forked from lijiext/lammps
Updates to the core-shell gpu pair styles
This commit is contained in:
Trung Nguyen
parent
3ae8fdccd8
commit
a9b794b2ab
|
|
@ -12,12 +12,14 @@ pair_style born/omp command :h3
|
|||
pair_style born/gpu command :h3
|
||||
pair_style born/coul/long command :h3
|
||||
pair_style born/coul/long/cs command :h3
|
||||
pair_style born/coul/long/cs/gpu command :h3
|
||||
pair_style born/coul/long/gpu command :h3
|
||||
pair_style born/coul/long/omp command :h3
|
||||
pair_style born/coul/msm command :h3
|
||||
pair_style born/coul/msm/omp command :h3
|
||||
pair_style born/coul/wolf command :h3
|
||||
pair_style born/coul/wolf/cs command :h3
|
||||
pair_style born/coul/wolf/cs/gpu command :h3
|
||||
pair_style born/coul/wolf/gpu command :h3
|
||||
pair_style born/coul/wolf/omp command :h3
|
||||
pair_style born/coul/dsf command :h3
|
||||
|
|
|
|||
|
|
@ -20,6 +20,7 @@ pair_style coul/dsf/kk command :h3
|
|||
pair_style coul/dsf/omp command :h3
|
||||
pair_style coul/long command :h3
|
||||
pair_style coul/long/cs command :h3
|
||||
pair_style coul/long/cs/gpu command :h3
|
||||
pair_style coul/long/omp command :h3
|
||||
pair_style coul/long/gpu command :h3
|
||||
pair_style coul/long/kk command :h3
|
||||
|
|
|
|||
|
|
@ -81,7 +81,8 @@ OBJS = $(OBJ_DIR)/lal_atom.o $(OBJ_DIR)/lal_ans.o \
|
|||
$(OBJ_DIR)/lal_dipole_long_lj.o $(OBJ_DIR)/lal_dipole_long_lj_ext.o \
|
||||
$(OBJ_DIR)/lal_lj_expand_coul_long.o $(OBJ_DIR)/lal_lj_expand_coul_long_ext.o \
|
||||
$(OBJ_DIR)/lal_coul_long_cs.o $(OBJ_DIR)/lal_coul_long_cs_ext.o \
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs.o $(OBJ_DIR)/lal_born_coul_long_cs_ext.o
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs.o $(OBJ_DIR)/lal_born_coul_long_cs_ext.o \
|
||||
$(OBJ_DIR)/lal_born_coul_wolf_cs.o $(OBJ_DIR)/lal_born_coul_wolf_cs_ext.o
|
||||
|
||||
CBNS = $(OBJ_DIR)/device.cubin $(OBJ_DIR)/device_cubin.h \
|
||||
$(OBJ_DIR)/atom.cubin $(OBJ_DIR)/atom_cubin.h \
|
||||
|
|
@ -141,7 +142,8 @@ CBNS = $(OBJ_DIR)/device.cubin $(OBJ_DIR)/device_cubin.h \
|
|||
$(OBJ_DIR)/dipole_long_lj.cubin $(OBJ_DIR)/dipole_long_lj_cubin.h \
|
||||
$(OBJ_DIR)/lj_expand_coul_long.cubin $(OBJ_DIR)/lj_expand_coul_long_cubin.h \
|
||||
$(OBJ_DIR)/coul_long_cs.cubin $(OBJ_DIR)/coul_long_cs_cubin.h \
|
||||
$(OBJ_DIR)/born_coul_long_cs.cubin $(OBJ_DIR)/born_coul_long_cs_cubin.h
|
||||
$(OBJ_DIR)/born_coul_long_cs.cubin $(OBJ_DIR)/born_coul_long_cs_cubin.h \
|
||||
$(OBJ_DIR)/born_coul_wolf_cs.cubin $(OBJ_DIR)/born_coul_wolf_cs_cubin.h
|
||||
|
||||
all: $(OBJ_DIR) $(GPU_LIB) $(EXECS)
|
||||
|
||||
|
|
@ -847,10 +849,10 @@ $(OBJ_DIR)/coul_long_cs.cubin: lal_coul_long_cs.cu lal_precision.h lal_preproces
|
|||
$(OBJ_DIR)/coul_long_cs_cubin.h: $(OBJ_DIR)/coul_long_cs.cubin $(OBJ_DIR)/coul_long_cs.cubin
|
||||
$(BIN2C) -c -n coul_long_cs $(OBJ_DIR)/coul_long_cs.cubin > $(OBJ_DIR)/coul_long_cs_cubin.h
|
||||
|
||||
$(OBJ_DIR)/lal_coul_long_cs.o: $(ALL_H) lal_coul_long_cs.h lal_coul_long_cs.cpp $(OBJ_DIR)/coul_long_cs_cubin.h $(OBJ_DIR)/lal_base_charge.o
|
||||
$(OBJ_DIR)/lal_coul_long_cs.o: $(ALL_H) lal_coul_long_cs.h lal_coul_long_cs.cpp $(OBJ_DIR)/coul_long_cs_cubin.h $(OBJ_DIR)/lal_base_charge.o $(OBJ_DIR)/lal_coul_long.o
|
||||
$(CUDR) -o $@ -c lal_coul_long_cs.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/lal_coul_long_cs_ext.o: $(ALL_H) lal_coul_long_cs.h lal_coul_long_cs_ext.cpp lal_base_charge.h
|
||||
$(OBJ_DIR)/lal_coul_long_cs_ext.o: $(ALL_H) lal_coul_long_cs.h lal_coul_long_cs_ext.cpp lal_coul_long.h
|
||||
$(CUDR) -o $@ -c lal_coul_long_cs_ext.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/born_coul_long_cs.cubin: lal_born_coul_long_cs.cu lal_precision.h lal_preprocessor.h
|
||||
|
|
@ -859,12 +861,24 @@ $(OBJ_DIR)/born_coul_long_cs.cubin: lal_born_coul_long_cs.cu lal_precision.h lal
|
|||
$(OBJ_DIR)/born_coul_long_cs_cubin.h: $(OBJ_DIR)/born_coul_long_cs.cubin $(OBJ_DIR)/born_coul_long_cs.cubin
|
||||
$(BIN2C) -c -n born_coul_long_cs $(OBJ_DIR)/born_coul_long_cs.cubin > $(OBJ_DIR)/born_coul_long_cs_cubin.h
|
||||
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs.o: $(ALL_H) lal_born_coul_long_cs.h lal_born_coul_long_cs.cpp $(OBJ_DIR)/born_coul_long_cs_cubin.h $(OBJ_DIR)/lal_base_charge.o
