| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
| |
Description:
Added flush dependencies between object header chunks. Also added a new object
header proxy object, which is a child dependency of all object header chunks.
All objects which are child dependencies of an object header should set up a
flush dependency on the object header proxy, so the dependency applies to all
object header chunks.
Tested: ummon; jam (fails ph5diff -v, unrelated), koala, ostrich (h5committest)
|
| |
|
|
|
|
|
|
| |
Bring r22251:22599 from trunk to revise_chunks branch.
Tested on:
Mac OSX/64 10.7.4 (amazon) w/gcc 4.7.1, C++ & FORTRAN
(Not h5committested yet, as this branch doesn't require it)
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
Description:
Adds SWMR support to the v2 b-trees when used as a chunk index. Because each
node keeps track of the total number of records reachable through each of its
children, nodes must be shadowed every time a descendent record is inserted or
removed. This implementation prevents this from happening, however, if the node
has already been shadowed since the last time the v2 b-tree header was flushed.
Also modified SWMR test to include v2 b-trees.
Tested: jam, koala, ostrich (h5committest), durandal
Note: There is a preexisting failure in the ph5diff (-v) test
|
| |
|
|
|
|
|
|
|
|
| |
Bring r20557:22085 from trunk to this branch, also fixing some other
issues/failures in the branch simultaneously. The h5repack tests are still
failing, but Neil will be checking into those, so the branch can be fully
functional again.
Tested on:
Mac OSX/64 10.7.3 (amazon) w/debug
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Description:
Two issues were found with the chunk cache's SWMR implementation:
1. Some places were not properly setting the rdcc field in the index udata
struct. Fixed. May eventually need to add an rdcc field to the idx_info
struct.
2. While recalculating the indices for chunks in a dataset being resized, it was
possible for the chunk cache to attempt to flush a chunk before all the
indices were updated, which caused problems when a node split and flush
dependencies needed to be updated, because the child nodes could not be found
in the chunk cache due to the index being out of date. Modified algorithm in
H5D_chunk_update_cache to keep a temporary list of entries that got "bumped",
and only evict them after all chunks have updated indices. Also modified
H5D_chunk_update_flush_dep and H5D_chunk_update_flush_dep to search this
temporary list when looking for the child chunk entry. As a side effect,
chunks are now more likely to remain in cache after a call to H5Dset_extent
(even without SWMR), so performance should improve slightly.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A) - Changes to add the new tool "h5watch" to High Level
- Changes to add three new High Level public routines for retrieving elements appended to a dataset
B) A patch from Mike M. to fix the H5Dflush/refresh problem:
src/H5Dpkg.h
src/H5Dint.c
src/H5D.c
C) Fix for an assertion failure in H5FS_sect_link_size() in H5FSsection.c
src/H5Dbtree2.c
h5committested; tested also on linew, duty, fred.
|
| |
|
|
|
|
|
|
| |
Minor code cleanups.
Tested on:
Mac OS X/32 10.6.4 (amazon) w/debug & production
(h5committest not required on this branch)
|
|
|
with >1 unlimited dimensions.
h5committested.
|
