mirror/hdf5
2
0
Fork 0
mirror of https://github.com/HDFGroup/hdf5.git synced 2024-12-09 07:32:32 +08:00
Code Issues Packages Projects Releases Wiki Activity
15691f8827
hdf5/testpar/t_mdset.c
John Mainzer c100b0bf26 [svn-r11470] Purpose: 
Repair synchronization bug in the metadata cache in PHDF5

Also repair numerous other bugs that surfaced in testing the
bug fix.


Description:

While operations modifying metadata must be collective, we allow
independant reads.  This allows metadata caches on different processes
to adjust to different sizes, and to place the entries on their dirty
lists in different orders.  Since only process 0 actually writes
metadata to disk (all other processes thought they did, but the writes
were discarded on the theory that they had to be collective), this made
it possible for another process to modify metadata, flush it, and then
read it back in in its original form (pre-modification) form.  The
possibilities for file corruption should be obvious.


Solution:

Make the policy that only process 0 can write to file explicit, and
visible to the metadata caches.  Thus only process 0 may flush dirty
entries -- all other caches must retain dirty entries until they are
informed by process 0 that the entries are clean.

Synchronization is handled by counting the bytes of dirty cache entries
created, and then synching up between the caches whenever the sum
exceeds an (eventually user specified) limit.  Dirty metadata creation
is consistent across all processes because all operations modifying
metadata must be collective.

This change uncovered may bugs which are repaired in this checkin.
It also required modification of H5HL and H5O to allocate file space
on insertion rather than on flush from cache.


Platforms tested:

H5committest, heping(parallel & serial)

Misc. update:
2005-09-27 00:20:11 -05:00

1692 lines
53 KiB
C
Raw Blame History

/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* 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. *
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
#include "testphdf5.h"
#define DIM 2
#define SIZE 32
#define NDATASET 4
#define GROUP_DEPTH 128
enum obj_type { is_group, is_dset };
int get_size(void);
void write_dataset(hid_t, hid_t, hid_t);
int read_dataset(hid_t, hid_t, hid_t);
void create_group_recursive(hid_t, hid_t, hid_t, int);
void recursive_read_group(hid_t, hid_t, hid_t, int);
void group_dataset_read(hid_t fid, int mpi_rank, int m);
void write_attribute(hid_t, int, int);
int read_attribute(hid_t, int, int);
int check_value(DATATYPE *, DATATYPE *, int);
void get_slab(hsize_t[], hsize_t[], hsize_t[], hsize_t[], int);
/*
* The size value computed by this function is used extensively in
* configuring tests for the current number of processes.
*
* This function was created as part of an effort to allow the
* test functions in this file to run on an arbitrary number of
* processors.
* JRM - 8/11/04
*/
int get_size(void)
{
int mpi_rank;
int mpi_size;
int size = SIZE;
MPI_Comm_rank (MPI_COMM_WORLD, &mpi_rank); /* needed for VRFY */
MPI_Comm_size (MPI_COMM_WORLD, &mpi_size);
if ( mpi_size > size ) {
if ( (mpi_size % 2) == 0 ) {
size = mpi_size;
} else {
size = mpi_size + 1;
}
}
VRFY((mpi_size <= size), "mpi_size <= size");
VRFY(((size % 2) == 0), "size isn't even");
return(size);
} /* get_size() */
/*
* Example of using PHDF5 to create ndatasets datasets. Each process write
* a slab of array to the file.
*
* Changes: Updated function to use a dynamically calculated size,
* instead of the old SIZE #define. This should allow it
* to function with an arbitrary number of processors.
*
* JRM - 8/11/04
*/
void multiple_dset_write(void)
{
int i, j, n, mpi_size, mpi_rank, size;
hid_t iof, plist, dataset, memspace, filespace;
hid_t dcpl; /* Dataset creation property list */
hbool_t use_gpfs = FALSE; /* Use GPFS hints */
hsize_t chunk_origin [DIM];
hsize_t chunk_dims [DIM], file_dims [DIM];
hsize_t count[DIM]={1,1};
double * outme = NULL;
double fill=1.0; /* Fill value */
char dname [100];
herr_t ret;
const H5Ptest_param_t *pt;
char *filename;
int ndatasets;
pt = GetTestParameters();
filename = pt->name;
ndatasets = pt->count;
size = get_size();
MPI_Comm_rank (MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size (MPI_COMM_WORLD, &mpi_size);
outme = HDmalloc((size_t)(size * size * sizeof(double)));
VRFY((outme != NULL), "HDmalloc succeeded for outme");
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
VRFY((plist>=0), "create_faccess_plist succeeded");
iof = H5Fcreate (filename, H5F_ACC_TRUNC, H5P_DEFAULT, plist);
VRFY((iof>=0), "H5Fcreate succeeded");
ret = H5Pclose (plist);
VRFY((ret>=0), "H5Pclose succeeded");
/* decide the hyperslab according to process number. */
get_slab(chunk_origin, chunk_dims, count, file_dims, size);
memspace = H5Screate_simple (DIM, chunk_dims, NULL);
filespace = H5Screate_simple (DIM, file_dims, NULL);
ret = H5Sselect_hyperslab (filespace, H5S_SELECT_SET, chunk_origin, chunk_dims, count, chunk_dims);
VRFY((ret>=0), "mdata hyperslab selection");
/* Create a dataset creation property list */
dcpl = H5Pcreate(H5P_DATASET_CREATE);
VRFY((dcpl>=0), "dataset creation property list succeeded");
ret=H5Pset_fill_value(dcpl, H5T_NATIVE_DOUBLE, &fill);
VRFY((ret>=0), "set fill-value succeeded");
for (n = 0; n < ndatasets; n++) {
sprintf (dname, "dataset %d", n);
dataset = H5Dcreate (iof, dname, H5T_NATIVE_DOUBLE, filespace, dcpl);
VRFY((dataset > 0), dname);
/* calculate data to write */
for (i = 0; i < size; i++)
for (j = 0; j < size; j++)
outme [(i * size) + j] = n*1000 + mpi_rank;
H5Dwrite (dataset, H5T_NATIVE_DOUBLE, memspace, filespace, H5P_DEFAULT, outme);
H5Dclose (dataset);
#ifdef BARRIER_CHECKS
if (! ((n+1) % 10)) {
printf("created %d datasets\n", n+1);
MPI_Barrier(MPI_COMM_WORLD);
}
#endif /* BARRIER_CHECKS */
}
H5Sclose (filespace);
H5Sclose (memspace);
H5Pclose (dcpl);
H5Fclose (iof);
HDfree(outme);
}
/* Example of using PHDF5 to create, write, and read compact dataset.
*
* Changes: Updated function to use a dynamically calculated size,
* instead of the old SIZE #define. This should allow it
* to function with an arbitrary number of processors.