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs.o: $(ALL_H) lal_born_coul_long_cs.h lal_born_coul_long_cs.cpp $(OBJ_DIR)/born_coul_long_cs_cubin.h $(OBJ_DIR)/lal_base_charge.o $(OBJ_DIR)/lal_born_coul_long.o
|
||||
$(CUDR) -o $@ -c lal_born_coul_long_cs.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs_ext.o: $(ALL_H) lal_born_coul_long_cs.h lal_born_coul_long_cs_ext.cpp lal_base_charge.h
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs_ext.o: $(ALL_H) lal_born_coul_long_cs.h lal_born_coul_long_cs_ext.cpp lal_born_coul_long.h
|
||||
$(CUDR) -o $@ -c lal_born_coul_long_cs_ext.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/born_coul_wolf_cs.cubin: lal_born_coul_wolf_cs.cu lal_precision.h lal_preprocessor.h
|
||||
$(CUDA) --cubin -DNV_KERNEL -o $@ lal_born_coul_wolf_cs.cu
|
||||
|
||||
$(OBJ_DIR)/born_coul_wolf_cs_cubin.h: $(OBJ_DIR)/born_coul_wolf_cs.cubin $(OBJ_DIR)/born_coul_wolf_cs.cubin
|
||||
$(BIN2C) -c -n born_coul_wolf_cs $(OBJ_DIR)/born_coul_wolf_cs.cubin > $(OBJ_DIR)/born_coul_wolf_cs_cubin.h
|
||||
|
||||
$(OBJ_DIR)/lal_born_coul_wolf_cs.o: $(ALL_H) lal_born_coul_wolf_cs.h lal_born_coul_wolf_cs.cpp $(OBJ_DIR)/born_coul_wolf_cs_cubin.h $(OBJ_DIR)/lal_base_charge.o $(OBJ_DIR)/lal_born_coul_wolf.o
|
||||
$(CUDR) -o $@ -c lal_born_coul_wolf_cs.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/lal_born_coul_wolf_cs_ext.o: $(ALL_H) lal_born_coul_wolf_cs.h lal_born_coul_wolf_cs_ext.cpp lal_born_coul_wolf.h
|
||||
$(CUDR) -o $@ -c lal_born_coul_wolf_cs_ext.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(BIN_DIR)/nvc_get_devices: ./geryon/ucl_get_devices.cpp $(NVD_H)
|
||||
$(CUDR) -o $@ ./geryon/ucl_get_devices.cpp -DUCL_CUDADR $(CUDA_LIB) -lcuda
|
||||
|
||||
|
|
|
|||
|
|
@ -70,7 +70,8 @@ OBJS = $(OBJ_DIR)/lal_atom.o $(OBJ_DIR)/lal_answer.o \
|
|||
$(OBJ_DIR)/lal_dipole_long_lj.o $(OBJ_DIR)/lal_dipole_long_lj_ext.o \
|
||||
$(OBJ_DIR)/lal_lj_expand_coul_long.o $(OBJ_DIR)/lal_lj_expand_coul_long_ext.o \
|
||||
$(OBJ_DIR)/lal_coul_long_cs.o $(OBJ_DIR)/lal_coul_long_cs_ext.o \
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs.o $(OBJ_DIR)/lal_born_coul_long_cs_ext.o
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs.o $(OBJ_DIR)/lal_born_coul_long_cs_ext.o \
|
||||
$(OBJ_DIR)/lal_born_coul_wolf_cs.o $(OBJ_DIR)/lal_born_coul_wolf_cs_ext.o
|
||||
|
||||
KERS = $(OBJ_DIR)/device_cl.h $(OBJ_DIR)/atom_cl.h \
|
||||
$(OBJ_DIR)/neighbor_cpu_cl.h $(OBJ_DIR)/pppm_cl.h \
|
||||
|
|
@ -101,7 +102,7 @@ KERS = $(OBJ_DIR)/device_cl.h $(OBJ_DIR)/atom_cl.h \
|
|||
$(OBJ_DIR)/lj_cubic_cl.h $(OBJ_DIR)/vashishta_cl.h \
|
||||
$(OBJ_DIR)/ufm_cl.h $(OBJ_DIR)/dipole_long_lj_cl.h \
|
||||
$(OBJ_DIR)/lj_expand_coul_long_cl.h $(OBJ_DIR)/coul_long_cs_cl.h \
|
||||
$(OBJ_DIR)/born_coul_long_cs_cl.h
|
||||
$(OBJ_DIR)/born_coul_long_cs_cl.h $(OBJ_DIR)/born_coul_wolf_cs_cl.h
|
||||
|
||||
|
||||
OCL_EXECS = $(BIN_DIR)/ocl_get_devices
|
||||
|
|
@ -615,21 +616,30 @@ $(OBJ_DIR)/lal_lj_expand_coul_long_ext.o: $(ALL_H) lal_lj_expand_coul_long.h lal
|
|||
$(OBJ_DIR)/coul_long_cs_cl.h: lal_coul_long_cs.cu $(PRE1_H)
|
||||
$(BSH) ./geryon/file_to_cstr.sh coul_long_cs $(PRE1_H) lal_coul_long_cs.cu $(OBJ_DIR)/coul_long_cs_cl.h;
|
||||
|
||||
$(OBJ_DIR)/lal_coul_long_cs.o: $(ALL_H) lal_coul_long_cs.h lal_coul_long_cs.cpp $(OBJ_DIR)/coul_long_cs_cl.h $(OBJ_DIR)/coul_long_cs_cl.h $(OBJ_DIR)/lal_base_charge.o
|
||||
$(OBJ_DIR)/lal_coul_long_cs.o: $(ALL_H) lal_coul_long_cs.h lal_coul_long_cs.cpp $(OBJ_DIR)/coul_long_cs_cl.h $(OBJ_DIR)/coul_long_cs_cl.h $(OBJ_DIR)/lal_base_charge.o $(OBJ_DIR)/lal_coul_long.o
|
||||
$(OCL) -o $@ -c lal_coul_long_cs.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/lal_coul_long_cs_ext.o: $(ALL_H) lal_coul_long_cs.h lal_coul_long_cs_ext.cpp lal_base_charge.h
|
||||
$(OBJ_DIR)/lal_coul_long_cs_ext.o: $(ALL_H) lal_coul_long_cs.h lal_coul_long_cs_ext.cpp lal_coul_long.h
|
||||
$(OCL) -o $@ -c lal_coul_long_cs_ext.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/born_coul_long_cs_cl.h: lal_born_coul_long_cs.cu $(PRE1_H)
|
||||
$(BSH) ./geryon/file_to_cstr.sh born_coul_long_cs $(PRE1_H) lal_born_coul_long_cs.cu $(OBJ_DIR)/born_coul_long_cs_cl.h;
|
||||
