-
Notifications
You must be signed in to change notification settings - Fork 0
/
newton.cpp
255 lines (224 loc) · 6.62 KB
/
newton.cpp
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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
#include "stream_util.h"
#include "memory_management.h"
#include "patch.h"
#include "patch_data.h"
#include "patch_force.h"
#include "domain_decomp.h"
#include "communication.h"
#include "initialize_random.h"
#include "read_magi.h"
#include "solver.h"
#include "write_vtk.h"
#include "command_line.h"
#include "timer_stack.h"
#if defined(NEWTONPP_ENABLE_SENSEI)
#include "insitu.h"
#endif
#include <iostream>
#include <vector>
#include <mpi.h>
#if defined(_OPENMP)
#include <omp.h>
#endif
#include <chrono>
using namespace std::literals;
using timer = std::chrono::high_resolution_clock;
#if defined(NEWTONPP_ENABLE_OMP)
#pragma message("the default allocator targets the GPU")
#else
#pragma message("the default allocator targets the CPU")
#endif
#if defined(NEWTONPP_ENABLE_CUDA)
#pragma message("Stream compact on the GPU")
#else
#pragma message("Stream compact on the CPU")
#endif
#if defined(NEWTONPP_GPU_DIRECT)
#pragma message("GPU direct is used for MPI communication")
#else
#pragma message("Data copy to the CPU for MPI communication")
#endif
#if defined(NEWTONPP_USE_OMP_LOOP)
#pragma message("OpenMP offload directive: target teams loop")
#else
#pragma message("OpenMP offload directive: target teams distribute parallel for")
#endif
int main(int argc, char **argv)
{
int rank = 0;
int n_ranks = 1;
MPI_Comm comm = MPI_COMM_WORLD;
#if 1
int thread_level = 0;
if ((MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &thread_level) != MPI_SUCCESS) ||
(thread_level != MPI_THREAD_MULTIPLE))
{
std::cerr << "Error: failed to initialize MPI" << std::endl;
return -1;
}
#else
MPI_Init(&argc, &argv);
#endif
MPI_Comm_rank(comm, &rank);
MPI_Comm_size(comm, &n_ranks);
timer_stack timer(rank == 0);
timer.push();
timer.push();
double h = 1e5; // time step size
double nfr = 0.; // distance for reduced representation
double eps = 0.; // the softening length
double G = 6.67408e-11; // the gravitational constant
long n_its = 0; // number of solver steps
long n_bodies = 0; // number of bodies
long part_int = 4; // how often to partition particles
const char *magi_h5 = nullptr; // where initial positions/velocities can be found
const char *magi_sum = nullptr; // where particle types can be found
const char *out_dir = nullptr; // directory to write results at
long io_int = 0; // how often to write resutls
const char *is_conf = nullptr; // sensei in situ configuration file
long is_int = 0; // how often to invoke in situ processing
#if defined(_OPENMP)
int num_devs = omp_get_num_devices();
#else
int num_devs = 1;
#endif
int start_dev = 0;
int dev_stride = 1;
if (parse_command_line(argc, argv, comm, num_devs, start_dev, dev_stride,
G, h, eps, nfr, n_its, n_bodies, part_int, magi_h5, magi_sum, out_dir,
io_int, is_conf, is_int))
return -1;
#if defined(_OPENMP)
// set the device to use
int dev = rank % num_devs * dev_stride + start_dev;
omp_set_default_device(dev);
//std::cerr << " === newton++ === : rank " << rank << " device " << dev << " of " << num_devs << std::endl;
#endif
// load the initial condition and initialize the bodies
patch_data pd;
patch_force pf;
std::vector<patch> patches;
timer.push();
#if defined(NEWTONPP_ENABLE_MAGI)
if (magi_h5)
{
// load the ic positions and velocities
patch dom;
if (magi_h5 && read_magi(comm, magi_h5, magi_sum, dom, pd))
return -1;
// decompose domain
patches = partition(dom, n_ranks);
assign_patches(patches, n_ranks);
// update partition
if (n_ranks > 1)
{
pf.resize(pd.size());
hamr::buffer<int> dest(def_alloc());
partition(comm, patches, pd, dest);
move2(comm, pd, pf, dest);
}
}
else
#endif
{
if (initialize_random(comm, n_bodies, patches, pd, nfr))
return -1;
}
timer.pop("read ic");
#if defined(NEWTONPP_ENABLE_SENSEI)
// initialize for in-situ
insitu_data is_data;
if (is_conf && is_int)
{
timer.push();
if (init_insitu(comm, is_conf, is_data))
return -1;
timer.pop("sensei init");
}
#endif
// write the domain decomp
if (io_int)
{
timer.push();
write_vtk(comm, patches, out_dir);
timer.pop("write dom");
}
// flag nearby patches
timer.push();
std::vector<int> nf;
near(patches, nfr, nf);
timer.pop("build tree");
// initialize forces
timer.push();
forces(comm, pd, pf, G, eps, nf);
timer.pop("init forces");
// write initial state
if (io_int)
{
timer.push();
write_vtk(comm, pd, pf, out_dir);
timer.pop("write part");
}
#if defined(NEWTONPP_ENABLE_SENSEI)
// process initial state
if (is_int && is_data)
{
timer.push();
if (update_insitu(comm, is_data, 0, 0, patches, pd, pf))
return -1;
timer.pop("sensei upd");
}
#endif
timer.pop("=== initialization ===");
// iterate
long it = 0;
while (it < n_its)
{
timer.push();
#if defined(_OPENMP)
omp_set_default_device(dev);
#endif
// update bodies
timer.push();
velocity_verlet(comm, pd, pf, G, h, eps, nf);
timer.pop("integrate part");
// update partition
if (part_int && (((it + 1) % part_int) == 0))
{
timer.push();
hamr::buffer<int> dest(def_alloc());
partition(comm, patches, pd, dest);
timer.pop_push("partition part");
move2(comm, pd, pf, dest);
timer.pop("move part");
}
// write current state
if (io_int && (((it + 1) % io_int) == 0))
{
timer.push();
write_vtk(comm, pd, pf, out_dir);
timer.pop("write part");
}
#if defined(NEWTONPP_ENABLE_SENSEI)
// process current state
if ((((it + 1) % is_int) == 0) && is_data)
{
timer.push();
if (update_insitu(comm, is_data, it, it*h, patches, pd, pf))
return -1;
timer.pop("sensei upd");
}
#endif
it += 1;
timer.pop("=== loop iteration ===");
}
#if defined(NEWTONPP_ENABLE_SENSEI)
// finalize in-situ processing
timer.push();
if (is_int && is_data && finalize_insitu(comm, is_data))
return -1;
timer.pop("fin sensei");
#endif
MPI_Finalize();
return 0;
}