Changelog for
blosc-devel-1.11.2-lp150.21.1.x86_64.rpm :
* Tue Apr 18 2017 toddrme2178AATTgmail.com- Update to 1.11.2 - Enabled use as a CMake subproject, exporting shared & static library targets for super-projects to use. See PRs #178, #179 and #180. Thanks to Kevin Murray. - Internal LZ4 codec updated to 1.7.5. - Internal Zstd codec updated to 1.1.2.
* Fri Sep 16 2016 toddrme2178AATTgmail.com- Update to 1.11.1 - Fixed a bug introduced in 1.11.0 and discovered by pandas test suite. This basically prevented to decompress buffers compressed with previous versions of C-Blosc. See: https://github.com/Blosc/python-blosc/issues/115- Update to 1.11.0 - Internal Zstd codec upgraded to 1.0.0. - New block size computation inherited from C-Blosc2. Benchmarks are saying that this benefits mainly to LZ4, LZ4HC, Zlib and Zstd codecs, both in speed and in compression ratios (although YMMV for your case). - Added the AATTrpath flag in Mac OSX for shared libraries. Fixes #175. - Added a fix for VS2008 - License changed from MIT to 3-clause BSD style.- Update to 1.10.2 - Force the use of --std=gnu99 when using gcc. Fixes #174.- Update to 1.10.1 - Removed an inconsistent check for C11 (__STDC_VERSION__ >= 201112L and _ISOC11_SOURCE) as this seem to pose problems on compilers doing different things in this check (e.g. clang). See https://github.com/Blosc/bloscpack/issues/50.- Update to 1.10.0 - Initial support for Zstandard (0.7.4). Zstandard (or Zstd for short) is a new compression library that allows better compression than Zlib, but that works typically faster (and some times much faster), making of it a good match for Blosc.- Update to 1.9.3 - Reverted a mistake introduced in 1.7.1. At that time, bit-shuffling was enabled for typesize == 1 (i.e. strings), but the change also included byte-shuffling accidentally. This only affected performance, but in a quite bad way (a copy was needed). This has been fixed and byte-shuffling is not active when typesize == 1 anymore.
* Thu Jul 07 2016 toddrme2178AATTgmail.com- Update to 1.9.2
* Check whether Blosc is actually initialized before blosc_init(), blosc_destroy() and blosc_free_resources(). This makes the library more resistant to different initialization cycles (e.g. https://github.com/stevengj/Blosc.jl/issues/19).- Update to 1.9.1
* The internal copies when clevel=0 are made now via memcpy(). At the beginning of C-Blosc development, benchmarks where saying that the internal, multi-threaded copies inside C-Blosc were faster than memcpy(), but 6 years later, memcpy() made greats strides in terms of efficiency. With this, you should expect an slight speed advantage (10% ~ 20%) when C-Blosc is used as a replacement of memcpy() (which should not be the most common scenario out there).
* Added a new DEACTIVATE_AVX2 cmake option to explicitly disable AVX2 at build-time. Thanks to James Bird.
* The ``make -jN`` for parallel compilation should work now. Thanks to James Bird.- Update to 1.9.0
* New blosc_get_nthreads() function to get the number of threads that will be used internally during compression/decompression (set by already existing blosc_set_nthreads()).
* New blosc_get_compressor() function to get the compressor that will be used internally during compression (set by already existing blosc_set_compressor()).
* New blosc_get_blocksize() function to get the internal blocksize to be used during compression (set by already existing blosc_set_blocksize()).
* Now, when the BLOSC_NOLOCK environment variable is set (to any value), the calls to blosc_compress() and blosc_decompress() will call blosc_compress_ctx() and blosc_decompress_ctx() under the hood so as to avoid the internal locks. See blosc.h for details. This allows multi-threaded apps calling the non _ctx() functions to avoid the internal locks in C-Blosc. For the not multi-threaded app though, it is in general slower to call the _ctx() functions so the use of BLOSC_NOLOCK is discouraged.
* In the same vein, from now on, when the BLOSC_NTHREADS environment variable is set to an integer, every call to blosc_compress() and blosc_decompress() will call blosc_set_nthreads(BLOSC_NTHREADS) before the actuall compression/decompression process. See blosc.h for details.
* Finally, if BLOSC_CLEVEL, BLOSC_SHUFFLE, BLOSC_TYPESIZE and/or BLOSC_COMPRESSOR variables are set in the environment, these will be also honored before calling blosc_compress().
