[AD] windows time functions

[AJ <aj@xxxxxxxxxx>]

/4.9.2/src/win/wtime.c

in function:

unsigned long al_current_time(void)
{
  /* XXX: Maybe use QueryPerformanceCounter? */
  return timeGetTime() - _al_win_initial_time;
}


more comments:

QueryPerformanceCounter, obtains its value from the CPU upon which the 
current thread is running.
If the thread switches to another CPU, it is possibly for the value to 
completely change.
There are currently no intelligent solutions to this problem.
Some solutions i have heard of:  
  - Mask the calling thread to a specific CPU, (which kind of defeats 
the point in having multi-processor machines).
  - Run it in a thread that is locked to a specific CPU, and use 
CriticalSections to access it.. ( which defeats the  
Query[Performace]Counter() )

QueryPerformanceCounter  is also documented as jumping in value which 
can create non-linear values being returned.



Problems with timeGetTime() :

unless you run the  timeBeginTime()  and timeEndTime()     
timeGetTime()'s resolution can be as much as 8-10ms sloppy.
which for anything like trying to sync a screen draw/flip  can cause 
significant error in calculations.


al_current_time()'s   wrap length.   49days :(
I know many (99.9%) of you are going to say "whose stupid enough to be 
using an allegro app for 49 days"...  well.. I do!

I do like the recently proposed solution on a.cc   using a struct { 
int32_t hi; uint32_t lo; }   when lo wraps hi++  then accessor functions 
_al_get_current_time32() {  return t.lo; }     and   
_al_get_current_time64() {  return (__int64)t; }


aj.