Page MenuHomec4science
Files F79335066

lal_eam_ext.cpp
No OneTemporary
Actions

File Metadata

Created
Sun, Aug 25, 21:39

lal_eam_ext.cpp
View Options

// **************************************************************************
// lal_eam_ext.cpp
// -------------------
// W. Michael Brown, Trung Dac Nguyen (ORNL)
//
// Class for acceleration of the eam pair style
//
// __________________________________________________________________________
// This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
// __________________________________________________________________________
//
// begin :
// email : brownw@ornl.gov nguyentd@ornl.gov
// ***************************************************************************/
#include <iostream>
#include <cassert>
#include <math.h>
#include "lal_eam.h"
using namespace std;
using namespace LAMMPS_AL;
static EAM<PRECISION,ACC_PRECISION> EAMMF;
// ---------------------------------------------------------------------------
// Allocate memory on host and device and copy constants to device
// ---------------------------------------------------------------------------
int eam_gpu_init(const int ntypes, double host_cutforcesq,
int **host_type2rhor, int **host_type2z2r, int *host_type2frho,
double ***host_rhor_spline, double ***host_z2r_spline,
double ***host_frho_spline,
double rdr, double rdrho, int nrhor,
int nrho, int nz2r, int nfrho, int nr,
const int nlocal, const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
int &gpu_mode, FILE *screen) {
EAMMF.clear();
gpu_mode=EAMMF.device->gpu_mode();
double gpu_split=EAMMF.device->particle_split();
int first_gpu=EAMMF.device->first_device();
int last_gpu=EAMMF.device->last_device();
int world_me=EAMMF.device->world_me();
int gpu_rank=EAMMF.device->gpu_rank();
int procs_per_gpu=EAMMF.device->procs_per_gpu();
EAMMF.device->init_message(screen,"eam",first_gpu,last_gpu);
bool message=false;
if (EAMMF.device->replica_me()==0 && screen)
message=true;
if (message) {
fprintf(screen,"Initializing GPU and compiling on process 0...");
fflush(screen);
}
int init_ok=0;
if (world_me==0)
init_ok=EAMMF.init(ntypes, host_cutforcesq,
host_type2rhor, host_type2z2r, host_type2frho,
host_rhor_spline, host_z2r_spline,
host_frho_spline,
rdr, rdrho, nrhor, nrho, nz2r, nfrho, nr,
nlocal, nall, 300, maxspecial,
cell_size, gpu_split, screen);
EAMMF.device->world_barrier();
if (message)
fprintf(screen,"Done.\n");
for (int i=0; i<procs_per_gpu; i++) {
if (message) {
if (last_gpu-first_gpu==0)
fprintf(screen,"Initializing GPU %d on core %d...",first_gpu,i);
else
fprintf(screen,"Initializing GPUs %d-%d on core %d...",first_gpu,
last_gpu,i);
fflush(screen);
}
if (gpu_rank==i && world_me!=0)
init_ok=EAMMF.init(ntypes, host_cutforcesq,
host_type2rhor, host_type2z2r, host_type2frho,
host_rhor_spline, host_z2r_spline,
host_frho_spline,
rdr, rdrho, nrhor, nrho, nz2r, nfrho, nr,
nlocal, nall, 300, maxspecial,
cell_size, gpu_split, screen);
EAMMF.device->gpu_barrier();
if (message)
fprintf(screen,"Done.\n");
}
if (message)
fprintf(screen,"\n");
if (init_ok==0)
EAMMF.estimate_gpu_overhead();
return init_ok;
}
void eam_gpu_clear() {
EAMMF.clear();
}
int ** eam_gpu_compute_energy_n(const int ago, const int inum_full,
const int nall, double **host_x, int *host_type,
double *sublo, double *subhi, int *tag, int **nspecial,
int **special, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
int **ilist, int **jnum, const double cpu_time,
bool &success, double *host_fp, double *boxlo,
double *prd, int &inum) {
return EAMMF.compute(ago, inum_full, nall, host_x, host_type, sublo,
subhi, tag, nspecial, special, eflag, vflag, eatom,
vatom, host_start, ilist, jnum, cpu_time, success,
host_fp, boxlo, prd, inum);
}
void eam_gpu_compute_energy(const int ago, const int inum_full, const int nall,
double **host_x, int *host_type, int *ilist, int *numj,
int **firstneigh, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
const double cpu_time, bool &success, double *host_fp,
const int nlocal, double *boxlo, double *prd) {
EAMMF.compute(ago,inum_full,nall,host_x,host_type,ilist,numj,
firstneigh,eflag,vflag,eatom,vatom,host_start,cpu_time,success,
host_fp,nlocal,boxlo,prd);
}
void eam_gpu_compute_n(const int ago, const int inum_full,
const int nall, double **host_x, int *host_type,
double *sublo, double *subhi, int *tag, int **nspecial,
int **special, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
int **ilist, int **jnum, const double cpu_time,
bool &success, double *host_fp, double *boxlo,
double *prd, int inum) {
EAMMF.compute2(NULL, eflag, vflag, eatom, vatom, host_fp);
}
void eam_gpu_compute(const int ago, const int inum_full, const int nall,
double **host_x, int *host_type, int *ilist, int *numj,
int **firstneigh, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
const double cpu_time, bool &success, double *host_fp,
const int nlocal, double *boxlo, double *prd) {
EAMMF.compute2(ilist, eflag, vflag, eatom, vatom, host_fp);
}
double eam_gpu_bytes() {
return EAMMF.host_memory_usage();
}

Event Timeline

c4science · Help