Optimizations to PME

plugins/amoeba/platforms/cuda/src/kernels/kCalculateAmoebaCudaPME.cu

@@ -1018,18 +1018,7 @@ void kCalculateAmoebaPMEInducedDipoleField(amoebaGpuContext amoebaGpu)
  */
 void kCalculateAmoebaPMEInducedDipoleForces(amoebaGpuContext amoebaGpu)
 {
-    // Perform PME for the induced dipoles.
-
     gpuContext gpu = amoebaGpu->gpuContext;
-    kGridSpreadInducedDipoles_kernel<<<10*gpu->sim.blocks, 64>>>();
-    LAUNCHERROR("kGridSpreadInducedDipoles");
-    cufftExecC2C(gpu->fftplan, gpu->psPmeGrid->_pDevData, gpu->psPmeGrid->_pDevData, CUFFT_FORWARD);
-    kAmoebaReciprocalConvolution_kernel<<<gpu->sim.blocks, gpu->sim.nonbond_threads_per_block>>>();
-    LAUNCHERROR("kAmoebaReciprocalConvolution");
-    cufftExecC2C(gpu->fftplan, gpu->psPmeGrid->_pDevData, gpu->psPmeGrid->_pDevData, CUFFT_INVERSE);
-    int potentialThreads = (gpu->sm_version >= SM_20 ? 256 : (gpu->sm_version >= SM_12 ? 128 : 64));
-    kComputeInducedPotentialFromGrid_kernel<<<gpu->sim.blocks, potentialThreads>>>();
-    LAUNCHERROR("kComputeInducedPotentialFromGrid");
     kComputeInducedDipoleForceAndEnergy_kernel<<<gpu->sim.blocks, gpu->sim.update_threads_per_block>>>();
     LAUNCHERROR("kComputeInducedDipoleForceAndEnergy");
     cudaComputeAmoebaMapTorquesAndAddTotalForce2(amoebaGpu, amoebaGpu->psTorque, gpu->psForce4);

plugins/amoeba/platforms/cuda/src/kernels/kCalculateAmoebaCudaPmeDirectElectrostatic.h

@@ -298,7 +298,8 @@ if( atomI == targetAtom ){
        
                 for (unsigned int j = 0; j < GRID; j++)
                 {
-           
+                    if ((flags&(1<<j)) != 0)
+                    {
                         unsigned int jIdx  = (flags == 0xFFFFFFFF) ? tj : j;
                         unsigned int atomJ = y + jIdx;
 
@@ -386,6 +387,7 @@ if( atomI == targetAtom ){
                                 psA[jIdx].torque[2] += sA[threadIdx.x].tempTorque[2] + sA[threadIdx.x+16].tempTorque[2];
                             }
                         }
+                    }
  
                     tj = (tj + 1) & (GRID - 1);
 

plugins/amoeba/platforms/cuda/src/kernels/kCalculateAmoebaCudaPmeFixedEField.h

@@ -228,6 +228,7 @@ if( atomI == targetAtom || targetAtom == (y+j) ){
 
                 for (unsigned int j = 0; j < GRID; j++){
 
+                    if ((flags&(1<<j)) != 0) {
                         unsigned int jIdx = (flags == 0xFFFFFFFF) ? tj : j;
                         if( bExclusionFlag ){
                             getMaskedDScaleFactor( jIdx, dScaleMask, &dScaleValue );
@@ -236,9 +237,9 @@ if( atomI == targetAtom || targetAtom == (y+j) ){
 
                         float4 ijField[3];
                         calculateFixedFieldRealSpacePairIxn_kernel( localParticle, psA[jIdx], dScaleValue, pScaleValue, ijField
-#ifdef AMOEBA_DEBUG
+    #ifdef AMOEBA_DEBUG
                                                     , pullBack
-#endif
+    #endif
                         );
 
                         unsigned int outOfBounds     = ( (atomI >= cAmoebaSim.numberOfAtoms) || ((y+jIdx) >= cAmoebaSim.numberOfAtoms) ) ? 1 : 0;
@@ -358,6 +359,7 @@ if( (atomI == targetAtom || (y + jIdx) == targetAtom) ){
             }
 }        
 #endif
+                    }
                     tj                  = (tj + 1) & (GRID - 1);
 
                 } // j-loop block

plugins/amoeba/platforms/cuda/src/kernels/kCalculateAmoebaCudaPmeMutualInducedField.h

@@ -231,16 +231,17 @@ if( atomI == targetAtom || (y+j) == targetAtom ){
     
                 for (unsigned int j = 0; j < GRID; j++)
                 {
-    
+                    if ((flags&(1<<j)) != 0)
+                    {
                         unsigned int jIdx = (flags == 0xFFFFFFFF) ? tj : j;
                         float4 ijField[3];
 
                         // load coords, charge, ...
 
                         calculatePmeDirectMutualInducedFieldPairIxn_kernel( localParticle, psA[jIdx], uscale, ijField
-#ifdef AMOEBA_DEBUG
+    #ifdef AMOEBA_DEBUG
         , pullBack
-#endif
+    #endif
            );
 
                         unsigned int mask   =  ( (atomI >= cAmoebaSim.numberOfAtoms) || ((y+jIdx) >= cAmoebaSim.numberOfAtoms) ) ? 0 : 1;
@@ -359,6 +360,7 @@ if( atomI == targetAtom || (y+jIdx) == targetAtom ){
             debugArray[index].w                = flag;
 }
 #endif
+                    }
     
                     tj                  = (tj + 1) & (GRID - 1);