Page MenuHomec4science
Files F68962911

pair_gayberne_gpu.cpp
No OneTemporary
Actions

File Metadata

Created
Sat, Jun 29, 16:36

pair_gayberne_gpu.cpp
View Options

/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
Copyright (2003) Sandia Corporation. Under the terms of Contract
DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
certain rights in this software. This software is distributed under
the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
/* ----------------------------------------------------------------------
Contributing author: Mike Brown (SNL)
------------------------------------------------------------------------- */
#include "math.h"
#include "stdio.h"
#include "stdlib.h"
#include "pair_gayberne_gpu.h"
#include "math_extra.h"
#include "atom.h"
#include "atom_vec.h"
#include "comm.h"
#include "force.h"
#include "neighbor.h"
#include "neigh_list.h"
#include "integrate.h"
#include "memory.h"
#include "error.h"
#include "neigh_request.h"
#include "universe.h"
#include <string>
#ifdef GB_GPU_OMP
#include "omp.h"
#endif
#define MIN(a,b) ((a) < (b) ? (a) : (b))
#define MAX(a,b) ((a) > (b) ? (a) : (b))
// External functions from cuda library for atom decomposition
bool gb_gpu_init(int &ij_size, const int ntypes, const double gamma,
const double upsilon, const double mu, double **shape,
double **well, double **cutsq, double **sigma,
double **epsilon, double *host_lshape, int **form,
double **host_lj1, double **host_lj2, double **host_lj3,
double **host_lj4, double **offset, double *special_lj,
const int max_nbors, const int thread, const int gpu_id);
void gb_gpu_clear(const int thread);
int * gb_gpu_reset_nbors(const int nall, const int nlocal, const int inum,
int *ilist, const int *numj, const int *type,
const int thread, bool &success);
void gb_gpu_nbors(const int *ij, const int num_ij, const bool eflag,
const int thread);
void gb_gpu_atom(double **host_x, double **host_quat, const int *host_type,
const bool rebuild, const int thread);
void gb_gpu_gayberne(const bool eflag, const bool vflag, const bool rebuild,
const int thread);
double gb_gpu_forces(double **f, double **tor, const int *ilist,
const bool eflag, const bool vflag, const bool eflag_atom,
const bool vflag_atom, double *eatom, double **vatom,
double *virial, const int thread);
std::string gb_gpu_name(const int i, const int max_nbors);
void gb_gpu_time(const int thread);
int gb_gpu_num_devices();
double gb_gpu_bytes();
using namespace LAMMPS_NS;
/* ---------------------------------------------------------------------- */
PairGayBerneGPU::PairGayBerneGPU(LAMMPS *lmp) : PairGayBerne(lmp), my_thread(0),
omp_chunk(0), nthreads(1),
multi_gpu_mode(ONE_NODE),
multi_gpu_param(0)
{
ij_new[0]=NULL;
}
/* ----------------------------------------------------------------------
free all arrays
------------------------------------------------------------------------- */
PairGayBerneGPU::~PairGayBerneGPU()
{
printf("\n\n-------------------------------------");
printf("--------------------------------\n");
printf(" GPU Time Stamps: ");
printf("\n-------------------------------------");
printf("--------------------------------\n");
gb_gpu_time(my_thread);
printf("-------------------------------------");
printf("--------------------------------\n\n");
#pragma omp parallel
{
#ifdef GB_GPU_OMP
int my_thread=omp_get_thread_num();
#endif
gb_gpu_clear(my_thread);
if (ij_new[my_thread]!=NULL)
