void ParticleSystem::applyForces()
{
if ( gravity.length() > 0.0f )
{
for ( int i = 0; i < mParticles.size(); ++i )
{
Particle* p = mParticles[i];
p->mAcc += gravity;
}
}
for ( int i = 0; i < mParticles.size(); ++i )
{
Particle* p = mParticles[i];
p->mAcc -= p->mVel * drag;
}
for ( int i = 0; i < mSprings.size(); i++ )
{
Spring* f = mSprings[i];
f->apply();
}
for ( int i = 0; i < attractions.size(); i++ )
{
Attraction* f = attractions[i];
f->apply();
}
for ( int i = 0; i < customForces.size(); i++ )
{
Force* f = customForces[i];
f->apply();
}
}
float RuntimeObject::TotalForceLength() const {
Force ForceMoyenne;
ForceMoyenne.SetX(TotalForceX());
ForceMoyenne.SetY(TotalForceY());
return ForceMoyenne.GetLength();
}
sf::Vector3f Force::produitVectoriel(Force vect)
{
sf::Vector3f vect3d;
vect3d.x = 0;
vect3d.y = 0;
vect3d.z = getForce().x*vect.getForce().y-getForce().y*vect.getForce().x;
return vect3d;
}
S_Chicken::S_Chicken( )
{
AABox* aabox = new AABox;
aabox->SetSize( CHICKEN_WIDTH_DEPTH, CHICKEN_HEIGHT, CHICKEN_WIDTH_DEPTH );
SetCollide( aabox );
Force* force = new Force;
force->SetMaxVelocity( 3.0f );
SetMoveForce( force );
}
void OpenSimContext::updateDisplayer(Force& f) {
// If muscle or force acting along a path then call respective updateDisplayer otherwise do nothing for now
realizeVelocity();
if (dynamic_cast<Muscle*>(&f)!= NULL)
return dynamic_cast<Muscle*>(&f)->updateDisplayer(*_configState);
if (f.hasGeometryPath()){
AbstractProperty& pathProp = f.updPropertyByName("GeometryPath");
Object& pathObj = pathProp.updValueAsObject();
return dynamic_cast<GeometryPath*>(&pathObj)->updateDisplayer(*_configState);
}
if (f.getDisplayer()!= NULL)
f.updateDisplayer(*_configState);
}
void calculateNewPosition(double delta_t) {
Force resultant = getResultantForce();
// cout << "Resultant Force " << resultant.getDir().getX() << " " << resultant.getDir().getY() << endl;
Vec2D accel(resultant.getX() / _mass, resultant.getY() / _mass);
// cout << "Acceleration " << accel.getX() << " " << accel.getY() << endl;
Vec2D newPos = _pos + _vel * delta_t;
Vec2D newVel = _vel + accel * delta_t;
// cout << "Position " << newPos.getX() << " " << newPos.getY() << endl;
// cout << "Velocity " << newVel.getX() << " " << newVel.getY() << endl;
setPos(newPos);
setVel(newVel);
}
ForceIndex adoptForce(Force& force) {
invalidateSubsystemTopologyCache();
const ForceIndex index((int) forces.size());
forces.push_back(new Force()); // grow
Force& f = *forces.back(); // refer to the empty handle we just created
force.disown(f); // transfer ownership to f
return index;
}
int ImbalanceGroup::options(int narg, char **arg)
{
Error *error = _lmp->error;
Force *force = _lmp->force;
Group *group = _lmp->group;
if (narg < 3) error->all(FLERR,"Illegal balance weight command");
_num = force->inumeric(FLERR,arg[0]);
if (_num < 1) error->all(FLERR,"Illegal balance weight command");
if (2*_num+1 > narg) error->all(FLERR,"Illegal balance weight command");
_id = new int[_num];
_factor = new double[_num];
for (int i = 0; i < _num; ++i) {
_id[i] = group->find(arg[2*i+1]);
if (_id[i] < 0)
error->all(FLERR,"Unknown group in balance weight command");
_factor[i] = force->numeric(FLERR,arg[2*i+2]);
}
return 2*_num+1;
}
static void irr_simd_firr(
const int addr,
_out_ double accout[3],
_out_ double jrkout[3],
_out_ int &nnbid)
{
const Particle *ptcl = ::ptcl;
const v2df tnow = ::vec_tnow;
const int nnb = list[addr].nnb;
Force force;
force.clear();
const Pred4 pri(Predictor(ptcl[addr], tnow));
for(int k=0; k<nnb; k++){
const int jaddr = list[addr].nb[k];
ptcl[jaddr].prefetch();
}
for(int k=0; k<nnb; k+=4){
const int *jptr = &(list[addr].nb[k]);
const Predictor p0(ptcl[jptr[0]], tnow);
const Predictor p1(ptcl[jptr[1]], tnow);
const Predictor p2(ptcl[jptr[2]], tnow);
const Predictor p3(ptcl[jptr[3]], tnow);
const Pred4 prj(p0, p1, p2, p3);
assert(0 == (size_t)jptr % 16);
v4sf idx = *(v4sf *)jptr;
force.calc_and_accum(pri, prj, idx);
}
force.write(accout, jrkout, nnbid);
}
void GroupT::CalcReaction()
{
double SumFx=0;
double SumFy=0;
double SumTorque=0;
for (list<Force *>::iterator f = s1->forces.begin(); f!=s1->forces.end(); f++)
{
Force *x = *f;
if(x->IsCalculatedx())
SumFx+=x->GetForcex();
if(x->IsCalculatedy())
SumFy+=x->GetForcey();
if(x->IsCalculatedt())
SumTorque+=x->GetForcet();
}
