<div dir="ltr">LIBXSMM 1.5.1 has been released, which is mainly a bug-fix release gaining its urgency from a fixed Fortran interface (SMM functionality) The issue is applicable to CP2K in particular (Fortran), where requesting a JIT kernel never returned a suitable PROCEDURE POINTER (always NULL): <a href="https://github.com/hfp/libxsmm/releases/tag/1.5" rel="nofollow" target="_blank">https://github.com/hfp/libxsmm/releases/tag/1.5.1</a>. Am Mittwoch, 5. Oktober 2016 15:01:25 UTC+2 schrieb Hans Pabst:<blockquote class="gmail_quote" style="margin: 0;margin-left: 0.8ex;border-left: 1px #ccc solid;padding-left: 1ex;"><div dir="ltr">LIBXSMM 1.5 has been released. You can read more about what's INTRODUCED (also see below), the CHANGES in general, and what has been FIXED: <a href="https://github.com/hfp/libxsmm/releases/tag/1.5" rel="nofollow" target="_blank" onmousedown="this.href='https://www.google.com/url?q\x3dhttps%3A%2F%2Fgithub.com%2Fhfp%2Flibxsmm%2Freleases%2Ftag%2F1.5\x26sa\x3dD\x26sntz\x3d1\x26usg\x3dAFQjCNHQdBABNTzQBMbJF_tMGmwC6bM0rg';return true;" onclick="this.href='https://www.google.com/url?q\x3dhttps%3A%2F%2Fgithub.com%2Fhfp%2Flibxsmm%2Freleases%2Ftag%2F1.5\x26sa\x3dD\x26sntz\x3d1\x26usg\x3dAFQjCNHQdBABNTzQBMbJF_tMGmwC6bM0rg';return true;">https://github.com/hfp/libxsmm/releases/tag/1.5</a>. The library was carefully validated, the SMM core functionality received fixes for issues which are not exposed by CP2K (but are present in previous releases of LIBXSMM). The validation was against a variety of applications; most relevant here are CP2K's regression tests. These tests have been even stronger by using LIBXSMM's <a href="https://github.com/hfp/libxsmm#call-wrapper" target="_blank" rel="nofollow" onmousedown="this.href='https://www.google.com/url?q\x3dhttps%3A%2F%2Fgithub.com%2Fhfp%2Flibxsmm%23call-wrapper\x26sa\x3dD\x26sntz\x3d1\x26usg\x3dAFQjCNFS1Yf27wBQjr8w-cCcXtms0rFAyQ';return true;" onclick="this.href='https://www.google.com/url?q\x3dhttps%3A%2F%2Fgithub.com%2Fhfp%2Flibxsmm%23call-wrapper\x26sa\x3dD\x26sntz\x3d1\x26usg\x3dAFQjCNFS1Yf27wBQjr8w-cCcXtms0rFAyQ';return true;">linker wrapper</a> to pass all GEMM calls through the LIBXSMM library including calls that are made by other libraries such as LAPACK. Most notable for the Supercomputing Centers might be support for the CRAY Compiling Environment (CCE), but also the support for PGI's compiler. Please note, the JIT code generation under Microsoft Windows is still pending, and due to missing support for the calling convention this applies equally to Cygwin. INTRODUCED <ul><li>New DNN API, sample code, and benchmarks (Googlenetv1, DeepBench, and Overfeat)</li><li>Enabled tiled GEMM support in static/dynamic wrapper; MT support via libxsmmext</li><li>More format variations of sparse matrix multiplication (dense/sparse etc.)</li><li>Sample code showing sparse matrix multiplication (PyFR examples collection)</li><li>Published synchronization layer (atomics, and simple/bare OS-thread/lock abstraction)</li><li>Introduced mini-API for optimized barrier implementation (general multicore support)</li><li>Introduced API for memory allocation (malloc interface); mostly exposed from internal API</li><li>Beside of Intel VTune, now Linux perf and jitdump are supported (Thank you Maciej D.!)</li><li>SPECFEM sample: received nicely written example contribution (Thank you Daniel P.!)</li><li>OSX (incl. "El Capitan") now supports Intel Compiler, Apple/Clang, and GNU GCC</li><li>CRAY's Compiling Environment (CCE) is now supported</li><li>PGI compiler is now supported</li></ul></div></blockquote></div>