoms()[blockSize*blockIndex];
             const vector<int>& neighbors = neighborList->getBlockNeighbors(blockIndex);
             const vector<char>& blockExclusions = neighborList->getBlockExclusions(blockIndex);
             for (int i = 0; i < (int) neighbors.size(); i++) {
                 int first = neighbors[i];
-                for (int k = 0; k < 4; k++) {
+                for (int k = 0; k < blockSize; k++) {
                     if ((blockExclusions[i] & (1<<k)) == 0) {
                         int second = blockAtom[k];
                         if (useExclusions && exclusions[first].find(second) != exclusions[first].end())
@@ -460,12 +462,13 @@ void CpuCustomGBForce::calculateChainRuleForces(ThreadData& data, int numAtoms,
             int blockIndex = gmx_atomic_fetch_add(reinterpret_cast<gmx_atomic_t*>(atomicCounter), 1);
             if (blockIndex >= neighborList->getNumBlocks())
                 break;
-            const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex];
+            const int blockSize = neighborList->getBlockSize();
+            const int* blockAtom = &neighborList->getSortedAtoms()[blockSize*blockIndex];
             const vector<int>& neighbors = neighborList->getBlockNeighbors(blockIndex);
             const vector<char>& blockExclusions = neighborList->getBlockExclusions(blockIndex);
             for (int i = 0; i < (int) neighbors.size(); i++) {
                 int first = neighbors[i];
-                for (int k = 0; k < 4; k++) {
+                for (int k = 0; k < blockSize; k++) {
                     if ((blockExclusions[i] & (1<<k)) == 0) {
                         int second = blockAtom[k];
                         bool isExcluded = (exclusions[first].find(second) != exclusions[first].end());

platforms/cpu/src/CpuCustomNonbondedForce.cpp

 
-/* Portions copyright (c) 2009-2014 Stanford University and Simbios.
+/* Portions copyright (c) 2009-2016 Stanford University and Simbios.
  * Contributors: Peter Eastman
  *
  * Permission is hereby granted, free of charge, to any person obtaining
@@ -190,7 +190,8 @@ void CpuCustomNonbondedForce::threadComputeForce(ThreadPool& threads, int thread
             int blockIndex = gmx_atomic_fetch_add(reinterpret_cast<gmx_atomic_t*>(atomicCounter), 1);
             if (blockIndex >= neighborList->getNumBlocks())
                 break;
-            const int* blockAtom = &neighborList->getSortedAtoms()[4*blockIndex];
+            const int blockSize = neighborList->getBlockSize();
+            const int* blockAtom = &neighborList->getSortedAtoms()[blockSize*blockIndex];
             const vector<int>& neighbors = neighborList->getBlockNeighbors(blockIndex);
             const vector<char>& exclusions = neighborList->getBlockExclusions(blockIndex);
             for (int i = 0; i < (int) neighbors.size(); i++) {
@@ -199,7 +200,7 @@ void CpuCustomNonbondedForce::threadComputeForce(ThreadPool& threads, int thread
                     ReferenceForce::setVariable(data.energyParticleParams[j*2], atomParameters[first][j]);
                     ReferenceForce::setVariable(data.forceParticleParams[j*2], atomParameters[first][j]);
                 }
-                for (int k = 0; k < 4; k++) {
+                for (int k = 0; k < blockSize; k++) {
                     if ((exclusions[i] & (1<<k)) == 0) {
                         int second = blockAtom[k];
                         for (int j = 0; j < (int) paramNames.size(); j++) {

platforms/cpu/src/CpuKernels.cpp

@@ -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) 2013-2015 Stanford University and the Authors.      *
+ * Portions copyright (c) 2013-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -230,6 +230,7 @@ CpuCalcForcesAndEnergyKernel::CpuCalcForcesAndEnergyKernel(std::string name, con
 
 void CpuCalcForcesAndEnergyKernel::initialize(const System& system) {
     referenceKernel.getAs<ReferenceCalcForcesAndEnergyKernel>().initialize(system);
+    lastPositions.resize(system.getNumParticles(), Vec3(1e10, 1e10, 1e10));
 }
 
 void CpuCalcForcesAndEnergyKernel::beginComputation(ContextImpl& context, bool includeForce, bool includeEnergy, int groups) {
@@ -237,11 +238,60 @@ void CpuCalcForcesAndEnergyKernel::beginComputation(ContextImpl& context, bool i
     
     // Convert positions to single precision and clear the forces.
 