*
* JRM - 8/11/04
*/
void compact_dataset(void)
{
int i, j, mpi_size, mpi_rank, size, err_num=0;
hbool_t use_gpfs = FALSE;
hid_t iof, plist, dcpl, dxpl, dataset, filespace;
hsize_t file_dims [DIM];
double * outme;
double * inme;
char dname[]="dataset";
herr_t ret;
const char *filename;
size = get_size();
for ( i = 0; i < DIM; i++ )
{
file_dims[i] = size;
}
MPI_Comm_rank (MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size (MPI_COMM_WORLD, &mpi_size);
outme = HDmalloc((size_t)(size * size * sizeof(double)));
VRFY((outme != NULL), "HDmalloc succeeded for outme");
inme = HDmalloc((size_t)(size * size * sizeof(double)));
VRFY((outme != NULL), "HDmalloc succeeded for inme");
filename = GetTestParameters();
VRFY((mpi_size <= size), "mpi_size <= size");
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
iof = H5Fcreate (filename, H5F_ACC_TRUNC, H5P_DEFAULT, plist);
/* Define data space */
filespace = H5Screate_simple (DIM, file_dims, NULL);
/* Create a compact dataset */
dcpl = H5Pcreate(H5P_DATASET_CREATE);
VRFY((dcpl>=0), "dataset creation property list succeeded");
ret=H5Pset_layout(dcpl, H5D_COMPACT);
VRFY((dcpl >= 0), "set property list for compact dataset");
ret=H5Pset_alloc_time(dcpl, H5D_ALLOC_TIME_EARLY);
VRFY((ret >= 0), "set space allocation time for compact dataset");
dataset = H5Dcreate (iof, dname, H5T_NATIVE_DOUBLE, filespace, dcpl);
VRFY((dataset >= 0), "H5Dcreate succeeded");
/* set up the collective transfer properties list */
dxpl = H5Pcreate (H5P_DATASET_XFER);
VRFY((dxpl >= 0), "");
ret=H5Pset_dxpl_mpio(dxpl, H5FD_MPIO_COLLECTIVE);
VRFY((ret >= 0), "H5Pcreate xfer succeeded");
/* Recalculate data to write. Each process writes the same data. */
for (i = 0; i < size; i++)
for (j = 0; j < size; j++)
outme[(i * size) + j] = (i+j)*1000;
ret=H5Dwrite (dataset, H5T_NATIVE_DOUBLE, H5S_ALL, H5S_ALL, dxpl, outme);
VRFY((ret >= 0), "H5Dwrite succeeded");
H5Pclose (dcpl);
H5Pclose (plist);
H5Dclose (dataset);
H5Sclose (filespace);
H5Fclose (iof);
/* Open the file and dataset, read and compare the data. */
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
iof = H5Fopen(filename, H5F_ACC_RDONLY, plist);
VRFY((iof >= 0), "H5Fopen succeeded");
/* set up the collective transfer properties list */
dxpl = H5Pcreate (H5P_DATASET_XFER);
VRFY((dxpl >= 0), "");
ret=H5Pset_dxpl_mpio(dxpl, H5FD_MPIO_COLLECTIVE);
VRFY((ret >= 0), "H5Pcreate xfer succeeded");
dataset = H5Dopen(iof, dname);
VRFY((dataset >= 0), "H5Dcreate succeeded");
ret = H5Dread(dataset, H5T_NATIVE_DOUBLE, H5S_ALL, H5S_ALL, dxpl, inme);
VRFY((ret >= 0), "H5Dread succeeded");
/* Verify data value */
for (i = 0; i < size; i++)
for (j = 0; j < size; j++)
if ( inme[(i * size) + j] != outme[(i * size) + j])
if(err_num++ < MAX_ERR_REPORT || VERBOSE_MED)
printf("Dataset Verify failed at [%d][%d]: expect %f, got %f\n", i, j, outme[(i * size) + j], inme[(i * size) + j]);
H5Pclose(plist);
H5Pclose(dxpl);
H5Dclose(dataset);
H5Fclose(iof);
HDfree(inme);
HDfree(outme);
}
/*
* Example of using PHDF5 to create, write, and read dataset and attribute
* of Null dataspace.
*
* Changes: Removed the assert that mpi_size <= the SIZE #define.
* As best I can tell, this assert isn't needed here,
* and in any case, the SIZE #define is being removed
* in an update of the functions in this file to run
* with an arbitrary number of processes.
*
* JRM - 8/24/04
*/
void null_dataset(void)
{
int mpi_size, mpi_rank;
hbool_t use_gpfs = FALSE;
hid_t iof, plist, dxpl, dataset, attr, sid;
unsigned uval=2; /* Buffer for writing to dataset */
int val=1; /* Buffer for writing to attribute */
int nelem;
char dname[]="dataset";
char attr_name[]="attribute";
herr_t ret;
const char *filename;
MPI_Comm_rank (MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size (MPI_COMM_WORLD, &mpi_size);
filename = GetTestParameters();
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL,
facc_type, use_gpfs);
iof = H5Fcreate (filename, H5F_ACC_TRUNC, H5P_DEFAULT, plist);
/* Define data space */
sid = H5Screate(H5S_NULL);
/* Check that the null dataspace actually has 0 elements */
nelem = H5Sget_simple_extent_npoints(sid);
VRFY((nelem== 0), "H5Sget_simple_extent_npoints");
/* Create a compact dataset */
dataset = H5Dcreate (iof, dname, H5T_NATIVE_UINT, sid, H5P_DEFAULT);
VRFY((dataset >= 0), "H5Dcreate succeeded");
/* set up the collective transfer properties list */
dxpl = H5Pcreate (H5P_DATASET_XFER);
VRFY((dxpl >= 0), "");
ret=H5Pset_dxpl_mpio(dxpl, H5FD_MPIO_COLLECTIVE);
VRFY((ret >= 0), "H5Pcreate xfer succeeded");
/* Write "nothing" to the dataset (with type conversion) */
ret=H5Dwrite (dataset, H5T_NATIVE_INT, H5S_ALL, H5S_ALL, dxpl, &uval);
VRFY((ret >= 0), "H5Dwrite succeeded");
/* Create an attribute for the group */
attr=H5Acreate(dataset,attr_name,H5T_NATIVE_UINT,sid,H5P_DEFAULT);
VRFY((attr>=0), "H5Acreate");
/* Write "nothing" to the attribute (with type conversion) */
ret = H5Awrite(attr, H5T_NATIVE_INT, &val);
VRFY((ret>=0), "H5Awrite");
H5Aclose (attr);
H5Dclose (dataset);
H5Pclose (plist);
H5Sclose (sid);
H5Fclose (iof);
/* Open the file and dataset, read and compare the data. */
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
iof = H5Fopen(filename, H5F_ACC_RDONLY, plist);
VRFY((iof >= 0), "H5Fopen succeeded");
/* set up the collective transfer properties list */
dxpl = H5Pcreate (H5P_DATASET_XFER);
VRFY((dxpl >= 0), "");
ret=H5Pset_dxpl_mpio(dxpl, H5FD_MPIO_COLLECTIVE);
VRFY((ret >= 0), "H5Pcreate xfer succeeded");
dataset = H5Dopen(iof, dname);
VRFY((dataset >= 0), "H5Dcreate succeeded");
/* Try reading from the dataset (make certain our buffer is unmodified) */
ret = H5Dread(dataset, H5T_NATIVE_UINT, H5S_ALL, H5S_ALL, dxpl, &uval);
VRFY((ret>=0), "H5Dread");
VRFY((uval==2), "H5Dread");
/* Open the attribute for the dataset */
attr=H5Aopen_name(dataset,attr_name);
VRFY((attr>=0), "H5Aopen_name");
/* Try reading from the attribute (make certain our buffer is unmodified) */ ret = H5Aread(attr, H5T_NATIVE_INT, &val);
VRFY((ret>=0), "H5Aread");
VRFY((val==1), "H5Aread");
H5Pclose(plist);
H5Pclose(dxpl);
H5Aclose (attr);
H5Dclose(dataset);
H5Fclose(iof);
}
/* Example of using PHDF5 to create "large" datasets. (>2GB, >4GB, >8GB)
* Actual data is _not_ written to these datasets. Dataspaces are exact
* sizes (2GB, 4GB, etc.), but the metadata for the file pushes the file over
* the boundary of interest.
