> Which makes it a perfectly fine benchmark to measure whether a particular compiler implements these optimisations.

No, because whether the optimizations are “implemented” doesn’t matter.

What matters is whether the optimizations are robust enough to trigger for real code, not just bogus dead loops.