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@14265 f3b2605a-c512-4ea7-a41b-209d697bcdaa

Browse Source
This commit is contained in:
sjplimp
2015-11-18 18:23:45 +00:00
parent e3c4db746c
commit f6c76f4623
9 changed files with 568 additions and 582 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
lib/gpu/lal_answer.h
View File

@ -9,7 +9,7 @@
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
__________________________________________________________________________
begin :
begin :
email : brownw@ornl.gov
***************************************************************************/
@ -49,14 +49,18 @@ class Answer {
inline void inum(const int n) { _inum=n; }
/// Return the maximum number of atoms that can be stored currently
inline int max_inum() const { return _max_local; }
/// Return the number of fields used for energy and virial
inline int ev_fields(const int mode) const {
return (mode == 1) ? _ev_fields : _e_fields;
}
/// Memory usage per atom in this class
int bytes_per_atom() const;
int bytes_per_atom() const;
/// Clear any previous data and set up for a new LAMMPS run
/** \param rot True if atom storage needs quaternions **/
bool init(const int inum, const bool charge, const bool rot, UCL_Device &dev);
/// Check if we have enough device storage and realloc if not
inline void resize(const int inum, bool &success) {
_inum=inum;
@ -67,14 +71,14 @@ class Answer {
_gpu_bytes=engv.device.row_bytes()+force.device.row_bytes();
}
}
/// If already initialized by another LAMMPS style, add fields as necessary
/** \param rot True if atom storage needs quaternions **/
bool add_fields(const bool charge, const bool rot);
/// Free all memory on host and device
void clear();
/// Return the total amount of host memory used by class in bytes
double host_memory_usage() const;
@ -92,12 +96,12 @@ class Answer {
inline double transfer_time() {
return time_answer.total_seconds();
}
/// Return the total time for data cast/pack
inline double cast_time() { return _time_cast; }
/// Return number of bytes used on device
inline double gpu_bytes() { return _gpu_bytes; }
inline double gpu_bytes() { return _gpu_bytes; }
// -------------------------COPY FROM GPU -------------------------------
@ -108,7 +112,7 @@ class Answer {
/// Copy answers from device into read buffer asynchronously
void copy_answers(const bool eflag, const bool vflag,
const bool ef_atom, const bool vf_atom, int *ilist);
/// Copy energy and virial data into LAMMPS memory
double energy_virial(double *eatom, double **vatom, double *virial);
@ -119,7 +123,7 @@ class Answer {
/// Add forces and torques from the GPU into a LAMMPS pointer
void get_answers(double **f, double **tor);
inline double get_answers(double **f, double **tor, double *eatom,
inline double get_answers(double **f, double **tor, double *eatom,
double **vatom, double *virial, double &ecoul) {
double ta=MPI_Wtime();
time_answer.sync_stop();
@ -130,7 +134,7 @@ class Answer {
_time_cast+=MPI_Wtime()-ts;
return evdw;
}
/// Return the time the CPU was idle waiting for GPU
inline double cpu_idle_time() { return _time_cpu_idle; }
@ -143,23 +147,23 @@ class Answer {
UCL_Vector<acctyp,acctyp> force;
/// Energy and virial per-atom storage
UCL_Vector<acctyp,acctyp> engv;
/// Device timers
UCL_Timer time_answer;
/// Geryon device
UCL_Device *dev;
private:
bool alloc(const int inum);
bool _allocated, _eflag, _vflag, _ef_atom, _vf_atom, _rot, _charge, _other;
int _max_local, _inum, _e_fields, _ev_fields, _ans_fields;
int *_ilist;
double _time_cast, _time_cpu_idle;
double _gpu_bytes;
bool _newton;
};

2
lib/gpu/lal_base_three.cpp
View File

@ -194,7 +194,7 @@ inline int BaseThreeT::build_nbor_list(const int inum, const int host_inum,
resize_atom(inum,nall,success);
resize_local(nall,host_inum,nbor->max_nbors(),success);
if (!success)
return 1;
return 0;
atom->cast_copy_x(host_x,host_type);
int mn;

6
lib/gpu/lal_base_three.h
View File

@ -13,8 +13,8 @@
email : brownw@ornl.gov
***************************************************************************/
#ifndef LAL_BASE_ATOMIC_H
#define LAL_BASE_ATOMIC_H
#ifndef LAL_BASE_THREE_H
#define LAL_BASE_THREE_H
#include "lal_device.h"
#include "lal_balance.h"
@ -28,7 +28,7 @@
#include "geryon/nvd_texture.h"
#endif
#define THREE_CONCURRENT
//#define THREE_CONCURRENT
namespace LAMMPS_AL {

2
lib/gpu/lal_sw.cu
View File

@ -37,7 +37,7 @@ texture<int4> sw3_tex;
#define THIRD (numtyp)0.66666667
#define THREE_CONCURRENT
//#define THREE_CONCURRENT
#if (ARCH < 300)

