> Just patched and rebuilt viafb CVS (from just now) and DirectFB CVS (from a > few days back) and it looks like it is working!
Hurrah! > Every now and then, there is a blip which is probably due to the odd > 'flip-to-flip' times I was seeing. What is my best way of trying to reduce > the latency? I'm not the best person to answer that one. I get a fairly uniform rate out of xine but I suspect it has some buffering in it. If vdr/softdevice has no buffering after the decoder, you might expect a slightly bursty frame rate. You would have to add some extra diagnostics to check whether the burstiness is so great that even now you occasionally miss the window. Another possibility is that your interrupt latency is occasionally too high to get the flip in on time. There isn't much time between the vblank interrupt and the 'deadline' for setting up the registers for the next frame. I've a suspicion that on my machine I occasionally miss. Difficult to solve that one, but I might investigate further at some point. The former possibility would cause a duplicate frame, whereas the latter would cause a duplicate field. If you can tell the difference, you could work out which problem you have. > Kernel pre-emption + low latency stuff? There isn't exactly a > lot else running on the PC! You could always try real-time scheduling (SCHED_FIFO et al) but if there isn't much else on the machine, that probably won't make much of a difference. > Every now and then, it will bail out with the following seg fault: > (!) [ 4491: 0.000] --> Caught signal 11 (at 0x318, invalid address) <-- > (!) [ 4542: 0.003] --> Caught signal 11 (at 0xb293c3d2, invalid address) > <-- > > I've yet to look into this with gdb but has anyone else seen this sort of > error? Well yes, but it could be caused by anything! See what GDB tells you. > I'll give it a while longer for testing and then I'll clean up my patch for > softdevice and pass it on to Stefan. Unfortunately, you're then dependent on some experimental patches to DirectFB and viafb. Don't know what to do about that because doing this properly might take a while. Mark _______________________________________________ directfb-users mailing list [email protected] http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-users
