* Thu Jul 14 2016 toddrme2178@gmail.com
- Update to 1.7.0
o New features
* Nonblocking requests now can be posted while the program is in the define
mode. This feature enables applications to add new variables and post the
nonblocking write requests without separating the codes by
ncmpi_enddef(). Note ncmpi_wait_all and ncmpi_wait must still be called
in the data mode.
* When using immutable write buffer in put APIs on a Little Endian machine,
the in-place byte swap operation performed internally in PnetCDF can
cause a fatal error (by trying to change the contents of an immutable
memory space). The solution of copying user's write request to a
temporary buffer and perform byte swap there conflicts with PnetCDF's
design principle of avoiding internal memory allocation as much as
possible. Since the size of immutable buffer in most cases is small,
PnetCDF now compromise the two by making a copy of write requests that is
less than 4KB. Users are warned that using immutable write buffer larger
than 4KB will still cause the fatal error in the PnetCDF default
configuration. The way to completely disable in-place byte swap is to
build PnetCDF with --disable-in-place-swap option at the configure time.
In this case, the internal memory allocation in PnetCDF will increase.
o Syntax changes
* In ncmpi_wait_all(), ncmpi_wait(), and ncmpi_cancel(), the API's third
argument, array of requests IDs, is changed to INOUT. Upon successful
completion or cancellation of individual nonblocking requests, the
corresponding request IDs are set to NC_REQ_NULL.
o New run-time environment variables
* PNETCDF_VERBOSE_DEBUG_MODE environment variable can be used to print the
location in the source code where the error code is originated, no matter
the error is intended or not. This run-time environment variable only
takes effect when PnetCDF is configure with debug mode, i.e.
- -enable-debug is used at the configure command line. Set this variable
to 1 to enable. Set it to 0 or keep it unset disables this mode. Default
is 0, i.e. disabled. Users are warned that enabling this mode may result
in a lot of debugging messages printed in stderr.
o New example programs
* examples/C/nonblocking_write_in_def.c shows an example of posting
nonblocking write requests in the define mode.
* examples/C/req_all.c shows an example of using NC_REQ_ALL to flush all
pending nonblocking requests without providing the requests IDs.
o New test program
* test/cdf_format/dim_cdf12.c tests defining maximal dimension size for
CDF-1 and CDF-2 file formats
* test/testcases/test_erange.c tests if the error code NC_ERANGE can be
correctly returned for two cases 1) get a NC_UBYTE value of 255 from a
netCDF file to a memory buffer of type signed char and 2) put a value
of -1 of signed char to a NC_UBYTE variable in a netCDF file
* test/testcases/check_type.c tests if the error codes can be correctly
returned when conflicted in-memory and extenal data types are used.
* test/testcases/put_parameter.f tests the use of immutable write buffer
(e.g. a buffer declared as PARAMETER). Note the buffer size must be
smaller than 4KB.
* test/nonblocking/i_varn_indef.c tests posting nonblocking requests in
define mode.
* test/nonblocking/req_all.c tests the use of NC_REQ_ALL for flushing
all pending nonblocking requests without providing the requests IDs.
* test/last_large_var.c tests the special case when there is no record
variable, the last fixed-size variable can be larger than 2GiB in size
if its starting file offset is less than 2GiB.
* test/testcases/buftype_free.c and test/testcases/buftype_freef.f test
the bug in r2160.
* testcases/add_var.c checks the starting file offsets of newly added
variables from re-entering the define mode.
* testcases/attrf.f checks NF_ERANGE is returned instead of coredump. This
is particularly for NAG Fortran compiler that may report "Arithmetics
exception".
* testcases/check_striping.c checks if the file striping unit and factor
returned from MPI-IO hints are consistent among processes.
* test/nonblocking/column_wise.c checks if PnetCDF detects interleaved
fileviews from multiple nonblocking requests and correctly breaks and
reconstructs the fileviews so the combined fileview is monotonic
non-decreasing in file offsets.
o New optimization
* Filling variables at ncmpi_enddef() is now done by aggregating all write
requests into one MPI collective write call. In v 1.6.1, this is done by
filling one variable at a time.
o New utility program
* ncoffsets reports the file offset information, including the starting and
ending file offsets, of variables stored in a netCDF file. ncoffsets is
compiled with gcc if gcc is presented on the build system. Additional
command-line options are: (-v) reports only for a selected list of
variables in interest, (-s) prints the variable sizes, (-g) outputs the
file space gap size from the end of previous variable, (-x) reports
whether there is a gap between any two adjacent fixed-size variables. See
the man page for descriptions of all command-line options and examples.
o Semantics updates
* All nonblocking APIs now take a NULL pointer for the request ID argument,
meaning users do not wish to keep track of the request ID. If NULL
request IDs are used, NC_REQ_ALL should be used when calling
ncmpi_wait_all/ncmpi_wait to commit all the pending nonblocking requests.
