This whole article was problematic to me because it starts out with:
Two processors, running in parallel, execute the following machine code
Processors don't "run in parallel." Among other things, the OS could be scheduling the threads on each CPU in any way it wants. I just can't think about the false hypothetical that the processors "run in parallel."
I'm in the process of going through the article more carefully to see if this erroneous way of thinking propogated. It seems like the author knows his stuff, though, so I'm guessing not.
However, one of the reasons I was uncomfortable with the setup (besides thread scheuling), is that I personally don't make any assumptions about cache coherency. To me, it could take thread 1 an arbitrary amount of time to see an update made to memory by thread 2, unless some primitive (like a semaphore... with "acquire and release semantics") is used.
Right? I mean, couldn't that account for what is happening just as much as instruction reordering? In theory, I'm fairly certain this is true. In practice, it depends on the specifics of cache coherency in x86-64; if anyone can comment on that, I'd appreciate it.
Two processors, running in parallel, execute the following machine code
Processors don't "run in parallel." Among other things, the OS could be scheduling the threads on each CPU in any way it wants. I just can't think about the false hypothetical that the processors "run in parallel."
I'm in the process of going through the article more carefully to see if this erroneous way of thinking propogated. It seems like the author knows his stuff, though, so I'm guessing not.