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs.o: $(ALL_H) lal_born_coul_long_cs.h lal_born_coul_long_cs.cpp $(OBJ_DIR)/born_coul_long_cs_cl.h $(OBJ_DIR)/born_coul_long_cs_cl.h $(OBJ_DIR)/lal_base_charge.o
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs.o: $(ALL_H) lal_born_coul_long_cs.h lal_born_coul_long_cs.cpp $(OBJ_DIR)/born_coul_long_cs_cl.h $(OBJ_DIR)/born_coul_long_cs_cl.h $(OBJ_DIR)/lal_base_charge.o $(OBJ_DIR)/lal_born_coul_long.o
|
||||
$(OCL) -o $@ -c lal_born_coul_long_cs.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs_ext.o: $(ALL_H) lal_born_coul_long_cs.h lal_born_coul_long_cs_ext.cpp lal_base_charge.h
|
||||
$(OBJ_DIR)/lal_born_coul_long_cs_ext.o: $(ALL_H) lal_born_coul_long_cs.h lal_born_coul_long_cs_ext.cpp lal_born_coul_long.h
|
||||
$(OCL) -o $@ -c lal_born_coul_long_cs_ext.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/born_coul_wolf_cs_cl.h: lal_born_coul_wolf_cs.cu $(PRE1_H)
|
||||
$(BSH) ./geryon/file_to_cstr.sh born_coul_wolf_cs $(PRE1_H) lal_born_coul_wolf_cs.cu $(OBJ_DIR)/born_coul_wolf_cs_cl.h;
|
||||
|
||||
$(OBJ_DIR)/lal_born_coul_wolf_cs.o: $(ALL_H) lal_born_coul_wolf_cs.h lal_born_coul_wolf_cs.cpp $(OBJ_DIR)/born_coul_wolf_cs_cl.h $(OBJ_DIR)/born_coul_wolf_cs_cl.h $(OBJ_DIR)/lal_base_charge.o $(OBJ_DIR)/lal_born_coul_wolf.o
|
||||
$(OCL) -o $@ -c lal_born_coul_wolf_cs.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(OBJ_DIR)/lal_born_coul_wolf_cs_ext.o: $(ALL_H) lal_born_coul_wolf_cs.h lal_born_coul_wolf_cs_ext.cpp lal_born_coul_wolf.h
|
||||
$(OCL) -o $@ -c lal_born_coul_wolf_cs_ext.cpp -I$(OBJ_DIR)
|
||||
|
||||
$(BIN_DIR)/ocl_get_devices: ./geryon/ucl_get_devices.cpp $(OCL_H)
|
||||
$(OCL) -o $@ ./geryon/ucl_get_devices.cpp -DUCL_OPENCL $(OCL_LINK)
|
||||
|
||||
|
|
|
|||
|
|
@ -78,7 +78,7 @@ class BornCoulLong : public BaseCharge<numtyp, acctyp> {
|
|||
|
||||
numtyp _cut_coulsq, _qqrd2e, _g_ewald;
|
||||
|
||||
private:
|
||||
protected:
|
||||
bool _allocated;
|
||||
void loop(const bool _eflag, const bool _vflag);
|
||||
};
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin :
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#ifdef USE_OPENCL
|
||||
|
|
@ -28,21 +28,6 @@ using namespace LAMMPS_AL;
|
|||
|
||||
extern Device<PRECISION,ACC_PRECISION> device;
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
BornCoulLongCST::BornCoulLongCS() : BaseCharge<numtyp,acctyp>(),
|
||||
_allocated(false) {
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
BornCoulLongCST::~BornCoulLongCS() {
|
||||
clear();
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
int BornCoulLongCST::bytes_per_atom(const int max_nbors) const {
|
||||
return this->bytes_per_atom_atomic(max_nbors);
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
int BornCoulLongCST::init(const int ntypes, double **host_cutsq, double **host_rhoinv,
|
||||
double **host_born1, double **host_born2, double **host_born3,
|
||||
|
|
@ -63,13 +48,13 @@ int BornCoulLongCST::init(const int ntypes, double **host_cutsq, double **host_r
|
|||
|
||||
// If atom type constants fit in shared memory use fast kernel
|
||||
int lj_types=ntypes;
|
||||
shared_types=false;
|
||||
this->shared_types=false;
|
||||
int max_shared_types=this->device->max_shared_types();
|
||||
if (lj_types<=max_shared_types && this->_block_size>=max_shared_types) {
|
||||
lj_types=max_shared_types;
|
||||
shared_types=true;
|
||||
this->shared_types=true;
|
||||
}
|
||||
_lj_types=lj_types;
|
||||
this->_lj_types=lj_types;
|
||||
|
||||
// Allocate a host write buffer for data initialization
|
||||
UCL_H_Vec<numtyp> host_write(lj_types*lj_types*32,*(this->ucl_device),
|
||||
|
|
@ -78,98 +63,33 @@ int BornCoulLongCST::init(const int ntypes, double **host_cutsq, double **host_r
|
|||
for (int i=0; i<lj_types*lj_types; i++)
|
||||
host_write[i]=0.0;
|
||||
|
||||
coeff1.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->atom->type_pack4(ntypes,lj_types,coeff1,host_write,host_rhoinv,
|
||||
this->coeff1.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->atom->type_pack4(ntypes,lj_types,this->coeff1,host_write,host_rhoinv,
|
||||
host_born1,host_born2,host_born3);
|
||||
|
||||
coeff2.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->atom->type_pack4(ntypes,lj_types,coeff2,host_write,host_a,host_c,
|
||||
