Fixed potential energy discrepancies in large systems between CUDA and...

Fixed potential energy discrepancies in large systems between CUDA and Reference platform. The energy calculation now uses the same method as that of reciprocalConvolution so it mirrors the reference implementation. OpenCL is fine.  

platforms/cuda/src/CudaKernels.cpp

@@ -1686,6 +1686,11 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF
         else
             cufftExecR2C(fftForward, (float*) originalPmeGrid->getDevicePointer(), (float2*) reciprocalPmeGrid->getDevicePointer());
 
+        if (includeEnergy) {
+            void* computeEnergyArgs[] = {&reciprocalPmeGrid->getDevicePointer(), &cu.getEnergyBuffer().getDevicePointer(), &pmeBsplineModuliX->getDevicePointer(), &pmeBsplineModuliY->getDevicePointer(), &pmeBsplineModuliZ->getDevicePointer(), cu.getPeriodicBoxSizePointer(), cu.getInvPeriodicBoxSizePointer()};
+            cu.executeKernel(pmeEvalEnergyKernel, computeEnergyArgs, cu.getNumAtoms());
+        }
+
         void* convolutionArgs[] = {&reciprocalPmeGrid->getDevicePointer(), &cu.getEnergyBuffer().getDevicePointer(), &pmeBsplineModuliX->getDevicePointer(), &pmeBsplineModuliY->getDevicePointer(), &pmeBsplineModuliZ->getDevicePointer(), cu.getPeriodicBoxSizePointer(), cu.getInvPeriodicBoxSizePointer()};
         cu.executeKernel(pmeConvolutionKernel, convolutionArgs, cu.getNumAtoms());
 
@@ -1694,11 +1699,7 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF
         else
             cufftExecC2R(fftBackward, (float2*) reciprocalPmeGrid->getDevicePointer(), (float*)  convolvedPmeGrid->getDevicePointer());
 
-        if (includeEnergy) {
-            void* computeEnergyArgs[] = {&originalPmeGrid->getDevicePointer(), &convolvedPmeGrid->getDevicePointer(), &cu.getEnergyBuffer().getDevicePointer() };
-            cu.executeKernel(pmeEvalEnergyKernel, computeEnergyArgs, cu.getNumAtoms());
-        }
-        
+
         void* interpolateArgs[] = {&cu.getPosq().getDevicePointer(), &cu.getForce().getDevicePointer(), &convolvedPmeGrid->getDevicePointer(),
                 cu.getPeriodicBoxSizePointer(), cu.getInvPeriodicBoxSizePointer()};
         cu.executeKernel(pmeInterpolateForceKernel, interpolateArgs, cu.getNumAtoms());