delete [] ij_new[my_thread];
}
}
/* ---------------------------------------------------------------------- */
void PairGayBerneGPU::compute(int eflag, int vflag)
{
if (eflag || vflag) ev_setup(eflag,vflag);
else evflag = vflag_fdotr = 0;
if (vflag_atom)
error->all("Per-atom virial not available with GPU Gay-Berne");
int nlocal = atom->nlocal;
int nall = nlocal + atom->nghost;
int inum = list->inum;
bool rebuild=false;
if (neighbor->ncalls > last_neighbor) {
last_neighbor=neighbor->ncalls;
rebuild=true;
}
#pragma omp parallel
{
bool success=true;
#ifdef GB_GPU_OMP
int my_thread=omp_get_thread_num();
if (rebuild) {
omp_chunk=static_cast<int>(ceil(static_cast<double>(inum)/nthreads));
if (my_thread==nthreads-1)
thread_inum[my_thread]=inum-(nthreads-1)*omp_chunk;
else
thread_inum[my_thread]=omp_chunk;
olist[my_thread]=gb_gpu_reset_nbors(nall, atom->nlocal,
thread_inum[my_thread],
list->ilist+omp_chunk*my_thread,
list->numneigh, atom->type, my_thread,
success);
}
#else
if (rebuild)
olist[my_thread]=gb_gpu_reset_nbors(nall, atom->nlocal, inum, list->ilist,
list->numneigh, atom->type, my_thread,
success);
#endif
if (!success)
error->one("Total # of atoms exceeds maximum allowed per GPGPU");
// copy atom data to GPU
gb_gpu_atom(atom->x,atom->quat,atom->type,rebuild,my_thread);
int i,j,ii,jj,jnum;
double factor_lj;
int *jlist;
if (rebuild==true) {
int num_ij = 0;
// loop over neighbors of my atoms
int *ijp=ij_new[my_thread];
#ifdef GB_GPU_OMP
int mgo=my_thread*omp_chunk;
int mgot=mgo+thread_inum[my_thread];
#else
int mgo=0, mgot=inum;
#endif
for (ii = mgo; ii<mgot; ii++) {
i = olist[my_thread][ii];
jlist = list->firstneigh[i];
jnum = list->numneigh[i];
for (jj = 0; jj < jnum; jj++) {
j = jlist[jj];
*ijp=j;
ijp++;
num_ij++;
if (num_ij==ij_size) {
gb_gpu_nbors(ij_new[my_thread],num_ij,eflag,my_thread);
ijp=ij_new[my_thread];
num_ij=0;
}
}
}
if (num_ij>0)
gb_gpu_nbors(ij_new[my_thread],num_ij,eflag,my_thread);
}
gb_gpu_gayberne(eflag,vflag,rebuild,my_thread);
double lvirial[6];
for (int i=0; i<6; i++) lvirial[i]=0.0;
double my_eng=gb_gpu_forces(atom->f,atom->torque,olist[my_thread],eflag,vflag,
eflag_atom, vflag_atom, eatom, vatom, lvirial,
my_thread);
#pragma omp critical
{
eng_vdwl+=my_eng;
virial[0]+=lvirial[0];
virial[1]+=lvirial[1];
virial[2]+=lvirial[2];
virial[3]+=lvirial[3];
virial[4]+=lvirial[4];
virial[5]+=lvirial[5];
}
} //End parallel
if (vflag_fdotr) virial_compute();
}
/* ----------------------------------------------------------------------
global settings
------------------------------------------------------------------------- */
void PairGayBerneGPU::settings(int narg, char **arg)
{
// strip off GPU keyword/value and send remaining args to parent
if (narg < 2) error->all("Illegal pair_style command");
// set multi_gpu_mode to one/node for multiple gpus on 1 node
// -- param is starting gpu id
// set multi_gpu_mode to one/gpu to select the same gpu id on every node
// -- param is id of gpu
// set multi_gpu_mode to multi/gpu to get ma
// -- param is number of gpus per node
if (strcmp(arg[0],"one/node") == 0)
multi_gpu_mode = ONE_NODE;
else if (strcmp(arg[0],"one/gpu") == 0)
multi_gpu_mode = ONE_GPU;
else if (strcmp(arg[0],"multi/gpu") == 0)
multi_gpu_mode = MULTI_GPU;
else error->all("Illegal pair_style command");
multi_gpu_param = atoi(arg[1]);
if (multi_gpu_mode == MULTI_GPU && multi_gpu_param < 1)
error->all("Illegal pair_style command");