this->coeff2.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->atom->type_pack4(ntypes,lj_types,this->coeff2,host_write,host_a,host_c,
|
||||
host_d,host_offset);
|
||||
|
||||
cutsq_sigma.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->atom->type_pack4(ntypes,lj_types,cutsq_sigma,host_write,host_cutsq,
|
||||
this->cutsq_sigma.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->atom->type_pack4(ntypes,lj_types,this->cutsq_sigma,host_write,host_cutsq,
|
||||
host_cut_ljsq,host_sigma);
|
||||
|
||||
sp_lj.alloc(8,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->sp_lj.alloc(8,*(this->ucl_device),UCL_READ_ONLY);
|
||||
for (int i=0; i<4; i++) {
|
||||
host_write[i]=host_special_lj[i];
|
||||
host_write[i+4]=host_special_coul[i];
|
||||
}
|
||||
ucl_copy(sp_lj,host_write,8,false);
|
||||
ucl_copy(this->sp_lj,host_write,8,false);
|
||||
|
||||
_cut_coulsq=host_cut_coulsq;
|
||||
_qqrd2e=qqrd2e;
|
||||
_g_ewald=g_ewald;
|
||||
this->_cut_coulsq=host_cut_coulsq;
|
||||
this->_qqrd2e=qqrd2e;
|
||||
this->_g_ewald=g_ewald;
|
||||
|
||||
_allocated=true;
|
||||
this->_max_bytes=coeff1.row_bytes()+coeff2.row_bytes()
|
||||
+cutsq_sigma.row_bytes()+sp_lj.row_bytes();
|
||||
this->_allocated=true;
|
||||
this->_max_bytes=this->coeff1.row_bytes()+this->coeff2.row_bytes()
|
||||
+this->cutsq_sigma.row_bytes()+this->sp_lj.row_bytes();
|
||||
return 0;
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
void BornCoulLongCST::clear() {
|
||||
if (!_allocated)
|
||||
return;
|
||||
_allocated=false;
|
||||
|
||||
coeff1.clear();
|
||||
coeff2.clear();
|
||||
cutsq_sigma.clear();
|
||||
sp_lj.clear();
|
||||
this->clear_atomic();
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
double BornCoulLongCST::host_memory_usage() const {
|
||||
return this->host_memory_usage_atomic()+sizeof(BornCoulLongCS<numtyp,acctyp>);
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Calculate energies, forces, and torques
|
||||
// ---------------------------------------------------------------------------
|
||||
template <class numtyp, class acctyp>
|
||||
void BornCoulLongCST::loop(const bool _eflag, const bool _vflag) {
|
||||
// Compute the block size and grid size to keep all cores busy
|
||||
const int BX=this->block_size();
|
||||
int eflag, vflag;
|
||||
if (_eflag)
|
||||
eflag=1;
|
||||
else
|
||||
eflag=0;
|
||||
|
||||
if (_vflag)
|
||||
vflag=1;
|
||||
else
|
||||
vflag=0;
|
||||
|
||||
int GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
|
||||
(BX/this->_threads_per_atom)));
|
||||
|
||||
int ainum=this->ans->inum();
|
||||
int nbor_pitch=this->nbor->nbor_pitch();
|
||||
this->time_pair.start();
|
||||
if (shared_types) {
|
||||
this->k_pair_fast.set_size(GX,BX);
|
||||
this->k_pair_fast.run(&this->atom->x, &coeff1, &coeff2, &sp_lj,
|
||||
&this->nbor->dev_nbor,
|
||||
&this->_nbor_data->begin(),
|
||||
&this->ans->force,
|
||||
&this->ans->engv, &eflag, &vflag,
|
||||
&ainum, &nbor_pitch, &this->atom->q,
|
||||
&cutsq_sigma, &_cut_coulsq, &_qqrd2e,
|
||||
&_g_ewald, &this->_threads_per_atom);
|
||||
} else {
|
||||
this->k_pair.set_size(GX,BX);
|
||||
this->k_pair.run(&this->atom->x, &coeff1, &coeff2, &_lj_types, &sp_lj,
|
||||
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
|
||||
&this->ans->force, &this->ans->engv,
|
||||
&eflag, &vflag, &ainum,
|
||||
&nbor_pitch, &this->atom->q,
|
||||
&cutsq_sigma, &_cut_coulsq,
|
||||
&_qqrd2e, &_g_ewald, &this->_threads_per_atom);
|
||||
}
|
||||
this->time_pair.stop();
|
||||
}
|
||||
|
||||
template class BornCoulLongCS<PRECISION,ACC_PRECISION>;
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
// __________________________________________________________________________
|
||||
//
|
||||
// begin : June 2018
|
||||
// email : trung.nguyen@northwestern.edu
|
||||
// email : ndactrung@gmail.com
|
||||
// ***************************************************************************/
|
||||
|
||||
#ifdef NV_KERNEL
|
||||
|
|
|
|||
|
|
@ -1,7 +1,7 @@
|
|||
/***************************************************************************
|
||||
born_coul_long.h
|
||||
born_coul_long_cs.h
|
||||
-------------------
|
||||
Trung Dac Nguyen (ORNL)
|
||||
Trung Dac Nguyen (Northwestern)
|
||||
|
||||
Class for acceleration of the born/coul/long/cs pair style.