-    InitForceTask task(context.getSystem().getNumParticles(), context, data);
+    int numParticles = context.getSystem().getNumParticles();
+    InitForceTask task(numParticles, context, data);
     data.threads.execute(task);
     data.threads.waitForThreads();
     if (!task.positionsValid)
         throw OpenMMException("Particle coordinate is nan");
+
+    // Determine whether we need to recompute the neighbor list.
+        
+    if (data.neighborList != NULL) {
+        double padding = data.paddedCutoff-data.cutoff;;
+        bool needRecompute = false;
+        double closeCutoff2 = 0.25*padding*padding;
+        double farCutoff2 = 0.5*padding*padding;
+        int maxNumMoved = numParticles/10;
+        vector<int> moved;
+        vector<RealVec>& posData = extractPositions(context);
+        for (int i = 0; i < numParticles; i++) {
+            RealVec delta = posData[i]-lastPositions[i];
+            double dist2 = delta.dot(delta);
+            if (dist2 > closeCutoff2) {
+                moved.push_back(i);
+                if (dist2 > farCutoff2 || moved.size() > maxNumMoved) {
+                    needRecompute = true;
+                    break;
+                }
+            }
+        }
+        if (!needRecompute && moved.size() > 0) {
+            // Some particles have moved further than half the padding distance.  Look for pairs
+            // that are missing from the neighbor list.
+
+            int numMoved = moved.size();
+            double cutoff2 = data.cutoff*data.cutoff;
+            double paddedCutoff2 = data.paddedCutoff*data.paddedCutoff;
+            for (int i = 1; i < numMoved && !needRecompute; i++)
+                for (int j = 0; j < i; j++) {
+                    RealVec delta = posData[moved[i]]-posData[moved[j]];
+                    if (delta.dot(delta) < cutoff2) {
+                        // These particles should interact.  See if they are in the neighbor list.
+                        
+                        RealVec oldDelta = lastPositions[moved[i]]-lastPositions[moved[j]];
+                        if (oldDelta.dot(oldDelta) > paddedCutoff2) {
+                            needRecompute = true;
+                            break;
+                        }
+                    }
+                }
+        }
+        if (needRecompute) {
+            data.neighborList->computeNeighborList(numParticles, data.posq, data.exclusions, extractBoxVectors(context), data.isPeriodic, data.paddedCutoff, data.threads);
+            lastPositions = posData;
+        }
+    }
 }
 
 double CpuCalcForcesAndEnergyKernel::finishComputation(ContextImpl& context, bool includeForce, bool includeEnergy, int groups, bool& valid) {
@@ -283,6 +333,7 @@ void CpuCalcHarmonicAngleForceKernel::initialize(const System& system, const Har
         angleParamArray[i][1] = (RealOpenMM) k;
     }
     bondForce.initialize(system.getNumParticles(), numAngles, 3, angleIndexArray, data.threads);
+    usePeriodic = force.usesPeriodicBoundaryConditions();
 }
 
 double CpuCalcHarmonicAngleForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
@@ -290,6 +341,8 @@ double CpuCalcHarmonicAngleForceKernel::execute(ContextImpl& context, bool inclu
     vector<RealVec>& forceData = extractForces(context);
     RealOpenMM energy = 0;
     ReferenceAngleBondIxn angleBond;
+    if (usePeriodic)
+        angleBond.setPeriodic(extractBoxVectors(context));
     bondForce.calculateForce(posData, angleParamArray, forceData, includeEnergy ? &energy : NULL, angleBond);
     return energy;
 }
@@ -343,6 +396,7 @@ void CpuCalcPeriodicTorsionForceKernel::initialize(const System& system, const P
         torsionParamArray[i][2] = (RealOpenMM) periodicity;
     }
     bondForce.initialize(system.getNumParticles(), numTorsions, 4, torsionIndexArray, data.threads);
+    usePeriodic = force.usesPeriodicBoundaryConditions();
 }
 
 double CpuCalcPeriodicTorsionForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
@@ -350,6 +404,8 @@ double CpuCalcPeriodicTorsionForceKernel::execute(ContextImpl& context, bool inc
     vector<RealVec>& forceData = extractForces(context);
     RealOpenMM energy = 0;
     ReferenceProperDihedralBond periodicTorsionBond;
+    if (usePeriodic)
+        periodicTorsionBond.setPeriodic(extractBoxVectors(context));
     bondForce.calculateForce(posData, torsionParamArray, forceData, includeEnergy ? &energy : NULL, periodicTorsionBond);
     return energy;
 }
@@ -407,6 +463,7 @@ void CpuCalcRBTorsionForceKernel::initialize(const System& system, const RBTorsi
         torsionParamArray[i][5] = (RealOpenMM) c5;
     }
     bondForce.initialize(system.getNumParticles(), numTorsions, 4, torsionIndexArray, data.threads);
+    usePeriodic = force.usesPeriodicBoundaryConditions();
 }
 
 double CpuCalcRBTorsionForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
@@ -414,6 +471,8 @@ double CpuCalcRBTorsionForceKernel::execute(ContextImpl& context, bool includeFo
     vector<RealVec>& forceData = extractForces(context);
     RealOpenMM energy = 0;
     ReferenceRbDihedralBond rbTorsionBond;
+    if (usePeriodic)
+        rbTorsionBond.setPeriodic(extractBoxVectors(context));
     bondForce.calculateForce(posData, torsionParamArray, forceData, includeEnergy ? &energy : NULL, rbTorsionBond);
     return energy;
 }
@@ -464,15 +523,11 @@ CpuNonbondedForce* createCpuNonbondedForceVec4();
 CpuNonbondedForce* createCpuNonbondedForceVec8();
 
 CpuCalcNonbondedForceKernel::CpuCalcNonbondedForceKernel(string name, const Platform& platform, CpuPlatform::PlatformData& data) : CalcNonbondedForceKernel(name, platform),
-        data(data), bonded14IndexArray(NULL), bonded14ParamArray(NULL), hasInitializedPme(false), neighborList(NULL), nonbonded(NULL) {
-    if (isVec8Supported()) {
-        neighborList = new CpuNeighborList(8);
+        data(data), bonded14IndexArray(NULL), bonded14ParamArray(NULL), hasInitializedPme(false), nonbonded(NULL) {
+    if (isVec8Supported())
         nonbonded = createCpuNonbondedForceVec8();
-    }
-    else {
-        neighborList = new CpuNeighborList(4);
+    else
         nonbonded = createCpuNonbondedForceVec4();
-    }
 }
 
 CpuCalcNonbondedForceKernel::~CpuCalcNonbondedForceKernel() {
@@ -486,8 +541,6 @@ CpuCalcNonbondedForceKernel::~CpuCalcNonbondedForceKernel() {
     }
     if (nonbonded != NULL)
         delete nonbonded;
-    if (neighborList != NULL)
-        delete neighborList;
 }
 
 void CpuCalcNonbondedForceKernel::initialize(const System& system, const NonbondedForce& force) {
@@ -547,6 +600,7 @@ void CpuCalcNonbondedForceKernel::initialize(const System& system, const Nonbond
     if (nonbondedMethod == NoCutoff)
         useSwitchingFunction = false;
     else {
+        data.requestNeighborList(nonbondedCutoff, 0.25*nonbondedCutoff, true, exclusions);
         useSwitchingFunction = force.getUseSwitchingFunction();
         switchingDistance = force.getSwitchingDistance();
     }
@@ -569,7 +623,6 @@ void CpuCalcNonbondedForceKernel::initialize(const System& system, const Nonbond
         dispersionCoefficient = NonbondedForceImpl::calcDispersionCorrection(system, force);
     else
         dispersionCoefficient = 0.0;
-    lastPositions.resize(numParticles, Vec3(1e10, 1e10, 1e10));
     data.isPeriodic = (nonbondedMethod == CutoffPeriodic || nonbondedMethod == Ewald || nonbondedMethod == PME);
 }
 