This feature relinquishes users from the responsibility of tracking the
IDs of pending requests.
* Using NC_REQ_ALL as the 2nd argument "num" in ncmpi_wait_all/ncmpi_wait
APIs will flush all the pending nonblocking requests. In this case, the
3rd and 4th arguments "array_of_requests" and "array_of_statuses" will be
ignored and thus these two arguments can be NULLs.
* Using NC_REQ_ALL in ncmpi_cancel() will cancel all the pending
nonblocking requests.
* Using NC_GET_REQ_ALL or NC_PUT_REQ_ALL in ncmpi_wait_all(), ncmpi_wait(),
and ncmpi_cancel() for all the pending get-only or put-only requests,
respectively.
o Other updates:
* Conform with netCDF on the maximal dimension size for CDF-2 file format
to be (2^32 - 4)
* NC_ERANGE checks have been removed from nc_test for text APIs and
variables that are defined as NC_CHAR type
* Add README.K-Computer build recipe for using Fujitsu MPI compilers on the
K computer at RIKEN in Japan
* Add README.INTEL build recipe for using Intel MPI compilers 4.x
* Build dependency rule is added for files configure and configure.in
* PnetCDF checks MPICC/MPICXX/MPIF77/MPIF90 instead of CC/CXX/F77/F90/FC.
If MPICC/MPICXX/MPIF77/MPIF90 are set, PnetCDF will ignore
CC/CXX/F77/F90/FC. If CC/CXX/F77/F90/FC is set instead of
MPICC/MPICXX/MPIF77/MPIF90, PnetCDF will now copy them to
MPICC/MPICXX/MPIF77/MPIF90.
* Enforce netCDF convention on error code priority: NC_ECHAR trumps
NC_EINVALCOORDS, NC_EEDGE, and NC_ESTRIDE.
* Return error code NC_EGLOBAL instead of NC_ENOTVAR for APIs where using
NC_GLOBAL as the variable ID argument is prohibited.
* All Fortran 77 test and example programs (files with .f and .F
extensions) have been revised to conform with 77 standard, and was tested
using pgf77.
* Now provides a pkg-config file
(http://www.freedesktop.org/wiki/Software/pkg-config/), making it
slightly easier to set the correct pnetcdf include and library paths.
o Bug fixes
* Fix the bus error of invalid address alignment when build with Fujitsu
compiler. See r2171 and r2180.
* Fix the bug for the special case when there is no record variable, the
last fixed-size variable can be larger than 2GiB in size if its starting
file offset is less than 2GiB. See r2166.
* Fix the nonblocking flexible APIs that fail to save (duplicate) the user
MPI derived data type that later is needed to unpack read data to the
user buffer (a call to MPI_Unpack). See r2160.
* Fix Fortran 77 constants nf_fill_uint and nf_fill_int64 (thanks Jim
Edwards) that pgf77-based MPI compiler does not like "_8" modifier.
See r2045 and r2051. The same issue for NAG Fortran compiler is also
resolved. See r2089 and r2093.
* In the example program examples/tutorial/pnetcdf-write-nb.c, the write
buffers used in two iput API calls should be different. See r2095.
* Fix the error reporting mechanism for NC_ERANGE, for when an arithmetic
overflow happens. Overflow checking is now performed before I/O.
* Wed Jul 15 2015 toddrme2178@gmail.com
- Update to version 1.6.1
o New features
* PnetCDF now supports fill mode. ncmpi_set_fill() can be used to set the
fill mode for the entire data set. Note the differences from netCDF:
1. The default mode in PnetCDF is NC_NOFILL.
2. Setting fill mode for the entire file or individual variables must be
done in define mode.
3. For non-record variables, they are filled at the time ncmpi_enddef()
is called.
4. For record variables, users must explicitly call ncmpi_fill_var_rec()
to fill one record at a time before writing to the record of that
variable.
o New APIs
* ncmpi_def_var_fill() sets fill mode for an individual variable. This API
must be called in define mode.
* ncmpi_inq_var_fill() inquires fill mode of a variable.
* ncmpi_fill_var_rec() is a collective API that fills a record of a record
variable. This API must be called at data mode.
* ncmpi_inq_default_format() for inquiring the default file format for
new file creation. Note the default format can be changed by
ncmpi_set_default_format().
* The above new API are also available in Fortran and C++ versions.
o New error code
* NC_ENOTRECVAR when attempting operation only for record variables
* NC_ENOTFILL when attempting to fill a variable when its fill mode is off
* NC_EMULTIDEFINE_FILL_MODE when inconsistent dataset fill mode is detected
* NC_EMULTIDEFINE_VAR_FILL_MODE when inconsistent variable fill mode is
detected
* NC_EMULTIDEFINE_VAR_FILL_VALUE when inconsistent variable fill value is
detected
* Fortran versions of the above error codes are also added.
o New example programs
* C/fill_mode.c shows the use of ncmpi_set_fill(), ncmpi_def_var_fill(),
ncmpi_inq_var_fill() and ncmpi_fill_var_rec()
F77/fill_mode.f is the Fortran version.
F90/fill_mode.f90 is the Fortran 90 version.
CXX/fill_mode.cpp is the C++ version.
* C/ghost_cell.c shows how to use varm API for writing from a user buffer
as a 2D array with ghost cells on both ends of every dimension.
o New test programs
* nc_test/tst_nofill.c borrowed from netCDF
* testcases/ivarn.c tests bug fix r2023 when the request IDs stored in
argument array_of_requests[] of ncmpi_wait_all() are not in an
increasing order.
o Other updates:
* Change the chunk size used for moving variable data when the file header
extent expands. The default is now 1MB. If the file's striping unit
size is known (from MPI-IO hint striping_unit) then the chunk size is
set to the striping unit size.
o Bug fixes
* Add missing F90 function overloading for f90mpi_put_var_all,
f90mpi_get_var_all, f90mpi_put_vard_all, and f90mpi_get_vard_all,
when the user buffer is a scalar.
* Fix when the request IDs passed in argument array_of_requests[] of
ncmpi_wait_all() are not in an increasing order. See r2023.
* Fix C++ compile error for converting NULL to string. See r2039.
- Update to version 1.6.0
o Format conformation updates:
* Conform with netCDF4 on CDF-1 and CDF-2 formats. The only difference now
between the two formats are the OFFSET item in the format spec (32 vs.
64 bit integers.) All names (variable, dimension, attribute) now allow
extended characters (eg. special2 and MUTF8).
o New APIs
* Nonblocking buffered varn API family.
For C, ncmpi_bput_varn_<type>()
For F77, nfmpi_bput_varn_<type>()
For F90, nf90mpi_bput_varn()
For C++, NcmpiVar::bputVarn()
* Nonblocking varn API family.
For C, ncmpi_iput_varn_<type>() and ncmpi_iget_varn_<type>()
For F77, nfmpi_iput_varn_<type>() and nfmpi_iget_varn_<type>()
For F90, nf90mpi_iput_varn() and nf90mpi_iget_varn()
For C++, NcmpiVar::iputVarn() and NcmpiVar::igetVarn()
* Blocking vard API family takes an argument of MPI derived data type that
describes the file access layout, as opposed to vara and vars APIs that
use start[] and count[].
For C, ncmpi_put_vard() and ncmpi_get_vard()
For F77, nfmpi_put_vard() and nfmpi_get_vard()
For F90, nf90mpi_put_vard() and nf90mpi_get_vard()
For C++, NcmpiVar::putVard() and NcmpiVar::getVard()
* Collective var1 API family
For C, ncmpi_put_var1_all() ncmpi_get_var1_all()
ncmpi_put_var1_<type>_all() ncmpi_get_var1_<type>_all()
For F77, nfmpi_put_var1_all() nfmpi_get_var1_all()
nfmpi_put_var1_<type>_all() nfmpi_get_var1_<type>_all()
For F90, nf90mpi_put_var_all() nf90mpi_get_var_all()
For C++, NcmpiVar::putVar_all() NcmpiVar::getVar_all()
* ncmpi_inq_buffer_size() returns the size of buffer previously attached
for use of bput APIs. With ncmpi_inq_buffer_usage() one can calculate
the space remaining for additional bput requests.
For F77, nfmpi_inq_buffer_size()
For F90, nf90mpi_inq_buffer_size()
For C++, NcmpiFile::Inq_buffer_size()
* ncmpi_inq_recsize() returns the size of record block, i.e. the sum of
single records of all record variables.
For F77, nfmpi_inq_recsize()
For F90, nf90mpi_inq_recsize()
For C++, NcmpiGroup::getRecSize()
* ncmpi_inq_num_rec_vars() and ncmpi_inq_num_fix_vars() report the number
of record and fixed-size variables, respectively.
For F77, nfmpi_inq_num_rec_vars() and nfmpi_inq_num_fix_vars()
For F90, nf90mpi_inq_num_rec_vars() and nf90mpi_inq_num_fix_vars()
For C++, NcmpiGroup::getRecVarCount() and NcmpiGroup::getFixVarCount()
o New PnetCDF hint
* pnetcdf_subfiling -- it can be set in an MPI info object or in the
environment variable PNETCDF_HINTS to enable/disable subfiling.
The value is either "enable" or "disable".
o PnetCDF hint priority
* The alignment hints set in the environment variable "PNETCDF_HINTS" have
the highest priority, which overwrite the alignment hints set in
ncmpi__enddef(), which overwrite the alignment hints set in the MPI_Info
object used in the call of ncmpi_create() and ncmpi_open().
o New error code
* NC_ESTRICTCDF2 for attempting CDF-5 operation on CDF-2 file. For
example, define a variable of type NC_INT64 in a CDF-2 file.
