| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
| |
Clean up code a bit, move new "terminate" VFD callback to new position
in H5FD_class_t struct.
Tested on:
Mac OS X/32 10.6.7 (amazon) w/debug & production
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bring file free space branch changes through r15795 into trunk, which
includes a fair bit of code cleanup & rearrangement along with a couple of
bug fixes also.
Tested on:
Mac OS X/32 10.5.5 (amazon) in debug mode
Mac OS X/32 10.5.5 (amazon) w/C++ & FORTRAN, w/threadsafe,
in production mode
FreeBSD/32 6.3 (duty) in debug mode
FreeBSD/64 6.3 (liberty) w/C++ & FORTRAN, in debug mode
Linux/32 2.6 (kagiso) w/PGI compilers, w/C++ & FORTRAN, w/threadsafe,
in debug mode
Linux/64-amd64 2.6 (smirom) w/Intel compilers w/default API=1.6.x,
w/C++ & FORTRAN, in production mode
Solaris/32 2.10 (linew) w/deprecated symbols disabled, w/C++ & FORTRAN,
w/szip filter, in production mode
Linux/64-ia64 2.6 (cobalt) w/Intel compilers, w/C++ & FORTRAN,
in production mode
Linux/64-ia64 2.4 (tg-login3) w/parallel, w/FORTRAN, in production mode
Linux/64-amd64 2.6 (abe) w/parallel, w/FORTRAN, in production mode
|
|
|
|
|
|
|
|
|
| |
copyright notice.
Tested platform:
Kagiso only since it is only a comment block change. If it works in one
machine, it should work in all, I hope. Still need to check the parallel
build on copper.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
optimization codes:
1. Provide another option for users to do independent IO with MPI file setview(collectively)
2. With the request of collective IO from users, using Independent IO with MPI file setview if we find collective IO is not good for the applications for IO per chunk(multi-chunk IO) case. Previously we used pure independent IO and that actually performed small IO(IO each row) for this case. The recent performance study suggested the independent IO with file setview can acheieve significantly better performance than collective IO when not many processes participate in the IO.
3. For applications that explicitly choose to do collective IO per chunk case, the library won't do any optimization(gather/broadcast) operations. The library simply passes the collective IO request to MPI-IO.
Tested at copper, kagiso, heping, mir and tungsten(cmpi and mpich)
Kagiso is using LAM, t_mpi test was broken even.
The cchunk10 test failed at heping and mir. I suspected it was an MPICH problem. Will investigate later.
Everything passed at copper.
at tungsten: the old cmpi bug(failed at esetw) is still there. Other tests passed.
Some sequential fheap tests failed at kagiso.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
New APIs to add for collective chunk IO
Description:
Three new APIs
H5Pset_dxpl_mpio_chunk_opt_ratio
H5Pset_dxpl_mpio_chunk_opt_num
H5Pset_dxpl_mpio_chunk_opt
for optional optimization choices from users.
Solution:
Haven't added tests yet, won't affect other parts of the library.
Will add tests after urgent investigations of memory leaking problems from NASA Aura team.
Platforms tested:
heping: both parallel and sequential
shanti
Misc. update:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Code cleanup
Description:
Add destructor to match constructor fr VFLs when they are shut down by the
library.
Solution:
Added H5FD_*_term() routines to "undo" changes made in H5FD_*_init()
routines.
Platforms tested:
IBM p690 (copper)
too minor to require h5committest
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug fix/optimization
Description:
Address slowdown in MPI-I/O file metadata operations that was introduced
mid-stream. We now _require_ a POSIX compliant parallel file system for the
MPI-I/O file driver (as well as for the MPI-POSIX file driver).
Also optimized file open operation when the file is being created by
reducing the number of collective & syncronizing calls.
Additionally, refactor the MPI routines into a common place, eliminating
duplicated code.
Platforms tested:
FreeBSD 4.9 (sleipnir) w/parallel
h5committest
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug fix & code cleanups
Description:
Change our use of MPI derived datatypes to not create datatypes with
"0-sized" lengths, which causes the LANL Q machine to hang.
