philipp/hdf5
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Files
555e79d05f3fb778c9345b3abd4e55a1f177df90
hdf5/release_docs/RELEASE.txt
jhendersonHDF b4090c7c1d Hdf5 1 10 parallel comp (#1644) 
* Hdf5 1 10 parallel compression merge
2022-04-14 16:19:02 -05:00

560 lines
22 KiB
Plaintext
Raw Blame History

HDF5 version 1.10.9-1 currently under development
================================================================================
INTRODUCTION
============
This document describes the differences between this release and the previous
HDF5 release. It contains information on the platforms tested and known
problems in this release. For more details check the HISTORY*.txt files in the
HDF5 source.
Note that documentation in the links below will be updated at the time of each
final release.
Links to HDF5 documentation can be found on The HDF5 web page:
https://portal.hdfgroup.org/display/HDF5/HDF5
The official HDF5 releases can be obtained from:
https://www.hdfgroup.org/downloads/hdf5/
Changes from Release to Release and New Features in the HDF5-1.10.x release series
can be found at:
https://portal.hdfgroup.org/display/HDF5/HDF5+Application+Developer%27s+Guide
If you have any questions or comments, please send them to the HDF Help Desk:
help@hdfgroup.org
CONTENTS
========
- New Features
- Support for new platforms and languages
- Bug Fixes since HDF5-1.10.8
- Supported Platforms
- Tested Configuration Features Summary
- More Tested Platforms
- Known Problems
- CMake vs. Autotools installations
New Features
============
Configuration:
-------------
- Added new option to the h5cc scripts produced by CMake.
Add -showconfig option to h5cc scripts to cat the
libhdf5-settings to the standard output.
(ADB - 2022/03/11)
- HDF5 memory allocation sanity checking is now off by default for
Autotools debug builds
HDF5 can be configured to perform sanity checking on internal memory
allocations by adding heap canaries to these allocations. However,
enabling this option can cause issues with external filter plugins
when working with (reallocating/freeing/allocating and passing back)
buffers.
Previously, this option was off by default for all CMake build types,
but only off by default for non-debug Autotools builds. Since debug
is the default build mode for HDF5 when built from source with
Autotools, this can result in surprising segfaults that don't occur
when an application is built against a release version of HDF5.
Therefore, this option is now off by default for all build types
across both CMake and Autotools.
(JTH - 2022/03/01)
- Refactored the utils folder.
Added subfolder test and moved the 'swmr_check_compat_vfd.c file'
from test into utils/test. Deleted the duplicate swmr_check_compat_vfd.c
file in hl/tools/h5watch folder. Also fixed vfd check options.
(ADB - 2021/10/18)
- Changed autotools and CMake configurations to derive both
compilation warnings-as-errors and warnings-only-warn configurations
from the same files, 'config/*/*error*'. Removed redundant files
'config/*/*noerror*'.
(DCY - 2021/09/29)
Library:
--------
- Several improvements to parallel compression feature, including:
* Improved support for collective I/O (for both writes and reads)
* Significant reduction of memory usage for the feature as a whole
* Reduction of copying of application data buffers passed to H5Dwrite
* Addition of support for incremental file space allocation for filtered
datasets created in parallel. Incremental file space allocation is the
default for these types of datasets (early file space allocation is
also still supported), while early file space allocation is still the
default (and only supported allocation time) for unfiltered datasets
created in parallel. Incremental file space allocation should help with
parallel HDF5 applications that wish to use fill values on filtered
datasets, but would typically avoid doing so since dataset creation in
parallel would often take an excessive amount of time. Since these
datasets previously used early file space allocation, HDF5 would
allocate space for and write fill values to every chunk in the dataset
at creation time, leading to noticeable overhead. Instead, with
incremental file space allocation, allocation of file space for chunks
and writing of fill values to those chunks will be delayed until each
individual chunk is initially written to.
