int main (int argc, char **argv)
{
int rootid;
int iam = -1;
int ret;
ret = GPTLinitialize ();
ret = GPTLprint_memusage ("before_MPI_Init");
ret = MPI_Init (&argc, &argv);
ret = GPTLprint_memusage ("after_MPI_Init ");
ret = MPI_Comm_rank (MPI_COMM_WORLD, &iam);
if (iam == 0)
rootid = iam;
ret = MPI_Bcast (&rootid, 1, MPI_INT, 0, MPI_COMM_WORLD);
ret = GPTLprint_memusage ("after_MPI_Bcast ");
ret = MPI_Finalize ();
}
/** Run Tests for NetCDF-4 Functions.
*
* @param argc argument count
* @param argv array of arguments
*/
int
main(int argc, char **argv)
{
int verbose = 1;
/** Zero-based rank of processor. */
int my_rank;
/** Number of processors involved in current execution. */
int ntasks;
/** Specifies the flavor of netCDF output format. */
int iotype;
/** Different output flavors. */
int format[NUM_NETCDF_FLAVORS] = {PIO_IOTYPE_PNETCDF,
PIO_IOTYPE_NETCDF,
PIO_IOTYPE_NETCDF4C,
PIO_IOTYPE_NETCDF4P};
/** Names for the output files. */
char filename[NUM_NETCDF_FLAVORS][NC_MAX_NAME + 1] = {"test_names_pnetcdf.nc",
"test_names_classic.nc",
"test_names_serial4.nc",
"test_names_parallel4.nc"};
/** Number of processors that will do IO. In this test we
* will do IO from all processors. */
int niotasks;
/** Stride in the mpi rank between io tasks. Always 1 in this
* test. */
int ioproc_stride = 1;
/** Number of the aggregator? Always 0 in this test. */
int numAggregator = 0;
/** Zero based rank of first processor to be used for I/O. */
int ioproc_start = 0;
/** The dimension IDs. */
int dimids[NDIM];
/** Array index per processing unit. */
PIO_Offset elements_per_pe;
/** The ID for the parallel I/O system. */
int iosysid;
/** The ncid of the netCDF file. */
int ncid = 0;
/** The ID of the netCDF varable. */
int varid;
/** Storage of netCDF-4 files (contiguous vs. chunked). */
int storage;
/** Chunksizes set in the file. */
size_t my_chunksize[NDIM];
/** The shuffle filter setting in the netCDF-4 test file. */
int shuffle;
/** Non-zero if deflate set for the variable in the netCDF-4 test file. */
int deflate;
/** The deflate level set for the variable in the netCDF-4 test file. */
int deflate_level;
/** Non-zero if fletcher32 filter is used for variable. */
int fletcher32;
/** Endianness of variable. */
int endianness;
/* Size of the file chunk cache. */
size_t chunk_cache_size;
/* Number of elements in file cache. */
size_t nelems;
/* File cache preemption. */
float preemption;
/* Size of the var chunk cache. */
size_t var_cache_size;
/* Number of elements in var cache. */
size_t var_cache_nelems;
/* Var cache preemption. */
float var_cache_preemption;
/** The I/O description ID. */
int ioid;
/** A buffer for sample data. */
float *buffer;
/** A buffer for reading data back from the file. */
int *read_buffer;
/** The decomposition mapping. */
PIO_Offset *compdof;
/** Return code. */
int ret;
/** Index for loops. */
int fmt, d, d1, i;
#ifdef TIMING
/* Initialize the GPTL timing library. */
if ((ret = GPTLinitialize ()))
return ret;
#endif
/* Initialize MPI. */
if ((ret = MPI_Init(&argc, &argv)))
MPIERR(ret);
/* Learn my rank and the total number of processors. */
if ((ret = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank)))
MPIERR(ret);
if ((ret = MPI_Comm_size(MPI_COMM_WORLD, &ntasks)))
MPIERR(ret);
/* Check that a valid number of processors was specified. */
if (!(ntasks == 1 || ntasks == 2 || ntasks == 4 ||
ntasks == 8 || ntasks == 16))
fprintf(stderr, "Number of processors must be 1, 2, 4, 8, or 16!\n");
if (verbose)
printf("%d: ParallelIO Library example1 running on %d processors.\n",
my_rank, ntasks);
/* keep things simple - 1 iotask per MPI process */
niotasks = ntasks;
/* Initialize the PIO IO system. This specifies how
* many and which processors are involved in I/O. */
if ((ret = PIOc_Init_Intracomm(MPI_COMM_WORLD, niotasks, ioproc_stride,
ioproc_start, PIO_REARR_SUBSET, &iosysid)))
ERR(ret);
/* Describe the decomposition. This is a 1-based array, so add 1! */
elements_per_pe = X_DIM_LEN * Y_DIM_LEN / ntasks;
if (!(compdof = malloc(elements_per_pe * sizeof(PIO_Offset))))
return PIO_ENOMEM;
for (i = 0; i < elements_per_pe; i++) {
compdof[i] = my_rank * elements_per_pe + i + 1;
}
/* Create the PIO decomposition for this test. */
if (verbose)
printf("rank: %d Creating decomposition...\n", my_rank);
if ((ret = PIOc_InitDecomp(iosysid, PIO_FLOAT, 2, &dim_len[1], (PIO_Offset)elements_per_pe,
compdof, &ioid, NULL, NULL, NULL)))
ERR(ret);
free(compdof);
/* How many flavors will we be running for? */
int num_flavors = 0;
int fmtidx = 0;
#ifdef _PNETCDF
num_flavors++;
format[fmtidx++] = PIO_IOTYPE_PNETCDF;
#endif
#ifdef _NETCDF
num_flavors++;
format[fmtidx++] = PIO_IOTYPE_NETCDF;
#endif
#ifdef _NETCDF4
num_flavors += 2;
format[fmtidx++] = PIO_IOTYPE_NETCDF4C;
format[fmtidx] = PIO_IOTYPE_NETCDF4P;
#endif
/* Use PIO to create the example file in each of the four
* available ways. */
for (fmt = 0; fmt < num_flavors; fmt++)
{
/* Create the netCDF output file. */
if (verbose)
printf("rank: %d Creating sample file %s with format %d...\n",
my_rank, filename[fmt], format[fmt]);
if ((ret = PIOc_createfile(iosysid, &ncid, &(format[fmt]), filename[fmt],
PIO_CLOBBER)))
ERR(ret);
/* Define netCDF dimensions and variable. */
if (verbose)
printf("rank: %d Defining netCDF metadata...\n", my_rank);
for (d = 0; d < NDIM; d++) {
if (verbose)
printf("rank: %d Defining netCDF dimension %s, length %d\n", my_rank,
dim_name[d], dim_len[d]);
if ((ret = PIOc_def_dim(ncid, dim_name[d], (PIO_Offset)dim_len[d], &dimids[d])))
ERR(ret);
}
/* Check the dimension names. */
if ((ret = check_dim_names(my_rank, ncid, verbose)))
ERR(ret);
/* Define a global attribute. */
int att_val = 42;
if ((ret = PIOc_put_att_int(ncid, NC_GLOBAL, ATT_NAME, NC_INT, 1, &att_val)))
ERR(ret);
/* Check the attribute name. */
if ((ret = check_att_name(my_rank, ncid, verbose)))
ERR(ret);
/* Define a variable. */
if ((ret = PIOc_def_var(ncid, VAR_NAME, PIO_FLOAT, NDIM, dimids, &varid)))
ERR(ret);
/* Check the variable name. */
if ((ret = check_var_name(my_rank, ncid, verbose)))
ERR(ret);
if ((ret = PIOc_enddef(ncid)))
ERR(ret);
/* Close the netCDF file. */
if (verbose)
printf("rank: %d Closing the sample data file...\n", my_rank);
if ((ret = PIOc_closefile(ncid)))
ERR(ret);
/* Put a barrier here to make verbose output look better. */
if ((ret = MPI_Barrier(MPI_COMM_WORLD)))
MPIERR(ret);
}
/* Free the PIO decomposition. */
if (verbose)
printf("rank: %d Freeing PIO decomposition...\n", my_rank);
if ((ret = PIOc_freedecomp(iosysid, ioid)))
ERR(ret);
/* Finalize the IO system. */
if (verbose)
printf("rank: %d Freeing PIO resources...\n", my_rank);
if ((ret = PIOc_finalize(iosysid)))
ERR(ret);
/* Finalize the MPI library. */
MPI_Finalize();
#ifdef TIMING
/* Finalize the GPTL timing library. */
if ((ret = GPTLfinalize ()))
return ret;
#endif
return 0;
}
/** Run Tests for NetCDF-4 Functions.
