tx258/lammps-gran-kokkos
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Enabled neighbor list build on the device with pair_style hybrid and hybrid/overlay

Browse Source
This commit is contained in:
Trung Nguyen
2019-04-16 23:30:25 -05:00
parent 898860328b
commit c55009a0ac
8 changed files with 80 additions and 27 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
lib/gpu/lal_device.cpp
View File

@ -301,16 +301,6 @@ int DeviceT::init(Answer<numtyp,acctyp> &ans, const bool charge,
if (!ans.init(ef_nlocal,charge,rot,*gpu))
return -3;
if (!nbor->init(&_neighbor_shared,ef_nlocal,host_nlocal,max_nbors,maxspecial,
*gpu,gpu_nbor,gpu_host,pre_cut, _block_cell_2d,
_block_cell_id, _block_nbor_build, threads_per_atom,
_warp_size, _time_device, compile_string()))
return -3;
if (_cell_size<0.0)
nbor->cell_size(cell_size,cell_size);
else
nbor->cell_size(_cell_size,cell_size);
_init_count++;
return 0;
}
@ -338,6 +328,39 @@ int DeviceT::init(Answer<numtyp,acctyp> &ans, const int nlocal,
return 0;
}
template <class numtyp, class acctyp>
int DeviceT::init_nbor(Neighbor *nbor, const int nlocal,
const int host_nlocal, const int nall,
const int maxspecial, const int gpu_host,
const int max_nbors, const double cell_size,
const bool pre_cut, const int threads_per_atom) {
int ef_nlocal=nlocal;
if (_particle_split<1.0 && _particle_split>0.0)
ef_nlocal=static_cast<int>(_particle_split*nlocal);
int gpu_nbor=0;
if (_gpu_mode==Device<numtyp,acctyp>::GPU_NEIGH)
gpu_nbor=1;
else if (_gpu_mode==Device<numtyp,acctyp>::GPU_HYB_NEIGH)
gpu_nbor=2;
#ifndef USE_CUDPP
if (gpu_nbor==1)
gpu_nbor=2;
#endif
if (!nbor->init(&_neighbor_shared,ef_nlocal,host_nlocal,max_nbors,maxspecial,
*gpu,gpu_nbor,gpu_host,pre_cut,_block_cell_2d,
_block_cell_id, _block_nbor_build, threads_per_atom,
_warp_size, _time_device, compile_string()))
return -3;
if (_cell_size<0.0)
nbor->cell_size(cell_size,cell_size);
else
nbor->cell_size(_cell_size,cell_size);
return 0;
}
template <class numtyp, class acctyp>
void DeviceT::set_single_precompute
(PPPM<numtyp,acctyp,float,_lgpu_float4> *pppm) {