O->R1->SetForce(-SumFx,-SumFy,0);
O->R2->SetForce(SumFx,SumFy,0);
}
static Vector3 getAngular(const Force & self) { return self.angular(); }
static void setLinear(Force & self, const Vector3 & f) { self.linear(f); }
static Vector3 getLinear(const Force & self ) { return self.linear(); }
static
boost::python::tuple
getinitargs(const Force & f)
{ return bp::make_tuple((Vector3)f.linear(),(Vector3)f.angular()); }
const Eigen::Matrix<double, 1, 1>
operator*( const Force& f ) const
{
return ref.axis.transpose()*f.angular();
}
typename Force::ConstLinear_t::template FixedSegmentReturnType<1>::Type
operator* (const Force & f) const
{ return f.linear().template segment<1>(axis); }
void RigidBody3DState::addForce( const Force& new_force )
{
m_forces.emplace_back( std::unique_ptr<Force>{ new_force.clone() } );
}
float Force::produitScalaire(Force vect)
{
return (getForce().x*vect.getForce().x+getForce().y*vect.getForce().y);
}
Force::Vector3 operator* (const Force & phi)
{
return phi.angular ();
}
static bool isApprox(const Force & self, const Force & other)
{ return self.isApprox(other); }
static void setRandom(Force & self) { self.setRandom(); }
static void setAngular(Force & self, const Vector3 & n) { self.angular(n); }
int main(int argc, char **argv)
{
//Common miniMD settings
In in;
in.datafile = NULL;
int me=0; //local MPI rank
int nprocs=1; //number of MPI ranks
int num_threads=32; //number of Threads per Block threads
int num_steps=-1; //number of timesteps (if -1 use value from lj.in)
int system_size=-1; //size of the system (if -1 use value from lj.in)
int check_safeexchange=0; //if 1 complain if atom moves further than 1 subdomain length between exchanges
int do_safeexchange=0; //if 1 use safe exchange mode [allows exchange over multiple subdomains]
int use_sse=0; //setting for SSE variant of miniMD only
int screen_yaml=0; //print yaml output to screen also
int yaml_output=0; //print yaml output
int halfneigh=0; //1: use half neighborlist; 0: use full neighborlist; -1: use original miniMD version half neighborlist force
char* input_file = NULL;
int ghost_newton = 0;
int skip_gpu = 999;
int neighbor_size = -1;
//OpenCL specific
int platform = 0;
int device = 0;
int subdevice = -1;
int ppn = 2;
int use_tex = 0;
int threads_per_atom = 1;
int map_device=0;
for(int i = 0; i < argc; i++) {
if((strcmp(argv[i], "-i") == 0) || (strcmp(argv[i], "--input_file") == 0)) {
input_file = argv[++i];
continue;
}
if((strcmp(argv[i],"-p")==0)||(strcmp(argv[i],"--platform")==0)) {platform=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-d")==0)||(strcmp(argv[i],"--device")==0)) {device=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-sd")==0)||(strcmp(argv[i],"--subdevice")==0)) {subdevice=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-sd_map")==0)||(strcmp(argv[i],"--subdevice_mapping")==0)) {subdevice=1-me%ppn; continue;}
if((strcmp(argv[i],"-ng")==0)||(strcmp(argv[i],"--num_gpus")==0)) {ppn=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-dm")==0)||(strcmp(argv[i],"--device_map")==0)) {map_device=1; continue;}
}
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &me);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
if(map_device) {device = me%ppn; if(device>=skip_gpu) device++;}
OpenCLWrapper* opencl = new OpenCLWrapper;
if( me == 0)
printf("# Platforms: %i\n",opencl->num_platforms);
printf("# Proc: %i using device %i\n",me,device);
opencl->Init(argc,argv,device,device+1,NULL,platform,subdevice);
int error = 0;
if(input_file == NULL)
error = input(in, "in.lj.miniMD");
else
error = input(in, input_file);
if (error)
{
MPI_Finalize();
exit(0);
}
for(int i=0;i<argc;i++)
{
if((strcmp(argv[i],"-t")==0)||(strcmp(argv[i],"--num_threads")==0)) {num_threads=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-n")==0)||(strcmp(argv[i],"--nsteps")==0)) {num_steps=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-s")==0)||(strcmp(argv[i],"--size")==0)) {system_size=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"--half_neigh")==0)) {halfneigh=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-sse")==0)) {use_sse=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"--check_exchange")==0)) {check_safeexchange=1; continue;}
if((strcmp(argv[i],"-o")==0)||(strcmp(argv[i],"--yaml_output")==0)) {yaml_output=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"--yaml_screen")==0)) {screen_yaml=1; continue;}