Also, get rid of "prefer MPI derived datatypes" environment variable since
it has no advantage.
Platforms tested:
FreeBSD 4.9 (sleipnir) w & w/o parallel
h5committest
|
|
|
|
|
|
|
|
|
|
|
|
| |
Update
Description:
Updated copyright statement in files which hadn't been updated yet.
Platforms tested:
Linux (Only comment change)
Misc. update:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug Fix
Description:
Metadata cache in parallel I/O can cause hangs in applications which
perform independent I/O on chunked datasets, because the metadata cache
can attempt to flush out dirty metadata from only a single process, instead
of collectively from all processes.
Solution:
Pass a dataset transfer property list down from every API function which
could possibly trigger metadata I/O.
Then, split the metadata cache into two sets of entries to allow dirty
metadata to be set aside when a hash table collision occurs during
independent I/O.
Platforms tested:
Tested h5committest {arabica (fortran), eirene (fortran, C++)
modi4 (parallel, fortran)}
FreeBSD 4.7 (sleipnir) serial & parallel
Misc. update:
Updated release_docs/RELEASE
|
|
|
|
|
|
|
|
|
|
|
|
| |
Purpose:
__DLL__ is a keyword in some platforms and __DLL__ is also defined as a macro for windows DLL applications.
That causes problems.
Description:
Solution:
Use H5_DLL*** to replace __DLL***__ at all header files.
Change the macro defination at H5api_adpt.h.
Platforms tested:
linux2.2.18smp, irix64, solaris 2.7 and windows 2000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug fix/Code cleanup/New Feature
Description:
Correct problems with writing fill-values to external storage and allocate
the data storage at the correct times.
Also, mostly straighten out the strange code which allocates and fills
raw data storage for datasets. Things are still a bit odd in that the
fill-values for chunked datasets are written when the space is allocated,
instead of in a separate routine, but there are two reasons for this:
it's inefficient (especially in parallel) to iterate through all the chunks
twice, and (more importantly) the space needed to store compressed chunks
isn't known until we've got a buffer of compressed fill-values ready to
write to the chunk.
Additionally, add in the H5D_SPACE_ALLOC_INCR and H5D_SPACE_ALLOC_DEFAULT
setting for the "space time", which incorporate the previous behavior of
the space allocation for chunked datasets.
The default settings for the different types of dataset storage are now
as follows:
Contiguous - Late
Chunked - Incremental
Compact - Early
This checkin also incorporates a change to the behavior of external data
storage in two ways - fill-values are _never_ written to external storage
(under the assumption that writing fill-values is triggered by allocating
space in an HDF5 file, and since space is not allocated in the file, the
fill-values should not be written) and external data files are now created
if they don't exist when data is written to them. The fill-value will
probably need to be revisited at some time in the future, this just seemed
like the safer course currently.
I think I cleaned up some compiler errors also, before getting bogged down
in the fixes for the space allocation and fill-values.
Platforms tested:
FreeBSD 4.6 (sleipnir) w/serial & parallel. Will be testing on IRIX64
6.5 (modi4) in serial & parallel shortly.
|
|
|
|
|
|
|
|
|
|
|
| |
Code improvement
Description:
Some small code cleanups and took out the code the was turning off the
metadata cache for parallel I/O (!)
Platforms tested:
IRIX64 6.5 (modi4) w/parallel
|
|
|
|
|
|
|
|
|
|
|
|
| |
Code optimization
Description:
Avoid creating MPI types (and thus requiring a MPI_File_set_view() call)
when contiguous selections are used for dataset I/O. This should be a
performance improvement for those sorts of selections.
Platforms tested:
Linux 2.2.x (eirene) w/parallel && IRIX64 6.5 (modi4) w/parallel & FORTRAN
|
|
|
|
|
|
|
|
|
|
|
|
| |
Code cleanup
Description:
Use dataset transfer property list to hold information about the MPI types
for the current transfer, instead of setting pseudo-global variables in
the file's struct.
