Thanks a lot for developing and running these microbenchmarks and writing it up!
One thing I noticed that there isn't made a separation between SVE and SVE2, and also not between SME and SME2. Since they all seem supersets of each other, it would be really nice to know how much the SVE, SVE2, SME and SME2 features are accelerated.
So it might be interesting to add some microbenchmarks to specifically target features supported by SVE, but not by SVE2, and some supported by SME, but not by SME2.
Thanks a lot for developing and running these microbenchmarks and writing it up!
One thing I noticed that there isn't made a separation between SVE and SVE2, and also not between SME and SME2. Since they all seem supersets of each other, it would be really nice to know how much the SVE, SVE2, SME and SME2 features are accelerated.
So it might be interesting to add some microbenchmarks to specifically target features supported by SVE, but not by SVE2, and some supported by SME, but not by SME2.