137
lib/gpu/lal_tersoff.cpp
View File

@ -33,10 +33,10 @@ TersoffT::Tersoff() : BaseThree<numtyp,acctyp>(), _allocated(false) {
}
template <class numtyp, class acctyp>
TersoffT::~Tersoff() {
TersoffT::~Tersoff() {
clear();
}
template <class numtyp, class acctyp>
int TersoffT::bytes_per_atom(const int max_nbors) const {
return this->bytes_per_atom_atomic(max_nbors);
@ -45,11 +45,11 @@ int TersoffT::bytes_per_atom(const int max_nbors) const {
template <class numtyp, class acctyp>
int TersoffT::init(const int ntypes, const int nlocal, const int nall, const int max_nbors,
const double cell_size, const double gpu_split, FILE *_screen,
int* host_map, const int nelements, int*** host_elem2param, const int nparams,
int* host_map, const int nelements, int*** host_elem2param, const int nparams,
const double* lam1, const double* lam2, const double* lam3,const double* powermint,
const double* biga, const double* bigb, const double* bigr, const double* bigd,
const double* c1, const double* c2, const double* c3, const double* c4,
const double* c, const double* d, const double* h, const double* gamma,
const double* biga, const double* bigb, const double* bigr, const double* bigd,
const double* c1, const double* c2, const double* c3, const double* c4,
const double* c, const double* d, const double* h, const double* gamma,
const double* beta, const double* powern, const double* host_cutsq)
{
int success;
@ -62,11 +62,7 @@ int TersoffT::init(const int ntypes, const int nlocal, const int nall, const int
int ef_nall=nall;
if (ef_nall==0)
ef_nall=2000;
_max_zij_size=static_cast<int>(static_cast<double>(ef_nall)*1.10);
_zetaij.alloc(_max_zij_size*max_nbors,*(this->ucl_device),UCL_READ_WRITE);
zeta_tex.get_texture(*(this->pair_program),"zeta_tex");
zeta_tex.bind_float(_zetaij,1);
_zetaij.alloc(ef_nall*max_nbors,*(this->ucl_device),UCL_READ_WRITE);
k_zeta.set_function(*(this->pair_program),"k_tersoff_zeta");
@ -87,74 +83,74 @@ int TersoffT::init(const int ntypes, const int nlocal, const int nall, const int
UCL_WRITE_ONLY);
for (int i=0; i<nparams; i++) {
dview[i].x=(numtyp)0;
dview[i].x=(numtyp)0;
dview[i].y=(numtyp)0;
dview[i].z=(numtyp)0;
dview[i].z=(numtyp)0;
dview[i].w=(numtyp)0;
}
// pack coefficients into arrays
ts1.alloc(nparams,*(this->ucl_device),UCL_READ_ONLY);
for (int i=0; i<nparams; i++) {
dview[i].x=static_cast<numtyp>(lam1[i]);
dview[i].y=static_cast<numtyp>(lam2[i]);
dview[i].z=static_cast<numtyp>(lam3[i]);
dview[i].w=static_cast<numtyp>(powermint[i]);
}
ucl_copy(ts1,dview,false);
ts1_tex.get_texture(*(this->pair_program),"ts1_tex");
ts1_tex.bind_float(ts1,4);
ts2.alloc(nparams,*(this->ucl_device),UCL_READ_ONLY);
for (int i=0; i<nparams; i++) {
dview[i].x=static_cast<numtyp>(biga[i]);
dview[i].y=static_cast<numtyp>(bigb[i]);
dview[i].z=static_cast<numtyp>(bigr[i]);
dview[i].w=static_cast<numtyp>(bigd[i]);
}
ucl_copy(ts2,dview,false);
ts2_tex.get_texture(*(this->pair_program),"ts2_tex");
ts2_tex.bind_float(ts2,4);
ts3.alloc(nparams,*(this->ucl_device),UCL_READ_ONLY);
for (int i=0; i<nparams; i++) {
dview[i].x=static_cast<numtyp>(c1[i]);
dview[i].y=static_cast<numtyp>(c2[i]);
dview[i].z=static_cast<numtyp>(c3[i]);
dview[i].w=static_cast<numtyp>(c4[i]);
}
ucl_copy(ts3,dview,false);
ts3_tex.get_texture(*(this->pair_program),"ts3_tex");
ts3_tex.bind_float(ts3,4);
ts4.alloc(nparams,*(this->ucl_device),UCL_READ_ONLY);
for (int i=0; i<nparams; i++) {
dview[i].x=static_cast<numtyp>(c[i]);
dview[i].y=static_cast<numtyp>(d[i]);
dview[i].z=static_cast<numtyp>(h[i]);
dview[i].w=static_cast<numtyp>(gamma[i]);
}
ucl_copy(ts4,dview,false);
ts4_tex.get_texture(*(this->pair_program),"ts4_tex");
ts4_tex.bind_float(ts4,4);
ts5.alloc(nparams,*(this->ucl_device),UCL_READ_ONLY);
for (int i=0; i<nparams; i++) {
dview[i].x=static_cast<numtyp>(beta[i]);
dview[i].y=static_cast<numtyp>(powern[i]);
dview[i].z=(numtyp)0;
dview[i].w=(numtyp)0;
}
ucl_copy(ts5,dview,false);
ts5_tex.get_texture(*(this->pair_program),"ts5_tex");
ts5_tex.bind_float(ts5,4);
@ -227,11 +223,11 @@ double TersoffT::host_memory_usage() const {
// Copy nbor list from host if necessary and then calculate forces, virials,..
// ---------------------------------------------------------------------------
template <class numtyp, class acctyp>
void TersoffT::compute(const int f_ago, const int nlocal, const int nall,
void TersoffT::compute(const int f_ago, const int nlocal, const int nall,
const int nlist, double **host_x, int *host_type,
int *ilist, int *numj, int **firstneigh,
int *ilist, int *numj, int **firstneigh,
const bool eflag, const bool vflag, const bool eatom,
const bool vatom, int &host_start,
const bool vatom, int &host_start,
const double cpu_time, bool &success) {
this->acc_timers();
if (nlist==0) {
@ -254,6 +250,7 @@ void TersoffT::compute(const int f_ago, const int nlocal, const int nall,
this->reset_nbors(nall, inum, nlist, ilist, numj, firstneigh, success);
if (!success)
return;
_max_nbors = this->nbor->max_nbor_loop(nlist,numj,ilist);
}
this->atom->cast_x_data(host_x,host_type);
@ -261,24 +258,28 @@ void TersoffT::compute(const int f_ago, const int nlocal, const int nall,
this->atom->add_x_data(host_x,host_type);
// re-allocate zetaij if necessary
if (nall>_max_zij_size) {
_max_nbors = this->nbor->max_nbor_loop(nlist,numj,ilist);
_max_zij_size=static_cast<int>(static_cast<double>(nall)*1.10);
_zetaij.resize(_max_nbors*_max_zij_size);
zeta_tex.bind_float(_zetaij,1);
if (nall*_max_nbors > _zetaij.cols()) {