Platforms tested:
Linux 2.2.x (eirene) w/parallel & IRIX64 6.5 (modi4) w/parallel & FORTRAN
|
|
|
|
|
|
|
|
|
|
|
|
| |
Code cleanup
Description:
Change MPI-I/O code to use the address of the dataset for the displacement,
instead of having a separate displacement value. Removed displacement
parameter from H5FD_mpio_setup parameters.
Platforms tested:
Linux 2.2.x (eirene) w/parallel & IRIX64 6.5 (modi4) w/parallel.
|
|
|
|
|
|
|
|
|
|
|
| |
New feature
Description:
Add 'closing' parameter to H5FDflush and VFL "flush" functions, per
http://hdf.ncsa.uiuc.edu/RFC/VFLFlush/VFLFlush.html
Platforms tested:
IRIX64 6.5 (modi4)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug fix/Code improvement.
Description:
Currently, the chunk data allocation routine invoked to allocate space for
the entire dataset is inefficient. It writes out each chunk in the dataset,
whether it is already allocated or not. Additionally, this happens not
only when it is created, but also anytime it is opened for writing, or the
dataset is extended. Worse, there's too much parallel I/O syncronization,
which slows things down even more.
Solution:
Only attempt to write out chunks that don't already exist. Additionally,
share the I/O writing between all the nodes, instead of writing everything
with process 0. Then, only block with MPI_Barrier if chunks were actually
created.
Platforms tested:
IRIX64 6.5 (modi4)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Performance enhancement
Description:
Doing an MPI_File_sync() just before a file is closed causing a large
performance loss.
Solution:
Add flag to MPI file driver to avoid performance the MPI_File_sync() when
the flag is set before a call to H5F_flush().
Platforms tested:
IRIX64 6.5 (modi4)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Code cleanup
Description:
The parallel I/O file driver is optimized to only write metadata with one
process (and broadcast the results to the other processes). This is
currently enabled by a separate call to H5FD_mpio_tas_allsame() before
each metadata write to the file. This can easily lead to problems where
the prelude function call is omitted before the actual write code or, in
a threaded environment, lead to race condititions where the value set is
reset before being used.
Solution:
Since we only want to write metadata from one process, key off of the 'type'
parameter (which has information about whether the data being written it
metadata or raw data) to H5FD_mpio_write() as the method for determining
whether to only write from one process or not.
Platforms tested:
IRIX64 6.5 (modi4)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug fix
Description:
This file is not C++ friendly/compliant because the protocols
are not bracketed by the #ifdef __cplusplus macro. This was
discovered by a user attempting to use C++ with enable-parallel.
Solution:
Though we are not supporting C++ in parallel mode yet, it is
simple to add the macro bracket. It would help if C++ is supported
in the future.
Platforms tested:
modi4 (64,n32) and eirene (mpich).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Update
Description:
Changed
#include <hdf_file.h>
construct to
#include "hdf_file.h"
so that the GNU compiler can more easily pick up the dependencies
which it places in the .depend and Dependencies files. Also
regenerated the Dependencies to go along with this.
Platforms tested:
Linux
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug Fix
Description:
When parallel I/O is turned on, there were some macros used in the H5D
routines which poked around in the H5F_t structure. This breaks the
privacy of that structure and ties the H5D code too tightly to the H5F_t
struct.
Solution:
Added a small function to retrieve the the value (driver_id) needed from
the H5F_t function.
Platforms tested:
Eyeballed only, Albert needs this right away...
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bug fix for parallel mode.
Description:
H5FD_mpio_tas_allsame was called for all cases, even when
MPIO is not used for access. That corrupted the internal
file handle structure.
Solution:
Define a macro, IS_H5FD_MPIO(f), for testing if f is opened with
MPIO access. Will call H5FD_mpio_tas_allsame only if this
condition is true.
Platform tested:
O2K, both -64 and -n32 modes.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
----------------------
./src/H5public.h
We undefine a bunch of things that could get redefined in the config
file because some customers have applications that include headers
from multiple packages, all of which might be using autoconf.
Include <stdint.h> for the C9x types.
