Comment by eru

I vaguely remember about some benchmarking project that deliberately randomised these compiler decisions, so that they could give you more stable estimates of how well your code actually performed, and not just how well you won or lost the linker lottery.