On Wed, 2 Jan 2002, Matt wrote: > I had thought of a potential problem that I hadn't considered with the > current setup. The problem is that timeSetEvent periodic for 16ms won't > work -- the poll time for QueryPerformanceCounter will increase by 0.667ms each > time through. This will quickly become wacky, because the high > performance counter will start to be perpetually behind, causing it to > poll for longer every time. This may be what is causing the general > wackiness during the intros and things. We should > instead start a non-periodic timeSetEvent after the Message is sent within > timeout(). > > I'm trying the new approach now, will let you know how it turns out.
Implemented this and CPU usage is much lower now. It seems to starve the main thread less as well, so things play a bit smoother than before. The tempo in complex pieces is still not perfect, but it is noticably improved on my machine. Chris/Alex, can you please test with this new code I'm about to check in? No need for extensive notes (no pun intended ;>), I just want to know if you notice the improvements. Thanks!
