int LJGROMACST::init(const int ntypes, double **host_cutsq,
double **host_lj1, double **host_lj2, double **host_lj3,
double **host_lj4, double *host_special_lj,
const int nlocal, const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen,
double **host_ljsw1, double **host_ljsw2, double **host_ljsw3,
double **host_ljsw4, double **host_ljsw5,
double **cut_inner, double **cut_inner_sq) {
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
_screen,lj_gromacs,"k_lj_gromacs");
if (success!=0)
return success;
// If atom type constants fit in shared memory use fast kernel
int lj_types=ntypes;
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;
}
_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),
UCL_WRITE_ONLY);
for (int i=0; i<lj_types*lj_types; i++)
host_write[i]=0.0;
lj1.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,lj_types,lj1,host_write,host_lj1,host_lj2,
host_cutsq,cut_inner_sq);
lj3.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,lj_types,lj3,host_write,host_lj3,host_lj4,
cut_inner,host_ljsw5);
ljsw.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,lj_types,ljsw,host_write,host_ljsw1,host_ljsw2,
host_ljsw3,host_ljsw4);
UCL_H_Vec<double> dview;
sp_lj.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
dview.view(host_special_lj,4,*(this->ucl_device));
ucl_copy(sp_lj,dview,false);
_allocated=true;
this->_max_bytes=lj1.row_bytes()+lj3.row_bytes()
+ljsw.row_bytes()+sp_lj.row_bytes();
return 0;
}
int BornT::init(const int ntypes, double **host_cutsq,
double **host_rhoinv, double **host_born1, double **host_born2,
double **host_born3, double **host_a, double **host_c,
double **host_d, double **host_sigma,
double **host_offset, double *host_special_lj,
const int nlocal, const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen) {
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
_screen,born,"k_born");
if (success!=0)
return success;
// If atom type constants fit in shared memory use fast kernel
int lj_types=ntypes;
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;
}
_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),
UCL_WRITE_OPTIMIZED);
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,
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,
host_d,host_offset);
cutsq_sigma.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack2(ntypes,lj_types,cutsq_sigma,host_write,host_cutsq,
host_sigma);
UCL_H_Vec<double> dview;
sp_lj.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
dview.view(host_special_lj,4,*(this->ucl_device));
ucl_copy(sp_lj,dview,false);
_allocated=true;
this->_max_bytes=coeff1.row_bytes()+coeff2.row_bytes()
+cutsq_sigma.row_bytes()+sp_lj.row_bytes();
return 0;
}
int DPDT::init(const int ntypes,
double **host_cutsq, double **host_a0,
double **host_gamma, double **host_sigma,
double **host_cut, double *host_special_lj,
const bool tstat_only,
const int nlocal, const int nall,
const int max_nbors, const int maxspecial,
const double cell_size,
const double gpu_split, FILE *_screen) {
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,_screen,dpd,"k_dpd");
if (success!=0)
return success;
// If atom type constants fit in shared memory use fast kernel
int lj_types=ntypes;
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;
}
_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),
UCL_WRITE_ONLY);
for (int i=0; i<lj_types*lj_types; i++)
host_write[i]=0.0;
coeff.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,lj_types,coeff,host_write,host_a0,host_gamma,
host_sigma,host_cut);
UCL_H_Vec<numtyp> host_rsq(lj_types*lj_types,*(this->ucl_device),
UCL_WRITE_ONLY);
cutsq.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack1(ntypes,lj_types,cutsq,host_rsq,host_cutsq);
UCL_H_Vec<double> dview;
sp_lj.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
dview.view(host_special_lj,4,*(this->ucl_device));
ucl_copy(sp_lj,dview,false);
_tstat_only = 0;
if (tstat_only) _tstat_only=1;
_allocated=true;
this->_max_bytes=coeff.row_bytes()+cutsq.row_bytes()+sp_lj.row_bytes();
return 0;
}
int CGCMMT::init(const int ntypes, double **host_cutsq,
int **host_cg_type, double **host_lj1,
