exchange_and_add inlined in g++ 4.0.2

[Peter Dimov]

Paolo Carlini wrote:

> A bit of additional info, in case you missed some of yesterday
> messages. Given the cost of a function call on P4 (~10 cycles) vs
> that of an atomic (~ 500-1000 cycles) it's unlikely that your profile
> would change much if the atomics are inlined.

Some configurations may suffer a 1000 cycles penalty (quad Xeons would be my 
guess - I've never observed that), but the average case isn't _that_ bad. On 
my Athlon I'm seeing the following numbers:

lock xadd, inline: 12 seconds
lock xadd, out of line: 15 seconds
non-atomic add, inline: 2 seconds
non-atomic add, out of line: 7 seconds

for 2^30 operations on a totally non-representative looping benchmark. :-)