Re: Portable memory barrier?

["Randy Brukardt" <randy@rrsoftware.com>]

"Jere" <jhb.chat@gmail.com> wrote in message 
news:8a968aae-79e4-421b-ba4b-e0a9a33ce0db@googlegroups.com...
...
>I am still digesting some of this, so forgive the simple question:

Don't bother, it's irrelevant if you trust your compiler vendor. If you 
don't, then you're out of luck on modern CPUs because mere mortals cannot 
understand them.

>What prevents the last 3 lines from being reordered like this:
>r1 := r1 + 1;
>(Write_Index'Address) := r1;
>(r3) := r4;

Assuming again that Write_Index is atomic as is necessary to make any 
guarentees, this translation is 100% wrong (can't put Write_Index into a 
register), so reordering is irrelevant. If the CPU is going to do it anyway, 
then the compiler has to do whatever the CPU vendor provided to prevent it.

If Write_Index isn't atomic, then you can make no assumptions about how the 
code is executed and hopefully you aren't (it is purely sequential code).

                              Randy.