optimizing variables

optimizing variables

[Jonne Zutt]

Dear gcc experts,

In my simulation tool, I use somebody else's run-time library for a
simulation kernel. This run-time library supports threads by allocating a
local array called 'stackarea' and doing all kinds of stuff and long jumps
etc. I don't quite understand.
The library is used many times, it's stable and all bugs are supposed to
be my own fault.

	/* Initialise the array, to fool compilers that it is being
	 * used, so they don't optimize it away.
	 * BTW don't write to the entire array, copy-on-write mechanism
	 * make that very slow. (!)
	 */
	stackarea[0]=0;
	// for(i=1; i<STACKSIZE-1; i++) stackarea[i]=stackarea[i+1];

The system works perfectly fine when compiled without any optimization.
Providing -O for the above file causes a segmentation fault for each
execution very quickly.

Adding the above commented for loop seems to solve this problem.

Now, here's my question:
- I have the feeling this statement stackarea[0]=0 might have fooled gcc
version 3 compiler, but my gcc version 4.0.1 DOES optimize this strange
array away. Is that correct?
- How can I easily check whether a variable has been optimized away, and
how can I avoid this? (e.g. should I write stackarea[random(0,N)] = 0 or
is there a more neat way).

Thanks in advance,
Jonne.

ps >
Just noticed that stackarea[0] = stackarea[STACKSIZE-1] = 0 also does not
"fool" my compiler (i.e. still results in a segmentation fault).

Re: optimizing variables

[John Love-Jensen]

Hi Jonne,

I was not able to reproduce the problem you are expriencing.  Could you
provide a compilable short example of the problem you are running into.

Please include the command line used to compile the source.

Also, what platform are you using?

Thanks,
--Eljay