*
* Changes: Removed the assert that mpi_size <= the SIZE #define.
* As best I can tell, this assert isn't needed here,
* and in any case, the SIZE #define is being removed
* in an update of the functions in this file to run
* with an arbitrary number of processes.
*
* JRM - 8/11/04
*/
void big_dataset(void)
{
int mpi_size, mpi_rank; /* MPI info */
hbool_t use_gpfs = FALSE; /* Don't use GPFS stuff for this test */
hid_t iof, /* File ID */
fapl, /* File access property list ID */
dataset, /* Dataset ID */
filespace; /* Dataset's dataspace ID */
hsize_t file_dims [4]; /* Dimensions of dataspace */
char dname[]="dataset"; /* Name of dataset */
MPI_Offset file_size; /* Size of file on disk */
herr_t ret; /* Generic return value */
const char *filename;
MPI_Comm_rank (MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size (MPI_COMM_WORLD, &mpi_size);
/* Verify MPI_Offset can handle larger than 2GB sizes */
VRFY((sizeof(MPI_Offset)>4), "sizeof(MPI_Offset)>4");
filename = GetTestParameters();
fapl = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
VRFY((fapl >= 0), "create_faccess_plist succeeded");
/*
* Create >2GB HDF5 file
*/
iof = H5Fcreate (filename, H5F_ACC_TRUNC, H5P_DEFAULT, fapl);
VRFY((iof >= 0), "H5Fcreate succeeded");
/* Define dataspace for 2GB dataspace */
file_dims[0]= 2;
file_dims[1]= 1024;
file_dims[2]= 1024;
file_dims[3]= 1024;
filespace = H5Screate_simple (4, file_dims, NULL);
VRFY((filespace >= 0), "H5Screate_simple succeeded");
dataset = H5Dcreate (iof, dname, H5T_NATIVE_UCHAR, filespace, H5P_DEFAULT);
VRFY((dataset >= 0), "H5Dcreate succeeded");
/* Close all file objects */
ret=H5Dclose (dataset);
VRFY((ret >= 0), "H5Dclose succeeded");
ret=H5Sclose (filespace);
VRFY((ret >= 0), "H5Sclose succeeded");
ret=H5Fclose (iof);
VRFY((ret >= 0), "H5Fclose succeeded");
/* Check that file of the correct size was created */
file_size=h5_mpi_get_file_size(filename, MPI_COMM_WORLD, MPI_INFO_NULL);
#ifndef WIN32
VRFY((file_size == 2147485696ULL), "File is correct size (~2GB)");
#endif
/*
* Create >4GB HDF5 file
*/
iof = H5Fcreate (filename, H5F_ACC_TRUNC, H5P_DEFAULT, fapl);
VRFY((iof >= 0), "H5Fcreate succeeded");
/* Define dataspace for 4GB dataspace */
file_dims[0]= 4;
file_dims[1]= 1024;
file_dims[2]= 1024;
file_dims[3]= 1024;
filespace = H5Screate_simple (4, file_dims, NULL);
VRFY((filespace >= 0), "H5Screate_simple succeeded");
dataset = H5Dcreate (iof, dname, H5T_NATIVE_UCHAR, filespace, H5P_DEFAULT);
VRFY((dataset >= 0), "H5Dcreate succeeded");
/* Close all file objects */
ret=H5Dclose (dataset);
VRFY((ret >= 0), "H5Dclose succeeded");
ret=H5Sclose (filespace);
VRFY((ret >= 0), "H5Sclose succeeded");
ret=H5Fclose (iof);
VRFY((ret >= 0), "H5Fclose succeeded");
/* Check that file of the correct size was created */
file_size=h5_mpi_get_file_size(filename, MPI_COMM_WORLD, MPI_INFO_NULL);
#ifndef WIN32
VRFY((file_size == 4294969344ULL), "File is correct size (~4GB)");
#endif
/*
* Create >8GB HDF5 file
*/
iof = H5Fcreate (filename, H5F_ACC_TRUNC, H5P_DEFAULT, fapl);
VRFY((iof >= 0), "H5Fcreate succeeded");
/* Define dataspace for 8GB dataspace */
file_dims[0]= 8;
file_dims[1]= 1024;
file_dims[2]= 1024;
file_dims[3]= 1024;
filespace = H5Screate_simple (4, file_dims, NULL);
VRFY((filespace >= 0), "H5Screate_simple succeeded");
dataset = H5Dcreate (iof, dname, H5T_NATIVE_UCHAR, filespace, H5P_DEFAULT);
VRFY((dataset >= 0), "H5Dcreate succeeded");
/* Close all file objects */
ret=H5Dclose (dataset);
VRFY((ret >= 0), "H5Dclose succeeded");
ret=H5Sclose (filespace);
VRFY((ret >= 0), "H5Sclose succeeded");
ret=H5Fclose (iof);
VRFY((ret >= 0), "H5Fclose succeeded");
/* Check that file of the correct size was created */
file_size=h5_mpi_get_file_size(filename, MPI_COMM_WORLD, MPI_INFO_NULL);
#ifndef WIN32
VRFY((file_size == 8589936640ULL), "File is correct size (~8GB)");
#endif
/* Close fapl */
ret=H5Pclose (fapl);
VRFY((ret >= 0), "H5Pclose succeeded");
}
/* Example of using PHDF5 to read a partial written dataset. The dataset does
* not have actual data written to the entire raw data area and relies on the
* default fill value of zeros to work correctly.
*
* Changes: Removed the assert that mpi_size <= the SIZE #define.
* As best I can tell, this assert isn't needed here,
* and in any case, the SIZE #define is being removed
* in an update of the functions in this file to run
* with an arbitrary number of processes.
*
* Also added code to free dynamically allocated buffers.