@@ -596,53 +649,8 @@ double CpuCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeFo
     double energy = (includeReciprocal ? ewaldSelfEnergy : 0.0);
     bool ewald  = (nonbondedMethod == Ewald);
     bool pme  = (nonbondedMethod == PME);
-    if (nonbondedMethod != NoCutoff) {
-        // Determine whether we need to recompute the neighbor list.
-        
-        double padding = 0.25*nonbondedCutoff;
-        bool needRecompute = false;
-        double closeCutoff2 = 0.25*padding*padding;
-        double farCutoff2 = 0.5*padding*padding;
-        int maxNumMoved = numParticles/10;
-        vector<int> moved;
-        for (int i = 0; i < numParticles; i++) {
-            RealVec delta = posData[i]-lastPositions[i];
-            double dist2 = delta.dot(delta);
-            if (dist2 > closeCutoff2) {
-                moved.push_back(i);
-                if (dist2 > farCutoff2 || moved.size() > maxNumMoved) {
-                    needRecompute = true;
-                    break;
-                }
-            }
-        }
-        if (!needRecompute && moved.size() > 0) {
-            // Some particles have moved further than half the padding distance.  Look for pairs
-            // that are missing from the neighbor list.
-
-            int numMoved = moved.size();
-            double cutoff2 = nonbondedCutoff*nonbondedCutoff;
-            double paddedCutoff2 = (nonbondedCutoff+padding)*(nonbondedCutoff+padding);
-            for (int i = 1; i < numMoved && !needRecompute; i++)
-                for (int j = 0; j < i; j++) {
-                    RealVec delta = posData[moved[i]]-posData[moved[j]];
-                    if (delta.dot(delta) < cutoff2) {
-                        // These particles should interact.  See if they are in the neighbor list.
-                        
-                        RealVec oldDelta = lastPositions[moved[i]]-lastPositions[moved[j]];
-                        if (oldDelta.dot(oldDelta) > paddedCutoff2) {
-                            needRecompute = true;
-                            break;
-                        }
-                    }
-                }
-        }
-        if (needRecompute) {
-            neighborList->computeNeighborList(numParticles, posq, exclusions, boxVectors, data.isPeriodic, nonbondedCutoff+padding, data.threads);
-            lastPositions = posData;
-        }
-        nonbonded->setUseCutoff(nonbondedCutoff, *neighborList, rfDielectric);
-    }
+    if (nonbondedMethod != NoCutoff)
+        nonbonded->setUseCutoff(nonbondedCutoff, *data.neighborList, rfDielectric);
     if (data.isPeriodic) {
         RealVec* boxVectors = extractBoxVectors(context);
         double minAllowedSize = 1.999999*nonbondedCutoff;
@@ -739,7 +747,7 @@ void CpuCalcNonbondedForceKernel::getPMEParameters(double& alpha, int& nx, int&
 }
 
 CpuCalcCustomNonbondedForceKernel::CpuCalcCustomNonbondedForceKernel(string name, const Platform& platform, CpuPlatform::PlatformData& data) :
-            CalcCustomNonbondedForceKernel(name, platform), data(data), forceCopy(NULL), neighborList(NULL), nonbonded(NULL) {
+            CalcCustomNonbondedForceKernel(name, platform), data(data), forceCopy(NULL), nonbonded(NULL) {
 }
 
 CpuCalcCustomNonbondedForceKernel::~CpuCalcCustomNonbondedForceKernel() {
@@ -748,8 +756,6 @@ CpuCalcCustomNonbondedForceKernel::~CpuCalcCustomNonbondedForceKernel() {
             delete[] particleParamArray[i];
         delete[] particleParamArray;
     }
-    if (neighborList != NULL)
-        delete neighborList;
     if (nonbonded != NULL)
         delete nonbonded;
     if (forceCopy != NULL)
@@ -786,7 +792,7 @@ void CpuCalcCustomNonbondedForceKernel::initialize(const System& system, const C
     if (nonbondedMethod == NoCutoff)
         useSwitchingFunction = false;
     else {
-        neighborList = new CpuNeighborList(4);
+        data.requestNeighborList(nonbondedCutoff, 0.25*nonbondedCutoff, true, exclusions);
         useSwitchingFunction = force.getUseSwitchingFunction();
         switchingDistance = force.getSwitchingDistance();
     }
@@ -852,10 +858,8 @@ double CpuCalcCustomNonbondedForceKernel::execute(ContextImpl& context, bool inc
     RealVec* boxVectors = extractBoxVectors(context);
     double energy = 0;
     bool periodic = (nonbondedMethod == CutoffPeriodic);
-    if (nonbondedMethod != NoCutoff) {
-        neighborList->computeNeighborList(numParticles, data.posq, exclusions, boxVectors, data.isPeriodic, nonbondedCutoff, data.threads);
-        nonbonded->setUseCutoff(nonbondedCutoff, *neighborList);
-    }
+    if (nonbondedMethod != NoCutoff)
+        nonbonded->setUseCutoff(nonbondedCutoff, *data.neighborList);
     if (periodic) {
         double minAllowedSize = 2*nonbondedCutoff;
         if (boxVectors[0][0] < minAllowedSize || boxVectors[1][1] < minAllowedSize || boxVectors[2][2] < minAllowedSize)
@@ -963,8 +967,6 @@ CpuCalcCustomGBForceKernel::~CpuCalcCustomGBForceKernel() {
             delete[] particleParamArray[i];
         delete[] particleParamArray;
     }
-    if (neighborList != NULL)
-        delete neighborList;
     if (ixn != NULL)
         delete ixn;
 }
@@ -1012,10 +1014,8 @@ void CpuCalcCustomGBForceKernel::initialize(const System& system, const CustomGB
         globalParameterNames.push_back(force.getGlobalParameterName(i));
     nonbondedMethod = CalcCustomGBForceKernel::NonbondedMethod(force.getNonbondedMethod());
     nonbondedCutoff = (RealOpenMM) force.getCutoffDistance();
-    if (nonbondedMethod == NoCutoff)
-        neighborList = NULL;
-    else
-        neighborList = new CpuNeighborList(4);
+    if (nonbondedMethod != NoCutoff)
+        data.requestNeighborList(nonbondedCutoff, 0.25*nonbondedCutoff, true, exclusions);
 
