An optimized BLAS library based on GotoBLAS2
OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version. The project is supported by the Lab of Parallel Software and Computational Science, ISCAS. http://www.rdcps.ac.cn
- Developed at science
- Sources inherited from project openSUSE:Factory
-
5
derived packages
- Download package
-
Checkout Package
osc -A https://api.opensuse.org checkout openSUSE:Leap:15.0:Staging:FactoryCandidates/openblas && cd $_ - Create Badge
Refresh
Refresh
Source Files
| Filename | Size | Changed |
|---|---|---|
| OpenBLAS-0.3.13.tar.gz | 0012437434 11.9 MB | |
| README.HPC.SUSE | 0000000889 889 Bytes | |
| README.SUSE | 0000000851 851 Bytes | |
| _constraints | 0000000373 373 Bytes | |
| _multibuild | 0000000180 180 Bytes | |
| openblas-noexecstack.patch | 0000001216 1.19 KB | |
| openblas-s390.patch | 0000001363 1.33 KB | |
| openblas.changes | 0000050175 49 KB | |
| openblas.spec | 0000014539 14.2 KB |
Revision 39 (latest revision is 61)
Dominique Leuenberger (dimstar_suse)
accepted
request 856635
from
Ismail Dönmez (namtrac)
(revision 39)
- Update to version 0.3.13
common:
* Added a generic bfloat16 SBGEMV kernel
* Fixed a potentially severe memory leak after fork in OpenMP builds
that was introduced in 0.3.12
* Added detection of the Fujitsu Fortran compiler
* Added detection of the (e)gfortran compiler on OpenBSD
* Added support for overriding the default name of the library independently
from symbol suffixing in the gmake builds (already supported in cmake)
RISC V:
* Added a RISC V port optimized for C910V
POWER:
* Added optimized POWER10 kernels for SAXPY, CAXPY, SDOT, DDOT and DGEMV_N
* Improved DGEMM performance on POWER10
* Improved STRSM and DTRSM performance on POWER9 and POWER10
* Fixed segmemtation faults in DYNAMIC_ARCH builds
* Fixed compilation with the PGI compiler
x86:
* Fixed compilation of kernels that require SSE2 intrinsics since 0.3.12
x86_64:
* Added an optimized bfloat16 SBGEMV kernel for SkylakeX and Cooperlake
* Improved the performance of SASUM and DASUM kernels through parallelization
* Improved the performance of SROT and DROT kernels
* Improved the performance of multithreaded xSYRK
* Fixed OpenMP builds that use the LLVM Clang compiler together with GNU gfortran
(where linking of both the LLVM libomp and GNU libgomp could lead to lockups or
wrong results)
* Fixed miscompilations by old gcc 4.6
* Fixed misdetection of AVX2 capability in some Sandybridge cpus
Comments 7
Please, update to OpenBLAS to 3.17, as some optimizations were causing segfaults and reverted. See: https://github.com/xianyi/OpenBLAS/blob/develop/Changelog.txt.
It seems that octave 6.3.0 build failures (on test phase) may be associated to this on Tumbleweed.
OK, this really needs a better explanation. I have Tumbleweed and Leap 15.3 installed on different computers, and on both of them I use the Science repo to provide some more packages I need. Downgrading to default versions (3.13 on Leap and 3.14 on Tumbleweed) fixes the problem with Octave, as I was able to build version 6.3 and also run the tests (make check-local) without unknown failures. It probably affects other numerical packages too.
Regards, André
@acobar: The comment section is not a bug tracker. https://bugzilla.opensuse.org/show_bug.cgi?id=1188608
Hi,
Now it is something less trivial (or, perhaps, more). Julia version 1.8.5 (and probably some older versions of 1.8.x series too) needs integer indexes of vectors and matrices to be 64 bits, that means, there is a need to build adjusted "flavors" of some libraries.
For openBLAS, the "flavored" version must be compiled with "INTERFACE64=1" option.
Of course, other libraries that are linked against openBLAS will need "flavored" versions too. SuiteSparse, qrupdate, sundials and arpack-ng, at least.
The relevant sources of information are: https://docs.octave.org/v7.3.0/Compiling-Octave-with-64_002dbit-Indexing.html https://savannah.gnu.org/bugs/?57771 https://github.com/Reference-LAPACK/lapack/issues/666
On my computer, I built "proto" packages with the needed adjustments, but, of course, the "ideal" way is to have an established standard for package and libraries names.
It should be noted that, once "flavored" libraries are built, not only we will be able to build Julia but also a special version of Octave.
May I collaborate with the effort to make it happen, let me know.
OK, I will fork the current project so that my modifications can be debated. Still, I have one doubt, how do I pass a definition to rpmbuild when using build.opensuse.org infrastructure?
What I mean is that, on my computer, to generate the binary package with 64 bits indexes, I use: rpmbuild --define "index64 1" openblas.spec
I have no idea how to do this under build.opensuse.org.
Branched. Will do the same on SuiteSparse, arpack-ng, qrupdate and sundials, but would appreciate if I get an input about what may be the best practices, names and definitions to be used, or if it seems to be OK the way I did.