int _nmax=static_cast<int>(static_cast<double>(nall)*1.10);
_zetaij.resize(_max_nbors*_nmax);
}
int _eflag;
if (eflag)
_eflag=1;
else
_eflag=0;
int ainum=nall;
int nbor_pitch=this->nbor->nbor_pitch();
int BX=this->block_pair();
int GX=static_cast<int>(ceil(static_cast<double>(ainum)/
(BX/this->_threads_per_atom)));
int GX=static_cast<int>(ceil(static_cast<double>(ainum)/
(BX/(JTHREADS*KTHREADS))));
this->k_zeta.set_size(GX,BX);
this->k_zeta.run(&this->atom->x, &ts1, &ts2, &ts4, &cutsq,
this->k_zeta.run(&this->atom->x, &ts1, &ts2, &ts3, &ts4, &ts5, &cutsq,
&map, &elem2param, &_nelements, &_nparams, &_zetaij,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&nall, &ainum, &nbor_pitch, &this->_threads_per_atom);
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&_eflag, &nall, &ainum, &nbor_pitch, &this->_threads_per_atom);
int evatom=0;
if (eatom || vatom)
@ -303,7 +304,7 @@ template <class numtyp, class acctyp>
int ** TersoffT::compute(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,
int **nspecial, tagint **special, const bool eflag,
const bool vflag, const bool eatom,
const bool vatom, int &host_start,
int **ilist, int **jnum,
@ -317,7 +318,7 @@ int ** TersoffT::compute(const int ago, const int inum_full,
this->zero_timers();
return NULL;
}
this->hd_balancer.balance(cpu_time);
int inum=this->hd_balancer.get_gpu_count(ago,inum_full);
this->ans->inum(inum);
@ -325,7 +326,7 @@ int ** TersoffT::compute(const int ago, const int inum_full,
this->ans2->inum(inum);
#endif
host_start=inum;
// Build neighbor list on GPU if necessary
if (ago==0) {
_max_nbors = this->build_nbor_list(inum, inum_full-inum, nall, host_x, host_type,
@ -342,23 +343,28 @@ int ** TersoffT::compute(const int ago, const int inum_full,
*jnum=this->nbor->host_acc.begin();
// re-allocate zetaij if necessary
if (nall>_max_zij_size) {
_max_zij_size=static_cast<int>(static_cast<double>(nall)*1.10);
_zetaij.resize(_max_nbors*_max_zij_size);
zeta_tex.bind_float(_zetaij,1);
if (nall*_max_nbors > _zetaij.cols()) {
int _nmax=static_cast<int>(static_cast<double>(nall)*1.10);
_zetaij.resize(_max_nbors*_nmax);
}
int _eflag;
if (eflag)
_eflag=1;
else
_eflag=0;
int ainum=nall;
int nbor_pitch=this->nbor->nbor_pitch();
int BX=this->block_pair();
int GX=static_cast<int>(ceil(static_cast<double>(ainum)/
(BX/this->_threads_per_atom)));
int GX=static_cast<int>(ceil(static_cast<double>(ainum)/
(BX/(JTHREADS*KTHREADS))));
this->k_zeta.set_size(GX,BX);
this->k_zeta.run(&this->atom->x, &ts1, &ts2, &ts4, &cutsq, &map,
&elem2param, &_nelements, &_nparams, &_zetaij,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&nall, &ainum, &nbor_pitch, &this->_threads_per_atom);
this->k_zeta.run(&this->atom->x, &ts1, &ts2, &ts3, &ts4, &ts5, &cutsq,
&map, &elem2param, &_nelements, &_nparams, &_zetaij,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&_eflag, &nall, &ainum, &nbor_pitch, &this->_threads_per_atom);
int evatom=0;
if (eatom || vatom)
@ -374,7 +380,7 @@ int ** TersoffT::compute(const int ago, const int inum_full,
this->device->add_ans_object(this->ans2);
#endif
this->hd_balancer.stop_timer();
return this->nbor->host_jlist.begin()-host_start;
}
@ -403,21 +409,21 @@ void TersoffT::loop(const bool _eflag, const bool _vflag, const int evatom) {
this->time_pair.start();
this->k_pair.set_size(GX,BX);
this->k_pair.run(&this->atom->x, &ts1, &ts2, &cutsq,
this->k_pair.run(&this->atom->x, &ts1, &ts2, &cutsq,
&map, &elem2param, &_nelements, &_nparams,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv,
&eflag, &vflag, &ainum, &nbor_pitch,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv,
&eflag, &vflag, &ainum, &nbor_pitch,
&this->_threads_per_atom);
BX=this->block_size();
GX=static_cast<int>(ceil(static_cast<double>(this->ans->inum())/
(BX/(KTHREADS*JTHREADS))));
(BX/(KTHREADS*JTHREADS))));
this->k_three_center.set_size(GX,BX);
this->k_three_center.run(&this->atom->x, &ts1, &ts2, &ts3, &ts4, &ts5, &cutsq,
this->k_three_center.run(&this->atom->x, &ts1, &ts2, &ts4, &cutsq,
&map, &elem2param, &_nelements, &_nparams, &_zetaij,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv, &eflag, &vflag, &ainum,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv, &eflag, &vflag, &ainum,
&nbor_pitch, &this->_threads_per_atom, &evatom);
Answer<numtyp,acctyp> *end_ans;
@ -428,20 +434,19 @@ void TersoffT::loop(const bool _eflag, const bool _vflag, const int evatom) {
#endif
if (evatom!=0) {
this->k_three_end_vatom.set_size(GX,BX);
this->k_three_end_vatom.run(&this->atom->x, &ts1, &ts2, &ts3, &ts4, &ts5, &cutsq,
&map, &elem2param, &_nelements, &_nparams, &_zetaij,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
this->k_three_end_vatom.run(&this->atom->x, &ts1, &ts2, &ts4, &cutsq,
&map, &elem2param, &_nelements, &_nparams, &_zetaij,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&end_ans->force, &end_ans->engv, &eflag, &vflag, &ainum,
&nbor_pitch, &this->_threads_per_atom);
} else {
this->k_three_end.set_size(GX,BX);
this->k_three_end.run(&this->atom->x, &ts1, &ts2, &ts3, &ts4, &ts5, &cutsq,
&map, &elem2param, &_nelements, &_nparams, &_zetaij,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&end_ans->force, &end_ans->engv, &eflag, &vflag, &ainum,
this->k_three_end.run(&this->atom->x, &ts1, &ts2, &ts4, &cutsq,
&map, &elem2param, &_nelements, &_nparams, &_zetaij,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&end_ans->force, &end_ans->engv, &eflag, &vflag, &ainum,
&nbor_pitch, &this->_threads_per_atom);
}
this->time_pair.stop();

