Charles Manning wrote:
On Saturday 02 May 2009 07:03:10 Gene Smith wrote:
With -O0 opt level, gdb refuses to step (S) over a delay loop like this:
code1; //<-- steps over this OK
for(c=0; c<10000; c++); //<--- hangs here
code2;
If I control-c and examine c it seems to be increasing slowly.
When I change the code structure like this:
code1;
for(c=0; c<10000; )
c++;
code2;
then I can step.
This is with embedded arm. See the same thing with arm-insight and
arm-gdb. Haven't tried this with native.
Is this to be expected?
The stepping in ARM, and perhaps other CPUs, is prone to getting stuck like
this.
To get out of this either:
* restructure code
* set break at the next line and continue.
* -if you're brave - try improve gdb :-).
It seems to be "stepping" 10000 times in the background to get over this
instruction so it will take a while. I think I verified this by setting
c to 9995 and it took just a short time to step to the next source line.
I think another "improvement" would be to eliminate erratic stepping
when optimization is applied (if that is possible).