Intel has kept a long “Optimization Reference Manual” to show developers how to optimize code for their latest CPU microarchitectures, but along with their latest manual update, there are now plenty of actual code examples to make it easier to do this. process of learning optimization techniques from Intel to take full advantage of their latest processors.

Aside from enabling open-source / Linux support and other key areas directly related to the deployment of their new hardware, Intel engineers are already doing a lot in the name of performance for open-source projects. as often by directly contributing to optimizations and the like. features to take advantage of their latest processor features in popular open source projects. We have covered such contributions from Intel countless times on Phoronix over the years.

From these open source code contributions directly from Intel as well as the open source code maintained by them in their projects as well as within oneAPI, independent developers can already glean a lot on optimization techniques and get the most out of it. their latest and greatest processors. There is also the Intel 64 / IA-32 Architectures Optimization Reference Manual, but that manual now comes with working (buildable) code examples as a much easier initial step in the learning curve of Intel code optimizations.

Intel’s latest optimization reference manual is available at software.intel.com while the new and exciting item is the intel / optimization manual on GitHub.

This new GitHub repository provides working code samples that accompany the optimization manual. All of these code samples can be easily built from scratch with the CMake build system on Linux and using any semi-recent code compiler on just about any Intel Haswell or newer processor.

These new code samples primarily involve AVX / AVX2 / FMA optimizations, INT8 deep learning inference, and the use of AVX-512 for best practices when targeting the latest Intel processors like Xeon Scalable Ice. Lake, Tiger Lake and Rocket Lake. It’s great to see these ongoing open contributions from Intel engineers outside of all of their other open source code contributions and other commitments.