Skip to content

GitLab

Commit e94e4d0a authored by Peter Eastman's avatar Peter Eastman
Browse files

Fixed bug computing B-spline moduli

parent a9b636aa
Show whitespace changes
Inline Side-by-side
Showing with 101 additions and 65 deletions
platforms/cuda/include/CudaKernels.h
View file @ e94e4d0a
......@@ -599,8 +599,8 @@ class CudaCalcNonbondedForceKernel : public CalcNonbondedForceKernel {
public:
CudaCalcNonbondedForceKernel(std::string name, const Platform& platform, CudaContext& cu, const System& system) : CalcNonbondedForceKernel(name, platform),
cu(cu), hasInitializedFFT(false), sigmaEpsilon(NULL), exceptionParams(NULL), cosSinSums(NULL), directPmeGrid(NULL), reciprocalPmeGrid(NULL),
pmeBsplineModuliX(NULL), pmeBsplineModuliY(NULL), pmeBsplineModuliZ(NULL), pmeAtomRange(NULL), pmeAtomGridIndex(NULL),
pmeEnergyBuffer(NULL), sort(NULL), dispersionFft(NULL), fft(NULL), pmeio(NULL) {
pmeBsplineModuliX(NULL), pmeBsplineModuliY(NULL), pmeBsplineModuliZ(NULL), pmeDispersionBsplineModuliX(NULL), pmeDispersionBsplineModuliY(NULL),
pmeDispersionBsplineModuliZ(NULL), pmeAtomRange(NULL), pmeAtomGridIndex(NULL), pmeEnergyBuffer(NULL), sort(NULL), dispersionFft(NULL), fft(NULL), pmeio(NULL) {
}
~CudaCalcNonbondedForceKernel();
/**
......@@ -672,6 +672,9 @@ private:
CudaArray* pmeBsplineModuliX;
CudaArray* pmeBsplineModuliY;
CudaArray* pmeBsplineModuliZ;
CudaArray* pmeDispersionBsplineModuliX;
CudaArray* pmeDispersionBsplineModuliY;
CudaArray* pmeDispersionBsplineModuliZ;
CudaArray* pmeAtomRange;
CudaArray* pmeAtomGridIndex;
CudaArray* pmeEnergyBuffer;
......
platforms/cuda/src/CudaKernels.cpp
View file @ e94e4d0a
......@@ -1607,6 +1607,12 @@ CudaCalcNonbondedForceKernel::~CudaCalcNonbondedForceKernel() {
delete pmeBsplineModuliY;
if (pmeBsplineModuliZ != NULL)
delete pmeBsplineModuliZ;
if (pmeDispersionBsplineModuliX != NULL)
delete pmeDispersionBsplineModuliX;
if (pmeDispersionBsplineModuliY != NULL)
delete pmeDispersionBsplineModuliY;
if (pmeDispersionBsplineModuliZ != NULL)
delete pmeDispersionBsplineModuliZ;
if (pmeAtomRange != NULL)
delete pmeAtomRange;
if (pmeAtomGridIndex != NULL)
......@@ -1861,6 +1867,11 @@ void CudaCalcNonbondedForceKernel::initialize(const System& system, const Nonbon
pmeBsplineModuliX = new CudaArray(cu, gridSizeX, elementSize, "pmeBsplineModuliX");
pmeBsplineModuliY = new CudaArray(cu, gridSizeY, elementSize, "pmeBsplineModuliY");
pmeBsplineModuliZ = new CudaArray(cu, gridSizeZ, elementSize, "pmeBsplineModuliZ");
if (doLJPME) {
pmeDispersionBsplineModuliX = new CudaArray(cu, dispersionGridSizeX, elementSize, "pmeDispersionBsplineModuliX");
pmeDispersionBsplineModuliY = new CudaArray(cu, dispersionGridSizeY, elementSize, "pmeDispersionBsplineModuliY");
pmeDispersionBsplineModuliZ = new CudaArray(cu, dispersionGridSizeZ, elementSize, "pmeDispersionBsplineModuliZ");
}
pmeAtomRange = CudaArray::create<int>(cu, gridSizeX*gridSizeY*gridSizeZ+1, "pmeAtomRange");
pmeAtomGridIndex = CudaArray::create<int2>(cu, numParticles, "pmeAtomGridIndex");
int energyElementSize = (cu.getUseDoublePrecision() || cu.getUseMixedPrecision() ? sizeof(double) : sizeof(float));
......@@ -1911,7 +1922,28 @@ void CudaCalcNonbondedForceKernel::initialize(const System& system, const Nonbon
hasInitializedFFT = true;
// Initialize the b-spline moduli.
int maxSize = max(max(gridSizeX, gridSizeY), gridSizeZ);
for (int grid = 0; grid < 2; grid++) {
int xsize, ysize, zsize;
CudaArray *xmoduli, *ymoduli, *zmoduli;
if (grid == 0) {
xsize = gridSizeX;
ysize = gridSizeY;
zsize = gridSizeZ;
xmoduli = pmeBsplineModuliX;
ymoduli = pmeBsplineModuliY;
zmoduli = pmeBsplineModuliZ;
}
else {
if (!doLJPME)
continue;
xsize = dispersionGridSizeX;
ysize = dispersionGridSizeY;
zsize = dispersionGridSizeZ;
xmoduli = pmeDispersionBsplineModuliX;
ymoduli = pmeDispersionBsplineModuliY;
zmoduli = pmeDispersionBsplineModuliZ;
}
int maxSize = max(max(xsize, ysize), zsize);
vector<double> data(PmeOrder);
vector<double> ddata(PmeOrder);
vector<double> bsplines_data(maxSize);
......@@ -1944,7 +1976,7 @@ void CudaCalcNonbondedForceKernel::initialize(const System& system, const Nonbon
// Evaluate the actual bspline moduli for X/Y/Z.
for(int dim = 0; dim < 3; dim++) {
int ndata = (dim == 0 ? gridSizeX : dim == 1 ? gridSizeY : gridSizeZ);
int ndata = (dim == 0 ? xsize : dim == 1 ? ysize : zsize);
vector<double> moduli(ndata);
for (int i = 0; i < ndata; i++) {
double sc = 0.0;
......@@ -1961,22 +1993,23 @@ void CudaCalcNonbondedForceKernel::initialize(const System& system, const Nonbon
moduli[i] = (moduli[i-1]+moduli[i+1])*0.5;
if (cu.getUseDoublePrecision()) {
if (dim == 0)
pmeBsplineModuliX->upload(moduli);
xmoduli->upload(moduli);
else if (dim == 1)
pmeBsplineModuliY->upload(moduli);
ymoduli->upload(moduli);
else
pmeBsplineModuliZ->upload(moduli);
zmoduli->upload(moduli);
}
else {
vector<float> modulif(ndata);
for (int i = 0; i < ndata; i++)
modulif[i] = (float) moduli[i];
if (dim == 0)
pmeBsplineModuliX->upload(modulif);
xmoduli->upload(modulif);
else if (dim == 1)
pmeBsplineModuliY->upload(modulif);
ymoduli->upload(modulif);
else
pmeBsplineModuliZ->upload(modulif);
zmoduli->upload(modulif);
}
}
}
}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment