[svn-r24730] Purpose: Fix bug and simplify code for multi-dataset I/O
Description:
When a single element is selected in a chunked dataset,the multi dataset I/O
code wouldn't add the chunk to the skip list. The prevented it from being
included in collective operations. The test was only passing because it only
tested single element I/O with point selection, which breaks collective
currently.
Refactored related parts of the code, fixing various potential issues.
Tested: jam
[svn-r24723] Purpose: Fix bug and improve performance (processing time) for multi-dataset I/O
Description: The previous implementation used very inefficient linear searches
of a skip list in a few places. In addition, it did not check to see if a chunk
was in the same dataset as the one it was looking for, which would cause a bug.
Added a new skiplist for chunks for each dataset involved in a multi dataset I/O
operation, indexed by chunk index, to fix both of these issues.
Tested: jam
[svn-r24607] Description:
Initial round of review updates.
Tested on:
Mac OSX/64 10.9.1 (amazon) w/parallel & serial
(h5committest not required on this branch)
[svn-r24602] Description:
Bring changes from trunk, through r24601.
Tested on:
Mac OSX/64 10.9.1 (amazon) w/parallel & FORTRAN
(h5committest not required on this branch)
[svn-r24451]
Add initial multi-dset test cases.
This file has two type of tests. One is feature test, the other is performance test. It can be control by code 'TEST_TYPE'.
This is initial stage of development. Need to refactor to fit for daily test framework.
[svn-r24447]
Code clean up work:
Final code clean up. Remove all DEBUG developing (JK_DBG) code. (was left for debug purpose). Also minor update checking parallel io mode for count==0 case code.
[svn-r24435] Code clean up work:
Removed all the multi chunk optimization code. These code were already removed with '#if 0'. Now erase all the code as a group. (gives easier back trace later if necessary).
[svn-r24434] Code clean up work:
Removed the rest of multi chunk optimization code. However all the multi chunk optimization code is removed with '#if 0'. They will be removed at the next stage of code clean. (it give easier back trace later if necessary).
[svn-r24430] Code clean up work:
Remove development code which is not needed anymore. These were just kept for back trace purpose until now. Remove as things work as expected.
[svn-r24424]
As part of rewiring single-dset H5Dread/H5Dwrite via multi-dset Read/Write path functions, cutoff some duplicated single-dset Read/Write path functions as result of the rewiring work since the pathway now go through multi-dset Read/Wrtie path functions instead.
Note:
Debugging code and extra code are still remained for debug and back-trace purpose.
Extra code will be cleaned up at wrapping up stage later when all things work.
Debugging messages are turned off, so there is no de...
[svn-r24398]
The main implemetation for Read IO for multiple dataset with H5Dread_multi() in Parallel mode (collective and Independent IO) and Serial mode.
Note:
Debugging code and extra code are still remained with ifdefs.
Extra code will be cleaned up at wrapping up stage later when all things work.
Debugging messages are turned off, so there is no debugging print outs when run
tests.
[svn-r24254]
Part of work for rewiring single-dset Write IO path through multi-dset Write IO path. Removed some single-dset Wrtie path which will not be needed. (currently just #if 0 around for debug purpose, will be removed later)
[svn-r24199] Changes:
- minor changes with 'will REMOVE' code for easier distinguish with other #ifdef lines. don't remove them yet for back trace and debugging purpose.
[svn-r24191] CHANGES:
- Removed multi-chunk-opt code and tests. (#ifdef JK_MCHUNK_OPT_REMOVE, #ifndef JK_TODO_MCHUNK_OPT)
- Make H5Dwrite go through multi-dset path.
- Undo the patch from Quincey , which checked in previous check-in. Code still left with #ifdef JK_FCLOSE_PATH, thus no effect. Just keep it now for later test. This is for testing purpose and will be removed when merge to trunk.
[svn-r24099] Applied a patch from Quincey to make Fclose faster by not doing turncate as default. Wraped with #ifndef JK_FCLOSE_PATCH.
This is for testing purpose and will be removed when merge to trunk, since it is not official yet.
[svn-r24051] 1. Fix to set ACTUAL_IO correctly when CONTIG and CHUNK dsets mixed for H5Dwrite_multi. (H5Dmpio.c)
2. Update to handle number of CONTIG dsets correctly. (H5Dio.c)
3. minor changes (rest file)
[svn-r23892]
The main functionality of writing to multiple dataset via H5Dwrite_multi() works in Parallel mode (collective and Independent IO).
This stage was given to Peter for testing via DAMSEL test cases.
Some more branched/sided functionalities need to be verified and worked on.
Working code, extra code and debugging code are added with ifdefs.
Those may be reused when working multi-dset read feature.
Extra code will be cleaned up at wrapping up stage later when all things work.
Debugg...
[svn-r29216] HDFFV-9692: rework h5D read/write vlen strings.
Removed other types of vlen read/write functions and added test. Other types of vlen will be addedlater.
[svn-r29213] HDFFV-9624: Add missing APIs, H5Pinsert.
Changed insert2, iterate2, create_class to *_nocb because of Java issues without those types of callbacks.
