Re: [AD] Add timeBeginPeriod(1) to wtimer.c

[AJ <aj@xxxxxxxxxx>]

i looked at the src code, to find out about the WaitForSingleObject()

how do you measure the results ?

lets presume for a moment that the theory about the kernel running more 
often to check on this thing is going to cost CPU cycles, what if it 
does!  i mean is the .001% really going to matter ?

now lets presume the WaitForSingleObject() gets checked whenever the CPU
is ready to check it, it will get done when it gets done.  which is how 
i think it actaully works.  if i request 1ms resolution, i *mean* i want
1ms resolution.  not some random number between 6 and 12ms.

does the continual calling of timeBeginPeriod()/End() going to cost more 
than just setting it once, and leaving it ?

how varied is your performance when using this mod ?

i set it in my main(), and leave it set.

aj.






Tobias Scheuer wrote:
> I think you're right. But my measurements have given different results,
> though I can't explain all the details. It might also depend on the machine
> and graphics adapter. Did you try the change on Windows on your PC? 
>
>
>
> > allegro uses timeGetTime() in its low_perf_thread  which would be 
> > effected, but in my opinion, a good way.
> > it uses a WaitForSingleObject() which will signal when the time is right 
> >  regardless of how much resoltuion was used to set the time.
> > i dont see how it can be effecting anything, if anything, it should be 
> > making everything more responsive.
> >
> > aj.
> >
> >
> >
> > Tobias Scheuer wrote:
> >
> > > I think it affects the timers more generally, so it also affects the
> >
> > timers
> >
> > > in allegro's separate timer threads. Didn't go into all the details, and
> >
> > I
> >
> > > don't have a decent profiler at home, so I couldn't find out where
> >
> > exactly
> >
> > > the slowdown comes from. Maybe I should try to Quantify this at work... 
> > >
> > >
> > >
> > > > it changes the resolution of the rest() and timeGetTime()  but it 
> > > > doesn't slow down the app itself.
> > > > how could it ?
> > > >
> > > >
> > > >
> > > > Tobias Scheuer wrote:
> > > >
> > > >
> > > > > I have tried that. The documentation says that each call to
> > > >
> > > > timeBeginPeriod
> > > >
> > > >
> > > > > has to be matched by a call to timeEndPeriod, so that#s why I have also
> > > > > called that function. If I leave that out, the timeBeginPeriod(1)
> > > >
> > > > remains
> > > >
> > > >
> > > > > active, which causes some overall slowdown of the whole program (as
> > > > > mentioned in the forum). 
> > > > >
> > > > >
> > > > >
> > > > >
> > > > > > timeBeginPeriod only needs to be called once for the whole process.
> > > > > > not each time rest is called.
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > > > > Hi all,
> > > > > > >
> > > > > > > as already discussed in the forum, I would propose to change the
> > > >
> > > > current
> > > >
> > > >
> > > > > > > implementation of tim_win32_rest in wtimer.c to the following:
> > > > > > >
> > > > > > > [code]
> > > > > > > /* tim_win32_rest:
> > > > > > > *  Rests the specified amount of milliseconds.
> > > > > > > */
> > > > > > > static void tim_win32_rest(unsigned int time, AL_METHOD(void,
> >
> > callback,
> >
> > > > > > > (void)))
> > > > > > > {
> > > > > > > unsigned int start;
> > > > > > > unsigned int ms = time;
> > > > > > >
> > > > > > > const MMRESULT rc = timeBeginPeriod(1);
> > > > > > > if (callback) {
> > > > > > >    start = timeGetTime();
> > > > > > >    while (timeGetTime() - start < ms)
> > > > > > >       (*callback)();
> > > > > > > }
> > > > > > > else {
> > > > > > >    Sleep(ms);
> > > > > > > }
> > > > > > > if( rc == TIMERR_NOERROR )
> > > > > > >   timeEndPeriod(1);
> > > > > > > }
> > > > > > > [/code]
> > > > > > >
> > > > > > > The new lines are the call to timeBeginPeriod(1); and the 
> > > > > > > correcponding call
> > > > > > > to timeEndPeriod(1); in case the first call was successful.
> > > > > > >
> > > > > > > This affects only windows, and increases the accuracy of calls to 
> > > > > > > rest(n).
> > > > > > >
> > > > > > > For the discussions, refer to the following threads:
> > > > > > >
> > > > > > > http://www.allegro.cc/forums/view_thread.php?_id=456146
> > > > > > > http://www.allegro.cc/forums/view_thread.php?_id=453116
> > > > > > >
> > > > > > > As this is my very first contribution, please forgive if anything is 
> > > > > > > not in
> > > > > > > the right format or otherwise unexpected.
> > > > > > >
> > > > > > > Thanks for consideration!
> > > > > > >
> > > > > > > Best regards Tobias
> > > > > > >
> > > > > > > --