Intel AMX Support Appears Ready For Linux 5.16

It's been over one year since Intel disclosed Advanced Matrix Extensions and began posting patches for bringing up AMX support under Linux in anticipation of Xeon Scalable "Sapphire Rapids" processors. While the compiler-side work to GCC and LLVM/Clang has been landing, finally with the forthcoming Linux 5.16 cycle that AMX support appears ready for landing.

Merged today to tip/tip.git's "x86/fpu" branch where kernel FPU changes are queued ahead of the next merge window, the last of the AMX enablement patches were queued up. Most notably, the work for actually enabling the AMX feature and being able to expose it to user-space via the new interface.

Unlike AVX-512 and earlier, user-space applications actually need to request the support from the kernel to be able to use Advanced Matrix Extensions functionality. The prctl interface aorund the dynamic XSTATE features is documented with this commit.

Since last year Intel open-source engineers sent out many revisions to the AMX patches while now it looks like it will be in place for this next kernel version. The Linux 5.16 merge window is opening up in a week or two while the stable Linux 5.16 kernel should debut by the start of January. Linux 5.16 in turn should be the kernel version that gets used by the likes of Ubuntu 22.04 LTS. Linux 5.16 is bringing many new features with AMX just being the latest. Xeon Sapphire Rapids processors with AMX support meanwhile should become available in Q2.

While waiting for the Linux 5.16 cycle to kickoff, the queued patches can be fetched for now from tip.git's x86/fpu branch.