* NC_ETYPESIZE when filetype size is bigger than the variable size
* NC_ETYPE_MISMATCH when the element type of filetype mismatches the
variable type
* NC_ETYPESIZE_MISMATCH when filetype size mismatches buffer type size
* NC_ENULLSTART when argument start is a NULL pointer
* NC_ENULLCOUNT when argument count is a NULL pointer
* NC_EINVAL_CMODE when invalid file create mode is set, (e.g. cannot have
both NC_64BIT_OFFSET & NC_64BIT_DATA. In PnetCDF 1.5.0 and earlier
versions, if both flags were set, then NC_64BIT_DATA triumphs
NC_64BIT_OFFSET.)
o New example programs
* C/bput_varn_uint.c and F77/bput_varn_int8.f show the use of
nonblocking bput_varn APIs
* C/i_varn_int64.c and F77/i_varn_real.f show the use of nonblocking
iput_varn and iget_varn APIs
* C/vard_int.c F77/vard_int.f F90/vard_int.f90 CXX/vard_int.cpp show the
use of vard API to write/read record and fixed-size variables.
* C/transpose.c shows the use of ncmpi_put_vara_int_all to write a 3D array
that is dimensional-transposed from the one stored in memory. Six
transposed 3D arrays are saved whose dimensions are organized as ZYX,
ZXY, YZX, YXZ, XZY, and XYZ. The C++, Fortran 77, and Fortran 90
versions are also provided.
o New test program
* nonblocking/wait_after_indep.c tests if ncmpi_end_indep_data() returns
properly when nonblocking APIs are called in independent data mode and
the wait call is made after exiting the independent mode.
* nonblocking/flexible_bput.c tests flexible bput APIs that use
noncontiguous buffer type, noncontiguous imap and require type conversion
* testcases/flexible2.c tests flexible APIs that perform type conversion
* testcases/flexible_varm.c tests flexible varm APIs
* testcases/varn_contig.c tests the case when the fileview is actually a
contiguous chunk. PnetCDF should be able to merge all.
* nonblocking/bput_varn_uint.c tests nonblocking bput_varn APIs
* nonblocking/i_varn_int64.c tests nonblocking iput_varn and iget_varn APIs
* test/testcases/test_vard.c test/testcases/test_vardf.f
test/testcases/test_vardf90.f90 test the new vard APIs.
* test/testcases/inq_recsize.c tests if one can correctly inquire the
size of record block from in a netCDF file. A similar program in F90,
named inq_recsizef.f90, is also added.
* In test/nc_test, the test programs below are borrowed from netCDF test
programs: t_nc.c tst_misc.c tst_norm.c tst_small.c tst_names.c
tst_atts3.c tst_atts.c
* test/testcases/one_record.c tests the special case defined in CDF-1 and
CDF-2 specification that "A special case: Where there is exactly one
record variable, we drop the requirement that each record be four-byte
aligned, so in this case there is no record padding."
* test/testcases/modes.c tests if correct error codes are returned when
various file create/open modes are used.
* Under test/testcases, varn_int.c varn_intf.f varn_real.f90 test varn APIs
* test/testcases/inq_num_vars.c test if one can correctly inquire the
numbers of record and fixed-size variables defined in a netCDF file.
A similar program in F90, named inq_num_varsf.f90, is also added.
* test/nonblocking/interleaved.c tests a combination of interleaved
file types. This is to test the bug fix in r1758.
o New optimization
* Prior to this release, PnetCDF wraps each MPI read/write call in put/get
APIs with two MPI_File_set_view(). One is before the MPI read/write call
to take advantage of MPI's fileview feature for accessing non-contiguous
file locations. The other is after the MPI read/write call to make the
whole file visible, as the root process may write to file header later
in the data mode and it alone cannot make a call to MPI_File_set_view()
because the function is collective.
In this release, the second MPI_File_set_view() has been removed. The
root process's fileview is changed to always keep the whole file header
visible. Saving a collective call to MPI_File_set_view() is expected to
improve some performance.
o Semantics updates
* Header consistency mechanism has been updated. See README.consistency for
details.
* The use of NC_SHARE is also revised. See README.consistency for details.
o Other updates:
* The subfiling feature is now disabled in ncmpidump and ncmpidiff until
a bug fix to allow reading the master file with the number of processes
that is smaller than the number of subfiles.
* The attribute names reserved for subfiling feature are now changed to
use a prefix of "_PnetCDF_SubFiling". The leading "_" underscore is a CDL
convention.
* The flexible APIs now allow argument buftype to be MPI_DATATYPE_NULL.
In this case, argument bufcount is ignored and argument buf's data type
must match the data type of variable defined in the file - no data
conversion will be done. This extension makes the flexible APIs be
able to correspond to the netCDF APIs whose names do not contain a data
type, e.g. nc_put_vara().
