[Tinyos-devel] Testing FTSP

[Miklos Maroti]

On Thu, Aug 14, 2008 at 11:28 PM, Branislav Kusy <kusy at stanford.edu> wrote:
>>>> inline uint32_t time16to32(uint16_t time, uint32_t recent_time)
>>>> {
>>>>  return recent_time + ((int16_t)time - (int16_t)receint_time);
>>>> }
>>>
>>> this will not work, for example for recent_time=0x18001 and time=0x7FFF,
>>> perhaps, you meant to say
>>>  return recent_time + (int16_t)(time - recent_time);
>>
>> These two are the same by the C conversion rules. Probably yours is
>> more readable, but there first the time is converted to 32 bits then
>> the 32-bit subtraction occurs then the result is converted to 16 bit
>> signed value which is extended to 32bit signed back (ignoring the high
>> 2 bytes) and added to the 32-bit signed value. Mine does a 16-bit
>> subtration, but I guess GCC optimizes both to the same OP codes.
>>
>> By the way, with your numbers:
>>
>> receint_time = 0x18001 uint32_t
>> time = 0x7FFF uint16_t
>> (int16_t)receint_time = 0x8001 int16_t
>> (int16_t)time = 0x7FFF int16_t
>> (int16_t)time - (int16_t)receint_time = 0xFFFE int16_t
>> before the + the (int16_t)time - (int16_t)receint_time is converted to
>> 0xFFFFFFFE int32_t
>
> no, it's converted to uint32_t and that's the problem
>
>> recent_time + ((int16_t)time - (int16_t)receint_time) = 0x00017FFF int32_t
>> which is converted to uint32_t
>
> $ cat foo.c
> #include<stdio.h>
> int main()
> {
>    unsigned int r = 0x18001;
>    unsigned short t = 0x7FFF;
>    printf("t:        \t%u\n", t);
>    printf("r:        \t%u\n", r);
>    printf("r+(_t-_r):\t%u\n", r+((short)t-(short)r));
>    printf("r+_(t-r): \t%u\n", r+(short)(t-r));
> }
>
>
> $ gcc foo.c -o foo && ./foo
> t:              32767
> r:              98305
> r+(_t-_r):      163839
> r+_(t-r):       98303
>
> :)
>
> brano
>

You made me install gcc :) Ok, the porblem is that everything is
converted to int IF the operand does not fit in an int. So the
((short)t-(short)r) ends up as an integer wihch is 32 bit on a PC. Try
this one to see what I mean:

#include<stdio.h>
int main()
{
   unsigned long r = 0x18001;
   unsigned int t = 0x7FFF;
   printf("t:        \t%lu\n", t);
   printf("r:        \t%lu\n", r);
   printf("r+(_t-_r):\t%lu\n", r+((int)t-(int)r));
   printf("r+_(t-r): \t%lu\n", r+(int)(t-r));
}

On the mote int is 16 bit, so I think the code works there, but I
stand corrected and your version is clearly more robust. Please use
that.

Miklos