SUSE's Adaptable Linux Platform Considers Requiring AVX-Capable x86_64 CPUs
With the work in recent years around x86_64 micro-architecture feature levels for the compiler stack, SUSE engineers had been eyeing a x86-64-v2 baseline over plain x86_64 as is currently targeted. The x86-64-v2 baseline means around Intel Nehalem CPUs and newer with requiring SSE3, SSE4.1, SSE4.2, SSSE3, and other newer instruction set extensions over the base x86_64 ISA. The x86-64-v2 target is what's used with the new Red Hat Enterprise Linux 9. But SUSE may push things up even higher and instead use x86-64-v3 as their base target.
With x86-64-v3, AVX, AVX2, BMI1, BMI2, F16C, FMA, LZCNT, MOVBE, and XSAVE all become required extensions. This takes the CPU requirements up to the era of Intel Haswell or on the AMD side around Excavator.
x86-64-v3 basically means Intel Haswell era x86 64-bit CPUs or newer would be their support requirement.
SUSE release manager Lubos Kocman has recently opened a public ticket over their (re)considering of x86_64 v3 for the ALP effort. There has been some push-back from the community over their v3 target plans and they may end up moving to v2 depending upon how the discussions play out.
Keep in mind this is for SUSE's Adaptable Linux Platform effort that isn't expected to ship until after a SUSE Linux Enterprise 15 SP5 (openSUSE Leap 15.5) with that update expected around May 2023 and thus still some time away till ALP -- unlike RHEL9 with x86-64-v2 that debuted as GA back in May. So in accommodating for the time until ALP will actually ship, x86-64-v3 may be less of an issue.
It will be interesting to see what is decided for the x86_64 micro-architecture feature level of the Adaptable Linux Platform. This follows the RHEL9 upping to v2, various Arch Linux efforts to raise their baseline or provide more optimized packages, and other efforts taking shape throughout the ecosystem. It will be interesting to see if Ubuntu or any other major players decide to (re)evaluate their baselines in the near future, which would be great to see or at least function multi-versioning and other compiler-assisted efforts to better exploit the capabilities of modern CPUs.