*
* JRM - 8/11/04
*/
void dataset_fillvalue(void)
{
int mpi_size, mpi_rank; /* MPI info */
hbool_t use_gpfs = FALSE; /* Don't use GPFS stuff for this test */
int err_num; /* Number of errors */
hid_t iof, /* File ID */
fapl, /* File access property list ID */
dxpl, /* Data transfer property list ID */
dataset, /* Dataset ID */
memspace, /* Memory dataspace ID */
filespace; /* Dataset's dataspace ID */
char dname[]="dataset"; /* Name of dataset */
hsize_t dset_dims[4] = {0, 6, 7, 8};
hsize_t req_start[4] = {0, 0, 0, 0};
hsize_t req_count[4] = {1, 6, 7, 8};
hsize_t dset_size; /* Dataset size */
int *rdata, *wdata; /* Buffers for data to read and write */
int *twdata, *trdata; /* Temporary pointer into buffer */
int acc, i, j, k, l; /* Local index variables */
herr_t ret; /* Generic return value */
const char *filename;
MPI_Comm_rank (MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size (MPI_COMM_WORLD, &mpi_size);
filename = GetTestParameters();
/* Set the dataset dimension to be one row more than number of processes */
/* and calculate the actual dataset size. */
dset_dims[0]=mpi_size+1;
dset_size=dset_dims[0]*dset_dims[1]*dset_dims[2]*dset_dims[3];
/* Allocate space for the buffers */
rdata=HDmalloc((size_t)(dset_size*sizeof(int)));
VRFY((rdata != NULL), "HDcalloc succeeded for read buffer");
wdata=HDmalloc((size_t)(dset_size*sizeof(int)));
VRFY((wdata != NULL), "HDmalloc succeeded for write buffer");
fapl = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
VRFY((fapl >= 0), "create_faccess_plist succeeded");
/*
* Create HDF5 file
*/
iof = H5Fcreate (filename, H5F_ACC_TRUNC, H5P_DEFAULT, fapl);
VRFY((iof >= 0), "H5Fcreate succeeded");
filespace = H5Screate_simple(4, dset_dims, NULL);
VRFY((filespace >= 0), "File H5Screate_simple succeeded");
dataset = H5Dcreate(iof, dname, H5T_NATIVE_INT, filespace, H5P_DEFAULT);
VRFY((dataset >= 0), "H5Dcreate succeeded");
memspace = H5Screate_simple(4, dset_dims, NULL);
VRFY((memspace >= 0), "Memory H5Screate_simple succeeded");
/*
* Read dataset before any data is written.
*/
/* set entire read buffer with the constant 2 */
HDmemset(rdata,2,(size_t)(dset_size*sizeof(int)));
/* Independently read the entire dataset back */
ret=H5Dread(dataset, H5T_NATIVE_INT, H5S_ALL, H5S_ALL, H5P_DEFAULT, rdata);
VRFY((ret >= 0), "H5Dread succeeded");
/* Verify all data read are the fill value 0 */
trdata=rdata;
err_num=0;
for (i=0; i<(int)dset_dims[0]; i++)
for (j=0; j<(int)dset_dims[1]; j++)
for (k=0; k<(int)dset_dims[2]; k++)
for (l=0; l<(int)dset_dims[3]; l++, twdata++, trdata++)
if( *trdata != 0)
if(err_num++ < MAX_ERR_REPORT || VERBOSE_MED)
printf("Dataset Verify failed at [%d][%d][%d][%d]: expect 0, got %d\n", i,j,k,l, *trdata);
if(err_num > MAX_ERR_REPORT && !VERBOSE_MED)
printf("[more errors ...]\n");
if(err_num){
printf("%d errors found in check_value\n", err_num);
nerrors++;
}
/* Barrier to ensure all processes have completed the above test. */
MPI_Barrier(MPI_COMM_WORLD);
/*
* Each process writes 1 row of data. Thus last row is not written.
*/
/* Create hyperslabs in memory and file dataspaces */
req_start[0]=mpi_rank;
ret=H5Sselect_hyperslab(filespace, H5S_SELECT_SET, req_start, NULL, req_count, NULL);
VRFY((ret >= 0), "H5Sselect_hyperslab succeeded on memory dataspace");
ret=H5Sselect_hyperslab(memspace, H5S_SELECT_SET, req_start, NULL, req_count, NULL);
VRFY((ret >= 0), "H5Sselect_hyperslab succeeded on memory dataspace");
/* Create DXPL for collective I/O */
dxpl = H5Pcreate (H5P_DATASET_XFER);
VRFY((dxpl >= 0), "H5Pcreate succeeded");
ret=H5Pset_dxpl_mpio(dxpl, H5FD_MPIO_COLLECTIVE);
VRFY((ret >= 0), "H5Pset_dxpl_mpio succeeded");
/* Fill write buffer with some values */
twdata=wdata;
for (i=0, acc=0; i<(int)dset_dims[0]; i++)
for (j=0; j<(int)dset_dims[1]; j++)
for (k=0; k<(int)dset_dims[2]; k++)
for (l=0; l<(int)dset_dims[3]; l++)
*twdata++ = acc++;
/* Collectively write a hyperslab of data to the dataset */
ret=H5Dwrite(dataset, H5T_NATIVE_INT, memspace, filespace, dxpl, wdata);
VRFY((ret >= 0), "H5Dwrite succeeded");
/* Barrier here, to allow MPI-posix I/O to sync */
MPI_Barrier(MPI_COMM_WORLD);
/*
* Read dataset after partial write.
*/
/* set entire read buffer with the constant 2 */
HDmemset(rdata,2,(size_t)(dset_size*sizeof(int)));
/* Independently read the entire dataset back */
ret=H5Dread(dataset, H5T_NATIVE_INT, H5S_ALL, H5S_ALL, H5P_DEFAULT, rdata);
VRFY((ret >= 0), "H5Dread succeeded");
/* Verify correct data read */
twdata=wdata;
trdata=rdata;
err_num=0;
for (i=0; i<(int)dset_dims[0]; i++)
for (j=0; j<(int)dset_dims[1]; j++)
for (k=0; k<(int)dset_dims[2]; k++)
for (l=0; l<(int)dset_dims[3]; l++, twdata++, trdata++)
if(i<mpi_size) {
if( *twdata != *trdata )
if(err_num++ < MAX_ERR_REPORT || VERBOSE_MED)
printf("Dataset Verify failed at [%d][%d][%d][%d]: expect %d, got %d\n", i,j,k,l, *twdata, *trdata);
} /* end if */
else {
if( *trdata != 0)
if(err_num++ < MAX_ERR_REPORT || VERBOSE_MED)
printf("Dataset Verify failed at [%d][%d][%d][%d]: expect 0, got %d\n", i,j,k,l, *trdata);
} /* end else */
if(err_num > MAX_ERR_REPORT && !VERBOSE_MED)
printf("[more errors ...]\n");
if(err_num){
printf("%d errors found in check_value\n", err_num);
nerrors++;
}
/* Close all file objects */
ret=H5Dclose (dataset);
VRFY((ret >= 0), "H5Dclose succeeded");
ret=H5Sclose (filespace);
VRFY((ret >= 0), "H5Sclose succeeded");
ret=H5Fclose (iof);
VRFY((ret >= 0), "H5Fclose succeeded");
/* Close memory dataspace */
ret=H5Sclose (memspace);
VRFY((ret >= 0), "H5Sclose succeeded");
/* Close dxpl */
ret=H5Pclose (dxpl);
VRFY((ret >= 0), "H5Pclose succeeded");
/* Close fapl */
ret=H5Pclose (fapl);
VRFY((ret >= 0), "H5Pclose succeeded");
/* free the buffers */
HDfree(rdata);
HDfree(wdata);
}
/* Write multiple groups with a chunked dataset in each group collectively.
* These groups and datasets are for testing independent read later.
*
* Changes: Updated function to use a dynamically calculated size,
* instead of the old SIZE #define. This should allow it
* to function with an arbitrary number of processors.