double **host_lj2, double **host_lj3,
double **host_lj4, double **host_offset,
double *host_special_lj, const int nlocal,
const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen) {
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
_screen,cg_cmm,"k_cg_cmm");
if (success!=0)
return success;
// If atom type constants fit in shared memory use fast kernel
int cmm_types=ntypes;
shared_types=false;
int max_shared_types=this->device->max_shared_types();
if (cmm_types<=max_shared_types && this->_block_size>=max_shared_types) {
cmm_types=max_shared_types;
shared_types=true;
}
_cmm_types=cmm_types;
// Allocate a host write buffer for data initialization
UCL_H_Vec<numtyp> host_write(cmm_types*cmm_types*32,*(this->ucl_device),
UCL_WRITE_ONLY);
for (int i=0; i<cmm_types*cmm_types; i++)
host_write[i]=0.0;
lj1.alloc(cmm_types*cmm_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,cmm_types,lj1,host_write,host_cutsq,
host_cg_type,host_lj1,host_lj2);
lj3.alloc(cmm_types*cmm_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,cmm_types,lj3,host_write,host_lj3,host_lj4,
host_offset);
UCL_H_Vec<double> dview;
sp_lj.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
dview.view(host_special_lj,4,*(this->ucl_device));
ucl_copy(sp_lj,dview,false);
_allocated=true;
this->_max_bytes=lj1.row_bytes()+lj3.row_bytes()+sp_lj.row_bytes();
return 0;
}
int MorseT::init(const int ntypes,
double **host_cutsq, double **host_morse1,
double **host_r0, double **host_alpha,
double **host_d0, double **host_offset,
double *host_special_lj, const int nlocal,
const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen) {
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,gpu_split,
_screen,morse,"k_morse");
if (success!=0)
return success;
// If atom type constants fit in shared memory use fast kernel
int types=ntypes;
shared_types=false;
int max_shared_types=this->device->max_shared_types();
if (types<=max_shared_types && this->_block_size>=max_shared_types) {
types=max_shared_types;
shared_types=true;
}
_types=types;
// Allocate a host write buffer for data initialization
UCL_H_Vec<numtyp> host_write(types*types*32,*(this->ucl_device),
UCL_WRITE_ONLY);
for (int i=0; i<types*types; i++)
host_write[i]=0.0;
mor1.alloc(types*types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack4(ntypes,types,mor1,host_write,host_cutsq,host_morse1,
host_r0,host_alpha);
mor2.alloc(types*types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack2(ntypes,types,mor2,host_write,host_d0,host_offset);
UCL_H_Vec<double> dview;
sp_lj.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
dview.view(host_special_lj,4,*(this->ucl_device));
ucl_copy(sp_lj,dview,false);
_allocated=true;
this->_max_bytes=mor1.row_bytes()+mor2.row_bytes()+sp_lj.row_bytes();
return 0;
}
int TableT::init(const int ntypes,
double **host_cutsq, double ***host_table_coeffs,
double **host_table_data,
double *host_special_lj, const int nlocal,
const int nall, const int max_nbors,
const int maxspecial, const double cell_size,
const double gpu_split, FILE *_screen,
int tabstyle, int ntables, int tablength) {
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,
gpu_split,_screen,table,"k_table");
if (success!=0)
return success;
k_pair_linear.set_function(*(this->pair_program),"k_table_linear");
k_pair_linear_fast.set_function(*(this->pair_program),"k_table_linear_fast");
k_pair_spline.set_function(*(this->pair_program),"k_table_spline");
k_pair_spline_fast.set_function(*(this->pair_program),"k_table_spline_fast");
k_pair_bitmap.set_function(*(this->pair_program),"k_table_bitmap");
k_pair_bitmap_fast.set_function(*(this->pair_program),"k_table_bitmap_fast");
_compiled_styles = true;
// If atom type constants fit in shared memory use fast kernel
int lj_types=ntypes;
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;
}
_lj_types=lj_types;
_tabstyle = tabstyle;
_ntables = ntables;
if (tabstyle != BITMAP) _tablength = tablength;
else _tablength = 1 << tablength;
// Allocate a host write buffer for data initialization
UCL_H_Vec<int> host_write_int(lj_types*lj_types,*(this->ucl_device),