* Calling blosc_init() before any other Blosc call, although recommended, is not necessary anymore. The idea is that you can use just the basic blosc_compress() and blosc_decompress() and control other parameters (nthreads, compressor, blocksize) by using environment variables (see above).- Update to 1.8.1
* Disable the use of __builtin_cpu_supports() for GCC 5.3.1 compatibility.- Update to 1.8.0
* The code is (again) compatible with VS2008 and VS2010. This is important for compatibility with Python 2.6/2.7/3.3/3.4.
* Introduced a new global lock during blosc_decompress() operation. As the blosc_compress() was already guarded by a global lock, this means that the compression/decompression is again thread safe. However, when using C-Blosc from multi-threaded environments, it is important to keep using the
*_ctx() functions for performance reasons. NOTE: _ctx() functions will be replaced by more powerful ones in C-Blosc 2.0.
* Wed Mar 09 2016 toddrme2178AATTgmail.com- Update to 1.7.1
* Fixed a bug preventing bitshuffle to work correctly on getitem(). Now, everything with bitshuffle seems to work correctly.
* Fixed the thread initialization for blosc_decompress_ctx(). Issue [#158]. Thanks to Chris Webers.
* Fixed a bug in the blocksize computation introduced in 1.7.0. This could have been creating segfaults.
* Allow bitshuffle to run on 1-byte typesizes.
* New parametrization of the blocksize to be independent of the typesize. This allows a smoother speed throughout all typesizes.
* lz4 and lz4hc codecs upgraded to 1.7.2 (from 1.7.0).
* When calling set_nthreads() but not actually changing the number of threads in the internal pool does not teardown and setup it anymore. PR #153. Thanks to Santi Villalba.- Update to 1.7.0
* Added a new \'bitshuffle\' filter so that the shuffle takes place at a bit level and not just at a byte one, which is what it does the previous \'shuffle\' filter.
* New acceleration mode for LZ4 and BloscLZ codecs that enters in operation with complevel < 9. This allows for an important boost in speed with minimal compression ratio loss. Francesc Alted.
* LZ4 codec updated to 1.7.0 (r130).
* PREFER_EXTERNAL_COMPLIBS cmake option has been removed and replaced by the more fine grained PREFER_EXTERNAL_LZ4, PREFER_EXTERNAL_SNAPPY and PREFER_EXTERNAL_ZLIB. In order to allow the use of the new API introduced in LZ4 1.7.0, PREFER_EXTERNAL_LZ4 has been set to OFF by default, whereas PREFER_EXTERNAL_SNAPPY and PREFER_EXTERNAL_ZLIB continues to be ON.
* Implemented SSE2 shuffle support for buffers containing a number of elements which is not a multiple of (typesize
* vectorsize). Jack Pappas.
* Added SSE2 shuffle/unshuffle routines for types larger than 16 bytes. Jack Pappas.
* \'test_basic\' suite has been split in components for a much better granularity on what\'s a possibly failing test. Also, lots of new tests have been added. Jack Pappas.
* Fixed compilation on non-Intel archs (tested on ARM). Zbyszek Szmek.
* Modifyied cmake files in order to inform that AVX2 on Visual Studio is supported only in 2013 update 2 and higher.
* Added a replacement for stdbool.h for Visual Studio < 2013.
* blosclz codec adds Win64/Intel as a platform supporting unaligned addressing. That leads to a speed-up of 2.2x in decompression.
* New blosc_get_version_string() function for retrieving the version of the c-blosc library. Useful when linking with dynamic libraries and one want to know its version.
* New example (win-dynamic-linking.c) that shows how to link a Blosc DLL dynamically in run-time (Windows only).
* The `context.threads_started` is initialized now when decompressing. This could cause crashes in case you decompressed before compressing (e.g. directly deserializing blosc buffers). AATTatchouprakov.
* The HDF5 filter has been removed from c-blosc and moved into its own repo at: https://github.com/Blosc/hdf5
* The MS Visual Studio 2008 has been tested with c-blosc for ensuring compatibility with extensions for Python 2.6 and up.- Remove update-included 0001-Fix-cmake-system-check.patch and 0002-Fix-compilation-on-unknown-archs.patch
* Mon Feb 15 2016 mkravecAATTsuse.com- Replaced long syntax with macros %cmake, %cmake_install, %ctest
* Fri Aug 07 2015 normandAATTlinux.vnet.ibm.com- Add two patches dragged from fedora to support non-intel archi 0001-Fix-cmake-system-check.patch 0002-Fix-compilation-on-unknown-archs.patch
* Tue May 19 2015 toddrme2178AATTgmail.com- Update to version 1.6.1:
* Fixed a subtle, but long-standing bug in the blosclz codec that could potentially overwrite an area beyond the output buffer.