*
* JRM - 8/16/04
*/
void collective_group_write(void)
{
int mpi_rank, mpi_size, size;
int i, j, m;
hbool_t use_gpfs = FALSE;
char gname[64], dname[32];
hid_t fid, gid, did, plist, dcpl, memspace, filespace;
DATATYPE * outme = NULL;
hsize_t chunk_origin[DIM];
hsize_t chunk_dims[DIM], file_dims[DIM], count[DIM];
hsize_t chunk_size[2]; /* Chunk dimensions - computed shortly */
herr_t ret1, ret2;
const H5Ptest_param_t *pt;
char *filename;
int ngroups;
pt = GetTestParameters();
filename = pt->name;
ngroups = pt->count;
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
size = get_size();
chunk_size[0] = (hsize_t)(size / 2);
chunk_size[1] = (hsize_t)(size / 2);
outme = HDmalloc((size_t)(size * size * sizeof(DATATYPE)));
VRFY((outme != NULL), "HDmalloc succeeded for outme");
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
fid = H5Fcreate(filename, H5F_ACC_TRUNC, H5P_DEFAULT, plist);
H5Pclose(plist);
/* decide the hyperslab according to process number. */
get_slab(chunk_origin, chunk_dims, count, file_dims, size);
/* select hyperslab in memory and file spaces. These two operations are
* identical since the datasets are the same. */
memspace = H5Screate_simple(DIM, file_dims, NULL);
ret1 = H5Sselect_hyperslab(memspace, H5S_SELECT_SET, chunk_origin,
chunk_dims, count, chunk_dims);
filespace = H5Screate_simple(DIM, file_dims, NULL);
ret2 = H5Sselect_hyperslab(filespace, H5S_SELECT_SET, chunk_origin,
chunk_dims, count, chunk_dims);
VRFY((memspace>=0), "memspace");
VRFY((filespace>=0), "filespace");
VRFY((ret1>=0), "mgroup memspace selection");
VRFY((ret2>=0), "mgroup filespace selection");
dcpl = H5Pcreate(H5P_DATASET_CREATE);
ret1 = H5Pset_chunk (dcpl, 2, chunk_size);
VRFY((dcpl>=0), "dataset creation property");
VRFY((ret1>=0), "set chunk for dataset creation property");
/* creates ngroups groups under the root group, writes chunked
* datasets in parallel. */
for(m = 0; m < ngroups; m++) {
sprintf(gname, "group%d", m);
gid = H5Gcreate(fid, gname, 0);
VRFY((gid > 0), gname);
sprintf(dname, "dataset%d", m);
did = H5Dcreate(gid, dname, H5T_NATIVE_INT, filespace, dcpl);
VRFY((did > 0), dname);
for(i=0; i < size; i++)
for(j=0; j < size; j++)
outme[(i * size) + j] = (i+j)*1000 + mpi_rank;
H5Dwrite(did, H5T_NATIVE_INT, memspace, filespace, H5P_DEFAULT,
outme);
H5Dclose(did);
H5Gclose(gid);
#ifdef BARRIER_CHECKS
if(! ((m+1) % 10)) {
printf("created %d groups\n", m+1);
MPI_Barrier(MPI_COMM_WORLD);
}
#endif /* BARRIER_CHECKS */
}
H5Pclose(dcpl);
H5Sclose(filespace);
H5Sclose(memspace);
H5Fclose(fid);
HDfree(outme);
}
/* Let two sets of processes open and read different groups and chunked
* datasets independently.
*/
void independent_group_read(void)
{
int mpi_rank, m;
hid_t plist, fid;
hbool_t use_gpfs = FALSE;
const H5Ptest_param_t *pt;
char *filename;
int ngroups;
pt = GetTestParameters();
filename = pt->name;
ngroups = pt->count;
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
fid = H5Fopen(filename, H5F_ACC_RDONLY, plist);
H5Pclose(plist);
/* open groups and read datasets. Odd number processes read even number
* groups from the end; even number processes read odd number groups
* from the beginning. */
if(mpi_rank%2==0) {
for(m=ngroups-1; m==0; m-=2)
group_dataset_read(fid, mpi_rank, m);
} else {
for(m=0; m<ngroups; m+=2)
group_dataset_read(fid, mpi_rank, m);
}
H5Fclose(fid);
}
/* Open and read datasets and compare data
*
* Changes: Updated function to use a dynamically calculated size,
* instead of the old SIZE #define. This should allow it
* to function with an arbitrary number of processors.
*
* Also added code to verify the results of dynamic memory
* allocations, and to free dynamically allocated memeory
* when we are done with it.
*
* JRM - 8/16/04
*/
void group_dataset_read(hid_t fid, int mpi_rank, int m)
{
int ret, i, j, size;
char gname[64], dname[32];
hid_t gid, did;
DATATYPE *outdata = NULL;
DATATYPE *indata = NULL;
size = get_size();
indata = (DATATYPE*)HDmalloc((size_t)(size * size * sizeof(DATATYPE)));
VRFY((indata != NULL), "HDmalloc succeeded for indata");
outdata = (DATATYPE*)HDmalloc((size_t)(size * size * sizeof(DATATYPE)));
VRFY((outdata != NULL), "HDmalloc succeeded for outdata");
/* open every group under root group. */
sprintf(gname, "group%d", m);
gid = H5Gopen(fid, gname);
VRFY((gid > 0), gname);
/* check the data. */
sprintf(dname, "dataset%d", m);
did = H5Dopen(gid, dname);
VRFY((did>0), dname);
H5Dread(did, H5T_NATIVE_INT, H5S_ALL, H5S_ALL, H5P_DEFAULT, indata);
/* this is the original value */
for(i=0; i<size; i++)
for(j=0; j<size; j++) {
outdata[(i * size) + j] = (i+j)*1000 + mpi_rank;
}
/* compare the original value(outdata) to the value in file(indata).*/
ret = check_value(indata, outdata, size);
VRFY((ret==0), "check the data");
H5Dclose(did);
H5Gclose(gid);
HDfree(indata);
HDfree(outdata);
}
/*
* Example of using PHDF5 to create multiple groups. Under the root group,
* it creates ngroups groups. Under the first group just created, it creates
* recursive subgroups of depth GROUP_DEPTH. In each created group, it
* generates NDATASETS datasets. Each process write a hyperslab of an array
* into the file. The structure is like
*
* root group
* |
* ---------------------------- ... ... ------------------------
* | | | ... ... | |
* group0*+' group1*+' group2*+' ... ... group ngroups*+'
* |
* 1st_child_group*'
* |
* 2nd_child_group*'
* |
* :
* :
* |
* GROUP_DEPTHth_child_group*'
*
* * means the group has dataset(s).
* + means the group has attribute(s).
* ' means the datasets in the groups have attribute(s).
*
* Changes: Updated function to use a dynamically calculated size,
* instead of the old SIZE #define. This should allow it
* to function with an arbitrary number of processors.
