ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@7180 f3b2605a-c512-4ea7-a41b-209d697bcdaa

Browse Source
This commit is contained in:
sjplimp
2011-10-24 17:49:41 +00:00
parent 5a03365db1
commit f3018d53bb
5 changed files with 50 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/USER-CUDA/atom_vec_atomic_cuda.cpp
View File

@ -286,14 +286,14 @@ int AtomVecAtomicCuda::pack_exchange(int dim, double *buf)
} }
if(max_nsend==0) grow_copylist(200); if(max_nsend==0) grow_copylist(200);
int nsend_atoms = Cuda_AtomVecAtomicCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer); int nsend_atoms = Cuda_AtomVecAtomicCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
if(nsend_atoms>max_nsend) {grow_copylist(nsend_atoms+100);} if(nsend_atoms>max_nsend) {grow_copylist(nsend_atoms+100);}
if(nsend_atoms*NCUDAEXCHANGE>*maxsend) if(nsend_atoms*NCUDAEXCHANGE>*maxsend)
{ {
grow_send((int) (nsend_atoms+100)*NCUDAEXCHANGE,buf_pointer,0); grow_send((int) (nsend_atoms+100)*NCUDAEXCHANGE,buf_pointer,0);
Cuda_AtomVecAtomicCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer); Cuda_AtomVecAtomicCuda_PackExchangeList(&cuda->shared_data,*maxsend,dim,*buf_pointer);
} }
int nlocal=atom->nlocal-nsend_atoms; int nlocal=atom->nlocal-nsend_atoms;
@ -395,6 +395,7 @@ int AtomVecAtomicCuda::unpack_exchange(double *buf)
} }
} }
cuda->shared_data.atom.nlocal=nlocal; cuda->shared_data.atom.nlocal=nlocal;
if(atom->nlocal!=nlocal)
cuda->shared_data.atom.update_nlocal=2; cuda->shared_data.atom.update_nlocal=2;
atom->nlocal=nlocal; atom->nlocal=nlocal;
mfirst+=m; mfirst+=m;

75
src/USER-CUDA/cuda.cpp
View File

@ -130,11 +130,11 @@ Cuda::Cuda(LAMMPS *lmp) : Pointers(lmp)
downloadtime=0; downloadtime=0;
dotiming=false; dotiming=false;
dotestatom = false; dotestatom = false;
testatom = 0; testatom = 0;
oncpu = true; oncpu = true;
self_comm = 0; self_comm = 0;
MYDBG( printf("# CUDA: Cuda::Cuda Done...\n");) MYDBG( printf("# CUDA: Cuda::Cuda Done...\n");)
//cCudaData<double, float, yx > //cCudaData<double, float, yx >
} }
@ -267,10 +267,10 @@ void Cuda::accelerator(int narg, char** arg)
cu_virial = 0; cu_virial = 0;
cu_eatom = 0; cu_eatom = 0;
cu_vatom = 0; cu_vatom = 0;
cu_radius = 0; cu_radius = 0;
cu_density = 0; cu_density = 0;
cu_omega = 0; cu_omega = 0;
cu_torque = 0; cu_torque = 0;
cu_special = 0; cu_special = 0;
cu_nspecial = 0; cu_nspecial = 0;
@ -299,8 +299,11 @@ void Cuda::setSharedDataZero()
shared_data.atom.q_flag = 0; shared_data.atom.q_flag = 0;
shared_data.atom.need_eatom = 0; shared_data.atom.need_eatom = 0;
shared_data.atom.need_vatom = 0; shared_data.atom.need_vatom = 0;
shared_data.atom.update_nmax = 1;
shared_data.atom.update_nlocal = 1;
shared_data.atom.update_neigh = 1;
shared_data.pair.cudable_force = 0; shared_data.pair.cudable_force = 0;
shared_data.pair.collect_forces_later = 0; shared_data.pair.collect_forces_later = 0;
shared_data.pair.use_block_per_atom = 0; shared_data.pair.use_block_per_atom = 0;
shared_data.pair.override_block_per_atom = -1; shared_data.pair.override_block_per_atom = -1;
@ -429,14 +432,6 @@ void Cuda::checkResize()
