Merge pull request #1057 from peastman/optimize

Very minor optimizations

platforms/cuda/include/CudaBondedUtilities.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) 2011-2012 Stanford University and the Authors.      *
+ * Portions copyright (c) 2011-2015 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -130,7 +130,7 @@ private:
     std::vector<std::vector<CudaArray*> > atomIndices;
     std::vector<std::string> prefixCode;
     std::vector<void*> kernelArgs;
-    int numForceBuffers, maxBonds;
+    int numForceBuffers, maxBonds, allGroups;
     bool hasInitializedKernels, hasInteractions;
 };
 

platforms/cuda/include/CudaKernels.h

@@ -1294,7 +1294,7 @@ private:
     double prevStepSize, energy;
     float energyFloat;
     int numGlobalVariables;
-    bool hasInitializedKernels, deviceValuesAreCurrent, deviceGlobalsAreCurrent, modifiesParameters, keNeedsForce;
+    bool hasInitializedKernels, deviceValuesAreCurrent, deviceGlobalsAreCurrent, modifiesParameters, keNeedsForce, hasAnyConstraints;
     mutable bool localValuesAreCurrent;
     CudaArray* globalValues;
     CudaArray* sumBuffer;

platforms/cuda/src/CudaBondedUtilities.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) 2011-2012 Stanford University and the Authors.      *
+ * Portions copyright (c) 2011-2015 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -34,7 +34,7 @@
 using namespace OpenMM;
 using namespace std;
 
-CudaBondedUtilities::CudaBondedUtilities(CudaContext& context) : context(context), numForceBuffers(0), maxBonds(0), hasInitializedKernels(false) {
+CudaBondedUtilities::CudaBondedUtilities(CudaContext& context) : context(context), numForceBuffers(0), maxBonds(0), allGroups(0), hasInitializedKernels(false) {
 }
 
 CudaBondedUtilities::~CudaBondedUtilities() {
@@ -48,6 +48,7 @@ void CudaBondedUtilities::addInteraction(const vector<vector<int> >& atoms, cons
         forceAtoms.push_back(atoms);
         forceSource.push_back(source);
         forceGroup.push_back(group);
+        allGroups |= 1<<group;
     }
 }
 
@@ -152,6 +153,8 @@ string CudaBondedUtilities::createForceSource(int forceIndex, int numBonds, int
 }
 
 void CudaBondedUtilities::computeInteractions(int groups) {
+    if ((groups&allGroups) == 0)
+        return;
     if (!hasInitializedKernels) {
         hasInitializedKernels = true;
         kernelArgs.push_back(&context.getForce().getDevicePointer());

platforms/cuda/src/CudaKernels.cpp

@@ -5877,8 +5877,10 @@ void CudaIntegrateCustomStepKernel::prepareForComputation(ContextImpl& context,
         
         // Determine how each step will represent the position (as just a value, or a value plus a delta).
         
+        hasAnyConstraints = (context.getSystem().getNumConstraints() > 0);
         vector<bool> storePosAsDelta(numSteps, false);
         vector<bool> loadPosAsDelta(numSteps, false);
+        if (hasAnyConstraints) {
             bool beforeConstrain = false;
             for (int step = numSteps-1; step >= 0; step--) {
                 if (stepType[step] == CustomIntegrator::ConstrainPositions)
@@ -5894,6 +5896,7 @@ void CudaIntegrateCustomStepKernel::prepareForComputation(ContextImpl& context,
                 if (stepType[step] == CustomIntegrator::ConstrainPositions)
                     storedAsDelta = false;
             }
+        }
         
         // Identify steps that can be merged into a single kernel.
         
@@ -6214,9 +6217,11 @@ void CudaIntegrateCustomStepKernel::execute(ContextImpl& context, CustomIntegrat
             context.updateContextState();
         }
         else if (stepType[step] == CustomIntegrator::ConstrainPositions) {
+            if (hasAnyConstraints) {
                 cu.getIntegrationUtilities().applyConstraints(integrator.getConstraintTolerance());
                 kernelArgs[step][0][1] = &posCorrection;
                 cu.executeKernel(kernels[step][0], &kernelArgs[step][0][0], numAtoms);
+            }
             cu.getIntegrationUtilities().computeVirtualSites();
         }
         else if (stepType[step] == CustomIntegrator::ConstrainVelocities) {

