hdf5/testpar/t_posix_compliant.c
Quincey Koziol 7be3afb278 [svn-r12440] Purpose:
Code cleanup

Description:
    Trim trailing whitespace in Makefile.am and C/C++ source files to make
diffing changes easier.

Platforms tested:
    None necessary, whitespace only change
2006-06-27 09:45:06 -05:00

898 lines
24 KiB
C

/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* Copyright by the Board of Trustees of the University of Illinois. *
* All rights reserved. *
* *
* This file is part of HDF5. The full HDF5 copyright notice, including *
* terms governing use, modification, and redistribution, is contained in *
* the files COPYING and Copyright.html. COPYING can be found at the root *
* of the source code distribution tree; Copyright.html can be found at the *
* root level of an installed copy of the electronic HDF5 document set and *
* is linked from the top-level documents page. It can also be found at *
* http://hdf.ncsa.uiuc.edu/HDF5/doc/Copyright.html. If you do not have *
* access to either file, you may request a copy from hdfhelp@ncsa.uiuc.edu. *
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
/* A series of tests for posix compliance
*
* These tests do increasingly complicated sets of writes followed by reads.
* POSIX standards say that any read that can be proven to occur after a write
* must include the data in that write. These tests attempt to verify whether the
* underlying filesystem and i/o layer provide such guarantees.
*
* There are two sets of tests, one which uses POSIX i/o (fread, fwrite) and one which
* uses MPI I/O (MPI_File_read, MPI_File_write). Each set has multiple sub-tests, which
* test varying patters of writes and reads.
*
*
* TODO:
* Add corresponding posix i/o tests for each MPI i/o test. Currently, not all of the
* MPI IO tests are implemented using fwrite/fread.
*
* Leon Arber
* larber@ncsa.uiuc.edu
*/
/* To compile this outside of the HDF5 library, you can do so by defining the
* macro STANDALONE. E.g.,
* mpicc -DSTANDALONE t_posix_compliant.c -o t_posix_compliant
* then run it as an MPI application. E.g.,
* mpirun -np 3 ./t_posix_compliant
*/
#include <unistd.h>
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <mpi.h>
#ifndef STANDALONE
#include "h5test.h"
#else
#define HDmalloc(sz) malloc(sz)
#define HDfree(p) free(p)
#define getenv_all(comm, root, name) getenv(name)
#endif
#define TESTFNAME "posix_test" /* test file name */
static char* testfile = NULL;
static int err_flag = 0;
static int max_err_print = 5;
/* globals needed for getopt
* Although they *should* be defined in unistd.h */
extern char *optarg;
extern int optind, opterr;
#define CHECK_SUCCESS(res) \
{ \
char err_got[MPI_MAX_ERROR_STRING]; \
int err_len; \
if(res != MPI_SUCCESS) \
{ \
MPI_Error_string(res, err_got, &err_len); \
fprintf(stderr, "Line %d, Error: %s\n", __LINE__, err_got); \
MPI_Abort(MPI_COMM_WORLD, -2); \
} \
}
#define PRINT_RESULT() \
{ \
int err_result; \
MPI_Reduce(&err_flag, &err_result, 1, MPI_INT, MPI_MIN, 0, MPI_COMM_WORLD); \