if((strcmp(argv[i], "-f") == 0) || (strcmp(argv[i], "--data_file") == 0)) {
if(in.datafile == NULL) in.datafile = new char[1000];
strcpy(in.datafile, argv[++i]);
continue;
}
if((strcmp(argv[i], "-u") == 0) || (strcmp(argv[i], "--units") == 0)) {
in.units = strcmp(argv[++i], "metal") == 0 ? 1 : 0;
continue;
}
if((strcmp(argv[i], "-p") == 0) || (strcmp(argv[i], "--force") == 0)) {
in.forcetype = strcmp(argv[++i], "eam") == 0 ? FORCEEAM : FORCELJ;
continue;
}
if((strcmp(argv[i], "-gn") == 0) || (strcmp(argv[i], "--ghost_newton") == 0)) {
ghost_newton = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "--skip_gpu") == 0)) {
skip_gpu = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-b") == 0) || (strcmp(argv[i], "--neigh_bins") == 0)) {
neighbor_size = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i],"-tex")==0)||(strcmp(argv[i],"--texture")==0)) {use_tex=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-tpa")==0)) {threads_per_atom=atoi(argv[++i]); continue;}
if((strcmp(argv[i],"-h")==0)||(strcmp(argv[i],"--help")==0))
{
printf("\n---------------------------------------------------------\n");
printf("-------------" VARIANT_STRING "------------\n");
printf("---------------------------------------------------------\n\n");
printf("miniMD is a simple, parallel molecular dynamics (MD) code,\n"
"which is part of the Mantevo project at Sandia National\n"
"Laboratories ( http://www.mantevo.org ).\n"
"The original authors of MPI based miniMD are Steve Plimpton ([email protected]) ,\n"
"Paul Crozier ([email protected]) with current versions \n"
"written by Christian Trott ([email protected]).\n\n");
printf("Commandline Options:\n");
printf("\n Execution configuration:\n");
printf("\t-t / --num_threads <threads>: set number of threads per block (default 32)\n");
printf("\t--half_neigh <int>: use half neighborlists (default 0)\n"
"\t 0: full neighborlist\n"
"\t 1: half neighborlist (not supported in OpenCL variant)\n"
"\t -1: original miniMD half neighborlist force \n"
"\t (not supported in OpenCL variant)\n");
printf("\t-d / --device <int>: select device (default 0)\n");
printf("\t-dm / --device_map: map devices to MPI ranks\n");
printf("\t-ng / --num_gpus <int>: give number of GPUs per Node (used in conjuction with -dm\n"
"\t to determine device id: 'id=mpi_rank%%ng' (default 2)\n");
printf("\t--skip_gpu <int>: skip the specified gpu when assigning devices to MPI ranks\n"
"\t used in conjunction with -dm (but must come first in arg list)\n");
printf("\t-p / --platform <int>: select platform (default 0)\n");
printf("\t-sse <sse_version>: use explicit sse intrinsics (use miniMD-SSE variant)\n");
printf("\t-gn / --ghost_newton <int>: set usage of newtons third law for ghost atoms\n"
"\t (only applicable with half neighborlists)\n");
printf("\n Simulation setup:\n");
printf("\t-i / --input_file <string>: set input file to be used (default: in.lj.miniMD)\n");
printf("\t-n / --nsteps <nsteps>: set number of timesteps for simulation\n");
printf("\t-s / --size <size>: set linear dimension of systembox and neighbor bins\n");
printf("\t-b / --neigh_bins <int>: set linear dimension of neighbor bin grid\n");
printf("\t-u / --units <string>: set units (lj or metal), see LAMMPS documentation\n");
printf("\t-p / --force <string>: set interaction model (lj or eam)\n");
printf("\t-f / --data_file <string>: read configuration from LAMMPS data file\n");
printf("\n Miscelaneous:\n");
printf("\t--check_exchange: check whether atoms moved further than subdomain width\n");
printf("\t--safe_exchange: perform exchange communication with all MPI processes\n"
"\t within rcut_neighbor (outer force cutoff)\n");
printf("\t--yaml_output <int>: level of yaml output (default 0)\n");
printf("\t--yaml_screen: write yaml output also to screen\n");
printf("\t-tex / --texture <int>: use texture cache in force kernel (default 0)\n");
printf("\t-h / --help: display this help message\n\n");
printf("---------------------------------------------------------\n\n");
exit(0);
}
}
Atom atom;
Force force;
Neighbor neighbor;
Integrate integrate;
Thermo thermo;
Comm comm;
Timer timer;
ThreadData threads;
if(in.forcetype == FORCEEAM) {
printf("ERROR: " VARIANT_STRING " does not yet support EAM simulations. Exiting.\n");
MPI_Finalize();
