Merge branch 'master' into gayberne

platforms/cuda/tests/TestCudaFFT3D.cpp

@@ -56,7 +56,7 @@ void testTransform(bool realToComplex, int xsize, int ysize, int zsize) {
     system.addParticle(0.0);
     CudaPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"), "false",
             platform.getPropertyDefaultValue(CudaPlatform::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()),
-            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), 1);
+            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", 1);
     CudaContext& context = *platformData.contexts[0];
     context.initialize();
     OpenMM_SFMT::SFMT sfmt;

platforms/cuda/tests/TestCudaNonbondedForce.cpp

@@ -118,9 +118,44 @@ void testReordering() {
     }
 }
 
+void testDeterministicForces() {
+    // Check that the CudaDeterministicForces property works correctly.
+    
+    const int numParticles = 1000;
+    System system;
+    system.setDefaultPeriodicBoxVectors(Vec3(6, 0, 0), Vec3(2.1, 6, 0), Vec3(-1.5, -0.5, 6));
+    NonbondedForce *nonbonded = new NonbondedForce();
+    nonbonded->setNonbondedMethod(NonbondedForce::PME);
+    system.addForce(nonbonded);
+    vector<Vec3> positions;
+    OpenMM_SFMT::SFMT sfmt;
+    init_gen_rand(0, sfmt);
+    for (int i = 0; i < numParticles; i++) {
+        system.addParticle(1.0);
+        nonbonded->addParticle(i%2 == 0 ? 1 : -1, 1, 0);
+        positions.push_back(Vec3(genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5)*6);
+    }
+    VerletIntegrator integrator(0.001);
+    map<string, string> properties;
+    properties[CudaPlatform::CudaDeterministicForces()] = "true";
+    Context context(system, integrator, platform, properties);
+    context.setPositions(positions);
+    State state1 = context.getState(State::Forces);
+    State state2 = context.getState(State::Forces);
+    
+    // All forces should be *exactly* equal.
+    
+    for (int i = 0; i < numParticles; i++) {
+        ASSERT_EQUAL(state1.getForces()[i][0], state2.getForces()[i][0]);
+        ASSERT_EQUAL(state1.getForces()[i][1], state2.getForces()[i][1]);
+        ASSERT_EQUAL(state1.getForces()[i][2], state2.getForces()[i][2]);
+    }
+}
+
 void runPlatformTests() {
     testParallelComputation(NonbondedForce::NoCutoff);
     testParallelComputation(NonbondedForce::Ewald);
     testParallelComputation(NonbondedForce::PME);
     testReordering();
+    testDeterministicForces();
 }

platforms/cuda/tests/TestCudaRandom.cpp

@@ -56,7 +56,7 @@ void testGaussian() {
         system.addParticle(1.0);
     CudaPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"), "false",
             platform.getPropertyDefaultValue(CudaPlatform::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()),
-            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), 1);
+            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", 1);
     CudaContext& context = *platformData.contexts[0];
     context.initialize();
     context.getIntegrationUtilities().initRandomNumberGenerator(0);

platforms/cuda/tests/TestCudaSort.cpp

@@ -66,7 +66,7 @@ void verifySorting(vector<float> array) {
     system.addParticle(0.0);
     CudaPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"), "false",
             platform.getPropertyDefaultValue(CudaPlatform::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()),
-            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), 1);
+            platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", 1);
     CudaContext& context = *platformData.contexts[0];
     context.initialize();
     CudaArray data(context, array.size(), 4, "sortData");

platforms/opencl/include/OpenCLKernels.h

@@ -156,7 +156,7 @@ public:
      * @param b      the vector defining the second edge of the periodic box
      * @param c      the vector defining the third edge of the periodic box
      */
-    void setPeriodicBoxVectors(ContextImpl& context, const Vec3& a, const Vec3& b, const Vec3& c) const;
+    void setPeriodicBoxVectors(ContextImpl& context, const Vec3& a, const Vec3& b, const Vec3& c);
     /**
      * Create a checkpoint recording the current state of the Context.
      * 
@@ -698,7 +698,7 @@ public:
      */
     void copyParametersToContext(ContextImpl& context, const CustomNonbondedForce& force);
 private:
-    void initInteractionGroups(const CustomNonbondedForce& force, const std::string& interactionSource);
+    void initInteractionGroups(const CustomNonbondedForce& force, const std::string& interactionSource, const std::vector<std::string>& tableTypes);
     OpenCLContext& cl;
     OpenCLParameterSet* params;
     OpenCLArray* globals;