|
||||
|
||||
|
|
@ -10,21 +10,21 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin :
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#ifndef LAL_BORN_COUL_LONG_CS_H
|
||||
#define LAL_BORN_COUL_LONG_CS_H
|
||||
|
||||
#include "lal_base_charge.h"
|
||||
#include "lal_born_coul_long.h"
|
||||
|
||||
namespace LAMMPS_AL {
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
class BornCoulLongCS : public BaseCharge<numtyp, acctyp> {
|
||||
class BornCoulLongCS : public BornCoulLong<numtyp, acctyp> {
|
||||
public:
|
||||
BornCoulLongCS();
|
||||
~BornCoulLongCS();
|
||||
BornCoulLongCS() {}
|
||||
~BornCoulLongCS() {}
|
||||
|
||||
/// Clear any previous data and set up for a new LAMMPS run
|
||||
/** \param max_nbors initial number of rows in the neighbor matrix
|
||||
|
|
@ -46,41 +46,6 @@ class BornCoulLongCS : public BaseCharge<numtyp, acctyp> {
|
|||
const double gpu_split, FILE *screen, double **host_cut_ljsq,
|
||||
const double host_cut_coulsq, double *host_special_coul,
|
||||
const double qqrd2e, const double g_ewald);
|
||||
|
||||
/// Clear all host and device data
|
||||
/** \note This is called at the beginning of the init() routine **/
|
||||
void clear();
|
||||
|
||||
/// Returns memory usage on device per atom
|
||||
int bytes_per_atom(const int max_nbors) const;
|
||||
|
||||
/// Total host memory used by library for pair style
|
||||
double host_memory_usage() const;
|
||||
|
||||
// --------------------------- TYPE DATA --------------------------
|
||||
|
||||
/// coeff1.x = rhoinv, coeff1.y = born1, coeff1.z = born2,
|
||||
/// coeff1.w = born3
|
||||
UCL_D_Vec<numtyp4> coeff1;
|
||||
/// coeff2.x = a, coeff2.y = c, coeff2.z = d, coeff2.w = offset
|
||||
UCL_D_Vec<numtyp4> coeff2;
|
||||
/// cutsq_sigma.x = cutsq, cutsq_sigma.y = cutsq_lj,
|
||||
/// cutsq_sigma.z = sigma
|
||||
UCL_D_Vec<numtyp4> cutsq_sigma;
|
||||
/// Special LJ values [0-3] and Special Coul values [4-7]
|
||||
UCL_D_Vec<numtyp> sp_lj;
|
||||
|
||||
/// If atom type constants fit in shared memory, use fast kernels
|
||||
bool shared_types;
|
||||
|
||||
/// Number of atom types
|
||||
int _lj_types;
|
||||
|
||||
numtyp _cut_coulsq, _qqrd2e, _g_ewald;
|
||||
|
||||
private:
|
||||
bool _allocated;
|
||||
void loop(const bool _eflag, const bool _vflag);
|
||||
};
|
||||
|
||||
}
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin :
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#include <iostream>
|
||||
|
|
|
|||
|
|
@ -57,7 +57,7 @@ int BornCoulWolfT::init(const int ntypes, double **host_cutsq, double **host_rho
|
|||
const double alf, const double e_shift, const double f_shift) {
|
||||
int success;
|
||||
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
|
||||
_screen,born_coul_wolf,"k_born_wolf");
|
||||
_screen,born_coul_wolf,"k_born_coul_wolf");
|
||||
if (success!=0)
|
||||
return success;
|
||||
|
||||
|
|
|
|||
|
|
@ -31,7 +31,7 @@ texture<int2> q_tex;
|
|||
|
||||
#define MY_PIS (acctyp)1.77245385090551602729
|
||||
|
||||
__kernel void k_born_wolf(const __global numtyp4 *restrict x_,
|
||||
__kernel void k_born_coul_wolf(const __global numtyp4 *restrict x_,
|
||||
const __global numtyp4 *restrict coeff1,
|
||||
const __global numtyp4 *restrict coeff2,
|
||||
const int lj_types,
|
||||
|
|
@ -165,7 +165,7 @@ __kernel void k_born_wolf(const __global numtyp4 *restrict x_,
|
|||
} // if ii
|
||||
}
|
||||
|
||||
__kernel void k_born_wolf_fast(const __global numtyp4 *restrict x_,
|
||||
__kernel void k_born_coul_wolf_fast(const __global numtyp4 *restrict x_,
|
||||
const __global numtyp4 *restrict coeff1_in,
|
||||
const __global numtyp4 *restrict coeff2_in,
|
||||
const __global numtyp *restrict sp_lj_in,
|
||||
|
|
|
|||
|
|
@ -13,8 +13,8 @@
|
|||
email : nguyentd@ornl.gov
|
||||
***************************************************************************/
|
||||
|
||||
#ifndef LAL_BORN_COUL_LONG_H
|
||||
#define LAL_BORN_COUL_LONG_H
|
||||
#ifndef LAL_BORN_COUL_WOLF_H
|
||||
#define LAL_BORN_COUL_WOLF_H
|
||||
|
||||
#include "lal_base_charge.h"
|
||||
|
||||
|
|
@ -79,7 +79,7 @@ class BornCoulWolf : public BaseCharge<numtyp, acctyp> {
|
|||
|
||||
numtyp _cut_coulsq,_qqrd2e,_alf,_e_shift,_f_shift;
|
||||
|
||||
private:
|
||||
protected:
|
||||
bool _allocated;
|
||||
void loop(const bool _eflag, const bool _vflag);
|
||||
};
|
||||
|
|
|
|||
|
|
@ -72,7 +72,7 @@ class CoulLong : public BaseCharge<numtyp, acctyp> {
|
|||
|
||||
numtyp _cut_coulsq, _qqrd2e, _g_ewald;
|
||||
|
||||
private:
|
||||
protected:
|
||||
bool _allocated;
|
||||
void loop(const bool _eflag, const bool _vflag);
|
||||
};
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin : June 2018
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#if defined(USE_OPENCL)
|
||||
|
|
@ -28,20 +28,6 @@ using namespace LAMMPS_AL;
|
|||
|
||||
extern Device<PRECISION,ACC_PRECISION> pair_gpu_device;
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
CoulLongCST::CoulLongCS() : BaseCharge<numtyp,acctyp>(), _allocated(false) {
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
CoulLongCST::~CoulLongCS() {
|
||||
clear();
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
int CoulLongCST::bytes_per_atom(const int max_nbors) const {
|
||||
return this->bytes_per_atom_atomic(max_nbors);
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
int CoulLongCST::init(const int ntypes, double **host_scale,
|
||||
const int nlocal, const int nall, const int max_nbors,
|
||||
|
|
@ -56,13 +42,13 @@ int CoulLongCST::init(const int ntypes, double **host_scale,
|
|||
return success;
|
||||
|
||||
int lj_types=ntypes;
|
||||
shared_types=false;
