> Practically speaking, is this sort of µop-dependent optimization implemented by compilers?
I suspect this specific thing is not; I don't think many people were aware of this anomaly until this week. I doubt the compilers are modeling it.
But in general, compilers will have a machine model that predicts the cost of different instructions. The backend will try to generate instruction sequences with low cost, so if you include this anomaly in the model it will probably work around it automatically.
This old example might interest you: a false dependency on the destination register for `popcnt` on some uarches led to compiler patches: https://stackoverflow.com/q/25078285/502399
I suspect this specific thing is not; I don't think many people were aware of this anomaly until this week. I doubt the compilers are modeling it.
But in general, compilers will have a machine model that predicts the cost of different instructions. The backend will try to generate instruction sequences with low cost, so if you include this anomaly in the model it will probably work around it automatically.
This old example might interest you: a false dependency on the destination register for `popcnt` on some uarches led to compiler patches: https://stackoverflow.com/q/25078285/502399