exit(0);
}
if(ghost_newton!=0)
{
if(me ==0 ) printf("ERROR: -ghost_newton %i is not supported in " VARIANT_STRING ". Exiting.\n",ghost_newton);
MPI_Finalize();
exit(0);
}
if(halfneigh!=0)
{
if(me ==0 ) printf("ERROR: -half_neigh %i is not supported in " VARIANT_STRING ". Exiting.\n",halfneigh);
MPI_Finalize();
exit(0);
}
if(halfneigh!=0)
{
if(me ==0 ) printf("ERROR: -half_neigh %i is not supported in " VARIANT_STRING ". Exiting.\n",halfneigh);
MPI_Finalize();
exit(0);
}
if(use_tex!=0)
{
if(me ==0 ) printf("ERROR: -tex %i is currently broken. Exiting.\n",use_tex);
MPI_Finalize();
exit(0);
}
if(use_sse)
{
#ifndef VARIANT_SSE
if(me ==0 ) printf("ERROR: Trying to run with -sse with miniMD reference version. Use SSE variant instead. Exiting.\n");
MPI_Finalize();
exit(0);
#endif
}
threads.mpi_me=me;
threads.mpi_num_threads=nprocs;
threads.omp_me=0;
threads.omp_num_threads=num_threads;
atom.threads = &threads;
comm.threads = &threads;
force.threads = &threads;
integrate.threads = &threads;
neighbor.threads = &threads;
thermo.threads = &threads;
opencl->blockdim = num_threads;
atom.threads_per_atom = threads_per_atom;
atom.use_tex = use_tex;
comm.do_safeexchange=do_safeexchange;
force.use_sse=use_sse;
neighbor.halfneigh=halfneigh;
compile_kernels(opencl);
integrate.opencl = opencl;
force.opencl = opencl;
neighbor.opencl = opencl;
atom.opencl = opencl;
comm.opencl = opencl;
if(num_steps > 0) in.ntimes = num_steps;
if(system_size > 0) {
in.nx = system_size;
in.ny = system_size;
in.nz = system_size;
}
if(neighbor_size > 0) {
neighbor.nbinx = neighbor_size;
neighbor.nbiny = neighbor_size;
neighbor.nbinz = neighbor_size;
}
if(neighbor_size < 0 && in.datafile == NULL) {
MMD_float neighscale = 5.0 / 6.0;
neighbor.nbinx = neighscale * in.nx;
neighbor.nbiny = neighscale * in.ny;
neighbor.nbinz = neighscale * in.ny;
}
if(neighbor_size < 0 && in.datafile)
neighbor.nbinx = -1;
if(neighbor.nbinx == 0) neighbor.nbinx = 1;
if(neighbor.nbiny == 0) neighbor.nbiny = 1;
if(neighbor.nbinz == 0) neighbor.nbinz = 1;
integrate.ntimes = in.ntimes;
integrate.dt = in.dt;
neighbor.every = in.neigh_every;
neighbor.cutneigh = in.neigh_cut;
force.cutforce = in.force_cut;
thermo.nstat = in.thermo_nstat;
if(me == 0)
printf("# Create System:\n");
if(in.datafile) {
read_lammps_data(atom, comm, neighbor, integrate, thermo, in.datafile, in.units);
MMD_float volume = atom.box.xprd * atom.box.yprd * atom.box.zprd;
in.rho = 1.0 * atom.natoms / volume;
force.setup();
} else {
create_box(atom, in.nx, in.ny, in.nz, in.rho);
comm.setup(neighbor.cutneigh, atom);
neighbor.setup(atom);
integrate.setup();
force.setup();
create_atoms(atom, in.nx, in.ny, in.nz, in.rho);
thermo.setup(in.rho, integrate, atom, in.units);
create_velocity(in.t_request, atom, thermo);
}
if(me == 0)
printf("# Done .... \n");
if(me == 0) {
fprintf(stdout, "# " VARIANT_STRING " output ...\n");
fprintf(stdout, "# Systemparameters: \n");
fprintf(stdout, "\t# MPI processes: %i\n", neighbor.threads->mpi_num_threads);
fprintf(stdout, "\t# OpenMP threads: %i\n", neighbor.threads->omp_num_threads);
fprintf(stdout, "\t# Inputfile: %s\n", input_file == 0 ? "in.lj.miniMD" : input_file);
fprintf(stdout, "\t# Datafile: %s\n", in.datafile ? in.datafile : "None");
fprintf(stdout, "\t# ForceStyle: %s\n", in.forcetype == FORCELJ ? "LJ" : "EAM");
fprintf(stdout, "\t# Units: %s\n", in.units == 0 ? "LJ" : "METAL");
fprintf(stdout, "\t# Atoms: %i\n", atom.natoms);
fprintf(stdout, "\t# System size: %2.2lf %2.2lf %2.2lf (unit cells: %i %i %i)\n", atom.box.xprd, atom.box.yprd, atom.box.zprd, in.nx, in.ny, in.nz);
fprintf(stdout, "\t# Density: %lf\n", in.rho);
fprintf(stdout, "\t# Force cutoff: %lf\n", force.cutforce);
fprintf(stdout, "\t# Neigh cutoff: %lf\n", neighbor.cutneigh);
fprintf(stdout, "\t# Half neighborlists: %i\n", neighbor.halfneigh);
fprintf(stdout, "\t# Neighbor bins: %i %i %i\n", neighbor.nbinx, neighbor.nbiny, neighbor.nbinz);
fprintf(stdout, "\t# Neighbor frequency: %i\n", neighbor.every);
fprintf(stdout, "\t# Timestep size: %lf\n", integrate.dt);
fprintf(stdout, "\t# Thermo frequency: %i\n", thermo.nstat);
fprintf(stdout, "\t# Ghost Newton: %i\n", ghost_newton);
fprintf(stdout, "\t# Use SSE intrinsics: %i\n", force.use_sse);
fprintf(stdout, "\t# Do safe exchange: %i\n", comm.do_safeexchange);