if( (rank == 0) && (err_result == 0) ) \
printf("PASSED\n"); \
fflush(stdout); \
fflush(stderr); \
err_flag = 0; \
}
static void vrfy_elements(int* a, int* b, int size, int rank);
static int find_writesize(int rank, int numprocs, int write_size);
/* All writes are to non-overlapping locations in the file
* Then, each task reads another tasks' data
* */
static int allwrite_allread_blocks(int numprocs, int rank, int write_size)
{
MPI_File fh = MPI_FILE_NULL;
int mpio_result;
int amode, i;
MPI_Offset offset = rank*write_size*sizeof(int);
MPI_Status Status;
int* writebuf = (int*)malloc(write_size*sizeof(int));
int* readbuf = (int*)malloc (write_size*sizeof(int));
for(i=0; i<write_size; i++)
writebuf[i] = i;
amode = MPI_MODE_CREATE | MPI_MODE_RDWR | MPI_MODE_DELETE_ON_CLOSE;
mpio_result = MPI_File_open(MPI_COMM_WORLD, testfile, amode,
MPI_INFO_NULL, &fh);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_File_write_at(fh, offset, writebuf, write_size, MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
MPI_Barrier(MPI_COMM_WORLD);
offset = ( (rank+(numprocs-1)) % numprocs)*write_size*sizeof(int);
mpio_result = MPI_File_read_at(fh, offset, readbuf, write_size, MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
vrfy_elements(writebuf, readbuf, write_size, rank);
mpio_result = MPI_File_close(&fh);
CHECK_SUCCESS(mpio_result);
HDfree(writebuf);
HDfree(readbuf);
return err_flag;
}
static int posix_allwrite_allread_blocks(int numprocs, int rank, int write_size)
{
int ret;
int i;
int offset = rank*write_size*sizeof(int);
int* writebuf = (int*)malloc(write_size*sizeof(int));
int* readbuf = (int*)malloc (write_size*sizeof(int));
FILE* file = NULL;
for(i=0; i<write_size; i++)
writebuf[i] = i;
if(rank==0)
file = fopen(testfile, "w+");
MPI_Barrier(MPI_COMM_WORLD);
if(rank != 0)
file = fopen(testfile, "r+");
if(file == NULL)
{
fprintf(stderr, "Could not create testfile\n");
MPI_Abort(MPI_COMM_WORLD, 1);
}
ret = fseek(file, offset, SEEK_SET);
if(ret == -1)
{
perror("fseek");
MPI_Abort(MPI_COMM_WORLD, 1);
}
ret = fwrite(writebuf, sizeof(int), write_size, file);
if(ret != write_size)
{
perror("fwrite");
MPI_Abort(MPI_COMM_WORLD, 1);
}
MPI_Barrier(MPI_COMM_WORLD);
offset = ( (rank+(numprocs-1)) % numprocs)*write_size*sizeof(int);
ret = fseek(file, offset, SEEK_SET);
if(ret == -1)
{
perror("fseek");
MPI_Abort(MPI_COMM_WORLD, 1);
}
ret = fread(readbuf, sizeof(int), write_size, file);
if( (ret == 0) && feof(file))
printf("Process %d: Error. Prematurely reached end of file\n", rank);
else if( (ret != write_size) && ferror(file))
{
perror("Error encountered in fread");
MPI_Abort(MPI_COMM_WORLD, 1);
}
vrfy_elements(writebuf, readbuf, write_size, rank);
fclose(file);
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
unlink(testfile);
HDfree(writebuf);
HDfree(readbuf);
return err_flag;
}
static int posix_onewrite_allread_blocks(int numprocs, int rank, int write_size)
{
int ret;
int i;
int offset = rank*write_size*sizeof(int);
int* writebuf = (int*)malloc(write_size*sizeof(int));
int* readbuf = (int*)malloc (write_size*sizeof(int));
