- In the C article he shows that LAPACK can be massively improved upon. That sounds very impressive, until you realize he hasn't mentioned which LAPACK/BLAS implementation he used. The default download from netlib is the vanilla version and is known to be very slow in general. Any benchmark dissing LAPACK is incomplete without comparison with OpenBLAS and Intel MKL. And instead we see the author pulling all sorts of overengineered tricks to top the performance of a vanilla LAPACK build (I mean there isn't even a mention of whether or not he used optimization flags while compiling LAPACK).
- In this python aritcle, again, python comes with highly developed scipy/numpy stack which can be coupled with a highly performant OpenBLAS or Intel MKL, but there is absolutely zero mention of the most popular scientific stack in the python community. Again, instead, we see a round-about and "clever" solution of trying to couple python with LLVM.