Re: Microcode is hard
The problem occurs only with active hyperthreading, i.e. when the CPU is executing two different instructions in parallel that are *supposed* to use non-interfering components of the processor. Hard to find that with simulators.
The code pattern to trigger this bug is also discouraged by Intel Optimization guides as being slow, so probably only coded very rarely by a compiler. Probably not a high priority on simulation experiments until now.