platforms/opencl/include/OpenCLBondedUtilities.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) 2011 Stanford University and the Authors.           *
+ * Portions copyright (c) 2011-2015 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -137,7 +137,7 @@ private:
     std::vector<OpenCLArray*> atomIndices;
     std::vector<OpenCLArray*> bufferIndices;
     std::vector<std::string> prefixCode;
-    int numForceBuffers, maxBonds;
+    int numForceBuffers, maxBonds, allGroups;
     bool hasInitializedKernels;
 };
 

platforms/opencl/include/OpenCLKernels.h

@@ -1283,7 +1283,7 @@ private:
     double prevStepSize, energy;
     float energyFloat;
     int numGlobalVariables;
-    bool hasInitializedKernels, deviceValuesAreCurrent, deviceGlobalsAreCurrent, modifiesParameters, keNeedsForce;
+    bool hasInitializedKernels, deviceValuesAreCurrent, deviceGlobalsAreCurrent, modifiesParameters, keNeedsForce, hasAnyConstraints;
     mutable bool localValuesAreCurrent;
     OpenCLArray* globalValues;
     OpenCLArray* sumBuffer;

platforms/opencl/src/OpenCLBondedUtilities.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) 2011-2012 Stanford University and the Authors.      *
+ * Portions copyright (c) 2011-2015 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -33,7 +33,7 @@
 using namespace OpenMM;
 using namespace std;
 
-OpenCLBondedUtilities::OpenCLBondedUtilities(OpenCLContext& context) : context(context), numForceBuffers(0), maxBonds(0), hasInitializedKernels(false) {
+OpenCLBondedUtilities::OpenCLBondedUtilities(OpenCLContext& context) : context(context), numForceBuffers(0), maxBonds(0), allGroups(0), hasInitializedKernels(false) {
 }
 
 OpenCLBondedUtilities::~OpenCLBondedUtilities() {
@@ -48,6 +48,7 @@ void OpenCLBondedUtilities::addInteraction(const vector<vector<int> >& atoms, co
         forceAtoms.push_back(atoms);
         forceSource.push_back(source);
         forceGroup.push_back(group);
+        allGroups |= 1<<group;
         int width = 1;
         while (width < (int) atoms[0].size())
             width *= 2;
@@ -73,7 +74,7 @@ void OpenCLBondedUtilities::initialize(const System& system) {
     if (numForces == 0)
         return;
     
-    // Build the lists of atom indicse and buffer indices.
+    // Build the lists of atom indices and buffer indices.
     
     vector<vector<cl_uint> > bufferVec(numForces);
     vector<vector<int> > bufferCounter(numForces, vector<int>(system.getNumParticles(), 0));
@@ -253,6 +254,8 @@ string OpenCLBondedUtilities::createForceSource(int forceIndex, int numBonds, in
 }
 
 void OpenCLBondedUtilities::computeInteractions(int groups) {
+    if ((groups&allGroups) == 0)
+        return;
     if (!hasInitializedKernels) {
         hasInitializedKernels = true;
         for (int i = 0; i < (int) forceSets.size(); i++) {

platforms/opencl/src/OpenCLKernels.cpp

@@ -6141,8 +6141,10 @@ void OpenCLIntegrateCustomStepKernel::prepareForComputation(ContextImpl& context
         
         // Determine how each step will represent the position (as just a value, or a value plus a delta).
         
+        hasAnyConstraints = (context.getSystem().getNumConstraints() > 0);
         vector<bool> storePosAsDelta(numSteps, false);
         vector<bool> loadPosAsDelta(numSteps, false);
+        if (hasAnyConstraints) {
             bool beforeConstrain = false;
             for (int step = numSteps-1; step >= 0; step--) {
                 if (stepType[step] == CustomIntegrator::ConstrainPositions)
@@ -6158,6 +6160,7 @@ void OpenCLIntegrateCustomStepKernel::prepareForComputation(ContextImpl& context
                 if (stepType[step] == CustomIntegrator::ConstrainPositions)
                     storedAsDelta = false;
             }
+        }
         
         // Identify steps that can be merged into a single kernel.
         
@@ -6478,8 +6481,10 @@ void OpenCLIntegrateCustomStepKernel::execute(ContextImpl& context, CustomIntegr
             context.updateContextState();
         }
         else if (stepType[step] == CustomIntegrator::ConstrainPositions) {
+            if (hasAnyConstraints) {
                 cl.getIntegrationUtilities().applyConstraints(integrator.getConstraintTolerance());
                 cl.executeKernel(kernels[step][0], numAtoms);
+            }
             cl.getIntegrationUtilities().computeVirtualSites();
         }
         else if (stepType[step] == CustomIntegrator::ConstrainVelocities) {