FILE* file = NULL;
for(i=0; i<write_size; i++)
writebuf[i] = i;
if(rank==0)
file = fopen(testfile, "w+");
MPI_Barrier(MPI_COMM_WORLD);
if(rank != 0)
file = fopen(testfile, "r+");
if(file == NULL)
{
fprintf(stderr, "Could not create testfile\n");
MPI_Abort(MPI_COMM_WORLD, 1);
}
if(rank == 0)
{
for(offset = 0; offset<numprocs*write_size*sizeof(int); offset+=(write_size*sizeof(int)))
{
ret = fseek(file, offset, SEEK_SET);
if(ret == -1)
{
perror("fseek");
MPI_Abort(MPI_COMM_WORLD, 1);
}
ret = fwrite(writebuf, sizeof(int), write_size, file);
if(ret != write_size)
{
perror("fwrite");
MPI_Abort(MPI_COMM_WORLD, 1);
}
}
}
MPI_Barrier(MPI_COMM_WORLD);
offset = rank*write_size*sizeof(int);
ret = fseek(file, offset, SEEK_SET);
if(ret == -1)
{
perror("fseek");
MPI_Abort(MPI_COMM_WORLD, 1);
}
ret = fread(readbuf, sizeof(int), write_size, file);
if( (ret == 0) && feof(file))
printf("Process %d: Error. Prematurely reached end of file\n", rank);
else if( (ret != write_size) && ferror(file))
{
perror("Error encountered in fread");
MPI_Abort(MPI_COMM_WORLD, 1);
}
vrfy_elements(writebuf, readbuf, write_size, rank);
fclose(file);
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
unlink(testfile);
HDfree(writebuf);
HDfree(readbuf);
return err_flag;
}
static int posix_onewrite_allread_interlaced(int numprocs, int rank, int write_size)
{
int ret;
int i, fill, index;
int offset = rank*write_size*sizeof(int);
int* writebuf = (int*)malloc(write_size*sizeof(int));
int* readbuf = (int*)malloc (write_size*sizeof(int));
FILE* file = NULL;
if(rank==0)
file = fopen(testfile, "w+");
MPI_Barrier(MPI_COMM_WORLD);
if(rank != 0)
file = fopen(testfile, "r+");
if(file == NULL)
{
fprintf(stderr, "Could not create testfile\n");
MPI_Abort(MPI_COMM_WORLD, 1);
}
if(rank == 0)
{
for(offset = 0; offset<numprocs*write_size*sizeof(int); offset+=(numprocs*sizeof(int)))
{
ret = fseek(file, offset, SEEK_SET);
if(ret == -1)
{
perror("fseek");
MPI_Abort(MPI_COMM_WORLD, 1);
}
fill = offset / (numprocs*sizeof(int));
for(i=0; i<numprocs; i++)
writebuf[i] = fill;
ret = fwrite(writebuf, sizeof(int), numprocs, file);
if(ret != numprocs)
{
perror("fwrite");
MPI_Abort(MPI_COMM_WORLD, 1);
}
}
}
MPI_Barrier(MPI_COMM_WORLD);
index = 0;
for(offset = rank*sizeof(int); offset<numprocs*write_size*sizeof(int); offset+=(numprocs*sizeof(int)))
{
ret = fseek(file, offset, SEEK_SET);
if(ret == -1)
{
perror("fseek");
MPI_Abort(MPI_COMM_WORLD, 1);
}
ret = fread(readbuf+index, sizeof(int), 1, file);
if( (ret == 0) && feof(file))
printf("Process %d: Error. Prematurely reached end of file\n", rank);
else if( (ret != 1) && ferror(file))
{
perror("Error encountered in fread");
MPI_Abort(MPI_COMM_WORLD, 1);
}
index++;
}
for(i=0; i<write_size; i++)
writebuf[i] = i;
vrfy_elements(writebuf, readbuf, write_size, rank);
fclose(file);
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
unlink(testfile);
HDfree(writebuf);
HDfree(readbuf);
return err_flag;
}
/* Each proc wites out 0 1 2 3 with displacement i, so file contents are:
* 0000 1111 2222 3333 etc. (with 4 procs)
*
* Each proc then reads in the whole file and verifies that the data is what it is supposed to be*/
static int allwrite_allread_interlaced(int numprocs, int rank, int write_size)
{
MPI_File fh = MPI_FILE_NULL;
int mpio_result;
int amode, i, counter = 0;
MPI_Datatype filetype;
MPI_Status Status;
int* writebuf = (int*)malloc(write_size*sizeof(int));
int* readbuf = (int*) malloc(numprocs*sizeof(int));
int offset=0;
for(i=0; i<write_size; i++)
writebuf[i] = i;
amode = MPI_MODE_CREATE | MPI_MODE_RDWR | MPI_MODE_DELETE_ON_CLOSE;
mpio_result = MPI_File_open(MPI_COMM_WORLD, testfile, amode, MPI_INFO_NULL, &fh);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_Type_vector(write_size, 1, numprocs, MPI_INT, &filetype);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_Type_commit(&filetype);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_File_set_view(fh, rank*sizeof(int), MPI_INT, filetype, "native", MPI_INFO_NULL);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_File_write(fh, writebuf, write_size, MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
MPI_Barrier(MPI_COMM_WORLD);
mpio_result = MPI_File_set_view(fh, 0, MPI_BYTE, MPI_BYTE, "native", MPI_INFO_NULL);
CHECK_SUCCESS(mpio_result);
for(offset = 0; offset<(write_size*numprocs*sizeof(int)); offset+=(numprocs*sizeof(int)))
{
mpio_result = MPI_File_read_at(fh, offset, readbuf, numprocs, MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
for(i=0; i<numprocs; i++)
{
if(writebuf[offset/(numprocs*sizeof(int))] != readbuf[i])
{
if( (rank == 0) && (counter == 0))
printf("\n");
if(counter++ < max_err_print)
fprintf(stderr, "Arrays do not match! Prcoess %d, element %d: [%d, %d]\n", rank, i, writebuf[offset/(numprocs*sizeof(int))], readbuf[i]);
else if(counter++ == max_err_print+1)
fprintf(stderr, "Printed %d errors. Omitting the rest\n", max_err_print);
err_flag = -1;
}
}
}
mpio_result = MPI_File_close(&fh);
CHECK_SUCCESS(mpio_result);
HDfree(writebuf);
HDfree(readbuf);
return err_flag;
}
/* Overlapping pattern works as follows (this test requires at least 2 procs:
* Writes:
* Task 0: 0 2 4 6 etc...
* Task 1: 1 3 5 7 etc...
* Task 2: 0 3 6 etc..
* Task 3: 0 4 8 etc...
*
* The above describes only the pattern of the elements being written. The actual
* number of elements written is going to be:
*
* Task i where i=(numprocs-1) writes write_size elements. All other tasks do:
* x = ((write_size-1)*numprocs)
* Task i's write_size is the smallest multiple of i<=x divided by i, with the exception
* of tasks 0 and 1, for whom i is 2, since they are writing the even and odd multiples.
*
* So, if there are 5 tasks with write_size=4, the resulting pattern of writes is:
*
* Task 0: 0 2 4 6 8 10 12 14
* Task 1: 1 3 5 7 9 11 13 15
* Task 2: 0 3 6 9 12 15
* Task 3: 0 4 8 12
* Task 4: 0 5 10 15
*
*
*
* * * All the entires that overlap will therefore be writing the same value
*
* At the end, all tasks read in the file and verify that it is correct should be
* (1,2...((numprocs-1)*WRTE_SIZE).