if(cu_atom->q_flag) if(cu_atom->q_flag)
{delete cu_q; cu_q = new cCudaData<double, F_FLOAT, x > ((double*)atom->q, & cu_atom->q , atom->nmax );}// cu_q->set_buffer(&(copy_buffer),&(copy_buffersize),true);} {delete cu_q; cu_q = new cCudaData<double, F_FLOAT, x > ((double*)atom->q, & cu_atom->q , atom->nmax );}// cu_q->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
/*
if(force->pair)
if(force->pair->eatom)
{delete cu_eatom; cu_eatom = new cCudaData<double, ENERGY_FLOAT, x > (force->pair->eatom, & cu_atom->eatom , atom->nmax );}// cu_eatom->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
if(force->pair)
if(force->pair->vatom)
{delete cu_vatom; cu_vatom = new cCudaData<double, ENERGY_FLOAT, yx > ((double*)force->pair->vatom, & cu_atom->vatom , atom->nmax,6 );}// cu_vatom->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
*/
if(atom->radius) if(atom->radius)
{ {
delete cu_radius; cu_radius = new cCudaData<double, X_FLOAT, x > (atom->radius , & cu_atom->radius , atom->nmax ); delete cu_radius; cu_radius = new cCudaData<double, X_FLOAT, x > (atom->radius , & cu_atom->radius , atom->nmax );
@ -444,11 +439,6 @@ void Cuda::checkResize()
delete cu_omega_rmass; cu_omega_rmass = new cCudaData<V_FLOAT, V_FLOAT, x> (omega_rmass , & cu_atom->omega_rmass , atom->nmax*4); delete cu_omega_rmass; cu_omega_rmass = new cCudaData<V_FLOAT, V_FLOAT, x> (omega_rmass , & cu_atom->omega_rmass , atom->nmax*4);
} }
/*
if(atom->density)
{delete cu_density; cu_density = new cCudaData<double, F_FLOAT, x > (atom->density , & cu_atom->density , atom->nmax );}
*/
if(atom->omega) if(atom->omega)
{delete cu_omega; cu_omega = new cCudaData<double, V_FLOAT, yx > (((double*) atom->omega) , & cu_atom->omega , atom->nmax,3 );} {delete cu_omega; cu_omega = new cCudaData<double, V_FLOAT, yx > (((double*) atom->omega) , & cu_atom->omega , atom->nmax,3 );}
@ -464,12 +454,10 @@ void Cuda::checkResize()
shared_data.atom.special_flag = neighbor->special_flag; shared_data.atom.special_flag = neighbor->special_flag;
shared_data.atom.molecular = atom->molecular; shared_data.atom.molecular = atom->molecular;
cu_atom->update_nmax = 2; cu_atom->update_nmax = 2;
cu_atom->nmax = atom->nmax; cu_atom->nmax = atom->nmax;
//delete [] x_type; x_type = new X_FLOAT4[atom->nmax];
delete cu_x_type; cu_x_type = new cCudaData<X_FLOAT, X_FLOAT, x> (x_type , & cu_atom->x_type , atom->nmax*4); delete cu_x_type; cu_x_type = new cCudaData<X_FLOAT, X_FLOAT, x> (x_type , & cu_atom->x_type , atom->nmax*4);
// shared_data.buffer_new = 2;
} }
if(((cu_xhold==NULL)||(cu_xhold->get_dim()[0]<neighbor->maxhold))&&neighbor->xhold) if(((cu_xhold==NULL)||(cu_xhold->get_dim()[0]<neighbor->maxhold))&&neighbor->xhold)
@ -488,6 +476,12 @@ void Cuda::checkResize()
{ {
cu_map_array = new cCudaData<int, int, x > (atom->get_map_array() , & cu_atom->map_array , atom->get_map_size() ); cu_map_array = new cCudaData<int, int, x > (atom->get_map_array() , & cu_atom->map_array , atom->get_map_size() );
} }
else
if(cu_map_array->dev_size()/sizeof(int)<atom->get_map_size())
{
delete cu_map_array;