*
* @param argc argument count
* @param argv array of arguments
*/
int
main(int argc, char **argv)
{
int verbose = 1;
/** Zero-based rank of processor. */
int my_rank;
/** Number of processors involved in current execution. */
int ntasks;
/** Different output flavors. The example file is written (and
* then read) four times. The first two flavors,
* parallel-netcdf, and netCDF serial, both produce a netCDF
* classic format file (but with different libraries). The
* last two produce netCDF4/HDF5 format files, written with
* and without using netCDF-4 parallel I/O. */
int format[NUM_NETCDF_FLAVORS] = {PIO_IOTYPE_PNETCDF,
PIO_IOTYPE_NETCDF,
PIO_IOTYPE_NETCDF4C,
PIO_IOTYPE_NETCDF4P};
/** Names for the output files. Two of them (pnetcdf and
* classic) will be in classic netCDF format, the others
* (serial4 and parallel4) will be in netCDF-4/HDF5
* format. All four can be read by the netCDF library, and all
* will contain the same contents. */
char filename[NUM_NETCDF_FLAVORS][NC_MAX_NAME + 1] = {"test_nc4_pnetcdf.nc",
"test_nc4_classic.nc",
"test_nc4_serial4.nc",
"test_nc4_parallel4.nc"};
/** Number of processors that will do IO. In this example we
* will do IO from all processors. */
int niotasks;
/** Stride in the mpi rank between io tasks. Always 1 in this
* example. */
int ioproc_stride = 1;
/** Number of the aggregator? Always 0 in this example. */
int numAggregator = 0;
/** Zero based rank of first processor to be used for I/O. */
int ioproc_start = 0;
/** Specifies the flavor of netCDF output format. */
int iotype;
/** The dimension IDs. */
int dimids[NDIM];
/** Array index per processing unit. This is the number of
* elements of the data array that will be handled by each
* processor. In this example there are 16 data elements. If the
* example is run on 4 processors, then arrIdxPerPe will be 4. */
PIO_Offset elements_per_pe;
/** The ID for the parallel I/O system. It is set by
* PIOc_Init_Intracomm(). It references an internal structure
* containing the general IO subsystem data and MPI
* structure. It is passed to PIOc_finalize() to free
* associated resources, after all I/O, but before
* MPI_Finalize is called. */
int iosysid;
/** The ncid of the netCDF file created in this example. */
int ncid = 0;
/** The ID of the netCDF varable in the example file. */
int varid;
/** The I/O description ID as passed back by PIOc_InitDecomp()
* and freed in PIOc_freedecomp(). */
int ioid;
/** A buffer for sample data. The size of this array will
* vary depending on how many processors are involved in the
* execution of the example code. It's length will be the same
* as elements_per_pe.*/
float *buffer;
/** A buffer for reading data back from the file. The size of
* this array will vary depending on how many processors are
* involved in the execution of the example code. It's length
* will be the same as elements_per_pe.*/
int *read_buffer;
/** A 1-D array which holds the decomposition mapping for this
* example. The size of this array will vary depending on how
* many processors are involved in the execution of the
* example code. It's length will be the same as
* elements_per_pe. */
PIO_Offset *compdof;
/** Return code. */
int ret;
#ifdef TIMING
/* Initialize the GPTL timing library. */
if ((ret = GPTLinitialize ()))
return ret;
#endif
/* Initialize MPI. */
if ((ret = MPI_Init(&argc, &argv)))
MPIERR(ret);
if ((ret = MPI_Errhandler_set(MPI_COMM_WORLD, MPI_ERRORS_RETURN)))
MPIERR(ret);
/* Learn my rank and the total number of processors. */
if ((ret = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank)))
MPIERR(ret);
if ((ret = MPI_Comm_size(MPI_COMM_WORLD, &ntasks)))
MPIERR(ret);
/* Check that a valid number of processors was specified. */
if (!(ntasks == 1 || ntasks == 2 || ntasks == 4 ||
ntasks == 8 || ntasks == 16))
fprintf(stderr, "Number of processors must be 1, 2, 4, 8, or 16!\n");
if (verbose)
printf("%d: ParallelIO Library example1 running on %d processors.\n",
my_rank, ntasks);
/* keep things simple - 1 iotask per MPI process */
niotasks = ntasks;
/* Initialize the PIO IO system. This specifies how
* many and which processors are involved in I/O. */
if ((ret = PIOc_Init_Intracomm(MPI_COMM_WORLD, niotasks, ioproc_stride,
ioproc_start, PIO_REARR_SUBSET, &iosysid)))
ERR(ret);
/* Describe the decomposition. This is a 1-based array, so add 1! */
elements_per_pe = X_DIM_LEN * Y_DIM_LEN / ntasks;
if (!(compdof = malloc(elements_per_pe * sizeof(PIO_Offset))))
return PIO_ENOMEM;
for (int i = 0; i < elements_per_pe; i++) {
compdof[i] = my_rank * elements_per_pe + i + 1;
}
/* Create the PIO decomposition for this test. */
if (verbose)
printf("rank: %d Creating decomposition...\n", my_rank);
if ((ret = PIOc_InitDecomp(iosysid, PIO_FLOAT, 2, &dim_len[1], (PIO_Offset)elements_per_pe,
compdof, &ioid, NULL, NULL, NULL)))
ERR(ret);
free(compdof);
#ifdef HAVE_MPE
/* Log with MPE that we are done with INIT. */
if ((ret = MPE_Log_event(event_num[END][INIT], 0, "end init")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Use PIO to create the example file in each of the four
* available ways. */
for (int fmt = 0; fmt < NUM_NETCDF_FLAVORS; fmt++)
{
#ifdef HAVE_MPE
/* Log with MPE that we are starting CREATE. */
if ((ret = MPE_Log_event(event_num[START][CREATE_PNETCDF+fmt], 0, "start create")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Create the netCDF output file. */
if (verbose)
printf("rank: %d Creating sample file %s with format %d...\n",
my_rank, filename[fmt], format[fmt]);
if ((ret = PIOc_createfile(iosysid, &ncid, &(format[fmt]), filename[fmt],
PIO_CLOBBER)))
ERR(ret);
/* Define netCDF dimensions and variable. */
if (verbose)
printf("rank: %d Defining netCDF metadata...\n", my_rank);
for (int d = 0; d < NDIM; d++) {
if (verbose)
printf("rank: %d Defining netCDF dimension %s, length %d\n", my_rank,
dim_name[d], dim_len[d]);
if ((ret = PIOc_def_dim(ncid, dim_name[d], (PIO_Offset)dim_len[d], &dimids[d])))
ERR(ret);
}
if ((ret = PIOc_def_var(ncid, VAR_NAME, PIO_FLOAT, NDIM, dimids, &varid)))
ERR(ret);
/* For netCDF-4 files, set the chunksize to improve performance. */
if (format[fmt] == PIO_IOTYPE_NETCDF4C || format[fmt] == PIO_IOTYPE_NETCDF4P)
{
if ((ret = PIOc_def_var_chunking(ncid, 0, NC_CHUNKED, chunksize)))
ERR(ret);
/** Check that the inq_var_chunking function works. */
int storage;
size_t my_chunksize[NDIM];
if ((ret = PIOc_inq_var_chunking(ncid, 0, &storage, my_chunksize)))
ERR(ret);
/** For serial netCDF-4, only processor rank 0 gets the answers. */
if (format[fmt] == PIO_IOTYPE_NETCDF4C && !my_rank ||
format[fmt] == PIO_IOTYPE_NETCDF4P)
{
if (storage != NC_CHUNKED)
ERR(ERR_AWFUL);
for (int d = 0; d < NDIM; d++)
if (my_chunksize[d] != chunksize[d])
ERR(ERR_AWFUL);
}
/* Check that the inv_var_deflate functions works. */
int shuffle;
int deflate;
int deflate_level;
if ((ret = PIOc_inq_var_deflate(ncid, 0, &shuffle, &deflate, &deflate_level)))
ERR(ret);
/** For serial netCDF-4, only processor rank 0 gets the
* answers. Also deflate is turned on by default */
if (format[fmt] == PIO_IOTYPE_NETCDF4C && !my_rank)
if (shuffle || !deflate || deflate_level != 1)
ERR(ERR_AWFUL);
/* For parallel netCDF, no compression available. :-( */
if (format[fmt] == PIO_IOTYPE_NETCDF4P)
if (shuffle || deflate)
ERR(ERR_AWFUL);
} else {
/* Trying to set chunking for non-netCDF-4 files results
* in the PIO_ENOTNC4 error. */
if ((ret = PIOc_def_var_chunking(ncid, 0, NC_CHUNKED, chunksize)) != PIO_ENOTNC4)
ERR(ERR_AWFUL);
}
if ((ret = PIOc_enddef(ncid)))
ERR(ret);
/* Close the netCDF file. */
if (verbose)
printf("rank: %d Closing the sample data file...\n", my_rank);
if ((ret = PIOc_closefile(ncid)))
ERR(ret);
}
/* Free the PIO decomposition. */
if (verbose)
printf("rank: %d Freeing PIO decomposition...\n", my_rank);
if ((ret = PIOc_freedecomp(iosysid, ioid)))
ERR(ret);
/* Finalize the IO system. */
if (verbose)
printf("rank: %d Freeing PIO resources...\n", my_rank);
if ((ret = PIOc_finalize(iosysid)))
ERR(ret);
/* Finalize the MPI library. */
MPI_Finalize();
#ifdef TIMING
/* Finalize the GPTL timing library. */
if ((ret = GPTLfinalize ()))
return ret;
#endif
return 0;
}
/** Main execution of code.
Executes the functions to:
- create a new examplePioClass instance
- initialize MPI and the ParallelIO libraries
- create the decomposition for this example
- create the netCDF output file
- define the variable in the file
- write data to the variable in the file using decomposition
- read the data back from the file using decomposition
- close the file
- clean up resources
The example can be run from the command line (on system that support it) like this:
<pre>
mpiexec -n 4 ./examplePio
</pre>
The sample file created by this program is a small netCDF file. It
has the following contents (as shown by ncdump) for a 4-processor
run:
<pre>
netcdf examplePio_c {
dimensions:
x = 16 ;
variables:
int foo(x) ;
data:
foo = 42, 42, 42, 42, 43, 43, 43, 43, 44, 44, 44, 44, 45, 45, 45, 45 ;
}
</pre>
@param [in] argc argument count (should be zero)
@param [in] argv argument array (should be NULL)
@retval examplePioClass* Pointer to self.