     // Create custom functions for the tabulated functions.
 
@@ -1112,8 +1112,7 @@ double CpuCalcCustomGBForceKernel::execute(ContextImpl& context, bool includeFor
         ixn->setPeriodic(extractBoxSize(context));
     if (nonbondedMethod != NoCutoff) {
         vector<set<int> > noExclusions(numParticles);
-        neighborList->computeNeighborList(numParticles, data.posq, exclusions, boxVectors, data.isPeriodic, nonbondedCutoff, data.threads);
-        ixn->setUseCutoff(nonbondedCutoff, *neighborList);
+        ixn->setUseCutoff(nonbondedCutoff, *data.neighborList);
     }
     map<string, double> globalParameters;
     for (int i = 0; i < (int) globalParameterNames.size(); i++)

platforms/cpu/src/CpuNeighborList.cpp

@@ -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) 2013-2015 Stanford University and the Authors.      *
+ * Portions copyright (c) 2013-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -158,8 +158,8 @@ public:
             float scale1 = floorf(yperiodic*recipBoxSize[1]);
             yperiodic -= periodicBoxVectors[1][0]*scale1;
         }
-        int y = min(ny-1, int(floorf(yperiodic / voxelSizeY)));
-        int z = min(nz-1, int(floorf(zperiodic / voxelSizeZ)));
+        int y = max(0, min(ny-1, int(floorf(yperiodic / voxelSizeY))));
+        int z = max(0, min(nz-1, int(floorf(zperiodic / voxelSizeZ))));
         
         return VoxelIndex(y, z);
     }
@@ -225,38 +225,78 @@ public:
                 if (usePeriodic)
                     voxelIndex.y = (y < 0 ? y+ny : (y >= ny ? y-ny : y));
                 float boxy = floor((float) y/ny);
-                float xoffset = (float) (usePeriodic ? boxy*periodicBoxVectors[1][0]+boxz*periodicBoxVectors[2][0] : 0);
                 
                 // Identify the range of atoms within this bin we need to search.  When using periodic boundary
                 // conditions, there may be two separate ranges.
                 