cu_map_array = new cCudaData<int, int, x > (atom->get_map_array() , & cu_atom->map_array , atom->get_map_size() );
}
} }
@ -512,11 +506,6 @@ void Cuda::checkResize()
if(atom->radius) if(atom->radius)
if(cu_radius->get_host_data() != atom->radius) cu_radius->set_host_data((double*) (atom->radius)); if(cu_radius->get_host_data() != atom->radius) cu_radius->set_host_data((double*) (atom->radius));
/*
if(atom->density)
if(cu_density->get_host_data() != atom->density) cu_density->set_host_data((double*) (atom->density));
*/
if(atom->omega) if(atom->omega)
if(cu_omega->get_host_data() != atom->omega) cu_omega->set_host_data((double*) (atom->omega)); if(cu_omega->get_host_data() != atom->omega) cu_omega->set_host_data((double*) (atom->omega));
@ -558,7 +547,7 @@ void Cuda::evsetup_eatom_vatom(int eflag_atom,int vflag_atom)
if(not cu_vatom) if(not cu_vatom)
cu_vatom = new cCudaData<double, ENERGY_FLOAT, yx > ((double*)force->pair->vatom, & (shared_data.atom.vatom) , atom->nmax ,6 );// cu_vatom->set_buffer(&(copy_buffer),&(copy_buffersize),true);} cu_vatom = new cCudaData<double, ENERGY_FLOAT, yx > ((double*)force->pair->vatom, & (shared_data.atom.vatom) , atom->nmax ,6 );// cu_vatom->set_buffer(&(copy_buffer),&(copy_buffersize),true);}
cu_vatom->set_host_data((double*)force->pair->vatom); cu_vatom->set_host_data((double*)force->pair->vatom);
cu_vatom->memset_device(0); cu_vatom->memset_device(0);
} }
} }
@ -579,16 +568,9 @@ void Cuda::uploadAll()
cu_image->upload(); cu_image->upload();
if(shared_data.atom.q_flag) cu_q ->upload(); if(shared_data.atom.q_flag) cu_q ->upload();
//printf("A3\n");
//if(shared_data.atom.need_eatom) cu_eatom->upload();
//printf("A4\n");
//if(shared_data.atom.need_vatom) cu_vatom->upload();
//printf("A5\n");
if(atom->rmass) cu_rmass->upload(); if(atom->rmass) cu_rmass->upload();
if(atom->radius) cu_radius->upload(); if(atom->radius) cu_radius->upload();
// if(atom->density) cu_density->upload();
if(atom->omega) cu_omega->upload(); if(atom->omega) cu_omega->upload();
if(atom->torque) cu_torque->upload(); if(atom->torque) cu_torque->upload();
if(atom->special) cu_special->upload(); if(atom->special) cu_special->upload();
@ -631,7 +613,6 @@ void Cuda::downloadAll()
if(atom->rmass) cu_rmass->download(); if(atom->rmass) cu_rmass->download();
if(atom->radius) cu_radius->download(); if(atom->radius) cu_radius->download();
// if(atom->density) cu_density->download();
if(atom->omega) cu_omega->download(); if(atom->omega) cu_omega->download();
if(atom->torque) cu_torque->download(); if(atom->torque) cu_torque->download();
if(atom->special) cu_special->download(); if(atom->special) cu_special->download();
@ -747,13 +728,13 @@ void Cuda::setTimingsZero()
shared_data.cuda_timings.neigh_special = 0; shared_data.cuda_timings.neigh_special = 0;
//PPPM //PPPM
shared_data.cuda_timings.pppm_particle_map; shared_data.cuda_timings.pppm_particle_map = 0;
shared_data.cuda_timings.pppm_make_rho; shared_data.cuda_timings.pppm_make_rho = 0;
shared_data.cuda_timings.pppm_brick2fft; shared_data.cuda_timings.pppm_brick2fft = 0;
shared_data.cuda_timings.pppm_poisson; shared_data.cuda_timings.pppm_poisson = 0;
