[cpp-threads] Yet another visibility question

[Alexander Terekhov]

On 1/11/07, Paul E. McKenney <paulmck at linux.vnet.ibm.com> wrote:

[... AMD x86-64 ...]

>        do {
>                l = x;
>        } while (cmpxchg(&x, l, l) != l);

Regarding providing illusion of remote write atomicity (under
assumption that AMD x86-64 is PC and not TSO) that's no better than

        l = cmpxchg(&x, 42, 42); // with LOCK prefix

because (quoting AMD64 Architecture Programmer's Manual Volume 3)

"CMPXCHG always does a read-modify-write on the memory operand. If the
compared operands were unequal, CMPXCHG writes the same value to the
memory operand that was read."

regards,
alexander.