|
||||
this->shared_types=false;
|
||||
int max_shared_types=this->device->max_shared_types();
|
||||
if (lj_types<=max_shared_types && this->_block_size>=max_shared_types) {
|
||||
lj_types=max_shared_types;
|
||||
shared_types=true;
|
||||
this->shared_types=true;
|
||||
}
|
||||
_lj_types=lj_types;
|
||||
this->_lj_types=lj_types;
|
||||
|
||||
// Allocate a host write buffer for data initialization
|
||||
UCL_H_Vec<numtyp> host_write(lj_types*lj_types*32,*(this->ucl_device),
|
||||
|
|
@ -71,88 +57,22 @@ int CoulLongCST::init(const int ntypes, double **host_scale,
|
|||
for (int i=0; i<lj_types*lj_types; i++)
|
||||
host_write[i]=0.0;
|
||||
|
||||
scale.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->atom->type_pack1(ntypes,lj_types,scale,host_write,host_scale);
|
||||
this->scale.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->atom->type_pack1(ntypes,lj_types,this->scale,host_write,host_scale);
|
||||
|
||||
sp_cl.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
|
||||
this->sp_cl.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
|
||||
for (int i=0; i<4; i++) {
|
||||
host_write[i]=host_special_coul[i];
|
||||
}
|
||||
ucl_copy(sp_cl,host_write,4,false);
|
||||
ucl_copy(this->sp_cl,host_write,4,false);
|
||||
|
||||
_cut_coulsq=host_cut_coulsq;
|
||||
_qqrd2e=qqrd2e;
|
||||
_g_ewald=g_ewald;
|
||||
this->_cut_coulsq=host_cut_coulsq;
|
||||
this->_qqrd2e=qqrd2e;
|
||||
this->_g_ewald=g_ewald;
|
||||
|
||||
_allocated=true;
|
||||
this->_max_bytes=scale.row_bytes()+sp_cl.row_bytes();
|
||||
this->_allocated=true;
|
||||
this->_max_bytes=this->scale.row_bytes()+this->sp_cl.row_bytes();
|
||||
return 0;
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
void CoulLongCST::reinit(const int ntypes, double **host_scale) {
|
||||
UCL_H_Vec<numtyp> hscale(_lj_types*_lj_types,*(this->ucl_device),
|
||||
UCL_WRITE_ONLY);
|
||||
this->atom->type_pack1(ntypes,_lj_types,scale,hscale,host_scale);
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
void CoulLongCST::clear() {
|
||||
if (!_allocated)
|
||||
return;
|
||||
_allocated=false;
|
||||
|
||||
scale.clear();
|
||||
sp_cl.clear();
|
||||
this->clear_atomic();
|
||||
}
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
double CoulLongCST::host_memory_usage() const {
|
||||
return this->host_memory_usage_atomic()+sizeof(CoulLongCS<numtyp,acctyp>);
|
||||
}
|
||||
|
||||
// ---------------------------------------------------------------------------
|
||||
// Calculate energies, forces, and torques
|
||||
// ---------------------------------------------------------------------------
|
||||
template <class numtyp, class acctyp>
|
||||
void CoulLongCST::loop(const bool _eflag, const bool _vflag) {
|
||||
// Compute the block size and grid size to keep all cores busy
|
||||
const int BX=this->block_size();
|
||||
int eflag, vflag;
|
||||
if (_eflag)
|
||||
eflag=1;
|
||||
else
|
||||
eflag=0;
|
||||
|
||||
if (_vflag)
|
||||
vflag=1;
|
||||
else
|
||||
vflag=0;
|
||||
|
||||
int GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
|
||||
(BX/this->_threads_per_atom)));
|
||||
|
||||
int ainum=this->ans->inum();
|
||||
int nbor_pitch=this->nbor->nbor_pitch();
|
||||
this->time_pair.start();
|
||||
if (shared_types) {
|
||||
this->k_pair_fast.set_size(GX,BX);
|
||||
this->k_pair_fast.run(&this->atom->x, &scale, &sp_cl,
|
||||
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
|
||||
&this->ans->force, &this->ans->engv,
|
||||
&eflag, &vflag, &ainum, &nbor_pitch,
|
||||
&this->atom->q, &_cut_coulsq, &_qqrd2e, &_g_ewald,
|
||||
&this->_threads_per_atom);
|
||||
} else {
|
||||
this->k_pair.set_size(GX,BX);
|
||||
this->k_pair.run(&this->atom->x, &scale, &_lj_types, &sp_cl,
|
||||
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
|
||||
&this->ans->force, &this->ans->engv, &eflag, &vflag,
|
||||
&ainum, &nbor_pitch, &this->atom->q, &_cut_coulsq,
|
||||
&_qqrd2e, &_g_ewald, &this->_threads_per_atom);
|
||||
}
|
||||
this->time_pair.stop();
|
||||
}
|
||||
|
||||
template class CoulLongCS<PRECISION,ACC_PRECISION>;
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
// __________________________________________________________________________
|
||||
//
|
||||
// begin : June 2018
|
||||
// email : trung.nguyen@northwestern.edu
|
||||
// email : ndactrung@gmail.com
|
||||
// ***************************************************************************/
|
||||
|
||||
#ifdef NV_KERNEL
|
||||
|
|
|
|||
|
|
@ -10,21 +10,21 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin : June 2018
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#ifndef LAL_COUL_LONG_CS_H
|
||||
#define LAL_COUL_LONG_CS_H
|
||||
|
||||
#include "lal_base_charge.h"
|
||||
#include "lal_coul_long.h"
|
||||
|
||||
namespace LAMMPS_AL {
|
||||
|
||||
template <class numtyp, class acctyp>
|
||||
class CoulLongCS : public BaseCharge<numtyp, acctyp> {
|
||||
class CoulLongCS : public CoulLong<numtyp, acctyp> {
|
||||
public:
|
||||
CoulLongCS();
|
||||
~CoulLongCS();
|
||||
CoulLongCS() {}
|
||||
~CoulLongCS() {}
|
||||
|
||||
/// Clear any previous data and set up for a new LAMMPS run
|
||||
/** \param max_nbors initial number of rows in the neighbor matrix
|
||||
|
|
@ -43,38 +43,6 @@ class CoulLongCS : public BaseCharge<numtyp, acctyp> {
|
|||
const double gpu_split, FILE *screen,
|
||||
const double host_cut_coulsq, double *host_special_coul,
|
||||
const double qqrd2e, const double g_ewald);
|
||||
|
||||
/// Send updated coeffs from host to device (to be compatible with fix adapt)
|
||||
void reinit(const int ntypes, double **scale);
|
||||
|
||||