*
* JRM - 8/16/04
*/
void multiple_group_write(void)
{
int mpi_rank, mpi_size, size;
int m;
hbool_t use_gpfs = FALSE;
char gname[64];
hid_t fid, gid, plist, memspace, filespace;
hsize_t chunk_origin[DIM];
hsize_t chunk_dims[DIM], file_dims[DIM], count[DIM];
herr_t ret;
const H5Ptest_param_t *pt;
char *filename;
int ngroups;
pt = GetTestParameters();
filename = pt->name;
ngroups = pt->count;
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
size = get_size();
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
fid = H5Fcreate(filename, H5F_ACC_TRUNC, H5P_DEFAULT, plist);
H5Pclose(plist);
/* decide the hyperslab according to process number. */
get_slab(chunk_origin, chunk_dims, count, file_dims, size);
/* select hyperslab in memory and file spaces. These two operations are
* identical since the datasets are the same. */
memspace = H5Screate_simple(DIM, file_dims, NULL);
VRFY((memspace>=0), "memspace");
ret = H5Sselect_hyperslab(memspace, H5S_SELECT_SET, chunk_origin,
chunk_dims, count, chunk_dims);
VRFY((ret>=0), "mgroup memspace selection");
filespace = H5Screate_simple(DIM, file_dims, NULL);
VRFY((filespace>=0), "filespace");
ret = H5Sselect_hyperslab(filespace, H5S_SELECT_SET, chunk_origin,
chunk_dims, count, chunk_dims);
VRFY((ret>=0), "mgroup filespace selection");
/* creates ngroups groups under the root group, writes datasets in
* parallel. */
for(m = 0; m < ngroups; m++) {
sprintf(gname, "group%d", m);
gid = H5Gcreate(fid, gname, 0);
VRFY((gid > 0), gname);
/* create attribute for these groups. */
write_attribute(gid, is_group, m);
if(m != 0)
write_dataset(memspace, filespace, gid);
H5Gclose(gid);
#ifdef BARRIER_CHECKS
if(! ((m+1) % 10)) {
printf("created %d groups\n", m+1);
MPI_Barrier(MPI_COMM_WORLD);
}
#endif /* BARRIER_CHECKS */
}
/* recursively creates subgroups under the first group. */
gid = H5Gopen(fid, "group0");
create_group_recursive(memspace, filespace, gid, 0);
ret = H5Gclose(gid);
VRFY((ret>=0), "H5Gclose");
ret = H5Sclose(filespace);
VRFY((ret>=0), "H5Sclose");
ret = H5Sclose(memspace);
VRFY((ret>=0), "H5Sclose");
ret = H5Fclose(fid);
VRFY((ret>=0), "H5Fclose");
}
/*
* In a group, creates NDATASETS datasets. Each process writes a hyperslab
* of a data array to the file.
*
* Changes: Updated function to use a dynamically calculated size,
* instead of the old SIZE #define. This should allow it
* to function with an arbitrary number of processors.
*
* JRM - 8/16/04
*/
void write_dataset(hid_t memspace, hid_t filespace, hid_t gid)
{
int i, j, n, size;
int mpi_rank, mpi_size;
char dname[32];
DATATYPE * outme = NULL;
hid_t did;
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
size = get_size();
outme = HDmalloc((size_t)(size * size * sizeof(double)));
VRFY((outme != NULL), "HDmalloc succeeded for outme");
for(n=0; n < NDATASET; n++) {
sprintf(dname, "dataset%d", n);
did = H5Dcreate(gid, dname, H5T_NATIVE_INT, filespace,
H5P_DEFAULT);
VRFY((did > 0), dname);
for(i=0; i < size; i++)
for(j=0; j < size; j++)
outme[(i * size) + j] = n*1000 + mpi_rank;
H5Dwrite(did, H5T_NATIVE_INT, memspace, filespace, H5P_DEFAULT,
outme);
/* create attribute for these datasets.*/
write_attribute(did, is_dset, n);
H5Dclose(did);
}
HDfree(outme);
}
/*
* Creates subgroups of depth GROUP_DEPTH recursively. Also writes datasets
* in parallel in each group.
*/
void create_group_recursive(hid_t memspace, hid_t filespace, hid_t gid,
int counter)
{
hid_t child_gid;
int mpi_rank;
char gname[64];
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
#ifdef BARRIER_CHECKS
if(! ((counter+1) % 10)) {
printf("created %dth child groups\n", counter+1);
MPI_Barrier(MPI_COMM_WORLD);
}
#endif /* BARRIER_CHECKS */
sprintf(gname, "%dth_child_group", counter+1);
child_gid = H5Gcreate(gid, gname, 0);
VRFY((child_gid > 0), gname);
/* write datasets in parallel. */
write_dataset(memspace, filespace, gid);
if( counter < GROUP_DEPTH )
create_group_recursive(memspace, filespace, child_gid, counter+1);
H5Gclose(child_gid);
}
/*
* This function is to verify the data from multiple group testing. It opens
* every dataset in every group and check their correctness.
*
* Changes: Updated function to use a dynamically calculated size,
* instead of the old SIZE #define. This should allow it
* to function with an arbitrary number of processors.
*
* JRM - 8/11/04
*/
void multiple_group_read(void)
{
int mpi_rank, mpi_size, error_num, size;
int m;
hbool_t use_gpfs = FALSE;
char gname[64];
hid_t plist, fid, gid, memspace, filespace;
hsize_t chunk_origin[DIM];
hsize_t chunk_dims[DIM], file_dims[DIM], count[DIM];
const H5Ptest_param_t *pt;
char *filename;
int ngroups;
pt = GetTestParameters();
filename = pt->name;
ngroups = pt->count;
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
size = get_size();
plist = create_faccess_plist(MPI_COMM_WORLD, MPI_INFO_NULL, facc_type, use_gpfs);
fid = H5Fopen(filename, H5F_ACC_RDONLY, plist);
H5Pclose(plist);
/* decide hyperslab for each process */
get_slab(chunk_origin, chunk_dims, count, file_dims, size);
/* select hyperslab for memory and file space */
memspace = H5Screate_simple(DIM, file_dims, NULL);
H5Sselect_hyperslab(memspace, H5S_SELECT_SET, chunk_origin, chunk_dims,
count, chunk_dims);
filespace = H5Screate_simple(DIM, file_dims, NULL);
H5Sselect_hyperslab(filespace, H5S_SELECT_SET, chunk_origin, chunk_dims,
count, chunk_dims);
/* open every group under root group. */
for(m=0; m<ngroups; m++) {
sprintf(gname, "group%d", m);
gid = H5Gopen(fid, gname);
VRFY((gid > 0), gname);
/* check the data. */
if(m != 0)
if( (error_num = read_dataset(memspace, filespace, gid))>0)
nerrors += error_num;
/* check attribute.*/
error_num = 0;
if( (error_num = read_attribute(gid, is_group, m))>0 )
nerrors += error_num;
H5Gclose(gid);
#ifdef BARRIER_CHECKS
if(!((m+1)%10))
MPI_Barrier(MPI_COMM_WORLD);
#endif /* BARRIER_CHECKS */
}
/* open all the groups in vertical direction. */
gid = H5Gopen(fid, "group0");
VRFY((gid>0), "group0");
recursive_read_group(memspace, filespace, gid, 0);
H5Gclose(gid);
H5Sclose(filespace);
H5Sclose(memspace);
H5Fclose(fid);
}
/*
* This function opens all the datasets in a certain, checks the data using
* dataset_vrfy function.
*
* Changes: Updated function to use a dynamically calculated size,
* instead of the old SIZE #define. This should allow it
* to function with an arbitrary number of processors.