703
lib/gpu/lal_tersoff.cu
View File

File diff suppressed because it is too large Load Diff

30
lib/gpu/lal_tersoff.h
View File

@ -30,25 +30,25 @@ class Tersoff : public BaseThree<numtyp, acctyp> {
/** \param max_nbors initial number of rows in the neighbor matrix
* \param cell_size cutoff + skin
* \param gpu_split fraction of particles handled by device
*
*
* Returns:
* - 0 if successfull
* - -1 if fix gpu not found
* - -3 if there is an out of memory error
* - -4 if the GPU library was not compiled for GPU
* - -5 Double precision is not supported on card **/
int init(const int ntypes, const int nlocal, const int nall, const int max_nbors,
int init(const int ntypes, const int nlocal, const int nall, const int max_nbors,
const double cell_size, const double gpu_split, FILE *screen,
int* host_map, const int nelements, int*** host_elem2param, const int nparams,
int* host_map, const int nelements, int*** host_elem2param, const int nparams,
const double* lam1, const double* lam2, const double* lam3,
const double* powermint, const double* biga, const double* bigb,
const double* bigr, const double* bigd, const double* c1, const double* c2,
const double* c3, const double* c4, const double* c, const double* d,
const double* h, const double* gamma, const double* beta,
const double* c3, const double* c4, const double* c, const double* d,
const double* h, const double* gamma, const double* beta,
const double* powern, const double* cutsq);
/// Pair loop with host neighboring
void compute(const int f_ago, const int inum_full, const int nall,
void compute(const int f_ago, const int inum_full, const int nall,
const int nlist, double **host_x, int *host_type,
int *ilist, int *numj, int **firstneigh, const bool eflag,
const bool vflag, const bool eatom, const bool vatom,
@ -58,10 +58,10 @@ class Tersoff : public BaseThree<numtyp, acctyp> {
int ** compute(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,
tagint **special, const bool eflag, const bool vflag,
const bool eatom, const bool vatom, int &host_start,
int **ilist, int **numj, const double cpu_time, bool &success);
/// Clear all host and device data
/** \note This is called at the beginning of the init() routine **/
void clear();
@ -77,7 +77,7 @@ class Tersoff : public BaseThree<numtyp, acctyp> {
/// If atom type constants fit in shared memory, use fast kernels
bool shared_types;
/// Number of atom types
/// Number of atom types
int _lj_types;
/// ts1.x = lam1, ts1.y = lam2, ts1.z = lam3, ts1.w = powermint
@ -97,13 +97,15 @@ class Tersoff : public BaseThree<numtyp, acctyp> {
UCL_D_Vec<int> map;
int _nparams,_nelements;
/// Per-atom arrays
UCL_D_Vec<numtyp> _zetaij;
/// Per-atom arrays:
/// zetaij.x = force, zetaij.y = prefactor, zetaij.z = evdwl,
/// zetaij.w = zetaij
UCL_D_Vec<numtyp4> _zetaij;
UCL_Kernel k_zeta;
UCL_Texture ts1_tex, ts2_tex, ts3_tex, ts4_tex, ts5_tex, zeta_tex;
UCL_Texture ts1_tex, ts2_tex, ts3_tex, ts4_tex, ts5_tex;
int _max_zij_size, _max_nbors;
int _max_nbors;
private:
bool _allocated;