./test/h5test.h
More flushing of stdout for when testing is redirected down a pipe.
./tools/h5ls.c
Added a `-S' or `--simple' switch which causes the output to be
simplified somewhat for easier parsing by other scripts. For instance,
characters are escaped using a very simple mechanism instead of C's
more complicated backslash notation, data doesn't have `{}' or `[]'
characters interspersed for compound and array types, and data is
printed with exactly one element per line. This switch is now used by
an HDF5-to-HTML CGI script being developed for the DMF people.
./tools/h5tools.c
./tools/h5tools.h
The repeat threshold which controls how strings are printed when a
character repeats a bunch of times is now settable at runtime instead
of compile time. The default is to show all characters, like
"abceeeeeeeeeeeeeeeeeeeeeeeeeeeeeefgh"
But if you set it to something like 5 then any sequence of 5 or more
characters is replaced by something shorter, like:
"abc" 'e'x30 "fgh" or
Added an `str_locale' property which describes how to escape special
characters in strings. The default is C-like escapes but an
alternative is ESCAPE_HTML which replaces all non-alphanumeric
characters with a 3-character HTML escape of the form `%XX'
Fixed a bug where empty strings didn't even have the quote characters
printed. Now empty strings show up as `""' instead of absolutely
nothing.
Added a `per_line' property which controls the maximum number of
elements which will appear per line of output. The default is infinity
but in practice the right margin causes line breaks. By setting the
`per_line' value to one and the right margin to a very large value one
can achieve output with exactly one element per line.
|
|
|
|
| |
It is working now. Still need to tie up some loose ends.
|
|
|
|
| |
can see it:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
----------------------
./src/H5D.c
./src/H5F.c
./src/H5FDmpio.c
./src/H5Fistore.c
./src/H5Fprivate.h
./src/H5Smpio.c
The `driver_id' for a file was accidently put in two
structs. I removed it from the H5F_file_t struct since it's
really an attribute of the VFL stuff.
More careful incrementing/decrementing the driver ID to fix a
memory leak.
./src/H5P.c
Rewrote H5Pcreate() in terms of H5P_copy() of a default
property list. This fixes some referencing counting bugs.
|
|
----------------------
This extensive change is the virtual file layer implementation. I've
ported and tested the sec2, family, and core drivers and only ported
the mpio driver (Albert will test it). So if you need MPIO I would
recommend sticking with the previous version for a while.
You will get a few compile warnings about split and stdio drivers not
being implemented and possibly tracing information not inserted in
some of the drivers. You can safely ignore them but I plan to fix
them.
I'm still working on the split driver because I just realized that it
needs a part of the VFL that isn't written yet.
Documentation is being updated also because there were some minor
changes (mostly just name changes). It should be available on my web
site later this week.
./MANIFEST
./src/Makefile.in
./src/hdf5.h
./src/H5Flow.c [REMOVED]
./src/H5Fstdio.c [REMOVED]
./src/H5Fsec2.c [REMOVED]
./src/H5Fsplit.c [REMOVED]
./src/H5Fmpio.c [REMOVED]
./src/H5Ffamily.c [REMOVED]
./src/H5Fcore.c [REMOVED]
./src/H5MFpublic.h [REMOVED]
./src/H5FD.c [NEW]
./src/H5FDcore.c [NEW]
./src/H5FDcore.h [NEW]
./src/H5FDfamily.c [NEW]
./src/H5FDfamily.h [NEW]
./src/H5FDmpio.c [NEW]
./src/H5FDmpio.h [NEW]
./src/H5FDprivate.h [NEW]
./src/H5FDpublic.h [NEW]
./src/H5FDsec2.c [NEW]
./src/H5FDsec2.h [NEW]
Removed/added files for virtual file layer.
./bin/trace
./src/H5.c
Removed unused public datatypes and added new VFL public
datatypes.
Changed an error message.
