[Tinyos-help] Bug In Arithmetic Ops

[Philip Levis]

On Nov 27, 2007, at 1:36 PM, Federico Fiorentin wrote:

> Bug In Arithmetic Ops
> I did more tests, and the problem still persists.
> It seems that the evaluation becames incorrect when the T1 operand  
> is bigger then 2^24.
>
> Below the operands T1, T2.
> T1 is the LocalTime.get() value. AlwaysT1 >= T2.
>
> The subtraction T1-T2  evaluated by mote, T1-T2 by Calc OpenOffice,
> then (T1-T2 by mote)-(T1-T2 by Calc) and last one logT1 in base 2.  
> The same data in the attachment.
>
> I'll send the code if you need, but I'm sure that there aren't  
> errors in it.
> I hope to find a solution.
> thanks
>
>     T1               T2       T1-T2 by mote   T1-T2 by Calc    
> Difference logT1/log2
>
>    16983908   16759804          224103        224104               
> -1      24,02
>

I am unable to reproduce this case in a simple, controlled test. The  
mote always returns 224104, for all combinations of uint32_t and  
nx_uint32_t operations. So there is something weird going on with  
your code. Try rewriting it to figure out what construction causes  
the problem. For example, rewrite

PollResp_ptr->globalTime = PollResp_ptr->eta[1] - T2

To be

uint32_t gTime = PollRsp_ptr->eta[1];
gTime -= T2;

Or directly access arrivalTimestamp rather than going through your  
structure. The generated assembly is very long (and inefficient), so  
hard to work through. But it is not an arithmetic problem.

Phil