Andy Freeman  <anamax@earthlink.net> wrote:
+---------------
| hs@paradise.nirvananet (Hartmann Schaffer) wrote:
| > it's been a while, but my understanding of synchronization points was
| > that after a sybchrobization point side effects in the preceeding
| > expression had to have an effect on subsequent calculations.  if
| > the compiler can determine that a side effect result is redundant, it
| > can discard them and the problem becomes mute
| 
| That's a huge "if".  How does a compiler know about side effects on
| other processes via memory accesses?
+---------------

You have to tell it. That's what the ANSI C "volatile" attribute is
all about. It tells the compiler that it *can't* depend on a location
being affected only by what it can prove about "local" code. For example,
consider "busy-waitng" on a flag that can be changed by an interrupt
routine:

	extern volatile done_flag;
	...
	while (!done_flag)
	  ; /*wait*/

Without the "volatile" declaration, the compiler would be completely
justified in treating the above exactly the same as this:

	if (!done_flag)	/* only need to test once, since never assigned to */
	  for(;;)
	    ;

which is an infinite loop if "done_flag" was false on entry.


-Rob

-----
Rob Warnock, 31-2-510		rpw3@sgi.com
SGI Network Engineering		http://reality.sgi.com/rpw3/
1600 Amphitheatre Pkwy.		Phone: 650-933-1673
Mountain View, CA  94043	PP-ASEL-IA