1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
|
/***************************************************************************
lj.cpp
-------------------
W. Michael Brown (ORNL)
Class for acceleration of the lj/cut pair style.
__________________________________________________________________________
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
__________________________________________________________________________
begin :
email : brownw@ornl.gov
***************************************************************************/
#if defined(USE_OPENCL)
#include "lj_cl.h"
#elif defined(USE_CUDART)
const char *lj=0;
#else
#include "lj_cubin.h"
#endif
#include "lal_lj.h"
#include <cassert>
namespace LAMMPS_AL {
#define LJT LJ<numtyp, acctyp>
extern Device<PRECISION,ACC_PRECISION> device;
template <class numtyp, class acctyp>
LJT::LJ() : BaseAtomic<numtyp,acctyp>(), _allocated(false) {
}
template <class numtyp, class acctyp>
LJT::~LJ() {
clear();
}
template <class numtyp, class acctyp>
int LJT::bytes_per_atom(const int max_nbors) const {
return this->bytes_per_atom_atomic(max_nbors);
}
template <class numtyp, class acctyp>
int LJT::init(const int ntypes,
double **host_cutsq, 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) {
const int max_shared_types=this->device->max_shared_types();
int onetype=0;
#ifdef USE_OPENCL
if (maxspecial==0)
for (int i=1; i<ntypes; i++)
for (int j=i; j<ntypes; j++)
if (host_cutsq[i][j]>0) {
if (onetype>0)
onetype=-1;
else if (onetype==0)
onetype=i*max_shared_types+j;
}
if (onetype<0) onetype=0;
#endif
int success;
success=this->init_atomic(nlocal,nall,max_nbors,maxspecial,cell_size,
gpu_split,_screen,lj,"k_lj",onetype);
if (success!=0)
return success;
// If atom type constants fit in shared memory use fast kernel
int lj_types=ntypes;
shared_types=false;
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);
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,
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;
}
template <class numtyp, class acctyp>
void LJT::reinit(const int ntypes, double **host_cutsq, double **host_lj1,
double **host_lj2, double **host_lj3,
double **host_lj4, double **host_offset) {
// 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;
this->atom->type_pack4(ntypes,_lj_types,lj1,host_write,host_lj1,host_lj2,
host_cutsq);
this->atom->type_pack4(ntypes,_lj_types,lj3,host_write,host_lj3,host_lj4,
host_offset);
}
template <class numtyp, class acctyp>
void LJT::clear() {
if (!_allocated)
return;
_allocated=false;
lj1.clear();
lj3.clear();
sp_lj.clear();
this->clear_atomic();
}
template <class numtyp, class acctyp>
double LJT::host_memory_usage() const {
return this->host_memory_usage_atomic()+sizeof(LJ<numtyp,acctyp>);
}
// ---------------------------------------------------------------------------
// Calculate energies, forces, and torques
// ---------------------------------------------------------------------------
template <class numtyp, class acctyp>
int LJT::loop(const int eflag, const int vflag) {
// Compute the block size and grid size to keep all cores busy
const int BX=this->block_size();
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_sel->set_size(GX,BX);
this->k_pair_sel->run(&this->atom->x, &lj1, &lj3, &sp_lj,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv, &eflag,
&vflag, &ainum, &nbor_pitch,
&this->_threads_per_atom);
} else {
this->k_pair.set_size(GX,BX);
this->k_pair.run(&this->atom->x, &lj1, &lj3, &_lj_types, &sp_lj,
&this->nbor->dev_nbor, &this->_nbor_data->begin(),
&this->ans->force, &this->ans->engv, &eflag, &vflag,
&ainum, &nbor_pitch, &this->_threads_per_atom);
}
this->time_pair.stop();
return GX;
}
template class LJ<PRECISION,ACC_PRECISION>;
}
|