                 float minx = centerPos[0];
                 float maxx = centerPos[0];
-                fvec4 offset(-xoffset, -yoffset+voxelSizeY*y+(usePeriodic ? 0.0f : miny), voxelSizeZ*z+(usePeriodic ? 0.0f : minz), 0);
-                for (int k = 0; k < (int) blockAtoms.size(); k++) {
-                    const float* atomPos = &sortedPositions[4*(blockSize*blockIndex+k)];
-                    fvec4 posVec(atomPos);
-                    fvec4 delta1 = offset-posVec;
-                    fvec4 delta2 = delta1+fvec4(0, voxelSizeY, voxelSizeZ, 0);
-                    if (usePeriodic) {
-                        delta1 -= round(delta1*invBoxSize)*boxSize;
-                        delta2 -= round(delta2*invBoxSize)*boxSize;
+                if (usePeriodic && triclinic) {
+                    for (int k = 0; k < (int) blockAtoms.size(); k++) {
+                        const float* atomPos = &sortedPositions[4*(blockSize*blockIndex+k)];
+                        fvec4 delta1(0, voxelSizeY*voxelIndex.y-atomPos[1], voxelSizeZ*voxelIndex.z-atomPos[2], 0);
+                        fvec4 delta2 = delta1+fvec4(0, voxelSizeY, 0, 0);
+                        fvec4 delta3 = delta1+fvec4(0, 0, voxelSizeZ, 0);
+                        fvec4 delta4 = delta1+fvec4(0, voxelSizeY, voxelSizeZ, 0);
+                        delta1 -= periodicBoxVec4[2]*floorf(delta1[2]*recipBoxSize[2]+0.5f);
+                        delta1 -= periodicBoxVec4[1]*floorf(delta1[1]*recipBoxSize[1]+0.5f);
+                        delta1 -= periodicBoxVec4[0]*floorf(delta1[0]*recipBoxSize[0]+0.5f);
+                        delta2 -= periodicBoxVec4[2]*floorf(delta2[2]*recipBoxSize[2]+0.5f);
+                        delta2 -= periodicBoxVec4[1]*floorf(delta2[1]*recipBoxSize[1]+0.5f);
+                        delta2 -= periodicBoxVec4[0]*floorf(delta2[0]*recipBoxSize[0]+0.5f);
+                        delta3 -= periodicBoxVec4[2]*floorf(delta3[2]*recipBoxSize[2]+0.5f);
+                        delta3 -= periodicBoxVec4[1]*floorf(delta3[1]*recipBoxSize[1]+0.5f);
+                        delta3 -= periodicBoxVec4[0]*floorf(delta3[0]*recipBoxSize[0]+0.5f);
+                        delta4 -= periodicBoxVec4[2]*floorf(delta4[2]*recipBoxSize[2]+0.5f);
+                        delta4 -= periodicBoxVec4[1]*floorf(delta4[1]*recipBoxSize[1]+0.5f);
+                        delta4 -= periodicBoxVec4[0]*floorf(delta4[0]*recipBoxSize[0]+0.5f);
+                        if (delta1[1] < 0 && delta1[1]+voxelSizeY > 0)
+                            delta1 = fvec4(delta1[0], 0, delta1[2], 0);
+                        if (delta1[2] < 0 && delta1[2]+voxelSizeZ > 0)
+                            delta1 = fvec4(delta1[0], delta1[1], 0, 0);
+                        if (delta3[1] < 0 && delta3[1]+voxelSizeY > 0)
+                            delta3 = fvec4(delta3[0], 0, delta3[2], 0);
+                        if (delta2[2] < 0 && delta2[2]+voxelSizeZ > 0)
+                            delta2 = fvec4(delta2[0], delta2[1], 0, 0);
+                        fvec4 delta = min(min(min(abs(delta1), abs(delta2)), abs(delta3)), abs(delta4));
+                        float dy = (voxelIndex.y == atomVoxelIndex[k].y ? 0.0f : delta[1]);
+                        float dz = (voxelIndex.z == atomVoxelIndex[k].z ? 0.0f : delta[2]);
+                        float dist2 = maxDistanceSquared-dy*dy-dz*dz;
+                        if (dist2 > 0) {
+                            float dist = sqrtf(dist2);
+                            minx = min(minx, atomPos[0]-dist-max(max(max(delta1[0], delta2[0]), delta3[0]), delta4[0]));
+                            maxx = max(maxx, atomPos[0]+dist-min(min(min(delta1[0], delta2[0]), delta3[0]), delta4[0]));
+                        }
                     }
-                    fvec4 delta = min(abs(delta1), abs(delta2));
-                    float dy = (y == atomVoxelIndex[k].y ? 0.0f : delta[1]);
-                    float dz = (z == atomVoxelIndex[k].z ? 0.0f : delta[2]);
-                    float dist2 = maxDistanceSquared-dy*dy-dz*dz;
-                    if (dist2 > 0) {
-                        float dist = sqrtf(dist2);
-                        minx = min(minx, atomPos[0]-dist-xoffset);
-                        maxx = max(maxx, atomPos[0]+dist-xoffset);
+                }
+                else {
+                    float xoffset = (float) (usePeriodic ? boxy*periodicBoxVectors[1][0]+boxz*periodicBoxVectors[2][0] : 0);
+                    fvec4 offset(-xoffset, -yoffset+voxelSizeY*y+(usePeriodic ? 0.0f : miny), voxelSizeZ*z+(usePeriodic ? 0.0f : minz), 0);
+                    for (int k = 0; k < (int) blockAtoms.size(); k++) {
+                        const float* atomPos = &sortedPositions[4*(blockSize*blockIndex+k)];
+                        fvec4 posVec(atomPos);
+                        fvec4 delta1 = offset-posVec;
+                        fvec4 delta2 = delta1+fvec4(0, voxelSizeY, voxelSizeZ, 0);
+                        if (usePeriodic) {
+                            delta1 -= round(delta1*invBoxSize)*boxSize;
+                            delta2 -= round(delta2*invBoxSize)*boxSize;
+                        }
+                        fvec4 delta = min(abs(delta1), abs(delta2));
+                        float dy = (y == atomVoxelIndex[k].y ? 0.0f : delta[1]);
+                        float dz = (z == atomVoxelIndex[k].z ? 0.0f : delta[2]);
+                        float dist2 = maxDistanceSquared-dy*dy-dz*dz;
+                        if (dist2 > 0) {
+                            float dist = sqrtf(dist2);
+                            minx = min(minx, atomPos[0]-dist-xoffset);
+                            maxx = max(maxx, atomPos[0]+dist-xoffset);
+                        }
                     }
                 }
                 if (minx == maxx)
                     continue;
-                bool needPeriodic = (centerPos[1]-blockWidth[1] < maxDistance || centerPos[1]+blockWidth[1] > periodicBoxSize[1]-maxDistance ||
-                                     centerPos[2]-blockWidth[2] < maxDistance || centerPos[2]+blockWidth[2] > periodicBoxSize[2]-maxDistance ||
-                                     minx < 0.0f || maxx > periodicBoxVectors[0][0]);
+                bool needPeriodic = usePeriodic && (centerPos[1]-blockWidth[1] < maxDistance || centerPos[1]+blockWidth[1] > periodicBoxSize[1]-maxDistance ||
+                                                    centerPos[2]-blockWidth[2] < maxDistance || centerPos[2]+blockWidth[2] > periodicBoxSize[2]-maxDistance ||
+                                                    minx < 0.0f || maxx > periodicBoxVectors[0][0]);
                 int numRanges;
                 int rangeStart[2];
                 int rangeEnd[2];
@@ -294,7 +334,7 @@ public:
                             continue;
                         