UCL_WRITE_ONLY);
for (int i=0; i<lj_types*lj_types; i++)
host_write_int[i] = 0;
tabindex.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
nshiftbits.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
nmask.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
for (int ix=1; ix<ntypes; ix++)
for (int iy=1; iy<ntypes; iy++)
host_write_int[ix*lj_types+iy] = (int)host_table_coeffs[ix][iy][0]; // tabindex
ucl_copy(tabindex,host_write_int,false);
for (int ix=1; ix<ntypes; ix++)
for (int iy=1; iy<ntypes; iy++)
host_write_int[ix*lj_types+iy] = (int)host_table_coeffs[ix][iy][1]; // nshiftbits
ucl_copy(nshiftbits,host_write_int,false);
for (int ix=1; ix<ntypes; ix++)
for (int iy=1; iy<ntypes; iy++)
host_write_int[ix*lj_types+iy] = (int)host_table_coeffs[ix][iy][2]; // nmask
ucl_copy(nmask,host_write_int,false);
UCL_H_Vec<numtyp4> host_write(lj_types*lj_types,*(this->ucl_device),
UCL_WRITE_ONLY);
coeff2.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
for (int ix=1; ix<ntypes; ix++)
for (int iy=1; iy<ntypes; iy++) {
host_write[ix*lj_types+iy].x = host_table_coeffs[ix][iy][3]; // innersq
host_write[ix*lj_types+iy].y = host_table_coeffs[ix][iy][4]; // invdelta
host_write[ix*lj_types+iy].z = host_table_coeffs[ix][iy][5]; // deltasq6
host_write[ix*lj_types+iy].w = (numtyp)0.0;
}
ucl_copy(coeff2,host_write,false);
// Allocate tablength arrays
UCL_H_Vec<numtyp4> host_write2(_ntables*_tablength,*(this->ucl_device),
UCL_WRITE_ONLY);
for (int i=0; i<_ntables*_tablength; i++) {
host_write2[i].x = 0.0;
host_write2[i].y = 0.0;
host_write2[i].z = 0.0;
host_write2[i].w = 0.0;
}
coeff3.alloc(_ntables*_tablength,*(this->ucl_device),UCL_READ_ONLY);
for (int n=0; n<_ntables; n++) {
if (tabstyle == LOOKUP) {
for (int k=0; k<_tablength-1; k++) {
host_write2[n*_tablength+k].x = (numtyp)0;
host_write2[n*_tablength+k].y = host_table_data[n][6*k+1]; // e
host_write2[n*_tablength+k].z = host_table_data[n][6*k+2]; // f
host_write2[n*_tablength+k].w = (numtyp)0;
}
} else if (tabstyle == LINEAR || tabstyle == SPLINE || tabstyle == BITMAP) {
for (int k=0; k<_tablength; k++) {
host_write2[n*_tablength+k].x = host_table_data[n][6*k+0]; // rsq
host_write2[n*_tablength+k].y = host_table_data[n][6*k+1]; // e
host_write2[n*_tablength+k].z = host_table_data[n][6*k+2]; // f
host_write2[n*_tablength+k].w = (numtyp)0;
}
}
}
ucl_copy(coeff3,host_write2,false);
coeff4.alloc(_ntables*_tablength,*(this->ucl_device),UCL_READ_ONLY);
for (int i=0; i<_ntables*_tablength; i++) {
host_write2[i].x = 0.0;
host_write2[i].y = 0.0;
host_write2[i].z = 0.0;
host_write2[i].w = 0.0;
}
for (int n=0; n<_ntables; n++) {
if (tabstyle == LINEAR) {
for (int k=0; k<_tablength-1; k++) {
host_write2[n*_tablength+k].x = (numtyp)0;
host_write2[n*_tablength+k].y = host_table_data[n][6*k+3]; // de
host_write2[n*_tablength+k].z = host_table_data[n][6*k+4]; // df
host_write2[n*_tablength+k].w = (numtyp)0;
}
} else if (tabstyle == SPLINE) {
for (int k=0; k<_tablength; k++) {
host_write2[n*_tablength+k].x = (numtyp)0;
host_write2[n*_tablength+k].y = host_table_data[n][6*k+3]; // e2
host_write2[n*_tablength+k].z = host_table_data[n][6*k+4]; // f2
host_write2[n*_tablength+k].w = (numtyp)0;
}
} else if (tabstyle == BITMAP) {
for (int k=0; k<_tablength; k++) {
host_write2[n*_tablength+k].x = (numtyp)0;
host_write2[n*_tablength+k].y = host_table_data[n][6*k+3]; // de
host_write2[n*_tablength+k].z = host_table_data[n][6*k+4]; // df
host_write2[n*_tablength+k].w = host_table_data[n][6*k+5]; // drsq
}
}
}
ucl_copy(coeff4,host_write2,false);
UCL_H_Vec<numtyp> host_rsq(lj_types*lj_types,*(this->ucl_device),
UCL_WRITE_ONLY);
cutsq.alloc(lj_types*lj_types,*(this->ucl_device),UCL_READ_ONLY);
this->atom->type_pack1(ntypes,lj_types,cutsq,host_rsq,host_cutsq);
UCL_H_Vec<double> dview;
sp_lj.alloc(4,*(this->ucl_device),UCL_READ_ONLY);
dview.view(host_special_lj,4,*(this->ucl_device));
ucl_copy(sp_lj,dview,false);
_allocated=true;
this->_max_bytes=tabindex.row_bytes()+nshiftbits.row_bytes()
+nmask.row_bytes()+coeff2.row_bytes()
+coeff3.row_bytes()+coeff4.row_bytes()+cutsq.row_bytes()
+sp_lj.row_bytes();
return 0;
}
