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
|
/*********************************************************************
*
* Copyright (C) 2014, Northwestern University and Argonne National Laboratory
* See COPYRIGHT notice in top-level directory.
*
*********************************************************************/
/* $Id$ */
/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* This example makes a number of nonblocking API calls, each writes a single
* column of a 2D integer array. Each process writes NX columns and any two
* consecutive columns are of nprocs columns distance apart from each other.
* In this case, the fileview of each process interleaves with all other
* processes.
* If simply concatenating fileviews of all the nonblocking calls will result
* in a fileview that violates the MPI-IO requirement on the fileview of which
* flattened file offsets must be monotonically non-decreasing. PnetCDF handles
* this case by breaking down each nonblocking call into a list of offset-length
* pairs, merging the pairs across multiple nonblocking calls, and sorting
* them into an increasing order. The sorted pairs are used to construct a
* fileview that meets the monotonically non-decreasing offset requirement,
* and thus the nonblocking requests can be serviced by a single MPI-IO call.
*
* The compile and run commands are given below, together with an ncmpidump of
* the output file.
*
* % mpicxx -O2 -o column_wise column_wise.cpp -lpnetcdf
* % mpiexec -l -n 4 ./column_wise /pvfs2/wkliao/testfile.nc
* 0: 0: myOff= 0 myNX= 4
* 1: 1: myOff= 4 myNX= 4
* 2: 2: myOff= 8 myNX= 4
* 3: 3: myOff= 12 myNX= 4
* 0: 0: start= 0 0 count= 10 1
* 1: 1: start= 0 1 count= 10 1
* 2: 2: start= 0 2 count= 10 1
* 3: 3: start= 0 3 count= 10 1
*
* % ncmpidump /pvfs2/wkliao/testfile.nc
* netcdf testfile {
* // file format: CDF-5 (big variables)
* dimensions:
* Y = 10 ;
* X = 16 ;
* variables:
* int var(Y, X) ;
* data:
*
* var =
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3,
* 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3, 0, 1, 2, 3 ;
* }
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
#include <stdio.h>
#include <stdlib.h>
#include <iostream>
using namespace std;
#include <string.h> /* strcpy(), strncpy() */
#include <unistd.h> /* getopt() */
#include <pnetcdf>
using namespace PnetCDF;
using namespace PnetCDF::exceptions;
#define NY 10
#define NX 4
static void
usage(char *argv0)
{
cerr <<
"Usage: %s [-h] | [-q] [file_name]\n"
" [-h] Print help\n"
" [-q] Quiet mode (reports when fail)\n"
" [filename] output netCDF file name\n"
<< argv0;
}
int main(int argc, char** argv)
{
extern int optind;
char filename[256];
int i, j, verbose=1, rank, nprocs, myNX, G_NX, myOff, num_reqs;
int *reqs, *sts, **buf;
vector<MPI_Offset> start(2), count(2);
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
/* get command-line arguments */
while ((i = getopt(argc, argv, "hq")) != EOF)
switch(i) {
case 'q': verbose = 0;
break;
case 'h':
default: if (rank==0) usage(argv[0]);
MPI_Finalize();
return 1;
}
if (argv[optind] == NULL) strcpy(filename, "testfile.nc");
else snprintf(filename, 256, "%s", argv[optind]);
try {
NcmpiFile nc(MPI_COMM_WORLD, filename, NcmpiFile::replace,
NcmpiFile::classic5);
/* the global array is NY * (NX * nprocs) */
G_NX = NX * nprocs;
myOff = NX * rank;
myNX = NX;
if (verbose) printf("%2d: myOff=%3d myNX=%3d\n",rank,myOff,myNX);
/* define dimensions Y and X */
vector<NcmpiDim> dimid(2);
dimid[0] = nc.addDim("Y", NY);
dimid[1] = nc.addDim("X", G_NX);
/* define a 2D variable of integer type */
NcmpiVar var = nc.addVar("var", ncmpiInt, dimid);
/* First, fill the entire array with zeros, using a blocking I/O.
Every process writes a subarray of size NY * myNX */
buf = (int**) malloc(sizeof(int*) * myNX);
buf[0] = (int*) calloc(NY * myNX, sizeof(int));
start[0] = 0; start[1] = myOff;
count[0] = NY; count[1] = myNX;
var.putVar_all(start, count, &buf[0][0]);
free(buf[0]);
nc.flush();
/* initialize the buffer with rank ID. Also make the case interesting,
by allocating buffers separately */
for (i=0; i<myNX; i++) {
buf[i] = (int*) malloc(sizeof(int) * NY);
for (j=0; j<NY; j++) buf[i][j] = rank;
}
reqs = (int*) malloc(sizeof(int) * myNX);
sts = (int*) malloc(sizeof(int) * myNX);
/* each proc writes myNX single columns of the 2D array */
start[0] = 0; start[1] = rank;
count[0] = NY; count[1] = 1;
if (verbose)
printf("%2d: start=%3lld %3lld count=%3lld %3lld\n",
rank, start[0],start[1], count[0],count[1]);
num_reqs = 0;
for (i=0; i<myNX; i++) {
var.iputVar(start, count, &buf[0][0], &reqs[num_reqs++]);
start[1] += nprocs;
}
nc.Wait_all(num_reqs, reqs, sts);
/* check status of all requests */
for (i=0; i<num_reqs; i++)
if (sts[i] != NC_NOERR)
printf("Error: nonblocking write fails on request %d (%s)\n",
i, ncmpi_strerror(sts[i]));
free(sts);
free(reqs);
for (i=0; i<myNX; i++) free(buf[i]);
free(buf);
/* file is close implicitly */
}
catch(NcmpiException& e) {
cout << e.what() << " error code=" << e.errorCode() << " Error!\n";
return 1;
}
/* check if there is any PnetCDF internal malloc residue */
MPI_Offset malloc_size, sum_size;
int err = ncmpi_inq_malloc_size(&malloc_size);
if (err == NC_NOERR) {
MPI_Reduce(&malloc_size, &sum_size, 1, MPI_OFFSET, MPI_SUM, 0, MPI_COMM_WORLD);
if (rank == 0 && sum_size > 0)
printf("heap memory allocated by PnetCDF internally has %lld bytes yet to be freed\n",
sum_size);
}
MPI_Finalize();
return 0;
}
|
