14 Dec
2009
14 Dec
'09
5:58 p.m.
On Mon, Dec 14, 2009 at 02:45:38PM -0500, Stephen Sinclair wrote:
I didn't read the others, but I looked at this one
and I think it's
important to point out some problems with it. He is claiming that you
cannot increment an integer atomically, which is true, but this
doesn't mean that the reads and writes, by themselves, are not atomic
operations.
True. OTOH having devoted cq. wasted part of my life
programming SPARC machines I can confirm Paul Davis'
comment on those: sig_atomic_t on these machines was
24 bits. But I *never* understood how on earth they
managed to create this anomaly on what was after all
a full 32-bit architecture.
A simple pattern I use to avoid the incr/decr problem
is to use two variables, each one of them being modified
(incremented) by only one thread. Instead of testing for
zero test for equality.
Ciao,
--
FA