RE: [avr-gcc-list] Gcc bug??

[Dave Hylands]

Based on the conditions you've described, it doesn't sound like volatile
is the problem.

The typical problem with volatile is that the ISR changes the value and
the non-ISR doesn't see the change because the variable has been
optimized into a register.

--
Dave Hylands
Vancouver, BC, Canada
http://www.DaveHylands.com/


> -----Original Message-----
> From: address@hidden 
> [mailto:address@hidden On Behalf Of Brian Cuthie
> Sent: Wednesday, March 03, 2004 5:21 AM
> To: 'Peter Hierholzer'; address@hidden
> Subject: RE: [avr-gcc-list] Gcc bug??
> 
> 
> 
> I'll do some tests and let you know. I know I did forget to 
> add volatile to at least one variable that I'm using in my 
> interrupt handler, and I fixed this already. Since I don't 
> recall which one I'll need to explicitly test this theory. 
> I'll be a little surprised (although happy) if this turns out 
> the be the cause. It's not quite the code I'd expect the 
> compiler to emit, even without the volatile keyword. I mean, 
> 0 is still 0.
> 
> Thanks
> 
> -brian  
> 
> > -----Original Message-----
> > From: Peter Hierholzer [mailto:address@hidden
> > Sent: Wednesday, March 03, 2004 5:52 AM
> > To: address@hidden
> > Cc: address@hidden
> > Subject: AW: [avr-gcc-list] Gcc bug??
> > 
> > > I'm having some problems with an app that uses AvrX and gcc
> > (v. 3.3.1)
> > > on an ATmega16. In the interrupt routine for the TWI  I
> > write a static
> > > with a code that indicates its current state. In an effort
> > to debug a
> > > different problem I compiled the code without any optimization.
> > > 
> > > While I did discover the original problem I had been hunting, I
> > > started to see some very unreliable behavior with the TWI. 
> > After some
> > > investigation I determined that the state variable
> > (mentioned above)
> > > was getting written with an unreasonable value (not a 
> valid state).
> > > The correct state value that should have been written into the 
> > > variable is "0". Further experiments revealed that memory 
> > corruption
> > > (at least directly into the variable's memory location) 
> was not the
> > > problem.
> > > 
> > > What I then determined is that the "0" register (r1) contained the
> > > value 8!
> > > This, not coincidentally (and looking at the disassembled 
> code) is 
> > > exactly the value that's being incorrectly written into the 
> > static state variable.
> > > Interestingly, turning on any level of optimization seems
> > to fix it,
> > > at least as observed by the TWI port handler not hanging.
> > > 
> > > I was wondering if anyone else has seen anything like this,
> > or has any
> > > advice or suggestions.
> > > 
> > > Thanks
> > > 
> > > -brian
> > 
> > 
> > Hi Brian,
> > 
> > Did you declare the state variable as volatile?
> > If you access a variable in your normal program AND in the
> > interrupt routine you must declare it as volatile. Otherwise 
> > the compiler is not aware that the variable is modified in 
> > the interrupt routine and generates wrong code.
> > 
> > Peter
> > 
> > 
> 
> 
> _______________________________________________
> avr-gcc-list mailing list
> address@hidden 
> http://www.avr1.org/mailman/listinfo/avr-gcc-> list
> 
> 


_______________________________________________
avr-gcc-list mailing list
address@hidden
http://www.avr1.org/mailman/listinfo/avr-gcc-list