platforms/opencl/include/OpenCLPlatform.h

@@ -58,14 +58,14 @@ public:
      * This is the name of the parameter for selecting which OpenCL device or devices to use.
      */
     static const std::string& OpenCLDeviceIndex() {
-        static const std::string key = "OpenCLDeviceIndex";
+        static const std::string key = "DeviceIndex";
         return key;
     }
     /**
      * This is the name of the parameter that reports the OpenCL device or devices being used.
      */
     static const std::string& OpenCLDeviceName() {
-        static const std::string key = "OpenCLDeviceName";
+        static const std::string key = "DeviceName";
         return key;
     }
     /**
@@ -86,21 +86,21 @@ public:
      * This is the name of the parameter for selecting what numerical precision to use.
      */
     static const std::string& OpenCLPrecision() {
-        static const std::string key = "OpenCLPrecision";
+        static const std::string key = "Precision";
         return key;
     }
     /**
      * This is the name of the parameter for selecting whether to use the CPU based PME calculation.
      */
     static const std::string& OpenCLUseCpuPme() {
-        static const std::string key = "OpenCLUseCpuPme";
+        static const std::string key = "UseCpuPme";
         return key;
     }
     /**
      * This is the name of the parameter for selecting whether to disable use of a separate stream for PME.
      */
     static const std::string& OpenCLDisablePmeStream() {
-        static const std::string key = "OpenCLDisablePmeStream";
+        static const std::string key = "DisablePmeStream";
         return key;
     }
 };

platforms/opencl/src/OpenCLContext.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) 2009-2015 Stanford University and the Authors.      *
+ * Portions copyright (c) 2009-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -84,7 +84,7 @@ OpenCLContext::OpenCLContext(const System& system, int platformIndex, int device
         useMixedPrecision = false;
     }
     else
-        throw OpenMMException("Illegal value for OpenCLPrecision: "+precision);
+        throw OpenMMException("Illegal value for Precision: "+precision);
     try {
         contextIndex = platformData.contexts.size();
         std::vector<cl::Platform> platforms;
@@ -105,7 +105,7 @@ OpenCLContext::OpenCLContext(const System& system, int platformIndex, int device
             vector<cl::Device> devices;
             platforms[j].getDevices(CL_DEVICE_TYPE_ALL, &devices);
             if (deviceIndex < -1 || deviceIndex >= (int) devices.size())
-                throw OpenMMException("Illegal value for OpenCLDeviceIndex: "+intToString(deviceIndex));
+                throw OpenMMException("Illegal value for DeviceIndex: "+intToString(deviceIndex));
 
             for (int i = 0; i < (int) devices.size(); i++) {
                 // If they supplied a valid deviceIndex, we only look through that one
@@ -113,6 +113,11 @@ OpenCLContext::OpenCLContext(const System& system, int platformIndex, int device
                     continue;
                 if (platformVendor == "Apple" && (devices[i].getInfo<CL_DEVICE_TYPE>() == CL_DEVICE_TYPE_CPU))
                     continue; // The CPU device on OS X won't work correctly.
+                if (useMixedPrecision || useDoublePrecision) {
+                    bool supportsDouble = (devices[i].getInfo<CL_DEVICE_EXTENSIONS>().find("cl_khr_fp64") != string::npos);
+                    if (!supportsDouble)
+                        continue; // This device does not support double precision.
+                }
                 int maxSize = devices[i].getInfo<CL_DEVICE_MAX_WORK_ITEM_SIZES>()[0];
                 int processingElementsPerComputeUnit = 8;
                 if (devices[i].getInfo<CL_DEVICE_TYPE>() != CL_DEVICE_TYPE_GPU) {

