@kargl
I disagree with the remarks you have provided.
I have been developing Structural FE software since the mid 70’s and have consistently tried to understand the best way to write Fortran code that suits the computer hardware available at the time.
It probably started when I read a paper by Mondkar and Powell : “Towards optimal in-core equation solving”. I have been trying to progress ever since.
I do not agree that trying to learn about a subject is just “academic fun”
Over the last decade, I have been trying to understand how to use OpenMP and AVX based CPU’s to improve solution performance, including the interaction of memory and cache for improving multi-threaded AVX efficiency. It is not a trivial challenge. Unfortunately, the use of the libraries you recommend would only be part of the solution.
What I have learnt is to not believe the advertising from hardware suppliers and compiler developers.