*
* JRM - 8/11/04
*/
int read_dataset(hid_t memspace, hid_t filespace, hid_t gid)
{
int i, j, n, mpi_rank, mpi_size, size, attr_errors=0, vrfy_errors=0;
char dname[32];
DATATYPE *outdata = NULL, *indata = NULL;
hid_t did;
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
size = get_size();
indata = (DATATYPE*)HDmalloc((size_t)(size * size * sizeof(DATATYPE)));
VRFY((indata != NULL), "HDmalloc succeeded for indata");
outdata = (DATATYPE*)HDmalloc((size_t)(size * size * sizeof(DATATYPE)));
VRFY((outdata != NULL), "HDmalloc succeeded for outdata");
for(n=0; n<NDATASET; n++) {
sprintf(dname, "dataset%d", n);
did = H5Dopen(gid, dname);
VRFY((did>0), dname);
H5Dread(did, H5T_NATIVE_INT, memspace, filespace, H5P_DEFAULT,
indata);
/* this is the original value */
for(i=0; i<size; i++)
for(j=0; j<size; j++) {
*outdata = n*1000 + mpi_rank;
outdata++;
}
outdata -= size * size;
/* compare the original value(outdata) to the value in file(indata).*/
vrfy_errors = check_value(indata, outdata, size);
/* check attribute.*/
if( (attr_errors = read_attribute(did, is_dset, n))>0 )
vrfy_errors += attr_errors;
H5Dclose(did);
}
HDfree(indata);
HDfree(outdata);
return vrfy_errors;
}
/*
* This recursive function opens all the groups in vertical direction and
* checks the data.
*/
void recursive_read_group(hid_t memspace, hid_t filespace, hid_t gid,
int counter)
{
hid_t child_gid;
int mpi_rank, err_num=0;
char gname[64];
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
#ifdef BARRIER_CHECKS
if((counter+1) % 10)
MPI_Barrier(MPI_COMM_WORLD);
#endif /* BARRIER_CHECKS */
if( (err_num = read_dataset(memspace, filespace, gid)) )
nerrors += err_num;
if( counter < GROUP_DEPTH ) {
sprintf(gname, "%dth_child_group", counter+1);
child_gid = H5Gopen(gid, gname);
VRFY((child_gid>0), gname);
recursive_read_group(memspace, filespace, child_gid, counter+1);
H5Gclose(child_gid);
}
}
/* Create and write attribute for a group or a dataset. For groups, attribute
* is a scalar datum; for dataset, it is a one-dimensional array.
*/
void write_attribute(hid_t obj_id, int this_type, int num)
{
hid_t sid, aid;
hsize_t dspace_dims[1]={8};
int i, mpi_rank, attr_data[8], dspace_rank=1;
char attr_name[32];
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
if(this_type == is_group) {
sprintf(attr_name, "Group Attribute %d", num);
sid = H5Screate(H5S_SCALAR);
aid = H5Acreate(obj_id, attr_name, H5T_NATIVE_INT, sid, H5P_DEFAULT);
H5Awrite(aid, H5T_NATIVE_INT, &num);
H5Aclose(aid);
H5Sclose(sid);
}
else if(this_type == is_dset) {
sprintf(attr_name, "Dataset Attribute %d", num);
for(i=0; i<8; i++)
attr_data[i] = i;
sid = H5Screate_simple(dspace_rank, dspace_dims, NULL);
aid = H5Acreate(obj_id, attr_name, H5T_NATIVE_INT, sid, H5P_DEFAULT);
H5Awrite(aid, H5T_NATIVE_INT, attr_data);
H5Aclose(aid);
H5Sclose(sid);
}
}
/* Read and verify attribute for group or dataset. */
int read_attribute(hid_t obj_id, int this_type, int num)
{
hid_t aid;
hsize_t group_block[2]={1,1}, dset_block[2]={1, 8};
int i, mpi_rank, in_num, in_data[8], out_data[8], vrfy_errors = 0;
char attr_name[32];
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
if(this_type == is_group) {
sprintf(attr_name, "Group Attribute %d", num);
aid = H5Aopen_name(obj_id, attr_name);
if(MAINPROCESS) {
H5Aread(aid, H5T_NATIVE_INT, &in_num);
vrfy_errors = dataset_vrfy(NULL, NULL, NULL, group_block,
&in_num, &num);
}
H5Aclose(aid);
}
else if(this_type == is_dset) {
sprintf(attr_name, "Dataset Attribute %d", num);
for(i=0; i<8; i++)
out_data[i] = i;
aid = H5Aopen_name(obj_id, attr_name);
if(MAINPROCESS) {
H5Aread(aid, H5T_NATIVE_INT, in_data);
vrfy_errors = dataset_vrfy(NULL, NULL, NULL, dset_block, in_data,
out_data);
}
H5Aclose(aid);
}
return vrfy_errors;
}
/* This functions compares the original data with the read-in data for its
* hyperslab part only by process ID.
*
* Changes: Modified function to use a passed in size parameter
* instead of the old SIZE #define. This should let us
* run with an arbitrary number of processes.
*
* JRM - 8/16/04
*/
int check_value(DATATYPE *indata, DATATYPE *outdata, int size)
{
int mpi_rank, mpi_size, err_num=0;
hsize_t i, j;
hsize_t chunk_origin[DIM];
hsize_t chunk_dims[DIM], count[DIM];
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
get_slab(chunk_origin, chunk_dims, count, NULL, size);
indata += chunk_origin[0]*size;
outdata += chunk_origin[0]*size;
for(i=chunk_origin[0]; i<(chunk_origin[0]+chunk_dims[0]); i++)
for(j=chunk_origin[1]; j<(chunk_origin[1]+chunk_dims[1]); j++) {
if( *indata != *outdata )
if(err_num++ < MAX_ERR_REPORT || VERBOSE_MED)
printf("Dataset Verify failed at [%lu][%lu](row %lu, col%lu): expect %d, got %d\n", (unsigned long)i, (unsigned long)j, (unsigned long)i, (unsigned long)j, *outdata, *indata);
}
if(err_num > MAX_ERR_REPORT && !VERBOSE_MED)
printf("[more errors ...]\n");
if(err_num)
printf("%d errors found in check_value\n", err_num);
return err_num;
}
/* Decide the portion of data chunk in dataset by process ID.
*
* Changes: Modified function to use a passed in size parameter
* instead of the old SIZE #define. This should let us
* run with an arbitrary number of processes.
*
* JRM - 8/11/04
*/
void get_slab(hsize_t chunk_origin[],
hsize_t chunk_dims[],
hsize_t count[],
hsize_t file_dims[],
int size)
{
int mpi_rank, mpi_size;
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
if(chunk_origin != NULL) {
chunk_origin[0] = mpi_rank * (size/mpi_size);
chunk_origin[1] = 0;
}
if(chunk_dims != NULL) {
chunk_dims[0] = size/mpi_size;
chunk_dims[1] = size;
}
if(file_dims != NULL)
file_dims[0] = file_dims[1] = size;
if(count != NULL)
count[0] = count[1] = 1;
}
/*
* This function is based on bug demonstration code provided by Thomas
* Guignon (thomas.guignon@ifp.fr), and is intended to verify the
* correctness of my fix for that bug.
