Continuing CUDA implementation of triclinic boxes for AmoebaMultipoleForce

plugins/amoeba/platforms/cuda/src/AmoebaCudaKernels.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) 2008-2013 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2015 Stanford University and the Authors.      *
  * Authors: Peter Eastman, Mark Friedrichs                                    *
  * Contributors:                                                              *
  *                                                                            *
@@ -801,7 +801,7 @@ CudaCalcAmoebaMultipoleForceKernel::CudaCalcAmoebaMultipoleForceKernel(std::stri
         diisCoefficients(NULL), inducedDipolePolar(NULL), inducedDipoleErrors(NULL), prevDipoles(NULL), prevDipolesPolar(NULL), prevDipolesGk(NULL),
         prevDipolesGkPolar(NULL), prevErrors(NULL), diisMatrix(NULL), polarizability(NULL), covalentFlags(NULL), polarizationGroupFlags(NULL),
         pmeGrid(NULL), pmeBsplineModuliX(NULL), pmeBsplineModuliY(NULL), pmeBsplineModuliZ(NULL), pmeIgrid(NULL), pmePhi(NULL),
-        pmePhid(NULL), pmePhip(NULL), pmePhidp(NULL), pmeAtomGridIndex(NULL), lastPositions(NULL), sort(NULL), gkKernel(NULL) {
+        pmePhid(NULL), pmePhip(NULL), pmePhidp(NULL), pmeCphi(NULL), pmeAtomGridIndex(NULL), lastPositions(NULL), sort(NULL), gkKernel(NULL) {
 }
 
 CudaCalcAmoebaMultipoleForceKernel::~CudaCalcAmoebaMultipoleForceKernel() {
@@ -876,6 +876,8 @@ CudaCalcAmoebaMultipoleForceKernel::~CudaCalcAmoebaMultipoleForceKernel() {
         delete pmePhip;
     if (pmePhidp != NULL)
         delete pmePhidp;
+    if (pmeCphi != NULL)
+        delete pmeCphi;
     if (pmeAtomGridIndex != NULL)
         delete pmeAtomGridIndex;
     if (lastPositions != NULL)
@@ -1192,6 +1194,7 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const
         CUmodule module = cu.createModule(CudaKernelSources::vectorOps+CudaAmoebaKernelSources::multipolePme, pmeDefines);
         pmeGridIndexKernel = cu.getKernel(module, "findAtomGridIndex");
         pmeTransformMultipolesKernel = cu.getKernel(module, "transformMultipolesToFractionalCoordinates");
+        pmeTransformPotentialKernel = cu.getKernel(module, "transformPotentialToCartesianCoordinates");
         pmeSpreadFixedMultipolesKernel = cu.getKernel(module, "gridSpreadFixedMultipoles");
         pmeSpreadInducedDipolesKernel = cu.getKernel(module, "gridSpreadInducedDipoles");
         pmeFinishSpreadChargeKernel = cu.getKernel(module, "finishSpreadCharge");
@@ -1219,6 +1222,7 @@ void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const
         pmePhid = new CudaArray(cu, 10*numMultipoles, elementSize, "pmePhid");
         pmePhip = new CudaArray(cu, 10*numMultipoles, elementSize, "pmePhip");
         pmePhidp = new CudaArray(cu, 20*numMultipoles, elementSize, "pmePhidp");
+        pmeCphi = new CudaArray(cu, 10*numMultipoles, elementSize, "pmeCphi");
         pmeAtomRange = CudaArray::create<int>(cu, gridSizeX*gridSizeY*gridSizeZ+1, "pmeAtomRange");
         pmeAtomGridIndex = CudaArray::create<int2>(cu, numMultipoles, "pmeAtomGridIndex");
         sort = new CudaSort(cu, new SortTrait(), cu.getNumAtoms());
@@ -1520,14 +1524,16 @@ double CudaCalcAmoebaMultipoleForceKernel::execute(ContextImpl& context, bool in
         