/// Clear all host and device data
|
||||
/** \note This is called at the beginning of the init() routine **/
|
||||
void clear();
|
||||
|
||||
/// Returns memory usage on device per atom
|
||||
int bytes_per_atom(const int max_nbors) const;
|
||||
|
||||
/// Total host memory used by library for pair style
|
||||
double host_memory_usage() const;
|
||||
|
||||
// --------------------------- TYPE DATA --------------------------
|
||||
|
||||
/// scale
|
||||
UCL_D_Vec<numtyp> scale;
|
||||
/// Special Coul values [0-3]
|
||||
UCL_D_Vec<numtyp> sp_cl;
|
||||
|
||||
/// If atom type constants fit in shared memory, use fast kernels
|
||||
bool shared_types;
|
||||
|
||||
/// Number of atom types
|
||||
int _lj_types;
|
||||
|
||||
numtyp _cut_coulsq, _qqrd2e, _g_ewald;
|
||||
|
||||
private:
|
||||
bool _allocated;
|
||||
void loop(const bool _eflag, const bool _vflag);
|
||||
};
|
||||
|
||||
}
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin : June 2018
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#include <iostream>
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin :
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#if defined(USE_OPENCL)
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
// __________________________________________________________________________
|
||||
//
|
||||
// begin :
|
||||
// email : trung.nguyen@northwestern.edu
|
||||
// email : ndactrung@gmail.com
|
||||
// ***************************************************************************/
|
||||
|
||||
#ifdef NV_KERNEL
|
||||
|
|
|
|||
|
|
@ -10,7 +10,7 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin :
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#ifndef LAL_LJ_EXPAND_COUL_LONG_H
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
/***************************************************************************
|
||||
lj_coul_long_ext.cpp
|
||||
lj_expand_coul_long_ext.cpp
|
||||
------------------------
|
||||
Trung Nguyen (Northwestern)
|
||||
|
||||
|
|
@ -10,7 +10,7 @@
|
|||
__________________________________________________________________________
|
||||
|
||||
begin :
|
||||
email : trung.nguyen@northwestern.edu
|
||||
email : ndactrung@gmail.com
|
||||
***************************************************************************/
|
||||
|
||||
#include <iostream>
|
||||
|
|
|
|||
|
|
@ -39,6 +39,8 @@ action pair_born_coul_long_cs_gpu.cpp pair_born_coul_long_cs.cpp
|
|||
action pair_born_coul_long_cs_gpu.h pair_born_coul_long_cs.cpp
|
||||
action pair_born_coul_wolf_gpu.cpp
|
||||
action pair_born_coul_wolf_gpu.h
|
||||
action pair_born_coul_wolf_cs_gpu.cpp pair_born_coul_wolf_cs.cpp
|
||||
action pair_born_coul_wolf_cs_gpu.h pair_born_coul_wolf_cs.cpp
|
||||
action pair_born_gpu.cpp
|
||||
action pair_born_gpu.h
|
||||
action pair_buck_coul_cut_gpu.cpp pair_buck_coul_cut.cpp
|
||||
|
|
|
|||
|
|
@ -0,0 +1,295 @@
|
|||
/* ----------------------------------------------------------------------
|
||||
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 authors: Trung Dac Nguyen (Northwestern)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include <cmath>
|
||||
#include <cstdio>
|
||||
#include <cstdlib>
|
||||
#include <cstring>
|
||||
#include "pair_born_coul_wolf_cs_gpu.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 "math_const.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "gpu_extra.h"
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
using namespace MathConst;
|
||||
|
||||
#define EPSILON 1.0e-20
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
int borncwcs_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
|
||||
double **host_born1, double **host_born2,
|
||||
double **host_born3, double **host_a, double **host_c,
|
||||
double **host_d, double **sigma, double **offset,
|
||||
double *special_lj, const int inum,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen,
|
||||
double **host_cut_ljsq, double host_cut_coulsq,
|
||||
double *host_special_coul, const double qqrd2e,
|
||||
const double alf, const double e_shift, const double f_shift);
|
||||
void borncwcs_gpu_clear();
|
||||
int ** borncwcs_gpu_compute_n(const int ago, const int inum_full, const int nall,
|
||||
double **host_x, int *host_type, double *sublo,
|
||||
double *subhi, tagint *tag, int **nspecial,
|
||||
tagint **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_q, double *boxlo,
|
||||
double *prd);
|
||||
void borncwcs_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_q,
|
||||
const int nlocal, double *boxlo, double *prd);
|
||||
double borncwcs_gpu_bytes();
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairBornCoulWolfCSGPU::PairBornCoulWolfCSGPU(LAMMPS *lmp) : PairBornCoulWolfCS(lmp),
|
||||
gpu_mode(GPU_FORCE)
|
||||
{
|
||||
respa_enable = 0;
|
||||
reinitflag = 0;
|
||||
cpu_time = 0.0;
|
||||
GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairBornCoulWolfCSGPU::~PairBornCoulWolfCSGPU()
|
||||
{
|
||||
borncwcs_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulWolfCSGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
int *ilist, *numneigh, **firstneigh;
|
||||
if (gpu_mode != GPU_FORCE) {
|
||||
inum = atom->nlocal;
|
||||
firstneigh = borncwcs_gpu_compute_n(neighbor->ago, inum, nall,
|
||||
atom->x, atom->type, domain->sublo,
|
||||
domain->subhi, atom->tag, atom->nspecial,
|
||||
atom->special, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start,
|
||||
&ilist, &numneigh, cpu_time, success,
|
||||
atom->q, domain->boxlo, domain->prd);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
borncwcs_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
|