fprintf(stdout, "\t# Size of float: %i\n\n",sizeof(MMD_float));
}
comm.exchange(atom);
comm.borders(atom);
atom.d_x->upload();
atom.d_v->upload();
//atom.d_vold->upload();
neighbor.build(atom);
if (me == 0) printf("# Starting dynamics ...\n");
if (me == 0) printf("# Timestep T U P Time\n");
thermo.compute(0,atom,neighbor,force,timer,comm);
force.compute(atom,neighbor,comm.me);
timer.barrier_start(TIME_TOTAL);
integrate.run(atom,force,neighbor,comm,thermo,timer);
timer.barrier_stop(TIME_TOTAL);
int natoms;
MPI_Allreduce(&atom.nlocal,&natoms,1,MPI_INT,MPI_SUM,MPI_COMM_WORLD);
thermo.compute(-1,atom,neighbor,force,timer,comm);
if(me == 0) {
double time_other=timer.array[TIME_TOTAL]-timer.array[TIME_FORCE]-timer.array[TIME_NEIGH]-timer.array[TIME_COMM];
printf("\n\n");
printf("# Performance Summary:\n");
printf("# MPI_proc OMP_threads nsteps natoms t_total t_force t_neigh t_comm t_other performance perf/thread grep_string t_extra\n");
printf("%i %i %i %i %lf %lf %lf %lf %lf %lf %lf PERF_SUMMARY %lf\n\n\n",
nprocs,num_threads,integrate.ntimes,natoms,
timer.array[TIME_TOTAL],timer.array[TIME_FORCE],timer.array[TIME_NEIGH],timer.array[TIME_COMM],time_other,
1.0*natoms*integrate.ntimes/timer.array[TIME_TOTAL],1.0*natoms*integrate.ntimes/timer.array[TIME_TOTAL]/nprocs/num_threads,timer.array[TIME_TEST]);
}
if(yaml_output)
output(in,atom,force,neighbor,comm,thermo,integrate,timer,screen_yaml);
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
delete opencl;
return 0;
}
int main(int argc, char** argv)
{
In in;
in.datafile = NULL;
int me = 0; //local MPI rank
int nprocs = 1; //number of MPI ranks
int num_threads = 1; //number of OpenMP threads
int num_steps = -1; //number of timesteps (if -1 use value from lj.in)
int system_size = -1; //size of the system (if -1 use value from lj.in)
int nx = -1;
int ny = -1;
int nz = -1;
int check_safeexchange = 0; //if 1 complain if atom moves further than 1 subdomain length between exchanges
int do_safeexchange = 0; //if 1 use safe exchange mode [allows exchange over multiple subdomains]
int use_sse = 0; //setting for SSE variant of miniMD only
int screen_yaml = 0; //print yaml output to screen also
int yaml_output = 0; //print yaml output
int halfneigh = 1; //1: use half neighborlist; 0: use full neighborlist; -1: use original miniMD version half neighborlist force
int teams = 1;
int device = 0;
int neighbor_size = -1;
char* input_file = NULL;
int ghost_newton = 1;
int sort = -1;
for(int i = 0; i < argc; i++) {
if((strcmp(argv[i], "-i") == 0) || (strcmp(argv[i], "--input_file") == 0)) {
input_file = argv[++i];
continue;
}
}
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &me);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
int error = 0;
if(input_file == NULL)
error = input(in, "in.lj.miniMD");
else
error = input(in, input_file);
if(error) {
MPI_Finalize();
exit(0);
}
for(int i = 0; i < argc; i++) {
if((strcmp(argv[i], "-t") == 0) || (strcmp(argv[i], "--num_threads") == 0)) {
num_threads = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "--teams") == 0)) {
teams = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-n") == 0) || (strcmp(argv[i], "--nsteps") == 0)) {
num_steps = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-s") == 0) || (strcmp(argv[i], "--size") == 0)) {
system_size = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-nx") == 0)) {
nx = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-ny") == 0)) {
ny = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-nz") == 0)) {
nz = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-b") == 0) || (strcmp(argv[i], "--neigh_bins") == 0)) {
neighbor_size = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "--half_neigh") == 0)) {
halfneigh = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-sse") == 0)) {
use_sse = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "--check_exchange") == 0)) {
check_safeexchange = 1;
continue;
}
if((strcmp(argv[i], "--sort") == 0)) {
sort = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-o") == 0) || (strcmp(argv[i], "--yaml_output") == 0)) {
yaml_output = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "--yaml_screen") == 0)) {
screen_yaml = 1;
continue;
}
if((strcmp(argv[i], "-f") == 0) || (strcmp(argv[i], "--data_file") == 0)) {
if(in.datafile == NULL) in.datafile = new char[1000];
strcpy(in.datafile, argv[++i]);
