-
Notifications
You must be signed in to change notification settings - Fork 2
/
osc.cpp
307 lines (276 loc) · 9.51 KB
/
osc.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
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
#include <mpi.h>
#include <omp.h>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <limits>
#define CONFIG_LEN 1024
// communication modes
#define COMM_SENDRECV 0
#define COMM_RMA_ACTV 1
#define COMM_RMA_FENC 2
#define COMM_RMA_LOCK 3
#define MEM_DATATYPE 0
#define MEM_CONTIG 1
#define RMA_PUT 0
#define RMA_GET 1
#define ALLOC_WIN 0
#define ALLOC_USR 1
#ifdef COMM_MODE
#define M_COMM COMM_MODE
#else
#define M_COMM COMM_SENDRECV
#endif
#ifdef MEM
#define M_MEM MEM
#else
#define M_MEM MEM_CONTIG
#endif
// the total communication size will be N^3 and split-up in NS^3 chunks, each chunk will be sent out // independently
#ifdef N
#define M_N N
#else
#define M_N 100
#endif
#ifdef NS
#define M_NS NS
#else
#define M_NS 5 // NS is the size of the subarray (exchanged as one block)
#endif
#ifdef RMA
#define M_RMA RMA
#else
#define M_RMA RMA_PUT
#endif
#ifdef ALLOC
#define M_ALLOC ALLOC
#else
#define M_ALLOC ALLOC_WIN
#endif
#define IWARM 2
#define IMAX 20
static int get_next_rank(const int rank, const int comm_size) {
return (rank + comm_size / 2) % comm_size;
}
static int get_prev_rank(const int rank, const int comm_size) {
return (rank + comm_size / 2) % comm_size;
}
int main(int argc, char** argv){
MPI_Init(&argc,&argv);
//--------------------------------------------------------------------------
// get usefull MPI information
int rank, comm_size;
MPI_Comm_rank(MPI_COMM_WORLD,&rank);
MPI_Comm_size(MPI_COMM_WORLD,&comm_size);
//// get the name and check if we are on the same node or not
//char name[MPI_MAX_PROCESSOR_NAME];
//int len;
//MPI_Get_processor_name(name, &len);
//printf("[rank %d] the world is %d ranks wide and my name is %s\n", rank, comm_size, name);
//fflush(stdout);
//--------------------------------------------------------------------------
// allocate the array of size [N x N x N]
int size = M_N * M_N * M_N;
#if (M_ALLOC == ALLOC_USR || M_COMM == COMM_SENDRECV)
int* array = (int*)malloc(sizeof(int) * size);
int* other = (int*)malloc(sizeof(int) * size);
MPI_Info info;
MPI_Info_create(&info);
MPI_Win window = MPI_WIN_NULL;
#if (M_RMA == RMA_GET)
MPI_Win_create(array, size * sizeof(int), sizeof(int), info, MPI_COMM_WORLD, &window);
#elif (M_RMA == RMA_PUT)
MPI_Win_create(other, size * sizeof(int), sizeof(int), info, MPI_COMM_WORLD, &window);
#endif
MPI_Info_free(&info);
#elif (M_ALLOC == ALLOC_WIN)
#if (M_RMA == RMA_GET)
int* array;
int* other = (int*)malloc(sizeof(int) * size);
MPI_Info info;
MPI_Info_create(&info);
MPI_Win window = MPI_WIN_NULL;
MPI_Win_allocate(size * sizeof(int), sizeof(int), info, MPI_COMM_WORLD,&array, &window);
MPI_Info_free(&info);