platforms/opencl/src/OpenCLExpressionUtilities.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) 2009-2015 Stanford University and the Authors.      *
+ * Portions copyright (c) 2009-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -254,7 +254,7 @@ void OpenCLExpressionUtilities::processExpression(stringstream& out, const Expre
                             for (int k = 3; k >= 0; k--)
                                 for (int m = 0; m < 4; m++) {
                                     int base = 4*m;
-                                    string suffix = suffixes[m];
+                                    string suffix = suffixes[k];
                                     out << "derivy[" << m << "] = da*derivy[" << m << "] + (3*c[" << (base+3) << "]" << suffix << "*db + 2*c[" << (base+2) << "]" << suffix << ")*db + c[" << (base+1) << "]" << suffix << ";\n";
                                 }
                             out << nodeNames[j] << " = derivy[0] + dc*(derivy[1] + dc*(derivy[2] + dc*derivy[3]));\n";
@@ -271,7 +271,7 @@ void OpenCLExpressionUtilities::processExpression(stringstream& out, const Expre
                             out << nodeNames[j] << " *= " << paramsFloat[11] << ";\n";
                         }
                         else
-                            throw OpenMMException("Unsupported derivative order for Continuous2DFunction");
+                            throw OpenMMException("Unsupported derivative order for Continuous3DFunction");
                     }
                     out << "}\n";
                 }

platforms/opencl/src/OpenCLPlatform.cpp

@@ -56,6 +56,11 @@ extern "C" OPENMM_EXPORT_OPENCL void registerPlatforms() {
 #endif
 
 OpenCLPlatform::OpenCLPlatform() {
+    deprecatedPropertyReplacements["OpenCLDeviceIndex"] = OpenCLDeviceIndex();
+    deprecatedPropertyReplacements["OpenCLDeviceName"] = OpenCLDeviceName();
+    deprecatedPropertyReplacements["OpenCLPrecision"] = OpenCLPrecision();
+    deprecatedPropertyReplacements["OpenCLUseCpuPme"] = OpenCLUseCpuPme();
+    deprecatedPropertyReplacements["OpenCLDisablePmeStream"] = OpenCLDisablePmeStream();
     OpenCLKernelFactory* factory = new OpenCLKernelFactory();
     registerKernelFactory(CalcForcesAndEnergyKernel::Name(), factory);
     registerKernelFactory(UpdateStateDataKernel::Name(), factory);
@@ -139,7 +144,10 @@ bool OpenCLPlatform::isPlatformSupported() {
 const string& OpenCLPlatform::getPropertyValue(const Context& context, const string& property) const {
     const ContextImpl& impl = getContextImpl(context);
     const PlatformData* data = reinterpret_cast<const PlatformData*>(impl.getPlatformData());
-    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 Platform::getPropertyValue(context, property);

platforms/opencl/src/kernels/angleForce.cl

 real4 v0 = pos2-pos1;
 real4 v1 = pos2-pos3;
+#if APPLY_PERIODIC
+APPLY_PERIODIC_TO_DELTA(v0)
+APPLY_PERIODIC_TO_DELTA(v1)
+#endif
 real4 cp = cross(v0, v1);
 real rp = cp.x*cp.x + cp.y*cp.y + cp.z*cp.z;
 rp = max(SQRT(rp), (real) 1.0e-06f);

platforms/opencl/src/kernels/bondForce.cl

 real4 delta = pos2-pos1;
+#if APPLY_PERIODIC
+APPLY_PERIODIC_TO_DELTA(delta)
+#endif
 real r = SQRT(delta.x*delta.x + delta.y*delta.y + delta.z*delta.z);
 COMPUTE_FORCE
 dEdR = (r > 0.0f) ? (dEdR / r) : 0.0f;

platforms/opencl/src/kernels/cmapTorsionForce.cl

@@ -5,6 +5,11 @@ const real PI = 3.14159265358979323846f;
 real4 v0a = (real4) (pos1.xyz-pos2.xyz, 0.0f);
 real4 v1a = (real4) (pos3.xyz-pos2.xyz, 0.0f);
 real4 v2a = (real4) (pos3.xyz-pos4.xyz, 0.0f);
+#if APPLY_PERIODIC
+APPLY_PERIODIC_TO_DELTA(v0a)
+APPLY_PERIODIC_TO_DELTA(v1a)
+APPLY_PERIODIC_TO_DELTA(v2a)
+#endif
 real4 cp0a = cross(v0a, v1a);
 real4 cp1a = cross(v1a, v2a);
 real cosangle = dot(normalize(cp0a), normalize(cp1a));
@@ -28,6 +33,11 @@ angleA = fmod(angleA+2.0f*PI, 2.0f*PI);
 real4 v0b = (real4) (pos5.xyz-pos6.xyz, 0.0f);
 real4 v1b = (real4) (pos7.xyz-pos6.xyz, 0.0f);
 real4 v2b = (real4) (pos7.xyz-pos8.xyz, 0.0f);
+#if APPLY_PERIODIC
+APPLY_PERIODIC_TO_DELTA(v0b)
+APPLY_PERIODIC_TO_DELTA(v1b)
+APPLY_PERIODIC_TO_DELTA(v2b)
+#endif
 real4 cp0b = cross(v0b, v1b);
 real4 cp1b = cross(v1b, v2b);
 cosangle = dot(normalize(cp0b), normalize(cp1b));