40
lib/gpu/lal_tersoff_ext.cpp
View File

@ -27,15 +27,15 @@ static Tersoff<PRECISION,ACC_PRECISION> TSMF;
// ---------------------------------------------------------------------------
// Allocate memory on host and device and copy constants to device
// ---------------------------------------------------------------------------
int tersoff_gpu_init(const int ntypes, const int inum, const int nall, const int max_nbors,
int tersoff_gpu_init(const int ntypes, const int inum, const int nall, const int max_nbors,
const double cell_size, int &gpu_mode, FILE *screen,
int* host_map, const int nelements, int*** host_elem2param, const int nparams,
const double* ts_lam1, const double* ts_lam2, const double* ts_lam3,
const double* ts_powermint, const double* ts_biga, const double* ts_bigb,
const double* ts_bigr, const double* ts_bigd,
const double* ts_c1, const double* ts_c2, const double* ts_c3, const double* ts_c4,
const double* ts_c, const double* ts_d, const double* ts_h,
const double* ts_gamma, const double* ts_beta,
const double* ts_bigr, const double* ts_bigd,
const double* ts_c1, const double* ts_c2, const double* ts_c3, const double* ts_c4,
const double* ts_c, const double* ts_d, const double* ts_h,
const double* ts_gamma, const double* ts_beta,
const double* ts_powern, const double* ts_cutsq) {
TSMF.clear();
gpu_mode=TSMF.device->gpu_mode();
@ -47,13 +47,9 @@ int tersoff_gpu_init(const int ntypes, const int inum, const int nall, const int
int procs_per_gpu=TSMF.device->procs_per_gpu();
// disable host/device split for now
if (gpu_split != 1.0)
if (gpu_split != 1.0)
return -8;
// disable multiple threads per atom for now
if (TSMF.device->threads_per_atom() != 1)
return -10;
TSMF.device->init_message(screen,"tersoff/gpu",first_gpu,last_gpu);
bool message=false;
@ -69,9 +65,9 @@ int tersoff_gpu_init(const int ntypes, const int inum, const int nall, const int
if (world_me==0)
init_ok=TSMF.init(ntypes, inum, nall, 300, cell_size, gpu_split, screen,
host_map, nelements, host_elem2param, nparams,
ts_lam1, ts_lam2, ts_lam3, ts_powermint,
ts_biga, ts_bigb, ts_bigr, ts_bigd,
ts_c1, ts_c2, ts_c3, ts_c4, ts_c, ts_d, ts_h,
ts_lam1, ts_lam2, ts_lam3, ts_powermint,
ts_biga, ts_bigb, ts_bigr, ts_bigd,
ts_c1, ts_c2, ts_c3, ts_c4, ts_c, ts_d, ts_h,
ts_gamma, ts_beta, ts_powern, ts_cutsq);
TSMF.device->world_barrier();
@ -90,13 +86,13 @@ int tersoff_gpu_init(const int ntypes, const int inum, const int nall, const int
if (gpu_rank==i && world_me!=0)
init_ok=TSMF.init(ntypes, inum, nall, 300, cell_size, gpu_split, screen,
host_map, nelements, host_elem2param, nparams,
ts_lam1, ts_lam2, ts_lam3, ts_powermint,
ts_biga, ts_bigb, ts_bigr, ts_bigd,
ts_c1, ts_c2, ts_c3, ts_c4, ts_c, ts_d, ts_h,
ts_lam1, ts_lam2, ts_lam3, ts_powermint,
ts_biga, ts_bigb, ts_bigr, ts_bigd,
ts_c1, ts_c2, ts_c3, ts_c4, ts_c, ts_d, ts_h,
ts_gamma, ts_beta, ts_powern, ts_cutsq);
TSMF.device->gpu_barrier();
if (message)
if (message)
fprintf(screen,"Done.\n");
}
if (message)
@ -121,12 +117,12 @@ int ** tersoff_gpu_compute_n(const int ago, const int inum_full,
return TSMF.compute(ago, inum_full, nall, host_x, host_type, sublo,
subhi, tag, nspecial, special, eflag, vflag, eatom,
vatom, host_start, ilist, jnum, cpu_time, success);
}
}
void tersoff_gpu_compute(const int ago, const int nlocal, const int nall,
const int nlist, double **host_x, int *host_type,
int *ilist, int *numj, int **firstneigh, const bool eflag,
const bool vflag, const bool eatom, const bool vatom,
void tersoff_gpu_compute(const int ago, const int nlocal, const int nall,
const int nlist, 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) {
TSMF.compute(ago,nlocal,nall,nlist,host_x,host_type,ilist,numj,
firstneigh,eflag,vflag,eatom,vatom,host_start,cpu_time,success);