shared_data.cuda_timings.pppm_fillbrick; shared_data.cuda_timings.pppm_fillbrick = 0;
shared_data.cuda_timings.pppm_fieldforce; shared_data.cuda_timings.pppm_fieldforce = 0;
shared_data.cuda_timings.pppm_compute; shared_data.cuda_timings.pppm_compute = 0;
CudaWrapper_CheckUploadTime(true); CudaWrapper_CheckUploadTime(true);
CudaWrapper_CheckDownloadTime(true); CudaWrapper_CheckDownloadTime(true);
@ -789,8 +770,8 @@ void Cuda::print_timings()
printf(" Exchange MPI \t %lf \n",shared_data.cuda_timings.comm_exchange_mpi); printf(" Exchange MPI \t %lf \n",shared_data.cuda_timings.comm_exchange_mpi);
printf(" Exchange Kernel Pack \t %lf \n",shared_data.cuda_timings.comm_exchange_kernel_pack); printf(" Exchange Kernel Pack \t %lf \n",shared_data.cuda_timings.comm_exchange_kernel_pack);
printf(" Exchange Kernel Unpack \t %lf \n",shared_data.cuda_timings.comm_exchange_kernel_unpack); printf(" Exchange Kernel Unpack \t %lf \n",shared_data.cuda_timings.comm_exchange_kernel_unpack);
printf(" Exchange Kernel Fill \t %lf \n",shared_data.cuda_timings.comm_exchange_kernel_fill); printf(" Exchange Kernel Fill \t %lf \n",shared_data.cuda_timings.comm_exchange_kernel_fill);
printf(" Exchange CPU Pack \t %lf \n",shared_data.cuda_timings.comm_exchange_cpu_pack); printf(" Exchange CPU Pack \t %lf \n",shared_data.cuda_timings.comm_exchange_cpu_pack);
printf(" Exchange Upload \t %lf \n",shared_data.cuda_timings.comm_exchange_upload); printf(" Exchange Upload \t %lf \n",shared_data.cuda_timings.comm_exchange_upload);
printf(" Exchange Download \t %lf \n",shared_data.cuda_timings.comm_exchange_download); printf(" Exchange Download \t %lf \n",shared_data.cuda_timings.comm_exchange_download);
printf("\n"); printf("\n");

1
src/USER-CUDA/cuda_neigh_list.cpp
View File

@ -111,6 +111,7 @@ void CudaNeighList::dev_alloc()
neighbors_inner = new int[sneighlist.maxlocal*sneighlist.maxneighbors]; neighbors_inner = new int[sneighlist.maxlocal*sneighlist.maxneighbors];
cu_neighbors_inner = new cCudaData<int, int, x> (neighbors_inner , & sneighlist.neighbors_inner , sneighlist.maxlocal*sneighlist.maxneighbors ); cu_neighbors_inner = new cCudaData<int, int, x> (neighbors_inner , & sneighlist.neighbors_inner , sneighlist.maxlocal*sneighlist.maxneighbors );
} }
cuda->shared_data.atom.update_neigh=2;
MYDBG( printf("# CUDA: CudaNeighList::dev_alloc() ... end\n"); ) MYDBG( printf("# CUDA: CudaNeighList::dev_alloc() ... end\n"); )
} }

1
src/USER-CUDA/neigh_full_cuda.cpp
View File

@ -250,6 +250,7 @@ void NeighborCuda::full_bin_cuda(NeighList *list)
}*/ }*/
list->cuda_list->cu_numneigh->download(); list->cuda_list->cu_numneigh->download();
list->cuda_list->cu_ilist->download(); list->cuda_list->cu_ilist->download();
cuda->shared_data.atom.update_neigh=2;
//printf("Done\n"); //printf("Done\n");
MYDBG(printf(" # CUDA::NeighFullBinCuda ... end\n");) MYDBG(printf(" # CUDA::NeighFullBinCuda ... end\n");)

37
src/USER-CUDA/verlet_cuda.cpp
View File

@ -564,6 +564,7 @@ void VerletCuda::run(int n)
cuda->shared_data.atom.reneigh_flag=0; cuda->shared_data.atom.reneigh_flag=0;
