8 replies to this topic

[Hw]

I'm having a problem with timings when doing a d3d app with borland cbuilder 4. It seems that windows' timers can't handle higher frequencies than ~20Hz (even if I set the interval to 1ms!). As directx offers no solution, how am I supposed to get a timer with freq in range of 50-100Hz? Any suggestions?

[VirtualNext]

Use the function

DWORD GetTickCount(VOID)

VirtualNext

[Hw]

Hmm... You're suggesting an infine loop in main routine? Damn, that's method I used years ago in DOS (before using the PIT)... Somehow, I just don't like the idea...

Oh well. Guess I just have to live with it.

[Sphet]

I think it ultimately becomes an infinite loop, unless you are doing a non-time-dependant stratedgy came. Just don't forget to service the message queue with PeekMessage and PostMessages. Take a look at the DirectX samples..

[Splat]

GetTickCount works as a backup, but since we are in Windows we might as well check for the existance of a high-resolution timer The following code should work:

int timerFrequency;
if (!QueryPerformanceTimer(&timerFrequency)) {
// Insert fall back code using GetTickCount here
}
else {
// A high resolution timer exists, the number of "ticks" per second is in timerFrequency
int currTime;
QueryPerformanceCounter(&currTime);
// Now we have the current tick count in currTime
}

So, basically you just need QueryPerformanceCounter and Frequency, and both can be included in with "winbase.h" (or i think if you use other WinAPI functions bay already be included)

[Hw]

In fact, I found multimedia timer, which makes things pretty easy with its 1ms resolution... Now my timer set to 100Hz gives me 110 counts per second... Oh well, guess you can't have everything. Just one question arises; do interrupts overlap in single thread? For example, if I'm doing software rendering called in event handler (slow), do I still get new events to move objects etc?, or should I make scene rendering a thread which is executed whenever needed?

[Niels]

I'd opt for the multithreaded approach, but then again, that's just how I work. Huge single threaded programs tend to become rather nasty, but ofcourse, you avoid the more curious thread-related bugs...

I personally find seperating the game in 4 threads connected with queues very usefull. One for "AI" (essentially anything complicated that require lots of processing and doesn't need to update with the same frequency as the framerate), one for rendering, one for UI and one for networking.

/Niels

[Splat]

Well, even though computers are much faster than they used to be, you still must be aware of speed traps. Multithreading for example. Yes, it makes code all nice and neat and logical (that it, until you get into locking/unlocking thread related problems which are hard to debug) but you must understand that for just about any game a single threaded main game loop is just about the optimal way to program it.

The problem with your multithreaded approach is that there are several more threads that you don't even seem. For example, Flip() spawns a thread, async WinSock spawns a thread, etc. At most I could definately see a need for a second thread that handle's non-framerate dependant low priority stuff like AI, but the rest you might as well put into a well-organized main loop that will in the long run save you debugging time, speed up your program, ease portability AND readability issues and just make for a better program

[Niels]

Ok, fanzy pants, beat this :
http://home.worldonline.dk/~noshit/scrshots.html

(35fps on a 90Mhz pentium with 4Meg VMem, 90fps on a 400 PII (limited by screen refresh rate))..

But you're right - unless you do a proper design, your threads will spend more time waiting for each other, than getting actual work done....

/Niels

[This message has been edited by Niels (edited October 14, 1999).]