platforms/opencl/src/kernels/customCentroidBond.cl

@@ -70,8 +70,11 @@ __kernel void computeGroupCenters(__global const real4* restrict posq, __global
 /**
  * Compute the difference between two vectors, setting the fourth component to the squared magnitude.
  */
-real4 delta(real4 vec1, real4 vec2) {
+real4 delta(real4 vec1, real4 vec2, bool periodic, real4 periodicBoxSize, real4 invPeriodicBoxSize, 
+        real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ) {
     real4 result = (real4) (vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0);
+    if (periodic)
+        APPLY_PERIODIC_TO_DELTA(result);
     result.w = result.x*result.x + result.y*result.y + result.z*result.z;
     return result;
 }
@@ -110,7 +113,7 @@ real4 computeCross(real4 vec1, real4 vec2) {
  * Compute the forces on groups based on the bonds.
  */
 __kernel void computeGroupForces(__global long* restrict groupForce, __global mixed* restrict energyBuffer, __global const real4* restrict centerPositions,
-        __global const int* restrict bondGroups
+        __global const int* restrict bondGroups, real4 periodicBoxSize, real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ
         EXTRA_ARGS) {
     mixed energy = 0;
     for (int index = get_global_id(0); index < NUM_BONDS; index += get_global_size(0)) {

platforms/opencl/src/kernels/customCompoundBond.cl

 /**
  * Compute the difference between two vectors, setting the fourth component to the squared magnitude.
  */
-real4 ccb_delta(real4 vec1, real4 vec2) {
+real4 ccb_delta(real4 vec1, real4 vec2, bool periodic, real4 periodicBoxSize, real4 invPeriodicBoxSize, 
+        real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ) {
     real4 result = (real4) (vec1.x-vec2.x, vec1.y-vec2.y, vec1.z-vec2.z, 0);
+    if (periodic)
+        APPLY_PERIODIC_TO_DELTA(result);
     result.w = result.x*result.x + result.y*result.y + result.z*result.z;
     return result;
 }

platforms/opencl/src/kernels/customManyParticle.cl

@@ -55,7 +55,7 @@ inline real4 computeCross(real4 vec1, real4 vec2) {
 /**
  * Determine whether a particular interaction is in the list of exclusions.
  */
-inline bool isInteractionExcluded(int atom1, int atom2, __global int* restrict exclusions, __global int* restrict exclusionStartIndex) {
+inline bool isInteractionExcluded(int atom1, int atom2, __global const int* restrict exclusions, __global const int* restrict exclusionStartIndex) {
     int first = exclusionStartIndex[atom1];
     int last = exclusionStartIndex[atom1+1];
     for (int i = last-1; i >= first; i--) {
@@ -174,7 +174,7 @@ __kernel void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, rea
         __global const real4* restrict posq, __global const real4* restrict blockCenter, __global const real4* restrict blockBoundingBox, __global int2* restrict neighborPairs,
         __global int* restrict numNeighborPairs, __global int* restrict numNeighborsForAtom, int maxNeighborPairs
 #ifdef USE_EXCLUSIONS
-        , __global int* restrict exclusions, __global int* restrict exclusionStartIndex
+        , __global const int* restrict exclusions, __global const int* restrict exclusionStartIndex
 #endif
         ) {
     __local real4 positionCache[FIND_NEIGHBORS_WORKGROUP_SIZE];
@@ -264,7 +264,9 @@ __kernel void findNeighbors(real4 periodicBoxSize, real4 invPeriodicBoxSize, rea
                 }
             }
         }
+        if (atom1 < NUM_ATOMS)
             numNeighborsForAtom[atom1] = totalNeighborsForAtom1;
+        SYNC_WARPS;
     }
 }
 