*
* In essence, the bug appeared when at least one process attempted to
* write a point selection -- for which collective I/O is not supported,
* and at least one other attempted to write some other type of selection
* for which collective I/O is supported.
*
* Since the processes did not compare notes before performing the I/O,
* some would attempt collective I/O while others performed independent
* I/O. A hang resulted.
*
* This function reproduces this situation. At present the test hangs
* on failure.
* JRM - 9/13/04
*
* Changes: None.
*/
#define N 4
void io_mode_confusion(void)
{
/*
* HDF5 APIs definitions
*/
const int rank = 1;
const char *dataset_name = "IntArray";
hid_t file_id, dset_id; /* file and dataset identifiers */
hid_t filespace, memspace; /* file and memory dataspace */
/* identifiers */
hsize_t dimsf[1]; /* dataset dimensions */
int data[N] = {1}; /* pointer to data buffer to write */
hsize_t coord[N] = {0L,1L,2L,3L};
hsize_t start[1];
hsize_t stride[1];
hsize_t count[1];
hsize_t block[1];
hid_t plist_id; /* property list identifier */
herr_t status;
/*
* MPI variables
*/
int mpi_size, mpi_rank;
/*
* test bed related variables
*/
const char * fcn_name = "io_mode_confusion";
const hbool_t verbose = FALSE;
const H5Ptest_param_t * pt;
char * filename;
pt = GetTestParameters();
filename = pt->name;
MPI_Comm_rank(MPI_COMM_WORLD, &mpi_rank);
MPI_Comm_size(MPI_COMM_WORLD, &mpi_size);
/*
* Set up file access property list with parallel I/O access
*/
if ( verbose )
HDfprintf(stdout, "%0d:%s: Setting up property list.\n",
mpi_rank, fcn_name);
plist_id = H5Pcreate(H5P_FILE_ACCESS);
VRFY((plist_id != -1), "H5Pcreate() failed");
status = H5Pset_fapl_mpio(plist_id, MPI_COMM_WORLD, MPI_INFO_NULL);
VRFY(( status >= 0 ), "H5Pset_fapl_mpio() failed");
/*
* Create a new file collectively and release property list identifier.
*/
if ( verbose )
HDfprintf(stdout, "%0d:%s: Creating new file.\n", mpi_rank, fcn_name);
file_id = H5Fcreate(filename, H5F_ACC_TRUNC, H5P_DEFAULT, plist_id);
VRFY(( file_id >= 0 ), "H5Fcreate() failed");
status = H5Pclose(plist_id);
VRFY(( status >= 0 ), "H5Pclose() failed");
/*
* Create the dataspace for the dataset.
*/
if ( verbose )
HDfprintf(stdout, "%0d:%s: Creating the dataspace for the dataset.\n",
mpi_rank, fcn_name);
dimsf[0] = N;
filespace = H5Screate_simple(rank, dimsf, NULL);
VRFY(( filespace >= 0 ), "H5Screate_simple() failed.");
/*
* Create the dataset with default properties and close filespace.
*/
if ( verbose )
HDfprintf(stdout,
"%0d:%s: Creating the dataset, and closing filespace.\n",
mpi_rank, fcn_name);
dset_id = H5Dcreate(file_id, dataset_name, H5T_NATIVE_INT, filespace,
H5P_DEFAULT);
VRFY(( dset_id >= 0 ), "H5Dcreate() failed");
status = H5Sclose(filespace);
VRFY(( status >= 0 ), "H5Sclose() failed");
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling H5Screate_simple().\n",
mpi_rank, fcn_name);
memspace = H5Screate_simple(rank, dimsf, NULL);
VRFY(( memspace >= 0 ), "H5Screate_simple() failed.");
if( mpi_rank == 0 ) {
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling H5Sselect_all(memspace).\n",
mpi_rank, fcn_name);
status = H5Sselect_all(memspace);
VRFY(( status >= 0 ), "H5Sselect_all() failed");
} else {
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling H5Sselect_none(memspace).\n",
mpi_rank, fcn_name);
status = H5Sselect_none(memspace);
VRFY(( status >= 0 ), "H5Sselect_none() failed");
}
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling MPI_Barrier().\n",
mpi_rank, fcn_name);
MPI_Barrier(MPI_COMM_WORLD);
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling H5Dget_space().\n",
mpi_rank, fcn_name);
filespace = H5Dget_space(dset_id);
VRFY(( filespace >= 0 ), "H5Dget_space() failed");
start[0] = 0L;
stride[0] = 1;
count[0] = 1;
block[0] = N;
if ( mpi_rank == 0 ) {
/* select all */
if ( verbose )
HDfprintf(stdout,
"%0d:%s: Calling H5Sselect_elements() -- set up hang?\n",
mpi_rank, fcn_name);
status = H5Sselect_elements(filespace, H5S_SELECT_SET, N,
(const hsize_t **)&coord);
VRFY(( status >= 0 ), "H5Sselect_elements() failed");
} else {
/* select nothing */
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling H5Sselect_none().\n",
mpi_rank, fcn_name);
status = H5Sselect_none(filespace);
VRFY(( status >= 0 ), "H5Sselect_none() failed");
}
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling MPI_Barrier().\n",
mpi_rank, fcn_name);
MPI_Barrier(MPI_COMM_WORLD);
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling H5Pcreate().\n", mpi_rank, fcn_name);
plist_id = H5Pcreate(H5P_DATASET_XFER);
VRFY(( plist_id != -1 ), "H5Pcreate() failed");
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling H5Pset_dxpl_mpio().\n",
mpi_rank, fcn_name);
status = H5Pset_dxpl_mpio(plist_id, H5FD_MPIO_COLLECTIVE);
VRFY(( status >= 0 ), "H5Pset_dxpl_mpio() failed");
if ( verbose )
HDfprintf(stdout, "%0d:%s: Calling H5Dwrite() -- hang here?.\n",
mpi_rank, fcn_name);
status = H5Dwrite(dset_id, H5T_NATIVE_INT, memspace, filespace,
plist_id, data);
if ( verbose )
HDfprintf(stdout, "%0d:%s: Returned from H5Dwrite(), status=%d.\n",
mpi_rank, fcn_name, status);
VRFY(( status >= 0 ), "H5Dwrite() failed");
/*
* Close/release resources.
*/
if ( verbose )
HDfprintf(stdout, "%0d:%s: Cleaning up from test.\n",
mpi_rank, fcn_name);
status = H5Dclose(dset_id);
VRFY(( status >= 0 ), "H5Dclose() failed");
status = H5Sclose(filespace);
VRFY(( status >= 0 ), "H5Dclose() failed");
status = H5Sclose(memspace);
VRFY(( status >= 0 ), "H5Sclose() failed");
status = H5Pclose(plist_id);
VRFY(( status >= 0 ), "H5Pclose() failed");
status = H5Fclose(file_id);
VRFY(( status >= 0 ), "H5Fclose() failed");
if ( verbose )
HDfprintf(stdout, "%0d:%s: Done.\n", mpi_rank);
return;
} /* io_mode_confusion() */
#undef N
/*=============================================================================
* End of t_mdset.c
*===========================================================================*/
Reference in New Issue View Git Blame Copy Permalink