* Type conversion between NC_BYTE and unsigned char no longer checks for
out of range error (NC_ERANGE). CDF file format specification make a
special case for this kind of data type conversion. See: "Note on byte
data" in the format specification.
* Conform with the CDF-2 and CDF-5 formats that names are normalized
according to Unicode NFC normalization rules during encoding as UTF-8
for storing in the file header.
* A new configure option, --enable-large-file-test, to enable testing
I/O on large files and large variables. Note the testing prgrams will
run very slowly.
o Bug fixes
* fix the bug for flexible get_varn API. When buftype is noncontiguous, the
bug forgot to "unpacks" the temporary buffer that reads data from file to
the user buffer using buftype.
* fix the bug in blocking flexible get APIs when buftype is noncontiguous,
swap is needed, type conversion is not. The bug sets a NULL pointer to
the read buffer and passes it to MPI_File_read functions. See r1815.
* fix the bug for the NetCDF special case when there is only one record
variable and the record size is not four-byte aligned. In this case,
NetCDF spec says the alignment must dropped (no padding). See r1791.
* fix the bug in nonblocking APIs when requests are resorted into
nonoverlapping groups and the first group of file types are interleaved,
it was mistakenly identified as non-interleaved. See r1758.
* fix the setting for PNETCDF_RELEASE_DATE in configure.in. Using read
command together with IFS does not work as expected in bash 4.3.11. See
http://lists.mcs.anl.gov/pipermail/parallel-netcdf/2014-July/001586.html
for further detailed discussion.
- Update to version 1.5.0
o New features
* A new configure option "--disable-in-place-swap" is added. It disables
the byte-swap operations running in-place on the user's write buffers.
The purpose of providing this option is to deal with the problem when a
Fortran program uses a immutable buffer for put APIs, e.g. the buffer is
declared as a PARAMETER, and in-place byte swap on this buffer causes
segmentation fault. See discussion threads of
http://lists.mcs.anl.gov/pipermail/parallel-netcdf/2013-July/001498.html
Impacts:
1. It takes effect only on Little Endian machines.
2. It only affects put/iput data APIs, but not attribute APIs.
3. The INTENT of buffer arguments in all Fortran 90 put/iput APIs will be
declared as "IN". Without this setting, the default is "INOUT".
4. It has an impact on performance, as an extra internal temporary buffer
will be allocated to copy data over from user's put buffer, so byte
swap can be run on the temporary buffer.
The default setting is to enable in-place byte swap. PnetCDF tries not to
allocate additional memory space, due to performance concern. Users are
discouraged to use Fortran PARAMETER buffers in put APIs.
* A new configure option "--enable-debug" is added. It enables a memory
allocation tracing mechanism internal in PnetCDF. In addition, it enables
the PnetCDF safe mode. (Note that setting the environment variable
PNETCDF_SAFE_MODE at the run time can still overwrite the safe mode.)
Default debug mode is disabled. When enabled, a user program can call
three new APIs below: ncmpi_inq_malloc_size, ncmpi_inq_malloc_max_size,
and ncmpi_inq_malloc_list to obtain the size in bytes of current memory
allocated internally. This feature uses a binary tree to manage all
malloc buffers, e.g. tsearch() and tdelete().
* Add three Fortran parameters for PnetCDF library version numbers:
PNETCDF_VERSION_MAJOR, PNETCDF_VERSION_MINOR, and PNETCDF_VERSION_SUB.
Similarly in C programs, these are defined in pnetcdf.h as constants.
o New APIs
* C++ APIs are now available. They are developed based on netCDF-4 C++
library. However, users are encouraged to use C APIs, instead C++.
* ncmpi_put_att, ncmpi_get_att, nfmpi_put_att, and nfmpi_get_att - these
APIs correspond to nc_put_att, nc_get_att, nf_put_att, and nf_get_att.
Note they are not the "flexible" APIs. Flexible APIs have an MPI derived
datatype argument.
* ncmpi__enddef, nfmpi__enddef, and nf90mpi_enddef - these APIs correspond
to netCDF nc__enddef, nf__enddef, and nf90_enddef (with additional
optional arguments).
* ncmpi_inq_file_info - the naming for ncmpi_get_file_info may cause
confusion, as "get" has a different meaning on PnetCDF. The correct
name should use "inq". However, ncmpi_get_file_info is kept for backward
compatibility.
* ncmpi_inq_striping, nfmpi_inq_striping, nf90mpi_inq_striping report the
file system striping settings of the opened file: striping size and
striping count, if the underneath MPI-IO can find their values from the
file system in use.
* ncmpi_inq_malloc_size, ncmpi_inq_malloc_max_size, ncmpi_inq_malloc_list
report the size in bytes of current memory allocated internally by
PnetCDF. Similar APIs for Fortran 77 and 90 are also available. These
APIs are enabled only when PnetCDF is configured with option
- -enable-debug. When this option is not enabled, calling these APIs will
return the error code NC_ENOTENABLED. These APIs are usefully for
debugging.
