Presented at this week's Linux Foundation Collaboration Summit in San Francisco was Noah Clemons of Intel talking about the company's Math Kernel Library. With it being performance-related, of course it catches our attention.
While not brand new, Intel's Math Kernel Library generates hand-tuned AVX (Advanced Vector Extensions) code with functions for common algorithms. Multiple operating systems are supported by this library and their Integrated Performance Primitives and they work with any compiler, the functions are thread-safe, and have minimal overhead. AVX support has been present since Intel Sandy Bridge CPUs and can lead to quite faster math operations given the right routines.
Intel MKL works with linear algebra, fast fourier transforms, vector math, vector random number generators, summar statistics, and data fitting. The Math Kernel Library is multi-core, multi-CPU, and cluster friendly while MKL 11.0 also introduces support for the Xeon Phi co-processor.
For those looking to learn more about Intel's Math Kernel Library and Intel Performance Primitives, check out the PDF slides that are 51 pages long from this week's Linux Foundation summit. This performance-optimized math library is available from software.intel.com.