> situation is i am running a multithreaded program that captures
> continuously from my FlyVideo98 tuner using read() then draws to a window
> using XShmPutImage(). this loops infinitely while another thread copies
> the data and performs some analysis on it.
I was getting around 4 fps with a Webcam 3 until I enabled compression
through the driver.
I'm using ioctl(VIDIOCGMBUF) and mmap to implement shared memory and then
ioctl(VIDIOCMCAPTURE) and ioctl(VIDIOCMSYNC) to read frames. I use two
frames in the following sequence:
// initialize
ioctl(VIDIOCMCAPTURE, frame0);
while {
ioctl(VIDIOCMCAPTURE, frame1);
ioctl(VIDIOCMSYNC, frame0);
ioctl(VIDIOCMCAPTURE, frame1);
ioctl(VIDIOCMSYNC, frame0);
}
Thus I alternate frames, sending a capture request to one and syncing to
the other.
The above pseudo-code is the "server" thread. The "client" thread copies
from the appropriate buffer into malloced memory for the application. As
a further optimization, the client thread does a pthread_cond_wait on the
condition that a new frame has arrived, instead of busy waiting in a loop.
I believe this saves a significant amount of processing, because the
thread is not scheduled until a new frame actually arrives.
Best regards,
Stephen
