Yes, as a long-term Lisper in research computing who knows a bit about linear algebra implementation and compilation. I don't see the comparisons I think are relevant to claims of how much more performant it's supposed to be. It's worth pointing out that you don't get the highest performance (close to peak) for things like normal GEMM on something like SKX a priori, at least from what I've seen. It needs experimentation, like fiddling with block sizes and prefetching.

An example I've asked about previously: When comparing a recursive implementation of an LAPACK operation with OpenBLAS, is that using the RELAPACK(?) implementation in OB, and if not, how does using a similar algorithm in C or Fortran compare? Generally, I want details of measurements.