* ncmpi_inq_files_opened reports the number of files that are currently
opened. Similar API for Fortran 77 and 90 are also available. The API
takes 2 arguments: number of files and array of file IDs. If the second
argument, array of IDs, is not NULL, then it will filled with the netCDF
dataset IDs. This API is useful for debugging.
o Syntax changes
* For all Fortran put/iput APIs, the INTENT of write buffer arguments is
changed to "INOUT" on Little Endian machines, if option
"--disable-in-place-swap" is not used at configuration. Otherwise, i.e.
on Big Endian machines or PnetCDF is configured with
"--disable-in-place-swap" on Little Endian machines, the INTENT is "IN".
o New PnetCDF hint
* nc_record_align_size - aligns the starting file offset of the record
variable section. Note this is for the entire section, not individual
records.
o New error code
* NC_ENOTENABLED indicates the API is available only when the corresponding
feature is enabled. For example, nfmpi_inq_malloc_size() returns this
error code when "--enable-debug" is not used at configure.
* NC_EBAD_FILE corresponds to MPI error code MPI_ERR_BAD_FILE, meaning
"Invalid file name (e.g., path name too long)."
* NC_ENO_SPACE corresponds to MPI error code MPI_ERR_NO_SPACE, meaning
"Not enough space."
* NC_EQUOTA corresponds to MPI error code MPI_ERR_QUOTA, meaning
"Quota exceeded."
* NC_EMULTIDEFINE_FNC_ARGS corresponds to MPI error code MPI_ERR_NOT_SAME,
meaning "inconsistent function arguments used in collective API."
o New run-time environment variables
* none
o New example programs
* Example programs now report if there is any PnetCDF internal malloc
residues yet to be freed, if --enable-debug option is used at
configure.
* Under examples/C, three examples are added: create_open.c, get_vara.c,
and global_attributes.c. File examples/README contains their
descriptions.
* Under examples/CXX, several example programs corresponding to those in
examples/C are added.
o New programs for I/O benchmarks
* none
o New test program
* Many test programs now report if there is any PnetCDF internal malloc
residues yet to be freed, if --enable-debug option is used at configure.
* add tests for flexible APIs. The tests borrow from nc_test/test_write.c
that tests nc_put_var1, nc_put_vara, nc_put_vars, and nc_put_varm.
Similarly for get APIs.
* testcases/record.c checks if the number of records is updated correctly.
It writes to a variable's 2nd record followed by the 1st record. A call
to ncmpi_inq_dim() or ncmpi_inq_dimlen() should report 2 records after
the writes complete.
* testcases/noclobber.c checks if error code NC_EEXIST can be returned
correctly when NC_NOCLOBBER modes is used in ncmpi_create and in the
meantime the file exists.
* Some test programs are developed to run in parallel. The test mechanism
for parallel runs is command "make ptest". Two environment variables
can be used to set the MPI run command and output file directory:
TEST_MPIRUN and TEST_OUTDIR. Their defaults are mpiexec and "." (current
directory), respectively. For example,
make ptest TEST_MPIRUN="aprun -n NP" TEST_OUTDIR=/scratch
Note the keyword "NP" will be replaced by the different numbers of
processes used to run the tests. The testing uses up to 8 MPI processes.
* A sample PBS script file is provided to test "make ptest" on machines
with a batch queue system: pbs.script. This example script can be
submitted from the build root directory (where you run "make" command to
build PnetCDF library).
* For cross compile environment (and batch queue system), the environment
variable TEST_SEQRUN can be used to set the MPI run command. For example,
the command for testing sequential programs:
make check TEST_SEQRUN="aprun -n 1" TEST_OUTDIR=/scratch
For non-cross compile environment, there is no need to set the environment
variables, as long as one can run the MPI executable without mpirun or
mpiexec.
o New optimization
* none
o New utility program
* pnetcdf_version prints the version information of the PnetCDF library and
command-line arguments used at configure
o Other updates:
* Revise FLASH-IO benchmark to use nonblocking APIs for both checkpoint and
plot writes. The control variable to switch between nonblocking and
blocking API is "use_nonblocking_io". Set it to .FALSE. in
flash_benchmark_io.F90 to switch to blocking APIs. Using nonblocking APIs
is now the default.
* To match ncdump, ncmpidump now only supports one input file.
* Makefiles are revised for better recursive make and fixed some build
target dependency for parallel make.
* File creation was revised for handling NC_CLOBBER and NC_NOCLOBBER modes.
On systems where Unix calls access() and unlink() are available, they
are used to check if a file exits and to delete an existing file.
* subfiling is now enabled by default. Users can use --disable-subfiling to
disable it
* man page of ncmpigen is updated to add the description for option "-v"
which lets users to specify the desired output file formats, e.g.
CDF-1, CDF-2, or CDF-5.