*/
int main(int argc, char* argv[])
{
/** Set to non-zero to get output to stdout. */
int verbose = 0;
/** Zero-based rank of processor. */
int my_rank;
/** Number of processors involved in current execution. */
int ntasks;
/** Different output flavors. The example file is written (and
* then read) four times. The first two flavors,
* parallel-netcdf, and netCDF serial, both produce a netCDF
* classic format file (but with different libraries). The
* last two produce netCDF4/HDF5 format files, written with
* and without using netCDF-4 parallel I/O. */
int format[NUM_NETCDF_FLAVORS] = {PIO_IOTYPE_PNETCDF,
PIO_IOTYPE_NETCDF,
PIO_IOTYPE_NETCDF4C,
PIO_IOTYPE_NETCDF4P};
/** Names for the output files. Two of them (pnetcdf and
* classic) will be in classic netCDF format, the others
* (serial4 and parallel4) will be in netCDF-4/HDF5
* format. All four can be read by the netCDF library, and all
* will contain the same contents. */
char filename[NUM_NETCDF_FLAVORS][NC_MAX_NAME + 1] = {"example2_pnetcdf.nc",
"example2_classic.nc",
"example2_serial4.nc",
"example2_parallel4.nc"};
/** Number of processors that will do IO. In this example we
* will do IO from all processors. */
int niotasks;
/** Stride in the mpi rank between io tasks. Always 1 in this
* example. */
int ioproc_stride = 1;
/** Number of the aggregator? Always 0 in this example. */
int numAggregator = 0;
/** Zero based rank of first processor to be used for I/O. */
int ioproc_start = 0;
/** Specifies the flavor of netCDF output format. */
int iotype;
/** The dimension IDs. */
int dimids[NDIM];
/** Array index per processing unit. This is the number of
* elements of the data array that will be handled by each
* processor. In this example there are 16 data elements. If the
* example is run on 4 processors, then arrIdxPerPe will be 4. */
PIO_Offset elements_per_pe;
/** The ID for the parallel I/O system. It is set by
* PIOc_Init_Intracomm(). It references an internal structure
* containing the general IO subsystem data and MPI
* structure. It is passed to PIOc_finalize() to free
* associated resources, after all I/O, but before
* MPI_Finalize is called. */
int iosysid;
/** The ncid of the netCDF file created in this example. */
int ncid = 0;
/** The ID of the netCDF varable in the example file. */
int varid;
/** The I/O description ID as passed back by PIOc_InitDecomp()
* and freed in PIOc_freedecomp(). */
int ioid;
/** A buffer for sample data. The size of this array will
* vary depending on how many processors are involved in the
* execution of the example code. It's length will be the same
* as elements_per_pe.*/
float *buffer;
/** A buffer for reading data back from the file. The size of
* this array will vary depending on how many processors are
* involved in the execution of the example code. It's length
* will be the same as elements_per_pe.*/
int *read_buffer;
/** A 1-D array which holds the decomposition mapping for this
* example. The size of this array will vary depending on how
* many processors are involved in the execution of the
* example code. It's length will be the same as
* elements_per_pe. */
PIO_Offset *compdof;
#ifdef HAVE_MPE
/** MPE event numbers used to track start and stop of
* different parts of the program for later display with
* Jumpshot. */
int event_num[2][NUM_EVENTS];
#endif /* HAVE_MPE */
/** Needed for command line processing. */
int c;
/* Parse command line. */
while ((c = getopt(argc, argv, "v")) != -1)
switch (c)
{
case 'v':
verbose++;
break;
default:
break;
}
#ifdef TIMING
/* Initialize the GPTL timing library. */
int ret;
if ((ret = GPTLinitialize ()))
return ret;
#endif
/* Initialize MPI. */
if ((ret = MPI_Init(&argc, &argv)))
MPIERR(ret);
if ((ret = MPI_Errhandler_set(MPI_COMM_WORLD, MPI_ERRORS_RETURN)))
MPIERR(ret);
/* Learn my rank and the total number of processors. */
if ((ret = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank)))
MPIERR(ret);
if ((ret = MPI_Comm_size(MPI_COMM_WORLD, &ntasks)))
MPIERR(ret);
/* Check that a valid number of processors was specified. */
if (!(ntasks == 1 || ntasks == 2 || ntasks == 4 ||
ntasks == 8 || ntasks == 16))
fprintf(stderr, "Number of processors must be 1, 2, 4, 8, or 16!\n");
if (verbose)
printf("%d: ParallelIO Library example1 running on %d processors.\n",
my_rank, ntasks);
#ifdef HAVE_MPE
/* Initialize MPE logging. */
if ((ret = MPE_Init_log()))
ERR(ret);
if (init_logging(my_rank, event_num))
ERR(ERR_LOGGING);
/* Log with MPE that we are starting INIT. */
if ((ret = MPE_Log_event(event_num[START][INIT], 0, "start init")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* keep things simple - 1 iotask per MPI process */
niotasks = ntasks;
/* Initialize the PIO IO system. This specifies how
* many and which processors are involved in I/O. */
if ((ret = PIOc_Init_Intracomm(MPI_COMM_WORLD, niotasks, ioproc_stride,
ioproc_start, PIO_REARR_SUBSET, &iosysid)))
ERR(ret);
/* Describe the decomposition. This is a 1-based array, so add 1! */
elements_per_pe = X_DIM_LEN * Y_DIM_LEN / ntasks;
if (!(compdof = malloc(elements_per_pe * sizeof(PIO_Offset))))
return PIO_ENOMEM;
for (int i = 0; i < elements_per_pe; i++) {
compdof[i] = my_rank * elements_per_pe + i + 1;
}
/* Create the PIO decomposition for this example. */
if (verbose)
printf("rank: %d Creating decomposition...\n", my_rank);
if ((ret = PIOc_InitDecomp(iosysid, PIO_FLOAT, 2, &dim_len[1], (PIO_Offset)elements_per_pe,
compdof, &ioid, NULL, NULL, NULL)))
ERR(ret);
free(compdof);
#ifdef HAVE_MPE
/* Log with MPE that we are done with INIT. */
if ((ret = MPE_Log_event(event_num[END][INIT], 0, "end init")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Use PIO to create the example file in each of the four
* available ways. */
for (int fmt = 0; fmt < NUM_NETCDF_FLAVORS; fmt++)
{
#ifdef HAVE_MPE
/* Log with MPE that we are starting CREATE. */
if ((ret = MPE_Log_event(event_num[START][CREATE_PNETCDF+fmt], 0, "start create")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Create the netCDF output file. */
if (verbose)
printf("rank: %d Creating sample file %s with format %d...\n",
my_rank, filename[fmt], format[fmt]);
if ((ret = PIOc_createfile(iosysid, &ncid, &(format[fmt]), filename[fmt],
PIO_CLOBBER)))
ERR(ret);
/* Define netCDF dimensions and variable. */
if (verbose)
printf("rank: %d Defining netCDF metadata...\n", my_rank);
for (int d = 0; d < NDIM; d++) {
if (verbose)
printf("rank: %d Defining netCDF dimension %s, length %d\n", my_rank,
dim_name[d], dim_len[d]);
if ((ret = PIOc_def_dim(ncid, dim_name[d], (PIO_Offset)dim_len[d], &dimids[d])))
ERR(ret);
}
if ((ret = PIOc_def_var(ncid, VAR_NAME, PIO_FLOAT, NDIM, dimids, &varid)))
ERR(ret);
/* For netCDF-4 files, set the chunksize to improve performance. */
if (format[fmt] == PIO_IOTYPE_NETCDF4C || format[fmt] == PIO_IOTYPE_NETCDF4P)
if ((ret = PIOc_def_var_chunking(ncid, 0, NC_CHUNKED, chunksize)))
ERR(ret);
if ((ret = PIOc_enddef(ncid)))
ERR(ret);
#ifdef HAVE_MPE
/* Log with MPE that we are done with CREATE. */
if ((ret = MPE_Log_event(event_num[END][CREATE_PNETCDF + fmt], 0, "end create")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Allocate space for sample data. */
if (!(buffer = malloc(elements_per_pe * sizeof(float))))
return PIO_ENOMEM;
/* Write data for each timestep. */
for (int ts = 0; ts < NUM_TIMESTEPS; ts++) {
#ifdef HAVE_MPE
/* Log with MPE that we are starting CALCULATE. */
if ((ret = MPE_Log_event(event_num[START][CALCULATE], 0, "start calculate")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Calculate sample data. Add some math function calls to make this slower. */
for (int i = 0; i < elements_per_pe; i++)
if ((ret = calculate_value(my_rank, ts, &buffer[i])))
ERR(ret);
#ifdef HAVE_MPE
/* Log with MPE that we are done with CALCULATE. */
if ((ret = MPE_Log_event(event_num[END][CALCULATE], 0, "end calculate")))
MPIERR(ret);
/* Log with MPE that we are starting WRITE. */
if ((ret = MPE_Log_event(event_num[START][WRITE], 0, "start write")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Write data to the file. */
if (verbose)
printf("rank: %d Writing sample data...\n", my_rank);
if ((ret = PIOc_setframe(ncid, varid, ts)))
ERR(ret);
if ((ret = PIOc_write_darray(ncid, varid, ioid, (PIO_Offset)elements_per_pe,
buffer, NULL)))
ERR(ret);
if ((ret = PIOc_sync(ncid)))
ERR(ret);
#ifdef HAVE_MPE
/* Log with MPE that we are done with WRITE. */
if ((ret = MPE_Log_event(event_num[END][WRITE], 0, "end write")))
MPIERR(ret);
#endif /* HAVE_MPE */
}
#ifdef HAVE_MPE
/* Log with MPE that we are starting CLOSE. */
if ((ret = MPE_Log_event(event_num[START][CLOSE], 0, "start close")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Free buffer space used in this example. */
free(buffer);
/* Close the netCDF file. */
if (verbose)
printf("rank: %d Closing the sample data file...\n", my_rank);
if ((ret = PIOc_closefile(ncid)))
ERR(ret);
#ifdef HAVE_MPE
/* Log with MPE that we are done with CLOSE. */
if ((ret = MPE_Log_event(event_num[END][CLOSE], 0, "end close")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* After each file is closed, make all processors wait so that
* all start creating the next file at the same time. */
if ((ret = MPI_Barrier(MPI_COMM_WORLD)))
MPIERR(ret);
}
#ifdef HAVE_MPE
/* Log with MPE that we are starting FREE. */
if ((ret = MPE_Log_event(event_num[START][FREE], 0, "start free")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Free the PIO decomposition. */
if (verbose)
printf("rank: %d Freeing PIO decomposition...\n", my_rank);
if ((ret = PIOc_freedecomp(iosysid, ioid)))
ERR(ret);
/* Finalize the IO system. */
if (verbose)
printf("rank: %d Freeing PIO resources...\n", my_rank);
if ((ret = PIOc_finalize(iosysid)))
ERR(ret);
#ifdef HAVE_MPE
/* Log with MPE that we are done with FREE. */
if ((ret = MPE_Log_event(event_num[END][FREE], 0, "end free")))
MPIERR(ret);
/* Log with MPE that we are starting READ. */
if ((ret = MPE_Log_event(event_num[START][READ], 0, "start read")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Check the output file. */
/* if (!my_rank) */
/* for (int fmt = 0; fmt < NUM_NETCDF_FLAVORS; fmt++) */
/* if ((ret = check_file(ntasks, filename[fmt]))) */
/* ERR(ret); */
#ifdef HAVE_MPE
/* Log with MPE that we are done with READ. */
if ((ret = MPE_Log_event(event_num[END][READ], 0, "end read")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* Finalize the MPI library. */
MPI_Finalize();
#ifdef TIMING
/* Finalize the GPTL timing library. */
if ((ret = GPTLfinalize ()))
return ret;
#endif
if (verbose)
printf("rank: %d SUCCESS!\n", my_rank);
return 0;
}
/* Fortran wrapper functions start here */
int gptlinitialize (void)
{
return GPTLinitialize ();
}
/** @brief Main execution of code.
