Yeah, on eoan I get: $ cat /proc/spl/kstat/zfs/fletcher_4_bench 0 0 0x01 -1 0 4250112835 2191051355493 implementation native byteswap scalar 7121128976 5682673600 superscalar 9132158207 6984205638 superscalar4 8279361738 6770847284 fastest superscalar superscalar
The same machine on bionic gets: 0 0 0x01 -1 0 67628130770 127422129063 implementation native byteswap scalar 7105261516 5016484387 superscalar 9725496554 7224297957 superscalar4 8153749237 6904111964 sse2 16325929140 9147142980 ssse3 16010951957 14542922899 avx2 24379617040 23016231817 fastest avx2 avx2
Bionic's algorithm using avx2 is 2.6x faster.
Yeah, on eoan I get: kstat/zfs/ fletcher_ 4_bench
$ cat /proc/spl/
0 0 0x01 -1 0 4250112835 2191051355493
implementation native byteswap
scalar 7121128976 5682673600
superscalar 9132158207 6984205638
superscalar4 8279361738 6770847284
fastest superscalar superscalar
The same machine on bionic gets:
0 0 0x01 -1 0 67628130770 127422129063
implementation native byteswap
scalar 7105261516 5016484387
superscalar 9725496554 7224297957
superscalar4 8153749237 6904111964
sse2 16325929140 9147142980
ssse3 16010951957 14542922899
avx2 24379617040 23016231817
fastest avx2 avx2
Bionic's algorithm using avx2 is 2.6x faster.