* Addition of support for HDF5's "don't filter partial edge chunks" flag
(https://portal.hdfgroup.org/display/HDF5/H5P_SET_CHUNK_OPTS)
* Addition of proper support for HDF5 fill values with the feature
* Addition of 'H5_HAVE_PARALLEL_FILTERED_WRITES' macro to H5pubconf.h
so HDF5 applications can determine at compile-time whether the feature
is available
* Addition of simple examples (ph5_filtered_writes.c and
ph5_filtered_writes_no_sel.c) under examples directory to demonstrate
usage of the feature
* Improved coverage of regression testing for the feature
(JTH - 2022/2/23)
Fortran Library:
----------------
-
C++ Library:
------------
-
Java Library:
-------------
-
Tools:
------
- h5repack added an optional verbose value for reporting R/W timing.
In addition to adding timing capture around the read/write calls in
h5repack, added help text to indicate how to show timing for read/write;
-v N, --verbose=N Verbose mode, print object information.
N - is an integer greater than 1, 2 displays read/write timing
(ADB - 2022/04/01)
High-Level APIs:
----------------
-
C Packet Table API:
-------------------
-
Internal header file:
---------------------
-
Documentation:
--------------
-
Support for new platforms, languages and compilers
==================================================
-
Bug Fixes since HDF5-1.10.7 release
===================================
Library
-------
- Fixed a metadata cache bug when resizing a pinned/protected cache entry
When resizing a pinned/protected cache entry, the metadata
cache code previously would wait until after resizing the
entry to attempt to log the newly-dirtied entry. This would
cause H5C_resize_entry to mark the entry as dirty and make
H5AC_resize_entry think that it doesn't need to add the
newly-dirtied entry to the dirty entries skiplist.
Thus, a subsequent H5AC__log_moved_entry would think it
needs to allocate a new entry for insertion into the dirty
entry skip list, since the entry doesn't exist on that list.
This causes an assertion failure, as the code to allocate a
new entry assumes that the entry is not dirty.
(JRM - 2022/02/28)
- Issue #1436 identified a problem with the H5_VERS_RELEASE check in the
H5check_version function.
Investigating the original fix, #812, we discovered some inconsistencies
with a new block added to check H5_VERS_RELEASE for incompatibilities.
This new block was not using the new warning text dealing with the
H5_VERS_RELEASE check and would cause the warning to be duplicated.
By removing the H5_VERS_RELEASE argument in the first check for
H5_VERS_MAJOR and H5_VERS_MINOR, the second check would only check
the H5_VERS_RELEASE for incompatible release versions. This adheres
to the statement that except for the develop branch, all release versions
in a major.minor maintenance branch should be compatible. The prerequisite
is that an application will not use any APIs not present in all release versions.
(ADB - 2022/03/11, #1438)
- Unified handling of collective metadata reads to correctly fix old bugs
Due to MPI-related issues occurring in HDF5 from mismanagement of the
status of collective metadata reads, they were forced to be disabled
during chunked dataset raw data I/O in the HDF5 1.10.5 release. This
wouldn't generally have affected application performance because HDF5
already disables collective metadata reads during chunk lookup, since
it is generally unlikely that the same chunks will be read by all MPI
ranks in the I/O operation. However, this was only a partial solution
that wasn't granular enough.
This change now unifies the handling of the file-global flag and the
API context-level flag for collective metadata reads in order to
simplify querying of the true status of collective metadata reads. Thus,
collective metadata reads are once again enabled for chunked dataset
raw data I/O, but manually controlled at places where some processing
occurs on MPI rank 0 only and would cause issues when collective
metadata reads are enabled.
(JTH - 2021/11/16, HDFFV-10501/HDFFV-10562)
- Fixed several potential MPI deadlocks in library failure conditions
In the parallel library, there were several places where MPI rank 0
could end up skipping past collective MPI operations when some failure
occurs in rank 0-specific processing. This would lead to deadlocks
where rank 0 completes an operation while other ranks wait in the
collective operation. These places have been rewritten to have rank 0
push an error and try to cleanup after the failure, then continue to
participate in the collective operation to the best of its ability.
(JTH - 2021/11/09)
- Fixed an issue with collective metadata reads being permanently disabled
after a dataset chunk lookup operation. This would usually cause a
mismatched MPI_Bcast and MPI_ERR_TRUNCATE issue in the library for
simple cases of H5Dcreate() -> H5Dwrite() -> H5Dcreate().
(JTH - 2021/11/08, HDFFV-11090)
Java Library
------------
-
Configuration
-------------
- Reworked corrected path searched by CMake find_package command
The install path for cmake find_package files had been changed to use
"share/cmake"
for all platforms. However setting the HDF5_ROOT variable failed to locate
the configuration files. The build variable HDF5_INSTALL_CMAKE_DIR is now
set to the <INSTALL_DIR>/cmake folder. The location of the configuration
files can still be specified by the "HDF5_DIR" variable.
(ADB - 2022/03/11)
Tools
-----
-
Fortran API
-----------
-
High-Level Library
------------------
- Fixed HL_test_packet, test for packet table vlen of vlen.
Incorrect length assignment.
(ADB - 2021/10/14)
Fortran High-Level APIs
-----------------------
-
Documentation
-------------
-
F90 APIs
--------
-
C++ APIs
--------
-
Testing
-------
-
Supported Platforms
===================
Linux 3.10.0-1127.10.1.el7 gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-39)
#1 SMP ppc64 GNU/Linux g++ (GCC) 4.8.5 20150623 (Red Hat 4.8.5-39)
(echidna) GNU Fortran (GCC) 4.8.5 20150623 (Red Hat 4.8.5-39)
Linux 2.6.32-754.31.1.el6 IBM XL C/C++ V13.1
#1 SMP ppc64 GNU/Linux IBM XL Fortran V15.1
(ostrich)
Linux 3.10.0-327.18.2.el7 GNU C (gcc), Fortran (gfortran), C++ (g++)
#1 SMP x86_64 GNU/Linux compilers:
(jelly/kituo/moohan) Version 4.8.5 20150623 (Red Hat 4.8.5-4)
Version 4.9.3, Version 5.3.0, Version 6.3.0,
Version 7.2.0, Version 8.3.0, Version 9.1.0
Intel(R) C (icc), C++ (icpc), Fortran (icc)
compilers:
Version 17.0.0.098 Build 20160721
MPICH 3.3 compiled with GCC 7.2.0
OpenMPI 4.0.0 compiled with GCC 7.2.0
SunOS 5.11 11.4.5.12.5.0 Sun C 5.15 SunOS_sparc 2017/05/30
32- and 64-bit Studio 12.6 Fortran 95 8.8 SunOS_sparc 2017/05/30
(hedgehog) Sun C++ 5.15 SunOS_sparc 2017/05/30
Windows 10 x64 Visual Studio 2015 w/ Intel Fortran 18 (cmake)
Visual Studio 2017 w/ Intel Fortran 19 (cmake)
Visual Studio 2019 w/ Intel Fortran 19 (cmake)
Visual Studio 2019 w/ MSMPI 10.1 (cmake)
macOS Mojave 10.14.6 Apple LLVM version 10.0.1 (clang-1001.0.46.4)
64-bit gfortran GNU Fortran (GCC) 6.3.0
(swallow) Intel icc/icpc/ifort version 19.0.4.233 20190416
Tested Configuration Features Summary
=====================================
In the tables below
y = tested
n = not tested in this release
C = Cluster
W = Workstation
x = not working in this release
dna = does not apply
( ) = footnote appears below second table
<blank> = testing incomplete on this feature or platform
Platform C F90/ F90 C++ zlib SZIP
parallel F2003 parallel
Solaris2.11 32-bit n y/y n y y y
Solaris2.11 64-bit n y/n n y y y
Windows 10 y y/y n y y y
Windows 10 x64 y y/y n y y y
Mac OS X El Capitan 10.11.6 64-bit n y/y n y y y
Mac OS Sierra 10.12.6 64-bit n y/y n y y y
Mac OS X High Sierra 10.13.6 64-bit n y/y n y y y
Mac OS X Mojave 10.14.6 64-bit n y/y n y y y
CentOS 7.2 Linux 2.6.32 x86_64 PGI n y/y n y y y
CentOS 7.2 Linux 2.6.32 x86_64 GNU y y/y y y y y
CentOS 7.2 Linux 2.6.32 x86_64 Intel n y/y n y y y
Linux 2.6.32-573.18.1.el6.ppc64 n y/y n y y y
Platform Shared Shared Shared Thread-
C libs F90 libs C++ libs safe
Solaris2.11 32-bit y y y y
Solaris2.11 64-bit y y y y
Windows 10 y y y y
Windows 10 x64 y y y y
Mac OS X El Capitan 10.11.6 64-bit y n y y
Mac OS Sierra 10.12.6 64-bit y n y y
Mac OS X High Sierra 10.13.6 64-bit y n y y
Mac OS X Mojave 10.14.6 64-bit y n y y
CentOS 7.2 Linux 2.6.32 x86_64 PGI y y y n
CentOS 7.2 Linux 2.6.32 x86_64 GNU y y y y
CentOS 7.2 Linux 2.6.32 x86_64 Intel y y y n
Linux 2.6.32-573.18.1.el6.ppc64 y y y n
Compiler versions for each platform are listed in the preceding
"Supported Platforms" table.
More Tested Platforms
=====================
The following platforms are not supported but have been tested for this release.
Linux 2.6.32-573.22.1.el6 GNU C (gcc), Fortran (gfortran), C++ (g++)
#1 SMP x86_64 GNU/Linux compilers:
(platypus) Version 4.4.7 20120313
Version 4.9.3, 5.3.0, 6.2.0
PGI C, Fortran, C++ for 64-bit target on
x86-64;
Version 19.10-0
MPICH 3.1.4 compiled with GCC 4.9.3
Linux 2.6.32-754.31.1.el6 gcc (GCC) 4.4.7 20120313 (Red Hat 4.4.7-18)
#1 SMP ppc64 GNU/Linux g++ (GCC) 4.4.7 20120313 (Red Hat 4.4.7-18)
(ostrich) GNU Fortran (GCC) 4.4.7 20120313 (Red Hat 4.4.7-18)
Linux 3.10.0-327.18.2.el7 GNU C (gcc) and C++ (g++) compilers
#1 SMP x86_64 GNU/Linux Version 4.8.5 20150623 (Red Hat 4.8.5-4)
(jelly) with NAG Fortran Compiler Release 6.1(Tozai)
GCC Version 7.1.0
OpenMPI 2.1.6-GCC-7.2.0-2.29,
3.1.3-GCC-7.2.0-2.29
Intel(R) C (icc) and C++ (icpc) compilers
Version 17.0.0.098 Build 20160721
with NAG Fortran Compiler Release 6.1(Tozai)
Linux 3.10.0-327.10.1.el7 MPICH 3.1.4 compiled with GCC 4.9.3
#1 SMP x86_64 GNU/Linux
(moohan)
Linux-3.10.0-1127.0.0.1chaos openmpi-4.0.0
#1 SMP x86_64 GNU/Linux clang/3.9.0, 8.0.1
(quartz) gcc/7.3.0, 8.1.0
intel/16.0.4
Linux-4.14.0-115.10.1.1 spectrum-mpi/rolling-release
#1 SMP ppc64le GNU/Linux clang/coral-2018.08.08
(lassen) gcc/7.3.1
xl/2019.02.07
Linux-4.12.14-150.52-default cray-mpich/7.7.10
#1 SMP x86_64 GNU/Linux gcc/7.3.0, 8.2.0
(cori) intel/19.0.3
Linux-4.4.180-94.107-default cray-mpich/7.7.6
# 1SMP x86_64 GNU/Linux gcc/7.2.0, 8.2.0
(mutrino) intel/17.0.4, 18.0.2, 19.0.4
Fedora33 5.10.10-200.fc33.x86_64
#1 SMP x86_64 GNU/Linux GNU gcc (GCC) 10.2.1 20201125 (Red Hat 10.2.1-9)
GNU Fortran (GCC) 10.2.1 20201125 (Red Hat 10.2.1-9)
clang version 11.0.0 (Fedora 11.0.0-2.fc33)
(cmake and autotools)
Ubuntu20.10 5.8.0-41-generic-x86_64
#46-Ubuntu SMP x86_64 GNU/Linux GNU gcc (GCC) 10.2.0-13ubuntu1
GNU Fortran (GCC) 10.2.0-13ubuntu1
(cmake and autotools)
SUSE15sp2 5.3.18-22-default
#1 SMP x86_64 GNU/Linux GNU gcc (SUSE Linux) 7.5.0
GNU Fortran (SUSE Linux) 7.5.0
clang version 7.0.1 (tags/RELEASE_701/final 349238)
(cmake and autotools)
Mac OS X El Capitan 10.11.6 Apple clang version 7.3.0 from Xcode 7.3
64-bit gfortran GNU Fortran (GCC) 5.2.0
(osx1011test) Intel icc/icpc/ifort version 16.0.2
macOS Sierra 10.12.6 Apple LLVM version 9.0.0 (clang-900.39.2)
64-bit gfortran GNU Fortran (GCC) 7.4.0
(kite) Intel icc/icpc/ifort version 17.0.2
macOS High Sierra 10.13.6 Apple LLVM version 10.0.0 (clang-1000.10.44.4)
64-bit gfortran GNU Fortran (GCC) 6.3.0
(bear) Intel icc/icpc/ifort version 19.0.4.233 20190416
SunOS 5.11 11.3 Sun C 5.15 SunOS_sparc
32- and 64-bit Sun Fortran 95 8.8 SunOS_sparc
(emu)
Known Problems
==============
Setting a variable-length dataset fill value will leak the memory allocated
