by lundman » Wed Jun 08, 2016 10:40 pm
First comment about SIMD, we are somewhat lucky that in XNU kernel we can mix SIMD with regular without any hoop-jumping, so we can vectorise our algorithms at any time. None of them are at the moment though. I was looking at the fletcher avx2 code from ZOL, but realised my hardware does not have AVX2 so I am unable to test it anyway.
As for EDON-R, the cryptographical concerns are probably valid, but it is only used as checksum in ZFS, so worries about collisions are probably not as drastic. But who knows.