Executes the functions to:
- create a new examplePioClass instance
- initialize MPI and the ParallelIO libraries
- create the decomposition for this example
- create the netCDF output file
- define the variable in the file
- write data to the variable in the file using decomposition
- read the data back from the file using decomposition
- close the file
- clean up resources
The example can be run from the command line (on system that support it) like this:
<pre>
mpiexec -n 4 ./examplePio
</pre>
The sample file created by this program is a small netCDF file. It
has the following contents (as shown by ncdump) for a 4-processor
run:
<pre>
netcdf examplePio_c {
dimensions:
x = 16 ;
variables:
int foo(x) ;
data:
foo = 42, 42, 42, 42, 43, 43, 43, 43, 44, 44, 44, 44, 45, 45, 45, 45 ;
}
</pre>
@param [in] argc argument count (should be zero)
@param [in] argv argument array (should be NULL)
@retval examplePioClass* Pointer to self.
*/
int main(int argc, char* argv[])
{
/** Set to non-zero to get output to stdout. */
int verbose = 0;
/** Zero-based rank of processor. */
int my_rank;
/** Number of processors involved in current execution. */
int ntasks;
/** Different output flavors. The example file is written (and
* then read) four times. The first two flavors,
* parallel-netcdf, and netCDF serial, both produce a netCDF
* classic format file (but with different libraries). The
* last two produce netCDF4/HDF5 format files, written with
* and without using netCDF-4 parallel I/O. */
int format[NUM_NETCDF_FLAVORS];
/** Number of processors that will do IO. In this example we
* will do IO from all processors. */
int niotasks;
/** Stride in the mpi rank between io tasks. Always 1 in this
* example. */
int ioproc_stride = 1;
/** Zero based rank of first processor to be used for I/O. */
int ioproc_start = 0;
/** The dimension ID. */
int dimid;
/** Array index per processing unit. This is the number of
* elements of the data array that will be handled by each
* processor. In this example there are 16 data elements. If the
* example is run on 4 processors, then arrIdxPerPe will be 4. */
PIO_Offset elements_per_pe;
/* Length of the dimensions in the data. This simple example
* uses one-dimensional data. The lenght along that dimension
* is DIM_LEN (16). */
int dim_len[1] = {DIM_LEN};
/** The ID for the parallel I/O system. It is set by
* PIOc_Init_Intracomm(). It references an internal structure
* containing the general IO subsystem data and MPI
* structure. It is passed to PIOc_finalize() to free
* associated resources, after all I/O, but before
* MPI_Finalize is called. */
int iosysid;
/** The ncid of the netCDF file created in this example. */
int ncid;
/** The ID of the netCDF varable in the example file. */
int varid;
/** The I/O description ID as passed back by PIOc_InitDecomp()
* and freed in PIOc_freedecomp(). */
int ioid;
/** A buffer for sample data. The size of this array will
* vary depending on how many processors are involved in the
* execution of the example code. It's length will be the same
* as elements_per_pe.*/
int *buffer;
/** A 1-D array which holds the decomposition mapping for this
* example. The size of this array will vary depending on how
* many processors are involved in the execution of the
* example code. It's length will be the same as
* elements_per_pe. */
PIO_Offset *compdof;
/** Test filename. */
char filename[NC_MAX_NAME + 1];
/** The number of netCDF flavors available in this build. */
int num_flavors = 0;
/** Used for command line processing. */
int c;
/** Return value. */
int ret;
/* Parse command line. */
while ((c = getopt(argc, argv, "v")) != -1)
switch (c)
{
case 'v':
verbose++;
break;
default:
break;
}
#ifdef TIMING
/* Initialize the GPTL timing library. */
if ((ret = GPTLinitialize ()))
return ret;
#endif
/* Initialize MPI. */
if ((ret = MPI_Init(&argc, &argv)))
MPIERR(ret);
if ((ret = MPI_Comm_set_errhandler(MPI_COMM_WORLD, MPI_ERRORS_RETURN)))
MPIERR(ret);
/* Learn my rank and the total number of processors. */
if ((ret = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank)))
MPIERR(ret);
if ((ret = MPI_Comm_size(MPI_COMM_WORLD, &ntasks)))
MPIERR(ret);
/* Check that a valid number of processors was specified. */
if (!(ntasks == 1 || ntasks == 2 || ntasks == 4 ||
ntasks == 8 || ntasks == 16))
fprintf(stderr, "Number of processors must be 1, 2, 4, 8, or 16!\n");
if (verbose)
printf("%d: ParallelIO Library example1 running on %d processors.\n",
my_rank, ntasks);
/* keep things simple - 1 iotask per MPI process */
niotasks = ntasks;
/* Turn on logging if available. */
/* PIOc_set_log_level(4); */
/* Change error handling to return errors. */
if ((ret = PIOc_set_iosystem_error_handling(PIO_DEFAULT, PIO_RETURN_ERROR, NULL)))
return ret;
/* Initialize the PIO IO system. This specifies how
* many and which processors are involved in I/O. */
if ((ret = PIOc_Init_Intracomm(MPI_COMM_WORLD, niotasks, ioproc_stride,
ioproc_start, PIO_REARR_SUBSET, &iosysid)))
ERR(ret);
/* Describe the decomposition. This is a 1-based array, so add 1! */
elements_per_pe = DIM_LEN / ntasks;
if (!(compdof = malloc(elements_per_pe * sizeof(PIO_Offset))))
return PIO_ENOMEM;
for (int i = 0; i < elements_per_pe; i++)
compdof[i] = my_rank * elements_per_pe + i + 1;
/* Create the PIO decomposition for this example. */
if (verbose)
printf("rank: %d Creating decomposition...\n", my_rank);
if ((ret = PIOc_InitDecomp(iosysid, PIO_INT, NDIM, dim_len, (PIO_Offset)elements_per_pe,
compdof, &ioid, NULL, NULL, NULL)))
ERR(ret);
free(compdof);
/* The number of favors may change with the build parameters. */
#ifdef _PNETCDF
format[num_flavors++] = PIO_IOTYPE_PNETCDF;
#endif
format[num_flavors++] = PIO_IOTYPE_NETCDF;
#ifdef _NETCDF4
format[num_flavors++] = PIO_IOTYPE_NETCDF4C;
format[num_flavors++] = PIO_IOTYPE_NETCDF4P;
#endif
/* Use PIO to create the example file in each of the four
* available ways. */
for (int fmt = 0; fmt < num_flavors; fmt++)
{
/* Create a filename. */
sprintf(filename, "example1_%d.nc", fmt);
/* Create the netCDF output file. */
if (verbose)
printf("rank: %d Creating sample file %s with format %d...\n",
my_rank, filename, format[fmt]);
if ((ret = PIOc_createfile(iosysid, &ncid, &(format[fmt]), filename,
PIO_CLOBBER)))
ERR(ret);
/* Define netCDF dimension and variable. */
if (verbose)
printf("rank: %d Defining netCDF metadata...\n", my_rank);
if ((ret = PIOc_def_dim(ncid, DIM_NAME, (PIO_Offset)dim_len[0], &dimid)))
ERR(ret);
if ((ret = PIOc_def_var(ncid, VAR_NAME, PIO_INT, NDIM, &dimid, &varid)))
ERR(ret);
if ((ret = PIOc_enddef(ncid)))
ERR(ret);
/* Prepare sample data. */
if (!(buffer = malloc(elements_per_pe * sizeof(int))))
return PIO_ENOMEM;
for (int i = 0; i < elements_per_pe; i++)
buffer[i] = START_DATA_VAL + my_rank;
/* Write data to the file. */
if (verbose)
printf("rank: %d Writing sample data...\n", my_rank);
if ((ret = PIOc_write_darray(ncid, varid, ioid, (PIO_Offset)elements_per_pe,
buffer, NULL)))
ERR(ret);
if ((ret = PIOc_sync(ncid)))
ERR(ret);
/* Free buffer space used in this example. */
free(buffer);
/* Close the netCDF file. */
if (verbose)
printf("rank: %d Closing the sample data file...\n", my_rank);
if ((ret = PIOc_closefile(ncid)))
ERR(ret);
}
/* Free the PIO decomposition. */
if (verbose)
printf("rank: %d Freeing PIO decomposition...\n", my_rank);
if ((ret = PIOc_freedecomp(iosysid, ioid)))
ERR(ret);
/* Finalize the IO system. */
if (verbose)
printf("rank: %d Freeing PIO resources...\n", my_rank);
if ((ret = PIOc_finalize(iosysid)))
ERR(ret);
/* Check the output file. */
if (!my_rank)
for (int fmt = 0; fmt < num_flavors; fmt++)
{
sprintf(filename, "example1_%d.nc", fmt);
if ((ret = check_file(ntasks, filename)))
ERR(ret);
}
/* Finalize the MPI library. */
MPI_Finalize();
#ifdef TIMING
/* Finalize the GPTL timing library. */