* */
static int allwrite_allread_overlap(int numprocs, int rank, int write_size)
{
MPI_File fh = MPI_FILE_NULL;
int mpio_result;
int amode, i, counter = 0;
MPI_Datatype filetype;
MPI_Status Status;
int* writebuf = (int*) malloc(write_size*(numprocs-1)*sizeof(int)); /* An upper bound...not all the elements will be written */
int* readbuf = (int*) malloc(write_size*(numprocs-1)*sizeof(int));
if(numprocs < 2)
{
fprintf(stderr, "The allwrite_allread_overlap test requires at least 2 procs\n");
return -1;
}
if(rank == 0)
{
for(i=0; i<write_size*(numprocs-1); i++)
writebuf[i] = 2*i;
}
else if(rank == 1)
{
for(i=0; i<write_size*(numprocs-1); i++)
writebuf[i] = (2*i)+1;
}
else
{
for(i=0; i<write_size*(numprocs-1); i++)
writebuf[i] = (rank+1)*i;
}
amode = MPI_MODE_CREATE | MPI_MODE_RDWR | MPI_MODE_DELETE_ON_CLOSE;
mpio_result = MPI_File_open(MPI_COMM_WORLD, testfile, amode, MPI_INFO_NULL, &fh);
CHECK_SUCCESS(mpio_result);
if( (rank == 0) || (rank == 1) )
mpio_result = MPI_Type_vector(write_size*(numprocs-1), 1, 2, MPI_INT, &filetype);
else
mpio_result = MPI_Type_vector(write_size*(numprocs-1), 1, rank+1, MPI_INT, &filetype);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_Type_commit(&filetype);
CHECK_SUCCESS(mpio_result);
if( rank == 1)
mpio_result = MPI_File_set_view(fh, sizeof(int), MPI_INT, filetype, "native", MPI_INFO_NULL);
else
mpio_result = MPI_File_set_view(fh, 0, MPI_INT, filetype, "native", MPI_INFO_NULL);
CHECK_SUCCESS(mpio_result);
if( rank == (numprocs - 1))
mpio_result = MPI_File_write(fh, writebuf, write_size, MPI_INT, &Status);
else
mpio_result = MPI_File_write(fh, writebuf, find_writesize(rank, numprocs, write_size), MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
MPI_Barrier(MPI_COMM_WORLD);
mpio_result = MPI_File_set_view(fh, 0, MPI_BYTE, MPI_BYTE, "native", MPI_INFO_NULL);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_File_read_at(fh, 0, readbuf, write_size*(numprocs-1), MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
for(i=0; i<write_size*(numprocs-1); i++)
{
if(i != readbuf[i])
{
if( (rank == 0) && (counter == 0))
printf("\n");
if(counter++ < max_err_print)
fprintf(stderr, "Arrays do not match! Prcoess %d, element %d: [%d, %d]\n", rank, i, i, readbuf[i]);
else if(counter++ == max_err_print+1)
fprintf(stderr, "Printed %d errors. Omitting the rest\n", max_err_print);
err_flag = -1;
}
}
mpio_result = MPI_File_close(&fh);
CHECK_SUCCESS(mpio_result);
HDfree(writebuf);
HDfree(readbuf);
return err_flag;
}
/* A random process writes out the following to the file:
* 0 1 2 3 4 0 1 2 3 4 0 1 2 3 4 (assuming write_size=5, and numprocs=3)
*
* Process i read's in write_size bytes at offset=i*write_size
*/
static int onewrite_allread_blocks(int numprocs, int rank, int write_size)
{
MPI_File fh = MPI_FILE_NULL;
int mpio_result;
int amode, i;
MPI_Status Status;
int* writebuf = (int*)malloc(write_size*sizeof(int));
int* readbuf = (int*)malloc (write_size*sizeof(int));
for(i=0; i<write_size; i++)
writebuf[i] = i;
amode = MPI_MODE_CREATE | MPI_MODE_RDWR | MPI_MODE_DELETE_ON_CLOSE;
mpio_result = MPI_File_open(MPI_COMM_WORLD, testfile, amode,
MPI_INFO_NULL, &fh);
CHECK_SUCCESS(mpio_result);
/* A random process writes out all the data */
if(rank == (rand() % numprocs))
{
for(i=0; i<numprocs; i++)
{
mpio_result = MPI_File_write_at(fh, write_size*i*sizeof(int), writebuf, write_size, MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
}
}
MPI_Barrier(MPI_COMM_WORLD);