continue;
}
if((strcmp(argv[i], "-u") == 0) || (strcmp(argv[i], "--units") == 0)) {
in.units = strcmp(argv[++i], "metal") == 0 ? 1 : 0;
continue;
}
if((strcmp(argv[i], "-p") == 0) || (strcmp(argv[i], "--force") == 0)) {
in.forcetype = strcmp(argv[++i], "eam") == 0 ? FORCEEAM : FORCELJ;
continue;
}
if((strcmp(argv[i], "-gn") == 0) || (strcmp(argv[i], "--ghost_newton") == 0)) {
ghost_newton = atoi(argv[++i]);
continue;
}
if((strcmp(argv[i], "-h") == 0) || (strcmp(argv[i], "--help") == 0)) {
printf("\n-----------------------------------------------------------------------------------------------------------\n");
printf("-------------" VARIANT_STRING "--------------------\n");
printf("-------------------------------------------------------------------------------------------------------------\n\n");
printf("miniMD is a simple, parallel molecular dynamics (MD) code,\n"
"which is part of the Mantevo project at Sandia National\n"
"Laboratories ( http://www.mantevo.org ).\n"
"The original authors of miniMD are Steve Plimpton ([email protected]) ,\n"
"Paul Crozier ([email protected]) with current\n"
"versions written by Christian Trott ([email protected]).\n\n");
printf("Commandline Options:\n");
printf("\n Execution configuration:\n");
printf("\t--teams <nteams>: set number of thread-teams used per MPI rank (default 1)\n");
printf("\t-t / --num_threads <threads>: set number of threads per thread-team (default 1)\n");
printf("\t--half_neigh <int>: use half neighborlists (default 1)\n"
"\t 0: full neighborlist\n"
"\t 1: half neighborlist\n"
"\t -1: original miniMD half neighborlist force (not OpenMP safe)\n");
printf("\t-d / --device <int>: choose device to use (only applicable for GPU execution)\n");
printf("\t-dm / --device_map: map devices to MPI ranks\n");
printf("\t-ng / --num_gpus <int>: give number of GPUs per Node (used in conjuction with -dm\n"
"\t to determine device id: 'id=mpi_rank%%ng' (default 2)\n");
printf("\t--skip_gpu <int>: skip the specified gpu when assigning devices to MPI ranks\n"
"\t used in conjunction with -dm (but must come first in arg list)\n");
printf("\t-sse <sse_version>: use explicit sse intrinsics (use miniMD-SSE variant)\n");
printf("\t-gn / --ghost_newton <int>: set usage of newtons third law for ghost atoms\n"
"\t (only applicable with half neighborlists)\n");
printf("\n Simulation setup:\n");
printf("\t-i / --input_file <string>: set input file to be used (default: in.lj.miniMD)\n");
printf("\t-n / --nsteps <int>: set number of timesteps for simulation\n");
printf("\t-s / --size <int>: set linear dimension of systembox\n");
printf("\t-nx/-ny/-nz <int>: set linear dimension of systembox in x/y/z direction\n");
printf("\t-b / --neigh_bins <int>: set linear dimension of neighbor bin grid\n");
printf("\t-u / --units <string>: set units (lj or metal), see LAMMPS documentation\n");
printf("\t-p / --force <string>: set interaction model (lj or eam)\n");
printf("\t-f / --data_file <string>: read configuration from LAMMPS data file\n");
printf("\n Miscelaneous:\n");
printf("\t--check_exchange: check whether atoms moved further than subdomain width\n");
printf("\t--safe_exchange: perform exchange communication with all MPI processes\n"
"\t within rcut_neighbor (outer force cutoff)\n");
printf("\t--sort <n>: resort atoms (simple bins) every <n> steps (default: use reneigh frequency; never=0)");
printf("\t-o / --yaml_output <int>: level of yaml output (default 1)\n");
printf("\t--yaml_screen: write yaml output also to screen\n");
printf("\t-h / --help: display this help message\n\n");
printf("---------------------------------------------------------\n\n");
exit(0);
}
}
Atom atom;
Neighbor neighbor;
Integrate integrate;
Thermo thermo;
Comm comm;
Timer timer;
ThreadData threads;
Force* force;
if(in.forcetype == FORCEEAM) {
force = (Force*) new ForceEAM();
if(ghost_newton == 1) {
if(me == 0)
printf("# EAM currently requires '--ghost_newton 0'; Changing setting now.\n");
ghost_newton = 0;
}
}
if(in.forcetype == FORCELJ) force = (Force*) new ForceLJ();
threads.mpi_me = me;
threads.mpi_num_threads = nprocs;
threads.omp_me = 0;
threads.omp_num_threads = num_threads;
atom.threads = &threads;
comm.threads = &threads;
force->threads = &threads;
integrate.threads = &threads;
neighbor.threads = &threads;
thermo.threads = &threads;
