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
|
/* StarPU --- Runtime system for heterogeneous multicore architectures.
*
* Copyright (C) 2009-2011 Université de Bordeaux 1
* Copyright (C) 2010 Mehdi Juhoor <mjuhoor@gmail.com>
* Copyright (C) 2010, 2011, 2012 Centre National de la Recherche Scientifique
*
* StarPU is free software; you can redistribute it and/or modify
* it under the terms of the GNU Lesser General Public License as published by
* the Free Software Foundation; either version 2.1 of the License, or (at
* your option) any later version.
*
* StarPU is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
*
* See the GNU Lesser General Public License in COPYING.LGPL for more details.
*/
#include <starpu_mpi.h>
#include "mpi_cholesky.h"
#include "mpi_cholesky_models.h"
/*
* Create the codelets
*/
static struct starpu_codelet cl11 =
{
.where = STARPU_CPU|STARPU_CUDA,
.cpu_funcs = {chol_cpu_codelet_update_u11, NULL},
#ifdef STARPU_USE_CUDA
.cuda_funcs = {chol_cublas_codelet_update_u11, NULL},
#endif
.nbuffers = 1,
.modes = {STARPU_RW},
.model = &chol_model_11
};
static struct starpu_codelet cl21 =
{
.where = STARPU_CPU|STARPU_CUDA,
.cpu_funcs = {chol_cpu_codelet_update_u21, NULL},
#ifdef STARPU_USE_CUDA
.cuda_funcs = {chol_cublas_codelet_update_u21, NULL},
#endif
.nbuffers = 2,
.modes = {STARPU_R, STARPU_RW},
.model = &chol_model_21
};
static struct starpu_codelet cl22 =
{
.where = STARPU_CPU|STARPU_CUDA,
.cpu_funcs = {chol_cpu_codelet_update_u22, NULL},
#ifdef STARPU_USE_CUDA
.cuda_funcs = {chol_cublas_codelet_update_u22, NULL},
#endif
.nbuffers = 3,
.modes = {STARPU_R, STARPU_R, STARPU_RW},
.model = &chol_model_22
};
/* Returns the MPI node number where data indexes index is */
int my_distrib(int x, int y, int nb_nodes)
{
return (x+y) % nb_nodes;
}
/*
* code to bootstrap the factorization
* and construct the DAG
*/
static void dw_cholesky(float ***matA, unsigned size, unsigned ld, unsigned nblocks, int rank, int nodes)
{
struct timeval start;
struct timeval end;
starpu_data_handle_t **data_handles;
int x, y;
/* create all the DAG nodes */
unsigned i,j,k;
data_handles = malloc(nblocks*sizeof(starpu_data_handle_t *));
for(x=0 ; x<nblocks ; x++) data_handles[x] = malloc(nblocks*sizeof(starpu_data_handle_t));
starpu_mpi_barrier(MPI_COMM_WORLD);
gettimeofday(&start, NULL);
for(x = 0; x < nblocks ; x++)
{
for (y = 0; y < nblocks; y++)
{
int mpi_rank = my_distrib(x, y, nodes);
if (mpi_rank == rank)
{
//fprintf(stderr, "[%d] Owning data[%d][%d]\n", rank, x, y);
starpu_matrix_data_register(&data_handles[x][y], 0, (uintptr_t)matA[x][y],
ld, size/nblocks, size/nblocks, sizeof(float));
}
/* TODO: make better test to only registering what is needed */
else
{
/* I don't own that index, but will need it for my computations */
//fprintf(stderr, "[%d] Neighbour of data[%d][%d]\n", rank, x, y);
starpu_matrix_data_register(&data_handles[x][y], -1, (uintptr_t)NULL,