         unsigned int maxTiles = nb.getInteractingTiles().getSize();
         void* gridIndexArgs[] = {&cu.getPosq().getDevicePointer(), &pmeAtomGridIndex->getDevicePointer(),
-            cu.getPeriodicBoxSizePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
+            cu.getPeriodicBoxVecXPointer(), cu.getPeriodicBoxVecYPointer(), cu.getPeriodicBoxVecZPointer(),
+            recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
         cu.executeKernel(pmeGridIndexKernel, gridIndexArgs, cu.getNumAtoms(), cu.ThreadBlockSize, cu.ThreadBlockSize*PmeOrder*PmeOrder*elementSize);
         sort->sort(*pmeAtomGridIndex);
         void* pmeTransformMultipolesArgs[] = {&labFrameDipoles->getDevicePointer(), &labFrameQuadrupoles->getDevicePointer(),
             &fracDipoles->getDevicePointer(), &fracQuadrupoles->getDevicePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
         cu.executeKernel(pmeTransformMultipolesKernel, pmeTransformMultipolesArgs, cu.getNumAtoms());
         void* pmeSpreadFixedMultipolesArgs[] = {&cu.getPosq().getDevicePointer(), &fracDipoles->getDevicePointer(), &fracQuadrupoles->getDevicePointer(),
-            &pmeGrid->getDevicePointer(), &pmeAtomGridIndex->getDevicePointer(),  cu.getPeriodicBoxSizePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
+            &pmeGrid->getDevicePointer(), &pmeAtomGridIndex->getDevicePointer(),  cu.getPeriodicBoxVecXPointer(), cu.getPeriodicBoxVecYPointer(), cu.getPeriodicBoxVecZPointer(),
+            recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
         cu.executeKernel(pmeSpreadFixedMultipolesKernel, pmeSpreadFixedMultipolesArgs, cu.getNumAtoms());
         void* finishSpreadArgs[] = {&pmeGrid->getDevicePointer()};
         if (cu.getUseDoublePrecision())
@@ -1547,9 +1553,11 @@ double CudaCalcAmoebaMultipoleForceKernel::execute(ContextImpl& context, bool in
             &fieldPolar ->getDevicePointer(), &cu.getPosq().getDevicePointer(), &labFrameDipoles->getDevicePointer(),
             cu.getPeriodicBoxSizePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2], &pmeAtomGridIndex->getDevicePointer()};
         cu.executeKernel(pmeFixedPotentialKernel, pmeFixedPotentialArgs, cu.getNumAtoms());
+        void* pmeTransformFixedPotentialArgs[] = {&pmePhi->getDevicePointer(), &pmeCphi->getDevicePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
+        cu.executeKernel(pmeTransformPotentialKernel, pmeTransformFixedPotentialArgs, cu.getNumAtoms());
         void* pmeFixedForceArgs[] = {&cu.getPosq().getDevicePointer(), &cu.getForce().getDevicePointer(), &torque->getDevicePointer(),
             &cu.getEnergyBuffer().getDevicePointer(), &labFrameDipoles->getDevicePointer(), &labFrameQuadrupoles->getDevicePointer(),
-            &fracDipoles->getDevicePointer(), &fracQuadrupoles->getDevicePointer(), &pmePhi->getDevicePointer(),
+            &fracDipoles->getDevicePointer(), &fracQuadrupoles->getDevicePointer(), &pmePhi->getDevicePointer(), &pmeCphi->getDevicePointer(),
             recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
         cu.executeKernel(pmeFixedForceKernel, pmeFixedForceArgs, cu.getNumAtoms());
         