force->epsilon = in.epsilon;
force->sigma = in.sigma;
force->sigma6 = in.sigma*in.sigma*in.sigma*in.sigma*in.sigma*in.sigma;
neighbor.ghost_newton = ghost_newton;
omp_set_num_threads(num_threads);
neighbor.timer = &timer;
force->timer = &timer;
comm.check_safeexchange = check_safeexchange;
comm.do_safeexchange = do_safeexchange;
force->use_sse = use_sse;
neighbor.halfneigh = halfneigh;
if(halfneigh < 0) force->use_oldcompute = 1;
if(use_sse) {
#ifdef VARIANT_REFERENCE
if(me == 0) printf("ERROR: Trying to run with -sse with miniMD reference version. Use SSE variant instead. Exiting.\n");
MPI_Finalize();
exit(0);
#endif
}
if(num_steps > 0) in.ntimes = num_steps;
if(system_size > 0) {
in.nx = system_size;
in.ny = system_size;
in.nz = system_size;
}
if(nx > 0) {
in.nx = nx;
if(ny > 0)
in.ny = ny;
else if(system_size < 0)
in.ny = nx;
if(nz > 0)
in.nz = nz;
else if(system_size < 0)
in.nz = nx;
}
if(neighbor_size > 0) {
neighbor.nbinx = neighbor_size;
neighbor.nbiny = neighbor_size;
neighbor.nbinz = neighbor_size;
}
if(neighbor_size < 0 && in.datafile == NULL) {
MMD_float neighscale = 5.0 / 6.0;
neighbor.nbinx = neighscale * in.nx;
neighbor.nbiny = neighscale * in.ny;
neighbor.nbinz = neighscale * in.nz;
}
if(neighbor_size < 0 && in.datafile)
neighbor.nbinx = -1;
if(neighbor.nbinx == 0) neighbor.nbinx = 1;
if(neighbor.nbiny == 0) neighbor.nbiny = 1;
if(neighbor.nbinz == 0) neighbor.nbinz = 1;
integrate.ntimes = in.ntimes;
integrate.dt = in.dt;
integrate.sort_every = sort>0?sort:(sort<0?in.neigh_every:0);
neighbor.every = in.neigh_every;
neighbor.cutneigh = in.neigh_cut;
force->cutforce = in.force_cut;
thermo.nstat = in.thermo_nstat;
if(me == 0)
printf("# Create System:\n");
if(in.datafile) {
read_lammps_data(atom, comm, neighbor, integrate, thermo, in.datafile, in.units);
MMD_float volume = atom.box.xprd * atom.box.yprd * atom.box.zprd;
in.rho = 1.0 * atom.natoms / volume;
force->setup();
if(in.forcetype == FORCEEAM) atom.mass = force->mass;
} else {
create_box(atom, in.nx, in.ny, in.nz, in.rho);
comm.setup(neighbor.cutneigh, atom);
neighbor.setup(atom);
integrate.setup();
force->setup();
if(in.forcetype == FORCEEAM) atom.mass = force->mass;
create_atoms(atom, in.nx, in.ny, in.nz, in.rho);
thermo.setup(in.rho, integrate, atom, in.units);
create_velocity(in.t_request, atom, thermo);
}
if(me == 0)
printf("# Done .... \n");
if(me == 0) {
fprintf(stdout, "# " VARIANT_STRING " output ...\n");
fprintf(stdout, "# Run Settings: \n");
fprintf(stdout, "\t# MPI processes: %i\n", neighbor.threads->mpi_num_threads);
fprintf(stdout, "\t# OpenMP threads: %i\n", neighbor.threads->omp_num_threads);
fprintf(stdout, "\t# Inputfile: %s\n", input_file == 0 ? "in.lj.miniMD" : input_file);
fprintf(stdout, "\t# Datafile: %s\n", in.datafile ? in.datafile : "None");
fprintf(stdout, "# Physics Settings: \n");
fprintf(stdout, "\t# ForceStyle: %s\n", in.forcetype == FORCELJ ? "LJ" : "EAM");
fprintf(stdout, "\t# Force Parameters: %2.2lf %2.2lf\n",in.epsilon,in.sigma);
fprintf(stdout, "\t# Units: %s\n", in.units == 0 ? "LJ" : "METAL");
fprintf(stdout, "\t# Atoms: %i\n", atom.natoms);
fprintf(stdout, "\t# System size: %2.2lf %2.2lf %2.2lf (unit cells: %i %i %i)\n", atom.box.xprd, atom.box.yprd, atom.box.zprd, in.nx, in.ny, in.nz);
fprintf(stdout, "\t# Density: %lf\n", in.rho);
fprintf(stdout, "\t# Force cutoff: %lf\n", force->cutforce);
fprintf(stdout, "\t# Timestep size: %lf\n", integrate.dt);
fprintf(stdout, "# Technical Settings: \n");
fprintf(stdout, "\t# Neigh cutoff: %lf\n", neighbor.cutneigh);
fprintf(stdout, "\t# Half neighborlists: %i\n", neighbor.halfneigh);
fprintf(stdout, "\t# Neighbor bins: %i %i %i\n", neighbor.nbinx, neighbor.nbiny, neighbor.nbinz);
fprintf(stdout, "\t# Neighbor frequency: %i\n", neighbor.every);
fprintf(stdout, "\t# Sorting frequency: %i\n", integrate.sort_every);
fprintf(stdout, "\t# Thermo frequency: %i\n", thermo.nstat);
fprintf(stdout, "\t# Ghost Newton: %i\n", ghost_newton);
fprintf(stdout, "\t# Use intrinsics: %i\n", force->use_sse);
fprintf(stdout, "\t# Do safe exchange: %i\n", comm.do_safeexchange);
fprintf(stdout, "\t# Size of float: %i\n\n", (int) sizeof(MMD_float));
}
comm.exchange(atom);
if(sort>0)
atom.sort(neighbor);
comm.borders(atom);
force->evflag = 1;
#pragma omp parallel
{
neighbor.build(atom);
force->compute(atom, neighbor, comm, me);
