While Intel Xeon "Sapphire Rapids" processors with Advanced Matrix Extensions are set for a Q2'22 ramp in production, one of the key new features that has yet to be properly plumbed in the mainline Linux kernel is for supporting AMX.
For more than one year now that has been various LLVM/Clang and GCC compiler patches as well as the kernel-side work around bringing up Advanced Matrix Extensions as one of the exciting additions for Sapphire Rapids. On Friday the Linux kernel work around supporting AMX was revised for an 11th time.
While the interesting AMX usage is on the compiler side and developers to make use of the new AMX capabilities, the kernel work around AMX is critical. The necessary kernel changes involve a new system call for applications to actually request access to use Advanced Matrix Extensions, handling for applications without AMX permissions, and other related changes.
With the new v11 AMX patches for the Linux, the permission requirements have been expanded to cover more AMX states, a new static allocation feature flag, simplifying various elements of the code, and other low-level code improvements. The latest Intel AMX patches for the kernel can be found via this mailing list thread. Hopefully this initial work will manage to get settled in the next few weeks to allow it time to hopefully appear in the 5.16 mainline kernel so this AMX support can begin working its way into release kernels and Linux distributions ahead of Sapphire Rapids processors becoming common next year.