#elif (M_RMA == RMA_PUT)
int* other;
int* array = (int*)malloc(sizeof(int) * size);
MPI_Info info;
MPI_Info_create(&info);
MPI_Win window = MPI_WIN_NULL;
MPI_Win_allocate(size * sizeof(int), sizeof(int), info, MPI_COMM_WORLD,&other, &window);
MPI_Info_free(&info);
#endif
#endif
for (int i = 0; i < size; ++i) {
array[i] = rank;
other[i] = -1;
}
// create the communication groups
//int n_in_group = (is_comm);
//int next_rank = (rank == rank_comm[0]) * rank_comm[1] + (rank == rank_comm[1]) * rank_comm[0];
//int prev_rank = (rank == rank_comm[0]) * rank_comm[1] + (rank == rank_comm[1]) * rank_comm[0];
//if (is_comm) {
// printf("[rank %d] I will access rank %d and be accessed by rank %d\n", rank, next_rank, prev_rank);
// fflush(stdout);
//}
int n_in_group = 1;
int prev_rank = get_prev_rank(rank, comm_size);
int next_rank = get_next_rank(rank, comm_size);
MPI_Group prev_group, next_group, global_group;
MPI_Comm_group(MPI_COMM_WORLD, &global_group);
MPI_Group_incl(global_group, n_in_group, &prev_rank, &prev_group);
MPI_Group_incl(global_group, n_in_group, &next_rank, &next_group);
MPI_Group_free(&global_group);
#if (M_COMM == COMM_SENDRECV)
const int nreq = (M_N / M_NS) * (M_N / M_NS) * (M_N / M_NS);
MPI_Request *sreq, *rreq;
sreq = (MPI_Request *)malloc(sizeof(MPI_Request) * nreq);
rreq = (MPI_Request *)malloc(sizeof(MPI_Request) * nreq);
#elif (M_COMM == COMM_RMA_LOCK)
MPI_Request sreq, rreq;
#endif
//--------------------------------------------------------------------------
double mean_time;
for (int iter = 0; iter <(IWARM + IMAX); ++iter) {
double tic = MPI_Wtime();
#if (M_COMM == COMM_RMA_ACTV)
MPI_Win_post(prev_group, 0, window);
MPI_Win_start(next_group, 0, window);
#elif (M_COMM == COMM_RMA_FENC)
MPI_Win_fence(0, window);
#elif (M_COMM == COMM_RMA_LOCK)
// notify the prev rank (the one accessing me) that I am ready
MPI_Irecv(NULL,0,MPI_BYTE,prev_rank,0,MPI_COMM_WORLD,&rreq);
// wait for the notification of the next rank (the one I access)
MPI_Issend(NULL,0,MPI_BYTE,next_rank,0,MPI_COMM_WORLD,&sreq);
MPI_Wait(&sreq,MPI_STATUS_IGNORE);
MPI_Win_lock(MPI_LOCK_SHARED,next_rank,0,window);
#endif
// for each submatrix of size NSxNSxNS
int ireq = 0;
for (int i2 = 0; i2 < M_N; i2 += M_NS) {
for (int i1 = 0; i1 < M_N; i1 += M_NS) {
for (int i0 = 0; i0 < M_N; i0 += M_NS) {
#if (M_MEM == MEM_DATATYPE)
// get the datatype corresponding to the sub-array
MPI_Datatype type1, type2;
MPI_Type_create_hvector(M_NS, M_NS, M_N * sizeof(int), MPI_INT, &type1);
MPI_Type_create_hvector(M_NS, 1, M_N * M_N * sizeof(int), type1,
&type2);
MPI_Type_commit(&type2);
MPI_Type_free(&type1);
int offset = i0 + M_N * (i1 + M_N * i2);
#elif (M_MEM == MEM_CONTIG)
MPI_Datatype type2;
MPI_Type_create_hvector(1, M_NS * M_NS * M_NS, sizeof(int),
MPI_INT, &type2);
MPI_Type_commit(&type2);
int offset = (M_NS * M_NS * M_NS) * ireq;
#endif