for the p field of the hvl_t struct. A fix is in progress for this.
HDFFV-10840
CMake files do not behave correctly with paths containing spaces.
Do not use spaces in paths because the required escaping for handling spaces
results in very complex and fragile build files.
ADB - 2019/05/07
At present, metadata cache images may not be generated by parallel
applications. Parallel applications can read files with metadata cache
images, but since this is a collective operation, a deadlock is possible
if one or more processes do not participate.
Two tests fail attempting collective writes with OpenMPI 3.0.0/GCC-7.2.0-2.29:
testphdf5 (ecdsetw, selnone, cchunk1, cchunk3, cchunk4, and actualio)
t_shapesame (sscontig2)
CPP ptable test fails on both VS2017 and VS2019 with Intel compiler, JIRA
issue: HDFFV-10628. This test will pass with VS2015 with Intel compiler.
The subsetting option in ph5diff currently will fail and should be avoided.
The subsetting option works correctly in serial h5diff.
Known problems in previous releases can be found in the HISTORY*.txt files
in the HDF5 source. Please report any new problems found to
help@hdfgroup.org.
CMake vs. Autotools installations
=================================
While both build systems produce similar results, there are differences.
Each system produces the same set of folders on linux (only CMake works
on standard Windows); bin, include, lib and share. Autotools places the
COPYING and RELEASE.txt file in the root folder, CMake places them in
the share folder.
The bin folder contains the tools and the build scripts. Additionally, CMake
creates dynamic versions of the tools with the suffix "-shared". Autotools
installs one set of tools depending on the "--enable-shared" configuration
option.
build scripts
-------------
Autotools: h5c++, h5cc, h5fc
CMake: h5c++, h5cc, h5hlc++, h5hlcc
The include folder holds the header files and the fortran mod files. CMake
places the fortran mod files into separate shared and static subfolders,
while Autotools places one set of mod files into the include folder. Because
CMake produces a tools library, the header files for tools will appear in
the include folder.
The lib folder contains the library files, and CMake adds the pkgconfig
subfolder with the hdf5*.pc files used by the bin/build scripts created by
the CMake build. CMake separates the C interface code from the fortran code by
creating C-stub libraries for each Fortran library. In addition, only CMake
installs the tools library. The names of the szip libraries are different
between the build systems.
The share folder will have the most differences because CMake builds include
a number of CMake specific files for support of CMake's find_package and support
for the HDF5 Examples CMake project.
Reference in New Issue View Git Blame Copy Permalink