if ((ret = GPTLfinalize ()))
return ret;
#endif
if (verbose)
printf("rank: %d SUCCESS!\n", my_rank);
return 0;
}
int main (int argc, char **argv)
{
char pname[MPI_MAX_PROCESSOR_NAME];
int iter;
int counter;
int c;
int tnum = 0;
int resultlen;
int ret;
double value;
extern char *optarg;
while ((c = getopt (argc, argv, "p:")) != -1) {
switch (c) {
case 'p':
if ((ret = GPTLevent_name_to_code (optarg, &counter)) != 0) {
printf ("Failure from GPTLevent_name_to_code\n");
return 1;
}
if (GPTLsetoption (counter, 1) < 0) {
printf ("Failure from GPTLsetoption (%s,1)\n", optarg);
return 1;
}
break;
default:
printf ("unknown option %c\n", c);
printf ("Usage: %s [-p option_name]\n", argv[0]);
return 2;
}
}
ret = GPTLsetoption (GPTLabort_on_error, 1);
ret = GPTLsetoption (GPTLoverhead, 1);
ret = GPTLsetoption (GPTLnarrowprint, 1);
if (MPI_Init (&argc, &argv) != MPI_SUCCESS) {
printf ("Failure from MPI_Init\n");
return 1;
}
ret = GPTLinitialize ();
ret = GPTLstart ("total");
ret = MPI_Comm_rank (MPI_COMM_WORLD, &iam);
ret = MPI_Comm_size (MPI_COMM_WORLD, &nproc);
ret = MPI_Get_processor_name (pname, &resultlen);
printf ("Rank %d is running on processor %s\n", iam, pname);
#ifdef THREADED_OMP
nthreads = omp_get_max_threads ();
#pragma omp parallel for private (iter, ret, tnum)
#endif
for (iter = 1; iter <= nthreads; iter++) {
#ifdef THREADED_OMP
tnum = omp_get_thread_num ();
#endif
printf ("Thread %d of rank %d on processor %s\n", tnum, iam, pname);
value = sub (iter);
}
ret = GPTLstop ("total");
ret = GPTLpr (iam);
if (iam == 0) {
printf ("summary: testing GPTLpr_summary...\n");
printf ("Number of threads was %d\n", nthreads);
printf ("Number of tasks was %d\n", nproc);
}
// NOTE: if ENABLE_PMPI is set, 2nd pr call below will show some extra send/recv calls
// due to MPI calls from within GPTLpr_summary_file
if (GPTLpr_summary (MPI_COMM_WORLD) != 0)
return 1;
if (GPTLpr_summary_file (MPI_COMM_WORLD, "timing.summary.duplicate") != 0)
return 1;
ret = MPI_Finalize ();
if (GPTLfinalize () != 0)
return 1;
return 0;
}
/** @brief Main execution of code.
Executes the functions to:
- create a new examplePioClass instance
- initialize MPI and the ParallelIO libraries
- create the decomposition for this example
- create the netCDF output file
- define the variable in the file
- write data to the variable in the file using decomposition
- read the data back from the file using decomposition
- close the file
- clean up resources
The example can be run from the command line (on system that support it) like this:
<pre>
mpiexec -n 4 ./examplePio
</pre>
The sample file created by this program is a small netCDF file. It
has the following contents (as shown by ncdump) for a 4-processor
run:
<pre>
netcdf examplePio_c {
dimensions:
x = 16 ;
variables:
int foo(x) ;
data:
foo = 42, 42, 42, 42, 43, 43, 43, 43, 44, 44, 44, 44, 45, 45, 45, 45 ;
}
</pre>
@param [in] argc argument count (should be zero)
@param [in] argv argument array (should be NULL)
@retval examplePioClass* Pointer to self.
*/
int main(int argc, char* argv[])
{
/** Set to non-zero to get output to stdout. */
int verbose = 0;
/** Zero-based rank of processor. */
int my_rank;
/** Number of processors involved in current execution. */
int ntasks;
/** Number of processors that will do IO. In this example we
* will do IO from all processors. */
int niotasks;
/** Stride in the mpi rank between io tasks. Always 1 in this
* example. */
int stride = 1;
/** Number of the aggregator? Always 0 in this example. */
int numAggregator = 0;
/** */
int optBase = 1;
/** Specifies the flavor of netCDF output format. */
int iotype;
/** The dimension ID. */
int pioDimId;
/** */
PIO_Offset ista;
/** */
PIO_Offset isto;
/** Array index per processing unit. This is the number of
* elements of the data array that will be handled by each
* processor. In this example there are 16 data elements. If the
* example is run on 4 processors, then arrIdxPerPe will be 4. */
PIO_Offset arrIdxPerPe;
/* Length of the dimension in data. */
int dimLen[1];
/** The ID for the parallel I/O system. It is set by
* PIOc_Init_Intracomm(). It references an internal structure
* containing the general IO subsystem data and MPI structure. */
int pio_io_system;
/** The ncid of the netCDF file created in this example. */
int pioFileDesc;
/** The ID of the netCDF varable in the example file. */
int pioVar;
/** The I/O description ID as passed back by PIOc_InitDecomp(). */
int iodescNCells;
/** A buffer for sample data. */
int *dataBuffer;
/** A buffer for reading data back from the file. */
int *readBuffer;
/** A 1-D array which holds the decomposition mapping for this
* example. */
PIO_Offset *compdof;
/** The example file name. */
char file_name[] = EXAMPLE_FILENAME;
/** Used for command line processing. */
int c;
/* Parse command line. */
while ((c = getopt(argc, argv, "v")) != -1)
switch (c)
{
case 'v':
verbose++;
break;
default:
break;
}
#ifdef TIMING
/* Initialize the GPTL timing library. */
int ret;
if ((ret = GPTLinitialize ()))
return ret;
#endif
/* Initialize MPI. */
if ((ret = MPI_Init(&argc, &argv)))
MPIERR(ret);
if ((ret = MPI_Errhandler_set(MPI_COMM_WORLD, MPI_ERRORS_RETURN)))
MPIERR(ret);
/* Learn my rank and the total number of processors. */
if ((ret = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank)))
MPIERR(ret);
if ((ret = MPI_Comm_size(MPI_COMM_WORLD, &ntasks)))
MPIERR(ret);
/* Check that a valid number of processors was specified. */
if (!(ntasks == 1 || ntasks == 2 || ntasks == 4 ||
ntasks == 8 || ntasks == 16))
fprintf(stderr, "Number of processors must be 1, 2, 4, 8, or 16!\n");
if (verbose)
printf("%d: ParallelIO Library example1 running on %d processors.\n",
my_rank, ntasks);
/* Initialize the ParallelIO library IO system. */
iotype = PIO_IOTYPE_NETCDF;
/* keep things simple - 1 iotask per MPI process */
niotasks = ntasks;
/* Initialize the IO system. */
if ((ret = PIOc_Init_Intracomm(MPI_COMM_WORLD, niotasks, stride, optBase,
PIO_REARR_SUBSET, &pio_io_system)))
ERR(ret);
/* Finalize the IO system. */
if ((ret = PIOc_finalize(pio_io_system)))
ERR(ret);
/* Finalize the MPI library. */
MPI_Finalize();
#ifdef TIMING
/* Finalize the GPTL timing library. */
if ((ret = GPTLfinalize ()))
return ret;
#endif
return 0;
}
int main (int argc, char **argv)
{
int ret;
int i, code;
long long pc[1]; /* papi counters */
double sum;
printf ("testpapi: Testing PAPI interface...\n");
printf ("%s: testing getting event code for PAPI_TOT_CYC...\n", argv[0]);
if ((ret = GPTLevent_name_to_code ("PAPI_TOT_CYC", &code)) != 0) {
printf ("Failure\n");
return 2;
}
printf ("Success\n");
printf ("%s: testing GPTLsetoption(PAPI_TOT_CYC,1)...\n", argv[0]);
if (GPTLsetoption (code, 1) != 0) {
printf ("Failure\n");
return 3;
}
printf ("Success\n");
printf ("%s: testing GPTLinitialize\n", argv[0]);
if ((ret = GPTLinitialize ()) != 0) {
printf ("Failure\n");
return 3;
}
printf ("Success\n");
printf ("%s: testing GPTLstart\n", argv[0]);
if ((ret = GPTLstart ("sum")) != 0) {
printf ("Unexpected failure from GPTLstart(\"sum\")\n");
return 3;
}
printf ("Success\n");
sum = 0.;
for (i = 0; i < 1000000; ++i)
sum += (double) i;
printf ("%s: testing GPTLstop\n", argv[0]);
if ((ret = GPTLstop ("sum")) != 0) {
printf ("Unexpected failure from GPTLstop(\"sum\")\n");
return 3;
}
printf ("Success\n");
printf ("%s: testing GPTLquerycounters...\n", argv[0]);
if (GPTLquerycounters ("sum", 0, pc) != 0) {
printf ("Failure\n");
return 4;
}
printf ("sum=%g\n",sum);
printf ("%s: testing reasonableness of PAPI counters...\n", argv[0]);
if (pc[0] < 1 || pc[0] > 1.e8) {
printf ("Suspicious PAPI_TOT_CYC value=%lld for 1e6 additions\n", pc[0]);
return 5;
} else {
printf ("Success\n");
}
printf ("%s: all tests successful\n", argv[0]);
return 0;
}
/** Run Tests for NetCDF-4 Functions.