||||
ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success, atom->q,
|
||||
atom->nlocal, domain->boxlo, domain->prd);
|
||||
}
|
||||
if (!success)
|
||||
error->one(FLERR,"Insufficient memory on accelerator");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulWolfCSGPU::init_style()
|
||||
{
|
||||
if (force->newton_pair)
|
||||
error->all(FLERR,
|
||||
"Cannot use newton pair with born/coul/wolf/cs/gpu pair style");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double maxcut = -1.0;
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
|
||||
cut = init_one(i,j);
|
||||
cut *= cut;
|
||||
if (cut > maxcut)
|
||||
maxcut = cut;
|
||||
cutsq[i][j] = cutsq[j][i] = cut;
|
||||
} else
|
||||
cutsq[i][j] = cutsq[j][i] = 0.0;
|
||||
}
|
||||
}
|
||||
double cell_size = sqrt(maxcut) + neighbor->skin;
|
||||
|
||||
cut_coulsq = cut_coul * cut_coul;
|
||||
|
||||
double e_shift = erfc(alf*cut_coul)/cut_coul;
|
||||
double f_shift = -(e_shift+ 2.0*alf/MY_PIS * exp(-alf*alf*cut_coul*cut_coul)) /
|
||||
cut_coul;
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
int success = borncwcs_gpu_init(atom->ntypes+1, cutsq, rhoinv,
|
||||
born1, born2, born3, a, c, d, sigma, offset,
|
||||
force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen, cut_ljsq,
|
||||
cut_coulsq, force->special_coul, force->qqrd2e,
|
||||
alf, e_shift, f_shift);
|
||||
GPU_EXTRA::check_flag(success,error,world);
|
||||
|
||||
if (gpu_mode == GPU_FORCE) {
|
||||
int irequest = neighbor->request(this,instance_me);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairBornCoulWolfCSGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + borncwcs_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairBornCoulWolfCSGPU::cpu_compute(int start, int inum, int eflag, int vflag,
|
||||
int *ilist, int *numneigh,
|
||||
int **firstneigh) {
|
||||
int i,j,ii,jj,jnum,itype,jtype;
|
||||
double xtmp,ytmp,ztmp,qtmp,delx,dely,delz,evdwl,ecoul,fpair;
|
||||
double rsq,r2inv,r6inv,forcecoul,forceborn,factor_coul,factor_lj;
|
||||
double erfcc,erfcd,v_sh,dvdrr,e_self,qisq;
|
||||
double prefactor;
|
||||
double r,rexp;
|
||||
int *jlist;
|
||||
|
||||
evdwl = ecoul = 0.0;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
double *q = atom->q;
|
||||
int *type = atom->type;
|
||||
int nlocal = atom->nlocal;
|
||||
double *special_coul = force->special_coul;
|
||||
double *special_lj = force->special_lj;
|
||||
double qqrd2e = force->qqrd2e;
|
||||
|
||||
double e_shift = erfc(alf*cut_coul)/cut_coul;
|
||||
double f_shift = -(e_shift+ 2.0*alf/MY_PIS * exp(-alf*alf*cut_coul*cut_coul)) /
|
||||
cut_coul;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
qtmp = q[i];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
qisq = qtmp*qtmp;
|
||||
e_self = -(e_shift/2.0 + alf/MY_PIS) * qisq*qqrd2e;
|
||||
if (evflag) ev_tally(i,i,nlocal,0,0.0,e_self,0.0,0.0,0.0,0.0);
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
factor_lj = special_lj[sbmask(j)];
|
||||
factor_coul = special_coul[sbmask(j)];
|
||||
j &= NEIGHMASK;
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
jtype = type[j];
|
||||
|
||||
if (rsq < cutsq[itype][jtype]) {
|
||||
rsq += EPSILON; // Add EPISLON for case: r = 0; Interaction must be removed by special bond
|
||||
r2inv = 1.0/rsq;
|
||||
|
||||
if (rsq < cut_coulsq) {
|
||||
r = sqrt(rsq);
|
||||
prefactor = qqrd2e*qtmp*q[j]/r;
|
||||
erfcc = erfc(alf*r);
|
||||
erfcd = exp(-alf*alf*r*r);
|
||||
v_sh = (erfcc - e_shift*r) * prefactor;
|
||||
dvdrr = (erfcc/rsq + 2.0*alf/MY_PIS * erfcd/r) + f_shift;
|
||||
forcecoul = dvdrr*rsq*prefactor;
|
||||
if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else forcecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
r = sqrt(rsq);
|
||||
rexp = exp((sigma[itype][jtype]-r)*rhoinv[itype][jtype]);
|
||||
forceborn = born1[itype][jtype]*r*rexp - born2[itype][jtype]*r6inv +
|
||||
born3[itype][jtype]*r2inv*r6inv;
|
||||
} else forceborn = 0.0;
|
||||
|
||||
fpair = (factor_coul*forcecoul + factor_lj*forceborn) * r2inv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
if (rsq < cut_coulsq) {
|
||||
ecoul = v_sh;
|
||||
if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else ecoul = 0.0;
|
||||
if (rsq < cut_ljsq[itype][jtype]) {
|
||||
evdwl = a[itype][jtype]*rexp - c[itype][jtype]*r6inv +
|
||||
d[itype][jtype]*r6inv*r2inv - offset[itype][jtype];
|
||||
evdwl *= factor_lj;
|
||||
} else evdwl = 0.0;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,ecoul,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -0,0 +1,58 @@
|
|||
/* -*- c++ -*- ----------------------------------------------------------
|
||||
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.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(born/coul/wolf/cs/gpu,PairBornCoulWolfCSGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_BORN_COUL_WOLF_CS_GPU_H
|
||||
#define LMP_PAIR_BORN_COUL_WOLF_CS_GPU_H
|
||||
|
||||
#include "pair_born_coul_wolf_cs.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairBornCoulWolfCSGPU : public PairBornCoulWolfCS {
|
||||
public:
|
||||
PairBornCoulWolfCSGPU(LAMMPS *lmp);
|
||||
~PairBornCoulWolfCSGPU();
|
||||
void cpu_compute(int, int, int, int, int *, int *, int **);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/* ERROR/WARNING messages:
|
||||
|
||||
E: Insufficient memory on accelerator
|
||||
|
||||
There is insufficient memory on one of the devices specified for the gpu
|
||||
package
|
||||
|
||||
E: Cannot use newton pair with born/coul/wolf/cs/gpu pair style
|
||||
|
||||
Self-explanatory.
|
||||
|
||||
*/
|
||||
Loading…
Reference in New Issue