CudaPmeRealSpaceCompute Class Reference

#include <CudaPmeSolverUtil.h>

Inheritance diagram for CudaPmeRealSpaceCompute:

Public Member Functions
	CudaPmeRealSpaceCompute (PmeGrid pmeGrid, const int jblock, const int kblock, int deviceID, cudaStream_t stream)
	~CudaPmeRealSpaceCompute ()
void	copyAtoms (const int numAtoms, const CudaAtom *atoms)
void	spreadCharge (Lattice &lattice)
void	gatherForce (Lattice &lattice, CudaForce *force)
void	gatherForceSetCallback (ComputePmeCUDADevice *devicePtr_in)
void	waitGatherForceDone ()
Public Member Functions inherited from PmeRealSpaceCompute
	PmeRealSpaceCompute (PmeGrid pmeGrid, const int jblock, const int kblock)
virtual	~PmeRealSpaceCompute ()
float *	getData ()
int	getDataSize ()

Additional Inherited Members
Static Public Member Functions inherited from PmeRealSpaceCompute
static double	calcGridCoord (const double x, const double recip11, const int nfftx)
static void	calcGridCoord (const double x, const double y, const double z, const double recip11, const double recip22, const double recip33, const int nfftx, const int nffty, const int nfftz, double &frx, double &fry, double &frz)
static void	calcGridCoord (const float x, const float y, const float z, const float recip11, const float recip22, const float recip33, const int nfftx, const int nffty, const int nfftz, float &frx, float &fry, float &frz)
static void	calcGridCoord (const float x, const float y, const float z, const int nfftx, const int nffty, const int nfftz, float &frx, float &fry, float &frz)
static void	calcGridCoord (const double x, const double y, const double z, const int nfftx, const int nffty, const int nfftz, double &frx, double &fry, double &frz)
Protected Attributes inherited from PmeRealSpaceCompute
int	numAtoms
PmeGrid	pmeGrid
int	y0
int	z0
int	xsize
int	ysize
int	zsize
int	dataSize
float *	data
const int	jblock
const int	kblock

Detailed Description

Definition at line 125 of file CudaPmeSolverUtil.h.

Constructor & Destructor Documentation

CudaPmeRealSpaceCompute::CudaPmeRealSpaceCompute	(	PmeGrid	pmeGrid,
		const int	jblock,
		const int	kblock,
		int	deviceID,
		cudaStream_t	stream
	)

Definition at line 584 of file CudaPmeSolverUtil.C.

References cudaCheck, PmeRealSpaceCompute::data, PmeRealSpaceCompute::dataSize, NAMD_bug(), PmeRealSpaceCompute::xsize, PmeRealSpaceCompute::ysize, and PmeRealSpaceCompute::zsize.

                                                                         : 
  PmeRealSpaceCompute(pmeGrid, jblock, kblock), deviceID(deviceID), stream(stream) {
  if (dataSize < xsize*ysize*zsize)
    NAMD_bug("CudaPmeRealSpaceCompute::CudaPmeRealSpaceCompute, insufficient dataSize");
  cudaCheck(cudaSetDevice(deviceID));
  d_atomsCapacity = 0;
  d_atoms = NULL;
  d_forceCapacity = 0;
  d_force = NULL;
  #ifdef NAMD_CUDA
  tex_data = NULL;
  tex_data_len = 0;
  #else
  grid_data = NULL;
  grid_data_len = 0;
  #endif
  allocate_device<float>(&data, dataSize);
  setupGridData(data, xsize*ysize*zsize);
  cudaCheck(cudaEventCreate(&gatherForceEvent));
}

CudaPmeRealSpaceCompute::~CudaPmeRealSpaceCompute

(

)

Definition at line 609 of file CudaPmeSolverUtil.C.

References cudaCheck, and PmeRealSpaceCompute::data.

                                                  {
  cudaCheck(cudaSetDevice(deviceID));
  if (d_atoms != NULL) deallocate_device<CudaAtom>(&d_atoms);
  if (d_force != NULL) deallocate_device<CudaForce>(&d_force);
  // if (d_patches != NULL) deallocate_device<PatchInfo>(&d_patches);
  // deallocate_device<double>(&d_selfEnergy);
  deallocate_device<float>(&data);
  cudaCheck(cudaEventDestroy(gatherForceEvent));
}

Member Function Documentation

void CudaPmeRealSpaceCompute::copyAtoms ( const int  numAtoms, const CudaAtom *  atoms  )

virtual

Implements PmeRealSpaceCompute.