mpio_result = MPI_File_read_at(fh, write_size*rank*sizeof(int), readbuf, write_size, MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
vrfy_elements(writebuf, readbuf, write_size, rank);
mpio_result = MPI_File_close(&fh);
CHECK_SUCCESS(mpio_result);
HDfree(writebuf);
HDfree(readbuf);
return err_flag;
}
/* Process zero writes out:
* 0000 1111 2222 3333 etc. (with 4 procs)
*
* Each proc reads out 0 1 2 3 starting at displacement i */
static int onewrite_allread_interlaced(int numprocs, int rank, int write_size)
{
MPI_File fh = MPI_FILE_NULL;
int mpio_result;
int amode, i;
MPI_Datatype filetype;
MPI_Status Status;
int* writebuf = (int*) malloc(numprocs*write_size*sizeof(int)); /* Upper bound, not all used */
int* readbuf = (int*)malloc (write_size*sizeof(int));
amode = MPI_MODE_CREATE | MPI_MODE_RDWR;
amode = MPI_MODE_CREATE | MPI_MODE_RDWR | MPI_MODE_DELETE_ON_CLOSE;
mpio_result = MPI_File_open(MPI_COMM_WORLD, testfile, amode, MPI_INFO_NULL, &fh);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_Type_vector(write_size, 1, numprocs, MPI_INT, &filetype);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_Type_commit(&filetype);
CHECK_SUCCESS(mpio_result);
if(rank == (rand() % numprocs))
{
for(i=0; i<write_size; i++)
{
int j;
for(j=0; j<numprocs; j++)
writebuf[j] = i;
mpio_result = MPI_File_write_at(fh, i*numprocs*sizeof(int), writebuf, numprocs, MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
}
}
MPI_Barrier(MPI_COMM_WORLD);
mpio_result = MPI_File_set_view(fh, rank*sizeof(int), MPI_INT, filetype, "native", MPI_INFO_NULL);
CHECK_SUCCESS(mpio_result);
mpio_result = MPI_File_read_at(fh, 0, readbuf, write_size, MPI_INT, &Status);
CHECK_SUCCESS(mpio_result);
for(i=0; i<write_size; i++)
writebuf[i] = i;
vrfy_elements(writebuf, readbuf, write_size, rank);
mpio_result = MPI_File_close(&fh);
CHECK_SUCCESS(mpio_result);
HDfree(writebuf);
HDfree(readbuf);
return err_flag;
}
int main(int argc, char* argv[])
{
int numprocs, rank, opt, mpi_tests=1, posix_tests=1;
int lb, ub, inc;
int write_size = 0;
char optstring[] = "h x m p: s: v:";
char *prefix;
err_flag = 0;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
while((opt = getopt(argc, argv, optstring)) != -1)
{
switch(opt)
{
case 'h':
if(rank == 0)
printf("Usage: %s [options]\n"
"-h prints this help message\n"
"-x run the posix i/o tests ONLY (default: posix and MPI)\n"
"-m run the mpi i/o tests ONLY (default: posix and MPI)\n"
"-s size Run the test for the specific size. Default is 1024, 4096, 16384, ..., 1048576\n"
"-p path specifies path for test file. Default is current directory\n"
"-v num Specifies number of unmatching entries to print (default 10, pass -1 for all)\n", argv[0]);
goto done;
case 'x':
mpi_tests = 0;
posix_tests = 1;
break;
case 'm':
mpi_tests = 1;
posix_tests = 0;
break;
case 'p':
/* need 2 extra--1 for the / and 1 for the terminating NULL. */
testfile = (char*) HDmalloc(strlen(optarg) + 2 + strlen(TESTFNAME));
strcpy(testfile, optarg);
/* Append a / just in case they didn't end their path with one */
strcat(testfile, "/" TESTFNAME);
break;
case 's':
write_size = atoi(optarg);
break;
case 'v':
max_err_print = atoi(optarg);
break;
}
}
if( (optind < argc) && (rank == 0))
fprintf(stderr, "Unkown command-line argument passed. Continuing anyway...\n");
if (!testfile){
/* Try environment variable if not given as option. */
prefix = getenv_all(MPI_COMM_WORLD, 0, "HDF5_PARAPREFIX");
if (prefix)
{
/* need 2 extra--1 for the / and 1 for the terminating NULL. */
testfile = (char*) HDmalloc(strlen(prefix) + 2 + strlen(TESTFNAME));