* Support for runtime detection of AVX2 and SSE2 SIMD instructions, allowing running AVX2 capable c-blosc libraries to run on machines with no AVX2 available (will use SSE2 instead).
* A new blocksize computation allows for better compression ratios for larger typesizes (> 8 bytes), without not penalizing the speed too much (at least on modern CPUs).- changes from version 1.6.0:
* Support for AVX2 is here! The benchmarks with a 4-core Intel Haswell machine tell that both compression and decompression are accelerated around a 10%, reaching peaks of 9.6 GB/s during compression and 26 GB/s during decompression (memcpy() speed for this machine is 7.5 GB/s for writes and 11.7 GB/s for reads). Many thanks to AATTlittlezhou for this nice work.
* Support for HPET (high precision timers) for the `bench` program. This is particularly important for microbenchmarks like bench is doing; since they take so little time to run, the granularity of a less-accurate timer may account for a significant portion of the runtime of the benchmark itself, skewing the results. Thanks to Jack Pappas.- changes from version 1.5.4:
* Updated to LZ4 1.6.0 (r128).
* Fix resource leak in t_blosc. Jack Pappas.
* Better checks during testing. Jack Pappas.
* Dynamically loadable HDF5 filter plugin. Kiyo Masui.- changes from version 1.5.3:
* Use llabs function (where available) instead of abs to avoid truncating the result. Jack Pappas.
* Use C11 aligned_alloc when it\'s available. Jack Pappas.
* Use the built-in stdint.h with MSVC when available. Jack Pappas.
* Only define the __SSE2__ symbol when compiling with MS Visual C++ and targeting x64 or x86 with the correct /arch flag set. This avoids re-defining the symbol which makes other compilers issue warnings. Jack Pappas.
* Reinitializing Blosc during a call to set_nthreads() so as to fix problems with contexts. Francesc Alted.- changes from version 1.5.2:
* Using blosc_compress_ctx() / blosc_decompress_ctx() inside the HDF5 compressor for allowing operation in multiprocess scenarios. See: https://github.com/PyTables/PyTables/issues/412 The drawback of this quick fix is that the Blosc filter will be only able to use a single thread until another solution can be devised.- changes from version 1.5.1:
* Updated to LZ4 1.5.0. Closes #74.
* Added the \'const\' qualifier to non SSE2 shuffle functions. Closes [#75].
* Explicitly call blosc_init() in HDF5 blosc_filter.c, fixing a segfault.
* Quite a few improvements in cmake files for HDF5 support. Thanks to Dana Robinson (The HDF Group).
* Variable \'class\' caused problems compiling the HDF5 filter with g++. Thanks to Laurent Chapon.
* Small improvements on docstrings of c-blosc main functions.- changes from version 1.5.0:
* Added new calls for allowing Blosc to be used
*simultaneously
* (i.e. lock free) from multi-threaded environments. The new functions are: - blosc_compress_ctx(...) - blosc_decompress_ctx(...) See the new docstrings in blosc.h for how to use them. The previous API should be completely unaffected. Thanks to Christopher Speller.
* Optimized copies during BloscLZ decompression. This can make BloscLZ to decompress up to 1.5x faster in some situations.
* LZ4 and LZ4HC compressors updated to version 1.3.1.
* Added an examples directory on how to link apps with Blosc.
* stdlib.h moved from blosc.c to blosc.h as suggested by Rob Lathm.
* Fix a warning for {snappy,lz4}-free compilation. Thanks to Andrew Schaaf.
* Several improvements for CMakeLists.txt (cmake).
* Fixing C99 compatibility warnings. Thanks to Christopher Speller.- specfile:
* update copyright year
* Tue Jul 22 2014 toddrme2178AATTgmail.com- Cleanup .so files
* Tue Jul 22 2014 toddrme2178AATTgmail.com- Use source url so source validation can run.- Change group- Cleanup post and postun
* Thu Jul 17 2014 toddrme2178AATTgmail.com- Update to 1.4.1
* Fixed a bug in blosc_getitem() introduced in 1.4.0. Added a test for blosc_getitem() as well.- Update to 1.4.0
* Support for non-Intel and non-SSE2 architectures has been added. In particular, the Raspberry Pi platform (ARM) has been tested and all tests pass here.
* Architectures requiring strict access alignment are supported as well. Due to this, arquitectures with a high penalty in accessing unaligned data (e.g. Raspberry Pi, ARMv6) can compress up to 2.5x faster.
* LZ4 has been updated to r119 (1.2.0) so as to fix a possible security breach.- Update to 1.3.6
* Updated to LZ4 r118 due to a (highly unlikely) security hole. For details see:
* Thu May 08 2014 toddrme2178AATTgmail.com- Initial version