>The latest quadsort version sorts 8 elements at a time. It's not really a sorting network,
That is precisely the definition of a sorting network. A sorting network is a fixed size sorter - that is all.
>it wouldn't have worked very well prior to branch prediction, or having a multitude of registers.
Both around for over 60 years...... In fact, both were in production hardware before any of these sorts were invented. Do you just make stuff up?
>So basically all the work that had been done for decades has become useless
Your baffling dismissiveness of things apparently you have little or no understanding of makes me highly disbelieving you're doing decent analysis on your sort. This is not the sign of a careful, thoughtful algorithm designer.
> plenty of people have checked and ran the benchmark themselves.
I don't doubt you have benchmarks where it's better - that's trivial to do for any sorting network idea - that is well trodden theory. I posted above how to make data that falls on either side of the argument. Making a sort that performs well over significant variety in real-world data without suddenly hitting bad downsides is the hard part. Making some datasets faster makes others slower when you're fiddling with number of comparisons dynamically.
That you are unaware of the past and dismiss it out of hand is not a good way to ensure you're not fooling yourself.