190
lib/gpu/lal_tersoff_extra.h
View File

@ -1,7 +1,7 @@
/// **************************************************************************
// tersoff_extra.h
// -------------------
// Trung Dac Nguyen
// Trung Dac Nguyen
//
// Device code for Tersoff math routines
//
@ -26,7 +26,7 @@
/* ---------------------------------------------------------------------- */
ucl_inline numtyp vec3_dot(const numtyp x[3], const numtyp y[3])
ucl_inline numtyp vec3_dot(const numtyp x[3], const numtyp y[3])
{
return (x[0]*y[0] + x[1]*y[1] + x[2]*y[2]);
}
@ -36,12 +36,12 @@ ucl_inline void vec3_add(const numtyp x[3], const numtyp y[3], numtyp z[3])
z[0] = x[0]+y[0]; z[1] = x[1]+y[1]; z[2] = x[2]+y[2];
}
ucl_inline void vec3_scale(const numtyp k, const numtyp x[3], numtyp y[3])
ucl_inline void vec3_scale(const numtyp k, const numtyp x[3], numtyp y[3])
{
y[0] = k*x[0]; y[1] = k*x[1]; y[2] = k*x[2];
}
ucl_inline void vec3_scaleadd(const numtyp k, const numtyp x[3],
ucl_inline void vec3_scaleadd(const numtyp k, const numtyp x[3],
const numtyp y[3], numtyp z[3])
{
z[0] = k*x[0]+y[0]; z[1] = k*x[1]+y[1]; z[2] = k*x[2]+y[2];
@ -51,14 +51,14 @@ ucl_inline void vec3_scaleadd(const numtyp k, const numtyp x[3],
ucl_inline numtyp ters_gijk(const numtyp costheta,
const numtyp param_c,
const numtyp param_d,
const numtyp param_d,
const numtyp param_h,
const numtyp param_gamma)
const numtyp param_gamma)
{
const numtyp ters_c = param_c * param_c;
const numtyp ters_d = param_d * param_d;
const numtyp hcth = param_h - costheta;
return param_gamma*((numtyp)1.0 + ters_c*ucl_recip(ters_d) -
return param_gamma*((numtyp)1.0 + ters_c*ucl_recip(ters_d) -
ters_c *ucl_recip(ters_d + hcth*hcth));
}
@ -66,9 +66,9 @@ ucl_inline numtyp ters_gijk(const numtyp costheta,
ucl_inline numtyp ters_gijk_d(const numtyp costheta,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_gamma)
const numtyp param_d,
const numtyp param_h,
const numtyp param_gamma)
{
const numtyp ters_c = param_c * param_c;
const numtyp ters_d = param_d * param_d;
@ -80,12 +80,12 @@ ucl_inline numtyp ters_gijk_d(const numtyp costheta,
/* ---------------------------------------------------------------------- */
ucl_inline void costheta_d(const numtyp rij_hat[3],
ucl_inline void costheta_d(const numtyp rij_hat[3],
const numtyp rij,
const numtyp rik_hat[3],
const numtyp rik_hat[3],
const numtyp rik,
numtyp *dri,
numtyp *drj,
numtyp *dri,
numtyp *drj,
numtyp *drk)
{
// first element is derivative wrt Ri, second wrt Rj, third wrt Rk
@ -131,85 +131,87 @@ ucl_inline numtyp ters_fa(const numtyp r,
const numtyp param_lam2)
{
if (r > param_bigr + param_bigd) return (numtyp)0.0;
return -param_bigb * ucl_exp(-param_lam2 * r) *
return -param_bigb * ucl_exp(-param_lam2 * r) *
ters_fc(r,param_bigr,param_bigd);
}
/* ---------------------------------------------------------------------- */
ucl_inline numtyp ters_fa_d(const numtyp r,
const numtyp param_bigb,
const numtyp param_bigb,
const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_lam2)
{
if (r > param_bigr + param_bigd) return (numtyp)0.0;
return param_bigb * ucl_exp(-param_lam2 * r) * (param_lam2 *
return param_bigb * ucl_exp(-param_lam2 * r) * (param_lam2 *
ters_fc(r,param_bigr,param_bigd) - ters_fc_d(r,param_bigr,param_bigd));
}
/* ---------------------------------------------------------------------- */
ucl_inline numtyp ters_bij(const numtyp zeta,
const numtyp param_beta,
const numtyp param_beta,
const numtyp param_powern,
const numtyp param_c1,
const numtyp param_c2,
const numtyp param_c1,
const numtyp param_c2,
const numtyp param_c3,
const numtyp param_c4)
{
numtyp tmp = param_beta * zeta;
if (tmp > param_c1) return ucl_rsqrt(tmp);
if (tmp > param_c2)
return ((numtyp)1.0 - ucl_powr(tmp,-param_powern) /
return ((numtyp)1.0 - ucl_powr(tmp,-param_powern) /
((numtyp)2.0*param_powern))*ucl_rsqrt(tmp);
if (tmp < param_c4) return (numtyp)1.0;
if (tmp < param_c3)
return (numtyp)1.0 - ucl_powr(tmp,param_powern)/((numtyp)2.0*param_powern);
return ucl_powr((numtyp)1.0 + ucl_powr(tmp,param_powern),
return ucl_powr((numtyp)1.0 + ucl_powr(tmp,param_powern),
(numtyp)-1.0/((numtyp)2.0*param_powern));
}
/* ---------------------------------------------------------------------- */
ucl_inline numtyp ters_bij_d(const numtyp zeta,
const numtyp param_beta,
const numtyp param_beta,
const numtyp param_powern,
const numtyp param_c1,
const numtyp param_c2,
const numtyp param_c1,
const numtyp param_c2,
const numtyp param_c3,
const numtyp param_c4)
{
numtyp tmp = param_beta * zeta;
if (tmp > param_c1)
if (tmp > param_c1)
return param_beta * (numtyp)-0.5*ucl_powr(tmp,(numtyp)-1.5);