cuda->shared_data.atom.update_nlocal=1; cuda->shared_data.atom.update_nlocal=1;
cuda->shared_data.atom.update_nmax=1; cuda->shared_data.atom.update_nmax=1;
cuda->shared_data.atom.update_neigh=1;
cuda->shared_data.domain.update=1; cuda->shared_data.domain.update=1;
cuda->shared_data.buffer_new=1; cuda->shared_data.buffer_new=1;
cuda->uploadtime=0; cuda->uploadtime=0;
@ -627,14 +628,12 @@ void VerletCuda::run(int n)
//start force calculation asynchronus //start force calculation asynchronus
cuda->shared_data.comm.comm_phase=1; cuda->shared_data.comm.comm_phase=1;
// printf("Pre Force Compute\n");
force->pair->compute(eflag, vflag); force->pair->compute(eflag, vflag);
timer->stamp(TIME_PAIR); timer->stamp(TIME_PAIR);
//CudaWrapper_Sync(); //CudaWrapper_Sync();
//download comm buffers from GPU, perform MPI communication and upload buffers again //download comm buffers from GPU, perform MPI communication and upload buffers again
clock_gettime(CLOCK_REALTIME,&starttime); clock_gettime(CLOCK_REALTIME,&starttime);
// printf("Pre forward_comm(2)\n");
comm->forward_comm(2); comm->forward_comm(2);
clock_gettime(CLOCK_REALTIME,&endtime); clock_gettime(CLOCK_REALTIME,&endtime);
cuda->shared_data.cuda_timings.comm_forward_total+= cuda->shared_data.cuda_timings.comm_forward_total+=
@ -642,16 +641,13 @@ void VerletCuda::run(int n)
timer->stamp(TIME_COMM); timer->stamp(TIME_COMM);
//wait for force calculation //wait for force calculation
//printf("Pre Synch\n");
CudaWrapper_Sync(); CudaWrapper_Sync();
timer->stamp(TIME_PAIR); timer->stamp(TIME_PAIR);
//unpack communication buffers //unpack communication buffers
clock_gettime(CLOCK_REALTIME,&starttime); clock_gettime(CLOCK_REALTIME,&starttime);
// printf("Pre forward_comm(3)\n");
comm->forward_comm(3); comm->forward_comm(3);
clock_gettime(CLOCK_REALTIME,&endtime); clock_gettime(CLOCK_REALTIME,&endtime);
// printf("Post forward_comm(3)\n");
cuda->shared_data.cuda_timings.comm_forward_total+= cuda->shared_data.cuda_timings.comm_forward_total+=
endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000; endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000;
@ -663,11 +659,9 @@ void VerletCuda::run(int n)
else else
{ {
//perform standard forward communication //perform standard forward communication
//printf("Forward_comm\n");
clock_gettime(CLOCK_REALTIME,&starttime); clock_gettime(CLOCK_REALTIME,&starttime);
comm->forward_comm(); comm->forward_comm();
clock_gettime(CLOCK_REALTIME,&endtime); clock_gettime(CLOCK_REALTIME,&endtime);
//printf("Forward_comm_done\n");
cuda->shared_data.cuda_timings.comm_forward_total+= cuda->shared_data.cuda_timings.comm_forward_total+=
endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000; endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000;
timer->stamp(TIME_COMM); timer->stamp(TIME_COMM);
@ -677,13 +671,13 @@ void VerletCuda::run(int n)
else else
{ {
int nlocalold=cuda->shared_data.atom.nlocal; int nlocalold=cuda->shared_data.atom.nlocal;
//if(firstreneigh) if(firstreneigh)
{ {
cuda->shared_data.atom.update_nlocal=1; cuda->shared_data.atom.update_nlocal=1;