./config/BlankForm
./config/dec-flags
./config/gnu-flags
./config/hpux10.20
./config/hpux9.03
./config/irix5.x
./config/irix6.x
./config/solaris2.x
./config/unicosmk
Removed the H5F_OPT_SEEK and H5F_LOW_DFLT constants from the
configuration since they're no longer applicable. The default
file driver is always the sec2 driver and it always optimizes
calls to lseek() or lseek64().
./config/depend.in
C preprocessor errors generated during automatic dependency
building are sent to /dev/null to prevent them from appearing
twice in the make output.
./src/H5AC.c
./src/H5B.c
./src/H5D.c
./src/H5F.c
./src/H5G.c
./src/H5Gent.c
./src/H5Gnode.c
./src/H5HG.c
./src/H5HL.c
./src/H5O.c
./src/H5Oattr.c
./src/H5Odtype.c
./src/H5Oefl.c
./src/H5Oshared.c
./src/H5T.c
./src/H5detect.c
./test/ohdr.c
Changed H5F_ADDR_UNDEF to HADDR_UNDEF to be more consistent
with the `haddr_t' datatype which is now a public type.
./src/H5D.c
./src/H5P.c
./src/H5Ppublic.h
./src/H5Tconv.c
./test/cmpd_dset.c
./test/dsets.c
./test/overhead.c
./test/tselect.c
./test/tvltypes.c
The H5P_DATASET_XFER constant was changed to H5P_DATA_XFER
because the properties apply to all types of I/O operations,
not just datasets.
./src/H5B.c
./src/H5Bprivate.h
./src/H5D.c
./src/H5Dpublic.h
./src/H5F.c
./src/H5Farray.c
./src/H5Fistore.c
./src/H5Fprivate.h
./src/H5Fpublic.h
./src/H5Gnode.c
./src/H5Gpkg.h
./src/H5HG.c
./src/H5HL.c
./src/H5O.c
./src/H5R.c
./src/H5Sall.c
./src/H5Shyper.c
./src/H5Smpio.c
./src/H5Spoint.c
./src/H5Sprivate.h
./test/big.c
./test/h5test.c
./test/istore.c
./testpar/t_dset.c
./testpar/t_file.c
./tools/h5debug.c
./tools/h5ls.c
Modified to work with the virtual file layer by calling H5FD_*
functions instead of H5F_low_* functions and by passing file
access and data transfer properties by object ID instead of
pointer.
Changed H5D_transfer_t to H5FD_mpio_xfer_t since the
COLLECTIVE vs. INDEPENDENT transfer mode is specific to the
MPIO file driver.
Moved MPIO-specific stuff into the MPIO driver.
./src/H5B.c
./src/H5D.c
./src/H5Fprivate.h
The H5F_mpio_* private functions were renamed and placed in
the H5FDmpio driver except those which appeared in H5Smpio.c.
./src/H5E.c
./src/H5Epublic.h
Added major error number H5E_VFL for virtual file layer
related errors.
./src/H5F.c
./src/H5Fprivate.h
Changed the logic that controls whether the boot block is
written. Instead of assuming that the first call to write the
boot block is only to allocate space, I've added a function
argument which makes this explicit.
Changed the way files are compared so that a driver-defined
comparison function can be called. Files which belong to
different drivers are always considered different.
Removed H5F_driver_t since file drivers are now identified by
object ID instead of a special non-user-extendible datatype.
Removed all the hard-coded low-level file properties which
have been replaced by the various file drivers.
./src/H5I.c
./src/H5Iprivate.h
Added the H5I_inc_ref() which was removed a few months ago
since we finally have a use for it.
./src/H5Ipublic.h
Added the H5I_VFL object ID type to identify file drivers in
the virtual file layer.
./src/H5MF.c
./src/H5MFprivate.h
Moved all the allocation/deallocation code into the virtual
file layer which allows file drivers to override much of it.
./src/H5P.c
./src/H5Ppublic.h
Moved file driver-specific code into the various file driver
files.
The H5Pcopy() and H5Pclose() functions make calls into the
virtual file driver to manage the memory for driver-specific
file access and data transfer properties.
./src/H5private.h
./src/H5public.h
The `haddr_t' type is now public.
./test/tfile.c
Added a few more comments.
|