strcpy(testfile, prefix);
/* Append a / just in case they didn't end their path with one */
strcat(testfile, "/" TESTFNAME);
}
else
{
testfile = strdup(TESTFNAME);
}
}
printf("Proc %d: testfile=%s\n", rank, testfile);
if(write_size == 0)
{
lb = 1024;
/* 1MB MPIO-IO overlapping is failing in copper. Lower it now pending
permenant fix for copper.*/
/* ub = 1024*1024;*/
ub = 1024*512;
inc = 4;
}
else
{
lb = write_size;
ub = write_size+1;
inc = 2;
}
#ifndef STANDALONE
/* set alarm. */
ALARM_ON;
#endif
for(write_size = lb; write_size <= ub; write_size*=inc)
{
if(rank == 0)
printf("\nTesting size %d\n", write_size);
if(mpi_tests)
{
if(rank == 0)
printf("Testing allwrite_allread_blocks with MPI IO\t\t"); fflush(stdout);
allwrite_allread_blocks(numprocs, rank, write_size/(numprocs*sizeof(int)));
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
printf("Testing allwrite_allread_interlaced with MPI IO\t\t"); fflush(stdout);
allwrite_allread_interlaced(numprocs, rank, write_size/(numprocs*sizeof(int)));
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
printf("Testing allwrite_allread_overlap with MPI IO\t\t"); fflush(stdout);
allwrite_allread_overlap(numprocs, rank, write_size);
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
printf("Testing onewrite_allread_blocks with MPI IO\t\t"); fflush(stdout);
onewrite_allread_blocks(numprocs, rank, write_size/(numprocs*sizeof(int)));
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
printf("Testing onewrite_allread_interlaced with MPI IO\t\t"); fflush(stdout);
onewrite_allread_interlaced(numprocs, rank, write_size/(numprocs*sizeof(int)));
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
}
if(posix_tests)
{
if(rank == 0)
printf("Testing allwrite_allread_blocks with POSIX IO\t\t"); fflush(stdout);
posix_allwrite_allread_blocks(numprocs, rank, write_size/(numprocs*sizeof(int)));
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
printf("Testing onewrite_allread_blocks with POSIX IO\t\t"); fflush(stdout);
posix_onewrite_allread_blocks(numprocs, rank, write_size/(numprocs*sizeof(int)));
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
if(rank == 0)
printf("Testing onewrite_allread_interlaced with POSIX IO\t"); fflush(stdout);
posix_onewrite_allread_interlaced(numprocs, rank, write_size/(numprocs*sizeof(int)));
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
/* if(rank == 0)
printf("Testing allwrite_allread_overlap with POSIX IO\t\t"); fflush(stdout);
posix_allwrite_allread_overlap(numprocs, rank, write_size);
PRINT_RESULT();
MPI_Barrier(MPI_COMM_WORLD);
*/
}
}
#ifndef STANDALONE
/* turn off alarm */
ALARM_OFF;
#endif
done:
if (testfile)
HDfree(testfile);
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
return 0;
}
static int find_writesize(int rank, int numprocs, int size)
{
/* Largest number in the file */
int tmp = (size-1)*numprocs;
int x = 0;
int write_size = 0;
/* Find largest multiple not greater than tmp */
while(x <= tmp)
{
if( (rank == 0) || (rank == 1) )
x+=2;
else
x += (rank+1);
write_size++;
}
return write_size;
}
static void vrfy_elements(int* a, int* b, int size, int rank)
{
int i, counter = 0;
for(i=0; i<size; i++)
{
if(a[i] != b[i])
{
if( (rank == 0) && (counter == 0))
printf("\n");
if(counter++ < max_err_print)
fprintf(stderr, "Arrays do not match! Prcoess %d, element %d: [%d, %d]\n", rank, i, a[i], b[i]);
else if(counter++ == max_err_print+1)
fprintf(stderr, "Printed %d errors. Omitting the rest\n", max_err_print);
err_flag = -1;
}
}
fflush(stderr);
fflush(stdout);
}