*
* @param argc argument count
* @param argv array of arguments
*/
int
main(int argc, char **argv)
{
int verbose = 1;
/** Zero-based rank of processor. */
int my_rank;
/** Number of processors involved in current execution. */
int ntasks;
/** Specifies the flavor of netCDF output format. */
int iotype;
/** Different output flavors. */
int format[NUM_NETCDF_FLAVORS] = {PIO_IOTYPE_PNETCDF,
PIO_IOTYPE_NETCDF,
PIO_IOTYPE_NETCDF4C,
PIO_IOTYPE_NETCDF4P};
/** Names for the output files. */
char filename[NUM_NETCDF_FLAVORS][NC_MAX_NAME + 1] = {"test_nc4_pnetcdf.nc",
"test_nc4_classic.nc",
"test_nc4_serial4.nc",
"test_nc4_parallel4.nc"};
/** Number of processors that will do IO. In this test we
* will do IO from all processors. */
int niotasks;
/** Stride in the mpi rank between io tasks. Always 1 in this
* test. */
int ioproc_stride = 1;
/** Number of the aggregator? Always 0 in this test. */
int numAggregator = 0;
/** Zero based rank of first processor to be used for I/O. */
int ioproc_start = 0;
/** The dimension IDs. */
int dimids[NDIM];
/** Array index per processing unit. */
PIO_Offset elements_per_pe;
/** The ID for the parallel I/O system. */
int iosysid;
/** The ncid of the netCDF file. */
int ncid = 0;
/** The ID of the netCDF varable. */
int varid;
/** Storage of netCDF-4 files (contiguous vs. chunked). */
int storage;
/** Chunksizes set in the file. */
PIO_Offset my_chunksize[NDIM];
/** The shuffle filter setting in the netCDF-4 test file. */
int shuffle;
/** Non-zero if deflate set for the variable in the netCDF-4 test file. */
int deflate;
/** The deflate level set for the variable in the netCDF-4 test file. */
int deflate_level;
/** Endianness of variable. */
int endianness;
/* Size of the var chunk cache. */
PIO_Offset var_cache_size;
/* Number of elements in var cache. */
PIO_Offset var_cache_nelems;
/* Var cache preemption. */
float var_cache_preemption;
/** The I/O description ID. */
int ioid;
/** A buffer for sample data. */
float *buffer;
/** A buffer for reading data back from the file. */
int *read_buffer;
/** The decomposition mapping. */
PIO_Offset *compdof;
/** Return code. */
int ret;
/** Index for loops. */
int fmt, d, d1, i;
/** For setting the chunk cache. */
PIO_Offset chunk_cache_size = 1024*1024;
PIO_Offset chunk_cache_nelems = 1024;
float chunk_cache_preemption = 0.5;
/* For reading the chunk cache. */
PIO_Offset chunk_cache_size_in;
PIO_Offset chunk_cache_nelems_in;
float chunk_cache_preemption_in;
char varname[15];
#ifdef TIMING
/* Initialize the GPTL timing library. */
if ((ret = GPTLinitialize ()))
return ret;
#endif
/* Initialize MPI. */
if ((ret = MPI_Init(&argc, &argv)))
MPIERR(ret);
/* Learn my rank and the total number of processors. */
if ((ret = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank)))
MPIERR(ret);
if ((ret = MPI_Comm_size(MPI_COMM_WORLD, &ntasks)))
MPIERR(ret);
/* Check that a valid number of processors was specified. */
if (!(ntasks == 1 || ntasks == 2 || ntasks == 4 ||
ntasks == 8 || ntasks == 16))
fprintf(stderr, "Number of processors must be 1, 2, 4, 8, or 16!\n");
if (verbose)
printf("%d: ParallelIO Library test_nc4 running on %d processors.\n",
my_rank, ntasks);
/* keep things simple - 1 iotask per MPI process */
niotasks = ntasks;
/* Initialize the PIO IO system. This specifies how
* many and which processors are involved in I/O. */
if ((ret = PIOc_Init_Intracomm(MPI_COMM_WORLD, niotasks, ioproc_stride,
ioproc_start, PIO_REARR_SUBSET, &iosysid)))
ERR(ret);
/* Describe the decomposition. This is a 1-based array, so add 1! */
elements_per_pe = X_DIM_LEN * Y_DIM_LEN / ntasks;
if (!(compdof = malloc(elements_per_pe * sizeof(PIO_Offset))))
return PIO_ENOMEM;
for (i = 0; i < elements_per_pe; i++) {
compdof[i] = my_rank * elements_per_pe + i + 1;
}
/* Create the PIO decomposition for this test. */
if (verbose)
printf("rank: %d Creating decomposition...\n", my_rank);
if ((ret = PIOc_InitDecomp(iosysid, PIO_FLOAT, 2, &dim_len[1], (PIO_Offset)elements_per_pe,
compdof, &ioid, NULL, NULL, NULL)))
ERR(ret);
free(compdof);
#ifdef HAVE_MPE
/* Log with MPE that we are done with INIT. */
if ((ret = MPE_Log_event(event_num[END][INIT], 0, "end init")))
MPIERR(ret);
#endif /* HAVE_MPE */
/* How many flavors will we be running for? */
int num_flavors = 0;
int fmtidx = 0;
#ifdef _PNETCDF
num_flavors++;
format[fmtidx++] = PIO_IOTYPE_PNETCDF;
#endif
#ifdef _NETCDF
num_flavors++;
format[fmtidx++] = PIO_IOTYPE_NETCDF;
#endif
#ifdef _NETCDF4
num_flavors += 2;
format[fmtidx++] = PIO_IOTYPE_NETCDF4C;
format[fmtidx] = PIO_IOTYPE_NETCDF4P;
#endif
/* Use PIO to create the example file in each of the four
* available ways. */
for (fmt = 0; fmt < num_flavors; fmt++)
{
#ifdef HAVE_MPE
/* Log with MPE that we are starting CREATE. */
if ((ret = MPE_Log_event(event_num[START][CREATE_PNETCDF+fmt], 0, "start create")))
MPIERR(ret);
#endif /* HAVE_MPE */
if (verbose)
printf("rank: %d Setting chunk cache for file %s with format %d...\n",
my_rank, filename[fmt], format[fmt]);
/* Try to set the chunk cache with invalid preemption to check error handling. */
chunk_cache_preemption = 50.0;
ret = PIOc_set_chunk_cache(iosysid, format[fmt], chunk_cache_size,
chunk_cache_nelems, chunk_cache_preemption);
if (format[fmt] == PIO_IOTYPE_NETCDF4C || format[fmt] == PIO_IOTYPE_NETCDF4P)
{
if (ret != NC_EINVAL)
ERR(ERR_AWFUL);
}
else
{
if (ret != NC_ENOTNC4)
ERR(ERR_AWFUL);
}
/* Try to set the chunk cache. */
chunk_cache_preemption = 0.5;
ret = PIOc_set_chunk_cache(iosysid, format[fmt], chunk_cache_size,
chunk_cache_nelems, chunk_cache_preemption);
/* Should only have worked for netCDF-4 iotypes. */
if (format[fmt] == PIO_IOTYPE_NETCDF4C || format[fmt] == PIO_IOTYPE_NETCDF4P)
{
if (ret != PIO_NOERR)
ERR(ret);
}
else
{
if (ret != PIO_ENOTNC4)
ERR(ERR_AWFUL);
}
/* Now check the chunk cache. */
ret = PIOc_get_chunk_cache(iosysid, format[fmt], &chunk_cache_size_in,
&chunk_cache_nelems_in, &chunk_cache_preemption_in);
/* Should only have worked for netCDF-4 iotypes. */
if (format[fmt] == PIO_IOTYPE_NETCDF4C || format[fmt] == PIO_IOTYPE_NETCDF4P)
{
/* Check that there was no error. */
if (ret != PIO_NOERR)
ERR(ret);
/* Check that we got the correct values. */
if (chunk_cache_size_in != chunk_cache_size || chunk_cache_nelems_in != chunk_cache_nelems ||
chunk_cache_preemption_in != chunk_cache_preemption)
ERR(ERR_AWFUL);
}
else
{
if (ret != PIO_ENOTNC4)
ERR(ERR_AWFUL);
}
/* Create the netCDF output file. */
if (verbose)
printf("rank: %d Creating sample file %s with format %d...\n",
my_rank, filename[fmt], format[fmt]);
if ((ret = PIOc_createfile(iosysid, &ncid, &(format[fmt]), filename[fmt],
PIO_CLOBBER)))
ERR(ret);
/* Set error handling. */
PIOc_Set_File_Error_Handling(ncid, PIO_BCAST_ERROR);
/* Define netCDF dimensions and variable. */
if (verbose)
printf("rank: %d Defining netCDF metadata...\n", my_rank);
for (d = 0; d < NDIM; d++) {
if (verbose)
printf("rank: %d Defining netCDF dimension %s, length %d\n", my_rank,
dim_name[d], dim_len[d]);
if ((ret = PIOc_def_dim(ncid, dim_name[d], (PIO_Offset)dim_len[d], &dimids[d])))
ERR(ret);
}
if (verbose)
printf("rank: %d Defining netCDF variable %s, ndims %d\n", my_rank, VAR_NAME, NDIM);
if ((ret = PIOc_def_var(ncid, VAR_NAME, PIO_FLOAT, NDIM, dimids, &varid)))
ERR(ret);
/* For netCDF-4 files, set the chunksize to improve performance. */
if (format[fmt] == PIO_IOTYPE_NETCDF4C || format[fmt] == PIO_IOTYPE_NETCDF4P)
{
if (verbose)
printf("rank: %d Defining chunksizes\n", my_rank);
if ((ret = PIOc_def_var_chunking(ncid, 0, NC_CHUNKED, chunksize)))
ERR(ret);
/** Check that the inq_varname function works. */
if (verbose)
printf("rank: %d Checking varname\n", my_rank);
ret = PIOc_inq_varname(ncid, 0, varname);
printf("rank: %d ret: %d varname: %s\n", my_rank, ret, varname);
/** Check that the inq_var_chunking function works. */
if (verbose)
printf("rank: %d Checking chunksizes\n");
if ((ret = PIOc_inq_var_chunking(ncid, 0, &storage, my_chunksize)))
ERR(ret);
if (verbose)
{
printf("rank: %d ret: %d storage: %d\n", my_rank, ret, storage);
for (d1 = 0; d1 < NDIM; d1++)
{
printf("chunksize[%d]=%d\n", d1, my_chunksize[d1]);
}
}
/** Check the answers. */
if (format[fmt] == PIO_IOTYPE_NETCDF4C ||
format[fmt] == PIO_IOTYPE_NETCDF4P)
{
if (storage != NC_CHUNKED)
ERR(ERR_AWFUL);
for (d1 = 0; d1 < NDIM; d1++)
if (my_chunksize[d1] != chunksize[d1])
ERR(ERR_AWFUL);
}
/* Check that the inq_var_deflate functions works. */