if (tmp > param_c2)
return param_beta * ((numtyp)-0.5*ucl_powr(tmp,(numtyp)-1.5) *
((numtyp)1.0 - (numtyp)0.5 * ((numtyp)1.0 + (numtyp)1.0 /
((numtyp)2.0 * param_powern)) * ucl_powr(tmp,-param_powern)));
// error in negligible 2nd term fixed 9/30/2015
// (1.0 - 0.5*(1.0 + 1.0/(2.0*param->powern)) *
((numtyp)1.0 - ((numtyp)1.0 + (numtyp)1.0 /((numtyp)2.0 * param_powern)) *
ucl_powr(tmp,-param_powern)));
if (tmp < param_c4) return (numtyp)0.0;
if (tmp < param_c3)
return (numtyp)-0.5*param_beta * ucl_powr(tmp,param_powern-(numtyp)1.0);
numtyp tmp_n = ucl_powr(tmp,param_powern);
return (numtyp)-0.5 * ucl_powr((numtyp)1.0+tmp_n, (numtyp) -
return (numtyp)-0.5 * ucl_powr((numtyp)1.0+tmp_n, (numtyp) -
(numtyp)1.0-((numtyp)1.0 / ((numtyp)2.0 * param_powern)))*tmp_n / zeta;
}
/* ---------------------------------------------------------------------- */
ucl_inline void ters_zetaterm_d(const numtyp prefactor,
const numtyp rij_hat[3],
const numtyp rij_hat[3],
const numtyp rij,
const numtyp rik_hat[3],
const numtyp rik_hat[3],
const numtyp rik,
const numtyp param_bigr,
const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_powermint,
const numtyp param_lam3,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_h,
const numtyp param_gamma,
numtyp dri[3],
numtyp drj[3],
@ -229,7 +231,7 @@ ucl_inline void ters_zetaterm_d(const numtyp prefactor,
else if (tmp < (numtyp)-69.0776) ex_delr = (numtyp)0.0;
else ex_delr = ucl_exp(tmp);
if ((int)param_powermint == 3)
if ((int)param_powermint == 3)
ex_delr_d = (numtyp)3.0*param_lam3*t*t*ex_delr;
else ex_delr_d = param_lam3 * ex_delr;
@ -269,17 +271,17 @@ ucl_inline void ters_zetaterm_d(const numtyp prefactor,
}
ucl_inline void ters_zetaterm_d_fi(const numtyp prefactor,
const numtyp rij_hat[3],
const numtyp rij_hat[3],
const numtyp rij,
const numtyp rik_hat[3],
const numtyp rik_hat[3],
const numtyp rik,
const numtyp param_bigr,
const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_powermint,
const numtyp param_lam3,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_h,
const numtyp param_gamma,
numtyp dri[3])
{
@ -297,7 +299,7 @@ ucl_inline void ters_zetaterm_d_fi(const numtyp prefactor,
else if (tmp < (numtyp)-69.0776) ex_delr = (numtyp)0.0;
else ex_delr = ucl_exp(tmp);
if ((int)param_powermint == 3)
if ((int)param_powermint == 3)
ex_delr_d = (numtyp)3.0*param_lam3*t*t*ex_delr;
else ex_delr_d = param_lam3 * ex_delr;
@ -319,17 +321,17 @@ ucl_inline void ters_zetaterm_d_fi(const numtyp prefactor,
}
ucl_inline void ters_zetaterm_d_fj(const numtyp prefactor,
const numtyp rij_hat[3],
const numtyp rij_hat[3],
const numtyp rij,
const numtyp rik_hat[3],
const numtyp rik_hat[3],
const numtyp rik,
const numtyp param_bigr,
const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_powermint,
const numtyp param_lam3,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_h,
const numtyp param_gamma,
numtyp drj[3])
{
@ -346,7 +348,7 @@ ucl_inline void ters_zetaterm_d_fj(const numtyp prefactor,
else if (tmp < (numtyp)-69.0776) ex_delr = (numtyp)0.0;
else ex_delr = ucl_exp(tmp);
if ((int)param_powermint == 3)
if ((int)param_powermint == 3)
ex_delr_d = (numtyp)3.0*param_lam3*t*t*ex_delr;
else ex_delr_d = param_lam3 * ex_delr;
@ -365,17 +367,17 @@ ucl_inline void ters_zetaterm_d_fj(const numtyp prefactor,
}
ucl_inline void ters_zetaterm_d_fk(const numtyp prefactor,
const numtyp rij_hat[3],
const numtyp rij_hat[3],
const numtyp rij,
const numtyp rik_hat[3],
const numtyp rik_hat[3],
const numtyp rik,
const numtyp param_bigr,
const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_powermint,
const numtyp param_lam3,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_h,
const numtyp param_gamma,
numtyp drk[3])
{
@ -393,7 +395,7 @@ ucl_inline void ters_zetaterm_d_fk(const numtyp prefactor,
else if (tmp < (numtyp)-69.0776) ex_delr = (numtyp)0.0;
else ex_delr = ucl_exp(tmp);
if ((int)param_powermint == 3)
if ((int)param_powermint == 3)
ex_delr_d = (numtyp)3.0*param_lam3*t*t*ex_delr;
else ex_delr_d = param_lam3 * ex_delr;
@ -419,20 +421,20 @@ ucl_inline void repulsive(const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_lam1,
const numtyp param_biga,
const numtyp rsq,
const int eflag,
const numtyp rsq,
const int eflag,
numtyp *ans)
{
numtyp r,tmp_fc,tmp_fc_d,tmp_exp;
r = ucl_sqrt(rsq);
tmp_fc = ters_fc(r,param_bigr,param_bigd);
tmp_fc_d = ters_fc_d(r,param_bigr,param_bigd);
tmp_exp = ucl_exp(-param_lam1 * r);
{
numtyp r,tmp_fc,tmp_fc_d,tmp_exp;
r = ucl_sqrt(rsq);
tmp_fc = ters_fc(r,param_bigr,param_bigd);
tmp_fc_d = ters_fc_d(r,param_bigr,param_bigd);
tmp_exp = ucl_exp(-param_lam1 * r);
