I see nothing new in this diff.  It is a repeat of the existing algorithm.

The existing algorithm is to look at stupid data points in the kernel,
to justify slowing the machine down drastically.

Then, if there suddenly is a need for high-performance, it takes 1/5th
of a second to notice.

There is no attempt to build a mechanism which will instantly provide
strong compute performance when there is a hint that it is needed.  If
there was such mechanism, then this would not need to look at stupid
variables, like "load".  The scheduler's load calculation is a complete
joke.  It does not do what you want.

We don't need tuned policies.  We need actual smart work, rather than
an algorithm which takes 1/5th of a second to make a machine responsive.


A key thing to notice is this is discussing power saving.  It is not
discussing making a machine useable.