@@ -307,6 +309,7 @@ __kernel void computeNeighborStartIndices(__global int* restrict numNeighborsFor
             numNeighborsForAtom[globalIndex] = 0; // Clear this so the next kernel can use it as a counter
         }
         globalOffset += posBuffer[get_local_size(0)-1];
+        barrier(CLK_LOCAL_MEM_FENCE);
     }
     if (get_local_id(0) == 0)
         neighborStartIndex[0] = 0;

platforms/opencl/src/kernels/torsionForce.cl

@@ -2,6 +2,11 @@ const real PI = 3.14159265358979323846f;
 real4 v0 = (real4) (pos1.xyz-pos2.xyz, 0.0f);
 real4 v1 = (real4) (pos3.xyz-pos2.xyz, 0.0f);
 real4 v2 = (real4) (pos3.xyz-pos4.xyz, 0.0f);
+#if APPLY_PERIODIC
+APPLY_PERIODIC_TO_DELTA(v0)
+APPLY_PERIODIC_TO_DELTA(v1)
+APPLY_PERIODIC_TO_DELTA(v2)
+#endif
 real4 cp0 = cross(v0, v1);
 real4 cp1 = cross(v1, v2);
 real cosangle = dot(normalize(cp0), normalize(cp1));

platforms/opencl/tests/OpenCLTests.h

@@ -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) 2015 Stanford University and the Authors.           *
+ * Portions copyright (c) 2015-2016 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -39,9 +39,9 @@ OpenMM::OpenCLPlatform platform;
 
 void initializeTests(int argc, char* argv[]) {
     if (argc > 1)
-        platform.setPropertyDefaultValue("OpenCLPrecision", std::string(argv[1]));
+        platform.setPropertyDefaultValue("Precision", std::string(argv[1]));
     if (argc > 2)
         platform.setPropertyDefaultValue("OpenCLPlatformIndex", std::string(argv[2]));
     if (argc > 3)
-        platform.setPropertyDefaultValue("OpenCLDeviceIndex", std::string(argv[3]));
+        platform.setPropertyDefaultValue("DeviceIndex", std::string(argv[3]));
 }

platforms/reference/include/ReferenceAngleBondIxn.h

@@ -33,6 +33,9 @@ class OPENMM_EXPORT ReferenceAngleBondIxn : public ReferenceBondIxn {
 
    private:
 
+        bool usePeriodic;
+        RealVec boxVectors[3];
+
    public:
 
       /**---------------------------------------------------------------------------------------
@@ -51,6 +54,16 @@ class OPENMM_EXPORT ReferenceAngleBondIxn : public ReferenceBondIxn {
 
        ~ReferenceAngleBondIxn();
 
+       /**---------------------------------------------------------------------------------------
+      
+         Set the force to use periodic boundary conditions.
+      
+         @param vectors    the vectors defining the periodic box
+      
+         --------------------------------------------------------------------------------------- */
+      
+      void setPeriodic(OpenMM::RealVec* vectors);
+
       /**---------------------------------------------------------------------------------------
 
          Get dEdR and energy term for angle bond

platforms/reference/include/ReferenceCMAPTorsionIxn.h

@@ -38,6 +38,8 @@ private:
     std::vector<std::vector<std::vector<RealOpenMM> > > coeff;
     std::vector<int> torsionMaps;
     std::vector<std::vector<int> > torsionIndices;
+    bool usePeriodic;
+    RealVec boxVectors[3];
 
     /**---------------------------------------------------------------------------------------
 
@@ -65,6 +67,16 @@ public:
                             const std::vector<int>& torsionMaps,
                             const std::vector<std::vector<int> >& torsionIndices);
 
+       /**---------------------------------------------------------------------------------------
+      
+         Set the force to use periodic boundary conditions.
+      
+         @param vectors    the vectors defining the periodic box
+      
+         --------------------------------------------------------------------------------------- */
+      
+       void setPeriodic(OpenMM::RealVec* vectors);
+
     /**---------------------------------------------------------------------------------------
 
        Calculate torsion interaction