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
|
/***************************************************************************
coul_debye_ext.cpp
-------------------
Trung Dac Nguyen
Functions for LAMMPS access to coul/debye acceleration routines.
__________________________________________________________________________
This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
__________________________________________________________________________
begin :
email : ndtrung@umich.edu
***************************************************************************/
#include <iostream>
#include <cassert>
#include <cmath>
#include "lal_coul_debye.h"
using namespace std;
using namespace LAMMPS_AL;
static CoulDebye<PRECISION,ACC_PRECISION> CDEMF;
// ---------------------------------------------------------------------------
// Allocate memory on host and device and copy constants to device
// ---------------------------------------------------------------------------
int cdebye_gpu_init(const int ntypes, double **host_scale, double **cutsq,
double *host_special_coul, const int inum,
const int nall, const int max_nbors, const int maxspecial,
const double cell_size, int &gpu_mode, FILE *screen,
const double qqrd2e, const double kappa) {
CDEMF.clear();
gpu_mode=CDEMF.device->gpu_mode();
double gpu_split=CDEMF.device->particle_split();
int first_gpu=CDEMF.device->first_device();
int last_gpu=CDEMF.device->last_device();
int world_me=CDEMF.device->world_me();
int gpu_rank=CDEMF.device->gpu_rank();
int procs_per_gpu=CDEMF.device->procs_per_gpu();
CDEMF.device->init_message(screen,"coul/debye",first_gpu,last_gpu);
bool message=false;
if (CDEMF.device->replica_me()==0 && screen)
message=true;
if (message) {
fprintf(screen,"Initializing GPU and compiling on process 0...");
fflush(screen);
}
int init_ok=0;
if (world_me==0)
init_ok=CDEMF.init(ntypes, host_scale, cutsq, host_special_coul, inum, nall, max_nbors,
maxspecial, cell_size, gpu_split, screen, qqrd2e, kappa);
CDEMF.device->world_barrier();
if (message)
fprintf(screen,"Done.\n");
for (int i=0; i<procs_per_gpu; i++) {
if (message) {
if (last_gpu-first_gpu==0)
fprintf(screen,"Initializing GPU %d on core %d...",first_gpu,i);
else
fprintf(screen,"Initializing GPUs %d-%d on core %d...",first_gpu,
last_gpu,i);
fflush(screen);
}
if (gpu_rank==i && world_me!=0)
init_ok=CDEMF.init(ntypes, host_scale, cutsq, host_special_coul, inum, nall, max_nbors,
maxspecial, cell_size, gpu_split, screen, qqrd2e, kappa);
CDEMF.device->gpu_barrier();
if (message)
fprintf(screen,"Done.\n");
}
if (message)
fprintf(screen,"\n");
if (init_ok==0)
CDEMF.estimate_gpu_overhead();
return init_ok;
}
// ---------------------------------------------------------------------------
// Copy updated constants to device
// ---------------------------------------------------------------------------
void cdebye_gpu_reinit(const int ntypes, double **host_scale) {
int world_me=CDEMF.device->world_me();
int gpu_rank=CDEMF.device->gpu_rank();
int procs_per_gpu=CDEMF.device->procs_per_gpu();
if (world_me==0)
CDEMF.reinit(ntypes, host_scale);
CDEMF.device->world_barrier();
for (int i=0; i<procs_per_gpu; i++) {
if (gpu_rank==i && world_me!=0)
CDEMF.reinit(ntypes, host_scale);
CDEMF.device->gpu_barrier();
}
}
void cdebye_gpu_clear() {
CDEMF.clear();
}
int** cdebye_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) {
return CDEMF.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,
host_q, boxlo, prd);
}
void cdebye_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) {
CDEMF.compute(ago,inum_full,nall,host_x,host_type,ilist,numj,firstneigh,eflag,
vflag,eatom,vatom,host_start,cpu_time,success,host_q,
nlocal,boxlo,prd);
}
double cdebye_gpu_bytes() {
return CDEMF.host_memory_usage();
}
|