                         fvec4 atomPos(&sortedPositions[4*sortedIndex]);
-                        fvec4 delta = atomPos-centerPos;
+                        fvec4 delta = atomPos-blockCenter;
                         if (periodicRectangular)
                             delta -= round(delta*invBoxSize)*boxSize;
                         else if (needPeriodic) {
@@ -468,6 +508,10 @@ int CpuNeighborList::getNumBlocks() const {
     return sortedAtoms.size()/blockSize;
 }
 
+int CpuNeighborList::getBlockSize() const {
+    return blockSize;
+}
+
 const std::vector<int>& CpuNeighborList::getSortedAtoms() const {
     return sortedAtoms;
 }

platforms/cpu/src/CpuNonbondedForce.cpp

@@ -386,13 +386,15 @@ void CpuNonbondedForce::threadComputeDirect(ThreadPool& threads, int threadIndex
                             float inverseR = 1/r;
                             float chargeProdOverR = scaledChargeI*posq[4*j+3]*inverseR;
                             float dEdR = chargeProdOverR*inverseR*inverseR;
-                            dEdR = dEdR * (erfAlphaR-(float)TWO_OVER_SQRT_PI*alphaR*(float)exp(-alphaR*alphaR));
+                            dEdR = dEdR * (erfAlphaR-TWO_OVER_SQRT_PI*alphaR*(float)exp(-alphaR*alphaR));
                             fvec4 result = deltaR*dEdR;
                             (fvec4(forces+4*i)-result).store(forces+4*i);
                             (fvec4(forces+4*j)+result).store(forces+4*j);
                             if (includeEnergy)
                                 threadEnergy[threadIndex] -= chargeProdOverR*erfAlphaR;
                         }
+                        else if (includeEnergy)
+                           threadEnergy[threadIndex] -= alphaEwald*TWO_OVER_SQRT_PI*scaledChargeI*posq[4*j+3];
                     }
                 }
             }

platforms/cpu/src/CpuPlatform.cpp

@@ -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) 2013-2014 Stanford University and the Authors.      *
+ * Portions copyright (c) 2013-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -34,6 +34,7 @@
 #include "CpuKernels.h"
 #include "CpuSETTLE.h"
 #include "ReferenceConstraints.h"
+#include "openmm/OpenMMException.h"
 #include "openmm/internal/hardware.h"
 #include "openmm/internal/vectorize.h"
 #include <sstream>
@@ -59,6 +60,7 @@ extern "C" OPENMM_EXPORT_CPU void registerPlatforms() {
 map<const ContextImpl*, CpuPlatform::PlatformData*> CpuPlatform::contextData;
 