PairGayBerne::settings(narg-2,&arg[2]);
}
/* ----------------------------------------------------------------------
init specific to this pair style
------------------------------------------------------------------------- */
void PairGayBerneGPU::init_style()
{
if (force->pair_match("gpu",0) == NULL)
error->all("Cannot use pair hybrid with multiple GPU pair styles");
if (!atom->quat_flag || !atom->torque_flag || !atom->avec->shape_type)
error->all("Pair gayberne requires atom attributes quat, torque, shape");
if (atom->radius_flag)
error->all("Pair gayberne cannot be used with atom attribute diameter");
// set the GPU ID
int my_gpu=comm->me+multi_gpu_param;
int ngpus=universe->nprocs;
if (multi_gpu_mode==ONE_GPU) {
my_gpu=multi_gpu_param;
ngpus=1;
} else if (multi_gpu_mode==MULTI_GPU) {
ngpus=multi_gpu_param;
my_gpu=comm->me%ngpus;
if (ngpus>universe->nprocs)
ngpus=universe->nprocs;
}
int irequest = neighbor->request(this);
// per-type shape precalculations
for (int i = 1; i <= atom->ntypes; i++) {
if (setwell[i]) {
double *one = atom->shape[i];
shape[i][0] = one[0]*one[0];
shape[i][1] = one[1]*one[1];
shape[i][2] = one[2]*one[2];
lshape[i] = (one[0]*one[1]+one[2]*one[2])*sqrt(one[0]*one[1]);
}
}
// Repeat cutsq calculation because done after call to init_style
double cut;
for (int i = 1; i <= atom->ntypes; i++)
for (int j = i; j <= atom->ntypes; j++) {
cut = init_one(i,j);
cutsq[i][j] = cutsq[j][i] = cut*cut;
}
// If compiled with OpenMP and only 1 proc, try to use multiple GPUs w/threads
#ifdef GB_GPU_OMP
if (multi_gpu_mode!=ONE_GPU)
nthreads=ngpus=gb_gpu_num_devices();
else
nthreads=ngpus=1;
if (nthreads>MAX_GPU_THREADS)
nthreads=MAX_GPU_THREADS;
omp_set_num_threads(nthreads);
#endif
#pragma omp parallel firstprivate(my_gpu)
{
#ifdef GB_GPU_OMP
int my_thread = omp_get_thread_num();
if (multi_gpu_mode!=ONE_GPU)
my_gpu=my_thread;
if (multi_gpu_mode==ONE_NODE)
my_gpu+=multi_gpu_param;
#endif
bool init_ok=gb_gpu_init(ij_size, atom->ntypes+1, gamma, upsilon, mu,
shape, well, cutsq, sigma, epsilon, lshape, form,
lj1, lj2, lj3, lj4, offset, force->special_lj,
neighbor->oneatom, my_thread, my_gpu);
if (!init_ok)
error->one("At least one process could not allocate a CUDA-enabled gpu");
if (ij_new[my_thread]!=NULL)
delete [] ij_new[my_thread];
ij_new[my_thread]=new int[ij_size];
}
last_neighbor = -1;
neighbor->requests[irequest]->half = 0;
neighbor->requests[irequest]->full = 1;
if (force->newton_pair)
error->all("Cannot use newton pair with GPU GayBerne pair style");
if (comm->me == 0 && screen) {
printf("\n-------------------------------------");
printf("-------------------------------------\n");
printf("- Using GPGPU acceleration for Gay-Berne:\n");
printf("-------------------------------------");
printf("-------------------------------------\n");
for (int i=0; i<ngpus; i++) {
int gpui=my_gpu;
if (multi_gpu_mode==ONE_NODE)
gpui=i+multi_gpu_param;
else if (multi_gpu_mode==MULTI_GPU)
gpui=i;
std::string gpu_string=gb_gpu_name(gpui,neighbor->oneatom);
printf("GPU %d: %s\n",gpui,gpu_string.c_str());
}
printf("-------------------------------------");
printf("-------------------------------------\n\n");
}
}
/* ---------------------------------------------------------------------- */
double PairGayBerneGPU::memory_usage()
{
double bytes=Pair::memory_usage()+nthreads*ij_size*sizeof(int);
return bytes+gb_gpu_bytes();
}

Event Timeline

c4science · Help