cuda->shared_data.atom.update_nmax=1; cuda->shared_data.atom.update_nmax=1;
firstreneigh=0; firstreneigh=0;
} }
cuda->shared_data.buffer_new=1; cuda->shared_data.buffer_new=1;
MYDBG( printf("# CUDA VerletCuda::iterate: neighbor\n"); ) MYDBG( printf("# CUDA VerletCuda::iterate: neighbor\n"); )
cuda->setDomainParams(); cuda->setDomainParams();
if(n_pre_exchange) modify->pre_exchange(); if(n_pre_exchange) modify->pre_exchange();
@ -759,10 +753,10 @@ void VerletCuda::run(int n)
cuda->shared_data.cuda_timings.test2+= cuda->shared_data.cuda_timings.test2+=
endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000; endtime.tv_sec-starttime.tv_sec+1.0*(endtime.tv_nsec-starttime.tv_nsec)/1000000000;
//rebuild neighbor list //rebuild neighbor list
test_atom(testatom,"Pre Neighbor"); test_atom(testatom,"Pre Neighbor");
neighbor->build(); neighbor->build();
timer->stamp(TIME_NEIGHBOR); timer->stamp(TIME_NEIGHBOR);
MYDBG( printf("# CUDA VerletCuda::iterate: neighbor done\n"); ) MYDBG( printf("# CUDA VerletCuda::iterate: neighbor done\n"); )
//if bonded interactions are used (in this case collect_forces_later is true), transfer data which only changes upon exchange/border routines from GPU to CPU //if bonded interactions are used (in this case collect_forces_later is true), transfer data which only changes upon exchange/border routines from GPU to CPU
@ -772,7 +766,7 @@ void VerletCuda::run(int n)
cuda->cu_tag->download(); cuda->cu_tag->download();
cuda->cu_type->download(); cuda->cu_type->download();
cuda->cu_mask->download(); cuda->cu_mask->download();
if(cuda->cu_q) cuda->cu_q->download(); if(cuda->cu_q) cuda->cu_q->download();
} }
cuda->shared_data.comm.comm_phase=3; cuda->shared_data.comm.comm_phase=3;
} }
@ -969,14 +963,16 @@ void VerletCuda::run(int n)
test_atom(testatom,"post output"); test_atom(testatom,"post output");
if(cuda->shared_data.atom.update_nlocal>0) if(cuda->shared_data.atom.update_nlocal>0)
cuda->shared_data.atom.update_nlocal--; cuda->shared_data.atom.update_nlocal--;
if(cuda->shared_data.atom.update_nmax>0) if(cuda->shared_data.atom.update_nmax>0)
cuda->shared_data.atom.update_nmax--; cuda->shared_data.atom.update_nmax--;
if(cuda->shared_data.domain.update>0) if(cuda->shared_data.atom.update_neigh>0)
cuda->shared_data.atom.update_neigh--;
if(cuda->shared_data.domain.update>0)
cuda->shared_data.domain.update--; cuda->shared_data.domain.update--;
if(cuda->shared_data.buffer_new>0) if(cuda->shared_data.buffer_new>0)
cuda->shared_data.buffer_new--; cuda->shared_data.buffer_new--;
cuda->shared_data.atom.reneigh_flag=0; cuda->shared_data.atom.reneigh_flag=0;
} }
@ -984,6 +980,7 @@ void VerletCuda::run(int n)
cuda->downloadAllNeighborLists(); cuda->downloadAllNeighborLists();
cuda->shared_data.atom.update_nlocal=1; cuda->shared_data.atom.update_nlocal=1;
cuda->shared_data.atom.update_nmax=1; cuda->shared_data.atom.update_nmax=1;
cuda->shared_data.atom.update_neigh=1;
cuda->shared_data.buffer_new=1; cuda->shared_data.buffer_new=1;
cuda->shared_data.domain.update=1; cuda->shared_data.domain.update=1;
cuda->oncpu = true; cuda->oncpu = true;