ld, size/nblocks, size/nblocks, sizeof(float));
}
if (data_handles[x][y])
{
starpu_data_set_rank(data_handles[x][y], mpi_rank);
starpu_data_set_tag(data_handles[x][y], (y*nblocks)+x);
}
}
}
for (k = 0; k < nblocks; k++)
{
int prio = STARPU_DEFAULT_PRIO;
if (!noprio) prio = STARPU_MAX_PRIO;
starpu_mpi_insert_task(MPI_COMM_WORLD, &cl11,
STARPU_PRIORITY, prio,
STARPU_RW, data_handles[k][k],
0);
for (j = k+1; j<nblocks; j++)
{
prio = STARPU_DEFAULT_PRIO;
if (!noprio&& (j == k+1)) prio = STARPU_MAX_PRIO;
starpu_mpi_insert_task(MPI_COMM_WORLD, &cl21,
STARPU_PRIORITY, prio,
STARPU_R, data_handles[k][k],
STARPU_RW, data_handles[k][j],
0);
for (i = k+1; i<nblocks; i++)
{
if (i <= j)
{
prio = STARPU_DEFAULT_PRIO;
if (!noprio && (i == k + 1) && (j == k +1) ) prio = STARPU_MAX_PRIO;
starpu_mpi_insert_task(MPI_COMM_WORLD, &cl22,
STARPU_PRIORITY, prio,
STARPU_R, data_handles[k][i],
STARPU_R, data_handles[k][j],
STARPU_RW, data_handles[i][j],
0);
}
}
}
}
starpu_task_wait_for_all();
for(x = 0; x < nblocks ; x++)
{
for (y = 0; y < nblocks; y++)
{
if (data_handles[x][y])
starpu_data_unregister(data_handles[x][y]);
}
free(data_handles[x]);
}
free(data_handles);
starpu_mpi_barrier(MPI_COMM_WORLD);
gettimeofday(&end, NULL);
if (rank == 0)
{
double timing = (double)((end.tv_sec - start.tv_sec)*1000000 + (end.tv_usec - start.tv_usec));
fprintf(stderr, "Computation took (in ms)\n");
fprintf(stdout, "%2.2f\n", timing/1000);
double flop = (1.0f*size*size*size)/3.0f;
fprintf(stderr, "Synthetic GFlops : %2.2f\n", (flop/timing/1000.0f));
}
}
int main(int argc, char **argv)
{
/* create a simple definite positive symetric matrix example
*
* Hilbert matrix : h(i,j) = 1/(i+j+1)
* */
float ***bmat;
int rank, nodes;
parse_args(argc, argv);
struct starpu_conf conf;
starpu_conf_init(&conf);
conf.sched_policy_name = "heft";
conf.calibrate = 1;
int ret = starpu_init(&conf);
STARPU_CHECK_RETURN_VALUE(ret, "starpu_init");
starpu_mpi_initialize_extended(&rank, &nodes);
starpu_helper_cublas_init();
unsigned i,j,x,y;
bmat = malloc(nblocks * sizeof(float *));
for(x=0 ; x<nblocks ; x++)
{
bmat[x] = malloc(nblocks * sizeof(float *));
for(y=0 ; y<nblocks ; y++)
{
int mpi_rank = my_distrib(x, y, nodes);
if (mpi_rank == rank)
{
starpu_malloc((void **)&bmat[x][y], BLOCKSIZE*BLOCKSIZE*sizeof(float));
for (i = 0; i < BLOCKSIZE; i++)
{
for (j = 0; j < BLOCKSIZE; j++)
{
bmat[x][y][j +i*BLOCKSIZE] = (1.0f/(1.0f+(i+(x*BLOCKSIZE)+j+(y*BLOCKSIZE)))) + ((i+(x*BLOCKSIZE) == j+(y*BLOCKSIZE))?1.0f*size:0.0f);
//mat[j +i*size] = ((i == j)?1.0f*size:0.0f);
}
}
}
}
}
dw_cholesky(bmat, size, size/nblocks, nblocks, rank, nodes);
starpu_mpi_shutdown();
for(x=0 ; x<nblocks ; x++)
{
for(y=0 ; y<nblocks ; y++)
{
int mpi_rank = my_distrib(x, y, nodes);
if (mpi_rank == rank)
{
starpu_free((void *)bmat[x][y]);
}
}
free(bmat[x]);
}
free(bmat);
starpu_helper_cublas_shutdown();
starpu_shutdown();
return 0;
}
|