Definition at line 640 of file CudaPmeSolverUtil.C.

References atoms, cudaCheck, and PmeRealSpaceCompute::numAtoms.

                                                                                 {
  cudaCheck(cudaSetDevice(deviceID));
  this->numAtoms = numAtoms;

  // Reallocate device arrays as neccessary
  reallocate_device<CudaAtom>(&d_atoms, &d_atomsCapacity, numAtoms, 1.5f);

  // Copy atom data to device
  copy_HtoD<CudaAtom>(atoms, d_atoms, numAtoms, stream);
}

void CudaPmeRealSpaceCompute::gatherForce ( Lattice &  lattice, CudaForce *  force  )

virtual

Implements PmeRealSpaceCompute.

Definition at line 751 of file CudaPmeSolverUtil.C.

References cudaCheck, gather_force(), PmeGrid::K1, PmeGrid::K2, PmeGrid::K3, PmeRealSpaceCompute::numAtoms, PmeGrid::order, PmeRealSpaceCompute::pmeGrid, PmeRealSpaceCompute::xsize, PmeRealSpaceCompute::y0, PmeGrid::yBlocks, PmeRealSpaceCompute::ysize, PmeRealSpaceCompute::z0, PmeGrid::zBlocks, and PmeRealSpaceCompute::zsize.

                                                                            {
  cudaCheck(cudaSetDevice(deviceID));

  // Re-allocate force array if needed
  reallocate_device<CudaForce>(&d_force, &d_forceCapacity, numAtoms, 1.5f);

  gather_force((const float4*)d_atoms, numAtoms,
    pmeGrid.K1, pmeGrid.K2, pmeGrid.K3,
    xsize, ysize, zsize, xsize, y0, z0, (pmeGrid.yBlocks == 1), (pmeGrid.zBlocks == 1),
    data, pmeGrid.order, (float3*)d_force, 
#ifdef NAMD_CUDA
    gridTexObj,
#endif
    stream);

  copy_DtoH<CudaForce>(d_force, force, numAtoms, stream);

  cudaCheck(cudaEventRecord(gatherForceEvent, stream));
}

void CudaPmeRealSpaceCompute::gatherForceSetCallback

(

ComputePmeCUDADevice *

devicePtr_in

)

Definition at line 703 of file CudaPmeSolverUtil.C.

References CcdCallBacksReset(), and cudaCheck.

                                                                                       {
  cudaCheck(cudaSetDevice(deviceID));
  devicePtr = devicePtr_in;
  checkCount = 0;
  CcdCallBacksReset(0, CmiWallTimer());
  // Set the call back at 0.1ms
  CcdCallFnAfter(cuda_gatherforce_check, this, 0.1);
}

void CudaPmeRealSpaceCompute::spreadCharge ( Lattice &  lattice )

virtual

Implements PmeRealSpaceCompute.

Definition at line 654 of file CudaPmeSolverUtil.C.

References cudaCheck, PmeRealSpaceCompute::data, PmeGrid::K1, PmeGrid::K2, PmeGrid::K3, PmeRealSpaceCompute::numAtoms, PmeGrid::order, PmeRealSpaceCompute::pmeGrid, spread_charge(), PmeRealSpaceCompute::xsize, PmeRealSpaceCompute::y0, PmeGrid::yBlocks, PmeRealSpaceCompute::ysize, PmeRealSpaceCompute::z0, PmeGrid::zBlocks, and PmeRealSpaceCompute::zsize.

                                                           {
  cudaCheck(cudaSetDevice(deviceID));

  // Clear grid
  clear_device_array<float>(data, xsize*ysize*zsize, stream);

  spread_charge((const float4*)d_atoms, numAtoms,
    pmeGrid.K1, pmeGrid.K2, pmeGrid.K3, xsize, ysize, zsize,
    xsize, y0, z0, (pmeGrid.yBlocks == 1), (pmeGrid.zBlocks == 1),
    data, pmeGrid.order, stream);

  // ncall++;

  // if (ncall == 1) writeRealToDisk(data, xsize*ysize*zsize, "data.txt");
}

void CudaPmeRealSpaceCompute::waitGatherForceDone

(

)

Definition at line 712 of file CudaPmeSolverUtil.C.

References cudaCheck.

                                                  {
  cudaCheck(cudaSetDevice(deviceID));
  cudaCheck(cudaEventSynchronize(gatherForceEvent));
}

---

The documentation for this class was generated from the following files:

• CudaPmeSolverUtil.h
• CudaPmeSolverUtil.C