}
if(neighbor.halfneigh && neighbor.ghost_newton)
comm.reverse_communicate(atom);
if(me == 0) printf("# Starting dynamics ...\n");
if(me == 0) printf("# Timestep T U P Time\n");
#pragma omp parallel
{
thermo.compute(0, atom, neighbor, force, timer, comm);
}
timer.barrier_start(TIME_TOTAL);
integrate.run(atom, force, neighbor, comm, thermo, timer);
timer.barrier_stop(TIME_TOTAL);
int natoms;
MPI_Allreduce(&atom.nlocal, &natoms, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
force->evflag = 1;
force->compute(atom, neighbor, comm, me);
if(neighbor.halfneigh && neighbor.ghost_newton)
comm.reverse_communicate(atom);
thermo.compute(-1, atom, neighbor, force, timer, comm);
if(me == 0) {
double time_other = timer.array[TIME_TOTAL] - timer.array[TIME_FORCE] - timer.array[TIME_NEIGH] - timer.array[TIME_COMM];
printf("\n\n");
printf("# Performance Summary:\n");
printf("# MPI_proc OMP_threads nsteps natoms t_total t_force t_neigh t_comm t_other performance perf/thread grep_string t_extra\n");
printf("%i %i %i %i %lf %lf %lf %lf %lf %lf %lf PERF_SUMMARY %lf\n\n\n",
nprocs, num_threads, integrate.ntimes, natoms,
timer.array[TIME_TOTAL], timer.array[TIME_FORCE], timer.array[TIME_NEIGH], timer.array[TIME_COMM], time_other,
1.0 * natoms * integrate.ntimes / timer.array[TIME_TOTAL], 1.0 * natoms * integrate.ntimes / timer.array[TIME_TOTAL] / nprocs / num_threads, timer.array[TIME_TEST]);
}
if(yaml_output)
output(in, atom, force, neighbor, comm, thermo, integrate, timer, screen_yaml);
delete force;
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
return 0;
}
static void setZero(Force & self) { self.setZero(); }
void Element::addForce(Force force) {
velX += force.getVelX();
velY += force.getVelY();
}
static Vector6 getVector(const Force & self) { return self.toVector(); }
void GroupTTT::CalcReaction()
{
double SumTorque1=0;
for (list<Force *>::iterator f = s1->forces.begin(); f!=s1->forces.end(); f++)
{
Force *Fi = *f;
double T=0;
double Fx=0;
double Fy=0;
if(Fi->IsCalculatedt())
T=Fi->GetForcet();
if(Fi->IsCalculatedx())
Fx=Fi->GetForcex();
if(Fi->IsCalculatedy())
Fy=Fi->GetForcey();
double leverx = -Fi->linear->x.GetTF0()+A->linear.x.GetTF0();
double levery = -Fi->linear->y.GetTF0()+A->linear.y.GetTF0();
SumTorque1+=T+Fx*leverx-Fy*levery;
}
double SumTorque2=0;
/*
dd:=(yb[i]-ya[i])*(xb[i]-xc[i])+(yb[i]-yc[i])*(xa[i]-xb[i]);
dd1:=-k3*(xb[i]-xc[i])+k4*(xa[i]-xb[i]);
dd2:=k4*(ya[i]-yb[i])+k3*(yb[i]-yc[i]);
*/
for (list<Force *>::iterator f = s2->forces.begin(); f!=s2->forces.end(); f++)
{
Force *Fi = *f;
double T=0;
double Fx=0;
double Fy=0;
if(Fi->IsCalculatedt())
T=Fi->GetForcet();
if(Fi->IsCalculatedx())
Fx=Fi->GetForcex();
if(Fi->IsCalculatedy())
Fy=Fi->GetForcey();
double leverx = -Fi->linear->x.GetTF0()+C->linear.x.GetTF0();
double levery = -Fi->linear->y.GetTF0()+C->linear.y.GetTF0();
SumTorque2+=T+Fx*leverx-Fy*levery;
}
/*
dd:=(yb[i]-ya[i])*(xb[i]-xc[i])+(yb[i]-yc[i])*(xa[i]-xb[i]);
dd1:=-k3*(xb[i]-xc[i])+k4*(xa[i]-xb[i]);
dd2:=k4*(ya[i]-yb[i])+k3*(yb[i]-yc[i]);
*/
double leverxAB = A->linear.x.GetTF0()-B->linear.x.GetTF0();
double leveryAB = A->linear.y.GetTF0()-B->linear.y.GetTF0();
double leverxCB = C->linear.x.GetTF0()-B->linear.x.GetTF0();
double leveryCB = C->linear.y.GetTF0()-B->linear.y.GetTF0();
double d=leveryAB*leverxCB-leverxAB*leveryCB;
double d1=SumTorque1*leverxCB+SumTorque2*leverxAB;
double d2=SumTorque2*leveryAB+SumTorque1*leveryCB;
double R21x=d1/d;
double R21y=d2/d;
B->R1->SetForce(R21x,R21y,0);
B->R2->SetForce(-R21x,-R21y,0);
double SumFx1=0;
double SumFy1=0;
for (list<Force *>::iterator f = s1->forces.begin(); f!=s1->forces.end(); f++)
{
Force *Fi = *f;
if(Fi->IsCalculatedx())
SumFx1+=Fi->GetForcex();
if(Fi->IsCalculatedy())
SumFy1+=Fi->GetForcey();
}
double RAx=-SumFx1;
double RAy=-SumFy1;
A->R1->SetForce(RAx,RAy,0);
A->R2->SetForce(-RAx,-RAy,0);
double SumFx2=0;
double SumFy2=0;
for (list<Force *>::iterator f = s2->forces.begin(); f!=s2->forces.end(); f++)
{
Force *x = *f;
if(x->IsCalculatedx())
SumFx2+=x->GetForcex();
if(x->IsCalculatedy())
SumFy2+=x->GetForcey();
}
double RCx=-SumFx2;
double RCy=-SumFy2;
C->R2->SetForce(RCx,RCy,0);
C->R1->SetForce(-RCx,-RCy,0);
}
void RigidSphere::addForce(Force f) {
velocity += f.buildVector();
}
JointForce operator* (const Force& f) const
{ return ref.S.transpose()*f.toVector(); }