 CpuPlatform::CpuPlatform() {
+    deprecatedPropertyReplacements["CpuThreads"] = CpuThreads();
     CpuKernelFactory* factory = new CpuKernelFactory();
     registerKernelFactory(CalcForcesAndEnergyKernel::Name(), factory);
     registerKernelFactory(CalcHarmonicAngleForceKernel::Name(), factory);
@@ -83,7 +85,10 @@ CpuPlatform::CpuPlatform() {
 const string& CpuPlatform::getPropertyValue(const Context& context, const string& property) const {
     const ContextImpl& impl = getContextImpl(context);
     const PlatformData& data = getPlatformData(impl);
-    map<string, string>::const_iterator value = data.propertyValues.find(property);
+    string propertyName = property;
+    if (deprecatedPropertyReplacements.find(property) != deprecatedPropertyReplacements.end())
+        propertyName = deprecatedPropertyReplacements.find(property)->second;
+    map<string, string>::const_iterator value = data.propertyValues.find(propertyName);
     if (value != data.propertyValues.end())
         return value->second;
     return ReferencePlatform::getPropertyValue(context, property);
@@ -131,7 +136,8 @@ const CpuPlatform::PlatformData& CpuPlatform::getPlatformData(const ContextImpl&
     return *contextData[&context];
 }
 
-CpuPlatform::PlatformData::PlatformData(int numParticles, int numThreads) : posq(4*numParticles), threads(numThreads) {
+CpuPlatform::PlatformData::PlatformData(int numParticles, int numThreads) : posq(4*numParticles), threads(numThreads),
+        neighborList(NULL), cutoff(0.0), paddedCutoff(0.0), anyExclusions(false) {
     numThreads = threads.getNumThreads();
     threadForce.resize(numThreads);
     for (int i = 0; i < numThreads; i++)
@@ -141,3 +147,27 @@ CpuPlatform::PlatformData::PlatformData(int numParticles, int numThreads) : posq
     threadsProperty << numThreads;
     propertyValues[CpuThreads()] = threadsProperty.str();
 }
+
+CpuPlatform::PlatformData::~PlatformData() {
+    if (neighborList != NULL)
+        delete neighborList;
+}
+
+bool isVec8Supported();
+
+void CpuPlatform::PlatformData::requestNeighborList(double cutoffDistance, double padding, bool useExclusions, vector<set<int> >& exclusionList) {
+    if (neighborList == NULL)
+        neighborList = new CpuNeighborList(isVec8Supported() ? 8 : 4);
+    if (cutoffDistance > cutoff)
+        cutoff = cutoffDistance;
+    if (cutoffDistance+padding > paddedCutoff)
+        paddedCutoff = cutoffDistance+padding;
+    if (useExclusions) {
+        if (anyExclusions && exclusions != exclusionList)
+            throw OpenMMException("All Forces must have identical exclusions");
+        else {
+            exclusions = exclusionList;
+            anyExclusions = true;
+        }
+    }
+}

platforms/cpu/tests/TestCpuCheckpoints.cpp

@@ -92,6 +92,15 @@ void testCheckpoint() {
     integrator.step(10);
     State s4 = context.getState(State::Positions | State::Velocities | State::Parameters);
     compareStates(s2, s4);
+    
+    // See if a checkpoint created from one Context can be loaded into a different one.
+    
+    VerletIntegrator integrator2(0.001);
+    Context context2(system, integrator2, platform);
+    stream1.seekg(0, stream1.beg);
+    context2.loadCheckpoint(stream1);
+    State s5 = context2.getState(State::Positions | State::Velocities | State::Parameters | State::Energy);
+    compareStates(s1, s5);
 }
 
 void runPlatformTests() {

platforms/cpu/tests/TestCpuNeighborList.cpp

@@ -53,14 +53,14 @@ void testNeighborList(bool periodic, bool triclinic) {
     const float cutoff = 2.0f;
     RealVec boxVectors[3];
     if (triclinic) {
-        boxVectors[0] = RealVec(20, 0, 0);
-        boxVectors[1] = RealVec(5, 15, 0);
-        boxVectors[2] = RealVec(-3, -7, 22);
+        boxVectors[0] = RealVec(10, 0, 0);
+        boxVectors[1] = RealVec(4, 9, 0);
+        boxVectors[2] = RealVec(-3, -3.5, 11);
     }
     else {
-        boxVectors[0] = RealVec(20, 0, 0);
-        boxVectors[1] = RealVec(0, 15, 0);
-        boxVectors[2] = RealVec(0, 0, 22);
+        boxVectors[0] = RealVec(10, 0, 0);
+        boxVectors[1] = RealVec(0, 9, 0);
+        boxVectors[2] = RealVec(0, 0, 11);
     }
     const float boxSize[3] = {(float) boxVectors[0][0], (float) boxVectors[1][1], (float) boxVectors[2][2]};
     const int blockSize = 8;