Don't use a macro that will silently fail when there are mistakes.

Don't use the functions added in cmake version >= 12

Add extra checks and fix REPEATS issue.

This is a simplified version of the submit script in order to simplify
the tester execution.

Use extern inline for functions intended to access from outside.

Remove serial code.
Add Blas conditional compilation.

This adds new events that will be needed when adding the extra versions.

Matmul functions to header and fix instrumentation entries.

Disable diagnostic 858
And add some const to dgemm_

The jacobi base independent compilation don't improve performance, and
now we use blas, so...

Now the multiple cpus are not run in the internal loop