| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
| |
Bring over rest of performance improvements for extending chunked datasets
and normalize against revise_chunks branch.
Tested on:
MacOSX/64 10.11.2 (amazon) w/serial & parallel)
(h5committest forthcoming)
|
|
|
|
|
|
|
|
| |
Minor change to align w/revise_chunks branch
Tested on:
MacOSX/64 10.11.2 (amazon) w/serial & parallel
(too minor for h5committest)
|
|
|
|
|
|
|
|
|
| |
Also purged obsolete /*ARGSUSED*/ from the library.
Tested on:
Ubuntu 15.10 (Linux 4.2.0 x86_64) gcc 5.2.1
serial only
(these changes have been in revise_chunks for a long time)
|
|
|
|
|
|
|
|
|
| |
Normalization for future single chunk indexing.
Tested on:
Ubuntu 15.10 (Linux 4.2.0 x86_64) gcc 5.2.1
serial only
(these changes have been in revise_chunks for a long time)
|
|
|
|
|
|
|
|
|
|
| |
Revisions to normalize revise chunks vs. trunk, plus push dataset metadata
changes into the cache immediately (instead of holding them within the dataset
data structures).
Tested on:
MacOSX/64 10.11.2 (amazon) w/serial & parallel
(h5committest not required on this branch)
|
|
|
|
|
|
|
|
| |
Normalization changes against revise_chunks branch.
Tested on:
MacOSX/64 10.11.1 (amazon) w/serial & parallel
(h5committest forthcoming)
|
|
|
|
|
| |
Add internal library callback to H5S_select_iterate to avoid having
to pass hid_t objects internally
|
|
|
|
|
|
|
|
|
|
|
| |
Refactor property list code to "deep copy" properties in the correct way,
retraining the rest of the library to copy & release things correctly. This
cleans up another batch of memory leaks, etc. within the library.
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial & parallel
Linux/32 2.6.x (jam) w/serial & parallel
(h5committest forthcoming)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Complete revamp of package initialization/shutdown mechanism in the library.
Each package now has a single init/term routine.
This new way should avoid packages being re-initialized during library
shutdown and is also be _much_ more proactive about giving feedback for
resource leaks internal to the library.
Introduces a new "module" header file for packages in the library
(e.g src/H5Fmodule.h) which sets up some necessary package configuration macros
for the FUNC_ENTER/LEAVE macros. (The VFL drivers have their own slightly
modified version of this header, src/H5FDdrvr_module.h)
Also cleaned up a bunch of resources leaks all across the library and tests,
along with addressing many warnings, as I encountered them.
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial & parallel
Linux/64 3.10.x (kituo) w/serial & parallel
Linux/64 2.6.x (ostrich) w/serial
|
|
|
|
|
|
|
|
|
|
| |
Align w/vds branch: Remove extraneous sieve buffer flush in chunk
flush callback, and move error check out of the middle of retrieving properties
for dataset.
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial & parallel
(h5committest forthcoming)
|
|
|
|
|
|
|
|
|
| |
Align w/vds branch: Move code for initializing contiguous datasets into
layout 'init' callback.
Tested on:
MacOSX/64 10.10.5 (amazon) w/serial & parallel
(h5committest forthcoming)
|
|
|
|
|
|
| |
selection as duplicate points are not checked to determine
that it is really an "entire chunk". The bug appears from running the mdset.c test. This is a short-term fix for the problem.
|
|
|
|
|
|
|
|
| |
* H5_ATTR_FORMAT(X,Y,Z) __attribute__((format(X, Y, Z)))
- Rename UNUSED attribute characterstic to H5_ATTR_UNUSED.
- Rename NORETURN attribute characterstic to H5_ATTR_NORETURN
tested with h5committest.
|
|
|
|
|
|
|
|
| |
Clean up H5D interface, to align w/v3 metadata cache changes
Tested on:
MacOSX/64 10.10.3 (amazon) w/serial & parallel
Linux/32 2.6.* (jam) w/serial & parallel
|
|
|
|
|
|
|
|
|
|
| |
Convert internal chunk structures to use 'scaled' coordinates instead of
absolute coordinates.
Tested on:
Mac OSX/64 10.10.3 (amazon) w/parallel & serial
Linux 2.6.x/32 (jam) w/parallel & serial
Linux 2.6.x/64 (koala) w/serial
|
|
|
|
|
|
| |
the arguments to be in a more logical order.
Tested on: h5committest
|
|
|
|
|
|
|
|
| |
Eliminate unneccessary 'rdcc' field from common chunk info struct.