if ((ret = PIOc_inq_var_deflate(ncid, 0, &shuffle, &deflate, &deflate_level)))
ERR(ret);
/** For serial netCDF-4 deflate is turned on by default */
if (format[fmt] == PIO_IOTYPE_NETCDF4C)
if (shuffle || !deflate || deflate_level != 1)
ERR(ERR_AWFUL);
/* For parallel netCDF-4, no compression available. :-( */
if (format[fmt] == PIO_IOTYPE_NETCDF4P)
if (shuffle || deflate)
ERR(ERR_AWFUL);
/* Check setting the chunk cache for the variable. */
printf("rank: %d PIOc_set_var_chunk_cache...\n", my_rank);
if ((ret = PIOc_set_var_chunk_cache(ncid, 0, VAR_CACHE_SIZE, VAR_CACHE_NELEMS,
VAR_CACHE_PREEMPTION)))
ERR(ret);
/* Check getting the chunk cache values for the variable. */
printf("rank: %d PIOc_get_var_chunk_cache...\n", my_rank);
if ((ret = PIOc_get_var_chunk_cache(ncid, 0, &var_cache_size, &var_cache_nelems,
&var_cache_preemption)))
ERR(ret);
PIO_Offset len;
if ((ret = PIOc_inq_dimlen(ncid, 0, &len)))
ERR(ret);
/* Check that we got expected values. */
printf("rank: %d var_cache_size = %d\n", my_rank, var_cache_size);
if (var_cache_size != VAR_CACHE_SIZE)
ERR(ERR_AWFUL);
if (var_cache_nelems != VAR_CACHE_NELEMS)
ERR(ERR_AWFUL);
if (var_cache_preemption != VAR_CACHE_PREEMPTION)
ERR(ERR_AWFUL);
} else {
/* Trying to set or inq netCDF-4 settings for non-netCDF-4
* files results in the PIO_ENOTNC4 error. */
if ((ret = PIOc_def_var_chunking(ncid, 0, NC_CHUNKED, chunksize)) != PIO_ENOTNC4)
ERR(ERR_AWFUL);
if ((ret = PIOc_inq_var_chunking(ncid, 0, &storage, my_chunksize)) != PIO_ENOTNC4)
ERR(ERR_AWFUL);
if ((ret = PIOc_inq_var_deflate(ncid, 0, &shuffle, &deflate, &deflate_level))
!= PIO_ENOTNC4)
ERR(ret);
if ((ret = PIOc_def_var_endian(ncid, 0, 1)) != PIO_ENOTNC4)
ERR(ret);
if ((ret = PIOc_inq_var_endian(ncid, 0, &endianness)) != PIO_ENOTNC4)
ERR(ret);
if ((ret = PIOc_set_var_chunk_cache(ncid, 0, VAR_CACHE_SIZE, VAR_CACHE_NELEMS,
VAR_CACHE_PREEMPTION)) != PIO_ENOTNC4)
ERR(ret);
if ((ret = PIOc_get_var_chunk_cache(ncid, 0, &var_cache_size, &var_cache_nelems,
&var_cache_preemption)) != PIO_ENOTNC4)
ERR(ret);
if ((ret = PIOc_set_chunk_cache(iosysid, format[fmt], chunk_cache_size, chunk_cache_nelems,
chunk_cache_preemption)) != PIO_ENOTNC4)
ERR(ret);
if ((ret = PIOc_get_chunk_cache(iosysid, format[fmt], &chunk_cache_size,
&chunk_cache_nelems, &chunk_cache_preemption)) != PIO_ENOTNC4)
ERR(ret);
}
if ((ret = PIOc_enddef(ncid)))
ERR(ret);
/* Close the netCDF file. */
if (verbose)
printf("rank: %d Closing the sample data file...\n", my_rank);
if ((ret = PIOc_closefile(ncid)))
ERR(ret);
}
/* Free the PIO decomposition. */
if (verbose)
printf("rank: %d Freeing PIO decomposition...\n", my_rank);
if ((ret = PIOc_freedecomp(iosysid, ioid)))
ERR(ret);
/* Finalize the IO system. */
if (verbose)
printf("rank: %d Freeing PIO resources...\n", my_rank);
if ((ret = PIOc_finalize(iosysid)))
ERR(ret);
/* Finalize the MPI library. */
MPI_Finalize();
#ifdef TIMING
/* Finalize the GPTL timing library. */
if ((ret = GPTLfinalize ()))
return ret;
#endif
return 0;
}
/* Write, then read, a simple example with darrays.
The sample file created by this program is a small netCDF file. It
has the following contents (as shown by ncdump):
<pre>
netcdf darray_no_async_iotype_1 {
dimensions:
unlimted = UNLIMITED ; // (2 currently)
x = 4 ;
y = 4 ;
variables:
int foo(unlimted, x, y) ;
data:
foo =
42, 42, 42, 42,
43, 43, 43, 43,
44, 44, 44, 44,
45, 45, 45, 45,
142, 142, 142, 142,
143, 143, 143, 143,
144, 144, 144, 144,
145, 145, 145, 145 ;
}
</pre>
*/
int main(int argc, char* argv[])
{
int my_rank; /* Zero-based rank of processor. */
int ntasks; /* Number of processors involved in current execution. */
int ioproc_stride = 1; /* Stride in the mpi rank between io tasks. */
int ioproc_start = 0; /* Rank of first task to be used for I/O. */
PIO_Offset elements_per_pe; /* Array elements per processing unit. */
int iosysid; /* The ID for the parallel I/O system. */
int ncid; /* The ncid of the netCDF file. */
int dimid[NDIM3]; /* The dimension ID. */
int varid; /* The ID of the netCDF varable. */
int ioid; /* The I/O description ID. */
char filename[NC_MAX_NAME + 1]; /* Test filename. */
int num_flavors = 0; /* Number of iotypes available in this build. */
int format[NUM_NETCDF_FLAVORS]; /* Different output flavors. */
int ret; /* Return value. */
#ifdef TIMING
/* Initialize the GPTL timing library. */
if ((ret = GPTLinitialize ()))
return ret;
#endif
/* Initialize MPI. */
if ((ret = MPI_Init(&argc, &argv)))
MPIERR(ret);
if ((ret = MPI_Comm_set_errhandler(MPI_COMM_WORLD, MPI_ERRORS_RETURN)))
MPIERR(ret);
/* Learn my rank and the total number of processors. */
if ((ret = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank)))
MPIERR(ret);
if ((ret = MPI_Comm_size(MPI_COMM_WORLD, &ntasks)))
MPIERR(ret);
/* Check that a valid number of processors was specified. */
if (ntasks != TARGET_NTASKS)
fprintf(stderr, "Number of processors must be 4!\n");
printf("%d: ParallelIO Library darray_no_async example running on %d processors.\n",
my_rank, ntasks);
/* Turn on logging. */
if ((ret = PIOc_set_log_level(LOG_LEVEL)))
return ret;
/* Initialize the PIO IO system. This specifies how many and
* which processors are involved in I/O. */
if ((ret = PIOc_Init_Intracomm(MPI_COMM_WORLD, 1, ioproc_stride,
ioproc_start, PIO_REARR_BOX, &iosysid)))
ERR(ret);
/* Describe the decomposition. */
elements_per_pe = DIM_LEN_X * DIM_LEN_Y / TARGET_NTASKS;
/* Allocate and initialize array of decomposition mapping. */
PIO_Offset compdof[elements_per_pe];
for (int i = 0; i < elements_per_pe; i++)
compdof[i] = my_rank * elements_per_pe + i;
/* Create the PIO decomposition for this example. Since this
* is a variable with an unlimited dimension, we want to
* create a 2-D composition which represents one record. */
printf("rank: %d Creating decomposition...\n", my_rank);
if ((ret = PIOc_init_decomp(iosysid, PIO_INT, NDIM3 - 1, &dim_len[1], elements_per_pe,
compdof, &ioid, 0, NULL, NULL)))
ERR(ret);
/* The number of favors may change with the build parameters. */
#ifdef _PNETCDF
format[num_flavors++] = PIO_IOTYPE_PNETCDF;
#endif
format[num_flavors++] = PIO_IOTYPE_NETCDF;
#ifdef _NETCDF4
format[num_flavors++] = PIO_IOTYPE_NETCDF4C;
format[num_flavors++] = PIO_IOTYPE_NETCDF4P;
#endif
/* Use PIO to create the example file in each of the four
* available ways. */
for (int fmt = 0; fmt < num_flavors; fmt++)
{
/* Create a filename. */
sprintf(filename, "darray_no_async_iotype_%d.nc", format[fmt]);
/* Create the netCDF output file. */
printf("rank: %d Creating sample file %s with format %d...\n",
my_rank, filename, format[fmt]);
if ((ret = PIOc_createfile(iosysid, &ncid, &(format[fmt]), filename, PIO_CLOBBER)))
ERR(ret);
/* Define netCDF dimension and variable. */
printf("rank: %d Defining netCDF metadata...\n", my_rank);
for (int d = 0; d < NDIM3; d++)
if ((ret = PIOc_def_dim(ncid, dim_name[d], dim_len[d], &dimid[d])))
ERR(ret);
if ((ret = PIOc_def_var(ncid, VAR_NAME, PIO_INT, NDIM3, dimid, &varid)))
ERR(ret);
if ((ret = PIOc_enddef(ncid)))
ERR(ret);
/* Allocate storage for sample data. */
int buffer[elements_per_pe];
/* Write each timestep. */
for (int t = 0; t < NUM_TIMESTEPS; t++)
{
/* Create some data for this timestep. */
for (int i = 0; i < elements_per_pe; i++)
buffer[i] = 100 * t + START_DATA_VAL + my_rank;
/* Write data to the file. */
printf("rank: %d Writing sample data...\n", my_rank);
if ((ret = PIOc_setframe(ncid, varid, t)))
ERR(ret);
if ((ret = PIOc_write_darray(ncid, varid, ioid, elements_per_pe, buffer, NULL)))
ERR(ret);
}
/* THis will cause all data to be written to disk. */
if ((ret = PIOc_sync(ncid)))
ERR(ret);
/* Close the netCDF file. */
printf("rank: %d Closing the sample data file...\n", my_rank);
if ((ret = PIOc_closefile(ncid)))
ERR(ret);
/* Check the output file. */
/* if ((ret = check_file(iosysid, ntasks, filename, format[fmt], elements_per_pe, */
/* my_rank, ioid))) */
/* ERR(ret); */
}
/* Free the PIO decomposition. */
printf("rank: %d Freeing PIO decomposition...\n", my_rank);
if ((ret = PIOc_freedecomp(iosysid, ioid)))
ERR(ret);
/* Finalize the IO system. */
printf("rank: %d Freeing PIO resources...\n", my_rank);
if ((ret = PIOc_finalize(iosysid)))
ERR(ret);
/* Finalize the MPI library. */
MPI_Finalize();
#ifdef TIMING
/* Finalize the GPTL timing library. */
if ((ret = GPTLfinalize ()))
return ret;
#endif
printf("rank: %d SUCCESS!\n", my_rank);
return 0;
}
/* Write, then read, a simple example with darrays.
