LLVM Clang 12 Adds Support For AVX-VNNI

The LLVM Clang compiler stack has merged its support for AVX-VNNI, the Vector Neural Network Instructions for AVX to complement the AVX-512 version.

Earlier this month GCC added AVX-VNNI support after the most recent Intel programmer's reference manual update outed this new VNNI variant without AVX-512. The Vector Neural Network Instructions are also known as DL BOOST and optimized to deliver greater deep learning performance out of CPUs for more efficient training and inference.

AVX-VNNI will initially be found with Intel Alder Lake on the desktop side towards the end of next year along with Xeon "Sapphire Rapids" server processors. These latest LLVM patches do reaffirm the Alder Lake / Sapphire Rapids introduction for AVX-VNNI with the support now being exposed there. Sapphire Rapids is really interesting on the instruction front with that also being where AMX (Advanced Matrix Extensions) are making their debut too along with various other new instructions. Besides being enabled for Alder Lake and Sapphire Rapids, there is also the -mavxvnni and -mno-avxvnni for individually toggling AVX-VNNI.

So as of this morning with this commit the LLVM/Clang 12 code slated for its release around February~March of next year has AVX-VNNI ready to go along with the other recently documented Intel CPU instructions.