Assorted optimizations

platforms/cpu/include/CpuNonbondedForce.h

@@ -176,9 +176,10 @@ private:
         pthread_cond_t startCondition, endCondition;
         pthread_mutex_t lock;
         // The following variables are used to make information accessible to the individual threads.
+        int numberOfAtoms;
         float* posq;
-        std::vector<std::pair<float, float> > atomParameters;        
-        std::vector<std::set<int> > exclusions;
+        std::pair<float, float> const* atomParameters;        
+        std::set<int> const* exclusions;
         bool includeEnergy;
 
         static const float TWO_OVER_SQRT_PI;

platforms/cpu/src/CpuNeighborList.cpp

@@ -117,13 +117,9 @@ public:
                     for (Voxel::const_iterator itemIter = voxel.begin(); itemIter != voxel.end(); ++itemIter) {
                         const int atomJ = itemIter->second;
 
-                        // Ignore self hits
-                        if (atomI >= atomJ)
-                            continue;
-                        
-                        // Ignore exclusions.
-                        if (exclusions[atomI].find(atomJ) != exclusions[atomI].end())
-                            continue;
+                        // Avoid duplicate entries.
+                        if (atomJ >= atomI)
+                            break;
                         
                         __m128 posJ = _mm_loadu_ps(itemIter->first);
                         __m128 delta = _mm_sub_ps(posJ, posI);
@@ -134,6 +130,11 @@ public:
                         float dSquared = _mm_cvtss_f32(_mm_dp_ps(delta, delta, 0x71));
                         if (dSquared > maxDistanceSquared)
                             continue;
+                        
+                        // Ignore exclusions.
+                        if (exclusions[atomI].find(atomJ) != exclusions[atomI].end())
+                            continue;
+
                         neighbors.push_back(make_pair(atomI, atomJ));
                     }
                 }

platforms/cpu/src/CpuNonbondedForce.cpp

@@ -340,9 +340,10 @@ void CpuNonbondedForce::calculateDirectIxn(int numberOfAtoms, float* posq, const
                 const vector<set<int> >& exclusions, float* forces, float* totalEnergy) {
     // Record the parameters for the threads.
     
+    this->numberOfAtoms = numberOfAtoms;
     this->posq = posq;
-    this->atomParameters = atomParameters;
-    this->exclusions = exclusions;
+    this->atomParameters = &atomParameters[0];
+    this->exclusions = &exclusions[0];
     includeEnergy = (totalEnergy != NULL);
     
     // Signal the threads to start running and wait for them to finish.
@@ -413,7 +414,6 @@ void CpuNonbondedForce::runThread(int index, vector<float>& threadForce, double&
         
         threadEnergy = 0;
         double* energyPtr = (includeEnergy ? &threadEnergy : NULL);
-        int numberOfAtoms = atomParameters.size();
         threadForce.resize(4*numberOfAtoms, 0.0f);
         for (int i = 0; i < 4*numberOfAtoms; i++)
             threadForce[i] = 0.0f;