The sample file created by this program is a small netCDF file. It
has the following contents (as shown by ncdump):
<pre>
netcdf darray_no_async_iotype_1 {
dimensions:
unlimted = UNLIMITED ; // (2 currently)
x = 4 ;
y = 4 ;
variables:
int foo(unlimted, x, y) ;
data:
foo =
42, 42, 42, 42,
43, 43, 43, 43,
44, 44, 44, 44,
45, 45, 45, 45,
142, 142, 142, 142,
143, 143, 143, 143,
144, 144, 144, 144,
145, 145, 145, 145 ;
}
</pre>
*/
int main(int argc, char* argv[])
{
int my_rank; /* Zero-based rank of processor. */
int ntasks; /* Number of processors involved in current execution. */
int iosysid; /* The ID for the parallel I/O system. */
/* int ncid; /\* The ncid of the netCDF file. *\/ */
/* int dimid[NDIM3]; /\* The dimension ID. *\/ */
/* int varid; /\* The ID of the netCDF varable. *\/ */
/* char filename[NC_MAX_NAME + 1]; /\* Test filename. *\/ */
/* int num_flavors = 0; /\* Number of iotypes available in this build. *\/ */
/* int format[NUM_NETCDF_FLAVORS]; /\* Different output flavors. *\/ */
int ret; /* Return value. */
#ifdef TIMING
/* Initialize the GPTL timing library. */
if ((ret = GPTLinitialize ()))
return ret;
#endif
/* Initialize MPI. */
if ((ret = MPI_Init(&argc, &argv)))
MPIERR(ret);
if ((ret = MPI_Comm_set_errhandler(MPI_COMM_WORLD, MPI_ERRORS_RETURN)))
MPIERR(ret);
/* Learn my rank and the total number of processors. */
if ((ret = MPI_Comm_rank(MPI_COMM_WORLD, &my_rank)))
MPIERR(ret);
if ((ret = MPI_Comm_size(MPI_COMM_WORLD, &ntasks)))
MPIERR(ret);
/* Check that a valid number of processors was specified. */
printf("%d: ParallelIO Library darray_async example running on %d processors.\n",
my_rank, ntasks);
if (ntasks != TARGET_NTASKS)
{
fprintf(stderr, "Number of processors must be %d!\n", TARGET_NTASKS);
return ERR_BAD;
}
/* Turn on logging. */
if ((ret = PIOc_set_log_level(LOG_LEVEL)))
return ret;
/* Num procs for computation. */
int num_procs2[COMPONENT_COUNT] = {4};
/* Is the current process a computation task? */
int comp_task = my_rank < NUM_IO_TASKS ? 0 : 1;
/* Initialize the IO system. */
if ((ret = PIOc_init_async(MPI_COMM_WORLD, NUM_IO_TASKS, NULL, COMPONENT_COUNT,
num_procs2, NULL, NULL, NULL, PIO_REARR_BOX, &iosysid)))
ERR(ret);
/* The rest of the code executes on computation tasks only. As
* PIO functions are called on the computation tasks, the
* async system will call them on the IO task. When the
* computation tasks call PIO_finalize(), the IO task will get
* a message to shut itself down. */
if (comp_task)
{
/* PIO_Offset elements_per_pe; /\* Array elements per processing unit. *\/ */
/* int ioid; /\* The I/O description ID. *\/ */
/* /\* How many elements on each computation task? *\/ */
/* elements_per_pe = DIM_LEN_X * DIM_LEN_Y / NUM_COMP_TASKS; */
/* /\* Allocate and initialize array of decomposition mapping. *\/ */
/* PIO_Offset compdof[elements_per_pe]; */
/* for (int i = 0; i < elements_per_pe; i++) */
/* compdof[i] = my_rank * elements_per_pe + i; */
/* /\* Create the PIO decomposition for this example. Since */
/* this is a variable with an unlimited dimension, we want */
/* to create a 2-D composition which represents one */
/* record. *\/ */
/* printf("rank: %d Creating decomposition...\n", my_rank); */
/* if ((ret = PIOc_init_decomp(iosysid, PIO_INT, NDIM3 - 1, &dim_len[1], elements_per_pe, */
/* compdof, &ioid, 0, NULL, NULL))) */
/* ERR(ret); */
/* /\* The number of favors may change with the build parameters. *\/ */
/* #ifdef _PNETCDF */
/* format[num_flavors++] = PIO_IOTYPE_PNETCDF; */
/* #endif */
/* format[num_flavors++] = PIO_IOTYPE_NETCDF; */
/* #ifdef _NETCDF4 */
/* format[num_flavors++] = PIO_IOTYPE_NETCDF4C; */
/* format[num_flavors++] = PIO_IOTYPE_NETCDF4P; */
/* #endif */
/* /\* Use PIO to create the example file in each of the four */
/* * available ways. *\/ */
/* for (int fmt = 0; fmt < num_flavors; fmt++) */
/* { */
/* /\* Create a filename. *\/ */
/* sprintf(filename, "darray_no_async_iotype_%d.nc", format[fmt]); */
/* /\* Create the netCDF output file. *\/ */
/* printf("rank: %d Creating sample file %s with format %d...\n", */
/* my_rank, filename, format[fmt]); */
/* if ((ret = PIOc_createfile(iosysid, &ncid, &(format[fmt]), filename, PIO_CLOBBER))) */
/* ERR(ret); */
/* /\* Define netCDF dimension and variable. *\/ */
/* printf("rank: %d Defining netCDF metadata...\n", my_rank); */
/* for (int d = 0; d < NDIM3; d++) */
/* if ((ret = PIOc_def_dim(ncid, dim_name[d], dim_len[d], &dimid[d]))) */
/* ERR(ret); */
/* if ((ret = PIOc_def_var(ncid, VAR_NAME, PIO_INT, NDIM3, dimid, &varid))) */
/* ERR(ret); */
/* if ((ret = PIOc_enddef(ncid))) */
/* ERR(ret); */
/* /\* Allocate storage for sample data. *\/ */
/* int buffer[elements_per_pe]; */
/* /\* Write each timestep. *\/ */
/* for (int t = 0; t < NUM_TIMESTEPS; t++) */
/* { */
/* /\* Create some data for this timestep. *\/ */
/* for (int i = 0; i < elements_per_pe; i++) */
/* buffer[i] = 100 * t + START_DATA_VAL + my_rank; */
/* /\* Write data to the file. *\/ */
/* printf("rank: %d Writing sample data...\n", my_rank); */
/* if ((ret = PIOc_setframe(ncid, varid, t))) */
/* ERR(ret); */
/* if ((ret = PIOc_write_darray(ncid, varid, ioid, elements_per_pe, buffer, NULL))) */
/* ERR(ret); */
/* } */
/* /\* THis will cause all data to be written to disk. *\/ */
/* if ((ret = PIOc_sync(ncid))) */
/* ERR(ret); */
/* /\* Close the netCDF file. *\/ */
/* printf("rank: %d Closing the sample data file...\n", my_rank); */
/* if ((ret = PIOc_closefile(ncid))) */
/* ERR(ret); */
/* /\* Check the output file. *\/ */
/* /\* if ((ret = check_file(iosysid, ntasks, filename, format[fmt], elements_per_pe, *\/ */
/* /\* my_rank, ioid))) *\/ */
/* /\* ERR(ret); *\/ */
/* } */
/* Free the PIO decomposition. */
/* printf("rank: %d Freeing PIO decomposition...\n", my_rank); */
/* if ((ret = PIOc_freedecomp(iosysid, ioid))) */
/* ERR(ret); */
/* Finalize the IO system. Only call this from the computation tasks. */
printf("%d %s Freeing PIO resources\n", my_rank, TEST_NAME);
if ((ret = PIOc_finalize(iosysid)))
ERR(ret);
} /* endif comp_task */
/* Finalize the MPI library. */
MPI_Finalize();
#ifdef TIMING
/* Finalize the GPTL timing library. */
if ((ret = GPTLfinalize ()))
return ret;
#endif
printf("rank: %d SUCCESS!\n", my_rank);
return 0;
}