@@ -1570,7 +1578,8 @@ double CudaCalcAmoebaMultipoleForceKernel::execute(ContextImpl& context, bool in
 
         cu.clearBuffer(*pmeGrid);
         void* pmeSpreadInducedDipolesArgs[] = {&cu.getPosq().getDevicePointer(), &inducedDipole->getDevicePointer(), &inducedDipolePolar->getDevicePointer(),
-            &pmeGrid->getDevicePointer(), &pmeAtomGridIndex->getDevicePointer(), cu.getPeriodicBoxSizePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
+            &pmeGrid->getDevicePointer(), &pmeAtomGridIndex->getDevicePointer(), cu.getPeriodicBoxVecXPointer(), cu.getPeriodicBoxVecYPointer(), cu.getPeriodicBoxVecZPointer(),
+            recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
         cu.executeKernel(pmeSpreadInducedDipolesKernel, pmeSpreadInducedDipolesArgs, cu.getNumAtoms());
         if (cu.getUseDoublePrecision())
             cu.executeKernel(pmeFinishSpreadChargeKernel, finishSpreadArgs, pmeGrid->getSize());
@@ -1634,11 +1643,13 @@ double CudaCalcAmoebaMultipoleForceKernel::execute(ContextImpl& context, bool in
             &labFrameDipoles->getDevicePointer(), &labFrameQuadrupoles->getDevicePointer(), &inducedDipole->getDevicePointer(),
             &inducedDipolePolar->getDevicePointer(), &dampingAndThole->getDevicePointer()};
         cu.executeKernel(electrostaticsKernel, electrostaticsArgs, numForceThreadBlocks*electrostaticsThreads, electrostaticsThreads);
+        void* pmeTransformInducedPotentialArgs[] = {&pmePhidp->getDevicePointer(), &pmeCphi->getDevicePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
+        cu.executeKernel(pmeTransformPotentialKernel, pmeTransformInducedPotentialArgs, cu.getNumAtoms());
         void* pmeInducedForceArgs[] = {&cu.getPosq().getDevicePointer(), &cu.getForce().getDevicePointer(), &torque->getDevicePointer(),
             &cu.getEnergyBuffer().getDevicePointer(), &labFrameDipoles->getDevicePointer(), &labFrameQuadrupoles->getDevicePointer(),
             &fracDipoles->getDevicePointer(), &fracQuadrupoles->getDevicePointer(),
             &inducedDipole->getDevicePointer(), &inducedDipolePolar->getDevicePointer(), &pmePhi->getDevicePointer(), &pmePhid->getDevicePointer(),
-            &pmePhip->getDevicePointer(), &pmePhidp->getDevicePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
+            &pmePhip->getDevicePointer(), &pmePhidp->getDevicePointer(), &pmeCphi->getDevicePointer(), recipBoxVectorPointer[0], recipBoxVectorPointer[1], recipBoxVectorPointer[2]};
         cu.executeKernel(pmeInducedForceKernel, pmeInducedForceArgs, cu.getNumAtoms());
     }
 

plugins/amoeba/platforms/cuda/src/AmoebaCudaKernels.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) 2008-2013 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2015 Stanford University and the Authors.      *
  * Authors: Mark Friedrichs, Peter Eastman                                    *
  * Contributors:                                                              *
  *                                                                            *
@@ -421,6 +421,7 @@ private:
     CudaArray* pmePhid;
     CudaArray* pmePhip;
     CudaArray* pmePhidp;
+    CudaArray* pmeCphi;
     CudaArray* pmeAtomRange;
     CudaArray* pmeAtomGridIndex;
     CudaArray* lastPositions;
@@ -430,7 +431,7 @@ private:
     CUfunction pmeGridIndexKernel, pmeSpreadFixedMultipolesKernel, pmeSpreadInducedDipolesKernel, pmeFinishSpreadChargeKernel, pmeConvolutionKernel;
     CUfunction pmeFixedPotentialKernel, pmeInducedPotentialKernel, pmeFixedForceKernel, pmeInducedForceKernel, pmeRecordInducedFieldDipolesKernel, computePotentialKernel;
     CUfunction recordDIISDipolesKernel, buildMatrixKernel;
-    CUfunction pmeTransformMultipolesKernel;
+    CUfunction pmeTransformMultipolesKernel, pmeTransformPotentialKernel;
     CudaCalcAmoebaGeneralizedKirkwoodForceKernel* gkKernel;
     static const int PmeOrder = 5;
     static const int MaxPrevDIISDipoles = 20;