New Release of LIBXSMM

[Hans Pabst]

*LIBXSMM 1.6.1 has been released* (
https://github.com/hfp/libxsmm/releases/tag/1.6.1 
<https://github.com/hfp/libxsmm/releases/tag/1.5>), which might be of 
interest for people who are running a security-hardened Linux kernel. 
LIBXSMM now supports JIT compilation in an enforcing SELinux environment. 
The v1.6 series focuses mostly on the CNN API for machine learning 
(Convolutional Neural Networks). However, there is also a new/experimental 
low-level API for sparse matrix multiplication (focuses on integration with 
other libraries). Most notable for this audience might be the native 
support for the Cray compiler (some fixed 8.3.x versions, and 8.4 series 
and beyond are supported); one may use "make CXX=CC CC=cc FC=ftn 
COMPATIBLE=1". Compatibility with the Cray Compiling Environment (CCE) has 
been generally improved. Also, OpenBLAS is now picked up in a more flexible 
fashion ("out-of-the-box") in typical environments ("module load 
openblas/dynamic/0.2.6"). The aforementioned COMPATIBLE key allows to omit 
sophisticated build settings, and thereby eventually enables "other 
compilers" (beyond GCC, Clang, and Intel).
<https://github.com/hfp/libxsmm/releases/tag/1.5>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20161205/9317e176/attachment.htm>