* flex, lex, bison, or yacc are no longer needed for building ncmpigen.
o Bug fixes
* fix ncmpigen.y on parsing CDL file to get the number of records. The bug
failed the command "make b-test", due to getting a wrong number of
records (current value for the unlimited dimension).
* fix the update for number of records when writing to a record that is
not the last record.
- Update to version 1.4.1
o Bug fix:
* Improve pnetcdf.inc portability for fixed/free-form Fortran programs
o Fortran API syntax changes
* For nfmpi_put_att and nf90mpi_put_att family, the intent modifier for
the put buffer arguments are now declared as INTENT(IN).
* For nfmpi_put_var* and nf90mpi_put_var family
+ On Big Endian machines, the the intent modifier for the put buffer
arguments are now declared as INTENT(IN).
+ On Little Endian machines, the the intent modifier for the put buffer
arguments are still declared as INTENT(INOUT). This is because PnetCDF
does in-place byte-swap on user's put buffer. If user's buffer is
declared as parameter, then segment fault can happen when PnetCDF
tries to byte-swap a read-only memory.
o Subfiling
* Subfiling is a new PnetCDF feature that divides a file transparently
into several smaller subfiles, each of which stores subarrays
in CDF file formats. The file name supplied by the users serves as a
master file that contains all metadata about array partitioning
information among the subfiles. Because data partitioning is made
transparently from users, data accessing is kept the same through
the conventional PnetCDF APIs and the master file.
* To enable this feature at configure time, add configure option
"--enable-subfiling".
- Update to version 1.4.0
o New APIs
* Fortran 90 APIs (adopted from netcdf-fortran-4.2). All F90 APIs have
prefix name "nf90mpi_". The APIs support function overloading.
* get/put_varn_<type> for reading/writing a list of subrequests (each is
specified by starts[i][ndims] and counts[i][ndims] for subrequest i.
ndims is the number of dimension of the variable) to a single variable.
* multiple put/get requests with explicit buffer type names:
ncmpi_mput_var_type(), ncmpi_mput_var1_type(), ncmpi_mput_vara_type(),
ncmpi_mput_vars_type(), ncmpi_mput_varm_type(). Similar for get APIs.
"type" is one of the followings: text, schar, uchar, short, ushort, int,
uint, long, float, double, longlong, or ulonglong.
* ncmpi_inq_nreqs() reports the number of pending nonblocking requests
* ncmpi_inq_header_size() reports the size of the file header
* ncmpi_inq_header_extent() reports the space currently allocated for the
file header, (also the file offset of the first variable)
* ncmpi_inq_put_size() reports the write amount committed by far
* ncmpi_inq_get_size() reports the read amount committed by far
* ncmpi_sync_numrecs() a collective API that can be called in independent
data mode to synchronize the number of records in memory across all
processes, and update to the file if NC_SHARE is set.
o Syntax change for Fortran put APIs
* intent of buffer argument in all Fortran APIs is changed to inout, as
byte-swap might be performed directly on the buffer. This change is
for performance consideration. For example, if the buffer is declared
as Fortran parameter, then compile will fail.
o New PnetCDF hint
* nc_header_read_chunk_size: PnetCDF reads the file headers in chunks. This
hint indicates the chunk size (in bytes). The default is 256 KB.
o New error code
* NC_EINTOVERFLOW reports the error of 4-byte integer overflow. This
usually happens due to MPI-IO data type constructor APIs' arguments using
4-byte integers.
* Error codes to report metadata defined inconsistently across processes.
o New run-time environment variables
* PNETCDF_SAFE_MODE environment variable can be used to enable/disable the
internal checking for data/argument consistency across all processes (by
calling collective MPI_Allreduce). Set it to 1 to enable the checking.
Default is 0, i.e. disabled.
* PNETCDF_HINTS environment variable can be used to pass the I/O hints to
PnetCDF library. Hints include both PnetCDF and MPI-IO hints.
PNETCDF_HINTS is a string of hints separated by ";" and each hint is in
the form of hint=value. E.g.
romio_ds_write=disable;nc_header_align_size=1048576
If this environment variable is set, it overrides any values that
were set by using calls to MPI_Info_set in the application code.
o New example programs
* example programs are now categorized into C, F77, and F90 directories
* nonblocking_write.f and nonblocking_write.f90 are the Fortran version of
nonblocking_write.c
* put_varn_float.c for using the new APIs ncmpi_put_varn_float_all()
put_varn_real.f and put_varn_real.f90 are the Fortran versions
* put_varn_int.c, put_varn_int.f, and put_varn_int.f90, for using the new
APIs ncmpi_put_varn_int_all() and nfmpi_put_varn_int_all()
* hints.c, hints.f, and hints.f90 for using PnetCDF hints
* flexible_api.c, flexible_api.f, and flexible_api.f90, for using blocking
and nonblocking flexible APIs
* mput.c for using ncmpi_mput_vara_all() to write a series of arbitrary
start[] and count[]
* block_cyclic.c, block_cyclic.f, and block_cyclic.f90 are for a
* -(block-cyclic) 2D partitioning pattern
* column_wise.c, for a *-cyclic 2D partitioning pattern
* put_vara.c, put_vara.f, and put_var.f90 for using for
nfmpi_put_vara_int_all()
o New programs for I/O benchmarks
* benchmarks/C/aggregation.c -- evaluate PnetCDF's performance on I/O
aggregation across multiple requests with different data access patterns.