#if (M_COMM == COMM_RMA_ACTV || M_COMM == COMM_RMA_FENC || M_COMM == COMM_RMA_LOCK)
#if (M_RMA == RMA_GET)
MPI_Get(other + offset, 1, type2, next_rank, offset, 1, type2, window);
#elif (M_RMA == RMA_PUT)
MPI_Put(array + offset, 1, type2, next_rank, offset, 1, type2, window);
#endif
#elif (M_COMM == COMM_SENDRECV)
MPI_Irecv(other + offset, 1, type2, prev_rank, 0, MPI_COMM_WORLD,
rreq + ireq);
MPI_Isend(array + offset, 1, type2, next_rank, 0, MPI_COMM_WORLD,
sreq + ireq);
#endif
ireq++;
// free the type
MPI_Type_free(&type2);
}
}
}
#if (M_COMM == COMM_RMA_ACTV)
MPI_Win_complete(window);
MPI_Win_wait(window);
#elif (M_COMM == COMM_RMA_FENC)
MPI_Win_fence(0, window);
#elif (M_COMM == COMM_SENDRECV)
MPI_Waitall(nreq, sreq, MPI_STATUSES_IGNORE);
MPI_Waitall(nreq, rreq, MPI_STATUSES_IGNORE);
#elif (M_COMM == COMM_RMA_LOCK)
// complete the previous recev request
MPI_Wait(&rreq,MPI_STATUS_IGNORE);
// the flush is not needed, the lock will do it
// MPI_Win_flush(next_rank, window);
MPI_Win_unlock(next_rank, window);
// notify the next rank (the one I access) that I am done
MPI_Irecv(NULL,0,MPI_BYTE,next_rank,1,MPI_COMM_WORLD,&rreq);
// wait for the notification of the prev rank (the one accessing my data)
MPI_Issend(NULL,0,MPI_BYTE,prev_rank,1,MPI_COMM_WORLD,&sreq);
MPI_Wait(&sreq,MPI_STATUS_IGNORE);
MPI_Wait(&rreq,MPI_STATUS_IGNORE);
#endif
double toc = MPI_Wtime();
if(iter >= IWARM){
mean_time += (toc - tic) / IMAX;
}
}
double mtime_global;
MPI_Reduce(&mean_time,&mtime_global,1,MPI_DOUBLE,MPI_SUM,0,MPI_COMM_WORLD);
mtime_global /= comm_size;
if (rank == 0) {
char config[CONFIG_LEN];
snprintf(config,CONFIG_LEN,"[%s, %s, %s, %s,",
#if (M_COMM == COMM_RMA_ACTV)
"RMA-ACTIV",
#elif (M_COMM == COMM_RMA_FENC)
"RMA-FENCE",
#elif (M_COMM == COMM_RMA_LOCK)
"RMA-LOCK ",
#elif (M_COMM == COMM_SENDRECV)
"SEND-RECV",
#endif
#if (M_ALLOC == ALLOC_WIN)
"ALLOC_WIN",
#elif (M_ALLOC == ALLOC_USR)
"ALLOC_USR",
#endif
#if (M_RMA == RMA_PUT)
"RMA-PUT",
#elif (M_RMA == RMA_GET)
"RMA-GET",
#endif
#if (M_MEM == MEM_DATATYPE)
"DATATYPE"
#elif (M_MEM == MEM_CONTIG)
"CONTIG"
#endif
);
fprintf(stdout,
"%s %d MSG, %6.ld B/MSG] time = %f ms > ttl bdw = %f GB/s\n",
config, (M_N / M_NS) * (M_N / M_NS) * (M_N / M_NS),
(M_NS * M_NS * M_NS * sizeof(int)),
mtime_global * 1e+3,
(M_N * M_N * M_N * sizeof(int)) / mtime_global / 1e+9);
fflush(stdout);
}
//--------------------------------------------------------------------------
// make sure we have the correct result
#if (M_COMM == COMM_SENDRECV)
free(sreq);
free(rreq);
#endif
MPI_Group_free(&next_group);
MPI_Group_free(&prev_group);
#if (M_ALLOC == ALLOC_USR || M_COMM == COMM_SENDRECV)
free(array);
free(other);
MPI_Win_free(&window);
#elif (M_ALLOC == ALLOC_WIN)
#if (M_RMA == RMA_PUT)
free(array);
MPI_Win_free(&window);
#elif (M_RMA == RMA_GET)
free(other);
MPI_Win_free(&window);
#endif
#endif
MPI_Finalize();
}