Tested on:
MacOSX/64 10.10.3 (amazon) w/serial & parallel
(Too minor to require h5commitest)
|
|
|
|
|
|
|
|
|
|
| |
Clean up chunk 'prune by extent' and flush code, and make the rehashing
routine 'update cache' hold onto more chunks, instead of evicting them as
aggressively.
Tested on:
MacOSX/64 10.10.3 (amazon) w/serial & parallel
(h5committest forthcoming)
|
|
|
|
|
|
|
|
|
| |
Align w/incoming chunk performance changes, including some small performance
improvements here, as well as a few minor general code cleanups.
Tested on:
Mac OSX/64 10.10.3 (amazon) w/serial & parallel
(Too minor to require h5committest)
|
|
|
|
|
|
|
|
| |
Minor code & warning cleanups.
Tested on:
MacOSX/64 10.10.2 (amazon) w/serial & parallel
Linux/32 2.6.18 (jam) w/serial & parallel
|
|
|
|
|
|
|
|
|
|
| |
Cache the dataset's rank & dimension sizes, instead of querying them
frequently, to speed up various checks & algorithms. Also, a few minor
cleanups.
Tested on:
MacOSX/64 10.10.2 (amazon) w/serial & parallel
Linux/32 2.6.18 (jam) w/serial & parallel
|
|
|
|
|
|
|
|
| |
Minor formatting cleanups and remove unused field from callback struct.
Tested on:
Mac OSX/64 10.10.2 (amazon) w/serial & parallel
(Too minor to require h5committest)
|
|
|
|
|
|
|
|
|
|
| |
Separate allocating chunk on disk from inserting the chunk record into the
index. This allows a "SWMR-safe" insert/update of chunks (the chunk is always
allocated -> written -> inserted/updated in the index).
Tested on:
Mac OSX/64 10.10.2 (amazon) w/parallel & serial
Linux/32 2.6.18 (jam) w/serial & parallel
|
|
|
|
|
|
|
|
| |
Minor typos & code cleanups
Tested on:
MacOSX/64 10.10.2 (amazon) w/serial & parallel
(Too minor to require h5committest)
|
|
|
|
|
|
|
|
| |
Minor code cleanups, to align with revise_chunks branch.
Tested on:
Mac OSX/64 10.10.2 (amazon) w/serial & parallel
Linux/32 2.6.x (jam) w/serial
|
|
|
|
|
|
|
|
|
|
| |
Revert r25273, 25283 & 25439 (the hyperslab improvement changes). They
are buggy and it's taking me a long time to correct the problem. I'll check
in a revised form of the changes when I've got them straightened out.
Tested on:
Mac OSX 10.10.0 (amazon) w/gcc 4.9.x, C++, FORTRAN
Linux 2.6.x (jam) w/parallel
|
| |
|
|
|
|
|
|
| |
Make the fill value operation for chunked datasets in parallel collective/fast.
tested with h5committest
|
|
|
|
| |
allocating space for zero-length chunked dataset and asserting.
|
|
|
|
|
|
|
|
|
|
|
| |
Bring in Chao/Neil/my changes to optimize hyperslab selection operations
further, along with 3 new public API routines: H5Scombine_hyperslab(),
H5Sselect_select() and H5Scombine_select(), along with many minor cleanups to
the code and fixing a few compiler warnings.
Tested on:
Mac OSX/64 10.9.3 w/gcc 4.9.x and parallel w/OpenMPI
(h5commttest forthcoming)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Bring changes from hdf5_1_8_coverity branch back to trunk:
r20878:
Issue 76: Check if H5Tget_nmembers(type) fails and simply return(FALSE). Also move printf to after check.
r20880:
Issue 192: Create ret_val var set to -1. Add out label for failures to jump to, return ret_val at bottom.
r20882:
Fixes for coverity:
1) bug #1679: remove dead code in test/mf.c
2) bug #1680: remove dead code in tools/lib/h5diff_dset.c
r20883:
Fix coverity issue 585
Description:
Changed variable "c" in processStrData in h5import.c to an int, to match fgetc
return value, and removed call to feof, instead checking if c == EOF.
Tested on:
MacOSX/64 10.9.3 (amazon) w/C++, FORTRAN & parallel
(too minor to require h5committest)
|
|
|
|
|
|
|
|
|
| |
Check in Mohamad's changes to support collective I/O on point selections,
along with some other minor cleanups.
Tested on:
Mac OSX/64 10.9.2 (amazon) w/parallel & serial
(h5committest forthcoming)
|
|
|
|
|
|
|
|
|
|
|
|
| |
project for
view objects. The addition of view objects in the fastforward project
is expected to be brough into the trunk sometimes in the future, which
is why we need to make this change.