// fforce
ans[0] = -param_biga*tmp_exp*(tmp_fc_d - tmp_fc*param_lam1)*ucl_recip(r);
ans[0] = -param_biga*tmp_exp*(tmp_fc_d - tmp_fc*param_lam1)*ucl_recip(r);
// eng
if (eflag) ans[1] = tmp_fc * param_biga * tmp_exp;
}
if (eflag) ans[1] = tmp_fc * param_biga * tmp_exp;
}
/* ---------------------------------------------------------------------- */
@ -441,7 +443,7 @@ ucl_inline numtyp zeta(const numtyp param_powermint,
const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_c,
const numtyp param_d,
const numtyp param_d,
const numtyp param_h,
const numtyp param_gamma,
const numtyp rsqij,
@ -464,23 +466,23 @@ ucl_inline numtyp zeta(const numtyp param_powermint,
else if (arg < (numtyp)-69.0776) ex_delr = (numtyp)0.0;
else ex_delr = ucl_exp(arg);
return ters_fc(rik,param_bigr,param_bigd) *
return ters_fc(rik,param_bigr,param_bigd) *
ters_gijk(costheta,param_c, param_d, param_h, param_gamma) * ex_delr;
}
/* ---------------------------------------------------------------------- */
ucl_inline void force_zeta(const numtyp param_bigb,
ucl_inline void force_zeta(const numtyp param_bigb,
const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_lam2,
const numtyp param_beta,
const numtyp param_beta,
const numtyp param_powern,
const numtyp param_c1,
const numtyp param_c2,
const numtyp param_c1,
const numtyp param_c2,
const numtyp param_c3,
const numtyp param_c4,
const numtyp rsq,
const numtyp rsq,
const numtyp zeta_ij,
const int eflag,
numtyp fpfeng[4])
@ -494,7 +496,7 @@ ucl_inline void force_zeta(const numtyp param_bigb,
param_c1,param_c2, param_c3, param_c4);
fpfeng[0] = (numtyp)0.5*bij*fa_d * ucl_recip(r); // fforce
fpfeng[1] = (numtyp)-0.5*fa * ters_bij_d(zeta_ij,param_beta, param_powern,
param_c1,param_c2, param_c3, param_c4); // prefactor
param_c1,param_c2, param_c3, param_c4); // prefactor
if (eflag) fpfeng[2] = (numtyp)0.5*bij*fa; // eng
}
@ -504,23 +506,23 @@ ucl_inline void force_zeta(const numtyp param_bigb,
use param_ijk cutoff for rik test
------------------------------------------------------------------------- */
ucl_inline void attractive(const numtyp param_bigr,
ucl_inline void attractive(const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_powermint,
const numtyp param_lam3,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_h,
const numtyp param_gamma,
const numtyp prefactor,
const numtyp rij,
const numtyp rijinv,
const numtyp rij,
const numtyp rijinv,
const numtyp rik,
const numtyp rikinv,
const numtyp delrij[3],
const numtyp rikinv,
const numtyp delrij[3],
const numtyp delrik[3],
numtyp fi[3],
numtyp fj[3],
numtyp fi[3],
numtyp fj[3],
numtyp fk[3])
{
numtyp rij_hat[3],rik_hat[3];
@ -531,20 +533,20 @@ ucl_inline void attractive(const numtyp param_bigr,
param_c, param_d, param_h, param_gamma, fi, fj, fk);
}
ucl_inline void attractive_fi(const numtyp param_bigr,
ucl_inline void attractive_fi(const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_powermint,
const numtyp param_lam3,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_h,
const numtyp param_gamma,
const numtyp prefactor,
const numtyp rij,
const numtyp rijinv,
const numtyp rij,
const numtyp rijinv,
const numtyp rik,
const numtyp rikinv,
const numtyp delrij[3],
const numtyp rikinv,
const numtyp delrij[3],
const numtyp delrik[3],
numtyp fi[3])
{
@ -556,20 +558,20 @@ ucl_inline void attractive_fi(const numtyp param_bigr,
param_c, param_d, param_h, param_gamma, fi);
}
ucl_inline void attractive_fj(const numtyp param_bigr,
ucl_inline void attractive_fj(const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_powermint,
const numtyp param_lam3,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_h,
const numtyp param_gamma,
const numtyp prefactor,
const numtyp rij,
const numtyp rijinv,
const numtyp rij,
const numtyp rijinv,
const numtyp rik,
const numtyp rikinv,
const numtyp delrij[3],
const numtyp rikinv,
const numtyp delrij[3],
const numtyp delrik[3],
numtyp fj[3])
{
@ -581,20 +583,20 @@ ucl_inline void attractive_fj(const numtyp param_bigr,
param_c, param_d, param_h, param_gamma, fj);
}
ucl_inline void attractive_fk(const numtyp param_bigr,
ucl_inline void attractive_fk(const numtyp param_bigr,
const numtyp param_bigd,
const numtyp param_powermint,
const numtyp param_lam3,
const numtyp param_c,
const numtyp param_d,
const numtyp param_h,
const numtyp param_h,
const numtyp param_gamma,
const numtyp prefactor,
const numtyp rij,
const numtyp rijinv,
const numtyp rij,
const numtyp rijinv,
const numtyp rik,
const numtyp rikinv,
const numtyp delrij[3],
const numtyp rikinv,
const numtyp delrij[3],
const numtyp delrik[3],
numtyp fk[3])
{