* benchmarks/C/write_block_read_column.c -- writes variables and reads back
using different data partitioning patterns
* benchmarks/FLASH-IO -- I/O kernel of FLASH, a reacting hydrodynamics code
developed at University of Chicago. This benchmark can be built
independently from PnetCDF release.
o New test program
* test/F90 contains test programs adopted from netcdf-fortran-4.2
* test/nf90_test contains test programs adopted from test/nf_test
* testcases/alignment_test.c -- test for header and fixed variable file
offset alignments when entering redef mode
* testcases/nonblocking.c -- test nonblocking APIs ncmpi_iput_vara_int()
and ncmpi_iget_vara_int()
* testcases/flexible.c -- test flexible API ncmpi_get_vara_int_all() using
an MPI derived data type created by MPI_Type_create_hindex()
* test/header/header_consistency.c -- test header inconsistency and see if
inconsistent metadata is overwritten by root's
o New semantics for attribute APIs in data mode
o New synchronization for number of records
o New optimization: I/O request aggregation
pointer).
o Other updates:
* configure.in and Makefile.in have been revised to detect MPI compilers
and other compile options automatically.
* A new configure option "--disable-file-sync" to disable calling file
sync. This is to be used when the underlying file system provides data
consistency control.
* add build recipe for IBM BGQ (e.g. Vesta/Mira/Cetus @ANL) in README.IBM
* add build recipe for CRAY-XE6, Hopper @NERSC in README.CRAY
* add build recipe for CRAY-XC30, Edison @NERSC in README.CRAY
* add build recipe for Endeavour @ NASA in README.SGI
* add declaration of flexible APIs for Fortran90
* "make testing" now hides most of the stdout. Use "make verbose_testing"
for verbose output.
* ncmpidump: add the command-line option "-k" to report the format of
netCDF file.
* ncvalid is renamed to ncmpivalid, a tool to validate the structure of
netCDF files for conforming with CDF formats.
* Fortran type NFMPI_OFFSET is removed. It was merely a shortcut to
integer(KIND=MPI_OFFSET_KIND). Some Fortran 77 compiler does not
recognize keyword KIND. In that case, please set MPIF77 to the MPI
Fortran 90 compiler, e.g. ./configure MPIF77=mpif90
* configure now automatically checks Fortran module compile flags
* Support additional Fortran netCDF data types: nf_ubyte, nf_ushort,
nf_uint, nf_uint64, nf90_ubyte, nf90_ushort, nf90_uint, nf90_uint64.
* Error codes and messages conform with netCDF's
o Bug fixes
* Argument unlimdimid of nfmpi_inq() returns -1 when no unlimited length
dimension has been defined (to conform nf_inq()).
* Argument varid of nfmpi_inq_varoffset() is fixed to be the C's varid
plus one.
* For collective APIs, many places have been changed to prevent program
from hanging if a subset of the processes got errors. The fix will allow
all processes participating the MPI collective calls in the PnetCDF, even
if errors are detected on a subset of processes.
* set the nonblocking request ID to NULL when the request length is zero
* report error when bogus request IDs are passed in ncmpi_wait
* when entering redef with different alignment hints, fixed-size
variables' file starting offsets will only be changed when it is bigger
than the old ones
* Fix some Fortran API intent in/out argument declarations
* ncmpi_def_var is fixed to detect if CDF-5 data types is used on CDF-1 or
CDF-2 files. Error code NC_ESTRICTNC3 will return.
- Update to version 1.3.1 (September 24, 2012)
This release is primarily a bug-fix release, tidying up a few issues and
incorporating some early feedback on our "buffered put" interface (see
http://trac.mcs.anl.gov/projects/parallel-netcdf/wiki/BufferedInterface for
more information)
- add a new API ncmpi_inq_buffer_usage/nfmpi_inq_buffer_usage for inquiring
the current usage of the internal buffer allocated by the "buffered"-put
APIs.
- bug fix to make bput APIs properly return error code NC_EINSUFFBUF.
- bug fixes for ncmpidump to avoid residue contents from a previous read
when it read beyond EOF.
- bug fixes in the the tutorial example codes.
- add more in-line comments for the tutorial example codes.
- add the error string for error code NC_ENOENT.
- Replace parallel-netcdf-1.3.0-destdir.patch with
parallel-netcdf-1.6.1-destdir.patch. The patch accomplishes the
same thing but is created new for the substantially different
makefiles.