Tested Manually on Jam and Ostrich.
Tested with h5commitest - Koala with intel compilers failed, but nothing had to do with those changes.
error on Koala: error while loading shared libraries: libirng.so
|
|
|
|
|
|
|
|
|
|
|
| |
Tested: jam, koala, ostrich (h5committest)
Log from r23939:
Fix unused variables warnings in trunk
Rename static functions in H5Adense.c, H5B2cache.c, H5Bcache.c
Switch these functions to use FUNC_ENTER_STATIC* macros
Tested: koala
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Description:
Failure to write fill values to the user's buffer when reading unallocated
chunks from datasets that have a fill value set to H5D_FILL_VALUE_DEFAULT.
A consequence of this was the reporting of spurious data values in h5dump
and h5diff output.
Solution:
Added check for fill value of H5D_FILL_VALUE_DEFAULT whenever fill value
setting was checked to decide whether or not to fill. This effectively
treats H5D_FILL_VALUE_DEFAULT the same as H5D_FILL_VALUE_USER rather than
H5D_FILL_VALUE_UNDEFINED.
Tested:
h5committest (platypus, ostrich, jam-pp, koala) ran successfully
(fillval test was modified to reveal and check for this error)
Note: RELEASE.txt was previously changed to reflect this fix. This
checkin was of a cleaner version of the code, but mostly to get this
full log message entered.
-This line, and those below, will be ignored--
M H5Dchunk.c
|
|
|
|
|
|
|
|
|
|
| |
Clean up warnings, switch library code to use Standard C/POSIX wrapper
macros, remove internal calls to API routines, update checkapi and checkposix
scripts.
Tested on:
Mac OSX/64 10.8.3 (amazon) w/C++ & FORTRAN
Big-Endian Linux/64 (ostrich)
|
|
|
|
| |
allocated/unallocated chunks.
|
|
|
|
| |
tested on Jam - too minor for committest
|
|
|
|
|
|
| |
related test and performance test.
Tested on koala.
|
|
|
|
|
|
| |
also added more test cases.
Tested on koala.
|
|
|
|
|
|
|
|
|
| |
Dectris people to test performance. The library's API function H5PSIdirect_write
is still in. I reorganized the internal library by adding H5D__pre_write function.
Tested on koala.
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Changes resulting from Klocwork static analysis tool, from Mark Miller
@ LLNL (miller86@llnl.gov).
Tested on:
Mac OS X/64 10.7.4 (amazon) w/debug, C++ & FORTRAN, using gcc 4.7.x
(too minor to require h5committest)
|
| |
| |
| |
| |
| |
| | |
comments. I added a performance benchmark program dectris_perf.c in the test/ directory.
Tested on koala and jam.
|
|/
|
|
| |
Tested on koala and jam.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Description:
When jumping out from between H5_BEGIN_TAG and H5_END_TAG macros using
HGOTO_ERROR or HGOTO_DONE, the previous metadata tag is not reset on the dxpl.
This could cause problems when, for example, calling H5Ocopy within an
H5Literate callback.
Added new HGOTO_ERROR_TAG and HGOTO_DONE_TAG macros which must be used in place
of the above between H5_BEGIN_TAG and H5_END_TAG.
Tested:
jam, koala, ostrich (h5committest), durandal
|
|
|
|
|
|
|
|
| |
Clean up more FUNC_ENTER/FUNC_LEAVE macros and move H5D & H5T code toward
the final design (as exemplified by the H5EA & H5FA code).
Tested on:
Mac OSX/64 10.7.3 (amazon) w/debug & parallel
|
|
|
|
|
|
|
|
| |
memory and caused seg fault. I added checks in two places to make sure the library returns error stack
when it fails to allocate memory. I didn't add any test to the test suite since there is no good way to test it. But I tested and verified the error stack by hand.
Tested on jam, koala, ostrich.
|
|
|
|
|
|
|
|
| |
Refactor function name macros and simplify the FUNC_ENTER macros, to clear
away the cruft and prepare for further cleanups.
Tested on:
Mac OSX/64 10.7.3 (amazon) w/debug, production & parallel
|
|
|
|
|
|
|
|
|
|
|
|
| |
Description:
When shrinking a chunked dataset, the library fills in the unused parts of
chunks that have been shrunk. The fill value buffer allocated for this purpose
had a maximum size of 1 MB, but the fill was performed in a single operation.
Therefore, if the amount of unused space in a chunk after being shrunk was
greater than 1 MB, the library would read off the end of the fill value buffer.
Changed the maximum fill buffer size to be equal to the chunk size.
Tested: durandal; jam, koala, heiwa (h5committest)
|