> Didn't need to deal with Spectre and Meltdown issues? If we made out-of-order work in ways where much more could be in flight and the compilers / assemblers w

> Didn't need to deal with Spectre and Meltdown issues? If we made out-of-order work in ways where much more could be in flight and the compilers / assemblers would know how to avoid stalls based on dependencies, or how to schedule dependencies? What if we took expensive operations, like semaphores / locks, and built solutions in to the chip? I'm pretty sure that these goals will conflict with one another at some point. For example, the way one solves Spectre/Meltdown issues in a principled way is by changing the hardware and system architecture to have some notion of "privacy-sensitive" data that shouldn't be speculated on. But this will unavoidably limit the scope of OOO and the amount of instructions that can be "in-flight" at any given time. For that matter, with modern chips, semaphores/locks are already implemented with hardware builtin operations, so you can't do that much better. Transactional memory is an interesting possibility but requires changes on the software side to work properly.