Both the hardware and LLVM have changed since 2012.
Now, load-based heuristic don't show big differences any more on OoO cores.
There is no regressons and improvements on spec2000/2006. (Cortex-A57, Core i5).
But There are two big improvements in test-suite.
|Benchmark Name||Exe time Opt/Ori|
And I didn't see notable regressions in test-suite.