Linus has already chimed in, and wants Zstd to just use the standard way of checking for CPU features instead of rolling their own. Apparently, that would be even faster, compiling to a single jump instruction.

this concerns both btrfs and zram, so this would be a huge boost?

I can’t wait for this fix to drop.

Is this relevant when using built-in zstd compression on ZFS?

Ace, its great to see so many significant performance improvements in the kernel lately